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 "SymbolizableObjectFile.h"
16 
17 #include "llvm/ADT/STLExtras.h"
18 #include "llvm/BinaryFormat/COFF.h"
19 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
20 #include "llvm/DebugInfo/PDB/PDB.h"
21 #include "llvm/DebugInfo/PDB/PDBContext.h"
22 #include "llvm/Demangle/Demangle.h"
23 #include "llvm/Object/COFF.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/Compression.h"
29 #include "llvm/Support/DataExtractor.h"
30 #include "llvm/Support/Errc.h"
31 #include "llvm/Support/FileSystem.h"
32 #include "llvm/Support/MemoryBuffer.h"
33 #include "llvm/Support/Path.h"
34 #include <algorithm>
35 #include <cassert>
36 #include <cstring>
37 
38 #if defined(_MSC_VER)
39 #include <Windows.h>
40 
41 // This must be included after 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 
54 Expected<DILineInfo>
55 LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
56                               object::SectionedAddress ModuleOffset,
57                               StringRef DWPName) {
58   SymbolizableModule *Info;
59   if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName, DWPName))
60     Info = InfoOrErr.get();
61   else
62     return InfoOrErr.takeError();
63 
64   // A null module means an error has already been reported. Return an empty
65   // result.
66   if (!Info)
67     return DILineInfo();
68 
69   // If the user is giving us relative addresses, add the preferred base of the
70   // object to the offset before we do the query. It's what DIContext expects.
71   if (Opts.RelativeAddresses)
72     ModuleOffset.Address += Info->getModulePreferredBase();
73 
74   DILineInfo LineInfo = Info->symbolizeCode(ModuleOffset, Opts.PrintFunctions,
75                                             Opts.UseSymbolTable);
76   if (Opts.Demangle)
77     LineInfo.FunctionName = DemangleName(LineInfo.FunctionName, Info);
78   return LineInfo;
79 }
80 
81 Expected<DIInliningInfo>
82 LLVMSymbolizer::symbolizeInlinedCode(const std::string &ModuleName,
83                                      object::SectionedAddress ModuleOffset,
84                                      StringRef DWPName) {
85   SymbolizableModule *Info;
86   if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName, DWPName))
87     Info = InfoOrErr.get();
88   else
89     return InfoOrErr.takeError();
90 
91   // A null module means an error has already been reported. Return an empty
92   // result.
93   if (!Info)
94     return DIInliningInfo();
95 
96   // If the user is giving us relative addresses, add the preferred base of the
97   // object to the offset before we do the query. It's what DIContext expects.
98   if (Opts.RelativeAddresses)
99     ModuleOffset.Address += Info->getModulePreferredBase();
100 
101   DIInliningInfo InlinedContext = Info->symbolizeInlinedCode(
102       ModuleOffset, Opts.PrintFunctions, Opts.UseSymbolTable);
103   if (Opts.Demangle) {
104     for (int i = 0, n = InlinedContext.getNumberOfFrames(); i < n; i++) {
105       auto *Frame = InlinedContext.getMutableFrame(i);
106       Frame->FunctionName = DemangleName(Frame->FunctionName, Info);
107     }
108   }
109   return InlinedContext;
110 }
111 
112 Expected<DIGlobal>
113 LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
114                               object::SectionedAddress ModuleOffset) {
115   SymbolizableModule *Info;
116   if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName))
117     Info = InfoOrErr.get();
118   else
119     return InfoOrErr.takeError();
120 
121   // A null module means an error has already been reported. Return an empty
122   // result.
123   if (!Info)
124     return DIGlobal();
125 
126   // If the user is giving us relative addresses, add the preferred base of
127   // the object to the offset before we do the query. It's what DIContext
128   // expects.
129   if (Opts.RelativeAddresses)
130     ModuleOffset.Address += Info->getModulePreferredBase();
131 
132   DIGlobal Global = Info->symbolizeData(ModuleOffset);
133   if (Opts.Demangle)
134     Global.Name = DemangleName(Global.Name, Info);
135   return Global;
136 }
137 
138 void LLVMSymbolizer::flush() {
139   ObjectForUBPathAndArch.clear();
140   BinaryForPath.clear();
141   ObjectPairForPathArch.clear();
142   Modules.clear();
143 }
144 
145 namespace {
146 
147 // For Path="/path/to/foo" and Basename="foo" assume that debug info is in
148 // /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
149 // For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
150 // /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
151 std::string getDarwinDWARFResourceForPath(
152     const std::string &Path, const std::string &Basename) {
153   SmallString<16> ResourceName = StringRef(Path);
154   if (sys::path::extension(Path) != ".dSYM") {
155     ResourceName += ".dSYM";
156   }
157   sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
158   sys::path::append(ResourceName, Basename);
159   return ResourceName.str();
160 }
161 
162 bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
163   ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
164       MemoryBuffer::getFileOrSTDIN(Path);
165   if (!MB)
166     return false;
167   return CRCHash == llvm::crc32(0, MB.get()->getBuffer());
168 }
169 
170 bool findDebugBinary(const std::string &OrigPath,
171                      const std::string &DebuglinkName, uint32_t CRCHash,
172                      const std::string &FallbackDebugPath,
173                      std::string &Result) {
174   SmallString<16> OrigDir(OrigPath);
175   llvm::sys::path::remove_filename(OrigDir);
176   SmallString<16> DebugPath = OrigDir;
177   // Try relative/path/to/original_binary/debuglink_name
178   llvm::sys::path::append(DebugPath, DebuglinkName);
179   if (checkFileCRC(DebugPath, CRCHash)) {
180     Result = DebugPath.str();
181     return true;
182   }
183   // Try relative/path/to/original_binary/.debug/debuglink_name
184   DebugPath = OrigDir;
185   llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
186   if (checkFileCRC(DebugPath, CRCHash)) {
187     Result = DebugPath.str();
188     return true;
189   }
190   // Make the path absolute so that lookups will go to
191   // "/usr/lib/debug/full/path/to/debug", not
192   // "/usr/lib/debug/to/debug"
193   llvm::sys::fs::make_absolute(OrigDir);
194   if (!FallbackDebugPath.empty()) {
195     // Try <FallbackDebugPath>/absolute/path/to/original_binary/debuglink_name
196     DebugPath = FallbackDebugPath;
197   } else {
198 #if defined(__NetBSD__)
199     // Try /usr/libdata/debug/absolute/path/to/original_binary/debuglink_name
200     DebugPath = "/usr/libdata/debug";
201 #else
202     // Try /usr/lib/debug/absolute/path/to/original_binary/debuglink_name
203     DebugPath = "/usr/lib/debug";
204 #endif
205   }
206   llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
207                           DebuglinkName);
208   if (checkFileCRC(DebugPath, CRCHash)) {
209     Result = DebugPath.str();
210     return true;
211   }
212   return false;
213 }
214 
215 bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
216                              uint32_t &CRCHash) {
217   if (!Obj)
218     return false;
219   for (const SectionRef &Section : Obj->sections()) {
220     StringRef Name;
221     Section.getName(Name);
222     Name = Name.substr(Name.find_first_not_of("._"));
223     if (Name == "gnu_debuglink") {
224       StringRef Data;
225       Section.getContents(Data);
226       DataExtractor DE(Data, Obj->isLittleEndian(), 0);
227       uint32_t Offset = 0;
228       if (const char *DebugNameStr = DE.getCStr(&Offset)) {
229         // 4-byte align the offset.
230         Offset = (Offset + 3) & ~0x3;
231         if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
232           DebugName = DebugNameStr;
233           CRCHash = DE.getU32(&Offset);
234           return true;
235         }
236       }
237       break;
238     }
239   }
240   return false;
241 }
242 
243 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
244                              const MachOObjectFile *Obj) {
245   ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
246   ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
247   if (dbg_uuid.empty() || bin_uuid.empty())
248     return false;
249   return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
250 }
251 
252 } // end anonymous namespace
253 
254 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
255     const MachOObjectFile *MachExeObj, const std::string &ArchName) {
256   // On Darwin we may find DWARF in separate object file in
257   // resource directory.
258   std::vector<std::string> DsymPaths;
259   StringRef Filename = sys::path::filename(ExePath);
260   DsymPaths.push_back(getDarwinDWARFResourceForPath(ExePath, Filename));
261   for (const auto &Path : Opts.DsymHints) {
262     DsymPaths.push_back(getDarwinDWARFResourceForPath(Path, Filename));
263   }
264   for (const auto &Path : DsymPaths) {
265     auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
266     if (!DbgObjOrErr) {
267       // Ignore errors, the file might not exist.
268       consumeError(DbgObjOrErr.takeError());
269       continue;
270     }
271     ObjectFile *DbgObj = DbgObjOrErr.get();
272     if (!DbgObj)
273       continue;
274     const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(DbgObj);
275     if (!MachDbgObj)
276       continue;
277     if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj))
278       return DbgObj;
279   }
280   return nullptr;
281 }
282 
283 ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
284                                                   const ObjectFile *Obj,
285                                                   const std::string &ArchName) {
286   std::string DebuglinkName;
287   uint32_t CRCHash;
288   std::string DebugBinaryPath;
289   if (!getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash))
290     return nullptr;
291   if (!findDebugBinary(Path, DebuglinkName, CRCHash, Opts.FallbackDebugPath,
292                        DebugBinaryPath))
293     return nullptr;
294   auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
295   if (!DbgObjOrErr) {
296     // Ignore errors, the file might not exist.
297     consumeError(DbgObjOrErr.takeError());
298     return nullptr;
299   }
300   return DbgObjOrErr.get();
301 }
302 
303 Expected<LLVMSymbolizer::ObjectPair>
304 LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
305                                       const std::string &ArchName) {
306   const auto &I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
307   if (I != ObjectPairForPathArch.end()) {
308     return I->second;
309   }
310 
311   auto ObjOrErr = getOrCreateObject(Path, ArchName);
312   if (!ObjOrErr) {
313     ObjectPairForPathArch.insert(std::make_pair(std::make_pair(Path, ArchName),
314                                                 ObjectPair(nullptr, nullptr)));
315     return ObjOrErr.takeError();
316   }
317 
318   ObjectFile *Obj = ObjOrErr.get();
319   assert(Obj != nullptr);
320   ObjectFile *DbgObj = nullptr;
321 
322   if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
323     DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
324   if (!DbgObj)
325     DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
326   if (!DbgObj)
327     DbgObj = Obj;
328   ObjectPair Res = std::make_pair(Obj, DbgObj);
329   ObjectPairForPathArch.insert(
330       std::make_pair(std::make_pair(Path, ArchName), Res));
331   return Res;
332 }
333 
334 Expected<ObjectFile *>
335 LLVMSymbolizer::getOrCreateObject(const std::string &Path,
336                                   const std::string &ArchName) {
337   const auto &I = BinaryForPath.find(Path);
338   Binary *Bin = nullptr;
339   if (I == BinaryForPath.end()) {
340     Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
341     if (!BinOrErr) {
342       BinaryForPath.insert(std::make_pair(Path, OwningBinary<Binary>()));
343       return BinOrErr.takeError();
344     }
345     Bin = BinOrErr->getBinary();
346     BinaryForPath.insert(std::make_pair(Path, std::move(BinOrErr.get())));
347   } else {
348     Bin = I->second.getBinary();
349   }
350 
351   if (!Bin)
352     return static_cast<ObjectFile *>(nullptr);
353 
354   if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Bin)) {
355     const auto &I = ObjectForUBPathAndArch.find(std::make_pair(Path, ArchName));
356     if (I != ObjectForUBPathAndArch.end()) {
357       return I->second.get();
358     }
359     Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
360         UB->getObjectForArch(ArchName);
361     if (!ObjOrErr) {
362       ObjectForUBPathAndArch.insert(std::make_pair(
363           std::make_pair(Path, ArchName), std::unique_ptr<ObjectFile>()));
364       return ObjOrErr.takeError();
365     }
366     ObjectFile *Res = ObjOrErr->get();
367     ObjectForUBPathAndArch.insert(std::make_pair(std::make_pair(Path, ArchName),
368                                                  std::move(ObjOrErr.get())));
369     return Res;
370   }
371   if (Bin->isObject()) {
372     return cast<ObjectFile>(Bin);
373   }
374   return errorCodeToError(object_error::arch_not_found);
375 }
376 
377 Expected<SymbolizableModule *>
378 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName,
379                                       StringRef DWPName) {
380   const auto &I = Modules.find(ModuleName);
381   if (I != Modules.end()) {
382     return I->second.get();
383   }
384   std::string BinaryName = ModuleName;
385   std::string ArchName = Opts.DefaultArch;
386   size_t ColonPos = ModuleName.find_last_of(':');
387   // Verify that substring after colon form a valid arch name.
388   if (ColonPos != std::string::npos) {
389     std::string ArchStr = ModuleName.substr(ColonPos + 1);
390     if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
391       BinaryName = ModuleName.substr(0, ColonPos);
392       ArchName = ArchStr;
393     }
394   }
395   auto ObjectsOrErr = getOrCreateObjectPair(BinaryName, ArchName);
396   if (!ObjectsOrErr) {
397     // Failed to find valid object file.
398     Modules.insert(
399         std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
400     return ObjectsOrErr.takeError();
401   }
402   ObjectPair Objects = ObjectsOrErr.get();
403 
404   std::unique_ptr<DIContext> Context;
405   // If this is a COFF object containing PDB info, use a PDBContext to
406   // symbolize. Otherwise, use DWARF.
407   if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
408     const codeview::DebugInfo *DebugInfo;
409     StringRef PDBFileName;
410     auto EC = CoffObject->getDebugPDBInfo(DebugInfo, PDBFileName);
411     if (!EC && DebugInfo != nullptr && !PDBFileName.empty()) {
412       using namespace pdb;
413       std::unique_ptr<IPDBSession> Session;
414       if (auto Err = loadDataForEXE(PDB_ReaderType::DIA,
415                                     Objects.first->getFileName(), Session)) {
416         Modules.insert(
417             std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
418         // Return along the PDB filename to provide more context
419         return createFileError(PDBFileName, std::move(Err));
420       }
421       Context.reset(new PDBContext(*CoffObject, std::move(Session)));
422     }
423   }
424   if (!Context)
425     Context = DWARFContext::create(*Objects.second, nullptr,
426                                    DWARFContext::defaultErrorHandler, DWPName);
427   assert(Context);
428   auto InfoOrErr =
429       SymbolizableObjectFile::create(Objects.first, std::move(Context));
430   std::unique_ptr<SymbolizableModule> SymMod;
431   if (InfoOrErr)
432     SymMod = std::move(InfoOrErr.get());
433   auto InsertResult =
434       Modules.insert(std::make_pair(ModuleName, std::move(SymMod)));
435   assert(InsertResult.second);
436   if (auto EC = InfoOrErr.getError())
437     return errorCodeToError(EC);
438   return InsertResult.first->second.get();
439 }
440 
441 namespace {
442 
443 // Undo these various manglings for Win32 extern "C" functions:
444 // cdecl       - _foo
445 // stdcall     - _foo@12
446 // fastcall    - @foo@12
447 // vectorcall  - foo@@12
448 // These are all different linkage names for 'foo'.
449 StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
450   // Remove any '_' or '@' prefix.
451   char Front = SymbolName.empty() ? '\0' : SymbolName[0];
452   if (Front == '_' || Front == '@')
453     SymbolName = SymbolName.drop_front();
454 
455   // Remove any '@[0-9]+' suffix.
456   if (Front != '?') {
457     size_t AtPos = SymbolName.rfind('@');
458     if (AtPos != StringRef::npos &&
459         std::all_of(SymbolName.begin() + AtPos + 1, SymbolName.end(),
460                     [](char C) { return C >= '0' && C <= '9'; })) {
461       SymbolName = SymbolName.substr(0, AtPos);
462     }
463   }
464 
465   // Remove any ending '@' for vectorcall.
466   if (SymbolName.endswith("@"))
467     SymbolName = SymbolName.drop_back();
468 
469   return SymbolName;
470 }
471 
472 } // end anonymous namespace
473 
474 std::string
475 LLVMSymbolizer::DemangleName(const std::string &Name,
476                              const SymbolizableModule *DbiModuleDescriptor) {
477   // We can spoil names of symbols with C linkage, so use an heuristic
478   // approach to check if the name should be demangled.
479   if (Name.substr(0, 2) == "_Z") {
480     int status = 0;
481     char *DemangledName = itaniumDemangle(Name.c_str(), nullptr, nullptr, &status);
482     if (status != 0)
483       return Name;
484     std::string Result = DemangledName;
485     free(DemangledName);
486     return Result;
487   }
488 
489 #if defined(_MSC_VER)
490   if (!Name.empty() && Name.front() == '?') {
491     // Only do MSVC C++ demangling on symbols starting with '?'.
492     char DemangledName[1024] = {0};
493     DWORD result = ::UnDecorateSymbolName(
494         Name.c_str(), DemangledName, 1023,
495         UNDNAME_NO_ACCESS_SPECIFIERS |       // Strip public, private, protected
496             UNDNAME_NO_ALLOCATION_LANGUAGE | // Strip __thiscall, __stdcall, etc
497             UNDNAME_NO_THROW_SIGNATURES |    // Strip throw() specifications
498             UNDNAME_NO_MEMBER_TYPE | // Strip virtual, static, etc specifiers
499             UNDNAME_NO_MS_KEYWORDS | // Strip all MS extension keywords
500             UNDNAME_NO_FUNCTION_RETURNS); // Strip function return types
501     return (result == 0) ? Name : std::string(DemangledName);
502   }
503 #endif
504   if (DbiModuleDescriptor && DbiModuleDescriptor->isWin32Module())
505     return std::string(demanglePE32ExternCFunc(Name));
506   return Name;
507 }
508 
509 } // namespace symbolize
510 } // namespace llvm
511