| File: | src/gnu/usr.bin/clang/liblldELF/../../../llvm/lld/ELF/Symbols.cpp |
| Warning: | line 265, column 5 Called C++ object pointer is null |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===- Symbols.cpp --------------------------------------------------------===// | |||
| 2 | // | |||
| 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 | |||
| 6 | // | |||
| 7 | //===----------------------------------------------------------------------===// | |||
| 8 | ||||
| 9 | #include "Symbols.h" | |||
| 10 | #include "InputFiles.h" | |||
| 11 | #include "InputSection.h" | |||
| 12 | #include "OutputSections.h" | |||
| 13 | #include "SyntheticSections.h" | |||
| 14 | #include "Target.h" | |||
| 15 | #include "Writer.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" | |||
| 21 | #include <cstring> | |||
| 22 | ||||
| 23 | using namespace llvm; | |||
| 24 | using namespace llvm::object; | |||
| 25 | using namespace llvm::ELF; | |||
| 26 | using namespace lld; | |||
| 27 | using namespace lld::elf; | |||
| 28 | ||||
| 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); | |||
| 34 | } | |||
| 35 | ||||
| 36 | std::string lld::toString(const elf::Symbol &sym) { | |||
| 37 | StringRef name = sym.getName(); | |||
| 38 | std::string ret = demangle(name); | |||
| 39 | ||||
| 40 | const char *suffix = sym.getVersionSuffix(); | |||
| 41 | if (*suffix == '@') | |||
| 42 | ret += suffix; | |||
| 43 | return ret; | |||
| 44 | } | |||
| 45 | ||||
| 46 | std::string lld::toELFString(const Archive::Symbol &b) { | |||
| 47 | return demangle(b.getName()); | |||
| 48 | } | |||
| 49 | ||||
| 50 | Defined *ElfSym::bss; | |||
| 51 | Defined *ElfSym::data; | |||
| 52 | Defined *ElfSym::etext1; | |||
| 53 | Defined *ElfSym::etext2; | |||
| 54 | Defined *ElfSym::edata1; | |||
| 55 | Defined *ElfSym::edata2; | |||
| 56 | Defined *ElfSym::end1; | |||
| 57 | Defined *ElfSym::end2; | |||
| 58 | Defined *ElfSym::globalOffsetTable; | |||
| 59 | Defined *ElfSym::mipsGp; | |||
| 60 | Defined *ElfSym::mipsGpDisp; | |||
| 61 | Defined *ElfSym::mipsLocalGp; | |||
| 62 | Defined *ElfSym::relaIpltStart; | |||
| 63 | Defined *ElfSym::relaIpltEnd; | |||
| 64 | Defined *ElfSym::riscvGlobalPointer; | |||
| 65 | Defined *ElfSym::tlsModuleBase; | |||
| 66 | DenseMap<const Symbol *, std::pair<const InputFile *, const InputFile *>> | |||
| 67 | elf::backwardReferences; | |||
| 68 | ||||
| 69 | static uint64_t getSymVA(const Symbol &sym, int64_t &addend) { | |||
| 70 | switch (sym.kind()) { | |||
| 71 | case Symbol::DefinedKind: { | |||
| 72 | auto &d = cast<Defined>(sym); | |||
| 73 | SectionBase *isec = d.section; | |||
| 74 | ||||
| 75 | // This is an absolute symbol. | |||
| 76 | if (!isec) | |||
| 77 | return d.value; | |||
| 78 | ||||
| 79 | assert(isec != &InputSection::discarded)((void)0); | |||
| 80 | isec = isec->repl; | |||
| 81 | ||||
| 82 | uint64_t offset = d.value; | |||
| 83 | ||||
| 84 | // An object in an SHF_MERGE section might be referenced via a | |||
| 85 | // section symbol (as a hack for reducing the number of local | |||
| 86 | // symbols). | |||
| 87 | // Depending on the addend, the reference via a section symbol | |||
| 88 | // refers to a different object in the merge section. | |||
| 89 | // Since the objects in the merge section are not necessarily | |||
| 90 | // contiguous in the output, the addend can thus affect the final | |||
| 91 | // VA in a non-linear way. | |||
| 92 | // To make this work, we incorporate the addend into the section | |||
| 93 | // offset (and zero out the addend for later processing) so that | |||
| 94 | // we find the right object in the section. | |||
| 95 | if (d.isSection()) { | |||
| 96 | offset += addend; | |||
| 97 | addend = 0; | |||
| 98 | } | |||
| 99 | ||||
| 100 | // In the typical case, this is actually very simple and boils | |||
| 101 | // down to adding together 3 numbers: | |||
| 102 | // 1. The address of the output section. | |||
| 103 | // 2. The offset of the input section within the output section. | |||
| 104 | // 3. The offset within the input section (this addition happens | |||
| 105 | // inside InputSection::getOffset). | |||
| 106 | // | |||
| 107 | // If you understand the data structures involved with this next | |||
| 108 | // line (and how they get built), then you have a pretty good | |||
| 109 | // understanding of the linker. | |||
| 110 | uint64_t va = isec->getVA(offset); | |||
| 111 | ||||
| 112 | // MIPS relocatable files can mix regular and microMIPS code. | |||
| 113 | // Linker needs to distinguish such code. To do so microMIPS | |||
| 114 | // symbols has the `STO_MIPS_MICROMIPS` flag in the `st_other` | |||
| 115 | // field. Unfortunately, the `MIPS::relocate()` method has | |||
| 116 | // a symbol value only. To pass type of the symbol (regular/microMIPS) | |||
| 117 | // to that routine as well as other places where we write | |||
| 118 | // a symbol value as-is (.dynamic section, `Elf_Ehdr::e_entry` | |||
| 119 | // field etc) do the same trick as compiler uses to mark microMIPS | |||
| 120 | // for CPU - set the less-significant bit. | |||
| 121 | if (config->emachine == EM_MIPS && isMicroMips() && | |||
| 122 | ((sym.stOther & STO_MIPS_MICROMIPS) || sym.needsPltAddr)) | |||
| 123 | va |= 1; | |||
| 124 | ||||
| 125 | if (d.isTls() && !config->relocatable) { | |||
| 126 | // Use the address of the TLS segment's first section rather than the | |||
| 127 | // segment's address, because segment addresses aren't initialized until | |||
| 128 | // after sections are finalized. (e.g. Measuring the size of .rela.dyn | |||
| 129 | // for Android relocation packing requires knowing TLS symbol addresses | |||
| 130 | // during section finalization.) | |||
| 131 | if (!Out::tlsPhdr || !Out::tlsPhdr->firstSec) | |||
| 132 | fatal(toString(d.file) + | |||
| 133 | " has an STT_TLS symbol but doesn't have an SHF_TLS section"); | |||
| 134 | return va - Out::tlsPhdr->firstSec->addr; | |||
| 135 | } | |||
| 136 | return va; | |||
| 137 | } | |||
| 138 | case Symbol::SharedKind: | |||
| 139 | case Symbol::UndefinedKind: | |||
| 140 | return 0; | |||
| 141 | case Symbol::LazyArchiveKind: | |||
| 142 | case Symbol::LazyObjectKind: | |||
| 143 | assert(sym.isUsedInRegularObj && "lazy symbol reached writer")((void)0); | |||
| 144 | return 0; | |||
| 145 | case Symbol::CommonKind: | |||
| 146 | llvm_unreachable("common symbol reached writer")__builtin_unreachable(); | |||
| 147 | case Symbol::PlaceholderKind: | |||
| 148 | llvm_unreachable("placeholder symbol reached writer")__builtin_unreachable(); | |||
| 149 | } | |||
| 150 | llvm_unreachable("invalid symbol kind")__builtin_unreachable(); | |||
| 151 | } | |||
| 152 | ||||
| 153 | uint64_t Symbol::getVA(int64_t addend) const { | |||
| 154 | uint64_t outVA = getSymVA(*this, addend); | |||
| 155 | return outVA + addend; | |||
| 156 | } | |||
| 157 | ||||
| 158 | uint64_t Symbol::getGotVA() const { | |||
| 159 | if (gotInIgot) | |||
| 160 | return in.igotPlt->getVA() + getGotPltOffset(); | |||
| 161 | return in.got->getVA() + getGotOffset(); | |||
| 162 | } | |||
| 163 | ||||
| 164 | uint64_t Symbol::getGotOffset() const { | |||
| 165 | return gotIndex * target->gotEntrySize; | |||
| 166 | } | |||
| 167 | ||||
| 168 | uint64_t Symbol::getGotPltVA() const { | |||
| 169 | if (isInIplt) | |||
| 170 | return in.igotPlt->getVA() + getGotPltOffset(); | |||
| 171 | return in.gotPlt->getVA() + getGotPltOffset(); | |||
| 172 | } | |||
| 173 | ||||
| 174 | uint64_t Symbol::getGotPltOffset() const { | |||
| 175 | if (isInIplt) | |||
| 176 | return pltIndex * target->gotEntrySize; | |||
| 177 | return (pltIndex + target->gotPltHeaderEntriesNum) * target->gotEntrySize; | |||
| 178 | } | |||
| 179 | ||||
| 180 | uint64_t Symbol::getPltVA() const { | |||
| 181 | uint64_t outVA = isInIplt | |||
| 182 | ? in.iplt->getVA() + pltIndex * target->ipltEntrySize | |||
| 183 | : in.plt->getVA() + in.plt->headerSize + | |||
| 184 | pltIndex * target->pltEntrySize; | |||
| 185 | ||||
| 186 | // While linking microMIPS code PLT code are always microMIPS | |||
| 187 | // code. Set the less-significant bit to track that fact. | |||
| 188 | // See detailed comment in the `getSymVA` function. | |||
| 189 | if (config->emachine == EM_MIPS && isMicroMips()) | |||
| 190 | outVA |= 1; | |||
| 191 | return outVA; | |||
| 192 | } | |||
| 193 | ||||
| 194 | uint64_t Symbol::getSize() const { | |||
| 195 | if (const auto *dr = dyn_cast<Defined>(this)) | |||
| 196 | return dr->size; | |||
| 197 | return cast<SharedSymbol>(this)->size; | |||
| 198 | } | |||
| 199 | ||||
| 200 | OutputSection *Symbol::getOutputSection() const { | |||
| 201 | if (auto *s = dyn_cast<Defined>(this)) { | |||
| 202 | if (auto *sec = s->section) | |||
| 203 | return sec->repl->getOutputSection(); | |||
| 204 | return nullptr; | |||
| 205 | } | |||
| 206 | return nullptr; | |||
| 207 | } | |||
| 208 | ||||
| 209 | // If a symbol name contains '@', the characters after that is | |||
| 210 | // a symbol version name. This function parses that. | |||
| 211 | void Symbol::parseSymbolVersion() { | |||
| 212 | // Return if localized by a local: pattern in a version script. | |||
| 213 | if (versionId == VER_NDX_LOCAL) | |||
| 214 | return; | |||
| 215 | StringRef s = getName(); | |||
| 216 | size_t pos = s.find('@'); | |||
| 217 | if (pos == 0 || pos == StringRef::npos) | |||
| 218 | return; | |||
| 219 | StringRef verstr = s.substr(pos + 1); | |||
| 220 | if (verstr.empty()) | |||
| 221 | return; | |||
| 222 | ||||
| 223 | // Truncate the symbol name so that it doesn't include the version string. | |||
| 224 | nameSize = pos; | |||
| 225 | ||||
| 226 | // If this is not in this DSO, it is not a definition. | |||
| 227 | if (!isDefined()) | |||
| 228 | return; | |||
| 229 | ||||
| 230 | // '@@' in a symbol name means the default version. | |||
| 231 | // It is usually the most recent one. | |||
| 232 | bool isDefault = (verstr[0] == '@'); | |||
| 233 | if (isDefault) | |||
| 234 | verstr = verstr.substr(1); | |||
| 235 | ||||
| 236 | for (const VersionDefinition &ver : namedVersionDefs()) { | |||
| 237 | if (ver.name != verstr) | |||
| 238 | continue; | |||
| 239 | ||||
| 240 | if (isDefault) | |||
| 241 | versionId = ver.id; | |||
| 242 | else | |||
| 243 | versionId = ver.id | VERSYM_HIDDEN; | |||
| 244 | return; | |||
| 245 | } | |||
| 246 | ||||
| 247 | // It is an error if the specified version is not defined. | |||
| 248 | // Usually version script is not provided when linking executable, | |||
| 249 | // but we may still want to override a versioned symbol from DSO, | |||
| 250 | // so we do not report error in this case. We also do not error | |||
| 251 | // if the symbol has a local version as it won't be in the dynamic | |||
| 252 | // symbol table. | |||
| 253 | if (config->shared && versionId != VER_NDX_LOCAL) | |||
| 254 | error(toString(file) + ": symbol " + s + " has undefined version " + | |||
| 255 | verstr); | |||
| 256 | } | |||
| 257 | ||||
| 258 | void Symbol::fetch() const { | |||
| 259 | if (auto *sym
| |||
| 260 | cast<ArchiveFile>(sym->file)->fetch(sym->sym); | |||
| 261 | return; | |||
| 262 | } | |||
| 263 | ||||
| 264 | if (auto *sym
| |||
| 265 | dyn_cast<LazyObjFile>(sym->file)->fetch(); | |||
| ||||
| 266 | return; | |||
| 267 | } | |||
| 268 | ||||
| 269 | llvm_unreachable("Symbol::fetch() is called on a non-lazy symbol")__builtin_unreachable(); | |||
| 270 | } | |||
| 271 | ||||
| 272 | MemoryBufferRef LazyArchive::getMemberBuffer() { | |||
| 273 | Archive::Child c = | |||
| 274 | CHECK(sym.getMember(),check2((sym.getMember()), [&] { return toString("could not get the member for symbol " + toELFString(sym)); }) | |||
| 275 | "could not get the member for symbol " + toELFString(sym))check2((sym.getMember()), [&] { return toString("could not get the member for symbol " + toELFString(sym)); }); | |||
| 276 | ||||
| 277 | return CHECK(c.getMemoryBufferRef(),check2((c.getMemoryBufferRef()), [&] { return toString("could not get the buffer for the member defining symbol " + toELFString(sym)); }) | |||
| 278 | "could not get the buffer for the member defining symbol " +check2((c.getMemoryBufferRef()), [&] { return toString("could not get the buffer for the member defining symbol " + toELFString(sym)); }) | |||
| 279 | toELFString(sym))check2((c.getMemoryBufferRef()), [&] { return toString("could not get the buffer for the member defining symbol " + toELFString(sym)); }); | |||
| 280 | } | |||
| 281 | ||||
| 282 | uint8_t Symbol::computeBinding() const { | |||
| 283 | if (config->relocatable) | |||
| 284 | return binding; | |||
| 285 | if ((visibility != STV_DEFAULT && visibility != STV_PROTECTED) || | |||
| 286 | (versionId == VER_NDX_LOCAL && !isLazy())) | |||
| 287 | return STB_LOCAL; | |||
| 288 | if (!config->gnuUnique && binding == STB_GNU_UNIQUE) | |||
| 289 | return STB_GLOBAL; | |||
| 290 | return binding; | |||
| 291 | } | |||
| 292 | ||||
| 293 | bool Symbol::includeInDynsym() const { | |||
| 294 | if (!config->hasDynSymTab) | |||
| 295 | return false; | |||
| 296 | if (computeBinding() == STB_LOCAL) | |||
| 297 | return false; | |||
| 298 | if (!isDefined() && !isCommon()) | |||
| 299 | // This should unconditionally return true, unfortunately glibc -static-pie | |||
| 300 | // expects undefined weak symbols not to exist in .dynsym, e.g. | |||
| 301 | // __pthread_mutex_lock reference in _dl_add_to_namespace_list, | |||
| 302 | // __pthread_initialize_minimal reference in csu/libc-start.c. | |||
| 303 | return !(config->noDynamicLinker && isUndefWeak()); | |||
| 304 | ||||
| 305 | return exportDynamic || inDynamicList; | |||
| 306 | } | |||
| 307 | ||||
| 308 | // Print out a log message for --trace-symbol. | |||
| 309 | void elf::printTraceSymbol(const Symbol *sym) { | |||
| 310 | std::string s; | |||
| 311 | if (sym->isUndefined()) | |||
| 312 | s = ": reference to "; | |||
| 313 | else if (sym->isLazy()) | |||
| 314 | s = ": lazy definition of "; | |||
| 315 | else if (sym->isShared()) | |||
| 316 | s = ": shared definition of "; | |||
| 317 | else if (sym->isCommon()) | |||
| 318 | s = ": common definition of "; | |||
| 319 | else | |||
| 320 | s = ": definition of "; | |||
| 321 | ||||
| 322 | message(toString(sym->file) + s + sym->getName()); | |||
| 323 | } | |||
| 324 | ||||
| 325 | void elf::maybeWarnUnorderableSymbol(const Symbol *sym) { | |||
| 326 | if (!config->warnSymbolOrdering) | |||
| 327 | return; | |||
| 328 | ||||
| 329 | // If UnresolvedPolicy::Ignore is used, no "undefined symbol" error/warning | |||
| 330 | // is emitted. It makes sense to not warn on undefined symbols. | |||
| 331 | // | |||
| 332 | // Note, ld.bfd --symbol-ordering-file= does not warn on undefined symbols, | |||
| 333 | // but we don't have to be compatible here. | |||
| 334 | if (sym->isUndefined() && | |||
| 335 | config->unresolvedSymbols == UnresolvedPolicy::Ignore) | |||
| 336 | return; | |||
| 337 | ||||
| 338 | const InputFile *file = sym->file; | |||
| 339 | auto *d = dyn_cast<Defined>(sym); | |||
| 340 | ||||
| 341 | auto report = [&](StringRef s) { warn(toString(file) + s + sym->getName()); }; | |||
| 342 | ||||
| 343 | if (sym->isUndefined()) | |||
| 344 | report(": unable to order undefined symbol: "); | |||
| 345 | else if (sym->isShared()) | |||
| 346 | report(": unable to order shared symbol: "); | |||
| 347 | else if (d && !d->section) | |||
| 348 | report(": unable to order absolute symbol: "); | |||
| 349 | else if (d && isa<OutputSection>(d->section)) | |||
| 350 | report(": unable to order synthetic symbol: "); | |||
| 351 | else if (d && !d->section->repl->isLive()) | |||
| 352 | report(": unable to order discarded symbol: "); | |||
| 353 | } | |||
| 354 | ||||
| 355 | // Returns true if a symbol can be replaced at load-time by a symbol | |||
| 356 | // with the same name defined in other ELF executable or DSO. | |||
| 357 | bool elf::computeIsPreemptible(const Symbol &sym) { | |||
| 358 | assert(!sym.isLocal())((void)0); | |||
| 359 | ||||
| 360 | // Only symbols with default visibility that appear in dynsym can be | |||
| 361 | // preempted. Symbols with protected visibility cannot be preempted. | |||
| 362 | if (!sym.includeInDynsym() || sym.visibility != STV_DEFAULT) | |||
| 363 | return false; | |||
| 364 | ||||
| 365 | // At this point copy relocations have not been created yet, so any | |||
| 366 | // symbol that is not defined locally is preemptible. | |||
| 367 | if (!sym.isDefined()) | |||
| 368 | return true; | |||
| 369 | ||||
| 370 | if (!config->shared) | |||
| 371 | return false; | |||
| 372 | ||||
| 373 | // If -Bsymbolic or --dynamic-list is specified, or -Bsymbolic-functions is | |||
| 374 | // specified and the symbol is STT_FUNC, the symbol is preemptible iff it is | |||
| 375 | // in the dynamic list. -Bsymbolic-non-weak-functions is a non-weak subset of | |||
| 376 | // -Bsymbolic-functions. | |||
| 377 | if (config->symbolic || | |||
| 378 | (config->bsymbolic == BsymbolicKind::Functions && sym.isFunc()) || | |||
| 379 | (config->bsymbolic == BsymbolicKind::NonWeakFunctions && sym.isFunc() && | |||
| 380 | sym.binding != STB_WEAK)) | |||
| 381 | return sym.inDynamicList; | |||
| 382 | return true; | |||
| 383 | } | |||
| 384 | ||||
| 385 | void elf::reportBackrefs() { | |||
| 386 | for (auto &it : backwardReferences) { | |||
| 387 | const Symbol &sym = *it.first; | |||
| 388 | std::string to = toString(it.second.second); | |||
| 389 | // Some libraries have known problems and can cause noise. Filter them out | |||
| 390 | // with --warn-backrefs-exclude=. to may look like *.o or *.a(*.o). | |||
| 391 | bool exclude = false; | |||
| 392 | for (const llvm::GlobPattern &pat : config->warnBackrefsExclude) | |||
| 393 | if (pat.match(to)) { | |||
| 394 | exclude = true; | |||
| 395 | break; | |||
| 396 | } | |||
| 397 | if (!exclude) | |||
| 398 | warn("backward reference detected: " + sym.getName() + " in " + | |||
| 399 | toString(it.second.first) + " refers to " + to); | |||
| 400 | } | |||
| 401 | } | |||
| 402 | ||||
| 403 | static uint8_t getMinVisibility(uint8_t va, uint8_t vb) { | |||
| 404 | if (va == STV_DEFAULT) | |||
| 405 | return vb; | |||
| 406 | if (vb == STV_DEFAULT) | |||
| 407 | return va; | |||
| 408 | return std::min(va, vb); | |||
| 409 | } | |||
| 410 | ||||
| 411 | // Merge symbol properties. | |||
| 412 | // | |||
| 413 | // When we have many symbols of the same name, we choose one of them, | |||
| 414 | // and that's the result of symbol resolution. However, symbols that | |||
| 415 | // were not chosen still affect some symbol properties. | |||
| 416 | void Symbol::mergeProperties(const Symbol &other) { | |||
| 417 | if (other.exportDynamic) | |||
| 418 | exportDynamic = true; | |||
| 419 | if (other.isUsedInRegularObj) | |||
| 420 | isUsedInRegularObj = true; | |||
| 421 | ||||
| 422 | // DSO symbols do not affect visibility in the output. | |||
| 423 | if (!other.isShared()) | |||
| 424 | visibility = getMinVisibility(visibility, other.visibility); | |||
| 425 | } | |||
| 426 | ||||
| 427 | void Symbol::resolve(const Symbol &other) { | |||
| 428 | mergeProperties(other); | |||
| 429 | ||||
| 430 | if (isPlaceholder()) { | |||
| ||||
| 431 | replace(other); | |||
| 432 | return; | |||
| 433 | } | |||
| 434 | ||||
| 435 | switch (other.kind()) { | |||
| 436 | case Symbol::UndefinedKind: | |||
| 437 | resolveUndefined(cast<Undefined>(other)); | |||
| 438 | break; | |||
| 439 | case Symbol::CommonKind: | |||
| 440 | resolveCommon(cast<CommonSymbol>(other)); | |||
| 441 | break; | |||
| 442 | case Symbol::DefinedKind: | |||
| 443 | resolveDefined(cast<Defined>(other)); | |||
| 444 | break; | |||
| 445 | case Symbol::LazyArchiveKind: | |||
| 446 | resolveLazy(cast<LazyArchive>(other)); | |||
| 447 | break; | |||
| 448 | case Symbol::LazyObjectKind: | |||
| 449 | resolveLazy(cast<LazyObject>(other)); | |||
| 450 | break; | |||
| 451 | case Symbol::SharedKind: | |||
| 452 | resolveShared(cast<SharedSymbol>(other)); | |||
| 453 | break; | |||
| 454 | case Symbol::PlaceholderKind: | |||
| 455 | llvm_unreachable("bad symbol kind")__builtin_unreachable(); | |||
| 456 | } | |||
| 457 | } | |||
| 458 | ||||
| 459 | void Symbol::resolveUndefined(const Undefined &other) { | |||
| 460 | // An undefined symbol with non default visibility must be satisfied | |||
| 461 | // in the same DSO. | |||
| 462 | // | |||
| 463 | // If this is a non-weak defined symbol in a discarded section, override the | |||
| 464 | // existing undefined symbol for better error message later. | |||
| 465 | if ((isShared() && other.visibility != STV_DEFAULT) || | |||
| 466 | (isUndefined() && other.binding != STB_WEAK && other.discardedSecIdx)) { | |||
| 467 | replace(other); | |||
| 468 | return; | |||
| 469 | } | |||
| 470 | ||||
| 471 | if (traced) | |||
| 472 | printTraceSymbol(&other); | |||
| 473 | ||||
| 474 | if (isLazy()) { | |||
| 475 | // An undefined weak will not fetch archive members. See comment on Lazy in | |||
| 476 | // Symbols.h for the details. | |||
| 477 | if (other.binding == STB_WEAK) { | |||
| 478 | binding = STB_WEAK; | |||
| 479 | type = other.type; | |||
| 480 | return; | |||
| 481 | } | |||
| 482 | ||||
| 483 | // Do extra check for --warn-backrefs. | |||
| 484 | // | |||
| 485 | // --warn-backrefs is an option to prevent an undefined reference from | |||
| 486 | // fetching an archive member written earlier in the command line. It can be | |||
| 487 | // used to keep compatibility with GNU linkers to some degree. | |||
| 488 | // I'll explain the feature and why you may find it useful in this comment. | |||
| 489 | // | |||
| 490 | // lld's symbol resolution semantics is more relaxed than traditional Unix | |||
| 491 | // linkers. For example, | |||
| 492 | // | |||
| 493 | // ld.lld foo.a bar.o | |||
| 494 | // | |||
| 495 | // succeeds even if bar.o contains an undefined symbol that has to be | |||
| 496 | // resolved by some object file in foo.a. Traditional Unix linkers don't | |||
| 497 | // allow this kind of backward reference, as they visit each file only once | |||
| 498 | // from left to right in the command line while resolving all undefined | |||
| 499 | // symbols at the moment of visiting. | |||
| 500 | // | |||
| 501 | // In the above case, since there's no undefined symbol when a linker visits | |||
| 502 | // foo.a, no files are pulled out from foo.a, and because the linker forgets | |||
| 503 | // about foo.a after visiting, it can't resolve undefined symbols in bar.o | |||
| 504 | // that could have been resolved otherwise. | |||
| 505 | // | |||
| 506 | // That lld accepts more relaxed form means that (besides it'd make more | |||
| 507 | // sense) you can accidentally write a command line or a build file that | |||
| 508 | // works only with lld, even if you have a plan to distribute it to wider | |||
| 509 | // users who may be using GNU linkers. With --warn-backrefs, you can detect | |||
| 510 | // a library order that doesn't work with other Unix linkers. | |||
| 511 | // | |||
| 512 | // The option is also useful to detect cyclic dependencies between static | |||
| 513 | // archives. Again, lld accepts | |||
| 514 | // | |||
| 515 | // ld.lld foo.a bar.a | |||
| 516 | // | |||
| 517 | // even if foo.a and bar.a depend on each other. With --warn-backrefs, it is | |||
| 518 | // handled as an error. | |||
| 519 | // | |||
| 520 | // Here is how the option works. We assign a group ID to each file. A file | |||
| 521 | // with a smaller group ID can pull out object files from an archive file | |||
| 522 | // with an equal or greater group ID. Otherwise, it is a reverse dependency | |||
| 523 | // and an error. | |||
| 524 | // | |||
| 525 | // A file outside --{start,end}-group gets a fresh ID when instantiated. All | |||
| 526 | // files within the same --{start,end}-group get the same group ID. E.g. | |||
| 527 | // | |||
| 528 | // ld.lld A B --start-group C D --end-group E | |||
| 529 | // | |||
| 530 | // A forms group 0. B form group 1. C and D (including their member object | |||
| 531 | // files) form group 2. E forms group 3. I think that you can see how this | |||
| 532 | // group assignment rule simulates the traditional linker's semantics. | |||
| 533 | bool backref = config->warnBackrefs && other.file && | |||
| 534 | file->groupId < other.file->groupId; | |||
| 535 | fetch(); | |||
| 536 | ||||
| 537 | // We don't report backward references to weak symbols as they can be | |||
| 538 | // overridden later. | |||
| 539 | // | |||
| 540 | // A traditional linker does not error for -ldef1 -lref -ldef2 (linking | |||
| 541 | // sandwich), where def2 may or may not be the same as def1. We don't want | |||
| 542 | // to warn for this case, so dismiss the warning if we see a subsequent lazy | |||
| 543 | // definition. this->file needs to be saved because in the case of LTO it | |||
| 544 | // may be reset to nullptr or be replaced with a file named lto.tmp. | |||
| 545 | if (backref && !isWeak()) | |||
| 546 | backwardReferences.try_emplace(this, std::make_pair(other.file, file)); | |||
| 547 | return; | |||
| 548 | } | |||
| 549 | ||||
| 550 | // Undefined symbols in a SharedFile do not change the binding. | |||
| 551 | if (dyn_cast_or_null<SharedFile>(other.file)) | |||
| 552 | return; | |||
| 553 | ||||
| 554 | if (isUndefined() || isShared()) { | |||
| 555 | // The binding will be weak if there is at least one reference and all are | |||
| 556 | // weak. The binding has one opportunity to change to weak: if the first | |||
| 557 | // reference is weak. | |||
| 558 | if (other.binding != STB_WEAK || !referenced) | |||
| 559 | binding = other.binding; | |||
| 560 | } | |||
| 561 | } | |||
| 562 | ||||
| 563 | // Using .symver foo,foo@@VER unfortunately creates two symbols: foo and | |||
| 564 | // foo@@VER. We want to effectively ignore foo, so give precedence to | |||
| 565 | // foo@@VER. | |||
| 566 | // FIXME: If users can transition to using | |||
| 567 | // .symver foo,foo@@@VER | |||
| 568 | // we can delete this hack. | |||
| 569 | static int compareVersion(StringRef a, StringRef b) { | |||
| 570 | bool x = a.contains("@@"); | |||
| 571 | bool y = b.contains("@@"); | |||
| 572 | if (!x && y) | |||
| 573 | return 1; | |||
| 574 | if (x && !y) | |||
| 575 | return -1; | |||
| 576 | return 0; | |||
| 577 | } | |||
| 578 | ||||
| 579 | // Compare two symbols. Return 1 if the new symbol should win, -1 if | |||
| 580 | // the new symbol should lose, or 0 if there is a conflict. | |||
| 581 | int Symbol::compare(const Symbol *other) const { | |||
| 582 | assert(other->isDefined() || other->isCommon())((void)0); | |||
| 583 | ||||
| 584 | if (!isDefined() && !isCommon()) | |||
| 585 | return 1; | |||
| 586 | ||||
| 587 | if (int cmp = compareVersion(getName(), other->getName())) | |||
| 588 | return cmp; | |||
| 589 | ||||
| 590 | if (other->isWeak()) | |||
| 591 | return -1; | |||
| 592 | ||||
| 593 | if (isWeak()) | |||
| 594 | return 1; | |||
| 595 | ||||
| 596 | if (isCommon() && other->isCommon()) { | |||
| 597 | if (config->warnCommon) | |||
| 598 | warn("multiple common of " + getName()); | |||
| 599 | return 0; | |||
| 600 | } | |||
| 601 | ||||
| 602 | if (isCommon()) { | |||
| 603 | if (config->warnCommon) | |||
| 604 | warn("common " + getName() + " is overridden"); | |||
| 605 | return 1; | |||
| 606 | } | |||
| 607 | ||||
| 608 | if (other->isCommon()) { | |||
| 609 | if (config->warnCommon) | |||
| 610 | warn("common " + getName() + " is overridden"); | |||
| 611 | return -1; | |||
| 612 | } | |||
| 613 | ||||
| 614 | auto *oldSym = cast<Defined>(this); | |||
| 615 | auto *newSym = cast<Defined>(other); | |||
| 616 | ||||
| 617 | if (dyn_cast_or_null<BitcodeFile>(other->file)) | |||
| 618 | return 0; | |||
| 619 | ||||
| 620 | if (!oldSym->section && !newSym->section && oldSym->value == newSym->value && | |||
| 621 | newSym->binding == STB_GLOBAL) | |||
| 622 | return -1; | |||
| 623 | ||||
| 624 | return 0; | |||
| 625 | } | |||
| 626 | ||||
| 627 | static void reportDuplicate(Symbol *sym, InputFile *newFile, | |||
| 628 | InputSectionBase *errSec, uint64_t errOffset) { | |||
| 629 | if (config->allowMultipleDefinition) | |||
| 630 | return; | |||
| 631 | ||||
| 632 | Defined *d = cast<Defined>(sym); | |||
| 633 | if (!d->section || !errSec) { | |||
| 634 | error("duplicate symbol: " + toString(*sym) + "\n>>> defined in " + | |||
| 635 | toString(sym->file) + "\n>>> defined in " + toString(newFile)); | |||
| 636 | return; | |||
| 637 | } | |||
| 638 | ||||
| 639 | // Construct and print an error message in the form of: | |||
| 640 | // | |||
| 641 | // ld.lld: error: duplicate symbol: foo | |||
| 642 | // >>> defined at bar.c:30 | |||
| 643 | // >>> bar.o (/home/alice/src/bar.o) | |||
| 644 | // >>> defined at baz.c:563 | |||
| 645 | // >>> baz.o in archive libbaz.a | |||
| 646 | auto *sec1 = cast<InputSectionBase>(d->section); | |||
| 647 | std::string src1 = sec1->getSrcMsg(*sym, d->value); | |||
| 648 | std::string obj1 = sec1->getObjMsg(d->value); | |||
| 649 | std::string src2 = errSec->getSrcMsg(*sym, errOffset); | |||
| 650 | std::string obj2 = errSec->getObjMsg(errOffset); | |||
| 651 | ||||
| 652 | std::string msg = "duplicate symbol: " + toString(*sym) + "\n>>> defined at "; | |||
| 653 | if (!src1.empty()) | |||
| 654 | msg += src1 + "\n>>> "; | |||
| 655 | msg += obj1 + "\n>>> defined at "; | |||
| 656 | if (!src2.empty()) | |||
| 657 | msg += src2 + "\n>>> "; | |||
| 658 | msg += obj2; | |||
| 659 | error(msg); | |||
| 660 | } | |||
| 661 | ||||
| 662 | void Symbol::resolveCommon(const CommonSymbol &other) { | |||
| 663 | int cmp = compare(&other); | |||
| 664 | if (cmp < 0) | |||
| 665 | return; | |||
| 666 | ||||
| 667 | if (cmp > 0) { | |||
| 668 | if (auto *s = dyn_cast<SharedSymbol>(this)) { | |||
| 669 | // Increase st_size if the shared symbol has a larger st_size. The shared | |||
| 670 | // symbol may be created from common symbols. The fact that some object | |||
| 671 | // files were linked into a shared object first should not change the | |||
| 672 | // regular rule that picks the largest st_size. | |||
| 673 | uint64_t size = s->size; | |||
| 674 | replace(other); | |||
| 675 | if (size > cast<CommonSymbol>(this)->size) | |||
| 676 | cast<CommonSymbol>(this)->size = size; | |||
| 677 | } else { | |||
| 678 | replace(other); | |||
| 679 | } | |||
| 680 | return; | |||
| 681 | } | |||
| 682 | ||||
| 683 | CommonSymbol *oldSym = cast<CommonSymbol>(this); | |||
| 684 | ||||
| 685 | oldSym->alignment = std::max(oldSym->alignment, other.alignment); | |||
| 686 | if (oldSym->size < other.size) { | |||
| 687 | oldSym->file = other.file; | |||
| 688 | oldSym->size = other.size; | |||
| 689 | } | |||
| 690 | } | |||
| 691 | ||||
| 692 | void Symbol::resolveDefined(const Defined &other) { | |||
| 693 | int cmp = compare(&other); | |||
| 694 | if (cmp > 0) | |||
| 695 | replace(other); | |||
| 696 | else if (cmp == 0) | |||
| 697 | reportDuplicate(this, other.file, | |||
| 698 | dyn_cast_or_null<InputSectionBase>(other.section), | |||
| 699 | other.value); | |||
| 700 | } | |||
| 701 | ||||
| 702 | template <class LazyT> | |||
| 703 | static void replaceCommon(Symbol &oldSym, const LazyT &newSym) { | |||
| 704 | backwardReferences.erase(&oldSym); | |||
| 705 | oldSym.replace(newSym); | |||
| 706 | newSym.fetch(); | |||
| 707 | } | |||
| 708 | ||||
| 709 | template <class LazyT> void Symbol::resolveLazy(const LazyT &other) { | |||
| 710 | // For common objects, we want to look for global or weak definitions that | |||
| 711 | // should be fetched as the canonical definition instead. | |||
| 712 | if (isCommon() && elf::config->fortranCommon) { | |||
| 713 | if (auto *laSym = dyn_cast<LazyArchive>(&other)) { | |||
| 714 | ArchiveFile *archive = cast<ArchiveFile>(laSym->file); | |||
| 715 | const Archive::Symbol &archiveSym = laSym->sym; | |||
| 716 | if (archive->shouldFetchForCommon(archiveSym)) { | |||
| 717 | replaceCommon(*this, other); | |||
| 718 | return; | |||
| 719 | } | |||
| 720 | } else if (auto *loSym = dyn_cast<LazyObject>(&other)) { | |||
| 721 | LazyObjFile *obj = cast<LazyObjFile>(loSym->file); | |||
| 722 | if (obj->shouldFetchForCommon(loSym->getName())) { | |||
| 723 | replaceCommon(*this, other); | |||
| 724 | return; | |||
| 725 | } | |||
| 726 | } | |||
| 727 | } | |||
| 728 | ||||
| 729 | if (!isUndefined()) { | |||
| 730 | // See the comment in resolveUndefined(). | |||
| 731 | if (isDefined()) | |||
| 732 | backwardReferences.erase(this); | |||
| 733 | return; | |||
| 734 | } | |||
| 735 | ||||
| 736 | // An undefined weak will not fetch archive members. See comment on Lazy in | |||
| 737 | // Symbols.h for the details. | |||
| 738 | if (isWeak()) { | |||
| 739 | uint8_t ty = type; | |||
| 740 | replace(other); | |||
| 741 | type = ty; | |||
| 742 | binding = STB_WEAK; | |||
| 743 | return; | |||
| 744 | } | |||
| 745 | ||||
| 746 | other.fetch(); | |||
| 747 | } | |||
| 748 | ||||
| 749 | void Symbol::resolveShared(const SharedSymbol &other) { | |||
| 750 | if (isCommon()) { | |||
| 751 | // See the comment in resolveCommon() above. | |||
| 752 | if (other.size > cast<CommonSymbol>(this)->size) | |||
| 753 | cast<CommonSymbol>(this)->size = other.size; | |||
| 754 | return; | |||
| 755 | } | |||
| 756 | if (visibility == STV_DEFAULT && (isUndefined() || isLazy())) { | |||
| 757 | // An undefined symbol with non default visibility must be satisfied | |||
| 758 | // in the same DSO. | |||
| 759 | uint8_t bind = binding; | |||
| 760 | replace(other); | |||
| 761 | binding = bind; | |||
| 762 | } else if (traced) | |||
| 763 | printTraceSymbol(&other); | |||
| 764 | } |