1 //===-- LLVMSymbolize.cpp -------------------------------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // Implementation for LLVM symbolization library.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "llvm/DebugInfo/Symbolize/Symbolize.h"
15
16 #include "SymbolizableObjectFile.h"
17
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/COFF.h"
24 #include "llvm/Object/ELFObjectFile.h"
25 #include "llvm/Object/MachO.h"
26 #include "llvm/Object/MachOUniversal.h"
27 #include "llvm/Support/COFF.h"
28 #include "llvm/Support/Casting.h"
29 #include "llvm/Support/Compression.h"
30 #include "llvm/Support/DataExtractor.h"
31 #include "llvm/Support/Errc.h"
32 #include "llvm/Support/FileSystem.h"
33 #include "llvm/Support/MemoryBuffer.h"
34 #include "llvm/Support/Path.h"
35 #include <algorithm>
36 #include <cassert>
37 #include <cstdlib>
38 #include <cstring>
39
40 #if defined(_MSC_VER)
41 #include <Windows.h>
42 #include <DbgHelp.h>
43 #pragma comment(lib, "dbghelp.lib")
44
45 // Windows.h conflicts with our COFF header definitions.
46 #ifdef IMAGE_FILE_MACHINE_I386
47 #undef IMAGE_FILE_MACHINE_I386
48 #endif
49 #endif
50
51 namespace llvm {
52 namespace symbolize {
53
symbolizeCode(const std::string & ModuleName,uint64_t ModuleOffset)54 Expected<DILineInfo> LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
55 uint64_t ModuleOffset) {
56 SymbolizableModule *Info;
57 if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName))
58 Info = InfoOrErr.get();
59 else
60 return InfoOrErr.takeError();
61
62 // A null module means an error has already been reported. Return an empty
63 // result.
64 if (!Info)
65 return DILineInfo();
66
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();
71
72 DILineInfo LineInfo = Info->symbolizeCode(ModuleOffset, Opts.PrintFunctions,
73 Opts.UseSymbolTable);
74 if (Opts.Demangle)
75 LineInfo.FunctionName = DemangleName(LineInfo.FunctionName, Info);
76 return LineInfo;
77 }
78
79 Expected<DIInliningInfo>
symbolizeInlinedCode(const std::string & ModuleName,uint64_t ModuleOffset)80 LLVMSymbolizer::symbolizeInlinedCode(const std::string &ModuleName,
81 uint64_t ModuleOffset) {
82 SymbolizableModule *Info;
83 if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName))
84 Info = InfoOrErr.get();
85 else
86 return InfoOrErr.takeError();
87
88 // A null module means an error has already been reported. Return an empty
89 // result.
90 if (!Info)
91 return DIInliningInfo();
92
93 // If the user is giving us relative addresses, add the preferred base of the
94 // object to the offset before we do the query. It's what DIContext expects.
95 if (Opts.RelativeAddresses)
96 ModuleOffset += Info->getModulePreferredBase();
97
98 DIInliningInfo InlinedContext = Info->symbolizeInlinedCode(
99 ModuleOffset, Opts.PrintFunctions, Opts.UseSymbolTable);
100 if (Opts.Demangle) {
101 for (int i = 0, n = InlinedContext.getNumberOfFrames(); i < n; i++) {
102 auto *Frame = InlinedContext.getMutableFrame(i);
103 Frame->FunctionName = DemangleName(Frame->FunctionName, Info);
104 }
105 }
106 return InlinedContext;
107 }
108
symbolizeData(const std::string & ModuleName,uint64_t ModuleOffset)109 Expected<DIGlobal> LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
110 uint64_t ModuleOffset) {
111 SymbolizableModule *Info;
112 if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName))
113 Info = InfoOrErr.get();
114 else
115 return InfoOrErr.takeError();
116
117 // A null module means an error has already been reported. Return an empty
118 // result.
119 if (!Info)
120 return DIGlobal();
121
122 // If the user is giving us relative addresses, add the preferred base of
123 // the object to the offset before we do the query. It's what DIContext
124 // expects.
125 if (Opts.RelativeAddresses)
126 ModuleOffset += Info->getModulePreferredBase();
127
128 DIGlobal Global = Info->symbolizeData(ModuleOffset);
129 if (Opts.Demangle)
130 Global.Name = DemangleName(Global.Name, Info);
131 return Global;
132 }
133
flush()134 void LLVMSymbolizer::flush() {
135 ObjectForUBPathAndArch.clear();
136 BinaryForPath.clear();
137 ObjectPairForPathArch.clear();
138 Modules.clear();
139 }
140
141 namespace {
142
143 // For Path="/path/to/foo" and Basename="foo" assume that debug info is in
144 // /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
145 // For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
146 // /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
getDarwinDWARFResourceForPath(const std::string & Path,const std::string & Basename)147 std::string getDarwinDWARFResourceForPath(
148 const std::string &Path, const std::string &Basename) {
149 SmallString<16> ResourceName = StringRef(Path);
150 if (sys::path::extension(Path) != ".dSYM") {
151 ResourceName += ".dSYM";
152 }
153 sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
154 sys::path::append(ResourceName, Basename);
155 return ResourceName.str();
156 }
157
checkFileCRC(StringRef Path,uint32_t CRCHash)158 bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
159 ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
160 MemoryBuffer::getFileOrSTDIN(Path);
161 if (!MB)
162 return false;
163 return !zlib::isAvailable() || CRCHash == zlib::crc32(MB.get()->getBuffer());
164 }
165
findDebugBinary(const std::string & OrigPath,const std::string & DebuglinkName,uint32_t CRCHash,std::string & Result)166 bool findDebugBinary(const std::string &OrigPath,
167 const std::string &DebuglinkName, uint32_t CRCHash,
168 std::string &Result) {
169 std::string OrigRealPath = OrigPath;
170 #if defined(HAVE_REALPATH)
171 if (char *RP = realpath(OrigPath.c_str(), nullptr)) {
172 OrigRealPath = RP;
173 free(RP);
174 }
175 #endif
176 SmallString<16> OrigDir(OrigRealPath);
177 llvm::sys::path::remove_filename(OrigDir);
178 SmallString<16> DebugPath = OrigDir;
179 // Try /path/to/original_binary/debuglink_name
180 llvm::sys::path::append(DebugPath, DebuglinkName);
181 if (checkFileCRC(DebugPath, CRCHash)) {
182 Result = DebugPath.str();
183 return true;
184 }
185 // Try /path/to/original_binary/.debug/debuglink_name
186 DebugPath = OrigRealPath;
187 llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
188 if (checkFileCRC(DebugPath, CRCHash)) {
189 Result = DebugPath.str();
190 return true;
191 }
192 // Try /usr/lib/debug/path/to/original_binary/debuglink_name
193 DebugPath = "/usr/lib/debug";
194 llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
195 DebuglinkName);
196 if (checkFileCRC(DebugPath, CRCHash)) {
197 Result = DebugPath.str();
198 return true;
199 }
200 return false;
201 }
202
getGNUDebuglinkContents(const ObjectFile * Obj,std::string & DebugName,uint32_t & CRCHash)203 bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
204 uint32_t &CRCHash) {
205 if (!Obj)
206 return false;
207 for (const SectionRef &Section : Obj->sections()) {
208 StringRef Name;
209 Section.getName(Name);
210 Name = Name.substr(Name.find_first_not_of("._"));
211 if (Name == "gnu_debuglink") {
212 StringRef Data;
213 Section.getContents(Data);
214 DataExtractor DE(Data, Obj->isLittleEndian(), 0);
215 uint32_t Offset = 0;
216 if (const char *DebugNameStr = DE.getCStr(&Offset)) {
217 // 4-byte align the offset.
218 Offset = (Offset + 3) & ~0x3;
219 if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
220 DebugName = DebugNameStr;
221 CRCHash = DE.getU32(&Offset);
222 return true;
223 }
224 }
225 break;
226 }
227 }
228 return false;
229 }
230
darwinDsymMatchesBinary(const MachOObjectFile * DbgObj,const MachOObjectFile * Obj)231 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
232 const MachOObjectFile *Obj) {
233 ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
234 ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
235 if (dbg_uuid.empty() || bin_uuid.empty())
236 return false;
237 return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
238 }
239
240 } // end anonymous namespace
241
lookUpDsymFile(const std::string & ExePath,const MachOObjectFile * MachExeObj,const std::string & ArchName)242 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
243 const MachOObjectFile *MachExeObj, const std::string &ArchName) {
244 // On Darwin we may find DWARF in separate object file in
245 // resource directory.
246 std::vector<std::string> DsymPaths;
247 StringRef Filename = sys::path::filename(ExePath);
248 DsymPaths.push_back(getDarwinDWARFResourceForPath(ExePath, Filename));
249 for (const auto &Path : Opts.DsymHints) {
250 DsymPaths.push_back(getDarwinDWARFResourceForPath(Path, Filename));
251 }
252 for (const auto &Path : DsymPaths) {
253 auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
254 if (!DbgObjOrErr) {
255 // Ignore errors, the file might not exist.
256 consumeError(DbgObjOrErr.takeError());
257 continue;
258 }
259 ObjectFile *DbgObj = DbgObjOrErr.get();
260 if (!DbgObj)
261 continue;
262 const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(DbgObj);
263 if (!MachDbgObj)
264 continue;
265 if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj))
266 return DbgObj;
267 }
268 return nullptr;
269 }
270
lookUpDebuglinkObject(const std::string & Path,const ObjectFile * Obj,const std::string & ArchName)271 ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
272 const ObjectFile *Obj,
273 const std::string &ArchName) {
274 std::string DebuglinkName;
275 uint32_t CRCHash;
276 std::string DebugBinaryPath;
277 if (!getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash))
278 return nullptr;
279 if (!findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath))
280 return nullptr;
281 auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
282 if (!DbgObjOrErr) {
283 // Ignore errors, the file might not exist.
284 consumeError(DbgObjOrErr.takeError());
285 return nullptr;
286 }
287 return DbgObjOrErr.get();
288 }
289
290 Expected<LLVMSymbolizer::ObjectPair>
getOrCreateObjectPair(const std::string & Path,const std::string & ArchName)291 LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
292 const std::string &ArchName) {
293 const auto &I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
294 if (I != ObjectPairForPathArch.end()) {
295 return I->second;
296 }
297
298 auto ObjOrErr = getOrCreateObject(Path, ArchName);
299 if (!ObjOrErr) {
300 ObjectPairForPathArch.insert(std::make_pair(std::make_pair(Path, ArchName),
301 ObjectPair(nullptr, nullptr)));
302 return ObjOrErr.takeError();
303 }
304
305 ObjectFile *Obj = ObjOrErr.get();
306 assert(Obj != nullptr);
307 ObjectFile *DbgObj = nullptr;
308
309 if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
310 DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
311 if (!DbgObj)
312 DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
313 if (!DbgObj)
314 DbgObj = Obj;
315 ObjectPair Res = std::make_pair(Obj, DbgObj);
316 ObjectPairForPathArch.insert(
317 std::make_pair(std::make_pair(Path, ArchName), Res));
318 return Res;
319 }
320
321 Expected<ObjectFile *>
getOrCreateObject(const std::string & Path,const std::string & ArchName)322 LLVMSymbolizer::getOrCreateObject(const std::string &Path,
323 const std::string &ArchName) {
324 const auto &I = BinaryForPath.find(Path);
325 Binary *Bin = nullptr;
326 if (I == BinaryForPath.end()) {
327 Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
328 if (!BinOrErr) {
329 BinaryForPath.insert(std::make_pair(Path, OwningBinary<Binary>()));
330 return BinOrErr.takeError();
331 }
332 Bin = BinOrErr->getBinary();
333 BinaryForPath.insert(std::make_pair(Path, std::move(BinOrErr.get())));
334 } else {
335 Bin = I->second.getBinary();
336 }
337
338 if (!Bin)
339 return static_cast<ObjectFile *>(nullptr);
340
341 if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Bin)) {
342 const auto &I = ObjectForUBPathAndArch.find(std::make_pair(Path, ArchName));
343 if (I != ObjectForUBPathAndArch.end()) {
344 return I->second.get();
345 }
346 Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
347 UB->getObjectForArch(ArchName);
348 if (!ObjOrErr) {
349 ObjectForUBPathAndArch.insert(std::make_pair(
350 std::make_pair(Path, ArchName), std::unique_ptr<ObjectFile>()));
351 return ObjOrErr.takeError();
352 }
353 ObjectFile *Res = ObjOrErr->get();
354 ObjectForUBPathAndArch.insert(std::make_pair(std::make_pair(Path, ArchName),
355 std::move(ObjOrErr.get())));
356 return Res;
357 }
358 if (Bin->isObject()) {
359 return cast<ObjectFile>(Bin);
360 }
361 return errorCodeToError(object_error::arch_not_found);
362 }
363
364 Expected<SymbolizableModule *>
getOrCreateModuleInfo(const std::string & ModuleName)365 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) {
366 const auto &I = Modules.find(ModuleName);
367 if (I != Modules.end()) {
368 return I->second.get();
369 }
370 std::string BinaryName = ModuleName;
371 std::string ArchName = Opts.DefaultArch;
372 size_t ColonPos = ModuleName.find_last_of(':');
373 // Verify that substring after colon form a valid arch name.
374 if (ColonPos != std::string::npos) {
375 std::string ArchStr = ModuleName.substr(ColonPos + 1);
376 if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
377 BinaryName = ModuleName.substr(0, ColonPos);
378 ArchName = ArchStr;
379 }
380 }
381 auto ObjectsOrErr = getOrCreateObjectPair(BinaryName, ArchName);
382 if (!ObjectsOrErr) {
383 // Failed to find valid object file.
384 Modules.insert(
385 std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
386 return ObjectsOrErr.takeError();
387 }
388 ObjectPair Objects = ObjectsOrErr.get();
389
390 std::unique_ptr<DIContext> Context;
391 // If this is a COFF object containing PDB info, use a PDBContext to
392 // symbolize. Otherwise, use DWARF.
393 if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
394 const debug_pdb_info *PDBInfo;
395 StringRef PDBFileName;
396 auto EC = CoffObject->getDebugPDBInfo(PDBInfo, PDBFileName);
397 if (!EC && PDBInfo != nullptr) {
398 using namespace pdb;
399 std::unique_ptr<IPDBSession> Session;
400 if (auto Err = loadDataForEXE(PDB_ReaderType::DIA,
401 Objects.first->getFileName(), Session)) {
402 Modules.insert(
403 std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
404 return std::move(Err);
405 }
406 Context.reset(new PDBContext(*CoffObject, std::move(Session)));
407 }
408 }
409 if (!Context)
410 Context.reset(new DWARFContextInMemory(*Objects.second));
411 assert(Context);
412 auto InfoOrErr =
413 SymbolizableObjectFile::create(Objects.first, std::move(Context));
414 std::unique_ptr<SymbolizableModule> SymMod;
415 if (InfoOrErr)
416 SymMod = std::move(InfoOrErr.get());
417 auto InsertResult =
418 Modules.insert(std::make_pair(ModuleName, std::move(SymMod)));
419 assert(InsertResult.second);
420 if (auto EC = InfoOrErr.getError())
421 return errorCodeToError(EC);
422 return InsertResult.first->second.get();
423 }
424
425 namespace {
426
427 // Undo these various manglings for Win32 extern "C" functions:
428 // cdecl - _foo
429 // stdcall - _foo@12
430 // fastcall - @foo@12
431 // vectorcall - foo@@12
432 // These are all different linkage names for 'foo'.
demanglePE32ExternCFunc(StringRef SymbolName)433 StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
434 // Remove any '_' or '@' prefix.
435 char Front = SymbolName.empty() ? '\0' : SymbolName[0];
436 if (Front == '_' || Front == '@')
437 SymbolName = SymbolName.drop_front();
438
439 // Remove any '@[0-9]+' suffix.
440 if (Front != '?') {
441 size_t AtPos = SymbolName.rfind('@');
442 if (AtPos != StringRef::npos &&
443 std::all_of(SymbolName.begin() + AtPos + 1, SymbolName.end(),
444 [](char C) { return C >= '0' && C <= '9'; })) {
445 SymbolName = SymbolName.substr(0, AtPos);
446 }
447 }
448
449 // Remove any ending '@' for vectorcall.
450 if (SymbolName.endswith("@"))
451 SymbolName = SymbolName.drop_back();
452
453 return SymbolName;
454 }
455
456 } // end anonymous namespace
457
458 #if !defined(_MSC_VER)
459 // Assume that __cxa_demangle is provided by libcxxabi (except for Windows).
460 extern "C" char *__cxa_demangle(const char *mangled_name, char *output_buffer,
461 size_t *length, int *status);
462 #endif
463
DemangleName(const std::string & Name,const SymbolizableModule * ModInfo)464 std::string LLVMSymbolizer::DemangleName(const std::string &Name,
465 const SymbolizableModule *ModInfo) {
466 #if !defined(_MSC_VER)
467 // We can spoil names of symbols with C linkage, so use an heuristic
468 // approach to check if the name should be demangled.
469 if (Name.substr(0, 2) == "_Z") {
470 int status = 0;
471 char *DemangledName = __cxa_demangle(Name.c_str(), nullptr, nullptr, &status);
472 if (status != 0)
473 return Name;
474 std::string Result = DemangledName;
475 free(DemangledName);
476 return Result;
477 }
478 #else
479 if (!Name.empty() && Name.front() == '?') {
480 // Only do MSVC C++ demangling on symbols starting with '?'.
481 char DemangledName[1024] = {0};
482 DWORD result = ::UnDecorateSymbolName(
483 Name.c_str(), DemangledName, 1023,
484 UNDNAME_NO_ACCESS_SPECIFIERS | // Strip public, private, protected
485 UNDNAME_NO_ALLOCATION_LANGUAGE | // Strip __thiscall, __stdcall, etc
486 UNDNAME_NO_THROW_SIGNATURES | // Strip throw() specifications
487 UNDNAME_NO_MEMBER_TYPE | // Strip virtual, static, etc specifiers
488 UNDNAME_NO_MS_KEYWORDS | // Strip all MS extension keywords
489 UNDNAME_NO_FUNCTION_RETURNS); // Strip function return types
490 return (result == 0) ? Name : std::string(DemangledName);
491 }
492 #endif
493 if (ModInfo && ModInfo->isWin32Module())
494 return std::string(demanglePE32ExternCFunc(Name));
495 return Name;
496 }
497
498 } // namespace symbolize
499 } // namespace llvm
500