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/LTO/LTO.h" 52 #include "llvm/Support/CommandLine.h" 53 #include "llvm/Support/Compression.h" 54 #include "llvm/Support/GlobPattern.h" 55 #include "llvm/Support/LEB128.h" 56 #include "llvm/Support/Path.h" 57 #include "llvm/Support/TarWriter.h" 58 #include "llvm/Support/TargetSelect.h" 59 #include "llvm/Support/TimeProfiler.h" 60 #include "llvm/Support/raw_ostream.h" 61 #include <cstdlib> 62 #include <utility> 63 64 using namespace llvm; 65 using namespace llvm::ELF; 66 using namespace llvm::object; 67 using namespace llvm::sys; 68 using namespace llvm::support; 69 70 namespace lld { 71 namespace elf { 72 73 Configuration *config; 74 LinkerDriver *driver; 75 76 static void setConfigs(opt::InputArgList &args); 77 static void readConfigs(opt::InputArgList &args); 78 79 bool link(ArrayRef<const char *> args, bool canExitEarly, raw_ostream &stdoutOS, 80 raw_ostream &stderrOS) { 81 lld::stdoutOS = &stdoutOS; 82 lld::stderrOS = &stderrOS; 83 84 errorHandler().logName = args::getFilenameWithoutExe(args[0]); 85 errorHandler().errorLimitExceededMsg = 86 "too many errors emitted, stopping now (use " 87 "-error-limit=0 to see all errors)"; 88 errorHandler().exitEarly = canExitEarly; 89 stderrOS.enable_colors(stderrOS.has_colors()); 90 91 inputSections.clear(); 92 outputSections.clear(); 93 binaryFiles.clear(); 94 bitcodeFiles.clear(); 95 objectFiles.clear(); 96 sharedFiles.clear(); 97 98 config = make<Configuration>(); 99 driver = make<LinkerDriver>(); 100 script = make<LinkerScript>(); 101 symtab = make<SymbolTable>(); 102 103 tar = nullptr; 104 memset(&in, 0, sizeof(in)); 105 106 partitions = {Partition()}; 107 108 SharedFile::vernauxNum = 0; 109 110 config->progName = args[0]; 111 112 driver->main(args); 113 114 // Exit immediately if we don't need to return to the caller. 115 // This saves time because the overhead of calling destructors 116 // for all globally-allocated objects is not negligible. 117 if (canExitEarly) 118 exitLld(errorCount() ? 1 : 0); 119 120 freeArena(); 121 return !errorCount(); 122 } 123 124 // Parses a linker -m option. 125 static std::tuple<ELFKind, uint16_t, uint8_t> parseEmulation(StringRef emul) { 126 uint8_t osabi = 0; 127 StringRef s = emul; 128 if (s.endswith("_fbsd")) { 129 s = s.drop_back(5); 130 osabi = ELFOSABI_FREEBSD; 131 } 132 133 std::pair<ELFKind, uint16_t> ret = 134 StringSwitch<std::pair<ELFKind, uint16_t>>(s) 135 .Cases("aarch64elf", "aarch64linux", "aarch64_elf64_le_vec", 136 {ELF64LEKind, EM_AARCH64}) 137 .Cases("armelf", "armelf_linux_eabi", {ELF32LEKind, EM_ARM}) 138 .Case("elf32_x86_64", {ELF32LEKind, EM_X86_64}) 139 .Cases("elf32btsmip", "elf32btsmipn32", {ELF32BEKind, EM_MIPS}) 140 .Cases("elf32ltsmip", "elf32ltsmipn32", {ELF32LEKind, EM_MIPS}) 141 .Case("elf32lriscv", {ELF32LEKind, EM_RISCV}) 142 .Cases("elf32ppc", "elf32ppclinux", {ELF32BEKind, EM_PPC}) 143 .Case("elf64btsmip", {ELF64BEKind, EM_MIPS}) 144 .Case("elf64ltsmip", {ELF64LEKind, EM_MIPS}) 145 .Case("elf64lriscv", {ELF64LEKind, EM_RISCV}) 146 .Case("elf64ppc", {ELF64BEKind, EM_PPC64}) 147 .Case("elf64lppc", {ELF64LEKind, EM_PPC64}) 148 .Cases("elf_amd64", "elf_x86_64", {ELF64LEKind, EM_X86_64}) 149 .Case("elf_i386", {ELF32LEKind, EM_386}) 150 .Case("elf_iamcu", {ELF32LEKind, EM_IAMCU}) 151 .Default({ELFNoneKind, EM_NONE}); 152 153 if (ret.first == ELFNoneKind) 154 error("unknown emulation: " + emul); 155 return std::make_tuple(ret.first, ret.second, osabi); 156 } 157 158 // Returns slices of MB by parsing MB as an archive file. 159 // Each slice consists of a member file in the archive. 160 std::vector<std::pair<MemoryBufferRef, uint64_t>> static getArchiveMembers( 161 MemoryBufferRef mb) { 162 std::unique_ptr<Archive> file = 163 CHECK(Archive::create(mb), 164 mb.getBufferIdentifier() + ": failed to parse archive"); 165 166 std::vector<std::pair<MemoryBufferRef, uint64_t>> v; 167 Error err = Error::success(); 168 bool addToTar = file->isThin() && tar; 169 for (const Archive::Child &c : file->children(err)) { 170 MemoryBufferRef mbref = 171 CHECK(c.getMemoryBufferRef(), 172 mb.getBufferIdentifier() + 173 ": could not get the buffer for a child of the archive"); 174 if (addToTar) 175 tar->append(relativeToRoot(check(c.getFullName())), mbref.getBuffer()); 176 v.push_back(std::make_pair(mbref, c.getChildOffset())); 177 } 178 if (err) 179 fatal(mb.getBufferIdentifier() + ": Archive::children failed: " + 180 toString(std::move(err))); 181 182 // Take ownership of memory buffers created for members of thin archives. 183 for (std::unique_ptr<MemoryBuffer> &mb : file->takeThinBuffers()) 184 make<std::unique_ptr<MemoryBuffer>>(std::move(mb)); 185 186 return v; 187 } 188 189 // Opens a file and create a file object. Path has to be resolved already. 190 void LinkerDriver::addFile(StringRef path, bool withLOption) { 191 using namespace sys::fs; 192 193 Optional<MemoryBufferRef> buffer = readFile(path); 194 if (!buffer.hasValue()) 195 return; 196 MemoryBufferRef mbref = *buffer; 197 198 if (config->formatBinary) { 199 files.push_back(make<BinaryFile>(mbref)); 200 return; 201 } 202 203 switch (identify_magic(mbref.getBuffer())) { 204 case file_magic::unknown: 205 readLinkerScript(mbref); 206 return; 207 case file_magic::archive: { 208 // Handle -whole-archive. 209 if (inWholeArchive) { 210 for (const auto &p : getArchiveMembers(mbref)) 211 files.push_back(createObjectFile(p.first, path, p.second)); 212 return; 213 } 214 215 std::unique_ptr<Archive> file = 216 CHECK(Archive::create(mbref), path + ": failed to parse archive"); 217 218 // If an archive file has no symbol table, it is likely that a user 219 // is attempting LTO and using a default ar command that doesn't 220 // understand the LLVM bitcode file. It is a pretty common error, so 221 // we'll handle it as if it had a symbol table. 222 if (!file->isEmpty() && !file->hasSymbolTable()) { 223 // Check if all members are bitcode files. If not, ignore, which is the 224 // default action without the LTO hack described above. 225 for (const std::pair<MemoryBufferRef, uint64_t> &p : 226 getArchiveMembers(mbref)) 227 if (identify_magic(p.first.getBuffer()) != file_magic::bitcode) { 228 error(path + ": archive has no index; run ranlib to add one"); 229 return; 230 } 231 232 for (const std::pair<MemoryBufferRef, uint64_t> &p : 233 getArchiveMembers(mbref)) 234 files.push_back(make<LazyObjFile>(p.first, path, p.second)); 235 return; 236 } 237 238 // Handle the regular case. 239 files.push_back(make<ArchiveFile>(std::move(file))); 240 return; 241 } 242 case file_magic::elf_shared_object: 243 if (config->isStatic || config->relocatable) { 244 error("attempted static link of dynamic object " + path); 245 return; 246 } 247 248 // DSOs usually have DT_SONAME tags in their ELF headers, and the 249 // sonames are used to identify DSOs. But if they are missing, 250 // they are identified by filenames. We don't know whether the new 251 // file has a DT_SONAME or not because we haven't parsed it yet. 252 // Here, we set the default soname for the file because we might 253 // need it later. 254 // 255 // If a file was specified by -lfoo, the directory part is not 256 // significant, as a user did not specify it. This behavior is 257 // compatible with GNU. 258 files.push_back( 259 make<SharedFile>(mbref, withLOption ? path::filename(path) : path)); 260 return; 261 case file_magic::bitcode: 262 case file_magic::elf_relocatable: 263 if (inLib) 264 files.push_back(make<LazyObjFile>(mbref, "", 0)); 265 else 266 files.push_back(createObjectFile(mbref)); 267 break; 268 default: 269 error(path + ": unknown file type"); 270 } 271 } 272 273 // Add a given library by searching it from input search paths. 274 void LinkerDriver::addLibrary(StringRef name) { 275 if (Optional<std::string> path = searchLibrary(name)) 276 addFile(*path, /*withLOption=*/true); 277 else 278 error("unable to find library -l" + name); 279 } 280 281 // This function is called on startup. We need this for LTO since 282 // LTO calls LLVM functions to compile bitcode files to native code. 283 // Technically this can be delayed until we read bitcode files, but 284 // we don't bother to do lazily because the initialization is fast. 285 static void initLLVM() { 286 InitializeAllTargets(); 287 InitializeAllTargetMCs(); 288 InitializeAllAsmPrinters(); 289 InitializeAllAsmParsers(); 290 } 291 292 // Some command line options or some combinations of them are not allowed. 293 // This function checks for such errors. 294 static void checkOptions() { 295 // The MIPS ABI as of 2016 does not support the GNU-style symbol lookup 296 // table which is a relatively new feature. 297 if (config->emachine == EM_MIPS && config->gnuHash) 298 error("the .gnu.hash section is not compatible with the MIPS target"); 299 300 if (config->fixCortexA53Errata843419 && config->emachine != EM_AARCH64) 301 error("--fix-cortex-a53-843419 is only supported on AArch64 targets"); 302 303 if (config->fixCortexA8 && config->emachine != EM_ARM) 304 error("--fix-cortex-a8 is only supported on ARM targets"); 305 306 if (config->tocOptimize && config->emachine != EM_PPC64) 307 error("--toc-optimize is only supported on the PowerPC64 target"); 308 309 if (config->pie && config->shared) 310 error("-shared and -pie may not be used together"); 311 312 if (!config->shared && !config->filterList.empty()) 313 error("-F may not be used without -shared"); 314 315 if (!config->shared && !config->auxiliaryList.empty()) 316 error("-f may not be used without -shared"); 317 318 if (!config->relocatable && !config->defineCommon) 319 error("-no-define-common not supported in non relocatable output"); 320 321 if (config->strip == StripPolicy::All && config->emitRelocs) 322 error("--strip-all and --emit-relocs may not be used together"); 323 324 if (config->zText && config->zIfuncNoplt) 325 error("-z text and -z ifunc-noplt may not be used together"); 326 327 if (config->relocatable) { 328 if (config->shared) 329 error("-r and -shared may not be used together"); 330 if (config->gcSections) 331 error("-r and --gc-sections may not be used together"); 332 if (config->gdbIndex) 333 error("-r and --gdb-index may not be used together"); 334 if (config->icf != ICFLevel::None) 335 error("-r and --icf may not be used together"); 336 if (config->pie) 337 error("-r and -pie may not be used together"); 338 if (config->exportDynamic) 339 error("-r and --export-dynamic may not be used together"); 340 } 341 342 if (config->executeOnly) { 343 if (config->emachine != EM_AARCH64) 344 error("-execute-only is only supported on AArch64 targets"); 345 346 if (config->singleRoRx && !script->hasSectionsCommand) 347 error("-execute-only and -no-rosegment cannot be used together"); 348 } 349 350 if (config->zRetpolineplt && config->zForceIbt) 351 error("-z force-ibt may not be used with -z retpolineplt"); 352 353 if (config->emachine != EM_AARCH64) { 354 if (config->zPacPlt) 355 error("-z pac-plt only supported on AArch64"); 356 if (config->zForceBti) 357 error("-z force-bti only supported on AArch64"); 358 } 359 } 360 361 static const char *getReproduceOption(opt::InputArgList &args) { 362 if (auto *arg = args.getLastArg(OPT_reproduce)) 363 return arg->getValue(); 364 return getenv("LLD_REPRODUCE"); 365 } 366 367 static bool hasZOption(opt::InputArgList &args, StringRef key) { 368 for (auto *arg : args.filtered(OPT_z)) 369 if (key == arg->getValue()) 370 return true; 371 return false; 372 } 373 374 static bool getZFlag(opt::InputArgList &args, StringRef k1, StringRef k2, 375 bool Default) { 376 for (auto *arg : args.filtered_reverse(OPT_z)) { 377 if (k1 == arg->getValue()) 378 return true; 379 if (k2 == arg->getValue()) 380 return false; 381 } 382 return Default; 383 } 384 385 static SeparateSegmentKind getZSeparate(opt::InputArgList &args) { 386 for (auto *arg : args.filtered_reverse(OPT_z)) { 387 StringRef v = arg->getValue(); 388 if (v == "noseparate-code") 389 return SeparateSegmentKind::None; 390 if (v == "separate-code") 391 return SeparateSegmentKind::Code; 392 if (v == "separate-loadable-segments") 393 return SeparateSegmentKind::Loadable; 394 } 395 return SeparateSegmentKind::None; 396 } 397 398 static GnuStackKind getZGnuStack(opt::InputArgList &args) { 399 for (auto *arg : args.filtered_reverse(OPT_z)) { 400 if (StringRef("execstack") == arg->getValue()) 401 return GnuStackKind::Exec; 402 if (StringRef("noexecstack") == arg->getValue()) 403 return GnuStackKind::NoExec; 404 if (StringRef("nognustack") == arg->getValue()) 405 return GnuStackKind::None; 406 } 407 408 return GnuStackKind::NoExec; 409 } 410 411 static bool isKnownZFlag(StringRef s) { 412 return s == "combreloc" || s == "copyreloc" || s == "defs" || 413 s == "execstack" || s == "force-bti" || s == "force-ibt" || 414 s == "global" || s == "hazardplt" || s == "ifunc-noplt" || 415 s == "initfirst" || s == "interpose" || 416 s == "keep-text-section-prefix" || s == "lazy" || s == "muldefs" || 417 s == "separate-code" || s == "separate-loadable-segments" || 418 s == "nocombreloc" || s == "nocopyreloc" || s == "nodefaultlib" || 419 s == "nodelete" || s == "nodlopen" || s == "noexecstack" || 420 s == "nognustack" || s == "nokeep-text-section-prefix" || 421 s == "norelro" || s == "noseparate-code" || s == "notext" || 422 s == "now" || s == "origin" || s == "pac-plt" || s == "relro" || 423 s == "retpolineplt" || s == "rodynamic" || s == "shstk" || 424 s == "text" || s == "undefs" || s == "wxneeded" || 425 s.startswith("common-page-size=") || s.startswith("max-page-size=") || 426 s.startswith("stack-size="); 427 } 428 429 // Report an error for an unknown -z option. 430 static void checkZOptions(opt::InputArgList &args) { 431 for (auto *arg : args.filtered(OPT_z)) 432 if (!isKnownZFlag(arg->getValue())) 433 error("unknown -z value: " + StringRef(arg->getValue())); 434 } 435 436 void LinkerDriver::main(ArrayRef<const char *> argsArr) { 437 ELFOptTable parser; 438 opt::InputArgList args = parser.parse(argsArr.slice(1)); 439 440 // Interpret this flag early because error() depends on them. 441 errorHandler().errorLimit = args::getInteger(args, OPT_error_limit, 20); 442 checkZOptions(args); 443 444 // Handle -help 445 if (args.hasArg(OPT_help)) { 446 printHelp(); 447 return; 448 } 449 450 // Handle -v or -version. 451 // 452 // A note about "compatible with GNU linkers" message: this is a hack for 453 // scripts generated by GNU Libtool 2.4.6 (released in February 2014 and 454 // still the newest version in March 2017) or earlier to recognize LLD as 455 // a GNU compatible linker. As long as an output for the -v option 456 // contains "GNU" or "with BFD", they recognize us as GNU-compatible. 457 // 458 // This is somewhat ugly hack, but in reality, we had no choice other 459 // than doing this. Considering the very long release cycle of Libtool, 460 // it is not easy to improve it to recognize LLD as a GNU compatible 461 // linker in a timely manner. Even if we can make it, there are still a 462 // lot of "configure" scripts out there that are generated by old version 463 // of Libtool. We cannot convince every software developer to migrate to 464 // the latest version and re-generate scripts. So we have this hack. 465 if (args.hasArg(OPT_v) || args.hasArg(OPT_version)) 466 message(getLLDVersion() + " (compatible with GNU linkers)"); 467 468 if (const char *path = getReproduceOption(args)) { 469 // Note that --reproduce is a debug option so you can ignore it 470 // if you are trying to understand the whole picture of the code. 471 Expected<std::unique_ptr<TarWriter>> errOrWriter = 472 TarWriter::create(path, path::stem(path)); 473 if (errOrWriter) { 474 tar = std::move(*errOrWriter); 475 tar->append("response.txt", createResponseFile(args)); 476 tar->append("version.txt", getLLDVersion() + "\n"); 477 } else { 478 error("--reproduce: " + toString(errOrWriter.takeError())); 479 } 480 } 481 482 readConfigs(args); 483 484 // The behavior of -v or --version is a bit strange, but this is 485 // needed for compatibility with GNU linkers. 486 if (args.hasArg(OPT_v) && !args.hasArg(OPT_INPUT)) 487 return; 488 if (args.hasArg(OPT_version)) 489 return; 490 491 // Initialize time trace profiler. 492 if (config->timeTraceEnabled) 493 timeTraceProfilerInitialize(config->timeTraceGranularity, config->progName); 494 495 { 496 llvm::TimeTraceScope timeScope("ExecuteLinker"); 497 498 initLLVM(); 499 createFiles(args); 500 if (errorCount()) 501 return; 502 503 inferMachineType(); 504 setConfigs(args); 505 checkOptions(); 506 if (errorCount()) 507 return; 508 509 // The Target instance handles target-specific stuff, such as applying 510 // relocations or writing a PLT section. It also contains target-dependent 511 // values such as a default image base address. 512 target = getTarget(); 513 514 switch (config->ekind) { 515 case ELF32LEKind: 516 link<ELF32LE>(args); 517 break; 518 case ELF32BEKind: 519 link<ELF32BE>(args); 520 break; 521 case ELF64LEKind: 522 link<ELF64LE>(args); 523 break; 524 case ELF64BEKind: 525 link<ELF64BE>(args); 526 break; 527 default: 528 llvm_unreachable("unknown Config->EKind"); 529 } 530 } 531 532 if (config->timeTraceEnabled) { 533 // Write the result of the time trace profiler. 534 std::string path = args.getLastArgValue(OPT_time_trace_file_eq).str(); 535 if (path.empty()) 536 path = (config->outputFile + ".time-trace").str(); 537 std::error_code ec; 538 raw_fd_ostream os(path, ec, sys::fs::OF_Text); 539 if (ec) { 540 error("cannot open " + path + ": " + ec.message()); 541 return; 542 } 543 timeTraceProfilerWrite(os); 544 timeTraceProfilerCleanup(); 545 } 546 } 547 548 static std::string getRpath(opt::InputArgList &args) { 549 std::vector<StringRef> v = args::getStrings(args, OPT_rpath); 550 return llvm::join(v.begin(), v.end(), ":"); 551 } 552 553 // Determines what we should do if there are remaining unresolved 554 // symbols after the name resolution. 555 static UnresolvedPolicy getUnresolvedSymbolPolicy(opt::InputArgList &args) { 556 UnresolvedPolicy errorOrWarn = args.hasFlag(OPT_error_unresolved_symbols, 557 OPT_warn_unresolved_symbols, true) 558 ? UnresolvedPolicy::ReportError 559 : UnresolvedPolicy::Warn; 560 561 // Process the last of -unresolved-symbols, -no-undefined or -z defs. 562 for (auto *arg : llvm::reverse(args)) { 563 switch (arg->getOption().getID()) { 564 case OPT_unresolved_symbols: { 565 StringRef s = arg->getValue(); 566 if (s == "ignore-all" || s == "ignore-in-object-files") 567 return UnresolvedPolicy::Ignore; 568 if (s == "ignore-in-shared-libs" || s == "report-all") 569 return errorOrWarn; 570 error("unknown --unresolved-symbols value: " + s); 571 continue; 572 } 573 case OPT_no_undefined: 574 return errorOrWarn; 575 case OPT_z: 576 if (StringRef(arg->getValue()) == "defs") 577 return errorOrWarn; 578 if (StringRef(arg->getValue()) == "undefs") 579 return UnresolvedPolicy::Ignore; 580 continue; 581 } 582 } 583 584 // -shared implies -unresolved-symbols=ignore-all because missing 585 // symbols are likely to be resolved at runtime using other DSOs. 586 if (config->shared) 587 return UnresolvedPolicy::Ignore; 588 return errorOrWarn; 589 } 590 591 static Target2Policy getTarget2(opt::InputArgList &args) { 592 StringRef s = args.getLastArgValue(OPT_target2, "got-rel"); 593 if (s == "rel") 594 return Target2Policy::Rel; 595 if (s == "abs") 596 return Target2Policy::Abs; 597 if (s == "got-rel") 598 return Target2Policy::GotRel; 599 error("unknown --target2 option: " + s); 600 return Target2Policy::GotRel; 601 } 602 603 static bool isOutputFormatBinary(opt::InputArgList &args) { 604 StringRef s = args.getLastArgValue(OPT_oformat, "elf"); 605 if (s == "binary") 606 return true; 607 if (!s.startswith("elf")) 608 error("unknown --oformat value: " + s); 609 return false; 610 } 611 612 static DiscardPolicy getDiscard(opt::InputArgList &args) { 613 if (args.hasArg(OPT_relocatable)) 614 return DiscardPolicy::None; 615 616 auto *arg = 617 args.getLastArg(OPT_discard_all, OPT_discard_locals, OPT_discard_none); 618 if (!arg) 619 return DiscardPolicy::Default; 620 if (arg->getOption().getID() == OPT_discard_all) 621 return DiscardPolicy::All; 622 if (arg->getOption().getID() == OPT_discard_locals) 623 return DiscardPolicy::Locals; 624 return DiscardPolicy::None; 625 } 626 627 static StringRef getDynamicLinker(opt::InputArgList &args) { 628 auto *arg = args.getLastArg(OPT_dynamic_linker, OPT_no_dynamic_linker); 629 if (!arg) 630 return ""; 631 if (arg->getOption().getID() == OPT_no_dynamic_linker) { 632 // --no-dynamic-linker suppresses undefined weak symbols in .dynsym 633 config->noDynamicLinker = true; 634 return ""; 635 } 636 return arg->getValue(); 637 } 638 639 static ICFLevel getICF(opt::InputArgList &args) { 640 auto *arg = args.getLastArg(OPT_icf_none, OPT_icf_safe, OPT_icf_all); 641 if (!arg || arg->getOption().getID() == OPT_icf_none) 642 return ICFLevel::None; 643 if (arg->getOption().getID() == OPT_icf_safe) 644 return ICFLevel::Safe; 645 return ICFLevel::All; 646 } 647 648 static StripPolicy getStrip(opt::InputArgList &args) { 649 if (args.hasArg(OPT_relocatable)) 650 return StripPolicy::None; 651 652 auto *arg = args.getLastArg(OPT_strip_all, OPT_strip_debug); 653 if (!arg) 654 return StripPolicy::None; 655 if (arg->getOption().getID() == OPT_strip_all) 656 return StripPolicy::All; 657 return StripPolicy::Debug; 658 } 659 660 static uint64_t parseSectionAddress(StringRef s, opt::InputArgList &args, 661 const opt::Arg &arg) { 662 uint64_t va = 0; 663 if (s.startswith("0x")) 664 s = s.drop_front(2); 665 if (!to_integer(s, va, 16)) 666 error("invalid argument: " + arg.getAsString(args)); 667 return va; 668 } 669 670 static StringMap<uint64_t> getSectionStartMap(opt::InputArgList &args) { 671 StringMap<uint64_t> ret; 672 for (auto *arg : args.filtered(OPT_section_start)) { 673 StringRef name; 674 StringRef addr; 675 std::tie(name, addr) = StringRef(arg->getValue()).split('='); 676 ret[name] = parseSectionAddress(addr, args, *arg); 677 } 678 679 if (auto *arg = args.getLastArg(OPT_Ttext)) 680 ret[".text"] = parseSectionAddress(arg->getValue(), args, *arg); 681 if (auto *arg = args.getLastArg(OPT_Tdata)) 682 ret[".data"] = parseSectionAddress(arg->getValue(), args, *arg); 683 if (auto *arg = args.getLastArg(OPT_Tbss)) 684 ret[".bss"] = parseSectionAddress(arg->getValue(), args, *arg); 685 return ret; 686 } 687 688 static SortSectionPolicy getSortSection(opt::InputArgList &args) { 689 StringRef s = args.getLastArgValue(OPT_sort_section); 690 if (s == "alignment") 691 return SortSectionPolicy::Alignment; 692 if (s == "name") 693 return SortSectionPolicy::Name; 694 if (!s.empty()) 695 error("unknown --sort-section rule: " + s); 696 return SortSectionPolicy::Default; 697 } 698 699 static OrphanHandlingPolicy getOrphanHandling(opt::InputArgList &args) { 700 StringRef s = args.getLastArgValue(OPT_orphan_handling, "place"); 701 if (s == "warn") 702 return OrphanHandlingPolicy::Warn; 703 if (s == "error") 704 return OrphanHandlingPolicy::Error; 705 if (s != "place") 706 error("unknown --orphan-handling mode: " + s); 707 return OrphanHandlingPolicy::Place; 708 } 709 710 // Parse --build-id or --build-id=<style>. We handle "tree" as a 711 // synonym for "sha1" because all our hash functions including 712 // -build-id=sha1 are actually tree hashes for performance reasons. 713 static std::pair<BuildIdKind, std::vector<uint8_t>> 714 getBuildId(opt::InputArgList &args) { 715 auto *arg = args.getLastArg(OPT_build_id, OPT_build_id_eq); 716 if (!arg) 717 return {BuildIdKind::None, {}}; 718 719 if (arg->getOption().getID() == OPT_build_id) 720 return {BuildIdKind::Fast, {}}; 721 722 StringRef s = arg->getValue(); 723 if (s == "fast") 724 return {BuildIdKind::Fast, {}}; 725 if (s == "md5") 726 return {BuildIdKind::Md5, {}}; 727 if (s == "sha1" || s == "tree") 728 return {BuildIdKind::Sha1, {}}; 729 if (s == "uuid") 730 return {BuildIdKind::Uuid, {}}; 731 if (s.startswith("0x")) 732 return {BuildIdKind::Hexstring, parseHex(s.substr(2))}; 733 734 if (s != "none") 735 error("unknown --build-id style: " + s); 736 return {BuildIdKind::None, {}}; 737 } 738 739 static std::pair<bool, bool> getPackDynRelocs(opt::InputArgList &args) { 740 StringRef s = args.getLastArgValue(OPT_pack_dyn_relocs, "none"); 741 if (s == "android") 742 return {true, false}; 743 if (s == "relr") 744 return {false, true}; 745 if (s == "android+relr") 746 return {true, true}; 747 748 if (s != "none") 749 error("unknown -pack-dyn-relocs format: " + s); 750 return {false, false}; 751 } 752 753 static void readCallGraph(MemoryBufferRef mb) { 754 // Build a map from symbol name to section 755 DenseMap<StringRef, Symbol *> map; 756 for (InputFile *file : objectFiles) 757 for (Symbol *sym : file->getSymbols()) 758 map[sym->getName()] = sym; 759 760 auto findSection = [&](StringRef name) -> InputSectionBase * { 761 Symbol *sym = map.lookup(name); 762 if (!sym) { 763 if (config->warnSymbolOrdering) 764 warn(mb.getBufferIdentifier() + ": no such symbol: " + name); 765 return nullptr; 766 } 767 maybeWarnUnorderableSymbol(sym); 768 769 if (Defined *dr = dyn_cast_or_null<Defined>(sym)) 770 return dyn_cast_or_null<InputSectionBase>(dr->section); 771 return nullptr; 772 }; 773 774 for (StringRef line : args::getLines(mb)) { 775 SmallVector<StringRef, 3> fields; 776 line.split(fields, ' '); 777 uint64_t count; 778 779 if (fields.size() != 3 || !to_integer(fields[2], count)) { 780 error(mb.getBufferIdentifier() + ": parse error"); 781 return; 782 } 783 784 if (InputSectionBase *from = findSection(fields[0])) 785 if (InputSectionBase *to = findSection(fields[1])) 786 config->callGraphProfile[std::make_pair(from, to)] += count; 787 } 788 } 789 790 template <class ELFT> static void readCallGraphsFromObjectFiles() { 791 for (auto file : objectFiles) { 792 auto *obj = cast<ObjFile<ELFT>>(file); 793 794 for (const Elf_CGProfile_Impl<ELFT> &cgpe : obj->cgProfile) { 795 auto *fromSym = dyn_cast<Defined>(&obj->getSymbol(cgpe.cgp_from)); 796 auto *toSym = dyn_cast<Defined>(&obj->getSymbol(cgpe.cgp_to)); 797 if (!fromSym || !toSym) 798 continue; 799 800 auto *from = dyn_cast_or_null<InputSectionBase>(fromSym->section); 801 auto *to = dyn_cast_or_null<InputSectionBase>(toSym->section); 802 if (from && to) 803 config->callGraphProfile[{from, to}] += cgpe.cgp_weight; 804 } 805 } 806 } 807 808 static bool getCompressDebugSections(opt::InputArgList &args) { 809 StringRef s = args.getLastArgValue(OPT_compress_debug_sections, "none"); 810 if (s == "none") 811 return false; 812 if (s != "zlib") 813 error("unknown --compress-debug-sections value: " + s); 814 if (!zlib::isAvailable()) 815 error("--compress-debug-sections: zlib is not available"); 816 return true; 817 } 818 819 static StringRef getAliasSpelling(opt::Arg *arg) { 820 if (const opt::Arg *alias = arg->getAlias()) 821 return alias->getSpelling(); 822 return arg->getSpelling(); 823 } 824 825 static std::pair<StringRef, StringRef> getOldNewOptions(opt::InputArgList &args, 826 unsigned id) { 827 auto *arg = args.getLastArg(id); 828 if (!arg) 829 return {"", ""}; 830 831 StringRef s = arg->getValue(); 832 std::pair<StringRef, StringRef> ret = s.split(';'); 833 if (ret.second.empty()) 834 error(getAliasSpelling(arg) + " expects 'old;new' format, but got " + s); 835 return ret; 836 } 837 838 // Parse the symbol ordering file and warn for any duplicate entries. 839 static std::vector<StringRef> getSymbolOrderingFile(MemoryBufferRef mb) { 840 SetVector<StringRef> names; 841 for (StringRef s : args::getLines(mb)) 842 if (!names.insert(s) && config->warnSymbolOrdering) 843 warn(mb.getBufferIdentifier() + ": duplicate ordered symbol: " + s); 844 845 return names.takeVector(); 846 } 847 848 static void parseClangOption(StringRef opt, const Twine &msg) { 849 std::string err; 850 raw_string_ostream os(err); 851 852 const char *argv[] = {config->progName.data(), opt.data()}; 853 if (cl::ParseCommandLineOptions(2, argv, "", &os)) 854 return; 855 os.flush(); 856 error(msg + ": " + StringRef(err).trim()); 857 } 858 859 // Initializes Config members by the command line options. 860 static void readConfigs(opt::InputArgList &args) { 861 errorHandler().verbose = args.hasArg(OPT_verbose); 862 errorHandler().fatalWarnings = 863 args.hasFlag(OPT_fatal_warnings, OPT_no_fatal_warnings, false); 864 errorHandler().vsDiagnostics = 865 args.hasArg(OPT_visual_studio_diagnostics_format, false); 866 threadsEnabled = args.hasFlag(OPT_threads, OPT_no_threads, true); 867 868 config->allowMultipleDefinition = 869 args.hasFlag(OPT_allow_multiple_definition, 870 OPT_no_allow_multiple_definition, false) || 871 hasZOption(args, "muldefs"); 872 config->allowShlibUndefined = 873 args.hasFlag(OPT_allow_shlib_undefined, OPT_no_allow_shlib_undefined, 874 args.hasArg(OPT_shared)); 875 config->auxiliaryList = args::getStrings(args, OPT_auxiliary); 876 config->bsymbolic = args.hasArg(OPT_Bsymbolic); 877 config->bsymbolicFunctions = args.hasArg(OPT_Bsymbolic_functions); 878 config->checkSections = 879 args.hasFlag(OPT_check_sections, OPT_no_check_sections, true); 880 config->chroot = args.getLastArgValue(OPT_chroot); 881 config->compressDebugSections = getCompressDebugSections(args); 882 config->cref = args.hasFlag(OPT_cref, OPT_no_cref, false); 883 config->defineCommon = args.hasFlag(OPT_define_common, OPT_no_define_common, 884 !args.hasArg(OPT_relocatable)); 885 config->demangle = args.hasFlag(OPT_demangle, OPT_no_demangle, true); 886 config->dependentLibraries = args.hasFlag(OPT_dependent_libraries, OPT_no_dependent_libraries, true); 887 config->disableVerify = args.hasArg(OPT_disable_verify); 888 config->discard = getDiscard(args); 889 config->dwoDir = args.getLastArgValue(OPT_plugin_opt_dwo_dir_eq); 890 config->dynamicLinker = getDynamicLinker(args); 891 config->ehFrameHdr = 892 args.hasFlag(OPT_eh_frame_hdr, OPT_no_eh_frame_hdr, false); 893 config->emitLLVM = args.hasArg(OPT_plugin_opt_emit_llvm, false); 894 config->emitRelocs = args.hasArg(OPT_emit_relocs); 895 config->callGraphProfileSort = args.hasFlag( 896 OPT_call_graph_profile_sort, OPT_no_call_graph_profile_sort, true); 897 config->enableNewDtags = 898 args.hasFlag(OPT_enable_new_dtags, OPT_disable_new_dtags, true); 899 config->entry = args.getLastArgValue(OPT_entry); 900 config->executeOnly = 901 args.hasFlag(OPT_execute_only, OPT_no_execute_only, false); 902 config->exportDynamic = 903 args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false); 904 config->filterList = args::getStrings(args, OPT_filter); 905 config->fini = args.getLastArgValue(OPT_fini, "_fini"); 906 config->fixCortexA53Errata843419 = args.hasArg(OPT_fix_cortex_a53_843419) && 907 !args.hasArg(OPT_relocatable); 908 config->fixCortexA8 = 909 args.hasArg(OPT_fix_cortex_a8) && !args.hasArg(OPT_relocatable); 910 config->gcSections = args.hasFlag(OPT_gc_sections, OPT_no_gc_sections, false); 911 config->gnuUnique = args.hasFlag(OPT_gnu_unique, OPT_no_gnu_unique, true); 912 config->gdbIndex = args.hasFlag(OPT_gdb_index, OPT_no_gdb_index, false); 913 config->icf = getICF(args); 914 config->ignoreDataAddressEquality = 915 args.hasArg(OPT_ignore_data_address_equality); 916 config->ignoreFunctionAddressEquality = 917 args.hasArg(OPT_ignore_function_address_equality); 918 config->init = args.getLastArgValue(OPT_init, "_init"); 919 config->ltoAAPipeline = args.getLastArgValue(OPT_lto_aa_pipeline); 920 config->ltoCSProfileGenerate = args.hasArg(OPT_lto_cs_profile_generate); 921 config->ltoCSProfileFile = args.getLastArgValue(OPT_lto_cs_profile_file); 922 config->ltoDebugPassManager = args.hasArg(OPT_lto_debug_pass_manager); 923 config->ltoNewPassManager = args.hasArg(OPT_lto_new_pass_manager); 924 config->ltoNewPmPasses = args.getLastArgValue(OPT_lto_newpm_passes); 925 config->ltoWholeProgramVisibility = 926 args.hasArg(OPT_lto_whole_program_visibility); 927 config->ltoo = args::getInteger(args, OPT_lto_O, 2); 928 config->ltoObjPath = args.getLastArgValue(OPT_lto_obj_path_eq); 929 config->ltoPartitions = args::getInteger(args, OPT_lto_partitions, 1); 930 config->ltoSampleProfile = args.getLastArgValue(OPT_lto_sample_profile); 931 config->mapFile = args.getLastArgValue(OPT_Map); 932 config->mipsGotSize = args::getInteger(args, OPT_mips_got_size, 0xfff0); 933 config->mergeArmExidx = 934 args.hasFlag(OPT_merge_exidx_entries, OPT_no_merge_exidx_entries, true); 935 config->mmapOutputFile = 936 args.hasFlag(OPT_mmap_output_file, OPT_no_mmap_output_file, true); 937 config->nmagic = args.hasFlag(OPT_nmagic, OPT_no_nmagic, false); 938 config->noinhibitExec = args.hasArg(OPT_noinhibit_exec); 939 config->nostdlib = args.hasArg(OPT_nostdlib); 940 config->oFormatBinary = isOutputFormatBinary(args); 941 config->omagic = args.hasFlag(OPT_omagic, OPT_no_omagic, false); 942 config->optRemarksFilename = args.getLastArgValue(OPT_opt_remarks_filename); 943 config->optRemarksPasses = args.getLastArgValue(OPT_opt_remarks_passes); 944 config->optRemarksWithHotness = args.hasArg(OPT_opt_remarks_with_hotness); 945 config->optRemarksFormat = args.getLastArgValue(OPT_opt_remarks_format); 946 config->optimize = args::getInteger(args, OPT_O, 1); 947 config->orphanHandling = getOrphanHandling(args); 948 config->outputFile = args.getLastArgValue(OPT_o); 949 config->pie = args.hasFlag(OPT_pie, OPT_no_pie, false); 950 config->printIcfSections = 951 args.hasFlag(OPT_print_icf_sections, OPT_no_print_icf_sections, false); 952 config->printGcSections = 953 args.hasFlag(OPT_print_gc_sections, OPT_no_print_gc_sections, false); 954 config->printSymbolOrder = 955 args.getLastArgValue(OPT_print_symbol_order); 956 config->rpath = getRpath(args); 957 config->relocatable = args.hasArg(OPT_relocatable); 958 config->saveTemps = args.hasArg(OPT_save_temps); 959 if (args.hasArg(OPT_shuffle_sections)) 960 config->shuffleSectionSeed = args::getInteger(args, OPT_shuffle_sections, 0); 961 config->searchPaths = args::getStrings(args, OPT_library_path); 962 config->sectionStartMap = getSectionStartMap(args); 963 config->shared = args.hasArg(OPT_shared); 964 config->singleRoRx = args.hasArg(OPT_no_rosegment); 965 config->soName = args.getLastArgValue(OPT_soname); 966 config->sortSection = getSortSection(args); 967 config->splitStackAdjustSize = args::getInteger(args, OPT_split_stack_adjust_size, 16384); 968 config->strip = getStrip(args); 969 config->sysroot = args.getLastArgValue(OPT_sysroot); 970 config->target1Rel = args.hasFlag(OPT_target1_rel, OPT_target1_abs, false); 971 config->target2 = getTarget2(args); 972 config->thinLTOCacheDir = args.getLastArgValue(OPT_thinlto_cache_dir); 973 config->thinLTOCachePolicy = CHECK( 974 parseCachePruningPolicy(args.getLastArgValue(OPT_thinlto_cache_policy)), 975 "--thinlto-cache-policy: invalid cache policy"); 976 config->thinLTOEmitImportsFiles = args.hasArg(OPT_thinlto_emit_imports_files); 977 config->thinLTOIndexOnly = args.hasArg(OPT_thinlto_index_only) || 978 args.hasArg(OPT_thinlto_index_only_eq); 979 config->thinLTOIndexOnlyArg = args.getLastArgValue(OPT_thinlto_index_only_eq); 980 config->thinLTOJobs = args::getInteger(args, OPT_thinlto_jobs, -1u); 981 config->thinLTOObjectSuffixReplace = 982 getOldNewOptions(args, OPT_thinlto_object_suffix_replace_eq); 983 config->thinLTOPrefixReplace = 984 getOldNewOptions(args, OPT_thinlto_prefix_replace_eq); 985 config->timeTraceEnabled = args.hasArg(OPT_time_trace); 986 config->timeTraceGranularity = 987 args::getInteger(args, OPT_time_trace_granularity, 500); 988 config->trace = args.hasArg(OPT_trace); 989 config->undefined = args::getStrings(args, OPT_undefined); 990 config->undefinedVersion = 991 args.hasFlag(OPT_undefined_version, OPT_no_undefined_version, true); 992 config->useAndroidRelrTags = args.hasFlag( 993 OPT_use_android_relr_tags, OPT_no_use_android_relr_tags, false); 994 config->unresolvedSymbols = getUnresolvedSymbolPolicy(args); 995 config->warnBackrefs = 996 args.hasFlag(OPT_warn_backrefs, OPT_no_warn_backrefs, false); 997 config->warnCommon = args.hasFlag(OPT_warn_common, OPT_no_warn_common, false); 998 config->warnIfuncTextrel = 999 args.hasFlag(OPT_warn_ifunc_textrel, OPT_no_warn_ifunc_textrel, false); 1000 config->warnSymbolOrdering = 1001 args.hasFlag(OPT_warn_symbol_ordering, OPT_no_warn_symbol_ordering, true); 1002 config->zCombreloc = getZFlag(args, "combreloc", "nocombreloc", true); 1003 config->zCopyreloc = getZFlag(args, "copyreloc", "nocopyreloc", true); 1004 config->zForceBti = hasZOption(args, "force-bti"); 1005 config->zForceIbt = hasZOption(args, "force-ibt"); 1006 config->zGlobal = hasZOption(args, "global"); 1007 config->zGnustack = getZGnuStack(args); 1008 config->zHazardplt = hasZOption(args, "hazardplt"); 1009 config->zIfuncNoplt = hasZOption(args, "ifunc-noplt"); 1010 config->zInitfirst = hasZOption(args, "initfirst"); 1011 config->zInterpose = hasZOption(args, "interpose"); 1012 config->zKeepTextSectionPrefix = getZFlag( 1013 args, "keep-text-section-prefix", "nokeep-text-section-prefix", false); 1014 config->zNodefaultlib = hasZOption(args, "nodefaultlib"); 1015 config->zNodelete = hasZOption(args, "nodelete"); 1016 config->zNodlopen = hasZOption(args, "nodlopen"); 1017 config->zNow = getZFlag(args, "now", "lazy", false); 1018 config->zOrigin = hasZOption(args, "origin"); 1019 config->zPacPlt = hasZOption(args, "pac-plt"); 1020 config->zRelro = getZFlag(args, "relro", "norelro", true); 1021 config->zRetpolineplt = hasZOption(args, "retpolineplt"); 1022 config->zRodynamic = hasZOption(args, "rodynamic"); 1023 config->zSeparate = getZSeparate(args); 1024 config->zShstk = hasZOption(args, "shstk"); 1025 config->zStackSize = args::getZOptionValue(args, OPT_z, "stack-size", 0); 1026 config->zText = getZFlag(args, "text", "notext", true); 1027 config->zWxneeded = hasZOption(args, "wxneeded"); 1028 1029 // Parse LTO options. 1030 if (auto *arg = args.getLastArg(OPT_plugin_opt_mcpu_eq)) 1031 parseClangOption(saver.save("-mcpu=" + StringRef(arg->getValue())), 1032 arg->getSpelling()); 1033 1034 for (auto *arg : args.filtered(OPT_plugin_opt)) 1035 parseClangOption(arg->getValue(), arg->getSpelling()); 1036 1037 // Parse -mllvm options. 1038 for (auto *arg : args.filtered(OPT_mllvm)) 1039 parseClangOption(arg->getValue(), arg->getSpelling()); 1040 1041 if (config->ltoo > 3) 1042 error("invalid optimization level for LTO: " + Twine(config->ltoo)); 1043 if (config->ltoPartitions == 0) 1044 error("--lto-partitions: number of threads must be > 0"); 1045 if (config->thinLTOJobs == 0) 1046 error("--thinlto-jobs: number of threads must be > 0"); 1047 1048 if (config->splitStackAdjustSize < 0) 1049 error("--split-stack-adjust-size: size must be >= 0"); 1050 1051 // The text segment is traditionally the first segment, whose address equals 1052 // the base address. However, lld places the R PT_LOAD first. -Ttext-segment 1053 // is an old-fashioned option that does not play well with lld's layout. 1054 // Suggest --image-base as a likely alternative. 1055 if (args.hasArg(OPT_Ttext_segment)) 1056 error("-Ttext-segment is not supported. Use --image-base if you " 1057 "intend to set the base address"); 1058 1059 // Parse ELF{32,64}{LE,BE} and CPU type. 1060 if (auto *arg = args.getLastArg(OPT_m)) { 1061 StringRef s = arg->getValue(); 1062 std::tie(config->ekind, config->emachine, config->osabi) = 1063 parseEmulation(s); 1064 config->mipsN32Abi = 1065 (s.startswith("elf32btsmipn32") || s.startswith("elf32ltsmipn32")); 1066 config->emulation = s; 1067 } 1068 1069 // Parse -hash-style={sysv,gnu,both}. 1070 if (auto *arg = args.getLastArg(OPT_hash_style)) { 1071 StringRef s = arg->getValue(); 1072 if (s == "sysv") 1073 config->sysvHash = true; 1074 else if (s == "gnu") 1075 config->gnuHash = true; 1076 else if (s == "both") 1077 config->sysvHash = config->gnuHash = true; 1078 else 1079 error("unknown -hash-style: " + s); 1080 } 1081 1082 if (args.hasArg(OPT_print_map)) 1083 config->mapFile = "-"; 1084 1085 // Page alignment can be disabled by the -n (--nmagic) and -N (--omagic). 1086 // As PT_GNU_RELRO relies on Paging, do not create it when we have disabled 1087 // it. 1088 if (config->nmagic || config->omagic) 1089 config->zRelro = false; 1090 1091 std::tie(config->buildId, config->buildIdVector) = getBuildId(args); 1092 1093 std::tie(config->androidPackDynRelocs, config->relrPackDynRelocs) = 1094 getPackDynRelocs(args); 1095 1096 if (auto *arg = args.getLastArg(OPT_symbol_ordering_file)){ 1097 if (args.hasArg(OPT_call_graph_ordering_file)) 1098 error("--symbol-ordering-file and --call-graph-order-file " 1099 "may not be used together"); 1100 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())){ 1101 config->symbolOrderingFile = getSymbolOrderingFile(*buffer); 1102 // Also need to disable CallGraphProfileSort to prevent 1103 // LLD order symbols with CGProfile 1104 config->callGraphProfileSort = false; 1105 } 1106 } 1107 1108 assert(config->versionDefinitions.empty()); 1109 config->versionDefinitions.push_back({"local", (uint16_t)VER_NDX_LOCAL, {}}); 1110 config->versionDefinitions.push_back( 1111 {"global", (uint16_t)VER_NDX_GLOBAL, {}}); 1112 1113 // If --retain-symbol-file is used, we'll keep only the symbols listed in 1114 // the file and discard all others. 1115 if (auto *arg = args.getLastArg(OPT_retain_symbols_file)) { 1116 config->versionDefinitions[VER_NDX_LOCAL].patterns.push_back( 1117 {"*", /*isExternCpp=*/false, /*hasWildcard=*/true}); 1118 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 1119 for (StringRef s : args::getLines(*buffer)) 1120 config->versionDefinitions[VER_NDX_GLOBAL].patterns.push_back( 1121 {s, /*isExternCpp=*/false, /*hasWildcard=*/false}); 1122 } 1123 1124 // Parses -dynamic-list and -export-dynamic-symbol. They make some 1125 // symbols private. Note that -export-dynamic takes precedence over them 1126 // as it says all symbols should be exported. 1127 if (!config->exportDynamic) { 1128 for (auto *arg : args.filtered(OPT_dynamic_list)) 1129 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 1130 readDynamicList(*buffer); 1131 1132 for (auto *arg : args.filtered(OPT_export_dynamic_symbol)) 1133 config->dynamicList.push_back( 1134 {arg->getValue(), /*isExternCpp=*/false, /*hasWildcard=*/false}); 1135 } 1136 1137 // If --export-dynamic-symbol=foo is given and symbol foo is defined in 1138 // an object file in an archive file, that object file should be pulled 1139 // out and linked. (It doesn't have to behave like that from technical 1140 // point of view, but this is needed for compatibility with GNU.) 1141 for (auto *arg : args.filtered(OPT_export_dynamic_symbol)) 1142 config->undefined.push_back(arg->getValue()); 1143 1144 for (auto *arg : args.filtered(OPT_version_script)) 1145 if (Optional<std::string> path = searchScript(arg->getValue())) { 1146 if (Optional<MemoryBufferRef> buffer = readFile(*path)) 1147 readVersionScript(*buffer); 1148 } else { 1149 error(Twine("cannot find version script ") + arg->getValue()); 1150 } 1151 } 1152 1153 // Some Config members do not directly correspond to any particular 1154 // command line options, but computed based on other Config values. 1155 // This function initialize such members. See Config.h for the details 1156 // of these values. 1157 static void setConfigs(opt::InputArgList &args) { 1158 ELFKind k = config->ekind; 1159 uint16_t m = config->emachine; 1160 1161 config->copyRelocs = (config->relocatable || config->emitRelocs); 1162 config->is64 = (k == ELF64LEKind || k == ELF64BEKind); 1163 config->isLE = (k == ELF32LEKind || k == ELF64LEKind); 1164 config->endianness = config->isLE ? endianness::little : endianness::big; 1165 config->isMips64EL = (k == ELF64LEKind && m == EM_MIPS); 1166 config->isPic = config->pie || config->shared; 1167 config->picThunk = args.hasArg(OPT_pic_veneer, config->isPic); 1168 config->wordsize = config->is64 ? 8 : 4; 1169 1170 // ELF defines two different ways to store relocation addends as shown below: 1171 // 1172 // Rel: Addends are stored to the location where relocations are applied. 1173 // Rela: Addends are stored as part of relocation entry. 1174 // 1175 // In other words, Rela makes it easy to read addends at the price of extra 1176 // 4 or 8 byte for each relocation entry. We don't know why ELF defined two 1177 // different mechanisms in the first place, but this is how the spec is 1178 // defined. 1179 // 1180 // You cannot choose which one, Rel or Rela, you want to use. Instead each 1181 // ABI defines which one you need to use. The following expression expresses 1182 // that. 1183 config->isRela = m == EM_AARCH64 || m == EM_AMDGPU || m == EM_HEXAGON || 1184 m == EM_PPC || m == EM_PPC64 || m == EM_RISCV || 1185 m == EM_X86_64; 1186 1187 // If the output uses REL relocations we must store the dynamic relocation 1188 // addends to the output sections. We also store addends for RELA relocations 1189 // if --apply-dynamic-relocs is used. 1190 // We default to not writing the addends when using RELA relocations since 1191 // any standard conforming tool can find it in r_addend. 1192 config->writeAddends = args.hasFlag(OPT_apply_dynamic_relocs, 1193 OPT_no_apply_dynamic_relocs, false) || 1194 !config->isRela; 1195 1196 config->tocOptimize = 1197 args.hasFlag(OPT_toc_optimize, OPT_no_toc_optimize, m == EM_PPC64); 1198 } 1199 1200 // Returns a value of "-format" option. 1201 static bool isFormatBinary(StringRef s) { 1202 if (s == "binary") 1203 return true; 1204 if (s == "elf" || s == "default") 1205 return false; 1206 error("unknown -format value: " + s + 1207 " (supported formats: elf, default, binary)"); 1208 return false; 1209 } 1210 1211 void LinkerDriver::createFiles(opt::InputArgList &args) { 1212 // For --{push,pop}-state. 1213 std::vector<std::tuple<bool, bool, bool>> stack; 1214 1215 // Iterate over argv to process input files and positional arguments. 1216 for (auto *arg : args) { 1217 switch (arg->getOption().getID()) { 1218 case OPT_library: 1219 addLibrary(arg->getValue()); 1220 break; 1221 case OPT_INPUT: 1222 addFile(arg->getValue(), /*withLOption=*/false); 1223 break; 1224 case OPT_defsym: { 1225 StringRef from; 1226 StringRef to; 1227 std::tie(from, to) = StringRef(arg->getValue()).split('='); 1228 if (from.empty() || to.empty()) 1229 error("-defsym: syntax error: " + StringRef(arg->getValue())); 1230 else 1231 readDefsym(from, MemoryBufferRef(to, "-defsym")); 1232 break; 1233 } 1234 case OPT_script: 1235 if (Optional<std::string> path = searchScript(arg->getValue())) { 1236 if (Optional<MemoryBufferRef> mb = readFile(*path)) 1237 readLinkerScript(*mb); 1238 break; 1239 } 1240 error(Twine("cannot find linker script ") + arg->getValue()); 1241 break; 1242 case OPT_as_needed: 1243 config->asNeeded = true; 1244 break; 1245 case OPT_format: 1246 config->formatBinary = isFormatBinary(arg->getValue()); 1247 break; 1248 case OPT_no_as_needed: 1249 config->asNeeded = false; 1250 break; 1251 case OPT_Bstatic: 1252 case OPT_omagic: 1253 case OPT_nmagic: 1254 config->isStatic = true; 1255 break; 1256 case OPT_Bdynamic: 1257 config->isStatic = false; 1258 break; 1259 case OPT_whole_archive: 1260 inWholeArchive = true; 1261 break; 1262 case OPT_no_whole_archive: 1263 inWholeArchive = false; 1264 break; 1265 case OPT_just_symbols: 1266 if (Optional<MemoryBufferRef> mb = readFile(arg->getValue())) { 1267 files.push_back(createObjectFile(*mb)); 1268 files.back()->justSymbols = true; 1269 } 1270 break; 1271 case OPT_start_group: 1272 if (InputFile::isInGroup) 1273 error("nested --start-group"); 1274 InputFile::isInGroup = true; 1275 break; 1276 case OPT_end_group: 1277 if (!InputFile::isInGroup) 1278 error("stray --end-group"); 1279 InputFile::isInGroup = false; 1280 ++InputFile::nextGroupId; 1281 break; 1282 case OPT_start_lib: 1283 if (inLib) 1284 error("nested --start-lib"); 1285 if (InputFile::isInGroup) 1286 error("may not nest --start-lib in --start-group"); 1287 inLib = true; 1288 InputFile::isInGroup = true; 1289 break; 1290 case OPT_end_lib: 1291 if (!inLib) 1292 error("stray --end-lib"); 1293 inLib = false; 1294 InputFile::isInGroup = false; 1295 ++InputFile::nextGroupId; 1296 break; 1297 case OPT_push_state: 1298 stack.emplace_back(config->asNeeded, config->isStatic, inWholeArchive); 1299 break; 1300 case OPT_pop_state: 1301 if (stack.empty()) { 1302 error("unbalanced --push-state/--pop-state"); 1303 break; 1304 } 1305 std::tie(config->asNeeded, config->isStatic, inWholeArchive) = stack.back(); 1306 stack.pop_back(); 1307 break; 1308 } 1309 } 1310 1311 if (files.empty() && errorCount() == 0) 1312 error("no input files"); 1313 } 1314 1315 // If -m <machine_type> was not given, infer it from object files. 1316 void LinkerDriver::inferMachineType() { 1317 if (config->ekind != ELFNoneKind) 1318 return; 1319 1320 for (InputFile *f : files) { 1321 if (f->ekind == ELFNoneKind) 1322 continue; 1323 config->ekind = f->ekind; 1324 config->emachine = f->emachine; 1325 config->osabi = f->osabi; 1326 config->mipsN32Abi = config->emachine == EM_MIPS && isMipsN32Abi(f); 1327 return; 1328 } 1329 error("target emulation unknown: -m or at least one .o file required"); 1330 } 1331 1332 // Parse -z max-page-size=<value>. The default value is defined by 1333 // each target. 1334 static uint64_t getMaxPageSize(opt::InputArgList &args) { 1335 uint64_t val = args::getZOptionValue(args, OPT_z, "max-page-size", 1336 target->defaultMaxPageSize); 1337 if (!isPowerOf2_64(val)) 1338 error("max-page-size: value isn't a power of 2"); 1339 if (config->nmagic || config->omagic) { 1340 if (val != target->defaultMaxPageSize) 1341 warn("-z max-page-size set, but paging disabled by omagic or nmagic"); 1342 return 1; 1343 } 1344 return val; 1345 } 1346 1347 // Parse -z common-page-size=<value>. The default value is defined by 1348 // each target. 1349 static uint64_t getCommonPageSize(opt::InputArgList &args) { 1350 uint64_t val = args::getZOptionValue(args, OPT_z, "common-page-size", 1351 target->defaultCommonPageSize); 1352 if (!isPowerOf2_64(val)) 1353 error("common-page-size: value isn't a power of 2"); 1354 if (config->nmagic || config->omagic) { 1355 if (val != target->defaultCommonPageSize) 1356 warn("-z common-page-size set, but paging disabled by omagic or nmagic"); 1357 return 1; 1358 } 1359 // commonPageSize can't be larger than maxPageSize. 1360 if (val > config->maxPageSize) 1361 val = config->maxPageSize; 1362 return val; 1363 } 1364 1365 // Parses -image-base option. 1366 static Optional<uint64_t> getImageBase(opt::InputArgList &args) { 1367 // Because we are using "Config->maxPageSize" here, this function has to be 1368 // called after the variable is initialized. 1369 auto *arg = args.getLastArg(OPT_image_base); 1370 if (!arg) 1371 return None; 1372 1373 StringRef s = arg->getValue(); 1374 uint64_t v; 1375 if (!to_integer(s, v)) { 1376 error("-image-base: number expected, but got " + s); 1377 return 0; 1378 } 1379 if ((v % config->maxPageSize) != 0) 1380 warn("-image-base: address isn't multiple of page size: " + s); 1381 return v; 1382 } 1383 1384 // Parses `--exclude-libs=lib,lib,...`. 1385 // The library names may be delimited by commas or colons. 1386 static DenseSet<StringRef> getExcludeLibs(opt::InputArgList &args) { 1387 DenseSet<StringRef> ret; 1388 for (auto *arg : args.filtered(OPT_exclude_libs)) { 1389 StringRef s = arg->getValue(); 1390 for (;;) { 1391 size_t pos = s.find_first_of(",:"); 1392 if (pos == StringRef::npos) 1393 break; 1394 ret.insert(s.substr(0, pos)); 1395 s = s.substr(pos + 1); 1396 } 1397 ret.insert(s); 1398 } 1399 return ret; 1400 } 1401 1402 // Handles the -exclude-libs option. If a static library file is specified 1403 // by the -exclude-libs option, all public symbols from the archive become 1404 // private unless otherwise specified by version scripts or something. 1405 // A special library name "ALL" means all archive files. 1406 // 1407 // This is not a popular option, but some programs such as bionic libc use it. 1408 static void excludeLibs(opt::InputArgList &args) { 1409 DenseSet<StringRef> libs = getExcludeLibs(args); 1410 bool all = libs.count("ALL"); 1411 1412 auto visit = [&](InputFile *file) { 1413 if (!file->archiveName.empty()) 1414 if (all || libs.count(path::filename(file->archiveName))) 1415 for (Symbol *sym : file->getSymbols()) 1416 if (!sym->isUndefined() && !sym->isLocal() && sym->file == file) 1417 sym->versionId = VER_NDX_LOCAL; 1418 }; 1419 1420 for (InputFile *file : objectFiles) 1421 visit(file); 1422 1423 for (BitcodeFile *file : bitcodeFiles) 1424 visit(file); 1425 } 1426 1427 // Force Sym to be entered in the output. Used for -u or equivalent. 1428 static void handleUndefined(Symbol *sym) { 1429 // Since a symbol may not be used inside the program, LTO may 1430 // eliminate it. Mark the symbol as "used" to prevent it. 1431 sym->isUsedInRegularObj = true; 1432 1433 if (sym->isLazy()) 1434 sym->fetch(); 1435 } 1436 1437 // As an extension to GNU linkers, lld supports a variant of `-u` 1438 // which accepts wildcard patterns. All symbols that match a given 1439 // pattern are handled as if they were given by `-u`. 1440 static void handleUndefinedGlob(StringRef arg) { 1441 Expected<GlobPattern> pat = GlobPattern::create(arg); 1442 if (!pat) { 1443 error("--undefined-glob: " + toString(pat.takeError())); 1444 return; 1445 } 1446 1447 std::vector<Symbol *> syms; 1448 for (Symbol *sym : symtab->symbols()) { 1449 // Calling Sym->fetch() from here is not safe because it may 1450 // add new symbols to the symbol table, invalidating the 1451 // current iterator. So we just keep a note. 1452 if (pat->match(sym->getName())) 1453 syms.push_back(sym); 1454 } 1455 1456 for (Symbol *sym : syms) 1457 handleUndefined(sym); 1458 } 1459 1460 static void handleLibcall(StringRef name) { 1461 Symbol *sym = symtab->find(name); 1462 if (!sym || !sym->isLazy()) 1463 return; 1464 1465 MemoryBufferRef mb; 1466 if (auto *lo = dyn_cast<LazyObject>(sym)) 1467 mb = lo->file->mb; 1468 else 1469 mb = cast<LazyArchive>(sym)->getMemberBuffer(); 1470 1471 if (isBitcode(mb)) 1472 sym->fetch(); 1473 } 1474 1475 // Replaces common symbols with defined symbols reside in .bss sections. 1476 // This function is called after all symbol names are resolved. As a 1477 // result, the passes after the symbol resolution won't see any 1478 // symbols of type CommonSymbol. 1479 static void replaceCommonSymbols() { 1480 for (Symbol *sym : symtab->symbols()) { 1481 auto *s = dyn_cast<CommonSymbol>(sym); 1482 if (!s) 1483 continue; 1484 1485 auto *bss = make<BssSection>("COMMON", s->size, s->alignment); 1486 bss->file = s->file; 1487 bss->markDead(); 1488 inputSections.push_back(bss); 1489 s->replace(Defined{s->file, s->getName(), s->binding, s->stOther, s->type, 1490 /*value=*/0, s->size, bss}); 1491 } 1492 } 1493 1494 // If all references to a DSO happen to be weak, the DSO is not added 1495 // to DT_NEEDED. If that happens, we need to eliminate shared symbols 1496 // created from the DSO. Otherwise, they become dangling references 1497 // that point to a non-existent DSO. 1498 static void demoteSharedSymbols() { 1499 for (Symbol *sym : symtab->symbols()) { 1500 auto *s = dyn_cast<SharedSymbol>(sym); 1501 if (!s || s->getFile().isNeeded) 1502 continue; 1503 1504 bool used = s->used; 1505 s->replace(Undefined{nullptr, s->getName(), STB_WEAK, s->stOther, s->type}); 1506 s->used = used; 1507 } 1508 } 1509 1510 // The section referred to by `s` is considered address-significant. Set the 1511 // keepUnique flag on the section if appropriate. 1512 static void markAddrsig(Symbol *s) { 1513 if (auto *d = dyn_cast_or_null<Defined>(s)) 1514 if (d->section) 1515 // We don't need to keep text sections unique under --icf=all even if they 1516 // are address-significant. 1517 if (config->icf == ICFLevel::Safe || !(d->section->flags & SHF_EXECINSTR)) 1518 d->section->keepUnique = true; 1519 } 1520 1521 // Record sections that define symbols mentioned in --keep-unique <symbol> 1522 // and symbols referred to by address-significance tables. These sections are 1523 // ineligible for ICF. 1524 template <class ELFT> 1525 static void findKeepUniqueSections(opt::InputArgList &args) { 1526 for (auto *arg : args.filtered(OPT_keep_unique)) { 1527 StringRef name = arg->getValue(); 1528 auto *d = dyn_cast_or_null<Defined>(symtab->find(name)); 1529 if (!d || !d->section) { 1530 warn("could not find symbol " + name + " to keep unique"); 1531 continue; 1532 } 1533 d->section->keepUnique = true; 1534 } 1535 1536 // --icf=all --ignore-data-address-equality means that we can ignore 1537 // the dynsym and address-significance tables entirely. 1538 if (config->icf == ICFLevel::All && config->ignoreDataAddressEquality) 1539 return; 1540 1541 // Symbols in the dynsym could be address-significant in other executables 1542 // or DSOs, so we conservatively mark them as address-significant. 1543 for (Symbol *sym : symtab->symbols()) 1544 if (sym->includeInDynsym()) 1545 markAddrsig(sym); 1546 1547 // Visit the address-significance table in each object file and mark each 1548 // referenced symbol as address-significant. 1549 for (InputFile *f : objectFiles) { 1550 auto *obj = cast<ObjFile<ELFT>>(f); 1551 ArrayRef<Symbol *> syms = obj->getSymbols(); 1552 if (obj->addrsigSec) { 1553 ArrayRef<uint8_t> contents = 1554 check(obj->getObj().getSectionContents(obj->addrsigSec)); 1555 const uint8_t *cur = contents.begin(); 1556 while (cur != contents.end()) { 1557 unsigned size; 1558 const char *err; 1559 uint64_t symIndex = decodeULEB128(cur, &size, contents.end(), &err); 1560 if (err) 1561 fatal(toString(f) + ": could not decode addrsig section: " + err); 1562 markAddrsig(syms[symIndex]); 1563 cur += size; 1564 } 1565 } else { 1566 // If an object file does not have an address-significance table, 1567 // conservatively mark all of its symbols as address-significant. 1568 for (Symbol *s : syms) 1569 markAddrsig(s); 1570 } 1571 } 1572 } 1573 1574 // This function reads a symbol partition specification section. These sections 1575 // are used to control which partition a symbol is allocated to. See 1576 // https://lld.llvm.org/Partitions.html for more details on partitions. 1577 template <typename ELFT> 1578 static void readSymbolPartitionSection(InputSectionBase *s) { 1579 // Read the relocation that refers to the partition's entry point symbol. 1580 Symbol *sym; 1581 if (s->areRelocsRela) 1582 sym = &s->getFile<ELFT>()->getRelocTargetSym(s->template relas<ELFT>()[0]); 1583 else 1584 sym = &s->getFile<ELFT>()->getRelocTargetSym(s->template rels<ELFT>()[0]); 1585 if (!isa<Defined>(sym) || !sym->includeInDynsym()) 1586 return; 1587 1588 StringRef partName = reinterpret_cast<const char *>(s->data().data()); 1589 for (Partition &part : partitions) { 1590 if (part.name == partName) { 1591 sym->partition = part.getNumber(); 1592 return; 1593 } 1594 } 1595 1596 // Forbid partitions from being used on incompatible targets, and forbid them 1597 // from being used together with various linker features that assume a single 1598 // set of output sections. 1599 if (script->hasSectionsCommand) 1600 error(toString(s->file) + 1601 ": partitions cannot be used with the SECTIONS command"); 1602 if (script->hasPhdrsCommands()) 1603 error(toString(s->file) + 1604 ": partitions cannot be used with the PHDRS command"); 1605 if (!config->sectionStartMap.empty()) 1606 error(toString(s->file) + ": partitions cannot be used with " 1607 "--section-start, -Ttext, -Tdata or -Tbss"); 1608 if (config->emachine == EM_MIPS) 1609 error(toString(s->file) + ": partitions cannot be used on this target"); 1610 1611 // Impose a limit of no more than 254 partitions. This limit comes from the 1612 // sizes of the Partition fields in InputSectionBase and Symbol, as well as 1613 // the amount of space devoted to the partition number in RankFlags. 1614 if (partitions.size() == 254) 1615 fatal("may not have more than 254 partitions"); 1616 1617 partitions.emplace_back(); 1618 Partition &newPart = partitions.back(); 1619 newPart.name = partName; 1620 sym->partition = newPart.getNumber(); 1621 } 1622 1623 static Symbol *addUndefined(StringRef name) { 1624 return symtab->addSymbol( 1625 Undefined{nullptr, name, STB_GLOBAL, STV_DEFAULT, 0}); 1626 } 1627 1628 // This function is where all the optimizations of link-time 1629 // optimization takes place. When LTO is in use, some input files are 1630 // not in native object file format but in the LLVM bitcode format. 1631 // This function compiles bitcode files into a few big native files 1632 // using LLVM functions and replaces bitcode symbols with the results. 1633 // Because all bitcode files that the program consists of are passed to 1634 // the compiler at once, it can do a whole-program optimization. 1635 template <class ELFT> void LinkerDriver::compileBitcodeFiles() { 1636 llvm::TimeTraceScope timeScope("LTO"); 1637 // Compile bitcode files and replace bitcode symbols. 1638 lto.reset(new BitcodeCompiler); 1639 for (BitcodeFile *file : bitcodeFiles) 1640 lto->add(*file); 1641 1642 for (InputFile *file : lto->compile()) { 1643 auto *obj = cast<ObjFile<ELFT>>(file); 1644 obj->parse(/*ignoreComdats=*/true); 1645 for (Symbol *sym : obj->getGlobalSymbols()) 1646 sym->parseSymbolVersion(); 1647 objectFiles.push_back(file); 1648 } 1649 } 1650 1651 // The --wrap option is a feature to rename symbols so that you can write 1652 // wrappers for existing functions. If you pass `-wrap=foo`, all 1653 // occurrences of symbol `foo` are resolved to `wrap_foo` (so, you are 1654 // expected to write `wrap_foo` function as a wrapper). The original 1655 // symbol becomes accessible as `real_foo`, so you can call that from your 1656 // wrapper. 1657 // 1658 // This data structure is instantiated for each -wrap option. 1659 struct WrappedSymbol { 1660 Symbol *sym; 1661 Symbol *real; 1662 Symbol *wrap; 1663 }; 1664 1665 // Handles -wrap option. 1666 // 1667 // This function instantiates wrapper symbols. At this point, they seem 1668 // like they are not being used at all, so we explicitly set some flags so 1669 // that LTO won't eliminate them. 1670 static std::vector<WrappedSymbol> addWrappedSymbols(opt::InputArgList &args) { 1671 std::vector<WrappedSymbol> v; 1672 DenseSet<StringRef> seen; 1673 1674 for (auto *arg : args.filtered(OPT_wrap)) { 1675 StringRef name = arg->getValue(); 1676 if (!seen.insert(name).second) 1677 continue; 1678 1679 Symbol *sym = symtab->find(name); 1680 if (!sym) 1681 continue; 1682 1683 Symbol *real = addUndefined(saver.save("__real_" + name)); 1684 Symbol *wrap = addUndefined(saver.save("__wrap_" + name)); 1685 v.push_back({sym, real, wrap}); 1686 1687 // We want to tell LTO not to inline symbols to be overwritten 1688 // because LTO doesn't know the final symbol contents after renaming. 1689 real->canInline = false; 1690 sym->canInline = false; 1691 1692 // Tell LTO not to eliminate these symbols. 1693 sym->isUsedInRegularObj = true; 1694 wrap->isUsedInRegularObj = true; 1695 } 1696 return v; 1697 } 1698 1699 // Do renaming for -wrap by updating pointers to symbols. 1700 // 1701 // When this function is executed, only InputFiles and symbol table 1702 // contain pointers to symbol objects. We visit them to replace pointers, 1703 // so that wrapped symbols are swapped as instructed by the command line. 1704 static void wrapSymbols(ArrayRef<WrappedSymbol> wrapped) { 1705 DenseMap<Symbol *, Symbol *> map; 1706 for (const WrappedSymbol &w : wrapped) { 1707 map[w.sym] = w.wrap; 1708 map[w.real] = w.sym; 1709 } 1710 1711 // Update pointers in input files. 1712 parallelForEach(objectFiles, [&](InputFile *file) { 1713 MutableArrayRef<Symbol *> syms = file->getMutableSymbols(); 1714 for (size_t i = 0, e = syms.size(); i != e; ++i) 1715 if (Symbol *s = map.lookup(syms[i])) 1716 syms[i] = s; 1717 }); 1718 1719 // Update pointers in the symbol table. 1720 for (const WrappedSymbol &w : wrapped) 1721 symtab->wrap(w.sym, w.real, w.wrap); 1722 } 1723 1724 // To enable CET (x86's hardware-assited control flow enforcement), each 1725 // source file must be compiled with -fcf-protection. Object files compiled 1726 // with the flag contain feature flags indicating that they are compatible 1727 // with CET. We enable the feature only when all object files are compatible 1728 // with CET. 1729 // 1730 // This is also the case with AARCH64's BTI and PAC which use the similar 1731 // GNU_PROPERTY_AARCH64_FEATURE_1_AND mechanism. 1732 template <class ELFT> static uint32_t getAndFeatures() { 1733 if (config->emachine != EM_386 && config->emachine != EM_X86_64 && 1734 config->emachine != EM_AARCH64) 1735 return 0; 1736 1737 uint32_t ret = -1; 1738 for (InputFile *f : objectFiles) { 1739 uint32_t features = cast<ObjFile<ELFT>>(f)->andFeatures; 1740 if (config->zForceBti && !(features & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)) { 1741 warn(toString(f) + ": -z force-bti: file does not have " 1742 "GNU_PROPERTY_AARCH64_FEATURE_1_BTI property"); 1743 features |= GNU_PROPERTY_AARCH64_FEATURE_1_BTI; 1744 } else if (config->zForceIbt && 1745 !(features & GNU_PROPERTY_X86_FEATURE_1_IBT)) { 1746 warn(toString(f) + ": -z force-ibt: file does not have " 1747 "GNU_PROPERTY_X86_FEATURE_1_IBT property"); 1748 features |= GNU_PROPERTY_X86_FEATURE_1_IBT; 1749 } 1750 if (config->zPacPlt && !(features & GNU_PROPERTY_AARCH64_FEATURE_1_PAC)) { 1751 warn(toString(f) + ": -z pac-plt: file does not have " 1752 "GNU_PROPERTY_AARCH64_FEATURE_1_PAC property"); 1753 features |= GNU_PROPERTY_AARCH64_FEATURE_1_PAC; 1754 } 1755 ret &= features; 1756 } 1757 1758 // Force enable Shadow Stack. 1759 if (config->zShstk) 1760 ret |= GNU_PROPERTY_X86_FEATURE_1_SHSTK; 1761 1762 return ret; 1763 } 1764 1765 // Do actual linking. Note that when this function is called, 1766 // all linker scripts have already been parsed. 1767 template <class ELFT> void LinkerDriver::link(opt::InputArgList &args) { 1768 llvm::TimeTraceScope timeScope("Link", StringRef("LinkerDriver::Link")); 1769 // If a -hash-style option was not given, set to a default value, 1770 // which varies depending on the target. 1771 if (!args.hasArg(OPT_hash_style)) { 1772 if (config->emachine == EM_MIPS) 1773 config->sysvHash = true; 1774 else 1775 config->sysvHash = config->gnuHash = true; 1776 } 1777 1778 // Default output filename is "a.out" by the Unix tradition. 1779 if (config->outputFile.empty()) 1780 config->outputFile = "a.out"; 1781 1782 // Fail early if the output file or map file is not writable. If a user has a 1783 // long link, e.g. due to a large LTO link, they do not wish to run it and 1784 // find that it failed because there was a mistake in their command-line. 1785 if (auto e = tryCreateFile(config->outputFile)) 1786 error("cannot open output file " + config->outputFile + ": " + e.message()); 1787 if (auto e = tryCreateFile(config->mapFile)) 1788 error("cannot open map file " + config->mapFile + ": " + e.message()); 1789 if (errorCount()) 1790 return; 1791 1792 // Use default entry point name if no name was given via the command 1793 // line nor linker scripts. For some reason, MIPS entry point name is 1794 // different from others. 1795 config->warnMissingEntry = 1796 (!config->entry.empty() || (!config->shared && !config->relocatable)); 1797 if (config->entry.empty() && !config->relocatable) 1798 config->entry = (config->emachine == EM_MIPS) ? "__start" : "_start"; 1799 1800 // Handle --trace-symbol. 1801 for (auto *arg : args.filtered(OPT_trace_symbol)) 1802 symtab->insert(arg->getValue())->traced = true; 1803 1804 // Add all files to the symbol table. This will add almost all 1805 // symbols that we need to the symbol table. This process might 1806 // add files to the link, via autolinking, these files are always 1807 // appended to the Files vector. 1808 { 1809 llvm::TimeTraceScope timeScope("Parse input files"); 1810 for (size_t i = 0; i < files.size(); ++i) 1811 parseFile(files[i]); 1812 } 1813 1814 // Now that we have every file, we can decide if we will need a 1815 // dynamic symbol table. 1816 // We need one if we were asked to export dynamic symbols or if we are 1817 // producing a shared library. 1818 // We also need one if any shared libraries are used and for pie executables 1819 // (probably because the dynamic linker needs it). 1820 config->hasDynSymTab = 1821 !sharedFiles.empty() || config->isPic || config->exportDynamic; 1822 1823 // Some symbols (such as __ehdr_start) are defined lazily only when there 1824 // are undefined symbols for them, so we add these to trigger that logic. 1825 for (StringRef name : script->referencedSymbols) 1826 addUndefined(name); 1827 1828 // Handle the `--undefined <sym>` options. 1829 for (StringRef arg : config->undefined) 1830 if (Symbol *sym = symtab->find(arg)) 1831 handleUndefined(sym); 1832 1833 // If an entry symbol is in a static archive, pull out that file now. 1834 if (Symbol *sym = symtab->find(config->entry)) 1835 handleUndefined(sym); 1836 1837 // Handle the `--undefined-glob <pattern>` options. 1838 for (StringRef pat : args::getStrings(args, OPT_undefined_glob)) 1839 handleUndefinedGlob(pat); 1840 1841 // Mark -init and -fini symbols so that the LTO doesn't eliminate them. 1842 if (Symbol *sym = symtab->find(config->init)) 1843 sym->isUsedInRegularObj = true; 1844 if (Symbol *sym = symtab->find(config->fini)) 1845 sym->isUsedInRegularObj = true; 1846 1847 // If any of our inputs are bitcode files, the LTO code generator may create 1848 // references to certain library functions that might not be explicit in the 1849 // bitcode file's symbol table. If any of those library functions are defined 1850 // in a bitcode file in an archive member, we need to arrange to use LTO to 1851 // compile those archive members by adding them to the link beforehand. 1852 // 1853 // However, adding all libcall symbols to the link can have undesired 1854 // consequences. For example, the libgcc implementation of 1855 // __sync_val_compare_and_swap_8 on 32-bit ARM pulls in an .init_array entry 1856 // that aborts the program if the Linux kernel does not support 64-bit 1857 // atomics, which would prevent the program from running even if it does not 1858 // use 64-bit atomics. 1859 // 1860 // Therefore, we only add libcall symbols to the link before LTO if we have 1861 // to, i.e. if the symbol's definition is in bitcode. Any other required 1862 // libcall symbols will be added to the link after LTO when we add the LTO 1863 // object file to the link. 1864 if (!bitcodeFiles.empty()) 1865 for (auto *s : lto::LTO::getRuntimeLibcallSymbols()) 1866 handleLibcall(s); 1867 1868 // Return if there were name resolution errors. 1869 if (errorCount()) 1870 return; 1871 1872 // We want to declare linker script's symbols early, 1873 // so that we can version them. 1874 // They also might be exported if referenced by DSOs. 1875 script->declareSymbols(); 1876 1877 // Handle the -exclude-libs option. 1878 if (args.hasArg(OPT_exclude_libs)) 1879 excludeLibs(args); 1880 1881 // Create elfHeader early. We need a dummy section in 1882 // addReservedSymbols to mark the created symbols as not absolute. 1883 Out::elfHeader = make<OutputSection>("", 0, SHF_ALLOC); 1884 Out::elfHeader->size = sizeof(typename ELFT::Ehdr); 1885 1886 // Create wrapped symbols for -wrap option. 1887 std::vector<WrappedSymbol> wrapped = addWrappedSymbols(args); 1888 1889 // We need to create some reserved symbols such as _end. Create them. 1890 if (!config->relocatable) 1891 addReservedSymbols(); 1892 1893 // Apply version scripts. 1894 // 1895 // For a relocatable output, version scripts don't make sense, and 1896 // parsing a symbol version string (e.g. dropping "@ver1" from a symbol 1897 // name "foo@ver1") rather do harm, so we don't call this if -r is given. 1898 if (!config->relocatable) 1899 symtab->scanVersionScript(); 1900 1901 // Do link-time optimization if given files are LLVM bitcode files. 1902 // This compiles bitcode files into real object files. 1903 // 1904 // With this the symbol table should be complete. After this, no new names 1905 // except a few linker-synthesized ones will be added to the symbol table. 1906 compileBitcodeFiles<ELFT>(); 1907 if (errorCount()) 1908 return; 1909 1910 // If -thinlto-index-only is given, we should create only "index 1911 // files" and not object files. Index file creation is already done 1912 // in addCombinedLTOObject, so we are done if that's the case. 1913 if (config->thinLTOIndexOnly) 1914 return; 1915 1916 // Likewise, --plugin-opt=emit-llvm is an option to make LTO create 1917 // an output file in bitcode and exit, so that you can just get a 1918 // combined bitcode file. 1919 if (config->emitLLVM) 1920 return; 1921 1922 // Apply symbol renames for -wrap. 1923 if (!wrapped.empty()) 1924 wrapSymbols(wrapped); 1925 1926 // Now that we have a complete list of input files. 1927 // Beyond this point, no new files are added. 1928 // Aggregate all input sections into one place. 1929 for (InputFile *f : objectFiles) 1930 for (InputSectionBase *s : f->getSections()) 1931 if (s && s != &InputSection::discarded) 1932 inputSections.push_back(s); 1933 for (BinaryFile *f : binaryFiles) 1934 for (InputSectionBase *s : f->getSections()) 1935 inputSections.push_back(cast<InputSection>(s)); 1936 1937 llvm::erase_if(inputSections, [](InputSectionBase *s) { 1938 if (s->type == SHT_LLVM_SYMPART) { 1939 readSymbolPartitionSection<ELFT>(s); 1940 return true; 1941 } 1942 1943 // We do not want to emit debug sections if --strip-all 1944 // or -strip-debug are given. 1945 if (config->strip == StripPolicy::None) 1946 return false; 1947 1948 if (isDebugSection(*s)) 1949 return true; 1950 if (auto *isec = dyn_cast<InputSection>(s)) 1951 if (InputSectionBase *rel = isec->getRelocatedSection()) 1952 if (isDebugSection(*rel)) 1953 return true; 1954 1955 return false; 1956 }); 1957 1958 // Now that the number of partitions is fixed, save a pointer to the main 1959 // partition. 1960 mainPart = &partitions[0]; 1961 1962 // Read .note.gnu.property sections from input object files which 1963 // contain a hint to tweak linker's and loader's behaviors. 1964 config->andFeatures = getAndFeatures<ELFT>(); 1965 1966 // The Target instance handles target-specific stuff, such as applying 1967 // relocations or writing a PLT section. It also contains target-dependent 1968 // values such as a default image base address. 1969 target = getTarget(); 1970 1971 config->eflags = target->calcEFlags(); 1972 // maxPageSize (sometimes called abi page size) is the maximum page size that 1973 // the output can be run on. For example if the OS can use 4k or 64k page 1974 // sizes then maxPageSize must be 64k for the output to be useable on both. 1975 // All important alignment decisions must use this value. 1976 config->maxPageSize = getMaxPageSize(args); 1977 // commonPageSize is the most common page size that the output will be run on. 1978 // For example if an OS can use 4k or 64k page sizes and 4k is more common 1979 // than 64k then commonPageSize is set to 4k. commonPageSize can be used for 1980 // optimizations such as DATA_SEGMENT_ALIGN in linker scripts. LLD's use of it 1981 // is limited to writing trap instructions on the last executable segment. 1982 config->commonPageSize = getCommonPageSize(args); 1983 1984 config->imageBase = getImageBase(args); 1985 1986 if (config->emachine == EM_ARM) { 1987 // FIXME: These warnings can be removed when lld only uses these features 1988 // when the input objects have been compiled with an architecture that 1989 // supports them. 1990 if (config->armHasBlx == false) 1991 warn("lld uses blx instruction, no object with architecture supporting " 1992 "feature detected"); 1993 } 1994 1995 // This adds a .comment section containing a version string. 1996 if (!config->relocatable) 1997 inputSections.push_back(createCommentSection()); 1998 1999 // Replace common symbols with regular symbols. 2000 replaceCommonSymbols(); 2001 2002 // Split SHF_MERGE and .eh_frame sections into pieces in preparation for garbage collection. 2003 splitSections<ELFT>(); 2004 2005 // Garbage collection and removal of shared symbols from unused shared objects. 2006 markLive<ELFT>(); 2007 demoteSharedSymbols(); 2008 2009 // Make copies of any input sections that need to be copied into each 2010 // partition. 2011 copySectionsIntoPartitions(); 2012 2013 // Create synthesized sections such as .got and .plt. This is called before 2014 // processSectionCommands() so that they can be placed by SECTIONS commands. 2015 createSyntheticSections<ELFT>(); 2016 2017 // Some input sections that are used for exception handling need to be moved 2018 // into synthetic sections. Do that now so that they aren't assigned to 2019 // output sections in the usual way. 2020 if (!config->relocatable) 2021 combineEhSections(); 2022 2023 // Create output sections described by SECTIONS commands. 2024 script->processSectionCommands(); 2025 2026 // Linker scripts control how input sections are assigned to output sections. 2027 // Input sections that were not handled by scripts are called "orphans", and 2028 // they are assigned to output sections by the default rule. Process that. 2029 script->addOrphanSections(); 2030 2031 // Migrate InputSectionDescription::sectionBases to sections. This includes 2032 // merging MergeInputSections into a single MergeSyntheticSection. From this 2033 // point onwards InputSectionDescription::sections should be used instead of 2034 // sectionBases. 2035 for (BaseCommand *base : script->sectionCommands) 2036 if (auto *sec = dyn_cast<OutputSection>(base)) 2037 sec->finalizeInputSections(); 2038 llvm::erase_if(inputSections, 2039 [](InputSectionBase *s) { return isa<MergeInputSection>(s); }); 2040 2041 // Two input sections with different output sections should not be folded. 2042 // ICF runs after processSectionCommands() so that we know the output sections. 2043 if (config->icf != ICFLevel::None) { 2044 findKeepUniqueSections<ELFT>(args); 2045 doIcf<ELFT>(); 2046 } 2047 2048 // Read the callgraph now that we know what was gced or icfed 2049 if (config->callGraphProfileSort) { 2050 if (auto *arg = args.getLastArg(OPT_call_graph_ordering_file)) 2051 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 2052 readCallGraph(*buffer); 2053 readCallGraphsFromObjectFiles<ELFT>(); 2054 } 2055 2056 // Write the result to the file. 2057 writeResult<ELFT>(); 2058 } 2059 2060 } // namespace elf 2061 } // namespace lld 2062