george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 1 | //===- FuzzerMutate.cpp - Mutate a test input -----------------------------===// |
| 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 | // Mutate a test input. |
| 10 | //===----------------------------------------------------------------------===// |
| 11 | |
| 12 | #include "FuzzerMutate.h" |
| 13 | #include "FuzzerCorpus.h" |
| 14 | #include "FuzzerDefs.h" |
| 15 | #include "FuzzerExtFunctions.h" |
| 16 | #include "FuzzerIO.h" |
| 17 | #include "FuzzerOptions.h" |
| 18 | |
| 19 | namespace fuzzer { |
| 20 | |
| 21 | const size_t Dictionary::kMaxDictSize; |
| 22 | |
| 23 | static void PrintASCII(const Word &W, const char *PrintAfter) { |
| 24 | PrintASCII(W.data(), W.size(), PrintAfter); |
| 25 | } |
| 26 | |
| 27 | MutationDispatcher::MutationDispatcher(Random &Rand, |
| 28 | const FuzzingOptions &Options) |
| 29 | : Rand(Rand), Options(Options) { |
| 30 | DefaultMutators.insert( |
| 31 | DefaultMutators.begin(), |
| 32 | { |
| 33 | {&MutationDispatcher::Mutate_EraseBytes, "EraseBytes"}, |
| 34 | {&MutationDispatcher::Mutate_InsertByte, "InsertByte"}, |
| 35 | {&MutationDispatcher::Mutate_InsertRepeatedBytes, |
| 36 | "InsertRepeatedBytes"}, |
| 37 | {&MutationDispatcher::Mutate_ChangeByte, "ChangeByte"}, |
| 38 | {&MutationDispatcher::Mutate_ChangeBit, "ChangeBit"}, |
| 39 | {&MutationDispatcher::Mutate_ShuffleBytes, "ShuffleBytes"}, |
| 40 | {&MutationDispatcher::Mutate_ChangeASCIIInteger, "ChangeASCIIInt"}, |
| 41 | {&MutationDispatcher::Mutate_ChangeBinaryInteger, "ChangeBinInt"}, |
| 42 | {&MutationDispatcher::Mutate_CopyPart, "CopyPart"}, |
| 43 | {&MutationDispatcher::Mutate_CrossOver, "CrossOver"}, |
| 44 | {&MutationDispatcher::Mutate_AddWordFromManualDictionary, |
| 45 | "ManualDict"}, |
| 46 | {&MutationDispatcher::Mutate_AddWordFromPersistentAutoDictionary, |
| 47 | "PersAutoDict"}, |
| 48 | }); |
| 49 | if(Options.UseCmp) |
| 50 | DefaultMutators.push_back( |
| 51 | {&MutationDispatcher::Mutate_AddWordFromTORC, "CMP"}); |
| 52 | |
| 53 | if (EF->LLVMFuzzerCustomMutator) |
| 54 | Mutators.push_back({&MutationDispatcher::Mutate_Custom, "Custom"}); |
| 55 | else |
| 56 | Mutators = DefaultMutators; |
| 57 | |
| 58 | if (EF->LLVMFuzzerCustomCrossOver) |
| 59 | Mutators.push_back( |
| 60 | {&MutationDispatcher::Mutate_CustomCrossOver, "CustomCrossOver"}); |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 61 | |
| 62 | // Initialize mutation statistic counters. |
| 63 | TotalMutations.resize(Mutators.size(), 0); |
| 64 | UsefulMutations.resize(Mutators.size(), 0); |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 65 | } |
| 66 | |
| 67 | static char RandCh(Random &Rand) { |
| 68 | if (Rand.RandBool()) return Rand(256); |
morehouse | a3809d5 | 2018-01-30 18:25:55 +0000 | [diff] [blame] | 69 | const char Special[] = "!*'();:@&=+$,/?%#[]012Az-`~.\xff\x00"; |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 70 | return Special[Rand(sizeof(Special) - 1)]; |
| 71 | } |
| 72 | |
| 73 | size_t MutationDispatcher::Mutate_Custom(uint8_t *Data, size_t Size, |
| 74 | size_t MaxSize) { |
| 75 | return EF->LLVMFuzzerCustomMutator(Data, Size, MaxSize, Rand.Rand()); |
| 76 | } |
| 77 | |
| 78 | size_t MutationDispatcher::Mutate_CustomCrossOver(uint8_t *Data, size_t Size, |
| 79 | size_t MaxSize) { |
| 80 | if (!Corpus || Corpus->size() < 2 || Size == 0) |
| 81 | return 0; |
| 82 | size_t Idx = Rand(Corpus->size()); |
| 83 | const Unit &Other = (*Corpus)[Idx]; |
| 84 | if (Other.empty()) |
| 85 | return 0; |
| 86 | CustomCrossOverInPlaceHere.resize(MaxSize); |
| 87 | auto &U = CustomCrossOverInPlaceHere; |
| 88 | size_t NewSize = EF->LLVMFuzzerCustomCrossOver( |
| 89 | Data, Size, Other.data(), Other.size(), U.data(), U.size(), Rand.Rand()); |
| 90 | if (!NewSize) |
| 91 | return 0; |
| 92 | assert(NewSize <= MaxSize && "CustomCrossOver returned overisized unit"); |
| 93 | memcpy(Data, U.data(), NewSize); |
| 94 | return NewSize; |
| 95 | } |
| 96 | |
| 97 | size_t MutationDispatcher::Mutate_ShuffleBytes(uint8_t *Data, size_t Size, |
| 98 | size_t MaxSize) { |
| 99 | if (Size > MaxSize || Size == 0) return 0; |
| 100 | size_t ShuffleAmount = |
| 101 | Rand(std::min(Size, (size_t)8)) + 1; // [1,8] and <= Size. |
| 102 | size_t ShuffleStart = Rand(Size - ShuffleAmount); |
| 103 | assert(ShuffleStart + ShuffleAmount <= Size); |
| 104 | std::shuffle(Data + ShuffleStart, Data + ShuffleStart + ShuffleAmount, Rand); |
| 105 | return Size; |
| 106 | } |
| 107 | |
| 108 | size_t MutationDispatcher::Mutate_EraseBytes(uint8_t *Data, size_t Size, |
| 109 | size_t MaxSize) { |
| 110 | if (Size <= 1) return 0; |
| 111 | size_t N = Rand(Size / 2) + 1; |
| 112 | assert(N < Size); |
| 113 | size_t Idx = Rand(Size - N + 1); |
| 114 | // Erase Data[Idx:Idx+N]. |
| 115 | memmove(Data + Idx, Data + Idx + N, Size - Idx - N); |
| 116 | // Printf("Erase: %zd %zd => %zd; Idx %zd\n", N, Size, Size - N, Idx); |
| 117 | return Size - N; |
| 118 | } |
| 119 | |
| 120 | size_t MutationDispatcher::Mutate_InsertByte(uint8_t *Data, size_t Size, |
| 121 | size_t MaxSize) { |
| 122 | if (Size >= MaxSize) return 0; |
| 123 | size_t Idx = Rand(Size + 1); |
| 124 | // Insert new value at Data[Idx]. |
| 125 | memmove(Data + Idx + 1, Data + Idx, Size - Idx); |
| 126 | Data[Idx] = RandCh(Rand); |
| 127 | return Size + 1; |
| 128 | } |
| 129 | |
| 130 | size_t MutationDispatcher::Mutate_InsertRepeatedBytes(uint8_t *Data, |
| 131 | size_t Size, |
| 132 | size_t MaxSize) { |
| 133 | const size_t kMinBytesToInsert = 3; |
| 134 | if (Size + kMinBytesToInsert >= MaxSize) return 0; |
| 135 | size_t MaxBytesToInsert = std::min(MaxSize - Size, (size_t)128); |
| 136 | size_t N = Rand(MaxBytesToInsert - kMinBytesToInsert + 1) + kMinBytesToInsert; |
| 137 | assert(Size + N <= MaxSize && N); |
| 138 | size_t Idx = Rand(Size + 1); |
| 139 | // Insert new values at Data[Idx]. |
| 140 | memmove(Data + Idx + N, Data + Idx, Size - Idx); |
| 141 | // Give preference to 0x00 and 0xff. |
| 142 | uint8_t Byte = Rand.RandBool() ? Rand(256) : (Rand.RandBool() ? 0 : 255); |
| 143 | for (size_t i = 0; i < N; i++) |
| 144 | Data[Idx + i] = Byte; |
| 145 | return Size + N; |
| 146 | } |
| 147 | |
| 148 | size_t MutationDispatcher::Mutate_ChangeByte(uint8_t *Data, size_t Size, |
| 149 | size_t MaxSize) { |
| 150 | if (Size > MaxSize) return 0; |
| 151 | size_t Idx = Rand(Size); |
| 152 | Data[Idx] = RandCh(Rand); |
| 153 | return Size; |
| 154 | } |
| 155 | |
| 156 | size_t MutationDispatcher::Mutate_ChangeBit(uint8_t *Data, size_t Size, |
| 157 | size_t MaxSize) { |
| 158 | if (Size > MaxSize) return 0; |
| 159 | size_t Idx = Rand(Size); |
| 160 | Data[Idx] ^= 1 << Rand(8); |
| 161 | return Size; |
| 162 | } |
| 163 | |
| 164 | size_t MutationDispatcher::Mutate_AddWordFromManualDictionary(uint8_t *Data, |
| 165 | size_t Size, |
| 166 | size_t MaxSize) { |
| 167 | return AddWordFromDictionary(ManualDictionary, Data, Size, MaxSize); |
| 168 | } |
| 169 | |
| 170 | size_t MutationDispatcher::ApplyDictionaryEntry(uint8_t *Data, size_t Size, |
| 171 | size_t MaxSize, |
| 172 | DictionaryEntry &DE) { |
| 173 | const Word &W = DE.GetW(); |
| 174 | bool UsePositionHint = DE.HasPositionHint() && |
| 175 | DE.GetPositionHint() + W.size() < Size && |
| 176 | Rand.RandBool(); |
| 177 | if (Rand.RandBool()) { // Insert W. |
| 178 | if (Size + W.size() > MaxSize) return 0; |
| 179 | size_t Idx = UsePositionHint ? DE.GetPositionHint() : Rand(Size + 1); |
| 180 | memmove(Data + Idx + W.size(), Data + Idx, Size - Idx); |
| 181 | memcpy(Data + Idx, W.data(), W.size()); |
| 182 | Size += W.size(); |
| 183 | } else { // Overwrite some bytes with W. |
| 184 | if (W.size() > Size) return 0; |
| 185 | size_t Idx = UsePositionHint ? DE.GetPositionHint() : Rand(Size - W.size()); |
| 186 | memcpy(Data + Idx, W.data(), W.size()); |
| 187 | } |
| 188 | return Size; |
| 189 | } |
| 190 | |
| 191 | // Somewhere in the past we have observed a comparison instructions |
| 192 | // with arguments Arg1 Arg2. This function tries to guess a dictionary |
| 193 | // entry that will satisfy that comparison. |
| 194 | // It first tries to find one of the arguments (possibly swapped) in the |
| 195 | // input and if it succeeds it creates a DE with a position hint. |
| 196 | // Otherwise it creates a DE with one of the arguments w/o a position hint. |
| 197 | DictionaryEntry MutationDispatcher::MakeDictionaryEntryFromCMP( |
| 198 | const void *Arg1, const void *Arg2, |
| 199 | const void *Arg1Mutation, const void *Arg2Mutation, |
| 200 | size_t ArgSize, const uint8_t *Data, |
| 201 | size_t Size) { |
| 202 | ScopedDoingMyOwnMemOrStr scoped_doing_my_own_mem_os_str; |
| 203 | bool HandleFirst = Rand.RandBool(); |
| 204 | const void *ExistingBytes, *DesiredBytes; |
| 205 | Word W; |
| 206 | const uint8_t *End = Data + Size; |
| 207 | for (int Arg = 0; Arg < 2; Arg++) { |
| 208 | ExistingBytes = HandleFirst ? Arg1 : Arg2; |
| 209 | DesiredBytes = HandleFirst ? Arg2Mutation : Arg1Mutation; |
| 210 | HandleFirst = !HandleFirst; |
| 211 | W.Set(reinterpret_cast<const uint8_t*>(DesiredBytes), ArgSize); |
| 212 | const size_t kMaxNumPositions = 8; |
| 213 | size_t Positions[kMaxNumPositions]; |
| 214 | size_t NumPositions = 0; |
| 215 | for (const uint8_t *Cur = Data; |
| 216 | Cur < End && NumPositions < kMaxNumPositions; Cur++) { |
| 217 | Cur = |
| 218 | (const uint8_t *)SearchMemory(Cur, End - Cur, ExistingBytes, ArgSize); |
| 219 | if (!Cur) break; |
| 220 | Positions[NumPositions++] = Cur - Data; |
| 221 | } |
| 222 | if (!NumPositions) continue; |
| 223 | return DictionaryEntry(W, Positions[Rand(NumPositions)]); |
| 224 | } |
| 225 | DictionaryEntry DE(W); |
| 226 | return DE; |
| 227 | } |
| 228 | |
| 229 | |
| 230 | template <class T> |
| 231 | DictionaryEntry MutationDispatcher::MakeDictionaryEntryFromCMP( |
| 232 | T Arg1, T Arg2, const uint8_t *Data, size_t Size) { |
| 233 | if (Rand.RandBool()) Arg1 = Bswap(Arg1); |
| 234 | if (Rand.RandBool()) Arg2 = Bswap(Arg2); |
| 235 | T Arg1Mutation = Arg1 + Rand(-1, 1); |
| 236 | T Arg2Mutation = Arg2 + Rand(-1, 1); |
| 237 | return MakeDictionaryEntryFromCMP(&Arg1, &Arg2, &Arg1Mutation, &Arg2Mutation, |
| 238 | sizeof(Arg1), Data, Size); |
| 239 | } |
| 240 | |
| 241 | DictionaryEntry MutationDispatcher::MakeDictionaryEntryFromCMP( |
| 242 | const Word &Arg1, const Word &Arg2, const uint8_t *Data, size_t Size) { |
| 243 | return MakeDictionaryEntryFromCMP(Arg1.data(), Arg2.data(), Arg1.data(), |
| 244 | Arg2.data(), Arg1.size(), Data, Size); |
| 245 | } |
| 246 | |
| 247 | size_t MutationDispatcher::Mutate_AddWordFromTORC( |
| 248 | uint8_t *Data, size_t Size, size_t MaxSize) { |
| 249 | Word W; |
| 250 | DictionaryEntry DE; |
| 251 | switch (Rand(4)) { |
| 252 | case 0: { |
| 253 | auto X = TPC.TORC8.Get(Rand.Rand()); |
| 254 | DE = MakeDictionaryEntryFromCMP(X.A, X.B, Data, Size); |
| 255 | } break; |
| 256 | case 1: { |
| 257 | auto X = TPC.TORC4.Get(Rand.Rand()); |
| 258 | if ((X.A >> 16) == 0 && (X.B >> 16) == 0 && Rand.RandBool()) |
| 259 | DE = MakeDictionaryEntryFromCMP((uint16_t)X.A, (uint16_t)X.B, Data, Size); |
| 260 | else |
| 261 | DE = MakeDictionaryEntryFromCMP(X.A, X.B, Data, Size); |
| 262 | } break; |
| 263 | case 2: { |
| 264 | auto X = TPC.TORCW.Get(Rand.Rand()); |
| 265 | DE = MakeDictionaryEntryFromCMP(X.A, X.B, Data, Size); |
| 266 | } break; |
| 267 | case 3: if (Options.UseMemmem) { |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 268 | auto X = TPC.MMT.Get(Rand.Rand()); |
| 269 | DE = DictionaryEntry(X); |
| 270 | } break; |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 271 | default: |
| 272 | assert(0); |
| 273 | } |
| 274 | if (!DE.GetW().size()) return 0; |
| 275 | Size = ApplyDictionaryEntry(Data, Size, MaxSize, DE); |
| 276 | if (!Size) return 0; |
| 277 | DictionaryEntry &DERef = |
| 278 | CmpDictionaryEntriesDeque[CmpDictionaryEntriesDequeIdx++ % |
| 279 | kCmpDictionaryEntriesDequeSize]; |
| 280 | DERef = DE; |
| 281 | CurrentDictionaryEntrySequence.push_back(&DERef); |
| 282 | return Size; |
| 283 | } |
| 284 | |
| 285 | size_t MutationDispatcher::Mutate_AddWordFromPersistentAutoDictionary( |
| 286 | uint8_t *Data, size_t Size, size_t MaxSize) { |
| 287 | return AddWordFromDictionary(PersistentAutoDictionary, Data, Size, MaxSize); |
| 288 | } |
| 289 | |
| 290 | size_t MutationDispatcher::AddWordFromDictionary(Dictionary &D, uint8_t *Data, |
| 291 | size_t Size, size_t MaxSize) { |
| 292 | if (Size > MaxSize) return 0; |
| 293 | if (D.empty()) return 0; |
| 294 | DictionaryEntry &DE = D[Rand(D.size())]; |
| 295 | Size = ApplyDictionaryEntry(Data, Size, MaxSize, DE); |
| 296 | if (!Size) return 0; |
| 297 | DE.IncUseCount(); |
| 298 | CurrentDictionaryEntrySequence.push_back(&DE); |
| 299 | return Size; |
| 300 | } |
| 301 | |
| 302 | // Overwrites part of To[0,ToSize) with a part of From[0,FromSize). |
| 303 | // Returns ToSize. |
| 304 | size_t MutationDispatcher::CopyPartOf(const uint8_t *From, size_t FromSize, |
| 305 | uint8_t *To, size_t ToSize) { |
| 306 | // Copy From[FromBeg, FromBeg + CopySize) into To[ToBeg, ToBeg + CopySize). |
| 307 | size_t ToBeg = Rand(ToSize); |
| 308 | size_t CopySize = Rand(ToSize - ToBeg) + 1; |
| 309 | assert(ToBeg + CopySize <= ToSize); |
| 310 | CopySize = std::min(CopySize, FromSize); |
| 311 | size_t FromBeg = Rand(FromSize - CopySize + 1); |
| 312 | assert(FromBeg + CopySize <= FromSize); |
| 313 | memmove(To + ToBeg, From + FromBeg, CopySize); |
| 314 | return ToSize; |
| 315 | } |
| 316 | |
| 317 | // Inserts part of From[0,ToSize) into To. |
| 318 | // Returns new size of To on success or 0 on failure. |
| 319 | size_t MutationDispatcher::InsertPartOf(const uint8_t *From, size_t FromSize, |
| 320 | uint8_t *To, size_t ToSize, |
| 321 | size_t MaxToSize) { |
| 322 | if (ToSize >= MaxToSize) return 0; |
| 323 | size_t AvailableSpace = MaxToSize - ToSize; |
| 324 | size_t MaxCopySize = std::min(AvailableSpace, FromSize); |
| 325 | size_t CopySize = Rand(MaxCopySize) + 1; |
| 326 | size_t FromBeg = Rand(FromSize - CopySize + 1); |
| 327 | assert(FromBeg + CopySize <= FromSize); |
| 328 | size_t ToInsertPos = Rand(ToSize + 1); |
| 329 | assert(ToInsertPos + CopySize <= MaxToSize); |
| 330 | size_t TailSize = ToSize - ToInsertPos; |
| 331 | if (To == From) { |
| 332 | MutateInPlaceHere.resize(MaxToSize); |
| 333 | memcpy(MutateInPlaceHere.data(), From + FromBeg, CopySize); |
| 334 | memmove(To + ToInsertPos + CopySize, To + ToInsertPos, TailSize); |
| 335 | memmove(To + ToInsertPos, MutateInPlaceHere.data(), CopySize); |
| 336 | } else { |
| 337 | memmove(To + ToInsertPos + CopySize, To + ToInsertPos, TailSize); |
| 338 | memmove(To + ToInsertPos, From + FromBeg, CopySize); |
| 339 | } |
| 340 | return ToSize + CopySize; |
| 341 | } |
| 342 | |
| 343 | size_t MutationDispatcher::Mutate_CopyPart(uint8_t *Data, size_t Size, |
| 344 | size_t MaxSize) { |
| 345 | if (Size > MaxSize || Size == 0) return 0; |
delcypher | 1fda131 | 2018-04-24 06:31:09 +0000 | [diff] [blame] | 346 | // If Size == MaxSize, `InsertPartOf(...)` will |
| 347 | // fail so there's no point using it in this case. |
| 348 | if (Size == MaxSize || Rand.RandBool()) |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 349 | return CopyPartOf(Data, Size, Data, Size); |
| 350 | else |
| 351 | return InsertPartOf(Data, Size, Data, Size, MaxSize); |
| 352 | } |
| 353 | |
| 354 | size_t MutationDispatcher::Mutate_ChangeASCIIInteger(uint8_t *Data, size_t Size, |
| 355 | size_t MaxSize) { |
| 356 | if (Size > MaxSize) return 0; |
| 357 | size_t B = Rand(Size); |
| 358 | while (B < Size && !isdigit(Data[B])) B++; |
| 359 | if (B == Size) return 0; |
| 360 | size_t E = B; |
| 361 | while (E < Size && isdigit(Data[E])) E++; |
| 362 | assert(B < E); |
| 363 | // now we have digits in [B, E). |
| 364 | // strtol and friends don't accept non-zero-teminated data, parse it manually. |
| 365 | uint64_t Val = Data[B] - '0'; |
| 366 | for (size_t i = B + 1; i < E; i++) |
| 367 | Val = Val * 10 + Data[i] - '0'; |
| 368 | |
| 369 | // Mutate the integer value. |
| 370 | switch(Rand(5)) { |
| 371 | case 0: Val++; break; |
| 372 | case 1: Val--; break; |
| 373 | case 2: Val /= 2; break; |
| 374 | case 3: Val *= 2; break; |
| 375 | case 4: Val = Rand(Val * Val); break; |
| 376 | default: assert(0); |
| 377 | } |
| 378 | // Just replace the bytes with the new ones, don't bother moving bytes. |
| 379 | for (size_t i = B; i < E; i++) { |
| 380 | size_t Idx = E + B - i - 1; |
| 381 | assert(Idx >= B && Idx < E); |
| 382 | Data[Idx] = (Val % 10) + '0'; |
| 383 | Val /= 10; |
| 384 | } |
| 385 | return Size; |
| 386 | } |
| 387 | |
| 388 | template<class T> |
| 389 | size_t ChangeBinaryInteger(uint8_t *Data, size_t Size, Random &Rand) { |
| 390 | if (Size < sizeof(T)) return 0; |
| 391 | size_t Off = Rand(Size - sizeof(T) + 1); |
| 392 | assert(Off + sizeof(T) <= Size); |
| 393 | T Val; |
| 394 | if (Off < 64 && !Rand(4)) { |
| 395 | Val = Size; |
| 396 | if (Rand.RandBool()) |
| 397 | Val = Bswap(Val); |
| 398 | } else { |
| 399 | memcpy(&Val, Data + Off, sizeof(Val)); |
| 400 | T Add = Rand(21); |
| 401 | Add -= 10; |
| 402 | if (Rand.RandBool()) |
| 403 | Val = Bswap(T(Bswap(Val) + Add)); // Add assuming different endiannes. |
| 404 | else |
| 405 | Val = Val + Add; // Add assuming current endiannes. |
| 406 | if (Add == 0 || Rand.RandBool()) // Maybe negate. |
| 407 | Val = -Val; |
| 408 | } |
| 409 | memcpy(Data + Off, &Val, sizeof(Val)); |
| 410 | return Size; |
| 411 | } |
| 412 | |
| 413 | size_t MutationDispatcher::Mutate_ChangeBinaryInteger(uint8_t *Data, |
| 414 | size_t Size, |
| 415 | size_t MaxSize) { |
| 416 | if (Size > MaxSize) return 0; |
| 417 | switch (Rand(4)) { |
| 418 | case 3: return ChangeBinaryInteger<uint64_t>(Data, Size, Rand); |
| 419 | case 2: return ChangeBinaryInteger<uint32_t>(Data, Size, Rand); |
| 420 | case 1: return ChangeBinaryInteger<uint16_t>(Data, Size, Rand); |
| 421 | case 0: return ChangeBinaryInteger<uint8_t>(Data, Size, Rand); |
| 422 | default: assert(0); |
| 423 | } |
| 424 | return 0; |
| 425 | } |
| 426 | |
| 427 | size_t MutationDispatcher::Mutate_CrossOver(uint8_t *Data, size_t Size, |
| 428 | size_t MaxSize) { |
| 429 | if (Size > MaxSize) return 0; |
| 430 | if (!Corpus || Corpus->size() < 2 || Size == 0) return 0; |
| 431 | size_t Idx = Rand(Corpus->size()); |
| 432 | const Unit &O = (*Corpus)[Idx]; |
| 433 | if (O.empty()) return 0; |
| 434 | MutateInPlaceHere.resize(MaxSize); |
| 435 | auto &U = MutateInPlaceHere; |
| 436 | size_t NewSize = 0; |
| 437 | switch(Rand(3)) { |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 438 | case 0: |
| 439 | NewSize = CrossOver(Data, Size, O.data(), O.size(), U.data(), U.size()); |
| 440 | break; |
| 441 | case 1: |
| 442 | NewSize = InsertPartOf(O.data(), O.size(), U.data(), U.size(), MaxSize); |
| 443 | if (!NewSize) |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 444 | NewSize = CopyPartOf(O.data(), O.size(), U.data(), U.size()); |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 445 | break; |
| 446 | case 2: |
| 447 | NewSize = CopyPartOf(O.data(), O.size(), U.data(), U.size()); |
| 448 | break; |
| 449 | default: assert(0); |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 450 | } |
| 451 | assert(NewSize > 0 && "CrossOver returned empty unit"); |
| 452 | assert(NewSize <= MaxSize && "CrossOver returned overisized unit"); |
| 453 | memcpy(Data, U.data(), NewSize); |
| 454 | return NewSize; |
| 455 | } |
| 456 | |
| 457 | void MutationDispatcher::StartMutationSequence() { |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 458 | CurrentMutatorIdxSequence.clear(); |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 459 | CurrentDictionaryEntrySequence.clear(); |
| 460 | } |
| 461 | |
| 462 | // Copy successful dictionary entries to PersistentAutoDictionary. |
| 463 | void MutationDispatcher::RecordSuccessfulMutationSequence() { |
| 464 | for (auto DE : CurrentDictionaryEntrySequence) { |
| 465 | // PersistentAutoDictionary.AddWithSuccessCountOne(DE); |
| 466 | DE->IncSuccessCount(); |
| 467 | assert(DE->GetW().size()); |
| 468 | // Linear search is fine here as this happens seldom. |
| 469 | if (!PersistentAutoDictionary.ContainsWord(DE->GetW())) |
| 470 | PersistentAutoDictionary.push_back({DE->GetW(), 1}); |
| 471 | } |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 472 | RecordUsefulMutations(); |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 473 | } |
| 474 | |
| 475 | void MutationDispatcher::PrintRecommendedDictionary() { |
george.karpenkov | fbfa45c | 2017-08-27 23:20:09 +0000 | [diff] [blame] | 476 | Vector<DictionaryEntry> V; |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 477 | for (auto &DE : PersistentAutoDictionary) |
| 478 | if (!ManualDictionary.ContainsWord(DE.GetW())) |
| 479 | V.push_back(DE); |
| 480 | if (V.empty()) return; |
| 481 | Printf("###### Recommended dictionary. ######\n"); |
| 482 | for (auto &DE: V) { |
| 483 | assert(DE.GetW().size()); |
| 484 | Printf("\""); |
| 485 | PrintASCII(DE.GetW(), "\""); |
| 486 | Printf(" # Uses: %zd\n", DE.GetUseCount()); |
| 487 | } |
| 488 | Printf("###### End of recommended dictionary. ######\n"); |
| 489 | } |
| 490 | |
| 491 | void MutationDispatcher::PrintMutationSequence() { |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 492 | Printf("MS: %zd ", CurrentMutatorIdxSequence.size()); |
| 493 | for (auto M : CurrentMutatorIdxSequence) |
| 494 | Printf("%s-", Mutators[M].Name); |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 495 | if (!CurrentDictionaryEntrySequence.empty()) { |
| 496 | Printf(" DE: "); |
| 497 | for (auto DE : CurrentDictionaryEntrySequence) { |
| 498 | Printf("\""); |
| 499 | PrintASCII(DE->GetW(), "\"-"); |
| 500 | } |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | size_t MutationDispatcher::Mutate(uint8_t *Data, size_t Size, size_t MaxSize) { |
| 505 | return MutateImpl(Data, Size, MaxSize, Mutators); |
| 506 | } |
| 507 | |
| 508 | size_t MutationDispatcher::DefaultMutate(uint8_t *Data, size_t Size, |
| 509 | size_t MaxSize) { |
| 510 | return MutateImpl(Data, Size, MaxSize, DefaultMutators); |
| 511 | } |
| 512 | |
| 513 | // Mutates Data in place, returns new size. |
| 514 | size_t MutationDispatcher::MutateImpl(uint8_t *Data, size_t Size, |
| 515 | size_t MaxSize, |
george.karpenkov | fbfa45c | 2017-08-27 23:20:09 +0000 | [diff] [blame] | 516 | Vector<Mutator> &Mutators) { |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 517 | assert(MaxSize > 0); |
| 518 | // Some mutations may fail (e.g. can't insert more bytes if Size == MaxSize), |
| 519 | // in which case they will return 0. |
| 520 | // Try several times before returning un-mutated data. |
| 521 | for (int Iter = 0; Iter < 100; Iter++) { |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 522 | size_t MutatorIdx = Rand(Mutators.size()); |
| 523 | auto M = Mutators[MutatorIdx]; |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 524 | size_t NewSize = (this->*(M.Fn))(Data, Size, MaxSize); |
| 525 | if (NewSize && NewSize <= MaxSize) { |
| 526 | if (Options.OnlyASCII) |
| 527 | ToASCII(Data, NewSize); |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 528 | CurrentMutatorIdxSequence.push_back(MutatorIdx); |
| 529 | TotalMutations[MutatorIdx]++; |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 530 | return NewSize; |
| 531 | } |
| 532 | } |
| 533 | *Data = ' '; |
| 534 | return 1; // Fallback, should not happen frequently. |
| 535 | } |
| 536 | |
| 537 | void MutationDispatcher::AddWordToManualDictionary(const Word &W) { |
| 538 | ManualDictionary.push_back( |
| 539 | {W, std::numeric_limits<size_t>::max()}); |
| 540 | } |
| 541 | |
morehouse | cfe1f44 | 2018-07-09 20:17:52 +0000 | [diff] [blame^] | 542 | void MutationDispatcher::RecordUsefulMutations() { |
| 543 | for (const size_t M : CurrentMutatorIdxSequence) |
| 544 | UsefulMutations[M]++; |
| 545 | } |
| 546 | |
| 547 | void MutationDispatcher::PrintMutationStats() { |
| 548 | Printf("\nstat::mutation_usefulness: "); |
| 549 | for (size_t i = 0; i < Mutators.size(); i++) { |
| 550 | double UsefulPercentage = |
| 551 | TotalMutations[i] |
| 552 | ? (100.0 * UsefulMutations[i]) / TotalMutations[i] |
| 553 | : 0; |
| 554 | Printf("%.3f", UsefulPercentage); |
| 555 | if (i < Mutators.size() - 1) |
| 556 | Printf(","); |
| 557 | } |
| 558 | Printf("\n"); |
| 559 | } |
| 560 | |
george.karpenkov | 29efa6d | 2017-08-21 23:25:50 +0000 | [diff] [blame] | 561 | } // namespace fuzzer |