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 (Error ExtractErr =
85               Header.extract(Context, Data, &Offset, SectionKind)) {
86         Context.getWarningHandler()(std::move(ExtractErr));
87         return nullptr;
88       }
89       if (!IndexEntry && IsDWO) {
90         const DWARFUnitIndex &Index = getDWARFUnitIndex(
91             Context, Header.isTypeUnit() ? DW_SECT_EXT_TYPES : DW_SECT_INFO);
92         if (Index) {
93           if (Header.isTypeUnit())
94             IndexEntry = Index.getFromHash(Header.getTypeHash());
95           else if (auto DWOId = Header.getDWOId())
96             IndexEntry = Index.getFromHash(*DWOId);
97         }
98         if (!IndexEntry)
99           IndexEntry = Index.getFromOffset(Header.getOffset());
100       }
101       if (IndexEntry && !Header.applyIndexEntry(IndexEntry))
102         return nullptr;
103       std::unique_ptr<DWARFUnit> U;
104       if (Header.isTypeUnit())
105         U = std::make_unique<DWARFTypeUnit>(Context, InfoSection, Header, DA,
106                                              RS, LocSection, SS, SOS, AOS, LS,
107                                              LE, IsDWO, *this);
108       else
109         U = std::make_unique<DWARFCompileUnit>(Context, InfoSection, Header,
110                                                 DA, RS, LocSection, SS, SOS,
111                                                 AOS, LS, LE, IsDWO, *this);
112       return U;
113     };
114   }
115   if (Lazy)
116     return;
117   // Find a reasonable insertion point within the vector.  We skip over
118   // (a) units from a different section, (b) units from the same section
119   // but with lower offset-within-section.  This keeps units in order
120   // within a section, although not necessarily within the object file,
121   // even if we do lazy parsing.
122   auto I = this->begin();
123   uint64_t Offset = 0;
124   while (Data.isValidOffset(Offset)) {
125     if (I != this->end() &&
126         (&(*I)->getInfoSection() != &Section || (*I)->getOffset() == Offset)) {
127       ++I;
128       continue;
129     }
130     auto U = Parser(Offset, SectionKind, &Section, nullptr);
131     // If parsing failed, we're done with this section.
132     if (!U)
133       break;
134     Offset = U->getNextUnitOffset();
135     I = std::next(this->insert(I, std::move(U)));
136   }
137 }
138 
addUnit(std::unique_ptr<DWARFUnit> Unit)139 DWARFUnit *DWARFUnitVector::addUnit(std::unique_ptr<DWARFUnit> Unit) {
140   auto I = llvm::upper_bound(*this, Unit,
141                              [](const std::unique_ptr<DWARFUnit> &LHS,
142                                 const std::unique_ptr<DWARFUnit> &RHS) {
143                                return LHS->getOffset() < RHS->getOffset();
144                              });
145   return this->insert(I, std::move(Unit))->get();
146 }
147 
getUnitForOffset(uint64_t Offset) const148 DWARFUnit *DWARFUnitVector::getUnitForOffset(uint64_t Offset) const {
149   auto end = begin() + getNumInfoUnits();
150   auto *CU =
151       std::upper_bound(begin(), end, Offset,
152                        [](uint64_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
153                          return LHS < RHS->getNextUnitOffset();
154                        });
155   if (CU != end && (*CU)->getOffset() <= Offset)
156     return CU->get();
157   return nullptr;
158 }
159 
160 DWARFUnit *
getUnitForIndexEntry(const DWARFUnitIndex::Entry & E)161 DWARFUnitVector::getUnitForIndexEntry(const DWARFUnitIndex::Entry &E) {
162   const auto *CUOff = E.getContribution(DW_SECT_INFO);
163   if (!CUOff)
164     return nullptr;
165 
166   uint64_t Offset = CUOff->getOffset();
167   auto end = begin() + getNumInfoUnits();
168 
169   auto *CU =
170       std::upper_bound(begin(), end, CUOff->getOffset(),
171                        [](uint64_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
172                          return LHS < RHS->getNextUnitOffset();
173                        });
174   if (CU != end && (*CU)->getOffset() <= Offset)
175     return CU->get();
176 
177   if (!Parser)
178     return nullptr;
179 
180   auto U = Parser(Offset, DW_SECT_INFO, nullptr, &E);
181   if (!U)
182     return nullptr;
183 
184   auto *NewCU = U.get();
185   this->insert(CU, std::move(U));
186   ++NumInfoUnits;
187   return NewCU;
188 }
189 
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)190 DWARFUnit::DWARFUnit(DWARFContext &DC, const DWARFSection &Section,
191                      const DWARFUnitHeader &Header, const DWARFDebugAbbrev *DA,
192                      const DWARFSection *RS, const DWARFSection *LocSection,
193                      StringRef SS, const DWARFSection &SOS,
194                      const DWARFSection *AOS, const DWARFSection &LS, bool LE,
195                      bool IsDWO, const DWARFUnitVector &UnitVector)
196     : Context(DC), InfoSection(Section), Header(Header), Abbrev(DA),
197       RangeSection(RS), LineSection(LS), StringSection(SS),
198       StringOffsetSection(SOS), AddrOffsetSection(AOS), IsLittleEndian(LE),
199       IsDWO(IsDWO), UnitVector(UnitVector) {
200   clear();
201 }
202 
203 DWARFUnit::~DWARFUnit() = default;
204 
getDebugInfoExtractor() const205 DWARFDataExtractor DWARFUnit::getDebugInfoExtractor() const {
206   return DWARFDataExtractor(Context.getDWARFObj(), InfoSection, IsLittleEndian,
207                             getAddressByteSize());
208 }
209 
210 std::optional<object::SectionedAddress>
getAddrOffsetSectionItem(uint32_t Index) const211 DWARFUnit::getAddrOffsetSectionItem(uint32_t Index) const {
212   if (!AddrOffsetSectionBase) {
213     auto R = Context.info_section_units();
214     // Surprising if a DWO file has more than one skeleton unit in it - this
215     // probably shouldn't be valid, but if a use case is found, here's where to
216     // support it (probably have to linearly search for the matching skeleton CU
217     // here)
218     if (IsDWO && hasSingleElement(R))
219       return (*R.begin())->getAddrOffsetSectionItem(Index);
220 
221     return std::nullopt;
222   }
223 
224   uint64_t Offset = *AddrOffsetSectionBase + Index * getAddressByteSize();
225   if (AddrOffsetSection->Data.size() < Offset + getAddressByteSize())
226     return std::nullopt;
227   DWARFDataExtractor DA(Context.getDWARFObj(), *AddrOffsetSection,
228                         IsLittleEndian, getAddressByteSize());
229   uint64_t Section;
230   uint64_t Address = DA.getRelocatedAddress(&Offset, &Section);
231   return {{Address, Section}};
232 }
233 
getStringOffsetSectionItem(uint32_t Index) const234 Expected<uint64_t> DWARFUnit::getStringOffsetSectionItem(uint32_t Index) const {
235   if (!StringOffsetsTableContribution)
236     return make_error<StringError>(
237         "DW_FORM_strx used without a valid string offsets table",
238         inconvertibleErrorCode());
239   unsigned ItemSize = getDwarfStringOffsetsByteSize();
240   uint64_t Offset = getStringOffsetsBase() + Index * ItemSize;
241   if (StringOffsetSection.Data.size() < Offset + ItemSize)
242     return make_error<StringError>("DW_FORM_strx uses index " + Twine(Index) +
243                                        ", which is too large",
244                                    inconvertibleErrorCode());
245   DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
246                         IsLittleEndian, 0);
247   return DA.getRelocatedValue(ItemSize, &Offset);
248 }
249 
extract(DWARFContext & Context,const DWARFDataExtractor & debug_info,uint64_t * offset_ptr,DWARFSectionKind SectionKind)250 Error DWARFUnitHeader::extract(DWARFContext &Context,
251                                const DWARFDataExtractor &debug_info,
252                                uint64_t *offset_ptr,
253                                DWARFSectionKind SectionKind) {
254   Offset = *offset_ptr;
255   Error Err = Error::success();
256   IndexEntry = nullptr;
257   std::tie(Length, FormParams.Format) =
258       debug_info.getInitialLength(offset_ptr, &Err);
259   FormParams.Version = debug_info.getU16(offset_ptr, &Err);
260   if (FormParams.Version >= 5) {
261     UnitType = debug_info.getU8(offset_ptr, &Err);
262     FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
263     AbbrOffset = debug_info.getRelocatedValue(
264         FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
265   } else {
266     AbbrOffset = debug_info.getRelocatedValue(
267         FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
268     FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
269     // Fake a unit type based on the section type.  This isn't perfect,
270     // but distinguishing compile and type units is generally enough.
271     if (SectionKind == DW_SECT_EXT_TYPES)
272       UnitType = DW_UT_type;
273     else
274       UnitType = DW_UT_compile;
275   }
276   if (isTypeUnit()) {
277     TypeHash = debug_info.getU64(offset_ptr, &Err);
278     TypeOffset = debug_info.getUnsigned(
279         offset_ptr, FormParams.getDwarfOffsetByteSize(), &Err);
280   } else if (UnitType == DW_UT_split_compile || UnitType == DW_UT_skeleton)
281     DWOId = debug_info.getU64(offset_ptr, &Err);
282 
283   if (Err)
284     return joinErrors(
285         createStringError(
286             errc::invalid_argument,
287             "DWARF unit at 0x%8.8" PRIx64 " cannot be parsed:", Offset),
288         std::move(Err));
289 
290   // Header fields all parsed, capture the size of this unit header.
291   assert(*offset_ptr - Offset <= 255 && "unexpected header size");
292   Size = uint8_t(*offset_ptr - Offset);
293   uint64_t NextCUOffset = Offset + getUnitLengthFieldByteSize() + getLength();
294 
295   if (!debug_info.isValidOffset(getNextUnitOffset() - 1))
296     return 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 
302   if (!DWARFContext::isSupportedVersion(getVersion()))
303     return createStringError(
304         errc::invalid_argument,
305         "DWARF unit at offset 0x%8.8" PRIx64 " "
306         "has unsupported version %" PRIu16 ", supported are 2-%u",
307         Offset, getVersion(), DWARFContext::getMaxSupportedVersion());
308 
309   // Type offset is unit-relative; should be after the header and before
310   // the end of the current unit.
311   if (isTypeUnit() && TypeOffset < Size)
312     return createStringError(errc::invalid_argument,
313                              "DWARF type unit at offset "
314                              "0x%8.8" PRIx64 " "
315                              "has its relocated type_offset 0x%8.8" PRIx64 " "
316                              "pointing inside the header",
317                              Offset, Offset + TypeOffset);
318 
319   if (isTypeUnit() && TypeOffset >= getUnitLengthFieldByteSize() + getLength())
320     return createStringError(
321         errc::invalid_argument,
322         "DWARF type unit from offset 0x%8.8" PRIx64 " incl. "
323         "to offset 0x%8.8" PRIx64 " excl. has its "
324         "relocated type_offset 0x%8.8" PRIx64 " pointing past the unit end",
325         Offset, NextCUOffset, Offset + TypeOffset);
326 
327   if (Error SizeErr = DWARFContext::checkAddressSizeSupported(
328           getAddressByteSize(), errc::invalid_argument,
329           "DWARF unit at offset 0x%8.8" PRIx64, Offset))
330     return SizeErr;
331 
332   // Keep track of the highest DWARF version we encounter across all units.
333   Context.setMaxVersionIfGreater(getVersion());
334   return Error::success();
335 }
336 
applyIndexEntry(const DWARFUnitIndex::Entry * Entry)337 bool DWARFUnitHeader::applyIndexEntry(const DWARFUnitIndex::Entry *Entry) {
338   assert(Entry);
339   assert(!IndexEntry);
340   IndexEntry = Entry;
341   if (AbbrOffset)
342     return false;
343   auto *UnitContrib = IndexEntry->getContribution();
344   if (!UnitContrib ||
345       UnitContrib->getLength() != (getLength() + getUnitLengthFieldByteSize()))
346     return false;
347   auto *AbbrEntry = IndexEntry->getContribution(DW_SECT_ABBREV);
348   if (!AbbrEntry)
349     return false;
350   AbbrOffset = AbbrEntry->getOffset();
351   return true;
352 }
353 
extractRangeList(uint64_t RangeListOffset,DWARFDebugRangeList & RangeList) const354 Error DWARFUnit::extractRangeList(uint64_t RangeListOffset,
355                                   DWARFDebugRangeList &RangeList) const {
356   // Require that compile unit is extracted.
357   assert(!DieArray.empty());
358   DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
359                                 IsLittleEndian, getAddressByteSize());
360   uint64_t ActualRangeListOffset = RangeSectionBase + RangeListOffset;
361   return RangeList.extract(RangesData, &ActualRangeListOffset);
362 }
363 
clear()364 void DWARFUnit::clear() {
365   Abbrevs = nullptr;
366   BaseAddr.reset();
367   RangeSectionBase = 0;
368   LocSectionBase = 0;
369   AddrOffsetSectionBase = std::nullopt;
370   SU = nullptr;
371   clearDIEs(false);
372   AddrDieMap.clear();
373   if (DWO)
374     DWO->clear();
375   DWO.reset();
376 }
377 
getCompilationDir()378 const char *DWARFUnit::getCompilationDir() {
379   return dwarf::toString(getUnitDIE().find(DW_AT_comp_dir), nullptr);
380 }
381 
extractDIEsToVector(bool AppendCUDie,bool AppendNonCUDies,std::vector<DWARFDebugInfoEntry> & Dies) const382 void DWARFUnit::extractDIEsToVector(
383     bool AppendCUDie, bool AppendNonCUDies,
384     std::vector<DWARFDebugInfoEntry> &Dies) const {
385   if (!AppendCUDie && !AppendNonCUDies)
386     return;
387 
388   // Set the offset to that of the first DIE and calculate the start of the
389   // next compilation unit header.
390   uint64_t DIEOffset = getOffset() + getHeaderSize();
391   uint64_t NextCUOffset = getNextUnitOffset();
392   DWARFDebugInfoEntry DIE;
393   DWARFDataExtractor DebugInfoData = getDebugInfoExtractor();
394   // The end offset has been already checked by DWARFUnitHeader::extract.
395   assert(DebugInfoData.isValidOffset(NextCUOffset - 1));
396   std::vector<uint32_t> Parents;
397   std::vector<uint32_t> PrevSiblings;
398   bool IsCUDie = true;
399 
400   assert(
401       ((AppendCUDie && Dies.empty()) || (!AppendCUDie && Dies.size() == 1)) &&
402       "Dies array is not empty");
403 
404   // Fill Parents and Siblings stacks with initial value.
405   Parents.push_back(UINT32_MAX);
406   if (!AppendCUDie)
407     Parents.push_back(0);
408   PrevSiblings.push_back(0);
409 
410   // Start to extract dies.
411   do {
412     assert(Parents.size() > 0 && "Empty parents stack");
413     assert((Parents.back() == UINT32_MAX || Parents.back() <= Dies.size()) &&
414            "Wrong parent index");
415 
416     // Extract die. Stop if any error occurred.
417     if (!DIE.extractFast(*this, &DIEOffset, DebugInfoData, NextCUOffset,
418                          Parents.back()))
419       break;
420 
421     // If previous sibling is remembered then update it`s SiblingIdx field.
422     if (PrevSiblings.back() > 0) {
423       assert(PrevSiblings.back() < Dies.size() &&
424              "Previous sibling index is out of Dies boundaries");
425       Dies[PrevSiblings.back()].setSiblingIdx(Dies.size());
426     }
427 
428     // Store die into the Dies vector.
429     if (IsCUDie) {
430       if (AppendCUDie)
431         Dies.push_back(DIE);
432       if (!AppendNonCUDies)
433         break;
434       // The average bytes per DIE entry has been seen to be
435       // around 14-20 so let's pre-reserve the needed memory for
436       // our DIE entries accordingly.
437       Dies.reserve(Dies.size() + getDebugInfoSize() / 14);
438     } else {
439       // Remember last previous sibling.
440       PrevSiblings.back() = Dies.size();
441 
442       Dies.push_back(DIE);
443     }
444 
445     // Check for new children scope.
446     if (const DWARFAbbreviationDeclaration *AbbrDecl =
447             DIE.getAbbreviationDeclarationPtr()) {
448       if (AbbrDecl->hasChildren()) {
449         if (AppendCUDie || !IsCUDie) {
450           assert(Dies.size() > 0 && "Dies does not contain any die");
451           Parents.push_back(Dies.size() - 1);
452           PrevSiblings.push_back(0);
453         }
454       } else if (IsCUDie)
455         // Stop if we have single compile unit die w/o children.
456         break;
457     } else {
458       // NULL DIE: finishes current children scope.
459       Parents.pop_back();
460       PrevSiblings.pop_back();
461     }
462 
463     if (IsCUDie)
464       IsCUDie = false;
465 
466     // Stop when compile unit die is removed from the parents stack.
467   } while (Parents.size() > 1);
468 }
469 
extractDIEsIfNeeded(bool CUDieOnly)470 void DWARFUnit::extractDIEsIfNeeded(bool CUDieOnly) {
471   if (Error e = tryExtractDIEsIfNeeded(CUDieOnly))
472     Context.getRecoverableErrorHandler()(std::move(e));
473 }
474 
tryExtractDIEsIfNeeded(bool CUDieOnly)475 Error DWARFUnit::tryExtractDIEsIfNeeded(bool CUDieOnly) {
476   if ((CUDieOnly && !DieArray.empty()) ||
477       DieArray.size() > 1)
478     return Error::success(); // Already parsed.
479 
480   bool HasCUDie = !DieArray.empty();
481   extractDIEsToVector(!HasCUDie, !CUDieOnly, DieArray);
482 
483   if (DieArray.empty())
484     return Error::success();
485 
486   // If CU DIE was just parsed, copy several attribute values from it.
487   if (HasCUDie)
488     return Error::success();
489 
490   DWARFDie UnitDie(this, &DieArray[0]);
491   if (std::optional<uint64_t> DWOId =
492           toUnsigned(UnitDie.find(DW_AT_GNU_dwo_id)))
493     Header.setDWOId(*DWOId);
494   if (!IsDWO) {
495     assert(AddrOffsetSectionBase == std::nullopt);
496     assert(RangeSectionBase == 0);
497     assert(LocSectionBase == 0);
498     AddrOffsetSectionBase = toSectionOffset(UnitDie.find(DW_AT_addr_base));
499     if (!AddrOffsetSectionBase)
500       AddrOffsetSectionBase =
501           toSectionOffset(UnitDie.find(DW_AT_GNU_addr_base));
502     RangeSectionBase = toSectionOffset(UnitDie.find(DW_AT_rnglists_base), 0);
503     LocSectionBase = toSectionOffset(UnitDie.find(DW_AT_loclists_base), 0);
504   }
505 
506   // In general, in DWARF v5 and beyond we derive the start of the unit's
507   // contribution to the string offsets table from the unit DIE's
508   // DW_AT_str_offsets_base attribute. Split DWARF units do not use this
509   // attribute, so we assume that there is a contribution to the string
510   // offsets table starting at offset 0 of the debug_str_offsets.dwo section.
511   // In both cases we need to determine the format of the contribution,
512   // which may differ from the unit's format.
513   DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
514                         IsLittleEndian, 0);
515   if (IsDWO || getVersion() >= 5) {
516     auto StringOffsetOrError =
517         IsDWO ? determineStringOffsetsTableContributionDWO(DA)
518               : determineStringOffsetsTableContribution(DA);
519     if (!StringOffsetOrError)
520       return createStringError(errc::invalid_argument,
521                                "invalid reference to or invalid content in "
522                                ".debug_str_offsets[.dwo]: " +
523                                    toString(StringOffsetOrError.takeError()));
524 
525     StringOffsetsTableContribution = *StringOffsetOrError;
526   }
527 
528   // DWARF v5 uses the .debug_rnglists and .debug_rnglists.dwo sections to
529   // describe address ranges.
530   if (getVersion() >= 5) {
531     // In case of DWP, the base offset from the index has to be added.
532     if (IsDWO) {
533       uint64_t ContributionBaseOffset = 0;
534       if (auto *IndexEntry = Header.getIndexEntry())
535         if (auto *Contrib = IndexEntry->getContribution(DW_SECT_RNGLISTS))
536           ContributionBaseOffset = Contrib->getOffset();
537       setRangesSection(
538           &Context.getDWARFObj().getRnglistsDWOSection(),
539           ContributionBaseOffset +
540               DWARFListTableHeader::getHeaderSize(Header.getFormat()));
541     } else
542       setRangesSection(&Context.getDWARFObj().getRnglistsSection(),
543                        toSectionOffset(UnitDie.find(DW_AT_rnglists_base),
544                                        DWARFListTableHeader::getHeaderSize(
545                                            Header.getFormat())));
546   }
547 
548   if (IsDWO) {
549     // If we are reading a package file, we need to adjust the location list
550     // data based on the index entries.
551     StringRef Data = Header.getVersion() >= 5
552                          ? Context.getDWARFObj().getLoclistsDWOSection().Data
553                          : Context.getDWARFObj().getLocDWOSection().Data;
554     if (auto *IndexEntry = Header.getIndexEntry())
555       if (const auto *C = IndexEntry->getContribution(
556               Header.getVersion() >= 5 ? DW_SECT_LOCLISTS : DW_SECT_EXT_LOC))
557         Data = Data.substr(C->getOffset(), C->getLength());
558 
559     DWARFDataExtractor DWARFData(Data, IsLittleEndian, getAddressByteSize());
560     LocTable =
561         std::make_unique<DWARFDebugLoclists>(DWARFData, Header.getVersion());
562     LocSectionBase = DWARFListTableHeader::getHeaderSize(Header.getFormat());
563   } else if (getVersion() >= 5) {
564     LocTable = std::make_unique<DWARFDebugLoclists>(
565         DWARFDataExtractor(Context.getDWARFObj(),
566                            Context.getDWARFObj().getLoclistsSection(),
567                            IsLittleEndian, getAddressByteSize()),
568         getVersion());
569   } else {
570     LocTable = std::make_unique<DWARFDebugLoc>(DWARFDataExtractor(
571         Context.getDWARFObj(), Context.getDWARFObj().getLocSection(),
572         IsLittleEndian, getAddressByteSize()));
573   }
574 
575   // Don't fall back to DW_AT_GNU_ranges_base: it should be ignored for
576   // skeleton CU DIE, so that DWARF users not aware of it are not broken.
577   return Error::success();
578 }
579 
parseDWO(StringRef DWOAlternativeLocation)580 bool DWARFUnit::parseDWO(StringRef DWOAlternativeLocation) {
581   if (IsDWO)
582     return false;
583   if (DWO)
584     return false;
585   DWARFDie UnitDie = getUnitDIE();
586   if (!UnitDie)
587     return false;
588   auto DWOFileName = getVersion() >= 5
589                          ? dwarf::toString(UnitDie.find(DW_AT_dwo_name))
590                          : dwarf::toString(UnitDie.find(DW_AT_GNU_dwo_name));
591   if (!DWOFileName)
592     return false;
593   auto CompilationDir = dwarf::toString(UnitDie.find(DW_AT_comp_dir));
594   SmallString<16> AbsolutePath;
595   if (sys::path::is_relative(*DWOFileName) && CompilationDir &&
596       *CompilationDir) {
597     sys::path::append(AbsolutePath, *CompilationDir);
598   }
599   sys::path::append(AbsolutePath, *DWOFileName);
600   auto DWOId = getDWOId();
601   if (!DWOId)
602     return false;
603   auto DWOContext = Context.getDWOContext(AbsolutePath);
604   if (!DWOContext) {
605     // Use the alternative location to get the DWARF context for the DWO object.
606     if (DWOAlternativeLocation.empty())
607       return false;
608     // If the alternative context does not correspond to the original DWO object
609     // (different hashes), the below 'getDWOCompileUnitForHash' call will catch
610     // the issue, with a returned null context.
611     DWOContext = Context.getDWOContext(DWOAlternativeLocation);
612     if (!DWOContext)
613       return false;
614   }
615 
616   DWARFCompileUnit *DWOCU = DWOContext->getDWOCompileUnitForHash(*DWOId);
617   if (!DWOCU)
618     return false;
619   DWO = std::shared_ptr<DWARFCompileUnit>(std::move(DWOContext), DWOCU);
620   DWO->setSkeletonUnit(this);
621   // Share .debug_addr and .debug_ranges section with compile unit in .dwo
622   if (AddrOffsetSectionBase)
623     DWO->setAddrOffsetSection(AddrOffsetSection, *AddrOffsetSectionBase);
624   if (getVersion() == 4) {
625     auto DWORangesBase = UnitDie.getRangesBaseAttribute();
626     DWO->setRangesSection(RangeSection, DWORangesBase.value_or(0));
627   }
628 
629   return true;
630 }
631 
clearDIEs(bool KeepCUDie)632 void DWARFUnit::clearDIEs(bool KeepCUDie) {
633   // Do not use resize() + shrink_to_fit() to free memory occupied by dies.
634   // shrink_to_fit() is a *non-binding* request to reduce capacity() to size().
635   // It depends on the implementation whether the request is fulfilled.
636   // Create a new vector with a small capacity and assign it to the DieArray to
637   // have previous contents freed.
638   DieArray = (KeepCUDie && !DieArray.empty())
639                  ? std::vector<DWARFDebugInfoEntry>({DieArray[0]})
640                  : std::vector<DWARFDebugInfoEntry>();
641 }
642 
643 Expected<DWARFAddressRangesVector>
findRnglistFromOffset(uint64_t Offset)644 DWARFUnit::findRnglistFromOffset(uint64_t Offset) {
645   if (getVersion() <= 4) {
646     DWARFDebugRangeList RangeList;
647     if (Error E = extractRangeList(Offset, RangeList))
648       return std::move(E);
649     return RangeList.getAbsoluteRanges(getBaseAddress());
650   }
651   DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
652                                 IsLittleEndian, Header.getAddressByteSize());
653   DWARFDebugRnglistTable RnglistTable;
654   auto RangeListOrError = RnglistTable.findList(RangesData, Offset);
655   if (RangeListOrError)
656     return RangeListOrError.get().getAbsoluteRanges(getBaseAddress(), *this);
657   return RangeListOrError.takeError();
658 }
659 
660 Expected<DWARFAddressRangesVector>
findRnglistFromIndex(uint32_t Index)661 DWARFUnit::findRnglistFromIndex(uint32_t Index) {
662   if (auto Offset = getRnglistOffset(Index))
663     return findRnglistFromOffset(*Offset);
664 
665   return createStringError(errc::invalid_argument,
666                            "invalid range list table index %d (possibly "
667                            "missing the entire range list table)",
668                            Index);
669 }
670 
collectAddressRanges()671 Expected<DWARFAddressRangesVector> DWARFUnit::collectAddressRanges() {
672   DWARFDie UnitDie = getUnitDIE();
673   if (!UnitDie)
674     return createStringError(errc::invalid_argument, "No unit DIE");
675 
676   // First, check if unit DIE describes address ranges for the whole unit.
677   auto CUDIERangesOrError = UnitDie.getAddressRanges();
678   if (!CUDIERangesOrError)
679     return createStringError(errc::invalid_argument,
680                              "decoding address ranges: %s",
681                              toString(CUDIERangesOrError.takeError()).c_str());
682   return *CUDIERangesOrError;
683 }
684 
685 Expected<DWARFLocationExpressionsVector>
findLoclistFromOffset(uint64_t Offset)686 DWARFUnit::findLoclistFromOffset(uint64_t Offset) {
687   DWARFLocationExpressionsVector Result;
688 
689   Error InterpretationError = Error::success();
690 
691   Error ParseError = getLocationTable().visitAbsoluteLocationList(
692       Offset, getBaseAddress(),
693       [this](uint32_t Index) { return getAddrOffsetSectionItem(Index); },
694       [&](Expected<DWARFLocationExpression> L) {
695         if (L)
696           Result.push_back(std::move(*L));
697         else
698           InterpretationError =
699               joinErrors(L.takeError(), std::move(InterpretationError));
700         return !InterpretationError;
701       });
702 
703   if (ParseError || InterpretationError)
704     return joinErrors(std::move(ParseError), std::move(InterpretationError));
705 
706   return Result;
707 }
708 
updateAddressDieMap(DWARFDie Die)709 void DWARFUnit::updateAddressDieMap(DWARFDie Die) {
710   if (Die.isSubroutineDIE()) {
711     auto DIERangesOrError = Die.getAddressRanges();
712     if (DIERangesOrError) {
713       for (const auto &R : DIERangesOrError.get()) {
714         // Ignore 0-sized ranges.
715         if (R.LowPC == R.HighPC)
716           continue;
717         auto B = AddrDieMap.upper_bound(R.LowPC);
718         if (B != AddrDieMap.begin() && R.LowPC < (--B)->second.first) {
719           // The range is a sub-range of existing ranges, we need to split the
720           // existing range.
721           if (R.HighPC < B->second.first)
722             AddrDieMap[R.HighPC] = B->second;
723           if (R.LowPC > B->first)
724             AddrDieMap[B->first].first = R.LowPC;
725         }
726         AddrDieMap[R.LowPC] = std::make_pair(R.HighPC, Die);
727       }
728     } else
729       llvm::consumeError(DIERangesOrError.takeError());
730   }
731   // Parent DIEs are added to the AddrDieMap prior to the Children DIEs to
732   // simplify the logic to update AddrDieMap. The child's range will always
733   // be equal or smaller than the parent's range. With this assumption, when
734   // adding one range into the map, it will at most split a range into 3
735   // sub-ranges.
736   for (DWARFDie Child = Die.getFirstChild(); Child; Child = Child.getSibling())
737     updateAddressDieMap(Child);
738 }
739 
getSubroutineForAddress(uint64_t Address)740 DWARFDie DWARFUnit::getSubroutineForAddress(uint64_t Address) {
741   extractDIEsIfNeeded(false);
742   if (AddrDieMap.empty())
743     updateAddressDieMap(getUnitDIE());
744   auto R = AddrDieMap.upper_bound(Address);
745   if (R == AddrDieMap.begin())
746     return DWARFDie();
747   // upper_bound's previous item contains Address.
748   --R;
749   if (Address >= R->second.first)
750     return DWARFDie();
751   return R->second.second;
752 }
753 
updateVariableDieMap(DWARFDie Die)754 void DWARFUnit::updateVariableDieMap(DWARFDie Die) {
755   for (DWARFDie Child : Die) {
756     if (isType(Child.getTag()))
757       continue;
758     updateVariableDieMap(Child);
759   }
760 
761   if (Die.getTag() != DW_TAG_variable)
762     return;
763 
764   Expected<DWARFLocationExpressionsVector> Locations =
765       Die.getLocations(DW_AT_location);
766   if (!Locations) {
767     // Missing DW_AT_location is fine here.
768     consumeError(Locations.takeError());
769     return;
770   }
771 
772   uint64_t Address = UINT64_MAX;
773 
774   for (const DWARFLocationExpression &Location : *Locations) {
775     uint8_t AddressSize = getAddressByteSize();
776     DataExtractor Data(Location.Expr, isLittleEndian(), AddressSize);
777     DWARFExpression Expr(Data, AddressSize);
778     auto It = Expr.begin();
779     if (It == Expr.end())
780       continue;
781 
782     // Match exactly the main sequence used to describe global variables:
783     // `DW_OP_addr[x] [+ DW_OP_plus_uconst]`. Currently, this is the sequence
784     // that LLVM produces for DILocalVariables and DIGlobalVariables. If, in
785     // future, the DWARF producer (`DwarfCompileUnit::addLocationAttribute()` is
786     // a good starting point) is extended to use further expressions, this code
787     // needs to be updated.
788     uint64_t LocationAddr;
789     if (It->getCode() == dwarf::DW_OP_addr) {
790       LocationAddr = It->getRawOperand(0);
791     } else if (It->getCode() == dwarf::DW_OP_addrx) {
792       uint64_t DebugAddrOffset = It->getRawOperand(0);
793       if (auto Pointer = getAddrOffsetSectionItem(DebugAddrOffset)) {
794         LocationAddr = Pointer->Address;
795       }
796     } else {
797       continue;
798     }
799 
800     // Read the optional 2nd operand, a DW_OP_plus_uconst.
801     if (++It != Expr.end()) {
802       if (It->getCode() != dwarf::DW_OP_plus_uconst)
803         continue;
804 
805       LocationAddr += It->getRawOperand(0);
806 
807       // Probe for a 3rd operand, if it exists, bail.
808       if (++It != Expr.end())
809         continue;
810     }
811 
812     Address = LocationAddr;
813     break;
814   }
815 
816   // Get the size of the global variable. If all else fails (i.e. the global has
817   // no type), then we use a size of one to still allow symbolization of the
818   // exact address.
819   uint64_t GVSize = 1;
820   if (Die.getAttributeValueAsReferencedDie(DW_AT_type))
821     if (std::optional<uint64_t> Size = Die.getTypeSize(getAddressByteSize()))
822       GVSize = *Size;
823 
824   if (Address != UINT64_MAX)
825     VariableDieMap[Address] = {Address + GVSize, Die};
826 }
827 
getVariableForAddress(uint64_t Address)828 DWARFDie DWARFUnit::getVariableForAddress(uint64_t Address) {
829   extractDIEsIfNeeded(false);
830 
831   auto RootDie = getUnitDIE();
832 
833   auto RootLookup = RootsParsedForVariables.insert(RootDie.getOffset());
834   if (RootLookup.second)
835     updateVariableDieMap(RootDie);
836 
837   auto R = VariableDieMap.upper_bound(Address);
838   if (R == VariableDieMap.begin())
839     return DWARFDie();
840 
841   // upper_bound's previous item contains Address.
842   --R;
843   if (Address >= R->second.first)
844     return DWARFDie();
845   return R->second.second;
846 }
847 
848 void
getInlinedChainForAddress(uint64_t Address,SmallVectorImpl<DWARFDie> & InlinedChain)849 DWARFUnit::getInlinedChainForAddress(uint64_t Address,
850                                      SmallVectorImpl<DWARFDie> &InlinedChain) {
851   assert(InlinedChain.empty());
852   // Try to look for subprogram DIEs in the DWO file.
853   parseDWO();
854   // First, find the subroutine that contains the given address (the leaf
855   // of inlined chain).
856   DWARFDie SubroutineDIE =
857       (DWO ? *DWO : *this).getSubroutineForAddress(Address);
858 
859   while (SubroutineDIE) {
860     if (SubroutineDIE.isSubprogramDIE()) {
861       InlinedChain.push_back(SubroutineDIE);
862       return;
863     }
864     if (SubroutineDIE.getTag() == DW_TAG_inlined_subroutine)
865       InlinedChain.push_back(SubroutineDIE);
866     SubroutineDIE  = SubroutineDIE.getParent();
867   }
868 }
869 
getDWARFUnitIndex(DWARFContext & Context,DWARFSectionKind Kind)870 const DWARFUnitIndex &llvm::getDWARFUnitIndex(DWARFContext &Context,
871                                               DWARFSectionKind Kind) {
872   if (Kind == DW_SECT_INFO)
873     return Context.getCUIndex();
874   assert(Kind == DW_SECT_EXT_TYPES);
875   return Context.getTUIndex();
876 }
877 
getParent(const DWARFDebugInfoEntry * Die)878 DWARFDie DWARFUnit::getParent(const DWARFDebugInfoEntry *Die) {
879   if (const DWARFDebugInfoEntry *Entry = getParentEntry(Die))
880     return DWARFDie(this, Entry);
881 
882   return DWARFDie();
883 }
884 
885 const DWARFDebugInfoEntry *
getParentEntry(const DWARFDebugInfoEntry * Die) const886 DWARFUnit::getParentEntry(const DWARFDebugInfoEntry *Die) const {
887   if (!Die)
888     return nullptr;
889   assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
890 
891   if (std::optional<uint32_t> ParentIdx = Die->getParentIdx()) {
892     assert(*ParentIdx < DieArray.size() &&
893            "ParentIdx is out of DieArray boundaries");
894     return getDebugInfoEntry(*ParentIdx);
895   }
896 
897   return nullptr;
898 }
899 
getSibling(const DWARFDebugInfoEntry * Die)900 DWARFDie DWARFUnit::getSibling(const DWARFDebugInfoEntry *Die) {
901   if (const DWARFDebugInfoEntry *Sibling = getSiblingEntry(Die))
902     return DWARFDie(this, Sibling);
903 
904   return DWARFDie();
905 }
906 
907 const DWARFDebugInfoEntry *
getSiblingEntry(const DWARFDebugInfoEntry * Die) const908 DWARFUnit::getSiblingEntry(const DWARFDebugInfoEntry *Die) const {
909   if (!Die)
910     return nullptr;
911   assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
912 
913   if (std::optional<uint32_t> SiblingIdx = Die->getSiblingIdx()) {
914     assert(*SiblingIdx < DieArray.size() &&
915            "SiblingIdx is out of DieArray boundaries");
916     return &DieArray[*SiblingIdx];
917   }
918 
919   return nullptr;
920 }
921 
getPreviousSibling(const DWARFDebugInfoEntry * Die)922 DWARFDie DWARFUnit::getPreviousSibling(const DWARFDebugInfoEntry *Die) {
923   if (const DWARFDebugInfoEntry *Sibling = getPreviousSiblingEntry(Die))
924     return DWARFDie(this, Sibling);
925 
926   return DWARFDie();
927 }
928 
929 const DWARFDebugInfoEntry *
getPreviousSiblingEntry(const DWARFDebugInfoEntry * Die) const930 DWARFUnit::getPreviousSiblingEntry(const DWARFDebugInfoEntry *Die) const {
931   if (!Die)
932     return nullptr;
933   assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
934 
935   std::optional<uint32_t> ParentIdx = Die->getParentIdx();
936   if (!ParentIdx)
937     // Die is a root die, there is no previous sibling.
938     return nullptr;
939 
940   assert(*ParentIdx < DieArray.size() &&
941          "ParentIdx is out of DieArray boundaries");
942   assert(getDIEIndex(Die) > 0 && "Die is a root die");
943 
944   uint32_t PrevDieIdx = getDIEIndex(Die) - 1;
945   if (PrevDieIdx == *ParentIdx)
946     // Immediately previous node is parent, there is no previous sibling.
947     return nullptr;
948 
949   while (DieArray[PrevDieIdx].getParentIdx() != *ParentIdx) {
950     PrevDieIdx = *DieArray[PrevDieIdx].getParentIdx();
951 
952     assert(PrevDieIdx < DieArray.size() &&
953            "PrevDieIdx is out of DieArray boundaries");
954     assert(PrevDieIdx >= *ParentIdx &&
955            "PrevDieIdx is not a child of parent of Die");
956   }
957 
958   return &DieArray[PrevDieIdx];
959 }
960 
getFirstChild(const DWARFDebugInfoEntry * Die)961 DWARFDie DWARFUnit::getFirstChild(const DWARFDebugInfoEntry *Die) {
962   if (const DWARFDebugInfoEntry *Child = getFirstChildEntry(Die))
963     return DWARFDie(this, Child);
964 
965   return DWARFDie();
966 }
967 
968 const DWARFDebugInfoEntry *
getFirstChildEntry(const DWARFDebugInfoEntry * Die) const969 DWARFUnit::getFirstChildEntry(const DWARFDebugInfoEntry *Die) const {
970   if (!Die)
971     return nullptr;
972   assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
973 
974   if (!Die->hasChildren())
975     return nullptr;
976 
977   // TODO: Instead of checking here for invalid die we might reject
978   // invalid dies at parsing stage(DWARFUnit::extractDIEsToVector).
979   // We do not want access out of bounds when parsing corrupted debug data.
980   size_t I = getDIEIndex(Die) + 1;
981   if (I >= DieArray.size())
982     return nullptr;
983   return &DieArray[I];
984 }
985 
getLastChild(const DWARFDebugInfoEntry * Die)986 DWARFDie DWARFUnit::getLastChild(const DWARFDebugInfoEntry *Die) {
987   if (const DWARFDebugInfoEntry *Child = getLastChildEntry(Die))
988     return DWARFDie(this, Child);
989 
990   return DWARFDie();
991 }
992 
993 const DWARFDebugInfoEntry *
getLastChildEntry(const DWARFDebugInfoEntry * Die) const994 DWARFUnit::getLastChildEntry(const DWARFDebugInfoEntry *Die) const {
995   if (!Die)
996     return nullptr;
997   assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
998 
999   if (!Die->hasChildren())
1000     return nullptr;
1001 
1002   if (std::optional<uint32_t> SiblingIdx = Die->getSiblingIdx()) {
1003     assert(*SiblingIdx < DieArray.size() &&
1004            "SiblingIdx is out of DieArray boundaries");
1005     assert(DieArray[*SiblingIdx - 1].getTag() == dwarf::DW_TAG_null &&
1006            "Bad end of children marker");
1007     return &DieArray[*SiblingIdx - 1];
1008   }
1009 
1010   // If SiblingIdx is set for non-root dies we could be sure that DWARF is
1011   // correct and "end of children marker" must be found. For root die we do not
1012   // have such a guarantee(parsing root die might be stopped if "end of children
1013   // marker" is missing, SiblingIdx is always zero for root die). That is why we
1014   // do not use assertion for checking for "end of children marker" for root
1015   // die.
1016 
1017   // TODO: Instead of checking here for invalid die we might reject
1018   // invalid dies at parsing stage(DWARFUnit::extractDIEsToVector).
1019   if (getDIEIndex(Die) == 0 && DieArray.size() > 1 &&
1020       DieArray.back().getTag() == dwarf::DW_TAG_null) {
1021     // For the unit die we might take last item from DieArray.
1022     assert(getDIEIndex(Die) ==
1023                getDIEIndex(const_cast<DWARFUnit *>(this)->getUnitDIE()) &&
1024            "Bad unit die");
1025     return &DieArray.back();
1026   }
1027 
1028   return nullptr;
1029 }
1030 
getAbbreviations() const1031 const DWARFAbbreviationDeclarationSet *DWARFUnit::getAbbreviations() const {
1032   if (!Abbrevs) {
1033     Expected<const DWARFAbbreviationDeclarationSet *> AbbrevsOrError =
1034         Abbrev->getAbbreviationDeclarationSet(getAbbreviationsOffset());
1035     if (!AbbrevsOrError) {
1036       // FIXME: We should propagate this error upwards.
1037       consumeError(AbbrevsOrError.takeError());
1038       return nullptr;
1039     }
1040     Abbrevs = *AbbrevsOrError;
1041   }
1042   return Abbrevs;
1043 }
1044 
getBaseAddress()1045 std::optional<object::SectionedAddress> DWARFUnit::getBaseAddress() {
1046   if (BaseAddr)
1047     return BaseAddr;
1048 
1049   DWARFDie UnitDie = (SU ? SU : this)->getUnitDIE();
1050   std::optional<DWARFFormValue> PC =
1051       UnitDie.find({DW_AT_low_pc, DW_AT_entry_pc});
1052   BaseAddr = toSectionedAddress(PC);
1053   return BaseAddr;
1054 }
1055 
1056 Expected<StrOffsetsContributionDescriptor>
validateContributionSize(DWARFDataExtractor & DA)1057 StrOffsetsContributionDescriptor::validateContributionSize(
1058     DWARFDataExtractor &DA) {
1059   uint8_t EntrySize = getDwarfOffsetByteSize();
1060   // In order to ensure that we don't read a partial record at the end of
1061   // the section we validate for a multiple of the entry size.
1062   uint64_t ValidationSize = alignTo(Size, EntrySize);
1063   // Guard against overflow.
1064   if (ValidationSize >= Size)
1065     if (DA.isValidOffsetForDataOfSize((uint32_t)Base, ValidationSize))
1066       return *this;
1067   return createStringError(errc::invalid_argument, "length exceeds section size");
1068 }
1069 
1070 // Look for a DWARF64-formatted contribution to the string offsets table
1071 // starting at a given offset and record it in a descriptor.
1072 static Expected<StrOffsetsContributionDescriptor>
parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)1073 parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
1074   if (!DA.isValidOffsetForDataOfSize(Offset, 16))
1075     return createStringError(errc::invalid_argument, "section offset exceeds section size");
1076 
1077   if (DA.getU32(&Offset) != dwarf::DW_LENGTH_DWARF64)
1078     return createStringError(errc::invalid_argument, "32 bit contribution referenced from a 64 bit unit");
1079 
1080   uint64_t Size = DA.getU64(&Offset);
1081   uint8_t Version = DA.getU16(&Offset);
1082   (void)DA.getU16(&Offset); // padding
1083   // The encoded length includes the 2-byte version field and the 2-byte
1084   // padding, so we need to subtract them out when we populate the descriptor.
1085   return StrOffsetsContributionDescriptor(Offset, Size - 4, Version, DWARF64);
1086 }
1087 
1088 // Look for a DWARF32-formatted contribution to the string offsets table
1089 // starting at a given offset and record it in a descriptor.
1090 static Expected<StrOffsetsContributionDescriptor>
parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)1091 parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
1092   if (!DA.isValidOffsetForDataOfSize(Offset, 8))
1093     return createStringError(errc::invalid_argument, "section offset exceeds section size");
1094 
1095   uint32_t ContributionSize = DA.getU32(&Offset);
1096   if (ContributionSize >= dwarf::DW_LENGTH_lo_reserved)
1097     return createStringError(errc::invalid_argument, "invalid length");
1098 
1099   uint8_t Version = DA.getU16(&Offset);
1100   (void)DA.getU16(&Offset); // padding
1101   // The encoded length includes the 2-byte version field and the 2-byte
1102   // padding, so we need to subtract them out when we populate the descriptor.
1103   return StrOffsetsContributionDescriptor(Offset, ContributionSize - 4, Version,
1104                                           DWARF32);
1105 }
1106 
1107 static Expected<StrOffsetsContributionDescriptor>
parseDWARFStringOffsetsTableHeader(DWARFDataExtractor & DA,llvm::dwarf::DwarfFormat Format,uint64_t Offset)1108 parseDWARFStringOffsetsTableHeader(DWARFDataExtractor &DA,
1109                                    llvm::dwarf::DwarfFormat Format,
1110                                    uint64_t Offset) {
1111   StrOffsetsContributionDescriptor Desc;
1112   switch (Format) {
1113   case dwarf::DwarfFormat::DWARF64: {
1114     if (Offset < 16)
1115       return createStringError(errc::invalid_argument, "insufficient space for 64 bit header prefix");
1116     auto DescOrError = parseDWARF64StringOffsetsTableHeader(DA, Offset - 16);
1117     if (!DescOrError)
1118       return DescOrError.takeError();
1119     Desc = *DescOrError;
1120     break;
1121   }
1122   case dwarf::DwarfFormat::DWARF32: {
1123     if (Offset < 8)
1124       return createStringError(errc::invalid_argument, "insufficient space for 32 bit header prefix");
1125     auto DescOrError = parseDWARF32StringOffsetsTableHeader(DA, Offset - 8);
1126     if (!DescOrError)
1127       return DescOrError.takeError();
1128     Desc = *DescOrError;
1129     break;
1130   }
1131   }
1132   return Desc.validateContributionSize(DA);
1133 }
1134 
1135 Expected<std::optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContribution(DWARFDataExtractor & DA)1136 DWARFUnit::determineStringOffsetsTableContribution(DWARFDataExtractor &DA) {
1137   assert(!IsDWO);
1138   auto OptOffset = toSectionOffset(getUnitDIE().find(DW_AT_str_offsets_base));
1139   if (!OptOffset)
1140     return std::nullopt;
1141   auto DescOrError =
1142       parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), *OptOffset);
1143   if (!DescOrError)
1144     return DescOrError.takeError();
1145   return *DescOrError;
1146 }
1147 
1148 Expected<std::optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA)1149 DWARFUnit::determineStringOffsetsTableContributionDWO(DWARFDataExtractor &DA) {
1150   assert(IsDWO);
1151   uint64_t Offset = 0;
1152   auto IndexEntry = Header.getIndexEntry();
1153   const auto *C =
1154       IndexEntry ? IndexEntry->getContribution(DW_SECT_STR_OFFSETS) : nullptr;
1155   if (C)
1156     Offset = C->getOffset();
1157   if (getVersion() >= 5) {
1158     if (DA.getData().data() == nullptr)
1159       return std::nullopt;
1160     Offset += Header.getFormat() == dwarf::DwarfFormat::DWARF32 ? 8 : 16;
1161     // Look for a valid contribution at the given offset.
1162     auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), Offset);
1163     if (!DescOrError)
1164       return DescOrError.takeError();
1165     return *DescOrError;
1166   }
1167   // Prior to DWARF v5, we derive the contribution size from the
1168   // index table (in a package file). In a .dwo file it is simply
1169   // the length of the string offsets section.
1170   StrOffsetsContributionDescriptor Desc;
1171   if (C)
1172     Desc = StrOffsetsContributionDescriptor(C->getOffset(), C->getLength(), 4,
1173                                             Header.getFormat());
1174   else if (!IndexEntry && !StringOffsetSection.Data.empty())
1175     Desc = StrOffsetsContributionDescriptor(0, StringOffsetSection.Data.size(),
1176                                             4, Header.getFormat());
1177   else
1178     return std::nullopt;
1179   auto DescOrError = Desc.validateContributionSize(DA);
1180   if (!DescOrError)
1181     return DescOrError.takeError();
1182   return *DescOrError;
1183 }
1184 
getRnglistOffset(uint32_t Index)1185 std::optional<uint64_t> DWARFUnit::getRnglistOffset(uint32_t Index) {
1186   DataExtractor RangesData(RangeSection->Data, IsLittleEndian,
1187                            getAddressByteSize());
1188   DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
1189                               IsLittleEndian, 0);
1190   if (std::optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
1191           RangesData, RangeSectionBase, getFormat(), Index))
1192     return *Off + RangeSectionBase;
1193   return std::nullopt;
1194 }
1195 
getLoclistOffset(uint32_t Index)1196 std::optional<uint64_t> DWARFUnit::getLoclistOffset(uint32_t Index) {
1197   if (std::optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
1198           LocTable->getData(), LocSectionBase, getFormat(), Index))
1199     return *Off + LocSectionBase;
1200   return std::nullopt;
1201 }
1202