1296465Sdelphij//===-- LLVMSymbolize.cpp -------------------------------------------------===// 2110010Smarkm// 3110010Smarkm// The LLVM Compiler Infrastructure 4142429Snectar// 5110010Smarkm// This file is distributed under the University of Illinois Open Source 6110010Smarkm// License. See LICENSE.TXT for details. 7110010Smarkm// 8110010Smarkm//===----------------------------------------------------------------------===// 9110010Smarkm// 10110010Smarkm// Implementation for LLVM symbolization library. 11110010Smarkm// 12110010Smarkm//===----------------------------------------------------------------------===// 13110010Smarkm 14110010Smarkm#include "LLVMSymbolize.h" 15110010Smarkm#include "llvm/ADT/STLExtras.h" 16110010Smarkm#include "llvm/Config/config.h" 17110010Smarkm#include "llvm/Object/MachO.h" 18110010Smarkm#include "llvm/Support/Casting.h" 19110010Smarkm#include "llvm/Support/Compression.h" 20215698Ssimon#include "llvm/Support/DataExtractor.h" 21215698Ssimon#include "llvm/Support/FileSystem.h" 22215698Ssimon#include "llvm/Support/MemoryBuffer.h" 23215698Ssimon#include "llvm/Support/Path.h" 24215698Ssimon 25110010Smarkm#include <sstream> 26110010Smarkm#include <stdlib.h> 27110010Smarkm 28110010Smarkmnamespace llvm { 29110010Smarkmnamespace symbolize { 30110010Smarkm 31110010Smarkmstatic bool error(error_code ec) { 32110010Smarkm if (!ec) 33110010Smarkm return false; 34110010Smarkm errs() << "LLVMSymbolizer: error reading file: " << ec.message() << ".\n"; 35110010Smarkm return true; 36110010Smarkm} 37110010Smarkm 38110010Smarkmstatic uint32_t 39110010SmarkmgetDILineInfoSpecifierFlags(const LLVMSymbolizer::Options &Opts) { 40110010Smarkm uint32_t Flags = llvm::DILineInfoSpecifier::FileLineInfo | 41279265Sdelphij llvm::DILineInfoSpecifier::AbsoluteFilePath; 42279265Sdelphij if (Opts.PrintFunctions) 43110010Smarkm Flags |= llvm::DILineInfoSpecifier::FunctionName; 44110010Smarkm return Flags; 45215698Ssimon} 46215698Ssimon 47215698Ssimonstatic void patchFunctionNameInDILineInfo(const std::string &NewFunctionName, 48215698Ssimon DILineInfo &LineInfo) { 49142429Snectar std::string FileName = LineInfo.getFileName(); 50215698Ssimon LineInfo = DILineInfo(StringRef(FileName), StringRef(NewFunctionName), 51142429Snectar LineInfo.getLine(), LineInfo.getColumn()); 52142429Snectar} 53279265Sdelphij 54279265SdelphijModuleInfo::ModuleInfo(ObjectFile *Obj, DIContext *DICtx) 55279265Sdelphij : Module(Obj), DebugInfoContext(DICtx) { 56110010Smarkm error_code ec; 57279265Sdelphij for (symbol_iterator si = Module->begin_symbols(), se = Module->end_symbols(); 58279265Sdelphij si != se; si.increment(ec)) { 59279265Sdelphij if (error(ec)) 60279265Sdelphij return; 61279265Sdelphij SymbolRef::Type SymbolType; 62279265Sdelphij if (error(si->getType(SymbolType))) 63215698Ssimon continue; 64279265Sdelphij if (SymbolType != SymbolRef::ST_Function && 65279265Sdelphij SymbolType != SymbolRef::ST_Data) 66279265Sdelphij continue; 67279265Sdelphij uint64_t SymbolAddress; 68279265Sdelphij if (error(si->getAddress(SymbolAddress)) || 69215698Ssimon SymbolAddress == UnknownAddressOrSize) 70279265Sdelphij continue; 71110010Smarkm uint64_t SymbolSize; 72110010Smarkm // Getting symbol size is linear for Mach-O files, so assume that symbol 73110010Smarkm // occupies the memory range up to the following symbol. 74110010Smarkm if (isa<MachOObjectFile>(Obj)) 75110010Smarkm SymbolSize = 0; 76110010Smarkm else if (error(si->getSize(SymbolSize)) || 77110010Smarkm SymbolSize == UnknownAddressOrSize) 78110010Smarkm continue; 79110010Smarkm StringRef SymbolName; 80110010Smarkm if (error(si->getName(SymbolName))) 81110010Smarkm continue; 82110010Smarkm // Mach-O symbol table names have leading underscore, skip it. 83110010Smarkm if (Module->isMachO() && SymbolName.size() > 0 && SymbolName[0] == '_') 84110010Smarkm SymbolName = SymbolName.drop_front(); 85110010Smarkm // FIXME: If a function has alias, there are two entries in symbol table 86110010Smarkm // with same address size. Make sure we choose the correct one. 87110010Smarkm SymbolMapTy &M = SymbolType == SymbolRef::ST_Function ? Functions : Objects; 88110010Smarkm SymbolDesc SD = { SymbolAddress, SymbolSize }; 89110010Smarkm M.insert(std::make_pair(SD, SymbolName)); 90110010Smarkm } 91110010Smarkm} 92110010Smarkm 93110010Smarkmbool ModuleInfo::getNameFromSymbolTable(SymbolRef::Type Type, uint64_t Address, 94110010Smarkm std::string &Name, uint64_t &Addr, 95110010Smarkm uint64_t &Size) const { 96110010Smarkm const SymbolMapTy &M = Type == SymbolRef::ST_Function ? Functions : Objects; 97110010Smarkm if (M.empty()) 98110010Smarkm return false; 99110010Smarkm SymbolDesc SD = { Address, Address }; 100110010Smarkm SymbolMapTy::const_iterator it = M.upper_bound(SD); 101110010Smarkm if (it == M.begin()) 102110010Smarkm return false; 103110010Smarkm --it; 104110010Smarkm if (it->first.Size != 0 && it->first.Addr + it->first.Size <= Address) 105110010Smarkm return false; 106110010Smarkm Name = it->second.str(); 107110010Smarkm Addr = it->first.Addr; 108110010Smarkm Size = it->first.Size; 109110010Smarkm return true; 110110010Smarkm} 111110010Smarkm 112110010SmarkmDILineInfo ModuleInfo::symbolizeCode( 113110010Smarkm uint64_t ModuleOffset, const LLVMSymbolizer::Options &Opts) const { 114110010Smarkm DILineInfo LineInfo; 115110010Smarkm if (DebugInfoContext) { 116110010Smarkm LineInfo = DebugInfoContext->getLineInfoForAddress( 117110010Smarkm ModuleOffset, getDILineInfoSpecifierFlags(Opts)); 118110010Smarkm } 119110010Smarkm // Override function name from symbol table if necessary. 120110010Smarkm if (Opts.PrintFunctions && Opts.UseSymbolTable) { 121110010Smarkm std::string FunctionName; 122110010Smarkm uint64_t Start, Size; 123110010Smarkm if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset, 124110010Smarkm FunctionName, Start, Size)) { 125110010Smarkm patchFunctionNameInDILineInfo(FunctionName, LineInfo); 126110010Smarkm } 127110010Smarkm } 128110010Smarkm return LineInfo; 129110010Smarkm} 130110010Smarkm 131110010SmarkmDIInliningInfo ModuleInfo::symbolizeInlinedCode( 132110010Smarkm uint64_t ModuleOffset, const LLVMSymbolizer::Options &Opts) const { 133142429Snectar DIInliningInfo InlinedContext; 134110010Smarkm if (DebugInfoContext) { 135110655Snectar InlinedContext = DebugInfoContext->getInliningInfoForAddress( 136296465Sdelphij ModuleOffset, getDILineInfoSpecifierFlags(Opts)); 137215698Ssimon } 138215698Ssimon // Make sure there is at least one frame in context. 139215698Ssimon if (InlinedContext.getNumberOfFrames() == 0) { 140215698Ssimon InlinedContext.addFrame(DILineInfo()); 141110010Smarkm } 142142429Snectar // Override the function name in lower frame with name from symbol table. 143110010Smarkm if (Opts.PrintFunctions && Opts.UseSymbolTable) { 144110010Smarkm DIInliningInfo PatchedInlinedContext; 145110010Smarkm for (uint32_t i = 0, n = InlinedContext.getNumberOfFrames(); i < n; i++) { 146110010Smarkm DILineInfo LineInfo = InlinedContext.getFrame(i); 147110010Smarkm if (i == n - 1) { 148110010Smarkm std::string FunctionName; 149110010Smarkm uint64_t Start, Size; 150110010Smarkm if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset, 151110010Smarkm FunctionName, Start, Size)) { 152110010Smarkm patchFunctionNameInDILineInfo(FunctionName, LineInfo); 153142429Snectar } 154110010Smarkm } 155111151Snectar PatchedInlinedContext.addFrame(LineInfo); 156110010Smarkm } 157110010Smarkm InlinedContext = PatchedInlinedContext; 158110010Smarkm } 159110010Smarkm return InlinedContext; 160110010Smarkm} 161110010Smarkm 162142429Snectarbool ModuleInfo::symbolizeData(uint64_t ModuleOffset, std::string &Name, 163110010Smarkm uint64_t &Start, uint64_t &Size) const { 164279265Sdelphij return getNameFromSymbolTable(SymbolRef::ST_Data, ModuleOffset, Name, Start, 165279265Sdelphij Size); 166110010Smarkm} 167110010Smarkm 168142429Snectarconst char LLVMSymbolizer::kBadString[] = "??"; 169110010Smarkm 170110010Smarkmstd::string LLVMSymbolizer::symbolizeCode(const std::string &ModuleName, 171110010Smarkm uint64_t ModuleOffset) { 172142429Snectar ModuleInfo *Info = getOrCreateModuleInfo(ModuleName); 173110010Smarkm if (Info == 0) 174110010Smarkm return printDILineInfo(DILineInfo()); 175110010Smarkm if (Opts.PrintInlining) { 176110010Smarkm DIInliningInfo InlinedContext = 177142429Snectar Info->symbolizeInlinedCode(ModuleOffset, Opts); 178110010Smarkm uint32_t FramesNum = InlinedContext.getNumberOfFrames(); 179110010Smarkm assert(FramesNum > 0); 180110010Smarkm std::string Result; 181110010Smarkm for (uint32_t i = 0; i < FramesNum; i++) { 182142429Snectar DILineInfo LineInfo = InlinedContext.getFrame(i); 183110010Smarkm Result += printDILineInfo(LineInfo); 184110010Smarkm } 185142429Snectar return Result; 186110010Smarkm } 187110010Smarkm DILineInfo LineInfo = Info->symbolizeCode(ModuleOffset, Opts); 188142429Snectar return printDILineInfo(LineInfo); 189110010Smarkm} 190110010Smarkm 191110010Smarkmstd::string LLVMSymbolizer::symbolizeData(const std::string &ModuleName, 192142429Snectar uint64_t ModuleOffset) { 193110010Smarkm std::string Name = kBadString; 194110010Smarkm uint64_t Start = 0; 195110010Smarkm uint64_t Size = 0; 196142429Snectar if (Opts.UseSymbolTable) { 197142429Snectar if (ModuleInfo *Info = getOrCreateModuleInfo(ModuleName)) { 198110010Smarkm if (Info->symbolizeData(ModuleOffset, Name, Start, Size) && Opts.Demangle) 199142429Snectar Name = DemangleGlobalName(Name); 200110010Smarkm } 201215698Ssimon } 202110010Smarkm std::stringstream ss; 203142429Snectar ss << Name << "\n" << Start << " " << Size << "\n"; 204110010Smarkm return ss.str(); 205110010Smarkm} 206110010Smarkm 207110010Smarkmvoid LLVMSymbolizer::flush() { 208142429Snectar DeleteContainerSeconds(Modules); 209111151Snectar DeleteContainerPointers(ParsedBinariesAndObjects); 210111151Snectar BinaryForPath.clear(); 211111151Snectar ObjectFileForArch.clear(); 212111151Snectar} 213111151Snectar 214110010Smarkmstatic std::string getDarwinDWARFResourceForPath(const std::string &Path) { 215110010Smarkm StringRef Basename = sys::path::filename(Path); 216110010Smarkm const std::string &DSymDirectory = Path + ".dSYM"; 217110010Smarkm SmallString<16> ResourceName = StringRef(DSymDirectory); 218110010Smarkm sys::path::append(ResourceName, "Contents", "Resources", "DWARF"); 219215698Ssimon sys::path::append(ResourceName, Basename); 220215698Ssimon return ResourceName.str(); 221110010Smarkm} 222142429Snectar 223110010Smarkmstatic bool checkFileCRC(StringRef Path, uint32_t CRCHash) { 224110010Smarkm OwningPtr<MemoryBuffer> MB; 225110010Smarkm if (MemoryBuffer::getFileOrSTDIN(Path, MB)) 226110010Smarkm return false; 227142429Snectar return !zlib::isAvailable() || CRCHash == zlib::crc32(MB->getBuffer()); 228142429Snectar} 229 230static bool findDebugBinary(const std::string &OrigPath, 231 const std::string &DebuglinkName, uint32_t CRCHash, 232 std::string &Result) { 233 std::string OrigRealPath = OrigPath; 234#if defined(HAVE_REALPATH) 235 if (char *RP = realpath(OrigPath.c_str(), NULL)) { 236 OrigRealPath = RP; 237 free(RP); 238 } 239#endif 240 SmallString<16> OrigDir(OrigRealPath); 241 llvm::sys::path::remove_filename(OrigDir); 242 SmallString<16> DebugPath = OrigDir; 243 // Try /path/to/original_binary/debuglink_name 244 llvm::sys::path::append(DebugPath, DebuglinkName); 245 if (checkFileCRC(DebugPath, CRCHash)) { 246 Result = DebugPath.str(); 247 return true; 248 } 249 // Try /path/to/original_binary/.debug/debuglink_name 250 DebugPath = OrigRealPath; 251 llvm::sys::path::append(DebugPath, ".debug", DebuglinkName); 252 if (checkFileCRC(DebugPath, CRCHash)) { 253 Result = DebugPath.str(); 254 return true; 255 } 256 // Try /usr/lib/debug/path/to/original_binary/debuglink_name 257 DebugPath = "/usr/lib/debug"; 258 llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir), 259 DebuglinkName); 260 if (checkFileCRC(DebugPath, CRCHash)) { 261 Result = DebugPath.str(); 262 return true; 263 } 264 return false; 265} 266 267static bool getGNUDebuglinkContents(const Binary *Bin, std::string &DebugName, 268 uint32_t &CRCHash) { 269 const ObjectFile *Obj = dyn_cast<ObjectFile>(Bin); 270 if (!Obj) 271 return false; 272 error_code EC; 273 for (section_iterator I = Obj->begin_sections(), E = Obj->end_sections(); 274 I != E; I.increment(EC)) { 275 StringRef Name; 276 I->getName(Name); 277 Name = Name.substr(Name.find_first_not_of("._")); 278 if (Name == "gnu_debuglink") { 279 StringRef Data; 280 I->getContents(Data); 281 DataExtractor DE(Data, Obj->isLittleEndian(), 0); 282 uint32_t Offset = 0; 283 if (const char *DebugNameStr = DE.getCStr(&Offset)) { 284 // 4-byte align the offset. 285 Offset = (Offset + 3) & ~0x3; 286 if (DE.isValidOffsetForDataOfSize(Offset, 4)) { 287 DebugName = DebugNameStr; 288 CRCHash = DE.getU32(&Offset); 289 return true; 290 } 291 } 292 break; 293 } 294 } 295 return false; 296} 297 298LLVMSymbolizer::BinaryPair 299LLVMSymbolizer::getOrCreateBinary(const std::string &Path) { 300 BinaryMapTy::iterator I = BinaryForPath.find(Path); 301 if (I != BinaryForPath.end()) 302 return I->second; 303 Binary *Bin = 0; 304 Binary *DbgBin = 0; 305 OwningPtr<Binary> ParsedBinary; 306 OwningPtr<Binary> ParsedDbgBinary; 307 if (!error(createBinary(Path, ParsedBinary))) { 308 // Check if it's a universal binary. 309 Bin = ParsedBinary.take(); 310 ParsedBinariesAndObjects.push_back(Bin); 311 if (Bin->isMachO() || Bin->isMachOUniversalBinary()) { 312 // On Darwin we may find DWARF in separate object file in 313 // resource directory. 314 const std::string &ResourcePath = 315 getDarwinDWARFResourceForPath(Path); 316 bool ResourceFileExists = false; 317 if (!sys::fs::exists(ResourcePath, ResourceFileExists) && 318 ResourceFileExists && 319 !error(createBinary(ResourcePath, ParsedDbgBinary))) { 320 DbgBin = ParsedDbgBinary.take(); 321 ParsedBinariesAndObjects.push_back(DbgBin); 322 } 323 } 324 // Try to locate the debug binary using .gnu_debuglink section. 325 if (DbgBin == 0) { 326 std::string DebuglinkName; 327 uint32_t CRCHash; 328 std::string DebugBinaryPath; 329 if (getGNUDebuglinkContents(Bin, DebuglinkName, CRCHash) && 330 findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath) && 331 !error(createBinary(DebugBinaryPath, ParsedDbgBinary))) { 332 DbgBin = ParsedDbgBinary.take(); 333 ParsedBinariesAndObjects.push_back(DbgBin); 334 } 335 } 336 } 337 if (DbgBin == 0) 338 DbgBin = Bin; 339 BinaryPair Res = std::make_pair(Bin, DbgBin); 340 BinaryForPath[Path] = Res; 341 return Res; 342} 343 344ObjectFile * 345LLVMSymbolizer::getObjectFileFromBinary(Binary *Bin, const std::string &ArchName) { 346 if (Bin == 0) 347 return 0; 348 ObjectFile *Res = 0; 349 if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(Bin)) { 350 ObjectFileForArchMapTy::iterator I = ObjectFileForArch.find( 351 std::make_pair(UB, ArchName)); 352 if (I != ObjectFileForArch.end()) 353 return I->second; 354 OwningPtr<ObjectFile> ParsedObj; 355 if (!UB->getObjectForArch(Triple(ArchName).getArch(), ParsedObj)) { 356 Res = ParsedObj.take(); 357 ParsedBinariesAndObjects.push_back(Res); 358 } 359 ObjectFileForArch[std::make_pair(UB, ArchName)] = Res; 360 } else if (Bin->isObject()) { 361 Res = cast<ObjectFile>(Bin); 362 } 363 return Res; 364} 365 366ModuleInfo * 367LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) { 368 ModuleMapTy::iterator I = Modules.find(ModuleName); 369 if (I != Modules.end()) 370 return I->second; 371 std::string BinaryName = ModuleName; 372 std::string ArchName = Opts.DefaultArch; 373 size_t ColonPos = ModuleName.find_last_of(':'); 374 // Verify that substring after colon form a valid arch name. 375 if (ColonPos != std::string::npos) { 376 std::string ArchStr = ModuleName.substr(ColonPos + 1); 377 if (Triple(ArchStr).getArch() != Triple::UnknownArch) { 378 BinaryName = ModuleName.substr(0, ColonPos); 379 ArchName = ArchStr; 380 } 381 } 382 BinaryPair Binaries = getOrCreateBinary(BinaryName); 383 ObjectFile *Obj = getObjectFileFromBinary(Binaries.first, ArchName); 384 ObjectFile *DbgObj = getObjectFileFromBinary(Binaries.second, ArchName); 385 386 if (Obj == 0) { 387 // Failed to find valid object file. 388 Modules.insert(make_pair(ModuleName, (ModuleInfo *)0)); 389 return 0; 390 } 391 DIContext *Context = DIContext::getDWARFContext(DbgObj); 392 assert(Context); 393 ModuleInfo *Info = new ModuleInfo(Obj, Context); 394 Modules.insert(make_pair(ModuleName, Info)); 395 return Info; 396} 397 398std::string LLVMSymbolizer::printDILineInfo(DILineInfo LineInfo) const { 399 // By default, DILineInfo contains "<invalid>" for function/filename it 400 // cannot fetch. We replace it to "??" to make our output closer to addr2line. 401 static const std::string kDILineInfoBadString = "<invalid>"; 402 std::stringstream Result; 403 if (Opts.PrintFunctions) { 404 std::string FunctionName = LineInfo.getFunctionName(); 405 if (FunctionName == kDILineInfoBadString) 406 FunctionName = kBadString; 407 else if (Opts.Demangle) 408 FunctionName = DemangleName(FunctionName); 409 Result << FunctionName << "\n"; 410 } 411 std::string Filename = LineInfo.getFileName(); 412 if (Filename == kDILineInfoBadString) 413 Filename = kBadString; 414 Result << Filename << ":" << LineInfo.getLine() << ":" << LineInfo.getColumn() 415 << "\n"; 416 return Result.str(); 417} 418 419#if !defined(_MSC_VER) 420// Assume that __cxa_demangle is provided by libcxxabi (except for Windows). 421extern "C" char *__cxa_demangle(const char *mangled_name, char *output_buffer, 422 size_t *length, int *status); 423#endif 424 425std::string LLVMSymbolizer::DemangleName(const std::string &Name) { 426#if !defined(_MSC_VER) 427 int status = 0; 428 char *DemangledName = __cxa_demangle(Name.c_str(), 0, 0, &status); 429 if (status != 0) 430 return Name; 431 std::string Result = DemangledName; 432 free(DemangledName); 433 return Result; 434#else 435 return Name; 436#endif 437} 438 439std::string LLVMSymbolizer::DemangleGlobalName(const std::string &Name) { 440 // We can spoil names of globals with C linkage, so use an heuristic 441 // approach to check if the name should be demangled. 442 return (Name.substr(0, 2) == "_Z") ? DemangleName(Name) : Name; 443} 444 445} // namespace symbolize 446} // namespace llvm 447