1 | //===- FuzzerDriver.cpp - FuzzerDriver function and flags -----------------===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | // FuzzerDriver and flag parsing. |
9 | //===----------------------------------------------------------------------===// |
10 | |
11 | #include "FuzzerCommand.h" |
12 | #include "FuzzerCorpus.h" |
13 | #include "FuzzerFork.h" |
14 | #include "FuzzerIO.h" |
15 | #include "FuzzerInterface.h" |
16 | #include "FuzzerInternal.h" |
17 | #include "FuzzerMerge.h" |
18 | #include "FuzzerMutate.h" |
19 | #include "FuzzerPlatform.h" |
20 | #include "FuzzerRandom.h" |
21 | #include "FuzzerTracePC.h" |
22 | #include <algorithm> |
23 | #include <atomic> |
24 | #include <chrono> |
25 | #include <cstdlib> |
26 | #include <cstring> |
27 | #include <mutex> |
28 | #include <string> |
29 | #include <thread> |
30 | #include <fstream> |
31 | |
32 | // This function should be present in the libFuzzer so that the client |
33 | // binary can test for its existence. |
34 | #if LIBFUZZER_MSVC |
35 | extern "C" void __libfuzzer_is_present() {} |
36 | #if defined(_M_IX86) || defined(__i386__) |
37 | #pragma comment(linker, "/include:___libfuzzer_is_present") |
38 | #else |
39 | #pragma comment(linker, "/include:__libfuzzer_is_present") |
40 | #endif |
41 | #else |
42 | extern "C" __attribute__((used)) void __libfuzzer_is_present() {} |
43 | #endif // LIBFUZZER_MSVC |
44 | |
45 | namespace fuzzer { |
46 | |
47 | // Program arguments. |
48 | struct FlagDescription { |
49 | const char *Name; |
50 | const char *Description; |
51 | int Default; |
52 | int *IntFlag; |
53 | const char **StrFlag; |
54 | unsigned int *UIntFlag; |
55 | }; |
56 | |
57 | struct { |
58 | #define FUZZER_DEPRECATED_FLAG(Name) |
59 | #define FUZZER_FLAG_INT(Name, Default, Description) int Name; |
60 | #define FUZZER_FLAG_UNSIGNED(Name, Default, Description) unsigned int Name; |
61 | #define FUZZER_FLAG_STRING(Name, Description) const char *Name; |
62 | #include "FuzzerFlags.def" |
63 | #undef FUZZER_DEPRECATED_FLAG |
64 | #undef FUZZER_FLAG_INT |
65 | #undef FUZZER_FLAG_UNSIGNED |
66 | #undef FUZZER_FLAG_STRING |
67 | } Flags; |
68 | |
69 | static const FlagDescription FlagDescriptions [] { |
70 | #define FUZZER_DEPRECATED_FLAG(Name) \ |
71 | {#Name, "Deprecated; don't use", 0, nullptr, nullptr, nullptr}, |
72 | #define FUZZER_FLAG_INT(Name, Default, Description) \ |
73 | {#Name, Description, Default, &Flags.Name, nullptr, nullptr}, |
74 | #define FUZZER_FLAG_UNSIGNED(Name, Default, Description) \ |
75 | {#Name, Description, static_cast<int>(Default), \ |
76 | nullptr, nullptr, &Flags.Name}, |
77 | #define FUZZER_FLAG_STRING(Name, Description) \ |
78 | {#Name, Description, 0, nullptr, &Flags.Name, nullptr}, |
79 | #include "FuzzerFlags.def" |
80 | #undef FUZZER_DEPRECATED_FLAG |
81 | #undef FUZZER_FLAG_INT |
82 | #undef FUZZER_FLAG_UNSIGNED |
83 | #undef FUZZER_FLAG_STRING |
84 | }; |
85 | |
86 | static const size_t kNumFlags = |
87 | sizeof(FlagDescriptions) / sizeof(FlagDescriptions[0]); |
88 | |
89 | static std::vector<std::string> *Inputs; |
90 | static std::string *ProgName; |
91 | |
92 | static void PrintHelp() { |
93 | Printf("Usage:\n" ); |
94 | auto Prog = ProgName->c_str(); |
95 | Printf("\nTo run fuzzing pass 0 or more directories.\n" ); |
96 | Printf("%s [-flag1=val1 [-flag2=val2 ...] ] [dir1 [dir2 ...] ]\n" , Prog); |
97 | |
98 | Printf("\nTo run individual tests without fuzzing pass 1 or more files:\n" ); |
99 | Printf("%s [-flag1=val1 [-flag2=val2 ...] ] file1 [file2 ...]\n" , Prog); |
100 | |
101 | Printf("\nFlags: (strictly in form -flag=value)\n" ); |
102 | size_t MaxFlagLen = 0; |
103 | for (size_t F = 0; F < kNumFlags; F++) |
104 | MaxFlagLen = std::max(strlen(FlagDescriptions[F].Name), MaxFlagLen); |
105 | |
106 | for (size_t F = 0; F < kNumFlags; F++) { |
107 | const auto &D = FlagDescriptions[F]; |
108 | if (strstr(D.Description, "internal flag" ) == D.Description) continue; |
109 | Printf(" %s" , D.Name); |
110 | for (size_t i = 0, n = MaxFlagLen - strlen(D.Name); i < n; i++) |
111 | Printf(" " ); |
112 | Printf("\t" ); |
113 | Printf("%d\t%s\n" , D.Default, D.Description); |
114 | } |
115 | Printf("\nFlags starting with '--' will be ignored and " |
116 | "will be passed verbatim to subprocesses.\n" ); |
117 | } |
118 | |
119 | static const char *FlagValue(const char *Param, const char *Name) { |
120 | size_t Len = strlen(Name); |
121 | if (Param[0] == '-' && strstr(Param + 1, Name) == Param + 1 && |
122 | Param[Len + 1] == '=') |
123 | return &Param[Len + 2]; |
124 | return nullptr; |
125 | } |
126 | |
127 | // Avoid calling stol as it triggers a bug in clang/glibc build. |
128 | static long MyStol(const char *Str) { |
129 | long Res = 0; |
130 | long Sign = 1; |
131 | if (*Str == '-') { |
132 | Str++; |
133 | Sign = -1; |
134 | } |
135 | for (size_t i = 0; Str[i]; i++) { |
136 | char Ch = Str[i]; |
137 | if (Ch < '0' || Ch > '9') |
138 | return Res; |
139 | Res = Res * 10 + (Ch - '0'); |
140 | } |
141 | return Res * Sign; |
142 | } |
143 | |
144 | static bool ParseOneFlag(const char *Param) { |
145 | if (Param[0] != '-') return false; |
146 | if (Param[1] == '-') { |
147 | static bool PrintedWarning = false; |
148 | if (!PrintedWarning) { |
149 | PrintedWarning = true; |
150 | Printf("INFO: libFuzzer ignores flags that start with '--'\n" ); |
151 | } |
152 | for (size_t F = 0; F < kNumFlags; F++) |
153 | if (FlagValue(Param + 1, FlagDescriptions[F].Name)) |
154 | Printf("WARNING: did you mean '%s' (single dash)?\n" , Param + 1); |
155 | return true; |
156 | } |
157 | for (size_t F = 0; F < kNumFlags; F++) { |
158 | const char *Name = FlagDescriptions[F].Name; |
159 | const char *Str = FlagValue(Param, Name); |
160 | if (Str) { |
161 | if (FlagDescriptions[F].IntFlag) { |
162 | auto Val = MyStol(Str); |
163 | *FlagDescriptions[F].IntFlag = static_cast<int>(Val); |
164 | if (Flags.verbosity >= 2) |
165 | Printf("Flag: %s %d\n" , Name, Val); |
166 | return true; |
167 | } else if (FlagDescriptions[F].UIntFlag) { |
168 | auto Val = std::stoul(Str); |
169 | *FlagDescriptions[F].UIntFlag = static_cast<unsigned int>(Val); |
170 | if (Flags.verbosity >= 2) |
171 | Printf("Flag: %s %u\n" , Name, Val); |
172 | return true; |
173 | } else if (FlagDescriptions[F].StrFlag) { |
174 | *FlagDescriptions[F].StrFlag = Str; |
175 | if (Flags.verbosity >= 2) |
176 | Printf("Flag: %s %s\n" , Name, Str); |
177 | return true; |
178 | } else { // Deprecated flag. |
179 | Printf("Flag: %s: deprecated, don't use\n" , Name); |
180 | return true; |
181 | } |
182 | } |
183 | } |
184 | Printf("\n\nWARNING: unrecognized flag '%s'; " |
185 | "use -help=1 to list all flags\n\n" , Param); |
186 | return true; |
187 | } |
188 | |
189 | // We don't use any library to minimize dependencies. |
190 | static void ParseFlags(const std::vector<std::string> &Args, |
191 | const ExternalFunctions *EF) { |
192 | for (size_t F = 0; F < kNumFlags; F++) { |
193 | if (FlagDescriptions[F].IntFlag) |
194 | *FlagDescriptions[F].IntFlag = FlagDescriptions[F].Default; |
195 | if (FlagDescriptions[F].UIntFlag) |
196 | *FlagDescriptions[F].UIntFlag = |
197 | static_cast<unsigned int>(FlagDescriptions[F].Default); |
198 | if (FlagDescriptions[F].StrFlag) |
199 | *FlagDescriptions[F].StrFlag = nullptr; |
200 | } |
201 | |
202 | // Disable len_control by default, if LLVMFuzzerCustomMutator is used. |
203 | if (EF->LLVMFuzzerCustomMutator) { |
204 | Flags.len_control = 0; |
205 | Printf("INFO: found LLVMFuzzerCustomMutator (%p). " |
206 | "Disabling -len_control by default.\n" , EF->LLVMFuzzerCustomMutator); |
207 | } |
208 | |
209 | Inputs = new std::vector<std::string>; |
210 | for (size_t A = 1; A < Args.size(); A++) { |
211 | if (ParseOneFlag(Args[A].c_str())) { |
212 | if (Flags.ignore_remaining_args) |
213 | break; |
214 | continue; |
215 | } |
216 | Inputs->push_back(Args[A]); |
217 | } |
218 | } |
219 | |
220 | static std::mutex Mu; |
221 | |
222 | static void PulseThread() { |
223 | while (true) { |
224 | SleepSeconds(600); |
225 | std::lock_guard<std::mutex> Lock(Mu); |
226 | Printf("pulse...\n" ); |
227 | } |
228 | } |
229 | |
230 | static void WorkerThread(const Command &BaseCmd, std::atomic<unsigned> *Counter, |
231 | unsigned NumJobs, std::atomic<bool> *HasErrors) { |
232 | ScopedDisableMsanInterceptorChecks S; |
233 | while (true) { |
234 | unsigned C = (*Counter)++; |
235 | if (C >= NumJobs) break; |
236 | std::string Log = "fuzz-" + std::to_string(C) + ".log" ; |
237 | Command Cmd(BaseCmd); |
238 | Cmd.setOutputFile(Log); |
239 | Cmd.combineOutAndErr(); |
240 | if (Flags.verbosity) { |
241 | std::string CommandLine = Cmd.toString(); |
242 | Printf("%s\n" , CommandLine.c_str()); |
243 | } |
244 | int ExitCode = ExecuteCommand(Cmd); |
245 | if (ExitCode != 0) |
246 | *HasErrors = true; |
247 | std::lock_guard<std::mutex> Lock(Mu); |
248 | Printf("================== Job %u exited with exit code %d ============\n" , |
249 | C, ExitCode); |
250 | fuzzer::CopyFileToErr(Log); |
251 | } |
252 | } |
253 | |
254 | static void ValidateDirectoryExists(const std::string &Path, |
255 | bool CreateDirectory) { |
256 | if (Path.empty()) { |
257 | Printf("ERROR: Provided directory path is an empty string\n" ); |
258 | exit(1); |
259 | } |
260 | |
261 | if (IsDirectory(Path)) |
262 | return; |
263 | |
264 | if (CreateDirectory) { |
265 | if (!MkDirRecursive(Path)) { |
266 | Printf("ERROR: Failed to create directory \"%s\"\n" , Path.c_str()); |
267 | exit(1); |
268 | } |
269 | return; |
270 | } |
271 | |
272 | Printf("ERROR: The required directory \"%s\" does not exist\n" , Path.c_str()); |
273 | exit(1); |
274 | } |
275 | |
276 | std::string CloneArgsWithoutX(const std::vector<std::string> &Args, |
277 | const char *X1, const char *X2) { |
278 | std::string Cmd; |
279 | for (auto &S : Args) { |
280 | if (FlagValue(S.c_str(), X1) || FlagValue(S.c_str(), X2)) |
281 | continue; |
282 | Cmd += S + " " ; |
283 | } |
284 | return Cmd; |
285 | } |
286 | |
287 | static int RunInMultipleProcesses(const std::vector<std::string> &Args, |
288 | unsigned NumWorkers, unsigned NumJobs) { |
289 | std::atomic<unsigned> Counter(0); |
290 | std::atomic<bool> HasErrors(false); |
291 | Command Cmd(Args); |
292 | Cmd.removeFlag("jobs" ); |
293 | Cmd.removeFlag("workers" ); |
294 | std::vector<std::thread> V; |
295 | std::thread Pulse(PulseThread); |
296 | Pulse.detach(); |
297 | V.resize(NumWorkers); |
298 | for (unsigned i = 0; i < NumWorkers; i++) { |
299 | V[i] = std::thread(WorkerThread, std::ref(Cmd), &Counter, NumJobs, |
300 | &HasErrors); |
301 | SetThreadName(V[i], "FuzzerWorker" ); |
302 | } |
303 | for (auto &T : V) |
304 | T.join(); |
305 | return HasErrors ? 1 : 0; |
306 | } |
307 | |
308 | static void RssThread(Fuzzer *F, size_t RssLimitMb) { |
309 | while (true) { |
310 | SleepSeconds(1); |
311 | size_t Peak = GetPeakRSSMb(); |
312 | if (Peak > RssLimitMb) |
313 | F->RssLimitCallback(); |
314 | } |
315 | } |
316 | |
317 | static void StartRssThread(Fuzzer *F, size_t RssLimitMb) { |
318 | if (!RssLimitMb) |
319 | return; |
320 | std::thread T(RssThread, F, RssLimitMb); |
321 | T.detach(); |
322 | } |
323 | |
324 | int RunOneTest(Fuzzer *F, const char *InputFilePath, size_t MaxLen) { |
325 | Unit U = FileToVector(InputFilePath); |
326 | if (MaxLen && MaxLen < U.size()) |
327 | U.resize(MaxLen); |
328 | F->ExecuteCallback(U.data(), U.size()); |
329 | if (Flags.print_full_coverage) { |
330 | // Leak detection is not needed when collecting full coverage data. |
331 | F->TPCUpdateObservedPCs(); |
332 | } else { |
333 | F->TryDetectingAMemoryLeak(U.data(), U.size(), true); |
334 | } |
335 | return 0; |
336 | } |
337 | |
338 | static bool AllInputsAreFiles() { |
339 | if (Inputs->empty()) return false; |
340 | for (auto &Path : *Inputs) |
341 | if (!IsFile(Path)) |
342 | return false; |
343 | return true; |
344 | } |
345 | |
346 | static std::string GetDedupTokenFromCmdOutput(const std::string &S) { |
347 | auto Beg = S.find("DEDUP_TOKEN:" ); |
348 | if (Beg == std::string::npos) |
349 | return "" ; |
350 | auto End = S.find('\n', Beg); |
351 | if (End == std::string::npos) |
352 | return "" ; |
353 | return S.substr(Beg, End - Beg); |
354 | } |
355 | |
356 | int CleanseCrashInput(const std::vector<std::string> &Args, |
357 | const FuzzingOptions &Options) { |
358 | if (Inputs->size() != 1 || !Flags.exact_artifact_path) { |
359 | Printf("ERROR: -cleanse_crash should be given one input file and" |
360 | " -exact_artifact_path\n" ); |
361 | exit(1); |
362 | } |
363 | std::string InputFilePath = Inputs->at(0); |
364 | std::string OutputFilePath = Flags.exact_artifact_path; |
365 | Command Cmd(Args); |
366 | Cmd.removeFlag("cleanse_crash" ); |
367 | |
368 | assert(Cmd.hasArgument(InputFilePath)); |
369 | Cmd.removeArgument(InputFilePath); |
370 | |
371 | auto TmpFilePath = TempPath("CleanseCrashInput" , ".repro" ); |
372 | Cmd.addArgument(TmpFilePath); |
373 | Cmd.setOutputFile(getDevNull()); |
374 | Cmd.combineOutAndErr(); |
375 | |
376 | std::string CurrentFilePath = InputFilePath; |
377 | auto U = FileToVector(CurrentFilePath); |
378 | size_t Size = U.size(); |
379 | |
380 | const std::vector<uint8_t> ReplacementBytes = {' ', 0xff}; |
381 | for (int NumAttempts = 0; NumAttempts < 5; NumAttempts++) { |
382 | bool Changed = false; |
383 | for (size_t Idx = 0; Idx < Size; Idx++) { |
384 | Printf("CLEANSE[%d]: Trying to replace byte %zd of %zd\n" , NumAttempts, |
385 | Idx, Size); |
386 | uint8_t OriginalByte = U[Idx]; |
387 | if (ReplacementBytes.end() != std::find(ReplacementBytes.begin(), |
388 | ReplacementBytes.end(), |
389 | OriginalByte)) |
390 | continue; |
391 | for (auto NewByte : ReplacementBytes) { |
392 | U[Idx] = NewByte; |
393 | WriteToFile(U, TmpFilePath); |
394 | auto ExitCode = ExecuteCommand(Cmd); |
395 | RemoveFile(TmpFilePath); |
396 | if (!ExitCode) { |
397 | U[Idx] = OriginalByte; |
398 | } else { |
399 | Changed = true; |
400 | Printf("CLEANSE: Replaced byte %zd with 0x%x\n" , Idx, NewByte); |
401 | WriteToFile(U, OutputFilePath); |
402 | break; |
403 | } |
404 | } |
405 | } |
406 | if (!Changed) break; |
407 | } |
408 | return 0; |
409 | } |
410 | |
411 | int MinimizeCrashInput(const std::vector<std::string> &Args, |
412 | const FuzzingOptions &Options) { |
413 | if (Inputs->size() != 1) { |
414 | Printf("ERROR: -minimize_crash should be given one input file\n" ); |
415 | exit(1); |
416 | } |
417 | std::string InputFilePath = Inputs->at(0); |
418 | Command BaseCmd(Args); |
419 | BaseCmd.removeFlag("minimize_crash" ); |
420 | BaseCmd.removeFlag("exact_artifact_path" ); |
421 | assert(BaseCmd.hasArgument(InputFilePath)); |
422 | BaseCmd.removeArgument(InputFilePath); |
423 | if (Flags.runs <= 0 && Flags.max_total_time == 0) { |
424 | Printf("INFO: you need to specify -runs=N or " |
425 | "-max_total_time=N with -minimize_crash=1\n" |
426 | "INFO: defaulting to -max_total_time=600\n" ); |
427 | BaseCmd.addFlag("max_total_time" , "600" ); |
428 | } |
429 | |
430 | BaseCmd.combineOutAndErr(); |
431 | |
432 | std::string CurrentFilePath = InputFilePath; |
433 | while (true) { |
434 | Unit U = FileToVector(CurrentFilePath); |
435 | Printf("CRASH_MIN: minimizing crash input: '%s' (%zd bytes)\n" , |
436 | CurrentFilePath.c_str(), U.size()); |
437 | |
438 | Command Cmd(BaseCmd); |
439 | Cmd.addArgument(CurrentFilePath); |
440 | |
441 | Printf("CRASH_MIN: executing: %s\n" , Cmd.toString().c_str()); |
442 | std::string CmdOutput; |
443 | bool Success = ExecuteCommand(Cmd, &CmdOutput); |
444 | if (Success) { |
445 | Printf("ERROR: the input %s did not crash\n" , CurrentFilePath.c_str()); |
446 | exit(1); |
447 | } |
448 | Printf("CRASH_MIN: '%s' (%zd bytes) caused a crash. Will try to minimize " |
449 | "it further\n" , |
450 | CurrentFilePath.c_str(), U.size()); |
451 | auto DedupToken1 = GetDedupTokenFromCmdOutput(CmdOutput); |
452 | if (!DedupToken1.empty()) |
453 | Printf("CRASH_MIN: DedupToken1: %s\n" , DedupToken1.c_str()); |
454 | |
455 | std::string ArtifactPath = |
456 | Flags.exact_artifact_path |
457 | ? Flags.exact_artifact_path |
458 | : Options.ArtifactPrefix + "minimized-from-" + Hash(U); |
459 | Cmd.addFlag("minimize_crash_internal_step" , "1" ); |
460 | Cmd.addFlag("exact_artifact_path" , ArtifactPath); |
461 | Printf("CRASH_MIN: executing: %s\n" , Cmd.toString().c_str()); |
462 | CmdOutput.clear(); |
463 | Success = ExecuteCommand(Cmd, &CmdOutput); |
464 | Printf("%s" , CmdOutput.c_str()); |
465 | if (Success) { |
466 | if (Flags.exact_artifact_path) { |
467 | CurrentFilePath = Flags.exact_artifact_path; |
468 | WriteToFile(U, CurrentFilePath); |
469 | } |
470 | Printf("CRASH_MIN: failed to minimize beyond %s (%zu bytes), exiting\n" , |
471 | CurrentFilePath.c_str(), U.size()); |
472 | break; |
473 | } |
474 | auto DedupToken2 = GetDedupTokenFromCmdOutput(CmdOutput); |
475 | if (!DedupToken2.empty()) |
476 | Printf("CRASH_MIN: DedupToken2: %s\n" , DedupToken2.c_str()); |
477 | |
478 | if (DedupToken1 != DedupToken2) { |
479 | if (Flags.exact_artifact_path) { |
480 | CurrentFilePath = Flags.exact_artifact_path; |
481 | WriteToFile(U, CurrentFilePath); |
482 | } |
483 | Printf("CRASH_MIN: mismatch in dedup tokens" |
484 | " (looks like a different bug). Won't minimize further\n" ); |
485 | break; |
486 | } |
487 | |
488 | CurrentFilePath = ArtifactPath; |
489 | Printf("*********************************\n" ); |
490 | } |
491 | return 0; |
492 | } |
493 | |
494 | int MinimizeCrashInputInternalStep(Fuzzer *F, InputCorpus *Corpus) { |
495 | assert(Inputs->size() == 1); |
496 | std::string InputFilePath = Inputs->at(0); |
497 | Unit U = FileToVector(InputFilePath); |
498 | Printf("INFO: Starting MinimizeCrashInputInternalStep: %zd\n" , U.size()); |
499 | if (U.size() < 2) { |
500 | Printf("INFO: The input is small enough, exiting\n" ); |
501 | exit(0); |
502 | } |
503 | F->SetMaxInputLen(U.size()); |
504 | F->SetMaxMutationLen(U.size() - 1); |
505 | F->MinimizeCrashLoop(U); |
506 | Printf("INFO: Done MinimizeCrashInputInternalStep, no crashes found\n" ); |
507 | exit(0); |
508 | } |
509 | |
510 | void Merge(Fuzzer *F, FuzzingOptions &Options, |
511 | const std::vector<std::string> &Args, |
512 | const std::vector<std::string> &Corpora, const char *CFPathOrNull) { |
513 | if (Corpora.size() < 2) { |
514 | Printf("INFO: Merge requires two or more corpus dirs\n" ); |
515 | exit(0); |
516 | } |
517 | |
518 | std::vector<SizedFile> OldCorpus, NewCorpus; |
519 | GetSizedFilesFromDir(Corpora[0], &OldCorpus); |
520 | for (size_t i = 1; i < Corpora.size(); i++) |
521 | GetSizedFilesFromDir(Corpora[i], &NewCorpus); |
522 | std::sort(OldCorpus.begin(), OldCorpus.end()); |
523 | std::sort(NewCorpus.begin(), NewCorpus.end()); |
524 | |
525 | std::string CFPath = CFPathOrNull ? CFPathOrNull : TempPath("Merge" , ".txt" ); |
526 | std::vector<std::string> NewFiles; |
527 | std::set<uint32_t> NewFeatures, NewCov; |
528 | CrashResistantMerge(Args, OldCorpus, NewCorpus, &NewFiles, {}, &NewFeatures, |
529 | {}, &NewCov, CFPath, true, Flags.set_cover_merge); |
530 | for (auto &Path : NewFiles) |
531 | F->WriteToOutputCorpus(FileToVector(Path, Options.MaxLen)); |
532 | // We are done, delete the control file if it was a temporary one. |
533 | if (!Flags.merge_control_file) |
534 | RemoveFile(CFPath); |
535 | |
536 | exit(0); |
537 | } |
538 | |
539 | int AnalyzeDictionary(Fuzzer *F, const std::vector<Unit> &Dict, |
540 | UnitVector &Corpus) { |
541 | Printf("Started dictionary minimization (up to %zu tests)\n" , |
542 | Dict.size() * Corpus.size() * 2); |
543 | |
544 | // Scores and usage count for each dictionary unit. |
545 | std::vector<int> Scores(Dict.size()); |
546 | std::vector<int> Usages(Dict.size()); |
547 | |
548 | std::vector<size_t> InitialFeatures; |
549 | std::vector<size_t> ModifiedFeatures; |
550 | for (auto &C : Corpus) { |
551 | // Get coverage for the testcase without modifications. |
552 | F->ExecuteCallback(C.data(), C.size()); |
553 | InitialFeatures.clear(); |
554 | TPC.CollectFeatures([&](size_t Feature) { |
555 | InitialFeatures.push_back(Feature); |
556 | }); |
557 | |
558 | for (size_t i = 0; i < Dict.size(); ++i) { |
559 | std::vector<uint8_t> Data = C; |
560 | auto StartPos = std::search(Data.begin(), Data.end(), |
561 | Dict[i].begin(), Dict[i].end()); |
562 | // Skip dictionary unit, if the testcase does not contain it. |
563 | if (StartPos == Data.end()) |
564 | continue; |
565 | |
566 | ++Usages[i]; |
567 | while (StartPos != Data.end()) { |
568 | // Replace all occurrences of dictionary unit in the testcase. |
569 | auto EndPos = StartPos + Dict[i].size(); |
570 | for (auto It = StartPos; It != EndPos; ++It) |
571 | *It ^= 0xFF; |
572 | |
573 | StartPos = std::search(EndPos, Data.end(), |
574 | Dict[i].begin(), Dict[i].end()); |
575 | } |
576 | |
577 | // Get coverage for testcase with masked occurrences of dictionary unit. |
578 | F->ExecuteCallback(Data.data(), Data.size()); |
579 | ModifiedFeatures.clear(); |
580 | TPC.CollectFeatures([&](size_t Feature) { |
581 | ModifiedFeatures.push_back(Feature); |
582 | }); |
583 | |
584 | if (InitialFeatures == ModifiedFeatures) |
585 | --Scores[i]; |
586 | else |
587 | Scores[i] += 2; |
588 | } |
589 | } |
590 | |
591 | Printf("###### Useless dictionary elements. ######\n" ); |
592 | for (size_t i = 0; i < Dict.size(); ++i) { |
593 | // Dictionary units with positive score are treated as useful ones. |
594 | if (Scores[i] > 0) |
595 | continue; |
596 | |
597 | Printf("\"" ); |
598 | PrintASCII(Dict[i].data(), Dict[i].size(), "\"" ); |
599 | Printf(" # Score: %d, Used: %d\n" , Scores[i], Usages[i]); |
600 | } |
601 | Printf("###### End of useless dictionary elements. ######\n" ); |
602 | return 0; |
603 | } |
604 | |
605 | std::vector<std::string> ParseSeedInuts(const char *seed_inputs) { |
606 | // Parse -seed_inputs=file1,file2,... or -seed_inputs=@seed_inputs_file |
607 | std::vector<std::string> Files; |
608 | if (!seed_inputs) return Files; |
609 | std::string SeedInputs; |
610 | if (Flags.seed_inputs[0] == '@') |
611 | SeedInputs = FileToString(Flags.seed_inputs + 1); // File contains list. |
612 | else |
613 | SeedInputs = Flags.seed_inputs; // seed_inputs contains the list. |
614 | if (SeedInputs.empty()) { |
615 | Printf("seed_inputs is empty or @file does not exist.\n" ); |
616 | exit(1); |
617 | } |
618 | // Parse SeedInputs. |
619 | size_t comma_pos = 0; |
620 | while ((comma_pos = SeedInputs.find_last_of(',')) != std::string::npos) { |
621 | Files.push_back(SeedInputs.substr(comma_pos + 1)); |
622 | SeedInputs = SeedInputs.substr(0, comma_pos); |
623 | } |
624 | Files.push_back(SeedInputs); |
625 | return Files; |
626 | } |
627 | |
628 | static std::vector<SizedFile> |
629 | ReadCorpora(const std::vector<std::string> &CorpusDirs, |
630 | const std::vector<std::string> &ExtraSeedFiles) { |
631 | std::vector<SizedFile> SizedFiles; |
632 | size_t LastNumFiles = 0; |
633 | for (auto &Dir : CorpusDirs) { |
634 | GetSizedFilesFromDir(Dir, &SizedFiles); |
635 | Printf("INFO: % 8zd files found in %s\n" , SizedFiles.size() - LastNumFiles, |
636 | Dir.c_str()); |
637 | LastNumFiles = SizedFiles.size(); |
638 | } |
639 | for (auto &File : ExtraSeedFiles) |
640 | if (auto Size = FileSize(File)) |
641 | SizedFiles.push_back({File, Size}); |
642 | return SizedFiles; |
643 | } |
644 | |
645 | int FuzzerDriver(int *argc, char ***argv, UserCallback Callback) { |
646 | using namespace fuzzer; |
647 | assert(argc && argv && "Argument pointers cannot be nullptr" ); |
648 | std::string Argv0((*argv)[0]); |
649 | EF = new ExternalFunctions(); |
650 | if (EF->LLVMFuzzerInitialize) |
651 | EF->LLVMFuzzerInitialize(argc, argv); |
652 | if (EF->__msan_scoped_disable_interceptor_checks) |
653 | EF->__msan_scoped_disable_interceptor_checks(); |
654 | const std::vector<std::string> Args(*argv, *argv + *argc); |
655 | assert(!Args.empty()); |
656 | ProgName = new std::string(Args[0]); |
657 | if (Argv0 != *ProgName) { |
658 | Printf("ERROR: argv[0] has been modified in LLVMFuzzerInitialize\n" ); |
659 | exit(1); |
660 | } |
661 | ParseFlags(Args, EF); |
662 | if (Flags.help) { |
663 | PrintHelp(); |
664 | return 0; |
665 | } |
666 | |
667 | if (Flags.close_fd_mask & 2) |
668 | DupAndCloseStderr(); |
669 | if (Flags.close_fd_mask & 1) |
670 | CloseStdout(); |
671 | |
672 | if (Flags.jobs > 0 && Flags.workers == 0) { |
673 | Flags.workers = std::min(NumberOfCpuCores() / 2, Flags.jobs); |
674 | if (Flags.workers > 1) |
675 | Printf("Running %u workers\n" , Flags.workers); |
676 | } |
677 | |
678 | if (Flags.workers > 0 && Flags.jobs > 0) |
679 | return RunInMultipleProcesses(Args, Flags.workers, Flags.jobs); |
680 | |
681 | FuzzingOptions Options; |
682 | Options.Verbosity = Flags.verbosity; |
683 | Options.MaxLen = Flags.max_len; |
684 | Options.LenControl = Flags.len_control; |
685 | Options.KeepSeed = Flags.keep_seed; |
686 | Options.UnitTimeoutSec = Flags.timeout; |
687 | Options.ErrorExitCode = Flags.error_exitcode; |
688 | Options.TimeoutExitCode = Flags.timeout_exitcode; |
689 | Options.IgnoreTimeouts = Flags.ignore_timeouts; |
690 | Options.IgnoreOOMs = Flags.ignore_ooms; |
691 | Options.IgnoreCrashes = Flags.ignore_crashes; |
692 | Options.MaxTotalTimeSec = Flags.max_total_time; |
693 | Options.DoCrossOver = Flags.cross_over; |
694 | Options.CrossOverUniformDist = Flags.cross_over_uniform_dist; |
695 | Options.MutateDepth = Flags.mutate_depth; |
696 | Options.ReduceDepth = Flags.reduce_depth; |
697 | Options.UseCounters = Flags.use_counters; |
698 | Options.UseMemmem = Flags.use_memmem; |
699 | Options.UseCmp = Flags.use_cmp; |
700 | Options.UseValueProfile = Flags.use_value_profile; |
701 | Options.Shrink = Flags.shrink; |
702 | Options.ReduceInputs = Flags.reduce_inputs; |
703 | Options.ShuffleAtStartUp = Flags.shuffle; |
704 | Options.PreferSmall = Flags.prefer_small; |
705 | Options.ReloadIntervalSec = Flags.reload; |
706 | Options.OnlyASCII = Flags.only_ascii; |
707 | Options.DetectLeaks = Flags.detect_leaks; |
708 | Options.PurgeAllocatorIntervalSec = Flags.purge_allocator_interval; |
709 | Options.TraceMalloc = Flags.trace_malloc; |
710 | Options.RssLimitMb = Flags.rss_limit_mb; |
711 | Options.MallocLimitMb = Flags.malloc_limit_mb; |
712 | if (!Options.MallocLimitMb) |
713 | Options.MallocLimitMb = Options.RssLimitMb; |
714 | if (Flags.runs >= 0) |
715 | Options.MaxNumberOfRuns = Flags.runs; |
716 | if (!Inputs->empty() && !Flags.minimize_crash_internal_step) { |
717 | // Ensure output corpus assumed to be the first arbitrary argument input |
718 | // is not a path to an existing file. |
719 | std::string OutputCorpusDir = (*Inputs)[0]; |
720 | if (!IsFile(OutputCorpusDir)) { |
721 | Options.OutputCorpus = OutputCorpusDir; |
722 | ValidateDirectoryExists(Options.OutputCorpus, Flags.create_missing_dirs); |
723 | } |
724 | } |
725 | Options.ReportSlowUnits = Flags.report_slow_units; |
726 | if (Flags.artifact_prefix) { |
727 | Options.ArtifactPrefix = Flags.artifact_prefix; |
728 | |
729 | // Since the prefix could be a full path to a file name prefix, assume |
730 | // that if the path ends with the platform's separator that a directory |
731 | // is desired |
732 | std::string ArtifactPathDir = Options.ArtifactPrefix; |
733 | if (!IsSeparator(ArtifactPathDir[ArtifactPathDir.length() - 1])) { |
734 | ArtifactPathDir = DirName(ArtifactPathDir); |
735 | } |
736 | ValidateDirectoryExists(ArtifactPathDir, Flags.create_missing_dirs); |
737 | } |
738 | if (Flags.exact_artifact_path) { |
739 | Options.ExactArtifactPath = Flags.exact_artifact_path; |
740 | ValidateDirectoryExists(DirName(Options.ExactArtifactPath), |
741 | Flags.create_missing_dirs); |
742 | } |
743 | std::vector<Unit> Dictionary; |
744 | if (Flags.dict) |
745 | if (!ParseDictionaryFile(FileToString(Flags.dict), &Dictionary)) |
746 | return 1; |
747 | if (Flags.verbosity > 0 && !Dictionary.empty()) |
748 | Printf("Dictionary: %zd entries\n" , Dictionary.size()); |
749 | bool RunIndividualFiles = AllInputsAreFiles(); |
750 | Options.SaveArtifacts = |
751 | !RunIndividualFiles || Flags.minimize_crash_internal_step; |
752 | Options.PrintNewCovPcs = Flags.print_pcs; |
753 | Options.PrintNewCovFuncs = Flags.print_funcs; |
754 | Options.PrintFinalStats = Flags.print_final_stats; |
755 | Options.PrintCorpusStats = Flags.print_corpus_stats; |
756 | Options.PrintCoverage = Flags.print_coverage; |
757 | Options.PrintFullCoverage = Flags.print_full_coverage; |
758 | if (Flags.exit_on_src_pos) |
759 | Options.ExitOnSrcPos = Flags.exit_on_src_pos; |
760 | if (Flags.exit_on_item) |
761 | Options.ExitOnItem = Flags.exit_on_item; |
762 | if (Flags.focus_function) |
763 | Options.FocusFunction = Flags.focus_function; |
764 | if (Flags.data_flow_trace) |
765 | Options.DataFlowTrace = Flags.data_flow_trace; |
766 | if (Flags.features_dir) { |
767 | Options.FeaturesDir = Flags.features_dir; |
768 | ValidateDirectoryExists(Options.FeaturesDir, Flags.create_missing_dirs); |
769 | } |
770 | if (Flags.mutation_graph_file) |
771 | Options.MutationGraphFile = Flags.mutation_graph_file; |
772 | if (Flags.collect_data_flow) |
773 | Options.CollectDataFlow = Flags.collect_data_flow; |
774 | if (Flags.stop_file) |
775 | Options.StopFile = Flags.stop_file; |
776 | Options.Entropic = Flags.entropic; |
777 | Options.EntropicFeatureFrequencyThreshold = |
778 | (size_t)Flags.entropic_feature_frequency_threshold; |
779 | Options.EntropicNumberOfRarestFeatures = |
780 | (size_t)Flags.entropic_number_of_rarest_features; |
781 | Options.EntropicScalePerExecTime = Flags.entropic_scale_per_exec_time; |
782 | if (!Options.FocusFunction.empty()) |
783 | Options.Entropic = false; // FocusFunction overrides entropic scheduling. |
784 | if (Options.Entropic) |
785 | Printf("INFO: Running with entropic power schedule (0x%zX, %zu).\n" , |
786 | Options.EntropicFeatureFrequencyThreshold, |
787 | Options.EntropicNumberOfRarestFeatures); |
788 | struct EntropicOptions Entropic; |
789 | Entropic.Enabled = Options.Entropic; |
790 | Entropic.FeatureFrequencyThreshold = |
791 | Options.EntropicFeatureFrequencyThreshold; |
792 | Entropic.NumberOfRarestFeatures = Options.EntropicNumberOfRarestFeatures; |
793 | Entropic.ScalePerExecTime = Options.EntropicScalePerExecTime; |
794 | |
795 | unsigned Seed = Flags.seed; |
796 | // Initialize Seed. |
797 | if (Seed == 0) |
798 | Seed = static_cast<unsigned>( |
799 | std::chrono::system_clock::now().time_since_epoch().count() + GetPid()); |
800 | if (Flags.verbosity) |
801 | Printf("INFO: Seed: %u\n" , Seed); |
802 | |
803 | if (Flags.collect_data_flow && Flags.data_flow_trace && !Flags.fork && |
804 | !(Flags.merge || Flags.set_cover_merge)) { |
805 | if (RunIndividualFiles) |
806 | return CollectDataFlow(Flags.collect_data_flow, Flags.data_flow_trace, |
807 | ReadCorpora({}, *Inputs)); |
808 | else |
809 | return CollectDataFlow(Flags.collect_data_flow, Flags.data_flow_trace, |
810 | ReadCorpora(*Inputs, {})); |
811 | } |
812 | |
813 | Random Rand(Seed); |
814 | auto *MD = new MutationDispatcher(Rand, Options); |
815 | auto *Corpus = new InputCorpus(Options.OutputCorpus, Entropic); |
816 | auto *F = new Fuzzer(Callback, *Corpus, *MD, Options); |
817 | |
818 | for (auto &U: Dictionary) |
819 | if (U.size() <= Word::GetMaxSize()) |
820 | MD->AddWordToManualDictionary(Word(U.data(), U.size())); |
821 | |
822 | // Threads are only supported by Chrome. Don't use them with emscripten |
823 | // for now. |
824 | #if !LIBFUZZER_EMSCRIPTEN |
825 | StartRssThread(F, Flags.rss_limit_mb); |
826 | #endif // LIBFUZZER_EMSCRIPTEN |
827 | |
828 | Options.HandleAbrt = Flags.handle_abrt; |
829 | Options.HandleAlrm = !Flags.minimize_crash; |
830 | Options.HandleBus = Flags.handle_bus; |
831 | Options.HandleFpe = Flags.handle_fpe; |
832 | Options.HandleIll = Flags.handle_ill; |
833 | Options.HandleInt = Flags.handle_int; |
834 | Options.HandleSegv = Flags.handle_segv; |
835 | Options.HandleTerm = Flags.handle_term; |
836 | Options.HandleXfsz = Flags.handle_xfsz; |
837 | Options.HandleUsr1 = Flags.handle_usr1; |
838 | Options.HandleUsr2 = Flags.handle_usr2; |
839 | Options.HandleWinExcept = Flags.handle_winexcept; |
840 | |
841 | SetSignalHandler(Options); |
842 | |
843 | std::atexit(Fuzzer::StaticExitCallback); |
844 | |
845 | if (Flags.minimize_crash) |
846 | return MinimizeCrashInput(Args, Options); |
847 | |
848 | if (Flags.minimize_crash_internal_step) |
849 | return MinimizeCrashInputInternalStep(F, Corpus); |
850 | |
851 | if (Flags.cleanse_crash) |
852 | return CleanseCrashInput(Args, Options); |
853 | |
854 | if (RunIndividualFiles) { |
855 | Options.SaveArtifacts = false; |
856 | int Runs = std::max(1, Flags.runs); |
857 | Printf("%s: Running %zd inputs %d time(s) each.\n" , ProgName->c_str(), |
858 | Inputs->size(), Runs); |
859 | for (auto &Path : *Inputs) { |
860 | auto StartTime = system_clock::now(); |
861 | Printf("Running: %s\n" , Path.c_str()); |
862 | for (int Iter = 0; Iter < Runs; Iter++) |
863 | RunOneTest(F, Path.c_str(), Options.MaxLen); |
864 | auto StopTime = system_clock::now(); |
865 | auto MS = duration_cast<milliseconds>(StopTime - StartTime).count(); |
866 | Printf("Executed %s in %ld ms\n" , Path.c_str(), (long)MS); |
867 | } |
868 | Printf("***\n" |
869 | "*** NOTE: fuzzing was not performed, you have only\n" |
870 | "*** executed the target code on a fixed set of inputs.\n" |
871 | "***\n" ); |
872 | F->PrintFinalStats(); |
873 | exit(0); |
874 | } |
875 | |
876 | Options.ForkCorpusGroups = Flags.fork_corpus_groups; |
877 | if (Flags.fork) |
878 | FuzzWithFork(F->GetMD().GetRand(), Options, Args, *Inputs, Flags.fork); |
879 | |
880 | if (Flags.merge || Flags.set_cover_merge) |
881 | Merge(F, Options, Args, *Inputs, Flags.merge_control_file); |
882 | |
883 | if (Flags.merge_inner) { |
884 | const size_t kDefaultMaxMergeLen = 1 << 20; |
885 | if (Options.MaxLen == 0) |
886 | F->SetMaxInputLen(kDefaultMaxMergeLen); |
887 | assert(Flags.merge_control_file); |
888 | F->CrashResistantMergeInternalStep(Flags.merge_control_file, |
889 | !strncmp(Flags.merge_inner, "2" , 1)); |
890 | exit(0); |
891 | } |
892 | |
893 | if (Flags.analyze_dict) { |
894 | size_t MaxLen = INT_MAX; // Large max length. |
895 | UnitVector InitialCorpus; |
896 | for (auto &Inp : *Inputs) { |
897 | Printf("Loading corpus dir: %s\n" , Inp.c_str()); |
898 | ReadDirToVectorOfUnits(Inp.c_str(), &InitialCorpus, nullptr, |
899 | MaxLen, /*ExitOnError=*/false); |
900 | } |
901 | |
902 | if (Dictionary.empty() || Inputs->empty()) { |
903 | Printf("ERROR: can't analyze dict without dict and corpus provided\n" ); |
904 | return 1; |
905 | } |
906 | if (AnalyzeDictionary(F, Dictionary, InitialCorpus)) { |
907 | Printf("Dictionary analysis failed\n" ); |
908 | exit(1); |
909 | } |
910 | Printf("Dictionary analysis succeeded\n" ); |
911 | exit(0); |
912 | } |
913 | |
914 | auto CorporaFiles = ReadCorpora(*Inputs, ParseSeedInuts(Flags.seed_inputs)); |
915 | F->Loop(CorporaFiles); |
916 | |
917 | if (Flags.verbosity) |
918 | Printf("Done %zd runs in %zd second(s)\n" , F->getTotalNumberOfRuns(), |
919 | F->secondsSinceProcessStartUp()); |
920 | F->PrintFinalStats(); |
921 | |
922 | exit(0); // Don't let F destroy itself. |
923 | } |
924 | |
925 | extern "C" ATTRIBUTE_INTERFACE int |
926 | LLVMFuzzerRunDriver(int *argc, char ***argv, |
927 | int (*UserCb)(const uint8_t *Data, size_t Size)) { |
928 | return FuzzerDriver(argc, argv, UserCb); |
929 | } |
930 | |
931 | // Storage for global ExternalFunctions object. |
932 | ExternalFunctions *EF = nullptr; |
933 | |
934 | } // namespace fuzzer |
935 | |