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 // By default, DILineInfo contains "<invalid>" for function/filename it
60 // cannot fetch. We replace it to "??" to make our output closer to addr2line.
61 static const char kDILineInfoBadString[] = "<invalid>";
63 const char LLVMSymbolizer::kBadString[] = "??";
65 std::string LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
66 uint64_t ModuleOffset) {
67 SymbolizableModule *Info = getOrCreateModuleInfo(ModuleName);
69 return printDILineInfo(DILineInfo(), Info);
71 // If the user is giving us relative addresses, add the preferred base of the
72 // object to the offset before we do the query. It's what DIContext expects.
73 if (Opts.RelativeAddresses)
74 ModuleOffset += Info->getModulePreferredBase();
76 if (Opts.PrintInlining) {
77 DIInliningInfo InlinedContext = Info->symbolizeInlinedCode(
78 ModuleOffset, Opts.PrintFunctions, Opts.UseSymbolTable);
79 uint32_t FramesNum = InlinedContext.getNumberOfFrames();
80 assert(FramesNum > 0);
82 for (uint32_t i = 0; i < FramesNum; i++) {
83 DILineInfo LineInfo = InlinedContext.getFrame(i);
84 Result += printDILineInfo(LineInfo, Info);
88 DILineInfo LineInfo = Info->symbolizeCode(ModuleOffset, Opts.PrintFunctions,
90 return printDILineInfo(LineInfo, Info);
93 std::string LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
94 uint64_t ModuleOffset) {
95 if (Opts.UseSymbolTable) {
96 if (SymbolizableModule *Info = getOrCreateModuleInfo(ModuleName)) {
97 // If the user is giving us relative addresses, add the preferred base of
98 // the object to the offset before we do the query. It's what DIContext
100 if (Opts.RelativeAddresses)
101 ModuleOffset += Info->getModulePreferredBase();
102 DIGlobal Global = Info->symbolizeData(ModuleOffset);
103 return printDIGlobal(Global, Info);
106 return printDIGlobal(DIGlobal(), nullptr);
109 void LLVMSymbolizer::flush() {
111 ObjectPairForPathArch.clear();
112 ObjectFileForArch.clear();
115 // For Path="/path/to/foo" and Basename="foo" assume that debug info is in
116 // /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
117 // For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
118 // /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
120 std::string getDarwinDWARFResourceForPath(
121 const std::string &Path, const std::string &Basename) {
122 SmallString<16> ResourceName = StringRef(Path);
123 if (sys::path::extension(Path) != ".dSYM") {
124 ResourceName += ".dSYM";
126 sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
127 sys::path::append(ResourceName, Basename);
128 return ResourceName.str();
131 static bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
132 ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
133 MemoryBuffer::getFileOrSTDIN(Path);
136 return !zlib::isAvailable() || CRCHash == zlib::crc32(MB.get()->getBuffer());
139 static bool findDebugBinary(const std::string &OrigPath,
140 const std::string &DebuglinkName, uint32_t CRCHash,
141 std::string &Result) {
142 std::string OrigRealPath = OrigPath;
143 #if defined(HAVE_REALPATH)
144 if (char *RP = realpath(OrigPath.c_str(), nullptr)) {
149 SmallString<16> OrigDir(OrigRealPath);
150 llvm::sys::path::remove_filename(OrigDir);
151 SmallString<16> DebugPath = OrigDir;
152 // Try /path/to/original_binary/debuglink_name
153 llvm::sys::path::append(DebugPath, DebuglinkName);
154 if (checkFileCRC(DebugPath, CRCHash)) {
155 Result = DebugPath.str();
158 // Try /path/to/original_binary/.debug/debuglink_name
159 DebugPath = OrigRealPath;
160 llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
161 if (checkFileCRC(DebugPath, CRCHash)) {
162 Result = DebugPath.str();
165 // Try /usr/lib/debug/path/to/original_binary/debuglink_name
166 DebugPath = "/usr/lib/debug";
167 llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
169 if (checkFileCRC(DebugPath, CRCHash)) {
170 Result = DebugPath.str();
176 static bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
180 for (const SectionRef &Section : Obj->sections()) {
182 Section.getName(Name);
183 Name = Name.substr(Name.find_first_not_of("._"));
184 if (Name == "gnu_debuglink") {
186 Section.getContents(Data);
187 DataExtractor DE(Data, Obj->isLittleEndian(), 0);
189 if (const char *DebugNameStr = DE.getCStr(&Offset)) {
190 // 4-byte align the offset.
191 Offset = (Offset + 3) & ~0x3;
192 if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
193 DebugName = DebugNameStr;
194 CRCHash = DE.getU32(&Offset);
205 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
206 const MachOObjectFile *Obj) {
207 ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
208 ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
209 if (dbg_uuid.empty() || bin_uuid.empty())
211 return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
214 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
215 const MachOObjectFile *MachExeObj, const std::string &ArchName) {
216 // On Darwin we may find DWARF in separate object file in
217 // resource directory.
218 std::vector<std::string> DsymPaths;
219 StringRef Filename = sys::path::filename(ExePath);
220 DsymPaths.push_back(getDarwinDWARFResourceForPath(ExePath, Filename));
221 for (const auto &Path : Opts.DsymHints) {
222 DsymPaths.push_back(getDarwinDWARFResourceForPath(Path, Filename));
224 for (const auto &path : DsymPaths) {
225 ErrorOr<OwningBinary<Binary>> BinaryOrErr = createBinary(path);
226 std::error_code EC = BinaryOrErr.getError();
227 if (EC != errc::no_such_file_or_directory && !error(EC)) {
228 OwningBinary<Binary> B = std::move(BinaryOrErr.get());
230 getObjectFileFromBinary(B.getBinary(), ArchName);
231 const MachOObjectFile *MachDbgObj =
232 dyn_cast<const MachOObjectFile>(DbgObj);
233 if (!MachDbgObj) continue;
234 if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj)) {
235 addOwningBinary(std::move(B));
243 LLVMSymbolizer::ObjectPair
244 LLVMSymbolizer::getOrCreateObjects(const std::string &Path,
245 const std::string &ArchName) {
246 const auto &I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
247 if (I != ObjectPairForPathArch.end())
249 ObjectFile *Obj = nullptr;
250 ObjectFile *DbgObj = nullptr;
251 ErrorOr<OwningBinary<Binary>> BinaryOrErr = createBinary(Path);
252 if (!error(BinaryOrErr.getError())) {
253 OwningBinary<Binary> &B = BinaryOrErr.get();
254 Obj = getObjectFileFromBinary(B.getBinary(), ArchName);
256 ObjectPair Res = std::make_pair(nullptr, nullptr);
257 ObjectPairForPathArch[std::make_pair(Path, ArchName)] = Res;
260 addOwningBinary(std::move(B));
261 if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
262 DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
263 // Try to locate the debug binary using .gnu_debuglink section.
265 std::string DebuglinkName;
267 std::string DebugBinaryPath;
268 if (getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash) &&
269 findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath)) {
270 BinaryOrErr = createBinary(DebugBinaryPath);
271 if (!error(BinaryOrErr.getError())) {
272 OwningBinary<Binary> B = std::move(BinaryOrErr.get());
273 DbgObj = getObjectFileFromBinary(B.getBinary(), ArchName);
274 addOwningBinary(std::move(B));
281 ObjectPair Res = std::make_pair(Obj, DbgObj);
282 ObjectPairForPathArch[std::make_pair(Path, ArchName)] = Res;
287 LLVMSymbolizer::getObjectFileFromBinary(Binary *Bin,
288 const std::string &ArchName) {
291 ObjectFile *Res = nullptr;
292 if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(Bin)) {
293 const auto &I = ObjectFileForArch.find(
294 std::make_pair(UB, ArchName));
295 if (I != ObjectFileForArch.end())
297 ErrorOr<std::unique_ptr<ObjectFile>> ParsedObj =
298 UB->getObjectForArch(ArchName);
300 Res = ParsedObj.get().get();
301 ParsedBinariesAndObjects.push_back(std::move(ParsedObj.get()));
303 ObjectFileForArch[std::make_pair(UB, ArchName)] = Res;
304 } else if (Bin->isObject()) {
305 Res = cast<ObjectFile>(Bin);
311 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) {
312 const auto &I = Modules.find(ModuleName);
313 if (I != Modules.end())
314 return I->second.get();
315 std::string BinaryName = ModuleName;
316 std::string ArchName = Opts.DefaultArch;
317 size_t ColonPos = ModuleName.find_last_of(':');
318 // Verify that substring after colon form a valid arch name.
319 if (ColonPos != std::string::npos) {
320 std::string ArchStr = ModuleName.substr(ColonPos + 1);
321 if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
322 BinaryName = ModuleName.substr(0, ColonPos);
326 ObjectPair Objects = getOrCreateObjects(BinaryName, ArchName);
328 if (!Objects.first) {
329 // Failed to find valid object file.
330 Modules.insert(std::make_pair(ModuleName, nullptr));
333 std::unique_ptr<DIContext> Context;
334 if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
335 // If this is a COFF object, assume it contains PDB debug information. If
336 // we don't find any we will fall back to the DWARF case.
337 std::unique_ptr<IPDBSession> Session;
338 PDB_ErrorCode Error = loadDataForEXE(PDB_ReaderType::DIA,
339 Objects.first->getFileName(), Session);
340 if (Error == PDB_ErrorCode::Success) {
341 Context.reset(new PDBContext(*CoffObject, std::move(Session)));
345 Context.reset(new DWARFContextInMemory(*Objects.second));
348 SymbolizableObjectFile::create(Objects.first, std::move(Context));
349 if (error(ErrOrInfo.getError())) {
350 Modules.insert(std::make_pair(ModuleName, nullptr));
353 SymbolizableModule *Res = ErrOrInfo.get().get();
354 Modules.insert(std::make_pair(ModuleName, std::move(ErrOrInfo.get())));
359 LLVMSymbolizer::printDILineInfo(DILineInfo LineInfo,
360 const SymbolizableModule *ModInfo) const {
361 std::stringstream Result;
362 if (Opts.PrintFunctions != FunctionNameKind::None) {
363 std::string FunctionName = LineInfo.FunctionName;
364 if (FunctionName == kDILineInfoBadString)
365 FunctionName = kBadString;
366 else if (Opts.Demangle)
367 FunctionName = DemangleName(FunctionName, ModInfo);
368 Result << FunctionName << "\n";
370 std::string Filename = LineInfo.FileName;
371 if (Filename == kDILineInfoBadString)
372 Filename = kBadString;
373 Result << Filename << ":" << LineInfo.Line << ":" << LineInfo.Column << "\n";
378 LLVMSymbolizer::printDIGlobal(DIGlobal Global,
379 const SymbolizableModule *ModInfo) const {
380 std::stringstream Result;
381 std::string Name = Global.Name;
382 if (Name == kDILineInfoBadString)
384 else if (Opts.Demangle)
385 Name = DemangleName(Name, ModInfo);
386 Result << Name << "\n";
387 Result << Global.Start << " " << Global.Size << "\n";
391 // Undo these various manglings for Win32 extern "C" functions:
394 // fastcall - @foo@12
395 // vectorcall - foo@@12
396 // These are all different linkage names for 'foo'.
397 static StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
398 // Remove any '_' or '@' prefix.
399 char Front = SymbolName.empty() ? '\0' : SymbolName[0];
400 if (Front == '_' || Front == '@')
401 SymbolName = SymbolName.drop_front();
403 // Remove any '@[0-9]+' suffix.
405 size_t AtPos = SymbolName.rfind('@');
406 if (AtPos != StringRef::npos &&
407 std::all_of(SymbolName.begin() + AtPos + 1, SymbolName.end(),
408 [](char C) { return C >= '0' && C <= '9'; })) {
409 SymbolName = SymbolName.substr(0, AtPos);
413 // Remove any ending '@' for vectorcall.
414 if (SymbolName.endswith("@"))
415 SymbolName = SymbolName.drop_back();
420 #if !defined(_MSC_VER)
421 // Assume that __cxa_demangle is provided by libcxxabi (except for Windows).
422 extern "C" char *__cxa_demangle(const char *mangled_name, char *output_buffer,
423 size_t *length, int *status);
426 std::string LLVMSymbolizer::DemangleName(const std::string &Name,
427 const SymbolizableModule *ModInfo) {
428 #if !defined(_MSC_VER)
429 // We can spoil names of symbols with C linkage, so use an heuristic
430 // approach to check if the name should be demangled.
431 if (Name.substr(0, 2) == "_Z") {
433 char *DemangledName = __cxa_demangle(Name.c_str(), nullptr, nullptr, &status);
436 std::string Result = DemangledName;
441 if (!Name.empty() && Name.front() == '?') {
442 // Only do MSVC C++ demangling on symbols starting with '?'.
443 char DemangledName[1024] = {0};
444 DWORD result = ::UnDecorateSymbolName(
445 Name.c_str(), DemangledName, 1023,
446 UNDNAME_NO_ACCESS_SPECIFIERS | // Strip public, private, protected
447 UNDNAME_NO_ALLOCATION_LANGUAGE | // Strip __thiscall, __stdcall, etc
448 UNDNAME_NO_THROW_SIGNATURES | // Strip throw() specifications
449 UNDNAME_NO_MEMBER_TYPE | // Strip virtual, static, etc specifiers
450 UNDNAME_NO_MS_KEYWORDS | // Strip all MS extension keywords
451 UNDNAME_NO_FUNCTION_RETURNS); // Strip function return types
452 return (result == 0) ? Name : std::string(DemangledName);
455 if (ModInfo && ModInfo->isWin32Module())
456 return std::string(demanglePE32ExternCFunc(Name));
460 } // namespace symbolize