blob: 83130444a16b836e1a967131c5198034a86e0a5b [file] [log] [blame]
Kostya Serebryany22526252015-05-11 21:16:27 +00001//===- FuzzerTraceState.cpp - Trace-based fuzzer mutator ------------------===//
Kostya Serebryany16d03bd2015-03-30 22:09:51 +00002//
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//===----------------------------------------------------------------------===//
Kostya Serebryany22526252015-05-11 21:16:27 +00009// 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//
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000026// DataFlowSanitizer (DFSan) is a tool for
27// generalised dynamic data flow (taint) analysis:
28// http://clang.llvm.org/docs/DataFlowSanitizer.html .
29//
Kostya Serebryany22526252015-05-11 21:16:27 +000030// The approach with DFSan-based fuzzing has some similarity to
31// "Taint-based Directed Whitebox Fuzzing"
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000032// 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//
Kostya Serebryany22526252015-05-11 21:16:27 +000037// Workflow with DFSan:
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000038// * lib/Fuzzer/Fuzzer*.cpp is compiled w/o any instrumentation.
Kostya Serebryany22526252015-05-11 21:16:27 +000039// * The code under test is compiled with DFSan *and* with instruction traces.
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000040// * 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::CurrentUnit) so that for any
45// 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.
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000049//
Kostya Serebryany22526252015-05-11 21:16:27 +000050// Parts of this code will not function when DFSan is not linked in.
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000051// 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
Kostya Serebryany22526252015-05-11 21:16:27 +000062/* Example of manual usage (-fsanitize=dataflow is optional):
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000063(
64 cd $LLVM/lib/Fuzzer/
65 clang -fPIC -c -g -O2 -std=c++11 Fuzzer*.cpp
Alexey Samsonov21a33812015-05-07 23:33:24 +000066 clang++ -O0 -std=c++11 -fsanitize-coverage=edge,trace-cmp \
Kostya Serebryany3befe942015-05-06 22:47:24 +000067 -fsanitize=dataflow \
Kostya Serebryany65f50862015-09-10 18:48:38 +000068 test/SimpleCmpTest.cpp Fuzzer*.o
69 ./a.out -use_traces=1
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000070)
71*/
72
Kostya Serebryany65f50862015-09-10 18:48:38 +000073#include "FuzzerDFSan.h"
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000074#include "FuzzerInternal.h"
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000075
Kostya Serebryanybeb24c32015-05-07 21:02:11 +000076#include <algorithm>
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000077#include <cstring>
Kostya Serebryany226b7342016-01-06 00:03:35 +000078#include <thread>
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000079#include <unordered_map>
80
Kostya Serebryany65f50862015-09-10 18:48:38 +000081#if !LLVM_FUZZER_SUPPORTS_DFSAN
82// Stubs for dfsan for platforms where dfsan does not exist and weak
83// functions don't work.
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000084extern "C" {
Kostya Serebryany65f50862015-09-10 18:48:38 +000085dfsan_label dfsan_create_label(const char *desc, void *userdata) { return 0; }
86void dfsan_set_label(dfsan_label label, void *addr, size_t size) {}
87void dfsan_add_label(dfsan_label label, void *addr, size_t size) {}
88const struct dfsan_label_info *dfsan_get_label_info(dfsan_label label) {
89 return nullptr;
90}
91dfsan_label dfsan_read_label(const void *addr, size_t size) { return 0; }
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000092} // extern "C"
Kostya Serebryany65f50862015-09-10 18:48:38 +000093#endif // !LLVM_FUZZER_SUPPORTS_DFSAN
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000094
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +000095namespace fuzzer {
96
Kostya Serebryany16d03bd2015-03-30 22:09:51 +000097// 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.
100enum 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
113template <class U, class S>
114bool 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
131static 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);
Kostya Serebryanyb74ba422015-07-30 02:33:45 +0000137 // Other size, ==
138 if (CmpType == ICMP_EQ) return Arg1 == Arg2;
Kostya Serebryany8ce74242015-08-01 01:42:51 +0000139 // assert(0 && "unsupported cmp and type size combination");
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000140 return true;
141}
142
143// As a simplification we use the range of input bytes instead of a set of input
144// bytes.
145struct LabelRange {
146 uint16_t Beg, End; // Range is [Beg, End), thus Beg==End is an empty range.
147
148 LabelRange(uint16_t Beg = 0, uint16_t End = 0) : Beg(Beg), End(End) {}
149
150 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 }
155 LabelRange &Join(LabelRange LR) {
156 return *this = Join(*this, LR);
157 }
158 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
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000165// For now, very simple: put Size bytes of Data at position Pos.
166struct TraceBasedMutation {
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000167 static const size_t kMaxSize = 28;
168 uint32_t Pos : 24;
169 uint32_t Size : 8;
170 uint8_t Data[kMaxSize];
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000171};
172
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000173const size_t TraceBasedMutation::kMaxSize;
174
Kostya Serebryany22526252015-05-11 21:16:27 +0000175class TraceState {
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000176 public:
Kostya Serebryanyb65805a2016-01-09 03:08:58 +0000177 TraceState(UserSuppliedFuzzer &USF,
178 const Fuzzer::FuzzingOptions &Options, const Unit &CurrentUnit)
179 : USF(USF), Options(Options), CurrentUnit(CurrentUnit) {
Kostya Serebryany226b7342016-01-06 00:03:35 +0000180 // Current trace collection is not thread-friendly and it probably
181 // does not have to be such, but at least we should not crash in presence
182 // of threads. So, just ignore all traces coming from all threads but one.
183 IsMyThread = true;
184 }
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000185
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000186 LabelRange GetLabelRange(dfsan_label L);
187 void DFSanCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType,
188 uint64_t Arg1, uint64_t Arg2, dfsan_label L1,
189 dfsan_label L2);
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000190 void DFSanMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
191 const uint8_t *Data2, dfsan_label L1,
192 dfsan_label L2);
Kostya Serebryanycd6a4662015-07-31 17:05:05 +0000193 void DFSanSwitchCallback(uint64_t PC, size_t ValSizeInBits, uint64_t Val,
194 size_t NumCases, uint64_t *Cases, dfsan_label L);
Kostya Serebryanye641dd62015-09-04 22:32:25 +0000195 void TraceCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType,
196 uint64_t Arg1, uint64_t Arg2);
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000197 void TraceMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
198 const uint8_t *Data2);
Kostya Serebryanyfb7d8d92015-07-31 01:33:06 +0000199
200 void TraceSwitchCallback(uintptr_t PC, size_t ValSizeInBits, uint64_t Val,
201 size_t NumCases, uint64_t *Cases);
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000202 int TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData,
203 size_t DataSize);
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000204 int TryToAddDesiredData(const uint8_t *PresentData,
205 const uint8_t *DesiredData, size_t DataSize);
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000206
207 void StartTraceRecording() {
Kostya Serebryany8817e862015-05-11 23:25:28 +0000208 if (!Options.UseTraces) return;
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000209 RecordingTraces = true;
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000210 NumMutations = 0;
Kostya Serebryanyb65805a2016-01-09 03:08:58 +0000211 USF.GetMD().ClearAutoDictionary();
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000212 }
213
Kostya Serebryanyb65805a2016-01-09 03:08:58 +0000214 void StopTraceRecording() {
215 if (!RecordingTraces) return;
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000216 RecordingTraces = false;
Kostya Serebryanyb65805a2016-01-09 03:08:58 +0000217 for (size_t i = 0; i < NumMutations; i++) {
218 auto &M = Mutations[i];
219 USF.GetMD().AddWordToAutoDictionary(Unit(M.Data, M.Data + M.Size), M.Pos);
220 }
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000221 }
222
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000223 void AddMutation(uint32_t Pos, uint32_t Size, const uint8_t *Data) {
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000224 if (NumMutations >= kMaxMutations) return;
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000225 assert(Size <= TraceBasedMutation::kMaxSize);
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000226 auto &M = Mutations[NumMutations++];
227 M.Pos = Pos;
228 M.Size = Size;
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000229 memcpy(M.Data, Data, Size);
230 }
231
232 void AddMutation(uint32_t Pos, uint32_t Size, uint64_t Data) {
233 assert(Size <= sizeof(Data));
234 AddMutation(Pos, Size, reinterpret_cast<uint8_t*>(&Data));
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000235 }
236
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000237 private:
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000238 bool IsTwoByteData(uint64_t Data) {
239 int64_t Signed = static_cast<int64_t>(Data);
240 Signed >>= 16;
241 return Signed == 0 || Signed == -1L;
242 }
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000243 bool RecordingTraces = false;
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000244 static const size_t kMaxMutations = 1 << 16;
245 size_t NumMutations;
246 TraceBasedMutation Mutations[kMaxMutations];
Kostya Serebryany4d623222015-11-18 01:08:30 +0000247 LabelRange LabelRanges[1 << (sizeof(dfsan_label) * 8)];
Kostya Serebryanyb65805a2016-01-09 03:08:58 +0000248 UserSuppliedFuzzer &USF;
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000249 const Fuzzer::FuzzingOptions &Options;
250 const Unit &CurrentUnit;
Kostya Serebryany226b7342016-01-06 00:03:35 +0000251 static thread_local bool IsMyThread;
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000252};
253
Kostya Serebryany226b7342016-01-06 00:03:35 +0000254thread_local bool TraceState::IsMyThread;
255
Kostya Serebryany22526252015-05-11 21:16:27 +0000256LabelRange TraceState::GetLabelRange(dfsan_label L) {
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000257 LabelRange &LR = LabelRanges[L];
258 if (LR.Beg < LR.End || L == 0)
259 return LR;
260 const dfsan_label_info *LI = dfsan_get_label_info(L);
261 if (LI->l1 || LI->l2)
262 return LR = LabelRange::Join(GetLabelRange(LI->l1), GetLabelRange(LI->l2));
263 return LR = LabelRange::Singleton(LI);
264}
265
Kostya Serebryany22526252015-05-11 21:16:27 +0000266void TraceState::DFSanCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType,
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000267 uint64_t Arg1, uint64_t Arg2, dfsan_label L1,
268 dfsan_label L2) {
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000269 assert(ReallyHaveDFSan());
Kostya Serebryany226b7342016-01-06 00:03:35 +0000270 if (!RecordingTraces || !IsMyThread) return;
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000271 if (L1 == 0 && L2 == 0)
272 return; // Not actionable.
273 if (L1 != 0 && L2 != 0)
274 return; // Probably still actionable.
275 bool Res = ComputeCmp(CmpSize, CmpType, Arg1, Arg2);
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000276 uint64_t Data = L1 ? Arg2 : Arg1;
277 LabelRange LR = L1 ? GetLabelRange(L1) : GetLabelRange(L2);
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000278
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000279 for (size_t Pos = LR.Beg; Pos + CmpSize <= LR.End; Pos++) {
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000280 AddMutation(Pos, CmpSize, Data);
281 AddMutation(Pos, CmpSize, Data + 1);
282 AddMutation(Pos, CmpSize, Data - 1);
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000283 }
284
285 if (CmpSize > LR.End - LR.Beg)
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000286 AddMutation(LR.Beg, (unsigned)(LR.End - LR.Beg), Data);
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000287
288
289 if (Options.Verbosity >= 3)
Kostya Serebryanyae7df1c2015-07-28 01:25:00 +0000290 Printf("DFSanCmpCallback: PC %lx S %zd T %zd A1 %llx A2 %llx R %d L1 %d L2 "
291 "%d MU %zd\n",
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000292 PC, CmpSize, CmpType, Arg1, Arg2, Res, L1, L2, NumMutations);
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000293}
294
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000295void TraceState::DFSanMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
296 const uint8_t *Data2, dfsan_label L1,
297 dfsan_label L2) {
298
299 assert(ReallyHaveDFSan());
300 if (!RecordingTraces || !IsMyThread) return;
301 if (L1 == 0 && L2 == 0)
302 return; // Not actionable.
303 if (L1 != 0 && L2 != 0)
304 return; // Probably still actionable.
305
306 const uint8_t *Data = L1 ? Data2 : Data1;
307 LabelRange LR = L1 ? GetLabelRange(L1) : GetLabelRange(L2);
308 for (size_t Pos = LR.Beg; Pos + CmpSize <= LR.End; Pos++) {
309 AddMutation(Pos, CmpSize, Data);
310 if (Options.Verbosity >= 3)
311 Printf("DFSanMemcmpCallback: Pos %d Size %d\n", Pos, CmpSize);
312 }
313}
314
Kostya Serebryanycd6a4662015-07-31 17:05:05 +0000315void TraceState::DFSanSwitchCallback(uint64_t PC, size_t ValSizeInBits,
316 uint64_t Val, size_t NumCases,
317 uint64_t *Cases, dfsan_label L) {
318 assert(ReallyHaveDFSan());
Kostya Serebryany226b7342016-01-06 00:03:35 +0000319 if (!RecordingTraces || !IsMyThread) return;
Kostya Serebryanycd6a4662015-07-31 17:05:05 +0000320 if (!L) return; // Not actionable.
321 LabelRange LR = GetLabelRange(L);
322 size_t ValSize = ValSizeInBits / 8;
Kostya Serebryanyfe7e41e2015-07-31 20:58:55 +0000323 bool TryShort = IsTwoByteData(Val);
324 for (size_t i = 0; i < NumCases; i++)
325 TryShort &= IsTwoByteData(Cases[i]);
326
327 for (size_t Pos = LR.Beg; Pos + ValSize <= LR.End; Pos++)
328 for (size_t i = 0; i < NumCases; i++)
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000329 AddMutation(Pos, ValSize, Cases[i]);
Kostya Serebryanyfe7e41e2015-07-31 20:58:55 +0000330
331 if (TryShort)
332 for (size_t Pos = LR.Beg; Pos + 2 <= LR.End; Pos++)
333 for (size_t i = 0; i < NumCases; i++)
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000334 AddMutation(Pos, 2, Cases[i]);
Kostya Serebryanyfe7e41e2015-07-31 20:58:55 +0000335
Kostya Serebryanycd6a4662015-07-31 17:05:05 +0000336 if (Options.Verbosity >= 3)
Kostya Serebryanyfe7e41e2015-07-31 20:58:55 +0000337 Printf("DFSanSwitchCallback: PC %lx Val %zd SZ %zd # %zd L %d: {%d, %d} "
338 "TryShort %d\n",
339 PC, Val, ValSize, NumCases, L, LR.Beg, LR.End, TryShort);
Kostya Serebryanycd6a4662015-07-31 17:05:05 +0000340}
341
Kostya Serebryany22526252015-05-11 21:16:27 +0000342int TraceState::TryToAddDesiredData(uint64_t PresentData, uint64_t DesiredData,
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000343 size_t DataSize) {
344 int Res = 0;
345 const uint8_t *Beg = CurrentUnit.data();
346 const uint8_t *End = Beg + CurrentUnit.size();
Kostya Serebryanye641dd62015-09-04 22:32:25 +0000347 for (const uint8_t *Cur = Beg; Cur < End; Cur++) {
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000348 Cur = (uint8_t *)memmem(Cur, End - Cur, &PresentData, DataSize);
349 if (!Cur)
350 break;
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000351 size_t Pos = Cur - Beg;
352 assert(Pos < CurrentUnit.size());
Kostya Serebryanye7583d22016-01-09 00:38:40 +0000353 AddMutation(Pos, DataSize, DesiredData);
354 AddMutation(Pos, DataSize, DesiredData + 1);
355 AddMutation(Pos, DataSize, DesiredData - 1);
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000356 Res++;
357 }
358 return Res;
359}
360
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000361int TraceState::TryToAddDesiredData(const uint8_t *PresentData,
362 const uint8_t *DesiredData,
363 size_t DataSize) {
364 int Res = 0;
365 const uint8_t *Beg = CurrentUnit.data();
366 const uint8_t *End = Beg + CurrentUnit.size();
367 for (const uint8_t *Cur = Beg; Cur < End; Cur++) {
368 Cur = (uint8_t *)memmem(Cur, End - Cur, PresentData, DataSize);
369 if (!Cur)
370 break;
371 size_t Pos = Cur - Beg;
372 assert(Pos < CurrentUnit.size());
373 AddMutation(Pos, DataSize, DesiredData);
374 Res++;
375 }
376 return Res;
377}
378
Kostya Serebryanye641dd62015-09-04 22:32:25 +0000379void TraceState::TraceCmpCallback(uintptr_t PC, size_t CmpSize, size_t CmpType,
380 uint64_t Arg1, uint64_t Arg2) {
Kostya Serebryany226b7342016-01-06 00:03:35 +0000381 if (!RecordingTraces || !IsMyThread) return;
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000382 int Added = 0;
383 if (Options.Verbosity >= 3)
Kostya Serebryany4b82de22015-09-08 20:40:10 +0000384 Printf("TraceCmp %zd/%zd: %p %zd %zd\n", CmpSize, CmpType, PC, Arg1, Arg2);
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000385 Added += TryToAddDesiredData(Arg1, Arg2, CmpSize);
386 Added += TryToAddDesiredData(Arg2, Arg1, CmpSize);
387 if (!Added && CmpSize == 4 && IsTwoByteData(Arg1) && IsTwoByteData(Arg2)) {
388 Added += TryToAddDesiredData(Arg1, Arg2, 2);
389 Added += TryToAddDesiredData(Arg2, Arg1, 2);
390 }
391}
392
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000393void TraceState::TraceMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
394 const uint8_t *Data2) {
395 if (!RecordingTraces || !IsMyThread) return;
396 CmpSize = std::min(CmpSize, TraceBasedMutation::kMaxSize);
397 TryToAddDesiredData(Data1, Data2, CmpSize);
398 TryToAddDesiredData(Data2, Data1, CmpSize);
399}
400
Kostya Serebryanyfb7d8d92015-07-31 01:33:06 +0000401void TraceState::TraceSwitchCallback(uintptr_t PC, size_t ValSizeInBits,
402 uint64_t Val, size_t NumCases,
403 uint64_t *Cases) {
Kostya Serebryany226b7342016-01-06 00:03:35 +0000404 if (!RecordingTraces || !IsMyThread) return;
Kostya Serebryanyfe7e41e2015-07-31 20:58:55 +0000405 size_t ValSize = ValSizeInBits / 8;
406 bool TryShort = IsTwoByteData(Val);
Kostya Serebryanyfb7d8d92015-07-31 01:33:06 +0000407 for (size_t i = 0; i < NumCases; i++)
Kostya Serebryanyfe7e41e2015-07-31 20:58:55 +0000408 TryShort &= IsTwoByteData(Cases[i]);
409
410 if (Options.Verbosity >= 3)
411 Printf("TraceSwitch: %p %zd # %zd; TryShort %d\n", PC, Val, NumCases,
412 TryShort);
413
414 for (size_t i = 0; i < NumCases; i++) {
415 TryToAddDesiredData(Val, Cases[i], ValSize);
416 if (TryShort)
417 TryToAddDesiredData(Val, Cases[i], 2);
418 }
Kostya Serebryanyfb7d8d92015-07-31 01:33:06 +0000419}
420
Kostya Serebryany22526252015-05-11 21:16:27 +0000421static TraceState *TS;
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000422
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000423void Fuzzer::StartTraceRecording() {
Kostya Serebryany22526252015-05-11 21:16:27 +0000424 if (!TS) return;
Kostya Serebryany4cc10d42015-08-05 23:02:57 +0000425 if (ReallyHaveDFSan())
426 for (size_t i = 0; i < static_cast<size_t>(Options.MaxLen); i++)
427 dfsan_set_label(i + 1, &CurrentUnit[i], 1);
Kostya Serebryany22526252015-05-11 21:16:27 +0000428 TS->StartTraceRecording();
Kostya Serebryanybeb24c32015-05-07 21:02:11 +0000429}
430
Kostya Serebryanyb65805a2016-01-09 03:08:58 +0000431void Fuzzer::StopTraceRecording() {
432 if (!TS) return;
433 TS->StopTraceRecording();
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000434}
435
Kostya Serebryany22526252015-05-11 21:16:27 +0000436void Fuzzer::InitializeTraceState() {
Kostya Serebryanyd8c54722015-05-12 01:58:34 +0000437 if (!Options.UseTraces) return;
Kostya Serebryanyb65805a2016-01-09 03:08:58 +0000438 TS = new TraceState(USF, Options, CurrentUnit);
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000439 CurrentUnit.resize(Options.MaxLen);
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000440 // The rest really requires DFSan.
Kostya Serebryanyd8c54722015-05-12 01:58:34 +0000441 if (!ReallyHaveDFSan()) return;
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000442 for (size_t i = 0; i < static_cast<size_t>(Options.MaxLen); i++) {
443 dfsan_label L = dfsan_create_label("input", (void*)(i + 1));
444 // We assume that no one else has called dfsan_create_label before.
Kostya Serebryanyd46369d2015-08-05 23:44:42 +0000445 if (L != i + 1) {
446 Printf("DFSan labels are not starting from 1, exiting\n");
447 exit(1);
448 }
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000449 }
450}
451
Kostya Serebryanyc9dc96b2015-07-30 21:22:22 +0000452static size_t InternalStrnlen(const char *S, size_t MaxLen) {
453 size_t Len = 0;
454 for (; Len < MaxLen && S[Len]; Len++) {}
455 return Len;
456}
457
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000458} // namespace fuzzer
459
Kostya Serebryany22526252015-05-11 21:16:27 +0000460using fuzzer::TS;
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000461
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000462extern "C" {
463void __dfsw___sanitizer_cov_trace_cmp(uint64_t SizeAndType, uint64_t Arg1,
464 uint64_t Arg2, dfsan_label L0,
465 dfsan_label L1, dfsan_label L2) {
Kostya Serebryany3fe76822015-05-29 20:31:17 +0000466 if (!TS) return;
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000467 assert(L0 == 0);
468 uintptr_t PC = reinterpret_cast<uintptr_t>(__builtin_return_address(0));
469 uint64_t CmpSize = (SizeAndType >> 32) / 8;
470 uint64_t Type = (SizeAndType << 32) >> 32;
Kostya Serebryany22526252015-05-11 21:16:27 +0000471 TS->DFSanCmpCallback(PC, CmpSize, Type, Arg1, Arg2, L1, L2);
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000472}
Kostya Serebryanya407dde2015-05-07 00:11:33 +0000473
Kostya Serebryanycd6a4662015-07-31 17:05:05 +0000474void __dfsw___sanitizer_cov_trace_switch(uint64_t Val, uint64_t *Cases,
475 dfsan_label L1, dfsan_label L2) {
476 if (!TS) return;
477 uintptr_t PC = reinterpret_cast<uintptr_t>(__builtin_return_address(0));
478 TS->DFSanSwitchCallback(PC, Cases[1], Val, Cases[0], Cases+2, L1);
479}
480
Kostya Serebryanya407dde2015-05-07 00:11:33 +0000481void dfsan_weak_hook_memcmp(void *caller_pc, const void *s1, const void *s2,
482 size_t n, dfsan_label s1_label,
483 dfsan_label s2_label, dfsan_label n_label) {
Kostya Serebryany3fe76822015-05-29 20:31:17 +0000484 if (!TS) return;
Kostya Serebryanya407dde2015-05-07 00:11:33 +0000485 dfsan_label L1 = dfsan_read_label(s1, n);
486 dfsan_label L2 = dfsan_read_label(s2, n);
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000487 TS->DFSanMemcmpCallback(n, reinterpret_cast<const uint8_t *>(s1),
488 reinterpret_cast<const uint8_t *>(s2), L1, L2);
Kostya Serebryanya407dde2015-05-07 00:11:33 +0000489}
Kostya Serebryany7d470cf2015-05-07 18:32:29 +0000490
Kostya Serebryanyb74ba422015-07-30 02:33:45 +0000491void dfsan_weak_hook_strncmp(void *caller_pc, const char *s1, const char *s2,
492 size_t n, dfsan_label s1_label,
493 dfsan_label s2_label, dfsan_label n_label) {
Kostya Serebryanyc9dc96b2015-07-30 21:22:22 +0000494 if (!TS) return;
Kostya Serebryanyc9dc96b2015-07-30 21:22:22 +0000495 n = std::min(n, fuzzer::InternalStrnlen(s1, n));
496 n = std::min(n, fuzzer::InternalStrnlen(s2, n));
Kostya Serebryanyc9dc96b2015-07-30 21:22:22 +0000497 dfsan_label L1 = dfsan_read_label(s1, n);
498 dfsan_label L2 = dfsan_read_label(s2, n);
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000499 TS->DFSanMemcmpCallback(n, reinterpret_cast<const uint8_t *>(s1),
500 reinterpret_cast<const uint8_t *>(s2), L1, L2);
Kostya Serebryanyb74ba422015-07-30 02:33:45 +0000501}
502
Kostya Serebryany7f4227d2015-08-05 18:23:01 +0000503void dfsan_weak_hook_strcmp(void *caller_pc, const char *s1, const char *s2,
504 dfsan_label s1_label, dfsan_label s2_label) {
505 if (!TS) return;
Kostya Serebryany7f4227d2015-08-05 18:23:01 +0000506 size_t Len1 = strlen(s1);
507 size_t Len2 = strlen(s2);
508 size_t N = std::min(Len1, Len2);
509 if (N <= 1) return; // Not interesting.
Kostya Serebryany7f4227d2015-08-05 18:23:01 +0000510 dfsan_label L1 = dfsan_read_label(s1, Len1);
511 dfsan_label L2 = dfsan_read_label(s2, Len2);
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000512 TS->DFSanMemcmpCallback(N, reinterpret_cast<const uint8_t *>(s1),
513 reinterpret_cast<const uint8_t *>(s2), L1, L2);
Kostya Serebryany7f4227d2015-08-05 18:23:01 +0000514}
515
Kostya Serebryany0e776a22015-07-30 01:34:58 +0000516void __sanitizer_weak_hook_memcmp(void *caller_pc, const void *s1,
517 const void *s2, size_t n) {
518 if (!TS) return;
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000519 if (n <= 1) return; // Not interesting.
520 TS->TraceMemcmpCallback(n, reinterpret_cast<const uint8_t *>(s1),
521 reinterpret_cast<const uint8_t *>(s2));
Kostya Serebryanyb74ba422015-07-30 02:33:45 +0000522}
523
524void __sanitizer_weak_hook_strncmp(void *caller_pc, const char *s1,
525 const char *s2, size_t n) {
Kostya Serebryanyc9dc96b2015-07-30 21:22:22 +0000526 if (!TS) return;
Kostya Serebryanycd6a4662015-07-31 17:05:05 +0000527 size_t Len1 = fuzzer::InternalStrnlen(s1, n);
528 size_t Len2 = fuzzer::InternalStrnlen(s2, n);
529 n = std::min(n, Len1);
530 n = std::min(n, Len2);
531 if (n <= 1) return; // Not interesting.
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000532 TS->TraceMemcmpCallback(n, reinterpret_cast<const uint8_t *>(s1),
533 reinterpret_cast<const uint8_t *>(s2));
Kostya Serebryany0e776a22015-07-30 01:34:58 +0000534}
535
Kostya Serebryany7f4227d2015-08-05 18:23:01 +0000536void __sanitizer_weak_hook_strcmp(void *caller_pc, const char *s1,
537 const char *s2) {
538 if (!TS) return;
Kostya Serebryany7f4227d2015-08-05 18:23:01 +0000539 size_t Len1 = strlen(s1);
540 size_t Len2 = strlen(s2);
541 size_t N = std::min(Len1, Len2);
542 if (N <= 1) return; // Not interesting.
Kostya Serebryanyc5733162016-01-09 01:39:55 +0000543 TS->TraceMemcmpCallback(N, reinterpret_cast<const uint8_t *>(s1),
544 reinterpret_cast<const uint8_t *>(s2));
Kostya Serebryany7f4227d2015-08-05 18:23:01 +0000545}
546
Kostya Serebryany3287d7a2015-09-30 22:22:37 +0000547__attribute__((visibility("default")))
Kostya Serebryany7d470cf2015-05-07 18:32:29 +0000548void __sanitizer_cov_trace_cmp(uint64_t SizeAndType, uint64_t Arg1,
549 uint64_t Arg2) {
Kostya Serebryany22526252015-05-11 21:16:27 +0000550 if (!TS) return;
Kostya Serebryany35959592015-07-28 00:59:53 +0000551 uintptr_t PC = reinterpret_cast<uintptr_t>(__builtin_return_address(0));
Kostya Serebryany5a99ecb2015-05-11 20:51:19 +0000552 uint64_t CmpSize = (SizeAndType >> 32) / 8;
553 uint64_t Type = (SizeAndType << 32) >> 32;
Kostya Serebryany35959592015-07-28 00:59:53 +0000554 TS->TraceCmpCallback(PC, CmpSize, Type, Arg1, Arg2);
Kostya Serebryany7d470cf2015-05-07 18:32:29 +0000555}
556
Kostya Serebryany3287d7a2015-09-30 22:22:37 +0000557__attribute__((visibility("default")))
Kostya Serebryanyfb7d8d92015-07-31 01:33:06 +0000558void __sanitizer_cov_trace_switch(uint64_t Val, uint64_t *Cases) {
559 if (!TS) return;
560 uintptr_t PC = reinterpret_cast<uintptr_t>(__builtin_return_address(0));
561 TS->TraceSwitchCallback(PC, Cases[1], Val, Cases[0], Cases + 2);
562}
563
Kostya Serebryany16d03bd2015-03-30 22:09:51 +0000564} // extern "C"