1 //===-- LLVMSymbolize.cpp -------------------------------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // Implementation for LLVM symbolization library.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/DebugInfo/Symbolize/Symbolize.h"
16 #include "SymbolizableObjectFile.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/Config/config.h"
20 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
21 #include "llvm/DebugInfo/PDB/PDB.h"
22 #include "llvm/DebugInfo/PDB/PDBContext.h"
23 #include "llvm/Object/ELFObjectFile.h"
24 #include "llvm/Object/MachO.h"
25 #include "llvm/Support/COFF.h"
26 #include "llvm/Support/Casting.h"
27 #include "llvm/Support/Compression.h"
28 #include "llvm/Support/DataExtractor.h"
29 #include "llvm/Support/Errc.h"
30 #include "llvm/Support/FileSystem.h"
31 #include "llvm/Support/MemoryBuffer.h"
32 #include "llvm/Support/Path.h"
39 #pragma comment(lib, "dbghelp.lib")
41 // Windows.h conflicts with our COFF header definitions.
42 #ifdef IMAGE_FILE_MACHINE_I386
43 #undef IMAGE_FILE_MACHINE_I386
50 // FIXME: Move this to llvm-symbolizer tool.
51 static bool error(std::error_code ec) {
54 errs() << "LLVMSymbolizer: error reading file: " << ec.message() << ".\n";
59 const char LLVMSymbolizer::kBadString[] = "??";
61 std::string LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
62 uint64_t ModuleOffset) {
63 SymbolizableModule *Info = getOrCreateModuleInfo(ModuleName);
65 return printDILineInfo(DILineInfo(), Info);
67 // If the user is giving us relative addresses, add the preferred base of the
68 // object to the offset before we do the query. It's what DIContext expects.
69 if (Opts.RelativeAddresses)
70 ModuleOffset += Info->getModulePreferredBase();
72 if (Opts.PrintInlining) {
73 DIInliningInfo InlinedContext = Info->symbolizeInlinedCode(
74 ModuleOffset, Opts.PrintFunctions, Opts.UseSymbolTable);
75 uint32_t FramesNum = InlinedContext.getNumberOfFrames();
76 assert(FramesNum > 0);
78 for (uint32_t i = 0; i < FramesNum; i++) {
79 DILineInfo LineInfo = InlinedContext.getFrame(i);
80 Result += printDILineInfo(LineInfo, Info);
84 DILineInfo LineInfo = Info->symbolizeCode(ModuleOffset, Opts.PrintFunctions,
86 return printDILineInfo(LineInfo, Info);
89 std::string LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
90 uint64_t ModuleOffset) {
91 std::string Name = kBadString;
94 if (Opts.UseSymbolTable) {
95 if (SymbolizableModule *Info = getOrCreateModuleInfo(ModuleName)) {
96 // If the user is giving us relative addresses, add the preferred base of
97 // the object to the offset before we do the query. It's what DIContext
99 if (Opts.RelativeAddresses)
100 ModuleOffset += Info->getModulePreferredBase();
101 if (Info->symbolizeData(ModuleOffset, Name, Start, Size) && Opts.Demangle)
102 Name = DemangleName(Name, Info);
105 std::stringstream ss;
106 ss << Name << "\n" << Start << " " << Size << "\n";
110 void LLVMSymbolizer::flush() {
112 ObjectPairForPathArch.clear();
113 ObjectFileForArch.clear();
116 // For Path="/path/to/foo" and Basename="foo" assume that debug info is in
117 // /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
118 // For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
119 // /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
121 std::string getDarwinDWARFResourceForPath(
122 const std::string &Path, const std::string &Basename) {
123 SmallString<16> ResourceName = StringRef(Path);
124 if (sys::path::extension(Path) != ".dSYM") {
125 ResourceName += ".dSYM";
127 sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
128 sys::path::append(ResourceName, Basename);
129 return ResourceName.str();
132 static bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
133 ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
134 MemoryBuffer::getFileOrSTDIN(Path);
137 return !zlib::isAvailable() || CRCHash == zlib::crc32(MB.get()->getBuffer());
140 static bool findDebugBinary(const std::string &OrigPath,
141 const std::string &DebuglinkName, uint32_t CRCHash,
142 std::string &Result) {
143 std::string OrigRealPath = OrigPath;
144 #if defined(HAVE_REALPATH)
145 if (char *RP = realpath(OrigPath.c_str(), nullptr)) {
150 SmallString<16> OrigDir(OrigRealPath);
151 llvm::sys::path::remove_filename(OrigDir);
152 SmallString<16> DebugPath = OrigDir;
153 // Try /path/to/original_binary/debuglink_name
154 llvm::sys::path::append(DebugPath, DebuglinkName);
155 if (checkFileCRC(DebugPath, CRCHash)) {
156 Result = DebugPath.str();
159 // Try /path/to/original_binary/.debug/debuglink_name
160 DebugPath = OrigRealPath;
161 llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
162 if (checkFileCRC(DebugPath, CRCHash)) {
163 Result = DebugPath.str();
166 // Try /usr/lib/debug/path/to/original_binary/debuglink_name
167 DebugPath = "/usr/lib/debug";
168 llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
170 if (checkFileCRC(DebugPath, CRCHash)) {
171 Result = DebugPath.str();
177 static bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
181 for (const SectionRef &Section : Obj->sections()) {
183 Section.getName(Name);
184 Name = Name.substr(Name.find_first_not_of("._"));
185 if (Name == "gnu_debuglink") {
187 Section.getContents(Data);
188 DataExtractor DE(Data, Obj->isLittleEndian(), 0);
190 if (const char *DebugNameStr = DE.getCStr(&Offset)) {
191 // 4-byte align the offset.
192 Offset = (Offset + 3) & ~0x3;
193 if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
194 DebugName = DebugNameStr;
195 CRCHash = DE.getU32(&Offset);
206 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
207 const MachOObjectFile *Obj) {
208 ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
209 ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
210 if (dbg_uuid.empty() || bin_uuid.empty())
212 return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
215 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
216 const MachOObjectFile *MachExeObj, const std::string &ArchName) {
217 // On Darwin we may find DWARF in separate object file in
218 // resource directory.
219 std::vector<std::string> DsymPaths;
220 StringRef Filename = sys::path::filename(ExePath);
221 DsymPaths.push_back(getDarwinDWARFResourceForPath(ExePath, Filename));
222 for (const auto &Path : Opts.DsymHints) {
223 DsymPaths.push_back(getDarwinDWARFResourceForPath(Path, Filename));
225 for (const auto &path : DsymPaths) {
226 ErrorOr<OwningBinary<Binary>> BinaryOrErr = createBinary(path);
227 std::error_code EC = BinaryOrErr.getError();
228 if (EC != errc::no_such_file_or_directory && !error(EC)) {
229 OwningBinary<Binary> B = std::move(BinaryOrErr.get());
231 getObjectFileFromBinary(B.getBinary(), ArchName);
232 const MachOObjectFile *MachDbgObj =
233 dyn_cast<const MachOObjectFile>(DbgObj);
234 if (!MachDbgObj) continue;
235 if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj)) {
236 addOwningBinary(std::move(B));
244 LLVMSymbolizer::ObjectPair
245 LLVMSymbolizer::getOrCreateObjects(const std::string &Path,
246 const std::string &ArchName) {
247 const auto &I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
248 if (I != ObjectPairForPathArch.end())
250 ObjectFile *Obj = nullptr;
251 ObjectFile *DbgObj = nullptr;
252 ErrorOr<OwningBinary<Binary>> BinaryOrErr = createBinary(Path);
253 if (!error(BinaryOrErr.getError())) {
254 OwningBinary<Binary> &B = BinaryOrErr.get();
255 Obj = getObjectFileFromBinary(B.getBinary(), ArchName);
257 ObjectPair Res = std::make_pair(nullptr, nullptr);
258 ObjectPairForPathArch[std::make_pair(Path, ArchName)] = Res;
261 addOwningBinary(std::move(B));
262 if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
263 DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
264 // Try to locate the debug binary using .gnu_debuglink section.
266 std::string DebuglinkName;
268 std::string DebugBinaryPath;
269 if (getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash) &&
270 findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath)) {
271 BinaryOrErr = createBinary(DebugBinaryPath);
272 if (!error(BinaryOrErr.getError())) {
273 OwningBinary<Binary> B = std::move(BinaryOrErr.get());
274 DbgObj = getObjectFileFromBinary(B.getBinary(), ArchName);
275 addOwningBinary(std::move(B));
282 ObjectPair Res = std::make_pair(Obj, DbgObj);
283 ObjectPairForPathArch[std::make_pair(Path, ArchName)] = Res;
288 LLVMSymbolizer::getObjectFileFromBinary(Binary *Bin,
289 const std::string &ArchName) {
292 ObjectFile *Res = nullptr;
293 if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(Bin)) {
294 const auto &I = ObjectFileForArch.find(
295 std::make_pair(UB, ArchName));
296 if (I != ObjectFileForArch.end())
298 ErrorOr<std::unique_ptr<ObjectFile>> ParsedObj =
299 UB->getObjectForArch(ArchName);
301 Res = ParsedObj.get().get();
302 ParsedBinariesAndObjects.push_back(std::move(ParsedObj.get()));
304 ObjectFileForArch[std::make_pair(UB, ArchName)] = Res;
305 } else if (Bin->isObject()) {
306 Res = cast<ObjectFile>(Bin);
312 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) {
313 const auto &I = Modules.find(ModuleName);
314 if (I != Modules.end())
315 return I->second.get();
316 std::string BinaryName = ModuleName;
317 std::string ArchName = Opts.DefaultArch;
318 size_t ColonPos = ModuleName.find_last_of(':');
319 // Verify that substring after colon form a valid arch name.
320 if (ColonPos != std::string::npos) {
321 std::string ArchStr = ModuleName.substr(ColonPos + 1);
322 if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
323 BinaryName = ModuleName.substr(0, ColonPos);
327 ObjectPair Objects = getOrCreateObjects(BinaryName, ArchName);
329 if (!Objects.first) {
330 // Failed to find valid object file.
331 Modules.insert(std::make_pair(ModuleName, nullptr));
334 std::unique_ptr<DIContext> Context;
335 if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
336 // If this is a COFF object, assume it contains PDB debug information. If
337 // we don't find any we will fall back to the DWARF case.
338 std::unique_ptr<IPDBSession> Session;
339 PDB_ErrorCode Error = loadDataForEXE(PDB_ReaderType::DIA,
340 Objects.first->getFileName(), Session);
341 if (Error == PDB_ErrorCode::Success) {
342 Context.reset(new PDBContext(*CoffObject, std::move(Session)));
346 Context.reset(new DWARFContextInMemory(*Objects.second));
349 SymbolizableObjectFile::create(Objects.first, std::move(Context));
350 if (error(ErrOrInfo.getError())) {
351 Modules.insert(std::make_pair(ModuleName, nullptr));
354 SymbolizableModule *Res = ErrOrInfo.get().get();
355 Modules.insert(std::make_pair(ModuleName, std::move(ErrOrInfo.get())));
360 LLVMSymbolizer::printDILineInfo(DILineInfo LineInfo,
361 const SymbolizableModule *ModInfo) const {
362 // By default, DILineInfo contains "<invalid>" for function/filename it
363 // cannot fetch. We replace it to "??" to make our output closer to addr2line.
364 static const std::string kDILineInfoBadString = "<invalid>";
365 std::stringstream Result;
366 if (Opts.PrintFunctions != FunctionNameKind::None) {
367 std::string FunctionName = LineInfo.FunctionName;
368 if (FunctionName == kDILineInfoBadString)
369 FunctionName = kBadString;
370 else if (Opts.Demangle)
371 FunctionName = DemangleName(FunctionName, ModInfo);
372 Result << FunctionName << "\n";
374 std::string Filename = LineInfo.FileName;
375 if (Filename == kDILineInfoBadString)
376 Filename = kBadString;
377 Result << Filename << ":" << LineInfo.Line << ":" << LineInfo.Column << "\n";
381 // Undo these various manglings for Win32 extern "C" functions:
384 // fastcall - @foo@12
385 // vectorcall - foo@@12
386 // These are all different linkage names for 'foo'.
387 static StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
388 // Remove any '_' or '@' prefix.
389 char Front = SymbolName.empty() ? '\0' : SymbolName[0];
390 if (Front == '_' || Front == '@')
391 SymbolName = SymbolName.drop_front();
393 // Remove any '@[0-9]+' suffix.
395 size_t AtPos = SymbolName.rfind('@');
396 if (AtPos != StringRef::npos &&
397 std::all_of(SymbolName.begin() + AtPos + 1, SymbolName.end(),
398 [](char C) { return C >= '0' && C <= '9'; })) {
399 SymbolName = SymbolName.substr(0, AtPos);
403 // Remove any ending '@' for vectorcall.
404 if (SymbolName.endswith("@"))
405 SymbolName = SymbolName.drop_back();
410 #if !defined(_MSC_VER)
411 // Assume that __cxa_demangle is provided by libcxxabi (except for Windows).
412 extern "C" char *__cxa_demangle(const char *mangled_name, char *output_buffer,
413 size_t *length, int *status);
416 std::string LLVMSymbolizer::DemangleName(const std::string &Name,
417 const SymbolizableModule *ModInfo) {
418 #if !defined(_MSC_VER)
419 // We can spoil names of symbols with C linkage, so use an heuristic
420 // approach to check if the name should be demangled.
421 if (Name.substr(0, 2) == "_Z") {
423 char *DemangledName = __cxa_demangle(Name.c_str(), nullptr, nullptr, &status);
426 std::string Result = DemangledName;
431 if (!Name.empty() && Name.front() == '?') {
432 // Only do MSVC C++ demangling on symbols starting with '?'.
433 char DemangledName[1024] = {0};
434 DWORD result = ::UnDecorateSymbolName(
435 Name.c_str(), DemangledName, 1023,
436 UNDNAME_NO_ACCESS_SPECIFIERS | // Strip public, private, protected
437 UNDNAME_NO_ALLOCATION_LANGUAGE | // Strip __thiscall, __stdcall, etc
438 UNDNAME_NO_THROW_SIGNATURES | // Strip throw() specifications
439 UNDNAME_NO_MEMBER_TYPE | // Strip virtual, static, etc specifiers
440 UNDNAME_NO_MS_KEYWORDS | // Strip all MS extension keywords
441 UNDNAME_NO_FUNCTION_RETURNS); // Strip function return types
442 return (result == 0) ? Name : std::string(DemangledName);
445 if (ModInfo->isWin32Module())
446 return std::string(demanglePE32ExternCFunc(Name));
450 } // namespace symbolize