1 //===- Driver.cpp ---------------------------------------------------------===// 2 // 3 // The LLVM Linker 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // The driver drives the entire linking process. It is responsible for 11 // parsing command line options and doing whatever it is instructed to do. 12 // 13 // One notable thing in the LLD's driver when compared to other linkers is 14 // that the LLD's driver is agnostic on the host operating system. 15 // Other linkers usually have implicit default values (such as a dynamic 16 // linker path or library paths) for each host OS. 17 // 18 // I don't think implicit default values are useful because they are 19 // usually explicitly specified by the compiler driver. They can even 20 // be harmful when you are doing cross-linking. Therefore, in LLD, we 21 // simply trust the compiler driver to pass all required options and 22 // don't try to make effort on our side. 23 // 24 //===----------------------------------------------------------------------===// 25 26 #include "Driver.h" 27 #include "Config.h" 28 #include "Filesystem.h" 29 #include "ICF.h" 30 #include "InputFiles.h" 31 #include "InputSection.h" 32 #include "LinkerScript.h" 33 #include "MarkLive.h" 34 #include "OutputSections.h" 35 #include "ScriptParser.h" 36 #include "SymbolTable.h" 37 #include "Symbols.h" 38 #include "SyntheticSections.h" 39 #include "Target.h" 40 #include "Writer.h" 41 #include "lld/Common/Args.h" 42 #include "lld/Common/Driver.h" 43 #include "lld/Common/ErrorHandler.h" 44 #include "lld/Common/Memory.h" 45 #include "lld/Common/Strings.h" 46 #include "lld/Common/TargetOptionsCommandFlags.h" 47 #include "lld/Common/Threads.h" 48 #include "lld/Common/Version.h" 49 #include "llvm/ADT/SetVector.h" 50 #include "llvm/ADT/StringExtras.h" 51 #include "llvm/ADT/StringSwitch.h" 52 #include "llvm/Support/CommandLine.h" 53 #include "llvm/Support/Compression.h" 54 #include "llvm/Support/LEB128.h" 55 #include "llvm/Support/Path.h" 56 #include "llvm/Support/TarWriter.h" 57 #include "llvm/Support/TargetSelect.h" 58 #include "llvm/Support/raw_ostream.h" 59 #include <cstdlib> 60 #include <utility> 61 62 using namespace llvm; 63 using namespace llvm::ELF; 64 using namespace llvm::object; 65 using namespace llvm::sys; 66 67 using namespace lld; 68 using namespace lld::elf; 69 70 Configuration *elf::Config; 71 LinkerDriver *elf::Driver; 72 73 static void setConfigs(opt::InputArgList &Args); 74 75 bool elf::link(ArrayRef<const char *> Args, bool CanExitEarly, 76 raw_ostream &Error) { 77 errorHandler().LogName = sys::path::filename(Args[0]); 78 errorHandler().ErrorLimitExceededMsg = 79 "too many errors emitted, stopping now (use " 80 "-error-limit=0 to see all errors)"; 81 errorHandler().ErrorOS = &Error; 82 errorHandler().ExitEarly = CanExitEarly; 83 errorHandler().ColorDiagnostics = Error.has_colors(); 84 85 InputSections.clear(); 86 OutputSections.clear(); 87 Tar = nullptr; 88 BinaryFiles.clear(); 89 BitcodeFiles.clear(); 90 ObjectFiles.clear(); 91 SharedFiles.clear(); 92 93 Config = make<Configuration>(); 94 Driver = make<LinkerDriver>(); 95 Script = make<LinkerScript>(); 96 Symtab = make<SymbolTable>(); 97 Config->ProgName = Args[0]; 98 99 Driver->main(Args); 100 101 // Exit immediately if we don't need to return to the caller. 102 // This saves time because the overhead of calling destructors 103 // for all globally-allocated objects is not negligible. 104 if (CanExitEarly) 105 exitLld(errorCount() ? 1 : 0); 106 107 freeArena(); 108 return !errorCount(); 109 } 110 111 // Parses a linker -m option. 112 static std::tuple<ELFKind, uint16_t, uint8_t> parseEmulation(StringRef Emul) { 113 uint8_t OSABI = 0; 114 StringRef S = Emul; 115 if (S.endswith("_fbsd")) { 116 S = S.drop_back(5); 117 OSABI = ELFOSABI_FREEBSD; 118 } 119 120 std::pair<ELFKind, uint16_t> Ret = 121 StringSwitch<std::pair<ELFKind, uint16_t>>(S) 122 .Cases("aarch64elf", "aarch64linux", "aarch64_elf64_le_vec", 123 {ELF64LEKind, EM_AARCH64}) 124 .Cases("armelf", "armelf_linux_eabi", {ELF32LEKind, EM_ARM}) 125 .Case("elf32_x86_64", {ELF32LEKind, EM_X86_64}) 126 .Cases("elf32btsmip", "elf32btsmipn32", {ELF32BEKind, EM_MIPS}) 127 .Cases("elf32ltsmip", "elf32ltsmipn32", {ELF32LEKind, EM_MIPS}) 128 .Case("elf32lriscv", {ELF32LEKind, EM_RISCV}) 129 .Case("elf32ppc", {ELF32BEKind, EM_PPC}) 130 .Case("elf64btsmip", {ELF64BEKind, EM_MIPS}) 131 .Case("elf64ltsmip", {ELF64LEKind, EM_MIPS}) 132 .Case("elf64lriscv", {ELF64LEKind, EM_RISCV}) 133 .Case("elf64ppc", {ELF64BEKind, EM_PPC64}) 134 .Case("elf64lppc", {ELF64LEKind, EM_PPC64}) 135 .Cases("elf_amd64", "elf_x86_64", {ELF64LEKind, EM_X86_64}) 136 .Case("elf_i386", {ELF32LEKind, EM_386}) 137 .Case("elf_iamcu", {ELF32LEKind, EM_IAMCU}) 138 .Default({ELFNoneKind, EM_NONE}); 139 140 if (Ret.first == ELFNoneKind) 141 error("unknown emulation: " + Emul); 142 return std::make_tuple(Ret.first, Ret.second, OSABI); 143 } 144 145 // Returns slices of MB by parsing MB as an archive file. 146 // Each slice consists of a member file in the archive. 147 std::vector<std::pair<MemoryBufferRef, uint64_t>> static getArchiveMembers( 148 MemoryBufferRef MB) { 149 std::unique_ptr<Archive> File = 150 CHECK(Archive::create(MB), 151 MB.getBufferIdentifier() + ": failed to parse archive"); 152 153 std::vector<std::pair<MemoryBufferRef, uint64_t>> V; 154 Error Err = Error::success(); 155 bool AddToTar = File->isThin() && Tar; 156 for (const ErrorOr<Archive::Child> &COrErr : File->children(Err)) { 157 Archive::Child C = 158 CHECK(COrErr, MB.getBufferIdentifier() + 159 ": could not get the child of the archive"); 160 MemoryBufferRef MBRef = 161 CHECK(C.getMemoryBufferRef(), 162 MB.getBufferIdentifier() + 163 ": could not get the buffer for a child of the archive"); 164 if (AddToTar) 165 Tar->append(relativeToRoot(check(C.getFullName())), MBRef.getBuffer()); 166 V.push_back(std::make_pair(MBRef, C.getChildOffset())); 167 } 168 if (Err) 169 fatal(MB.getBufferIdentifier() + ": Archive::children failed: " + 170 toString(std::move(Err))); 171 172 // Take ownership of memory buffers created for members of thin archives. 173 for (std::unique_ptr<MemoryBuffer> &MB : File->takeThinBuffers()) 174 make<std::unique_ptr<MemoryBuffer>>(std::move(MB)); 175 176 return V; 177 } 178 179 // Opens a file and create a file object. Path has to be resolved already. 180 void LinkerDriver::addFile(StringRef Path, bool WithLOption) { 181 using namespace sys::fs; 182 183 Optional<MemoryBufferRef> Buffer = readFile(Path); 184 if (!Buffer.hasValue()) 185 return; 186 MemoryBufferRef MBRef = *Buffer; 187 188 if (Config->FormatBinary) { 189 Files.push_back(make<BinaryFile>(MBRef)); 190 return; 191 } 192 193 switch (identify_magic(MBRef.getBuffer())) { 194 case file_magic::unknown: 195 readLinkerScript(MBRef); 196 return; 197 case file_magic::archive: { 198 // Handle -whole-archive. 199 if (InWholeArchive) { 200 for (const auto &P : getArchiveMembers(MBRef)) 201 Files.push_back(createObjectFile(P.first, Path, P.second)); 202 return; 203 } 204 205 std::unique_ptr<Archive> File = 206 CHECK(Archive::create(MBRef), Path + ": failed to parse archive"); 207 208 // If an archive file has no symbol table, it is likely that a user 209 // is attempting LTO and using a default ar command that doesn't 210 // understand the LLVM bitcode file. It is a pretty common error, so 211 // we'll handle it as if it had a symbol table. 212 if (!File->isEmpty() && !File->hasSymbolTable()) { 213 for (const auto &P : getArchiveMembers(MBRef)) 214 Files.push_back(make<LazyObjFile>(P.first, Path, P.second)); 215 return; 216 } 217 218 // Handle the regular case. 219 Files.push_back(make<ArchiveFile>(std::move(File))); 220 return; 221 } 222 case file_magic::elf_shared_object: 223 if (Config->Relocatable) { 224 error("attempted static link of dynamic object " + Path); 225 return; 226 } 227 228 // DSOs usually have DT_SONAME tags in their ELF headers, and the 229 // sonames are used to identify DSOs. But if they are missing, 230 // they are identified by filenames. We don't know whether the new 231 // file has a DT_SONAME or not because we haven't parsed it yet. 232 // Here, we set the default soname for the file because we might 233 // need it later. 234 // 235 // If a file was specified by -lfoo, the directory part is not 236 // significant, as a user did not specify it. This behavior is 237 // compatible with GNU. 238 Files.push_back( 239 createSharedFile(MBRef, WithLOption ? path::filename(Path) : Path)); 240 return; 241 case file_magic::bitcode: 242 case file_magic::elf_relocatable: 243 if (InLib) 244 Files.push_back(make<LazyObjFile>(MBRef, "", 0)); 245 else 246 Files.push_back(createObjectFile(MBRef)); 247 break; 248 default: 249 error(Path + ": unknown file type"); 250 } 251 } 252 253 // Add a given library by searching it from input search paths. 254 void LinkerDriver::addLibrary(StringRef Name) { 255 if (Optional<std::string> Path = searchLibrary(Name)) 256 addFile(*Path, /*WithLOption=*/true); 257 else 258 error("unable to find library -l" + Name); 259 } 260 261 // This function is called on startup. We need this for LTO since 262 // LTO calls LLVM functions to compile bitcode files to native code. 263 // Technically this can be delayed until we read bitcode files, but 264 // we don't bother to do lazily because the initialization is fast. 265 static void initLLVM() { 266 InitializeAllTargets(); 267 InitializeAllTargetMCs(); 268 InitializeAllAsmPrinters(); 269 InitializeAllAsmParsers(); 270 } 271 272 // Some command line options or some combinations of them are not allowed. 273 // This function checks for such errors. 274 static void checkOptions(opt::InputArgList &Args) { 275 // The MIPS ABI as of 2016 does not support the GNU-style symbol lookup 276 // table which is a relatively new feature. 277 if (Config->EMachine == EM_MIPS && Config->GnuHash) 278 error("the .gnu.hash section is not compatible with the MIPS target."); 279 280 if (Config->FixCortexA53Errata843419 && Config->EMachine != EM_AARCH64) 281 error("--fix-cortex-a53-843419 is only supported on AArch64 targets."); 282 283 if (Config->Pie && Config->Shared) 284 error("-shared and -pie may not be used together"); 285 286 if (!Config->Shared && !Config->FilterList.empty()) 287 error("-F may not be used without -shared"); 288 289 if (!Config->Shared && !Config->AuxiliaryList.empty()) 290 error("-f may not be used without -shared"); 291 292 if (!Config->Relocatable && !Config->DefineCommon) 293 error("-no-define-common not supported in non relocatable output"); 294 295 if (Config->Relocatable) { 296 if (Config->Shared) 297 error("-r and -shared may not be used together"); 298 if (Config->GcSections) 299 error("-r and --gc-sections may not be used together"); 300 if (Config->GdbIndex) 301 error("-r and --gdb-index may not be used together"); 302 if (Config->ICF != ICFLevel::None) 303 error("-r and --icf may not be used together"); 304 if (Config->Pie) 305 error("-r and -pie may not be used together"); 306 } 307 308 if (Config->ExecuteOnly) { 309 if (Config->EMachine != EM_AARCH64) 310 error("-execute-only is only supported on AArch64 targets"); 311 312 if (Config->SingleRoRx && !Script->HasSectionsCommand) 313 error("-execute-only and -no-rosegment cannot be used together"); 314 } 315 } 316 317 static const char *getReproduceOption(opt::InputArgList &Args) { 318 if (auto *Arg = Args.getLastArg(OPT_reproduce)) 319 return Arg->getValue(); 320 return getenv("LLD_REPRODUCE"); 321 } 322 323 static bool hasZOption(opt::InputArgList &Args, StringRef Key) { 324 for (auto *Arg : Args.filtered(OPT_z)) 325 if (Key == Arg->getValue()) 326 return true; 327 return false; 328 } 329 330 static bool getZFlag(opt::InputArgList &Args, StringRef K1, StringRef K2, 331 bool Default) { 332 for (auto *Arg : Args.filtered_reverse(OPT_z)) { 333 if (K1 == Arg->getValue()) 334 return true; 335 if (K2 == Arg->getValue()) 336 return false; 337 } 338 return Default; 339 } 340 341 static bool isKnown(StringRef S) { 342 return S == "combreloc" || S == "copyreloc" || S == "defs" || 343 S == "execstack" || S == "hazardplt" || S == "initfirst" || 344 S == "keep-text-section-prefix" || S == "lazy" || S == "muldefs" || 345 S == "nocombreloc" || S == "nocopyreloc" || S == "nodelete" || 346 S == "nodlopen" || S == "noexecstack" || 347 S == "nokeep-text-section-prefix" || S == "norelro" || S == "notext" || 348 S == "now" || S == "origin" || S == "relro" || S == "retpolineplt" || 349 S == "rodynamic" || S == "text" || S == "wxneeded" || 350 S.startswith("max-page-size=") || S.startswith("stack-size="); 351 } 352 353 // Report an error for an unknown -z option. 354 static void checkZOptions(opt::InputArgList &Args) { 355 for (auto *Arg : Args.filtered(OPT_z)) 356 if (!isKnown(Arg->getValue())) 357 error("unknown -z value: " + StringRef(Arg->getValue())); 358 } 359 360 void LinkerDriver::main(ArrayRef<const char *> ArgsArr) { 361 ELFOptTable Parser; 362 opt::InputArgList Args = Parser.parse(ArgsArr.slice(1)); 363 364 // Interpret this flag early because error() depends on them. 365 errorHandler().ErrorLimit = args::getInteger(Args, OPT_error_limit, 20); 366 367 // Handle -help 368 if (Args.hasArg(OPT_help)) { 369 printHelp(); 370 return; 371 } 372 373 // Handle -v or -version. 374 // 375 // A note about "compatible with GNU linkers" message: this is a hack for 376 // scripts generated by GNU Libtool 2.4.6 (released in February 2014 and 377 // still the newest version in March 2017) or earlier to recognize LLD as 378 // a GNU compatible linker. As long as an output for the -v option 379 // contains "GNU" or "with BFD", they recognize us as GNU-compatible. 380 // 381 // This is somewhat ugly hack, but in reality, we had no choice other 382 // than doing this. Considering the very long release cycle of Libtool, 383 // it is not easy to improve it to recognize LLD as a GNU compatible 384 // linker in a timely manner. Even if we can make it, there are still a 385 // lot of "configure" scripts out there that are generated by old version 386 // of Libtool. We cannot convince every software developer to migrate to 387 // the latest version and re-generate scripts. So we have this hack. 388 if (Args.hasArg(OPT_v) || Args.hasArg(OPT_version)) 389 message(getLLDVersion() + " (compatible with GNU linkers)"); 390 391 // The behavior of -v or --version is a bit strange, but this is 392 // needed for compatibility with GNU linkers. 393 if (Args.hasArg(OPT_v) && !Args.hasArg(OPT_INPUT)) 394 return; 395 if (Args.hasArg(OPT_version)) 396 return; 397 398 if (const char *Path = getReproduceOption(Args)) { 399 // Note that --reproduce is a debug option so you can ignore it 400 // if you are trying to understand the whole picture of the code. 401 Expected<std::unique_ptr<TarWriter>> ErrOrWriter = 402 TarWriter::create(Path, path::stem(Path)); 403 if (ErrOrWriter) { 404 Tar = ErrOrWriter->get(); 405 Tar->append("response.txt", createResponseFile(Args)); 406 Tar->append("version.txt", getLLDVersion() + "\n"); 407 make<std::unique_ptr<TarWriter>>(std::move(*ErrOrWriter)); 408 } else { 409 error(Twine("--reproduce: failed to open ") + Path + ": " + 410 toString(ErrOrWriter.takeError())); 411 } 412 } 413 414 readConfigs(Args); 415 checkZOptions(Args); 416 initLLVM(); 417 createFiles(Args); 418 if (errorCount()) 419 return; 420 421 inferMachineType(); 422 setConfigs(Args); 423 checkOptions(Args); 424 if (errorCount()) 425 return; 426 427 switch (Config->EKind) { 428 case ELF32LEKind: 429 link<ELF32LE>(Args); 430 return; 431 case ELF32BEKind: 432 link<ELF32BE>(Args); 433 return; 434 case ELF64LEKind: 435 link<ELF64LE>(Args); 436 return; 437 case ELF64BEKind: 438 link<ELF64BE>(Args); 439 return; 440 default: 441 llvm_unreachable("unknown Config->EKind"); 442 } 443 } 444 445 static std::string getRpath(opt::InputArgList &Args) { 446 std::vector<StringRef> V = args::getStrings(Args, OPT_rpath); 447 return llvm::join(V.begin(), V.end(), ":"); 448 } 449 450 // Determines what we should do if there are remaining unresolved 451 // symbols after the name resolution. 452 static UnresolvedPolicy getUnresolvedSymbolPolicy(opt::InputArgList &Args) { 453 UnresolvedPolicy ErrorOrWarn = Args.hasFlag(OPT_error_unresolved_symbols, 454 OPT_warn_unresolved_symbols, true) 455 ? UnresolvedPolicy::ReportError 456 : UnresolvedPolicy::Warn; 457 458 // Process the last of -unresolved-symbols, -no-undefined or -z defs. 459 for (auto *Arg : llvm::reverse(Args)) { 460 switch (Arg->getOption().getID()) { 461 case OPT_unresolved_symbols: { 462 StringRef S = Arg->getValue(); 463 if (S == "ignore-all" || S == "ignore-in-object-files") 464 return UnresolvedPolicy::Ignore; 465 if (S == "ignore-in-shared-libs" || S == "report-all") 466 return ErrorOrWarn; 467 error("unknown --unresolved-symbols value: " + S); 468 continue; 469 } 470 case OPT_no_undefined: 471 return ErrorOrWarn; 472 case OPT_z: 473 if (StringRef(Arg->getValue()) == "defs") 474 return ErrorOrWarn; 475 continue; 476 } 477 } 478 479 // -shared implies -unresolved-symbols=ignore-all because missing 480 // symbols are likely to be resolved at runtime using other DSOs. 481 if (Config->Shared) 482 return UnresolvedPolicy::Ignore; 483 return ErrorOrWarn; 484 } 485 486 static Target2Policy getTarget2(opt::InputArgList &Args) { 487 StringRef S = Args.getLastArgValue(OPT_target2, "got-rel"); 488 if (S == "rel") 489 return Target2Policy::Rel; 490 if (S == "abs") 491 return Target2Policy::Abs; 492 if (S == "got-rel") 493 return Target2Policy::GotRel; 494 error("unknown --target2 option: " + S); 495 return Target2Policy::GotRel; 496 } 497 498 static bool isOutputFormatBinary(opt::InputArgList &Args) { 499 StringRef S = Args.getLastArgValue(OPT_oformat, "elf"); 500 if (S == "binary") 501 return true; 502 if (!S.startswith("elf")) 503 error("unknown --oformat value: " + S); 504 return false; 505 } 506 507 static DiscardPolicy getDiscard(opt::InputArgList &Args) { 508 if (Args.hasArg(OPT_relocatable)) 509 return DiscardPolicy::None; 510 511 auto *Arg = 512 Args.getLastArg(OPT_discard_all, OPT_discard_locals, OPT_discard_none); 513 if (!Arg) 514 return DiscardPolicy::Default; 515 if (Arg->getOption().getID() == OPT_discard_all) 516 return DiscardPolicy::All; 517 if (Arg->getOption().getID() == OPT_discard_locals) 518 return DiscardPolicy::Locals; 519 return DiscardPolicy::None; 520 } 521 522 static StringRef getDynamicLinker(opt::InputArgList &Args) { 523 auto *Arg = Args.getLastArg(OPT_dynamic_linker, OPT_no_dynamic_linker); 524 if (!Arg || Arg->getOption().getID() == OPT_no_dynamic_linker) 525 return ""; 526 return Arg->getValue(); 527 } 528 529 static ICFLevel getICF(opt::InputArgList &Args) { 530 auto *Arg = Args.getLastArg(OPT_icf_none, OPT_icf_safe, OPT_icf_all); 531 if (!Arg || Arg->getOption().getID() == OPT_icf_none) 532 return ICFLevel::None; 533 if (Arg->getOption().getID() == OPT_icf_safe) 534 return ICFLevel::Safe; 535 return ICFLevel::All; 536 } 537 538 static StripPolicy getStrip(opt::InputArgList &Args) { 539 if (Args.hasArg(OPT_relocatable)) 540 return StripPolicy::None; 541 542 auto *Arg = Args.getLastArg(OPT_strip_all, OPT_strip_debug); 543 if (!Arg) 544 return StripPolicy::None; 545 if (Arg->getOption().getID() == OPT_strip_all) 546 return StripPolicy::All; 547 return StripPolicy::Debug; 548 } 549 550 static uint64_t parseSectionAddress(StringRef S, const opt::Arg &Arg) { 551 uint64_t VA = 0; 552 if (S.startswith("0x")) 553 S = S.drop_front(2); 554 if (!to_integer(S, VA, 16)) 555 error("invalid argument: " + toString(Arg)); 556 return VA; 557 } 558 559 static StringMap<uint64_t> getSectionStartMap(opt::InputArgList &Args) { 560 StringMap<uint64_t> Ret; 561 for (auto *Arg : Args.filtered(OPT_section_start)) { 562 StringRef Name; 563 StringRef Addr; 564 std::tie(Name, Addr) = StringRef(Arg->getValue()).split('='); 565 Ret[Name] = parseSectionAddress(Addr, *Arg); 566 } 567 568 if (auto *Arg = Args.getLastArg(OPT_Ttext)) 569 Ret[".text"] = parseSectionAddress(Arg->getValue(), *Arg); 570 if (auto *Arg = Args.getLastArg(OPT_Tdata)) 571 Ret[".data"] = parseSectionAddress(Arg->getValue(), *Arg); 572 if (auto *Arg = Args.getLastArg(OPT_Tbss)) 573 Ret[".bss"] = parseSectionAddress(Arg->getValue(), *Arg); 574 return Ret; 575 } 576 577 static SortSectionPolicy getSortSection(opt::InputArgList &Args) { 578 StringRef S = Args.getLastArgValue(OPT_sort_section); 579 if (S == "alignment") 580 return SortSectionPolicy::Alignment; 581 if (S == "name") 582 return SortSectionPolicy::Name; 583 if (!S.empty()) 584 error("unknown --sort-section rule: " + S); 585 return SortSectionPolicy::Default; 586 } 587 588 static OrphanHandlingPolicy getOrphanHandling(opt::InputArgList &Args) { 589 StringRef S = Args.getLastArgValue(OPT_orphan_handling, "place"); 590 if (S == "warn") 591 return OrphanHandlingPolicy::Warn; 592 if (S == "error") 593 return OrphanHandlingPolicy::Error; 594 if (S != "place") 595 error("unknown --orphan-handling mode: " + S); 596 return OrphanHandlingPolicy::Place; 597 } 598 599 // Parse --build-id or --build-id=<style>. We handle "tree" as a 600 // synonym for "sha1" because all our hash functions including 601 // -build-id=sha1 are actually tree hashes for performance reasons. 602 static std::pair<BuildIdKind, std::vector<uint8_t>> 603 getBuildId(opt::InputArgList &Args) { 604 auto *Arg = Args.getLastArg(OPT_build_id, OPT_build_id_eq); 605 if (!Arg) 606 return {BuildIdKind::None, {}}; 607 608 if (Arg->getOption().getID() == OPT_build_id) 609 return {BuildIdKind::Fast, {}}; 610 611 StringRef S = Arg->getValue(); 612 if (S == "fast") 613 return {BuildIdKind::Fast, {}}; 614 if (S == "md5") 615 return {BuildIdKind::Md5, {}}; 616 if (S == "sha1" || S == "tree") 617 return {BuildIdKind::Sha1, {}}; 618 if (S == "uuid") 619 return {BuildIdKind::Uuid, {}}; 620 if (S.startswith("0x")) 621 return {BuildIdKind::Hexstring, parseHex(S.substr(2))}; 622 623 if (S != "none") 624 error("unknown --build-id style: " + S); 625 return {BuildIdKind::None, {}}; 626 } 627 628 static std::pair<bool, bool> getPackDynRelocs(opt::InputArgList &Args) { 629 StringRef S = Args.getLastArgValue(OPT_pack_dyn_relocs, "none"); 630 if (S == "android") 631 return {true, false}; 632 if (S == "relr") 633 return {false, true}; 634 if (S == "android+relr") 635 return {true, true}; 636 637 if (S != "none") 638 error("unknown -pack-dyn-relocs format: " + S); 639 return {false, false}; 640 } 641 642 static void readCallGraph(MemoryBufferRef MB) { 643 // Build a map from symbol name to section 644 DenseMap<StringRef, const Symbol *> SymbolNameToSymbol; 645 for (InputFile *File : ObjectFiles) 646 for (Symbol *Sym : File->getSymbols()) 647 SymbolNameToSymbol[Sym->getName()] = Sym; 648 649 auto FindSection = [&](StringRef SymName) -> InputSectionBase * { 650 const Symbol *Sym = SymbolNameToSymbol.lookup(SymName); 651 if (Sym) 652 warnUnorderableSymbol(Sym); 653 else if (Config->WarnSymbolOrdering) 654 warn(MB.getBufferIdentifier() + ": no such symbol: " + SymName); 655 656 if (const Defined *DR = dyn_cast_or_null<Defined>(Sym)) 657 return dyn_cast_or_null<InputSectionBase>(DR->Section); 658 return nullptr; 659 }; 660 661 for (StringRef L : args::getLines(MB)) { 662 SmallVector<StringRef, 3> Fields; 663 L.split(Fields, ' '); 664 uint64_t Count; 665 if (Fields.size() != 3 || !to_integer(Fields[2], Count)) 666 fatal(MB.getBufferIdentifier() + ": parse error"); 667 668 if (const InputSectionBase *FromSB = FindSection(Fields[0])) 669 if (const InputSectionBase *ToSB = FindSection(Fields[1])) 670 Config->CallGraphProfile[std::make_pair(FromSB, ToSB)] += Count; 671 } 672 } 673 674 static bool getCompressDebugSections(opt::InputArgList &Args) { 675 StringRef S = Args.getLastArgValue(OPT_compress_debug_sections, "none"); 676 if (S == "none") 677 return false; 678 if (S != "zlib") 679 error("unknown --compress-debug-sections value: " + S); 680 if (!zlib::isAvailable()) 681 error("--compress-debug-sections: zlib is not available"); 682 return true; 683 } 684 685 static std::pair<StringRef, StringRef> getOldNewOptions(opt::InputArgList &Args, 686 unsigned Id) { 687 auto *Arg = Args.getLastArg(Id); 688 if (!Arg) 689 return {"", ""}; 690 691 StringRef S = Arg->getValue(); 692 std::pair<StringRef, StringRef> Ret = S.split(';'); 693 if (Ret.second.empty()) 694 error(Arg->getSpelling() + " expects 'old;new' format, but got " + S); 695 return Ret; 696 } 697 698 // Parse the symbol ordering file and warn for any duplicate entries. 699 static std::vector<StringRef> getSymbolOrderingFile(MemoryBufferRef MB) { 700 SetVector<StringRef> Names; 701 for (StringRef S : args::getLines(MB)) 702 if (!Names.insert(S) && Config->WarnSymbolOrdering) 703 warn(MB.getBufferIdentifier() + ": duplicate ordered symbol: " + S); 704 705 return Names.takeVector(); 706 } 707 708 static void parseClangOption(StringRef Opt, const Twine &Msg) { 709 std::string Err; 710 raw_string_ostream OS(Err); 711 712 const char *Argv[] = {Config->ProgName.data(), Opt.data()}; 713 if (cl::ParseCommandLineOptions(2, Argv, "", &OS)) 714 return; 715 OS.flush(); 716 error(Msg + ": " + StringRef(Err).trim()); 717 } 718 719 // Initializes Config members by the command line options. 720 void LinkerDriver::readConfigs(opt::InputArgList &Args) { 721 errorHandler().Verbose = Args.hasArg(OPT_verbose); 722 errorHandler().FatalWarnings = 723 Args.hasFlag(OPT_fatal_warnings, OPT_no_fatal_warnings, false); 724 ThreadsEnabled = Args.hasFlag(OPT_threads, OPT_no_threads, true); 725 726 Config->AllowMultipleDefinition = 727 Args.hasFlag(OPT_allow_multiple_definition, 728 OPT_no_allow_multiple_definition, false) || 729 hasZOption(Args, "muldefs"); 730 Config->AuxiliaryList = args::getStrings(Args, OPT_auxiliary); 731 Config->Bsymbolic = Args.hasArg(OPT_Bsymbolic); 732 Config->BsymbolicFunctions = Args.hasArg(OPT_Bsymbolic_functions); 733 Config->CheckSections = 734 Args.hasFlag(OPT_check_sections, OPT_no_check_sections, true); 735 Config->Chroot = Args.getLastArgValue(OPT_chroot); 736 Config->CompressDebugSections = getCompressDebugSections(Args); 737 Config->Cref = Args.hasFlag(OPT_cref, OPT_no_cref, false); 738 Config->DefineCommon = Args.hasFlag(OPT_define_common, OPT_no_define_common, 739 !Args.hasArg(OPT_relocatable)); 740 Config->Demangle = Args.hasFlag(OPT_demangle, OPT_no_demangle, true); 741 Config->DisableVerify = Args.hasArg(OPT_disable_verify); 742 Config->Discard = getDiscard(Args); 743 Config->DwoDir = Args.getLastArgValue(OPT_plugin_opt_dwo_dir_eq); 744 Config->DynamicLinker = getDynamicLinker(Args); 745 Config->EhFrameHdr = 746 Args.hasFlag(OPT_eh_frame_hdr, OPT_no_eh_frame_hdr, false); 747 Config->EmitRelocs = Args.hasArg(OPT_emit_relocs); 748 Config->EnableNewDtags = 749 Args.hasFlag(OPT_enable_new_dtags, OPT_disable_new_dtags, true); 750 Config->Entry = Args.getLastArgValue(OPT_entry); 751 Config->ExecuteOnly = 752 Args.hasFlag(OPT_execute_only, OPT_no_execute_only, false); 753 Config->ExportDynamic = 754 Args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false); 755 Config->FilterList = args::getStrings(Args, OPT_filter); 756 Config->Fini = Args.getLastArgValue(OPT_fini, "_fini"); 757 Config->FixCortexA53Errata843419 = Args.hasArg(OPT_fix_cortex_a53_843419); 758 Config->GcSections = Args.hasFlag(OPT_gc_sections, OPT_no_gc_sections, false); 759 Config->GnuUnique = Args.hasFlag(OPT_gnu_unique, OPT_no_gnu_unique, true); 760 Config->GdbIndex = Args.hasFlag(OPT_gdb_index, OPT_no_gdb_index, false); 761 Config->ICF = getICF(Args); 762 Config->IgnoreDataAddressEquality = 763 Args.hasArg(OPT_ignore_data_address_equality); 764 Config->IgnoreFunctionAddressEquality = 765 Args.hasArg(OPT_ignore_function_address_equality); 766 Config->Init = Args.getLastArgValue(OPT_init, "_init"); 767 Config->LTOAAPipeline = Args.getLastArgValue(OPT_lto_aa_pipeline); 768 Config->LTODebugPassManager = Args.hasArg(OPT_lto_debug_pass_manager); 769 Config->LTONewPassManager = Args.hasArg(OPT_lto_new_pass_manager); 770 Config->LTONewPmPasses = Args.getLastArgValue(OPT_lto_newpm_passes); 771 Config->LTOO = args::getInteger(Args, OPT_lto_O, 2); 772 Config->LTOObjPath = Args.getLastArgValue(OPT_plugin_opt_obj_path_eq); 773 Config->LTOPartitions = args::getInteger(Args, OPT_lto_partitions, 1); 774 Config->LTOSampleProfile = Args.getLastArgValue(OPT_lto_sample_profile); 775 Config->MapFile = Args.getLastArgValue(OPT_Map); 776 Config->MipsGotSize = args::getInteger(Args, OPT_mips_got_size, 0xfff0); 777 Config->MergeArmExidx = 778 Args.hasFlag(OPT_merge_exidx_entries, OPT_no_merge_exidx_entries, true); 779 Config->NoinhibitExec = Args.hasArg(OPT_noinhibit_exec); 780 Config->Nostdlib = Args.hasArg(OPT_nostdlib); 781 Config->OFormatBinary = isOutputFormatBinary(Args); 782 Config->Omagic = Args.hasFlag(OPT_omagic, OPT_no_omagic, false); 783 Config->OptRemarksFilename = Args.getLastArgValue(OPT_opt_remarks_filename); 784 Config->OptRemarksWithHotness = Args.hasArg(OPT_opt_remarks_with_hotness); 785 Config->Optimize = args::getInteger(Args, OPT_O, 1); 786 Config->OrphanHandling = getOrphanHandling(Args); 787 Config->OutputFile = Args.getLastArgValue(OPT_o); 788 Config->Pie = Args.hasFlag(OPT_pie, OPT_no_pie, false); 789 Config->PrintIcfSections = 790 Args.hasFlag(OPT_print_icf_sections, OPT_no_print_icf_sections, false); 791 Config->PrintGcSections = 792 Args.hasFlag(OPT_print_gc_sections, OPT_no_print_gc_sections, false); 793 Config->Rpath = getRpath(Args); 794 Config->Relocatable = Args.hasArg(OPT_relocatable); 795 Config->SaveTemps = Args.hasArg(OPT_save_temps); 796 Config->SearchPaths = args::getStrings(Args, OPT_library_path); 797 Config->SectionStartMap = getSectionStartMap(Args); 798 Config->Shared = Args.hasArg(OPT_shared); 799 Config->SingleRoRx = Args.hasArg(OPT_no_rosegment); 800 Config->SoName = Args.getLastArgValue(OPT_soname); 801 Config->SortSection = getSortSection(Args); 802 Config->Strip = getStrip(Args); 803 Config->Sysroot = Args.getLastArgValue(OPT_sysroot); 804 Config->Target1Rel = Args.hasFlag(OPT_target1_rel, OPT_target1_abs, false); 805 Config->Target2 = getTarget2(Args); 806 Config->ThinLTOCacheDir = Args.getLastArgValue(OPT_thinlto_cache_dir); 807 Config->ThinLTOCachePolicy = CHECK( 808 parseCachePruningPolicy(Args.getLastArgValue(OPT_thinlto_cache_policy)), 809 "--thinlto-cache-policy: invalid cache policy"); 810 Config->ThinLTOEmitImportsFiles = 811 Args.hasArg(OPT_plugin_opt_thinlto_emit_imports_files); 812 Config->ThinLTOIndexOnly = Args.hasArg(OPT_plugin_opt_thinlto_index_only) || 813 Args.hasArg(OPT_plugin_opt_thinlto_index_only_eq); 814 Config->ThinLTOIndexOnlyArg = 815 Args.getLastArgValue(OPT_plugin_opt_thinlto_index_only_eq); 816 Config->ThinLTOJobs = args::getInteger(Args, OPT_thinlto_jobs, -1u); 817 Config->ThinLTOObjectSuffixReplace = 818 getOldNewOptions(Args, OPT_plugin_opt_thinlto_object_suffix_replace_eq); 819 Config->ThinLTOPrefixReplace = 820 getOldNewOptions(Args, OPT_plugin_opt_thinlto_prefix_replace_eq); 821 Config->Trace = Args.hasArg(OPT_trace); 822 Config->Undefined = args::getStrings(Args, OPT_undefined); 823 Config->UndefinedVersion = 824 Args.hasFlag(OPT_undefined_version, OPT_no_undefined_version, true); 825 Config->UseAndroidRelrTags = Args.hasFlag( 826 OPT_use_android_relr_tags, OPT_no_use_android_relr_tags, false); 827 Config->UnresolvedSymbols = getUnresolvedSymbolPolicy(Args); 828 Config->WarnBackrefs = 829 Args.hasFlag(OPT_warn_backrefs, OPT_no_warn_backrefs, false); 830 Config->WarnCommon = Args.hasFlag(OPT_warn_common, OPT_no_warn_common, false); 831 Config->WarnSymbolOrdering = 832 Args.hasFlag(OPT_warn_symbol_ordering, OPT_no_warn_symbol_ordering, true); 833 Config->ZCombreloc = getZFlag(Args, "combreloc", "nocombreloc", true); 834 Config->ZCopyreloc = getZFlag(Args, "copyreloc", "nocopyreloc", true); 835 Config->ZExecstack = getZFlag(Args, "execstack", "noexecstack", false); 836 Config->ZHazardplt = hasZOption(Args, "hazardplt"); 837 Config->ZInitfirst = hasZOption(Args, "initfirst"); 838 Config->ZKeepTextSectionPrefix = getZFlag( 839 Args, "keep-text-section-prefix", "nokeep-text-section-prefix", false); 840 Config->ZNodelete = hasZOption(Args, "nodelete"); 841 Config->ZNodlopen = hasZOption(Args, "nodlopen"); 842 Config->ZNow = getZFlag(Args, "now", "lazy", false); 843 Config->ZOrigin = hasZOption(Args, "origin"); 844 Config->ZRelro = getZFlag(Args, "relro", "norelro", true); 845 Config->ZRetpolineplt = hasZOption(Args, "retpolineplt"); 846 Config->ZRodynamic = hasZOption(Args, "rodynamic"); 847 Config->ZStackSize = args::getZOptionValue(Args, OPT_z, "stack-size", 0); 848 Config->ZText = getZFlag(Args, "text", "notext", true); 849 Config->ZWxneeded = hasZOption(Args, "wxneeded"); 850 851 // Parse LTO options. 852 if (auto *Arg = Args.getLastArg(OPT_plugin_opt_mcpu_eq)) 853 parseClangOption(Saver.save("-mcpu=" + StringRef(Arg->getValue())), 854 Arg->getSpelling()); 855 856 for (auto *Arg : Args.filtered(OPT_plugin_opt)) 857 parseClangOption(Arg->getValue(), Arg->getSpelling()); 858 859 // Parse -mllvm options. 860 for (auto *Arg : Args.filtered(OPT_mllvm)) 861 parseClangOption(Arg->getValue(), Arg->getSpelling()); 862 863 if (Config->LTOO > 3) 864 error("invalid optimization level for LTO: " + Twine(Config->LTOO)); 865 if (Config->LTOPartitions == 0) 866 error("--lto-partitions: number of threads must be > 0"); 867 if (Config->ThinLTOJobs == 0) 868 error("--thinlto-jobs: number of threads must be > 0"); 869 870 // Parse ELF{32,64}{LE,BE} and CPU type. 871 if (auto *Arg = Args.getLastArg(OPT_m)) { 872 StringRef S = Arg->getValue(); 873 std::tie(Config->EKind, Config->EMachine, Config->OSABI) = 874 parseEmulation(S); 875 Config->MipsN32Abi = (S == "elf32btsmipn32" || S == "elf32ltsmipn32"); 876 Config->Emulation = S; 877 } 878 879 // Parse -hash-style={sysv,gnu,both}. 880 if (auto *Arg = Args.getLastArg(OPT_hash_style)) { 881 StringRef S = Arg->getValue(); 882 if (S == "sysv") 883 Config->SysvHash = true; 884 else if (S == "gnu") 885 Config->GnuHash = true; 886 else if (S == "both") 887 Config->SysvHash = Config->GnuHash = true; 888 else 889 error("unknown -hash-style: " + S); 890 } 891 892 if (Args.hasArg(OPT_print_map)) 893 Config->MapFile = "-"; 894 895 // --omagic is an option to create old-fashioned executables in which 896 // .text segments are writable. Today, the option is still in use to 897 // create special-purpose programs such as boot loaders. It doesn't 898 // make sense to create PT_GNU_RELRO for such executables. 899 if (Config->Omagic) 900 Config->ZRelro = false; 901 902 std::tie(Config->BuildId, Config->BuildIdVector) = getBuildId(Args); 903 904 std::tie(Config->AndroidPackDynRelocs, Config->RelrPackDynRelocs) = 905 getPackDynRelocs(Args); 906 907 if (auto *Arg = Args.getLastArg(OPT_symbol_ordering_file)) 908 if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue())) 909 Config->SymbolOrderingFile = getSymbolOrderingFile(*Buffer); 910 911 // If --retain-symbol-file is used, we'll keep only the symbols listed in 912 // the file and discard all others. 913 if (auto *Arg = Args.getLastArg(OPT_retain_symbols_file)) { 914 Config->DefaultSymbolVersion = VER_NDX_LOCAL; 915 if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue())) 916 for (StringRef S : args::getLines(*Buffer)) 917 Config->VersionScriptGlobals.push_back( 918 {S, /*IsExternCpp*/ false, /*HasWildcard*/ false}); 919 } 920 921 bool HasExportDynamic = 922 Args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false); 923 924 // Parses -dynamic-list and -export-dynamic-symbol. They make some 925 // symbols private. Note that -export-dynamic takes precedence over them 926 // as it says all symbols should be exported. 927 if (!HasExportDynamic) { 928 for (auto *Arg : Args.filtered(OPT_dynamic_list)) 929 if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue())) 930 readDynamicList(*Buffer); 931 932 for (auto *Arg : Args.filtered(OPT_export_dynamic_symbol)) 933 Config->DynamicList.push_back( 934 {Arg->getValue(), /*IsExternCpp*/ false, /*HasWildcard*/ false}); 935 } 936 937 // If --export-dynamic-symbol=foo is given and symbol foo is defined in 938 // an object file in an archive file, that object file should be pulled 939 // out and linked. (It doesn't have to behave like that from technical 940 // point of view, but this is needed for compatibility with GNU.) 941 for (auto *Arg : Args.filtered(OPT_export_dynamic_symbol)) 942 Config->Undefined.push_back(Arg->getValue()); 943 944 for (auto *Arg : Args.filtered(OPT_version_script)) 945 if (Optional<std::string> Path = searchScript(Arg->getValue())) { 946 if (Optional<MemoryBufferRef> Buffer = readFile(*Path)) 947 readVersionScript(*Buffer); 948 } else { 949 error(Twine("cannot find version script ") + Arg->getValue()); 950 } 951 } 952 953 // Some Config members do not directly correspond to any particular 954 // command line options, but computed based on other Config values. 955 // This function initialize such members. See Config.h for the details 956 // of these values. 957 static void setConfigs(opt::InputArgList &Args) { 958 ELFKind Kind = Config->EKind; 959 uint16_t Machine = Config->EMachine; 960 961 Config->CopyRelocs = (Config->Relocatable || Config->EmitRelocs); 962 Config->Is64 = (Kind == ELF64LEKind || Kind == ELF64BEKind); 963 Config->IsLE = (Kind == ELF32LEKind || Kind == ELF64LEKind); 964 Config->Endianness = 965 Config->IsLE ? support::endianness::little : support::endianness::big; 966 Config->IsMips64EL = (Kind == ELF64LEKind && Machine == EM_MIPS); 967 Config->Pic = Config->Pie || Config->Shared; 968 Config->Wordsize = Config->Is64 ? 8 : 4; 969 970 // There is an ILP32 ABI for x86-64, although it's not very popular. 971 // It is called the x32 ABI. 972 bool IsX32 = (Kind == ELF32LEKind && Machine == EM_X86_64); 973 974 // ELF defines two different ways to store relocation addends as shown below: 975 // 976 // Rel: Addends are stored to the location where relocations are applied. 977 // Rela: Addends are stored as part of relocation entry. 978 // 979 // In other words, Rela makes it easy to read addends at the price of extra 980 // 4 or 8 byte for each relocation entry. We don't know why ELF defined two 981 // different mechanisms in the first place, but this is how the spec is 982 // defined. 983 // 984 // You cannot choose which one, Rel or Rela, you want to use. Instead each 985 // ABI defines which one you need to use. The following expression expresses 986 // that. 987 Config->IsRela = 988 (Config->Is64 || IsX32 || Machine == EM_PPC || Machine == EM_RISCV) && 989 Machine != EM_MIPS; 990 991 // If the output uses REL relocations we must store the dynamic relocation 992 // addends to the output sections. We also store addends for RELA relocations 993 // if --apply-dynamic-relocs is used. 994 // We default to not writing the addends when using RELA relocations since 995 // any standard conforming tool can find it in r_addend. 996 Config->WriteAddends = Args.hasFlag(OPT_apply_dynamic_relocs, 997 OPT_no_apply_dynamic_relocs, false) || 998 !Config->IsRela; 999 } 1000 1001 // Returns a value of "-format" option. 1002 static bool isFormatBinary(StringRef S) { 1003 if (S == "binary") 1004 return true; 1005 if (S == "elf" || S == "default") 1006 return false; 1007 error("unknown -format value: " + S + 1008 " (supported formats: elf, default, binary)"); 1009 return false; 1010 } 1011 1012 void LinkerDriver::createFiles(opt::InputArgList &Args) { 1013 // For --{push,pop}-state. 1014 std::vector<std::tuple<bool, bool, bool>> Stack; 1015 1016 // Iterate over argv to process input files and positional arguments. 1017 for (auto *Arg : Args) { 1018 switch (Arg->getOption().getUnaliasedOption().getID()) { 1019 case OPT_library: 1020 addLibrary(Arg->getValue()); 1021 break; 1022 case OPT_INPUT: 1023 addFile(Arg->getValue(), /*WithLOption=*/false); 1024 break; 1025 case OPT_defsym: { 1026 StringRef From; 1027 StringRef To; 1028 std::tie(From, To) = StringRef(Arg->getValue()).split('='); 1029 if (From.empty() || To.empty()) 1030 error("-defsym: syntax error: " + StringRef(Arg->getValue())); 1031 else 1032 readDefsym(From, MemoryBufferRef(To, "-defsym")); 1033 break; 1034 } 1035 case OPT_script: 1036 if (Optional<std::string> Path = searchScript(Arg->getValue())) { 1037 if (Optional<MemoryBufferRef> MB = readFile(*Path)) 1038 readLinkerScript(*MB); 1039 break; 1040 } 1041 error(Twine("cannot find linker script ") + Arg->getValue()); 1042 break; 1043 case OPT_as_needed: 1044 Config->AsNeeded = true; 1045 break; 1046 case OPT_format: 1047 Config->FormatBinary = isFormatBinary(Arg->getValue()); 1048 break; 1049 case OPT_no_as_needed: 1050 Config->AsNeeded = false; 1051 break; 1052 case OPT_Bstatic: 1053 Config->Static = true; 1054 break; 1055 case OPT_Bdynamic: 1056 Config->Static = false; 1057 break; 1058 case OPT_whole_archive: 1059 InWholeArchive = true; 1060 break; 1061 case OPT_no_whole_archive: 1062 InWholeArchive = false; 1063 break; 1064 case OPT_just_symbols: 1065 if (Optional<MemoryBufferRef> MB = readFile(Arg->getValue())) { 1066 Files.push_back(createObjectFile(*MB)); 1067 Files.back()->JustSymbols = true; 1068 } 1069 break; 1070 case OPT_start_group: 1071 if (InputFile::IsInGroup) 1072 error("nested --start-group"); 1073 InputFile::IsInGroup = true; 1074 break; 1075 case OPT_end_group: 1076 if (!InputFile::IsInGroup) 1077 error("stray --end-group"); 1078 InputFile::IsInGroup = false; 1079 ++InputFile::NextGroupId; 1080 break; 1081 case OPT_start_lib: 1082 if (InLib) 1083 error("nested --start-lib"); 1084 if (InputFile::IsInGroup) 1085 error("may not nest --start-lib in --start-group"); 1086 InLib = true; 1087 InputFile::IsInGroup = true; 1088 break; 1089 case OPT_end_lib: 1090 if (!InLib) 1091 error("stray --end-lib"); 1092 InLib = false; 1093 InputFile::IsInGroup = false; 1094 ++InputFile::NextGroupId; 1095 break; 1096 case OPT_push_state: 1097 Stack.emplace_back(Config->AsNeeded, Config->Static, InWholeArchive); 1098 break; 1099 case OPT_pop_state: 1100 if (Stack.empty()) { 1101 error("unbalanced --push-state/--pop-state"); 1102 break; 1103 } 1104 std::tie(Config->AsNeeded, Config->Static, InWholeArchive) = Stack.back(); 1105 Stack.pop_back(); 1106 break; 1107 } 1108 } 1109 1110 if (Files.empty() && errorCount() == 0) 1111 error("no input files"); 1112 } 1113 1114 // If -m <machine_type> was not given, infer it from object files. 1115 void LinkerDriver::inferMachineType() { 1116 if (Config->EKind != ELFNoneKind) 1117 return; 1118 1119 for (InputFile *F : Files) { 1120 if (F->EKind == ELFNoneKind) 1121 continue; 1122 Config->EKind = F->EKind; 1123 Config->EMachine = F->EMachine; 1124 Config->OSABI = F->OSABI; 1125 Config->MipsN32Abi = Config->EMachine == EM_MIPS && isMipsN32Abi(F); 1126 return; 1127 } 1128 error("target emulation unknown: -m or at least one .o file required"); 1129 } 1130 1131 // Parse -z max-page-size=<value>. The default value is defined by 1132 // each target. 1133 static uint64_t getMaxPageSize(opt::InputArgList &Args) { 1134 uint64_t Val = args::getZOptionValue(Args, OPT_z, "max-page-size", 1135 Target->DefaultMaxPageSize); 1136 if (!isPowerOf2_64(Val)) 1137 error("max-page-size: value isn't a power of 2"); 1138 return Val; 1139 } 1140 1141 // Parses -image-base option. 1142 static Optional<uint64_t> getImageBase(opt::InputArgList &Args) { 1143 // Because we are using "Config->MaxPageSize" here, this function has to be 1144 // called after the variable is initialized. 1145 auto *Arg = Args.getLastArg(OPT_image_base); 1146 if (!Arg) 1147 return None; 1148 1149 StringRef S = Arg->getValue(); 1150 uint64_t V; 1151 if (!to_integer(S, V)) { 1152 error("-image-base: number expected, but got " + S); 1153 return 0; 1154 } 1155 if ((V % Config->MaxPageSize) != 0) 1156 warn("-image-base: address isn't multiple of page size: " + S); 1157 return V; 1158 } 1159 1160 // Parses `--exclude-libs=lib,lib,...`. 1161 // The library names may be delimited by commas or colons. 1162 static DenseSet<StringRef> getExcludeLibs(opt::InputArgList &Args) { 1163 DenseSet<StringRef> Ret; 1164 for (auto *Arg : Args.filtered(OPT_exclude_libs)) { 1165 StringRef S = Arg->getValue(); 1166 for (;;) { 1167 size_t Pos = S.find_first_of(",:"); 1168 if (Pos == StringRef::npos) 1169 break; 1170 Ret.insert(S.substr(0, Pos)); 1171 S = S.substr(Pos + 1); 1172 } 1173 Ret.insert(S); 1174 } 1175 return Ret; 1176 } 1177 1178 // Handles the -exclude-libs option. If a static library file is specified 1179 // by the -exclude-libs option, all public symbols from the archive become 1180 // private unless otherwise specified by version scripts or something. 1181 // A special library name "ALL" means all archive files. 1182 // 1183 // This is not a popular option, but some programs such as bionic libc use it. 1184 template <class ELFT> 1185 static void excludeLibs(opt::InputArgList &Args) { 1186 DenseSet<StringRef> Libs = getExcludeLibs(Args); 1187 bool All = Libs.count("ALL"); 1188 1189 auto Visit = [&](InputFile *File) { 1190 if (!File->ArchiveName.empty()) 1191 if (All || Libs.count(path::filename(File->ArchiveName))) 1192 for (Symbol *Sym : File->getSymbols()) 1193 if (!Sym->isLocal() && Sym->File == File) 1194 Sym->VersionId = VER_NDX_LOCAL; 1195 }; 1196 1197 for (InputFile *File : ObjectFiles) 1198 Visit(File); 1199 1200 for (BitcodeFile *File : BitcodeFiles) 1201 Visit(File); 1202 } 1203 1204 // Force Sym to be entered in the output. Used for -u or equivalent. 1205 template <class ELFT> static void handleUndefined(StringRef Name) { 1206 Symbol *Sym = Symtab->find(Name); 1207 if (!Sym) 1208 return; 1209 1210 // Since symbol S may not be used inside the program, LTO may 1211 // eliminate it. Mark the symbol as "used" to prevent it. 1212 Sym->IsUsedInRegularObj = true; 1213 1214 if (Sym->isLazy()) 1215 Symtab->fetchLazy<ELFT>(Sym); 1216 } 1217 1218 template <class ELFT> static void handleLibcall(StringRef Name) { 1219 Symbol *Sym = Symtab->find(Name); 1220 if (!Sym || !Sym->isLazy()) 1221 return; 1222 1223 MemoryBufferRef MB; 1224 if (auto *LO = dyn_cast<LazyObject>(Sym)) 1225 MB = LO->File->MB; 1226 else 1227 MB = cast<LazyArchive>(Sym)->getMemberBuffer(); 1228 1229 if (isBitcode(MB)) 1230 Symtab->fetchLazy<ELFT>(Sym); 1231 } 1232 1233 template <class ELFT> static bool shouldDemote(Symbol &Sym) { 1234 // If all references to a DSO happen to be weak, the DSO is not added to 1235 // DT_NEEDED. If that happens, we need to eliminate shared symbols created 1236 // from the DSO. Otherwise, they become dangling references that point to a 1237 // non-existent DSO. 1238 if (auto *S = dyn_cast<SharedSymbol>(&Sym)) 1239 return !S->getFile<ELFT>().IsNeeded; 1240 1241 // We are done processing archives, so lazy symbols that were used but not 1242 // found can be converted to undefined. We could also just delete the other 1243 // lazy symbols, but that seems to be more work than it is worth. 1244 return Sym.isLazy() && Sym.IsUsedInRegularObj; 1245 } 1246 1247 // Some files, such as .so or files between -{start,end}-lib may be removed 1248 // after their symbols are added to the symbol table. If that happens, we 1249 // need to remove symbols that refer files that no longer exist, so that 1250 // they won't appear in the symbol table of the output file. 1251 // 1252 // We remove symbols by demoting them to undefined symbol. 1253 template <class ELFT> static void demoteSymbols() { 1254 for (Symbol *Sym : Symtab->getSymbols()) { 1255 if (shouldDemote<ELFT>(*Sym)) { 1256 bool Used = Sym->Used; 1257 replaceSymbol<Undefined>(Sym, nullptr, Sym->getName(), Sym->Binding, 1258 Sym->StOther, Sym->Type); 1259 Sym->Used = Used; 1260 } 1261 } 1262 } 1263 1264 // The section referred to by S is considered address-significant. Set the 1265 // KeepUnique flag on the section if appropriate. 1266 static void markAddrsig(Symbol *S) { 1267 if (auto *D = dyn_cast_or_null<Defined>(S)) 1268 if (D->Section) 1269 // We don't need to keep text sections unique under --icf=all even if they 1270 // are address-significant. 1271 if (Config->ICF == ICFLevel::Safe || !(D->Section->Flags & SHF_EXECINSTR)) 1272 D->Section->KeepUnique = true; 1273 } 1274 1275 // Record sections that define symbols mentioned in --keep-unique <symbol> 1276 // and symbols referred to by address-significance tables. These sections are 1277 // ineligible for ICF. 1278 template <class ELFT> 1279 static void findKeepUniqueSections(opt::InputArgList &Args) { 1280 for (auto *Arg : Args.filtered(OPT_keep_unique)) { 1281 StringRef Name = Arg->getValue(); 1282 auto *D = dyn_cast_or_null<Defined>(Symtab->find(Name)); 1283 if (!D || !D->Section) { 1284 warn("could not find symbol " + Name + " to keep unique"); 1285 continue; 1286 } 1287 D->Section->KeepUnique = true; 1288 } 1289 1290 // --icf=all --ignore-data-address-equality means that we can ignore 1291 // the dynsym and address-significance tables entirely. 1292 if (Config->ICF == ICFLevel::All && Config->IgnoreDataAddressEquality) 1293 return; 1294 1295 // Symbols in the dynsym could be address-significant in other executables 1296 // or DSOs, so we conservatively mark them as address-significant. 1297 for (Symbol *S : Symtab->getSymbols()) 1298 if (S->includeInDynsym()) 1299 markAddrsig(S); 1300 1301 // Visit the address-significance table in each object file and mark each 1302 // referenced symbol as address-significant. 1303 for (InputFile *F : ObjectFiles) { 1304 auto *Obj = cast<ObjFile<ELFT>>(F); 1305 ArrayRef<Symbol *> Syms = Obj->getSymbols(); 1306 if (Obj->AddrsigSec) { 1307 ArrayRef<uint8_t> Contents = 1308 check(Obj->getObj().getSectionContents(Obj->AddrsigSec)); 1309 const uint8_t *Cur = Contents.begin(); 1310 while (Cur != Contents.end()) { 1311 unsigned Size; 1312 const char *Err; 1313 uint64_t SymIndex = decodeULEB128(Cur, &Size, Contents.end(), &Err); 1314 if (Err) 1315 fatal(toString(F) + ": could not decode addrsig section: " + Err); 1316 markAddrsig(Syms[SymIndex]); 1317 Cur += Size; 1318 } 1319 } else { 1320 // If an object file does not have an address-significance table, 1321 // conservatively mark all of its symbols as address-significant. 1322 for (Symbol *S : Syms) 1323 markAddrsig(S); 1324 } 1325 } 1326 } 1327 1328 // The --wrap option is a feature to rename symbols so that you can write 1329 // wrappers for existing functions. If you pass `-wrap=foo`, all 1330 // occurrences of symbol `foo` are resolved to `wrap_foo` (so, you are 1331 // expected to write `wrap_foo` function as a wrapper). The original 1332 // symbol becomes accessible as `real_foo`, so you can call that from your 1333 // wrapper. 1334 // 1335 // This data structure is instantiated for each -wrap option. 1336 struct WrappedSymbol { 1337 Symbol *Sym; 1338 Symbol *Real; 1339 Symbol *Wrap; 1340 }; 1341 1342 // Handles -wrap option. 1343 // 1344 // This function instantiates wrapper symbols. At this point, they seem 1345 // like they are not being used at all, so we explicitly set some flags so 1346 // that LTO won't eliminate them. 1347 template <class ELFT> 1348 static std::vector<WrappedSymbol> addWrappedSymbols(opt::InputArgList &Args) { 1349 std::vector<WrappedSymbol> V; 1350 DenseSet<StringRef> Seen; 1351 1352 for (auto *Arg : Args.filtered(OPT_wrap)) { 1353 StringRef Name = Arg->getValue(); 1354 if (!Seen.insert(Name).second) 1355 continue; 1356 1357 Symbol *Sym = Symtab->find(Name); 1358 if (!Sym) 1359 continue; 1360 1361 Symbol *Real = Symtab->addUndefined<ELFT>(Saver.save("__real_" + Name)); 1362 Symbol *Wrap = Symtab->addUndefined<ELFT>(Saver.save("__wrap_" + Name)); 1363 V.push_back({Sym, Real, Wrap}); 1364 1365 // We want to tell LTO not to inline symbols to be overwritten 1366 // because LTO doesn't know the final symbol contents after renaming. 1367 Real->CanInline = false; 1368 Sym->CanInline = false; 1369 1370 // Tell LTO not to eliminate these symbols. 1371 Sym->IsUsedInRegularObj = true; 1372 Wrap->IsUsedInRegularObj = true; 1373 } 1374 return V; 1375 } 1376 1377 // Do renaming for -wrap by updating pointers to symbols. 1378 // 1379 // When this function is executed, only InputFiles and symbol table 1380 // contain pointers to symbol objects. We visit them to replace pointers, 1381 // so that wrapped symbols are swapped as instructed by the command line. 1382 template <class ELFT> static void wrapSymbols(ArrayRef<WrappedSymbol> Wrapped) { 1383 DenseMap<Symbol *, Symbol *> Map; 1384 for (const WrappedSymbol &W : Wrapped) { 1385 Map[W.Sym] = W.Wrap; 1386 Map[W.Real] = W.Sym; 1387 } 1388 1389 // Update pointers in input files. 1390 parallelForEach(ObjectFiles, [&](InputFile *File) { 1391 std::vector<Symbol *> &Syms = File->getMutableSymbols(); 1392 for (size_t I = 0, E = Syms.size(); I != E; ++I) 1393 if (Symbol *S = Map.lookup(Syms[I])) 1394 Syms[I] = S; 1395 }); 1396 1397 // Update pointers in the symbol table. 1398 for (const WrappedSymbol &W : Wrapped) 1399 Symtab->wrap(W.Sym, W.Real, W.Wrap); 1400 } 1401 1402 static const char *LibcallRoutineNames[] = { 1403 #define HANDLE_LIBCALL(code, name) name, 1404 #include "llvm/IR/RuntimeLibcalls.def" 1405 #undef HANDLE_LIBCALL 1406 }; 1407 1408 // Do actual linking. Note that when this function is called, 1409 // all linker scripts have already been parsed. 1410 template <class ELFT> void LinkerDriver::link(opt::InputArgList &Args) { 1411 Target = getTarget(); 1412 1413 Config->MaxPageSize = getMaxPageSize(Args); 1414 Config->ImageBase = getImageBase(Args); 1415 1416 // If a -hash-style option was not given, set to a default value, 1417 // which varies depending on the target. 1418 if (!Args.hasArg(OPT_hash_style)) { 1419 if (Config->EMachine == EM_MIPS) 1420 Config->SysvHash = true; 1421 else 1422 Config->SysvHash = Config->GnuHash = true; 1423 } 1424 1425 // Default output filename is "a.out" by the Unix tradition. 1426 if (Config->OutputFile.empty()) 1427 Config->OutputFile = "a.out"; 1428 1429 // Fail early if the output file or map file is not writable. If a user has a 1430 // long link, e.g. due to a large LTO link, they do not wish to run it and 1431 // find that it failed because there was a mistake in their command-line. 1432 if (auto E = tryCreateFile(Config->OutputFile)) 1433 error("cannot open output file " + Config->OutputFile + ": " + E.message()); 1434 if (auto E = tryCreateFile(Config->MapFile)) 1435 error("cannot open map file " + Config->MapFile + ": " + E.message()); 1436 if (errorCount()) 1437 return; 1438 1439 // Use default entry point name if no name was given via the command 1440 // line nor linker scripts. For some reason, MIPS entry point name is 1441 // different from others. 1442 Config->WarnMissingEntry = 1443 (!Config->Entry.empty() || (!Config->Shared && !Config->Relocatable)); 1444 if (Config->Entry.empty() && !Config->Relocatable) 1445 Config->Entry = (Config->EMachine == EM_MIPS) ? "__start" : "_start"; 1446 1447 // Handle --trace-symbol. 1448 for (auto *Arg : Args.filtered(OPT_trace_symbol)) 1449 Symtab->trace(Arg->getValue()); 1450 1451 // Add all files to the symbol table. This will add almost all 1452 // symbols that we need to the symbol table. 1453 for (InputFile *F : Files) 1454 Symtab->addFile<ELFT>(F); 1455 1456 // Now that we have every file, we can decide if we will need a 1457 // dynamic symbol table. 1458 // We need one if we were asked to export dynamic symbols or if we are 1459 // producing a shared library. 1460 // We also need one if any shared libraries are used and for pie executables 1461 // (probably because the dynamic linker needs it). 1462 Config->HasDynSymTab = 1463 !SharedFiles.empty() || Config->Pic || Config->ExportDynamic; 1464 1465 // Some symbols (such as __ehdr_start) are defined lazily only when there 1466 // are undefined symbols for them, so we add these to trigger that logic. 1467 for (StringRef Sym : Script->ReferencedSymbols) 1468 Symtab->addUndefined<ELFT>(Sym); 1469 1470 // Handle the `--undefined <sym>` options. 1471 for (StringRef S : Config->Undefined) 1472 handleUndefined<ELFT>(S); 1473 1474 // If an entry symbol is in a static archive, pull out that file now. 1475 handleUndefined<ELFT>(Config->Entry); 1476 1477 // If any of our inputs are bitcode files, the LTO code generator may create 1478 // references to certain library functions that might not be explicit in the 1479 // bitcode file's symbol table. If any of those library functions are defined 1480 // in a bitcode file in an archive member, we need to arrange to use LTO to 1481 // compile those archive members by adding them to the link beforehand. 1482 // 1483 // However, adding all libcall symbols to the link can have undesired 1484 // consequences. For example, the libgcc implementation of 1485 // __sync_val_compare_and_swap_8 on 32-bit ARM pulls in an .init_array entry 1486 // that aborts the program if the Linux kernel does not support 64-bit 1487 // atomics, which would prevent the program from running even if it does not 1488 // use 64-bit atomics. 1489 // 1490 // Therefore, we only add libcall symbols to the link before LTO if we have 1491 // to, i.e. if the symbol's definition is in bitcode. Any other required 1492 // libcall symbols will be added to the link after LTO when we add the LTO 1493 // object file to the link. 1494 if (!BitcodeFiles.empty()) 1495 for (const char *S : LibcallRoutineNames) 1496 handleLibcall<ELFT>(S); 1497 1498 // Return if there were name resolution errors. 1499 if (errorCount()) 1500 return; 1501 1502 // Now when we read all script files, we want to finalize order of linker 1503 // script commands, which can be not yet final because of INSERT commands. 1504 Script->processInsertCommands(); 1505 1506 // We want to declare linker script's symbols early, 1507 // so that we can version them. 1508 // They also might be exported if referenced by DSOs. 1509 Script->declareSymbols(); 1510 1511 // Handle the -exclude-libs option. 1512 if (Args.hasArg(OPT_exclude_libs)) 1513 excludeLibs<ELFT>(Args); 1514 1515 // Create ElfHeader early. We need a dummy section in 1516 // addReservedSymbols to mark the created symbols as not absolute. 1517 Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC); 1518 Out::ElfHeader->Size = sizeof(typename ELFT::Ehdr); 1519 1520 // We need to create some reserved symbols such as _end. Create them. 1521 if (!Config->Relocatable) 1522 addReservedSymbols(); 1523 1524 // Apply version scripts. 1525 // 1526 // For a relocatable output, version scripts don't make sense, and 1527 // parsing a symbol version string (e.g. dropping "@ver1" from a symbol 1528 // name "foo@ver1") rather do harm, so we don't call this if -r is given. 1529 if (!Config->Relocatable) 1530 Symtab->scanVersionScript(); 1531 1532 // Create wrapped symbols for -wrap option. 1533 std::vector<WrappedSymbol> Wrapped = addWrappedSymbols<ELFT>(Args); 1534 1535 // Do link-time optimization if given files are LLVM bitcode files. 1536 // This compiles bitcode files into real object files. 1537 // 1538 // With this the symbol table should be complete. After this, no new names 1539 // except a few linker-synthesized ones will be added to the symbol table. 1540 Symtab->addCombinedLTOObject<ELFT>(); 1541 if (errorCount()) 1542 return; 1543 1544 // If -thinlto-index-only is given, we should create only "index 1545 // files" and not object files. Index file creation is already done 1546 // in addCombinedLTOObject, so we are done if that's the case. 1547 if (Config->ThinLTOIndexOnly) 1548 return; 1549 1550 // Apply symbol renames for -wrap. 1551 if (!Wrapped.empty()) 1552 wrapSymbols<ELFT>(Wrapped); 1553 1554 // Now that we have a complete list of input files. 1555 // Beyond this point, no new files are added. 1556 // Aggregate all input sections into one place. 1557 for (InputFile *F : ObjectFiles) 1558 for (InputSectionBase *S : F->getSections()) 1559 if (S && S != &InputSection::Discarded) 1560 InputSections.push_back(S); 1561 for (BinaryFile *F : BinaryFiles) 1562 for (InputSectionBase *S : F->getSections()) 1563 InputSections.push_back(cast<InputSection>(S)); 1564 1565 // We do not want to emit debug sections if --strip-all 1566 // or -strip-debug are given. 1567 if (Config->Strip != StripPolicy::None) 1568 llvm::erase_if(InputSections, [](InputSectionBase *S) { 1569 return S->Name.startswith(".debug") || S->Name.startswith(".zdebug"); 1570 }); 1571 1572 Config->EFlags = Target->calcEFlags(); 1573 1574 if (Config->EMachine == EM_ARM) { 1575 // FIXME: These warnings can be removed when lld only uses these features 1576 // when the input objects have been compiled with an architecture that 1577 // supports them. 1578 if (Config->ARMHasBlx == false) 1579 warn("lld uses blx instruction, no object with architecture supporting " 1580 "feature detected."); 1581 } 1582 1583 // This adds a .comment section containing a version string. We have to add it 1584 // before decompressAndMergeSections because the .comment section is a 1585 // mergeable section. 1586 if (!Config->Relocatable) 1587 InputSections.push_back(createCommentSection()); 1588 1589 // Do size optimizations: garbage collection, merging of SHF_MERGE sections 1590 // and identical code folding. 1591 decompressSections(); 1592 splitSections<ELFT>(); 1593 markLive<ELFT>(); 1594 demoteSymbols<ELFT>(); 1595 mergeSections(); 1596 if (Config->ICF != ICFLevel::None) { 1597 findKeepUniqueSections<ELFT>(Args); 1598 doIcf<ELFT>(); 1599 } 1600 1601 // Read the callgraph now that we know what was gced or icfed 1602 if (auto *Arg = Args.getLastArg(OPT_call_graph_ordering_file)) 1603 if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue())) 1604 readCallGraph(*Buffer); 1605 1606 // Write the result to the file. 1607 writeResult<ELFT>(); 1608 } 1609