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::getPPC64LongBranchOffset() const {
166 assert(ppc64BranchltIndex != 0xffff);
167 return ppc64BranchltIndex * config->wordsize;
170 uint64_t Symbol::getPltVA() const {
171 PltSection *plt = isInIplt ? in.iplt : in.plt;
173 plt->getVA() + plt->headerSize + pltIndex * target->pltEntrySize;
174 // While linking microMIPS code PLT code are always microMIPS
175 // code. Set the less-significant bit to track that fact.
176 // See detailed comment in the `getSymVA` function.
177 if (config->emachine == EM_MIPS && isMicroMips())
182 uint64_t Symbol::getPPC64LongBranchTableVA() const {
183 assert(ppc64BranchltIndex != 0xffff);
184 return in.ppc64LongBranchTarget->getVA() +
185 ppc64BranchltIndex * config->wordsize;
188 uint64_t Symbol::getSize() const {
189 if (const auto *dr = dyn_cast<Defined>(this))
191 return cast<SharedSymbol>(this)->size;
194 OutputSection *Symbol::getOutputSection() const {
195 if (auto *s = dyn_cast<Defined>(this)) {
196 if (auto *sec = s->section)
197 return sec->repl->getOutputSection();
203 // If a symbol name contains '@', the characters after that is
204 // a symbol version name. This function parses that.
205 void Symbol::parseSymbolVersion() {
206 StringRef s = getName();
207 size_t pos = s.find('@');
208 if (pos == 0 || pos == StringRef::npos)
210 StringRef verstr = s.substr(pos + 1);
214 // Truncate the symbol name so that it doesn't include the version string.
217 // If this is not in this DSO, it is not a definition.
221 // '@@' in a symbol name means the default version.
222 // It is usually the most recent one.
223 bool isDefault = (verstr[0] == '@');
225 verstr = verstr.substr(1);
227 for (const VersionDefinition &ver : namedVersionDefs()) {
228 if (ver.name != verstr)
234 versionId = ver.id | VERSYM_HIDDEN;
238 // It is an error if the specified version is not defined.
239 // Usually version script is not provided when linking executable,
240 // but we may still want to override a versioned symbol from DSO,
241 // so we do not report error in this case. We also do not error
242 // if the symbol has a local version as it won't be in the dynamic
244 if (config->shared && versionId != VER_NDX_LOCAL)
245 error(toString(file) + ": symbol " + s + " has undefined version " +
249 void Symbol::fetch() const {
250 if (auto *sym = dyn_cast<LazyArchive>(this)) {
251 cast<ArchiveFile>(sym->file)->fetch(sym->sym);
255 if (auto *sym = dyn_cast<LazyObject>(this)) {
256 dyn_cast<LazyObjFile>(sym->file)->fetch();
260 llvm_unreachable("Symbol::fetch() is called on a non-lazy symbol");
263 MemoryBufferRef LazyArchive::getMemberBuffer() {
265 CHECK(sym.getMember(),
266 "could not get the member for symbol " + toELFString(sym));
268 return CHECK(c.getMemoryBufferRef(),
269 "could not get the buffer for the member defining symbol " +
273 uint8_t Symbol::computeBinding() const {
274 if (config->relocatable)
276 if ((visibility != STV_DEFAULT && visibility != STV_PROTECTED) ||
277 versionId == VER_NDX_LOCAL)
279 if (!config->gnuUnique && binding == STB_GNU_UNIQUE)
284 bool Symbol::includeInDynsym() const {
285 if (!config->hasDynSymTab)
287 if (computeBinding() == STB_LOCAL)
290 // If a PIE binary was not linked against any shared libraries, then we can
291 // safely drop weak undef symbols from .dynsym.
292 if (isUndefWeak() && config->pie && sharedFiles.empty())
295 return isUndefined() || isShared() || exportDynamic || inDynamicList;
298 // Print out a log message for --trace-symbol.
299 void printTraceSymbol(const Symbol *sym) {
301 if (sym->isUndefined())
302 s = ": reference to ";
303 else if (sym->isLazy())
304 s = ": lazy definition of ";
305 else if (sym->isShared())
306 s = ": shared definition of ";
307 else if (sym->isCommon())
308 s = ": common definition of ";
310 s = ": definition of ";
312 message(toString(sym->file) + s + sym->getName());
315 void maybeWarnUnorderableSymbol(const Symbol *sym) {
316 if (!config->warnSymbolOrdering)
319 // If UnresolvedPolicy::Ignore is used, no "undefined symbol" error/warning
320 // is emitted. It makes sense to not warn on undefined symbols.
322 // Note, ld.bfd --symbol-ordering-file= does not warn on undefined symbols,
323 // but we don't have to be compatible here.
324 if (sym->isUndefined() &&
325 config->unresolvedSymbols == UnresolvedPolicy::Ignore)
328 const InputFile *file = sym->file;
329 auto *d = dyn_cast<Defined>(sym);
331 auto report = [&](StringRef s) { warn(toString(file) + s + sym->getName()); };
333 if (sym->isUndefined())
334 report(": unable to order undefined symbol: ");
335 else if (sym->isShared())
336 report(": unable to order shared symbol: ");
337 else if (d && !d->section)
338 report(": unable to order absolute symbol: ");
339 else if (d && isa<OutputSection>(d->section))
340 report(": unable to order synthetic symbol: ");
341 else if (d && !d->section->repl->isLive())
342 report(": unable to order discarded symbol: ");
345 static uint8_t getMinVisibility(uint8_t va, uint8_t vb) {
346 if (va == STV_DEFAULT)
348 if (vb == STV_DEFAULT)
350 return std::min(va, vb);
353 // Merge symbol properties.
355 // When we have many symbols of the same name, we choose one of them,
356 // and that's the result of symbol resolution. However, symbols that
357 // were not chosen still affect some symbol properties.
358 void Symbol::mergeProperties(const Symbol &other) {
359 if (other.exportDynamic)
360 exportDynamic = true;
361 if (other.isUsedInRegularObj)
362 isUsedInRegularObj = true;
364 // DSO symbols do not affect visibility in the output.
365 if (!other.isShared())
366 visibility = getMinVisibility(visibility, other.visibility);
369 void Symbol::resolve(const Symbol &other) {
370 mergeProperties(other);
372 if (isPlaceholder()) {
377 switch (other.kind()) {
378 case Symbol::UndefinedKind:
379 resolveUndefined(cast<Undefined>(other));
381 case Symbol::CommonKind:
382 resolveCommon(cast<CommonSymbol>(other));
384 case Symbol::DefinedKind:
385 resolveDefined(cast<Defined>(other));
387 case Symbol::LazyArchiveKind:
388 resolveLazy(cast<LazyArchive>(other));
390 case Symbol::LazyObjectKind:
391 resolveLazy(cast<LazyObject>(other));
393 case Symbol::SharedKind:
394 resolveShared(cast<SharedSymbol>(other));
396 case Symbol::PlaceholderKind:
397 llvm_unreachable("bad symbol kind");
401 void Symbol::resolveUndefined(const Undefined &other) {
402 // An undefined symbol with non default visibility must be satisfied
405 // If this is a non-weak defined symbol in a discarded section, override the
406 // existing undefined symbol for better error message later.
407 if ((isShared() && other.visibility != STV_DEFAULT) ||
408 (isUndefined() && other.binding != STB_WEAK && other.discardedSecIdx)) {
414 printTraceSymbol(&other);
417 // An undefined weak will not fetch archive members. See comment on Lazy in
418 // Symbols.h for the details.
419 if (other.binding == STB_WEAK) {
425 // Do extra check for --warn-backrefs.
427 // --warn-backrefs is an option to prevent an undefined reference from
428 // fetching an archive member written earlier in the command line. It can be
429 // used to keep compatibility with GNU linkers to some degree.
430 // I'll explain the feature and why you may find it useful in this comment.
432 // lld's symbol resolution semantics is more relaxed than traditional Unix
433 // linkers. For example,
435 // ld.lld foo.a bar.o
437 // succeeds even if bar.o contains an undefined symbol that has to be
438 // resolved by some object file in foo.a. Traditional Unix linkers don't
439 // allow this kind of backward reference, as they visit each file only once
440 // from left to right in the command line while resolving all undefined
441 // symbols at the moment of visiting.
443 // In the above case, since there's no undefined symbol when a linker visits
444 // foo.a, no files are pulled out from foo.a, and because the linker forgets
445 // about foo.a after visiting, it can't resolve undefined symbols in bar.o
446 // that could have been resolved otherwise.
448 // That lld accepts more relaxed form means that (besides it'd make more
449 // sense) you can accidentally write a command line or a build file that
450 // works only with lld, even if you have a plan to distribute it to wider
451 // users who may be using GNU linkers. With --warn-backrefs, you can detect
452 // a library order that doesn't work with other Unix linkers.
454 // The option is also useful to detect cyclic dependencies between static
455 // archives. Again, lld accepts
457 // ld.lld foo.a bar.a
459 // even if foo.a and bar.a depend on each other. With --warn-backrefs, it is
460 // handled as an error.
462 // Here is how the option works. We assign a group ID to each file. A file
463 // with a smaller group ID can pull out object files from an archive file
464 // with an equal or greater group ID. Otherwise, it is a reverse dependency
467 // A file outside --{start,end}-group gets a fresh ID when instantiated. All
468 // files within the same --{start,end}-group get the same group ID. E.g.
470 // ld.lld A B --start-group C D --end-group E
472 // A forms group 0. B form group 1. C and D (including their member object
473 // files) form group 2. E forms group 3. I think that you can see how this
474 // group assignment rule simulates the traditional linker's semantics.
475 bool backref = config->warnBackrefs && other.file &&
476 file->groupId < other.file->groupId;
479 // We don't report backward references to weak symbols as they can be
481 if (backref && !isWeak())
482 warn("backward reference detected: " + other.getName() + " in " +
483 toString(other.file) + " refers to " + toString(file));
487 // Undefined symbols in a SharedFile do not change the binding.
488 if (dyn_cast_or_null<SharedFile>(other.file))
491 if (isUndefined() || isShared()) {
492 // The binding will be weak if there is at least one reference and all are
493 // weak. The binding has one opportunity to change to weak: if the first
494 // reference is weak.
495 if (other.binding != STB_WEAK || !referenced)
496 binding = other.binding;
501 // Using .symver foo,foo@@VER unfortunately creates two symbols: foo and
502 // foo@@VER. We want to effectively ignore foo, so give precedence to
504 // FIXME: If users can transition to using
505 // .symver foo,foo@@@VER
506 // we can delete this hack.
507 static int compareVersion(StringRef a, StringRef b) {
508 bool x = a.contains("@@");
509 bool y = b.contains("@@");
517 // Compare two symbols. Return 1 if the new symbol should win, -1 if
518 // the new symbol should lose, or 0 if there is a conflict.
519 int Symbol::compare(const Symbol *other) const {
520 assert(other->isDefined() || other->isCommon());
522 if (!isDefined() && !isCommon())
525 if (int cmp = compareVersion(getName(), other->getName()))
534 if (isCommon() && other->isCommon()) {
535 if (config->warnCommon)
536 warn("multiple common of " + getName());
541 if (config->warnCommon)
542 warn("common " + getName() + " is overridden");
546 if (other->isCommon()) {
547 if (config->warnCommon)
548 warn("common " + getName() + " is overridden");
552 auto *oldSym = cast<Defined>(this);
553 auto *newSym = cast<Defined>(other);
555 if (dyn_cast_or_null<BitcodeFile>(other->file))
558 if (!oldSym->section && !newSym->section && oldSym->value == newSym->value &&
559 newSym->binding == STB_GLOBAL)
565 static void reportDuplicate(Symbol *sym, InputFile *newFile,
566 InputSectionBase *errSec, uint64_t errOffset) {
567 if (config->allowMultipleDefinition)
570 Defined *d = cast<Defined>(sym);
571 if (!d->section || !errSec) {
572 error("duplicate symbol: " + toString(*sym) + "\n>>> defined in " +
573 toString(sym->file) + "\n>>> defined in " + toString(newFile));
577 // Construct and print an error message in the form of:
579 // ld.lld: error: duplicate symbol: foo
580 // >>> defined at bar.c:30
581 // >>> bar.o (/home/alice/src/bar.o)
582 // >>> defined at baz.c:563
583 // >>> baz.o in archive libbaz.a
584 auto *sec1 = cast<InputSectionBase>(d->section);
585 std::string src1 = sec1->getSrcMsg(*sym, d->value);
586 std::string obj1 = sec1->getObjMsg(d->value);
587 std::string src2 = errSec->getSrcMsg(*sym, errOffset);
588 std::string obj2 = errSec->getObjMsg(errOffset);
590 std::string msg = "duplicate symbol: " + toString(*sym) + "\n>>> defined at ";
592 msg += src1 + "\n>>> ";
593 msg += obj1 + "\n>>> defined at ";
595 msg += src2 + "\n>>> ";
600 void Symbol::resolveCommon(const CommonSymbol &other) {
601 int cmp = compare(&other);
610 CommonSymbol *oldSym = cast<CommonSymbol>(this);
612 oldSym->alignment = std::max(oldSym->alignment, other.alignment);
613 if (oldSym->size < other.size) {
614 oldSym->file = other.file;
615 oldSym->size = other.size;
619 void Symbol::resolveDefined(const Defined &other) {
620 int cmp = compare(&other);
624 reportDuplicate(this, other.file,
625 dyn_cast_or_null<InputSectionBase>(other.section),
629 template <class LazyT> void Symbol::resolveLazy(const LazyT &other) {
633 // An undefined weak will not fetch archive members. See comment on Lazy in
634 // Symbols.h for the details.
646 void Symbol::resolveShared(const SharedSymbol &other) {
647 if (visibility == STV_DEFAULT && (isUndefined() || isLazy())) {
648 // An undefined symbol with non default visibility must be satisfied
650 uint8_t bind = binding;