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 using namespace lld::elf;
30 Defined *ElfSym::etext1;
31 Defined *ElfSym::etext2;
32 Defined *ElfSym::edata1;
33 Defined *ElfSym::edata2;
34 Defined *ElfSym::end1;
35 Defined *ElfSym::end2;
36 Defined *ElfSym::globalOffsetTable;
37 Defined *ElfSym::mipsGp;
38 Defined *ElfSym::mipsGpDisp;
39 Defined *ElfSym::mipsLocalGp;
40 Defined *ElfSym::relaIpltStart;
41 Defined *ElfSym::relaIpltEnd;
42 Defined *ElfSym::riscvGlobalPointer;
43 Defined *ElfSym::tlsModuleBase;
45 // Returns a symbol for an error message.
46 static std::string demangle(StringRef symName) {
48 if (Optional<std::string> s = demangleItanium(symName))
53 std::string toString(const Symbol &b) { return demangle(b.getName()); }
54 std::string toELFString(const Archive::Symbol &b) {
55 return demangle(b.getName());
59 static uint64_t getSymVA(const Symbol &sym, int64_t &addend) {
61 case Symbol::DefinedKind: {
62 auto &d = cast<Defined>(sym);
63 SectionBase *isec = d.section;
65 // This is an absolute symbol.
69 assert(isec != &InputSection::discarded);
72 uint64_t offset = d.value;
74 // An object in an SHF_MERGE section might be referenced via a
75 // section symbol (as a hack for reducing the number of local
77 // Depending on the addend, the reference via a section symbol
78 // refers to a different object in the merge section.
79 // Since the objects in the merge section are not necessarily
80 // contiguous in the output, the addend can thus affect the final
81 // VA in a non-linear way.
82 // To make this work, we incorporate the addend into the section
83 // offset (and zero out the addend for later processing) so that
84 // we find the right object in the section.
90 // In the typical case, this is actually very simple and boils
91 // down to adding together 3 numbers:
92 // 1. The address of the output section.
93 // 2. The offset of the input section within the output section.
94 // 3. The offset within the input section (this addition happens
95 // inside InputSection::getOffset).
97 // If you understand the data structures involved with this next
98 // line (and how they get built), then you have a pretty good
99 // understanding of the linker.
100 uint64_t va = isec->getVA(offset);
102 // MIPS relocatable files can mix regular and microMIPS code.
103 // Linker needs to distinguish such code. To do so microMIPS
104 // symbols has the `STO_MIPS_MICROMIPS` flag in the `st_other`
105 // field. Unfortunately, the `MIPS::relocateOne()` method has
106 // a symbol value only. To pass type of the symbol (regular/microMIPS)
107 // to that routine as well as other places where we write
108 // a symbol value as-is (.dynamic section, `Elf_Ehdr::e_entry`
109 // field etc) do the same trick as compiler uses to mark microMIPS
110 // for CPU - set the less-significant bit.
111 if (config->emachine == EM_MIPS && isMicroMips() &&
112 ((sym.stOther & STO_MIPS_MICROMIPS) || sym.needsPltAddr))
115 if (d.isTls() && !config->relocatable) {
116 // Use the address of the TLS segment's first section rather than the
117 // segment's address, because segment addresses aren't initialized until
118 // after sections are finalized. (e.g. Measuring the size of .rela.dyn
119 // for Android relocation packing requires knowing TLS symbol addresses
120 // during section finalization.)
121 if (!Out::tlsPhdr || !Out::tlsPhdr->firstSec)
122 fatal(toString(d.file) +
123 " has an STT_TLS symbol but doesn't have an SHF_TLS section");
124 return va - Out::tlsPhdr->firstSec->addr;
128 case Symbol::SharedKind:
129 case Symbol::UndefinedKind:
131 case Symbol::LazyArchiveKind:
132 case Symbol::LazyObjectKind:
133 assert(sym.isUsedInRegularObj && "lazy symbol reached writer");
135 case Symbol::CommonKind:
136 llvm_unreachable("common symbol reached writer");
137 case Symbol::PlaceholderKind:
138 llvm_unreachable("placeholder symbol reached writer");
140 llvm_unreachable("invalid symbol kind");
143 uint64_t Symbol::getVA(int64_t addend) const {
144 uint64_t outVA = getSymVA(*this, addend);
145 return outVA + addend;
148 uint64_t Symbol::getGotVA() const {
150 return in.igotPlt->getVA() + getGotPltOffset();
151 return in.got->getVA() + getGotOffset();
154 uint64_t Symbol::getGotOffset() const { return gotIndex * config->wordsize; }
156 uint64_t Symbol::getGotPltVA() const {
158 return in.igotPlt->getVA() + getGotPltOffset();
159 return in.gotPlt->getVA() + getGotPltOffset();
162 uint64_t Symbol::getGotPltOffset() const {
164 return pltIndex * config->wordsize;
165 return (pltIndex + target->gotPltHeaderEntriesNum) * config->wordsize;
168 uint64_t Symbol::getPPC64LongBranchOffset() const {
169 assert(ppc64BranchltIndex != 0xffff);
170 return ppc64BranchltIndex * config->wordsize;
173 uint64_t Symbol::getPltVA() const {
174 PltSection *plt = isInIplt ? in.iplt : in.plt;
176 plt->getVA() + plt->headerSize + pltIndex * target->pltEntrySize;
177 // While linking microMIPS code PLT code are always microMIPS
178 // code. Set the less-significant bit to track that fact.
179 // See detailed comment in the `getSymVA` function.
180 if (config->emachine == EM_MIPS && isMicroMips())
185 uint64_t Symbol::getPPC64LongBranchTableVA() const {
186 assert(ppc64BranchltIndex != 0xffff);
187 return in.ppc64LongBranchTarget->getVA() +
188 ppc64BranchltIndex * config->wordsize;
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 (VersionDefinition &ver : config->versionDefinitions) {
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)
281 if (versionId == VER_NDX_LOCAL && isDefined() && !isPreemptible)
283 if (!config->gnuUnique && binding == STB_GNU_UNIQUE)
288 bool Symbol::includeInDynsym() const {
289 if (!config->hasDynSymTab)
291 if (computeBinding() == STB_LOCAL)
294 // If a PIE binary was not linked against any shared libraries, then we can
295 // safely drop weak undef symbols from .dynsym.
296 if (isUndefWeak() && config->pie && sharedFiles.empty())
299 return isUndefined() || isShared() || exportDynamic;
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 static uint8_t getMinVisibility(uint8_t va, uint8_t vb) {
350 if (va == STV_DEFAULT)
352 if (vb == STV_DEFAULT)
354 return std::min(va, vb);
357 // Merge symbol properties.
359 // When we have many symbols of the same name, we choose one of them,
360 // and that's the result of symbol resolution. However, symbols that
361 // were not chosen still affect some symbol properties.
362 void Symbol::mergeProperties(const Symbol &other) {
363 if (other.exportDynamic)
364 exportDynamic = true;
365 if (other.isUsedInRegularObj)
366 isUsedInRegularObj = true;
368 // DSO symbols do not affect visibility in the output.
369 if (!other.isShared())
370 visibility = getMinVisibility(visibility, other.visibility);
373 void Symbol::resolve(const Symbol &other) {
374 mergeProperties(other);
376 if (isPlaceholder()) {
381 switch (other.kind()) {
382 case Symbol::UndefinedKind:
383 resolveUndefined(cast<Undefined>(other));
385 case Symbol::CommonKind:
386 resolveCommon(cast<CommonSymbol>(other));
388 case Symbol::DefinedKind:
389 resolveDefined(cast<Defined>(other));
391 case Symbol::LazyArchiveKind:
392 resolveLazy(cast<LazyArchive>(other));
394 case Symbol::LazyObjectKind:
395 resolveLazy(cast<LazyObject>(other));
397 case Symbol::SharedKind:
398 resolveShared(cast<SharedSymbol>(other));
400 case Symbol::PlaceholderKind:
401 llvm_unreachable("bad symbol kind");
405 void Symbol::resolveUndefined(const Undefined &other) {
406 // An undefined symbol with non default visibility must be satisfied
409 // If this is a non-weak defined symbol in a discarded section, override the
410 // existing undefined symbol for better error message later.
411 if ((isShared() && other.visibility != STV_DEFAULT) ||
412 (isUndefined() && other.binding != STB_WEAK && other.discardedSecIdx)) {
418 printTraceSymbol(&other);
421 // An undefined weak will not fetch archive members. See comment on Lazy in
422 // Symbols.h for the details.
423 if (other.binding == STB_WEAK) {
429 // Do extra check for --warn-backrefs.
431 // --warn-backrefs is an option to prevent an undefined reference from
432 // fetching an archive member written earlier in the command line. It can be
433 // used to keep compatibility with GNU linkers to some degree.
434 // I'll explain the feature and why you may find it useful in this comment.
436 // lld's symbol resolution semantics is more relaxed than traditional Unix
437 // linkers. For example,
439 // ld.lld foo.a bar.o
441 // succeeds even if bar.o contains an undefined symbol that has to be
442 // resolved by some object file in foo.a. Traditional Unix linkers don't
443 // allow this kind of backward reference, as they visit each file only once
444 // from left to right in the command line while resolving all undefined
445 // symbols at the moment of visiting.
447 // In the above case, since there's no undefined symbol when a linker visits
448 // foo.a, no files are pulled out from foo.a, and because the linker forgets
449 // about foo.a after visiting, it can't resolve undefined symbols in bar.o
450 // that could have been resolved otherwise.
452 // That lld accepts more relaxed form means that (besides it'd make more
453 // sense) you can accidentally write a command line or a build file that
454 // works only with lld, even if you have a plan to distribute it to wider
455 // users who may be using GNU linkers. With --warn-backrefs, you can detect
456 // a library order that doesn't work with other Unix linkers.
458 // The option is also useful to detect cyclic dependencies between static
459 // archives. Again, lld accepts
461 // ld.lld foo.a bar.a
463 // even if foo.a and bar.a depend on each other. With --warn-backrefs, it is
464 // handled as an error.
466 // Here is how the option works. We assign a group ID to each file. A file
467 // with a smaller group ID can pull out object files from an archive file
468 // with an equal or greater group ID. Otherwise, it is a reverse dependency
471 // A file outside --{start,end}-group gets a fresh ID when instantiated. All
472 // files within the same --{start,end}-group get the same group ID. E.g.
474 // ld.lld A B --start-group C D --end-group E
476 // A forms group 0. B form group 1. C and D (including their member object
477 // files) form group 2. E forms group 3. I think that you can see how this
478 // group assignment rule simulates the traditional linker's semantics.
479 bool backref = config->warnBackrefs && other.file &&
480 file->groupId < other.file->groupId;
483 // We don't report backward references to weak symbols as they can be
485 if (backref && !isWeak())
486 warn("backward reference detected: " + other.getName() + " in " +
487 toString(other.file) + " refers to " + toString(file));
491 // Undefined symbols in a SharedFile do not change the binding.
492 if (dyn_cast_or_null<SharedFile>(other.file))
496 // The binding may "upgrade" from weak to non-weak.
497 if (other.binding != STB_WEAK)
498 binding = other.binding;
499 } else if (auto *s = dyn_cast<SharedSymbol>(this)) {
500 // The binding of a SharedSymbol will be weak if there is at least one
501 // reference and all are weak. The binding has one opportunity to change to
502 // weak: if the first reference is weak.
503 if (other.binding != STB_WEAK || !s->referenced)
504 binding = other.binding;
505 s->referenced = true;
509 // Using .symver foo,foo@@VER unfortunately creates two symbols: foo and
510 // foo@@VER. We want to effectively ignore foo, so give precedence to
512 // FIXME: If users can transition to using
513 // .symver foo,foo@@@VER
514 // we can delete this hack.
515 static int compareVersion(StringRef a, StringRef b) {
516 bool x = a.contains("@@");
517 bool y = b.contains("@@");
525 // Compare two symbols. Return 1 if the new symbol should win, -1 if
526 // the new symbol should lose, or 0 if there is a conflict.
527 int Symbol::compare(const Symbol *other) const {
528 assert(other->isDefined() || other->isCommon());
530 if (!isDefined() && !isCommon())
533 if (int cmp = compareVersion(getName(), other->getName()))
542 if (isCommon() && other->isCommon()) {
543 if (config->warnCommon)
544 warn("multiple common of " + getName());
549 if (config->warnCommon)
550 warn("common " + getName() + " is overridden");
554 if (other->isCommon()) {
555 if (config->warnCommon)
556 warn("common " + getName() + " is overridden");
560 auto *oldSym = cast<Defined>(this);
561 auto *newSym = cast<Defined>(other);
563 if (other->file && isa<BitcodeFile>(other->file))
566 if (!oldSym->section && !newSym->section && oldSym->value == newSym->value &&
567 newSym->binding == STB_GLOBAL)
573 static void reportDuplicate(Symbol *sym, InputFile *newFile,
574 InputSectionBase *errSec, uint64_t errOffset) {
575 if (config->allowMultipleDefinition)
578 Defined *d = cast<Defined>(sym);
579 if (!d->section || !errSec) {
580 error("duplicate symbol: " + toString(*sym) + "\n>>> defined in " +
581 toString(sym->file) + "\n>>> defined in " + toString(newFile));
585 // Construct and print an error message in the form of:
587 // ld.lld: error: duplicate symbol: foo
588 // >>> defined at bar.c:30
589 // >>> bar.o (/home/alice/src/bar.o)
590 // >>> defined at baz.c:563
591 // >>> baz.o in archive libbaz.a
592 auto *sec1 = cast<InputSectionBase>(d->section);
593 std::string src1 = sec1->getSrcMsg(*sym, d->value);
594 std::string obj1 = sec1->getObjMsg(d->value);
595 std::string src2 = errSec->getSrcMsg(*sym, errOffset);
596 std::string obj2 = errSec->getObjMsg(errOffset);
598 std::string msg = "duplicate symbol: " + toString(*sym) + "\n>>> defined at ";
600 msg += src1 + "\n>>> ";
601 msg += obj1 + "\n>>> defined at ";
603 msg += src2 + "\n>>> ";
608 void Symbol::resolveCommon(const CommonSymbol &other) {
609 int cmp = compare(&other);
618 CommonSymbol *oldSym = cast<CommonSymbol>(this);
620 oldSym->alignment = std::max(oldSym->alignment, other.alignment);
621 if (oldSym->size < other.size) {
622 oldSym->file = other.file;
623 oldSym->size = other.size;
627 void Symbol::resolveDefined(const Defined &other) {
628 int cmp = compare(&other);
632 reportDuplicate(this, other.file,
633 dyn_cast_or_null<InputSectionBase>(other.section),
637 template <class LazyT> void Symbol::resolveLazy(const LazyT &other) {
641 // An undefined weak will not fetch archive members. See comment on Lazy in
642 // Symbols.h for the details.
654 void Symbol::resolveShared(const SharedSymbol &other) {
655 if (visibility == STV_DEFAULT && (isUndefined() || isLazy())) {
656 // An undefined symbol with non default visibility must be satisfied
658 uint8_t bind = binding;
661 cast<SharedSymbol>(this)->referenced = true;