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 <fstream>
28#include <functional>
29#include <mutex>
30#include <string>
31#include <thread>
32
33// This function should be present in the libFuzzer so that the client
34// binary can test for its existence.
35#if LIBFUZZER_MSVC
36extern "C" void __libfuzzer_is_present() {}
37#if defined(_M_IX86) || defined(__i386__)
38#pragma comment(linker, "/include:___libfuzzer_is_present")
39#else
40#pragma comment(linker, "/include:__libfuzzer_is_present")
41#endif
42#else
43extern "C" __attribute__((used)) void __libfuzzer_is_present() {}
44#endif // LIBFUZZER_MSVC
45
46namespace fuzzer {
47
48// Program arguments.
49struct FlagDescription {
50 const char *Name;
51 const char *Description;
52 int Default;
53 int *IntFlag;
54 const char **StrFlag;
55 unsigned int *UIntFlag;
56};
57
58struct {
59#define FUZZER_DEPRECATED_FLAG(Name)
60#define FUZZER_FLAG_INT(Name, Default, Description) int Name;
61#define FUZZER_FLAG_UNSIGNED(Name, Default, Description) unsigned int Name;
62#define FUZZER_FLAG_STRING(Name, Description) const char *Name;
63#include "FuzzerFlags.def"
64#undef FUZZER_DEPRECATED_FLAG
65#undef FUZZER_FLAG_INT
66#undef FUZZER_FLAG_UNSIGNED
67#undef FUZZER_FLAG_STRING
68} Flags;
69
70static const FlagDescription FlagDescriptions [] {
71#define FUZZER_DEPRECATED_FLAG(Name) \
72 {#Name, "Deprecated; don't use", 0, nullptr, nullptr, nullptr},
73#define FUZZER_FLAG_INT(Name, Default, Description) \
74 {#Name, Description, Default, &Flags.Name, nullptr, nullptr},
75#define FUZZER_FLAG_UNSIGNED(Name, Default, Description) \
76 {#Name, Description, static_cast<int>(Default), \
77 nullptr, nullptr, &Flags.Name},
78#define FUZZER_FLAG_STRING(Name, Description) \
79 {#Name, Description, 0, nullptr, &Flags.Name, nullptr},
80#include "FuzzerFlags.def"
81#undef FUZZER_DEPRECATED_FLAG
82#undef FUZZER_FLAG_INT
83#undef FUZZER_FLAG_UNSIGNED
84#undef FUZZER_FLAG_STRING
85};
86
87static const size_t kNumFlags =
88 sizeof(FlagDescriptions) / sizeof(FlagDescriptions[0]);
89
90static std::vector<std::string> *Inputs;
91static std::string *ProgName;
92
93static void PrintHelp() {
94 Printf(Fmt: "Usage:\n");
95 auto Prog = ProgName->c_str();
96 Printf(Fmt: "\nTo run fuzzing pass 0 or more directories.\n");
97 Printf(Fmt: "%s [-flag1=val1 [-flag2=val2 ...] ] [dir1 [dir2 ...] ]\n", Prog);
98
99 Printf(Fmt: "\nTo run individual tests without fuzzing pass 1 or more files:\n");
100 Printf(Fmt: "%s [-flag1=val1 [-flag2=val2 ...] ] file1 [file2 ...]\n", Prog);
101
102 Printf(Fmt: "\nFlags: (strictly in form -flag=value)\n");
103 size_t MaxFlagLen = 0;
104 for (size_t F = 0; F < kNumFlags; F++)
105 MaxFlagLen = std::max(a: strlen(s: FlagDescriptions[F].Name), b: MaxFlagLen);
106
107 for (size_t F = 0; F < kNumFlags; F++) {
108 const auto &D = FlagDescriptions[F];
109 if (strstr(haystack: D.Description, needle: "internal flag") == D.Description) continue;
110 Printf(Fmt: " %s", D.Name);
111 for (size_t i = 0, n = MaxFlagLen - strlen(s: D.Name); i < n; i++)
112 Printf(Fmt: " ");
113 Printf(Fmt: "\t");
114 Printf(Fmt: "%d\t%s\n", D.Default, D.Description);
115 }
116 Printf(Fmt: "\nFlags starting with '--' will be ignored and "
117 "will be passed verbatim to subprocesses.\n");
118}
119
120static const char *FlagValue(const char *Param, const char *Name) {
121 size_t Len = strlen(s: Name);
122 if (Param[0] == '-' && strstr(haystack: Param + 1, needle: Name) == Param + 1 &&
123 Param[Len + 1] == '=')
124 return &Param[Len + 2];
125 return nullptr;
126}
127
128// Avoid calling stol as it triggers a bug in clang/glibc build.
129static long MyStol(const char *Str) {
130 long Res = 0;
131 long Sign = 1;
132 if (*Str == '-') {
133 Str++;
134 Sign = -1;
135 }
136 for (size_t i = 0; Str[i]; i++) {
137 char Ch = Str[i];
138 if (Ch < '0' || Ch > '9')
139 return Res;
140 Res = Res * 10 + (Ch - '0');
141 }
142 return Res * Sign;
143}
144
145static bool ParseOneFlag(const char *Param) {
146 if (Param[0] != '-') return false;
147 if (Param[1] == '-') {
148 static bool PrintedWarning = false;
149 if (!PrintedWarning) {
150 PrintedWarning = true;
151 Printf(Fmt: "INFO: libFuzzer ignores flags that start with '--'\n");
152 }
153 for (size_t F = 0; F < kNumFlags; F++)
154 if (FlagValue(Param: Param + 1, Name: FlagDescriptions[F].Name))
155 Printf(Fmt: "WARNING: did you mean '%s' (single dash)?\n", Param + 1);
156 return true;
157 }
158 for (size_t F = 0; F < kNumFlags; F++) {
159 const char *Name = FlagDescriptions[F].Name;
160 const char *Str = FlagValue(Param, Name);
161 if (Str) {
162 if (FlagDescriptions[F].IntFlag) {
163 auto Val = MyStol(Str);
164 *FlagDescriptions[F].IntFlag = static_cast<int>(Val);
165 if (Flags.verbosity >= 2)
166 Printf(Fmt: "Flag: %s %d\n", Name, Val);
167 return true;
168 } else if (FlagDescriptions[F].UIntFlag) {
169 auto Val = std::stoul(str: Str);
170 *FlagDescriptions[F].UIntFlag = static_cast<unsigned int>(Val);
171 if (Flags.verbosity >= 2)
172 Printf(Fmt: "Flag: %s %u\n", Name, Val);
173 return true;
174 } else if (FlagDescriptions[F].StrFlag) {
175 *FlagDescriptions[F].StrFlag = Str;
176 if (Flags.verbosity >= 2)
177 Printf(Fmt: "Flag: %s %s\n", Name, Str);
178 return true;
179 } else { // Deprecated flag.
180 Printf(Fmt: "Flag: %s: deprecated, don't use\n", Name);
181 return true;
182 }
183 }
184 }
185 Printf(Fmt: "\n\nWARNING: unrecognized flag '%s'; "
186 "use -help=1 to list all flags\n\n", Param);
187 return true;
188}
189
190// We don't use any library to minimize dependencies.
191static void ParseFlags(const std::vector<std::string> &Args,
192 const ExternalFunctions *EF) {
193 for (size_t F = 0; F < kNumFlags; F++) {
194 if (FlagDescriptions[F].IntFlag)
195 *FlagDescriptions[F].IntFlag = FlagDescriptions[F].Default;
196 if (FlagDescriptions[F].UIntFlag)
197 *FlagDescriptions[F].UIntFlag =
198 static_cast<unsigned int>(FlagDescriptions[F].Default);
199 if (FlagDescriptions[F].StrFlag)
200 *FlagDescriptions[F].StrFlag = nullptr;
201 }
202
203 // Disable len_control by default, if LLVMFuzzerCustomMutator is used.
204 if (EF->LLVMFuzzerCustomMutator) {
205 Flags.len_control = 0;
206 Printf(Fmt: "INFO: found LLVMFuzzerCustomMutator (%p). "
207 "Disabling -len_control by default.\n", EF->LLVMFuzzerCustomMutator);
208 }
209
210 Inputs = new std::vector<std::string>;
211 for (size_t A = 1; A < Args.size(); A++) {
212 if (ParseOneFlag(Param: Args[A].c_str())) {
213 if (Flags.ignore_remaining_args)
214 break;
215 continue;
216 }
217 Inputs->push_back(x: Args[A]);
218 }
219}
220
221static std::mutex Mu;
222
223static void PulseThread() {
224 while (true) {
225 SleepSeconds(Seconds: 600);
226 std::lock_guard<std::mutex> Lock(Mu);
227 Printf(Fmt: "pulse...\n");
228 }
229}
230
231static void WorkerThread(const Command &BaseCmd, std::atomic<unsigned> *Counter,
232 unsigned NumJobs, std::atomic<bool> *HasErrors) {
233 ScopedDisableMsanInterceptorChecks S;
234 while (true) {
235 unsigned C = (*Counter)++;
236 if (C >= NumJobs) break;
237 std::string Log = "fuzz-" + std::to_string(val: C) + ".log";
238 Command Cmd(BaseCmd);
239 Cmd.setOutputFile(Log);
240 Cmd.combineOutAndErr();
241 if (Flags.verbosity) {
242 std::string CommandLine = Cmd.toString();
243 Printf(Fmt: "%s\n", CommandLine.c_str());
244 }
245 int ExitCode = ExecuteCommand(Cmd);
246 if (ExitCode != 0)
247 *HasErrors = true;
248 std::lock_guard<std::mutex> Lock(Mu);
249 Printf(Fmt: "================== Job %u exited with exit code %d ============\n",
250 C, ExitCode);
251 fuzzer::CopyFileToErr(Path: Log);
252 }
253}
254
255static void ValidateDirectoryExists(const std::string &Path,
256 bool CreateDirectory) {
257 if (Path.empty()) {
258 Printf(Fmt: "ERROR: Provided directory path is an empty string\n");
259 exit(status: 1);
260 }
261
262 if (IsDirectory(Path))
263 return;
264
265 if (CreateDirectory) {
266 if (!MkDirRecursive(Dir: Path)) {
267 Printf(Fmt: "ERROR: Failed to create directory \"%s\"\n", Path.c_str());
268 exit(status: 1);
269 }
270 return;
271 }
272
273 Printf(Fmt: "ERROR: The required directory \"%s\" does not exist\n", Path.c_str());
274 exit(status: 1);
275}
276
277std::string CloneArgsWithoutX(const std::vector<std::string> &Args,
278 const char *X1, const char *X2) {
279 std::string Cmd;
280 for (auto &S : Args) {
281 if (FlagValue(Param: S.c_str(), Name: X1) || FlagValue(Param: S.c_str(), Name: X2))
282 continue;
283 Cmd += S + " ";
284 }
285 return Cmd;
286}
287
288static int RunInMultipleProcesses(const std::vector<std::string> &Args,
289 unsigned NumWorkers, unsigned NumJobs) {
290 std::atomic<unsigned> Counter(0);
291 std::atomic<bool> HasErrors(false);
292 Command Cmd(Args);
293 Cmd.removeFlag(Flag: "jobs");
294 Cmd.removeFlag(Flag: "workers");
295 std::vector<std::thread> V;
296 std::thread Pulse(PulseThread);
297 Pulse.detach();
298 V.resize(new_size: NumWorkers);
299 for (unsigned i = 0; i < NumWorkers; i++) {
300 V[i] = std::thread(WorkerThread, std::ref(t&: Cmd), &Counter, NumJobs,
301 &HasErrors);
302 SetThreadName(thread&: V[i], name: "FuzzerWorker");
303 }
304 for (auto &T : V)
305 T.join();
306 return HasErrors ? 1 : 0;
307}
308
309static void RssThread(Fuzzer *F, size_t RssLimitMb) {
310 while (true) {
311 SleepSeconds(Seconds: 1);
312 size_t Peak = GetPeakRSSMb();
313 if (Peak > RssLimitMb)
314 F->RssLimitCallback();
315 }
316}
317
318static void StartRssThread(Fuzzer *F, size_t RssLimitMb) {
319 if (!RssLimitMb)
320 return;
321 std::thread T(RssThread, F, RssLimitMb);
322 T.detach();
323}
324
325int RunOneTest(Fuzzer *F, const char *InputFilePath, size_t MaxLen) {
326 Unit U = FileToVector(Path: InputFilePath);
327 if (MaxLen && MaxLen < U.size())
328 U.resize(new_size: MaxLen);
329 F->ExecuteCallback(Data: U.data(), Size: U.size());
330 if (Flags.print_full_coverage) {
331 // Leak detection is not needed when collecting full coverage data.
332 F->TPCUpdateObservedPCs();
333 } else {
334 F->TryDetectingAMemoryLeak(Data: U.data(), Size: U.size(), DuringInitialCorpusExecution: true);
335 }
336 return 0;
337}
338
339static bool AllInputsAreFiles() {
340 if (Inputs->empty()) return false;
341 for (auto &Path : *Inputs)
342 if (!IsFile(Path))
343 return false;
344 return true;
345}
346
347static std::string GetDedupTokenFromCmdOutput(const std::string &S) {
348 auto Beg = S.find(s: "DEDUP_TOKEN:");
349 if (Beg == std::string::npos)
350 return "";
351 auto End = S.find(c: '\n', pos: Beg);
352 if (End == std::string::npos)
353 return "";
354 return S.substr(pos: Beg, n: End - Beg);
355}
356
357int CleanseCrashInput(const std::vector<std::string> &Args,
358 const FuzzingOptions &Options) {
359 if (Inputs->size() != 1 || !Flags.exact_artifact_path) {
360 Printf(Fmt: "ERROR: -cleanse_crash should be given one input file and"
361 " -exact_artifact_path\n");
362 exit(status: 1);
363 }
364 std::string InputFilePath = Inputs->at(n: 0);
365 std::string OutputFilePath = Flags.exact_artifact_path;
366 Command Cmd(Args);
367 Cmd.removeFlag(Flag: "cleanse_crash");
368
369 assert(Cmd.hasArgument(InputFilePath));
370 Cmd.removeArgument(Arg: InputFilePath);
371
372 auto TmpFilePath = TempPath(Prefix: "CleanseCrashInput", Extension: ".repro");
373 Cmd.addArgument(Arg: TmpFilePath);
374 Cmd.setOutputFile(getDevNull());
375 Cmd.combineOutAndErr();
376
377 std::string CurrentFilePath = InputFilePath;
378 auto U = FileToVector(Path: CurrentFilePath);
379 size_t Size = U.size();
380
381 const std::vector<uint8_t> ReplacementBytes = {' ', 0xff};
382 for (int NumAttempts = 0; NumAttempts < 5; NumAttempts++) {
383 bool Changed = false;
384 for (size_t Idx = 0; Idx < Size; Idx++) {
385 Printf(Fmt: "CLEANSE[%d]: Trying to replace byte %zd of %zd\n", NumAttempts,
386 Idx, Size);
387 uint8_t OriginalByte = U[Idx];
388 if (ReplacementBytes.end() != std::find(first: ReplacementBytes.begin(),
389 last: ReplacementBytes.end(),
390 val: OriginalByte))
391 continue;
392 for (auto NewByte : ReplacementBytes) {
393 U[Idx] = NewByte;
394 WriteToFile(U, Path: TmpFilePath);
395 auto ExitCode = ExecuteCommand(Cmd);
396 RemoveFile(Path: TmpFilePath);
397 if (!ExitCode) {
398 U[Idx] = OriginalByte;
399 } else {
400 Changed = true;
401 Printf(Fmt: "CLEANSE: Replaced byte %zd with 0x%x\n", Idx, NewByte);
402 WriteToFile(U, Path: OutputFilePath);
403 break;
404 }
405 }
406 }
407 if (!Changed) break;
408 }
409 return 0;
410}
411
412int MinimizeCrashInput(const std::vector<std::string> &Args,
413 const FuzzingOptions &Options) {
414 if (Inputs->size() != 1) {
415 Printf(Fmt: "ERROR: -minimize_crash should be given one input file\n");
416 exit(status: 1);
417 }
418 std::string InputFilePath = Inputs->at(n: 0);
419 Command BaseCmd(Args);
420 BaseCmd.removeFlag(Flag: "minimize_crash");
421 BaseCmd.removeFlag(Flag: "exact_artifact_path");
422 assert(BaseCmd.hasArgument(InputFilePath));
423 BaseCmd.removeArgument(Arg: InputFilePath);
424 if (Flags.runs <= 0 && Flags.max_total_time == 0) {
425 Printf(Fmt: "INFO: you need to specify -runs=N or "
426 "-max_total_time=N with -minimize_crash=1\n"
427 "INFO: defaulting to -max_total_time=600\n");
428 BaseCmd.addFlag(Flag: "max_total_time", Value: "600");
429 }
430
431 BaseCmd.combineOutAndErr();
432
433 std::string CurrentFilePath = InputFilePath;
434 while (true) {
435 Unit U = FileToVector(Path: CurrentFilePath);
436 Printf(Fmt: "CRASH_MIN: minimizing crash input: '%s' (%zd bytes)\n",
437 CurrentFilePath.c_str(), U.size());
438
439 Command Cmd(BaseCmd);
440 Cmd.addArgument(Arg: CurrentFilePath);
441
442 Printf(Fmt: "CRASH_MIN: executing: %s\n", Cmd.toString().c_str());
443 std::string CmdOutput;
444 bool Success = ExecuteCommand(Cmd, CmdOutput: &CmdOutput);
445 if (Success) {
446 Printf(Fmt: "ERROR: the input %s did not crash\n", CurrentFilePath.c_str());
447 exit(status: 1);
448 }
449 Printf(Fmt: "CRASH_MIN: '%s' (%zd bytes) caused a crash. Will try to minimize "
450 "it further\n",
451 CurrentFilePath.c_str(), U.size());
452 auto DedupToken1 = GetDedupTokenFromCmdOutput(S: CmdOutput);
453 if (!DedupToken1.empty())
454 Printf(Fmt: "CRASH_MIN: DedupToken1: %s\n", DedupToken1.c_str());
455
456 std::string ArtifactPath =
457 Flags.exact_artifact_path
458 ? Flags.exact_artifact_path
459 : Options.ArtifactPrefix + "minimized-from-" + Hash(U);
460 Cmd.addFlag(Flag: "minimize_crash_internal_step", Value: "1");
461 Cmd.addFlag(Flag: "exact_artifact_path", Value: ArtifactPath);
462 Printf(Fmt: "CRASH_MIN: executing: %s\n", Cmd.toString().c_str());
463 CmdOutput.clear();
464 Success = ExecuteCommand(Cmd, CmdOutput: &CmdOutput);
465 Printf(Fmt: "%s", CmdOutput.c_str());
466 if (Success) {
467 if (Flags.exact_artifact_path) {
468 CurrentFilePath = Flags.exact_artifact_path;
469 WriteToFile(U, Path: CurrentFilePath);
470 }
471 Printf(Fmt: "CRASH_MIN: failed to minimize beyond %s (%zu bytes), exiting\n",
472 CurrentFilePath.c_str(), U.size());
473 break;
474 }
475 auto DedupToken2 = GetDedupTokenFromCmdOutput(S: CmdOutput);
476 if (!DedupToken2.empty())
477 Printf(Fmt: "CRASH_MIN: DedupToken2: %s\n", DedupToken2.c_str());
478
479 if (DedupToken1 != DedupToken2) {
480 if (Flags.exact_artifact_path) {
481 CurrentFilePath = Flags.exact_artifact_path;
482 WriteToFile(U, Path: CurrentFilePath);
483 }
484 Printf(Fmt: "CRASH_MIN: mismatch in dedup tokens"
485 " (looks like a different bug). Won't minimize further\n");
486 break;
487 }
488
489 CurrentFilePath = ArtifactPath;
490 Printf(Fmt: "*********************************\n");
491 }
492 return 0;
493}
494
495int MinimizeCrashInputInternalStep(Fuzzer *F, InputCorpus *Corpus) {
496 assert(Inputs->size() == 1);
497 std::string InputFilePath = Inputs->at(n: 0);
498 Unit U = FileToVector(Path: InputFilePath);
499 Printf(Fmt: "INFO: Starting MinimizeCrashInputInternalStep: %zd\n", U.size());
500 if (U.size() < 2) {
501 Printf(Fmt: "INFO: The input is small enough, exiting\n");
502 exit(status: 0);
503 }
504 F->SetMaxInputLen(U.size());
505 F->SetMaxMutationLen(U.size() - 1);
506 F->MinimizeCrashLoop(U);
507 Printf(Fmt: "INFO: Done MinimizeCrashInputInternalStep, no crashes found\n");
508 exit(status: 0);
509}
510
511void Merge(Fuzzer *F, FuzzingOptions &Options,
512 const std::vector<std::string> &Args,
513 const std::vector<std::string> &Corpora, const char *CFPathOrNull) {
514 if (Corpora.size() < 2) {
515 Printf(Fmt: "INFO: Merge requires two or more corpus dirs\n");
516 exit(status: 0);
517 }
518
519 std::vector<SizedFile> OldCorpus, NewCorpus;
520 GetSizedFilesFromDir(Dir: Corpora[0], V: &OldCorpus);
521 for (size_t i = 1; i < Corpora.size(); i++)
522 GetSizedFilesFromDir(Dir: Corpora[i], V: &NewCorpus);
523 std::sort(first: OldCorpus.begin(), last: OldCorpus.end());
524 std::sort(first: NewCorpus.begin(), last: NewCorpus.end());
525
526 std::string CFPath = CFPathOrNull ? CFPathOrNull : TempPath(Prefix: "Merge", Extension: ".txt");
527 std::vector<std::string> NewFiles;
528 std::set<uint32_t> NewFeatures, NewCov;
529 CrashResistantMerge(Args, OldCorpus, NewCorpus, NewFiles: &NewFiles, InitialFeatures: {}, NewFeatures: &NewFeatures,
530 InitialCov: {}, NewCov: &NewCov, CFPath, Verbose: true, IsSetCoverMerge: Flags.set_cover_merge);
531 for (auto &Path : NewFiles)
532 F->WriteToOutputCorpus(U: FileToVector(Path, MaxSize: Options.MaxLen));
533 // We are done, delete the control file if it was a temporary one.
534 if (!Flags.merge_control_file)
535 RemoveFile(Path: CFPath);
536
537 exit(status: 0);
538}
539
540int AnalyzeDictionary(Fuzzer *F, const std::vector<Unit> &Dict,
541 UnitVector &Corpus) {
542 Printf(Fmt: "Started dictionary minimization (up to %zu tests)\n",
543 Dict.size() * Corpus.size() * 2);
544
545 // Scores and usage count for each dictionary unit.
546 std::vector<int> Scores(Dict.size());
547 std::vector<int> Usages(Dict.size());
548
549 std::vector<size_t> InitialFeatures;
550 std::vector<size_t> ModifiedFeatures;
551 for (auto &C : Corpus) {
552 // Get coverage for the testcase without modifications.
553 F->ExecuteCallback(Data: C.data(), Size: C.size());
554 InitialFeatures.clear();
555 TPC.CollectFeatures(HandleFeature: [&](size_t Feature) {
556 InitialFeatures.push_back(x: Feature);
557 });
558
559 for (size_t i = 0; i < Dict.size(); ++i) {
560 std::vector<uint8_t> Data = C;
561 auto StartPos = std::search(first1: Data.begin(), last1: Data.end(),
562 first2: Dict[i].begin(), last2: Dict[i].end());
563 // Skip dictionary unit, if the testcase does not contain it.
564 if (StartPos == Data.end())
565 continue;
566
567 ++Usages[i];
568 while (StartPos != Data.end()) {
569 // Replace all occurrences of dictionary unit in the testcase.
570 auto EndPos = StartPos + Dict[i].size();
571 for (auto It = StartPos; It != EndPos; ++It)
572 *It ^= 0xFF;
573
574 StartPos = std::search(first1: EndPos, last1: Data.end(),
575 first2: Dict[i].begin(), last2: Dict[i].end());
576 }
577
578 // Get coverage for testcase with masked occurrences of dictionary unit.
579 F->ExecuteCallback(Data: Data.data(), Size: Data.size());
580 ModifiedFeatures.clear();
581 TPC.CollectFeatures(HandleFeature: [&](size_t Feature) {
582 ModifiedFeatures.push_back(x: Feature);
583 });
584
585 if (InitialFeatures == ModifiedFeatures)
586 --Scores[i];
587 else
588 Scores[i] += 2;
589 }
590 }
591
592 Printf(Fmt: "###### Useless dictionary elements. ######\n");
593 for (size_t i = 0; i < Dict.size(); ++i) {
594 // Dictionary units with positive score are treated as useful ones.
595 if (Scores[i] > 0)
596 continue;
597
598 Printf(Fmt: "\"");
599 PrintASCII(Data: Dict[i].data(), Size: Dict[i].size(), PrintAfter: "\"");
600 Printf(Fmt: " # Score: %d, Used: %d\n", Scores[i], Usages[i]);
601 }
602 Printf(Fmt: "###### End of useless dictionary elements. ######\n");
603 return 0;
604}
605
606std::vector<std::string> ParseSeedInuts(const char *seed_inputs) {
607 // Parse -seed_inputs=file1,file2,... or -seed_inputs=@seed_inputs_file
608 std::vector<std::string> Files;
609 if (!seed_inputs) return Files;
610 std::string SeedInputs;
611 if (Flags.seed_inputs[0] == '@')
612 SeedInputs = FileToString(Path: Flags.seed_inputs + 1); // File contains list.
613 else
614 SeedInputs = Flags.seed_inputs; // seed_inputs contains the list.
615 if (SeedInputs.empty()) {
616 Printf(Fmt: "seed_inputs is empty or @file does not exist.\n");
617 exit(status: 1);
618 }
619 // Parse SeedInputs.
620 size_t comma_pos = 0;
621 while ((comma_pos = SeedInputs.find_last_of(c: ',')) != std::string::npos) {
622 Files.push_back(x: SeedInputs.substr(pos: comma_pos + 1));
623 SeedInputs = SeedInputs.substr(pos: 0, n: comma_pos);
624 }
625 Files.push_back(x: SeedInputs);
626 return Files;
627}
628
629static std::vector<SizedFile>
630ReadCorpora(const std::vector<std::string> &CorpusDirs,
631 const std::vector<std::string> &ExtraSeedFiles) {
632 std::vector<SizedFile> SizedFiles;
633 size_t LastNumFiles = 0;
634 for (auto &Dir : CorpusDirs) {
635 GetSizedFilesFromDir(Dir, V: &SizedFiles);
636 Printf(Fmt: "INFO: % 8zd files found in %s\n", SizedFiles.size() - LastNumFiles,
637 Dir.c_str());
638 LastNumFiles = SizedFiles.size();
639 }
640 for (auto &File : ExtraSeedFiles)
641 if (auto Size = FileSize(Path: File))
642 SizedFiles.push_back(x: {.File: File, .Size: Size});
643 return SizedFiles;
644}
645
646int FuzzerDriver(int *argc, char ***argv, UserCallback Callback) {
647 using namespace fuzzer;
648 assert(argc && argv && "Argument pointers cannot be nullptr");
649 std::string Argv0((*argv)[0]);
650 EF = new ExternalFunctions();
651 if (EF->LLVMFuzzerInitialize)
652 EF->LLVMFuzzerInitialize(argc, argv);
653 if (EF->__msan_scoped_disable_interceptor_checks)
654 EF->__msan_scoped_disable_interceptor_checks();
655 const std::vector<std::string> Args(*argv, *argv + *argc);
656 assert(!Args.empty());
657 ProgName = new std::string(Args[0]);
658 if (Argv0 != *ProgName) {
659 Printf(Fmt: "ERROR: argv[0] has been modified in LLVMFuzzerInitialize\n");
660 exit(status: 1);
661 }
662 ParseFlags(Args, EF);
663 if (Flags.help) {
664 PrintHelp();
665 return 0;
666 }
667
668 if (Flags.close_fd_mask & 2)
669 DupAndCloseStderr();
670 if (Flags.close_fd_mask & 1)
671 CloseStdout();
672
673 if (Flags.jobs > 0 && Flags.workers == 0) {
674 Flags.workers = std::min(a: NumberOfCpuCores() / 2, b: Flags.jobs);
675 if (Flags.workers > 1)
676 Printf(Fmt: "Running %u workers\n", Flags.workers);
677 }
678
679 if (Flags.workers > 0 && Flags.jobs > 0)
680 return RunInMultipleProcesses(Args, NumWorkers: Flags.workers, NumJobs: Flags.jobs);
681
682 FuzzingOptions Options;
683 Options.Verbosity = Flags.verbosity;
684 Options.MaxLen = Flags.max_len;
685 Options.LenControl = Flags.len_control;
686 Options.KeepSeed = Flags.keep_seed;
687 Options.UnitTimeoutSec = Flags.timeout;
688 Options.ErrorExitCode = Flags.error_exitcode;
689 Options.TimeoutExitCode = Flags.timeout_exitcode;
690 Options.IgnoreTimeouts = Flags.ignore_timeouts;
691 Options.IgnoreOOMs = Flags.ignore_ooms;
692 Options.IgnoreCrashes = Flags.ignore_crashes;
693 Options.MaxTotalTimeSec = Flags.max_total_time;
694 Options.DoCrossOver = Flags.cross_over;
695 Options.CrossOverUniformDist = Flags.cross_over_uniform_dist;
696 Options.MutateDepth = Flags.mutate_depth;
697 Options.ReduceDepth = Flags.reduce_depth;
698 Options.UseCounters = Flags.use_counters;
699 Options.UseMemmem = Flags.use_memmem;
700 Options.UseCmp = Flags.use_cmp;
701 Options.UseValueProfile = Flags.use_value_profile;
702 Options.Shrink = Flags.shrink;
703 Options.ReduceInputs = Flags.reduce_inputs;
704 Options.ShuffleAtStartUp = Flags.shuffle;
705 Options.PreferSmall = Flags.prefer_small;
706 Options.ReloadIntervalSec = Flags.reload;
707 Options.OnlyASCII = Flags.only_ascii;
708 Options.DetectLeaks = Flags.detect_leaks;
709 Options.PurgeAllocatorIntervalSec = Flags.purge_allocator_interval;
710 Options.TraceMalloc = Flags.trace_malloc;
711 Options.RssLimitMb = Flags.rss_limit_mb;
712 Options.MallocLimitMb = Flags.malloc_limit_mb;
713 if (!Options.MallocLimitMb)
714 Options.MallocLimitMb = Options.RssLimitMb;
715 if (Flags.runs >= 0)
716 Options.MaxNumberOfRuns = Flags.runs;
717 if (!Inputs->empty() && !Flags.minimize_crash_internal_step) {
718 // Ensure output corpus assumed to be the first arbitrary argument input
719 // is not a path to an existing file.
720 std::string OutputCorpusDir = (*Inputs)[0];
721 if (!IsFile(Path: OutputCorpusDir)) {
722 Options.OutputCorpus = OutputCorpusDir;
723 ValidateDirectoryExists(Path: Options.OutputCorpus, CreateDirectory: Flags.create_missing_dirs);
724 }
725 }
726 Options.ReportSlowUnits = Flags.report_slow_units;
727 if (Flags.artifact_prefix) {
728 Options.ArtifactPrefix = Flags.artifact_prefix;
729
730 // Since the prefix could be a full path to a file name prefix, assume
731 // that if the path ends with the platform's separator that a directory
732 // is desired
733 std::string ArtifactPathDir = Options.ArtifactPrefix;
734 if (!IsSeparator(C: ArtifactPathDir[ArtifactPathDir.length() - 1])) {
735 ArtifactPathDir = DirName(FileName: ArtifactPathDir);
736 }
737 ValidateDirectoryExists(Path: ArtifactPathDir, CreateDirectory: Flags.create_missing_dirs);
738 }
739 if (Flags.exact_artifact_path) {
740 Options.ExactArtifactPath = Flags.exact_artifact_path;
741 ValidateDirectoryExists(Path: DirName(FileName: Options.ExactArtifactPath),
742 CreateDirectory: Flags.create_missing_dirs);
743 }
744 std::vector<Unit> Dictionary;
745 if (Flags.dict)
746 if (!ParseDictionaryFile(Text: FileToString(Path: Flags.dict), Units: &Dictionary))
747 return 1;
748 if (Flags.verbosity > 0 && !Dictionary.empty())
749 Printf(Fmt: "Dictionary: %zd entries\n", Dictionary.size());
750 bool RunIndividualFiles = AllInputsAreFiles();
751 Options.SaveArtifacts =
752 !RunIndividualFiles || Flags.minimize_crash_internal_step;
753 Options.PrintNewCovPcs = Flags.print_pcs;
754 Options.PrintNewCovFuncs = Flags.print_funcs;
755 Options.PrintFinalStats = Flags.print_final_stats;
756 Options.PrintCorpusStats = Flags.print_corpus_stats;
757 Options.PrintCoverage = Flags.print_coverage;
758 Options.PrintFullCoverage = Flags.print_full_coverage;
759 if (Flags.exit_on_src_pos)
760 Options.ExitOnSrcPos = Flags.exit_on_src_pos;
761 if (Flags.exit_on_item)
762 Options.ExitOnItem = Flags.exit_on_item;
763 if (Flags.focus_function)
764 Options.FocusFunction = Flags.focus_function;
765 if (Flags.data_flow_trace)
766 Options.DataFlowTrace = Flags.data_flow_trace;
767 if (Flags.features_dir) {
768 Options.FeaturesDir = Flags.features_dir;
769 ValidateDirectoryExists(Path: Options.FeaturesDir, CreateDirectory: Flags.create_missing_dirs);
770 }
771 if (Flags.mutation_graph_file)
772 Options.MutationGraphFile = Flags.mutation_graph_file;
773 if (Flags.collect_data_flow)
774 Options.CollectDataFlow = Flags.collect_data_flow;
775 if (Flags.stop_file)
776 Options.StopFile = Flags.stop_file;
777 Options.Entropic = Flags.entropic;
778 Options.EntropicFeatureFrequencyThreshold =
779 (size_t)Flags.entropic_feature_frequency_threshold;
780 Options.EntropicNumberOfRarestFeatures =
781 (size_t)Flags.entropic_number_of_rarest_features;
782 Options.EntropicScalePerExecTime = Flags.entropic_scale_per_exec_time;
783 if (!Options.FocusFunction.empty())
784 Options.Entropic = false; // FocusFunction overrides entropic scheduling.
785 if (Options.Entropic)
786 Printf(Fmt: "INFO: Running with entropic power schedule (0x%zX, %zu).\n",
787 Options.EntropicFeatureFrequencyThreshold,
788 Options.EntropicNumberOfRarestFeatures);
789 struct EntropicOptions Entropic;
790 Entropic.Enabled = Options.Entropic;
791 Entropic.FeatureFrequencyThreshold =
792 Options.EntropicFeatureFrequencyThreshold;
793 Entropic.NumberOfRarestFeatures = Options.EntropicNumberOfRarestFeatures;
794 Entropic.ScalePerExecTime = Options.EntropicScalePerExecTime;
795
796 unsigned Seed = Flags.seed;
797 // Initialize Seed.
798 if (Seed == 0)
799 Seed = static_cast<unsigned>(
800 std::chrono::system_clock::now().time_since_epoch().count() + GetPid());
801 if (Flags.verbosity)
802 Printf(Fmt: "INFO: Seed: %u\n", Seed);
803
804 if (Flags.collect_data_flow && Flags.data_flow_trace && !Flags.fork &&
805 !(Flags.merge || Flags.set_cover_merge)) {
806 if (RunIndividualFiles)
807 return CollectDataFlow(DFTBinary: Flags.collect_data_flow, DirPath: Flags.data_flow_trace,
808 CorporaFiles: ReadCorpora(CorpusDirs: {}, ExtraSeedFiles: *Inputs));
809 else
810 return CollectDataFlow(DFTBinary: Flags.collect_data_flow, DirPath: Flags.data_flow_trace,
811 CorporaFiles: ReadCorpora(CorpusDirs: *Inputs, ExtraSeedFiles: {}));
812 }
813
814 Random Rand(Seed);
815 auto *MD = new MutationDispatcher(Rand, Options);
816 auto *Corpus = new InputCorpus(Options.OutputCorpus, Entropic);
817 auto *F = new Fuzzer(Callback, *Corpus, *MD, Options);
818
819 for (auto &U: Dictionary)
820 if (U.size() <= Word::GetMaxSize())
821 MD->AddWordToManualDictionary(W: Word(U.data(), U.size()));
822
823 // Threads are only supported by Chrome. Don't use them with emscripten
824 // for now.
825#if !LIBFUZZER_EMSCRIPTEN
826 StartRssThread(F, RssLimitMb: Flags.rss_limit_mb);
827#endif // LIBFUZZER_EMSCRIPTEN
828
829 Options.HandleAbrt = Flags.handle_abrt;
830 Options.HandleAlrm = !Flags.minimize_crash;
831 Options.HandleBus = Flags.handle_bus;
832 Options.HandleFpe = Flags.handle_fpe;
833 Options.HandleIll = Flags.handle_ill;
834 Options.HandleInt = Flags.handle_int;
835 Options.HandleSegv = Flags.handle_segv;
836 Options.HandleTerm = Flags.handle_term;
837 Options.HandleXfsz = Flags.handle_xfsz;
838 Options.HandleUsr1 = Flags.handle_usr1;
839 Options.HandleUsr2 = Flags.handle_usr2;
840 Options.HandleWinExcept = Flags.handle_winexcept;
841
842 SetSignalHandler(Options);
843
844 std::atexit(func: Fuzzer::StaticExitCallback);
845
846 if (Flags.minimize_crash)
847 return MinimizeCrashInput(Args, Options);
848
849 if (Flags.minimize_crash_internal_step)
850 return MinimizeCrashInputInternalStep(F, Corpus);
851
852 if (Flags.cleanse_crash)
853 return CleanseCrashInput(Args, Options);
854
855 if (RunIndividualFiles) {
856 Options.SaveArtifacts = false;
857 int Runs = std::max(a: 1, b: Flags.runs);
858 Printf(Fmt: "%s: Running %zd inputs %d time(s) each.\n", ProgName->c_str(),
859 Inputs->size(), Runs);
860 for (auto &Path : *Inputs) {
861 auto StartTime = system_clock::now();
862 Printf(Fmt: "Running: %s\n", Path.c_str());
863 for (int Iter = 0; Iter < Runs; Iter++)
864 RunOneTest(F, InputFilePath: Path.c_str(), MaxLen: Options.MaxLen);
865 auto StopTime = system_clock::now();
866 auto MS = duration_cast<milliseconds>(d: StopTime - StartTime).count();
867 Printf(Fmt: "Executed %s in %ld ms\n", Path.c_str(), (long)MS);
868 }
869 Printf(Fmt: "***\n"
870 "*** NOTE: fuzzing was not performed, you have only\n"
871 "*** executed the target code on a fixed set of inputs.\n"
872 "***\n");
873 F->PrintFinalStats();
874 exit(status: 0);
875 }
876
877 Options.ForkCorpusGroups = Flags.fork_corpus_groups;
878 if (Flags.fork)
879 FuzzWithFork(Rand&: F->GetMD().GetRand(), Options, Args, CorpusDirs: *Inputs, NumJobs: Flags.fork);
880
881 if (Flags.merge || Flags.set_cover_merge)
882 Merge(F, Options, Args, Corpora: *Inputs, CFPathOrNull: Flags.merge_control_file);
883
884 if (Flags.merge_inner) {
885 const size_t kDefaultMaxMergeLen = 1 << 20;
886 if (Options.MaxLen == 0)
887 F->SetMaxInputLen(kDefaultMaxMergeLen);
888 assert(Flags.merge_control_file);
889 F->CrashResistantMergeInternalStep(ControlFilePath: Flags.merge_control_file,
890 IsSetCoverMerge: !strncmp(s1: Flags.merge_inner, s2: "2", n: 1));
891 exit(status: 0);
892 }
893
894 if (Flags.analyze_dict) {
895 size_t MaxLen = INT_MAX; // Large max length.
896 UnitVector InitialCorpus;
897 for (auto &Inp : *Inputs) {
898 Printf(Fmt: "Loading corpus dir: %s\n", Inp.c_str());
899 ReadDirToVectorOfUnits(Path: Inp.c_str(), V: &InitialCorpus, Epoch: nullptr,
900 MaxSize: MaxLen, /*ExitOnError=*/false);
901 }
902
903 if (Dictionary.empty() || Inputs->empty()) {
904 Printf(Fmt: "ERROR: can't analyze dict without dict and corpus provided\n");
905 return 1;
906 }
907 if (AnalyzeDictionary(F, Dict: Dictionary, Corpus&: InitialCorpus)) {
908 Printf(Fmt: "Dictionary analysis failed\n");
909 exit(status: 1);
910 }
911 Printf(Fmt: "Dictionary analysis succeeded\n");
912 exit(status: 0);
913 }
914
915 auto CorporaFiles = ReadCorpora(CorpusDirs: *Inputs, ExtraSeedFiles: ParseSeedInuts(seed_inputs: Flags.seed_inputs));
916 F->Loop(CorporaFiles);
917
918 if (Flags.verbosity)
919 Printf(Fmt: "Done %zd runs in %zd second(s)\n", F->getTotalNumberOfRuns(),
920 F->secondsSinceProcessStartUp());
921 F->PrintFinalStats();
922
923 exit(status: 0); // Don't let F destroy itself.
924}
925
926extern "C" ATTRIBUTE_INTERFACE int
927LLVMFuzzerRunDriver(int *argc, char ***argv,
928 int (*UserCb)(const uint8_t *Data, size_t Size)) {
929 return FuzzerDriver(argc, argv, Callback: UserCb);
930}
931
932// Storage for global ExternalFunctions object.
933ExternalFunctions *EF = nullptr;
934
935} // namespace fuzzer
936

source code of compiler-rt/lib/fuzzer/FuzzerDriver.cpp