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/BTF/BTFContext.h"
17#include "llvm/DebugInfo/DWARF/DWARFContext.h"
18#include "llvm/DebugInfo/PDB/PDB.h"
19#include "llvm/DebugInfo/PDB/PDBContext.h"
20#include "llvm/DebugInfo/Symbolize/SymbolizableObjectFile.h"
21#include "llvm/Demangle/Demangle.h"
22#include "llvm/Object/BuildID.h"
23#include "llvm/Object/COFF.h"
24#include "llvm/Object/ELFObjectFile.h"
25#include "llvm/Object/MachO.h"
26#include "llvm/Object/MachOUniversal.h"
27#include "llvm/Support/CRC.h"
28#include "llvm/Support/Casting.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
38namespace llvm {
39namespace codeview {
40union DebugInfo;
41}
42namespace symbolize {
43
44LLVMSymbolizer::LLVMSymbolizer() = default;
45
46LLVMSymbolizer::LLVMSymbolizer(const Options &Opts)
47 : Opts(Opts),
48 BIDFetcher(std::make_unique<BuildIDFetcher>(args: Opts.DebugFileDirectory)) {}
49
50LLVMSymbolizer::~LLVMSymbolizer() = default;
51
52template <typename T>
53Expected<DILineInfo>
54LLVMSymbolizer::symbolizeCodeCommon(const T &ModuleSpecifier,
55 object::SectionedAddress ModuleOffset) {
56
57 auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
58 if (!InfoOrErr)
59 return InfoOrErr.takeError();
60
61 SymbolizableModule *Info = *InfoOrErr;
62
63 // A null module means an error has already been reported. Return an empty
64 // result.
65 if (!Info)
66 return DILineInfo();
67
68 // If the user is giving us relative addresses, add the preferred base of the
69 // object to the offset before we do the query. It's what DIContext expects.
70 if (Opts.RelativeAddresses)
71 ModuleOffset.Address += Info->getModulePreferredBase();
72
73 DILineInfo LineInfo = Info->symbolizeCode(
74 ModuleOffset, LineInfoSpecifier: DILineInfoSpecifier(Opts.PathStyle, Opts.PrintFunctions),
75 UseSymbolTable: Opts.UseSymbolTable);
76 if (Opts.Demangle)
77 LineInfo.FunctionName = DemangleName(Name: LineInfo.FunctionName, DbiModuleDescriptor: Info);
78 return LineInfo;
79}
80
81Expected<DILineInfo>
82LLVMSymbolizer::symbolizeCode(const ObjectFile &Obj,
83 object::SectionedAddress ModuleOffset) {
84 return symbolizeCodeCommon(ModuleSpecifier: Obj, ModuleOffset);
85}
86
87Expected<DILineInfo>
88LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
89 object::SectionedAddress ModuleOffset) {
90 return symbolizeCodeCommon(ModuleSpecifier: ModuleName, ModuleOffset);
91}
92
93Expected<DILineInfo>
94LLVMSymbolizer::symbolizeCode(ArrayRef<uint8_t> BuildID,
95 object::SectionedAddress ModuleOffset) {
96 return symbolizeCodeCommon(ModuleSpecifier: BuildID, ModuleOffset);
97}
98
99template <typename T>
100Expected<DIInliningInfo> LLVMSymbolizer::symbolizeInlinedCodeCommon(
101 const T &ModuleSpecifier, object::SectionedAddress ModuleOffset) {
102 auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
103 if (!InfoOrErr)
104 return InfoOrErr.takeError();
105
106 SymbolizableModule *Info = *InfoOrErr;
107
108 // A null module means an error has already been reported. Return an empty
109 // result.
110 if (!Info)
111 return DIInliningInfo();
112
113 // If the user is giving us relative addresses, add the preferred base of the
114 // object to the offset before we do the query. It's what DIContext expects.
115 if (Opts.RelativeAddresses)
116 ModuleOffset.Address += Info->getModulePreferredBase();
117
118 DIInliningInfo InlinedContext = Info->symbolizeInlinedCode(
119 ModuleOffset, LineInfoSpecifier: DILineInfoSpecifier(Opts.PathStyle, Opts.PrintFunctions),
120 UseSymbolTable: Opts.UseSymbolTable);
121 if (Opts.Demangle) {
122 for (int i = 0, n = InlinedContext.getNumberOfFrames(); i < n; i++) {
123 auto *Frame = InlinedContext.getMutableFrame(Index: i);
124 Frame->FunctionName = DemangleName(Name: Frame->FunctionName, DbiModuleDescriptor: Info);
125 }
126 }
127 return InlinedContext;
128}
129
130Expected<DIInliningInfo>
131LLVMSymbolizer::symbolizeInlinedCode(const ObjectFile &Obj,
132 object::SectionedAddress ModuleOffset) {
133 return symbolizeInlinedCodeCommon(ModuleSpecifier: Obj, ModuleOffset);
134}
135
136Expected<DIInliningInfo>
137LLVMSymbolizer::symbolizeInlinedCode(const std::string &ModuleName,
138 object::SectionedAddress ModuleOffset) {
139 return symbolizeInlinedCodeCommon(ModuleSpecifier: ModuleName, ModuleOffset);
140}
141
142Expected<DIInliningInfo>
143LLVMSymbolizer::symbolizeInlinedCode(ArrayRef<uint8_t> BuildID,
144 object::SectionedAddress ModuleOffset) {
145 return symbolizeInlinedCodeCommon(ModuleSpecifier: BuildID, ModuleOffset);
146}
147
148template <typename T>
149Expected<DIGlobal>
150LLVMSymbolizer::symbolizeDataCommon(const T &ModuleSpecifier,
151 object::SectionedAddress ModuleOffset) {
152
153 auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
154 if (!InfoOrErr)
155 return InfoOrErr.takeError();
156
157 SymbolizableModule *Info = *InfoOrErr;
158 // A null module means an error has already been reported. Return an empty
159 // result.
160 if (!Info)
161 return DIGlobal();
162
163 // If the user is giving us relative addresses, add the preferred base of
164 // the object to the offset before we do the query. It's what DIContext
165 // expects.
166 if (Opts.RelativeAddresses)
167 ModuleOffset.Address += Info->getModulePreferredBase();
168
169 DIGlobal Global = Info->symbolizeData(ModuleOffset);
170 if (Opts.Demangle)
171 Global.Name = DemangleName(Name: Global.Name, DbiModuleDescriptor: Info);
172 return Global;
173}
174
175Expected<DIGlobal>
176LLVMSymbolizer::symbolizeData(const ObjectFile &Obj,
177 object::SectionedAddress ModuleOffset) {
178 return symbolizeDataCommon(ModuleSpecifier: Obj, ModuleOffset);
179}
180
181Expected<DIGlobal>
182LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
183 object::SectionedAddress ModuleOffset) {
184 return symbolizeDataCommon(ModuleSpecifier: ModuleName, ModuleOffset);
185}
186
187Expected<DIGlobal>
188LLVMSymbolizer::symbolizeData(ArrayRef<uint8_t> BuildID,
189 object::SectionedAddress ModuleOffset) {
190 return symbolizeDataCommon(ModuleSpecifier: BuildID, ModuleOffset);
191}
192
193template <typename T>
194Expected<std::vector<DILocal>>
195LLVMSymbolizer::symbolizeFrameCommon(const T &ModuleSpecifier,
196 object::SectionedAddress ModuleOffset) {
197 auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
198 if (!InfoOrErr)
199 return InfoOrErr.takeError();
200
201 SymbolizableModule *Info = *InfoOrErr;
202 // A null module means an error has already been reported. Return an empty
203 // result.
204 if (!Info)
205 return std::vector<DILocal>();
206
207 // If the user is giving us relative addresses, add the preferred base of
208 // the object to the offset before we do the query. It's what DIContext
209 // expects.
210 if (Opts.RelativeAddresses)
211 ModuleOffset.Address += Info->getModulePreferredBase();
212
213 return Info->symbolizeFrame(ModuleOffset);
214}
215
216Expected<std::vector<DILocal>>
217LLVMSymbolizer::symbolizeFrame(const ObjectFile &Obj,
218 object::SectionedAddress ModuleOffset) {
219 return symbolizeFrameCommon(ModuleSpecifier: Obj, ModuleOffset);
220}
221
222Expected<std::vector<DILocal>>
223LLVMSymbolizer::symbolizeFrame(const std::string &ModuleName,
224 object::SectionedAddress ModuleOffset) {
225 return symbolizeFrameCommon(ModuleSpecifier: ModuleName, ModuleOffset);
226}
227
228Expected<std::vector<DILocal>>
229LLVMSymbolizer::symbolizeFrame(ArrayRef<uint8_t> BuildID,
230 object::SectionedAddress ModuleOffset) {
231 return symbolizeFrameCommon(ModuleSpecifier: BuildID, ModuleOffset);
232}
233
234template <typename T>
235Expected<std::vector<DILineInfo>>
236LLVMSymbolizer::findSymbolCommon(const T &ModuleSpecifier, StringRef Symbol,
237 uint64_t Offset) {
238 auto InfoOrErr = getOrCreateModuleInfo(ModuleSpecifier);
239 if (!InfoOrErr)
240 return InfoOrErr.takeError();
241
242 SymbolizableModule *Info = *InfoOrErr;
243 std::vector<DILineInfo> Result;
244
245 // A null module means an error has already been reported. Return an empty
246 // result.
247 if (!Info)
248 return Result;
249
250 for (object::SectionedAddress A : Info->findSymbol(Symbol, Offset)) {
251 DILineInfo LineInfo = Info->symbolizeCode(
252 ModuleOffset: A, LineInfoSpecifier: DILineInfoSpecifier(Opts.PathStyle, Opts.PrintFunctions),
253 UseSymbolTable: Opts.UseSymbolTable);
254 if (LineInfo.FileName != DILineInfo::BadString) {
255 if (Opts.Demangle)
256 LineInfo.FunctionName = DemangleName(Name: LineInfo.FunctionName, DbiModuleDescriptor: Info);
257 Result.push_back(x: LineInfo);
258 }
259 }
260
261 return Result;
262}
263
264Expected<std::vector<DILineInfo>>
265LLVMSymbolizer::findSymbol(const ObjectFile &Obj, StringRef Symbol,
266 uint64_t Offset) {
267 return findSymbolCommon(ModuleSpecifier: Obj, Symbol, Offset);
268}
269
270Expected<std::vector<DILineInfo>>
271LLVMSymbolizer::findSymbol(const std::string &ModuleName, StringRef Symbol,
272 uint64_t Offset) {
273 return findSymbolCommon(ModuleSpecifier: ModuleName, Symbol, Offset);
274}
275
276Expected<std::vector<DILineInfo>>
277LLVMSymbolizer::findSymbol(ArrayRef<uint8_t> BuildID, StringRef Symbol,
278 uint64_t Offset) {
279 return findSymbolCommon(ModuleSpecifier: BuildID, Symbol, Offset);
280}
281
282void LLVMSymbolizer::flush() {
283 ObjectForUBPathAndArch.clear();
284 LRUBinaries.clear();
285 CacheSize = 0;
286 BinaryForPath.clear();
287 ObjectPairForPathArch.clear();
288 Modules.clear();
289 BuildIDPaths.clear();
290}
291
292namespace {
293
294// For Path="/path/to/foo" and Basename="foo" assume that debug info is in
295// /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
296// For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
297// /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
298std::string getDarwinDWARFResourceForPath(const std::string &Path,
299 const std::string &Basename) {
300 SmallString<16> ResourceName = StringRef(Path);
301 if (sys::path::extension(path: Path) != ".dSYM") {
302 ResourceName += ".dSYM";
303 }
304 sys::path::append(path&: ResourceName, a: "Contents", b: "Resources", c: "DWARF");
305 sys::path::append(path&: ResourceName, a: Basename);
306 return std::string(ResourceName);
307}
308
309bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
310 ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
311 MemoryBuffer::getFileOrSTDIN(Filename: Path);
312 if (!MB)
313 return false;
314 return CRCHash == llvm::crc32(Data: arrayRefFromStringRef(Input: MB.get()->getBuffer()));
315}
316
317bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
318 uint32_t &CRCHash) {
319 if (!Obj)
320 return false;
321 for (const SectionRef &Section : Obj->sections()) {
322 StringRef Name;
323 consumeError(Err: Section.getName().moveInto(Value&: Name));
324
325 Name = Name.substr(Start: Name.find_first_not_of(Chars: "._"));
326 if (Name == "gnu_debuglink") {
327 Expected<StringRef> ContentsOrErr = Section.getContents();
328 if (!ContentsOrErr) {
329 consumeError(Err: ContentsOrErr.takeError());
330 return false;
331 }
332 DataExtractor DE(*ContentsOrErr, Obj->isLittleEndian(), 0);
333 uint64_t Offset = 0;
334 if (const char *DebugNameStr = DE.getCStr(OffsetPtr: &Offset)) {
335 // 4-byte align the offset.
336 Offset = (Offset + 3) & ~0x3;
337 if (DE.isValidOffsetForDataOfSize(offset: Offset, length: 4)) {
338 DebugName = DebugNameStr;
339 CRCHash = DE.getU32(offset_ptr: &Offset);
340 return true;
341 }
342 }
343 break;
344 }
345 }
346 return false;
347}
348
349bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
350 const MachOObjectFile *Obj) {
351 ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
352 ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
353 if (dbg_uuid.empty() || bin_uuid.empty())
354 return false;
355 return !memcmp(s1: dbg_uuid.data(), s2: bin_uuid.data(), n: dbg_uuid.size());
356}
357
358} // end anonymous namespace
359
360ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
361 const MachOObjectFile *MachExeObj,
362 const std::string &ArchName) {
363 // On Darwin we may find DWARF in separate object file in
364 // resource directory.
365 std::vector<std::string> DsymPaths;
366 StringRef Filename = sys::path::filename(path: ExePath);
367 DsymPaths.push_back(
368 x: getDarwinDWARFResourceForPath(Path: ExePath, Basename: std::string(Filename)));
369 for (const auto &Path : Opts.DsymHints) {
370 DsymPaths.push_back(
371 x: getDarwinDWARFResourceForPath(Path, Basename: std::string(Filename)));
372 }
373 for (const auto &Path : DsymPaths) {
374 auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
375 if (!DbgObjOrErr) {
376 // Ignore errors, the file might not exist.
377 consumeError(Err: DbgObjOrErr.takeError());
378 continue;
379 }
380 ObjectFile *DbgObj = DbgObjOrErr.get();
381 if (!DbgObj)
382 continue;
383 const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(Val: DbgObj);
384 if (!MachDbgObj)
385 continue;
386 if (darwinDsymMatchesBinary(DbgObj: MachDbgObj, Obj: MachExeObj))
387 return DbgObj;
388 }
389 return nullptr;
390}
391
392ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
393 const ObjectFile *Obj,
394 const std::string &ArchName) {
395 std::string DebuglinkName;
396 uint32_t CRCHash;
397 std::string DebugBinaryPath;
398 if (!getGNUDebuglinkContents(Obj, DebugName&: DebuglinkName, CRCHash))
399 return nullptr;
400 if (!findDebugBinary(OrigPath: Path, DebuglinkName, CRCHash, Result&: DebugBinaryPath))
401 return nullptr;
402 auto DbgObjOrErr = getOrCreateObject(Path: DebugBinaryPath, ArchName);
403 if (!DbgObjOrErr) {
404 // Ignore errors, the file might not exist.
405 consumeError(Err: DbgObjOrErr.takeError());
406 return nullptr;
407 }
408 return DbgObjOrErr.get();
409}
410
411ObjectFile *LLVMSymbolizer::lookUpBuildIDObject(const std::string &Path,
412 const ELFObjectFileBase *Obj,
413 const std::string &ArchName) {
414 auto BuildID = getBuildID(Obj);
415 if (BuildID.size() < 2)
416 return nullptr;
417 std::string DebugBinaryPath;
418 if (!getOrFindDebugBinary(BuildID, Result&: DebugBinaryPath))
419 return nullptr;
420 auto DbgObjOrErr = getOrCreateObject(Path: DebugBinaryPath, ArchName);
421 if (!DbgObjOrErr) {
422 consumeError(Err: DbgObjOrErr.takeError());
423 return nullptr;
424 }
425 return DbgObjOrErr.get();
426}
427
428bool LLVMSymbolizer::findDebugBinary(const std::string &OrigPath,
429 const std::string &DebuglinkName,
430 uint32_t CRCHash, std::string &Result) {
431 SmallString<16> OrigDir(OrigPath);
432 llvm::sys::path::remove_filename(path&: OrigDir);
433 SmallString<16> DebugPath = OrigDir;
434 // Try relative/path/to/original_binary/debuglink_name
435 llvm::sys::path::append(path&: DebugPath, a: DebuglinkName);
436 if (checkFileCRC(Path: DebugPath, CRCHash)) {
437 Result = std::string(DebugPath);
438 return true;
439 }
440 // Try relative/path/to/original_binary/.debug/debuglink_name
441 DebugPath = OrigDir;
442 llvm::sys::path::append(path&: DebugPath, a: ".debug", b: DebuglinkName);
443 if (checkFileCRC(Path: DebugPath, CRCHash)) {
444 Result = std::string(DebugPath);
445 return true;
446 }
447 // Make the path absolute so that lookups will go to
448 // "/usr/lib/debug/full/path/to/debug", not
449 // "/usr/lib/debug/to/debug"
450 llvm::sys::fs::make_absolute(path&: OrigDir);
451 if (!Opts.FallbackDebugPath.empty()) {
452 // Try <FallbackDebugPath>/absolute/path/to/original_binary/debuglink_name
453 DebugPath = Opts.FallbackDebugPath;
454 } else {
455#if defined(__NetBSD__)
456 // Try /usr/libdata/debug/absolute/path/to/original_binary/debuglink_name
457 DebugPath = "/usr/libdata/debug";
458#else
459 // Try /usr/lib/debug/absolute/path/to/original_binary/debuglink_name
460 DebugPath = "/usr/lib/debug";
461#endif
462 }
463 llvm::sys::path::append(path&: DebugPath, a: llvm::sys::path::relative_path(path: OrigDir),
464 b: DebuglinkName);
465 if (checkFileCRC(Path: DebugPath, CRCHash)) {
466 Result = std::string(DebugPath);
467 return true;
468 }
469 return false;
470}
471
472static StringRef getBuildIDStr(ArrayRef<uint8_t> BuildID) {
473 return StringRef(reinterpret_cast<const char *>(BuildID.data()),
474 BuildID.size());
475}
476
477bool LLVMSymbolizer::getOrFindDebugBinary(const ArrayRef<uint8_t> BuildID,
478 std::string &Result) {
479 StringRef BuildIDStr = getBuildIDStr(BuildID);
480 auto I = BuildIDPaths.find(Key: BuildIDStr);
481 if (I != BuildIDPaths.end()) {
482 Result = I->second;
483 return true;
484 }
485 if (!BIDFetcher)
486 return false;
487 if (std::optional<std::string> Path = BIDFetcher->fetch(BuildID)) {
488 Result = *Path;
489 auto InsertResult = BuildIDPaths.insert(KV: {BuildIDStr, Result});
490 assert(InsertResult.second);
491 (void)InsertResult;
492 return true;
493 }
494
495 return false;
496}
497
498Expected<LLVMSymbolizer::ObjectPair>
499LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
500 const std::string &ArchName) {
501 auto I = ObjectPairForPathArch.find(x: std::make_pair(x: Path, y: ArchName));
502 if (I != ObjectPairForPathArch.end()) {
503 recordAccess(Bin&: BinaryForPath.find(x: Path)->second);
504 return I->second;
505 }
506
507 auto ObjOrErr = getOrCreateObject(Path, ArchName);
508 if (!ObjOrErr) {
509 ObjectPairForPathArch.emplace(args: std::make_pair(x: Path, y: ArchName),
510 args: ObjectPair(nullptr, nullptr));
511 return ObjOrErr.takeError();
512 }
513
514 ObjectFile *Obj = ObjOrErr.get();
515 assert(Obj != nullptr);
516 ObjectFile *DbgObj = nullptr;
517
518 if (auto MachObj = dyn_cast<const MachOObjectFile>(Val: Obj))
519 DbgObj = lookUpDsymFile(ExePath: Path, MachExeObj: MachObj, ArchName);
520 else if (auto ELFObj = dyn_cast<const ELFObjectFileBase>(Val: Obj))
521 DbgObj = lookUpBuildIDObject(Path, Obj: ELFObj, ArchName);
522 if (!DbgObj)
523 DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
524 if (!DbgObj)
525 DbgObj = Obj;
526 ObjectPair Res = std::make_pair(x&: Obj, y&: DbgObj);
527 std::string DbgObjPath = DbgObj->getFileName().str();
528 auto Pair =
529 ObjectPairForPathArch.emplace(args: std::make_pair(x: Path, y: ArchName), args&: Res);
530 BinaryForPath.find(x: DbgObjPath)->second.pushEvictor(Evictor: [this, I = Pair.first]() {
531 ObjectPairForPathArch.erase(position: I);
532 });
533 return Res;
534}
535
536Expected<ObjectFile *>
537LLVMSymbolizer::getOrCreateObject(const std::string &Path,
538 const std::string &ArchName) {
539 Binary *Bin;
540 auto Pair = BinaryForPath.emplace(args: Path, args: OwningBinary<Binary>());
541 if (!Pair.second) {
542 Bin = Pair.first->second->getBinary();
543 recordAccess(Bin&: Pair.first->second);
544 } else {
545 Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
546 if (!BinOrErr)
547 return BinOrErr.takeError();
548
549 CachedBinary &CachedBin = Pair.first->second;
550 CachedBin = std::move(BinOrErr.get());
551 CachedBin.pushEvictor(Evictor: [this, I = Pair.first]() { BinaryForPath.erase(position: I); });
552 LRUBinaries.push_back(Node&: CachedBin);
553 CacheSize += CachedBin.size();
554 Bin = CachedBin->getBinary();
555 }
556
557 if (!Bin)
558 return static_cast<ObjectFile *>(nullptr);
559
560 if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Val: Bin)) {
561 auto I = ObjectForUBPathAndArch.find(x: std::make_pair(x: Path, y: ArchName));
562 if (I != ObjectForUBPathAndArch.end())
563 return I->second.get();
564
565 Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
566 UB->getMachOObjectForArch(ArchName);
567 if (!ObjOrErr) {
568 ObjectForUBPathAndArch.emplace(args: std::make_pair(x: Path, y: ArchName),
569 args: std::unique_ptr<ObjectFile>());
570 return ObjOrErr.takeError();
571 }
572 ObjectFile *Res = ObjOrErr->get();
573 auto Pair = ObjectForUBPathAndArch.emplace(args: std::make_pair(x: Path, y: ArchName),
574 args: std::move(ObjOrErr.get()));
575 BinaryForPath.find(x: Path)->second.pushEvictor(
576 Evictor: [this, Iter = Pair.first]() { ObjectForUBPathAndArch.erase(position: Iter); });
577 return Res;
578 }
579 if (Bin->isObject()) {
580 return cast<ObjectFile>(Val: Bin);
581 }
582 return errorCodeToError(EC: object_error::arch_not_found);
583}
584
585Expected<SymbolizableModule *>
586LLVMSymbolizer::createModuleInfo(const ObjectFile *Obj,
587 std::unique_ptr<DIContext> Context,
588 StringRef ModuleName) {
589 auto InfoOrErr = SymbolizableObjectFile::create(Obj, DICtx: std::move(Context),
590 UntagAddresses: Opts.UntagAddresses);
591 std::unique_ptr<SymbolizableModule> SymMod;
592 if (InfoOrErr)
593 SymMod = std::move(*InfoOrErr);
594 auto InsertResult = Modules.insert(
595 x: std::make_pair(x: std::string(ModuleName), y: std::move(SymMod)));
596 assert(InsertResult.second);
597 if (!InfoOrErr)
598 return InfoOrErr.takeError();
599 return InsertResult.first->second.get();
600}
601
602Expected<SymbolizableModule *>
603LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) {
604 std::string BinaryName = ModuleName;
605 std::string ArchName = Opts.DefaultArch;
606 size_t ColonPos = ModuleName.find_last_of(c: ':');
607 // Verify that substring after colon form a valid arch name.
608 if (ColonPos != std::string::npos) {
609 std::string ArchStr = ModuleName.substr(pos: ColonPos + 1);
610 if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
611 BinaryName = ModuleName.substr(pos: 0, n: ColonPos);
612 ArchName = ArchStr;
613 }
614 }
615
616 auto I = Modules.find(x: ModuleName);
617 if (I != Modules.end()) {
618 recordAccess(Bin&: BinaryForPath.find(x: BinaryName)->second);
619 return I->second.get();
620 }
621
622 auto ObjectsOrErr = getOrCreateObjectPair(Path: BinaryName, ArchName);
623 if (!ObjectsOrErr) {
624 // Failed to find valid object file.
625 Modules.emplace(args: ModuleName, args: std::unique_ptr<SymbolizableModule>());
626 return ObjectsOrErr.takeError();
627 }
628 ObjectPair Objects = ObjectsOrErr.get();
629
630 std::unique_ptr<DIContext> Context;
631 // If this is a COFF object containing PDB info and not containing DWARF
632 // section, use a PDBContext to symbolize. Otherwise, use DWARF.
633 if (auto CoffObject = dyn_cast<COFFObjectFile>(Val: Objects.first)) {
634 const codeview::DebugInfo *DebugInfo;
635 StringRef PDBFileName;
636 auto EC = CoffObject->getDebugPDBInfo(Info&: DebugInfo, PDBFileName);
637 // Use DWARF if there're DWARF sections.
638 bool HasDwarf =
639 llvm::any_of(Range: Objects.first->sections(), P: [](SectionRef Section) -> bool {
640 if (Expected<StringRef> SectionName = Section.getName())
641 return SectionName.get() == ".debug_info";
642 return false;
643 });
644 if (!EC && !HasDwarf && DebugInfo != nullptr && !PDBFileName.empty()) {
645 using namespace pdb;
646 std::unique_ptr<IPDBSession> Session;
647
648 PDB_ReaderType ReaderType =
649 Opts.UseDIA ? PDB_ReaderType::DIA : PDB_ReaderType::Native;
650 if (auto Err = loadDataForEXE(Type: ReaderType, Path: Objects.first->getFileName(),
651 Session)) {
652 Modules.emplace(args: ModuleName, args: std::unique_ptr<SymbolizableModule>());
653 // Return along the PDB filename to provide more context
654 return createFileError(F: PDBFileName, E: std::move(Err));
655 }
656 Context.reset(p: new PDBContext(*CoffObject, std::move(Session)));
657 }
658 }
659 if (!Context)
660 Context = DWARFContext::create(
661 Obj: *Objects.second, RelocAction: DWARFContext::ProcessDebugRelocations::Process,
662 L: nullptr, DWPName: Opts.DWPName);
663 auto ModuleOrErr =
664 createModuleInfo(Obj: Objects.first, Context: std::move(Context), ModuleName);
665 if (ModuleOrErr) {
666 auto I = Modules.find(x: ModuleName);
667 BinaryForPath.find(x: BinaryName)->second.pushEvictor(Evictor: [this, I]() {
668 Modules.erase(position: I);
669 });
670 }
671 return ModuleOrErr;
672}
673
674// For BPF programs .BTF.ext section contains line numbers information,
675// use it if regular DWARF is not available (e.g. for stripped binary).
676static bool useBTFContext(const ObjectFile &Obj) {
677 return Obj.makeTriple().isBPF() && !Obj.hasDebugInfo() &&
678 BTFParser::hasBTFSections(Obj);
679}
680
681Expected<SymbolizableModule *>
682LLVMSymbolizer::getOrCreateModuleInfo(const ObjectFile &Obj) {
683 StringRef ObjName = Obj.getFileName();
684 auto I = Modules.find(x: ObjName);
685 if (I != Modules.end())
686 return I->second.get();
687
688 std::unique_ptr<DIContext> Context;
689 if (useBTFContext(Obj))
690 Context = BTFContext::create(Obj);
691 else
692 Context = DWARFContext::create(Obj);
693 // FIXME: handle COFF object with PDB info to use PDBContext
694 return createModuleInfo(Obj: &Obj, Context: std::move(Context), ModuleName: ObjName);
695}
696
697Expected<SymbolizableModule *>
698LLVMSymbolizer::getOrCreateModuleInfo(ArrayRef<uint8_t> BuildID) {
699 std::string Path;
700 if (!getOrFindDebugBinary(BuildID, Result&: Path)) {
701 return createStringError(EC: errc::no_such_file_or_directory,
702 S: "could not find build ID");
703 }
704 return getOrCreateModuleInfo(ModuleName: Path);
705}
706
707namespace {
708
709// Undo these various manglings for Win32 extern "C" functions:
710// cdecl - _foo
711// stdcall - _foo@12
712// fastcall - @foo@12
713// vectorcall - foo@@12
714// These are all different linkage names for 'foo'.
715StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
716 char Front = SymbolName.empty() ? '\0' : SymbolName[0];
717
718 // Remove any '@[0-9]+' suffix.
719 bool HasAtNumSuffix = false;
720 if (Front != '?') {
721 size_t AtPos = SymbolName.rfind(C: '@');
722 if (AtPos != StringRef::npos &&
723 all_of(Range: drop_begin(RangeOrContainer&: SymbolName, N: AtPos + 1), P: isDigit)) {
724 SymbolName = SymbolName.substr(Start: 0, N: AtPos);
725 HasAtNumSuffix = true;
726 }
727 }
728
729 // Remove any ending '@' for vectorcall.
730 bool IsVectorCall = false;
731 if (HasAtNumSuffix && SymbolName.ends_with(Suffix: "@")) {
732 SymbolName = SymbolName.drop_back();
733 IsVectorCall = true;
734 }
735
736 // If not vectorcall, remove any '_' or '@' prefix.
737 if (!IsVectorCall && (Front == '_' || Front == '@'))
738 SymbolName = SymbolName.drop_front();
739
740 return SymbolName;
741}
742
743} // end anonymous namespace
744
745std::string
746LLVMSymbolizer::DemangleName(const std::string &Name,
747 const SymbolizableModule *DbiModuleDescriptor) {
748 std::string Result;
749 if (nonMicrosoftDemangle(MangledName: Name, Result))
750 return Result;
751
752 if (!Name.empty() && Name.front() == '?') {
753 // Only do MSVC C++ demangling on symbols starting with '?'.
754 int status = 0;
755 char *DemangledName = microsoftDemangle(
756 mangled_name: Name, n_read: nullptr, status: &status,
757 Flags: MSDemangleFlags(MSDF_NoAccessSpecifier | MSDF_NoCallingConvention |
758 MSDF_NoMemberType | MSDF_NoReturnType));
759 if (status != 0)
760 return Name;
761 Result = DemangledName;
762 free(ptr: DemangledName);
763 return Result;
764 }
765
766 if (DbiModuleDescriptor && DbiModuleDescriptor->isWin32Module()) {
767 std::string DemangledCName(demanglePE32ExternCFunc(SymbolName: Name));
768 // On i386 Windows, the C name mangling for different calling conventions
769 // may also be applied on top of the Itanium or Rust name mangling.
770 if (nonMicrosoftDemangle(MangledName: DemangledCName, Result))
771 return Result;
772 return DemangledCName;
773 }
774 return Name;
775}
776
777void LLVMSymbolizer::recordAccess(CachedBinary &Bin) {
778 if (Bin->getBinary())
779 LRUBinaries.splice(I: LRUBinaries.end(), L2&: LRUBinaries, Node: Bin.getIterator());
780}
781
782void LLVMSymbolizer::pruneCache() {
783 // Evict the LRU binary until the max cache size is reached or there's <= 1
784 // item in the cache. The MRU binary is always kept to avoid thrashing if it's
785 // larger than the cache size.
786 while (CacheSize > Opts.MaxCacheSize && !LRUBinaries.empty() &&
787 std::next(x: LRUBinaries.begin()) != LRUBinaries.end()) {
788 CachedBinary &Bin = LRUBinaries.front();
789 CacheSize -= Bin.size();
790 LRUBinaries.pop_front();
791 Bin.evict();
792 }
793}
794
795void CachedBinary::pushEvictor(std::function<void()> NewEvictor) {
796 if (Evictor) {
797 this->Evictor = [OldEvictor = std::move(this->Evictor),
798 NewEvictor = std::move(NewEvictor)]() {
799 NewEvictor();
800 OldEvictor();
801 };
802 } else {
803 this->Evictor = std::move(NewEvictor);
804 }
805}
806
807} // namespace symbolize
808} // namespace llvm
809