X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FFuzzer%2FFuzzerTraceState.cpp;h=36fc6dcfd23e8393fe17bfc91a57cb870d5b6ae3;hp=75b77ca9bf048a44734c69103de167a6bc3e6454;hb=38c36e3216d68c8907fb52001a64b5930097a6e2;hpb=33a9a09cd77dfb4282da4c4e76d37f825c773a50 diff --git a/lib/Fuzzer/FuzzerTraceState.cpp b/lib/Fuzzer/FuzzerTraceState.cpp index 75b77ca9bf0..36fc6dcfd23 100644 --- a/lib/Fuzzer/FuzzerTraceState.cpp +++ b/lib/Fuzzer/FuzzerTraceState.cpp @@ -46,8 +46,6 @@ // * The __dfsw_* functions (implemented in this file) record the // parameters (i.e. the application data and the corresponding taint labels) // in a global state. -// * Fuzzer::ApplyTraceBasedMutation() tries to use the data recorded -// by __dfsw_* hooks to guide the fuzzing towards new application states. // // Parts of this code will not function when DFSan is not linked in. // Instead of using ifdefs and thus requiring a separate build of lib/Fuzzer @@ -67,37 +65,35 @@ clang -fPIC -c -g -O2 -std=c++11 Fuzzer*.cpp clang++ -O0 -std=c++11 -fsanitize-coverage=edge,trace-cmp \ -fsanitize=dataflow \ - test/dfsan/DFSanSimpleCmpTest.cpp Fuzzer*.o - ./a.out + test/SimpleCmpTest.cpp Fuzzer*.o + ./a.out -use_traces=1 ) */ +#include "FuzzerDFSan.h" #include "FuzzerInternal.h" -#include #include #include +#include #include +#if !LLVM_FUZZER_SUPPORTS_DFSAN +// Stubs for dfsan for platforms where dfsan does not exist and weak +// functions don't work. extern "C" { -__attribute__((weak)) -dfsan_label dfsan_create_label(const char *desc, void *userdata); -__attribute__((weak)) -void dfsan_set_label(dfsan_label label, void *addr, size_t size); -__attribute__((weak)) -void dfsan_add_label(dfsan_label label, void *addr, size_t size); -__attribute__((weak)) -const struct dfsan_label_info *dfsan_get_label_info(dfsan_label label); -__attribute__((weak)) -dfsan_label dfsan_read_label(const void *addr, size_t size); +dfsan_label dfsan_create_label(const char *desc, void *userdata) { return 0; } +void dfsan_set_label(dfsan_label label, void *addr, size_t size) {} +void dfsan_add_label(dfsan_label label, void *addr, size_t size) {} +const struct dfsan_label_info *dfsan_get_label_info(dfsan_label label) { + return nullptr; +} +dfsan_label dfsan_read_label(const void *addr, size_t size) { return 0; } } // extern "C" +#endif // !LLVM_FUZZER_SUPPORTS_DFSAN namespace fuzzer { -static bool ReallyHaveDFSan() { - return &dfsan_create_label != nullptr; -} - // These values are copied from include/llvm/IR/InstrTypes.h. // We do not include the LLVM headers here to remain independent. // If these values ever change, an assertion in ComputeCmp will fail. @@ -138,7 +134,9 @@ 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; } @@ -166,38 +164,94 @@ struct LabelRange { // For now, very simple: put Size bytes of Data at position Pos. struct TraceBasedMutation { - size_t Pos; - size_t Size; - uint64_t Data; + static const size_t kMaxSize = 28; + uint32_t Pos : 24; + uint32_t Size : 8; + uint8_t Data[kMaxSize]; }; +static void PrintDataByte(uint8_t Byte) { + if (Byte == '\\') + Printf("\\\\"); + else if (Byte == '"') + Printf("\\\""); + else if (Byte >= 32 && Byte < 127) + Printf("%c", Byte); + else + Printf("\\x02x", Byte); +} + +static void PrintData(const uint8_t *Data, size_t Size) { + Printf("\""); + for (size_t i = 0; i < Size; i++) { + PrintDataByte(Data[i]); + } + Printf("\""); +} + +const size_t TraceBasedMutation::kMaxSize; + class TraceState { public: - TraceState(const Fuzzer::FuzzingOptions &Options, const Unit &CurrentUnit) - : Options(Options), CurrentUnit(CurrentUnit) {} + TraceState(UserSuppliedFuzzer &USF, + const Fuzzer::FuzzingOptions &Options, const Unit &CurrentUnit) + : USF(USF), Options(Options), CurrentUnit(CurrentUnit) { + // Current trace collection is not thread-friendly and it probably + // does not have to be such, but at least we should not crash in presence + // of threads. So, just ignore all traces coming from all threads but one. + IsMyThread = true; + } LabelRange GetLabelRange(dfsan_label L); 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, - uint64_t Arg2); + void DFSanMemcmpCallback(size_t CmpSize, const uint8_t *Data1, + const uint8_t *Data2, dfsan_label L1, + dfsan_label L2); + 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 TraceMemcmpCallback(size_t CmpSize, const uint8_t *Data1, + const uint8_t *Data2); + + 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); + int TryToAddDesiredData(const uint8_t *PresentData, + const uint8_t *DesiredData, size_t DataSize); void StartTraceRecording() { if (!Options.UseTraces) return; RecordingTraces = true; - Mutations.clear(); + NumMutations = 0; + USF.GetMD().ClearAutoDictionary(); } - size_t StopTraceRecording(FuzzerRandomBase &Rand) { + void StopTraceRecording() { + if (!RecordingTraces) return; RecordingTraces = false; - std::random_shuffle(Mutations.begin(), Mutations.end(), Rand); - return Mutations.size(); + for (size_t i = 0; i < NumMutations; i++) { + auto &M = Mutations[i]; + USF.GetMD().AddWordToAutoDictionary(Unit(M.Data, M.Data + M.Size), M.Pos); + } + } + + void AddMutation(uint32_t Pos, uint32_t Size, const uint8_t *Data) { + if (NumMutations >= kMaxMutations) return; + assert(Size <= TraceBasedMutation::kMaxSize); + auto &M = Mutations[NumMutations++]; + M.Pos = Pos; + M.Size = Size; + memcpy(M.Data, Data, Size); } - void ApplyTraceBasedMutation(size_t Idx, fuzzer::Unit *U); + void AddMutation(uint32_t Pos, uint32_t Size, uint64_t Data) { + assert(Size <= sizeof(Data)); + AddMutation(Pos, Size, reinterpret_cast(&Data)); + } private: bool IsTwoByteData(uint64_t Data) { @@ -206,12 +260,18 @@ class TraceState { return Signed == 0 || Signed == -1L; } bool RecordingTraces = false; - std::vector Mutations; - LabelRange LabelRanges[1 << (sizeof(dfsan_label) * 8)] = {}; + static const size_t kMaxMutations = 1 << 16; + size_t NumMutations; + TraceBasedMutation Mutations[kMaxMutations]; + LabelRange LabelRanges[1 << (sizeof(dfsan_label) * 8)]; + UserSuppliedFuzzer &USF; const Fuzzer::FuzzingOptions &Options; const Unit &CurrentUnit; + static thread_local bool IsMyThread; }; +thread_local bool TraceState::IsMyThread; + LabelRange TraceState::GetLabelRange(dfsan_label L) { LabelRange &LR = LabelRanges[L]; if (LR.Beg < LR.End || L == 0) @@ -222,20 +282,11 @@ LabelRange TraceState::GetLabelRange(dfsan_label L) { return LR = LabelRange::Singleton(LI); } -void TraceState::ApplyTraceBasedMutation(size_t Idx, fuzzer::Unit *U) { - assert(Idx < Mutations.size()); - auto &M = Mutations[Idx]; - if (Options.Verbosity >= 3) - 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); -} - void TraceState::DFSanCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType, uint64_t Arg1, uint64_t Arg2, dfsan_label L1, dfsan_label L2) { assert(ReallyHaveDFSan()); - if (!RecordingTraces) return; + if (!RecordingTraces || !IsMyThread) return; if (L1 == 0 && L2 == 0) return; // Not actionable. if (L1 != 0 && L2 != 0) @@ -245,18 +296,66 @@ void TraceState::DFSanCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType, LabelRange LR = L1 ? GetLabelRange(L1) : GetLabelRange(L2); for (size_t Pos = LR.Beg; Pos + CmpSize <= LR.End; Pos++) { - Mutations.push_back({Pos, CmpSize, Data}); - Mutations.push_back({Pos, CmpSize, Data + 1}); - Mutations.push_back({Pos, CmpSize, Data - 1}); + AddMutation(Pos, CmpSize, Data); + AddMutation(Pos, CmpSize, Data + 1); + AddMutation(Pos, CmpSize, Data - 1); } if (CmpSize > LR.End - LR.Beg) - Mutations.push_back({LR.Beg, (unsigned)(LR.End - LR.Beg), Data}); + AddMutation(LR.Beg, (unsigned)(LR.End - LR.Beg), Data); + + + if (Options.Verbosity >= 3) + 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, NumMutations); +} + +void TraceState::DFSanMemcmpCallback(size_t CmpSize, const uint8_t *Data1, + const uint8_t *Data2, dfsan_label L1, + dfsan_label L2) { + assert(ReallyHaveDFSan()); + if (!RecordingTraces || !IsMyThread) return; + if (L1 == 0 && L2 == 0) + return; // Not actionable. + if (L1 != 0 && L2 != 0) + return; // Probably still actionable. + + const uint8_t *Data = L1 ? Data2 : Data1; + LabelRange LR = L1 ? GetLabelRange(L1) : GetLabelRange(L2); + for (size_t Pos = LR.Beg; Pos + CmpSize <= LR.End; Pos++) { + AddMutation(Pos, CmpSize, Data); + if (Options.Verbosity >= 3) + Printf("DFSanMemcmpCallback: Pos %d Size %d\n", Pos, CmpSize); + } +} + +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 || !IsMyThread) 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++) + AddMutation(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++) + AddMutation(Pos, 2, Cases[i]); if (Options.Verbosity >= 3) - Printf("DFSAN: 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()); + 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, @@ -264,27 +363,44 @@ int TraceState::TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData, int Res = 0; const uint8_t *Beg = CurrentUnit.data(); const uint8_t *End = Beg + CurrentUnit.size(); - for (const uint8_t *Cur = Beg; Cur < End; Cur += DataSize) { + for (const uint8_t *Cur = Beg; Cur < End; Cur++) { Cur = (uint8_t *)memmem(Cur, End - Cur, &PresentData, DataSize); if (!Cur) break; size_t Pos = Cur - Beg; assert(Pos < CurrentUnit.size()); - Mutations.push_back({Pos, DataSize, DesiredData}); - Mutations.push_back({Pos, DataSize, DesiredData + 1}); - Mutations.push_back({Pos, DataSize, DesiredData - 1}); - Cur += DataSize; + AddMutation(Pos, DataSize, DesiredData); + AddMutation(Pos, DataSize, DesiredData + 1); + AddMutation(Pos, DataSize, DesiredData - 1); Res++; } return Res; } -void TraceState::TraceCmpCallback(size_t CmpSize, size_t CmpType, uint64_t Arg1, - uint64_t Arg2) { - if (!RecordingTraces) return; +int TraceState::TryToAddDesiredData(const uint8_t *PresentData, + const uint8_t *DesiredData, + size_t DataSize) { + int Res = 0; + const uint8_t *Beg = CurrentUnit.data(); + const uint8_t *End = Beg + CurrentUnit.size(); + for (const uint8_t *Cur = Beg; Cur < End; Cur++) { + Cur = (uint8_t *)memmem(Cur, End - Cur, PresentData, DataSize); + if (!Cur) + break; + size_t Pos = Cur - Beg; + assert(Pos < CurrentUnit.size()); + AddMutation(Pos, DataSize, DesiredData); + Res++; + } + return Res; +} + +void TraceState::TraceCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType, + uint64_t Arg1, uint64_t Arg2) { + if (!RecordingTraces || !IsMyThread) return; int Added = 0; if (Options.Verbosity >= 3) - Printf("TraceCmp: %zd %zd\n", Arg1, Arg2); + Printf("TraceCmp %zd/%zd: %p %zd %zd\n", CmpSize, CmpType, PC, Arg1, Arg2); Added += TryToAddDesiredData(Arg1, Arg2, CmpSize); Added += TryToAddDesiredData(Arg2, Arg1, CmpSize); if (!Added && CmpSize == 4 && IsTwoByteData(Arg1) && IsTwoByteData(Arg2)) { @@ -293,37 +409,77 @@ void TraceState::TraceCmpCallback(size_t CmpSize, size_t CmpType, uint64_t Arg1, } } +void TraceState::TraceMemcmpCallback(size_t CmpSize, const uint8_t *Data1, + const uint8_t *Data2) { + if (!RecordingTraces || !IsMyThread) return; + CmpSize = std::min(CmpSize, TraceBasedMutation::kMaxSize); + int Added2 = TryToAddDesiredData(Data1, Data2, CmpSize); + int Added1 = TryToAddDesiredData(Data2, Data1, CmpSize); + if ((Added1 || Added2) && Options.Verbosity >= 3) { + Printf("MemCmp Added %d%d: ", Added1, Added2); + if (Added1) PrintData(Data1, CmpSize); + if (Added2) PrintData(Data2, CmpSize); + Printf("\n"); + } +} + +void TraceState::TraceSwitchCallback(uintptr_t PC, size_t ValSizeInBits, + uint64_t Val, size_t NumCases, + uint64_t *Cases) { + if (!RecordingTraces || !IsMyThread) 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(USF.GetRand()); -} - -void Fuzzer::ApplyTraceBasedMutation(size_t Idx, Unit *U) { - assert(TS); - TS->ApplyTraceBasedMutation(Idx, U); +void Fuzzer::StopTraceRecording() { + if (!TS) return; + TS->StopTraceRecording(); } void Fuzzer::InitializeTraceState() { if (!Options.UseTraces) return; - TS = new TraceState(Options, CurrentUnit); + TS = new TraceState(USF, Options, CurrentUnit); CurrentUnit.resize(Options.MaxLen); // The rest really requires DFSan. 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); + if (L != i + 1) { + Printf("DFSan labels are not starting from 1, exiting\n"); + exit(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; @@ -340,26 +496,94 @@ 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) { 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); + TS->DFSanMemcmpCallback(n, reinterpret_cast(s1), + reinterpret_cast(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; + n = std::min(n, fuzzer::InternalStrnlen(s1, n)); + n = std::min(n, fuzzer::InternalStrnlen(s2, n)); + dfsan_label L1 = dfsan_read_label(s1, n); + dfsan_label L2 = dfsan_read_label(s2, n); + TS->DFSanMemcmpCallback(n, reinterpret_cast(s1), + reinterpret_cast(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; + size_t Len1 = strlen(s1); + size_t Len2 = strlen(s2); + size_t N = std::min(Len1, Len2); + if (N <= 1) return; // Not interesting. + dfsan_label L1 = dfsan_read_label(s1, Len1); + dfsan_label L2 = dfsan_read_label(s2, Len2); + TS->DFSanMemcmpCallback(N, reinterpret_cast(s1), + reinterpret_cast(s2), L1, L2); +} + +void __sanitizer_weak_hook_memcmp(void *caller_pc, const void *s1, + const void *s2, size_t n) { + if (!TS) return; + if (n <= 1) return; // Not interesting. + TS->TraceMemcmpCallback(n, reinterpret_cast(s1), + reinterpret_cast(s2)); +} + +void __sanitizer_weak_hook_strncmp(void *caller_pc, const char *s1, + const char *s2, size_t n) { + if (!TS) return; + 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. + TS->TraceMemcmpCallback(n, reinterpret_cast(s1), + reinterpret_cast(s2)); +} + +void __sanitizer_weak_hook_strcmp(void *caller_pc, const char *s1, + const char *s2) { + if (!TS) return; + size_t Len1 = strlen(s1); + size_t Len2 = strlen(s2); + size_t N = std::min(Len1, Len2); + if (N <= 1) return; // Not interesting. + TS->TraceMemcmpCallback(N, reinterpret_cast(s1), + reinterpret_cast(s2)); +} + +__attribute__((visibility("default"))) 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); +} + +__attribute__((visibility("default"))) +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"