1 //===- DWARFAcceleratorTable.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/DWARFAcceleratorTable.h"
11 
12 #include "llvm/ADT/SmallVector.h"
13 #include "llvm/BinaryFormat/Dwarf.h"
14 #include "llvm/DebugInfo/DWARF/DWARFRelocMap.h"
15 #include "llvm/Support/Compiler.h"
16 #include "llvm/Support/DJB.h"
17 #include "llvm/Support/Errc.h"
18 #include "llvm/Support/Format.h"
19 #include "llvm/Support/FormatVariadic.h"
20 #include "llvm/Support/ScopedPrinter.h"
21 #include "llvm/Support/raw_ostream.h"
22 #include <cstddef>
23 #include <cstdint>
24 #include <utility>
25 
26 using namespace llvm;
27 
28 namespace {
29 struct Atom {
30   unsigned Value;
31 };
32 
operator <<(raw_ostream & OS,const Atom & A)33 static raw_ostream &operator<<(raw_ostream &OS, const Atom &A) {
34   StringRef Str = dwarf::AtomTypeString(A.Value);
35   if (!Str.empty())
36     return OS << Str;
37   return OS << "DW_ATOM_unknown_" << format("%x", A.Value);
38 }
39 } // namespace
40 
formatAtom(unsigned Atom)41 static Atom formatAtom(unsigned Atom) { return {Atom}; }
42 
43 DWARFAcceleratorTable::~DWARFAcceleratorTable() = default;
44 
extract()45 llvm::Error AppleAcceleratorTable::extract() {
46   uint32_t Offset = 0;
47 
48   // Check that we can at least read the header.
49   if (!AccelSection.isValidOffset(offsetof(Header, HeaderDataLength) + 4))
50     return createStringError(errc::illegal_byte_sequence,
51                              "Section too small: cannot read header.");
52 
53   Hdr.Magic = AccelSection.getU32(&Offset);
54   Hdr.Version = AccelSection.getU16(&Offset);
55   Hdr.HashFunction = AccelSection.getU16(&Offset);
56   Hdr.BucketCount = AccelSection.getU32(&Offset);
57   Hdr.HashCount = AccelSection.getU32(&Offset);
58   Hdr.HeaderDataLength = AccelSection.getU32(&Offset);
59 
60   // Check that we can read all the hashes and offsets from the
61   // section (see SourceLevelDebugging.rst for the structure of the index).
62   // We need to substract one because we're checking for an *offset* which is
63   // equal to the size for an empty table and hence pointer after the section.
64   if (!AccelSection.isValidOffset(sizeof(Hdr) + Hdr.HeaderDataLength +
65                                   Hdr.BucketCount * 4 + Hdr.HashCount * 8 - 1))
66     return createStringError(
67         errc::illegal_byte_sequence,
68         "Section too small: cannot read buckets and hashes.");
69 
70   HdrData.DIEOffsetBase = AccelSection.getU32(&Offset);
71   uint32_t NumAtoms = AccelSection.getU32(&Offset);
72 
73   for (unsigned i = 0; i < NumAtoms; ++i) {
74     uint16_t AtomType = AccelSection.getU16(&Offset);
75     auto AtomForm = static_cast<dwarf::Form>(AccelSection.getU16(&Offset));
76     HdrData.Atoms.push_back(std::make_pair(AtomType, AtomForm));
77   }
78 
79   IsValid = true;
80   return Error::success();
81 }
82 
getNumBuckets()83 uint32_t AppleAcceleratorTable::getNumBuckets() { return Hdr.BucketCount; }
getNumHashes()84 uint32_t AppleAcceleratorTable::getNumHashes() { return Hdr.HashCount; }
getSizeHdr()85 uint32_t AppleAcceleratorTable::getSizeHdr() { return sizeof(Hdr); }
getHeaderDataLength()86 uint32_t AppleAcceleratorTable::getHeaderDataLength() {
87   return Hdr.HeaderDataLength;
88 }
89 
90 ArrayRef<std::pair<AppleAcceleratorTable::HeaderData::AtomType,
91                    AppleAcceleratorTable::HeaderData::Form>>
getAtomsDesc()92 AppleAcceleratorTable::getAtomsDesc() {
93   return HdrData.Atoms;
94 }
95 
validateForms()96 bool AppleAcceleratorTable::validateForms() {
97   for (auto Atom : getAtomsDesc()) {
98     DWARFFormValue FormValue(Atom.second);
99     switch (Atom.first) {
100     case dwarf::DW_ATOM_die_offset:
101     case dwarf::DW_ATOM_die_tag:
102     case dwarf::DW_ATOM_type_flags:
103       if ((!FormValue.isFormClass(DWARFFormValue::FC_Constant) &&
104            !FormValue.isFormClass(DWARFFormValue::FC_Flag)) ||
105           FormValue.getForm() == dwarf::DW_FORM_sdata)
106         return false;
107       break;
108     default:
109       break;
110     }
111   }
112   return true;
113 }
114 
115 std::pair<uint32_t, dwarf::Tag>
readAtoms(uint32_t & HashDataOffset)116 AppleAcceleratorTable::readAtoms(uint32_t &HashDataOffset) {
117   uint32_t DieOffset = dwarf::DW_INVALID_OFFSET;
118   dwarf::Tag DieTag = dwarf::DW_TAG_null;
119   dwarf::FormParams FormParams = {Hdr.Version, 0, dwarf::DwarfFormat::DWARF32};
120 
121   for (auto Atom : getAtomsDesc()) {
122     DWARFFormValue FormValue(Atom.second);
123     FormValue.extractValue(AccelSection, &HashDataOffset, FormParams);
124     switch (Atom.first) {
125     case dwarf::DW_ATOM_die_offset:
126       DieOffset = *FormValue.getAsUnsignedConstant();
127       break;
128     case dwarf::DW_ATOM_die_tag:
129       DieTag = (dwarf::Tag)*FormValue.getAsUnsignedConstant();
130       break;
131     default:
132       break;
133     }
134   }
135   return {DieOffset, DieTag};
136 }
137 
dump(ScopedPrinter & W) const138 void AppleAcceleratorTable::Header::dump(ScopedPrinter &W) const {
139   DictScope HeaderScope(W, "Header");
140   W.printHex("Magic", Magic);
141   W.printHex("Version", Version);
142   W.printHex("Hash function", HashFunction);
143   W.printNumber("Bucket count", BucketCount);
144   W.printNumber("Hashes count", HashCount);
145   W.printNumber("HeaderData length", HeaderDataLength);
146 }
147 
extractOffset(Optional<DWARFFormValue> Value) const148 Optional<uint64_t> AppleAcceleratorTable::HeaderData::extractOffset(
149     Optional<DWARFFormValue> Value) const {
150   if (!Value)
151     return None;
152 
153   switch (Value->getForm()) {
154   case dwarf::DW_FORM_ref1:
155   case dwarf::DW_FORM_ref2:
156   case dwarf::DW_FORM_ref4:
157   case dwarf::DW_FORM_ref8:
158   case dwarf::DW_FORM_ref_udata:
159     return Value->getRawUValue() + DIEOffsetBase;
160   default:
161     return Value->getAsSectionOffset();
162   }
163 }
164 
dumpName(ScopedPrinter & W,SmallVectorImpl<DWARFFormValue> & AtomForms,uint32_t * DataOffset) const165 bool AppleAcceleratorTable::dumpName(ScopedPrinter &W,
166                                      SmallVectorImpl<DWARFFormValue> &AtomForms,
167                                      uint32_t *DataOffset) const {
168   dwarf::FormParams FormParams = {Hdr.Version, 0, dwarf::DwarfFormat::DWARF32};
169   uint32_t NameOffset = *DataOffset;
170   if (!AccelSection.isValidOffsetForDataOfSize(*DataOffset, 4)) {
171     W.printString("Incorrectly terminated list.");
172     return false;
173   }
174   unsigned StringOffset = AccelSection.getRelocatedValue(4, DataOffset);
175   if (!StringOffset)
176     return false; // End of list
177 
178   DictScope NameScope(W, ("Name@0x" + Twine::utohexstr(NameOffset)).str());
179   W.startLine() << format("String: 0x%08x", StringOffset);
180   W.getOStream() << " \"" << StringSection.getCStr(&StringOffset) << "\"\n";
181 
182   unsigned NumData = AccelSection.getU32(DataOffset);
183   for (unsigned Data = 0; Data < NumData; ++Data) {
184     ListScope DataScope(W, ("Data " + Twine(Data)).str());
185     unsigned i = 0;
186     for (auto &Atom : AtomForms) {
187       W.startLine() << format("Atom[%d]: ", i);
188       if (Atom.extractValue(AccelSection, DataOffset, FormParams)) {
189         Atom.dump(W.getOStream());
190         if (Optional<uint64_t> Val = Atom.getAsUnsignedConstant()) {
191           StringRef Str = dwarf::AtomValueString(HdrData.Atoms[i].first, *Val);
192           if (!Str.empty())
193             W.getOStream() << " (" << Str << ")";
194         }
195       } else
196         W.getOStream() << "Error extracting the value";
197       W.getOStream() << "\n";
198       i++;
199     }
200   }
201   return true; // more entries follow
202 }
203 
dump(raw_ostream & OS) const204 LLVM_DUMP_METHOD void AppleAcceleratorTable::dump(raw_ostream &OS) const {
205   if (!IsValid)
206     return;
207 
208   ScopedPrinter W(OS);
209 
210   Hdr.dump(W);
211 
212   W.printNumber("DIE offset base", HdrData.DIEOffsetBase);
213   W.printNumber("Number of atoms", uint64_t(HdrData.Atoms.size()));
214   SmallVector<DWARFFormValue, 3> AtomForms;
215   {
216     ListScope AtomsScope(W, "Atoms");
217     unsigned i = 0;
218     for (const auto &Atom : HdrData.Atoms) {
219       DictScope AtomScope(W, ("Atom " + Twine(i++)).str());
220       W.startLine() << "Type: " << formatAtom(Atom.first) << '\n';
221       W.startLine() << "Form: " << formatv("{0}", Atom.second) << '\n';
222       AtomForms.push_back(DWARFFormValue(Atom.second));
223     }
224   }
225 
226   // Now go through the actual tables and dump them.
227   uint32_t Offset = sizeof(Hdr) + Hdr.HeaderDataLength;
228   unsigned HashesBase = Offset + Hdr.BucketCount * 4;
229   unsigned OffsetsBase = HashesBase + Hdr.HashCount * 4;
230 
231   for (unsigned Bucket = 0; Bucket < Hdr.BucketCount; ++Bucket) {
232     unsigned Index = AccelSection.getU32(&Offset);
233 
234     ListScope BucketScope(W, ("Bucket " + Twine(Bucket)).str());
235     if (Index == UINT32_MAX) {
236       W.printString("EMPTY");
237       continue;
238     }
239 
240     for (unsigned HashIdx = Index; HashIdx < Hdr.HashCount; ++HashIdx) {
241       unsigned HashOffset = HashesBase + HashIdx*4;
242       unsigned OffsetsOffset = OffsetsBase + HashIdx*4;
243       uint32_t Hash = AccelSection.getU32(&HashOffset);
244 
245       if (Hash % Hdr.BucketCount != Bucket)
246         break;
247 
248       unsigned DataOffset = AccelSection.getU32(&OffsetsOffset);
249       ListScope HashScope(W, ("Hash 0x" + Twine::utohexstr(Hash)).str());
250       if (!AccelSection.isValidOffset(DataOffset)) {
251         W.printString("Invalid section offset");
252         continue;
253       }
254       while (dumpName(W, AtomForms, &DataOffset))
255         /*empty*/;
256     }
257   }
258 }
259 
Entry(const AppleAcceleratorTable::HeaderData & HdrData)260 AppleAcceleratorTable::Entry::Entry(
261     const AppleAcceleratorTable::HeaderData &HdrData)
262     : HdrData(&HdrData) {
263   Values.reserve(HdrData.Atoms.size());
264   for (const auto &Atom : HdrData.Atoms)
265     Values.push_back(DWARFFormValue(Atom.second));
266 }
267 
extract(const AppleAcceleratorTable & AccelTable,uint32_t * Offset)268 void AppleAcceleratorTable::Entry::extract(
269     const AppleAcceleratorTable &AccelTable, uint32_t *Offset) {
270 
271   dwarf::FormParams FormParams = {AccelTable.Hdr.Version, 0,
272                                   dwarf::DwarfFormat::DWARF32};
273   for (auto &Atom : Values)
274     Atom.extractValue(AccelTable.AccelSection, Offset, FormParams);
275 }
276 
277 Optional<DWARFFormValue>
lookup(HeaderData::AtomType Atom) const278 AppleAcceleratorTable::Entry::lookup(HeaderData::AtomType Atom) const {
279   assert(HdrData && "Dereferencing end iterator?");
280   assert(HdrData->Atoms.size() == Values.size());
281   for (const auto &Tuple : zip_first(HdrData->Atoms, Values)) {
282     if (std::get<0>(Tuple).first == Atom)
283       return std::get<1>(Tuple);
284   }
285   return None;
286 }
287 
getDIESectionOffset() const288 Optional<uint64_t> AppleAcceleratorTable::Entry::getDIESectionOffset() const {
289   return HdrData->extractOffset(lookup(dwarf::DW_ATOM_die_offset));
290 }
291 
getCUOffset() const292 Optional<uint64_t> AppleAcceleratorTable::Entry::getCUOffset() const {
293   return HdrData->extractOffset(lookup(dwarf::DW_ATOM_cu_offset));
294 }
295 
getTag() const296 Optional<dwarf::Tag> AppleAcceleratorTable::Entry::getTag() const {
297   Optional<DWARFFormValue> Tag = lookup(dwarf::DW_ATOM_die_tag);
298   if (!Tag)
299     return None;
300   if (Optional<uint64_t> Value = Tag->getAsUnsignedConstant())
301     return dwarf::Tag(*Value);
302   return None;
303 }
304 
ValueIterator(const AppleAcceleratorTable & AccelTable,unsigned Offset)305 AppleAcceleratorTable::ValueIterator::ValueIterator(
306     const AppleAcceleratorTable &AccelTable, unsigned Offset)
307     : AccelTable(&AccelTable), Current(AccelTable.HdrData), DataOffset(Offset) {
308   if (!AccelTable.AccelSection.isValidOffsetForDataOfSize(DataOffset, 4))
309     return;
310 
311   // Read the first entry.
312   NumData = AccelTable.AccelSection.getU32(&DataOffset);
313   Next();
314 }
315 
Next()316 void AppleAcceleratorTable::ValueIterator::Next() {
317   assert(NumData > 0 && "attempted to increment iterator past the end");
318   auto &AccelSection = AccelTable->AccelSection;
319   if (Data >= NumData ||
320       !AccelSection.isValidOffsetForDataOfSize(DataOffset, 4)) {
321     NumData = 0;
322     DataOffset = 0;
323     return;
324   }
325   Current.extract(*AccelTable, &DataOffset);
326   ++Data;
327 }
328 
329 iterator_range<AppleAcceleratorTable::ValueIterator>
equal_range(StringRef Key) const330 AppleAcceleratorTable::equal_range(StringRef Key) const {
331   if (!IsValid)
332     return make_range(ValueIterator(), ValueIterator());
333 
334   // Find the bucket.
335   unsigned HashValue = djbHash(Key);
336   unsigned Bucket = HashValue % Hdr.BucketCount;
337   unsigned BucketBase = sizeof(Hdr) + Hdr.HeaderDataLength;
338   unsigned HashesBase = BucketBase + Hdr.BucketCount * 4;
339   unsigned OffsetsBase = HashesBase + Hdr.HashCount * 4;
340 
341   unsigned BucketOffset = BucketBase + Bucket * 4;
342   unsigned Index = AccelSection.getU32(&BucketOffset);
343 
344   // Search through all hashes in the bucket.
345   for (unsigned HashIdx = Index; HashIdx < Hdr.HashCount; ++HashIdx) {
346     unsigned HashOffset = HashesBase + HashIdx * 4;
347     unsigned OffsetsOffset = OffsetsBase + HashIdx * 4;
348     uint32_t Hash = AccelSection.getU32(&HashOffset);
349 
350     if (Hash % Hdr.BucketCount != Bucket)
351       // We are already in the next bucket.
352       break;
353 
354     unsigned DataOffset = AccelSection.getU32(&OffsetsOffset);
355     unsigned StringOffset = AccelSection.getRelocatedValue(4, &DataOffset);
356     if (!StringOffset)
357       break;
358 
359     // Finally, compare the key.
360     if (Key == StringSection.getCStr(&StringOffset))
361       return make_range({*this, DataOffset}, ValueIterator());
362   }
363   return make_range(ValueIterator(), ValueIterator());
364 }
365 
dump(ScopedPrinter & W) const366 void DWARFDebugNames::Header::dump(ScopedPrinter &W) const {
367   DictScope HeaderScope(W, "Header");
368   W.printHex("Length", UnitLength);
369   W.printNumber("Version", Version);
370   W.printHex("Padding", Padding);
371   W.printNumber("CU count", CompUnitCount);
372   W.printNumber("Local TU count", LocalTypeUnitCount);
373   W.printNumber("Foreign TU count", ForeignTypeUnitCount);
374   W.printNumber("Bucket count", BucketCount);
375   W.printNumber("Name count", NameCount);
376   W.printHex("Abbreviations table size", AbbrevTableSize);
377   W.startLine() << "Augmentation: '" << AugmentationString << "'\n";
378 }
379 
extract(const DWARFDataExtractor & AS,uint32_t * Offset)380 llvm::Error DWARFDebugNames::Header::extract(const DWARFDataExtractor &AS,
381                                              uint32_t *Offset) {
382   // Check that we can read the fixed-size part.
383   if (!AS.isValidOffset(*Offset + sizeof(HeaderPOD) - 1))
384     return createStringError(errc::illegal_byte_sequence,
385                              "Section too small: cannot read header.");
386 
387   UnitLength = AS.getU32(Offset);
388   Version = AS.getU16(Offset);
389   Padding = AS.getU16(Offset);
390   CompUnitCount = AS.getU32(Offset);
391   LocalTypeUnitCount = AS.getU32(Offset);
392   ForeignTypeUnitCount = AS.getU32(Offset);
393   BucketCount = AS.getU32(Offset);
394   NameCount = AS.getU32(Offset);
395   AbbrevTableSize = AS.getU32(Offset);
396   AugmentationStringSize = alignTo(AS.getU32(Offset), 4);
397 
398   if (!AS.isValidOffsetForDataOfSize(*Offset, AugmentationStringSize))
399     return createStringError(
400         errc::illegal_byte_sequence,
401         "Section too small: cannot read header augmentation.");
402   AugmentationString.resize(AugmentationStringSize);
403   AS.getU8(Offset, reinterpret_cast<uint8_t *>(AugmentationString.data()),
404            AugmentationStringSize);
405   return Error::success();
406 }
407 
dump(ScopedPrinter & W) const408 void DWARFDebugNames::Abbrev::dump(ScopedPrinter &W) const {
409   DictScope AbbrevScope(W, ("Abbreviation 0x" + Twine::utohexstr(Code)).str());
410   W.startLine() << formatv("Tag: {0}\n", Tag);
411 
412   for (const auto &Attr : Attributes)
413     W.startLine() << formatv("{0}: {1}\n", Attr.Index, Attr.Form);
414 }
415 
sentinelAttrEnc()416 static constexpr DWARFDebugNames::AttributeEncoding sentinelAttrEnc() {
417   return {dwarf::Index(0), dwarf::Form(0)};
418 }
419 
isSentinel(const DWARFDebugNames::AttributeEncoding & AE)420 static bool isSentinel(const DWARFDebugNames::AttributeEncoding &AE) {
421   return AE == sentinelAttrEnc();
422 }
423 
sentinelAbbrev()424 static DWARFDebugNames::Abbrev sentinelAbbrev() {
425   return DWARFDebugNames::Abbrev(0, dwarf::Tag(0), {});
426 }
427 
isSentinel(const DWARFDebugNames::Abbrev & Abbr)428 static bool isSentinel(const DWARFDebugNames::Abbrev &Abbr) {
429   return Abbr.Code == 0;
430 }
431 
getEmptyKey()432 DWARFDebugNames::Abbrev DWARFDebugNames::AbbrevMapInfo::getEmptyKey() {
433   return sentinelAbbrev();
434 }
435 
getTombstoneKey()436 DWARFDebugNames::Abbrev DWARFDebugNames::AbbrevMapInfo::getTombstoneKey() {
437   return DWARFDebugNames::Abbrev(~0, dwarf::Tag(0), {});
438 }
439 
440 Expected<DWARFDebugNames::AttributeEncoding>
extractAttributeEncoding(uint32_t * Offset)441 DWARFDebugNames::NameIndex::extractAttributeEncoding(uint32_t *Offset) {
442   if (*Offset >= EntriesBase) {
443     return createStringError(errc::illegal_byte_sequence,
444                              "Incorrectly terminated abbreviation table.");
445   }
446 
447   uint32_t Index = Section.AccelSection.getULEB128(Offset);
448   uint32_t Form = Section.AccelSection.getULEB128(Offset);
449   return AttributeEncoding(dwarf::Index(Index), dwarf::Form(Form));
450 }
451 
452 Expected<std::vector<DWARFDebugNames::AttributeEncoding>>
extractAttributeEncodings(uint32_t * Offset)453 DWARFDebugNames::NameIndex::extractAttributeEncodings(uint32_t *Offset) {
454   std::vector<AttributeEncoding> Result;
455   for (;;) {
456     auto AttrEncOr = extractAttributeEncoding(Offset);
457     if (!AttrEncOr)
458       return AttrEncOr.takeError();
459     if (isSentinel(*AttrEncOr))
460       return std::move(Result);
461 
462     Result.emplace_back(*AttrEncOr);
463   }
464 }
465 
466 Expected<DWARFDebugNames::Abbrev>
extractAbbrev(uint32_t * Offset)467 DWARFDebugNames::NameIndex::extractAbbrev(uint32_t *Offset) {
468   if (*Offset >= EntriesBase) {
469     return createStringError(errc::illegal_byte_sequence,
470                              "Incorrectly terminated abbreviation table.");
471   }
472 
473   uint32_t Code = Section.AccelSection.getULEB128(Offset);
474   if (Code == 0)
475     return sentinelAbbrev();
476 
477   uint32_t Tag = Section.AccelSection.getULEB128(Offset);
478   auto AttrEncOr = extractAttributeEncodings(Offset);
479   if (!AttrEncOr)
480     return AttrEncOr.takeError();
481   return Abbrev(Code, dwarf::Tag(Tag), std::move(*AttrEncOr));
482 }
483 
extract()484 Error DWARFDebugNames::NameIndex::extract() {
485   const DWARFDataExtractor &AS = Section.AccelSection;
486   uint32_t Offset = Base;
487   if (Error E = Hdr.extract(AS, &Offset))
488     return E;
489 
490   CUsBase = Offset;
491   Offset += Hdr.CompUnitCount * 4;
492   Offset += Hdr.LocalTypeUnitCount * 4;
493   Offset += Hdr.ForeignTypeUnitCount * 8;
494   BucketsBase = Offset;
495   Offset += Hdr.BucketCount * 4;
496   HashesBase = Offset;
497   if (Hdr.BucketCount > 0)
498     Offset += Hdr.NameCount * 4;
499   StringOffsetsBase = Offset;
500   Offset += Hdr.NameCount * 4;
501   EntryOffsetsBase = Offset;
502   Offset += Hdr.NameCount * 4;
503 
504   if (!AS.isValidOffsetForDataOfSize(Offset, Hdr.AbbrevTableSize))
505     return createStringError(errc::illegal_byte_sequence,
506                              "Section too small: cannot read abbreviations.");
507 
508   EntriesBase = Offset + Hdr.AbbrevTableSize;
509 
510   for (;;) {
511     auto AbbrevOr = extractAbbrev(&Offset);
512     if (!AbbrevOr)
513       return AbbrevOr.takeError();
514     if (isSentinel(*AbbrevOr))
515       return Error::success();
516 
517     if (!Abbrevs.insert(std::move(*AbbrevOr)).second)
518       return createStringError(errc::invalid_argument,
519                                "Duplicate abbreviation code.");
520   }
521 }
Entry(const NameIndex & NameIdx,const Abbrev & Abbr)522 DWARFDebugNames::Entry::Entry(const NameIndex &NameIdx, const Abbrev &Abbr)
523     : NameIdx(&NameIdx), Abbr(&Abbr) {
524   // This merely creates form values. It is up to the caller
525   // (NameIndex::getEntry) to populate them.
526   Values.reserve(Abbr.Attributes.size());
527   for (const auto &Attr : Abbr.Attributes)
528     Values.emplace_back(Attr.Form);
529 }
530 
531 Optional<DWARFFormValue>
lookup(dwarf::Index Index) const532 DWARFDebugNames::Entry::lookup(dwarf::Index Index) const {
533   assert(Abbr->Attributes.size() == Values.size());
534   for (const auto &Tuple : zip_first(Abbr->Attributes, Values)) {
535     if (std::get<0>(Tuple).Index == Index)
536       return std::get<1>(Tuple);
537   }
538   return None;
539 }
540 
getDIEUnitOffset() const541 Optional<uint64_t> DWARFDebugNames::Entry::getDIEUnitOffset() const {
542   if (Optional<DWARFFormValue> Off = lookup(dwarf::DW_IDX_die_offset))
543     return Off->getAsReferenceUVal();
544   return None;
545 }
546 
getCUIndex() const547 Optional<uint64_t> DWARFDebugNames::Entry::getCUIndex() const {
548   if (Optional<DWARFFormValue> Off = lookup(dwarf::DW_IDX_compile_unit))
549     return Off->getAsUnsignedConstant();
550   // In a per-CU index, the entries without a DW_IDX_compile_unit attribute
551   // implicitly refer to the single CU.
552   if (NameIdx->getCUCount() == 1)
553     return 0;
554   return None;
555 }
556 
getCUOffset() const557 Optional<uint64_t> DWARFDebugNames::Entry::getCUOffset() const {
558   Optional<uint64_t> Index = getCUIndex();
559   if (!Index || *Index >= NameIdx->getCUCount())
560     return None;
561   return NameIdx->getCUOffset(*Index);
562 }
563 
dump(ScopedPrinter & W) const564 void DWARFDebugNames::Entry::dump(ScopedPrinter &W) const {
565   W.printHex("Abbrev", Abbr->Code);
566   W.startLine() << formatv("Tag: {0}\n", Abbr->Tag);
567   assert(Abbr->Attributes.size() == Values.size());
568   for (const auto &Tuple : zip_first(Abbr->Attributes, Values)) {
569     W.startLine() << formatv("{0}: ", std::get<0>(Tuple).Index);
570     std::get<1>(Tuple).dump(W.getOStream());
571     W.getOStream() << '\n';
572   }
573 }
574 
575 char DWARFDebugNames::SentinelError::ID;
convertToErrorCode() const576 std::error_code DWARFDebugNames::SentinelError::convertToErrorCode() const {
577   return inconvertibleErrorCode();
578 }
579 
getCUOffset(uint32_t CU) const580 uint32_t DWARFDebugNames::NameIndex::getCUOffset(uint32_t CU) const {
581   assert(CU < Hdr.CompUnitCount);
582   uint32_t Offset = CUsBase + 4 * CU;
583   return Section.AccelSection.getRelocatedValue(4, &Offset);
584 }
585 
getLocalTUOffset(uint32_t TU) const586 uint32_t DWARFDebugNames::NameIndex::getLocalTUOffset(uint32_t TU) const {
587   assert(TU < Hdr.LocalTypeUnitCount);
588   uint32_t Offset = CUsBase + Hdr.CompUnitCount * 4;
589   return Section.AccelSection.getRelocatedValue(4, &Offset);
590 }
591 
getForeignTUSignature(uint32_t TU) const592 uint64_t DWARFDebugNames::NameIndex::getForeignTUSignature(uint32_t TU) const {
593   assert(TU < Hdr.ForeignTypeUnitCount);
594   uint32_t Offset = CUsBase + (Hdr.CompUnitCount + Hdr.LocalTypeUnitCount) * 4;
595   return Section.AccelSection.getU64(&Offset);
596 }
597 
598 Expected<DWARFDebugNames::Entry>
getEntry(uint32_t * Offset) const599 DWARFDebugNames::NameIndex::getEntry(uint32_t *Offset) const {
600   const DWARFDataExtractor &AS = Section.AccelSection;
601   if (!AS.isValidOffset(*Offset))
602     return createStringError(errc::illegal_byte_sequence,
603                              "Incorrectly terminated entry list.");
604 
605   uint32_t AbbrevCode = AS.getULEB128(Offset);
606   if (AbbrevCode == 0)
607     return make_error<SentinelError>();
608 
609   const auto AbbrevIt = Abbrevs.find_as(AbbrevCode);
610   if (AbbrevIt == Abbrevs.end())
611     return createStringError(errc::invalid_argument, "Invalid abbreviation.");
612 
613   Entry E(*this, *AbbrevIt);
614 
615   dwarf::FormParams FormParams = {Hdr.Version, 0, dwarf::DwarfFormat::DWARF32};
616   for (auto &Value : E.Values) {
617     if (!Value.extractValue(AS, Offset, FormParams))
618       return createStringError(errc::io_error,
619                                "Error extracting index attribute values.");
620   }
621   return std::move(E);
622 }
623 
624 DWARFDebugNames::NameTableEntry
getNameTableEntry(uint32_t Index) const625 DWARFDebugNames::NameIndex::getNameTableEntry(uint32_t Index) const {
626   assert(0 < Index && Index <= Hdr.NameCount);
627   uint32_t StringOffsetOffset = StringOffsetsBase + 4 * (Index - 1);
628   uint32_t EntryOffsetOffset = EntryOffsetsBase + 4 * (Index - 1);
629   const DWARFDataExtractor &AS = Section.AccelSection;
630 
631   uint32_t StringOffset = AS.getRelocatedValue(4, &StringOffsetOffset);
632   uint32_t EntryOffset = AS.getU32(&EntryOffsetOffset);
633   EntryOffset += EntriesBase;
634   return {Section.StringSection, Index, StringOffset, EntryOffset};
635 }
636 
637 uint32_t
getBucketArrayEntry(uint32_t Bucket) const638 DWARFDebugNames::NameIndex::getBucketArrayEntry(uint32_t Bucket) const {
639   assert(Bucket < Hdr.BucketCount);
640   uint32_t BucketOffset = BucketsBase + 4 * Bucket;
641   return Section.AccelSection.getU32(&BucketOffset);
642 }
643 
getHashArrayEntry(uint32_t Index) const644 uint32_t DWARFDebugNames::NameIndex::getHashArrayEntry(uint32_t Index) const {
645   assert(0 < Index && Index <= Hdr.NameCount);
646   uint32_t HashOffset = HashesBase + 4 * (Index - 1);
647   return Section.AccelSection.getU32(&HashOffset);
648 }
649 
650 // Returns true if we should continue scanning for entries, false if this is the
651 // last (sentinel) entry). In case of a parsing error we also return false, as
652 // it's not possible to recover this entry list (but the other lists may still
653 // parse OK).
dumpEntry(ScopedPrinter & W,uint32_t * Offset) const654 bool DWARFDebugNames::NameIndex::dumpEntry(ScopedPrinter &W,
655                                            uint32_t *Offset) const {
656   uint32_t EntryId = *Offset;
657   auto EntryOr = getEntry(Offset);
658   if (!EntryOr) {
659     handleAllErrors(EntryOr.takeError(), [](const SentinelError &) {},
660                     [&W](const ErrorInfoBase &EI) { EI.log(W.startLine()); });
661     return false;
662   }
663 
664   DictScope EntryScope(W, ("Entry @ 0x" + Twine::utohexstr(EntryId)).str());
665   EntryOr->dump(W);
666   return true;
667 }
668 
dumpName(ScopedPrinter & W,const NameTableEntry & NTE,Optional<uint32_t> Hash) const669 void DWARFDebugNames::NameIndex::dumpName(ScopedPrinter &W,
670                                           const NameTableEntry &NTE,
671                                           Optional<uint32_t> Hash) const {
672   DictScope NameScope(W, ("Name " + Twine(NTE.getIndex())).str());
673   if (Hash)
674     W.printHex("Hash", *Hash);
675 
676   W.startLine() << format("String: 0x%08x", NTE.getStringOffset());
677   W.getOStream() << " \"" << NTE.getString() << "\"\n";
678 
679   uint32_t EntryOffset = NTE.getEntryOffset();
680   while (dumpEntry(W, &EntryOffset))
681     /*empty*/;
682 }
683 
dumpCUs(ScopedPrinter & W) const684 void DWARFDebugNames::NameIndex::dumpCUs(ScopedPrinter &W) const {
685   ListScope CUScope(W, "Compilation Unit offsets");
686   for (uint32_t CU = 0; CU < Hdr.CompUnitCount; ++CU)
687     W.startLine() << format("CU[%u]: 0x%08x\n", CU, getCUOffset(CU));
688 }
689 
dumpLocalTUs(ScopedPrinter & W) const690 void DWARFDebugNames::NameIndex::dumpLocalTUs(ScopedPrinter &W) const {
691   if (Hdr.LocalTypeUnitCount == 0)
692     return;
693 
694   ListScope TUScope(W, "Local Type Unit offsets");
695   for (uint32_t TU = 0; TU < Hdr.LocalTypeUnitCount; ++TU)
696     W.startLine() << format("LocalTU[%u]: 0x%08x\n", TU, getLocalTUOffset(TU));
697 }
698 
dumpForeignTUs(ScopedPrinter & W) const699 void DWARFDebugNames::NameIndex::dumpForeignTUs(ScopedPrinter &W) const {
700   if (Hdr.ForeignTypeUnitCount == 0)
701     return;
702 
703   ListScope TUScope(W, "Foreign Type Unit signatures");
704   for (uint32_t TU = 0; TU < Hdr.ForeignTypeUnitCount; ++TU) {
705     W.startLine() << format("ForeignTU[%u]: 0x%016" PRIx64 "\n", TU,
706                             getForeignTUSignature(TU));
707   }
708 }
709 
dumpAbbreviations(ScopedPrinter & W) const710 void DWARFDebugNames::NameIndex::dumpAbbreviations(ScopedPrinter &W) const {
711   ListScope AbbrevsScope(W, "Abbreviations");
712   for (const auto &Abbr : Abbrevs)
713     Abbr.dump(W);
714 }
715 
dumpBucket(ScopedPrinter & W,uint32_t Bucket) const716 void DWARFDebugNames::NameIndex::dumpBucket(ScopedPrinter &W,
717                                             uint32_t Bucket) const {
718   ListScope BucketScope(W, ("Bucket " + Twine(Bucket)).str());
719   uint32_t Index = getBucketArrayEntry(Bucket);
720   if (Index == 0) {
721     W.printString("EMPTY");
722     return;
723   }
724   if (Index > Hdr.NameCount) {
725     W.printString("Name index is invalid");
726     return;
727   }
728 
729   for (; Index <= Hdr.NameCount; ++Index) {
730     uint32_t Hash = getHashArrayEntry(Index);
731     if (Hash % Hdr.BucketCount != Bucket)
732       break;
733 
734     dumpName(W, getNameTableEntry(Index), Hash);
735   }
736 }
737 
dump(ScopedPrinter & W) const738 LLVM_DUMP_METHOD void DWARFDebugNames::NameIndex::dump(ScopedPrinter &W) const {
739   DictScope UnitScope(W, ("Name Index @ 0x" + Twine::utohexstr(Base)).str());
740   Hdr.dump(W);
741   dumpCUs(W);
742   dumpLocalTUs(W);
743   dumpForeignTUs(W);
744   dumpAbbreviations(W);
745 
746   if (Hdr.BucketCount > 0) {
747     for (uint32_t Bucket = 0; Bucket < Hdr.BucketCount; ++Bucket)
748       dumpBucket(W, Bucket);
749     return;
750   }
751 
752   W.startLine() << "Hash table not present\n";
753   for (NameTableEntry NTE : *this)
754     dumpName(W, NTE, None);
755 }
756 
extract()757 llvm::Error DWARFDebugNames::extract() {
758   uint32_t Offset = 0;
759   while (AccelSection.isValidOffset(Offset)) {
760     NameIndex Next(*this, Offset);
761     if (llvm::Error E = Next.extract())
762       return E;
763     Offset = Next.getNextUnitOffset();
764     NameIndices.push_back(std::move(Next));
765   }
766   return Error::success();
767 }
768 
769 iterator_range<DWARFDebugNames::ValueIterator>
equal_range(StringRef Key) const770 DWARFDebugNames::NameIndex::equal_range(StringRef Key) const {
771   return make_range(ValueIterator(*this, Key), ValueIterator());
772 }
773 
dump(raw_ostream & OS) const774 LLVM_DUMP_METHOD void DWARFDebugNames::dump(raw_ostream &OS) const {
775   ScopedPrinter W(OS);
776   for (const NameIndex &NI : NameIndices)
777     NI.dump(W);
778 }
779 
780 Optional<uint32_t>
findEntryOffsetInCurrentIndex()781 DWARFDebugNames::ValueIterator::findEntryOffsetInCurrentIndex() {
782   const Header &Hdr = CurrentIndex->Hdr;
783   if (Hdr.BucketCount == 0) {
784     // No Hash Table, We need to search through all names in the Name Index.
785     for (NameTableEntry NTE : *CurrentIndex) {
786       if (NTE.getString() == Key)
787         return NTE.getEntryOffset();
788     }
789     return None;
790   }
791 
792   // The Name Index has a Hash Table, so use that to speed up the search.
793   // Compute the Key Hash, if it has not been done already.
794   if (!Hash)
795     Hash = caseFoldingDjbHash(Key);
796   uint32_t Bucket = *Hash % Hdr.BucketCount;
797   uint32_t Index = CurrentIndex->getBucketArrayEntry(Bucket);
798   if (Index == 0)
799     return None; // Empty bucket
800 
801   for (; Index <= Hdr.NameCount; ++Index) {
802     uint32_t Hash = CurrentIndex->getHashArrayEntry(Index);
803     if (Hash % Hdr.BucketCount != Bucket)
804       return None; // End of bucket
805 
806     NameTableEntry NTE = CurrentIndex->getNameTableEntry(Index);
807     if (NTE.getString() == Key)
808       return NTE.getEntryOffset();
809   }
810   return None;
811 }
812 
getEntryAtCurrentOffset()813 bool DWARFDebugNames::ValueIterator::getEntryAtCurrentOffset() {
814   auto EntryOr = CurrentIndex->getEntry(&DataOffset);
815   if (!EntryOr) {
816     consumeError(EntryOr.takeError());
817     return false;
818   }
819   CurrentEntry = std::move(*EntryOr);
820   return true;
821 }
822 
findInCurrentIndex()823 bool DWARFDebugNames::ValueIterator::findInCurrentIndex() {
824   Optional<uint32_t> Offset = findEntryOffsetInCurrentIndex();
825   if (!Offset)
826     return false;
827   DataOffset = *Offset;
828   return getEntryAtCurrentOffset();
829 }
830 
searchFromStartOfCurrentIndex()831 void DWARFDebugNames::ValueIterator::searchFromStartOfCurrentIndex() {
832   for (const NameIndex *End = CurrentIndex->Section.NameIndices.end();
833        CurrentIndex != End; ++CurrentIndex) {
834     if (findInCurrentIndex())
835       return;
836   }
837   setEnd();
838 }
839 
next()840 void DWARFDebugNames::ValueIterator::next() {
841   assert(CurrentIndex && "Incrementing an end() iterator?");
842 
843   // First try the next entry in the current Index.
844   if (getEntryAtCurrentOffset())
845     return;
846 
847   // If we're a local iterator or we have reached the last Index, we're done.
848   if (IsLocal || CurrentIndex == &CurrentIndex->Section.NameIndices.back()) {
849     setEnd();
850     return;
851   }
852 
853   // Otherwise, try the next index.
854   ++CurrentIndex;
855   searchFromStartOfCurrentIndex();
856 }
857 
ValueIterator(const DWARFDebugNames & AccelTable,StringRef Key)858 DWARFDebugNames::ValueIterator::ValueIterator(const DWARFDebugNames &AccelTable,
859                                               StringRef Key)
860     : CurrentIndex(AccelTable.NameIndices.begin()), IsLocal(false), Key(Key) {
861   searchFromStartOfCurrentIndex();
862 }
863 
ValueIterator(const DWARFDebugNames::NameIndex & NI,StringRef Key)864 DWARFDebugNames::ValueIterator::ValueIterator(
865     const DWARFDebugNames::NameIndex &NI, StringRef Key)
866     : CurrentIndex(&NI), IsLocal(true), Key(Key) {
867   if (!findInCurrentIndex())
868     setEnd();
869 }
870 
871 iterator_range<DWARFDebugNames::ValueIterator>
equal_range(StringRef Key) const872 DWARFDebugNames::equal_range(StringRef Key) const {
873   if (NameIndices.empty())
874     return make_range(ValueIterator(), ValueIterator());
875   return make_range(ValueIterator(*this, Key), ValueIterator());
876 }
877 
878 const DWARFDebugNames::NameIndex *
getCUNameIndex(uint32_t CUOffset)879 DWARFDebugNames::getCUNameIndex(uint32_t CUOffset) {
880   if (CUToNameIndex.size() == 0 && NameIndices.size() > 0) {
881     for (const auto &NI : *this) {
882       for (uint32_t CU = 0; CU < NI.getCUCount(); ++CU)
883         CUToNameIndex.try_emplace(NI.getCUOffset(CU), &NI);
884     }
885   }
886   return CUToNameIndex.lookup(CUOffset);
887 }
888