1 //===- DWARFVerifier.cpp --------------------------------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 
10 #include "llvm/DebugInfo/DWARF/DWARFVerifier.h"
11 #include "llvm/ADT/SmallSet.h"
12 #include "llvm/DebugInfo/DWARF/DWARFCompileUnit.h"
13 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
14 #include "llvm/DebugInfo/DWARF/DWARFDebugLine.h"
15 #include "llvm/DebugInfo/DWARF/DWARFDie.h"
16 #include "llvm/DebugInfo/DWARF/DWARFExpression.h"
17 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
18 #include "llvm/DebugInfo/DWARF/DWARFSection.h"
19 #include "llvm/Support/DJB.h"
20 #include "llvm/Support/FormatVariadic.h"
21 #include "llvm/Support/WithColor.h"
22 #include "llvm/Support/raw_ostream.h"
23 #include <map>
24 #include <set>
25 #include <vector>
26 
27 using namespace llvm;
28 using namespace dwarf;
29 using namespace object;
30 
31 DWARFVerifier::DieRangeInfo::address_range_iterator
32 DWARFVerifier::DieRangeInfo::insert(const DWARFAddressRange &R) {
33   auto Begin = Ranges.begin();
34   auto End = Ranges.end();
35   auto Pos = std::lower_bound(Begin, End, R);
36 
37   if (Pos != End) {
38     if (Pos->intersects(R))
39       return Pos;
40     if (Pos != Begin) {
41       auto Iter = Pos - 1;
42       if (Iter->intersects(R))
43         return Iter;
44     }
45   }
46 
47   Ranges.insert(Pos, R);
48   return Ranges.end();
49 }
50 
51 DWARFVerifier::DieRangeInfo::die_range_info_iterator
52 DWARFVerifier::DieRangeInfo::insert(const DieRangeInfo &RI) {
53   auto End = Children.end();
54   auto Iter = Children.begin();
55   while (Iter != End) {
56     if (Iter->intersects(RI))
57       return Iter;
58     ++Iter;
59   }
60   Children.insert(RI);
61   return Children.end();
62 }
63 
64 bool DWARFVerifier::DieRangeInfo::contains(const DieRangeInfo &RHS) const {
65   // Both list of ranges are sorted so we can make this fast.
66 
67   if (Ranges.empty() || RHS.Ranges.empty())
68     return false;
69 
70   // Since the ranges are sorted we can advance where we start searching with
71   // this object's ranges as we traverse RHS.Ranges.
72   auto End = Ranges.end();
73   auto Iter = findRange(RHS.Ranges.front());
74 
75   // Now linearly walk the ranges in this object and see if they contain each
76   // ranges from RHS.Ranges.
77   for (const auto &R : RHS.Ranges) {
78     while (Iter != End) {
79       if (Iter->contains(R))
80         break;
81       ++Iter;
82     }
83     if (Iter == End)
84       return false;
85   }
86   return true;
87 }
88 
89 bool DWARFVerifier::DieRangeInfo::intersects(const DieRangeInfo &RHS) const {
90   if (Ranges.empty() || RHS.Ranges.empty())
91     return false;
92 
93   auto End = Ranges.end();
94   auto Iter = findRange(RHS.Ranges.front());
95   for (const auto &R : RHS.Ranges) {
96     if(Iter == End)
97       return false;
98     if (R.HighPC <= Iter->LowPC)
99       continue;
100     while (Iter != End) {
101       if (Iter->intersects(R))
102         return true;
103       ++Iter;
104     }
105   }
106 
107   return false;
108 }
109 
110 bool DWARFVerifier::verifyUnitHeader(const DWARFDataExtractor DebugInfoData,
111                                      uint32_t *Offset, unsigned UnitIndex,
112                                      uint8_t &UnitType, bool &isUnitDWARF64) {
113   uint32_t AbbrOffset, Length;
114   uint8_t AddrSize = 0;
115   uint16_t Version;
116   bool Success = true;
117 
118   bool ValidLength = false;
119   bool ValidVersion = false;
120   bool ValidAddrSize = false;
121   bool ValidType = true;
122   bool ValidAbbrevOffset = true;
123 
124   uint32_t OffsetStart = *Offset;
125   Length = DebugInfoData.getU32(Offset);
126   if (Length == UINT32_MAX) {
127     isUnitDWARF64 = true;
128     OS << format(
129         "Unit[%d] is in 64-bit DWARF format; cannot verify from this point.\n",
130         UnitIndex);
131     return false;
132   }
133   Version = DebugInfoData.getU16(Offset);
134 
135   if (Version >= 5) {
136     UnitType = DebugInfoData.getU8(Offset);
137     AddrSize = DebugInfoData.getU8(Offset);
138     AbbrOffset = DebugInfoData.getU32(Offset);
139     ValidType = dwarf::isUnitType(UnitType);
140   } else {
141     UnitType = 0;
142     AbbrOffset = DebugInfoData.getU32(Offset);
143     AddrSize = DebugInfoData.getU8(Offset);
144   }
145 
146   if (!DCtx.getDebugAbbrev()->getAbbreviationDeclarationSet(AbbrOffset))
147     ValidAbbrevOffset = false;
148 
149   ValidLength = DebugInfoData.isValidOffset(OffsetStart + Length + 3);
150   ValidVersion = DWARFContext::isSupportedVersion(Version);
151   ValidAddrSize = AddrSize == 4 || AddrSize == 8;
152   if (!ValidLength || !ValidVersion || !ValidAddrSize || !ValidAbbrevOffset ||
153       !ValidType) {
154     Success = false;
155     error() << format("Units[%d] - start offset: 0x%08x \n", UnitIndex,
156                       OffsetStart);
157     if (!ValidLength)
158       note() << "The length for this unit is too "
159             "large for the .debug_info provided.\n";
160     if (!ValidVersion)
161       note() << "The 16 bit unit header version is not valid.\n";
162     if (!ValidType)
163       note() << "The unit type encoding is not valid.\n";
164     if (!ValidAbbrevOffset)
165       note() << "The offset into the .debug_abbrev section is "
166             "not valid.\n";
167     if (!ValidAddrSize)
168       note() << "The address size is unsupported.\n";
169   }
170   *Offset = OffsetStart + Length + 4;
171   return Success;
172 }
173 
174 bool DWARFVerifier::verifyUnitContents(DWARFUnit &Unit, uint8_t UnitType) {
175   uint32_t NumUnitErrors = 0;
176   unsigned NumDies = Unit.getNumDIEs();
177   for (unsigned I = 0; I < NumDies; ++I) {
178     auto Die = Unit.getDIEAtIndex(I);
179     if (Die.getTag() == DW_TAG_null)
180       continue;
181     for (auto AttrValue : Die.attributes()) {
182       NumUnitErrors += verifyDebugInfoAttribute(Die, AttrValue);
183       NumUnitErrors += verifyDebugInfoForm(Die, AttrValue);
184     }
185   }
186 
187   DWARFDie Die = Unit.getUnitDIE(/* ExtractUnitDIEOnly = */ false);
188   if (!Die) {
189     error() << "Compilation unit without DIE.\n";
190     NumUnitErrors++;
191     return NumUnitErrors == 0;
192   }
193 
194   if (!dwarf::isUnitType(Die.getTag())) {
195     error() << "Compilation unit root DIE is not a unit DIE: "
196             << dwarf::TagString(Die.getTag()) << ".\n";
197     NumUnitErrors++;
198   }
199 
200   if (UnitType != 0 &&
201       !DWARFUnit::isMatchingUnitTypeAndTag(UnitType, Die.getTag())) {
202     error() << "Compilation unit type (" << dwarf::UnitTypeString(UnitType)
203             << ") and root DIE (" << dwarf::TagString(Die.getTag())
204             << ") do not match.\n";
205     NumUnitErrors++;
206   }
207 
208   DieRangeInfo RI;
209   NumUnitErrors += verifyDieRanges(Die, RI);
210 
211   return NumUnitErrors == 0;
212 }
213 
214 unsigned DWARFVerifier::verifyAbbrevSection(const DWARFDebugAbbrev *Abbrev) {
215   unsigned NumErrors = 0;
216   if (Abbrev) {
217     const DWARFAbbreviationDeclarationSet *AbbrDecls =
218         Abbrev->getAbbreviationDeclarationSet(0);
219     for (auto AbbrDecl : *AbbrDecls) {
220       SmallDenseSet<uint16_t> AttributeSet;
221       for (auto Attribute : AbbrDecl.attributes()) {
222         auto Result = AttributeSet.insert(Attribute.Attr);
223         if (!Result.second) {
224           error() << "Abbreviation declaration contains multiple "
225                   << AttributeString(Attribute.Attr) << " attributes.\n";
226           AbbrDecl.dump(OS);
227           ++NumErrors;
228         }
229       }
230     }
231   }
232   return NumErrors;
233 }
234 
235 bool DWARFVerifier::handleDebugAbbrev() {
236   OS << "Verifying .debug_abbrev...\n";
237 
238   const DWARFObject &DObj = DCtx.getDWARFObj();
239   bool noDebugAbbrev = DObj.getAbbrevSection().empty();
240   bool noDebugAbbrevDWO = DObj.getAbbrevDWOSection().empty();
241 
242   if (noDebugAbbrev && noDebugAbbrevDWO) {
243     return true;
244   }
245 
246   unsigned NumErrors = 0;
247   if (!noDebugAbbrev)
248     NumErrors += verifyAbbrevSection(DCtx.getDebugAbbrev());
249 
250   if (!noDebugAbbrevDWO)
251     NumErrors += verifyAbbrevSection(DCtx.getDebugAbbrevDWO());
252   return NumErrors == 0;
253 }
254 
255 bool DWARFVerifier::handleDebugInfo() {
256   OS << "Verifying .debug_info Unit Header Chain...\n";
257 
258   const DWARFObject &DObj = DCtx.getDWARFObj();
259   DWARFDataExtractor DebugInfoData(DObj, DObj.getInfoSection(),
260                                    DCtx.isLittleEndian(), 0);
261   uint32_t NumDebugInfoErrors = 0;
262   uint32_t OffsetStart = 0, Offset = 0, UnitIdx = 0;
263   uint8_t UnitType = 0;
264   bool isUnitDWARF64 = false;
265   bool isHeaderChainValid = true;
266   bool hasDIE = DebugInfoData.isValidOffset(Offset);
267   DWARFUnitVector UnitVector{};
268   while (hasDIE) {
269     OffsetStart = Offset;
270     if (!verifyUnitHeader(DebugInfoData, &Offset, UnitIdx, UnitType,
271                           isUnitDWARF64)) {
272       isHeaderChainValid = false;
273       if (isUnitDWARF64)
274         break;
275     } else {
276       DWARFUnitHeader Header;
277       Header.extract(DCtx, DebugInfoData, &OffsetStart);
278       std::unique_ptr<DWARFUnit> Unit;
279       switch (UnitType) {
280       case dwarf::DW_UT_type:
281       case dwarf::DW_UT_split_type: {
282         Unit.reset(new DWARFTypeUnit(
283             DCtx, DObj.getInfoSection(), Header, DCtx.getDebugAbbrev(),
284             &DObj.getRangeSection(), DObj.getStringSection(),
285             DObj.getStringOffsetSection(), &DObj.getAppleObjCSection(),
286             DObj.getLineSection(), DCtx.isLittleEndian(), false, UnitVector));
287         break;
288       }
289       case dwarf::DW_UT_skeleton:
290       case dwarf::DW_UT_split_compile:
291       case dwarf::DW_UT_compile:
292       case dwarf::DW_UT_partial:
293       // UnitType = 0 means that we are
294       // verifying a compile unit in DWARF v4.
295       case 0: {
296         Unit.reset(new DWARFCompileUnit(
297             DCtx, DObj.getInfoSection(), Header, DCtx.getDebugAbbrev(),
298             &DObj.getRangeSection(), DObj.getStringSection(),
299             DObj.getStringOffsetSection(), &DObj.getAppleObjCSection(),
300             DObj.getLineSection(), DCtx.isLittleEndian(), false, UnitVector));
301         break;
302       }
303       default: { llvm_unreachable("Invalid UnitType."); }
304       }
305       if (!verifyUnitContents(*Unit, UnitType))
306         ++NumDebugInfoErrors;
307     }
308     hasDIE = DebugInfoData.isValidOffset(Offset);
309     ++UnitIdx;
310   }
311   if (UnitIdx == 0 && !hasDIE) {
312     warn() << ".debug_info is empty.\n";
313     isHeaderChainValid = true;
314   }
315   NumDebugInfoErrors += verifyDebugInfoReferences();
316   return (isHeaderChainValid && NumDebugInfoErrors == 0);
317 }
318 
319 unsigned DWARFVerifier::verifyDieRanges(const DWARFDie &Die,
320                                         DieRangeInfo &ParentRI) {
321   unsigned NumErrors = 0;
322 
323   if (!Die.isValid())
324     return NumErrors;
325 
326   auto RangesOrError = Die.getAddressRanges();
327   if (!RangesOrError) {
328     // FIXME: Report the error.
329     ++NumErrors;
330     llvm::consumeError(RangesOrError.takeError());
331     return NumErrors;
332   }
333 
334   DWARFAddressRangesVector Ranges = RangesOrError.get();
335   // Build RI for this DIE and check that ranges within this DIE do not
336   // overlap.
337   DieRangeInfo RI(Die);
338   for (auto Range : Ranges) {
339     if (!Range.valid()) {
340       ++NumErrors;
341       error() << "Invalid address range " << Range << "\n";
342       continue;
343     }
344 
345     // Verify that ranges don't intersect.
346     const auto IntersectingRange = RI.insert(Range);
347     if (IntersectingRange != RI.Ranges.end()) {
348       ++NumErrors;
349       error() << "DIE has overlapping address ranges: " << Range << " and "
350               << *IntersectingRange << "\n";
351       break;
352     }
353   }
354 
355   // Verify that children don't intersect.
356   const auto IntersectingChild = ParentRI.insert(RI);
357   if (IntersectingChild != ParentRI.Children.end()) {
358     ++NumErrors;
359     error() << "DIEs have overlapping address ranges:";
360     Die.dump(OS, 0);
361     IntersectingChild->Die.dump(OS, 0);
362     OS << "\n";
363   }
364 
365   // Verify that ranges are contained within their parent.
366   bool ShouldBeContained = !Ranges.empty() && !ParentRI.Ranges.empty() &&
367                            !(Die.getTag() == DW_TAG_subprogram &&
368                              ParentRI.Die.getTag() == DW_TAG_subprogram);
369   if (ShouldBeContained && !ParentRI.contains(RI)) {
370     ++NumErrors;
371     error() << "DIE address ranges are not contained in its parent's ranges:";
372     ParentRI.Die.dump(OS, 0);
373     Die.dump(OS, 2);
374     OS << "\n";
375   }
376 
377   // Recursively check children.
378   for (DWARFDie Child : Die)
379     NumErrors += verifyDieRanges(Child, RI);
380 
381   return NumErrors;
382 }
383 
384 unsigned DWARFVerifier::verifyDebugInfoAttribute(const DWARFDie &Die,
385                                                  DWARFAttribute &AttrValue) {
386   unsigned NumErrors = 0;
387   auto ReportError = [&](const Twine &TitleMsg) {
388     ++NumErrors;
389     error() << TitleMsg << '\n';
390     Die.dump(OS, 0, DumpOpts);
391     OS << "\n";
392   };
393 
394   const DWARFObject &DObj = DCtx.getDWARFObj();
395   const auto Attr = AttrValue.Attr;
396   switch (Attr) {
397   case DW_AT_ranges:
398     // Make sure the offset in the DW_AT_ranges attribute is valid.
399     if (auto SectionOffset = AttrValue.Value.getAsSectionOffset()) {
400       if (*SectionOffset >= DObj.getRangeSection().Data.size())
401         ReportError("DW_AT_ranges offset is beyond .debug_ranges bounds:");
402       break;
403     }
404     ReportError("DIE has invalid DW_AT_ranges encoding:");
405     break;
406   case DW_AT_stmt_list:
407     // Make sure the offset in the DW_AT_stmt_list attribute is valid.
408     if (auto SectionOffset = AttrValue.Value.getAsSectionOffset()) {
409       if (*SectionOffset >= DObj.getLineSection().Data.size())
410         ReportError("DW_AT_stmt_list offset is beyond .debug_line bounds: " +
411                     llvm::formatv("{0:x8}", *SectionOffset));
412       break;
413     }
414     ReportError("DIE has invalid DW_AT_stmt_list encoding:");
415     break;
416   case DW_AT_location: {
417     auto VerifyLocationExpr = [&](StringRef D) {
418       DWARFUnit *U = Die.getDwarfUnit();
419       DataExtractor Data(D, DCtx.isLittleEndian(), 0);
420       DWARFExpression Expression(Data, U->getVersion(),
421                                  U->getAddressByteSize());
422       bool Error = llvm::any_of(Expression, [](DWARFExpression::Operation &Op) {
423         return Op.isError();
424       });
425       if (Error)
426         ReportError("DIE contains invalid DWARF expression:");
427     };
428     if (Optional<ArrayRef<uint8_t>> Expr = AttrValue.Value.getAsBlock()) {
429       // Verify inlined location.
430       VerifyLocationExpr(llvm::toStringRef(*Expr));
431     } else if (auto LocOffset = AttrValue.Value.getAsSectionOffset()) {
432       // Verify location list.
433       if (auto DebugLoc = DCtx.getDebugLoc())
434         if (auto LocList = DebugLoc->getLocationListAtOffset(*LocOffset))
435           for (const auto &Entry : LocList->Entries)
436             VerifyLocationExpr({Entry.Loc.data(), Entry.Loc.size()});
437     }
438     break;
439   }
440 
441   default:
442     break;
443   }
444   return NumErrors;
445 }
446 
447 unsigned DWARFVerifier::verifyDebugInfoForm(const DWARFDie &Die,
448                                             DWARFAttribute &AttrValue) {
449   const DWARFObject &DObj = DCtx.getDWARFObj();
450   unsigned NumErrors = 0;
451   const auto Form = AttrValue.Value.getForm();
452   switch (Form) {
453   case DW_FORM_ref1:
454   case DW_FORM_ref2:
455   case DW_FORM_ref4:
456   case DW_FORM_ref8:
457   case DW_FORM_ref_udata: {
458     // Verify all CU relative references are valid CU offsets.
459     Optional<uint64_t> RefVal = AttrValue.Value.getAsReference();
460     assert(RefVal);
461     if (RefVal) {
462       auto DieCU = Die.getDwarfUnit();
463       auto CUSize = DieCU->getNextUnitOffset() - DieCU->getOffset();
464       auto CUOffset = AttrValue.Value.getRawUValue();
465       if (CUOffset >= CUSize) {
466         ++NumErrors;
467         error() << FormEncodingString(Form) << " CU offset "
468                 << format("0x%08" PRIx64, CUOffset)
469                 << " is invalid (must be less than CU size of "
470                 << format("0x%08" PRIx32, CUSize) << "):\n";
471         Die.dump(OS, 0, DumpOpts);
472         OS << "\n";
473       } else {
474         // Valid reference, but we will verify it points to an actual
475         // DIE later.
476         ReferenceToDIEOffsets[*RefVal].insert(Die.getOffset());
477       }
478     }
479     break;
480   }
481   case DW_FORM_ref_addr: {
482     // Verify all absolute DIE references have valid offsets in the
483     // .debug_info section.
484     Optional<uint64_t> RefVal = AttrValue.Value.getAsReference();
485     assert(RefVal);
486     if (RefVal) {
487       if (*RefVal >= DObj.getInfoSection().Data.size()) {
488         ++NumErrors;
489         error() << "DW_FORM_ref_addr offset beyond .debug_info "
490                    "bounds:\n";
491         Die.dump(OS, 0, DumpOpts);
492         OS << "\n";
493       } else {
494         // Valid reference, but we will verify it points to an actual
495         // DIE later.
496         ReferenceToDIEOffsets[*RefVal].insert(Die.getOffset());
497       }
498     }
499     break;
500   }
501   case DW_FORM_strp: {
502     auto SecOffset = AttrValue.Value.getAsSectionOffset();
503     assert(SecOffset); // DW_FORM_strp is a section offset.
504     if (SecOffset && *SecOffset >= DObj.getStringSection().size()) {
505       ++NumErrors;
506       error() << "DW_FORM_strp offset beyond .debug_str bounds:\n";
507       Die.dump(OS, 0, DumpOpts);
508       OS << "\n";
509     }
510     break;
511   }
512   default:
513     break;
514   }
515   return NumErrors;
516 }
517 
518 unsigned DWARFVerifier::verifyDebugInfoReferences() {
519   // Take all references and make sure they point to an actual DIE by
520   // getting the DIE by offset and emitting an error
521   OS << "Verifying .debug_info references...\n";
522   unsigned NumErrors = 0;
523   for (auto Pair : ReferenceToDIEOffsets) {
524     auto Die = DCtx.getDIEForOffset(Pair.first);
525     if (Die)
526       continue;
527     ++NumErrors;
528     error() << "invalid DIE reference " << format("0x%08" PRIx64, Pair.first)
529             << ". Offset is in between DIEs:\n";
530     for (auto Offset : Pair.second) {
531       auto ReferencingDie = DCtx.getDIEForOffset(Offset);
532       ReferencingDie.dump(OS, 0, DumpOpts);
533       OS << "\n";
534     }
535     OS << "\n";
536   }
537   return NumErrors;
538 }
539 
540 void DWARFVerifier::verifyDebugLineStmtOffsets() {
541   std::map<uint64_t, DWARFDie> StmtListToDie;
542   for (const auto &CU : DCtx.compile_units()) {
543     auto Die = CU->getUnitDIE();
544     // Get the attribute value as a section offset. No need to produce an
545     // error here if the encoding isn't correct because we validate this in
546     // the .debug_info verifier.
547     auto StmtSectionOffset = toSectionOffset(Die.find(DW_AT_stmt_list));
548     if (!StmtSectionOffset)
549       continue;
550     const uint32_t LineTableOffset = *StmtSectionOffset;
551     auto LineTable = DCtx.getLineTableForUnit(CU.get());
552     if (LineTableOffset < DCtx.getDWARFObj().getLineSection().Data.size()) {
553       if (!LineTable) {
554         ++NumDebugLineErrors;
555         error() << ".debug_line[" << format("0x%08" PRIx32, LineTableOffset)
556                 << "] was not able to be parsed for CU:\n";
557         Die.dump(OS, 0, DumpOpts);
558         OS << '\n';
559         continue;
560       }
561     } else {
562       // Make sure we don't get a valid line table back if the offset is wrong.
563       assert(LineTable == nullptr);
564       // Skip this line table as it isn't valid. No need to create an error
565       // here because we validate this in the .debug_info verifier.
566       continue;
567     }
568     auto Iter = StmtListToDie.find(LineTableOffset);
569     if (Iter != StmtListToDie.end()) {
570       ++NumDebugLineErrors;
571       error() << "two compile unit DIEs, "
572               << format("0x%08" PRIx32, Iter->second.getOffset()) << " and "
573               << format("0x%08" PRIx32, Die.getOffset())
574               << ", have the same DW_AT_stmt_list section offset:\n";
575       Iter->second.dump(OS, 0, DumpOpts);
576       Die.dump(OS, 0, DumpOpts);
577       OS << '\n';
578       // Already verified this line table before, no need to do it again.
579       continue;
580     }
581     StmtListToDie[LineTableOffset] = Die;
582   }
583 }
584 
585 void DWARFVerifier::verifyDebugLineRows() {
586   for (const auto &CU : DCtx.compile_units()) {
587     auto Die = CU->getUnitDIE();
588     auto LineTable = DCtx.getLineTableForUnit(CU.get());
589     // If there is no line table we will have created an error in the
590     // .debug_info verifier or in verifyDebugLineStmtOffsets().
591     if (!LineTable)
592       continue;
593 
594     // Verify prologue.
595     uint32_t MaxFileIndex = LineTable->Prologue.FileNames.size();
596     uint32_t MaxDirIndex = LineTable->Prologue.IncludeDirectories.size();
597     uint32_t FileIndex = 1;
598     StringMap<uint16_t> FullPathMap;
599     for (const auto &FileName : LineTable->Prologue.FileNames) {
600       // Verify directory index.
601       if (FileName.DirIdx > MaxDirIndex) {
602         ++NumDebugLineErrors;
603         error() << ".debug_line["
604                 << format("0x%08" PRIx64,
605                           *toSectionOffset(Die.find(DW_AT_stmt_list)))
606                 << "].prologue.file_names[" << FileIndex
607                 << "].dir_idx contains an invalid index: " << FileName.DirIdx
608                 << "\n";
609       }
610 
611       // Check file paths for duplicates.
612       std::string FullPath;
613       const bool HasFullPath = LineTable->getFileNameByIndex(
614           FileIndex, CU->getCompilationDir(),
615           DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FullPath);
616       assert(HasFullPath && "Invalid index?");
617       (void)HasFullPath;
618       auto It = FullPathMap.find(FullPath);
619       if (It == FullPathMap.end())
620         FullPathMap[FullPath] = FileIndex;
621       else if (It->second != FileIndex) {
622         warn() << ".debug_line["
623                << format("0x%08" PRIx64,
624                          *toSectionOffset(Die.find(DW_AT_stmt_list)))
625                << "].prologue.file_names[" << FileIndex
626                << "] is a duplicate of file_names[" << It->second << "]\n";
627       }
628 
629       FileIndex++;
630     }
631 
632     // Verify rows.
633     uint64_t PrevAddress = 0;
634     uint32_t RowIndex = 0;
635     for (const auto &Row : LineTable->Rows) {
636       // Verify row address.
637       if (Row.Address < PrevAddress) {
638         ++NumDebugLineErrors;
639         error() << ".debug_line["
640                 << format("0x%08" PRIx64,
641                           *toSectionOffset(Die.find(DW_AT_stmt_list)))
642                 << "] row[" << RowIndex
643                 << "] decreases in address from previous row:\n";
644 
645         DWARFDebugLine::Row::dumpTableHeader(OS);
646         if (RowIndex > 0)
647           LineTable->Rows[RowIndex - 1].dump(OS);
648         Row.dump(OS);
649         OS << '\n';
650       }
651 
652       // Verify file index.
653       if (Row.File > MaxFileIndex) {
654         ++NumDebugLineErrors;
655         error() << ".debug_line["
656                 << format("0x%08" PRIx64,
657                           *toSectionOffset(Die.find(DW_AT_stmt_list)))
658                 << "][" << RowIndex << "] has invalid file index " << Row.File
659                 << " (valid values are [1," << MaxFileIndex << "]):\n";
660         DWARFDebugLine::Row::dumpTableHeader(OS);
661         Row.dump(OS);
662         OS << '\n';
663       }
664       if (Row.EndSequence)
665         PrevAddress = 0;
666       else
667         PrevAddress = Row.Address;
668       ++RowIndex;
669     }
670   }
671 }
672 
673 bool DWARFVerifier::handleDebugLine() {
674   NumDebugLineErrors = 0;
675   OS << "Verifying .debug_line...\n";
676   verifyDebugLineStmtOffsets();
677   verifyDebugLineRows();
678   return NumDebugLineErrors == 0;
679 }
680 
681 unsigned DWARFVerifier::verifyAppleAccelTable(const DWARFSection *AccelSection,
682                                               DataExtractor *StrData,
683                                               const char *SectionName) {
684   unsigned NumErrors = 0;
685   DWARFDataExtractor AccelSectionData(DCtx.getDWARFObj(), *AccelSection,
686                                       DCtx.isLittleEndian(), 0);
687   AppleAcceleratorTable AccelTable(AccelSectionData, *StrData);
688 
689   OS << "Verifying " << SectionName << "...\n";
690 
691   // Verify that the fixed part of the header is not too short.
692   if (!AccelSectionData.isValidOffset(AccelTable.getSizeHdr())) {
693     error() << "Section is too small to fit a section header.\n";
694     return 1;
695   }
696 
697   // Verify that the section is not too short.
698   if (Error E = AccelTable.extract()) {
699     error() << toString(std::move(E)) << '\n';
700     return 1;
701   }
702 
703   // Verify that all buckets have a valid hash index or are empty.
704   uint32_t NumBuckets = AccelTable.getNumBuckets();
705   uint32_t NumHashes = AccelTable.getNumHashes();
706 
707   uint32_t BucketsOffset =
708       AccelTable.getSizeHdr() + AccelTable.getHeaderDataLength();
709   uint32_t HashesBase = BucketsOffset + NumBuckets * 4;
710   uint32_t OffsetsBase = HashesBase + NumHashes * 4;
711   for (uint32_t BucketIdx = 0; BucketIdx < NumBuckets; ++BucketIdx) {
712     uint32_t HashIdx = AccelSectionData.getU32(&BucketsOffset);
713     if (HashIdx >= NumHashes && HashIdx != UINT32_MAX) {
714       error() << format("Bucket[%d] has invalid hash index: %u.\n", BucketIdx,
715                         HashIdx);
716       ++NumErrors;
717     }
718   }
719   uint32_t NumAtoms = AccelTable.getAtomsDesc().size();
720   if (NumAtoms == 0) {
721     error() << "No atoms: failed to read HashData.\n";
722     return 1;
723   }
724   if (!AccelTable.validateForms()) {
725     error() << "Unsupported form: failed to read HashData.\n";
726     return 1;
727   }
728 
729   for (uint32_t HashIdx = 0; HashIdx < NumHashes; ++HashIdx) {
730     uint32_t HashOffset = HashesBase + 4 * HashIdx;
731     uint32_t DataOffset = OffsetsBase + 4 * HashIdx;
732     uint32_t Hash = AccelSectionData.getU32(&HashOffset);
733     uint32_t HashDataOffset = AccelSectionData.getU32(&DataOffset);
734     if (!AccelSectionData.isValidOffsetForDataOfSize(HashDataOffset,
735                                                      sizeof(uint64_t))) {
736       error() << format("Hash[%d] has invalid HashData offset: 0x%08x.\n",
737                         HashIdx, HashDataOffset);
738       ++NumErrors;
739     }
740 
741     uint32_t StrpOffset;
742     uint32_t StringOffset;
743     uint32_t StringCount = 0;
744     unsigned Offset;
745     unsigned Tag;
746     while ((StrpOffset = AccelSectionData.getU32(&HashDataOffset)) != 0) {
747       const uint32_t NumHashDataObjects =
748           AccelSectionData.getU32(&HashDataOffset);
749       for (uint32_t HashDataIdx = 0; HashDataIdx < NumHashDataObjects;
750            ++HashDataIdx) {
751         std::tie(Offset, Tag) = AccelTable.readAtoms(HashDataOffset);
752         auto Die = DCtx.getDIEForOffset(Offset);
753         if (!Die) {
754           const uint32_t BucketIdx =
755               NumBuckets ? (Hash % NumBuckets) : UINT32_MAX;
756           StringOffset = StrpOffset;
757           const char *Name = StrData->getCStr(&StringOffset);
758           if (!Name)
759             Name = "<NULL>";
760 
761           error() << format(
762               "%s Bucket[%d] Hash[%d] = 0x%08x "
763               "Str[%u] = 0x%08x "
764               "DIE[%d] = 0x%08x is not a valid DIE offset for \"%s\".\n",
765               SectionName, BucketIdx, HashIdx, Hash, StringCount, StrpOffset,
766               HashDataIdx, Offset, Name);
767 
768           ++NumErrors;
769           continue;
770         }
771         if ((Tag != dwarf::DW_TAG_null) && (Die.getTag() != Tag)) {
772           error() << "Tag " << dwarf::TagString(Tag)
773                   << " in accelerator table does not match Tag "
774                   << dwarf::TagString(Die.getTag()) << " of DIE[" << HashDataIdx
775                   << "].\n";
776           ++NumErrors;
777         }
778       }
779       ++StringCount;
780     }
781   }
782   return NumErrors;
783 }
784 
785 unsigned
786 DWARFVerifier::verifyDebugNamesCULists(const DWARFDebugNames &AccelTable) {
787   // A map from CU offset to the (first) Name Index offset which claims to index
788   // this CU.
789   DenseMap<uint32_t, uint32_t> CUMap;
790   const uint32_t NotIndexed = std::numeric_limits<uint32_t>::max();
791 
792   CUMap.reserve(DCtx.getNumCompileUnits());
793   for (const auto &CU : DCtx.compile_units())
794     CUMap[CU->getOffset()] = NotIndexed;
795 
796   unsigned NumErrors = 0;
797   for (const DWARFDebugNames::NameIndex &NI : AccelTable) {
798     if (NI.getCUCount() == 0) {
799       error() << formatv("Name Index @ {0:x} does not index any CU\n",
800                          NI.getUnitOffset());
801       ++NumErrors;
802       continue;
803     }
804     for (uint32_t CU = 0, End = NI.getCUCount(); CU < End; ++CU) {
805       uint32_t Offset = NI.getCUOffset(CU);
806       auto Iter = CUMap.find(Offset);
807 
808       if (Iter == CUMap.end()) {
809         error() << formatv(
810             "Name Index @ {0:x} references a non-existing CU @ {1:x}\n",
811             NI.getUnitOffset(), Offset);
812         ++NumErrors;
813         continue;
814       }
815 
816       if (Iter->second != NotIndexed) {
817         error() << formatv("Name Index @ {0:x} references a CU @ {1:x}, but "
818                           "this CU is already indexed by Name Index @ {2:x}\n",
819                           NI.getUnitOffset(), Offset, Iter->second);
820         continue;
821       }
822       Iter->second = NI.getUnitOffset();
823     }
824   }
825 
826   for (const auto &KV : CUMap) {
827     if (KV.second == NotIndexed)
828       warn() << formatv("CU @ {0:x} not covered by any Name Index\n", KV.first);
829   }
830 
831   return NumErrors;
832 }
833 
834 unsigned
835 DWARFVerifier::verifyNameIndexBuckets(const DWARFDebugNames::NameIndex &NI,
836                                       const DataExtractor &StrData) {
837   struct BucketInfo {
838     uint32_t Bucket;
839     uint32_t Index;
840 
841     constexpr BucketInfo(uint32_t Bucket, uint32_t Index)
842         : Bucket(Bucket), Index(Index) {}
843     bool operator<(const BucketInfo &RHS) const { return Index < RHS.Index; };
844   };
845 
846   uint32_t NumErrors = 0;
847   if (NI.getBucketCount() == 0) {
848     warn() << formatv("Name Index @ {0:x} does not contain a hash table.\n",
849                       NI.getUnitOffset());
850     return NumErrors;
851   }
852 
853   // Build up a list of (Bucket, Index) pairs. We use this later to verify that
854   // each Name is reachable from the appropriate bucket.
855   std::vector<BucketInfo> BucketStarts;
856   BucketStarts.reserve(NI.getBucketCount() + 1);
857   for (uint32_t Bucket = 0, End = NI.getBucketCount(); Bucket < End; ++Bucket) {
858     uint32_t Index = NI.getBucketArrayEntry(Bucket);
859     if (Index > NI.getNameCount()) {
860       error() << formatv("Bucket {0} of Name Index @ {1:x} contains invalid "
861                          "value {2}. Valid range is [0, {3}].\n",
862                          Bucket, NI.getUnitOffset(), Index, NI.getNameCount());
863       ++NumErrors;
864       continue;
865     }
866     if (Index > 0)
867       BucketStarts.emplace_back(Bucket, Index);
868   }
869 
870   // If there were any buckets with invalid values, skip further checks as they
871   // will likely produce many errors which will only confuse the actual root
872   // problem.
873   if (NumErrors > 0)
874     return NumErrors;
875 
876   // Sort the list in the order of increasing "Index" entries.
877   array_pod_sort(BucketStarts.begin(), BucketStarts.end());
878 
879   // Insert a sentinel entry at the end, so we can check that the end of the
880   // table is covered in the loop below.
881   BucketStarts.emplace_back(NI.getBucketCount(), NI.getNameCount() + 1);
882 
883   // Loop invariant: NextUncovered is the (1-based) index of the first Name
884   // which is not reachable by any of the buckets we processed so far (and
885   // hasn't been reported as uncovered).
886   uint32_t NextUncovered = 1;
887   for (const BucketInfo &B : BucketStarts) {
888     // Under normal circumstances B.Index be equal to NextUncovered, but it can
889     // be less if a bucket points to names which are already known to be in some
890     // bucket we processed earlier. In that case, we won't trigger this error,
891     // but report the mismatched hash value error instead. (We know the hash
892     // will not match because we have already verified that the name's hash
893     // puts it into the previous bucket.)
894     if (B.Index > NextUncovered) {
895       error() << formatv("Name Index @ {0:x}: Name table entries [{1}, {2}] "
896                          "are not covered by the hash table.\n",
897                          NI.getUnitOffset(), NextUncovered, B.Index - 1);
898       ++NumErrors;
899     }
900     uint32_t Idx = B.Index;
901 
902     // The rest of the checks apply only to non-sentinel entries.
903     if (B.Bucket == NI.getBucketCount())
904       break;
905 
906     // This triggers if a non-empty bucket points to a name with a mismatched
907     // hash. Clients are likely to interpret this as an empty bucket, because a
908     // mismatched hash signals the end of a bucket, but if this is indeed an
909     // empty bucket, the producer should have signalled this by marking the
910     // bucket as empty.
911     uint32_t FirstHash = NI.getHashArrayEntry(Idx);
912     if (FirstHash % NI.getBucketCount() != B.Bucket) {
913       error() << formatv(
914           "Name Index @ {0:x}: Bucket {1} is not empty but points to a "
915           "mismatched hash value {2:x} (belonging to bucket {3}).\n",
916           NI.getUnitOffset(), B.Bucket, FirstHash,
917           FirstHash % NI.getBucketCount());
918       ++NumErrors;
919     }
920 
921     // This find the end of this bucket and also verifies that all the hashes in
922     // this bucket are correct by comparing the stored hashes to the ones we
923     // compute ourselves.
924     while (Idx <= NI.getNameCount()) {
925       uint32_t Hash = NI.getHashArrayEntry(Idx);
926       if (Hash % NI.getBucketCount() != B.Bucket)
927         break;
928 
929       const char *Str = NI.getNameTableEntry(Idx).getString();
930       if (caseFoldingDjbHash(Str) != Hash) {
931         error() << formatv("Name Index @ {0:x}: String ({1}) at index {2} "
932                            "hashes to {3:x}, but "
933                            "the Name Index hash is {4:x}\n",
934                            NI.getUnitOffset(), Str, Idx,
935                            caseFoldingDjbHash(Str), Hash);
936         ++NumErrors;
937       }
938 
939       ++Idx;
940     }
941     NextUncovered = std::max(NextUncovered, Idx);
942   }
943   return NumErrors;
944 }
945 
946 unsigned DWARFVerifier::verifyNameIndexAttribute(
947     const DWARFDebugNames::NameIndex &NI, const DWARFDebugNames::Abbrev &Abbr,
948     DWARFDebugNames::AttributeEncoding AttrEnc) {
949   StringRef FormName = dwarf::FormEncodingString(AttrEnc.Form);
950   if (FormName.empty()) {
951     error() << formatv("NameIndex @ {0:x}: Abbreviation {1:x}: {2} uses an "
952                        "unknown form: {3}.\n",
953                        NI.getUnitOffset(), Abbr.Code, AttrEnc.Index,
954                        AttrEnc.Form);
955     return 1;
956   }
957 
958   if (AttrEnc.Index == DW_IDX_type_hash) {
959     if (AttrEnc.Form != dwarf::DW_FORM_data8) {
960       error() << formatv(
961           "NameIndex @ {0:x}: Abbreviation {1:x}: DW_IDX_type_hash "
962           "uses an unexpected form {2} (should be {3}).\n",
963           NI.getUnitOffset(), Abbr.Code, AttrEnc.Form, dwarf::DW_FORM_data8);
964       return 1;
965     }
966   }
967 
968   // A list of known index attributes and their expected form classes.
969   // DW_IDX_type_hash is handled specially in the check above, as it has a
970   // specific form (not just a form class) we should expect.
971   struct FormClassTable {
972     dwarf::Index Index;
973     DWARFFormValue::FormClass Class;
974     StringLiteral ClassName;
975   };
976   static constexpr FormClassTable Table[] = {
977       {dwarf::DW_IDX_compile_unit, DWARFFormValue::FC_Constant, {"constant"}},
978       {dwarf::DW_IDX_type_unit, DWARFFormValue::FC_Constant, {"constant"}},
979       {dwarf::DW_IDX_die_offset, DWARFFormValue::FC_Reference, {"reference"}},
980       {dwarf::DW_IDX_parent, DWARFFormValue::FC_Constant, {"constant"}},
981   };
982 
983   ArrayRef<FormClassTable> TableRef(Table);
984   auto Iter = find_if(TableRef, [AttrEnc](const FormClassTable &T) {
985     return T.Index == AttrEnc.Index;
986   });
987   if (Iter == TableRef.end()) {
988     warn() << formatv("NameIndex @ {0:x}: Abbreviation {1:x} contains an "
989                       "unknown index attribute: {2}.\n",
990                       NI.getUnitOffset(), Abbr.Code, AttrEnc.Index);
991     return 0;
992   }
993 
994   if (!DWARFFormValue(AttrEnc.Form).isFormClass(Iter->Class)) {
995     error() << formatv("NameIndex @ {0:x}: Abbreviation {1:x}: {2} uses an "
996                        "unexpected form {3} (expected form class {4}).\n",
997                        NI.getUnitOffset(), Abbr.Code, AttrEnc.Index,
998                        AttrEnc.Form, Iter->ClassName);
999     return 1;
1000   }
1001   return 0;
1002 }
1003 
1004 unsigned
1005 DWARFVerifier::verifyNameIndexAbbrevs(const DWARFDebugNames::NameIndex &NI) {
1006   if (NI.getLocalTUCount() + NI.getForeignTUCount() > 0) {
1007     warn() << formatv("Name Index @ {0:x}: Verifying indexes of type units is "
1008                       "not currently supported.\n",
1009                       NI.getUnitOffset());
1010     return 0;
1011   }
1012 
1013   unsigned NumErrors = 0;
1014   for (const auto &Abbrev : NI.getAbbrevs()) {
1015     StringRef TagName = dwarf::TagString(Abbrev.Tag);
1016     if (TagName.empty()) {
1017       warn() << formatv("NameIndex @ {0:x}: Abbreviation {1:x} references an "
1018                         "unknown tag: {2}.\n",
1019                         NI.getUnitOffset(), Abbrev.Code, Abbrev.Tag);
1020     }
1021     SmallSet<unsigned, 5> Attributes;
1022     for (const auto &AttrEnc : Abbrev.Attributes) {
1023       if (!Attributes.insert(AttrEnc.Index).second) {
1024         error() << formatv("NameIndex @ {0:x}: Abbreviation {1:x} contains "
1025                            "multiple {2} attributes.\n",
1026                            NI.getUnitOffset(), Abbrev.Code, AttrEnc.Index);
1027         ++NumErrors;
1028         continue;
1029       }
1030       NumErrors += verifyNameIndexAttribute(NI, Abbrev, AttrEnc);
1031     }
1032 
1033     if (NI.getCUCount() > 1 && !Attributes.count(dwarf::DW_IDX_compile_unit)) {
1034       error() << formatv("NameIndex @ {0:x}: Indexing multiple compile units "
1035                          "and abbreviation {1:x} has no {2} attribute.\n",
1036                          NI.getUnitOffset(), Abbrev.Code,
1037                          dwarf::DW_IDX_compile_unit);
1038       ++NumErrors;
1039     }
1040     if (!Attributes.count(dwarf::DW_IDX_die_offset)) {
1041       error() << formatv(
1042           "NameIndex @ {0:x}: Abbreviation {1:x} has no {2} attribute.\n",
1043           NI.getUnitOffset(), Abbrev.Code, dwarf::DW_IDX_die_offset);
1044       ++NumErrors;
1045     }
1046   }
1047   return NumErrors;
1048 }
1049 
1050 static SmallVector<StringRef, 2> getNames(const DWARFDie &DIE) {
1051   SmallVector<StringRef, 2> Result;
1052   if (const char *Str = DIE.getName(DINameKind::ShortName))
1053     Result.emplace_back(Str);
1054   else if (DIE.getTag() == dwarf::DW_TAG_namespace)
1055     Result.emplace_back("(anonymous namespace)");
1056 
1057   if (const char *Str = DIE.getName(DINameKind::LinkageName)) {
1058     if (Result.empty() || Result[0] != Str)
1059       Result.emplace_back(Str);
1060   }
1061 
1062   return Result;
1063 }
1064 
1065 unsigned DWARFVerifier::verifyNameIndexEntries(
1066     const DWARFDebugNames::NameIndex &NI,
1067     const DWARFDebugNames::NameTableEntry &NTE) {
1068   // Verifying type unit indexes not supported.
1069   if (NI.getLocalTUCount() + NI.getForeignTUCount() > 0)
1070     return 0;
1071 
1072   const char *CStr = NTE.getString();
1073   if (!CStr) {
1074     error() << formatv(
1075         "Name Index @ {0:x}: Unable to get string associated with name {1}.\n",
1076         NI.getUnitOffset(), NTE.getIndex());
1077     return 1;
1078   }
1079   StringRef Str(CStr);
1080 
1081   unsigned NumErrors = 0;
1082   unsigned NumEntries = 0;
1083   uint32_t EntryID = NTE.getEntryOffset();
1084   uint32_t NextEntryID = EntryID;
1085   Expected<DWARFDebugNames::Entry> EntryOr = NI.getEntry(&NextEntryID);
1086   for (; EntryOr; ++NumEntries, EntryID = NextEntryID,
1087                                 EntryOr = NI.getEntry(&NextEntryID)) {
1088     uint32_t CUIndex = *EntryOr->getCUIndex();
1089     if (CUIndex > NI.getCUCount()) {
1090       error() << formatv("Name Index @ {0:x}: Entry @ {1:x} contains an "
1091                          "invalid CU index ({2}).\n",
1092                          NI.getUnitOffset(), EntryID, CUIndex);
1093       ++NumErrors;
1094       continue;
1095     }
1096     uint32_t CUOffset = NI.getCUOffset(CUIndex);
1097     uint64_t DIEOffset = CUOffset + *EntryOr->getDIEUnitOffset();
1098     DWARFDie DIE = DCtx.getDIEForOffset(DIEOffset);
1099     if (!DIE) {
1100       error() << formatv("Name Index @ {0:x}: Entry @ {1:x} references a "
1101                          "non-existing DIE @ {2:x}.\n",
1102                          NI.getUnitOffset(), EntryID, DIEOffset);
1103       ++NumErrors;
1104       continue;
1105     }
1106     if (DIE.getDwarfUnit()->getOffset() != CUOffset) {
1107       error() << formatv("Name Index @ {0:x}: Entry @ {1:x}: mismatched CU of "
1108                          "DIE @ {2:x}: index - {3:x}; debug_info - {4:x}.\n",
1109                          NI.getUnitOffset(), EntryID, DIEOffset, CUOffset,
1110                          DIE.getDwarfUnit()->getOffset());
1111       ++NumErrors;
1112     }
1113     if (DIE.getTag() != EntryOr->tag()) {
1114       error() << formatv("Name Index @ {0:x}: Entry @ {1:x}: mismatched Tag of "
1115                          "DIE @ {2:x}: index - {3}; debug_info - {4}.\n",
1116                          NI.getUnitOffset(), EntryID, DIEOffset, EntryOr->tag(),
1117                          DIE.getTag());
1118       ++NumErrors;
1119     }
1120 
1121     auto EntryNames = getNames(DIE);
1122     if (!is_contained(EntryNames, Str)) {
1123       error() << formatv("Name Index @ {0:x}: Entry @ {1:x}: mismatched Name "
1124                          "of DIE @ {2:x}: index - {3}; debug_info - {4}.\n",
1125                          NI.getUnitOffset(), EntryID, DIEOffset, Str,
1126                          make_range(EntryNames.begin(), EntryNames.end()));
1127       ++NumErrors;
1128     }
1129   }
1130   handleAllErrors(EntryOr.takeError(),
1131                   [&](const DWARFDebugNames::SentinelError &) {
1132                     if (NumEntries > 0)
1133                       return;
1134                     error() << formatv("Name Index @ {0:x}: Name {1} ({2}) is "
1135                                        "not associated with any entries.\n",
1136                                        NI.getUnitOffset(), NTE.getIndex(), Str);
1137                     ++NumErrors;
1138                   },
1139                   [&](const ErrorInfoBase &Info) {
1140                     error()
1141                         << formatv("Name Index @ {0:x}: Name {1} ({2}): {3}\n",
1142                                    NI.getUnitOffset(), NTE.getIndex(), Str,
1143                                    Info.message());
1144                     ++NumErrors;
1145                   });
1146   return NumErrors;
1147 }
1148 
1149 static bool isVariableIndexable(const DWARFDie &Die, DWARFContext &DCtx) {
1150   Optional<DWARFFormValue> Location = Die.findRecursively(DW_AT_location);
1151   if (!Location)
1152     return false;
1153 
1154   auto ContainsInterestingOperators = [&](StringRef D) {
1155     DWARFUnit *U = Die.getDwarfUnit();
1156     DataExtractor Data(D, DCtx.isLittleEndian(), U->getAddressByteSize());
1157     DWARFExpression Expression(Data, U->getVersion(), U->getAddressByteSize());
1158     return any_of(Expression, [](DWARFExpression::Operation &Op) {
1159       return !Op.isError() && (Op.getCode() == DW_OP_addr ||
1160                                Op.getCode() == DW_OP_form_tls_address ||
1161                                Op.getCode() == DW_OP_GNU_push_tls_address);
1162     });
1163   };
1164 
1165   if (Optional<ArrayRef<uint8_t>> Expr = Location->getAsBlock()) {
1166     // Inlined location.
1167     if (ContainsInterestingOperators(toStringRef(*Expr)))
1168       return true;
1169   } else if (Optional<uint64_t> Offset = Location->getAsSectionOffset()) {
1170     // Location list.
1171     if (const DWARFDebugLoc *DebugLoc = DCtx.getDebugLoc()) {
1172       if (const DWARFDebugLoc::LocationList *LocList =
1173               DebugLoc->getLocationListAtOffset(*Offset)) {
1174         if (any_of(LocList->Entries, [&](const DWARFDebugLoc::Entry &E) {
1175               return ContainsInterestingOperators({E.Loc.data(), E.Loc.size()});
1176             }))
1177           return true;
1178       }
1179     }
1180   }
1181   return false;
1182 }
1183 
1184 unsigned DWARFVerifier::verifyNameIndexCompleteness(
1185     const DWARFDie &Die, const DWARFDebugNames::NameIndex &NI) {
1186 
1187   // First check, if the Die should be indexed. The code follows the DWARF v5
1188   // wording as closely as possible.
1189 
1190   // "All non-defining declarations (that is, debugging information entries
1191   // with a DW_AT_declaration attribute) are excluded."
1192   if (Die.find(DW_AT_declaration))
1193     return 0;
1194 
1195   // "DW_TAG_namespace debugging information entries without a DW_AT_name
1196   // attribute are included with the name “(anonymous namespace)”.
1197   // All other debugging information entries without a DW_AT_name attribute
1198   // are excluded."
1199   // "If a subprogram or inlined subroutine is included, and has a
1200   // DW_AT_linkage_name attribute, there will be an additional index entry for
1201   // the linkage name."
1202   auto EntryNames = getNames(Die);
1203   if (EntryNames.empty())
1204     return 0;
1205 
1206   // We deviate from the specification here, which says:
1207   // "The name index must contain an entry for each debugging information entry
1208   // that defines a named subprogram, label, variable, type, or namespace,
1209   // subject to ..."
1210   // Instead whitelisting all TAGs representing a "type" or a "subprogram", to
1211   // make sure we catch any missing items, we instead blacklist all TAGs that we
1212   // know shouldn't be indexed.
1213   switch (Die.getTag()) {
1214   // Compile units and modules have names but shouldn't be indexed.
1215   case DW_TAG_compile_unit:
1216   case DW_TAG_module:
1217     return 0;
1218 
1219   // Function and template parameters are not globally visible, so we shouldn't
1220   // index them.
1221   case DW_TAG_formal_parameter:
1222   case DW_TAG_template_value_parameter:
1223   case DW_TAG_template_type_parameter:
1224   case DW_TAG_GNU_template_parameter_pack:
1225   case DW_TAG_GNU_template_template_param:
1226     return 0;
1227 
1228   // Object members aren't globally visible.
1229   case DW_TAG_member:
1230     return 0;
1231 
1232   // According to a strict reading of the specification, enumerators should not
1233   // be indexed (and LLVM currently does not do that). However, this causes
1234   // problems for the debuggers, so we may need to reconsider this.
1235   case DW_TAG_enumerator:
1236     return 0;
1237 
1238   // Imported declarations should not be indexed according to the specification
1239   // and LLVM currently does not do that.
1240   case DW_TAG_imported_declaration:
1241     return 0;
1242 
1243   // "DW_TAG_subprogram, DW_TAG_inlined_subroutine, and DW_TAG_label debugging
1244   // information entries without an address attribute (DW_AT_low_pc,
1245   // DW_AT_high_pc, DW_AT_ranges, or DW_AT_entry_pc) are excluded."
1246   case DW_TAG_subprogram:
1247   case DW_TAG_inlined_subroutine:
1248   case DW_TAG_label:
1249     if (Die.findRecursively(
1250             {DW_AT_low_pc, DW_AT_high_pc, DW_AT_ranges, DW_AT_entry_pc}))
1251       break;
1252     return 0;
1253 
1254   // "DW_TAG_variable debugging information entries with a DW_AT_location
1255   // attribute that includes a DW_OP_addr or DW_OP_form_tls_address operator are
1256   // included; otherwise, they are excluded."
1257   //
1258   // LLVM extension: We also add DW_OP_GNU_push_tls_address to this list.
1259   case DW_TAG_variable:
1260     if (isVariableIndexable(Die, DCtx))
1261       break;
1262     return 0;
1263 
1264   default:
1265     break;
1266   }
1267 
1268   // Now we know that our Die should be present in the Index. Let's check if
1269   // that's the case.
1270   unsigned NumErrors = 0;
1271   uint64_t DieUnitOffset = Die.getOffset() - Die.getDwarfUnit()->getOffset();
1272   for (StringRef Name : EntryNames) {
1273     if (none_of(NI.equal_range(Name), [&](const DWARFDebugNames::Entry &E) {
1274           return E.getDIEUnitOffset() == DieUnitOffset;
1275         })) {
1276       error() << formatv("Name Index @ {0:x}: Entry for DIE @ {1:x} ({2}) with "
1277                          "name {3} missing.\n",
1278                          NI.getUnitOffset(), Die.getOffset(), Die.getTag(),
1279                          Name);
1280       ++NumErrors;
1281     }
1282   }
1283   return NumErrors;
1284 }
1285 
1286 unsigned DWARFVerifier::verifyDebugNames(const DWARFSection &AccelSection,
1287                                          const DataExtractor &StrData) {
1288   unsigned NumErrors = 0;
1289   DWARFDataExtractor AccelSectionData(DCtx.getDWARFObj(), AccelSection,
1290                                       DCtx.isLittleEndian(), 0);
1291   DWARFDebugNames AccelTable(AccelSectionData, StrData);
1292 
1293   OS << "Verifying .debug_names...\n";
1294 
1295   // This verifies that we can read individual name indices and their
1296   // abbreviation tables.
1297   if (Error E = AccelTable.extract()) {
1298     error() << toString(std::move(E)) << '\n';
1299     return 1;
1300   }
1301 
1302   NumErrors += verifyDebugNamesCULists(AccelTable);
1303   for (const auto &NI : AccelTable)
1304     NumErrors += verifyNameIndexBuckets(NI, StrData);
1305   for (const auto &NI : AccelTable)
1306     NumErrors += verifyNameIndexAbbrevs(NI);
1307 
1308   // Don't attempt Entry validation if any of the previous checks found errors
1309   if (NumErrors > 0)
1310     return NumErrors;
1311   for (const auto &NI : AccelTable)
1312     for (DWARFDebugNames::NameTableEntry NTE : NI)
1313       NumErrors += verifyNameIndexEntries(NI, NTE);
1314 
1315   if (NumErrors > 0)
1316     return NumErrors;
1317 
1318   for (const std::unique_ptr<DWARFUnit> &U : DCtx.compile_units()) {
1319     if (const DWARFDebugNames::NameIndex *NI =
1320             AccelTable.getCUNameIndex(U->getOffset())) {
1321       auto *CU = cast<DWARFCompileUnit>(U.get());
1322       for (const DWARFDebugInfoEntry &Die : CU->dies())
1323         NumErrors += verifyNameIndexCompleteness(DWARFDie(CU, &Die), *NI);
1324     }
1325   }
1326   return NumErrors;
1327 }
1328 
1329 bool DWARFVerifier::handleAccelTables() {
1330   const DWARFObject &D = DCtx.getDWARFObj();
1331   DataExtractor StrData(D.getStringSection(), DCtx.isLittleEndian(), 0);
1332   unsigned NumErrors = 0;
1333   if (!D.getAppleNamesSection().Data.empty())
1334     NumErrors +=
1335         verifyAppleAccelTable(&D.getAppleNamesSection(), &StrData, ".apple_names");
1336   if (!D.getAppleTypesSection().Data.empty())
1337     NumErrors +=
1338         verifyAppleAccelTable(&D.getAppleTypesSection(), &StrData, ".apple_types");
1339   if (!D.getAppleNamespacesSection().Data.empty())
1340     NumErrors += verifyAppleAccelTable(&D.getAppleNamespacesSection(), &StrData,
1341                                   ".apple_namespaces");
1342   if (!D.getAppleObjCSection().Data.empty())
1343     NumErrors +=
1344         verifyAppleAccelTable(&D.getAppleObjCSection(), &StrData, ".apple_objc");
1345 
1346   if (!D.getDebugNamesSection().Data.empty())
1347     NumErrors += verifyDebugNames(D.getDebugNamesSection(), StrData);
1348   return NumErrors == 0;
1349 }
1350 
1351 raw_ostream &DWARFVerifier::error() const { return WithColor::error(OS); }
1352 
1353 raw_ostream &DWARFVerifier::warn() const { return WithColor::warning(OS); }
1354 
1355 raw_ostream &DWARFVerifier::note() const { return WithColor::note(OS); }
1356