1 //===- InputFiles.cpp -----------------------------------------------------===//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 #include "InputFiles.h"
11 #include "InputSection.h"
12 #include "LinkerScript.h"
13 #include "SymbolTable.h"
15 #include "SyntheticSections.h"
16 #include "lld/Common/DWARF.h"
17 #include "lld/Common/ErrorHandler.h"
18 #include "lld/Common/Memory.h"
19 #include "llvm/ADT/STLExtras.h"
20 #include "llvm/CodeGen/Analysis.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/Endian.h"
29 #include "llvm/Support/Path.h"
30 #include "llvm/Support/TarWriter.h"
31 #include "llvm/Support/raw_ostream.h"
34 using namespace llvm::ELF;
35 using namespace llvm::object;
36 using namespace llvm::sys;
37 using namespace llvm::sys::fs;
38 using namespace llvm::support::endian;
41 // Returns "<internal>", "foo.a(bar.o)" or "baz.o".
42 std::string toString(const elf::InputFile *f) {
46 if (f->toStringCache.empty()) {
47 if (f->archiveName.empty())
48 f->toStringCache = f->getName();
50 f->toStringCache = (f->archiveName + "(" + f->getName() + ")").str();
52 return f->toStringCache;
56 bool InputFile::isInGroup;
57 uint32_t InputFile::nextGroupId;
58 std::vector<BinaryFile *> binaryFiles;
59 std::vector<BitcodeFile *> bitcodeFiles;
60 std::vector<LazyObjFile *> lazyObjFiles;
61 std::vector<InputFile *> objectFiles;
62 std::vector<SharedFile *> sharedFiles;
64 std::unique_ptr<TarWriter> tar;
66 static ELFKind getELFKind(MemoryBufferRef mb, StringRef archiveName) {
69 std::tie(size, endian) = getElfArchType(mb.getBuffer());
71 auto report = [&](StringRef msg) {
72 StringRef filename = mb.getBufferIdentifier();
73 if (archiveName.empty())
74 fatal(filename + ": " + msg);
76 fatal(archiveName + "(" + filename + "): " + msg);
79 if (!mb.getBuffer().startswith(ElfMagic))
80 report("not an ELF file");
81 if (endian != ELFDATA2LSB && endian != ELFDATA2MSB)
82 report("corrupted ELF file: invalid data encoding");
83 if (size != ELFCLASS32 && size != ELFCLASS64)
84 report("corrupted ELF file: invalid file class");
86 size_t bufSize = mb.getBuffer().size();
87 if ((size == ELFCLASS32 && bufSize < sizeof(Elf32_Ehdr)) ||
88 (size == ELFCLASS64 && bufSize < sizeof(Elf64_Ehdr)))
89 report("corrupted ELF file: file is too short");
91 if (size == ELFCLASS32)
92 return (endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind;
93 return (endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind;
96 InputFile::InputFile(Kind k, MemoryBufferRef m)
97 : mb(m), groupId(nextGroupId), fileKind(k) {
98 // All files within the same --{start,end}-group get the same group ID.
99 // Otherwise, a new file will get a new group ID.
104 Optional<MemoryBufferRef> readFile(StringRef path) {
105 // The --chroot option changes our virtual root directory.
106 // This is useful when you are dealing with files created by --reproduce.
107 if (!config->chroot.empty() && path.startswith("/"))
108 path = saver.save(config->chroot + path);
112 auto mbOrErr = MemoryBuffer::getFile(path, -1, false);
113 if (auto ec = mbOrErr.getError()) {
114 error("cannot open " + path + ": " + ec.message());
118 std::unique_ptr<MemoryBuffer> &mb = *mbOrErr;
119 MemoryBufferRef mbref = mb->getMemBufferRef();
120 make<std::unique_ptr<MemoryBuffer>>(std::move(mb)); // take MB ownership
123 tar->append(relativeToRoot(path), mbref.getBuffer());
127 // All input object files must be for the same architecture
128 // (e.g. it does not make sense to link x86 object files with
129 // MIPS object files.) This function checks for that error.
130 static bool isCompatible(InputFile *file) {
131 if (!file->isElf() && !isa<BitcodeFile>(file))
134 if (file->ekind == config->ekind && file->emachine == config->emachine) {
135 if (config->emachine != EM_MIPS)
137 if (isMipsN32Abi(file) == config->mipsN32Abi)
141 if (!config->emulation.empty()) {
142 error(toString(file) + " is incompatible with " + config->emulation);
147 if (!objectFiles.empty())
148 existing = objectFiles[0];
149 else if (!sharedFiles.empty())
150 existing = sharedFiles[0];
152 existing = bitcodeFiles[0];
154 error(toString(file) + " is incompatible with " + toString(existing));
158 template <class ELFT> static void doParseFile(InputFile *file) {
159 if (!isCompatible(file))
163 if (auto *f = dyn_cast<BinaryFile>(file)) {
164 binaryFiles.push_back(f);
170 if (auto *f = dyn_cast<ArchiveFile>(file)) {
176 if (auto *f = dyn_cast<LazyObjFile>(file)) {
177 lazyObjFiles.push_back(f);
183 message(toString(file));
186 if (auto *f = dyn_cast<SharedFile>(file)) {
192 if (auto *f = dyn_cast<BitcodeFile>(file)) {
193 bitcodeFiles.push_back(f);
198 // Regular object file
199 objectFiles.push_back(file);
200 cast<ObjFile<ELFT>>(file)->parse();
203 // Add symbols in File to the symbol table.
204 void parseFile(InputFile *file) {
205 switch (config->ekind) {
207 doParseFile<ELF32LE>(file);
210 doParseFile<ELF32BE>(file);
213 doParseFile<ELF64LE>(file);
216 doParseFile<ELF64BE>(file);
219 llvm_unreachable("unknown ELFT");
223 // Concatenates arguments to construct a string representing an error location.
224 static std::string createFileLineMsg(StringRef path, unsigned line) {
225 std::string filename = path::filename(path);
226 std::string lineno = ":" + std::to_string(line);
227 if (filename == path)
228 return filename + lineno;
229 return filename + lineno + " (" + path.str() + lineno + ")";
232 template <class ELFT>
233 static std::string getSrcMsgAux(ObjFile<ELFT> &file, const Symbol &sym,
234 InputSectionBase &sec, uint64_t offset) {
235 // In DWARF, functions and variables are stored to different places.
236 // First, lookup a function for a given offset.
237 if (Optional<DILineInfo> info = file.getDILineInfo(&sec, offset))
238 return createFileLineMsg(info->FileName, info->Line);
240 // If it failed, lookup again as a variable.
241 if (Optional<std::pair<std::string, unsigned>> fileLine =
242 file.getVariableLoc(sym.getName()))
243 return createFileLineMsg(fileLine->first, fileLine->second);
245 // File.sourceFile contains STT_FILE symbol, and that is a last resort.
246 return file.sourceFile;
249 std::string InputFile::getSrcMsg(const Symbol &sym, InputSectionBase &sec,
251 if (kind() != ObjKind)
253 switch (config->ekind) {
255 llvm_unreachable("Invalid kind");
257 return getSrcMsgAux(cast<ObjFile<ELF32LE>>(*this), sym, sec, offset);
259 return getSrcMsgAux(cast<ObjFile<ELF32BE>>(*this), sym, sec, offset);
261 return getSrcMsgAux(cast<ObjFile<ELF64LE>>(*this), sym, sec, offset);
263 return getSrcMsgAux(cast<ObjFile<ELF64BE>>(*this), sym, sec, offset);
267 template <class ELFT> void ObjFile<ELFT>::initializeDwarf() {
268 dwarf = make<DWARFCache>(std::make_unique<DWARFContext>(
269 std::make_unique<LLDDwarfObj<ELFT>>(this)));
272 // Returns the pair of file name and line number describing location of data
273 // object (variable, array, etc) definition.
274 template <class ELFT>
275 Optional<std::pair<std::string, unsigned>>
276 ObjFile<ELFT>::getVariableLoc(StringRef name) {
277 llvm::call_once(initDwarfLine, [this]() { initializeDwarf(); });
279 return dwarf->getVariableLoc(name);
282 // Returns source line information for a given offset
283 // using DWARF debug info.
284 template <class ELFT>
285 Optional<DILineInfo> ObjFile<ELFT>::getDILineInfo(InputSectionBase *s,
287 llvm::call_once(initDwarfLine, [this]() { initializeDwarf(); });
289 // Detect SectionIndex for specified section.
290 uint64_t sectionIndex = object::SectionedAddress::UndefSection;
291 ArrayRef<InputSectionBase *> sections = s->file->getSections();
292 for (uint64_t curIndex = 0; curIndex < sections.size(); ++curIndex) {
293 if (s == sections[curIndex]) {
294 sectionIndex = curIndex;
299 // Use fake address calculated by adding section file offset and offset in
300 // section. See comments for ObjectInfo class.
301 return dwarf->getDILineInfo(s->getOffsetInFile() + offset, sectionIndex);
304 ELFFileBase::ELFFileBase(Kind k, MemoryBufferRef mb) : InputFile(k, mb) {
305 ekind = getELFKind(mb, "");
321 llvm_unreachable("getELFKind");
325 template <typename Elf_Shdr>
326 static const Elf_Shdr *findSection(ArrayRef<Elf_Shdr> sections, uint32_t type) {
327 for (const Elf_Shdr &sec : sections)
328 if (sec.sh_type == type)
333 template <class ELFT> void ELFFileBase::init() {
334 using Elf_Shdr = typename ELFT::Shdr;
335 using Elf_Sym = typename ELFT::Sym;
337 // Initialize trivial attributes.
338 const ELFFile<ELFT> &obj = getObj<ELFT>();
339 emachine = obj.getHeader()->e_machine;
340 osabi = obj.getHeader()->e_ident[llvm::ELF::EI_OSABI];
341 abiVersion = obj.getHeader()->e_ident[llvm::ELF::EI_ABIVERSION];
343 ArrayRef<Elf_Shdr> sections = CHECK(obj.sections(), this);
345 // Find a symbol table.
347 (identify_magic(mb.getBuffer()) == file_magic::elf_shared_object);
348 const Elf_Shdr *symtabSec =
349 findSection(sections, isDSO ? SHT_DYNSYM : SHT_SYMTAB);
354 // Initialize members corresponding to a symbol table.
355 firstGlobal = symtabSec->sh_info;
357 ArrayRef<Elf_Sym> eSyms = CHECK(obj.symbols(symtabSec), this);
358 if (firstGlobal == 0 || firstGlobal > eSyms.size())
359 fatal(toString(this) + ": invalid sh_info in symbol table");
361 elfSyms = reinterpret_cast<const void *>(eSyms.data());
362 numELFSyms = eSyms.size();
363 stringTable = CHECK(obj.getStringTableForSymtab(*symtabSec, sections), this);
366 template <class ELFT>
367 uint32_t ObjFile<ELFT>::getSectionIndex(const Elf_Sym &sym) const {
369 this->getObj().getSectionIndex(&sym, getELFSyms<ELFT>(), shndxTable),
373 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() {
374 if (this->symbols.empty())
376 return makeArrayRef(this->symbols).slice(1, this->firstGlobal - 1);
379 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getGlobalSymbols() {
380 return makeArrayRef(this->symbols).slice(this->firstGlobal);
383 template <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) {
384 // Read a section table. justSymbols is usually false.
385 if (this->justSymbols)
386 initializeJustSymbols();
388 initializeSections(ignoreComdats);
390 // Read a symbol table.
394 // Sections with SHT_GROUP and comdat bits define comdat section groups.
395 // They are identified and deduplicated by group name. This function
396 // returns a group name.
397 template <class ELFT>
398 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> sections,
399 const Elf_Shdr &sec) {
400 typename ELFT::SymRange symbols = this->getELFSyms<ELFT>();
401 if (sec.sh_info >= symbols.size())
402 fatal(toString(this) + ": invalid symbol index");
403 const typename ELFT::Sym &sym = symbols[sec.sh_info];
404 StringRef signature = CHECK(sym.getName(this->stringTable), this);
406 // As a special case, if a symbol is a section symbol and has no name,
407 // we use a section name as a signature.
409 // Such SHT_GROUP sections are invalid from the perspective of the ELF
410 // standard, but GNU gold 1.14 (the newest version as of July 2017) or
411 // older produce such sections as outputs for the -r option, so we need
412 // a bug-compatibility.
413 if (signature.empty() && sym.getType() == STT_SECTION)
414 return getSectionName(sec);
418 template <class ELFT>
419 bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &sec, StringRef name) {
420 // On a regular link we don't merge sections if -O0 (default is -O1). This
421 // sometimes makes the linker significantly faster, although the output will
424 // Doing the same for -r would create a problem as it would combine sections
425 // with different sh_entsize. One option would be to just copy every SHF_MERGE
426 // section as is to the output. While this would produce a valid ELF file with
427 // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when
428 // they see two .debug_str. We could have separate logic for combining
429 // SHF_MERGE sections based both on their name and sh_entsize, but that seems
430 // to be more trouble than it is worth. Instead, we just use the regular (-O1)
432 if (config->optimize == 0 && !config->relocatable)
435 // A mergeable section with size 0 is useless because they don't have
436 // any data to merge. A mergeable string section with size 0 can be
437 // argued as invalid because it doesn't end with a null character.
438 // We'll avoid a mess by handling them as if they were non-mergeable.
439 if (sec.sh_size == 0)
442 // Check for sh_entsize. The ELF spec is not clear about the zero
443 // sh_entsize. It says that "the member [sh_entsize] contains 0 if
444 // the section does not hold a table of fixed-size entries". We know
445 // that Rust 1.13 produces a string mergeable section with a zero
446 // sh_entsize. Here we just accept it rather than being picky about it.
447 uint64_t entSize = sec.sh_entsize;
450 if (sec.sh_size % entSize)
451 fatal(toString(this) + ":(" + name + "): SHF_MERGE section size (" +
452 Twine(sec.sh_size) + ") must be a multiple of sh_entsize (" +
453 Twine(entSize) + ")");
455 uint64_t flags = sec.sh_flags;
456 if (!(flags & SHF_MERGE))
458 if (flags & SHF_WRITE)
459 fatal(toString(this) + ":(" + name +
460 "): writable SHF_MERGE section is not supported");
465 // This is for --just-symbols.
467 // --just-symbols is a very minor feature that allows you to link your
468 // output against other existing program, so that if you load both your
469 // program and the other program into memory, your output can refer the
470 // other program's symbols.
472 // When the option is given, we link "just symbols". The section table is
473 // initialized with null pointers.
474 template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() {
475 ArrayRef<Elf_Shdr> sections = CHECK(this->getObj().sections(), this);
476 this->sections.resize(sections.size());
479 // An ELF object file may contain a `.deplibs` section. If it exists, the
480 // section contains a list of library specifiers such as `m` for libm. This
481 // function resolves a given name by finding the first matching library checking
482 // the various ways that a library can be specified to LLD. This ELF extension
483 // is a form of autolinking and is called `dependent libraries`. It is currently
484 // unique to LLVM and lld.
485 static void addDependentLibrary(StringRef specifier, const InputFile *f) {
486 if (!config->dependentLibraries)
488 if (fs::exists(specifier))
489 driver->addFile(specifier, /*withLOption=*/false);
490 else if (Optional<std::string> s = findFromSearchPaths(specifier))
491 driver->addFile(*s, /*withLOption=*/true);
492 else if (Optional<std::string> s = searchLibraryBaseName(specifier))
493 driver->addFile(*s, /*withLOption=*/true);
496 ": unable to find library from dependent library specifier: " +
500 // Record the membership of a section group so that in the garbage collection
501 // pass, section group members are kept or discarded as a unit.
502 template <class ELFT>
503 static void handleSectionGroup(ArrayRef<InputSectionBase *> sections,
504 ArrayRef<typename ELFT::Word> entries) {
505 bool hasAlloc = false;
506 for (uint32_t index : entries.slice(1)) {
507 if (index >= sections.size())
509 if (InputSectionBase *s = sections[index])
510 if (s != &InputSection::discarded && s->flags & SHF_ALLOC)
514 // If any member has the SHF_ALLOC flag, the whole group is subject to garbage
515 // collection. See the comment in markLive(). This rule retains .debug_types
516 // and .rela.debug_types.
520 // Connect the members in a circular doubly-linked list via
521 // nextInSectionGroup.
522 InputSectionBase *head;
523 InputSectionBase *prev = nullptr;
524 for (uint32_t index : entries.slice(1)) {
525 InputSectionBase *s = sections[index];
526 if (!s || s == &InputSection::discarded)
529 prev->nextInSectionGroup = s;
535 prev->nextInSectionGroup = head;
538 template <class ELFT>
539 void ObjFile<ELFT>::initializeSections(bool ignoreComdats) {
540 const ELFFile<ELFT> &obj = this->getObj();
542 ArrayRef<Elf_Shdr> objSections = CHECK(obj.sections(), this);
543 uint64_t size = objSections.size();
544 this->sections.resize(size);
545 this->sectionStringTable =
546 CHECK(obj.getSectionStringTable(objSections), this);
548 std::vector<ArrayRef<Elf_Word>> selectedGroups;
550 for (size_t i = 0, e = objSections.size(); i < e; ++i) {
551 if (this->sections[i] == &InputSection::discarded)
553 const Elf_Shdr &sec = objSections[i];
555 if (sec.sh_type == ELF::SHT_LLVM_CALL_GRAPH_PROFILE)
557 check(obj.template getSectionContentsAsArray<Elf_CGProfile>(&sec));
559 // SHF_EXCLUDE'ed sections are discarded by the linker. However,
560 // if -r is given, we'll let the final link discard such sections.
561 // This is compatible with GNU.
562 if ((sec.sh_flags & SHF_EXCLUDE) && !config->relocatable) {
563 if (sec.sh_type == SHT_LLVM_ADDRSIG) {
564 // We ignore the address-significance table if we know that the object
565 // file was created by objcopy or ld -r. This is because these tools
566 // will reorder the symbols in the symbol table, invalidating the data
567 // in the address-significance table, which refers to symbols by index.
568 if (sec.sh_link != 0)
569 this->addrsigSec = &sec;
570 else if (config->icf == ICFLevel::Safe)
571 warn(toString(this) + ": --icf=safe is incompatible with object "
572 "files created using objcopy or ld -r");
574 this->sections[i] = &InputSection::discarded;
578 switch (sec.sh_type) {
580 // De-duplicate section groups by their signatures.
581 StringRef signature = getShtGroupSignature(objSections, sec);
582 this->sections[i] = &InputSection::discarded;
585 ArrayRef<Elf_Word> entries =
586 CHECK(obj.template getSectionContentsAsArray<Elf_Word>(&sec), this);
588 fatal(toString(this) + ": empty SHT_GROUP");
590 // The first word of a SHT_GROUP section contains flags. Currently,
591 // the standard defines only "GRP_COMDAT" flag for the COMDAT group.
592 // An group with the empty flag doesn't define anything; such sections
597 if (entries[0] != GRP_COMDAT)
598 fatal(toString(this) + ": unsupported SHT_GROUP format");
602 symtab->comdatGroups.try_emplace(CachedHashStringRef(signature), this)
605 if (config->relocatable)
606 this->sections[i] = createInputSection(sec);
607 selectedGroups.push_back(entries);
611 // Otherwise, discard group members.
612 for (uint32_t secIndex : entries.slice(1)) {
613 if (secIndex >= size)
614 fatal(toString(this) +
615 ": invalid section index in group: " + Twine(secIndex));
616 this->sections[secIndex] = &InputSection::discarded;
620 case SHT_SYMTAB_SHNDX:
621 shndxTable = CHECK(obj.getSHNDXTable(sec, objSections), this);
628 this->sections[i] = createInputSection(sec);
632 // This block handles SHF_LINK_ORDER.
633 for (size_t i = 0, e = objSections.size(); i < e; ++i) {
634 if (this->sections[i] == &InputSection::discarded)
636 const Elf_Shdr &sec = objSections[i];
637 if (!(sec.sh_flags & SHF_LINK_ORDER))
640 // .ARM.exidx sections have a reverse dependency on the InputSection they
641 // have a SHF_LINK_ORDER dependency, this is identified by the sh_link.
642 InputSectionBase *linkSec = nullptr;
643 if (sec.sh_link < this->sections.size())
644 linkSec = this->sections[sec.sh_link];
646 fatal(toString(this) + ": invalid sh_link index: " + Twine(sec.sh_link));
648 InputSection *isec = cast<InputSection>(this->sections[i]);
649 linkSec->dependentSections.push_back(isec);
650 if (!isa<InputSection>(linkSec))
651 error("a section " + isec->name +
652 " with SHF_LINK_ORDER should not refer a non-regular section: " +
656 for (ArrayRef<Elf_Word> entries : selectedGroups)
657 handleSectionGroup<ELFT>(this->sections, entries);
660 // For ARM only, to set the EF_ARM_ABI_FLOAT_SOFT or EF_ARM_ABI_FLOAT_HARD
661 // flag in the ELF Header we need to look at Tag_ABI_VFP_args to find out how
662 // the input objects have been compiled.
663 static void updateARMVFPArgs(const ARMAttributeParser &attributes,
664 const InputFile *f) {
665 if (!attributes.hasAttribute(ARMBuildAttrs::ABI_VFP_args))
666 // If an ABI tag isn't present then it is implicitly given the value of 0
667 // which maps to ARMBuildAttrs::BaseAAPCS. However many assembler files,
668 // including some in glibc that don't use FP args (and should have value 3)
669 // don't have the attribute so we do not consider an implicit value of 0
673 unsigned vfpArgs = attributes.getAttributeValue(ARMBuildAttrs::ABI_VFP_args);
676 case ARMBuildAttrs::BaseAAPCS:
677 arg = ARMVFPArgKind::Base;
679 case ARMBuildAttrs::HardFPAAPCS:
680 arg = ARMVFPArgKind::VFP;
682 case ARMBuildAttrs::ToolChainFPPCS:
683 // Tool chain specific convention that conforms to neither AAPCS variant.
684 arg = ARMVFPArgKind::ToolChain;
686 case ARMBuildAttrs::CompatibleFPAAPCS:
687 // Object compatible with all conventions.
690 error(toString(f) + ": unknown Tag_ABI_VFP_args value: " + Twine(vfpArgs));
693 // Follow ld.bfd and error if there is a mix of calling conventions.
694 if (config->armVFPArgs != arg && config->armVFPArgs != ARMVFPArgKind::Default)
695 error(toString(f) + ": incompatible Tag_ABI_VFP_args");
697 config->armVFPArgs = arg;
700 // The ARM support in lld makes some use of instructions that are not available
701 // on all ARM architectures. Namely:
702 // - Use of BLX instruction for interworking between ARM and Thumb state.
703 // - Use of the extended Thumb branch encoding in relocation.
704 // - Use of the MOVT/MOVW instructions in Thumb Thunks.
705 // The ARM Attributes section contains information about the architecture chosen
706 // at compile time. We follow the convention that if at least one input object
707 // is compiled with an architecture that supports these features then lld is
708 // permitted to use them.
709 static void updateSupportedARMFeatures(const ARMAttributeParser &attributes) {
710 if (!attributes.hasAttribute(ARMBuildAttrs::CPU_arch))
712 auto arch = attributes.getAttributeValue(ARMBuildAttrs::CPU_arch);
714 case ARMBuildAttrs::Pre_v4:
715 case ARMBuildAttrs::v4:
716 case ARMBuildAttrs::v4T:
717 // Architectures prior to v5 do not support BLX instruction
719 case ARMBuildAttrs::v5T:
720 case ARMBuildAttrs::v5TE:
721 case ARMBuildAttrs::v5TEJ:
722 case ARMBuildAttrs::v6:
723 case ARMBuildAttrs::v6KZ:
724 case ARMBuildAttrs::v6K:
725 config->armHasBlx = true;
726 // Architectures used in pre-Cortex processors do not support
727 // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception
728 // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do.
731 // All other Architectures have BLX and extended branch encoding
732 config->armHasBlx = true;
733 config->armJ1J2BranchEncoding = true;
734 if (arch != ARMBuildAttrs::v6_M && arch != ARMBuildAttrs::v6S_M)
735 // All Architectures used in Cortex processors with the exception
736 // of v6-M and v6S-M have the MOVT and MOVW instructions.
737 config->armHasMovtMovw = true;
742 // If a source file is compiled with x86 hardware-assisted call flow control
743 // enabled, the generated object file contains feature flags indicating that
744 // fact. This function reads the feature flags and returns it.
746 // Essentially we want to read a single 32-bit value in this function, but this
747 // function is rather complicated because the value is buried deep inside a
748 // .note.gnu.property section.
750 // The section consists of one or more NOTE records. Each NOTE record consists
751 // of zero or more type-length-value fields. We want to find a field of a
752 // certain type. It seems a bit too much to just store a 32-bit value, perhaps
753 // the ABI is unnecessarily complicated.
754 template <class ELFT>
755 static uint32_t readAndFeatures(ObjFile<ELFT> *obj, ArrayRef<uint8_t> data) {
756 using Elf_Nhdr = typename ELFT::Nhdr;
757 using Elf_Note = typename ELFT::Note;
759 uint32_t featuresSet = 0;
760 while (!data.empty()) {
761 // Read one NOTE record.
762 if (data.size() < sizeof(Elf_Nhdr))
763 fatal(toString(obj) + ": .note.gnu.property: section too short");
765 auto *nhdr = reinterpret_cast<const Elf_Nhdr *>(data.data());
766 if (data.size() < nhdr->getSize())
767 fatal(toString(obj) + ": .note.gnu.property: section too short");
769 Elf_Note note(*nhdr);
770 if (nhdr->n_type != NT_GNU_PROPERTY_TYPE_0 || note.getName() != "GNU") {
771 data = data.slice(nhdr->getSize());
775 uint32_t featureAndType = config->emachine == EM_AARCH64
776 ? GNU_PROPERTY_AARCH64_FEATURE_1_AND
777 : GNU_PROPERTY_X86_FEATURE_1_AND;
779 // Read a body of a NOTE record, which consists of type-length-value fields.
780 ArrayRef<uint8_t> desc = note.getDesc();
781 while (!desc.empty()) {
783 fatal(toString(obj) + ": .note.gnu.property: section too short");
785 uint32_t type = read32le(desc.data());
786 uint32_t size = read32le(desc.data() + 4);
788 if (type == featureAndType) {
789 // We found a FEATURE_1_AND field. There may be more than one of these
790 // in a .note.gnu.property section, for a relocatable object we
791 // accumulate the bits set.
792 featuresSet |= read32le(desc.data() + 8);
795 // On 64-bit, a payload may be followed by a 4-byte padding to make its
796 // size a multiple of 8.
798 size = alignTo(size, 8);
800 desc = desc.slice(size + 8); // +8 for Type and Size
803 // Go to next NOTE record to look for more FEATURE_1_AND descriptions.
804 data = data.slice(nhdr->getSize());
810 template <class ELFT>
811 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &sec) {
812 uint32_t idx = sec.sh_info;
813 if (idx >= this->sections.size())
814 fatal(toString(this) + ": invalid relocated section index: " + Twine(idx));
815 InputSectionBase *target = this->sections[idx];
817 // Strictly speaking, a relocation section must be included in the
818 // group of the section it relocates. However, LLVM 3.3 and earlier
819 // would fail to do so, so we gracefully handle that case.
820 if (target == &InputSection::discarded)
824 fatal(toString(this) + ": unsupported relocation reference");
828 // Create a regular InputSection class that has the same contents
829 // as a given section.
830 static InputSection *toRegularSection(MergeInputSection *sec) {
831 return make<InputSection>(sec->file, sec->flags, sec->type, sec->alignment,
832 sec->data(), sec->name);
835 template <class ELFT>
836 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &sec) {
837 StringRef name = getSectionName(sec);
839 switch (sec.sh_type) {
840 case SHT_ARM_ATTRIBUTES: {
841 if (config->emachine != EM_ARM)
843 ARMAttributeParser attributes;
844 ArrayRef<uint8_t> contents = check(this->getObj().getSectionContents(&sec));
845 attributes.Parse(contents, /*isLittle*/ config->ekind == ELF32LEKind);
846 updateSupportedARMFeatures(attributes);
847 updateARMVFPArgs(attributes, this);
849 // FIXME: Retain the first attribute section we see. The eglibc ARM
850 // dynamic loaders require the presence of an attribute section for dlopen
851 // to work. In a full implementation we would merge all attribute sections.
852 if (in.armAttributes == nullptr) {
853 in.armAttributes = make<InputSection>(*this, sec, name);
854 return in.armAttributes;
856 return &InputSection::discarded;
858 case SHT_LLVM_DEPENDENT_LIBRARIES: {
859 if (config->relocatable)
861 ArrayRef<char> data =
862 CHECK(this->getObj().template getSectionContentsAsArray<char>(&sec), this);
863 if (!data.empty() && data.back() != '\0') {
864 error(toString(this) +
865 ": corrupted dependent libraries section (unterminated string): " +
867 return &InputSection::discarded;
869 for (const char *d = data.begin(), *e = data.end(); d < e;) {
871 addDependentLibrary(s, this);
874 return &InputSection::discarded;
878 // Find a relocation target section and associate this section with that.
879 // Target may have been discarded if it is in a different section group
880 // and the group is discarded, even though it's a violation of the
881 // spec. We handle that situation gracefully by discarding dangling
882 // relocation sections.
883 InputSectionBase *target = getRelocTarget(sec);
887 // ELF spec allows mergeable sections with relocations, but they are
888 // rare, and it is in practice hard to merge such sections by contents,
889 // because applying relocations at end of linking changes section
890 // contents. So, we simply handle such sections as non-mergeable ones.
891 // Degrading like this is acceptable because section merging is optional.
892 if (auto *ms = dyn_cast<MergeInputSection>(target)) {
893 target = toRegularSection(ms);
894 this->sections[sec.sh_info] = target;
897 // This section contains relocation information.
898 // If -r is given, we do not interpret or apply relocation
899 // but just copy relocation sections to output.
900 if (config->relocatable) {
901 InputSection *relocSec = make<InputSection>(*this, sec, name);
902 // We want to add a dependency to target, similar like we do for
903 // -emit-relocs below. This is useful for the case when linker script
904 // contains the "/DISCARD/". It is perhaps uncommon to use a script with
905 // -r, but we faced it in the Linux kernel and have to handle such case
907 target->dependentSections.push_back(relocSec);
911 if (target->firstRelocation)
912 fatal(toString(this) +
913 ": multiple relocation sections to one section are not supported");
915 if (sec.sh_type == SHT_RELA) {
916 ArrayRef<Elf_Rela> rels = CHECK(getObj().relas(&sec), this);
917 target->firstRelocation = rels.begin();
918 target->numRelocations = rels.size();
919 target->areRelocsRela = true;
921 ArrayRef<Elf_Rel> rels = CHECK(getObj().rels(&sec), this);
922 target->firstRelocation = rels.begin();
923 target->numRelocations = rels.size();
924 target->areRelocsRela = false;
926 assert(isUInt<31>(target->numRelocations));
928 // Relocation sections processed by the linker are usually removed
929 // from the output, so returning `nullptr` for the normal case.
930 // However, if -emit-relocs is given, we need to leave them in the output.
931 // (Some post link analysis tools need this information.)
932 if (config->emitRelocs) {
933 InputSection *relocSec = make<InputSection>(*this, sec, name);
934 // We will not emit relocation section if target was discarded.
935 target->dependentSections.push_back(relocSec);
942 // The GNU linker uses .note.GNU-stack section as a marker indicating
943 // that the code in the object file does not expect that the stack is
944 // executable (in terms of NX bit). If all input files have the marker,
945 // the GNU linker adds a PT_GNU_STACK segment to tells the loader to
946 // make the stack non-executable. Most object files have this section as
949 // But making the stack non-executable is a norm today for security
950 // reasons. Failure to do so may result in a serious security issue.
951 // Therefore, we make LLD always add PT_GNU_STACK unless it is
952 // explicitly told to do otherwise (by -z execstack). Because the stack
953 // executable-ness is controlled solely by command line options,
954 // .note.GNU-stack sections are simply ignored.
955 if (name == ".note.GNU-stack")
956 return &InputSection::discarded;
958 // Object files that use processor features such as Intel Control-Flow
959 // Enforcement (CET) or AArch64 Branch Target Identification BTI, use a
960 // .note.gnu.property section containing a bitfield of feature bits like the
961 // GNU_PROPERTY_X86_FEATURE_1_IBT flag. Read a bitmap containing the flag.
963 // Since we merge bitmaps from multiple object files to create a new
964 // .note.gnu.property containing a single AND'ed bitmap, we discard an input
965 // file's .note.gnu.property section.
966 if (name == ".note.gnu.property") {
967 ArrayRef<uint8_t> contents = check(this->getObj().getSectionContents(&sec));
968 this->andFeatures = readAndFeatures(this, contents);
969 return &InputSection::discarded;
972 // Split stacks is a feature to support a discontiguous stack,
973 // commonly used in the programming language Go. For the details,
974 // see https://gcc.gnu.org/wiki/SplitStacks. An object file compiled
975 // for split stack will include a .note.GNU-split-stack section.
976 if (name == ".note.GNU-split-stack") {
977 if (config->relocatable) {
978 error("cannot mix split-stack and non-split-stack in a relocatable link");
979 return &InputSection::discarded;
981 this->splitStack = true;
982 return &InputSection::discarded;
985 // An object file cmpiled for split stack, but where some of the
986 // functions were compiled with the no_split_stack_attribute will
987 // include a .note.GNU-no-split-stack section.
988 if (name == ".note.GNU-no-split-stack") {
989 this->someNoSplitStack = true;
990 return &InputSection::discarded;
993 // The linkonce feature is a sort of proto-comdat. Some glibc i386 object
994 // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce
995 // sections. Drop those sections to avoid duplicate symbol errors.
996 // FIXME: This is glibc PR20543, we should remove this hack once that has been
997 // fixed for a while.
998 if (name == ".gnu.linkonce.t.__x86.get_pc_thunk.bx" ||
999 name == ".gnu.linkonce.t.__i686.get_pc_thunk.bx")
1000 return &InputSection::discarded;
1002 // If we are creating a new .build-id section, strip existing .build-id
1003 // sections so that the output won't have more than one .build-id.
1004 // This is not usually a problem because input object files normally don't
1005 // have .build-id sections, but you can create such files by
1006 // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it.
1007 if (name == ".note.gnu.build-id" && config->buildId != BuildIdKind::None)
1008 return &InputSection::discarded;
1010 // The linker merges EH (exception handling) frames and creates a
1011 // .eh_frame_hdr section for runtime. So we handle them with a special
1012 // class. For relocatable outputs, they are just passed through.
1013 if (name == ".eh_frame" && !config->relocatable)
1014 return make<EhInputSection>(*this, sec, name);
1016 if (shouldMerge(sec, name))
1017 return make<MergeInputSection>(*this, sec, name);
1018 return make<InputSection>(*this, sec, name);
1021 template <class ELFT>
1022 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &sec) {
1023 return CHECK(getObj().getSectionName(&sec, sectionStringTable), this);
1026 // Initialize this->Symbols. this->Symbols is a parallel array as
1027 // its corresponding ELF symbol table.
1028 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() {
1029 ArrayRef<Elf_Sym> eSyms = this->getELFSyms<ELFT>();
1030 this->symbols.resize(eSyms.size());
1032 // Our symbol table may have already been partially initialized
1033 // because of LazyObjFile.
1034 for (size_t i = 0, end = eSyms.size(); i != end; ++i)
1035 if (!this->symbols[i] && eSyms[i].getBinding() != STB_LOCAL)
1037 symtab->insert(CHECK(eSyms[i].getName(this->stringTable), this));
1039 // Fill this->Symbols. A symbol is either local or global.
1040 for (size_t i = 0, end = eSyms.size(); i != end; ++i) {
1041 const Elf_Sym &eSym = eSyms[i];
1043 // Read symbol attributes.
1044 uint32_t secIdx = getSectionIndex(eSym);
1045 if (secIdx >= this->sections.size())
1046 fatal(toString(this) + ": invalid section index: " + Twine(secIdx));
1048 InputSectionBase *sec = this->sections[secIdx];
1049 uint8_t binding = eSym.getBinding();
1050 uint8_t stOther = eSym.st_other;
1051 uint8_t type = eSym.getType();
1052 uint64_t value = eSym.st_value;
1053 uint64_t size = eSym.st_size;
1054 StringRefZ name = this->stringTable.data() + eSym.st_name;
1056 // Handle local symbols. Local symbols are not added to the symbol
1057 // table because they are not visible from other object files. We
1058 // allocate symbol instances and add their pointers to Symbols.
1059 if (binding == STB_LOCAL) {
1060 if (eSym.getType() == STT_FILE)
1061 sourceFile = CHECK(eSym.getName(this->stringTable), this);
1063 if (this->stringTable.size() <= eSym.st_name)
1064 fatal(toString(this) + ": invalid symbol name offset");
1066 if (eSym.st_shndx == SHN_UNDEF)
1067 this->symbols[i] = make<Undefined>(this, name, binding, stOther, type);
1068 else if (sec == &InputSection::discarded)
1069 this->symbols[i] = make<Undefined>(this, name, binding, stOther, type,
1070 /*DiscardedSecIdx=*/secIdx);
1073 make<Defined>(this, name, binding, stOther, type, value, size, sec);
1077 // Handle global undefined symbols.
1078 if (eSym.st_shndx == SHN_UNDEF) {
1079 this->symbols[i]->resolve(Undefined{this, name, binding, stOther, type});
1080 this->symbols[i]->referenced = true;
1084 // Handle global common symbols.
1085 if (eSym.st_shndx == SHN_COMMON) {
1086 if (value == 0 || value >= UINT32_MAX)
1087 fatal(toString(this) + ": common symbol '" + StringRef(name.data) +
1088 "' has invalid alignment: " + Twine(value));
1089 this->symbols[i]->resolve(
1090 CommonSymbol{this, name, binding, stOther, type, value, size});
1094 // If a defined symbol is in a discarded section, handle it as if it
1095 // were an undefined symbol. Such symbol doesn't comply with the
1096 // standard, but in practice, a .eh_frame often directly refer
1097 // COMDAT member sections, and if a comdat group is discarded, some
1098 // defined symbol in a .eh_frame becomes dangling symbols.
1099 if (sec == &InputSection::discarded) {
1100 this->symbols[i]->resolve(
1101 Undefined{this, name, binding, stOther, type, secIdx});
1105 // Handle global defined symbols.
1106 if (binding == STB_GLOBAL || binding == STB_WEAK ||
1107 binding == STB_GNU_UNIQUE) {
1108 this->symbols[i]->resolve(
1109 Defined{this, name, binding, stOther, type, value, size, sec});
1113 fatal(toString(this) + ": unexpected binding: " + Twine((int)binding));
1117 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&file)
1118 : InputFile(ArchiveKind, file->getMemoryBufferRef()),
1119 file(std::move(file)) {}
1121 void ArchiveFile::parse() {
1122 for (const Archive::Symbol &sym : file->symbols())
1123 symtab->addSymbol(LazyArchive{*this, sym});
1126 // Returns a buffer pointing to a member file containing a given symbol.
1127 void ArchiveFile::fetch(const Archive::Symbol &sym) {
1129 CHECK(sym.getMember(), toString(this) +
1130 ": could not get the member for symbol " +
1133 if (!seen.insert(c.getChildOffset()).second)
1136 MemoryBufferRef mb =
1137 CHECK(c.getMemoryBufferRef(),
1139 ": could not get the buffer for the member defining symbol " +
1142 if (tar && c.getParent()->isThin())
1143 tar->append(relativeToRoot(CHECK(c.getFullName(), this)), mb.getBuffer());
1145 InputFile *file = createObjectFile(
1146 mb, getName(), c.getParent()->isThin() ? 0 : c.getChildOffset());
1147 file->groupId = groupId;
1151 unsigned SharedFile::vernauxNum;
1153 // Parse the version definitions in the object file if present, and return a
1154 // vector whose nth element contains a pointer to the Elf_Verdef for version
1155 // identifier n. Version identifiers that are not definitions map to nullptr.
1156 template <typename ELFT>
1157 static std::vector<const void *> parseVerdefs(const uint8_t *base,
1158 const typename ELFT::Shdr *sec) {
1162 // We cannot determine the largest verdef identifier without inspecting
1163 // every Elf_Verdef, but both bfd and gold assign verdef identifiers
1164 // sequentially starting from 1, so we predict that the largest identifier
1165 // will be verdefCount.
1166 unsigned verdefCount = sec->sh_info;
1167 std::vector<const void *> verdefs(verdefCount + 1);
1169 // Build the Verdefs array by following the chain of Elf_Verdef objects
1170 // from the start of the .gnu.version_d section.
1171 const uint8_t *verdef = base + sec->sh_offset;
1172 for (unsigned i = 0; i != verdefCount; ++i) {
1173 auto *curVerdef = reinterpret_cast<const typename ELFT::Verdef *>(verdef);
1174 verdef += curVerdef->vd_next;
1175 unsigned verdefIndex = curVerdef->vd_ndx;
1176 verdefs.resize(verdefIndex + 1);
1177 verdefs[verdefIndex] = curVerdef;
1182 // We do not usually care about alignments of data in shared object
1183 // files because the loader takes care of it. However, if we promote a
1184 // DSO symbol to point to .bss due to copy relocation, we need to keep
1185 // the original alignment requirements. We infer it in this function.
1186 template <typename ELFT>
1187 static uint64_t getAlignment(ArrayRef<typename ELFT::Shdr> sections,
1188 const typename ELFT::Sym &sym) {
1189 uint64_t ret = UINT64_MAX;
1191 ret = 1ULL << countTrailingZeros((uint64_t)sym.st_value);
1192 if (0 < sym.st_shndx && sym.st_shndx < sections.size())
1193 ret = std::min<uint64_t>(ret, sections[sym.st_shndx].sh_addralign);
1194 return (ret > UINT32_MAX) ? 0 : ret;
1197 // Fully parse the shared object file.
1199 // This function parses symbol versions. If a DSO has version information,
1200 // the file has a ".gnu.version_d" section which contains symbol version
1201 // definitions. Each symbol is associated to one version through a table in
1202 // ".gnu.version" section. That table is a parallel array for the symbol
1203 // table, and each table entry contains an index in ".gnu.version_d".
1205 // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for
1206 // VER_NDX_GLOBAL. There's no table entry for these special versions in
1207 // ".gnu.version_d".
1209 // The file format for symbol versioning is perhaps a bit more complicated
1210 // than necessary, but you can easily understand the code if you wrap your
1211 // head around the data structure described above.
1212 template <class ELFT> void SharedFile::parse() {
1213 using Elf_Dyn = typename ELFT::Dyn;
1214 using Elf_Shdr = typename ELFT::Shdr;
1215 using Elf_Sym = typename ELFT::Sym;
1216 using Elf_Verdef = typename ELFT::Verdef;
1217 using Elf_Versym = typename ELFT::Versym;
1219 ArrayRef<Elf_Dyn> dynamicTags;
1220 const ELFFile<ELFT> obj = this->getObj<ELFT>();
1221 ArrayRef<Elf_Shdr> sections = CHECK(obj.sections(), this);
1223 const Elf_Shdr *versymSec = nullptr;
1224 const Elf_Shdr *verdefSec = nullptr;
1226 // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d.
1227 for (const Elf_Shdr &sec : sections) {
1228 switch (sec.sh_type) {
1233 CHECK(obj.template getSectionContentsAsArray<Elf_Dyn>(&sec), this);
1235 case SHT_GNU_versym:
1238 case SHT_GNU_verdef:
1244 if (versymSec && numELFSyms == 0) {
1245 error("SHT_GNU_versym should be associated with symbol table");
1249 // Search for a DT_SONAME tag to initialize this->soName.
1250 for (const Elf_Dyn &dyn : dynamicTags) {
1251 if (dyn.d_tag == DT_NEEDED) {
1252 uint64_t val = dyn.getVal();
1253 if (val >= this->stringTable.size())
1254 fatal(toString(this) + ": invalid DT_NEEDED entry");
1255 dtNeeded.push_back(this->stringTable.data() + val);
1256 } else if (dyn.d_tag == DT_SONAME) {
1257 uint64_t val = dyn.getVal();
1258 if (val >= this->stringTable.size())
1259 fatal(toString(this) + ": invalid DT_SONAME entry");
1260 soName = this->stringTable.data() + val;
1264 // DSOs are uniquified not by filename but by soname.
1265 DenseMap<StringRef, SharedFile *>::iterator it;
1267 std::tie(it, wasInserted) = symtab->soNames.try_emplace(soName, this);
1269 // If a DSO appears more than once on the command line with and without
1270 // --as-needed, --no-as-needed takes precedence over --as-needed because a
1271 // user can add an extra DSO with --no-as-needed to force it to be added to
1272 // the dependency list.
1273 it->second->isNeeded |= isNeeded;
1277 sharedFiles.push_back(this);
1279 verdefs = parseVerdefs<ELFT>(obj.base(), verdefSec);
1281 // Parse ".gnu.version" section which is a parallel array for the symbol
1282 // table. If a given file doesn't have a ".gnu.version" section, we use
1284 size_t size = numELFSyms - firstGlobal;
1285 std::vector<uint32_t> versyms(size, VER_NDX_GLOBAL);
1287 ArrayRef<Elf_Versym> versym =
1288 CHECK(obj.template getSectionContentsAsArray<Elf_Versym>(versymSec),
1290 .slice(firstGlobal);
1291 for (size_t i = 0; i < size; ++i)
1292 versyms[i] = versym[i].vs_index;
1295 // System libraries can have a lot of symbols with versions. Using a
1296 // fixed buffer for computing the versions name (foo@ver) can save a
1297 // lot of allocations.
1298 SmallString<0> versionedNameBuffer;
1300 // Add symbols to the symbol table.
1301 ArrayRef<Elf_Sym> syms = this->getGlobalELFSyms<ELFT>();
1302 for (size_t i = 0; i < syms.size(); ++i) {
1303 const Elf_Sym &sym = syms[i];
1305 // ELF spec requires that all local symbols precede weak or global
1306 // symbols in each symbol table, and the index of first non-local symbol
1307 // is stored to sh_info. If a local symbol appears after some non-local
1308 // symbol, that's a violation of the spec.
1309 StringRef name = CHECK(sym.getName(this->stringTable), this);
1310 if (sym.getBinding() == STB_LOCAL) {
1311 warn("found local symbol '" + name +
1312 "' in global part of symbol table in file " + toString(this));
1316 if (sym.isUndefined()) {
1317 Symbol *s = symtab->addSymbol(
1318 Undefined{this, name, sym.getBinding(), sym.st_other, sym.getType()});
1319 s->exportDynamic = true;
1323 // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly
1324 // assigns VER_NDX_LOCAL to this section global symbol. Here is a
1325 // workaround for this bug.
1326 uint32_t idx = versyms[i] & ~VERSYM_HIDDEN;
1327 if (config->emachine == EM_MIPS && idx == VER_NDX_LOCAL &&
1331 uint32_t alignment = getAlignment<ELFT>(sections, sym);
1332 if (!(versyms[i] & VERSYM_HIDDEN)) {
1333 symtab->addSymbol(SharedSymbol{*this, name, sym.getBinding(),
1334 sym.st_other, sym.getType(), sym.st_value,
1335 sym.st_size, alignment, idx});
1338 // Also add the symbol with the versioned name to handle undefined symbols
1339 // with explicit versions.
1340 if (idx == VER_NDX_GLOBAL)
1343 if (idx >= verdefs.size() || idx == VER_NDX_LOCAL) {
1344 error("corrupt input file: version definition index " + Twine(idx) +
1345 " for symbol " + name + " is out of bounds\n>>> defined in " +
1351 this->stringTable.data() +
1352 reinterpret_cast<const Elf_Verdef *>(verdefs[idx])->getAux()->vda_name;
1353 versionedNameBuffer.clear();
1354 name = (name + "@" + verName).toStringRef(versionedNameBuffer);
1355 symtab->addSymbol(SharedSymbol{*this, saver.save(name), sym.getBinding(),
1356 sym.st_other, sym.getType(), sym.st_value,
1357 sym.st_size, alignment, idx});
1361 static ELFKind getBitcodeELFKind(const Triple &t) {
1362 if (t.isLittleEndian())
1363 return t.isArch64Bit() ? ELF64LEKind : ELF32LEKind;
1364 return t.isArch64Bit() ? ELF64BEKind : ELF32BEKind;
1367 static uint8_t getBitcodeMachineKind(StringRef path, const Triple &t) {
1368 switch (t.getArch()) {
1369 case Triple::aarch64:
1371 case Triple::amdgcn:
1380 case Triple::mipsel:
1381 case Triple::mips64:
1382 case Triple::mips64el:
1384 case Triple::msp430:
1389 case Triple::ppc64le:
1391 case Triple::riscv32:
1392 case Triple::riscv64:
1395 return t.isOSIAMCU() ? EM_IAMCU : EM_386;
1396 case Triple::x86_64:
1399 error(path + ": could not infer e_machine from bitcode target triple " +
1405 BitcodeFile::BitcodeFile(MemoryBufferRef mb, StringRef archiveName,
1406 uint64_t offsetInArchive)
1407 : InputFile(BitcodeKind, mb) {
1408 this->archiveName = archiveName;
1410 std::string path = mb.getBufferIdentifier().str();
1411 if (config->thinLTOIndexOnly)
1412 path = replaceThinLTOSuffix(mb.getBufferIdentifier());
1414 // ThinLTO assumes that all MemoryBufferRefs given to it have a unique
1415 // name. If two archives define two members with the same name, this
1416 // causes a collision which result in only one of the objects being taken
1417 // into consideration at LTO time (which very likely causes undefined
1418 // symbols later in the link stage). So we append file offset to make
1420 StringRef name = archiveName.empty()
1422 : saver.save(archiveName + "(" + path + " at " +
1423 utostr(offsetInArchive) + ")");
1424 MemoryBufferRef mbref(mb.getBuffer(), name);
1426 obj = CHECK(lto::InputFile::create(mbref), this);
1428 Triple t(obj->getTargetTriple());
1429 ekind = getBitcodeELFKind(t);
1430 emachine = getBitcodeMachineKind(mb.getBufferIdentifier(), t);
1433 static uint8_t mapVisibility(GlobalValue::VisibilityTypes gvVisibility) {
1434 switch (gvVisibility) {
1435 case GlobalValue::DefaultVisibility:
1437 case GlobalValue::HiddenVisibility:
1439 case GlobalValue::ProtectedVisibility:
1440 return STV_PROTECTED;
1442 llvm_unreachable("unknown visibility");
1445 template <class ELFT>
1446 static Symbol *createBitcodeSymbol(const std::vector<bool> &keptComdats,
1447 const lto::InputFile::Symbol &objSym,
1449 StringRef name = saver.save(objSym.getName());
1450 uint8_t binding = objSym.isWeak() ? STB_WEAK : STB_GLOBAL;
1451 uint8_t type = objSym.isTLS() ? STT_TLS : STT_NOTYPE;
1452 uint8_t visibility = mapVisibility(objSym.getVisibility());
1453 bool canOmitFromDynSym = objSym.canBeOmittedFromSymbolTable();
1455 int c = objSym.getComdatIndex();
1456 if (objSym.isUndefined() || (c != -1 && !keptComdats[c])) {
1457 Undefined newSym(&f, name, binding, visibility, type);
1458 if (canOmitFromDynSym)
1459 newSym.exportDynamic = false;
1460 Symbol *ret = symtab->addSymbol(newSym);
1461 ret->referenced = true;
1465 if (objSym.isCommon())
1466 return symtab->addSymbol(
1467 CommonSymbol{&f, name, binding, visibility, STT_OBJECT,
1468 objSym.getCommonAlignment(), objSym.getCommonSize()});
1470 Defined newSym(&f, name, binding, visibility, type, 0, 0, nullptr);
1471 if (canOmitFromDynSym)
1472 newSym.exportDynamic = false;
1473 return symtab->addSymbol(newSym);
1476 template <class ELFT> void BitcodeFile::parse() {
1477 std::vector<bool> keptComdats;
1478 for (StringRef s : obj->getComdatTable())
1479 keptComdats.push_back(
1480 symtab->comdatGroups.try_emplace(CachedHashStringRef(s), this).second);
1482 for (const lto::InputFile::Symbol &objSym : obj->symbols())
1483 symbols.push_back(createBitcodeSymbol<ELFT>(keptComdats, objSym, *this));
1485 for (auto l : obj->getDependentLibraries())
1486 addDependentLibrary(l, this);
1489 void BinaryFile::parse() {
1490 ArrayRef<uint8_t> data = arrayRefFromStringRef(mb.getBuffer());
1491 auto *section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS,
1493 sections.push_back(section);
1495 // For each input file foo that is embedded to a result as a binary
1496 // blob, we define _binary_foo_{start,end,size} symbols, so that
1497 // user programs can access blobs by name. Non-alphanumeric
1498 // characters in a filename are replaced with underscore.
1499 std::string s = "_binary_" + mb.getBufferIdentifier().str();
1500 for (size_t i = 0; i < s.size(); ++i)
1504 symtab->addSymbol(Defined{nullptr, saver.save(s + "_start"), STB_GLOBAL,
1505 STV_DEFAULT, STT_OBJECT, 0, 0, section});
1506 symtab->addSymbol(Defined{nullptr, saver.save(s + "_end"), STB_GLOBAL,
1507 STV_DEFAULT, STT_OBJECT, data.size(), 0, section});
1508 symtab->addSymbol(Defined{nullptr, saver.save(s + "_size"), STB_GLOBAL,
1509 STV_DEFAULT, STT_OBJECT, data.size(), 0, nullptr});
1512 InputFile *createObjectFile(MemoryBufferRef mb, StringRef archiveName,
1513 uint64_t offsetInArchive) {
1515 return make<BitcodeFile>(mb, archiveName, offsetInArchive);
1517 switch (getELFKind(mb, archiveName)) {
1519 return make<ObjFile<ELF32LE>>(mb, archiveName);
1521 return make<ObjFile<ELF32BE>>(mb, archiveName);
1523 return make<ObjFile<ELF64LE>>(mb, archiveName);
1525 return make<ObjFile<ELF64BE>>(mb, archiveName);
1527 llvm_unreachable("getELFKind");
1531 void LazyObjFile::fetch() {
1532 if (mb.getBuffer().empty())
1535 InputFile *file = createObjectFile(mb, archiveName, offsetInArchive);
1536 file->groupId = groupId;
1540 // Copy symbol vector so that the new InputFile doesn't have to
1541 // insert the same defined symbols to the symbol table again.
1542 file->symbols = std::move(symbols);
1547 template <class ELFT> void LazyObjFile::parse() {
1548 using Elf_Sym = typename ELFT::Sym;
1550 // A lazy object file wraps either a bitcode file or an ELF file.
1551 if (isBitcode(this->mb)) {
1552 std::unique_ptr<lto::InputFile> obj =
1553 CHECK(lto::InputFile::create(this->mb), this);
1554 for (const lto::InputFile::Symbol &sym : obj->symbols()) {
1555 if (sym.isUndefined())
1557 symtab->addSymbol(LazyObject{*this, saver.save(sym.getName())});
1562 if (getELFKind(this->mb, archiveName) != config->ekind) {
1563 error("incompatible file: " + this->mb.getBufferIdentifier());
1567 // Find a symbol table.
1568 ELFFile<ELFT> obj = check(ELFFile<ELFT>::create(mb.getBuffer()));
1569 ArrayRef<typename ELFT::Shdr> sections = CHECK(obj.sections(), this);
1571 for (const typename ELFT::Shdr &sec : sections) {
1572 if (sec.sh_type != SHT_SYMTAB)
1575 // A symbol table is found.
1576 ArrayRef<Elf_Sym> eSyms = CHECK(obj.symbols(&sec), this);
1577 uint32_t firstGlobal = sec.sh_info;
1578 StringRef strtab = CHECK(obj.getStringTableForSymtab(sec, sections), this);
1579 this->symbols.resize(eSyms.size());
1581 // Get existing symbols or insert placeholder symbols.
1582 for (size_t i = firstGlobal, end = eSyms.size(); i != end; ++i)
1583 if (eSyms[i].st_shndx != SHN_UNDEF)
1584 this->symbols[i] = symtab->insert(CHECK(eSyms[i].getName(strtab), this));
1586 // Replace existing symbols with LazyObject symbols.
1588 // resolve() may trigger this->fetch() if an existing symbol is an
1589 // undefined symbol. If that happens, this LazyObjFile has served
1590 // its purpose, and we can exit from the loop early.
1591 for (Symbol *sym : this->symbols) {
1594 sym->resolve(LazyObject{*this, sym->getName()});
1596 // MemoryBuffer is emptied if this file is instantiated as ObjFile.
1597 if (mb.getBuffer().empty())
1604 std::string replaceThinLTOSuffix(StringRef path) {
1605 StringRef suffix = config->thinLTOObjectSuffixReplace.first;
1606 StringRef repl = config->thinLTOObjectSuffixReplace.second;
1608 if (path.consume_back(suffix))
1609 return (path + repl).str();
1613 template void BitcodeFile::parse<ELF32LE>();
1614 template void BitcodeFile::parse<ELF32BE>();
1615 template void BitcodeFile::parse<ELF64LE>();
1616 template void BitcodeFile::parse<ELF64BE>();
1618 template void LazyObjFile::parse<ELF32LE>();
1619 template void LazyObjFile::parse<ELF32BE>();
1620 template void LazyObjFile::parse<ELF64LE>();
1621 template void LazyObjFile::parse<ELF64BE>();
1623 template class ObjFile<ELF32LE>;
1624 template class ObjFile<ELF32BE>;
1625 template class ObjFile<ELF64LE>;
1626 template class ObjFile<ELF64BE>;
1628 template void SharedFile::parse<ELF32LE>();
1629 template void SharedFile::parse<ELF32BE>();
1630 template void SharedFile::parse<ELF64LE>();
1631 template void SharedFile::parse<ELF64BE>();