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/SymbolizableObjectFile.h"
20 #include "llvm/Demangle/Demangle.h"
21 #include "llvm/Object/BuildID.h"
22 #include "llvm/Object/COFF.h"
23 #include "llvm/Object/ELFObjectFile.h"
24 #include "llvm/Object/MachO.h"
25 #include "llvm/Object/MachOUniversal.h"
26 #include "llvm/Support/CRC.h"
27 #include "llvm/Support/Casting.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"
33 #include <algorithm>
34 #include <cassert>
35 #include <cstring>
36 
37 namespace llvm {
38 namespace codeview {
39 union DebugInfo;
40 }
41 namespace symbolize {
42 
43 LLVMSymbolizer::LLVMSymbolizer() = default;
44 
LLVMSymbolizer(const Options & Opts)45 LLVMSymbolizer::LLVMSymbolizer(const Options &Opts)
46     : Opts(Opts),
47       BIDFetcher(std::make_unique<BuildIDFetcher>(Opts.DebugFileDirectory)) {}
48 
49 LLVMSymbolizer::~LLVMSymbolizer() = default;
50 
51 template <typename T>
52 Expected<DILineInfo>
symbolizeCodeCommon(const T & ModuleSpecifier,object::SectionedAddress ModuleOffset)53 LLVMSymbolizer::symbolizeCodeCommon(const T &ModuleSpecifier,
54                                     object::SectionedAddress ModuleOffset) {
55 
56   auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
57   if (!InfoOrErr)
58     return InfoOrErr.takeError();
59 
60   SymbolizableModule *Info = *InfoOrErr;
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.Address += Info->getModulePreferredBase();
71 
72   DILineInfo LineInfo = Info->symbolizeCode(
73       ModuleOffset, DILineInfoSpecifier(Opts.PathStyle, Opts.PrintFunctions),
74       Opts.UseSymbolTable);
75   if (Opts.Demangle)
76     LineInfo.FunctionName = DemangleName(LineInfo.FunctionName, Info);
77   return LineInfo;
78 }
79 
80 Expected<DILineInfo>
symbolizeCode(const ObjectFile & Obj,object::SectionedAddress ModuleOffset)81 LLVMSymbolizer::symbolizeCode(const ObjectFile &Obj,
82                               object::SectionedAddress ModuleOffset) {
83   return symbolizeCodeCommon(Obj, ModuleOffset);
84 }
85 
86 Expected<DILineInfo>
symbolizeCode(const std::string & ModuleName,object::SectionedAddress ModuleOffset)87 LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
88                               object::SectionedAddress ModuleOffset) {
89   return symbolizeCodeCommon(ModuleName, ModuleOffset);
90 }
91 
92 Expected<DILineInfo>
symbolizeCode(ArrayRef<uint8_t> BuildID,object::SectionedAddress ModuleOffset)93 LLVMSymbolizer::symbolizeCode(ArrayRef<uint8_t> BuildID,
94                               object::SectionedAddress ModuleOffset) {
95   return symbolizeCodeCommon(BuildID, ModuleOffset);
96 }
97 
98 template <typename T>
symbolizeInlinedCodeCommon(const T & ModuleSpecifier,object::SectionedAddress ModuleOffset)99 Expected<DIInliningInfo> LLVMSymbolizer::symbolizeInlinedCodeCommon(
100     const T &ModuleSpecifier, object::SectionedAddress ModuleOffset) {
101   auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
102   if (!InfoOrErr)
103     return InfoOrErr.takeError();
104 
105   SymbolizableModule *Info = *InfoOrErr;
106 
107   // A null module means an error has already been reported. Return an empty
108   // result.
109   if (!Info)
110     return DIInliningInfo();
111 
112   // If the user is giving us relative addresses, add the preferred base of the
113   // object to the offset before we do the query. It's what DIContext expects.
114   if (Opts.RelativeAddresses)
115     ModuleOffset.Address += Info->getModulePreferredBase();
116 
117   DIInliningInfo InlinedContext = Info->symbolizeInlinedCode(
118       ModuleOffset, DILineInfoSpecifier(Opts.PathStyle, Opts.PrintFunctions),
119       Opts.UseSymbolTable);
120   if (Opts.Demangle) {
121     for (int i = 0, n = InlinedContext.getNumberOfFrames(); i < n; i++) {
122       auto *Frame = InlinedContext.getMutableFrame(i);
123       Frame->FunctionName = DemangleName(Frame->FunctionName, Info);
124     }
125   }
126   return InlinedContext;
127 }
128 
129 Expected<DIInliningInfo>
symbolizeInlinedCode(const ObjectFile & Obj,object::SectionedAddress ModuleOffset)130 LLVMSymbolizer::symbolizeInlinedCode(const ObjectFile &Obj,
131                                      object::SectionedAddress ModuleOffset) {
132   return symbolizeInlinedCodeCommon(Obj, ModuleOffset);
133 }
134 
135 Expected<DIInliningInfo>
symbolizeInlinedCode(const std::string & ModuleName,object::SectionedAddress ModuleOffset)136 LLVMSymbolizer::symbolizeInlinedCode(const std::string &ModuleName,
137                                      object::SectionedAddress ModuleOffset) {
138   return symbolizeInlinedCodeCommon(ModuleName, ModuleOffset);
139 }
140 
141 Expected<DIInliningInfo>
symbolizeInlinedCode(ArrayRef<uint8_t> BuildID,object::SectionedAddress ModuleOffset)142 LLVMSymbolizer::symbolizeInlinedCode(ArrayRef<uint8_t> BuildID,
143                                      object::SectionedAddress ModuleOffset) {
144   return symbolizeInlinedCodeCommon(BuildID, ModuleOffset);
145 }
146 
147 template <typename T>
148 Expected<DIGlobal>
symbolizeDataCommon(const T & ModuleSpecifier,object::SectionedAddress ModuleOffset)149 LLVMSymbolizer::symbolizeDataCommon(const T &ModuleSpecifier,
150                                     object::SectionedAddress ModuleOffset) {
151 
152   auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
153   if (!InfoOrErr)
154     return InfoOrErr.takeError();
155 
156   SymbolizableModule *Info = *InfoOrErr;
157   // A null module means an error has already been reported. Return an empty
158   // result.
159   if (!Info)
160     return DIGlobal();
161 
162   // If the user is giving us relative addresses, add the preferred base of
163   // the object to the offset before we do the query. It's what DIContext
164   // expects.
165   if (Opts.RelativeAddresses)
166     ModuleOffset.Address += Info->getModulePreferredBase();
167 
168   DIGlobal Global = Info->symbolizeData(ModuleOffset);
169   if (Opts.Demangle)
170     Global.Name = DemangleName(Global.Name, Info);
171   return Global;
172 }
173 
174 Expected<DIGlobal>
symbolizeData(const ObjectFile & Obj,object::SectionedAddress ModuleOffset)175 LLVMSymbolizer::symbolizeData(const ObjectFile &Obj,
176                               object::SectionedAddress ModuleOffset) {
177   return symbolizeDataCommon(Obj, ModuleOffset);
178 }
179 
180 Expected<DIGlobal>
symbolizeData(const std::string & ModuleName,object::SectionedAddress ModuleOffset)181 LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
182                               object::SectionedAddress ModuleOffset) {
183   return symbolizeDataCommon(ModuleName, ModuleOffset);
184 }
185 
186 Expected<DIGlobal>
symbolizeData(ArrayRef<uint8_t> BuildID,object::SectionedAddress ModuleOffset)187 LLVMSymbolizer::symbolizeData(ArrayRef<uint8_t> BuildID,
188                               object::SectionedAddress ModuleOffset) {
189   return symbolizeDataCommon(BuildID, ModuleOffset);
190 }
191 
192 template <typename T>
193 Expected<std::vector<DILocal>>
symbolizeFrameCommon(const T & ModuleSpecifier,object::SectionedAddress ModuleOffset)194 LLVMSymbolizer::symbolizeFrameCommon(const T &ModuleSpecifier,
195                                      object::SectionedAddress ModuleOffset) {
196   auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
197   if (!InfoOrErr)
198     return InfoOrErr.takeError();
199 
200   SymbolizableModule *Info = *InfoOrErr;
201   // A null module means an error has already been reported. Return an empty
202   // result.
203   if (!Info)
204     return std::vector<DILocal>();
205 
206   // If the user is giving us relative addresses, add the preferred base of
207   // the object to the offset before we do the query. It's what DIContext
208   // expects.
209   if (Opts.RelativeAddresses)
210     ModuleOffset.Address += Info->getModulePreferredBase();
211 
212   return Info->symbolizeFrame(ModuleOffset);
213 }
214 
215 Expected<std::vector<DILocal>>
symbolizeFrame(const ObjectFile & Obj,object::SectionedAddress ModuleOffset)216 LLVMSymbolizer::symbolizeFrame(const ObjectFile &Obj,
217                                object::SectionedAddress ModuleOffset) {
218   return symbolizeFrameCommon(Obj, ModuleOffset);
219 }
220 
221 Expected<std::vector<DILocal>>
symbolizeFrame(const std::string & ModuleName,object::SectionedAddress ModuleOffset)222 LLVMSymbolizer::symbolizeFrame(const std::string &ModuleName,
223                                object::SectionedAddress ModuleOffset) {
224   return symbolizeFrameCommon(ModuleName, ModuleOffset);
225 }
226 
227 Expected<std::vector<DILocal>>
symbolizeFrame(ArrayRef<uint8_t> BuildID,object::SectionedAddress ModuleOffset)228 LLVMSymbolizer::symbolizeFrame(ArrayRef<uint8_t> BuildID,
229                                object::SectionedAddress ModuleOffset) {
230   return symbolizeFrameCommon(BuildID, ModuleOffset);
231 }
232 
flush()233 void LLVMSymbolizer::flush() {
234   ObjectForUBPathAndArch.clear();
235   LRUBinaries.clear();
236   CacheSize = 0;
237   BinaryForPath.clear();
238   ObjectPairForPathArch.clear();
239   Modules.clear();
240   BuildIDPaths.clear();
241 }
242 
243 namespace {
244 
245 // For Path="/path/to/foo" and Basename="foo" assume that debug info is in
246 // /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
247 // For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
248 // /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
getDarwinDWARFResourceForPath(const std::string & Path,const std::string & Basename)249 std::string getDarwinDWARFResourceForPath(const std::string &Path,
250                                           const std::string &Basename) {
251   SmallString<16> ResourceName = StringRef(Path);
252   if (sys::path::extension(Path) != ".dSYM") {
253     ResourceName += ".dSYM";
254   }
255   sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
256   sys::path::append(ResourceName, Basename);
257   return std::string(ResourceName.str());
258 }
259 
checkFileCRC(StringRef Path,uint32_t CRCHash)260 bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
261   ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
262       MemoryBuffer::getFileOrSTDIN(Path);
263   if (!MB)
264     return false;
265   return CRCHash == llvm::crc32(arrayRefFromStringRef(MB.get()->getBuffer()));
266 }
267 
getGNUDebuglinkContents(const ObjectFile * Obj,std::string & DebugName,uint32_t & CRCHash)268 bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
269                              uint32_t &CRCHash) {
270   if (!Obj)
271     return false;
272   for (const SectionRef &Section : Obj->sections()) {
273     StringRef Name;
274     consumeError(Section.getName().moveInto(Name));
275 
276     Name = Name.substr(Name.find_first_not_of("._"));
277     if (Name == "gnu_debuglink") {
278       Expected<StringRef> ContentsOrErr = Section.getContents();
279       if (!ContentsOrErr) {
280         consumeError(ContentsOrErr.takeError());
281         return false;
282       }
283       DataExtractor DE(*ContentsOrErr, Obj->isLittleEndian(), 0);
284       uint64_t Offset = 0;
285       if (const char *DebugNameStr = DE.getCStr(&Offset)) {
286         // 4-byte align the offset.
287         Offset = (Offset + 3) & ~0x3;
288         if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
289           DebugName = DebugNameStr;
290           CRCHash = DE.getU32(&Offset);
291           return true;
292         }
293       }
294       break;
295     }
296   }
297   return false;
298 }
299 
darwinDsymMatchesBinary(const MachOObjectFile * DbgObj,const MachOObjectFile * Obj)300 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
301                              const MachOObjectFile *Obj) {
302   ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
303   ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
304   if (dbg_uuid.empty() || bin_uuid.empty())
305     return false;
306   return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
307 }
308 
309 } // end anonymous namespace
310 
lookUpDsymFile(const std::string & ExePath,const MachOObjectFile * MachExeObj,const std::string & ArchName)311 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
312                                            const MachOObjectFile *MachExeObj,
313                                            const std::string &ArchName) {
314   // On Darwin we may find DWARF in separate object file in
315   // resource directory.
316   std::vector<std::string> DsymPaths;
317   StringRef Filename = sys::path::filename(ExePath);
318   DsymPaths.push_back(
319       getDarwinDWARFResourceForPath(ExePath, std::string(Filename)));
320   for (const auto &Path : Opts.DsymHints) {
321     DsymPaths.push_back(
322         getDarwinDWARFResourceForPath(Path, std::string(Filename)));
323   }
324   for (const auto &Path : DsymPaths) {
325     auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
326     if (!DbgObjOrErr) {
327       // Ignore errors, the file might not exist.
328       consumeError(DbgObjOrErr.takeError());
329       continue;
330     }
331     ObjectFile *DbgObj = DbgObjOrErr.get();
332     if (!DbgObj)
333       continue;
334     const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(DbgObj);
335     if (!MachDbgObj)
336       continue;
337     if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj))
338       return DbgObj;
339   }
340   return nullptr;
341 }
342 
lookUpDebuglinkObject(const std::string & Path,const ObjectFile * Obj,const std::string & ArchName)343 ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
344                                                   const ObjectFile *Obj,
345                                                   const std::string &ArchName) {
346   std::string DebuglinkName;
347   uint32_t CRCHash;
348   std::string DebugBinaryPath;
349   if (!getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash))
350     return nullptr;
351   if (!findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath))
352     return nullptr;
353   auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
354   if (!DbgObjOrErr) {
355     // Ignore errors, the file might not exist.
356     consumeError(DbgObjOrErr.takeError());
357     return nullptr;
358   }
359   return DbgObjOrErr.get();
360 }
361 
lookUpBuildIDObject(const std::string & Path,const ELFObjectFileBase * Obj,const std::string & ArchName)362 ObjectFile *LLVMSymbolizer::lookUpBuildIDObject(const std::string &Path,
363                                                 const ELFObjectFileBase *Obj,
364                                                 const std::string &ArchName) {
365   auto BuildID = getBuildID(Obj);
366   if (!BuildID)
367     return nullptr;
368   if (BuildID->size() < 2)
369     return nullptr;
370   std::string DebugBinaryPath;
371   if (!getOrFindDebugBinary(*BuildID, DebugBinaryPath))
372     return nullptr;
373   auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
374   if (!DbgObjOrErr) {
375     consumeError(DbgObjOrErr.takeError());
376     return nullptr;
377   }
378   return DbgObjOrErr.get();
379 }
380 
findDebugBinary(const std::string & OrigPath,const std::string & DebuglinkName,uint32_t CRCHash,std::string & Result)381 bool LLVMSymbolizer::findDebugBinary(const std::string &OrigPath,
382                                      const std::string &DebuglinkName,
383                                      uint32_t CRCHash, std::string &Result) {
384   SmallString<16> OrigDir(OrigPath);
385   llvm::sys::path::remove_filename(OrigDir);
386   SmallString<16> DebugPath = OrigDir;
387   // Try relative/path/to/original_binary/debuglink_name
388   llvm::sys::path::append(DebugPath, DebuglinkName);
389   if (checkFileCRC(DebugPath, CRCHash)) {
390     Result = std::string(DebugPath.str());
391     return true;
392   }
393   // Try relative/path/to/original_binary/.debug/debuglink_name
394   DebugPath = OrigDir;
395   llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
396   if (checkFileCRC(DebugPath, CRCHash)) {
397     Result = std::string(DebugPath.str());
398     return true;
399   }
400   // Make the path absolute so that lookups will go to
401   // "/usr/lib/debug/full/path/to/debug", not
402   // "/usr/lib/debug/to/debug"
403   llvm::sys::fs::make_absolute(OrigDir);
404   if (!Opts.FallbackDebugPath.empty()) {
405     // Try <FallbackDebugPath>/absolute/path/to/original_binary/debuglink_name
406     DebugPath = Opts.FallbackDebugPath;
407   } else {
408 #if defined(__NetBSD__)
409     // Try /usr/libdata/debug/absolute/path/to/original_binary/debuglink_name
410     DebugPath = "/usr/libdata/debug";
411 #else
412     // Try /usr/lib/debug/absolute/path/to/original_binary/debuglink_name
413     DebugPath = "/usr/lib/debug";
414 #endif
415   }
416   llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
417                           DebuglinkName);
418   if (checkFileCRC(DebugPath, CRCHash)) {
419     Result = std::string(DebugPath.str());
420     return true;
421   }
422   return false;
423 }
424 
getBuildIDStr(ArrayRef<uint8_t> BuildID)425 static StringRef getBuildIDStr(ArrayRef<uint8_t> BuildID) {
426   return StringRef(reinterpret_cast<const char *>(BuildID.data()),
427                    BuildID.size());
428 }
429 
getOrFindDebugBinary(const ArrayRef<uint8_t> BuildID,std::string & Result)430 bool LLVMSymbolizer::getOrFindDebugBinary(const ArrayRef<uint8_t> BuildID,
431                                           std::string &Result) {
432   StringRef BuildIDStr = getBuildIDStr(BuildID);
433   auto I = BuildIDPaths.find(BuildIDStr);
434   if (I != BuildIDPaths.end()) {
435     Result = I->second;
436     return true;
437   }
438   if (!BIDFetcher)
439     return false;
440   if (std::optional<std::string> Path = BIDFetcher->fetch(BuildID)) {
441     Result = *Path;
442     auto InsertResult = BuildIDPaths.insert({BuildIDStr, Result});
443     assert(InsertResult.second);
444     (void)InsertResult;
445     return true;
446   }
447 
448   return false;
449 }
450 
451 Expected<LLVMSymbolizer::ObjectPair>
getOrCreateObjectPair(const std::string & Path,const std::string & ArchName)452 LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
453                                       const std::string &ArchName) {
454   auto I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
455   if (I != ObjectPairForPathArch.end()) {
456     recordAccess(BinaryForPath.find(Path)->second);
457     return I->second;
458   }
459 
460   auto ObjOrErr = getOrCreateObject(Path, ArchName);
461   if (!ObjOrErr) {
462     ObjectPairForPathArch.emplace(std::make_pair(Path, ArchName),
463                                   ObjectPair(nullptr, nullptr));
464     return ObjOrErr.takeError();
465   }
466 
467   ObjectFile *Obj = ObjOrErr.get();
468   assert(Obj != nullptr);
469   ObjectFile *DbgObj = nullptr;
470 
471   if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
472     DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
473   else if (auto ELFObj = dyn_cast<const ELFObjectFileBase>(Obj))
474     DbgObj = lookUpBuildIDObject(Path, ELFObj, ArchName);
475   if (!DbgObj)
476     DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
477   if (!DbgObj)
478     DbgObj = Obj;
479   ObjectPair Res = std::make_pair(Obj, DbgObj);
480   std::string DbgObjPath = DbgObj->getFileName().str();
481   auto Pair =
482       ObjectPairForPathArch.emplace(std::make_pair(Path, ArchName), Res);
483   BinaryForPath.find(DbgObjPath)->second.pushEvictor([this, I = Pair.first]() {
484     ObjectPairForPathArch.erase(I);
485   });
486   return Res;
487 }
488 
489 Expected<ObjectFile *>
getOrCreateObject(const std::string & Path,const std::string & ArchName)490 LLVMSymbolizer::getOrCreateObject(const std::string &Path,
491                                   const std::string &ArchName) {
492   Binary *Bin;
493   auto Pair = BinaryForPath.emplace(Path, OwningBinary<Binary>());
494   if (!Pair.second) {
495     Bin = Pair.first->second->getBinary();
496     recordAccess(Pair.first->second);
497   } else {
498     Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
499     if (!BinOrErr)
500       return BinOrErr.takeError();
501 
502     CachedBinary &CachedBin = Pair.first->second;
503     CachedBin = std::move(BinOrErr.get());
504     CachedBin.pushEvictor([this, I = Pair.first]() { BinaryForPath.erase(I); });
505     LRUBinaries.push_back(CachedBin);
506     CacheSize += CachedBin.size();
507     Bin = CachedBin->getBinary();
508   }
509 
510   if (!Bin)
511     return static_cast<ObjectFile *>(nullptr);
512 
513   if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Bin)) {
514     auto I = ObjectForUBPathAndArch.find(std::make_pair(Path, ArchName));
515     if (I != ObjectForUBPathAndArch.end())
516       return I->second.get();
517 
518     Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
519         UB->getMachOObjectForArch(ArchName);
520     if (!ObjOrErr) {
521       ObjectForUBPathAndArch.emplace(std::make_pair(Path, ArchName),
522                                      std::unique_ptr<ObjectFile>());
523       return ObjOrErr.takeError();
524     }
525     ObjectFile *Res = ObjOrErr->get();
526     auto Pair = ObjectForUBPathAndArch.emplace(std::make_pair(Path, ArchName),
527                                                std::move(ObjOrErr.get()));
528     BinaryForPath.find(Path)->second.pushEvictor(
529         [this, Iter = Pair.first]() { ObjectForUBPathAndArch.erase(Iter); });
530     return Res;
531   }
532   if (Bin->isObject()) {
533     return cast<ObjectFile>(Bin);
534   }
535   return errorCodeToError(object_error::arch_not_found);
536 }
537 
538 Expected<SymbolizableModule *>
createModuleInfo(const ObjectFile * Obj,std::unique_ptr<DIContext> Context,StringRef ModuleName)539 LLVMSymbolizer::createModuleInfo(const ObjectFile *Obj,
540                                  std::unique_ptr<DIContext> Context,
541                                  StringRef ModuleName) {
542   auto InfoOrErr = SymbolizableObjectFile::create(Obj, std::move(Context),
543                                                   Opts.UntagAddresses);
544   std::unique_ptr<SymbolizableModule> SymMod;
545   if (InfoOrErr)
546     SymMod = std::move(*InfoOrErr);
547   auto InsertResult = Modules.insert(
548       std::make_pair(std::string(ModuleName), std::move(SymMod)));
549   assert(InsertResult.second);
550   if (!InfoOrErr)
551     return InfoOrErr.takeError();
552   return InsertResult.first->second.get();
553 }
554 
555 Expected<SymbolizableModule *>
getOrCreateModuleInfo(const std::string & ModuleName)556 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) {
557   std::string BinaryName = ModuleName;
558   std::string ArchName = Opts.DefaultArch;
559   size_t ColonPos = ModuleName.find_last_of(':');
560   // Verify that substring after colon form a valid arch name.
561   if (ColonPos != std::string::npos) {
562     std::string ArchStr = ModuleName.substr(ColonPos + 1);
563     if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
564       BinaryName = ModuleName.substr(0, ColonPos);
565       ArchName = ArchStr;
566     }
567   }
568 
569   auto I = Modules.find(ModuleName);
570   if (I != Modules.end()) {
571     recordAccess(BinaryForPath.find(BinaryName)->second);
572     return I->second.get();
573   }
574 
575   auto ObjectsOrErr = getOrCreateObjectPair(BinaryName, ArchName);
576   if (!ObjectsOrErr) {
577     // Failed to find valid object file.
578     Modules.emplace(ModuleName, std::unique_ptr<SymbolizableModule>());
579     return ObjectsOrErr.takeError();
580   }
581   ObjectPair Objects = ObjectsOrErr.get();
582 
583   std::unique_ptr<DIContext> Context;
584   // If this is a COFF object containing PDB info, use a PDBContext to
585   // symbolize. Otherwise, use DWARF.
586   if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
587     const codeview::DebugInfo *DebugInfo;
588     StringRef PDBFileName;
589     auto EC = CoffObject->getDebugPDBInfo(DebugInfo, PDBFileName);
590     if (!EC && DebugInfo != nullptr && !PDBFileName.empty()) {
591       using namespace pdb;
592       std::unique_ptr<IPDBSession> Session;
593 
594       PDB_ReaderType ReaderType =
595           Opts.UseDIA ? PDB_ReaderType::DIA : PDB_ReaderType::Native;
596       if (auto Err = loadDataForEXE(ReaderType, Objects.first->getFileName(),
597                                     Session)) {
598         Modules.emplace(ModuleName, std::unique_ptr<SymbolizableModule>());
599         // Return along the PDB filename to provide more context
600         return createFileError(PDBFileName, std::move(Err));
601       }
602       Context.reset(new PDBContext(*CoffObject, std::move(Session)));
603     }
604   }
605   if (!Context)
606     Context = DWARFContext::create(
607         *Objects.second, DWARFContext::ProcessDebugRelocations::Process,
608         nullptr, Opts.DWPName);
609   auto ModuleOrErr =
610       createModuleInfo(Objects.first, std::move(Context), ModuleName);
611   if (ModuleOrErr) {
612     auto I = Modules.find(ModuleName);
613     BinaryForPath.find(BinaryName)->second.pushEvictor([this, I]() {
614       Modules.erase(I);
615     });
616   }
617   return ModuleOrErr;
618 }
619 
620 Expected<SymbolizableModule *>
getOrCreateModuleInfo(const ObjectFile & Obj)621 LLVMSymbolizer::getOrCreateModuleInfo(const ObjectFile &Obj) {
622   StringRef ObjName = Obj.getFileName();
623   auto I = Modules.find(ObjName);
624   if (I != Modules.end())
625     return I->second.get();
626 
627   std::unique_ptr<DIContext> Context = DWARFContext::create(Obj);
628   // FIXME: handle COFF object with PDB info to use PDBContext
629   return createModuleInfo(&Obj, std::move(Context), ObjName);
630 }
631 
632 Expected<SymbolizableModule *>
getOrCreateModuleInfo(ArrayRef<uint8_t> BuildID)633 LLVMSymbolizer::getOrCreateModuleInfo(ArrayRef<uint8_t> BuildID) {
634   std::string Path;
635   if (!getOrFindDebugBinary(BuildID, Path)) {
636     return createStringError(errc::no_such_file_or_directory,
637                              Twine("could not find build ID '") +
638                                  toHex(BuildID) + "'");
639   }
640   return getOrCreateModuleInfo(Path);
641 }
642 
643 namespace {
644 
645 // Undo these various manglings for Win32 extern "C" functions:
646 // cdecl       - _foo
647 // stdcall     - _foo@12
648 // fastcall    - @foo@12
649 // vectorcall  - foo@@12
650 // These are all different linkage names for 'foo'.
demanglePE32ExternCFunc(StringRef SymbolName)651 StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
652   char Front = SymbolName.empty() ? '\0' : SymbolName[0];
653 
654   // Remove any '@[0-9]+' suffix.
655   bool HasAtNumSuffix = false;
656   if (Front != '?') {
657     size_t AtPos = SymbolName.rfind('@');
658     if (AtPos != StringRef::npos &&
659         all_of(drop_begin(SymbolName, AtPos + 1), isDigit)) {
660       SymbolName = SymbolName.substr(0, AtPos);
661       HasAtNumSuffix = true;
662     }
663   }
664 
665   // Remove any ending '@' for vectorcall.
666   bool IsVectorCall = false;
667   if (HasAtNumSuffix && SymbolName.endswith("@")) {
668     SymbolName = SymbolName.drop_back();
669     IsVectorCall = true;
670   }
671 
672   // If not vectorcall, remove any '_' or '@' prefix.
673   if (!IsVectorCall && (Front == '_' || Front == '@'))
674     SymbolName = SymbolName.drop_front();
675 
676   return SymbolName;
677 }
678 
679 } // end anonymous namespace
680 
681 std::string
DemangleName(const std::string & Name,const SymbolizableModule * DbiModuleDescriptor)682 LLVMSymbolizer::DemangleName(const std::string &Name,
683                              const SymbolizableModule *DbiModuleDescriptor) {
684   std::string Result;
685   if (nonMicrosoftDemangle(Name.c_str(), Result))
686     return Result;
687 
688   if (!Name.empty() && Name.front() == '?') {
689     // Only do MSVC C++ demangling on symbols starting with '?'.
690     int status = 0;
691     char *DemangledName = microsoftDemangle(
692         Name.c_str(), nullptr, nullptr, nullptr, &status,
693         MSDemangleFlags(MSDF_NoAccessSpecifier | MSDF_NoCallingConvention |
694                         MSDF_NoMemberType | MSDF_NoReturnType));
695     if (status != 0)
696       return Name;
697     Result = DemangledName;
698     free(DemangledName);
699     return Result;
700   }
701 
702   if (DbiModuleDescriptor && DbiModuleDescriptor->isWin32Module()) {
703     std::string DemangledCName(demanglePE32ExternCFunc(Name));
704     // On i386 Windows, the C name mangling for different calling conventions
705     // may also be applied on top of the Itanium or Rust name mangling.
706     if (nonMicrosoftDemangle(DemangledCName.c_str(), Result))
707       return Result;
708     return DemangledCName;
709   }
710   return Name;
711 }
712 
recordAccess(CachedBinary & Bin)713 void LLVMSymbolizer::recordAccess(CachedBinary &Bin) {
714   if (Bin->getBinary())
715     LRUBinaries.splice(LRUBinaries.end(), LRUBinaries, Bin.getIterator());
716 }
717 
pruneCache()718 void LLVMSymbolizer::pruneCache() {
719   // Evict the LRU binary until the max cache size is reached or there's <= 1
720   // item in the cache. The MRU binary is always kept to avoid thrashing if it's
721   // larger than the cache size.
722   while (CacheSize > Opts.MaxCacheSize && !LRUBinaries.empty() &&
723          std::next(LRUBinaries.begin()) != LRUBinaries.end()) {
724     CachedBinary &Bin = LRUBinaries.front();
725     CacheSize -= Bin.size();
726     LRUBinaries.pop_front();
727     Bin.evict();
728   }
729 }
730 
pushEvictor(std::function<void ()> NewEvictor)731 void CachedBinary::pushEvictor(std::function<void()> NewEvictor) {
732   if (Evictor) {
733     this->Evictor = [OldEvictor = std::move(this->Evictor),
734                      NewEvictor = std::move(NewEvictor)]() {
735       NewEvictor();
736       OldEvictor();
737     };
738   } else {
739     this->Evictor = std::move(NewEvictor);
740   }
741 }
742 
743 } // namespace symbolize
744 } // namespace llvm
745