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/Object/COFF.h"
25 #include "llvm/Object/ELFObjectFile.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>
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>
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 
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 
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.
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 
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 
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 = OrigRealPath;
190   llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
191   if (checkFileCRC(DebugPath, CRCHash)) {
192     Result = DebugPath.str();
193     return true;
194   }
195   // Try /usr/lib/debug/path/to/original_binary/debuglink_name
196   DebugPath = "/usr/lib/debug";
197   llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
198                           DebuglinkName);
199   if (checkFileCRC(DebugPath, CRCHash)) {
200     Result = DebugPath.str();
201     return true;
202   }
203   return false;
204 }
205 
206 bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
207                              uint32_t &CRCHash) {
208   if (!Obj)
209     return false;
210   for (const SectionRef &Section : Obj->sections()) {
211     StringRef Name;
212     Section.getName(Name);
213     Name = Name.substr(Name.find_first_not_of("._"));
214     if (Name == "gnu_debuglink") {
215       StringRef Data;
216       Section.getContents(Data);
217       DataExtractor DE(Data, Obj->isLittleEndian(), 0);
218       uint32_t Offset = 0;
219       if (const char *DebugNameStr = DE.getCStr(&Offset)) {
220         // 4-byte align the offset.
221         Offset = (Offset + 3) & ~0x3;
222         if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
223           DebugName = DebugNameStr;
224           CRCHash = DE.getU32(&Offset);
225           return true;
226         }
227       }
228       break;
229     }
230   }
231   return false;
232 }
233 
234 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
235                              const MachOObjectFile *Obj) {
236   ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
237   ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
238   if (dbg_uuid.empty() || bin_uuid.empty())
239     return false;
240   return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
241 }
242 
243 } // end anonymous namespace
244 
245 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
246     const MachOObjectFile *MachExeObj, const std::string &ArchName) {
247   // On Darwin we may find DWARF in separate object file in
248   // resource directory.
249   std::vector<std::string> DsymPaths;
250   StringRef Filename = sys::path::filename(ExePath);
251   DsymPaths.push_back(getDarwinDWARFResourceForPath(ExePath, Filename));
252   for (const auto &Path : Opts.DsymHints) {
253     DsymPaths.push_back(getDarwinDWARFResourceForPath(Path, Filename));
254   }
255   for (const auto &Path : DsymPaths) {
256     auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
257     if (!DbgObjOrErr) {
258       // Ignore errors, the file might not exist.
259       consumeError(DbgObjOrErr.takeError());
260       continue;
261     }
262     ObjectFile *DbgObj = DbgObjOrErr.get();
263     if (!DbgObj)
264       continue;
265     const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(DbgObj);
266     if (!MachDbgObj)
267       continue;
268     if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj))
269       return DbgObj;
270   }
271   return nullptr;
272 }
273 
274 ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
275                                                   const ObjectFile *Obj,
276                                                   const std::string &ArchName) {
277   std::string DebuglinkName;
278   uint32_t CRCHash;
279   std::string DebugBinaryPath;
280   if (!getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash))
281     return nullptr;
282   if (!findDebugBinary(Path, DebuglinkName, CRCHash, DebugBinaryPath))
283     return nullptr;
284   auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
285   if (!DbgObjOrErr) {
286     // Ignore errors, the file might not exist.
287     consumeError(DbgObjOrErr.takeError());
288     return nullptr;
289   }
290   return DbgObjOrErr.get();
291 }
292 
293 Expected<LLVMSymbolizer::ObjectPair>
294 LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
295                                       const std::string &ArchName) {
296   const auto &I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
297   if (I != ObjectPairForPathArch.end()) {
298     return I->second;
299   }
300 
301   auto ObjOrErr = getOrCreateObject(Path, ArchName);
302   if (!ObjOrErr) {
303     ObjectPairForPathArch.insert(std::make_pair(std::make_pair(Path, ArchName),
304                                                 ObjectPair(nullptr, nullptr)));
305     return ObjOrErr.takeError();
306   }
307 
308   ObjectFile *Obj = ObjOrErr.get();
309   assert(Obj != nullptr);
310   ObjectFile *DbgObj = nullptr;
311 
312   if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
313     DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
314   if (!DbgObj)
315     DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
316   if (!DbgObj)
317     DbgObj = Obj;
318   ObjectPair Res = std::make_pair(Obj, DbgObj);
319   ObjectPairForPathArch.insert(
320       std::make_pair(std::make_pair(Path, ArchName), Res));
321   return Res;
322 }
323 
324 Expected<ObjectFile *>
325 LLVMSymbolizer::getOrCreateObject(const std::string &Path,
326                                   const std::string &ArchName) {
327   const auto &I = BinaryForPath.find(Path);
328   Binary *Bin = nullptr;
329   if (I == BinaryForPath.end()) {
330     Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
331     if (!BinOrErr) {
332       BinaryForPath.insert(std::make_pair(Path, OwningBinary<Binary>()));
333       return BinOrErr.takeError();
334     }
335     Bin = BinOrErr->getBinary();
336     BinaryForPath.insert(std::make_pair(Path, std::move(BinOrErr.get())));
337   } else {
338     Bin = I->second.getBinary();
339   }
340 
341   if (!Bin)
342     return static_cast<ObjectFile *>(nullptr);
343 
344   if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Bin)) {
345     const auto &I = ObjectForUBPathAndArch.find(std::make_pair(Path, ArchName));
346     if (I != ObjectForUBPathAndArch.end()) {
347       return I->second.get();
348     }
349     Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
350         UB->getObjectForArch(ArchName);
351     if (!ObjOrErr) {
352       ObjectForUBPathAndArch.insert(std::make_pair(
353           std::make_pair(Path, ArchName), std::unique_ptr<ObjectFile>()));
354       return ObjOrErr.takeError();
355     }
356     ObjectFile *Res = ObjOrErr->get();
357     ObjectForUBPathAndArch.insert(std::make_pair(std::make_pair(Path, ArchName),
358                                                  std::move(ObjOrErr.get())));
359     return Res;
360   }
361   if (Bin->isObject()) {
362     return cast<ObjectFile>(Bin);
363   }
364   return errorCodeToError(object_error::arch_not_found);
365 }
366 
367 Expected<SymbolizableModule *>
368 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName,
369                                       StringRef DWPName) {
370   const auto &I = Modules.find(ModuleName);
371   if (I != Modules.end()) {
372     return I->second.get();
373   }
374   std::string BinaryName = ModuleName;
375   std::string ArchName = Opts.DefaultArch;
376   size_t ColonPos = ModuleName.find_last_of(':');
377   // Verify that substring after colon form a valid arch name.
378   if (ColonPos != std::string::npos) {
379     std::string ArchStr = ModuleName.substr(ColonPos + 1);
380     if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
381       BinaryName = ModuleName.substr(0, ColonPos);
382       ArchName = ArchStr;
383     }
384   }
385   auto ObjectsOrErr = getOrCreateObjectPair(BinaryName, ArchName);
386   if (!ObjectsOrErr) {
387     // Failed to find valid object file.
388     Modules.insert(
389         std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
390     return ObjectsOrErr.takeError();
391   }
392   ObjectPair Objects = ObjectsOrErr.get();
393 
394   std::unique_ptr<DIContext> Context;
395   // If this is a COFF object containing PDB info, use a PDBContext to
396   // symbolize. Otherwise, use DWARF.
397   if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
398     const codeview::DebugInfo *DebugInfo;
399     StringRef PDBFileName;
400     auto EC = CoffObject->getDebugPDBInfo(DebugInfo, PDBFileName);
401     if (!EC && DebugInfo != nullptr && !PDBFileName.empty()) {
402       using namespace pdb;
403       std::unique_ptr<IPDBSession> Session;
404       if (auto Err = loadDataForEXE(PDB_ReaderType::DIA,
405                                     Objects.first->getFileName(), Session)) {
406         Modules.insert(
407             std::make_pair(ModuleName, std::unique_ptr<SymbolizableModule>()));
408         return std::move(Err);
409       }
410       Context.reset(new PDBContext(*CoffObject, std::move(Session)));
411     }
412   }
413   if (!Context)
414     Context = DWARFContext::create(*Objects.second, nullptr,
415                                    DWARFContext::defaultErrorHandler, DWPName);
416   assert(Context);
417   auto InfoOrErr =
418       SymbolizableObjectFile::create(Objects.first, std::move(Context));
419   std::unique_ptr<SymbolizableModule> SymMod;
420   if (InfoOrErr)
421     SymMod = std::move(InfoOrErr.get());
422   auto InsertResult =
423       Modules.insert(std::make_pair(ModuleName, std::move(SymMod)));
424   assert(InsertResult.second);
425   if (auto EC = InfoOrErr.getError())
426     return errorCodeToError(EC);
427   return InsertResult.first->second.get();
428 }
429 
430 namespace {
431 
432 // Undo these various manglings for Win32 extern "C" functions:
433 // cdecl       - _foo
434 // stdcall     - _foo@12
435 // fastcall    - @foo@12
436 // vectorcall  - foo@@12
437 // These are all different linkage names for 'foo'.
438 StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
439   // Remove any '_' or '@' prefix.
440   char Front = SymbolName.empty() ? '\0' : SymbolName[0];
441   if (Front == '_' || Front == '@')
442     SymbolName = SymbolName.drop_front();
443 
444   // Remove any '@[0-9]+' suffix.
445   if (Front != '?') {
446     size_t AtPos = SymbolName.rfind('@');
447     if (AtPos != StringRef::npos &&
448         std::all_of(SymbolName.begin() + AtPos + 1, SymbolName.end(),
449                     [](char C) { return C >= '0' && C <= '9'; })) {
450       SymbolName = SymbolName.substr(0, AtPos);
451     }
452   }
453 
454   // Remove any ending '@' for vectorcall.
455   if (SymbolName.endswith("@"))
456     SymbolName = SymbolName.drop_back();
457 
458   return SymbolName;
459 }
460 
461 } // end anonymous namespace
462 
463 #if !defined(_MSC_VER)
464 // Assume that __cxa_demangle is provided by libcxxabi (except for Windows).
465 extern "C" char *__cxa_demangle(const char *mangled_name, char *output_buffer,
466                                 size_t *length, int *status);
467 #endif
468 
469 std::string
470 LLVMSymbolizer::DemangleName(const std::string &Name,
471                              const SymbolizableModule *DbiModuleDescriptor) {
472 #if !defined(_MSC_VER)
473   // We can spoil names of symbols with C linkage, so use an heuristic
474   // approach to check if the name should be demangled.
475   if (Name.substr(0, 2) == "_Z") {
476     int status = 0;
477     char *DemangledName = __cxa_demangle(Name.c_str(), nullptr, nullptr, &status);
478     if (status != 0)
479       return Name;
480     std::string Result = DemangledName;
481     free(DemangledName);
482     return Result;
483   }
484 #else
485   if (!Name.empty() && Name.front() == '?') {
486     // Only do MSVC C++ demangling on symbols starting with '?'.
487     char DemangledName[1024] = {0};
488     DWORD result = ::UnDecorateSymbolName(
489         Name.c_str(), DemangledName, 1023,
490         UNDNAME_NO_ACCESS_SPECIFIERS |       // Strip public, private, protected
491             UNDNAME_NO_ALLOCATION_LANGUAGE | // Strip __thiscall, __stdcall, etc
492             UNDNAME_NO_THROW_SIGNATURES |    // Strip throw() specifications
493             UNDNAME_NO_MEMBER_TYPE | // Strip virtual, static, etc specifiers
494             UNDNAME_NO_MS_KEYWORDS | // Strip all MS extension keywords
495             UNDNAME_NO_FUNCTION_RETURNS); // Strip function return types
496     return (result == 0) ? Name : std::string(DemangledName);
497   }
498 #endif
499   if (DbiModuleDescriptor && DbiModuleDescriptor->isWin32Module())
500     return std::string(demanglePE32ExternCFunc(Name));
501   return Name;
502 }
503 
504 } // namespace symbolize
505 } // namespace llvm
506