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/Path.h"
23 using namespace llvm::object;
24 using namespace llvm::ELF;
27 // Returns a symbol for an error message.
28 static std::string demangle(StringRef symName) {
29 if (elf::config->demangle)
30 return demangleItanium(symName);
34 std::string toString(const elf::Symbol &b) { return demangle(b.getName()); }
35 std::string toELFString(const Archive::Symbol &b) {
36 return demangle(b.getName());
41 Defined *ElfSym::etext1;
42 Defined *ElfSym::etext2;
43 Defined *ElfSym::edata1;
44 Defined *ElfSym::edata2;
45 Defined *ElfSym::end1;
46 Defined *ElfSym::end2;
47 Defined *ElfSym::globalOffsetTable;
48 Defined *ElfSym::mipsGp;
49 Defined *ElfSym::mipsGpDisp;
50 Defined *ElfSym::mipsLocalGp;
51 Defined *ElfSym::relaIpltStart;
52 Defined *ElfSym::relaIpltEnd;
53 Defined *ElfSym::riscvGlobalPointer;
54 Defined *ElfSym::tlsModuleBase;
56 static uint64_t getSymVA(const Symbol &sym, int64_t &addend) {
58 case Symbol::DefinedKind: {
59 auto &d = cast<Defined>(sym);
60 SectionBase *isec = d.section;
62 // This is an absolute symbol.
66 assert(isec != &InputSection::discarded);
69 uint64_t offset = d.value;
71 // An object in an SHF_MERGE section might be referenced via a
72 // section symbol (as a hack for reducing the number of local
74 // Depending on the addend, the reference via a section symbol
75 // refers to a different object in the merge section.
76 // Since the objects in the merge section are not necessarily
77 // contiguous in the output, the addend can thus affect the final
78 // VA in a non-linear way.
79 // To make this work, we incorporate the addend into the section
80 // offset (and zero out the addend for later processing) so that
81 // we find the right object in the section.
87 // In the typical case, this is actually very simple and boils
88 // down to adding together 3 numbers:
89 // 1. The address of the output section.
90 // 2. The offset of the input section within the output section.
91 // 3. The offset within the input section (this addition happens
92 // inside InputSection::getOffset).
94 // If you understand the data structures involved with this next
95 // line (and how they get built), then you have a pretty good
96 // understanding of the linker.
97 uint64_t va = isec->getVA(offset);
99 // MIPS relocatable files can mix regular and microMIPS code.
100 // Linker needs to distinguish such code. To do so microMIPS
101 // symbols has the `STO_MIPS_MICROMIPS` flag in the `st_other`
102 // field. Unfortunately, the `MIPS::relocateOne()` method has
103 // a symbol value only. To pass type of the symbol (regular/microMIPS)
104 // to that routine as well as other places where we write
105 // a symbol value as-is (.dynamic section, `Elf_Ehdr::e_entry`
106 // field etc) do the same trick as compiler uses to mark microMIPS
107 // for CPU - set the less-significant bit.
108 if (config->emachine == EM_MIPS && isMicroMips() &&
109 ((sym.stOther & STO_MIPS_MICROMIPS) || sym.needsPltAddr))
112 if (d.isTls() && !config->relocatable) {
113 // Use the address of the TLS segment's first section rather than the
114 // segment's address, because segment addresses aren't initialized until
115 // after sections are finalized. (e.g. Measuring the size of .rela.dyn
116 // for Android relocation packing requires knowing TLS symbol addresses
117 // during section finalization.)
118 if (!Out::tlsPhdr || !Out::tlsPhdr->firstSec)
119 fatal(toString(d.file) +
120 " has an STT_TLS symbol but doesn't have an SHF_TLS section");
121 return va - Out::tlsPhdr->firstSec->addr;
125 case Symbol::SharedKind:
126 case Symbol::UndefinedKind:
128 case Symbol::LazyArchiveKind:
129 case Symbol::LazyObjectKind:
130 assert(sym.isUsedInRegularObj && "lazy symbol reached writer");
132 case Symbol::CommonKind:
133 llvm_unreachable("common symbol reached writer");
134 case Symbol::PlaceholderKind:
135 llvm_unreachable("placeholder symbol reached writer");
137 llvm_unreachable("invalid symbol kind");
140 uint64_t Symbol::getVA(int64_t addend) const {
141 uint64_t outVA = getSymVA(*this, addend);
142 return outVA + addend;
145 uint64_t Symbol::getGotVA() const {
147 return in.igotPlt->getVA() + getGotPltOffset();
148 return in.got->getVA() + getGotOffset();
151 uint64_t Symbol::getGotOffset() const { return gotIndex * config->wordsize; }
153 uint64_t Symbol::getGotPltVA() const {
155 return in.igotPlt->getVA() + getGotPltOffset();
156 return in.gotPlt->getVA() + getGotPltOffset();
159 uint64_t Symbol::getGotPltOffset() const {
161 return pltIndex * config->wordsize;
162 return (pltIndex + target->gotPltHeaderEntriesNum) * config->wordsize;
165 uint64_t Symbol::getPltVA() const {
166 uint64_t outVA = isInIplt
167 ? in.iplt->getVA() + pltIndex * target->ipltEntrySize
168 : in.plt->getVA() + in.plt->headerSize +
169 pltIndex * target->pltEntrySize;
171 // While linking microMIPS code PLT code are always microMIPS
172 // code. Set the less-significant bit to track that fact.
173 // See detailed comment in the `getSymVA` function.
174 if (config->emachine == EM_MIPS && isMicroMips())
179 uint64_t Symbol::getSize() const {
180 if (const auto *dr = dyn_cast<Defined>(this))
182 return cast<SharedSymbol>(this)->size;
185 OutputSection *Symbol::getOutputSection() const {
186 if (auto *s = dyn_cast<Defined>(this)) {
187 if (auto *sec = s->section)
188 return sec->repl->getOutputSection();
194 // If a symbol name contains '@', the characters after that is
195 // a symbol version name. This function parses that.
196 void Symbol::parseSymbolVersion() {
197 StringRef s = getName();
198 size_t pos = s.find('@');
199 if (pos == 0 || pos == StringRef::npos)
201 StringRef verstr = s.substr(pos + 1);
205 // Truncate the symbol name so that it doesn't include the version string.
208 // If this is not in this DSO, it is not a definition.
212 // '@@' in a symbol name means the default version.
213 // It is usually the most recent one.
214 bool isDefault = (verstr[0] == '@');
216 verstr = verstr.substr(1);
218 for (const VersionDefinition &ver : namedVersionDefs()) {
219 if (ver.name != verstr)
225 versionId = ver.id | VERSYM_HIDDEN;
229 // It is an error if the specified version is not defined.
230 // Usually version script is not provided when linking executable,
231 // but we may still want to override a versioned symbol from DSO,
232 // so we do not report error in this case. We also do not error
233 // if the symbol has a local version as it won't be in the dynamic
235 if (config->shared && versionId != VER_NDX_LOCAL)
236 error(toString(file) + ": symbol " + s + " has undefined version " +
240 void Symbol::fetch() const {
241 if (auto *sym = dyn_cast<LazyArchive>(this)) {
242 cast<ArchiveFile>(sym->file)->fetch(sym->sym);
246 if (auto *sym = dyn_cast<LazyObject>(this)) {
247 dyn_cast<LazyObjFile>(sym->file)->fetch();
251 llvm_unreachable("Symbol::fetch() is called on a non-lazy symbol");
254 MemoryBufferRef LazyArchive::getMemberBuffer() {
256 CHECK(sym.getMember(),
257 "could not get the member for symbol " + toELFString(sym));
259 return CHECK(c.getMemoryBufferRef(),
260 "could not get the buffer for the member defining symbol " +
264 uint8_t Symbol::computeBinding() const {
265 if (config->relocatable)
267 if ((visibility != STV_DEFAULT && visibility != STV_PROTECTED) ||
268 versionId == VER_NDX_LOCAL)
270 if (!config->gnuUnique && binding == STB_GNU_UNIQUE)
275 bool Symbol::includeInDynsym() const {
276 if (!config->hasDynSymTab)
278 if (computeBinding() == STB_LOCAL)
280 if (!isDefined() && !isCommon())
281 // This should unconditionally return true, unfortunately glibc -static-pie
282 // expects undefined weak symbols not to exist in .dynsym, e.g.
283 // __pthread_mutex_lock reference in _dl_add_to_namespace_list,
284 // __pthread_initialize_minimal reference in csu/libc-start.c.
285 return !(config->noDynamicLinker && isUndefWeak());
287 return exportDynamic || inDynamicList;
290 // Print out a log message for --trace-symbol.
291 void printTraceSymbol(const Symbol *sym) {
293 if (sym->isUndefined())
294 s = ": reference to ";
295 else if (sym->isLazy())
296 s = ": lazy definition of ";
297 else if (sym->isShared())
298 s = ": shared definition of ";
299 else if (sym->isCommon())
300 s = ": common definition of ";
302 s = ": definition of ";
304 message(toString(sym->file) + s + sym->getName());
307 void maybeWarnUnorderableSymbol(const Symbol *sym) {
308 if (!config->warnSymbolOrdering)
311 // If UnresolvedPolicy::Ignore is used, no "undefined symbol" error/warning
312 // is emitted. It makes sense to not warn on undefined symbols.
314 // Note, ld.bfd --symbol-ordering-file= does not warn on undefined symbols,
315 // but we don't have to be compatible here.
316 if (sym->isUndefined() &&
317 config->unresolvedSymbols == UnresolvedPolicy::Ignore)
320 const InputFile *file = sym->file;
321 auto *d = dyn_cast<Defined>(sym);
323 auto report = [&](StringRef s) { warn(toString(file) + s + sym->getName()); };
325 if (sym->isUndefined())
326 report(": unable to order undefined symbol: ");
327 else if (sym->isShared())
328 report(": unable to order shared symbol: ");
329 else if (d && !d->section)
330 report(": unable to order absolute symbol: ");
331 else if (d && isa<OutputSection>(d->section))
332 report(": unable to order synthetic symbol: ");
333 else if (d && !d->section->repl->isLive())
334 report(": unable to order discarded symbol: ");
337 // Returns true if a symbol can be replaced at load-time by a symbol
338 // with the same name defined in other ELF executable or DSO.
339 bool computeIsPreemptible(const Symbol &sym) {
340 assert(!sym.isLocal());
342 // Only symbols with default visibility that appear in dynsym can be
343 // preempted. Symbols with protected visibility cannot be preempted.
344 if (!sym.includeInDynsym() || sym.visibility != STV_DEFAULT)
347 // At this point copy relocations have not been created yet, so any
348 // symbol that is not defined locally is preemptible.
349 if (!sym.isDefined())
355 // If the dynamic list is present, it specifies preemptable symbols in a DSO.
356 if (config->hasDynamicList)
357 return sym.inDynamicList;
359 // -Bsymbolic means that definitions are not preempted.
360 if (config->bsymbolic || (config->bsymbolicFunctions && sym.isFunc()))
365 static uint8_t getMinVisibility(uint8_t va, uint8_t vb) {
366 if (va == STV_DEFAULT)
368 if (vb == STV_DEFAULT)
370 return std::min(va, vb);
373 // Merge symbol properties.
375 // When we have many symbols of the same name, we choose one of them,
376 // and that's the result of symbol resolution. However, symbols that
377 // were not chosen still affect some symbol properties.
378 void Symbol::mergeProperties(const Symbol &other) {
379 if (other.exportDynamic)
380 exportDynamic = true;
381 if (other.isUsedInRegularObj)
382 isUsedInRegularObj = true;
384 // DSO symbols do not affect visibility in the output.
385 if (!other.isShared())
386 visibility = getMinVisibility(visibility, other.visibility);
389 void Symbol::resolve(const Symbol &other) {
390 mergeProperties(other);
392 if (isPlaceholder()) {
397 switch (other.kind()) {
398 case Symbol::UndefinedKind:
399 resolveUndefined(cast<Undefined>(other));
401 case Symbol::CommonKind:
402 resolveCommon(cast<CommonSymbol>(other));
404 case Symbol::DefinedKind:
405 resolveDefined(cast<Defined>(other));
407 case Symbol::LazyArchiveKind:
408 resolveLazy(cast<LazyArchive>(other));
410 case Symbol::LazyObjectKind:
411 resolveLazy(cast<LazyObject>(other));
413 case Symbol::SharedKind:
414 resolveShared(cast<SharedSymbol>(other));
416 case Symbol::PlaceholderKind:
417 llvm_unreachable("bad symbol kind");
421 void Symbol::resolveUndefined(const Undefined &other) {
422 // An undefined symbol with non default visibility must be satisfied
425 // If this is a non-weak defined symbol in a discarded section, override the
426 // existing undefined symbol for better error message later.
427 if ((isShared() && other.visibility != STV_DEFAULT) ||
428 (isUndefined() && other.binding != STB_WEAK && other.discardedSecIdx)) {
434 printTraceSymbol(&other);
437 // An undefined weak will not fetch archive members. See comment on Lazy in
438 // Symbols.h for the details.
439 if (other.binding == STB_WEAK) {
445 // Do extra check for --warn-backrefs.
447 // --warn-backrefs is an option to prevent an undefined reference from
448 // fetching an archive member written earlier in the command line. It can be
449 // used to keep compatibility with GNU linkers to some degree.
450 // I'll explain the feature and why you may find it useful in this comment.
452 // lld's symbol resolution semantics is more relaxed than traditional Unix
453 // linkers. For example,
455 // ld.lld foo.a bar.o
457 // succeeds even if bar.o contains an undefined symbol that has to be
458 // resolved by some object file in foo.a. Traditional Unix linkers don't
459 // allow this kind of backward reference, as they visit each file only once
460 // from left to right in the command line while resolving all undefined
461 // symbols at the moment of visiting.
463 // In the above case, since there's no undefined symbol when a linker visits
464 // foo.a, no files are pulled out from foo.a, and because the linker forgets
465 // about foo.a after visiting, it can't resolve undefined symbols in bar.o
466 // that could have been resolved otherwise.
468 // That lld accepts more relaxed form means that (besides it'd make more
469 // sense) you can accidentally write a command line or a build file that
470 // works only with lld, even if you have a plan to distribute it to wider
471 // users who may be using GNU linkers. With --warn-backrefs, you can detect
472 // a library order that doesn't work with other Unix linkers.
474 // The option is also useful to detect cyclic dependencies between static
475 // archives. Again, lld accepts
477 // ld.lld foo.a bar.a
479 // even if foo.a and bar.a depend on each other. With --warn-backrefs, it is
480 // handled as an error.
482 // Here is how the option works. We assign a group ID to each file. A file
483 // with a smaller group ID can pull out object files from an archive file
484 // with an equal or greater group ID. Otherwise, it is a reverse dependency
487 // A file outside --{start,end}-group gets a fresh ID when instantiated. All
488 // files within the same --{start,end}-group get the same group ID. E.g.
490 // ld.lld A B --start-group C D --end-group E
492 // A forms group 0. B form group 1. C and D (including their member object
493 // files) form group 2. E forms group 3. I think that you can see how this
494 // group assignment rule simulates the traditional linker's semantics.
495 bool backref = config->warnBackrefs && other.file &&
496 file->groupId < other.file->groupId;
499 // We don't report backward references to weak symbols as they can be
501 if (backref && !isWeak())
502 warn("backward reference detected: " + other.getName() + " in " +
503 toString(other.file) + " refers to " + toString(file));
507 // Undefined symbols in a SharedFile do not change the binding.
508 if (dyn_cast_or_null<SharedFile>(other.file))
511 if (isUndefined() || isShared()) {
512 // The binding will be weak if there is at least one reference and all are
513 // weak. The binding has one opportunity to change to weak: if the first
514 // reference is weak.
515 if (other.binding != STB_WEAK || !referenced)
516 binding = other.binding;
521 // Using .symver foo,foo@@VER unfortunately creates two symbols: foo and
522 // foo@@VER. We want to effectively ignore foo, so give precedence to
524 // FIXME: If users can transition to using
525 // .symver foo,foo@@@VER
526 // we can delete this hack.
527 static int compareVersion(StringRef a, StringRef b) {
528 bool x = a.contains("@@");
529 bool y = b.contains("@@");
537 // Compare two symbols. Return 1 if the new symbol should win, -1 if
538 // the new symbol should lose, or 0 if there is a conflict.
539 int Symbol::compare(const Symbol *other) const {
540 assert(other->isDefined() || other->isCommon());
542 if (!isDefined() && !isCommon())
545 if (int cmp = compareVersion(getName(), other->getName()))
554 if (isCommon() && other->isCommon()) {
555 if (config->warnCommon)
556 warn("multiple common of " + getName());
561 if (config->warnCommon)
562 warn("common " + getName() + " is overridden");
566 if (other->isCommon()) {
567 if (config->warnCommon)
568 warn("common " + getName() + " is overridden");
572 auto *oldSym = cast<Defined>(this);
573 auto *newSym = cast<Defined>(other);
575 if (dyn_cast_or_null<BitcodeFile>(other->file))
578 if (!oldSym->section && !newSym->section && oldSym->value == newSym->value &&
579 newSym->binding == STB_GLOBAL)
585 static void reportDuplicate(Symbol *sym, InputFile *newFile,
586 InputSectionBase *errSec, uint64_t errOffset) {
587 if (config->allowMultipleDefinition)
590 Defined *d = cast<Defined>(sym);
591 if (!d->section || !errSec) {
592 error("duplicate symbol: " + toString(*sym) + "\n>>> defined in " +
593 toString(sym->file) + "\n>>> defined in " + toString(newFile));
597 // Construct and print an error message in the form of:
599 // ld.lld: error: duplicate symbol: foo
600 // >>> defined at bar.c:30
601 // >>> bar.o (/home/alice/src/bar.o)
602 // >>> defined at baz.c:563
603 // >>> baz.o in archive libbaz.a
604 auto *sec1 = cast<InputSectionBase>(d->section);
605 std::string src1 = sec1->getSrcMsg(*sym, d->value);
606 std::string obj1 = sec1->getObjMsg(d->value);
607 std::string src2 = errSec->getSrcMsg(*sym, errOffset);
608 std::string obj2 = errSec->getObjMsg(errOffset);
610 std::string msg = "duplicate symbol: " + toString(*sym) + "\n>>> defined at ";
612 msg += src1 + "\n>>> ";
613 msg += obj1 + "\n>>> defined at ";
615 msg += src2 + "\n>>> ";
620 void Symbol::resolveCommon(const CommonSymbol &other) {
621 int cmp = compare(&other);
626 if (auto *s = dyn_cast<SharedSymbol>(this)) {
627 // Increase st_size if the shared symbol has a larger st_size. The shared
628 // symbol may be created from common symbols. The fact that some object
629 // files were linked into a shared object first should not change the
630 // regular rule that picks the largest st_size.
631 uint64_t size = s->size;
633 if (size > cast<CommonSymbol>(this)->size)
634 cast<CommonSymbol>(this)->size = size;
641 CommonSymbol *oldSym = cast<CommonSymbol>(this);
643 oldSym->alignment = std::max(oldSym->alignment, other.alignment);
644 if (oldSym->size < other.size) {
645 oldSym->file = other.file;
646 oldSym->size = other.size;
650 void Symbol::resolveDefined(const Defined &other) {
651 int cmp = compare(&other);
655 reportDuplicate(this, other.file,
656 dyn_cast_or_null<InputSectionBase>(other.section),
660 template <class LazyT> void Symbol::resolveLazy(const LazyT &other) {
664 // An undefined weak will not fetch archive members. See comment on Lazy in
665 // Symbols.h for the details.
677 void Symbol::resolveShared(const SharedSymbol &other) {
679 // See the comment in resolveCommon() above.
680 if (other.size > cast<CommonSymbol>(this)->size)
681 cast<CommonSymbol>(this)->size = other.size;
684 if (visibility == STV_DEFAULT && (isUndefined() || isLazy())) {
685 // An undefined symbol with non default visibility must be satisfied
687 uint8_t bind = binding;