1 //===-- LLVMSymbolize.cpp -------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // Implementation for LLVM symbolization library.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/DebugInfo/Symbolize/Symbolize.h"
14 
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
17 #include "llvm/DebugInfo/PDB/PDB.h"
18 #include "llvm/DebugInfo/PDB/PDBContext.h"
19 #include "llvm/DebugInfo/Symbolize/DIFetcher.h"
20 #include "llvm/DebugInfo/Symbolize/SymbolizableObjectFile.h"
21 #include "llvm/Demangle/Demangle.h"
22 #include "llvm/Object/COFF.h"
23 #include "llvm/Object/ELF.h"
24 #include "llvm/Object/ELFObjectFile.h"
25 #include "llvm/Object/MachO.h"
26 #include "llvm/Object/MachOUniversal.h"
27 #include "llvm/Support/CRC.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 <cstring>
38 
39 namespace llvm {
40 namespace symbolize {
41 
42 LLVMSymbolizer::LLVMSymbolizer(const Options &Opts) : Opts(Opts) {}
43 
44 LLVMSymbolizer::~LLVMSymbolizer() = default;
45 
46 template <typename T>
47 Expected<DILineInfo>
48 LLVMSymbolizer::symbolizeCodeCommon(const T &ModuleSpecifier,
49                                     object::SectionedAddress ModuleOffset) {
50 
51   auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
52   if (!InfoOrErr)
53     return InfoOrErr.takeError();
54 
55   SymbolizableModule *Info = *InfoOrErr;
56 
57   // A null module means an error has already been reported. Return an empty
58   // result.
59   if (!Info)
60     return DILineInfo();
61 
62   // If the user is giving us relative addresses, add the preferred base of the
63   // object to the offset before we do the query. It's what DIContext expects.
64   if (Opts.RelativeAddresses)
65     ModuleOffset.Address += Info->getModulePreferredBase();
66 
67   DILineInfo LineInfo = Info->symbolizeCode(
68       ModuleOffset, DILineInfoSpecifier(Opts.PathStyle, Opts.PrintFunctions),
69       Opts.UseSymbolTable);
70   if (Opts.Demangle)
71     LineInfo.FunctionName = DemangleName(LineInfo.FunctionName, Info);
72   return LineInfo;
73 }
74 
75 Expected<DILineInfo>
76 LLVMSymbolizer::symbolizeCode(const ObjectFile &Obj,
77                               object::SectionedAddress ModuleOffset) {
78   return symbolizeCodeCommon(Obj, ModuleOffset);
79 }
80 
81 Expected<DILineInfo>
82 LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
83                               object::SectionedAddress ModuleOffset) {
84   return symbolizeCodeCommon(ModuleName, ModuleOffset);
85 }
86 
87 Expected<DILineInfo>
88 LLVMSymbolizer::symbolizeCode(ArrayRef<uint8_t> BuildID,
89                               object::SectionedAddress ModuleOffset) {
90   return symbolizeCodeCommon(BuildID, ModuleOffset);
91 }
92 
93 template <typename T>
94 Expected<DIInliningInfo> LLVMSymbolizer::symbolizeInlinedCodeCommon(
95     const T &ModuleSpecifier, object::SectionedAddress ModuleOffset) {
96   auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
97   if (!InfoOrErr)
98     return InfoOrErr.takeError();
99 
100   SymbolizableModule *Info = *InfoOrErr;
101 
102   // A null module means an error has already been reported. Return an empty
103   // result.
104   if (!Info)
105     return DIInliningInfo();
106 
107   // If the user is giving us relative addresses, add the preferred base of the
108   // object to the offset before we do the query. It's what DIContext expects.
109   if (Opts.RelativeAddresses)
110     ModuleOffset.Address += Info->getModulePreferredBase();
111 
112   DIInliningInfo InlinedContext = Info->symbolizeInlinedCode(
113       ModuleOffset, DILineInfoSpecifier(Opts.PathStyle, Opts.PrintFunctions),
114       Opts.UseSymbolTable);
115   if (Opts.Demangle) {
116     for (int i = 0, n = InlinedContext.getNumberOfFrames(); i < n; i++) {
117       auto *Frame = InlinedContext.getMutableFrame(i);
118       Frame->FunctionName = DemangleName(Frame->FunctionName, Info);
119     }
120   }
121   return InlinedContext;
122 }
123 
124 Expected<DIInliningInfo>
125 LLVMSymbolizer::symbolizeInlinedCode(const ObjectFile &Obj,
126                                      object::SectionedAddress ModuleOffset) {
127   return symbolizeInlinedCodeCommon(Obj, ModuleOffset);
128 }
129 
130 Expected<DIInliningInfo>
131 LLVMSymbolizer::symbolizeInlinedCode(const std::string &ModuleName,
132                                      object::SectionedAddress ModuleOffset) {
133   return symbolizeInlinedCodeCommon(ModuleName, ModuleOffset);
134 }
135 
136 Expected<DIInliningInfo>
137 LLVMSymbolizer::symbolizeInlinedCode(ArrayRef<uint8_t> BuildID,
138                                      object::SectionedAddress ModuleOffset) {
139   return symbolizeInlinedCodeCommon(BuildID, ModuleOffset);
140 }
141 
142 template <typename T>
143 Expected<DIGlobal>
144 LLVMSymbolizer::symbolizeDataCommon(const T &ModuleSpecifier,
145                                     object::SectionedAddress ModuleOffset) {
146 
147   auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
148   if (!InfoOrErr)
149     return InfoOrErr.takeError();
150 
151   SymbolizableModule *Info = *InfoOrErr;
152   // A null module means an error has already been reported. Return an empty
153   // result.
154   if (!Info)
155     return DIGlobal();
156 
157   // If the user is giving us relative addresses, add the preferred base of
158   // the object to the offset before we do the query. It's what DIContext
159   // expects.
160   if (Opts.RelativeAddresses)
161     ModuleOffset.Address += Info->getModulePreferredBase();
162 
163   DIGlobal Global = Info->symbolizeData(ModuleOffset);
164   if (Opts.Demangle)
165     Global.Name = DemangleName(Global.Name, Info);
166   return Global;
167 }
168 
169 Expected<DIGlobal>
170 LLVMSymbolizer::symbolizeData(const ObjectFile &Obj,
171                               object::SectionedAddress ModuleOffset) {
172   return symbolizeDataCommon(Obj, ModuleOffset);
173 }
174 
175 Expected<DIGlobal>
176 LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
177                               object::SectionedAddress ModuleOffset) {
178   return symbolizeDataCommon(ModuleName, ModuleOffset);
179 }
180 
181 Expected<DIGlobal>
182 LLVMSymbolizer::symbolizeData(ArrayRef<uint8_t> BuildID,
183                               object::SectionedAddress ModuleOffset) {
184   return symbolizeDataCommon(BuildID, ModuleOffset);
185 }
186 
187 template <typename T>
188 Expected<std::vector<DILocal>>
189 LLVMSymbolizer::symbolizeFrameCommon(const T &ModuleSpecifier,
190                                      object::SectionedAddress ModuleOffset) {
191   auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
192   if (!InfoOrErr)
193     return InfoOrErr.takeError();
194 
195   SymbolizableModule *Info = *InfoOrErr;
196   // A null module means an error has already been reported. Return an empty
197   // result.
198   if (!Info)
199     return std::vector<DILocal>();
200 
201   // If the user is giving us relative addresses, add the preferred base of
202   // the object to the offset before we do the query. It's what DIContext
203   // expects.
204   if (Opts.RelativeAddresses)
205     ModuleOffset.Address += Info->getModulePreferredBase();
206 
207   return Info->symbolizeFrame(ModuleOffset);
208 }
209 
210 Expected<std::vector<DILocal>>
211 LLVMSymbolizer::symbolizeFrame(const ObjectFile &Obj,
212                                object::SectionedAddress ModuleOffset) {
213   return symbolizeFrameCommon(Obj, ModuleOffset);
214 }
215 
216 Expected<std::vector<DILocal>>
217 LLVMSymbolizer::symbolizeFrame(const std::string &ModuleName,
218                                object::SectionedAddress ModuleOffset) {
219   return symbolizeFrameCommon(ModuleName, ModuleOffset);
220 }
221 
222 Expected<std::vector<DILocal>>
223 LLVMSymbolizer::symbolizeFrame(ArrayRef<uint8_t> BuildID,
224                                object::SectionedAddress ModuleOffset) {
225   return symbolizeFrameCommon(BuildID, ModuleOffset);
226 }
227 
228 void LLVMSymbolizer::flush() {
229   ObjectForUBPathAndArch.clear();
230   BinaryForPath.clear();
231   ObjectPairForPathArch.clear();
232   Modules.clear();
233   BuildIDPaths.clear();
234 }
235 
236 namespace {
237 
238 // For Path="/path/to/foo" and Basename="foo" assume that debug info is in
239 // /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
240 // For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
241 // /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
242 std::string getDarwinDWARFResourceForPath(const std::string &Path,
243                                           const std::string &Basename) {
244   SmallString<16> ResourceName = StringRef(Path);
245   if (sys::path::extension(Path) != ".dSYM") {
246     ResourceName += ".dSYM";
247   }
248   sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
249   sys::path::append(ResourceName, Basename);
250   return std::string(ResourceName.str());
251 }
252 
253 bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
254   ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
255       MemoryBuffer::getFileOrSTDIN(Path);
256   if (!MB)
257     return false;
258   return CRCHash == llvm::crc32(arrayRefFromStringRef(MB.get()->getBuffer()));
259 }
260 
261 bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
262                              uint32_t &CRCHash) {
263   if (!Obj)
264     return false;
265   for (const SectionRef &Section : Obj->sections()) {
266     StringRef Name;
267     consumeError(Section.getName().moveInto(Name));
268 
269     Name = Name.substr(Name.find_first_not_of("._"));
270     if (Name == "gnu_debuglink") {
271       Expected<StringRef> ContentsOrErr = Section.getContents();
272       if (!ContentsOrErr) {
273         consumeError(ContentsOrErr.takeError());
274         return false;
275       }
276       DataExtractor DE(*ContentsOrErr, Obj->isLittleEndian(), 0);
277       uint64_t Offset = 0;
278       if (const char *DebugNameStr = DE.getCStr(&Offset)) {
279         // 4-byte align the offset.
280         Offset = (Offset + 3) & ~0x3;
281         if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
282           DebugName = DebugNameStr;
283           CRCHash = DE.getU32(&Offset);
284           return true;
285         }
286       }
287       break;
288     }
289   }
290   return false;
291 }
292 
293 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
294                              const MachOObjectFile *Obj) {
295   ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
296   ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
297   if (dbg_uuid.empty() || bin_uuid.empty())
298     return false;
299   return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
300 }
301 
302 template <typename ELFT>
303 Optional<ArrayRef<uint8_t>> getBuildID(const ELFFile<ELFT> &Obj) {
304   auto PhdrsOrErr = Obj.program_headers();
305   if (!PhdrsOrErr) {
306     consumeError(PhdrsOrErr.takeError());
307     return {};
308   }
309   for (const auto &P : *PhdrsOrErr) {
310     if (P.p_type != ELF::PT_NOTE)
311       continue;
312     Error Err = Error::success();
313     for (auto N : Obj.notes(P, Err))
314       if (N.getType() == ELF::NT_GNU_BUILD_ID &&
315           N.getName() == ELF::ELF_NOTE_GNU)
316         return N.getDesc();
317     consumeError(std::move(Err));
318   }
319   return {};
320 }
321 
322 Optional<ArrayRef<uint8_t>> getBuildID(const ELFObjectFileBase *Obj) {
323   Optional<ArrayRef<uint8_t>> BuildID;
324   if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Obj))
325     BuildID = getBuildID(O->getELFFile());
326   else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Obj))
327     BuildID = getBuildID(O->getELFFile());
328   else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Obj))
329     BuildID = getBuildID(O->getELFFile());
330   else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Obj))
331     BuildID = getBuildID(O->getELFFile());
332   else
333     llvm_unreachable("unsupported file format");
334   return BuildID;
335 }
336 
337 } // end anonymous namespace
338 
339 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
340                                            const MachOObjectFile *MachExeObj,
341                                            const std::string &ArchName) {
342   // On Darwin we may find DWARF in separate object file in
343   // resource directory.
344   std::vector<std::string> DsymPaths;
345   StringRef Filename = sys::path::filename(ExePath);
346   DsymPaths.push_back(
347       getDarwinDWARFResourceForPath(ExePath, std::string(Filename)));
348   for (const auto &Path : Opts.DsymHints) {
349     DsymPaths.push_back(
350         getDarwinDWARFResourceForPath(Path, std::string(Filename)));
351   }
352   for (const auto &Path : DsymPaths) {
353     auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
354     if (!DbgObjOrErr) {
355       // Ignore errors, the file might not exist.
356       consumeError(DbgObjOrErr.takeError());
357       continue;
358     }
359     ObjectFile *DbgObj = DbgObjOrErr.get();
360     if (!DbgObj)
361       continue;
362     const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(DbgObj);
363     if (!MachDbgObj)
364       continue;
365     if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj))
366       return DbgObj;
367   }
368   return nullptr;
369 }
370 
371 ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
372                                                   const ObjectFile *Obj,
373                                                   const std::string &ArchName) {
374   std::string DebuglinkName;
375   uint32_t CRCHash;
376   std::string DebugBinaryPath;
377   if (!getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash))
378     return nullptr;
379   if (!findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath))
380     return nullptr;
381   auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
382   if (!DbgObjOrErr) {
383     // Ignore errors, the file might not exist.
384     consumeError(DbgObjOrErr.takeError());
385     return nullptr;
386   }
387   return DbgObjOrErr.get();
388 }
389 
390 ObjectFile *LLVMSymbolizer::lookUpBuildIDObject(const std::string &Path,
391                                                 const ELFObjectFileBase *Obj,
392                                                 const std::string &ArchName) {
393   auto BuildID = getBuildID(Obj);
394   if (!BuildID)
395     return nullptr;
396   if (BuildID->size() < 2)
397     return nullptr;
398   std::string DebugBinaryPath;
399   if (!getOrFindDebugBinary(*BuildID, DebugBinaryPath))
400     return nullptr;
401   auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
402   if (!DbgObjOrErr) {
403     consumeError(DbgObjOrErr.takeError());
404     return nullptr;
405   }
406   return DbgObjOrErr.get();
407 }
408 
409 bool LLVMSymbolizer::findDebugBinary(const std::string &OrigPath,
410                                      const std::string &DebuglinkName,
411                                      uint32_t CRCHash, std::string &Result) {
412   SmallString<16> OrigDir(OrigPath);
413   llvm::sys::path::remove_filename(OrigDir);
414   SmallString<16> DebugPath = OrigDir;
415   // Try relative/path/to/original_binary/debuglink_name
416   llvm::sys::path::append(DebugPath, DebuglinkName);
417   if (checkFileCRC(DebugPath, CRCHash)) {
418     Result = std::string(DebugPath.str());
419     return true;
420   }
421   // Try relative/path/to/original_binary/.debug/debuglink_name
422   DebugPath = OrigDir;
423   llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
424   if (checkFileCRC(DebugPath, CRCHash)) {
425     Result = std::string(DebugPath.str());
426     return true;
427   }
428   // Make the path absolute so that lookups will go to
429   // "/usr/lib/debug/full/path/to/debug", not
430   // "/usr/lib/debug/to/debug"
431   llvm::sys::fs::make_absolute(OrigDir);
432   if (!Opts.FallbackDebugPath.empty()) {
433     // Try <FallbackDebugPath>/absolute/path/to/original_binary/debuglink_name
434     DebugPath = Opts.FallbackDebugPath;
435   } else {
436 #if defined(__NetBSD__)
437     // Try /usr/libdata/debug/absolute/path/to/original_binary/debuglink_name
438     DebugPath = "/usr/libdata/debug";
439 #else
440     // Try /usr/lib/debug/absolute/path/to/original_binary/debuglink_name
441     DebugPath = "/usr/lib/debug";
442 #endif
443   }
444   llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
445                           DebuglinkName);
446   if (checkFileCRC(DebugPath, CRCHash)) {
447     Result = std::string(DebugPath.str());
448     return true;
449   }
450   return false;
451 }
452 
453 static StringRef getBuildIDStr(ArrayRef<uint8_t> BuildID) {
454   return StringRef(reinterpret_cast<const char *>(BuildID.data()),
455                    BuildID.size());
456 }
457 
458 bool LLVMSymbolizer::getOrFindDebugBinary(const ArrayRef<uint8_t> BuildID,
459                                           std::string &Result) {
460   StringRef BuildIDStr = getBuildIDStr(BuildID);
461   auto I = BuildIDPaths.find(BuildIDStr);
462   if (I != BuildIDPaths.end()) {
463     Result = I->second;
464     return true;
465   }
466   auto recordPath = [&](StringRef Path) {
467     Result = Path.str();
468     auto InsertResult = BuildIDPaths.insert({BuildIDStr, Result});
469     assert(InsertResult.second);
470     (void)InsertResult;
471   };
472 
473   Optional<std::string> Path;
474   Path = LocalDIFetcher(Opts.DebugFileDirectory).fetchBuildID(BuildID);
475   if (Path) {
476     recordPath(*Path);
477     return true;
478   }
479 
480   // Try caller-provided debug info fetchers.
481   for (const std::unique_ptr<DIFetcher> &Fetcher : DIFetchers) {
482     Path = Fetcher->fetchBuildID(BuildID);
483     if (Path) {
484       recordPath(*Path);
485       return true;
486     }
487   }
488 
489   return false;
490 }
491 
492 Expected<LLVMSymbolizer::ObjectPair>
493 LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
494                                       const std::string &ArchName) {
495   auto I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
496   if (I != ObjectPairForPathArch.end())
497     return I->second;
498 
499   auto ObjOrErr = getOrCreateObject(Path, ArchName);
500   if (!ObjOrErr) {
501     ObjectPairForPathArch.emplace(std::make_pair(Path, ArchName),
502                                   ObjectPair(nullptr, nullptr));
503     return ObjOrErr.takeError();
504   }
505 
506   ObjectFile *Obj = ObjOrErr.get();
507   assert(Obj != nullptr);
508   ObjectFile *DbgObj = nullptr;
509 
510   if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
511     DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
512   else if (auto ELFObj = dyn_cast<const ELFObjectFileBase>(Obj))
513     DbgObj = lookUpBuildIDObject(Path, ELFObj, ArchName);
514   if (!DbgObj)
515     DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
516   if (!DbgObj)
517     DbgObj = Obj;
518   ObjectPair Res = std::make_pair(Obj, DbgObj);
519   ObjectPairForPathArch.emplace(std::make_pair(Path, ArchName), Res);
520   return Res;
521 }
522 
523 Expected<ObjectFile *>
524 LLVMSymbolizer::getOrCreateObject(const std::string &Path,
525                                   const std::string &ArchName) {
526   Binary *Bin;
527   auto Pair = BinaryForPath.emplace(Path, OwningBinary<Binary>());
528   if (!Pair.second) {
529     Bin = Pair.first->second.getBinary();
530   } else {
531     Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
532     if (!BinOrErr)
533       return BinOrErr.takeError();
534     Pair.first->second = std::move(BinOrErr.get());
535     Bin = Pair.first->second.getBinary();
536   }
537 
538   if (!Bin)
539     return static_cast<ObjectFile *>(nullptr);
540 
541   if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Bin)) {
542     auto I = ObjectForUBPathAndArch.find(std::make_pair(Path, ArchName));
543     if (I != ObjectForUBPathAndArch.end())
544       return I->second.get();
545 
546     Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
547         UB->getMachOObjectForArch(ArchName);
548     if (!ObjOrErr) {
549       ObjectForUBPathAndArch.emplace(std::make_pair(Path, ArchName),
550                                      std::unique_ptr<ObjectFile>());
551       return ObjOrErr.takeError();
552     }
553     ObjectFile *Res = ObjOrErr->get();
554     ObjectForUBPathAndArch.emplace(std::make_pair(Path, ArchName),
555                                    std::move(ObjOrErr.get()));
556     return Res;
557   }
558   if (Bin->isObject()) {
559     return cast<ObjectFile>(Bin);
560   }
561   return errorCodeToError(object_error::arch_not_found);
562 }
563 
564 Expected<SymbolizableModule *>
565 LLVMSymbolizer::createModuleInfo(const ObjectFile *Obj,
566                                  std::unique_ptr<DIContext> Context,
567                                  StringRef ModuleName) {
568   auto InfoOrErr = SymbolizableObjectFile::create(Obj, std::move(Context),
569                                                   Opts.UntagAddresses);
570   std::unique_ptr<SymbolizableModule> SymMod;
571   if (InfoOrErr)
572     SymMod = std::move(*InfoOrErr);
573   auto InsertResult = Modules.insert(
574       std::make_pair(std::string(ModuleName), std::move(SymMod)));
575   assert(InsertResult.second);
576   if (!InfoOrErr)
577     return InfoOrErr.takeError();
578   return InsertResult.first->second.get();
579 }
580 
581 Expected<SymbolizableModule *>
582 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) {
583   auto I = Modules.find(ModuleName);
584   if (I != Modules.end())
585     return I->second.get();
586 
587   std::string BinaryName = ModuleName;
588   std::string ArchName = Opts.DefaultArch;
589   size_t ColonPos = ModuleName.find_last_of(':');
590   // Verify that substring after colon form a valid arch name.
591   if (ColonPos != std::string::npos) {
592     std::string ArchStr = ModuleName.substr(ColonPos + 1);
593     if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
594       BinaryName = ModuleName.substr(0, ColonPos);
595       ArchName = ArchStr;
596     }
597   }
598   auto ObjectsOrErr = getOrCreateObjectPair(BinaryName, ArchName);
599   if (!ObjectsOrErr) {
600     // Failed to find valid object file.
601     Modules.emplace(ModuleName, std::unique_ptr<SymbolizableModule>());
602     return ObjectsOrErr.takeError();
603   }
604   ObjectPair Objects = ObjectsOrErr.get();
605 
606   std::unique_ptr<DIContext> Context;
607   // If this is a COFF object containing PDB info, use a PDBContext to
608   // symbolize. Otherwise, use DWARF.
609   if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
610     const codeview::DebugInfo *DebugInfo;
611     StringRef PDBFileName;
612     auto EC = CoffObject->getDebugPDBInfo(DebugInfo, PDBFileName);
613     if (!EC && DebugInfo != nullptr && !PDBFileName.empty()) {
614       using namespace pdb;
615       std::unique_ptr<IPDBSession> Session;
616 
617       PDB_ReaderType ReaderType =
618           Opts.UseDIA ? PDB_ReaderType::DIA : PDB_ReaderType::Native;
619       if (auto Err = loadDataForEXE(ReaderType, Objects.first->getFileName(),
620                                     Session)) {
621         Modules.emplace(ModuleName, std::unique_ptr<SymbolizableModule>());
622         // Return along the PDB filename to provide more context
623         return createFileError(PDBFileName, std::move(Err));
624       }
625       Context.reset(new PDBContext(*CoffObject, std::move(Session)));
626     }
627   }
628   if (!Context)
629     Context = DWARFContext::create(
630         *Objects.second, DWARFContext::ProcessDebugRelocations::Process,
631         nullptr, Opts.DWPName);
632   return createModuleInfo(Objects.first, std::move(Context), ModuleName);
633 }
634 
635 Expected<SymbolizableModule *>
636 LLVMSymbolizer::getOrCreateModuleInfo(const ObjectFile &Obj) {
637   StringRef ObjName = Obj.getFileName();
638   auto I = Modules.find(ObjName);
639   if (I != Modules.end())
640     return I->second.get();
641 
642   std::unique_ptr<DIContext> Context = DWARFContext::create(Obj);
643   // FIXME: handle COFF object with PDB info to use PDBContext
644   return createModuleInfo(&Obj, std::move(Context), ObjName);
645 }
646 
647 Expected<SymbolizableModule *>
648 LLVMSymbolizer::getOrCreateModuleInfo(ArrayRef<uint8_t> BuildID) {
649   std::string Path;
650   if (!getOrFindDebugBinary(BuildID, Path)) {
651     return createStringError(errc::no_such_file_or_directory,
652                              Twine("could not find build ID '") +
653                                  toHex(BuildID) + "'");
654   }
655   return getOrCreateModuleInfo(Path);
656 }
657 
658 namespace {
659 
660 // Undo these various manglings for Win32 extern "C" functions:
661 // cdecl       - _foo
662 // stdcall     - _foo@12
663 // fastcall    - @foo@12
664 // vectorcall  - foo@@12
665 // These are all different linkage names for 'foo'.
666 StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
667   // Remove any '_' or '@' prefix.
668   char Front = SymbolName.empty() ? '\0' : SymbolName[0];
669   if (Front == '_' || Front == '@')
670     SymbolName = SymbolName.drop_front();
671 
672   // Remove any '@[0-9]+' suffix.
673   if (Front != '?') {
674     size_t AtPos = SymbolName.rfind('@');
675     if (AtPos != StringRef::npos &&
676         all_of(drop_begin(SymbolName, AtPos + 1), isDigit))
677       SymbolName = SymbolName.substr(0, AtPos);
678   }
679 
680   // Remove any ending '@' for vectorcall.
681   if (SymbolName.endswith("@"))
682     SymbolName = SymbolName.drop_back();
683 
684   return SymbolName;
685 }
686 
687 } // end anonymous namespace
688 
689 std::string
690 LLVMSymbolizer::DemangleName(const std::string &Name,
691                              const SymbolizableModule *DbiModuleDescriptor) {
692   std::string Result;
693   if (nonMicrosoftDemangle(Name.c_str(), Result))
694     return Result;
695 
696   if (!Name.empty() && Name.front() == '?') {
697     // Only do MSVC C++ demangling on symbols starting with '?'.
698     int status = 0;
699     char *DemangledName = microsoftDemangle(
700         Name.c_str(), nullptr, nullptr, nullptr, &status,
701         MSDemangleFlags(MSDF_NoAccessSpecifier | MSDF_NoCallingConvention |
702                         MSDF_NoMemberType | MSDF_NoReturnType));
703     if (status != 0)
704       return Name;
705     Result = DemangledName;
706     free(DemangledName);
707     return Result;
708   }
709 
710   if (DbiModuleDescriptor && DbiModuleDescriptor->isWin32Module())
711     return std::string(demanglePE32ExternCFunc(Name));
712   return Name;
713 }
714 
715 } // namespace symbolize
716 } // namespace llvm
717