[lit] Add argument check: --timeout must be non-negative integer
[llvm-core.git] / tools / sancov / sancov.cpp
blobe0cc8a7cf8494f732603da2b8831611d7222e4cd
1 //===-- sancov.cpp --------------------------------------------------------===//
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 // This file is a command-line tool for reading and analyzing sanitizer
9 // coverage.
10 //===----------------------------------------------------------------------===//
11 #include "llvm/ADT/STLExtras.h"
12 #include "llvm/ADT/StringExtras.h"
13 #include "llvm/ADT/Twine.h"
14 #include "llvm/DebugInfo/Symbolize/Symbolize.h"
15 #include "llvm/MC/MCAsmInfo.h"
16 #include "llvm/MC/MCContext.h"
17 #include "llvm/MC/MCDisassembler/MCDisassembler.h"
18 #include "llvm/MC/MCInst.h"
19 #include "llvm/MC/MCInstrAnalysis.h"
20 #include "llvm/MC/MCInstrInfo.h"
21 #include "llvm/MC/MCObjectFileInfo.h"
22 #include "llvm/MC/MCRegisterInfo.h"
23 #include "llvm/MC/MCSubtargetInfo.h"
24 #include "llvm/Object/Archive.h"
25 #include "llvm/Object/Binary.h"
26 #include "llvm/Object/COFF.h"
27 #include "llvm/Object/MachO.h"
28 #include "llvm/Object/ObjectFile.h"
29 #include "llvm/Support/Casting.h"
30 #include "llvm/Support/CommandLine.h"
31 #include "llvm/Support/Errc.h"
32 #include "llvm/Support/ErrorOr.h"
33 #include "llvm/Support/FileSystem.h"
34 #include "llvm/Support/JSON.h"
35 #include "llvm/Support/MD5.h"
36 #include "llvm/Support/ManagedStatic.h"
37 #include "llvm/Support/MemoryBuffer.h"
38 #include "llvm/Support/Path.h"
39 #include "llvm/Support/PrettyStackTrace.h"
40 #include "llvm/Support/Regex.h"
41 #include "llvm/Support/SHA1.h"
42 #include "llvm/Support/Signals.h"
43 #include "llvm/Support/SourceMgr.h"
44 #include "llvm/Support/SpecialCaseList.h"
45 #include "llvm/Support/TargetRegistry.h"
46 #include "llvm/Support/TargetSelect.h"
47 #include "llvm/Support/YAMLParser.h"
48 #include "llvm/Support/raw_ostream.h"
50 #include <set>
51 #include <vector>
53 using namespace llvm;
55 namespace {
57 // --------- COMMAND LINE FLAGS ---------
59 enum ActionType {
60 CoveredFunctionsAction,
61 HtmlReportAction,
62 MergeAction,
63 NotCoveredFunctionsAction,
64 PrintAction,
65 PrintCovPointsAction,
66 StatsAction,
67 SymbolizeAction
70 cl::opt<ActionType> Action(
71 cl::desc("Action (required)"), cl::Required,
72 cl::values(
73 clEnumValN(PrintAction, "print", "Print coverage addresses"),
74 clEnumValN(PrintCovPointsAction, "print-coverage-pcs",
75 "Print coverage instrumentation points addresses."),
76 clEnumValN(CoveredFunctionsAction, "covered-functions",
77 "Print all covered funcions."),
78 clEnumValN(NotCoveredFunctionsAction, "not-covered-functions",
79 "Print all not covered funcions."),
80 clEnumValN(StatsAction, "print-coverage-stats",
81 "Print coverage statistics."),
82 clEnumValN(HtmlReportAction, "html-report",
83 "REMOVED. Use -symbolize & coverage-report-server.py."),
84 clEnumValN(SymbolizeAction, "symbolize",
85 "Produces a symbolized JSON report from binary report."),
86 clEnumValN(MergeAction, "merge", "Merges reports.")));
88 static cl::list<std::string>
89 ClInputFiles(cl::Positional, cl::OneOrMore,
90 cl::desc("<action> <binary files...> <.sancov files...> "
91 "<.symcov files...>"));
93 static cl::opt<bool> ClDemangle("demangle", cl::init(true),
94 cl::desc("Print demangled function name."));
96 static cl::opt<bool>
97 ClSkipDeadFiles("skip-dead-files", cl::init(true),
98 cl::desc("Do not list dead source files in reports."));
100 static cl::opt<std::string> ClStripPathPrefix(
101 "strip_path_prefix", cl::init(""),
102 cl::desc("Strip this prefix from file paths in reports."));
104 static cl::opt<std::string>
105 ClBlacklist("blacklist", cl::init(""),
106 cl::desc("Blacklist file (sanitizer blacklist format)."));
108 static cl::opt<bool> ClUseDefaultBlacklist(
109 "use_default_blacklist", cl::init(true), cl::Hidden,
110 cl::desc("Controls if default blacklist should be used."));
112 static const char *const DefaultBlacklistStr = "fun:__sanitizer_.*\n"
113 "src:/usr/include/.*\n"
114 "src:.*/libc\\+\\+/.*\n";
116 // --------- FORMAT SPECIFICATION ---------
118 struct FileHeader {
119 uint32_t Bitness;
120 uint32_t Magic;
123 static const uint32_t BinCoverageMagic = 0xC0BFFFFF;
124 static const uint32_t Bitness32 = 0xFFFFFF32;
125 static const uint32_t Bitness64 = 0xFFFFFF64;
127 static const Regex SancovFileRegex("(.*)\\.[0-9]+\\.sancov");
128 static const Regex SymcovFileRegex(".*\\.symcov");
130 // --------- MAIN DATASTRUCTURES ----------
132 // Contents of .sancov file: list of coverage point addresses that were
133 // executed.
134 struct RawCoverage {
135 explicit RawCoverage(std::unique_ptr<std::set<uint64_t>> Addrs)
136 : Addrs(std::move(Addrs)) {}
138 // Read binary .sancov file.
139 static ErrorOr<std::unique_ptr<RawCoverage>>
140 read(const std::string &FileName);
142 std::unique_ptr<std::set<uint64_t>> Addrs;
145 // Coverage point has an opaque Id and corresponds to multiple source locations.
146 struct CoveragePoint {
147 explicit CoveragePoint(const std::string &Id) : Id(Id) {}
149 std::string Id;
150 SmallVector<DILineInfo, 1> Locs;
153 // Symcov file content: set of covered Ids plus information about all available
154 // coverage points.
155 struct SymbolizedCoverage {
156 // Read json .symcov file.
157 static std::unique_ptr<SymbolizedCoverage> read(const std::string &InputFile);
159 std::set<std::string> CoveredIds;
160 std::string BinaryHash;
161 std::vector<CoveragePoint> Points;
164 struct CoverageStats {
165 size_t AllPoints;
166 size_t CovPoints;
167 size_t AllFns;
168 size_t CovFns;
171 // --------- ERROR HANDLING ---------
173 static void fail(const llvm::Twine &E) {
174 errs() << "ERROR: " << E << "\n";
175 exit(1);
178 static void failIf(bool B, const llvm::Twine &E) {
179 if (B)
180 fail(E);
183 static void failIfError(std::error_code Error) {
184 if (!Error)
185 return;
186 errs() << "ERROR: " << Error.message() << "(" << Error.value() << ")\n";
187 exit(1);
190 template <typename T> static void failIfError(const ErrorOr<T> &E) {
191 failIfError(E.getError());
194 static void failIfError(Error Err) {
195 if (Err) {
196 logAllUnhandledErrors(std::move(Err), errs(), "ERROR: ");
197 exit(1);
201 template <typename T> static void failIfError(Expected<T> &E) {
202 failIfError(E.takeError());
205 static void failIfNotEmpty(const llvm::Twine &E) {
206 if (E.str().empty())
207 return;
208 fail(E);
211 template <typename T>
212 static void failIfEmpty(const std::unique_ptr<T> &Ptr,
213 const std::string &Message) {
214 if (Ptr.get())
215 return;
216 fail(Message);
219 // ----------- Coverage I/O ----------
220 template <typename T>
221 static void readInts(const char *Start, const char *End,
222 std::set<uint64_t> *Ints) {
223 const T *S = reinterpret_cast<const T *>(Start);
224 const T *E = reinterpret_cast<const T *>(End);
225 std::copy(S, E, std::inserter(*Ints, Ints->end()));
228 ErrorOr<std::unique_ptr<RawCoverage>>
229 RawCoverage::read(const std::string &FileName) {
230 ErrorOr<std::unique_ptr<MemoryBuffer>> BufOrErr =
231 MemoryBuffer::getFile(FileName);
232 if (!BufOrErr)
233 return BufOrErr.getError();
234 std::unique_ptr<MemoryBuffer> Buf = std::move(BufOrErr.get());
235 if (Buf->getBufferSize() < 8) {
236 errs() << "File too small (<8): " << Buf->getBufferSize() << '\n';
237 return make_error_code(errc::illegal_byte_sequence);
239 const FileHeader *Header =
240 reinterpret_cast<const FileHeader *>(Buf->getBufferStart());
242 if (Header->Magic != BinCoverageMagic) {
243 errs() << "Wrong magic: " << Header->Magic << '\n';
244 return make_error_code(errc::illegal_byte_sequence);
247 auto Addrs = std::make_unique<std::set<uint64_t>>();
249 switch (Header->Bitness) {
250 case Bitness64:
251 readInts<uint64_t>(Buf->getBufferStart() + 8, Buf->getBufferEnd(),
252 Addrs.get());
253 break;
254 case Bitness32:
255 readInts<uint32_t>(Buf->getBufferStart() + 8, Buf->getBufferEnd(),
256 Addrs.get());
257 break;
258 default:
259 errs() << "Unsupported bitness: " << Header->Bitness << '\n';
260 return make_error_code(errc::illegal_byte_sequence);
263 // Ignore slots that are zero, so a runtime implementation is not required
264 // to compactify the data.
265 Addrs->erase(0);
267 return std::unique_ptr<RawCoverage>(new RawCoverage(std::move(Addrs)));
270 // Print coverage addresses.
271 raw_ostream &operator<<(raw_ostream &OS, const RawCoverage &CoverageData) {
272 for (auto Addr : *CoverageData.Addrs) {
273 OS << "0x";
274 OS.write_hex(Addr);
275 OS << "\n";
277 return OS;
280 static raw_ostream &operator<<(raw_ostream &OS, const CoverageStats &Stats) {
281 OS << "all-edges: " << Stats.AllPoints << "\n";
282 OS << "cov-edges: " << Stats.CovPoints << "\n";
283 OS << "all-functions: " << Stats.AllFns << "\n";
284 OS << "cov-functions: " << Stats.CovFns << "\n";
285 return OS;
288 // Output symbolized information for coverage points in JSON.
289 // Format:
290 // {
291 // '<file_name>' : {
292 // '<function_name>' : {
293 // '<point_id'> : '<line_number>:'<column_number'.
294 // ....
295 // }
296 // }
297 // }
298 static void operator<<(json::OStream &W,
299 const std::vector<CoveragePoint> &Points) {
300 // Group points by file.
301 std::map<std::string, std::vector<const CoveragePoint *>> PointsByFile;
302 for (const auto &Point : Points) {
303 for (const DILineInfo &Loc : Point.Locs) {
304 PointsByFile[Loc.FileName].push_back(&Point);
308 for (const auto &P : PointsByFile) {
309 std::string FileName = P.first;
310 std::map<std::string, std::vector<const CoveragePoint *>> PointsByFn;
311 for (auto PointPtr : P.second) {
312 for (const DILineInfo &Loc : PointPtr->Locs) {
313 PointsByFn[Loc.FunctionName].push_back(PointPtr);
317 W.attributeObject(P.first, [&] {
318 // Group points by function.
319 for (const auto &P : PointsByFn) {
320 std::string FunctionName = P.first;
321 std::set<std::string> WrittenIds;
323 W.attributeObject(FunctionName, [&] {
324 for (const CoveragePoint *Point : P.second) {
325 for (const auto &Loc : Point->Locs) {
326 if (Loc.FileName != FileName || Loc.FunctionName != FunctionName)
327 continue;
328 if (WrittenIds.find(Point->Id) != WrittenIds.end())
329 continue;
331 // Output <point_id> : "<line>:<col>".
332 WrittenIds.insert(Point->Id);
333 W.attribute(Point->Id,
334 (utostr(Loc.Line) + ":" + utostr(Loc.Column)));
343 static void operator<<(json::OStream &W, const SymbolizedCoverage &C) {
344 W.object([&] {
345 W.attributeArray("covered-points", [&] {
346 for (const std::string &P : C.CoveredIds) {
347 W.value(P);
350 W.attribute("binary-hash", C.BinaryHash);
351 W.attributeObject("point-symbol-info", [&] { W << C.Points; });
355 static std::string parseScalarString(yaml::Node *N) {
356 SmallString<64> StringStorage;
357 yaml::ScalarNode *S = dyn_cast<yaml::ScalarNode>(N);
358 failIf(!S, "expected string");
359 return S->getValue(StringStorage);
362 std::unique_ptr<SymbolizedCoverage>
363 SymbolizedCoverage::read(const std::string &InputFile) {
364 auto Coverage(std::make_unique<SymbolizedCoverage>());
366 std::map<std::string, CoveragePoint> Points;
367 ErrorOr<std::unique_ptr<MemoryBuffer>> BufOrErr =
368 MemoryBuffer::getFile(InputFile);
369 failIfError(BufOrErr);
371 SourceMgr SM;
372 yaml::Stream S(**BufOrErr, SM);
374 yaml::document_iterator DI = S.begin();
375 failIf(DI == S.end(), "empty document: " + InputFile);
376 yaml::Node *Root = DI->getRoot();
377 failIf(!Root, "expecting root node: " + InputFile);
378 yaml::MappingNode *Top = dyn_cast<yaml::MappingNode>(Root);
379 failIf(!Top, "expecting mapping node: " + InputFile);
381 for (auto &KVNode : *Top) {
382 auto Key = parseScalarString(KVNode.getKey());
384 if (Key == "covered-points") {
385 yaml::SequenceNode *Points =
386 dyn_cast<yaml::SequenceNode>(KVNode.getValue());
387 failIf(!Points, "expected array: " + InputFile);
389 for (auto I = Points->begin(), E = Points->end(); I != E; ++I) {
390 Coverage->CoveredIds.insert(parseScalarString(&*I));
392 } else if (Key == "binary-hash") {
393 Coverage->BinaryHash = parseScalarString(KVNode.getValue());
394 } else if (Key == "point-symbol-info") {
395 yaml::MappingNode *PointSymbolInfo =
396 dyn_cast<yaml::MappingNode>(KVNode.getValue());
397 failIf(!PointSymbolInfo, "expected mapping node: " + InputFile);
399 for (auto &FileKVNode : *PointSymbolInfo) {
400 auto Filename = parseScalarString(FileKVNode.getKey());
402 yaml::MappingNode *FileInfo =
403 dyn_cast<yaml::MappingNode>(FileKVNode.getValue());
404 failIf(!FileInfo, "expected mapping node: " + InputFile);
406 for (auto &FunctionKVNode : *FileInfo) {
407 auto FunctionName = parseScalarString(FunctionKVNode.getKey());
409 yaml::MappingNode *FunctionInfo =
410 dyn_cast<yaml::MappingNode>(FunctionKVNode.getValue());
411 failIf(!FunctionInfo, "expected mapping node: " + InputFile);
413 for (auto &PointKVNode : *FunctionInfo) {
414 auto PointId = parseScalarString(PointKVNode.getKey());
415 auto Loc = parseScalarString(PointKVNode.getValue());
417 size_t ColonPos = Loc.find(':');
418 failIf(ColonPos == std::string::npos, "expected ':': " + InputFile);
420 auto LineStr = Loc.substr(0, ColonPos);
421 auto ColStr = Loc.substr(ColonPos + 1, Loc.size());
423 if (Points.find(PointId) == Points.end())
424 Points.insert(std::make_pair(PointId, CoveragePoint(PointId)));
426 DILineInfo LineInfo;
427 LineInfo.FileName = Filename;
428 LineInfo.FunctionName = FunctionName;
429 char *End;
430 LineInfo.Line = std::strtoul(LineStr.c_str(), &End, 10);
431 LineInfo.Column = std::strtoul(ColStr.c_str(), &End, 10);
433 CoveragePoint *CoveragePoint = &Points.find(PointId)->second;
434 CoveragePoint->Locs.push_back(LineInfo);
438 } else {
439 errs() << "Ignoring unknown key: " << Key << "\n";
443 for (auto &KV : Points) {
444 Coverage->Points.push_back(KV.second);
447 return Coverage;
450 // ---------- MAIN FUNCTIONALITY ----------
452 std::string stripPathPrefix(std::string Path) {
453 if (ClStripPathPrefix.empty())
454 return Path;
455 size_t Pos = Path.find(ClStripPathPrefix);
456 if (Pos == std::string::npos)
457 return Path;
458 return Path.substr(Pos + ClStripPathPrefix.size());
461 static std::unique_ptr<symbolize::LLVMSymbolizer> createSymbolizer() {
462 symbolize::LLVMSymbolizer::Options SymbolizerOptions;
463 SymbolizerOptions.Demangle = ClDemangle;
464 SymbolizerOptions.UseSymbolTable = true;
465 return std::unique_ptr<symbolize::LLVMSymbolizer>(
466 new symbolize::LLVMSymbolizer(SymbolizerOptions));
469 static std::string normalizeFilename(const std::string &FileName) {
470 SmallString<256> S(FileName);
471 sys::path::remove_dots(S, /* remove_dot_dot */ true);
472 return stripPathPrefix(S.str().str());
475 class Blacklists {
476 public:
477 Blacklists()
478 : DefaultBlacklist(createDefaultBlacklist()),
479 UserBlacklist(createUserBlacklist()) {}
481 bool isBlacklisted(const DILineInfo &I) {
482 if (DefaultBlacklist &&
483 DefaultBlacklist->inSection("sancov", "fun", I.FunctionName))
484 return true;
485 if (DefaultBlacklist &&
486 DefaultBlacklist->inSection("sancov", "src", I.FileName))
487 return true;
488 if (UserBlacklist &&
489 UserBlacklist->inSection("sancov", "fun", I.FunctionName))
490 return true;
491 if (UserBlacklist && UserBlacklist->inSection("sancov", "src", I.FileName))
492 return true;
493 return false;
496 private:
497 static std::unique_ptr<SpecialCaseList> createDefaultBlacklist() {
498 if (!ClUseDefaultBlacklist)
499 return std::unique_ptr<SpecialCaseList>();
500 std::unique_ptr<MemoryBuffer> MB =
501 MemoryBuffer::getMemBuffer(DefaultBlacklistStr);
502 std::string Error;
503 auto Blacklist = SpecialCaseList::create(MB.get(), Error);
504 failIfNotEmpty(Error);
505 return Blacklist;
508 static std::unique_ptr<SpecialCaseList> createUserBlacklist() {
509 if (ClBlacklist.empty())
510 return std::unique_ptr<SpecialCaseList>();
512 return SpecialCaseList::createOrDie({{ClBlacklist}});
514 std::unique_ptr<SpecialCaseList> DefaultBlacklist;
515 std::unique_ptr<SpecialCaseList> UserBlacklist;
518 static std::vector<CoveragePoint>
519 getCoveragePoints(const std::string &ObjectFile,
520 const std::set<uint64_t> &Addrs,
521 const std::set<uint64_t> &CoveredAddrs) {
522 std::vector<CoveragePoint> Result;
523 auto Symbolizer(createSymbolizer());
524 Blacklists B;
526 std::set<std::string> CoveredFiles;
527 if (ClSkipDeadFiles) {
528 for (auto Addr : CoveredAddrs) {
529 // TODO: it would be neccessary to set proper section index here.
530 // object::SectionedAddress::UndefSection works for only absolute
531 // addresses.
532 object::SectionedAddress ModuleAddress = {
533 Addr, object::SectionedAddress::UndefSection};
535 auto LineInfo = Symbolizer->symbolizeCode(ObjectFile, ModuleAddress);
536 failIfError(LineInfo);
537 CoveredFiles.insert(LineInfo->FileName);
538 auto InliningInfo =
539 Symbolizer->symbolizeInlinedCode(ObjectFile, ModuleAddress);
540 failIfError(InliningInfo);
541 for (uint32_t I = 0; I < InliningInfo->getNumberOfFrames(); ++I) {
542 auto FrameInfo = InliningInfo->getFrame(I);
543 CoveredFiles.insert(FrameInfo.FileName);
548 for (auto Addr : Addrs) {
549 std::set<DILineInfo> Infos; // deduplicate debug info.
551 // TODO: it would be neccessary to set proper section index here.
552 // object::SectionedAddress::UndefSection works for only absolute addresses.
553 object::SectionedAddress ModuleAddress = {
554 Addr, object::SectionedAddress::UndefSection};
556 auto LineInfo = Symbolizer->symbolizeCode(ObjectFile, ModuleAddress);
557 failIfError(LineInfo);
558 if (ClSkipDeadFiles &&
559 CoveredFiles.find(LineInfo->FileName) == CoveredFiles.end())
560 continue;
561 LineInfo->FileName = normalizeFilename(LineInfo->FileName);
562 if (B.isBlacklisted(*LineInfo))
563 continue;
565 auto Id = utohexstr(Addr, true);
566 auto Point = CoveragePoint(Id);
567 Infos.insert(*LineInfo);
568 Point.Locs.push_back(*LineInfo);
570 auto InliningInfo =
571 Symbolizer->symbolizeInlinedCode(ObjectFile, ModuleAddress);
572 failIfError(InliningInfo);
573 for (uint32_t I = 0; I < InliningInfo->getNumberOfFrames(); ++I) {
574 auto FrameInfo = InliningInfo->getFrame(I);
575 if (ClSkipDeadFiles &&
576 CoveredFiles.find(FrameInfo.FileName) == CoveredFiles.end())
577 continue;
578 FrameInfo.FileName = normalizeFilename(FrameInfo.FileName);
579 if (B.isBlacklisted(FrameInfo))
580 continue;
581 if (Infos.find(FrameInfo) == Infos.end()) {
582 Infos.insert(FrameInfo);
583 Point.Locs.push_back(FrameInfo);
587 Result.push_back(Point);
590 return Result;
593 static bool isCoveragePointSymbol(StringRef Name) {
594 return Name == "__sanitizer_cov" || Name == "__sanitizer_cov_with_check" ||
595 Name == "__sanitizer_cov_trace_func_enter" ||
596 Name == "__sanitizer_cov_trace_pc_guard" ||
597 // Mac has '___' prefix
598 Name == "___sanitizer_cov" || Name == "___sanitizer_cov_with_check" ||
599 Name == "___sanitizer_cov_trace_func_enter" ||
600 Name == "___sanitizer_cov_trace_pc_guard";
603 // Locate __sanitizer_cov* function addresses inside the stubs table on MachO.
604 static void findMachOIndirectCovFunctions(const object::MachOObjectFile &O,
605 std::set<uint64_t> *Result) {
606 MachO::dysymtab_command Dysymtab = O.getDysymtabLoadCommand();
607 MachO::symtab_command Symtab = O.getSymtabLoadCommand();
609 for (const auto &Load : O.load_commands()) {
610 if (Load.C.cmd == MachO::LC_SEGMENT_64) {
611 MachO::segment_command_64 Seg = O.getSegment64LoadCommand(Load);
612 for (unsigned J = 0; J < Seg.nsects; ++J) {
613 MachO::section_64 Sec = O.getSection64(Load, J);
615 uint32_t SectionType = Sec.flags & MachO::SECTION_TYPE;
616 if (SectionType == MachO::S_SYMBOL_STUBS) {
617 uint32_t Stride = Sec.reserved2;
618 uint32_t Cnt = Sec.size / Stride;
619 uint32_t N = Sec.reserved1;
620 for (uint32_t J = 0; J < Cnt && N + J < Dysymtab.nindirectsyms; J++) {
621 uint32_t IndirectSymbol =
622 O.getIndirectSymbolTableEntry(Dysymtab, N + J);
623 uint64_t Addr = Sec.addr + J * Stride;
624 if (IndirectSymbol < Symtab.nsyms) {
625 object::SymbolRef Symbol = *(O.getSymbolByIndex(IndirectSymbol));
626 Expected<StringRef> Name = Symbol.getName();
627 failIfError(Name);
628 if (isCoveragePointSymbol(Name.get())) {
629 Result->insert(Addr);
636 if (Load.C.cmd == MachO::LC_SEGMENT) {
637 errs() << "ERROR: 32 bit MachO binaries not supported\n";
642 // Locate __sanitizer_cov* function addresses that are used for coverage
643 // reporting.
644 static std::set<uint64_t>
645 findSanitizerCovFunctions(const object::ObjectFile &O) {
646 std::set<uint64_t> Result;
648 for (const object::SymbolRef &Symbol : O.symbols()) {
649 Expected<uint64_t> AddressOrErr = Symbol.getAddress();
650 failIfError(AddressOrErr);
651 uint64_t Address = AddressOrErr.get();
653 Expected<StringRef> NameOrErr = Symbol.getName();
654 failIfError(NameOrErr);
655 StringRef Name = NameOrErr.get();
657 if (!(Symbol.getFlags() & object::BasicSymbolRef::SF_Undefined) &&
658 isCoveragePointSymbol(Name)) {
659 Result.insert(Address);
663 if (const auto *CO = dyn_cast<object::COFFObjectFile>(&O)) {
664 for (const object::ExportDirectoryEntryRef &Export :
665 CO->export_directories()) {
666 uint32_t RVA;
667 std::error_code EC = Export.getExportRVA(RVA);
668 failIfError(EC);
670 StringRef Name;
671 EC = Export.getSymbolName(Name);
672 failIfError(EC);
674 if (isCoveragePointSymbol(Name))
675 Result.insert(CO->getImageBase() + RVA);
679 if (const auto *MO = dyn_cast<object::MachOObjectFile>(&O)) {
680 findMachOIndirectCovFunctions(*MO, &Result);
683 return Result;
686 static uint64_t getPreviousInstructionPc(uint64_t PC,
687 Triple TheTriple) {
688 if (TheTriple.isARM()) {
689 return (PC - 3) & (~1);
690 } else if (TheTriple.isAArch64()) {
691 return PC - 4;
692 } else if (TheTriple.isMIPS()) {
693 return PC - 8;
694 } else {
695 return PC - 1;
699 // Locate addresses of all coverage points in a file. Coverage point
700 // is defined as the 'address of instruction following __sanitizer_cov
701 // call - 1'.
702 static void getObjectCoveragePoints(const object::ObjectFile &O,
703 std::set<uint64_t> *Addrs) {
704 Triple TheTriple("unknown-unknown-unknown");
705 TheTriple.setArch(Triple::ArchType(O.getArch()));
706 auto TripleName = TheTriple.getTriple();
708 std::string Error;
709 const Target *TheTarget = TargetRegistry::lookupTarget(TripleName, Error);
710 failIfNotEmpty(Error);
712 std::unique_ptr<const MCSubtargetInfo> STI(
713 TheTarget->createMCSubtargetInfo(TripleName, "", ""));
714 failIfEmpty(STI, "no subtarget info for target " + TripleName);
716 std::unique_ptr<const MCRegisterInfo> MRI(
717 TheTarget->createMCRegInfo(TripleName));
718 failIfEmpty(MRI, "no register info for target " + TripleName);
720 std::unique_ptr<const MCAsmInfo> AsmInfo(
721 TheTarget->createMCAsmInfo(*MRI, TripleName));
722 failIfEmpty(AsmInfo, "no asm info for target " + TripleName);
724 std::unique_ptr<const MCObjectFileInfo> MOFI(new MCObjectFileInfo);
725 MCContext Ctx(AsmInfo.get(), MRI.get(), MOFI.get());
726 std::unique_ptr<MCDisassembler> DisAsm(
727 TheTarget->createMCDisassembler(*STI, Ctx));
728 failIfEmpty(DisAsm, "no disassembler info for target " + TripleName);
730 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo());
731 failIfEmpty(MII, "no instruction info for target " + TripleName);
733 std::unique_ptr<const MCInstrAnalysis> MIA(
734 TheTarget->createMCInstrAnalysis(MII.get()));
735 failIfEmpty(MIA, "no instruction analysis info for target " + TripleName);
737 auto SanCovAddrs = findSanitizerCovFunctions(O);
738 if (SanCovAddrs.empty())
739 fail("__sanitizer_cov* functions not found");
741 for (object::SectionRef Section : O.sections()) {
742 if (Section.isVirtual() || !Section.isText()) // llvm-objdump does the same.
743 continue;
744 uint64_t SectionAddr = Section.getAddress();
745 uint64_t SectSize = Section.getSize();
746 if (!SectSize)
747 continue;
749 Expected<StringRef> BytesStr = Section.getContents();
750 failIfError(BytesStr);
751 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef(*BytesStr);
753 for (uint64_t Index = 0, Size = 0; Index < Section.getSize();
754 Index += Size) {
755 MCInst Inst;
756 if (!DisAsm->getInstruction(Inst, Size, Bytes.slice(Index),
757 SectionAddr + Index, nulls(), nulls())) {
758 if (Size == 0)
759 Size = 1;
760 continue;
762 uint64_t Addr = Index + SectionAddr;
763 // Sanitizer coverage uses the address of the next instruction - 1.
764 uint64_t CovPoint = getPreviousInstructionPc(Addr + Size, TheTriple);
765 uint64_t Target;
766 if (MIA->isCall(Inst) &&
767 MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target) &&
768 SanCovAddrs.find(Target) != SanCovAddrs.end())
769 Addrs->insert(CovPoint);
774 static void
775 visitObjectFiles(const object::Archive &A,
776 function_ref<void(const object::ObjectFile &)> Fn) {
777 Error Err = Error::success();
778 for (auto &C : A.children(Err)) {
779 Expected<std::unique_ptr<object::Binary>> ChildOrErr = C.getAsBinary();
780 failIfError(ChildOrErr);
781 if (auto *O = dyn_cast<object::ObjectFile>(&*ChildOrErr.get()))
782 Fn(*O);
783 else
784 failIfError(object::object_error::invalid_file_type);
786 failIfError(std::move(Err));
789 static void
790 visitObjectFiles(const std::string &FileName,
791 function_ref<void(const object::ObjectFile &)> Fn) {
792 Expected<object::OwningBinary<object::Binary>> BinaryOrErr =
793 object::createBinary(FileName);
794 if (!BinaryOrErr)
795 failIfError(BinaryOrErr);
797 object::Binary &Binary = *BinaryOrErr.get().getBinary();
798 if (object::Archive *A = dyn_cast<object::Archive>(&Binary))
799 visitObjectFiles(*A, Fn);
800 else if (object::ObjectFile *O = dyn_cast<object::ObjectFile>(&Binary))
801 Fn(*O);
802 else
803 failIfError(object::object_error::invalid_file_type);
806 static std::set<uint64_t>
807 findSanitizerCovFunctions(const std::string &FileName) {
808 std::set<uint64_t> Result;
809 visitObjectFiles(FileName, [&](const object::ObjectFile &O) {
810 auto Addrs = findSanitizerCovFunctions(O);
811 Result.insert(Addrs.begin(), Addrs.end());
813 return Result;
816 // Locate addresses of all coverage points in a file. Coverage point
817 // is defined as the 'address of instruction following __sanitizer_cov
818 // call - 1'.
819 static std::set<uint64_t> findCoveragePointAddrs(const std::string &FileName) {
820 std::set<uint64_t> Result;
821 visitObjectFiles(FileName, [&](const object::ObjectFile &O) {
822 getObjectCoveragePoints(O, &Result);
824 return Result;
827 static void printCovPoints(const std::string &ObjFile, raw_ostream &OS) {
828 for (uint64_t Addr : findCoveragePointAddrs(ObjFile)) {
829 OS << "0x";
830 OS.write_hex(Addr);
831 OS << "\n";
835 static ErrorOr<bool> isCoverageFile(const std::string &FileName) {
836 auto ShortFileName = llvm::sys::path::filename(FileName);
837 if (!SancovFileRegex.match(ShortFileName))
838 return false;
840 ErrorOr<std::unique_ptr<MemoryBuffer>> BufOrErr =
841 MemoryBuffer::getFile(FileName);
842 if (!BufOrErr) {
843 errs() << "Warning: " << BufOrErr.getError().message() << "("
844 << BufOrErr.getError().value()
845 << "), filename: " << llvm::sys::path::filename(FileName) << "\n";
846 return BufOrErr.getError();
848 std::unique_ptr<MemoryBuffer> Buf = std::move(BufOrErr.get());
849 if (Buf->getBufferSize() < 8) {
850 return false;
852 const FileHeader *Header =
853 reinterpret_cast<const FileHeader *>(Buf->getBufferStart());
854 return Header->Magic == BinCoverageMagic;
857 static bool isSymbolizedCoverageFile(const std::string &FileName) {
858 auto ShortFileName = llvm::sys::path::filename(FileName);
859 return SymcovFileRegex.match(ShortFileName);
862 static std::unique_ptr<SymbolizedCoverage>
863 symbolize(const RawCoverage &Data, const std::string ObjectFile) {
864 auto Coverage = std::make_unique<SymbolizedCoverage>();
866 ErrorOr<std::unique_ptr<MemoryBuffer>> BufOrErr =
867 MemoryBuffer::getFile(ObjectFile);
868 failIfError(BufOrErr);
869 SHA1 Hasher;
870 Hasher.update((*BufOrErr)->getBuffer());
871 Coverage->BinaryHash = toHex(Hasher.final());
873 Blacklists B;
874 auto Symbolizer(createSymbolizer());
876 for (uint64_t Addr : *Data.Addrs) {
877 // TODO: it would be neccessary to set proper section index here.
878 // object::SectionedAddress::UndefSection works for only absolute addresses.
879 auto LineInfo = Symbolizer->symbolizeCode(
880 ObjectFile, {Addr, object::SectionedAddress::UndefSection});
881 failIfError(LineInfo);
882 if (B.isBlacklisted(*LineInfo))
883 continue;
885 Coverage->CoveredIds.insert(utohexstr(Addr, true));
888 std::set<uint64_t> AllAddrs = findCoveragePointAddrs(ObjectFile);
889 if (!std::includes(AllAddrs.begin(), AllAddrs.end(), Data.Addrs->begin(),
890 Data.Addrs->end())) {
891 fail("Coverage points in binary and .sancov file do not match.");
893 Coverage->Points = getCoveragePoints(ObjectFile, AllAddrs, *Data.Addrs);
894 return Coverage;
897 struct FileFn {
898 bool operator<(const FileFn &RHS) const {
899 return std::tie(FileName, FunctionName) <
900 std::tie(RHS.FileName, RHS.FunctionName);
903 std::string FileName;
904 std::string FunctionName;
907 static std::set<FileFn>
908 computeFunctions(const std::vector<CoveragePoint> &Points) {
909 std::set<FileFn> Fns;
910 for (const auto &Point : Points) {
911 for (const auto &Loc : Point.Locs) {
912 Fns.insert(FileFn{Loc.FileName, Loc.FunctionName});
915 return Fns;
918 static std::set<FileFn>
919 computeNotCoveredFunctions(const SymbolizedCoverage &Coverage) {
920 auto Fns = computeFunctions(Coverage.Points);
922 for (const auto &Point : Coverage.Points) {
923 if (Coverage.CoveredIds.find(Point.Id) == Coverage.CoveredIds.end())
924 continue;
926 for (const auto &Loc : Point.Locs) {
927 Fns.erase(FileFn{Loc.FileName, Loc.FunctionName});
931 return Fns;
934 static std::set<FileFn>
935 computeCoveredFunctions(const SymbolizedCoverage &Coverage) {
936 auto AllFns = computeFunctions(Coverage.Points);
937 std::set<FileFn> Result;
939 for (const auto &Point : Coverage.Points) {
940 if (Coverage.CoveredIds.find(Point.Id) == Coverage.CoveredIds.end())
941 continue;
943 for (const auto &Loc : Point.Locs) {
944 Result.insert(FileFn{Loc.FileName, Loc.FunctionName});
948 return Result;
951 typedef std::map<FileFn, std::pair<uint32_t, uint32_t>> FunctionLocs;
952 // finds first location in a file for each function.
953 static FunctionLocs resolveFunctions(const SymbolizedCoverage &Coverage,
954 const std::set<FileFn> &Fns) {
955 FunctionLocs Result;
956 for (const auto &Point : Coverage.Points) {
957 for (const auto &Loc : Point.Locs) {
958 FileFn Fn = FileFn{Loc.FileName, Loc.FunctionName};
959 if (Fns.find(Fn) == Fns.end())
960 continue;
962 auto P = std::make_pair(Loc.Line, Loc.Column);
963 auto I = Result.find(Fn);
964 if (I == Result.end() || I->second > P) {
965 Result[Fn] = P;
969 return Result;
972 static void printFunctionLocs(const FunctionLocs &FnLocs, raw_ostream &OS) {
973 for (const auto &P : FnLocs) {
974 OS << stripPathPrefix(P.first.FileName) << ":" << P.second.first << " "
975 << P.first.FunctionName << "\n";
978 CoverageStats computeStats(const SymbolizedCoverage &Coverage) {
979 CoverageStats Stats = {Coverage.Points.size(), Coverage.CoveredIds.size(),
980 computeFunctions(Coverage.Points).size(),
981 computeCoveredFunctions(Coverage).size()};
982 return Stats;
985 // Print list of covered functions.
986 // Line format: <file_name>:<line> <function_name>
987 static void printCoveredFunctions(const SymbolizedCoverage &CovData,
988 raw_ostream &OS) {
989 auto CoveredFns = computeCoveredFunctions(CovData);
990 printFunctionLocs(resolveFunctions(CovData, CoveredFns), OS);
993 // Print list of not covered functions.
994 // Line format: <file_name>:<line> <function_name>
995 static void printNotCoveredFunctions(const SymbolizedCoverage &CovData,
996 raw_ostream &OS) {
997 auto NotCoveredFns = computeNotCoveredFunctions(CovData);
998 printFunctionLocs(resolveFunctions(CovData, NotCoveredFns), OS);
1001 // Read list of files and merges their coverage info.
1002 static void readAndPrintRawCoverage(const std::vector<std::string> &FileNames,
1003 raw_ostream &OS) {
1004 std::vector<std::unique_ptr<RawCoverage>> Covs;
1005 for (const auto &FileName : FileNames) {
1006 auto Cov = RawCoverage::read(FileName);
1007 if (!Cov)
1008 continue;
1009 OS << *Cov.get();
1013 static std::unique_ptr<SymbolizedCoverage>
1014 merge(const std::vector<std::unique_ptr<SymbolizedCoverage>> &Coverages) {
1015 if (Coverages.empty())
1016 return nullptr;
1018 auto Result = std::make_unique<SymbolizedCoverage>();
1020 for (size_t I = 0; I < Coverages.size(); ++I) {
1021 const SymbolizedCoverage &Coverage = *Coverages[I];
1022 std::string Prefix;
1023 if (Coverages.size() > 1) {
1024 // prefix is not needed when there's only one file.
1025 Prefix = utostr(I);
1028 for (const auto &Id : Coverage.CoveredIds) {
1029 Result->CoveredIds.insert(Prefix + Id);
1032 for (const auto &CovPoint : Coverage.Points) {
1033 CoveragePoint NewPoint(CovPoint);
1034 NewPoint.Id = Prefix + CovPoint.Id;
1035 Result->Points.push_back(NewPoint);
1039 if (Coverages.size() == 1) {
1040 Result->BinaryHash = Coverages[0]->BinaryHash;
1043 return Result;
1046 static std::unique_ptr<SymbolizedCoverage>
1047 readSymbolizeAndMergeCmdArguments(std::vector<std::string> FileNames) {
1048 std::vector<std::unique_ptr<SymbolizedCoverage>> Coverages;
1051 // Short name => file name.
1052 std::map<std::string, std::string> ObjFiles;
1053 std::string FirstObjFile;
1054 std::set<std::string> CovFiles;
1056 // Partition input values into coverage/object files.
1057 for (const auto &FileName : FileNames) {
1058 if (isSymbolizedCoverageFile(FileName)) {
1059 Coverages.push_back(SymbolizedCoverage::read(FileName));
1062 auto ErrorOrIsCoverage = isCoverageFile(FileName);
1063 if (!ErrorOrIsCoverage)
1064 continue;
1065 if (ErrorOrIsCoverage.get()) {
1066 CovFiles.insert(FileName);
1067 } else {
1068 auto ShortFileName = llvm::sys::path::filename(FileName);
1069 if (ObjFiles.find(ShortFileName) != ObjFiles.end()) {
1070 fail("Duplicate binary file with a short name: " + ShortFileName);
1073 ObjFiles[ShortFileName] = FileName;
1074 if (FirstObjFile.empty())
1075 FirstObjFile = FileName;
1079 SmallVector<StringRef, 2> Components;
1081 // Object file => list of corresponding coverage file names.
1082 std::map<std::string, std::vector<std::string>> CoverageByObjFile;
1083 for (const auto &FileName : CovFiles) {
1084 auto ShortFileName = llvm::sys::path::filename(FileName);
1085 auto Ok = SancovFileRegex.match(ShortFileName, &Components);
1086 if (!Ok) {
1087 fail("Can't match coverage file name against "
1088 "<module_name>.<pid>.sancov pattern: " +
1089 FileName);
1092 auto Iter = ObjFiles.find(Components[1]);
1093 if (Iter == ObjFiles.end()) {
1094 fail("Object file for coverage not found: " + FileName);
1097 CoverageByObjFile[Iter->second].push_back(FileName);
1100 for (const auto &Pair : ObjFiles) {
1101 auto FileName = Pair.second;
1102 if (CoverageByObjFile.find(FileName) == CoverageByObjFile.end())
1103 errs() << "WARNING: No coverage file for " << FileName << "\n";
1106 // Read raw coverage and symbolize it.
1107 for (const auto &Pair : CoverageByObjFile) {
1108 if (findSanitizerCovFunctions(Pair.first).empty()) {
1109 errs()
1110 << "WARNING: Ignoring " << Pair.first
1111 << " and its coverage because __sanitizer_cov* functions were not "
1112 "found.\n";
1113 continue;
1116 for (const std::string &CoverageFile : Pair.second) {
1117 auto DataOrError = RawCoverage::read(CoverageFile);
1118 failIfError(DataOrError);
1119 Coverages.push_back(symbolize(*DataOrError.get(), Pair.first));
1124 return merge(Coverages);
1127 } // namespace
1129 int main(int Argc, char **Argv) {
1130 // Print stack trace if we signal out.
1131 sys::PrintStackTraceOnErrorSignal(Argv[0]);
1132 PrettyStackTraceProgram X(Argc, Argv);
1133 llvm_shutdown_obj Y; // Call llvm_shutdown() on exit.
1135 llvm::InitializeAllTargetInfos();
1136 llvm::InitializeAllTargetMCs();
1137 llvm::InitializeAllDisassemblers();
1139 cl::ParseCommandLineOptions(Argc, Argv,
1140 "Sanitizer Coverage Processing Tool (sancov)\n\n"
1141 " This tool can extract various coverage-related information from: \n"
1142 " coverage-instrumented binary files, raw .sancov files and their "
1143 "symbolized .symcov version.\n"
1144 " Depending on chosen action the tool expects different input files:\n"
1145 " -print-coverage-pcs - coverage-instrumented binary files\n"
1146 " -print-coverage - .sancov files\n"
1147 " <other actions> - .sancov files & corresponding binary "
1148 "files, .symcov files\n"
1151 // -print doesn't need object files.
1152 if (Action == PrintAction) {
1153 readAndPrintRawCoverage(ClInputFiles, outs());
1154 return 0;
1155 } else if (Action == PrintCovPointsAction) {
1156 // -print-coverage-points doesn't need coverage files.
1157 for (const std::string &ObjFile : ClInputFiles) {
1158 printCovPoints(ObjFile, outs());
1160 return 0;
1163 auto Coverage = readSymbolizeAndMergeCmdArguments(ClInputFiles);
1164 failIf(!Coverage, "No valid coverage files given.");
1166 switch (Action) {
1167 case CoveredFunctionsAction: {
1168 printCoveredFunctions(*Coverage, outs());
1169 return 0;
1171 case NotCoveredFunctionsAction: {
1172 printNotCoveredFunctions(*Coverage, outs());
1173 return 0;
1175 case StatsAction: {
1176 outs() << computeStats(*Coverage);
1177 return 0;
1179 case MergeAction:
1180 case SymbolizeAction: { // merge & symbolize are synonims.
1181 json::OStream W(outs(), 2);
1182 W << *Coverage;
1183 return 0;
1185 case HtmlReportAction:
1186 errs() << "-html-report option is removed: "
1187 "use -symbolize & coverage-report-server.py instead\n";
1188 return 1;
1189 case PrintAction:
1190 case PrintCovPointsAction:
1191 llvm_unreachable("unsupported action");