• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //===- FuzzerTraceState.cpp - Trace-based fuzzer mutator ------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 // This file implements a mutation algorithm based on instruction traces and
10 // on taint analysis feedback from DFSan.
11 //
12 // Instruction traces are special hooks inserted by the compiler around
13 // interesting instructions. Currently supported traces:
14 //   * __sanitizer_cov_trace_cmp -- inserted before every ICMP instruction,
15 //    receives the type, size and arguments of ICMP.
16 //
17 // Every time a traced event is intercepted we analyse the data involved
18 // in the event and suggest a mutation for future executions.
19 // For example if 4 bytes of data that derive from input bytes {4,5,6,7}
20 // are compared with a constant 12345,
21 // we try to insert 12345, 12344, 12346 into bytes
22 // {4,5,6,7} of the next fuzzed inputs.
23 //
24 // The fuzzer can work only with the traces, or with both traces and DFSan.
25 //
26 // DataFlowSanitizer (DFSan) is a tool for
27 // generalised dynamic data flow (taint) analysis:
28 // http://clang.llvm.org/docs/DataFlowSanitizer.html .
29 //
30 // The approach with DFSan-based fuzzing has some similarity to
31 // "Taint-based Directed Whitebox Fuzzing"
32 // by Vijay Ganesh & Tim Leek & Martin Rinard:
33 // http://dspace.mit.edu/openaccess-disseminate/1721.1/59320,
34 // but it uses a full blown LLVM IR taint analysis and separate instrumentation
35 // to analyze all of the "attack points" at once.
36 //
37 // Workflow with DFSan:
38 //   * lib/Fuzzer/Fuzzer*.cpp is compiled w/o any instrumentation.
39 //   * The code under test is compiled with DFSan *and* with instruction traces.
40 //   * Every call to HOOK(a,b) is replaced by DFSan with
41 //     __dfsw_HOOK(a, b, label(a), label(b)) so that __dfsw_HOOK
42 //     gets all the taint labels for the arguments.
43 //   * At the Fuzzer startup we assign a unique DFSan label
44 //     to every byte of the input string (Fuzzer::CurrentUnitData) so that
45 //     for any chunk of data we know which input bytes it has derived from.
46 //   * The __dfsw_* functions (implemented in this file) record the
47 //     parameters (i.e. the application data and the corresponding taint labels)
48 //     in a global state.
49 //
50 // Parts of this code will not function when DFSan is not linked in.
51 // Instead of using ifdefs and thus requiring a separate build of lib/Fuzzer
52 // we redeclare the dfsan_* interface functions as weak and check if they
53 // are nullptr before calling.
54 // If this approach proves to be useful we may add attribute(weak) to the
55 // dfsan declarations in dfsan_interface.h
56 //
57 // This module is in the "proof of concept" stage.
58 // It is capable of solving only the simplest puzzles
59 // like test/dfsan/DFSanSimpleCmpTest.cpp.
60 //===----------------------------------------------------------------------===//
61 
62 /* Example of manual usage (-fsanitize=dataflow is optional):
63 (
64   cd $LLVM/lib/Fuzzer/
65   clang  -fPIC -c -g -O2 -std=c++11 Fuzzer*.cpp
66   clang++ -O0 -std=c++11 -fsanitize-coverage=edge,trace-cmp \
67     -fsanitize=dataflow \
68     test/SimpleCmpTest.cpp Fuzzer*.o
69   ./a.out -use_traces=1
70 )
71 */
72 
73 #include "FuzzerDFSan.h"
74 #include "FuzzerInternal.h"
75 
76 #include <algorithm>
77 #include <cstring>
78 #include <thread>
79 #include <map>
80 
81 #if !LLVM_FUZZER_SUPPORTS_DFSAN
82 // Stubs for dfsan for platforms where dfsan does not exist and weak
83 // functions don't work.
84 extern "C" {
dfsan_create_label(const char * desc,void * userdata)85 dfsan_label dfsan_create_label(const char *desc, void *userdata) { return 0; }
dfsan_set_label(dfsan_label label,void * addr,size_t size)86 void dfsan_set_label(dfsan_label label, void *addr, size_t size) {}
dfsan_add_label(dfsan_label label,void * addr,size_t size)87 void dfsan_add_label(dfsan_label label, void *addr, size_t size) {}
dfsan_get_label_info(dfsan_label label)88 const struct dfsan_label_info *dfsan_get_label_info(dfsan_label label) {
89   return nullptr;
90 }
dfsan_read_label(const void * addr,size_t size)91 dfsan_label dfsan_read_label(const void *addr, size_t size) { return 0; }
92 }  // extern "C"
93 #endif  // !LLVM_FUZZER_SUPPORTS_DFSAN
94 
95 namespace fuzzer {
96 
97 // These values are copied from include/llvm/IR/InstrTypes.h.
98 // We do not include the LLVM headers here to remain independent.
99 // If these values ever change, an assertion in ComputeCmp will fail.
100 enum Predicate {
101   ICMP_EQ = 32,  ///< equal
102   ICMP_NE = 33,  ///< not equal
103   ICMP_UGT = 34, ///< unsigned greater than
104   ICMP_UGE = 35, ///< unsigned greater or equal
105   ICMP_ULT = 36, ///< unsigned less than
106   ICMP_ULE = 37, ///< unsigned less or equal
107   ICMP_SGT = 38, ///< signed greater than
108   ICMP_SGE = 39, ///< signed greater or equal
109   ICMP_SLT = 40, ///< signed less than
110   ICMP_SLE = 41, ///< signed less or equal
111 };
112 
113 template <class U, class S>
ComputeCmp(size_t CmpType,U Arg1,U Arg2)114 bool ComputeCmp(size_t CmpType, U Arg1, U Arg2) {
115   switch(CmpType) {
116     case ICMP_EQ : return Arg1 == Arg2;
117     case ICMP_NE : return Arg1 != Arg2;
118     case ICMP_UGT: return Arg1 > Arg2;
119     case ICMP_UGE: return Arg1 >= Arg2;
120     case ICMP_ULT: return Arg1 < Arg2;
121     case ICMP_ULE: return Arg1 <= Arg2;
122     case ICMP_SGT: return (S)Arg1 > (S)Arg2;
123     case ICMP_SGE: return (S)Arg1 >= (S)Arg2;
124     case ICMP_SLT: return (S)Arg1 < (S)Arg2;
125     case ICMP_SLE: return (S)Arg1 <= (S)Arg2;
126     default: assert(0 && "unsupported CmpType");
127   }
128   return false;
129 }
130 
ComputeCmp(size_t CmpSize,size_t CmpType,uint64_t Arg1,uint64_t Arg2)131 static bool ComputeCmp(size_t CmpSize, size_t CmpType, uint64_t Arg1,
132                        uint64_t Arg2) {
133   if (CmpSize == 8) return ComputeCmp<uint64_t, int64_t>(CmpType, Arg1, Arg2);
134   if (CmpSize == 4) return ComputeCmp<uint32_t, int32_t>(CmpType, Arg1, Arg2);
135   if (CmpSize == 2) return ComputeCmp<uint16_t, int16_t>(CmpType, Arg1, Arg2);
136   if (CmpSize == 1) return ComputeCmp<uint8_t, int8_t>(CmpType, Arg1, Arg2);
137   // Other size, ==
138   if (CmpType == ICMP_EQ) return Arg1 == Arg2;
139   // assert(0 && "unsupported cmp and type size combination");
140   return true;
141 }
142 
143 // As a simplification we use the range of input bytes instead of a set of input
144 // bytes.
145 struct LabelRange {
146   uint16_t Beg, End;  // Range is [Beg, End), thus Beg==End is an empty range.
147 
LabelRangefuzzer::LabelRange148   LabelRange(uint16_t Beg = 0, uint16_t End = 0) : Beg(Beg), End(End) {}
149 
Joinfuzzer::LabelRange150   static LabelRange Join(LabelRange LR1, LabelRange LR2) {
151     if (LR1.Beg == LR1.End) return LR2;
152     if (LR2.Beg == LR2.End) return LR1;
153     return {std::min(LR1.Beg, LR2.Beg), std::max(LR1.End, LR2.End)};
154   }
Joinfuzzer::LabelRange155   LabelRange &Join(LabelRange LR) {
156     return *this = Join(*this, LR);
157   }
Singletonfuzzer::LabelRange158   static LabelRange Singleton(const dfsan_label_info *LI) {
159     uint16_t Idx = (uint16_t)(uintptr_t)LI->userdata;
160     assert(Idx > 0);
161     return {(uint16_t)(Idx - 1), Idx};
162   }
163 };
164 
165 // For now, very simple: put Size bytes of Data at position Pos.
166 struct TraceBasedMutation {
167   uint32_t Pos;
168   Word W;
169 };
170 
171 // Declared as static globals for faster checks inside the hooks.
172 static bool RecordingTraces = false;
173 static bool RecordingMemcmp = false;
174 
175 class TraceState {
176 public:
TraceState(MutationDispatcher & MD,const FuzzingOptions & Options,const Fuzzer * F)177   TraceState(MutationDispatcher &MD, const FuzzingOptions &Options,
178              const Fuzzer *F)
179       : MD(MD), Options(Options), F(F) {}
180 
181   LabelRange GetLabelRange(dfsan_label L);
182   void DFSanCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType,
183                         uint64_t Arg1, uint64_t Arg2, dfsan_label L1,
184                         dfsan_label L2);
185   void DFSanMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
186                            const uint8_t *Data2, dfsan_label L1,
187                            dfsan_label L2);
188   void DFSanSwitchCallback(uint64_t PC, size_t ValSizeInBits, uint64_t Val,
189                            size_t NumCases, uint64_t *Cases, dfsan_label L);
190   void TraceCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType,
191                         uint64_t Arg1, uint64_t Arg2);
192   void TraceMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
193                            const uint8_t *Data2);
194 
195   void TraceSwitchCallback(uintptr_t PC, size_t ValSizeInBits, uint64_t Val,
196                            size_t NumCases, uint64_t *Cases);
197   int TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData,
198                           size_t DataSize);
199   int TryToAddDesiredData(const uint8_t *PresentData,
200                           const uint8_t *DesiredData, size_t DataSize);
201 
StartTraceRecording()202   void StartTraceRecording() {
203     if (!Options.UseTraces && !Options.UseMemcmp)
204       return;
205     RecordingTraces = Options.UseTraces;
206     RecordingMemcmp = Options.UseMemcmp;
207     NumMutations = 0;
208     MD.ClearAutoDictionary();
209   }
210 
StopTraceRecording()211   void StopTraceRecording() {
212     if (!RecordingTraces && !RecordingMemcmp)
213       return;
214     RecordingTraces = false;
215     RecordingMemcmp = false;
216     for (size_t i = 0; i < NumMutations; i++) {
217       auto &M = Mutations[i];
218       if (Options.Verbosity >= 2) {
219         AutoDictUnitCounts[M.W]++;
220         AutoDictAdds++;
221         if ((AutoDictAdds & (AutoDictAdds - 1)) == 0) {
222           typedef std::pair<size_t, Word> CU;
223           std::vector<CU> CountedUnits;
224           for (auto &I : AutoDictUnitCounts)
225             CountedUnits.push_back(std::make_pair(I.second, I.first));
226           std::sort(CountedUnits.begin(), CountedUnits.end(),
227                     [](const CU &a, const CU &b) { return a.first > b.first; });
228           Printf("AutoDict:\n");
229           for (auto &I : CountedUnits) {
230             Printf("   %zd ", I.first);
231             PrintASCII(I.second);
232             Printf("\n");
233           }
234         }
235       }
236       MD.AddWordToAutoDictionary(M.W, M.Pos);
237     }
238   }
239 
AddMutation(uint32_t Pos,uint32_t Size,const uint8_t * Data)240   void AddMutation(uint32_t Pos, uint32_t Size, const uint8_t *Data) {
241     if (NumMutations >= kMaxMutations) return;
242     auto &M = Mutations[NumMutations++];
243     M.Pos = Pos;
244     M.W.Set(Data, Size);
245   }
246 
AddMutation(uint32_t Pos,uint32_t Size,uint64_t Data)247   void AddMutation(uint32_t Pos, uint32_t Size, uint64_t Data) {
248     assert(Size <= sizeof(Data));
249     AddMutation(Pos, Size, reinterpret_cast<uint8_t*>(&Data));
250   }
251 
EnsureDfsanLabels(size_t Size)252   void EnsureDfsanLabels(size_t Size) {
253     for (; LastDfsanLabel < Size; LastDfsanLabel++) {
254       dfsan_label L = dfsan_create_label("input", (void *)(LastDfsanLabel + 1));
255       // We assume that no one else has called dfsan_create_label before.
256       if (L != LastDfsanLabel + 1) {
257         Printf("DFSan labels are not starting from 1, exiting\n");
258         exit(1);
259       }
260     }
261   }
262 
263  private:
IsTwoByteData(uint64_t Data)264   bool IsTwoByteData(uint64_t Data) {
265     int64_t Signed = static_cast<int64_t>(Data);
266     Signed >>= 16;
267     return Signed == 0 || Signed == -1L;
268   }
269 
270   // We don't want to create too many trace-based mutations as it is both
271   // expensive and useless. So after some number of mutations is collected,
272   // start rejecting some of them. The more there are mutations the more we
273   // reject.
WantToHandleOneMoreMutation()274   bool WantToHandleOneMoreMutation() {
275     const size_t FirstN = 64;
276     // Gladly handle first N mutations.
277     if (NumMutations <= FirstN) return true;
278     size_t Diff = NumMutations - FirstN;
279     size_t DiffLog = sizeof(long) * 8 - __builtin_clzl((long)Diff);
280     assert(DiffLog > 0 && DiffLog < 64);
281     bool WantThisOne = MD.GetRand()(1 << DiffLog) == 0;  // 1 out of DiffLog.
282     return WantThisOne;
283   }
284 
285   static const size_t kMaxMutations = 1 << 16;
286   size_t NumMutations;
287   TraceBasedMutation Mutations[kMaxMutations];
288   LabelRange LabelRanges[1 << (sizeof(dfsan_label) * 8)];
289   size_t LastDfsanLabel = 0;
290   MutationDispatcher &MD;
291   const FuzzingOptions Options;
292   const Fuzzer *F;
293   std::map<Word, size_t> AutoDictUnitCounts;
294   size_t AutoDictAdds = 0;
295 };
296 
297 
GetLabelRange(dfsan_label L)298 LabelRange TraceState::GetLabelRange(dfsan_label L) {
299   LabelRange &LR = LabelRanges[L];
300   if (LR.Beg < LR.End || L == 0)
301     return LR;
302   const dfsan_label_info *LI = dfsan_get_label_info(L);
303   if (LI->l1 || LI->l2)
304     return LR = LabelRange::Join(GetLabelRange(LI->l1), GetLabelRange(LI->l2));
305   return LR = LabelRange::Singleton(LI);
306 }
307 
DFSanCmpCallback(uintptr_t PC,size_t CmpSize,size_t CmpType,uint64_t Arg1,uint64_t Arg2,dfsan_label L1,dfsan_label L2)308 void TraceState::DFSanCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType,
309                                   uint64_t Arg1, uint64_t Arg2, dfsan_label L1,
310                                   dfsan_label L2) {
311   assert(ReallyHaveDFSan());
312   if (!RecordingTraces || !F->InFuzzingThread()) return;
313   if (L1 == 0 && L2 == 0)
314     return;  // Not actionable.
315   if (L1 != 0 && L2 != 0)
316     return;  // Probably still actionable.
317   bool Res = ComputeCmp(CmpSize, CmpType, Arg1, Arg2);
318   uint64_t Data = L1 ? Arg2 : Arg1;
319   LabelRange LR = L1 ? GetLabelRange(L1) : GetLabelRange(L2);
320 
321   for (size_t Pos = LR.Beg; Pos + CmpSize <= LR.End; Pos++) {
322     AddMutation(Pos, CmpSize, Data);
323     AddMutation(Pos, CmpSize, Data + 1);
324     AddMutation(Pos, CmpSize, Data - 1);
325   }
326 
327   if (CmpSize > (size_t)(LR.End - LR.Beg))
328     AddMutation(LR.Beg, (unsigned)(LR.End - LR.Beg), Data);
329 
330 
331   if (Options.Verbosity >= 3)
332     Printf("DFSanCmpCallback: PC %lx S %zd T %zd A1 %llx A2 %llx R %d L1 %d L2 "
333            "%d MU %zd\n",
334            PC, CmpSize, CmpType, Arg1, Arg2, Res, L1, L2, NumMutations);
335 }
336 
DFSanMemcmpCallback(size_t CmpSize,const uint8_t * Data1,const uint8_t * Data2,dfsan_label L1,dfsan_label L2)337 void TraceState::DFSanMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
338                                      const uint8_t *Data2, dfsan_label L1,
339                                      dfsan_label L2) {
340 
341   assert(ReallyHaveDFSan());
342   if (!RecordingMemcmp || !F->InFuzzingThread()) return;
343   if (L1 == 0 && L2 == 0)
344     return;  // Not actionable.
345   if (L1 != 0 && L2 != 0)
346     return;  // Probably still actionable.
347 
348   const uint8_t *Data = L1 ? Data2 : Data1;
349   LabelRange LR = L1 ? GetLabelRange(L1) : GetLabelRange(L2);
350   for (size_t Pos = LR.Beg; Pos + CmpSize <= LR.End; Pos++) {
351     AddMutation(Pos, CmpSize, Data);
352     if (Options.Verbosity >= 3)
353       Printf("DFSanMemcmpCallback: Pos %d Size %d\n", Pos, CmpSize);
354   }
355 }
356 
DFSanSwitchCallback(uint64_t PC,size_t ValSizeInBits,uint64_t Val,size_t NumCases,uint64_t * Cases,dfsan_label L)357 void TraceState::DFSanSwitchCallback(uint64_t PC, size_t ValSizeInBits,
358                                      uint64_t Val, size_t NumCases,
359                                      uint64_t *Cases, dfsan_label L) {
360   assert(ReallyHaveDFSan());
361   if (!RecordingTraces || !F->InFuzzingThread()) return;
362   if (!L) return;  // Not actionable.
363   LabelRange LR = GetLabelRange(L);
364   size_t ValSize = ValSizeInBits / 8;
365   bool TryShort = IsTwoByteData(Val);
366   for (size_t i = 0; i < NumCases; i++)
367     TryShort &= IsTwoByteData(Cases[i]);
368 
369   for (size_t Pos = LR.Beg; Pos + ValSize <= LR.End; Pos++)
370     for (size_t i = 0; i < NumCases; i++)
371       AddMutation(Pos, ValSize, Cases[i]);
372 
373   if (TryShort)
374     for (size_t Pos = LR.Beg; Pos + 2 <= LR.End; Pos++)
375       for (size_t i = 0; i < NumCases; i++)
376         AddMutation(Pos, 2, Cases[i]);
377 
378   if (Options.Verbosity >= 3)
379     Printf("DFSanSwitchCallback: PC %lx Val %zd SZ %zd # %zd L %d: {%d, %d} "
380            "TryShort %d\n",
381            PC, Val, ValSize, NumCases, L, LR.Beg, LR.End, TryShort);
382 }
383 
TryToAddDesiredData(uint64_t PresentData,uint64_t DesiredData,size_t DataSize)384 int TraceState::TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData,
385                                     size_t DataSize) {
386   if (NumMutations >= kMaxMutations || !WantToHandleOneMoreMutation()) return 0;
387   const uint8_t *UnitData;
388   auto UnitSize = F->GetCurrentUnitInFuzzingThead(&UnitData);
389   int Res = 0;
390   const uint8_t *Beg = UnitData;
391   const uint8_t *End = Beg + UnitSize;
392   for (const uint8_t *Cur = Beg; Cur < End; Cur++) {
393     Cur = (uint8_t *)memmem(Cur, End - Cur, &PresentData, DataSize);
394     if (!Cur)
395       break;
396     size_t Pos = Cur - Beg;
397     assert(Pos < UnitSize);
398     AddMutation(Pos, DataSize, DesiredData);
399     AddMutation(Pos, DataSize, DesiredData + 1);
400     AddMutation(Pos, DataSize, DesiredData - 1);
401     Res++;
402   }
403   return Res;
404 }
405 
TryToAddDesiredData(const uint8_t * PresentData,const uint8_t * DesiredData,size_t DataSize)406 int TraceState::TryToAddDesiredData(const uint8_t *PresentData,
407                                     const uint8_t *DesiredData,
408                                     size_t DataSize) {
409   if (NumMutations >= kMaxMutations || !WantToHandleOneMoreMutation()) return 0;
410   const uint8_t *UnitData;
411   auto UnitSize = F->GetCurrentUnitInFuzzingThead(&UnitData);
412   int Res = 0;
413   const uint8_t *Beg = UnitData;
414   const uint8_t *End = Beg + UnitSize;
415   for (const uint8_t *Cur = Beg; Cur < End; Cur++) {
416     Cur = (uint8_t *)memmem(Cur, End - Cur, PresentData, DataSize);
417     if (!Cur)
418       break;
419     size_t Pos = Cur - Beg;
420     assert(Pos < UnitSize);
421     AddMutation(Pos, DataSize, DesiredData);
422     Res++;
423   }
424   return Res;
425 }
426 
TraceCmpCallback(uintptr_t PC,size_t CmpSize,size_t CmpType,uint64_t Arg1,uint64_t Arg2)427 void TraceState::TraceCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType,
428                                   uint64_t Arg1, uint64_t Arg2) {
429   if (!RecordingTraces || !F->InFuzzingThread()) return;
430   if ((CmpType == ICMP_EQ || CmpType == ICMP_NE) && Arg1 == Arg2)
431     return;  // No reason to mutate.
432   int Added = 0;
433   Added += TryToAddDesiredData(Arg1, Arg2, CmpSize);
434   Added += TryToAddDesiredData(Arg2, Arg1, CmpSize);
435   if (!Added && CmpSize == 4 && IsTwoByteData(Arg1) && IsTwoByteData(Arg2)) {
436     Added += TryToAddDesiredData(Arg1, Arg2, 2);
437     Added += TryToAddDesiredData(Arg2, Arg1, 2);
438   }
439   if (Options.Verbosity >= 3 && Added)
440     Printf("TraceCmp %zd/%zd: %p %zd %zd\n", CmpSize, CmpType, PC, Arg1, Arg2);
441 }
442 
TraceMemcmpCallback(size_t CmpSize,const uint8_t * Data1,const uint8_t * Data2)443 void TraceState::TraceMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
444                                      const uint8_t *Data2) {
445   if (!RecordingMemcmp || !F->InFuzzingThread()) return;
446   CmpSize = std::min(CmpSize, Word::GetMaxSize());
447   int Added2 = TryToAddDesiredData(Data1, Data2, CmpSize);
448   int Added1 = TryToAddDesiredData(Data2, Data1, CmpSize);
449   if ((Added1 || Added2) && Options.Verbosity >= 3) {
450     Printf("MemCmp Added %d%d: ", Added1, Added2);
451     if (Added1) PrintASCII(Data1, CmpSize);
452     if (Added2) PrintASCII(Data2, CmpSize);
453     Printf("\n");
454   }
455 }
456 
TraceSwitchCallback(uintptr_t PC,size_t ValSizeInBits,uint64_t Val,size_t NumCases,uint64_t * Cases)457 void TraceState::TraceSwitchCallback(uintptr_t PC, size_t ValSizeInBits,
458                                      uint64_t Val, size_t NumCases,
459                                      uint64_t *Cases) {
460   if (!RecordingTraces || !F->InFuzzingThread()) return;
461   size_t ValSize = ValSizeInBits / 8;
462   bool TryShort = IsTwoByteData(Val);
463   for (size_t i = 0; i < NumCases; i++)
464     TryShort &= IsTwoByteData(Cases[i]);
465 
466   if (Options.Verbosity >= 3)
467     Printf("TraceSwitch: %p %zd # %zd; TryShort %d\n", PC, Val, NumCases,
468            TryShort);
469 
470   for (size_t i = 0; i < NumCases; i++) {
471     TryToAddDesiredData(Val, Cases[i], ValSize);
472     if (TryShort)
473       TryToAddDesiredData(Val, Cases[i], 2);
474   }
475 }
476 
477 static TraceState *TS;
478 
StartTraceRecording()479 void Fuzzer::StartTraceRecording() {
480   if (!TS) return;
481   TS->StartTraceRecording();
482 }
483 
StopTraceRecording()484 void Fuzzer::StopTraceRecording() {
485   if (!TS) return;
486   TS->StopTraceRecording();
487 }
488 
AssignTaintLabels(uint8_t * Data,size_t Size)489 void Fuzzer::AssignTaintLabels(uint8_t *Data, size_t Size) {
490   if (!Options.UseTraces && !Options.UseMemcmp) return;
491   if (!ReallyHaveDFSan()) return;
492   TS->EnsureDfsanLabels(Size);
493   for (size_t i = 0; i < Size; i++)
494     dfsan_set_label(i + 1, &Data[i], 1);
495 }
496 
InitializeTraceState()497 void Fuzzer::InitializeTraceState() {
498   if (!Options.UseTraces && !Options.UseMemcmp) return;
499   TS = new TraceState(MD, Options, this);
500 }
501 
InternalStrnlen(const char * S,size_t MaxLen)502 static size_t InternalStrnlen(const char *S, size_t MaxLen) {
503   size_t Len = 0;
504   for (; Len < MaxLen && S[Len]; Len++) {}
505   return Len;
506 }
507 
508 }  // namespace fuzzer
509 
510 using fuzzer::TS;
511 using fuzzer::RecordingTraces;
512 using fuzzer::RecordingMemcmp;
513 
514 extern "C" {
__dfsw___sanitizer_cov_trace_cmp(uint64_t SizeAndType,uint64_t Arg1,uint64_t Arg2,dfsan_label L0,dfsan_label L1,dfsan_label L2)515 void __dfsw___sanitizer_cov_trace_cmp(uint64_t SizeAndType, uint64_t Arg1,
516                                       uint64_t Arg2, dfsan_label L0,
517                                       dfsan_label L1, dfsan_label L2) {
518   if (!RecordingTraces) return;
519   assert(L0 == 0);
520   uintptr_t PC = reinterpret_cast<uintptr_t>(__builtin_return_address(0));
521   uint64_t CmpSize = (SizeAndType >> 32) / 8;
522   uint64_t Type = (SizeAndType << 32) >> 32;
523   TS->DFSanCmpCallback(PC, CmpSize, Type, Arg1, Arg2, L1, L2);
524 }
525 
__dfsw___sanitizer_cov_trace_switch(uint64_t Val,uint64_t * Cases,dfsan_label L1,dfsan_label L2)526 void __dfsw___sanitizer_cov_trace_switch(uint64_t Val, uint64_t *Cases,
527                                          dfsan_label L1, dfsan_label L2) {
528   if (!RecordingTraces) return;
529   uintptr_t PC = reinterpret_cast<uintptr_t>(__builtin_return_address(0));
530   TS->DFSanSwitchCallback(PC, Cases[1], Val, Cases[0], Cases+2, L1);
531 }
532 
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)533 void dfsan_weak_hook_memcmp(void *caller_pc, const void *s1, const void *s2,
534                             size_t n, dfsan_label s1_label,
535                             dfsan_label s2_label, dfsan_label n_label) {
536   if (!RecordingMemcmp) return;
537   dfsan_label L1 = dfsan_read_label(s1, n);
538   dfsan_label L2 = dfsan_read_label(s2, n);
539   TS->DFSanMemcmpCallback(n, reinterpret_cast<const uint8_t *>(s1),
540                           reinterpret_cast<const uint8_t *>(s2), L1, L2);
541 }
542 
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)543 void dfsan_weak_hook_strncmp(void *caller_pc, const char *s1, const char *s2,
544                              size_t n, dfsan_label s1_label,
545                              dfsan_label s2_label, dfsan_label n_label) {
546   if (!RecordingMemcmp) return;
547   n = std::min(n, fuzzer::InternalStrnlen(s1, n));
548   n = std::min(n, fuzzer::InternalStrnlen(s2, n));
549   dfsan_label L1 = dfsan_read_label(s1, n);
550   dfsan_label L2 = dfsan_read_label(s2, n);
551   TS->DFSanMemcmpCallback(n, reinterpret_cast<const uint8_t *>(s1),
552                           reinterpret_cast<const uint8_t *>(s2), L1, L2);
553 }
554 
dfsan_weak_hook_strcmp(void * caller_pc,const char * s1,const char * s2,dfsan_label s1_label,dfsan_label s2_label)555 void dfsan_weak_hook_strcmp(void *caller_pc, const char *s1, const char *s2,
556                             dfsan_label s1_label, dfsan_label s2_label) {
557   if (!RecordingMemcmp) return;
558   size_t Len1 = strlen(s1);
559   size_t Len2 = strlen(s2);
560   size_t N = std::min(Len1, Len2);
561   if (N <= 1) return;  // Not interesting.
562   dfsan_label L1 = dfsan_read_label(s1, Len1);
563   dfsan_label L2 = dfsan_read_label(s2, Len2);
564   TS->DFSanMemcmpCallback(N, reinterpret_cast<const uint8_t *>(s1),
565                           reinterpret_cast<const uint8_t *>(s2), L1, L2);
566 }
567 
568 // We may need to avoid defining weak hooks to stay compatible with older clang.
569 #ifndef LLVM_FUZZER_DEFINES_SANITIZER_WEAK_HOOOKS
570 # define LLVM_FUZZER_DEFINES_SANITIZER_WEAK_HOOOKS 1
571 #endif
572 
573 #if LLVM_FUZZER_DEFINES_SANITIZER_WEAK_HOOOKS
__sanitizer_weak_hook_memcmp(void * caller_pc,const void * s1,const void * s2,size_t n,int result)574 void __sanitizer_weak_hook_memcmp(void *caller_pc, const void *s1,
575                                   const void *s2, size_t n, int result) {
576   if (!RecordingMemcmp) return;
577   if (result == 0) return;  // No reason to mutate.
578   if (n <= 1) return;  // Not interesting.
579   TS->TraceMemcmpCallback(n, reinterpret_cast<const uint8_t *>(s1),
580                           reinterpret_cast<const uint8_t *>(s2));
581 }
582 
__sanitizer_weak_hook_strncmp(void * caller_pc,const char * s1,const char * s2,size_t n,int result)583 void __sanitizer_weak_hook_strncmp(void *caller_pc, const char *s1,
584                                    const char *s2, size_t n, int result) {
585   if (!RecordingMemcmp) return;
586   if (result == 0) return;  // No reason to mutate.
587   size_t Len1 = fuzzer::InternalStrnlen(s1, n);
588   size_t Len2 = fuzzer::InternalStrnlen(s2, n);
589   n = std::min(n, Len1);
590   n = std::min(n, Len2);
591   if (n <= 1) return;  // Not interesting.
592   TS->TraceMemcmpCallback(n, reinterpret_cast<const uint8_t *>(s1),
593                           reinterpret_cast<const uint8_t *>(s2));
594 }
595 
__sanitizer_weak_hook_strcmp(void * caller_pc,const char * s1,const char * s2,int result)596 void __sanitizer_weak_hook_strcmp(void *caller_pc, const char *s1,
597                                    const char *s2, int result) {
598   if (!RecordingMemcmp) return;
599   if (result == 0) return;  // No reason to mutate.
600   size_t Len1 = strlen(s1);
601   size_t Len2 = strlen(s2);
602   size_t N = std::min(Len1, Len2);
603   if (N <= 1) return;  // Not interesting.
604   TS->TraceMemcmpCallback(N, reinterpret_cast<const uint8_t *>(s1),
605                           reinterpret_cast<const uint8_t *>(s2));
606 }
607 
608 #endif  // LLVM_FUZZER_DEFINES_SANITIZER_WEAK_HOOOKS
609 
610 __attribute__((visibility("default")))
__sanitizer_cov_trace_cmp(uint64_t SizeAndType,uint64_t Arg1,uint64_t Arg2)611 void __sanitizer_cov_trace_cmp(uint64_t SizeAndType, uint64_t Arg1,
612                                uint64_t Arg2) {
613   if (!RecordingTraces) return;
614   uintptr_t PC = reinterpret_cast<uintptr_t>(__builtin_return_address(0));
615   uint64_t CmpSize = (SizeAndType >> 32) / 8;
616   uint64_t Type = (SizeAndType << 32) >> 32;
617   TS->TraceCmpCallback(PC, CmpSize, Type, Arg1, Arg2);
618 }
619 
620 __attribute__((visibility("default")))
__sanitizer_cov_trace_switch(uint64_t Val,uint64_t * Cases)621 void __sanitizer_cov_trace_switch(uint64_t Val, uint64_t *Cases) {
622   if (!RecordingTraces) return;
623   uintptr_t PC = reinterpret_cast<uintptr_t>(__builtin_return_address(0));
624   TS->TraceSwitchCallback(PC, Cases[1], Val, Cases[0], Cases + 2);
625 }
626 
627 }  // extern "C"
628