1 //===-- LLVMSymbolize.cpp -------------------------------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // Implementation for LLVM symbolization library.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/DebugInfo/Symbolize/Symbolize.h"
15 
16 #include "SymbolizableObjectFile.h"
17 
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/BinaryFormat/COFF.h"
20 #include "llvm/Config/config.h"
21 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
22 #include "llvm/DebugInfo/PDB/PDB.h"
23 #include "llvm/DebugInfo/PDB/PDBContext.h"
24 #include "llvm/Demangle/Demangle.h"
25 #include "llvm/Object/COFF.h"
26 #include "llvm/Object/MachO.h"
27 #include "llvm/Object/MachOUniversal.h"
28 #include "llvm/Support/Casting.h"
29 #include "llvm/Support/Compression.h"
30 #include "llvm/Support/DataExtractor.h"
31 #include "llvm/Support/Errc.h"
32 #include "llvm/Support/FileSystem.h"
33 #include "llvm/Support/MemoryBuffer.h"
34 #include "llvm/Support/Path.h"
35 #include <algorithm>
36 #include <cassert>
37 #include <cstdlib>
38 #include <cstring>
39 
40 #if defined(_MSC_VER)
41 #include <Windows.h>
42 
43 // This must be included after windows.h.
44 #include <DbgHelp.h>
45 #pragma comment(lib, "dbghelp.lib")
46 
47 // Windows.h conflicts with our COFF header definitions.
48 #ifdef IMAGE_FILE_MACHINE_I386
49 #undef IMAGE_FILE_MACHINE_I386
50 #endif
51 #endif
52 
53 namespace llvm {
54 namespace symbolize {
55 
56 Expected<DILineInfo>
symbolizeCode(const std::string & ModuleName,uint64_t ModuleOffset,StringRef DWPName)57 LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
58                               uint64_t ModuleOffset, StringRef DWPName) {
59   SymbolizableModule *Info;
60   if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName, DWPName))
61     Info = InfoOrErr.get();
62   else
63     return InfoOrErr.takeError();
64 
65   // A null module means an error has already been reported. Return an empty
66   // result.
67   if (!Info)
68     return DILineInfo();
69 
70   // If the user is giving us relative addresses, add the preferred base of the
71   // object to the offset before we do the query. It's what DIContext expects.
72   if (Opts.RelativeAddresses)
73     ModuleOffset += Info->getModulePreferredBase();
74 
75   DILineInfo LineInfo = Info->symbolizeCode(ModuleOffset, Opts.PrintFunctions,
76                                             Opts.UseSymbolTable);
77   if (Opts.Demangle)
78     LineInfo.FunctionName = DemangleName(LineInfo.FunctionName, Info);
79   return LineInfo;
80 }
81 
82 Expected<DIInliningInfo>
symbolizeInlinedCode(const std::string & ModuleName,uint64_t ModuleOffset,StringRef DWPName)83 LLVMSymbolizer::symbolizeInlinedCode(const std::string &ModuleName,
84                                      uint64_t ModuleOffset, 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 += 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 
symbolizeData(const std::string & ModuleName,uint64_t ModuleOffset)112 Expected<DIGlobal> LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
113                                                  uint64_t ModuleOffset) {
114   SymbolizableModule *Info;
115   if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName))
116     Info = InfoOrErr.get();
117   else
118     return InfoOrErr.takeError();
119 
120   // A null module means an error has already been reported. Return an empty
121   // result.
122   if (!Info)
123     return DIGlobal();
124 
125   // If the user is giving us relative addresses, add the preferred base of
126   // the object to the offset before we do the query. It's what DIContext
127   // expects.
128   if (Opts.RelativeAddresses)
129     ModuleOffset += Info->getModulePreferredBase();
130 
131   DIGlobal Global = Info->symbolizeData(ModuleOffset);
132   if (Opts.Demangle)
133     Global.Name = DemangleName(Global.Name, Info);
134   return Global;
135 }
136 
flush()137 void LLVMSymbolizer::flush() {
138   ObjectForUBPathAndArch.clear();
139   BinaryForPath.clear();
140   ObjectPairForPathArch.clear();
141   Modules.clear();
142 }
143 
144 namespace {
145 
146 // For Path="/path/to/foo" and Basename="foo" assume that debug info is in
147 // /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
148 // For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
149 // /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
getDarwinDWARFResourceForPath(const std::string & Path,const std::string & Basename)150 std::string getDarwinDWARFResourceForPath(
151     const std::string &Path, const std::string &Basename) {
152   SmallString<16> ResourceName = StringRef(Path);
153   if (sys::path::extension(Path) != ".dSYM") {
154     ResourceName += ".dSYM";
155   }
156   sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
157   sys::path::append(ResourceName, Basename);
158   return ResourceName.str();
159 }
160 
checkFileCRC(StringRef Path,uint32_t CRCHash)161 bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
162   ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
163       MemoryBuffer::getFileOrSTDIN(Path);
164   if (!MB)
165     return false;
166   return !zlib::isAvailable() || CRCHash == zlib::crc32(MB.get()->getBuffer());
167 }
168 
findDebugBinary(const std::string & OrigPath,const std::string & DebuglinkName,uint32_t CRCHash,std::string & Result)169 bool findDebugBinary(const std::string &OrigPath,
170                      const std::string &DebuglinkName, uint32_t CRCHash,
171                      std::string &Result) {
172   std::string OrigRealPath = OrigPath;
173 #if defined(HAVE_REALPATH)
174   if (char *RP = realpath(OrigPath.c_str(), nullptr)) {
175     OrigRealPath = RP;
176     free(RP);
177   }
178 #endif
179   SmallString<16> OrigDir(OrigRealPath);
180   llvm::sys::path::remove_filename(OrigDir);
181   SmallString<16> DebugPath = OrigDir;
182   // Try /path/to/original_binary/debuglink_name
183   llvm::sys::path::append(DebugPath, DebuglinkName);
184   if (checkFileCRC(DebugPath, CRCHash)) {
185     Result = DebugPath.str();
186     return true;
187   }
188   // Try /path/to/original_binary/.debug/debuglink_name
189   DebugPath = OrigDir;
190   llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
191   if (checkFileCRC(DebugPath, CRCHash)) {
192     Result = DebugPath.str();
193     return true;
194   }
195 #if defined(__NetBSD__)
196   // Try /usr/libdata/debug/path/to/original_binary/debuglink_name
197   DebugPath = "/usr/libdata/debug";
198 #else
199   // Try /usr/lib/debug/path/to/original_binary/debuglink_name
200   DebugPath = "/usr/lib/debug";
201 #endif
202   llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
203                           DebuglinkName);
204   if (checkFileCRC(DebugPath, CRCHash)) {
205     Result = DebugPath.str();
206     return true;
207   }
208   return false;
209 }
210 
getGNUDebuglinkContents(const ObjectFile * Obj,std::string & DebugName,uint32_t & CRCHash)211 bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
212                              uint32_t &CRCHash) {
213   if (!Obj)
214     return false;
215   for (const SectionRef &Section : Obj->sections()) {
216     StringRef Name;
217     Section.getName(Name);
218     Name = Name.substr(Name.find_first_not_of("._"));
219     if (Name == "gnu_debuglink") {
220       StringRef Data;
221       Section.getContents(Data);
222       DataExtractor DE(Data, Obj->isLittleEndian(), 0);
223       uint32_t Offset = 0;
224       if (const char *DebugNameStr = DE.getCStr(&Offset)) {
225         // 4-byte align the offset.
226         Offset = (Offset + 3) & ~0x3;
227         if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
228           DebugName = DebugNameStr;
229           CRCHash = DE.getU32(&Offset);
230           return true;
231         }
232       }
233       break;
234     }
235   }
236   return false;
237 }
238 
darwinDsymMatchesBinary(const MachOObjectFile * DbgObj,const MachOObjectFile * Obj)239 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
240                              const MachOObjectFile *Obj) {
241   ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
242   ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
243   if (dbg_uuid.empty() || bin_uuid.empty())
244     return false;
245   return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
246 }
247 
248 } // end anonymous namespace
249 
lookUpDsymFile(const std::string & ExePath,const MachOObjectFile * MachExeObj,const std::string & ArchName)250 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
251     const MachOObjectFile *MachExeObj, const std::string &ArchName) {
252   // On Darwin we may find DWARF in separate object file in
253   // resource directory.
254   std::vector<std::string> DsymPaths;
255   StringRef Filename = sys::path::filename(ExePath);
256   DsymPaths.push_back(getDarwinDWARFResourceForPath(ExePath, Filename));
257   for (const auto &Path : Opts.DsymHints) {
258     DsymPaths.push_back(getDarwinDWARFResourceForPath(Path, Filename));
259   }
260   for (const auto &Path : DsymPaths) {
261     auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
262     if (!DbgObjOrErr) {
263       // Ignore errors, the file might not exist.
264       consumeError(DbgObjOrErr.takeError());
265       continue;
266     }
267     ObjectFile *DbgObj = DbgObjOrErr.get();
268     if (!DbgObj)
269       continue;
270     const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(DbgObj);
271     if (!MachDbgObj)
272       continue;
273     if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj))
274       return DbgObj;
275   }
276   return nullptr;
277 }
278 
lookUpDebuglinkObject(const std::string & Path,const ObjectFile * Obj,const std::string & ArchName)279 ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
280                                                   const ObjectFile *Obj,
281                                                   const std::string &ArchName) {
282   std::string DebuglinkName;
283   uint32_t CRCHash;
284   std::string DebugBinaryPath;
285   if (!getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash))
286     return nullptr;
287   if (!findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath))
288     return nullptr;
289   auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
290   if (!DbgObjOrErr) {
291     // Ignore errors, the file might not exist.
292     consumeError(DbgObjOrErr.takeError());
293     return nullptr;
294   }
295   return DbgObjOrErr.get();
296 }
297 
298 Expected<LLVMSymbolizer::ObjectPair>
getOrCreateObjectPair(const std::string & Path,const std::string & ArchName)299 LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
300                                       const std::string &ArchName) {
301   const auto &I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
302   if (I != ObjectPairForPathArch.end()) {
303     return I->second;
304   }
305 
306   auto ObjOrErr = getOrCreateObject(Path, ArchName);
307   if (!ObjOrErr) {
308     ObjectPairForPathArch.insert(std::make_pair(std::make_pair(Path, ArchName),
309                                                 ObjectPair(nullptr, nullptr)));
310     return ObjOrErr.takeError();
311   }
312 
313   ObjectFile *Obj = ObjOrErr.get();
314   assert(Obj != nullptr);
315   ObjectFile *DbgObj = nullptr;
316 
317   if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
318     DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
319   if (!DbgObj)
320     DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
321   if (!DbgObj)
322     DbgObj = Obj;
323   ObjectPair Res = std::make_pair(Obj, DbgObj);
324   ObjectPairForPathArch.insert(
325       std::make_pair(std::make_pair(Path, ArchName), Res));
326   return Res;
327 }
328 
329 Expected<ObjectFile *>
getOrCreateObject(const std::string & Path,const std::string & ArchName)330 LLVMSymbolizer::getOrCreateObject(const std::string &Path,
331                                   const std::string &ArchName) {
332   const auto &I = BinaryForPath.find(Path);
333   Binary *Bin = nullptr;
334   if (I == BinaryForPath.end()) {
335     Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
336     if (!BinOrErr) {
337       BinaryForPath.insert(std::make_pair(Path, OwningBinary<Binary>()));
338       return BinOrErr.takeError();
339     }
340     Bin = BinOrErr->getBinary();
341     BinaryForPath.insert(std::make_pair(Path, std::move(BinOrErr.get())));
342   } else {
343     Bin = I->second.getBinary();
344   }
345 
346   if (!Bin)
347     return static_cast<ObjectFile *>(nullptr);
348 
349   if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Bin)) {
350     const auto &I = ObjectForUBPathAndArch.find(std::make_pair(Path, ArchName));
351     if (I != ObjectForUBPathAndArch.end()) {
352       return I->second.get();
353     }
354     Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
355         UB->getObjectForArch(ArchName);
356     if (!ObjOrErr) {
357       ObjectForUBPathAndArch.insert(std::make_pair(
358           std::make_pair(Path, ArchName), std::unique_ptr<ObjectFile>()));
359       return ObjOrErr.takeError();
360     }
361     ObjectFile *Res = ObjOrErr->get();
362     ObjectForUBPathAndArch.insert(std::make_pair(std::make_pair(Path, ArchName),
363                                                  std::move(ObjOrErr.get())));
364     return Res;
365   }
366   if (Bin->isObject()) {
367     return cast<ObjectFile>(Bin);
368   }
369   return errorCodeToError(object_error::arch_not_found);
370 }
371 
372 Expected<SymbolizableModule *>
getOrCreateModuleInfo(const std::string & ModuleName,StringRef DWPName)373 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName,
374                                       StringRef DWPName) {
375   const auto &I = Modules.find(ModuleName);
376   if (I != Modules.end()) {
377     return I->second.get();
378   }
379   std::string BinaryName = ModuleName;
380   std::string ArchName = Opts.DefaultArch;
381   size_t ColonPos = ModuleName.find_last_of(':');
382   // Verify that substring after colon form a valid arch name.
383   if (ColonPos != std::string::npos) {
384     std::string ArchStr = ModuleName.substr(ColonPos + 1);
385     if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
386       BinaryName = ModuleName.substr(0, ColonPos);
387       ArchName = ArchStr;
388     }
389   }
390   auto ObjectsOrErr = getOrCreateObjectPair(BinaryName, ArchName);
391   if (!ObjectsOrErr) {
392     // Failed to find valid object file.
393     Modules.insert(
394         std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
395     return ObjectsOrErr.takeError();
396   }
397   ObjectPair Objects = ObjectsOrErr.get();
398 
399   std::unique_ptr<DIContext> Context;
400   // If this is a COFF object containing PDB info, use a PDBContext to
401   // symbolize. Otherwise, use DWARF.
402   if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
403     const codeview::DebugInfo *DebugInfo;
404     StringRef PDBFileName;
405     auto EC = CoffObject->getDebugPDBInfo(DebugInfo, PDBFileName);
406     if (!EC && DebugInfo != nullptr && !PDBFileName.empty()) {
407       using namespace pdb;
408       std::unique_ptr<IPDBSession> Session;
409       if (auto Err = loadDataForEXE(PDB_ReaderType::DIA,
410                                     Objects.first->getFileName(), Session)) {
411         Modules.insert(
412             std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
413         return std::move(Err);
414       }
415       Context.reset(new PDBContext(*CoffObject, std::move(Session)));
416     }
417   }
418   if (!Context)
419     Context = DWARFContext::create(*Objects.second, nullptr,
420                                    DWARFContext::defaultErrorHandler, DWPName);
421   assert(Context);
422   auto InfoOrErr =
423       SymbolizableObjectFile::create(Objects.first, std::move(Context));
424   std::unique_ptr<SymbolizableModule> SymMod;
425   if (InfoOrErr)
426     SymMod = std::move(InfoOrErr.get());
427   auto InsertResult =
428       Modules.insert(std::make_pair(ModuleName, std::move(SymMod)));
429   assert(InsertResult.second);
430   if (auto EC = InfoOrErr.getError())
431     return errorCodeToError(EC);
432   return InsertResult.first->second.get();
433 }
434 
435 namespace {
436 
437 // Undo these various manglings for Win32 extern "C" functions:
438 // cdecl       - _foo
439 // stdcall     - _foo@12
440 // fastcall    - @foo@12
441 // vectorcall  - foo@@12
442 // These are all different linkage names for 'foo'.
demanglePE32ExternCFunc(StringRef SymbolName)443 StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
444   // Remove any '_' or '@' prefix.
445   char Front = SymbolName.empty() ? '\0' : SymbolName[0];
446   if (Front == '_' || Front == '@')
447     SymbolName = SymbolName.drop_front();
448 
449   // Remove any '@[0-9]+' suffix.
450   if (Front != '?') {
451     size_t AtPos = SymbolName.rfind('@');
452     if (AtPos != StringRef::npos &&
453         std::all_of(SymbolName.begin() + AtPos + 1, SymbolName.end(),
454                     [](char C) { return C >= '0' && C <= '9'; })) {
455       SymbolName = SymbolName.substr(0, AtPos);
456     }
457   }
458 
459   // Remove any ending '@' for vectorcall.
460   if (SymbolName.endswith("@"))
461     SymbolName = SymbolName.drop_back();
462 
463   return SymbolName;
464 }
465 
466 } // end anonymous namespace
467 
468 std::string
DemangleName(const std::string & Name,const SymbolizableModule * DbiModuleDescriptor)469 LLVMSymbolizer::DemangleName(const std::string &Name,
470                              const SymbolizableModule *DbiModuleDescriptor) {
471   // We can spoil names of symbols with C linkage, so use an heuristic
472   // approach to check if the name should be demangled.
473   if (Name.substr(0, 2) == "_Z") {
474     int status = 0;
475     char *DemangledName = itaniumDemangle(Name.c_str(), nullptr, nullptr, &status);
476     if (status != 0)
477       return Name;
478     std::string Result = DemangledName;
479     free(DemangledName);
480     return Result;
481   }
482 
483 #if defined(_MSC_VER)
484   if (!Name.empty() && Name.front() == '?') {
485     // Only do MSVC C++ demangling on symbols starting with '?'.
486     char DemangledName[1024] = {0};
487     DWORD result = ::UnDecorateSymbolName(
488         Name.c_str(), DemangledName, 1023,
489         UNDNAME_NO_ACCESS_SPECIFIERS |       // Strip public, private, protected
490             UNDNAME_NO_ALLOCATION_LANGUAGE | // Strip __thiscall, __stdcall, etc
491             UNDNAME_NO_THROW_SIGNATURES |    // Strip throw() specifications
492             UNDNAME_NO_MEMBER_TYPE | // Strip virtual, static, etc specifiers
493             UNDNAME_NO_MS_KEYWORDS | // Strip all MS extension keywords
494             UNDNAME_NO_FUNCTION_RETURNS); // Strip function return types
495     return (result == 0) ? Name : std::string(DemangledName);
496   }
497 #endif
498   if (DbiModuleDescriptor && DbiModuleDescriptor->isWin32Module())
499     return std::string(demanglePE32ExternCFunc(Name));
500   return Name;
501 }
502 
503 } // namespace symbolize
504 } // namespace llvm
505