diff options
Diffstat (limited to 'sandbox/linux')
159 files changed, 30633 insertions, 0 deletions
diff --git a/sandbox/linux/BUILD.gn b/sandbox/linux/BUILD.gn new file mode 100644 index 0000000000..a1a77204f5 --- /dev/null +++ b/sandbox/linux/BUILD.gn @@ -0,0 +1,382 @@ +# Copyright 2014 The Chromium Authors. All rights reserved. +# Use of this source code is governed by a BSD-style license that can be +# found in the LICENSE file. + +import("//build/config/features.gni") +import("//testing/test.gni") + +declare_args() { + compile_suid_client = is_linux + + compile_credentials = is_linux + + # On Android, use plain GTest. + use_base_test_suite = is_linux +} + +# We have two principal targets: sandbox and sandbox_linux_unittests +# All other targets are listed as dependencies. +# There is one notable exception: for historical reasons, chrome_sandbox is +# the setuid sandbox and is its own target. + +group("sandbox") { + deps = [ + ":sandbox_services", + ] + + if (compile_suid_client) { + deps += [ ":suid_sandbox_client" ] + } + if (use_seccomp_bpf) { + deps += [ + ":seccomp_bpf", + ":seccomp_bpf_helpers", + ] + } +} + +source_set("sandbox_linux_test_utils") { + testonly = true + sources = [ + "tests/sandbox_test_runner.cc", + "tests/sandbox_test_runner.h", + "tests/sandbox_test_runner_function_pointer.cc", + "tests/sandbox_test_runner_function_pointer.h", + "tests/test_utils.cc", + "tests/test_utils.h", + "tests/unit_tests.cc", + "tests/unit_tests.h", + ] + + deps = [ + "//testing/gtest", + ] + + if (use_seccomp_bpf) { + sources += [ + "seccomp-bpf/bpf_tester_compatibility_delegate.h", + "seccomp-bpf/bpf_tests.h", + "seccomp-bpf/sandbox_bpf_test_runner.cc", + "seccomp-bpf/sandbox_bpf_test_runner.h", + ] + deps += [ ":seccomp_bpf" ] + } + + if (use_base_test_suite) { + deps += [ "//base/test:test_support" ] + defines = [ "SANDBOX_USES_BASE_TEST_SUITE" ] + } +} + +# Sources shared by sandbox_linux_unittests and sandbox_linux_jni_unittests. +source_set("sandbox_linux_unittests_sources") { + testonly = true + + sources = [ + "services/proc_util_unittest.cc", + "services/resource_limits_unittests.cc", + "services/scoped_process_unittest.cc", + "services/syscall_wrappers_unittest.cc", + "services/thread_helpers_unittests.cc", + "services/yama_unittests.cc", + "syscall_broker/broker_file_permission_unittest.cc", + "syscall_broker/broker_process_unittest.cc", + "tests/main.cc", + "tests/scoped_temporary_file.cc", + "tests/scoped_temporary_file.h", + "tests/scoped_temporary_file_unittest.cc", + "tests/test_utils_unittest.cc", + "tests/unit_tests_unittest.cc", + ] + + deps = [ + ":sandbox", + ":sandbox_linux_test_utils", + "//base", + "//testing/gtest", + ] + + if (use_base_test_suite) { + deps += [ "//base/test:test_support" ] + defines = [ "SANDBOX_USES_BASE_TEST_SUITE" ] + } + + if (is_linux) { + # Don't use this on Android. + libs = [ "rt" ] + } + + if (compile_suid_client) { + sources += [ + "suid/client/setuid_sandbox_client_unittest.cc", + "suid/client/setuid_sandbox_host_unittest.cc", + ] + } + if (use_seccomp_bpf) { + sources += [ + "bpf_dsl/bpf_dsl_unittest.cc", + "bpf_dsl/codegen_unittest.cc", + "bpf_dsl/cons_unittest.cc", + "bpf_dsl/syscall_set_unittest.cc", + "integration_tests/bpf_dsl_seccomp_unittest.cc", + "integration_tests/seccomp_broker_process_unittest.cc", + "seccomp-bpf-helpers/baseline_policy_unittest.cc", + "seccomp-bpf-helpers/syscall_parameters_restrictions_unittests.cc", + "seccomp-bpf/bpf_tests_unittest.cc", + "seccomp-bpf/errorcode_unittest.cc", + "seccomp-bpf/sandbox_bpf_unittest.cc", + "seccomp-bpf/syscall_unittest.cc", + "seccomp-bpf/trap_unittest.cc", + ] + } + if (compile_credentials) { + sources += [ + "integration_tests/namespace_unix_domain_socket_unittest.cc", + "services/credentials_unittest.cc", + "services/namespace_utils_unittest.cc", + ] + + if (use_base_test_suite) { + # Tests that use advanced features not available in stock GTest. + sources += [ "services/namespace_sandbox_unittest.cc" ] + } + + # For credentials_unittest.cc + configs += [ "//build/config/linux:libcap" ] + } +} + +# The main sandboxing test target. +test("sandbox_linux_unittests") { + deps = [ + ":sandbox_linux_unittests_sources", + ] +} + +# This target is the shared library used by Android APK (i.e. +# JNI-friendly) tests. +shared_library("sandbox_linux_jni_unittests") { + testonly = true + deps = [ + ":sandbox_linux_unittests_sources", + ] + if (is_android) { + deps += [ "//testing/android/native_test:native_test_native_code" ] + } +} + +component("seccomp_bpf") { + sources = [ + "bpf_dsl/bpf_dsl.cc", + "bpf_dsl/bpf_dsl.h", + "bpf_dsl/bpf_dsl_forward.h", + "bpf_dsl/bpf_dsl_impl.h", + "bpf_dsl/codegen.cc", + "bpf_dsl/codegen.h", + "bpf_dsl/cons.h", + "bpf_dsl/dump_bpf.cc", + "bpf_dsl/dump_bpf.h", + "bpf_dsl/linux_syscall_ranges.h", + "bpf_dsl/policy.cc", + "bpf_dsl/policy.h", + "bpf_dsl/policy_compiler.cc", + "bpf_dsl/policy_compiler.h", + "bpf_dsl/seccomp_macros.h", + "bpf_dsl/syscall_set.cc", + "bpf_dsl/syscall_set.h", + "bpf_dsl/trap_registry.h", + "bpf_dsl/verifier.cc", + "bpf_dsl/verifier.h", + "seccomp-bpf/die.cc", + "seccomp-bpf/die.h", + "seccomp-bpf/errorcode.cc", + "seccomp-bpf/errorcode.h", + "seccomp-bpf/sandbox_bpf.cc", + "seccomp-bpf/sandbox_bpf.h", + "seccomp-bpf/syscall.cc", + "seccomp-bpf/syscall.h", + "seccomp-bpf/trap.cc", + "seccomp-bpf/trap.h", + ] + defines = [ "SANDBOX_IMPLEMENTATION" ] + + deps = [ + ":sandbox_services", + ":sandbox_services_headers", + "//base", + ] +} + +component("seccomp_bpf_helpers") { + sources = [ + "seccomp-bpf-helpers/baseline_policy.cc", + "seccomp-bpf-helpers/baseline_policy.h", + "seccomp-bpf-helpers/sigsys_handlers.cc", + "seccomp-bpf-helpers/sigsys_handlers.h", + "seccomp-bpf-helpers/syscall_parameters_restrictions.cc", + "seccomp-bpf-helpers/syscall_parameters_restrictions.h", + "seccomp-bpf-helpers/syscall_sets.cc", + "seccomp-bpf-helpers/syscall_sets.h", + ] + defines = [ "SANDBOX_IMPLEMENTATION" ] + + deps = [ + "//base", + ":sandbox_services", + ":seccomp_bpf", + ] +} + +if (is_linux) { + # The setuid sandbox for Linux. + executable("chrome_sandbox") { + sources = [ + "suid/common/sandbox.h", + "suid/common/suid_unsafe_environment_variables.h", + "suid/process_util.h", + "suid/process_util_linux.c", + "suid/sandbox.c", + ] + + cflags = [ + # For ULLONG_MAX + "-std=gnu99", + + # These files have a suspicious comparison. + # TODO fix this and re-enable this warning. + "-Wno-sign-compare", + ] + } +} + +component("sandbox_services") { + sources = [ + "services/init_process_reaper.cc", + "services/init_process_reaper.h", + "services/proc_util.cc", + "services/proc_util.h", + "services/resource_limits.cc", + "services/resource_limits.h", + "services/scoped_process.cc", + "services/scoped_process.h", + "services/syscall_wrappers.cc", + "services/syscall_wrappers.h", + "services/thread_helpers.cc", + "services/thread_helpers.h", + "services/yama.cc", + "services/yama.h", + "syscall_broker/broker_channel.cc", + "syscall_broker/broker_channel.h", + "syscall_broker/broker_client.cc", + "syscall_broker/broker_client.h", + "syscall_broker/broker_common.h", + "syscall_broker/broker_file_permission.cc", + "syscall_broker/broker_file_permission.h", + "syscall_broker/broker_host.cc", + "syscall_broker/broker_host.h", + "syscall_broker/broker_policy.cc", + "syscall_broker/broker_policy.h", + "syscall_broker/broker_process.cc", + "syscall_broker/broker_process.h", + ] + + defines = [ "SANDBOX_IMPLEMENTATION" ] + + deps = [ + "//base", + ] + + if (compile_credentials) { + sources += [ + "services/credentials.cc", + "services/credentials.h", + "services/namespace_sandbox.cc", + "services/namespace_sandbox.h", + "services/namespace_utils.cc", + "services/namespace_utils.h", + ] + + deps += [ ":sandbox_services_headers" ] + } +} + +source_set("sandbox_services_headers") { + sources = [ + "system_headers/arm64_linux_syscalls.h", + "system_headers/arm64_linux_ucontext.h", + "system_headers/arm_linux_syscalls.h", + "system_headers/arm_linux_ucontext.h", + "system_headers/i386_linux_ucontext.h", + "system_headers/linux_futex.h", + "system_headers/linux_seccomp.h", + "system_headers/linux_signal.h", + "system_headers/linux_syscalls.h", + "system_headers/linux_time.h", + "system_headers/linux_ucontext.h", + "system_headers/x86_32_linux_syscalls.h", + "system_headers/x86_64_linux_syscalls.h", + ] +} + +# We make this its own target so that it does not interfere with our tests. +source_set("libc_urandom_override") { + sources = [ + "services/libc_urandom_override.cc", + "services/libc_urandom_override.h", + ] + deps = [ + "//base", + ] +} + +if (compile_suid_client) { + component("suid_sandbox_client") { + sources = [ + "suid/client/setuid_sandbox_client.cc", + "suid/client/setuid_sandbox_client.h", + "suid/client/setuid_sandbox_host.cc", + "suid/client/setuid_sandbox_host.h", + "suid/common/sandbox.h", + "suid/common/suid_unsafe_environment_variables.h", + ] + defines = [ "SANDBOX_IMPLEMENTATION" ] + + deps = [ + ":sandbox_services", + "//base", + ] + } +} + +if (is_android) { + # TODO(GYP) enable this. Needs an android_strip wrapper python script. + #action("sandbox_linux_unittests_stripped") { + # script = "android_stip.py" + # + # in_file = "$root_out_dir/sandbox_linux_unittests" + # + # out_file = "$root_out_dir/sandbox_linux_unittests_stripped" + # outputs = [ out_file ] + # + # args = [ + # rebase_path(in_file, root_build_dir), + # "-o", rebase_path(out_file, root_build_dir), + # ] + # + # deps = [ + # ":sandbox_linux_unittests", + # ] + #} + # TODO(GYP) convert this. + # { + # 'target_name': 'sandbox_linux_jni_unittests_apk', + # 'type': 'none', + # 'variables': { + # 'test_suite_name': 'sandbox_linux_jni_unittests', + # }, + # 'dependencies': [ + # 'sandbox_linux_jni_unittests', + # ], + # 'includes': [ '../../build/apk_test.gypi' ], + # } +} diff --git a/sandbox/linux/DEPS b/sandbox/linux/DEPS new file mode 100644 index 0000000000..3912859344 --- /dev/null +++ b/sandbox/linux/DEPS @@ -0,0 +1,25 @@ +include_rules = [ + # First, exclude everything. + # Exclude a few dependencies that are included in the root DEPS and that we + # don't need. + # Sadly, there is no way to exclude all root DEPS since the root has no name. + "-ipc", + "-library_loaders", + "-third_party", + "-url", + # Make sure that each subdirectory has to declare its dependencies in + # sandbox/ explicitly. + "-sandbox/linux", + + # Second, add what we want to allow. + # Anything included from sandbox/linux must be declared after this line or in + # a more specific DEPS file. + # base/, build/ and testing/ are already included in the global DEPS file, + # but be explicit. + "+base", + "+build", + "+testing", + "+sandbox/sandbox_export.h", + # Everyone can use tests/ + "+sandbox/linux/tests", +] diff --git a/sandbox/linux/OWNERS b/sandbox/linux/OWNERS new file mode 100644 index 0000000000..f39e96736d --- /dev/null +++ b/sandbox/linux/OWNERS @@ -0,0 +1,3 @@ +jln@chromium.org +jorgelo@chromium.org +mdempsky@chromium.org diff --git a/sandbox/linux/bpf_dsl/DEPS b/sandbox/linux/bpf_dsl/DEPS new file mode 100644 index 0000000000..be37a129c2 --- /dev/null +++ b/sandbox/linux/bpf_dsl/DEPS @@ -0,0 +1,5 @@ +include_rules = [ + # TODO(mdempsky): Eliminate cyclic dependency on seccomp-bpf. + "+sandbox/linux/seccomp-bpf", + "+sandbox/linux/system_headers", +] diff --git a/sandbox/linux/bpf_dsl/bpf_dsl.cc b/sandbox/linux/bpf_dsl/bpf_dsl.cc new file mode 100644 index 0000000000..3a35903ec9 --- /dev/null +++ b/sandbox/linux/bpf_dsl/bpf_dsl.cc @@ -0,0 +1,363 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" + +#include <limits> + +#include "base/logging.h" +#include "base/memory/ref_counted.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_impl.h" +#include "sandbox/linux/bpf_dsl/policy_compiler.h" +#include "sandbox/linux/seccomp-bpf/die.h" +#include "sandbox/linux/seccomp-bpf/errorcode.h" + +namespace sandbox { +namespace bpf_dsl { +namespace { + +intptr_t BPFFailure(const struct arch_seccomp_data&, void* aux) { + SANDBOX_DIE(static_cast<char*>(aux)); +} + +class AllowResultExprImpl : public internal::ResultExprImpl { + public: + AllowResultExprImpl() {} + + ErrorCode Compile(PolicyCompiler* pc) const override { + return ErrorCode(ErrorCode::ERR_ALLOWED); + } + + bool IsAllow() const override { return true; } + + private: + ~AllowResultExprImpl() override {} + + DISALLOW_COPY_AND_ASSIGN(AllowResultExprImpl); +}; + +class ErrorResultExprImpl : public internal::ResultExprImpl { + public: + explicit ErrorResultExprImpl(int err) : err_(err) { + CHECK(err_ >= ErrorCode::ERR_MIN_ERRNO && err_ <= ErrorCode::ERR_MAX_ERRNO); + } + + ErrorCode Compile(PolicyCompiler* pc) const override { + return pc->Error(err_); + } + + bool IsDeny() const override { return true; } + + private: + ~ErrorResultExprImpl() override {} + + int err_; + + DISALLOW_COPY_AND_ASSIGN(ErrorResultExprImpl); +}; + +class TraceResultExprImpl : public internal::ResultExprImpl { + public: + TraceResultExprImpl(uint16_t aux) : aux_(aux) {} + + ErrorCode Compile(PolicyCompiler* pc) const override { + return ErrorCode(ErrorCode::ERR_TRACE + aux_); + } + + private: + ~TraceResultExprImpl() override {} + + uint16_t aux_; + + DISALLOW_COPY_AND_ASSIGN(TraceResultExprImpl); +}; + +class TrapResultExprImpl : public internal::ResultExprImpl { + public: + TrapResultExprImpl(TrapRegistry::TrapFnc func, const void* arg, bool safe) + : func_(func), arg_(arg), safe_(safe) { + DCHECK(func_); + } + + ErrorCode Compile(PolicyCompiler* pc) const override { + return pc->Trap(func_, arg_, safe_); + } + + bool HasUnsafeTraps() const override { return safe_ == false; } + + bool IsDeny() const override { return true; } + + private: + ~TrapResultExprImpl() override {} + + TrapRegistry::TrapFnc func_; + const void* arg_; + bool safe_; + + DISALLOW_COPY_AND_ASSIGN(TrapResultExprImpl); +}; + +class IfThenResultExprImpl : public internal::ResultExprImpl { + public: + IfThenResultExprImpl(const BoolExpr& cond, + const ResultExpr& then_result, + const ResultExpr& else_result) + : cond_(cond), then_result_(then_result), else_result_(else_result) {} + + ErrorCode Compile(PolicyCompiler* pc) const override { + return cond_->Compile( + pc, then_result_->Compile(pc), else_result_->Compile(pc)); + } + + bool HasUnsafeTraps() const override { + return then_result_->HasUnsafeTraps() || else_result_->HasUnsafeTraps(); + } + + private: + ~IfThenResultExprImpl() override {} + + BoolExpr cond_; + ResultExpr then_result_; + ResultExpr else_result_; + + DISALLOW_COPY_AND_ASSIGN(IfThenResultExprImpl); +}; + +class ConstBoolExprImpl : public internal::BoolExprImpl { + public: + ConstBoolExprImpl(bool value) : value_(value) {} + + ErrorCode Compile(PolicyCompiler* pc, + ErrorCode true_ec, + ErrorCode false_ec) const override { + return value_ ? true_ec : false_ec; + } + + private: + ~ConstBoolExprImpl() override {} + + bool value_; + + DISALLOW_COPY_AND_ASSIGN(ConstBoolExprImpl); +}; + +class PrimitiveBoolExprImpl : public internal::BoolExprImpl { + public: + PrimitiveBoolExprImpl(int argno, + ErrorCode::ArgType is_32bit, + uint64_t mask, + uint64_t value) + : argno_(argno), is_32bit_(is_32bit), mask_(mask), value_(value) {} + + ErrorCode Compile(PolicyCompiler* pc, + ErrorCode true_ec, + ErrorCode false_ec) const override { + return pc->CondMaskedEqual( + argno_, is_32bit_, mask_, value_, true_ec, false_ec); + } + + private: + ~PrimitiveBoolExprImpl() override {} + + int argno_; + ErrorCode::ArgType is_32bit_; + uint64_t mask_; + uint64_t value_; + + DISALLOW_COPY_AND_ASSIGN(PrimitiveBoolExprImpl); +}; + +class NegateBoolExprImpl : public internal::BoolExprImpl { + public: + explicit NegateBoolExprImpl(const BoolExpr& cond) : cond_(cond) {} + + ErrorCode Compile(PolicyCompiler* pc, + ErrorCode true_ec, + ErrorCode false_ec) const override { + return cond_->Compile(pc, false_ec, true_ec); + } + + private: + ~NegateBoolExprImpl() override {} + + BoolExpr cond_; + + DISALLOW_COPY_AND_ASSIGN(NegateBoolExprImpl); +}; + +class AndBoolExprImpl : public internal::BoolExprImpl { + public: + AndBoolExprImpl(const BoolExpr& lhs, const BoolExpr& rhs) + : lhs_(lhs), rhs_(rhs) {} + + ErrorCode Compile(PolicyCompiler* pc, + ErrorCode true_ec, + ErrorCode false_ec) const override { + return lhs_->Compile(pc, rhs_->Compile(pc, true_ec, false_ec), false_ec); + } + + private: + ~AndBoolExprImpl() override {} + + BoolExpr lhs_; + BoolExpr rhs_; + + DISALLOW_COPY_AND_ASSIGN(AndBoolExprImpl); +}; + +class OrBoolExprImpl : public internal::BoolExprImpl { + public: + OrBoolExprImpl(const BoolExpr& lhs, const BoolExpr& rhs) + : lhs_(lhs), rhs_(rhs) {} + + ErrorCode Compile(PolicyCompiler* pc, + ErrorCode true_ec, + ErrorCode false_ec) const override { + return lhs_->Compile(pc, true_ec, rhs_->Compile(pc, true_ec, false_ec)); + } + + private: + ~OrBoolExprImpl() override {} + + BoolExpr lhs_; + BoolExpr rhs_; + + DISALLOW_COPY_AND_ASSIGN(OrBoolExprImpl); +}; + +} // namespace + +namespace internal { + +bool ResultExprImpl::HasUnsafeTraps() const { + return false; +} + +bool ResultExprImpl::IsAllow() const { + return false; +} + +bool ResultExprImpl::IsDeny() const { + return false; +} + +uint64_t DefaultMask(size_t size) { + switch (size) { + case 4: + return std::numeric_limits<uint32_t>::max(); + case 8: + return std::numeric_limits<uint64_t>::max(); + default: + CHECK(false) << "Unimplemented DefaultMask case"; + return 0; + } +} + +BoolExpr ArgEq(int num, size_t size, uint64_t mask, uint64_t val) { + CHECK(size == 4 || size == 8); + + // TODO(mdempsky): Should we just always use TP_64BIT? + const ErrorCode::ArgType arg_type = + (size == 4) ? ErrorCode::TP_32BIT : ErrorCode::TP_64BIT; + + return BoolExpr(new const PrimitiveBoolExprImpl(num, arg_type, mask, val)); +} + +} // namespace internal + +ResultExpr Allow() { + return ResultExpr(new const AllowResultExprImpl()); +} + +ResultExpr Error(int err) { + return ResultExpr(new const ErrorResultExprImpl(err)); +} + +ResultExpr Kill(const char* msg) { + return Trap(BPFFailure, msg); +} + +ResultExpr Trace(uint16_t aux) { + return ResultExpr(new const TraceResultExprImpl(aux)); +} + +ResultExpr Trap(TrapRegistry::TrapFnc trap_func, const void* aux) { + return ResultExpr( + new const TrapResultExprImpl(trap_func, aux, true /* safe */)); +} + +ResultExpr UnsafeTrap(TrapRegistry::TrapFnc trap_func, const void* aux) { + return ResultExpr( + new const TrapResultExprImpl(trap_func, aux, false /* unsafe */)); +} + +BoolExpr BoolConst(bool value) { + return BoolExpr(new const ConstBoolExprImpl(value)); +} + +BoolExpr operator!(const BoolExpr& cond) { + return BoolExpr(new const NegateBoolExprImpl(cond)); +} + +BoolExpr operator&&(const BoolExpr& lhs, const BoolExpr& rhs) { + return BoolExpr(new const AndBoolExprImpl(lhs, rhs)); +} + +BoolExpr operator||(const BoolExpr& lhs, const BoolExpr& rhs) { + return BoolExpr(new const OrBoolExprImpl(lhs, rhs)); +} + +Elser If(const BoolExpr& cond, const ResultExpr& then_result) { + return Elser(nullptr).ElseIf(cond, then_result); +} + +Elser::Elser(cons::List<Clause> clause_list) : clause_list_(clause_list) { +} + +Elser::Elser(const Elser& elser) : clause_list_(elser.clause_list_) { +} + +Elser::~Elser() { +} + +Elser Elser::ElseIf(const BoolExpr& cond, const ResultExpr& then_result) const { + return Elser(Cons(std::make_pair(cond, then_result), clause_list_)); +} + +ResultExpr Elser::Else(const ResultExpr& else_result) const { + // We finally have the default result expression for this + // if/then/else sequence. Also, we've already accumulated all + // if/then pairs into a list of reverse order (i.e., lower priority + // conditions are listed before higher priority ones). E.g., an + // expression like + // + // If(b1, e1).ElseIf(b2, e2).ElseIf(b3, e3).Else(e4) + // + // will have built up a list like + // + // [(b3, e3), (b2, e2), (b1, e1)]. + // + // Now that we have e4, we can walk the list and create a ResultExpr + // tree like: + // + // expr = e4 + // expr = (b3 ? e3 : expr) = (b3 ? e3 : e4) + // expr = (b2 ? e2 : expr) = (b2 ? e2 : (b3 ? e3 : e4)) + // expr = (b1 ? e1 : expr) = (b1 ? e1 : (b2 ? e2 : (b3 ? e3 : e4))) + // + // and end up with an appropriately chained tree. + + ResultExpr expr = else_result; + for (const Clause& clause : clause_list_) { + expr = ResultExpr( + new const IfThenResultExprImpl(clause.first, clause.second, expr)); + } + return expr; +} + +} // namespace bpf_dsl +} // namespace sandbox + +template class scoped_refptr<const sandbox::bpf_dsl::internal::BoolExprImpl>; +template class scoped_refptr<const sandbox::bpf_dsl::internal::ResultExprImpl>; diff --git a/sandbox/linux/bpf_dsl/bpf_dsl.h b/sandbox/linux/bpf_dsl/bpf_dsl.h new file mode 100644 index 0000000000..365e9b5466 --- /dev/null +++ b/sandbox/linux/bpf_dsl/bpf_dsl.h @@ -0,0 +1,317 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_BPF_DSL_H_ +#define SANDBOX_LINUX_BPF_DSL_BPF_DSL_H_ + +#include <stdint.h> + +#include <utility> +#include <vector> + +#include "base/macros.h" +#include "base/memory/ref_counted.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_forward.h" +#include "sandbox/linux/bpf_dsl/cons.h" +#include "sandbox/linux/bpf_dsl/trap_registry.h" +#include "sandbox/sandbox_export.h" + +// The sandbox::bpf_dsl namespace provides a domain-specific language +// to make writing BPF policies more expressive. In general, the +// object types all have value semantics (i.e., they can be copied +// around, returned from or passed to function calls, etc. without any +// surprising side effects), though not all support assignment. +// +// An idiomatic and demonstrative (albeit silly) example of this API +// would be: +// +// #include "sandbox/linux/bpf_dsl/bpf_dsl.h" +// +// using namespace sandbox::bpf_dsl; +// +// class SillyPolicy : public Policy { +// public: +// SillyPolicy() {} +// ~SillyPolicy() override {} +// ResultExpr EvaluateSyscall(int sysno) const override { +// if (sysno == __NR_fcntl) { +// Arg<int> fd(0), cmd(1); +// Arg<unsigned long> flags(2); +// const uint64_t kGoodFlags = O_ACCMODE | O_NONBLOCK; +// return If(fd == 0 && cmd == F_SETFL && (flags & ~kGoodFlags) == 0, +// Allow()) +// .ElseIf(cmd == F_DUPFD || cmd == F_DUPFD_CLOEXEC, +// Error(EMFILE)) +// .Else(Trap(SetFlagHandler, NULL)); +// } else { +// return Allow(); +// } +// } +// +// private: +// DISALLOW_COPY_AND_ASSIGN(SillyPolicy); +// }; +// +// More generally, the DSL currently supports the following grammar: +// +// result = Allow() | Error(errno) | Kill(msg) | Trace(aux) +// | Trap(trap_func, aux) | UnsafeTrap(trap_func, aux) +// | If(bool, result)[.ElseIf(bool, result)].Else(result) +// | Switch(arg)[.Case(val, result)].Default(result) +// bool = BoolConst(boolean) | !bool | bool && bool | bool || bool +// | arg == val | arg != val +// arg = Arg<T>(num) | arg & mask +// +// The semantics of each function and operator are intended to be +// intuitive, but are described in more detail below. +// +// (Credit to Sean Parent's "Inheritance is the Base Class of Evil" +// talk at Going Native 2013 for promoting value semantics via shared +// pointers to immutable state.) + +namespace sandbox { +namespace bpf_dsl { + +// ResultExpr is an opaque reference to an immutable result expression tree. +typedef scoped_refptr<const internal::ResultExprImpl> ResultExpr; + +// BoolExpr is an opaque reference to an immutable boolean expression tree. +typedef scoped_refptr<const internal::BoolExprImpl> BoolExpr; + +// Allow specifies a result that the system call should be allowed to +// execute normally. +SANDBOX_EXPORT ResultExpr Allow(); + +// Error specifies a result that the system call should fail with +// error number |err|. As a special case, Error(0) will result in the +// system call appearing to have succeeded, but without having any +// side effects. +SANDBOX_EXPORT ResultExpr Error(int err); + +// Kill specifies a result to kill the program and print an error message. +SANDBOX_EXPORT ResultExpr Kill(const char* msg); + +// Trace specifies a result to notify a tracing process via the +// PTRACE_EVENT_SECCOMP event and allow it to change or skip the system call. +// The value of |aux| will be available to the tracer via PTRACE_GETEVENTMSG. +SANDBOX_EXPORT ResultExpr Trace(uint16_t aux); + +// Trap specifies a result that the system call should be handled by +// trapping back into userspace and invoking |trap_func|, passing +// |aux| as the second parameter. +SANDBOX_EXPORT ResultExpr + Trap(TrapRegistry::TrapFnc trap_func, const void* aux); + +// UnsafeTrap is like Trap, except the policy is marked as "unsafe" +// and allowed to use SandboxSyscall to invoke any system call. +// +// NOTE: This feature, by definition, disables all security features of +// the sandbox. It should never be used in production, but it can be +// very useful to diagnose code that is incompatible with the sandbox. +// If even a single system call returns "UnsafeTrap", the security of +// entire sandbox should be considered compromised. +SANDBOX_EXPORT ResultExpr + UnsafeTrap(TrapRegistry::TrapFnc trap_func, const void* aux); + +// BoolConst converts a bool value into a BoolExpr. +SANDBOX_EXPORT BoolExpr BoolConst(bool value); + +// Various ways to combine boolean expressions into more complex expressions. +// They follow standard boolean algebra laws. +SANDBOX_EXPORT BoolExpr operator!(const BoolExpr& cond); +SANDBOX_EXPORT BoolExpr operator&&(const BoolExpr& lhs, const BoolExpr& rhs); +SANDBOX_EXPORT BoolExpr operator||(const BoolExpr& lhs, const BoolExpr& rhs); + +template <typename T> +class SANDBOX_EXPORT Arg { + public: + // Initializes the Arg to represent the |num|th system call + // argument (indexed from 0), which is of type |T|. + explicit Arg(int num); + + Arg(const Arg& arg) : num_(arg.num_), mask_(arg.mask_) {} + + // Returns an Arg representing the current argument, but after + // bitwise-and'ing it with |rhs|. + friend Arg operator&(const Arg& lhs, uint64_t rhs) { + return Arg(lhs.num_, lhs.mask_ & rhs); + } + + // Returns a boolean expression comparing whether the system call argument + // (after applying any bitmasks, if appropriate) equals |rhs|. + friend BoolExpr operator==(const Arg& lhs, T rhs) { return lhs.EqualTo(rhs); } + + // Returns a boolean expression comparing whether the system call argument + // (after applying any bitmasks, if appropriate) does not equal |rhs|. + friend BoolExpr operator!=(const Arg& lhs, T rhs) { return !(lhs == rhs); } + + private: + Arg(int num, uint64_t mask) : num_(num), mask_(mask) {} + + BoolExpr EqualTo(T val) const; + + int num_; + uint64_t mask_; + + DISALLOW_ASSIGN(Arg); +}; + +// If begins a conditional result expression predicated on the +// specified boolean expression. +SANDBOX_EXPORT Elser If(const BoolExpr& cond, const ResultExpr& then_result); + +class SANDBOX_EXPORT Elser { + public: + Elser(const Elser& elser); + ~Elser(); + + // ElseIf extends the conditional result expression with another + // "if then" clause, predicated on the specified boolean expression. + Elser ElseIf(const BoolExpr& cond, const ResultExpr& then_result) const; + + // Else terminates a conditional result expression using |else_result| as + // the default fallback result expression. + ResultExpr Else(const ResultExpr& else_result) const; + + private: + typedef std::pair<BoolExpr, ResultExpr> Clause; + + explicit Elser(cons::List<Clause> clause_list); + + cons::List<Clause> clause_list_; + + friend Elser If(const BoolExpr&, const ResultExpr&); + template <typename T> + friend Caser<T> Switch(const Arg<T>&); + DISALLOW_ASSIGN(Elser); +}; + +// Switch begins a switch expression dispatched according to the +// specified argument value. +template <typename T> +SANDBOX_EXPORT Caser<T> Switch(const Arg<T>& arg); + +template <typename T> +class SANDBOX_EXPORT Caser { + public: + Caser(const Caser<T>& caser) : arg_(caser.arg_), elser_(caser.elser_) {} + ~Caser() {} + + // Case adds a single-value "case" clause to the switch. + Caser<T> Case(T value, ResultExpr result) const; + + // Cases adds a multiple-value "case" clause to the switch. + // See also the SANDBOX_BPF_DSL_CASES macro below for a more idiomatic way + // of using this function. + Caser<T> Cases(const std::vector<T>& values, ResultExpr result) const; + + // Terminate the switch with a "default" clause. + ResultExpr Default(ResultExpr result) const; + + private: + Caser(const Arg<T>& arg, Elser elser) : arg_(arg), elser_(elser) {} + + Arg<T> arg_; + Elser elser_; + + template <typename U> + friend Caser<U> Switch(const Arg<U>&); + DISALLOW_ASSIGN(Caser); +}; + +// Recommended usage is to put +// #define CASES SANDBOX_BPF_DSL_CASES +// near the top of the .cc file (e.g., nearby any "using" statements), then +// use like: +// Switch(arg).CASES((3, 5, 7), result)...; +#define SANDBOX_BPF_DSL_CASES(values, result) \ + Cases(SANDBOX_BPF_DSL_CASES_HELPER values, result) + +// Helper macro to construct a std::vector from an initializer list. +// TODO(mdempsky): Convert to use C++11 initializer lists instead. +#define SANDBOX_BPF_DSL_CASES_HELPER(value, ...) \ + ({ \ + const __typeof__(value) bpf_dsl_cases_values[] = {value, __VA_ARGS__}; \ + std::vector<__typeof__(value)>( \ + bpf_dsl_cases_values, \ + bpf_dsl_cases_values + arraysize(bpf_dsl_cases_values)); \ + }) + +// ===================================================================== +// Official API ends here. +// ===================================================================== + +namespace internal { + +// Make argument-dependent lookup work. This is necessary because although +// BoolExpr is defined in bpf_dsl, since it's merely a typedef for +// scoped_refptr<const internal::BoolExplImpl>, argument-dependent lookup only +// searches the "internal" nested namespace. +using bpf_dsl::operator!; +using bpf_dsl::operator||; +using bpf_dsl::operator&&; + +// Returns a boolean expression that represents whether system call +// argument |num| of size |size| is equal to |val|, when masked +// according to |mask|. Users should use the Arg template class below +// instead of using this API directly. +SANDBOX_EXPORT BoolExpr + ArgEq(int num, size_t size, uint64_t mask, uint64_t val); + +// Returns the default mask for a system call argument of the specified size. +SANDBOX_EXPORT uint64_t DefaultMask(size_t size); + +} // namespace internal + +template <typename T> +Arg<T>::Arg(int num) + : num_(num), mask_(internal::DefaultMask(sizeof(T))) { +} + +// Definition requires ArgEq to have been declared. Moved out-of-line +// to minimize how much internal clutter users have to ignore while +// reading the header documentation. +// +// Additionally, we use this helper member function to avoid linker errors +// caused by defining operator== out-of-line. For a more detailed explanation, +// see http://www.parashift.com/c++-faq-lite/template-friends.html. +template <typename T> +BoolExpr Arg<T>::EqualTo(T val) const { + return internal::ArgEq(num_, sizeof(T), mask_, static_cast<uint64_t>(val)); +} + +template <typename T> +SANDBOX_EXPORT Caser<T> Switch(const Arg<T>& arg) { + return Caser<T>(arg, Elser(nullptr)); +} + +template <typename T> +Caser<T> Caser<T>::Case(T value, ResultExpr result) const { + return SANDBOX_BPF_DSL_CASES((value), result); +} + +template <typename T> +Caser<T> Caser<T>::Cases(const std::vector<T>& values, + ResultExpr result) const { + // Theoretically we could evaluate arg_ just once and emit a more efficient + // dispatch table, but for now we simply translate into an equivalent + // If/ElseIf/Else chain. + + typedef typename std::vector<T>::const_iterator Iter; + BoolExpr test = BoolConst(false); + for (Iter i = values.begin(), end = values.end(); i != end; ++i) { + test = test || (arg_ == *i); + } + return Caser<T>(arg_, elser_.ElseIf(test, result)); +} + +template <typename T> +ResultExpr Caser<T>::Default(ResultExpr result) const { + return elser_.Else(result); +} + +} // namespace bpf_dsl +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_BPF_DSL_H_ diff --git a/sandbox/linux/bpf_dsl/bpf_dsl_forward.h b/sandbox/linux/bpf_dsl/bpf_dsl_forward.h new file mode 100644 index 0000000000..183038990a --- /dev/null +++ b/sandbox/linux/bpf_dsl/bpf_dsl_forward.h @@ -0,0 +1,42 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_BPF_DSL_FORWARD_H_ +#define SANDBOX_LINUX_BPF_DSL_BPF_DSL_FORWARD_H_ + +#include "base/memory/ref_counted.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { +namespace bpf_dsl { + +// The bpf_dsl_forward.h header provides forward declarations for the +// types defined in bpf_dsl.h. It's intended for use in user headers +// that need to reference bpf_dsl types, but don't require definitions. + +namespace internal { +class ResultExprImpl; +class BoolExprImpl; +} + +typedef scoped_refptr<const internal::ResultExprImpl> ResultExpr; +typedef scoped_refptr<const internal::BoolExprImpl> BoolExpr; + +template <typename T> +class Arg; + +class Elser; + +template <typename T> +class Caser; + +} // namespace bpf_dsl +} // namespace sandbox + +extern template class SANDBOX_EXPORT + scoped_refptr<const sandbox::bpf_dsl::internal::BoolExprImpl>; +extern template class SANDBOX_EXPORT + scoped_refptr<const sandbox::bpf_dsl::internal::ResultExprImpl>; + +#endif // SANDBOX_LINUX_BPF_DSL_BPF_DSL_FORWARD_H_ diff --git a/sandbox/linux/bpf_dsl/bpf_dsl_impl.h b/sandbox/linux/bpf_dsl/bpf_dsl_impl.h new file mode 100644 index 0000000000..2ffaf79c2f --- /dev/null +++ b/sandbox/linux/bpf_dsl/bpf_dsl_impl.h @@ -0,0 +1,69 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_BPF_DSL_IMPL_H_ +#define SANDBOX_LINUX_BPF_DSL_BPF_DSL_IMPL_H_ + +#include "base/macros.h" +#include "base/memory/ref_counted.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { +class ErrorCode; + +namespace bpf_dsl { +class PolicyCompiler; + +namespace internal { + +// Internal interface implemented by BoolExpr implementations. +class BoolExprImpl : public base::RefCounted<BoolExprImpl> { + public: + // Compile uses |pc| to construct an ErrorCode that conditionally continues + // to either |true_ec| or |false_ec|, depending on whether the represented + // boolean expression is true or false. + virtual ErrorCode Compile(PolicyCompiler* pc, + ErrorCode true_ec, + ErrorCode false_ec) const = 0; + + protected: + BoolExprImpl() {} + virtual ~BoolExprImpl() {} + + private: + friend class base::RefCounted<BoolExprImpl>; + DISALLOW_COPY_AND_ASSIGN(BoolExprImpl); +}; + +// Internal interface implemented by ResultExpr implementations. +class ResultExprImpl : public base::RefCounted<ResultExprImpl> { + public: + // Compile uses |pc| to construct an ErrorCode analogous to the represented + // result expression. + virtual ErrorCode Compile(PolicyCompiler* pc) const = 0; + + // HasUnsafeTraps returns whether the result expression is or recursively + // contains an unsafe trap expression. + virtual bool HasUnsafeTraps() const; + + // IsAllow returns whether the result expression is an "allow" result. + virtual bool IsAllow() const; + + // IsAllow returns whether the result expression is a "deny" result. + virtual bool IsDeny() const; + + protected: + ResultExprImpl() {} + virtual ~ResultExprImpl() {} + + private: + friend class base::RefCounted<ResultExprImpl>; + DISALLOW_COPY_AND_ASSIGN(ResultExprImpl); +}; + +} // namespace internal +} // namespace bpf_dsl +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_BPF_DSL_IMPL_H_ diff --git a/sandbox/linux/bpf_dsl/bpf_dsl_unittest.cc b/sandbox/linux/bpf_dsl/bpf_dsl_unittest.cc new file mode 100644 index 0000000000..398ec59ef1 --- /dev/null +++ b/sandbox/linux/bpf_dsl/bpf_dsl_unittest.cc @@ -0,0 +1,486 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" + +#include <errno.h> +#include <fcntl.h> +#include <netinet/in.h> +#include <sys/socket.h> +#include <sys/syscall.h> +#include <sys/utsname.h> +#include <unistd.h> + +#include <map> +#include <utility> + +#include "base/files/scoped_file.h" +#include "base/macros.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_impl.h" +#include "sandbox/linux/bpf_dsl/codegen.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/bpf_dsl/policy_compiler.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" +#include "sandbox/linux/bpf_dsl/trap_registry.h" +#include "sandbox/linux/bpf_dsl/verifier.h" +#include "sandbox/linux/seccomp-bpf/errorcode.h" +#include "sandbox/linux/system_headers/linux_filter.h" +#include "testing/gtest/include/gtest/gtest.h" + +#define CASES SANDBOX_BPF_DSL_CASES + +namespace sandbox { +namespace bpf_dsl { +namespace { + +// Helper function to construct fake arch_seccomp_data objects. +struct arch_seccomp_data FakeSyscall(int nr, + uint64_t p0 = 0, + uint64_t p1 = 0, + uint64_t p2 = 0, + uint64_t p3 = 0, + uint64_t p4 = 0, + uint64_t p5 = 0) { + // Made up program counter for syscall address. + const uint64_t kFakePC = 0x543210; + + struct arch_seccomp_data data = { + nr, + SECCOMP_ARCH, + kFakePC, + { + p0, p1, p2, p3, p4, p5, + }, + }; + + return data; +} + +class FakeTrapRegistry : public TrapRegistry { + public: + FakeTrapRegistry() : map_() {} + virtual ~FakeTrapRegistry() {} + + uint16_t Add(TrapFnc fnc, const void* aux, bool safe) override { + EXPECT_TRUE(safe); + + const uint16_t next_id = map_.size() + 1; + return map_.insert(std::make_pair(Key(fnc, aux), next_id)).first->second; + } + + bool EnableUnsafeTraps() override { + ADD_FAILURE() << "Unimplemented"; + return false; + } + + private: + using Key = std::pair<TrapFnc, const void*>; + + std::map<Key, uint16_t> map_; + + DISALLOW_COPY_AND_ASSIGN(FakeTrapRegistry); +}; + +intptr_t FakeTrapFuncOne(const arch_seccomp_data& data, void* aux) { return 1; } +intptr_t FakeTrapFuncTwo(const arch_seccomp_data& data, void* aux) { return 2; } + +// Test that FakeTrapRegistry correctly assigns trap IDs to trap handlers. +TEST(FakeTrapRegistry, TrapIDs) { + struct { + TrapRegistry::TrapFnc fnc; + const void* aux; + } funcs[] = { + {FakeTrapFuncOne, nullptr}, + {FakeTrapFuncTwo, nullptr}, + {FakeTrapFuncOne, funcs}, + {FakeTrapFuncTwo, funcs}, + }; + + FakeTrapRegistry traps; + + // Add traps twice to test that IDs are reused correctly. + for (int i = 0; i < 2; ++i) { + for (size_t j = 0; j < arraysize(funcs); ++j) { + // Trap IDs start at 1. + EXPECT_EQ(j + 1, traps.Add(funcs[j].fnc, funcs[j].aux, true)); + } + } +} + +class PolicyEmulator { + public: + explicit PolicyEmulator(const Policy* policy) : program_(), traps_() { + program_ = *PolicyCompiler(policy, &traps_).Compile(true /* verify */); + } + ~PolicyEmulator() {} + + uint32_t Emulate(const struct arch_seccomp_data& data) const { + const char* err = nullptr; + uint32_t res = Verifier::EvaluateBPF(program_, data, &err); + if (err) { + ADD_FAILURE() << err; + return 0; + } + return res; + } + + void ExpectAllow(const struct arch_seccomp_data& data) const { + EXPECT_EQ(SECCOMP_RET_ALLOW, Emulate(data)); + } + + void ExpectErrno(uint16_t err, const struct arch_seccomp_data& data) const { + EXPECT_EQ(SECCOMP_RET_ERRNO | err, Emulate(data)); + } + + private: + CodeGen::Program program_; + FakeTrapRegistry traps_; + + DISALLOW_COPY_AND_ASSIGN(PolicyEmulator); +}; + +class BasicPolicy : public Policy { + public: + BasicPolicy() {} + ~BasicPolicy() override {} + ResultExpr EvaluateSyscall(int sysno) const override { + if (sysno == __NR_getpgid) { + const Arg<pid_t> pid(0); + return If(pid == 0, Error(EPERM)).Else(Error(EINVAL)); + } + if (sysno == __NR_setuid) { + const Arg<uid_t> uid(0); + return If(uid != 42, Error(ESRCH)).Else(Error(ENOMEM)); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(BasicPolicy); +}; + +TEST(BPFDSL, Basic) { + BasicPolicy policy; + PolicyEmulator emulator(&policy); + + emulator.ExpectErrno(EPERM, FakeSyscall(__NR_getpgid, 0)); + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_getpgid, 1)); + + emulator.ExpectErrno(ENOMEM, FakeSyscall(__NR_setuid, 42)); + emulator.ExpectErrno(ESRCH, FakeSyscall(__NR_setuid, 43)); +} + +/* On IA-32, socketpair() is implemented via socketcall(). :-( */ +#if !defined(ARCH_CPU_X86) +class BooleanLogicPolicy : public Policy { + public: + BooleanLogicPolicy() {} + ~BooleanLogicPolicy() override {} + ResultExpr EvaluateSyscall(int sysno) const override { + if (sysno == __NR_socketpair) { + const Arg<int> domain(0), type(1), protocol(2); + return If(domain == AF_UNIX && + (type == SOCK_STREAM || type == SOCK_DGRAM) && + protocol == 0, + Error(EPERM)).Else(Error(EINVAL)); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(BooleanLogicPolicy); +}; + +TEST(BPFDSL, BooleanLogic) { + BooleanLogicPolicy policy; + PolicyEmulator emulator(&policy); + + const intptr_t kFakeSV = 0x12345; + + // Acceptable combinations that should return EPERM. + emulator.ExpectErrno( + EPERM, FakeSyscall(__NR_socketpair, AF_UNIX, SOCK_STREAM, 0, kFakeSV)); + emulator.ExpectErrno( + EPERM, FakeSyscall(__NR_socketpair, AF_UNIX, SOCK_DGRAM, 0, kFakeSV)); + + // Combinations that are invalid for only one reason; should return EINVAL. + emulator.ExpectErrno( + EINVAL, FakeSyscall(__NR_socketpair, AF_INET, SOCK_STREAM, 0, kFakeSV)); + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_socketpair, AF_UNIX, + SOCK_SEQPACKET, 0, kFakeSV)); + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_socketpair, AF_UNIX, + SOCK_STREAM, IPPROTO_TCP, kFakeSV)); + + // Completely unacceptable combination; should also return EINVAL. + emulator.ExpectErrno( + EINVAL, FakeSyscall(__NR_socketpair, AF_INET, SOCK_SEQPACKET, IPPROTO_UDP, + kFakeSV)); +} +#endif // !ARCH_CPU_X86 + +class MoreBooleanLogicPolicy : public Policy { + public: + MoreBooleanLogicPolicy() {} + ~MoreBooleanLogicPolicy() override {} + ResultExpr EvaluateSyscall(int sysno) const override { + if (sysno == __NR_setresuid) { + const Arg<uid_t> ruid(0), euid(1), suid(2); + return If(ruid == 0 || euid == 0 || suid == 0, Error(EPERM)) + .ElseIf(ruid == 1 && euid == 1 && suid == 1, Error(EAGAIN)) + .Else(Error(EINVAL)); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(MoreBooleanLogicPolicy); +}; + +TEST(BPFDSL, MoreBooleanLogic) { + MoreBooleanLogicPolicy policy; + PolicyEmulator emulator(&policy); + + // Expect EPERM if any set to 0. + emulator.ExpectErrno(EPERM, FakeSyscall(__NR_setresuid, 0, 5, 5)); + emulator.ExpectErrno(EPERM, FakeSyscall(__NR_setresuid, 5, 0, 5)); + emulator.ExpectErrno(EPERM, FakeSyscall(__NR_setresuid, 5, 5, 0)); + + // Expect EAGAIN if all set to 1. + emulator.ExpectErrno(EAGAIN, FakeSyscall(__NR_setresuid, 1, 1, 1)); + + // Expect EINVAL for anything else. + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_setresuid, 5, 1, 1)); + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_setresuid, 1, 5, 1)); + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_setresuid, 1, 1, 5)); + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_setresuid, 3, 4, 5)); +} + +static const uintptr_t kDeadBeefAddr = + static_cast<uintptr_t>(0xdeadbeefdeadbeefULL); + +class ArgSizePolicy : public Policy { + public: + ArgSizePolicy() {} + ~ArgSizePolicy() override {} + ResultExpr EvaluateSyscall(int sysno) const override { + if (sysno == __NR_uname) { + const Arg<uintptr_t> addr(0); + return If(addr == kDeadBeefAddr, Error(EPERM)).Else(Allow()); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(ArgSizePolicy); +}; + +TEST(BPFDSL, ArgSizeTest) { + ArgSizePolicy policy; + PolicyEmulator emulator(&policy); + + emulator.ExpectAllow(FakeSyscall(__NR_uname, 0)); + emulator.ExpectErrno(EPERM, FakeSyscall(__NR_uname, kDeadBeefAddr)); +} + +#if 0 +// TODO(mdempsky): This is really an integration test. + +class TrappingPolicy : public Policy { + public: + TrappingPolicy() {} + ~TrappingPolicy() override {} + ResultExpr EvaluateSyscall(int sysno) const override { + if (sysno == __NR_uname) { + return Trap(UnameTrap, &count_); + } + return Allow(); + } + + private: + static intptr_t count_; + + static intptr_t UnameTrap(const struct arch_seccomp_data& data, void* aux) { + BPF_ASSERT_EQ(&count_, aux); + return ++count_; + } + + DISALLOW_COPY_AND_ASSIGN(TrappingPolicy); +}; + +intptr_t TrappingPolicy::count_; + +BPF_TEST_C(BPFDSL, TrapTest, TrappingPolicy) { + ASSERT_SYSCALL_RESULT(1, uname, NULL); + ASSERT_SYSCALL_RESULT(2, uname, NULL); + ASSERT_SYSCALL_RESULT(3, uname, NULL); +} +#endif + +class MaskingPolicy : public Policy { + public: + MaskingPolicy() {} + ~MaskingPolicy() override {} + ResultExpr EvaluateSyscall(int sysno) const override { + if (sysno == __NR_setuid) { + const Arg<uid_t> uid(0); + return If((uid & 0xf) == 0, Error(EINVAL)).Else(Error(EACCES)); + } + if (sysno == __NR_setgid) { + const Arg<gid_t> gid(0); + return If((gid & 0xf0) == 0xf0, Error(EINVAL)).Else(Error(EACCES)); + } + if (sysno == __NR_setpgid) { + const Arg<pid_t> pid(0); + return If((pid & 0xa5) == 0xa0, Error(EINVAL)).Else(Error(EACCES)); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(MaskingPolicy); +}; + +TEST(BPFDSL, MaskTest) { + MaskingPolicy policy; + PolicyEmulator emulator(&policy); + + for (uid_t uid = 0; uid < 0x100; ++uid) { + const int expect_errno = (uid & 0xf) == 0 ? EINVAL : EACCES; + emulator.ExpectErrno(expect_errno, FakeSyscall(__NR_setuid, uid)); + } + + for (gid_t gid = 0; gid < 0x100; ++gid) { + const int expect_errno = (gid & 0xf0) == 0xf0 ? EINVAL : EACCES; + emulator.ExpectErrno(expect_errno, FakeSyscall(__NR_setgid, gid)); + } + + for (pid_t pid = 0; pid < 0x100; ++pid) { + const int expect_errno = (pid & 0xa5) == 0xa0 ? EINVAL : EACCES; + emulator.ExpectErrno(expect_errno, FakeSyscall(__NR_setpgid, pid, 0)); + } +} + +class ElseIfPolicy : public Policy { + public: + ElseIfPolicy() {} + ~ElseIfPolicy() override {} + ResultExpr EvaluateSyscall(int sysno) const override { + if (sysno == __NR_setuid) { + const Arg<uid_t> uid(0); + return If((uid & 0xfff) == 0, Error(0)) + .ElseIf((uid & 0xff0) == 0, Error(EINVAL)) + .ElseIf((uid & 0xf00) == 0, Error(EEXIST)) + .Else(Error(EACCES)); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(ElseIfPolicy); +}; + +TEST(BPFDSL, ElseIfTest) { + ElseIfPolicy policy; + PolicyEmulator emulator(&policy); + + emulator.ExpectErrno(0, FakeSyscall(__NR_setuid, 0)); + + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_setuid, 0x0001)); + emulator.ExpectErrno(EINVAL, FakeSyscall(__NR_setuid, 0x0002)); + + emulator.ExpectErrno(EEXIST, FakeSyscall(__NR_setuid, 0x0011)); + emulator.ExpectErrno(EEXIST, FakeSyscall(__NR_setuid, 0x0022)); + + emulator.ExpectErrno(EACCES, FakeSyscall(__NR_setuid, 0x0111)); + emulator.ExpectErrno(EACCES, FakeSyscall(__NR_setuid, 0x0222)); +} + +class SwitchPolicy : public Policy { + public: + SwitchPolicy() {} + ~SwitchPolicy() override {} + ResultExpr EvaluateSyscall(int sysno) const override { + if (sysno == __NR_fcntl) { + const Arg<int> cmd(1); + const Arg<unsigned long> long_arg(2); + return Switch(cmd) + .CASES((F_GETFL, F_GETFD), Error(ENOENT)) + .Case(F_SETFD, If(long_arg == O_CLOEXEC, Allow()).Else(Error(EINVAL))) + .Case(F_SETFL, Error(EPERM)) + .Default(Error(EACCES)); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(SwitchPolicy); +}; + +TEST(BPFDSL, SwitchTest) { + SwitchPolicy policy; + PolicyEmulator emulator(&policy); + + const int kFakeSockFD = 42; + + emulator.ExpectErrno(ENOENT, FakeSyscall(__NR_fcntl, kFakeSockFD, F_GETFD)); + emulator.ExpectErrno(ENOENT, FakeSyscall(__NR_fcntl, kFakeSockFD, F_GETFL)); + + emulator.ExpectAllow( + FakeSyscall(__NR_fcntl, kFakeSockFD, F_SETFD, O_CLOEXEC)); + emulator.ExpectErrno(EINVAL, + FakeSyscall(__NR_fcntl, kFakeSockFD, F_SETFD, 0)); + + emulator.ExpectErrno(EPERM, + FakeSyscall(__NR_fcntl, kFakeSockFD, F_SETFL, O_RDONLY)); + + emulator.ExpectErrno(EACCES, + FakeSyscall(__NR_fcntl, kFakeSockFD, F_DUPFD, 0)); +} + +static intptr_t DummyTrap(const struct arch_seccomp_data& data, void* aux) { + return 0; +} + +TEST(BPFDSL, IsAllowDeny) { + ResultExpr allow = Allow(); + EXPECT_TRUE(allow->IsAllow()); + EXPECT_FALSE(allow->IsDeny()); + + ResultExpr error = Error(ENOENT); + EXPECT_FALSE(error->IsAllow()); + EXPECT_TRUE(error->IsDeny()); + + ResultExpr trace = Trace(42); + EXPECT_FALSE(trace->IsAllow()); + EXPECT_FALSE(trace->IsDeny()); + + ResultExpr trap = Trap(DummyTrap, nullptr); + EXPECT_FALSE(trap->IsAllow()); + EXPECT_TRUE(trap->IsDeny()); + + const Arg<int> arg(0); + ResultExpr maybe = If(arg == 0, Allow()).Else(Error(EPERM)); + EXPECT_FALSE(maybe->IsAllow()); + EXPECT_FALSE(maybe->IsDeny()); +} + +TEST(BPFDSL, HasUnsafeTraps) { + ResultExpr allow = Allow(); + EXPECT_FALSE(allow->HasUnsafeTraps()); + + ResultExpr safe = Trap(DummyTrap, nullptr); + EXPECT_FALSE(safe->HasUnsafeTraps()); + + ResultExpr unsafe = UnsafeTrap(DummyTrap, nullptr); + EXPECT_TRUE(unsafe->HasUnsafeTraps()); + + const Arg<int> arg(0); + ResultExpr maybe = If(arg == 0, allow).Else(unsafe); + EXPECT_TRUE(maybe->HasUnsafeTraps()); +} + +} // namespace +} // namespace bpf_dsl +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/codegen.cc b/sandbox/linux/bpf_dsl/codegen.cc new file mode 100644 index 0000000000..2d5c8e406e --- /dev/null +++ b/sandbox/linux/bpf_dsl/codegen.cc @@ -0,0 +1,159 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/codegen.h" + +#include <limits> +#include <utility> + +#include "base/logging.h" +#include "sandbox/linux/system_headers/linux_filter.h" + +// This CodeGen implementation strives for simplicity while still +// generating acceptable BPF programs under typical usage patterns +// (e.g., by PolicyCompiler). +// +// The key to its simplicity is that BPF programs only support forward +// jumps/branches, which allows constraining the DAG construction API +// to make instruction nodes immutable. Immutable nodes admits a +// simple greedy approach of emitting new instructions as needed and +// then reusing existing ones that have already been emitted. This +// cleanly avoids any need to compute basic blocks or apply +// topological sorting because the API effectively sorts instructions +// for us (e.g., before MakeInstruction() can be called to emit a +// branch instruction, it must have already been called for each +// branch path). +// +// This greedy algorithm is not without (theoretical) weakness though: +// +// 1. In the general case, we don't eliminate dead code. If needed, +// we could trace back through the program in Compile() and elide +// any unneeded instructions, but in practice we only emit live +// instructions anyway. +// +// 2. By not dividing instructions into basic blocks and sorting, we +// lose an opportunity to move non-branch/non-return instructions +// adjacent to their successor instructions, which means we might +// need to emit additional jumps. But in practice, they'll +// already be nearby as long as callers don't go out of their way +// to interleave MakeInstruction() calls for unrelated code +// sequences. + +namespace sandbox { + +// kBranchRange is the maximum value that can be stored in +// sock_filter's 8-bit jt and jf fields. +const size_t kBranchRange = std::numeric_limits<uint8_t>::max(); + +const CodeGen::Node CodeGen::kNullNode; + +CodeGen::CodeGen() : program_(), equivalent_(), memos_() { +} + +CodeGen::~CodeGen() { +} + +void CodeGen::Compile(CodeGen::Node head, Program* out) { + DCHECK(out); + out->assign(program_.rbegin() + Offset(head), program_.rend()); +} + +CodeGen::Node CodeGen::MakeInstruction(uint16_t code, + uint32_t k, + Node jt, + Node jf) { + // To avoid generating redundant code sequences, we memoize the + // results from AppendInstruction(). + auto res = memos_.insert(std::make_pair(MemoKey(code, k, jt, jf), kNullNode)); + CodeGen::Node* node = &res.first->second; + if (res.second) { // Newly inserted memo entry. + *node = AppendInstruction(code, k, jt, jf); + } + return *node; +} + +CodeGen::Node CodeGen::AppendInstruction(uint16_t code, + uint32_t k, + Node jt, + Node jf) { + if (BPF_CLASS(code) == BPF_JMP) { + CHECK_NE(BPF_JA, BPF_OP(code)) << "CodeGen inserts JAs as needed"; + + // Optimally adding jumps is rather tricky, so we use a quick + // approximation: by artificially reducing |jt|'s range, |jt| will + // stay within its true range even if we add a jump for |jf|. + jt = WithinRange(jt, kBranchRange - 1); + jf = WithinRange(jf, kBranchRange); + return Append(code, k, Offset(jt), Offset(jf)); + } + + CHECK_EQ(kNullNode, jf) << "Non-branch instructions shouldn't provide jf"; + if (BPF_CLASS(code) == BPF_RET) { + CHECK_EQ(kNullNode, jt) << "Return instructions shouldn't provide jt"; + } else { + // For non-branch/non-return instructions, execution always + // proceeds to the next instruction; so we need to arrange for + // that to be |jt|. + jt = WithinRange(jt, 0); + CHECK_EQ(0U, Offset(jt)) << "ICE: Failed to setup next instruction"; + } + return Append(code, k, 0, 0); +} + +CodeGen::Node CodeGen::WithinRange(Node target, size_t range) { + // Just use |target| if it's already within range. + if (Offset(target) <= range) { + return target; + } + + // Alternatively, look for an equivalent instruction within range. + if (Offset(equivalent_.at(target)) <= range) { + return equivalent_.at(target); + } + + // Otherwise, fall back to emitting a jump instruction. + Node jump = Append(BPF_JMP | BPF_JA, Offset(target), 0, 0); + equivalent_.at(target) = jump; + return jump; +} + +CodeGen::Node CodeGen::Append(uint16_t code, uint32_t k, size_t jt, size_t jf) { + if (BPF_CLASS(code) == BPF_JMP && BPF_OP(code) != BPF_JA) { + CHECK_LE(jt, kBranchRange); + CHECK_LE(jf, kBranchRange); + } else { + CHECK_EQ(0U, jt); + CHECK_EQ(0U, jf); + } + + CHECK_LT(program_.size(), static_cast<size_t>(BPF_MAXINSNS)); + CHECK_EQ(program_.size(), equivalent_.size()); + + Node res = program_.size(); + program_.push_back(sock_filter{ + code, static_cast<uint8_t>(jt), static_cast<uint8_t>(jf), k}); + equivalent_.push_back(res); + return res; +} + +size_t CodeGen::Offset(Node target) const { + CHECK_LT(target, program_.size()) << "Bogus offset target node"; + return (program_.size() - 1) - target; +} + +// TODO(mdempsky): Move into a general base::Tuple helper library. +bool CodeGen::MemoKeyLess::operator()(const MemoKey& lhs, + const MemoKey& rhs) const { + if (base::get<0>(lhs) != base::get<0>(rhs)) + return base::get<0>(lhs) < base::get<0>(rhs); + if (base::get<1>(lhs) != base::get<1>(rhs)) + return base::get<1>(lhs) < base::get<1>(rhs); + if (base::get<2>(lhs) != base::get<2>(rhs)) + return base::get<2>(lhs) < base::get<2>(rhs); + if (base::get<3>(lhs) != base::get<3>(rhs)) + return base::get<3>(lhs) < base::get<3>(rhs); + return false; +} + +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/codegen.h b/sandbox/linux/bpf_dsl/codegen.h new file mode 100644 index 0000000000..9d898030b9 --- /dev/null +++ b/sandbox/linux/bpf_dsl/codegen.h @@ -0,0 +1,123 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_CODEGEN_H__ +#define SANDBOX_LINUX_BPF_DSL_CODEGEN_H__ + +#include <stddef.h> +#include <stdint.h> + +#include <map> +#include <vector> + +#include "base/macros.h" +#include "base/tuple.h" +#include "sandbox/sandbox_export.h" + +struct sock_filter; + +namespace sandbox { + +// The code generator implements a basic assembler that can convert a +// graph of BPF instructions into a well-formed array of BPF +// instructions. Most notably, it ensures that jumps are always +// forward and don't exceed the limit of 255 instructions imposed by +// the instruction set. +// +// Callers would typically create a new CodeGen object and then use it +// to build a DAG of instruction nodes. They'll eventually call +// Compile() to convert this DAG to a Program. +// +// CodeGen gen; +// CodeGen::Node allow, branch, dag; +// +// allow = +// gen.MakeInstruction(BPF_RET+BPF_K, +// ErrorCode(ErrorCode::ERR_ALLOWED).err())); +// branch = +// gen.MakeInstruction(BPF_JMP+BPF_EQ+BPF_K, __NR_getpid, +// Trap(GetPidHandler, NULL), allow); +// dag = +// gen.MakeInstruction(BPF_LD+BPF_W+BPF_ABS, +// offsetof(struct arch_seccomp_data, nr), branch); +// +// // Simplified code follows; in practice, it is important to avoid calling +// // any C++ destructors after starting the sandbox. +// CodeGen::Program program; +// gen.Compile(dag, program); +// const struct sock_fprog prog = { +// static_cast<unsigned short>(program->size()), &program[0] }; +// prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); +// +class SANDBOX_EXPORT CodeGen { + public: + // A vector of BPF instructions that need to be installed as a filter + // program in the kernel. + typedef std::vector<struct sock_filter> Program; + + // Node represents a node within the instruction DAG being compiled. + using Node = Program::size_type; + + // kNullNode represents the "null" node; i.e., the reserved node + // value guaranteed to not equal any actual nodes. + static const Node kNullNode = -1; + + CodeGen(); + ~CodeGen(); + + // MakeInstruction creates a node representing the specified + // instruction, or returns and existing equivalent node if one + // exists. For details on the possible parameters refer to + // https://www.kernel.org/doc/Documentation/networking/filter.txt. + // TODO(mdempsky): Reconsider using default arguments here. + Node MakeInstruction(uint16_t code, + uint32_t k, + Node jt = kNullNode, + Node jf = kNullNode); + + // Compile linearizes the instruction DAG rooted at |head| into a + // program that can be executed by a BPF virtual machine. + void Compile(Node head, Program* program); + + private: + using MemoKey = base::Tuple<uint16_t, uint32_t, Node, Node>; + struct MemoKeyLess { + bool operator()(const MemoKey& lhs, const MemoKey& rhs) const; + }; + + // AppendInstruction adds a new instruction, ensuring that |jt| and + // |jf| are within range as necessary for |code|. + Node AppendInstruction(uint16_t code, uint32_t k, Node jt, Node jf); + + // WithinRange returns a node equivalent to |next| that is at most + // |range| instructions away from the (logical) beginning of the + // program. + Node WithinRange(Node next, size_t range); + + // Append appends a new instruction to the physical end (i.e., + // logical beginning) of |program_|. + Node Append(uint16_t code, uint32_t k, size_t jt, size_t jf); + + // Offset returns how many instructions exist in |program_| after |target|. + size_t Offset(Node target) const; + + // NOTE: program_ is the compiled program in *reverse*, so that + // indices remain stable as we add instructions. + Program program_; + + // equivalent_ stores the most recent semantically-equivalent node for each + // instruction in program_. A node is defined as semantically-equivalent to N + // if it has the same instruction code and constant as N and its successor + // nodes (if any) are semantically-equivalent to N's successor nodes, or + // if it's an unconditional jump to a node semantically-equivalent to N. + std::vector<Node> equivalent_; + + std::map<MemoKey, Node, MemoKeyLess> memos_; + + DISALLOW_COPY_AND_ASSIGN(CodeGen); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_CODEGEN_H__ diff --git a/sandbox/linux/bpf_dsl/codegen_unittest.cc b/sandbox/linux/bpf_dsl/codegen_unittest.cc new file mode 100644 index 0000000000..5961822123 --- /dev/null +++ b/sandbox/linux/bpf_dsl/codegen_unittest.cc @@ -0,0 +1,402 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/codegen.h" + +#include <map> +#include <utility> +#include <vector> + +#include "base/macros.h" +#include "base/md5.h" +#include "base/strings/string_piece.h" +#include "sandbox/linux/system_headers/linux_filter.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { +namespace { + +// Hash provides an abstraction for building "hash trees" from BPF +// control flow graphs, and efficiently identifying equivalent graphs. +// +// For simplicity, we use MD5, because base happens to provide a +// convenient API for its use. However, any collision-resistant hash +// should suffice. +class Hash { + public: + static const Hash kZero; + + Hash() : digest_() {} + + Hash(uint16_t code, + uint32_t k, + const Hash& jt = kZero, + const Hash& jf = kZero) + : digest_() { + base::MD5Context ctx; + base::MD5Init(&ctx); + HashValue(&ctx, code); + HashValue(&ctx, k); + HashValue(&ctx, jt); + HashValue(&ctx, jf); + base::MD5Final(&digest_, &ctx); + } + + Hash(const Hash& hash) = default; + Hash& operator=(const Hash& rhs) = default; + + friend bool operator==(const Hash& lhs, const Hash& rhs) { + return lhs.Base16() == rhs.Base16(); + } + friend bool operator!=(const Hash& lhs, const Hash& rhs) { + return !(lhs == rhs); + } + + private: + template <typename T> + void HashValue(base::MD5Context* ctx, const T& value) { + base::MD5Update(ctx, + base::StringPiece(reinterpret_cast<const char*>(&value), + sizeof(value))); + } + + std::string Base16() const { + return base::MD5DigestToBase16(digest_); + } + + base::MD5Digest digest_; +}; + +const Hash Hash::kZero; + +// Sanity check that equality and inequality work on Hash as required. +TEST(CodeGen, HashSanity) { + std::vector<Hash> hashes; + + // Push a bunch of logically distinct hashes. + hashes.push_back(Hash::kZero); + for (int i = 0; i < 4; ++i) { + hashes.push_back(Hash(i & 1, i & 2)); + } + for (int i = 0; i < 16; ++i) { + hashes.push_back(Hash(i & 1, i & 2, Hash(i & 4, i & 8))); + } + for (int i = 0; i < 64; ++i) { + hashes.push_back( + Hash(i & 1, i & 2, Hash(i & 4, i & 8), Hash(i & 16, i & 32))); + } + + for (const Hash& a : hashes) { + for (const Hash& b : hashes) { + // Hashes should equal themselves, but not equal all others. + if (&a == &b) { + EXPECT_EQ(a, b); + } else { + EXPECT_NE(a, b); + } + } + } +} + +// ProgramTest provides a fixture for writing compiling sample +// programs with CodeGen and verifying the linearized output matches +// the input DAG. +class ProgramTest : public ::testing::Test { + protected: + ProgramTest() : gen_(), node_hashes_() {} + + // MakeInstruction calls CodeGen::MakeInstruction() and associated + // the returned address with a hash of the instruction. + CodeGen::Node MakeInstruction(uint16_t code, + uint32_t k, + CodeGen::Node jt = CodeGen::kNullNode, + CodeGen::Node jf = CodeGen::kNullNode) { + CodeGen::Node res = gen_.MakeInstruction(code, k, jt, jf); + EXPECT_NE(CodeGen::kNullNode, res); + + Hash digest(code, k, Lookup(jt), Lookup(jf)); + auto it = node_hashes_.insert(std::make_pair(res, digest)); + EXPECT_EQ(digest, it.first->second); + + return res; + } + + // RunTest compiles the program and verifies that the output matches + // what is expected. It should be called at the end of each program + // test case. + void RunTest(CodeGen::Node head) { + // Compile the program + CodeGen::Program program; + gen_.Compile(head, &program); + + // Walk the program backwards, and compute the hash for each instruction. + std::vector<Hash> prog_hashes(program.size()); + for (size_t i = program.size(); i > 0; --i) { + const sock_filter& insn = program.at(i - 1); + Hash& hash = prog_hashes.at(i - 1); + + if (BPF_CLASS(insn.code) == BPF_JMP) { + if (BPF_OP(insn.code) == BPF_JA) { + // The compiler adds JA instructions as needed, so skip them. + hash = prog_hashes.at(i + insn.k); + } else { + hash = Hash(insn.code, insn.k, prog_hashes.at(i + insn.jt), + prog_hashes.at(i + insn.jf)); + } + } else if (BPF_CLASS(insn.code) == BPF_RET) { + hash = Hash(insn.code, insn.k); + } else { + hash = Hash(insn.code, insn.k, prog_hashes.at(i)); + } + } + + EXPECT_EQ(Lookup(head), prog_hashes.at(0)); + } + + private: + const Hash& Lookup(CodeGen::Node next) const { + if (next == CodeGen::kNullNode) { + return Hash::kZero; + } + auto it = node_hashes_.find(next); + if (it == node_hashes_.end()) { + ADD_FAILURE() << "No hash found for node " << next; + return Hash::kZero; + } + return it->second; + } + + CodeGen gen_; + std::map<CodeGen::Node, Hash> node_hashes_; + + DISALLOW_COPY_AND_ASSIGN(ProgramTest); +}; + +TEST_F(ProgramTest, OneInstruction) { + // Create the most basic valid BPF program: + // RET 0 + CodeGen::Node head = MakeInstruction(BPF_RET + BPF_K, 0); + RunTest(head); +} + +TEST_F(ProgramTest, SimpleBranch) { + // Create a program with a single branch: + // JUMP if eq 42 then $0 else $1 + // 0: RET 1 + // 1: RET 0 + CodeGen::Node head = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 42, + MakeInstruction(BPF_RET + BPF_K, 1), + MakeInstruction(BPF_RET + BPF_K, 0)); + RunTest(head); +} + +TEST_F(ProgramTest, AtypicalBranch) { + // Create a program with a single branch: + // JUMP if eq 42 then $0 else $0 + // 0: RET 0 + + CodeGen::Node ret = MakeInstruction(BPF_RET + BPF_K, 0); + CodeGen::Node head = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 42, ret, ret); + + // N.B.: As the instructions in both sides of the branch are already + // the same object, we do not actually have any "mergeable" branches. + // This needs to be reflected in our choice of "flags". + RunTest(head); +} + +TEST_F(ProgramTest, Complex) { + // Creates a basic BPF program that we'll use to test some of the code: + // JUMP if eq 42 the $0 else $1 (insn6) + // 0: LD 23 (insn5) + // 1: JUMP if eq 42 then $2 else $4 (insn4) + // 2: JUMP to $3 (insn2) + // 3: LD 42 (insn1) + // RET 42 (insn0) + // 4: LD 42 (insn3) + // RET 42 (insn3+) + CodeGen::Node insn0 = MakeInstruction(BPF_RET + BPF_K, 42); + CodeGen::Node insn1 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 42, insn0); + CodeGen::Node insn2 = insn1; // Implicit JUMP + + // We explicitly duplicate instructions to test that they're merged. + CodeGen::Node insn3 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 42, + MakeInstruction(BPF_RET + BPF_K, 42)); + EXPECT_EQ(insn2, insn3); + + CodeGen::Node insn4 = + MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 42, insn2, insn3); + CodeGen::Node insn5 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 23, insn4); + + // Force a basic block that ends in neither a jump instruction nor a return + // instruction. It only contains "insn5". This exercises one of the less + // common code paths in the topo-sort algorithm. + // This also gives us a diamond-shaped pattern in our graph, which stresses + // another aspect of the topo-sort algorithm (namely, the ability to + // correctly count the incoming branches for subtrees that are not disjunct). + CodeGen::Node insn6 = + MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 42, insn5, insn4); + + RunTest(insn6); +} + +TEST_F(ProgramTest, ConfusingTails) { + // This simple program demonstrates https://crbug.com/351103/ + // The two "LOAD 0" instructions are blocks of their own. MergeTails() could + // be tempted to merge them since they are the same. However, they are + // not mergeable because they fall-through to non semantically equivalent + // blocks. + // Without the fix for this bug, this program should trigger the check in + // CompileAndCompare: the serialized graphs from the program and its compiled + // version will differ. + // + // 0) LOAD 1 // ??? + // 1) if A == 0x1; then JMP 2 else JMP 3 + // 2) LOAD 0 // System call number + // 3) if A == 0x2; then JMP 4 else JMP 5 + // 4) LOAD 0 // System call number + // 5) if A == 0x1; then JMP 6 else JMP 7 + // 6) RET 0 + // 7) RET 1 + + CodeGen::Node i7 = MakeInstruction(BPF_RET + BPF_K, 1); + CodeGen::Node i6 = MakeInstruction(BPF_RET + BPF_K, 0); + CodeGen::Node i5 = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 1, i6, i7); + CodeGen::Node i4 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 0, i5); + CodeGen::Node i3 = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 2, i4, i5); + CodeGen::Node i2 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 0, i3); + CodeGen::Node i1 = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 1, i2, i3); + CodeGen::Node i0 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 1, i1); + + RunTest(i0); +} + +TEST_F(ProgramTest, ConfusingTailsBasic) { + // Without the fix for https://crbug.com/351103/, (see + // SampleProgramConfusingTails()), this would generate a cyclic graph and + // crash as the two "LOAD 0" instructions would get merged. + // + // 0) LOAD 1 // ??? + // 1) if A == 0x1; then JMP 2 else JMP 3 + // 2) LOAD 0 // System call number + // 3) if A == 0x2; then JMP 4 else JMP 5 + // 4) LOAD 0 // System call number + // 5) RET 1 + + CodeGen::Node i5 = MakeInstruction(BPF_RET + BPF_K, 1); + CodeGen::Node i4 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 0, i5); + CodeGen::Node i3 = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 2, i4, i5); + CodeGen::Node i2 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 0, i3); + CodeGen::Node i1 = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 1, i2, i3); + CodeGen::Node i0 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 1, i1); + + RunTest(i0); +} + +TEST_F(ProgramTest, ConfusingTailsMergeable) { + // This is similar to SampleProgramConfusingTails(), except that + // instructions 2 and 4 are now RET instructions. + // In PointerCompare(), this exercises the path where two blocks are of the + // same length and identical and the last instruction is a JMP or RET, so the + // following blocks don't need to be looked at and the blocks are mergeable. + // + // 0) LOAD 1 // ??? + // 1) if A == 0x1; then JMP 2 else JMP 3 + // 2) RET 42 + // 3) if A == 0x2; then JMP 4 else JMP 5 + // 4) RET 42 + // 5) if A == 0x1; then JMP 6 else JMP 7 + // 6) RET 0 + // 7) RET 1 + + CodeGen::Node i7 = MakeInstruction(BPF_RET + BPF_K, 1); + CodeGen::Node i6 = MakeInstruction(BPF_RET + BPF_K, 0); + CodeGen::Node i5 = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 1, i6, i7); + CodeGen::Node i4 = MakeInstruction(BPF_RET + BPF_K, 42); + CodeGen::Node i3 = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 2, i4, i5); + CodeGen::Node i2 = MakeInstruction(BPF_RET + BPF_K, 42); + CodeGen::Node i1 = MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 1, i2, i3); + CodeGen::Node i0 = MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 1, i1); + + RunTest(i0); +} + +TEST_F(ProgramTest, InstructionFolding) { + // Check that simple instructions are folded as expected. + CodeGen::Node a = MakeInstruction(BPF_RET + BPF_K, 0); + EXPECT_EQ(a, MakeInstruction(BPF_RET + BPF_K, 0)); + CodeGen::Node b = MakeInstruction(BPF_RET + BPF_K, 1); + EXPECT_EQ(a, MakeInstruction(BPF_RET + BPF_K, 0)); + EXPECT_EQ(b, MakeInstruction(BPF_RET + BPF_K, 1)); + EXPECT_EQ(b, MakeInstruction(BPF_RET + BPF_K, 1)); + + // Check that complex sequences are folded too. + CodeGen::Node c = + MakeInstruction(BPF_LD + BPF_W + BPF_ABS, 0, + MakeInstruction(BPF_JMP + BPF_JSET + BPF_K, 0x100, a, b)); + EXPECT_EQ(c, MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, 0, + MakeInstruction(BPF_JMP + BPF_JSET + BPF_K, 0x100, a, b))); + + RunTest(c); +} + +TEST_F(ProgramTest, FarBranches) { + // BPF instructions use 8-bit fields for branch offsets, which means + // branch targets must be within 255 instructions of the branch + // instruction. CodeGen abstracts away this detail by inserting jump + // instructions as needed, which we test here by generating programs + // that should trigger any interesting boundary conditions. + + // Populate with 260 initial instruction nodes. + std::vector<CodeGen::Node> nodes; + nodes.push_back(MakeInstruction(BPF_RET + BPF_K, 0)); + for (size_t i = 1; i < 260; ++i) { + nodes.push_back( + MakeInstruction(BPF_ALU + BPF_ADD + BPF_K, i, nodes.back())); + } + + // Exhaustively test branch offsets near BPF's limits. + for (size_t jt = 250; jt < 260; ++jt) { + for (size_t jf = 250; jf < 260; ++jf) { + nodes.push_back(MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 0, + nodes.rbegin()[jt], nodes.rbegin()[jf])); + RunTest(nodes.back()); + } + } +} + +TEST_F(ProgramTest, JumpReuse) { + // As a code size optimization, we try to reuse jumps when possible + // instead of emitting new ones. Here we make sure that optimization + // is working as intended. + // + // NOTE: To simplify testing, we rely on implementation details + // about what CodeGen::Node values indicate (i.e., vector indices), + // but CodeGen users should treat them as opaque values. + + // Populate with 260 initial instruction nodes. + std::vector<CodeGen::Node> nodes; + nodes.push_back(MakeInstruction(BPF_RET + BPF_K, 0)); + for (size_t i = 1; i < 260; ++i) { + nodes.push_back( + MakeInstruction(BPF_ALU + BPF_ADD + BPF_K, i, nodes.back())); + } + + // Branching to nodes[0] and nodes[1] should require 3 new + // instructions: two far jumps plus the branch itself. + CodeGen::Node one = + MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 0, nodes[0], nodes[1]); + EXPECT_EQ(nodes.back() + 3, one); // XXX: Implementation detail! + RunTest(one); + + // Branching again to the same target nodes should require only one + // new instruction, as we can reuse the previous branch's jumps. + CodeGen::Node two = + MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, 1, nodes[0], nodes[1]); + EXPECT_EQ(one + 1, two); // XXX: Implementation detail! + RunTest(two); +} + +} // namespace +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/cons.h b/sandbox/linux/bpf_dsl/cons.h new file mode 100644 index 0000000000..fa47c140ff --- /dev/null +++ b/sandbox/linux/bpf_dsl/cons.h @@ -0,0 +1,138 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_CONS_H_ +#define SANDBOX_LINUX_BPF_DSL_CONS_H_ + +#include "base/memory/ref_counted.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { +namespace cons { + +// Namespace cons provides an abstraction for immutable "cons list" +// data structures as commonly provided in functional programming +// languages like Lisp or Haskell. +// +// A cons list is a linked list consisting of "cells", each of which +// have a "head" and a "tail" element. A cell's head element contains +// a user specified value, while the tail element contains a (possibly +// null) pointer to another cell. +// +// An empty list (idiomatically referred to as "nil") can be +// constructed as "cons::List<Foo>()" or simply as "nullptr" if Foo +// can be inferred from context (e.g., calling a function that has a +// "cons::List<Foo>" parameter). +// +// Existing lists (including empty lists) can be extended by +// prepending new values to the front using the "Cons(head, tail)" +// function, which will allocate a new cons cell. Notably, cons lists +// support creating multiple lists that share a common tail sequence. +// +// Lastly, lists support iteration via C++11's range-based for loop +// construct. +// +// Examples: +// +// // basic construction +// const cons::List<char> kNil = nullptr; +// cons::List<char> ba = Cons('b', Cons('a', kNil)); +// +// // common tail sequence +// cons::List<char> cba = Cons('c', ba); +// cons::List<char> dba = Cons('d', ba); +// +// // iteration +// for (const char& ch : cba) { +// // iterates 'c', 'b', 'a' +// } +// for (const char& ch : dba) { +// // iterates 'd', 'b', 'a' +// } + +// Forward declarations. +template <typename T> +class Cell; +template <typename T> +class ListIterator; + +// List represents a (possibly null) pointer to a cons cell. +template <typename T> +using List = scoped_refptr<const Cell<T>>; + +// Cons extends a cons list by prepending a new value to the front. +template <typename T> +List<T> Cons(const T& head, const List<T>& tail) { + return List<T>(new const Cell<T>(head, tail)); +} + +// Cell represents an individual "cons cell" within a cons list. +template <typename T> +class Cell : public base::RefCounted<Cell<T>> { + public: + Cell(const T& head, const List<T>& tail) : head_(head), tail_(tail) {} + + // Head returns this cell's head element. + const T& head() const { return head_; } + + // Tail returns this cell's tail element. + const List<T>& tail() const { return tail_; } + + private: + virtual ~Cell() {} + + T head_; + List<T> tail_; + + friend class base::RefCounted<Cell<T>>; + DISALLOW_COPY_AND_ASSIGN(Cell); +}; + +// Begin returns a list iterator pointing to the first element of the +// cons list. It's provided to support range-based for loops. +template <typename T> +ListIterator<T> begin(const List<T>& list) { + return ListIterator<T>(list); +} + +// End returns a list iterator pointing to the "past-the-end" element +// of the cons list (i.e., nil). It's provided to support range-based +// for loops. +template <typename T> +ListIterator<T> end(const List<T>& list) { + return ListIterator<T>(); +} + +// ListIterator provides C++ forward iterator semantics for traversing +// a cons list. +template <typename T> +class ListIterator { + public: + ListIterator() : list_() {} + explicit ListIterator(const List<T>& list) : list_(list) {} + + const T& operator*() const { return list_->head(); } + + ListIterator& operator++() { + list_ = list_->tail(); + return *this; + } + + friend bool operator==(const ListIterator& lhs, const ListIterator& rhs) { + return lhs.list_ == rhs.list_; + } + + private: + List<T> list_; +}; + +template <typename T> +bool operator!=(const ListIterator<T>& lhs, const ListIterator<T>& rhs) { + return !(lhs == rhs); +} + +} // namespace cons +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_CONS_H_ diff --git a/sandbox/linux/bpf_dsl/cons_unittest.cc b/sandbox/linux/bpf_dsl/cons_unittest.cc new file mode 100644 index 0000000000..ea2ba2f8dc --- /dev/null +++ b/sandbox/linux/bpf_dsl/cons_unittest.cc @@ -0,0 +1,33 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/cons.h" + +#include <string> + +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { +namespace { + +std::string Join(cons::List<char> char_list) { + std::string res; + for (const char& ch : char_list) { + res.push_back(ch); + } + return res; +} + +TEST(ConsListTest, Basic) { + cons::List<char> ba = Cons('b', Cons('a', cons::List<char>())); + EXPECT_EQ("ba", Join(ba)); + + cons::List<char> cba = Cons('c', ba); + cons::List<char> dba = Cons('d', ba); + EXPECT_EQ("cba", Join(cba)); + EXPECT_EQ("dba", Join(dba)); +} + +} // namespace +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/dump_bpf.cc b/sandbox/linux/bpf_dsl/dump_bpf.cc new file mode 100644 index 0000000000..d0c8f75073 --- /dev/null +++ b/sandbox/linux/bpf_dsl/dump_bpf.cc @@ -0,0 +1,109 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/dump_bpf.h" + +#include <stdio.h> + +#include "sandbox/linux/bpf_dsl/codegen.h" +#include "sandbox/linux/bpf_dsl/trap_registry.h" +#include "sandbox/linux/system_headers/linux_filter.h" +#include "sandbox/linux/system_headers/linux_seccomp.h" + +namespace sandbox { +namespace bpf_dsl { + +void DumpBPF::PrintProgram(const CodeGen::Program& program) { + for (CodeGen::Program::const_iterator iter = program.begin(); + iter != program.end(); + ++iter) { + int ip = (int)(iter - program.begin()); + fprintf(stderr, "%3d) ", ip); + switch (BPF_CLASS(iter->code)) { + case BPF_LD: + if (iter->code == BPF_LD + BPF_W + BPF_ABS) { + fprintf(stderr, "LOAD %d // ", (int)iter->k); + if (iter->k == offsetof(struct arch_seccomp_data, nr)) { + fprintf(stderr, "System call number\n"); + } else if (iter->k == offsetof(struct arch_seccomp_data, arch)) { + fprintf(stderr, "Architecture\n"); + } else if (iter->k == + offsetof(struct arch_seccomp_data, instruction_pointer)) { + fprintf(stderr, "Instruction pointer (LSB)\n"); + } else if (iter->k == + offsetof(struct arch_seccomp_data, instruction_pointer) + + 4) { + fprintf(stderr, "Instruction pointer (MSB)\n"); + } else if (iter->k >= offsetof(struct arch_seccomp_data, args) && + iter->k < offsetof(struct arch_seccomp_data, args) + 48 && + (iter->k - offsetof(struct arch_seccomp_data, args)) % 4 == + 0) { + fprintf( + stderr, + "Argument %d (%cSB)\n", + (int)(iter->k - offsetof(struct arch_seccomp_data, args)) / 8, + (iter->k - offsetof(struct arch_seccomp_data, args)) % 8 ? 'M' + : 'L'); + } else { + fprintf(stderr, "???\n"); + } + } else { + fprintf(stderr, "LOAD ???\n"); + } + break; + case BPF_JMP: + if (BPF_OP(iter->code) == BPF_JA) { + fprintf(stderr, "JMP %d\n", ip + iter->k + 1); + } else { + fprintf(stderr, "if A %s 0x%x; then JMP %d else JMP %d\n", + BPF_OP(iter->code) == BPF_JSET ? "&" : + BPF_OP(iter->code) == BPF_JEQ ? "==" : + BPF_OP(iter->code) == BPF_JGE ? ">=" : + BPF_OP(iter->code) == BPF_JGT ? ">" : "???", + (int)iter->k, + ip + iter->jt + 1, ip + iter->jf + 1); + } + break; + case BPF_RET: + fprintf(stderr, "RET 0x%x // ", iter->k); + if ((iter->k & SECCOMP_RET_ACTION) == SECCOMP_RET_TRAP) { + fprintf(stderr, "Trap #%d\n", iter->k & SECCOMP_RET_DATA); + } else if ((iter->k & SECCOMP_RET_ACTION) == SECCOMP_RET_ERRNO) { + fprintf(stderr, "errno = %d\n", iter->k & SECCOMP_RET_DATA); + } else if ((iter->k & SECCOMP_RET_ACTION) == SECCOMP_RET_TRACE) { + fprintf(stderr, "Trace #%d\n", iter->k & SECCOMP_RET_DATA); + } else if (iter->k == SECCOMP_RET_ALLOW) { + fprintf(stderr, "Allowed\n"); + } else { + fprintf(stderr, "???\n"); + } + break; + case BPF_ALU: + if (BPF_OP(iter->code) == BPF_NEG) { + fprintf(stderr, "A := -A\n"); + } else { + fprintf(stderr, "A := A %s 0x%x\n", + BPF_OP(iter->code) == BPF_ADD ? "+" : + BPF_OP(iter->code) == BPF_SUB ? "-" : + BPF_OP(iter->code) == BPF_MUL ? "*" : + BPF_OP(iter->code) == BPF_DIV ? "/" : + BPF_OP(iter->code) == BPF_MOD ? "%" : + BPF_OP(iter->code) == BPF_OR ? "|" : + BPF_OP(iter->code) == BPF_XOR ? "^" : + BPF_OP(iter->code) == BPF_AND ? "&" : + BPF_OP(iter->code) == BPF_LSH ? "<<" : + BPF_OP(iter->code) == BPF_RSH ? ">>" : "???", + (int)iter->k); + } + break; + default: + fprintf(stderr, "???\n"); + break; + } + } + return; +} + +} // namespace bpf_dsl +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/dump_bpf.h b/sandbox/linux/bpf_dsl/dump_bpf.h new file mode 100644 index 0000000000..cd12be793d --- /dev/null +++ b/sandbox/linux/bpf_dsl/dump_bpf.h @@ -0,0 +1,18 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/codegen.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { +namespace bpf_dsl { + +class SANDBOX_EXPORT DumpBPF { + public: + // PrintProgram writes |program| in a human-readable format to stderr. + static void PrintProgram(const CodeGen::Program& program); +}; + +} // namespace bpf_dsl +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/linux_syscall_ranges.h b/sandbox/linux/bpf_dsl/linux_syscall_ranges.h new file mode 100644 index 0000000000..a747770c78 --- /dev/null +++ b/sandbox/linux/bpf_dsl/linux_syscall_ranges.h @@ -0,0 +1,57 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_LINUX_SYSCALL_RANGES_H_ +#define SANDBOX_LINUX_BPF_DSL_LINUX_SYSCALL_RANGES_H_ + +#if defined(__x86_64__) + +#define MIN_SYSCALL 0u +#define MAX_PUBLIC_SYSCALL 1024u +#define MAX_SYSCALL MAX_PUBLIC_SYSCALL + +#elif defined(__i386__) + +#define MIN_SYSCALL 0u +#define MAX_PUBLIC_SYSCALL 1024u +#define MAX_SYSCALL MAX_PUBLIC_SYSCALL + +#elif defined(__arm__) && (defined(__thumb__) || defined(__ARM_EABI__)) + +// ARM EABI includes "ARM private" system calls starting at |__ARM_NR_BASE|, +// and a "ghost syscall private to the kernel", cmpxchg, +// at |__ARM_NR_BASE+0x00fff0|. +// See </arch/arm/include/asm/unistd.h> in the Linux kernel. + +// __NR_SYSCALL_BASE is 0 in thumb and ARM EABI. +#define MIN_SYSCALL 0u +#define MAX_PUBLIC_SYSCALL (MIN_SYSCALL + 1024u) +// __ARM_NR_BASE is __NR_SYSCALL_BASE + 0xf0000u +#define MIN_PRIVATE_SYSCALL 0xf0000u +#define MAX_PRIVATE_SYSCALL (MIN_PRIVATE_SYSCALL + 16u) +#define MIN_GHOST_SYSCALL (MIN_PRIVATE_SYSCALL + 0xfff0u) +#define MAX_SYSCALL (MIN_GHOST_SYSCALL + 4u) + +#elif defined(__mips__) && (_MIPS_SIM == _ABIO32) + +#include <asm/unistd.h> // for __NR_O32_Linux and __NR_Linux_syscalls +#define MIN_SYSCALL __NR_O32_Linux +#define MAX_PUBLIC_SYSCALL (MIN_SYSCALL + __NR_Linux_syscalls) +#define MAX_SYSCALL MAX_PUBLIC_SYSCALL + +#elif defined(__mips__) && (_MIPS_SIM == _ABI64) + +#error "Add support to header file" + +#elif defined(__aarch64__) + +#define MIN_SYSCALL 0u +#define MAX_PUBLIC_SYSCALL 279u +#define MAX_SYSCALL MAX_PUBLIC_SYSCALL + +#else +#error "Unsupported architecture" +#endif + +#endif // SANDBOX_LINUX_BPF_DSL_LINUX_SYSCALL_RANGES_H_ diff --git a/sandbox/linux/bpf_dsl/policy.cc b/sandbox/linux/bpf_dsl/policy.cc new file mode 100644 index 0000000000..c20edc6da8 --- /dev/null +++ b/sandbox/linux/bpf_dsl/policy.cc @@ -0,0 +1,19 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/policy.h" + +#include <errno.h> + +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" + +namespace sandbox { +namespace bpf_dsl { + +ResultExpr Policy::InvalidSyscall() const { + return Error(ENOSYS); +} + +} // namespace bpf_dsl +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/policy.h b/sandbox/linux/bpf_dsl/policy.h new file mode 100644 index 0000000000..6c67589456 --- /dev/null +++ b/sandbox/linux/bpf_dsl/policy.h @@ -0,0 +1,37 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_POLICY_H_ +#define SANDBOX_LINUX_BPF_DSL_POLICY_H_ + +#include "base/macros.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_forward.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { +namespace bpf_dsl { + +// Interface to implement to define a BPF sandbox policy. +class SANDBOX_EXPORT Policy { + public: + Policy() {} + virtual ~Policy() {} + + // User extension point for writing custom sandbox policies. + // The returned ResultExpr will control how the kernel responds to the + // specified system call number. + virtual ResultExpr EvaluateSyscall(int sysno) const = 0; + + // Optional overload for specifying alternate behavior for invalid + // system calls. The default is to return ENOSYS. + virtual ResultExpr InvalidSyscall() const; + + private: + DISALLOW_COPY_AND_ASSIGN(Policy); +}; + +} // namespace bpf_dsl +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_POLICY_H_ diff --git a/sandbox/linux/bpf_dsl/policy_compiler.cc b/sandbox/linux/bpf_dsl/policy_compiler.cc new file mode 100644 index 0000000000..f38232f85f --- /dev/null +++ b/sandbox/linux/bpf_dsl/policy_compiler.cc @@ -0,0 +1,499 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/policy_compiler.h" + +#include <errno.h> +#include <sys/syscall.h> + +#include <limits> + +#include "base/logging.h" +#include "base/macros.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_impl.h" +#include "sandbox/linux/bpf_dsl/codegen.h" +#include "sandbox/linux/bpf_dsl/dump_bpf.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" +#include "sandbox/linux/bpf_dsl/syscall_set.h" +#include "sandbox/linux/bpf_dsl/verifier.h" +#include "sandbox/linux/seccomp-bpf/errorcode.h" +#include "sandbox/linux/system_headers/linux_filter.h" +#include "sandbox/linux/system_headers/linux_seccomp.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" + +namespace sandbox { +namespace bpf_dsl { + +namespace { + +#if defined(__i386__) || defined(__x86_64__) +const bool kIsIntel = true; +#else +const bool kIsIntel = false; +#endif +#if defined(__x86_64__) && defined(__ILP32__) +const bool kIsX32 = true; +#else +const bool kIsX32 = false; +#endif + +const int kSyscallsRequiredForUnsafeTraps[] = { + __NR_rt_sigprocmask, + __NR_rt_sigreturn, +#if defined(__NR_sigprocmask) + __NR_sigprocmask, +#endif +#if defined(__NR_sigreturn) + __NR_sigreturn, +#endif +}; + +bool HasExactlyOneBit(uint64_t x) { + // Common trick; e.g., see http://stackoverflow.com/a/108329. + return x != 0 && (x & (x - 1)) == 0; +} + +// A Trap() handler that returns an "errno" value. The value is encoded +// in the "aux" parameter. +intptr_t ReturnErrno(const struct arch_seccomp_data&, void* aux) { + // TrapFnc functions report error by following the native kernel convention + // of returning an exit code in the range of -1..-4096. They do not try to + // set errno themselves. The glibc wrapper that triggered the SIGSYS will + // ultimately do so for us. + int err = reinterpret_cast<intptr_t>(aux) & SECCOMP_RET_DATA; + return -err; +} + +bool HasUnsafeTraps(const Policy* policy) { + DCHECK(policy); + for (uint32_t sysnum : SyscallSet::ValidOnly()) { + if (policy->EvaluateSyscall(sysnum)->HasUnsafeTraps()) { + return true; + } + } + return policy->InvalidSyscall()->HasUnsafeTraps(); +} + +} // namespace + +struct PolicyCompiler::Range { + uint32_t from; + CodeGen::Node node; +}; + +PolicyCompiler::PolicyCompiler(const Policy* policy, TrapRegistry* registry) + : policy_(policy), + registry_(registry), + escapepc_(0), + conds_(), + gen_(), + has_unsafe_traps_(HasUnsafeTraps(policy_)) { + DCHECK(policy); +} + +PolicyCompiler::~PolicyCompiler() { +} + +scoped_ptr<CodeGen::Program> PolicyCompiler::Compile(bool verify) { + CHECK(policy_->InvalidSyscall()->IsDeny()) + << "Policies should deny invalid system calls"; + + // If our BPF program has unsafe traps, enable support for them. + if (has_unsafe_traps_) { + CHECK_NE(0U, escapepc_) << "UnsafeTrap() requires a valid escape PC"; + + for (int sysnum : kSyscallsRequiredForUnsafeTraps) { + CHECK(policy_->EvaluateSyscall(sysnum)->IsAllow()) + << "Policies that use UnsafeTrap() must unconditionally allow all " + "required system calls"; + } + + CHECK(registry_->EnableUnsafeTraps()) + << "We'd rather die than enable unsafe traps"; + } + + // Assemble the BPF filter program. + scoped_ptr<CodeGen::Program> program(new CodeGen::Program()); + gen_.Compile(AssemblePolicy(), program.get()); + + // Make sure compilation resulted in a BPF program that executes + // correctly. Otherwise, there is an internal error in our BPF compiler. + // There is really nothing the caller can do until the bug is fixed. + if (verify) { + const char* err = nullptr; + if (!Verifier::VerifyBPF(this, *program, *policy_, &err)) { + DumpBPF::PrintProgram(*program); + LOG(FATAL) << err; + } + } + + return program.Pass(); +} + +void PolicyCompiler::DangerousSetEscapePC(uint64_t escapepc) { + escapepc_ = escapepc; +} + +CodeGen::Node PolicyCompiler::AssemblePolicy() { + // A compiled policy consists of three logical parts: + // 1. Check that the "arch" field matches the expected architecture. + // 2. If the policy involves unsafe traps, check if the syscall was + // invoked by Syscall::Call, and then allow it unconditionally. + // 3. Check the system call number and jump to the appropriate compiled + // system call policy number. + return CheckArch(MaybeAddEscapeHatch(DispatchSyscall())); +} + +CodeGen::Node PolicyCompiler::CheckArch(CodeGen::Node passed) { + // If the architecture doesn't match SECCOMP_ARCH, disallow the + // system call. + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, SECCOMP_ARCH_IDX, + gen_.MakeInstruction( + BPF_JMP + BPF_JEQ + BPF_K, SECCOMP_ARCH, passed, + CompileResult(Kill("Invalid audit architecture in BPF filter")))); +} + +CodeGen::Node PolicyCompiler::MaybeAddEscapeHatch(CodeGen::Node rest) { + // If no unsafe traps, then simply return |rest|. + if (!has_unsafe_traps_) { + return rest; + } + + // We already enabled unsafe traps in Compile, but enable them again to give + // the trap registry a second chance to complain before we add the backdoor. + CHECK(registry_->EnableUnsafeTraps()); + + // Allow system calls, if they originate from our magic return address. + const uint32_t lopc = static_cast<uint32_t>(escapepc_); + const uint32_t hipc = static_cast<uint32_t>(escapepc_ >> 32); + + // BPF cannot do native 64-bit comparisons, so we have to compare + // both 32-bit halves of the instruction pointer. If they match what + // we expect, we return ERR_ALLOWED. If either or both don't match, + // we continue evalutating the rest of the sandbox policy. + // + // For simplicity, we check the full 64-bit instruction pointer even + // on 32-bit architectures. + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, SECCOMP_IP_LSB_IDX, + gen_.MakeInstruction( + BPF_JMP + BPF_JEQ + BPF_K, lopc, + gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, SECCOMP_IP_MSB_IDX, + gen_.MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, hipc, + CompileResult(Allow()), rest)), + rest)); +} + +CodeGen::Node PolicyCompiler::DispatchSyscall() { + // Evaluate all possible system calls and group their ErrorCodes into + // ranges of identical codes. + Ranges ranges; + FindRanges(&ranges); + + // Compile the system call ranges to an optimized BPF jumptable + CodeGen::Node jumptable = AssembleJumpTable(ranges.begin(), ranges.end()); + + // Grab the system call number, so that we can check it and then + // execute the jump table. + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, SECCOMP_NR_IDX, CheckSyscallNumber(jumptable)); +} + +CodeGen::Node PolicyCompiler::CheckSyscallNumber(CodeGen::Node passed) { + if (kIsIntel) { + // On Intel architectures, verify that system call numbers are in the + // expected number range. + CodeGen::Node invalidX32 = + CompileResult(Kill("Illegal mixing of system call ABIs")); + if (kIsX32) { + // The newer x32 API always sets bit 30. + return gen_.MakeInstruction( + BPF_JMP + BPF_JSET + BPF_K, 0x40000000, passed, invalidX32); + } else { + // The older i386 and x86-64 APIs clear bit 30 on all system calls. + return gen_.MakeInstruction( + BPF_JMP + BPF_JSET + BPF_K, 0x40000000, invalidX32, passed); + } + } + + // TODO(mdempsky): Similar validation for other architectures? + return passed; +} + +void PolicyCompiler::FindRanges(Ranges* ranges) { + // Please note that "struct seccomp_data" defines system calls as a signed + // int32_t, but BPF instructions always operate on unsigned quantities. We + // deal with this disparity by enumerating from MIN_SYSCALL to MAX_SYSCALL, + // and then verifying that the rest of the number range (both positive and + // negative) all return the same ErrorCode. + const CodeGen::Node invalid_node = CompileResult(policy_->InvalidSyscall()); + uint32_t old_sysnum = 0; + CodeGen::Node old_node = + SyscallSet::IsValid(old_sysnum) + ? CompileResult(policy_->EvaluateSyscall(old_sysnum)) + : invalid_node; + + for (uint32_t sysnum : SyscallSet::All()) { + CodeGen::Node node = + SyscallSet::IsValid(sysnum) + ? CompileResult(policy_->EvaluateSyscall(static_cast<int>(sysnum))) + : invalid_node; + // N.B., here we rely on CodeGen folding (i.e., returning the same + // node value for) identical code sequences, otherwise our jump + // table will blow up in size. + if (node != old_node) { + ranges->push_back(Range{old_sysnum, old_node}); + old_sysnum = sysnum; + old_node = node; + } + } + ranges->push_back(Range{old_sysnum, old_node}); +} + +CodeGen::Node PolicyCompiler::AssembleJumpTable(Ranges::const_iterator start, + Ranges::const_iterator stop) { + // We convert the list of system call ranges into jump table that performs + // a binary search over the ranges. + // As a sanity check, we need to have at least one distinct ranges for us + // to be able to build a jump table. + CHECK(start < stop) << "Invalid iterator range"; + const auto n = stop - start; + if (n == 1) { + // If we have narrowed things down to a single range object, we can + // return from the BPF filter program. + return start->node; + } + + // Pick the range object that is located at the mid point of our list. + // We compare our system call number against the lowest valid system call + // number in this range object. If our number is lower, it is outside of + // this range object. If it is greater or equal, it might be inside. + Ranges::const_iterator mid = start + n / 2; + + // Sub-divide the list of ranges and continue recursively. + CodeGen::Node jf = AssembleJumpTable(start, mid); + CodeGen::Node jt = AssembleJumpTable(mid, stop); + return gen_.MakeInstruction(BPF_JMP + BPF_JGE + BPF_K, mid->from, jt, jf); +} + +CodeGen::Node PolicyCompiler::CompileResult(const ResultExpr& res) { + return RetExpression(res->Compile(this)); +} + +CodeGen::Node PolicyCompiler::RetExpression(const ErrorCode& err) { + switch (err.error_type()) { + case ErrorCode::ET_COND: + return CondExpression(err); + case ErrorCode::ET_SIMPLE: + case ErrorCode::ET_TRAP: + return gen_.MakeInstruction(BPF_RET + BPF_K, err.err()); + default: + LOG(FATAL) + << "ErrorCode is not suitable for returning from a BPF program"; + return CodeGen::kNullNode; + } +} + +CodeGen::Node PolicyCompiler::CondExpression(const ErrorCode& cond) { + // Sanity check that |cond| makes sense. + CHECK(cond.argno_ >= 0 && cond.argno_ < 6) << "Invalid argument number " + << cond.argno_; + CHECK(cond.width_ == ErrorCode::TP_32BIT || + cond.width_ == ErrorCode::TP_64BIT) + << "Invalid argument width " << cond.width_; + CHECK_NE(0U, cond.mask_) << "Zero mask is invalid"; + CHECK_EQ(cond.value_, cond.value_ & cond.mask_) + << "Value contains masked out bits"; + if (sizeof(void*) == 4) { + CHECK_EQ(ErrorCode::TP_32BIT, cond.width_) + << "Invalid width on 32-bit platform"; + } + if (cond.width_ == ErrorCode::TP_32BIT) { + CHECK_EQ(0U, cond.mask_ >> 32) << "Mask exceeds argument size"; + CHECK_EQ(0U, cond.value_ >> 32) << "Value exceeds argument size"; + } + + CodeGen::Node passed = RetExpression(*cond.passed_); + CodeGen::Node failed = RetExpression(*cond.failed_); + + // We want to emit code to check "(arg & mask) == value" where arg, mask, and + // value are 64-bit values, but the BPF machine is only 32-bit. We implement + // this by independently testing the upper and lower 32-bits and continuing to + // |passed| if both evaluate true, or to |failed| if either evaluate false. + return CondExpressionHalf(cond, + UpperHalf, + CondExpressionHalf(cond, LowerHalf, passed, failed), + failed); +} + +CodeGen::Node PolicyCompiler::CondExpressionHalf(const ErrorCode& cond, + ArgHalf half, + CodeGen::Node passed, + CodeGen::Node failed) { + if (cond.width_ == ErrorCode::TP_32BIT && half == UpperHalf) { + // Special logic for sanity checking the upper 32-bits of 32-bit system + // call arguments. + + // TODO(mdempsky): Compile Unexpected64bitArgument() just per program. + CodeGen::Node invalid_64bit = RetExpression(Unexpected64bitArgument()); + + const uint32_t upper = SECCOMP_ARG_MSB_IDX(cond.argno_); + const uint32_t lower = SECCOMP_ARG_LSB_IDX(cond.argno_); + + if (sizeof(void*) == 4) { + // On 32-bit platforms, the upper 32-bits should always be 0: + // LDW [upper] + // JEQ 0, passed, invalid + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, + upper, + gen_.MakeInstruction( + BPF_JMP + BPF_JEQ + BPF_K, 0, passed, invalid_64bit)); + } + + // On 64-bit platforms, the upper 32-bits may be 0 or ~0; but we only allow + // ~0 if the sign bit of the lower 32-bits is set too: + // LDW [upper] + // JEQ 0, passed, (next) + // JEQ ~0, (next), invalid + // LDW [lower] + // JSET (1<<31), passed, invalid + // + // TODO(mdempsky): The JSET instruction could perhaps jump to passed->next + // instead, as the first instruction of passed should be "LDW [lower]". + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, + upper, + gen_.MakeInstruction( + BPF_JMP + BPF_JEQ + BPF_K, + 0, + passed, + gen_.MakeInstruction( + BPF_JMP + BPF_JEQ + BPF_K, + std::numeric_limits<uint32_t>::max(), + gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, + lower, + gen_.MakeInstruction(BPF_JMP + BPF_JSET + BPF_K, + 1U << 31, + passed, + invalid_64bit)), + invalid_64bit))); + } + + const uint32_t idx = (half == UpperHalf) ? SECCOMP_ARG_MSB_IDX(cond.argno_) + : SECCOMP_ARG_LSB_IDX(cond.argno_); + const uint32_t mask = (half == UpperHalf) ? cond.mask_ >> 32 : cond.mask_; + const uint32_t value = (half == UpperHalf) ? cond.value_ >> 32 : cond.value_; + + // Emit a suitable instruction sequence for (arg & mask) == value. + + // For (arg & 0) == 0, just return passed. + if (mask == 0) { + CHECK_EQ(0U, value); + return passed; + } + + // For (arg & ~0) == value, emit: + // LDW [idx] + // JEQ value, passed, failed + if (mask == std::numeric_limits<uint32_t>::max()) { + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, + idx, + gen_.MakeInstruction(BPF_JMP + BPF_JEQ + BPF_K, value, passed, failed)); + } + + // For (arg & mask) == 0, emit: + // LDW [idx] + // JSET mask, failed, passed + // (Note: failed and passed are intentionally swapped.) + if (value == 0) { + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, + idx, + gen_.MakeInstruction(BPF_JMP + BPF_JSET + BPF_K, mask, failed, passed)); + } + + // For (arg & x) == x where x is a single-bit value, emit: + // LDW [idx] + // JSET mask, passed, failed + if (mask == value && HasExactlyOneBit(mask)) { + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, + idx, + gen_.MakeInstruction(BPF_JMP + BPF_JSET + BPF_K, mask, passed, failed)); + } + + // Generic fallback: + // LDW [idx] + // AND mask + // JEQ value, passed, failed + return gen_.MakeInstruction( + BPF_LD + BPF_W + BPF_ABS, + idx, + gen_.MakeInstruction( + BPF_ALU + BPF_AND + BPF_K, + mask, + gen_.MakeInstruction( + BPF_JMP + BPF_JEQ + BPF_K, value, passed, failed))); +} + +ErrorCode PolicyCompiler::Unexpected64bitArgument() { + return Kill("Unexpected 64bit argument detected")->Compile(this); +} + +ErrorCode PolicyCompiler::Error(int err) { + if (has_unsafe_traps_) { + // When inside an UnsafeTrap() callback, we want to allow all system calls. + // This means, we must conditionally disable the sandbox -- and that's not + // something that kernel-side BPF filters can do, as they cannot inspect + // any state other than the syscall arguments. + // But if we redirect all error handlers to user-space, then we can easily + // make this decision. + // The performance penalty for this extra round-trip to user-space is not + // actually that bad, as we only ever pay it for denied system calls; and a + // typical program has very few of these. + return Trap(ReturnErrno, reinterpret_cast<void*>(err), true); + } + + return ErrorCode(err); +} + +ErrorCode PolicyCompiler::Trap(TrapRegistry::TrapFnc fnc, + const void* aux, + bool safe) { + uint16_t trap_id = registry_->Add(fnc, aux, safe); + return ErrorCode(trap_id, fnc, aux, safe); +} + +bool PolicyCompiler::IsRequiredForUnsafeTrap(int sysno) { + for (size_t i = 0; i < arraysize(kSyscallsRequiredForUnsafeTraps); ++i) { + if (sysno == kSyscallsRequiredForUnsafeTraps[i]) { + return true; + } + } + return false; +} + +ErrorCode PolicyCompiler::CondMaskedEqual(int argno, + ErrorCode::ArgType width, + uint64_t mask, + uint64_t value, + const ErrorCode& passed, + const ErrorCode& failed) { + return ErrorCode(argno, + width, + mask, + value, + &*conds_.insert(passed).first, + &*conds_.insert(failed).first); +} + +} // namespace bpf_dsl +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/policy_compiler.h b/sandbox/linux/bpf_dsl/policy_compiler.h new file mode 100644 index 0000000000..df38d4ccbc --- /dev/null +++ b/sandbox/linux/bpf_dsl/policy_compiler.h @@ -0,0 +1,159 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_POLICY_COMPILER_H_ +#define SANDBOX_LINUX_BPF_DSL_POLICY_COMPILER_H_ + +#include <stdint.h> + +#include <map> +#include <set> +#include <vector> + +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_forward.h" +#include "sandbox/linux/bpf_dsl/codegen.h" +#include "sandbox/linux/seccomp-bpf/errorcode.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { +namespace bpf_dsl { +class Policy; + +// PolicyCompiler implements the bpf_dsl compiler, allowing users to +// transform bpf_dsl policies into BPF programs to be executed by the +// Linux kernel. +class SANDBOX_EXPORT PolicyCompiler { + public: + PolicyCompiler(const Policy* policy, TrapRegistry* registry); + ~PolicyCompiler(); + + // Compile registers any trap handlers needed by the policy and + // compiles the policy to a BPF program, which it returns. + scoped_ptr<CodeGen::Program> Compile(bool verify); + + // DangerousSetEscapePC sets the "escape PC" that is allowed to issue any + // system calls, regardless of policy. + void DangerousSetEscapePC(uint64_t escapepc); + + // Error returns an ErrorCode to indicate the system call should fail with + // the specified error number. + ErrorCode Error(int err); + + // Trap returns an ErrorCode to indicate the system call should + // instead invoke a trap handler. + ErrorCode Trap(TrapRegistry::TrapFnc fnc, const void* aux, bool safe); + + // UnsafeTraps require some syscalls to always be allowed. + // This helper function returns true for these calls. + static bool IsRequiredForUnsafeTrap(int sysno); + + // We can also use ErrorCode to request evaluation of a conditional + // statement based on inspection of system call parameters. + // This method wrap an ErrorCode object around the conditional statement. + // Argument "argno" (1..6) will be bitwise-AND'd with "mask" and compared + // to "value"; if equal, then "passed" will be returned, otherwise "failed". + // If "is32bit" is set, the argument must in the range of 0x0..(1u << 32 - 1) + // If it is outside this range, the sandbox treats the system call just + // the same as any other ABI violation (i.e. it aborts with an error + // message). + ErrorCode CondMaskedEqual(int argno, + ErrorCode::ArgType is_32bit, + uint64_t mask, + uint64_t value, + const ErrorCode& passed, + const ErrorCode& failed); + + // Returns the fatal ErrorCode that is used to indicate that somebody + // attempted to pass a 64bit value in a 32bit system call argument. + // This method is primarily needed for testing purposes. + ErrorCode Unexpected64bitArgument(); + + private: + struct Range; + typedef std::vector<Range> Ranges; + typedef std::set<ErrorCode, struct ErrorCode::LessThan> Conds; + + // Used by CondExpressionHalf to track which half of the argument it's + // emitting instructions for. + enum ArgHalf { + LowerHalf, + UpperHalf, + }; + + // Compile the configured policy into a complete instruction sequence. + CodeGen::Node AssemblePolicy(); + + // Return an instruction sequence that checks the + // arch_seccomp_data's "arch" field is valid, and then passes + // control to |passed| if so. + CodeGen::Node CheckArch(CodeGen::Node passed); + + // If |has_unsafe_traps_| is true, returns an instruction sequence + // that allows all system calls from |escapepc_|, and otherwise + // passes control to |rest|. Otherwise, simply returns |rest|. + CodeGen::Node MaybeAddEscapeHatch(CodeGen::Node rest); + + // Return an instruction sequence that loads and checks the system + // call number, performs a binary search, and then dispatches to an + // appropriate instruction sequence compiled from the current + // policy. + CodeGen::Node DispatchSyscall(); + + // Return an instruction sequence that checks the system call number + // (expected to be loaded in register A) and if valid, passes + // control to |passed| (with register A still valid). + CodeGen::Node CheckSyscallNumber(CodeGen::Node passed); + + // Finds all the ranges of system calls that need to be handled. Ranges are + // sorted in ascending order of system call numbers. There are no gaps in the + // ranges. System calls with identical ErrorCodes are coalesced into a single + // range. + void FindRanges(Ranges* ranges); + + // Returns a BPF program snippet that implements a jump table for the + // given range of system call numbers. This function runs recursively. + CodeGen::Node AssembleJumpTable(Ranges::const_iterator start, + Ranges::const_iterator stop); + + // CompileResult compiles an individual result expression into a + // CodeGen node. + CodeGen::Node CompileResult(const ResultExpr& res); + + // Returns a BPF program snippet that makes the BPF filter program exit + // with the given ErrorCode "err". N.B. the ErrorCode may very well be a + // conditional expression; if so, this function will recursively call + // CondExpression() and possibly RetExpression() to build a complex set of + // instructions. + CodeGen::Node RetExpression(const ErrorCode& err); + + // Returns a BPF program that evaluates the conditional expression in + // "cond" and returns the appropriate value from the BPF filter program. + // This function recursively calls RetExpression(); it should only ever be + // called from RetExpression(). + CodeGen::Node CondExpression(const ErrorCode& cond); + + // Returns a BPF program that evaluates half of a conditional expression; + // it should only ever be called from CondExpression(). + CodeGen::Node CondExpressionHalf(const ErrorCode& cond, + ArgHalf half, + CodeGen::Node passed, + CodeGen::Node failed); + + const Policy* policy_; + TrapRegistry* registry_; + uint64_t escapepc_; + + Conds conds_; + CodeGen gen_; + bool has_unsafe_traps_; + + DISALLOW_COPY_AND_ASSIGN(PolicyCompiler); +}; + +} // namespace bpf_dsl +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_POLICY_COMPILER_H_ diff --git a/sandbox/linux/bpf_dsl/seccomp_macros.h b/sandbox/linux/bpf_dsl/seccomp_macros.h new file mode 100644 index 0000000000..ca28c1d7cd --- /dev/null +++ b/sandbox/linux/bpf_dsl/seccomp_macros.h @@ -0,0 +1,295 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_SECCOMP_MACROS_H_ +#define SANDBOX_LINUX_BPF_DSL_SECCOMP_MACROS_H_ + +#include <sys/cdefs.h> +// Old Bionic versions do not have sys/user.h. The if can be removed once we no +// longer need to support these old Bionic versions. +// All x86_64 builds use a new enough bionic to have sys/user.h. +#if !defined(__BIONIC__) || defined(__x86_64__) +#include <sys/types.h> // Fix for gcc 4.7, make sure __uint16_t is defined. +#if !defined(__native_client_nonsfi__) +#include <sys/user.h> +#endif +#if defined(__mips__) +// sys/user.h in eglibc misses size_t definition +#include <stddef.h> +#endif +#endif + +#include "sandbox/linux/system_headers/linux_seccomp.h" // For AUDIT_ARCH_* + +// Impose some reasonable maximum BPF program size. Realistically, the +// kernel probably has much lower limits. But by limiting to less than +// 30 bits, we can ease requirements on some of our data types. +#define SECCOMP_MAX_PROGRAM_SIZE (1<<30) + +#if defined(__i386__) +#define SECCOMP_ARCH AUDIT_ARCH_I386 + +#define SECCOMP_REG(_ctx, _reg) ((_ctx)->uc_mcontext.gregs[(_reg)]) +#define SECCOMP_RESULT(_ctx) SECCOMP_REG(_ctx, REG_EAX) +#define SECCOMP_SYSCALL(_ctx) SECCOMP_REG(_ctx, REG_EAX) +#define SECCOMP_IP(_ctx) SECCOMP_REG(_ctx, REG_EIP) +#define SECCOMP_PARM1(_ctx) SECCOMP_REG(_ctx, REG_EBX) +#define SECCOMP_PARM2(_ctx) SECCOMP_REG(_ctx, REG_ECX) +#define SECCOMP_PARM3(_ctx) SECCOMP_REG(_ctx, REG_EDX) +#define SECCOMP_PARM4(_ctx) SECCOMP_REG(_ctx, REG_ESI) +#define SECCOMP_PARM5(_ctx) SECCOMP_REG(_ctx, REG_EDI) +#define SECCOMP_PARM6(_ctx) SECCOMP_REG(_ctx, REG_EBP) +#define SECCOMP_NR_IDX (offsetof(struct arch_seccomp_data, nr)) +#define SECCOMP_ARCH_IDX (offsetof(struct arch_seccomp_data, arch)) +#define SECCOMP_IP_MSB_IDX (offsetof(struct arch_seccomp_data, \ + instruction_pointer) + 4) +#define SECCOMP_IP_LSB_IDX (offsetof(struct arch_seccomp_data, \ + instruction_pointer) + 0) +#define SECCOMP_ARG_MSB_IDX(nr) (offsetof(struct arch_seccomp_data, args) + \ + 8*(nr) + 4) +#define SECCOMP_ARG_LSB_IDX(nr) (offsetof(struct arch_seccomp_data, args) + \ + 8*(nr) + 0) + + +#if defined(__BIONIC__) || defined(__native_client_nonsfi__) +// Old Bionic versions and PNaCl toolchain don't have sys/user.h, so we just +// define regs_struct directly. This can be removed once we no longer need to +// support these old Bionic versions and PNaCl toolchain. +struct regs_struct { + long int ebx; + long int ecx; + long int edx; + long int esi; + long int edi; + long int ebp; + long int eax; + long int xds; + long int xes; + long int xfs; + long int xgs; + long int orig_eax; + long int eip; + long int xcs; + long int eflags; + long int esp; + long int xss; +}; +#else +typedef user_regs_struct regs_struct; +#endif + +#define SECCOMP_PT_RESULT(_regs) (_regs).eax +#define SECCOMP_PT_SYSCALL(_regs) (_regs).orig_eax +#define SECCOMP_PT_IP(_regs) (_regs).eip +#define SECCOMP_PT_PARM1(_regs) (_regs).ebx +#define SECCOMP_PT_PARM2(_regs) (_regs).ecx +#define SECCOMP_PT_PARM3(_regs) (_regs).edx +#define SECCOMP_PT_PARM4(_regs) (_regs).esi +#define SECCOMP_PT_PARM5(_regs) (_regs).edi +#define SECCOMP_PT_PARM6(_regs) (_regs).ebp + +#elif defined(__x86_64__) +#define SECCOMP_ARCH AUDIT_ARCH_X86_64 + +#define SECCOMP_REG(_ctx, _reg) ((_ctx)->uc_mcontext.gregs[(_reg)]) +#define SECCOMP_RESULT(_ctx) SECCOMP_REG(_ctx, REG_RAX) +#define SECCOMP_SYSCALL(_ctx) SECCOMP_REG(_ctx, REG_RAX) +#define SECCOMP_IP(_ctx) SECCOMP_REG(_ctx, REG_RIP) +#define SECCOMP_PARM1(_ctx) SECCOMP_REG(_ctx, REG_RDI) +#define SECCOMP_PARM2(_ctx) SECCOMP_REG(_ctx, REG_RSI) +#define SECCOMP_PARM3(_ctx) SECCOMP_REG(_ctx, REG_RDX) +#define SECCOMP_PARM4(_ctx) SECCOMP_REG(_ctx, REG_R10) +#define SECCOMP_PARM5(_ctx) SECCOMP_REG(_ctx, REG_R8) +#define SECCOMP_PARM6(_ctx) SECCOMP_REG(_ctx, REG_R9) +#define SECCOMP_NR_IDX (offsetof(struct arch_seccomp_data, nr)) +#define SECCOMP_ARCH_IDX (offsetof(struct arch_seccomp_data, arch)) +#define SECCOMP_IP_MSB_IDX (offsetof(struct arch_seccomp_data, \ + instruction_pointer) + 4) +#define SECCOMP_IP_LSB_IDX (offsetof(struct arch_seccomp_data, \ + instruction_pointer) + 0) +#define SECCOMP_ARG_MSB_IDX(nr) (offsetof(struct arch_seccomp_data, args) + \ + 8*(nr) + 4) +#define SECCOMP_ARG_LSB_IDX(nr) (offsetof(struct arch_seccomp_data, args) + \ + 8*(nr) + 0) + +typedef user_regs_struct regs_struct; +#define SECCOMP_PT_RESULT(_regs) (_regs).rax +#define SECCOMP_PT_SYSCALL(_regs) (_regs).orig_rax +#define SECCOMP_PT_IP(_regs) (_regs).rip +#define SECCOMP_PT_PARM1(_regs) (_regs).rdi +#define SECCOMP_PT_PARM2(_regs) (_regs).rsi +#define SECCOMP_PT_PARM3(_regs) (_regs).rdx +#define SECCOMP_PT_PARM4(_regs) (_regs).r10 +#define SECCOMP_PT_PARM5(_regs) (_regs).r8 +#define SECCOMP_PT_PARM6(_regs) (_regs).r9 + +#elif defined(__arm__) && (defined(__thumb__) || defined(__ARM_EABI__)) +#define SECCOMP_ARCH AUDIT_ARCH_ARM + +// ARM sigcontext_t is different from i386/x86_64. +// See </arch/arm/include/asm/sigcontext.h> in the Linux kernel. +#define SECCOMP_REG(_ctx, _reg) ((_ctx)->uc_mcontext.arm_##_reg) +// ARM EABI syscall convention. +#define SECCOMP_RESULT(_ctx) SECCOMP_REG(_ctx, r0) +#define SECCOMP_SYSCALL(_ctx) SECCOMP_REG(_ctx, r7) +#define SECCOMP_IP(_ctx) SECCOMP_REG(_ctx, pc) +#define SECCOMP_PARM1(_ctx) SECCOMP_REG(_ctx, r0) +#define SECCOMP_PARM2(_ctx) SECCOMP_REG(_ctx, r1) +#define SECCOMP_PARM3(_ctx) SECCOMP_REG(_ctx, r2) +#define SECCOMP_PARM4(_ctx) SECCOMP_REG(_ctx, r3) +#define SECCOMP_PARM5(_ctx) SECCOMP_REG(_ctx, r4) +#define SECCOMP_PARM6(_ctx) SECCOMP_REG(_ctx, r5) +#define SECCOMP_NR_IDX (offsetof(struct arch_seccomp_data, nr)) +#define SECCOMP_ARCH_IDX (offsetof(struct arch_seccomp_data, arch)) +#define SECCOMP_IP_MSB_IDX (offsetof(struct arch_seccomp_data, \ + instruction_pointer) + 4) +#define SECCOMP_IP_LSB_IDX (offsetof(struct arch_seccomp_data, \ + instruction_pointer) + 0) +#define SECCOMP_ARG_MSB_IDX(nr) (offsetof(struct arch_seccomp_data, args) + \ + 8*(nr) + 4) +#define SECCOMP_ARG_LSB_IDX(nr) (offsetof(struct arch_seccomp_data, args) + \ + 8*(nr) + 0) + +#if defined(__BIONIC__) || defined(__native_client_nonsfi__) +// Old Bionic versions and PNaCl toolchain don't have sys/user.h, so we just +// define regs_struct directly. This can be removed once we no longer need to +// support these old Bionic versions and PNaCl toolchain. +struct regs_struct { + unsigned long uregs[18]; +}; +#else +typedef user_regs regs_struct; +#endif + +#define REG_cpsr uregs[16] +#define REG_pc uregs[15] +#define REG_lr uregs[14] +#define REG_sp uregs[13] +#define REG_ip uregs[12] +#define REG_fp uregs[11] +#define REG_r10 uregs[10] +#define REG_r9 uregs[9] +#define REG_r8 uregs[8] +#define REG_r7 uregs[7] +#define REG_r6 uregs[6] +#define REG_r5 uregs[5] +#define REG_r4 uregs[4] +#define REG_r3 uregs[3] +#define REG_r2 uregs[2] +#define REG_r1 uregs[1] +#define REG_r0 uregs[0] +#define REG_ORIG_r0 uregs[17] + +#define SECCOMP_PT_RESULT(_regs) (_regs).REG_r0 +#define SECCOMP_PT_SYSCALL(_regs) (_regs).REG_r7 +#define SECCOMP_PT_IP(_regs) (_regs).REG_pc +#define SECCOMP_PT_PARM1(_regs) (_regs).REG_r0 +#define SECCOMP_PT_PARM2(_regs) (_regs).REG_r1 +#define SECCOMP_PT_PARM3(_regs) (_regs).REG_r2 +#define SECCOMP_PT_PARM4(_regs) (_regs).REG_r3 +#define SECCOMP_PT_PARM5(_regs) (_regs).REG_r4 +#define SECCOMP_PT_PARM6(_regs) (_regs).REG_r5 + +#elif defined(__mips__) && (_MIPS_SIM == _MIPS_SIM_ABI32) +#define SECCOMP_ARCH AUDIT_ARCH_MIPSEL +#define SYSCALL_EIGHT_ARGS +// MIPS sigcontext_t is different from i386/x86_64 and ARM. +// See </arch/mips/include/uapi/asm/sigcontext.h> in the Linux kernel. +#define SECCOMP_REG(_ctx, _reg) ((_ctx)->uc_mcontext.gregs[_reg]) +// Based on MIPS o32 ABI syscall convention. +// On MIPS, when indirect syscall is being made (syscall(__NR_foo)), +// real identificator (__NR_foo) is not in v0, but in a0 +#define SECCOMP_RESULT(_ctx) SECCOMP_REG(_ctx, 2) +#define SECCOMP_SYSCALL(_ctx) SECCOMP_REG(_ctx, 2) +#define SECCOMP_IP(_ctx) (_ctx)->uc_mcontext.pc +#define SECCOMP_PARM1(_ctx) SECCOMP_REG(_ctx, 4) +#define SECCOMP_PARM2(_ctx) SECCOMP_REG(_ctx, 5) +#define SECCOMP_PARM3(_ctx) SECCOMP_REG(_ctx, 6) +#define SECCOMP_PARM4(_ctx) SECCOMP_REG(_ctx, 7) +// Only the first 4 arguments of syscall are in registers. +// The rest are on the stack. +#define SECCOMP_STACKPARM(_ctx, n) (((long *)SECCOMP_REG(_ctx, 29))[(n)]) +#define SECCOMP_PARM5(_ctx) SECCOMP_STACKPARM(_ctx, 4) +#define SECCOMP_PARM6(_ctx) SECCOMP_STACKPARM(_ctx, 5) +#define SECCOMP_PARM7(_ctx) SECCOMP_STACKPARM(_ctx, 6) +#define SECCOMP_PARM8(_ctx) SECCOMP_STACKPARM(_ctx, 7) +#define SECCOMP_NR_IDX (offsetof(struct arch_seccomp_data, nr)) +#define SECCOMP_ARCH_IDX (offsetof(struct arch_seccomp_data, arch)) +#define SECCOMP_IP_MSB_IDX (offsetof(struct arch_seccomp_data, \ + instruction_pointer) + 4) +#define SECCOMP_IP_LSB_IDX (offsetof(struct arch_seccomp_data, \ + instruction_pointer) + 0) +#define SECCOMP_ARG_MSB_IDX(nr) (offsetof(struct arch_seccomp_data, args) + \ + 8*(nr) + 4) +#define SECCOMP_ARG_LSB_IDX(nr) (offsetof(struct arch_seccomp_data, args) + \ + 8*(nr) + 0) + +// On Mips we don't have structures like user_regs or user_regs_struct in +// sys/user.h that we could use, so we just define regs_struct directly. +struct regs_struct { + unsigned long long regs[32]; +}; + +#define REG_a3 regs[7] +#define REG_a2 regs[6] +#define REG_a1 regs[5] +#define REG_a0 regs[4] +#define REG_v1 regs[3] +#define REG_v0 regs[2] + +#define SECCOMP_PT_RESULT(_regs) (_regs).REG_v0 +#define SECCOMP_PT_SYSCALL(_regs) (_regs).REG_v0 +#define SECCOMP_PT_PARM1(_regs) (_regs).REG_a0 +#define SECCOMP_PT_PARM2(_regs) (_regs).REG_a1 +#define SECCOMP_PT_PARM3(_regs) (_regs).REG_a2 +#define SECCOMP_PT_PARM4(_regs) (_regs).REG_a3 + +#elif defined(__aarch64__) +struct regs_struct { + unsigned long long regs[31]; + unsigned long long sp; + unsigned long long pc; + unsigned long long pstate; +}; + +#define SECCOMP_ARCH AUDIT_ARCH_AARCH64 + +#define SECCOMP_REG(_ctx, _reg) ((_ctx)->uc_mcontext.regs[_reg]) + +#define SECCOMP_RESULT(_ctx) SECCOMP_REG(_ctx, 0) +#define SECCOMP_SYSCALL(_ctx) SECCOMP_REG(_ctx, 8) +#define SECCOMP_IP(_ctx) (_ctx)->uc_mcontext.pc +#define SECCOMP_PARM1(_ctx) SECCOMP_REG(_ctx, 0) +#define SECCOMP_PARM2(_ctx) SECCOMP_REG(_ctx, 1) +#define SECCOMP_PARM3(_ctx) SECCOMP_REG(_ctx, 2) +#define SECCOMP_PARM4(_ctx) SECCOMP_REG(_ctx, 3) +#define SECCOMP_PARM5(_ctx) SECCOMP_REG(_ctx, 4) +#define SECCOMP_PARM6(_ctx) SECCOMP_REG(_ctx, 5) + +#define SECCOMP_NR_IDX (offsetof(struct arch_seccomp_data, nr)) +#define SECCOMP_ARCH_IDX (offsetof(struct arch_seccomp_data, arch)) +#define SECCOMP_IP_MSB_IDX \ + (offsetof(struct arch_seccomp_data, instruction_pointer) + 4) +#define SECCOMP_IP_LSB_IDX \ + (offsetof(struct arch_seccomp_data, instruction_pointer) + 0) +#define SECCOMP_ARG_MSB_IDX(nr) \ + (offsetof(struct arch_seccomp_data, args) + 8 * (nr) + 4) +#define SECCOMP_ARG_LSB_IDX(nr) \ + (offsetof(struct arch_seccomp_data, args) + 8 * (nr) + 0) + +#define SECCOMP_PT_RESULT(_regs) (_regs).regs[0] +#define SECCOMP_PT_SYSCALL(_regs) (_regs).regs[8] +#define SECCOMP_PT_IP(_regs) (_regs).pc +#define SECCOMP_PT_PARM1(_regs) (_regs).regs[0] +#define SECCOMP_PT_PARM2(_regs) (_regs).regs[1] +#define SECCOMP_PT_PARM3(_regs) (_regs).regs[2] +#define SECCOMP_PT_PARM4(_regs) (_regs).regs[3] +#define SECCOMP_PT_PARM5(_regs) (_regs).regs[4] +#define SECCOMP_PT_PARM6(_regs) (_regs).regs[5] +#else +#error Unsupported target platform + +#endif + +#endif // SANDBOX_LINUX_BPF_DSL_SECCOMP_MACROS_H_ diff --git a/sandbox/linux/bpf_dsl/syscall_set.cc b/sandbox/linux/bpf_dsl/syscall_set.cc new file mode 100644 index 0000000000..47810e99ac --- /dev/null +++ b/sandbox/linux/bpf_dsl/syscall_set.cc @@ -0,0 +1,144 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/syscall_set.h" + +#include "base/logging.h" +#include "base/macros.h" +#include "sandbox/linux/bpf_dsl/linux_syscall_ranges.h" + +namespace sandbox { + +namespace { + +#if defined(__mips__) && (_MIPS_SIM == _MIPS_SIM_ABI32) +// This is true for Mips O32 ABI. +static_assert(MIN_SYSCALL == __NR_Linux, "min syscall number should be 4000"); +#else +// This true for supported architectures (Intel and ARM EABI). +static_assert(MIN_SYSCALL == 0u, + "min syscall should always be zero"); +#endif + +// SyscallRange represents an inclusive range of system call numbers. +struct SyscallRange { + uint32_t first; + uint32_t last; +}; + +const SyscallRange kValidSyscallRanges[] = { + // First we iterate up to MAX_PUBLIC_SYSCALL, which is equal to MAX_SYSCALL + // on Intel architectures, but leaves room for private syscalls on ARM. + {MIN_SYSCALL, MAX_PUBLIC_SYSCALL}, +#if defined(__arm__) + // ARM EABI includes "ARM private" system calls starting at + // MIN_PRIVATE_SYSCALL, and a "ghost syscall private to the kernel" at + // MIN_GHOST_SYSCALL. + {MIN_PRIVATE_SYSCALL, MAX_PRIVATE_SYSCALL}, + {MIN_GHOST_SYSCALL, MAX_SYSCALL}, +#endif +}; + +} // namespace + +SyscallSet::Iterator SyscallSet::begin() const { + return Iterator(set_, false); +} + +SyscallSet::Iterator SyscallSet::end() const { + return Iterator(set_, true); +} + +bool SyscallSet::IsValid(uint32_t num) { + for (const SyscallRange& range : kValidSyscallRanges) { + if (num >= range.first && num <= range.last) { + return true; + } + } + return false; +} + +bool operator==(const SyscallSet& lhs, const SyscallSet& rhs) { + return (lhs.set_ == rhs.set_); +} + +SyscallSet::Iterator::Iterator(Set set, bool done) + : set_(set), done_(done), num_(0) { + // If the set doesn't contain 0, we need to skip to the next element. + if (!done && set_ == (IsValid(num_) ? Set::INVALID_ONLY : Set::VALID_ONLY)) { + ++*this; + } +} + +uint32_t SyscallSet::Iterator::operator*() const { + DCHECK(!done_); + return num_; +} + +SyscallSet::Iterator& SyscallSet::Iterator::operator++() { + DCHECK(!done_); + + num_ = NextSyscall(); + if (num_ == 0) { + done_ = true; + } + + return *this; +} + +// NextSyscall returns the next system call in the iterated system +// call set after |num_|, or 0 if no such system call exists. +uint32_t SyscallSet::Iterator::NextSyscall() const { + const bool want_valid = (set_ != Set::INVALID_ONLY); + const bool want_invalid = (set_ != Set::VALID_ONLY); + + for (const SyscallRange& range : kValidSyscallRanges) { + if (want_invalid && range.first > 0 && num_ < range.first - 1) { + // Even when iterating invalid syscalls, we only include the end points; + // so skip directly to just before the next (valid) range. + return range.first - 1; + } + if (want_valid && num_ < range.first) { + return range.first; + } + if (want_valid && num_ < range.last) { + return num_ + 1; + } + if (want_invalid && num_ <= range.last) { + return range.last + 1; + } + } + + if (want_invalid) { + // BPF programs only ever operate on unsigned quantities. So, + // that's how we iterate; we return values from + // 0..0xFFFFFFFFu. But there are places, where the kernel might + // interpret system call numbers as signed quantities, so the + // boundaries between signed and unsigned values are potential + // problem cases. We want to explicitly return these values from + // our iterator. + if (num_ < 0x7FFFFFFFu) + return 0x7FFFFFFFu; + if (num_ < 0x80000000u) + return 0x80000000u; + + if (num_ < 0xFFFFFFFFu) + return 0xFFFFFFFFu; + } + + return 0; +} + +bool operator==(const SyscallSet::Iterator& lhs, + const SyscallSet::Iterator& rhs) { + DCHECK(lhs.set_ == rhs.set_); + return (lhs.done_ == rhs.done_) && (lhs.num_ == rhs.num_); +} + +bool operator!=(const SyscallSet::Iterator& lhs, + const SyscallSet::Iterator& rhs) { + return !(lhs == rhs); +} + +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/syscall_set.h b/sandbox/linux/bpf_dsl/syscall_set.h new file mode 100644 index 0000000000..b9f076d932 --- /dev/null +++ b/sandbox/linux/bpf_dsl/syscall_set.h @@ -0,0 +1,103 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_SYSCALL_SET_H__ +#define SANDBOX_LINUX_BPF_DSL_SYSCALL_SET_H__ + +#include <stdint.h> + +#include <iterator> + +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// Iterates over the entire system call range from 0..0xFFFFFFFFu. This +// iterator is aware of how system calls look like and will skip quickly +// over ranges that can't contain system calls. It iterates more slowly +// whenever it reaches a range that is potentially problematic, returning +// the last invalid value before a valid range of system calls, and the +// first invalid value after a valid range of syscalls. It iterates over +// individual values whenever it is in the normal range for system calls +// (typically MIN_SYSCALL..MAX_SYSCALL). +// +// Example usage: +// for (uint32_t sysnum : SyscallSet::All()) { +// // Do something with sysnum. +// } +class SANDBOX_EXPORT SyscallSet { + public: + class Iterator; + + SyscallSet(const SyscallSet& ss) : set_(ss.set_) {} + ~SyscallSet() {} + + Iterator begin() const; + Iterator end() const; + + // All returns a SyscallSet that contains both valid and invalid + // system call numbers. + static SyscallSet All() { return SyscallSet(Set::ALL); } + + // ValidOnly returns a SyscallSet that contains only valid system + // call numbers. + static SyscallSet ValidOnly() { return SyscallSet(Set::VALID_ONLY); } + + // InvalidOnly returns a SyscallSet that contains only invalid + // system call numbers, but still omits numbers in the middle of a + // range of invalid system call numbers. + static SyscallSet InvalidOnly() { return SyscallSet(Set::INVALID_ONLY); } + + // IsValid returns whether |num| specifies a valid system call + // number. + static bool IsValid(uint32_t num); + + private: + enum class Set { ALL, VALID_ONLY, INVALID_ONLY }; + + explicit SyscallSet(Set set) : set_(set) {} + + Set set_; + + friend bool operator==(const SyscallSet&, const SyscallSet&); + DISALLOW_ASSIGN(SyscallSet); +}; + +SANDBOX_EXPORT bool operator==(const SyscallSet& lhs, const SyscallSet& rhs); + +// Iterator provides C++ input iterator semantics for traversing a +// SyscallSet. +class SyscallSet::Iterator + : public std::iterator<std::input_iterator_tag, uint32_t> { + public: + Iterator(const Iterator& it) + : set_(it.set_), done_(it.done_), num_(it.num_) {} + ~Iterator() {} + + uint32_t operator*() const; + Iterator& operator++(); + + private: + Iterator(Set set, bool done); + + uint32_t NextSyscall() const; + + Set set_; + bool done_; + uint32_t num_; + + friend SyscallSet; + friend bool operator==(const Iterator&, const Iterator&); + DISALLOW_ASSIGN(Iterator); +}; + +SANDBOX_EXPORT bool operator==(const SyscallSet::Iterator& lhs, + const SyscallSet::Iterator& rhs); +SANDBOX_EXPORT bool operator!=(const SyscallSet::Iterator& lhs, + const SyscallSet::Iterator& rhs); + +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_SYSCALL_SET_H__ diff --git a/sandbox/linux/bpf_dsl/syscall_set_unittest.cc b/sandbox/linux/bpf_dsl/syscall_set_unittest.cc new file mode 100644 index 0000000000..fafb6f6f73 --- /dev/null +++ b/sandbox/linux/bpf_dsl/syscall_set_unittest.cc @@ -0,0 +1,124 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/syscall_set.h" + +#include <stdint.h> + +#include "sandbox/linux/bpf_dsl/linux_syscall_ranges.h" +#include "sandbox/linux/tests/unit_tests.h" + +namespace sandbox { + +namespace { + +const SyscallSet kSyscallSets[] = { + SyscallSet::All(), + SyscallSet::InvalidOnly(), +}; + +SANDBOX_TEST(SyscallSet, Monotonous) { + for (const SyscallSet& set : kSyscallSets) { + uint32_t prev = 0; + bool have_prev = false; + for (uint32_t sysnum : set) { + if (have_prev) { + SANDBOX_ASSERT(sysnum > prev); + } else if (set == SyscallSet::All()) { + // The iterator should start at 0. + SANDBOX_ASSERT(sysnum == 0); + } + + prev = sysnum; + have_prev = true; + } + + // The iterator should always return 0xFFFFFFFFu as the last value. + SANDBOX_ASSERT(have_prev); + SANDBOX_ASSERT(prev == 0xFFFFFFFFu); + } +} + +// AssertRange checks that SyscallIterator produces all system call +// numbers in the inclusive range [min, max]. +void AssertRange(uint32_t min, uint32_t max) { + SANDBOX_ASSERT(min < max); + uint32_t prev = min - 1; + for (uint32_t sysnum : SyscallSet::All()) { + if (sysnum >= min && sysnum <= max) { + SANDBOX_ASSERT(prev == sysnum - 1); + prev = sysnum; + } + } + SANDBOX_ASSERT(prev == max); +} + +SANDBOX_TEST(SyscallSet, ValidSyscallRanges) { + AssertRange(MIN_SYSCALL, MAX_PUBLIC_SYSCALL); +#if defined(__arm__) + AssertRange(MIN_PRIVATE_SYSCALL, MAX_PRIVATE_SYSCALL); + AssertRange(MIN_GHOST_SYSCALL, MAX_SYSCALL); +#endif +} + +SANDBOX_TEST(SyscallSet, InvalidSyscalls) { + static const uint32_t kExpected[] = { +#if defined(__mips__) + 0, + MIN_SYSCALL - 1, +#endif + MAX_PUBLIC_SYSCALL + 1, +#if defined(__arm__) + MIN_PRIVATE_SYSCALL - 1, + MAX_PRIVATE_SYSCALL + 1, + MIN_GHOST_SYSCALL - 1, + MAX_SYSCALL + 1, +#endif + 0x7FFFFFFFu, + 0x80000000u, + 0xFFFFFFFFu, + }; + + for (const SyscallSet& set : kSyscallSets) { + size_t i = 0; + for (uint32_t sysnum : set) { + if (!SyscallSet::IsValid(sysnum)) { + SANDBOX_ASSERT(i < arraysize(kExpected)); + SANDBOX_ASSERT(kExpected[i] == sysnum); + ++i; + } + } + SANDBOX_ASSERT(i == arraysize(kExpected)); + } +} + +SANDBOX_TEST(SyscallSet, ValidOnlyIsOnlyValid) { + for (uint32_t sysnum : SyscallSet::ValidOnly()) { + SANDBOX_ASSERT(SyscallSet::IsValid(sysnum)); + } +} + +SANDBOX_TEST(SyscallSet, InvalidOnlyIsOnlyInvalid) { + for (uint32_t sysnum : SyscallSet::InvalidOnly()) { + SANDBOX_ASSERT(!SyscallSet::IsValid(sysnum)); + } +} + +SANDBOX_TEST(SyscallSet, AllIsValidOnlyPlusInvalidOnly) { + std::vector<uint32_t> merged; + const SyscallSet valid_only = SyscallSet::ValidOnly(); + const SyscallSet invalid_only = SyscallSet::InvalidOnly(); + std::merge(valid_only.begin(), + valid_only.end(), + invalid_only.begin(), + invalid_only.end(), + std::back_inserter(merged)); + + const SyscallSet all = SyscallSet::All(); + SANDBOX_ASSERT(merged == std::vector<uint32_t>(all.begin(), all.end())); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/trap_registry.h b/sandbox/linux/bpf_dsl/trap_registry.h new file mode 100644 index 0000000000..0a5d2f14cc --- /dev/null +++ b/sandbox/linux/bpf_dsl/trap_registry.h @@ -0,0 +1,73 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_TRAP_REGISTRY_H_ +#define SANDBOX_LINUX_BPF_DSL_TRAP_REGISTRY_H_ + +#include <stdint.h> + +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// This must match the kernel's seccomp_data structure. +struct arch_seccomp_data { + int nr; + uint32_t arch; + uint64_t instruction_pointer; + uint64_t args[6]; +}; + +namespace bpf_dsl { + +// TrapRegistry provides an interface for registering "trap handlers" +// by associating them with non-zero 16-bit trap IDs. Trap IDs should +// remain valid for the lifetime of the trap registry. +class SANDBOX_EXPORT TrapRegistry { + public: + // TrapFnc is a pointer to a function that fulfills the trap handler + // function signature. + // + // Trap handlers follow the calling convention of native system + // calls; e.g., to report an error, they return an exit code in the + // range -1..-4096 instead of directly modifying errno. However, + // modifying errno is harmless, as the original value will be + // restored afterwards. + // + // Trap handlers are executed from signal context and possibly an + // async-signal context, so they must be async-signal safe: + // http://pubs.opengroup.org/onlinepubs/009695399/functions/xsh_chap02_04.html + typedef intptr_t (*TrapFnc)(const struct arch_seccomp_data& args, void* aux); + + // Add registers the specified trap handler tuple and returns a + // non-zero trap ID that uniquely identifies the tuple for the life + // time of the trap registry. If the same tuple is registered + // multiple times, the same value will be returned each time. + virtual uint16_t Add(TrapFnc fnc, const void* aux, bool safe) = 0; + + // EnableUnsafeTraps tries to enable unsafe traps and returns + // whether it was successful. This is a one-way operation. + // + // CAUTION: Enabling unsafe traps effectively defeats the security + // guarantees provided by the sandbox policy. TrapRegistry + // implementations should ensure unsafe traps are only enabled + // during testing. + virtual bool EnableUnsafeTraps() = 0; + + protected: + TrapRegistry() {} + + // TrapRegistry's destructor is intentionally non-virtual so that + // implementations can omit their destructor. Instead we protect against + // misuse by marking it protected. + ~TrapRegistry() {} + + DISALLOW_COPY_AND_ASSIGN(TrapRegistry); +}; + +} // namespace bpf_dsl +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_TRAP_REGISTRY_H_ diff --git a/sandbox/linux/bpf_dsl/verifier.cc b/sandbox/linux/bpf_dsl/verifier.cc new file mode 100644 index 0000000000..417c663e30 --- /dev/null +++ b/sandbox/linux/bpf_dsl/verifier.cc @@ -0,0 +1,396 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/bpf_dsl/verifier.h" + +#include <string.h> + +#include <limits> + +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_impl.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/bpf_dsl/policy_compiler.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" +#include "sandbox/linux/bpf_dsl/syscall_set.h" +#include "sandbox/linux/seccomp-bpf/errorcode.h" +#include "sandbox/linux/system_headers/linux_filter.h" +#include "sandbox/linux/system_headers/linux_seccomp.h" + +namespace sandbox { +namespace bpf_dsl { + +namespace { + +const uint64_t kLower32Bits = std::numeric_limits<uint32_t>::max(); +const uint64_t kUpper32Bits = static_cast<uint64_t>(kLower32Bits) << 32; + +struct State { + State(const std::vector<struct sock_filter>& p, + const struct arch_seccomp_data& d) + : program(p), data(d), ip(0), accumulator(0), acc_is_valid(false) {} + const std::vector<struct sock_filter>& program; + const struct arch_seccomp_data& data; + unsigned int ip; + uint32_t accumulator; + bool acc_is_valid; + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(State); +}; + +uint32_t EvaluateErrorCode(bpf_dsl::PolicyCompiler* compiler, + const ErrorCode& code, + const struct arch_seccomp_data& data) { + if (code.error_type() == ErrorCode::ET_SIMPLE || + code.error_type() == ErrorCode::ET_TRAP) { + return code.err(); + } else if (code.error_type() == ErrorCode::ET_COND) { + if (code.width() == ErrorCode::TP_32BIT && + (data.args[code.argno()] >> 32) && + (data.args[code.argno()] & 0xFFFFFFFF80000000ull) != + 0xFFFFFFFF80000000ull) { + return compiler->Unexpected64bitArgument().err(); + } + bool equal = (data.args[code.argno()] & code.mask()) == code.value(); + return EvaluateErrorCode(compiler, equal ? *code.passed() : *code.failed(), + data); + } else { + return SECCOMP_RET_INVALID; + } +} + +bool VerifyErrorCode(bpf_dsl::PolicyCompiler* compiler, + const std::vector<struct sock_filter>& program, + struct arch_seccomp_data* data, + const ErrorCode& root_code, + const ErrorCode& code, + const char** err) { + if (code.error_type() == ErrorCode::ET_SIMPLE || + code.error_type() == ErrorCode::ET_TRAP) { + const uint32_t computed_ret = Verifier::EvaluateBPF(program, *data, err); + if (*err) { + return false; + } + const uint32_t policy_ret = EvaluateErrorCode(compiler, root_code, *data); + if (computed_ret != policy_ret) { + // For efficiency's sake, we'd much rather compare "computed_ret" + // against "code.err()". This works most of the time, but it doesn't + // always work for nested conditional expressions. The test values + // that we generate on the fly to probe expressions can trigger + // code flow decisions in multiple nodes of the decision tree, and the + // only way to compute the correct error code in that situation is by + // calling EvaluateErrorCode(). + *err = "Exit code from BPF program doesn't match"; + return false; + } + } else if (code.error_type() == ErrorCode::ET_COND) { + if (code.argno() < 0 || code.argno() >= 6) { + *err = "Invalid argument number in error code"; + return false; + } + + // TODO(mdempsky): The test values generated here try to provide good + // coverage for generated BPF instructions while avoiding combinatorial + // explosion on large policies. Ideally we would instead take a fuzzing-like + // approach and generate a bounded number of test cases regardless of policy + // size. + + // Verify that we can check a value for simple equality. + data->args[code.argno()] = code.value(); + if (!VerifyErrorCode(compiler, program, data, root_code, *code.passed(), + err)) { + return false; + } + + // If mask ignores any bits, verify that setting those bits is still + // detected as equality. + uint64_t ignored_bits = ~code.mask(); + if (code.width() == ErrorCode::TP_32BIT) { + ignored_bits = static_cast<uint32_t>(ignored_bits); + } + if ((ignored_bits & kLower32Bits) != 0) { + data->args[code.argno()] = code.value() | (ignored_bits & kLower32Bits); + if (!VerifyErrorCode(compiler, program, data, root_code, *code.passed(), + err)) { + return false; + } + } + if ((ignored_bits & kUpper32Bits) != 0) { + data->args[code.argno()] = code.value() | (ignored_bits & kUpper32Bits); + if (!VerifyErrorCode(compiler, program, data, root_code, *code.passed(), + err)) { + return false; + } + } + + // Verify that changing bits included in the mask is detected as inequality. + if ((code.mask() & kLower32Bits) != 0) { + data->args[code.argno()] = code.value() ^ (code.mask() & kLower32Bits); + if (!VerifyErrorCode(compiler, program, data, root_code, *code.failed(), + err)) { + return false; + } + } + if ((code.mask() & kUpper32Bits) != 0) { + data->args[code.argno()] = code.value() ^ (code.mask() & kUpper32Bits); + if (!VerifyErrorCode(compiler, program, data, root_code, *code.failed(), + err)) { + return false; + } + } + + if (code.width() == ErrorCode::TP_32BIT) { + // For 32-bit system call arguments, we emit additional instructions to + // validate the upper 32-bits. Here we test that validation. + + // Arbitrary 64-bit values should be rejected. + data->args[code.argno()] = 1ULL << 32; + if (!VerifyErrorCode(compiler, program, data, root_code, + compiler->Unexpected64bitArgument(), err)) { + return false; + } + + // Upper 32-bits set without the MSB of the lower 32-bits set should be + // rejected too. + data->args[code.argno()] = kUpper32Bits; + if (!VerifyErrorCode(compiler, program, data, root_code, + compiler->Unexpected64bitArgument(), err)) { + return false; + } + } + } else { + *err = "Attempting to return invalid error code from BPF program"; + return false; + } + return true; +} + +void Ld(State* state, const struct sock_filter& insn, const char** err) { + if (BPF_SIZE(insn.code) != BPF_W || BPF_MODE(insn.code) != BPF_ABS || + insn.jt != 0 || insn.jf != 0) { + *err = "Invalid BPF_LD instruction"; + return; + } + if (insn.k < sizeof(struct arch_seccomp_data) && (insn.k & 3) == 0) { + // We only allow loading of properly aligned 32bit quantities. + memcpy(&state->accumulator, + reinterpret_cast<const char*>(&state->data) + insn.k, 4); + } else { + *err = "Invalid operand in BPF_LD instruction"; + return; + } + state->acc_is_valid = true; + return; +} + +void Jmp(State* state, const struct sock_filter& insn, const char** err) { + if (BPF_OP(insn.code) == BPF_JA) { + if (state->ip + insn.k + 1 >= state->program.size() || + state->ip + insn.k + 1 <= state->ip) { + compilation_failure: + *err = "Invalid BPF_JMP instruction"; + return; + } + state->ip += insn.k; + } else { + if (BPF_SRC(insn.code) != BPF_K || !state->acc_is_valid || + state->ip + insn.jt + 1 >= state->program.size() || + state->ip + insn.jf + 1 >= state->program.size()) { + goto compilation_failure; + } + switch (BPF_OP(insn.code)) { + case BPF_JEQ: + if (state->accumulator == insn.k) { + state->ip += insn.jt; + } else { + state->ip += insn.jf; + } + break; + case BPF_JGT: + if (state->accumulator > insn.k) { + state->ip += insn.jt; + } else { + state->ip += insn.jf; + } + break; + case BPF_JGE: + if (state->accumulator >= insn.k) { + state->ip += insn.jt; + } else { + state->ip += insn.jf; + } + break; + case BPF_JSET: + if (state->accumulator & insn.k) { + state->ip += insn.jt; + } else { + state->ip += insn.jf; + } + break; + default: + goto compilation_failure; + } + } +} + +uint32_t Ret(State*, const struct sock_filter& insn, const char** err) { + if (BPF_SRC(insn.code) != BPF_K) { + *err = "Invalid BPF_RET instruction"; + return 0; + } + return insn.k; +} + +void Alu(State* state, const struct sock_filter& insn, const char** err) { + if (BPF_OP(insn.code) == BPF_NEG) { + state->accumulator = -state->accumulator; + return; + } else { + if (BPF_SRC(insn.code) != BPF_K) { + *err = "Unexpected source operand in arithmetic operation"; + return; + } + switch (BPF_OP(insn.code)) { + case BPF_ADD: + state->accumulator += insn.k; + break; + case BPF_SUB: + state->accumulator -= insn.k; + break; + case BPF_MUL: + state->accumulator *= insn.k; + break; + case BPF_DIV: + if (!insn.k) { + *err = "Illegal division by zero"; + break; + } + state->accumulator /= insn.k; + break; + case BPF_MOD: + if (!insn.k) { + *err = "Illegal division by zero"; + break; + } + state->accumulator %= insn.k; + break; + case BPF_OR: + state->accumulator |= insn.k; + break; + case BPF_XOR: + state->accumulator ^= insn.k; + break; + case BPF_AND: + state->accumulator &= insn.k; + break; + case BPF_LSH: + if (insn.k > 32) { + *err = "Illegal shift operation"; + break; + } + state->accumulator <<= insn.k; + break; + case BPF_RSH: + if (insn.k > 32) { + *err = "Illegal shift operation"; + break; + } + state->accumulator >>= insn.k; + break; + default: + *err = "Invalid operator in arithmetic operation"; + break; + } + } +} + +} // namespace + +bool Verifier::VerifyBPF(bpf_dsl::PolicyCompiler* compiler, + const std::vector<struct sock_filter>& program, + const bpf_dsl::Policy& policy, + const char** err) { + *err = NULL; + for (uint32_t sysnum : SyscallSet::All()) { + // We ideally want to iterate over the full system call range and values + // just above and just below this range. This gives us the full result set + // of the "evaluators". + // On Intel systems, this can fail in a surprising way, as a cleared bit 30 + // indicates either i386 or x86-64; and a set bit 30 indicates x32. And + // unless we pay attention to setting this bit correctly, an early check in + // our BPF program will make us fail with a misleading error code. + struct arch_seccomp_data data = {static_cast<int>(sysnum), + static_cast<uint32_t>(SECCOMP_ARCH)}; +#if defined(__i386__) || defined(__x86_64__) +#if defined(__x86_64__) && defined(__ILP32__) + if (!(sysnum & 0x40000000u)) { + continue; + } +#else + if (sysnum & 0x40000000u) { + continue; + } +#endif +#endif + ErrorCode code = SyscallSet::IsValid(sysnum) + ? policy.EvaluateSyscall(sysnum)->Compile(compiler) + : policy.InvalidSyscall()->Compile(compiler); + if (!VerifyErrorCode(compiler, program, &data, code, code, err)) { + return false; + } + } + return true; +} + +uint32_t Verifier::EvaluateBPF(const std::vector<struct sock_filter>& program, + const struct arch_seccomp_data& data, + const char** err) { + *err = NULL; + if (program.size() < 1 || program.size() >= SECCOMP_MAX_PROGRAM_SIZE) { + *err = "Invalid program length"; + return 0; + } + for (State state(program, data); !*err; ++state.ip) { + if (state.ip >= program.size()) { + *err = "Invalid instruction pointer in BPF program"; + break; + } + const struct sock_filter& insn = program[state.ip]; + switch (BPF_CLASS(insn.code)) { + case BPF_LD: + Ld(&state, insn, err); + break; + case BPF_JMP: + Jmp(&state, insn, err); + break; + case BPF_RET: { + uint32_t r = Ret(&state, insn, err); + switch (r & SECCOMP_RET_ACTION) { + case SECCOMP_RET_TRAP: + case SECCOMP_RET_ERRNO: + case SECCOMP_RET_TRACE: + case SECCOMP_RET_ALLOW: + break; + case SECCOMP_RET_KILL: // We don't ever generate this + case SECCOMP_RET_INVALID: // Should never show up in BPF program + default: + *err = "Unexpected return code found in BPF program"; + return 0; + } + return r; + } + case BPF_ALU: + Alu(&state, insn, err); + break; + default: + *err = "Unexpected instruction in BPF program"; + break; + } + } + return 0; +} + +} // namespace bpf_dsl +} // namespace sandbox diff --git a/sandbox/linux/bpf_dsl/verifier.h b/sandbox/linux/bpf_dsl/verifier.h new file mode 100644 index 0000000000..b0435d1aa1 --- /dev/null +++ b/sandbox/linux/bpf_dsl/verifier.h @@ -0,0 +1,57 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_BPF_DSL_VERIFIER_H__ +#define SANDBOX_LINUX_BPF_DSL_VERIFIER_H__ + +#include <stdint.h> + +#include <vector> + +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +struct sock_filter; + +namespace sandbox { +struct arch_seccomp_data; + +namespace bpf_dsl { +class Policy; +class PolicyCompiler; + +class SANDBOX_EXPORT Verifier { + public: + // Evaluate the BPF program for all possible inputs and verify that it + // computes the correct result. We use the "evaluators" to determine + // the full set of possible inputs that we have to iterate over. + // Returns success, if the BPF filter accurately reflects the rules + // set by the "evaluators". + // Upon success, "err" is set to NULL. Upon failure, it contains a static + // error message that does not need to be free()'d. + static bool VerifyBPF(bpf_dsl::PolicyCompiler* compiler, + const std::vector<struct sock_filter>& program, + const bpf_dsl::Policy& policy, + const char** err); + + // Evaluate a given BPF program for a particular set of system call + // parameters. If evaluation failed for any reason, "err" will be set to + // a non-NULL error string. Otherwise, the BPF program's result will be + // returned by the function and "err" is NULL. + // We do not actually implement the full BPF state machine, but only the + // parts that can actually be generated by our BPF compiler. If this code + // is used for purposes other than verifying the output of the sandbox's + // BPF compiler, we might have to extend this BPF interpreter. + static uint32_t EvaluateBPF(const std::vector<struct sock_filter>& program, + const struct arch_seccomp_data& data, + const char** err); + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(Verifier); +}; + +} // namespace bpf_dsl +} // namespace sandbox + +#endif // SANDBOX_LINUX_BPF_DSL_VERIFIER_H__ diff --git a/sandbox/linux/integration_tests/DEPS b/sandbox/linux/integration_tests/DEPS new file mode 100644 index 0000000000..d50729cea3 --- /dev/null +++ b/sandbox/linux/integration_tests/DEPS @@ -0,0 +1,7 @@ +include_rules = [ + "+sandbox/linux/bpf_dsl", + "+sandbox/linux/seccomp-bpf", + "+sandbox/linux/services", + "+sandbox/linux/syscall_broker", + "+sandbox/linux/system_headers", +] diff --git a/sandbox/linux/integration_tests/bpf_dsl_seccomp_unittest.cc b/sandbox/linux/integration_tests/bpf_dsl_seccomp_unittest.cc new file mode 100644 index 0000000000..e884774146 --- /dev/null +++ b/sandbox/linux/integration_tests/bpf_dsl_seccomp_unittest.cc @@ -0,0 +1,2259 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include <errno.h> +#include <fcntl.h> +#include <pthread.h> +#include <sched.h> +#include <signal.h> +#include <sys/prctl.h> +#include <sys/ptrace.h> +#include <sys/syscall.h> +#include <sys/time.h> +#include <sys/types.h> +#include <sys/utsname.h> +#include <unistd.h> +#include <sys/socket.h> + +#if defined(ANDROID) +// Work-around for buggy headers in Android's NDK +#define __user +#endif +#include <linux/futex.h> + +#include "base/bind.h" +#include "base/logging.h" +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "base/posix/eintr_wrapper.h" +#include "base/synchronization/waitable_event.h" +#include "base/sys_info.h" +#include "base/threading/thread.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/linux_syscall_ranges.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" +#include "sandbox/linux/seccomp-bpf/bpf_tests.h" +#include "sandbox/linux/seccomp-bpf/die.h" +#include "sandbox/linux/seccomp-bpf/errorcode.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/seccomp-bpf/syscall.h" +#include "sandbox/linux/seccomp-bpf/trap.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/services/thread_helpers.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" +#include "sandbox/linux/tests/scoped_temporary_file.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +// Workaround for Android's prctl.h file. +#ifndef PR_GET_ENDIAN +#define PR_GET_ENDIAN 19 +#endif +#ifndef PR_CAPBSET_READ +#define PR_CAPBSET_READ 23 +#define PR_CAPBSET_DROP 24 +#endif + +namespace sandbox { +namespace bpf_dsl { + +namespace { + +const int kExpectedReturnValue = 42; +const char kSandboxDebuggingEnv[] = "CHROME_SANDBOX_DEBUGGING"; + +// Set the global environment to allow the use of UnsafeTrap() policies. +void EnableUnsafeTraps() { + // The use of UnsafeTrap() causes us to print a warning message. This is + // generally desirable, but it results in the unittest failing, as it doesn't + // expect any messages on "stderr". So, temporarily disable messages. The + // BPF_TEST() is guaranteed to turn messages back on, after the policy + // function has completed. + setenv(kSandboxDebuggingEnv, "t", 0); + Die::SuppressInfoMessages(true); +} + +// BPF_TEST does a lot of the boiler-plate code around setting up a +// policy and optional passing data between the caller, the policy and +// any Trap() handlers. This is great for writing short and concise tests, +// and it helps us accidentally forgetting any of the crucial steps in +// setting up the sandbox. But it wouldn't hurt to have at least one test +// that explicitly walks through all these steps. + +intptr_t IncreaseCounter(const struct arch_seccomp_data& args, void* aux) { + BPF_ASSERT(aux); + int* counter = static_cast<int*>(aux); + return (*counter)++; +} + +class VerboseAPITestingPolicy : public Policy { + public: + explicit VerboseAPITestingPolicy(int* counter_ptr) + : counter_ptr_(counter_ptr) {} + ~VerboseAPITestingPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + if (sysno == __NR_uname) { + return Trap(IncreaseCounter, counter_ptr_); + } + return Allow(); + } + + private: + int* counter_ptr_; + + DISALLOW_COPY_AND_ASSIGN(VerboseAPITestingPolicy); +}; + +SANDBOX_TEST(SandboxBPF, DISABLE_ON_TSAN(VerboseAPITesting)) { + if (SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::SINGLE_THREADED)) { + static int counter = 0; + + SandboxBPF sandbox(new VerboseAPITestingPolicy(&counter)); + BPF_ASSERT(sandbox.StartSandbox(SandboxBPF::SeccompLevel::SINGLE_THREADED)); + + BPF_ASSERT_EQ(0, counter); + BPF_ASSERT_EQ(0, syscall(__NR_uname, 0)); + BPF_ASSERT_EQ(1, counter); + BPF_ASSERT_EQ(1, syscall(__NR_uname, 0)); + BPF_ASSERT_EQ(2, counter); + } +} + +// A simple blacklist test + +class BlacklistNanosleepPolicy : public Policy { + public: + BlacklistNanosleepPolicy() {} + ~BlacklistNanosleepPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + switch (sysno) { + case __NR_nanosleep: + return Error(EACCES); + default: + return Allow(); + } + } + + static void AssertNanosleepFails() { + const struct timespec ts = {0, 0}; + errno = 0; + BPF_ASSERT_EQ(-1, HANDLE_EINTR(syscall(__NR_nanosleep, &ts, NULL))); + BPF_ASSERT_EQ(EACCES, errno); + } + + private: + DISALLOW_COPY_AND_ASSIGN(BlacklistNanosleepPolicy); +}; + +BPF_TEST_C(SandboxBPF, ApplyBasicBlacklistPolicy, BlacklistNanosleepPolicy) { + BlacklistNanosleepPolicy::AssertNanosleepFails(); +} + +BPF_TEST_C(SandboxBPF, UseVsyscall, BlacklistNanosleepPolicy) { + time_t current_time; + // time() is implemented as a vsyscall. With an older glibc, with + // vsyscall=emulate and some versions of the seccomp BPF patch + // we may get SIGKILL-ed. Detect this! + BPF_ASSERT_NE(static_cast<time_t>(-1), time(¤t_time)); +} + +// Now do a simple whitelist test + +class WhitelistGetpidPolicy : public Policy { + public: + WhitelistGetpidPolicy() {} + ~WhitelistGetpidPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + switch (sysno) { + case __NR_getpid: + case __NR_exit_group: + return Allow(); + default: + return Error(ENOMEM); + } + } + + private: + DISALLOW_COPY_AND_ASSIGN(WhitelistGetpidPolicy); +}; + +BPF_TEST_C(SandboxBPF, ApplyBasicWhitelistPolicy, WhitelistGetpidPolicy) { + // getpid() should be allowed + errno = 0; + BPF_ASSERT(sys_getpid() > 0); + BPF_ASSERT(errno == 0); + + // getpgid() should be denied + BPF_ASSERT(getpgid(0) == -1); + BPF_ASSERT(errno == ENOMEM); +} + +// A simple blacklist policy, with a SIGSYS handler +intptr_t EnomemHandler(const struct arch_seccomp_data& args, void* aux) { + // We also check that the auxiliary data is correct + SANDBOX_ASSERT(aux); + *(static_cast<int*>(aux)) = kExpectedReturnValue; + return -ENOMEM; +} + +class BlacklistNanosleepTrapPolicy : public Policy { + public: + explicit BlacklistNanosleepTrapPolicy(int* aux) : aux_(aux) {} + ~BlacklistNanosleepTrapPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + switch (sysno) { + case __NR_nanosleep: + return Trap(EnomemHandler, aux_); + default: + return Allow(); + } + } + + private: + int* aux_; + + DISALLOW_COPY_AND_ASSIGN(BlacklistNanosleepTrapPolicy); +}; + +BPF_TEST(SandboxBPF, + BasicBlacklistWithSigsys, + BlacklistNanosleepTrapPolicy, + int /* (*BPF_AUX) */) { + // getpid() should work properly + errno = 0; + BPF_ASSERT(sys_getpid() > 0); + BPF_ASSERT(errno == 0); + + // Our Auxiliary Data, should be reset by the signal handler + *BPF_AUX = -1; + const struct timespec ts = {0, 0}; + BPF_ASSERT(syscall(__NR_nanosleep, &ts, NULL) == -1); + BPF_ASSERT(errno == ENOMEM); + + // We expect the signal handler to modify AuxData + BPF_ASSERT(*BPF_AUX == kExpectedReturnValue); +} + +// A simple test that verifies we can return arbitrary errno values. + +class ErrnoTestPolicy : public Policy { + public: + ErrnoTestPolicy() {} + ~ErrnoTestPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + DISALLOW_COPY_AND_ASSIGN(ErrnoTestPolicy); +}; + +ResultExpr ErrnoTestPolicy::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + switch (sysno) { + case __NR_dup3: // dup2 is a wrapper of dup3 in android +#if defined(__NR_dup2) + case __NR_dup2: +#endif + // Pretend that dup2() worked, but don't actually do anything. + return Error(0); + case __NR_setuid: +#if defined(__NR_setuid32) + case __NR_setuid32: +#endif + // Return errno = 1. + return Error(1); + case __NR_setgid: +#if defined(__NR_setgid32) + case __NR_setgid32: +#endif + // Return maximum errno value (typically 4095). + return Error(ErrorCode::ERR_MAX_ERRNO); + case __NR_uname: + // Return errno = 42; + return Error(42); + default: + return Allow(); + } +} + +BPF_TEST_C(SandboxBPF, ErrnoTest, ErrnoTestPolicy) { + // Verify that dup2() returns success, but doesn't actually run. + int fds[4]; + BPF_ASSERT(pipe(fds) == 0); + BPF_ASSERT(pipe(fds + 2) == 0); + BPF_ASSERT(dup2(fds[2], fds[0]) == 0); + char buf[1] = {}; + BPF_ASSERT(write(fds[1], "\x55", 1) == 1); + BPF_ASSERT(write(fds[3], "\xAA", 1) == 1); + BPF_ASSERT(read(fds[0], buf, 1) == 1); + + // If dup2() executed, we will read \xAA, but it dup2() has been turned + // into a no-op by our policy, then we will read \x55. + BPF_ASSERT(buf[0] == '\x55'); + + // Verify that we can return the minimum and maximum errno values. + errno = 0; + BPF_ASSERT(setuid(0) == -1); + BPF_ASSERT(errno == 1); + + // On Android, errno is only supported up to 255, otherwise errno + // processing is skipped. + // We work around this (crbug.com/181647). + if (sandbox::IsAndroid() && setgid(0) != -1) { + errno = 0; + BPF_ASSERT(setgid(0) == -ErrorCode::ERR_MAX_ERRNO); + BPF_ASSERT(errno == 0); + } else { + errno = 0; + BPF_ASSERT(setgid(0) == -1); + BPF_ASSERT(errno == ErrorCode::ERR_MAX_ERRNO); + } + + // Finally, test an errno in between the minimum and maximum. + errno = 0; + struct utsname uts_buf; + BPF_ASSERT(uname(&uts_buf) == -1); + BPF_ASSERT(errno == 42); +} + +// Testing the stacking of two sandboxes + +class StackingPolicyPartOne : public Policy { + public: + StackingPolicyPartOne() {} + ~StackingPolicyPartOne() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + switch (sysno) { + case __NR_getppid: { + const Arg<int> arg(0); + return If(arg == 0, Allow()).Else(Error(EPERM)); + } + default: + return Allow(); + } + } + + private: + DISALLOW_COPY_AND_ASSIGN(StackingPolicyPartOne); +}; + +class StackingPolicyPartTwo : public Policy { + public: + StackingPolicyPartTwo() {} + ~StackingPolicyPartTwo() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + switch (sysno) { + case __NR_getppid: { + const Arg<int> arg(0); + return If(arg == 0, Error(EINVAL)).Else(Allow()); + } + default: + return Allow(); + } + } + + private: + DISALLOW_COPY_AND_ASSIGN(StackingPolicyPartTwo); +}; + +BPF_TEST_C(SandboxBPF, StackingPolicy, StackingPolicyPartOne) { + errno = 0; + BPF_ASSERT(syscall(__NR_getppid, 0) > 0); + BPF_ASSERT(errno == 0); + + BPF_ASSERT(syscall(__NR_getppid, 1) == -1); + BPF_ASSERT(errno == EPERM); + + // Stack a second sandbox with its own policy. Verify that we can further + // restrict filters, but we cannot relax existing filters. + SandboxBPF sandbox(new StackingPolicyPartTwo()); + BPF_ASSERT(sandbox.StartSandbox(SandboxBPF::SeccompLevel::SINGLE_THREADED)); + + errno = 0; + BPF_ASSERT(syscall(__NR_getppid, 0) == -1); + BPF_ASSERT(errno == EINVAL); + + BPF_ASSERT(syscall(__NR_getppid, 1) == -1); + BPF_ASSERT(errno == EPERM); +} + +// A more complex, but synthetic policy. This tests the correctness of the BPF +// program by iterating through all syscalls and checking for an errno that +// depends on the syscall number. Unlike the Verifier, this exercises the BPF +// interpreter in the kernel. + +// We try to make sure we exercise optimizations in the BPF compiler. We make +// sure that the compiler can have an opportunity to coalesce syscalls with +// contiguous numbers and we also make sure that disjoint sets can return the +// same errno. +int SysnoToRandomErrno(int sysno) { + // Small contiguous sets of 3 system calls return an errno equal to the + // index of that set + 1 (so that we never return a NUL errno). + return ((sysno & ~3) >> 2) % 29 + 1; +} + +class SyntheticPolicy : public Policy { + public: + SyntheticPolicy() {} + ~SyntheticPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + if (sysno == __NR_exit_group || sysno == __NR_write) { + // exit_group() is special, we really need it to work. + // write() is needed for BPF_ASSERT() to report a useful error message. + return Allow(); + } + return Error(SysnoToRandomErrno(sysno)); + } + + private: + DISALLOW_COPY_AND_ASSIGN(SyntheticPolicy); +}; + +BPF_TEST_C(SandboxBPF, SyntheticPolicy, SyntheticPolicy) { + // Ensure that that kExpectedReturnValue + syscallnumber + 1 does not int + // overflow. + BPF_ASSERT(std::numeric_limits<int>::max() - kExpectedReturnValue - 1 >= + static_cast<int>(MAX_PUBLIC_SYSCALL)); + + for (int syscall_number = static_cast<int>(MIN_SYSCALL); + syscall_number <= static_cast<int>(MAX_PUBLIC_SYSCALL); + ++syscall_number) { + if (syscall_number == __NR_exit_group || syscall_number == __NR_write) { + // exit_group() is special + continue; + } + errno = 0; + BPF_ASSERT(syscall(syscall_number) == -1); + BPF_ASSERT(errno == SysnoToRandomErrno(syscall_number)); + } +} + +#if defined(__arm__) +// A simple policy that tests whether ARM private system calls are supported +// by our BPF compiler and by the BPF interpreter in the kernel. + +// For ARM private system calls, return an errno equal to their offset from +// MIN_PRIVATE_SYSCALL plus 1 (to avoid NUL errno). +int ArmPrivateSysnoToErrno(int sysno) { + if (sysno >= static_cast<int>(MIN_PRIVATE_SYSCALL) && + sysno <= static_cast<int>(MAX_PRIVATE_SYSCALL)) { + return (sysno - MIN_PRIVATE_SYSCALL) + 1; + } else { + return ENOSYS; + } +} + +class ArmPrivatePolicy : public Policy { + public: + ArmPrivatePolicy() {} + ~ArmPrivatePolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + // Start from |__ARM_NR_set_tls + 1| so as not to mess with actual + // ARM private system calls. + if (sysno >= static_cast<int>(__ARM_NR_set_tls + 1) && + sysno <= static_cast<int>(MAX_PRIVATE_SYSCALL)) { + return Error(ArmPrivateSysnoToErrno(sysno)); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(ArmPrivatePolicy); +}; + +BPF_TEST_C(SandboxBPF, ArmPrivatePolicy, ArmPrivatePolicy) { + for (int syscall_number = static_cast<int>(__ARM_NR_set_tls + 1); + syscall_number <= static_cast<int>(MAX_PRIVATE_SYSCALL); + ++syscall_number) { + errno = 0; + BPF_ASSERT(syscall(syscall_number) == -1); + BPF_ASSERT(errno == ArmPrivateSysnoToErrno(syscall_number)); + } +} +#endif // defined(__arm__) + +intptr_t CountSyscalls(const struct arch_seccomp_data& args, void* aux) { + // Count all invocations of our callback function. + ++*reinterpret_cast<int*>(aux); + + // Verify that within the callback function all filtering is temporarily + // disabled. + BPF_ASSERT(sys_getpid() > 1); + + // Verify that we can now call the underlying system call without causing + // infinite recursion. + return SandboxBPF::ForwardSyscall(args); +} + +class GreyListedPolicy : public Policy { + public: + explicit GreyListedPolicy(int* aux) : aux_(aux) { + // Set the global environment for unsafe traps once. + EnableUnsafeTraps(); + } + ~GreyListedPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + // Some system calls must always be allowed, if our policy wants to make + // use of UnsafeTrap() + if (SandboxBPF::IsRequiredForUnsafeTrap(sysno)) { + return Allow(); + } else if (sysno == __NR_getpid) { + // Disallow getpid() + return Error(EPERM); + } else { + // Allow (and count) all other system calls. + return UnsafeTrap(CountSyscalls, aux_); + } + } + + private: + int* aux_; + + DISALLOW_COPY_AND_ASSIGN(GreyListedPolicy); +}; + +BPF_TEST(SandboxBPF, GreyListedPolicy, GreyListedPolicy, int /* (*BPF_AUX) */) { + BPF_ASSERT(sys_getpid() == -1); + BPF_ASSERT(errno == EPERM); + BPF_ASSERT(*BPF_AUX == 0); + BPF_ASSERT(syscall(__NR_geteuid) == syscall(__NR_getuid)); + BPF_ASSERT(*BPF_AUX == 2); + char name[17] = {}; + BPF_ASSERT(!syscall(__NR_prctl, + PR_GET_NAME, + name, + (void*)NULL, + (void*)NULL, + (void*)NULL)); + BPF_ASSERT(*BPF_AUX == 3); + BPF_ASSERT(*name); +} + +SANDBOX_TEST(SandboxBPF, EnableUnsafeTrapsInSigSysHandler) { + // Disabling warning messages that could confuse our test framework. + setenv(kSandboxDebuggingEnv, "t", 0); + Die::SuppressInfoMessages(true); + + unsetenv(kSandboxDebuggingEnv); + SANDBOX_ASSERT(Trap::Registry()->EnableUnsafeTraps() == false); + setenv(kSandboxDebuggingEnv, "", 1); + SANDBOX_ASSERT(Trap::Registry()->EnableUnsafeTraps() == false); + setenv(kSandboxDebuggingEnv, "t", 1); + SANDBOX_ASSERT(Trap::Registry()->EnableUnsafeTraps() == true); +} + +intptr_t PrctlHandler(const struct arch_seccomp_data& args, void*) { + if (args.args[0] == PR_CAPBSET_DROP && static_cast<int>(args.args[1]) == -1) { + // prctl(PR_CAPBSET_DROP, -1) is never valid. The kernel will always + // return an error. But our handler allows this call. + return 0; + } else { + return SandboxBPF::ForwardSyscall(args); + } +} + +class PrctlPolicy : public Policy { + public: + PrctlPolicy() {} + ~PrctlPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + setenv(kSandboxDebuggingEnv, "t", 0); + Die::SuppressInfoMessages(true); + + if (sysno == __NR_prctl) { + // Handle prctl() inside an UnsafeTrap() + return UnsafeTrap(PrctlHandler, NULL); + } + + // Allow all other system calls. + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(PrctlPolicy); +}; + +BPF_TEST_C(SandboxBPF, ForwardSyscall, PrctlPolicy) { + // This call should never be allowed. But our policy will intercept it and + // let it pass successfully. + BPF_ASSERT( + !prctl(PR_CAPBSET_DROP, -1, (void*)NULL, (void*)NULL, (void*)NULL)); + + // Verify that the call will fail, if it makes it all the way to the kernel. + BPF_ASSERT( + prctl(PR_CAPBSET_DROP, -2, (void*)NULL, (void*)NULL, (void*)NULL) == -1); + + // And verify that other uses of prctl() work just fine. + char name[17] = {}; + BPF_ASSERT(!syscall(__NR_prctl, + PR_GET_NAME, + name, + (void*)NULL, + (void*)NULL, + (void*)NULL)); + BPF_ASSERT(*name); + + // Finally, verify that system calls other than prctl() are completely + // unaffected by our policy. + struct utsname uts = {}; + BPF_ASSERT(!uname(&uts)); + BPF_ASSERT(!strcmp(uts.sysname, "Linux")); +} + +intptr_t AllowRedirectedSyscall(const struct arch_seccomp_data& args, void*) { + return SandboxBPF::ForwardSyscall(args); +} + +class RedirectAllSyscallsPolicy : public Policy { + public: + RedirectAllSyscallsPolicy() {} + ~RedirectAllSyscallsPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + DISALLOW_COPY_AND_ASSIGN(RedirectAllSyscallsPolicy); +}; + +ResultExpr RedirectAllSyscallsPolicy::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + setenv(kSandboxDebuggingEnv, "t", 0); + Die::SuppressInfoMessages(true); + + // Some system calls must always be allowed, if our policy wants to make + // use of UnsafeTrap() + if (SandboxBPF::IsRequiredForUnsafeTrap(sysno)) + return Allow(); + return UnsafeTrap(AllowRedirectedSyscall, NULL); +} + +#if !defined(ADDRESS_SANITIZER) +// ASan does not allow changing the signal handler for SIGBUS, and treats it as +// a fatal signal. + +int bus_handler_fd_ = -1; + +void SigBusHandler(int, siginfo_t* info, void* void_context) { + BPF_ASSERT(write(bus_handler_fd_, "\x55", 1) == 1); +} + +BPF_TEST_C(SandboxBPF, SigBus, RedirectAllSyscallsPolicy) { + // We use the SIGBUS bit in the signal mask as a thread-local boolean + // value in the implementation of UnsafeTrap(). This is obviously a bit + // of a hack that could conceivably interfere with code that uses SIGBUS + // in more traditional ways. This test verifies that basic functionality + // of SIGBUS is not impacted, but it is certainly possibly to construe + // more complex uses of signals where our use of the SIGBUS mask is not + // 100% transparent. This is expected behavior. + int fds[2]; + BPF_ASSERT(socketpair(AF_UNIX, SOCK_STREAM, 0, fds) == 0); + bus_handler_fd_ = fds[1]; + struct sigaction sa = {}; + sa.sa_sigaction = SigBusHandler; + sa.sa_flags = SA_SIGINFO; + BPF_ASSERT(sigaction(SIGBUS, &sa, NULL) == 0); + raise(SIGBUS); + char c = '\000'; + BPF_ASSERT(read(fds[0], &c, 1) == 1); + BPF_ASSERT(close(fds[0]) == 0); + BPF_ASSERT(close(fds[1]) == 0); + BPF_ASSERT(c == 0x55); +} +#endif // !defined(ADDRESS_SANITIZER) + +BPF_TEST_C(SandboxBPF, SigMask, RedirectAllSyscallsPolicy) { + // Signal masks are potentially tricky to handle. For instance, if we + // ever tried to update them from inside a Trap() or UnsafeTrap() handler, + // the call to sigreturn() at the end of the signal handler would undo + // all of our efforts. So, it makes sense to test that sigprocmask() + // works, even if we have a policy in place that makes use of UnsafeTrap(). + // In practice, this works because we force sigprocmask() to be handled + // entirely in the kernel. + sigset_t mask0, mask1, mask2; + + // Call sigprocmask() to verify that SIGUSR2 wasn't blocked, if we didn't + // change the mask (it shouldn't have been, as it isn't blocked by default + // in POSIX). + // + // Use SIGUSR2 because Android seems to use SIGUSR1 for some purpose. + sigemptyset(&mask0); + BPF_ASSERT(!sigprocmask(SIG_BLOCK, &mask0, &mask1)); + BPF_ASSERT(!sigismember(&mask1, SIGUSR2)); + + // Try again, and this time we verify that we can block it. This + // requires a second call to sigprocmask(). + sigaddset(&mask0, SIGUSR2); + BPF_ASSERT(!sigprocmask(SIG_BLOCK, &mask0, NULL)); + BPF_ASSERT(!sigprocmask(SIG_BLOCK, NULL, &mask2)); + BPF_ASSERT(sigismember(&mask2, SIGUSR2)); +} + +BPF_TEST_C(SandboxBPF, UnsafeTrapWithErrno, RedirectAllSyscallsPolicy) { + // An UnsafeTrap() (or for that matter, a Trap()) has to report error + // conditions by returning an exit code in the range -1..-4096. This + // should happen automatically if using ForwardSyscall(). If the TrapFnc() + // uses some other method to make system calls, then it is responsible + // for computing the correct return code. + // This test verifies that ForwardSyscall() does the correct thing. + + // The glibc system wrapper will ultimately set errno for us. So, from normal + // userspace, all of this should be completely transparent. + errno = 0; + BPF_ASSERT(close(-1) == -1); + BPF_ASSERT(errno == EBADF); + + // Explicitly avoid the glibc wrapper. This is not normally the way anybody + // would make system calls, but it allows us to verify that we don't + // accidentally mess with errno, when we shouldn't. + errno = 0; + struct arch_seccomp_data args = {}; + args.nr = __NR_close; + args.args[0] = -1; + BPF_ASSERT(SandboxBPF::ForwardSyscall(args) == -EBADF); + BPF_ASSERT(errno == 0); +} + +// Simple test demonstrating how to use SandboxBPF::Cond() + +class SimpleCondTestPolicy : public Policy { + public: + SimpleCondTestPolicy() {} + ~SimpleCondTestPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + DISALLOW_COPY_AND_ASSIGN(SimpleCondTestPolicy); +}; + +ResultExpr SimpleCondTestPolicy::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + + // We deliberately return unusual errno values upon failure, so that we + // can uniquely test for these values. In a "real" policy, you would want + // to return more traditional values. + int flags_argument_position = -1; + switch (sysno) { +#if defined(__NR_open) + case __NR_open: + flags_argument_position = 1; +#endif + case __NR_openat: { // open can be a wrapper for openat(2). + if (sysno == __NR_openat) + flags_argument_position = 2; + + // Allow opening files for reading, but don't allow writing. + static_assert(O_RDONLY == 0, "O_RDONLY must be all zero bits"); + const Arg<int> flags(flags_argument_position); + return If((flags & O_ACCMODE) != 0, Error(EROFS)).Else(Allow()); + } + case __NR_prctl: { + // Allow prctl(PR_SET_DUMPABLE) and prctl(PR_GET_DUMPABLE), but + // disallow everything else. + const Arg<int> option(0); + return If(option == PR_SET_DUMPABLE || option == PR_GET_DUMPABLE, Allow()) + .Else(Error(ENOMEM)); + } + default: + return Allow(); + } +} + +BPF_TEST_C(SandboxBPF, SimpleCondTest, SimpleCondTestPolicy) { + int fd; + BPF_ASSERT((fd = open("/proc/self/comm", O_RDWR)) == -1); + BPF_ASSERT(errno == EROFS); + BPF_ASSERT((fd = open("/proc/self/comm", O_RDONLY)) >= 0); + close(fd); + + int ret; + BPF_ASSERT((ret = prctl(PR_GET_DUMPABLE)) >= 0); + BPF_ASSERT(prctl(PR_SET_DUMPABLE, 1 - ret) == 0); + BPF_ASSERT(prctl(PR_GET_ENDIAN, &ret) == -1); + BPF_ASSERT(errno == ENOMEM); +} + +// This test exercises the SandboxBPF::Cond() method by building a complex +// tree of conditional equality operations. It then makes system calls and +// verifies that they return the values that we expected from our BPF +// program. +class EqualityStressTest { + public: + EqualityStressTest() { + // We want a deterministic test + srand(0); + + // Iterates over system call numbers and builds a random tree of + // equality tests. + // We are actually constructing a graph of ArgValue objects. This + // graph will later be used to a) compute our sandbox policy, and + // b) drive the code that verifies the output from the BPF program. + static_assert( + kNumTestCases < (int)(MAX_PUBLIC_SYSCALL - MIN_SYSCALL - 10), + "kNumTestCases must be significantly smaller than the number " + "of system calls"); + for (int sysno = MIN_SYSCALL, end = kNumTestCases; sysno < end; ++sysno) { + if (IsReservedSyscall(sysno)) { + // Skip reserved system calls. This ensures that our test frame + // work isn't impacted by the fact that we are overriding + // a lot of different system calls. + ++end; + arg_values_.push_back(NULL); + } else { + arg_values_.push_back( + RandomArgValue(rand() % kMaxArgs, 0, rand() % kMaxArgs)); + } + } + } + + ~EqualityStressTest() { + for (std::vector<ArgValue*>::iterator iter = arg_values_.begin(); + iter != arg_values_.end(); + ++iter) { + DeleteArgValue(*iter); + } + } + + ResultExpr Policy(int sysno) { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + if (sysno < 0 || sysno >= (int)arg_values_.size() || + IsReservedSyscall(sysno)) { + // We only return ErrorCode values for the system calls that + // are part of our test data. Every other system call remains + // allowed. + return Allow(); + } else { + // ToErrorCode() turns an ArgValue object into an ErrorCode that is + // suitable for use by a sandbox policy. + return ToErrorCode(arg_values_[sysno]); + } + } + + void VerifyFilter() { + // Iterate over all system calls. Skip the system calls that have + // previously been determined as being reserved. + for (int sysno = 0; sysno < (int)arg_values_.size(); ++sysno) { + if (!arg_values_[sysno]) { + // Skip reserved system calls. + continue; + } + // Verify that system calls return the values that we expect them to + // return. This involves passing different combinations of system call + // parameters in order to exercise all possible code paths through the + // BPF filter program. + // We arbitrarily start by setting all six system call arguments to + // zero. And we then recursive traverse our tree of ArgValues to + // determine the necessary combinations of parameters. + intptr_t args[6] = {}; + Verify(sysno, args, *arg_values_[sysno]); + } + } + + private: + struct ArgValue { + int argno; // Argument number to inspect. + int size; // Number of test cases (must be > 0). + struct Tests { + uint32_t k_value; // Value to compare syscall arg against. + int err; // If non-zero, errno value to return. + struct ArgValue* arg_value; // Otherwise, more args needs inspecting. + }* tests; + int err; // If none of the tests passed, this is what + struct ArgValue* arg_value; // we'll return (this is the "else" branch). + }; + + bool IsReservedSyscall(int sysno) { + // There are a handful of system calls that we should never use in our + // test cases. These system calls are needed to allow the test framework + // to run properly. + // If we wanted to write fully generic code, there are more system calls + // that could be listed here, and it is quite difficult to come up with a + // truly comprehensive list. After all, we are deliberately making system + // calls unavailable. In practice, we have a pretty good idea of the system + // calls that will be made by this particular test. So, this small list is + // sufficient. But if anybody copy'n'pasted this code for other uses, they + // would have to review that the list. + return sysno == __NR_read || sysno == __NR_write || sysno == __NR_exit || + sysno == __NR_exit_group || sysno == __NR_restart_syscall; + } + + ArgValue* RandomArgValue(int argno, int args_mask, int remaining_args) { + // Create a new ArgValue and fill it with random data. We use as bit mask + // to keep track of the system call parameters that have previously been + // set; this ensures that we won't accidentally define a contradictory + // set of equality tests. + struct ArgValue* arg_value = new ArgValue(); + args_mask |= 1 << argno; + arg_value->argno = argno; + + // Apply some restrictions on just how complex our tests can be. + // Otherwise, we end up with a BPF program that is too complicated for + // the kernel to load. + int fan_out = kMaxFanOut; + if (remaining_args > 3) { + fan_out = 1; + } else if (remaining_args > 2) { + fan_out = 2; + } + + // Create a couple of different test cases with randomized values that + // we want to use when comparing system call parameter number "argno". + arg_value->size = rand() % fan_out + 1; + arg_value->tests = new ArgValue::Tests[arg_value->size]; + + uint32_t k_value = rand(); + for (int n = 0; n < arg_value->size; ++n) { + // Ensure that we have unique values + k_value += rand() % (RAND_MAX / (kMaxFanOut + 1)) + 1; + + // There are two possible types of nodes. Either this is a leaf node; + // in that case, we have completed all the equality tests that we + // wanted to perform, and we can now compute a random "errno" value that + // we should return. Or this is part of a more complex boolean + // expression; in that case, we have to recursively add tests for some + // of system call parameters that we have not yet included in our + // tests. + arg_value->tests[n].k_value = k_value; + if (!remaining_args || (rand() & 1)) { + arg_value->tests[n].err = (rand() % 1000) + 1; + arg_value->tests[n].arg_value = NULL; + } else { + arg_value->tests[n].err = 0; + arg_value->tests[n].arg_value = + RandomArgValue(RandomArg(args_mask), args_mask, remaining_args - 1); + } + } + // Finally, we have to define what we should return if none of the + // previous equality tests pass. Again, we can either deal with a leaf + // node, or we can randomly add another couple of tests. + if (!remaining_args || (rand() & 1)) { + arg_value->err = (rand() % 1000) + 1; + arg_value->arg_value = NULL; + } else { + arg_value->err = 0; + arg_value->arg_value = + RandomArgValue(RandomArg(args_mask), args_mask, remaining_args - 1); + } + // We have now built a new (sub-)tree of ArgValues defining a set of + // boolean expressions for testing random system call arguments against + // random values. Return this tree to our caller. + return arg_value; + } + + int RandomArg(int args_mask) { + // Compute a random system call parameter number. + int argno = rand() % kMaxArgs; + + // Make sure that this same parameter number has not previously been + // used. Otherwise, we could end up with a test that is impossible to + // satisfy (e.g. args[0] == 1 && args[0] == 2). + while (args_mask & (1 << argno)) { + argno = (argno + 1) % kMaxArgs; + } + return argno; + } + + void DeleteArgValue(ArgValue* arg_value) { + // Delete an ArgValue and all of its child nodes. This requires + // recursively descending into the tree. + if (arg_value) { + if (arg_value->size) { + for (int n = 0; n < arg_value->size; ++n) { + if (!arg_value->tests[n].err) { + DeleteArgValue(arg_value->tests[n].arg_value); + } + } + delete[] arg_value->tests; + } + if (!arg_value->err) { + DeleteArgValue(arg_value->arg_value); + } + delete arg_value; + } + } + + ResultExpr ToErrorCode(ArgValue* arg_value) { + // Compute the ResultExpr that should be returned, if none of our + // tests succeed (i.e. the system call parameter doesn't match any + // of the values in arg_value->tests[].k_value). + ResultExpr err; + if (arg_value->err) { + // If this was a leaf node, return the errno value that we expect to + // return from the BPF filter program. + err = Error(arg_value->err); + } else { + // If this wasn't a leaf node yet, recursively descend into the rest + // of the tree. This will end up adding a few more SandboxBPF::Cond() + // tests to our ErrorCode. + err = ToErrorCode(arg_value->arg_value); + } + + // Now, iterate over all the test cases that we want to compare against. + // This builds a chain of SandboxBPF::Cond() tests + // (aka "if ... elif ... elif ... elif ... fi") + for (int n = arg_value->size; n-- > 0;) { + ResultExpr matched; + // Again, we distinguish between leaf nodes and subtrees. + if (arg_value->tests[n].err) { + matched = Error(arg_value->tests[n].err); + } else { + matched = ToErrorCode(arg_value->tests[n].arg_value); + } + // For now, all of our tests are limited to 32bit. + // We have separate tests that check the behavior of 32bit vs. 64bit + // conditional expressions. + const Arg<uint32_t> arg(arg_value->argno); + err = If(arg == arg_value->tests[n].k_value, matched).Else(err); + } + return err; + } + + void Verify(int sysno, intptr_t* args, const ArgValue& arg_value) { + uint32_t mismatched = 0; + // Iterate over all the k_values in arg_value.tests[] and verify that + // we see the expected return values from system calls, when we pass + // the k_value as a parameter in a system call. + for (int n = arg_value.size; n-- > 0;) { + mismatched += arg_value.tests[n].k_value; + args[arg_value.argno] = arg_value.tests[n].k_value; + if (arg_value.tests[n].err) { + VerifyErrno(sysno, args, arg_value.tests[n].err); + } else { + Verify(sysno, args, *arg_value.tests[n].arg_value); + } + } + // Find a k_value that doesn't match any of the k_values in + // arg_value.tests[]. In most cases, the current value of "mismatched" + // would fit this requirement. But on the off-chance that it happens + // to collide, we double-check. + try_again: + for (int n = arg_value.size; n-- > 0;) { + if (mismatched == arg_value.tests[n].k_value) { + ++mismatched; + goto try_again; + } + } + // Now verify that we see the expected return value from system calls, + // if we pass a value that doesn't match any of the conditions (i.e. this + // is testing the "else" clause of the conditions). + args[arg_value.argno] = mismatched; + if (arg_value.err) { + VerifyErrno(sysno, args, arg_value.err); + } else { + Verify(sysno, args, *arg_value.arg_value); + } + // Reset args[arg_value.argno]. This is not technically needed, but it + // makes it easier to reason about the correctness of our tests. + args[arg_value.argno] = 0; + } + + void VerifyErrno(int sysno, intptr_t* args, int err) { + // We installed BPF filters that return different errno values + // based on the system call number and the parameters that we decided + // to pass in. Verify that this condition holds true. + BPF_ASSERT( + Syscall::Call( + sysno, args[0], args[1], args[2], args[3], args[4], args[5]) == + -err); + } + + // Vector of ArgValue trees. These trees define all the possible boolean + // expressions that we want to turn into a BPF filter program. + std::vector<ArgValue*> arg_values_; + + // Don't increase these values. We are pushing the limits of the maximum + // BPF program that the kernel will allow us to load. If the values are + // increased too much, the test will start failing. +#if defined(__aarch64__) + static const int kNumTestCases = 30; +#else + static const int kNumTestCases = 40; +#endif + static const int kMaxFanOut = 3; + static const int kMaxArgs = 6; +}; + +class EqualityStressTestPolicy : public Policy { + public: + explicit EqualityStressTestPolicy(EqualityStressTest* aux) : aux_(aux) {} + ~EqualityStressTestPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + return aux_->Policy(sysno); + } + + private: + EqualityStressTest* aux_; + + DISALLOW_COPY_AND_ASSIGN(EqualityStressTestPolicy); +}; + +BPF_TEST(SandboxBPF, + EqualityTests, + EqualityStressTestPolicy, + EqualityStressTest /* (*BPF_AUX) */) { + BPF_AUX->VerifyFilter(); +} + +class EqualityArgumentWidthPolicy : public Policy { + public: + EqualityArgumentWidthPolicy() {} + ~EqualityArgumentWidthPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + DISALLOW_COPY_AND_ASSIGN(EqualityArgumentWidthPolicy); +}; + +ResultExpr EqualityArgumentWidthPolicy::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + if (sysno == __NR_uname) { + const Arg<int> option(0); + const Arg<uint32_t> arg32(1); + const Arg<uint64_t> arg64(1); + return Switch(option) + .Case(0, If(arg32 == 0x55555555, Error(1)).Else(Error(2))) +#if __SIZEOF_POINTER__ > 4 + .Case(1, If(arg64 == 0x55555555AAAAAAAAULL, Error(1)).Else(Error(2))) +#endif + .Default(Error(3)); + } + return Allow(); +} + +BPF_TEST_C(SandboxBPF, EqualityArgumentWidth, EqualityArgumentWidthPolicy) { + BPF_ASSERT(Syscall::Call(__NR_uname, 0, 0x55555555) == -1); + BPF_ASSERT(Syscall::Call(__NR_uname, 0, 0xAAAAAAAA) == -2); +#if __SIZEOF_POINTER__ > 4 + // On 32bit machines, there is no way to pass a 64bit argument through the + // syscall interface. So, we have to skip the part of the test that requires + // 64bit arguments. + BPF_ASSERT(Syscall::Call(__NR_uname, 1, 0x55555555AAAAAAAAULL) == -1); + BPF_ASSERT(Syscall::Call(__NR_uname, 1, 0x5555555500000000ULL) == -2); + BPF_ASSERT(Syscall::Call(__NR_uname, 1, 0x5555555511111111ULL) == -2); + BPF_ASSERT(Syscall::Call(__NR_uname, 1, 0x11111111AAAAAAAAULL) == -2); +#endif +} + +#if __SIZEOF_POINTER__ > 4 +// On 32bit machines, there is no way to pass a 64bit argument through the +// syscall interface. So, we have to skip the part of the test that requires +// 64bit arguments. +BPF_DEATH_TEST_C(SandboxBPF, + EqualityArgumentUnallowed64bit, + DEATH_MESSAGE("Unexpected 64bit argument detected"), + EqualityArgumentWidthPolicy) { + Syscall::Call(__NR_uname, 0, 0x5555555555555555ULL); +} +#endif + +class EqualityWithNegativeArgumentsPolicy : public Policy { + public: + EqualityWithNegativeArgumentsPolicy() {} + ~EqualityWithNegativeArgumentsPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + if (sysno == __NR_uname) { + // TODO(mdempsky): This currently can't be Arg<int> because then + // 0xFFFFFFFF will be treated as a (signed) int, and then when + // Arg::EqualTo casts it to uint64_t, it will be sign extended. + const Arg<unsigned> arg(0); + return If(arg == 0xFFFFFFFF, Error(1)).Else(Error(2)); + } + return Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(EqualityWithNegativeArgumentsPolicy); +}; + +BPF_TEST_C(SandboxBPF, + EqualityWithNegativeArguments, + EqualityWithNegativeArgumentsPolicy) { + BPF_ASSERT(Syscall::Call(__NR_uname, 0xFFFFFFFF) == -1); + BPF_ASSERT(Syscall::Call(__NR_uname, -1) == -1); + BPF_ASSERT(Syscall::Call(__NR_uname, -1LL) == -1); +} + +#if __SIZEOF_POINTER__ > 4 +BPF_DEATH_TEST_C(SandboxBPF, + EqualityWithNegative64bitArguments, + DEATH_MESSAGE("Unexpected 64bit argument detected"), + EqualityWithNegativeArgumentsPolicy) { + // When expecting a 32bit system call argument, we look at the MSB of the + // 64bit value and allow both "0" and "-1". But the latter is allowed only + // iff the LSB was negative. So, this death test should error out. + BPF_ASSERT(Syscall::Call(__NR_uname, 0xFFFFFFFF00000000LL) == -1); +} +#endif + +class AllBitTestPolicy : public Policy { + public: + AllBitTestPolicy() {} + ~AllBitTestPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + static ResultExpr HasAllBits32(uint32_t bits); + static ResultExpr HasAllBits64(uint64_t bits); + + DISALLOW_COPY_AND_ASSIGN(AllBitTestPolicy); +}; + +ResultExpr AllBitTestPolicy::HasAllBits32(uint32_t bits) { + if (bits == 0) { + return Error(1); + } + const Arg<uint32_t> arg(1); + return If((arg & bits) == bits, Error(1)).Else(Error(0)); +} + +ResultExpr AllBitTestPolicy::HasAllBits64(uint64_t bits) { + if (bits == 0) { + return Error(1); + } + const Arg<uint64_t> arg(1); + return If((arg & bits) == bits, Error(1)).Else(Error(0)); +} + +ResultExpr AllBitTestPolicy::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + // Test masked-equality cases that should trigger the "has all bits" + // peephole optimizations. We try to find bitmasks that could conceivably + // touch corner cases. + // For all of these tests, we override the uname(). We can make use with + // a single system call number, as we use the first system call argument to + // select the different bit masks that we want to test against. + if (sysno == __NR_uname) { + const Arg<int> option(0); + return Switch(option) + .Case(0, HasAllBits32(0x0)) + .Case(1, HasAllBits32(0x1)) + .Case(2, HasAllBits32(0x3)) + .Case(3, HasAllBits32(0x80000000)) +#if __SIZEOF_POINTER__ > 4 + .Case(4, HasAllBits64(0x0)) + .Case(5, HasAllBits64(0x1)) + .Case(6, HasAllBits64(0x3)) + .Case(7, HasAllBits64(0x80000000)) + .Case(8, HasAllBits64(0x100000000ULL)) + .Case(9, HasAllBits64(0x300000000ULL)) + .Case(10, HasAllBits64(0x100000001ULL)) +#endif + .Default(Kill("Invalid test case number")); + } + return Allow(); +} + +// Define a macro that performs tests using our test policy. +// NOTE: Not all of the arguments in this macro are actually used! +// They are here just to serve as documentation of the conditions +// implemented in the test policy. +// Most notably, "op" and "mask" are unused by the macro. If you want +// to make changes to these values, you will have to edit the +// test policy instead. +#define BITMASK_TEST(testcase, arg, op, mask, expected_value) \ + BPF_ASSERT(Syscall::Call(__NR_uname, (testcase), (arg)) == (expected_value)) + +// Our uname() system call returns ErrorCode(1) for success and +// ErrorCode(0) for failure. Syscall::Call() turns this into an +// exit code of -1 or 0. +#define EXPECT_FAILURE 0 +#define EXPECT_SUCCESS -1 + +// A couple of our tests behave differently on 32bit and 64bit systems, as +// there is no way for a 32bit system call to pass in a 64bit system call +// argument "arg". +// We expect these tests to succeed on 64bit systems, but to tail on 32bit +// systems. +#define EXPT64_SUCCESS (sizeof(void*) > 4 ? EXPECT_SUCCESS : EXPECT_FAILURE) +BPF_TEST_C(SandboxBPF, AllBitTests, AllBitTestPolicy) { + // 32bit test: all of 0x0 (should always be true) + BITMASK_TEST( 0, 0, ALLBITS32, 0, EXPECT_SUCCESS); + BITMASK_TEST( 0, 1, ALLBITS32, 0, EXPECT_SUCCESS); + BITMASK_TEST( 0, 3, ALLBITS32, 0, EXPECT_SUCCESS); + BITMASK_TEST( 0, 0xFFFFFFFFU, ALLBITS32, 0, EXPECT_SUCCESS); + BITMASK_TEST( 0, -1LL, ALLBITS32, 0, EXPECT_SUCCESS); + + // 32bit test: all of 0x1 + BITMASK_TEST( 1, 0, ALLBITS32, 0x1, EXPECT_FAILURE); + BITMASK_TEST( 1, 1, ALLBITS32, 0x1, EXPECT_SUCCESS); + BITMASK_TEST( 1, 2, ALLBITS32, 0x1, EXPECT_FAILURE); + BITMASK_TEST( 1, 3, ALLBITS32, 0x1, EXPECT_SUCCESS); + + // 32bit test: all of 0x3 + BITMASK_TEST( 2, 0, ALLBITS32, 0x3, EXPECT_FAILURE); + BITMASK_TEST( 2, 1, ALLBITS32, 0x3, EXPECT_FAILURE); + BITMASK_TEST( 2, 2, ALLBITS32, 0x3, EXPECT_FAILURE); + BITMASK_TEST( 2, 3, ALLBITS32, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 2, 7, ALLBITS32, 0x3, EXPECT_SUCCESS); + + // 32bit test: all of 0x80000000 + BITMASK_TEST( 3, 0, ALLBITS32, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 3, 0x40000000U, ALLBITS32, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 3, 0x80000000U, ALLBITS32, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 3, 0xC0000000U, ALLBITS32, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 3, -0x80000000LL, ALLBITS32, 0x80000000, EXPECT_SUCCESS); + +#if __SIZEOF_POINTER__ > 4 + // 64bit test: all of 0x0 (should always be true) + BITMASK_TEST( 4, 0, ALLBITS64, 0, EXPECT_SUCCESS); + BITMASK_TEST( 4, 1, ALLBITS64, 0, EXPECT_SUCCESS); + BITMASK_TEST( 4, 3, ALLBITS64, 0, EXPECT_SUCCESS); + BITMASK_TEST( 4, 0xFFFFFFFFU, ALLBITS64, 0, EXPECT_SUCCESS); + BITMASK_TEST( 4, 0x100000000LL, ALLBITS64, 0, EXPECT_SUCCESS); + BITMASK_TEST( 4, 0x300000000LL, ALLBITS64, 0, EXPECT_SUCCESS); + BITMASK_TEST( 4,0x8000000000000000LL, ALLBITS64, 0, EXPECT_SUCCESS); + BITMASK_TEST( 4, -1LL, ALLBITS64, 0, EXPECT_SUCCESS); + + // 64bit test: all of 0x1 + BITMASK_TEST( 5, 0, ALLBITS64, 1, EXPECT_FAILURE); + BITMASK_TEST( 5, 1, ALLBITS64, 1, EXPECT_SUCCESS); + BITMASK_TEST( 5, 2, ALLBITS64, 1, EXPECT_FAILURE); + BITMASK_TEST( 5, 3, ALLBITS64, 1, EXPECT_SUCCESS); + BITMASK_TEST( 5, 0x100000000LL, ALLBITS64, 1, EXPECT_FAILURE); + BITMASK_TEST( 5, 0x100000001LL, ALLBITS64, 1, EXPECT_SUCCESS); + BITMASK_TEST( 5, 0x100000002LL, ALLBITS64, 1, EXPECT_FAILURE); + BITMASK_TEST( 5, 0x100000003LL, ALLBITS64, 1, EXPECT_SUCCESS); + + // 64bit test: all of 0x3 + BITMASK_TEST( 6, 0, ALLBITS64, 3, EXPECT_FAILURE); + BITMASK_TEST( 6, 1, ALLBITS64, 3, EXPECT_FAILURE); + BITMASK_TEST( 6, 2, ALLBITS64, 3, EXPECT_FAILURE); + BITMASK_TEST( 6, 3, ALLBITS64, 3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 7, ALLBITS64, 3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 0x100000000LL, ALLBITS64, 3, EXPECT_FAILURE); + BITMASK_TEST( 6, 0x100000001LL, ALLBITS64, 3, EXPECT_FAILURE); + BITMASK_TEST( 6, 0x100000002LL, ALLBITS64, 3, EXPECT_FAILURE); + BITMASK_TEST( 6, 0x100000003LL, ALLBITS64, 3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 0x100000007LL, ALLBITS64, 3, EXPECT_SUCCESS); + + // 64bit test: all of 0x80000000 + BITMASK_TEST( 7, 0, ALLBITS64, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 7, 0x40000000U, ALLBITS64, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 7, 0x80000000U, ALLBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, 0xC0000000U, ALLBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, -0x80000000LL, ALLBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, 0x100000000LL, ALLBITS64, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 7, 0x140000000LL, ALLBITS64, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 7, 0x180000000LL, ALLBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, 0x1C0000000LL, ALLBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, -0x180000000LL, ALLBITS64, 0x80000000, EXPECT_SUCCESS); + + // 64bit test: all of 0x100000000 + BITMASK_TEST( 8, 0x000000000LL, ALLBITS64,0x100000000, EXPECT_FAILURE); + BITMASK_TEST( 8, 0x100000000LL, ALLBITS64,0x100000000, EXPT64_SUCCESS); + BITMASK_TEST( 8, 0x200000000LL, ALLBITS64,0x100000000, EXPECT_FAILURE); + BITMASK_TEST( 8, 0x300000000LL, ALLBITS64,0x100000000, EXPT64_SUCCESS); + BITMASK_TEST( 8, 0x000000001LL, ALLBITS64,0x100000000, EXPECT_FAILURE); + BITMASK_TEST( 8, 0x100000001LL, ALLBITS64,0x100000000, EXPT64_SUCCESS); + BITMASK_TEST( 8, 0x200000001LL, ALLBITS64,0x100000000, EXPECT_FAILURE); + BITMASK_TEST( 8, 0x300000001LL, ALLBITS64,0x100000000, EXPT64_SUCCESS); + + // 64bit test: all of 0x300000000 + BITMASK_TEST( 9, 0x000000000LL, ALLBITS64,0x300000000, EXPECT_FAILURE); + BITMASK_TEST( 9, 0x100000000LL, ALLBITS64,0x300000000, EXPECT_FAILURE); + BITMASK_TEST( 9, 0x200000000LL, ALLBITS64,0x300000000, EXPECT_FAILURE); + BITMASK_TEST( 9, 0x300000000LL, ALLBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x700000000LL, ALLBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x000000001LL, ALLBITS64,0x300000000, EXPECT_FAILURE); + BITMASK_TEST( 9, 0x100000001LL, ALLBITS64,0x300000000, EXPECT_FAILURE); + BITMASK_TEST( 9, 0x200000001LL, ALLBITS64,0x300000000, EXPECT_FAILURE); + BITMASK_TEST( 9, 0x300000001LL, ALLBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x700000001LL, ALLBITS64,0x300000000, EXPT64_SUCCESS); + + // 64bit test: all of 0x100000001 + BITMASK_TEST(10, 0x000000000LL, ALLBITS64,0x100000001, EXPECT_FAILURE); + BITMASK_TEST(10, 0x000000001LL, ALLBITS64,0x100000001, EXPECT_FAILURE); + BITMASK_TEST(10, 0x100000000LL, ALLBITS64,0x100000001, EXPECT_FAILURE); + BITMASK_TEST(10, 0x100000001LL, ALLBITS64,0x100000001, EXPT64_SUCCESS); + BITMASK_TEST(10, 0xFFFFFFFFU, ALLBITS64,0x100000001, EXPECT_FAILURE); + BITMASK_TEST(10, -1L, ALLBITS64,0x100000001, EXPT64_SUCCESS); +#endif +} + +class AnyBitTestPolicy : public Policy { + public: + AnyBitTestPolicy() {} + ~AnyBitTestPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + static ResultExpr HasAnyBits32(uint32_t); + static ResultExpr HasAnyBits64(uint64_t); + + DISALLOW_COPY_AND_ASSIGN(AnyBitTestPolicy); +}; + +ResultExpr AnyBitTestPolicy::HasAnyBits32(uint32_t bits) { + if (bits == 0) { + return Error(0); + } + const Arg<uint32_t> arg(1); + return If((arg & bits) != 0, Error(1)).Else(Error(0)); +} + +ResultExpr AnyBitTestPolicy::HasAnyBits64(uint64_t bits) { + if (bits == 0) { + return Error(0); + } + const Arg<uint64_t> arg(1); + return If((arg & bits) != 0, Error(1)).Else(Error(0)); +} + +ResultExpr AnyBitTestPolicy::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + // Test masked-equality cases that should trigger the "has any bits" + // peephole optimizations. We try to find bitmasks that could conceivably + // touch corner cases. + // For all of these tests, we override the uname(). We can make use with + // a single system call number, as we use the first system call argument to + // select the different bit masks that we want to test against. + if (sysno == __NR_uname) { + const Arg<int> option(0); + return Switch(option) + .Case(0, HasAnyBits32(0x0)) + .Case(1, HasAnyBits32(0x1)) + .Case(2, HasAnyBits32(0x3)) + .Case(3, HasAnyBits32(0x80000000)) +#if __SIZEOF_POINTER__ > 4 + .Case(4, HasAnyBits64(0x0)) + .Case(5, HasAnyBits64(0x1)) + .Case(6, HasAnyBits64(0x3)) + .Case(7, HasAnyBits64(0x80000000)) + .Case(8, HasAnyBits64(0x100000000ULL)) + .Case(9, HasAnyBits64(0x300000000ULL)) + .Case(10, HasAnyBits64(0x100000001ULL)) +#endif + .Default(Kill("Invalid test case number")); + } + return Allow(); +} + +BPF_TEST_C(SandboxBPF, AnyBitTests, AnyBitTestPolicy) { + // 32bit test: any of 0x0 (should always be false) + BITMASK_TEST( 0, 0, ANYBITS32, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 0, 1, ANYBITS32, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 0, 3, ANYBITS32, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 0, 0xFFFFFFFFU, ANYBITS32, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 0, -1LL, ANYBITS32, 0x0, EXPECT_FAILURE); + + // 32bit test: any of 0x1 + BITMASK_TEST( 1, 0, ANYBITS32, 0x1, EXPECT_FAILURE); + BITMASK_TEST( 1, 1, ANYBITS32, 0x1, EXPECT_SUCCESS); + BITMASK_TEST( 1, 2, ANYBITS32, 0x1, EXPECT_FAILURE); + BITMASK_TEST( 1, 3, ANYBITS32, 0x1, EXPECT_SUCCESS); + + // 32bit test: any of 0x3 + BITMASK_TEST( 2, 0, ANYBITS32, 0x3, EXPECT_FAILURE); + BITMASK_TEST( 2, 1, ANYBITS32, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 2, 2, ANYBITS32, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 2, 3, ANYBITS32, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 2, 7, ANYBITS32, 0x3, EXPECT_SUCCESS); + + // 32bit test: any of 0x80000000 + BITMASK_TEST( 3, 0, ANYBITS32, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 3, 0x40000000U, ANYBITS32, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 3, 0x80000000U, ANYBITS32, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 3, 0xC0000000U, ANYBITS32, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 3, -0x80000000LL, ANYBITS32, 0x80000000, EXPECT_SUCCESS); + +#if __SIZEOF_POINTER__ > 4 + // 64bit test: any of 0x0 (should always be false) + BITMASK_TEST( 4, 0, ANYBITS64, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 4, 1, ANYBITS64, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 4, 3, ANYBITS64, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 4, 0xFFFFFFFFU, ANYBITS64, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 4, 0x100000000LL, ANYBITS64, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 4, 0x300000000LL, ANYBITS64, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 4,0x8000000000000000LL, ANYBITS64, 0x0, EXPECT_FAILURE); + BITMASK_TEST( 4, -1LL, ANYBITS64, 0x0, EXPECT_FAILURE); + + // 64bit test: any of 0x1 + BITMASK_TEST( 5, 0, ANYBITS64, 0x1, EXPECT_FAILURE); + BITMASK_TEST( 5, 1, ANYBITS64, 0x1, EXPECT_SUCCESS); + BITMASK_TEST( 5, 2, ANYBITS64, 0x1, EXPECT_FAILURE); + BITMASK_TEST( 5, 3, ANYBITS64, 0x1, EXPECT_SUCCESS); + BITMASK_TEST( 5, 0x100000001LL, ANYBITS64, 0x1, EXPECT_SUCCESS); + BITMASK_TEST( 5, 0x100000000LL, ANYBITS64, 0x1, EXPECT_FAILURE); + BITMASK_TEST( 5, 0x100000002LL, ANYBITS64, 0x1, EXPECT_FAILURE); + BITMASK_TEST( 5, 0x100000003LL, ANYBITS64, 0x1, EXPECT_SUCCESS); + + // 64bit test: any of 0x3 + BITMASK_TEST( 6, 0, ANYBITS64, 0x3, EXPECT_FAILURE); + BITMASK_TEST( 6, 1, ANYBITS64, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 2, ANYBITS64, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 3, ANYBITS64, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 7, ANYBITS64, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 0x100000000LL, ANYBITS64, 0x3, EXPECT_FAILURE); + BITMASK_TEST( 6, 0x100000001LL, ANYBITS64, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 0x100000002LL, ANYBITS64, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 0x100000003LL, ANYBITS64, 0x3, EXPECT_SUCCESS); + BITMASK_TEST( 6, 0x100000007LL, ANYBITS64, 0x3, EXPECT_SUCCESS); + + // 64bit test: any of 0x80000000 + BITMASK_TEST( 7, 0, ANYBITS64, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 7, 0x40000000U, ANYBITS64, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 7, 0x80000000U, ANYBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, 0xC0000000U, ANYBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, -0x80000000LL, ANYBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, 0x100000000LL, ANYBITS64, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 7, 0x140000000LL, ANYBITS64, 0x80000000, EXPECT_FAILURE); + BITMASK_TEST( 7, 0x180000000LL, ANYBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, 0x1C0000000LL, ANYBITS64, 0x80000000, EXPECT_SUCCESS); + BITMASK_TEST( 7, -0x180000000LL, ANYBITS64, 0x80000000, EXPECT_SUCCESS); + + // 64bit test: any of 0x100000000 + BITMASK_TEST( 8, 0x000000000LL, ANYBITS64,0x100000000, EXPECT_FAILURE); + BITMASK_TEST( 8, 0x100000000LL, ANYBITS64,0x100000000, EXPT64_SUCCESS); + BITMASK_TEST( 8, 0x200000000LL, ANYBITS64,0x100000000, EXPECT_FAILURE); + BITMASK_TEST( 8, 0x300000000LL, ANYBITS64,0x100000000, EXPT64_SUCCESS); + BITMASK_TEST( 8, 0x000000001LL, ANYBITS64,0x100000000, EXPECT_FAILURE); + BITMASK_TEST( 8, 0x100000001LL, ANYBITS64,0x100000000, EXPT64_SUCCESS); + BITMASK_TEST( 8, 0x200000001LL, ANYBITS64,0x100000000, EXPECT_FAILURE); + BITMASK_TEST( 8, 0x300000001LL, ANYBITS64,0x100000000, EXPT64_SUCCESS); + + // 64bit test: any of 0x300000000 + BITMASK_TEST( 9, 0x000000000LL, ANYBITS64,0x300000000, EXPECT_FAILURE); + BITMASK_TEST( 9, 0x100000000LL, ANYBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x200000000LL, ANYBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x300000000LL, ANYBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x700000000LL, ANYBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x000000001LL, ANYBITS64,0x300000000, EXPECT_FAILURE); + BITMASK_TEST( 9, 0x100000001LL, ANYBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x200000001LL, ANYBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x300000001LL, ANYBITS64,0x300000000, EXPT64_SUCCESS); + BITMASK_TEST( 9, 0x700000001LL, ANYBITS64,0x300000000, EXPT64_SUCCESS); + + // 64bit test: any of 0x100000001 + BITMASK_TEST( 10, 0x000000000LL, ANYBITS64,0x100000001, EXPECT_FAILURE); + BITMASK_TEST( 10, 0x000000001LL, ANYBITS64,0x100000001, EXPECT_SUCCESS); + BITMASK_TEST( 10, 0x100000000LL, ANYBITS64,0x100000001, EXPT64_SUCCESS); + BITMASK_TEST( 10, 0x100000001LL, ANYBITS64,0x100000001, EXPECT_SUCCESS); + BITMASK_TEST( 10, 0xFFFFFFFFU, ANYBITS64,0x100000001, EXPECT_SUCCESS); + BITMASK_TEST( 10, -1L, ANYBITS64,0x100000001, EXPECT_SUCCESS); +#endif +} + +class MaskedEqualTestPolicy : public Policy { + public: + MaskedEqualTestPolicy() {} + ~MaskedEqualTestPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + static ResultExpr MaskedEqual32(uint32_t mask, uint32_t value); + static ResultExpr MaskedEqual64(uint64_t mask, uint64_t value); + + DISALLOW_COPY_AND_ASSIGN(MaskedEqualTestPolicy); +}; + +ResultExpr MaskedEqualTestPolicy::MaskedEqual32(uint32_t mask, uint32_t value) { + const Arg<uint32_t> arg(1); + return If((arg & mask) == value, Error(1)).Else(Error(0)); +} + +ResultExpr MaskedEqualTestPolicy::MaskedEqual64(uint64_t mask, uint64_t value) { + const Arg<uint64_t> arg(1); + return If((arg & mask) == value, Error(1)).Else(Error(0)); +} + +ResultExpr MaskedEqualTestPolicy::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + + if (sysno == __NR_uname) { + const Arg<int> option(0); + return Switch(option) + .Case(0, MaskedEqual32(0x00ff00ff, 0x005500aa)) +#if __SIZEOF_POINTER__ > 4 + .Case(1, MaskedEqual64(0x00ff00ff00000000, 0x005500aa00000000)) + .Case(2, MaskedEqual64(0x00ff00ff00ff00ff, 0x005500aa005500aa)) +#endif + .Default(Kill("Invalid test case number")); + } + + return Allow(); +} + +#define MASKEQ_TEST(rulenum, arg, expected_result) \ + BPF_ASSERT(Syscall::Call(__NR_uname, (rulenum), (arg)) == (expected_result)) + +BPF_TEST_C(SandboxBPF, MaskedEqualTests, MaskedEqualTestPolicy) { + // Allowed: 0x__55__aa + MASKEQ_TEST(0, 0x00000000, EXPECT_FAILURE); + MASKEQ_TEST(0, 0x00000001, EXPECT_FAILURE); + MASKEQ_TEST(0, 0x00000003, EXPECT_FAILURE); + MASKEQ_TEST(0, 0x00000100, EXPECT_FAILURE); + MASKEQ_TEST(0, 0x00000300, EXPECT_FAILURE); + MASKEQ_TEST(0, 0x005500aa, EXPECT_SUCCESS); + MASKEQ_TEST(0, 0x005500ab, EXPECT_FAILURE); + MASKEQ_TEST(0, 0x005600aa, EXPECT_FAILURE); + MASKEQ_TEST(0, 0x005501aa, EXPECT_SUCCESS); + MASKEQ_TEST(0, 0x005503aa, EXPECT_SUCCESS); + MASKEQ_TEST(0, 0x555500aa, EXPECT_SUCCESS); + MASKEQ_TEST(0, 0xaa5500aa, EXPECT_SUCCESS); + +#if __SIZEOF_POINTER__ > 4 + // Allowed: 0x__55__aa________ + MASKEQ_TEST(1, 0x0000000000000000, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x0000000000000010, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x0000000000000050, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x0000000100000000, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x0000000300000000, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x0000010000000000, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x0000030000000000, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x005500aa00000000, EXPECT_SUCCESS); + MASKEQ_TEST(1, 0x005500ab00000000, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x005600aa00000000, EXPECT_FAILURE); + MASKEQ_TEST(1, 0x005501aa00000000, EXPECT_SUCCESS); + MASKEQ_TEST(1, 0x005503aa00000000, EXPECT_SUCCESS); + MASKEQ_TEST(1, 0x555500aa00000000, EXPECT_SUCCESS); + MASKEQ_TEST(1, 0xaa5500aa00000000, EXPECT_SUCCESS); + MASKEQ_TEST(1, 0xaa5500aa00000000, EXPECT_SUCCESS); + MASKEQ_TEST(1, 0xaa5500aa0000cafe, EXPECT_SUCCESS); + + // Allowed: 0x__55__aa__55__aa + MASKEQ_TEST(2, 0x0000000000000000, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x0000000000000010, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x0000000000000050, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x0000000100000000, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x0000000300000000, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x0000010000000000, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x0000030000000000, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x00000000005500aa, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x005500aa00000000, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x005500aa005500aa, EXPECT_SUCCESS); + MASKEQ_TEST(2, 0x005500aa005700aa, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x005700aa005500aa, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x005500aa004500aa, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x004500aa005500aa, EXPECT_FAILURE); + MASKEQ_TEST(2, 0x005512aa005500aa, EXPECT_SUCCESS); + MASKEQ_TEST(2, 0x005500aa005534aa, EXPECT_SUCCESS); + MASKEQ_TEST(2, 0xff5500aa0055ffaa, EXPECT_SUCCESS); +#endif +} + +intptr_t PthreadTrapHandler(const struct arch_seccomp_data& args, void* aux) { + if (args.args[0] != (CLONE_CHILD_CLEARTID | CLONE_CHILD_SETTID | SIGCHLD)) { + // We expect to get called for an attempt to fork(). No need to log that + // call. But if we ever get called for anything else, we want to verbosely + // print as much information as possible. + const char* msg = (const char*)aux; + printf( + "Clone() was called with unexpected arguments\n" + " nr: %d\n" + " 1: 0x%llX\n" + " 2: 0x%llX\n" + " 3: 0x%llX\n" + " 4: 0x%llX\n" + " 5: 0x%llX\n" + " 6: 0x%llX\n" + "%s\n", + args.nr, + (long long)args.args[0], + (long long)args.args[1], + (long long)args.args[2], + (long long)args.args[3], + (long long)args.args[4], + (long long)args.args[5], + msg); + } + return -EPERM; +} + +class PthreadPolicyEquality : public Policy { + public: + PthreadPolicyEquality() {} + ~PthreadPolicyEquality() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + DISALLOW_COPY_AND_ASSIGN(PthreadPolicyEquality); +}; + +ResultExpr PthreadPolicyEquality::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + // This policy allows creating threads with pthread_create(). But it + // doesn't allow any other uses of clone(). Most notably, it does not + // allow callers to implement fork() or vfork() by passing suitable flags + // to the clone() system call. + if (sysno == __NR_clone) { + // We have seen two different valid combinations of flags. Glibc + // uses the more modern flags, sets the TLS from the call to clone(), and + // uses futexes to monitor threads. Android's C run-time library, doesn't + // do any of this, but it sets the obsolete (and no-op) CLONE_DETACHED. + // More recent versions of Android don't set CLONE_DETACHED anymore, so + // the last case accounts for that. + // The following policy is very strict. It only allows the exact masks + // that we have seen in known implementations. It is probably somewhat + // stricter than what we would want to do. + const uint64_t kGlibcCloneMask = CLONE_VM | CLONE_FS | CLONE_FILES | + CLONE_SIGHAND | CLONE_THREAD | + CLONE_SYSVSEM | CLONE_SETTLS | + CLONE_PARENT_SETTID | CLONE_CHILD_CLEARTID; + const uint64_t kBaseAndroidCloneMask = CLONE_VM | CLONE_FS | CLONE_FILES | + CLONE_SIGHAND | CLONE_THREAD | + CLONE_SYSVSEM; + const Arg<unsigned long> flags(0); + return If(flags == kGlibcCloneMask || + flags == (kBaseAndroidCloneMask | CLONE_DETACHED) || + flags == kBaseAndroidCloneMask, + Allow()).Else(Trap(PthreadTrapHandler, "Unknown mask")); + } + + return Allow(); +} + +class PthreadPolicyBitMask : public Policy { + public: + PthreadPolicyBitMask() {} + ~PthreadPolicyBitMask() override {} + + ResultExpr EvaluateSyscall(int sysno) const override; + + private: + static BoolExpr HasAnyBits(const Arg<unsigned long>& arg, unsigned long bits); + static BoolExpr HasAllBits(const Arg<unsigned long>& arg, unsigned long bits); + + DISALLOW_COPY_AND_ASSIGN(PthreadPolicyBitMask); +}; + +BoolExpr PthreadPolicyBitMask::HasAnyBits(const Arg<unsigned long>& arg, + unsigned long bits) { + return (arg & bits) != 0; +} + +BoolExpr PthreadPolicyBitMask::HasAllBits(const Arg<unsigned long>& arg, + unsigned long bits) { + return (arg & bits) == bits; +} + +ResultExpr PthreadPolicyBitMask::EvaluateSyscall(int sysno) const { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + // This policy allows creating threads with pthread_create(). But it + // doesn't allow any other uses of clone(). Most notably, it does not + // allow callers to implement fork() or vfork() by passing suitable flags + // to the clone() system call. + if (sysno == __NR_clone) { + // We have seen two different valid combinations of flags. Glibc + // uses the more modern flags, sets the TLS from the call to clone(), and + // uses futexes to monitor threads. Android's C run-time library, doesn't + // do any of this, but it sets the obsolete (and no-op) CLONE_DETACHED. + // The following policy allows for either combination of flags, but it + // is generally a little more conservative than strictly necessary. We + // err on the side of rather safe than sorry. + // Very noticeably though, we disallow fork() (which is often just a + // wrapper around clone()). + const unsigned long kMandatoryFlags = CLONE_VM | CLONE_FS | CLONE_FILES | + CLONE_SIGHAND | CLONE_THREAD | + CLONE_SYSVSEM; + const unsigned long kFutexFlags = + CLONE_SETTLS | CLONE_PARENT_SETTID | CLONE_CHILD_CLEARTID; + const unsigned long kNoopFlags = CLONE_DETACHED; + const unsigned long kKnownFlags = + kMandatoryFlags | kFutexFlags | kNoopFlags; + + const Arg<unsigned long> flags(0); + return If(HasAnyBits(flags, ~kKnownFlags), + Trap(PthreadTrapHandler, "Unexpected CLONE_XXX flag found")) + .ElseIf(!HasAllBits(flags, kMandatoryFlags), + Trap(PthreadTrapHandler, + "Missing mandatory CLONE_XXX flags " + "when creating new thread")) + .ElseIf( + !HasAllBits(flags, kFutexFlags) && HasAnyBits(flags, kFutexFlags), + Trap(PthreadTrapHandler, + "Must set either all or none of the TLS and futex bits in " + "call to clone()")) + .Else(Allow()); + } + + return Allow(); +} + +static void* ThreadFnc(void* arg) { + ++*reinterpret_cast<int*>(arg); + Syscall::Call(__NR_futex, arg, FUTEX_WAKE, 1, 0, 0, 0); + return NULL; +} + +static void PthreadTest() { + // Attempt to start a joinable thread. This should succeed. + pthread_t thread; + int thread_ran = 0; + BPF_ASSERT(!pthread_create(&thread, NULL, ThreadFnc, &thread_ran)); + BPF_ASSERT(!pthread_join(thread, NULL)); + BPF_ASSERT(thread_ran); + + // Attempt to start a detached thread. This should succeed. + thread_ran = 0; + pthread_attr_t attr; + BPF_ASSERT(!pthread_attr_init(&attr)); + BPF_ASSERT(!pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED)); + BPF_ASSERT(!pthread_create(&thread, &attr, ThreadFnc, &thread_ran)); + BPF_ASSERT(!pthread_attr_destroy(&attr)); + while (Syscall::Call(__NR_futex, &thread_ran, FUTEX_WAIT, 0, 0, 0, 0) == + -EINTR) { + } + BPF_ASSERT(thread_ran); + + // Attempt to fork() a process using clone(). This should fail. We use the + // same flags that glibc uses when calling fork(). But we don't actually + // try calling the fork() implementation in the C run-time library, as + // run-time libraries other than glibc might call __NR_fork instead of + // __NR_clone, and that would introduce a bogus test failure. + int pid; + BPF_ASSERT(Syscall::Call(__NR_clone, + CLONE_CHILD_CLEARTID | CLONE_CHILD_SETTID | SIGCHLD, + 0, + 0, + &pid) == -EPERM); +} + +BPF_TEST_C(SandboxBPF, PthreadEquality, PthreadPolicyEquality) { + PthreadTest(); +} + +BPF_TEST_C(SandboxBPF, PthreadBitMask, PthreadPolicyBitMask) { + PthreadTest(); +} + +// libc might not define these even though the kernel supports it. +#ifndef PTRACE_O_TRACESECCOMP +#define PTRACE_O_TRACESECCOMP 0x00000080 +#endif + +#ifdef PTRACE_EVENT_SECCOMP +#define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP) +#else +// When Debian/Ubuntu backported seccomp-bpf support into earlier kernels, they +// changed the value of PTRACE_EVENT_SECCOMP from 7 to 8, since 7 was taken by +// PTRACE_EVENT_STOP (upstream chose to renumber PTRACE_EVENT_STOP to 128). If +// PTRACE_EVENT_SECCOMP isn't defined, we have no choice but to consider both +// values here. +#define IS_SECCOMP_EVENT(status) ((status >> 16) == 7 || (status >> 16) == 8) +#endif + +#if defined(__arm__) +#ifndef PTRACE_SET_SYSCALL +#define PTRACE_SET_SYSCALL 23 +#endif +#endif + +#if defined(__aarch64__) +#ifndef PTRACE_GETREGS +#define PTRACE_GETREGS 12 +#endif +#endif + +#if defined(__aarch64__) +#ifndef PTRACE_SETREGS +#define PTRACE_SETREGS 13 +#endif +#endif + +// Changes the syscall to run for a child being sandboxed using seccomp-bpf with +// PTRACE_O_TRACESECCOMP. Should only be called when the child is stopped on +// PTRACE_EVENT_SECCOMP. +// +// regs should contain the current set of registers of the child, obtained using +// PTRACE_GETREGS. +// +// Depending on the architecture, this may modify regs, so the caller is +// responsible for committing these changes using PTRACE_SETREGS. +long SetSyscall(pid_t pid, regs_struct* regs, int syscall_number) { +#if defined(__arm__) + // On ARM, the syscall is changed using PTRACE_SET_SYSCALL. We cannot use the + // libc ptrace call as the request parameter is an enum, and + // PTRACE_SET_SYSCALL may not be in the enum. + return syscall(__NR_ptrace, PTRACE_SET_SYSCALL, pid, NULL, syscall_number); +#endif + + SECCOMP_PT_SYSCALL(*regs) = syscall_number; + return 0; +} + +const uint16_t kTraceData = 0xcc; + +class TraceAllPolicy : public Policy { + public: + TraceAllPolicy() {} + ~TraceAllPolicy() override {} + + ResultExpr EvaluateSyscall(int system_call_number) const override { + return Trace(kTraceData); + } + + private: + DISALLOW_COPY_AND_ASSIGN(TraceAllPolicy); +}; + +SANDBOX_TEST(SandboxBPF, DISABLE_ON_TSAN(SeccompRetTrace)) { + if (!SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::SINGLE_THREADED)) { + return; + } + +// This test is disabled on arm due to a kernel bug. +// See https://code.google.com/p/chromium/issues/detail?id=383977 +#if defined(__arm__) || defined(__aarch64__) + printf("This test is currently disabled on ARM32/64 due to a kernel bug."); + return; +#endif + +#if defined(__mips__) + // TODO: Figure out how to support specificity of handling indirect syscalls + // in this test and enable it. + printf("This test is currently disabled on MIPS."); + return; +#endif + + pid_t pid = fork(); + BPF_ASSERT_NE(-1, pid); + if (pid == 0) { + pid_t my_pid = getpid(); + BPF_ASSERT_NE(-1, ptrace(PTRACE_TRACEME, -1, NULL, NULL)); + BPF_ASSERT_EQ(0, raise(SIGSTOP)); + SandboxBPF sandbox(new TraceAllPolicy); + BPF_ASSERT(sandbox.StartSandbox(SandboxBPF::SeccompLevel::SINGLE_THREADED)); + + // getpid is allowed. + BPF_ASSERT_EQ(my_pid, sys_getpid()); + + // write to stdout is skipped and returns a fake value. + BPF_ASSERT_EQ(kExpectedReturnValue, + syscall(__NR_write, STDOUT_FILENO, "A", 1)); + + // kill is rewritten to exit(kExpectedReturnValue). + syscall(__NR_kill, my_pid, SIGKILL); + + // Should not be reached. + BPF_ASSERT(false); + } + + int status; + BPF_ASSERT(HANDLE_EINTR(waitpid(pid, &status, WUNTRACED)) != -1); + BPF_ASSERT(WIFSTOPPED(status)); + + BPF_ASSERT_NE(-1, + ptrace(PTRACE_SETOPTIONS, + pid, + NULL, + reinterpret_cast<void*>(PTRACE_O_TRACESECCOMP))); + BPF_ASSERT_NE(-1, ptrace(PTRACE_CONT, pid, NULL, NULL)); + while (true) { + BPF_ASSERT(HANDLE_EINTR(waitpid(pid, &status, 0)) != -1); + if (WIFEXITED(status) || WIFSIGNALED(status)) { + BPF_ASSERT(WIFEXITED(status)); + BPF_ASSERT_EQ(kExpectedReturnValue, WEXITSTATUS(status)); + break; + } + + if (!WIFSTOPPED(status) || WSTOPSIG(status) != SIGTRAP || + !IS_SECCOMP_EVENT(status)) { + BPF_ASSERT_NE(-1, ptrace(PTRACE_CONT, pid, NULL, NULL)); + continue; + } + + unsigned long data; + BPF_ASSERT_NE(-1, ptrace(PTRACE_GETEVENTMSG, pid, NULL, &data)); + BPF_ASSERT_EQ(kTraceData, data); + + regs_struct regs; + BPF_ASSERT_NE(-1, ptrace(PTRACE_GETREGS, pid, NULL, ®s)); + switch (SECCOMP_PT_SYSCALL(regs)) { + case __NR_write: + // Skip writes to stdout, make it return kExpectedReturnValue. Allow + // writes to stderr so that BPF_ASSERT messages show up. + if (SECCOMP_PT_PARM1(regs) == STDOUT_FILENO) { + BPF_ASSERT_NE(-1, SetSyscall(pid, ®s, -1)); + SECCOMP_PT_RESULT(regs) = kExpectedReturnValue; + BPF_ASSERT_NE(-1, ptrace(PTRACE_SETREGS, pid, NULL, ®s)); + } + break; + + case __NR_kill: + // Rewrite to exit(kExpectedReturnValue). + BPF_ASSERT_NE(-1, SetSyscall(pid, ®s, __NR_exit)); + SECCOMP_PT_PARM1(regs) = kExpectedReturnValue; + BPF_ASSERT_NE(-1, ptrace(PTRACE_SETREGS, pid, NULL, ®s)); + break; + + default: + // Allow all other syscalls. + break; + } + + BPF_ASSERT_NE(-1, ptrace(PTRACE_CONT, pid, NULL, NULL)); + } +} + +// Android does not expose pread64 nor pwrite64. +#if !defined(OS_ANDROID) + +bool FullPwrite64(int fd, const char* buffer, size_t count, off64_t offset) { + while (count > 0) { + const ssize_t transfered = + HANDLE_EINTR(pwrite64(fd, buffer, count, offset)); + if (transfered <= 0 || static_cast<size_t>(transfered) > count) { + return false; + } + count -= transfered; + buffer += transfered; + offset += transfered; + } + return true; +} + +bool FullPread64(int fd, char* buffer, size_t count, off64_t offset) { + while (count > 0) { + const ssize_t transfered = HANDLE_EINTR(pread64(fd, buffer, count, offset)); + if (transfered <= 0 || static_cast<size_t>(transfered) > count) { + return false; + } + count -= transfered; + buffer += transfered; + offset += transfered; + } + return true; +} + +bool pread_64_was_forwarded = false; + +class TrapPread64Policy : public Policy { + public: + TrapPread64Policy() {} + ~TrapPread64Policy() override {} + + ResultExpr EvaluateSyscall(int system_call_number) const override { + // Set the global environment for unsafe traps once. + if (system_call_number == MIN_SYSCALL) { + EnableUnsafeTraps(); + } + + if (system_call_number == __NR_pread64) { + return UnsafeTrap(ForwardPreadHandler, NULL); + } + return Allow(); + } + + private: + static intptr_t ForwardPreadHandler(const struct arch_seccomp_data& args, + void* aux) { + BPF_ASSERT(args.nr == __NR_pread64); + pread_64_was_forwarded = true; + + return SandboxBPF::ForwardSyscall(args); + } + + DISALLOW_COPY_AND_ASSIGN(TrapPread64Policy); +}; + +// pread(2) takes a 64 bits offset. On 32 bits systems, it will be split +// between two arguments. In this test, we make sure that ForwardSyscall() can +// forward it properly. +BPF_TEST_C(SandboxBPF, Pread64, TrapPread64Policy) { + ScopedTemporaryFile temp_file; + const uint64_t kLargeOffset = (static_cast<uint64_t>(1) << 32) | 0xBEEF; + const char kTestString[] = "This is a test!"; + BPF_ASSERT(FullPwrite64( + temp_file.fd(), kTestString, sizeof(kTestString), kLargeOffset)); + + char read_test_string[sizeof(kTestString)] = {0}; + BPF_ASSERT(FullPread64(temp_file.fd(), + read_test_string, + sizeof(read_test_string), + kLargeOffset)); + BPF_ASSERT_EQ(0, memcmp(kTestString, read_test_string, sizeof(kTestString))); + BPF_ASSERT(pread_64_was_forwarded); +} + +#endif // !defined(OS_ANDROID) + +void* TsyncApplyToTwoThreadsFunc(void* cond_ptr) { + base::WaitableEvent* event = static_cast<base::WaitableEvent*>(cond_ptr); + + // Wait for the main thread to signal that the filter has been applied. + if (!event->IsSignaled()) { + event->Wait(); + } + + BPF_ASSERT(event->IsSignaled()); + + BlacklistNanosleepPolicy::AssertNanosleepFails(); + + return NULL; +} + +SANDBOX_TEST(SandboxBPF, Tsync) { + const bool supports_multi_threaded = SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::MULTI_THREADED); +// On Chrome OS tsync is mandatory. +#if defined(OS_CHROMEOS) + if (base::SysInfo::IsRunningOnChromeOS()) { + BPF_ASSERT_EQ(true, supports_multi_threaded); + } +// else a Chrome OS build not running on a Chrome OS device e.g. Chrome bots. +// In this case fall through. +#endif + if (!supports_multi_threaded) { + return; + } + + base::WaitableEvent event(true, false); + + // Create a thread on which to invoke the blocked syscall. + pthread_t thread; + BPF_ASSERT_EQ( + 0, pthread_create(&thread, NULL, &TsyncApplyToTwoThreadsFunc, &event)); + + // Test that nanoseelp success. + const struct timespec ts = {0, 0}; + BPF_ASSERT_EQ(0, HANDLE_EINTR(syscall(__NR_nanosleep, &ts, NULL))); + + // Engage the sandbox. + SandboxBPF sandbox(new BlacklistNanosleepPolicy()); + BPF_ASSERT(sandbox.StartSandbox(SandboxBPF::SeccompLevel::MULTI_THREADED)); + + // This thread should have the filter applied as well. + BlacklistNanosleepPolicy::AssertNanosleepFails(); + + // Signal the condition to invoke the system call. + event.Signal(); + + // Wait for the thread to finish. + BPF_ASSERT_EQ(0, pthread_join(thread, NULL)); +} + +class AllowAllPolicy : public Policy { + public: + AllowAllPolicy() {} + ~AllowAllPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { return Allow(); } + + private: + DISALLOW_COPY_AND_ASSIGN(AllowAllPolicy); +}; + +SANDBOX_DEATH_TEST( + SandboxBPF, + StartMultiThreadedAsSingleThreaded, + DEATH_MESSAGE( + ThreadHelpers::GetAssertSingleThreadedErrorMessageForTests())) { + base::Thread thread("sandbox.linux.StartMultiThreadedAsSingleThreaded"); + BPF_ASSERT(thread.Start()); + + SandboxBPF sandbox(new AllowAllPolicy()); + BPF_ASSERT(!sandbox.StartSandbox(SandboxBPF::SeccompLevel::SINGLE_THREADED)); +} + +// http://crbug.com/407357 +#if !defined(THREAD_SANITIZER) +SANDBOX_DEATH_TEST( + SandboxBPF, + StartSingleThreadedAsMultiThreaded, + DEATH_MESSAGE( + "Cannot start sandbox; process may be single-threaded when " + "reported as not")) { + SandboxBPF sandbox(new AllowAllPolicy()); + BPF_ASSERT(!sandbox.StartSandbox(SandboxBPF::SeccompLevel::MULTI_THREADED)); +} +#endif // !defined(THREAD_SANITIZER) + +// A stub handler for the UnsafeTrap. Never called. +intptr_t NoOpHandler(const struct arch_seccomp_data& args, void*) { + return -1; +} + +class UnsafeTrapWithCondPolicy : public Policy { + public: + UnsafeTrapWithCondPolicy() {} + ~UnsafeTrapWithCondPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + setenv(kSandboxDebuggingEnv, "t", 0); + Die::SuppressInfoMessages(true); + + if (SandboxBPF::IsRequiredForUnsafeTrap(sysno)) + return Allow(); + + switch (sysno) { + case __NR_uname: { + const Arg<uint32_t> arg(0); + return If(arg == 0, Allow()).Else(Error(EPERM)); + } + case __NR_setgid: { + const Arg<uint32_t> arg(0); + return Switch(arg) + .Case(100, Error(ENOMEM)) + .Case(200, Error(ENOSYS)) + .Default(Error(EPERM)); + } + case __NR_close: + case __NR_exit_group: + case __NR_write: + return Allow(); + case __NR_getppid: + return UnsafeTrap(NoOpHandler, NULL); + default: + return Error(EPERM); + } + } + + private: + DISALLOW_COPY_AND_ASSIGN(UnsafeTrapWithCondPolicy); +}; + +BPF_TEST_C(SandboxBPF, UnsafeTrapWithCond, UnsafeTrapWithCondPolicy) { + BPF_ASSERT_EQ(-1, syscall(__NR_uname, 0)); + BPF_ASSERT_EQ(EFAULT, errno); + + BPF_ASSERT_EQ(-1, syscall(__NR_uname, 1)); + BPF_ASSERT_EQ(EPERM, errno); + + BPF_ASSERT_EQ(-1, syscall(__NR_setgid, 100)); + BPF_ASSERT_EQ(ENOMEM, errno); + + BPF_ASSERT_EQ(-1, syscall(__NR_setgid, 200)); + BPF_ASSERT_EQ(ENOSYS, errno); + + BPF_ASSERT_EQ(-1, syscall(__NR_setgid, 300)); + BPF_ASSERT_EQ(EPERM, errno); +} + +} // namespace + +} // namespace bpf_dsl +} // namespace sandbox diff --git a/sandbox/linux/integration_tests/namespace_unix_domain_socket_unittest.cc b/sandbox/linux/integration_tests/namespace_unix_domain_socket_unittest.cc new file mode 100644 index 0000000000..9d79bff1c6 --- /dev/null +++ b/sandbox/linux/integration_tests/namespace_unix_domain_socket_unittest.cc @@ -0,0 +1,267 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include <sched.h> +#include <stdio.h> +#include <string.h> +#include <sys/socket.h> +#include <sys/syscall.h> +#include <sys/wait.h> +#include <unistd.h> + +#include <vector> + +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/memory/scoped_vector.h" +#include "base/posix/eintr_wrapper.h" +#include "base/posix/unix_domain_socket_linux.h" +#include "base/process/process.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/tests/unit_tests.h" + +// Additional tests for base's UnixDomainSocket to make sure it behaves +// correctly in the presence of sandboxing functionality (e.g., receiving +// PIDs across namespaces). + +namespace sandbox { + +namespace { + +const char kHello[] = "hello"; + +// If the calling process isn't root, then try using unshare(CLONE_NEWUSER) +// to fake it. +void FakeRoot() { + // If we're already root, then allow test to proceed. + if (geteuid() == 0) + return; + + // Otherwise hope the kernel supports unprivileged namespaces. + if (unshare(CLONE_NEWUSER) == 0) + return; + + printf("Permission to use CLONE_NEWPID missing; skipping test.\n"); + UnitTests::IgnoreThisTest(); +} + +void WaitForExit(pid_t pid) { + int status; + CHECK_EQ(pid, HANDLE_EINTR(waitpid(pid, &status, 0))); + CHECK(WIFEXITED(status)); + CHECK_EQ(0, WEXITSTATUS(status)); +} + +base::ProcessId GetParentProcessId(base::ProcessId pid) { + // base::GetParentProcessId() is defined as taking a ProcessHandle instead of + // a ProcessId, even though it's a POSIX-only function and IDs and Handles + // are both simply pid_t on POSIX... :/ + base::Process process = base::Process::Open(pid); + CHECK(process.IsValid()); + base::ProcessId ret = base::GetParentProcessId(process.Handle()); + return ret; +} + +// SendHello sends a "hello" to socket fd, and then blocks until the recipient +// acknowledges it by calling RecvHello. +void SendHello(int fd) { + int pipe_fds[2]; + CHECK_EQ(0, pipe(pipe_fds)); + base::ScopedFD read_pipe(pipe_fds[0]); + base::ScopedFD write_pipe(pipe_fds[1]); + + std::vector<int> send_fds; + send_fds.push_back(write_pipe.get()); + CHECK(base::UnixDomainSocket::SendMsg(fd, kHello, sizeof(kHello), send_fds)); + + write_pipe.reset(); + + // Block until receiver closes their end of the pipe. + char ch; + CHECK_EQ(0, HANDLE_EINTR(read(read_pipe.get(), &ch, 1))); +} + +// RecvHello receives and acknowledges a "hello" on socket fd, and returns the +// process ID of the sender in sender_pid. Optionally, write_pipe can be used +// to return a file descriptor, and the acknowledgement will be delayed until +// the descriptor is closed. +// (Implementation details: SendHello allocates a new pipe, sends us the writing +// end alongside the "hello" message, and then blocks until we close the writing +// end of the pipe.) +void RecvHello(int fd, + base::ProcessId* sender_pid, + base::ScopedFD* write_pipe = NULL) { + // Extra receiving buffer space to make sure we really received only + // sizeof(kHello) bytes and it wasn't just truncated to fit the buffer. + char buf[sizeof(kHello) + 1]; + ScopedVector<base::ScopedFD> message_fds; + ssize_t n = base::UnixDomainSocket::RecvMsgWithPid( + fd, buf, sizeof(buf), &message_fds, sender_pid); + CHECK_EQ(sizeof(kHello), static_cast<size_t>(n)); + CHECK_EQ(0, memcmp(buf, kHello, sizeof(kHello))); + CHECK_EQ(1U, message_fds.size()); + if (write_pipe) + write_pipe->swap(*message_fds[0]); +} + +// Check that receiving PIDs works across a fork(). +SANDBOX_TEST(UnixDomainSocketTest, Fork) { + int fds[2]; + CHECK_EQ(0, socketpair(AF_UNIX, SOCK_SEQPACKET, 0, fds)); + base::ScopedFD recv_sock(fds[0]); + base::ScopedFD send_sock(fds[1]); + + CHECK(base::UnixDomainSocket::EnableReceiveProcessId(recv_sock.get())); + + const pid_t pid = fork(); + CHECK_NE(-1, pid); + if (pid == 0) { + // Child process. + recv_sock.reset(); + SendHello(send_sock.get()); + _exit(0); + } + + // Parent process. + send_sock.reset(); + + base::ProcessId sender_pid; + RecvHello(recv_sock.get(), &sender_pid); + CHECK_EQ(pid, sender_pid); + + WaitForExit(pid); +} + +// Similar to Fork above, but forking the child into a new pid namespace. +SANDBOX_TEST(UnixDomainSocketTest, Namespace) { + FakeRoot(); + + int fds[2]; + CHECK_EQ(0, socketpair(AF_UNIX, SOCK_SEQPACKET, 0, fds)); + base::ScopedFD recv_sock(fds[0]); + base::ScopedFD send_sock(fds[1]); + + CHECK(base::UnixDomainSocket::EnableReceiveProcessId(recv_sock.get())); + + const pid_t pid = sys_clone(CLONE_NEWPID | SIGCHLD, 0, 0, 0, 0); + CHECK_NE(-1, pid); + if (pid == 0) { + // Child process. + recv_sock.reset(); + + // Check that we think we're pid 1 in our new namespace. + CHECK_EQ(1, sys_getpid()); + + SendHello(send_sock.get()); + _exit(0); + } + + // Parent process. + send_sock.reset(); + + base::ProcessId sender_pid; + RecvHello(recv_sock.get(), &sender_pid); + CHECK_EQ(pid, sender_pid); + + WaitForExit(pid); +} + +// Again similar to Fork, but now with nested PID namespaces. +SANDBOX_TEST(UnixDomainSocketTest, DoubleNamespace) { + FakeRoot(); + + int fds[2]; + CHECK_EQ(0, socketpair(AF_UNIX, SOCK_SEQPACKET, 0, fds)); + base::ScopedFD recv_sock(fds[0]); + base::ScopedFD send_sock(fds[1]); + + CHECK(base::UnixDomainSocket::EnableReceiveProcessId(recv_sock.get())); + + const pid_t pid = sys_clone(CLONE_NEWPID | SIGCHLD, 0, 0, 0, 0); + CHECK_NE(-1, pid); + if (pid == 0) { + // Child process. + recv_sock.reset(); + + const pid_t pid2 = sys_clone(CLONE_NEWPID | SIGCHLD, 0, 0, 0, 0); + CHECK_NE(-1, pid2); + + if (pid2 != 0) { + // Wait for grandchild to run to completion; see comments below. + WaitForExit(pid2); + + // Fallthrough once grandchild has sent its hello and exited. + } + + // Check that we think we're pid 1. + CHECK_EQ(1, sys_getpid()); + + SendHello(send_sock.get()); + _exit(0); + } + + // Parent process. + send_sock.reset(); + + // We have two messages to receive: first from the grand-child, + // then from the child. + for (unsigned iteration = 0; iteration < 2; ++iteration) { + base::ProcessId sender_pid; + base::ScopedFD pipe_fd; + RecvHello(recv_sock.get(), &sender_pid, &pipe_fd); + + // We need our child and grandchild processes to both be alive for + // GetParentProcessId() to return a valid pid, hence the pipe trickery. + // (On the first iteration, grandchild is blocked reading from the pipe + // until we close it, and child is blocked waiting for grandchild to exit.) + switch (iteration) { + case 0: // Grandchild's message + // Check that sender_pid refers to our grandchild by checking that pid + // (our child) is its parent. + CHECK_EQ(pid, GetParentProcessId(sender_pid)); + break; + case 1: // Child's message + CHECK_EQ(pid, sender_pid); + break; + default: + NOTREACHED(); + } + } + + WaitForExit(pid); +} + +// Tests that GetPeerPid() returns 0 if the peer does not exist in caller's +// namespace. +SANDBOX_TEST(UnixDomainSocketTest, ImpossiblePid) { + FakeRoot(); + + int fds[2]; + CHECK_EQ(0, socketpair(AF_UNIX, SOCK_SEQPACKET, 0, fds)); + base::ScopedFD send_sock(fds[0]); + base::ScopedFD recv_sock(fds[1]); + + CHECK(base::UnixDomainSocket::EnableReceiveProcessId(recv_sock.get())); + + const pid_t pid = sys_clone(CLONE_NEWPID | SIGCHLD, 0, 0, 0, 0); + CHECK_NE(-1, pid); + if (pid == 0) { + // Child process. + send_sock.reset(); + + base::ProcessId sender_pid; + RecvHello(recv_sock.get(), &sender_pid); + CHECK_EQ(0, sender_pid); + _exit(0); + } + + // Parent process. + recv_sock.reset(); + SendHello(send_sock.get()); + WaitForExit(pid); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/integration_tests/seccomp_broker_process_unittest.cc b/sandbox/linux/integration_tests/seccomp_broker_process_unittest.cc new file mode 100644 index 0000000000..9aa320997b --- /dev/null +++ b/sandbox/linux/integration_tests/seccomp_broker_process_unittest.cc @@ -0,0 +1,180 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include <sys/types.h> +#include <sys/stat.h> +#include <fcntl.h> +#include <unistd.h> + +#include <vector> + +#include "base/bind.h" +#include "base/memory/scoped_ptr.h" +#include "base/posix/eintr_wrapper.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" +#include "sandbox/linux/seccomp-bpf/bpf_tests.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/syscall_broker/broker_file_permission.h" +#include "sandbox/linux/syscall_broker/broker_process.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +using bpf_dsl::Allow; +using bpf_dsl::ResultExpr; +using bpf_dsl::Trap; + +bool NoOpCallback() { + return true; +} + +// Test a trap handler that makes use of a broker process to open(). + +class InitializedOpenBroker { + public: + InitializedOpenBroker() : initialized_(false) { + std::vector<syscall_broker::BrokerFilePermission> permissions; + permissions.push_back( + syscall_broker::BrokerFilePermission::ReadOnly("/proc/allowed")); + permissions.push_back( + syscall_broker::BrokerFilePermission::ReadOnly("/proc/cpuinfo")); + + broker_process_.reset( + new syscall_broker::BrokerProcess(EPERM, permissions)); + BPF_ASSERT(broker_process() != NULL); + BPF_ASSERT(broker_process_->Init(base::Bind(&NoOpCallback))); + + initialized_ = true; + } + bool initialized() { return initialized_; } + class syscall_broker::BrokerProcess* broker_process() { + return broker_process_.get(); + } + + private: + bool initialized_; + scoped_ptr<class syscall_broker::BrokerProcess> broker_process_; + DISALLOW_COPY_AND_ASSIGN(InitializedOpenBroker); +}; + +intptr_t BrokerOpenTrapHandler(const struct arch_seccomp_data& args, + void* aux) { + BPF_ASSERT(aux); + syscall_broker::BrokerProcess* broker_process = + static_cast<syscall_broker::BrokerProcess*>(aux); + switch (args.nr) { + case __NR_faccessat: // access is a wrapper of faccessat in android + BPF_ASSERT(static_cast<int>(args.args[0]) == AT_FDCWD); + return broker_process->Access(reinterpret_cast<const char*>(args.args[1]), + static_cast<int>(args.args[2])); +#if defined(__NR_access) + case __NR_access: + return broker_process->Access(reinterpret_cast<const char*>(args.args[0]), + static_cast<int>(args.args[1])); +#endif +#if defined(__NR_open) + case __NR_open: + return broker_process->Open(reinterpret_cast<const char*>(args.args[0]), + static_cast<int>(args.args[1])); +#endif + case __NR_openat: + // We only call open() so if we arrive here, it's because glibc uses + // the openat() system call. + BPF_ASSERT(static_cast<int>(args.args[0]) == AT_FDCWD); + return broker_process->Open(reinterpret_cast<const char*>(args.args[1]), + static_cast<int>(args.args[2])); + default: + BPF_ASSERT(false); + return -ENOSYS; + } +} + +class DenyOpenPolicy : public bpf_dsl::Policy { + public: + explicit DenyOpenPolicy(InitializedOpenBroker* iob) : iob_(iob) {} + ~DenyOpenPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + + switch (sysno) { + case __NR_faccessat: +#if defined(__NR_access) + case __NR_access: +#endif +#if defined(__NR_open) + case __NR_open: +#endif + case __NR_openat: + // We get a InitializedOpenBroker class, but our trap handler wants + // the syscall_broker::BrokerProcess object. + return Trap(BrokerOpenTrapHandler, iob_->broker_process()); + default: + return Allow(); + } + } + + private: + InitializedOpenBroker* iob_; + + DISALLOW_COPY_AND_ASSIGN(DenyOpenPolicy); +}; + +// We use a InitializedOpenBroker class, so that we can run unsandboxed +// code in its constructor, which is the only way to do so in a BPF_TEST. +BPF_TEST(SandboxBPF, + UseOpenBroker, + DenyOpenPolicy, + InitializedOpenBroker /* (*BPF_AUX) */) { + BPF_ASSERT(BPF_AUX->initialized()); + syscall_broker::BrokerProcess* broker_process = BPF_AUX->broker_process(); + BPF_ASSERT(broker_process != NULL); + + // First, use the broker "manually" + BPF_ASSERT(broker_process->Open("/proc/denied", O_RDONLY) == -EPERM); + BPF_ASSERT(broker_process->Access("/proc/denied", R_OK) == -EPERM); + BPF_ASSERT(broker_process->Open("/proc/allowed", O_RDONLY) == -ENOENT); + BPF_ASSERT(broker_process->Access("/proc/allowed", R_OK) == -ENOENT); + + // Now use glibc's open() as an external library would. + BPF_ASSERT(open("/proc/denied", O_RDONLY) == -1); + BPF_ASSERT(errno == EPERM); + + BPF_ASSERT(open("/proc/allowed", O_RDONLY) == -1); + BPF_ASSERT(errno == ENOENT); + + // Also test glibc's openat(), some versions of libc use it transparently + // instead of open(). + BPF_ASSERT(openat(AT_FDCWD, "/proc/denied", O_RDONLY) == -1); + BPF_ASSERT(errno == EPERM); + + BPF_ASSERT(openat(AT_FDCWD, "/proc/allowed", O_RDONLY) == -1); + BPF_ASSERT(errno == ENOENT); + + // And test glibc's access(). + BPF_ASSERT(access("/proc/denied", R_OK) == -1); + BPF_ASSERT(errno == EPERM); + + BPF_ASSERT(access("/proc/allowed", R_OK) == -1); + BPF_ASSERT(errno == ENOENT); + + // This is also white listed and does exist. + int cpu_info_access = access("/proc/cpuinfo", R_OK); + BPF_ASSERT(cpu_info_access == 0); + int cpu_info_fd = open("/proc/cpuinfo", O_RDONLY); + BPF_ASSERT(cpu_info_fd >= 0); + char buf[1024]; + BPF_ASSERT(read(cpu_info_fd, buf, sizeof(buf)) > 0); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/sandbox_linux.gypi b/sandbox/linux/sandbox_linux.gypi new file mode 100644 index 0000000000..a7bd259d8a --- /dev/null +++ b/sandbox/linux/sandbox_linux.gypi @@ -0,0 +1,416 @@ +# Copyright (c) 2012 The Chromium Authors. All rights reserved. +# Use of this source code is governed by a BSD-style license that can be +# found in the LICENSE file. + +{ + 'variables': { + 'conditions': [ + ['OS=="linux"', { + 'compile_suid_client': 1, + 'compile_credentials': 1, + 'use_base_test_suite': 1, + }, { + 'compile_suid_client': 0, + 'compile_credentials': 0, + 'use_base_test_suite': 0, + }], + ['OS=="linux" and (target_arch=="ia32" or target_arch=="x64" or ' + 'target_arch=="mipsel")', { + 'compile_seccomp_bpf_demo': 1, + }, { + 'compile_seccomp_bpf_demo': 0, + }], + ], + }, + 'target_defaults': { + 'target_conditions': [ + # All linux/ files will automatically be excluded on Android + # so make sure we re-include them explicitly. + ['OS == "android"', { + 'sources/': [ + ['include', '^linux/'], + ], + }], + ], + }, + 'targets': [ + # We have two principal targets: sandbox and sandbox_linux_unittests + # All other targets are listed as dependencies. + # There is one notable exception: for historical reasons, chrome_sandbox is + # the setuid sandbox and is its own target. + { + 'target_name': 'sandbox', + 'type': 'none', + 'dependencies': [ + 'sandbox_services', + ], + 'conditions': [ + [ 'compile_suid_client==1', { + 'dependencies': [ + 'suid_sandbox_client', + ], + }], + # Compile seccomp BPF when we support it. + [ 'use_seccomp_bpf==1', { + 'dependencies': [ + 'seccomp_bpf', + 'seccomp_bpf_helpers', + ], + }], + ], + }, + { + 'target_name': 'sandbox_linux_test_utils', + 'type': 'static_library', + 'dependencies': [ + '../testing/gtest.gyp:gtest', + ], + 'include_dirs': [ + '../..', + ], + 'sources': [ + 'tests/sandbox_test_runner.cc', + 'tests/sandbox_test_runner.h', + 'tests/sandbox_test_runner_function_pointer.cc', + 'tests/sandbox_test_runner_function_pointer.h', + 'tests/test_utils.cc', + 'tests/test_utils.h', + 'tests/unit_tests.cc', + 'tests/unit_tests.h', + ], + 'conditions': [ + [ 'use_seccomp_bpf==1', { + 'sources': [ + 'seccomp-bpf/bpf_tester_compatibility_delegate.h', + 'seccomp-bpf/bpf_tests.h', + 'seccomp-bpf/sandbox_bpf_test_runner.cc', + 'seccomp-bpf/sandbox_bpf_test_runner.h', + ], + 'dependencies': [ + 'seccomp_bpf', + ] + }], + [ 'use_base_test_suite==1', { + 'dependencies': [ + '../base/base.gyp:test_support_base', + ], + 'defines': [ + 'SANDBOX_USES_BASE_TEST_SUITE', + ], + }], + ], + }, + { + # The main sandboxing test target. + 'target_name': 'sandbox_linux_unittests', + 'includes': [ + 'sandbox_linux_test_sources.gypi', + ], + 'type': 'executable', + }, + { + # This target is the shared library used by Android APK (i.e. + # JNI-friendly) tests. + 'target_name': 'sandbox_linux_jni_unittests', + 'includes': [ + 'sandbox_linux_test_sources.gypi', + ], + 'type': 'shared_library', + 'conditions': [ + [ 'OS == "android"', { + 'dependencies': [ + '../testing/android/native_test.gyp:native_test_native_code', + ], + }], + ], + }, + { + 'target_name': 'seccomp_bpf', + 'type': '<(component)', + 'sources': [ + 'bpf_dsl/bpf_dsl.cc', + 'bpf_dsl/bpf_dsl.h', + 'bpf_dsl/bpf_dsl_forward.h', + 'bpf_dsl/bpf_dsl_impl.h', + 'bpf_dsl/codegen.cc', + 'bpf_dsl/codegen.h', + 'bpf_dsl/cons.h', + 'bpf_dsl/dump_bpf.cc', + 'bpf_dsl/dump_bpf.h', + 'bpf_dsl/linux_syscall_ranges.h', + 'bpf_dsl/policy.cc', + 'bpf_dsl/policy.h', + 'bpf_dsl/policy_compiler.cc', + 'bpf_dsl/policy_compiler.h', + 'bpf_dsl/seccomp_macros.h', + 'bpf_dsl/seccomp_macros.h', + 'bpf_dsl/syscall_set.cc', + 'bpf_dsl/syscall_set.h', + 'bpf_dsl/trap_registry.h', + 'bpf_dsl/verifier.cc', + 'bpf_dsl/verifier.h', + 'seccomp-bpf/die.cc', + 'seccomp-bpf/die.h', + 'seccomp-bpf/errorcode.cc', + 'seccomp-bpf/errorcode.h', + 'seccomp-bpf/sandbox_bpf.cc', + 'seccomp-bpf/sandbox_bpf.h', + 'seccomp-bpf/syscall.cc', + 'seccomp-bpf/syscall.h', + 'seccomp-bpf/trap.cc', + 'seccomp-bpf/trap.h', + ], + 'dependencies': [ + '../base/base.gyp:base', + 'sandbox_services', + 'sandbox_services_headers', + ], + 'defines': [ + 'SANDBOX_IMPLEMENTATION', + ], + 'includes': [ + # Disable LTO due to compiler bug + # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57703 + '../../build/android/disable_lto.gypi', + ], + 'include_dirs': [ + '../..', + ], + }, + { + 'target_name': 'seccomp_bpf_helpers', + 'type': '<(component)', + 'sources': [ + 'seccomp-bpf-helpers/baseline_policy.cc', + 'seccomp-bpf-helpers/baseline_policy.h', + 'seccomp-bpf-helpers/sigsys_handlers.cc', + 'seccomp-bpf-helpers/sigsys_handlers.h', + 'seccomp-bpf-helpers/syscall_parameters_restrictions.cc', + 'seccomp-bpf-helpers/syscall_parameters_restrictions.h', + 'seccomp-bpf-helpers/syscall_sets.cc', + 'seccomp-bpf-helpers/syscall_sets.h', + ], + 'dependencies': [ + '../base/base.gyp:base', + 'sandbox_services', + 'seccomp_bpf', + ], + 'defines': [ + 'SANDBOX_IMPLEMENTATION', + ], + 'include_dirs': [ + '../..', + ], + }, + { + # The setuid sandbox, for Linux + 'target_name': 'chrome_sandbox', + 'type': 'executable', + 'sources': [ + 'suid/common/sandbox.h', + 'suid/common/suid_unsafe_environment_variables.h', + 'suid/process_util.h', + 'suid/process_util_linux.c', + 'suid/sandbox.c', + ], + 'cflags': [ + # For ULLONG_MAX + '-std=gnu99', + ], + 'include_dirs': [ + '../..', + ], + # Do not use any sanitizer tools with this binary. http://crbug.com/382766 + 'cflags/': [ + ['exclude', '-fsanitize'], + ], + 'ldflags/': [ + ['exclude', '-fsanitize'], + ], + }, + { 'target_name': 'sandbox_services', + 'type': '<(component)', + 'sources': [ + 'services/init_process_reaper.cc', + 'services/init_process_reaper.h', + 'services/proc_util.cc', + 'services/proc_util.h', + 'services/resource_limits.cc', + 'services/resource_limits.h', + 'services/scoped_process.cc', + 'services/scoped_process.h', + 'services/syscall_wrappers.cc', + 'services/syscall_wrappers.h', + 'services/thread_helpers.cc', + 'services/thread_helpers.h', + 'services/yama.cc', + 'services/yama.h', + 'syscall_broker/broker_channel.cc', + 'syscall_broker/broker_channel.h', + 'syscall_broker/broker_client.cc', + 'syscall_broker/broker_client.h', + 'syscall_broker/broker_common.h', + 'syscall_broker/broker_file_permission.cc', + 'syscall_broker/broker_file_permission.h', + 'syscall_broker/broker_host.cc', + 'syscall_broker/broker_host.h', + 'syscall_broker/broker_policy.cc', + 'syscall_broker/broker_policy.h', + 'syscall_broker/broker_process.cc', + 'syscall_broker/broker_process.h', + ], + 'dependencies': [ + '../base/base.gyp:base', + ], + 'defines': [ + 'SANDBOX_IMPLEMENTATION', + ], + 'conditions': [ + ['compile_credentials==1', { + 'sources': [ + 'services/credentials.cc', + 'services/credentials.h', + 'services/namespace_sandbox.cc', + 'services/namespace_sandbox.h', + 'services/namespace_utils.cc', + 'services/namespace_utils.h', + ], + 'dependencies': [ + # for capability.h. + 'sandbox_services_headers', + ], + }], + ], + 'include_dirs': [ + '..', + ], + }, + { 'target_name': 'sandbox_services_headers', + 'type': 'none', + 'sources': [ + 'system_headers/arm64_linux_syscalls.h', + 'system_headers/arm64_linux_ucontext.h', + 'system_headers/arm_linux_syscalls.h', + 'system_headers/arm_linux_ucontext.h', + 'system_headers/capability.h', + 'system_headers/i386_linux_ucontext.h', + 'system_headers/linux_futex.h', + 'system_headers/linux_seccomp.h', + 'system_headers/linux_syscalls.h', + 'system_headers/linux_time.h', + 'system_headers/linux_ucontext.h', + 'system_headers/mips_linux_syscalls.h', + 'system_headers/mips_linux_ucontext.h', + 'system_headers/x86_32_linux_syscalls.h', + 'system_headers/x86_64_linux_syscalls.h', + ], + 'include_dirs': [ + '..', + ], + }, + { + # We make this its own target so that it does not interfere + # with our tests. + 'target_name': 'libc_urandom_override', + 'type': 'static_library', + 'sources': [ + 'services/libc_urandom_override.cc', + 'services/libc_urandom_override.h', + ], + 'dependencies': [ + '../base/base.gyp:base', + ], + 'include_dirs': [ + '..', + ], + }, + { + 'target_name': 'suid_sandbox_client', + 'type': '<(component)', + 'sources': [ + 'suid/common/sandbox.h', + 'suid/common/suid_unsafe_environment_variables.h', + 'suid/client/setuid_sandbox_client.cc', + 'suid/client/setuid_sandbox_client.h', + 'suid/client/setuid_sandbox_host.cc', + 'suid/client/setuid_sandbox_host.h', + ], + 'defines': [ + 'SANDBOX_IMPLEMENTATION', + ], + 'dependencies': [ + '../base/base.gyp:base', + 'sandbox_services', + ], + 'include_dirs': [ + '..', + ], + }, + ], + 'conditions': [ + [ 'OS=="android"', { + 'targets': [ + { + 'target_name': 'sandbox_linux_unittests_stripped', + 'type': 'none', + 'dependencies': [ 'sandbox_linux_unittests' ], + 'actions': [{ + 'action_name': 'strip sandbox_linux_unittests', + 'inputs': [ '<(PRODUCT_DIR)/sandbox_linux_unittests' ], + 'outputs': [ '<(PRODUCT_DIR)/sandbox_linux_unittests_stripped' ], + 'action': [ '<(android_strip)', '<@(_inputs)', '-o', '<@(_outputs)' ], + }], + }, + { + 'target_name': 'sandbox_linux_unittests_deps', + 'type': 'none', + 'dependencies': [ + 'sandbox_linux_unittests_stripped', + ], + # For the component build, ensure dependent shared libraries are + # stripped and put alongside sandbox_linux_unittests to simplify pushing + # to the device. + 'variables': { + 'output_dir': '<(PRODUCT_DIR)/sandbox_linux_unittests_deps/', + 'native_binary': '<(PRODUCT_DIR)/sandbox_linux_unittests_stripped', + 'include_main_binary': 0, + }, + 'includes': [ + '../../build/android/native_app_dependencies.gypi' + ], + }], + }], + [ 'OS=="android"', { + 'targets': [ + { + 'target_name': 'sandbox_linux_jni_unittests_apk', + 'type': 'none', + 'variables': { + 'test_suite_name': 'sandbox_linux_jni_unittests', + }, + 'dependencies': [ + 'sandbox_linux_jni_unittests', + ], + 'includes': [ '../../build/apk_test.gypi' ], + } + ], + }], + ['test_isolation_mode != "noop"', { + 'targets': [ + { + 'target_name': 'sandbox_linux_unittests_run', + 'type': 'none', + 'dependencies': [ + 'sandbox_linux_unittests', + ], + 'includes': [ + '../../build/isolate.gypi', + ], + 'sources': [ + '../sandbox_linux_unittests.isolate', + ], + }, + ], + }], + ], +} diff --git a/sandbox/linux/sandbox_linux_nacl_nonsfi.gyp b/sandbox/linux/sandbox_linux_nacl_nonsfi.gyp new file mode 100644 index 0000000000..87ad06ccdc --- /dev/null +++ b/sandbox/linux/sandbox_linux_nacl_nonsfi.gyp @@ -0,0 +1,88 @@ +# Copyright 2015 The Chromium Authors. All rights reserved. +# Use of this source code is governed by a BSD-style license that can be +# found in the LICENSE file. + +{ + 'variables': { + 'chromium_code': 1, + }, + 'includes': [ + '../../build/common_untrusted.gypi', + ], + 'conditions': [ + ['disable_nacl==0 and disable_nacl_untrusted==0', { + 'targets': [ + { + 'target_name': 'sandbox_linux_nacl_nonsfi', + 'type': 'none', + 'variables': { + 'nacl_untrusted_build': 1, + 'nlib_target': 'libsandbox_linux_nacl_nonsfi.a', + 'build_glibc': 0, + 'build_newlib': 0, + 'build_irt': 0, + 'build_pnacl_newlib': 0, + 'build_nonsfi_helper': 1, + + 'sources': [ + # This is the subset of linux build target, needed for + # nacl_helper_nonsfi's sandbox implementation. + 'bpf_dsl/bpf_dsl.cc', + 'bpf_dsl/codegen.cc', + 'bpf_dsl/dump_bpf.cc', + 'bpf_dsl/policy.cc', + 'bpf_dsl/policy_compiler.cc', + 'bpf_dsl/syscall_set.cc', + 'bpf_dsl/verifier.cc', + 'seccomp-bpf-helpers/sigsys_handlers.cc', + 'seccomp-bpf-helpers/syscall_parameters_restrictions.cc', + 'seccomp-bpf/die.cc', + 'seccomp-bpf/errorcode.cc', + 'seccomp-bpf/sandbox_bpf.cc', + 'seccomp-bpf/syscall.cc', + 'seccomp-bpf/trap.cc', + 'services/credentials.cc', + 'services/namespace_sandbox.cc', + 'services/namespace_utils.cc', + 'services/proc_util.cc', + 'services/resource_limits.cc', + 'services/syscall_wrappers.cc', + 'services/thread_helpers.cc', + 'suid/client/setuid_sandbox_client.cc', + ], + }, + 'dependencies': [ + '../../base/base_nacl.gyp:base_nacl_nonsfi', + ], + }, + ], + }], + + ['disable_nacl==0 and disable_nacl_untrusted==0 and enable_nacl_nonsfi_test==1', { + 'targets': [ + { + 'target_name': 'sandbox_linux_test_utils_nacl_nonsfi', + 'type': 'none', + 'variables': { + 'nacl_untrusted_build': 1, + 'nlib_target': 'libsandbox_linux_test_utils_nacl_nonsfi.a', + 'build_glibc': 0, + 'build_newlib': 0, + 'build_irt': 0, + 'build_pnacl_newlib': 0, + 'build_nonsfi_helper': 1, + + 'sources': [ + 'seccomp-bpf/sandbox_bpf_test_runner.cc', + 'tests/sandbox_test_runner.cc', + 'tests/unit_tests.cc', + ], + }, + 'dependencies': [ + '../../testing/gtest_nacl.gyp:gtest_nacl', + ], + }, + ], + }], + ], +} diff --git a/sandbox/linux/sandbox_linux_test_sources.gypi b/sandbox/linux/sandbox_linux_test_sources.gypi new file mode 100644 index 0000000000..82d7532056 --- /dev/null +++ b/sandbox/linux/sandbox_linux_test_sources.gypi @@ -0,0 +1,84 @@ +# Copyright (c) 2012 The Chromium Authors. All rights reserved. +# Use of this source code is governed by a BSD-style license that can be +# found in the LICENSE file. + +# Tests need to be compiled in the same link unit, so we have to list them +# in a separate .gypi file. +{ + 'dependencies': [ + 'sandbox', + 'sandbox_linux_test_utils', + 'sandbox_services', + '../base/base.gyp:base', + '../testing/gtest.gyp:gtest', + ], + 'include_dirs': [ + '../..', + ], + 'sources': [ + 'services/proc_util_unittest.cc', + 'services/scoped_process_unittest.cc', + 'services/resource_limits_unittests.cc', + 'services/syscall_wrappers_unittest.cc', + 'services/thread_helpers_unittests.cc', + 'services/yama_unittests.cc', + 'syscall_broker/broker_file_permission_unittest.cc', + 'syscall_broker/broker_process_unittest.cc', + 'tests/main.cc', + 'tests/scoped_temporary_file.cc', + 'tests/scoped_temporary_file.h', + 'tests/scoped_temporary_file_unittest.cc', + 'tests/test_utils_unittest.cc', + 'tests/unit_tests_unittest.cc', + ], + 'conditions': [ + [ 'compile_suid_client==1', { + 'sources': [ + 'suid/client/setuid_sandbox_client_unittest.cc', + 'suid/client/setuid_sandbox_host_unittest.cc', + ], + }], + [ 'use_seccomp_bpf==1', { + 'sources': [ + 'bpf_dsl/bpf_dsl_unittest.cc', + 'bpf_dsl/codegen_unittest.cc', + 'bpf_dsl/cons_unittest.cc', + 'bpf_dsl/syscall_set_unittest.cc', + 'integration_tests/bpf_dsl_seccomp_unittest.cc', + 'integration_tests/seccomp_broker_process_unittest.cc', + 'seccomp-bpf-helpers/baseline_policy_unittest.cc', + 'seccomp-bpf-helpers/syscall_parameters_restrictions_unittests.cc', + 'seccomp-bpf/bpf_tests_unittest.cc', + 'seccomp-bpf/errorcode_unittest.cc', + 'seccomp-bpf/sandbox_bpf_unittest.cc', + 'seccomp-bpf/syscall_unittest.cc', + 'seccomp-bpf/trap_unittest.cc', + ], + }], + [ 'compile_credentials==1', { + 'sources': [ + 'integration_tests/namespace_unix_domain_socket_unittest.cc', + 'services/credentials_unittest.cc', + 'services/namespace_utils_unittest.cc', + ], + 'dependencies': [ + '../build/linux/system.gyp:libcap' + ], + 'conditions': [ + [ 'use_base_test_suite==1', { + 'sources': [ + 'services/namespace_sandbox_unittest.cc', + ] + }] + ], + }], + [ 'use_base_test_suite==1', { + 'dependencies': [ + '../base/base.gyp:test_support_base', + ], + 'defines': [ + 'SANDBOX_USES_BASE_TEST_SUITE', + ], + }], + ], +} diff --git a/sandbox/linux/seccomp-bpf-helpers/DEPS b/sandbox/linux/seccomp-bpf-helpers/DEPS new file mode 100644 index 0000000000..4419fd1da3 --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/DEPS @@ -0,0 +1,7 @@ +include_rules = [ + "+sandbox/linux/bpf_dsl", + "+sandbox/linux/seccomp-bpf", + "+sandbox/linux/services", + "+sandbox/linux/system_headers", + "+third_party/lss/linux_syscall_support.h", +] diff --git a/sandbox/linux/seccomp-bpf-helpers/baseline_policy.cc b/sandbox/linux/seccomp-bpf-helpers/baseline_policy.cc new file mode 100644 index 0000000000..8c679a3d41 --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/baseline_policy.cc @@ -0,0 +1,270 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf-helpers/baseline_policy.h" + +#include <errno.h> +#include <sys/mman.h> +#include <sys/socket.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <unistd.h> + +#include "base/logging.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.h" +#include "sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.h" +#include "sandbox/linux/seccomp-bpf-helpers/syscall_sets.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" + +// Changing this implementation will have an effect on *all* policies. +// Currently this means: Renderer/Worker, GPU, Flash and NaCl. + +using sandbox::bpf_dsl::Allow; +using sandbox::bpf_dsl::Arg; +using sandbox::bpf_dsl::Error; +using sandbox::bpf_dsl::If; +using sandbox::bpf_dsl::ResultExpr; + +namespace sandbox { + +namespace { + +bool IsBaselinePolicyAllowed(int sysno) { + return SyscallSets::IsAllowedAddressSpaceAccess(sysno) || + SyscallSets::IsAllowedBasicScheduler(sysno) || + SyscallSets::IsAllowedEpoll(sysno) || + SyscallSets::IsAllowedFileSystemAccessViaFd(sysno) || + SyscallSets::IsAllowedFutex(sysno) || + SyscallSets::IsAllowedGeneralIo(sysno) || + SyscallSets::IsAllowedGetOrModifySocket(sysno) || + SyscallSets::IsAllowedGettime(sysno) || + SyscallSets::IsAllowedProcessStartOrDeath(sysno) || + SyscallSets::IsAllowedSignalHandling(sysno) || + SyscallSets::IsGetSimpleId(sysno) || + SyscallSets::IsKernelInternalApi(sysno) || +#if defined(__arm__) + SyscallSets::IsArmPrivate(sysno) || +#endif +#if defined(__mips__) + SyscallSets::IsMipsPrivate(sysno) || +#endif + SyscallSets::IsAllowedOperationOnFd(sysno); +} + +// System calls that will trigger the crashing SIGSYS handler. +bool IsBaselinePolicyWatched(int sysno) { + return SyscallSets::IsAdminOperation(sysno) || + SyscallSets::IsAdvancedScheduler(sysno) || + SyscallSets::IsAdvancedTimer(sysno) || + SyscallSets::IsAsyncIo(sysno) || + SyscallSets::IsDebug(sysno) || + SyscallSets::IsEventFd(sysno) || + SyscallSets::IsExtendedAttributes(sysno) || + SyscallSets::IsFaNotify(sysno) || + SyscallSets::IsFsControl(sysno) || + SyscallSets::IsGlobalFSViewChange(sysno) || + SyscallSets::IsGlobalProcessEnvironment(sysno) || + SyscallSets::IsGlobalSystemStatus(sysno) || + SyscallSets::IsInotify(sysno) || + SyscallSets::IsKernelModule(sysno) || + SyscallSets::IsKeyManagement(sysno) || + SyscallSets::IsKill(sysno) || + SyscallSets::IsMessageQueue(sysno) || + SyscallSets::IsMisc(sysno) || +#if defined(__x86_64__) + SyscallSets::IsNetworkSocketInformation(sysno) || +#endif + SyscallSets::IsNuma(sysno) || + SyscallSets::IsPrctl(sysno) || + SyscallSets::IsProcessGroupOrSession(sysno) || +#if defined(__i386__) || defined(__mips__) + SyscallSets::IsSocketCall(sysno) || +#endif +#if defined(__arm__) + SyscallSets::IsArmPciConfig(sysno) || +#endif +#if defined(__mips__) + SyscallSets::IsMipsMisc(sysno) || +#endif + SyscallSets::IsTimer(sysno); +} + +// |fs_denied_errno| is the errno return for denied filesystem access. +ResultExpr EvaluateSyscallImpl(int fs_denied_errno, + pid_t current_pid, + int sysno) { +#if defined(ADDRESS_SANITIZER) || defined(THREAD_SANITIZER) || \ + defined(MEMORY_SANITIZER) + // TCGETS is required by the sanitizers on failure. + if (sysno == __NR_ioctl) { + return RestrictIoctl(); + } + + if (sysno == __NR_sched_getaffinity) { + return Allow(); + } + + // Used when RSS limiting is enabled in sanitizers. + if (sysno == __NR_getrusage) { + return RestrictGetrusage(); + } + + if (sysno == __NR_sigaltstack) { + // Required for better stack overflow detection in ASan. Disallowed in + // non-ASan builds. + return Allow(); + } +#endif // defined(ADDRESS_SANITIZER) || defined(THREAD_SANITIZER) || + // defined(MEMORY_SANITIZER) + + if (IsBaselinePolicyAllowed(sysno)) { + return Allow(); + } + +#if defined(OS_ANDROID) + // Needed for thread creation. + if (sysno == __NR_sigaltstack) + return Allow(); +#endif + + if (sysno == __NR_clock_gettime) { + return RestrictClockID(); + } + + if (sysno == __NR_clone) { + return RestrictCloneToThreadsAndEPERMFork(); + } + + if (sysno == __NR_fcntl) + return RestrictFcntlCommands(); + +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + if (sysno == __NR_fcntl64) + return RestrictFcntlCommands(); +#endif + +#if !defined(__aarch64__) + // fork() is never used as a system call (clone() is used instead), but we + // have seen it in fallback code on Android. + if (sysno == __NR_fork) { + return Error(EPERM); + } +#endif + + if (sysno == __NR_futex) + return RestrictFutex(); + + if (sysno == __NR_set_robust_list) + return Error(EPERM); + + if (sysno == __NR_getpriority || sysno ==__NR_setpriority) + return RestrictGetSetpriority(current_pid); + + if (sysno == __NR_madvise) { + // Only allow MADV_DONTNEED (aka MADV_FREE). + const Arg<int> advice(2); + return If(advice == MADV_DONTNEED, Allow()).Else(Error(EPERM)); + } + +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) || \ + defined(__aarch64__) + if (sysno == __NR_mmap) + return RestrictMmapFlags(); +#endif + +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + if (sysno == __NR_mmap2) + return RestrictMmapFlags(); +#endif + + if (sysno == __NR_mprotect) + return RestrictMprotectFlags(); + + if (sysno == __NR_prctl) + return RestrictPrctl(); + +#if defined(__x86_64__) || defined(__arm__) || defined(__mips__) || \ + defined(__aarch64__) + if (sysno == __NR_socketpair) { + // Only allow AF_UNIX, PF_UNIX. Crash if anything else is seen. + static_assert(AF_UNIX == PF_UNIX, + "af_unix and pf_unix should not be different"); + const Arg<int> domain(0); + return If(domain == AF_UNIX, Allow()).Else(CrashSIGSYS()); + } +#endif + + if (SyscallSets::IsKill(sysno)) { + return RestrictKillTarget(current_pid, sysno); + } + + if (SyscallSets::IsFileSystem(sysno) || + SyscallSets::IsCurrentDirectory(sysno)) { + return Error(fs_denied_errno); + } + + if (SyscallSets::IsSeccomp(sysno)) + return Error(EPERM); + + if (SyscallSets::IsAnySystemV(sysno)) { + return Error(EPERM); + } + + if (SyscallSets::IsUmask(sysno) || + SyscallSets::IsDeniedFileSystemAccessViaFd(sysno) || + SyscallSets::IsDeniedGetOrModifySocket(sysno) || + SyscallSets::IsProcessPrivilegeChange(sysno)) { + return Error(EPERM); + } + +#if defined(__i386__) || defined(__mips__) + if (SyscallSets::IsSocketCall(sysno)) + return RestrictSocketcallCommand(); +#endif + + if (IsBaselinePolicyWatched(sysno)) { + // Previously unseen syscalls. TODO(jln): some of these should + // be denied gracefully right away. + return CrashSIGSYS(); + } + + // In any other case crash the program with our SIGSYS handler. + return CrashSIGSYS(); +} + +} // namespace. + +// Unfortunately C++03 doesn't allow delegated constructors. +// Call other constructor when C++11 lands. +BaselinePolicy::BaselinePolicy() : BaselinePolicy(EPERM) {} + +BaselinePolicy::BaselinePolicy(int fs_denied_errno) + : fs_denied_errno_(fs_denied_errno), policy_pid_(sys_getpid()) { +} + +BaselinePolicy::~BaselinePolicy() { + // Make sure that this policy is created, used and destroyed by a single + // process. + DCHECK_EQ(sys_getpid(), policy_pid_); +} + +ResultExpr BaselinePolicy::EvaluateSyscall(int sysno) const { + // Sanity check that we're only called with valid syscall numbers. + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + // Make sure that this policy is used in the creating process. + if (1 == sysno) { + DCHECK_EQ(sys_getpid(), policy_pid_); + } + return EvaluateSyscallImpl(fs_denied_errno_, policy_pid_, sysno); +} + +ResultExpr BaselinePolicy::InvalidSyscall() const { + return CrashSIGSYS(); +} + +} // namespace sandbox. diff --git a/sandbox/linux/seccomp-bpf-helpers/baseline_policy.h b/sandbox/linux/seccomp-bpf-helpers/baseline_policy.h new file mode 100644 index 0000000000..4169d9c3e2 --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/baseline_policy.h @@ -0,0 +1,48 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_HELPERS_BASELINE_POLICY_H_ +#define SANDBOX_LINUX_SECCOMP_BPF_HELPERS_BASELINE_POLICY_H_ + +#include "sandbox/linux/bpf_dsl/bpf_dsl_forward.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// This is a helper to build seccomp-bpf policies, i.e. policies for a sandbox +// that reduces the Linux kernel's attack surface. Given its nature, it doesn't +// have a clear semantics and is mostly "implementation-defined". +// +// This class implements the Policy interface with a "baseline" +// policy for use within Chromium. +// The "baseline" policy is somewhat arbitrary. All Chromium policies are an +// alteration of it, and it represents a reasonable common ground to run most +// code in a sandboxed environment. +// A baseline policy is only valid for the process for which this object was +// instantiated (so do not fork() and use it in a child). +class SANDBOX_EXPORT BaselinePolicy : public bpf_dsl::Policy { + public: + BaselinePolicy(); + // |fs_denied_errno| is the errno returned when a filesystem access system + // call is denied. + explicit BaselinePolicy(int fs_denied_errno); + ~BaselinePolicy() override; + + bpf_dsl::ResultExpr EvaluateSyscall(int system_call_number) const override; + bpf_dsl::ResultExpr InvalidSyscall() const override; + pid_t policy_pid() const { return policy_pid_; } + + private: + int fs_denied_errno_; + + // The PID that the policy applies to (should be equal to the current pid). + pid_t policy_pid_; + + DISALLOW_COPY_AND_ASSIGN(BaselinePolicy); +}; + +} // namespace sandbox. + +#endif // SANDBOX_LINUX_SECCOMP_BPF_HELPERS_BASELINE_POLICY_H_ diff --git a/sandbox/linux/seccomp-bpf-helpers/baseline_policy_unittest.cc b/sandbox/linux/seccomp-bpf-helpers/baseline_policy_unittest.cc new file mode 100644 index 0000000000..614849f61c --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/baseline_policy_unittest.cc @@ -0,0 +1,334 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf-helpers/baseline_policy.h" + +#include <errno.h> +#include <fcntl.h> +#include <sched.h> +#include <signal.h> +#include <string.h> +#include <sys/prctl.h> +#include <sys/resource.h> +#include <sys/socket.h> +#include <sys/stat.h> +#include <sys/syscall.h> +#include <sys/time.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <time.h> +#include <unistd.h> + +#include "base/files/scoped_file.h" +#include "base/macros.h" +#include "base/posix/eintr_wrapper.h" +#include "base/threading/thread.h" +#include "build/build_config.h" +#include "sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.h" +#include "sandbox/linux/seccomp-bpf/bpf_tests.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/seccomp-bpf/syscall.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/services/thread_helpers.h" +#include "sandbox/linux/system_headers/linux_futex.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" +#include "sandbox/linux/tests/test_utils.h" +#include "sandbox/linux/tests/unit_tests.h" + +namespace sandbox { + +namespace { + +// This also tests that read(), write() and fstat() are allowed. +void TestPipeOrSocketPair(base::ScopedFD read_end, base::ScopedFD write_end) { + BPF_ASSERT_LE(0, read_end.get()); + BPF_ASSERT_LE(0, write_end.get()); + struct stat stat_buf; + int sys_ret = fstat(read_end.get(), &stat_buf); + BPF_ASSERT_EQ(0, sys_ret); + BPF_ASSERT(S_ISFIFO(stat_buf.st_mode) || S_ISSOCK(stat_buf.st_mode)); + + const ssize_t kTestTransferSize = 4; + static const char kTestString[kTestTransferSize] = {'T', 'E', 'S', 'T'}; + ssize_t transfered = 0; + + transfered = + HANDLE_EINTR(write(write_end.get(), kTestString, kTestTransferSize)); + BPF_ASSERT_EQ(kTestTransferSize, transfered); + char read_buf[kTestTransferSize + 1] = {0}; + transfered = HANDLE_EINTR(read(read_end.get(), read_buf, sizeof(read_buf))); + BPF_ASSERT_EQ(kTestTransferSize, transfered); + BPF_ASSERT_EQ(0, memcmp(kTestString, read_buf, kTestTransferSize)); +} + +// Test that a few easy-to-test system calls are allowed. +BPF_TEST_C(BaselinePolicy, BaselinePolicyBasicAllowed, BaselinePolicy) { + BPF_ASSERT_EQ(0, sched_yield()); + + int pipefd[2]; + int sys_ret = pipe(pipefd); + BPF_ASSERT_EQ(0, sys_ret); + TestPipeOrSocketPair(base::ScopedFD(pipefd[0]), base::ScopedFD(pipefd[1])); + + BPF_ASSERT_LE(1, getpid()); + BPF_ASSERT_LE(0, getuid()); +} + +BPF_TEST_C(BaselinePolicy, FchmodErrno, BaselinePolicy) { + int ret = fchmod(-1, 07777); + BPF_ASSERT_EQ(-1, ret); + // Without the sandbox, this would EBADF instead. + BPF_ASSERT_EQ(EPERM, errno); +} + +BPF_TEST_C(BaselinePolicy, ForkErrno, BaselinePolicy) { + errno = 0; + pid_t pid = fork(); + const int fork_errno = errno; + TestUtils::HandlePostForkReturn(pid); + + BPF_ASSERT_EQ(-1, pid); + BPF_ASSERT_EQ(EPERM, fork_errno); +} + +pid_t ForkX86Glibc() { + static pid_t ptid; + return sys_clone(CLONE_PARENT_SETTID | SIGCHLD, nullptr, &ptid, nullptr, + nullptr); +} + +BPF_TEST_C(BaselinePolicy, ForkX86Eperm, BaselinePolicy) { + errno = 0; + pid_t pid = ForkX86Glibc(); + const int fork_errno = errno; + TestUtils::HandlePostForkReturn(pid); + + BPF_ASSERT_EQ(-1, pid); + BPF_ASSERT_EQ(EPERM, fork_errno); +} + +pid_t ForkARMGlibc() { + static pid_t ctid; + return sys_clone(CLONE_CHILD_SETTID | CLONE_CHILD_CLEARTID | SIGCHLD, nullptr, + nullptr, &ctid, nullptr); +} + +BPF_TEST_C(BaselinePolicy, ForkArmEperm, BaselinePolicy) { + errno = 0; + pid_t pid = ForkARMGlibc(); + const int fork_errno = errno; + TestUtils::HandlePostForkReturn(pid); + + BPF_ASSERT_EQ(-1, pid); + BPF_ASSERT_EQ(EPERM, fork_errno); +} + +BPF_TEST_C(BaselinePolicy, CreateThread, BaselinePolicy) { + base::Thread thread("sandbox_tests"); + BPF_ASSERT(thread.Start()); +} + +BPF_DEATH_TEST_C(BaselinePolicy, + DisallowedCloneFlagCrashes, + DEATH_SEGV_MESSAGE(GetCloneErrorMessageContentForTests()), + BaselinePolicy) { + pid_t pid = sys_clone(CLONE_THREAD | SIGCHLD); + TestUtils::HandlePostForkReturn(pid); +} + +BPF_DEATH_TEST_C(BaselinePolicy, + DisallowedKillCrashes, + DEATH_SEGV_MESSAGE(GetKillErrorMessageContentForTests()), + BaselinePolicy) { + BPF_ASSERT_NE(1, getpid()); + kill(1, 0); + _exit(0); +} + +BPF_TEST_C(BaselinePolicy, CanKillSelf, BaselinePolicy) { + int sys_ret = kill(getpid(), 0); + BPF_ASSERT_EQ(0, sys_ret); +} + +BPF_TEST_C(BaselinePolicy, Socketpair, BaselinePolicy) { + int sv[2]; + int sys_ret = socketpair(AF_UNIX, SOCK_DGRAM, 0, sv); + BPF_ASSERT_EQ(0, sys_ret); + TestPipeOrSocketPair(base::ScopedFD(sv[0]), base::ScopedFD(sv[1])); + + sys_ret = socketpair(AF_UNIX, SOCK_SEQPACKET, 0, sv); + BPF_ASSERT_EQ(0, sys_ret); + TestPipeOrSocketPair(base::ScopedFD(sv[0]), base::ScopedFD(sv[1])); +} + +// Not all architectures can restrict the domain for socketpair(). +#if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) +BPF_DEATH_TEST_C(BaselinePolicy, + SocketpairWrongDomain, + DEATH_SEGV_MESSAGE(GetErrorMessageContentForTests()), + BaselinePolicy) { + int sv[2]; + ignore_result(socketpair(AF_INET, SOCK_STREAM, 0, sv)); + _exit(1); +} +#endif // defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) + +BPF_TEST_C(BaselinePolicy, EPERM_open, BaselinePolicy) { + errno = 0; + int sys_ret = open("/proc/cpuinfo", O_RDONLY); + BPF_ASSERT_EQ(-1, sys_ret); + BPF_ASSERT_EQ(EPERM, errno); +} + +BPF_TEST_C(BaselinePolicy, EPERM_access, BaselinePolicy) { + errno = 0; + int sys_ret = access("/proc/cpuinfo", R_OK); + BPF_ASSERT_EQ(-1, sys_ret); + BPF_ASSERT_EQ(EPERM, errno); +} + +BPF_TEST_C(BaselinePolicy, EPERM_getcwd, BaselinePolicy) { + errno = 0; + char buf[1024]; + char* cwd = getcwd(buf, sizeof(buf)); + BPF_ASSERT_EQ(NULL, cwd); + BPF_ASSERT_EQ(EPERM, errno); +} + +BPF_DEATH_TEST_C(BaselinePolicy, + SIGSYS_InvalidSyscall, + DEATH_SEGV_MESSAGE(GetErrorMessageContentForTests()), + BaselinePolicy) { + Syscall::InvalidCall(); +} + +// A failing test using this macro could be problematic since we perform +// system calls by passing "0" as every argument. +// The kernel could SIGSEGV the process or the system call itself could reboot +// the machine. Some thoughts have been given when hand-picking the system +// calls below to limit any potential side effects outside of the current +// process. +#define TEST_BASELINE_SIGSYS(sysno) \ + BPF_DEATH_TEST_C(BaselinePolicy, \ + SIGSYS_##sysno, \ + DEATH_SEGV_MESSAGE(GetErrorMessageContentForTests()), \ + BaselinePolicy) { \ + syscall(sysno, 0, 0, 0, 0, 0, 0); \ + _exit(1); \ + } + +TEST_BASELINE_SIGSYS(__NR_acct); +TEST_BASELINE_SIGSYS(__NR_chroot); +TEST_BASELINE_SIGSYS(__NR_fanotify_init); +TEST_BASELINE_SIGSYS(__NR_fgetxattr); +TEST_BASELINE_SIGSYS(__NR_getcpu); +TEST_BASELINE_SIGSYS(__NR_getitimer); +TEST_BASELINE_SIGSYS(__NR_init_module); +TEST_BASELINE_SIGSYS(__NR_io_cancel); +TEST_BASELINE_SIGSYS(__NR_keyctl); +TEST_BASELINE_SIGSYS(__NR_mq_open); +TEST_BASELINE_SIGSYS(__NR_ptrace); +TEST_BASELINE_SIGSYS(__NR_sched_setaffinity); +TEST_BASELINE_SIGSYS(__NR_setpgid); +TEST_BASELINE_SIGSYS(__NR_swapon); +TEST_BASELINE_SIGSYS(__NR_sysinfo); +TEST_BASELINE_SIGSYS(__NR_syslog); +TEST_BASELINE_SIGSYS(__NR_timer_create); + +#if !defined(__aarch64__) +TEST_BASELINE_SIGSYS(__NR_eventfd); +TEST_BASELINE_SIGSYS(__NR_inotify_init); +TEST_BASELINE_SIGSYS(__NR_vserver); +#endif + +BPF_DEATH_TEST_C(BaselinePolicy, + FutexWithRequeuePriorityInheritence, + DEATH_SEGV_MESSAGE(GetFutexErrorMessageContentForTests()), + BaselinePolicy) { + syscall(__NR_futex, NULL, FUTEX_CMP_REQUEUE_PI, 0, NULL, NULL, 0); + _exit(1); +} + +BPF_DEATH_TEST_C(BaselinePolicy, + FutexWithRequeuePriorityInheritencePrivate, + DEATH_SEGV_MESSAGE(GetFutexErrorMessageContentForTests()), + BaselinePolicy) { + syscall(__NR_futex, NULL, FUTEX_CMP_REQUEUE_PI_PRIVATE, 0, NULL, NULL, 0); + _exit(1); +} + +BPF_DEATH_TEST_C(BaselinePolicy, + FutexWithUnlockPIPrivate, + DEATH_SEGV_MESSAGE(GetFutexErrorMessageContentForTests()), + BaselinePolicy) { + syscall(__NR_futex, NULL, FUTEX_UNLOCK_PI_PRIVATE, 0, NULL, NULL, 0); + _exit(1); +} + +BPF_TEST_C(BaselinePolicy, PrctlDumpable, BaselinePolicy) { + const int is_dumpable = prctl(PR_GET_DUMPABLE, 0, 0, 0, 0); + BPF_ASSERT(is_dumpable == 1 || is_dumpable == 0); + const int prctl_ret = prctl(PR_SET_DUMPABLE, is_dumpable, 0, 0, 0, 0); + BPF_ASSERT_EQ(0, prctl_ret); +} + +// Workaround incomplete Android headers. +#if !defined(PR_CAPBSET_READ) +#define PR_CAPBSET_READ 23 +#endif + +BPF_DEATH_TEST_C(BaselinePolicy, + PrctlSigsys, + DEATH_SEGV_MESSAGE(GetPrctlErrorMessageContentForTests()), + BaselinePolicy) { + prctl(PR_CAPBSET_READ, 0, 0, 0, 0); + _exit(1); +} + +BPF_TEST_C(BaselinePolicy, GetOrSetPriority, BaselinePolicy) { + errno = 0; + const int original_prio = getpriority(PRIO_PROCESS, 0); + // Check errno instead of the return value since this system call can return + // -1 as a valid value. + BPF_ASSERT_EQ(0, errno); + + errno = 0; + int rc = getpriority(PRIO_PROCESS, getpid()); + BPF_ASSERT_EQ(0, errno); + + rc = getpriority(PRIO_PROCESS, getpid() + 1); + BPF_ASSERT_EQ(-1, rc); + BPF_ASSERT_EQ(EPERM, errno); + + rc = setpriority(PRIO_PROCESS, 0, original_prio); + BPF_ASSERT_EQ(0, rc); + + rc = setpriority(PRIO_PROCESS, getpid(), original_prio); + BPF_ASSERT_EQ(0, rc); + + errno = 0; + rc = setpriority(PRIO_PROCESS, getpid() + 1, original_prio); + BPF_ASSERT_EQ(-1, rc); + BPF_ASSERT_EQ(EPERM, errno); +} + +BPF_DEATH_TEST_C(BaselinePolicy, + GetPrioritySigsys, + DEATH_SEGV_MESSAGE(GetErrorMessageContentForTests()), + BaselinePolicy) { + getpriority(PRIO_USER, 0); + _exit(1); +} + +BPF_DEATH_TEST_C(BaselinePolicy, + ClockGettimeWithDisallowedClockCrashes, + DEATH_SEGV_MESSAGE(sandbox::GetErrorMessageContentForTests()), + BaselinePolicy) { + struct timespec ts; + clock_gettime(CLOCK_MONOTONIC_RAW, &ts); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.cc b/sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.cc new file mode 100644 index 0000000000..05250d147f --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.cc @@ -0,0 +1,297 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// Note: any code in this file MUST be async-signal safe. + +#include "sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.h" + +#include <sys/syscall.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/seccomp-bpf/syscall.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" + +#if defined(__mips__) +// __NR_Linux, is defined in <asm/unistd.h>. +#include <asm/unistd.h> +#endif + +#define SECCOMP_MESSAGE_COMMON_CONTENT "seccomp-bpf failure" +#define SECCOMP_MESSAGE_CLONE_CONTENT "clone() failure" +#define SECCOMP_MESSAGE_PRCTL_CONTENT "prctl() failure" +#define SECCOMP_MESSAGE_IOCTL_CONTENT "ioctl() failure" +#define SECCOMP_MESSAGE_KILL_CONTENT "(tg)kill() failure" +#define SECCOMP_MESSAGE_FUTEX_CONTENT "futex() failure" + +namespace { + +inline bool IsArchitectureX86_64() { +#if defined(__x86_64__) + return true; +#else + return false; +#endif +} + +// Write |error_message| to stderr. Similar to RawLog(), but a bit more careful +// about async-signal safety. |size| is the size to write and should typically +// not include a terminating \0. +void WriteToStdErr(const char* error_message, size_t size) { + while (size > 0) { + // TODO(jln): query the current policy to check if send() is available and + // use it to perform a non-blocking write. + const int ret = HANDLE_EINTR(write(STDERR_FILENO, error_message, size)); + // We can't handle any type of error here. + if (ret <= 0 || static_cast<size_t>(ret) > size) break; + size -= ret; + error_message += ret; + } +} + +// Invalid syscall values are truncated to zero. +// On architectures where base value is zero (Intel and Arm), +// syscall number is the same as offset from base. +// This function returns values between 0 and 1023 on all architectures. +// On architectures where base value is different than zero (currently only +// Mips), we are truncating valid syscall values to offset from base. +uint32_t SyscallNumberToOffsetFromBase(uint32_t sysno) { +#if defined(__mips__) + // On MIPS syscall numbers are in different range than on x86 and ARM. + // Valid MIPS O32 ABI syscall __NR_syscall will be truncated to zero for + // simplicity. + sysno = sysno - __NR_Linux; +#endif + + if (sysno >= 1024) + sysno = 0; + + return sysno; +} + +// Print a seccomp-bpf failure to handle |sysno| to stderr in an +// async-signal safe way. +void PrintSyscallError(uint32_t sysno) { + if (sysno >= 1024) + sysno = 0; + // TODO(markus): replace with async-signal safe snprintf when available. + const size_t kNumDigits = 4; + char sysno_base10[kNumDigits]; + uint32_t rem = sysno; + uint32_t mod = 0; + for (int i = kNumDigits - 1; i >= 0; i--) { + mod = rem % 10; + rem /= 10; + sysno_base10[i] = '0' + mod; + } +#if defined(__mips__) && (_MIPS_SIM == _MIPS_SIM_ABI32) + static const char kSeccompErrorPrefix[] = __FILE__ + ":**CRASHING**:" SECCOMP_MESSAGE_COMMON_CONTENT " in syscall 4000 + "; +#else + static const char kSeccompErrorPrefix[] = + __FILE__":**CRASHING**:" SECCOMP_MESSAGE_COMMON_CONTENT " in syscall "; +#endif + static const char kSeccompErrorPostfix[] = "\n"; + WriteToStdErr(kSeccompErrorPrefix, sizeof(kSeccompErrorPrefix) - 1); + WriteToStdErr(sysno_base10, sizeof(sysno_base10)); + WriteToStdErr(kSeccompErrorPostfix, sizeof(kSeccompErrorPostfix) - 1); +} + +} // namespace. + +namespace sandbox { + +intptr_t CrashSIGSYS_Handler(const struct arch_seccomp_data& args, void* aux) { + uint32_t syscall = SyscallNumberToOffsetFromBase(args.nr); + + PrintSyscallError(syscall); + + // Encode 8-bits of the 1st two arguments too, so we can discern which socket + // type, which fcntl, ... etc., without being likely to hit a mapped + // address. + // Do not encode more bits here without thinking about increasing the + // likelihood of collision with mapped pages. + syscall |= ((args.args[0] & 0xffUL) << 12); + syscall |= ((args.args[1] & 0xffUL) << 20); + // Purposefully dereference the syscall as an address so it'll show up very + // clearly and easily in crash dumps. + volatile char* addr = reinterpret_cast<volatile char*>(syscall); + *addr = '\0'; + // In case we hit a mapped address, hit the null page with just the syscall, + // for paranoia. + syscall &= 0xfffUL; + addr = reinterpret_cast<volatile char*>(syscall); + *addr = '\0'; + for (;;) + _exit(1); +} + +// TODO(jln): refactor the reporting functions. + +intptr_t SIGSYSCloneFailure(const struct arch_seccomp_data& args, void* aux) { + static const char kSeccompCloneError[] = + __FILE__":**CRASHING**:" SECCOMP_MESSAGE_CLONE_CONTENT "\n"; + WriteToStdErr(kSeccompCloneError, sizeof(kSeccompCloneError) - 1); + // "flags" is the first argument in the kernel's clone(). + // Mark as volatile to be able to find the value on the stack in a minidump. + volatile uint64_t clone_flags = args.args[0]; + volatile char* addr; + if (IsArchitectureX86_64()) { + addr = reinterpret_cast<volatile char*>(clone_flags & 0xFFFFFF); + *addr = '\0'; + } + // Hit the NULL page if this fails to fault. + addr = reinterpret_cast<volatile char*>(clone_flags & 0xFFF); + *addr = '\0'; + for (;;) + _exit(1); +} + +intptr_t SIGSYSPrctlFailure(const struct arch_seccomp_data& args, + void* /* aux */) { + static const char kSeccompPrctlError[] = + __FILE__":**CRASHING**:" SECCOMP_MESSAGE_PRCTL_CONTENT "\n"; + WriteToStdErr(kSeccompPrctlError, sizeof(kSeccompPrctlError) - 1); + // Mark as volatile to be able to find the value on the stack in a minidump. + volatile uint64_t option = args.args[0]; + volatile char* addr = + reinterpret_cast<volatile char*>(option & 0xFFF); + *addr = '\0'; + for (;;) + _exit(1); +} + +intptr_t SIGSYSIoctlFailure(const struct arch_seccomp_data& args, + void* /* aux */) { + static const char kSeccompIoctlError[] = + __FILE__":**CRASHING**:" SECCOMP_MESSAGE_IOCTL_CONTENT "\n"; + WriteToStdErr(kSeccompIoctlError, sizeof(kSeccompIoctlError) - 1); + // Make "request" volatile so that we can see it on the stack in a minidump. + volatile uint64_t request = args.args[1]; + volatile char* addr = reinterpret_cast<volatile char*>(request & 0xFFFF); + *addr = '\0'; + // Hit the NULL page if this fails. + addr = reinterpret_cast<volatile char*>(request & 0xFFF); + *addr = '\0'; + for (;;) + _exit(1); +} + +intptr_t SIGSYSKillFailure(const struct arch_seccomp_data& args, + void* /* aux */) { + static const char kSeccompKillError[] = + __FILE__":**CRASHING**:" SECCOMP_MESSAGE_KILL_CONTENT "\n"; + WriteToStdErr(kSeccompKillError, sizeof(kSeccompKillError) - 1); + // Make "pid" volatile so that we can see it on the stack in a minidump. + volatile uint64_t my_pid = sys_getpid(); + volatile char* addr = reinterpret_cast<volatile char*>(my_pid & 0xFFF); + *addr = '\0'; + for (;;) + _exit(1); +} + +intptr_t SIGSYSFutexFailure(const struct arch_seccomp_data& args, + void* /* aux */) { + static const char kSeccompFutexError[] = + __FILE__ ":**CRASHING**:" SECCOMP_MESSAGE_FUTEX_CONTENT "\n"; + WriteToStdErr(kSeccompFutexError, sizeof(kSeccompFutexError) - 1); + volatile int futex_op = args.args[1]; + volatile char* addr = reinterpret_cast<volatile char*>(futex_op & 0xFFF); + *addr = '\0'; + for (;;) + _exit(1); +} + +intptr_t SIGSYSSchedHandler(const struct arch_seccomp_data& args, + void* aux) { + switch (args.nr) { + case __NR_sched_getaffinity: + case __NR_sched_getattr: + case __NR_sched_getparam: + case __NR_sched_getscheduler: + case __NR_sched_rr_get_interval: + case __NR_sched_setaffinity: + case __NR_sched_setattr: + case __NR_sched_setparam: + case __NR_sched_setscheduler: + const pid_t tid = sys_gettid(); + // The first argument is the pid. If is our thread id, then replace it + // with 0, which is equivalent and allowed by the policy. + if (args.args[0] == static_cast<uint64_t>(tid)) { + return Syscall::Call(args.nr, + 0, + static_cast<intptr_t>(args.args[1]), + static_cast<intptr_t>(args.args[2]), + static_cast<intptr_t>(args.args[3]), + static_cast<intptr_t>(args.args[4]), + static_cast<intptr_t>(args.args[5])); + } + break; + } + + CrashSIGSYS_Handler(args, aux); + + // Should never be reached. + RAW_CHECK(false); + return -ENOSYS; +} + +bpf_dsl::ResultExpr CrashSIGSYS() { + return bpf_dsl::Trap(CrashSIGSYS_Handler, NULL); +} + +bpf_dsl::ResultExpr CrashSIGSYSClone() { + return bpf_dsl::Trap(SIGSYSCloneFailure, NULL); +} + +bpf_dsl::ResultExpr CrashSIGSYSPrctl() { + return bpf_dsl::Trap(SIGSYSPrctlFailure, NULL); +} + +bpf_dsl::ResultExpr CrashSIGSYSIoctl() { + return bpf_dsl::Trap(SIGSYSIoctlFailure, NULL); +} + +bpf_dsl::ResultExpr CrashSIGSYSKill() { + return bpf_dsl::Trap(SIGSYSKillFailure, NULL); +} + +bpf_dsl::ResultExpr CrashSIGSYSFutex() { + return bpf_dsl::Trap(SIGSYSFutexFailure, NULL); +} + +bpf_dsl::ResultExpr RewriteSchedSIGSYS() { + return bpf_dsl::Trap(SIGSYSSchedHandler, NULL); +} + +const char* GetErrorMessageContentForTests() { + return SECCOMP_MESSAGE_COMMON_CONTENT; +} + +const char* GetCloneErrorMessageContentForTests() { + return SECCOMP_MESSAGE_CLONE_CONTENT; +} + +const char* GetPrctlErrorMessageContentForTests() { + return SECCOMP_MESSAGE_PRCTL_CONTENT; +} + +const char* GetIoctlErrorMessageContentForTests() { + return SECCOMP_MESSAGE_IOCTL_CONTENT; +} + +const char* GetKillErrorMessageContentForTests() { + return SECCOMP_MESSAGE_KILL_CONTENT; +} + +const char* GetFutexErrorMessageContentForTests() { + return SECCOMP_MESSAGE_FUTEX_CONTENT; +} + +} // namespace sandbox. diff --git a/sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.h b/sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.h new file mode 100644 index 0000000000..c64e994172 --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.h @@ -0,0 +1,82 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SIGSYS_HANDLERS_H_ +#define SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SIGSYS_HANDLERS_H_ + +#include <stdint.h> + +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_forward.h" +#include "sandbox/sandbox_export.h" + +// The handlers are suitable for use in Trap() error codes. They are +// guaranteed to be async-signal safe. +// See sandbox/linux/seccomp-bpf/trap.h to see how they work. + +namespace sandbox { + +struct arch_seccomp_data; + +// This handler will crash the currently running process. The crashing address +// will be the number of the current system call, extracted from |args|. +// This handler will also print to stderr the number of the crashing syscall. +SANDBOX_EXPORT intptr_t + CrashSIGSYS_Handler(const struct arch_seccomp_data& args, void* aux); + +// The following three handlers are suitable to report failures with the +// clone(), prctl() and ioctl() system calls respectively. + +// The crashing address will be (clone_flags & 0xFFFFFF), where clone_flags is +// the clone(2) argument, extracted from |args|. +SANDBOX_EXPORT intptr_t + SIGSYSCloneFailure(const struct arch_seccomp_data& args, void* aux); +// The crashing address will be (option & 0xFFF), where option is the prctl(2) +// argument. +SANDBOX_EXPORT intptr_t + SIGSYSPrctlFailure(const struct arch_seccomp_data& args, void* aux); +// The crashing address will be request & 0xFFFF, where request is the ioctl(2) +// argument. +SANDBOX_EXPORT intptr_t + SIGSYSIoctlFailure(const struct arch_seccomp_data& args, void* aux); +// The crashing address will be (pid & 0xFFF), where pid is the first +// argument (and can be a tid). +SANDBOX_EXPORT intptr_t + SIGSYSKillFailure(const struct arch_seccomp_data& args, void* aux); +// The crashing address will be (op & 0xFFF), where op is the second +// argument. +SANDBOX_EXPORT intptr_t + SIGSYSFutexFailure(const struct arch_seccomp_data& args, void* aux); +// If the syscall is not being called on the current tid, crashes in the same +// way as CrashSIGSYS_Handler. Otherwise, returns the result of calling the +// syscall with the pid argument set to 0 (which for these calls means the +// current thread). The following syscalls are supported: +// +// sched_getaffinity(), sched_getattr(), sched_getparam(), sched_getscheduler(), +// sched_rr_get_interval(), sched_setaffinity(), sched_setattr(), +// sched_setparam(), sched_setscheduler() +SANDBOX_EXPORT intptr_t + SIGSYSSchedHandler(const struct arch_seccomp_data& args, void* aux); + +// Variants of the above functions for use with bpf_dsl. +SANDBOX_EXPORT bpf_dsl::ResultExpr CrashSIGSYS(); +SANDBOX_EXPORT bpf_dsl::ResultExpr CrashSIGSYSClone(); +SANDBOX_EXPORT bpf_dsl::ResultExpr CrashSIGSYSPrctl(); +SANDBOX_EXPORT bpf_dsl::ResultExpr CrashSIGSYSIoctl(); +SANDBOX_EXPORT bpf_dsl::ResultExpr CrashSIGSYSKill(); +SANDBOX_EXPORT bpf_dsl::ResultExpr CrashSIGSYSFutex(); +SANDBOX_EXPORT bpf_dsl::ResultExpr RewriteSchedSIGSYS(); + +// Following four functions return substrings of error messages used +// in the above four functions. They are useful in death tests. +SANDBOX_EXPORT const char* GetErrorMessageContentForTests(); +SANDBOX_EXPORT const char* GetCloneErrorMessageContentForTests(); +SANDBOX_EXPORT const char* GetPrctlErrorMessageContentForTests(); +SANDBOX_EXPORT const char* GetIoctlErrorMessageContentForTests(); +SANDBOX_EXPORT const char* GetKillErrorMessageContentForTests(); +SANDBOX_EXPORT const char* GetFutexErrorMessageContentForTests(); + +} // namespace sandbox. + +#endif // SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SIGSYS_HANDLERS_H_ diff --git a/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.cc b/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.cc new file mode 100644 index 0000000000..58ffb843a8 --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.cc @@ -0,0 +1,319 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.h" + +#include <errno.h> +#include <fcntl.h> +#include <fcntl.h> +#include <linux/net.h> +#include <sched.h> +#include <signal.h> +#include <stdint.h> +#include <sys/mman.h> +#include <sys/prctl.h> +#include <sys/resource.h> +#include <sys/stat.h> +#include <sys/time.h> +#include <sys/types.h> +#include <time.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/macros.h" +#include "base/time/time.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" +#include "sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/system_headers/linux_futex.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" +#include "sandbox/linux/system_headers/linux_time.h" + +// PNaCl toolchain does not provide sys/ioctl.h header. +#if !defined(OS_NACL_NONSFI) +#include <sys/ioctl.h> +#endif + +#if defined(OS_ANDROID) + +#if !defined(F_DUPFD_CLOEXEC) +#define F_DUPFD_CLOEXEC (F_LINUX_SPECIFIC_BASE + 6) +#endif + +// https://android.googlesource.com/platform/bionic/+/lollipop-release/libc/private/bionic_prctl.h +#if !defined(PR_SET_VMA) +#define PR_SET_VMA 0x53564d41 +#endif + +// https://android.googlesource.com/platform/system/core/+/lollipop-release/libcutils/sched_policy.c +#if !defined(PR_SET_TIMERSLACK_PID) +#define PR_SET_TIMERSLACK_PID 41 +#endif + +#endif // defined(OS_ANDROID) + +#if defined(__arm__) && !defined(MAP_STACK) +#define MAP_STACK 0x20000 // Daisy build environment has old headers. +#endif + +#if defined(__mips__) && !defined(MAP_STACK) +#define MAP_STACK 0x40000 +#endif +namespace { + +inline bool IsArchitectureX86_64() { +#if defined(__x86_64__) + return true; +#else + return false; +#endif +} + +inline bool IsArchitectureI386() { +#if defined(__i386__) + return true; +#else + return false; +#endif +} + +inline bool IsAndroid() { +#if defined(OS_ANDROID) + return true; +#else + return false; +#endif +} + +inline bool IsArchitectureMips() { +#if defined(__mips__) + return true; +#else + return false; +#endif +} + +} // namespace. + +#define CASES SANDBOX_BPF_DSL_CASES + +using sandbox::bpf_dsl::Allow; +using sandbox::bpf_dsl::Arg; +using sandbox::bpf_dsl::BoolExpr; +using sandbox::bpf_dsl::Error; +using sandbox::bpf_dsl::If; +using sandbox::bpf_dsl::ResultExpr; + +namespace sandbox { + +#if !defined(OS_NACL_NONSFI) +// Allow Glibc's and Android pthread creation flags, crash on any other +// thread creation attempts and EPERM attempts to use neither +// CLONE_VM, nor CLONE_THREAD, which includes all fork() implementations. +ResultExpr RestrictCloneToThreadsAndEPERMFork() { + const Arg<unsigned long> flags(0); + + // TODO(mdempsky): Extend DSL to support (flags & ~mask1) == mask2. + const uint64_t kAndroidCloneMask = CLONE_VM | CLONE_FS | CLONE_FILES | + CLONE_SIGHAND | CLONE_THREAD | + CLONE_SYSVSEM; + const uint64_t kObsoleteAndroidCloneMask = kAndroidCloneMask | CLONE_DETACHED; + + const uint64_t kGlibcPthreadFlags = + CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND | CLONE_THREAD | + CLONE_SYSVSEM | CLONE_SETTLS | CLONE_PARENT_SETTID | CLONE_CHILD_CLEARTID; + const BoolExpr glibc_test = flags == kGlibcPthreadFlags; + + const BoolExpr android_test = flags == kAndroidCloneMask || + flags == kObsoleteAndroidCloneMask || + flags == kGlibcPthreadFlags; + + return If(IsAndroid() ? android_test : glibc_test, Allow()) + .ElseIf((flags & (CLONE_VM | CLONE_THREAD)) == 0, Error(EPERM)) + .Else(CrashSIGSYSClone()); +} + +ResultExpr RestrictPrctl() { + // Will need to add seccomp compositing in the future. PR_SET_PTRACER is + // used by breakpad but not needed anymore. + const Arg<int> option(0); + return Switch(option) + .CASES((PR_GET_NAME, PR_SET_NAME, PR_GET_DUMPABLE, PR_SET_DUMPABLE), + Allow()) +#if defined(OS_ANDROID) + .CASES((PR_SET_VMA, PR_SET_TIMERSLACK_PID), Allow()) +#endif + .Default(CrashSIGSYSPrctl()); +} + +ResultExpr RestrictIoctl() { + const Arg<int> request(1); + return Switch(request).CASES((TCGETS, FIONREAD), Allow()).Default( + CrashSIGSYSIoctl()); +} + +ResultExpr RestrictMmapFlags() { + // The flags you see are actually the allowed ones, and the variable is a + // "denied" mask because of the negation operator. + // Significantly, we don't permit MAP_HUGETLB, or the newer flags such as + // MAP_POPULATE. + // TODO(davidung), remove MAP_DENYWRITE with updated Tegra libraries. + const uint64_t kAllowedMask = MAP_SHARED | MAP_PRIVATE | MAP_ANONYMOUS | + MAP_STACK | MAP_NORESERVE | MAP_FIXED | + MAP_DENYWRITE; + const Arg<int> flags(3); + return If((flags & ~kAllowedMask) == 0, Allow()).Else(CrashSIGSYS()); +} + +ResultExpr RestrictMprotectFlags() { + // The flags you see are actually the allowed ones, and the variable is a + // "denied" mask because of the negation operator. + // Significantly, we don't permit weird undocumented flags such as + // PROT_GROWSDOWN. + const uint64_t kAllowedMask = PROT_READ | PROT_WRITE | PROT_EXEC; + const Arg<int> prot(2); + return If((prot & ~kAllowedMask) == 0, Allow()).Else(CrashSIGSYS()); +} + +ResultExpr RestrictFcntlCommands() { + // We also restrict the flags in F_SETFL. We don't want to permit flags with + // a history of trouble such as O_DIRECT. The flags you see are actually the + // allowed ones, and the variable is a "denied" mask because of the negation + // operator. + // Glibc overrides the kernel's O_LARGEFILE value. Account for this. + uint64_t kOLargeFileFlag = O_LARGEFILE; + if (IsArchitectureX86_64() || IsArchitectureI386() || IsArchitectureMips()) + kOLargeFileFlag = 0100000; + + const Arg<int> cmd(1); + const Arg<long> long_arg(2); + + const uint64_t kAllowedMask = O_ACCMODE | O_APPEND | O_NONBLOCK | O_SYNC | + kOLargeFileFlag | O_CLOEXEC | O_NOATIME; + return Switch(cmd) + .CASES((F_GETFL, + F_GETFD, + F_SETFD, + F_SETLK, + F_SETLKW, + F_GETLK, + F_DUPFD, + F_DUPFD_CLOEXEC), + Allow()) + .Case(F_SETFL, + If((long_arg & ~kAllowedMask) == 0, Allow()).Else(CrashSIGSYS())) + .Default(CrashSIGSYS()); +} + +#if defined(__i386__) || defined(__mips__) +ResultExpr RestrictSocketcallCommand() { + // Unfortunately, we are unable to restrict the first parameter to + // socketpair(2). Whilst initially sounding bad, it's noteworthy that very + // few protocols actually support socketpair(2). The scary call that we're + // worried about, socket(2), remains blocked. + const Arg<int> call(0); + return Switch(call) + .CASES((SYS_SOCKETPAIR, + SYS_SHUTDOWN, + SYS_RECV, + SYS_SEND, + SYS_RECVFROM, + SYS_SENDTO, + SYS_RECVMSG, + SYS_SENDMSG), + Allow()) + .Default(Error(EPERM)); +} +#endif + +ResultExpr RestrictKillTarget(pid_t target_pid, int sysno) { + switch (sysno) { + case __NR_kill: + case __NR_tgkill: { + const Arg<pid_t> pid(0); + return If(pid == target_pid, Allow()).Else(CrashSIGSYSKill()); + } + case __NR_tkill: + return CrashSIGSYSKill(); + default: + NOTREACHED(); + return CrashSIGSYS(); + } +} + +ResultExpr RestrictFutex() { + const uint64_t kAllowedFutexFlags = FUTEX_PRIVATE_FLAG | FUTEX_CLOCK_REALTIME; + const Arg<int> op(1); + return Switch(op & ~kAllowedFutexFlags) + .CASES((FUTEX_WAIT, + FUTEX_WAKE, + FUTEX_REQUEUE, + FUTEX_CMP_REQUEUE, + FUTEX_WAKE_OP, + FUTEX_WAIT_BITSET, + FUTEX_WAKE_BITSET), + Allow()) + .Default(CrashSIGSYSFutex()); +} + +ResultExpr RestrictGetSetpriority(pid_t target_pid) { + const Arg<int> which(0); + const Arg<int> who(1); + return If(which == PRIO_PROCESS, + If(who == 0 || who == target_pid, Allow()).Else(Error(EPERM))) + .Else(CrashSIGSYS()); +} + +ResultExpr RestrictSchedTarget(pid_t target_pid, int sysno) { + switch (sysno) { + case __NR_sched_getaffinity: + case __NR_sched_getattr: + case __NR_sched_getparam: + case __NR_sched_getscheduler: + case __NR_sched_rr_get_interval: + case __NR_sched_setaffinity: + case __NR_sched_setattr: + case __NR_sched_setparam: + case __NR_sched_setscheduler: { + const Arg<pid_t> pid(0); + return If(pid == 0 || pid == target_pid, Allow()) + .Else(RewriteSchedSIGSYS()); + } + default: + NOTREACHED(); + return CrashSIGSYS(); + } +} + +ResultExpr RestrictPrlimit64(pid_t target_pid) { + const Arg<pid_t> pid(0); + return If(pid == 0 || pid == target_pid, Allow()).Else(CrashSIGSYS()); +} + +ResultExpr RestrictGetrusage() { + const Arg<int> who(0); + return If(who == RUSAGE_SELF, Allow()).Else(CrashSIGSYS()); +} +#endif // !defined(OS_NACL_NONSFI) + +ResultExpr RestrictClockID() { + static_assert(4 == sizeof(clockid_t), "clockid_t is not 32bit"); + const Arg<clockid_t> clockid(0); + return If( +#if defined(OS_CHROMEOS) + // Allow the special clock for Chrome OS used by Chrome tracing. + clockid == base::TraceTicks::kClockSystemTrace || +#endif + clockid == CLOCK_MONOTONIC || + clockid == CLOCK_MONOTONIC_COARSE || + clockid == CLOCK_PROCESS_CPUTIME_ID || + clockid == CLOCK_REALTIME || + clockid == CLOCK_REALTIME_COARSE || + clockid == CLOCK_THREAD_CPUTIME_ID, + Allow()).Else(CrashSIGSYS()); +} + +} // namespace sandbox. diff --git a/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.h b/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.h new file mode 100644 index 0000000000..9eb35d10e0 --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.h @@ -0,0 +1,100 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SYSCALL_PARAMETERS_RESTRICTIONS_H_ +#define SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SYSCALL_PARAMETERS_RESTRICTIONS_H_ + +#include <unistd.h> + +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl_forward.h" +#include "sandbox/sandbox_export.h" + +// These are helpers to build seccomp-bpf policies, i.e. policies for a +// sandbox that reduces the Linux kernel's attack surface. They return a +// bpf_dsl::ResultExpr suitable to restrict certain system call parameters. + +namespace sandbox { + +// Allow clone(2) for threads. +// Reject fork(2) attempts with EPERM. +// Don't restrict on ASAN. +// Crash if anything else is attempted. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictCloneToThreadsAndEPERMFork(); + +// Allow PR_SET_NAME, PR_SET_DUMPABLE, PR_GET_DUMPABLE. +// Crash if anything else is attempted. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictPrctl(); + +// Allow TCGETS and FIONREAD. +// Crash if anything else is attempted. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictIoctl(); + +// Restrict the flags argument in mmap(2). +// Only allow: MAP_SHARED | MAP_PRIVATE | MAP_ANONYMOUS | +// MAP_STACK | MAP_NORESERVE | MAP_FIXED | MAP_DENYWRITE. +// Crash if any other flag is used. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictMmapFlags(); + +// Restrict the prot argument in mprotect(2). +// Only allow: PROT_READ | PROT_WRITE | PROT_EXEC. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictMprotectFlags(); + +// Restrict fcntl(2) cmd argument to: +// We allow F_GETFL, F_SETFL, F_GETFD, F_SETFD, F_DUPFD, F_DUPFD_CLOEXEC, +// F_SETLK, F_SETLKW and F_GETLK. +// Also, in F_SETFL, restrict the allowed flags to: O_ACCMODE | O_APPEND | +// O_NONBLOCK | O_SYNC | O_LARGEFILE | O_CLOEXEC | O_NOATIME. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictFcntlCommands(); + +#if defined(__i386__) || defined(__mips__) +// Restrict socketcall(2) to only allow socketpair(2), send(2), recv(2), +// sendto(2), recvfrom(2), shutdown(2), sendmsg(2) and recvmsg(2). +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictSocketcallCommand(); +#endif + +// Restrict |sysno| (which must be kill, tkill or tgkill) by allowing tgkill or +// kill iff the first parameter is |target_pid|, crashing otherwise or if +// |sysno| is tkill. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictKillTarget(pid_t target_pid, + int sysno); + +// Crash if FUTEX_CMP_REQUEUE_PI is used in the second argument of futex(2). +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictFutex(); + +// Crash if |which| is not PRIO_PROCESS. EPERM if |who| is not 0, neither +// |target_pid| while calling setpriority(2) / getpriority(2). +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictGetSetpriority(pid_t target_pid); + +// Restricts |pid| for sched_* syscalls which take a pid as the first argument. +// We only allow calling these syscalls if the pid argument is equal to the pid +// of the sandboxed process or 0 (indicating the current thread). The following +// syscalls are supported: +// +// sched_getaffinity(), sched_getattr(), sched_getparam(), sched_getscheduler(), +// sched_rr_get_interval(), sched_setaffinity(), sched_setattr(), +// sched_setparam(), sched_setscheduler() +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictSchedTarget(pid_t target_pid, + int sysno); + +// Restricts the |pid| argument of prlimit64 to 0 (meaning the calling process) +// or target_pid. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictPrlimit64(pid_t target_pid); + +// Restricts the |who| argument of getrusage to RUSAGE_SELF (meaning the calling +// process). +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictGetrusage(); + +// Restrict |clk_id| for clock_getres(), clock_gettime() and clock_settime(). +// We allow accessing only CLOCK_MONOTONIC, CLOCK_PROCESS_CPUTIME_ID, +// CLOCK_REALTIME, and CLOCK_THREAD_CPUTIME_ID. In particular, this disallows +// access to arbitrary per-{process,thread} CPU-time clock IDs (such as those +// returned by {clock,pthread}_getcpuclockid), which can leak information +// about the state of the host OS. +// On Chrome OS, base::TraceTicks::kClockSystemTrace is also allowed. +SANDBOX_EXPORT bpf_dsl::ResultExpr RestrictClockID(); + +} // namespace sandbox. + +#endif // SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SYSCALL_PARAMETERS_RESTRICTIONS_H_ diff --git a/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions_unittests.cc b/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions_unittests.cc new file mode 100644 index 0000000000..aaed480d69 --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions_unittests.cc @@ -0,0 +1,282 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf-helpers/syscall_parameters_restrictions.h" + +#include <errno.h> +#include <sched.h> +#include <sys/resource.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <time.h> +#include <unistd.h> + +#include "base/bind.h" +#include "base/synchronization/waitable_event.h" +#include "base/sys_info.h" +#include "base/threading/thread.h" +#include "base/time/time.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.h" +#include "sandbox/linux/seccomp-bpf/bpf_tests.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/seccomp-bpf/syscall.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" +#include "sandbox/linux/system_headers/linux_time.h" +#include "sandbox/linux/tests/unit_tests.h" + +#if !defined(OS_ANDROID) +#include "third_party/lss/linux_syscall_support.h" // for MAKE_PROCESS_CPUCLOCK +#endif + +namespace sandbox { + +namespace { + +// NOTE: most of the parameter restrictions are tested in +// baseline_policy_unittest.cc as a more end-to-end test. + +using sandbox::bpf_dsl::Allow; +using sandbox::bpf_dsl::ResultExpr; + +class RestrictClockIdPolicy : public bpf_dsl::Policy { + public: + RestrictClockIdPolicy() {} + ~RestrictClockIdPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + switch (sysno) { + case __NR_clock_gettime: + case __NR_clock_getres: + return RestrictClockID(); + default: + return Allow(); + } + } +}; + +void CheckClock(clockid_t clockid) { + struct timespec ts; + ts.tv_sec = -1; + ts.tv_nsec = -1; + BPF_ASSERT_EQ(0, clock_getres(clockid, &ts)); + BPF_ASSERT_EQ(0, ts.tv_sec); + BPF_ASSERT_LE(0, ts.tv_nsec); + ts.tv_sec = -1; + ts.tv_nsec = -1; + BPF_ASSERT_EQ(0, clock_gettime(clockid, &ts)); + BPF_ASSERT_LE(0, ts.tv_sec); + BPF_ASSERT_LE(0, ts.tv_nsec); +} + +BPF_TEST_C(ParameterRestrictions, + clock_gettime_allowed, + RestrictClockIdPolicy) { + CheckClock(CLOCK_MONOTONIC); + CheckClock(CLOCK_MONOTONIC_COARSE); + CheckClock(CLOCK_PROCESS_CPUTIME_ID); + CheckClock(CLOCK_REALTIME); + CheckClock(CLOCK_REALTIME_COARSE); + CheckClock(CLOCK_THREAD_CPUTIME_ID); +} + +BPF_DEATH_TEST_C(ParameterRestrictions, + clock_gettime_crash_monotonic_raw, + DEATH_SEGV_MESSAGE(sandbox::GetErrorMessageContentForTests()), + RestrictClockIdPolicy) { + struct timespec ts; + clock_gettime(CLOCK_MONOTONIC_RAW, &ts); +} + +#if defined(OS_CHROMEOS) + +// A custom BPF tester delegate to run IsRunningOnChromeOS() before +// the sandbox is enabled because we cannot run it with non-SFI BPF +// sandbox enabled. +class ClockSystemTesterDelegate : public sandbox::BPFTesterDelegate { + public: + ClockSystemTesterDelegate() + : is_running_on_chromeos_(base::SysInfo::IsRunningOnChromeOS()) {} + ~ClockSystemTesterDelegate() override {} + + scoped_ptr<sandbox::bpf_dsl::Policy> GetSandboxBPFPolicy() override { + return scoped_ptr<sandbox::bpf_dsl::Policy>(new RestrictClockIdPolicy()); + } + void RunTestFunction() override { + if (is_running_on_chromeos_) { + CheckClock(base::TraceTicks::kClockSystemTrace); + } else { + struct timespec ts; + // kClockSystemTrace is 11, which is CLOCK_THREAD_CPUTIME_ID of + // the init process (pid=1). If kernel supports this feature, + // this may succeed even if this is not running on Chrome OS. We + // just check this clock_gettime call does not crash. + clock_gettime(base::TraceTicks::kClockSystemTrace, &ts); + } + } + + private: + const bool is_running_on_chromeos_; + DISALLOW_COPY_AND_ASSIGN(ClockSystemTesterDelegate); +}; + +BPF_TEST_D(BPFTest, BPFTestWithDelegateClass, ClockSystemTesterDelegate); + +#elif defined(OS_LINUX) + +BPF_DEATH_TEST_C(ParameterRestrictions, + clock_gettime_crash_system_trace, + DEATH_SEGV_MESSAGE(sandbox::GetErrorMessageContentForTests()), + RestrictClockIdPolicy) { + struct timespec ts; + clock_gettime(base::TraceTicks::kClockSystemTrace, &ts); +} + +#endif // defined(OS_CHROMEOS) + +#if !defined(OS_ANDROID) +BPF_DEATH_TEST_C(ParameterRestrictions, + clock_gettime_crash_cpu_clock, + DEATH_SEGV_MESSAGE(sandbox::GetErrorMessageContentForTests()), + RestrictClockIdPolicy) { + // We can't use clock_getcpuclockid() because it's not implemented in newlib, + // and it might not work inside the sandbox anyway. + const pid_t kInitPID = 1; + const clockid_t kInitCPUClockID = + MAKE_PROCESS_CPUCLOCK(kInitPID, CPUCLOCK_SCHED); + + struct timespec ts; + clock_gettime(kInitCPUClockID, &ts); +} +#endif // !defined(OS_ANDROID) + +class RestrictSchedPolicy : public bpf_dsl::Policy { + public: + RestrictSchedPolicy() {} + ~RestrictSchedPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + switch (sysno) { + case __NR_sched_getparam: + return RestrictSchedTarget(getpid(), sysno); + default: + return Allow(); + } + } +}; + +void CheckSchedGetParam(pid_t pid, struct sched_param* param) { + BPF_ASSERT_EQ(0, sched_getparam(pid, param)); +} + +void SchedGetParamThread(base::WaitableEvent* thread_run) { + const pid_t pid = getpid(); + const pid_t tid = sys_gettid(); + BPF_ASSERT_NE(pid, tid); + + struct sched_param current_pid_param; + CheckSchedGetParam(pid, ¤t_pid_param); + + struct sched_param zero_param; + CheckSchedGetParam(0, &zero_param); + + struct sched_param tid_param; + CheckSchedGetParam(tid, &tid_param); + + BPF_ASSERT_EQ(zero_param.sched_priority, tid_param.sched_priority); + + // Verify that the SIGSYS handler sets errno properly. + errno = 0; + BPF_ASSERT_EQ(-1, sched_getparam(tid, NULL)); + BPF_ASSERT_EQ(EINVAL, errno); + + thread_run->Signal(); +} + +BPF_TEST_C(ParameterRestrictions, + sched_getparam_allowed, + RestrictSchedPolicy) { + base::WaitableEvent thread_run(true, false); + // Run the actual test in a new thread so that the current pid and tid are + // different. + base::Thread getparam_thread("sched_getparam_thread"); + BPF_ASSERT(getparam_thread.Start()); + getparam_thread.message_loop()->PostTask( + FROM_HERE, base::Bind(&SchedGetParamThread, &thread_run)); + BPF_ASSERT(thread_run.TimedWait(base::TimeDelta::FromMilliseconds(5000))); + getparam_thread.Stop(); +} + +BPF_DEATH_TEST_C(ParameterRestrictions, + sched_getparam_crash_non_zero, + DEATH_SEGV_MESSAGE(sandbox::GetErrorMessageContentForTests()), + RestrictSchedPolicy) { + const pid_t kInitPID = 1; + struct sched_param param; + sched_getparam(kInitPID, ¶m); +} + +class RestrictPrlimit64Policy : public bpf_dsl::Policy { + public: + RestrictPrlimit64Policy() {} + ~RestrictPrlimit64Policy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + switch (sysno) { + case __NR_prlimit64: + return RestrictPrlimit64(getpid()); + default: + return Allow(); + } + } +}; + +BPF_TEST_C(ParameterRestrictions, prlimit64_allowed, RestrictPrlimit64Policy) { + BPF_ASSERT_EQ(0, sys_prlimit64(0, RLIMIT_AS, NULL, NULL)); + BPF_ASSERT_EQ(0, sys_prlimit64(getpid(), RLIMIT_AS, NULL, NULL)); +} + +BPF_DEATH_TEST_C(ParameterRestrictions, + prlimit64_crash_not_self, + DEATH_SEGV_MESSAGE(sandbox::GetErrorMessageContentForTests()), + RestrictPrlimit64Policy) { + const pid_t kInitPID = 1; + BPF_ASSERT_NE(kInitPID, getpid()); + sys_prlimit64(kInitPID, RLIMIT_AS, NULL, NULL); +} + +class RestrictGetrusagePolicy : public bpf_dsl::Policy { + public: + RestrictGetrusagePolicy() {} + ~RestrictGetrusagePolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + switch (sysno) { + case __NR_getrusage: + return RestrictGetrusage(); + default: + return Allow(); + } + } +}; + +BPF_TEST_C(ParameterRestrictions, getrusage_allowed, RestrictGetrusagePolicy) { + struct rusage usage; + BPF_ASSERT_EQ(0, getrusage(RUSAGE_SELF, &usage)); +} + +BPF_DEATH_TEST_C(ParameterRestrictions, + getrusage_crash_not_self, + DEATH_SEGV_MESSAGE(sandbox::GetErrorMessageContentForTests()), + RestrictGetrusagePolicy) { + struct rusage usage; + getrusage(RUSAGE_CHILDREN, &usage); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf-helpers/syscall_sets.cc b/sandbox/linux/seccomp-bpf-helpers/syscall_sets.cc new file mode 100644 index 0000000000..c217d47e2d --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/syscall_sets.cc @@ -0,0 +1,1060 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf-helpers/syscall_sets.h" + +#include "build/build_config.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" + +namespace sandbox { + +// The functions below cover all existing i386, x86_64, and ARM system calls; +// excluding syscalls made obsolete in ARM EABI. +// The implicitly defined sets form a partition of the sets of +// system calls. + +bool SyscallSets::IsKill(int sysno) { + switch (sysno) { + case __NR_kill: + case __NR_tgkill: + case __NR_tkill: // Deprecated. + return true; + default: + return false; + } +} + +bool SyscallSets::IsAllowedGettime(int sysno) { + switch (sysno) { + case __NR_gettimeofday: +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_time: +#endif + return true; + case __NR_adjtimex: // Privileged. + case __NR_clock_adjtime: // Privileged. + case __NR_clock_getres: // Could be allowed. + case __NR_clock_gettime: + case __NR_clock_nanosleep: // Could be allowed. + case __NR_clock_settime: // Privileged. +#if defined(__i386__) || defined(__mips__) + case __NR_ftime: // Obsolete. +#endif + case __NR_settimeofday: // Privileged. +#if defined(__i386__) || defined(__mips__) + case __NR_stime: +#endif + default: + return false; + } +} + +bool SyscallSets::IsCurrentDirectory(int sysno) { + switch (sysno) { + case __NR_getcwd: + case __NR_chdir: + case __NR_fchdir: + return true; + default: + return false; + } +} + +bool SyscallSets::IsUmask(int sysno) { + switch (sysno) { + case __NR_umask: + return true; + default: + return false; + } +} + +// System calls that directly access the file system. They might acquire +// a new file descriptor or otherwise perform an operation directly +// via a path. +// Both EPERM and ENOENT are valid errno unless otherwise noted in comment. +bool SyscallSets::IsFileSystem(int sysno) { + switch (sysno) { +#if !defined(__aarch64__) + case __NR_access: // EPERM not a valid errno. + case __NR_chmod: + case __NR_chown: +#if defined(__i386__) || defined(__arm__) + case __NR_chown32: +#endif + case __NR_creat: + case __NR_futimesat: // Should be called utimesat ? + case __NR_lchown: + case __NR_link: + case __NR_lstat: // EPERM not a valid errno. + case __NR_mkdir: + case __NR_mknod: + case __NR_open: + case __NR_readlink: // EPERM not a valid errno. + case __NR_rename: + case __NR_rmdir: + case __NR_stat: // EPERM not a valid errno. + case __NR_symlink: + case __NR_unlink: + case __NR_uselib: // Neither EPERM, nor ENOENT are valid errno. + case __NR_ustat: // Same as above. Deprecated. + case __NR_utimes: +#endif // !defined(__aarch64__) + + case __NR_execve: + case __NR_faccessat: // EPERM not a valid errno. + case __NR_fchmodat: + case __NR_fchownat: // Should be called chownat ? +#if defined(__x86_64__) || defined(__aarch64__) + case __NR_newfstatat: // fstatat(). EPERM not a valid errno. +#elif defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_fstatat64: +#endif +#if defined(__i386__) || defined(__arm__) + case __NR_lchown32: +#endif + case __NR_linkat: + case __NR_lookup_dcookie: // ENOENT not a valid errno. + +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_lstat64: +#endif +#if defined(__i386__) || defined(__arm__) || defined(__x86_64__) + case __NR_memfd_create: +#endif + case __NR_mkdirat: + case __NR_mknodat: +#if defined(__i386__) + case __NR_oldlstat: + case __NR_oldstat: +#endif + case __NR_openat: + case __NR_readlinkat: + case __NR_renameat: + case __NR_renameat2: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_stat64: +#endif + case __NR_statfs: // EPERM not a valid errno. +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_statfs64: +#endif + case __NR_symlinkat: + case __NR_truncate: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_truncate64: +#endif + case __NR_unlinkat: +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_utime: +#endif + case __NR_utimensat: // New. + return true; + default: + return false; + } +} + +bool SyscallSets::IsAllowedFileSystemAccessViaFd(int sysno) { + switch (sysno) { + case __NR_fstat: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_fstat64: +#endif + return true; +// TODO(jln): these should be denied gracefully as well (moved below). +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_fadvise64: // EPERM not a valid errno. +#endif +#if defined(__i386__) + case __NR_fadvise64_64: +#endif +#if defined(__arm__) + case __NR_arm_fadvise64_64: +#endif + case __NR_fdatasync: // EPERM not a valid errno. + case __NR_flock: // EPERM not a valid errno. + case __NR_fstatfs: // Give information about the whole filesystem. +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_fstatfs64: +#endif + case __NR_fsync: // EPERM not a valid errno. +#if defined(__i386__) + case __NR_oldfstat: +#endif +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_sync_file_range: // EPERM not a valid errno. +#elif defined(__arm__) + case __NR_arm_sync_file_range: // EPERM not a valid errno. +#endif + default: + return false; + } +} + +// EPERM is a good errno for any of these. +bool SyscallSets::IsDeniedFileSystemAccessViaFd(int sysno) { + switch (sysno) { + case __NR_fallocate: + case __NR_fchmod: + case __NR_fchown: + case __NR_ftruncate: +#if defined(__i386__) || defined(__arm__) + case __NR_fchown32: +#endif +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_ftruncate64: +#endif +#if !defined(__aarch64__) + case __NR_getdents: // EPERM not a valid errno. +#endif + case __NR_getdents64: // EPERM not a valid errno. +#if defined(__i386__) || defined(__mips__) + case __NR_readdir: +#endif + return true; + default: + return false; + } +} + +bool SyscallSets::IsGetSimpleId(int sysno) { + switch (sysno) { + case __NR_capget: + case __NR_getegid: + case __NR_geteuid: + case __NR_getgid: + case __NR_getgroups: + case __NR_getpid: + case __NR_getppid: + case __NR_getresgid: + case __NR_getsid: + case __NR_gettid: + case __NR_getuid: + case __NR_getresuid: +#if defined(__i386__) || defined(__arm__) + case __NR_getegid32: + case __NR_geteuid32: + case __NR_getgid32: + case __NR_getgroups32: + case __NR_getresgid32: + case __NR_getresuid32: + case __NR_getuid32: +#endif + return true; + default: + return false; + } +} + +bool SyscallSets::IsProcessPrivilegeChange(int sysno) { + switch (sysno) { + case __NR_capset: +#if defined(__i386__) || defined(__x86_64__) + case __NR_ioperm: // Intel privilege. + case __NR_iopl: // Intel privilege. +#endif + case __NR_setfsgid: + case __NR_setfsuid: + case __NR_setgid: + case __NR_setgroups: + case __NR_setregid: + case __NR_setresgid: + case __NR_setresuid: + case __NR_setreuid: + case __NR_setuid: +#if defined(__i386__) || defined(__arm__) + case __NR_setfsgid32: + case __NR_setfsuid32: + case __NR_setgid32: + case __NR_setgroups32: + case __NR_setregid32: + case __NR_setresgid32: + case __NR_setresuid32: + case __NR_setreuid32: + case __NR_setuid32: +#endif + return true; + default: + return false; + } +} + +bool SyscallSets::IsProcessGroupOrSession(int sysno) { + switch (sysno) { + case __NR_setpgid: +#if !defined(__aarch64__) + case __NR_getpgrp: +#endif + case __NR_setsid: + case __NR_getpgid: + return true; + default: + return false; + } +} + +bool SyscallSets::IsAllowedSignalHandling(int sysno) { + switch (sysno) { + case __NR_rt_sigaction: + case __NR_rt_sigprocmask: + case __NR_rt_sigreturn: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_sigaction: + case __NR_sigprocmask: + case __NR_sigreturn: +#endif + return true; + case __NR_rt_sigpending: + case __NR_rt_sigqueueinfo: + case __NR_rt_sigsuspend: + case __NR_rt_sigtimedwait: + case __NR_rt_tgsigqueueinfo: + case __NR_sigaltstack: +#if !defined(__aarch64__) + case __NR_signalfd: +#endif + case __NR_signalfd4: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_sigpending: + case __NR_sigsuspend: +#endif +#if defined(__i386__) || defined(__mips__) + case __NR_signal: + case __NR_sgetmask: // Obsolete. + case __NR_ssetmask: +#endif + default: + return false; + } +} + +bool SyscallSets::IsAllowedOperationOnFd(int sysno) { + switch (sysno) { + case __NR_close: + case __NR_dup: +#if !defined(__aarch64__) + case __NR_dup2: +#endif + case __NR_dup3: +#if defined(__x86_64__) || defined(__arm__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_shutdown: +#endif + return true; + case __NR_fcntl: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_fcntl64: +#endif + default: + return false; + } +} + +bool SyscallSets::IsKernelInternalApi(int sysno) { + switch (sysno) { + case __NR_restart_syscall: +#if defined(__arm__) + case __ARM_NR_cmpxchg: +#endif + return true; + default: + return false; + } +} + +// This should be thought through in conjunction with IsFutex(). +bool SyscallSets::IsAllowedProcessStartOrDeath(int sysno) { + switch (sysno) { + case __NR_exit: + case __NR_exit_group: + case __NR_wait4: + case __NR_waitid: +#if defined(__i386__) + case __NR_waitpid: +#endif + return true; + case __NR_clone: // Should be parameter-restricted. + case __NR_setns: // Privileged. +#if !defined(__aarch64__) + case __NR_fork: +#endif +#if defined(__i386__) || defined(__x86_64__) + case __NR_get_thread_area: +#endif +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_set_thread_area: +#endif + case __NR_set_tid_address: + case __NR_unshare: +#if !defined(__mips__) && !defined(__aarch64__) + case __NR_vfork: +#endif + default: + return false; + } +} + +// It's difficult to restrict those, but there is attack surface here. +bool SyscallSets::IsAllowedFutex(int sysno) { + switch (sysno) { + case __NR_get_robust_list: + case __NR_set_robust_list: + case __NR_futex: + default: + return false; + } +} + +bool SyscallSets::IsAllowedEpoll(int sysno) { + switch (sysno) { +#if !defined(__aarch64__) + case __NR_epoll_create: + case __NR_epoll_wait: +#endif + case __NR_epoll_create1: + case __NR_epoll_ctl: + return true; + default: +#if defined(__x86_64__) + case __NR_epoll_ctl_old: +#endif + case __NR_epoll_pwait: +#if defined(__x86_64__) + case __NR_epoll_wait_old: +#endif + return false; + } +} + +bool SyscallSets::IsAllowedGetOrModifySocket(int sysno) { + switch (sysno) { +#if !defined(__aarch64__) + case __NR_pipe: +#endif + case __NR_pipe2: + return true; + default: +#if defined(__x86_64__) || defined(__arm__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_socketpair: // We will want to inspect its argument. +#endif + return false; + } +} + +bool SyscallSets::IsDeniedGetOrModifySocket(int sysno) { + switch (sysno) { +#if defined(__x86_64__) || defined(__arm__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_accept: + case __NR_accept4: + case __NR_bind: + case __NR_connect: + case __NR_socket: + case __NR_listen: + return true; +#endif + default: + return false; + } +} + +#if defined(__i386__) || defined(__mips__) +// Big multiplexing system call for sockets. +bool SyscallSets::IsSocketCall(int sysno) { + switch (sysno) { + case __NR_socketcall: + return true; + default: + return false; + } +} +#endif + +#if defined(__x86_64__) || defined(__arm__) || defined(__mips__) +bool SyscallSets::IsNetworkSocketInformation(int sysno) { + switch (sysno) { + case __NR_getpeername: + case __NR_getsockname: + case __NR_getsockopt: + case __NR_setsockopt: + return true; + default: + return false; + } +} +#endif + +bool SyscallSets::IsAllowedAddressSpaceAccess(int sysno) { + switch (sysno) { + case __NR_brk: + case __NR_mlock: + case __NR_munlock: + case __NR_munmap: + return true; + case __NR_madvise: + case __NR_mincore: + case __NR_mlockall: +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_mmap: +#endif +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_mmap2: +#endif +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_modify_ldt: +#endif + case __NR_mprotect: + case __NR_mremap: + case __NR_msync: + case __NR_munlockall: + case __NR_readahead: + case __NR_remap_file_pages: +#if defined(__i386__) + case __NR_vm86: + case __NR_vm86old: +#endif + default: + return false; + } +} + +bool SyscallSets::IsAllowedGeneralIo(int sysno) { + switch (sysno) { + case __NR_lseek: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR__llseek: +#endif +#if !defined(__aarch64__) + case __NR_poll: +#endif + case __NR_ppoll: + case __NR_pselect6: + case __NR_read: + case __NR_readv: +#if defined(__arm__) || defined(__mips__) + case __NR_recv: +#endif +#if defined(__x86_64__) || defined(__arm__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_recvfrom: // Could specify source. + case __NR_recvmsg: // Could specify source. +#endif +#if defined(__i386__) || defined(__x86_64__) + case __NR_select: +#endif +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR__newselect: +#endif +#if defined(__arm__) + case __NR_send: +#endif +#if defined(__x86_64__) || defined(__arm__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_sendmsg: // Could specify destination. + case __NR_sendto: // Could specify destination. +#endif + case __NR_write: + case __NR_writev: + return true; + case __NR_ioctl: // Can be very powerful. + case __NR_pread64: + case __NR_preadv: + case __NR_pwrite64: + case __NR_pwritev: + case __NR_recvmmsg: // Could specify source. + case __NR_sendfile: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_sendfile64: +#endif + case __NR_sendmmsg: // Could specify destination. + case __NR_splice: + case __NR_tee: + case __NR_vmsplice: + default: + return false; + } +} + +bool SyscallSets::IsPrctl(int sysno) { + switch (sysno) { +#if defined(__x86_64__) + case __NR_arch_prctl: +#endif + case __NR_prctl: + return true; + default: + return false; + } +} + +bool SyscallSets::IsSeccomp(int sysno) { + switch (sysno) { + case __NR_seccomp: + return true; + default: + return false; + } +} + +bool SyscallSets::IsAllowedBasicScheduler(int sysno) { + switch (sysno) { + case __NR_sched_yield: +#if !defined(__aarch64__) + case __NR_pause: +#endif + case __NR_nanosleep: + return true; + case __NR_getpriority: +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_nice: +#endif + case __NR_setpriority: + default: + return false; + } +} + +bool SyscallSets::IsAdminOperation(int sysno) { + switch (sysno) { +#if defined(__i386__) || defined(__arm__) || defined(__mips__) + case __NR_bdflush: +#endif + case __NR_kexec_load: + case __NR_reboot: + case __NR_setdomainname: + case __NR_sethostname: + case __NR_syslog: + return true; + default: + return false; + } +} + +bool SyscallSets::IsKernelModule(int sysno) { + switch (sysno) { +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_create_module: + case __NR_get_kernel_syms: // Should ENOSYS. + case __NR_query_module: +#endif + case __NR_delete_module: + case __NR_init_module: + case __NR_finit_module: + return true; + default: + return false; + } +} + +bool SyscallSets::IsGlobalFSViewChange(int sysno) { + switch (sysno) { + case __NR_pivot_root: + case __NR_chroot: + case __NR_sync: + return true; + default: + return false; + } +} + +bool SyscallSets::IsFsControl(int sysno) { + switch (sysno) { + case __NR_mount: + case __NR_nfsservctl: + case __NR_quotactl: + case __NR_swapoff: + case __NR_swapon: +#if defined(__i386__) || defined(__mips__) + case __NR_umount: +#endif + case __NR_umount2: + return true; + default: + return false; + } +} + +bool SyscallSets::IsNuma(int sysno) { + switch (sysno) { + case __NR_get_mempolicy: + case __NR_getcpu: + case __NR_mbind: +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_migrate_pages: +#endif + case __NR_move_pages: + case __NR_set_mempolicy: + return true; + default: + return false; + } +} + +bool SyscallSets::IsMessageQueue(int sysno) { + switch (sysno) { + case __NR_mq_getsetattr: + case __NR_mq_notify: + case __NR_mq_open: + case __NR_mq_timedreceive: + case __NR_mq_timedsend: + case __NR_mq_unlink: + return true; + default: + return false; + } +} + +bool SyscallSets::IsGlobalProcessEnvironment(int sysno) { + switch (sysno) { + case __NR_acct: // Privileged. +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) || \ + defined(__aarch64__) + case __NR_getrlimit: +#endif +#if defined(__i386__) || defined(__arm__) + case __NR_ugetrlimit: +#endif +#if defined(__i386__) || defined(__mips__) + case __NR_ulimit: +#endif + case __NR_getrusage: + case __NR_personality: // Can change its personality as well. + case __NR_prlimit64: // Like setrlimit / getrlimit. + case __NR_setrlimit: + case __NR_times: + return true; + default: + return false; + } +} + +bool SyscallSets::IsDebug(int sysno) { + switch (sysno) { + case __NR_ptrace: + case __NR_process_vm_readv: + case __NR_process_vm_writev: + case __NR_kcmp: + return true; + default: + return false; + } +} + +bool SyscallSets::IsGlobalSystemStatus(int sysno) { + switch (sysno) { +#if !defined(__aarch64__) + case __NR__sysctl: + case __NR_sysfs: +#endif + case __NR_sysinfo: + case __NR_uname: +#if defined(__i386__) + case __NR_olduname: + case __NR_oldolduname: +#endif + return true; + default: + return false; + } +} + +bool SyscallSets::IsEventFd(int sysno) { + switch (sysno) { +#if !defined(__aarch64__) + case __NR_eventfd: +#endif + case __NR_eventfd2: + return true; + default: + return false; + } +} + +// Asynchronous I/O API. +bool SyscallSets::IsAsyncIo(int sysno) { + switch (sysno) { + case __NR_io_cancel: + case __NR_io_destroy: + case __NR_io_getevents: + case __NR_io_setup: + case __NR_io_submit: + return true; + default: + return false; + } +} + +bool SyscallSets::IsKeyManagement(int sysno) { + switch (sysno) { + case __NR_add_key: + case __NR_keyctl: + case __NR_request_key: + return true; + default: + return false; + } +} + +#if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) +bool SyscallSets::IsSystemVSemaphores(int sysno) { + switch (sysno) { + case __NR_semctl: + case __NR_semget: + case __NR_semop: + case __NR_semtimedop: + return true; + default: + return false; + } +} +#endif + +#if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) +// These give a lot of ambient authority and bypass the setuid sandbox. +bool SyscallSets::IsSystemVSharedMemory(int sysno) { + switch (sysno) { + case __NR_shmat: + case __NR_shmctl: + case __NR_shmdt: + case __NR_shmget: + return true; + default: + return false; + } +} +#endif + +#if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) +bool SyscallSets::IsSystemVMessageQueue(int sysno) { + switch (sysno) { + case __NR_msgctl: + case __NR_msgget: + case __NR_msgrcv: + case __NR_msgsnd: + return true; + default: + return false; + } +} +#endif + +#if defined(__i386__) || defined(__mips__) +// Big system V multiplexing system call. +bool SyscallSets::IsSystemVIpc(int sysno) { + switch (sysno) { + case __NR_ipc: + return true; + default: + return false; + } +} +#endif + +bool SyscallSets::IsAnySystemV(int sysno) { +#if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) + return IsSystemVMessageQueue(sysno) || IsSystemVSemaphores(sysno) || + IsSystemVSharedMemory(sysno); +#elif defined(__i386__) || defined(__mips__) + return IsSystemVIpc(sysno); +#endif +} + +bool SyscallSets::IsAdvancedScheduler(int sysno) { + switch (sysno) { + case __NR_ioprio_get: // IO scheduler. + case __NR_ioprio_set: + case __NR_sched_get_priority_max: + case __NR_sched_get_priority_min: + case __NR_sched_getaffinity: + case __NR_sched_getattr: + case __NR_sched_getparam: + case __NR_sched_getscheduler: + case __NR_sched_rr_get_interval: + case __NR_sched_setaffinity: + case __NR_sched_setattr: + case __NR_sched_setparam: + case __NR_sched_setscheduler: + return true; + default: + return false; + } +} + +bool SyscallSets::IsInotify(int sysno) { + switch (sysno) { + case __NR_inotify_add_watch: +#if !defined(__aarch64__) + case __NR_inotify_init: +#endif + case __NR_inotify_init1: + case __NR_inotify_rm_watch: + return true; + default: + return false; + } +} + +bool SyscallSets::IsFaNotify(int sysno) { + switch (sysno) { + case __NR_fanotify_init: + case __NR_fanotify_mark: + return true; + default: + return false; + } +} + +bool SyscallSets::IsTimer(int sysno) { + switch (sysno) { + case __NR_getitimer: +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_alarm: +#endif + case __NR_setitimer: + return true; + default: + return false; + } +} + +bool SyscallSets::IsAdvancedTimer(int sysno) { + switch (sysno) { + case __NR_timer_create: + case __NR_timer_delete: + case __NR_timer_getoverrun: + case __NR_timer_gettime: + case __NR_timer_settime: + case __NR_timerfd_create: + case __NR_timerfd_gettime: + case __NR_timerfd_settime: + return true; + default: + return false; + } +} + +bool SyscallSets::IsExtendedAttributes(int sysno) { + switch (sysno) { + case __NR_fgetxattr: + case __NR_flistxattr: + case __NR_fremovexattr: + case __NR_fsetxattr: + case __NR_getxattr: + case __NR_lgetxattr: + case __NR_listxattr: + case __NR_llistxattr: + case __NR_lremovexattr: + case __NR_lsetxattr: + case __NR_removexattr: + case __NR_setxattr: + return true; + default: + return false; + } +} + +// Various system calls that need to be researched. +// TODO(jln): classify this better. +bool SyscallSets::IsMisc(int sysno) { + switch (sysno) { +#if !defined(__mips__) + case __NR_getrandom: +#endif + case __NR_name_to_handle_at: + case __NR_open_by_handle_at: + case __NR_perf_event_open: + case __NR_syncfs: + case __NR_vhangup: +// The system calls below are not implemented. +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_afs_syscall: +#endif +#if defined(__i386__) || defined(__mips__) + case __NR_break: +#endif +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_getpmsg: +#endif +#if defined(__i386__) || defined(__mips__) + case __NR_gtty: + case __NR_idle: + case __NR_lock: + case __NR_mpx: + case __NR_prof: + case __NR_profil: +#endif +#if defined(__i386__) || defined(__x86_64__) || defined(__mips__) + case __NR_putpmsg: +#endif +#if defined(__x86_64__) + case __NR_security: +#endif +#if defined(__i386__) || defined(__mips__) + case __NR_stty: +#endif +#if defined(__x86_64__) + case __NR_tuxcall: +#endif +#if !defined(__aarch64__) + case __NR_vserver: +#endif + return true; + default: + return false; + } +} + +#if defined(__arm__) +bool SyscallSets::IsArmPciConfig(int sysno) { + switch (sysno) { + case __NR_pciconfig_iobase: + case __NR_pciconfig_read: + case __NR_pciconfig_write: + return true; + default: + return false; + } +} + +bool SyscallSets::IsArmPrivate(int sysno) { + switch (sysno) { + case __ARM_NR_breakpoint: + case __ARM_NR_cacheflush: + case __ARM_NR_set_tls: + case __ARM_NR_usr26: + case __ARM_NR_usr32: + return true; + default: + return false; + } +} +#endif // defined(__arm__) + +#if defined(__mips__) +bool SyscallSets::IsMipsPrivate(int sysno) { + switch (sysno) { + case __NR_cacheflush: + case __NR_cachectl: + return true; + default: + return false; + } +} + +bool SyscallSets::IsMipsMisc(int sysno) { + switch (sysno) { + case __NR_sysmips: + case __NR_unused150: + return true; + default: + return false; + } +} +#endif // defined(__mips__) +} // namespace sandbox. diff --git a/sandbox/linux/seccomp-bpf-helpers/syscall_sets.h b/sandbox/linux/seccomp-bpf-helpers/syscall_sets.h new file mode 100644 index 0000000000..5ba6335a95 --- /dev/null +++ b/sandbox/linux/seccomp-bpf-helpers/syscall_sets.h @@ -0,0 +1,112 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SYSCALL_SETS_H_ +#define SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SYSCALL_SETS_H_ + +#include "base/macros.h" +#include "build/build_config.h" +#include "sandbox/sandbox_export.h" + +// These are helpers to build seccomp-bpf policies, i.e. policies for a +// sandbox that reduces the Linux kernel's attack surface. Given their +// nature, they don't have any clear semantics and are completely +// "implementation-defined". + +namespace sandbox { + +class SANDBOX_EXPORT SyscallSets { + public: + static bool IsKill(int sysno); + static bool IsAllowedGettime(int sysno); + static bool IsCurrentDirectory(int sysno); + static bool IsUmask(int sysno); + // System calls that directly access the file system. They might acquire + // a new file descriptor or otherwise perform an operation directly + // via a path. + static bool IsFileSystem(int sysno); + static bool IsAllowedFileSystemAccessViaFd(int sysno); + static bool IsDeniedFileSystemAccessViaFd(int sysno); + static bool IsGetSimpleId(int sysno); + static bool IsProcessPrivilegeChange(int sysno); + static bool IsProcessGroupOrSession(int sysno); + static bool IsAllowedSignalHandling(int sysno); + static bool IsAllowedOperationOnFd(int sysno); + static bool IsKernelInternalApi(int sysno); + // This should be thought through in conjunction with IsFutex(). + static bool IsAllowedProcessStartOrDeath(int sysno); + // It's difficult to restrict those, but there is attack surface here. + static bool IsAllowedFutex(int sysno); + static bool IsAllowedEpoll(int sysno); + static bool IsAllowedGetOrModifySocket(int sysno); + static bool IsDeniedGetOrModifySocket(int sysno); + +#if defined(__i386__) || defined(__mips__) + // Big multiplexing system call for sockets. + static bool IsSocketCall(int sysno); +#endif + +#if defined(__x86_64__) || defined(__arm__) || defined(__mips__) || \ + defined(__aarch64__) + static bool IsNetworkSocketInformation(int sysno); +#endif + + static bool IsAllowedAddressSpaceAccess(int sysno); + static bool IsAllowedGeneralIo(int sysno); + static bool IsPrctl(int sysno); + static bool IsSeccomp(int sysno); + static bool IsAllowedBasicScheduler(int sysno); + static bool IsAdminOperation(int sysno); + static bool IsKernelModule(int sysno); + static bool IsGlobalFSViewChange(int sysno); + static bool IsFsControl(int sysno); + static bool IsNuma(int sysno); + static bool IsMessageQueue(int sysno); + static bool IsGlobalProcessEnvironment(int sysno); + static bool IsDebug(int sysno); + static bool IsGlobalSystemStatus(int sysno); + static bool IsEventFd(int sysno); + // Asynchronous I/O API. + static bool IsAsyncIo(int sysno); + static bool IsKeyManagement(int sysno); +#if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) + static bool IsSystemVSemaphores(int sysno); +#endif +#if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) + // These give a lot of ambient authority and bypass the setuid sandbox. + static bool IsSystemVSharedMemory(int sysno); +#endif + +#if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) + static bool IsSystemVMessageQueue(int sysno); +#endif + +#if defined(__i386__) || defined(__mips__) + // Big system V multiplexing system call. + static bool IsSystemVIpc(int sysno); +#endif + + static bool IsAnySystemV(int sysno); + static bool IsAdvancedScheduler(int sysno); + static bool IsInotify(int sysno); + static bool IsFaNotify(int sysno); + static bool IsTimer(int sysno); + static bool IsAdvancedTimer(int sysno); + static bool IsExtendedAttributes(int sysno); + static bool IsMisc(int sysno); +#if defined(__arm__) + static bool IsArmPciConfig(int sysno); + static bool IsArmPrivate(int sysno); +#endif // defined(__arm__) +#if defined(__mips__) + static bool IsMipsPrivate(int sysno); + static bool IsMipsMisc(int sysno); +#endif // defined(__mips__) + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(SyscallSets); +}; + +} // namespace sandbox. + +#endif // SANDBOX_LINUX_SECCOMP_BPF_HELPERS_SYSCALL_SETS_H_ diff --git a/sandbox/linux/seccomp-bpf/DEPS b/sandbox/linux/seccomp-bpf/DEPS new file mode 100644 index 0000000000..149c463b06 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/DEPS @@ -0,0 +1,5 @@ +include_rules = [ + "+sandbox/linux/bpf_dsl", + "+sandbox/linux/services", + "+sandbox/linux/system_headers", +] diff --git a/sandbox/linux/seccomp-bpf/bpf_tester_compatibility_delegate.h b/sandbox/linux/seccomp-bpf/bpf_tester_compatibility_delegate.h new file mode 100644 index 0000000000..7736c1506f --- /dev/null +++ b/sandbox/linux/seccomp-bpf/bpf_tester_compatibility_delegate.h @@ -0,0 +1,54 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_BPF_TESTER_COMPATIBILITY_DELEGATE_H_ +#define SANDBOX_LINUX_SECCOMP_BPF_BPF_TESTER_COMPATIBILITY_DELEGATE_H_ + +#include "base/memory/scoped_ptr.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf_test_runner.h" + +namespace sandbox { + +// This templated class allows building a BPFTesterDelegate from a +// deprecated-style BPF policy (that is a SyscallEvaluator function pointer, +// instead of a SandboxBPFPolicy class), specified in |policy_function| and a +// function pointer to a test in |test_function|. +// This allows both the policy and the test function to take a pointer to an +// object of type "Aux" as a parameter. This is used to implement the BPF_TEST +// macro and should generally not be used directly. +template <class Policy, class Aux> +class BPFTesterCompatibilityDelegate : public BPFTesterDelegate { + public: + typedef void (*TestFunction)(Aux*); + + explicit BPFTesterCompatibilityDelegate(TestFunction test_function) + : aux_(), test_function_(test_function) {} + + ~BPFTesterCompatibilityDelegate() override {} + + scoped_ptr<bpf_dsl::Policy> GetSandboxBPFPolicy() override { + // The current method is guaranteed to only run in the child process + // running the test. In this process, the current object is guaranteed + // to live forever. So it's ok to pass aux_pointer_for_policy_ to + // the policy, which could in turn pass it to the kernel via Trap(). + return scoped_ptr<bpf_dsl::Policy>(new Policy(&aux_)); + } + + void RunTestFunction() override { + // Run the actual test. + // The current object is guaranteed to live forever in the child process + // where this will run. + test_function_(&aux_); + } + + private: + Aux aux_; + TestFunction test_function_; + + DISALLOW_COPY_AND_ASSIGN(BPFTesterCompatibilityDelegate); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SECCOMP_BPF_BPF_TESTER_COMPATIBILITY_DELEGATE_H_ diff --git a/sandbox/linux/seccomp-bpf/bpf_tests.h b/sandbox/linux/seccomp-bpf/bpf_tests.h new file mode 100644 index 0000000000..cc4debd4c3 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/bpf_tests.h @@ -0,0 +1,122 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_BPF_TESTS_H__ +#define SANDBOX_LINUX_SECCOMP_BPF_BPF_TESTS_H__ + +#include "base/logging.h" +#include "base/macros.h" +#include "build/build_config.h" +#include "sandbox/linux/seccomp-bpf/bpf_tester_compatibility_delegate.h" +#include "sandbox/linux/tests/unit_tests.h" + +namespace sandbox { + +// BPF_TEST_C() is a special version of SANDBOX_TEST(). It runs a test function +// in a sub-process, under a seccomp-bpf policy specified in +// |bpf_policy_class_name| without failing on configurations that are allowed +// to not support seccomp-bpf in their kernels. +// This is the preferred format for new BPF tests. |bpf_policy_class_name| is a +// class name (which will be default-constructed) that implements the +// Policy interface. +// The test function's body can simply follow. Test functions should use +// the BPF_ASSERT macros defined below, not GTEST's macros. The use of +// CHECK* macros is supported but less robust. +#define BPF_TEST_C(test_case_name, test_name, bpf_policy_class_name) \ + BPF_DEATH_TEST_C( \ + test_case_name, test_name, DEATH_SUCCESS(), bpf_policy_class_name) + +// Identical to BPF_TEST_C but allows to specify the nature of death. +#define BPF_DEATH_TEST_C( \ + test_case_name, test_name, death, bpf_policy_class_name) \ + void BPF_TEST_C_##test_name(); \ + TEST(test_case_name, DISABLE_ON_TSAN(test_name)) { \ + sandbox::SandboxBPFTestRunner bpf_test_runner( \ + new sandbox::BPFTesterSimpleDelegate<bpf_policy_class_name>( \ + BPF_TEST_C_##test_name)); \ + sandbox::UnitTests::RunTestInProcess(&bpf_test_runner, death); \ + } \ + void BPF_TEST_C_##test_name() + +// This form of BPF_TEST is a little verbose and should be reserved for complex +// tests where a lot of control is required. +// |bpf_tester_delegate_class| must be a classname implementing the +// BPFTesterDelegate interface. +#define BPF_TEST_D(test_case_name, test_name, bpf_tester_delegate_class) \ + BPF_DEATH_TEST_D( \ + test_case_name, test_name, DEATH_SUCCESS(), bpf_tester_delegate_class) + +// Identical to BPF_TEST_D but allows to specify the nature of death. +#define BPF_DEATH_TEST_D( \ + test_case_name, test_name, death, bpf_tester_delegate_class) \ + TEST(test_case_name, DISABLE_ON_TSAN(test_name)) { \ + sandbox::SandboxBPFTestRunner bpf_test_runner( \ + new bpf_tester_delegate_class()); \ + sandbox::UnitTests::RunTestInProcess(&bpf_test_runner, death); \ + } + +// Assertions are handled exactly the same as with a normal SANDBOX_TEST() +#define BPF_ASSERT SANDBOX_ASSERT +#define BPF_ASSERT_EQ(x, y) BPF_ASSERT((x) == (y)) +#define BPF_ASSERT_NE(x, y) BPF_ASSERT((x) != (y)) +#define BPF_ASSERT_LT(x, y) BPF_ASSERT((x) < (y)) +#define BPF_ASSERT_GT(x, y) BPF_ASSERT((x) > (y)) +#define BPF_ASSERT_LE(x, y) BPF_ASSERT((x) <= (y)) +#define BPF_ASSERT_GE(x, y) BPF_ASSERT((x) >= (y)) + +// This form of BPF_TEST is now discouraged (but still allowed) in favor of +// BPF_TEST_D and BPF_TEST_C. +// The |policy| parameter should be a Policy subclass. +// BPF_TEST() takes a C++ data type as an fourth parameter. A variable +// of this type will be allocated and a pointer to it will be +// available within the test function as "BPF_AUX". The pointer will +// also be passed as an argument to the policy's constructor. Policies +// would typically use it as an argument to SandboxBPF::Trap(), if +// they want to communicate data between the BPF_TEST() and a Trap() +// function. The life-time of this object is the same as the life-time +// of the process running under the seccomp-bpf policy. +// |aux| must not be void. +#define BPF_TEST(test_case_name, test_name, policy, aux) \ + BPF_DEATH_TEST(test_case_name, test_name, DEATH_SUCCESS(), policy, aux) + +// A BPF_DEATH_TEST is just the same as a BPF_TEST, but it assumes that the +// test will fail with a particular known error condition. Use the DEATH_XXX() +// macros from unit_tests.h to specify the expected error condition. +#define BPF_DEATH_TEST(test_case_name, test_name, death, policy, aux) \ + void BPF_TEST_##test_name(aux* BPF_AUX); \ + TEST(test_case_name, DISABLE_ON_TSAN(test_name)) { \ + sandbox::SandboxBPFTestRunner bpf_test_runner( \ + new sandbox::BPFTesterCompatibilityDelegate<policy, aux>( \ + BPF_TEST_##test_name)); \ + sandbox::UnitTests::RunTestInProcess(&bpf_test_runner, death); \ + } \ + void BPF_TEST_##test_name(aux* BPF_AUX) + +// This class takes a simple function pointer as a constructor parameter and a +// class name as a template parameter to implement the BPFTesterDelegate +// interface which can be used to build BPF unittests with +// the SandboxBPFTestRunner class. +template <class PolicyClass> +class BPFTesterSimpleDelegate : public BPFTesterDelegate { + public: + explicit BPFTesterSimpleDelegate(void (*test_function)(void)) + : test_function_(test_function) {} + ~BPFTesterSimpleDelegate() override {} + + scoped_ptr<bpf_dsl::Policy> GetSandboxBPFPolicy() override { + return scoped_ptr<bpf_dsl::Policy>(new PolicyClass()); + } + void RunTestFunction() override { + DCHECK(test_function_); + test_function_(); + } + + private: + void (*test_function_)(void); + DISALLOW_COPY_AND_ASSIGN(BPFTesterSimpleDelegate); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SECCOMP_BPF_BPF_TESTS_H__ diff --git a/sandbox/linux/seccomp-bpf/bpf_tests_unittest.cc b/sandbox/linux/seccomp-bpf/bpf_tests_unittest.cc new file mode 100644 index 0000000000..63e1814c90 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/bpf_tests_unittest.cc @@ -0,0 +1,153 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/bpf_tests.h" + +#include <errno.h> +#include <sys/ptrace.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +using sandbox::bpf_dsl::Allow; +using sandbox::bpf_dsl::Error; +using sandbox::bpf_dsl::ResultExpr; + +namespace sandbox { + +namespace { + +class FourtyTwo { + public: + static const int kMagicValue = 42; + FourtyTwo() : value_(kMagicValue) {} + int value() { return value_; } + + private: + int value_; + DISALLOW_COPY_AND_ASSIGN(FourtyTwo); +}; + +class EmptyClassTakingPolicy : public bpf_dsl::Policy { + public: + explicit EmptyClassTakingPolicy(FourtyTwo* fourty_two) { + BPF_ASSERT(fourty_two); + BPF_ASSERT(FourtyTwo::kMagicValue == fourty_two->value()); + } + ~EmptyClassTakingPolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + return Allow(); + } +}; + +BPF_TEST(BPFTest, + BPFAUXPointsToClass, + EmptyClassTakingPolicy, + FourtyTwo /* *BPF_AUX */) { + // BPF_AUX should point to an instance of FourtyTwo. + BPF_ASSERT(BPF_AUX); + BPF_ASSERT(FourtyTwo::kMagicValue == BPF_AUX->value()); +} + +void DummyTestFunction(FourtyTwo *fourty_two) { +} + +TEST(BPFTest, BPFTesterCompatibilityDelegateLeakTest) { + // Don't do anything, simply gives dynamic tools an opportunity to detect + // leaks. + { + BPFTesterCompatibilityDelegate<EmptyClassTakingPolicy, FourtyTwo> + simple_delegate(DummyTestFunction); + } + { + // Test polymorphism. + scoped_ptr<BPFTesterDelegate> simple_delegate( + new BPFTesterCompatibilityDelegate<EmptyClassTakingPolicy, FourtyTwo>( + DummyTestFunction)); + } +} + +class EnosysPtracePolicy : public bpf_dsl::Policy { + public: + EnosysPtracePolicy() { my_pid_ = sys_getpid(); } + ~EnosysPtracePolicy() override { + // Policies should be able to bind with the process on which they are + // created. They should never be created in a parent process. + BPF_ASSERT_EQ(my_pid_, sys_getpid()); + } + + ResultExpr EvaluateSyscall(int system_call_number) const override { + CHECK(SandboxBPF::IsValidSyscallNumber(system_call_number)); + if (system_call_number == __NR_ptrace) { + // The EvaluateSyscall function should run in the process that created + // the current object. + BPF_ASSERT_EQ(my_pid_, sys_getpid()); + return Error(ENOSYS); + } else { + return Allow(); + } + } + + private: + pid_t my_pid_; + DISALLOW_COPY_AND_ASSIGN(EnosysPtracePolicy); +}; + +class BasicBPFTesterDelegate : public BPFTesterDelegate { + public: + BasicBPFTesterDelegate() {} + ~BasicBPFTesterDelegate() override {} + + scoped_ptr<bpf_dsl::Policy> GetSandboxBPFPolicy() override { + return scoped_ptr<bpf_dsl::Policy>(new EnosysPtracePolicy()); + } + void RunTestFunction() override { + errno = 0; + int ret = ptrace(PTRACE_TRACEME, -1, NULL, NULL); + BPF_ASSERT(-1 == ret); + BPF_ASSERT(ENOSYS == errno); + } + + private: + DISALLOW_COPY_AND_ASSIGN(BasicBPFTesterDelegate); +}; + +// This is the most powerful and complex way to create a BPF test, but it +// requires a full class definition (BasicBPFTesterDelegate). +BPF_TEST_D(BPFTest, BPFTestWithDelegateClass, BasicBPFTesterDelegate); + +// This is the simplest form of BPF tests. +BPF_TEST_C(BPFTest, BPFTestWithInlineTest, EnosysPtracePolicy) { + errno = 0; + int ret = ptrace(PTRACE_TRACEME, -1, NULL, NULL); + BPF_ASSERT(-1 == ret); + BPF_ASSERT(ENOSYS == errno); +} + +const char kHelloMessage[] = "Hello"; + +BPF_DEATH_TEST_C(BPFTest, + BPFDeathTestWithInlineTest, + DEATH_MESSAGE(kHelloMessage), + EnosysPtracePolicy) { + LOG(ERROR) << kHelloMessage; + _exit(1); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/die.cc b/sandbox/linux/seccomp-bpf/die.cc new file mode 100644 index 0000000000..3baf1f13d9 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/die.cc @@ -0,0 +1,93 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/die.h" + +#include <errno.h> +#include <signal.h> +#include <stdio.h> +#include <sys/prctl.h> +#include <sys/syscall.h> +#include <unistd.h> + +#include <string> + +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "sandbox/linux/seccomp-bpf/syscall.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/system_headers/linux_signal.h" + +namespace sandbox { + +void Die::ExitGroup() { + // exit_group() should exit our program. After all, it is defined as a + // function that doesn't return. But things can theoretically go wrong. + // Especially, since we are dealing with system call filters. Continuing + // execution would be very bad in most cases where ExitGroup() gets called. + // So, we'll try a few other strategies too. + Syscall::Call(__NR_exit_group, 1); + + // We have no idea what our run-time environment looks like. So, signal + // handlers might or might not do the right thing. Try to reset settings + // to a defined state; but we have not way to verify whether we actually + // succeeded in doing so. Nonetheless, triggering a fatal signal could help + // us terminate. + struct sigaction sa = {}; + sa.sa_handler = LINUX_SIG_DFL; + sa.sa_flags = LINUX_SA_RESTART; + sys_sigaction(LINUX_SIGSEGV, &sa, nullptr); + Syscall::Call(__NR_prctl, PR_SET_DUMPABLE, (void*)0, (void*)0, (void*)0); + if (*(volatile char*)0) { + } + + // If there is no way for us to ask for the program to exit, the next + // best thing we can do is to loop indefinitely. Maybe, somebody will notice + // and file a bug... + // We in fact retry the system call inside of our loop so that it will + // stand out when somebody tries to diagnose the problem by using "strace". + for (;;) { + Syscall::Call(__NR_exit_group, 1); + } +} + +void Die::SandboxDie(const char* msg, const char* file, int line) { + if (simple_exit_) { + LogToStderr(msg, file, line); + } else { + logging::LogMessage(file, line, logging::LOG_FATAL).stream() << msg; + } + ExitGroup(); +} + +void Die::RawSandboxDie(const char* msg) { + if (!msg) + msg = ""; + RAW_LOG(FATAL, msg); + ExitGroup(); +} + +void Die::SandboxInfo(const char* msg, const char* file, int line) { + if (!suppress_info_) { + logging::LogMessage(file, line, logging::LOG_INFO).stream() << msg; + } +} + +void Die::LogToStderr(const char* msg, const char* file, int line) { + if (msg) { + char buf[40]; + snprintf(buf, sizeof(buf), "%d", line); + std::string s = std::string(file) + ":" + buf + ":" + msg + "\n"; + + // No need to loop. Short write()s are unlikely and if they happen we + // probably prefer them over a loop that blocks. + ignore_result( + HANDLE_EINTR(Syscall::Call(__NR_write, 2, s.c_str(), s.length()))); + } +} + +bool Die::simple_exit_ = false; +bool Die::suppress_info_ = false; + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/die.h b/sandbox/linux/seccomp-bpf/die.h new file mode 100644 index 0000000000..b3f3f72c2f --- /dev/null +++ b/sandbox/linux/seccomp-bpf/die.h @@ -0,0 +1,68 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_DIE_H__ +#define SANDBOX_LINUX_SECCOMP_BPF_DIE_H__ + +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// This is the main API for using this file. Prints a error message and +// exits with a fatal error. This is not async-signal safe. +#define SANDBOX_DIE(m) sandbox::Die::SandboxDie(m, __FILE__, __LINE__) + +// An async signal safe version of the same API. Won't print the filename +// and line numbers. +#define RAW_SANDBOX_DIE(m) sandbox::Die::RawSandboxDie(m) + +// Adds an informational message to the log file or stderr as appropriate. +#define SANDBOX_INFO(m) sandbox::Die::SandboxInfo(m, __FILE__, __LINE__) + +class SANDBOX_EXPORT Die { + public: + // Terminate the program, even if the current sandbox policy prevents some + // of the more commonly used functions used for exiting. + // Most users would want to call SANDBOX_DIE() instead, as it logs extra + // information. But calling ExitGroup() is correct and in some rare cases + // preferable. So, we make it part of the public API. + static void ExitGroup() __attribute__((noreturn)); + + // This method gets called by SANDBOX_DIE(). There is normally no reason + // to call it directly unless you are defining your own exiting macro. + static void SandboxDie(const char* msg, const char* file, int line) + __attribute__((noreturn)); + + static void RawSandboxDie(const char* msg) __attribute__((noreturn)); + + // This method gets called by SANDBOX_INFO(). There is normally no reason + // to call it directly unless you are defining your own logging macro. + static void SandboxInfo(const char* msg, const char* file, int line); + + // Writes a message to stderr. Used as a fall-back choice, if we don't have + // any other way to report an error. + static void LogToStderr(const char* msg, const char* file, int line); + + // We generally want to run all exit handlers. This means, on SANDBOX_DIE() + // we should be calling LOG(FATAL). But there are some situations where + // we just need to print a message and then terminate. This would typically + // happen in cases where we consume the error message internally (e.g. in + // unit tests or in the supportsSeccompSandbox() method). + static void EnableSimpleExit() { simple_exit_ = true; } + + // Sometimes we need to disable all informational messages (e.g. from within + // unittests). + static void SuppressInfoMessages(bool flag) { suppress_info_ = flag; } + + private: + static bool simple_exit_; + static bool suppress_info_; + + DISALLOW_IMPLICIT_CONSTRUCTORS(Die); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SECCOMP_BPF_DIE_H__ diff --git a/sandbox/linux/seccomp-bpf/errorcode.cc b/sandbox/linux/seccomp-bpf/errorcode.cc new file mode 100644 index 0000000000..9bb3ddb648 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/errorcode.cc @@ -0,0 +1,115 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/errorcode.h" + +#include "sandbox/linux/seccomp-bpf/die.h" +#include "sandbox/linux/system_headers/linux_seccomp.h" + +namespace sandbox { + +ErrorCode::ErrorCode() : error_type_(ET_INVALID), err_(SECCOMP_RET_INVALID) { +} + +ErrorCode::ErrorCode(int err) { + switch (err) { + case ERR_ALLOWED: + err_ = SECCOMP_RET_ALLOW; + error_type_ = ET_SIMPLE; + break; + case ERR_MIN_ERRNO... ERR_MAX_ERRNO: + err_ = SECCOMP_RET_ERRNO + err; + error_type_ = ET_SIMPLE; + break; + default: + if ((err & ~SECCOMP_RET_DATA) == ERR_TRACE) { + err_ = SECCOMP_RET_TRACE + (err & SECCOMP_RET_DATA); + error_type_ = ET_SIMPLE; + break; + } + SANDBOX_DIE("Invalid use of ErrorCode object"); + } +} + +ErrorCode::ErrorCode(uint16_t trap_id, + Trap::TrapFnc fnc, + const void* aux, + bool safe) + : error_type_(ET_TRAP), + fnc_(fnc), + aux_(const_cast<void*>(aux)), + safe_(safe), + err_(SECCOMP_RET_TRAP + trap_id) { +} + +ErrorCode::ErrorCode(int argno, + ArgType width, + uint64_t mask, + uint64_t value, + const ErrorCode* passed, + const ErrorCode* failed) + : error_type_(ET_COND), + mask_(mask), + value_(value), + argno_(argno), + width_(width), + passed_(passed), + failed_(failed), + err_(SECCOMP_RET_INVALID) { +} + +bool ErrorCode::Equals(const ErrorCode& err) const { + if (error_type_ == ET_INVALID || err.error_type_ == ET_INVALID) { + SANDBOX_DIE("Dereferencing invalid ErrorCode"); + } + if (error_type_ != err.error_type_) { + return false; + } + if (error_type_ == ET_SIMPLE || error_type_ == ET_TRAP) { + return err_ == err.err_; + } else if (error_type_ == ET_COND) { + return mask_ == err.mask_ && value_ == err.value_ && argno_ == err.argno_ && + width_ == err.width_ && passed_->Equals(*err.passed_) && + failed_->Equals(*err.failed_); + } else { + SANDBOX_DIE("Corrupted ErrorCode"); + } +} + +bool ErrorCode::LessThan(const ErrorCode& err) const { + // Implementing a "LessThan()" operator allows us to use ErrorCode objects + // as keys in STL containers; most notably, it also allows us to put them + // into std::set<>. Actual ordering is not important as long as it is + // deterministic. + if (error_type_ == ET_INVALID || err.error_type_ == ET_INVALID) { + SANDBOX_DIE("Dereferencing invalid ErrorCode"); + } + if (error_type_ != err.error_type_) { + return error_type_ < err.error_type_; + } else { + if (error_type_ == ET_SIMPLE || error_type_ == ET_TRAP) { + return err_ < err.err_; + } else if (error_type_ == ET_COND) { + if (mask_ != err.mask_) { + return mask_ < err.mask_; + } else if (value_ != err.value_) { + return value_ < err.value_; + } else if (argno_ != err.argno_) { + return argno_ < err.argno_; + } else if (width_ != err.width_) { + return width_ < err.width_; + } else if (!passed_->Equals(*err.passed_)) { + return passed_->LessThan(*err.passed_); + } else if (!failed_->Equals(*err.failed_)) { + return failed_->LessThan(*err.failed_); + } else { + return false; + } + } else { + SANDBOX_DIE("Corrupted ErrorCode"); + } + } +} + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/errorcode.h b/sandbox/linux/seccomp-bpf/errorcode.h new file mode 100644 index 0000000000..d88777313e --- /dev/null +++ b/sandbox/linux/seccomp-bpf/errorcode.h @@ -0,0 +1,203 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_ERRORCODE_H__ +#define SANDBOX_LINUX_SECCOMP_BPF_ERRORCODE_H__ + +#include "sandbox/linux/seccomp-bpf/trap.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { +namespace bpf_dsl { +class PolicyCompiler; +} + +// This class holds all the possible values that can be returned by a sandbox +// policy. +// We can either wrap a symbolic ErrorCode (i.e. ERR_XXX enum values), an +// errno value (in the range 0..4095), a pointer to a TrapFnc callback +// handling a SECCOMP_RET_TRAP trap, or a complex constraint. +// All of the commonly used values are stored in the "err_" field. So, code +// that is using the ErrorCode class typically operates on a single 32bit +// field. +// +// TODO(mdempsky): Nuke from orbit. The only reason this class still +// exists is for Verifier, which will eventually be replaced by a true +// BPF symbolic evaluator and constraint solver. +class SANDBOX_EXPORT ErrorCode { + public: + enum { + // Allow this system call. The value of ERR_ALLOWED is pretty much + // completely arbitrary. But we want to pick it so that is is unlikely + // to be passed in accidentally, when the user intended to return an + // "errno" (see below) value instead. + ERR_ALLOWED = 0x04000000, + + // If the progress is being ptraced with PTRACE_O_TRACESECCOMP, then the + // tracer will be notified of a PTRACE_EVENT_SECCOMP and allowed to change + // or skip the system call. The lower 16 bits of err will be available to + // the tracer via PTRACE_GETEVENTMSG. + ERR_TRACE = 0x08000000, + + // Deny the system call with a particular "errno" value. + // N.B.: It is also possible to return "0" here. That would normally + // indicate success, but it won't actually run the system call. + // This is very different from return ERR_ALLOWED. + ERR_MIN_ERRNO = 0, +#if defined(__mips__) + // MIPS only supports errno up to 1133 + ERR_MAX_ERRNO = 1133, +#else + // TODO(markus): Android only supports errno up to 255 + // (crbug.com/181647). + ERR_MAX_ERRNO = 4095, +#endif + }; + + // While BPF filter programs always operate on 32bit quantities, the kernel + // always sees system call arguments as 64bit values. This statement is true + // no matter whether the host system is natively operating in 32bit or 64bit. + // The BPF compiler hides the fact that BPF instructions cannot directly + // access 64bit quantities. But policies are still advised to specify whether + // a system call expects a 32bit or a 64bit quantity. + enum ArgType { + // When passed as an argument to SandboxBPF::Cond(), TP_32BIT requests that + // the conditional test should operate on the 32bit part of the system call + // argument. + // On 64bit architectures, this verifies that user space did not pass + // a 64bit value as an argument to the system call. If it did, that will be + // interpreted as an attempt at breaking the sandbox and results in the + // program getting terminated. + // In other words, only perform a 32bit test, if you are sure this + // particular system call would never legitimately take a 64bit + // argument. + // Implementation detail: TP_32BIT does two things. 1) it restricts the + // conditional test to operating on the LSB only, and 2) it adds code to + // the BPF filter program verifying that the MSB the kernel received from + // user space is either 0, or 0xFFFFFFFF; the latter is acceptable, iff bit + // 31 was set in the system call argument. It deals with 32bit arguments + // having been sign extended. + TP_32BIT, + + // When passed as an argument to SandboxBPF::Cond(), TP_64BIT requests that + // the conditional test should operate on the full 64bit argument. It is + // generally harmless to perform a 64bit test on 32bit systems, as the + // kernel will always see the top 32 bits of all arguments as zero'd out. + // This approach has the desirable property that for tests of pointer + // values, we can always use TP_64BIT no matter the host architecture. + // But of course, that also means, it is possible to write conditional + // policies that turn into no-ops on 32bit systems; this is by design. + TP_64BIT, + }; + + // Deprecated. + enum Operation { + // Test whether the system call argument is equal to the operand. + OP_EQUAL, + + // Tests a system call argument against a bit mask. + // The "ALL_BITS" variant performs this test: "arg & mask == mask" + // This implies that a mask of zero always results in a passing test. + // The "ANY_BITS" variant performs this test: "arg & mask != 0" + // This implies that a mask of zero always results in a failing test. + OP_HAS_ALL_BITS, + OP_HAS_ANY_BITS, + }; + + enum ErrorType { + ET_INVALID, + ET_SIMPLE, + ET_TRAP, + ET_COND, + }; + + // We allow the default constructor, as it makes the ErrorCode class + // much easier to use. But if we ever encounter an invalid ErrorCode + // when compiling a BPF filter, we deliberately generate an invalid + // program that will get flagged both by our Verifier class and by + // the Linux kernel. + ErrorCode(); + explicit ErrorCode(int err); + + // For all practical purposes, ErrorCodes are treated as if they were + // structs. The copy constructor and assignment operator are trivial and + // we do not need to explicitly specify them. + // Most notably, it is in fact perfectly OK to directly copy the passed_ and + // failed_ field. They only ever get set by our private constructor, and the + // callers handle life-cycle management for these objects. + + // Destructor + ~ErrorCode() {} + + bool Equals(const ErrorCode& err) const; + bool LessThan(const ErrorCode& err) const; + + uint32_t err() const { return err_; } + ErrorType error_type() const { return error_type_; } + + bool safe() const { return safe_; } + + uint64_t mask() const { return mask_; } + uint64_t value() const { return value_; } + int argno() const { return argno_; } + ArgType width() const { return width_; } + const ErrorCode* passed() const { return passed_; } + const ErrorCode* failed() const { return failed_; } + + struct LessThan { + bool operator()(const ErrorCode& a, const ErrorCode& b) const { + return a.LessThan(b); + } + }; + + private: + friend bpf_dsl::PolicyCompiler; + friend class CodeGen; + friend class SandboxBPF; + friend class Trap; + + // If we are wrapping a callback, we must assign a unique id. This id is + // how the kernel tells us which one of our different SECCOMP_RET_TRAP + // cases has been triggered. + ErrorCode(uint16_t trap_id, Trap::TrapFnc fnc, const void* aux, bool safe); + + // Some system calls require inspection of arguments. This constructor + // allows us to specify additional constraints. + ErrorCode(int argno, + ArgType width, + uint64_t mask, + uint64_t value, + const ErrorCode* passed, + const ErrorCode* failed); + + ErrorType error_type_; + + union { + // Fields needed for SECCOMP_RET_TRAP callbacks + struct { + Trap::TrapFnc fnc_; // Callback function and arg, if trap was + void* aux_; // triggered by the kernel's BPF filter. + bool safe_; // Keep sandbox active while calling fnc_() + }; + + // Fields needed when inspecting additional arguments. + struct { + uint64_t mask_; // Mask that we are comparing under. + uint64_t value_; // Value that we are comparing with. + int argno_; // Syscall arg number that we are inspecting. + ArgType width_; // Whether we are looking at a 32/64bit value. + const ErrorCode* passed_; // Value to be returned if comparison passed, + const ErrorCode* failed_; // or if it failed. + }; + }; + + // 32bit field used for all possible types of ErrorCode values. This is + // the value that uniquely identifies any ErrorCode and it (typically) can + // be emitted directly into a BPF filter program. + uint32_t err_; +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SECCOMP_BPF_ERRORCODE_H__ diff --git a/sandbox/linux/seccomp-bpf/errorcode_unittest.cc b/sandbox/linux/seccomp-bpf/errorcode_unittest.cc new file mode 100644 index 0000000000..6b5491ee4a --- /dev/null +++ b/sandbox/linux/seccomp-bpf/errorcode_unittest.cc @@ -0,0 +1,120 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/errorcode.h" + +#include <errno.h> + +#include "base/macros.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/bpf_dsl/policy_compiler.h" +#include "sandbox/linux/seccomp-bpf/trap.h" +#include "sandbox/linux/system_headers/linux_seccomp.h" +#include "sandbox/linux/tests/unit_tests.h" + +namespace sandbox { + +namespace { + +class DummyPolicy : public bpf_dsl::Policy { + public: + DummyPolicy() {} + ~DummyPolicy() override {} + + bpf_dsl::ResultExpr EvaluateSyscall(int sysno) const override { + return bpf_dsl::Allow(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(DummyPolicy); +}; + +SANDBOX_TEST(ErrorCode, ErrnoConstructor) { + ErrorCode e0; + SANDBOX_ASSERT(e0.err() == SECCOMP_RET_INVALID); + + ErrorCode e1(ErrorCode::ERR_ALLOWED); + SANDBOX_ASSERT(e1.err() == SECCOMP_RET_ALLOW); + + ErrorCode e2(EPERM); + SANDBOX_ASSERT(e2.err() == SECCOMP_RET_ERRNO + EPERM); + + DummyPolicy dummy_policy; + bpf_dsl::PolicyCompiler compiler(&dummy_policy, Trap::Registry()); + ErrorCode e3 = compiler.Trap(NULL, NULL, true /* safe */); + SANDBOX_ASSERT((e3.err() & SECCOMP_RET_ACTION) == SECCOMP_RET_TRAP); + + uint16_t data = 0xdead; + ErrorCode e4(ErrorCode::ERR_TRACE + data); + SANDBOX_ASSERT(e4.err() == SECCOMP_RET_TRACE + data); +} + +SANDBOX_DEATH_TEST(ErrorCode, + InvalidSeccompRetTrace, + DEATH_MESSAGE("Invalid use of ErrorCode object")) { + // Should die if the trace data does not fit in 16 bits. + ErrorCode e(ErrorCode::ERR_TRACE + (1 << 16)); +} + +SANDBOX_TEST(ErrorCode, Trap) { + DummyPolicy dummy_policy; + bpf_dsl::PolicyCompiler compiler(&dummy_policy, Trap::Registry()); + ErrorCode e0 = compiler.Trap(NULL, "a", true /* safe */); + ErrorCode e1 = compiler.Trap(NULL, "b", true /* safe */); + SANDBOX_ASSERT((e0.err() & SECCOMP_RET_DATA) + 1 == + (e1.err() & SECCOMP_RET_DATA)); + + ErrorCode e2 = compiler.Trap(NULL, "a", true /* safe */); + SANDBOX_ASSERT((e0.err() & SECCOMP_RET_DATA) == + (e2.err() & SECCOMP_RET_DATA)); +} + +SANDBOX_TEST(ErrorCode, Equals) { + ErrorCode e1(ErrorCode::ERR_ALLOWED); + ErrorCode e2(ErrorCode::ERR_ALLOWED); + SANDBOX_ASSERT(e1.Equals(e1)); + SANDBOX_ASSERT(e1.Equals(e2)); + SANDBOX_ASSERT(e2.Equals(e1)); + + ErrorCode e3(EPERM); + SANDBOX_ASSERT(!e1.Equals(e3)); + + DummyPolicy dummy_policy; + bpf_dsl::PolicyCompiler compiler(&dummy_policy, Trap::Registry()); + ErrorCode e4 = compiler.Trap(NULL, "a", true /* safe */); + ErrorCode e5 = compiler.Trap(NULL, "b", true /* safe */); + ErrorCode e6 = compiler.Trap(NULL, "a", true /* safe */); + SANDBOX_ASSERT(!e1.Equals(e4)); + SANDBOX_ASSERT(!e3.Equals(e4)); + SANDBOX_ASSERT(!e5.Equals(e4)); + SANDBOX_ASSERT( e6.Equals(e4)); +} + +SANDBOX_TEST(ErrorCode, LessThan) { + ErrorCode e1(ErrorCode::ERR_ALLOWED); + ErrorCode e2(ErrorCode::ERR_ALLOWED); + SANDBOX_ASSERT(!e1.LessThan(e1)); + SANDBOX_ASSERT(!e1.LessThan(e2)); + SANDBOX_ASSERT(!e2.LessThan(e1)); + + ErrorCode e3(EPERM); + SANDBOX_ASSERT(!e1.LessThan(e3)); + SANDBOX_ASSERT( e3.LessThan(e1)); + + DummyPolicy dummy_policy; + bpf_dsl::PolicyCompiler compiler(&dummy_policy, Trap::Registry()); + ErrorCode e4 = compiler.Trap(NULL, "a", true /* safe */); + ErrorCode e5 = compiler.Trap(NULL, "b", true /* safe */); + ErrorCode e6 = compiler.Trap(NULL, "a", true /* safe */); + SANDBOX_ASSERT(e1.LessThan(e4)); + SANDBOX_ASSERT(e3.LessThan(e4)); + SANDBOX_ASSERT(e4.LessThan(e5)); + SANDBOX_ASSERT(!e4.LessThan(e6)); + SANDBOX_ASSERT(!e6.LessThan(e4)); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/sandbox_bpf.cc b/sandbox/linux/seccomp-bpf/sandbox_bpf.cc new file mode 100644 index 0000000000..239043eb27 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/sandbox_bpf.cc @@ -0,0 +1,279 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" + +// Some headers on Android are missing cdefs: crbug.com/172337. +// (We can't use OS_ANDROID here since build_config.h is not included). +#if defined(ANDROID) +#include <sys/cdefs.h> +#endif + +#include <errno.h> +#include <sys/prctl.h> +#include <sys/types.h> +#include <unistd.h> + +#include "base/compiler_specific.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "base/posix/eintr_wrapper.h" +#include "base/third_party/valgrind/valgrind.h" +#include "sandbox/linux/bpf_dsl/codegen.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/bpf_dsl/policy_compiler.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" +#include "sandbox/linux/bpf_dsl/syscall_set.h" +#include "sandbox/linux/seccomp-bpf/die.h" +#include "sandbox/linux/seccomp-bpf/syscall.h" +#include "sandbox/linux/seccomp-bpf/trap.h" +#include "sandbox/linux/services/proc_util.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/services/thread_helpers.h" +#include "sandbox/linux/system_headers/linux_filter.h" +#include "sandbox/linux/system_headers/linux_seccomp.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" + +namespace sandbox { + +namespace { + +bool IsRunningOnValgrind() { return RUNNING_ON_VALGRIND; } + +bool IsSingleThreaded(int proc_fd) { + return ThreadHelpers::IsSingleThreaded(proc_fd); +} + +// Check if the kernel supports seccomp-filter (a.k.a. seccomp mode 2) via +// prctl(). +bool KernelSupportsSeccompBPF() { + errno = 0; + const int rv = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, nullptr); + + if (rv == -1 && EFAULT == errno) { + return true; + } + return false; +} + +// LG introduced a buggy syscall, sys_set_media_ext, with the same number as +// seccomp. Return true if the current kernel has this buggy syscall. +// +// We want this to work with upcoming versions of seccomp, so we pass bogus +// flags that are unlikely to ever be used by the kernel. A normal kernel would +// return -EINVAL, but a buggy LG kernel would return 1. +bool KernelHasLGBug() { +#if defined(OS_ANDROID) + // sys_set_media will see this as NULL, which should be a safe (non-crashing) + // way to invoke it. A genuine seccomp syscall will see it as + // SECCOMP_SET_MODE_STRICT. + const unsigned int operation = 0; + // Chosen by fair dice roll. Guaranteed to be random. + const unsigned int flags = 0xf7a46a5c; + const int rv = sys_seccomp(operation, flags, nullptr); + // A genuine kernel would return -EINVAL (which would set rv to -1 and errno + // to EINVAL), or at the very least return some kind of error (which would + // set rv to -1). Any other behavior indicates that whatever code received + // our syscall was not the real seccomp. + if (rv != -1) { + return true; + } +#endif // defined(OS_ANDROID) + + return false; +} + +// Check if the kernel supports seccomp-filter via the seccomp system call +// and the TSYNC feature to enable seccomp on all threads. +bool KernelSupportsSeccompTsync() { + if (KernelHasLGBug()) { + return false; + } + + errno = 0; + const int rv = + sys_seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, nullptr); + + if (rv == -1 && errno == EFAULT) { + return true; + } else { + // TODO(jln): turn these into DCHECK after 417888 is considered fixed. + CHECK_EQ(-1, rv); + CHECK(ENOSYS == errno || EINVAL == errno); + return false; + } +} + +uint64_t EscapePC() { + intptr_t rv = Syscall::Call(-1); + if (rv == -1 && errno == ENOSYS) { + return 0; + } + return static_cast<uint64_t>(static_cast<uintptr_t>(rv)); +} + +} // namespace + +SandboxBPF::SandboxBPF(bpf_dsl::Policy* policy) + : proc_fd_(), sandbox_has_started_(false), policy_(policy) { +} + +SandboxBPF::~SandboxBPF() { +} + +// static +bool SandboxBPF::SupportsSeccompSandbox(SeccompLevel level) { + // Never pretend to support seccomp with Valgrind, as it + // throws the tool off. + if (IsRunningOnValgrind()) { + return false; + } + + switch (level) { + case SeccompLevel::SINGLE_THREADED: + return KernelSupportsSeccompBPF(); + case SeccompLevel::MULTI_THREADED: + return KernelSupportsSeccompTsync(); + } + NOTREACHED(); + return false; +} + +bool SandboxBPF::StartSandbox(SeccompLevel seccomp_level) { + DCHECK(policy_); + CHECK(seccomp_level == SeccompLevel::SINGLE_THREADED || + seccomp_level == SeccompLevel::MULTI_THREADED); + + if (sandbox_has_started_) { + SANDBOX_DIE( + "Cannot repeatedly start sandbox. Create a separate Sandbox " + "object instead."); + return false; + } + + if (!proc_fd_.is_valid()) { + SetProcFd(ProcUtil::OpenProc()); + } + + const bool supports_tsync = KernelSupportsSeccompTsync(); + + if (seccomp_level == SeccompLevel::SINGLE_THREADED) { + // Wait for /proc/self/task/ to update if needed and assert the + // process is single threaded. + ThreadHelpers::AssertSingleThreaded(proc_fd_.get()); + } else if (seccomp_level == SeccompLevel::MULTI_THREADED) { + if (IsSingleThreaded(proc_fd_.get())) { + SANDBOX_DIE("Cannot start sandbox; " + "process may be single-threaded when reported as not"); + return false; + } + if (!supports_tsync) { + SANDBOX_DIE("Cannot start sandbox; kernel does not support synchronizing " + "filters for a threadgroup"); + return false; + } + } + + // We no longer need access to any files in /proc. We want to do this + // before installing the filters, just in case that our policy denies + // close(). + if (proc_fd_.is_valid()) { + proc_fd_.reset(); + } + + // Install the filters. + InstallFilter(supports_tsync || + seccomp_level == SeccompLevel::MULTI_THREADED); + + return true; +} + +void SandboxBPF::SetProcFd(base::ScopedFD proc_fd) { + proc_fd_.swap(proc_fd); +} + +// static +bool SandboxBPF::IsValidSyscallNumber(int sysnum) { + return SyscallSet::IsValid(sysnum); +} + +// static +bool SandboxBPF::IsRequiredForUnsafeTrap(int sysno) { + return bpf_dsl::PolicyCompiler::IsRequiredForUnsafeTrap(sysno); +} + +// static +intptr_t SandboxBPF::ForwardSyscall(const struct arch_seccomp_data& args) { + return Syscall::Call( + args.nr, static_cast<intptr_t>(args.args[0]), + static_cast<intptr_t>(args.args[1]), static_cast<intptr_t>(args.args[2]), + static_cast<intptr_t>(args.args[3]), static_cast<intptr_t>(args.args[4]), + static_cast<intptr_t>(args.args[5])); +} + +scoped_ptr<CodeGen::Program> SandboxBPF::AssembleFilter( + bool force_verification) { +#if !defined(NDEBUG) + force_verification = true; +#endif + DCHECK(policy_); + + bpf_dsl::PolicyCompiler compiler(policy_.get(), Trap::Registry()); + if (Trap::SandboxDebuggingAllowedByUser()) { + compiler.DangerousSetEscapePC(EscapePC()); + } + return compiler.Compile(force_verification); +} + +void SandboxBPF::InstallFilter(bool must_sync_threads) { + // We want to be very careful in not imposing any requirements on the + // policies that are set with SetSandboxPolicy(). This means, as soon as + // the sandbox is active, we shouldn't be relying on libraries that could + // be making system calls. This, for example, means we should avoid + // using the heap and we should avoid using STL functions. + // Temporarily copy the contents of the "program" vector into a + // stack-allocated array; and then explicitly destroy that object. + // This makes sure we don't ex- or implicitly call new/delete after we + // installed the BPF filter program in the kernel. Depending on the + // system memory allocator that is in effect, these operators can result + // in system calls to things like munmap() or brk(). + CodeGen::Program* program = AssembleFilter(false).release(); + + struct sock_filter bpf[program->size()]; + const struct sock_fprog prog = {static_cast<unsigned short>(program->size()), + bpf}; + memcpy(bpf, &(*program)[0], sizeof(bpf)); + delete program; + + // Make an attempt to release memory that is no longer needed here, rather + // than in the destructor. Try to avoid as much as possible to presume of + // what will be possible to do in the new (sandboxed) execution environment. + policy_.reset(); + + if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { + SANDBOX_DIE("Kernel refuses to enable no-new-privs"); + } + + // Install BPF filter program. If the thread state indicates multi-threading + // support, then the kernel hass the seccomp system call. Otherwise, fall + // back on prctl, which requires the process to be single-threaded. + if (must_sync_threads) { + int rv = + sys_seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, &prog); + if (rv) { + SANDBOX_DIE( + "Kernel refuses to turn on and synchronize threads for BPF filters"); + } + } else { + if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog)) { + SANDBOX_DIE("Kernel refuses to turn on BPF filters"); + } + } + + sandbox_has_started_ = true; +} + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/sandbox_bpf.h b/sandbox/linux/seccomp-bpf/sandbox_bpf.h new file mode 100644 index 0000000000..96cceb5648 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/sandbox_bpf.h @@ -0,0 +1,118 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_SANDBOX_BPF_H_ +#define SANDBOX_LINUX_SECCOMP_BPF_SANDBOX_BPF_H_ + +#include <stdint.h> + +#include "base/files/scoped_file.h" +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "sandbox/linux/bpf_dsl/codegen.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { +struct arch_seccomp_data; +namespace bpf_dsl { +class Policy; +} + +// This class can be used to apply a syscall sandboxing policy expressed in a +// bpf_dsl::Policy object to the current process. +// Syscall sandboxing policies get inherited by subprocesses and, once applied, +// can never be removed for the lifetime of the process. +class SANDBOX_EXPORT SandboxBPF { + public: + enum class SeccompLevel { + SINGLE_THREADED, + MULTI_THREADED, + }; + + // Ownership of |policy| is transfered here to the sandbox object. + // nullptr is allowed for unit tests. + explicit SandboxBPF(bpf_dsl::Policy* policy); + // NOTE: Setting a policy and starting the sandbox is a one-way operation. + // The kernel does not provide any option for unloading a loaded sandbox. The + // sandbox remains engaged even when the object is destructed. + ~SandboxBPF(); + + // Detect if the kernel supports the specified seccomp level. + // See StartSandbox() for a description of these. + static bool SupportsSeccompSandbox(SeccompLevel level); + + // This is the main public entry point. It sets up the resources needed by + // the sandbox, and enters Seccomp mode. + // The calling process must provide a |level| to tell the sandbox which type + // of kernel support it should engage. + // SINGLE_THREADED will only sandbox the calling thread. Since it would be a + // security risk, the sandbox will also check that the current process is + // single threaded and crash if it isn't the case. + // MULTI_THREADED requires more recent kernel support and allows to sandbox + // all the threads of the current process. Be mindful of potential races, + // with other threads using disallowed system calls either before or after + // the sandbox is engaged. + // + // It is possible to stack multiple sandboxes by creating separate "Sandbox" + // objects and calling "StartSandbox()" on each of them. Please note, that + // this requires special care, though, as newly stacked sandboxes can never + // relax restrictions imposed by earlier sandboxes. Furthermore, installing + // a new policy requires making system calls, that might already be + // disallowed. + // Finally, stacking does add more kernel overhead than having a single + // combined policy. So, it should only be used if there are no alternatives. + bool StartSandbox(SeccompLevel level) WARN_UNUSED_RESULT; + + // The sandbox needs to be able to access files in "/proc/self/". If + // this directory is not accessible when "StartSandbox()" gets called, the + // caller must provide an already opened file descriptor by calling + // "SetProcFd()". + // The sandbox becomes the new owner of this file descriptor and will + // close it when "StartSandbox()" executes or when the sandbox object + // disappears. + void SetProcFd(base::ScopedFD proc_fd); + + // Checks whether a particular system call number is valid on the current + // architecture. + static bool IsValidSyscallNumber(int sysnum); + + // UnsafeTraps require some syscalls to always be allowed. + // This helper function returns true for these calls. + static bool IsRequiredForUnsafeTrap(int sysno); + + // From within an UnsafeTrap() it is often useful to be able to execute + // the system call that triggered the trap. The ForwardSyscall() method + // makes this easy. It is more efficient than calling glibc's syscall() + // function, as it avoid the extra round-trip to the signal handler. And + // it automatically does the correct thing to report kernel-style error + // conditions, rather than setting errno. See the comments for TrapFnc for + // details. In other words, the return value from ForwardSyscall() is + // directly suitable as a return value for a trap handler. + static intptr_t ForwardSyscall(const struct arch_seccomp_data& args); + + // Assembles a BPF filter program from the current policy. After calling this + // function, you must not call any other sandboxing function. + // Typically, AssembleFilter() is only used by unit tests and by sandbox + // internals. It should not be used by production code. + // For performance reasons, we normally only run the assembled BPF program + // through the verifier, iff the program was built in debug mode. + // But by setting "force_verification", the caller can request that the + // verifier is run unconditionally. This is useful for unittests. + scoped_ptr<CodeGen::Program> AssembleFilter(bool force_verification); + + private: + // Assembles and installs a filter based on the policy that has previously + // been configured with SetSandboxPolicy(). + void InstallFilter(bool must_sync_threads); + + base::ScopedFD proc_fd_; + bool sandbox_has_started_; + scoped_ptr<bpf_dsl::Policy> policy_; + + DISALLOW_COPY_AND_ASSIGN(SandboxBPF); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SECCOMP_BPF_SANDBOX_BPF_H_ diff --git a/sandbox/linux/seccomp-bpf/sandbox_bpf_test_runner.cc b/sandbox/linux/seccomp-bpf/sandbox_bpf_test_runner.cc new file mode 100644 index 0000000000..321ea9a8ee --- /dev/null +++ b/sandbox/linux/seccomp-bpf/sandbox_bpf_test_runner.cc @@ -0,0 +1,65 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/sandbox_bpf_test_runner.h" + +#include <fcntl.h> + +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/seccomp-bpf/die.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/system_headers/linux_filter.h" +#include "sandbox/linux/tests/unit_tests.h" + +namespace sandbox { + +SandboxBPFTestRunner::SandboxBPFTestRunner( + BPFTesterDelegate* bpf_tester_delegate) + : bpf_tester_delegate_(bpf_tester_delegate) { +} + +SandboxBPFTestRunner::~SandboxBPFTestRunner() { +} + +void SandboxBPFTestRunner::Run() { + DCHECK(bpf_tester_delegate_); + sandbox::Die::EnableSimpleExit(); + + scoped_ptr<bpf_dsl::Policy> policy = + bpf_tester_delegate_->GetSandboxBPFPolicy(); + + if (sandbox::SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::SINGLE_THREADED)) { + // Initialize and then start the sandbox with our custom policy + sandbox::SandboxBPF sandbox(policy.release()); + SANDBOX_ASSERT(sandbox.StartSandbox( + sandbox::SandboxBPF::SeccompLevel::SINGLE_THREADED)); + + // Run the actual test. + bpf_tester_delegate_->RunTestFunction(); + } else { + printf("This BPF test is not fully running in this configuration!\n"); + // Android and Valgrind are the only configurations where we accept not + // having kernel BPF support. + if (!IsAndroid() && !IsRunningOnValgrind()) { + const bool seccomp_bpf_is_supported = false; + SANDBOX_ASSERT(seccomp_bpf_is_supported); + } + // Call the compiler and verify the policy. That's the least we can do, + // if we don't have kernel support. + sandbox::SandboxBPF sandbox(policy.release()); + sandbox.AssembleFilter(true /* force_verification */); + sandbox::UnitTests::IgnoreThisTest(); + } +} + +bool SandboxBPFTestRunner::ShouldCheckForLeaks() const { + // LSAN requires being able to use ptrace() and other system calls that could + // be denied. + return false; +} + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/sandbox_bpf_test_runner.h b/sandbox/linux/seccomp-bpf/sandbox_bpf_test_runner.h new file mode 100644 index 0000000000..fef6240d74 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/sandbox_bpf_test_runner.h @@ -0,0 +1,61 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_SANDBOX_BPF_TEST_RUNNER_H_ +#define SANDBOX_LINUX_SECCOMP_BPF_SANDBOX_BPF_TEST_RUNNER_H_ + +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "sandbox/linux/tests/sandbox_test_runner.h" + +namespace sandbox { +namespace bpf_dsl { +class Policy; +} + +// To create a SandboxBPFTestRunner object, one needs to implement this +// interface and pass an instance to the SandboxBPFTestRunner constructor. +// In the child process running the test, the BPFTesterDelegate object is +// guaranteed to not be destroyed until the child process terminates. +class BPFTesterDelegate { + public: + BPFTesterDelegate() {} + virtual ~BPFTesterDelegate() {} + + // This will instanciate a policy suitable for the test we want to run. It is + // guaranteed to only be called from the child process that will run the + // test. + virtual scoped_ptr<bpf_dsl::Policy> GetSandboxBPFPolicy() = 0; + // This will be called from a child process with the BPF sandbox turned on. + virtual void RunTestFunction() = 0; + + private: + DISALLOW_COPY_AND_ASSIGN(BPFTesterDelegate); +}; + +// This class implements the SandboxTestRunner interface and Run() will +// initialize a seccomp-bpf sandbox (specified by |bpf_tester_delegate|) and +// run a test function (via |bpf_tester_delegate|) if the current kernel +// configuration allows it. If it can not run the test under seccomp-bpf, +// Run() will still compile the policy which should allow to get some coverage +// under tools such as Valgrind. +class SandboxBPFTestRunner : public SandboxTestRunner { + public: + // This constructor takes ownership of the |bpf_tester_delegate| object. + // (It doesn't take a scoped_ptr since they make polymorphism verbose). + explicit SandboxBPFTestRunner(BPFTesterDelegate* bpf_tester_delegate); + ~SandboxBPFTestRunner() override; + + void Run() override; + + bool ShouldCheckForLeaks() const override; + + private: + scoped_ptr<BPFTesterDelegate> bpf_tester_delegate_; + DISALLOW_COPY_AND_ASSIGN(SandboxBPFTestRunner); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SECCOMP_BPF_SANDBOX_BPF_TEST_RUNNER_H_ diff --git a/sandbox/linux/seccomp-bpf/sandbox_bpf_unittest.cc b/sandbox/linux/seccomp-bpf/sandbox_bpf_unittest.cc new file mode 100644 index 0000000000..580cad2525 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/sandbox_bpf_unittest.cc @@ -0,0 +1,85 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" + +#include <fcntl.h> +#include <unistd.h> + +#include <iostream> + +#include "base/files/scoped_file.h" +#include "base/posix/eintr_wrapper.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { +namespace { + +// NOTE: most tests for the SandboxBPF class are currently in +// integration_tests/. + +TEST(SandboxBPF, CreateDestroy) { + // Give an opportunity to dynamic tools to perform some simple testing. + SandboxBPF sandbox(nullptr); + SandboxBPF* sandbox_ptr = new SandboxBPF(nullptr); + delete sandbox_ptr; +} + +// This test should execute no matter whether we have kernel support. So, +// we make it a TEST() instead of a BPF_TEST(). +TEST(SandboxBPF, DISABLE_ON_TSAN(CallSupports)) { + // We check that we don't crash, but it's ok if the kernel doesn't + // support it. + bool seccomp_bpf_supported = SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::SINGLE_THREADED); + bool seccomp_bpf_tsync_supported = SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::MULTI_THREADED); + // We want to log whether or not seccomp BPF is actually supported + // since actual test coverage depends on it. + std::cout << "Seccomp BPF supported (single thread): " + << (seccomp_bpf_supported ? "true." : "false.") << "\n"; + std::cout << "Seccomp BPF supported (multi thread): " + << (seccomp_bpf_tsync_supported ? "true." : "false.") << "\n"; + std::cout << "Pointer size: " << sizeof(void*) << "\n"; +} + +SANDBOX_TEST(SandboxBPF, DISABLE_ON_TSAN(CallSupportsTwice)) { + bool single1 = SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::SINGLE_THREADED); + bool single2 = SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::SINGLE_THREADED); + ASSERT_EQ(single1, single2); + bool multi1 = SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::MULTI_THREADED); + bool multi2 = SandboxBPF::SupportsSeccompSandbox( + SandboxBPF::SeccompLevel::MULTI_THREADED); + ASSERT_EQ(multi1, multi2); + + // Multi threaded support implies single threaded support. + if (multi1) { + ASSERT_TRUE(single1); + } +} + +TEST(SandboxBPF, ProcTaskFdDescriptorGetsClosed) { + int pipe_fds[2]; + ASSERT_EQ(0, pipe(pipe_fds)); + base::ScopedFD read_end(pipe_fds[0]); + base::ScopedFD write_end(pipe_fds[1]); + + { + SandboxBPF sandbox(nullptr); + sandbox.SetProcFd(write_end.Pass()); + } + + ASSERT_EQ(0, fcntl(read_end.get(), F_SETFL, O_NONBLOCK)); + char c; + // Check that the sandbox closed the write_end (read will EOF instead of + // returning EWOULDBLOCK). + ASSERT_EQ(0, read(read_end.get(), &c, 1)); +} + +} // namespace +} // sandbox diff --git a/sandbox/linux/seccomp-bpf/syscall.cc b/sandbox/linux/seccomp-bpf/syscall.cc new file mode 100644 index 0000000000..bc6461f117 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/syscall.cc @@ -0,0 +1,421 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/syscall.h" + +#include <errno.h> +#include <stdint.h> + +#include "base/logging.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" + +namespace sandbox { + +namespace { + +#if defined(ARCH_CPU_X86_FAMILY) || defined(ARCH_CPU_ARM_FAMILY) || \ + defined(ARCH_CPU_MIPS_FAMILY) +// Number that's not currently used by any Linux kernel ABIs. +const int kInvalidSyscallNumber = 0x351d3; +#else +#error Unrecognized architecture +#endif + +asm(// We need to be able to tell the kernel exactly where we made a + // system call. The C++ compiler likes to sometimes clone or + // inline code, which would inadvertently end up duplicating + // the entry point. + // "gcc" can suppress code duplication with suitable function + // attributes, but "clang" doesn't have this ability. + // The "clang" developer mailing list suggested that the correct + // and portable solution is a file-scope assembly block. + // N.B. We do mark our code as a proper function so that backtraces + // work correctly. But we make absolutely no attempt to use the + // ABI's calling conventions for passing arguments. We will only + // ever be called from assembly code and thus can pick more + // suitable calling conventions. +#if defined(__i386__) + ".text\n" + ".align 16, 0x90\n" + ".type SyscallAsm, @function\n" + "SyscallAsm:.cfi_startproc\n" + // Check if "%eax" is negative. If so, do not attempt to make a + // system call. Instead, compute the return address that is visible + // to the kernel after we execute "int $0x80". This address can be + // used as a marker that BPF code inspects. + "test %eax, %eax\n" + "jge 1f\n" + // Always, make sure that our code is position-independent, or + // address space randomization might not work on i386. This means, + // we can't use "lea", but instead have to rely on "call/pop". + "call 0f; .cfi_adjust_cfa_offset 4\n" + "0:pop %eax; .cfi_adjust_cfa_offset -4\n" + "addl $2f-0b, %eax\n" + "ret\n" + // Save register that we don't want to clobber. On i386, we need to + // save relatively aggressively, as there are a couple or registers + // that are used internally (e.g. %ebx for position-independent + // code, and %ebp for the frame pointer), and as we need to keep at + // least a few registers available for the register allocator. + "1:push %esi; .cfi_adjust_cfa_offset 4; .cfi_rel_offset esi, 0\n" + "push %edi; .cfi_adjust_cfa_offset 4; .cfi_rel_offset edi, 0\n" + "push %ebx; .cfi_adjust_cfa_offset 4; .cfi_rel_offset ebx, 0\n" + "push %ebp; .cfi_adjust_cfa_offset 4; .cfi_rel_offset ebp, 0\n" + // Copy entries from the array holding the arguments into the + // correct CPU registers. + "movl 0(%edi), %ebx\n" + "movl 4(%edi), %ecx\n" + "movl 8(%edi), %edx\n" + "movl 12(%edi), %esi\n" + "movl 20(%edi), %ebp\n" + "movl 16(%edi), %edi\n" + // Enter the kernel. + "int $0x80\n" + // This is our "magic" return address that the BPF filter sees. + "2:" + // Restore any clobbered registers that we didn't declare to the + // compiler. + "pop %ebp; .cfi_restore ebp; .cfi_adjust_cfa_offset -4\n" + "pop %ebx; .cfi_restore ebx; .cfi_adjust_cfa_offset -4\n" + "pop %edi; .cfi_restore edi; .cfi_adjust_cfa_offset -4\n" + "pop %esi; .cfi_restore esi; .cfi_adjust_cfa_offset -4\n" + "ret\n" + ".cfi_endproc\n" + "9:.size SyscallAsm, 9b-SyscallAsm\n" +#elif defined(__x86_64__) + ".text\n" + ".align 16, 0x90\n" + ".type SyscallAsm, @function\n" + "SyscallAsm:.cfi_startproc\n" + // Check if "%rdi" is negative. If so, do not attempt to make a + // system call. Instead, compute the return address that is visible + // to the kernel after we execute "syscall". This address can be + // used as a marker that BPF code inspects. + "test %rdi, %rdi\n" + "jge 1f\n" + // Always make sure that our code is position-independent, or the + // linker will throw a hissy fit on x86-64. + "lea 2f(%rip), %rax\n" + "ret\n" + // Now we load the registers used to pass arguments to the system + // call: system call number in %rax, and arguments in %rdi, %rsi, + // %rdx, %r10, %r8, %r9. Note: These are all caller-save registers + // (only %rbx, %rbp, %rsp, and %r12-%r15 are callee-save), so no + // need to worry here about spilling registers or CFI directives. + "1:movq %rdi, %rax\n" + "movq 0(%rsi), %rdi\n" + "movq 16(%rsi), %rdx\n" + "movq 24(%rsi), %r10\n" + "movq 32(%rsi), %r8\n" + "movq 40(%rsi), %r9\n" + "movq 8(%rsi), %rsi\n" + // Enter the kernel. + "syscall\n" + // This is our "magic" return address that the BPF filter sees. + "2:ret\n" + ".cfi_endproc\n" + "9:.size SyscallAsm, 9b-SyscallAsm\n" +#elif defined(__arm__) + // Throughout this file, we use the same mode (ARM vs. thumb) + // that the C++ compiler uses. This means, when transfering control + // from C++ to assembly code, we do not need to switch modes (e.g. + // by using the "bx" instruction). It also means that our assembly + // code should not be invoked directly from code that lives in + // other compilation units, as we don't bother implementing thumb + // interworking. That's OK, as we don't make any of the assembly + // symbols public. They are all local to this file. + ".text\n" + ".align 2\n" + ".type SyscallAsm, %function\n" +#if defined(__thumb__) + ".thumb_func\n" +#else + ".arm\n" +#endif + "SyscallAsm:\n" +#if !defined(__native_client_nonsfi__) + // .fnstart and .fnend pseudo operations creates unwind table. + // It also creates a reference to the symbol __aeabi_unwind_cpp_pr0, which + // is not provided by PNaCl toolchain. Disable it. + ".fnstart\n" +#endif + "@ args = 0, pretend = 0, frame = 8\n" + "@ frame_needed = 1, uses_anonymous_args = 0\n" +#if defined(__thumb__) + ".cfi_startproc\n" + "push {r7, lr}\n" + ".save {r7, lr}\n" + ".cfi_offset 14, -4\n" + ".cfi_offset 7, -8\n" + ".cfi_def_cfa_offset 8\n" +#else + "stmfd sp!, {fp, lr}\n" + "add fp, sp, #4\n" +#endif + // Check if "r0" is negative. If so, do not attempt to make a + // system call. Instead, compute the return address that is visible + // to the kernel after we execute "swi 0". This address can be + // used as a marker that BPF code inspects. + "cmp r0, #0\n" + "bge 1f\n" + "adr r0, 2f\n" + "b 2f\n" + // We declared (almost) all clobbered registers to the compiler. On + // ARM there is no particular register pressure. So, we can go + // ahead and directly copy the entries from the arguments array + // into the appropriate CPU registers. + "1:ldr r5, [r6, #20]\n" + "ldr r4, [r6, #16]\n" + "ldr r3, [r6, #12]\n" + "ldr r2, [r6, #8]\n" + "ldr r1, [r6, #4]\n" + "mov r7, r0\n" + "ldr r0, [r6, #0]\n" + // Enter the kernel + "swi 0\n" +// Restore the frame pointer. Also restore the program counter from +// the link register; this makes us return to the caller. +#if defined(__thumb__) + "2:pop {r7, pc}\n" + ".cfi_endproc\n" +#else + "2:ldmfd sp!, {fp, pc}\n" +#endif +#if !defined(__native_client_nonsfi__) + // Do not use .fnstart and .fnend for PNaCl toolchain. See above comment, + // for more details. + ".fnend\n" +#endif + "9:.size SyscallAsm, 9b-SyscallAsm\n" +#elif defined(__mips__) + ".text\n" + ".align 4\n" + ".type SyscallAsm, @function\n" + "SyscallAsm:.ent SyscallAsm\n" + ".frame $sp, 40, $ra\n" + ".set push\n" + ".set noreorder\n" + "addiu $sp, $sp, -40\n" + "sw $ra, 36($sp)\n" + // Check if "v0" is negative. If so, do not attempt to make a + // system call. Instead, compute the return address that is visible + // to the kernel after we execute "syscall". This address can be + // used as a marker that BPF code inspects. + "bgez $v0, 1f\n" + " nop\n" + "la $v0, 2f\n" + "b 2f\n" + " nop\n" + // On MIPS first four arguments go to registers a0 - a3 and any + // argument after that goes to stack. We can go ahead and directly + // copy the entries from the arguments array into the appropriate + // CPU registers and on the stack. + "1:lw $a3, 28($a0)\n" + "lw $a2, 24($a0)\n" + "lw $a1, 20($a0)\n" + "lw $t0, 16($a0)\n" + "sw $a3, 28($sp)\n" + "sw $a2, 24($sp)\n" + "sw $a1, 20($sp)\n" + "sw $t0, 16($sp)\n" + "lw $a3, 12($a0)\n" + "lw $a2, 8($a0)\n" + "lw $a1, 4($a0)\n" + "lw $a0, 0($a0)\n" + // Enter the kernel + "syscall\n" + // This is our "magic" return address that the BPF filter sees. + // Restore the return address from the stack. + "2:lw $ra, 36($sp)\n" + "jr $ra\n" + " addiu $sp, $sp, 40\n" + ".set pop\n" + ".end SyscallAsm\n" + ".size SyscallAsm,.-SyscallAsm\n" +#elif defined(__aarch64__) + ".text\n" + ".align 2\n" + ".type SyscallAsm, %function\n" + "SyscallAsm:\n" + ".cfi_startproc\n" + "cmp x0, #0\n" + "b.ge 1f\n" + "adr x0,2f\n" + "b 2f\n" + "1:ldr x5, [x6, #40]\n" + "ldr x4, [x6, #32]\n" + "ldr x3, [x6, #24]\n" + "ldr x2, [x6, #16]\n" + "ldr x1, [x6, #8]\n" + "mov x8, x0\n" + "ldr x0, [x6, #0]\n" + // Enter the kernel + "svc 0\n" + "2:ret\n" + ".cfi_endproc\n" + ".size SyscallAsm, .-SyscallAsm\n" +#endif + ); // asm + +#if defined(__x86_64__) +extern "C" { +intptr_t SyscallAsm(intptr_t nr, const intptr_t args[6]); +} +#endif + +} // namespace + +intptr_t Syscall::InvalidCall() { + // Explicitly pass eight zero arguments just in case. + return Call(kInvalidSyscallNumber, 0, 0, 0, 0, 0, 0, 0, 0); +} + +intptr_t Syscall::Call(int nr, + intptr_t p0, + intptr_t p1, + intptr_t p2, + intptr_t p3, + intptr_t p4, + intptr_t p5, + intptr_t p6, + intptr_t p7) { + // We rely on "intptr_t" to be the exact size as a "void *". This is + // typically true, but just in case, we add a check. The language + // specification allows platforms some leeway in cases, where + // "sizeof(void *)" is not the same as "sizeof(void (*)())". We expect + // that this would only be an issue for IA64, which we are currently not + // planning on supporting. And it is even possible that this would work + // on IA64, but for lack of actual hardware, I cannot test. + static_assert(sizeof(void*) == sizeof(intptr_t), + "pointer types and intptr_t must be exactly the same size"); + + // TODO(nedeljko): Enable use of more than six parameters on architectures + // where that makes sense. +#if defined(__mips__) + const intptr_t args[8] = {p0, p1, p2, p3, p4, p5, p6, p7}; +#else + DCHECK_EQ(p6, 0) << " Support for syscalls with more than six arguments not " + "added for this architecture"; + DCHECK_EQ(p7, 0) << " Support for syscalls with more than six arguments not " + "added for this architecture"; + const intptr_t args[6] = {p0, p1, p2, p3, p4, p5}; +#endif // defined(__mips__) + +// Invoke our file-scope assembly code. The constraints have been picked +// carefully to match what the rest of the assembly code expects in input, +// output, and clobbered registers. +#if defined(__i386__) + intptr_t ret = nr; + asm volatile( + "call SyscallAsm\n" + // N.B. These are not the calling conventions normally used by the ABI. + : "=a"(ret) + : "0"(ret), "D"(args) + : "cc", "esp", "memory", "ecx", "edx"); +#elif defined(__x86_64__) + intptr_t ret = SyscallAsm(nr, args); +#elif defined(__arm__) + intptr_t ret; + { + register intptr_t inout __asm__("r0") = nr; + register const intptr_t* data __asm__("r6") = args; + asm volatile( + "bl SyscallAsm\n" + // N.B. These are not the calling conventions normally used by the ABI. + : "=r"(inout) + : "0"(inout), "r"(data) + : "cc", + "lr", + "memory", + "r1", + "r2", + "r3", + "r4", + "r5" +#if !defined(__thumb__) + // In thumb mode, we cannot use "r7" as a general purpose register, as + // it is our frame pointer. We have to manually manage and preserve + // it. + // In ARM mode, we have a dedicated frame pointer register and "r7" is + // thus available as a general purpose register. We don't preserve it, + // but instead mark it as clobbered. + , + "r7" +#endif // !defined(__thumb__) + ); + ret = inout; + } +#elif defined(__mips__) + int err_status; + intptr_t ret = Syscall::SandboxSyscallRaw(nr, args, &err_status); + + if (err_status) { + // On error, MIPS returns errno from syscall instead of -errno. + // The purpose of this negation is for SandboxSyscall() to behave + // more like it would on other architectures. + ret = -ret; + } +#elif defined(__aarch64__) + intptr_t ret; + { + register intptr_t inout __asm__("x0") = nr; + register const intptr_t* data __asm__("x6") = args; + asm volatile("bl SyscallAsm\n" + : "=r"(inout) + : "0"(inout), "r"(data) + : "memory", "x1", "x2", "x3", "x4", "x5", "x8", "x30"); + ret = inout; + } + +#else +#error "Unimplemented architecture" +#endif + return ret; +} + +void Syscall::PutValueInUcontext(intptr_t ret_val, ucontext_t* ctx) { +#if defined(__mips__) + // Mips ABI states that on error a3 CPU register has non zero value and if + // there is no error, it should be zero. + if (ret_val <= -1 && ret_val >= -4095) { + // |ret_val| followes the Syscall::Call() convention of being -errno on + // errors. In order to write correct value to return register this sign + // needs to be changed back. + ret_val = -ret_val; + SECCOMP_PARM4(ctx) = 1; + } else + SECCOMP_PARM4(ctx) = 0; +#endif + SECCOMP_RESULT(ctx) = static_cast<greg_t>(ret_val); +} + +#if defined(__mips__) +intptr_t Syscall::SandboxSyscallRaw(int nr, + const intptr_t* args, + intptr_t* err_ret) { + register intptr_t ret __asm__("v0") = nr; + // a3 register becomes non zero on error. + register intptr_t err_stat __asm__("a3") = 0; + { + register const intptr_t* data __asm__("a0") = args; + asm volatile( + "la $t9, SyscallAsm\n" + "jalr $t9\n" + " nop\n" + : "=r"(ret), "=r"(err_stat) + : "0"(ret), + "r"(data) + // a2 is in the clober list so inline assembly can not change its + // value. + : "memory", "ra", "t9", "a2"); + } + + // Set an error status so it can be used outside of this function + *err_ret = err_stat; + + return ret; +} +#endif // defined(__mips__) + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/syscall.h b/sandbox/linux/seccomp-bpf/syscall.h new file mode 100644 index 0000000000..ccfc88dcb3 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/syscall.h @@ -0,0 +1,166 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_SYSCALL_H__ +#define SANDBOX_LINUX_SECCOMP_BPF_SYSCALL_H__ + +#include <signal.h> +#include <stdint.h> + +#include "base/macros.h" +#include "sandbox/linux/system_headers/linux_signal.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// This purely static class can be used to perform system calls with some +// low-level control. +class SANDBOX_EXPORT Syscall { + public: + // InvalidCall() invokes Call() with a platform-appropriate syscall + // number that is guaranteed to not be implemented (i.e., normally + // returns -ENOSYS). + // This is primarily meant to be useful for writing sandbox policy + // unit tests. + static intptr_t InvalidCall(); + + // System calls can take up to six parameters (up to eight on some + // architectures). Traditionally, glibc + // implements this property by using variadic argument lists. This works, but + // confuses modern tools such as valgrind, because we are nominally passing + // uninitialized data whenever we call through this function and pass less + // than the full six arguments. + // So, instead, we use C++'s template system to achieve a very similar + // effect. C++ automatically sets the unused parameters to zero for us, and + // it also does the correct type expansion (e.g. from 32bit to 64bit) where + // necessary. + // We have to use C-style cast operators as we want to be able to accept both + // integer and pointer types. + template <class T0, + class T1, + class T2, + class T3, + class T4, + class T5, + class T6, + class T7> + static inline intptr_t + Call(int nr, T0 p0, T1 p1, T2 p2, T3 p3, T4 p4, T5 p5, T6 p6, T7 p7) { + return Call(nr, + (intptr_t)p0, + (intptr_t)p1, + (intptr_t)p2, + (intptr_t)p3, + (intptr_t)p4, + (intptr_t)p5, + (intptr_t)p6, + (intptr_t)p7); + } + + template <class T0, + class T1, + class T2, + class T3, + class T4, + class T5, + class T6> + static inline intptr_t + Call(int nr, T0 p0, T1 p1, T2 p2, T3 p3, T4 p4, T5 p5, T6 p6) { + return Call(nr, + (intptr_t)p0, + (intptr_t)p1, + (intptr_t)p2, + (intptr_t)p3, + (intptr_t)p4, + (intptr_t)p5, + (intptr_t)p6, + 0); + } + + template <class T0, class T1, class T2, class T3, class T4, class T5> + static inline intptr_t + Call(int nr, T0 p0, T1 p1, T2 p2, T3 p3, T4 p4, T5 p5) { + return Call(nr, + (intptr_t)p0, + (intptr_t)p1, + (intptr_t)p2, + (intptr_t)p3, + (intptr_t)p4, + (intptr_t)p5, + 0, + 0); + } + + template <class T0, class T1, class T2, class T3, class T4> + static inline intptr_t Call(int nr, T0 p0, T1 p1, T2 p2, T3 p3, T4 p4) { + return Call(nr, p0, p1, p2, p3, p4, 0, 0, 0); + } + + template <class T0, class T1, class T2, class T3> + static inline intptr_t Call(int nr, T0 p0, T1 p1, T2 p2, T3 p3) { + return Call(nr, p0, p1, p2, p3, 0, 0, 0, 0); + } + + template <class T0, class T1, class T2> + static inline intptr_t Call(int nr, T0 p0, T1 p1, T2 p2) { + return Call(nr, p0, p1, p2, 0, 0, 0, 0, 0); + } + + template <class T0, class T1> + static inline intptr_t Call(int nr, T0 p0, T1 p1) { + return Call(nr, p0, p1, 0, 0, 0, 0, 0, 0); + } + + template <class T0> + static inline intptr_t Call(int nr, T0 p0) { + return Call(nr, p0, 0, 0, 0, 0, 0, 0, 0); + } + + static inline intptr_t Call(int nr) { + return Call(nr, 0, 0, 0, 0, 0, 0, 0, 0); + } + + // Set the registers in |ctx| to match what they would be after a system call + // returning |ret_val|. |ret_val| must follow the Syscall::Call() convention + // of being -errno on errors. + static void PutValueInUcontext(intptr_t ret_val, ucontext_t* ctx); + + private: + // This performs system call |nr| with the arguments p0 to p7 from a constant + // userland address, which is for instance observable by seccomp-bpf filters. + // The constant userland address from which these system calls are made will + // be returned if |nr| is passed as -1. + // On error, this function will return a value between -1 and -4095 which + // should be interpreted as -errno. + static intptr_t Call(int nr, + intptr_t p0, + intptr_t p1, + intptr_t p2, + intptr_t p3, + intptr_t p4, + intptr_t p5, + intptr_t p6, + intptr_t p7); + +#if defined(__mips__) + // This function basically does on MIPS what SandboxSyscall() is doing on + // other architectures. However, because of specificity of MIPS regarding + // handling syscall errors, SandboxSyscall() is made as a wrapper for this + // function in order for SandboxSyscall() to behave more like on other + // architectures on places where return value from SandboxSyscall() is used + // directly (like in most tests). + // The syscall "nr" is called with arguments that are set in an array on which + // pointer "args" points to and an information weather there is an error or no + // is returned to SandboxSyscall() by err_stat. + static intptr_t SandboxSyscallRaw(int nr, + const intptr_t* args, + intptr_t* err_stat); +#endif // defined(__mips__) + + DISALLOW_IMPLICIT_CONSTRUCTORS(Syscall); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SECCOMP_BPF_SYSCALL_H__ diff --git a/sandbox/linux/seccomp-bpf/syscall_unittest.cc b/sandbox/linux/seccomp-bpf/syscall_unittest.cc new file mode 100644 index 0000000000..5fdee6c495 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/syscall_unittest.cc @@ -0,0 +1,240 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/syscall.h" + +#include <asm/unistd.h> +#include <fcntl.h> +#include <sys/mman.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <unistd.h> + +#include <vector> + +#include "base/posix/eintr_wrapper.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/bpf_dsl.h" +#include "sandbox/linux/bpf_dsl/policy.h" +#include "sandbox/linux/seccomp-bpf/bpf_tests.h" +#include "sandbox/linux/seccomp-bpf/sandbox_bpf.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +using sandbox::bpf_dsl::Allow; +using sandbox::bpf_dsl::ResultExpr; +using sandbox::bpf_dsl::Trap; + +namespace sandbox { + +namespace { + +// Different platforms use different symbols for the six-argument version +// of the mmap() system call. Test for the correct symbol at compile time. +#ifdef __NR_mmap2 +const int kMMapNr = __NR_mmap2; +#else +const int kMMapNr = __NR_mmap; +#endif + +TEST(Syscall, InvalidCallReturnsENOSYS) { + EXPECT_EQ(-ENOSYS, Syscall::InvalidCall()); +} + +TEST(Syscall, WellKnownEntryPoint) { +// Test that Syscall::Call(-1) is handled specially. Don't do this on ARM, +// where syscall(-1) crashes with SIGILL. Not running the test is fine, as we +// are still testing ARM code in the next set of tests. +#if !defined(__arm__) && !defined(__aarch64__) + EXPECT_NE(Syscall::Call(-1), syscall(-1)); +#endif + +// If possible, test that Syscall::Call(-1) returns the address right +// after +// a kernel entry point. +#if defined(__i386__) + EXPECT_EQ(0x80CDu, ((uint16_t*)Syscall::Call(-1))[-1]); // INT 0x80 +#elif defined(__x86_64__) + EXPECT_EQ(0x050Fu, ((uint16_t*)Syscall::Call(-1))[-1]); // SYSCALL +#elif defined(__arm__) +#if defined(__thumb__) + EXPECT_EQ(0xDF00u, ((uint16_t*)Syscall::Call(-1))[-1]); // SWI 0 +#else + EXPECT_EQ(0xEF000000u, ((uint32_t*)Syscall::Call(-1))[-1]); // SVC 0 +#endif +#elif defined(__mips__) + // Opcode for MIPS sycall is in the lower 16-bits + EXPECT_EQ(0x0cu, (((uint32_t*)Syscall::Call(-1))[-1]) & 0x0000FFFF); +#elif defined(__aarch64__) + EXPECT_EQ(0xD4000001u, ((uint32_t*)Syscall::Call(-1))[-1]); // SVC 0 +#else +#warning Incomplete test case; need port for target platform +#endif +} + +TEST(Syscall, TrivialSyscallNoArgs) { + // Test that we can do basic system calls + EXPECT_EQ(Syscall::Call(__NR_getpid), syscall(__NR_getpid)); +} + +TEST(Syscall, TrivialSyscallOneArg) { + int new_fd; + // Duplicate standard error and close it. + ASSERT_GE(new_fd = Syscall::Call(__NR_dup, 2), 0); + int close_return_value = IGNORE_EINTR(Syscall::Call(__NR_close, new_fd)); + ASSERT_EQ(close_return_value, 0); +} + +TEST(Syscall, TrivialFailingSyscall) { + errno = -42; + int ret = Syscall::Call(__NR_dup, -1); + ASSERT_EQ(-EBADF, ret); + // Verify that Syscall::Call does not touch errno. + ASSERT_EQ(-42, errno); +} + +// SIGSYS trap handler that will be called on __NR_uname. +intptr_t CopySyscallArgsToAux(const struct arch_seccomp_data& args, void* aux) { + // |aux| is our BPF_AUX pointer. + std::vector<uint64_t>* const seen_syscall_args = + static_cast<std::vector<uint64_t>*>(aux); + BPF_ASSERT(arraysize(args.args) == 6); + seen_syscall_args->assign(args.args, args.args + arraysize(args.args)); + return -ENOMEM; +} + +class CopyAllArgsOnUnamePolicy : public bpf_dsl::Policy { + public: + explicit CopyAllArgsOnUnamePolicy(std::vector<uint64_t>* aux) : aux_(aux) {} + ~CopyAllArgsOnUnamePolicy() override {} + + ResultExpr EvaluateSyscall(int sysno) const override { + DCHECK(SandboxBPF::IsValidSyscallNumber(sysno)); + if (sysno == __NR_uname) { + return Trap(CopySyscallArgsToAux, aux_); + } else { + return Allow(); + } + } + + private: + std::vector<uint64_t>* aux_; + + DISALLOW_COPY_AND_ASSIGN(CopyAllArgsOnUnamePolicy); +}; + +// We are testing Syscall::Call() by making use of a BPF filter that +// allows us +// to inspect the system call arguments that the kernel saw. +BPF_TEST(Syscall, + SyntheticSixArgs, + CopyAllArgsOnUnamePolicy, + std::vector<uint64_t> /* (*BPF_AUX) */) { + const int kExpectedValue = 42; + // In this test we only pass integers to the kernel. We might want to make + // additional tests to try other types. What we will see depends on + // implementation details of kernel BPF filters and we will need to document + // the expected behavior very clearly. + int syscall_args[6]; + for (size_t i = 0; i < arraysize(syscall_args); ++i) { + syscall_args[i] = kExpectedValue + i; + } + + // We could use pretty much any system call we don't need here. uname() is + // nice because it doesn't have any dangerous side effects. + BPF_ASSERT(Syscall::Call(__NR_uname, + syscall_args[0], + syscall_args[1], + syscall_args[2], + syscall_args[3], + syscall_args[4], + syscall_args[5]) == -ENOMEM); + + // We expect the trap handler to have copied the 6 arguments. + BPF_ASSERT(BPF_AUX->size() == 6); + + // Don't loop here so that we can see which argument does cause the failure + // easily from the failing line. + // uint64_t is the type passed to our SIGSYS handler. + BPF_ASSERT((*BPF_AUX)[0] == static_cast<uint64_t>(syscall_args[0])); + BPF_ASSERT((*BPF_AUX)[1] == static_cast<uint64_t>(syscall_args[1])); + BPF_ASSERT((*BPF_AUX)[2] == static_cast<uint64_t>(syscall_args[2])); + BPF_ASSERT((*BPF_AUX)[3] == static_cast<uint64_t>(syscall_args[3])); + BPF_ASSERT((*BPF_AUX)[4] == static_cast<uint64_t>(syscall_args[4])); + BPF_ASSERT((*BPF_AUX)[5] == static_cast<uint64_t>(syscall_args[5])); +} + +TEST(Syscall, ComplexSyscallSixArgs) { + int fd; + ASSERT_LE(0, + fd = Syscall::Call(__NR_openat, AT_FDCWD, "/dev/null", O_RDWR, 0L)); + + // Use mmap() to allocate some read-only memory + char* addr0; + ASSERT_NE( + (char*)NULL, + addr0 = reinterpret_cast<char*>(Syscall::Call(kMMapNr, + (void*)NULL, + 4096, + PROT_READ, + MAP_PRIVATE | MAP_ANONYMOUS, + fd, + 0L))); + + // Try to replace the existing mapping with a read-write mapping + char* addr1; + ASSERT_EQ(addr0, + addr1 = reinterpret_cast<char*>( + Syscall::Call(kMMapNr, + addr0, + 4096L, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, + fd, + 0L))); + ++*addr1; // This should not seg fault + + // Clean up + EXPECT_EQ(0, Syscall::Call(__NR_munmap, addr1, 4096L)); + EXPECT_EQ(0, IGNORE_EINTR(Syscall::Call(__NR_close, fd))); + + // Check that the offset argument (i.e. the sixth argument) is processed + // correctly. + ASSERT_GE( + fd = Syscall::Call(__NR_openat, AT_FDCWD, "/proc/self/exe", O_RDONLY, 0L), + 0); + char* addr2, *addr3; + ASSERT_NE((char*)NULL, + addr2 = reinterpret_cast<char*>(Syscall::Call( + kMMapNr, (void*)NULL, 8192L, PROT_READ, MAP_PRIVATE, fd, 0L))); + ASSERT_NE((char*)NULL, + addr3 = reinterpret_cast<char*>(Syscall::Call(kMMapNr, + (void*)NULL, + 4096L, + PROT_READ, + MAP_PRIVATE, + fd, +#if defined(__NR_mmap2) + 1L +#else + 4096L +#endif + ))); + EXPECT_EQ(0, memcmp(addr2 + 4096, addr3, 4096)); + + // Just to be absolutely on the safe side, also verify that the file + // contents matches what we are getting from a read() operation. + char buf[8192]; + EXPECT_EQ(8192, Syscall::Call(__NR_read, fd, buf, 8192L)); + EXPECT_EQ(0, memcmp(addr2, buf, 8192)); + + // Clean up + EXPECT_EQ(0, Syscall::Call(__NR_munmap, addr2, 8192L)); + EXPECT_EQ(0, Syscall::Call(__NR_munmap, addr3, 4096L)); + EXPECT_EQ(0, IGNORE_EINTR(Syscall::Call(__NR_close, fd))); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/trap.cc b/sandbox/linux/seccomp-bpf/trap.cc new file mode 100644 index 0000000000..8f559e53b1 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/trap.cc @@ -0,0 +1,390 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/trap.h" + +#include <errno.h> +#include <signal.h> +#include <string.h> +#include <sys/syscall.h> + +#include <algorithm> +#include <limits> + +#include "base/compiler_specific.h" +#include "base/logging.h" +#include "build/build_config.h" +#include "sandbox/linux/bpf_dsl/seccomp_macros.h" +#include "sandbox/linux/seccomp-bpf/die.h" +#include "sandbox/linux/seccomp-bpf/syscall.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/system_headers/linux_seccomp.h" +#include "sandbox/linux/system_headers/linux_signal.h" + +namespace { + +struct arch_sigsys { + void* ip; + int nr; + unsigned int arch; +}; + +const int kCapacityIncrement = 20; + +// Unsafe traps can only be turned on, if the user explicitly allowed them +// by setting the CHROME_SANDBOX_DEBUGGING environment variable. +const char kSandboxDebuggingEnv[] = "CHROME_SANDBOX_DEBUGGING"; + +// We need to tell whether we are performing a "normal" callback, or +// whether we were called recursively from within a UnsafeTrap() callback. +// This is a little tricky to do, because we need to somehow get access to +// per-thread data from within a signal context. Normal TLS storage is not +// safely accessible at this time. We could roll our own, but that involves +// a lot of complexity. Instead, we co-opt one bit in the signal mask. +// If BUS is blocked, we assume that we have been called recursively. +// There is a possibility for collision with other code that needs to do +// this, but in practice the risks are low. +// If SIGBUS turns out to be a problem, we could instead co-opt one of the +// realtime signals. There are plenty of them. Unfortunately, there is no +// way to mark a signal as allocated. So, the potential for collision is +// possibly even worse. +bool GetIsInSigHandler(const ucontext_t* ctx) { + // Note: on Android, sigismember does not take a pointer to const. + return sigismember(const_cast<sigset_t*>(&ctx->uc_sigmask), LINUX_SIGBUS); +} + +void SetIsInSigHandler() { + sigset_t mask; + if (sigemptyset(&mask) || sigaddset(&mask, LINUX_SIGBUS) || + sandbox::sys_sigprocmask(LINUX_SIG_BLOCK, &mask, NULL)) { + SANDBOX_DIE("Failed to block SIGBUS"); + } +} + +bool IsDefaultSignalAction(const struct sigaction& sa) { + if (sa.sa_flags & SA_SIGINFO || sa.sa_handler != SIG_DFL) { + return false; + } + return true; +} + +} // namespace + +namespace sandbox { + +Trap::Trap() + : trap_array_(NULL), + trap_array_size_(0), + trap_array_capacity_(0), + has_unsafe_traps_(false) { + // Set new SIGSYS handler + struct sigaction sa = {}; + // In some toolchain, sa_sigaction is not declared in struct sigaction. + // So, here cast the pointer to the sa_handler's type. This works because + // |sa_handler| and |sa_sigaction| shares the same memory. + sa.sa_handler = reinterpret_cast<void (*)(int)>(SigSysAction); + sa.sa_flags = LINUX_SA_SIGINFO | LINUX_SA_NODEFER; + struct sigaction old_sa = {}; + if (sys_sigaction(LINUX_SIGSYS, &sa, &old_sa) < 0) { + SANDBOX_DIE("Failed to configure SIGSYS handler"); + } + + if (!IsDefaultSignalAction(old_sa)) { + static const char kExistingSIGSYSMsg[] = + "Existing signal handler when trying to install SIGSYS. SIGSYS needs " + "to be reserved for seccomp-bpf."; + DLOG(FATAL) << kExistingSIGSYSMsg; + LOG(ERROR) << kExistingSIGSYSMsg; + } + + // Unmask SIGSYS + sigset_t mask; + if (sigemptyset(&mask) || sigaddset(&mask, LINUX_SIGSYS) || + sys_sigprocmask(LINUX_SIG_UNBLOCK, &mask, NULL)) { + SANDBOX_DIE("Failed to configure SIGSYS handler"); + } +} + +bpf_dsl::TrapRegistry* Trap::Registry() { + // Note: This class is not thread safe. It is the caller's responsibility + // to avoid race conditions. Normally, this is a non-issue as the sandbox + // can only be initialized if there are no other threads present. + // Also, this is not a normal singleton. Once created, the global trap + // object must never be destroyed again. + if (!global_trap_) { + global_trap_ = new Trap(); + if (!global_trap_) { + SANDBOX_DIE("Failed to allocate global trap handler"); + } + } + return global_trap_; +} + +void Trap::SigSysAction(int nr, LinuxSigInfo* info, void* void_context) { + if (info) { + MSAN_UNPOISON(info, sizeof(*info)); + } + + // Obtain the signal context. This, most notably, gives us access to + // all CPU registers at the time of the signal. + ucontext_t* ctx = reinterpret_cast<ucontext_t*>(void_context); + if (ctx) { + MSAN_UNPOISON(ctx, sizeof(*ctx)); + } + + if (!global_trap_) { + RAW_SANDBOX_DIE( + "This can't happen. Found no global singleton instance " + "for Trap() handling."); + } + global_trap_->SigSys(nr, info, ctx); +} + +void Trap::SigSys(int nr, LinuxSigInfo* info, ucontext_t* ctx) { + // Signal handlers should always preserve "errno". Otherwise, we could + // trigger really subtle bugs. + const int old_errno = errno; + + // Various sanity checks to make sure we actually received a signal + // triggered by a BPF filter. If something else triggered SIGSYS + // (e.g. kill()), there is really nothing we can do with this signal. + if (nr != LINUX_SIGSYS || info->si_code != SYS_SECCOMP || !ctx || + info->si_errno <= 0 || + static_cast<size_t>(info->si_errno) > trap_array_size_) { + // ATI drivers seem to send SIGSYS, so this cannot be FATAL. + // See crbug.com/178166. + // TODO(jln): add a DCHECK or move back to FATAL. + RAW_LOG(ERROR, "Unexpected SIGSYS received."); + errno = old_errno; + return; + } + + + // Obtain the siginfo information that is specific to SIGSYS. Unfortunately, + // most versions of glibc don't include this information in siginfo_t. So, + // we need to explicitly copy it into a arch_sigsys structure. + struct arch_sigsys sigsys; + memcpy(&sigsys, &info->_sifields, sizeof(sigsys)); + +#if defined(__mips__) + // When indirect syscall (syscall(__NR_foo, ...)) is made on Mips, the + // number in register SECCOMP_SYSCALL(ctx) is always __NR_syscall and the + // real number of a syscall (__NR_foo) is in SECCOMP_PARM1(ctx) + bool sigsys_nr_is_bad = sigsys.nr != static_cast<int>(SECCOMP_SYSCALL(ctx)) && + sigsys.nr != static_cast<int>(SECCOMP_PARM1(ctx)); +#else + bool sigsys_nr_is_bad = sigsys.nr != static_cast<int>(SECCOMP_SYSCALL(ctx)); +#endif + + // Some more sanity checks. + if (sigsys.ip != reinterpret_cast<void*>(SECCOMP_IP(ctx)) || + sigsys_nr_is_bad || sigsys.arch != SECCOMP_ARCH) { + // TODO(markus): + // SANDBOX_DIE() can call LOG(FATAL). This is not normally async-signal + // safe and can lead to bugs. We should eventually implement a different + // logging and reporting mechanism that is safe to be called from + // the sigSys() handler. + RAW_SANDBOX_DIE("Sanity checks are failing after receiving SIGSYS."); + } + + intptr_t rc; + if (has_unsafe_traps_ && GetIsInSigHandler(ctx)) { + errno = old_errno; + if (sigsys.nr == __NR_clone) { + RAW_SANDBOX_DIE("Cannot call clone() from an UnsafeTrap() handler."); + } +#if defined(__mips__) + // Mips supports up to eight arguments for syscall. + // However, seccomp bpf can filter only up to six arguments, so using eight + // arguments has sense only when using UnsafeTrap() handler. + rc = Syscall::Call(SECCOMP_SYSCALL(ctx), + SECCOMP_PARM1(ctx), + SECCOMP_PARM2(ctx), + SECCOMP_PARM3(ctx), + SECCOMP_PARM4(ctx), + SECCOMP_PARM5(ctx), + SECCOMP_PARM6(ctx), + SECCOMP_PARM7(ctx), + SECCOMP_PARM8(ctx)); +#else + rc = Syscall::Call(SECCOMP_SYSCALL(ctx), + SECCOMP_PARM1(ctx), + SECCOMP_PARM2(ctx), + SECCOMP_PARM3(ctx), + SECCOMP_PARM4(ctx), + SECCOMP_PARM5(ctx), + SECCOMP_PARM6(ctx)); +#endif // defined(__mips__) + } else { + const TrapKey& trap = trap_array_[info->si_errno - 1]; + if (!trap.safe) { + SetIsInSigHandler(); + } + + // Copy the seccomp-specific data into a arch_seccomp_data structure. This + // is what we are showing to TrapFnc callbacks that the system call + // evaluator registered with the sandbox. + struct arch_seccomp_data data = { + static_cast<int>(SECCOMP_SYSCALL(ctx)), + SECCOMP_ARCH, + reinterpret_cast<uint64_t>(sigsys.ip), + {static_cast<uint64_t>(SECCOMP_PARM1(ctx)), + static_cast<uint64_t>(SECCOMP_PARM2(ctx)), + static_cast<uint64_t>(SECCOMP_PARM3(ctx)), + static_cast<uint64_t>(SECCOMP_PARM4(ctx)), + static_cast<uint64_t>(SECCOMP_PARM5(ctx)), + static_cast<uint64_t>(SECCOMP_PARM6(ctx))}}; + + // Now call the TrapFnc callback associated with this particular instance + // of SECCOMP_RET_TRAP. + rc = trap.fnc(data, const_cast<void*>(trap.aux)); + } + + // Update the CPU register that stores the return code of the system call + // that we just handled, and restore "errno" to the value that it had + // before entering the signal handler. + Syscall::PutValueInUcontext(rc, ctx); + errno = old_errno; + + return; +} + +bool Trap::TrapKey::operator<(const TrapKey& o) const { + if (fnc != o.fnc) { + return fnc < o.fnc; + } else if (aux != o.aux) { + return aux < o.aux; + } else { + return safe < o.safe; + } +} + +uint16_t Trap::Add(TrapFnc fnc, const void* aux, bool safe) { + if (!safe && !SandboxDebuggingAllowedByUser()) { + // Unless the user set the CHROME_SANDBOX_DEBUGGING environment variable, + // we never return an ErrorCode that is marked as "unsafe". This also + // means, the BPF compiler will never emit code that allow unsafe system + // calls to by-pass the filter (because they use the magic return address + // from Syscall::Call(-1)). + + // This SANDBOX_DIE() can optionally be removed. It won't break security, + // but it might make error messages from the BPF compiler a little harder + // to understand. Removing the SANDBOX_DIE() allows callers to easily check + // whether unsafe traps are supported (by checking whether the returned + // ErrorCode is ET_INVALID). + SANDBOX_DIE( + "Cannot use unsafe traps unless CHROME_SANDBOX_DEBUGGING " + "is enabled"); + + return 0; + } + + // Each unique pair of TrapFnc and auxiliary data make up a distinct instance + // of a SECCOMP_RET_TRAP. + TrapKey key(fnc, aux, safe); + + // We return unique identifiers together with SECCOMP_RET_TRAP. This allows + // us to associate trap with the appropriate handler. The kernel allows us + // identifiers in the range from 0 to SECCOMP_RET_DATA (0xFFFF). We want to + // avoid 0, as it could be confused for a trap without any specific id. + // The nice thing about sequentially numbered identifiers is that we can also + // trivially look them up from our signal handler without making any system + // calls that might be async-signal-unsafe. + // In order to do so, we store all of our traps in a C-style trap_array_. + + TrapIds::const_iterator iter = trap_ids_.find(key); + if (iter != trap_ids_.end()) { + // We have seen this pair before. Return the same id that we assigned + // earlier. + return iter->second; + } + + // This is a new pair. Remember it and assign a new id. + if (trap_array_size_ >= SECCOMP_RET_DATA /* 0xFFFF */ || + trap_array_size_ >= std::numeric_limits<uint16_t>::max()) { + // In practice, this is pretty much impossible to trigger, as there + // are other kernel limitations that restrict overall BPF program sizes. + SANDBOX_DIE("Too many SECCOMP_RET_TRAP callback instances"); + } + + // Our callers ensure that there are no other threads accessing trap_array_ + // concurrently (typically this is done by ensuring that we are single- + // threaded while the sandbox is being set up). But we nonetheless are + // modifying a live data structure that could be accessed any time a + // system call is made; as system calls could be triggering SIGSYS. + // So, we have to be extra careful that we update trap_array_ atomically. + // In particular, this means we shouldn't be using realloc() to resize it. + // Instead, we allocate a new array, copy the values, and then switch the + // pointer. We only really care about the pointer being updated atomically + // and the data that is pointed to being valid, as these are the only + // values accessed from the signal handler. It is OK if trap_array_size_ + // is inconsistent with the pointer, as it is monotonously increasing. + // Also, we only care about compiler barriers, as the signal handler is + // triggered synchronously from a system call. We don't have to protect + // against issues with the memory model or with completely asynchronous + // events. + if (trap_array_size_ >= trap_array_capacity_) { + trap_array_capacity_ += kCapacityIncrement; + TrapKey* old_trap_array = trap_array_; + TrapKey* new_trap_array = new TrapKey[trap_array_capacity_]; + std::copy_n(old_trap_array, trap_array_size_, new_trap_array); + + // Language specs are unclear on whether the compiler is allowed to move + // the "delete[]" above our preceding assignments and/or memory moves, + // iff the compiler believes that "delete[]" doesn't have any other + // global side-effects. + // We insert optimization barriers to prevent this from happening. + // The first barrier is probably not needed, but better be explicit in + // what we want to tell the compiler. + // The clang developer mailing list couldn't answer whether this is a + // legitimate worry; but they at least thought that the barrier is + // sufficient to prevent the (so far hypothetical) problem of re-ordering + // of instructions by the compiler. + // + // TODO(mdempsky): Try to clean this up using base/atomicops or C++11 + // atomics; see crbug.com/414363. + asm volatile("" : "=r"(new_trap_array) : "0"(new_trap_array) : "memory"); + trap_array_ = new_trap_array; + asm volatile("" : "=r"(trap_array_) : "0"(trap_array_) : "memory"); + + delete[] old_trap_array; + } + + uint16_t id = trap_array_size_ + 1; + trap_ids_[key] = id; + trap_array_[trap_array_size_] = key; + trap_array_size_++; + return id; +} + +bool Trap::SandboxDebuggingAllowedByUser() { + const char* debug_flag = getenv(kSandboxDebuggingEnv); + return debug_flag && *debug_flag; +} + +bool Trap::EnableUnsafeTraps() { + if (!has_unsafe_traps_) { + // Unsafe traps are a one-way fuse. Once enabled, they can never be turned + // off again. + // We only allow enabling unsafe traps, if the user explicitly set an + // appropriate environment variable. This prevents bugs that accidentally + // disable all sandboxing for all users. + if (SandboxDebuggingAllowedByUser()) { + // We only ever print this message once, when we enable unsafe traps the + // first time. + SANDBOX_INFO("WARNING! Disabling sandbox for debugging purposes"); + has_unsafe_traps_ = true; + } else { + SANDBOX_INFO( + "Cannot disable sandbox and use unsafe traps unless " + "CHROME_SANDBOX_DEBUGGING is turned on first"); + } + } + // Returns the, possibly updated, value of has_unsafe_traps_. + return has_unsafe_traps_; +} + +Trap* Trap::global_trap_; + +} // namespace sandbox diff --git a/sandbox/linux/seccomp-bpf/trap.h b/sandbox/linux/seccomp-bpf/trap.h new file mode 100644 index 0000000000..50ac3fd1c3 --- /dev/null +++ b/sandbox/linux/seccomp-bpf/trap.h @@ -0,0 +1,85 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SECCOMP_BPF_TRAP_H__ +#define SANDBOX_LINUX_SECCOMP_BPF_TRAP_H__ + +#include <stdint.h> + +#include <map> + +#include "base/macros.h" +#include "sandbox/linux/bpf_dsl/trap_registry.h" +#include "sandbox/linux/system_headers/linux_signal.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// The Trap class allows a BPF filter program to branch out to user space by +// raising a SIGSYS signal. +// N.B.: This class does not perform any synchronization operations. If +// modifications are made to any of the traps, it is the caller's +// responsibility to ensure that this happens in a thread-safe fashion. +// Preferably, that means that no other threads should be running at that +// time. For the purposes of our sandbox, this assertion should always be +// true. Threads are incompatible with the seccomp sandbox anyway. +class SANDBOX_EXPORT Trap : public bpf_dsl::TrapRegistry { + public: + uint16_t Add(TrapFnc fnc, const void* aux, bool safe) override; + + bool EnableUnsafeTraps() override; + + // Registry returns the trap registry used by Trap's SIGSYS handler, + // creating it if necessary. + static bpf_dsl::TrapRegistry* Registry(); + + // SandboxDebuggingAllowedByUser returns whether the + // "CHROME_SANDBOX_DEBUGGING" environment variable is set. + static bool SandboxDebuggingAllowedByUser(); + + private: + struct TrapKey { + TrapKey() : fnc(NULL), aux(NULL), safe(false) {} + TrapKey(TrapFnc f, const void* a, bool s) : fnc(f), aux(a), safe(s) {} + TrapFnc fnc; + const void* aux; + bool safe; + bool operator<(const TrapKey&) const; + }; + typedef std::map<TrapKey, uint16_t> TrapIds; + + // Our constructor is private. A shared global instance is created + // automatically as needed. + Trap(); + + // The destructor is unimplemented as destroying this object would + // break subsequent system calls that trigger a SIGSYS. + ~Trap() = delete; + + static void SigSysAction(int nr, LinuxSigInfo* info, void* void_context); + + // Make sure that SigSys is not inlined in order to get slightly better crash + // dumps. + void SigSys(int nr, LinuxSigInfo* info, ucontext_t* ctx) + __attribute__((noinline)); + // We have a global singleton that handles all of our SIGSYS traps. This + // variable must never be deallocated after it has been set up initially, as + // there is no way to reset in-kernel BPF filters that generate SIGSYS + // events. + static Trap* global_trap_; + + TrapIds trap_ids_; // Maps from TrapKeys to numeric ids + TrapKey* trap_array_; // Array of TrapKeys indexed by ids + size_t trap_array_size_; // Currently used size of array + size_t trap_array_capacity_; // Currently allocated capacity of array + bool has_unsafe_traps_; // Whether unsafe traps have been enabled + + // Copying and assigning is unimplemented. It doesn't make sense for a + // singleton. + DISALLOW_COPY_AND_ASSIGN(Trap); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SECCOMP_BPF_TRAP_H__ diff --git a/sandbox/linux/seccomp-bpf/trap_unittest.cc b/sandbox/linux/seccomp-bpf/trap_unittest.cc new file mode 100644 index 0000000000..99f94bfb3a --- /dev/null +++ b/sandbox/linux/seccomp-bpf/trap_unittest.cc @@ -0,0 +1,28 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/seccomp-bpf/trap.h" + +#include <signal.h> + +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { +namespace { + +SANDBOX_TEST_ALLOW_NOISE(Trap, SigSysAction) { + // This creates a global Trap instance, and registers the signal handler + // (Trap::SigSysAction). + Trap::Registry(); + + // Send SIGSYS to self. If signal handler (SigSysAction) is not registered, + // the process will be terminated with status code -SIGSYS. + // Note that, SigSysAction handler would output an error message + // "Unexpected SIGSYS received." so it is necessary to allow the noise. + raise(SIGSYS); +} + +} // namespace +} // namespace sandbox diff --git a/sandbox/linux/services/DEPS b/sandbox/linux/services/DEPS new file mode 100644 index 0000000000..70d9b18aa1 --- /dev/null +++ b/sandbox/linux/services/DEPS @@ -0,0 +1,3 @@ +include_rules = [ + "+sandbox/linux/system_headers", +] diff --git a/sandbox/linux/services/credentials.cc b/sandbox/linux/services/credentials.cc new file mode 100644 index 0000000000..35bb4dcbd7 --- /dev/null +++ b/sandbox/linux/services/credentials.cc @@ -0,0 +1,299 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/credentials.h" + +#include <errno.h> +#include <signal.h> +#include <stdint.h> +#include <stdio.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include "base/bind.h" +#include "base/files/file_path.h" +#include "base/files/file_util.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "base/process/launch.h" +#include "base/template_util.h" +#include "base/third_party/valgrind/valgrind.h" +#include "build/build_config.h" +#include "sandbox/linux/services/namespace_utils.h" +#include "sandbox/linux/services/proc_util.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/services/thread_helpers.h" +#include "sandbox/linux/system_headers/capability.h" +#include "sandbox/linux/system_headers/linux_signal.h" + +namespace sandbox { + +namespace { + +bool IsRunningOnValgrind() { return RUNNING_ON_VALGRIND; } + +// Checks that the set of RES-uids and the set of RES-gids have +// one element each and return that element in |resuid| and |resgid| +// respectively. It's ok to pass NULL as one or both of the ids. +bool GetRESIds(uid_t* resuid, gid_t* resgid) { + uid_t ruid, euid, suid; + gid_t rgid, egid, sgid; + PCHECK(sys_getresuid(&ruid, &euid, &suid) == 0); + PCHECK(sys_getresgid(&rgid, &egid, &sgid) == 0); + const bool uids_are_equal = (ruid == euid) && (ruid == suid); + const bool gids_are_equal = (rgid == egid) && (rgid == sgid); + if (!uids_are_equal || !gids_are_equal) return false; + if (resuid) *resuid = euid; + if (resgid) *resgid = egid; + return true; +} + +const int kExitSuccess = 0; + +int ChrootToSelfFdinfo(void*) { + RAW_CHECK(sys_chroot("/proc/self/fdinfo/") == 0); + + // CWD is essentially an implicit file descriptor, so be careful to not + // leave it behind. + RAW_CHECK(chdir("/") == 0); + _exit(kExitSuccess); +} + +// chroot() to an empty dir that is "safe". To be safe, it must not contain +// any subdirectory (chroot-ing there would allow a chroot escape) and it must +// be impossible to create an empty directory there. +// We achieve this by doing the following: +// 1. We create a new process sharing file system information. +// 2. In the child, we chroot to /proc/self/fdinfo/ +// This is already "safe", since fdinfo/ does not contain another directory and +// one cannot create another directory there. +// 3. The process dies +// After (3) happens, the directory is not available anymore in /proc. +bool ChrootToSafeEmptyDir() { + // We need to chroot to a fdinfo that is unique to a process and have that + // process die. + // 1. We don't want to simply fork() because duplicating the page tables is + // slow with a big address space. + // 2. We do not use a regular thread (that would unshare CLONE_FILES) because + // when we are in a PID namespace, we cannot easily get a handle to the + // /proc/tid directory for the thread (since /proc may not be aware of the + // PID namespace). With a process, we can just use /proc/self. + pid_t pid = -1; + char stack_buf[PTHREAD_STACK_MIN]; +#if defined(ARCH_CPU_X86_FAMILY) || defined(ARCH_CPU_ARM_FAMILY) || \ + defined(ARCH_CPU_MIPS64_FAMILY) || defined(ARCH_CPU_MIPS_FAMILY) + // The stack grows downward. + void* stack = stack_buf + sizeof(stack_buf); +#else +#error "Unsupported architecture" +#endif + + pid = clone(ChrootToSelfFdinfo, stack, + CLONE_VM | CLONE_VFORK | CLONE_FS | LINUX_SIGCHLD, nullptr, + nullptr, nullptr, nullptr); + PCHECK(pid != -1); + + int status = -1; + PCHECK(HANDLE_EINTR(waitpid(pid, &status, 0)) == pid); + + return WIFEXITED(status) && WEXITSTATUS(status) == kExitSuccess; +} + +// CHECK() that an attempt to move to a new user namespace raised an expected +// errno. +void CheckCloneNewUserErrno(int error) { + // EPERM can happen if already in a chroot. EUSERS if too many nested + // namespaces are used. EINVAL for kernels that don't support the feature. + // Valgrind will ENOSYS unshare(). + PCHECK(error == EPERM || error == EUSERS || error == EINVAL || + error == ENOSYS); +} + +// Converts a Capability to the corresponding Linux CAP_XXX value. +int CapabilityToKernelValue(Credentials::Capability cap) { + switch (cap) { + case Credentials::Capability::SYS_CHROOT: + return CAP_SYS_CHROOT; + case Credentials::Capability::SYS_ADMIN: + return CAP_SYS_ADMIN; + } + + LOG(FATAL) << "Invalid Capability: " << static_cast<int>(cap); + return 0; +} + +} // namespace. + +// static +bool Credentials::DropAllCapabilities(int proc_fd) { + if (!SetCapabilities(proc_fd, std::vector<Capability>())) { + return false; + } + + CHECK(!HasAnyCapability()); + return true; +} + +// static +bool Credentials::DropAllCapabilities() { + base::ScopedFD proc_fd(ProcUtil::OpenProc()); + return Credentials::DropAllCapabilities(proc_fd.get()); +} + +// static +bool Credentials::DropAllCapabilitiesOnCurrentThread() { + return SetCapabilitiesOnCurrentThread(std::vector<Capability>()); +} + +// static +bool Credentials::SetCapabilitiesOnCurrentThread( + const std::vector<Capability>& caps) { + struct cap_hdr hdr = {}; + hdr.version = _LINUX_CAPABILITY_VERSION_3; + struct cap_data data[_LINUX_CAPABILITY_U32S_3] = {{}}; + + // Initially, cap has no capability flags set. Enable the effective and + // permitted flags only for the requested capabilities. + for (const Capability cap : caps) { + const int cap_num = CapabilityToKernelValue(cap); + const size_t index = CAP_TO_INDEX(cap_num); + const uint32_t mask = CAP_TO_MASK(cap_num); + data[index].effective |= mask; + data[index].permitted |= mask; + } + + return sys_capset(&hdr, data) == 0; +} + +// static +bool Credentials::SetCapabilities(int proc_fd, + const std::vector<Capability>& caps) { + DCHECK_LE(0, proc_fd); + +#if !defined(THREAD_SANITIZER) + // With TSAN, accept to break the security model as it is a testing + // configuration. + CHECK(ThreadHelpers::IsSingleThreaded(proc_fd)); +#endif + + return SetCapabilitiesOnCurrentThread(caps); +} + +bool Credentials::HasAnyCapability() { + struct cap_hdr hdr = {}; + hdr.version = _LINUX_CAPABILITY_VERSION_3; + struct cap_data data[_LINUX_CAPABILITY_U32S_3] = {{}}; + + PCHECK(sys_capget(&hdr, data) == 0); + + for (size_t i = 0; i < arraysize(data); ++i) { + if (data[i].effective || data[i].permitted || data[i].inheritable) { + return true; + } + } + + return false; +} + +bool Credentials::HasCapability(Capability cap) { + struct cap_hdr hdr = {}; + hdr.version = _LINUX_CAPABILITY_VERSION_3; + struct cap_data data[_LINUX_CAPABILITY_U32S_3] = {{}}; + + PCHECK(sys_capget(&hdr, data) == 0); + + const int cap_num = CapabilityToKernelValue(cap); + const size_t index = CAP_TO_INDEX(cap_num); + const uint32_t mask = CAP_TO_MASK(cap_num); + + return (data[index].effective | data[index].permitted | + data[index].inheritable) & + mask; +} + +// static +bool Credentials::CanCreateProcessInNewUserNS() { + // Valgrind will let clone(2) pass-through, but doesn't support unshare(), + // so always consider UserNS unsupported there. + if (IsRunningOnValgrind()) { + return false; + } + +#if defined(THREAD_SANITIZER) + // With TSAN, processes will always have threads running and can never + // enter a new user namespace with MoveToNewUserNS(). + return false; +#endif + + // This is roughly a fork(). + const pid_t pid = sys_clone(CLONE_NEWUSER | SIGCHLD, 0, 0, 0, 0); + + if (pid == -1) { + CheckCloneNewUserErrno(errno); + return false; + } + + // The parent process could have had threads. In the child, these threads + // have disappeared. Make sure to not do anything in the child, as this is a + // fragile execution environment. + if (pid == 0) { + _exit(kExitSuccess); + } + + // Always reap the child. + int status = -1; + PCHECK(HANDLE_EINTR(waitpid(pid, &status, 0)) == pid); + CHECK(WIFEXITED(status)); + CHECK_EQ(kExitSuccess, WEXITSTATUS(status)); + + // clone(2) succeeded, we can use CLONE_NEWUSER. + return true; +} + +bool Credentials::MoveToNewUserNS() { + uid_t uid; + gid_t gid; + if (!GetRESIds(&uid, &gid)) { + // If all the uids (or gids) are not equal to each other, the security + // model will most likely confuse the caller, abort. + DVLOG(1) << "uids or gids differ!"; + return false; + } + int ret = sys_unshare(CLONE_NEWUSER); + if (ret) { + const int unshare_errno = errno; + VLOG(1) << "Looks like unprivileged CLONE_NEWUSER may not be available " + << "on this kernel."; + CheckCloneNewUserErrno(unshare_errno); + return false; + } + + if (NamespaceUtils::KernelSupportsDenySetgroups()) { + PCHECK(NamespaceUtils::DenySetgroups()); + } + + // The current {r,e,s}{u,g}id is now an overflow id (c.f. + // /proc/sys/kernel/overflowuid). Setup the uid and gid maps. + DCHECK(GetRESIds(NULL, NULL)); + const char kGidMapFile[] = "/proc/self/gid_map"; + const char kUidMapFile[] = "/proc/self/uid_map"; + PCHECK(NamespaceUtils::WriteToIdMapFile(kGidMapFile, gid)); + PCHECK(NamespaceUtils::WriteToIdMapFile(kUidMapFile, uid)); + DCHECK(GetRESIds(NULL, NULL)); + return true; +} + +bool Credentials::DropFileSystemAccess(int proc_fd) { + CHECK_LE(0, proc_fd); + + CHECK(ChrootToSafeEmptyDir()); + CHECK(!base::DirectoryExists(base::FilePath("/proc"))); + CHECK(!ProcUtil::HasOpenDirectory(proc_fd)); + // We never let this function fail. + return true; +} + +} // namespace sandbox. diff --git a/sandbox/linux/services/credentials.h b/sandbox/linux/services/credentials.h new file mode 100644 index 0000000000..0001dc7328 --- /dev/null +++ b/sandbox/linux/services/credentials.h @@ -0,0 +1,104 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_CREDENTIALS_H_ +#define SANDBOX_LINUX_SERVICES_CREDENTIALS_H_ + +#include "build/build_config.h" +// Link errors are tedious to track, raise a compile-time error instead. +#if defined(OS_ANDROID) +#error "Android is not supported." +#endif // defined(OS_ANDROID). + +#include <string> +#include <vector> + +#include "base/compiler_specific.h" +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "sandbox/linux/system_headers/capability.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// This class should be used to manipulate the current process' credentials. +// It is currently a stub used to manipulate POSIX.1e capabilities as +// implemented by the Linux kernel. +class SANDBOX_EXPORT Credentials { + public: + // For brevity, we only expose enums for the subset of capabilities we use. + // This can be expanded as the need arises. + enum class Capability { + SYS_CHROOT, + SYS_ADMIN, + }; + + // Drop all capabilities in the effective, inheritable and permitted sets for + // the current thread. For security reasons, since capabilities are + // per-thread, the caller is responsible for ensuring it is single-threaded + // when calling this API. + // |proc_fd| must be a file descriptor to /proc/ and remains owned by + // the caller. + static bool DropAllCapabilities(int proc_fd) WARN_UNUSED_RESULT; + // A similar API which assumes that it can open /proc/self/ by itself. + static bool DropAllCapabilities() WARN_UNUSED_RESULT; + // Sets the effective and permitted capability sets for the current thread to + // the list of capabiltiies in |caps|. All other capability flags are cleared. + static bool SetCapabilities(int proc_fd, + const std::vector<Capability>& caps) + WARN_UNUSED_RESULT; + + // Versions of the above functions which do not check that the process is + // single-threaded. After calling these functions, capabilities of other + // threads will not be changed. This is dangerous, do not use unless you nkow + // what you are doing. + static bool DropAllCapabilitiesOnCurrentThread() WARN_UNUSED_RESULT; + static bool SetCapabilitiesOnCurrentThread( + const std::vector<Capability>& caps) WARN_UNUSED_RESULT; + + // Returns true if the current thread has either the effective, permitted, or + // inheritable flag set for the given capability. + static bool HasCapability(Capability cap); + + // Return true iff there is any capability in any of the capabilities sets + // of the current thread. + static bool HasAnyCapability(); + + // Returns whether the kernel supports CLONE_NEWUSER and whether it would be + // possible to immediately move to a new user namespace. There is no point + // in using this method right before calling MoveToNewUserNS(), simply call + // MoveToNewUserNS() immediately. This method is only useful to test the + // ability to move to a user namespace ahead of time. + static bool CanCreateProcessInNewUserNS(); + + // Move the current process to a new "user namespace" as supported by Linux + // 3.8+ (CLONE_NEWUSER). + // The uid map will be set-up so that the perceived uid and gid will not + // change. + // If this call succeeds, the current process will be granted a full set of + // capabilities in the new namespace. + // This will fail if the process is not mono-threaded. + static bool MoveToNewUserNS() WARN_UNUSED_RESULT; + + // Remove the ability of the process to access the file system. File + // descriptors which are already open prior to calling this API remain + // available. + // The implementation currently uses chroot(2) and requires CAP_SYS_CHROOT. + // CAP_SYS_CHROOT can be acquired by using the MoveToNewUserNS() API. + // |proc_fd| must be a file descriptor to /proc/ and must be the only open + // directory file descriptor of the process. + // + // CRITICAL: + // - the caller must close |proc_fd| eventually or access to the file + // system can be recovered. + // - DropAllCapabilities() must be called to prevent escapes. + static bool DropFileSystemAccess(int proc_fd) WARN_UNUSED_RESULT; + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(Credentials); +}; + +} // namespace sandbox. + +#endif // SANDBOX_LINUX_SERVICES_CREDENTIALS_H_ diff --git a/sandbox/linux/services/credentials_unittest.cc b/sandbox/linux/services/credentials_unittest.cc new file mode 100644 index 0000000000..6b93c86c3e --- /dev/null +++ b/sandbox/linux/services/credentials_unittest.cc @@ -0,0 +1,242 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/credentials.h" + +#include <errno.h> +#include <fcntl.h> +#include <stdio.h> +#include <sys/capability.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> + +#include <vector> + +#include "base/files/file_path.h" +#include "base/files/file_util.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "sandbox/linux/services/proc_util.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/system_headers/capability.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +struct CapFreeDeleter { + inline void operator()(cap_t cap) const { + int ret = cap_free(cap); + CHECK_EQ(0, ret); + } +}; + +// Wrapper to manage libcap2's cap_t type. +typedef scoped_ptr<typeof(*((cap_t)0)), CapFreeDeleter> ScopedCap; + +bool WorkingDirectoryIsRoot() { + char current_dir[PATH_MAX]; + char* cwd = getcwd(current_dir, sizeof(current_dir)); + PCHECK(cwd); + if (strcmp("/", cwd)) return false; + + // The current directory is the root. Add a few paranoid checks. + struct stat current; + CHECK_EQ(0, stat(".", ¤t)); + struct stat parrent; + CHECK_EQ(0, stat("..", &parrent)); + CHECK_EQ(current.st_dev, parrent.st_dev); + CHECK_EQ(current.st_ino, parrent.st_ino); + CHECK_EQ(current.st_mode, parrent.st_mode); + CHECK_EQ(current.st_uid, parrent.st_uid); + CHECK_EQ(current.st_gid, parrent.st_gid); + return true; +} + +SANDBOX_TEST(Credentials, DropAllCaps) { + CHECK(Credentials::DropAllCapabilities()); + CHECK(!Credentials::HasAnyCapability()); +} + +SANDBOX_TEST(Credentials, MoveToNewUserNS) { + CHECK(Credentials::DropAllCapabilities()); + bool moved_to_new_ns = Credentials::MoveToNewUserNS(); + fprintf(stdout, + "Unprivileged CLONE_NEWUSER supported: %s\n", + moved_to_new_ns ? "true." : "false."); + fflush(stdout); + if (!moved_to_new_ns) { + fprintf(stdout, "This kernel does not support unprivileged namespaces. " + "USERNS tests will succeed without running.\n"); + fflush(stdout); + return; + } + CHECK(Credentials::HasAnyCapability()); + CHECK(Credentials::DropAllCapabilities()); + CHECK(!Credentials::HasAnyCapability()); +} + +SANDBOX_TEST(Credentials, CanCreateProcessInNewUserNS) { + CHECK(Credentials::DropAllCapabilities()); + bool user_ns_supported = Credentials::CanCreateProcessInNewUserNS(); + bool moved_to_new_ns = Credentials::MoveToNewUserNS(); + CHECK_EQ(user_ns_supported, moved_to_new_ns); +} + +SANDBOX_TEST(Credentials, UidIsPreserved) { + CHECK(Credentials::DropAllCapabilities()); + uid_t old_ruid, old_euid, old_suid; + gid_t old_rgid, old_egid, old_sgid; + PCHECK(0 == getresuid(&old_ruid, &old_euid, &old_suid)); + PCHECK(0 == getresgid(&old_rgid, &old_egid, &old_sgid)); + // Probably missing kernel support. + if (!Credentials::MoveToNewUserNS()) return; + uid_t new_ruid, new_euid, new_suid; + PCHECK(0 == getresuid(&new_ruid, &new_euid, &new_suid)); + CHECK(old_ruid == new_ruid); + CHECK(old_euid == new_euid); + CHECK(old_suid == new_suid); + + gid_t new_rgid, new_egid, new_sgid; + PCHECK(0 == getresgid(&new_rgid, &new_egid, &new_sgid)); + CHECK(old_rgid == new_rgid); + CHECK(old_egid == new_egid); + CHECK(old_sgid == new_sgid); +} + +bool NewUserNSCycle() { + if (!Credentials::MoveToNewUserNS() || + !Credentials::HasAnyCapability() || + !Credentials::DropAllCapabilities() || + Credentials::HasAnyCapability()) { + return false; + } + return true; +} + +SANDBOX_TEST(Credentials, NestedUserNS) { + CHECK(Credentials::DropAllCapabilities()); + // Probably missing kernel support. + if (!Credentials::MoveToNewUserNS()) return; + CHECK(Credentials::DropAllCapabilities()); + // As of 3.12, the kernel has a limit of 32. See create_user_ns(). + const int kNestLevel = 10; + for (int i = 0; i < kNestLevel; ++i) { + CHECK(NewUserNSCycle()) << "Creating new user NS failed at iteration " + << i << "."; + } +} + +// Test the WorkingDirectoryIsRoot() helper. +SANDBOX_TEST(Credentials, CanDetectRoot) { + PCHECK(0 == chdir("/proc/")); + CHECK(!WorkingDirectoryIsRoot()); + PCHECK(0 == chdir("/")); + CHECK(WorkingDirectoryIsRoot()); +} + +// Disabled on ASAN because of crbug.com/451603. +SANDBOX_TEST(Credentials, DISABLE_ON_ASAN(DropFileSystemAccessIsSafe)) { + CHECK(Credentials::DropAllCapabilities()); + // Probably missing kernel support. + if (!Credentials::MoveToNewUserNS()) return; + CHECK(Credentials::DropFileSystemAccess(ProcUtil::OpenProc().get())); + CHECK(!base::DirectoryExists(base::FilePath("/proc"))); + CHECK(WorkingDirectoryIsRoot()); + CHECK(base::IsDirectoryEmpty(base::FilePath("/"))); + // We want the chroot to never have a subdirectory. A subdirectory + // could allow a chroot escape. + CHECK_NE(0, mkdir("/test", 0700)); +} + +// Check that after dropping filesystem access and dropping privileges +// it is not possible to regain capabilities. +SANDBOX_TEST(Credentials, DISABLE_ON_ASAN(CannotRegainPrivileges)) { + base::ScopedFD proc_fd(ProcUtil::OpenProc()); + CHECK(Credentials::DropAllCapabilities(proc_fd.get())); + // Probably missing kernel support. + if (!Credentials::MoveToNewUserNS()) return; + CHECK(Credentials::DropFileSystemAccess(proc_fd.get())); + CHECK(Credentials::DropAllCapabilities(proc_fd.get())); + + // The kernel should now prevent us from regaining capabilities because we + // are in a chroot. + CHECK(!Credentials::CanCreateProcessInNewUserNS()); + CHECK(!Credentials::MoveToNewUserNS()); +} + +SANDBOX_TEST(Credentials, SetCapabilities) { + // Probably missing kernel support. + if (!Credentials::MoveToNewUserNS()) + return; + + base::ScopedFD proc_fd(ProcUtil::OpenProc()); + + CHECK(Credentials::HasCapability(Credentials::Capability::SYS_ADMIN)); + CHECK(Credentials::HasCapability(Credentials::Capability::SYS_CHROOT)); + + std::vector<Credentials::Capability> caps; + caps.push_back(Credentials::Capability::SYS_CHROOT); + CHECK(Credentials::SetCapabilities(proc_fd.get(), caps)); + + CHECK(!Credentials::HasCapability(Credentials::Capability::SYS_ADMIN)); + CHECK(Credentials::HasCapability(Credentials::Capability::SYS_CHROOT)); + + const std::vector<Credentials::Capability> no_caps; + CHECK(Credentials::SetCapabilities(proc_fd.get(), no_caps)); + CHECK(!Credentials::HasAnyCapability()); +} + +SANDBOX_TEST(Credentials, SetCapabilitiesAndChroot) { + // Probably missing kernel support. + if (!Credentials::MoveToNewUserNS()) + return; + + base::ScopedFD proc_fd(ProcUtil::OpenProc()); + + CHECK(Credentials::HasCapability(Credentials::Capability::SYS_CHROOT)); + PCHECK(chroot("/") == 0); + + std::vector<Credentials::Capability> caps; + caps.push_back(Credentials::Capability::SYS_CHROOT); + CHECK(Credentials::SetCapabilities(proc_fd.get(), caps)); + PCHECK(chroot("/") == 0); + + CHECK(Credentials::DropAllCapabilities()); + PCHECK(chroot("/") == -1 && errno == EPERM); +} + +SANDBOX_TEST(Credentials, SetCapabilitiesMatchesLibCap2) { + // Probably missing kernel support. + if (!Credentials::MoveToNewUserNS()) + return; + + base::ScopedFD proc_fd(ProcUtil::OpenProc()); + + std::vector<Credentials::Capability> caps; + caps.push_back(Credentials::Capability::SYS_CHROOT); + CHECK(Credentials::SetCapabilities(proc_fd.get(), caps)); + + ScopedCap actual_cap(cap_get_proc()); + PCHECK(actual_cap != nullptr); + + ScopedCap expected_cap(cap_init()); + PCHECK(expected_cap != nullptr); + + const cap_value_t allowed_cap = CAP_SYS_CHROOT; + for (const cap_flag_t flag : {CAP_EFFECTIVE, CAP_PERMITTED}) { + PCHECK(cap_set_flag(expected_cap.get(), flag, 1, &allowed_cap, CAP_SET) == + 0); + } + + CHECK_EQ(0, cap_compare(expected_cap.get(), actual_cap.get())); +} + +} // namespace. + +} // namespace sandbox. diff --git a/sandbox/linux/services/init_process_reaper.cc b/sandbox/linux/services/init_process_reaper.cc new file mode 100644 index 0000000000..2e0b90b7b5 --- /dev/null +++ b/sandbox/linux/services/init_process_reaper.cc @@ -0,0 +1,101 @@ +// Copyright 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/init_process_reaper.h" + +#include <signal.h> +#include <string.h> +#include <sys/socket.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include "base/callback.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" + +namespace sandbox { + +namespace { + +void DoNothingSignalHandler(int signal) {} + +} // namespace + +bool CreateInitProcessReaper(base::Closure* post_fork_parent_callback) { + int sync_fds[2]; + // We want to use send, so we can't use a pipe + if (socketpair(AF_UNIX, SOCK_STREAM, 0, sync_fds)) { + PLOG(ERROR) << "Failed to create socketpair"; + return false; + } + pid_t child_pid = fork(); + if (child_pid == -1) { + int close_ret; + close_ret = IGNORE_EINTR(close(sync_fds[0])); + DPCHECK(!close_ret); + close_ret = IGNORE_EINTR(close(sync_fds[1])); + DPCHECK(!close_ret); + return false; + } + if (child_pid) { + // In the parent, assuming the role of an init process. + // The disposition for SIGCHLD cannot be SIG_IGN or wait() will only return + // once all of our childs are dead. Since we're init we need to reap childs + // as they come. + struct sigaction action; + memset(&action, 0, sizeof(action)); + action.sa_handler = &DoNothingSignalHandler; + CHECK(sigaction(SIGCHLD, &action, NULL) == 0); + + int close_ret; + close_ret = IGNORE_EINTR(close(sync_fds[0])); + DPCHECK(!close_ret); + close_ret = shutdown(sync_fds[1], SHUT_RD); + DPCHECK(!close_ret); + if (post_fork_parent_callback) + post_fork_parent_callback->Run(); + // Tell the child to continue + CHECK(HANDLE_EINTR(send(sync_fds[1], "C", 1, MSG_NOSIGNAL)) == 1); + close_ret = IGNORE_EINTR(close(sync_fds[1])); + DPCHECK(!close_ret); + + for (;;) { + // Loop until we have reaped our one natural child + siginfo_t reaped_child_info; + int wait_ret = + HANDLE_EINTR(waitid(P_ALL, 0, &reaped_child_info, WEXITED)); + if (wait_ret) + _exit(1); + if (reaped_child_info.si_pid == child_pid) { + int exit_code = 0; + // We're done waiting + if (reaped_child_info.si_code == CLD_EXITED) { + exit_code = reaped_child_info.si_status; + } + // Exit with the same exit code as our parent. Exit with 0 if we got + // signaled. + _exit(exit_code); + } + } + } else { + // The child needs to wait for the parent to run the callback to avoid a + // race condition. + int close_ret; + close_ret = IGNORE_EINTR(close(sync_fds[1])); + DPCHECK(!close_ret); + close_ret = shutdown(sync_fds[0], SHUT_WR); + DPCHECK(!close_ret); + char should_continue; + int read_ret = HANDLE_EINTR(read(sync_fds[0], &should_continue, 1)); + close_ret = IGNORE_EINTR(close(sync_fds[0])); + DPCHECK(!close_ret); + if (read_ret == 1) + return true; + else + return false; + } +} + +} // namespace sandbox. diff --git a/sandbox/linux/services/init_process_reaper.h b/sandbox/linux/services/init_process_reaper.h new file mode 100644 index 0000000000..840f6fcda7 --- /dev/null +++ b/sandbox/linux/services/init_process_reaper.h @@ -0,0 +1,25 @@ +// Copyright 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_INIT_PROCESS_REAPER_H_ +#define SANDBOX_LINUX_SERVICES_INIT_PROCESS_REAPER_H_ + +#include "base/callback_forward.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// The current process will fork(). The parent will become a process reaper +// like init(1). The child will continue normally (after this function +// returns). +// If not NULL, |post_fork_parent_callback| will run in the parent almost +// immediately after fork(). +// Since this function calls fork(), it's very important that the caller has +// only one thread running. +SANDBOX_EXPORT bool CreateInitProcessReaper( + base::Closure* post_fork_parent_callback); + +} // namespace sandbox. + +#endif // SANDBOX_LINUX_SERVICES_INIT_PROCESS_REAPER_H_ diff --git a/sandbox/linux/services/libc_urandom_override.cc b/sandbox/linux/services/libc_urandom_override.cc new file mode 100644 index 0000000000..33bb25d6b1 --- /dev/null +++ b/sandbox/linux/services/libc_urandom_override.cc @@ -0,0 +1,236 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/libc_urandom_override.h" + +#include <dlfcn.h> +#include <pthread.h> +#include <stdio.h> +#include <sys/stat.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "base/rand_util.h" + +// Note: this file is used by the zygote and nacl_helper. + +#if !defined(HAVE_XSTAT) && defined(LIBC_GLIBC) +#define HAVE_XSTAT 1 +#endif + +namespace sandbox { + +static bool g_override_urandom = false; + +// TODO(sergeyu): Currently InitLibcUrandomOverrides() doesn't work properly +// under ASan or MSan - it crashes content_unittests. Make sure it works +// properly and enable it here. http://crbug.com/123263 +#if !(defined(ADDRESS_SANITIZER) || defined(MEMORY_SANITIZER)) +static void InitLibcFileIOFunctions(); +static pthread_once_t g_libc_file_io_funcs_guard = PTHREAD_ONCE_INIT; +#endif + +void InitLibcUrandomOverrides() { + // Make sure /dev/urandom is open. + base::GetUrandomFD(); + g_override_urandom = true; + +#if !(defined(ADDRESS_SANITIZER) || defined(MEMORY_SANITIZER)) + CHECK_EQ(0, pthread_once(&g_libc_file_io_funcs_guard, + InitLibcFileIOFunctions)); +#endif +} + +#if !(defined(ADDRESS_SANITIZER) || defined(MEMORY_SANITIZER)) + +static const char kUrandomDevPath[] = "/dev/urandom"; + +typedef FILE* (*FopenFunction)(const char* path, const char* mode); + +static FopenFunction g_libc_fopen = NULL; +static FopenFunction g_libc_fopen64 = NULL; + +#if HAVE_XSTAT +typedef int (*XstatFunction)(int version, const char *path, struct stat *buf); +typedef int (*Xstat64Function)(int version, const char *path, + struct stat64 *buf); + +static XstatFunction g_libc_xstat = NULL; +static Xstat64Function g_libc_xstat64 = NULL; +#else +typedef int (*StatFunction)(const char *path, struct stat *buf); +typedef int (*Stat64Function)(const char *path, struct stat64 *buf); + +static StatFunction g_libc_stat = NULL; +static Stat64Function g_libc_stat64 = NULL; +#endif // HAVE_XSTAT + +// Find the libc's real fopen* and *stat* functions. This should only be +// called once, and should be guarded by g_libc_file_io_funcs_guard. +static void InitLibcFileIOFunctions() { + g_libc_fopen = reinterpret_cast<FopenFunction>( + dlsym(RTLD_NEXT, "fopen")); + g_libc_fopen64 = reinterpret_cast<FopenFunction>( + dlsym(RTLD_NEXT, "fopen64")); + + if (!g_libc_fopen) { + LOG(FATAL) << "Failed to get fopen() from libc."; + } else if (!g_libc_fopen64) { +#if !defined(OS_OPENBSD) && !defined(OS_FREEBSD) + LOG(WARNING) << "Failed to get fopen64() from libc. Using fopen() instead."; +#endif // !defined(OS_OPENBSD) && !defined(OS_FREEBSD) + g_libc_fopen64 = g_libc_fopen; + } + +#if HAVE_XSTAT + g_libc_xstat = reinterpret_cast<XstatFunction>( + dlsym(RTLD_NEXT, "__xstat")); + g_libc_xstat64 = reinterpret_cast<Xstat64Function>( + dlsym(RTLD_NEXT, "__xstat64")); + + if (!g_libc_xstat) { + LOG(FATAL) << "Failed to get __xstat() from libc."; + } + if (!g_libc_xstat64) { + LOG(FATAL) << "Failed to get __xstat64() from libc."; + } +#else + g_libc_stat = reinterpret_cast<StatFunction>( + dlsym(RTLD_NEXT, "stat")); + g_libc_stat64 = reinterpret_cast<Stat64Function>( + dlsym(RTLD_NEXT, "stat64")); + + if (!g_libc_stat) { + LOG(FATAL) << "Failed to get stat() from libc."; + } + if (!g_libc_stat64) { + LOG(FATAL) << "Failed to get stat64() from libc."; + } +#endif // HAVE_XSTAT +} + +// fopen() and fopen64() are intercepted here so that NSS can open +// /dev/urandom to seed its random number generator. NSS is used by +// remoting in the sendbox. + +// fopen() call may be redirected to fopen64() in stdio.h using +// __REDIRECT(), which sets asm name for fopen() to "fopen64". This +// means that we cannot override fopen() directly here. Instead the +// the code below defines fopen_override() function with asm name +// "fopen", so that all references to fopen() will resolve to this +// function. +__attribute__ ((__visibility__("default"))) +FILE* fopen_override(const char* path, const char* mode) __asm__ ("fopen"); + +__attribute__ ((__visibility__("default"))) +FILE* fopen_override(const char* path, const char* mode) { + if (g_override_urandom && strcmp(path, kUrandomDevPath) == 0) { + int fd = HANDLE_EINTR(dup(base::GetUrandomFD())); + if (fd < 0) { + PLOG(ERROR) << "dup() failed."; + return NULL; + } + return fdopen(fd, mode); + } else { + CHECK_EQ(0, pthread_once(&g_libc_file_io_funcs_guard, + InitLibcFileIOFunctions)); + return g_libc_fopen(path, mode); + } +} + +__attribute__ ((__visibility__("default"))) +FILE* fopen64(const char* path, const char* mode) { + if (g_override_urandom && strcmp(path, kUrandomDevPath) == 0) { + int fd = HANDLE_EINTR(dup(base::GetUrandomFD())); + if (fd < 0) { + PLOG(ERROR) << "dup() failed."; + return NULL; + } + return fdopen(fd, mode); + } else { + CHECK_EQ(0, pthread_once(&g_libc_file_io_funcs_guard, + InitLibcFileIOFunctions)); + return g_libc_fopen64(path, mode); + } +} + +// The stat() family of functions are subject to the same problem as +// fopen(), so we have to use the same trick to override them. + +#if HAVE_XSTAT + +__attribute__ ((__visibility__("default"))) +int xstat_override(int version, + const char *path, + struct stat *buf) __asm__ ("__xstat"); + +__attribute__ ((__visibility__("default"))) +int xstat_override(int version, const char *path, struct stat *buf) { + if (g_override_urandom && strcmp(path, kUrandomDevPath) == 0) { + int result = __fxstat(version, base::GetUrandomFD(), buf); + return result; + } else { + CHECK_EQ(0, pthread_once(&g_libc_file_io_funcs_guard, + InitLibcFileIOFunctions)); + return g_libc_xstat(version, path, buf); + } +} + +__attribute__ ((__visibility__("default"))) +int xstat64_override(int version, + const char *path, + struct stat64 *buf) __asm__ ("__xstat64"); + +__attribute__ ((__visibility__("default"))) +int xstat64_override(int version, const char *path, struct stat64 *buf) { + if (g_override_urandom && strcmp(path, kUrandomDevPath) == 0) { + int result = __fxstat64(version, base::GetUrandomFD(), buf); + return result; + } else { + CHECK_EQ(0, pthread_once(&g_libc_file_io_funcs_guard, + InitLibcFileIOFunctions)); + return g_libc_xstat64(version, path, buf); + } +} + +#else + +__attribute__ ((__visibility__("default"))) +int stat_override(const char *path, + struct stat *buf) __asm__ ("stat"); + +__attribute__ ((__visibility__("default"))) +int stat_override(const char *path, struct stat *buf) { + if (g_override_urandom && strcmp(path, kUrandomDevPath) == 0) { + int result = fstat(base::GetUrandomFD(), buf); + return result; + } else { + CHECK_EQ(0, pthread_once(&g_libc_file_io_funcs_guard, + InitLibcFileIOFunctions)); + return g_libc_stat(path, buf); + } +} + +__attribute__ ((__visibility__("default"))) +int stat64_override(const char *path, + struct stat64 *buf) __asm__ ("stat64"); + +__attribute__ ((__visibility__("default"))) +int stat64_override(const char *path, struct stat64 *buf) { + if (g_override_urandom && strcmp(path, kUrandomDevPath) == 0) { + int result = fstat64(base::GetUrandomFD(), buf); + return result; + } else { + CHECK_EQ(0, pthread_once(&g_libc_file_io_funcs_guard, + InitLibcFileIOFunctions)); + return g_libc_stat64(path, buf); + } +} + +#endif // HAVE_XSTAT + +#endif // !(defined(ADDRESS_SANITIZER) || defined(MEMORY_SANITIZER)) + +} // namespace content diff --git a/sandbox/linux/services/libc_urandom_override.h b/sandbox/linux/services/libc_urandom_override.h new file mode 100644 index 0000000000..86212f8bc4 --- /dev/null +++ b/sandbox/linux/services/libc_urandom_override.h @@ -0,0 +1,14 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_LIBC_URANDOM_OVERRIDE_H_ +#define SANDBOX_LINUX_SERVICES_LIBC_URANDOM_OVERRIDE_H_ + +namespace sandbox { + +void InitLibcUrandomOverrides(); + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_LIBC_URANDOM_OVERRIDE_H_ diff --git a/sandbox/linux/services/namespace_sandbox.cc b/sandbox/linux/services/namespace_sandbox.cc new file mode 100644 index 0000000000..23796446f3 --- /dev/null +++ b/sandbox/linux/services/namespace_sandbox.cc @@ -0,0 +1,208 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/namespace_sandbox.h" + +#include <sched.h> +#include <signal.h> +#include <stdlib.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <utility> +#include <vector> + +#include "base/command_line.h" +#include "base/environment.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/macros.h" +#include "base/posix/eintr_wrapper.h" +#include "base/process/launch.h" +#include "base/process/process.h" +#include "sandbox/linux/services/credentials.h" +#include "sandbox/linux/services/namespace_utils.h" + +namespace sandbox { + +namespace { + +const char kSandboxUSERNSEnvironmentVarName[] = "SBX_USER_NS"; +const char kSandboxPIDNSEnvironmentVarName[] = "SBX_PID_NS"; +const char kSandboxNETNSEnvironmentVarName[] = "SBX_NET_NS"; + +#if !defined(OS_NACL_NONSFI) +class WriteUidGidMapDelegate : public base::LaunchOptions::PreExecDelegate { + public: + WriteUidGidMapDelegate() + : uid_(getuid()), + gid_(getgid()), + supports_deny_setgroups_( + NamespaceUtils::KernelSupportsDenySetgroups()) {} + + ~WriteUidGidMapDelegate() override {} + + void RunAsyncSafe() override { + if (supports_deny_setgroups_) { + RAW_CHECK(NamespaceUtils::DenySetgroups()); + } + RAW_CHECK(NamespaceUtils::WriteToIdMapFile("/proc/self/uid_map", uid_)); + RAW_CHECK(NamespaceUtils::WriteToIdMapFile("/proc/self/gid_map", gid_)); + } + + private: + const uid_t uid_; + const gid_t gid_; + const bool supports_deny_setgroups_; + DISALLOW_COPY_AND_ASSIGN(WriteUidGidMapDelegate); +}; + +void SetEnvironForNamespaceType(base::EnvironmentMap* environ, + base::NativeEnvironmentString env_var, + bool value) { + // An empty string causes the env var to be unset in the child process. + (*environ)[env_var] = value ? "1" : ""; +} + +// Linux supports up to 64 signals. This should be updated if that ever changes. +int g_signal_exit_codes[64]; + +void TerminationSignalHandler(int sig) { + // Return a special exit code so that the process is detected as terminated by + // a signal. + const size_t sig_idx = static_cast<size_t>(sig); + if (sig_idx < arraysize(g_signal_exit_codes)) { + _exit(g_signal_exit_codes[sig_idx]); + } + + _exit(NamespaceSandbox::kDefaultExitCode); +} +#endif // !defined(OS_NACL_NONSFI) + +} // namespace + +#if !defined(OS_NACL_NONSFI) +// static +base::Process NamespaceSandbox::LaunchProcess( + const base::CommandLine& cmdline, + const base::LaunchOptions& options) { + return LaunchProcess(cmdline.argv(), options); +} + +// static +base::Process NamespaceSandbox::LaunchProcess( + const std::vector<std::string>& argv, + const base::LaunchOptions& options) { + int clone_flags = 0; + int ns_types[] = {CLONE_NEWUSER, CLONE_NEWPID, CLONE_NEWNET}; + for (const int ns_type : ns_types) { + if (NamespaceUtils::KernelSupportsUnprivilegedNamespace(ns_type)) { + clone_flags |= ns_type; + } + } + CHECK(clone_flags & CLONE_NEWUSER); + + // These fields may not be set by the caller. + CHECK(options.pre_exec_delegate == nullptr); + CHECK_EQ(0, options.clone_flags); + + WriteUidGidMapDelegate write_uid_gid_map_delegate; + + base::LaunchOptions launch_options = options; + launch_options.pre_exec_delegate = &write_uid_gid_map_delegate; + launch_options.clone_flags = clone_flags; + + const std::pair<int, const char*> clone_flag_environ[] = { + std::make_pair(CLONE_NEWUSER, kSandboxUSERNSEnvironmentVarName), + std::make_pair(CLONE_NEWPID, kSandboxPIDNSEnvironmentVarName), + std::make_pair(CLONE_NEWNET, kSandboxNETNSEnvironmentVarName), + }; + + base::EnvironmentMap* environ = &launch_options.environ; + for (const auto& entry : clone_flag_environ) { + const int flag = entry.first; + const char* environ_name = entry.second; + SetEnvironForNamespaceType(environ, environ_name, clone_flags & flag); + } + + return base::LaunchProcess(argv, launch_options); +} + +// static +pid_t NamespaceSandbox::ForkInNewPidNamespace(bool drop_capabilities_in_child) { + const pid_t pid = + base::ForkWithFlags(CLONE_NEWPID | SIGCHLD, nullptr, nullptr); + if (pid < 0) { + return pid; + } + + if (pid == 0) { + DCHECK_EQ(1, getpid()); + if (drop_capabilities_in_child) { + // Since we just forked, we are single-threaded, so this should be safe. + CHECK(Credentials::DropAllCapabilitiesOnCurrentThread()); + } + return 0; + } + + return pid; +} + +// static +void NamespaceSandbox::InstallDefaultTerminationSignalHandlers() { + static const int kDefaultTermSignals[] = { + SIGHUP, SIGINT, SIGABRT, SIGQUIT, SIGPIPE, SIGTERM, SIGUSR1, SIGUSR2, + }; + + for (const int sig : kDefaultTermSignals) { + InstallTerminationSignalHandler(sig, kDefaultExitCode); + } +} + +// static +bool NamespaceSandbox::InstallTerminationSignalHandler( + int sig, + int exit_code) { + struct sigaction old_action; + PCHECK(sigaction(sig, nullptr, &old_action) == 0); + + if (old_action.sa_flags & SA_SIGINFO && + old_action.sa_sigaction != nullptr) { + return false; + } else if (old_action.sa_handler != SIG_DFL) { + return false; + } + + const size_t sig_idx = static_cast<size_t>(sig); + CHECK_LT(sig_idx, arraysize(g_signal_exit_codes)); + + DCHECK_GE(exit_code, 0); + DCHECK_LT(exit_code, 256); + + g_signal_exit_codes[sig_idx] = exit_code; + + struct sigaction action = {}; + action.sa_handler = &TerminationSignalHandler; + PCHECK(sigaction(sig, &action, nullptr) == 0); + return true; +} +#endif // !defined(OS_NACL_NONSFI) + +// static +bool NamespaceSandbox::InNewUserNamespace() { + return getenv(kSandboxUSERNSEnvironmentVarName) != nullptr; +} + +// static +bool NamespaceSandbox::InNewPidNamespace() { + return getenv(kSandboxPIDNSEnvironmentVarName) != nullptr; +} + +// static +bool NamespaceSandbox::InNewNetNamespace() { + return getenv(kSandboxNETNSEnvironmentVarName) != nullptr; +} + +} // namespace sandbox diff --git a/sandbox/linux/services/namespace_sandbox.h b/sandbox/linux/services/namespace_sandbox.h new file mode 100644 index 0000000000..80097fb16a --- /dev/null +++ b/sandbox/linux/services/namespace_sandbox.h @@ -0,0 +1,101 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_NAMESPACE_SANDBOX_H_ +#define SANDBOX_LINUX_SERVICES_NAMESPACE_SANDBOX_H_ + +#include <sys/types.h> + +#include <string> +#include <vector> + +#include "base/command_line.h" +#include "base/macros.h" +#include "base/process/launch.h" +#include "base/process/process.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// Helper class for starting a process inside a new user, PID, and network +// namespace. Before using a namespace sandbox, check for namespaces support +// using Credentials::CanCreateProcessInNewUserNS. +// +// A typical use for "A" launching a sandboxed process "B" would be: +// 1. A sets up a command line and launch options for process B. +// 2. A launches B with LaunchProcess. +// 3. B should be prepared to assume the role of init(1). In particular, apart +// from SIGKILL and SIGSTOP, B cannot receive any signal for which it does +// not have an explicit signal handler registered. +// If B dies, all the processes in the namespace will die. +// B can fork() and the parent can assume the role of init(1), by using +// CreateInitProcessReaper(). +// 4. B chroots using Credentials::MoveToNewUserNS() and +// Credentials::DropFileSystemAccess() +// 5. B drops capabilities gained by entering the new user namespace with +// Credentials::DropAllCapabilities(). +class SANDBOX_EXPORT NamespaceSandbox { + public: +#if !defined(OS_NACL_NONSFI) + static const int kDefaultExitCode = 1; + + // Launch a new process inside its own user/PID/network namespaces (depending + // on kernel support). Requires at a minimum that user namespaces are + // supported (use Credentials::CanCreateProcessInNewUserNS to check this). + // + // pre_exec_delegate and clone_flags fields of LaunchOptions should be nullptr + // and 0, respectively, since this function makes a copy of options and + // overrides them. + static base::Process LaunchProcess(const base::CommandLine& cmdline, + const base::LaunchOptions& options); + static base::Process LaunchProcess(const std::vector<std::string>& argv, + const base::LaunchOptions& options); + + // Forks a process in its own PID namespace. The child process is the init + // process inside of the PID namespace, so if the child needs to fork further, + // it should call CreateInitProcessReaper, which turns the init process into a + // reaper process. + // + // Otherwise, the child should setup handlers for signals which should + // terminate the process using InstallDefaultTerminationSignalHandlers or + // InstallTerminationSignalHandler. This works around the fact that init + // processes ignore such signals unless they have an explicit handler set. + // + // This function requries CAP_SYS_ADMIN. If |drop_capabilities_in_child| is + // true, then capabilities are dropped in the child. + static pid_t ForkInNewPidNamespace(bool drop_capabilities_in_child); + + // Installs a signal handler for: + // + // SIGHUP, SIGINT, SIGABRT, SIGQUIT, SIGPIPE, SIGTERM, SIGUSR1, SIGUSR2 + // + // that exits with kDefaultExitCode. These are signals whose default action is + // to terminate the program (apart from SIGILL, SIGFPE, and SIGSEGV, which + // will still terminate the process if e.g. an illegal instruction is + // encountered, etc.). + // + // If any of these already had a signal handler installed, this function will + // not override them. + static void InstallDefaultTerminationSignalHandlers(); + + // Installs a signal handler for |sig| which exits with |exit_code|. If a + // signal handler was already present for |sig|, does nothing and returns + // false. + static bool InstallTerminationSignalHandler(int sig, int exit_code); +#endif // !defined(OS_NACL_NONSFI) + + // Returns whether the namespace sandbox created a new user, PID, and network + // namespace. In particular, InNewUserNamespace should return true iff the + // process was started via this class. + static bool InNewUserNamespace(); + static bool InNewPidNamespace(); + static bool InNewNetNamespace(); + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(NamespaceSandbox); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_NAMESPACE_SANDBOX_H_ diff --git a/sandbox/linux/services/namespace_sandbox_unittest.cc b/sandbox/linux/services/namespace_sandbox_unittest.cc new file mode 100644 index 0000000000..547ef6728c --- /dev/null +++ b/sandbox/linux/services/namespace_sandbox_unittest.cc @@ -0,0 +1,217 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/namespace_sandbox.h" + +#include <signal.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include <string> +#include <utility> + +#include "base/command_line.h" +#include "base/files/file_enumerator.h" +#include "base/files/file_path.h" +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "base/process/launch.h" +#include "base/process/process.h" +#include "base/test/multiprocess_test.h" +#include "sandbox/linux/services/credentials.h" +#include "sandbox/linux/services/namespace_utils.h" +#include "sandbox/linux/services/proc_util.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" +#include "testing/multiprocess_func_list.h" + +namespace sandbox { + +namespace { + +bool RootDirectoryIsEmpty() { + base::FilePath root("/"); + int file_type = + base::FileEnumerator::DIRECTORIES | base::FileEnumerator::FILES; + base::FileEnumerator enumerator_before(root, false, file_type); + return enumerator_before.Next().empty(); +} + +class NamespaceSandboxTest : public base::MultiProcessTest { + public: + void TestProc(const std::string& procname) { + if (!Credentials::CanCreateProcessInNewUserNS()) { + return; + } + + base::FileHandleMappingVector fds_to_remap = { + std::make_pair(STDOUT_FILENO, STDOUT_FILENO), + std::make_pair(STDERR_FILENO, STDERR_FILENO), + }; + base::LaunchOptions launch_options; + launch_options.fds_to_remap = &fds_to_remap; + + base::Process process = + NamespaceSandbox::LaunchProcess(MakeCmdLine(procname), launch_options); + ASSERT_TRUE(process.IsValid()); + + const int kDummyExitCode = 42; + int exit_code = kDummyExitCode; + EXPECT_TRUE(process.WaitForExit(&exit_code)); + EXPECT_EQ(0, exit_code); + } +}; + +MULTIPROCESS_TEST_MAIN(SimpleChildProcess) { + scoped_ptr<base::Environment> env(base::Environment::Create()); + bool in_user_ns = NamespaceSandbox::InNewUserNamespace(); + bool in_pid_ns = NamespaceSandbox::InNewPidNamespace(); + bool in_net_ns = NamespaceSandbox::InNewNetNamespace(); + CHECK(in_user_ns); + CHECK_EQ(in_pid_ns, + NamespaceUtils::KernelSupportsUnprivilegedNamespace(CLONE_NEWPID)); + CHECK_EQ(in_net_ns, + NamespaceUtils::KernelSupportsUnprivilegedNamespace(CLONE_NEWNET)); + if (in_pid_ns) { + CHECK_EQ(1, getpid()); + } + return 0; +} + +TEST_F(NamespaceSandboxTest, BasicUsage) { + TestProc("SimpleChildProcess"); +} + +MULTIPROCESS_TEST_MAIN(ChrootMe) { + CHECK(!RootDirectoryIsEmpty()); + CHECK(sandbox::Credentials::MoveToNewUserNS()); + CHECK(sandbox::Credentials::DropFileSystemAccess(ProcUtil::OpenProc().get())); + CHECK(RootDirectoryIsEmpty()); + return 0; +} + +// Temporarily disabled on ASAN due to crbug.com/451603. +TEST_F(NamespaceSandboxTest, DISABLE_ON_ASAN(ChrootAndDropCapabilities)) { + TestProc("ChrootMe"); +} + +MULTIPROCESS_TEST_MAIN(NestedNamespaceSandbox) { + base::FileHandleMappingVector fds_to_remap = { + std::make_pair(STDOUT_FILENO, STDOUT_FILENO), + std::make_pair(STDERR_FILENO, STDERR_FILENO), + }; + base::LaunchOptions launch_options; + launch_options.fds_to_remap = &fds_to_remap; + base::Process process = NamespaceSandbox::LaunchProcess( + base::CommandLine(base::FilePath("/bin/true")), launch_options); + CHECK(process.IsValid()); + + const int kDummyExitCode = 42; + int exit_code = kDummyExitCode; + CHECK(process.WaitForExit(&exit_code)); + CHECK_EQ(0, exit_code); + return 0; +} + +TEST_F(NamespaceSandboxTest, NestedNamespaceSandbox) { + TestProc("NestedNamespaceSandbox"); +} + +const int kNormalExitCode = 0; +const int kSignalTerminationExitCode = 255; + +// Ensure that CHECK(false) is distinguishable from _exit(kNormalExitCode). +// Allowing noise since CHECK(false) will write a stack trace to stderr. +SANDBOX_TEST_ALLOW_NOISE(ForkInNewPidNamespace, CheckDoesNotReturnZero) { + if (!Credentials::CanCreateProcessInNewUserNS()) { + return; + } + + CHECK(sandbox::Credentials::MoveToNewUserNS()); + const pid_t pid = NamespaceSandbox::ForkInNewPidNamespace( + /*drop_capabilities_in_child=*/true); + CHECK_GE(pid, 0); + + if (pid == 0) { + CHECK(false); + _exit(kNormalExitCode); + } + + int status; + PCHECK(waitpid(pid, &status, 0) == pid); + if (WIFEXITED(status)) { + CHECK_NE(kNormalExitCode, WEXITSTATUS(status)); + } +} + +SANDBOX_TEST(ForkInNewPidNamespace, BasicUsage) { + if (!Credentials::CanCreateProcessInNewUserNS()) { + return; + } + + CHECK(sandbox::Credentials::MoveToNewUserNS()); + const pid_t pid = NamespaceSandbox::ForkInNewPidNamespace( + /*drop_capabilities_in_child=*/true); + CHECK_GE(pid, 0); + + if (pid == 0) { + CHECK_EQ(1, getpid()); + CHECK(!Credentials::HasAnyCapability()); + _exit(kNormalExitCode); + } + + int status; + PCHECK(waitpid(pid, &status, 0) == pid); + CHECK(WIFEXITED(status)); + CHECK_EQ(kNormalExitCode, WEXITSTATUS(status)); +} + +SANDBOX_TEST(ForkInNewPidNamespace, ExitWithSignal) { + if (!Credentials::CanCreateProcessInNewUserNS()) { + return; + } + + CHECK(sandbox::Credentials::MoveToNewUserNS()); + const pid_t pid = NamespaceSandbox::ForkInNewPidNamespace( + /*drop_capabilities_in_child=*/true); + CHECK_GE(pid, 0); + + if (pid == 0) { + CHECK_EQ(1, getpid()); + CHECK(!Credentials::HasAnyCapability()); + CHECK(NamespaceSandbox::InstallTerminationSignalHandler( + SIGTERM, kSignalTerminationExitCode)); + while (true) { + raise(SIGTERM); + } + } + + int status; + PCHECK(waitpid(pid, &status, 0) == pid); + CHECK(WIFEXITED(status)); + CHECK_EQ(kSignalTerminationExitCode, WEXITSTATUS(status)); +} + +volatile sig_atomic_t signal_handler_called; +void ExitSuccessfully(int sig) { + signal_handler_called = 1; +} + +SANDBOX_TEST(InstallTerminationSignalHandler, DoesNotOverrideExistingHandlers) { + struct sigaction action = {}; + action.sa_handler = &ExitSuccessfully; + PCHECK(sigaction(SIGUSR1, &action, nullptr) == 0); + + NamespaceSandbox::InstallDefaultTerminationSignalHandlers(); + CHECK(!NamespaceSandbox::InstallTerminationSignalHandler( + SIGUSR1, kSignalTerminationExitCode)); + + raise(SIGUSR1); + CHECK_EQ(1, signal_handler_called); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/services/namespace_utils.cc b/sandbox/linux/services/namespace_utils.cc new file mode 100644 index 0000000000..82a544453f --- /dev/null +++ b/sandbox/linux/services/namespace_utils.cc @@ -0,0 +1,117 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/namespace_utils.h" + +#include <fcntl.h> +#include <sched.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <unistd.h> + +#include <string> + +#include "base/files/file_path.h" +#include "base/files/file_util.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "base/process/launch.h" +#include "base/strings/safe_sprintf.h" +#include "base/third_party/valgrind/valgrind.h" + +namespace sandbox { + +namespace { +bool IsRunningOnValgrind() { + return RUNNING_ON_VALGRIND; +} + +const char kProcSelfSetgroups[] = "/proc/self/setgroups"; +} // namespace + +// static +bool NamespaceUtils::WriteToIdMapFile(const char* map_file, generic_id_t id) { + // This function needs to be async-signal-safe, as it may be called in between + // fork and exec. + + int fd = HANDLE_EINTR(open(map_file, O_WRONLY)); + if (fd == -1) { + return false; + } + + const generic_id_t inside_id = id; + const generic_id_t outside_id = id; + + char mapping[64]; + const ssize_t len = + base::strings::SafeSPrintf(mapping, "%d %d 1\n", inside_id, outside_id); + const ssize_t rc = HANDLE_EINTR(write(fd, mapping, len)); + RAW_CHECK(IGNORE_EINTR(close(fd)) == 0); + return rc == len; +} + +// static +bool NamespaceUtils::KernelSupportsUnprivilegedNamespace(int type) { + // Valgrind will let clone(2) pass-through, but doesn't support unshare(), + // so always consider namespaces unsupported there. + if (IsRunningOnValgrind()) { + return false; + } + + // As of Linux 3.8, /proc/self/ns/* files exist for all namespace types. Since + // user namespaces were added in 3.8, it is OK to rely on the existence of + // /proc/self/ns/*. + if (!base::PathExists(base::FilePath("/proc/self/ns/user"))) { + return false; + } + + const char* path; + switch (type) { + case CLONE_NEWUSER: + return true; + case CLONE_NEWIPC: + path = "/proc/self/ns/ipc"; + break; + case CLONE_NEWNET: + path = "/proc/self/ns/net"; + break; + case CLONE_NEWNS: + path = "/proc/self/ns/mnt"; + break; + case CLONE_NEWPID: + path = "/proc/self/ns/pid"; + break; + case CLONE_NEWUTS: + path = "/proc/self/ns/uts"; + break; + default: + NOTREACHED(); + return false; + } + + return base::PathExists(base::FilePath(path)); +} + +// static +bool NamespaceUtils::KernelSupportsDenySetgroups() { + return base::PathExists(base::FilePath(kProcSelfSetgroups)); +} + +// static +bool NamespaceUtils::DenySetgroups() { + // This function needs to be async-signal-safe. + int fd = HANDLE_EINTR(open(kProcSelfSetgroups, O_WRONLY)); + if (fd == -1) { + return false; + } + + static const char kDeny[] = "deny"; + const ssize_t len = sizeof(kDeny) - 1; + const ssize_t rc = HANDLE_EINTR(write(fd, kDeny, len)); + RAW_CHECK(IGNORE_EINTR(close(fd)) == 0); + return rc == len; +} + +} // namespace sandbox diff --git a/sandbox/linux/services/namespace_utils.h b/sandbox/linux/services/namespace_utils.h new file mode 100644 index 0000000000..f3c88a9452 --- /dev/null +++ b/sandbox/linux/services/namespace_utils.h @@ -0,0 +1,53 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_NAMESPACE_UTILS_H_ +#define SANDBOX_LINUX_SERVICES_NAMESPACE_UTILS_H_ + +#include <sys/types.h> + +#include "base/compiler_specific.h" +#include "base/macros.h" +#include "base/template_util.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// Utility functions for using Linux namepaces. +class SANDBOX_EXPORT NamespaceUtils { + public: + COMPILE_ASSERT((base::is_same<uid_t, gid_t>::value), UidAndGidAreSameType); + // generic_id_t can be used for either uid_t or gid_t. + typedef uid_t generic_id_t; + + // Write a uid or gid mapping from |id| to |id| in |map_file|. This function + // is async-signal-safe. + static bool WriteToIdMapFile(const char* map_file, + generic_id_t id) WARN_UNUSED_RESULT; + + // Returns true if unprivileged namespaces of type |type| is supported + // (meaning that both CLONE_NEWUSER and type are are supported). |type| must + // be one of CLONE_NEWIPC, CLONE_NEWNET, CLONE_NEWNS, CLONE_NEWPID, + // CLONE_NEWUSER, or CLONE_NEWUTS. This relies on access to /proc, so it will + // not work from within a sandbox. + static bool KernelSupportsUnprivilegedNamespace(int type); + + // Returns true if the kernel supports denying setgroups in a user namespace. + // On kernels where this is supported, DenySetgroups must be called before a + // gid mapping can be added. + static bool KernelSupportsDenySetgroups(); + + // Disables setgroups() within the current user namespace. On Linux 3.18.2 and + // later, this is required in order to write to /proc/self/gid_map without + // having CAP_SETGID. Callers can determine whether is this needed with + // KernelSupportsDenySetgroups. This function is async-signal-safe. + static bool DenySetgroups() WARN_UNUSED_RESULT; + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(NamespaceUtils); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_NAMESPACE_UTILS_H_ diff --git a/sandbox/linux/services/namespace_utils_unittest.cc b/sandbox/linux/services/namespace_utils_unittest.cc new file mode 100644 index 0000000000..41ed7e89a6 --- /dev/null +++ b/sandbox/linux/services/namespace_utils_unittest.cc @@ -0,0 +1,72 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/namespace_utils.h" + +#include <errno.h> +#include <sched.h> +#include <sys/types.h> +#include <sys/wait.h> + +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "base/process/launch.h" +#include "sandbox/linux/services/credentials.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +SANDBOX_TEST(NamespaceUtils, KernelSupportsUnprivilegedNamespace) { + const bool can_create_user_ns = Credentials::CanCreateProcessInNewUserNS(); + const bool supports_user_ns = + NamespaceUtils::KernelSupportsUnprivilegedNamespace(CLONE_NEWUSER); + // can_create_user_ns implies supports_user_ns, but the converse is not + // necessarily true, as creating a user namespace can fail for various + // reasons. + if (can_create_user_ns) { + SANDBOX_ASSERT(supports_user_ns); + } +} + +SANDBOX_TEST(NamespaceUtils, WriteToIdMapFile) { + if (!Credentials::CanCreateProcessInNewUserNS()) { + return; + } + + const uid_t uid = getuid(); + const gid_t gid = getgid(); + + const bool supports_deny_setgroups = + NamespaceUtils::KernelSupportsDenySetgroups(); + + const pid_t pid = + base::ForkWithFlags(CLONE_NEWUSER | SIGCHLD, nullptr, nullptr); + ASSERT_NE(-1, pid); + if (pid == 0) { + if (supports_deny_setgroups) { + RAW_CHECK(NamespaceUtils::DenySetgroups()); + } + + RAW_CHECK(getuid() != uid); + RAW_CHECK(NamespaceUtils::WriteToIdMapFile("/proc/self/uid_map", uid)); + RAW_CHECK(getuid() == uid); + + RAW_CHECK(getgid() != gid); + RAW_CHECK(NamespaceUtils::WriteToIdMapFile("/proc/self/gid_map", gid)); + RAW_CHECK(getgid() == gid); + + _exit(0); + } + + int status = 42; + SANDBOX_ASSERT_EQ(pid, HANDLE_EINTR(waitpid(pid, &status, 0))); + SANDBOX_ASSERT_EQ(0, status); +} + +} // namespace. + +} // namespace sandbox. diff --git a/sandbox/linux/services/proc_util.cc b/sandbox/linux/services/proc_util.cc new file mode 100644 index 0000000000..d3f755f9a1 --- /dev/null +++ b/sandbox/linux/services/proc_util.cc @@ -0,0 +1,119 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/proc_util.h" + +#include <dirent.h> +#include <errno.h> +#include <fcntl.h> +#include <string.h> +#include <sys/stat.h> +#include <sys/types.h> + +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "base/posix/eintr_wrapper.h" +#include "base/strings/string_number_conversions.h" + +namespace sandbox { +namespace { + +struct DIRCloser { + void operator()(DIR* d) const { + DCHECK(d); + PCHECK(0 == closedir(d)); + } +}; + +typedef scoped_ptr<DIR, DIRCloser> ScopedDIR; + +base::ScopedFD OpenDirectory(const char* path) { + DCHECK(path); + base::ScopedFD directory_fd( + HANDLE_EINTR(open(path, O_RDONLY | O_DIRECTORY | O_CLOEXEC))); + PCHECK(directory_fd.is_valid()); + return directory_fd.Pass(); +} + +} // namespace + +int ProcUtil::CountOpenFds(int proc_fd) { + DCHECK_LE(0, proc_fd); + int proc_self_fd = HANDLE_EINTR( + openat(proc_fd, "self/fd/", O_DIRECTORY | O_RDONLY | O_CLOEXEC)); + PCHECK(0 <= proc_self_fd); + + // Ownership of proc_self_fd is transferred here, it must not be closed + // or modified afterwards except via dir. + ScopedDIR dir(fdopendir(proc_self_fd)); + CHECK(dir); + + int count = 0; + struct dirent e; + struct dirent* de; + while (!readdir_r(dir.get(), &e, &de) && de) { + if (strcmp(e.d_name, ".") == 0 || strcmp(e.d_name, "..") == 0) { + continue; + } + + int fd_num; + CHECK(base::StringToInt(e.d_name, &fd_num)); + if (fd_num == proc_fd || fd_num == proc_self_fd) { + continue; + } + + ++count; + } + return count; +} + +bool ProcUtil::HasOpenDirectory(int proc_fd) { + DCHECK_LE(0, proc_fd); + int proc_self_fd = + openat(proc_fd, "self/fd/", O_DIRECTORY | O_RDONLY | O_CLOEXEC); + + PCHECK(0 <= proc_self_fd); + + // Ownership of proc_self_fd is transferred here, it must not be closed + // or modified afterwards except via dir. + ScopedDIR dir(fdopendir(proc_self_fd)); + CHECK(dir); + + struct dirent e; + struct dirent* de; + while (!readdir_r(dir.get(), &e, &de) && de) { + if (strcmp(e.d_name, ".") == 0 || strcmp(e.d_name, "..") == 0) { + continue; + } + + int fd_num; + CHECK(base::StringToInt(e.d_name, &fd_num)); + if (fd_num == proc_fd || fd_num == proc_self_fd) { + continue; + } + + struct stat s; + // It's OK to use proc_self_fd here, fstatat won't modify it. + CHECK(fstatat(proc_self_fd, e.d_name, &s, 0) == 0); + if (S_ISDIR(s.st_mode)) { + return true; + } + } + + // No open unmanaged directories found. + return false; +} + +bool ProcUtil::HasOpenDirectory() { + base::ScopedFD proc_fd( + HANDLE_EINTR(open("/proc/", O_DIRECTORY | O_RDONLY | O_CLOEXEC))); + return HasOpenDirectory(proc_fd.get()); +} + +//static +base::ScopedFD ProcUtil::OpenProc() { + return OpenDirectory("/proc/"); +} + +} // namespace sandbox diff --git a/sandbox/linux/services/proc_util.h b/sandbox/linux/services/proc_util.h new file mode 100644 index 0000000000..bc14c5ef2a --- /dev/null +++ b/sandbox/linux/services/proc_util.h @@ -0,0 +1,42 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_PROC_UTIL_H_ +#define SANDBOX_LINUX_SERVICES_PROC_UTIL_H_ + +#include "base/files/scoped_file.h" +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +class SANDBOX_EXPORT ProcUtil { + public: + // Returns the number of file descriptors in the current process's FD + // table, excluding |proc_fd|, which should be a file descriptor for + // /proc/. + static int CountOpenFds(int proc_fd); + + // Checks whether the current process has any directory file descriptor open. + // Directory file descriptors are "capabilities" that would let a process use + // system calls such as openat() to bypass restrictions such as + // DropFileSystemAccess(). + // Sometimes it's useful to call HasOpenDirectory() after file system access + // has been dropped. In this case, |proc_fd| should be a file descriptor to + // /proc/. The file descriptor in |proc_fd| will be ignored by + // HasOpenDirectory() and remains owned by the caller. It is very important + // for the caller to close it. + static bool HasOpenDirectory(int proc_fd) WARN_UNUSED_RESULT; + static bool HasOpenDirectory() WARN_UNUSED_RESULT; + + // Open /proc/ or crash if not possible. + static base::ScopedFD OpenProc(); + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(ProcUtil); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_PROC_UTIL_H_ diff --git a/sandbox/linux/services/proc_util_unittest.cc b/sandbox/linux/services/proc_util_unittest.cc new file mode 100644 index 0000000000..bf25151956 --- /dev/null +++ b/sandbox/linux/services/proc_util_unittest.cc @@ -0,0 +1,62 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/proc_util.h" + +#include <fcntl.h> +#include <unistd.h> + +#include "base/files/scoped_file.h" +#include "base/posix/eintr_wrapper.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +TEST(ProcUtil, CountOpenFds) { + base::ScopedFD proc_fd(open("/proc/", O_RDONLY | O_DIRECTORY)); + ASSERT_TRUE(proc_fd.is_valid()); + int fd_count = ProcUtil::CountOpenFds(proc_fd.get()); + int fd = open("/dev/null", O_RDONLY); + ASSERT_LE(0, fd); + EXPECT_EQ(fd_count + 1, ProcUtil::CountOpenFds(proc_fd.get())); + ASSERT_EQ(0, IGNORE_EINTR(close(fd))); + EXPECT_EQ(fd_count, ProcUtil::CountOpenFds(proc_fd.get())); +} + +TEST(ProcUtil, HasOpenDirectory) { + // No open directory should exist at startup. + EXPECT_FALSE(ProcUtil::HasOpenDirectory()); + { + // Have a "/proc" file descriptor around. + int proc_fd = open("/proc/", O_RDONLY | O_DIRECTORY); + base::ScopedFD proc_fd_closer(proc_fd); + EXPECT_TRUE(ProcUtil::HasOpenDirectory()); + } + EXPECT_FALSE(ProcUtil::HasOpenDirectory()); +} + +TEST(ProcUtil, HasOpenDirectoryWithFD) { + int proc_fd = open("/proc/", O_RDONLY | O_DIRECTORY); + base::ScopedFD proc_fd_closer(proc_fd); + ASSERT_LE(0, proc_fd); + + // Don't pass |proc_fd|, an open directory (proc_fd) should + // be detected. + EXPECT_TRUE(ProcUtil::HasOpenDirectory()); + // Pass |proc_fd| and no open directory should be detected. + EXPECT_FALSE(ProcUtil::HasOpenDirectory(proc_fd)); + + { + // Have a directory file descriptor around. + int open_directory_fd = open("/proc/self/", O_RDONLY | O_DIRECTORY); + base::ScopedFD open_directory_fd_closer(open_directory_fd); + EXPECT_TRUE(ProcUtil::HasOpenDirectory(proc_fd)); + } + + // The "/proc/" file descriptor should now be closed, |proc_fd| is the + // only directory file descriptor open. + EXPECT_FALSE(ProcUtil::HasOpenDirectory(proc_fd)); +} + +} // namespace sandbox diff --git a/sandbox/linux/services/resource_limits.cc b/sandbox/linux/services/resource_limits.cc new file mode 100644 index 0000000000..1ec11295d1 --- /dev/null +++ b/sandbox/linux/services/resource_limits.cc @@ -0,0 +1,26 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/resource_limits.h" + +#include <sys/resource.h> +#include <sys/time.h> + +#include <algorithm> + +namespace sandbox { + +// static +bool ResourceLimits::Lower(int resource, rlim_t limit) { + struct rlimit old_rlimit; + if (getrlimit(resource, &old_rlimit)) + return false; + // Make sure we don't raise the existing limit. + const struct rlimit new_rlimit = {std::min(old_rlimit.rlim_cur, limit), + std::min(old_rlimit.rlim_max, limit)}; + int rc = setrlimit(resource, &new_rlimit); + return rc == 0; +} + +} // namespace sandbox diff --git a/sandbox/linux/services/resource_limits.h b/sandbox/linux/services/resource_limits.h new file mode 100644 index 0000000000..3464dab679 --- /dev/null +++ b/sandbox/linux/services/resource_limits.h @@ -0,0 +1,29 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_RESOURCE_LIMITS_H_ +#define SANDBOX_LINUX_SERVICES_RESOURCE_LIMITS_H_ + +#include <sys/resource.h> + +#include "base/compiler_specific.h" +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// This class provides a small wrapper around setrlimit(). +class SANDBOX_EXPORT ResourceLimits { + public: + // Lower the soft and hard limit of |resource| to |limit|. If the current + // limit is lower than |limit|, keep it. + static bool Lower(int resource, rlim_t limit) WARN_UNUSED_RESULT; + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(ResourceLimits); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_RESOURCE_LIMITS_H_ diff --git a/sandbox/linux/services/resource_limits_unittests.cc b/sandbox/linux/services/resource_limits_unittests.cc new file mode 100644 index 0000000000..910c740f7b --- /dev/null +++ b/sandbox/linux/services/resource_limits_unittests.cc @@ -0,0 +1,43 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/resource_limits.h" + +#include <errno.h> +#include <sys/resource.h> +#include <sys/time.h> +#include <unistd.h> + +#include "base/logging.h" +#include "sandbox/linux/tests/test_utils.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +// Fails on Android: crbug.com/459158 +#if !defined(OS_ANDROID) +#define MAYBE_NoFork DISABLE_ON_ASAN(NoFork) +#else +#define MAYBE_NoFork DISABLED_NoFork +#endif // OS_ANDROID + +// Not being able to fork breaks LeakSanitizer, so disable on +// all ASAN builds. +SANDBOX_TEST(ResourceLimits, MAYBE_NoFork) { + // Make sure that fork will fail with EAGAIN. + SANDBOX_ASSERT(ResourceLimits::Lower(RLIMIT_NPROC, 0)); + errno = 0; + pid_t pid = fork(); + // Reap any child if fork succeeded. + TestUtils::HandlePostForkReturn(pid); + SANDBOX_ASSERT_EQ(-1, pid); + CHECK_EQ(EAGAIN, errno); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/services/scoped_process.cc b/sandbox/linux/services/scoped_process.cc new file mode 100644 index 0000000000..65af4873a4 --- /dev/null +++ b/sandbox/linux/services/scoped_process.cc @@ -0,0 +1,119 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/scoped_process.h" + +#include <fcntl.h> +#include <signal.h> +#include <sys/stat.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include "base/callback.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "build/build_config.h" +#include "sandbox/linux/services/syscall_wrappers.h" +#include "sandbox/linux/services/thread_helpers.h" + +namespace sandbox { + +namespace { + +const char kSynchronisationChar[] = "D"; + +void WaitForever() { + while(true) { + pause(); + } +} + +} // namespace + +ScopedProcess::ScopedProcess(const base::Closure& child_callback) + : child_process_id_(-1), process_id_(getpid()) { + PCHECK(0 == pipe(pipe_fds_)); +#if !defined(THREAD_SANITIZER) + // Make sure that we can safely fork(). + CHECK(ThreadHelpers::IsSingleThreaded()); +#endif + child_process_id_ = fork(); + PCHECK(0 <= child_process_id_); + + if (0 == child_process_id_) { + PCHECK(0 == IGNORE_EINTR(close(pipe_fds_[0]))); + pipe_fds_[0] = -1; + child_callback.Run(); + // Notify the parent that the closure has run. + CHECK_EQ(1, HANDLE_EINTR(write(pipe_fds_[1], kSynchronisationChar, 1))); + WaitForever(); + NOTREACHED(); + _exit(1); + } + + PCHECK(0 == IGNORE_EINTR(close(pipe_fds_[1]))); + pipe_fds_[1] = -1; +} + +ScopedProcess::~ScopedProcess() { + CHECK(IsOriginalProcess()); + if (child_process_id_ >= 0) { + PCHECK(0 == kill(child_process_id_, SIGKILL)); + siginfo_t process_info; + + PCHECK(0 == HANDLE_EINTR( + waitid(P_PID, child_process_id_, &process_info, WEXITED))); + } + if (pipe_fds_[0] >= 0) { + PCHECK(0 == IGNORE_EINTR(close(pipe_fds_[0]))); + } + if (pipe_fds_[1] >= 0) { + PCHECK(0 == IGNORE_EINTR(close(pipe_fds_[1]))); + } +} + +int ScopedProcess::WaitForExit(bool* got_signaled) { + DCHECK(got_signaled); + CHECK(IsOriginalProcess()); + siginfo_t process_info; + // WNOWAIT to make sure that the destructor can wait on the child. + int ret = HANDLE_EINTR( + waitid(P_PID, child_process_id_, &process_info, WEXITED | WNOWAIT)); + PCHECK(0 == ret) << "Did something else wait on the child?"; + + if (process_info.si_code == CLD_EXITED) { + *got_signaled = false; + } else if (process_info.si_code == CLD_KILLED || + process_info.si_code == CLD_DUMPED) { + *got_signaled = true; + } else { + CHECK(false) << "ScopedProcess needs to be extended for si_code " + << process_info.si_code; + } + return process_info.si_status; +} + +bool ScopedProcess::WaitForClosureToRun() { + char c = 0; + int ret = HANDLE_EINTR(read(pipe_fds_[0], &c, 1)); + PCHECK(ret >= 0); + if (0 == ret) + return false; + + CHECK_EQ(c, kSynchronisationChar[0]); + return true; +} + +// It would be problematic if after a fork(), another process would start using +// this object. +// This method allows to assert it is not happening. +bool ScopedProcess::IsOriginalProcess() { + // Make a direct syscall to bypass glibc caching of PIDs. + pid_t pid = sys_getpid(); + return pid == process_id_; +} + +} // namespace sandbox diff --git a/sandbox/linux/services/scoped_process.h b/sandbox/linux/services/scoped_process.h new file mode 100644 index 0000000000..bddbd5529b --- /dev/null +++ b/sandbox/linux/services/scoped_process.h @@ -0,0 +1,55 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_SCOPED_PROCESS_H_ +#define SANDBOX_LINUX_SERVICES_SCOPED_PROCESS_H_ + +#include "base/callback_forward.h" +#include "base/macros.h" +#include "base/process/process_handle.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// fork() a child process that will run a Closure. +// After the Closure has run, the child will pause forever. If this object +// is detroyed, the child will be destroyed, even if the closure did not +// finish running. It's ok to signal the child from outside of this class to +// destroy it. +// This class cannot be instanciated from a multi-threaded process, as it needs +// to fork(). +class SANDBOX_EXPORT ScopedProcess { + public: + // A new process will be created and |child_callback| will run in the child + // process. This callback is allowed to terminate the process or to simply + // return. If the callback returns, the process will wait forever. + explicit ScopedProcess(const base::Closure& child_callback); + ~ScopedProcess(); + + // Wait for the process to exit. + // |got_signaled| tells how to interpret the return value: either as an exit + // code, or as a signal number. + // When this returns, the process will still not have been reaped and will + // survive as a zombie for the lifetime of this object. This method can be + // called multiple times. + int WaitForExit(bool* got_signaled); + + // Wait for the |child_callback| passed at construction to run. Return false + // if |child_callback| did not finish running and we know it never will (for + // instance the child crashed or used _exit()). + bool WaitForClosureToRun(); + base::ProcessId GetPid() { return child_process_id_; } + + private: + bool IsOriginalProcess(); + + base::ProcessId child_process_id_; + base::ProcessId process_id_; + int pipe_fds_[2]; + DISALLOW_COPY_AND_ASSIGN(ScopedProcess); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_SCOPED_PROCESS_H_ diff --git a/sandbox/linux/services/scoped_process_unittest.cc b/sandbox/linux/services/scoped_process_unittest.cc new file mode 100644 index 0000000000..8bd2847997 --- /dev/null +++ b/sandbox/linux/services/scoped_process_unittest.cc @@ -0,0 +1,130 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/scoped_process.h" + +#include <errno.h> +#include <fcntl.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include "base/bind.h" +#include "base/callback.h" +#include "base/files/file_util.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "base/threading/platform_thread.h" +#include "base/time/time.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +void DoExit() { _exit(0); } + +void ExitWithCode(int exit_code) { _exit(exit_code); } + +void RaiseAndExit(int signal) { + PCHECK(0 == raise(signal)); + _exit(0); +} + +void DoNothing() {} + +TEST(ScopedProcess, ScopedProcessNormalExit) { + const int kCustomExitCode = 12; + ScopedProcess process(base::Bind(&ExitWithCode, kCustomExitCode)); + bool got_signaled = true; + int exit_code = process.WaitForExit(&got_signaled); + EXPECT_FALSE(got_signaled); + EXPECT_EQ(kCustomExitCode, exit_code); + + // Verify that WaitForExit() can be called multiple times on the same + // process. + bool got_signaled2 = true; + int exit_code2 = process.WaitForExit(&got_signaled2); + EXPECT_FALSE(got_signaled2); + EXPECT_EQ(kCustomExitCode, exit_code2); +} + +// Disable this test on Android, SIGABRT is funky there. +TEST(ScopedProcess, DISABLE_ON_ANDROID(ScopedProcessAbort)) { + PCHECK(SIG_ERR != signal(SIGABRT, SIG_DFL)); + ScopedProcess process(base::Bind(&RaiseAndExit, SIGABRT)); + bool got_signaled = false; + int exit_code = process.WaitForExit(&got_signaled); + EXPECT_TRUE(got_signaled); + EXPECT_EQ(SIGABRT, exit_code); +} + +TEST(ScopedProcess, ScopedProcessSignaled) { + ScopedProcess process(base::Bind(&DoNothing)); + bool got_signaled = false; + ASSERT_EQ(0, kill(process.GetPid(), SIGKILL)); + int exit_code = process.WaitForExit(&got_signaled); + EXPECT_TRUE(got_signaled); + EXPECT_EQ(SIGKILL, exit_code); +} + +TEST(ScopedProcess, DiesForReal) { + int pipe_fds[2]; + ASSERT_EQ(0, pipe(pipe_fds)); + base::ScopedFD read_end_closer(pipe_fds[0]); + base::ScopedFD write_end_closer(pipe_fds[1]); + + { ScopedProcess process(base::Bind(&DoExit)); } + + // Close writing end of the pipe. + write_end_closer.reset(); + pipe_fds[1] = -1; + + ASSERT_EQ(0, fcntl(pipe_fds[0], F_SETFL, O_NONBLOCK)); + char c; + // If the child process is dead for real, there will be no writing end + // for this pipe left and read will EOF instead of returning EWOULDBLOCK. + ASSERT_EQ(0, read(pipe_fds[0], &c, 1)); +} + +TEST(ScopedProcess, SynchronizationBasic) { + ScopedProcess process1(base::Bind(&DoNothing)); + EXPECT_TRUE(process1.WaitForClosureToRun()); + + ScopedProcess process2(base::Bind(&DoExit)); + // The closure didn't finish running normally. This case is simple enough + // that process.WaitForClosureToRun() should return false, even though the + // API does not guarantees that it will return at all. + EXPECT_FALSE(process2.WaitForClosureToRun()); +} + +void SleepInMsAndWriteOneByte(int time_to_sleep, int fd) { + base::PlatformThread::Sleep(base::TimeDelta::FromMilliseconds(time_to_sleep)); + CHECK(1 == write(fd, "1", 1)); +} + +TEST(ScopedProcess, SynchronizationWorks) { + int pipe_fds[2]; + ASSERT_EQ(0, pipe(pipe_fds)); + base::ScopedFD read_end_closer(pipe_fds[0]); + base::ScopedFD write_end_closer(pipe_fds[1]); + + // Start a process with a closure that takes a little bit to run. + ScopedProcess process( + base::Bind(&SleepInMsAndWriteOneByte, 100, pipe_fds[1])); + EXPECT_TRUE(process.WaitForClosureToRun()); + + // Verify that the closure did, indeed, run. + ASSERT_EQ(0, fcntl(pipe_fds[0], F_SETFL, O_NONBLOCK)); + char c = 0; + EXPECT_EQ(1, read(pipe_fds[0], &c, 1)); + EXPECT_EQ('1', c); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/services/syscall_wrappers.cc b/sandbox/linux/services/syscall_wrappers.cc new file mode 100644 index 0000000000..264eb6842d --- /dev/null +++ b/sandbox/linux/services/syscall_wrappers.cc @@ -0,0 +1,246 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/syscall_wrappers.h" + +#include <pthread.h> +#include <sched.h> +#include <setjmp.h> +#include <sys/resource.h> +#include <sys/syscall.h> +#include <sys/time.h> +#include <sys/types.h> +#include <unistd.h> +#include <cstring> + +#include "base/compiler_specific.h" +#include "base/logging.h" +#include "base/third_party/valgrind/valgrind.h" +#include "build/build_config.h" +#include "sandbox/linux/system_headers/capability.h" +#include "sandbox/linux/system_headers/linux_signal.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" + +namespace sandbox { + +pid_t sys_getpid(void) { + return syscall(__NR_getpid); +} + +pid_t sys_gettid(void) { + return syscall(__NR_gettid); +} + +long sys_clone(unsigned long flags, + decltype(nullptr) child_stack, + pid_t* ptid, + pid_t* ctid, + decltype(nullptr) tls) { + const bool clone_tls_used = flags & CLONE_SETTLS; + const bool invalid_ctid = + (flags & (CLONE_CHILD_SETTID | CLONE_CHILD_CLEARTID)) && !ctid; + const bool invalid_ptid = (flags & CLONE_PARENT_SETTID) && !ptid; + + // We do not support CLONE_VM. + const bool clone_vm_used = flags & CLONE_VM; + if (clone_tls_used || invalid_ctid || invalid_ptid || clone_vm_used) { + RAW_LOG(FATAL, "Invalid usage of sys_clone"); + } + + if (ptid) MSAN_UNPOISON(ptid, sizeof(*ptid)); + if (ctid) MSAN_UNPOISON(ctid, sizeof(*ctid)); + // See kernel/fork.c in Linux. There is different ordering of sys_clone + // parameters depending on CONFIG_CLONE_BACKWARDS* configuration options. +#if defined(ARCH_CPU_X86_64) + return syscall(__NR_clone, flags, child_stack, ptid, ctid, tls); +#elif defined(ARCH_CPU_X86) || defined(ARCH_CPU_ARM_FAMILY) || \ + defined(ARCH_CPU_MIPS_FAMILY) || defined(ARCH_CPU_MIPS64_FAMILY) + // CONFIG_CLONE_BACKWARDS defined. + return syscall(__NR_clone, flags, child_stack, ptid, tls, ctid); +#endif +} + +long sys_clone(unsigned long flags) { + return sys_clone(flags, nullptr, nullptr, nullptr, nullptr); +} + +void sys_exit_group(int status) { + syscall(__NR_exit_group, status); +} + +int sys_seccomp(unsigned int operation, + unsigned int flags, + const struct sock_fprog* args) { + return syscall(__NR_seccomp, operation, flags, args); +} + +int sys_prlimit64(pid_t pid, + int resource, + const struct rlimit64* new_limit, + struct rlimit64* old_limit) { + int res = syscall(__NR_prlimit64, pid, resource, new_limit, old_limit); + if (res == 0 && old_limit) MSAN_UNPOISON(old_limit, sizeof(*old_limit)); + return res; +} + +int sys_capget(cap_hdr* hdrp, cap_data* datap) { + int res = syscall(__NR_capget, hdrp, datap); + if (res == 0) { + if (hdrp) MSAN_UNPOISON(hdrp, sizeof(*hdrp)); + if (datap) MSAN_UNPOISON(datap, sizeof(*datap)); + } + return res; +} + +int sys_capset(cap_hdr* hdrp, const cap_data* datap) { + return syscall(__NR_capset, hdrp, datap); +} + +int sys_getresuid(uid_t* ruid, uid_t* euid, uid_t* suid) { + int res; +#if defined(ARCH_CPU_X86) || defined(ARCH_CPU_ARMEL) + // On 32-bit x86 or 32-bit arm, getresuid supports 16bit values only. + // Use getresuid32 instead. + res = syscall(__NR_getresuid32, ruid, euid, suid); +#else + res = syscall(__NR_getresuid, ruid, euid, suid); +#endif + if (res == 0) { + if (ruid) MSAN_UNPOISON(ruid, sizeof(*ruid)); + if (euid) MSAN_UNPOISON(euid, sizeof(*euid)); + if (suid) MSAN_UNPOISON(suid, sizeof(*suid)); + } + return res; +} + +int sys_getresgid(gid_t* rgid, gid_t* egid, gid_t* sgid) { + int res; +#if defined(ARCH_CPU_X86) || defined(ARCH_CPU_ARMEL) + // On 32-bit x86 or 32-bit arm, getresgid supports 16bit values only. + // Use getresgid32 instead. + res = syscall(__NR_getresgid32, rgid, egid, sgid); +#else + res = syscall(__NR_getresgid, rgid, egid, sgid); +#endif + if (res == 0) { + if (rgid) MSAN_UNPOISON(rgid, sizeof(*rgid)); + if (egid) MSAN_UNPOISON(egid, sizeof(*egid)); + if (sgid) MSAN_UNPOISON(sgid, sizeof(*sgid)); + } + return res; +} + +int sys_chroot(const char* path) { + return syscall(__NR_chroot, path); +} + +int sys_unshare(int flags) { + return syscall(__NR_unshare, flags); +} + +int sys_sigprocmask(int how, const sigset_t* set, decltype(nullptr) oldset) { + // In some toolchain (in particular Android and PNaCl toolchain), + // sigset_t is 32 bits, but Linux ABI requires 64 bits. + uint64_t linux_value = 0; + std::memcpy(&linux_value, set, std::min(sizeof(sigset_t), sizeof(uint64_t))); + return syscall(__NR_rt_sigprocmask, how, &linux_value, nullptr, + sizeof(linux_value)); +} + +#if (defined(MEMORY_SANITIZER) || defined(THREAD_SANITIZER) || \ + (defined(ARCH_CPU_X86_64) && !defined(__clang__))) && \ + !defined(OS_NACL_NONSFI) +// If MEMORY_SANITIZER or THREAD_SANITIZER is enabled, it is necessary to call +// sigaction() here, rather than the direct syscall (sys_sigaction() defined +// by ourselves). +// It is because, if MEMORY_SANITIZER or THREAD_SANITIZER is enabled, sigaction +// is wrapped, and |act->sa_handler| is injected in order to unpoisonize the +// memory passed via callback's arguments for MEMORY_SANITIZER, or handle +// signals to check thread consistency for THREAD_SANITIZER. Please see +// msan_interceptors.cc and tsan_interceptors.cc for more details. +// So, specifically, if MEMORY_SANITIZER is enabled while the direct syscall is +// used, as MEMORY_SANITIZER does not know about it, sigaction() invocation in +// other places would be broken (in more precise, returned |oldact| would have +// a broken |sa_handler| callback). +// Practically, it would break NaCl's signal handler installation. +// cf) native_client/src/trusted/service_runtime/linux/nacl_signal.c. +// As for THREAD_SANITIZER, the intercepted signal handlers are processed more +// in other libc functions' interceptors (such as for raise()), so that it +// would not work properly. +// +// Also on x86_64 architecture, we need naked function for rt_sigreturn. +// However, there is no simple way to define it with GCC. Note that the body +// of function is actually very small (only two instructions), but we need to +// define much debug information in addition, otherwise backtrace() used by +// base::StackTrace would not work so that some tests would fail. +// +// When this is built with PNaCl toolchain, we should always use sys_sigaction +// below, because sigaction() provided by the toolchain is incompatible with +// Linux's ABI. So, otherwise, it would just fail. Note that it is not +// necessary to think about sigaction() invocation in other places even with +// MEMORY_SANITIZER or THREAD_SANITIZER, because it would just fail there. +int sys_sigaction(int signum, + const struct sigaction* act, + struct sigaction* oldact) { + return sigaction(signum, act, oldact); +} +#else +// struct sigaction is different ABI from the Linux's. +struct KernelSigAction { + void (*kernel_handler)(int); + uint32_t sa_flags; + void (*sa_restorer)(void); + uint64_t sa_mask; +}; + +// On X86_64 arch, it is necessary to set sa_restorer always. +#if defined(ARCH_CPU_X86_64) +#if !defined(SA_RESTORER) +#define SA_RESTORER 0x04000000 +#endif + +// rt_sigreturn is a special system call that interacts with the user land +// stack. Thus, here prologue must not be created, which implies syscall() +// does not work properly, too. Note that rt_sigreturn will never return. +static __attribute__((naked)) void sys_rt_sigreturn() { + // Just invoke rt_sigreturn system call. + asm volatile ("syscall\n" + :: "a"(__NR_rt_sigreturn)); +} +#endif + +int sys_sigaction(int signum, + const struct sigaction* act, + struct sigaction* oldact) { + KernelSigAction kernel_act = {}; + if (act) { + kernel_act.kernel_handler = act->sa_handler; + std::memcpy(&kernel_act.sa_mask, &act->sa_mask, + std::min(sizeof(kernel_act.sa_mask), sizeof(act->sa_mask))); + kernel_act.sa_flags = act->sa_flags; + +#if defined(ARCH_CPU_X86_64) + if (!(kernel_act.sa_flags & SA_RESTORER)) { + kernel_act.sa_flags |= SA_RESTORER; + kernel_act.sa_restorer = sys_rt_sigreturn; + } +#endif + } + + KernelSigAction kernel_oldact = {}; + int result = syscall(__NR_rt_sigaction, signum, act ? &kernel_act : nullptr, + oldact ? &kernel_oldact : nullptr, sizeof(uint64_t)); + if (result == 0 && oldact) { + oldact->sa_handler = kernel_oldact.kernel_handler; + sigemptyset(&oldact->sa_mask); + std::memcpy(&oldact->sa_mask, &kernel_oldact.sa_mask, + std::min(sizeof(kernel_act.sa_mask), sizeof(act->sa_mask))); + oldact->sa_flags = kernel_oldact.sa_flags; + } + return result; +} + +#endif // defined(MEMORY_SANITIZER) + +} // namespace sandbox diff --git a/sandbox/linux/services/syscall_wrappers.h b/sandbox/linux/services/syscall_wrappers.h new file mode 100644 index 0000000000..581425a367 --- /dev/null +++ b/sandbox/linux/services/syscall_wrappers.h @@ -0,0 +1,83 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_SYSCALL_WRAPPERS_H_ +#define SANDBOX_LINUX_SERVICES_SYSCALL_WRAPPERS_H_ + +#include <signal.h> +#include <stdint.h> +#include <sys/types.h> + +#include "sandbox/sandbox_export.h" + +struct sock_fprog; +struct rlimit64; +struct cap_hdr; +struct cap_data; + +namespace sandbox { + +// Provide direct system call wrappers for a few common system calls. +// These are guaranteed to perform a system call and do not rely on things such +// as caching the current pid (c.f. getpid()) unless otherwise specified. + +SANDBOX_EXPORT pid_t sys_getpid(void); + +SANDBOX_EXPORT pid_t sys_gettid(void); + +SANDBOX_EXPORT long sys_clone(unsigned long flags); + +// |regs| is not supported and must be passed as nullptr. |child_stack| must be +// nullptr, since otherwise this function cannot safely return. As a +// consequence, this function does not support CLONE_VM. +SANDBOX_EXPORT long sys_clone(unsigned long flags, + decltype(nullptr) child_stack, + pid_t* ptid, + pid_t* ctid, + decltype(nullptr) regs); + +SANDBOX_EXPORT void sys_exit_group(int status); + +// The official system call takes |args| as void* (in order to be extensible), +// but add more typing for the cases that are currently used. +SANDBOX_EXPORT int sys_seccomp(unsigned int operation, + unsigned int flags, + const struct sock_fprog* args); + +// Some libcs do not expose a prlimit64 wrapper. +SANDBOX_EXPORT int sys_prlimit64(pid_t pid, + int resource, + const struct rlimit64* new_limit, + struct rlimit64* old_limit); + +// Some libcs do not expose capget/capset wrappers. We want to use these +// directly in order to avoid pulling in libcap2. +SANDBOX_EXPORT int sys_capget(struct cap_hdr* hdrp, struct cap_data* datap); +SANDBOX_EXPORT int sys_capset(struct cap_hdr* hdrp, + const struct cap_data* datap); + +// Some libcs do not expose getresuid/getresgid wrappers. +SANDBOX_EXPORT int sys_getresuid(uid_t* ruid, uid_t* euid, uid_t* suid); +SANDBOX_EXPORT int sys_getresgid(gid_t* rgid, gid_t* egid, gid_t* sgid); + +// Some libcs do not expose a chroot wrapper. +SANDBOX_EXPORT int sys_chroot(const char* path); + +// Some libcs do not expose a unshare wrapper. +SANDBOX_EXPORT int sys_unshare(int flags); + +// Some libcs do not expose a sigprocmask. Note that oldset must be a nullptr, +// because of some ABI gap between toolchain's and Linux's. +SANDBOX_EXPORT int sys_sigprocmask(int how, + const sigset_t* set, + decltype(nullptr) oldset); + +// Some libcs do not expose a sigaction(). +SANDBOX_EXPORT int sys_sigaction(int signum, + const struct sigaction* act, + struct sigaction* oldact); + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_SYSCALL_WRAPPERS_H_ diff --git a/sandbox/linux/services/syscall_wrappers_unittest.cc b/sandbox/linux/services/syscall_wrappers_unittest.cc new file mode 100644 index 0000000000..1878ff3fe7 --- /dev/null +++ b/sandbox/linux/services/syscall_wrappers_unittest.cc @@ -0,0 +1,99 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/syscall_wrappers.h" + +#include <sys/syscall.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> +#include <cstring> + +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "base/third_party/valgrind/valgrind.h" +#include "build/build_config.h" +#include "sandbox/linux/system_headers/linux_signal.h" +#include "sandbox/linux/tests/test_utils.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +TEST(SyscallWrappers, BasicSyscalls) { + EXPECT_EQ(getpid(), sys_getpid()); +} + +TEST(SyscallWrappers, CloneBasic) { + pid_t child = sys_clone(SIGCHLD); + TestUtils::HandlePostForkReturn(child); + EXPECT_LT(0, child); +} + +TEST(SyscallWrappers, CloneParentSettid) { + pid_t ptid = 0; + pid_t child = sys_clone(CLONE_PARENT_SETTID | SIGCHLD, nullptr, &ptid, + nullptr, nullptr); + TestUtils::HandlePostForkReturn(child); + EXPECT_LT(0, child); + EXPECT_EQ(child, ptid); +} + +TEST(SyscallWrappers, CloneChildSettid) { + pid_t ctid = 0; + pid_t pid = + sys_clone(CLONE_CHILD_SETTID | SIGCHLD, nullptr, nullptr, &ctid, nullptr); + + const int kSuccessExit = 0; + if (0 == pid) { + // In child. + if (sys_getpid() == ctid) + _exit(kSuccessExit); + _exit(1); + } + + ASSERT_NE(-1, pid); + int status = 0; + ASSERT_EQ(pid, HANDLE_EINTR(waitpid(pid, &status, 0))); + ASSERT_TRUE(WIFEXITED(status)); + EXPECT_EQ(kSuccessExit, WEXITSTATUS(status)); +} + +TEST(SyscallWrappers, GetRESUid) { + uid_t ruid, euid, suid; + uid_t sys_ruid, sys_euid, sys_suid; + ASSERT_EQ(0, getresuid(&ruid, &euid, &suid)); + ASSERT_EQ(0, sys_getresuid(&sys_ruid, &sys_euid, &sys_suid)); + EXPECT_EQ(ruid, sys_ruid); + EXPECT_EQ(euid, sys_euid); + EXPECT_EQ(suid, sys_suid); +} + +TEST(SyscallWrappers, GetRESGid) { + gid_t rgid, egid, sgid; + gid_t sys_rgid, sys_egid, sys_sgid; + ASSERT_EQ(0, getresgid(&rgid, &egid, &sgid)); + ASSERT_EQ(0, sys_getresgid(&sys_rgid, &sys_egid, &sys_sgid)); + EXPECT_EQ(rgid, sys_rgid); + EXPECT_EQ(egid, sys_egid); + EXPECT_EQ(sgid, sys_sgid); +} + +TEST(SyscallWrappers, LinuxSigSet) { + sigset_t sigset; + ASSERT_EQ(0, sigemptyset(&sigset)); + ASSERT_EQ(0, sigaddset(&sigset, LINUX_SIGSEGV)); + ASSERT_EQ(0, sigaddset(&sigset, LINUX_SIGBUS)); + uint64_t linux_sigset = 0; + std::memcpy(&linux_sigset, &sigset, + std::min(sizeof(sigset), sizeof(linux_sigset))); + EXPECT_EQ((1ULL << (LINUX_SIGSEGV - 1)) | (1ULL << (LINUX_SIGBUS - 1)), + linux_sigset); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/services/thread_helpers.cc b/sandbox/linux/services/thread_helpers.cc new file mode 100644 index 0000000000..80766a9bc5 --- /dev/null +++ b/sandbox/linux/services/thread_helpers.cc @@ -0,0 +1,157 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/thread_helpers.h" + +#include <errno.h> +#include <fcntl.h> +#include <signal.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <unistd.h> + +#include <string> + +#include "base/bind.h" +#include "base/callback.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "base/strings/string_number_conversions.h" +#include "base/threading/platform_thread.h" +#include "base/threading/thread.h" +#include "sandbox/linux/services/proc_util.h" + +namespace sandbox { + +namespace { + +const char kAssertSingleThreadedError[] = + "Current process is not mono-threaded!"; + +bool IsSingleThreadedImpl(int proc_fd) { + CHECK_LE(0, proc_fd); + struct stat task_stat; + int fstat_ret = fstatat(proc_fd, "self/task/", &task_stat, 0); + PCHECK(0 == fstat_ret); + + // At least "..", "." and the current thread should be present. + CHECK_LE(3UL, task_stat.st_nlink); + // Counting threads via /proc/self/task could be racy. For the purpose of + // determining if the current proces is monothreaded it works: if at any + // time it becomes monothreaded, it'll stay so. + return task_stat.st_nlink == 3; +} + +bool IsThreadPresentInProcFS(int proc_fd, + const std::string& thread_id_dir_str) { + struct stat task_stat; + const int fstat_ret = + fstatat(proc_fd, thread_id_dir_str.c_str(), &task_stat, 0); + if (fstat_ret < 0) { + PCHECK(ENOENT == errno); + return false; + } + return true; +} + +// Run |cb| in a loop until it returns false. Every time |cb| runs, sleep +// for an exponentially increasing amount of time. |cb| is expected to return +// false very quickly and this will crash if it doesn't happen within ~64ms on +// Debug builds (2s on Release builds). +// This is guaranteed to not sleep more than twice as much as the bare minimum +// amount of time. +void RunWhileTrue(const base::Callback<bool(void)>& cb) { +#if defined(NDEBUG) + // In Release mode, crash after 30 iterations, which means having spent + // roughly 2s in + // nanosleep(2) cumulatively. + const unsigned int kMaxIterations = 30U; +#else + // In practice, this never goes through more than a couple iterations. In + // debug mode, crash after 64ms (+ eventually 25 times the granularity of + // the clock) in nanosleep(2). This ensures that this is not becoming too + // slow. + const unsigned int kMaxIterations = 25U; +#endif + + // Run |cb| with an exponential back-off, sleeping 2^iterations nanoseconds + // in nanosleep(2). + // Note: the clock may not allow for nanosecond granularity, in this case the + // first iterations would sleep a tiny bit more instead, which would not + // change the calculations significantly. + for (unsigned int i = 0; i < kMaxIterations; ++i) { + if (!cb.Run()) { + return; + } + + // Increase the waiting time exponentially. + struct timespec ts = {0, 1L << i /* nanoseconds */}; + PCHECK(0 == HANDLE_EINTR(nanosleep(&ts, &ts))); + } + + LOG(FATAL) << kAssertSingleThreadedError << " (iterations: " << kMaxIterations + << ")"; + + NOTREACHED(); +} + +bool IsMultiThreaded(int proc_fd) { + return !ThreadHelpers::IsSingleThreaded(proc_fd); +} + +} // namespace + +// static +bool ThreadHelpers::IsSingleThreaded(int proc_fd) { + DCHECK_LE(0, proc_fd); + return IsSingleThreadedImpl(proc_fd); +} + +// static +bool ThreadHelpers::IsSingleThreaded() { + base::ScopedFD task_fd(ProcUtil::OpenProc()); + return IsSingleThreaded(task_fd.get()); +} + +// static +void ThreadHelpers::AssertSingleThreaded(int proc_fd) { + DCHECK_LE(0, proc_fd); + const base::Callback<bool(void)> cb = base::Bind(&IsMultiThreaded, proc_fd); + RunWhileTrue(cb); +} + +void ThreadHelpers::AssertSingleThreaded() { + base::ScopedFD task_fd(ProcUtil::OpenProc()); + AssertSingleThreaded(task_fd.get()); +} + +// static +bool ThreadHelpers::StopThreadAndWatchProcFS(int proc_fd, + base::Thread* thread) { + DCHECK_LE(0, proc_fd); + DCHECK(thread); + const base::PlatformThreadId thread_id = thread->thread_id(); + const std::string thread_id_dir_str = + "self/task/" + base::IntToString(thread_id) + "/"; + + // The kernel is at liberty to wake the thread id futex before updating + // /proc. Following Stop(), the thread is joined, but entries in /proc may + // not have been updated. + thread->Stop(); + + const base::Callback<bool(void)> cb = + base::Bind(&IsThreadPresentInProcFS, proc_fd, thread_id_dir_str); + + RunWhileTrue(cb); + + return true; +} + +// static +const char* ThreadHelpers::GetAssertSingleThreadedErrorMessageForTests() { + return kAssertSingleThreadedError; +} + +} // namespace sandbox diff --git a/sandbox/linux/services/thread_helpers.h b/sandbox/linux/services/thread_helpers.h new file mode 100644 index 0000000000..f4abdffd03 --- /dev/null +++ b/sandbox/linux/services/thread_helpers.h @@ -0,0 +1,43 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_THREAD_HELPERS_H_ +#define SANDBOX_LINUX_SERVICES_THREAD_HELPERS_H_ + +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +namespace base { class Thread; } + +namespace sandbox { + +class SANDBOX_EXPORT ThreadHelpers { + public: + // Check whether the current process is single threaded. |proc_fd| + // must be a file descriptor to /proc/ and remains owned by the + // caller. + static bool IsSingleThreaded(int proc_fd); + static bool IsSingleThreaded(); + + // Crash if the current process is not single threaded. This will wait + // on /proc to be updated. In the case where this doesn't crash, this will + // return promptly. In the case where this does crash, this will first wait + // for a few ms in Debug mode, a few seconds in Release mode. + static void AssertSingleThreaded(int proc_fd); + static void AssertSingleThreaded(); + + // Stop |thread| and ensure that it does not have an entry in + // /proc/self/task/ from the point of view of the current thread. This is + // the way to stop threads before calling IsSingleThreaded(). + static bool StopThreadAndWatchProcFS(int proc_fd, base::Thread* thread); + + static const char* GetAssertSingleThreadedErrorMessageForTests(); + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(ThreadHelpers); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_THREAD_HELPERS_H_ diff --git a/sandbox/linux/services/thread_helpers_unittests.cc b/sandbox/linux/services/thread_helpers_unittests.cc new file mode 100644 index 0000000000..7357a0cfa7 --- /dev/null +++ b/sandbox/linux/services/thread_helpers_unittests.cc @@ -0,0 +1,147 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/thread_helpers.h" + +#include <errno.h> +#include <fcntl.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "base/posix/eintr_wrapper.h" +#include "base/process/process_metrics.h" +#include "base/threading/platform_thread.h" +#include "base/threading/thread.h" +#include "build/build_config.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +using base::PlatformThread; + +namespace sandbox { + +namespace { + +// These tests fail under ThreadSanitizer, see http://crbug.com/342305 +#if !defined(THREAD_SANITIZER) + +int GetRaceTestIterations() { + if (IsRunningOnValgrind()) { + return 2; + } else { + return 1000; + } +} + +class ScopedProc { + public: + ScopedProc() : fd_(-1) { + fd_ = open("/proc/", O_RDONLY | O_DIRECTORY); + CHECK_LE(0, fd_); + } + + ~ScopedProc() { PCHECK(0 == IGNORE_EINTR(close(fd_))); } + + int fd() { return fd_; } + + private: + int fd_; + DISALLOW_COPY_AND_ASSIGN(ScopedProc); +}; + +TEST(ThreadHelpers, IsSingleThreadedBasic) { + ScopedProc proc_fd; + ASSERT_TRUE(ThreadHelpers::IsSingleThreaded(proc_fd.fd())); + ASSERT_TRUE(ThreadHelpers::IsSingleThreaded()); + + base::Thread thread("sandbox_tests"); + ASSERT_TRUE(thread.Start()); + ASSERT_FALSE(ThreadHelpers::IsSingleThreaded(proc_fd.fd())); + ASSERT_FALSE(ThreadHelpers::IsSingleThreaded()); + // Explicitly stop the thread here to not pollute the next test. + ASSERT_TRUE(ThreadHelpers::StopThreadAndWatchProcFS(proc_fd.fd(), &thread)); +} + +SANDBOX_TEST(ThreadHelpers, AssertSingleThreaded) { + ScopedProc proc_fd; + SANDBOX_ASSERT(ThreadHelpers::IsSingleThreaded(proc_fd.fd())); + SANDBOX_ASSERT(ThreadHelpers::IsSingleThreaded()); + + ThreadHelpers::AssertSingleThreaded(proc_fd.fd()); + ThreadHelpers::AssertSingleThreaded(); +} + +TEST(ThreadHelpers, IsSingleThreadedIterated) { + ScopedProc proc_fd; + ASSERT_TRUE(ThreadHelpers::IsSingleThreaded(proc_fd.fd())); + + // Iterate to check for race conditions. + for (int i = 0; i < GetRaceTestIterations(); ++i) { + base::Thread thread("sandbox_tests"); + ASSERT_TRUE(thread.Start()); + ASSERT_FALSE(ThreadHelpers::IsSingleThreaded(proc_fd.fd())); + // Explicitly stop the thread here to not pollute the next test. + ASSERT_TRUE(ThreadHelpers::StopThreadAndWatchProcFS(proc_fd.fd(), &thread)); + } +} + +TEST(ThreadHelpers, IsSingleThreadedStartAndStop) { + ScopedProc proc_fd; + ASSERT_TRUE(ThreadHelpers::IsSingleThreaded(proc_fd.fd())); + + base::Thread thread("sandbox_tests"); + // This is testing for a race condition, so iterate. + // Manually, this has been tested with more that 1M iterations. + for (int i = 0; i < GetRaceTestIterations(); ++i) { + ASSERT_TRUE(thread.Start()); + ASSERT_FALSE(ThreadHelpers::IsSingleThreaded(proc_fd.fd())); + + ASSERT_TRUE(ThreadHelpers::StopThreadAndWatchProcFS(proc_fd.fd(), &thread)); + ASSERT_TRUE(ThreadHelpers::IsSingleThreaded(proc_fd.fd())); + ASSERT_EQ(1, base::GetNumberOfThreads(base::GetCurrentProcessHandle())); + } +} + +SANDBOX_TEST(ThreadHelpers, AssertSingleThreadedAfterThreadStopped) { + SANDBOX_ASSERT(ThreadHelpers::IsSingleThreaded()); + + base::Thread thread1("sandbox_tests"); + base::Thread thread2("sandbox_tests"); + + for (int i = 0; i < GetRaceTestIterations(); ++i) { + SANDBOX_ASSERT(thread1.Start()); + SANDBOX_ASSERT(thread2.Start()); + SANDBOX_ASSERT(!ThreadHelpers::IsSingleThreaded()); + + thread1.Stop(); + thread2.Stop(); + // This will wait on /proc/ to reflect the state of threads in the + // process. + ThreadHelpers::AssertSingleThreaded(); + SANDBOX_ASSERT(ThreadHelpers::IsSingleThreaded()); + } +} + +// Only run this test in Debug mode, where AssertSingleThreaded() will return +// in less than 64ms. +#if !defined(NDEBUG) +SANDBOX_DEATH_TEST( + ThreadHelpers, + AssertSingleThreadedDies, + DEATH_MESSAGE( + ThreadHelpers::GetAssertSingleThreadedErrorMessageForTests())) { + base::Thread thread1("sandbox_tests"); + SANDBOX_ASSERT(thread1.Start()); + ThreadHelpers::AssertSingleThreaded(); +} +#endif // !defined(NDEBUG) + +#endif // !defined(THREAD_SANITIZER) + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/services/yama.cc b/sandbox/linux/services/yama.cc new file mode 100644 index 0000000000..151f4bd340 --- /dev/null +++ b/sandbox/linux/services/yama.cc @@ -0,0 +1,115 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/services/yama.h" + +#include <fcntl.h> +#include <sys/prctl.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> + +#include "base/files/file_util.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" + +#if !defined(PR_SET_PTRACER_ANY) +#define PR_SET_PTRACER_ANY ((unsigned long)-1) +#endif + +#if !defined(PR_SET_PTRACER) +#define PR_SET_PTRACER 0x59616d61 +#endif + +namespace sandbox { + +namespace { + +// Enable or disable the Yama ptracers restrictions. +// Return false if Yama is not present on this kernel. +bool SetYamaPtracersRestriction(bool enable_restrictions) { + unsigned long set_ptracer_arg; + if (enable_restrictions) { + set_ptracer_arg = 0; + } else { + set_ptracer_arg = PR_SET_PTRACER_ANY; + } + + const int ret = prctl(PR_SET_PTRACER, set_ptracer_arg); + const int prctl_errno = errno; + + if (0 == ret) { + return true; + } else { + // ENOSYS or EINVAL means Yama is not in the current kernel. + CHECK(ENOSYS == prctl_errno || EINVAL == prctl_errno); + return false; + } +} + +bool CanAccessProcFS() { + static const char kProcfsKernelSysPath[] = "/proc/sys/kernel/"; + int ret = access(kProcfsKernelSysPath, F_OK); + if (ret) { + return false; + } + return true; +} + +} // namespace + +// static +bool Yama::RestrictPtracersToAncestors() { + return SetYamaPtracersRestriction(true /* enable_restrictions */); +} + +// static +bool Yama::DisableYamaRestrictions() { + return SetYamaPtracersRestriction(false /* enable_restrictions */); +} + +// static +int Yama::GetStatus() { + if (!CanAccessProcFS()) { + return 0; + } + + static const char kPtraceScopePath[] = "/proc/sys/kernel/yama/ptrace_scope"; + + base::ScopedFD yama_scope(HANDLE_EINTR(open(kPtraceScopePath, O_RDONLY))); + + if (!yama_scope.is_valid()) { + const int open_errno = errno; + DCHECK(ENOENT == open_errno); + // The status is known, yama is not present. + return STATUS_KNOWN; + } + + char yama_scope_value = 0; + ssize_t num_read = HANDLE_EINTR(read(yama_scope.get(), &yama_scope_value, 1)); + PCHECK(1 == num_read); + + switch (yama_scope_value) { + case '0': + return STATUS_KNOWN | STATUS_PRESENT; + case '1': + return STATUS_KNOWN | STATUS_PRESENT | STATUS_ENFORCING; + case '2': + case '3': + return STATUS_KNOWN | STATUS_PRESENT | STATUS_ENFORCING | + STATUS_STRICT_ENFORCING; + default: + NOTREACHED(); + return 0; + } +} + +// static +bool Yama::IsPresent() { return GetStatus() & STATUS_PRESENT; } + +// static +bool Yama::IsEnforcing() { return GetStatus() & STATUS_ENFORCING; } + +} // namespace sandbox diff --git a/sandbox/linux/services/yama.h b/sandbox/linux/services/yama.h new file mode 100644 index 0000000000..e6c5c45b2a --- /dev/null +++ b/sandbox/linux/services/yama.h @@ -0,0 +1,57 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_YAMA_H_ +#define SANDBOX_LINUX_SERVICES_YAMA_H_ + +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// Yama is a LSM kernel module which can restrict ptrace(). +// This class provides ways to detect if Yama is present and enabled +// and to restrict which processes can ptrace the current process. +class SANDBOX_EXPORT Yama { + public: + // This enum should be used to set or check a bitmask. + // A value of 0 would indicate that the status is not known. + enum GlobalStatus { + STATUS_KNOWN = 1 << 0, + STATUS_PRESENT = 1 << 1, + STATUS_ENFORCING = 1 << 2, + // STATUS_STRICT_ENFORCING corresponds to either mode 2 or mode 3 of Yama. + // Ptrace could be entirely denied, or restricted to CAP_SYS_PTRACE + // and PTRACE_TRACEME. + STATUS_STRICT_ENFORCING = 1 << 3 + }; + + // Restrict who can ptrace() the current process to its ancestors. + // If this succeeds, then Yama is available on this kernel. + // However, Yama may not be enforcing at this time. + static bool RestrictPtracersToAncestors(); + + // Disable Yama restrictions for the current process. + // This will fail if Yama is not available on this kernel. + // This is meant for testing only. If you need this, implement + // a per-pid authorization instead. + static bool DisableYamaRestrictions(); + + // Checks if Yama is currently in enforcing mode for the machine (not the + // current process). This requires access to the filesystem and will use + // /proc/sys/kernel/yama/ptrace_scope. + static int GetStatus(); + + // Helper for checking for STATUS_PRESENT in GetStatus(). + static bool IsPresent(); + // Helper for checkking for STATUS_ENFORCING in GetStatus(). + static bool IsEnforcing(); + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(Yama); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_YAMA_H_ diff --git a/sandbox/linux/services/yama_unittests.cc b/sandbox/linux/services/yama_unittests.cc new file mode 100644 index 0000000000..204cfd6a44 --- /dev/null +++ b/sandbox/linux/services/yama_unittests.cc @@ -0,0 +1,172 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include <errno.h> +#include <fcntl.h> +#include <sys/ptrace.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> + +#include "base/bind.h" +#include "base/compiler_specific.h" +#include "base/posix/eintr_wrapper.h" +#include "base/strings/string_util.h" +#include "base/sys_info.h" +#include "sandbox/linux/services/scoped_process.h" +#include "sandbox/linux/services/yama.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +bool HasLinux32Bug() { +#if defined(__i386__) + // On 3.2 kernels, yama doesn't work for 32-bit binaries on 64-bit kernels. + // This is fixed in 3.4. + bool is_kernel_64bit = + base::SysInfo::OperatingSystemArchitecture() == "x86_64"; + bool is_linux = base::SysInfo::OperatingSystemName() == "Linux"; + bool is_3_dot_2 = base::StartsWithASCII( + base::SysInfo::OperatingSystemVersion(), "3.2", /*case_sensitive=*/false); + if (is_kernel_64bit && is_linux && is_3_dot_2) + return true; +#endif // defined(__i386__) + return false; +} + +bool CanPtrace(pid_t pid) { + int ret; + ret = ptrace(PTRACE_ATTACH, pid, NULL, NULL); + if (ret == -1) { + CHECK_EQ(EPERM, errno); + return false; + } + // Wait for the process to be stopped so that it can be detached. + siginfo_t process_info; + int wait_ret = HANDLE_EINTR(waitid(P_PID, pid, &process_info, WSTOPPED)); + PCHECK(0 == wait_ret); + PCHECK(0 == ptrace(PTRACE_DETACH, pid, NULL, NULL)); + return true; +} + +// _exit(0) if pid can be ptraced by the current process. +// _exit(1) otherwise. +void ExitZeroIfCanPtrace(pid_t pid) { + if (CanPtrace(pid)) { + _exit(0); + } else { + _exit(1); + } +} + +bool CanSubProcessPtrace(pid_t pid) { + ScopedProcess process(base::Bind(&ExitZeroIfCanPtrace, pid)); + bool signaled; + int exit_code = process.WaitForExit(&signaled); + CHECK(!signaled); + return 0 == exit_code; +} + +// The tests below assume that the system-level configuration will not change +// while they run. + +TEST(Yama, GetStatus) { + int status1 = Yama::GetStatus(); + + // Check that the value is a possible bitmask. + ASSERT_LE(0, status1); + ASSERT_GE(Yama::STATUS_KNOWN | Yama::STATUS_PRESENT | Yama::STATUS_ENFORCING | + Yama::STATUS_STRICT_ENFORCING, + status1); + + // The status should not just be a random value. + int status2 = Yama::GetStatus(); + EXPECT_EQ(status1, status2); + + // This test is not running sandboxed, there is no reason to not know the + // status. + EXPECT_NE(0, Yama::STATUS_KNOWN & status1); + + if (status1 & Yama::STATUS_STRICT_ENFORCING) { + // If Yama is strictly enforcing, it is also enforcing. + EXPECT_TRUE(status1 & Yama::STATUS_ENFORCING); + } + + if (status1 & Yama::STATUS_ENFORCING) { + // If Yama is enforcing, Yama is present. + EXPECT_NE(0, status1 & Yama::STATUS_PRESENT); + } + + // Verify that the helper functions work as intended. + EXPECT_EQ(static_cast<bool>(status1 & Yama::STATUS_ENFORCING), + Yama::IsEnforcing()); + EXPECT_EQ(static_cast<bool>(status1 & Yama::STATUS_PRESENT), + Yama::IsPresent()); + + fprintf(stdout, + "Yama present: %s - enforcing: %s\n", + Yama::IsPresent() ? "Y" : "N", + Yama::IsEnforcing() ? "Y" : "N"); +} + +SANDBOX_TEST(Yama, RestrictPtraceSucceedsWhenYamaPresent) { + // This call will succeed iff Yama is present. + bool restricted = Yama::RestrictPtracersToAncestors(); + CHECK_EQ(restricted, Yama::IsPresent()); +} + +// Attempts to enable or disable Yama restrictions. +void SetYamaRestrictions(bool enable_restriction) { + if (enable_restriction) { + Yama::RestrictPtracersToAncestors(); + } else { + Yama::DisableYamaRestrictions(); + } +} + +TEST(Yama, RestrictPtraceWorks) { + if (HasLinux32Bug()) + return; + + ScopedProcess process1(base::Bind(&SetYamaRestrictions, true)); + ASSERT_TRUE(process1.WaitForClosureToRun()); + + if (Yama::IsEnforcing()) { + // A sibling process cannot ptrace process1. + ASSERT_FALSE(CanSubProcessPtrace(process1.GetPid())); + } + + if (!(Yama::GetStatus() & Yama::STATUS_STRICT_ENFORCING)) { + // However, parent can ptrace process1. + ASSERT_TRUE(CanPtrace(process1.GetPid())); + + // A sibling can ptrace process2 which disables any Yama protection. + ScopedProcess process2(base::Bind(&SetYamaRestrictions, false)); + ASSERT_TRUE(process2.WaitForClosureToRun()); + ASSERT_TRUE(CanSubProcessPtrace(process2.GetPid())); + } +} + +void DoNothing() {} + +SANDBOX_TEST(Yama, RestrictPtraceIsDefault) { + if (!Yama::IsPresent() || HasLinux32Bug()) + return; + + CHECK(Yama::DisableYamaRestrictions()); + ScopedProcess process1(base::Bind(&DoNothing)); + + if (Yama::IsEnforcing()) { + // Check that process1 is protected by Yama, even though it has + // been created from a process that disabled Yama. + CHECK(!CanSubProcessPtrace(process1.GetPid())); + } +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/suid/client/DEPS b/sandbox/linux/suid/client/DEPS new file mode 100644 index 0000000000..99a337d772 --- /dev/null +++ b/sandbox/linux/suid/client/DEPS @@ -0,0 +1,3 @@ +include_rules = [ + "+sandbox/linux/services", +] diff --git a/sandbox/linux/suid/client/setuid_sandbox_client.cc b/sandbox/linux/suid/client/setuid_sandbox_client.cc new file mode 100644 index 0000000000..12ef7f9f40 --- /dev/null +++ b/sandbox/linux/suid/client/setuid_sandbox_client.cc @@ -0,0 +1,151 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/suid/client/setuid_sandbox_client.h" + +#include <fcntl.h> +#include <sys/stat.h> +#include <sys/wait.h> +#include <unistd.h> + +#include <string> + +#include "base/environment.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "base/strings/string_number_conversions.h" +#include "sandbox/linux/suid/common/sandbox.h" + +namespace { + +bool IsFileSystemAccessDenied() { + base::ScopedFD root_dir(HANDLE_EINTR(open("/", O_RDONLY))); + return !root_dir.is_valid(); +} + +int GetHelperApi(base::Environment* env) { + std::string api_string; + int api_number = 0; // Assume API version 0 if no environment was found. + if (env->GetVar(sandbox::kSandboxEnvironmentApiProvides, &api_string) && + !base::StringToInt(api_string, &api_number)) { + // It's an error if we could not convert the API number. + api_number = -1; + } + return api_number; +} + +// Convert |var_name| from the environment |env| to an int. +// Return -1 if the variable does not exist or the value cannot be converted. +int EnvToInt(base::Environment* env, const char* var_name) { + std::string var_string; + int var_value = -1; + if (env->GetVar(var_name, &var_string) && + !base::StringToInt(var_string, &var_value)) { + var_value = -1; + } + return var_value; +} + +pid_t GetHelperPID(base::Environment* env) { + return EnvToInt(env, sandbox::kSandboxHelperPidEnvironmentVarName); +} + +// Get the IPC file descriptor used to communicate with the setuid helper. +int GetIPCDescriptor(base::Environment* env) { + return EnvToInt(env, sandbox::kSandboxDescriptorEnvironmentVarName); +} + +} // namespace + +namespace sandbox { + +SetuidSandboxClient* SetuidSandboxClient::Create() { + base::Environment* environment(base::Environment::Create()); + CHECK(environment); + return new SetuidSandboxClient(environment); +} + +SetuidSandboxClient::SetuidSandboxClient(base::Environment* env) + : env_(env), sandboxed_(false) { +} + +SetuidSandboxClient::~SetuidSandboxClient() { +} + +void SetuidSandboxClient::CloseDummyFile() { + // When we're launched through the setuid sandbox, SetupLaunchOptions + // arranges for kZygoteIdFd to be a dummy file descriptor to satisfy an + // ancient setuid sandbox ABI requirement. However, the descriptor is no + // longer needed, so we can simply close it right away now. + CHECK(IsSuidSandboxChild()); + + // Sanity check that kZygoteIdFd refers to a pipe. + struct stat st; + PCHECK(0 == fstat(kZygoteIdFd, &st)); + CHECK(S_ISFIFO(st.st_mode)); + + PCHECK(0 == IGNORE_EINTR(close(kZygoteIdFd))); +} + +bool SetuidSandboxClient::ChrootMe() { + int ipc_fd = GetIPCDescriptor(env_.get()); + + if (ipc_fd < 0) { + LOG(ERROR) << "Failed to obtain the sandbox IPC descriptor"; + return false; + } + + if (HANDLE_EINTR(write(ipc_fd, &kMsgChrootMe, 1)) != 1) { + PLOG(ERROR) << "Failed to write to chroot pipe"; + return false; + } + + // We need to reap the chroot helper process in any event. + pid_t helper_pid = GetHelperPID(env_.get()); + // If helper_pid is -1 we wait for any child. + if (HANDLE_EINTR(waitpid(helper_pid, NULL, 0)) < 0) { + PLOG(ERROR) << "Failed to wait for setuid helper to die"; + return false; + } + + char reply; + if (HANDLE_EINTR(read(ipc_fd, &reply, 1)) != 1) { + PLOG(ERROR) << "Failed to read from chroot pipe"; + return false; + } + + if (reply != kMsgChrootSuccessful) { + LOG(ERROR) << "Error code reply from chroot helper"; + return false; + } + + // We now consider ourselves "fully sandboxed" as far as the + // setuid sandbox is concerned. + CHECK(IsFileSystemAccessDenied()); + sandboxed_ = true; + return true; +} + +bool SetuidSandboxClient::IsSuidSandboxUpToDate() const { + return GetHelperApi(env_.get()) == kSUIDSandboxApiNumber; +} + +bool SetuidSandboxClient::IsSuidSandboxChild() const { + return GetIPCDescriptor(env_.get()) >= 0; +} + +bool SetuidSandboxClient::IsInNewPIDNamespace() const { + return env_->HasVar(kSandboxPIDNSEnvironmentVarName); +} + +bool SetuidSandboxClient::IsInNewNETNamespace() const { + return env_->HasVar(kSandboxNETNSEnvironmentVarName); +} + +bool SetuidSandboxClient::IsSandboxed() const { + return sandboxed_; +} + +} // namespace sandbox diff --git a/sandbox/linux/suid/client/setuid_sandbox_client.h b/sandbox/linux/suid/client/setuid_sandbox_client.h new file mode 100644 index 0000000000..026894fc27 --- /dev/null +++ b/sandbox/linux/suid/client/setuid_sandbox_client.h @@ -0,0 +1,71 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SUID_SETUID_SANDBOX_CLIENT_H_ +#define SANDBOX_LINUX_SUID_SETUID_SANDBOX_CLIENT_H_ + +#include "base/environment.h" +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// Helper class to use the setuid sandbox. This class is to be used +// after being executed through the setuid helper. +// This class is difficult to use. It has been created by refactoring very old +// code scathered through the Chromium code base. +// +// A typical use for "A" launching a sandboxed process "B" would be: +// (Steps 1 through 4 are described in setuid_sandbox_host.h.) +// 5. B uses CloseDummyFile() to close the dummy file descriptor. +// 6. B performs various initializations that require access to the file +// system. +// 6.b (optional) B uses sandbox::Credentials::HasOpenDirectory() to verify +// that no directory is kept open (which would allow bypassing the setuid +// sandbox). +// 7. B should be prepared to assume the role of init(1). In particular, B +// cannot receive any signal from any other process, excluding SIGKILL. +// If B dies, all the processes in the namespace will die. +// B can fork() and the parent can assume the role of init(1), by using +// sandbox::CreateInitProcessReaper(). +// 8. B requests being chroot-ed through ChrootMe() and +// requests other sandboxing status via the status functions. +class SANDBOX_EXPORT SetuidSandboxClient { + public: + // All instantation should go through this factory method. + static SetuidSandboxClient* Create(); + ~SetuidSandboxClient(); + + // Close the dummy file descriptor leftover from the sandbox ABI. + void CloseDummyFile(); + // Ask the setuid helper over the setuid sandbox IPC channel to chroot() us + // to an empty directory. + // Will only work if we have been launched through the setuid helper. + bool ChrootMe(); + + // Did we get launched through an up to date setuid binary ? + bool IsSuidSandboxUpToDate() const; + // Did we get launched through the setuid helper ? + bool IsSuidSandboxChild() const; + // Did the setuid helper create a new PID namespace ? + bool IsInNewPIDNamespace() const; + // Did the setuid helper create a new network namespace ? + bool IsInNewNETNamespace() const; + // Are we done and fully sandboxed ? + bool IsSandboxed() const; + + private: + explicit SetuidSandboxClient(base::Environment* env); + + // Holds the environment. Will never be NULL. + scoped_ptr<base::Environment> env_; + bool sandboxed_; + + DISALLOW_COPY_AND_ASSIGN(SetuidSandboxClient); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SUID_SETUID_SANDBOX_CLIENT_H_ diff --git a/sandbox/linux/suid/client/setuid_sandbox_client_unittest.cc b/sandbox/linux/suid/client/setuid_sandbox_client_unittest.cc new file mode 100644 index 0000000000..2acd8fb5fc --- /dev/null +++ b/sandbox/linux/suid/client/setuid_sandbox_client_unittest.cc @@ -0,0 +1,46 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/suid/client/setuid_sandbox_client.h" + +#include "base/environment.h" +#include "base/memory/scoped_ptr.h" +#include "base/strings/string_number_conversions.h" +#include "sandbox/linux/suid/common/sandbox.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +TEST(SetuidSandboxClient, SandboxedClientAPI) { + scoped_ptr<base::Environment> env(base::Environment::Create()); + EXPECT_TRUE(env != NULL); + + scoped_ptr<SetuidSandboxClient> + sandbox_client(SetuidSandboxClient::Create()); + EXPECT_TRUE(sandbox_client != NULL); + + // Set-up a fake environment as if we went through the setuid sandbox. + EXPECT_TRUE(env->SetVar(kSandboxEnvironmentApiProvides, + base::IntToString(kSUIDSandboxApiNumber))); + EXPECT_TRUE(env->SetVar(kSandboxDescriptorEnvironmentVarName, "1")); + EXPECT_TRUE(env->SetVar(kSandboxPIDNSEnvironmentVarName, "1")); + EXPECT_TRUE(env->UnSetVar(kSandboxNETNSEnvironmentVarName)); + + // Check the API. + EXPECT_TRUE(sandbox_client->IsSuidSandboxUpToDate()); + EXPECT_TRUE(sandbox_client->IsSuidSandboxChild()); + EXPECT_TRUE(sandbox_client->IsInNewPIDNamespace()); + EXPECT_FALSE(sandbox_client->IsInNewNETNamespace()); + + // Forge an incorrect API version and check. + EXPECT_TRUE(env->SetVar(kSandboxEnvironmentApiProvides, + base::IntToString(kSUIDSandboxApiNumber + 1))); + EXPECT_FALSE(sandbox_client->IsSuidSandboxUpToDate()); + // We didn't go through the actual sandboxing mechanism as it is + // very hard in a unit test. + EXPECT_FALSE(sandbox_client->IsSandboxed()); +} + +} // namespace sandbox + diff --git a/sandbox/linux/suid/client/setuid_sandbox_host.cc b/sandbox/linux/suid/client/setuid_sandbox_host.cc new file mode 100644 index 0000000000..71171ebd4f --- /dev/null +++ b/sandbox/linux/suid/client/setuid_sandbox_host.cc @@ -0,0 +1,195 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/suid/client/setuid_sandbox_host.h" + +#include <fcntl.h> +#include <stdlib.h> +#include <sys/stat.h> +#include <unistd.h> + +#include <string> +#include <utility> + +#include "base/command_line.h" +#include "base/environment.h" +#include "base/files/file_path.h" +#include "base/files/file_util.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "base/path_service.h" +#include "base/posix/eintr_wrapper.h" +#include "base/process/launch.h" +#include "base/process/process_metrics.h" +#include "base/strings/string_number_conversions.h" +#include "sandbox/linux/suid/common/sandbox.h" +#include "sandbox/linux/suid/common/suid_unsafe_environment_variables.h" + +namespace { + +// Set an environment variable that reflects the API version we expect from the +// setuid sandbox. Old versions of the sandbox will ignore this. +void SetSandboxAPIEnvironmentVariable(base::Environment* env) { + env->SetVar(sandbox::kSandboxEnvironmentApiRequest, + base::IntToString(sandbox::kSUIDSandboxApiNumber)); +} + +// Unset environment variables that are expected to be set by the setuid +// sandbox. This is to allow nesting of one instance of the SUID sandbox +// inside another. +void UnsetExpectedEnvironmentVariables(base::EnvironmentMap* env_map) { + DCHECK(env_map); + const base::NativeEnvironmentString environment_vars[] = { + sandbox::kSandboxDescriptorEnvironmentVarName, + sandbox::kSandboxHelperPidEnvironmentVarName, + sandbox::kSandboxEnvironmentApiProvides, + sandbox::kSandboxPIDNSEnvironmentVarName, + sandbox::kSandboxNETNSEnvironmentVarName, + }; + + for (size_t i = 0; i < arraysize(environment_vars); ++i) { + // Setting values in EnvironmentMap to an empty-string will make + // sure that they get unset from the environment via AlterEnvironment(). + (*env_map)[environment_vars[i]] = base::NativeEnvironmentString(); + } +} + +// Wrapper around a shared C function. +// Returns the "saved" environment variable name corresponding to |envvar| +// in a new string or NULL. +std::string* CreateSavedVariableName(const char* env_var) { + char* const saved_env_var = SandboxSavedEnvironmentVariable(env_var); + if (!saved_env_var) + return NULL; + std::string* saved_env_var_copy = new std::string(saved_env_var); + // SandboxSavedEnvironmentVariable is the C function that we wrap and uses + // malloc() to allocate memory. + free(saved_env_var); + return saved_env_var_copy; +} + +// The ELF loader will clear many environment variables so we save them to +// different names here so that the SUID sandbox can resolve them for the +// renderer. +void SaveSUIDUnsafeEnvironmentVariables(base::Environment* env) { + for (unsigned i = 0; kSUIDUnsafeEnvironmentVariables[i]; ++i) { + const char* env_var = kSUIDUnsafeEnvironmentVariables[i]; + // Get the saved environment variable corresponding to envvar. + scoped_ptr<std::string> saved_env_var(CreateSavedVariableName(env_var)); + if (saved_env_var == NULL) + continue; + + std::string value; + if (env->GetVar(env_var, &value)) + env->SetVar(saved_env_var->c_str(), value); + else + env->UnSetVar(saved_env_var->c_str()); + } +} + +const char* GetDevelSandboxPath() { + return getenv("CHROME_DEVEL_SANDBOX"); +} + +} // namespace + +namespace sandbox { + +SetuidSandboxHost* SetuidSandboxHost::Create() { + base::Environment* environment(base::Environment::Create()); + CHECK(environment); + return new SetuidSandboxHost(environment); +} + +SetuidSandboxHost::SetuidSandboxHost(base::Environment* env) : env_(env) { +} + +SetuidSandboxHost::~SetuidSandboxHost() { +} + +// Check if CHROME_DEVEL_SANDBOX is set but empty. This currently disables +// the setuid sandbox. TODO(jln): fix this (crbug.com/245376). +bool SetuidSandboxHost::IsDisabledViaEnvironment() { + const char* devel_sandbox_path = GetDevelSandboxPath(); + if (devel_sandbox_path && '\0' == *devel_sandbox_path) { + return true; + } + return false; +} + +base::FilePath SetuidSandboxHost::GetSandboxBinaryPath() { + base::FilePath sandbox_binary; + base::FilePath exe_dir; + if (PathService::Get(base::DIR_EXE, &exe_dir)) { + base::FilePath sandbox_candidate = exe_dir.AppendASCII("chrome-sandbox"); + if (base::PathExists(sandbox_candidate)) + sandbox_binary = sandbox_candidate; + } + + // In user-managed builds, including development builds, an environment + // variable is required to enable the sandbox. See + // http://code.google.com/p/chromium/wiki/LinuxSUIDSandboxDevelopment + struct stat st; + if (sandbox_binary.empty() && stat(base::kProcSelfExe, &st) == 0 && + st.st_uid == getuid()) { + const char* devel_sandbox_path = GetDevelSandboxPath(); + if (devel_sandbox_path) { + sandbox_binary = base::FilePath(devel_sandbox_path); + } + } + + return sandbox_binary; +} + +void SetuidSandboxHost::PrependWrapper(base::CommandLine* cmd_line) { + std::string sandbox_binary(GetSandboxBinaryPath().value()); + struct stat st; + if (sandbox_binary.empty() || stat(sandbox_binary.c_str(), &st) != 0) { + LOG(FATAL) << "The SUID sandbox helper binary is missing: " + << sandbox_binary << " Aborting now. See " + "https://code.google.com/p/chromium/wiki/" + "LinuxSUIDSandboxDevelopment."; + } + + if (access(sandbox_binary.c_str(), X_OK) != 0 || (st.st_uid != 0) || + ((st.st_mode & S_ISUID) == 0) || ((st.st_mode & S_IXOTH)) == 0) { + LOG(FATAL) << "The SUID sandbox helper binary was found, but is not " + "configured correctly. Rather than run without sandboxing " + "I'm aborting now. You need to make sure that " + << sandbox_binary << " is owned by root and has mode 4755."; + } + + cmd_line->PrependWrapper(sandbox_binary); +} + +void SetuidSandboxHost::SetupLaunchOptions( + base::LaunchOptions* options, + base::FileHandleMappingVector* fds_to_remap, + base::ScopedFD* dummy_fd) { + DCHECK(options); + DCHECK(fds_to_remap); + + // Launching a setuid binary requires PR_SET_NO_NEW_PRIVS to not be used. + options->allow_new_privs = true; + UnsetExpectedEnvironmentVariables(&options->environ); + + // Set dummy_fd to the reading end of a closed pipe. + int pipe_fds[2]; + PCHECK(0 == pipe(pipe_fds)); + PCHECK(0 == IGNORE_EINTR(close(pipe_fds[1]))); + dummy_fd->reset(pipe_fds[0]); + + // We no longer need a dummy socket for discovering the child's PID, + // but the sandbox is still hard-coded to expect a file descriptor at + // kZygoteIdFd. Fixing this requires a sandbox API change. :( + fds_to_remap->push_back(std::make_pair(dummy_fd->get(), kZygoteIdFd)); +} + +void SetuidSandboxHost::SetupLaunchEnvironment() { + SaveSUIDUnsafeEnvironmentVariables(env_.get()); + SetSandboxAPIEnvironmentVariable(env_.get()); +} + +} // namespace sandbox diff --git a/sandbox/linux/suid/client/setuid_sandbox_host.h b/sandbox/linux/suid/client/setuid_sandbox_host.h new file mode 100644 index 0000000000..6788892441 --- /dev/null +++ b/sandbox/linux/suid/client/setuid_sandbox_host.h @@ -0,0 +1,70 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SUID_SETUID_SANDBOX_HOST_H_ +#define SANDBOX_LINUX_SUID_SETUID_SANDBOX_HOST_H_ + +#include "base/files/file_path.h" +#include "base/files/scoped_file.h" +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "base/process/launch.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +// Helper class to use the setuid sandbox. This class is to be used +// before launching the setuid helper. +// This class is difficult to use. It has been created by refactoring very old +// code scathered through the Chromium code base. +// +// A typical use for "A" launching a sandboxed process "B" would be: +// 1. A calls SetupLaunchEnvironment() +// 2. A sets up a base::CommandLine and then amends it with +// PrependWrapper() (or manually, by relying on GetSandboxBinaryPath()). +// 3. A uses SetupLaunchOptions() to arrange for a dummy descriptor for the +// setuid sandbox ABI. +// 4. A launches B with base::LaunchProcess, using the amended +// base::CommandLine. +// (The remaining steps are described within setuid_sandbox_client.h.) +class SANDBOX_EXPORT SetuidSandboxHost { + public: + // All instantation should go through this factory method. + static SetuidSandboxHost* Create(); + ~SetuidSandboxHost(); + + // The setuid sandbox may still be disabled via the environment. + // This is tracked in crbug.com/245376. + bool IsDisabledViaEnvironment(); + // Get the sandbox binary path. This method knows about the + // CHROME_DEVEL_SANDBOX environment variable used for user-managed builds. If + // the sandbox binary cannot be found, it will return an empty FilePath. + base::FilePath GetSandboxBinaryPath(); + // Modify |cmd_line| to launch via the setuid sandbox. Crash if the setuid + // sandbox binary cannot be found. |cmd_line| must not be NULL. + void PrependWrapper(base::CommandLine* cmd_line); + // Set-up the launch options for launching via the setuid sandbox. Caller is + // responsible for keeping |dummy_fd| alive until LaunchProcess() completes. + // |options| and |fds_to_remap| must not be NULL. + // (Keeping |dummy_fd| alive is an unfortunate historical artifact of the + // chrome-sandbox ABI.) + void SetupLaunchOptions(base::LaunchOptions* options, + base::FileHandleMappingVector* fds_to_remap, + base::ScopedFD* dummy_fd); + // Set-up the environment. This should be done prior to launching the setuid + // helper. + void SetupLaunchEnvironment(); + + private: + explicit SetuidSandboxHost(base::Environment* env); + + // Holds the environment. Will never be NULL. + scoped_ptr<base::Environment> env_; + + DISALLOW_COPY_AND_ASSIGN(SetuidSandboxHost); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SUID_SETUID_SANDBOX_HOST_H_ diff --git a/sandbox/linux/suid/client/setuid_sandbox_host_unittest.cc b/sandbox/linux/suid/client/setuid_sandbox_host_unittest.cc new file mode 100644 index 0000000000..8415abb064 --- /dev/null +++ b/sandbox/linux/suid/client/setuid_sandbox_host_unittest.cc @@ -0,0 +1,72 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/suid/client/setuid_sandbox_host.h" + +#include <string> + +#include "base/environment.h" +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "base/strings/string_number_conversions.h" +#include "sandbox/linux/suid/common/sandbox.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +TEST(SetuidSandboxHost, SetupLaunchEnvironment) { + const char kTestValue[] = "This is a test"; + scoped_ptr<base::Environment> env(base::Environment::Create()); + EXPECT_TRUE(env != NULL); + + std::string saved_ld_preload; + bool environment_had_ld_preload; + // First, back-up the real LD_PRELOAD if any. + environment_had_ld_preload = env->GetVar("LD_PRELOAD", &saved_ld_preload); + // Setup environment variables to save or not save. + EXPECT_TRUE(env->SetVar("LD_PRELOAD", kTestValue)); + EXPECT_TRUE(env->UnSetVar("LD_ORIGIN_PATH")); + + scoped_ptr<SetuidSandboxHost> sandbox_host(SetuidSandboxHost::Create()); + EXPECT_TRUE(sandbox_host != NULL); + + // Make sure the environment is clean. + EXPECT_TRUE(env->UnSetVar(kSandboxEnvironmentApiRequest)); + EXPECT_TRUE(env->UnSetVar(kSandboxEnvironmentApiProvides)); + + sandbox_host->SetupLaunchEnvironment(); + + // Check if the requested API environment was set. + std::string api_request; + EXPECT_TRUE(env->GetVar(kSandboxEnvironmentApiRequest, &api_request)); + int api_request_num; + EXPECT_TRUE(base::StringToInt(api_request, &api_request_num)); + EXPECT_EQ(api_request_num, kSUIDSandboxApiNumber); + + // Now check if LD_PRELOAD was saved to SANDBOX_LD_PRELOAD. + std::string sandbox_ld_preload; + EXPECT_TRUE(env->GetVar("SANDBOX_LD_PRELOAD", &sandbox_ld_preload)); + EXPECT_EQ(sandbox_ld_preload, kTestValue); + + // Check that LD_ORIGIN_PATH was not saved. + EXPECT_FALSE(env->HasVar("SANDBOX_LD_ORIGIN_PATH")); + + // We should not forget to restore LD_PRELOAD at the end, or this environment + // variable will affect the next running tests! + if (environment_had_ld_preload) { + EXPECT_TRUE(env->SetVar("LD_PRELOAD", saved_ld_preload)); + } else { + EXPECT_TRUE(env->UnSetVar("LD_PRELOAD")); + } +} + +// This test doesn't accomplish much, but will make sure that analysis tools +// will run this codepath. +TEST(SetuidSandboxHost, GetSandboxBinaryPath) { + scoped_ptr<SetuidSandboxHost> setuid_sandbox_host( + SetuidSandboxHost::Create()); + ignore_result(setuid_sandbox_host->GetSandboxBinaryPath()); +} + +} // namespace sandbox diff --git a/sandbox/linux/suid/common/sandbox.h b/sandbox/linux/suid/common/sandbox.h new file mode 100644 index 0000000000..99eb7b5120 --- /dev/null +++ b/sandbox/linux/suid/common/sandbox.h @@ -0,0 +1,41 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SUID_SANDBOX_H_ +#define SANDBOX_LINUX_SUID_SANDBOX_H_ + +#if defined(__cplusplus) +namespace sandbox { +#endif + +// These are command line switches that may be used by other programs +// (e.g. Chrome) to construct a command line for the sandbox. +static const char kSuidSandboxGetApiSwitch[] = "--get-api"; +static const char kAdjustOOMScoreSwitch[] = "--adjust-oom-score"; + +static const char kSandboxDescriptorEnvironmentVarName[] = "SBX_D"; +static const char kSandboxHelperPidEnvironmentVarName[] = "SBX_HELPER_PID"; + +static const long kSUIDSandboxApiNumber = 1; +static const char kSandboxEnvironmentApiRequest[] = "SBX_CHROME_API_RQ"; +static const char kSandboxEnvironmentApiProvides[] = "SBX_CHROME_API_PRV"; + +// This number must be kept in sync with common/zygote_commands_linux.h +static const int kZygoteIdFd = 7; + +// These are the magic byte values which the sandboxed process uses to request +// that it be chrooted. +static const char kMsgChrootMe = 'C'; +static const char kMsgChrootSuccessful = 'O'; + +// These are set if we have respectively switched to a new PID or NET namespace +// by going through the setuid binary helper. +static const char kSandboxPIDNSEnvironmentVarName[] = "SBX_PID_NS"; +static const char kSandboxNETNSEnvironmentVarName[] = "SBX_NET_NS"; + +#if defined(__cplusplus) +} // namespace sandbox +#endif + +#endif // SANDBOX_LINUX_SUID_SANDBOX_H_ diff --git a/sandbox/linux/suid/common/suid_unsafe_environment_variables.h b/sandbox/linux/suid/common/suid_unsafe_environment_variables.h new file mode 100644 index 0000000000..33ba4b6ab7 --- /dev/null +++ b/sandbox/linux/suid/common/suid_unsafe_environment_variables.h @@ -0,0 +1,73 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// This is a list of environment variables which the ELF loader unsets when +// loading a SUID binary. Because they are unset rather than just ignored, they +// aren't passed to child processes of SUID processes either. +// +// We need to save these environment variables before running a SUID sandbox +// and restore them before running child processes (but after dropping root). +// +// List gathered from glibc sources (00ebd7ed58df389a78e41dece058048725cb585e): +// sysdeps/unix/sysv/linux/i386/dl-librecon.h +// sysdeps/generic/unsecvars.h + +#ifndef SANDBOX_LINUX_SUID_SUID_UNSAFE_ENVIRONMENT_VARIABLES_H_ +#define SANDBOX_LINUX_SUID_SUID_UNSAFE_ENVIRONMENT_VARIABLES_H_ + +#include <stdint.h> +#include <stdlib.h> // malloc +#include <string.h> // memcpy + +static const char* kSUIDUnsafeEnvironmentVariables[] = { + "LD_AOUT_LIBRARY_PATH", + "LD_AOUT_PRELOAD", + "GCONV_PATH", + "GETCONF_DIR", + "HOSTALIASES", + "LD_AUDIT", + "LD_DEBUG", + "LD_DEBUG_OUTPUT", + "LD_DYNAMIC_WEAK", + "LD_LIBRARY_PATH", + "LD_ORIGIN_PATH", + "LD_PRELOAD", + "LD_PROFILE", + "LD_SHOW_AUXV", + "LD_USE_LOAD_BIAS", + "LOCALDOMAIN", + "LOCPATH", + "MALLOC_TRACE", + "NIS_PATH", + "NLSPATH", + "RESOLV_HOST_CONF", + "RES_OPTIONS", + "TMPDIR", + "TZDIR", + NULL, +}; + +// Return a malloc allocated string containing the 'saved' environment variable +// name for a given environment variable. +static inline char* SandboxSavedEnvironmentVariable(const char* envvar) { + const size_t envvar_len = strlen(envvar); + const size_t kMaxSizeT = (size_t) -1; + + if (envvar_len > kMaxSizeT - 1 - 8) + return NULL; + + const size_t saved_envvarlen = envvar_len + 1 /* NUL terminator */ + + 8 /* strlen("SANDBOX_") */; + char* const saved_envvar = (char*) malloc(saved_envvarlen); + if (!saved_envvar) + return NULL; + + memcpy(saved_envvar, "SANDBOX_", 8); + memcpy(saved_envvar + 8, envvar, envvar_len); + saved_envvar[8 + envvar_len] = 0; + + return saved_envvar; +} + +#endif // SANDBOX_LINUX_SUID_SUID_UNSAFE_ENVIRONMENT_VARIABLES_H_ diff --git a/sandbox/linux/suid/process_util.h b/sandbox/linux/suid/process_util.h new file mode 100644 index 0000000000..9fb9a8791a --- /dev/null +++ b/sandbox/linux/suid/process_util.h @@ -0,0 +1,30 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// The following is duplicated from base/process_utils.h. +// We shouldn't link against C++ code in a setuid binary. + +#ifndef SANDBOX_LINUX_SUID_PROCESS_UTIL_H_ +#define SANDBOX_LINUX_SUID_PROCESS_UTIL_H_ + +#include <stdbool.h> +#include <sys/types.h> + +// This adjusts /proc/process/oom_score_adj so the Linux OOM killer +// will prefer certain process types over others. The range for the +// adjustment is [-1000, 1000], with [0, 1000] being user accessible. +// +// If the Linux system isn't new enough to use oom_score_adj, then we +// try to set the older oom_adj value instead, scaling the score to +// the required range of [0, 15]. This may result in some aliasing of +// values, of course. +bool AdjustOOMScore(pid_t process, int score); + +// This adjusts /sys/kernel/mm/chromeos-low_mem/margin so that +// the kernel notifies us that we are low on memory when less than +// |margin_mb| megabytes are available. Setting |margin_mb| to -1 +// turns off low memory notification. +bool AdjustLowMemoryMargin(int64_t margin_mb); + +#endif // SANDBOX_LINUX_SUID_PROCESS_UTIL_H_ diff --git a/sandbox/linux/suid/process_util_linux.c b/sandbox/linux/suid/process_util_linux.c new file mode 100644 index 0000000000..8d9a53c3a4 --- /dev/null +++ b/sandbox/linux/suid/process_util_linux.c @@ -0,0 +1,78 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// The following is the C version of code from base/process_utils_linux.cc. +// We shouldn't link against C++ code in a setuid binary. + +// Needed for O_DIRECTORY, must be defined before fcntl.h is included +// (and it can be included earlier than the explicit #include below +// in some versions of glibc). +#define _GNU_SOURCE + +#include "sandbox/linux/suid/process_util.h" + +#include <fcntl.h> +#include <inttypes.h> +#include <limits.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> + +// Ranges for the current (oom_score_adj) and previous (oom_adj) +// flavors of OOM score. +static const int kMaxOomScore = 1000; +static const int kMaxOldOomScore = 15; + +// NOTE: This is not the only version of this function in the source: +// the base library (in process_util_linux.cc) also has its own C++ version. +bool AdjustOOMScore(pid_t process, int score) { + if (score < 0 || score > kMaxOomScore) + return false; + + char oom_adj[27]; // "/proc/" + log_10(2**64) + "\0" + // 6 + 20 + 1 = 27 + snprintf(oom_adj, sizeof(oom_adj), "/proc/%" PRIdMAX, (intmax_t)process); + + const int dirfd = open(oom_adj, O_RDONLY | O_DIRECTORY); + if (dirfd < 0) + return false; + + struct stat statbuf; + if (fstat(dirfd, &statbuf) < 0) { + close(dirfd); + return false; + } + if (getuid() != statbuf.st_uid) { + close(dirfd); + return false; + } + + int fd = openat(dirfd, "oom_score_adj", O_WRONLY); + if (fd < 0) { + // We failed to open oom_score_adj, so let's try for the older + // oom_adj file instead. + fd = openat(dirfd, "oom_adj", O_WRONLY); + if (fd < 0) { + // Nope, that doesn't work either. + return false; + } else { + // If we're using the old oom_adj file, the allowed range is now + // [0, kMaxOldOomScore], so we scale the score. This may result in some + // aliasing of values, of course. + score = score * kMaxOldOomScore / kMaxOomScore; + } + } + close(dirfd); + + char buf[11]; // 0 <= |score| <= kMaxOomScore; using log_10(2**32) + 1 size + snprintf(buf, sizeof(buf), "%d", score); + size_t len = strlen(buf); + + ssize_t bytes_written = write(fd, buf, len); + close(fd); + return (bytes_written == (ssize_t)len); +} diff --git a/sandbox/linux/suid/sandbox.c b/sandbox/linux/suid/sandbox.c new file mode 100644 index 0000000000..3049ae5211 --- /dev/null +++ b/sandbox/linux/suid/sandbox.c @@ -0,0 +1,480 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// http://code.google.com/p/chromium/wiki/LinuxSUIDSandbox + +#include "sandbox/linux/suid/common/sandbox.h" + +#define _GNU_SOURCE +#include <asm/unistd.h> +#include <errno.h> +#include <fcntl.h> +#include <limits.h> +#include <sched.h> +#include <signal.h> +#include <stdarg.h> +#include <stdbool.h> +#include <stdint.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <sys/prctl.h> +#include <sys/resource.h> +#include <sys/socket.h> +#include <sys/stat.h> +#include <sys/time.h> +#include <sys/types.h> +#include <sys/vfs.h> +#include <sys/wait.h> +#include <unistd.h> + +#include "sandbox/linux/suid/common/suid_unsafe_environment_variables.h" +#include "sandbox/linux/suid/process_util.h" + +#if !defined(CLONE_NEWPID) +#define CLONE_NEWPID 0x20000000 +#endif +#if !defined(CLONE_NEWNET) +#define CLONE_NEWNET 0x40000000 +#endif + +static bool DropRoot(); + +#define HANDLE_EINTR(x) TEMP_FAILURE_RETRY(x) + +static void FatalError(const char* msg, ...) + __attribute__((noreturn, format(printf, 1, 2))); + +static void FatalError(const char* msg, ...) { + va_list ap; + va_start(ap, msg); + + vfprintf(stderr, msg, ap); + fprintf(stderr, ": %s\n", strerror(errno)); + fflush(stderr); + va_end(ap); + _exit(1); +} + +static void ExitWithErrorSignalHandler(int signal) { + const char msg[] = "\nThe setuid sandbox got signaled, exiting.\n"; + if (-1 == write(2, msg, sizeof(msg) - 1)) { + // Do nothing. + } + + _exit(1); +} + +// We will chroot() to the helper's /proc/self directory. Anything there will +// not exist anymore if we make sure to wait() for the helper. +// +// /proc/self/fdinfo or /proc/self/fd are especially safe and will be empty +// even if the helper survives as a zombie. +// +// There is very little reason to use fdinfo/ instead of fd/ but we are +// paranoid. fdinfo/ only exists since 2.6.22 so we allow fallback to fd/ +#define SAFE_DIR "/proc/self/fdinfo" +#define SAFE_DIR2 "/proc/self/fd" + +static bool SpawnChrootHelper() { + int sv[2]; + if (socketpair(AF_UNIX, SOCK_STREAM, 0, sv) == -1) { + perror("socketpair"); + return false; + } + + char* safedir = NULL; + struct stat sdir_stat; + if (!stat(SAFE_DIR, &sdir_stat) && S_ISDIR(sdir_stat.st_mode)) { + safedir = SAFE_DIR; + } else if (!stat(SAFE_DIR2, &sdir_stat) && S_ISDIR(sdir_stat.st_mode)) { + safedir = SAFE_DIR2; + } else { + fprintf(stderr, "Could not find %s\n", SAFE_DIR2); + return false; + } + + const pid_t pid = syscall(__NR_clone, CLONE_FS | SIGCHLD, 0, 0, 0); + + if (pid == -1) { + perror("clone"); + close(sv[0]); + close(sv[1]); + return false; + } + + if (pid == 0) { + // We share our files structure with an untrusted process. As a security in + // depth measure, we make sure that we can't open anything by mistake. + // TODO(agl): drop CAP_SYS_RESOURCE / use SECURE_NOROOT + + const struct rlimit nofile = {0, 0}; + if (setrlimit(RLIMIT_NOFILE, &nofile)) + FatalError("Setting RLIMIT_NOFILE"); + + if (close(sv[1])) + FatalError("close"); + + // wait for message + char msg; + ssize_t bytes; + do { + bytes = read(sv[0], &msg, 1); + } while (bytes == -1 && errno == EINTR); + + if (bytes == 0) + _exit(0); + if (bytes != 1) + FatalError("read"); + + // do chrooting + if (msg != kMsgChrootMe) + FatalError("Unknown message from sandboxed process"); + + // sanity check + if (chdir(safedir)) + FatalError("Cannot chdir into /proc/ directory"); + + if (chroot(safedir)) + FatalError("Cannot chroot into /proc/ directory"); + + if (chdir("/")) + FatalError("Cannot chdir to / after chroot"); + + const char reply = kMsgChrootSuccessful; + do { + bytes = write(sv[0], &reply, 1); + } while (bytes == -1 && errno == EINTR); + + if (bytes != 1) + FatalError("Writing reply"); + + _exit(0); + // We now become a zombie. /proc/self/fd(info) is now an empty dir and we + // are chrooted there. + // Our (unprivileged) parent should not even be able to open "." or "/" + // since they would need to pass the ptrace() check. If our parent wait() + // for us, our root directory will completely disappear. + } + + if (close(sv[0])) { + close(sv[1]); + perror("close"); + return false; + } + + // In the parent process, we install an environment variable containing the + // number of the file descriptor. + char desc_str[64]; + int printed = snprintf(desc_str, sizeof(desc_str), "%u", sv[1]); + if (printed < 0 || printed >= (int)sizeof(desc_str)) { + fprintf(stderr, "Failed to snprintf\n"); + return false; + } + + if (setenv(kSandboxDescriptorEnvironmentVarName, desc_str, 1)) { + perror("setenv"); + close(sv[1]); + return false; + } + + // We also install an environment variable containing the pid of the child + char helper_pid_str[64]; + printed = snprintf(helper_pid_str, sizeof(helper_pid_str), "%u", pid); + if (printed < 0 || printed >= (int)sizeof(helper_pid_str)) { + fprintf(stderr, "Failed to snprintf\n"); + return false; + } + + if (setenv(kSandboxHelperPidEnvironmentVarName, helper_pid_str, 1)) { + perror("setenv"); + close(sv[1]); + return false; + } + + return true; +} + +// Block until child_pid exits, then exit. Try to preserve the exit code. +static void WaitForChildAndExit(pid_t child_pid) { + int exit_code = -1; + siginfo_t reaped_child_info; + + // Don't "Core" on SIGABRT. SIGABRT is sent by the Chrome OS session manager + // when things are hanging. + // Here, the current process is going to waitid() and _exit(), so there is no + // point in generating a crash report. The child process is the one + // blocking us. + if (signal(SIGABRT, ExitWithErrorSignalHandler) == SIG_ERR) { + FatalError("Failed to change signal handler"); + } + + int wait_ret = + HANDLE_EINTR(waitid(P_PID, child_pid, &reaped_child_info, WEXITED)); + + if (!wait_ret && reaped_child_info.si_pid == child_pid) { + if (reaped_child_info.si_code == CLD_EXITED) { + exit_code = reaped_child_info.si_status; + } else { + // Exit with code 0 if the child got signaled. + exit_code = 0; + } + } + _exit(exit_code); +} + +static bool MoveToNewNamespaces() { + // These are the sets of flags which we'll try, in order. + const int kCloneExtraFlags[] = {CLONE_NEWPID | CLONE_NEWNET, CLONE_NEWPID, }; + + // We need to close kZygoteIdFd before the child can continue. We use this + // socketpair to tell the child when to continue; + int sync_fds[2]; + if (socketpair(AF_UNIX, SOCK_STREAM, 0, sync_fds)) { + FatalError("Failed to create a socketpair"); + } + + for (size_t i = 0; i < sizeof(kCloneExtraFlags) / sizeof(kCloneExtraFlags[0]); + i++) { + pid_t pid = syscall(__NR_clone, SIGCHLD | kCloneExtraFlags[i], 0, 0, 0); + const int clone_errno = errno; + + if (pid > 0) { + if (!DropRoot()) { + FatalError("Could not drop privileges"); + } else { + if (close(sync_fds[0]) || shutdown(sync_fds[1], SHUT_RD)) + FatalError("Could not close socketpair"); + // The kZygoteIdFd needs to be closed in the parent before + // Zygote gets started. + if (close(kZygoteIdFd)) + FatalError("close"); + // Tell our child to continue + if (HANDLE_EINTR(send(sync_fds[1], "C", 1, MSG_NOSIGNAL)) != 1) + FatalError("send"); + if (close(sync_fds[1])) + FatalError("close"); + // We want to keep a full process tree and we don't want our childs to + // be reparented to (the outer PID namespace) init. So we wait for it. + WaitForChildAndExit(pid); + } + // NOTREACHED + FatalError("Not reached"); + } + + if (pid == 0) { + if (close(sync_fds[1]) || shutdown(sync_fds[0], SHUT_WR)) + FatalError("Could not close socketpair"); + + // Wait for the parent to confirm it closed kZygoteIdFd before we + // continue + char should_continue; + if (HANDLE_EINTR(read(sync_fds[0], &should_continue, 1)) != 1) + FatalError("Read on socketpair"); + if (close(sync_fds[0])) + FatalError("close"); + + if (kCloneExtraFlags[i] & CLONE_NEWPID) { + setenv(kSandboxPIDNSEnvironmentVarName, "", 1 /* overwrite */); + } else { + unsetenv(kSandboxPIDNSEnvironmentVarName); + } + + if (kCloneExtraFlags[i] & CLONE_NEWNET) { + setenv(kSandboxNETNSEnvironmentVarName, "", 1 /* overwrite */); + } else { + unsetenv(kSandboxNETNSEnvironmentVarName); + } + + break; + } + + // If EINVAL then the system doesn't support the requested flags, so + // continue to try a different set. + // On any other errno value the system *does* support these flags but + // something went wrong, hence we bail with an error message rather then + // provide less security. + if (errno != EINVAL) { + fprintf(stderr, "Failed to move to new namespace:"); + if (kCloneExtraFlags[i] & CLONE_NEWPID) { + fprintf(stderr, " PID namespaces supported,"); + } + if (kCloneExtraFlags[i] & CLONE_NEWNET) { + fprintf(stderr, " Network namespace supported,"); + } + fprintf(stderr, " but failed: errno = %s\n", strerror(clone_errno)); + return false; + } + } + + // If the system doesn't support NEWPID then we carry on anyway. + return true; +} + +static bool DropRoot() { + if (prctl(PR_SET_DUMPABLE, 0, 0, 0, 0)) { + perror("prctl(PR_SET_DUMPABLE)"); + return false; + } + + if (prctl(PR_GET_DUMPABLE, 0, 0, 0, 0)) { + perror("Still dumpable after prctl(PR_SET_DUMPABLE)"); + return false; + } + + gid_t rgid, egid, sgid; + if (getresgid(&rgid, &egid, &sgid)) { + perror("getresgid"); + return false; + } + + if (setresgid(rgid, rgid, rgid)) { + perror("setresgid"); + return false; + } + + uid_t ruid, euid, suid; + if (getresuid(&ruid, &euid, &suid)) { + perror("getresuid"); + return false; + } + + if (setresuid(ruid, ruid, ruid)) { + perror("setresuid"); + return false; + } + + return true; +} + +static bool SetupChildEnvironment() { + unsigned i; + + // ld.so may have cleared several environment variables because we are SUID. + // However, the child process might need them so zygote_host_linux.cc saves a + // copy in SANDBOX_$x. This is safe because we have dropped root by this + // point, so we can only exec a binary with the permissions of the user who + // ran us in the first place. + + for (i = 0; kSUIDUnsafeEnvironmentVariables[i]; ++i) { + const char* const envvar = kSUIDUnsafeEnvironmentVariables[i]; + char* const saved_envvar = SandboxSavedEnvironmentVariable(envvar); + if (!saved_envvar) + return false; + + const char* const value = getenv(saved_envvar); + if (value) { + setenv(envvar, value, 1 /* overwrite */); + unsetenv(saved_envvar); + } + + free(saved_envvar); + } + + return true; +} + +bool CheckAndExportApiVersion() { + // Check the environment to see if a specific API version was requested. + // assume version 0 if none. + long api_number = -1; + char* api_string = getenv(kSandboxEnvironmentApiRequest); + if (!api_string) { + api_number = 0; + } else { + errno = 0; + char* endptr = NULL; + api_number = strtol(api_string, &endptr, 10); + if (!endptr || *endptr || errno != 0) + return false; + } + + // Warn only for now. + if (api_number != kSUIDSandboxApiNumber) { + fprintf( + stderr, + "The setuid sandbox provides API version %ld, " + "but you need %ld\n" + "Please read " + "https://code.google.com/p/chromium/wiki/LinuxSUIDSandboxDevelopment." + "\n\n", + kSUIDSandboxApiNumber, + api_number); + } + + // Export our version so that the sandboxed process can verify it did not + // use an old sandbox. + char version_string[64]; + snprintf( + version_string, sizeof(version_string), "%ld", kSUIDSandboxApiNumber); + if (setenv(kSandboxEnvironmentApiProvides, version_string, 1)) { + perror("setenv"); + return false; + } + + return true; +} + +int main(int argc, char** argv) { + if (argc <= 1) { + if (argc <= 0) { + return 1; + } + + fprintf(stderr, "Usage: %s <renderer process> <args...>\n", argv[0]); + return 1; + } + + // Allow someone to query our API version + if (argc == 2 && 0 == strcmp(argv[1], kSuidSandboxGetApiSwitch)) { + printf("%ld\n", kSUIDSandboxApiNumber); + return 0; + } + + // We cannot adjust /proc/pid/oom_adj for sandboxed renderers + // because those files are owned by root. So we need a helper here. + if (argc == 4 && (0 == strcmp(argv[1], kAdjustOOMScoreSwitch))) { + char* endptr = NULL; + long score; + errno = 0; + unsigned long pid_ul = strtoul(argv[2], &endptr, 10); + if (pid_ul == ULONG_MAX || !endptr || *endptr || errno != 0) + return 1; + pid_t pid = pid_ul; + endptr = NULL; + errno = 0; + score = strtol(argv[3], &endptr, 10); + if (score == LONG_MAX || score == LONG_MIN || !endptr || *endptr || + errno != 0) { + return 1; + } + return AdjustOOMScore(pid, score); + } + + // Protect the core setuid sandbox functionality with an API version + if (!CheckAndExportApiVersion()) { + return 1; + } + + if (geteuid() != 0) { + fprintf(stderr, + "The setuid sandbox is not running as root. Common causes:\n" + " * An unprivileged process using ptrace on it, like a debugger.\n" + " * A parent process set prctl(PR_SET_NO_NEW_PRIVS, ...)\n"); + } + + if (!MoveToNewNamespaces()) + return 1; + if (!SpawnChrootHelper()) + return 1; + if (!DropRoot()) + return 1; + if (!SetupChildEnvironment()) + return 1; + + execv(argv[1], &argv[1]); + FatalError("execv failed"); + + return 1; +} diff --git a/sandbox/linux/syscall_broker/DEPS b/sandbox/linux/syscall_broker/DEPS new file mode 100644 index 0000000000..70d9b18aa1 --- /dev/null +++ b/sandbox/linux/syscall_broker/DEPS @@ -0,0 +1,3 @@ +include_rules = [ + "+sandbox/linux/system_headers", +] diff --git a/sandbox/linux/syscall_broker/broker_channel.cc b/sandbox/linux/syscall_broker/broker_channel.cc new file mode 100644 index 0000000000..fa0f7615fc --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_channel.cc @@ -0,0 +1,35 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/syscall_broker/broker_channel.h" + +#include <sys/socket.h> +#include <sys/types.h> + +#include "base/logging.h" + +namespace sandbox { + +namespace syscall_broker { + +// static +void BrokerChannel::CreatePair(EndPoint* reader, EndPoint* writer) { + DCHECK(reader); + DCHECK(writer); + int socket_pair[2]; + // Use SOCK_SEQPACKET, to preserve message boundaries but we also want to be + // notified (recvmsg should return and not block) when the connection has + // been broken which could mean that the other end has been closed. + PCHECK(0 == socketpair(AF_UNIX, SOCK_SEQPACKET, 0, socket_pair)); + + reader->reset(socket_pair[0]); + PCHECK(0 == shutdown(reader->get(), SHUT_WR)); + + writer->reset(socket_pair[1]); + PCHECK(0 == shutdown(writer->get(), SHUT_RD)); +} + +} // namespace syscall_broker + +} // namespace sandbox diff --git a/sandbox/linux/syscall_broker/broker_channel.h b/sandbox/linux/syscall_broker/broker_channel.h new file mode 100644 index 0000000000..2abdba413a --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_channel.h @@ -0,0 +1,31 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSCALL_BROKER_BROKER_CHANNEL_H_ +#define SANDBOX_LINUX_SYSCALL_BROKER_BROKER_CHANNEL_H_ + +#include "base/files/scoped_file.h" +#include "base/macros.h" + +namespace sandbox { + +namespace syscall_broker { + +// A small class to create a pipe-like communication channel. It is based on a +// SOCK_SEQPACKET unix socket, which is connection-based and guaranteed to +// preserve message boundaries. +class BrokerChannel { + public: + typedef base::ScopedFD EndPoint; + static void CreatePair(EndPoint* reader, EndPoint* writer); + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(BrokerChannel); +}; + +} // namespace syscall_broker + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SYSCALL_BROKER_BROKER_CHANNEL_H_ diff --git a/sandbox/linux/syscall_broker/broker_client.cc b/sandbox/linux/syscall_broker/broker_client.cc new file mode 100644 index 0000000000..760cf59b3c --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_client.cc @@ -0,0 +1,144 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/syscall_broker/broker_client.h" + +#include <errno.h> +#include <fcntl.h> +#include <sys/stat.h> +#include <sys/socket.h> +#include <sys/types.h> + +#include "build/build_config.h" +#include "base/logging.h" +#include "base/pickle.h" +#include "base/posix/unix_domain_socket_linux.h" +#include "sandbox/linux/syscall_broker/broker_channel.h" +#include "sandbox/linux/syscall_broker/broker_common.h" +#include "sandbox/linux/syscall_broker/broker_policy.h" + +#if defined(OS_ANDROID) && !defined(MSG_CMSG_CLOEXEC) +#define MSG_CMSG_CLOEXEC 0x40000000 +#endif + +namespace sandbox { + +namespace syscall_broker { + +// Make a remote system call over IPC for syscalls that take a path and flags +// as arguments, currently open() and access(). +// Will return -errno like a real system call. +// This function needs to be async signal safe. +int BrokerClient::PathAndFlagsSyscall(IPCCommand syscall_type, + const char* pathname, + int flags) const { + int recvmsg_flags = 0; + RAW_CHECK(syscall_type == COMMAND_OPEN || syscall_type == COMMAND_ACCESS); + if (!pathname) + return -EFAULT; + + // For this "remote system call" to work, we need to handle any flag that + // cannot be sent over a Unix socket in a special way. + // See the comments around kCurrentProcessOpenFlagsMask. + if (syscall_type == COMMAND_OPEN && (flags & kCurrentProcessOpenFlagsMask)) { + // This implementation only knows about O_CLOEXEC, someone needs to look at + // this code if other flags are added. + RAW_CHECK(kCurrentProcessOpenFlagsMask == O_CLOEXEC); + recvmsg_flags |= MSG_CMSG_CLOEXEC; + flags &= ~O_CLOEXEC; + } + + // There is no point in forwarding a request that we know will be denied. + // Of course, the real security check needs to be on the other side of the + // IPC. + if (fast_check_in_client_) { + if (syscall_type == COMMAND_OPEN && + !broker_policy_.GetFileNameIfAllowedToOpen( + pathname, flags, NULL /* file_to_open */, + NULL /* unlink_after_open */)) { + return -broker_policy_.denied_errno(); + } + if (syscall_type == COMMAND_ACCESS && + !broker_policy_.GetFileNameIfAllowedToAccess(pathname, flags, NULL)) { + return -broker_policy_.denied_errno(); + } + } + + base::Pickle write_pickle; + write_pickle.WriteInt(syscall_type); + write_pickle.WriteString(pathname); + write_pickle.WriteInt(flags); + RAW_CHECK(write_pickle.size() <= kMaxMessageLength); + + int returned_fd = -1; + uint8_t reply_buf[kMaxMessageLength]; + + // Send a request (in write_pickle) as well that will include a new + // temporary socketpair (created internally by SendRecvMsg()). + // Then read the reply on this new socketpair in reply_buf and put an + // eventual attached file descriptor in |returned_fd|. + ssize_t msg_len = base::UnixDomainSocket::SendRecvMsgWithFlags( + ipc_channel_.get(), reply_buf, sizeof(reply_buf), recvmsg_flags, + &returned_fd, write_pickle); + if (msg_len <= 0) { + if (!quiet_failures_for_tests_) + RAW_LOG(ERROR, "Could not make request to broker process"); + return -ENOMEM; + } + + base::Pickle read_pickle(reinterpret_cast<char*>(reply_buf), msg_len); + base::PickleIterator iter(read_pickle); + int return_value = -1; + // Now deserialize the return value and eventually return the file + // descriptor. + if (iter.ReadInt(&return_value)) { + switch (syscall_type) { + case COMMAND_ACCESS: + // We should never have a fd to return. + RAW_CHECK(returned_fd == -1); + return return_value; + case COMMAND_OPEN: + if (return_value < 0) { + RAW_CHECK(returned_fd == -1); + return return_value; + } else { + // We have a real file descriptor to return. + RAW_CHECK(returned_fd >= 0); + return returned_fd; + } + default: + RAW_LOG(ERROR, "Unsupported command"); + return -ENOSYS; + } + } else { + RAW_LOG(ERROR, "Could not read pickle"); + NOTREACHED(); + return -ENOMEM; + } +} + +BrokerClient::BrokerClient(const BrokerPolicy& broker_policy, + BrokerChannel::EndPoint ipc_channel, + bool fast_check_in_client, + bool quiet_failures_for_tests) + : broker_policy_(broker_policy), + ipc_channel_(ipc_channel.Pass()), + fast_check_in_client_(fast_check_in_client), + quiet_failures_for_tests_(quiet_failures_for_tests) { +} + +BrokerClient::~BrokerClient() { +} + +int BrokerClient::Access(const char* pathname, int mode) const { + return PathAndFlagsSyscall(COMMAND_ACCESS, pathname, mode); +} + +int BrokerClient::Open(const char* pathname, int flags) const { + return PathAndFlagsSyscall(COMMAND_OPEN, pathname, flags); +} + +} // namespace syscall_broker + +} // namespace sandbox diff --git a/sandbox/linux/syscall_broker/broker_client.h b/sandbox/linux/syscall_broker/broker_client.h new file mode 100644 index 0000000000..2dfef8150c --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_client.h @@ -0,0 +1,75 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSCALL_BROKER_BROKER_CLIENT_H_ +#define SANDBOX_LINUX_SYSCALL_BROKER_BROKER_CLIENT_H_ + +#include "base/macros.h" +#include "sandbox/linux/syscall_broker/broker_channel.h" +#include "sandbox/linux/syscall_broker/broker_common.h" + +namespace sandbox { + +namespace syscall_broker { + +class BrokerPolicy; + +// This class can be embedded in a sandboxed process and can be +// used to perform certain system calls in another, presumably +// non-sandboxed process (which embeds BrokerHost). +// A key feature of this class is the ability to use some of its methods in a +// thread-safe and async-signal safe way. The goal is to be able to use it to +// replace the open() or access() system calls happening anywhere in a process +// (as allowed for instance by seccomp-bpf's SIGSYS mechanism). +class BrokerClient { + public: + // |policy| needs to match the policy used by BrokerHost. This + // allows to predict some of the requests which will be denied + // and save an IPC round trip. + // |ipc_channel| needs to be a suitable SOCK_SEQPACKET unix socket. + // |fast_check_in_client| should be set to true and + // |quiet_failures_for_tests| to false unless you are writing tests. + BrokerClient(const BrokerPolicy& policy, + BrokerChannel::EndPoint ipc_channel, + bool fast_check_in_client, + bool quiet_failures_for_tests); + ~BrokerClient(); + + // Can be used in place of access(). + // X_OK will always return an error in practice since the broker process + // doesn't support execute permissions. + // It's similar to the access() system call and will return -errno on errors. + // This is async signal safe. + int Access(const char* pathname, int mode) const; + // Can be used in place of open(). + // The implementation only supports certain white listed flags and will + // return -EPERM on other flags. + // It's similar to the open() system call and will return -errno on errors. + // This is async signal safe. + int Open(const char* pathname, int flags) const; + + // Get the file descriptor used for IPC. This is used for tests. + int GetIPCDescriptor() const { return ipc_channel_.get(); } + + private: + const BrokerPolicy& broker_policy_; + const BrokerChannel::EndPoint ipc_channel_; + const bool fast_check_in_client_; // Whether to forward a request that we + // know will be denied to the broker. (Used + // for tests). + const bool quiet_failures_for_tests_; // Disable certain error message when + // testing for failures. + + int PathAndFlagsSyscall(IPCCommand syscall_type, + const char* pathname, + int flags) const; + + DISALLOW_COPY_AND_ASSIGN(BrokerClient); +}; + +} // namespace syscall_broker + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SYSCALL_BROKER_BROKER_CLIENT_H_ diff --git a/sandbox/linux/syscall_broker/broker_common.h b/sandbox/linux/syscall_broker/broker_common.h new file mode 100644 index 0000000000..25aafa7ed2 --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_common.h @@ -0,0 +1,41 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSCALL_BROKER_BROKER_COMMON_H_ +#define SANDBOX_LINUX_SYSCALL_BROKER_BROKER_COMMON_H_ + +#include <fcntl.h> +#include <stddef.h> + +namespace sandbox { + +namespace syscall_broker { + +const size_t kMaxMessageLength = 4096; + +// Some flags are local to the current process and cannot be sent over a Unix +// socket. They need special treatment from the client. +// O_CLOEXEC is tricky because in theory another thread could call execve() +// before special treatment is made on the client, so a client needs to call +// recvmsg(2) with MSG_CMSG_CLOEXEC. +// To make things worse, there are two CLOEXEC related flags, FD_CLOEXEC (see +// F_GETFD in fcntl(2)) and O_CLOEXEC (see F_GETFL in fcntl(2)). O_CLOEXEC +// doesn't affect the semantics on execve(), it's merely a note that the +// descriptor was originally opened with O_CLOEXEC as a flag. And it is sent +// over unix sockets just fine, so a receiver that would (incorrectly) look at +// O_CLOEXEC instead of FD_CLOEXEC may be tricked in thinking that the file +// descriptor will or won't be closed on execve(). +const int kCurrentProcessOpenFlagsMask = O_CLOEXEC; + +enum IPCCommand { + COMMAND_INVALID = 0, + COMMAND_OPEN, + COMMAND_ACCESS, +}; + +} // namespace syscall_broker + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SYSCALL_BROKER_BROKER_COMMON_H_ diff --git a/sandbox/linux/syscall_broker/broker_file_permission.cc b/sandbox/linux/syscall_broker/broker_file_permission.cc new file mode 100644 index 0000000000..beceda93f5 --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_file_permission.cc @@ -0,0 +1,243 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/syscall_broker/broker_file_permission.h" + +#include <fcntl.h> +#include <string.h> + +#include <string> + +#include "base/logging.h" +#include "sandbox/linux/syscall_broker/broker_common.h" + +namespace sandbox { + +namespace syscall_broker { + +// Async signal safe +bool BrokerFilePermission::ValidatePath(const char* path) { + if (!path) + return false; + + const size_t len = strlen(path); + // No empty paths + if (len == 0) + return false; + // Paths must be absolute and not relative + if (path[0] != '/') + return false; + // No trailing / (but "/" is valid) + if (len > 1 && path[len - 1] == '/') + return false; + // No trailing /.. + if (len >= 3 && path[len - 3] == '/' && path[len - 2] == '.' && + path[len - 1] == '.') + return false; + // No /../ anywhere + for (size_t i = 0; i < len; i++) { + if (path[i] == '/' && (len - i) > 3) { + if (path[i + 1] == '.' && path[i + 2] == '.' && path[i + 3] == '/') { + return false; + } + } + } + return true; +} + +// Async signal safe +// Calls std::string::c_str(), strncmp and strlen. All these +// methods are async signal safe in common standard libs. +// TODO(leecam): remove dependency on std::string +bool BrokerFilePermission::MatchPath(const char* requested_filename) const { + const char* path = path_.c_str(); + if ((recursive_ && strncmp(requested_filename, path, strlen(path)) == 0)) { + // Note: This prefix match will allow any path under the whitelisted + // path, for any number of directory levels. E.g. if the whitelisted + // path is /good/ then the following will be permitted by the policy. + // /good/file1 + // /good/folder/file2 + // /good/folder/folder2/file3 + // If an attacker could make 'folder' a symlink to ../../ they would have + // access to the entire filesystem. + // Whitelisting with multiple depths is useful, e.g /proc/ but + // the system needs to ensure symlinks can not be created! + // That said if an attacker can convert any of the absolute paths + // to a symlink they can control any file on the system also. + return true; + } else if (strcmp(requested_filename, path) == 0) { + return true; + } + return false; +} + +// Async signal safe. +// External call to std::string::c_str() is +// called in MatchPath. +// TODO(leecam): remove dependency on std::string +bool BrokerFilePermission::CheckAccess(const char* requested_filename, + int mode, + const char** file_to_access) const { + // First, check if |mode| is existence, ability to read or ability + // to write. We do not support X_OK. + if (mode != F_OK && mode & ~(R_OK | W_OK)) { + return false; + } + + if (!ValidatePath(requested_filename)) + return false; + + if (!MatchPath(requested_filename)) { + return false; + } + bool allowed = false; + switch (mode) { + case F_OK: + if (allow_read_ || allow_write_) + allowed = true; + break; + case R_OK: + if (allow_read_) + allowed = true; + break; + case W_OK: + if (allow_write_) + allowed = true; + break; + case R_OK | W_OK: + if (allow_read_ && allow_write_) + allowed = true; + break; + default: + return false; + } + + if (allowed && file_to_access) { + if (!recursive_) + *file_to_access = path_.c_str(); + else + *file_to_access = requested_filename; + } + return allowed; +} + +// Async signal safe. +// External call to std::string::c_str() is +// called in MatchPath. +// TODO(leecam): remove dependency on std::string +bool BrokerFilePermission::CheckOpen(const char* requested_filename, + int flags, + const char** file_to_open, + bool* unlink_after_open) const { + if (!ValidatePath(requested_filename)) + return false; + + if (!MatchPath(requested_filename)) { + return false; + } + + // First, check the access mode is valid. + const int access_mode = flags & O_ACCMODE; + if (access_mode != O_RDONLY && access_mode != O_WRONLY && + access_mode != O_RDWR) { + return false; + } + + // Check if read is allowed + if (!allow_read_ && (access_mode == O_RDONLY || access_mode == O_RDWR)) { + return false; + } + + // Check if write is allowed + if (!allow_write_ && (access_mode == O_WRONLY || access_mode == O_RDWR)) { + return false; + } + + // Check if file creation is allowed. + if (!allow_create_ && (flags & O_CREAT)) { + return false; + } + + // If O_CREAT is present, ensure O_EXCL + if ((flags & O_CREAT) && !(flags & O_EXCL)) { + return false; + } + + // If this file is to be unlinked, ensure it's created. + if (unlink_ && !(flags & O_CREAT)) { + return false; + } + + // Some flags affect the behavior of the current process. We don't support + // them and don't allow them for now. + if (flags & kCurrentProcessOpenFlagsMask) { + return false; + } + + // Now check that all the flags are known to us. + const int creation_and_status_flags = flags & ~O_ACCMODE; + + const int known_flags = O_APPEND | O_ASYNC | O_CLOEXEC | O_CREAT | O_DIRECT | + O_DIRECTORY | O_EXCL | O_LARGEFILE | O_NOATIME | + O_NOCTTY | O_NOFOLLOW | O_NONBLOCK | O_NDELAY | + O_SYNC | O_TRUNC; + + const int unknown_flags = ~known_flags; + const bool has_unknown_flags = creation_and_status_flags & unknown_flags; + + if (has_unknown_flags) + return false; + + if (file_to_open) { + if (!recursive_) + *file_to_open = path_.c_str(); + else + *file_to_open = requested_filename; + } + if (unlink_after_open) + *unlink_after_open = unlink_; + + return true; +} + +const char* BrokerFilePermission::GetErrorMessageForTests() { + static char kInvalidBrokerFileString[] = "Invalid BrokerFilePermission"; + return kInvalidBrokerFileString; +} + +BrokerFilePermission::BrokerFilePermission(const std::string& path, + bool recursive, + bool unlink, + bool allow_read, + bool allow_write, + bool allow_create) + : path_(path), + recursive_(recursive), + unlink_(unlink), + allow_read_(allow_read), + allow_write_(allow_write), + allow_create_(allow_create) { + // Validate this permission and die if invalid! + + // Must have enough length for a '/' + CHECK(path_.length() > 0) << GetErrorMessageForTests(); + // Whitelisted paths must be absolute. + CHECK(path_[0] == '/') << GetErrorMessageForTests(); + + // Don't allow unlinking on creation without create permission + if (unlink_) { + CHECK(allow_create) << GetErrorMessageForTests(); + } + const char last_char = *(path_.rbegin()); + // Recursive paths must have a trailing slash + if (recursive_) { + CHECK(last_char == '/') << GetErrorMessageForTests(); + } else { + CHECK(last_char != '/') << GetErrorMessageForTests(); + } +} + +} // namespace syscall_broker + +} // namespace sandbox
\ No newline at end of file diff --git a/sandbox/linux/syscall_broker/broker_file_permission.h b/sandbox/linux/syscall_broker/broker_file_permission.h new file mode 100644 index 0000000000..03300d1d74 --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_file_permission.h @@ -0,0 +1,119 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSCALL_BROKER_BROKER_FILE_PERMISSION_H_ +#define SANDBOX_LINUX_SYSCALL_BROKER_BROKER_FILE_PERMISSION_H_ + +#include <string> + +#include "base/macros.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +namespace syscall_broker { + +// BrokerFilePermission defines a path for whitelisting. +// Pick the correct static factory method to create a permission. +// CheckOpen and CheckAccess are async signal safe. +// Constuction and Destruction are not async signal safe. +// |path| is the path to be whitelisted. +class SANDBOX_EXPORT BrokerFilePermission { + public: + ~BrokerFilePermission() {} + BrokerFilePermission(const BrokerFilePermission&) = default; + BrokerFilePermission& operator=(const BrokerFilePermission&) = default; + + static BrokerFilePermission ReadOnly(const std::string& path) { + return BrokerFilePermission(path, false, false, true, false, false); + } + + static BrokerFilePermission ReadOnlyRecursive(const std::string& path) { + return BrokerFilePermission(path, true, false, true, false, false); + } + + static BrokerFilePermission WriteOnly(const std::string& path) { + return BrokerFilePermission(path, false, false, false, true, false); + } + + static BrokerFilePermission ReadWrite(const std::string& path) { + return BrokerFilePermission(path, false, false, true, true, false); + } + + static BrokerFilePermission ReadWriteCreate(const std::string& path) { + return BrokerFilePermission(path, false, false, true, true, true); + } + + static BrokerFilePermission ReadWriteCreateUnlink(const std::string& path) { + return BrokerFilePermission(path, false, true, true, true, true); + } + + static BrokerFilePermission ReadWriteCreateUnlinkRecursive( + const std::string& path) { + return BrokerFilePermission(path, true, true, true, true, true); + } + + // Returns true if |requested_filename| is allowed to be opened + // by this permission. + // If |file_to_open| is not NULL it is set to point to either + // the |requested_filename| in the case of a recursive match, + // or a pointer the matched path in the whitelist if an absolute + // match. + // If not NULL |unlink_after_open| is set to point to true if the + // caller should unlink the path after openning. + // Async signal safe if |file_to_open| is NULL. + bool CheckOpen(const char* requested_filename, + int flags, + const char** file_to_open, + bool* unlink_after_open) const; + // Returns true if |requested_filename| is allowed to be accessed + // by this permission as per access(2). + // If |file_to_open| is not NULL it is set to point to either + // the |requested_filename| in the case of a recursive match, + // or a pointer to the matched path in the whitelist if an absolute + // match. + // |mode| is per mode argument of access(2). + // Async signal safe if |file_to_access| is NULL + bool CheckAccess(const char* requested_filename, + int mode, + const char** file_to_access) const; + + private: + friend class BrokerFilePermissionTester; + BrokerFilePermission(const std::string& path, + bool recursive, + bool unlink, + bool allow_read, + bool allow_write, + bool allow_create); + + // ValidatePath checks |path| and returns true if these conditions are met + // * Greater than 0 length + // * Is an absolute path + // * No trailing slash + // * No /../ path traversal + static bool ValidatePath(const char* path); + + // MatchPath returns true if |requested_filename| is covered by this instance + bool MatchPath(const char* requested_filename) const; + + // Used in by BrokerFilePermissionTester for tests. + static const char* GetErrorMessageForTests(); + + // These are not const as std::vector requires copy-assignment and this class + // is stored in vectors. All methods are marked const so + // the compiler will still enforce no changes outside of the constructor. + std::string path_; + bool recursive_; // Allow everything under this path. |path| must be a dir. + bool unlink_; // unlink after opening. + bool allow_read_; + bool allow_write_; + bool allow_create_; +}; + +} // namespace syscall_broker + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SYSCALL_BROKER_BROKER_FILE_PERMISSION_H_
\ No newline at end of file diff --git a/sandbox/linux/syscall_broker/broker_file_permission_unittest.cc b/sandbox/linux/syscall_broker/broker_file_permission_unittest.cc new file mode 100644 index 0000000000..b58a901cde --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_file_permission_unittest.cc @@ -0,0 +1,262 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/syscall_broker/broker_file_permission.h" + +#include <fcntl.h> +#include <string.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/macros.h" +#include "sandbox/linux/tests/test_utils.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace syscall_broker { + +class BrokerFilePermissionTester { + public: + static bool ValidatePath(const char* path) { + return BrokerFilePermission::ValidatePath(path); + } + static const char* GetErrorMessage() { + return BrokerFilePermission::GetErrorMessageForTests(); + } + + private: + DISALLOW_COPY_AND_ASSIGN(BrokerFilePermissionTester); +}; + +namespace { + +// Creation tests are DEATH tests as a bad permission causes termination. +SANDBOX_TEST(BrokerFilePermission, CreateGood) { + const char kPath[] = "/tmp/good"; + BrokerFilePermission perm = BrokerFilePermission::ReadOnly(kPath); +} + +SANDBOX_TEST(BrokerFilePermission, CreateGoodRecursive) { + const char kPath[] = "/tmp/good/"; + BrokerFilePermission perm = BrokerFilePermission::ReadOnlyRecursive(kPath); +} + +SANDBOX_DEATH_TEST( + BrokerFilePermission, + CreateBad, + DEATH_MESSAGE(BrokerFilePermissionTester::GetErrorMessage())) { + const char kPath[] = "/tmp/bad/"; + BrokerFilePermission perm = BrokerFilePermission::ReadOnly(kPath); +} + +SANDBOX_DEATH_TEST( + BrokerFilePermission, + CreateBadRecursive, + DEATH_MESSAGE(BrokerFilePermissionTester::GetErrorMessage())) { + const char kPath[] = "/tmp/bad"; + BrokerFilePermission perm = BrokerFilePermission::ReadOnlyRecursive(kPath); +} + +SANDBOX_DEATH_TEST( + BrokerFilePermission, + CreateBadNotAbs, + DEATH_MESSAGE(BrokerFilePermissionTester::GetErrorMessage())) { + const char kPath[] = "tmp/bad"; + BrokerFilePermission perm = BrokerFilePermission::ReadOnly(kPath); +} + +SANDBOX_DEATH_TEST( + BrokerFilePermission, + CreateBadEmpty, + DEATH_MESSAGE(BrokerFilePermissionTester::GetErrorMessage())) { + const char kPath[] = ""; + BrokerFilePermission perm = BrokerFilePermission::ReadOnly(kPath); +} + +// CheckPerm tests |path| against |perm| given |access_flags|. +// If |create| is true then file creation is tested for success. +void CheckPerm(const BrokerFilePermission& perm, + const char* path, + int access_flags, + bool create) { + const char* file_to_open = NULL; + + ASSERT_FALSE(perm.CheckAccess(path, X_OK, NULL)); + ASSERT_TRUE(perm.CheckAccess(path, F_OK, NULL)); + // check bad perms + switch (access_flags) { + case O_RDONLY: + ASSERT_TRUE(perm.CheckOpen(path, O_RDONLY, &file_to_open, NULL)); + ASSERT_FALSE(perm.CheckOpen(path, O_WRONLY, &file_to_open, NULL)); + ASSERT_FALSE(perm.CheckOpen(path, O_RDWR, &file_to_open, NULL)); + ASSERT_TRUE(perm.CheckAccess(path, R_OK, NULL)); + ASSERT_FALSE(perm.CheckAccess(path, W_OK, NULL)); + break; + case O_WRONLY: + ASSERT_FALSE(perm.CheckOpen(path, O_RDONLY, &file_to_open, NULL)); + ASSERT_TRUE(perm.CheckOpen(path, O_WRONLY, &file_to_open, NULL)); + ASSERT_FALSE(perm.CheckOpen(path, O_RDWR, &file_to_open, NULL)); + ASSERT_FALSE(perm.CheckAccess(path, R_OK, NULL)); + ASSERT_TRUE(perm.CheckAccess(path, W_OK, NULL)); + break; + case O_RDWR: + ASSERT_TRUE(perm.CheckOpen(path, O_RDONLY, &file_to_open, NULL)); + ASSERT_TRUE(perm.CheckOpen(path, O_WRONLY, &file_to_open, NULL)); + ASSERT_TRUE(perm.CheckOpen(path, O_RDWR, &file_to_open, NULL)); + ASSERT_TRUE(perm.CheckAccess(path, R_OK, NULL)); + ASSERT_TRUE(perm.CheckAccess(path, W_OK, NULL)); + break; + default: + // Bad test case + NOTREACHED(); + } + +// O_SYNC can be defined as (__O_SYNC|O_DSYNC) +#ifdef O_DSYNC + const int kSyncFlag = O_SYNC & ~O_DSYNC; +#else + const int kSyncFlag = O_SYNC; +#endif + + const int kNumberOfBitsInOAccMode = 2; + static_assert(O_ACCMODE == ((1 << kNumberOfBitsInOAccMode) - 1), + "incorrect number of bits"); + // check every possible flag and act accordingly. + // Skipping AccMode bits as they are present in every case. + for (int i = kNumberOfBitsInOAccMode; i < 32; i++) { + int flag = 1 << i; + switch (flag) { + case O_APPEND: + case O_ASYNC: + case O_DIRECT: + case O_DIRECTORY: +#ifdef O_DSYNC + case O_DSYNC: +#endif + case O_EXCL: + case O_LARGEFILE: + case O_NOATIME: + case O_NOCTTY: + case O_NOFOLLOW: + case O_NONBLOCK: +#if (O_NONBLOCK != O_NDELAY) + case O_NDELAY: +#endif + case kSyncFlag: + case O_TRUNC: + ASSERT_TRUE( + perm.CheckOpen(path, access_flags | flag, &file_to_open, NULL)); + break; + case O_CLOEXEC: + case O_CREAT: + default: + ASSERT_FALSE( + perm.CheckOpen(path, access_flags | flag, &file_to_open, NULL)); + } + } + if (create) { + bool unlink; + ASSERT_TRUE(perm.CheckOpen(path, O_CREAT | O_EXCL | access_flags, + &file_to_open, &unlink)); + ASSERT_FALSE(unlink); + } else { + ASSERT_FALSE(perm.CheckOpen(path, O_CREAT | O_EXCL | access_flags, + &file_to_open, NULL)); + } +} + +TEST(BrokerFilePermission, ReadOnly) { + const char kPath[] = "/tmp/good"; + BrokerFilePermission perm = BrokerFilePermission::ReadOnly(kPath); + CheckPerm(perm, kPath, O_RDONLY, false); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerFilePermission, ReadOnlyRecursive) { + const char kPath[] = "/tmp/good/"; + const char kPathFile[] = "/tmp/good/file"; + BrokerFilePermission perm = BrokerFilePermission::ReadOnlyRecursive(kPath); + CheckPerm(perm, kPathFile, O_RDONLY, false); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerFilePermission, WriteOnly) { + const char kPath[] = "/tmp/good"; + BrokerFilePermission perm = BrokerFilePermission::WriteOnly(kPath); + CheckPerm(perm, kPath, O_WRONLY, false); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerFilePermission, ReadWrite) { + const char kPath[] = "/tmp/good"; + BrokerFilePermission perm = BrokerFilePermission::ReadWrite(kPath); + CheckPerm(perm, kPath, O_RDWR, false); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerFilePermission, ReadWriteCreate) { + const char kPath[] = "/tmp/good"; + BrokerFilePermission perm = BrokerFilePermission::ReadWriteCreate(kPath); + CheckPerm(perm, kPath, O_RDWR, true); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +void CheckUnlink(BrokerFilePermission& perm, + const char* path, + int access_flags) { + bool unlink; + ASSERT_FALSE(perm.CheckOpen(path, access_flags, NULL, &unlink)); + ASSERT_FALSE(perm.CheckOpen(path, access_flags | O_CREAT, NULL, &unlink)); + ASSERT_TRUE( + perm.CheckOpen(path, access_flags | O_CREAT | O_EXCL, NULL, &unlink)); + ASSERT_TRUE(unlink); +} + +TEST(BrokerFilePermission, ReadWriteCreateUnlink) { + const char kPath[] = "/tmp/good"; + BrokerFilePermission perm = + BrokerFilePermission::ReadWriteCreateUnlink(kPath); + CheckUnlink(perm, kPath, O_RDWR); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerFilePermission, ReadWriteCreateUnlinkRecursive) { + const char kPath[] = "/tmp/good/"; + const char kPathFile[] = "/tmp/good/file"; + BrokerFilePermission perm = + BrokerFilePermission::ReadWriteCreateUnlinkRecursive(kPath); + CheckUnlink(perm, kPathFile, O_RDWR); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerFilePermission, ValidatePath) { + EXPECT_TRUE(BrokerFilePermissionTester::ValidatePath("/path")); + EXPECT_TRUE(BrokerFilePermissionTester::ValidatePath("/")); + EXPECT_TRUE(BrokerFilePermissionTester::ValidatePath("/..path")); + + EXPECT_FALSE(BrokerFilePermissionTester::ValidatePath("")); + EXPECT_FALSE(BrokerFilePermissionTester::ValidatePath("bad")); + EXPECT_FALSE(BrokerFilePermissionTester::ValidatePath("/bad/")); + EXPECT_FALSE(BrokerFilePermissionTester::ValidatePath("bad/")); + EXPECT_FALSE(BrokerFilePermissionTester::ValidatePath("/bad/..")); + EXPECT_FALSE(BrokerFilePermissionTester::ValidatePath("/bad/../bad")); + EXPECT_FALSE(BrokerFilePermissionTester::ValidatePath("/../bad")); +} + +} // namespace + +} // namespace syscall_broker + +} // namespace sandbox diff --git a/sandbox/linux/syscall_broker/broker_host.cc b/sandbox/linux/syscall_broker/broker_host.cc new file mode 100644 index 0000000000..e5957ed224 --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_host.cc @@ -0,0 +1,231 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/syscall_broker/broker_host.h" + +#include <fcntl.h> +#include <sys/socket.h> +#include <sys/stat.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/pickle.h" +#include "base/posix/eintr_wrapper.h" +#include "base/posix/unix_domain_socket_linux.h" +#include "base/third_party/valgrind/valgrind.h" +#include "sandbox/linux/syscall_broker/broker_common.h" +#include "sandbox/linux/syscall_broker/broker_policy.h" +#include "sandbox/linux/system_headers/linux_syscalls.h" + +namespace sandbox { + +namespace syscall_broker { + +namespace { + +bool IsRunningOnValgrind() { + return RUNNING_ON_VALGRIND; +} + +// A little open(2) wrapper to handle some oddities for us. In the general case +// make a direct system call since we want to keep in control of the broker +// process' system calls profile to be able to loosely sandbox it. +int sys_open(const char* pathname, int flags) { + // Hardcode mode to rw------- when creating files. + int mode; + if (flags & O_CREAT) { + mode = 0600; + } else { + mode = 0; + } + if (IsRunningOnValgrind()) { + // Valgrind does not support AT_FDCWD, just use libc's open() in this case. + return open(pathname, flags, mode); + } else { + return syscall(__NR_openat, AT_FDCWD, pathname, flags, mode); + } +} + +// Open |requested_filename| with |flags| if allowed by our policy. +// Write the syscall return value (-errno) to |write_pickle| and append +// a file descriptor to |opened_files| if relevant. +void OpenFileForIPC(const BrokerPolicy& policy, + const std::string& requested_filename, + int flags, + base::Pickle* write_pickle, + std::vector<int>* opened_files) { + DCHECK(write_pickle); + DCHECK(opened_files); + const char* file_to_open = NULL; + bool unlink_after_open = false; + const bool safe_to_open_file = policy.GetFileNameIfAllowedToOpen( + requested_filename.c_str(), flags, &file_to_open, &unlink_after_open); + + if (safe_to_open_file) { + CHECK(file_to_open); + int opened_fd = sys_open(file_to_open, flags); + if (opened_fd < 0) { + write_pickle->WriteInt(-errno); + } else { + // Success. + if (unlink_after_open) { + unlink(file_to_open); + } + opened_files->push_back(opened_fd); + write_pickle->WriteInt(0); + } + } else { + write_pickle->WriteInt(-policy.denied_errno()); + } +} + +// Perform access(2) on |requested_filename| with mode |mode| if allowed by our +// policy. Write the syscall return value (-errno) to |write_pickle|. +void AccessFileForIPC(const BrokerPolicy& policy, + const std::string& requested_filename, + int mode, + base::Pickle* write_pickle) { + DCHECK(write_pickle); + const char* file_to_access = NULL; + const bool safe_to_access_file = policy.GetFileNameIfAllowedToAccess( + requested_filename.c_str(), mode, &file_to_access); + + if (safe_to_access_file) { + CHECK(file_to_access); + int access_ret = access(file_to_access, mode); + int access_errno = errno; + if (!access_ret) + write_pickle->WriteInt(0); + else + write_pickle->WriteInt(-access_errno); + } else { + write_pickle->WriteInt(-policy.denied_errno()); + } +} + +// Handle a |command_type| request contained in |iter| and send the reply +// on |reply_ipc|. +// Currently COMMAND_OPEN and COMMAND_ACCESS are supported. +bool HandleRemoteCommand(const BrokerPolicy& policy, + IPCCommand command_type, + int reply_ipc, + base::PickleIterator iter) { + // Currently all commands have two arguments: filename and flags. + std::string requested_filename; + int flags = 0; + if (!iter.ReadString(&requested_filename) || !iter.ReadInt(&flags)) + return false; + + base::Pickle write_pickle; + std::vector<int> opened_files; + + switch (command_type) { + case COMMAND_ACCESS: + AccessFileForIPC(policy, requested_filename, flags, &write_pickle); + break; + case COMMAND_OPEN: + OpenFileForIPC( + policy, requested_filename, flags, &write_pickle, &opened_files); + break; + default: + LOG(ERROR) << "Invalid IPC command"; + break; + } + + CHECK_LE(write_pickle.size(), kMaxMessageLength); + ssize_t sent = base::UnixDomainSocket::SendMsg( + reply_ipc, write_pickle.data(), write_pickle.size(), opened_files); + + // Close anything we have opened in this process. + for (std::vector<int>::iterator it = opened_files.begin(); + it != opened_files.end(); + ++it) { + int ret = IGNORE_EINTR(close(*it)); + DCHECK(!ret) << "Could not close file descriptor"; + } + + if (sent <= 0) { + LOG(ERROR) << "Could not send IPC reply"; + return false; + } + return true; +} + +} // namespace + +BrokerHost::BrokerHost(const BrokerPolicy& broker_policy, + BrokerChannel::EndPoint ipc_channel) + : broker_policy_(broker_policy), ipc_channel_(ipc_channel.Pass()) { +} + +BrokerHost::~BrokerHost() { +} + +// Handle a request on the IPC channel ipc_channel_. +// A request should have a file descriptor attached on which we will reply and +// that we will then close. +// A request should start with an int that will be used as the command type. +BrokerHost::RequestStatus BrokerHost::HandleRequest() const { + ScopedVector<base::ScopedFD> fds; + char buf[kMaxMessageLength]; + errno = 0; + const ssize_t msg_len = base::UnixDomainSocket::RecvMsg( + ipc_channel_.get(), buf, sizeof(buf), &fds); + + if (msg_len == 0 || (msg_len == -1 && errno == ECONNRESET)) { + // EOF from the client, or the client died, we should die. + return RequestStatus::LOST_CLIENT; + } + + // The client should send exactly one file descriptor, on which we + // will write the reply. + // TODO(mdempsky): ScopedVector doesn't have 'at()', only 'operator[]'. + if (msg_len < 0 || fds.size() != 1 || fds[0]->get() < 0) { + PLOG(ERROR) << "Error reading message from the client"; + return RequestStatus::FAILURE; + } + + base::ScopedFD temporary_ipc(fds[0]->Pass()); + + base::Pickle pickle(buf, msg_len); + base::PickleIterator iter(pickle); + int command_type; + if (iter.ReadInt(&command_type)) { + bool command_handled = false; + // Go through all the possible IPC messages. + switch (command_type) { + case COMMAND_ACCESS: + case COMMAND_OPEN: + // We reply on the file descriptor sent to us via the IPC channel. + command_handled = HandleRemoteCommand( + broker_policy_, static_cast<IPCCommand>(command_type), + temporary_ipc.get(), iter); + break; + default: + NOTREACHED(); + break; + } + + if (command_handled) { + return RequestStatus::SUCCESS; + } else { + return RequestStatus::FAILURE; + } + + NOTREACHED(); + } + + LOG(ERROR) << "Error parsing IPC request"; + return RequestStatus::FAILURE; +} + +} // namespace syscall_broker + +} // namespace sandbox diff --git a/sandbox/linux/syscall_broker/broker_host.h b/sandbox/linux/syscall_broker/broker_host.h new file mode 100644 index 0000000000..9866507d1c --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_host.h @@ -0,0 +1,41 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSCALL_BROKER_BROKER_HOST_H_ +#define SANDBOX_LINUX_SYSCALL_BROKER_BROKER_HOST_H_ + +#include "base/macros.h" +#include "sandbox/linux/syscall_broker/broker_channel.h" + +namespace sandbox { + +namespace syscall_broker { + +class BrokerPolicy; + +// The BrokerHost class should be embedded in a (presumably not sandboxed) +// process. It will honor IPC requests from a BrokerClient sent over +// |ipc_channel| according to |broker_policy|. +class BrokerHost { + public: + enum class RequestStatus { LOST_CLIENT = 0, SUCCESS, FAILURE }; + + BrokerHost(const BrokerPolicy& broker_policy, + BrokerChannel::EndPoint ipc_channel); + ~BrokerHost(); + + RequestStatus HandleRequest() const; + + private: + const BrokerPolicy& broker_policy_; + const BrokerChannel::EndPoint ipc_channel_; + + DISALLOW_COPY_AND_ASSIGN(BrokerHost); +}; + +} // namespace syscall_broker + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SYSCALL_BROKER_BROKER_HOST_H_ diff --git a/sandbox/linux/syscall_broker/broker_policy.cc b/sandbox/linux/syscall_broker/broker_policy.cc new file mode 100644 index 0000000000..d9f69e3b81 --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_policy.cc @@ -0,0 +1,99 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/syscall_broker/broker_policy.h" + +#include <fcntl.h> +#include <stdint.h> +#include <string.h> + +#include <string> +#include <vector> + +#include "base/logging.h" +#include "sandbox/linux/syscall_broker/broker_common.h" + +namespace sandbox { +namespace syscall_broker { + +BrokerPolicy::BrokerPolicy(int denied_errno, + const std::vector<BrokerFilePermission>& permissions) + : denied_errno_(denied_errno), + permissions_(permissions), + num_of_permissions_(permissions.size()) { + // The spec guarantees vectors store their elements contiguously + // so set up a pointer to array of element so it can be used + // in async signal safe code instead of vector operations. + if (num_of_permissions_ > 0) { + permissions_array_ = &permissions_[0]; + } else { + permissions_array_ = NULL; + } +} + +BrokerPolicy::~BrokerPolicy() { +} + +// Check if calling access() should be allowed on |requested_filename| with +// mode |requested_mode|. +// Note: access() being a system call to check permissions, this can get a bit +// confusing. We're checking if calling access() should even be allowed with +// the same policy we would use for open(). +// If |file_to_access| is not NULL, we will return the matching pointer from +// the whitelist. For paranoia a caller should then use |file_to_access|. See +// GetFileNameIfAllowedToOpen() for more explanation. +// return true if calling access() on this file should be allowed, false +// otherwise. +// Async signal safe if and only if |file_to_access| is NULL. +bool BrokerPolicy::GetFileNameIfAllowedToAccess( + const char* requested_filename, + int requested_mode, + const char** file_to_access) const { + if (file_to_access && *file_to_access) { + // Make sure that callers never pass a non-empty string. In case callers + // wrongly forget to check the return value and look at the string + // instead, this could catch bugs. + RAW_LOG(FATAL, "*file_to_access should be NULL"); + return false; + } + for (size_t i = 0; i < num_of_permissions_; i++) { + if (permissions_array_[i].CheckAccess(requested_filename, requested_mode, + file_to_access)) { + return true; + } + } + return false; +} + +// Check if |requested_filename| can be opened with flags |requested_flags|. +// If |file_to_open| is not NULL, we will return the matching pointer from the +// whitelist. For paranoia, a caller should then use |file_to_open| rather +// than |requested_filename|, so that it never attempts to open an +// attacker-controlled file name, even if an attacker managed to fool the +// string comparison mechanism. +// Return true if opening should be allowed, false otherwise. +// Async signal safe if and only if |file_to_open| is NULL. +bool BrokerPolicy::GetFileNameIfAllowedToOpen(const char* requested_filename, + int requested_flags, + const char** file_to_open, + bool* unlink_after_open) const { + if (file_to_open && *file_to_open) { + // Make sure that callers never pass a non-empty string. In case callers + // wrongly forget to check the return value and look at the string + // instead, this could catch bugs. + RAW_LOG(FATAL, "*file_to_open should be NULL"); + return false; + } + for (size_t i = 0; i < num_of_permissions_; i++) { + if (permissions_array_[i].CheckOpen(requested_filename, requested_flags, + file_to_open, unlink_after_open)) { + return true; + } + } + return false; +} + +} // namespace syscall_broker + +} // namespace sandbox diff --git a/sandbox/linux/syscall_broker/broker_policy.h b/sandbox/linux/syscall_broker/broker_policy.h new file mode 100644 index 0000000000..d5146edc06 --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_policy.h @@ -0,0 +1,87 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSCALL_BROKER_BROKER_POLICY_H_ +#define SANDBOX_LINUX_SYSCALL_BROKER_BROKER_POLICY_H_ + +#include <string> +#include <vector> + +#include "base/macros.h" + +#include "sandbox/linux/syscall_broker/broker_file_permission.h" + +namespace sandbox { +namespace syscall_broker { + +// BrokerPolicy allows to define the security policy enforced by a +// BrokerHost. The BrokerHost will evaluate requests sent over its +// IPC channel according to the BrokerPolicy. +// Some of the methods of this class can be used in an async-signal safe +// way. +class BrokerPolicy { + public: + // |denied_errno| is the error code returned when IPC requests for system + // calls such as open() or access() are denied because a file is not in the + // whitelist. EACCESS would be a typical value. + // |permissions| is a list of BrokerPermission objects that define + // what the broker will allow. + BrokerPolicy(int denied_errno, + const std::vector<BrokerFilePermission>& permissions); + + ~BrokerPolicy(); + + // Check if calling access() should be allowed on |requested_filename| with + // mode |requested_mode|. + // Note: access() being a system call to check permissions, this can get a bit + // confusing. We're checking if calling access() should even be allowed with + // If |file_to_open| is not NULL, a pointer to the path will be returned. + // In the case of a recursive match, this will be the requested_filename, + // otherwise it will return the matching pointer from the + // whitelist. For paranoia a caller should then use |file_to_access|. See + // GetFileNameIfAllowedToOpen() for more explanation. + // return true if calling access() on this file should be allowed, false + // otherwise. + // Async signal safe if and only if |file_to_access| is NULL. + bool GetFileNameIfAllowedToAccess(const char* requested_filename, + int requested_mode, + const char** file_to_access) const; + + // Check if |requested_filename| can be opened with flags |requested_flags|. + // If |file_to_open| is not NULL, a pointer to the path will be returned. + // In the case of a recursive match, this will be the requested_filename, + // otherwise it will return the matching pointer from the + // whitelist. For paranoia, a caller should then use |file_to_open| rather + // than |requested_filename|, so that it never attempts to open an + // attacker-controlled file name, even if an attacker managed to fool the + // string comparison mechanism. + // |unlink_after_open| if not NULL will be set to point to true if the + // policy requests the caller unlink the path after opening. + // Return true if opening should be allowed, false otherwise. + // Async signal safe if and only if |file_to_open| is NULL. + bool GetFileNameIfAllowedToOpen(const char* requested_filename, + int requested_flags, + const char** file_to_open, + bool* unlink_after_open) const; + int denied_errno() const { return denied_errno_; } + + private: + const int denied_errno_; + // The permissions_ vector is used as storage for the BrokerFilePermission + // objects but is not referenced outside of the constructor as + // vectors are unfriendly in async signal safe code. + const std::vector<BrokerFilePermission> permissions_; + // permissions_array_ is set up to point to the backing store of + // permissions_ and is used in async signal safe methods. + const BrokerFilePermission* permissions_array_; + const size_t num_of_permissions_; + + DISALLOW_COPY_AND_ASSIGN(BrokerPolicy); +}; + +} // namespace syscall_broker + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SYSCALL_BROKER_BROKER_POLICY_H_ diff --git a/sandbox/linux/syscall_broker/broker_process.cc b/sandbox/linux/syscall_broker/broker_process.cc new file mode 100644 index 0000000000..81131cc4e0 --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_process.cc @@ -0,0 +1,120 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/syscall_broker/broker_process.h" + +#include <fcntl.h> +#include <signal.h> +#include <sys/stat.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include <algorithm> +#include <string> +#include <vector> + +#include "base/callback.h" +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "base/posix/eintr_wrapper.h" +#include "base/process/process_metrics.h" +#include "build/build_config.h" +#include "sandbox/linux/syscall_broker/broker_channel.h" +#include "sandbox/linux/syscall_broker/broker_client.h" +#include "sandbox/linux/syscall_broker/broker_host.h" + +namespace sandbox { + +namespace syscall_broker { + +BrokerProcess::BrokerProcess( + int denied_errno, + const std::vector<syscall_broker::BrokerFilePermission>& permissions, + bool fast_check_in_client, + bool quiet_failures_for_tests) + : initialized_(false), + fast_check_in_client_(fast_check_in_client), + quiet_failures_for_tests_(quiet_failures_for_tests), + broker_pid_(-1), + policy_(denied_errno, permissions) { +} + +BrokerProcess::~BrokerProcess() { + if (initialized_) { + if (broker_client_.get()) { + // Closing the socket should be enough to notify the child to die, + // unless it has been duplicated. + CloseChannel(); + } + PCHECK(0 == kill(broker_pid_, SIGKILL)); + siginfo_t process_info; + // Reap the child. + int ret = HANDLE_EINTR(waitid(P_PID, broker_pid_, &process_info, WEXITED)); + PCHECK(0 == ret); + } +} + +bool BrokerProcess::Init( + const base::Callback<bool(void)>& broker_process_init_callback) { + CHECK(!initialized_); + BrokerChannel::EndPoint ipc_reader; + BrokerChannel::EndPoint ipc_writer; + BrokerChannel::CreatePair(&ipc_reader, &ipc_writer); + +#if !defined(THREAD_SANITIZER) + DCHECK_EQ(1, base::GetNumberOfThreads(base::GetCurrentProcessHandle())); +#endif + int child_pid = fork(); + if (child_pid == -1) { + return false; + } + if (child_pid) { + // We are the parent and we have just forked our broker process. + ipc_reader.reset(); + broker_pid_ = child_pid; + broker_client_.reset(new BrokerClient(policy_, ipc_writer.Pass(), + fast_check_in_client_, + quiet_failures_for_tests_)); + initialized_ = true; + return true; + } else { + // We are the broker process. Make sure to close the writer's end so that + // we get notified if the client disappears. + ipc_writer.reset(); + CHECK(broker_process_init_callback.Run()); + BrokerHost broker_host(policy_, ipc_reader.Pass()); + for (;;) { + switch (broker_host.HandleRequest()) { + case BrokerHost::RequestStatus::LOST_CLIENT: + _exit(1); + case BrokerHost::RequestStatus::SUCCESS: + case BrokerHost::RequestStatus::FAILURE: + continue; + } + } + _exit(1); + } + NOTREACHED(); + return false; +} + +void BrokerProcess::CloseChannel() { + broker_client_.reset(); +} + +int BrokerProcess::Access(const char* pathname, int mode) const { + RAW_CHECK(initialized_); + return broker_client_->Access(pathname, mode); +} + +int BrokerProcess::Open(const char* pathname, int flags) const { + RAW_CHECK(initialized_); + return broker_client_->Open(pathname, flags); +} + +} // namespace syscall_broker + +} // namespace sandbox. diff --git a/sandbox/linux/syscall_broker/broker_process.h b/sandbox/linux/syscall_broker/broker_process.h new file mode 100644 index 0000000000..8a512a0c12 --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_process.h @@ -0,0 +1,94 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SERVICES_BROKER_PROCESS_H_ +#define SANDBOX_LINUX_SERVICES_BROKER_PROCESS_H_ + +#include <string> +#include <vector> + +#include "base/callback_forward.h" +#include "base/macros.h" +#include "base/memory/scoped_ptr.h" +#include "base/pickle.h" +#include "base/process/process.h" +#include "sandbox/linux/syscall_broker/broker_policy.h" +#include "sandbox/sandbox_export.h" + +namespace sandbox { + +namespace syscall_broker { + +class BrokerClient; +class BrokerFilePermission; + +// Create a new "broker" process to which we can send requests via an IPC +// channel by forking the current process. +// This is a low level IPC mechanism that is suitable to be called from a +// signal handler. +// A process would typically create a broker process before entering +// sandboxing. +// 1. BrokerProcess open_broker(read_whitelist, write_whitelist); +// 2. CHECK(open_broker.Init(NULL)); +// 3. Enable sandbox. +// 4. Use open_broker.Open() to open files. +class SANDBOX_EXPORT BrokerProcess { + public: + // |denied_errno| is the error code returned when methods such as Open() + // or Access() are invoked on a file which is not in the whitelist. EACCESS + // would be a typical value. + // |allowed_r_files| and |allowed_w_files| are white lists of files that can + // be opened later via the Open() API, respectively for reading and writing. + // A file available read-write should be listed in both. + // |fast_check_in_client| and |quiet_failures_for_tests| are reserved for + // unit tests, don't use it. + + BrokerProcess( + int denied_errno, + const std::vector<syscall_broker::BrokerFilePermission>& permissions, + bool fast_check_in_client = true, + bool quiet_failures_for_tests = false); + + ~BrokerProcess(); + // Will initialize the broker process. There should be no threads at this + // point, since we need to fork(). + // broker_process_init_callback will be called in the new broker process, + // after fork() returns. + bool Init(const base::Callback<bool(void)>& broker_process_init_callback); + + // Can be used in place of access(). Will be async signal safe. + // X_OK will always return an error in practice since the broker process + // doesn't support execute permissions. + // It's similar to the access() system call and will return -errno on errors. + int Access(const char* pathname, int mode) const; + // Can be used in place of open(). Will be async signal safe. + // The implementation only supports certain white listed flags and will + // return -EPERM on other flags. + // It's similar to the open() system call and will return -errno on errors. + int Open(const char* pathname, int flags) const; + + int broker_pid() const { return broker_pid_; } + + private: + friend class BrokerProcessTestHelper; + + // Close the IPC channel with the other party. This should only be used + // by tests an none of the class methods should be used afterwards. + void CloseChannel(); + + bool initialized_; // Whether we've been through Init() yet. + const bool fast_check_in_client_; + const bool quiet_failures_for_tests_; + pid_t broker_pid_; // The PID of the broker (child). + syscall_broker::BrokerPolicy policy_; // The sandboxing policy. + scoped_ptr<syscall_broker::BrokerClient> broker_client_; + + DISALLOW_COPY_AND_ASSIGN(BrokerProcess); +}; + +} // namespace syscall_broker + +} // namespace sandbox + +#endif // SANDBOX_LINUX_SERVICES_BROKER_PROCESS_H_ diff --git a/sandbox/linux/syscall_broker/broker_process_unittest.cc b/sandbox/linux/syscall_broker/broker_process_unittest.cc new file mode 100644 index 0000000000..9ad0e719de --- /dev/null +++ b/sandbox/linux/syscall_broker/broker_process_unittest.cc @@ -0,0 +1,656 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/syscall_broker/broker_process.h" + +#include <errno.h> +#include <fcntl.h> +#include <poll.h> +#include <sys/resource.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include <algorithm> +#include <string> +#include <vector> + +#include "base/bind.h" +#include "base/files/file_util.h" +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/memory/scoped_ptr.h" +#include "base/posix/eintr_wrapper.h" +#include "base/posix/unix_domain_socket_linux.h" +#include "sandbox/linux/syscall_broker/broker_client.h" +#include "sandbox/linux/tests/scoped_temporary_file.h" +#include "sandbox/linux/tests/test_utils.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace syscall_broker { + +class BrokerProcessTestHelper { + public: + static void CloseChannel(BrokerProcess* broker) { broker->CloseChannel(); } + // Get the client's IPC descriptor to send IPC requests directly. + // TODO(jln): refator tests to get rid of this. + static int GetIPCDescriptor(const BrokerProcess* broker) { + return broker->broker_client_->GetIPCDescriptor(); + } +}; + +namespace { + +bool NoOpCallback() { + return true; +} + +} // namespace + +TEST(BrokerProcess, CreateAndDestroy) { + std::vector<BrokerFilePermission> permissions; + permissions.push_back(BrokerFilePermission::ReadOnly("/proc/cpuinfo")); + + scoped_ptr<BrokerProcess> open_broker(new BrokerProcess(EPERM, permissions)); + ASSERT_TRUE(open_broker->Init(base::Bind(&NoOpCallback))); + + ASSERT_TRUE(TestUtils::CurrentProcessHasChildren()); + // Destroy the broker and check it has exited properly. + open_broker.reset(); + ASSERT_FALSE(TestUtils::CurrentProcessHasChildren()); +} + +TEST(BrokerProcess, TestOpenAccessNull) { + std::vector<BrokerFilePermission> empty; + BrokerProcess open_broker(EPERM, empty); + ASSERT_TRUE(open_broker.Init(base::Bind(&NoOpCallback))); + + int fd = open_broker.Open(NULL, O_RDONLY); + ASSERT_EQ(fd, -EFAULT); + + int ret = open_broker.Access(NULL, F_OK); + ASSERT_EQ(ret, -EFAULT); +} + +void TestOpenFilePerms(bool fast_check_in_client, int denied_errno) { + const char kR_WhiteListed[] = "/proc/DOESNOTEXIST1"; + // We can't debug the init process, and shouldn't be able to access + // its auxv file. + const char kR_WhiteListedButDenied[] = "/proc/1/auxv"; + const char kW_WhiteListed[] = "/proc/DOESNOTEXIST2"; + const char kRW_WhiteListed[] = "/proc/DOESNOTEXIST3"; + const char k_NotWhitelisted[] = "/proc/DOESNOTEXIST4"; + + std::vector<BrokerFilePermission> permissions; + permissions.push_back(BrokerFilePermission::ReadOnly(kR_WhiteListed)); + permissions.push_back( + BrokerFilePermission::ReadOnly(kR_WhiteListedButDenied)); + permissions.push_back(BrokerFilePermission::WriteOnly(kW_WhiteListed)); + permissions.push_back(BrokerFilePermission::ReadWrite(kRW_WhiteListed)); + + BrokerProcess open_broker(denied_errno, permissions, fast_check_in_client); + ASSERT_TRUE(open_broker.Init(base::Bind(&NoOpCallback))); + + int fd = -1; + fd = open_broker.Open(kR_WhiteListed, O_RDONLY); + ASSERT_EQ(fd, -ENOENT); + fd = open_broker.Open(kR_WhiteListed, O_WRONLY); + ASSERT_EQ(fd, -denied_errno); + fd = open_broker.Open(kR_WhiteListed, O_RDWR); + ASSERT_EQ(fd, -denied_errno); + int ret = -1; + ret = open_broker.Access(kR_WhiteListed, F_OK); + ASSERT_EQ(ret, -ENOENT); + ret = open_broker.Access(kR_WhiteListed, R_OK); + ASSERT_EQ(ret, -ENOENT); + ret = open_broker.Access(kR_WhiteListed, W_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kR_WhiteListed, R_OK | W_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kR_WhiteListed, X_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kR_WhiteListed, R_OK | X_OK); + ASSERT_EQ(ret, -denied_errno); + + // Android sometimes runs tests as root. + // This part of the test requires a process that doesn't have + // CAP_DAC_OVERRIDE. We check against a root euid as a proxy for that. + if (geteuid()) { + fd = open_broker.Open(kR_WhiteListedButDenied, O_RDONLY); + // The broker process will allow this, but the normal permission system + // won't. + ASSERT_EQ(fd, -EACCES); + fd = open_broker.Open(kR_WhiteListedButDenied, O_WRONLY); + ASSERT_EQ(fd, -denied_errno); + fd = open_broker.Open(kR_WhiteListedButDenied, O_RDWR); + ASSERT_EQ(fd, -denied_errno); + ret = open_broker.Access(kR_WhiteListedButDenied, F_OK); + // The normal permission system will let us check that the file exists. + ASSERT_EQ(ret, 0); + ret = open_broker.Access(kR_WhiteListedButDenied, R_OK); + ASSERT_EQ(ret, -EACCES); + ret = open_broker.Access(kR_WhiteListedButDenied, W_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kR_WhiteListedButDenied, R_OK | W_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kR_WhiteListedButDenied, X_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kR_WhiteListedButDenied, R_OK | X_OK); + ASSERT_EQ(ret, -denied_errno); + } + + fd = open_broker.Open(kW_WhiteListed, O_RDONLY); + ASSERT_EQ(fd, -denied_errno); + fd = open_broker.Open(kW_WhiteListed, O_WRONLY); + ASSERT_EQ(fd, -ENOENT); + fd = open_broker.Open(kW_WhiteListed, O_RDWR); + ASSERT_EQ(fd, -denied_errno); + ret = open_broker.Access(kW_WhiteListed, F_OK); + ASSERT_EQ(ret, -ENOENT); + ret = open_broker.Access(kW_WhiteListed, R_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kW_WhiteListed, W_OK); + ASSERT_EQ(ret, -ENOENT); + ret = open_broker.Access(kW_WhiteListed, R_OK | W_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kW_WhiteListed, X_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kW_WhiteListed, R_OK | X_OK); + ASSERT_EQ(ret, -denied_errno); + + fd = open_broker.Open(kRW_WhiteListed, O_RDONLY); + ASSERT_EQ(fd, -ENOENT); + fd = open_broker.Open(kRW_WhiteListed, O_WRONLY); + ASSERT_EQ(fd, -ENOENT); + fd = open_broker.Open(kRW_WhiteListed, O_RDWR); + ASSERT_EQ(fd, -ENOENT); + ret = open_broker.Access(kRW_WhiteListed, F_OK); + ASSERT_EQ(ret, -ENOENT); + ret = open_broker.Access(kRW_WhiteListed, R_OK); + ASSERT_EQ(ret, -ENOENT); + ret = open_broker.Access(kRW_WhiteListed, W_OK); + ASSERT_EQ(ret, -ENOENT); + ret = open_broker.Access(kRW_WhiteListed, R_OK | W_OK); + ASSERT_EQ(ret, -ENOENT); + ret = open_broker.Access(kRW_WhiteListed, X_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(kRW_WhiteListed, R_OK | X_OK); + ASSERT_EQ(ret, -denied_errno); + + fd = open_broker.Open(k_NotWhitelisted, O_RDONLY); + ASSERT_EQ(fd, -denied_errno); + fd = open_broker.Open(k_NotWhitelisted, O_WRONLY); + ASSERT_EQ(fd, -denied_errno); + fd = open_broker.Open(k_NotWhitelisted, O_RDWR); + ASSERT_EQ(fd, -denied_errno); + ret = open_broker.Access(k_NotWhitelisted, F_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(k_NotWhitelisted, R_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(k_NotWhitelisted, W_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(k_NotWhitelisted, R_OK | W_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(k_NotWhitelisted, X_OK); + ASSERT_EQ(ret, -denied_errno); + ret = open_broker.Access(k_NotWhitelisted, R_OK | X_OK); + ASSERT_EQ(ret, -denied_errno); + + // We have some extra sanity check for clearly wrong values. + fd = open_broker.Open(kRW_WhiteListed, O_RDONLY | O_WRONLY | O_RDWR); + ASSERT_EQ(fd, -denied_errno); + + // It makes no sense to allow O_CREAT in a 2-parameters open. Ensure this + // is denied. + fd = open_broker.Open(kRW_WhiteListed, O_RDWR | O_CREAT); + ASSERT_EQ(fd, -denied_errno); +} + +// Run the same thing twice. The second time, we make sure that no security +// check is performed on the client. +TEST(BrokerProcess, OpenFilePermsWithClientCheck) { + TestOpenFilePerms(true /* fast_check_in_client */, EPERM); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerProcess, OpenOpenFilePermsNoClientCheck) { + TestOpenFilePerms(false /* fast_check_in_client */, EPERM); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +// Run the same twice again, but with ENOENT instead of EPERM. +TEST(BrokerProcess, OpenFilePermsWithClientCheckNoEnt) { + TestOpenFilePerms(true /* fast_check_in_client */, ENOENT); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerProcess, OpenOpenFilePermsNoClientCheckNoEnt) { + TestOpenFilePerms(false /* fast_check_in_client */, ENOENT); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +void TestBadPaths(bool fast_check_in_client) { + const char kFileCpuInfo[] = "/proc/cpuinfo"; + const char kNotAbsPath[] = "proc/cpuinfo"; + const char kDotDotStart[] = "/../proc/cpuinfo"; + const char kDotDotMiddle[] = "/proc/self/../cpuinfo"; + const char kDotDotEnd[] = "/proc/.."; + const char kTrailingSlash[] = "/proc/"; + + std::vector<BrokerFilePermission> permissions; + + permissions.push_back(BrokerFilePermission::ReadOnlyRecursive("/proc/")); + scoped_ptr<BrokerProcess> open_broker( + new BrokerProcess(EPERM, permissions, fast_check_in_client)); + ASSERT_TRUE(open_broker->Init(base::Bind(&NoOpCallback))); + // Open cpuinfo via the broker. + int cpuinfo_fd = open_broker->Open(kFileCpuInfo, O_RDONLY); + base::ScopedFD cpuinfo_fd_closer(cpuinfo_fd); + ASSERT_GE(cpuinfo_fd, 0); + + int fd = -1; + int can_access; + + can_access = open_broker->Access(kNotAbsPath, R_OK); + ASSERT_EQ(can_access, -EPERM); + fd = open_broker->Open(kNotAbsPath, O_RDONLY); + ASSERT_EQ(fd, -EPERM); + + can_access = open_broker->Access(kDotDotStart, R_OK); + ASSERT_EQ(can_access, -EPERM); + fd = open_broker->Open(kDotDotStart, O_RDONLY); + ASSERT_EQ(fd, -EPERM); + + can_access = open_broker->Access(kDotDotMiddle, R_OK); + ASSERT_EQ(can_access, -EPERM); + fd = open_broker->Open(kDotDotMiddle, O_RDONLY); + ASSERT_EQ(fd, -EPERM); + + can_access = open_broker->Access(kDotDotEnd, R_OK); + ASSERT_EQ(can_access, -EPERM); + fd = open_broker->Open(kDotDotEnd, O_RDONLY); + ASSERT_EQ(fd, -EPERM); + + can_access = open_broker->Access(kTrailingSlash, R_OK); + ASSERT_EQ(can_access, -EPERM); + fd = open_broker->Open(kTrailingSlash, O_RDONLY); + ASSERT_EQ(fd, -EPERM); +} + +TEST(BrokerProcess, BadPathsClientCheck) { + TestBadPaths(true /* fast_check_in_client */); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerProcess, BadPathsNoClientCheck) { + TestBadPaths(false /* fast_check_in_client */); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +void TestOpenCpuinfo(bool fast_check_in_client, bool recursive) { + const char kFileCpuInfo[] = "/proc/cpuinfo"; + const char kDirProc[] = "/proc/"; + + std::vector<BrokerFilePermission> permissions; + if (recursive) + permissions.push_back(BrokerFilePermission::ReadOnlyRecursive(kDirProc)); + else + permissions.push_back(BrokerFilePermission::ReadOnly(kFileCpuInfo)); + + scoped_ptr<BrokerProcess> open_broker( + new BrokerProcess(EPERM, permissions, fast_check_in_client)); + ASSERT_TRUE(open_broker->Init(base::Bind(&NoOpCallback))); + + int fd = -1; + fd = open_broker->Open(kFileCpuInfo, O_RDWR); + base::ScopedFD fd_closer(fd); + ASSERT_EQ(fd, -EPERM); + + // Check we can read /proc/cpuinfo. + int can_access = open_broker->Access(kFileCpuInfo, R_OK); + ASSERT_EQ(can_access, 0); + can_access = open_broker->Access(kFileCpuInfo, W_OK); + ASSERT_EQ(can_access, -EPERM); + // Check we can not write /proc/cpuinfo. + + // Open cpuinfo via the broker. + int cpuinfo_fd = open_broker->Open(kFileCpuInfo, O_RDONLY); + base::ScopedFD cpuinfo_fd_closer(cpuinfo_fd); + ASSERT_GE(cpuinfo_fd, 0); + char buf[3]; + memset(buf, 0, sizeof(buf)); + int read_len1 = read(cpuinfo_fd, buf, sizeof(buf)); + ASSERT_GT(read_len1, 0); + + // Open cpuinfo directly. + int cpuinfo_fd2 = open(kFileCpuInfo, O_RDONLY); + base::ScopedFD cpuinfo_fd2_closer(cpuinfo_fd2); + ASSERT_GE(cpuinfo_fd2, 0); + char buf2[3]; + memset(buf2, 1, sizeof(buf2)); + int read_len2 = read(cpuinfo_fd2, buf2, sizeof(buf2)); + ASSERT_GT(read_len1, 0); + + // The following is not guaranteed true, but will be in practice. + ASSERT_EQ(read_len1, read_len2); + // Compare the cpuinfo as returned by the broker with the one we opened + // ourselves. + ASSERT_EQ(memcmp(buf, buf2, read_len1), 0); + + ASSERT_TRUE(TestUtils::CurrentProcessHasChildren()); + open_broker.reset(); + ASSERT_FALSE(TestUtils::CurrentProcessHasChildren()); +} + +// Run this test 4 times. With and without the check in client +// and using a recursive path. +TEST(BrokerProcess, OpenCpuinfoWithClientCheck) { + TestOpenCpuinfo(true /* fast_check_in_client */, false /* not recursive */); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerProcess, OpenCpuinfoNoClientCheck) { + TestOpenCpuinfo(false /* fast_check_in_client */, false /* not recursive */); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerProcess, OpenCpuinfoWithClientCheckRecursive) { + TestOpenCpuinfo(true /* fast_check_in_client */, true /* recursive */); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerProcess, OpenCpuinfoNoClientCheckRecursive) { + TestOpenCpuinfo(false /* fast_check_in_client */, true /* recursive */); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerProcess, OpenFileRW) { + ScopedTemporaryFile tempfile; + const char* tempfile_name = tempfile.full_file_name(); + + std::vector<BrokerFilePermission> permissions; + permissions.push_back(BrokerFilePermission::ReadWrite(tempfile_name)); + + BrokerProcess open_broker(EPERM, permissions); + ASSERT_TRUE(open_broker.Init(base::Bind(&NoOpCallback))); + + // Check we can access that file with read or write. + int can_access = open_broker.Access(tempfile_name, R_OK | W_OK); + ASSERT_EQ(can_access, 0); + + int tempfile2 = -1; + tempfile2 = open_broker.Open(tempfile_name, O_RDWR); + ASSERT_GE(tempfile2, 0); + + // Write to the descriptor opened by the broker. + char test_text[] = "TESTTESTTEST"; + ssize_t len = write(tempfile2, test_text, sizeof(test_text)); + ASSERT_EQ(len, static_cast<ssize_t>(sizeof(test_text))); + + // Read back from the original file descriptor what we wrote through + // the descriptor provided by the broker. + char buf[1024]; + len = read(tempfile.fd(), buf, sizeof(buf)); + + ASSERT_EQ(len, static_cast<ssize_t>(sizeof(test_text))); + ASSERT_EQ(memcmp(test_text, buf, sizeof(test_text)), 0); + + ASSERT_EQ(close(tempfile2), 0); +} + +// SANDBOX_TEST because the process could die with a SIGPIPE +// and we want this to happen in a subprocess. +SANDBOX_TEST(BrokerProcess, BrokerDied) { + const char kCpuInfo[] = "/proc/cpuinfo"; + std::vector<BrokerFilePermission> permissions; + permissions.push_back(BrokerFilePermission::ReadOnly(kCpuInfo)); + + BrokerProcess open_broker(EPERM, permissions, true /* fast_check_in_client */, + true /* quiet_failures_for_tests */); + SANDBOX_ASSERT(open_broker.Init(base::Bind(&NoOpCallback))); + const pid_t broker_pid = open_broker.broker_pid(); + SANDBOX_ASSERT(kill(broker_pid, SIGKILL) == 0); + + // Now we check that the broker has been signaled, but do not reap it. + siginfo_t process_info; + SANDBOX_ASSERT(HANDLE_EINTR(waitid( + P_PID, broker_pid, &process_info, WEXITED | WNOWAIT)) == + 0); + SANDBOX_ASSERT(broker_pid == process_info.si_pid); + SANDBOX_ASSERT(CLD_KILLED == process_info.si_code); + SANDBOX_ASSERT(SIGKILL == process_info.si_status); + + // Check that doing Open with a dead broker won't SIGPIPE us. + SANDBOX_ASSERT(open_broker.Open(kCpuInfo, O_RDONLY) == -ENOMEM); + SANDBOX_ASSERT(open_broker.Access(kCpuInfo, O_RDONLY) == -ENOMEM); +} + +void TestOpenComplexFlags(bool fast_check_in_client) { + const char kCpuInfo[] = "/proc/cpuinfo"; + std::vector<BrokerFilePermission> permissions; + permissions.push_back(BrokerFilePermission::ReadOnly(kCpuInfo)); + + BrokerProcess open_broker(EPERM, permissions, fast_check_in_client); + ASSERT_TRUE(open_broker.Init(base::Bind(&NoOpCallback))); + // Test that we do the right thing for O_CLOEXEC and O_NONBLOCK. + int fd = -1; + int ret = 0; + fd = open_broker.Open(kCpuInfo, O_RDONLY); + ASSERT_GE(fd, 0); + ret = fcntl(fd, F_GETFL); + ASSERT_NE(-1, ret); + // The descriptor shouldn't have the O_CLOEXEC attribute, nor O_NONBLOCK. + ASSERT_EQ(0, ret & (O_CLOEXEC | O_NONBLOCK)); + ASSERT_EQ(0, close(fd)); + + fd = open_broker.Open(kCpuInfo, O_RDONLY | O_CLOEXEC); + ASSERT_GE(fd, 0); + ret = fcntl(fd, F_GETFD); + ASSERT_NE(-1, ret); + // Important: use F_GETFD, not F_GETFL. The O_CLOEXEC flag in F_GETFL + // is actually not used by the kernel. + ASSERT_TRUE(FD_CLOEXEC & ret); + ASSERT_EQ(0, close(fd)); + + fd = open_broker.Open(kCpuInfo, O_RDONLY | O_NONBLOCK); + ASSERT_GE(fd, 0); + ret = fcntl(fd, F_GETFL); + ASSERT_NE(-1, ret); + ASSERT_TRUE(O_NONBLOCK & ret); + ASSERT_EQ(0, close(fd)); +} + +TEST(BrokerProcess, OpenComplexFlagsWithClientCheck) { + TestOpenComplexFlags(true /* fast_check_in_client */); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +TEST(BrokerProcess, OpenComplexFlagsNoClientCheck) { + TestOpenComplexFlags(false /* fast_check_in_client */); + // Don't do anything here, so that ASSERT works in the subfunction as + // expected. +} + +// We need to allow noise because the broker will log when it receives our +// bogus IPCs. +SANDBOX_TEST_ALLOW_NOISE(BrokerProcess, RecvMsgDescriptorLeak) { + // Android creates a socket on first use of the LOG call. + // We need to ensure this socket is open before we + // begin the test. + LOG(INFO) << "Ensure Android LOG socket is allocated"; + + // Find the four lowest available file descriptors. + int available_fds[4]; + SANDBOX_ASSERT(0 == pipe(available_fds)); + SANDBOX_ASSERT(0 == pipe(available_fds + 2)); + + // Save one FD to send to the broker later, and close the others. + base::ScopedFD message_fd(available_fds[0]); + for (size_t i = 1; i < arraysize(available_fds); i++) { + SANDBOX_ASSERT(0 == IGNORE_EINTR(close(available_fds[i]))); + } + + // Lower our file descriptor limit to just allow three more file descriptors + // to be allocated. (N.B., RLIMIT_NOFILE doesn't limit the number of file + // descriptors a process can have: it only limits the highest value that can + // be assigned to newly-created descriptors allocated by the process.) + const rlim_t fd_limit = + 1 + + *std::max_element(available_fds, + available_fds + arraysize(available_fds)); + + // Valgrind doesn't allow changing the hard descriptor limit, so we only + // change the soft descriptor limit here. + struct rlimit rlim; + SANDBOX_ASSERT(0 == getrlimit(RLIMIT_NOFILE, &rlim)); + SANDBOX_ASSERT(fd_limit <= rlim.rlim_cur); + rlim.rlim_cur = fd_limit; + SANDBOX_ASSERT(0 == setrlimit(RLIMIT_NOFILE, &rlim)); + + static const char kCpuInfo[] = "/proc/cpuinfo"; + std::vector<BrokerFilePermission> permissions; + permissions.push_back(BrokerFilePermission::ReadOnly(kCpuInfo)); + + BrokerProcess open_broker(EPERM, permissions); + SANDBOX_ASSERT(open_broker.Init(base::Bind(&NoOpCallback))); + + const int ipc_fd = BrokerProcessTestHelper::GetIPCDescriptor(&open_broker); + SANDBOX_ASSERT(ipc_fd >= 0); + + static const char kBogus[] = "not a pickle"; + std::vector<int> fds; + fds.push_back(message_fd.get()); + + // The broker process should only have a couple spare file descriptors + // available, but for good measure we send it fd_limit bogus IPCs anyway. + for (rlim_t i = 0; i < fd_limit; ++i) { + SANDBOX_ASSERT( + base::UnixDomainSocket::SendMsg(ipc_fd, kBogus, sizeof(kBogus), fds)); + } + + const int fd = open_broker.Open(kCpuInfo, O_RDONLY); + SANDBOX_ASSERT(fd >= 0); + SANDBOX_ASSERT(0 == IGNORE_EINTR(close(fd))); +} + +bool CloseFD(int fd) { + PCHECK(0 == IGNORE_EINTR(close(fd))); + return true; +} + +// Return true if the other end of the |reader| pipe was closed, +// false if |timeout_in_seconds| was reached or another event +// or error occured. +bool WaitForClosedPipeWriter(int reader, int timeout_in_ms) { + struct pollfd poll_fd = {reader, POLLIN | POLLRDHUP, 0}; + const int num_events = HANDLE_EINTR(poll(&poll_fd, 1, timeout_in_ms)); + if (1 == num_events && poll_fd.revents | POLLHUP) + return true; + return false; +} + +// Closing the broker client's IPC channel should terminate the broker +// process. +TEST(BrokerProcess, BrokerDiesOnClosedChannel) { + std::vector<BrokerFilePermission> permissions; + permissions.push_back(BrokerFilePermission::ReadOnly("/proc/cpuinfo")); + + // Get the writing end of a pipe into the broker (child) process so + // that we can reliably detect when it dies. + int lifeline_fds[2]; + PCHECK(0 == pipe(lifeline_fds)); + + BrokerProcess open_broker(EPERM, permissions, true /* fast_check_in_client */, + false /* quiet_failures_for_tests */); + ASSERT_TRUE(open_broker.Init(base::Bind(&CloseFD, lifeline_fds[0]))); + // Make sure the writing end only exists in the broker process. + CloseFD(lifeline_fds[1]); + base::ScopedFD reader(lifeline_fds[0]); + + const pid_t broker_pid = open_broker.broker_pid(); + + // This should cause the broker process to exit. + BrokerProcessTestHelper::CloseChannel(&open_broker); + + const int kTimeoutInMilliseconds = 5000; + const bool broker_lifeline_closed = + WaitForClosedPipeWriter(reader.get(), kTimeoutInMilliseconds); + // If the broker exited, its lifeline fd should be closed. + ASSERT_TRUE(broker_lifeline_closed); + // Now check that the broker has exited, but do not reap it. + siginfo_t process_info; + ASSERT_EQ(0, HANDLE_EINTR(waitid(P_PID, broker_pid, &process_info, + WEXITED | WNOWAIT))); + EXPECT_EQ(broker_pid, process_info.si_pid); + EXPECT_EQ(CLD_EXITED, process_info.si_code); + EXPECT_EQ(1, process_info.si_status); +} + +TEST(BrokerProcess, CreateFile) { + std::string temp_str; + { + ScopedTemporaryFile tmp_file; + temp_str = tmp_file.full_file_name(); + } + const char* tempfile_name = temp_str.c_str(); + + std::vector<BrokerFilePermission> permissions; + permissions.push_back(BrokerFilePermission::ReadWriteCreate(tempfile_name)); + + BrokerProcess open_broker(EPERM, permissions); + ASSERT_TRUE(open_broker.Init(base::Bind(&NoOpCallback))); + + int fd = -1; + + // Try without O_EXCL + fd = open_broker.Open(tempfile_name, O_RDWR | O_CREAT); + ASSERT_EQ(fd, -EPERM); + + const char kTestText[] = "TESTTESTTEST"; + // Create a file + fd = open_broker.Open(tempfile_name, O_RDWR | O_CREAT | O_EXCL); + ASSERT_GE(fd, 0); + { + base::ScopedFD scoped_fd(fd); + + // Confirm fail if file exists + int bad_fd = open_broker.Open(tempfile_name, O_RDWR | O_CREAT | O_EXCL); + ASSERT_EQ(bad_fd, -EEXIST); + + // Write to the descriptor opened by the broker. + + ssize_t len = HANDLE_EINTR(write(fd, kTestText, sizeof(kTestText))); + ASSERT_EQ(len, static_cast<ssize_t>(sizeof(kTestText))); + } + + int fd_check = open(tempfile_name, O_RDONLY); + ASSERT_GE(fd_check, 0); + { + base::ScopedFD scoped_fd(fd_check); + char buf[1024]; + ssize_t len = HANDLE_EINTR(read(fd_check, buf, sizeof(buf))); + + ASSERT_EQ(len, static_cast<ssize_t>(sizeof(kTestText))); + ASSERT_EQ(memcmp(kTestText, buf, sizeof(kTestText)), 0); + } +} + +} // namespace syscall_broker + +} // namespace sandbox diff --git a/sandbox/linux/system_headers/arm64_linux_syscalls.h b/sandbox/linux/system_headers/arm64_linux_syscalls.h new file mode 100644 index 0000000000..8acb2d1000 --- /dev/null +++ b/sandbox/linux/system_headers/arm64_linux_syscalls.h @@ -0,0 +1,1062 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_ARM64_LINUX_SYSCALLS_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_ARM64_LINUX_SYSCALLS_H_ + +#include <asm-generic/unistd.h> + +#if !defined(__NR_io_setup) +#define __NR_io_setup 0 +#endif + +#if !defined(__NR_io_destroy) +#define __NR_io_destroy 1 +#endif + +#if !defined(__NR_io_submit) +#define __NR_io_submit 2 +#endif + +#if !defined(__NR_io_cancel) +#define __NR_io_cancel 3 +#endif + +#if !defined(__NR_io_getevents) +#define __NR_io_getevents 4 +#endif + +#if !defined(__NR_setxattr) +#define __NR_setxattr 5 +#endif + +#if !defined(__NR_lsetxattr) +#define __NR_lsetxattr 6 +#endif + +#if !defined(__NR_fsetxattr) +#define __NR_fsetxattr 7 +#endif + +#if !defined(__NR_getxattr) +#define __NR_getxattr 8 +#endif + +#if !defined(__NR_lgetxattr) +#define __NR_lgetxattr 9 +#endif + +#if !defined(__NR_fgetxattr) +#define __NR_fgetxattr 10 +#endif + +#if !defined(__NR_listxattr) +#define __NR_listxattr 11 +#endif + +#if !defined(__NR_llistxattr) +#define __NR_llistxattr 12 +#endif + +#if !defined(__NR_flistxattr) +#define __NR_flistxattr 13 +#endif + +#if !defined(__NR_removexattr) +#define __NR_removexattr 14 +#endif + +#if !defined(__NR_lremovexattr) +#define __NR_lremovexattr 15 +#endif + +#if !defined(__NR_fremovexattr) +#define __NR_fremovexattr 16 +#endif + +#if !defined(__NR_getcwd) +#define __NR_getcwd 17 +#endif + +#if !defined(__NR_lookup_dcookie) +#define __NR_lookup_dcookie 18 +#endif + +#if !defined(__NR_eventfd2) +#define __NR_eventfd2 19 +#endif + +#if !defined(__NR_epoll_create1) +#define __NR_epoll_create1 20 +#endif + +#if !defined(__NR_epoll_ctl) +#define __NR_epoll_ctl 21 +#endif + +#if !defined(__NR_epoll_pwait) +#define __NR_epoll_pwait 22 +#endif + +#if !defined(__NR_dup) +#define __NR_dup 23 +#endif + +#if !defined(__NR_dup3) +#define __NR_dup3 24 +#endif + +#if !defined(__NR_fcntl) +#define __NR_fcntl 25 +#endif + +#if !defined(__NR_inotify_init1) +#define __NR_inotify_init1 26 +#endif + +#if !defined(__NR_inotify_add_watch) +#define __NR_inotify_add_watch 27 +#endif + +#if !defined(__NR_inotify_rm_watch) +#define __NR_inotify_rm_watch 28 +#endif + +#if !defined(__NR_ioctl) +#define __NR_ioctl 29 +#endif + +#if !defined(__NR_ioprio_set) +#define __NR_ioprio_set 30 +#endif + +#if !defined(__NR_ioprio_get) +#define __NR_ioprio_get 31 +#endif + +#if !defined(__NR_flock) +#define __NR_flock 32 +#endif + +#if !defined(__NR_mknodat) +#define __NR_mknodat 33 +#endif + +#if !defined(__NR_mkdirat) +#define __NR_mkdirat 34 +#endif + +#if !defined(__NR_unlinkat) +#define __NR_unlinkat 35 +#endif + +#if !defined(__NR_symlinkat) +#define __NR_symlinkat 36 +#endif + +#if !defined(__NR_linkat) +#define __NR_linkat 37 +#endif + +#if !defined(__NR_renameat) +#define __NR_renameat 38 +#endif + +#if !defined(__NR_umount2) +#define __NR_umount2 39 +#endif + +#if !defined(__NR_mount) +#define __NR_mount 40 +#endif + +#if !defined(__NR_pivot_root) +#define __NR_pivot_root 41 +#endif + +#if !defined(__NR_nfsservctl) +#define __NR_nfsservctl 42 +#endif + +#if !defined(__NR_statfs) +#define __NR_statfs 43 +#endif + +#if !defined(__NR_fstatfs) +#define __NR_fstatfs 44 +#endif + +#if !defined(__NR_truncate) +#define __NR_truncate 45 +#endif + +#if !defined(__NR_ftruncate) +#define __NR_ftruncate 46 +#endif + +#if !defined(__NR_fallocate) +#define __NR_fallocate 47 +#endif + +#if !defined(__NR_faccessat) +#define __NR_faccessat 48 +#endif + +#if !defined(__NR_chdir) +#define __NR_chdir 49 +#endif + +#if !defined(__NR_fchdir) +#define __NR_fchdir 50 +#endif + +#if !defined(__NR_chroot) +#define __NR_chroot 51 +#endif + +#if !defined(__NR_fchmod) +#define __NR_fchmod 52 +#endif + +#if !defined(__NR_fchmodat) +#define __NR_fchmodat 53 +#endif + +#if !defined(__NR_fchownat) +#define __NR_fchownat 54 +#endif + +#if !defined(__NR_fchown) +#define __NR_fchown 55 +#endif + +#if !defined(__NR_openat) +#define __NR_openat 56 +#endif + +#if !defined(__NR_close) +#define __NR_close 57 +#endif + +#if !defined(__NR_vhangup) +#define __NR_vhangup 58 +#endif + +#if !defined(__NR_pipe2) +#define __NR_pipe2 59 +#endif + +#if !defined(__NR_quotactl) +#define __NR_quotactl 60 +#endif + +#if !defined(__NR_getdents64) +#define __NR_getdents64 61 +#endif + +#if !defined(__NR_lseek) +#define __NR_lseek 62 +#endif + +#if !defined(__NR_read) +#define __NR_read 63 +#endif + +#if !defined(__NR_write) +#define __NR_write 64 +#endif + +#if !defined(__NR_readv) +#define __NR_readv 65 +#endif + +#if !defined(__NR_writev) +#define __NR_writev 66 +#endif + +#if !defined(__NR_pread64) +#define __NR_pread64 67 +#endif + +#if !defined(__NR_pwrite64) +#define __NR_pwrite64 68 +#endif + +#if !defined(__NR_preadv) +#define __NR_preadv 69 +#endif + +#if !defined(__NR_pwritev) +#define __NR_pwritev 70 +#endif + +#if !defined(__NR_sendfile) +#define __NR_sendfile 71 +#endif + +#if !defined(__NR_pselect6) +#define __NR_pselect6 72 +#endif + +#if !defined(__NR_ppoll) +#define __NR_ppoll 73 +#endif + +#if !defined(__NR_signalfd4) +#define __NR_signalfd4 74 +#endif + +#if !defined(__NR_vmsplice) +#define __NR_vmsplice 75 +#endif + +#if !defined(__NR_splice) +#define __NR_splice 76 +#endif + +#if !defined(__NR_tee) +#define __NR_tee 77 +#endif + +#if !defined(__NR_readlinkat) +#define __NR_readlinkat 78 +#endif + +#if !defined(__NR_newfstatat) +#define __NR_newfstatat 79 +#endif + +#if !defined(__NR_fstat) +#define __NR_fstat 80 +#endif + +#if !defined(__NR_sync) +#define __NR_sync 81 +#endif + +#if !defined(__NR_fsync) +#define __NR_fsync 82 +#endif + +#if !defined(__NR_fdatasync) +#define __NR_fdatasync 83 +#endif + +#if !defined(__NR_sync_file_range) +#define __NR_sync_file_range 84 +#endif + +#if !defined(__NR_timerfd_create) +#define __NR_timerfd_create 85 +#endif + +#if !defined(__NR_timerfd_settime) +#define __NR_timerfd_settime 86 +#endif + +#if !defined(__NR_timerfd_gettime) +#define __NR_timerfd_gettime 87 +#endif + +#if !defined(__NR_utimensat) +#define __NR_utimensat 88 +#endif + +#if !defined(__NR_acct) +#define __NR_acct 89 +#endif + +#if !defined(__NR_capget) +#define __NR_capget 90 +#endif + +#if !defined(__NR_capset) +#define __NR_capset 91 +#endif + +#if !defined(__NR_personality) +#define __NR_personality 92 +#endif + +#if !defined(__NR_exit) +#define __NR_exit 93 +#endif + +#if !defined(__NR_exit_group) +#define __NR_exit_group 94 +#endif + +#if !defined(__NR_waitid) +#define __NR_waitid 95 +#endif + +#if !defined(__NR_set_tid_address) +#define __NR_set_tid_address 96 +#endif + +#if !defined(__NR_unshare) +#define __NR_unshare 97 +#endif + +#if !defined(__NR_futex) +#define __NR_futex 98 +#endif + +#if !defined(__NR_set_robust_list) +#define __NR_set_robust_list 99 +#endif + +#if !defined(__NR_get_robust_list) +#define __NR_get_robust_list 100 +#endif + +#if !defined(__NR_nanosleep) +#define __NR_nanosleep 101 +#endif + +#if !defined(__NR_getitimer) +#define __NR_getitimer 102 +#endif + +#if !defined(__NR_setitimer) +#define __NR_setitimer 103 +#endif + +#if !defined(__NR_kexec_load) +#define __NR_kexec_load 104 +#endif + +#if !defined(__NR_init_module) +#define __NR_init_module 105 +#endif + +#if !defined(__NR_delete_module) +#define __NR_delete_module 106 +#endif + +#if !defined(__NR_timer_create) +#define __NR_timer_create 107 +#endif + +#if !defined(__NR_timer_gettime) +#define __NR_timer_gettime 108 +#endif + +#if !defined(__NR_timer_getoverrun) +#define __NR_timer_getoverrun 109 +#endif + +#if !defined(__NR_timer_settime) +#define __NR_timer_settime 110 +#endif + +#if !defined(__NR_timer_delete) +#define __NR_timer_delete 111 +#endif + +#if !defined(__NR_clock_settime) +#define __NR_clock_settime 112 +#endif + +#if !defined(__NR_clock_gettime) +#define __NR_clock_gettime 113 +#endif + +#if !defined(__NR_clock_getres) +#define __NR_clock_getres 114 +#endif + +#if !defined(__NR_clock_nanosleep) +#define __NR_clock_nanosleep 115 +#endif + +#if !defined(__NR_syslog) +#define __NR_syslog 116 +#endif + +#if !defined(__NR_ptrace) +#define __NR_ptrace 117 +#endif + +#if !defined(__NR_sched_setparam) +#define __NR_sched_setparam 118 +#endif + +#if !defined(__NR_sched_setscheduler) +#define __NR_sched_setscheduler 119 +#endif + +#if !defined(__NR_sched_getscheduler) +#define __NR_sched_getscheduler 120 +#endif + +#if !defined(__NR_sched_getparam) +#define __NR_sched_getparam 121 +#endif + +#if !defined(__NR_sched_setaffinity) +#define __NR_sched_setaffinity 122 +#endif + +#if !defined(__NR_sched_getaffinity) +#define __NR_sched_getaffinity 123 +#endif + +#if !defined(__NR_sched_yield) +#define __NR_sched_yield 124 +#endif + +#if !defined(__NR_sched_get_priority_max) +#define __NR_sched_get_priority_max 125 +#endif + +#if !defined(__NR_sched_get_priority_min) +#define __NR_sched_get_priority_min 126 +#endif + +#if !defined(__NR_sched_rr_get_interval) +#define __NR_sched_rr_get_interval 127 +#endif + +#if !defined(__NR_restart_syscall) +#define __NR_restart_syscall 128 +#endif + +#if !defined(__NR_kill) +#define __NR_kill 129 +#endif + +#if !defined(__NR_tkill) +#define __NR_tkill 130 +#endif + +#if !defined(__NR_tgkill) +#define __NR_tgkill 131 +#endif + +#if !defined(__NR_sigaltstack) +#define __NR_sigaltstack 132 +#endif + +#if !defined(__NR_rt_sigsuspend) +#define __NR_rt_sigsuspend 133 +#endif + +#if !defined(__NR_rt_sigaction) +#define __NR_rt_sigaction 134 +#endif + +#if !defined(__NR_rt_sigprocmask) +#define __NR_rt_sigprocmask 135 +#endif + +#if !defined(__NR_rt_sigpending) +#define __NR_rt_sigpending 136 +#endif + +#if !defined(__NR_rt_sigtimedwait) +#define __NR_rt_sigtimedwait 137 +#endif + +#if !defined(__NR_rt_sigqueueinfo) +#define __NR_rt_sigqueueinfo 138 +#endif + +#if !defined(__NR_rt_sigreturn) +#define __NR_rt_sigreturn 139 +#endif + +#if !defined(__NR_setpriority) +#define __NR_setpriority 140 +#endif + +#if !defined(__NR_getpriority) +#define __NR_getpriority 141 +#endif + +#if !defined(__NR_reboot) +#define __NR_reboot 142 +#endif + +#if !defined(__NR_setregid) +#define __NR_setregid 143 +#endif + +#if !defined(__NR_setgid) +#define __NR_setgid 144 +#endif + +#if !defined(__NR_setreuid) +#define __NR_setreuid 145 +#endif + +#if !defined(__NR_setuid) +#define __NR_setuid 146 +#endif + +#if !defined(__NR_setresuid) +#define __NR_setresuid 147 +#endif + +#if !defined(__NR_getresuid) +#define __NR_getresuid 148 +#endif + +#if !defined(__NR_setresgid) +#define __NR_setresgid 149 +#endif + +#if !defined(__NR_getresgid) +#define __NR_getresgid 150 +#endif + +#if !defined(__NR_setfsuid) +#define __NR_setfsuid 151 +#endif + +#if !defined(__NR_setfsgid) +#define __NR_setfsgid 152 +#endif + +#if !defined(__NR_times) +#define __NR_times 153 +#endif + +#if !defined(__NR_setpgid) +#define __NR_setpgid 154 +#endif + +#if !defined(__NR_getpgid) +#define __NR_getpgid 155 +#endif + +#if !defined(__NR_getsid) +#define __NR_getsid 156 +#endif + +#if !defined(__NR_setsid) +#define __NR_setsid 157 +#endif + +#if !defined(__NR_getgroups) +#define __NR_getgroups 158 +#endif + +#if !defined(__NR_setgroups) +#define __NR_setgroups 159 +#endif + +#if !defined(__NR_uname) +#define __NR_uname 160 +#endif + +#if !defined(__NR_sethostname) +#define __NR_sethostname 161 +#endif + +#if !defined(__NR_setdomainname) +#define __NR_setdomainname 162 +#endif + +#if !defined(__NR_getrlimit) +#define __NR_getrlimit 163 +#endif + +#if !defined(__NR_setrlimit) +#define __NR_setrlimit 164 +#endif + +#if !defined(__NR_getrusage) +#define __NR_getrusage 165 +#endif + +#if !defined(__NR_umask) +#define __NR_umask 166 +#endif + +#if !defined(__NR_prctl) +#define __NR_prctl 167 +#endif + +#if !defined(__NR_getcpu) +#define __NR_getcpu 168 +#endif + +#if !defined(__NR_gettimeofday) +#define __NR_gettimeofday 169 +#endif + +#if !defined(__NR_settimeofday) +#define __NR_settimeofday 170 +#endif + +#if !defined(__NR_adjtimex) +#define __NR_adjtimex 171 +#endif + +#if !defined(__NR_getpid) +#define __NR_getpid 172 +#endif + +#if !defined(__NR_getppid) +#define __NR_getppid 173 +#endif + +#if !defined(__NR_getuid) +#define __NR_getuid 174 +#endif + +#if !defined(__NR_geteuid) +#define __NR_geteuid 175 +#endif + +#if !defined(__NR_getgid) +#define __NR_getgid 176 +#endif + +#if !defined(__NR_getegid) +#define __NR_getegid 177 +#endif + +#if !defined(__NR_gettid) +#define __NR_gettid 178 +#endif + +#if !defined(__NR_sysinfo) +#define __NR_sysinfo 179 +#endif + +#if !defined(__NR_mq_open) +#define __NR_mq_open 180 +#endif + +#if !defined(__NR_mq_unlink) +#define __NR_mq_unlink 181 +#endif + +#if !defined(__NR_mq_timedsend) +#define __NR_mq_timedsend 182 +#endif + +#if !defined(__NR_mq_timedreceive) +#define __NR_mq_timedreceive 183 +#endif + +#if !defined(__NR_mq_notify) +#define __NR_mq_notify 184 +#endif + +#if !defined(__NR_mq_getsetattr) +#define __NR_mq_getsetattr 185 +#endif + +#if !defined(__NR_msgget) +#define __NR_msgget 186 +#endif + +#if !defined(__NR_msgctl) +#define __NR_msgctl 187 +#endif + +#if !defined(__NR_msgrcv) +#define __NR_msgrcv 188 +#endif + +#if !defined(__NR_msgsnd) +#define __NR_msgsnd 189 +#endif + +#if !defined(__NR_semget) +#define __NR_semget 190 +#endif + +#if !defined(__NR_semctl) +#define __NR_semctl 191 +#endif + +#if !defined(__NR_semtimedop) +#define __NR_semtimedop 192 +#endif + +#if !defined(__NR_semop) +#define __NR_semop 193 +#endif + +#if !defined(__NR_shmget) +#define __NR_shmget 194 +#endif + +#if !defined(__NR_shmctl) +#define __NR_shmctl 195 +#endif + +#if !defined(__NR_shmat) +#define __NR_shmat 196 +#endif + +#if !defined(__NR_shmdt) +#define __NR_shmdt 197 +#endif + +#if !defined(__NR_socket) +#define __NR_socket 198 +#endif + +#if !defined(__NR_socketpair) +#define __NR_socketpair 199 +#endif + +#if !defined(__NR_bind) +#define __NR_bind 200 +#endif + +#if !defined(__NR_listen) +#define __NR_listen 201 +#endif + +#if !defined(__NR_accept) +#define __NR_accept 202 +#endif + +#if !defined(__NR_connect) +#define __NR_connect 203 +#endif + +#if !defined(__NR_getsockname) +#define __NR_getsockname 204 +#endif + +#if !defined(__NR_getpeername) +#define __NR_getpeername 205 +#endif + +#if !defined(__NR_sendto) +#define __NR_sendto 206 +#endif + +#if !defined(__NR_recvfrom) +#define __NR_recvfrom 207 +#endif + +#if !defined(__NR_setsockopt) +#define __NR_setsockopt 208 +#endif + +#if !defined(__NR_getsockopt) +#define __NR_getsockopt 209 +#endif + +#if !defined(__NR_shutdown) +#define __NR_shutdown 210 +#endif + +#if !defined(__NR_sendmsg) +#define __NR_sendmsg 211 +#endif + +#if !defined(__NR_recvmsg) +#define __NR_recvmsg 212 +#endif + +#if !defined(__NR_readahead) +#define __NR_readahead 213 +#endif + +#if !defined(__NR_brk) +#define __NR_brk 214 +#endif + +#if !defined(__NR_munmap) +#define __NR_munmap 215 +#endif + +#if !defined(__NR_mremap) +#define __NR_mremap 216 +#endif + +#if !defined(__NR_add_key) +#define __NR_add_key 217 +#endif + +#if !defined(__NR_request_key) +#define __NR_request_key 218 +#endif + +#if !defined(__NR_keyctl) +#define __NR_keyctl 219 +#endif + +#if !defined(__NR_clone) +#define __NR_clone 220 +#endif + +#if !defined(__NR_execve) +#define __NR_execve 221 +#endif + +#if !defined(__NR_mmap) +#define __NR_mmap 222 +#endif + +#if !defined(__NR_fadvise64) +#define __NR_fadvise64 223 +#endif + +#if !defined(__NR_swapon) +#define __NR_swapon 224 +#endif + +#if !defined(__NR_swapoff) +#define __NR_swapoff 225 +#endif + +#if !defined(__NR_mprotect) +#define __NR_mprotect 226 +#endif + +#if !defined(__NR_msync) +#define __NR_msync 227 +#endif + +#if !defined(__NR_mlock) +#define __NR_mlock 228 +#endif + +#if !defined(__NR_munlock) +#define __NR_munlock 229 +#endif + +#if !defined(__NR_mlockall) +#define __NR_mlockall 230 +#endif + +#if !defined(__NR_munlockall) +#define __NR_munlockall 231 +#endif + +#if !defined(__NR_mincore) +#define __NR_mincore 232 +#endif + +#if !defined(__NR_madvise) +#define __NR_madvise 233 +#endif + +#if !defined(__NR_remap_file_pages) +#define __NR_remap_file_pages 234 +#endif + +#if !defined(__NR_mbind) +#define __NR_mbind 235 +#endif + +#if !defined(__NR_get_mempolicy) +#define __NR_get_mempolicy 236 +#endif + +#if !defined(__NR_set_mempolicy) +#define __NR_set_mempolicy 237 +#endif + +#if !defined(__NR_migrate_pages) +#define __NR_migrate_pages 238 +#endif + +#if !defined(__NR_move_pages) +#define __NR_move_pages 239 +#endif + +#if !defined(__NR_rt_tgsigqueueinfo) +#define __NR_rt_tgsigqueueinfo 240 +#endif + +#if !defined(__NR_perf_event_open) +#define __NR_perf_event_open 241 +#endif + +#if !defined(__NR_accept4) +#define __NR_accept4 242 +#endif + +#if !defined(__NR_recvmmsg) +#define __NR_recvmmsg 243 +#endif + +#if !defined(__NR_wait4) +#define __NR_wait4 260 +#endif + +#if !defined(__NR_prlimit64) +#define __NR_prlimit64 261 +#endif + +#if !defined(__NR_fanotify_init) +#define __NR_fanotify_init 262 +#endif + +#if !defined(__NR_fanotify_mark) +#define __NR_fanotify_mark 263 +#endif + +#if !defined(__NR_name_to_handle_at) +#define __NR_name_to_handle_at 264 +#endif + +#if !defined(__NR_open_by_handle_at) +#define __NR_open_by_handle_at 265 +#endif + +#if !defined(__NR_clock_adjtime) +#define __NR_clock_adjtime 266 +#endif + +#if !defined(__NR_syncfs) +#define __NR_syncfs 267 +#endif + +#if !defined(__NR_setns) +#define __NR_setns 268 +#endif + +#if !defined(__NR_sendmmsg) +#define __NR_sendmmsg 269 +#endif + +#if !defined(__NR_process_vm_readv) +#define __NR_process_vm_readv 270 +#endif + +#if !defined(__NR_process_vm_writev) +#define __NR_process_vm_writev 271 +#endif + +#if !defined(__NR_kcmp) +#define __NR_kcmp 272 +#endif + +#if !defined(__NR_finit_module) +#define __NR_finit_module 273 +#endif + +#if !defined(__NR_sched_setattr) +#define __NR_sched_setattr 274 +#endif + +#if !defined(__NR_sched_getattr) +#define __NR_sched_getattr 275 +#endif + +#if !defined(__NR_renameat2) +#define __NR_renameat2 276 +#endif + +#if !defined(__NR_seccomp) +#define __NR_seccomp 277 +#endif + +#if !defined(__NR_getrandom) +#define __NR_getrandom 278 +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_ARM64_LINUX_SYSCALLS_H_ diff --git a/sandbox/linux/system_headers/arm64_linux_ucontext.h b/sandbox/linux/system_headers/arm64_linux_ucontext.h new file mode 100644 index 0000000000..46e0407599 --- /dev/null +++ b/sandbox/linux/system_headers/arm64_linux_ucontext.h @@ -0,0 +1,29 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_ARM64_LINUX_UCONTEXT_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_ARM64_LINUX_UCONTEXT_H_ + +#if !defined(__BIONIC_HAVE_UCONTEXT_T) +#include <asm/sigcontext.h> +#include <signal.h> +// We also need greg_t for the sandbox, include it in this header as well. +typedef uint64_t greg_t; + +struct ucontext_t { + unsigned long uc_flags; + struct ucontext* uc_link; + stack_t uc_stack; + sigset_t uc_sigmask; + /* glibc uses a 1024-bit sigset_t */ + uint8_t unused[1024 / 8 - sizeof(sigset_t)]; + /* last for future expansion */ + struct sigcontext uc_mcontext; +}; + +#else +#include <sys/ucontext.h> +#endif // __BIONIC_HAVE_UCONTEXT_T + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_ARM64_LINUX_UCONTEXT_H_ diff --git a/sandbox/linux/system_headers/arm_linux_syscalls.h b/sandbox/linux/system_headers/arm_linux_syscalls.h new file mode 100644 index 0000000000..1addd53843 --- /dev/null +++ b/sandbox/linux/system_headers/arm_linux_syscalls.h @@ -0,0 +1,1418 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// Generated from the Linux kernel's calls.S. +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_ARM_LINUX_SYSCALLS_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_ARM_LINUX_SYSCALLS_H_ + +#if !defined(__arm__) || !defined(__ARM_EABI__) +#error "Including header on wrong architecture" +#endif + +#if !defined(__NR_SYSCALL_BASE) +// On ARM EABI arch, __NR_SYSCALL_BASE is 0. +#define __NR_SYSCALL_BASE 0 +#endif + +// This syscall list has holes, because ARM EABI makes some syscalls obsolete. + +#if !defined(__NR_restart_syscall) +#define __NR_restart_syscall (__NR_SYSCALL_BASE+0) +#endif + +#if !defined(__NR_exit) +#define __NR_exit (__NR_SYSCALL_BASE+1) +#endif + +#if !defined(__NR_fork) +#define __NR_fork (__NR_SYSCALL_BASE+2) +#endif + +#if !defined(__NR_read) +#define __NR_read (__NR_SYSCALL_BASE+3) +#endif + +#if !defined(__NR_write) +#define __NR_write (__NR_SYSCALL_BASE+4) +#endif + +#if !defined(__NR_open) +#define __NR_open (__NR_SYSCALL_BASE+5) +#endif + +#if !defined(__NR_close) +#define __NR_close (__NR_SYSCALL_BASE+6) +#endif + +#if !defined(__NR_creat) +#define __NR_creat (__NR_SYSCALL_BASE+8) +#endif + +#if !defined(__NR_link) +#define __NR_link (__NR_SYSCALL_BASE+9) +#endif + +#if !defined(__NR_unlink) +#define __NR_unlink (__NR_SYSCALL_BASE+10) +#endif + +#if !defined(__NR_execve) +#define __NR_execve (__NR_SYSCALL_BASE+11) +#endif + +#if !defined(__NR_chdir) +#define __NR_chdir (__NR_SYSCALL_BASE+12) +#endif + +#if !defined(__NR_mknod) +#define __NR_mknod (__NR_SYSCALL_BASE+14) +#endif + +#if !defined(__NR_chmod) +#define __NR_chmod (__NR_SYSCALL_BASE+15) +#endif + +#if !defined(__NR_lchown) +#define __NR_lchown (__NR_SYSCALL_BASE+16) +#endif + +#if !defined(__NR_lseek) +#define __NR_lseek (__NR_SYSCALL_BASE+19) +#endif + +#if !defined(__NR_getpid) +#define __NR_getpid (__NR_SYSCALL_BASE+20) +#endif + +#if !defined(__NR_mount) +#define __NR_mount (__NR_SYSCALL_BASE+21) +#endif + +#if !defined(__NR_setuid) +#define __NR_setuid (__NR_SYSCALL_BASE+23) +#endif + +#if !defined(__NR_getuid) +#define __NR_getuid (__NR_SYSCALL_BASE+24) +#endif + +#if !defined(__NR_ptrace) +#define __NR_ptrace (__NR_SYSCALL_BASE+26) +#endif + +#if !defined(__NR_pause) +#define __NR_pause (__NR_SYSCALL_BASE+29) +#endif + +#if !defined(__NR_access) +#define __NR_access (__NR_SYSCALL_BASE+33) +#endif + +#if !defined(__NR_nice) +#define __NR_nice (__NR_SYSCALL_BASE+34) +#endif + +#if !defined(__NR_sync) +#define __NR_sync (__NR_SYSCALL_BASE+36) +#endif + +#if !defined(__NR_kill) +#define __NR_kill (__NR_SYSCALL_BASE+37) +#endif + +#if !defined(__NR_rename) +#define __NR_rename (__NR_SYSCALL_BASE+38) +#endif + +#if !defined(__NR_mkdir) +#define __NR_mkdir (__NR_SYSCALL_BASE+39) +#endif + +#if !defined(__NR_rmdir) +#define __NR_rmdir (__NR_SYSCALL_BASE+40) +#endif + +#if !defined(__NR_dup) +#define __NR_dup (__NR_SYSCALL_BASE+41) +#endif + +#if !defined(__NR_pipe) +#define __NR_pipe (__NR_SYSCALL_BASE+42) +#endif + +#if !defined(__NR_times) +#define __NR_times (__NR_SYSCALL_BASE+43) +#endif + +#if !defined(__NR_brk) +#define __NR_brk (__NR_SYSCALL_BASE+45) +#endif + +#if !defined(__NR_setgid) +#define __NR_setgid (__NR_SYSCALL_BASE+46) +#endif + +#if !defined(__NR_getgid) +#define __NR_getgid (__NR_SYSCALL_BASE+47) +#endif + +#if !defined(__NR_geteuid) +#define __NR_geteuid (__NR_SYSCALL_BASE+49) +#endif + +#if !defined(__NR_getegid) +#define __NR_getegid (__NR_SYSCALL_BASE+50) +#endif + +#if !defined(__NR_acct) +#define __NR_acct (__NR_SYSCALL_BASE+51) +#endif + +#if !defined(__NR_umount2) +#define __NR_umount2 (__NR_SYSCALL_BASE+52) +#endif + +#if !defined(__NR_ioctl) +#define __NR_ioctl (__NR_SYSCALL_BASE+54) +#endif + +#if !defined(__NR_fcntl) +#define __NR_fcntl (__NR_SYSCALL_BASE+55) +#endif + +#if !defined(__NR_setpgid) +#define __NR_setpgid (__NR_SYSCALL_BASE+57) +#endif + +#if !defined(__NR_umask) +#define __NR_umask (__NR_SYSCALL_BASE+60) +#endif + +#if !defined(__NR_chroot) +#define __NR_chroot (__NR_SYSCALL_BASE+61) +#endif + +#if !defined(__NR_ustat) +#define __NR_ustat (__NR_SYSCALL_BASE+62) +#endif + +#if !defined(__NR_dup2) +#define __NR_dup2 (__NR_SYSCALL_BASE+63) +#endif + +#if !defined(__NR_getppid) +#define __NR_getppid (__NR_SYSCALL_BASE+64) +#endif + +#if !defined(__NR_getpgrp) +#define __NR_getpgrp (__NR_SYSCALL_BASE+65) +#endif + +#if !defined(__NR_setsid) +#define __NR_setsid (__NR_SYSCALL_BASE+66) +#endif + +#if !defined(__NR_sigaction) +#define __NR_sigaction (__NR_SYSCALL_BASE+67) +#endif + +#if !defined(__NR_setreuid) +#define __NR_setreuid (__NR_SYSCALL_BASE+70) +#endif + +#if !defined(__NR_setregid) +#define __NR_setregid (__NR_SYSCALL_BASE+71) +#endif + +#if !defined(__NR_sigsuspend) +#define __NR_sigsuspend (__NR_SYSCALL_BASE+72) +#endif + +#if !defined(__NR_sigpending) +#define __NR_sigpending (__NR_SYSCALL_BASE+73) +#endif + +#if !defined(__NR_sethostname) +#define __NR_sethostname (__NR_SYSCALL_BASE+74) +#endif + +#if !defined(__NR_setrlimit) +#define __NR_setrlimit (__NR_SYSCALL_BASE+75) +#endif + +#if !defined(__NR_getrusage) +#define __NR_getrusage (__NR_SYSCALL_BASE+77) +#endif + +#if !defined(__NR_gettimeofday) +#define __NR_gettimeofday (__NR_SYSCALL_BASE+78) +#endif + +#if !defined(__NR_settimeofday) +#define __NR_settimeofday (__NR_SYSCALL_BASE+79) +#endif + +#if !defined(__NR_getgroups) +#define __NR_getgroups (__NR_SYSCALL_BASE+80) +#endif + +#if !defined(__NR_setgroups) +#define __NR_setgroups (__NR_SYSCALL_BASE+81) +#endif + +#if !defined(__NR_symlink) +#define __NR_symlink (__NR_SYSCALL_BASE+83) +#endif + +#if !defined(__NR_readlink) +#define __NR_readlink (__NR_SYSCALL_BASE+85) +#endif + +#if !defined(__NR_uselib) +#define __NR_uselib (__NR_SYSCALL_BASE+86) +#endif + +#if !defined(__NR_swapon) +#define __NR_swapon (__NR_SYSCALL_BASE+87) +#endif + +#if !defined(__NR_reboot) +#define __NR_reboot (__NR_SYSCALL_BASE+88) +#endif + +#if !defined(__NR_munmap) +#define __NR_munmap (__NR_SYSCALL_BASE+91) +#endif + +#if !defined(__NR_truncate) +#define __NR_truncate (__NR_SYSCALL_BASE+92) +#endif + +#if !defined(__NR_ftruncate) +#define __NR_ftruncate (__NR_SYSCALL_BASE+93) +#endif + +#if !defined(__NR_fchmod) +#define __NR_fchmod (__NR_SYSCALL_BASE+94) +#endif + +#if !defined(__NR_fchown) +#define __NR_fchown (__NR_SYSCALL_BASE+95) +#endif + +#if !defined(__NR_getpriority) +#define __NR_getpriority (__NR_SYSCALL_BASE+96) +#endif + +#if !defined(__NR_setpriority) +#define __NR_setpriority (__NR_SYSCALL_BASE+97) +#endif + +#if !defined(__NR_statfs) +#define __NR_statfs (__NR_SYSCALL_BASE+99) +#endif + +#if !defined(__NR_fstatfs) +#define __NR_fstatfs (__NR_SYSCALL_BASE+100) +#endif + +#if !defined(__NR_syslog) +#define __NR_syslog (__NR_SYSCALL_BASE+103) +#endif + +#if !defined(__NR_setitimer) +#define __NR_setitimer (__NR_SYSCALL_BASE+104) +#endif + +#if !defined(__NR_getitimer) +#define __NR_getitimer (__NR_SYSCALL_BASE+105) +#endif + +#if !defined(__NR_stat) +#define __NR_stat (__NR_SYSCALL_BASE+106) +#endif + +#if !defined(__NR_lstat) +#define __NR_lstat (__NR_SYSCALL_BASE+107) +#endif + +#if !defined(__NR_fstat) +#define __NR_fstat (__NR_SYSCALL_BASE+108) +#endif + +#if !defined(__NR_vhangup) +#define __NR_vhangup (__NR_SYSCALL_BASE+111) +#endif + +#if !defined(__NR_wait4) +#define __NR_wait4 (__NR_SYSCALL_BASE+114) +#endif + +#if !defined(__NR_swapoff) +#define __NR_swapoff (__NR_SYSCALL_BASE+115) +#endif + +#if !defined(__NR_sysinfo) +#define __NR_sysinfo (__NR_SYSCALL_BASE+116) +#endif + +#if !defined(__NR_fsync) +#define __NR_fsync (__NR_SYSCALL_BASE+118) +#endif + +#if !defined(__NR_sigreturn) +#define __NR_sigreturn (__NR_SYSCALL_BASE+119) +#endif + +#if !defined(__NR_clone) +#define __NR_clone (__NR_SYSCALL_BASE+120) +#endif + +#if !defined(__NR_setdomainname) +#define __NR_setdomainname (__NR_SYSCALL_BASE+121) +#endif + +#if !defined(__NR_uname) +#define __NR_uname (__NR_SYSCALL_BASE+122) +#endif + +#if !defined(__NR_adjtimex) +#define __NR_adjtimex (__NR_SYSCALL_BASE+124) +#endif + +#if !defined(__NR_mprotect) +#define __NR_mprotect (__NR_SYSCALL_BASE+125) +#endif + +#if !defined(__NR_sigprocmask) +#define __NR_sigprocmask (__NR_SYSCALL_BASE+126) +#endif + +#if !defined(__NR_init_module) +#define __NR_init_module (__NR_SYSCALL_BASE+128) +#endif + +#if !defined(__NR_delete_module) +#define __NR_delete_module (__NR_SYSCALL_BASE+129) +#endif + +#if !defined(__NR_quotactl) +#define __NR_quotactl (__NR_SYSCALL_BASE+131) +#endif + +#if !defined(__NR_getpgid) +#define __NR_getpgid (__NR_SYSCALL_BASE+132) +#endif + +#if !defined(__NR_fchdir) +#define __NR_fchdir (__NR_SYSCALL_BASE+133) +#endif + +#if !defined(__NR_bdflush) +#define __NR_bdflush (__NR_SYSCALL_BASE+134) +#endif + +#if !defined(__NR_sysfs) +#define __NR_sysfs (__NR_SYSCALL_BASE+135) +#endif + +#if !defined(__NR_personality) +#define __NR_personality (__NR_SYSCALL_BASE+136) +#endif + +#if !defined(__NR_setfsuid) +#define __NR_setfsuid (__NR_SYSCALL_BASE+138) +#endif + +#if !defined(__NR_setfsgid) +#define __NR_setfsgid (__NR_SYSCALL_BASE+139) +#endif + +#if !defined(__NR__llseek) +#define __NR__llseek (__NR_SYSCALL_BASE+140) +#endif + +#if !defined(__NR_getdents) +#define __NR_getdents (__NR_SYSCALL_BASE+141) +#endif + +#if !defined(__NR__newselect) +#define __NR__newselect (__NR_SYSCALL_BASE+142) +#endif + +#if !defined(__NR_flock) +#define __NR_flock (__NR_SYSCALL_BASE+143) +#endif + +#if !defined(__NR_msync) +#define __NR_msync (__NR_SYSCALL_BASE+144) +#endif + +#if !defined(__NR_readv) +#define __NR_readv (__NR_SYSCALL_BASE+145) +#endif + +#if !defined(__NR_writev) +#define __NR_writev (__NR_SYSCALL_BASE+146) +#endif + +#if !defined(__NR_getsid) +#define __NR_getsid (__NR_SYSCALL_BASE+147) +#endif + +#if !defined(__NR_fdatasync) +#define __NR_fdatasync (__NR_SYSCALL_BASE+148) +#endif + +#if !defined(__NR__sysctl) +#define __NR__sysctl (__NR_SYSCALL_BASE+149) +#endif + +#if !defined(__NR_mlock) +#define __NR_mlock (__NR_SYSCALL_BASE+150) +#endif + +#if !defined(__NR_munlock) +#define __NR_munlock (__NR_SYSCALL_BASE+151) +#endif + +#if !defined(__NR_mlockall) +#define __NR_mlockall (__NR_SYSCALL_BASE+152) +#endif + +#if !defined(__NR_munlockall) +#define __NR_munlockall (__NR_SYSCALL_BASE+153) +#endif + +#if !defined(__NR_sched_setparam) +#define __NR_sched_setparam (__NR_SYSCALL_BASE+154) +#endif + +#if !defined(__NR_sched_getparam) +#define __NR_sched_getparam (__NR_SYSCALL_BASE+155) +#endif + +#if !defined(__NR_sched_setscheduler) +#define __NR_sched_setscheduler (__NR_SYSCALL_BASE+156) +#endif + +#if !defined(__NR_sched_getscheduler) +#define __NR_sched_getscheduler (__NR_SYSCALL_BASE+157) +#endif + +#if !defined(__NR_sched_yield) +#define __NR_sched_yield (__NR_SYSCALL_BASE+158) +#endif + +#if !defined(__NR_sched_get_priority_max) +#define __NR_sched_get_priority_max (__NR_SYSCALL_BASE+159) +#endif + +#if !defined(__NR_sched_get_priority_min) +#define __NR_sched_get_priority_min (__NR_SYSCALL_BASE+160) +#endif + +#if !defined(__NR_sched_rr_get_interval) +#define __NR_sched_rr_get_interval (__NR_SYSCALL_BASE+161) +#endif + +#if !defined(__NR_nanosleep) +#define __NR_nanosleep (__NR_SYSCALL_BASE+162) +#endif + +#if !defined(__NR_mremap) +#define __NR_mremap (__NR_SYSCALL_BASE+163) +#endif + +#if !defined(__NR_setresuid) +#define __NR_setresuid (__NR_SYSCALL_BASE+164) +#endif + +#if !defined(__NR_getresuid) +#define __NR_getresuid (__NR_SYSCALL_BASE+165) +#endif + +#if !defined(__NR_poll) +#define __NR_poll (__NR_SYSCALL_BASE+168) +#endif + +#if !defined(__NR_nfsservctl) +#define __NR_nfsservctl (__NR_SYSCALL_BASE+169) +#endif + +#if !defined(__NR_setresgid) +#define __NR_setresgid (__NR_SYSCALL_BASE+170) +#endif + +#if !defined(__NR_getresgid) +#define __NR_getresgid (__NR_SYSCALL_BASE+171) +#endif + +#if !defined(__NR_prctl) +#define __NR_prctl (__NR_SYSCALL_BASE+172) +#endif + +#if !defined(__NR_rt_sigreturn) +#define __NR_rt_sigreturn (__NR_SYSCALL_BASE+173) +#endif + +#if !defined(__NR_rt_sigaction) +#define __NR_rt_sigaction (__NR_SYSCALL_BASE+174) +#endif + +#if !defined(__NR_rt_sigprocmask) +#define __NR_rt_sigprocmask (__NR_SYSCALL_BASE+175) +#endif + +#if !defined(__NR_rt_sigpending) +#define __NR_rt_sigpending (__NR_SYSCALL_BASE+176) +#endif + +#if !defined(__NR_rt_sigtimedwait) +#define __NR_rt_sigtimedwait (__NR_SYSCALL_BASE+177) +#endif + +#if !defined(__NR_rt_sigqueueinfo) +#define __NR_rt_sigqueueinfo (__NR_SYSCALL_BASE+178) +#endif + +#if !defined(__NR_rt_sigsuspend) +#define __NR_rt_sigsuspend (__NR_SYSCALL_BASE+179) +#endif + +#if !defined(__NR_pread64) +#define __NR_pread64 (__NR_SYSCALL_BASE+180) +#endif + +#if !defined(__NR_pwrite64) +#define __NR_pwrite64 (__NR_SYSCALL_BASE+181) +#endif + +#if !defined(__NR_chown) +#define __NR_chown (__NR_SYSCALL_BASE+182) +#endif + +#if !defined(__NR_getcwd) +#define __NR_getcwd (__NR_SYSCALL_BASE+183) +#endif + +#if !defined(__NR_capget) +#define __NR_capget (__NR_SYSCALL_BASE+184) +#endif + +#if !defined(__NR_capset) +#define __NR_capset (__NR_SYSCALL_BASE+185) +#endif + +#if !defined(__NR_sigaltstack) +#define __NR_sigaltstack (__NR_SYSCALL_BASE+186) +#endif + +#if !defined(__NR_sendfile) +#define __NR_sendfile (__NR_SYSCALL_BASE+187) +#endif + +#if !defined(__NR_vfork) +#define __NR_vfork (__NR_SYSCALL_BASE+190) +#endif + +#if !defined(__NR_ugetrlimit) +#define __NR_ugetrlimit (__NR_SYSCALL_BASE+191) +#endif + +#if !defined(__NR_mmap2) +#define __NR_mmap2 (__NR_SYSCALL_BASE+192) +#endif + +#if !defined(__NR_truncate64) +#define __NR_truncate64 (__NR_SYSCALL_BASE+193) +#endif + +#if !defined(__NR_ftruncate64) +#define __NR_ftruncate64 (__NR_SYSCALL_BASE+194) +#endif + +#if !defined(__NR_stat64) +#define __NR_stat64 (__NR_SYSCALL_BASE+195) +#endif + +#if !defined(__NR_lstat64) +#define __NR_lstat64 (__NR_SYSCALL_BASE+196) +#endif + +#if !defined(__NR_fstat64) +#define __NR_fstat64 (__NR_SYSCALL_BASE+197) +#endif + +#if !defined(__NR_lchown32) +#define __NR_lchown32 (__NR_SYSCALL_BASE+198) +#endif + +#if !defined(__NR_getuid32) +#define __NR_getuid32 (__NR_SYSCALL_BASE+199) +#endif + +#if !defined(__NR_getgid32) +#define __NR_getgid32 (__NR_SYSCALL_BASE+200) +#endif + +#if !defined(__NR_geteuid32) +#define __NR_geteuid32 (__NR_SYSCALL_BASE+201) +#endif + +#if !defined(__NR_getegid32) +#define __NR_getegid32 (__NR_SYSCALL_BASE+202) +#endif + +#if !defined(__NR_setreuid32) +#define __NR_setreuid32 (__NR_SYSCALL_BASE+203) +#endif + +#if !defined(__NR_setregid32) +#define __NR_setregid32 (__NR_SYSCALL_BASE+204) +#endif + +#if !defined(__NR_getgroups32) +#define __NR_getgroups32 (__NR_SYSCALL_BASE+205) +#endif + +#if !defined(__NR_setgroups32) +#define __NR_setgroups32 (__NR_SYSCALL_BASE+206) +#endif + +#if !defined(__NR_fchown32) +#define __NR_fchown32 (__NR_SYSCALL_BASE+207) +#endif + +#if !defined(__NR_setresuid32) +#define __NR_setresuid32 (__NR_SYSCALL_BASE+208) +#endif + +#if !defined(__NR_getresuid32) +#define __NR_getresuid32 (__NR_SYSCALL_BASE+209) +#endif + +#if !defined(__NR_setresgid32) +#define __NR_setresgid32 (__NR_SYSCALL_BASE+210) +#endif + +#if !defined(__NR_getresgid32) +#define __NR_getresgid32 (__NR_SYSCALL_BASE+211) +#endif + +#if !defined(__NR_chown32) +#define __NR_chown32 (__NR_SYSCALL_BASE+212) +#endif + +#if !defined(__NR_setuid32) +#define __NR_setuid32 (__NR_SYSCALL_BASE+213) +#endif + +#if !defined(__NR_setgid32) +#define __NR_setgid32 (__NR_SYSCALL_BASE+214) +#endif + +#if !defined(__NR_setfsuid32) +#define __NR_setfsuid32 (__NR_SYSCALL_BASE+215) +#endif + +#if !defined(__NR_setfsgid32) +#define __NR_setfsgid32 (__NR_SYSCALL_BASE+216) +#endif + +#if !defined(__NR_getdents64) +#define __NR_getdents64 (__NR_SYSCALL_BASE+217) +#endif + +#if !defined(__NR_pivot_root) +#define __NR_pivot_root (__NR_SYSCALL_BASE+218) +#endif + +#if !defined(__NR_mincore) +#define __NR_mincore (__NR_SYSCALL_BASE+219) +#endif + +#if !defined(__NR_madvise) +#define __NR_madvise (__NR_SYSCALL_BASE+220) +#endif + +#if !defined(__NR_fcntl64) +#define __NR_fcntl64 (__NR_SYSCALL_BASE+221) +#endif + +#if !defined(__NR_gettid) +#define __NR_gettid (__NR_SYSCALL_BASE+224) +#endif + +#if !defined(__NR_readahead) +#define __NR_readahead (__NR_SYSCALL_BASE+225) +#endif + +#if !defined(__NR_setxattr) +#define __NR_setxattr (__NR_SYSCALL_BASE+226) +#endif + +#if !defined(__NR_lsetxattr) +#define __NR_lsetxattr (__NR_SYSCALL_BASE+227) +#endif + +#if !defined(__NR_fsetxattr) +#define __NR_fsetxattr (__NR_SYSCALL_BASE+228) +#endif + +#if !defined(__NR_getxattr) +#define __NR_getxattr (__NR_SYSCALL_BASE+229) +#endif + +#if !defined(__NR_lgetxattr) +#define __NR_lgetxattr (__NR_SYSCALL_BASE+230) +#endif + +#if !defined(__NR_fgetxattr) +#define __NR_fgetxattr (__NR_SYSCALL_BASE+231) +#endif + +#if !defined(__NR_listxattr) +#define __NR_listxattr (__NR_SYSCALL_BASE+232) +#endif + +#if !defined(__NR_llistxattr) +#define __NR_llistxattr (__NR_SYSCALL_BASE+233) +#endif + +#if !defined(__NR_flistxattr) +#define __NR_flistxattr (__NR_SYSCALL_BASE+234) +#endif + +#if !defined(__NR_removexattr) +#define __NR_removexattr (__NR_SYSCALL_BASE+235) +#endif + +#if !defined(__NR_lremovexattr) +#define __NR_lremovexattr (__NR_SYSCALL_BASE+236) +#endif + +#if !defined(__NR_fremovexattr) +#define __NR_fremovexattr (__NR_SYSCALL_BASE+237) +#endif + +#if !defined(__NR_tkill) +#define __NR_tkill (__NR_SYSCALL_BASE+238) +#endif + +#if !defined(__NR_sendfile64) +#define __NR_sendfile64 (__NR_SYSCALL_BASE+239) +#endif + +#if !defined(__NR_futex) +#define __NR_futex (__NR_SYSCALL_BASE+240) +#endif + +#if !defined(__NR_sched_setaffinity) +#define __NR_sched_setaffinity (__NR_SYSCALL_BASE+241) +#endif + +#if !defined(__NR_sched_getaffinity) +#define __NR_sched_getaffinity (__NR_SYSCALL_BASE+242) +#endif + +#if !defined(__NR_io_setup) +#define __NR_io_setup (__NR_SYSCALL_BASE+243) +#endif + +#if !defined(__NR_io_destroy) +#define __NR_io_destroy (__NR_SYSCALL_BASE+244) +#endif + +#if !defined(__NR_io_getevents) +#define __NR_io_getevents (__NR_SYSCALL_BASE+245) +#endif + +#if !defined(__NR_io_submit) +#define __NR_io_submit (__NR_SYSCALL_BASE+246) +#endif + +#if !defined(__NR_io_cancel) +#define __NR_io_cancel (__NR_SYSCALL_BASE+247) +#endif + +#if !defined(__NR_exit_group) +#define __NR_exit_group (__NR_SYSCALL_BASE+248) +#endif + +#if !defined(__NR_lookup_dcookie) +#define __NR_lookup_dcookie (__NR_SYSCALL_BASE+249) +#endif + +#if !defined(__NR_epoll_create) +#define __NR_epoll_create (__NR_SYSCALL_BASE+250) +#endif + +#if !defined(__NR_epoll_ctl) +#define __NR_epoll_ctl (__NR_SYSCALL_BASE+251) +#endif + +#if !defined(__NR_epoll_wait) +#define __NR_epoll_wait (__NR_SYSCALL_BASE+252) +#endif + +#if !defined(__NR_remap_file_pages) +#define __NR_remap_file_pages (__NR_SYSCALL_BASE+253) +#endif + +#if !defined(__NR_set_tid_address) +#define __NR_set_tid_address (__NR_SYSCALL_BASE+256) +#endif + +#if !defined(__NR_timer_create) +#define __NR_timer_create (__NR_SYSCALL_BASE+257) +#endif + +#if !defined(__NR_timer_settime) +#define __NR_timer_settime (__NR_SYSCALL_BASE+258) +#endif + +#if !defined(__NR_timer_gettime) +#define __NR_timer_gettime (__NR_SYSCALL_BASE+259) +#endif + +#if !defined(__NR_timer_getoverrun) +#define __NR_timer_getoverrun (__NR_SYSCALL_BASE+260) +#endif + +#if !defined(__NR_timer_delete) +#define __NR_timer_delete (__NR_SYSCALL_BASE+261) +#endif + +#if !defined(__NR_clock_settime) +#define __NR_clock_settime (__NR_SYSCALL_BASE+262) +#endif + +#if !defined(__NR_clock_gettime) +#define __NR_clock_gettime (__NR_SYSCALL_BASE+263) +#endif + +#if !defined(__NR_clock_getres) +#define __NR_clock_getres (__NR_SYSCALL_BASE+264) +#endif + +#if !defined(__NR_clock_nanosleep) +#define __NR_clock_nanosleep (__NR_SYSCALL_BASE+265) +#endif + +#if !defined(__NR_statfs64) +#define __NR_statfs64 (__NR_SYSCALL_BASE+266) +#endif + +#if !defined(__NR_fstatfs64) +#define __NR_fstatfs64 (__NR_SYSCALL_BASE+267) +#endif + +#if !defined(__NR_tgkill) +#define __NR_tgkill (__NR_SYSCALL_BASE+268) +#endif + +#if !defined(__NR_utimes) +#define __NR_utimes (__NR_SYSCALL_BASE+269) +#endif + +#if !defined(__NR_arm_fadvise64_64) +#define __NR_arm_fadvise64_64 (__NR_SYSCALL_BASE+270) +#endif + +#if !defined(__NR_pciconfig_iobase) +#define __NR_pciconfig_iobase (__NR_SYSCALL_BASE+271) +#endif + +#if !defined(__NR_pciconfig_read) +#define __NR_pciconfig_read (__NR_SYSCALL_BASE+272) +#endif + +#if !defined(__NR_pciconfig_write) +#define __NR_pciconfig_write (__NR_SYSCALL_BASE+273) +#endif + +#if !defined(__NR_mq_open) +#define __NR_mq_open (__NR_SYSCALL_BASE+274) +#endif + +#if !defined(__NR_mq_unlink) +#define __NR_mq_unlink (__NR_SYSCALL_BASE+275) +#endif + +#if !defined(__NR_mq_timedsend) +#define __NR_mq_timedsend (__NR_SYSCALL_BASE+276) +#endif + +#if !defined(__NR_mq_timedreceive) +#define __NR_mq_timedreceive (__NR_SYSCALL_BASE+277) +#endif + +#if !defined(__NR_mq_notify) +#define __NR_mq_notify (__NR_SYSCALL_BASE+278) +#endif + +#if !defined(__NR_mq_getsetattr) +#define __NR_mq_getsetattr (__NR_SYSCALL_BASE+279) +#endif + +#if !defined(__NR_waitid) +#define __NR_waitid (__NR_SYSCALL_BASE+280) +#endif + +#if !defined(__NR_socket) +#define __NR_socket (__NR_SYSCALL_BASE+281) +#endif + +#if !defined(__NR_bind) +#define __NR_bind (__NR_SYSCALL_BASE+282) +#endif + +#if !defined(__NR_connect) +#define __NR_connect (__NR_SYSCALL_BASE+283) +#endif + +#if !defined(__NR_listen) +#define __NR_listen (__NR_SYSCALL_BASE+284) +#endif + +#if !defined(__NR_accept) +#define __NR_accept (__NR_SYSCALL_BASE+285) +#endif + +#if !defined(__NR_getsockname) +#define __NR_getsockname (__NR_SYSCALL_BASE+286) +#endif + +#if !defined(__NR_getpeername) +#define __NR_getpeername (__NR_SYSCALL_BASE+287) +#endif + +#if !defined(__NR_socketpair) +#define __NR_socketpair (__NR_SYSCALL_BASE+288) +#endif + +#if !defined(__NR_send) +#define __NR_send (__NR_SYSCALL_BASE+289) +#endif + +#if !defined(__NR_sendto) +#define __NR_sendto (__NR_SYSCALL_BASE+290) +#endif + +#if !defined(__NR_recv) +#define __NR_recv (__NR_SYSCALL_BASE+291) +#endif + +#if !defined(__NR_recvfrom) +#define __NR_recvfrom (__NR_SYSCALL_BASE+292) +#endif + +#if !defined(__NR_shutdown) +#define __NR_shutdown (__NR_SYSCALL_BASE+293) +#endif + +#if !defined(__NR_setsockopt) +#define __NR_setsockopt (__NR_SYSCALL_BASE+294) +#endif + +#if !defined(__NR_getsockopt) +#define __NR_getsockopt (__NR_SYSCALL_BASE+295) +#endif + +#if !defined(__NR_sendmsg) +#define __NR_sendmsg (__NR_SYSCALL_BASE+296) +#endif + +#if !defined(__NR_recvmsg) +#define __NR_recvmsg (__NR_SYSCALL_BASE+297) +#endif + +#if !defined(__NR_semop) +#define __NR_semop (__NR_SYSCALL_BASE+298) +#endif + +#if !defined(__NR_semget) +#define __NR_semget (__NR_SYSCALL_BASE+299) +#endif + +#if !defined(__NR_semctl) +#define __NR_semctl (__NR_SYSCALL_BASE+300) +#endif + +#if !defined(__NR_msgsnd) +#define __NR_msgsnd (__NR_SYSCALL_BASE+301) +#endif + +#if !defined(__NR_msgrcv) +#define __NR_msgrcv (__NR_SYSCALL_BASE+302) +#endif + +#if !defined(__NR_msgget) +#define __NR_msgget (__NR_SYSCALL_BASE+303) +#endif + +#if !defined(__NR_msgctl) +#define __NR_msgctl (__NR_SYSCALL_BASE+304) +#endif + +#if !defined(__NR_shmat) +#define __NR_shmat (__NR_SYSCALL_BASE+305) +#endif + +#if !defined(__NR_shmdt) +#define __NR_shmdt (__NR_SYSCALL_BASE+306) +#endif + +#if !defined(__NR_shmget) +#define __NR_shmget (__NR_SYSCALL_BASE+307) +#endif + +#if !defined(__NR_shmctl) +#define __NR_shmctl (__NR_SYSCALL_BASE+308) +#endif + +#if !defined(__NR_add_key) +#define __NR_add_key (__NR_SYSCALL_BASE+309) +#endif + +#if !defined(__NR_request_key) +#define __NR_request_key (__NR_SYSCALL_BASE+310) +#endif + +#if !defined(__NR_keyctl) +#define __NR_keyctl (__NR_SYSCALL_BASE+311) +#endif + +#if !defined(__NR_semtimedop) +#define __NR_semtimedop (__NR_SYSCALL_BASE+312) +#endif + +#if !defined(__NR_vserver) +#define __NR_vserver (__NR_SYSCALL_BASE+313) +#endif + +#if !defined(__NR_ioprio_set) +#define __NR_ioprio_set (__NR_SYSCALL_BASE+314) +#endif + +#if !defined(__NR_ioprio_get) +#define __NR_ioprio_get (__NR_SYSCALL_BASE+315) +#endif + +#if !defined(__NR_inotify_init) +#define __NR_inotify_init (__NR_SYSCALL_BASE+316) +#endif + +#if !defined(__NR_inotify_add_watch) +#define __NR_inotify_add_watch (__NR_SYSCALL_BASE+317) +#endif + +#if !defined(__NR_inotify_rm_watch) +#define __NR_inotify_rm_watch (__NR_SYSCALL_BASE+318) +#endif + +#if !defined(__NR_mbind) +#define __NR_mbind (__NR_SYSCALL_BASE+319) +#endif + +#if !defined(__NR_get_mempolicy) +#define __NR_get_mempolicy (__NR_SYSCALL_BASE+320) +#endif + +#if !defined(__NR_set_mempolicy) +#define __NR_set_mempolicy (__NR_SYSCALL_BASE+321) +#endif + +#if !defined(__NR_openat) +#define __NR_openat (__NR_SYSCALL_BASE+322) +#endif + +#if !defined(__NR_mkdirat) +#define __NR_mkdirat (__NR_SYSCALL_BASE+323) +#endif + +#if !defined(__NR_mknodat) +#define __NR_mknodat (__NR_SYSCALL_BASE+324) +#endif + +#if !defined(__NR_fchownat) +#define __NR_fchownat (__NR_SYSCALL_BASE+325) +#endif + +#if !defined(__NR_futimesat) +#define __NR_futimesat (__NR_SYSCALL_BASE+326) +#endif + +#if !defined(__NR_fstatat64) +#define __NR_fstatat64 (__NR_SYSCALL_BASE+327) +#endif + +#if !defined(__NR_unlinkat) +#define __NR_unlinkat (__NR_SYSCALL_BASE+328) +#endif + +#if !defined(__NR_renameat) +#define __NR_renameat (__NR_SYSCALL_BASE+329) +#endif + +#if !defined(__NR_linkat) +#define __NR_linkat (__NR_SYSCALL_BASE+330) +#endif + +#if !defined(__NR_symlinkat) +#define __NR_symlinkat (__NR_SYSCALL_BASE+331) +#endif + +#if !defined(__NR_readlinkat) +#define __NR_readlinkat (__NR_SYSCALL_BASE+332) +#endif + +#if !defined(__NR_fchmodat) +#define __NR_fchmodat (__NR_SYSCALL_BASE+333) +#endif + +#if !defined(__NR_faccessat) +#define __NR_faccessat (__NR_SYSCALL_BASE+334) +#endif + +#if !defined(__NR_pselect6) +#define __NR_pselect6 (__NR_SYSCALL_BASE+335) +#endif + +#if !defined(__NR_ppoll) +#define __NR_ppoll (__NR_SYSCALL_BASE+336) +#endif + +#if !defined(__NR_unshare) +#define __NR_unshare (__NR_SYSCALL_BASE+337) +#endif + +#if !defined(__NR_set_robust_list) +#define __NR_set_robust_list (__NR_SYSCALL_BASE+338) +#endif + +#if !defined(__NR_get_robust_list) +#define __NR_get_robust_list (__NR_SYSCALL_BASE+339) +#endif + +#if !defined(__NR_splice) +#define __NR_splice (__NR_SYSCALL_BASE+340) +#endif + +#if !defined(__NR_arm_sync_file_range) +#define __NR_arm_sync_file_range (__NR_SYSCALL_BASE+341) +#endif + +#if !defined(__NR_sync_file_range2) +#define __NR_sync_file_range2 (__NR_SYSCALL_BASE+341) +#endif + +#if !defined(__NR_tee) +#define __NR_tee (__NR_SYSCALL_BASE+342) +#endif + +#if !defined(__NR_vmsplice) +#define __NR_vmsplice (__NR_SYSCALL_BASE+343) +#endif + +#if !defined(__NR_move_pages) +#define __NR_move_pages (__NR_SYSCALL_BASE+344) +#endif + +#if !defined(__NR_getcpu) +#define __NR_getcpu (__NR_SYSCALL_BASE+345) +#endif + +#if !defined(__NR_epoll_pwait) +#define __NR_epoll_pwait (__NR_SYSCALL_BASE+346) +#endif + +#if !defined(__NR_kexec_load) +#define __NR_kexec_load (__NR_SYSCALL_BASE+347) +#endif + +#if !defined(__NR_utimensat) +#define __NR_utimensat (__NR_SYSCALL_BASE+348) +#endif + +#if !defined(__NR_signalfd) +#define __NR_signalfd (__NR_SYSCALL_BASE+349) +#endif + +#if !defined(__NR_timerfd_create) +#define __NR_timerfd_create (__NR_SYSCALL_BASE+350) +#endif + +#if !defined(__NR_eventfd) +#define __NR_eventfd (__NR_SYSCALL_BASE+351) +#endif + +#if !defined(__NR_fallocate) +#define __NR_fallocate (__NR_SYSCALL_BASE+352) +#endif + +#if !defined(__NR_timerfd_settime) +#define __NR_timerfd_settime (__NR_SYSCALL_BASE+353) +#endif + +#if !defined(__NR_timerfd_gettime) +#define __NR_timerfd_gettime (__NR_SYSCALL_BASE+354) +#endif + +#if !defined(__NR_signalfd4) +#define __NR_signalfd4 (__NR_SYSCALL_BASE+355) +#endif + +#if !defined(__NR_eventfd2) +#define __NR_eventfd2 (__NR_SYSCALL_BASE+356) +#endif + +#if !defined(__NR_epoll_create1) +#define __NR_epoll_create1 (__NR_SYSCALL_BASE+357) +#endif + +#if !defined(__NR_dup3) +#define __NR_dup3 (__NR_SYSCALL_BASE+358) +#endif + +#if !defined(__NR_pipe2) +#define __NR_pipe2 (__NR_SYSCALL_BASE+359) +#endif + +#if !defined(__NR_inotify_init1) +#define __NR_inotify_init1 (__NR_SYSCALL_BASE+360) +#endif + +#if !defined(__NR_preadv) +#define __NR_preadv (__NR_SYSCALL_BASE+361) +#endif + +#if !defined(__NR_pwritev) +#define __NR_pwritev (__NR_SYSCALL_BASE+362) +#endif + +#if !defined(__NR_rt_tgsigqueueinfo) +#define __NR_rt_tgsigqueueinfo (__NR_SYSCALL_BASE+363) +#endif + +#if !defined(__NR_perf_event_open) +#define __NR_perf_event_open (__NR_SYSCALL_BASE+364) +#endif + +#if !defined(__NR_recvmmsg) +#define __NR_recvmmsg (__NR_SYSCALL_BASE+365) +#endif + +#if !defined(__NR_accept4) +#define __NR_accept4 (__NR_SYSCALL_BASE+366) +#endif + +#if !defined(__NR_fanotify_init) +#define __NR_fanotify_init (__NR_SYSCALL_BASE+367) +#endif + +#if !defined(__NR_fanotify_mark) +#define __NR_fanotify_mark (__NR_SYSCALL_BASE+368) +#endif + +#if !defined(__NR_prlimit64) +#define __NR_prlimit64 (__NR_SYSCALL_BASE+369) +#endif + +#if !defined(__NR_name_to_handle_at) +#define __NR_name_to_handle_at (__NR_SYSCALL_BASE+370) +#endif + +#if !defined(__NR_open_by_handle_at) +#define __NR_open_by_handle_at (__NR_SYSCALL_BASE+371) +#endif + +#if !defined(__NR_clock_adjtime) +#define __NR_clock_adjtime (__NR_SYSCALL_BASE+372) +#endif + +#if !defined(__NR_syncfs) +#define __NR_syncfs (__NR_SYSCALL_BASE+373) +#endif + +#if !defined(__NR_sendmmsg) +#define __NR_sendmmsg (__NR_SYSCALL_BASE+374) +#endif + +#if !defined(__NR_setns) +#define __NR_setns (__NR_SYSCALL_BASE+375) +#endif + +#if !defined(__NR_process_vm_readv) +#define __NR_process_vm_readv (__NR_SYSCALL_BASE+376) +#endif + +#if !defined(__NR_process_vm_writev) +#define __NR_process_vm_writev (__NR_SYSCALL_BASE+377) +#endif + +#if !defined(__NR_kcmp) +#define __NR_kcmp (__NR_SYSCALL_BASE+378) +#endif + +#if !defined(__NR_finit_module) +#define __NR_finit_module (__NR_SYSCALL_BASE+379) +#endif + +#if !defined(__NR_sched_setattr) +#define __NR_sched_setattr (__NR_SYSCALL_BASE+380) +#endif + +#if !defined(__NR_sched_getattr) +#define __NR_sched_getattr (__NR_SYSCALL_BASE+381) +#endif + +#if !defined(__NR_renameat2) +#define __NR_renameat2 (__NR_SYSCALL_BASE+382) +#endif + +#if !defined(__NR_seccomp) +#define __NR_seccomp (__NR_SYSCALL_BASE+383) +#endif + +#if !defined(__NR_getrandom) +#define __NR_getrandom (__NR_SYSCALL_BASE+384) +#endif + +#if !defined(__NR_memfd_create) +#define __NR_memfd_create (__NR_SYSCALL_BASE+385) +#endif + +// ARM private syscalls. +#if !defined(__ARM_NR_BASE) +#define __ARM_NR_BASE (__NR_SYSCALL_BASE + 0xF0000) +#endif + +#if !defined(__ARM_NR_breakpoint) +#define __ARM_NR_breakpoint (__ARM_NR_BASE+1) +#endif + +#if !defined(__ARM_NR_cacheflush) +#define __ARM_NR_cacheflush (__ARM_NR_BASE+2) +#endif + +#if !defined(__ARM_NR_usr26) +#define __ARM_NR_usr26 (__ARM_NR_BASE+3) +#endif + +#if !defined(__ARM_NR_usr32) +#define __ARM_NR_usr32 (__ARM_NR_BASE+4) +#endif + +#if !defined(__ARM_NR_set_tls) +#define __ARM_NR_set_tls (__ARM_NR_BASE+5) +#endif + +// ARM kernel private syscall. +#if !defined(__ARM_NR_cmpxchg) +#define __ARM_NR_cmpxchg (__ARM_NR_BASE+0x00fff0) +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_ARM_LINUX_SYSCALLS_H_ diff --git a/sandbox/linux/system_headers/arm_linux_ucontext.h b/sandbox/linux/system_headers/arm_linux_ucontext.h new file mode 100644 index 0000000000..0eb723a236 --- /dev/null +++ b/sandbox/linux/system_headers/arm_linux_ucontext.h @@ -0,0 +1,67 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_ARM_LINUX_UCONTEXT_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_ARM_LINUX_UCONTEXT_H_ + +#if !defined(__BIONIC_HAVE_UCONTEXT_T) +#if !defined(__native_client_nonsfi__) +#include <asm/sigcontext.h> +#else +// In PNaCl toolchain, sigcontext and stack_t is not defined. So here declare +// them. +struct sigcontext { + unsigned long trap_no; + unsigned long error_code; + unsigned long oldmask; + unsigned long arm_r0; + unsigned long arm_r1; + unsigned long arm_r2; + unsigned long arm_r3; + unsigned long arm_r4; + unsigned long arm_r5; + unsigned long arm_r6; + unsigned long arm_r7; + unsigned long arm_r8; + unsigned long arm_r9; + unsigned long arm_r10; + unsigned long arm_fp; + unsigned long arm_ip; + unsigned long arm_sp; + unsigned long arm_lr; + unsigned long arm_pc; + unsigned long arm_cpsr; + unsigned long fault_address; +}; + +typedef struct sigaltstack { + void* ss_sp; + int ss_flags; + size_t ss_size; +} stack_t; + +#endif + +// We also need greg_t for the sandbox, include it in this header as well. +typedef unsigned long greg_t; + +// typedef unsigned long sigset_t; +typedef struct ucontext { + unsigned long uc_flags; + struct ucontext* uc_link; + stack_t uc_stack; + struct sigcontext uc_mcontext; + sigset_t uc_sigmask; + /* Allow for uc_sigmask growth. Glibc uses a 1024-bit sigset_t. */ + int __not_used[32 - (sizeof(sigset_t) / sizeof(int))]; + /* Last for extensibility. Eight byte aligned because some + coprocessors require eight byte alignment. */ + unsigned long uc_regspace[128] __attribute__((__aligned__(8))); +} ucontext_t; + +#else +#include <sys/ucontext.h> +#endif // __BIONIC_HAVE_UCONTEXT_T + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_ARM_LINUX_UCONTEXT_H_ diff --git a/sandbox/linux/system_headers/capability.h b/sandbox/linux/system_headers/capability.h new file mode 100644 index 0000000000..f91fcf78ac --- /dev/null +++ b/sandbox/linux/system_headers/capability.h @@ -0,0 +1,42 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_CAPABILITY_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_CAPABILITY_H_ + +#include <stdint.h> + +// The following macros are taken from linux/capability.h. +// We only support capability version 3, which was introduced in Linux 2.6.26. +#ifndef _LINUX_CAPABILITY_VERSION_3 +#define _LINUX_CAPABILITY_VERSION_3 0x20080522 +#endif +#ifndef _LINUX_CAPABILITY_U32S_3 +#define _LINUX_CAPABILITY_U32S_3 2 +#endif +#ifndef CAP_TO_INDEX +#define CAP_TO_INDEX(x) ((x) >> 5) // 1 << 5 == bits in __u32 +#endif +#ifndef CAP_TO_MASK +#define CAP_TO_MASK(x) (1 << ((x) & 31)) // mask for indexed __u32 +#endif +#ifndef CAP_SYS_CHROOT +#define CAP_SYS_CHROOT 18 +#endif +#ifndef CAP_SYS_ADMIN +#define CAP_SYS_ADMIN 21 +#endif + +struct cap_hdr { + uint32_t version; + int pid; +}; + +struct cap_data { + uint32_t effective; + uint32_t permitted; + uint32_t inheritable; +}; + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_CAPABILITY_H_ diff --git a/sandbox/linux/system_headers/i386_linux_ucontext.h b/sandbox/linux/system_headers/i386_linux_ucontext.h new file mode 100644 index 0000000000..61d9f7a9b8 --- /dev/null +++ b/sandbox/linux/system_headers/i386_linux_ucontext.h @@ -0,0 +1,93 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_ANDROID_I386_UCONTEXT_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_ANDROID_I386_UCONTEXT_H_ + +// We do something compatible with glibc. Hopefully, at some point Android will +// provide that for us, and __BIONIC_HAVE_UCONTEXT_T should be defined. +// This is mostly copied from breakpad (common/android/include/sys/ucontext.h), +// except we do use sigset_t for uc_sigmask instead of a custom type. + +#if !defined(__BIONIC_HAVE_UCONTEXT_T) +#if !defined(__native_client_nonsfi__) +#include <asm/sigcontext.h> +#else +// In PNaCl toolchain, sigcontext is not defined. So here declare it. +typedef struct sigaltstack { + void* ss_sp; + int ss_flags; + size_t ss_size; +} stack_t; +#endif + +/* 80-bit floating-point register */ +struct _libc_fpreg { + unsigned short significand[4]; + unsigned short exponent; +}; + +/* Simple floating-point state, see FNSTENV instruction */ +struct _libc_fpstate { + unsigned long cw; + unsigned long sw; + unsigned long tag; + unsigned long ipoff; + unsigned long cssel; + unsigned long dataoff; + unsigned long datasel; + struct _libc_fpreg _st[8]; + unsigned long status; +}; + +typedef uint32_t greg_t; + +typedef struct { + uint32_t gregs[19]; + struct _libc_fpstate* fpregs; + uint32_t oldmask; + uint32_t cr2; +} mcontext_t; + +enum { + REG_GS = 0, + REG_FS, + REG_ES, + REG_DS, + REG_EDI, + REG_ESI, + REG_EBP, + REG_ESP, + REG_EBX, + REG_EDX, + REG_ECX, + REG_EAX, + REG_TRAPNO, + REG_ERR, + REG_EIP, + REG_CS, + REG_EFL, + REG_UESP, + REG_SS, +}; + +typedef struct ucontext { + uint32_t uc_flags; + struct ucontext* uc_link; + stack_t uc_stack; + mcontext_t uc_mcontext; + // Android and PNaCl toolchain's sigset_t has only 32 bits, though Linux + // ABI requires 64 bits. + union { + sigset_t uc_sigmask; + uint32_t kernel_sigmask[2]; + }; + struct _libc_fpstate __fpregs_mem; +} ucontext_t; + +#else +#include <sys/ucontext.h> +#endif // __BIONIC_HAVE_UCONTEXT_T + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_ANDROID_I386_UCONTEXT_H_ diff --git a/sandbox/linux/system_headers/linux_filter.h b/sandbox/linux/system_headers/linux_filter.h new file mode 100644 index 0000000000..b23b6eb0c1 --- /dev/null +++ b/sandbox/linux/system_headers/linux_filter.h @@ -0,0 +1,140 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_FILTER_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_FILTER_H_ + +#include <stdint.h> + +// The following structs and macros are taken from linux/filter.h, +// as some toolchain does not expose them. +struct sock_filter { + uint16_t code; + uint8_t jt; + uint8_t jf; + uint32_t k; +}; + +struct sock_fprog { + uint16_t len; + struct sock_filter *filter; +}; + +#ifndef BPF_CLASS +#define BPF_CLASS(code) ((code) & 0x07) +#endif + +#ifndef BPF_LD +#define BPF_LD 0x00 +#endif + +#ifndef BPF_ALU +#define BPF_ALU 0x04 +#endif + +#ifndef BPF_JMP +#define BPF_JMP 0x05 +#endif + +#ifndef BPF_RET +#define BPF_RET 0x06 +#endif + +#ifndef BPF_SIZE +#define BPF_SIZE(code) ((code) & 0x18) +#endif + +#ifndef BPF_W +#define BPF_W 0x00 +#endif + +#ifndef BPF_MODE +#define BPF_MODE(code) ((code) & 0xe0) +#endif + +#ifndef BPF_ABS +#define BPF_ABS 0x20 +#endif + +#ifndef BPF_OP +#define BPF_OP(code) ((code) & 0xf0) +#endif + +#ifndef BPF_ADD +#define BPF_ADD 0x00 +#endif + +#ifndef BPF_SUB +#define BPF_SUB 0x10 +#endif + +#ifndef BPF_MUL +#define BPF_MUL 0x20 +#endif + +#ifndef BPF_DIV +#define BPF_DIV 0x30 +#endif + +#ifndef BPF_OR +#define BPF_OR 0x40 +#endif + +#ifndef BPF_AND +#define BPF_AND 0x50 +#endif + +#ifndef BPF_LSH +#define BPF_LSH 0x60 +#endif + +#ifndef BPF_RSH +#define BPF_RSH 0x70 +#endif + +#ifndef BPF_NEG +#define BPF_NEG 0x80 +#endif + +#ifndef BPF_MOD +#define BPF_MOD 0x90 +#endif + +#ifndef BPF_XOR +#define BPF_XOR 0xA0 +#endif + +#ifndef BPF_JA +#define BPF_JA 0x00 +#endif + +#ifndef BPF_JEQ +#define BPF_JEQ 0x10 +#endif + +#ifndef BPF_JGT +#define BPF_JGT 0x20 +#endif + +#ifndef BPF_JGE +#define BPF_JGE 0x30 +#endif + +#ifndef BPF_JSET +#define BPF_JSET 0x40 +#endif + +#ifndef BPF_SRC +#define BPF_SRC(code) ((code) & 0x08) +#endif + +#ifndef BPF_K +#define BPF_K 0x00 +#endif + +#ifndef BPF_MAXINSNS +#define BPF_MAXINSNS 4096 +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_FILTER_H_ diff --git a/sandbox/linux/system_headers/linux_futex.h b/sandbox/linux/system_headers/linux_futex.h new file mode 100644 index 0000000000..4e28403336 --- /dev/null +++ b/sandbox/linux/system_headers/linux_futex.h @@ -0,0 +1,84 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_FUTEX_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_FUTEX_H_ + +#if !defined(__native_client_nonsfi__) +#include <linux/futex.h> +#endif // !defined(__native_client_nonsfi__) + +#if !defined(FUTEX_WAIT) +#define FUTEX_WAIT 0 +#endif + +#if !defined(FUTEX_WAKE) +#define FUTEX_WAKE 1 +#endif + +#if !defined(FUTEX_FD) +#define FUTEX_FD 2 +#endif + +#if !defined(FUTEX_REQUEUE) +#define FUTEX_REQUEUE 3 +#endif + +#if !defined(FUTEX_CMP_REQUEUE) +#define FUTEX_CMP_REQUEUE 4 +#endif + +#if !defined(FUTEX_WAKE_OP) +#define FUTEX_WAKE_OP 5 +#endif + +#if !defined(FUTEX_LOCK_PI) +#define FUTEX_LOCK_PI 6 +#endif + +#if !defined(FUTEX_UNLOCK_PI) +#define FUTEX_UNLOCK_PI 7 +#endif + +#if !defined(FUTEX_TRYLOCK_PI) +#define FUTEX_TRYLOCK_PI 8 +#endif + +#if !defined(FUTEX_WAIT_BITSET) +#define FUTEX_WAIT_BITSET 9 +#endif + +#if !defined(FUTEX_WAKE_BITSET) +#define FUTEX_WAKE_BITSET 10 +#endif + +#if !defined(FUTEX_WAIT_REQUEUE_PI) +#define FUTEX_WAIT_REQUEUE_PI 11 +#endif + +#if !defined(FUTEX_CMP_REQUEUE_PI) +#define FUTEX_CMP_REQUEUE_PI 12 +#endif + +#if !defined(FUTEX_PRIVATE_FLAG) +#define FUTEX_PRIVATE_FLAG 128 +#endif + +#if !defined FUTEX_CLOCK_REALTIME +#define FUTEX_CLOCK_REALTIME 256 +#endif + +#if !defined(FUTEX_CMD_MASK) +#define FUTEX_CMD_MASK ~(FUTEX_PRIVATE_FLAG | FUTEX_CLOCK_REALTIME) +#endif + +#if !defined(FUTEX_CMP_REQUEUE_PI_PRIVATE) +#define FUTEX_CMP_REQUEUE_PI_PRIVATE (FUTEX_CMP_REQUEUE_PI | FUTEX_PRIVATE_FLAG) +#endif + +#if !defined(FUTEX_UNLOCK_PI_PRIVATE) +#define FUTEX_UNLOCK_PI_PRIVATE (FUTEX_UNLOCK_PI | FUTEX_PRIVATE_FLAG) +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_FUTEX_H_ diff --git a/sandbox/linux/system_headers/linux_seccomp.h b/sandbox/linux/system_headers/linux_seccomp.h new file mode 100644 index 0000000000..3deb3d2253 --- /dev/null +++ b/sandbox/linux/system_headers/linux_seccomp.h @@ -0,0 +1,107 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SECCOMP_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SECCOMP_H_ + +// The Seccomp2 kernel ABI is not part of older versions of glibc. +// As we can't break compilation with these versions of the library, +// we explicitly define all missing symbols. +// If we ever decide that we can now rely on system headers, the following +// include files should be enabled: +// #include <linux/audit.h> +// #include <linux/seccomp.h> + +// For audit.h +#ifndef EM_ARM +#define EM_ARM 40 +#endif +#ifndef EM_386 +#define EM_386 3 +#endif +#ifndef EM_X86_64 +#define EM_X86_64 62 +#endif +#ifndef EM_MIPS +#define EM_MIPS 8 +#endif +#ifndef EM_AARCH64 +#define EM_AARCH64 183 +#endif + +#ifndef __AUDIT_ARCH_64BIT +#define __AUDIT_ARCH_64BIT 0x80000000 +#endif +#ifndef __AUDIT_ARCH_LE +#define __AUDIT_ARCH_LE 0x40000000 +#endif +#ifndef AUDIT_ARCH_ARM +#define AUDIT_ARCH_ARM (EM_ARM|__AUDIT_ARCH_LE) +#endif +#ifndef AUDIT_ARCH_I386 +#define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE) +#endif +#ifndef AUDIT_ARCH_X86_64 +#define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE) +#endif +#ifndef AUDIT_ARCH_MIPSEL +#define AUDIT_ARCH_MIPSEL (EM_MIPS|__AUDIT_ARCH_LE) +#endif +#ifndef AUDIT_ARCH_AARCH64 +#define AUDIT_ARCH_AARCH64 (EM_AARCH64 | __AUDIT_ARCH_64BIT | __AUDIT_ARCH_LE) +#endif + +// For prctl.h +#ifndef PR_SET_SECCOMP +#define PR_SET_SECCOMP 22 +#define PR_GET_SECCOMP 21 +#endif +#ifndef PR_SET_NO_NEW_PRIVS +#define PR_SET_NO_NEW_PRIVS 38 +#define PR_GET_NO_NEW_PRIVS 39 +#endif +#ifndef IPC_64 +#define IPC_64 0x0100 +#endif + +// In order to build will older tool chains, we currently have to avoid +// including <linux/seccomp.h>. Until that can be fixed (if ever). Rely on +// our own definitions of the seccomp kernel ABI. +#ifndef SECCOMP_MODE_FILTER +#define SECCOMP_MODE_DISABLED 0 +#define SECCOMP_MODE_STRICT 1 +#define SECCOMP_MODE_FILTER 2 // User user-supplied filter +#endif + +#ifndef SECCOMP_SET_MODE_STRICT +#define SECCOMP_SET_MODE_STRICT 0 +#endif +#ifndef SECCOMP_SET_MODE_FILTER +#define SECCOMP_SET_MODE_FILTER 1 +#endif +#ifndef SECCOMP_FILTER_FLAG_TSYNC +#define SECCOMP_FILTER_FLAG_TSYNC 1 +#endif + +#ifndef SECCOMP_RET_KILL +// Return values supported for BPF filter programs. Please note that the +// "illegal" SECCOMP_RET_INVALID is not supported by the kernel, should only +// ever be used internally, and would result in the kernel killing our process. +#define SECCOMP_RET_KILL 0x00000000U // Kill the task immediately +#define SECCOMP_RET_INVALID 0x00010000U // Illegal return value +#define SECCOMP_RET_TRAP 0x00030000U // Disallow and force a SIGSYS +#define SECCOMP_RET_ERRNO 0x00050000U // Returns an errno +#define SECCOMP_RET_TRACE 0x7ff00000U // Pass to a tracer or disallow +#define SECCOMP_RET_ALLOW 0x7fff0000U // Allow +#define SECCOMP_RET_ACTION 0xffff0000U // Masks for the return value +#define SECCOMP_RET_DATA 0x0000ffffU // sections +#else +#define SECCOMP_RET_INVALID 0x00010000U // Illegal return value +#endif + +#ifndef SYS_SECCOMP +#define SYS_SECCOMP 1 +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SECCOMP_H_ diff --git a/sandbox/linux/system_headers/linux_signal.h b/sandbox/linux/system_headers/linux_signal.h new file mode 100644 index 0000000000..5db7fc5ea1 --- /dev/null +++ b/sandbox/linux/system_headers/linux_signal.h @@ -0,0 +1,73 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SIGNAL_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SIGNAL_H_ + +// NOTE: On some toolchains, signal related ABI is incompatible with Linux's +// (not undefined, but defined different values and in different memory +// layouts). So, fill the gap here. + +#if defined(__native_client_nonsfi__) +#if !defined(__i386__) && !defined(__arm__) +#error "Unsupported platform" +#endif + +#include <signal.h> + +#define LINUX_SIGBUS 7 // 10 in PNaCl toolchain. +#define LINUX_SIGSEGV 11 // 11 in PNaCl toolchain. Defined for consistency. +#define LINUX_SIGCHLD 17 // 20 in PNaCl toolchain. +#define LINUX_SIGSYS 31 // 12 in PNaCl toolchain. + +#define LINUX_SIG_BLOCK 0 // 1 in PNaCl toolchain. +#define LINUX_SIG_UNBLOCK 1 // 2 in PNaCl toolchain. + +#define LINUX_SA_SIGINFO 4 // 2 in PNaCl toolchain. +#define LINUX_SA_NODEFER 0x40000000 // Undefined in PNaCl toolchain. +#define LINUX_SA_RESTART 0x10000000 // Undefined in PNaCl toolchain. + +#define LINUX_SIG_DFL 0 // In PNaCl toolchain, unneeded cast is applied. + +struct LinuxSigInfo { + int si_signo; + int si_errno; + int si_code; + + // Extra data is followed by the |si_code|. The length depends on the + // signal number. + char _sifields[1]; +}; + +#include "sandbox/linux/system_headers/linux_ucontext.h" + +#else // !defined(__native_client_nonsfi__) + +// Just alias the toolchain's value. +#include <signal.h> + +#define LINUX_SIGBUS SIGBUS +#define LINUX_SIGSEGV SIGSEGV +#define LINUX_SIGCHLD SIGCHLD +#define LINUX_SIGSYS SIGSYS + +#define LINUX_SIG_BLOCK SIG_BLOCK +#define LINUX_SIG_UNBLOCK SIG_UNBLOCK + +#define LINUX_SA_SIGINFO SA_SIGINFO +#define LINUX_SA_NODEFER SA_NODEFER +#define LINUX_SA_RESTART SA_RESTART + +#define LINUX_SIG_DFL SIG_DFL + +typedef siginfo_t LinuxSigInfo; + +#if defined(__ANDROID__) +// Android's signal.h doesn't define ucontext etc. +#include "sandbox/linux/system_headers/linux_ucontext.h" +#endif // defined(__ANDROID__) + +#endif // !defined(__native_client_nonsfi__) + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SIGNAL_H_ diff --git a/sandbox/linux/system_headers/linux_syscalls.h b/sandbox/linux/system_headers/linux_syscalls.h new file mode 100644 index 0000000000..2b441e47ea --- /dev/null +++ b/sandbox/linux/system_headers/linux_syscalls.h @@ -0,0 +1,37 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// This header will be kept up to date so that we can compile system-call +// policies even when system headers are old. +// System call numbers are accessible through __NR_syscall_name. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SYSCALLS_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SYSCALLS_H_ + +#if defined(__x86_64__) +#include "sandbox/linux/system_headers/x86_64_linux_syscalls.h" +#endif + +#if defined(__i386__) +#include "sandbox/linux/system_headers/x86_32_linux_syscalls.h" +#endif + +#if defined(__arm__) && defined(__ARM_EABI__) +#include "sandbox/linux/system_headers/arm_linux_syscalls.h" +#endif + +#if defined(__mips__) && (_MIPS_SIM == _ABIO32) +#include "sandbox/linux/system_headers/mips_linux_syscalls.h" +#endif + +#if defined(__mips__) && (_MIPS_SIM == _ABI64) +#include "sandbox/linux/system_headers/mips64_linux_syscalls.h" +#endif + +#if defined(__aarch64__) +#include "sandbox/linux/system_headers/arm64_linux_syscalls.h" +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_SYSCALLS_H_ + diff --git a/sandbox/linux/system_headers/linux_time.h b/sandbox/linux/system_headers/linux_time.h new file mode 100644 index 0000000000..e6c8112b86 --- /dev/null +++ b/sandbox/linux/system_headers/linux_time.h @@ -0,0 +1,18 @@ +// Copyright 2015 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_TIME_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_TIME_H_ + +#include <time.h> + +#if !defined(CLOCK_REALTIME_COARSE) +#define CLOCK_REALTIME_COARSE 5 +#endif + +#if !defined(CLOCK_MONOTONIC_COARSE) +#define CLOCK_MONOTONIC_COARSE 6 +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_TIME_H_ diff --git a/sandbox/linux/system_headers/linux_ucontext.h b/sandbox/linux/system_headers/linux_ucontext.h new file mode 100644 index 0000000000..ea4d8a6c1f --- /dev/null +++ b/sandbox/linux/system_headers/linux_ucontext.h @@ -0,0 +1,28 @@ +// Copyright (c) 2013 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_UCONTEXT_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_UCONTEXT_H_ + +#if defined(__ANDROID__) || defined(__native_client_nonsfi__) + +#if defined(__arm__) +#include "sandbox/linux/system_headers/arm_linux_ucontext.h" +#elif defined(__i386__) +#include "sandbox/linux/system_headers/i386_linux_ucontext.h" +#elif defined(__x86_64__) +#include "sandbox/linux/system_headers/x86_64_linux_ucontext.h" +#elif defined(__mips__) +#include "sandbox/linux/system_headers/mips_linux_ucontext.h" +#elif defined(__aarch64__) +#include "sandbox/linux/system_headers/arm64_linux_ucontext.h" +#else +#error "No support for your architecture in Android or PNaCl header" +#endif + +#else // defined(__ANDROID__) || defined(__native_client_nonsfi__) +#error "The header file included on non Android and non PNaCl." +#endif // defined(__ANDROID__) || defined(__native_client_nonsfi__) + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_LINUX_UCONTEXT_H_ diff --git a/sandbox/linux/system_headers/mips64_linux_syscalls.h b/sandbox/linux/system_headers/mips64_linux_syscalls.h new file mode 100644 index 0000000000..d003124284 --- /dev/null +++ b/sandbox/linux/system_headers/mips64_linux_syscalls.h @@ -0,0 +1,1266 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// Generated from the Linux kernel's calls.S. +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_MIPS64_LINUX_SYSCALLS_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_MIPS64_LINUX_SYSCALLS_H_ + +#if !defined(__mips__) || (_MIPS_SIM != _ABI64) +#error "Including header on wrong architecture" +#endif + +// __NR_Linux, is defined in <asm/unistd.h>. +#include <asm/unistd.h> + +#if !defined(__NR_read) +#define __NR_read (__NR_Linux + 0) +#endif + +#if !defined(__NR_write) +#define __NR_write (__NR_Linux + 1) +#endif + +#if !defined(__NR_open) +#define __NR_open (__NR_Linux + 2) +#endif + +#if !defined(__NR_close) +#define __NR_close (__NR_Linux + 3) +#endif + +#if !defined(__NR_stat) +#define __NR_stat (__NR_Linux + 4) +#endif + +#if !defined(__NR_fstat) +#define __NR_fstat (__NR_Linux + 5) +#endif + +#if !defined(__NR_lstat) +#define __NR_lstat (__NR_Linux + 6) +#endif + +#if !defined(__NR_poll) +#define __NR_poll (__NR_Linux + 7) +#endif + +#if !defined(__NR_lseek) +#define __NR_lseek (__NR_Linux + 8) +#endif + +#if !defined(__NR_mmap) +#define __NR_mmap (__NR_Linux + 9) +#endif + +#if !defined(__NR_mprotect) +#define __NR_mprotect (__NR_Linux + 10) +#endif + +#if !defined(__NR_munmap) +#define __NR_munmap (__NR_Linux + 11) +#endif + +#if !defined(__NR_brk) +#define __NR_brk (__NR_Linux + 12) +#endif + +#if !defined(__NR_rt_sigaction) +#define __NR_rt_sigaction (__NR_Linux + 13) +#endif + +#if !defined(__NR_rt_sigprocmask) +#define __NR_rt_sigprocmask (__NR_Linux + 14) +#endif + +#if !defined(__NR_ioctl) +#define __NR_ioctl (__NR_Linux + 15) +#endif + +#if !defined(__NR_pread64) +#define __NR_pread64 (__NR_Linux + 16) +#endif + +#if !defined(__NR_pwrite64) +#define __NR_pwrite64 (__NR_Linux + 17) +#endif + +#if !defined(__NR_readv) +#define __NR_readv (__NR_Linux + 18) +#endif + +#if !defined(__NR_writev) +#define __NR_writev (__NR_Linux + 19) +#endif + +#if !defined(__NR_access) +#define __NR_access (__NR_Linux + 20) +#endif + +#if !defined(__NR_pipe) +#define __NR_pipe (__NR_Linux + 21) +#endif + +#if !defined(__NR__newselect) +#define __NR__newselect (__NR_Linux + 22) +#endif + +#if !defined(__NR_sched_yield) +#define __NR_sched_yield (__NR_Linux + 23) +#endif + +#if !defined(__NR_mremap) +#define __NR_mremap (__NR_Linux + 24) +#endif + +#if !defined(__NR_msync) +#define __NR_msync (__NR_Linux + 25) +#endif + +#if !defined(__NR_mincore) +#define __NR_mincore (__NR_Linux + 26) +#endif + +#if !defined(__NR_madvise) +#define __NR_madvise (__NR_Linux + 27) +#endif + +#if !defined(__NR_shmget) +#define __NR_shmget (__NR_Linux + 28) +#endif + +#if !defined(__NR_shmat) +#define __NR_shmat (__NR_Linux + 29) +#endif + +#if !defined(__NR_shmctl) +#define __NR_shmctl (__NR_Linux + 30) +#endif + +#if !defined(__NR_dup) +#define __NR_dup (__NR_Linux + 31) +#endif + +#if !defined(__NR_dup2) +#define __NR_dup2 (__NR_Linux + 32) +#endif + +#if !defined(__NR_pause) +#define __NR_pause (__NR_Linux + 33) +#endif + +#if !defined(__NR_nanosleep) +#define __NR_nanosleep (__NR_Linux + 34) +#endif + +#if !defined(__NR_getitimer) +#define __NR_getitimer (__NR_Linux + 35) +#endif + +#if !defined(__NR_setitimer) +#define __NR_setitimer (__NR_Linux + 36) +#endif + +#if !defined(__NR_alarm) +#define __NR_alarm (__NR_Linux + 37) +#endif + +#if !defined(__NR_getpid) +#define __NR_getpid (__NR_Linux + 38) +#endif + +#if !defined(__NR_sendfile) +#define __NR_sendfile (__NR_Linux + 39) +#endif + +#if !defined(__NR_socket) +#define __NR_socket (__NR_Linux + 40) +#endif + +#if !defined(__NR_connect) +#define __NR_connect (__NR_Linux + 41) +#endif + +#if !defined(__NR_accept) +#define __NR_accept (__NR_Linux + 42) +#endif + +#if !defined(__NR_sendto) +#define __NR_sendto (__NR_Linux + 43) +#endif + +#if !defined(__NR_recvfrom) +#define __NR_recvfrom (__NR_Linux + 44) +#endif + +#if !defined(__NR_sendmsg) +#define __NR_sendmsg (__NR_Linux + 45) +#endif + +#if !defined(__NR_recvmsg) +#define __NR_recvmsg (__NR_Linux + 46) +#endif + +#if !defined(__NR_shutdown) +#define __NR_shutdown (__NR_Linux + 47) +#endif + +#if !defined(__NR_bind) +#define __NR_bind (__NR_Linux + 48) +#endif + +#if !defined(__NR_listen) +#define __NR_listen (__NR_Linux + 49) +#endif + +#if !defined(__NR_getsockname) +#define __NR_getsockname (__NR_Linux + 50) +#endif + +#if !defined(__NR_getpeername) +#define __NR_getpeername (__NR_Linux + 51) +#endif + +#if !defined(__NR_socketpair) +#define __NR_socketpair (__NR_Linux + 52) +#endif + +#if !defined(__NR_setsockopt) +#define __NR_setsockopt (__NR_Linux + 53) +#endif + +#if !defined(__NR_getsockopt) +#define __NR_getsockopt (__NR_Linux + 54) +#endif + +#if !defined(__NR_clone) +#define __NR_clone (__NR_Linux + 55) +#endif + +#if !defined(__NR_fork) +#define __NR_fork (__NR_Linux + 56) +#endif + +#if !defined(__NR_execve) +#define __NR_execve (__NR_Linux + 57) +#endif + +#if !defined(__NR_exit) +#define __NR_exit (__NR_Linux + 58) +#endif + +#if !defined(__NR_wait4) +#define __NR_wait4 (__NR_Linux + 59) +#endif + +#if !defined(__NR_kill) +#define __NR_kill (__NR_Linux + 60) +#endif + +#if !defined(__NR_uname) +#define __NR_uname (__NR_Linux + 61) +#endif + +#if !defined(__NR_semget) +#define __NR_semget (__NR_Linux + 62) +#endif + +#if !defined(__NR_semop) +#define __NR_semop (__NR_Linux + 63) +#endif + +#if !defined(__NR_semctl) +#define __NR_semctl (__NR_Linux + 64) +#endif + +#if !defined(__NR_shmdt) +#define __NR_shmdt (__NR_Linux + 65) +#endif + +#if !defined(__NR_msgget) +#define __NR_msgget (__NR_Linux + 66) +#endif + +#if !defined(__NR_msgsnd) +#define __NR_msgsnd (__NR_Linux + 67) +#endif + +#if !defined(__NR_msgrcv) +#define __NR_msgrcv (__NR_Linux + 68) +#endif + +#if !defined(__NR_msgctl) +#define __NR_msgctl (__NR_Linux + 69) +#endif + +#if !defined(__NR_fcntl) +#define __NR_fcntl (__NR_Linux + 70) +#endif + +#if !defined(__NR_flock) +#define __NR_flock (__NR_Linux + 71) +#endif + +#if !defined(__NR_fsync) +#define __NR_fsync (__NR_Linux + 72) +#endif + +#if !defined(__NR_fdatasync) +#define __NR_fdatasync (__NR_Linux + 73) +#endif + +#if !defined(__NR_truncate) +#define __NR_truncate (__NR_Linux + 74) +#endif + +#if !defined(__NR_ftruncate) +#define __NR_ftruncate (__NR_Linux + 75) +#endif + +#if !defined(__NR_getdents) +#define __NR_getdents (__NR_Linux + 76) +#endif + +#if !defined(__NR_getcwd) +#define __NR_getcwd (__NR_Linux + 77) +#endif + +#if !defined(__NR_chdir) +#define __NR_chdir (__NR_Linux + 78) +#endif + +#if !defined(__NR_fchdir) +#define __NR_fchdir (__NR_Linux + 79) +#endif + +#if !defined(__NR_rename) +#define __NR_rename (__NR_Linux + 80) +#endif + +#if !defined(__NR_mkdir) +#define __NR_mkdir (__NR_Linux + 81) +#endif + +#if !defined(__NR_rmdir) +#define __NR_rmdir (__NR_Linux + 82) +#endif + +#if !defined(__NR_creat) +#define __NR_creat (__NR_Linux + 83) +#endif + +#if !defined(__NR_link) +#define __NR_link (__NR_Linux + 84) +#endif + +#if !defined(__NR_unlink) +#define __NR_unlink (__NR_Linux + 85) +#endif + +#if !defined(__NR_symlink) +#define __NR_symlink (__NR_Linux + 86) +#endif + +#if !defined(__NR_readlink) +#define __NR_readlink (__NR_Linux + 87) +#endif + +#if !defined(__NR_chmod) +#define __NR_chmod (__NR_Linux + 88) +#endif + +#if !defined(__NR_fchmod) +#define __NR_fchmod (__NR_Linux + 89) +#endif + +#if !defined(__NR_chown) +#define __NR_chown (__NR_Linux + 90) +#endif + +#if !defined(__NR_fchown) +#define __NR_fchown (__NR_Linux + 91) +#endif + +#if !defined(__NR_lchown) +#define __NR_lchown (__NR_Linux + 92) +#endif + +#if !defined(__NR_umask) +#define __NR_umask (__NR_Linux + 93) +#endif + +#if !defined(__NR_gettimeofday) +#define __NR_gettimeofday (__NR_Linux + 94) +#endif + +#if !defined(__NR_getrlimit) +#define __NR_getrlimit (__NR_Linux + 95) +#endif + +#if !defined(__NR_getrusage) +#define __NR_getrusage (__NR_Linux + 96) +#endif + +#if !defined(__NR_sysinfo) +#define __NR_sysinfo (__NR_Linux + 97) +#endif + +#if !defined(__NR_times) +#define __NR_times (__NR_Linux + 98) +#endif + +#if !defined(__NR_ptrace) +#define __NR_ptrace (__NR_Linux + 99) +#endif + +#if !defined(__NR_getuid) +#define __NR_getuid (__NR_Linux + 100) +#endif + +#if !defined(__NR_syslog) +#define __NR_syslog (__NR_Linux + 101) +#endif + +#if !defined(__NR_getgid) +#define __NR_getgid (__NR_Linux + 102) +#endif + +#if !defined(__NR_setuid) +#define __NR_setuid (__NR_Linux + 103) +#endif + +#if !defined(__NR_setgid) +#define __NR_setgid (__NR_Linux + 104) +#endif + +#if !defined(__NR_geteuid) +#define __NR_geteuid (__NR_Linux + 105) +#endif + +#if !defined(__NR_getegid) +#define __NR_getegid (__NR_Linux + 106) +#endif + +#if !defined(__NR_setpgid) +#define __NR_setpgid (__NR_Linux + 107) +#endif + +#if !defined(__NR_getppid) +#define __NR_getppid (__NR_Linux + 108) +#endif + +#if !defined(__NR_getpgrp) +#define __NR_getpgrp (__NR_Linux + 109) +#endif + +#if !defined(__NR_setsid) +#define __NR_setsid (__NR_Linux + 110) +#endif + +#if !defined(__NR_setreuid) +#define __NR_setreuid (__NR_Linux + 111) +#endif + +#if !defined(__NR_setregid) +#define __NR_setregid (__NR_Linux + 112) +#endif + +#if !defined(__NR_getgroups) +#define __NR_getgroups (__NR_Linux + 113) +#endif + +#if !defined(__NR_setgroups) +#define __NR_setgroups (__NR_Linux + 114) +#endif + +#if !defined(__NR_setresuid) +#define __NR_setresuid (__NR_Linux + 115) +#endif + +#if !defined(__NR_getresuid) +#define __NR_getresuid (__NR_Linux + 116) +#endif + +#if !defined(__NR_setresgid) +#define __NR_setresgid (__NR_Linux + 117) +#endif + +#if !defined(__NR_getresgid) +#define __NR_getresgid (__NR_Linux + 118) +#endif + +#if !defined(__NR_getpgid) +#define __NR_getpgid (__NR_Linux + 119) +#endif + +#if !defined(__NR_setfsuid) +#define __NR_setfsuid (__NR_Linux + 120) +#endif + +#if !defined(__NR_setfsgid) +#define __NR_setfsgid (__NR_Linux + 121) +#endif + +#if !defined(__NR_getsid) +#define __NR_getsid (__NR_Linux + 122) +#endif + +#if !defined(__NR_capget) +#define __NR_capget (__NR_Linux + 123) +#endif + +#if !defined(__NR_capset) +#define __NR_capset (__NR_Linux + 124) +#endif + +#if !defined(__NR_rt_sigpending) +#define __NR_rt_sigpending (__NR_Linux + 125) +#endif + +#if !defined(__NR_rt_sigtimedwait) +#define __NR_rt_sigtimedwait (__NR_Linux + 126) +#endif + +#if !defined(__NR_rt_sigqueueinfo) +#define __NR_rt_sigqueueinfo (__NR_Linux + 127) +#endif + +#if !defined(__NR_rt_sigsuspend) +#define __NR_rt_sigsuspend (__NR_Linux + 128) +#endif + +#if !defined(__NR_sigaltstack) +#define __NR_sigaltstack (__NR_Linux + 129) +#endif + +#if !defined(__NR_utime) +#define __NR_utime (__NR_Linux + 130) +#endif + +#if !defined(__NR_mknod) +#define __NR_mknod (__NR_Linux + 131) +#endif + +#if !defined(__NR_personality) +#define __NR_personality (__NR_Linux + 132) +#endif + +#if !defined(__NR_ustat) +#define __NR_ustat (__NR_Linux + 133) +#endif + +#if !defined(__NR_statfs) +#define __NR_statfs (__NR_Linux + 134) +#endif + +#if !defined(__NR_fstatfs) +#define __NR_fstatfs (__NR_Linux + 135) +#endif + +#if !defined(__NR_sysfs) +#define __NR_sysfs (__NR_Linux + 136) +#endif + +#if !defined(__NR_getpriority) +#define __NR_getpriority (__NR_Linux + 137) +#endif + +#if !defined(__NR_setpriority) +#define __NR_setpriority (__NR_Linux + 138) +#endif + +#if !defined(__NR_sched_setparam) +#define __NR_sched_setparam (__NR_Linux + 139) +#endif + +#if !defined(__NR_sched_getparam) +#define __NR_sched_getparam (__NR_Linux + 140) +#endif + +#if !defined(__NR_sched_setscheduler) +#define __NR_sched_setscheduler (__NR_Linux + 141) +#endif + +#if !defined(__NR_sched_getscheduler) +#define __NR_sched_getscheduler (__NR_Linux + 142) +#endif + +#if !defined(__NR_sched_get_priority_max) +#define __NR_sched_get_priority_max (__NR_Linux + 143) +#endif + +#if !defined(__NR_sched_get_priority_min) +#define __NR_sched_get_priority_min (__NR_Linux + 144) +#endif + +#if !defined(__NR_sched_rr_get_interval) +#define __NR_sched_rr_get_interval (__NR_Linux + 145) +#endif + +#if !defined(__NR_mlock) +#define __NR_mlock (__NR_Linux + 146) +#endif + +#if !defined(__NR_munlock) +#define __NR_munlock (__NR_Linux + 147) +#endif + +#if !defined(__NR_mlockall) +#define __NR_mlockall (__NR_Linux + 148) +#endif + +#if !defined(__NR_munlockall) +#define __NR_munlockall (__NR_Linux + 149) +#endif + +#if !defined(__NR_vhangup) +#define __NR_vhangup (__NR_Linux + 150) +#endif + +#if !defined(__NR_pivot_root) +#define __NR_pivot_root (__NR_Linux + 151) +#endif + +#if !defined(__NR__sysctl) +#define __NR__sysctl (__NR_Linux + 152) +#endif + +#if !defined(__NR_prctl) +#define __NR_prctl (__NR_Linux + 153) +#endif + +#if !defined(__NR_adjtimex) +#define __NR_adjtimex (__NR_Linux + 154) +#endif + +#if !defined(__NR_setrlimit) +#define __NR_setrlimit (__NR_Linux + 155) +#endif + +#if !defined(__NR_chroot) +#define __NR_chroot (__NR_Linux + 156) +#endif + +#if !defined(__NR_sync) +#define __NR_sync (__NR_Linux + 157) +#endif + +#if !defined(__NR_acct) +#define __NR_acct (__NR_Linux + 158) +#endif + +#if !defined(__NR_settimeofday) +#define __NR_settimeofday (__NR_Linux + 159) +#endif + +#if !defined(__NR_mount) +#define __NR_mount (__NR_Linux + 160) +#endif + +#if !defined(__NR_umount2) +#define __NR_umount2 (__NR_Linux + 161) +#endif + +#if !defined(__NR_swapon) +#define __NR_swapon (__NR_Linux + 162) +#endif + +#if !defined(__NR_swapoff) +#define __NR_swapoff (__NR_Linux + 163) +#endif + +#if !defined(__NR_reboot) +#define __NR_reboot (__NR_Linux + 164) +#endif + +#if !defined(__NR_sethostname) +#define __NR_sethostname (__NR_Linux + 165) +#endif + +#if !defined(__NR_setdomainname) +#define __NR_setdomainname (__NR_Linux + 166) +#endif + +#if !defined(__NR_create_module) +#define __NR_create_module (__NR_Linux + 167) +#endif + +#if !defined(__NR_init_module) +#define __NR_init_module (__NR_Linux + 168) +#endif + +#if !defined(__NR_delete_module) +#define __NR_delete_module (__NR_Linux + 169) +#endif + +#if !defined(__NR_get_kernel_syms) +#define __NR_get_kernel_syms (__NR_Linux + 170) +#endif + +#if !defined(__NR_query_module) +#define __NR_query_module (__NR_Linux + 171) +#endif + +#if !defined(__NR_quotactl) +#define __NR_quotactl (__NR_Linux + 172) +#endif + +#if !defined(__NR_nfsservctl) +#define __NR_nfsservctl (__NR_Linux + 173) +#endif + +#if !defined(__NR_getpmsg) +#define __NR_getpmsg (__NR_Linux + 174) +#endif + +#if !defined(__NR_putpmsg) +#define __NR_putpmsg (__NR_Linux + 175) +#endif + +#if !defined(__NR_afs_syscall) +#define __NR_afs_syscall (__NR_Linux + 176) +#endif + +#if !defined(__NR_reserved177) +#define __NR_reserved177 (__NR_Linux + 177) +#endif + +#if !defined(__NR_gettid) +#define __NR_gettid (__NR_Linux + 178) +#endif + +#if !defined(__NR_readahead) +#define __NR_readahead (__NR_Linux + 179) +#endif + +#if !defined(__NR_setxattr) +#define __NR_setxattr (__NR_Linux + 180) +#endif + +#if !defined(__NR_lsetxattr) +#define __NR_lsetxattr (__NR_Linux + 181) +#endif + +#if !defined(__NR_fsetxattr) +#define __NR_fsetxattr (__NR_Linux + 182) +#endif + +#if !defined(__NR_getxattr) +#define __NR_getxattr (__NR_Linux + 183) +#endif + +#if !defined(__NR_lgetxattr) +#define __NR_lgetxattr (__NR_Linux + 184) +#endif + +#if !defined(__NR_fgetxattr) +#define __NR_fgetxattr (__NR_Linux + 185) +#endif + +#if !defined(__NR_listxattr) +#define __NR_listxattr (__NR_Linux + 186) +#endif + +#if !defined(__NR_llistxattr) +#define __NR_llistxattr (__NR_Linux + 187) +#endif + +#if !defined(__NR_flistxattr) +#define __NR_flistxattr (__NR_Linux + 188) +#endif + +#if !defined(__NR_removexattr) +#define __NR_removexattr (__NR_Linux + 189) +#endif + +#if !defined(__NR_lremovexattr) +#define __NR_lremovexattr (__NR_Linux + 190) +#endif + +#if !defined(__NR_fremovexattr) +#define __NR_fremovexattr (__NR_Linux + 191) +#endif + +#if !defined(__NR_tkill) +#define __NR_tkill (__NR_Linux + 192) +#endif + +#if !defined(__NR_reserved193) +#define __NR_reserved193 (__NR_Linux + 193) +#endif + +#if !defined(__NR_futex) +#define __NR_futex (__NR_Linux + 194) +#endif + +#if !defined(__NR_sched_setaffinity) +#define __NR_sched_setaffinity (__NR_Linux + 195) +#endif + +#if !defined(__NR_sched_getaffinity) +#define __NR_sched_getaffinity (__NR_Linux + 196) +#endif + +#if !defined(__NR_cacheflush) +#define __NR_cacheflush (__NR_Linux + 197) +#endif + +#if !defined(__NR_cachectl) +#define __NR_cachectl (__NR_Linux + 198) +#endif + +#if !defined(__NR_sysmips) +#define __NR_sysmips (__NR_Linux + 199) +#endif + +#if !defined(__NR_io_setup) +#define __NR_io_setup (__NR_Linux + 200) +#endif + +#if !defined(__NR_io_destroy) +#define __NR_io_destroy (__NR_Linux + 201) +#endif + +#if !defined(__NR_io_getevents) +#define __NR_io_getevents (__NR_Linux + 202) +#endif + +#if !defined(__NR_io_submit) +#define __NR_io_submit (__NR_Linux + 203) +#endif + +#if !defined(__NR_io_cancel) +#define __NR_io_cancel (__NR_Linux + 204) +#endif + +#if !defined(__NR_exit_group) +#define __NR_exit_group (__NR_Linux + 205) +#endif + +#if !defined(__NR_lookup_dcookie) +#define __NR_lookup_dcookie (__NR_Linux + 206) +#endif + +#if !defined(__NR_epoll_create) +#define __NR_epoll_create (__NR_Linux + 207) +#endif + +#if !defined(__NR_epoll_ctl) +#define __NR_epoll_ctl (__NR_Linux + 208) +#endif + +#if !defined(__NR_epoll_wait) +#define __NR_epoll_wait (__NR_Linux + 209) +#endif + +#if !defined(__NR_remap_file_pages) +#define __NR_remap_file_pages (__NR_Linux + 210) +#endif + +#if !defined(__NR_rt_sigreturn) +#define __NR_rt_sigreturn (__NR_Linux + 211) +#endif + +#if !defined(__NR_set_tid_address) +#define __NR_set_tid_address (__NR_Linux + 212) +#endif + +#if !defined(__NR_restart_syscall) +#define __NR_restart_syscall (__NR_Linux + 213) +#endif + +#if !defined(__NR_semtimedop) +#define __NR_semtimedop (__NR_Linux + 214) +#endif + +#if !defined(__NR_fadvise64) +#define __NR_fadvise64 (__NR_Linux + 215) +#endif + +#if !defined(__NR_timer_create) +#define __NR_timer_create (__NR_Linux + 216) +#endif + +#if !defined(__NR_timer_settime) +#define __NR_timer_settime (__NR_Linux + 217) +#endif + +#if !defined(__NR_timer_gettime) +#define __NR_timer_gettime (__NR_Linux + 218) +#endif + +#if !defined(__NR_timer_getoverrun) +#define __NR_timer_getoverrun (__NR_Linux + 219) +#endif + +#if !defined(__NR_timer_delete) +#define __NR_timer_delete (__NR_Linux + 220) +#endif + +#if !defined(__NR_clock_settime) +#define __NR_clock_settime (__NR_Linux + 221) +#endif + +#if !defined(__NR_clock_gettime) +#define __NR_clock_gettime (__NR_Linux + 222) +#endif + +#if !defined(__NR_clock_getres) +#define __NR_clock_getres (__NR_Linux + 223) +#endif + +#if !defined(__NR_clock_nanosleep) +#define __NR_clock_nanosleep (__NR_Linux + 224) +#endif + +#if !defined(__NR_tgkill) +#define __NR_tgkill (__NR_Linux + 225) +#endif + +#if !defined(__NR_utimes) +#define __NR_utimes (__NR_Linux + 226) +#endif + +#if !defined(__NR_mbind) +#define __NR_mbind (__NR_Linux + 227) +#endif + +#if !defined(__NR_get_mempolicy) +#define __NR_get_mempolicy (__NR_Linux + 228) +#endif + +#if !defined(__NR_set_mempolicy) +#define __NR_set_mempolicy (__NR_Linux + 229) +#endif + +#if !defined(__NR_mq_open) +#define __NR_mq_open (__NR_Linux + 230) +#endif + +#if !defined(__NR_mq_unlink) +#define __NR_mq_unlink (__NR_Linux + 231) +#endif + +#if !defined(__NR_mq_timedsend) +#define __NR_mq_timedsend (__NR_Linux + 232) +#endif + +#if !defined(__NR_mq_timedreceive) +#define __NR_mq_timedreceive (__NR_Linux + 233) +#endif + +#if !defined(__NR_mq_notify) +#define __NR_mq_notify (__NR_Linux + 234) +#endif + +#if !defined(__NR_mq_getsetattr) +#define __NR_mq_getsetattr (__NR_Linux + 235) +#endif + +#if !defined(__NR_vserver) +#define __NR_vserver (__NR_Linux + 236) +#endif + +#if !defined(__NR_waitid) +#define __NR_waitid (__NR_Linux + 237) +#endif + +/* #define __NR_sys_setaltroot (__NR_Linux + 238) */ + +#if !defined(__NR_add_key) +#define __NR_add_key (__NR_Linux + 239) +#endif + +#if !defined(__NR_request_key) +#define __NR_request_key (__NR_Linux + 240) +#endif + +#if !defined(__NR_keyctl) +#define __NR_keyctl (__NR_Linux + 241) +#endif + +#if !defined(__NR_set_thread_area) +#define __NR_set_thread_area (__NR_Linux + 242) +#endif + +#if !defined(__NR_inotify_init) +#define __NR_inotify_init (__NR_Linux + 243) +#endif + +#if !defined(__NR_inotify_add_watch) +#define __NR_inotify_add_watch (__NR_Linux + 244) +#endif + +#if !defined(__NR_inotify_rm_watch) +#define __NR_inotify_rm_watch (__NR_Linux + 245) +#endif + +#if !defined(__NR_migrate_pages) +#define __NR_migrate_pages (__NR_Linux + 246) +#endif + +#if !defined(__NR_openat) +#define __NR_openat (__NR_Linux + 247) +#endif + +#if !defined(__NR_mkdirat) +#define __NR_mkdirat (__NR_Linux + 248) +#endif + +#if !defined(__NR_mknodat) +#define __NR_mknodat (__NR_Linux + 249) +#endif + +#if !defined(__NR_fchownat) +#define __NR_fchownat (__NR_Linux + 250) +#endif + +#if !defined(__NR_futimesat) +#define __NR_futimesat (__NR_Linux + 251) +#endif + +#if !defined(__NR_newfstatat) +#define __NR_newfstatat (__NR_Linux + 252) +#endif + +#if !defined(__NR_unlinkat) +#define __NR_unlinkat (__NR_Linux + 253) +#endif + +#if !defined(__NR_renameat) +#define __NR_renameat (__NR_Linux + 254) +#endif + +#if !defined(__NR_linkat) +#define __NR_linkat (__NR_Linux + 255) +#endif + +#if !defined(__NR_symlinkat) +#define __NR_symlinkat (__NR_Linux + 256) +#endif + +#if !defined(__NR_readlinkat) +#define __NR_readlinkat (__NR_Linux + 257) +#endif + +#if !defined(__NR_fchmodat) +#define __NR_fchmodat (__NR_Linux + 258) +#endif + +#if !defined(__NR_faccessat) +#define __NR_faccessat (__NR_Linux + 259) +#endif + +#if !defined(__NR_pselect6) +#define __NR_pselect6 (__NR_Linux + 260) +#endif + +#if !defined(__NR_ppoll) +#define __NR_ppoll (__NR_Linux + 261) +#endif + +#if !defined(__NR_unshare) +#define __NR_unshare (__NR_Linux + 262) +#endif + +#if !defined(__NR_splice) +#define __NR_splice (__NR_Linux + 263) +#endif + +#if !defined(__NR_sync_file_range) +#define __NR_sync_file_range (__NR_Linux + 264) +#endif + +#if !defined(__NR_tee) +#define __NR_tee (__NR_Linux + 265) +#endif + +#if !defined(__NR_vmsplice) +#define __NR_vmsplice (__NR_Linux + 266) +#endif + +#if !defined(__NR_move_pages) +#define __NR_move_pages (__NR_Linux + 267) +#endif + +#if !defined(__NR_set_robust_list) +#define __NR_set_robust_list (__NR_Linux + 268) +#endif + +#if !defined(__NR_get_robust_list) +#define __NR_get_robust_list (__NR_Linux + 269) +#endif + +#if !defined(__NR_kexec_load) +#define __NR_kexec_load (__NR_Linux + 270) +#endif + +#if !defined(__NR_getcpu) +#define __NR_getcpu (__NR_Linux + 271) +#endif + +#if !defined(__NR_epoll_pwait) +#define __NR_epoll_pwait (__NR_Linux + 272) +#endif + +#if !defined(__NR_ioprio_set) +#define __NR_ioprio_set (__NR_Linux + 273) +#endif + +#if !defined(__NR_ioprio_get) +#define __NR_ioprio_get (__NR_Linux + 274) +#endif + +#if !defined(__NR_utimensat) +#define __NR_utimensat (__NR_Linux + 275) +#endif + +#if !defined(__NR_signalfd) +#define __NR_signalfd (__NR_Linux + 276) +#endif + +#if !defined(__NR_timerfd) +#define __NR_timerfd (__NR_Linux + 277) +#endif + +#if !defined(__NR_eventfd) +#define __NR_eventfd (__NR_Linux + 278) +#endif + +#if !defined(__NR_fallocate) +#define __NR_fallocate (__NR_Linux + 279) +#endif + +#if !defined(__NR_timerfd_create) +#define __NR_timerfd_create (__NR_Linux + 280) +#endif + +#if !defined(__NR_timerfd_gettime) +#define __NR_timerfd_gettime (__NR_Linux + 281) +#endif + +#if !defined(__NR_timerfd_settime) +#define __NR_timerfd_settime (__NR_Linux + 282) +#endif + +#if !defined(__NR_signalfd4) +#define __NR_signalfd4 (__NR_Linux + 283) +#endif + +#if !defined(__NR_eventfd2) +#define __NR_eventfd2 (__NR_Linux + 284) +#endif + +#if !defined(__NR_epoll_create1) +#define __NR_epoll_create1 (__NR_Linux + 285) +#endif + +#if !defined(__NR_dup3) +#define __NR_dup3 (__NR_Linux + 286) +#endif + +#if !defined(__NR_pipe2) +#define __NR_pipe2 (__NR_Linux + 287) +#endif + +#if !defined(__NR_inotify_init1) +#define __NR_inotify_init1 (__NR_Linux + 288) +#endif + +#if !defined(__NR_preadv) +#define __NR_preadv (__NR_Linux + 289) +#endif + +#if !defined(__NR_pwritev) +#define __NR_pwritev (__NR_Linux + 290) +#endif + +#if !defined(__NR_rt_tgsigqueueinfo) +#define __NR_rt_tgsigqueueinfo (__NR_Linux + 291) +#endif + +#if !defined(__NR_perf_event_open) +#define __NR_perf_event_open (__NR_Linux + 292) +#endif + +#if !defined(__NR_accept4) +#define __NR_accept4 (__NR_Linux + 293) +#endif + +#if !defined(__NR_recvmmsg) +#define __NR_recvmmsg (__NR_Linux + 294) +#endif + +#if !defined(__NR_fanotify_init) +#define __NR_fanotify_init (__NR_Linux + 295) +#endif + +#if !defined(__NR_fanotify_mark) +#define __NR_fanotify_mark (__NR_Linux + 296) +#endif + +#if !defined(__NR_prlimit64) +#define __NR_prlimit64 (__NR_Linux + 297) +#endif + +#if !defined(__NR_name_to_handle_at) +#define __NR_name_to_handle_at (__NR_Linux + 298) +#endif + +#if !defined(__NR_open_by_handle_at) +#define __NR_open_by_handle_at (__NR_Linux + 299) +#endif + +#if !defined(__NR_clock_adjtime) +#define __NR_clock_adjtime (__NR_Linux + 300) +#endif + +#if !defined(__NR_syncfs) +#define __NR_syncfs (__NR_Linux + 301) +#endif + +#if !defined(__NR_sendmmsg) +#define __NR_sendmmsg (__NR_Linux + 302) +#endif + +#if !defined(__NR_setns) +#define __NR_setns (__NR_Linux + 303) +#endif + +#if !defined(__NR_process_vm_readv) +#define __NR_process_vm_readv (__NR_Linux + 304) +#endif + +#if !defined(__NR_process_vm_writev) +#define __NR_process_vm_writev (__NR_Linux + 305) +#endif + +#if !defined(__NR_kcmp) +#define __NR_kcmp (__NR_Linux + 306) +#endif + +#if !defined(__NR_finit_module) +#define __NR_finit_module (__NR_Linux + 307) +#endif + +#if !defined(__NR_getdents64) +#define __NR_getdents64 (__NR_Linux + 308) +#endif + +#if !defined(__NR_sched_setattr) +#define __NR_sched_setattr (__NR_Linux + 309) +#endif + +#if !defined(__NR_sched_getattr) +#define __NR_sched_getattr (__NR_Linux + 310) +#endif + +#if !defined(__NR_renameat2) +#define __NR_renameat2 (__NR_Linux + 311) +#endif + +#if !defined(__NR_seccomp) +#define __NR_seccomp (__NR_Linux + 312) +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_MIPS64_LINUX_SYSCALLS_H_ diff --git a/sandbox/linux/system_headers/mips_linux_syscalls.h b/sandbox/linux/system_headers/mips_linux_syscalls.h new file mode 100644 index 0000000000..eb1717aad9 --- /dev/null +++ b/sandbox/linux/system_headers/mips_linux_syscalls.h @@ -0,0 +1,1428 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// Generated from the Linux kernel's calls.S. +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_MIPS_LINUX_SYSCALLS_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_MIPS_LINUX_SYSCALLS_H_ + +#if !defined(__mips__) || (_MIPS_SIM != _ABIO32) +#error "Including header on wrong architecture" +#endif + +// __NR_Linux, is defined in <asm/unistd.h>. +#include <asm/unistd.h> + +#if !defined(__NR_syscall) +#define __NR_syscall (__NR_Linux + 0) +#endif + +#if !defined(__NR_exit) +#define __NR_exit (__NR_Linux + 1) +#endif + +#if !defined(__NR_fork) +#define __NR_fork (__NR_Linux + 2) +#endif + +#if !defined(__NR_read) +#define __NR_read (__NR_Linux + 3) +#endif + +#if !defined(__NR_write) +#define __NR_write (__NR_Linux + 4) +#endif + +#if !defined(__NR_open) +#define __NR_open (__NR_Linux + 5) +#endif + +#if !defined(__NR_close) +#define __NR_close (__NR_Linux + 6) +#endif + +#if !defined(__NR_waitpid) +#define __NR_waitpid (__NR_Linux + 7) +#endif + +#if !defined(__NR_creat) +#define __NR_creat (__NR_Linux + 8) +#endif + +#if !defined(__NR_link) +#define __NR_link (__NR_Linux + 9) +#endif + +#if !defined(__NR_unlink) +#define __NR_unlink (__NR_Linux + 10) +#endif + +#if !defined(__NR_execve) +#define __NR_execve (__NR_Linux + 11) +#endif + +#if !defined(__NR_chdir) +#define __NR_chdir (__NR_Linux + 12) +#endif + +#if !defined(__NR_time) +#define __NR_time (__NR_Linux + 13) +#endif + +#if !defined(__NR_mknod) +#define __NR_mknod (__NR_Linux + 14) +#endif + +#if !defined(__NR_chmod) +#define __NR_chmod (__NR_Linux + 15) +#endif + +#if !defined(__NR_lchown) +#define __NR_lchown (__NR_Linux + 16) +#endif + +#if !defined(__NR_break) +#define __NR_break (__NR_Linux + 17) +#endif + +#if !defined(__NR_unused18) +#define __NR_unused18 (__NR_Linux + 18) +#endif + +#if !defined(__NR_lseek) +#define __NR_lseek (__NR_Linux + 19) +#endif + +#if !defined(__NR_getpid) +#define __NR_getpid (__NR_Linux + 20) +#endif + +#if !defined(__NR_mount) +#define __NR_mount (__NR_Linux + 21) +#endif + +#if !defined(__NR_umount) +#define __NR_umount (__NR_Linux + 22) +#endif + +#if !defined(__NR_setuid) +#define __NR_setuid (__NR_Linux + 23) +#endif + +#if !defined(__NR_getuid) +#define __NR_getuid (__NR_Linux + 24) +#endif + +#if !defined(__NR_stime) +#define __NR_stime (__NR_Linux + 25) +#endif + +#if !defined(__NR_ptrace) +#define __NR_ptrace (__NR_Linux + 26) +#endif + +#if !defined(__NR_alarm) +#define __NR_alarm (__NR_Linux + 27) +#endif + +#if !defined(__NR_unused28) +#define __NR_unused28 (__NR_Linux + 28) +#endif + +#if !defined(__NR_pause) +#define __NR_pause (__NR_Linux + 29) +#endif + +#if !defined(__NR_utime) +#define __NR_utime (__NR_Linux + 30) +#endif + +#if !defined(__NR_stty) +#define __NR_stty (__NR_Linux + 31) +#endif + +#if !defined(__NR_gtty) +#define __NR_gtty (__NR_Linux + 32) +#endif + +#if !defined(__NR_access) +#define __NR_access (__NR_Linux + 33) +#endif + +#if !defined(__NR_nice) +#define __NR_nice (__NR_Linux + 34) +#endif + +#if !defined(__NR_ftime) +#define __NR_ftime (__NR_Linux + 35) +#endif + +#if !defined(__NR_sync) +#define __NR_sync (__NR_Linux + 36) +#endif + +#if !defined(__NR_kill) +#define __NR_kill (__NR_Linux + 37) +#endif + +#if !defined(__NR_rename) +#define __NR_rename (__NR_Linux + 38) +#endif + +#if !defined(__NR_mkdir) +#define __NR_mkdir (__NR_Linux + 39) +#endif + +#if !defined(__NR_rmdir) +#define __NR_rmdir (__NR_Linux + 40) +#endif + +#if !defined(__NR_dup) +#define __NR_dup (__NR_Linux + 41) +#endif + +#if !defined(__NR_pipe) +#define __NR_pipe (__NR_Linux + 42) +#endif + +#if !defined(__NR_times) +#define __NR_times (__NR_Linux + 43) +#endif + +#if !defined(__NR_prof) +#define __NR_prof (__NR_Linux + 44) +#endif + +#if !defined(__NR_brk) +#define __NR_brk (__NR_Linux + 45) +#endif + +#if !defined(__NR_setgid) +#define __NR_setgid (__NR_Linux + 46) +#endif + +#if !defined(__NR_getgid) +#define __NR_getgid (__NR_Linux + 47) +#endif + +#if !defined(__NR_signal) +#define __NR_signal (__NR_Linux + 48) +#endif + +#if !defined(__NR_geteuid) +#define __NR_geteuid (__NR_Linux + 49) +#endif + +#if !defined(__NR_getegid) +#define __NR_getegid (__NR_Linux + 50) +#endif + +#if !defined(__NR_acct) +#define __NR_acct (__NR_Linux + 51) +#endif + +#if !defined(__NR_umount2) +#define __NR_umount2 (__NR_Linux + 52) +#endif + +#if !defined(__NR_lock) +#define __NR_lock (__NR_Linux + 53) +#endif + +#if !defined(__NR_ioctl) +#define __NR_ioctl (__NR_Linux + 54) +#endif + +#if !defined(__NR_fcntl) +#define __NR_fcntl (__NR_Linux + 55) +#endif + +#if !defined(__NR_mpx) +#define __NR_mpx (__NR_Linux + 56) +#endif + +#if !defined(__NR_setpgid) +#define __NR_setpgid (__NR_Linux + 57) +#endif + +#if !defined(__NR_ulimit) +#define __NR_ulimit (__NR_Linux + 58) +#endif + +#if !defined(__NR_unused59) +#define __NR_unused59 (__NR_Linux + 59) +#endif + +#if !defined(__NR_umask) +#define __NR_umask (__NR_Linux + 60) +#endif + +#if !defined(__NR_chroot) +#define __NR_chroot (__NR_Linux + 61) +#endif + +#if !defined(__NR_ustat) +#define __NR_ustat (__NR_Linux + 62) +#endif + +#if !defined(__NR_dup2) +#define __NR_dup2 (__NR_Linux + 63) +#endif + +#if !defined(__NR_getppid) +#define __NR_getppid (__NR_Linux + 64) +#endif + +#if !defined(__NR_getpgrp) +#define __NR_getpgrp (__NR_Linux + 65) +#endif + +#if !defined(__NR_setsid) +#define __NR_setsid (__NR_Linux + 66) +#endif + +#if !defined(__NR_sigaction) +#define __NR_sigaction (__NR_Linux + 67) +#endif + +#if !defined(__NR_sgetmask) +#define __NR_sgetmask (__NR_Linux + 68) +#endif + +#if !defined(__NR_ssetmask) +#define __NR_ssetmask (__NR_Linux + 69) +#endif + +#if !defined(__NR_setreuid) +#define __NR_setreuid (__NR_Linux + 70) +#endif + +#if !defined(__NR_setregid) +#define __NR_setregid (__NR_Linux + 71) +#endif + +#if !defined(__NR_sigsuspend) +#define __NR_sigsuspend (__NR_Linux + 72) +#endif + +#if !defined(__NR_sigpending) +#define __NR_sigpending (__NR_Linux + 73) +#endif + +#if !defined(__NR_sethostname) +#define __NR_sethostname (__NR_Linux + 74) +#endif + +#if !defined(__NR_setrlimit) +#define __NR_setrlimit (__NR_Linux + 75) +#endif + +#if !defined(__NR_getrlimit) +#define __NR_getrlimit (__NR_Linux + 76) +#endif + +#if !defined(__NR_getrusage) +#define __NR_getrusage (__NR_Linux + 77) +#endif + +#if !defined(__NR_gettimeofday) +#define __NR_gettimeofday (__NR_Linux + 78) +#endif + +#if !defined(__NR_settimeofday) +#define __NR_settimeofday (__NR_Linux + 79) +#endif + +#if !defined(__NR_getgroups) +#define __NR_getgroups (__NR_Linux + 80) +#endif + +#if !defined(__NR_setgroups) +#define __NR_setgroups (__NR_Linux + 81) +#endif + +#if !defined(__NR_reserved82) +#define __NR_reserved82 (__NR_Linux + 82) +#endif + +#if !defined(__NR_symlink) +#define __NR_symlink (__NR_Linux + 83) +#endif + +#if !defined(__NR_unused84) +#define __NR_unused84 (__NR_Linux + 84) +#endif + +#if !defined(__NR_readlink) +#define __NR_readlink (__NR_Linux + 85) +#endif + +#if !defined(__NR_uselib) +#define __NR_uselib (__NR_Linux + 86) +#endif + +#if !defined(__NR_swapon) +#define __NR_swapon (__NR_Linux + 87) +#endif + +#if !defined(__NR_reboot) +#define __NR_reboot (__NR_Linux + 88) +#endif + +#if !defined(__NR_readdir) +#define __NR_readdir (__NR_Linux + 89) +#endif + +#if !defined(__NR_mmap) +#define __NR_mmap (__NR_Linux + 90) +#endif + +#if !defined(__NR_munmap) +#define __NR_munmap (__NR_Linux + 91) +#endif + +#if !defined(__NR_truncate) +#define __NR_truncate (__NR_Linux + 92) +#endif + +#if !defined(__NR_ftruncate) +#define __NR_ftruncate (__NR_Linux + 93) +#endif + +#if !defined(__NR_fchmod) +#define __NR_fchmod (__NR_Linux + 94) +#endif + +#if !defined(__NR_fchown) +#define __NR_fchown (__NR_Linux + 95) +#endif + +#if !defined(__NR_getpriority) +#define __NR_getpriority (__NR_Linux + 96) +#endif + +#if !defined(__NR_setpriority) +#define __NR_setpriority (__NR_Linux + 97) +#endif + +#if !defined(__NR_profil) +#define __NR_profil (__NR_Linux + 98) +#endif + +#if !defined(__NR_statfs) +#define __NR_statfs (__NR_Linux + 99) +#endif + +#if !defined(__NR_fstatfs) +#define __NR_fstatfs (__NR_Linux + 100) +#endif + +#if !defined(__NR_ioperm) +#define __NR_ioperm (__NR_Linux + 101) +#endif + +#if !defined(__NR_socketcall) +#define __NR_socketcall (__NR_Linux + 102) +#endif + +#if !defined(__NR_syslog) +#define __NR_syslog (__NR_Linux + 103) +#endif + +#if !defined(__NR_setitimer) +#define __NR_setitimer (__NR_Linux + 104) +#endif + +#if !defined(__NR_getitimer) +#define __NR_getitimer (__NR_Linux + 105) +#endif + +#if !defined(__NR_stat) +#define __NR_stat (__NR_Linux + 106) +#endif + +#if !defined(__NR_lstat) +#define __NR_lstat (__NR_Linux + 107) +#endif + +#if !defined(__NR_fstat) +#define __NR_fstat (__NR_Linux + 108) +#endif + +#if !defined(__NR_unused109) +#define __NR_unused109 (__NR_Linux + 109) +#endif + +#if !defined(__NR_iopl) +#define __NR_iopl (__NR_Linux + 110) +#endif + +#if !defined(__NR_vhangup) +#define __NR_vhangup (__NR_Linux + 111) +#endif + +#if !defined(__NR_idle) +#define __NR_idle (__NR_Linux + 112) +#endif + +#if !defined(__NR_vm86) +#define __NR_vm86 (__NR_Linux + 113) +#endif + +#if !defined(__NR_wait4) +#define __NR_wait4 (__NR_Linux + 114) +#endif + +#if !defined(__NR_swapoff) +#define __NR_swapoff (__NR_Linux + 115) +#endif + +#if !defined(__NR_sysinfo) +#define __NR_sysinfo (__NR_Linux + 116) +#endif + +#if !defined(__NR_ipc) +#define __NR_ipc (__NR_Linux + 117) +#endif + +#if !defined(__NR_fsync) +#define __NR_fsync (__NR_Linux + 118) +#endif + +#if !defined(__NR_sigreturn) +#define __NR_sigreturn (__NR_Linux + 119) +#endif + +#if !defined(__NR_clone) +#define __NR_clone (__NR_Linux + 120) +#endif + +#if !defined(__NR_setdomainname) +#define __NR_setdomainname (__NR_Linux + 121) +#endif + +#if !defined(__NR_uname) +#define __NR_uname (__NR_Linux + 122) +#endif + +#if !defined(__NR_modify_ldt) +#define __NR_modify_ldt (__NR_Linux + 123) +#endif + +#if !defined(__NR_adjtimex) +#define __NR_adjtimex (__NR_Linux + 124) +#endif + +#if !defined(__NR_mprotect) +#define __NR_mprotect (__NR_Linux + 125) +#endif + +#if !defined(__NR_sigprocmask) +#define __NR_sigprocmask (__NR_Linux + 126) +#endif + +#if !defined(__NR_create_module) +#define __NR_create_module (__NR_Linux + 127) +#endif + +#if !defined(__NR_init_module) +#define __NR_init_module (__NR_Linux + 128) +#endif + +#if !defined(__NR_delete_module) +#define __NR_delete_module (__NR_Linux + 129) +#endif + +#if !defined(__NR_get_kernel_syms) +#define __NR_get_kernel_syms (__NR_Linux + 130) +#endif + +#if !defined(__NR_quotactl) +#define __NR_quotactl (__NR_Linux + 131) +#endif + +#if !defined(__NR_getpgid) +#define __NR_getpgid (__NR_Linux + 132) +#endif + +#if !defined(__NR_fchdir) +#define __NR_fchdir (__NR_Linux + 133) +#endif + +#if !defined(__NR_bdflush) +#define __NR_bdflush (__NR_Linux + 134) +#endif + +#if !defined(__NR_sysfs) +#define __NR_sysfs (__NR_Linux + 135) +#endif + +#if !defined(__NR_personality) +#define __NR_personality (__NR_Linux + 136) +#endif + +#if !defined(__NR_afs_syscall) +#define __NR_afs_syscall \ + (__NR_Linux + 137) /* Syscall for Andrew File System \ + */ +#endif + +#if !defined(__NR_setfsuid) +#define __NR_setfsuid (__NR_Linux + 138) +#endif + +#if !defined(__NR_setfsgid) +#define __NR_setfsgid (__NR_Linux + 139) +#endif + +#if !defined(__NR__llseek) +#define __NR__llseek (__NR_Linux + 140) +#endif + +#if !defined(__NR_getdents) +#define __NR_getdents (__NR_Linux + 141) +#endif + +#if !defined(__NR__newselect) +#define __NR__newselect (__NR_Linux + 142) +#endif + +#if !defined(__NR_flock) +#define __NR_flock (__NR_Linux + 143) +#endif + +#if !defined(__NR_msync) +#define __NR_msync (__NR_Linux + 144) +#endif + +#if !defined(__NR_readv) +#define __NR_readv (__NR_Linux + 145) +#endif + +#if !defined(__NR_writev) +#define __NR_writev (__NR_Linux + 146) +#endif + +#if !defined(__NR_cacheflush) +#define __NR_cacheflush (__NR_Linux + 147) +#endif + +#if !defined(__NR_cachectl) +#define __NR_cachectl (__NR_Linux + 148) +#endif + +#if !defined(__NR_sysmips) +#define __NR_sysmips (__NR_Linux + 149) +#endif + +#if !defined(__NR_unused150) +#define __NR_unused150 (__NR_Linux + 150) +#endif + +#if !defined(__NR_getsid) +#define __NR_getsid (__NR_Linux + 151) +#endif + +#if !defined(__NR_fdatasync) +#define __NR_fdatasync (__NR_Linux + 152) +#endif + +#if !defined(__NR__sysctl) +#define __NR__sysctl (__NR_Linux + 153) +#endif + +#if !defined(__NR_mlock) +#define __NR_mlock (__NR_Linux + 154) +#endif + +#if !defined(__NR_munlock) +#define __NR_munlock (__NR_Linux + 155) +#endif + +#if !defined(__NR_mlockall) +#define __NR_mlockall (__NR_Linux + 156) +#endif + +#if !defined(__NR_munlockall) +#define __NR_munlockall (__NR_Linux + 157) +#endif + +#if !defined(__NR_sched_setparam) +#define __NR_sched_setparam (__NR_Linux + 158) +#endif + +#if !defined(__NR_sched_getparam) +#define __NR_sched_getparam (__NR_Linux + 159) +#endif + +#if !defined(__NR_sched_setscheduler) +#define __NR_sched_setscheduler (__NR_Linux + 160) +#endif + +#if !defined(__NR_sched_getscheduler) +#define __NR_sched_getscheduler (__NR_Linux + 161) +#endif + +#if !defined(__NR_sched_yield) +#define __NR_sched_yield (__NR_Linux + 162) +#endif + +#if !defined(__NR_sched_get_priority_max) +#define __NR_sched_get_priority_max (__NR_Linux + 163) +#endif + +#if !defined(__NR_sched_get_priority_min) +#define __NR_sched_get_priority_min (__NR_Linux + 164) +#endif + +#if !defined(__NR_sched_rr_get_interval) +#define __NR_sched_rr_get_interval (__NR_Linux + 165) +#endif + +#if !defined(__NR_nanosleep) +#define __NR_nanosleep (__NR_Linux + 166) +#endif + +#if !defined(__NR_mremap) +#define __NR_mremap (__NR_Linux + 167) +#endif + +#if !defined(__NR_accept) +#define __NR_accept (__NR_Linux + 168) +#endif + +#if !defined(__NR_bind) +#define __NR_bind (__NR_Linux + 169) +#endif + +#if !defined(__NR_connect) +#define __NR_connect (__NR_Linux + 170) +#endif + +#if !defined(__NR_getpeername) +#define __NR_getpeername (__NR_Linux + 171) +#endif + +#if !defined(__NR_getsockname) +#define __NR_getsockname (__NR_Linux + 172) +#endif + +#if !defined(__NR_getsockopt) +#define __NR_getsockopt (__NR_Linux + 173) +#endif + +#if !defined(__NR_listen) +#define __NR_listen (__NR_Linux + 174) +#endif + +#if !defined(__NR_recv) +#define __NR_recv (__NR_Linux + 175) +#endif + +#if !defined(__NR_recvfrom) +#define __NR_recvfrom (__NR_Linux + 176) +#endif + +#if !defined(__NR_recvmsg) +#define __NR_recvmsg (__NR_Linux + 177) +#endif + +#if !defined(__NR_send) +#define __NR_send (__NR_Linux + 178) +#endif + +#if !defined(__NR_sendmsg) +#define __NR_sendmsg (__NR_Linux + 179) +#endif + +#if !defined(__NR_sendto) +#define __NR_sendto (__NR_Linux + 180) +#endif + +#if !defined(__NR_setsockopt) +#define __NR_setsockopt (__NR_Linux + 181) +#endif + +#if !defined(__NR_shutdown) +#define __NR_shutdown (__NR_Linux + 182) +#endif + +#if !defined(__NR_socket) +#define __NR_socket (__NR_Linux + 183) +#endif + +#if !defined(__NR_socketpair) +#define __NR_socketpair (__NR_Linux + 184) +#endif + +#if !defined(__NR_setresuid) +#define __NR_setresuid (__NR_Linux + 185) +#endif + +#if !defined(__NR_getresuid) +#define __NR_getresuid (__NR_Linux + 186) +#endif + +#if !defined(__NR_query_module) +#define __NR_query_module (__NR_Linux + 187) +#endif + +#if !defined(__NR_poll) +#define __NR_poll (__NR_Linux + 188) +#endif + +#if !defined(__NR_nfsservctl) +#define __NR_nfsservctl (__NR_Linux + 189) +#endif + +#if !defined(__NR_setresgid) +#define __NR_setresgid (__NR_Linux + 190) +#endif + +#if !defined(__NR_getresgid) +#define __NR_getresgid (__NR_Linux + 191) +#endif + +#if !defined(__NR_prctl) +#define __NR_prctl (__NR_Linux + 192) +#endif + +#if !defined(__NR_rt_sigreturn) +#define __NR_rt_sigreturn (__NR_Linux + 193) +#endif + +#if !defined(__NR_rt_sigaction) +#define __NR_rt_sigaction (__NR_Linux + 194) +#endif + +#if !defined(__NR_rt_sigprocmask) +#define __NR_rt_sigprocmask (__NR_Linux + 195) +#endif + +#if !defined(__NR_rt_sigpending) +#define __NR_rt_sigpending (__NR_Linux + 196) +#endif + +#if !defined(__NR_rt_sigtimedwait) +#define __NR_rt_sigtimedwait (__NR_Linux + 197) +#endif + +#if !defined(__NR_rt_sigqueueinfo) +#define __NR_rt_sigqueueinfo (__NR_Linux + 198) +#endif + +#if !defined(__NR_rt_sigsuspend) +#define __NR_rt_sigsuspend (__NR_Linux + 199) +#endif + +#if !defined(__NR_pread64) +#define __NR_pread64 (__NR_Linux + 200) +#endif + +#if !defined(__NR_pwrite64) +#define __NR_pwrite64 (__NR_Linux + 201) +#endif + +#if !defined(__NR_chown) +#define __NR_chown (__NR_Linux + 202) +#endif + +#if !defined(__NR_getcwd) +#define __NR_getcwd (__NR_Linux + 203) +#endif + +#if !defined(__NR_capget) +#define __NR_capget (__NR_Linux + 204) +#endif + +#if !defined(__NR_capset) +#define __NR_capset (__NR_Linux + 205) +#endif + +#if !defined(__NR_sigaltstack) +#define __NR_sigaltstack (__NR_Linux + 206) +#endif + +#if !defined(__NR_sendfile) +#define __NR_sendfile (__NR_Linux + 207) +#endif + +#if !defined(__NR_getpmsg) +#define __NR_getpmsg (__NR_Linux + 208) +#endif + +#if !defined(__NR_putpmsg) +#define __NR_putpmsg (__NR_Linux + 209) +#endif + +#if !defined(__NR_mmap2) +#define __NR_mmap2 (__NR_Linux + 210) +#endif + +#if !defined(__NR_truncate64) +#define __NR_truncate64 (__NR_Linux + 211) +#endif + +#if !defined(__NR_ftruncate64) +#define __NR_ftruncate64 (__NR_Linux + 212) +#endif + +#if !defined(__NR_stat64) +#define __NR_stat64 (__NR_Linux + 213) +#endif + +#if !defined(__NR_lstat64) +#define __NR_lstat64 (__NR_Linux + 214) +#endif + +#if !defined(__NR_fstat64) +#define __NR_fstat64 (__NR_Linux + 215) +#endif + +#if !defined(__NR_pivot_root) +#define __NR_pivot_root (__NR_Linux + 216) +#endif + +#if !defined(__NR_mincore) +#define __NR_mincore (__NR_Linux + 217) +#endif + +#if !defined(__NR_madvise) +#define __NR_madvise (__NR_Linux + 218) +#endif + +#if !defined(__NR_getdents64) +#define __NR_getdents64 (__NR_Linux + 219) +#endif + +#if !defined(__NR_fcntl64) +#define __NR_fcntl64 (__NR_Linux + 220) +#endif + +#if !defined(__NR_reserved221) +#define __NR_reserved221 (__NR_Linux + 221) +#endif + +#if !defined(__NR_gettid) +#define __NR_gettid (__NR_Linux + 222) +#endif + +#if !defined(__NR_readahead) +#define __NR_readahead (__NR_Linux + 223) +#endif + +#if !defined(__NR_setxattr) +#define __NR_setxattr (__NR_Linux + 224) +#endif + +#if !defined(__NR_lsetxattr) +#define __NR_lsetxattr (__NR_Linux + 225) +#endif + +#if !defined(__NR_fsetxattr) +#define __NR_fsetxattr (__NR_Linux + 226) +#endif + +#if !defined(__NR_getxattr) +#define __NR_getxattr (__NR_Linux + 227) +#endif + +#if !defined(__NR_lgetxattr) +#define __NR_lgetxattr (__NR_Linux + 228) +#endif + +#if !defined(__NR_fgetxattr) +#define __NR_fgetxattr (__NR_Linux + 229) +#endif + +#if !defined(__NR_listxattr) +#define __NR_listxattr (__NR_Linux + 230) +#endif + +#if !defined(__NR_llistxattr) +#define __NR_llistxattr (__NR_Linux + 231) +#endif + +#if !defined(__NR_flistxattr) +#define __NR_flistxattr (__NR_Linux + 232) +#endif + +#if !defined(__NR_removexattr) +#define __NR_removexattr (__NR_Linux + 233) +#endif + +#if !defined(__NR_lremovexattr) +#define __NR_lremovexattr (__NR_Linux + 234) +#endif + +#if !defined(__NR_fremovexattr) +#define __NR_fremovexattr (__NR_Linux + 235) +#endif + +#if !defined(__NR_tkill) +#define __NR_tkill (__NR_Linux + 236) +#endif + +#if !defined(__NR_sendfile64) +#define __NR_sendfile64 (__NR_Linux + 237) +#endif + +#if !defined(__NR_futex) +#define __NR_futex (__NR_Linux + 238) +#endif + +#if !defined(__NR_sched_setaffinity) +#define __NR_sched_setaffinity (__NR_Linux + 239) +#endif + +#if !defined(__NR_sched_getaffinity) +#define __NR_sched_getaffinity (__NR_Linux + 240) +#endif + +#if !defined(__NR_io_setup) +#define __NR_io_setup (__NR_Linux + 241) +#endif + +#if !defined(__NR_io_destroy) +#define __NR_io_destroy (__NR_Linux + 242) +#endif + +#if !defined(__NR_io_getevents) +#define __NR_io_getevents (__NR_Linux + 243) +#endif + +#if !defined(__NR_io_submit) +#define __NR_io_submit (__NR_Linux + 244) +#endif + +#if !defined(__NR_io_cancel) +#define __NR_io_cancel (__NR_Linux + 245) +#endif + +#if !defined(__NR_exit_group) +#define __NR_exit_group (__NR_Linux + 246) +#endif + +#if !defined(__NR_lookup_dcookie) +#define __NR_lookup_dcookie (__NR_Linux + 247) +#endif + +#if !defined(__NR_epoll_create) +#define __NR_epoll_create (__NR_Linux + 248) +#endif + +#if !defined(__NR_epoll_ctl) +#define __NR_epoll_ctl (__NR_Linux + 249) +#endif + +#if !defined(__NR_epoll_wait) +#define __NR_epoll_wait (__NR_Linux + 250) +#endif + +#if !defined(__NR_remap_file_pages) +#define __NR_remap_file_pages (__NR_Linux + 251) +#endif + +#if !defined(__NR_set_tid_address) +#define __NR_set_tid_address (__NR_Linux + 252) +#endif + +#if !defined(__NR_restart_syscall) +#define __NR_restart_syscall (__NR_Linux + 253) +#endif + +#if !defined(__NR_fadvise64) +#define __NR_fadvise64 (__NR_Linux + 254) +#endif + +#if !defined(__NR_statfs64) +#define __NR_statfs64 (__NR_Linux + 255) +#endif + +#if !defined(__NR_fstatfs64) +#define __NR_fstatfs64 (__NR_Linux + 256) +#endif + +#if !defined(__NR_timer_create) +#define __NR_timer_create (__NR_Linux + 257) +#endif + +#if !defined(__NR_timer_settime) +#define __NR_timer_settime (__NR_Linux + 258) +#endif + +#if !defined(__NR_timer_gettime) +#define __NR_timer_gettime (__NR_Linux + 259) +#endif + +#if !defined(__NR_timer_getoverrun) +#define __NR_timer_getoverrun (__NR_Linux + 260) +#endif + +#if !defined(__NR_timer_delete) +#define __NR_timer_delete (__NR_Linux + 261) +#endif + +#if !defined(__NR_clock_settime) +#define __NR_clock_settime (__NR_Linux + 262) +#endif + +#if !defined(__NR_clock_gettime) +#define __NR_clock_gettime (__NR_Linux + 263) +#endif + +#if !defined(__NR_clock_getres) +#define __NR_clock_getres (__NR_Linux + 264) +#endif + +#if !defined(__NR_clock_nanosleep) +#define __NR_clock_nanosleep (__NR_Linux + 265) +#endif + +#if !defined(__NR_tgkill) +#define __NR_tgkill (__NR_Linux + 266) +#endif + +#if !defined(__NR_utimes) +#define __NR_utimes (__NR_Linux + 267) +#endif + +#if !defined(__NR_mbind) +#define __NR_mbind (__NR_Linux + 268) +#endif + +#if !defined(__NR_get_mempolicy) +#define __NR_get_mempolicy (__NR_Linux + 269) +#endif + +#if !defined(__NR_set_mempolicy) +#define __NR_set_mempolicy (__NR_Linux + 270) +#endif + +#if !defined(__NR_mq_open) +#define __NR_mq_open (__NR_Linux + 271) +#endif + +#if !defined(__NR_mq_unlink) +#define __NR_mq_unlink (__NR_Linux + 272) +#endif + +#if !defined(__NR_mq_timedsend) +#define __NR_mq_timedsend (__NR_Linux + 273) +#endif + +#if !defined(__NR_mq_timedreceive) +#define __NR_mq_timedreceive (__NR_Linux + 274) +#endif + +#if !defined(__NR_mq_notify) +#define __NR_mq_notify (__NR_Linux + 275) +#endif + +#if !defined(__NR_mq_getsetattr) +#define __NR_mq_getsetattr (__NR_Linux + 276) +#endif + +#if !defined(__NR_vserver) +#define __NR_vserver (__NR_Linux + 277) +#endif + +#if !defined(__NR_waitid) +#define __NR_waitid (__NR_Linux + 278) +#endif + +/* #define __NR_sys_setaltroot (__NR_Linux + 279) */ + +#if !defined(__NR_add_key) +#define __NR_add_key (__NR_Linux + 280) +#endif + +#if !defined(__NR_request_key) +#define __NR_request_key (__NR_Linux + 281) +#endif + +#if !defined(__NR_keyctl) +#define __NR_keyctl (__NR_Linux + 282) +#endif + +#if !defined(__NR_set_thread_area) +#define __NR_set_thread_area (__NR_Linux + 283) +#endif + +#if !defined(__NR_inotify_init) +#define __NR_inotify_init (__NR_Linux + 284) +#endif + +#if !defined(__NR_inotify_add_watch) +#define __NR_inotify_add_watch (__NR_Linux + 285) +#endif + +#if !defined(__NR_inotify_rm_watch) +#define __NR_inotify_rm_watch (__NR_Linux + 286) +#endif + +#if !defined(__NR_migrate_pages) +#define __NR_migrate_pages (__NR_Linux + 287) +#endif + +#if !defined(__NR_openat) +#define __NR_openat (__NR_Linux + 288) +#endif + +#if !defined(__NR_mkdirat) +#define __NR_mkdirat (__NR_Linux + 289) +#endif + +#if !defined(__NR_mknodat) +#define __NR_mknodat (__NR_Linux + 290) +#endif + +#if !defined(__NR_fchownat) +#define __NR_fchownat (__NR_Linux + 291) +#endif + +#if !defined(__NR_futimesat) +#define __NR_futimesat (__NR_Linux + 292) +#endif + +#if !defined(__NR_fstatat64) +#define __NR_fstatat64 (__NR_Linux + 293) +#endif + +#if !defined(__NR_unlinkat) +#define __NR_unlinkat (__NR_Linux + 294) +#endif + +#if !defined(__NR_renameat) +#define __NR_renameat (__NR_Linux + 295) +#endif + +#if !defined(__NR_linkat) +#define __NR_linkat (__NR_Linux + 296) +#endif + +#if !defined(__NR_symlinkat) +#define __NR_symlinkat (__NR_Linux + 297) +#endif + +#if !defined(__NR_readlinkat) +#define __NR_readlinkat (__NR_Linux + 298) +#endif + +#if !defined(__NR_fchmodat) +#define __NR_fchmodat (__NR_Linux + 299) +#endif + +#if !defined(__NR_faccessat) +#define __NR_faccessat (__NR_Linux + 300) +#endif + +#if !defined(__NR_pselect6) +#define __NR_pselect6 (__NR_Linux + 301) +#endif + +#if !defined(__NR_ppoll) +#define __NR_ppoll (__NR_Linux + 302) +#endif + +#if !defined(__NR_unshare) +#define __NR_unshare (__NR_Linux + 303) +#endif + +#if !defined(__NR_splice) +#define __NR_splice (__NR_Linux + 304) +#endif + +#if !defined(__NR_sync_file_range) +#define __NR_sync_file_range (__NR_Linux + 305) +#endif + +#if !defined(__NR_tee) +#define __NR_tee (__NR_Linux + 306) +#endif + +#if !defined(__NR_vmsplice) +#define __NR_vmsplice (__NR_Linux + 307) +#endif + +#if !defined(__NR_move_pages) +#define __NR_move_pages (__NR_Linux + 308) +#endif + +#if !defined(__NR_set_robust_list) +#define __NR_set_robust_list (__NR_Linux + 309) +#endif + +#if !defined(__NR_get_robust_list) +#define __NR_get_robust_list (__NR_Linux + 310) +#endif + +#if !defined(__NR_kexec_load) +#define __NR_kexec_load (__NR_Linux + 311) +#endif + +#if !defined(__NR_getcpu) +#define __NR_getcpu (__NR_Linux + 312) +#endif + +#if !defined(__NR_epoll_pwait) +#define __NR_epoll_pwait (__NR_Linux + 313) +#endif + +#if !defined(__NR_ioprio_set) +#define __NR_ioprio_set (__NR_Linux + 314) +#endif + +#if !defined(__NR_ioprio_get) +#define __NR_ioprio_get (__NR_Linux + 315) +#endif + +#if !defined(__NR_utimensat) +#define __NR_utimensat (__NR_Linux + 316) +#endif + +#if !defined(__NR_signalfd) +#define __NR_signalfd (__NR_Linux + 317) +#endif + +#if !defined(__NR_timerfd) +#define __NR_timerfd (__NR_Linux + 318) +#endif + +#if !defined(__NR_eventfd) +#define __NR_eventfd (__NR_Linux + 319) +#endif + +#if !defined(__NR_eventfd) +#define __NR_eventfd (__NR_Linux + 320) +#endif + +#if !defined(__NR_timerfd_create) +#define __NR_timerfd_create (__NR_Linux + 321) +#endif + +#if !defined(__NR_timerfd_gettime) +#define __NR_timerfd_gettime (__NR_Linux + 322) +#endif + +#if !defined(__NR_timerfd_settime) +#define __NR_timerfd_settime (__NR_Linux + 323) +#endif + +#if !defined(__NR_signalfd4) +#define __NR_signalfd4 (__NR_Linux + 324) +#endif + +#if !defined(__NR_eventfd2) +#define __NR_eventfd2 (__NR_Linux + 325) +#endif + +#if !defined(__NR_epoll_create1) +#define __NR_epoll_create1 (__NR_Linux + 326) +#endif + +#if !defined(__NR_dup3) +#define __NR_dup3 (__NR_Linux + 327) +#endif + +#if !defined(__NR_pipe2) +#define __NR_pipe2 (__NR_Linux + 328) +#endif + +#if !defined(__NR_inotify_init1) +#define __NR_inotify_init1 (__NR_Linux + 329) +#endif + +#if !defined(__NR_preadv) +#define __NR_preadv (__NR_Linux + 330) +#endif + +#if !defined(__NR_pwritev) +#define __NR_pwritev (__NR_Linux + 331) +#endif + +#if !defined(__NR_rt_tgsigqueueinfo) +#define __NR_rt_tgsigqueueinfo (__NR_Linux + 332) +#endif + +#if !defined(__NR_perf_event_open) +#define __NR_perf_event_open (__NR_Linux + 333) +#endif + +#if !defined(__NR_accept4) +#define __NR_accept4 (__NR_Linux + 334) +#endif + +#if !defined(__NR_recvmmsg) +#define __NR_recvmmsg (__NR_Linux + 335) +#endif + +#if !defined(__NR_fanotify_init) +#define __NR_fanotify_init (__NR_Linux + 336) +#endif + +#if !defined(__NR_fanotify_mark) +#define __NR_fanotify_mark (__NR_Linux + 337) +#endif + +#if !defined(__NR_prlimit64) +#define __NR_prlimit64 (__NR_Linux + 338) +#endif + +#if !defined(__NR_name_to_handle_at) +#define __NR_name_to_handle_at (__NR_Linux + 339) +#endif + +#if !defined(__NR_open_by_handle_at) +#define __NR_open_by_handle_at (__NR_Linux + 340) +#endif + +#if !defined(__NR_clock_adjtime) +#define __NR_clock_adjtime (__NR_Linux + 341) +#endif + +#if !defined(__NR_syncfs) +#define __NR_syncfs (__NR_Linux + 342) +#endif + +#if !defined(__NR_sendmmsg) +#define __NR_sendmmsg (__NR_Linux + 343) +#endif + +#if !defined(__NR_setns) +#define __NR_setns (__NR_Linux + 344) +#endif + +#if !defined(__NR_process_vm_readv) +#define __NR_process_vm_readv (__NR_Linux + 345) +#endif + +#if !defined(__NR_process_vm_writev) +#define __NR_process_vm_writev (__NR_Linux + 346) +#endif + +#if !defined(__NR_kcmp) +#define __NR_kcmp (__NR_Linux + 347) +#endif + +#if !defined(__NR_finit_module) +#define __NR_finit_module (__NR_Linux + 348) +#endif + +#if !defined(__NR_sched_setattr) +#define __NR_sched_setattr (__NR_Linux + 349) +#endif + +#if !defined(__NR_sched_getattr) +#define __NR_sched_getattr (__NR_Linux + 350) +#endif + +#if !defined(__NR_renameat2) +#define __NR_renameat2 (__NR_Linux + 351) +#endif + +#if !defined(__NR_seccomp) +#define __NR_seccomp (__NR_Linux + 352) +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_MIPS_LINUX_SYSCALLS_H_ diff --git a/sandbox/linux/system_headers/mips_linux_ucontext.h b/sandbox/linux/system_headers/mips_linux_ucontext.h new file mode 100644 index 0000000000..27b3763522 --- /dev/null +++ b/sandbox/linux/system_headers/mips_linux_ucontext.h @@ -0,0 +1,51 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_MIPS_LINUX_UCONTEXT_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_MIPS_LINUX_UCONTEXT_H_ + +// This is mostly copied from breakpad (common/android/include/sys/ucontext.h), +// except we do use sigset_t for uc_sigmask instead of a custom type. +#if !defined(__BIONIC_HAVE_UCONTEXT_T) +// Ensure that 'stack_t' is defined. +#include <asm/signal.h> + +// We also need greg_t for the sandbox, include it in this header as well. +typedef unsigned long greg_t; + +typedef struct { + uint32_t regmask; + uint32_t status; + uint64_t pc; + uint64_t gregs[32]; + uint64_t fpregs[32]; + uint32_t acx; + uint32_t fpc_csr; + uint32_t fpc_eir; + uint32_t used_math; + uint32_t dsp; + uint64_t mdhi; + uint64_t mdlo; + uint32_t hi1; + uint32_t lo1; + uint32_t hi2; + uint32_t lo2; + uint32_t hi3; + uint32_t lo3; +} mcontext_t; + +typedef struct ucontext { + uint32_t uc_flags; + struct ucontext* uc_link; + stack_t uc_stack; + mcontext_t uc_mcontext; + sigset_t uc_sigmask; + // Other fields are not used by Google Breakpad. Don't define them. +} ucontext_t; + +#else +#include <sys/ucontext.h> +#endif // __BIONIC_HAVE_UCONTEXT_T + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_MIPS_LINUX_UCONTEXT_H_ diff --git a/sandbox/linux/system_headers/x86_32_linux_syscalls.h b/sandbox/linux/system_headers/x86_32_linux_syscalls.h new file mode 100644 index 0000000000..a6afc62d99 --- /dev/null +++ b/sandbox/linux/system_headers/x86_32_linux_syscalls.h @@ -0,0 +1,1426 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// Generated from the Linux kernel's syscall_32.tbl. +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_X86_32_LINUX_SYSCALLS_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_X86_32_LINUX_SYSCALLS_H_ + +#if !defined(__i386__) +#error "Including header on wrong architecture" +#endif + +#if !defined(__NR_restart_syscall) +#define __NR_restart_syscall 0 +#endif + +#if !defined(__NR_exit) +#define __NR_exit 1 +#endif + +#if !defined(__NR_fork) +#define __NR_fork 2 +#endif + +#if !defined(__NR_read) +#define __NR_read 3 +#endif + +#if !defined(__NR_write) +#define __NR_write 4 +#endif + +#if !defined(__NR_open) +#define __NR_open 5 +#endif + +#if !defined(__NR_close) +#define __NR_close 6 +#endif + +#if !defined(__NR_waitpid) +#define __NR_waitpid 7 +#endif + +#if !defined(__NR_creat) +#define __NR_creat 8 +#endif + +#if !defined(__NR_link) +#define __NR_link 9 +#endif + +#if !defined(__NR_unlink) +#define __NR_unlink 10 +#endif + +#if !defined(__NR_execve) +#define __NR_execve 11 +#endif + +#if !defined(__NR_chdir) +#define __NR_chdir 12 +#endif + +#if !defined(__NR_time) +#define __NR_time 13 +#endif + +#if !defined(__NR_mknod) +#define __NR_mknod 14 +#endif + +#if !defined(__NR_chmod) +#define __NR_chmod 15 +#endif + +#if !defined(__NR_lchown) +#define __NR_lchown 16 +#endif + +#if !defined(__NR_break) +#define __NR_break 17 +#endif + +#if !defined(__NR_oldstat) +#define __NR_oldstat 18 +#endif + +#if !defined(__NR_lseek) +#define __NR_lseek 19 +#endif + +#if !defined(__NR_getpid) +#define __NR_getpid 20 +#endif + +#if !defined(__NR_mount) +#define __NR_mount 21 +#endif + +#if !defined(__NR_umount) +#define __NR_umount 22 +#endif + +#if !defined(__NR_setuid) +#define __NR_setuid 23 +#endif + +#if !defined(__NR_getuid) +#define __NR_getuid 24 +#endif + +#if !defined(__NR_stime) +#define __NR_stime 25 +#endif + +#if !defined(__NR_ptrace) +#define __NR_ptrace 26 +#endif + +#if !defined(__NR_alarm) +#define __NR_alarm 27 +#endif + +#if !defined(__NR_oldfstat) +#define __NR_oldfstat 28 +#endif + +#if !defined(__NR_pause) +#define __NR_pause 29 +#endif + +#if !defined(__NR_utime) +#define __NR_utime 30 +#endif + +#if !defined(__NR_stty) +#define __NR_stty 31 +#endif + +#if !defined(__NR_gtty) +#define __NR_gtty 32 +#endif + +#if !defined(__NR_access) +#define __NR_access 33 +#endif + +#if !defined(__NR_nice) +#define __NR_nice 34 +#endif + +#if !defined(__NR_ftime) +#define __NR_ftime 35 +#endif + +#if !defined(__NR_sync) +#define __NR_sync 36 +#endif + +#if !defined(__NR_kill) +#define __NR_kill 37 +#endif + +#if !defined(__NR_rename) +#define __NR_rename 38 +#endif + +#if !defined(__NR_mkdir) +#define __NR_mkdir 39 +#endif + +#if !defined(__NR_rmdir) +#define __NR_rmdir 40 +#endif + +#if !defined(__NR_dup) +#define __NR_dup 41 +#endif + +#if !defined(__NR_pipe) +#define __NR_pipe 42 +#endif + +#if !defined(__NR_times) +#define __NR_times 43 +#endif + +#if !defined(__NR_prof) +#define __NR_prof 44 +#endif + +#if !defined(__NR_brk) +#define __NR_brk 45 +#endif + +#if !defined(__NR_setgid) +#define __NR_setgid 46 +#endif + +#if !defined(__NR_getgid) +#define __NR_getgid 47 +#endif + +#if !defined(__NR_signal) +#define __NR_signal 48 +#endif + +#if !defined(__NR_geteuid) +#define __NR_geteuid 49 +#endif + +#if !defined(__NR_getegid) +#define __NR_getegid 50 +#endif + +#if !defined(__NR_acct) +#define __NR_acct 51 +#endif + +#if !defined(__NR_umount2) +#define __NR_umount2 52 +#endif + +#if !defined(__NR_lock) +#define __NR_lock 53 +#endif + +#if !defined(__NR_ioctl) +#define __NR_ioctl 54 +#endif + +#if !defined(__NR_fcntl) +#define __NR_fcntl 55 +#endif + +#if !defined(__NR_mpx) +#define __NR_mpx 56 +#endif + +#if !defined(__NR_setpgid) +#define __NR_setpgid 57 +#endif + +#if !defined(__NR_ulimit) +#define __NR_ulimit 58 +#endif + +#if !defined(__NR_oldolduname) +#define __NR_oldolduname 59 +#endif + +#if !defined(__NR_umask) +#define __NR_umask 60 +#endif + +#if !defined(__NR_chroot) +#define __NR_chroot 61 +#endif + +#if !defined(__NR_ustat) +#define __NR_ustat 62 +#endif + +#if !defined(__NR_dup2) +#define __NR_dup2 63 +#endif + +#if !defined(__NR_getppid) +#define __NR_getppid 64 +#endif + +#if !defined(__NR_getpgrp) +#define __NR_getpgrp 65 +#endif + +#if !defined(__NR_setsid) +#define __NR_setsid 66 +#endif + +#if !defined(__NR_sigaction) +#define __NR_sigaction 67 +#endif + +#if !defined(__NR_sgetmask) +#define __NR_sgetmask 68 +#endif + +#if !defined(__NR_ssetmask) +#define __NR_ssetmask 69 +#endif + +#if !defined(__NR_setreuid) +#define __NR_setreuid 70 +#endif + +#if !defined(__NR_setregid) +#define __NR_setregid 71 +#endif + +#if !defined(__NR_sigsuspend) +#define __NR_sigsuspend 72 +#endif + +#if !defined(__NR_sigpending) +#define __NR_sigpending 73 +#endif + +#if !defined(__NR_sethostname) +#define __NR_sethostname 74 +#endif + +#if !defined(__NR_setrlimit) +#define __NR_setrlimit 75 +#endif + +#if !defined(__NR_getrlimit) +#define __NR_getrlimit 76 +#endif + +#if !defined(__NR_getrusage) +#define __NR_getrusage 77 +#endif + +#if !defined(__NR_gettimeofday) +#define __NR_gettimeofday 78 +#endif + +#if !defined(__NR_settimeofday) +#define __NR_settimeofday 79 +#endif + +#if !defined(__NR_getgroups) +#define __NR_getgroups 80 +#endif + +#if !defined(__NR_setgroups) +#define __NR_setgroups 81 +#endif + +#if !defined(__NR_select) +#define __NR_select 82 +#endif + +#if !defined(__NR_symlink) +#define __NR_symlink 83 +#endif + +#if !defined(__NR_oldlstat) +#define __NR_oldlstat 84 +#endif + +#if !defined(__NR_readlink) +#define __NR_readlink 85 +#endif + +#if !defined(__NR_uselib) +#define __NR_uselib 86 +#endif + +#if !defined(__NR_swapon) +#define __NR_swapon 87 +#endif + +#if !defined(__NR_reboot) +#define __NR_reboot 88 +#endif + +#if !defined(__NR_readdir) +#define __NR_readdir 89 +#endif + +#if !defined(__NR_mmap) +#define __NR_mmap 90 +#endif + +#if !defined(__NR_munmap) +#define __NR_munmap 91 +#endif + +#if !defined(__NR_truncate) +#define __NR_truncate 92 +#endif + +#if !defined(__NR_ftruncate) +#define __NR_ftruncate 93 +#endif + +#if !defined(__NR_fchmod) +#define __NR_fchmod 94 +#endif + +#if !defined(__NR_fchown) +#define __NR_fchown 95 +#endif + +#if !defined(__NR_getpriority) +#define __NR_getpriority 96 +#endif + +#if !defined(__NR_setpriority) +#define __NR_setpriority 97 +#endif + +#if !defined(__NR_profil) +#define __NR_profil 98 +#endif + +#if !defined(__NR_statfs) +#define __NR_statfs 99 +#endif + +#if !defined(__NR_fstatfs) +#define __NR_fstatfs 100 +#endif + +#if !defined(__NR_ioperm) +#define __NR_ioperm 101 +#endif + +#if !defined(__NR_socketcall) +#define __NR_socketcall 102 +#endif + +#if !defined(__NR_syslog) +#define __NR_syslog 103 +#endif + +#if !defined(__NR_setitimer) +#define __NR_setitimer 104 +#endif + +#if !defined(__NR_getitimer) +#define __NR_getitimer 105 +#endif + +#if !defined(__NR_stat) +#define __NR_stat 106 +#endif + +#if !defined(__NR_lstat) +#define __NR_lstat 107 +#endif + +#if !defined(__NR_fstat) +#define __NR_fstat 108 +#endif + +#if !defined(__NR_olduname) +#define __NR_olduname 109 +#endif + +#if !defined(__NR_iopl) +#define __NR_iopl 110 +#endif + +#if !defined(__NR_vhangup) +#define __NR_vhangup 111 +#endif + +#if !defined(__NR_idle) +#define __NR_idle 112 +#endif + +#if !defined(__NR_vm86old) +#define __NR_vm86old 113 +#endif + +#if !defined(__NR_wait4) +#define __NR_wait4 114 +#endif + +#if !defined(__NR_swapoff) +#define __NR_swapoff 115 +#endif + +#if !defined(__NR_sysinfo) +#define __NR_sysinfo 116 +#endif + +#if !defined(__NR_ipc) +#define __NR_ipc 117 +#endif + +#if !defined(__NR_fsync) +#define __NR_fsync 118 +#endif + +#if !defined(__NR_sigreturn) +#define __NR_sigreturn 119 +#endif + +#if !defined(__NR_clone) +#define __NR_clone 120 +#endif + +#if !defined(__NR_setdomainname) +#define __NR_setdomainname 121 +#endif + +#if !defined(__NR_uname) +#define __NR_uname 122 +#endif + +#if !defined(__NR_modify_ldt) +#define __NR_modify_ldt 123 +#endif + +#if !defined(__NR_adjtimex) +#define __NR_adjtimex 124 +#endif + +#if !defined(__NR_mprotect) +#define __NR_mprotect 125 +#endif + +#if !defined(__NR_sigprocmask) +#define __NR_sigprocmask 126 +#endif + +#if !defined(__NR_create_module) +#define __NR_create_module 127 +#endif + +#if !defined(__NR_init_module) +#define __NR_init_module 128 +#endif + +#if !defined(__NR_delete_module) +#define __NR_delete_module 129 +#endif + +#if !defined(__NR_get_kernel_syms) +#define __NR_get_kernel_syms 130 +#endif + +#if !defined(__NR_quotactl) +#define __NR_quotactl 131 +#endif + +#if !defined(__NR_getpgid) +#define __NR_getpgid 132 +#endif + +#if !defined(__NR_fchdir) +#define __NR_fchdir 133 +#endif + +#if !defined(__NR_bdflush) +#define __NR_bdflush 134 +#endif + +#if !defined(__NR_sysfs) +#define __NR_sysfs 135 +#endif + +#if !defined(__NR_personality) +#define __NR_personality 136 +#endif + +#if !defined(__NR_afs_syscall) +#define __NR_afs_syscall 137 +#endif + +#if !defined(__NR_setfsuid) +#define __NR_setfsuid 138 +#endif + +#if !defined(__NR_setfsgid) +#define __NR_setfsgid 139 +#endif + +#if !defined(__NR__llseek) +#define __NR__llseek 140 +#endif + +#if !defined(__NR_getdents) +#define __NR_getdents 141 +#endif + +#if !defined(__NR__newselect) +#define __NR__newselect 142 +#endif + +#if !defined(__NR_flock) +#define __NR_flock 143 +#endif + +#if !defined(__NR_msync) +#define __NR_msync 144 +#endif + +#if !defined(__NR_readv) +#define __NR_readv 145 +#endif + +#if !defined(__NR_writev) +#define __NR_writev 146 +#endif + +#if !defined(__NR_getsid) +#define __NR_getsid 147 +#endif + +#if !defined(__NR_fdatasync) +#define __NR_fdatasync 148 +#endif + +#if !defined(__NR__sysctl) +#define __NR__sysctl 149 +#endif + +#if !defined(__NR_mlock) +#define __NR_mlock 150 +#endif + +#if !defined(__NR_munlock) +#define __NR_munlock 151 +#endif + +#if !defined(__NR_mlockall) +#define __NR_mlockall 152 +#endif + +#if !defined(__NR_munlockall) +#define __NR_munlockall 153 +#endif + +#if !defined(__NR_sched_setparam) +#define __NR_sched_setparam 154 +#endif + +#if !defined(__NR_sched_getparam) +#define __NR_sched_getparam 155 +#endif + +#if !defined(__NR_sched_setscheduler) +#define __NR_sched_setscheduler 156 +#endif + +#if !defined(__NR_sched_getscheduler) +#define __NR_sched_getscheduler 157 +#endif + +#if !defined(__NR_sched_yield) +#define __NR_sched_yield 158 +#endif + +#if !defined(__NR_sched_get_priority_max) +#define __NR_sched_get_priority_max 159 +#endif + +#if !defined(__NR_sched_get_priority_min) +#define __NR_sched_get_priority_min 160 +#endif + +#if !defined(__NR_sched_rr_get_interval) +#define __NR_sched_rr_get_interval 161 +#endif + +#if !defined(__NR_nanosleep) +#define __NR_nanosleep 162 +#endif + +#if !defined(__NR_mremap) +#define __NR_mremap 163 +#endif + +#if !defined(__NR_setresuid) +#define __NR_setresuid 164 +#endif + +#if !defined(__NR_getresuid) +#define __NR_getresuid 165 +#endif + +#if !defined(__NR_vm86) +#define __NR_vm86 166 +#endif + +#if !defined(__NR_query_module) +#define __NR_query_module 167 +#endif + +#if !defined(__NR_poll) +#define __NR_poll 168 +#endif + +#if !defined(__NR_nfsservctl) +#define __NR_nfsservctl 169 +#endif + +#if !defined(__NR_setresgid) +#define __NR_setresgid 170 +#endif + +#if !defined(__NR_getresgid) +#define __NR_getresgid 171 +#endif + +#if !defined(__NR_prctl) +#define __NR_prctl 172 +#endif + +#if !defined(__NR_rt_sigreturn) +#define __NR_rt_sigreturn 173 +#endif + +#if !defined(__NR_rt_sigaction) +#define __NR_rt_sigaction 174 +#endif + +#if !defined(__NR_rt_sigprocmask) +#define __NR_rt_sigprocmask 175 +#endif + +#if !defined(__NR_rt_sigpending) +#define __NR_rt_sigpending 176 +#endif + +#if !defined(__NR_rt_sigtimedwait) +#define __NR_rt_sigtimedwait 177 +#endif + +#if !defined(__NR_rt_sigqueueinfo) +#define __NR_rt_sigqueueinfo 178 +#endif + +#if !defined(__NR_rt_sigsuspend) +#define __NR_rt_sigsuspend 179 +#endif + +#if !defined(__NR_pread64) +#define __NR_pread64 180 +#endif + +#if !defined(__NR_pwrite64) +#define __NR_pwrite64 181 +#endif + +#if !defined(__NR_chown) +#define __NR_chown 182 +#endif + +#if !defined(__NR_getcwd) +#define __NR_getcwd 183 +#endif + +#if !defined(__NR_capget) +#define __NR_capget 184 +#endif + +#if !defined(__NR_capset) +#define __NR_capset 185 +#endif + +#if !defined(__NR_sigaltstack) +#define __NR_sigaltstack 186 +#endif + +#if !defined(__NR_sendfile) +#define __NR_sendfile 187 +#endif + +#if !defined(__NR_getpmsg) +#define __NR_getpmsg 188 +#endif + +#if !defined(__NR_putpmsg) +#define __NR_putpmsg 189 +#endif + +#if !defined(__NR_vfork) +#define __NR_vfork 190 +#endif + +#if !defined(__NR_ugetrlimit) +#define __NR_ugetrlimit 191 +#endif + +#if !defined(__NR_mmap2) +#define __NR_mmap2 192 +#endif + +#if !defined(__NR_truncate64) +#define __NR_truncate64 193 +#endif + +#if !defined(__NR_ftruncate64) +#define __NR_ftruncate64 194 +#endif + +#if !defined(__NR_stat64) +#define __NR_stat64 195 +#endif + +#if !defined(__NR_lstat64) +#define __NR_lstat64 196 +#endif + +#if !defined(__NR_fstat64) +#define __NR_fstat64 197 +#endif + +#if !defined(__NR_lchown32) +#define __NR_lchown32 198 +#endif + +#if !defined(__NR_getuid32) +#define __NR_getuid32 199 +#endif + +#if !defined(__NR_getgid32) +#define __NR_getgid32 200 +#endif + +#if !defined(__NR_geteuid32) +#define __NR_geteuid32 201 +#endif + +#if !defined(__NR_getegid32) +#define __NR_getegid32 202 +#endif + +#if !defined(__NR_setreuid32) +#define __NR_setreuid32 203 +#endif + +#if !defined(__NR_setregid32) +#define __NR_setregid32 204 +#endif + +#if !defined(__NR_getgroups32) +#define __NR_getgroups32 205 +#endif + +#if !defined(__NR_setgroups32) +#define __NR_setgroups32 206 +#endif + +#if !defined(__NR_fchown32) +#define __NR_fchown32 207 +#endif + +#if !defined(__NR_setresuid32) +#define __NR_setresuid32 208 +#endif + +#if !defined(__NR_getresuid32) +#define __NR_getresuid32 209 +#endif + +#if !defined(__NR_setresgid32) +#define __NR_setresgid32 210 +#endif + +#if !defined(__NR_getresgid32) +#define __NR_getresgid32 211 +#endif + +#if !defined(__NR_chown32) +#define __NR_chown32 212 +#endif + +#if !defined(__NR_setuid32) +#define __NR_setuid32 213 +#endif + +#if !defined(__NR_setgid32) +#define __NR_setgid32 214 +#endif + +#if !defined(__NR_setfsuid32) +#define __NR_setfsuid32 215 +#endif + +#if !defined(__NR_setfsgid32) +#define __NR_setfsgid32 216 +#endif + +#if !defined(__NR_pivot_root) +#define __NR_pivot_root 217 +#endif + +#if !defined(__NR_mincore) +#define __NR_mincore 218 +#endif + +#if !defined(__NR_madvise) +#define __NR_madvise 219 +#endif + +#if !defined(__NR_getdents64) +#define __NR_getdents64 220 +#endif + +#if !defined(__NR_fcntl64) +#define __NR_fcntl64 221 +#endif + +#if !defined(__NR_gettid) +#define __NR_gettid 224 +#endif + +#if !defined(__NR_readahead) +#define __NR_readahead 225 +#endif + +#if !defined(__NR_setxattr) +#define __NR_setxattr 226 +#endif + +#if !defined(__NR_lsetxattr) +#define __NR_lsetxattr 227 +#endif + +#if !defined(__NR_fsetxattr) +#define __NR_fsetxattr 228 +#endif + +#if !defined(__NR_getxattr) +#define __NR_getxattr 229 +#endif + +#if !defined(__NR_lgetxattr) +#define __NR_lgetxattr 230 +#endif + +#if !defined(__NR_fgetxattr) +#define __NR_fgetxattr 231 +#endif + +#if !defined(__NR_listxattr) +#define __NR_listxattr 232 +#endif + +#if !defined(__NR_llistxattr) +#define __NR_llistxattr 233 +#endif + +#if !defined(__NR_flistxattr) +#define __NR_flistxattr 234 +#endif + +#if !defined(__NR_removexattr) +#define __NR_removexattr 235 +#endif + +#if !defined(__NR_lremovexattr) +#define __NR_lremovexattr 236 +#endif + +#if !defined(__NR_fremovexattr) +#define __NR_fremovexattr 237 +#endif + +#if !defined(__NR_tkill) +#define __NR_tkill 238 +#endif + +#if !defined(__NR_sendfile64) +#define __NR_sendfile64 239 +#endif + +#if !defined(__NR_futex) +#define __NR_futex 240 +#endif + +#if !defined(__NR_sched_setaffinity) +#define __NR_sched_setaffinity 241 +#endif + +#if !defined(__NR_sched_getaffinity) +#define __NR_sched_getaffinity 242 +#endif + +#if !defined(__NR_set_thread_area) +#define __NR_set_thread_area 243 +#endif + +#if !defined(__NR_get_thread_area) +#define __NR_get_thread_area 244 +#endif + +#if !defined(__NR_io_setup) +#define __NR_io_setup 245 +#endif + +#if !defined(__NR_io_destroy) +#define __NR_io_destroy 246 +#endif + +#if !defined(__NR_io_getevents) +#define __NR_io_getevents 247 +#endif + +#if !defined(__NR_io_submit) +#define __NR_io_submit 248 +#endif + +#if !defined(__NR_io_cancel) +#define __NR_io_cancel 249 +#endif + +#if !defined(__NR_fadvise64) +#define __NR_fadvise64 250 +#endif + +#if !defined(__NR_exit_group) +#define __NR_exit_group 252 +#endif + +#if !defined(__NR_lookup_dcookie) +#define __NR_lookup_dcookie 253 +#endif + +#if !defined(__NR_epoll_create) +#define __NR_epoll_create 254 +#endif + +#if !defined(__NR_epoll_ctl) +#define __NR_epoll_ctl 255 +#endif + +#if !defined(__NR_epoll_wait) +#define __NR_epoll_wait 256 +#endif + +#if !defined(__NR_remap_file_pages) +#define __NR_remap_file_pages 257 +#endif + +#if !defined(__NR_set_tid_address) +#define __NR_set_tid_address 258 +#endif + +#if !defined(__NR_timer_create) +#define __NR_timer_create 259 +#endif + +#if !defined(__NR_timer_settime) +#define __NR_timer_settime 260 +#endif + +#if !defined(__NR_timer_gettime) +#define __NR_timer_gettime 261 +#endif + +#if !defined(__NR_timer_getoverrun) +#define __NR_timer_getoverrun 262 +#endif + +#if !defined(__NR_timer_delete) +#define __NR_timer_delete 263 +#endif + +#if !defined(__NR_clock_settime) +#define __NR_clock_settime 264 +#endif + +#if !defined(__NR_clock_gettime) +#define __NR_clock_gettime 265 +#endif + +#if !defined(__NR_clock_getres) +#define __NR_clock_getres 266 +#endif + +#if !defined(__NR_clock_nanosleep) +#define __NR_clock_nanosleep 267 +#endif + +#if !defined(__NR_statfs64) +#define __NR_statfs64 268 +#endif + +#if !defined(__NR_fstatfs64) +#define __NR_fstatfs64 269 +#endif + +#if !defined(__NR_tgkill) +#define __NR_tgkill 270 +#endif + +#if !defined(__NR_utimes) +#define __NR_utimes 271 +#endif + +#if !defined(__NR_fadvise64_64) +#define __NR_fadvise64_64 272 +#endif + +#if !defined(__NR_vserver) +#define __NR_vserver 273 +#endif + +#if !defined(__NR_mbind) +#define __NR_mbind 274 +#endif + +#if !defined(__NR_get_mempolicy) +#define __NR_get_mempolicy 275 +#endif + +#if !defined(__NR_set_mempolicy) +#define __NR_set_mempolicy 276 +#endif + +#if !defined(__NR_mq_open) +#define __NR_mq_open 277 +#endif + +#if !defined(__NR_mq_unlink) +#define __NR_mq_unlink 278 +#endif + +#if !defined(__NR_mq_timedsend) +#define __NR_mq_timedsend 279 +#endif + +#if !defined(__NR_mq_timedreceive) +#define __NR_mq_timedreceive 280 +#endif + +#if !defined(__NR_mq_notify) +#define __NR_mq_notify 281 +#endif + +#if !defined(__NR_mq_getsetattr) +#define __NR_mq_getsetattr 282 +#endif + +#if !defined(__NR_kexec_load) +#define __NR_kexec_load 283 +#endif + +#if !defined(__NR_waitid) +#define __NR_waitid 284 +#endif + +#if !defined(__NR_add_key) +#define __NR_add_key 286 +#endif + +#if !defined(__NR_request_key) +#define __NR_request_key 287 +#endif + +#if !defined(__NR_keyctl) +#define __NR_keyctl 288 +#endif + +#if !defined(__NR_ioprio_set) +#define __NR_ioprio_set 289 +#endif + +#if !defined(__NR_ioprio_get) +#define __NR_ioprio_get 290 +#endif + +#if !defined(__NR_inotify_init) +#define __NR_inotify_init 291 +#endif + +#if !defined(__NR_inotify_add_watch) +#define __NR_inotify_add_watch 292 +#endif + +#if !defined(__NR_inotify_rm_watch) +#define __NR_inotify_rm_watch 293 +#endif + +#if !defined(__NR_migrate_pages) +#define __NR_migrate_pages 294 +#endif + +#if !defined(__NR_openat) +#define __NR_openat 295 +#endif + +#if !defined(__NR_mkdirat) +#define __NR_mkdirat 296 +#endif + +#if !defined(__NR_mknodat) +#define __NR_mknodat 297 +#endif + +#if !defined(__NR_fchownat) +#define __NR_fchownat 298 +#endif + +#if !defined(__NR_futimesat) +#define __NR_futimesat 299 +#endif + +#if !defined(__NR_fstatat64) +#define __NR_fstatat64 300 +#endif + +#if !defined(__NR_unlinkat) +#define __NR_unlinkat 301 +#endif + +#if !defined(__NR_renameat) +#define __NR_renameat 302 +#endif + +#if !defined(__NR_linkat) +#define __NR_linkat 303 +#endif + +#if !defined(__NR_symlinkat) +#define __NR_symlinkat 304 +#endif + +#if !defined(__NR_readlinkat) +#define __NR_readlinkat 305 +#endif + +#if !defined(__NR_fchmodat) +#define __NR_fchmodat 306 +#endif + +#if !defined(__NR_faccessat) +#define __NR_faccessat 307 +#endif + +#if !defined(__NR_pselect6) +#define __NR_pselect6 308 +#endif + +#if !defined(__NR_ppoll) +#define __NR_ppoll 309 +#endif + +#if !defined(__NR_unshare) +#define __NR_unshare 310 +#endif + +#if !defined(__NR_set_robust_list) +#define __NR_set_robust_list 311 +#endif + +#if !defined(__NR_get_robust_list) +#define __NR_get_robust_list 312 +#endif + +#if !defined(__NR_splice) +#define __NR_splice 313 +#endif + +#if !defined(__NR_sync_file_range) +#define __NR_sync_file_range 314 +#endif + +#if !defined(__NR_tee) +#define __NR_tee 315 +#endif + +#if !defined(__NR_vmsplice) +#define __NR_vmsplice 316 +#endif + +#if !defined(__NR_move_pages) +#define __NR_move_pages 317 +#endif + +#if !defined(__NR_getcpu) +#define __NR_getcpu 318 +#endif + +#if !defined(__NR_epoll_pwait) +#define __NR_epoll_pwait 319 +#endif + +#if !defined(__NR_utimensat) +#define __NR_utimensat 320 +#endif + +#if !defined(__NR_signalfd) +#define __NR_signalfd 321 +#endif + +#if !defined(__NR_timerfd_create) +#define __NR_timerfd_create 322 +#endif + +#if !defined(__NR_eventfd) +#define __NR_eventfd 323 +#endif + +#if !defined(__NR_fallocate) +#define __NR_fallocate 324 +#endif + +#if !defined(__NR_timerfd_settime) +#define __NR_timerfd_settime 325 +#endif + +#if !defined(__NR_timerfd_gettime) +#define __NR_timerfd_gettime 326 +#endif + +#if !defined(__NR_signalfd4) +#define __NR_signalfd4 327 +#endif + +#if !defined(__NR_eventfd2) +#define __NR_eventfd2 328 +#endif + +#if !defined(__NR_epoll_create1) +#define __NR_epoll_create1 329 +#endif + +#if !defined(__NR_dup3) +#define __NR_dup3 330 +#endif + +#if !defined(__NR_pipe2) +#define __NR_pipe2 331 +#endif + +#if !defined(__NR_inotify_init1) +#define __NR_inotify_init1 332 +#endif + +#if !defined(__NR_preadv) +#define __NR_preadv 333 +#endif + +#if !defined(__NR_pwritev) +#define __NR_pwritev 334 +#endif + +#if !defined(__NR_rt_tgsigqueueinfo) +#define __NR_rt_tgsigqueueinfo 335 +#endif + +#if !defined(__NR_perf_event_open) +#define __NR_perf_event_open 336 +#endif + +#if !defined(__NR_recvmmsg) +#define __NR_recvmmsg 337 +#endif + +#if !defined(__NR_fanotify_init) +#define __NR_fanotify_init 338 +#endif + +#if !defined(__NR_fanotify_mark) +#define __NR_fanotify_mark 339 +#endif + +#if !defined(__NR_prlimit64) +#define __NR_prlimit64 340 +#endif + +#if !defined(__NR_name_to_handle_at) +#define __NR_name_to_handle_at 341 +#endif + +#if !defined(__NR_open_by_handle_at) +#define __NR_open_by_handle_at 342 +#endif + +#if !defined(__NR_clock_adjtime) +#define __NR_clock_adjtime 343 +#endif + +#if !defined(__NR_syncfs) +#define __NR_syncfs 344 +#endif + +#if !defined(__NR_sendmmsg) +#define __NR_sendmmsg 345 +#endif + +#if !defined(__NR_setns) +#define __NR_setns 346 +#endif + +#if !defined(__NR_process_vm_readv) +#define __NR_process_vm_readv 347 +#endif + +#if !defined(__NR_process_vm_writev) +#define __NR_process_vm_writev 348 +#endif + +#if !defined(__NR_kcmp) +#define __NR_kcmp 349 +#endif + +#if !defined(__NR_finit_module) +#define __NR_finit_module 350 +#endif + +#if !defined(__NR_sched_setattr) +#define __NR_sched_setattr 351 +#endif + +#if !defined(__NR_sched_getattr) +#define __NR_sched_getattr 352 +#endif + +#if !defined(__NR_renameat2) +#define __NR_renameat2 353 +#endif + +#if !defined(__NR_seccomp) +#define __NR_seccomp 354 +#endif + +#if !defined(__NR_getrandom) +#define __NR_getrandom 355 +#endif + +#if !defined(__NR_memfd_create) +#define __NR_memfd_create 356 +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_X86_32_LINUX_SYSCALLS_H_ + diff --git a/sandbox/linux/system_headers/x86_64_linux_syscalls.h b/sandbox/linux/system_headers/x86_64_linux_syscalls.h new file mode 100644 index 0000000000..349504aee4 --- /dev/null +++ b/sandbox/linux/system_headers/x86_64_linux_syscalls.h @@ -0,0 +1,1294 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// Generated from the Linux kernel's syscall_64.tbl. +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_X86_64_LINUX_SYSCALLS_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_X86_64_LINUX_SYSCALLS_H_ + +#if !defined(__x86_64__) +#error "Including header on wrong architecture" +#endif + +#if !defined(__NR_read) +#define __NR_read 0 +#endif + +#if !defined(__NR_write) +#define __NR_write 1 +#endif + +#if !defined(__NR_open) +#define __NR_open 2 +#endif + +#if !defined(__NR_close) +#define __NR_close 3 +#endif + +#if !defined(__NR_stat) +#define __NR_stat 4 +#endif + +#if !defined(__NR_fstat) +#define __NR_fstat 5 +#endif + +#if !defined(__NR_lstat) +#define __NR_lstat 6 +#endif + +#if !defined(__NR_poll) +#define __NR_poll 7 +#endif + +#if !defined(__NR_lseek) +#define __NR_lseek 8 +#endif + +#if !defined(__NR_mmap) +#define __NR_mmap 9 +#endif + +#if !defined(__NR_mprotect) +#define __NR_mprotect 10 +#endif + +#if !defined(__NR_munmap) +#define __NR_munmap 11 +#endif + +#if !defined(__NR_brk) +#define __NR_brk 12 +#endif + +#if !defined(__NR_rt_sigaction) +#define __NR_rt_sigaction 13 +#endif + +#if !defined(__NR_rt_sigprocmask) +#define __NR_rt_sigprocmask 14 +#endif + +#if !defined(__NR_rt_sigreturn) +#define __NR_rt_sigreturn 15 +#endif + +#if !defined(__NR_ioctl) +#define __NR_ioctl 16 +#endif + +#if !defined(__NR_pread64) +#define __NR_pread64 17 +#endif + +#if !defined(__NR_pwrite64) +#define __NR_pwrite64 18 +#endif + +#if !defined(__NR_readv) +#define __NR_readv 19 +#endif + +#if !defined(__NR_writev) +#define __NR_writev 20 +#endif + +#if !defined(__NR_access) +#define __NR_access 21 +#endif + +#if !defined(__NR_pipe) +#define __NR_pipe 22 +#endif + +#if !defined(__NR_select) +#define __NR_select 23 +#endif + +#if !defined(__NR_sched_yield) +#define __NR_sched_yield 24 +#endif + +#if !defined(__NR_mremap) +#define __NR_mremap 25 +#endif + +#if !defined(__NR_msync) +#define __NR_msync 26 +#endif + +#if !defined(__NR_mincore) +#define __NR_mincore 27 +#endif + +#if !defined(__NR_madvise) +#define __NR_madvise 28 +#endif + +#if !defined(__NR_shmget) +#define __NR_shmget 29 +#endif + +#if !defined(__NR_shmat) +#define __NR_shmat 30 +#endif + +#if !defined(__NR_shmctl) +#define __NR_shmctl 31 +#endif + +#if !defined(__NR_dup) +#define __NR_dup 32 +#endif + +#if !defined(__NR_dup2) +#define __NR_dup2 33 +#endif + +#if !defined(__NR_pause) +#define __NR_pause 34 +#endif + +#if !defined(__NR_nanosleep) +#define __NR_nanosleep 35 +#endif + +#if !defined(__NR_getitimer) +#define __NR_getitimer 36 +#endif + +#if !defined(__NR_alarm) +#define __NR_alarm 37 +#endif + +#if !defined(__NR_setitimer) +#define __NR_setitimer 38 +#endif + +#if !defined(__NR_getpid) +#define __NR_getpid 39 +#endif + +#if !defined(__NR_sendfile) +#define __NR_sendfile 40 +#endif + +#if !defined(__NR_socket) +#define __NR_socket 41 +#endif + +#if !defined(__NR_connect) +#define __NR_connect 42 +#endif + +#if !defined(__NR_accept) +#define __NR_accept 43 +#endif + +#if !defined(__NR_sendto) +#define __NR_sendto 44 +#endif + +#if !defined(__NR_recvfrom) +#define __NR_recvfrom 45 +#endif + +#if !defined(__NR_sendmsg) +#define __NR_sendmsg 46 +#endif + +#if !defined(__NR_recvmsg) +#define __NR_recvmsg 47 +#endif + +#if !defined(__NR_shutdown) +#define __NR_shutdown 48 +#endif + +#if !defined(__NR_bind) +#define __NR_bind 49 +#endif + +#if !defined(__NR_listen) +#define __NR_listen 50 +#endif + +#if !defined(__NR_getsockname) +#define __NR_getsockname 51 +#endif + +#if !defined(__NR_getpeername) +#define __NR_getpeername 52 +#endif + +#if !defined(__NR_socketpair) +#define __NR_socketpair 53 +#endif + +#if !defined(__NR_setsockopt) +#define __NR_setsockopt 54 +#endif + +#if !defined(__NR_getsockopt) +#define __NR_getsockopt 55 +#endif + +#if !defined(__NR_clone) +#define __NR_clone 56 +#endif + +#if !defined(__NR_fork) +#define __NR_fork 57 +#endif + +#if !defined(__NR_vfork) +#define __NR_vfork 58 +#endif + +#if !defined(__NR_execve) +#define __NR_execve 59 +#endif + +#if !defined(__NR_exit) +#define __NR_exit 60 +#endif + +#if !defined(__NR_wait4) +#define __NR_wait4 61 +#endif + +#if !defined(__NR_kill) +#define __NR_kill 62 +#endif + +#if !defined(__NR_uname) +#define __NR_uname 63 +#endif + +#if !defined(__NR_semget) +#define __NR_semget 64 +#endif + +#if !defined(__NR_semop) +#define __NR_semop 65 +#endif + +#if !defined(__NR_semctl) +#define __NR_semctl 66 +#endif + +#if !defined(__NR_shmdt) +#define __NR_shmdt 67 +#endif + +#if !defined(__NR_msgget) +#define __NR_msgget 68 +#endif + +#if !defined(__NR_msgsnd) +#define __NR_msgsnd 69 +#endif + +#if !defined(__NR_msgrcv) +#define __NR_msgrcv 70 +#endif + +#if !defined(__NR_msgctl) +#define __NR_msgctl 71 +#endif + +#if !defined(__NR_fcntl) +#define __NR_fcntl 72 +#endif + +#if !defined(__NR_flock) +#define __NR_flock 73 +#endif + +#if !defined(__NR_fsync) +#define __NR_fsync 74 +#endif + +#if !defined(__NR_fdatasync) +#define __NR_fdatasync 75 +#endif + +#if !defined(__NR_truncate) +#define __NR_truncate 76 +#endif + +#if !defined(__NR_ftruncate) +#define __NR_ftruncate 77 +#endif + +#if !defined(__NR_getdents) +#define __NR_getdents 78 +#endif + +#if !defined(__NR_getcwd) +#define __NR_getcwd 79 +#endif + +#if !defined(__NR_chdir) +#define __NR_chdir 80 +#endif + +#if !defined(__NR_fchdir) +#define __NR_fchdir 81 +#endif + +#if !defined(__NR_rename) +#define __NR_rename 82 +#endif + +#if !defined(__NR_mkdir) +#define __NR_mkdir 83 +#endif + +#if !defined(__NR_rmdir) +#define __NR_rmdir 84 +#endif + +#if !defined(__NR_creat) +#define __NR_creat 85 +#endif + +#if !defined(__NR_link) +#define __NR_link 86 +#endif + +#if !defined(__NR_unlink) +#define __NR_unlink 87 +#endif + +#if !defined(__NR_symlink) +#define __NR_symlink 88 +#endif + +#if !defined(__NR_readlink) +#define __NR_readlink 89 +#endif + +#if !defined(__NR_chmod) +#define __NR_chmod 90 +#endif + +#if !defined(__NR_fchmod) +#define __NR_fchmod 91 +#endif + +#if !defined(__NR_chown) +#define __NR_chown 92 +#endif + +#if !defined(__NR_fchown) +#define __NR_fchown 93 +#endif + +#if !defined(__NR_lchown) +#define __NR_lchown 94 +#endif + +#if !defined(__NR_umask) +#define __NR_umask 95 +#endif + +#if !defined(__NR_gettimeofday) +#define __NR_gettimeofday 96 +#endif + +#if !defined(__NR_getrlimit) +#define __NR_getrlimit 97 +#endif + +#if !defined(__NR_getrusage) +#define __NR_getrusage 98 +#endif + +#if !defined(__NR_sysinfo) +#define __NR_sysinfo 99 +#endif + +#if !defined(__NR_times) +#define __NR_times 100 +#endif + +#if !defined(__NR_ptrace) +#define __NR_ptrace 101 +#endif + +#if !defined(__NR_getuid) +#define __NR_getuid 102 +#endif + +#if !defined(__NR_syslog) +#define __NR_syslog 103 +#endif + +#if !defined(__NR_getgid) +#define __NR_getgid 104 +#endif + +#if !defined(__NR_setuid) +#define __NR_setuid 105 +#endif + +#if !defined(__NR_setgid) +#define __NR_setgid 106 +#endif + +#if !defined(__NR_geteuid) +#define __NR_geteuid 107 +#endif + +#if !defined(__NR_getegid) +#define __NR_getegid 108 +#endif + +#if !defined(__NR_setpgid) +#define __NR_setpgid 109 +#endif + +#if !defined(__NR_getppid) +#define __NR_getppid 110 +#endif + +#if !defined(__NR_getpgrp) +#define __NR_getpgrp 111 +#endif + +#if !defined(__NR_setsid) +#define __NR_setsid 112 +#endif + +#if !defined(__NR_setreuid) +#define __NR_setreuid 113 +#endif + +#if !defined(__NR_setregid) +#define __NR_setregid 114 +#endif + +#if !defined(__NR_getgroups) +#define __NR_getgroups 115 +#endif + +#if !defined(__NR_setgroups) +#define __NR_setgroups 116 +#endif + +#if !defined(__NR_setresuid) +#define __NR_setresuid 117 +#endif + +#if !defined(__NR_getresuid) +#define __NR_getresuid 118 +#endif + +#if !defined(__NR_setresgid) +#define __NR_setresgid 119 +#endif + +#if !defined(__NR_getresgid) +#define __NR_getresgid 120 +#endif + +#if !defined(__NR_getpgid) +#define __NR_getpgid 121 +#endif + +#if !defined(__NR_setfsuid) +#define __NR_setfsuid 122 +#endif + +#if !defined(__NR_setfsgid) +#define __NR_setfsgid 123 +#endif + +#if !defined(__NR_getsid) +#define __NR_getsid 124 +#endif + +#if !defined(__NR_capget) +#define __NR_capget 125 +#endif + +#if !defined(__NR_capset) +#define __NR_capset 126 +#endif + +#if !defined(__NR_rt_sigpending) +#define __NR_rt_sigpending 127 +#endif + +#if !defined(__NR_rt_sigtimedwait) +#define __NR_rt_sigtimedwait 128 +#endif + +#if !defined(__NR_rt_sigqueueinfo) +#define __NR_rt_sigqueueinfo 129 +#endif + +#if !defined(__NR_rt_sigsuspend) +#define __NR_rt_sigsuspend 130 +#endif + +#if !defined(__NR_sigaltstack) +#define __NR_sigaltstack 131 +#endif + +#if !defined(__NR_utime) +#define __NR_utime 132 +#endif + +#if !defined(__NR_mknod) +#define __NR_mknod 133 +#endif + +#if !defined(__NR_uselib) +#define __NR_uselib 134 +#endif + +#if !defined(__NR_personality) +#define __NR_personality 135 +#endif + +#if !defined(__NR_ustat) +#define __NR_ustat 136 +#endif + +#if !defined(__NR_statfs) +#define __NR_statfs 137 +#endif + +#if !defined(__NR_fstatfs) +#define __NR_fstatfs 138 +#endif + +#if !defined(__NR_sysfs) +#define __NR_sysfs 139 +#endif + +#if !defined(__NR_getpriority) +#define __NR_getpriority 140 +#endif + +#if !defined(__NR_setpriority) +#define __NR_setpriority 141 +#endif + +#if !defined(__NR_sched_setparam) +#define __NR_sched_setparam 142 +#endif + +#if !defined(__NR_sched_getparam) +#define __NR_sched_getparam 143 +#endif + +#if !defined(__NR_sched_setscheduler) +#define __NR_sched_setscheduler 144 +#endif + +#if !defined(__NR_sched_getscheduler) +#define __NR_sched_getscheduler 145 +#endif + +#if !defined(__NR_sched_get_priority_max) +#define __NR_sched_get_priority_max 146 +#endif + +#if !defined(__NR_sched_get_priority_min) +#define __NR_sched_get_priority_min 147 +#endif + +#if !defined(__NR_sched_rr_get_interval) +#define __NR_sched_rr_get_interval 148 +#endif + +#if !defined(__NR_mlock) +#define __NR_mlock 149 +#endif + +#if !defined(__NR_munlock) +#define __NR_munlock 150 +#endif + +#if !defined(__NR_mlockall) +#define __NR_mlockall 151 +#endif + +#if !defined(__NR_munlockall) +#define __NR_munlockall 152 +#endif + +#if !defined(__NR_vhangup) +#define __NR_vhangup 153 +#endif + +#if !defined(__NR_modify_ldt) +#define __NR_modify_ldt 154 +#endif + +#if !defined(__NR_pivot_root) +#define __NR_pivot_root 155 +#endif + +#if !defined(__NR__sysctl) +#define __NR__sysctl 156 +#endif + +#if !defined(__NR_prctl) +#define __NR_prctl 157 +#endif + +#if !defined(__NR_arch_prctl) +#define __NR_arch_prctl 158 +#endif + +#if !defined(__NR_adjtimex) +#define __NR_adjtimex 159 +#endif + +#if !defined(__NR_setrlimit) +#define __NR_setrlimit 160 +#endif + +#if !defined(__NR_chroot) +#define __NR_chroot 161 +#endif + +#if !defined(__NR_sync) +#define __NR_sync 162 +#endif + +#if !defined(__NR_acct) +#define __NR_acct 163 +#endif + +#if !defined(__NR_settimeofday) +#define __NR_settimeofday 164 +#endif + +#if !defined(__NR_mount) +#define __NR_mount 165 +#endif + +#if !defined(__NR_umount2) +#define __NR_umount2 166 +#endif + +#if !defined(__NR_swapon) +#define __NR_swapon 167 +#endif + +#if !defined(__NR_swapoff) +#define __NR_swapoff 168 +#endif + +#if !defined(__NR_reboot) +#define __NR_reboot 169 +#endif + +#if !defined(__NR_sethostname) +#define __NR_sethostname 170 +#endif + +#if !defined(__NR_setdomainname) +#define __NR_setdomainname 171 +#endif + +#if !defined(__NR_iopl) +#define __NR_iopl 172 +#endif + +#if !defined(__NR_ioperm) +#define __NR_ioperm 173 +#endif + +#if !defined(__NR_create_module) +#define __NR_create_module 174 +#endif + +#if !defined(__NR_init_module) +#define __NR_init_module 175 +#endif + +#if !defined(__NR_delete_module) +#define __NR_delete_module 176 +#endif + +#if !defined(__NR_get_kernel_syms) +#define __NR_get_kernel_syms 177 +#endif + +#if !defined(__NR_query_module) +#define __NR_query_module 178 +#endif + +#if !defined(__NR_quotactl) +#define __NR_quotactl 179 +#endif + +#if !defined(__NR_nfsservctl) +#define __NR_nfsservctl 180 +#endif + +#if !defined(__NR_getpmsg) +#define __NR_getpmsg 181 +#endif + +#if !defined(__NR_putpmsg) +#define __NR_putpmsg 182 +#endif + +#if !defined(__NR_afs_syscall) +#define __NR_afs_syscall 183 +#endif + +#if !defined(__NR_tuxcall) +#define __NR_tuxcall 184 +#endif + +#if !defined(__NR_security) +#define __NR_security 185 +#endif + +#if !defined(__NR_gettid) +#define __NR_gettid 186 +#endif + +#if !defined(__NR_readahead) +#define __NR_readahead 187 +#endif + +#if !defined(__NR_setxattr) +#define __NR_setxattr 188 +#endif + +#if !defined(__NR_lsetxattr) +#define __NR_lsetxattr 189 +#endif + +#if !defined(__NR_fsetxattr) +#define __NR_fsetxattr 190 +#endif + +#if !defined(__NR_getxattr) +#define __NR_getxattr 191 +#endif + +#if !defined(__NR_lgetxattr) +#define __NR_lgetxattr 192 +#endif + +#if !defined(__NR_fgetxattr) +#define __NR_fgetxattr 193 +#endif + +#if !defined(__NR_listxattr) +#define __NR_listxattr 194 +#endif + +#if !defined(__NR_llistxattr) +#define __NR_llistxattr 195 +#endif + +#if !defined(__NR_flistxattr) +#define __NR_flistxattr 196 +#endif + +#if !defined(__NR_removexattr) +#define __NR_removexattr 197 +#endif + +#if !defined(__NR_lremovexattr) +#define __NR_lremovexattr 198 +#endif + +#if !defined(__NR_fremovexattr) +#define __NR_fremovexattr 199 +#endif + +#if !defined(__NR_tkill) +#define __NR_tkill 200 +#endif + +#if !defined(__NR_time) +#define __NR_time 201 +#endif + +#if !defined(__NR_futex) +#define __NR_futex 202 +#endif + +#if !defined(__NR_sched_setaffinity) +#define __NR_sched_setaffinity 203 +#endif + +#if !defined(__NR_sched_getaffinity) +#define __NR_sched_getaffinity 204 +#endif + +#if !defined(__NR_set_thread_area) +#define __NR_set_thread_area 205 +#endif + +#if !defined(__NR_io_setup) +#define __NR_io_setup 206 +#endif + +#if !defined(__NR_io_destroy) +#define __NR_io_destroy 207 +#endif + +#if !defined(__NR_io_getevents) +#define __NR_io_getevents 208 +#endif + +#if !defined(__NR_io_submit) +#define __NR_io_submit 209 +#endif + +#if !defined(__NR_io_cancel) +#define __NR_io_cancel 210 +#endif + +#if !defined(__NR_get_thread_area) +#define __NR_get_thread_area 211 +#endif + +#if !defined(__NR_lookup_dcookie) +#define __NR_lookup_dcookie 212 +#endif + +#if !defined(__NR_epoll_create) +#define __NR_epoll_create 213 +#endif + +#if !defined(__NR_epoll_ctl_old) +#define __NR_epoll_ctl_old 214 +#endif + +#if !defined(__NR_epoll_wait_old) +#define __NR_epoll_wait_old 215 +#endif + +#if !defined(__NR_remap_file_pages) +#define __NR_remap_file_pages 216 +#endif + +#if !defined(__NR_getdents64) +#define __NR_getdents64 217 +#endif + +#if !defined(__NR_set_tid_address) +#define __NR_set_tid_address 218 +#endif + +#if !defined(__NR_restart_syscall) +#define __NR_restart_syscall 219 +#endif + +#if !defined(__NR_semtimedop) +#define __NR_semtimedop 220 +#endif + +#if !defined(__NR_fadvise64) +#define __NR_fadvise64 221 +#endif + +#if !defined(__NR_timer_create) +#define __NR_timer_create 222 +#endif + +#if !defined(__NR_timer_settime) +#define __NR_timer_settime 223 +#endif + +#if !defined(__NR_timer_gettime) +#define __NR_timer_gettime 224 +#endif + +#if !defined(__NR_timer_getoverrun) +#define __NR_timer_getoverrun 225 +#endif + +#if !defined(__NR_timer_delete) +#define __NR_timer_delete 226 +#endif + +#if !defined(__NR_clock_settime) +#define __NR_clock_settime 227 +#endif + +#if !defined(__NR_clock_gettime) +#define __NR_clock_gettime 228 +#endif + +#if !defined(__NR_clock_getres) +#define __NR_clock_getres 229 +#endif + +#if !defined(__NR_clock_nanosleep) +#define __NR_clock_nanosleep 230 +#endif + +#if !defined(__NR_exit_group) +#define __NR_exit_group 231 +#endif + +#if !defined(__NR_epoll_wait) +#define __NR_epoll_wait 232 +#endif + +#if !defined(__NR_epoll_ctl) +#define __NR_epoll_ctl 233 +#endif + +#if !defined(__NR_tgkill) +#define __NR_tgkill 234 +#endif + +#if !defined(__NR_utimes) +#define __NR_utimes 235 +#endif + +#if !defined(__NR_vserver) +#define __NR_vserver 236 +#endif + +#if !defined(__NR_mbind) +#define __NR_mbind 237 +#endif + +#if !defined(__NR_set_mempolicy) +#define __NR_set_mempolicy 238 +#endif + +#if !defined(__NR_get_mempolicy) +#define __NR_get_mempolicy 239 +#endif + +#if !defined(__NR_mq_open) +#define __NR_mq_open 240 +#endif + +#if !defined(__NR_mq_unlink) +#define __NR_mq_unlink 241 +#endif + +#if !defined(__NR_mq_timedsend) +#define __NR_mq_timedsend 242 +#endif + +#if !defined(__NR_mq_timedreceive) +#define __NR_mq_timedreceive 243 +#endif + +#if !defined(__NR_mq_notify) +#define __NR_mq_notify 244 +#endif + +#if !defined(__NR_mq_getsetattr) +#define __NR_mq_getsetattr 245 +#endif + +#if !defined(__NR_kexec_load) +#define __NR_kexec_load 246 +#endif + +#if !defined(__NR_waitid) +#define __NR_waitid 247 +#endif + +#if !defined(__NR_add_key) +#define __NR_add_key 248 +#endif + +#if !defined(__NR_request_key) +#define __NR_request_key 249 +#endif + +#if !defined(__NR_keyctl) +#define __NR_keyctl 250 +#endif + +#if !defined(__NR_ioprio_set) +#define __NR_ioprio_set 251 +#endif + +#if !defined(__NR_ioprio_get) +#define __NR_ioprio_get 252 +#endif + +#if !defined(__NR_inotify_init) +#define __NR_inotify_init 253 +#endif + +#if !defined(__NR_inotify_add_watch) +#define __NR_inotify_add_watch 254 +#endif + +#if !defined(__NR_inotify_rm_watch) +#define __NR_inotify_rm_watch 255 +#endif + +#if !defined(__NR_migrate_pages) +#define __NR_migrate_pages 256 +#endif + +#if !defined(__NR_openat) +#define __NR_openat 257 +#endif + +#if !defined(__NR_mkdirat) +#define __NR_mkdirat 258 +#endif + +#if !defined(__NR_mknodat) +#define __NR_mknodat 259 +#endif + +#if !defined(__NR_fchownat) +#define __NR_fchownat 260 +#endif + +#if !defined(__NR_futimesat) +#define __NR_futimesat 261 +#endif + +#if !defined(__NR_newfstatat) +#define __NR_newfstatat 262 +#endif + +#if !defined(__NR_unlinkat) +#define __NR_unlinkat 263 +#endif + +#if !defined(__NR_renameat) +#define __NR_renameat 264 +#endif + +#if !defined(__NR_linkat) +#define __NR_linkat 265 +#endif + +#if !defined(__NR_symlinkat) +#define __NR_symlinkat 266 +#endif + +#if !defined(__NR_readlinkat) +#define __NR_readlinkat 267 +#endif + +#if !defined(__NR_fchmodat) +#define __NR_fchmodat 268 +#endif + +#if !defined(__NR_faccessat) +#define __NR_faccessat 269 +#endif + +#if !defined(__NR_pselect6) +#define __NR_pselect6 270 +#endif + +#if !defined(__NR_ppoll) +#define __NR_ppoll 271 +#endif + +#if !defined(__NR_unshare) +#define __NR_unshare 272 +#endif + +#if !defined(__NR_set_robust_list) +#define __NR_set_robust_list 273 +#endif + +#if !defined(__NR_get_robust_list) +#define __NR_get_robust_list 274 +#endif + +#if !defined(__NR_splice) +#define __NR_splice 275 +#endif + +#if !defined(__NR_tee) +#define __NR_tee 276 +#endif + +#if !defined(__NR_sync_file_range) +#define __NR_sync_file_range 277 +#endif + +#if !defined(__NR_vmsplice) +#define __NR_vmsplice 278 +#endif + +#if !defined(__NR_move_pages) +#define __NR_move_pages 279 +#endif + +#if !defined(__NR_utimensat) +#define __NR_utimensat 280 +#endif + +#if !defined(__NR_epoll_pwait) +#define __NR_epoll_pwait 281 +#endif + +#if !defined(__NR_signalfd) +#define __NR_signalfd 282 +#endif + +#if !defined(__NR_timerfd_create) +#define __NR_timerfd_create 283 +#endif + +#if !defined(__NR_eventfd) +#define __NR_eventfd 284 +#endif + +#if !defined(__NR_fallocate) +#define __NR_fallocate 285 +#endif + +#if !defined(__NR_timerfd_settime) +#define __NR_timerfd_settime 286 +#endif + +#if !defined(__NR_timerfd_gettime) +#define __NR_timerfd_gettime 287 +#endif + +#if !defined(__NR_accept4) +#define __NR_accept4 288 +#endif + +#if !defined(__NR_signalfd4) +#define __NR_signalfd4 289 +#endif + +#if !defined(__NR_eventfd2) +#define __NR_eventfd2 290 +#endif + +#if !defined(__NR_epoll_create1) +#define __NR_epoll_create1 291 +#endif + +#if !defined(__NR_dup3) +#define __NR_dup3 292 +#endif + +#if !defined(__NR_pipe2) +#define __NR_pipe2 293 +#endif + +#if !defined(__NR_inotify_init1) +#define __NR_inotify_init1 294 +#endif + +#if !defined(__NR_preadv) +#define __NR_preadv 295 +#endif + +#if !defined(__NR_pwritev) +#define __NR_pwritev 296 +#endif + +#if !defined(__NR_rt_tgsigqueueinfo) +#define __NR_rt_tgsigqueueinfo 297 +#endif + +#if !defined(__NR_perf_event_open) +#define __NR_perf_event_open 298 +#endif + +#if !defined(__NR_recvmmsg) +#define __NR_recvmmsg 299 +#endif + +#if !defined(__NR_fanotify_init) +#define __NR_fanotify_init 300 +#endif + +#if !defined(__NR_fanotify_mark) +#define __NR_fanotify_mark 301 +#endif + +#if !defined(__NR_prlimit64) +#define __NR_prlimit64 302 +#endif + +#if !defined(__NR_name_to_handle_at) +#define __NR_name_to_handle_at 303 +#endif + +#if !defined(__NR_open_by_handle_at) +#define __NR_open_by_handle_at 304 +#endif + +#if !defined(__NR_clock_adjtime) +#define __NR_clock_adjtime 305 +#endif + +#if !defined(__NR_syncfs) +#define __NR_syncfs 306 +#endif + +#if !defined(__NR_sendmmsg) +#define __NR_sendmmsg 307 +#endif + +#if !defined(__NR_setns) +#define __NR_setns 308 +#endif + +#if !defined(__NR_getcpu) +#define __NR_getcpu 309 +#endif + +#if !defined(__NR_process_vm_readv) +#define __NR_process_vm_readv 310 +#endif + +#if !defined(__NR_process_vm_writev) +#define __NR_process_vm_writev 311 +#endif + +#if !defined(__NR_kcmp) +#define __NR_kcmp 312 +#endif + +#if !defined(__NR_finit_module) +#define __NR_finit_module 313 +#endif + +#if !defined(__NR_sched_setattr) +#define __NR_sched_setattr 314 +#endif + +#if !defined(__NR_sched_getattr) +#define __NR_sched_getattr 315 +#endif + +#if !defined(__NR_renameat2) +#define __NR_renameat2 316 +#endif + +#if !defined(__NR_seccomp) +#define __NR_seccomp 317 +#endif + +#if !defined(__NR_getrandom) +#define __NR_getrandom 318 +#endif + +#if !defined(__NR_memfd_create) +#define __NR_memfd_create 319 +#endif + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_X86_64_LINUX_SYSCALLS_H_ + diff --git a/sandbox/linux/system_headers/x86_64_linux_ucontext.h b/sandbox/linux/system_headers/x86_64_linux_ucontext.h new file mode 100644 index 0000000000..57b8919a9c --- /dev/null +++ b/sandbox/linux/system_headers/x86_64_linux_ucontext.h @@ -0,0 +1,88 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_SYSTEM_HEADERS_X86_64_LINUX_UCONTEXT_H_ +#define SANDBOX_LINUX_SYSTEM_HEADERS_X86_64_LINUX_UCONTEXT_H_ + +// We do something compatible with glibc. Hopefully, at some point Android will +// provide that for us, and __BIONIC_HAVE_UCONTEXT_T should be defined. +// Spec: +// http://refspecs.linuxfoundation.org/LSB_4.1.0/LSB-Core-AMD64/LSB-Core-AMD64/libc-ddefs.html#AEN5668 + +#if !defined(__BIONIC_HAVE_UCONTEXT_T) +#include <asm/sigcontext.h> + +struct _libc_fpxreg { + unsigned short significand[4]; + unsigned short exponent; + unsigned short padding[3]; +}; + +struct _libc_xmmreg { + uint32_t element[4]; +}; + +struct _libc_fpstate { + uint16_t cwd; + uint16_t swd; + uint16_t twd; + uint16_t fop; + uint64_t rip; + uint64_t rdp; + uint32_t mxcsr; + uint32_t mxcsr_mask; + struct _libc_fpxreg _st[8]; + struct _libc_xmmreg _xmm[16]; + uint32_t padding[24]; +}; + +typedef uint64_t greg_t; + +typedef struct { + greg_t gregs[23]; + struct _libc_fpstate* fpregs; + unsigned long __reserved1[8]; +} mcontext_t; + +enum { + REG_R8 = 0, + REG_R9, + REG_R10, + REG_R11, + REG_R12, + REG_R13, + REG_R14, + REG_R15, + REG_RDI, + REG_RSI, + REG_RBP, + REG_RBX, + REG_RDX, + REG_RAX, + REG_RCX, + REG_RSP, + REG_RIP, + REG_EFL, + REG_CSGSFS, + REG_ERR, + REG_TRAPNO, + REG_OLDMASK, + REG_CR2, + NGREG, +}; + +typedef struct ucontext { + unsigned long uc_flags; + struct ucontext* uc_link; + stack_t uc_stack; + mcontext_t uc_mcontext; + sigset_t uc_sigmask; + struct _libc_fpstate __fpregs_mem; +} ucontext_t; + +#else +#include <sys/ucontext.h> +#endif // __BIONIC_HAVE_UCONTEXT_T + +#endif // SANDBOX_LINUX_SYSTEM_HEADERS_X86_64_LINUX_UCONTEXT_H_ diff --git a/sandbox/linux/tests/main.cc b/sandbox/linux/tests/main.cc new file mode 100644 index 0000000000..caeddee32c --- /dev/null +++ b/sandbox/linux/tests/main.cc @@ -0,0 +1,82 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "base/at_exit.h" +#include "base/base_switches.h" +#include "base/command_line.h" +#include "base/files/file_path.h" +#include "base/files/file_util.h" +#include "base/logging.h" +#include "base/test/test_suite.h" +#include "build/build_config.h" +#include "sandbox/linux/tests/test_utils.h" +#include "sandbox/linux/tests/unit_tests.h" +#include "testing/gtest/include/gtest/gtest.h" +#include "testing/multiprocess_func_list.h" + +namespace sandbox { +namespace { + +// Check for leaks in our tests. +void RunPostTestsChecks(const base::FilePath& orig_cwd) { + if (TestUtils::CurrentProcessHasChildren()) { + LOG(FATAL) << "One of the tests created a child that was not waited for. " + << "Please, clean up after your tests!"; + } + + base::FilePath cwd; + CHECK(GetCurrentDirectory(&cwd)); + if (orig_cwd != cwd) { + LOG(FATAL) << "One of the tests changed the current working directory. " + << "Please, clean up after your tests!"; + } +} + +} // namespace +} // namespace sandbox + +#if !defined(SANDBOX_USES_BASE_TEST_SUITE) +void UnitTestAssertHandler(const std::string& str) { + _exit(1); +} +#endif + +int main(int argc, char* argv[]) { + base::CommandLine::Init(argc, argv); + std::string client_func; +#if defined(SANDBOX_USES_BASE_TEST_SUITE) + client_func = base::CommandLine::ForCurrentProcess()->GetSwitchValueASCII( + switches::kTestChildProcess); +#endif + if (!client_func.empty()) { + base::AtExitManager exit_manager; + return multi_process_function_list::InvokeChildProcessTest(client_func); + } + + base::FilePath orig_cwd; + CHECK(GetCurrentDirectory(&orig_cwd)); + +#if !defined(SANDBOX_USES_BASE_TEST_SUITE) + // The use of Callbacks requires an AtExitManager. + base::AtExitManager exit_manager; + testing::InitGoogleTest(&argc, argv); + // Death tests rely on LOG(FATAL) triggering an exit (the default behavior is + // SIGABRT). The normal test launcher does this at initialization, but since + // we still do not use this on Android, we must install the handler ourselves. + logging::SetLogAssertHandler(UnitTestAssertHandler); +#endif + // Always go through re-execution for death tests. + // This makes gtest only marginally slower for us and has the + // additional side effect of getting rid of gtest warnings about fork() + // safety. + ::testing::FLAGS_gtest_death_test_style = "threadsafe"; +#if !defined(SANDBOX_USES_BASE_TEST_SUITE) + int tests_result = RUN_ALL_TESTS(); +#else + int tests_result = base::RunUnitTestsUsingBaseTestSuite(argc, argv); +#endif + + sandbox::RunPostTestsChecks(orig_cwd); + return tests_result; +} diff --git a/sandbox/linux/tests/sandbox_test_runner.cc b/sandbox/linux/tests/sandbox_test_runner.cc new file mode 100644 index 0000000000..b099b97289 --- /dev/null +++ b/sandbox/linux/tests/sandbox_test_runner.cc @@ -0,0 +1,19 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/tests/sandbox_test_runner.h" + +namespace sandbox { + +SandboxTestRunner::SandboxTestRunner() { +} + +SandboxTestRunner::~SandboxTestRunner() { +} + +bool SandboxTestRunner::ShouldCheckForLeaks() const { + return true; +} + +} // namespace sandbox diff --git a/sandbox/linux/tests/sandbox_test_runner.h b/sandbox/linux/tests/sandbox_test_runner.h new file mode 100644 index 0000000000..3155b74008 --- /dev/null +++ b/sandbox/linux/tests/sandbox_test_runner.h @@ -0,0 +1,30 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_TESTS_SANDBOX_TEST_RUNNER_H_ +#define SANDBOX_LINUX_TESTS_SANDBOX_TEST_RUNNER_H_ + +#include "base/macros.h" + +namespace sandbox { + +// A simple "runner" class to implement tests. +class SandboxTestRunner { + public: + SandboxTestRunner(); + virtual ~SandboxTestRunner(); + + virtual void Run() = 0; + + // Override to decide whether or not to check for leaks with LSAN + // (if built with LSAN and LSAN is enabled). + virtual bool ShouldCheckForLeaks() const; + + private: + DISALLOW_COPY_AND_ASSIGN(SandboxTestRunner); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_TESTS_SANDBOX_TEST_RUNNER_H_ diff --git a/sandbox/linux/tests/sandbox_test_runner_function_pointer.cc b/sandbox/linux/tests/sandbox_test_runner_function_pointer.cc new file mode 100644 index 0000000000..69e05ac4e0 --- /dev/null +++ b/sandbox/linux/tests/sandbox_test_runner_function_pointer.cc @@ -0,0 +1,25 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/tests/sandbox_test_runner_function_pointer.h" + +#include "base/logging.h" +#include "build/build_config.h" + +namespace sandbox { + +SandboxTestRunnerFunctionPointer::SandboxTestRunnerFunctionPointer( + void (*function_to_run)(void)) + : function_to_run_(function_to_run) { +} + +SandboxTestRunnerFunctionPointer::~SandboxTestRunnerFunctionPointer() { +} + +void SandboxTestRunnerFunctionPointer::Run() { + DCHECK(function_to_run_); + function_to_run_(); +} + +} // namespace sandbox diff --git a/sandbox/linux/tests/sandbox_test_runner_function_pointer.h b/sandbox/linux/tests/sandbox_test_runner_function_pointer.h new file mode 100644 index 0000000000..cadd07c248 --- /dev/null +++ b/sandbox/linux/tests/sandbox_test_runner_function_pointer.h @@ -0,0 +1,26 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_TESTS_SANDBOX_TEST_RUNNER_FUNCTION_POINTER_H_ +#define SANDBOX_LINUX_TESTS_SANDBOX_TEST_RUNNER_FUNCTION_POINTER_H_ + +#include "base/macros.h" +#include "sandbox/linux/tests/sandbox_test_runner.h" + +namespace sandbox { + +class SandboxTestRunnerFunctionPointer : public SandboxTestRunner { + public: + SandboxTestRunnerFunctionPointer(void (*function_to_run)(void)); + ~SandboxTestRunnerFunctionPointer() override; + void Run() override; + + private: + void (*function_to_run_)(void); + DISALLOW_COPY_AND_ASSIGN(SandboxTestRunnerFunctionPointer); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_TESTS_SANDBOX_TEST_RUNNER__FUNCTION_POINTER_H_ diff --git a/sandbox/linux/tests/scoped_temporary_file.cc b/sandbox/linux/tests/scoped_temporary_file.cc new file mode 100644 index 0000000000..1f2d66fd6b --- /dev/null +++ b/sandbox/linux/tests/scoped_temporary_file.cc @@ -0,0 +1,35 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/tests/scoped_temporary_file.h" + +#include <stdlib.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/macros.h" +#include "base/posix/eintr_wrapper.h" +#include "build/build_config.h" + +namespace sandbox { + +ScopedTemporaryFile::ScopedTemporaryFile() : fd_(-1) { +#if defined(OS_ANDROID) + static const char file_template[] = "/data/local/tmp/ScopedTempFileXXXXXX"; +#else + static const char file_template[] = "/tmp/ScopedTempFileXXXXXX"; +#endif // defined(OS_ANDROID) + static_assert(sizeof(full_file_name_) >= sizeof(file_template), + "full_file_name is not large enough"); + memcpy(full_file_name_, file_template, sizeof(file_template)); + fd_ = mkstemp(full_file_name_); + CHECK_LE(0, fd_); +} + +ScopedTemporaryFile::~ScopedTemporaryFile() { + CHECK_EQ(0, unlink(full_file_name_)); + CHECK_EQ(0, IGNORE_EINTR(close(fd_))); +} + +} // namespace sandbox diff --git a/sandbox/linux/tests/scoped_temporary_file.h b/sandbox/linux/tests/scoped_temporary_file.h new file mode 100644 index 0000000000..0734130055 --- /dev/null +++ b/sandbox/linux/tests/scoped_temporary_file.h @@ -0,0 +1,30 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_TESTS_SCOPED_TEMPORARY_FILE_H_ +#define SANDBOX_LINUX_TESTS_SCOPED_TEMPORARY_FILE_H_ + +#include "base/macros.h" + +namespace sandbox { +// Creates and open a temporary file on creation and closes +// and removes it on destruction. +// Unlike base/ helpers, this does not require JNI on Android. +class ScopedTemporaryFile { + public: + ScopedTemporaryFile(); + ~ScopedTemporaryFile(); + + int fd() const { return fd_; } + const char* full_file_name() const { return full_file_name_; } + + private: + int fd_; + char full_file_name_[128]; + DISALLOW_COPY_AND_ASSIGN(ScopedTemporaryFile); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_TESTS_SCOPED_TEMPORARY_FILE_H_ diff --git a/sandbox/linux/tests/scoped_temporary_file_unittest.cc b/sandbox/linux/tests/scoped_temporary_file_unittest.cc new file mode 100644 index 0000000000..44a2ecb1ae --- /dev/null +++ b/sandbox/linux/tests/scoped_temporary_file_unittest.cc @@ -0,0 +1,76 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/tests/scoped_temporary_file.h" + +#include <errno.h> +#include <fcntl.h> +#include <stdlib.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> + +#include "base/files/scoped_file.h" +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +bool FullWrite(int fd, const char* buffer, size_t count) { + while (count > 0) { + const ssize_t transfered = HANDLE_EINTR(write(fd, buffer, count)); + if (transfered <= 0 || static_cast<size_t>(transfered) > count) { + return false; + } + count -= transfered; + buffer += transfered; + } + return true; +} + +bool FullRead(int fd, char* buffer, size_t count) { + while (count > 0) { + const ssize_t transfered = HANDLE_EINTR(read(fd, buffer, count)); + if (transfered <= 0 || static_cast<size_t>(transfered) > count) { + return false; + } + count -= transfered; + buffer += transfered; + } + return true; +} + +TEST(ScopedTemporaryFile, Basics) { + std::string temp_file_name; + { + ScopedTemporaryFile temp_file_1; + const char kTestString[] = "This is a test"; + ASSERT_LE(0, temp_file_1.fd()); + + temp_file_name = temp_file_1.full_file_name(); + base::ScopedFD temp_file_2(open(temp_file_1.full_file_name(), O_RDONLY)); + ASSERT_TRUE(temp_file_2.is_valid()); + + ASSERT_TRUE(FullWrite(temp_file_1.fd(), kTestString, sizeof(kTestString))); + + char test_string_read[sizeof(kTestString)] = {0}; + ASSERT_TRUE(FullRead( + temp_file_2.get(), test_string_read, sizeof(test_string_read))); + ASSERT_EQ(0, memcmp(kTestString, test_string_read, sizeof(kTestString))); + } + + errno = 0; + struct stat buf; + ASSERT_EQ(-1, stat(temp_file_name.c_str(), &buf)); + ASSERT_EQ(ENOENT, errno); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/tests/test_utils.cc b/sandbox/linux/tests/test_utils.cc new file mode 100644 index 0000000000..747bad27a5 --- /dev/null +++ b/sandbox/linux/tests/test_utils.cc @@ -0,0 +1,42 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/tests/test_utils.h" + +#include <errno.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" + +namespace sandbox { + +bool TestUtils::CurrentProcessHasChildren() { + siginfo_t process_info; + int ret = HANDLE_EINTR( + waitid(P_ALL, 0, &process_info, WEXITED | WNOHANG | WNOWAIT)); + if (-1 == ret) { + PCHECK(ECHILD == errno); + return false; + } else { + return true; + } +} + +void TestUtils::HandlePostForkReturn(pid_t pid) { + const int kChildExitCode = 1; + if (pid > 0) { + int status = 0; + PCHECK(pid == HANDLE_EINTR(waitpid(pid, &status, 0))); + CHECK(WIFEXITED(status)); + CHECK_EQ(kChildExitCode, WEXITSTATUS(status)); + } else if (pid == 0) { + _exit(kChildExitCode); + } +} + +} // namespace sandbox diff --git a/sandbox/linux/tests/test_utils.h b/sandbox/linux/tests/test_utils.h new file mode 100644 index 0000000000..7cf9749fe4 --- /dev/null +++ b/sandbox/linux/tests/test_utils.h @@ -0,0 +1,29 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_TESTS_TEST_UTILS_H_ +#define SANDBOX_LINUX_TESTS_TEST_UTILS_H_ + +#include <sys/types.h> + +#include "base/macros.h" + +namespace sandbox { + +// This class provide small helpers to help writing tests. +class TestUtils { + public: + static bool CurrentProcessHasChildren(); + // |pid| is the return value of a fork()-like call. This + // makes sure that if fork() succeeded the child exits + // and the parent waits for it. + static void HandlePostForkReturn(pid_t pid); + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(TestUtils); +}; + +} // namespace sandbox + +#endif // SANDBOX_LINUX_TESTS_TEST_UTILS_H_ diff --git a/sandbox/linux/tests/test_utils_unittest.cc b/sandbox/linux/tests/test_utils_unittest.cc new file mode 100644 index 0000000000..0f86e616e9 --- /dev/null +++ b/sandbox/linux/tests/test_utils_unittest.cc @@ -0,0 +1,24 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "sandbox/linux/tests/test_utils.h" + +#include <sys/types.h> +#include <unistd.h> + +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +namespace { + +// Check that HandlePostForkReturn works. +TEST(TestUtils, HandlePostForkReturn) { + pid_t pid = fork(); + TestUtils::HandlePostForkReturn(pid); +} + +} // namespace + +} // namespace sandbox diff --git a/sandbox/linux/tests/unit_tests.cc b/sandbox/linux/tests/unit_tests.cc new file mode 100644 index 0000000000..4973c41fbd --- /dev/null +++ b/sandbox/linux/tests/unit_tests.cc @@ -0,0 +1,354 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include <fcntl.h> +#include <poll.h> +#include <signal.h> +#include <stdio.h> +#include <sys/resource.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <sys/time.h> +#include <time.h> +#include <unistd.h> + +#include "base/debug/leak_annotations.h" +#include "base/files/file_util.h" +#include "base/posix/eintr_wrapper.h" +#include "base/third_party/valgrind/valgrind.h" +#include "build/build_config.h" +#include "sandbox/linux/tests/unit_tests.h" + +// Specifically, PNaCl toolchain does not have this flag. +#if !defined(POLLRDHUP) +#define POLLRDHUP 0x2000 +#endif + +namespace { +std::string TestFailedMessage(const std::string& msg) { + return msg.empty() ? std::string() : "Actual test failure: " + msg; +} + +int GetSubProcessTimeoutTimeInSeconds() { + // Previously 10s, but that timed out (just) on Chromecast. + return 12; +} + +// Returns the number of threads of the current process or -1. +int CountThreads() { + struct stat task_stat; + int task_d = stat("/proc/self/task", &task_stat); + // task_stat.st_nlink should be the number of tasks + 2 (accounting for + // "." and "..". + if (task_d != 0 || task_stat.st_nlink < 3) + return -1; + const int num_threads = task_stat.st_nlink - 2; + return num_threads; +} + +} // namespace + +namespace sandbox { + +bool IsAndroid() { +#if defined(OS_ANDROID) + return true; +#else + return false; +#endif +} + +bool IsArchitectureArm() { +#if defined(ARCH_CPU_ARM_FAMILY) + return true; +#else + return false; +#endif +} + +// TODO(jln): figure out why base/.../dynamic_annotations.h's +// RunningOnValgrind() cannot link. +bool IsRunningOnValgrind() { return RUNNING_ON_VALGRIND; } + +static const int kExpectedValue = 42; +static const int kIgnoreThisTest = 43; +static const int kExitWithAssertionFailure = 1; +#if !defined(OS_NACL_NONSFI) +static const int kExitForTimeout = 2; +#endif + +#if defined(SANDBOX_USES_BASE_TEST_SUITE) +// This is due to StackDumpSignalHandler() performing _exit(1). +// TODO(jln): get rid of the collision with kExitWithAssertionFailure. +const int kExitAfterSIGSEGV = 1; +#endif + +// PNaCl toolchain's signal ABIs are incompatible with Linux's. +// So, for simplicity, just drop the "timeout" feature from unittest framework +// with relying on the buildbot's timeout feature. +#if !defined(OS_NACL_NONSFI) +static void SigAlrmHandler(int) { + const char failure_message[] = "Timeout reached!\n"; + // Make sure that we never block here. + if (!fcntl(2, F_SETFL, O_NONBLOCK)) { + ignore_result(write(2, failure_message, sizeof(failure_message) - 1)); + } + _exit(kExitForTimeout); +} + +// Set a timeout with a handler that will automatically fail the +// test. +static void SetProcessTimeout(int time_in_seconds) { + struct sigaction act = {}; + act.sa_handler = SigAlrmHandler; + SANDBOX_ASSERT(sigemptyset(&act.sa_mask) == 0); + act.sa_flags = 0; + + struct sigaction old_act; + SANDBOX_ASSERT(sigaction(SIGALRM, &act, &old_act) == 0); + + // We don't implemenet signal chaining, so make sure that nothing else + // is expecting to handle SIGALRM. + SANDBOX_ASSERT((old_act.sa_flags & SA_SIGINFO) == 0); + SANDBOX_ASSERT(old_act.sa_handler == SIG_DFL); + sigset_t sigalrm_set; + SANDBOX_ASSERT(sigemptyset(&sigalrm_set) == 0); + SANDBOX_ASSERT(sigaddset(&sigalrm_set, SIGALRM) == 0); + SANDBOX_ASSERT(sigprocmask(SIG_UNBLOCK, &sigalrm_set, NULL) == 0); + SANDBOX_ASSERT(alarm(time_in_seconds) == 0); // There should be no previous + // alarm. +} +#endif // !defined(OS_NACL_NONSFI) + +// Runs a test in a sub-process. This is necessary for most of the code +// in the BPF sandbox, as it potentially makes global state changes and as +// it also tends to raise fatal errors, if the code has been used in an +// insecure manner. +void UnitTests::RunTestInProcess(SandboxTestRunner* test_runner, + DeathCheck death, + const void* death_aux) { + CHECK(test_runner); + // We need to fork(), so we can't be multi-threaded, as threads could hold + // locks. + int num_threads = CountThreads(); +#if !defined(THREAD_SANITIZER) + const int kNumExpectedThreads = 1; +#else + // Under TSAN, there is a special helper thread. It should be completely + // invisible to our testing, so we ignore it. It should be ok to fork() + // with this thread. It's currently buggy, but it's the best we can do until + // there is a way to delay the start of the thread + // (https://code.google.com/p/thread-sanitizer/issues/detail?id=19). + const int kNumExpectedThreads = 2; +#endif + + // The kernel is at liberty to wake a thread id futex before updating /proc. + // If another test running in the same process has stopped a thread, it may + // appear as still running in /proc. + // We poll /proc, with an exponential back-off. At most, we'll sleep around + // 2^iterations nanoseconds in nanosleep(). + for (unsigned int iteration = 0; iteration < 30; iteration++) { + struct timespec ts = {0, 1L << iteration /* nanoseconds */}; + PCHECK(0 == HANDLE_EINTR(nanosleep(&ts, &ts))); + num_threads = CountThreads(); + if (kNumExpectedThreads == num_threads) + break; + } + + ASSERT_EQ(kNumExpectedThreads, num_threads) + << "Running sandbox tests with multiple threads " + << "is not supported and will make the tests flaky."; + int fds[2]; + ASSERT_EQ(0, pipe(fds)); + // Check that our pipe is not on one of the standard file descriptor. + SANDBOX_ASSERT(fds[0] > 2 && fds[1] > 2); + + pid_t pid; + ASSERT_LE(0, (pid = fork())); + if (!pid) { + // In child process + // Redirect stderr to our pipe. This way, we can capture all error + // messages, if we decide we want to do so in our tests. + SANDBOX_ASSERT(dup2(fds[1], 2) == 2); + SANDBOX_ASSERT(!close(fds[0])); + SANDBOX_ASSERT(!close(fds[1])); + + // Don't set a timeout if running on Valgrind, since it's generally much + // slower. + if (!IsRunningOnValgrind()) { +#if !defined(OS_NACL_NONSFI) + SetProcessTimeout(GetSubProcessTimeoutTimeInSeconds()); +#endif + } + + // Disable core files. They are not very useful for our individual test + // cases. + struct rlimit no_core = {0}; + setrlimit(RLIMIT_CORE, &no_core); + + test_runner->Run(); + if (test_runner->ShouldCheckForLeaks()) { +#if defined(LEAK_SANITIZER) + __lsan_do_leak_check(); +#endif + } + _exit(kExpectedValue); + } + + close(fds[1]); + std::vector<char> msg_buf; + ssize_t rc; + + // Make sure read() will never block as we'll use poll() to + // block with a timeout instead. + const int fcntl_ret = fcntl(fds[0], F_SETFL, O_NONBLOCK); + ASSERT_EQ(0, fcntl_ret); + struct pollfd poll_fd = {fds[0], POLLIN | POLLRDHUP, 0}; + + int poll_ret; + // We prefer the SIGALRM timeout to trigger in the child than this timeout + // so we double the common value here. + int poll_timeout = GetSubProcessTimeoutTimeInSeconds() * 2 * 1000; + while ((poll_ret = poll(&poll_fd, 1, poll_timeout) > 0)) { + const size_t kCapacity = 256; + const size_t len = msg_buf.size(); + msg_buf.resize(len + kCapacity); + rc = HANDLE_EINTR(read(fds[0], &msg_buf[len], kCapacity)); + msg_buf.resize(len + std::max(rc, static_cast<ssize_t>(0))); + if (rc <= 0) + break; + } + ASSERT_NE(poll_ret, -1) << "poll() failed"; + ASSERT_NE(poll_ret, 0) << "Timeout while reading child state"; + close(fds[0]); + std::string msg(msg_buf.begin(), msg_buf.end()); + + int status = 0; + int waitpid_returned = HANDLE_EINTR(waitpid(pid, &status, 0)); + ASSERT_EQ(pid, waitpid_returned) << TestFailedMessage(msg); + + // At run-time, we sometimes decide that a test shouldn't actually + // run (e.g. when testing sandbox features on a kernel that doesn't + // have sandboxing support). When that happens, don't attempt to + // call the "death" function, as it might be looking for a + // death-test condition that would never have triggered. + if (!WIFEXITED(status) || WEXITSTATUS(status) != kIgnoreThisTest || + !msg.empty()) { + // We use gtest's ASSERT_XXX() macros instead of the DeathCheck + // functions. This means, on failure, "return" is called. This + // only works correctly, if the call of the "death" callback is + // the very last thing in our function. + death(status, msg, death_aux); + } +} + +void UnitTests::DeathSuccess(int status, const std::string& msg, const void*) { + std::string details(TestFailedMessage(msg)); + + bool subprocess_terminated_normally = WIFEXITED(status); + ASSERT_TRUE(subprocess_terminated_normally) << details; + int subprocess_exit_status = WEXITSTATUS(status); + ASSERT_EQ(kExpectedValue, subprocess_exit_status) << details; + bool subprocess_exited_but_printed_messages = !msg.empty(); + EXPECT_FALSE(subprocess_exited_but_printed_messages) << details; +} + +void UnitTests::DeathSuccessAllowNoise(int status, + const std::string& msg, + const void*) { + std::string details(TestFailedMessage(msg)); + + bool subprocess_terminated_normally = WIFEXITED(status); + ASSERT_TRUE(subprocess_terminated_normally) << details; + int subprocess_exit_status = WEXITSTATUS(status); + ASSERT_EQ(kExpectedValue, subprocess_exit_status) << details; +} + +void UnitTests::DeathMessage(int status, + const std::string& msg, + const void* aux) { + std::string details(TestFailedMessage(msg)); + const char* expected_msg = static_cast<const char*>(aux); + + bool subprocess_terminated_normally = WIFEXITED(status); + ASSERT_TRUE(subprocess_terminated_normally) << "Exit status: " << status + << " " << details; + int subprocess_exit_status = WEXITSTATUS(status); + ASSERT_EQ(1, subprocess_exit_status) << details; + + bool subprocess_exited_without_matching_message = + msg.find(expected_msg) == std::string::npos; + +// In official builds CHECK messages are dropped, so look for SIGABRT. +// See https://code.google.com/p/chromium/issues/detail?id=437312 +#if defined(OFFICIAL_BUILD) && defined(NDEBUG) && !defined(OS_ANDROID) + if (subprocess_exited_without_matching_message) { + static const char kSigAbortMessage[] = "Received signal 6"; + subprocess_exited_without_matching_message = + msg.find(kSigAbortMessage) == std::string::npos; + } +#endif + EXPECT_FALSE(subprocess_exited_without_matching_message) << details; +} + +void UnitTests::DeathSEGVMessage(int status, + const std::string& msg, + const void* aux) { + std::string details(TestFailedMessage(msg)); + const char* expected_msg = static_cast<const char*>(aux); + +#if !defined(SANDBOX_USES_BASE_TEST_SUITE) + const bool subprocess_got_sigsegv = + WIFSIGNALED(status) && (SIGSEGV == WTERMSIG(status)); +#else + // This hack is required when a signal handler is installed + // for SEGV that will _exit(1). + const bool subprocess_got_sigsegv = + WIFEXITED(status) && (kExitAfterSIGSEGV == WEXITSTATUS(status)); +#endif + + ASSERT_TRUE(subprocess_got_sigsegv) << "Exit status: " << status + << " " << details; + + bool subprocess_exited_without_matching_message = + msg.find(expected_msg) == std::string::npos; + EXPECT_FALSE(subprocess_exited_without_matching_message) << details; +} + +void UnitTests::DeathExitCode(int status, + const std::string& msg, + const void* aux) { + int expected_exit_code = static_cast<int>(reinterpret_cast<intptr_t>(aux)); + std::string details(TestFailedMessage(msg)); + + bool subprocess_terminated_normally = WIFEXITED(status); + ASSERT_TRUE(subprocess_terminated_normally) << details; + int subprocess_exit_status = WEXITSTATUS(status); + ASSERT_EQ(expected_exit_code, subprocess_exit_status) << details; +} + +void UnitTests::DeathBySignal(int status, + const std::string& msg, + const void* aux) { + int expected_signo = static_cast<int>(reinterpret_cast<intptr_t>(aux)); + std::string details(TestFailedMessage(msg)); + + bool subprocess_terminated_by_signal = WIFSIGNALED(status); + ASSERT_TRUE(subprocess_terminated_by_signal) << details; + int subprocess_signal_number = WTERMSIG(status); + ASSERT_EQ(expected_signo, subprocess_signal_number) << details; +} + +void UnitTests::AssertionFailure(const char* expr, const char* file, int line) { + fprintf(stderr, "%s:%d:%s", file, line, expr); + fflush(stderr); + _exit(kExitWithAssertionFailure); +} + +void UnitTests::IgnoreThisTest() { + fflush(stderr); + _exit(kIgnoreThisTest); +} + +} // namespace diff --git a/sandbox/linux/tests/unit_tests.h b/sandbox/linux/tests/unit_tests.h new file mode 100644 index 0000000000..5a7116e932 --- /dev/null +++ b/sandbox/linux/tests/unit_tests.h @@ -0,0 +1,201 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef SANDBOX_LINUX_TESTS_UNIT_TESTS_H_ +#define SANDBOX_LINUX_TESTS_UNIT_TESTS_H_ + +#include "base/macros.h" +#include "build/build_config.h" +#include "sandbox/linux/tests/sandbox_test_runner_function_pointer.h" +#include "testing/gtest/include/gtest/gtest.h" + +namespace sandbox { + +// Has this been compiled to run on Android? +bool IsAndroid(); + +bool IsArchitectureArm(); + +// Is Valgrind currently being used? +bool IsRunningOnValgrind(); + +#if defined(ADDRESS_SANITIZER) +#define DISABLE_ON_ASAN(test_name) DISABLED_##test_name +#else +#define DISABLE_ON_ASAN(test_name) test_name +#endif // defined(ADDRESS_SANITIZER) + +#if defined(LEAK_SANITIZER) +#define DISABLE_ON_LSAN(test_name) DISABLED_##test_name +#else +#define DISABLE_ON_LSAN(test_name) test_name +#endif + +#if defined(THREAD_SANITIZER) +#define DISABLE_ON_TSAN(test_name) DISABLED_##test_name +#else +#define DISABLE_ON_TSAN(test_name) test_name +#endif // defined(THREAD_SANITIZER) + +#if defined(ADDRESS_SANITIZER) || defined(MEMORY_SANITIZER) || \ + defined(THREAD_SANITIZER) || defined(LEAK_SANITIZER) || \ + defined(UNDEFINED_SANITIZER) || defined(SANITIZER_COVERAGE) +#define DISABLE_ON_SANITIZERS(test_name) DISABLED_##test_name +#else +#define DISABLE_ON_SANITIZERS(test_name) test_name +#endif + +#if defined(OS_ANDROID) +#define DISABLE_ON_ANDROID(test_name) DISABLED_##test_name +#else +#define DISABLE_ON_ANDROID(test_name) test_name +#endif + +// While it is perfectly OK for a complex test to provide its own DeathCheck +// function. Most death tests have very simple requirements. These tests should +// use one of the predefined DEATH_XXX macros as an argument to +// SANDBOX_DEATH_TEST(). You can check for a (sub-)string in the output of the +// test, for a particular exit code, or for a particular death signal. +// NOTE: If you do decide to write your own DeathCheck, make sure to use +// gtests's ASSERT_XXX() macros instead of SANDBOX_ASSERT(). See +// unit_tests.cc for examples. +#define DEATH_SUCCESS() sandbox::UnitTests::DeathSuccess, NULL +#define DEATH_SUCCESS_ALLOW_NOISE() \ + sandbox::UnitTests::DeathSuccessAllowNoise, NULL +#define DEATH_MESSAGE(msg) \ + sandbox::UnitTests::DeathMessage, \ + static_cast<const void*>(static_cast<const char*>(msg)) +#define DEATH_SEGV_MESSAGE(msg) \ + sandbox::UnitTests::DeathSEGVMessage, \ + static_cast<const void*>(static_cast<const char*>(msg)) +#define DEATH_EXIT_CODE(rc) \ + sandbox::UnitTests::DeathExitCode, \ + reinterpret_cast<void*>(static_cast<intptr_t>(rc)) +#define DEATH_BY_SIGNAL(s) \ + sandbox::UnitTests::DeathBySignal, \ + reinterpret_cast<void*>(static_cast<intptr_t>(s)) + +// A SANDBOX_DEATH_TEST is just like a SANDBOX_TEST (see below), but it assumes +// that the test actually dies. The death test only passes if the death occurs +// in the expected fashion, as specified by "death" and "death_aux". These two +// parameters are typically set to one of the DEATH_XXX() macros. +#define SANDBOX_DEATH_TEST(test_case_name, test_name, death) \ + void TEST_##test_name(void); \ + TEST(test_case_name, test_name) { \ + SandboxTestRunnerFunctionPointer sandbox_test_runner(TEST_##test_name); \ + sandbox::UnitTests::RunTestInProcess(&sandbox_test_runner, death); \ + } \ + void TEST_##test_name(void) + +// Define a new test case that runs inside of a GTest death test. This is +// necessary, as most of our tests by definition make global and irreversible +// changes to the system (i.e. they install a sandbox). GTest provides death +// tests as a tool to isolate global changes from the rest of the tests. +#define SANDBOX_TEST(test_case_name, test_name) \ + SANDBOX_DEATH_TEST(test_case_name, test_name, DEATH_SUCCESS()) + +// SANDBOX_TEST_ALLOW_NOISE is just like SANDBOX_TEST, except it does not +// consider log error messages printed by the test to be test failures. +#define SANDBOX_TEST_ALLOW_NOISE(test_case_name, test_name) \ + SANDBOX_DEATH_TEST(test_case_name, test_name, DEATH_SUCCESS_ALLOW_NOISE()) + +// Simple assertion macro that is compatible with running inside of a death +// test. We unfortunately cannot use any of the GTest macros. +#define SANDBOX_STR(x) #x +#define SANDBOX_ASSERT(expr) \ + ((expr) ? static_cast<void>(0) : sandbox::UnitTests::AssertionFailure( \ + SANDBOX_STR(expr), __FILE__, __LINE__)) + +#define SANDBOX_ASSERT_EQ(x, y) SANDBOX_ASSERT((x) == (y)) +#define SANDBOX_ASSERT_NE(x, y) SANDBOX_ASSERT((x) != (y)) +#define SANDBOX_ASSERT_LT(x, y) SANDBOX_ASSERT((x) < (y)) +#define SANDBOX_ASSERT_GT(x, y) SANDBOX_ASSERT((x) > (y)) +#define SANDBOX_ASSERT_LE(x, y) SANDBOX_ASSERT((x) <= (y)) +#define SANDBOX_ASSERT_GE(x, y) SANDBOX_ASSERT((x) >= (y)) + +// This class allows to run unittests in their own process. The main method is +// RunTestInProcess(). +class UnitTests { + public: + typedef void (*DeathCheck)(int status, + const std::string& msg, + const void* aux); + + // Runs a test inside a short-lived process. Do not call this function + // directly. It is automatically invoked by SANDBOX_TEST(). Most sandboxing + // functions make global irreversible changes to the execution environment + // and must therefore execute in their own isolated process. + // |test_runner| must implement the SandboxTestRunner interface and will run + // in a subprocess. + // Note: since the child process (created with fork()) will never return from + // RunTestInProcess(), |test_runner| is guaranteed to exist for the lifetime + // of the child process. + static void RunTestInProcess(SandboxTestRunner* test_runner, + DeathCheck death, + const void* death_aux); + + // Report a useful error message and terminate the current SANDBOX_TEST(). + // Calling this function from outside a SANDBOX_TEST() is unlikely to do + // anything useful. + static void AssertionFailure(const char* expr, const char* file, int line); + + // Sometimes we determine at run-time that a test should be disabled. + // Call this method if we want to return from a test and completely + // ignore its results. + // You should not call this method, if the test already ran any test-relevant + // code. Most notably, you should not call it, you already wrote any messages + // to stderr. + static void IgnoreThisTest(); + + // A DeathCheck method that verifies that the test completed succcessfully. + // This is the default test mode for SANDBOX_TEST(). The "aux" parameter + // of this DeathCheck is unused (and thus unnamed) + static void DeathSuccess(int status, const std::string& msg, const void*); + + // A DeathCheck method that verifies that the test completed succcessfully + // allowing for log error messages. + static void DeathSuccessAllowNoise(int status, + const std::string& msg, + const void*); + + // A DeathCheck method that verifies that the test completed with error + // code "1" and printed a message containing a particular substring. The + // "aux" pointer should point to a C-string containing the expected error + // message. This method is useful for checking assertion failures such as + // in SANDBOX_ASSERT() and/or SANDBOX_DIE(). + static void DeathMessage(int status, const std::string& msg, const void* aux); + + // Like DeathMessage() but the process must be terminated with a segmentation + // fault. + // Implementation detail: On Linux (but not on Android), this does check for + // the return value of our default signal handler rather than for the actual + // reception of a SIGSEGV. + // TODO(jln): make this more robust. + static void DeathSEGVMessage(int status, + const std::string& msg, + const void* aux); + + // A DeathCheck method that verifies that the test completed with a + // particular exit code. If the test output any messages to stderr, they are + // silently ignored. The expected exit code should be passed in by + // casting the its "int" value to a "void *", which is then used for "aux". + static void DeathExitCode(int status, + const std::string& msg, + const void* aux); + + // A DeathCheck method that verifies that the test was terminated by a + // particular signal. If the test output any messages to stderr, they are + // silently ignore. The expected signal number should be passed in by + // casting the its "int" value to a "void *", which is then used for "aux". + static void DeathBySignal(int status, + const std::string& msg, + const void* aux); + + private: + DISALLOW_IMPLICIT_CONSTRUCTORS(UnitTests); +}; + +} // namespace + +#endif // SANDBOX_LINUX_TESTS_UNIT_TESTS_H_ diff --git a/sandbox/linux/tests/unit_tests_unittest.cc b/sandbox/linux/tests/unit_tests_unittest.cc new file mode 100644 index 0000000000..57799b14c0 --- /dev/null +++ b/sandbox/linux/tests/unit_tests_unittest.cc @@ -0,0 +1,62 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include <signal.h> +#include <stdlib.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include "base/logging.h" +#include "base/posix/eintr_wrapper.h" +#include "sandbox/linux/tests/unit_tests.h" + +namespace sandbox { + +namespace { + +// Let's not use any of the "magic" values used internally in unit_tests.cc, +// such as kExpectedValue. +const int kExpectedExitCode = 100; + +SANDBOX_DEATH_TEST(UnitTests, + DeathExitCode, + DEATH_EXIT_CODE(kExpectedExitCode)) { + _exit(kExpectedExitCode); +} + +const int kExpectedSignalNumber = SIGKILL; + +SANDBOX_DEATH_TEST(UnitTests, + DeathBySignal, + DEATH_BY_SIGNAL(kExpectedSignalNumber)) { + raise(kExpectedSignalNumber); +} + +SANDBOX_DEATH_TEST(UnitTests, + DeathWithMessage, + DEATH_MESSAGE("Hello")) { + LOG(ERROR) << "Hello"; + _exit(1); +} + +SANDBOX_DEATH_TEST(UnitTests, + SEGVDeathWithMessage, + DEATH_SEGV_MESSAGE("Hello")) { + LOG(ERROR) << "Hello"; + while (1) { + volatile char* addr = reinterpret_cast<volatile char*>(NULL); + *addr = '\0'; + } + + _exit(2); +} + +SANDBOX_TEST_ALLOW_NOISE(UnitTests, NoisyTest) { + LOG(ERROR) << "The cow says moo!"; +} + +} // namespace + +} // namespace sandbox |