1 //===- Driver.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 // The driver drives the entire linking process. It is responsible for 10 // parsing command line options and doing whatever it is instructed to do. 11 // 12 // One notable thing in the LLD's driver when compared to other linkers is 13 // that the LLD's driver is agnostic on the host operating system. 14 // Other linkers usually have implicit default values (such as a dynamic 15 // linker path or library paths) for each host OS. 16 // 17 // I don't think implicit default values are useful because they are 18 // usually explicitly specified by the compiler driver. They can even 19 // be harmful when you are doing cross-linking. Therefore, in LLD, we 20 // simply trust the compiler driver to pass all required options and 21 // don't try to make effort on our side. 22 // 23 //===----------------------------------------------------------------------===// 24 25 #include "Driver.h" 26 #include "Config.h" 27 #include "ICF.h" 28 #include "InputFiles.h" 29 #include "InputSection.h" 30 #include "LinkerScript.h" 31 #include "MarkLive.h" 32 #include "OutputSections.h" 33 #include "ScriptParser.h" 34 #include "SymbolTable.h" 35 #include "Symbols.h" 36 #include "SyntheticSections.h" 37 #include "Target.h" 38 #include "Writer.h" 39 #include "lld/Common/Args.h" 40 #include "lld/Common/Driver.h" 41 #include "lld/Common/ErrorHandler.h" 42 #include "lld/Common/Filesystem.h" 43 #include "lld/Common/Memory.h" 44 #include "lld/Common/Strings.h" 45 #include "lld/Common/TargetOptionsCommandFlags.h" 46 #include "lld/Common/Threads.h" 47 #include "lld/Common/Version.h" 48 #include "llvm/ADT/SetVector.h" 49 #include "llvm/ADT/StringExtras.h" 50 #include "llvm/ADT/StringSwitch.h" 51 #include "llvm/Support/CommandLine.h" 52 #include "llvm/Support/Compression.h" 53 #include "llvm/Support/GlobPattern.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 using namespace llvm::support; 67 68 using namespace lld; 69 using namespace lld::elf; 70 71 Configuration *elf::config; 72 LinkerDriver *elf::driver; 73 74 static void setConfigs(opt::InputArgList &args); 75 static void readConfigs(opt::InputArgList &args); 76 77 bool elf::link(ArrayRef<const char *> args, bool canExitEarly, 78 raw_ostream &error) { 79 errorHandler().logName = args::getFilenameWithoutExe(args[0]); 80 errorHandler().errorLimitExceededMsg = 81 "too many errors emitted, stopping now (use " 82 "-error-limit=0 to see all errors)"; 83 errorHandler().errorOS = &error; 84 errorHandler().exitEarly = canExitEarly; 85 errorHandler().colorDiagnostics = error.has_colors(); 86 87 inputSections.clear(); 88 outputSections.clear(); 89 binaryFiles.clear(); 90 bitcodeFiles.clear(); 91 objectFiles.clear(); 92 sharedFiles.clear(); 93 94 config = make<Configuration>(); 95 driver = make<LinkerDriver>(); 96 script = make<LinkerScript>(); 97 symtab = make<SymbolTable>(); 98 99 tar = nullptr; 100 memset(&in, 0, sizeof(in)); 101 102 partitions = {Partition()}; 103 104 SharedFile::vernauxNum = 0; 105 106 config->progName = args[0]; 107 108 driver->main(args); 109 110 // Exit immediately if we don't need to return to the caller. 111 // This saves time because the overhead of calling destructors 112 // for all globally-allocated objects is not negligible. 113 if (canExitEarly) 114 exitLld(errorCount() ? 1 : 0); 115 116 freeArena(); 117 return !errorCount(); 118 } 119 120 // Parses a linker -m option. 121 static std::tuple<ELFKind, uint16_t, uint8_t> parseEmulation(StringRef emul) { 122 uint8_t osabi = 0; 123 StringRef s = emul; 124 if (s.endswith("_fbsd")) { 125 s = s.drop_back(5); 126 osabi = ELFOSABI_FREEBSD; 127 } 128 129 std::pair<ELFKind, uint16_t> ret = 130 StringSwitch<std::pair<ELFKind, uint16_t>>(s) 131 .Cases("aarch64elf", "aarch64linux", "aarch64_elf64_le_vec", 132 {ELF64LEKind, EM_AARCH64}) 133 .Cases("armelf", "armelf_linux_eabi", {ELF32LEKind, EM_ARM}) 134 .Case("elf32_x86_64", {ELF32LEKind, EM_X86_64}) 135 .Cases("elf32btsmip", "elf32btsmipn32", {ELF32BEKind, EM_MIPS}) 136 .Cases("elf32ltsmip", "elf32ltsmipn32", {ELF32LEKind, EM_MIPS}) 137 .Case("elf32lriscv", {ELF32LEKind, EM_RISCV}) 138 .Cases("elf32ppc", "elf32ppclinux", {ELF32BEKind, EM_PPC}) 139 .Case("elf64btsmip", {ELF64BEKind, EM_MIPS}) 140 .Case("elf64ltsmip", {ELF64LEKind, EM_MIPS}) 141 .Case("elf64lriscv", {ELF64LEKind, EM_RISCV}) 142 .Case("elf64ppc", {ELF64BEKind, EM_PPC64}) 143 .Case("elf64lppc", {ELF64LEKind, EM_PPC64}) 144 .Cases("elf_amd64", "elf_x86_64", {ELF64LEKind, EM_X86_64}) 145 .Case("elf_i386", {ELF32LEKind, EM_386}) 146 .Case("elf_iamcu", {ELF32LEKind, EM_IAMCU}) 147 .Default({ELFNoneKind, EM_NONE}); 148 149 if (ret.first == ELFNoneKind) 150 error("unknown emulation: " + emul); 151 return std::make_tuple(ret.first, ret.second, osabi); 152 } 153 154 // Returns slices of MB by parsing MB as an archive file. 155 // Each slice consists of a member file in the archive. 156 std::vector<std::pair<MemoryBufferRef, uint64_t>> static getArchiveMembers( 157 MemoryBufferRef mb) { 158 std::unique_ptr<Archive> file = 159 CHECK(Archive::create(mb), 160 mb.getBufferIdentifier() + ": failed to parse archive"); 161 162 std::vector<std::pair<MemoryBufferRef, uint64_t>> v; 163 Error err = Error::success(); 164 bool addToTar = file->isThin() && tar; 165 for (const ErrorOr<Archive::Child> &cOrErr : file->children(err)) { 166 Archive::Child c = 167 CHECK(cOrErr, mb.getBufferIdentifier() + 168 ": could not get the child of the archive"); 169 MemoryBufferRef mbref = 170 CHECK(c.getMemoryBufferRef(), 171 mb.getBufferIdentifier() + 172 ": could not get the buffer for a child of the archive"); 173 if (addToTar) 174 tar->append(relativeToRoot(check(c.getFullName())), mbref.getBuffer()); 175 v.push_back(std::make_pair(mbref, c.getChildOffset())); 176 } 177 if (err) 178 fatal(mb.getBufferIdentifier() + ": Archive::children failed: " + 179 toString(std::move(err))); 180 181 // Take ownership of memory buffers created for members of thin archives. 182 for (std::unique_ptr<MemoryBuffer> &mb : file->takeThinBuffers()) 183 make<std::unique_ptr<MemoryBuffer>>(std::move(mb)); 184 185 return v; 186 } 187 188 // Opens a file and create a file object. Path has to be resolved already. 189 void LinkerDriver::addFile(StringRef path, bool withLOption) { 190 using namespace sys::fs; 191 192 Optional<MemoryBufferRef> buffer = readFile(path); 193 if (!buffer.hasValue()) 194 return; 195 MemoryBufferRef mbref = *buffer; 196 197 if (config->formatBinary) { 198 files.push_back(make<BinaryFile>(mbref)); 199 return; 200 } 201 202 switch (identify_magic(mbref.getBuffer())) { 203 case file_magic::unknown: 204 readLinkerScript(mbref); 205 return; 206 case file_magic::archive: { 207 // Handle -whole-archive. 208 if (inWholeArchive) { 209 for (const auto &p : getArchiveMembers(mbref)) 210 files.push_back(createObjectFile(p.first, path, p.second)); 211 return; 212 } 213 214 std::unique_ptr<Archive> file = 215 CHECK(Archive::create(mbref), path + ": failed to parse archive"); 216 217 // If an archive file has no symbol table, it is likely that a user 218 // is attempting LTO and using a default ar command that doesn't 219 // understand the LLVM bitcode file. It is a pretty common error, so 220 // we'll handle it as if it had a symbol table. 221 if (!file->isEmpty() && !file->hasSymbolTable()) { 222 // Check if all members are bitcode files. If not, ignore, which is the 223 // default action without the LTO hack described above. 224 for (const std::pair<MemoryBufferRef, uint64_t> &p : 225 getArchiveMembers(mbref)) 226 if (identify_magic(p.first.getBuffer()) != file_magic::bitcode) { 227 error(path + ": archive has no index; run ranlib to add one"); 228 return; 229 } 230 231 for (const std::pair<MemoryBufferRef, uint64_t> &p : 232 getArchiveMembers(mbref)) 233 files.push_back(make<LazyObjFile>(p.first, path, p.second)); 234 return; 235 } 236 237 // Handle the regular case. 238 files.push_back(make<ArchiveFile>(std::move(file))); 239 return; 240 } 241 case file_magic::elf_shared_object: 242 if (config->isStatic || config->relocatable) { 243 error("attempted static link of dynamic object " + path); 244 return; 245 } 246 247 // DSOs usually have DT_SONAME tags in their ELF headers, and the 248 // sonames are used to identify DSOs. But if they are missing, 249 // they are identified by filenames. We don't know whether the new 250 // file has a DT_SONAME or not because we haven't parsed it yet. 251 // Here, we set the default soname for the file because we might 252 // need it later. 253 // 254 // If a file was specified by -lfoo, the directory part is not 255 // significant, as a user did not specify it. This behavior is 256 // compatible with GNU. 257 files.push_back( 258 make<SharedFile>(mbref, withLOption ? path::filename(path) : path)); 259 return; 260 case file_magic::bitcode: 261 case file_magic::elf_relocatable: 262 if (inLib) 263 files.push_back(make<LazyObjFile>(mbref, "", 0)); 264 else 265 files.push_back(createObjectFile(mbref)); 266 break; 267 default: 268 error(path + ": unknown file type"); 269 } 270 } 271 272 // Add a given library by searching it from input search paths. 273 void LinkerDriver::addLibrary(StringRef name) { 274 if (Optional<std::string> path = searchLibrary(name)) 275 addFile(*path, /*withLOption=*/true); 276 else 277 error("unable to find library -l" + name); 278 } 279 280 // This function is called on startup. We need this for LTO since 281 // LTO calls LLVM functions to compile bitcode files to native code. 282 // Technically this can be delayed until we read bitcode files, but 283 // we don't bother to do lazily because the initialization is fast. 284 static void initLLVM() { 285 InitializeAllTargets(); 286 InitializeAllTargetMCs(); 287 InitializeAllAsmPrinters(); 288 InitializeAllAsmParsers(); 289 } 290 291 // Some command line options or some combinations of them are not allowed. 292 // This function checks for such errors. 293 static void checkOptions() { 294 // The MIPS ABI as of 2016 does not support the GNU-style symbol lookup 295 // table which is a relatively new feature. 296 if (config->emachine == EM_MIPS && config->gnuHash) 297 error("the .gnu.hash section is not compatible with the MIPS target"); 298 299 if (config->fixCortexA53Errata843419 && config->emachine != EM_AARCH64) 300 error("--fix-cortex-a53-843419 is only supported on AArch64 targets"); 301 302 if (config->tocOptimize && config->emachine != EM_PPC64) 303 error("--toc-optimize is only supported on the PowerPC64 target"); 304 305 if (config->pie && config->shared) 306 error("-shared and -pie may not be used together"); 307 308 if (!config->shared && !config->filterList.empty()) 309 error("-F may not be used without -shared"); 310 311 if (!config->shared && !config->auxiliaryList.empty()) 312 error("-f may not be used without -shared"); 313 314 if (!config->relocatable && !config->defineCommon) 315 error("-no-define-common not supported in non relocatable output"); 316 317 if (config->zText && config->zIfuncNoplt) 318 error("-z text and -z ifunc-noplt may not be used together"); 319 320 if (config->relocatable) { 321 if (config->shared) 322 error("-r and -shared may not be used together"); 323 if (config->gcSections) 324 error("-r and --gc-sections may not be used together"); 325 if (config->gdbIndex) 326 error("-r and --gdb-index may not be used together"); 327 if (config->icf != ICFLevel::None) 328 error("-r and --icf may not be used together"); 329 if (config->pie) 330 error("-r and -pie may not be used together"); 331 } 332 333 if (config->executeOnly) { 334 if (config->emachine != EM_AARCH64) 335 error("-execute-only is only supported on AArch64 targets"); 336 337 if (config->singleRoRx && !script->hasSectionsCommand) 338 error("-execute-only and -no-rosegment cannot be used together"); 339 } 340 341 if (config->zRetpolineplt && config->requireCET) 342 error("--require-cet may not be used with -z retpolineplt"); 343 344 if (config->emachine != EM_AARCH64) { 345 if (config->pacPlt) 346 error("--pac-plt only supported on AArch64"); 347 if (config->forceBTI) 348 error("--force-bti only supported on AArch64"); 349 } 350 } 351 352 static const char *getReproduceOption(opt::InputArgList &args) { 353 if (auto *arg = args.getLastArg(OPT_reproduce)) 354 return arg->getValue(); 355 return getenv("LLD_REPRODUCE"); 356 } 357 358 static bool hasZOption(opt::InputArgList &args, StringRef key) { 359 for (auto *arg : args.filtered(OPT_z)) 360 if (key == arg->getValue()) 361 return true; 362 return false; 363 } 364 365 static bool getZFlag(opt::InputArgList &args, StringRef k1, StringRef k2, 366 bool Default) { 367 for (auto *arg : args.filtered_reverse(OPT_z)) { 368 if (k1 == arg->getValue()) 369 return true; 370 if (k2 == arg->getValue()) 371 return false; 372 } 373 return Default; 374 } 375 376 static bool isKnownZFlag(StringRef s) { 377 return s == "combreloc" || s == "copyreloc" || s == "defs" || 378 s == "execstack" || s == "global" || s == "hazardplt" || 379 s == "ifunc-noplt" || s == "initfirst" || s == "interpose" || 380 s == "keep-text-section-prefix" || s == "lazy" || s == "muldefs" || 381 s == "separate-code" || s == "nocombreloc" || s == "nocopyreloc" || 382 s == "nodefaultlib" || s == "nodelete" || s == "nodlopen" || 383 s == "noexecstack" || s == "nokeep-text-section-prefix" || 384 s == "norelro" || s == "noseparate-code" || s == "notext" || 385 s == "now" || s == "origin" || s == "relro" || s == "retpolineplt" || 386 s == "rodynamic" || s == "text" || s == "wxneeded" || 387 s.startswith("common-page-size") || s.startswith("max-page-size=") || 388 s.startswith("stack-size="); 389 } 390 391 // Report an error for an unknown -z option. 392 static void checkZOptions(opt::InputArgList &args) { 393 for (auto *arg : args.filtered(OPT_z)) 394 if (!isKnownZFlag(arg->getValue())) 395 error("unknown -z value: " + StringRef(arg->getValue())); 396 } 397 398 void LinkerDriver::main(ArrayRef<const char *> argsArr) { 399 ELFOptTable parser; 400 opt::InputArgList args = parser.parse(argsArr.slice(1)); 401 402 // Interpret this flag early because error() depends on them. 403 errorHandler().errorLimit = args::getInteger(args, OPT_error_limit, 20); 404 checkZOptions(args); 405 406 // Handle -help 407 if (args.hasArg(OPT_help)) { 408 printHelp(); 409 return; 410 } 411 412 // Handle -v or -version. 413 // 414 // A note about "compatible with GNU linkers" message: this is a hack for 415 // scripts generated by GNU Libtool 2.4.6 (released in February 2014 and 416 // still the newest version in March 2017) or earlier to recognize LLD as 417 // a GNU compatible linker. As long as an output for the -v option 418 // contains "GNU" or "with BFD", they recognize us as GNU-compatible. 419 // 420 // This is somewhat ugly hack, but in reality, we had no choice other 421 // than doing this. Considering the very long release cycle of Libtool, 422 // it is not easy to improve it to recognize LLD as a GNU compatible 423 // linker in a timely manner. Even if we can make it, there are still a 424 // lot of "configure" scripts out there that are generated by old version 425 // of Libtool. We cannot convince every software developer to migrate to 426 // the latest version and re-generate scripts. So we have this hack. 427 if (args.hasArg(OPT_v) || args.hasArg(OPT_version)) 428 message(getLLDVersion() + " (compatible with GNU linkers)"); 429 430 if (const char *path = getReproduceOption(args)) { 431 // Note that --reproduce is a debug option so you can ignore it 432 // if you are trying to understand the whole picture of the code. 433 Expected<std::unique_ptr<TarWriter>> errOrWriter = 434 TarWriter::create(path, path::stem(path)); 435 if (errOrWriter) { 436 tar = std::move(*errOrWriter); 437 tar->append("response.txt", createResponseFile(args)); 438 tar->append("version.txt", getLLDVersion() + "\n"); 439 } else { 440 error("--reproduce: " + toString(errOrWriter.takeError())); 441 } 442 } 443 444 readConfigs(args); 445 446 // The behavior of -v or --version is a bit strange, but this is 447 // needed for compatibility with GNU linkers. 448 if (args.hasArg(OPT_v) && !args.hasArg(OPT_INPUT)) 449 return; 450 if (args.hasArg(OPT_version)) 451 return; 452 453 initLLVM(); 454 createFiles(args); 455 if (errorCount()) 456 return; 457 458 inferMachineType(); 459 setConfigs(args); 460 checkOptions(); 461 if (errorCount()) 462 return; 463 464 // The Target instance handles target-specific stuff, such as applying 465 // relocations or writing a PLT section. It also contains target-dependent 466 // values such as a default image base address. 467 target = getTarget(); 468 469 switch (config->ekind) { 470 case ELF32LEKind: 471 link<ELF32LE>(args); 472 return; 473 case ELF32BEKind: 474 link<ELF32BE>(args); 475 return; 476 case ELF64LEKind: 477 link<ELF64LE>(args); 478 return; 479 case ELF64BEKind: 480 link<ELF64BE>(args); 481 return; 482 default: 483 llvm_unreachable("unknown Config->EKind"); 484 } 485 } 486 487 static std::string getRpath(opt::InputArgList &args) { 488 std::vector<StringRef> v = args::getStrings(args, OPT_rpath); 489 return llvm::join(v.begin(), v.end(), ":"); 490 } 491 492 // Determines what we should do if there are remaining unresolved 493 // symbols after the name resolution. 494 static UnresolvedPolicy getUnresolvedSymbolPolicy(opt::InputArgList &args) { 495 UnresolvedPolicy errorOrWarn = args.hasFlag(OPT_error_unresolved_symbols, 496 OPT_warn_unresolved_symbols, true) 497 ? UnresolvedPolicy::ReportError 498 : UnresolvedPolicy::Warn; 499 500 // Process the last of -unresolved-symbols, -no-undefined or -z defs. 501 for (auto *arg : llvm::reverse(args)) { 502 switch (arg->getOption().getID()) { 503 case OPT_unresolved_symbols: { 504 StringRef s = arg->getValue(); 505 if (s == "ignore-all" || s == "ignore-in-object-files") 506 return UnresolvedPolicy::Ignore; 507 if (s == "ignore-in-shared-libs" || s == "report-all") 508 return errorOrWarn; 509 error("unknown --unresolved-symbols value: " + s); 510 continue; 511 } 512 case OPT_no_undefined: 513 return errorOrWarn; 514 case OPT_z: 515 if (StringRef(arg->getValue()) == "defs") 516 return errorOrWarn; 517 continue; 518 } 519 } 520 521 // -shared implies -unresolved-symbols=ignore-all because missing 522 // symbols are likely to be resolved at runtime using other DSOs. 523 if (config->shared) 524 return UnresolvedPolicy::Ignore; 525 return errorOrWarn; 526 } 527 528 static Target2Policy getTarget2(opt::InputArgList &args) { 529 StringRef s = args.getLastArgValue(OPT_target2, "got-rel"); 530 if (s == "rel") 531 return Target2Policy::Rel; 532 if (s == "abs") 533 return Target2Policy::Abs; 534 if (s == "got-rel") 535 return Target2Policy::GotRel; 536 error("unknown --target2 option: " + s); 537 return Target2Policy::GotRel; 538 } 539 540 static bool isOutputFormatBinary(opt::InputArgList &args) { 541 StringRef s = args.getLastArgValue(OPT_oformat, "elf"); 542 if (s == "binary") 543 return true; 544 if (!s.startswith("elf")) 545 error("unknown --oformat value: " + s); 546 return false; 547 } 548 549 static DiscardPolicy getDiscard(opt::InputArgList &args) { 550 if (args.hasArg(OPT_relocatable)) 551 return DiscardPolicy::None; 552 553 auto *arg = 554 args.getLastArg(OPT_discard_all, OPT_discard_locals, OPT_discard_none); 555 if (!arg) 556 return DiscardPolicy::Default; 557 if (arg->getOption().getID() == OPT_discard_all) 558 return DiscardPolicy::All; 559 if (arg->getOption().getID() == OPT_discard_locals) 560 return DiscardPolicy::Locals; 561 return DiscardPolicy::None; 562 } 563 564 static StringRef getDynamicLinker(opt::InputArgList &args) { 565 auto *arg = args.getLastArg(OPT_dynamic_linker, OPT_no_dynamic_linker); 566 if (!arg || arg->getOption().getID() == OPT_no_dynamic_linker) 567 return ""; 568 return arg->getValue(); 569 } 570 571 static ICFLevel getICF(opt::InputArgList &args) { 572 auto *arg = args.getLastArg(OPT_icf_none, OPT_icf_safe, OPT_icf_all); 573 if (!arg || arg->getOption().getID() == OPT_icf_none) 574 return ICFLevel::None; 575 if (arg->getOption().getID() == OPT_icf_safe) 576 return ICFLevel::Safe; 577 return ICFLevel::All; 578 } 579 580 static StripPolicy getStrip(opt::InputArgList &args) { 581 if (args.hasArg(OPT_relocatable)) 582 return StripPolicy::None; 583 584 auto *arg = args.getLastArg(OPT_strip_all, OPT_strip_debug); 585 if (!arg) 586 return StripPolicy::None; 587 if (arg->getOption().getID() == OPT_strip_all) 588 return StripPolicy::All; 589 return StripPolicy::Debug; 590 } 591 592 static uint64_t parseSectionAddress(StringRef s, opt::InputArgList &args, 593 const opt::Arg &arg) { 594 uint64_t va = 0; 595 if (s.startswith("0x")) 596 s = s.drop_front(2); 597 if (!to_integer(s, va, 16)) 598 error("invalid argument: " + arg.getAsString(args)); 599 return va; 600 } 601 602 static StringMap<uint64_t> getSectionStartMap(opt::InputArgList &args) { 603 StringMap<uint64_t> ret; 604 for (auto *arg : args.filtered(OPT_section_start)) { 605 StringRef name; 606 StringRef addr; 607 std::tie(name, addr) = StringRef(arg->getValue()).split('='); 608 ret[name] = parseSectionAddress(addr, args, *arg); 609 } 610 611 if (auto *arg = args.getLastArg(OPT_Ttext)) 612 ret[".text"] = parseSectionAddress(arg->getValue(), args, *arg); 613 if (auto *arg = args.getLastArg(OPT_Tdata)) 614 ret[".data"] = parseSectionAddress(arg->getValue(), args, *arg); 615 if (auto *arg = args.getLastArg(OPT_Tbss)) 616 ret[".bss"] = parseSectionAddress(arg->getValue(), args, *arg); 617 return ret; 618 } 619 620 static SortSectionPolicy getSortSection(opt::InputArgList &args) { 621 StringRef s = args.getLastArgValue(OPT_sort_section); 622 if (s == "alignment") 623 return SortSectionPolicy::Alignment; 624 if (s == "name") 625 return SortSectionPolicy::Name; 626 if (!s.empty()) 627 error("unknown --sort-section rule: " + s); 628 return SortSectionPolicy::Default; 629 } 630 631 static OrphanHandlingPolicy getOrphanHandling(opt::InputArgList &args) { 632 StringRef s = args.getLastArgValue(OPT_orphan_handling, "place"); 633 if (s == "warn") 634 return OrphanHandlingPolicy::Warn; 635 if (s == "error") 636 return OrphanHandlingPolicy::Error; 637 if (s != "place") 638 error("unknown --orphan-handling mode: " + s); 639 return OrphanHandlingPolicy::Place; 640 } 641 642 // Parse --build-id or --build-id=<style>. We handle "tree" as a 643 // synonym for "sha1" because all our hash functions including 644 // -build-id=sha1 are actually tree hashes for performance reasons. 645 static std::pair<BuildIdKind, std::vector<uint8_t>> 646 getBuildId(opt::InputArgList &args) { 647 auto *arg = args.getLastArg(OPT_build_id, OPT_build_id_eq); 648 if (!arg) 649 return {BuildIdKind::None, {}}; 650 651 if (arg->getOption().getID() == OPT_build_id) 652 return {BuildIdKind::Fast, {}}; 653 654 StringRef s = arg->getValue(); 655 if (s == "fast") 656 return {BuildIdKind::Fast, {}}; 657 if (s == "md5") 658 return {BuildIdKind::Md5, {}}; 659 if (s == "sha1" || s == "tree") 660 return {BuildIdKind::Sha1, {}}; 661 if (s == "uuid") 662 return {BuildIdKind::Uuid, {}}; 663 if (s.startswith("0x")) 664 return {BuildIdKind::Hexstring, parseHex(s.substr(2))}; 665 666 if (s != "none") 667 error("unknown --build-id style: " + s); 668 return {BuildIdKind::None, {}}; 669 } 670 671 static std::pair<bool, bool> getPackDynRelocs(opt::InputArgList &args) { 672 StringRef s = args.getLastArgValue(OPT_pack_dyn_relocs, "none"); 673 if (s == "android") 674 return {true, false}; 675 if (s == "relr") 676 return {false, true}; 677 if (s == "android+relr") 678 return {true, true}; 679 680 if (s != "none") 681 error("unknown -pack-dyn-relocs format: " + s); 682 return {false, false}; 683 } 684 685 static void readCallGraph(MemoryBufferRef mb) { 686 // Build a map from symbol name to section 687 DenseMap<StringRef, Symbol *> map; 688 for (InputFile *file : objectFiles) 689 for (Symbol *sym : file->getSymbols()) 690 map[sym->getName()] = sym; 691 692 auto findSection = [&](StringRef name) -> InputSectionBase * { 693 Symbol *sym = map.lookup(name); 694 if (!sym) { 695 if (config->warnSymbolOrdering) 696 warn(mb.getBufferIdentifier() + ": no such symbol: " + name); 697 return nullptr; 698 } 699 maybeWarnUnorderableSymbol(sym); 700 701 if (Defined *dr = dyn_cast_or_null<Defined>(sym)) 702 return dyn_cast_or_null<InputSectionBase>(dr->section); 703 return nullptr; 704 }; 705 706 for (StringRef line : args::getLines(mb)) { 707 SmallVector<StringRef, 3> fields; 708 line.split(fields, ' '); 709 uint64_t count; 710 711 if (fields.size() != 3 || !to_integer(fields[2], count)) { 712 error(mb.getBufferIdentifier() + ": parse error"); 713 return; 714 } 715 716 if (InputSectionBase *from = findSection(fields[0])) 717 if (InputSectionBase *to = findSection(fields[1])) 718 config->callGraphProfile[std::make_pair(from, to)] += count; 719 } 720 } 721 722 template <class ELFT> static void readCallGraphsFromObjectFiles() { 723 for (auto file : objectFiles) { 724 auto *obj = cast<ObjFile<ELFT>>(file); 725 726 for (const Elf_CGProfile_Impl<ELFT> &cgpe : obj->cgProfile) { 727 auto *fromSym = dyn_cast<Defined>(&obj->getSymbol(cgpe.cgp_from)); 728 auto *toSym = dyn_cast<Defined>(&obj->getSymbol(cgpe.cgp_to)); 729 if (!fromSym || !toSym) 730 continue; 731 732 auto *from = dyn_cast_or_null<InputSectionBase>(fromSym->section); 733 auto *to = dyn_cast_or_null<InputSectionBase>(toSym->section); 734 if (from && to) 735 config->callGraphProfile[{from, to}] += cgpe.cgp_weight; 736 } 737 } 738 } 739 740 static bool getCompressDebugSections(opt::InputArgList &args) { 741 StringRef s = args.getLastArgValue(OPT_compress_debug_sections, "none"); 742 if (s == "none") 743 return false; 744 if (s != "zlib") 745 error("unknown --compress-debug-sections value: " + s); 746 if (!zlib::isAvailable()) 747 error("--compress-debug-sections: zlib is not available"); 748 return true; 749 } 750 751 static std::pair<StringRef, StringRef> getOldNewOptions(opt::InputArgList &args, 752 unsigned id) { 753 auto *arg = args.getLastArg(id); 754 if (!arg) 755 return {"", ""}; 756 757 StringRef s = arg->getValue(); 758 std::pair<StringRef, StringRef> ret = s.split(';'); 759 if (ret.second.empty()) 760 error(arg->getSpelling() + " expects 'old;new' format, but got " + s); 761 return ret; 762 } 763 764 // Parse the symbol ordering file and warn for any duplicate entries. 765 static std::vector<StringRef> getSymbolOrderingFile(MemoryBufferRef mb) { 766 SetVector<StringRef> names; 767 for (StringRef s : args::getLines(mb)) 768 if (!names.insert(s) && config->warnSymbolOrdering) 769 warn(mb.getBufferIdentifier() + ": duplicate ordered symbol: " + s); 770 771 return names.takeVector(); 772 } 773 774 static void parseClangOption(StringRef opt, const Twine &msg) { 775 std::string err; 776 raw_string_ostream os(err); 777 778 const char *argv[] = {config->progName.data(), opt.data()}; 779 if (cl::ParseCommandLineOptions(2, argv, "", &os)) 780 return; 781 os.flush(); 782 error(msg + ": " + StringRef(err).trim()); 783 } 784 785 // Initializes Config members by the command line options. 786 static void readConfigs(opt::InputArgList &args) { 787 errorHandler().verbose = args.hasArg(OPT_verbose); 788 errorHandler().fatalWarnings = 789 args.hasFlag(OPT_fatal_warnings, OPT_no_fatal_warnings, false); 790 errorHandler().vsDiagnostics = 791 args.hasArg(OPT_visual_studio_diagnostics_format, false); 792 threadsEnabled = args.hasFlag(OPT_threads, OPT_no_threads, true); 793 794 config->allowMultipleDefinition = 795 args.hasFlag(OPT_allow_multiple_definition, 796 OPT_no_allow_multiple_definition, false) || 797 hasZOption(args, "muldefs"); 798 config->allowShlibUndefined = 799 args.hasFlag(OPT_allow_shlib_undefined, OPT_no_allow_shlib_undefined, 800 args.hasArg(OPT_shared)); 801 config->auxiliaryList = args::getStrings(args, OPT_auxiliary); 802 config->bsymbolic = args.hasArg(OPT_Bsymbolic); 803 config->bsymbolicFunctions = args.hasArg(OPT_Bsymbolic_functions); 804 config->checkSections = 805 args.hasFlag(OPT_check_sections, OPT_no_check_sections, true); 806 config->chroot = args.getLastArgValue(OPT_chroot); 807 config->compressDebugSections = getCompressDebugSections(args); 808 config->cref = args.hasFlag(OPT_cref, OPT_no_cref, false); 809 config->defineCommon = args.hasFlag(OPT_define_common, OPT_no_define_common, 810 !args.hasArg(OPT_relocatable)); 811 config->demangle = args.hasFlag(OPT_demangle, OPT_no_demangle, true); 812 config->dependentLibraries = args.hasFlag(OPT_dependent_libraries, OPT_no_dependent_libraries, true); 813 config->disableVerify = args.hasArg(OPT_disable_verify); 814 config->discard = getDiscard(args); 815 config->dwoDir = args.getLastArgValue(OPT_plugin_opt_dwo_dir_eq); 816 config->dynamicLinker = getDynamicLinker(args); 817 config->ehFrameHdr = 818 args.hasFlag(OPT_eh_frame_hdr, OPT_no_eh_frame_hdr, false); 819 config->emitLLVM = args.hasArg(OPT_plugin_opt_emit_llvm, false); 820 config->emitRelocs = args.hasArg(OPT_emit_relocs); 821 config->callGraphProfileSort = args.hasFlag( 822 OPT_call_graph_profile_sort, OPT_no_call_graph_profile_sort, true); 823 config->enableNewDtags = 824 args.hasFlag(OPT_enable_new_dtags, OPT_disable_new_dtags, true); 825 config->entry = args.getLastArgValue(OPT_entry); 826 config->executeOnly = 827 args.hasFlag(OPT_execute_only, OPT_no_execute_only, false); 828 config->exportDynamic = 829 args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false); 830 config->filterList = args::getStrings(args, OPT_filter); 831 config->fini = args.getLastArgValue(OPT_fini, "_fini"); 832 config->fixCortexA53Errata843419 = args.hasArg(OPT_fix_cortex_a53_843419); 833 config->forceBTI = args.hasArg(OPT_force_bti); 834 config->requireCET = args.hasArg(OPT_require_cet); 835 config->gcSections = args.hasFlag(OPT_gc_sections, OPT_no_gc_sections, false); 836 config->gnuUnique = args.hasFlag(OPT_gnu_unique, OPT_no_gnu_unique, true); 837 config->gdbIndex = args.hasFlag(OPT_gdb_index, OPT_no_gdb_index, false); 838 config->icf = getICF(args); 839 config->ignoreDataAddressEquality = 840 args.hasArg(OPT_ignore_data_address_equality); 841 config->ignoreFunctionAddressEquality = 842 args.hasArg(OPT_ignore_function_address_equality); 843 config->init = args.getLastArgValue(OPT_init, "_init"); 844 config->ltoAAPipeline = args.getLastArgValue(OPT_lto_aa_pipeline); 845 config->ltoCSProfileGenerate = args.hasArg(OPT_lto_cs_profile_generate); 846 config->ltoCSProfileFile = args.getLastArgValue(OPT_lto_cs_profile_file); 847 config->ltoDebugPassManager = args.hasArg(OPT_lto_debug_pass_manager); 848 config->ltoNewPassManager = args.hasArg(OPT_lto_new_pass_manager); 849 config->ltoNewPmPasses = args.getLastArgValue(OPT_lto_newpm_passes); 850 config->ltoo = args::getInteger(args, OPT_lto_O, 2); 851 config->ltoObjPath = args.getLastArgValue(OPT_plugin_opt_obj_path_eq); 852 config->ltoPartitions = args::getInteger(args, OPT_lto_partitions, 1); 853 config->ltoSampleProfile = args.getLastArgValue(OPT_lto_sample_profile); 854 config->mapFile = args.getLastArgValue(OPT_Map); 855 config->mipsGotSize = args::getInteger(args, OPT_mips_got_size, 0xfff0); 856 config->mergeArmExidx = 857 args.hasFlag(OPT_merge_exidx_entries, OPT_no_merge_exidx_entries, true); 858 config->nmagic = args.hasFlag(OPT_nmagic, OPT_no_nmagic, false); 859 config->noinhibitExec = args.hasArg(OPT_noinhibit_exec); 860 config->nostdlib = args.hasArg(OPT_nostdlib); 861 config->oFormatBinary = isOutputFormatBinary(args); 862 config->omagic = args.hasFlag(OPT_omagic, OPT_no_omagic, false); 863 config->optRemarksFilename = args.getLastArgValue(OPT_opt_remarks_filename); 864 config->optRemarksPasses = args.getLastArgValue(OPT_opt_remarks_passes); 865 config->optRemarksWithHotness = args.hasArg(OPT_opt_remarks_with_hotness); 866 config->optRemarksFormat = args.getLastArgValue(OPT_opt_remarks_format); 867 config->optimize = args::getInteger(args, OPT_O, 1); 868 config->orphanHandling = getOrphanHandling(args); 869 config->outputFile = args.getLastArgValue(OPT_o); 870 config->pacPlt = args.hasArg(OPT_pac_plt); 871 config->pie = args.hasFlag(OPT_pie, OPT_no_pie, false); 872 config->printIcfSections = 873 args.hasFlag(OPT_print_icf_sections, OPT_no_print_icf_sections, false); 874 config->printGcSections = 875 args.hasFlag(OPT_print_gc_sections, OPT_no_print_gc_sections, false); 876 config->printSymbolOrder = 877 args.getLastArgValue(OPT_print_symbol_order); 878 config->rpath = getRpath(args); 879 config->relocatable = args.hasArg(OPT_relocatable); 880 config->saveTemps = args.hasArg(OPT_save_temps); 881 config->searchPaths = args::getStrings(args, OPT_library_path); 882 config->sectionStartMap = getSectionStartMap(args); 883 config->shared = args.hasArg(OPT_shared); 884 config->singleRoRx = args.hasArg(OPT_no_rosegment); 885 config->soName = args.getLastArgValue(OPT_soname); 886 config->sortSection = getSortSection(args); 887 config->splitStackAdjustSize = args::getInteger(args, OPT_split_stack_adjust_size, 16384); 888 config->strip = getStrip(args); 889 config->sysroot = args.getLastArgValue(OPT_sysroot); 890 config->target1Rel = args.hasFlag(OPT_target1_rel, OPT_target1_abs, false); 891 config->target2 = getTarget2(args); 892 config->thinLTOCacheDir = args.getLastArgValue(OPT_thinlto_cache_dir); 893 config->thinLTOCachePolicy = CHECK( 894 parseCachePruningPolicy(args.getLastArgValue(OPT_thinlto_cache_policy)), 895 "--thinlto-cache-policy: invalid cache policy"); 896 config->thinLTOEmitImportsFiles = 897 args.hasArg(OPT_plugin_opt_thinlto_emit_imports_files); 898 config->thinLTOIndexOnly = args.hasArg(OPT_plugin_opt_thinlto_index_only) || 899 args.hasArg(OPT_plugin_opt_thinlto_index_only_eq); 900 config->thinLTOIndexOnlyArg = 901 args.getLastArgValue(OPT_plugin_opt_thinlto_index_only_eq); 902 config->thinLTOJobs = args::getInteger(args, OPT_thinlto_jobs, -1u); 903 config->thinLTOObjectSuffixReplace = 904 getOldNewOptions(args, OPT_plugin_opt_thinlto_object_suffix_replace_eq); 905 config->thinLTOPrefixReplace = 906 getOldNewOptions(args, OPT_plugin_opt_thinlto_prefix_replace_eq); 907 config->trace = args.hasArg(OPT_trace); 908 config->undefined = args::getStrings(args, OPT_undefined); 909 config->undefinedVersion = 910 args.hasFlag(OPT_undefined_version, OPT_no_undefined_version, true); 911 config->useAndroidRelrTags = args.hasFlag( 912 OPT_use_android_relr_tags, OPT_no_use_android_relr_tags, false); 913 config->unresolvedSymbols = getUnresolvedSymbolPolicy(args); 914 config->warnBackrefs = 915 args.hasFlag(OPT_warn_backrefs, OPT_no_warn_backrefs, false); 916 config->warnCommon = args.hasFlag(OPT_warn_common, OPT_no_warn_common, false); 917 config->warnIfuncTextrel = 918 args.hasFlag(OPT_warn_ifunc_textrel, OPT_no_warn_ifunc_textrel, false); 919 config->warnSymbolOrdering = 920 args.hasFlag(OPT_warn_symbol_ordering, OPT_no_warn_symbol_ordering, true); 921 config->zCombreloc = getZFlag(args, "combreloc", "nocombreloc", true); 922 config->zCopyreloc = getZFlag(args, "copyreloc", "nocopyreloc", true); 923 config->zExecstack = getZFlag(args, "execstack", "noexecstack", false); 924 config->zGlobal = hasZOption(args, "global"); 925 config->zHazardplt = hasZOption(args, "hazardplt"); 926 config->zIfuncNoplt = hasZOption(args, "ifunc-noplt"); 927 config->zInitfirst = hasZOption(args, "initfirst"); 928 config->zInterpose = hasZOption(args, "interpose"); 929 config->zKeepTextSectionPrefix = getZFlag( 930 args, "keep-text-section-prefix", "nokeep-text-section-prefix", false); 931 config->zNodefaultlib = hasZOption(args, "nodefaultlib"); 932 config->zNodelete = hasZOption(args, "nodelete"); 933 config->zNodlopen = hasZOption(args, "nodlopen"); 934 config->zNow = getZFlag(args, "now", "lazy", false); 935 config->zOrigin = hasZOption(args, "origin"); 936 config->zRelro = getZFlag(args, "relro", "norelro", true); 937 config->zRetpolineplt = hasZOption(args, "retpolineplt"); 938 config->zRodynamic = hasZOption(args, "rodynamic"); 939 config->zSeparateCode = getZFlag(args, "separate-code", "noseparate-code", false); 940 config->zStackSize = args::getZOptionValue(args, OPT_z, "stack-size", 0); 941 config->zText = getZFlag(args, "text", "notext", true); 942 config->zWxneeded = hasZOption(args, "wxneeded"); 943 944 // Parse LTO options. 945 if (auto *arg = args.getLastArg(OPT_plugin_opt_mcpu_eq)) 946 parseClangOption(saver.save("-mcpu=" + StringRef(arg->getValue())), 947 arg->getSpelling()); 948 949 for (auto *arg : args.filtered(OPT_plugin_opt)) 950 parseClangOption(arg->getValue(), arg->getSpelling()); 951 952 // Parse -mllvm options. 953 for (auto *arg : args.filtered(OPT_mllvm)) 954 parseClangOption(arg->getValue(), arg->getSpelling()); 955 956 if (config->ltoo > 3) 957 error("invalid optimization level for LTO: " + Twine(config->ltoo)); 958 if (config->ltoPartitions == 0) 959 error("--lto-partitions: number of threads must be > 0"); 960 if (config->thinLTOJobs == 0) 961 error("--thinlto-jobs: number of threads must be > 0"); 962 963 if (config->splitStackAdjustSize < 0) 964 error("--split-stack-adjust-size: size must be >= 0"); 965 966 // Parse ELF{32,64}{LE,BE} and CPU type. 967 if (auto *arg = args.getLastArg(OPT_m)) { 968 StringRef s = arg->getValue(); 969 std::tie(config->ekind, config->emachine, config->osabi) = 970 parseEmulation(s); 971 config->mipsN32Abi = (s == "elf32btsmipn32" || s == "elf32ltsmipn32"); 972 config->emulation = s; 973 } 974 975 // Parse -hash-style={sysv,gnu,both}. 976 if (auto *arg = args.getLastArg(OPT_hash_style)) { 977 StringRef s = arg->getValue(); 978 if (s == "sysv") 979 config->sysvHash = true; 980 else if (s == "gnu") 981 config->gnuHash = true; 982 else if (s == "both") 983 config->sysvHash = config->gnuHash = true; 984 else 985 error("unknown -hash-style: " + s); 986 } 987 988 if (args.hasArg(OPT_print_map)) 989 config->mapFile = "-"; 990 991 // Page alignment can be disabled by the -n (--nmagic) and -N (--omagic). 992 // As PT_GNU_RELRO relies on Paging, do not create it when we have disabled 993 // it. 994 if (config->nmagic || config->omagic) 995 config->zRelro = false; 996 997 std::tie(config->buildId, config->buildIdVector) = getBuildId(args); 998 999 std::tie(config->androidPackDynRelocs, config->relrPackDynRelocs) = 1000 getPackDynRelocs(args); 1001 1002 if (auto *arg = args.getLastArg(OPT_symbol_ordering_file)){ 1003 if (args.hasArg(OPT_call_graph_ordering_file)) 1004 error("--symbol-ordering-file and --call-graph-order-file " 1005 "may not be used together"); 1006 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())){ 1007 config->symbolOrderingFile = getSymbolOrderingFile(*buffer); 1008 // Also need to disable CallGraphProfileSort to prevent 1009 // LLD order symbols with CGProfile 1010 config->callGraphProfileSort = false; 1011 } 1012 } 1013 1014 // If --retain-symbol-file is used, we'll keep only the symbols listed in 1015 // the file and discard all others. 1016 if (auto *arg = args.getLastArg(OPT_retain_symbols_file)) { 1017 config->defaultSymbolVersion = VER_NDX_LOCAL; 1018 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 1019 for (StringRef s : args::getLines(*buffer)) 1020 config->versionScriptGlobals.push_back( 1021 {s, /*IsExternCpp*/ false, /*HasWildcard*/ false}); 1022 } 1023 1024 bool hasExportDynamic = 1025 args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false); 1026 1027 // Parses -dynamic-list and -export-dynamic-symbol. They make some 1028 // symbols private. Note that -export-dynamic takes precedence over them 1029 // as it says all symbols should be exported. 1030 if (!hasExportDynamic) { 1031 for (auto *arg : args.filtered(OPT_dynamic_list)) 1032 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 1033 readDynamicList(*buffer); 1034 1035 for (auto *arg : args.filtered(OPT_export_dynamic_symbol)) 1036 config->dynamicList.push_back( 1037 {arg->getValue(), /*IsExternCpp*/ false, /*HasWildcard*/ false}); 1038 } 1039 1040 // If --export-dynamic-symbol=foo is given and symbol foo is defined in 1041 // an object file in an archive file, that object file should be pulled 1042 // out and linked. (It doesn't have to behave like that from technical 1043 // point of view, but this is needed for compatibility with GNU.) 1044 for (auto *arg : args.filtered(OPT_export_dynamic_symbol)) 1045 config->undefined.push_back(arg->getValue()); 1046 1047 for (auto *arg : args.filtered(OPT_version_script)) 1048 if (Optional<std::string> path = searchScript(arg->getValue())) { 1049 if (Optional<MemoryBufferRef> buffer = readFile(*path)) 1050 readVersionScript(*buffer); 1051 } else { 1052 error(Twine("cannot find version script ") + arg->getValue()); 1053 } 1054 } 1055 1056 // Some Config members do not directly correspond to any particular 1057 // command line options, but computed based on other Config values. 1058 // This function initialize such members. See Config.h for the details 1059 // of these values. 1060 static void setConfigs(opt::InputArgList &args) { 1061 ELFKind k = config->ekind; 1062 uint16_t m = config->emachine; 1063 1064 config->copyRelocs = (config->relocatable || config->emitRelocs); 1065 config->is64 = (k == ELF64LEKind || k == ELF64BEKind); 1066 config->isLE = (k == ELF32LEKind || k == ELF64LEKind); 1067 config->endianness = config->isLE ? endianness::little : endianness::big; 1068 config->isMips64EL = (k == ELF64LEKind && m == EM_MIPS); 1069 config->isPic = config->pie || config->shared; 1070 config->picThunk = args.hasArg(OPT_pic_veneer, config->isPic); 1071 config->wordsize = config->is64 ? 8 : 4; 1072 1073 // ELF defines two different ways to store relocation addends as shown below: 1074 // 1075 // Rel: Addends are stored to the location where relocations are applied. 1076 // Rela: Addends are stored as part of relocation entry. 1077 // 1078 // In other words, Rela makes it easy to read addends at the price of extra 1079 // 4 or 8 byte for each relocation entry. We don't know why ELF defined two 1080 // different mechanisms in the first place, but this is how the spec is 1081 // defined. 1082 // 1083 // You cannot choose which one, Rel or Rela, you want to use. Instead each 1084 // ABI defines which one you need to use. The following expression expresses 1085 // that. 1086 config->isRela = m == EM_AARCH64 || m == EM_AMDGPU || m == EM_HEXAGON || 1087 m == EM_PPC || m == EM_PPC64 || m == EM_RISCV || 1088 m == EM_X86_64; 1089 1090 // If the output uses REL relocations we must store the dynamic relocation 1091 // addends to the output sections. We also store addends for RELA relocations 1092 // if --apply-dynamic-relocs is used. 1093 // We default to not writing the addends when using RELA relocations since 1094 // any standard conforming tool can find it in r_addend. 1095 config->writeAddends = args.hasFlag(OPT_apply_dynamic_relocs, 1096 OPT_no_apply_dynamic_relocs, false) || 1097 !config->isRela; 1098 1099 config->tocOptimize = 1100 args.hasFlag(OPT_toc_optimize, OPT_no_toc_optimize, m == EM_PPC64); 1101 } 1102 1103 // Returns a value of "-format" option. 1104 static bool isFormatBinary(StringRef s) { 1105 if (s == "binary") 1106 return true; 1107 if (s == "elf" || s == "default") 1108 return false; 1109 error("unknown -format value: " + s + 1110 " (supported formats: elf, default, binary)"); 1111 return false; 1112 } 1113 1114 void LinkerDriver::createFiles(opt::InputArgList &args) { 1115 // For --{push,pop}-state. 1116 std::vector<std::tuple<bool, bool, bool>> stack; 1117 1118 // Iterate over argv to process input files and positional arguments. 1119 for (auto *arg : args) { 1120 switch (arg->getOption().getID()) { 1121 case OPT_library: 1122 addLibrary(arg->getValue()); 1123 break; 1124 case OPT_INPUT: 1125 addFile(arg->getValue(), /*withLOption=*/false); 1126 break; 1127 case OPT_defsym: { 1128 StringRef from; 1129 StringRef to; 1130 std::tie(from, to) = StringRef(arg->getValue()).split('='); 1131 if (from.empty() || to.empty()) 1132 error("-defsym: syntax error: " + StringRef(arg->getValue())); 1133 else 1134 readDefsym(from, MemoryBufferRef(to, "-defsym")); 1135 break; 1136 } 1137 case OPT_script: 1138 if (Optional<std::string> path = searchScript(arg->getValue())) { 1139 if (Optional<MemoryBufferRef> mb = readFile(*path)) 1140 readLinkerScript(*mb); 1141 break; 1142 } 1143 error(Twine("cannot find linker script ") + arg->getValue()); 1144 break; 1145 case OPT_as_needed: 1146 config->asNeeded = true; 1147 break; 1148 case OPT_format: 1149 config->formatBinary = isFormatBinary(arg->getValue()); 1150 break; 1151 case OPT_no_as_needed: 1152 config->asNeeded = false; 1153 break; 1154 case OPT_Bstatic: 1155 case OPT_omagic: 1156 case OPT_nmagic: 1157 config->isStatic = true; 1158 break; 1159 case OPT_Bdynamic: 1160 config->isStatic = false; 1161 break; 1162 case OPT_whole_archive: 1163 inWholeArchive = true; 1164 break; 1165 case OPT_no_whole_archive: 1166 inWholeArchive = false; 1167 break; 1168 case OPT_just_symbols: 1169 if (Optional<MemoryBufferRef> mb = readFile(arg->getValue())) { 1170 files.push_back(createObjectFile(*mb)); 1171 files.back()->justSymbols = true; 1172 } 1173 break; 1174 case OPT_start_group: 1175 if (InputFile::isInGroup) 1176 error("nested --start-group"); 1177 InputFile::isInGroup = true; 1178 break; 1179 case OPT_end_group: 1180 if (!InputFile::isInGroup) 1181 error("stray --end-group"); 1182 InputFile::isInGroup = false; 1183 ++InputFile::nextGroupId; 1184 break; 1185 case OPT_start_lib: 1186 if (inLib) 1187 error("nested --start-lib"); 1188 if (InputFile::isInGroup) 1189 error("may not nest --start-lib in --start-group"); 1190 inLib = true; 1191 InputFile::isInGroup = true; 1192 break; 1193 case OPT_end_lib: 1194 if (!inLib) 1195 error("stray --end-lib"); 1196 inLib = false; 1197 InputFile::isInGroup = false; 1198 ++InputFile::nextGroupId; 1199 break; 1200 case OPT_push_state: 1201 stack.emplace_back(config->asNeeded, config->isStatic, inWholeArchive); 1202 break; 1203 case OPT_pop_state: 1204 if (stack.empty()) { 1205 error("unbalanced --push-state/--pop-state"); 1206 break; 1207 } 1208 std::tie(config->asNeeded, config->isStatic, inWholeArchive) = stack.back(); 1209 stack.pop_back(); 1210 break; 1211 } 1212 } 1213 1214 if (files.empty() && errorCount() == 0) 1215 error("no input files"); 1216 } 1217 1218 // If -m <machine_type> was not given, infer it from object files. 1219 void LinkerDriver::inferMachineType() { 1220 if (config->ekind != ELFNoneKind) 1221 return; 1222 1223 for (InputFile *f : files) { 1224 if (f->ekind == ELFNoneKind) 1225 continue; 1226 config->ekind = f->ekind; 1227 config->emachine = f->emachine; 1228 config->osabi = f->osabi; 1229 config->mipsN32Abi = config->emachine == EM_MIPS && isMipsN32Abi(f); 1230 return; 1231 } 1232 error("target emulation unknown: -m or at least one .o file required"); 1233 } 1234 1235 // Parse -z max-page-size=<value>. The default value is defined by 1236 // each target. 1237 static uint64_t getMaxPageSize(opt::InputArgList &args) { 1238 uint64_t val = args::getZOptionValue(args, OPT_z, "max-page-size", 1239 target->defaultMaxPageSize); 1240 if (!isPowerOf2_64(val)) 1241 error("max-page-size: value isn't a power of 2"); 1242 if (config->nmagic || config->omagic) { 1243 if (val != target->defaultMaxPageSize) 1244 warn("-z max-page-size set, but paging disabled by omagic or nmagic"); 1245 return 1; 1246 } 1247 return val; 1248 } 1249 1250 // Parse -z common-page-size=<value>. The default value is defined by 1251 // each target. 1252 static uint64_t getCommonPageSize(opt::InputArgList &args) { 1253 uint64_t val = args::getZOptionValue(args, OPT_z, "common-page-size", 1254 target->defaultCommonPageSize); 1255 if (!isPowerOf2_64(val)) 1256 error("common-page-size: value isn't a power of 2"); 1257 if (config->nmagic || config->omagic) { 1258 if (val != target->defaultCommonPageSize) 1259 warn("-z common-page-size set, but paging disabled by omagic or nmagic"); 1260 return 1; 1261 } 1262 // commonPageSize can't be larger than maxPageSize. 1263 if (val > config->maxPageSize) 1264 val = config->maxPageSize; 1265 return val; 1266 } 1267 1268 // Parses -image-base option. 1269 static Optional<uint64_t> getImageBase(opt::InputArgList &args) { 1270 // Because we are using "Config->maxPageSize" here, this function has to be 1271 // called after the variable is initialized. 1272 auto *arg = args.getLastArg(OPT_image_base); 1273 if (!arg) 1274 return None; 1275 1276 StringRef s = arg->getValue(); 1277 uint64_t v; 1278 if (!to_integer(s, v)) { 1279 error("-image-base: number expected, but got " + s); 1280 return 0; 1281 } 1282 if ((v % config->maxPageSize) != 0) 1283 warn("-image-base: address isn't multiple of page size: " + s); 1284 return v; 1285 } 1286 1287 // Parses `--exclude-libs=lib,lib,...`. 1288 // The library names may be delimited by commas or colons. 1289 static DenseSet<StringRef> getExcludeLibs(opt::InputArgList &args) { 1290 DenseSet<StringRef> ret; 1291 for (auto *arg : args.filtered(OPT_exclude_libs)) { 1292 StringRef s = arg->getValue(); 1293 for (;;) { 1294 size_t pos = s.find_first_of(",:"); 1295 if (pos == StringRef::npos) 1296 break; 1297 ret.insert(s.substr(0, pos)); 1298 s = s.substr(pos + 1); 1299 } 1300 ret.insert(s); 1301 } 1302 return ret; 1303 } 1304 1305 // Handles the -exclude-libs option. If a static library file is specified 1306 // by the -exclude-libs option, all public symbols from the archive become 1307 // private unless otherwise specified by version scripts or something. 1308 // A special library name "ALL" means all archive files. 1309 // 1310 // This is not a popular option, but some programs such as bionic libc use it. 1311 static void excludeLibs(opt::InputArgList &args) { 1312 DenseSet<StringRef> libs = getExcludeLibs(args); 1313 bool all = libs.count("ALL"); 1314 1315 auto visit = [&](InputFile *file) { 1316 if (!file->archiveName.empty()) 1317 if (all || libs.count(path::filename(file->archiveName))) 1318 for (Symbol *sym : file->getSymbols()) 1319 if (!sym->isLocal() && sym->file == file) 1320 sym->versionId = VER_NDX_LOCAL; 1321 }; 1322 1323 for (InputFile *file : objectFiles) 1324 visit(file); 1325 1326 for (BitcodeFile *file : bitcodeFiles) 1327 visit(file); 1328 } 1329 1330 // Force Sym to be entered in the output. Used for -u or equivalent. 1331 static void handleUndefined(Symbol *sym) { 1332 // Since a symbol may not be used inside the program, LTO may 1333 // eliminate it. Mark the symbol as "used" to prevent it. 1334 sym->isUsedInRegularObj = true; 1335 1336 if (sym->isLazy()) 1337 sym->fetch(); 1338 } 1339 1340 // As an extention to GNU linkers, lld supports a variant of `-u` 1341 // which accepts wildcard patterns. All symbols that match a given 1342 // pattern are handled as if they were given by `-u`. 1343 static void handleUndefinedGlob(StringRef arg) { 1344 Expected<GlobPattern> pat = GlobPattern::create(arg); 1345 if (!pat) { 1346 error("--undefined-glob: " + toString(pat.takeError())); 1347 return; 1348 } 1349 1350 std::vector<Symbol *> syms; 1351 symtab->forEachSymbol([&](Symbol *sym) { 1352 // Calling Sym->fetch() from here is not safe because it may 1353 // add new symbols to the symbol table, invalidating the 1354 // current iterator. So we just keep a note. 1355 if (pat->match(sym->getName())) 1356 syms.push_back(sym); 1357 }); 1358 1359 for (Symbol *sym : syms) 1360 handleUndefined(sym); 1361 } 1362 1363 static void handleLibcall(StringRef name) { 1364 Symbol *sym = symtab->find(name); 1365 if (!sym || !sym->isLazy()) 1366 return; 1367 1368 MemoryBufferRef mb; 1369 if (auto *lo = dyn_cast<LazyObject>(sym)) 1370 mb = lo->file->mb; 1371 else 1372 mb = cast<LazyArchive>(sym)->getMemberBuffer(); 1373 1374 if (isBitcode(mb)) 1375 sym->fetch(); 1376 } 1377 1378 // Replaces common symbols with defined symbols reside in .bss sections. 1379 // This function is called after all symbol names are resolved. As a 1380 // result, the passes after the symbol resolution won't see any 1381 // symbols of type CommonSymbol. 1382 static void replaceCommonSymbols() { 1383 symtab->forEachSymbol([](Symbol *sym) { 1384 auto *s = dyn_cast<CommonSymbol>(sym); 1385 if (!s) 1386 return; 1387 1388 auto *bss = make<BssSection>("COMMON", s->size, s->alignment); 1389 bss->file = s->file; 1390 bss->markDead(); 1391 inputSections.push_back(bss); 1392 s->replace(Defined{s->file, s->getName(), s->binding, s->stOther, s->type, 1393 /*value=*/0, s->size, bss}); 1394 }); 1395 } 1396 1397 // If all references to a DSO happen to be weak, the DSO is not added 1398 // to DT_NEEDED. If that happens, we need to eliminate shared symbols 1399 // created from the DSO. Otherwise, they become dangling references 1400 // that point to a non-existent DSO. 1401 static void demoteSharedSymbols() { 1402 symtab->forEachSymbol([](Symbol *sym) { 1403 auto *s = dyn_cast<SharedSymbol>(sym); 1404 if (!s || s->getFile().isNeeded) 1405 return; 1406 1407 bool used = s->used; 1408 s->replace(Undefined{nullptr, s->getName(), STB_WEAK, s->stOther, s->type}); 1409 s->used = used; 1410 }); 1411 } 1412 1413 // The section referred to by `s` is considered address-significant. Set the 1414 // keepUnique flag on the section if appropriate. 1415 static void markAddrsig(Symbol *s) { 1416 if (auto *d = dyn_cast_or_null<Defined>(s)) 1417 if (d->section) 1418 // We don't need to keep text sections unique under --icf=all even if they 1419 // are address-significant. 1420 if (config->icf == ICFLevel::Safe || !(d->section->flags & SHF_EXECINSTR)) 1421 d->section->keepUnique = true; 1422 } 1423 1424 // Record sections that define symbols mentioned in --keep-unique <symbol> 1425 // and symbols referred to by address-significance tables. These sections are 1426 // ineligible for ICF. 1427 template <class ELFT> 1428 static void findKeepUniqueSections(opt::InputArgList &args) { 1429 for (auto *arg : args.filtered(OPT_keep_unique)) { 1430 StringRef name = arg->getValue(); 1431 auto *d = dyn_cast_or_null<Defined>(symtab->find(name)); 1432 if (!d || !d->section) { 1433 warn("could not find symbol " + name + " to keep unique"); 1434 continue; 1435 } 1436 d->section->keepUnique = true; 1437 } 1438 1439 // --icf=all --ignore-data-address-equality means that we can ignore 1440 // the dynsym and address-significance tables entirely. 1441 if (config->icf == ICFLevel::All && config->ignoreDataAddressEquality) 1442 return; 1443 1444 // Symbols in the dynsym could be address-significant in other executables 1445 // or DSOs, so we conservatively mark them as address-significant. 1446 symtab->forEachSymbol([&](Symbol *sym) { 1447 if (sym->includeInDynsym()) 1448 markAddrsig(sym); 1449 }); 1450 1451 // Visit the address-significance table in each object file and mark each 1452 // referenced symbol as address-significant. 1453 for (InputFile *f : objectFiles) { 1454 auto *obj = cast<ObjFile<ELFT>>(f); 1455 ArrayRef<Symbol *> syms = obj->getSymbols(); 1456 if (obj->addrsigSec) { 1457 ArrayRef<uint8_t> contents = 1458 check(obj->getObj().getSectionContents(obj->addrsigSec)); 1459 const uint8_t *cur = contents.begin(); 1460 while (cur != contents.end()) { 1461 unsigned size; 1462 const char *err; 1463 uint64_t symIndex = decodeULEB128(cur, &size, contents.end(), &err); 1464 if (err) 1465 fatal(toString(f) + ": could not decode addrsig section: " + err); 1466 markAddrsig(syms[symIndex]); 1467 cur += size; 1468 } 1469 } else { 1470 // If an object file does not have an address-significance table, 1471 // conservatively mark all of its symbols as address-significant. 1472 for (Symbol *s : syms) 1473 markAddrsig(s); 1474 } 1475 } 1476 } 1477 1478 // This function reads a symbol partition specification section. These sections 1479 // are used to control which partition a symbol is allocated to. See 1480 // https://lld.llvm.org/Partitions.html for more details on partitions. 1481 template <typename ELFT> 1482 static void readSymbolPartitionSection(InputSectionBase *s) { 1483 // Read the relocation that refers to the partition's entry point symbol. 1484 Symbol *sym; 1485 if (s->areRelocsRela) 1486 sym = &s->getFile<ELFT>()->getRelocTargetSym(s->template relas<ELFT>()[0]); 1487 else 1488 sym = &s->getFile<ELFT>()->getRelocTargetSym(s->template rels<ELFT>()[0]); 1489 if (!isa<Defined>(sym) || !sym->includeInDynsym()) 1490 return; 1491 1492 StringRef partName = reinterpret_cast<const char *>(s->data().data()); 1493 for (Partition &part : partitions) { 1494 if (part.name == partName) { 1495 sym->partition = part.getNumber(); 1496 return; 1497 } 1498 } 1499 1500 // Forbid partitions from being used on incompatible targets, and forbid them 1501 // from being used together with various linker features that assume a single 1502 // set of output sections. 1503 if (script->hasSectionsCommand) 1504 error(toString(s->file) + 1505 ": partitions cannot be used with the SECTIONS command"); 1506 if (script->hasPhdrsCommands()) 1507 error(toString(s->file) + 1508 ": partitions cannot be used with the PHDRS command"); 1509 if (!config->sectionStartMap.empty()) 1510 error(toString(s->file) + ": partitions cannot be used with " 1511 "--section-start, -Ttext, -Tdata or -Tbss"); 1512 if (config->emachine == EM_MIPS) 1513 error(toString(s->file) + ": partitions cannot be used on this target"); 1514 1515 // Impose a limit of no more than 254 partitions. This limit comes from the 1516 // sizes of the Partition fields in InputSectionBase and Symbol, as well as 1517 // the amount of space devoted to the partition number in RankFlags. 1518 if (partitions.size() == 254) 1519 fatal("may not have more than 254 partitions"); 1520 1521 partitions.emplace_back(); 1522 Partition &newPart = partitions.back(); 1523 newPart.name = partName; 1524 sym->partition = newPart.getNumber(); 1525 } 1526 1527 static Symbol *addUndefined(StringRef name) { 1528 return symtab->addSymbol( 1529 Undefined{nullptr, name, STB_GLOBAL, STV_DEFAULT, 0}); 1530 } 1531 1532 // This function is where all the optimizations of link-time 1533 // optimization takes place. When LTO is in use, some input files are 1534 // not in native object file format but in the LLVM bitcode format. 1535 // This function compiles bitcode files into a few big native files 1536 // using LLVM functions and replaces bitcode symbols with the results. 1537 // Because all bitcode files that the program consists of are passed to 1538 // the compiler at once, it can do a whole-program optimization. 1539 template <class ELFT> void LinkerDriver::compileBitcodeFiles() { 1540 // Compile bitcode files and replace bitcode symbols. 1541 lto.reset(new BitcodeCompiler); 1542 for (BitcodeFile *file : bitcodeFiles) 1543 lto->add(*file); 1544 1545 for (InputFile *file : lto->compile()) { 1546 auto *obj = cast<ObjFile<ELFT>>(file); 1547 obj->parse(/*ignoreComdats=*/true); 1548 for (Symbol *sym : obj->getGlobalSymbols()) 1549 sym->parseSymbolVersion(); 1550 objectFiles.push_back(file); 1551 } 1552 } 1553 1554 // The --wrap option is a feature to rename symbols so that you can write 1555 // wrappers for existing functions. If you pass `-wrap=foo`, all 1556 // occurrences of symbol `foo` are resolved to `wrap_foo` (so, you are 1557 // expected to write `wrap_foo` function as a wrapper). The original 1558 // symbol becomes accessible as `real_foo`, so you can call that from your 1559 // wrapper. 1560 // 1561 // This data structure is instantiated for each -wrap option. 1562 struct WrappedSymbol { 1563 Symbol *sym; 1564 Symbol *real; 1565 Symbol *wrap; 1566 }; 1567 1568 // Handles -wrap option. 1569 // 1570 // This function instantiates wrapper symbols. At this point, they seem 1571 // like they are not being used at all, so we explicitly set some flags so 1572 // that LTO won't eliminate them. 1573 static std::vector<WrappedSymbol> addWrappedSymbols(opt::InputArgList &args) { 1574 std::vector<WrappedSymbol> v; 1575 DenseSet<StringRef> seen; 1576 1577 for (auto *arg : args.filtered(OPT_wrap)) { 1578 StringRef name = arg->getValue(); 1579 if (!seen.insert(name).second) 1580 continue; 1581 1582 Symbol *sym = symtab->find(name); 1583 if (!sym) 1584 continue; 1585 1586 Symbol *real = addUndefined(saver.save("__real_" + name)); 1587 Symbol *wrap = addUndefined(saver.save("__wrap_" + name)); 1588 v.push_back({sym, real, wrap}); 1589 1590 // We want to tell LTO not to inline symbols to be overwritten 1591 // because LTO doesn't know the final symbol contents after renaming. 1592 real->canInline = false; 1593 sym->canInline = false; 1594 1595 // Tell LTO not to eliminate these symbols. 1596 sym->isUsedInRegularObj = true; 1597 wrap->isUsedInRegularObj = true; 1598 } 1599 return v; 1600 } 1601 1602 // Do renaming for -wrap by updating pointers to symbols. 1603 // 1604 // When this function is executed, only InputFiles and symbol table 1605 // contain pointers to symbol objects. We visit them to replace pointers, 1606 // so that wrapped symbols are swapped as instructed by the command line. 1607 static void wrapSymbols(ArrayRef<WrappedSymbol> wrapped) { 1608 DenseMap<Symbol *, Symbol *> map; 1609 for (const WrappedSymbol &w : wrapped) { 1610 map[w.sym] = w.wrap; 1611 map[w.real] = w.sym; 1612 } 1613 1614 // Update pointers in input files. 1615 parallelForEach(objectFiles, [&](InputFile *file) { 1616 MutableArrayRef<Symbol *> syms = file->getMutableSymbols(); 1617 for (size_t i = 0, e = syms.size(); i != e; ++i) 1618 if (Symbol *s = map.lookup(syms[i])) 1619 syms[i] = s; 1620 }); 1621 1622 // Update pointers in the symbol table. 1623 for (const WrappedSymbol &w : wrapped) 1624 symtab->wrap(w.sym, w.real, w.wrap); 1625 } 1626 1627 // To enable CET (x86's hardware-assited control flow enforcement), each 1628 // source file must be compiled with -fcf-protection. Object files compiled 1629 // with the flag contain feature flags indicating that they are compatible 1630 // with CET. We enable the feature only when all object files are compatible 1631 // with CET. 1632 // 1633 // This function returns the merged feature flags. If 0, we cannot enable CET. 1634 // This is also the case with AARCH64's BTI and PAC which use the similar 1635 // GNU_PROPERTY_AARCH64_FEATURE_1_AND mechanism. 1636 // 1637 // Note that the CET-aware PLT is not implemented yet. We do error 1638 // check only. 1639 template <class ELFT> static uint32_t getAndFeatures() { 1640 if (config->emachine != EM_386 && config->emachine != EM_X86_64 && 1641 config->emachine != EM_AARCH64) 1642 return 0; 1643 1644 uint32_t ret = -1; 1645 for (InputFile *f : objectFiles) { 1646 uint32_t features = cast<ObjFile<ELFT>>(f)->andFeatures; 1647 if (config->forceBTI && !(features & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)) { 1648 warn(toString(f) + ": --force-bti: file does not have BTI property"); 1649 features |= GNU_PROPERTY_AARCH64_FEATURE_1_BTI; 1650 } else if (!features && config->requireCET) 1651 error(toString(f) + ": --require-cet: file is not compatible with CET"); 1652 ret &= features; 1653 } 1654 1655 // Force enable pointer authentication Plt, we don't warn in this case as 1656 // this does not require support in the object for correctness. 1657 if (config->pacPlt) 1658 ret |= GNU_PROPERTY_AARCH64_FEATURE_1_PAC; 1659 1660 return ret; 1661 } 1662 1663 static const char *libcallRoutineNames[] = { 1664 #define HANDLE_LIBCALL(code, name) name, 1665 #include "llvm/IR/RuntimeLibcalls.def" 1666 #undef HANDLE_LIBCALL 1667 }; 1668 1669 // Do actual linking. Note that when this function is called, 1670 // all linker scripts have already been parsed. 1671 template <class ELFT> void LinkerDriver::link(opt::InputArgList &args) { 1672 // If a -hash-style option was not given, set to a default value, 1673 // which varies depending on the target. 1674 if (!args.hasArg(OPT_hash_style)) { 1675 if (config->emachine == EM_MIPS) 1676 config->sysvHash = true; 1677 else 1678 config->sysvHash = config->gnuHash = true; 1679 } 1680 1681 // Default output filename is "a.out" by the Unix tradition. 1682 if (config->outputFile.empty()) 1683 config->outputFile = "a.out"; 1684 1685 // Fail early if the output file or map file is not writable. If a user has a 1686 // long link, e.g. due to a large LTO link, they do not wish to run it and 1687 // find that it failed because there was a mistake in their command-line. 1688 if (auto e = tryCreateFile(config->outputFile)) 1689 error("cannot open output file " + config->outputFile + ": " + e.message()); 1690 if (auto e = tryCreateFile(config->mapFile)) 1691 error("cannot open map file " + config->mapFile + ": " + e.message()); 1692 if (errorCount()) 1693 return; 1694 1695 // Use default entry point name if no name was given via the command 1696 // line nor linker scripts. For some reason, MIPS entry point name is 1697 // different from others. 1698 config->warnMissingEntry = 1699 (!config->entry.empty() || (!config->shared && !config->relocatable)); 1700 if (config->entry.empty() && !config->relocatable) 1701 config->entry = (config->emachine == EM_MIPS) ? "__start" : "_start"; 1702 1703 // Handle --trace-symbol. 1704 for (auto *arg : args.filtered(OPT_trace_symbol)) 1705 symtab->insert(arg->getValue())->traced = true; 1706 1707 // Add all files to the symbol table. This will add almost all 1708 // symbols that we need to the symbol table. This process might 1709 // add files to the link, via autolinking, these files are always 1710 // appended to the Files vector. 1711 for (size_t i = 0; i < files.size(); ++i) 1712 parseFile(files[i]); 1713 1714 // Now that we have every file, we can decide if we will need a 1715 // dynamic symbol table. 1716 // We need one if we were asked to export dynamic symbols or if we are 1717 // producing a shared library. 1718 // We also need one if any shared libraries are used and for pie executables 1719 // (probably because the dynamic linker needs it). 1720 config->hasDynSymTab = 1721 !sharedFiles.empty() || config->isPic || config->exportDynamic; 1722 1723 // Some symbols (such as __ehdr_start) are defined lazily only when there 1724 // are undefined symbols for them, so we add these to trigger that logic. 1725 for (StringRef name : script->referencedSymbols) 1726 addUndefined(name); 1727 1728 // Handle the `--undefined <sym>` options. 1729 for (StringRef arg : config->undefined) 1730 if (Symbol *sym = symtab->find(arg)) 1731 handleUndefined(sym); 1732 1733 // If an entry symbol is in a static archive, pull out that file now. 1734 if (Symbol *sym = symtab->find(config->entry)) 1735 handleUndefined(sym); 1736 1737 // Handle the `--undefined-glob <pattern>` options. 1738 for (StringRef pat : args::getStrings(args, OPT_undefined_glob)) 1739 handleUndefinedGlob(pat); 1740 1741 // If any of our inputs are bitcode files, the LTO code generator may create 1742 // references to certain library functions that might not be explicit in the 1743 // bitcode file's symbol table. If any of those library functions are defined 1744 // in a bitcode file in an archive member, we need to arrange to use LTO to 1745 // compile those archive members by adding them to the link beforehand. 1746 // 1747 // However, adding all libcall symbols to the link can have undesired 1748 // consequences. For example, the libgcc implementation of 1749 // __sync_val_compare_and_swap_8 on 32-bit ARM pulls in an .init_array entry 1750 // that aborts the program if the Linux kernel does not support 64-bit 1751 // atomics, which would prevent the program from running even if it does not 1752 // use 64-bit atomics. 1753 // 1754 // Therefore, we only add libcall symbols to the link before LTO if we have 1755 // to, i.e. if the symbol's definition is in bitcode. Any other required 1756 // libcall symbols will be added to the link after LTO when we add the LTO 1757 // object file to the link. 1758 if (!bitcodeFiles.empty()) 1759 for (const char *s : libcallRoutineNames) 1760 handleLibcall(s); 1761 1762 // Return if there were name resolution errors. 1763 if (errorCount()) 1764 return; 1765 1766 // Now when we read all script files, we want to finalize order of linker 1767 // script commands, which can be not yet final because of INSERT commands. 1768 script->processInsertCommands(); 1769 1770 // We want to declare linker script's symbols early, 1771 // so that we can version them. 1772 // They also might be exported if referenced by DSOs. 1773 script->declareSymbols(); 1774 1775 // Handle the -exclude-libs option. 1776 if (args.hasArg(OPT_exclude_libs)) 1777 excludeLibs(args); 1778 1779 // Create elfHeader early. We need a dummy section in 1780 // addReservedSymbols to mark the created symbols as not absolute. 1781 Out::elfHeader = make<OutputSection>("", 0, SHF_ALLOC); 1782 Out::elfHeader->size = sizeof(typename ELFT::Ehdr); 1783 1784 // Create wrapped symbols for -wrap option. 1785 std::vector<WrappedSymbol> wrapped = addWrappedSymbols(args); 1786 1787 // We need to create some reserved symbols such as _end. Create them. 1788 if (!config->relocatable) 1789 addReservedSymbols(); 1790 1791 // Apply version scripts. 1792 // 1793 // For a relocatable output, version scripts don't make sense, and 1794 // parsing a symbol version string (e.g. dropping "@ver1" from a symbol 1795 // name "foo@ver1") rather do harm, so we don't call this if -r is given. 1796 if (!config->relocatable) 1797 symtab->scanVersionScript(); 1798 1799 // Do link-time optimization if given files are LLVM bitcode files. 1800 // This compiles bitcode files into real object files. 1801 // 1802 // With this the symbol table should be complete. After this, no new names 1803 // except a few linker-synthesized ones will be added to the symbol table. 1804 compileBitcodeFiles<ELFT>(); 1805 if (errorCount()) 1806 return; 1807 1808 // If -thinlto-index-only is given, we should create only "index 1809 // files" and not object files. Index file creation is already done 1810 // in addCombinedLTOObject, so we are done if that's the case. 1811 if (config->thinLTOIndexOnly) 1812 return; 1813 1814 // Likewise, --plugin-opt=emit-llvm is an option to make LTO create 1815 // an output file in bitcode and exit, so that you can just get a 1816 // combined bitcode file. 1817 if (config->emitLLVM) 1818 return; 1819 1820 // Apply symbol renames for -wrap. 1821 if (!wrapped.empty()) 1822 wrapSymbols(wrapped); 1823 1824 // Now that we have a complete list of input files. 1825 // Beyond this point, no new files are added. 1826 // Aggregate all input sections into one place. 1827 for (InputFile *f : objectFiles) 1828 for (InputSectionBase *s : f->getSections()) 1829 if (s && s != &InputSection::discarded) 1830 inputSections.push_back(s); 1831 for (BinaryFile *f : binaryFiles) 1832 for (InputSectionBase *s : f->getSections()) 1833 inputSections.push_back(cast<InputSection>(s)); 1834 1835 llvm::erase_if(inputSections, [](InputSectionBase *s) { 1836 if (s->type == SHT_LLVM_SYMPART) { 1837 readSymbolPartitionSection<ELFT>(s); 1838 return true; 1839 } 1840 1841 // We do not want to emit debug sections if --strip-all 1842 // or -strip-debug are given. 1843 return config->strip != StripPolicy::None && 1844 (s->name.startswith(".debug") || s->name.startswith(".zdebug")); 1845 }); 1846 1847 // Now that the number of partitions is fixed, save a pointer to the main 1848 // partition. 1849 mainPart = &partitions[0]; 1850 1851 // Read .note.gnu.property sections from input object files which 1852 // contain a hint to tweak linker's and loader's behaviors. 1853 config->andFeatures = getAndFeatures<ELFT>(); 1854 1855 // The Target instance handles target-specific stuff, such as applying 1856 // relocations or writing a PLT section. It also contains target-dependent 1857 // values such as a default image base address. 1858 target = getTarget(); 1859 1860 config->eflags = target->calcEFlags(); 1861 // maxPageSize (sometimes called abi page size) is the maximum page size that 1862 // the output can be run on. For example if the OS can use 4k or 64k page 1863 // sizes then maxPageSize must be 64k for the output to be useable on both. 1864 // All important alignment decisions must use this value. 1865 config->maxPageSize = getMaxPageSize(args); 1866 // commonPageSize is the most common page size that the output will be run on. 1867 // For example if an OS can use 4k or 64k page sizes and 4k is more common 1868 // than 64k then commonPageSize is set to 4k. commonPageSize can be used for 1869 // optimizations such as DATA_SEGMENT_ALIGN in linker scripts. LLD's use of it 1870 // is limited to writing trap instructions on the last executable segment. 1871 config->commonPageSize = getCommonPageSize(args); 1872 1873 config->imageBase = getImageBase(args); 1874 1875 if (config->emachine == EM_ARM) { 1876 // FIXME: These warnings can be removed when lld only uses these features 1877 // when the input objects have been compiled with an architecture that 1878 // supports them. 1879 if (config->armHasBlx == false) 1880 warn("lld uses blx instruction, no object with architecture supporting " 1881 "feature detected"); 1882 } 1883 1884 // This adds a .comment section containing a version string. We have to add it 1885 // before mergeSections because the .comment section is a mergeable section. 1886 if (!config->relocatable) 1887 inputSections.push_back(createCommentSection()); 1888 1889 // Replace common symbols with regular symbols. 1890 replaceCommonSymbols(); 1891 1892 // Do size optimizations: garbage collection, merging of SHF_MERGE sections 1893 // and identical code folding. 1894 splitSections<ELFT>(); 1895 markLive<ELFT>(); 1896 demoteSharedSymbols(); 1897 mergeSections(); 1898 if (config->icf != ICFLevel::None) { 1899 findKeepUniqueSections<ELFT>(args); 1900 doIcf<ELFT>(); 1901 } 1902 1903 // Read the callgraph now that we know what was gced or icfed 1904 if (config->callGraphProfileSort) { 1905 if (auto *arg = args.getLastArg(OPT_call_graph_ordering_file)) 1906 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 1907 readCallGraph(*buffer); 1908 readCallGraphsFromObjectFiles<ELFT>(); 1909 } 1910 1911 // Write the result to the file. 1912 writeResult<ELFT>(); 1913 } 1914