1 //===- FuzzerDataFlowTrace.cpp - DataFlowTrace                ---*- C++ -* ===//
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 // fuzzer::DataFlowTrace
9 //===----------------------------------------------------------------------===//
10 
11 #include "FuzzerDataFlowTrace.h"
12 
13 #include "FuzzerCommand.h"
14 #include "FuzzerIO.h"
15 #include "FuzzerRandom.h"
16 #include "FuzzerSHA1.h"
17 #include "FuzzerUtil.h"
18 
19 #include <cstdlib>
20 #include <fstream>
21 #include <numeric>
22 #include <queue>
23 #include <sstream>
24 #include <string>
25 #include <unordered_map>
26 #include <unordered_set>
27 #include <vector>
28 
29 namespace fuzzer {
30 static const char *kFunctionsTxt = "functions.txt";
31 
AppendCoverage(const std::string & S)32 bool BlockCoverage::AppendCoverage(const std::string &S) {
33   std::stringstream SS(S);
34   return AppendCoverage(SS);
35 }
36 
37 // Coverage lines have this form:
38 // CN X Y Z T
39 // where N is the number of the function, T is the total number of instrumented
40 // BBs, and X,Y,Z, if present, are the indecies of covered BB.
41 // BB #0, which is the entry block, is not explicitly listed.
AppendCoverage(std::istream & IN)42 bool BlockCoverage::AppendCoverage(std::istream &IN) {
43   std::string L;
44   while (std::getline(IN, L, '\n')) {
45     if (L.empty())
46       continue;
47     std::stringstream SS(L.c_str() + 1);
48     size_t FunctionId  = 0;
49     SS >> FunctionId;
50     if (L[0] == 'F') {
51       FunctionsWithDFT.insert(FunctionId);
52       continue;
53     }
54     if (L[0] != 'C') continue;
55     Vector<uint32_t> CoveredBlocks;
56     while (true) {
57       uint32_t BB = 0;
58       SS >> BB;
59       if (!SS) break;
60       CoveredBlocks.push_back(BB);
61     }
62     if (CoveredBlocks.empty()) return false;
63     // Ensures no CoverageVector is longer than UINT32_MAX.
64     uint32_t NumBlocks = CoveredBlocks.back();
65     CoveredBlocks.pop_back();
66     for (auto BB : CoveredBlocks)
67       if (BB >= NumBlocks) return false;
68     auto It = Functions.find(FunctionId);
69     auto &Counters =
70         It == Functions.end()
71             ? Functions.insert({FunctionId, Vector<uint32_t>(NumBlocks)})
72                   .first->second
73             : It->second;
74 
75     if (Counters.size() != NumBlocks) return false;  // wrong number of blocks.
76 
77     Counters[0]++;
78     for (auto BB : CoveredBlocks)
79       Counters[BB]++;
80   }
81   return true;
82 }
83 
84 // Assign weights to each function.
85 // General principles:
86 //   * any uncovered function gets weight 0.
87 //   * a function with lots of uncovered blocks gets bigger weight.
88 //   * a function with a less frequently executed code gets bigger weight.
FunctionWeights(size_t NumFunctions) const89 Vector<double> BlockCoverage::FunctionWeights(size_t NumFunctions) const {
90   Vector<double> Res(NumFunctions);
91   for (auto It : Functions) {
92     auto FunctionID = It.first;
93     auto Counters = It.second;
94     assert(FunctionID < NumFunctions);
95     auto &Weight = Res[FunctionID];
96     // Give higher weight if the function has a DFT.
97     Weight = FunctionsWithDFT.count(FunctionID) ? 1000. : 1;
98     // Give higher weight to functions with less frequently seen basic blocks.
99     Weight /= SmallestNonZeroCounter(Counters);
100     // Give higher weight to functions with the most uncovered basic blocks.
101     Weight *= NumberOfUncoveredBlocks(Counters) + 1;
102   }
103   return Res;
104 }
105 
ReadCoverage(const std::string & DirPath)106 void DataFlowTrace::ReadCoverage(const std::string &DirPath) {
107   Vector<SizedFile> Files;
108   GetSizedFilesFromDir(DirPath, &Files);
109   for (auto &SF : Files) {
110     auto Name = Basename(SF.File);
111     if (Name == kFunctionsTxt) continue;
112     if (!CorporaHashes.count(Name)) continue;
113     std::ifstream IF(SF.File);
114     Coverage.AppendCoverage(IF);
115   }
116 }
117 
DFTStringAppendToVector(Vector<uint8_t> * DFT,const std::string & DFTString)118 static void DFTStringAppendToVector(Vector<uint8_t> *DFT,
119                                     const std::string &DFTString) {
120   assert(DFT->size() == DFTString.size());
121   for (size_t I = 0, Len = DFT->size(); I < Len; I++)
122     (*DFT)[I] = DFTString[I] == '1';
123 }
124 
125 // converts a string of '0' and '1' into a Vector<uint8_t>
DFTStringToVector(const std::string & DFTString)126 static Vector<uint8_t> DFTStringToVector(const std::string &DFTString) {
127   Vector<uint8_t> DFT(DFTString.size());
128   DFTStringAppendToVector(&DFT, DFTString);
129   return DFT;
130 }
131 
ParseError(const char * Err,const std::string & Line)132 static bool ParseError(const char *Err, const std::string &Line) {
133   Printf("DataFlowTrace: parse error: %s: Line: %s\n", Err, Line.c_str());
134   return false;
135 }
136 
137 // TODO(metzman): replace std::string with std::string_view for
138 // better performance. Need to figure our how to use string_view on Windows.
ParseDFTLine(const std::string & Line,size_t * FunctionNum,std::string * DFTString)139 static bool ParseDFTLine(const std::string &Line, size_t *FunctionNum,
140                          std::string *DFTString) {
141   if (!Line.empty() && Line[0] != 'F')
142     return false; // Ignore coverage.
143   size_t SpacePos = Line.find(' ');
144   if (SpacePos == std::string::npos)
145     return ParseError("no space in the trace line", Line);
146   if (Line.empty() || Line[0] != 'F')
147     return ParseError("the trace line doesn't start with 'F'", Line);
148   *FunctionNum = std::atol(Line.c_str() + 1);
149   const char *Beg = Line.c_str() + SpacePos + 1;
150   const char *End = Line.c_str() + Line.size();
151   assert(Beg < End);
152   size_t Len = End - Beg;
153   for (size_t I = 0; I < Len; I++) {
154     if (Beg[I] != '0' && Beg[I] != '1')
155       return ParseError("the trace should contain only 0 or 1", Line);
156   }
157   *DFTString = Beg;
158   return true;
159 }
160 
Init(const std::string & DirPath,std::string * FocusFunction,Vector<SizedFile> & CorporaFiles,Random & Rand)161 bool DataFlowTrace::Init(const std::string &DirPath, std::string *FocusFunction,
162                          Vector<SizedFile> &CorporaFiles, Random &Rand) {
163   if (DirPath.empty()) return false;
164   Printf("INFO: DataFlowTrace: reading from '%s'\n", DirPath.c_str());
165   Vector<SizedFile> Files;
166   GetSizedFilesFromDir(DirPath, &Files);
167   std::string L;
168   size_t FocusFuncIdx = SIZE_MAX;
169   Vector<std::string> FunctionNames;
170 
171   // Collect the hashes of the corpus files.
172   for (auto &SF : CorporaFiles)
173     CorporaHashes.insert(Hash(FileToVector(SF.File)));
174 
175   // Read functions.txt
176   std::ifstream IF(DirPlusFile(DirPath, kFunctionsTxt));
177   size_t NumFunctions = 0;
178   while (std::getline(IF, L, '\n')) {
179     FunctionNames.push_back(L);
180     NumFunctions++;
181     if (*FocusFunction == L)
182       FocusFuncIdx = NumFunctions - 1;
183   }
184   if (!NumFunctions)
185     return false;
186 
187   if (*FocusFunction == "auto") {
188     // AUTOFOCUS works like this:
189     // * reads the coverage data from the DFT files.
190     // * assigns weights to functions based on coverage.
191     // * chooses a random function according to the weights.
192     ReadCoverage(DirPath);
193     auto Weights = Coverage.FunctionWeights(NumFunctions);
194     Vector<double> Intervals(NumFunctions + 1);
195     std::iota(Intervals.begin(), Intervals.end(), 0);
196     auto Distribution = std::piecewise_constant_distribution<double>(
197         Intervals.begin(), Intervals.end(), Weights.begin());
198     FocusFuncIdx = static_cast<size_t>(Distribution(Rand));
199     *FocusFunction = FunctionNames[FocusFuncIdx];
200     assert(FocusFuncIdx < NumFunctions);
201     Printf("INFO: AUTOFOCUS: %zd %s\n", FocusFuncIdx,
202            FunctionNames[FocusFuncIdx].c_str());
203     for (size_t i = 0; i < NumFunctions; i++) {
204       if (Weights[i] == 0.0)
205         continue;
206       Printf("  [%zd] W %g\tBB-tot %u\tBB-cov %u\tEntryFreq %u:\t%s\n", i,
207              Weights[i], Coverage.GetNumberOfBlocks(i),
208              Coverage.GetNumberOfCoveredBlocks(i), Coverage.GetCounter(i, 0),
209              FunctionNames[i].c_str());
210     }
211   }
212 
213   if (!NumFunctions || FocusFuncIdx == SIZE_MAX || Files.size() <= 1)
214     return false;
215 
216   // Read traces.
217   size_t NumTraceFiles = 0;
218   size_t NumTracesWithFocusFunction = 0;
219   for (auto &SF : Files) {
220     auto Name = Basename(SF.File);
221     if (Name == kFunctionsTxt) continue;
222     if (!CorporaHashes.count(Name)) continue;  // not in the corpus.
223     NumTraceFiles++;
224     // Printf("=== %s\n", Name.c_str());
225     std::ifstream IF(SF.File);
226     while (std::getline(IF, L, '\n')) {
227       size_t FunctionNum = 0;
228       std::string DFTString;
229       if (ParseDFTLine(L, &FunctionNum, &DFTString) &&
230           FunctionNum == FocusFuncIdx) {
231         NumTracesWithFocusFunction++;
232 
233         if (FunctionNum >= NumFunctions)
234           return ParseError("N is greater than the number of functions", L);
235         Traces[Name] = DFTStringToVector(DFTString);
236         // Print just a few small traces.
237         if (NumTracesWithFocusFunction <= 3 && DFTString.size() <= 16)
238           Printf("%s => |%s|\n", Name.c_str(), std::string(DFTString).c_str());
239         break; // No need to parse the following lines.
240       }
241     }
242   }
243   Printf("INFO: DataFlowTrace: %zd trace files, %zd functions, "
244          "%zd traces with focus function\n",
245          NumTraceFiles, NumFunctions, NumTracesWithFocusFunction);
246   return NumTraceFiles > 0;
247 }
248 
CollectDataFlow(const std::string & DFTBinary,const std::string & DirPath,const Vector<SizedFile> & CorporaFiles)249 int CollectDataFlow(const std::string &DFTBinary, const std::string &DirPath,
250                     const Vector<SizedFile> &CorporaFiles) {
251   Printf("INFO: collecting data flow: bin: %s dir: %s files: %zd\n",
252          DFTBinary.c_str(), DirPath.c_str(), CorporaFiles.size());
253   if (CorporaFiles.empty()) {
254     Printf("ERROR: can't collect data flow without corpus provided.");
255     return 1;
256   }
257 
258   static char DFSanEnv[] = "DFSAN_OPTIONS=warn_unimplemented=0";
259   putenv(DFSanEnv);
260   MkDir(DirPath);
261   for (auto &F : CorporaFiles) {
262     // For every input F we need to collect the data flow and the coverage.
263     // Data flow collection may fail if we request too many DFSan tags at once.
264     // So, we start from requesting all tags in range [0,Size) and if that fails
265     // we then request tags in [0,Size/2) and [Size/2, Size), and so on.
266     // Function number => DFT.
267     auto OutPath = DirPlusFile(DirPath, Hash(FileToVector(F.File)));
268     std::unordered_map<size_t, Vector<uint8_t>> DFTMap;
269     std::unordered_set<std::string> Cov;
270     Command Cmd;
271     Cmd.addArgument(DFTBinary);
272     Cmd.addArgument(F.File);
273     Cmd.addArgument(OutPath);
274     Printf("CMD: %s\n", Cmd.toString().c_str());
275     ExecuteCommand(Cmd);
276   }
277   // Write functions.txt if it's currently empty or doesn't exist.
278   auto FunctionsTxtPath = DirPlusFile(DirPath, kFunctionsTxt);
279   if (FileToString(FunctionsTxtPath).empty()) {
280     Command Cmd;
281     Cmd.addArgument(DFTBinary);
282     Cmd.setOutputFile(FunctionsTxtPath);
283     ExecuteCommand(Cmd);
284   }
285   return 0;
286 }
287 
288 }  // namespace fuzzer
289