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/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/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 using namespace lld::elf;
43 bool InputFile::isInGroup;
44 uint32_t InputFile::nextGroupId;
45 std::vector<BinaryFile *> elf::binaryFiles;
46 std::vector<BitcodeFile *> elf::bitcodeFiles;
47 std::vector<LazyObjFile *> elf::lazyObjFiles;
48 std::vector<InputFile *> elf::objectFiles;
49 std::vector<SharedFile *> elf::sharedFiles;
51 std::unique_ptr<TarWriter> elf::tar;
53 static ELFKind getELFKind(MemoryBufferRef mb, StringRef archiveName) {
56 std::tie(size, endian) = getElfArchType(mb.getBuffer());
58 auto report = [&](StringRef msg) {
59 StringRef filename = mb.getBufferIdentifier();
60 if (archiveName.empty())
61 fatal(filename + ": " + msg);
63 fatal(archiveName + "(" + filename + "): " + msg);
66 if (!mb.getBuffer().startswith(ElfMagic))
67 report("not an ELF file");
68 if (endian != ELFDATA2LSB && endian != ELFDATA2MSB)
69 report("corrupted ELF file: invalid data encoding");
70 if (size != ELFCLASS32 && size != ELFCLASS64)
71 report("corrupted ELF file: invalid file class");
73 size_t bufSize = mb.getBuffer().size();
74 if ((size == ELFCLASS32 && bufSize < sizeof(Elf32_Ehdr)) ||
75 (size == ELFCLASS64 && bufSize < sizeof(Elf64_Ehdr)))
76 report("corrupted ELF file: file is too short");
78 if (size == ELFCLASS32)
79 return (endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind;
80 return (endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind;
83 InputFile::InputFile(Kind k, MemoryBufferRef m)
84 : mb(m), groupId(nextGroupId), fileKind(k) {
85 // All files within the same --{start,end}-group get the same group ID.
86 // Otherwise, a new file will get a new group ID.
91 Optional<MemoryBufferRef> elf::readFile(StringRef path) {
92 // The --chroot option changes our virtual root directory.
93 // This is useful when you are dealing with files created by --reproduce.
94 if (!config->chroot.empty() && path.startswith("/"))
95 path = saver.save(config->chroot + path);
99 auto mbOrErr = MemoryBuffer::getFile(path, -1, false);
100 if (auto ec = mbOrErr.getError()) {
101 error("cannot open " + path + ": " + ec.message());
105 std::unique_ptr<MemoryBuffer> &mb = *mbOrErr;
106 MemoryBufferRef mbref = mb->getMemBufferRef();
107 make<std::unique_ptr<MemoryBuffer>>(std::move(mb)); // take MB ownership
110 tar->append(relativeToRoot(path), mbref.getBuffer());
114 // All input object files must be for the same architecture
115 // (e.g. it does not make sense to link x86 object files with
116 // MIPS object files.) This function checks for that error.
117 static bool isCompatible(InputFile *file) {
118 if (!file->isElf() && !isa<BitcodeFile>(file))
121 if (file->ekind == config->ekind && file->emachine == config->emachine) {
122 if (config->emachine != EM_MIPS)
124 if (isMipsN32Abi(file) == config->mipsN32Abi)
128 if (!config->emulation.empty()) {
129 error(toString(file) + " is incompatible with " + config->emulation);
132 if (!objectFiles.empty())
133 existing = objectFiles[0];
134 else if (!sharedFiles.empty())
135 existing = sharedFiles[0];
137 existing = bitcodeFiles[0];
139 error(toString(file) + " is incompatible with " + toString(existing));
145 template <class ELFT> static void doParseFile(InputFile *file) {
146 if (!isCompatible(file))
150 if (auto *f = dyn_cast<BinaryFile>(file)) {
151 binaryFiles.push_back(f);
157 if (auto *f = dyn_cast<ArchiveFile>(file)) {
163 if (auto *f = dyn_cast<LazyObjFile>(file)) {
164 lazyObjFiles.push_back(f);
170 message(toString(file));
173 if (auto *f = dyn_cast<SharedFile>(file)) {
179 if (auto *f = dyn_cast<BitcodeFile>(file)) {
180 bitcodeFiles.push_back(f);
185 // Regular object file
186 objectFiles.push_back(file);
187 cast<ObjFile<ELFT>>(file)->parse();
190 // Add symbols in File to the symbol table.
191 void elf::parseFile(InputFile *file) {
192 switch (config->ekind) {
194 doParseFile<ELF32LE>(file);
197 doParseFile<ELF32BE>(file);
200 doParseFile<ELF64LE>(file);
203 doParseFile<ELF64BE>(file);
206 llvm_unreachable("unknown ELFT");
210 // Concatenates arguments to construct a string representing an error location.
211 static std::string createFileLineMsg(StringRef path, unsigned line) {
212 std::string filename = path::filename(path);
213 std::string lineno = ":" + std::to_string(line);
214 if (filename == path)
215 return filename + lineno;
216 return filename + lineno + " (" + path.str() + lineno + ")";
219 template <class ELFT>
220 static std::string getSrcMsgAux(ObjFile<ELFT> &file, const Symbol &sym,
221 InputSectionBase &sec, uint64_t offset) {
222 // In DWARF, functions and variables are stored to different places.
223 // First, lookup a function for a given offset.
224 if (Optional<DILineInfo> info = file.getDILineInfo(&sec, offset))
225 return createFileLineMsg(info->FileName, info->Line);
227 // If it failed, lookup again as a variable.
228 if (Optional<std::pair<std::string, unsigned>> fileLine =
229 file.getVariableLoc(sym.getName()))
230 return createFileLineMsg(fileLine->first, fileLine->second);
232 // File.sourceFile contains STT_FILE symbol, and that is a last resort.
233 return file.sourceFile;
236 std::string InputFile::getSrcMsg(const Symbol &sym, InputSectionBase &sec,
238 if (kind() != ObjKind)
240 switch (config->ekind) {
242 llvm_unreachable("Invalid kind");
244 return getSrcMsgAux(cast<ObjFile<ELF32LE>>(*this), sym, sec, offset);
246 return getSrcMsgAux(cast<ObjFile<ELF32BE>>(*this), sym, sec, offset);
248 return getSrcMsgAux(cast<ObjFile<ELF64LE>>(*this), sym, sec, offset);
250 return getSrcMsgAux(cast<ObjFile<ELF64BE>>(*this), sym, sec, offset);
254 template <class ELFT> void ObjFile<ELFT>::initializeDwarf() {
255 dwarf = llvm::make_unique<DWARFContext>(make_unique<LLDDwarfObj<ELFT>>(this));
256 for (std::unique_ptr<DWARFUnit> &cu : dwarf->compile_units()) {
257 auto report = [](Error err) {
258 handleAllErrors(std::move(err),
259 [](ErrorInfoBase &info) { warn(info.message()); });
261 Expected<const DWARFDebugLine::LineTable *> expectedLT =
262 dwarf->getLineTableForUnit(cu.get(), report);
263 const DWARFDebugLine::LineTable *lt = nullptr;
267 report(expectedLT.takeError());
270 lineTables.push_back(lt);
272 // Loop over variable records and insert them to variableLoc.
273 for (const auto &entry : cu->dies()) {
274 DWARFDie die(cu.get(), &entry);
275 // Skip all tags that are not variables.
276 if (die.getTag() != dwarf::DW_TAG_variable)
279 // Skip if a local variable because we don't need them for generating
280 // error messages. In general, only non-local symbols can fail to be
282 if (!dwarf::toUnsigned(die.find(dwarf::DW_AT_external), 0))
285 // Get the source filename index for the variable.
286 unsigned file = dwarf::toUnsigned(die.find(dwarf::DW_AT_decl_file), 0);
287 if (!lt->hasFileAtIndex(file))
290 // Get the line number on which the variable is declared.
291 unsigned line = dwarf::toUnsigned(die.find(dwarf::DW_AT_decl_line), 0);
293 // Here we want to take the variable name to add it into variableLoc.
294 // Variable can have regular and linkage name associated. At first, we try
295 // to get linkage name as it can be different, for example when we have
296 // two variables in different namespaces of the same object. Use common
297 // name otherwise, but handle the case when it also absent in case if the
298 // input object file lacks some debug info.
300 dwarf::toString(die.find(dwarf::DW_AT_linkage_name),
301 dwarf::toString(die.find(dwarf::DW_AT_name), ""));
303 variableLoc.insert({name, {lt, file, line}});
308 // Returns the pair of file name and line number describing location of data
309 // object (variable, array, etc) definition.
310 template <class ELFT>
311 Optional<std::pair<std::string, unsigned>>
312 ObjFile<ELFT>::getVariableLoc(StringRef name) {
313 llvm::call_once(initDwarfLine, [this]() { initializeDwarf(); });
315 // Return if we have no debug information about data object.
316 auto it = variableLoc.find(name);
317 if (it == variableLoc.end())
320 // Take file name string from line table.
321 std::string fileName;
322 if (!it->second.lt->getFileNameByIndex(
324 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, fileName))
327 return std::make_pair(fileName, it->second.line);
330 // Returns source line information for a given offset
331 // using DWARF debug info.
332 template <class ELFT>
333 Optional<DILineInfo> ObjFile<ELFT>::getDILineInfo(InputSectionBase *s,
335 llvm::call_once(initDwarfLine, [this]() { initializeDwarf(); });
337 // Detect SectionIndex for specified section.
338 uint64_t sectionIndex = object::SectionedAddress::UndefSection;
339 ArrayRef<InputSectionBase *> sections = s->file->getSections();
340 for (uint64_t curIndex = 0; curIndex < sections.size(); ++curIndex) {
341 if (s == sections[curIndex]) {
342 sectionIndex = curIndex;
347 // Use fake address calcuated by adding section file offset and offset in
348 // section. See comments for ObjectInfo class.
350 for (const llvm::DWARFDebugLine::LineTable *lt : lineTables) {
351 if (lt->getFileLineInfoForAddress(
352 {s->getOffsetInFile() + offset, sectionIndex}, nullptr,
353 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, info))
359 // Returns "<internal>", "foo.a(bar.o)" or "baz.o".
360 std::string lld::toString(const InputFile *f) {
364 if (f->toStringCache.empty()) {
365 if (f->archiveName.empty())
366 f->toStringCache = f->getName();
368 f->toStringCache = (f->archiveName + "(" + f->getName() + ")").str();
370 return f->toStringCache;
373 ELFFileBase::ELFFileBase(Kind k, MemoryBufferRef mb) : InputFile(k, mb) {
374 ekind = getELFKind(mb, "");
390 llvm_unreachable("getELFKind");
394 template <typename Elf_Shdr>
395 static const Elf_Shdr *findSection(ArrayRef<Elf_Shdr> sections, uint32_t type) {
396 for (const Elf_Shdr &sec : sections)
397 if (sec.sh_type == type)
402 template <class ELFT> void ELFFileBase::init() {
403 using Elf_Shdr = typename ELFT::Shdr;
404 using Elf_Sym = typename ELFT::Sym;
406 // Initialize trivial attributes.
407 const ELFFile<ELFT> &obj = getObj<ELFT>();
408 emachine = obj.getHeader()->e_machine;
409 osabi = obj.getHeader()->e_ident[llvm::ELF::EI_OSABI];
410 abiVersion = obj.getHeader()->e_ident[llvm::ELF::EI_ABIVERSION];
412 ArrayRef<Elf_Shdr> sections = CHECK(obj.sections(), this);
414 // Find a symbol table.
416 (identify_magic(mb.getBuffer()) == file_magic::elf_shared_object);
417 const Elf_Shdr *symtabSec =
418 findSection(sections, isDSO ? SHT_DYNSYM : SHT_SYMTAB);
423 // Initialize members corresponding to a symbol table.
424 firstGlobal = symtabSec->sh_info;
426 ArrayRef<Elf_Sym> eSyms = CHECK(obj.symbols(symtabSec), this);
427 if (firstGlobal == 0 || firstGlobal > eSyms.size())
428 fatal(toString(this) + ": invalid sh_info in symbol table");
430 elfSyms = reinterpret_cast<const void *>(eSyms.data());
431 numELFSyms = eSyms.size();
432 stringTable = CHECK(obj.getStringTableForSymtab(*symtabSec, sections), this);
435 template <class ELFT>
436 uint32_t ObjFile<ELFT>::getSectionIndex(const Elf_Sym &sym) const {
438 this->getObj().getSectionIndex(&sym, getELFSyms<ELFT>(), shndxTable),
442 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() {
443 if (this->symbols.empty())
445 return makeArrayRef(this->symbols).slice(1, this->firstGlobal - 1);
448 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getGlobalSymbols() {
449 return makeArrayRef(this->symbols).slice(this->firstGlobal);
452 template <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) {
453 // Read a section table. justSymbols is usually false.
454 if (this->justSymbols)
455 initializeJustSymbols();
457 initializeSections(ignoreComdats);
459 // Read a symbol table.
463 // Sections with SHT_GROUP and comdat bits define comdat section groups.
464 // They are identified and deduplicated by group name. This function
465 // returns a group name.
466 template <class ELFT>
467 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> sections,
468 const Elf_Shdr &sec) {
469 typename ELFT::SymRange symbols = this->getELFSyms<ELFT>();
470 if (sec.sh_info >= symbols.size())
471 fatal(toString(this) + ": invalid symbol index");
472 const typename ELFT::Sym &sym = symbols[sec.sh_info];
473 StringRef signature = CHECK(sym.getName(this->stringTable), this);
475 // As a special case, if a symbol is a section symbol and has no name,
476 // we use a section name as a signature.
478 // Such SHT_GROUP sections are invalid from the perspective of the ELF
479 // standard, but GNU gold 1.14 (the newest version as of July 2017) or
480 // older produce such sections as outputs for the -r option, so we need
481 // a bug-compatibility.
482 if (signature.empty() && sym.getType() == STT_SECTION)
483 return getSectionName(sec);
487 template <class ELFT> bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &sec) {
488 // On a regular link we don't merge sections if -O0 (default is -O1). This
489 // sometimes makes the linker significantly faster, although the output will
492 // Doing the same for -r would create a problem as it would combine sections
493 // with different sh_entsize. One option would be to just copy every SHF_MERGE
494 // section as is to the output. While this would produce a valid ELF file with
495 // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when
496 // they see two .debug_str. We could have separate logic for combining
497 // SHF_MERGE sections based both on their name and sh_entsize, but that seems
498 // to be more trouble than it is worth. Instead, we just use the regular (-O1)
500 if (config->optimize == 0 && !config->relocatable)
503 // A mergeable section with size 0 is useless because they don't have
504 // any data to merge. A mergeable string section with size 0 can be
505 // argued as invalid because it doesn't end with a null character.
506 // We'll avoid a mess by handling them as if they were non-mergeable.
507 if (sec.sh_size == 0)
510 // Check for sh_entsize. The ELF spec is not clear about the zero
511 // sh_entsize. It says that "the member [sh_entsize] contains 0 if
512 // the section does not hold a table of fixed-size entries". We know
513 // that Rust 1.13 produces a string mergeable section with a zero
514 // sh_entsize. Here we just accept it rather than being picky about it.
515 uint64_t entSize = sec.sh_entsize;
518 if (sec.sh_size % entSize)
519 fatal(toString(this) +
520 ": SHF_MERGE section size must be a multiple of sh_entsize");
522 uint64_t flags = sec.sh_flags;
523 if (!(flags & SHF_MERGE))
525 if (flags & SHF_WRITE)
526 fatal(toString(this) + ": writable SHF_MERGE section is not supported");
531 // This is for --just-symbols.
533 // --just-symbols is a very minor feature that allows you to link your
534 // output against other existing program, so that if you load both your
535 // program and the other program into memory, your output can refer the
536 // other program's symbols.
538 // When the option is given, we link "just symbols". The section table is
539 // initialized with null pointers.
540 template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() {
541 ArrayRef<Elf_Shdr> sections = CHECK(this->getObj().sections(), this);
542 this->sections.resize(sections.size());
545 // An ELF object file may contain a `.deplibs` section. If it exists, the
546 // section contains a list of library specifiers such as `m` for libm. This
547 // function resolves a given name by finding the first matching library checking
548 // the various ways that a library can be specified to LLD. This ELF extension
549 // is a form of autolinking and is called `dependent libraries`. It is currently
550 // unique to LLVM and lld.
551 static void addDependentLibrary(StringRef specifier, const InputFile *f) {
552 if (!config->dependentLibraries)
554 if (fs::exists(specifier))
555 driver->addFile(specifier, /*withLOption=*/false);
556 else if (Optional<std::string> s = findFromSearchPaths(specifier))
557 driver->addFile(*s, /*withLOption=*/true);
558 else if (Optional<std::string> s = searchLibraryBaseName(specifier))
559 driver->addFile(*s, /*withLOption=*/true);
562 ": unable to find library from dependent library specifier: " +
566 template <class ELFT>
567 void ObjFile<ELFT>::initializeSections(bool ignoreComdats) {
568 const ELFFile<ELFT> &obj = this->getObj();
570 ArrayRef<Elf_Shdr> objSections = CHECK(obj.sections(), this);
571 uint64_t size = objSections.size();
572 this->sections.resize(size);
573 this->sectionStringTable =
574 CHECK(obj.getSectionStringTable(objSections), this);
576 for (size_t i = 0, e = objSections.size(); i < e; i++) {
577 if (this->sections[i] == &InputSection::discarded)
579 const Elf_Shdr &sec = objSections[i];
581 if (sec.sh_type == ELF::SHT_LLVM_CALL_GRAPH_PROFILE)
583 check(obj.template getSectionContentsAsArray<Elf_CGProfile>(&sec));
585 // SHF_EXCLUDE'ed sections are discarded by the linker. However,
586 // if -r is given, we'll let the final link discard such sections.
587 // This is compatible with GNU.
588 if ((sec.sh_flags & SHF_EXCLUDE) && !config->relocatable) {
589 if (sec.sh_type == SHT_LLVM_ADDRSIG) {
590 // We ignore the address-significance table if we know that the object
591 // file was created by objcopy or ld -r. This is because these tools
592 // will reorder the symbols in the symbol table, invalidating the data
593 // in the address-significance table, which refers to symbols by index.
594 if (sec.sh_link != 0)
595 this->addrsigSec = &sec;
596 else if (config->icf == ICFLevel::Safe)
597 warn(toString(this) + ": --icf=safe is incompatible with object "
598 "files created using objcopy or ld -r");
600 this->sections[i] = &InputSection::discarded;
604 switch (sec.sh_type) {
606 // De-duplicate section groups by their signatures.
607 StringRef signature = getShtGroupSignature(objSections, sec);
608 this->sections[i] = &InputSection::discarded;
611 ArrayRef<Elf_Word> entries =
612 CHECK(obj.template getSectionContentsAsArray<Elf_Word>(&sec), this);
614 fatal(toString(this) + ": empty SHT_GROUP");
616 // The first word of a SHT_GROUP section contains flags. Currently,
617 // the standard defines only "GRP_COMDAT" flag for the COMDAT group.
618 // An group with the empty flag doesn't define anything; such sections
623 if (entries[0] != GRP_COMDAT)
624 fatal(toString(this) + ": unsupported SHT_GROUP format");
628 symtab->comdatGroups.try_emplace(CachedHashStringRef(signature), this)
631 if (config->relocatable)
632 this->sections[i] = createInputSection(sec);
636 // Otherwise, discard group members.
637 for (uint32_t secIndex : entries.slice(1)) {
638 if (secIndex >= size)
639 fatal(toString(this) +
640 ": invalid section index in group: " + Twine(secIndex));
641 this->sections[secIndex] = &InputSection::discarded;
645 case SHT_SYMTAB_SHNDX:
646 shndxTable = CHECK(obj.getSHNDXTable(sec, objSections), this);
653 this->sections[i] = createInputSection(sec);
656 // .ARM.exidx sections have a reverse dependency on the InputSection they
657 // have a SHF_LINK_ORDER dependency, this is identified by the sh_link.
658 if (sec.sh_flags & SHF_LINK_ORDER) {
659 InputSectionBase *linkSec = nullptr;
660 if (sec.sh_link < this->sections.size())
661 linkSec = this->sections[sec.sh_link];
663 fatal(toString(this) +
664 ": invalid sh_link index: " + Twine(sec.sh_link));
666 InputSection *isec = cast<InputSection>(this->sections[i]);
667 linkSec->dependentSections.push_back(isec);
668 if (!isa<InputSection>(linkSec))
669 error("a section " + isec->name +
670 " with SHF_LINK_ORDER should not refer a non-regular "
677 // For ARM only, to set the EF_ARM_ABI_FLOAT_SOFT or EF_ARM_ABI_FLOAT_HARD
678 // flag in the ELF Header we need to look at Tag_ABI_VFP_args to find out how
679 // the input objects have been compiled.
680 static void updateARMVFPArgs(const ARMAttributeParser &attributes,
681 const InputFile *f) {
682 if (!attributes.hasAttribute(ARMBuildAttrs::ABI_VFP_args))
683 // If an ABI tag isn't present then it is implicitly given the value of 0
684 // which maps to ARMBuildAttrs::BaseAAPCS. However many assembler files,
685 // including some in glibc that don't use FP args (and should have value 3)
686 // don't have the attribute so we do not consider an implicit value of 0
690 unsigned vfpArgs = attributes.getAttributeValue(ARMBuildAttrs::ABI_VFP_args);
693 case ARMBuildAttrs::BaseAAPCS:
694 arg = ARMVFPArgKind::Base;
696 case ARMBuildAttrs::HardFPAAPCS:
697 arg = ARMVFPArgKind::VFP;
699 case ARMBuildAttrs::ToolChainFPPCS:
700 // Tool chain specific convention that conforms to neither AAPCS variant.
701 arg = ARMVFPArgKind::ToolChain;
703 case ARMBuildAttrs::CompatibleFPAAPCS:
704 // Object compatible with all conventions.
707 error(toString(f) + ": unknown Tag_ABI_VFP_args value: " + Twine(vfpArgs));
710 // Follow ld.bfd and error if there is a mix of calling conventions.
711 if (config->armVFPArgs != arg && config->armVFPArgs != ARMVFPArgKind::Default)
712 error(toString(f) + ": incompatible Tag_ABI_VFP_args");
714 config->armVFPArgs = arg;
717 // The ARM support in lld makes some use of instructions that are not available
718 // on all ARM architectures. Namely:
719 // - Use of BLX instruction for interworking between ARM and Thumb state.
720 // - Use of the extended Thumb branch encoding in relocation.
721 // - Use of the MOVT/MOVW instructions in Thumb Thunks.
722 // The ARM Attributes section contains information about the architecture chosen
723 // at compile time. We follow the convention that if at least one input object
724 // is compiled with an architecture that supports these features then lld is
725 // permitted to use them.
726 static void updateSupportedARMFeatures(const ARMAttributeParser &attributes) {
727 if (!attributes.hasAttribute(ARMBuildAttrs::CPU_arch))
729 auto arch = attributes.getAttributeValue(ARMBuildAttrs::CPU_arch);
731 case ARMBuildAttrs::Pre_v4:
732 case ARMBuildAttrs::v4:
733 case ARMBuildAttrs::v4T:
734 // Architectures prior to v5 do not support BLX instruction
736 case ARMBuildAttrs::v5T:
737 case ARMBuildAttrs::v5TE:
738 case ARMBuildAttrs::v5TEJ:
739 case ARMBuildAttrs::v6:
740 case ARMBuildAttrs::v6KZ:
741 case ARMBuildAttrs::v6K:
742 config->armHasBlx = true;
743 // Architectures used in pre-Cortex processors do not support
744 // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception
745 // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do.
748 // All other Architectures have BLX and extended branch encoding
749 config->armHasBlx = true;
750 config->armJ1J2BranchEncoding = true;
751 if (arch != ARMBuildAttrs::v6_M && arch != ARMBuildAttrs::v6S_M)
752 // All Architectures used in Cortex processors with the exception
753 // of v6-M and v6S-M have the MOVT and MOVW instructions.
754 config->armHasMovtMovw = true;
759 // If a source file is compiled with x86 hardware-assisted call flow control
760 // enabled, the generated object file contains feature flags indicating that
761 // fact. This function reads the feature flags and returns it.
763 // Essentially we want to read a single 32-bit value in this function, but this
764 // function is rather complicated because the value is buried deep inside a
765 // .note.gnu.property section.
767 // The section consists of one or more NOTE records. Each NOTE record consists
768 // of zero or more type-length-value fields. We want to find a field of a
769 // certain type. It seems a bit too much to just store a 32-bit value, perhaps
770 // the ABI is unnecessarily complicated.
771 template <class ELFT>
772 static uint32_t readAndFeatures(ObjFile<ELFT> *obj, ArrayRef<uint8_t> data) {
773 using Elf_Nhdr = typename ELFT::Nhdr;
774 using Elf_Note = typename ELFT::Note;
776 uint32_t featuresSet = 0;
777 while (!data.empty()) {
778 // Read one NOTE record.
779 if (data.size() < sizeof(Elf_Nhdr))
780 fatal(toString(obj) + ": .note.gnu.property: section too short");
782 auto *nhdr = reinterpret_cast<const Elf_Nhdr *>(data.data());
783 if (data.size() < nhdr->getSize())
784 fatal(toString(obj) + ": .note.gnu.property: section too short");
786 Elf_Note note(*nhdr);
787 if (nhdr->n_type != NT_GNU_PROPERTY_TYPE_0 || note.getName() != "GNU") {
788 data = data.slice(nhdr->getSize());
792 uint32_t featureAndType = config->emachine == EM_AARCH64
793 ? GNU_PROPERTY_AARCH64_FEATURE_1_AND
794 : GNU_PROPERTY_X86_FEATURE_1_AND;
796 // Read a body of a NOTE record, which consists of type-length-value fields.
797 ArrayRef<uint8_t> desc = note.getDesc();
798 while (!desc.empty()) {
800 fatal(toString(obj) + ": .note.gnu.property: section too short");
802 uint32_t type = read32le(desc.data());
803 uint32_t size = read32le(desc.data() + 4);
805 if (type == featureAndType) {
806 // We found a FEATURE_1_AND field. There may be more than one of these
807 // in a .note.gnu.propery section, for a relocatable object we
808 // accumulate the bits set.
809 featuresSet |= read32le(desc.data() + 8);
812 // On 64-bit, a payload may be followed by a 4-byte padding to make its
813 // size a multiple of 8.
815 size = alignTo(size, 8);
817 desc = desc.slice(size + 8); // +8 for Type and Size
820 // Go to next NOTE record to look for more FEATURE_1_AND descriptions.
821 data = data.slice(nhdr->getSize());
827 template <class ELFT>
828 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &sec) {
829 uint32_t idx = sec.sh_info;
830 if (idx >= this->sections.size())
831 fatal(toString(this) + ": invalid relocated section index: " + Twine(idx));
832 InputSectionBase *target = this->sections[idx];
834 // Strictly speaking, a relocation section must be included in the
835 // group of the section it relocates. However, LLVM 3.3 and earlier
836 // would fail to do so, so we gracefully handle that case.
837 if (target == &InputSection::discarded)
841 fatal(toString(this) + ": unsupported relocation reference");
845 // Create a regular InputSection class that has the same contents
846 // as a given section.
847 static InputSection *toRegularSection(MergeInputSection *sec) {
848 return make<InputSection>(sec->file, sec->flags, sec->type, sec->alignment,
849 sec->data(), sec->name);
852 template <class ELFT>
853 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &sec) {
854 StringRef name = getSectionName(sec);
856 switch (sec.sh_type) {
857 case SHT_ARM_ATTRIBUTES: {
858 if (config->emachine != EM_ARM)
860 ARMAttributeParser attributes;
861 ArrayRef<uint8_t> contents = check(this->getObj().getSectionContents(&sec));
862 attributes.Parse(contents, /*isLittle*/ config->ekind == ELF32LEKind);
863 updateSupportedARMFeatures(attributes);
864 updateARMVFPArgs(attributes, this);
866 // FIXME: Retain the first attribute section we see. The eglibc ARM
867 // dynamic loaders require the presence of an attribute section for dlopen
868 // to work. In a full implementation we would merge all attribute sections.
869 if (in.armAttributes == nullptr) {
870 in.armAttributes = make<InputSection>(*this, sec, name);
871 return in.armAttributes;
873 return &InputSection::discarded;
875 case SHT_LLVM_DEPENDENT_LIBRARIES: {
876 if (config->relocatable)
878 ArrayRef<char> data =
879 CHECK(this->getObj().template getSectionContentsAsArray<char>(&sec), this);
880 if (!data.empty() && data.back() != '\0') {
881 error(toString(this) +
882 ": corrupted dependent libraries section (unterminated string): " +
884 return &InputSection::discarded;
886 for (const char *d = data.begin(), *e = data.end(); d < e;) {
888 addDependentLibrary(s, this);
891 return &InputSection::discarded;
895 // Find a relocation target section and associate this section with that.
896 // Target may have been discarded if it is in a different section group
897 // and the group is discarded, even though it's a violation of the
898 // spec. We handle that situation gracefully by discarding dangling
899 // relocation sections.
900 InputSectionBase *target = getRelocTarget(sec);
904 // This section contains relocation information.
905 // If -r is given, we do not interpret or apply relocation
906 // but just copy relocation sections to output.
907 if (config->relocatable) {
908 InputSection *relocSec = make<InputSection>(*this, sec, name);
909 // We want to add a dependency to target, similar like we do for
910 // -emit-relocs below. This is useful for the case when linker script
911 // contains the "/DISCARD/". It is perhaps uncommon to use a script with
912 // -r, but we faced it in the Linux kernel and have to handle such case
914 target->dependentSections.push_back(relocSec);
918 if (target->firstRelocation)
919 fatal(toString(this) +
920 ": multiple relocation sections to one section are not supported");
922 // ELF spec allows mergeable sections with relocations, but they are
923 // rare, and it is in practice hard to merge such sections by contents,
924 // because applying relocations at end of linking changes section
925 // contents. So, we simply handle such sections as non-mergeable ones.
926 // Degrading like this is acceptable because section merging is optional.
927 if (auto *ms = dyn_cast<MergeInputSection>(target)) {
928 target = toRegularSection(ms);
929 this->sections[sec.sh_info] = target;
932 if (sec.sh_type == SHT_RELA) {
933 ArrayRef<Elf_Rela> rels = CHECK(getObj().relas(&sec), this);
934 target->firstRelocation = rels.begin();
935 target->numRelocations = rels.size();
936 target->areRelocsRela = true;
938 ArrayRef<Elf_Rel> rels = CHECK(getObj().rels(&sec), this);
939 target->firstRelocation = rels.begin();
940 target->numRelocations = rels.size();
941 target->areRelocsRela = false;
943 assert(isUInt<31>(target->numRelocations));
945 // Relocation sections processed by the linker are usually removed
946 // from the output, so returning `nullptr` for the normal case.
947 // However, if -emit-relocs is given, we need to leave them in the output.
948 // (Some post link analysis tools need this information.)
949 if (config->emitRelocs) {
950 InputSection *relocSec = make<InputSection>(*this, sec, name);
951 // We will not emit relocation section if target was discarded.
952 target->dependentSections.push_back(relocSec);
959 // The GNU linker uses .note.GNU-stack section as a marker indicating
960 // that the code in the object file does not expect that the stack is
961 // executable (in terms of NX bit). If all input files have the marker,
962 // the GNU linker adds a PT_GNU_STACK segment to tells the loader to
963 // make the stack non-executable. Most object files have this section as
966 // But making the stack non-executable is a norm today for security
967 // reasons. Failure to do so may result in a serious security issue.
968 // Therefore, we make LLD always add PT_GNU_STACK unless it is
969 // explicitly told to do otherwise (by -z execstack). Because the stack
970 // executable-ness is controlled solely by command line options,
971 // .note.GNU-stack sections are simply ignored.
972 if (name == ".note.GNU-stack")
973 return &InputSection::discarded;
975 // Object files that use processor features such as Intel Control-Flow
976 // Enforcement (CET) or AArch64 Branch Target Identification BTI, use a
977 // .note.gnu.property section containing a bitfield of feature bits like the
978 // GNU_PROPERTY_X86_FEATURE_1_IBT flag. Read a bitmap containing the flag.
980 // Since we merge bitmaps from multiple object files to create a new
981 // .note.gnu.property containing a single AND'ed bitmap, we discard an input
982 // file's .note.gnu.property section.
983 if (name == ".note.gnu.property") {
984 ArrayRef<uint8_t> contents = check(this->getObj().getSectionContents(&sec));
985 this->andFeatures = readAndFeatures(this, contents);
986 return &InputSection::discarded;
989 // Split stacks is a feature to support a discontiguous stack,
990 // commonly used in the programming language Go. For the details,
991 // see https://gcc.gnu.org/wiki/SplitStacks. An object file compiled
992 // for split stack will include a .note.GNU-split-stack section.
993 if (name == ".note.GNU-split-stack") {
994 if (config->relocatable) {
995 error("cannot mix split-stack and non-split-stack in a relocatable link");
996 return &InputSection::discarded;
998 this->splitStack = true;
999 return &InputSection::discarded;
1002 // An object file cmpiled for split stack, but where some of the
1003 // functions were compiled with the no_split_stack_attribute will
1004 // include a .note.GNU-no-split-stack section.
1005 if (name == ".note.GNU-no-split-stack") {
1006 this->someNoSplitStack = true;
1007 return &InputSection::discarded;
1010 // The linkonce feature is a sort of proto-comdat. Some glibc i386 object
1011 // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce
1012 // sections. Drop those sections to avoid duplicate symbol errors.
1013 // FIXME: This is glibc PR20543, we should remove this hack once that has been
1014 // fixed for a while.
1015 if (name == ".gnu.linkonce.t.__x86.get_pc_thunk.bx" ||
1016 name == ".gnu.linkonce.t.__i686.get_pc_thunk.bx")
1017 return &InputSection::discarded;
1019 // If we are creating a new .build-id section, strip existing .build-id
1020 // sections so that the output won't have more than one .build-id.
1021 // This is not usually a problem because input object files normally don't
1022 // have .build-id sections, but you can create such files by
1023 // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it.
1024 if (name == ".note.gnu.build-id" && config->buildId != BuildIdKind::None)
1025 return &InputSection::discarded;
1027 // The linker merges EH (exception handling) frames and creates a
1028 // .eh_frame_hdr section for runtime. So we handle them with a special
1029 // class. For relocatable outputs, they are just passed through.
1030 if (name == ".eh_frame" && !config->relocatable)
1031 return make<EhInputSection>(*this, sec, name);
1033 if (shouldMerge(sec))
1034 return make<MergeInputSection>(*this, sec, name);
1035 return make<InputSection>(*this, sec, name);
1038 template <class ELFT>
1039 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &sec) {
1040 return CHECK(getObj().getSectionName(&sec, sectionStringTable), this);
1043 // Initialize this->Symbols. this->Symbols is a parallel array as
1044 // its corresponding ELF symbol table.
1045 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() {
1046 ArrayRef<Elf_Sym> eSyms = this->getELFSyms<ELFT>();
1047 this->symbols.resize(eSyms.size());
1049 // Our symbol table may have already been partially initialized
1050 // because of LazyObjFile.
1051 for (size_t i = 0, end = eSyms.size(); i != end; ++i)
1052 if (!this->symbols[i] && eSyms[i].getBinding() != STB_LOCAL)
1054 symtab->insert(CHECK(eSyms[i].getName(this->stringTable), this));
1056 // Fill this->Symbols. A symbol is either local or global.
1057 for (size_t i = 0, end = eSyms.size(); i != end; ++i) {
1058 const Elf_Sym &eSym = eSyms[i];
1060 // Read symbol attributes.
1061 uint32_t secIdx = getSectionIndex(eSym);
1062 if (secIdx >= this->sections.size())
1063 fatal(toString(this) + ": invalid section index: " + Twine(secIdx));
1065 InputSectionBase *sec = this->sections[secIdx];
1066 uint8_t binding = eSym.getBinding();
1067 uint8_t stOther = eSym.st_other;
1068 uint8_t type = eSym.getType();
1069 uint64_t value = eSym.st_value;
1070 uint64_t size = eSym.st_size;
1071 StringRefZ name = this->stringTable.data() + eSym.st_name;
1073 // Handle local symbols. Local symbols are not added to the symbol
1074 // table because they are not visible from other object files. We
1075 // allocate symbol instances and add their pointers to Symbols.
1076 if (binding == STB_LOCAL) {
1077 if (eSym.getType() == STT_FILE)
1078 sourceFile = CHECK(eSym.getName(this->stringTable), this);
1080 if (this->stringTable.size() <= eSym.st_name)
1081 fatal(toString(this) + ": invalid symbol name offset");
1083 if (eSym.st_shndx == SHN_UNDEF)
1084 this->symbols[i] = make<Undefined>(this, name, binding, stOther, type);
1085 else if (sec == &InputSection::discarded)
1086 this->symbols[i] = make<Undefined>(this, name, binding, stOther, type,
1087 /*DiscardedSecIdx=*/secIdx);
1090 make<Defined>(this, name, binding, stOther, type, value, size, sec);
1094 // Handle global undefined symbols.
1095 if (eSym.st_shndx == SHN_UNDEF) {
1096 this->symbols[i]->resolve(Undefined{this, name, binding, stOther, type});
1100 // Handle global common symbols.
1101 if (eSym.st_shndx == SHN_COMMON) {
1102 if (value == 0 || value >= UINT32_MAX)
1103 fatal(toString(this) + ": common symbol '" + StringRef(name.data) +
1104 "' has invalid alignment: " + Twine(value));
1105 this->symbols[i]->resolve(
1106 CommonSymbol{this, name, binding, stOther, type, value, size});
1110 // If a defined symbol is in a discarded section, handle it as if it
1111 // were an undefined symbol. Such symbol doesn't comply with the
1112 // standard, but in practice, a .eh_frame often directly refer
1113 // COMDAT member sections, and if a comdat group is discarded, some
1114 // defined symbol in a .eh_frame becomes dangling symbols.
1115 if (sec == &InputSection::discarded) {
1116 this->symbols[i]->resolve(
1117 Undefined{this, name, binding, stOther, type, secIdx});
1121 // Handle global defined symbols.
1122 if (binding == STB_GLOBAL || binding == STB_WEAK ||
1123 binding == STB_GNU_UNIQUE) {
1124 this->symbols[i]->resolve(
1125 Defined{this, name, binding, stOther, type, value, size, sec});
1129 fatal(toString(this) + ": unexpected binding: " + Twine((int)binding));
1133 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&file)
1134 : InputFile(ArchiveKind, file->getMemoryBufferRef()),
1135 file(std::move(file)) {}
1137 void ArchiveFile::parse() {
1138 for (const Archive::Symbol &sym : file->symbols())
1139 symtab->addSymbol(LazyArchive{*this, sym});
1142 // Returns a buffer pointing to a member file containing a given symbol.
1143 void ArchiveFile::fetch(const Archive::Symbol &sym) {
1145 CHECK(sym.getMember(), toString(this) +
1146 ": could not get the member for symbol " +
1149 if (!seen.insert(c.getChildOffset()).second)
1152 MemoryBufferRef mb =
1153 CHECK(c.getMemoryBufferRef(),
1155 ": could not get the buffer for the member defining symbol " +
1158 if (tar && c.getParent()->isThin())
1159 tar->append(relativeToRoot(CHECK(c.getFullName(), this)), mb.getBuffer());
1161 InputFile *file = createObjectFile(
1162 mb, getName(), c.getParent()->isThin() ? 0 : c.getChildOffset());
1163 file->groupId = groupId;
1167 unsigned SharedFile::vernauxNum;
1169 // Parse the version definitions in the object file if present, and return a
1170 // vector whose nth element contains a pointer to the Elf_Verdef for version
1171 // identifier n. Version identifiers that are not definitions map to nullptr.
1172 template <typename ELFT>
1173 static std::vector<const void *> parseVerdefs(const uint8_t *base,
1174 const typename ELFT::Shdr *sec) {
1178 // We cannot determine the largest verdef identifier without inspecting
1179 // every Elf_Verdef, but both bfd and gold assign verdef identifiers
1180 // sequentially starting from 1, so we predict that the largest identifier
1181 // will be verdefCount.
1182 unsigned verdefCount = sec->sh_info;
1183 std::vector<const void *> verdefs(verdefCount + 1);
1185 // Build the Verdefs array by following the chain of Elf_Verdef objects
1186 // from the start of the .gnu.version_d section.
1187 const uint8_t *verdef = base + sec->sh_offset;
1188 for (unsigned i = 0; i != verdefCount; ++i) {
1189 auto *curVerdef = reinterpret_cast<const typename ELFT::Verdef *>(verdef);
1190 verdef += curVerdef->vd_next;
1191 unsigned verdefIndex = curVerdef->vd_ndx;
1192 verdefs.resize(verdefIndex + 1);
1193 verdefs[verdefIndex] = curVerdef;
1198 // We do not usually care about alignments of data in shared object
1199 // files because the loader takes care of it. However, if we promote a
1200 // DSO symbol to point to .bss due to copy relocation, we need to keep
1201 // the original alignment requirements. We infer it in this function.
1202 template <typename ELFT>
1203 static uint64_t getAlignment(ArrayRef<typename ELFT::Shdr> sections,
1204 const typename ELFT::Sym &sym) {
1205 uint64_t ret = UINT64_MAX;
1207 ret = 1ULL << countTrailingZeros((uint64_t)sym.st_value);
1208 if (0 < sym.st_shndx && sym.st_shndx < sections.size())
1209 ret = std::min<uint64_t>(ret, sections[sym.st_shndx].sh_addralign);
1210 return (ret > UINT32_MAX) ? 0 : ret;
1213 // Fully parse the shared object file.
1215 // This function parses symbol versions. If a DSO has version information,
1216 // the file has a ".gnu.version_d" section which contains symbol version
1217 // definitions. Each symbol is associated to one version through a table in
1218 // ".gnu.version" section. That table is a parallel array for the symbol
1219 // table, and each table entry contains an index in ".gnu.version_d".
1221 // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for
1222 // VER_NDX_GLOBAL. There's no table entry for these special versions in
1223 // ".gnu.version_d".
1225 // The file format for symbol versioning is perhaps a bit more complicated
1226 // than necessary, but you can easily understand the code if you wrap your
1227 // head around the data structure described above.
1228 template <class ELFT> void SharedFile::parse() {
1229 using Elf_Dyn = typename ELFT::Dyn;
1230 using Elf_Shdr = typename ELFT::Shdr;
1231 using Elf_Sym = typename ELFT::Sym;
1232 using Elf_Verdef = typename ELFT::Verdef;
1233 using Elf_Versym = typename ELFT::Versym;
1235 ArrayRef<Elf_Dyn> dynamicTags;
1236 const ELFFile<ELFT> obj = this->getObj<ELFT>();
1237 ArrayRef<Elf_Shdr> sections = CHECK(obj.sections(), this);
1239 const Elf_Shdr *versymSec = nullptr;
1240 const Elf_Shdr *verdefSec = nullptr;
1242 // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d.
1243 for (const Elf_Shdr &sec : sections) {
1244 switch (sec.sh_type) {
1249 CHECK(obj.template getSectionContentsAsArray<Elf_Dyn>(&sec), this);
1251 case SHT_GNU_versym:
1254 case SHT_GNU_verdef:
1260 if (versymSec && numELFSyms == 0) {
1261 error("SHT_GNU_versym should be associated with symbol table");
1265 // Search for a DT_SONAME tag to initialize this->soName.
1266 for (const Elf_Dyn &dyn : dynamicTags) {
1267 if (dyn.d_tag == DT_NEEDED) {
1268 uint64_t val = dyn.getVal();
1269 if (val >= this->stringTable.size())
1270 fatal(toString(this) + ": invalid DT_NEEDED entry");
1271 dtNeeded.push_back(this->stringTable.data() + val);
1272 } else if (dyn.d_tag == DT_SONAME) {
1273 uint64_t val = dyn.getVal();
1274 if (val >= this->stringTable.size())
1275 fatal(toString(this) + ": invalid DT_SONAME entry");
1276 soName = this->stringTable.data() + val;
1280 // DSOs are uniquified not by filename but by soname.
1281 DenseMap<StringRef, SharedFile *>::iterator it;
1283 std::tie(it, wasInserted) = symtab->soNames.try_emplace(soName, this);
1285 // If a DSO appears more than once on the command line with and without
1286 // --as-needed, --no-as-needed takes precedence over --as-needed because a
1287 // user can add an extra DSO with --no-as-needed to force it to be added to
1288 // the dependency list.
1289 it->second->isNeeded |= isNeeded;
1293 sharedFiles.push_back(this);
1295 verdefs = parseVerdefs<ELFT>(obj.base(), verdefSec);
1297 // Parse ".gnu.version" section which is a parallel array for the symbol
1298 // table. If a given file doesn't have a ".gnu.version" section, we use
1300 size_t size = numELFSyms - firstGlobal;
1301 std::vector<uint32_t> versyms(size, VER_NDX_GLOBAL);
1303 ArrayRef<Elf_Versym> versym =
1304 CHECK(obj.template getSectionContentsAsArray<Elf_Versym>(versymSec),
1306 .slice(firstGlobal);
1307 for (size_t i = 0; i < size; ++i)
1308 versyms[i] = versym[i].vs_index;
1311 // System libraries can have a lot of symbols with versions. Using a
1312 // fixed buffer for computing the versions name (foo@ver) can save a
1313 // lot of allocations.
1314 SmallString<0> versionedNameBuffer;
1316 // Add symbols to the symbol table.
1317 ArrayRef<Elf_Sym> syms = this->getGlobalELFSyms<ELFT>();
1318 for (size_t i = 0; i < syms.size(); ++i) {
1319 const Elf_Sym &sym = syms[i];
1321 // ELF spec requires that all local symbols precede weak or global
1322 // symbols in each symbol table, and the index of first non-local symbol
1323 // is stored to sh_info. If a local symbol appears after some non-local
1324 // symbol, that's a violation of the spec.
1325 StringRef name = CHECK(sym.getName(this->stringTable), this);
1326 if (sym.getBinding() == STB_LOCAL) {
1327 warn("found local symbol '" + name +
1328 "' in global part of symbol table in file " + toString(this));
1332 if (sym.isUndefined()) {
1333 Symbol *s = symtab->addSymbol(
1334 Undefined{this, name, sym.getBinding(), sym.st_other, sym.getType()});
1335 s->exportDynamic = true;
1339 // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly
1340 // assigns VER_NDX_LOCAL to this section global symbol. Here is a
1341 // workaround for this bug.
1342 uint32_t idx = versyms[i] & ~VERSYM_HIDDEN;
1343 if (config->emachine == EM_MIPS && idx == VER_NDX_LOCAL &&
1347 uint32_t alignment = getAlignment<ELFT>(sections, sym);
1348 if (!(versyms[i] & VERSYM_HIDDEN)) {
1349 symtab->addSymbol(SharedSymbol{*this, name, sym.getBinding(),
1350 sym.st_other, sym.getType(), sym.st_value,
1351 sym.st_size, alignment, idx});
1354 // Also add the symbol with the versioned name to handle undefined symbols
1355 // with explicit versions.
1356 if (idx == VER_NDX_GLOBAL)
1359 if (idx >= verdefs.size() || idx == VER_NDX_LOCAL) {
1360 error("corrupt input file: version definition index " + Twine(idx) +
1361 " for symbol " + name + " is out of bounds\n>>> defined in " +
1367 this->stringTable.data() +
1368 reinterpret_cast<const Elf_Verdef *>(verdefs[idx])->getAux()->vda_name;
1369 versionedNameBuffer.clear();
1370 name = (name + "@" + verName).toStringRef(versionedNameBuffer);
1371 symtab->addSymbol(SharedSymbol{*this, saver.save(name), sym.getBinding(),
1372 sym.st_other, sym.getType(), sym.st_value,
1373 sym.st_size, alignment, idx});
1377 static ELFKind getBitcodeELFKind(const Triple &t) {
1378 if (t.isLittleEndian())
1379 return t.isArch64Bit() ? ELF64LEKind : ELF32LEKind;
1380 return t.isArch64Bit() ? ELF64BEKind : ELF32BEKind;
1383 static uint8_t getBitcodeMachineKind(StringRef path, const Triple &t) {
1384 switch (t.getArch()) {
1385 case Triple::aarch64:
1387 case Triple::amdgcn:
1396 case Triple::mipsel:
1397 case Triple::mips64:
1398 case Triple::mips64el:
1400 case Triple::msp430:
1405 case Triple::ppc64le:
1407 case Triple::riscv32:
1408 case Triple::riscv64:
1411 return t.isOSIAMCU() ? EM_IAMCU : EM_386;
1412 case Triple::x86_64:
1415 error(path + ": could not infer e_machine from bitcode target triple " +
1421 BitcodeFile::BitcodeFile(MemoryBufferRef mb, StringRef archiveName,
1422 uint64_t offsetInArchive)
1423 : InputFile(BitcodeKind, mb) {
1424 this->archiveName = archiveName;
1426 std::string path = mb.getBufferIdentifier().str();
1427 if (config->thinLTOIndexOnly)
1428 path = replaceThinLTOSuffix(mb.getBufferIdentifier());
1430 // ThinLTO assumes that all MemoryBufferRefs given to it have a unique
1431 // name. If two archives define two members with the same name, this
1432 // causes a collision which result in only one of the objects being taken
1433 // into consideration at LTO time (which very likely causes undefined
1434 // symbols later in the link stage). So we append file offset to make
1436 StringRef name = archiveName.empty()
1438 : saver.save(archiveName + "(" + path + " at " +
1439 utostr(offsetInArchive) + ")");
1440 MemoryBufferRef mbref(mb.getBuffer(), name);
1442 obj = CHECK(lto::InputFile::create(mbref), this);
1444 Triple t(obj->getTargetTriple());
1445 ekind = getBitcodeELFKind(t);
1446 emachine = getBitcodeMachineKind(mb.getBufferIdentifier(), t);
1449 static uint8_t mapVisibility(GlobalValue::VisibilityTypes gvVisibility) {
1450 switch (gvVisibility) {
1451 case GlobalValue::DefaultVisibility:
1453 case GlobalValue::HiddenVisibility:
1455 case GlobalValue::ProtectedVisibility:
1456 return STV_PROTECTED;
1458 llvm_unreachable("unknown visibility");
1461 template <class ELFT>
1462 static Symbol *createBitcodeSymbol(const std::vector<bool> &keptComdats,
1463 const lto::InputFile::Symbol &objSym,
1465 StringRef name = saver.save(objSym.getName());
1466 uint8_t binding = objSym.isWeak() ? STB_WEAK : STB_GLOBAL;
1467 uint8_t type = objSym.isTLS() ? STT_TLS : STT_NOTYPE;
1468 uint8_t visibility = mapVisibility(objSym.getVisibility());
1469 bool canOmitFromDynSym = objSym.canBeOmittedFromSymbolTable();
1471 int c = objSym.getComdatIndex();
1472 if (objSym.isUndefined() || (c != -1 && !keptComdats[c])) {
1473 Undefined New(&f, name, binding, visibility, type);
1474 if (canOmitFromDynSym)
1475 New.exportDynamic = false;
1476 return symtab->addSymbol(New);
1479 if (objSym.isCommon())
1480 return symtab->addSymbol(
1481 CommonSymbol{&f, name, binding, visibility, STT_OBJECT,
1482 objSym.getCommonAlignment(), objSym.getCommonSize()});
1484 Defined New(&f, name, binding, visibility, type, 0, 0, nullptr);
1485 if (canOmitFromDynSym)
1486 New.exportDynamic = false;
1487 return symtab->addSymbol(New);
1490 template <class ELFT> void BitcodeFile::parse() {
1491 std::vector<bool> keptComdats;
1492 for (StringRef s : obj->getComdatTable())
1493 keptComdats.push_back(
1494 symtab->comdatGroups.try_emplace(CachedHashStringRef(s), this).second);
1496 for (const lto::InputFile::Symbol &objSym : obj->symbols())
1497 symbols.push_back(createBitcodeSymbol<ELFT>(keptComdats, objSym, *this));
1499 for (auto l : obj->getDependentLibraries())
1500 addDependentLibrary(l, this);
1503 void BinaryFile::parse() {
1504 ArrayRef<uint8_t> data = arrayRefFromStringRef(mb.getBuffer());
1505 auto *section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS,
1507 sections.push_back(section);
1509 // For each input file foo that is embedded to a result as a binary
1510 // blob, we define _binary_foo_{start,end,size} symbols, so that
1511 // user programs can access blobs by name. Non-alphanumeric
1512 // characters in a filename are replaced with underscore.
1513 std::string s = "_binary_" + mb.getBufferIdentifier().str();
1514 for (size_t i = 0; i < s.size(); ++i)
1518 symtab->addSymbol(Defined{nullptr, saver.save(s + "_start"), STB_GLOBAL,
1519 STV_DEFAULT, STT_OBJECT, 0, 0, section});
1520 symtab->addSymbol(Defined{nullptr, saver.save(s + "_end"), STB_GLOBAL,
1521 STV_DEFAULT, STT_OBJECT, data.size(), 0, section});
1522 symtab->addSymbol(Defined{nullptr, saver.save(s + "_size"), STB_GLOBAL,
1523 STV_DEFAULT, STT_OBJECT, data.size(), 0, nullptr});
1526 InputFile *elf::createObjectFile(MemoryBufferRef mb, StringRef archiveName,
1527 uint64_t offsetInArchive) {
1529 return make<BitcodeFile>(mb, archiveName, offsetInArchive);
1531 switch (getELFKind(mb, archiveName)) {
1533 return make<ObjFile<ELF32LE>>(mb, archiveName);
1535 return make<ObjFile<ELF32BE>>(mb, archiveName);
1537 return make<ObjFile<ELF64LE>>(mb, archiveName);
1539 return make<ObjFile<ELF64BE>>(mb, archiveName);
1541 llvm_unreachable("getELFKind");
1545 void LazyObjFile::fetch() {
1546 if (mb.getBuffer().empty())
1549 InputFile *file = createObjectFile(mb, archiveName, offsetInArchive);
1550 file->groupId = groupId;
1554 // Copy symbol vector so that the new InputFile doesn't have to
1555 // insert the same defined symbols to the symbol table again.
1556 file->symbols = std::move(symbols);
1561 template <class ELFT> void LazyObjFile::parse() {
1562 using Elf_Sym = typename ELFT::Sym;
1564 // A lazy object file wraps either a bitcode file or an ELF file.
1565 if (isBitcode(this->mb)) {
1566 std::unique_ptr<lto::InputFile> obj =
1567 CHECK(lto::InputFile::create(this->mb), this);
1568 for (const lto::InputFile::Symbol &sym : obj->symbols()) {
1569 if (sym.isUndefined())
1571 symtab->addSymbol(LazyObject{*this, saver.save(sym.getName())});
1576 if (getELFKind(this->mb, archiveName) != config->ekind) {
1577 error("incompatible file: " + this->mb.getBufferIdentifier());
1581 // Find a symbol table.
1582 ELFFile<ELFT> obj = check(ELFFile<ELFT>::create(mb.getBuffer()));
1583 ArrayRef<typename ELFT::Shdr> sections = CHECK(obj.sections(), this);
1585 for (const typename ELFT::Shdr &sec : sections) {
1586 if (sec.sh_type != SHT_SYMTAB)
1589 // A symbol table is found.
1590 ArrayRef<Elf_Sym> eSyms = CHECK(obj.symbols(&sec), this);
1591 uint32_t firstGlobal = sec.sh_info;
1592 StringRef strtab = CHECK(obj.getStringTableForSymtab(sec, sections), this);
1593 this->symbols.resize(eSyms.size());
1595 // Get existing symbols or insert placeholder symbols.
1596 for (size_t i = firstGlobal, end = eSyms.size(); i != end; ++i)
1597 if (eSyms[i].st_shndx != SHN_UNDEF)
1598 this->symbols[i] = symtab->insert(CHECK(eSyms[i].getName(strtab), this));
1600 // Replace existing symbols with LazyObject symbols.
1602 // resolve() may trigger this->fetch() if an existing symbol is an
1603 // undefined symbol. If that happens, this LazyObjFile has served
1604 // its purpose, and we can exit from the loop early.
1605 for (Symbol *sym : this->symbols) {
1608 sym->resolve(LazyObject{*this, sym->getName()});
1610 // MemoryBuffer is emptied if this file is instantiated as ObjFile.
1611 if (mb.getBuffer().empty())
1618 std::string elf::replaceThinLTOSuffix(StringRef path) {
1619 StringRef suffix = config->thinLTOObjectSuffixReplace.first;
1620 StringRef repl = config->thinLTOObjectSuffixReplace.second;
1622 if (path.consume_back(suffix))
1623 return (path + repl).str();
1627 template void BitcodeFile::parse<ELF32LE>();
1628 template void BitcodeFile::parse<ELF32BE>();
1629 template void BitcodeFile::parse<ELF64LE>();
1630 template void BitcodeFile::parse<ELF64BE>();
1632 template void LazyObjFile::parse<ELF32LE>();
1633 template void LazyObjFile::parse<ELF32BE>();
1634 template void LazyObjFile::parse<ELF64LE>();
1635 template void LazyObjFile::parse<ELF64BE>();
1637 template class elf::ObjFile<ELF32LE>;
1638 template class elf::ObjFile<ELF32BE>;
1639 template class elf::ObjFile<ELF64LE>;
1640 template class elf::ObjFile<ELF64BE>;
1642 template void SharedFile::parse<ELF32LE>();
1643 template void SharedFile::parse<ELF32BE>();
1644 template void SharedFile::parse<ELF64LE>();
1645 template void SharedFile::parse<ELF64BE>();