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 "LLVMSymbolize.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/Config/config.h"
17 #include "llvm/Object/ELFObjectFile.h"
18 #include "llvm/Object/MachO.h"
19 #include "llvm/Support/Casting.h"
20 #include "llvm/Support/Compression.h"
21 #include "llvm/Support/DataExtractor.h"
22 #include "llvm/Support/FileSystem.h"
23 #include "llvm/Support/MemoryBuffer.h"
24 #include "llvm/Support/Path.h"
31 static bool error(error_code ec) {
34 errs() << "LLVMSymbolizer: error reading file: " << ec.message() << ".\n";
38 static DILineInfoSpecifier
39 getDILineInfoSpecifier(const LLVMSymbolizer::Options &Opts) {
40 return DILineInfoSpecifier(
41 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath,
42 Opts.PrintFunctions ? DILineInfoSpecifier::FunctionNameKind::LinkageName
43 : DILineInfoSpecifier::FunctionNameKind::None);
46 ModuleInfo::ModuleInfo(ObjectFile *Obj, DIContext *DICtx)
47 : Module(Obj), DebugInfoContext(DICtx) {
48 for (const SymbolRef &Symbol : Module->symbols()) {
51 bool NoSymbolTable = (Module->symbol_begin() == Module->symbol_end());
52 if (NoSymbolTable && Module->isELF()) {
53 // Fallback to dynamic symbol table, if regular symbol table is stripped.
54 std::pair<symbol_iterator, symbol_iterator> IDyn =
55 getELFDynamicSymbolIterators(Module);
56 for (symbol_iterator si = IDyn.first, se = IDyn.second; si != se; ++si) {
62 void ModuleInfo::addSymbol(const SymbolRef &Symbol) {
63 SymbolRef::Type SymbolType;
64 if (error(Symbol.getType(SymbolType)))
66 if (SymbolType != SymbolRef::ST_Function && SymbolType != SymbolRef::ST_Data)
68 uint64_t SymbolAddress;
69 if (error(Symbol.getAddress(SymbolAddress)) ||
70 SymbolAddress == UnknownAddressOrSize)
73 // Getting symbol size is linear for Mach-O files, so assume that symbol
74 // occupies the memory range up to the following symbol.
75 if (isa<MachOObjectFile>(Module))
77 else if (error(Symbol.getSize(SymbolSize)) ||
78 SymbolSize == UnknownAddressOrSize)
81 if (error(Symbol.getName(SymbolName)))
83 // Mach-O symbol table names have leading underscore, skip it.
84 if (Module->isMachO() && SymbolName.size() > 0 && SymbolName[0] == '_')
85 SymbolName = SymbolName.drop_front();
86 // FIXME: If a function has alias, there are two entries in symbol table
87 // with same address size. Make sure we choose the correct one.
88 SymbolMapTy &M = SymbolType == SymbolRef::ST_Function ? Functions : Objects;
89 SymbolDesc SD = { SymbolAddress, SymbolSize };
90 M.insert(std::make_pair(SD, SymbolName));
93 bool ModuleInfo::getNameFromSymbolTable(SymbolRef::Type Type, uint64_t Address,
94 std::string &Name, uint64_t &Addr,
95 uint64_t &Size) const {
96 const SymbolMapTy &M = Type == SymbolRef::ST_Function ? Functions : Objects;
99 SymbolDesc SD = { Address, Address };
100 SymbolMapTy::const_iterator it = M.upper_bound(SD);
104 if (it->first.Size != 0 && it->first.Addr + it->first.Size <= Address)
106 Name = it->second.str();
107 Addr = it->first.Addr;
108 Size = it->first.Size;
112 DILineInfo ModuleInfo::symbolizeCode(
113 uint64_t ModuleOffset, const LLVMSymbolizer::Options &Opts) const {
115 if (DebugInfoContext) {
116 LineInfo = DebugInfoContext->getLineInfoForAddress(
117 ModuleOffset, getDILineInfoSpecifier(Opts));
119 // Override function name from symbol table if necessary.
120 if (Opts.PrintFunctions && Opts.UseSymbolTable) {
121 std::string FunctionName;
122 uint64_t Start, Size;
123 if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset,
124 FunctionName, Start, Size)) {
125 LineInfo.FunctionName = FunctionName;
131 DIInliningInfo ModuleInfo::symbolizeInlinedCode(
132 uint64_t ModuleOffset, const LLVMSymbolizer::Options &Opts) const {
133 DIInliningInfo InlinedContext;
134 if (DebugInfoContext) {
135 InlinedContext = DebugInfoContext->getInliningInfoForAddress(
136 ModuleOffset, getDILineInfoSpecifier(Opts));
138 // Make sure there is at least one frame in context.
139 if (InlinedContext.getNumberOfFrames() == 0) {
140 InlinedContext.addFrame(DILineInfo());
142 // Override the function name in lower frame with name from symbol table.
143 if (Opts.PrintFunctions && Opts.UseSymbolTable) {
144 DIInliningInfo PatchedInlinedContext;
145 for (uint32_t i = 0, n = InlinedContext.getNumberOfFrames(); i < n; i++) {
146 DILineInfo LineInfo = InlinedContext.getFrame(i);
148 std::string FunctionName;
149 uint64_t Start, Size;
150 if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset,
151 FunctionName, Start, Size)) {
152 LineInfo.FunctionName = FunctionName;
155 PatchedInlinedContext.addFrame(LineInfo);
157 InlinedContext = PatchedInlinedContext;
159 return InlinedContext;
162 bool ModuleInfo::symbolizeData(uint64_t ModuleOffset, std::string &Name,
163 uint64_t &Start, uint64_t &Size) const {
164 return getNameFromSymbolTable(SymbolRef::ST_Data, ModuleOffset, Name, Start,
168 const char LLVMSymbolizer::kBadString[] = "??";
170 std::string LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
171 uint64_t ModuleOffset) {
172 ModuleInfo *Info = getOrCreateModuleInfo(ModuleName);
174 return printDILineInfo(DILineInfo());
175 if (Opts.PrintInlining) {
176 DIInliningInfo InlinedContext =
177 Info->symbolizeInlinedCode(ModuleOffset, Opts);
178 uint32_t FramesNum = InlinedContext.getNumberOfFrames();
179 assert(FramesNum > 0);
181 for (uint32_t i = 0; i < FramesNum; i++) {
182 DILineInfo LineInfo = InlinedContext.getFrame(i);
183 Result += printDILineInfo(LineInfo);
187 DILineInfo LineInfo = Info->symbolizeCode(ModuleOffset, Opts);
188 return printDILineInfo(LineInfo);
191 std::string LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
192 uint64_t ModuleOffset) {
193 std::string Name = kBadString;
196 if (Opts.UseSymbolTable) {
197 if (ModuleInfo *Info = getOrCreateModuleInfo(ModuleName)) {
198 if (Info->symbolizeData(ModuleOffset, Name, Start, Size) && Opts.Demangle)
199 Name = DemangleName(Name);
202 std::stringstream ss;
203 ss << Name << "\n" << Start << " " << Size << "\n";
207 void LLVMSymbolizer::flush() {
208 DeleteContainerSeconds(Modules);
209 BinaryForPath.clear();
210 ObjectFileForArch.clear();
213 static std::string getDarwinDWARFResourceForPath(const std::string &Path) {
214 StringRef Basename = sys::path::filename(Path);
215 const std::string &DSymDirectory = Path + ".dSYM";
216 SmallString<16> ResourceName = StringRef(DSymDirectory);
217 sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
218 sys::path::append(ResourceName, Basename);
219 return ResourceName.str();
222 static bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
223 std::unique_ptr<MemoryBuffer> MB;
224 if (MemoryBuffer::getFileOrSTDIN(Path, MB))
226 return !zlib::isAvailable() || CRCHash == zlib::crc32(MB->getBuffer());
229 static bool findDebugBinary(const std::string &OrigPath,
230 const std::string &DebuglinkName, uint32_t CRCHash,
231 std::string &Result) {
232 std::string OrigRealPath = OrigPath;
233 #if defined(HAVE_REALPATH)
234 if (char *RP = realpath(OrigPath.c_str(), nullptr)) {
239 SmallString<16> OrigDir(OrigRealPath);
240 llvm::sys::path::remove_filename(OrigDir);
241 SmallString<16> DebugPath = OrigDir;
242 // Try /path/to/original_binary/debuglink_name
243 llvm::sys::path::append(DebugPath, DebuglinkName);
244 if (checkFileCRC(DebugPath, CRCHash)) {
245 Result = DebugPath.str();
248 // Try /path/to/original_binary/.debug/debuglink_name
249 DebugPath = OrigRealPath;
250 llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
251 if (checkFileCRC(DebugPath, CRCHash)) {
252 Result = DebugPath.str();
255 // Try /usr/lib/debug/path/to/original_binary/debuglink_name
256 DebugPath = "/usr/lib/debug";
257 llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
259 if (checkFileCRC(DebugPath, CRCHash)) {
260 Result = DebugPath.str();
266 static bool getGNUDebuglinkContents(const Binary *Bin, std::string &DebugName,
268 const ObjectFile *Obj = dyn_cast<ObjectFile>(Bin);
271 for (const SectionRef &Section : Obj->sections()) {
273 Section.getName(Name);
274 Name = Name.substr(Name.find_first_not_of("._"));
275 if (Name == "gnu_debuglink") {
277 Section.getContents(Data);
278 DataExtractor DE(Data, Obj->isLittleEndian(), 0);
280 if (const char *DebugNameStr = DE.getCStr(&Offset)) {
281 // 4-byte align the offset.
282 Offset = (Offset + 3) & ~0x3;
283 if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
284 DebugName = DebugNameStr;
285 CRCHash = DE.getU32(&Offset);
295 LLVMSymbolizer::BinaryPair
296 LLVMSymbolizer::getOrCreateBinary(const std::string &Path) {
297 BinaryMapTy::iterator I = BinaryForPath.find(Path);
298 if (I != BinaryForPath.end())
300 Binary *Bin = nullptr;
301 Binary *DbgBin = nullptr;
302 ErrorOr<Binary *> BinaryOrErr = createBinary(Path);
303 if (!error(BinaryOrErr.getError())) {
304 std::unique_ptr<Binary> ParsedBinary(BinaryOrErr.get());
305 // Check if it's a universal binary.
306 Bin = ParsedBinary.get();
307 ParsedBinariesAndObjects.push_back(std::move(ParsedBinary));
308 if (Bin->isMachO() || Bin->isMachOUniversalBinary()) {
309 // On Darwin we may find DWARF in separate object file in
310 // resource directory.
311 const std::string &ResourcePath =
312 getDarwinDWARFResourceForPath(Path);
313 BinaryOrErr = createBinary(ResourcePath);
314 error_code EC = BinaryOrErr.getError();
315 if (EC != errc::no_such_file_or_directory && !error(EC)) {
316 DbgBin = BinaryOrErr.get();
317 ParsedBinariesAndObjects.push_back(std::unique_ptr<Binary>(DbgBin));
320 // Try to locate the debug binary using .gnu_debuglink section.
322 std::string DebuglinkName;
324 std::string DebugBinaryPath;
325 if (getGNUDebuglinkContents(Bin, DebuglinkName, CRCHash) &&
326 findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath)) {
327 BinaryOrErr = createBinary(DebugBinaryPath);
328 if (!error(BinaryOrErr.getError())) {
329 DbgBin = BinaryOrErr.get();
330 ParsedBinariesAndObjects.push_back(std::unique_ptr<Binary>(DbgBin));
337 BinaryPair Res = std::make_pair(Bin, DbgBin);
338 BinaryForPath[Path] = Res;
343 LLVMSymbolizer::getObjectFileFromBinary(Binary *Bin, const std::string &ArchName) {
346 ObjectFile *Res = nullptr;
347 if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(Bin)) {
348 ObjectFileForArchMapTy::iterator I = ObjectFileForArch.find(
349 std::make_pair(UB, ArchName));
350 if (I != ObjectFileForArch.end())
352 std::unique_ptr<ObjectFile> ParsedObj;
353 if (!UB->getObjectForArch(Triple(ArchName).getArch(), ParsedObj)) {
354 Res = ParsedObj.get();
355 ParsedBinariesAndObjects.push_back(std::move(ParsedObj));
357 ObjectFileForArch[std::make_pair(UB, ArchName)] = Res;
358 } else if (Bin->isObject()) {
359 Res = cast<ObjectFile>(Bin);
365 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) {
366 ModuleMapTy::iterator I = Modules.find(ModuleName);
367 if (I != Modules.end())
369 std::string BinaryName = ModuleName;
370 std::string ArchName = Opts.DefaultArch;
371 size_t ColonPos = ModuleName.find_last_of(':');
372 // Verify that substring after colon form a valid arch name.
373 if (ColonPos != std::string::npos) {
374 std::string ArchStr = ModuleName.substr(ColonPos + 1);
375 if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
376 BinaryName = ModuleName.substr(0, ColonPos);
380 BinaryPair Binaries = getOrCreateBinary(BinaryName);
381 ObjectFile *Obj = getObjectFileFromBinary(Binaries.first, ArchName);
382 ObjectFile *DbgObj = getObjectFileFromBinary(Binaries.second, ArchName);
385 // Failed to find valid object file.
386 Modules.insert(make_pair(ModuleName, (ModuleInfo *)nullptr));
389 DIContext *Context = DIContext::getDWARFContext(DbgObj);
391 ModuleInfo *Info = new ModuleInfo(Obj, Context);
392 Modules.insert(make_pair(ModuleName, Info));
396 std::string LLVMSymbolizer::printDILineInfo(DILineInfo LineInfo) const {
397 // By default, DILineInfo contains "<invalid>" for function/filename it
398 // cannot fetch. We replace it to "??" to make our output closer to addr2line.
399 static const std::string kDILineInfoBadString = "<invalid>";
400 std::stringstream Result;
401 if (Opts.PrintFunctions) {
402 std::string FunctionName = LineInfo.FunctionName;
403 if (FunctionName == kDILineInfoBadString)
404 FunctionName = kBadString;
405 else if (Opts.Demangle)
406 FunctionName = DemangleName(FunctionName);
407 Result << FunctionName << "\n";
409 std::string Filename = LineInfo.FileName;
410 if (Filename == kDILineInfoBadString)
411 Filename = kBadString;
412 Result << Filename << ":" << LineInfo.Line << ":" << LineInfo.Column << "\n";
416 #if !defined(_MSC_VER)
417 // Assume that __cxa_demangle is provided by libcxxabi (except for Windows).
418 extern "C" char *__cxa_demangle(const char *mangled_name, char *output_buffer,
419 size_t *length, int *status);
422 std::string LLVMSymbolizer::DemangleName(const std::string &Name) {
423 #if !defined(_MSC_VER)
424 // We can spoil names of symbols with C linkage, so use an heuristic
425 // approach to check if the name should be demangled.
426 if (Name.substr(0, 2) != "_Z")
429 char *DemangledName = __cxa_demangle(Name.c_str(), nullptr, nullptr, &status);
432 std::string Result = DemangledName;
440 } // namespace symbolize