Runtime flags for malloc bisection.

Reviewers: kcc, pcc

Subscribers: kubamracek, mgorny, jdoerfert, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58162

Change-Id: I1c17d53f07f7954f670b675d9739d358348b1ac4
git-svn-id: https://llvm.org/svn/llvm-project/compiler-rt/trunk@354156 91177308-0d34-0410-b5e6-96231b3b80d8

8 files changed, 141 insertions, 19 deletions

diff --git a/lib/hwasan/CMakeLists.txt b/lib/hwasan/CMakeLists.txt
index e1704d1be..ed94d3f53 100644
--- a/lib/hwasan/CMakeLists.txt
+++ b/lib/hwasan/CMakeLists.txt
@@ -26,6 +26,7 @@ set(HWASAN_RTL_HEADERS
   hwasan_flags.h
   hwasan_flags.inc
   hwasan_interface_internal.h
+  hwasan_malloc_bisect.h
   hwasan_mapping.h
   hwasan_poisoning.h
   hwasan_report.h
diff --git a/lib/hwasan/hwasan_allocator.cc b/lib/hwasan/hwasan_allocator.cc
index 4a4de3e45..09f1ab94e 100644
--- a/lib/hwasan/hwasan_allocator.cc
+++ b/lib/hwasan/hwasan_allocator.cc
@@ -17,6 +17,7 @@
 #include "hwasan.h"
 #include "hwasan_allocator.h"
 #include "hwasan_mapping.h"
+#include "hwasan_malloc_bisect.h"
 #include "hwasan_thread.h"
 #include "hwasan_report.h"
 
@@ -181,7 +182,7 @@ static void *HwasanAllocate(StackTrace *stack, uptr orig_size, uptr alignment,
   // retag to 0.
   if ((flags()->tag_in_malloc || flags()->tag_in_free) &&
       atomic_load_relaxed(&hwasan_allocator_tagging_enabled)) {
-    tag_t tag = flags()->tag_in_malloc
+    tag_t tag = flags()->tag_in_malloc && malloc_bisect(stack, orig_size)
                     ? (t ? t->GenerateRandomTag() : kFallbackAllocTag)
                     : 0;
     user_ptr = (void *)TagMemoryAligned((uptr)user_ptr, size, tag);
@@ -247,7 +248,7 @@ static void HwasanDeallocate(StackTrace *stack, void *tagged_ptr) {
         Min(TaggedSize(orig_size), (uptr)flags()->max_free_fill_size);
     internal_memset(aligned_ptr, flags()->free_fill_byte, fill_size);
   }
-  if (flags()->tag_in_free &&
+  if (flags()->tag_in_free && malloc_bisect(stack, 0) &&
       atomic_load_relaxed(&hwasan_allocator_tagging_enabled))
     TagMemoryAligned(reinterpret_cast<uptr>(aligned_ptr), TaggedSize(orig_size),
                      t ? t->GenerateRandomTag() : kFallbackFreeTag);
diff --git a/lib/hwasan/hwasan_flags.inc b/lib/hwasan/hwasan_flags.inc
index f0cf151a4..01fdad87a 100644
--- a/lib/hwasan/hwasan_flags.inc
+++ b/lib/hwasan/hwasan_flags.inc
@@ -85,3 +85,16 @@ HWASAN_FLAG(int, stack_history_size, 1024,
             "The number of stack frames remembered per thread. "
             "Affects the quality of stack-related reports, but not the ability "
             "to find bugs.")
+
+// Malloc / free bisection. Only tag malloc and free calls when a hash of
+// allocation size and stack trace is between malloc_bisect_left and
+// malloc_bisect_right (both inclusive). [0, 0] range is special and disables
+// bisection (i.e. everything is tagged). Once the range is narrowed down
+// enough, use malloc_bisect_dump to see interesting allocations.
+HWASAN_FLAG(uptr, malloc_bisect_left, 0,
+            "Left bound of malloc bisection, inclusive.")
+HWASAN_FLAG(uptr, malloc_bisect_right, 0,
+            "Right bound of malloc bisection, inclusive.")
+HWASAN_FLAG(bool, malloc_bisect_dump, false,
+            "Print all allocations within [malloc_bisect_left, "
+            "malloc_bisect_right] range ")
diff --git a/lib/hwasan/hwasan_malloc_bisect.h b/lib/hwasan/hwasan_malloc_bisect.h
new file mode 100644
index 000000000..eaf124aab
--- /dev/null
+++ b/lib/hwasan/hwasan_malloc_bisect.h
@@ -0,0 +1,50 @@
+//===-- hwasan_malloc_bisect.h ----------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of HWAddressSanitizer.
+//
+//===----------------------------------------------------------------------===//
+
+#include "sanitizer_common/sanitizer_hash.h"
+#include "hwasan.h"
+
+namespace __hwasan {
+
+static u32 malloc_hash(StackTrace *stack, uptr orig_size) {
+  uptr len = Min(stack->size, (unsigned)7);
+  MurMur2HashBuilder H(len);
+  H.add(orig_size);
+  // Start with frame #1 to skip __sanitizer_malloc frame, which is
+  // (a) almost always the same (well, could be operator new or new[])
+  // (b) can change hashes when compiler-rt is rebuilt, invalidating previous
+  // bisection results.
+  // Because of ASLR, use only offset inside the page.
+  for (uptr i = 1; i < len; ++i) H.add(((u32)stack->trace[i]) & 0xFFF);
+  return H.get();
+}
+
+static INLINE bool malloc_bisect(StackTrace *stack, uptr orig_size) {
+  uptr left = flags()->malloc_bisect_left;
+  uptr right = flags()->malloc_bisect_right;
+  if (LIKELY(left == 0 && right == 0))
+    return true;
+  if (!stack)
+    return true;
+  // Allow malloc_bisect_right > (u32)(-1) to avoid spelling the latter in
+  // decimal.
+  uptr h = (uptr)malloc_hash(stack, orig_size);
+  if (h < left || h > right)
+    return false;
+  if (flags()->malloc_bisect_dump) {
+    Printf("[alloc] %u %zu\n", h, orig_size);
+    stack->Print();
+  }
+  return true;
+}
+
+}  // namespace __hwasan
diff --git a/lib/sanitizer_common/CMakeLists.txt b/lib/sanitizer_common/CMakeLists.txt
index f7945918a..d37804572 100644
--- a/lib/sanitizer_common/CMakeLists.txt
+++ b/lib/sanitizer_common/CMakeLists.txt
@@ -143,6 +143,7 @@ set(SANITIZER_IMPL_HEADERS
   sanitizer_freebsd.h
   sanitizer_fuchsia.h
   sanitizer_getauxval.h
+  sanitizer_hash.h
   sanitizer_interceptors_ioctl_netbsd.inc
   sanitizer_interface_internal.h
   sanitizer_internal_defs.h
diff --git a/lib/sanitizer_common/sanitizer_hash.h b/lib/sanitizer_common/sanitizer_hash.h
new file mode 100644
index 000000000..3d97dcc5d
--- /dev/null
+++ b/lib/sanitizer_common/sanitizer_hash.h
@@ -0,0 +1,43 @@
+//===-- sanitizer_common.h --------------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a simple hash function.
+//===----------------------------------------------------------------------===//
+
+#ifndef SANITIZER_HASH_H
+#define SANITIZER_HASH_H
+
+#include "sanitizer_internal_defs.h"
+
+namespace __sanitizer {
+class MurMur2HashBuilder {
+  static const u32 m = 0x5bd1e995;
+  static const u32 seed = 0x9747b28c;
+  static const u32 r = 24;
+  u32 h;
+
+ public:
+  explicit MurMur2HashBuilder(u32 init = 0) { h = seed ^ init; }
+  void add(u32 k) {
+    k *= m;
+    k ^= k >> r;
+    k *= m;
+    h *= m;
+    h ^= k;
+  }
+  u32 get() {
+    u32 x = h;
+    x ^= x >> 13;
+    x *= m;
+    x ^= x >> 15;
+    return x;
+  }
+};
+}  //namespace __sanitizer
+
+#endif  // SANITIZER_HASH_H
diff --git a/lib/sanitizer_common/sanitizer_stackdepot.cc b/lib/sanitizer_common/sanitizer_stackdepot.cc
index a38cb9dee..1cdedfa32 100644
--- a/lib/sanitizer_common/sanitizer_stackdepot.cc
+++ b/lib/sanitizer_common/sanitizer_stackdepot.cc
@@ -13,6 +13,7 @@
 #include "sanitizer_stackdepot.h"
 
 #include "sanitizer_common.h"
+#include "sanitizer_hash.h"
 #include "sanitizer_stackdepotbase.h"
 
 namespace __sanitizer {
@@ -49,23 +50,9 @@ struct StackDepotNode {
     return sizeof(StackDepotNode) + (args.size - 1) * sizeof(uptr);
   }
   static u32 hash(const args_type &args) {
-    // murmur2
-    const u32 m = 0x5bd1e995;
-    const u32 seed = 0x9747b28c;
-    const u32 r = 24;
-    u32 h = seed ^ (args.size * sizeof(uptr));
-    for (uptr i = 0; i < args.size; i++) {
-      u32 k = args.trace[i];
-      k *= m;
-      k ^= k >> r;
-      k *= m;
-      h *= m;
-      h ^= k;
-    }
-    h ^= h >> 13;
-    h *= m;
-    h ^= h >> 15;
-    return h;
+    MurMur2HashBuilder H(args.size * sizeof(uptr));
+    for (uptr i = 0; i < args.size; i++) H.add(args.trace[i]);
+    return H.get();
   }
   static bool is_valid(const args_type &args) {
     return args.size > 0 && args.trace;
diff --git a/test/hwasan/TestCases/malloc_bisect.c b/test/hwasan/TestCases/malloc_bisect.c
new file mode 100644
index 000000000..51cbbfe2a
--- /dev/null
+++ b/test/hwasan/TestCases/malloc_bisect.c
@@ -0,0 +1,26 @@
+// RUN: %clang_hwasan -O0 %s -o %t
+// RUN: %env_hwasan_opts=malloc_bisect_left=0,malloc_bisect_right=0          not %run %t 2>&1 | \
+// RUN:     FileCheck %s --check-prefix=CRASH
+// RUN: %env_hwasan_opts=malloc_bisect_left=1000,malloc_bisect_right=999         %run %t 2>&1
+// RUN: %env_hwasan_opts=malloc_bisect_left=0,malloc_bisect_right=4294967295 not %run %t 2>&1 | \
+// RUN:     FileCheck %s --check-prefix=CRASH
+// RUN: %env_hwasan_opts=malloc_bisect_left=0,malloc_bisect_right=4294967295,malloc_bisect_dump=1 not %run %t 2>&1 | \
+// RUN:     FileCheck %s --check-prefixes=CRASH,DUMP
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <sanitizer/hwasan_interface.h>
+
+int main() {
+  __hwasan_enable_allocator_tagging();
+  // DUMP: [alloc] {{.*}} 10{{$}}
+  // DUMP: in main{{.*}}malloc_bisect.c
+  char * volatile p = (char*)malloc(10);
+  // CRASH: HWAddressSanitizer: tag-mismatch on address
+  // CRASH: in main{{.*}}malloc_bisect.c
+  char volatile x = p[16];
+  free(p);
+  __hwasan_disable_allocator_tagging();
+
+  return 0;
+}