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/Casting.h"
27 #include "llvm/Support/Compression.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 #if defined(_MSC_VER)
38 #include <Windows.h>
39 
40 // This must be included after windows.h.
41 #include <DbgHelp.h>
42 #pragma comment(lib, "dbghelp.lib")
43 
44 // Windows.h conflicts with our COFF header definitions.
45 #ifdef IMAGE_FILE_MACHINE_I386
46 #undef IMAGE_FILE_MACHINE_I386
47 #endif
48 #endif
49 
50 namespace llvm {
51 namespace symbolize {
52 
53 Expected<DILineInfo>
54 LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
55                               uint64_t ModuleOffset, StringRef DWPName) {
56   SymbolizableModule *Info;
57   if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName, DWPName))
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>
80 LLVMSymbolizer::symbolizeInlinedCode(const std::string &ModuleName,
81                                      uint64_t ModuleOffset, StringRef DWPName) {
82   SymbolizableModule *Info;
83   if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName, DWPName))
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 
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 
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.
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 
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 
166 bool findDebugBinary(const std::string &OrigPath,
167                      const std::string &DebuglinkName, uint32_t CRCHash,
168                      std::string &Result) {
169   SmallString<16> OrigDir(OrigPath);
170   llvm::sys::path::remove_filename(OrigDir);
171   SmallString<16> DebugPath = OrigDir;
172   // Try /path/to/original_binary/debuglink_name
173   llvm::sys::path::append(DebugPath, DebuglinkName);
174   if (checkFileCRC(DebugPath, CRCHash)) {
175     Result = DebugPath.str();
176     return true;
177   }
178   // Try /path/to/original_binary/.debug/debuglink_name
179   DebugPath = OrigDir;
180   llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
181   if (checkFileCRC(DebugPath, CRCHash)) {
182     Result = DebugPath.str();
183     return true;
184   }
185 #if defined(__NetBSD__)
186   // Try /usr/libdata/debug/path/to/original_binary/debuglink_name
187   DebugPath = "/usr/libdata/debug";
188 #else
189   // Try /usr/lib/debug/path/to/original_binary/debuglink_name
190   DebugPath = "/usr/lib/debug";
191 #endif
192   llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
193                           DebuglinkName);
194   if (checkFileCRC(DebugPath, CRCHash)) {
195     Result = DebugPath.str();
196     return true;
197   }
198   return false;
199 }
200 
201 bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
202                              uint32_t &CRCHash) {
203   if (!Obj)
204     return false;
205   for (const SectionRef &Section : Obj->sections()) {
206     StringRef Name;
207     Section.getName(Name);
208     Name = Name.substr(Name.find_first_not_of("._"));
209     if (Name == "gnu_debuglink") {
210       StringRef Data;
211       Section.getContents(Data);
212       DataExtractor DE(Data, Obj->isLittleEndian(), 0);
213       uint32_t Offset = 0;
214       if (const char *DebugNameStr = DE.getCStr(&Offset)) {
215         // 4-byte align the offset.
216         Offset = (Offset + 3) & ~0x3;
217         if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
218           DebugName = DebugNameStr;
219           CRCHash = DE.getU32(&Offset);
220           return true;
221         }
222       }
223       break;
224     }
225   }
226   return false;
227 }
228 
229 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
230                              const MachOObjectFile *Obj) {
231   ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
232   ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
233   if (dbg_uuid.empty() || bin_uuid.empty())
234     return false;
235   return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
236 }
237 
238 } // end anonymous namespace
239 
240 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
241     const MachOObjectFile *MachExeObj, const std::string &ArchName) {
242   // On Darwin we may find DWARF in separate object file in
243   // resource directory.
244   std::vector<std::string> DsymPaths;
245   StringRef Filename = sys::path::filename(ExePath);
246   DsymPaths.push_back(getDarwinDWARFResourceForPath(ExePath, Filename));
247   for (const auto &Path : Opts.DsymHints) {
248     DsymPaths.push_back(getDarwinDWARFResourceForPath(Path, Filename));
249   }
250   for (const auto &Path : DsymPaths) {
251     auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
252     if (!DbgObjOrErr) {
253       // Ignore errors, the file might not exist.
254       consumeError(DbgObjOrErr.takeError());
255       continue;
256     }
257     ObjectFile *DbgObj = DbgObjOrErr.get();
258     if (!DbgObj)
259       continue;
260     const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(DbgObj);
261     if (!MachDbgObj)
262       continue;
263     if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj))
264       return DbgObj;
265   }
266   return nullptr;
267 }
268 
269 ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
270                                                   const ObjectFile *Obj,
271                                                   const std::string &ArchName) {
272   std::string DebuglinkName;
273   uint32_t CRCHash;
274   std::string DebugBinaryPath;
275   if (!getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash))
276     return nullptr;
277   if (!findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath))
278     return nullptr;
279   auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
280   if (!DbgObjOrErr) {
281     // Ignore errors, the file might not exist.
282     consumeError(DbgObjOrErr.takeError());
283     return nullptr;
284   }
285   return DbgObjOrErr.get();
286 }
287 
288 Expected<LLVMSymbolizer::ObjectPair>
289 LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
290                                       const std::string &ArchName) {
291   const auto &I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
292   if (I != ObjectPairForPathArch.end()) {
293     return I->second;
294   }
295 
296   auto ObjOrErr = getOrCreateObject(Path, ArchName);
297   if (!ObjOrErr) {
298     ObjectPairForPathArch.insert(std::make_pair(std::make_pair(Path, ArchName),
299                                                 ObjectPair(nullptr, nullptr)));
300     return ObjOrErr.takeError();
301   }
302 
303   ObjectFile *Obj = ObjOrErr.get();
304   assert(Obj != nullptr);
305   ObjectFile *DbgObj = nullptr;
306 
307   if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
308     DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
309   if (!DbgObj)
310     DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
311   if (!DbgObj)
312     DbgObj = Obj;
313   ObjectPair Res = std::make_pair(Obj, DbgObj);
314   ObjectPairForPathArch.insert(
315       std::make_pair(std::make_pair(Path, ArchName), Res));
316   return Res;
317 }
318 
319 Expected<ObjectFile *>
320 LLVMSymbolizer::getOrCreateObject(const std::string &Path,
321                                   const std::string &ArchName) {
322   const auto &I = BinaryForPath.find(Path);
323   Binary *Bin = nullptr;
324   if (I == BinaryForPath.end()) {
325     Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
326     if (!BinOrErr) {
327       BinaryForPath.insert(std::make_pair(Path, OwningBinary<Binary>()));
328       return BinOrErr.takeError();
329     }
330     Bin = BinOrErr->getBinary();
331     BinaryForPath.insert(std::make_pair(Path, std::move(BinOrErr.get())));
332   } else {
333     Bin = I->second.getBinary();
334   }
335 
336   if (!Bin)
337     return static_cast<ObjectFile *>(nullptr);
338 
339   if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Bin)) {
340     const auto &I = ObjectForUBPathAndArch.find(std::make_pair(Path, ArchName));
341     if (I != ObjectForUBPathAndArch.end()) {
342       return I->second.get();
343     }
344     Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
345         UB->getObjectForArch(ArchName);
346     if (!ObjOrErr) {
347       ObjectForUBPathAndArch.insert(std::make_pair(
348           std::make_pair(Path, ArchName), std::unique_ptr<ObjectFile>()));
349       return ObjOrErr.takeError();
350     }
351     ObjectFile *Res = ObjOrErr->get();
352     ObjectForUBPathAndArch.insert(std::make_pair(std::make_pair(Path, ArchName),
353                                                  std::move(ObjOrErr.get())));
354     return Res;
355   }
356   if (Bin->isObject()) {
357     return cast<ObjectFile>(Bin);
358   }
359   return errorCodeToError(object_error::arch_not_found);
360 }
361 
362 Expected<SymbolizableModule *>
363 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName,
364                                       StringRef DWPName) {
365   const auto &I = Modules.find(ModuleName);
366   if (I != Modules.end()) {
367     return I->second.get();
368   }
369   std::string BinaryName = ModuleName;
370   std::string ArchName = Opts.DefaultArch;
371   size_t ColonPos = ModuleName.find_last_of(':');
372   // Verify that substring after colon form a valid arch name.
373   if (ColonPos != std::string::npos) {
374     std::string ArchStr = ModuleName.substr(ColonPos + 1);
375     if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
376       BinaryName = ModuleName.substr(0, ColonPos);
377       ArchName = ArchStr;
378     }
379   }
380   auto ObjectsOrErr = getOrCreateObjectPair(BinaryName, ArchName);
381   if (!ObjectsOrErr) {
382     // Failed to find valid object file.
383     Modules.insert(
384         std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
385     return ObjectsOrErr.takeError();
386   }
387   ObjectPair Objects = ObjectsOrErr.get();
388 
389   std::unique_ptr<DIContext> Context;
390   // If this is a COFF object containing PDB info, use a PDBContext to
391   // symbolize. Otherwise, use DWARF.
392   if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
393     const codeview::DebugInfo *DebugInfo;
394     StringRef PDBFileName;
395     auto EC = CoffObject->getDebugPDBInfo(DebugInfo, PDBFileName);
396     if (!EC && DebugInfo != nullptr && !PDBFileName.empty()) {
397       using namespace pdb;
398       std::unique_ptr<IPDBSession> Session;
399       if (auto Err = loadDataForEXE(PDB_ReaderType::DIA,
400                                     Objects.first->getFileName(), Session)) {
401         Modules.insert(
402             std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
403         // Return along the PDB filename to provide more context
404         return createFileError(PDBFileName, std::move(Err));
405       }
406       Context.reset(new PDBContext(*CoffObject, std::move(Session)));
407     }
408   }
409   if (!Context)
410     Context = DWARFContext::create(*Objects.second, nullptr,
411                                    DWARFContext::defaultErrorHandler, DWPName);
412   assert(Context);
413   auto InfoOrErr =
414       SymbolizableObjectFile::create(Objects.first, std::move(Context));
415   std::unique_ptr<SymbolizableModule> SymMod;
416   if (InfoOrErr)
417     SymMod = std::move(InfoOrErr.get());
418   auto InsertResult =
419       Modules.insert(std::make_pair(ModuleName, std::move(SymMod)));
420   assert(InsertResult.second);
421   if (auto EC = InfoOrErr.getError())
422     return errorCodeToError(EC);
423   return InsertResult.first->second.get();
424 }
425 
426 namespace {
427 
428 // Undo these various manglings for Win32 extern "C" functions:
429 // cdecl       - _foo
430 // stdcall     - _foo@12
431 // fastcall    - @foo@12
432 // vectorcall  - foo@@12
433 // These are all different linkage names for 'foo'.
434 StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
435   // Remove any '_' or '@' prefix.
436   char Front = SymbolName.empty() ? '\0' : SymbolName[0];
437   if (Front == '_' || Front == '@')
438     SymbolName = SymbolName.drop_front();
439 
440   // Remove any '@[0-9]+' suffix.
441   if (Front != '?') {
442     size_t AtPos = SymbolName.rfind('@');
443     if (AtPos != StringRef::npos &&
444         std::all_of(SymbolName.begin() + AtPos + 1, SymbolName.end(),
445                     [](char C) { return C >= '0' && C <= '9'; })) {
446       SymbolName = SymbolName.substr(0, AtPos);
447     }
448   }
449 
450   // Remove any ending '@' for vectorcall.
451   if (SymbolName.endswith("@"))
452     SymbolName = SymbolName.drop_back();
453 
454   return SymbolName;
455 }
456 
457 } // end anonymous namespace
458 
459 std::string
460 LLVMSymbolizer::DemangleName(const std::string &Name,
461                              const SymbolizableModule *DbiModuleDescriptor) {
462   // We can spoil names of symbols with C linkage, so use an heuristic
463   // approach to check if the name should be demangled.
464   if (Name.substr(0, 2) == "_Z") {
465     int status = 0;
466     char *DemangledName = itaniumDemangle(Name.c_str(), nullptr, nullptr, &status);
467     if (status != 0)
468       return Name;
469     std::string Result = DemangledName;
470     free(DemangledName);
471     return Result;
472   }
473 
474 #if defined(_MSC_VER)
475   if (!Name.empty() && Name.front() == '?') {
476     // Only do MSVC C++ demangling on symbols starting with '?'.
477     char DemangledName[1024] = {0};
478     DWORD result = ::UnDecorateSymbolName(
479         Name.c_str(), DemangledName, 1023,
480         UNDNAME_NO_ACCESS_SPECIFIERS |       // Strip public, private, protected
481             UNDNAME_NO_ALLOCATION_LANGUAGE | // Strip __thiscall, __stdcall, etc
482             UNDNAME_NO_THROW_SIGNATURES |    // Strip throw() specifications
483             UNDNAME_NO_MEMBER_TYPE | // Strip virtual, static, etc specifiers
484             UNDNAME_NO_MS_KEYWORDS | // Strip all MS extension keywords
485             UNDNAME_NO_FUNCTION_RETURNS); // Strip function return types
486     return (result == 0) ? Name : std::string(DemangledName);
487   }
488 #endif
489   if (DbiModuleDescriptor && DbiModuleDescriptor->isWin32Module())
490     return std::string(demanglePE32ExternCFunc(Name));
491   return Name;
492 }
493 
494 } // namespace symbolize
495 } // namespace llvm
496