| 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 "Driver.h" |
| 11 | #include "InputFiles.h" |
| 12 | #include "InputSection.h" |
| 13 | #include "OutputSections.h" |
| 14 | #include "SymbolTable.h" |
| 15 | #include "SyntheticSections.h" |
| 16 | #include "Target.h" |
| 17 | #include "Writer.h" |
| 18 | #include "lld/Common/Strings.h" |
| 19 | #include "llvm/Demangle/Demangle.h" |
| 20 | #include "llvm/Support/Compiler.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 | static_assert(sizeof(SymbolUnion) <= 64, "SymbolUnion too large" ); |
| 30 | |
| 31 | template <typename T> struct AssertSymbol { |
| 32 | static_assert(std::is_trivially_destructible<T>(), |
| 33 | "Symbol types must be trivially destructible" ); |
| 34 | static_assert(sizeof(T) <= sizeof(SymbolUnion), "SymbolUnion too small" ); |
| 35 | static_assert(alignof(T) <= alignof(SymbolUnion), |
| 36 | "SymbolUnion not aligned enough" ); |
| 37 | }; |
| 38 | |
| 39 | [[maybe_unused]] static inline void assertSymbols() { |
| 40 | AssertSymbol<Defined>(); |
| 41 | AssertSymbol<CommonSymbol>(); |
| 42 | AssertSymbol<Undefined>(); |
| 43 | AssertSymbol<SharedSymbol>(); |
| 44 | AssertSymbol<LazySymbol>(); |
| 45 | } |
| 46 | |
| 47 | // Returns a symbol for an error message. |
| 48 | static std::string maybeDemangleSymbol(Ctx &ctx, StringRef symName) { |
| 49 | return ctx.arg.demangle ? demangle(MangledName: symName.str()) : symName.str(); |
| 50 | } |
| 51 | |
| 52 | std::string elf::toStr(Ctx &ctx, const elf::Symbol &sym) { |
| 53 | StringRef name = sym.getName(); |
| 54 | std::string ret = maybeDemangleSymbol(ctx, symName: name); |
| 55 | |
| 56 | const char *suffix = sym.getVersionSuffix(); |
| 57 | if (*suffix == '@') |
| 58 | ret += suffix; |
| 59 | return ret; |
| 60 | } |
| 61 | |
| 62 | const ELFSyncStream &elf::operator<<(const ELFSyncStream &s, |
| 63 | const Symbol *sym) { |
| 64 | return s << toStr(ctx&: s.ctx, sym: *sym); |
| 65 | } |
| 66 | |
| 67 | static uint64_t getSymVA(Ctx &ctx, const Symbol &sym, int64_t addend) { |
| 68 | switch (sym.kind()) { |
| 69 | case Symbol::DefinedKind: { |
| 70 | auto &d = cast<Defined>(Val: sym); |
| 71 | SectionBase *isec = d.section; |
| 72 | |
| 73 | // This is an absolute symbol. |
| 74 | if (!isec) |
| 75 | return d.value; |
| 76 | |
| 77 | assert(isec != &InputSection::discarded); |
| 78 | |
| 79 | uint64_t offset = d.value; |
| 80 | |
| 81 | // An object in an SHF_MERGE section might be referenced via a |
| 82 | // section symbol (as a hack for reducing the number of local |
| 83 | // symbols). |
| 84 | // Depending on the addend, the reference via a section symbol |
| 85 | // refers to a different object in the merge section. |
| 86 | // Since the objects in the merge section are not necessarily |
| 87 | // contiguous in the output, the addend can thus affect the final |
| 88 | // VA in a non-linear way. |
| 89 | // To make this work, we incorporate the addend into the section |
| 90 | // offset (and zero out the addend for later processing) so that |
| 91 | // we find the right object in the section. |
| 92 | if (d.isSection()) { |
| 93 | offset += addend; |
| 94 | if (auto *ms = dyn_cast<MergeInputSection>(Val: isec); |
| 95 | ms && offset >= ms->content().size()) { |
| 96 | if (offset > ms->content().size()) |
| 97 | Err(ctx) << ms << ": offset 0x" << Twine::utohexstr(Val: offset) |
| 98 | << " is outside the section" ; |
| 99 | return 0; |
| 100 | } |
| 101 | } |
| 102 | |
| 103 | // In the typical case, this is actually very simple and boils |
| 104 | // down to adding together 3 numbers: |
| 105 | // 1. The address of the output section. |
| 106 | // 2. The offset of the input section within the output section. |
| 107 | // 3. The offset within the input section (this addition happens |
| 108 | // inside InputSection::getOffset). |
| 109 | // |
| 110 | // If you understand the data structures involved with this next |
| 111 | // line (and how they get built), then you have a pretty good |
| 112 | // understanding of the linker. |
| 113 | uint64_t va = isec->getVA(offset); |
| 114 | if (d.isSection()) |
| 115 | va -= addend; |
| 116 | |
| 117 | // MIPS relocatable files can mix regular and microMIPS code. |
| 118 | // Linker needs to distinguish such code. To do so microMIPS |
| 119 | // symbols has the `STO_MIPS_MICROMIPS` flag in the `st_other` |
| 120 | // field. Unfortunately, the `MIPS::relocate()` method has |
| 121 | // a symbol value only. To pass type of the symbol (regular/microMIPS) |
| 122 | // to that routine as well as other places where we write |
| 123 | // a symbol value as-is (.dynamic section, `Elf_Ehdr::e_entry` |
| 124 | // field etc) do the same trick as compiler uses to mark microMIPS |
| 125 | // for CPU - set the less-significant bit. |
| 126 | if (ctx.arg.emachine == EM_MIPS && isMicroMips(ctx) && |
| 127 | ((sym.stOther & STO_MIPS_MICROMIPS) || sym.hasFlag(bit: NEEDS_COPY))) |
| 128 | va |= 1; |
| 129 | |
| 130 | if (d.isTls() && !ctx.arg.relocatable) { |
| 131 | // Use the address of the TLS segment's first section rather than the |
| 132 | // segment's address, because segment addresses aren't initialized until |
| 133 | // after sections are finalized. (e.g. Measuring the size of .rela.dyn |
| 134 | // for Android relocation packing requires knowing TLS symbol addresses |
| 135 | // during section finalization.) |
| 136 | if (!ctx.tlsPhdr || !ctx.tlsPhdr->firstSec) { |
| 137 | Err(ctx) << d.file |
| 138 | << " has an STT_TLS symbol but doesn't have a PT_TLS segment" ; |
| 139 | return 0; |
| 140 | } |
| 141 | return va - ctx.tlsPhdr->firstSec->addr; |
| 142 | } |
| 143 | return va; |
| 144 | } |
| 145 | case Symbol::SharedKind: |
| 146 | case Symbol::UndefinedKind: |
| 147 | return 0; |
| 148 | case Symbol::LazyKind: |
| 149 | llvm_unreachable("lazy symbol reached writer" ); |
| 150 | case Symbol::CommonKind: |
| 151 | llvm_unreachable("common symbol reached writer" ); |
| 152 | case Symbol::PlaceholderKind: |
| 153 | llvm_unreachable("placeholder symbol reached writer" ); |
| 154 | } |
| 155 | llvm_unreachable("invalid symbol kind" ); |
| 156 | } |
| 157 | |
| 158 | uint64_t Symbol::getVA(Ctx &ctx, int64_t addend) const { |
| 159 | return getSymVA(ctx, sym: *this, addend) + addend; |
| 160 | } |
| 161 | |
| 162 | uint64_t Symbol::getGotVA(Ctx &ctx) const { |
| 163 | if (gotInIgot) |
| 164 | return ctx.in.igotPlt->getVA() + getGotPltOffset(ctx); |
| 165 | return ctx.in.got->getVA() + getGotOffset(ctx); |
| 166 | } |
| 167 | |
| 168 | uint64_t Symbol::getGotOffset(Ctx &ctx) const { |
| 169 | return getGotIdx(ctx) * ctx.target->gotEntrySize; |
| 170 | } |
| 171 | |
| 172 | uint64_t Symbol::getGotPltVA(Ctx &ctx) const { |
| 173 | if (isInIplt) |
| 174 | return ctx.in.igotPlt->getVA() + getGotPltOffset(ctx); |
| 175 | return ctx.in.gotPlt->getVA() + getGotPltOffset(ctx); |
| 176 | } |
| 177 | |
| 178 | uint64_t Symbol::getGotPltOffset(Ctx &ctx) const { |
| 179 | if (isInIplt) |
| 180 | return getPltIdx(ctx) * ctx.target->gotEntrySize; |
| 181 | return (getPltIdx(ctx) + ctx.target->gotPltHeaderEntriesNum) * |
| 182 | ctx.target->gotEntrySize; |
| 183 | } |
| 184 | |
| 185 | uint64_t Symbol::getPltVA(Ctx &ctx) const { |
| 186 | uint64_t outVA = isInIplt ? ctx.in.iplt->getVA() + |
| 187 | getPltIdx(ctx) * ctx.target->ipltEntrySize |
| 188 | : ctx.in.plt->getVA() + ctx.in.plt->headerSize + |
| 189 | getPltIdx(ctx) * ctx.target->pltEntrySize; |
| 190 | |
| 191 | // While linking microMIPS code PLT code are always microMIPS |
| 192 | // code. Set the less-significant bit to track that fact. |
| 193 | // See detailed comment in the `getSymVA` function. |
| 194 | if (ctx.arg.emachine == EM_MIPS && isMicroMips(ctx)) |
| 195 | outVA |= 1; |
| 196 | return outVA; |
| 197 | } |
| 198 | |
| 199 | uint64_t Symbol::getSize() const { |
| 200 | if (const auto *dr = dyn_cast<Defined>(Val: this)) |
| 201 | return dr->size; |
| 202 | return cast<SharedSymbol>(Val: this)->size; |
| 203 | } |
| 204 | |
| 205 | OutputSection *Symbol::getOutputSection() const { |
| 206 | if (auto *s = dyn_cast<Defined>(Val: this)) { |
| 207 | if (auto *sec = s->section) |
| 208 | return sec->getOutputSection(); |
| 209 | return nullptr; |
| 210 | } |
| 211 | return nullptr; |
| 212 | } |
| 213 | |
| 214 | // If a symbol name contains '@', the characters after that is |
| 215 | // a symbol version name. This function parses that. |
| 216 | void Symbol::parseSymbolVersion(Ctx &ctx) { |
| 217 | // Return if localized by a local: pattern in a version script. |
| 218 | if (versionId == VER_NDX_LOCAL) |
| 219 | return; |
| 220 | StringRef s = getName(); |
| 221 | size_t pos = s.find(C: '@'); |
| 222 | if (pos == StringRef::npos) |
| 223 | return; |
| 224 | StringRef verstr = s.substr(Start: pos + 1); |
| 225 | |
| 226 | // Truncate the symbol name so that it doesn't include the version string. |
| 227 | nameSize = pos; |
| 228 | |
| 229 | if (verstr.empty()) |
| 230 | return; |
| 231 | |
| 232 | // If this is not in this DSO, it is not a definition. |
| 233 | if (!isDefined()) |
| 234 | return; |
| 235 | |
| 236 | // '@@' in a symbol name means the default version. |
| 237 | // It is usually the most recent one. |
| 238 | bool isDefault = (verstr[0] == '@'); |
| 239 | if (isDefault) |
| 240 | verstr = verstr.substr(Start: 1); |
| 241 | |
| 242 | for (const VersionDefinition &ver : namedVersionDefs(ctx)) { |
| 243 | if (ver.name != verstr) |
| 244 | continue; |
| 245 | |
| 246 | // Like GNU ld, localize a versioned symbol (foo@v1 or foo@@v1) if a |
| 247 | // local: pattern in its own node v1 matches foo and no global: pattern |
| 248 | // does. |
| 249 | StringRef base = s.take_front(N: pos); |
| 250 | std::string demangled = demangle(MangledName: base); |
| 251 | auto matches = [&](ArrayRef<SymbolVersion> pats) { |
| 252 | for (const SymbolVersion &pat : pats) { |
| 253 | StringRef name = pat.isExternCpp ? StringRef(demangled) : base; |
| 254 | if (pat.hasWildcard ? SingleStringMatcher(pat.name).match(s: name) |
| 255 | : pat.name == name) |
| 256 | return true; |
| 257 | } |
| 258 | return false; |
| 259 | }; |
| 260 | if (!matches(ver.nonLocalPatterns) && matches(ver.localPatterns)) |
| 261 | versionId = VER_NDX_LOCAL; |
| 262 | else |
| 263 | versionId = isDefault ? ver.id : ver.id | VERSYM_HIDDEN; |
| 264 | return; |
| 265 | } |
| 266 | |
| 267 | // It is an error if the specified version is not defined. |
| 268 | // Usually version script is not provided when linking executable, |
| 269 | // but we may still want to override a versioned symbol from DSO, |
| 270 | // so we do not report error in this case. We also do not error |
| 271 | // if the symbol has a local version as it won't be in the dynamic |
| 272 | // symbol table. |
| 273 | if (ctx.arg.shared && versionId != VER_NDX_LOCAL) |
| 274 | ErrAlways(ctx) << file << ": symbol " << s << " has undefined version " |
| 275 | << verstr; |
| 276 | } |
| 277 | |
| 278 | void Symbol::(Ctx &ctx) const { |
| 279 | assert(file->lazy); |
| 280 | file->lazy = false; |
| 281 | parseFile(ctx, file); |
| 282 | } |
| 283 | |
| 284 | uint8_t Symbol::computeBinding(Ctx &ctx) const { |
| 285 | auto v = visibility(); |
| 286 | if ((v != STV_DEFAULT && v != STV_PROTECTED) || versionId == VER_NDX_LOCAL) |
| 287 | return STB_LOCAL; |
| 288 | if (binding == STB_GNU_UNIQUE && !ctx.arg.gnuUnique) |
| 289 | return STB_GLOBAL; |
| 290 | return binding; |
| 291 | } |
| 292 | |
| 293 | // Print out a log message for --trace-symbol. |
| 294 | void elf::printTraceSymbol(const Symbol &sym, StringRef name) { |
| 295 | std::string s; |
| 296 | if (sym.isUndefined()) |
| 297 | s = ": reference to " ; |
| 298 | else if (sym.isLazy()) |
| 299 | s = ": lazy definition of " ; |
| 300 | else if (sym.isShared()) |
| 301 | s = ": shared definition of " ; |
| 302 | else if (sym.isCommon()) |
| 303 | s = ": common definition of " ; |
| 304 | else |
| 305 | s = ": definition of " ; |
| 306 | |
| 307 | Msg(ctx&: sym.file->ctx) << sym.file << s << name; |
| 308 | } |
| 309 | |
| 310 | static void (Ctx &ctx, const InputFile *reference, |
| 311 | const InputFile &, const Symbol &sym) { |
| 312 | ctx.whyExtractRecords.emplace_back(Args: toStr(ctx, f: reference), Args: &extracted, Args: sym); |
| 313 | } |
| 314 | |
| 315 | void elf::maybeWarnUnorderableSymbol(Ctx &ctx, const Symbol *sym) { |
| 316 | if (!ctx.arg.warnSymbolOrdering) |
| 317 | return; |
| 318 | |
| 319 | // If UnresolvedPolicy::Ignore is used, no "undefined symbol" error/warning is |
| 320 | // emitted. It makes sense to not warn on undefined symbols (excluding those |
| 321 | // demoted by demoteSymbols). |
| 322 | // |
| 323 | // Note, ld.bfd --symbol-ordering-file= does not warn on undefined symbols, |
| 324 | // but we don't have to be compatible here. |
| 325 | if (sym->isUndefined() && !cast<Undefined>(Val: sym)->discardedSecIdx && |
| 326 | ctx.arg.unresolvedSymbols == UnresolvedPolicy::Ignore) |
| 327 | return; |
| 328 | |
| 329 | const InputFile *file = sym->file; |
| 330 | auto report = [&](StringRef s) { Warn(ctx) << file << s << sym->getName(); }; |
| 331 | if (auto *d = dyn_cast<Defined>(Val: sym)) { |
| 332 | if (!d->section) |
| 333 | report(": unable to order absolute symbol: " ); |
| 334 | else if (isa<OutputSection>(Val: d->section)) |
| 335 | report(": unable to order synthetic symbol: " ); |
| 336 | else if (!d->section->isLive()) |
| 337 | report(": unable to order discarded symbol: " ); |
| 338 | } else if (sym->isUndefined()) { |
| 339 | if (cast<Undefined>(Val: sym)->discardedSecIdx) |
| 340 | report(": unable to order discarded symbol: " ); |
| 341 | else |
| 342 | report(": unable to order undefined symbol: " ); |
| 343 | } else { |
| 344 | assert(sym->isShared()); |
| 345 | report(": unable to order shared symbol: " ); |
| 346 | } |
| 347 | } |
| 348 | |
| 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(Ctx &ctx, const Symbol &sym) { |
| 352 | assert(!sym.isLocal() || sym.isPlaceholder()); |
| 353 | |
| 354 | // Only symbols with default visibility that appear in dynsym can be |
| 355 | // preempted. Symbols with protected visibility cannot be preempted. |
| 356 | if (sym.visibility() != STV_DEFAULT) |
| 357 | return false; |
| 358 | |
| 359 | // At this point copy relocations have not been created yet. |
| 360 | // Shared symbols are preemptible. Undefined symbols are preemptible |
| 361 | // when zDynamicUndefined (default in dynamic linking). Weakness is not |
| 362 | // checked, though undefined non-weak would typically trigger relocation |
| 363 | // errors unless options like -z undefs are used. |
| 364 | if (!sym.isDefined()) |
| 365 | return !sym.isUndefined() || ctx.arg.zDynamicUndefined; |
| 366 | |
| 367 | if (!ctx.arg.shared) |
| 368 | return false; |
| 369 | |
| 370 | // If -Bsymbolic or --dynamic-list is specified, or -Bsymbolic-functions is |
| 371 | // specified and the symbol is STT_FUNC, the symbol is preemptible iff it is |
| 372 | // in the dynamic list. -Bsymbolic-non-weak-functions is a non-weak subset of |
| 373 | // -Bsymbolic-functions. |
| 374 | if (ctx.arg.symbolic || |
| 375 | (ctx.arg.bsymbolic == BsymbolicKind::NonWeak && |
| 376 | sym.binding != STB_WEAK) || |
| 377 | (ctx.arg.bsymbolic == BsymbolicKind::Functions && sym.isFunc()) || |
| 378 | (ctx.arg.bsymbolic == BsymbolicKind::NonWeakFunctions && sym.isFunc() && |
| 379 | sym.binding != STB_WEAK)) |
| 380 | return sym.inDynamicList; |
| 381 | return true; |
| 382 | } |
| 383 | |
| 384 | void elf::parseVersionAndComputeIsPreemptible(Ctx &ctx) { |
| 385 | // Symbol themselves might know their versions because symbols |
| 386 | // can contain versions in the form of <name>@<version>. |
| 387 | // Let them parse and update their names to exclude version suffix. |
| 388 | // In addition, compute isExported and isPreemptible. |
| 389 | for (Symbol *sym : ctx.symtab->getSymbols()) { |
| 390 | if (sym->hasVersionSuffix) |
| 391 | sym->parseSymbolVersion(ctx); |
| 392 | if (sym->computeBinding(ctx) == STB_LOCAL) { |
| 393 | sym->isExported = false; |
| 394 | continue; |
| 395 | } |
| 396 | if (!sym->isDefined() && !sym->isCommon()) { |
| 397 | sym->isPreemptible = computeIsPreemptible(ctx, sym: *sym); |
| 398 | } else if (ctx.arg.exportDynamic && |
| 399 | (sym->isUsedInRegularObj || !sym->ltoCanOmit)) { |
| 400 | sym->isExported = true; |
| 401 | sym->isPreemptible = computeIsPreemptible(ctx, sym: *sym); |
| 402 | } |
| 403 | } |
| 404 | } |
| 405 | |
| 406 | // Merge symbol properties. |
| 407 | // |
| 408 | // When we have many symbols of the same name, we choose one of them, |
| 409 | // and that's the result of symbol resolution. However, symbols that |
| 410 | // were not chosen still affect some symbol properties. |
| 411 | void Symbol::mergeProperties(const Symbol &other) { |
| 412 | // DSO symbols do not affect visibility in the output. |
| 413 | if (!other.isShared() && other.visibility() != STV_DEFAULT) { |
| 414 | uint8_t v = visibility(), ov = other.visibility(); |
| 415 | setVisibility(v == STV_DEFAULT ? ov : std::min(a: v, b: ov)); |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | void Symbol::resolve(Ctx &ctx, const Undefined &other) { |
| 420 | if (other.visibility() != STV_DEFAULT) { |
| 421 | uint8_t v = visibility(), ov = other.visibility(); |
| 422 | setVisibility(v == STV_DEFAULT ? ov : std::min(a: v, b: ov)); |
| 423 | } |
| 424 | // An undefined symbol with non default visibility must be satisfied |
| 425 | // in the same DSO. |
| 426 | // |
| 427 | // If this is a non-weak defined symbol in a discarded section, override the |
| 428 | // existing undefined symbol for better error message later. |
| 429 | if (isPlaceholder() || (isShared() && other.visibility() != STV_DEFAULT) || |
| 430 | (isUndefined() && other.binding != STB_WEAK && other.discardedSecIdx)) { |
| 431 | other.overwrite(sym&: *this); |
| 432 | return; |
| 433 | } |
| 434 | |
| 435 | if (traced) |
| 436 | printTraceSymbol(sym: other, name: getName()); |
| 437 | |
| 438 | if (isLazy()) { |
| 439 | // An undefined weak will not extract archive members. See comment on Lazy |
| 440 | // in Symbols.h for the details. |
| 441 | if (other.binding == STB_WEAK) { |
| 442 | binding = STB_WEAK; |
| 443 | type = other.type; |
| 444 | return; |
| 445 | } |
| 446 | |
| 447 | // Do extra check for --warn-backrefs. |
| 448 | // |
| 449 | // --warn-backrefs is an option to prevent an undefined reference from |
| 450 | // extracting an archive member written earlier in the command line. It can |
| 451 | // be used to keep compatibility with GNU linkers to some degree. I'll |
| 452 | // explain the feature and why you may find it useful in this comment. |
| 453 | // |
| 454 | // lld's symbol resolution semantics is more relaxed than traditional Unix |
| 455 | // linkers. For example, |
| 456 | // |
| 457 | // ld.lld foo.a bar.o |
| 458 | // |
| 459 | // succeeds even if bar.o contains an undefined symbol that has to be |
| 460 | // resolved by some object file in foo.a. Traditional Unix linkers don't |
| 461 | // allow this kind of backward reference, as they visit each file only once |
| 462 | // from left to right in the command line while resolving all undefined |
| 463 | // symbols at the moment of visiting. |
| 464 | // |
| 465 | // In the above case, since there's no undefined symbol when a linker visits |
| 466 | // foo.a, no files are pulled out from foo.a, and because the linker forgets |
| 467 | // about foo.a after visiting, it can't resolve undefined symbols in bar.o |
| 468 | // that could have been resolved otherwise. |
| 469 | // |
| 470 | // That lld accepts more relaxed form means that (besides it'd make more |
| 471 | // sense) you can accidentally write a command line or a build file that |
| 472 | // works only with lld, even if you have a plan to distribute it to wider |
| 473 | // users who may be using GNU linkers. With --warn-backrefs, you can detect |
| 474 | // a library order that doesn't work with other Unix linkers. |
| 475 | // |
| 476 | // The option is also useful to detect cyclic dependencies between static |
| 477 | // archives. Again, lld accepts |
| 478 | // |
| 479 | // ld.lld foo.a bar.a |
| 480 | // |
| 481 | // even if foo.a and bar.a depend on each other. With --warn-backrefs, it is |
| 482 | // handled as an error. |
| 483 | // |
| 484 | // Here is how the option works. We assign a group ID to each file. A file |
| 485 | // with a smaller group ID can pull out object files from an archive file |
| 486 | // with an equal or greater group ID. Otherwise, it is a reverse dependency |
| 487 | // and an error. |
| 488 | // |
| 489 | // A file outside --{start,end}-group gets a fresh ID when instantiated. All |
| 490 | // files within the same --{start,end}-group get the same group ID. E.g. |
| 491 | // |
| 492 | // ld.lld A B --start-group C D --end-group E |
| 493 | // |
| 494 | // A forms group 0. B form group 1. C and D (including their member object |
| 495 | // files) form group 2. E forms group 3. I think that you can see how this |
| 496 | // group assignment rule simulates the traditional linker's semantics. |
| 497 | bool backref = ctx.arg.warnBackrefs && file->groupId < other.file->groupId; |
| 498 | extract(ctx); |
| 499 | |
| 500 | if (!ctx.arg.whyExtract.empty()) |
| 501 | recordWhyExtract(ctx, reference: other.file, extracted: *file, sym: *this); |
| 502 | |
| 503 | // We don't report backward references to weak symbols as they can be |
| 504 | // overridden later. |
| 505 | // |
| 506 | // A traditional linker does not error for -ldef1 -lref -ldef2 (linking |
| 507 | // sandwich), where def2 may or may not be the same as def1. We don't want |
| 508 | // to warn for this case, so dismiss the warning if we see a subsequent lazy |
| 509 | // definition. this->file needs to be saved because in the case of LTO it |
| 510 | // may be reset to internalFile or be replaced with a file named lto.tmp. |
| 511 | if (backref && !isWeak()) |
| 512 | ctx.backwardReferences.try_emplace(Key: this, |
| 513 | Args: std::make_pair(x: other.file, y&: file)); |
| 514 | return; |
| 515 | } |
| 516 | |
| 517 | // Undefined symbols in a SharedFile do not change the binding. |
| 518 | if (isa<SharedFile>(Val: other.file)) |
| 519 | return; |
| 520 | |
| 521 | if (isUndefined() || isShared()) { |
| 522 | // The binding will be weak if there is at least one reference and all are |
| 523 | // weak. The binding has one opportunity to change to weak: if the first |
| 524 | // reference is weak. |
| 525 | if (other.binding != STB_WEAK || !referenced) |
| 526 | binding = other.binding; |
| 527 | // -u creates a placeholder Undefined (internalFile, STT_NOTYPE). |
| 528 | // Adopt the real file and type from the object file's undefined. |
| 529 | if (file == ctx.internalFile) { |
| 530 | file = other.file; |
| 531 | type = other.type; |
| 532 | } |
| 533 | } |
| 534 | } |
| 535 | |
| 536 | // Compare two symbols. Return true if the new symbol should win. |
| 537 | bool Symbol::shouldReplace(Ctx &ctx, const Defined &other) const { |
| 538 | if (LLVM_UNLIKELY(isCommon())) { |
| 539 | if (ctx.arg.warnCommon) |
| 540 | Warn(ctx) << "common " << getName() << " is overridden" ; |
| 541 | return !other.isWeak(); |
| 542 | } |
| 543 | if (!isDefined()) |
| 544 | return true; |
| 545 | |
| 546 | // Incoming STB_GLOBAL overrides STB_WEAK/STB_GNU_UNIQUE. -fgnu-unique changes |
| 547 | // some vague linkage data in COMDAT from STB_WEAK to STB_GNU_UNIQUE. Treat |
| 548 | // STB_GNU_UNIQUE like STB_WEAK so that we prefer the first among all |
| 549 | // STB_WEAK/STB_GNU_UNIQUE copies. If we prefer an incoming STB_GNU_UNIQUE to |
| 550 | // an existing STB_WEAK, there may be discarded section errors because the |
| 551 | // selected copy may be in a non-prevailing COMDAT. |
| 552 | return !isGlobal() && other.isGlobal(); |
| 553 | } |
| 554 | |
| 555 | void elf::reportDuplicate(Ctx &ctx, const Symbol &sym, const InputFile *newFile, |
| 556 | InputSectionBase *errSec, uint64_t errOffset) { |
| 557 | if (ctx.arg.allowMultipleDefinition) |
| 558 | return; |
| 559 | // In glibc<2.32, crti.o has .gnu.linkonce.t.__x86.get_pc_thunk.bx, which |
| 560 | // is sort of proto-comdat. There is actually no duplicate if we have |
| 561 | // full support for .gnu.linkonce. |
| 562 | const Defined *d = dyn_cast<Defined>(Val: &sym); |
| 563 | if (!d || d->getName() == "__x86.get_pc_thunk.bx" ) |
| 564 | return; |
| 565 | // Allow absolute symbols with the same value for GNU ld compatibility. |
| 566 | if (!d->section && !errSec && errOffset && d->value == errOffset) |
| 567 | return; |
| 568 | if (!d->section || !errSec) { |
| 569 | Err(ctx) << "duplicate symbol: " << &sym << "\n>>> defined in " << sym.file |
| 570 | << "\n>>> defined in " << newFile; |
| 571 | return; |
| 572 | } |
| 573 | |
| 574 | // Construct and print an error message in the form of: |
| 575 | // |
| 576 | // ld.lld: error: duplicate symbol: foo |
| 577 | // >>> defined at bar.c:30 |
| 578 | // >>> bar.o (/home/alice/src/bar.o) |
| 579 | // >>> defined at baz.c:563 |
| 580 | // >>> baz.o in archive libbaz.a |
| 581 | auto *sec1 = cast<InputSectionBase>(Val: d->section); |
| 582 | auto diag = Err(ctx); |
| 583 | diag << "duplicate symbol: " << &sym << "\n>>> defined at " ; |
| 584 | auto tell = diag.tell(); |
| 585 | diag << sec1->getSrcMsg(sym, offset: d->value); |
| 586 | if (tell != diag.tell()) |
| 587 | diag << "\n>>> " ; |
| 588 | diag << sec1->getObjMsg(offset: d->value) << "\n>>> defined at " ; |
| 589 | tell = diag.tell(); |
| 590 | diag << errSec->getSrcMsg(sym, offset: errOffset); |
| 591 | if (tell != diag.tell()) |
| 592 | diag << "\n>>> " ; |
| 593 | diag << errSec->getObjMsg(offset: errOffset); |
| 594 | } |
| 595 | |
| 596 | void Symbol::checkDuplicate(Ctx &ctx, const Defined &other) const { |
| 597 | if (!isWeak() && !other.isWeak()) |
| 598 | reportDuplicate(ctx, sym: *this, newFile: other.file, |
| 599 | errSec: dyn_cast_or_null<InputSectionBase>(Val: other.section), |
| 600 | errOffset: other.value); |
| 601 | } |
| 602 | |
| 603 | void Symbol::resolve(Ctx &ctx, const CommonSymbol &other) { |
| 604 | if (other.visibility() != STV_DEFAULT) { |
| 605 | uint8_t v = visibility(), ov = other.visibility(); |
| 606 | setVisibility(v == STV_DEFAULT ? ov : std::min(a: v, b: ov)); |
| 607 | } |
| 608 | if (isDefined() && !isWeak()) { |
| 609 | if (ctx.arg.warnCommon) |
| 610 | Warn(ctx) << "common " << getName() << " is overridden" ; |
| 611 | return; |
| 612 | } |
| 613 | |
| 614 | if (CommonSymbol *oldSym = dyn_cast<CommonSymbol>(Val: this)) { |
| 615 | if (ctx.arg.warnCommon) |
| 616 | Warn(ctx) << "multiple common of " << getName(); |
| 617 | oldSym->alignment = std::max(a: oldSym->alignment, b: other.alignment); |
| 618 | if (oldSym->size < other.size) { |
| 619 | oldSym->file = other.file; |
| 620 | oldSym->size = other.size; |
| 621 | } |
| 622 | return; |
| 623 | } |
| 624 | |
| 625 | if (auto *s = dyn_cast<SharedSymbol>(Val: this)) { |
| 626 | // Increase st_size if the shared symbol has a larger st_size. The shared |
| 627 | // symbol may be created from common symbols. The fact that some object |
| 628 | // files were linked into a shared object first should not change the |
| 629 | // regular rule that picks the largest st_size. |
| 630 | uint64_t size = s->size; |
| 631 | other.overwrite(sym&: *this); |
| 632 | if (size > cast<CommonSymbol>(Val: this)->size) |
| 633 | cast<CommonSymbol>(Val: this)->size = size; |
| 634 | } else { |
| 635 | other.overwrite(sym&: *this); |
| 636 | } |
| 637 | } |
| 638 | |
| 639 | void Symbol::resolve(Ctx &ctx, const Defined &other) { |
| 640 | if (other.visibility() != STV_DEFAULT) { |
| 641 | uint8_t v = visibility(), ov = other.visibility(); |
| 642 | setVisibility(v == STV_DEFAULT ? ov : std::min(a: v, b: ov)); |
| 643 | } |
| 644 | if (shouldReplace(ctx, other)) |
| 645 | other.overwrite(sym&: *this); |
| 646 | } |
| 647 | |
| 648 | void Symbol::resolve(Ctx &ctx, const LazySymbol &other) { |
| 649 | if (isPlaceholder()) { |
| 650 | other.overwrite(sym&: *this); |
| 651 | return; |
| 652 | } |
| 653 | |
| 654 | if (LLVM_UNLIKELY(!isUndefined())) { |
| 655 | // See the comment in resolve(Ctx &, const Undefined &). |
| 656 | if (isDefined()) { |
| 657 | ctx.backwardReferences.erase(Val: this); |
| 658 | } else if (isCommon() && ctx.arg.fortranCommon && |
| 659 | other.file->shouldExtractForCommon(name: getName())) { |
| 660 | // For common objects, we want to look for global or weak definitions that |
| 661 | // should be extracted as the canonical definition instead. |
| 662 | ctx.backwardReferences.erase(Val: this); |
| 663 | other.overwrite(sym&: *this); |
| 664 | other.extract(ctx); |
| 665 | } |
| 666 | return; |
| 667 | } |
| 668 | |
| 669 | // An undefined weak will not extract archive members. See comment on Lazy in |
| 670 | // Symbols.h for the details. |
| 671 | if (isWeak()) { |
| 672 | uint8_t ty = type; |
| 673 | other.overwrite(sym&: *this); |
| 674 | type = ty; |
| 675 | binding = STB_WEAK; |
| 676 | return; |
| 677 | } |
| 678 | |
| 679 | const InputFile *oldFile = file; |
| 680 | other.extract(ctx); |
| 681 | if (!ctx.arg.whyExtract.empty()) |
| 682 | recordWhyExtract(ctx, reference: oldFile, extracted: *file, sym: *this); |
| 683 | } |
| 684 | |
| 685 | void Symbol::resolve(Ctx &ctx, const SharedSymbol &other) { |
| 686 | isExported = true; |
| 687 | if (isPlaceholder()) { |
| 688 | other.overwrite(sym&: *this); |
| 689 | return; |
| 690 | } |
| 691 | if (isCommon()) { |
| 692 | // See the comment in resolveCommon() above. |
| 693 | if (other.size > cast<CommonSymbol>(Val: this)->size) |
| 694 | cast<CommonSymbol>(Val: this)->size = other.size; |
| 695 | return; |
| 696 | } |
| 697 | if (visibility() == STV_DEFAULT && (isUndefined() || isLazy())) { |
| 698 | // An undefined symbol with non default visibility must be satisfied |
| 699 | // in the same DSO. |
| 700 | uint8_t bind = binding; |
| 701 | other.overwrite(sym&: *this); |
| 702 | binding = bind; |
| 703 | } else if (traced) |
| 704 | printTraceSymbol(sym: other, name: getName()); |
| 705 | } |
| 706 | |
| 707 | void Defined::overwrite(Symbol &sym) const { |
| 708 | if (isa_and_nonnull<SharedFile>(Val: sym.file)) |
| 709 | sym.versionId = VER_NDX_GLOBAL; |
| 710 | Symbol::overwrite(sym, k: DefinedKind); |
| 711 | auto &s = static_cast<Defined &>(sym); |
| 712 | s.value = value; |
| 713 | s.size = size; |
| 714 | s.section = section; |
| 715 | } |
| 716 | |