1 //===- Writer.cpp ---------------------------------------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #include "Writer.h" 10 #include "Config.h" 11 #include "InputFiles.h" 12 #include "InputSection.h" 13 #include "MergedOutputSection.h" 14 #include "OutputSection.h" 15 #include "OutputSegment.h" 16 #include "SymbolTable.h" 17 #include "Symbols.h" 18 #include "SyntheticSections.h" 19 #include "Target.h" 20 #include "UnwindInfoSection.h" 21 22 #include "lld/Common/ErrorHandler.h" 23 #include "lld/Common/Memory.h" 24 #include "llvm/BinaryFormat/MachO.h" 25 #include "llvm/Config/llvm-config.h" 26 #include "llvm/Support/LEB128.h" 27 #include "llvm/Support/MathExtras.h" 28 #include "llvm/Support/Path.h" 29 #include "llvm/Support/xxhash.h" 30 31 #include <algorithm> 32 33 using namespace llvm; 34 using namespace llvm::MachO; 35 using namespace llvm::sys; 36 using namespace lld; 37 using namespace lld::macho; 38 39 namespace { 40 class LCUuid; 41 42 class Writer { 43 public: 44 Writer() : buffer(errorHandler().outputBuffer) {} 45 46 void scanRelocations(); 47 void scanSymbols(); 48 void createOutputSections(); 49 void createLoadCommands(); 50 void assignAddresses(OutputSegment *); 51 52 void openFile(); 53 void writeSections(); 54 void writeUuid(); 55 56 void run(); 57 58 std::unique_ptr<FileOutputBuffer> &buffer; 59 uint64_t addr = 0; 60 uint64_t fileOff = 0; 61 MachHeaderSection *header = nullptr; 62 StringTableSection *stringTableSection = nullptr; 63 SymtabSection *symtabSection = nullptr; 64 IndirectSymtabSection *indirectSymtabSection = nullptr; 65 UnwindInfoSection *unwindInfoSection = nullptr; 66 LCUuid *uuidCommand = nullptr; 67 }; 68 69 // LC_DYLD_INFO_ONLY stores the offsets of symbol import/export information. 70 class LCDyldInfo : public LoadCommand { 71 public: 72 LCDyldInfo(RebaseSection *rebaseSection, BindingSection *bindingSection, 73 WeakBindingSection *weakBindingSection, 74 LazyBindingSection *lazyBindingSection, 75 ExportSection *exportSection) 76 : rebaseSection(rebaseSection), bindingSection(bindingSection), 77 weakBindingSection(weakBindingSection), 78 lazyBindingSection(lazyBindingSection), exportSection(exportSection) {} 79 80 uint32_t getSize() const override { return sizeof(dyld_info_command); } 81 82 void writeTo(uint8_t *buf) const override { 83 auto *c = reinterpret_cast<dyld_info_command *>(buf); 84 c->cmd = LC_DYLD_INFO_ONLY; 85 c->cmdsize = getSize(); 86 if (rebaseSection->isNeeded()) { 87 c->rebase_off = rebaseSection->fileOff; 88 c->rebase_size = rebaseSection->getFileSize(); 89 } 90 if (bindingSection->isNeeded()) { 91 c->bind_off = bindingSection->fileOff; 92 c->bind_size = bindingSection->getFileSize(); 93 } 94 if (weakBindingSection->isNeeded()) { 95 c->weak_bind_off = weakBindingSection->fileOff; 96 c->weak_bind_size = weakBindingSection->getFileSize(); 97 } 98 if (lazyBindingSection->isNeeded()) { 99 c->lazy_bind_off = lazyBindingSection->fileOff; 100 c->lazy_bind_size = lazyBindingSection->getFileSize(); 101 } 102 if (exportSection->isNeeded()) { 103 c->export_off = exportSection->fileOff; 104 c->export_size = exportSection->getFileSize(); 105 } 106 } 107 108 RebaseSection *rebaseSection; 109 BindingSection *bindingSection; 110 WeakBindingSection *weakBindingSection; 111 LazyBindingSection *lazyBindingSection; 112 ExportSection *exportSection; 113 }; 114 115 class LCDysymtab : public LoadCommand { 116 public: 117 LCDysymtab(SymtabSection *symtabSection, 118 IndirectSymtabSection *indirectSymtabSection) 119 : symtabSection(symtabSection), 120 indirectSymtabSection(indirectSymtabSection) {} 121 122 uint32_t getSize() const override { return sizeof(dysymtab_command); } 123 124 void writeTo(uint8_t *buf) const override { 125 auto *c = reinterpret_cast<dysymtab_command *>(buf); 126 c->cmd = LC_DYSYMTAB; 127 c->cmdsize = getSize(); 128 129 c->ilocalsym = 0; 130 c->iextdefsym = c->nlocalsym = symtabSection->getNumLocalSymbols(); 131 c->nextdefsym = symtabSection->getNumExternalSymbols(); 132 c->iundefsym = c->iextdefsym + c->nextdefsym; 133 c->nundefsym = symtabSection->getNumUndefinedSymbols(); 134 135 c->indirectsymoff = indirectSymtabSection->fileOff; 136 c->nindirectsyms = indirectSymtabSection->getNumSymbols(); 137 } 138 139 SymtabSection *symtabSection; 140 IndirectSymtabSection *indirectSymtabSection; 141 }; 142 143 class LCSegment : public LoadCommand { 144 public: 145 LCSegment(StringRef name, OutputSegment *seg) : name(name), seg(seg) {} 146 147 uint32_t getSize() const override { 148 return sizeof(segment_command_64) + 149 seg->numNonHiddenSections() * sizeof(section_64); 150 } 151 152 void writeTo(uint8_t *buf) const override { 153 auto *c = reinterpret_cast<segment_command_64 *>(buf); 154 buf += sizeof(segment_command_64); 155 156 c->cmd = LC_SEGMENT_64; 157 c->cmdsize = getSize(); 158 memcpy(c->segname, name.data(), name.size()); 159 c->fileoff = seg->fileOff; 160 c->maxprot = seg->maxProt; 161 c->initprot = seg->initProt; 162 163 if (seg->getSections().empty()) 164 return; 165 166 c->vmaddr = seg->firstSection()->addr; 167 c->vmsize = 168 seg->lastSection()->addr + seg->lastSection()->getSize() - c->vmaddr; 169 c->nsects = seg->numNonHiddenSections(); 170 171 for (OutputSection *osec : seg->getSections()) { 172 if (!isZeroFill(osec->flags)) { 173 assert(osec->fileOff >= seg->fileOff); 174 c->filesize = std::max( 175 c->filesize, osec->fileOff + osec->getFileSize() - seg->fileOff); 176 } 177 178 if (osec->isHidden()) 179 continue; 180 181 auto *sectHdr = reinterpret_cast<section_64 *>(buf); 182 buf += sizeof(section_64); 183 184 memcpy(sectHdr->sectname, osec->name.data(), osec->name.size()); 185 memcpy(sectHdr->segname, name.data(), name.size()); 186 187 sectHdr->addr = osec->addr; 188 sectHdr->offset = osec->fileOff; 189 sectHdr->align = Log2_32(osec->align); 190 sectHdr->flags = osec->flags; 191 sectHdr->size = osec->getSize(); 192 sectHdr->reserved1 = osec->reserved1; 193 sectHdr->reserved2 = osec->reserved2; 194 } 195 } 196 197 private: 198 StringRef name; 199 OutputSegment *seg; 200 }; 201 202 class LCMain : public LoadCommand { 203 uint32_t getSize() const override { return sizeof(entry_point_command); } 204 205 void writeTo(uint8_t *buf) const override { 206 auto *c = reinterpret_cast<entry_point_command *>(buf); 207 c->cmd = LC_MAIN; 208 c->cmdsize = getSize(); 209 210 if (config->entry->isInStubs()) 211 c->entryoff = 212 in.stubs->fileOff + config->entry->stubsIndex * target->stubSize; 213 else 214 c->entryoff = config->entry->getFileOffset(); 215 216 c->stacksize = 0; 217 } 218 }; 219 220 class LCSymtab : public LoadCommand { 221 public: 222 LCSymtab(SymtabSection *symtabSection, StringTableSection *stringTableSection) 223 : symtabSection(symtabSection), stringTableSection(stringTableSection) {} 224 225 uint32_t getSize() const override { return sizeof(symtab_command); } 226 227 void writeTo(uint8_t *buf) const override { 228 auto *c = reinterpret_cast<symtab_command *>(buf); 229 c->cmd = LC_SYMTAB; 230 c->cmdsize = getSize(); 231 c->symoff = symtabSection->fileOff; 232 c->nsyms = symtabSection->getNumSymbols(); 233 c->stroff = stringTableSection->fileOff; 234 c->strsize = stringTableSection->getFileSize(); 235 } 236 237 SymtabSection *symtabSection = nullptr; 238 StringTableSection *stringTableSection = nullptr; 239 }; 240 241 // There are several dylib load commands that share the same structure: 242 // * LC_LOAD_DYLIB 243 // * LC_ID_DYLIB 244 // * LC_REEXPORT_DYLIB 245 class LCDylib : public LoadCommand { 246 public: 247 LCDylib(LoadCommandType type, StringRef path, 248 uint32_t compatibilityVersion = 0, uint32_t currentVersion = 0) 249 : type(type), path(path), compatibilityVersion(compatibilityVersion), 250 currentVersion(currentVersion) { 251 instanceCount++; 252 } 253 254 uint32_t getSize() const override { 255 return alignTo(sizeof(dylib_command) + path.size() + 1, 8); 256 } 257 258 void writeTo(uint8_t *buf) const override { 259 auto *c = reinterpret_cast<dylib_command *>(buf); 260 buf += sizeof(dylib_command); 261 262 c->cmd = type; 263 c->cmdsize = getSize(); 264 c->dylib.name = sizeof(dylib_command); 265 c->dylib.timestamp = 0; 266 c->dylib.compatibility_version = compatibilityVersion; 267 c->dylib.current_version = currentVersion; 268 269 memcpy(buf, path.data(), path.size()); 270 buf[path.size()] = '\0'; 271 } 272 273 static uint32_t getInstanceCount() { return instanceCount; } 274 275 private: 276 LoadCommandType type; 277 StringRef path; 278 uint32_t compatibilityVersion; 279 uint32_t currentVersion; 280 static uint32_t instanceCount; 281 }; 282 283 uint32_t LCDylib::instanceCount = 0; 284 285 class LCLoadDylinker : public LoadCommand { 286 public: 287 uint32_t getSize() const override { 288 return alignTo(sizeof(dylinker_command) + path.size() + 1, 8); 289 } 290 291 void writeTo(uint8_t *buf) const override { 292 auto *c = reinterpret_cast<dylinker_command *>(buf); 293 buf += sizeof(dylinker_command); 294 295 c->cmd = LC_LOAD_DYLINKER; 296 c->cmdsize = getSize(); 297 c->name = sizeof(dylinker_command); 298 299 memcpy(buf, path.data(), path.size()); 300 buf[path.size()] = '\0'; 301 } 302 303 private: 304 // Recent versions of Darwin won't run any binary that has dyld at a 305 // different location. 306 const StringRef path = "/usr/lib/dyld"; 307 }; 308 309 class LCRPath : public LoadCommand { 310 public: 311 LCRPath(StringRef path) : path(path) {} 312 313 uint32_t getSize() const override { 314 return alignTo(sizeof(rpath_command) + path.size() + 1, WordSize); 315 } 316 317 void writeTo(uint8_t *buf) const override { 318 auto *c = reinterpret_cast<rpath_command *>(buf); 319 buf += sizeof(rpath_command); 320 321 c->cmd = LC_RPATH; 322 c->cmdsize = getSize(); 323 c->path = sizeof(rpath_command); 324 325 memcpy(buf, path.data(), path.size()); 326 buf[path.size()] = '\0'; 327 } 328 329 private: 330 StringRef path; 331 }; 332 333 class LCBuildVersion : public LoadCommand { 334 public: 335 LCBuildVersion(const PlatformInfo &platform) : platform(platform) {} 336 337 const int ntools = 1; 338 339 uint32_t getSize() const override { 340 return sizeof(build_version_command) + ntools * sizeof(build_tool_version); 341 } 342 343 void writeTo(uint8_t *buf) const override { 344 auto *c = reinterpret_cast<build_version_command *>(buf); 345 c->cmd = LC_BUILD_VERSION; 346 c->cmdsize = getSize(); 347 c->platform = static_cast<uint32_t>(platform.kind); 348 c->minos = ((platform.minimum.getMajor() << 020) | 349 (platform.minimum.getMinor().getValueOr(0) << 010) | 350 platform.minimum.getSubminor().getValueOr(0)); 351 c->sdk = ((platform.sdk.getMajor() << 020) | 352 (platform.sdk.getMinor().getValueOr(0) << 010) | 353 platform.sdk.getSubminor().getValueOr(0)); 354 c->ntools = ntools; 355 auto *t = reinterpret_cast<build_tool_version *>(&c[1]); 356 t->tool = TOOL_LD; 357 t->version = (LLVM_VERSION_MAJOR << 020) | (LLVM_VERSION_MINOR << 010) | 358 LLVM_VERSION_PATCH; 359 } 360 361 const PlatformInfo &platform; 362 }; 363 364 // Stores a unique identifier for the output file based on an MD5 hash of its 365 // contents. In order to hash the contents, we must first write them, but 366 // LC_UUID itself must be part of the written contents in order for all the 367 // offsets to be calculated correctly. We resolve this circular paradox by 368 // first writing an LC_UUID with an all-zero UUID, then updating the UUID with 369 // its real value later. 370 class LCUuid : public LoadCommand { 371 public: 372 uint32_t getSize() const override { return sizeof(uuid_command); } 373 374 void writeTo(uint8_t *buf) const override { 375 auto *c = reinterpret_cast<uuid_command *>(buf); 376 c->cmd = LC_UUID; 377 c->cmdsize = getSize(); 378 uuidBuf = c->uuid; 379 } 380 381 void writeUuid(uint64_t digest) const { 382 // xxhash only gives us 8 bytes, so put some fixed data in the other half. 383 static_assert(sizeof(uuid_command::uuid) == 16, "unexpected uuid size"); 384 memcpy(uuidBuf, "LLD\xa1UU1D", 8); 385 memcpy(uuidBuf + 8, &digest, 8); 386 387 // RFC 4122 conformance. We need to fix 4 bits in byte 6 and 2 bits in 388 // byte 8. Byte 6 is already fine due to the fixed data we put in. We don't 389 // want to lose bits of the digest in byte 8, so swap that with a byte of 390 // fixed data that happens to have the right bits set. 391 std::swap(uuidBuf[3], uuidBuf[8]); 392 393 // Claim that this is an MD5-based hash. It isn't, but this signals that 394 // this is not a time-based and not a random hash. MD5 seems like the least 395 // bad lie we can put here. 396 assert((uuidBuf[6] & 0xf0) == 0x30 && "See RFC 4122 Sections 4.2.2, 4.1.3"); 397 assert((uuidBuf[8] & 0xc0) == 0x80 && "See RFC 4122 Section 4.2.2"); 398 } 399 400 mutable uint8_t *uuidBuf; 401 }; 402 403 } // namespace 404 405 static void prepareSymbolRelocation(lld::macho::Symbol *sym, 406 const InputSection *isec, const Reloc &r) { 407 const TargetInfo::RelocAttrs &relocAttrs = target->getRelocAttrs(r.type); 408 409 if (relocAttrs.hasAttr(RelocAttrBits::BRANCH)) { 410 prepareBranchTarget(sym); 411 } else if (relocAttrs.hasAttr(RelocAttrBits::GOT | RelocAttrBits::LOAD)) { 412 if (needsBinding(sym)) 413 in.got->addEntry(sym); 414 } else if (relocAttrs.hasAttr(RelocAttrBits::GOT)) { 415 in.got->addEntry(sym); 416 } else if (relocAttrs.hasAttr(RelocAttrBits::TLV | RelocAttrBits::LOAD)) { 417 if (needsBinding(sym)) 418 in.tlvPointers->addEntry(sym); 419 } else if (relocAttrs.hasAttr(RelocAttrBits::TLV)) { 420 // References from thread-local variable sections are treated as offsets 421 // relative to the start of the referent section, and therefore have no 422 // need of rebase opcodes. 423 if (!(isThreadLocalVariables(isec->flags) && isa<Defined>(sym))) 424 addNonLazyBindingEntries(sym, isec, r.offset, r.addend); 425 } 426 } 427 428 void Writer::scanRelocations() { 429 for (InputSection *isec : inputSections) { 430 if (isec->segname == segment_names::ld) { 431 prepareCompactUnwind(isec); 432 continue; 433 } 434 435 for (Reloc &r : isec->relocs) { 436 if (target->hasAttr(r.type, RelocAttrBits::SUBTRAHEND)) 437 continue; 438 if (auto *sym = r.referent.dyn_cast<lld::macho::Symbol *>()) { 439 if (auto *undefined = dyn_cast<Undefined>(sym)) 440 treatUndefinedSymbol(*undefined); 441 else if (target->validateSymbolRelocation(sym, isec, r)) 442 prepareSymbolRelocation(sym, isec, r); 443 } else { 444 assert(r.referent.is<InputSection *>()); 445 if (!r.pcrel) 446 in.rebase->addEntry(isec, r.offset); 447 } 448 } 449 } 450 } 451 452 void Writer::scanSymbols() { 453 for (const macho::Symbol *sym : symtab->getSymbols()) { 454 if (const auto *defined = dyn_cast<Defined>(sym)) { 455 if (defined->overridesWeakDef) 456 in.weakBinding->addNonWeakDefinition(defined); 457 } else if (const auto *dysym = dyn_cast<DylibSymbol>(sym)) { 458 dysym->getFile()->refState = 459 std::max(dysym->getFile()->refState, dysym->refState); 460 } 461 } 462 } 463 464 void Writer::createLoadCommands() { 465 uint8_t segIndex = 0; 466 for (OutputSegment *seg : outputSegments) { 467 in.header->addLoadCommand(make<LCSegment>(seg->name, seg)); 468 seg->index = segIndex++; 469 } 470 471 in.header->addLoadCommand(make<LCDyldInfo>( 472 in.rebase, in.binding, in.weakBinding, in.lazyBinding, in.exports)); 473 in.header->addLoadCommand(make<LCSymtab>(symtabSection, stringTableSection)); 474 in.header->addLoadCommand( 475 make<LCDysymtab>(symtabSection, indirectSymtabSection)); 476 for (StringRef path : config->runtimePaths) 477 in.header->addLoadCommand(make<LCRPath>(path)); 478 479 switch (config->outputType) { 480 case MH_EXECUTE: 481 in.header->addLoadCommand(make<LCLoadDylinker>()); 482 in.header->addLoadCommand(make<LCMain>()); 483 break; 484 case MH_DYLIB: 485 in.header->addLoadCommand(make<LCDylib>(LC_ID_DYLIB, config->installName, 486 config->dylibCompatibilityVersion, 487 config->dylibCurrentVersion)); 488 break; 489 case MH_BUNDLE: 490 break; 491 default: 492 llvm_unreachable("unhandled output file type"); 493 } 494 495 uuidCommand = make<LCUuid>(); 496 in.header->addLoadCommand(uuidCommand); 497 498 in.header->addLoadCommand(make<LCBuildVersion>(config->platform)); 499 500 int64_t dylibOrdinal = 1; 501 for (InputFile *file : inputFiles) { 502 if (auto *dylibFile = dyn_cast<DylibFile>(file)) { 503 if (dylibFile->isBundleLoader) { 504 dylibFile->ordinal = MachO::BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE; 505 // Shortcut since bundle-loader does not re-export the symbols. 506 507 dylibFile->reexport = false; 508 continue; 509 } 510 511 dylibFile->ordinal = dylibOrdinal++; 512 LoadCommandType lcType = 513 dylibFile->forceWeakImport || dylibFile->refState == RefState::Weak 514 ? LC_LOAD_WEAK_DYLIB 515 : LC_LOAD_DYLIB; 516 in.header->addLoadCommand(make<LCDylib>(lcType, dylibFile->dylibName, 517 dylibFile->compatibilityVersion, 518 dylibFile->currentVersion)); 519 520 if (dylibFile->reexport) 521 in.header->addLoadCommand( 522 make<LCDylib>(LC_REEXPORT_DYLIB, dylibFile->dylibName)); 523 } 524 } 525 526 const uint32_t MACOS_MAXPATHLEN = 1024; 527 config->headerPad = std::max( 528 config->headerPad, (config->headerPadMaxInstallNames 529 ? LCDylib::getInstanceCount() * MACOS_MAXPATHLEN 530 : 0)); 531 } 532 533 static size_t getSymbolPriority(const SymbolPriorityEntry &entry, 534 const InputFile *f) { 535 // We don't use toString(InputFile *) here because it returns the full path 536 // for object files, and we only want the basename. 537 StringRef filename; 538 if (f->archiveName.empty()) 539 filename = path::filename(f->getName()); 540 else 541 filename = saver.save(path::filename(f->archiveName) + "(" + 542 path::filename(f->getName()) + ")"); 543 return std::max(entry.objectFiles.lookup(filename), entry.anyObjectFile); 544 } 545 546 // Each section gets assigned the priority of the highest-priority symbol it 547 // contains. 548 static DenseMap<const InputSection *, size_t> buildInputSectionPriorities() { 549 DenseMap<const InputSection *, size_t> sectionPriorities; 550 551 if (config->priorities.empty()) 552 return sectionPriorities; 553 554 auto addSym = [&](Defined &sym) { 555 auto it = config->priorities.find(sym.getName()); 556 if (it == config->priorities.end()) 557 return; 558 559 SymbolPriorityEntry &entry = it->second; 560 size_t &priority = sectionPriorities[sym.isec]; 561 priority = std::max(priority, getSymbolPriority(entry, sym.isec->file)); 562 }; 563 564 // TODO: Make sure this handles weak symbols correctly. 565 for (InputFile *file : inputFiles) 566 if (isa<ObjFile>(file)) 567 for (lld::macho::Symbol *sym : file->symbols) 568 if (auto *d = dyn_cast<Defined>(sym)) 569 addSym(*d); 570 571 return sectionPriorities; 572 } 573 574 static int segmentOrder(OutputSegment *seg) { 575 return StringSwitch<int>(seg->name) 576 .Case(segment_names::pageZero, -4) 577 .Case(segment_names::text, -3) 578 .Case(segment_names::dataConst, -2) 579 .Case(segment_names::data, -1) 580 // Make sure __LINKEDIT is the last segment (i.e. all its hidden 581 // sections must be ordered after other sections). 582 .Case(segment_names::linkEdit, std::numeric_limits<int>::max()) 583 .Default(0); 584 } 585 586 static int sectionOrder(OutputSection *osec) { 587 StringRef segname = osec->parent->name; 588 // Sections are uniquely identified by their segment + section name. 589 if (segname == segment_names::text) { 590 return StringSwitch<int>(osec->name) 591 .Case(section_names::header, -4) 592 .Case(section_names::text, -3) 593 .Case(section_names::stubs, -2) 594 .Case(section_names::stubHelper, -1) 595 .Case(section_names::unwindInfo, std::numeric_limits<int>::max() - 1) 596 .Case(section_names::ehFrame, std::numeric_limits<int>::max()) 597 .Default(0); 598 } else if (segname == segment_names::data) { 599 // For each thread spawned, dyld will initialize its TLVs by copying the 600 // address range from the start of the first thread-local data section to 601 // the end of the last one. We therefore arrange these sections contiguously 602 // to minimize the amount of memory used. Additionally, since zerofill 603 // sections must be at the end of their segments, and since TLV data 604 // sections can be zerofills, we end up putting all TLV data sections at the 605 // end of the segment. 606 switch (sectionType(osec->flags)) { 607 case S_THREAD_LOCAL_REGULAR: 608 return std::numeric_limits<int>::max() - 2; 609 case S_THREAD_LOCAL_ZEROFILL: 610 return std::numeric_limits<int>::max() - 1; 611 case S_ZEROFILL: 612 return std::numeric_limits<int>::max(); 613 default: 614 return StringSwitch<int>(osec->name) 615 .Case(section_names::laSymbolPtr, -2) 616 .Case(section_names::data, -1) 617 .Default(0); 618 } 619 } else if (segname == segment_names::linkEdit) { 620 return StringSwitch<int>(osec->name) 621 .Case(section_names::rebase, -8) 622 .Case(section_names::binding, -7) 623 .Case(section_names::weakBinding, -6) 624 .Case(section_names::lazyBinding, -5) 625 .Case(section_names::export_, -4) 626 .Case(section_names::symbolTable, -3) 627 .Case(section_names::indirectSymbolTable, -2) 628 .Case(section_names::stringTable, -1) 629 .Default(0); 630 } 631 // ZeroFill sections must always be the at the end of their segments, 632 // otherwise subsequent sections may get overwritten with zeroes at runtime. 633 if (sectionType(osec->flags) == S_ZEROFILL) 634 return std::numeric_limits<int>::max(); 635 return 0; 636 } 637 638 template <typename T, typename F> 639 static std::function<bool(T, T)> compareByOrder(F ord) { 640 return [=](T a, T b) { return ord(a) < ord(b); }; 641 } 642 643 // Sorting only can happen once all outputs have been collected. Here we sort 644 // segments, output sections within each segment, and input sections within each 645 // output segment. 646 static void sortSegmentsAndSections() { 647 llvm::stable_sort(outputSegments, 648 compareByOrder<OutputSegment *>(segmentOrder)); 649 650 DenseMap<const InputSection *, size_t> isecPriorities = 651 buildInputSectionPriorities(); 652 653 uint32_t sectionIndex = 0; 654 for (OutputSegment *seg : outputSegments) { 655 seg->sortOutputSections(compareByOrder<OutputSection *>(sectionOrder)); 656 for (OutputSection *osec : seg->getSections()) { 657 // Now that the output sections are sorted, assign the final 658 // output section indices. 659 if (!osec->isHidden()) 660 osec->index = ++sectionIndex; 661 662 if (!firstTLVDataSection && isThreadLocalData(osec->flags)) 663 firstTLVDataSection = osec; 664 665 if (!isecPriorities.empty()) { 666 if (auto *merged = dyn_cast<MergedOutputSection>(osec)) { 667 llvm::stable_sort(merged->inputs, 668 [&](InputSection *a, InputSection *b) { 669 return isecPriorities[a] > isecPriorities[b]; 670 }); 671 } 672 } 673 } 674 } 675 } 676 677 void Writer::createOutputSections() { 678 // First, create hidden sections 679 stringTableSection = make<StringTableSection>(); 680 unwindInfoSection = make<UnwindInfoSection>(); // TODO(gkm): only when no -r 681 symtabSection = make<SymtabSection>(*stringTableSection); 682 indirectSymtabSection = make<IndirectSymtabSection>(); 683 684 switch (config->outputType) { 685 case MH_EXECUTE: 686 make<PageZeroSection>(); 687 break; 688 case MH_DYLIB: 689 case MH_BUNDLE: 690 break; 691 default: 692 llvm_unreachable("unhandled output file type"); 693 } 694 695 // Then merge input sections into output sections. 696 MapVector<std::pair<StringRef, StringRef>, MergedOutputSection *> 697 mergedOutputSections; 698 for (InputSection *isec : inputSections) { 699 MergedOutputSection *&osec = 700 mergedOutputSections[{isec->segname, isec->name}]; 701 if (osec == nullptr) 702 osec = make<MergedOutputSection>(isec->name); 703 osec->mergeInput(isec); 704 } 705 706 for (const auto &it : mergedOutputSections) { 707 StringRef segname = it.first.first; 708 MergedOutputSection *osec = it.second; 709 if (unwindInfoSection && segname == segment_names::ld) { 710 assert(osec->name == section_names::compactUnwind); 711 unwindInfoSection->setCompactUnwindSection(osec); 712 } else { 713 getOrCreateOutputSegment(segname)->addOutputSection(osec); 714 } 715 } 716 717 for (SyntheticSection *ssec : syntheticSections) { 718 auto it = mergedOutputSections.find({ssec->segname, ssec->name}); 719 if (it == mergedOutputSections.end()) { 720 if (ssec->isNeeded()) 721 getOrCreateOutputSegment(ssec->segname)->addOutputSection(ssec); 722 } else { 723 error("section from " + toString(it->second->firstSection()->file) + 724 " conflicts with synthetic section " + ssec->segname + "," + 725 ssec->name); 726 } 727 } 728 } 729 730 void Writer::assignAddresses(OutputSegment *seg) { 731 uint64_t pageSize = target->getPageSize(); 732 addr = alignTo(addr, pageSize); 733 fileOff = alignTo(fileOff, pageSize); 734 seg->fileOff = fileOff; 735 736 for (OutputSection *osec : seg->getSections()) { 737 if (!osec->isNeeded()) 738 continue; 739 addr = alignTo(addr, osec->align); 740 fileOff = alignTo(fileOff, osec->align); 741 osec->addr = addr; 742 osec->fileOff = isZeroFill(osec->flags) ? 0 : fileOff; 743 osec->finalize(); 744 745 addr += osec->getSize(); 746 fileOff += osec->getFileSize(); 747 } 748 } 749 750 void Writer::openFile() { 751 Expected<std::unique_ptr<FileOutputBuffer>> bufferOrErr = 752 FileOutputBuffer::create(config->outputFile, fileOff, 753 FileOutputBuffer::F_executable); 754 755 if (!bufferOrErr) 756 error("failed to open " + config->outputFile + ": " + 757 llvm::toString(bufferOrErr.takeError())); 758 else 759 buffer = std::move(*bufferOrErr); 760 } 761 762 void Writer::writeSections() { 763 uint8_t *buf = buffer->getBufferStart(); 764 for (OutputSegment *seg : outputSegments) 765 for (OutputSection *osec : seg->getSections()) 766 osec->writeTo(buf + osec->fileOff); 767 } 768 769 void Writer::writeUuid() { 770 uint64_t digest = 771 xxHash64({buffer->getBufferStart(), buffer->getBufferEnd()}); 772 uuidCommand->writeUuid(digest); 773 } 774 775 void Writer::run() { 776 // dyld requires __LINKEDIT segment to always exist (even if empty). 777 OutputSegment *linkEditSegment = 778 getOrCreateOutputSegment(segment_names::linkEdit); 779 780 prepareBranchTarget(config->entry); 781 scanRelocations(); 782 if (in.stubHelper->isNeeded()) 783 in.stubHelper->setup(); 784 scanSymbols(); 785 786 // Sort and assign sections to their respective segments. No more sections nor 787 // segments may be created after these methods run. 788 createOutputSections(); 789 sortSegmentsAndSections(); 790 791 createLoadCommands(); 792 793 // Ensure that segments (and the sections they contain) are allocated 794 // addresses in ascending order, which dyld requires. 795 // 796 // Note that at this point, __LINKEDIT sections are empty, but we need to 797 // determine addresses of other segments/sections before generating its 798 // contents. 799 for (OutputSegment *seg : outputSegments) 800 if (seg != linkEditSegment) 801 assignAddresses(seg); 802 803 // Fill __LINKEDIT contents. 804 in.rebase->finalizeContents(); 805 in.binding->finalizeContents(); 806 in.weakBinding->finalizeContents(); 807 in.lazyBinding->finalizeContents(); 808 in.exports->finalizeContents(); 809 symtabSection->finalizeContents(); 810 indirectSymtabSection->finalizeContents(); 811 812 // Now that __LINKEDIT is filled out, do a proper calculation of its 813 // addresses and offsets. 814 assignAddresses(linkEditSegment); 815 816 openFile(); 817 if (errorCount()) 818 return; 819 820 writeSections(); 821 writeUuid(); 822 823 if (auto e = buffer->commit()) 824 error("failed to write to the output file: " + toString(std::move(e))); 825 } 826 827 void macho::writeResult() { Writer().run(); } 828 829 void macho::createSyntheticSections() { 830 in.header = make<MachHeaderSection>(); 831 in.rebase = make<RebaseSection>(); 832 in.binding = make<BindingSection>(); 833 in.weakBinding = make<WeakBindingSection>(); 834 in.lazyBinding = make<LazyBindingSection>(); 835 in.exports = make<ExportSection>(); 836 in.got = make<GotSection>(); 837 in.tlvPointers = make<TlvPointerSection>(); 838 in.lazyPointers = make<LazyPointerSection>(); 839 in.stubs = make<StubsSection>(); 840 in.stubHelper = make<StubHelperSection>(); 841 in.imageLoaderCache = make<ImageLoaderCacheSection>(); 842 } 843 844 OutputSection *macho::firstTLVDataSection = nullptr; 845