1 //===- InputFiles.cpp -----------------------------------------------------===//
2 //
3 //                             The LLVM Linker
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 "InputFiles.h"
11 #include "InputSection.h"
12 #include "LinkerScript.h"
13 #include "SymbolTable.h"
14 #include "Symbols.h"
15 #include "SyntheticSections.h"
16 #include "lld/Common/ErrorHandler.h"
17 #include "lld/Common/Memory.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/CodeGen/Analysis.h"
20 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
21 #include "llvm/IR/LLVMContext.h"
22 #include "llvm/IR/Module.h"
23 #include "llvm/LTO/LTO.h"
24 #include "llvm/MC/StringTableBuilder.h"
25 #include "llvm/Object/ELFObjectFile.h"
26 #include "llvm/Support/ARMAttributeParser.h"
27 #include "llvm/Support/ARMBuildAttributes.h"
28 #include "llvm/Support/Path.h"
29 #include "llvm/Support/TarWriter.h"
30 #include "llvm/Support/raw_ostream.h"
31 
32 using namespace llvm;
33 using namespace llvm::ELF;
34 using namespace llvm::object;
35 using namespace llvm::sys;
36 using namespace llvm::sys::fs;
37 
38 using namespace lld;
39 using namespace lld::elf;
40 
41 bool InputFile::IsInGroup;
42 uint32_t InputFile::NextGroupId;
43 std::vector<BinaryFile *> elf::BinaryFiles;
44 std::vector<BitcodeFile *> elf::BitcodeFiles;
45 std::vector<InputFile *> elf::ObjectFiles;
46 std::vector<InputFile *> elf::SharedFiles;
47 
48 TarWriter *elf::Tar;
49 
50 InputFile::InputFile(Kind K, MemoryBufferRef M)
51     : MB(M), GroupId(NextGroupId), FileKind(K) {
52   // All files within the same --{start,end}-group get the same group ID.
53   // Otherwise, a new file will get a new group ID.
54   if (!IsInGroup)
55     ++NextGroupId;
56 }
57 
58 Optional<MemoryBufferRef> elf::readFile(StringRef Path) {
59   // The --chroot option changes our virtual root directory.
60   // This is useful when you are dealing with files created by --reproduce.
61   if (!Config->Chroot.empty() && Path.startswith("/"))
62     Path = Saver.save(Config->Chroot + Path);
63 
64   log(Path);
65 
66   auto MBOrErr = MemoryBuffer::getFile(Path);
67   if (auto EC = MBOrErr.getError()) {
68     error("cannot open " + Path + ": " + EC.message());
69     return None;
70   }
71 
72   std::unique_ptr<MemoryBuffer> &MB = *MBOrErr;
73   MemoryBufferRef MBRef = MB->getMemBufferRef();
74   make<std::unique_ptr<MemoryBuffer>>(std::move(MB)); // take MB ownership
75 
76   if (Tar)
77     Tar->append(relativeToRoot(Path), MBRef.getBuffer());
78   return MBRef;
79 }
80 
81 // Concatenates arguments to construct a string representing an error location.
82 static std::string createFileLineMsg(StringRef Path, unsigned Line) {
83   std::string Filename = path::filename(Path);
84   std::string Lineno = ":" + std::to_string(Line);
85   if (Filename == Path)
86     return Filename + Lineno;
87   return Filename + Lineno + " (" + Path.str() + Lineno + ")";
88 }
89 
90 template <class ELFT>
91 static std::string getSrcMsgAux(ObjFile<ELFT> &File, const Symbol &Sym,
92                                 InputSectionBase &Sec, uint64_t Offset) {
93   // In DWARF, functions and variables are stored to different places.
94   // First, lookup a function for a given offset.
95   if (Optional<DILineInfo> Info = File.getDILineInfo(&Sec, Offset))
96     return createFileLineMsg(Info->FileName, Info->Line);
97 
98   // If it failed, lookup again as a variable.
99   if (Optional<std::pair<std::string, unsigned>> FileLine =
100           File.getVariableLoc(Sym.getName()))
101     return createFileLineMsg(FileLine->first, FileLine->second);
102 
103   // File.SourceFile contains STT_FILE symbol, and that is a last resort.
104   return File.SourceFile;
105 }
106 
107 std::string InputFile::getSrcMsg(const Symbol &Sym, InputSectionBase &Sec,
108                                  uint64_t Offset) {
109   if (kind() != ObjKind)
110     return "";
111   switch (Config->EKind) {
112   default:
113     llvm_unreachable("Invalid kind");
114   case ELF32LEKind:
115     return getSrcMsgAux(cast<ObjFile<ELF32LE>>(*this), Sym, Sec, Offset);
116   case ELF32BEKind:
117     return getSrcMsgAux(cast<ObjFile<ELF32BE>>(*this), Sym, Sec, Offset);
118   case ELF64LEKind:
119     return getSrcMsgAux(cast<ObjFile<ELF64LE>>(*this), Sym, Sec, Offset);
120   case ELF64BEKind:
121     return getSrcMsgAux(cast<ObjFile<ELF64BE>>(*this), Sym, Sec, Offset);
122   }
123 }
124 
125 template <class ELFT> void ObjFile<ELFT>::initializeDwarf() {
126   Dwarf = llvm::make_unique<DWARFContext>(make_unique<LLDDwarfObj<ELFT>>(this));
127   const DWARFObject &Obj = Dwarf->getDWARFObj();
128   DwarfLine.reset(new DWARFDebugLine);
129   DWARFDataExtractor LineData(Obj, Obj.getLineSection(), Config->IsLE,
130                               Config->Wordsize);
131 
132   for (std::unique_ptr<DWARFCompileUnit> &CU : Dwarf->compile_units()) {
133     const DWARFDebugLine::LineTable *LT = Dwarf->getLineTableForUnit(CU.get());
134     if (!LT)
135       continue;
136     LineTables.push_back(LT);
137 
138     // Loop over variable records and insert them to VariableLoc.
139     for (const auto &Entry : CU->dies()) {
140       DWARFDie Die(CU.get(), &Entry);
141       // Skip all tags that are not variables.
142       if (Die.getTag() != dwarf::DW_TAG_variable)
143         continue;
144 
145       // Skip if a local variable because we don't need them for generating
146       // error messages. In general, only non-local symbols can fail to be
147       // linked.
148       if (!dwarf::toUnsigned(Die.find(dwarf::DW_AT_external), 0))
149         continue;
150 
151       // Get the source filename index for the variable.
152       unsigned File = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_file), 0);
153       if (!LT->hasFileAtIndex(File))
154         continue;
155 
156       // Get the line number on which the variable is declared.
157       unsigned Line = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_line), 0);
158 
159       // Get the name of the variable and add the collected information to
160       // VariableLoc. Usually Name is non-empty, but it can be empty if the
161       // input object file lacks some debug info.
162       StringRef Name = dwarf::toString(Die.find(dwarf::DW_AT_name), "");
163       if (!Name.empty())
164         VariableLoc.insert({Name, {LT, File, Line}});
165     }
166   }
167 }
168 
169 // Returns the pair of file name and line number describing location of data
170 // object (variable, array, etc) definition.
171 template <class ELFT>
172 Optional<std::pair<std::string, unsigned>>
173 ObjFile<ELFT>::getVariableLoc(StringRef Name) {
174   llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); });
175 
176   // Return if we have no debug information about data object.
177   auto It = VariableLoc.find(Name);
178   if (It == VariableLoc.end())
179     return None;
180 
181   // Take file name string from line table.
182   std::string FileName;
183   if (!It->second.LT->getFileNameByIndex(
184           It->second.File, nullptr,
185           DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FileName))
186     return None;
187 
188   return std::make_pair(FileName, It->second.Line);
189 }
190 
191 // Returns source line information for a given offset
192 // using DWARF debug info.
193 template <class ELFT>
194 Optional<DILineInfo> ObjFile<ELFT>::getDILineInfo(InputSectionBase *S,
195                                                   uint64_t Offset) {
196   llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); });
197 
198   // Use fake address calcuated by adding section file offset and offset in
199   // section. See comments for ObjectInfo class.
200   DILineInfo Info;
201   for (const llvm::DWARFDebugLine::LineTable *LT : LineTables)
202     if (LT->getFileLineInfoForAddress(
203             S->getOffsetInFile() + Offset, nullptr,
204             DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, Info))
205       return Info;
206   return None;
207 }
208 
209 // Returns source line information for a given offset using DWARF debug info.
210 template <class ELFT>
211 std::string ObjFile<ELFT>::getLineInfo(InputSectionBase *S, uint64_t Offset) {
212   if (Optional<DILineInfo> Info = getDILineInfo(S, Offset))
213     return Info->FileName + ":" + std::to_string(Info->Line);
214   return "";
215 }
216 
217 // Returns "<internal>", "foo.a(bar.o)" or "baz.o".
218 std::string lld::toString(const InputFile *F) {
219   if (!F)
220     return "<internal>";
221 
222   if (F->ToStringCache.empty()) {
223     if (F->ArchiveName.empty())
224       F->ToStringCache = F->getName();
225     else
226       F->ToStringCache = (F->ArchiveName + "(" + F->getName() + ")").str();
227   }
228   return F->ToStringCache;
229 }
230 
231 template <class ELFT>
232 ELFFileBase<ELFT>::ELFFileBase(Kind K, MemoryBufferRef MB) : InputFile(K, MB) {
233   if (ELFT::TargetEndianness == support::little)
234     EKind = ELFT::Is64Bits ? ELF64LEKind : ELF32LEKind;
235   else
236     EKind = ELFT::Is64Bits ? ELF64BEKind : ELF32BEKind;
237 
238   EMachine = getObj().getHeader()->e_machine;
239   OSABI = getObj().getHeader()->e_ident[llvm::ELF::EI_OSABI];
240 }
241 
242 template <class ELFT>
243 typename ELFT::SymRange ELFFileBase<ELFT>::getGlobalELFSyms() {
244   return makeArrayRef(ELFSyms.begin() + FirstGlobal, ELFSyms.end());
245 }
246 
247 template <class ELFT>
248 uint32_t ELFFileBase<ELFT>::getSectionIndex(const Elf_Sym &Sym) const {
249   return CHECK(getObj().getSectionIndex(&Sym, ELFSyms, SymtabSHNDX), this);
250 }
251 
252 template <class ELFT>
253 void ELFFileBase<ELFT>::initSymtab(ArrayRef<Elf_Shdr> Sections,
254                                    const Elf_Shdr *Symtab) {
255   FirstGlobal = Symtab->sh_info;
256   ELFSyms = CHECK(getObj().symbols(Symtab), this);
257   if (FirstGlobal == 0 || FirstGlobal > ELFSyms.size())
258     fatal(toString(this) + ": invalid sh_info in symbol table");
259 
260   StringTable =
261       CHECK(getObj().getStringTableForSymtab(*Symtab, Sections), this);
262 }
263 
264 template <class ELFT>
265 ObjFile<ELFT>::ObjFile(MemoryBufferRef M, StringRef ArchiveName)
266     : ELFFileBase<ELFT>(Base::ObjKind, M) {
267   this->ArchiveName = ArchiveName;
268 }
269 
270 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() {
271   if (this->Symbols.empty())
272     return {};
273   return makeArrayRef(this->Symbols).slice(1, this->FirstGlobal - 1);
274 }
275 
276 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getGlobalSymbols() {
277   return makeArrayRef(this->Symbols).slice(this->FirstGlobal);
278 }
279 
280 template <class ELFT>
281 void ObjFile<ELFT>::parse(DenseSet<CachedHashStringRef> &ComdatGroups) {
282   // Read a section table. JustSymbols is usually false.
283   if (this->JustSymbols)
284     initializeJustSymbols();
285   else
286     initializeSections(ComdatGroups);
287 
288   // Read a symbol table.
289   initializeSymbols();
290 }
291 
292 // Sections with SHT_GROUP and comdat bits define comdat section groups.
293 // They are identified and deduplicated by group name. This function
294 // returns a group name.
295 template <class ELFT>
296 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> Sections,
297                                               const Elf_Shdr &Sec) {
298   // Group signatures are stored as symbol names in object files.
299   // sh_info contains a symbol index, so we fetch a symbol and read its name.
300   if (this->ELFSyms.empty())
301     this->initSymtab(
302         Sections, CHECK(object::getSection<ELFT>(Sections, Sec.sh_link), this));
303 
304   const Elf_Sym *Sym =
305       CHECK(object::getSymbol<ELFT>(this->ELFSyms, Sec.sh_info), this);
306   StringRef Signature = CHECK(Sym->getName(this->StringTable), this);
307 
308   // As a special case, if a symbol is a section symbol and has no name,
309   // we use a section name as a signature.
310   //
311   // Such SHT_GROUP sections are invalid from the perspective of the ELF
312   // standard, but GNU gold 1.14 (the newest version as of July 2017) or
313   // older produce such sections as outputs for the -r option, so we need
314   // a bug-compatibility.
315   if (Signature.empty() && Sym->getType() == STT_SECTION)
316     return getSectionName(Sec);
317   return Signature;
318 }
319 
320 template <class ELFT>
321 ArrayRef<typename ObjFile<ELFT>::Elf_Word>
322 ObjFile<ELFT>::getShtGroupEntries(const Elf_Shdr &Sec) {
323   const ELFFile<ELFT> &Obj = this->getObj();
324   ArrayRef<Elf_Word> Entries =
325       CHECK(Obj.template getSectionContentsAsArray<Elf_Word>(&Sec), this);
326   if (Entries.empty() || Entries[0] != GRP_COMDAT)
327     fatal(toString(this) + ": unsupported SHT_GROUP format");
328   return Entries.slice(1);
329 }
330 
331 template <class ELFT> bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &Sec) {
332   // On a regular link we don't merge sections if -O0 (default is -O1). This
333   // sometimes makes the linker significantly faster, although the output will
334   // be bigger.
335   //
336   // Doing the same for -r would create a problem as it would combine sections
337   // with different sh_entsize. One option would be to just copy every SHF_MERGE
338   // section as is to the output. While this would produce a valid ELF file with
339   // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when
340   // they see two .debug_str. We could have separate logic for combining
341   // SHF_MERGE sections based both on their name and sh_entsize, but that seems
342   // to be more trouble than it is worth. Instead, we just use the regular (-O1)
343   // logic for -r.
344   if (Config->Optimize == 0 && !Config->Relocatable)
345     return false;
346 
347   // A mergeable section with size 0 is useless because they don't have
348   // any data to merge. A mergeable string section with size 0 can be
349   // argued as invalid because it doesn't end with a null character.
350   // We'll avoid a mess by handling them as if they were non-mergeable.
351   if (Sec.sh_size == 0)
352     return false;
353 
354   // Check for sh_entsize. The ELF spec is not clear about the zero
355   // sh_entsize. It says that "the member [sh_entsize] contains 0 if
356   // the section does not hold a table of fixed-size entries". We know
357   // that Rust 1.13 produces a string mergeable section with a zero
358   // sh_entsize. Here we just accept it rather than being picky about it.
359   uint64_t EntSize = Sec.sh_entsize;
360   if (EntSize == 0)
361     return false;
362   if (Sec.sh_size % EntSize)
363     fatal(toString(this) +
364           ": SHF_MERGE section size must be a multiple of sh_entsize");
365 
366   uint64_t Flags = Sec.sh_flags;
367   if (!(Flags & SHF_MERGE))
368     return false;
369   if (Flags & SHF_WRITE)
370     fatal(toString(this) + ": writable SHF_MERGE section is not supported");
371 
372   return true;
373 }
374 
375 // This is for --just-symbols.
376 //
377 // --just-symbols is a very minor feature that allows you to link your
378 // output against other existing program, so that if you load both your
379 // program and the other program into memory, your output can refer the
380 // other program's symbols.
381 //
382 // When the option is given, we link "just symbols". The section table is
383 // initialized with null pointers.
384 template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() {
385   ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this);
386   this->Sections.resize(ObjSections.size());
387 
388   for (const Elf_Shdr &Sec : ObjSections) {
389     if (Sec.sh_type != SHT_SYMTAB)
390       continue;
391     this->initSymtab(ObjSections, &Sec);
392     return;
393   }
394 }
395 
396 template <class ELFT>
397 void ObjFile<ELFT>::initializeSections(
398     DenseSet<CachedHashStringRef> &ComdatGroups) {
399   const ELFFile<ELFT> &Obj = this->getObj();
400 
401   ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this);
402   uint64_t Size = ObjSections.size();
403   this->Sections.resize(Size);
404   this->SectionStringTable =
405       CHECK(Obj.getSectionStringTable(ObjSections), this);
406 
407   for (size_t I = 0, E = ObjSections.size(); I < E; I++) {
408     if (this->Sections[I] == &InputSection::Discarded)
409       continue;
410     const Elf_Shdr &Sec = ObjSections[I];
411 
412     // SHF_EXCLUDE'ed sections are discarded by the linker. However,
413     // if -r is given, we'll let the final link discard such sections.
414     // This is compatible with GNU.
415     if ((Sec.sh_flags & SHF_EXCLUDE) && !Config->Relocatable) {
416       this->Sections[I] = &InputSection::Discarded;
417       continue;
418     }
419 
420     switch (Sec.sh_type) {
421     case SHT_GROUP: {
422       // De-duplicate section groups by their signatures.
423       StringRef Signature = getShtGroupSignature(ObjSections, Sec);
424       bool IsNew = ComdatGroups.insert(CachedHashStringRef(Signature)).second;
425       this->Sections[I] = &InputSection::Discarded;
426 
427       // If it is a new section group, we want to keep group members.
428       // Group leader sections, which contain indices of group members, are
429       // discarded because they are useless beyond this point. The only
430       // exception is the -r option because in order to produce re-linkable
431       // object files, we want to pass through basically everything.
432       if (IsNew) {
433         if (Config->Relocatable)
434           this->Sections[I] = createInputSection(Sec);
435         continue;
436       }
437 
438       // Otherwise, discard group members.
439       for (uint32_t SecIndex : getShtGroupEntries(Sec)) {
440         if (SecIndex >= Size)
441           fatal(toString(this) +
442                 ": invalid section index in group: " + Twine(SecIndex));
443         this->Sections[SecIndex] = &InputSection::Discarded;
444       }
445       break;
446     }
447     case SHT_SYMTAB:
448       this->initSymtab(ObjSections, &Sec);
449       break;
450     case SHT_SYMTAB_SHNDX:
451       this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, ObjSections), this);
452       break;
453     case SHT_STRTAB:
454     case SHT_NULL:
455       break;
456     default:
457       this->Sections[I] = createInputSection(Sec);
458     }
459 
460     // .ARM.exidx sections have a reverse dependency on the InputSection they
461     // have a SHF_LINK_ORDER dependency, this is identified by the sh_link.
462     if (Sec.sh_flags & SHF_LINK_ORDER) {
463       if (Sec.sh_link >= this->Sections.size())
464         fatal(toString(this) +
465               ": invalid sh_link index: " + Twine(Sec.sh_link));
466 
467       InputSectionBase *LinkSec = this->Sections[Sec.sh_link];
468       InputSection *IS = cast<InputSection>(this->Sections[I]);
469       LinkSec->DependentSections.push_back(IS);
470       if (!isa<InputSection>(LinkSec))
471         error("a section " + IS->Name +
472               " with SHF_LINK_ORDER should not refer a non-regular "
473               "section: " +
474               toString(LinkSec));
475     }
476   }
477 }
478 
479 // The ARM support in lld makes some use of instructions that are not available
480 // on all ARM architectures. Namely:
481 // - Use of BLX instruction for interworking between ARM and Thumb state.
482 // - Use of the extended Thumb branch encoding in relocation.
483 // - Use of the MOVT/MOVW instructions in Thumb Thunks.
484 // The ARM Attributes section contains information about the architecture chosen
485 // at compile time. We follow the convention that if at least one input object
486 // is compiled with an architecture that supports these features then lld is
487 // permitted to use them.
488 static void updateSupportedARMFeatures(const ARMAttributeParser &Attributes) {
489   if (!Attributes.hasAttribute(ARMBuildAttrs::CPU_arch))
490     return;
491   auto Arch = Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch);
492   switch (Arch) {
493   case ARMBuildAttrs::Pre_v4:
494   case ARMBuildAttrs::v4:
495   case ARMBuildAttrs::v4T:
496     // Architectures prior to v5 do not support BLX instruction
497     break;
498   case ARMBuildAttrs::v5T:
499   case ARMBuildAttrs::v5TE:
500   case ARMBuildAttrs::v5TEJ:
501   case ARMBuildAttrs::v6:
502   case ARMBuildAttrs::v6KZ:
503   case ARMBuildAttrs::v6K:
504     Config->ARMHasBlx = true;
505     // Architectures used in pre-Cortex processors do not support
506     // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception
507     // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do.
508     break;
509   default:
510     // All other Architectures have BLX and extended branch encoding
511     Config->ARMHasBlx = true;
512     Config->ARMJ1J2BranchEncoding = true;
513     if (Arch != ARMBuildAttrs::v6_M && Arch != ARMBuildAttrs::v6S_M)
514       // All Architectures used in Cortex processors with the exception
515       // of v6-M and v6S-M have the MOVT and MOVW instructions.
516       Config->ARMHasMovtMovw = true;
517     break;
518   }
519 }
520 
521 template <class ELFT>
522 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &Sec) {
523   uint32_t Idx = Sec.sh_info;
524   if (Idx >= this->Sections.size())
525     fatal(toString(this) + ": invalid relocated section index: " + Twine(Idx));
526   InputSectionBase *Target = this->Sections[Idx];
527 
528   // Strictly speaking, a relocation section must be included in the
529   // group of the section it relocates. However, LLVM 3.3 and earlier
530   // would fail to do so, so we gracefully handle that case.
531   if (Target == &InputSection::Discarded)
532     return nullptr;
533 
534   if (!Target)
535     fatal(toString(this) + ": unsupported relocation reference");
536   return Target;
537 }
538 
539 // Create a regular InputSection class that has the same contents
540 // as a given section.
541 static InputSection *toRegularSection(MergeInputSection *Sec) {
542   return make<InputSection>(Sec->File, Sec->Flags, Sec->Type, Sec->Alignment,
543                             Sec->Data, Sec->Name);
544 }
545 
546 template <class ELFT>
547 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &Sec) {
548   StringRef Name = getSectionName(Sec);
549 
550   switch (Sec.sh_type) {
551   case SHT_ARM_ATTRIBUTES: {
552     if (Config->EMachine != EM_ARM)
553       break;
554     ARMAttributeParser Attributes;
555     ArrayRef<uint8_t> Contents = check(this->getObj().getSectionContents(&Sec));
556     Attributes.Parse(Contents, /*isLittle*/ Config->EKind == ELF32LEKind);
557     updateSupportedARMFeatures(Attributes);
558     // FIXME: Retain the first attribute section we see. The eglibc ARM
559     // dynamic loaders require the presence of an attribute section for dlopen
560     // to work. In a full implementation we would merge all attribute sections.
561     if (InX::ARMAttributes == nullptr) {
562       InX::ARMAttributes = make<InputSection>(*this, Sec, Name);
563       return InX::ARMAttributes;
564     }
565     return &InputSection::Discarded;
566   }
567   case SHT_RELA:
568   case SHT_REL: {
569     // Find a relocation target section and associate this section with that.
570     // Target may have been discarded if it is in a different section group
571     // and the group is discarded, even though it's a violation of the
572     // spec. We handle that situation gracefully by discarding dangling
573     // relocation sections.
574     InputSectionBase *Target = getRelocTarget(Sec);
575     if (!Target)
576       return nullptr;
577 
578     // This section contains relocation information.
579     // If -r is given, we do not interpret or apply relocation
580     // but just copy relocation sections to output.
581     if (Config->Relocatable)
582       return make<InputSection>(*this, Sec, Name);
583 
584     if (Target->FirstRelocation)
585       fatal(toString(this) +
586             ": multiple relocation sections to one section are not supported");
587 
588     // ELF spec allows mergeable sections with relocations, but they are
589     // rare, and it is in practice hard to merge such sections by contents,
590     // because applying relocations at end of linking changes section
591     // contents. So, we simply handle such sections as non-mergeable ones.
592     // Degrading like this is acceptable because section merging is optional.
593     if (auto *MS = dyn_cast<MergeInputSection>(Target)) {
594       Target = toRegularSection(MS);
595       this->Sections[Sec.sh_info] = Target;
596     }
597 
598     if (Sec.sh_type == SHT_RELA) {
599       ArrayRef<Elf_Rela> Rels = CHECK(this->getObj().relas(&Sec), this);
600       Target->FirstRelocation = Rels.begin();
601       Target->NumRelocations = Rels.size();
602       Target->AreRelocsRela = true;
603     } else {
604       ArrayRef<Elf_Rel> Rels = CHECK(this->getObj().rels(&Sec), this);
605       Target->FirstRelocation = Rels.begin();
606       Target->NumRelocations = Rels.size();
607       Target->AreRelocsRela = false;
608     }
609     assert(isUInt<31>(Target->NumRelocations));
610 
611     // Relocation sections processed by the linker are usually removed
612     // from the output, so returning `nullptr` for the normal case.
613     // However, if -emit-relocs is given, we need to leave them in the output.
614     // (Some post link analysis tools need this information.)
615     if (Config->EmitRelocs) {
616       InputSection *RelocSec = make<InputSection>(*this, Sec, Name);
617       // We will not emit relocation section if target was discarded.
618       Target->DependentSections.push_back(RelocSec);
619       return RelocSec;
620     }
621     return nullptr;
622   }
623   }
624 
625   // The GNU linker uses .note.GNU-stack section as a marker indicating
626   // that the code in the object file does not expect that the stack is
627   // executable (in terms of NX bit). If all input files have the marker,
628   // the GNU linker adds a PT_GNU_STACK segment to tells the loader to
629   // make the stack non-executable. Most object files have this section as
630   // of 2017.
631   //
632   // But making the stack non-executable is a norm today for security
633   // reasons. Failure to do so may result in a serious security issue.
634   // Therefore, we make LLD always add PT_GNU_STACK unless it is
635   // explicitly told to do otherwise (by -z execstack). Because the stack
636   // executable-ness is controlled solely by command line options,
637   // .note.GNU-stack sections are simply ignored.
638   if (Name == ".note.GNU-stack")
639     return &InputSection::Discarded;
640 
641   // Split stacks is a feature to support a discontiguous stack. At least
642   // as of 2017, it seems that the feature is not being used widely.
643   // Only GNU gold supports that. We don't. For the details about that,
644   // see https://gcc.gnu.org/wiki/SplitStacks
645   if (Name == ".note.GNU-split-stack") {
646     error(toString(this) +
647           ": object file compiled with -fsplit-stack is not supported");
648     return &InputSection::Discarded;
649   }
650 
651   // The linkonce feature is a sort of proto-comdat. Some glibc i386 object
652   // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce
653   // sections. Drop those sections to avoid duplicate symbol errors.
654   // FIXME: This is glibc PR20543, we should remove this hack once that has been
655   // fixed for a while.
656   if (Name.startswith(".gnu.linkonce."))
657     return &InputSection::Discarded;
658 
659   // If we are creating a new .build-id section, strip existing .build-id
660   // sections so that the output won't have more than one .build-id.
661   // This is not usually a problem because input object files normally don't
662   // have .build-id sections, but you can create such files by
663   // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it.
664   if (Name == ".note.gnu.build-id" && Config->BuildId != BuildIdKind::None)
665     return &InputSection::Discarded;
666 
667   // The linker merges EH (exception handling) frames and creates a
668   // .eh_frame_hdr section for runtime. So we handle them with a special
669   // class. For relocatable outputs, they are just passed through.
670   if (Name == ".eh_frame" && !Config->Relocatable)
671     return make<EhInputSection>(*this, Sec, Name);
672 
673   if (shouldMerge(Sec))
674     return make<MergeInputSection>(*this, Sec, Name);
675   return make<InputSection>(*this, Sec, Name);
676 }
677 
678 template <class ELFT>
679 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &Sec) {
680   return CHECK(this->getObj().getSectionName(&Sec, SectionStringTable), this);
681 }
682 
683 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() {
684   this->Symbols.reserve(this->ELFSyms.size());
685   for (const Elf_Sym &Sym : this->ELFSyms)
686     this->Symbols.push_back(createSymbol(&Sym));
687 }
688 
689 template <class ELFT> Symbol *ObjFile<ELFT>::createSymbol(const Elf_Sym *Sym) {
690   int Binding = Sym->getBinding();
691 
692   uint32_t SecIdx = this->getSectionIndex(*Sym);
693   if (SecIdx >= this->Sections.size())
694     fatal(toString(this) + ": invalid section index: " + Twine(SecIdx));
695 
696   InputSectionBase *Sec = this->Sections[SecIdx];
697   uint8_t StOther = Sym->st_other;
698   uint8_t Type = Sym->getType();
699   uint64_t Value = Sym->st_value;
700   uint64_t Size = Sym->st_size;
701 
702   if (Binding == STB_LOCAL) {
703     if (Sym->getType() == STT_FILE)
704       SourceFile = CHECK(Sym->getName(this->StringTable), this);
705 
706     if (this->StringTable.size() <= Sym->st_name)
707       fatal(toString(this) + ": invalid symbol name offset");
708 
709     StringRefZ Name = this->StringTable.data() + Sym->st_name;
710     if (Sym->st_shndx == SHN_UNDEF)
711       return make<Undefined>(this, Name, Binding, StOther, Type);
712 
713     return make<Defined>(this, Name, Binding, StOther, Type, Value, Size, Sec);
714   }
715 
716   StringRef Name = CHECK(Sym->getName(this->StringTable), this);
717 
718   switch (Sym->st_shndx) {
719   case SHN_UNDEF:
720     return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type,
721                                       /*CanOmitFromDynSym=*/false, this);
722   case SHN_COMMON:
723     if (Value == 0 || Value >= UINT32_MAX)
724       fatal(toString(this) + ": common symbol '" + Name +
725             "' has invalid alignment: " + Twine(Value));
726     return Symtab->addCommon(Name, Size, Value, Binding, StOther, Type, *this);
727   }
728 
729   switch (Binding) {
730   default:
731     fatal(toString(this) + ": unexpected binding: " + Twine(Binding));
732   case STB_GLOBAL:
733   case STB_WEAK:
734   case STB_GNU_UNIQUE:
735     if (Sec == &InputSection::Discarded)
736       return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type,
737                                         /*CanOmitFromDynSym=*/false, this);
738     return Symtab->addRegular(Name, StOther, Type, Value, Size, Binding, Sec,
739                               this);
740   }
741 }
742 
743 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&File)
744     : InputFile(ArchiveKind, File->getMemoryBufferRef()),
745       File(std::move(File)) {}
746 
747 template <class ELFT> void ArchiveFile::parse() {
748   for (const Archive::Symbol &Sym : File->symbols())
749     Symtab->addLazyArchive<ELFT>(Sym.getName(), *this, Sym);
750 }
751 
752 // Returns a buffer pointing to a member file containing a given symbol.
753 InputFile *ArchiveFile::fetch(const Archive::Symbol &Sym) {
754   Archive::Child C =
755       CHECK(Sym.getMember(), toString(this) +
756                                  ": could not get the member for symbol " +
757                                  Sym.getName());
758 
759   if (!Seen.insert(C.getChildOffset()).second)
760     return nullptr;
761 
762   MemoryBufferRef MB =
763       CHECK(C.getMemoryBufferRef(),
764             toString(this) +
765                 ": could not get the buffer for the member defining symbol " +
766                 Sym.getName());
767 
768   if (Tar && C.getParent()->isThin())
769     Tar->append(relativeToRoot(CHECK(C.getFullName(), this)), MB.getBuffer());
770 
771   InputFile *File = createObjectFile(
772       MB, getName(), C.getParent()->isThin() ? 0 : C.getChildOffset());
773   File->GroupId = GroupId;
774   return File;
775 }
776 
777 template <class ELFT>
778 SharedFile<ELFT>::SharedFile(MemoryBufferRef M, StringRef DefaultSoName)
779     : ELFFileBase<ELFT>(Base::SharedKind, M), SoName(DefaultSoName),
780       IsNeeded(!Config->AsNeeded) {}
781 
782 // Partially parse the shared object file so that we can call
783 // getSoName on this object.
784 template <class ELFT> void SharedFile<ELFT>::parseSoName() {
785   const Elf_Shdr *DynamicSec = nullptr;
786   const ELFFile<ELFT> Obj = this->getObj();
787   ArrayRef<Elf_Shdr> Sections = CHECK(Obj.sections(), this);
788 
789   // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d.
790   for (const Elf_Shdr &Sec : Sections) {
791     switch (Sec.sh_type) {
792     default:
793       continue;
794     case SHT_DYNSYM:
795       this->initSymtab(Sections, &Sec);
796       break;
797     case SHT_DYNAMIC:
798       DynamicSec = &Sec;
799       break;
800     case SHT_SYMTAB_SHNDX:
801       this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, Sections), this);
802       break;
803     case SHT_GNU_versym:
804       this->VersymSec = &Sec;
805       break;
806     case SHT_GNU_verdef:
807       this->VerdefSec = &Sec;
808       break;
809     }
810   }
811 
812   if (this->VersymSec && this->ELFSyms.empty())
813     error("SHT_GNU_versym should be associated with symbol table");
814 
815   // Search for a DT_SONAME tag to initialize this->SoName.
816   if (!DynamicSec)
817     return;
818   ArrayRef<Elf_Dyn> Arr =
819       CHECK(Obj.template getSectionContentsAsArray<Elf_Dyn>(DynamicSec), this);
820   for (const Elf_Dyn &Dyn : Arr) {
821     if (Dyn.d_tag == DT_SONAME) {
822       uint64_t Val = Dyn.getVal();
823       if (Val >= this->StringTable.size())
824         fatal(toString(this) + ": invalid DT_SONAME entry");
825       SoName = this->StringTable.data() + Val;
826       return;
827     }
828   }
829 }
830 
831 // Parses ".gnu.version" section which is a parallel array for the symbol table.
832 // If a given file doesn't have ".gnu.version" section, returns VER_NDX_GLOBAL.
833 template <class ELFT> std::vector<uint32_t> SharedFile<ELFT>::parseVersyms() {
834   size_t Size = this->ELFSyms.size() - this->FirstGlobal;
835   if (!VersymSec)
836     return std::vector<uint32_t>(Size, VER_NDX_GLOBAL);
837 
838   const char *Base = this->MB.getBuffer().data();
839   const Elf_Versym *Versym =
840       reinterpret_cast<const Elf_Versym *>(Base + VersymSec->sh_offset) +
841       this->FirstGlobal;
842 
843   std::vector<uint32_t> Ret(Size);
844   for (size_t I = 0; I < Size; ++I)
845     Ret[I] = Versym[I].vs_index;
846   return Ret;
847 }
848 
849 // Parse the version definitions in the object file if present. Returns a vector
850 // whose nth element contains a pointer to the Elf_Verdef for version identifier
851 // n. Version identifiers that are not definitions map to nullptr.
852 template <class ELFT>
853 std::vector<const typename ELFT::Verdef *> SharedFile<ELFT>::parseVerdefs() {
854   if (!VerdefSec)
855     return {};
856 
857   // We cannot determine the largest verdef identifier without inspecting
858   // every Elf_Verdef, but both bfd and gold assign verdef identifiers
859   // sequentially starting from 1, so we predict that the largest identifier
860   // will be VerdefCount.
861   unsigned VerdefCount = VerdefSec->sh_info;
862   std::vector<const Elf_Verdef *> Verdefs(VerdefCount + 1);
863 
864   // Build the Verdefs array by following the chain of Elf_Verdef objects
865   // from the start of the .gnu.version_d section.
866   const char *Base = this->MB.getBuffer().data();
867   const char *Verdef = Base + VerdefSec->sh_offset;
868   for (unsigned I = 0; I != VerdefCount; ++I) {
869     auto *CurVerdef = reinterpret_cast<const Elf_Verdef *>(Verdef);
870     Verdef += CurVerdef->vd_next;
871     unsigned VerdefIndex = CurVerdef->vd_ndx;
872     if (Verdefs.size() <= VerdefIndex)
873       Verdefs.resize(VerdefIndex + 1);
874     Verdefs[VerdefIndex] = CurVerdef;
875   }
876 
877   return Verdefs;
878 }
879 
880 // We do not usually care about alignments of data in shared object
881 // files because the loader takes care of it. However, if we promote a
882 // DSO symbol to point to .bss due to copy relocation, we need to keep
883 // the original alignment requirements. We infer it in this function.
884 template <class ELFT>
885 uint32_t SharedFile<ELFT>::getAlignment(ArrayRef<Elf_Shdr> Sections,
886                                         const Elf_Sym &Sym) {
887   uint64_t Ret = 1;
888   if (Sym.st_value)
889     Ret = 1ULL << countTrailingZeros((uint64_t)Sym.st_value);
890   if (0 < Sym.st_shndx && Sym.st_shndx < Sections.size())
891     Ret = std::min<uint64_t>(Ret, Sections[Sym.st_shndx].sh_addralign);
892 
893   if (Ret > UINT32_MAX)
894     error(toString(this) + ": alignment too large: " +
895           CHECK(Sym.getName(this->StringTable), this));
896   return Ret;
897 }
898 
899 // Fully parse the shared object file. This must be called after parseSoName().
900 //
901 // This function parses symbol versions. If a DSO has version information,
902 // the file has a ".gnu.version_d" section which contains symbol version
903 // definitions. Each symbol is associated to one version through a table in
904 // ".gnu.version" section. That table is a parallel array for the symbol
905 // table, and each table entry contains an index in ".gnu.version_d".
906 //
907 // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for
908 // VER_NDX_GLOBAL. There's no table entry for these special versions in
909 // ".gnu.version_d".
910 //
911 // The file format for symbol versioning is perhaps a bit more complicated
912 // than necessary, but you can easily understand the code if you wrap your
913 // head around the data structure described above.
914 template <class ELFT> void SharedFile<ELFT>::parseRest() {
915   Verdefs = parseVerdefs();                       // parse .gnu.version_d
916   std::vector<uint32_t> Versyms = parseVersyms(); // parse .gnu.version
917   ArrayRef<Elf_Shdr> Sections = CHECK(this->getObj().sections(), this);
918 
919   // System libraries can have a lot of symbols with versions. Using a
920   // fixed buffer for computing the versions name (foo@ver) can save a
921   // lot of allocations.
922   SmallString<0> VersionedNameBuffer;
923 
924   // Add symbols to the symbol table.
925   ArrayRef<Elf_Sym> Syms = this->getGlobalELFSyms();
926   for (size_t I = 0; I < Syms.size(); ++I) {
927     const Elf_Sym &Sym = Syms[I];
928 
929     StringRef Name = CHECK(Sym.getName(this->StringTable), this);
930     if (Sym.isUndefined()) {
931       Symbol *S = Symtab->addUndefined<ELFT>(Name, Sym.getBinding(),
932                                              Sym.st_other, Sym.getType(),
933                                              /*CanOmitFromDynSym=*/false, this);
934       S->ExportDynamic = true;
935       continue;
936     }
937 
938     // ELF spec requires that all local symbols precede weak or global
939     // symbols in each symbol table, and the index of first non-local symbol
940     // is stored to sh_info. If a local symbol appears after some non-local
941     // symbol, that's a violation of the spec.
942     if (Sym.getBinding() == STB_LOCAL) {
943       warn("found local symbol '" + Name +
944            "' in global part of symbol table in file " + toString(this));
945       continue;
946     }
947 
948     // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly
949     // assigns VER_NDX_LOCAL to this section global symbol. Here is a
950     // workaround for this bug.
951     uint32_t Idx = Versyms[I] & ~VERSYM_HIDDEN;
952     if (Config->EMachine == EM_MIPS && Idx == VER_NDX_LOCAL &&
953         Name == "_gp_disp")
954       continue;
955 
956     uint64_t Alignment = getAlignment(Sections, Sym);
957     if (!(Versyms[I] & VERSYM_HIDDEN))
958       Symtab->addShared(Name, *this, Sym, Alignment, Idx);
959 
960     // Also add the symbol with the versioned name to handle undefined symbols
961     // with explicit versions.
962     if (Idx == VER_NDX_GLOBAL)
963       continue;
964 
965     if (Idx >= Verdefs.size() || Idx == VER_NDX_LOCAL) {
966       error("corrupt input file: version definition index " + Twine(Idx) +
967             " for symbol " + Name + " is out of bounds\n>>> defined in " +
968             toString(this));
969       continue;
970     }
971 
972     StringRef VerName =
973         this->StringTable.data() + Verdefs[Idx]->getAux()->vda_name;
974     VersionedNameBuffer.clear();
975     Name = (Name + "@" + VerName).toStringRef(VersionedNameBuffer);
976     Symtab->addShared(Saver.save(Name), *this, Sym, Alignment, Idx);
977   }
978 }
979 
980 static ELFKind getBitcodeELFKind(const Triple &T) {
981   if (T.isLittleEndian())
982     return T.isArch64Bit() ? ELF64LEKind : ELF32LEKind;
983   return T.isArch64Bit() ? ELF64BEKind : ELF32BEKind;
984 }
985 
986 static uint8_t getBitcodeMachineKind(StringRef Path, const Triple &T) {
987   switch (T.getArch()) {
988   case Triple::aarch64:
989     return EM_AARCH64;
990   case Triple::arm:
991   case Triple::thumb:
992     return EM_ARM;
993   case Triple::avr:
994     return EM_AVR;
995   case Triple::mips:
996   case Triple::mipsel:
997   case Triple::mips64:
998   case Triple::mips64el:
999     return EM_MIPS;
1000   case Triple::ppc:
1001     return EM_PPC;
1002   case Triple::ppc64:
1003     return EM_PPC64;
1004   case Triple::x86:
1005     return T.isOSIAMCU() ? EM_IAMCU : EM_386;
1006   case Triple::x86_64:
1007     return EM_X86_64;
1008   default:
1009     fatal(Path + ": could not infer e_machine from bitcode target triple " +
1010           T.str());
1011   }
1012 }
1013 
1014 BitcodeFile::BitcodeFile(MemoryBufferRef MB, StringRef ArchiveName,
1015                          uint64_t OffsetInArchive)
1016     : InputFile(BitcodeKind, MB) {
1017   this->ArchiveName = ArchiveName;
1018 
1019   // Here we pass a new MemoryBufferRef which is identified by ArchiveName
1020   // (the fully resolved path of the archive) + member name + offset of the
1021   // member in the archive.
1022   // ThinLTO uses the MemoryBufferRef identifier to access its internal
1023   // data structures and if two archives define two members with the same name,
1024   // this causes a collision which result in only one of the objects being
1025   // taken into consideration at LTO time (which very likely causes undefined
1026   // symbols later in the link stage).
1027   MemoryBufferRef MBRef(MB.getBuffer(),
1028                         Saver.save(ArchiveName + MB.getBufferIdentifier() +
1029                                    utostr(OffsetInArchive)));
1030   Obj = CHECK(lto::InputFile::create(MBRef), this);
1031 
1032   Triple T(Obj->getTargetTriple());
1033   EKind = getBitcodeELFKind(T);
1034   EMachine = getBitcodeMachineKind(MB.getBufferIdentifier(), T);
1035 }
1036 
1037 static uint8_t mapVisibility(GlobalValue::VisibilityTypes GvVisibility) {
1038   switch (GvVisibility) {
1039   case GlobalValue::DefaultVisibility:
1040     return STV_DEFAULT;
1041   case GlobalValue::HiddenVisibility:
1042     return STV_HIDDEN;
1043   case GlobalValue::ProtectedVisibility:
1044     return STV_PROTECTED;
1045   }
1046   llvm_unreachable("unknown visibility");
1047 }
1048 
1049 template <class ELFT>
1050 static Symbol *createBitcodeSymbol(const std::vector<bool> &KeptComdats,
1051                                    const lto::InputFile::Symbol &ObjSym,
1052                                    BitcodeFile &F) {
1053   StringRef NameRef = Saver.save(ObjSym.getName());
1054   uint32_t Binding = ObjSym.isWeak() ? STB_WEAK : STB_GLOBAL;
1055 
1056   uint8_t Type = ObjSym.isTLS() ? STT_TLS : STT_NOTYPE;
1057   uint8_t Visibility = mapVisibility(ObjSym.getVisibility());
1058   bool CanOmitFromDynSym = ObjSym.canBeOmittedFromSymbolTable();
1059 
1060   int C = ObjSym.getComdatIndex();
1061   if (C != -1 && !KeptComdats[C])
1062     return Symtab->addUndefined<ELFT>(NameRef, Binding, Visibility, Type,
1063                                       CanOmitFromDynSym, &F);
1064 
1065   if (ObjSym.isUndefined())
1066     return Symtab->addUndefined<ELFT>(NameRef, Binding, Visibility, Type,
1067                                       CanOmitFromDynSym, &F);
1068 
1069   if (ObjSym.isCommon())
1070     return Symtab->addCommon(NameRef, ObjSym.getCommonSize(),
1071                              ObjSym.getCommonAlignment(), Binding, Visibility,
1072                              STT_OBJECT, F);
1073 
1074   return Symtab->addBitcode(NameRef, Binding, Visibility, Type,
1075                             CanOmitFromDynSym, F);
1076 }
1077 
1078 template <class ELFT>
1079 void BitcodeFile::parse(DenseSet<CachedHashStringRef> &ComdatGroups) {
1080   std::vector<bool> KeptComdats;
1081   for (StringRef S : Obj->getComdatTable())
1082     KeptComdats.push_back(ComdatGroups.insert(CachedHashStringRef(S)).second);
1083 
1084   for (const lto::InputFile::Symbol &ObjSym : Obj->symbols())
1085     Symbols.push_back(createBitcodeSymbol<ELFT>(KeptComdats, ObjSym, *this));
1086 }
1087 
1088 static ELFKind getELFKind(MemoryBufferRef MB) {
1089   unsigned char Size;
1090   unsigned char Endian;
1091   std::tie(Size, Endian) = getElfArchType(MB.getBuffer());
1092 
1093   if (Endian != ELFDATA2LSB && Endian != ELFDATA2MSB)
1094     fatal(MB.getBufferIdentifier() + ": invalid data encoding");
1095   if (Size != ELFCLASS32 && Size != ELFCLASS64)
1096     fatal(MB.getBufferIdentifier() + ": invalid file class");
1097 
1098   size_t BufSize = MB.getBuffer().size();
1099   if ((Size == ELFCLASS32 && BufSize < sizeof(Elf32_Ehdr)) ||
1100       (Size == ELFCLASS64 && BufSize < sizeof(Elf64_Ehdr)))
1101     fatal(MB.getBufferIdentifier() + ": file is too short");
1102 
1103   if (Size == ELFCLASS32)
1104     return (Endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind;
1105   return (Endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind;
1106 }
1107 
1108 void BinaryFile::parse() {
1109   ArrayRef<uint8_t> Data = toArrayRef(MB.getBuffer());
1110   auto *Section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS,
1111                                      8, Data, ".data");
1112   Sections.push_back(Section);
1113 
1114   // For each input file foo that is embedded to a result as a binary
1115   // blob, we define _binary_foo_{start,end,size} symbols, so that
1116   // user programs can access blobs by name. Non-alphanumeric
1117   // characters in a filename are replaced with underscore.
1118   std::string S = "_binary_" + MB.getBufferIdentifier().str();
1119   for (size_t I = 0; I < S.size(); ++I)
1120     if (!isAlnum(S[I]))
1121       S[I] = '_';
1122 
1123   Symtab->addRegular(Saver.save(S + "_start"), STV_DEFAULT, STT_OBJECT, 0, 0,
1124                      STB_GLOBAL, Section, nullptr);
1125   Symtab->addRegular(Saver.save(S + "_end"), STV_DEFAULT, STT_OBJECT,
1126                      Data.size(), 0, STB_GLOBAL, Section, nullptr);
1127   Symtab->addRegular(Saver.save(S + "_size"), STV_DEFAULT, STT_OBJECT,
1128                      Data.size(), 0, STB_GLOBAL, nullptr, nullptr);
1129 }
1130 
1131 static bool isBitcode(MemoryBufferRef MB) {
1132   using namespace sys::fs;
1133   return identify_magic(MB.getBuffer()) == file_magic::bitcode;
1134 }
1135 
1136 InputFile *elf::createObjectFile(MemoryBufferRef MB, StringRef ArchiveName,
1137                                  uint64_t OffsetInArchive) {
1138   if (isBitcode(MB))
1139     return make<BitcodeFile>(MB, ArchiveName, OffsetInArchive);
1140 
1141   switch (getELFKind(MB)) {
1142   case ELF32LEKind:
1143     return make<ObjFile<ELF32LE>>(MB, ArchiveName);
1144   case ELF32BEKind:
1145     return make<ObjFile<ELF32BE>>(MB, ArchiveName);
1146   case ELF64LEKind:
1147     return make<ObjFile<ELF64LE>>(MB, ArchiveName);
1148   case ELF64BEKind:
1149     return make<ObjFile<ELF64BE>>(MB, ArchiveName);
1150   default:
1151     llvm_unreachable("getELFKind");
1152   }
1153 }
1154 
1155 InputFile *elf::createSharedFile(MemoryBufferRef MB, StringRef DefaultSoName) {
1156   switch (getELFKind(MB)) {
1157   case ELF32LEKind:
1158     return make<SharedFile<ELF32LE>>(MB, DefaultSoName);
1159   case ELF32BEKind:
1160     return make<SharedFile<ELF32BE>>(MB, DefaultSoName);
1161   case ELF64LEKind:
1162     return make<SharedFile<ELF64LE>>(MB, DefaultSoName);
1163   case ELF64BEKind:
1164     return make<SharedFile<ELF64BE>>(MB, DefaultSoName);
1165   default:
1166     llvm_unreachable("getELFKind");
1167   }
1168 }
1169 
1170 MemoryBufferRef LazyObjFile::getBuffer() {
1171   if (Seen)
1172     return MemoryBufferRef();
1173   Seen = true;
1174   return MB;
1175 }
1176 
1177 InputFile *LazyObjFile::fetch() {
1178   MemoryBufferRef MBRef = getBuffer();
1179   if (MBRef.getBuffer().empty())
1180     return nullptr;
1181 
1182   InputFile *File = createObjectFile(MBRef, ArchiveName, OffsetInArchive);
1183   File->GroupId = GroupId;
1184   return File;
1185 }
1186 
1187 template <class ELFT> void LazyObjFile::parse() {
1188   // A lazy object file wraps either a bitcode file or an ELF file.
1189   if (isBitcode(this->MB)) {
1190     std::unique_ptr<lto::InputFile> Obj =
1191         CHECK(lto::InputFile::create(this->MB), this);
1192     for (const lto::InputFile::Symbol &Sym : Obj->symbols())
1193       if (!Sym.isUndefined())
1194         Symtab->addLazyObject<ELFT>(Saver.save(Sym.getName()), *this);
1195     return;
1196   }
1197 
1198   switch (getELFKind(this->MB)) {
1199   case ELF32LEKind:
1200     addElfSymbols<ELF32LE>();
1201     return;
1202   case ELF32BEKind:
1203     addElfSymbols<ELF32BE>();
1204     return;
1205   case ELF64LEKind:
1206     addElfSymbols<ELF64LE>();
1207     return;
1208   case ELF64BEKind:
1209     addElfSymbols<ELF64BE>();
1210     return;
1211   default:
1212     llvm_unreachable("getELFKind");
1213   }
1214 }
1215 
1216 template <class ELFT> void LazyObjFile::addElfSymbols() {
1217   ELFFile<ELFT> Obj = check(ELFFile<ELFT>::create(MB.getBuffer()));
1218   ArrayRef<typename ELFT::Shdr> Sections = CHECK(Obj.sections(), this);
1219 
1220   for (const typename ELFT::Shdr &Sec : Sections) {
1221     if (Sec.sh_type != SHT_SYMTAB)
1222       continue;
1223 
1224     typename ELFT::SymRange Syms = CHECK(Obj.symbols(&Sec), this);
1225     uint32_t FirstGlobal = Sec.sh_info;
1226     StringRef StringTable =
1227         CHECK(Obj.getStringTableForSymtab(Sec, Sections), this);
1228 
1229     for (const typename ELFT::Sym &Sym : Syms.slice(FirstGlobal))
1230       if (Sym.st_shndx != SHN_UNDEF)
1231         Symtab->addLazyObject<ELFT>(CHECK(Sym.getName(StringTable), this),
1232                                     *this);
1233     return;
1234   }
1235 }
1236 
1237 template void ArchiveFile::parse<ELF32LE>();
1238 template void ArchiveFile::parse<ELF32BE>();
1239 template void ArchiveFile::parse<ELF64LE>();
1240 template void ArchiveFile::parse<ELF64BE>();
1241 
1242 template void BitcodeFile::parse<ELF32LE>(DenseSet<CachedHashStringRef> &);
1243 template void BitcodeFile::parse<ELF32BE>(DenseSet<CachedHashStringRef> &);
1244 template void BitcodeFile::parse<ELF64LE>(DenseSet<CachedHashStringRef> &);
1245 template void BitcodeFile::parse<ELF64BE>(DenseSet<CachedHashStringRef> &);
1246 
1247 template void LazyObjFile::parse<ELF32LE>();
1248 template void LazyObjFile::parse<ELF32BE>();
1249 template void LazyObjFile::parse<ELF64LE>();
1250 template void LazyObjFile::parse<ELF64BE>();
1251 
1252 template class elf::ELFFileBase<ELF32LE>;
1253 template class elf::ELFFileBase<ELF32BE>;
1254 template class elf::ELFFileBase<ELF64LE>;
1255 template class elf::ELFFileBase<ELF64BE>;
1256 
1257 template class elf::ObjFile<ELF32LE>;
1258 template class elf::ObjFile<ELF32BE>;
1259 template class elf::ObjFile<ELF64LE>;
1260 template class elf::ObjFile<ELF64BE>;
1261 
1262 template class elf::SharedFile<ELF32LE>;
1263 template class elf::SharedFile<ELF32BE>;
1264 template class elf::SharedFile<ELF64LE>;
1265 template class elf::SharedFile<ELF64BE>;
1266