diff options
author | van Hauser <vh@thc.org> | 2021-01-21 12:03:21 +0100 |
---|---|---|
committer | GitHub <noreply@github.com> | 2021-01-21 12:03:21 +0100 |
commit | fc3f06caec887b807e13eeb8be4c9899ec04036a (patch) | |
tree | 1884eab7e5e1db9d4c9a00439fdb1589f15bc410 /instrumentation | |
parent | b39b87b2f720926645ac5ae23fd49dec7e94c7e9 (diff) | |
parent | 0d472adef0cbe68cec128b7b15e508f0bb05455d (diff) | |
download | AFLplusplus-fc3f06caec887b807e13eeb8be4c9899ec04036a.tar.gz |
Merge pull request #689 from AFLplusplus/cmplog_merge
cmplog merge
Diffstat (limited to 'instrumentation')
-rw-r--r-- | instrumentation/afl-compiler-rt.o.c | 281 | ||||
-rw-r--r-- | instrumentation/cmplog-instructions-pass.cc | 590 | ||||
-rw-r--r-- | instrumentation/llvm-ngram-coverage.h | 2 |
3 files changed, 747 insertions, 126 deletions
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c index e31bff16..14da4caa 100644 --- a/instrumentation/afl-compiler-rt.o.c +++ b/instrumentation/afl-compiler-rt.o.c @@ -161,7 +161,7 @@ void send_forkserver_error(int error) { u32 status; if (!error || error > 0xffff) return; status = (FS_OPT_ERROR | FS_OPT_SET_ERROR(error)); - if (write(FORKSRV_FD + 1, (char *)&status, 4) != 4) return; + if (write(FORKSRV_FD + 1, (char *)&status, 4) != 4) { return; } } @@ -544,11 +544,11 @@ static void __afl_start_snapshots(void) { if (__afl_dictionary_len && __afl_dictionary) status |= FS_OPT_AUTODICT; memcpy(tmp, &status, 4); - if (write(FORKSRV_FD + 1, tmp, 4) != 4) return; + if (write(FORKSRV_FD + 1, tmp, 4) != 4) { return; } if (__afl_sharedmem_fuzzing || (__afl_dictionary_len && __afl_dictionary)) { - if (read(FORKSRV_FD, &was_killed, 4) != 4) _exit(1); + if (read(FORKSRV_FD, &was_killed, 4) != 4) { _exit(1); } if (getenv("AFL_DEBUG")) { @@ -1207,22 +1207,33 @@ void __sanitizer_cov_trace_pc_guard_init(uint32_t *start, uint32_t *stop) { ///// CmpLog instrumentation -void __cmplog_ins_hook1(uint8_t arg1, uint8_t arg2) { +void __cmplog_ins_hook1(uint8_t arg1, uint8_t arg2, uint8_t attr) { - if (unlikely(!__afl_cmp_map)) return; + // fprintf(stderr, "hook1 arg0=%02x arg1=%02x attr=%u\n", + // (u8) arg1, (u8) arg2, attr); + + if (unlikely(!__afl_cmp_map || arg1 == arg2)) return; uintptr_t k = (uintptr_t)__builtin_return_address(0); k = (k >> 4) ^ (k << 8); k &= CMP_MAP_W - 1; - __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + u32 hits; - u32 hits = __afl_cmp_map->headers[k].hits; - __afl_cmp_map->headers[k].hits = hits + 1; - // if (!__afl_cmp_map->headers[k].cnt) - // __afl_cmp_map->headers[k].cnt = __afl_cmp_counter++; + if (__afl_cmp_map->headers[k].type != CMP_TYPE_INS) { - __afl_cmp_map->headers[k].shape = 0; + __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + hits = 0; + __afl_cmp_map->headers[k].hits = 1; + __afl_cmp_map->headers[k].shape = 0; + + } else { + + hits = __afl_cmp_map->headers[k].hits++; + + } + + __afl_cmp_map->headers[k].attribute = attr; hits &= CMP_MAP_H - 1; __afl_cmp_map->log[k][hits].v0 = arg1; @@ -1230,20 +1241,36 @@ void __cmplog_ins_hook1(uint8_t arg1, uint8_t arg2) { } -void __cmplog_ins_hook2(uint16_t arg1, uint16_t arg2) { +void __cmplog_ins_hook2(uint16_t arg1, uint16_t arg2, uint8_t attr) { - if (unlikely(!__afl_cmp_map)) return; + if (unlikely(!__afl_cmp_map || arg1 == arg2)) return; uintptr_t k = (uintptr_t)__builtin_return_address(0); k = (k >> 4) ^ (k << 8); k &= CMP_MAP_W - 1; - __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + u32 hits; + + if (__afl_cmp_map->headers[k].type != CMP_TYPE_INS) { + + __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + hits = 0; + __afl_cmp_map->headers[k].hits = 1; + __afl_cmp_map->headers[k].shape = 1; + + } else { - u32 hits = __afl_cmp_map->headers[k].hits; - __afl_cmp_map->headers[k].hits = hits + 1; + hits = __afl_cmp_map->headers[k].hits++; - __afl_cmp_map->headers[k].shape = 1; + if (!__afl_cmp_map->headers[k].shape) { + + __afl_cmp_map->headers[k].shape = 1; + + } + + } + + __afl_cmp_map->headers[k].attribute = attr; hits &= CMP_MAP_H - 1; __afl_cmp_map->log[k][hits].v0 = arg1; @@ -1251,20 +1278,38 @@ void __cmplog_ins_hook2(uint16_t arg1, uint16_t arg2) { } -void __cmplog_ins_hook4(uint32_t arg1, uint32_t arg2) { +void __cmplog_ins_hook4(uint32_t arg1, uint32_t arg2, uint8_t attr) { - if (unlikely(!__afl_cmp_map)) return; + // fprintf(stderr, "hook4 arg0=%x arg1=%x attr=%u\n", arg1, arg2, attr); + + if (unlikely(!__afl_cmp_map || arg1 == arg2)) return; uintptr_t k = (uintptr_t)__builtin_return_address(0); k = (k >> 4) ^ (k << 8); k &= CMP_MAP_W - 1; - __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + u32 hits; + + if (__afl_cmp_map->headers[k].type != CMP_TYPE_INS) { + + __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + hits = 0; + __afl_cmp_map->headers[k].hits = 1; + __afl_cmp_map->headers[k].shape = 3; + + } else { + + hits = __afl_cmp_map->headers[k].hits++; + + if (__afl_cmp_map->headers[k].shape < 3) { + + __afl_cmp_map->headers[k].shape = 3; - u32 hits = __afl_cmp_map->headers[k].hits; - __afl_cmp_map->headers[k].hits = hits + 1; + } + + } - __afl_cmp_map->headers[k].shape = 3; + __afl_cmp_map->headers[k].attribute = attr; hits &= CMP_MAP_H - 1; __afl_cmp_map->log[k][hits].v0 = arg1; @@ -1272,20 +1317,38 @@ void __cmplog_ins_hook4(uint32_t arg1, uint32_t arg2) { } -void __cmplog_ins_hook8(uint64_t arg1, uint64_t arg2) { +void __cmplog_ins_hook8(uint64_t arg1, uint64_t arg2, uint8_t attr) { - if (unlikely(!__afl_cmp_map)) return; + // fprintf(stderr, "hook8 arg0=%lx arg1=%lx attr=%u\n", arg1, arg2, attr); + + if (unlikely(!__afl_cmp_map || arg1 == arg2)) return; uintptr_t k = (uintptr_t)__builtin_return_address(0); k = (k >> 4) ^ (k << 8); k &= CMP_MAP_W - 1; - __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + u32 hits; + + if (__afl_cmp_map->headers[k].type != CMP_TYPE_INS) { + + __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + hits = 0; + __afl_cmp_map->headers[k].hits = 1; + __afl_cmp_map->headers[k].shape = 7; - u32 hits = __afl_cmp_map->headers[k].hits; - __afl_cmp_map->headers[k].hits = hits + 1; + } else { + + hits = __afl_cmp_map->headers[k].hits++; + + if (__afl_cmp_map->headers[k].shape < 7) { + + __afl_cmp_map->headers[k].shape = 7; + + } + + } - __afl_cmp_map->headers[k].shape = 7; + __afl_cmp_map->headers[k].attribute = attr; hits &= CMP_MAP_H - 1; __afl_cmp_map->log[k][hits].v0 = arg1; @@ -1293,16 +1356,110 @@ void __cmplog_ins_hook8(uint64_t arg1, uint64_t arg2) { } +#ifdef WORD_SIZE_64 +// support for u24 to u120 via llvm _ExitInt(). size is in bytes minus 1 +void __cmplog_ins_hookN(uint128_t arg1, uint128_t arg2, uint8_t attr, + uint8_t size) { + + // fprintf(stderr, "hookN arg0=%llx:%llx arg1=%llx:%llx bytes=%u attr=%u\n", + // (u64)(arg1 >> 64), (u64)arg1, (u64)(arg2 >> 64), (u64)arg2, size + 1, + // attr); + + if (unlikely(!__afl_cmp_map || arg1 == arg2)) return; + + uintptr_t k = (uintptr_t)__builtin_return_address(0); + k = (k >> 4) ^ (k << 8); + k &= CMP_MAP_W - 1; + + u32 hits; + + if (__afl_cmp_map->headers[k].type != CMP_TYPE_INS) { + + __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + hits = 0; + __afl_cmp_map->headers[k].hits = 1; + __afl_cmp_map->headers[k].shape = size; + + } else { + + hits = __afl_cmp_map->headers[k].hits++; + + if (__afl_cmp_map->headers[k].shape < size) { + + __afl_cmp_map->headers[k].shape = size; + + } + + } + + __afl_cmp_map->headers[k].attribute = attr; + + hits &= CMP_MAP_H - 1; + __afl_cmp_map->log[k][hits].v0 = (u64)arg1; + __afl_cmp_map->log[k][hits].v1 = (u64)arg2; + + if (size > 7) { + + __afl_cmp_map->log[k][hits].v0_128 = (u64)(arg1 >> 64); + __afl_cmp_map->log[k][hits].v1_128 = (u64)(arg2 >> 64); + + } + +} + +void __cmplog_ins_hook16(uint128_t arg1, uint128_t arg2, uint8_t attr) { + + if (unlikely(!__afl_cmp_map)) return; + + uintptr_t k = (uintptr_t)__builtin_return_address(0); + k = (k >> 4) ^ (k << 8); + k &= CMP_MAP_W - 1; + + u32 hits; + + if (__afl_cmp_map->headers[k].type != CMP_TYPE_INS) { + + __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + hits = 0; + __afl_cmp_map->headers[k].hits = 1; + __afl_cmp_map->headers[k].shape = 15; + + } else { + + hits = __afl_cmp_map->headers[k].hits++; + + if (__afl_cmp_map->headers[k].shape < 15) { + + __afl_cmp_map->headers[k].shape = 15; + + } + + } + + __afl_cmp_map->headers[k].attribute = attr; + + hits &= CMP_MAP_H - 1; + __afl_cmp_map->log[k][hits].v0 = (u64)arg1; + __afl_cmp_map->log[k][hits].v1 = (u64)arg2; + __afl_cmp_map->log[k][hits].v0_128 = (u64)(arg1 >> 64); + __afl_cmp_map->log[k][hits].v1_128 = (u64)(arg2 >> 64); + +} + +#endif + #if defined(__APPLE__) #pragma weak __sanitizer_cov_trace_const_cmp1 = __cmplog_ins_hook1 #pragma weak __sanitizer_cov_trace_const_cmp2 = __cmplog_ins_hook2 #pragma weak __sanitizer_cov_trace_const_cmp4 = __cmplog_ins_hook4 #pragma weak __sanitizer_cov_trace_const_cmp8 = __cmplog_ins_hook8 + #pragma weak __sanitizer_cov_trace_const_cmp16 = __cmplog_ins_hook16 #pragma weak __sanitizer_cov_trace_cmp1 = __cmplog_ins_hook1 #pragma weak __sanitizer_cov_trace_cmp2 = __cmplog_ins_hook2 #pragma weak __sanitizer_cov_trace_cmp4 = __cmplog_ins_hook4 #pragma weak __sanitizer_cov_trace_cmp8 = __cmplog_ins_hook8 + #pragma weak __sanitizer_cov_trace_cmp16 = __cmplog_ins_hook16 #else void __sanitizer_cov_trace_const_cmp1(uint8_t arg1, uint8_t arg2) __attribute__((alias("__cmplog_ins_hook1"))); @@ -1312,6 +1469,10 @@ void __sanitizer_cov_trace_const_cmp4(uint32_t arg1, uint32_t arg2) __attribute__((alias("__cmplog_ins_hook4"))); void __sanitizer_cov_trace_const_cmp8(uint64_t arg1, uint64_t arg2) __attribute__((alias("__cmplog_ins_hook8"))); + #ifdef WORD_SIZE_64 +void __sanitizer_cov_trace_const_cmp16(uint128_t arg1, uint128_t arg2) + __attribute__((alias("__cmplog_ins_hook16"))); + #endif void __sanitizer_cov_trace_cmp1(uint8_t arg1, uint8_t arg2) __attribute__((alias("__cmplog_ins_hook1"))); @@ -1321,6 +1482,10 @@ void __sanitizer_cov_trace_cmp4(uint32_t arg1, uint32_t arg2) __attribute__((alias("__cmplog_ins_hook4"))); void __sanitizer_cov_trace_cmp8(uint64_t arg1, uint64_t arg2) __attribute__((alias("__cmplog_ins_hook8"))); + #ifdef WORD_SIZE_64 +void __sanitizer_cov_trace_cmp16(uint128_t arg1, uint128_t arg2) + __attribute__((alias("__cmplog_ins_hook16"))); + #endif #endif /* defined(__APPLE__) */ void __sanitizer_cov_trace_switch(uint64_t val, uint64_t *cases) { @@ -1333,12 +1498,28 @@ void __sanitizer_cov_trace_switch(uint64_t val, uint64_t *cases) { k = (k >> 4) ^ (k << 8); k &= CMP_MAP_W - 1; - __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + u32 hits; - u32 hits = __afl_cmp_map->headers[k].hits; - __afl_cmp_map->headers[k].hits = hits + 1; + if (__afl_cmp_map->headers[k].type != CMP_TYPE_INS) { - __afl_cmp_map->headers[k].shape = 7; + __afl_cmp_map->headers[k].type = CMP_TYPE_INS; + hits = 0; + __afl_cmp_map->headers[k].hits = 1; + __afl_cmp_map->headers[k].shape = 7; + + } else { + + hits = __afl_cmp_map->headers[k].hits++; + + if (__afl_cmp_map->headers[k].shape < 7) { + + __afl_cmp_map->headers[k].shape = 7; + + } + + } + + __afl_cmp_map->headers[k].attribute = 1; hits &= CMP_MAP_H - 1; __afl_cmp_map->log[k][hits].v0 = val; @@ -1364,6 +1545,18 @@ static int area_is_mapped(void *ptr, size_t len) { void __cmplog_rtn_hook(u8 *ptr1, u8 *ptr2) { + /* + u32 i; + if (!area_is_mapped(ptr1, 32) || !area_is_mapped(ptr2, 32)) return; + fprintf(stderr, "rtn arg0="); + for (i = 0; i < 8; i++) + fprintf(stderr, "%02x", ptr1[i]); + fprintf(stderr, " arg1="); + for (i = 0; i < 8; i++) + fprintf(stderr, "%02x", ptr2[i]); + fprintf(stderr, "\n"); + */ + if (unlikely(!__afl_cmp_map)) return; if (!area_is_mapped(ptr1, 32) || !area_is_mapped(ptr2, 32)) return; @@ -1372,12 +1565,26 @@ void __cmplog_rtn_hook(u8 *ptr1, u8 *ptr2) { k = (k >> 4) ^ (k << 8); k &= CMP_MAP_W - 1; - __afl_cmp_map->headers[k].type = CMP_TYPE_RTN; + u32 hits; + + if (__afl_cmp_map->headers[k].type != CMP_TYPE_RTN) { - u32 hits = __afl_cmp_map->headers[k].hits; - __afl_cmp_map->headers[k].hits = hits + 1; + __afl_cmp_map->headers[k].type = CMP_TYPE_RTN; + hits = 0; + __afl_cmp_map->headers[k].hits = 1; + __afl_cmp_map->headers[k].shape = 31; - __afl_cmp_map->headers[k].shape = 31; + } else { + + hits = __afl_cmp_map->headers[k].hits++; + + if (__afl_cmp_map->headers[k].shape < 31) { + + __afl_cmp_map->headers[k].shape = 31; + + } + + } hits &= CMP_MAP_RTN_H - 1; __builtin_memcpy(((struct cmpfn_operands *)__afl_cmp_map->log[k])[hits].v0, diff --git a/instrumentation/cmplog-instructions-pass.cc b/instrumentation/cmplog-instructions-pass.cc index 3499ccf0..6ce1832f 100644 --- a/instrumentation/cmplog-instructions-pass.cc +++ b/instrumentation/cmplog-instructions-pass.cc @@ -85,9 +85,25 @@ class CmpLogInstructions : public ModulePass { char CmpLogInstructions::ID = 0; +template <class Iterator> +Iterator Unique(Iterator first, Iterator last) { + + while (first != last) { + + Iterator next(first); + last = std::remove(++next, last, *first); + first = next; + + } + + return last; + +} + bool CmpLogInstructions::hookInstrs(Module &M) { std::vector<Instruction *> icomps; + std::vector<SwitchInst *> switches; LLVMContext & C = M.getContext(); Type * VoidTy = Type::getVoidTy(C); @@ -95,13 +111,15 @@ bool CmpLogInstructions::hookInstrs(Module &M) { IntegerType *Int16Ty = IntegerType::getInt16Ty(C); IntegerType *Int32Ty = IntegerType::getInt32Ty(C); IntegerType *Int64Ty = IntegerType::getInt64Ty(C); + IntegerType *Int128Ty = IntegerType::getInt128Ty(C); #if LLVM_VERSION_MAJOR < 9 Constant * #else FunctionCallee #endif - c1 = M.getOrInsertFunction("__cmplog_ins_hook1", VoidTy, Int8Ty, Int8Ty + c1 = M.getOrInsertFunction("__cmplog_ins_hook1", VoidTy, Int8Ty, Int8Ty, + Int8Ty #if LLVM_VERSION_MAJOR < 5 , NULL @@ -118,7 +136,8 @@ bool CmpLogInstructions::hookInstrs(Module &M) { #else FunctionCallee #endif - c2 = M.getOrInsertFunction("__cmplog_ins_hook2", VoidTy, Int16Ty, Int16Ty + c2 = M.getOrInsertFunction("__cmplog_ins_hook2", VoidTy, Int16Ty, Int16Ty, + Int8Ty #if LLVM_VERSION_MAJOR < 5 , NULL @@ -135,7 +154,8 @@ bool CmpLogInstructions::hookInstrs(Module &M) { #else FunctionCallee #endif - c4 = M.getOrInsertFunction("__cmplog_ins_hook4", VoidTy, Int32Ty, Int32Ty + c4 = M.getOrInsertFunction("__cmplog_ins_hook4", VoidTy, Int32Ty, Int32Ty, + Int8Ty #if LLVM_VERSION_MAJOR < 5 , NULL @@ -152,7 +172,8 @@ bool CmpLogInstructions::hookInstrs(Module &M) { #else FunctionCallee #endif - c8 = M.getOrInsertFunction("__cmplog_ins_hook8", VoidTy, Int64Ty, Int64Ty + c8 = M.getOrInsertFunction("__cmplog_ins_hook8", VoidTy, Int64Ty, Int64Ty, + Int8Ty #if LLVM_VERSION_MAJOR < 5 , NULL @@ -164,6 +185,42 @@ bool CmpLogInstructions::hookInstrs(Module &M) { FunctionCallee cmplogHookIns8 = c8; #endif +#if LLVM_VERSION_MAJOR < 9 + Constant * +#else + FunctionCallee +#endif + c16 = M.getOrInsertFunction("__cmplog_ins_hook16", VoidTy, Int128Ty, + Int128Ty, Int8Ty +#if LLVM_VERSION_MAJOR < 5 + , + NULL +#endif + ); +#if LLVM_VERSION_MAJOR < 9 + Function *cmplogHookIns16 = cast<Function>(c16); +#else + FunctionCallee cmplogHookIns16 = c16; +#endif + +#if LLVM_VERSION_MAJOR < 9 + Constant * +#else + FunctionCallee +#endif + cN = M.getOrInsertFunction("__cmplog_ins_hookN", VoidTy, Int128Ty, + Int128Ty, Int8Ty, Int8Ty +#if LLVM_VERSION_MAJOR < 5 + , + NULL +#endif + ); +#if LLVM_VERSION_MAJOR < 9 + Function *cmplogHookInsN = cast<Function>(cN); +#else + FunctionCallee cmplogHookInsN = cN; +#endif + /* iterate over all functions, bbs and instruction and add suitable calls */ for (auto &F : M) { @@ -174,35 +231,16 @@ bool CmpLogInstructions::hookInstrs(Module &M) { for (auto &IN : BB) { CmpInst *selectcmpInst = nullptr; - if ((selectcmpInst = dyn_cast<CmpInst>(&IN))) { - if (selectcmpInst->getPredicate() == CmpInst::ICMP_EQ || - selectcmpInst->getPredicate() == CmpInst::ICMP_NE || - selectcmpInst->getPredicate() == CmpInst::ICMP_UGT || - selectcmpInst->getPredicate() == CmpInst::ICMP_SGT || - selectcmpInst->getPredicate() == CmpInst::ICMP_ULT || - selectcmpInst->getPredicate() == CmpInst::ICMP_SLT || - selectcmpInst->getPredicate() == CmpInst::ICMP_UGE || - selectcmpInst->getPredicate() == CmpInst::ICMP_SGE || - selectcmpInst->getPredicate() == CmpInst::ICMP_ULE || - selectcmpInst->getPredicate() == CmpInst::ICMP_SLE || - selectcmpInst->getPredicate() == CmpInst::FCMP_OGE || - selectcmpInst->getPredicate() == CmpInst::FCMP_UGE || - selectcmpInst->getPredicate() == CmpInst::FCMP_OLE || - selectcmpInst->getPredicate() == CmpInst::FCMP_ULE || - selectcmpInst->getPredicate() == CmpInst::FCMP_OGT || - selectcmpInst->getPredicate() == CmpInst::FCMP_UGT || - selectcmpInst->getPredicate() == CmpInst::FCMP_OLT || - selectcmpInst->getPredicate() == CmpInst::FCMP_ULT || - selectcmpInst->getPredicate() == CmpInst::FCMP_UEQ || - selectcmpInst->getPredicate() == CmpInst::FCMP_OEQ || - selectcmpInst->getPredicate() == CmpInst::FCMP_UNE || - selectcmpInst->getPredicate() == CmpInst::FCMP_ONE) { - - icomps.push_back(selectcmpInst); + icomps.push_back(selectcmpInst); - } + } + + SwitchInst *switchInst = nullptr; + if ((switchInst = dyn_cast<SwitchInst>(BB.getTerminator()))) { + + if (switchInst->getNumCases() > 1) { switches.push_back(switchInst); } } @@ -212,101 +250,477 @@ bool CmpLogInstructions::hookInstrs(Module &M) { } - if (!icomps.size()) return false; - // if (!be_quiet) errs() << "Hooking " << icomps.size() << " cmp - // instructions\n"; + // unique the collected switches + switches.erase(Unique(switches.begin(), switches.end()), switches.end()); + + // Instrument switch values for cmplog + if (switches.size()) { + + if (!be_quiet) + errs() << "Hooking " << switches.size() << " switch instructions\n"; + + for (auto &SI : switches) { + + Value * Val = SI->getCondition(); + unsigned int max_size = Val->getType()->getIntegerBitWidth(), cast_size; + unsigned char do_cast = 0; + + if (!SI->getNumCases() || max_size <= 8) { + + // if (!be_quiet) errs() << "skip trivial switch..\n"; + continue; + + } + + IRBuilder<> IRB(SI->getParent()); + IRB.SetInsertPoint(SI); + + if (max_size % 8) { + + max_size = (((max_size / 8) + 1) * 8); + do_cast = 1; + + } + + if (max_size > 128) { + + if (!be_quiet) { + + fprintf(stderr, + "Cannot handle this switch bit size: %u (truncating)\n", + max_size); + + } + + max_size = 128; + do_cast = 1; + + } + + // do we need to cast? + switch (max_size) { + + case 8: + case 16: + case 32: + case 64: + case 128: + cast_size = max_size; + break; + default: + cast_size = 128; + do_cast = 1; + + } + + Value *CompareTo = Val; + + if (do_cast) { + + ConstantInt *cint = dyn_cast<ConstantInt>(Val); + if (cint) { + + uint64_t val = cint->getZExtValue(); + // fprintf(stderr, "ConstantInt: %lu\n", val); + switch (cast_size) { + + case 8: + CompareTo = ConstantInt::get(Int8Ty, val); + break; + case 16: + CompareTo = ConstantInt::get(Int16Ty, val); + break; + case 32: + CompareTo = ConstantInt::get(Int32Ty, val); + break; + case 64: + CompareTo = ConstantInt::get(Int64Ty, val); + break; + case 128: + CompareTo = ConstantInt::get(Int128Ty, val); + break; + + } + + } else { + + CompareTo = IRB.CreateBitCast(Val, IntegerType::get(C, cast_size)); + + } + + } + + for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end(); i != e; + ++i) { + +#if LLVM_VERSION_MAJOR < 5 + ConstantInt *cint = i.getCaseValue(); +#else + ConstantInt *cint = i->getCaseValue(); +#endif + + if (cint) { + + std::vector<Value *> args; + args.push_back(CompareTo); + + Value *new_param = cint; - for (auto &selectcmpInst : icomps) { + if (do_cast) { - IRBuilder<> IRB(selectcmpInst->getParent()); - IRB.SetInsertPoint(selectcmpInst); + uint64_t val = cint->getZExtValue(); + // fprintf(stderr, "ConstantInt: %lu\n", val); + switch (cast_size) { - auto op0 = selectcmpInst->getOperand(0); - auto op1 = selectcmpInst->getOperand(1); + case 8: + new_param = ConstantInt::get(Int8Ty, val); + break; + case 16: + new_param = ConstantInt::get(Int16Ty, val); + break; + case 32: + new_param = ConstantInt::get(Int32Ty, val); + break; + case 64: + new_param = ConstantInt::get(Int64Ty, val); + break; + case 128: + new_param = ConstantInt::get(Int128Ty, val); + break; - IntegerType * intTyOp0 = NULL; - IntegerType * intTyOp1 = NULL; - unsigned max_size = 0; - std::vector<Value *> args; + } - if (selectcmpInst->getOpcode() == Instruction::FCmp) { + } + + if (new_param) { + + args.push_back(new_param); + ConstantInt *attribute = ConstantInt::get(Int8Ty, 1); + args.push_back(attribute); + if (cast_size != max_size) { + + ConstantInt *bitsize = + ConstantInt::get(Int8Ty, (max_size / 8) - 1); + args.push_back(bitsize); + + } + + switch (cast_size) { + + case 8: + IRB.CreateCall(cmplogHookIns1, args); + break; + case 16: + IRB.CreateCall(cmplogHookIns2, args); + break; + case 32: + IRB.CreateCall(cmplogHookIns4, args); + break; + case 64: + IRB.CreateCall(cmplogHookIns8, args); + break; + case 128: +#ifdef WORD_SIZE_64 + if (max_size == 128) { + + IRB.CreateCall(cmplogHookIns16, args); + + } else { + + IRB.CreateCall(cmplogHookInsN, args); + + } - auto ty0 = op0->getType(); - if (ty0->isHalfTy() +#endif + break; + default: + break; + + } + + } + + } + + } + + } + + } + + if (icomps.size()) { + + // if (!be_quiet) errs() << "Hooking " << icomps.size() << + // " cmp instructions\n"; + + for (auto &selectcmpInst : icomps) { + + IRBuilder<> IRB(selectcmpInst->getParent()); + IRB.SetInsertPoint(selectcmpInst); + + Value *op0 = selectcmpInst->getOperand(0); + Value *op1 = selectcmpInst->getOperand(1); + + IntegerType * intTyOp0 = NULL; + IntegerType * intTyOp1 = NULL; + unsigned max_size = 0, cast_size = 0; + unsigned char attr = 0, do_cast = 0; + std::vector<Value *> args; + + CmpInst *cmpInst = dyn_cast<CmpInst>(selectcmpInst); + + if (!cmpInst) { continue; } + + switch (cmpInst->getPredicate()) { + + case CmpInst::ICMP_NE: + case CmpInst::FCMP_UNE: + case CmpInst::FCMP_ONE: + break; + case CmpInst::ICMP_EQ: + case CmpInst::FCMP_UEQ: + case CmpInst::FCMP_OEQ: + attr += 1; + break; + case CmpInst::ICMP_UGT: + case CmpInst::ICMP_SGT: + case CmpInst::FCMP_OGT: + case CmpInst::FCMP_UGT: + attr += 2; + break; + case CmpInst::ICMP_UGE: + case CmpInst::ICMP_SGE: + case CmpInst::FCMP_OGE: + case CmpInst::FCMP_UGE: + attr += 3; + break; + case CmpInst::ICMP_ULT: + case CmpInst::ICMP_SLT: + case CmpInst::FCMP_OLT: + case CmpInst::FCMP_ULT: + attr += 4; + break; + case CmpInst::ICMP_ULE: + case CmpInst::ICMP_SLE: + case CmpInst::FCMP_OLE: + case CmpInst::FCMP_ULE: + attr += 5; + break; + default: + break; + + } + + if (selectcmpInst->getOpcode() == Instruction::FCmp) { + + auto ty0 = op0->getType(); + if (ty0->isHalfTy() #if LLVM_VERSION_MAJOR >= 11 - || ty0->isBFloatTy() + || ty0->isBFloatTy() #endif - ) - max_size = 16; - else if (ty0->isFloatTy()) - max_size = 32; - else if (ty0->isDoubleTy()) - max_size = 64; + ) + max_size = 16; + else if (ty0->isFloatTy()) + max_size = 32; + else if (ty0->isDoubleTy()) + max_size = 64; + else if (ty0->isX86_FP80Ty()) + max_size = 80; + else if (ty0->isFP128Ty() || ty0->isPPC_FP128Ty()) + max_size = 128; + + attr += 8; + do_cast = 1; - if (max_size) { + } else { - Value *V0 = IRB.CreateBitCast(op0, IntegerType::get(C, max_size)); - intTyOp0 = dyn_cast<IntegerType>(V0->getType()); - Value *V1 = IRB.CreateBitCast(op1, IntegerType::get(C, max_size)); - intTyOp1 = dyn_cast<IntegerType>(V1->getType()); + intTyOp0 = dyn_cast<IntegerType>(op0->getType()); + intTyOp1 = dyn_cast<IntegerType>(op1->getType()); if (intTyOp0 && intTyOp1) { max_size = intTyOp0->getBitWidth() > intTyOp1->getBitWidth() ? intTyOp0->getBitWidth() : intTyOp1->getBitWidth(); - args.push_back(V0); - args.push_back(V1); - } else { + } + + } + + if (!max_size) { continue; } + + // _ExtInt() with non-8th values + if (max_size % 8) { + + max_size = (((max_size / 8) + 1) * 8); + do_cast = 1; + + } + + if (max_size > 128) { - max_size = 0; + if (!be_quiet) { + + fprintf(stderr, + "Cannot handle this compare bit size: %u (truncating)\n", + max_size); } + max_size = 128; + do_cast = 1; + } - } else { + // do we need to cast? + switch (max_size) { + + case 8: + case 16: + case 32: + case 64: + case 128: + cast_size = max_size; + break; + default: + cast_size = 128; + do_cast = 1; - intTyOp0 = dyn_cast<IntegerType>(op0->getType()); - intTyOp1 = dyn_cast<IntegerType>(op1->getType()); + } + + if (do_cast) { + + // F*cking LLVM optimized out any kind of bitcasts of ConstantInt values + // creating illegal calls. WTF. So we have to work around this. + + ConstantInt *cint = dyn_cast<ConstantInt>(op0); + if (cint) { + + uint64_t val = cint->getZExtValue(); + // fprintf(stderr, "ConstantInt: %lu\n", val); + ConstantInt *new_param = NULL; + switch (cast_size) { + + case 8: + new_param = ConstantInt::get(Int8Ty, val); + break; + case 16: + new_param = ConstantInt::get(Int16Ty, val); + break; + case 32: + new_param = ConstantInt::get(Int32Ty, val); + break; + case 64: + new_param = ConstantInt::get(Int64Ty, val); + break; + case 128: + new_param = ConstantInt::get(Int128Ty, val); + break; - if (intTyOp0 && intTyOp1) { + } + + if (!new_param) { continue; } + args.push_back(new_param); + + } else { + + Value *V0 = IRB.CreateBitCast(op0, IntegerType::get(C, cast_size)); + args.push_back(V0); + + } + + cint = dyn_cast<ConstantInt>(op1); + if (cint) { + + uint64_t val = cint->getZExtValue(); + ConstantInt *new_param = NULL; + switch (cast_size) { + + case 8: + new_param = ConstantInt::get(Int8Ty, val); + break; + case 16: + new_param = ConstantInt::get(Int16Ty, val); + break; + case 32: + new_param = ConstantInt::get(Int32Ty, val); + break; + case 64: + new_param = ConstantInt::get(Int64Ty, val); + break; + case 128: + new_param = ConstantInt::get(Int128Ty, val); + break; + + } + + if (!new_param) { continue; } + args.push_back(new_param); + + } else { + + Value *V1 = IRB.CreateBitCast(op1, IntegerType::get(C, cast_size)); + args.push_back(V1); + + } + + } else { - max_size = intTyOp0->getBitWidth() > intTyOp1->getBitWidth() - ? intTyOp0->getBitWidth() - : intTyOp1->getBitWidth(); args.push_back(op0); args.push_back(op1); } - } + ConstantInt *attribute = ConstantInt::get(Int8Ty, attr); + args.push_back(attribute); + + if (cast_size != max_size) { + + ConstantInt *bitsize = ConstantInt::get(Int8Ty, (max_size / 8) - 1); + args.push_back(bitsize); + + } + + // fprintf(stderr, "_ExtInt(%u) castTo %u with attr %u didcast %u\n", + // max_size, cast_size, attr, do_cast); + + switch (cast_size) { - if (max_size < 8 || max_size > 64 || !intTyOp0 || !intTyOp1) continue; - - switch (max_size) { - - case 8: - IRB.CreateCall(cmplogHookIns1, args); - break; - case 16: - IRB.CreateCall(cmplogHookIns2, args); - break; - case 32: - IRB.CreateCall(cmplogHookIns4, args); - break; - case 64: - IRB.CreateCall(cmplogHookIns8, args); - break; - default: - break; + case 8: + IRB.CreateCall(cmplogHookIns1, args); + break; + case 16: + IRB.CreateCall(cmplogHookIns2, args); + break; + case 32: + IRB.CreateCall(cmplogHookIns4, args); + break; + case 64: + IRB.CreateCall(cmplogHookIns8, args); + break; + case 128: + if (max_size == 128) { + + IRB.CreateCall(cmplogHookIns16, args); + + } else { + + IRB.CreateCall(cmplogHookInsN, args); + + } + + break; + + } } } - return true; + if (switches.size() || icomps.size()) + return true; + else + return false; } diff --git a/instrumentation/llvm-ngram-coverage.h b/instrumentation/llvm-ngram-coverage.h index 12b666e9..666839c8 100644 --- a/instrumentation/llvm-ngram-coverage.h +++ b/instrumentation/llvm-ngram-coverage.h @@ -1,7 +1,7 @@ #ifndef AFL_NGRAM_CONFIG_H #define AFL_NGRAM_CONFIG_H -#include "../config.h" +#include "types.h" #if (MAP_SIZE_POW2 <= 16) typedef u16 PREV_LOC_T; |