1 //===- FuzzerMerge.cpp - merging corpora ----------------------------------===//
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 // Merging corpora.
9 //===----------------------------------------------------------------------===//
10 
11 #include "FuzzerCommand.h"
12 #include "FuzzerMerge.h"
13 #include "FuzzerIO.h"
14 #include "FuzzerInternal.h"
15 #include "FuzzerTracePC.h"
16 #include "FuzzerUtil.h"
17 
18 #include <fstream>
19 #include <iterator>
20 #include <set>
21 #include <sstream>
22 
23 namespace fuzzer {
24 
25 bool Merger::Parse(const std::string &Str, bool ParseCoverage) {
26   std::istringstream SS(Str);
27   return Parse(SS, ParseCoverage);
28 }
29 
30 void Merger::ParseOrExit(std::istream &IS, bool ParseCoverage) {
31   if (!Parse(IS, ParseCoverage)) {
32     Printf("MERGE: failed to parse the control file (unexpected error)\n");
33     exit(1);
34   }
35 }
36 
37 // The control file example:
38 //
39 // 3 # The number of inputs
40 // 1 # The number of inputs in the first corpus, <= the previous number
41 // file0
42 // file1
43 // file2  # One file name per line.
44 // STARTED 0 123  # FileID, file size
45 // FT 0 1 4 6 8  # FileID COV1 COV2 ...
46 // COV 0 7 8 9 # FileID COV1 COV1
47 // STARTED 1 456  # If FT is missing, the input crashed while processing.
48 // STARTED 2 567
49 // FT 2 8 9
50 // COV 2 11 12
51 bool Merger::Parse(std::istream &IS, bool ParseCoverage) {
52   LastFailure.clear();
53   std::string Line;
54 
55   // Parse NumFiles.
56   if (!std::getline(IS, Line, '\n')) return false;
57   std::istringstream L1(Line);
58   size_t NumFiles = 0;
59   L1 >> NumFiles;
60   if (NumFiles == 0 || NumFiles > 10000000) return false;
61 
62   // Parse NumFilesInFirstCorpus.
63   if (!std::getline(IS, Line, '\n')) return false;
64   std::istringstream L2(Line);
65   NumFilesInFirstCorpus = NumFiles + 1;
66   L2 >> NumFilesInFirstCorpus;
67   if (NumFilesInFirstCorpus > NumFiles) return false;
68 
69   // Parse file names.
70   Files.resize(NumFiles);
71   for (size_t i = 0; i < NumFiles; i++)
72     if (!std::getline(IS, Files[i].Name, '\n'))
73       return false;
74 
75   // Parse STARTED, FT, and COV lines.
76   size_t ExpectedStartMarker = 0;
77   const size_t kInvalidStartMarker = -1;
78   size_t LastSeenStartMarker = kInvalidStartMarker;
79   Vector<uint32_t> TmpFeatures;
80   Set<uint32_t> PCs;
81   while (std::getline(IS, Line, '\n')) {
82     std::istringstream ISS1(Line);
83     std::string Marker;
84     size_t N;
85     ISS1 >> Marker;
86     ISS1 >> N;
87     if (Marker == "STARTED") {
88       // STARTED FILE_ID FILE_SIZE
89       if (ExpectedStartMarker != N)
90         return false;
91       ISS1 >> Files[ExpectedStartMarker].Size;
92       LastSeenStartMarker = ExpectedStartMarker;
93       assert(ExpectedStartMarker < Files.size());
94       ExpectedStartMarker++;
95     } else if (Marker == "FT") {
96       // FT FILE_ID COV1 COV2 COV3 ...
97       size_t CurrentFileIdx = N;
98       if (CurrentFileIdx != LastSeenStartMarker)
99         return false;
100       LastSeenStartMarker = kInvalidStartMarker;
101       if (ParseCoverage) {
102         TmpFeatures.clear();  // use a vector from outer scope to avoid resizes.
103         while (ISS1 >> N)
104           TmpFeatures.push_back(N);
105         std::sort(TmpFeatures.begin(), TmpFeatures.end());
106         Files[CurrentFileIdx].Features = TmpFeatures;
107       }
108     } else if (Marker == "COV") {
109       size_t CurrentFileIdx = N;
110       if (ParseCoverage)
111         while (ISS1 >> N)
112           if (PCs.insert(N).second)
113             Files[CurrentFileIdx].Cov.push_back(N);
114     } else {
115       return false;
116     }
117   }
118   if (LastSeenStartMarker != kInvalidStartMarker)
119     LastFailure = Files[LastSeenStartMarker].Name;
120 
121   FirstNotProcessedFile = ExpectedStartMarker;
122   return true;
123 }
124 
125 size_t Merger::ApproximateMemoryConsumption() const  {
126   size_t Res = 0;
127   for (const auto &F: Files)
128     Res += sizeof(F) + F.Features.size() * sizeof(F.Features[0]);
129   return Res;
130 }
131 
132 // Decides which files need to be merged (add those to NewFiles).
133 // Returns the number of new features added.
134 size_t Merger::Merge(const Set<uint32_t> &InitialFeatures,
135                      Set<uint32_t> *NewFeatures,
136                      const Set<uint32_t> &InitialCov, Set<uint32_t> *NewCov,
137                      Vector<std::string> *NewFiles) {
138   NewFiles->clear();
139   assert(NumFilesInFirstCorpus <= Files.size());
140   Set<uint32_t> AllFeatures = InitialFeatures;
141 
142   // What features are in the initial corpus?
143   for (size_t i = 0; i < NumFilesInFirstCorpus; i++) {
144     auto &Cur = Files[i].Features;
145     AllFeatures.insert(Cur.begin(), Cur.end());
146   }
147   // Remove all features that we already know from all other inputs.
148   for (size_t i = NumFilesInFirstCorpus; i < Files.size(); i++) {
149     auto &Cur = Files[i].Features;
150     Vector<uint32_t> Tmp;
151     std::set_difference(Cur.begin(), Cur.end(), AllFeatures.begin(),
152                         AllFeatures.end(), std::inserter(Tmp, Tmp.begin()));
153     Cur.swap(Tmp);
154   }
155 
156   // Sort. Give preference to
157   //   * smaller files
158   //   * files with more features.
159   std::sort(Files.begin() + NumFilesInFirstCorpus, Files.end(),
160             [&](const MergeFileInfo &a, const MergeFileInfo &b) -> bool {
161               if (a.Size != b.Size)
162                 return a.Size < b.Size;
163               return a.Features.size() > b.Features.size();
164             });
165 
166   // One greedy pass: add the file's features to AllFeatures.
167   // If new features were added, add this file to NewFiles.
168   for (size_t i = NumFilesInFirstCorpus; i < Files.size(); i++) {
169     auto &Cur = Files[i].Features;
170     // Printf("%s -> sz %zd ft %zd\n", Files[i].Name.c_str(),
171     //       Files[i].Size, Cur.size());
172     bool FoundNewFeatures = false;
173     for (auto Fe: Cur) {
174       if (AllFeatures.insert(Fe).second) {
175         FoundNewFeatures = true;
176         NewFeatures->insert(Fe);
177       }
178     }
179     if (FoundNewFeatures)
180       NewFiles->push_back(Files[i].Name);
181     for (auto Cov : Files[i].Cov)
182       if (InitialCov.find(Cov) == InitialCov.end())
183         NewCov->insert(Cov);
184   }
185   return NewFeatures->size();
186 }
187 
188 Set<uint32_t> Merger::AllFeatures() const {
189   Set<uint32_t> S;
190   for (auto &File : Files)
191     S.insert(File.Features.begin(), File.Features.end());
192   return S;
193 }
194 
195 // Inner process. May crash if the target crashes.
196 void Fuzzer::CrashResistantMergeInternalStep(const std::string &CFPath) {
197   Printf("MERGE-INNER: using the control file '%s'\n", CFPath.c_str());
198   Merger M;
199   std::ifstream IF(CFPath);
200   M.ParseOrExit(IF, false);
201   IF.close();
202   if (!M.LastFailure.empty())
203     Printf("MERGE-INNER: '%s' caused a failure at the previous merge step\n",
204            M.LastFailure.c_str());
205 
206   Printf("MERGE-INNER: %zd total files;"
207          " %zd processed earlier; will process %zd files now\n",
208          M.Files.size(), M.FirstNotProcessedFile,
209          M.Files.size() - M.FirstNotProcessedFile);
210 
211   std::ofstream OF(CFPath, std::ofstream::out | std::ofstream::app);
212   Set<size_t> AllFeatures;
213   auto PrintStatsWrapper = [this, &AllFeatures](const char* Where) {
214     this->PrintStats(Where, "\n", 0, AllFeatures.size());
215   };
216   Set<const TracePC::PCTableEntry *> AllPCs;
217   for (size_t i = M.FirstNotProcessedFile; i < M.Files.size(); i++) {
218     Fuzzer::MaybeExitGracefully();
219     auto U = FileToVector(M.Files[i].Name);
220     if (U.size() > MaxInputLen) {
221       U.resize(MaxInputLen);
222       U.shrink_to_fit();
223     }
224     std::ostringstream StartedLine;
225     // Write the pre-run marker.
226     OF << "STARTED " << i << " " << U.size() << "\n";
227     OF.flush();  // Flush is important since Command::Execute may crash.
228     // Run.
229     TPC.ResetMaps();
230     ExecuteCallback(U.data(), U.size());
231     // Collect coverage. We are iterating over the files in this order:
232     // * First, files in the initial corpus ordered by size, smallest first.
233     // * Then, all other files, smallest first.
234     // So it makes no sense to record all features for all files, instead we
235     // only record features that were not seen before.
236     Set<size_t> UniqFeatures;
237     TPC.CollectFeatures([&](size_t Feature) {
238       if (AllFeatures.insert(Feature).second)
239         UniqFeatures.insert(Feature);
240     });
241     TPC.UpdateObservedPCs();
242     // Show stats.
243     if (!(TotalNumberOfRuns & (TotalNumberOfRuns - 1)))
244       PrintStatsWrapper("pulse ");
245     if (TotalNumberOfRuns == M.NumFilesInFirstCorpus)
246       PrintStatsWrapper("LOADED");
247     // Write the post-run marker and the coverage.
248     OF << "FT " << i;
249     for (size_t F : UniqFeatures)
250       OF << " " << F;
251     OF << "\n";
252     OF << "COV " << i;
253     TPC.ForEachObservedPC([&](const TracePC::PCTableEntry *TE) {
254       if (AllPCs.insert(TE).second)
255         OF << " " << TPC.PCTableEntryIdx(TE);
256     });
257     OF << "\n";
258     OF.flush();
259   }
260   PrintStatsWrapper("DONE  ");
261 }
262 
263 static void WriteNewControlFile(const std::string &CFPath,
264                                 const Vector<SizedFile> &OldCorpus,
265                                 const Vector<SizedFile> &NewCorpus) {
266   RemoveFile(CFPath);
267   std::ofstream ControlFile(CFPath);
268   ControlFile << (OldCorpus.size() + NewCorpus.size()) << "\n";
269   ControlFile << OldCorpus.size() << "\n";
270   for (auto &SF: OldCorpus)
271     ControlFile << SF.File << "\n";
272   for (auto &SF: NewCorpus)
273     ControlFile << SF.File << "\n";
274   if (!ControlFile) {
275     Printf("MERGE-OUTER: failed to write to the control file: %s\n",
276            CFPath.c_str());
277     exit(1);
278   }
279 }
280 
281 // Outer process. Does not call the target code and thus should not fail.
282 void CrashResistantMerge(const Vector<std::string> &Args,
283                          const Vector<SizedFile> &OldCorpus,
284                          const Vector<SizedFile> &NewCorpus,
285                          Vector<std::string> *NewFiles,
286                          const Set<uint32_t> &InitialFeatures,
287                          Set<uint32_t> *NewFeatures,
288                          const Set<uint32_t> &InitialCov,
289                          Set<uint32_t> *NewCov,
290                          const std::string &CFPath,
291                          bool V /*Verbose*/) {
292   if (NewCorpus.empty() && OldCorpus.empty()) return;  // Nothing to merge.
293   size_t NumAttempts = 0;
294   if (FileSize(CFPath)) {
295     VPrintf(V, "MERGE-OUTER: non-empty control file provided: '%s'\n",
296            CFPath.c_str());
297     Merger M;
298     std::ifstream IF(CFPath);
299     if (M.Parse(IF, /*ParseCoverage=*/false)) {
300       VPrintf(V, "MERGE-OUTER: control file ok, %zd files total,"
301              " first not processed file %zd\n",
302              M.Files.size(), M.FirstNotProcessedFile);
303       if (!M.LastFailure.empty())
304         VPrintf(V, "MERGE-OUTER: '%s' will be skipped as unlucky "
305                "(merge has stumbled on it the last time)\n",
306                M.LastFailure.c_str());
307       if (M.FirstNotProcessedFile >= M.Files.size()) {
308         VPrintf(
309             V, "MERGE-OUTER: nothing to do, merge has been completed before\n");
310         exit(0);
311       }
312 
313       NumAttempts = M.Files.size() - M.FirstNotProcessedFile;
314     } else {
315       VPrintf(V, "MERGE-OUTER: bad control file, will overwrite it\n");
316     }
317   }
318 
319   if (!NumAttempts) {
320     // The supplied control file is empty or bad, create a fresh one.
321     NumAttempts = OldCorpus.size() + NewCorpus.size();
322     VPrintf(V, "MERGE-OUTER: %zd files, %zd in the initial corpus\n",
323             NumAttempts, OldCorpus.size());
324     WriteNewControlFile(CFPath, OldCorpus, NewCorpus);
325   }
326 
327   // Execute the inner process until it passes.
328   // Every inner process should execute at least one input.
329   Command BaseCmd(Args);
330   BaseCmd.removeFlag("merge");
331   BaseCmd.removeFlag("fork");
332   BaseCmd.removeFlag("collect_data_flow");
333   for (size_t Attempt = 1; Attempt <= NumAttempts; Attempt++) {
334     Fuzzer::MaybeExitGracefully();
335     VPrintf(V, "MERGE-OUTER: attempt %zd\n", Attempt);
336     Command Cmd(BaseCmd);
337     Cmd.addFlag("merge_control_file", CFPath);
338     Cmd.addFlag("merge_inner", "1");
339     if (!V) {
340       Cmd.setOutputFile(getDevNull());
341       Cmd.combineOutAndErr();
342     }
343     auto ExitCode = ExecuteCommand(Cmd);
344     if (!ExitCode) {
345       VPrintf(V, "MERGE-OUTER: succesfull in %zd attempt(s)\n", Attempt);
346       break;
347     }
348   }
349   // Read the control file and do the merge.
350   Merger M;
351   std::ifstream IF(CFPath);
352   IF.seekg(0, IF.end);
353   VPrintf(V, "MERGE-OUTER: the control file has %zd bytes\n",
354           (size_t)IF.tellg());
355   IF.seekg(0, IF.beg);
356   M.ParseOrExit(IF, true);
357   IF.close();
358   VPrintf(V,
359           "MERGE-OUTER: consumed %zdMb (%zdMb rss) to parse the control file\n",
360           M.ApproximateMemoryConsumption() >> 20, GetPeakRSSMb());
361   M.Merge(InitialFeatures, NewFeatures, InitialCov, NewCov, NewFiles);
362   VPrintf(V, "MERGE-OUTER: %zd new files with %zd new features added; "
363           "%zd new coverage edges\n",
364          NewFiles->size(), NewFeatures->size(), NewCov->size());
365 }
366 
367 } // namespace fuzzer
368