1 //===- DWARFUnit.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 #include "llvm/DebugInfo/DWARF/DWARFUnit.h"
10 #include "llvm/ADT/SmallString.h"
11 #include "llvm/ADT/StringRef.h"
12 #include "llvm/BinaryFormat/Dwarf.h"
13 #include "llvm/DebugInfo/DWARF/DWARFAbbreviationDeclaration.h"
14 #include "llvm/DebugInfo/DWARF/DWARFCompileUnit.h"
15 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
16 #include "llvm/DebugInfo/DWARF/DWARFDebugAbbrev.h"
17 #include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
18 #include "llvm/DebugInfo/DWARF/DWARFDebugLoc.h"
19 #include "llvm/DebugInfo/DWARF/DWARFDebugRangeList.h"
20 #include "llvm/DebugInfo/DWARF/DWARFDebugRnglists.h"
21 #include "llvm/DebugInfo/DWARF/DWARFDie.h"
22 #include "llvm/DebugInfo/DWARF/DWARFExpression.h"
23 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
24 #include "llvm/DebugInfo/DWARF/DWARFListTable.h"
25 #include "llvm/DebugInfo/DWARF/DWARFObject.h"
26 #include "llvm/DebugInfo/DWARF/DWARFSection.h"
27 #include "llvm/DebugInfo/DWARF/DWARFTypeUnit.h"
28 #include "llvm/Object/ObjectFile.h"
29 #include "llvm/Support/DataExtractor.h"
30 #include "llvm/Support/Errc.h"
31 #include "llvm/Support/Path.h"
32 #include <algorithm>
33 #include <cassert>
34 #include <cstddef>
35 #include <cstdint>
36 #include <utility>
37 #include <vector>
38
39 using namespace llvm;
40 using namespace dwarf;
41
addUnitsForSection(DWARFContext & C,const DWARFSection & Section,DWARFSectionKind SectionKind)42 void DWARFUnitVector::addUnitsForSection(DWARFContext &C,
43 const DWARFSection &Section,
44 DWARFSectionKind SectionKind) {
45 const DWARFObject &D = C.getDWARFObj();
46 addUnitsImpl(C, D, Section, C.getDebugAbbrev(), &D.getRangesSection(),
47 &D.getLocSection(), D.getStrSection(),
48 D.getStrOffsetsSection(), &D.getAddrSection(),
49 D.getLineSection(), D.isLittleEndian(), false, false,
50 SectionKind);
51 }
52
addUnitsForDWOSection(DWARFContext & C,const DWARFSection & DWOSection,DWARFSectionKind SectionKind,bool Lazy)53 void DWARFUnitVector::addUnitsForDWOSection(DWARFContext &C,
54 const DWARFSection &DWOSection,
55 DWARFSectionKind SectionKind,
56 bool Lazy) {
57 const DWARFObject &D = C.getDWARFObj();
58 addUnitsImpl(C, D, DWOSection, C.getDebugAbbrevDWO(), &D.getRangesDWOSection(),
59 &D.getLocDWOSection(), D.getStrDWOSection(),
60 D.getStrOffsetsDWOSection(), &D.getAddrSection(),
61 D.getLineDWOSection(), C.isLittleEndian(), true, Lazy,
62 SectionKind);
63 }
64
addUnitsImpl(DWARFContext & Context,const DWARFObject & Obj,const DWARFSection & Section,const DWARFDebugAbbrev * DA,const DWARFSection * RS,const DWARFSection * LocSection,StringRef SS,const DWARFSection & SOS,const DWARFSection * AOS,const DWARFSection & LS,bool LE,bool IsDWO,bool Lazy,DWARFSectionKind SectionKind)65 void DWARFUnitVector::addUnitsImpl(
66 DWARFContext &Context, const DWARFObject &Obj, const DWARFSection &Section,
67 const DWARFDebugAbbrev *DA, const DWARFSection *RS,
68 const DWARFSection *LocSection, StringRef SS, const DWARFSection &SOS,
69 const DWARFSection *AOS, const DWARFSection &LS, bool LE, bool IsDWO,
70 bool Lazy, DWARFSectionKind SectionKind) {
71 DWARFDataExtractor Data(Obj, Section, LE, 0);
72 // Lazy initialization of Parser, now that we have all section info.
73 if (!Parser) {
74 Parser = [=, &Context, &Obj, &Section, &SOS,
75 &LS](uint64_t Offset, DWARFSectionKind SectionKind,
76 const DWARFSection *CurSection,
77 const DWARFUnitIndex::Entry *IndexEntry)
78 -> std::unique_ptr<DWARFUnit> {
79 const DWARFSection &InfoSection = CurSection ? *CurSection : Section;
80 DWARFDataExtractor Data(Obj, InfoSection, LE, 0);
81 if (!Data.isValidOffset(Offset))
82 return nullptr;
83 DWARFUnitHeader Header;
84 if (!Header.extract(Context, Data, &Offset, SectionKind))
85 return nullptr;
86 if (!IndexEntry && IsDWO) {
87 const DWARFUnitIndex &Index = getDWARFUnitIndex(
88 Context, Header.isTypeUnit() ? DW_SECT_EXT_TYPES : DW_SECT_INFO);
89 if (Index) {
90 if (Header.isTypeUnit())
91 IndexEntry = Index.getFromHash(Header.getTypeHash());
92 else if (auto DWOId = Header.getDWOId())
93 IndexEntry = Index.getFromHash(*DWOId);
94 }
95 if (!IndexEntry)
96 IndexEntry = Index.getFromOffset(Header.getOffset());
97 }
98 if (IndexEntry && !Header.applyIndexEntry(IndexEntry))
99 return nullptr;
100 std::unique_ptr<DWARFUnit> U;
101 if (Header.isTypeUnit())
102 U = std::make_unique<DWARFTypeUnit>(Context, InfoSection, Header, DA,
103 RS, LocSection, SS, SOS, AOS, LS,
104 LE, IsDWO, *this);
105 else
106 U = std::make_unique<DWARFCompileUnit>(Context, InfoSection, Header,
107 DA, RS, LocSection, SS, SOS,
108 AOS, LS, LE, IsDWO, *this);
109 return U;
110 };
111 }
112 if (Lazy)
113 return;
114 // Find a reasonable insertion point within the vector. We skip over
115 // (a) units from a different section, (b) units from the same section
116 // but with lower offset-within-section. This keeps units in order
117 // within a section, although not necessarily within the object file,
118 // even if we do lazy parsing.
119 auto I = this->begin();
120 uint64_t Offset = 0;
121 while (Data.isValidOffset(Offset)) {
122 if (I != this->end() &&
123 (&(*I)->getInfoSection() != &Section || (*I)->getOffset() == Offset)) {
124 ++I;
125 continue;
126 }
127 auto U = Parser(Offset, SectionKind, &Section, nullptr);
128 // If parsing failed, we're done with this section.
129 if (!U)
130 break;
131 Offset = U->getNextUnitOffset();
132 I = std::next(this->insert(I, std::move(U)));
133 }
134 }
135
addUnit(std::unique_ptr<DWARFUnit> Unit)136 DWARFUnit *DWARFUnitVector::addUnit(std::unique_ptr<DWARFUnit> Unit) {
137 auto I = llvm::upper_bound(*this, Unit,
138 [](const std::unique_ptr<DWARFUnit> &LHS,
139 const std::unique_ptr<DWARFUnit> &RHS) {
140 return LHS->getOffset() < RHS->getOffset();
141 });
142 return this->insert(I, std::move(Unit))->get();
143 }
144
getUnitForOffset(uint64_t Offset) const145 DWARFUnit *DWARFUnitVector::getUnitForOffset(uint64_t Offset) const {
146 auto end = begin() + getNumInfoUnits();
147 auto *CU =
148 std::upper_bound(begin(), end, Offset,
149 [](uint64_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
150 return LHS < RHS->getNextUnitOffset();
151 });
152 if (CU != end && (*CU)->getOffset() <= Offset)
153 return CU->get();
154 return nullptr;
155 }
156
157 DWARFUnit *
getUnitForIndexEntry(const DWARFUnitIndex::Entry & E)158 DWARFUnitVector::getUnitForIndexEntry(const DWARFUnitIndex::Entry &E) {
159 const auto *CUOff = E.getContribution(DW_SECT_INFO);
160 if (!CUOff)
161 return nullptr;
162
163 uint64_t Offset = CUOff->getOffset();
164 auto end = begin() + getNumInfoUnits();
165
166 auto *CU =
167 std::upper_bound(begin(), end, CUOff->getOffset(),
168 [](uint64_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
169 return LHS < RHS->getNextUnitOffset();
170 });
171 if (CU != end && (*CU)->getOffset() <= Offset)
172 return CU->get();
173
174 if (!Parser)
175 return nullptr;
176
177 auto U = Parser(Offset, DW_SECT_INFO, nullptr, &E);
178 if (!U)
179 U = nullptr;
180
181 auto *NewCU = U.get();
182 this->insert(CU, std::move(U));
183 ++NumInfoUnits;
184 return NewCU;
185 }
186
DWARFUnit(DWARFContext & DC,const DWARFSection & Section,const DWARFUnitHeader & Header,const DWARFDebugAbbrev * DA,const DWARFSection * RS,const DWARFSection * LocSection,StringRef SS,const DWARFSection & SOS,const DWARFSection * AOS,const DWARFSection & LS,bool LE,bool IsDWO,const DWARFUnitVector & UnitVector)187 DWARFUnit::DWARFUnit(DWARFContext &DC, const DWARFSection &Section,
188 const DWARFUnitHeader &Header, const DWARFDebugAbbrev *DA,
189 const DWARFSection *RS, const DWARFSection *LocSection,
190 StringRef SS, const DWARFSection &SOS,
191 const DWARFSection *AOS, const DWARFSection &LS, bool LE,
192 bool IsDWO, const DWARFUnitVector &UnitVector)
193 : Context(DC), InfoSection(Section), Header(Header), Abbrev(DA),
194 RangeSection(RS), LineSection(LS), StringSection(SS),
195 StringOffsetSection(SOS), AddrOffsetSection(AOS), IsLittleEndian(LE),
196 IsDWO(IsDWO), UnitVector(UnitVector) {
197 clear();
198 }
199
200 DWARFUnit::~DWARFUnit() = default;
201
getDebugInfoExtractor() const202 DWARFDataExtractor DWARFUnit::getDebugInfoExtractor() const {
203 return DWARFDataExtractor(Context.getDWARFObj(), InfoSection, IsLittleEndian,
204 getAddressByteSize());
205 }
206
207 std::optional<object::SectionedAddress>
getAddrOffsetSectionItem(uint32_t Index) const208 DWARFUnit::getAddrOffsetSectionItem(uint32_t Index) const {
209 if (!AddrOffsetSectionBase) {
210 auto R = Context.info_section_units();
211 // Surprising if a DWO file has more than one skeleton unit in it - this
212 // probably shouldn't be valid, but if a use case is found, here's where to
213 // support it (probably have to linearly search for the matching skeleton CU
214 // here)
215 if (IsDWO && hasSingleElement(R))
216 return (*R.begin())->getAddrOffsetSectionItem(Index);
217
218 return std::nullopt;
219 }
220
221 uint64_t Offset = *AddrOffsetSectionBase + Index * getAddressByteSize();
222 if (AddrOffsetSection->Data.size() < Offset + getAddressByteSize())
223 return std::nullopt;
224 DWARFDataExtractor DA(Context.getDWARFObj(), *AddrOffsetSection,
225 IsLittleEndian, getAddressByteSize());
226 uint64_t Section;
227 uint64_t Address = DA.getRelocatedAddress(&Offset, &Section);
228 return {{Address, Section}};
229 }
230
getStringOffsetSectionItem(uint32_t Index) const231 Expected<uint64_t> DWARFUnit::getStringOffsetSectionItem(uint32_t Index) const {
232 if (!StringOffsetsTableContribution)
233 return make_error<StringError>(
234 "DW_FORM_strx used without a valid string offsets table",
235 inconvertibleErrorCode());
236 unsigned ItemSize = getDwarfStringOffsetsByteSize();
237 uint64_t Offset = getStringOffsetsBase() + Index * ItemSize;
238 if (StringOffsetSection.Data.size() < Offset + ItemSize)
239 return make_error<StringError>("DW_FORM_strx uses index " + Twine(Index) +
240 ", which is too large",
241 inconvertibleErrorCode());
242 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
243 IsLittleEndian, 0);
244 return DA.getRelocatedValue(ItemSize, &Offset);
245 }
246
extract(DWARFContext & Context,const DWARFDataExtractor & debug_info,uint64_t * offset_ptr,DWARFSectionKind SectionKind)247 bool DWARFUnitHeader::extract(DWARFContext &Context,
248 const DWARFDataExtractor &debug_info,
249 uint64_t *offset_ptr,
250 DWARFSectionKind SectionKind) {
251 Offset = *offset_ptr;
252 Error Err = Error::success();
253 IndexEntry = nullptr;
254 std::tie(Length, FormParams.Format) =
255 debug_info.getInitialLength(offset_ptr, &Err);
256 FormParams.Version = debug_info.getU16(offset_ptr, &Err);
257 if (FormParams.Version >= 5) {
258 UnitType = debug_info.getU8(offset_ptr, &Err);
259 FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
260 AbbrOffset = debug_info.getRelocatedValue(
261 FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
262 } else {
263 AbbrOffset = debug_info.getRelocatedValue(
264 FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
265 FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
266 // Fake a unit type based on the section type. This isn't perfect,
267 // but distinguishing compile and type units is generally enough.
268 if (SectionKind == DW_SECT_EXT_TYPES)
269 UnitType = DW_UT_type;
270 else
271 UnitType = DW_UT_compile;
272 }
273 if (isTypeUnit()) {
274 TypeHash = debug_info.getU64(offset_ptr, &Err);
275 TypeOffset = debug_info.getUnsigned(
276 offset_ptr, FormParams.getDwarfOffsetByteSize(), &Err);
277 } else if (UnitType == DW_UT_split_compile || UnitType == DW_UT_skeleton)
278 DWOId = debug_info.getU64(offset_ptr, &Err);
279
280 if (Err) {
281 Context.getWarningHandler()(joinErrors(
282 createStringError(
283 errc::invalid_argument,
284 "DWARF unit at 0x%8.8" PRIx64 " cannot be parsed:", Offset),
285 std::move(Err)));
286 return false;
287 }
288
289 // Header fields all parsed, capture the size of this unit header.
290 assert(*offset_ptr - Offset <= 255 && "unexpected header size");
291 Size = uint8_t(*offset_ptr - Offset);
292 uint64_t NextCUOffset = Offset + getUnitLengthFieldByteSize() + getLength();
293
294 if (!debug_info.isValidOffset(getNextUnitOffset() - 1)) {
295 Context.getWarningHandler()(
296 createStringError(errc::invalid_argument,
297 "DWARF unit from offset 0x%8.8" PRIx64 " incl. "
298 "to offset 0x%8.8" PRIx64 " excl. "
299 "extends past section size 0x%8.8zx",
300 Offset, NextCUOffset, debug_info.size()));
301 return false;
302 }
303
304 if (!DWARFContext::isSupportedVersion(getVersion())) {
305 Context.getWarningHandler()(createStringError(
306 errc::invalid_argument,
307 "DWARF unit at offset 0x%8.8" PRIx64 " "
308 "has unsupported version %" PRIu16 ", supported are 2-%u",
309 Offset, getVersion(), DWARFContext::getMaxSupportedVersion()));
310 return false;
311 }
312
313 // Type offset is unit-relative; should be after the header and before
314 // the end of the current unit.
315 if (isTypeUnit() && TypeOffset < Size) {
316 Context.getWarningHandler()(
317 createStringError(errc::invalid_argument,
318 "DWARF type unit at offset "
319 "0x%8.8" PRIx64 " "
320 "has its relocated type_offset 0x%8.8" PRIx64 " "
321 "pointing inside the header",
322 Offset, Offset + TypeOffset));
323 return false;
324 }
325 if (isTypeUnit() &&
326 TypeOffset >= getUnitLengthFieldByteSize() + getLength()) {
327 Context.getWarningHandler()(createStringError(
328 errc::invalid_argument,
329 "DWARF type unit from offset 0x%8.8" PRIx64 " incl. "
330 "to offset 0x%8.8" PRIx64 " excl. has its "
331 "relocated type_offset 0x%8.8" PRIx64 " pointing past the unit end",
332 Offset, NextCUOffset, Offset + TypeOffset));
333 return false;
334 }
335
336 if (Error SizeErr = DWARFContext::checkAddressSizeSupported(
337 getAddressByteSize(), errc::invalid_argument,
338 "DWARF unit at offset 0x%8.8" PRIx64, Offset)) {
339 Context.getWarningHandler()(std::move(SizeErr));
340 return false;
341 }
342
343 // Keep track of the highest DWARF version we encounter across all units.
344 Context.setMaxVersionIfGreater(getVersion());
345 return true;
346 }
347
applyIndexEntry(const DWARFUnitIndex::Entry * Entry)348 bool DWARFUnitHeader::applyIndexEntry(const DWARFUnitIndex::Entry *Entry) {
349 assert(Entry);
350 assert(!IndexEntry);
351 IndexEntry = Entry;
352 if (AbbrOffset)
353 return false;
354 auto *UnitContrib = IndexEntry->getContribution();
355 if (!UnitContrib ||
356 UnitContrib->getLength() != (getLength() + getUnitLengthFieldByteSize()))
357 return false;
358 auto *AbbrEntry = IndexEntry->getContribution(DW_SECT_ABBREV);
359 if (!AbbrEntry)
360 return false;
361 AbbrOffset = AbbrEntry->getOffset();
362 return true;
363 }
364
extractRangeList(uint64_t RangeListOffset,DWARFDebugRangeList & RangeList) const365 Error DWARFUnit::extractRangeList(uint64_t RangeListOffset,
366 DWARFDebugRangeList &RangeList) const {
367 // Require that compile unit is extracted.
368 assert(!DieArray.empty());
369 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
370 IsLittleEndian, getAddressByteSize());
371 uint64_t ActualRangeListOffset = RangeSectionBase + RangeListOffset;
372 return RangeList.extract(RangesData, &ActualRangeListOffset);
373 }
374
clear()375 void DWARFUnit::clear() {
376 Abbrevs = nullptr;
377 BaseAddr.reset();
378 RangeSectionBase = 0;
379 LocSectionBase = 0;
380 AddrOffsetSectionBase = std::nullopt;
381 SU = nullptr;
382 clearDIEs(false);
383 AddrDieMap.clear();
384 if (DWO)
385 DWO->clear();
386 DWO.reset();
387 }
388
getCompilationDir()389 const char *DWARFUnit::getCompilationDir() {
390 return dwarf::toString(getUnitDIE().find(DW_AT_comp_dir), nullptr);
391 }
392
extractDIEsToVector(bool AppendCUDie,bool AppendNonCUDies,std::vector<DWARFDebugInfoEntry> & Dies) const393 void DWARFUnit::extractDIEsToVector(
394 bool AppendCUDie, bool AppendNonCUDies,
395 std::vector<DWARFDebugInfoEntry> &Dies) const {
396 if (!AppendCUDie && !AppendNonCUDies)
397 return;
398
399 // Set the offset to that of the first DIE and calculate the start of the
400 // next compilation unit header.
401 uint64_t DIEOffset = getOffset() + getHeaderSize();
402 uint64_t NextCUOffset = getNextUnitOffset();
403 DWARFDebugInfoEntry DIE;
404 DWARFDataExtractor DebugInfoData = getDebugInfoExtractor();
405 // The end offset has been already checked by DWARFUnitHeader::extract.
406 assert(DebugInfoData.isValidOffset(NextCUOffset - 1));
407 std::vector<uint32_t> Parents;
408 std::vector<uint32_t> PrevSiblings;
409 bool IsCUDie = true;
410
411 assert(
412 ((AppendCUDie && Dies.empty()) || (!AppendCUDie && Dies.size() == 1)) &&
413 "Dies array is not empty");
414
415 // Fill Parents and Siblings stacks with initial value.
416 Parents.push_back(UINT32_MAX);
417 if (!AppendCUDie)
418 Parents.push_back(0);
419 PrevSiblings.push_back(0);
420
421 // Start to extract dies.
422 do {
423 assert(Parents.size() > 0 && "Empty parents stack");
424 assert((Parents.back() == UINT32_MAX || Parents.back() <= Dies.size()) &&
425 "Wrong parent index");
426
427 // Extract die. Stop if any error occurred.
428 if (!DIE.extractFast(*this, &DIEOffset, DebugInfoData, NextCUOffset,
429 Parents.back()))
430 break;
431
432 // If previous sibling is remembered then update it`s SiblingIdx field.
433 if (PrevSiblings.back() > 0) {
434 assert(PrevSiblings.back() < Dies.size() &&
435 "Previous sibling index is out of Dies boundaries");
436 Dies[PrevSiblings.back()].setSiblingIdx(Dies.size());
437 }
438
439 // Store die into the Dies vector.
440 if (IsCUDie) {
441 if (AppendCUDie)
442 Dies.push_back(DIE);
443 if (!AppendNonCUDies)
444 break;
445 // The average bytes per DIE entry has been seen to be
446 // around 14-20 so let's pre-reserve the needed memory for
447 // our DIE entries accordingly.
448 Dies.reserve(Dies.size() + getDebugInfoSize() / 14);
449 } else {
450 // Remember last previous sibling.
451 PrevSiblings.back() = Dies.size();
452
453 Dies.push_back(DIE);
454 }
455
456 // Check for new children scope.
457 if (const DWARFAbbreviationDeclaration *AbbrDecl =
458 DIE.getAbbreviationDeclarationPtr()) {
459 if (AbbrDecl->hasChildren()) {
460 if (AppendCUDie || !IsCUDie) {
461 assert(Dies.size() > 0 && "Dies does not contain any die");
462 Parents.push_back(Dies.size() - 1);
463 PrevSiblings.push_back(0);
464 }
465 } else if (IsCUDie)
466 // Stop if we have single compile unit die w/o children.
467 break;
468 } else {
469 // NULL DIE: finishes current children scope.
470 Parents.pop_back();
471 PrevSiblings.pop_back();
472 }
473
474 if (IsCUDie)
475 IsCUDie = false;
476
477 // Stop when compile unit die is removed from the parents stack.
478 } while (Parents.size() > 1);
479 }
480
extractDIEsIfNeeded(bool CUDieOnly)481 void DWARFUnit::extractDIEsIfNeeded(bool CUDieOnly) {
482 if (Error e = tryExtractDIEsIfNeeded(CUDieOnly))
483 Context.getRecoverableErrorHandler()(std::move(e));
484 }
485
tryExtractDIEsIfNeeded(bool CUDieOnly)486 Error DWARFUnit::tryExtractDIEsIfNeeded(bool CUDieOnly) {
487 if ((CUDieOnly && !DieArray.empty()) ||
488 DieArray.size() > 1)
489 return Error::success(); // Already parsed.
490
491 bool HasCUDie = !DieArray.empty();
492 extractDIEsToVector(!HasCUDie, !CUDieOnly, DieArray);
493
494 if (DieArray.empty())
495 return Error::success();
496
497 // If CU DIE was just parsed, copy several attribute values from it.
498 if (HasCUDie)
499 return Error::success();
500
501 DWARFDie UnitDie(this, &DieArray[0]);
502 if (std::optional<uint64_t> DWOId =
503 toUnsigned(UnitDie.find(DW_AT_GNU_dwo_id)))
504 Header.setDWOId(*DWOId);
505 if (!IsDWO) {
506 assert(AddrOffsetSectionBase == std::nullopt);
507 assert(RangeSectionBase == 0);
508 assert(LocSectionBase == 0);
509 AddrOffsetSectionBase = toSectionOffset(UnitDie.find(DW_AT_addr_base));
510 if (!AddrOffsetSectionBase)
511 AddrOffsetSectionBase =
512 toSectionOffset(UnitDie.find(DW_AT_GNU_addr_base));
513 RangeSectionBase = toSectionOffset(UnitDie.find(DW_AT_rnglists_base), 0);
514 LocSectionBase = toSectionOffset(UnitDie.find(DW_AT_loclists_base), 0);
515 }
516
517 // In general, in DWARF v5 and beyond we derive the start of the unit's
518 // contribution to the string offsets table from the unit DIE's
519 // DW_AT_str_offsets_base attribute. Split DWARF units do not use this
520 // attribute, so we assume that there is a contribution to the string
521 // offsets table starting at offset 0 of the debug_str_offsets.dwo section.
522 // In both cases we need to determine the format of the contribution,
523 // which may differ from the unit's format.
524 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
525 IsLittleEndian, 0);
526 if (IsDWO || getVersion() >= 5) {
527 auto StringOffsetOrError =
528 IsDWO ? determineStringOffsetsTableContributionDWO(DA)
529 : determineStringOffsetsTableContribution(DA);
530 if (!StringOffsetOrError)
531 return createStringError(errc::invalid_argument,
532 "invalid reference to or invalid content in "
533 ".debug_str_offsets[.dwo]: " +
534 toString(StringOffsetOrError.takeError()));
535
536 StringOffsetsTableContribution = *StringOffsetOrError;
537 }
538
539 // DWARF v5 uses the .debug_rnglists and .debug_rnglists.dwo sections to
540 // describe address ranges.
541 if (getVersion() >= 5) {
542 // In case of DWP, the base offset from the index has to be added.
543 if (IsDWO) {
544 uint64_t ContributionBaseOffset = 0;
545 if (auto *IndexEntry = Header.getIndexEntry())
546 if (auto *Contrib = IndexEntry->getContribution(DW_SECT_RNGLISTS))
547 ContributionBaseOffset = Contrib->getOffset();
548 setRangesSection(
549 &Context.getDWARFObj().getRnglistsDWOSection(),
550 ContributionBaseOffset +
551 DWARFListTableHeader::getHeaderSize(Header.getFormat()));
552 } else
553 setRangesSection(&Context.getDWARFObj().getRnglistsSection(),
554 toSectionOffset(UnitDie.find(DW_AT_rnglists_base),
555 DWARFListTableHeader::getHeaderSize(
556 Header.getFormat())));
557 }
558
559 if (IsDWO) {
560 // If we are reading a package file, we need to adjust the location list
561 // data based on the index entries.
562 StringRef Data = Header.getVersion() >= 5
563 ? Context.getDWARFObj().getLoclistsDWOSection().Data
564 : Context.getDWARFObj().getLocDWOSection().Data;
565 if (auto *IndexEntry = Header.getIndexEntry())
566 if (const auto *C = IndexEntry->getContribution(
567 Header.getVersion() >= 5 ? DW_SECT_LOCLISTS : DW_SECT_EXT_LOC))
568 Data = Data.substr(C->getOffset(), C->getLength());
569
570 DWARFDataExtractor DWARFData(Data, IsLittleEndian, getAddressByteSize());
571 LocTable =
572 std::make_unique<DWARFDebugLoclists>(DWARFData, Header.getVersion());
573 LocSectionBase = DWARFListTableHeader::getHeaderSize(Header.getFormat());
574 } else if (getVersion() >= 5) {
575 LocTable = std::make_unique<DWARFDebugLoclists>(
576 DWARFDataExtractor(Context.getDWARFObj(),
577 Context.getDWARFObj().getLoclistsSection(),
578 IsLittleEndian, getAddressByteSize()),
579 getVersion());
580 } else {
581 LocTable = std::make_unique<DWARFDebugLoc>(DWARFDataExtractor(
582 Context.getDWARFObj(), Context.getDWARFObj().getLocSection(),
583 IsLittleEndian, getAddressByteSize()));
584 }
585
586 // Don't fall back to DW_AT_GNU_ranges_base: it should be ignored for
587 // skeleton CU DIE, so that DWARF users not aware of it are not broken.
588 return Error::success();
589 }
590
parseDWO(StringRef DWOAlternativeLocation)591 bool DWARFUnit::parseDWO(StringRef DWOAlternativeLocation) {
592 if (IsDWO)
593 return false;
594 if (DWO)
595 return false;
596 DWARFDie UnitDie = getUnitDIE();
597 if (!UnitDie)
598 return false;
599 auto DWOFileName = getVersion() >= 5
600 ? dwarf::toString(UnitDie.find(DW_AT_dwo_name))
601 : dwarf::toString(UnitDie.find(DW_AT_GNU_dwo_name));
602 if (!DWOFileName)
603 return false;
604 auto CompilationDir = dwarf::toString(UnitDie.find(DW_AT_comp_dir));
605 SmallString<16> AbsolutePath;
606 if (sys::path::is_relative(*DWOFileName) && CompilationDir &&
607 *CompilationDir) {
608 sys::path::append(AbsolutePath, *CompilationDir);
609 }
610 sys::path::append(AbsolutePath, *DWOFileName);
611 auto DWOId = getDWOId();
612 if (!DWOId)
613 return false;
614 auto DWOContext = Context.getDWOContext(AbsolutePath);
615 if (!DWOContext) {
616 // Use the alternative location to get the DWARF context for the DWO object.
617 if (DWOAlternativeLocation.empty())
618 return false;
619 // If the alternative context does not correspond to the original DWO object
620 // (different hashes), the below 'getDWOCompileUnitForHash' call will catch
621 // the issue, with a returned null context.
622 DWOContext = Context.getDWOContext(DWOAlternativeLocation);
623 if (!DWOContext)
624 return false;
625 }
626
627 DWARFCompileUnit *DWOCU = DWOContext->getDWOCompileUnitForHash(*DWOId);
628 if (!DWOCU)
629 return false;
630 DWO = std::shared_ptr<DWARFCompileUnit>(std::move(DWOContext), DWOCU);
631 DWO->setSkeletonUnit(this);
632 // Share .debug_addr and .debug_ranges section with compile unit in .dwo
633 if (AddrOffsetSectionBase)
634 DWO->setAddrOffsetSection(AddrOffsetSection, *AddrOffsetSectionBase);
635 if (getVersion() == 4) {
636 auto DWORangesBase = UnitDie.getRangesBaseAttribute();
637 DWO->setRangesSection(RangeSection, DWORangesBase.value_or(0));
638 }
639
640 return true;
641 }
642
clearDIEs(bool KeepCUDie)643 void DWARFUnit::clearDIEs(bool KeepCUDie) {
644 // Do not use resize() + shrink_to_fit() to free memory occupied by dies.
645 // shrink_to_fit() is a *non-binding* request to reduce capacity() to size().
646 // It depends on the implementation whether the request is fulfilled.
647 // Create a new vector with a small capacity and assign it to the DieArray to
648 // have previous contents freed.
649 DieArray = (KeepCUDie && !DieArray.empty())
650 ? std::vector<DWARFDebugInfoEntry>({DieArray[0]})
651 : std::vector<DWARFDebugInfoEntry>();
652 }
653
654 Expected<DWARFAddressRangesVector>
findRnglistFromOffset(uint64_t Offset)655 DWARFUnit::findRnglistFromOffset(uint64_t Offset) {
656 if (getVersion() <= 4) {
657 DWARFDebugRangeList RangeList;
658 if (Error E = extractRangeList(Offset, RangeList))
659 return std::move(E);
660 return RangeList.getAbsoluteRanges(getBaseAddress());
661 }
662 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
663 IsLittleEndian, Header.getAddressByteSize());
664 DWARFDebugRnglistTable RnglistTable;
665 auto RangeListOrError = RnglistTable.findList(RangesData, Offset);
666 if (RangeListOrError)
667 return RangeListOrError.get().getAbsoluteRanges(getBaseAddress(), *this);
668 return RangeListOrError.takeError();
669 }
670
671 Expected<DWARFAddressRangesVector>
findRnglistFromIndex(uint32_t Index)672 DWARFUnit::findRnglistFromIndex(uint32_t Index) {
673 if (auto Offset = getRnglistOffset(Index))
674 return findRnglistFromOffset(*Offset);
675
676 return createStringError(errc::invalid_argument,
677 "invalid range list table index %d (possibly "
678 "missing the entire range list table)",
679 Index);
680 }
681
collectAddressRanges()682 Expected<DWARFAddressRangesVector> DWARFUnit::collectAddressRanges() {
683 DWARFDie UnitDie = getUnitDIE();
684 if (!UnitDie)
685 return createStringError(errc::invalid_argument, "No unit DIE");
686
687 // First, check if unit DIE describes address ranges for the whole unit.
688 auto CUDIERangesOrError = UnitDie.getAddressRanges();
689 if (!CUDIERangesOrError)
690 return createStringError(errc::invalid_argument,
691 "decoding address ranges: %s",
692 toString(CUDIERangesOrError.takeError()).c_str());
693 return *CUDIERangesOrError;
694 }
695
696 Expected<DWARFLocationExpressionsVector>
findLoclistFromOffset(uint64_t Offset)697 DWARFUnit::findLoclistFromOffset(uint64_t Offset) {
698 DWARFLocationExpressionsVector Result;
699
700 Error InterpretationError = Error::success();
701
702 Error ParseError = getLocationTable().visitAbsoluteLocationList(
703 Offset, getBaseAddress(),
704 [this](uint32_t Index) { return getAddrOffsetSectionItem(Index); },
705 [&](Expected<DWARFLocationExpression> L) {
706 if (L)
707 Result.push_back(std::move(*L));
708 else
709 InterpretationError =
710 joinErrors(L.takeError(), std::move(InterpretationError));
711 return !InterpretationError;
712 });
713
714 if (ParseError || InterpretationError)
715 return joinErrors(std::move(ParseError), std::move(InterpretationError));
716
717 return Result;
718 }
719
updateAddressDieMap(DWARFDie Die)720 void DWARFUnit::updateAddressDieMap(DWARFDie Die) {
721 if (Die.isSubroutineDIE()) {
722 auto DIERangesOrError = Die.getAddressRanges();
723 if (DIERangesOrError) {
724 for (const auto &R : DIERangesOrError.get()) {
725 // Ignore 0-sized ranges.
726 if (R.LowPC == R.HighPC)
727 continue;
728 auto B = AddrDieMap.upper_bound(R.LowPC);
729 if (B != AddrDieMap.begin() && R.LowPC < (--B)->second.first) {
730 // The range is a sub-range of existing ranges, we need to split the
731 // existing range.
732 if (R.HighPC < B->second.first)
733 AddrDieMap[R.HighPC] = B->second;
734 if (R.LowPC > B->first)
735 AddrDieMap[B->first].first = R.LowPC;
736 }
737 AddrDieMap[R.LowPC] = std::make_pair(R.HighPC, Die);
738 }
739 } else
740 llvm::consumeError(DIERangesOrError.takeError());
741 }
742 // Parent DIEs are added to the AddrDieMap prior to the Children DIEs to
743 // simplify the logic to update AddrDieMap. The child's range will always
744 // be equal or smaller than the parent's range. With this assumption, when
745 // adding one range into the map, it will at most split a range into 3
746 // sub-ranges.
747 for (DWARFDie Child = Die.getFirstChild(); Child; Child = Child.getSibling())
748 updateAddressDieMap(Child);
749 }
750
getSubroutineForAddress(uint64_t Address)751 DWARFDie DWARFUnit::getSubroutineForAddress(uint64_t Address) {
752 extractDIEsIfNeeded(false);
753 if (AddrDieMap.empty())
754 updateAddressDieMap(getUnitDIE());
755 auto R = AddrDieMap.upper_bound(Address);
756 if (R == AddrDieMap.begin())
757 return DWARFDie();
758 // upper_bound's previous item contains Address.
759 --R;
760 if (Address >= R->second.first)
761 return DWARFDie();
762 return R->second.second;
763 }
764
updateVariableDieMap(DWARFDie Die)765 void DWARFUnit::updateVariableDieMap(DWARFDie Die) {
766 for (DWARFDie Child : Die) {
767 if (isType(Child.getTag()))
768 continue;
769 updateVariableDieMap(Child);
770 }
771
772 if (Die.getTag() != DW_TAG_variable)
773 return;
774
775 Expected<DWARFLocationExpressionsVector> Locations =
776 Die.getLocations(DW_AT_location);
777 if (!Locations) {
778 // Missing DW_AT_location is fine here.
779 consumeError(Locations.takeError());
780 return;
781 }
782
783 uint64_t Address = UINT64_MAX;
784
785 for (const DWARFLocationExpression &Location : *Locations) {
786 uint8_t AddressSize = getAddressByteSize();
787 DataExtractor Data(Location.Expr, /*IsLittleEndian=*/true, AddressSize);
788 DWARFExpression Expr(Data, AddressSize);
789 auto It = Expr.begin();
790 if (It == Expr.end())
791 continue;
792
793 // Match exactly the main sequence used to describe global variables:
794 // `DW_OP_addr[x] [+ DW_OP_plus_uconst]`. Currently, this is the sequence
795 // that LLVM produces for DILocalVariables and DIGlobalVariables. If, in
796 // future, the DWARF producer (`DwarfCompileUnit::addLocationAttribute()` is
797 // a good starting point) is extended to use further expressions, this code
798 // needs to be updated.
799 uint64_t LocationAddr;
800 if (It->getCode() == dwarf::DW_OP_addr) {
801 LocationAddr = It->getRawOperand(0);
802 } else if (It->getCode() == dwarf::DW_OP_addrx) {
803 uint64_t DebugAddrOffset = It->getRawOperand(0);
804 if (auto Pointer = getAddrOffsetSectionItem(DebugAddrOffset)) {
805 LocationAddr = Pointer->Address;
806 }
807 } else {
808 continue;
809 }
810
811 // Read the optional 2nd operand, a DW_OP_plus_uconst.
812 if (++It != Expr.end()) {
813 if (It->getCode() != dwarf::DW_OP_plus_uconst)
814 continue;
815
816 LocationAddr += It->getRawOperand(0);
817
818 // Probe for a 3rd operand, if it exists, bail.
819 if (++It != Expr.end())
820 continue;
821 }
822
823 Address = LocationAddr;
824 break;
825 }
826
827 // Get the size of the global variable. If all else fails (i.e. the global has
828 // no type), then we use a size of one to still allow symbolization of the
829 // exact address.
830 uint64_t GVSize = 1;
831 if (DWARFDie BaseType = Die.getAttributeValueAsReferencedDie(DW_AT_type))
832 if (std::optional<uint64_t> Size = Die.getTypeSize(getAddressByteSize()))
833 GVSize = *Size;
834
835 if (Address != UINT64_MAX)
836 VariableDieMap[Address] = {Address + GVSize, Die};
837 }
838
getVariableForAddress(uint64_t Address)839 DWARFDie DWARFUnit::getVariableForAddress(uint64_t Address) {
840 extractDIEsIfNeeded(false);
841
842 auto RootDie = getUnitDIE();
843
844 auto RootLookup = RootsParsedForVariables.insert(RootDie.getOffset());
845 if (RootLookup.second)
846 updateVariableDieMap(RootDie);
847
848 auto R = VariableDieMap.upper_bound(Address);
849 if (R == VariableDieMap.begin())
850 return DWARFDie();
851
852 // upper_bound's previous item contains Address.
853 --R;
854 if (Address >= R->second.first)
855 return DWARFDie();
856 return R->second.second;
857 }
858
859 void
getInlinedChainForAddress(uint64_t Address,SmallVectorImpl<DWARFDie> & InlinedChain)860 DWARFUnit::getInlinedChainForAddress(uint64_t Address,
861 SmallVectorImpl<DWARFDie> &InlinedChain) {
862 assert(InlinedChain.empty());
863 // Try to look for subprogram DIEs in the DWO file.
864 parseDWO();
865 // First, find the subroutine that contains the given address (the leaf
866 // of inlined chain).
867 DWARFDie SubroutineDIE =
868 (DWO ? *DWO : *this).getSubroutineForAddress(Address);
869
870 while (SubroutineDIE) {
871 if (SubroutineDIE.isSubprogramDIE()) {
872 InlinedChain.push_back(SubroutineDIE);
873 return;
874 }
875 if (SubroutineDIE.getTag() == DW_TAG_inlined_subroutine)
876 InlinedChain.push_back(SubroutineDIE);
877 SubroutineDIE = SubroutineDIE.getParent();
878 }
879 }
880
getDWARFUnitIndex(DWARFContext & Context,DWARFSectionKind Kind)881 const DWARFUnitIndex &llvm::getDWARFUnitIndex(DWARFContext &Context,
882 DWARFSectionKind Kind) {
883 if (Kind == DW_SECT_INFO)
884 return Context.getCUIndex();
885 assert(Kind == DW_SECT_EXT_TYPES);
886 return Context.getTUIndex();
887 }
888
getParent(const DWARFDebugInfoEntry * Die)889 DWARFDie DWARFUnit::getParent(const DWARFDebugInfoEntry *Die) {
890 if (const DWARFDebugInfoEntry *Entry = getParentEntry(Die))
891 return DWARFDie(this, Entry);
892
893 return DWARFDie();
894 }
895
896 const DWARFDebugInfoEntry *
getParentEntry(const DWARFDebugInfoEntry * Die) const897 DWARFUnit::getParentEntry(const DWARFDebugInfoEntry *Die) const {
898 if (!Die)
899 return nullptr;
900 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
901
902 if (std::optional<uint32_t> ParentIdx = Die->getParentIdx()) {
903 assert(*ParentIdx < DieArray.size() &&
904 "ParentIdx is out of DieArray boundaries");
905 return getDebugInfoEntry(*ParentIdx);
906 }
907
908 return nullptr;
909 }
910
getSibling(const DWARFDebugInfoEntry * Die)911 DWARFDie DWARFUnit::getSibling(const DWARFDebugInfoEntry *Die) {
912 if (const DWARFDebugInfoEntry *Sibling = getSiblingEntry(Die))
913 return DWARFDie(this, Sibling);
914
915 return DWARFDie();
916 }
917
918 const DWARFDebugInfoEntry *
getSiblingEntry(const DWARFDebugInfoEntry * Die) const919 DWARFUnit::getSiblingEntry(const DWARFDebugInfoEntry *Die) const {
920 if (!Die)
921 return nullptr;
922 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
923
924 if (std::optional<uint32_t> SiblingIdx = Die->getSiblingIdx()) {
925 assert(*SiblingIdx < DieArray.size() &&
926 "SiblingIdx is out of DieArray boundaries");
927 return &DieArray[*SiblingIdx];
928 }
929
930 return nullptr;
931 }
932
getPreviousSibling(const DWARFDebugInfoEntry * Die)933 DWARFDie DWARFUnit::getPreviousSibling(const DWARFDebugInfoEntry *Die) {
934 if (const DWARFDebugInfoEntry *Sibling = getPreviousSiblingEntry(Die))
935 return DWARFDie(this, Sibling);
936
937 return DWARFDie();
938 }
939
940 const DWARFDebugInfoEntry *
getPreviousSiblingEntry(const DWARFDebugInfoEntry * Die) const941 DWARFUnit::getPreviousSiblingEntry(const DWARFDebugInfoEntry *Die) const {
942 if (!Die)
943 return nullptr;
944 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
945
946 std::optional<uint32_t> ParentIdx = Die->getParentIdx();
947 if (!ParentIdx)
948 // Die is a root die, there is no previous sibling.
949 return nullptr;
950
951 assert(*ParentIdx < DieArray.size() &&
952 "ParentIdx is out of DieArray boundaries");
953 assert(getDIEIndex(Die) > 0 && "Die is a root die");
954
955 uint32_t PrevDieIdx = getDIEIndex(Die) - 1;
956 if (PrevDieIdx == *ParentIdx)
957 // Immediately previous node is parent, there is no previous sibling.
958 return nullptr;
959
960 while (DieArray[PrevDieIdx].getParentIdx() != *ParentIdx) {
961 PrevDieIdx = *DieArray[PrevDieIdx].getParentIdx();
962
963 assert(PrevDieIdx < DieArray.size() &&
964 "PrevDieIdx is out of DieArray boundaries");
965 assert(PrevDieIdx >= *ParentIdx &&
966 "PrevDieIdx is not a child of parent of Die");
967 }
968
969 return &DieArray[PrevDieIdx];
970 }
971
getFirstChild(const DWARFDebugInfoEntry * Die)972 DWARFDie DWARFUnit::getFirstChild(const DWARFDebugInfoEntry *Die) {
973 if (const DWARFDebugInfoEntry *Child = getFirstChildEntry(Die))
974 return DWARFDie(this, Child);
975
976 return DWARFDie();
977 }
978
979 const DWARFDebugInfoEntry *
getFirstChildEntry(const DWARFDebugInfoEntry * Die) const980 DWARFUnit::getFirstChildEntry(const DWARFDebugInfoEntry *Die) const {
981 if (!Die)
982 return nullptr;
983 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
984
985 if (!Die->hasChildren())
986 return nullptr;
987
988 // TODO: Instead of checking here for invalid die we might reject
989 // invalid dies at parsing stage(DWARFUnit::extractDIEsToVector).
990 // We do not want access out of bounds when parsing corrupted debug data.
991 size_t I = getDIEIndex(Die) + 1;
992 if (I >= DieArray.size())
993 return nullptr;
994 return &DieArray[I];
995 }
996
getLastChild(const DWARFDebugInfoEntry * Die)997 DWARFDie DWARFUnit::getLastChild(const DWARFDebugInfoEntry *Die) {
998 if (const DWARFDebugInfoEntry *Child = getLastChildEntry(Die))
999 return DWARFDie(this, Child);
1000
1001 return DWARFDie();
1002 }
1003
1004 const DWARFDebugInfoEntry *
getLastChildEntry(const DWARFDebugInfoEntry * Die) const1005 DWARFUnit::getLastChildEntry(const DWARFDebugInfoEntry *Die) const {
1006 if (!Die)
1007 return nullptr;
1008 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
1009
1010 if (!Die->hasChildren())
1011 return nullptr;
1012
1013 if (std::optional<uint32_t> SiblingIdx = Die->getSiblingIdx()) {
1014 assert(*SiblingIdx < DieArray.size() &&
1015 "SiblingIdx is out of DieArray boundaries");
1016 assert(DieArray[*SiblingIdx - 1].getTag() == dwarf::DW_TAG_null &&
1017 "Bad end of children marker");
1018 return &DieArray[*SiblingIdx - 1];
1019 }
1020
1021 // If SiblingIdx is set for non-root dies we could be sure that DWARF is
1022 // correct and "end of children marker" must be found. For root die we do not
1023 // have such a guarantee(parsing root die might be stopped if "end of children
1024 // marker" is missing, SiblingIdx is always zero for root die). That is why we
1025 // do not use assertion for checking for "end of children marker" for root
1026 // die.
1027
1028 // TODO: Instead of checking here for invalid die we might reject
1029 // invalid dies at parsing stage(DWARFUnit::extractDIEsToVector).
1030 if (getDIEIndex(Die) == 0 && DieArray.size() > 1 &&
1031 DieArray.back().getTag() == dwarf::DW_TAG_null) {
1032 // For the unit die we might take last item from DieArray.
1033 assert(getDIEIndex(Die) ==
1034 getDIEIndex(const_cast<DWARFUnit *>(this)->getUnitDIE()) &&
1035 "Bad unit die");
1036 return &DieArray.back();
1037 }
1038
1039 return nullptr;
1040 }
1041
getAbbreviations() const1042 const DWARFAbbreviationDeclarationSet *DWARFUnit::getAbbreviations() const {
1043 if (!Abbrevs)
1044 Abbrevs = Abbrev->getAbbreviationDeclarationSet(getAbbreviationsOffset());
1045 return Abbrevs;
1046 }
1047
getBaseAddress()1048 std::optional<object::SectionedAddress> DWARFUnit::getBaseAddress() {
1049 if (BaseAddr)
1050 return BaseAddr;
1051
1052 DWARFDie UnitDie = getUnitDIE();
1053 std::optional<DWARFFormValue> PC =
1054 UnitDie.find({DW_AT_low_pc, DW_AT_entry_pc});
1055 BaseAddr = toSectionedAddress(PC);
1056 return BaseAddr;
1057 }
1058
1059 Expected<StrOffsetsContributionDescriptor>
validateContributionSize(DWARFDataExtractor & DA)1060 StrOffsetsContributionDescriptor::validateContributionSize(
1061 DWARFDataExtractor &DA) {
1062 uint8_t EntrySize = getDwarfOffsetByteSize();
1063 // In order to ensure that we don't read a partial record at the end of
1064 // the section we validate for a multiple of the entry size.
1065 uint64_t ValidationSize = alignTo(Size, EntrySize);
1066 // Guard against overflow.
1067 if (ValidationSize >= Size)
1068 if (DA.isValidOffsetForDataOfSize((uint32_t)Base, ValidationSize))
1069 return *this;
1070 return createStringError(errc::invalid_argument, "length exceeds section size");
1071 }
1072
1073 // Look for a DWARF64-formatted contribution to the string offsets table
1074 // starting at a given offset and record it in a descriptor.
1075 static Expected<StrOffsetsContributionDescriptor>
parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)1076 parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
1077 if (!DA.isValidOffsetForDataOfSize(Offset, 16))
1078 return createStringError(errc::invalid_argument, "section offset exceeds section size");
1079
1080 if (DA.getU32(&Offset) != dwarf::DW_LENGTH_DWARF64)
1081 return createStringError(errc::invalid_argument, "32 bit contribution referenced from a 64 bit unit");
1082
1083 uint64_t Size = DA.getU64(&Offset);
1084 uint8_t Version = DA.getU16(&Offset);
1085 (void)DA.getU16(&Offset); // padding
1086 // The encoded length includes the 2-byte version field and the 2-byte
1087 // padding, so we need to subtract them out when we populate the descriptor.
1088 return StrOffsetsContributionDescriptor(Offset, Size - 4, Version, DWARF64);
1089 }
1090
1091 // Look for a DWARF32-formatted contribution to the string offsets table
1092 // starting at a given offset and record it in a descriptor.
1093 static Expected<StrOffsetsContributionDescriptor>
parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)1094 parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
1095 if (!DA.isValidOffsetForDataOfSize(Offset, 8))
1096 return createStringError(errc::invalid_argument, "section offset exceeds section size");
1097
1098 uint32_t ContributionSize = DA.getU32(&Offset);
1099 if (ContributionSize >= dwarf::DW_LENGTH_lo_reserved)
1100 return createStringError(errc::invalid_argument, "invalid length");
1101
1102 uint8_t Version = DA.getU16(&Offset);
1103 (void)DA.getU16(&Offset); // padding
1104 // The encoded length includes the 2-byte version field and the 2-byte
1105 // padding, so we need to subtract them out when we populate the descriptor.
1106 return StrOffsetsContributionDescriptor(Offset, ContributionSize - 4, Version,
1107 DWARF32);
1108 }
1109
1110 static Expected<StrOffsetsContributionDescriptor>
parseDWARFStringOffsetsTableHeader(DWARFDataExtractor & DA,llvm::dwarf::DwarfFormat Format,uint64_t Offset)1111 parseDWARFStringOffsetsTableHeader(DWARFDataExtractor &DA,
1112 llvm::dwarf::DwarfFormat Format,
1113 uint64_t Offset) {
1114 StrOffsetsContributionDescriptor Desc;
1115 switch (Format) {
1116 case dwarf::DwarfFormat::DWARF64: {
1117 if (Offset < 16)
1118 return createStringError(errc::invalid_argument, "insufficient space for 64 bit header prefix");
1119 auto DescOrError = parseDWARF64StringOffsetsTableHeader(DA, Offset - 16);
1120 if (!DescOrError)
1121 return DescOrError.takeError();
1122 Desc = *DescOrError;
1123 break;
1124 }
1125 case dwarf::DwarfFormat::DWARF32: {
1126 if (Offset < 8)
1127 return createStringError(errc::invalid_argument, "insufficient space for 32 bit header prefix");
1128 auto DescOrError = parseDWARF32StringOffsetsTableHeader(DA, Offset - 8);
1129 if (!DescOrError)
1130 return DescOrError.takeError();
1131 Desc = *DescOrError;
1132 break;
1133 }
1134 }
1135 return Desc.validateContributionSize(DA);
1136 }
1137
1138 Expected<std::optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContribution(DWARFDataExtractor & DA)1139 DWARFUnit::determineStringOffsetsTableContribution(DWARFDataExtractor &DA) {
1140 assert(!IsDWO);
1141 auto OptOffset = toSectionOffset(getUnitDIE().find(DW_AT_str_offsets_base));
1142 if (!OptOffset)
1143 return std::nullopt;
1144 auto DescOrError =
1145 parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), *OptOffset);
1146 if (!DescOrError)
1147 return DescOrError.takeError();
1148 return *DescOrError;
1149 }
1150
1151 Expected<std::optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA)1152 DWARFUnit::determineStringOffsetsTableContributionDWO(DWARFDataExtractor &DA) {
1153 assert(IsDWO);
1154 uint64_t Offset = 0;
1155 auto IndexEntry = Header.getIndexEntry();
1156 const auto *C =
1157 IndexEntry ? IndexEntry->getContribution(DW_SECT_STR_OFFSETS) : nullptr;
1158 if (C)
1159 Offset = C->getOffset();
1160 if (getVersion() >= 5) {
1161 if (DA.getData().data() == nullptr)
1162 return std::nullopt;
1163 Offset += Header.getFormat() == dwarf::DwarfFormat::DWARF32 ? 8 : 16;
1164 // Look for a valid contribution at the given offset.
1165 auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), Offset);
1166 if (!DescOrError)
1167 return DescOrError.takeError();
1168 return *DescOrError;
1169 }
1170 // Prior to DWARF v5, we derive the contribution size from the
1171 // index table (in a package file). In a .dwo file it is simply
1172 // the length of the string offsets section.
1173 StrOffsetsContributionDescriptor Desc;
1174 if (C)
1175 Desc = StrOffsetsContributionDescriptor(C->getOffset(), C->getLength(), 4,
1176 Header.getFormat());
1177 else if (!IndexEntry && !StringOffsetSection.Data.empty())
1178 Desc = StrOffsetsContributionDescriptor(0, StringOffsetSection.Data.size(),
1179 4, Header.getFormat());
1180 else
1181 return std::nullopt;
1182 auto DescOrError = Desc.validateContributionSize(DA);
1183 if (!DescOrError)
1184 return DescOrError.takeError();
1185 return *DescOrError;
1186 }
1187
getRnglistOffset(uint32_t Index)1188 std::optional<uint64_t> DWARFUnit::getRnglistOffset(uint32_t Index) {
1189 DataExtractor RangesData(RangeSection->Data, IsLittleEndian,
1190 getAddressByteSize());
1191 DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
1192 IsLittleEndian, 0);
1193 if (std::optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
1194 RangesData, RangeSectionBase, getFormat(), Index))
1195 return *Off + RangeSectionBase;
1196 return std::nullopt;
1197 }
1198
getLoclistOffset(uint32_t Index)1199 std::optional<uint64_t> DWARFUnit::getLoclistOffset(uint32_t Index) {
1200 if (std::optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
1201 LocTable->getData(), LocSectionBase, getFormat(), Index))
1202 return *Off + LocSectionBase;
1203 return std::nullopt;
1204 }
1205