X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FFuzzer%2FFuzzerTraceState.cpp;h=c6f7c9e5a952dda858e6604f6a63a2b4b9a59f60;hp=c04084903809b82439d4817508fb71323d0c9494;hb=5707219e9494f14a01ad0ddf0e55a91af1723f7a;hpb=051ef86497b5c292465101af684d5d0708cc37bc diff --git a/lib/Fuzzer/FuzzerTraceState.cpp b/lib/Fuzzer/FuzzerTraceState.cpp index c0408490380..c6f7c9e5a95 100644 --- a/lib/Fuzzer/FuzzerTraceState.cpp +++ b/lib/Fuzzer/FuzzerTraceState.cpp @@ -77,7 +77,6 @@ #include #include -#include #include extern "C" { @@ -139,7 +138,9 @@ static bool ComputeCmp(size_t CmpSize, size_t CmpType, uint64_t Arg1, if (CmpSize == 4) return ComputeCmp(CmpType, Arg1, Arg2); if (CmpSize == 2) return ComputeCmp(CmpType, Arg1, Arg2); if (CmpSize == 1) return ComputeCmp(CmpType, Arg1, Arg2); - assert(0 && "unsupported type size"); + // Other size, == + if (CmpType == ICMP_EQ) return Arg1 == Arg2; + // assert(0 && "unsupported cmp and type size combination"); return true; } @@ -165,9 +166,37 @@ struct LabelRange { } }; -std::ostream &operator<<(std::ostream &os, const LabelRange &LR) { - return os << "[" << LR.Beg << "," << LR.End << ")"; -} +// A passport for a CMP site. We want to keep track of where the given CMP is +// and how many times it is evaluated to true or false. +struct CmpSitePassport { + uintptr_t PC; + size_t Counter[2]; + + bool IsInterestingCmpTarget() { + static const size_t kRareEnough = 50; + size_t C0 = Counter[0]; + size_t C1 = Counter[1]; + return C0 > kRareEnough * (C1 + 1) || C1 > kRareEnough * (C0 + 1); + } +}; + +// For now, just keep a simple imprecise hash table PC => CmpSitePassport. +// Potentially, will need to have a compiler support to have a precise mapping +// and also thread-safety. +struct CmpSitePassportTable { + static const size_t kSize = 99991; // Prime. + CmpSitePassport Passports[kSize]; + + CmpSitePassport *GetPassport(uintptr_t PC) { + uintptr_t Idx = PC & kSize; + CmpSitePassport *Res = &Passports[Idx]; + if (Res->PC == 0) // Not thread safe. + Res->PC = PC; + return Res->PC == PC ? Res : nullptr; + } +}; + +static CmpSitePassportTable CSPTable; // Zero initialized. // For now, very simple: put Size bytes of Data at position Pos. struct TraceBasedMutation { @@ -185,20 +214,26 @@ class TraceState { void DFSanCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType, uint64_t Arg1, uint64_t Arg2, dfsan_label L1, dfsan_label L2); - void TraceCmpCallback(size_t CmpSize, size_t CmpType, uint64_t Arg1, + void DFSanSwitchCallback(uint64_t PC, size_t ValSizeInBits, uint64_t Val, + size_t NumCases, uint64_t *Cases, dfsan_label L); + void TraceCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType, uint64_t Arg1, uint64_t Arg2); + + void TraceSwitchCallback(uintptr_t PC, size_t ValSizeInBits, uint64_t Val, + size_t NumCases, uint64_t *Cases); int TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData, size_t DataSize); void StartTraceRecording() { + if (!Options.UseTraces) return; RecordingTraces = true; Mutations.clear(); } - size_t StopTraceRecording() { + size_t StopTraceRecording(FuzzerRandomBase &Rand) { RecordingTraces = false; - std::random_shuffle(Mutations.begin(), Mutations.end()); - return Mutations.size(); + std::random_shuffle(Mutations.begin(), Mutations.end(), Rand); + return std::min(Mutations.size(), 128UL); } void ApplyTraceBasedMutation(size_t Idx, fuzzer::Unit *U); @@ -230,7 +265,7 @@ void TraceState::ApplyTraceBasedMutation(size_t Idx, fuzzer::Unit *U) { assert(Idx < Mutations.size()); auto &M = Mutations[Idx]; if (Options.Verbosity >= 3) - std::cerr << "TBM " << M.Pos << " " << M.Size << " " << M.Data << "\n"; + Printf("TBM %zd %zd %zd\n", M.Pos, M.Size, M.Data); if (M.Pos + M.Size > U->size()) return; memcpy(U->data() + M.Pos, &M.Data, M.Size); } @@ -259,16 +294,36 @@ void TraceState::DFSanCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType, if (Options.Verbosity >= 3) - std::cerr << "DFSAN:" - << " PC " << std::hex << PC << std::dec - << " S " << CmpSize - << " T " << CmpType - << " A1 " << Arg1 << " A2 " << Arg2 << " R " << Res - << " L" << L1 - << " L" << L2 - << " R" << LR - << " MU " << Mutations.size() - << "\n"; + Printf("DFSanCmpCallback: PC %lx S %zd T %zd A1 %llx A2 %llx R %d L1 %d L2 " + "%d MU %zd\n", + PC, CmpSize, CmpType, Arg1, Arg2, Res, L1, L2, Mutations.size()); +} + +void TraceState::DFSanSwitchCallback(uint64_t PC, size_t ValSizeInBits, + uint64_t Val, size_t NumCases, + uint64_t *Cases, dfsan_label L) { + assert(ReallyHaveDFSan()); + if (!RecordingTraces) return; + if (!L) return; // Not actionable. + LabelRange LR = GetLabelRange(L); + size_t ValSize = ValSizeInBits / 8; + bool TryShort = IsTwoByteData(Val); + for (size_t i = 0; i < NumCases; i++) + TryShort &= IsTwoByteData(Cases[i]); + + for (size_t Pos = LR.Beg; Pos + ValSize <= LR.End; Pos++) + for (size_t i = 0; i < NumCases; i++) + Mutations.push_back({Pos, ValSize, Cases[i]}); + + if (TryShort) + for (size_t Pos = LR.Beg; Pos + 2 <= LR.End; Pos++) + for (size_t i = 0; i < NumCases; i++) + Mutations.push_back({Pos, 2, Cases[i]}); + + if (Options.Verbosity >= 3) + Printf("DFSanSwitchCallback: PC %lx Val %zd SZ %zd # %zd L %d: {%d, %d} " + "TryShort %d\n", + PC, Val, ValSize, NumCases, L, LR.Beg, LR.End, TryShort); } int TraceState::TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData, @@ -280,9 +335,9 @@ int TraceState::TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData, Cur = (uint8_t *)memmem(Cur, End - Cur, &PresentData, DataSize); if (!Cur) break; - // std::cerr << "Cur " << (void*)Cur << "\n"; size_t Pos = Cur - Beg; assert(Pos < CurrentUnit.size()); + if (Mutations.size() > 100000U) return Res; // Just in case. Mutations.push_back({Pos, DataSize, DesiredData}); Mutations.push_back({Pos, DataSize, DesiredData + 1}); Mutations.push_back({Pos, DataSize, DesiredData - 1}); @@ -292,12 +347,20 @@ int TraceState::TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData, return Res; } -void TraceState::TraceCmpCallback(size_t CmpSize, size_t CmpType, uint64_t Arg1, +void TraceState::TraceCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType, uint64_t Arg1, uint64_t Arg2) { - if (!Options.UseTraces) return; + if (!RecordingTraces) return; int Added = 0; + CmpSitePassport *CSP = CSPTable.GetPassport(PC); + if (!CSP) return; + CSP->Counter[ComputeCmp(CmpSize, CmpType, Arg1, Arg2)]++; + size_t C0 = CSP->Counter[0]; + size_t C1 = CSP->Counter[1]; + // FIXME: is this a good idea or a bad? + // if (!CSP->IsInterestingCmpTarget()) + // return; if (Options.Verbosity >= 3) - std::cerr << "TraceCmp: " << Arg1 << " " << Arg2 << "\n"; + Printf("TraceCmp: %p %zd/%zd; %zd %zd\n", CSP->PC, C0, C1, Arg1, Arg2); Added += TryToAddDesiredData(Arg1, Arg2, CmpSize); Added += TryToAddDesiredData(Arg2, Arg1, CmpSize); if (!Added && CmpSize == 4 && IsTwoByteData(Arg1) && IsTwoByteData(Arg2)) { @@ -306,16 +369,40 @@ void TraceState::TraceCmpCallback(size_t CmpSize, size_t CmpType, uint64_t Arg1, } } +void TraceState::TraceSwitchCallback(uintptr_t PC, size_t ValSizeInBits, + uint64_t Val, size_t NumCases, + uint64_t *Cases) { + if (!RecordingTraces) return; + size_t ValSize = ValSizeInBits / 8; + bool TryShort = IsTwoByteData(Val); + for (size_t i = 0; i < NumCases; i++) + TryShort &= IsTwoByteData(Cases[i]); + + if (Options.Verbosity >= 3) + Printf("TraceSwitch: %p %zd # %zd; TryShort %d\n", PC, Val, NumCases, + TryShort); + + for (size_t i = 0; i < NumCases; i++) { + TryToAddDesiredData(Val, Cases[i], ValSize); + if (TryShort) + TryToAddDesiredData(Val, Cases[i], 2); + } + +} + static TraceState *TS; void Fuzzer::StartTraceRecording() { if (!TS) return; + if (ReallyHaveDFSan()) + for (size_t i = 0; i < static_cast(Options.MaxLen); i++) + dfsan_set_label(i + 1, &CurrentUnit[i], 1); TS->StartTraceRecording(); } size_t Fuzzer::StopTraceRecording() { if (!TS) return 0; - return TS->StopTraceRecording(); + return TS->StopTraceRecording(USF.GetRand()); } void Fuzzer::ApplyTraceBasedMutation(size_t Idx, Unit *U) { @@ -324,19 +411,24 @@ void Fuzzer::ApplyTraceBasedMutation(size_t Idx, Unit *U) { } void Fuzzer::InitializeTraceState() { - if (!Options.UseTraces && !Options.UseDFSan) return; + if (!Options.UseTraces) return; TS = new TraceState(Options, CurrentUnit); CurrentUnit.resize(Options.MaxLen); // The rest really requires DFSan. - if (!ReallyHaveDFSan() || !Options.UseDFSan) return; + if (!ReallyHaveDFSan()) return; for (size_t i = 0; i < static_cast(Options.MaxLen); i++) { dfsan_label L = dfsan_create_label("input", (void*)(i + 1)); // We assume that no one else has called dfsan_create_label before. assert(L == i + 1); - dfsan_set_label(L, &CurrentUnit[i], 1); } } +static size_t InternalStrnlen(const char *S, size_t MaxLen) { + size_t Len = 0; + for (; Len < MaxLen && S[Len]; Len++) {} + return Len; +} + } // namespace fuzzer using fuzzer::TS; @@ -345,7 +437,7 @@ extern "C" { void __dfsw___sanitizer_cov_trace_cmp(uint64_t SizeAndType, uint64_t Arg1, uint64_t Arg2, dfsan_label L0, dfsan_label L1, dfsan_label L2) { - assert(TS); + if (!TS) return; assert(L0 == 0); uintptr_t PC = reinterpret_cast(__builtin_return_address(0)); uint64_t CmpSize = (SizeAndType >> 32) / 8; @@ -353,12 +445,35 @@ void __dfsw___sanitizer_cov_trace_cmp(uint64_t SizeAndType, uint64_t Arg1, TS->DFSanCmpCallback(PC, CmpSize, Type, Arg1, Arg2, L1, L2); } +void __dfsw___sanitizer_cov_trace_switch(uint64_t Val, uint64_t *Cases, + dfsan_label L1, dfsan_label L2) { + if (!TS) return; + uintptr_t PC = reinterpret_cast(__builtin_return_address(0)); + TS->DFSanSwitchCallback(PC, Cases[1], Val, Cases[0], Cases+2, L1); +} + void dfsan_weak_hook_memcmp(void *caller_pc, const void *s1, const void *s2, size_t n, dfsan_label s1_label, dfsan_label s2_label, dfsan_label n_label) { - assert(TS); + if (!TS) return; + uintptr_t PC = reinterpret_cast(caller_pc); + uint64_t S1 = 0, S2 = 0; + // Simplification: handle only first 8 bytes. + memcpy(&S1, s1, std::min(n, sizeof(S1))); + memcpy(&S2, s2, std::min(n, sizeof(S2))); + dfsan_label L1 = dfsan_read_label(s1, n); + dfsan_label L2 = dfsan_read_label(s2, n); + TS->DFSanCmpCallback(PC, n, fuzzer::ICMP_EQ, S1, S2, L1, L2); +} + +void dfsan_weak_hook_strncmp(void *caller_pc, const char *s1, const char *s2, + size_t n, dfsan_label s1_label, + dfsan_label s2_label, dfsan_label n_label) { + if (!TS) return; uintptr_t PC = reinterpret_cast(caller_pc); uint64_t S1 = 0, S2 = 0; + n = std::min(n, fuzzer::InternalStrnlen(s1, n)); + n = std::min(n, fuzzer::InternalStrnlen(s2, n)); // Simplification: handle only first 8 bytes. memcpy(&S1, s1, std::min(n, sizeof(S1))); memcpy(&S2, s2, std::min(n, sizeof(S2))); @@ -367,12 +482,79 @@ void dfsan_weak_hook_memcmp(void *caller_pc, const void *s1, const void *s2, TS->DFSanCmpCallback(PC, n, fuzzer::ICMP_EQ, S1, S2, L1, L2); } +void dfsan_weak_hook_strcmp(void *caller_pc, const char *s1, const char *s2, + dfsan_label s1_label, dfsan_label s2_label) { + if (!TS) return; + uintptr_t PC = reinterpret_cast(caller_pc); + uint64_t S1 = 0, S2 = 0; + size_t Len1 = strlen(s1); + size_t Len2 = strlen(s2); + size_t N = std::min(Len1, Len2); + if (N <= 1) return; // Not interesting. + // Simplification: handle only first 8 bytes. + memcpy(&S1, s1, std::min(N, sizeof(S1))); + memcpy(&S2, s2, std::min(N, sizeof(S2))); + dfsan_label L1 = dfsan_read_label(s1, Len1); + dfsan_label L2 = dfsan_read_label(s2, Len2); + TS->DFSanCmpCallback(PC, N, fuzzer::ICMP_EQ, S1, S2, L1, L2); +} + +void __sanitizer_weak_hook_memcmp(void *caller_pc, const void *s1, + const void *s2, size_t n) { + if (!TS) return; + uintptr_t PC = reinterpret_cast(caller_pc); + uint64_t S1 = 0, S2 = 0; + // Simplification: handle only first 8 bytes. + memcpy(&S1, s1, std::min(n, sizeof(S1))); + memcpy(&S2, s2, std::min(n, sizeof(S2))); + TS->TraceCmpCallback(PC, n, fuzzer::ICMP_EQ, S1, S2); +} + +void __sanitizer_weak_hook_strncmp(void *caller_pc, const char *s1, + const char *s2, size_t n) { + if (!TS) return; + uintptr_t PC = reinterpret_cast(caller_pc); + uint64_t S1 = 0, S2 = 0; + size_t Len1 = fuzzer::InternalStrnlen(s1, n); + size_t Len2 = fuzzer::InternalStrnlen(s2, n); + n = std::min(n, Len1); + n = std::min(n, Len2); + if (n <= 1) return; // Not interesting. + // Simplification: handle only first 8 bytes. + memcpy(&S1, s1, std::min(n, sizeof(S1))); + memcpy(&S2, s2, std::min(n, sizeof(S2))); + TS->TraceCmpCallback(PC, n, fuzzer::ICMP_EQ, S1, S2); +} + +void __sanitizer_weak_hook_strcmp(void *caller_pc, const char *s1, + const char *s2) { + if (!TS) return; + uintptr_t PC = reinterpret_cast(caller_pc); + uint64_t S1 = 0, S2 = 0; + size_t Len1 = strlen(s1); + size_t Len2 = strlen(s2); + size_t N = std::min(Len1, Len2); + if (N <= 1) return; // Not interesting. + // Simplification: handle only first 8 bytes. + memcpy(&S1, s1, std::min(N, sizeof(S1))); + memcpy(&S2, s2, std::min(N, sizeof(S2))); + TS->TraceCmpCallback(PC, N, fuzzer::ICMP_EQ, S1, S2); +} + + void __sanitizer_cov_trace_cmp(uint64_t SizeAndType, uint64_t Arg1, uint64_t Arg2) { if (!TS) return; + uintptr_t PC = reinterpret_cast(__builtin_return_address(0)); uint64_t CmpSize = (SizeAndType >> 32) / 8; uint64_t Type = (SizeAndType << 32) >> 32; - TS->TraceCmpCallback(CmpSize, Type, Arg1, Arg2); + TS->TraceCmpCallback(PC, CmpSize, Type, Arg1, Arg2); +} + +void __sanitizer_cov_trace_switch(uint64_t Val, uint64_t *Cases) { + if (!TS) return; + uintptr_t PC = reinterpret_cast(__builtin_return_address(0)); + TS->TraceSwitchCallback(PC, Cases[1], Val, Cases[0], Cases + 2); } } // extern "C"