1 //===- Symbols.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 //===----------------------------------------------------------------------===//
10 #include "InputFiles.h"
11 #include "InputSection.h"
12 #include "OutputSections.h"
13 #include "SyntheticSections.h"
16 #include "lld/Common/ErrorHandler.h"
17 #include "lld/Common/Strings.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/Support/FileSystem.h"
20 #include "llvm/Support/Path.h"
24 using namespace llvm::object;
25 using namespace llvm::ELF;
27 using namespace lld::elf;
29 // Returns a symbol for an error message.
30 static std::string demangle(StringRef symName) {
31 if (elf::config->demangle)
32 return demangleItanium(symName);
33 return std::string(symName);
36 std::string lld::toString(const elf::Symbol &sym) {
37 StringRef name = sym.getName();
38 std::string ret = demangle(name);
40 const char *suffix = sym.getVersionSuffix();
46 std::string lld::toELFString(const Archive::Symbol &b) {
47 return demangle(b.getName());
51 Defined *ElfSym::etext1;
52 Defined *ElfSym::etext2;
53 Defined *ElfSym::edata1;
54 Defined *ElfSym::edata2;
55 Defined *ElfSym::end1;
56 Defined *ElfSym::end2;
57 Defined *ElfSym::globalOffsetTable;
58 Defined *ElfSym::mipsGp;
59 Defined *ElfSym::mipsGpDisp;
60 Defined *ElfSym::mipsLocalGp;
61 Defined *ElfSym::relaIpltStart;
62 Defined *ElfSym::relaIpltEnd;
63 Defined *ElfSym::riscvGlobalPointer;
64 Defined *ElfSym::tlsModuleBase;
65 DenseMap<const Symbol *, std::pair<const InputFile *, const InputFile *>>
66 elf::backwardReferences;
68 static uint64_t getSymVA(const Symbol &sym, int64_t &addend) {
70 case Symbol::DefinedKind: {
71 auto &d = cast<Defined>(sym);
72 SectionBase *isec = d.section;
74 // This is an absolute symbol.
78 assert(isec != &InputSection::discarded);
81 uint64_t offset = d.value;
83 // An object in an SHF_MERGE section might be referenced via a
84 // section symbol (as a hack for reducing the number of local
86 // Depending on the addend, the reference via a section symbol
87 // refers to a different object in the merge section.
88 // Since the objects in the merge section are not necessarily
89 // contiguous in the output, the addend can thus affect the final
90 // VA in a non-linear way.
91 // To make this work, we incorporate the addend into the section
92 // offset (and zero out the addend for later processing) so that
93 // we find the right object in the section.
99 // In the typical case, this is actually very simple and boils
100 // down to adding together 3 numbers:
101 // 1. The address of the output section.
102 // 2. The offset of the input section within the output section.
103 // 3. The offset within the input section (this addition happens
104 // inside InputSection::getOffset).
106 // If you understand the data structures involved with this next
107 // line (and how they get built), then you have a pretty good
108 // understanding of the linker.
109 uint64_t va = isec->getVA(offset);
111 // MIPS relocatable files can mix regular and microMIPS code.
112 // Linker needs to distinguish such code. To do so microMIPS
113 // symbols has the `STO_MIPS_MICROMIPS` flag in the `st_other`
114 // field. Unfortunately, the `MIPS::relocate()` method has
115 // a symbol value only. To pass type of the symbol (regular/microMIPS)
116 // to that routine as well as other places where we write
117 // a symbol value as-is (.dynamic section, `Elf_Ehdr::e_entry`
118 // field etc) do the same trick as compiler uses to mark microMIPS
119 // for CPU - set the less-significant bit.
120 if (config->emachine == EM_MIPS && isMicroMips() &&
121 ((sym.stOther & STO_MIPS_MICROMIPS) || sym.needsPltAddr))
124 if (d.isTls() && !config->relocatable) {
125 // Use the address of the TLS segment's first section rather than the
126 // segment's address, because segment addresses aren't initialized until
127 // after sections are finalized. (e.g. Measuring the size of .rela.dyn
128 // for Android relocation packing requires knowing TLS symbol addresses
129 // during section finalization.)
130 if (!Out::tlsPhdr || !Out::tlsPhdr->firstSec)
131 fatal(toString(d.file) +
132 " has an STT_TLS symbol but doesn't have an SHF_TLS section");
133 return va - Out::tlsPhdr->firstSec->addr;
137 case Symbol::SharedKind:
138 case Symbol::UndefinedKind:
140 case Symbol::LazyArchiveKind:
141 case Symbol::LazyObjectKind:
142 assert(sym.isUsedInRegularObj && "lazy symbol reached writer");
144 case Symbol::CommonKind:
145 llvm_unreachable("common symbol reached writer");
146 case Symbol::PlaceholderKind:
147 llvm_unreachable("placeholder symbol reached writer");
149 llvm_unreachable("invalid symbol kind");
152 uint64_t Symbol::getVA(int64_t addend) const {
153 uint64_t outVA = getSymVA(*this, addend);
154 return outVA + addend;
157 uint64_t Symbol::getGotVA() const {
159 return in.igotPlt->getVA() + getGotPltOffset();
160 return in.got->getVA() + getGotOffset();
163 uint64_t Symbol::getGotOffset() const { return gotIndex * config->wordsize; }
165 uint64_t Symbol::getGotPltVA() const {
167 return in.igotPlt->getVA() + getGotPltOffset();
168 return in.gotPlt->getVA() + getGotPltOffset();
171 uint64_t Symbol::getGotPltOffset() const {
173 return pltIndex * config->wordsize;
174 return (pltIndex + target->gotPltHeaderEntriesNum) * config->wordsize;
177 uint64_t Symbol::getPltVA() const {
178 uint64_t outVA = isInIplt
179 ? in.iplt->getVA() + pltIndex * target->ipltEntrySize
180 : in.plt->getVA() + in.plt->headerSize +
181 pltIndex * target->pltEntrySize;
183 // While linking microMIPS code PLT code are always microMIPS
184 // code. Set the less-significant bit to track that fact.
185 // See detailed comment in the `getSymVA` function.
186 if (config->emachine == EM_MIPS && isMicroMips())
191 uint64_t Symbol::getSize() const {
192 if (const auto *dr = dyn_cast<Defined>(this))
194 return cast<SharedSymbol>(this)->size;
197 OutputSection *Symbol::getOutputSection() const {
198 if (auto *s = dyn_cast<Defined>(this)) {
199 if (auto *sec = s->section)
200 return sec->repl->getOutputSection();
206 // If a symbol name contains '@', the characters after that is
207 // a symbol version name. This function parses that.
208 void Symbol::parseSymbolVersion() {
209 StringRef s = getName();
210 size_t pos = s.find('@');
211 if (pos == 0 || pos == StringRef::npos)
213 StringRef verstr = s.substr(pos + 1);
217 // Truncate the symbol name so that it doesn't include the version string.
220 // If this is not in this DSO, it is not a definition.
224 // '@@' in a symbol name means the default version.
225 // It is usually the most recent one.
226 bool isDefault = (verstr[0] == '@');
228 verstr = verstr.substr(1);
230 for (const VersionDefinition &ver : namedVersionDefs()) {
231 if (ver.name != verstr)
237 versionId = ver.id | VERSYM_HIDDEN;
241 // It is an error if the specified version is not defined.
242 // Usually version script is not provided when linking executable,
243 // but we may still want to override a versioned symbol from DSO,
244 // so we do not report error in this case. We also do not error
245 // if the symbol has a local version as it won't be in the dynamic
247 if (config->shared && versionId != VER_NDX_LOCAL)
248 error(toString(file) + ": symbol " + s + " has undefined version " +
252 void Symbol::fetch() const {
253 if (auto *sym = dyn_cast<LazyArchive>(this)) {
254 cast<ArchiveFile>(sym->file)->fetch(sym->sym);
258 if (auto *sym = dyn_cast<LazyObject>(this)) {
259 dyn_cast<LazyObjFile>(sym->file)->fetch();
263 llvm_unreachable("Symbol::fetch() is called on a non-lazy symbol");
266 MemoryBufferRef LazyArchive::getMemberBuffer() {
268 CHECK(sym.getMember(),
269 "could not get the member for symbol " + toELFString(sym));
271 return CHECK(c.getMemoryBufferRef(),
272 "could not get the buffer for the member defining symbol " +
276 uint8_t Symbol::computeBinding() const {
277 if (config->relocatable)
279 if ((visibility != STV_DEFAULT && visibility != STV_PROTECTED) ||
280 (versionId == VER_NDX_LOCAL && isDefined()))
282 if (!config->gnuUnique && binding == STB_GNU_UNIQUE)
287 bool Symbol::includeInDynsym() const {
288 if (!config->hasDynSymTab)
290 if (computeBinding() == STB_LOCAL)
292 if (!isDefined() && !isCommon())
293 // This should unconditionally return true, unfortunately glibc -static-pie
294 // expects undefined weak symbols not to exist in .dynsym, e.g.
295 // __pthread_mutex_lock reference in _dl_add_to_namespace_list,
296 // __pthread_initialize_minimal reference in csu/libc-start.c.
297 return !(config->noDynamicLinker && isUndefWeak());
299 return exportDynamic || inDynamicList;
302 // Print out a log message for --trace-symbol.
303 void elf::printTraceSymbol(const Symbol *sym) {
305 if (sym->isUndefined())
306 s = ": reference to ";
307 else if (sym->isLazy())
308 s = ": lazy definition of ";
309 else if (sym->isShared())
310 s = ": shared definition of ";
311 else if (sym->isCommon())
312 s = ": common definition of ";
314 s = ": definition of ";
316 message(toString(sym->file) + s + sym->getName());
319 void elf::maybeWarnUnorderableSymbol(const Symbol *sym) {
320 if (!config->warnSymbolOrdering)
323 // If UnresolvedPolicy::Ignore is used, no "undefined symbol" error/warning
324 // is emitted. It makes sense to not warn on undefined symbols.
326 // Note, ld.bfd --symbol-ordering-file= does not warn on undefined symbols,
327 // but we don't have to be compatible here.
328 if (sym->isUndefined() &&
329 config->unresolvedSymbols == UnresolvedPolicy::Ignore)
332 const InputFile *file = sym->file;
333 auto *d = dyn_cast<Defined>(sym);
335 auto report = [&](StringRef s) { warn(toString(file) + s + sym->getName()); };
337 if (sym->isUndefined())
338 report(": unable to order undefined symbol: ");
339 else if (sym->isShared())
340 report(": unable to order shared symbol: ");
341 else if (d && !d->section)
342 report(": unable to order absolute symbol: ");
343 else if (d && isa<OutputSection>(d->section))
344 report(": unable to order synthetic symbol: ");
345 else if (d && !d->section->repl->isLive())
346 report(": unable to order discarded symbol: ");
349 // Returns true if a symbol can be replaced at load-time by a symbol
350 // with the same name defined in other ELF executable or DSO.
351 bool elf::computeIsPreemptible(const Symbol &sym) {
352 assert(!sym.isLocal());
354 // Only symbols with default visibility that appear in dynsym can be
355 // preempted. Symbols with protected visibility cannot be preempted.
356 if (!sym.includeInDynsym() || sym.visibility != STV_DEFAULT)
359 // At this point copy relocations have not been created yet, so any
360 // symbol that is not defined locally is preemptible.
361 if (!sym.isDefined())
367 // If -Bsymbolic or --dynamic-list is specified, or -Bsymbolic-functions is
368 // specified and the symbol is STT_FUNC, the symbol is preemptible iff it is
369 // in the dynamic list.
370 if (config->symbolic || (config->bsymbolicFunctions && sym.isFunc()))
371 return sym.inDynamicList;
375 void elf::reportBackrefs() {
376 for (auto &it : backwardReferences) {
377 const Symbol &sym = *it.first;
378 std::string to = toString(it.second.second);
379 // Some libraries have known problems and can cause noise. Filter them out
380 // with --warn-backrefs-exclude=. to may look like *.o or *.a(*.o).
381 bool exclude = false;
382 for (const llvm::GlobPattern &pat : config->warnBackrefsExclude)
388 warn("backward reference detected: " + sym.getName() + " in " +
389 toString(it.second.first) + " refers to " + to);
393 static uint8_t getMinVisibility(uint8_t va, uint8_t vb) {
394 if (va == STV_DEFAULT)
396 if (vb == STV_DEFAULT)
398 return std::min(va, vb);
401 // Merge symbol properties.
403 // When we have many symbols of the same name, we choose one of them,
404 // and that's the result of symbol resolution. However, symbols that
405 // were not chosen still affect some symbol properties.
406 void Symbol::mergeProperties(const Symbol &other) {
407 if (other.exportDynamic)
408 exportDynamic = true;
409 if (other.isUsedInRegularObj)
410 isUsedInRegularObj = true;
412 // DSO symbols do not affect visibility in the output.
413 if (!other.isShared())
414 visibility = getMinVisibility(visibility, other.visibility);
417 void Symbol::resolve(const Symbol &other) {
418 mergeProperties(other);
420 if (isPlaceholder()) {
425 switch (other.kind()) {
426 case Symbol::UndefinedKind:
427 resolveUndefined(cast<Undefined>(other));
429 case Symbol::CommonKind:
430 resolveCommon(cast<CommonSymbol>(other));
432 case Symbol::DefinedKind:
433 resolveDefined(cast<Defined>(other));
435 case Symbol::LazyArchiveKind:
436 resolveLazy(cast<LazyArchive>(other));
438 case Symbol::LazyObjectKind:
439 resolveLazy(cast<LazyObject>(other));
441 case Symbol::SharedKind:
442 resolveShared(cast<SharedSymbol>(other));
444 case Symbol::PlaceholderKind:
445 llvm_unreachable("bad symbol kind");
449 void Symbol::resolveUndefined(const Undefined &other) {
450 // An undefined symbol with non default visibility must be satisfied
453 // If this is a non-weak defined symbol in a discarded section, override the
454 // existing undefined symbol for better error message later.
455 if ((isShared() && other.visibility != STV_DEFAULT) ||
456 (isUndefined() && other.binding != STB_WEAK && other.discardedSecIdx)) {
462 printTraceSymbol(&other);
465 // An undefined weak will not fetch archive members. See comment on Lazy in
466 // Symbols.h for the details.
467 if (other.binding == STB_WEAK) {
473 // Do extra check for --warn-backrefs.
475 // --warn-backrefs is an option to prevent an undefined reference from
476 // fetching an archive member written earlier in the command line. It can be
477 // used to keep compatibility with GNU linkers to some degree.
478 // I'll explain the feature and why you may find it useful in this comment.
480 // lld's symbol resolution semantics is more relaxed than traditional Unix
481 // linkers. For example,
483 // ld.lld foo.a bar.o
485 // succeeds even if bar.o contains an undefined symbol that has to be
486 // resolved by some object file in foo.a. Traditional Unix linkers don't
487 // allow this kind of backward reference, as they visit each file only once
488 // from left to right in the command line while resolving all undefined
489 // symbols at the moment of visiting.
491 // In the above case, since there's no undefined symbol when a linker visits
492 // foo.a, no files are pulled out from foo.a, and because the linker forgets
493 // about foo.a after visiting, it can't resolve undefined symbols in bar.o
494 // that could have been resolved otherwise.
496 // That lld accepts more relaxed form means that (besides it'd make more
497 // sense) you can accidentally write a command line or a build file that
498 // works only with lld, even if you have a plan to distribute it to wider
499 // users who may be using GNU linkers. With --warn-backrefs, you can detect
500 // a library order that doesn't work with other Unix linkers.
502 // The option is also useful to detect cyclic dependencies between static
503 // archives. Again, lld accepts
505 // ld.lld foo.a bar.a
507 // even if foo.a and bar.a depend on each other. With --warn-backrefs, it is
508 // handled as an error.
510 // Here is how the option works. We assign a group ID to each file. A file
511 // with a smaller group ID can pull out object files from an archive file
512 // with an equal or greater group ID. Otherwise, it is a reverse dependency
515 // A file outside --{start,end}-group gets a fresh ID when instantiated. All
516 // files within the same --{start,end}-group get the same group ID. E.g.
518 // ld.lld A B --start-group C D --end-group E
520 // A forms group 0. B form group 1. C and D (including their member object
521 // files) form group 2. E forms group 3. I think that you can see how this
522 // group assignment rule simulates the traditional linker's semantics.
523 bool backref = config->warnBackrefs && other.file &&
524 file->groupId < other.file->groupId;
527 // We don't report backward references to weak symbols as they can be
530 // A traditional linker does not error for -ldef1 -lref -ldef2 (linking
531 // sandwich), where def2 may or may not be the same as def1. We don't want
532 // to warn for this case, so dismiss the warning if we see a subsequent lazy
533 // definition. this->file needs to be saved because in the case of LTO it
534 // may be reset to nullptr or be replaced with a file named lto.tmp.
535 if (backref && !isWeak())
536 backwardReferences.try_emplace(this, std::make_pair(other.file, file));
540 // Undefined symbols in a SharedFile do not change the binding.
541 if (dyn_cast_or_null<SharedFile>(other.file))
544 if (isUndefined() || isShared()) {
545 // The binding will be weak if there is at least one reference and all are
546 // weak. The binding has one opportunity to change to weak: if the first
547 // reference is weak.
548 if (other.binding != STB_WEAK || !referenced)
549 binding = other.binding;
553 // Using .symver foo,foo@@VER unfortunately creates two symbols: foo and
554 // foo@@VER. We want to effectively ignore foo, so give precedence to
556 // FIXME: If users can transition to using
557 // .symver foo,foo@@@VER
558 // we can delete this hack.
559 static int compareVersion(StringRef a, StringRef b) {
560 bool x = a.contains("@@");
561 bool y = b.contains("@@");
569 // Compare two symbols. Return 1 if the new symbol should win, -1 if
570 // the new symbol should lose, or 0 if there is a conflict.
571 int Symbol::compare(const Symbol *other) const {
572 assert(other->isDefined() || other->isCommon());
574 if (!isDefined() && !isCommon())
577 if (int cmp = compareVersion(getName(), other->getName()))
586 if (isCommon() && other->isCommon()) {
587 if (config->warnCommon)
588 warn("multiple common of " + getName());
593 if (config->warnCommon)
594 warn("common " + getName() + " is overridden");
598 if (other->isCommon()) {
599 if (config->warnCommon)
600 warn("common " + getName() + " is overridden");
604 auto *oldSym = cast<Defined>(this);
605 auto *newSym = cast<Defined>(other);
607 if (dyn_cast_or_null<BitcodeFile>(other->file))
610 if (!oldSym->section && !newSym->section && oldSym->value == newSym->value &&
611 newSym->binding == STB_GLOBAL)
617 static void reportDuplicate(Symbol *sym, InputFile *newFile,
618 InputSectionBase *errSec, uint64_t errOffset) {
619 if (config->allowMultipleDefinition)
622 Defined *d = cast<Defined>(sym);
623 if (!d->section || !errSec) {
624 error("duplicate symbol: " + toString(*sym) + "\n>>> defined in " +
625 toString(sym->file) + "\n>>> defined in " + toString(newFile));
629 // Construct and print an error message in the form of:
631 // ld.lld: error: duplicate symbol: foo
632 // >>> defined at bar.c:30
633 // >>> bar.o (/home/alice/src/bar.o)
634 // >>> defined at baz.c:563
635 // >>> baz.o in archive libbaz.a
636 auto *sec1 = cast<InputSectionBase>(d->section);
637 std::string src1 = sec1->getSrcMsg(*sym, d->value);
638 std::string obj1 = sec1->getObjMsg(d->value);
639 std::string src2 = errSec->getSrcMsg(*sym, errOffset);
640 std::string obj2 = errSec->getObjMsg(errOffset);
642 std::string msg = "duplicate symbol: " + toString(*sym) + "\n>>> defined at ";
644 msg += src1 + "\n>>> ";
645 msg += obj1 + "\n>>> defined at ";
647 msg += src2 + "\n>>> ";
652 void Symbol::resolveCommon(const CommonSymbol &other) {
653 int cmp = compare(&other);
658 if (auto *s = dyn_cast<SharedSymbol>(this)) {
659 // Increase st_size if the shared symbol has a larger st_size. The shared
660 // symbol may be created from common symbols. The fact that some object
661 // files were linked into a shared object first should not change the
662 // regular rule that picks the largest st_size.
663 uint64_t size = s->size;
665 if (size > cast<CommonSymbol>(this)->size)
666 cast<CommonSymbol>(this)->size = size;
673 CommonSymbol *oldSym = cast<CommonSymbol>(this);
675 oldSym->alignment = std::max(oldSym->alignment, other.alignment);
676 if (oldSym->size < other.size) {
677 oldSym->file = other.file;
678 oldSym->size = other.size;
682 void Symbol::resolveDefined(const Defined &other) {
683 int cmp = compare(&other);
687 reportDuplicate(this, other.file,
688 dyn_cast_or_null<InputSectionBase>(other.section),
692 template <class LazyT>
693 static void replaceCommon(Symbol &oldSym, const LazyT &newSym) {
694 backwardReferences.erase(&oldSym);
695 oldSym.replace(newSym);
699 template <class LazyT> void Symbol::resolveLazy(const LazyT &other) {
700 // For common objects, we want to look for global or weak definitions that
701 // should be fetched as the cannonical definition instead.
702 if (isCommon() && elf::config->fortranCommon) {
703 if (auto *laSym = dyn_cast<LazyArchive>(&other)) {
704 ArchiveFile *archive = cast<ArchiveFile>(laSym->file);
705 const Archive::Symbol &archiveSym = laSym->sym;
706 if (archive->shouldFetchForCommon(archiveSym)) {
707 replaceCommon(*this, other);
710 } else if (auto *loSym = dyn_cast<LazyObject>(&other)) {
711 LazyObjFile *obj = cast<LazyObjFile>(loSym->file);
712 if (obj->shouldFetchForCommon(loSym->getName())) {
713 replaceCommon(*this, other);
719 if (!isUndefined()) {
720 // See the comment in resolveUndefined().
722 backwardReferences.erase(this);
726 // An undefined weak will not fetch archive members. See comment on Lazy in
727 // Symbols.h for the details.
739 void Symbol::resolveShared(const SharedSymbol &other) {
741 // See the comment in resolveCommon() above.
742 if (other.size > cast<CommonSymbol>(this)->size)
743 cast<CommonSymbol>(this)->size = other.size;
746 if (visibility == STV_DEFAULT && (isUndefined() || isLazy())) {
747 // An undefined symbol with non default visibility must be satisfied
749 uint8_t bind = binding;
753 printTraceSymbol(&other);