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 "ConcatOutputSection.h"
11 #include "Config.h"
12 #include "InputFiles.h"
13 #include "InputSection.h"
14 #include "MapFile.h"
15 #include "OutputSection.h"
16 #include "OutputSegment.h"
17 #include "SectionPriorities.h"
18 #include "SymbolTable.h"
19 #include "Symbols.h"
20 #include "SyntheticSections.h"
21 #include "Target.h"
22 #include "UnwindInfoSection.h"
23 
24 #include "lld/Common/Arrays.h"
25 #include "lld/Common/CommonLinkerContext.h"
26 #include "llvm/BinaryFormat/MachO.h"
27 #include "llvm/Config/llvm-config.h"
28 #include "llvm/Support/LEB128.h"
29 #include "llvm/Support/Parallel.h"
30 #include "llvm/Support/Path.h"
31 #include "llvm/Support/ThreadPool.h"
32 #include "llvm/Support/TimeProfiler.h"
33 #include "llvm/Support/xxhash.h"
34 
35 #include <algorithm>
36 
37 using namespace llvm;
38 using namespace llvm::MachO;
39 using namespace llvm::sys;
40 using namespace lld;
41 using namespace lld::macho;
42 
43 namespace {
44 class LCUuid;
45 
46 class Writer {
47 public:
Writer()48   Writer() : buffer(errorHandler().outputBuffer) {}
49 
50   void treatSpecialUndefineds();
51   void scanRelocations();
52   void scanSymbols();
53   template <class LP> void createOutputSections();
54   template <class LP> void createLoadCommands();
55   void finalizeAddresses();
56   void finalizeLinkEditSegment();
57   void assignAddresses(OutputSegment *);
58 
59   void openFile();
60   void writeSections();
61   void applyOptimizationHints();
62   void buildFixupChains();
63   void writeUuid();
64   void writeCodeSignature();
65   void writeOutputFile();
66 
67   template <class LP> void run();
68 
69   ThreadPool threadPool;
70   std::unique_ptr<FileOutputBuffer> &buffer;
71   uint64_t addr = 0;
72   uint64_t fileOff = 0;
73   MachHeaderSection *header = nullptr;
74   StringTableSection *stringTableSection = nullptr;
75   SymtabSection *symtabSection = nullptr;
76   IndirectSymtabSection *indirectSymtabSection = nullptr;
77   CodeSignatureSection *codeSignatureSection = nullptr;
78   DataInCodeSection *dataInCodeSection = nullptr;
79   FunctionStartsSection *functionStartsSection = nullptr;
80 
81   LCUuid *uuidCommand = nullptr;
82   OutputSegment *linkEditSegment = nullptr;
83 };
84 
85 // LC_DYLD_INFO_ONLY stores the offsets of symbol import/export information.
86 class LCDyldInfo final : public LoadCommand {
87 public:
LCDyldInfo(RebaseSection * rebaseSection,BindingSection * bindingSection,WeakBindingSection * weakBindingSection,LazyBindingSection * lazyBindingSection,ExportSection * exportSection)88   LCDyldInfo(RebaseSection *rebaseSection, BindingSection *bindingSection,
89              WeakBindingSection *weakBindingSection,
90              LazyBindingSection *lazyBindingSection,
91              ExportSection *exportSection)
92       : rebaseSection(rebaseSection), bindingSection(bindingSection),
93         weakBindingSection(weakBindingSection),
94         lazyBindingSection(lazyBindingSection), exportSection(exportSection) {}
95 
getSize() const96   uint32_t getSize() const override { return sizeof(dyld_info_command); }
97 
writeTo(uint8_t * buf) const98   void writeTo(uint8_t *buf) const override {
99     auto *c = reinterpret_cast<dyld_info_command *>(buf);
100     c->cmd = LC_DYLD_INFO_ONLY;
101     c->cmdsize = getSize();
102     if (rebaseSection->isNeeded()) {
103       c->rebase_off = rebaseSection->fileOff;
104       c->rebase_size = rebaseSection->getFileSize();
105     }
106     if (bindingSection->isNeeded()) {
107       c->bind_off = bindingSection->fileOff;
108       c->bind_size = bindingSection->getFileSize();
109     }
110     if (weakBindingSection->isNeeded()) {
111       c->weak_bind_off = weakBindingSection->fileOff;
112       c->weak_bind_size = weakBindingSection->getFileSize();
113     }
114     if (lazyBindingSection->isNeeded()) {
115       c->lazy_bind_off = lazyBindingSection->fileOff;
116       c->lazy_bind_size = lazyBindingSection->getFileSize();
117     }
118     if (exportSection->isNeeded()) {
119       c->export_off = exportSection->fileOff;
120       c->export_size = exportSection->getFileSize();
121     }
122   }
123 
124   RebaseSection *rebaseSection;
125   BindingSection *bindingSection;
126   WeakBindingSection *weakBindingSection;
127   LazyBindingSection *lazyBindingSection;
128   ExportSection *exportSection;
129 };
130 
131 class LCSubFramework final : public LoadCommand {
132 public:
LCSubFramework(StringRef umbrella)133   LCSubFramework(StringRef umbrella) : umbrella(umbrella) {}
134 
getSize() const135   uint32_t getSize() const override {
136     return alignToPowerOf2(sizeof(sub_framework_command) + umbrella.size() + 1,
137                            target->wordSize);
138   }
139 
writeTo(uint8_t * buf) const140   void writeTo(uint8_t *buf) const override {
141     auto *c = reinterpret_cast<sub_framework_command *>(buf);
142     buf += sizeof(sub_framework_command);
143 
144     c->cmd = LC_SUB_FRAMEWORK;
145     c->cmdsize = getSize();
146     c->umbrella = sizeof(sub_framework_command);
147 
148     memcpy(buf, umbrella.data(), umbrella.size());
149     buf[umbrella.size()] = '\0';
150   }
151 
152 private:
153   const StringRef umbrella;
154 };
155 
156 class LCFunctionStarts final : public LoadCommand {
157 public:
LCFunctionStarts(FunctionStartsSection * functionStartsSection)158   explicit LCFunctionStarts(FunctionStartsSection *functionStartsSection)
159       : functionStartsSection(functionStartsSection) {}
160 
getSize() const161   uint32_t getSize() const override { return sizeof(linkedit_data_command); }
162 
writeTo(uint8_t * buf) const163   void writeTo(uint8_t *buf) const override {
164     auto *c = reinterpret_cast<linkedit_data_command *>(buf);
165     c->cmd = LC_FUNCTION_STARTS;
166     c->cmdsize = getSize();
167     c->dataoff = functionStartsSection->fileOff;
168     c->datasize = functionStartsSection->getFileSize();
169   }
170 
171 private:
172   FunctionStartsSection *functionStartsSection;
173 };
174 
175 class LCDataInCode final : public LoadCommand {
176 public:
LCDataInCode(DataInCodeSection * dataInCodeSection)177   explicit LCDataInCode(DataInCodeSection *dataInCodeSection)
178       : dataInCodeSection(dataInCodeSection) {}
179 
getSize() const180   uint32_t getSize() const override { return sizeof(linkedit_data_command); }
181 
writeTo(uint8_t * buf) const182   void writeTo(uint8_t *buf) const override {
183     auto *c = reinterpret_cast<linkedit_data_command *>(buf);
184     c->cmd = LC_DATA_IN_CODE;
185     c->cmdsize = getSize();
186     c->dataoff = dataInCodeSection->fileOff;
187     c->datasize = dataInCodeSection->getFileSize();
188   }
189 
190 private:
191   DataInCodeSection *dataInCodeSection;
192 };
193 
194 class LCDysymtab final : public LoadCommand {
195 public:
LCDysymtab(SymtabSection * symtabSection,IndirectSymtabSection * indirectSymtabSection)196   LCDysymtab(SymtabSection *symtabSection,
197              IndirectSymtabSection *indirectSymtabSection)
198       : symtabSection(symtabSection),
199         indirectSymtabSection(indirectSymtabSection) {}
200 
getSize() const201   uint32_t getSize() const override { return sizeof(dysymtab_command); }
202 
writeTo(uint8_t * buf) const203   void writeTo(uint8_t *buf) const override {
204     auto *c = reinterpret_cast<dysymtab_command *>(buf);
205     c->cmd = LC_DYSYMTAB;
206     c->cmdsize = getSize();
207 
208     c->ilocalsym = 0;
209     c->iextdefsym = c->nlocalsym = symtabSection->getNumLocalSymbols();
210     c->nextdefsym = symtabSection->getNumExternalSymbols();
211     c->iundefsym = c->iextdefsym + c->nextdefsym;
212     c->nundefsym = symtabSection->getNumUndefinedSymbols();
213 
214     c->indirectsymoff = indirectSymtabSection->fileOff;
215     c->nindirectsyms = indirectSymtabSection->getNumSymbols();
216   }
217 
218   SymtabSection *symtabSection;
219   IndirectSymtabSection *indirectSymtabSection;
220 };
221 
222 template <class LP> class LCSegment final : public LoadCommand {
223 public:
LCSegment(StringRef name,OutputSegment * seg)224   LCSegment(StringRef name, OutputSegment *seg) : name(name), seg(seg) {}
225 
getSize() const226   uint32_t getSize() const override {
227     return sizeof(typename LP::segment_command) +
228            seg->numNonHiddenSections() * sizeof(typename LP::section);
229   }
230 
writeTo(uint8_t * buf) const231   void writeTo(uint8_t *buf) const override {
232     using SegmentCommand = typename LP::segment_command;
233     using SectionHeader = typename LP::section;
234 
235     auto *c = reinterpret_cast<SegmentCommand *>(buf);
236     buf += sizeof(SegmentCommand);
237 
238     c->cmd = LP::segmentLCType;
239     c->cmdsize = getSize();
240     memcpy(c->segname, name.data(), name.size());
241     c->fileoff = seg->fileOff;
242     c->maxprot = seg->maxProt;
243     c->initprot = seg->initProt;
244 
245     c->vmaddr = seg->addr;
246     c->vmsize = seg->vmSize;
247     c->filesize = seg->fileSize;
248     c->nsects = seg->numNonHiddenSections();
249     c->flags = seg->flags;
250 
251     for (const OutputSection *osec : seg->getSections()) {
252       if (osec->isHidden())
253         continue;
254 
255       auto *sectHdr = reinterpret_cast<SectionHeader *>(buf);
256       buf += sizeof(SectionHeader);
257 
258       memcpy(sectHdr->sectname, osec->name.data(), osec->name.size());
259       memcpy(sectHdr->segname, name.data(), name.size());
260 
261       sectHdr->addr = osec->addr;
262       sectHdr->offset = osec->fileOff;
263       sectHdr->align = Log2_32(osec->align);
264       sectHdr->flags = osec->flags;
265       sectHdr->size = osec->getSize();
266       sectHdr->reserved1 = osec->reserved1;
267       sectHdr->reserved2 = osec->reserved2;
268     }
269   }
270 
271 private:
272   StringRef name;
273   OutputSegment *seg;
274 };
275 
276 class LCMain final : public LoadCommand {
getSize() const277   uint32_t getSize() const override {
278     return sizeof(structs::entry_point_command);
279   }
280 
writeTo(uint8_t * buf) const281   void writeTo(uint8_t *buf) const override {
282     auto *c = reinterpret_cast<structs::entry_point_command *>(buf);
283     c->cmd = LC_MAIN;
284     c->cmdsize = getSize();
285 
286     if (config->entry->isInStubs())
287       c->entryoff =
288           in.stubs->fileOff + config->entry->stubsIndex * target->stubSize;
289     else
290       c->entryoff = config->entry->getVA() - in.header->addr;
291 
292     c->stacksize = 0;
293   }
294 };
295 
296 class LCSymtab final : public LoadCommand {
297 public:
LCSymtab(SymtabSection * symtabSection,StringTableSection * stringTableSection)298   LCSymtab(SymtabSection *symtabSection, StringTableSection *stringTableSection)
299       : symtabSection(symtabSection), stringTableSection(stringTableSection) {}
300 
getSize() const301   uint32_t getSize() const override { return sizeof(symtab_command); }
302 
writeTo(uint8_t * buf) const303   void writeTo(uint8_t *buf) const override {
304     auto *c = reinterpret_cast<symtab_command *>(buf);
305     c->cmd = LC_SYMTAB;
306     c->cmdsize = getSize();
307     c->symoff = symtabSection->fileOff;
308     c->nsyms = symtabSection->getNumSymbols();
309     c->stroff = stringTableSection->fileOff;
310     c->strsize = stringTableSection->getFileSize();
311   }
312 
313   SymtabSection *symtabSection = nullptr;
314   StringTableSection *stringTableSection = nullptr;
315 };
316 
317 // There are several dylib load commands that share the same structure:
318 //   * LC_LOAD_DYLIB
319 //   * LC_ID_DYLIB
320 //   * LC_REEXPORT_DYLIB
321 class LCDylib final : public LoadCommand {
322 public:
LCDylib(LoadCommandType type,StringRef path,uint32_t compatibilityVersion=0,uint32_t currentVersion=0)323   LCDylib(LoadCommandType type, StringRef path,
324           uint32_t compatibilityVersion = 0, uint32_t currentVersion = 0)
325       : type(type), path(path), compatibilityVersion(compatibilityVersion),
326         currentVersion(currentVersion) {
327     instanceCount++;
328   }
329 
getSize() const330   uint32_t getSize() const override {
331     return alignToPowerOf2(sizeof(dylib_command) + path.size() + 1,
332                            target->wordSize);
333   }
334 
writeTo(uint8_t * buf) const335   void writeTo(uint8_t *buf) const override {
336     auto *c = reinterpret_cast<dylib_command *>(buf);
337     buf += sizeof(dylib_command);
338 
339     c->cmd = type;
340     c->cmdsize = getSize();
341     c->dylib.name = sizeof(dylib_command);
342     c->dylib.timestamp = 0;
343     c->dylib.compatibility_version = compatibilityVersion;
344     c->dylib.current_version = currentVersion;
345 
346     memcpy(buf, path.data(), path.size());
347     buf[path.size()] = '\0';
348   }
349 
getInstanceCount()350   static uint32_t getInstanceCount() { return instanceCount; }
resetInstanceCount()351   static void resetInstanceCount() { instanceCount = 0; }
352 
353 private:
354   LoadCommandType type;
355   StringRef path;
356   uint32_t compatibilityVersion;
357   uint32_t currentVersion;
358   static uint32_t instanceCount;
359 };
360 
361 uint32_t LCDylib::instanceCount = 0;
362 
363 class LCLoadDylinker final : public LoadCommand {
364 public:
getSize() const365   uint32_t getSize() const override {
366     return alignToPowerOf2(sizeof(dylinker_command) + path.size() + 1,
367                            target->wordSize);
368   }
369 
writeTo(uint8_t * buf) const370   void writeTo(uint8_t *buf) const override {
371     auto *c = reinterpret_cast<dylinker_command *>(buf);
372     buf += sizeof(dylinker_command);
373 
374     c->cmd = LC_LOAD_DYLINKER;
375     c->cmdsize = getSize();
376     c->name = sizeof(dylinker_command);
377 
378     memcpy(buf, path.data(), path.size());
379     buf[path.size()] = '\0';
380   }
381 
382 private:
383   // Recent versions of Darwin won't run any binary that has dyld at a
384   // different location.
385   const StringRef path = "/usr/lib/dyld";
386 };
387 
388 class LCRPath final : public LoadCommand {
389 public:
LCRPath(StringRef path)390   explicit LCRPath(StringRef path) : path(path) {}
391 
getSize() const392   uint32_t getSize() const override {
393     return alignToPowerOf2(sizeof(rpath_command) + path.size() + 1,
394                            target->wordSize);
395   }
396 
writeTo(uint8_t * buf) const397   void writeTo(uint8_t *buf) const override {
398     auto *c = reinterpret_cast<rpath_command *>(buf);
399     buf += sizeof(rpath_command);
400 
401     c->cmd = LC_RPATH;
402     c->cmdsize = getSize();
403     c->path = sizeof(rpath_command);
404 
405     memcpy(buf, path.data(), path.size());
406     buf[path.size()] = '\0';
407   }
408 
409 private:
410   StringRef path;
411 };
412 
413 class LCDyldEnv final : public LoadCommand {
414 public:
LCDyldEnv(StringRef name)415   explicit LCDyldEnv(StringRef name) : name(name) {}
416 
getSize() const417   uint32_t getSize() const override {
418     return alignToPowerOf2(sizeof(dyld_env_command) + name.size() + 1,
419                            target->wordSize);
420   }
421 
writeTo(uint8_t * buf) const422   void writeTo(uint8_t *buf) const override {
423     auto *c = reinterpret_cast<dyld_env_command *>(buf);
424     buf += sizeof(dyld_env_command);
425 
426     c->cmd = LC_DYLD_ENVIRONMENT;
427     c->cmdsize = getSize();
428     c->name = sizeof(dyld_env_command);
429 
430     memcpy(buf, name.data(), name.size());
431     buf[name.size()] = '\0';
432   }
433 
434 private:
435   StringRef name;
436 };
437 
438 class LCMinVersion final : public LoadCommand {
439 public:
LCMinVersion(const PlatformInfo & platformInfo)440   explicit LCMinVersion(const PlatformInfo &platformInfo)
441       : platformInfo(platformInfo) {}
442 
getSize() const443   uint32_t getSize() const override { return sizeof(version_min_command); }
444 
writeTo(uint8_t * buf) const445   void writeTo(uint8_t *buf) const override {
446     auto *c = reinterpret_cast<version_min_command *>(buf);
447     switch (platformInfo.target.Platform) {
448     case PLATFORM_MACOS:
449       c->cmd = LC_VERSION_MIN_MACOSX;
450       break;
451     case PLATFORM_IOS:
452     case PLATFORM_IOSSIMULATOR:
453       c->cmd = LC_VERSION_MIN_IPHONEOS;
454       break;
455     case PLATFORM_TVOS:
456     case PLATFORM_TVOSSIMULATOR:
457       c->cmd = LC_VERSION_MIN_TVOS;
458       break;
459     case PLATFORM_WATCHOS:
460     case PLATFORM_WATCHOSSIMULATOR:
461       c->cmd = LC_VERSION_MIN_WATCHOS;
462       break;
463     default:
464       llvm_unreachable("invalid platform");
465       break;
466     }
467     c->cmdsize = getSize();
468     c->version = encodeVersion(platformInfo.target.MinDeployment);
469     c->sdk = encodeVersion(platformInfo.sdk);
470   }
471 
472 private:
473   const PlatformInfo &platformInfo;
474 };
475 
476 class LCBuildVersion final : public LoadCommand {
477 public:
LCBuildVersion(const PlatformInfo & platformInfo)478   explicit LCBuildVersion(const PlatformInfo &platformInfo)
479       : platformInfo(platformInfo) {}
480 
481   const int ntools = 1;
482 
getSize() const483   uint32_t getSize() const override {
484     return sizeof(build_version_command) + ntools * sizeof(build_tool_version);
485   }
486 
writeTo(uint8_t * buf) const487   void writeTo(uint8_t *buf) const override {
488     auto *c = reinterpret_cast<build_version_command *>(buf);
489     c->cmd = LC_BUILD_VERSION;
490     c->cmdsize = getSize();
491 
492     c->platform = static_cast<uint32_t>(platformInfo.target.Platform);
493     c->minos = encodeVersion(platformInfo.target.MinDeployment);
494     c->sdk = encodeVersion(platformInfo.sdk);
495 
496     c->ntools = ntools;
497     auto *t = reinterpret_cast<build_tool_version *>(&c[1]);
498     t->tool = TOOL_LLD;
499     t->version = encodeVersion(VersionTuple(
500         LLVM_VERSION_MAJOR, LLVM_VERSION_MINOR, LLVM_VERSION_PATCH));
501   }
502 
503 private:
504   const PlatformInfo &platformInfo;
505 };
506 
507 // Stores a unique identifier for the output file based on an MD5 hash of its
508 // contents. In order to hash the contents, we must first write them, but
509 // LC_UUID itself must be part of the written contents in order for all the
510 // offsets to be calculated correctly. We resolve this circular paradox by
511 // first writing an LC_UUID with an all-zero UUID, then updating the UUID with
512 // its real value later.
513 class LCUuid final : public LoadCommand {
514 public:
getSize() const515   uint32_t getSize() const override { return sizeof(uuid_command); }
516 
writeTo(uint8_t * buf) const517   void writeTo(uint8_t *buf) const override {
518     auto *c = reinterpret_cast<uuid_command *>(buf);
519     c->cmd = LC_UUID;
520     c->cmdsize = getSize();
521     uuidBuf = c->uuid;
522   }
523 
writeUuid(uint64_t digest) const524   void writeUuid(uint64_t digest) const {
525     // xxhash only gives us 8 bytes, so put some fixed data in the other half.
526     static_assert(sizeof(uuid_command::uuid) == 16, "unexpected uuid size");
527     memcpy(uuidBuf, "LLD\xa1UU1D", 8);
528     memcpy(uuidBuf + 8, &digest, 8);
529 
530     // RFC 4122 conformance. We need to fix 4 bits in byte 6 and 2 bits in
531     // byte 8. Byte 6 is already fine due to the fixed data we put in. We don't
532     // want to lose bits of the digest in byte 8, so swap that with a byte of
533     // fixed data that happens to have the right bits set.
534     std::swap(uuidBuf[3], uuidBuf[8]);
535 
536     // Claim that this is an MD5-based hash. It isn't, but this signals that
537     // this is not a time-based and not a random hash. MD5 seems like the least
538     // bad lie we can put here.
539     assert((uuidBuf[6] & 0xf0) == 0x30 && "See RFC 4122 Sections 4.2.2, 4.1.3");
540     assert((uuidBuf[8] & 0xc0) == 0x80 && "See RFC 4122 Section 4.2.2");
541   }
542 
543   mutable uint8_t *uuidBuf;
544 };
545 
546 template <class LP> class LCEncryptionInfo final : public LoadCommand {
547 public:
getSize() const548   uint32_t getSize() const override {
549     return sizeof(typename LP::encryption_info_command);
550   }
551 
writeTo(uint8_t * buf) const552   void writeTo(uint8_t *buf) const override {
553     using EncryptionInfo = typename LP::encryption_info_command;
554     auto *c = reinterpret_cast<EncryptionInfo *>(buf);
555     buf += sizeof(EncryptionInfo);
556     c->cmd = LP::encryptionInfoLCType;
557     c->cmdsize = getSize();
558     c->cryptoff = in.header->getSize();
559     auto it = find_if(outputSegments, [](const OutputSegment *seg) {
560       return seg->name == segment_names::text;
561     });
562     assert(it != outputSegments.end());
563     c->cryptsize = (*it)->fileSize - c->cryptoff;
564   }
565 };
566 
567 class LCCodeSignature final : public LoadCommand {
568 public:
LCCodeSignature(CodeSignatureSection * section)569   LCCodeSignature(CodeSignatureSection *section) : section(section) {}
570 
getSize() const571   uint32_t getSize() const override { return sizeof(linkedit_data_command); }
572 
writeTo(uint8_t * buf) const573   void writeTo(uint8_t *buf) const override {
574     auto *c = reinterpret_cast<linkedit_data_command *>(buf);
575     c->cmd = LC_CODE_SIGNATURE;
576     c->cmdsize = getSize();
577     c->dataoff = static_cast<uint32_t>(section->fileOff);
578     c->datasize = section->getSize();
579   }
580 
581   CodeSignatureSection *section;
582 };
583 
584 class LCExportsTrie final : public LoadCommand {
585 public:
LCExportsTrie(ExportSection * section)586   LCExportsTrie(ExportSection *section) : section(section) {}
587 
getSize() const588   uint32_t getSize() const override { return sizeof(linkedit_data_command); }
589 
writeTo(uint8_t * buf) const590   void writeTo(uint8_t *buf) const override {
591     auto *c = reinterpret_cast<linkedit_data_command *>(buf);
592     c->cmd = LC_DYLD_EXPORTS_TRIE;
593     c->cmdsize = getSize();
594     c->dataoff = section->fileOff;
595     c->datasize = section->getSize();
596   }
597 
598   ExportSection *section;
599 };
600 
601 class LCChainedFixups final : public LoadCommand {
602 public:
LCChainedFixups(ChainedFixupsSection * section)603   LCChainedFixups(ChainedFixupsSection *section) : section(section) {}
604 
getSize() const605   uint32_t getSize() const override { return sizeof(linkedit_data_command); }
606 
writeTo(uint8_t * buf) const607   void writeTo(uint8_t *buf) const override {
608     auto *c = reinterpret_cast<linkedit_data_command *>(buf);
609     c->cmd = LC_DYLD_CHAINED_FIXUPS;
610     c->cmdsize = getSize();
611     c->dataoff = section->fileOff;
612     c->datasize = section->getSize();
613   }
614 
615   ChainedFixupsSection *section;
616 };
617 
618 } // namespace
619 
treatSpecialUndefineds()620 void Writer::treatSpecialUndefineds() {
621   if (config->entry)
622     if (auto *undefined = dyn_cast<Undefined>(config->entry))
623       treatUndefinedSymbol(*undefined, "the entry point");
624 
625   // FIXME: This prints symbols that are undefined both in input files and
626   // via -u flag twice.
627   for (const Symbol *sym : config->explicitUndefineds) {
628     if (const auto *undefined = dyn_cast<Undefined>(sym))
629       treatUndefinedSymbol(*undefined, "-u");
630   }
631   // Literal exported-symbol names must be defined, but glob
632   // patterns need not match.
633   for (const CachedHashStringRef &cachedName :
634        config->exportedSymbols.literals) {
635     if (const Symbol *sym = symtab->find(cachedName))
636       if (const auto *undefined = dyn_cast<Undefined>(sym))
637         treatUndefinedSymbol(*undefined, "-exported_symbol(s_list)");
638   }
639 }
640 
prepareSymbolRelocation(Symbol * sym,const InputSection * isec,const lld::macho::Reloc & r)641 static void prepareSymbolRelocation(Symbol *sym, const InputSection *isec,
642                                     const lld::macho::Reloc &r) {
643   assert(sym->isLive());
644   const RelocAttrs &relocAttrs = target->getRelocAttrs(r.type);
645 
646   if (relocAttrs.hasAttr(RelocAttrBits::BRANCH)) {
647     if (needsBinding(sym))
648       in.stubs->addEntry(sym);
649   } else if (relocAttrs.hasAttr(RelocAttrBits::GOT)) {
650     if (relocAttrs.hasAttr(RelocAttrBits::POINTER) || needsBinding(sym))
651       in.got->addEntry(sym);
652   } else if (relocAttrs.hasAttr(RelocAttrBits::TLV)) {
653     if (needsBinding(sym))
654       in.tlvPointers->addEntry(sym);
655   } else if (relocAttrs.hasAttr(RelocAttrBits::UNSIGNED)) {
656     // References from thread-local variable sections are treated as offsets
657     // relative to the start of the referent section, and therefore have no
658     // need of rebase opcodes.
659     if (!(isThreadLocalVariables(isec->getFlags()) && isa<Defined>(sym)))
660       addNonLazyBindingEntries(sym, isec, r.offset, r.addend);
661   }
662 }
663 
scanRelocations()664 void Writer::scanRelocations() {
665   TimeTraceScope timeScope("Scan relocations");
666 
667   // This can't use a for-each loop: It calls treatUndefinedSymbol(), which can
668   // add to inputSections, which invalidates inputSections's iterators.
669   for (size_t i = 0; i < inputSections.size(); ++i) {
670     ConcatInputSection *isec = inputSections[i];
671 
672     if (isec->shouldOmitFromOutput())
673       continue;
674 
675     for (auto it = isec->relocs.begin(); it != isec->relocs.end(); ++it) {
676       lld::macho::Reloc &r = *it;
677 
678       // Canonicalize the referent so that later accesses in Writer won't
679       // have to worry about it.
680       if (auto *referentIsec = r.referent.dyn_cast<InputSection *>())
681         r.referent = referentIsec->canonical();
682 
683       if (target->hasAttr(r.type, RelocAttrBits::SUBTRAHEND)) {
684         // Skip over the following UNSIGNED relocation -- it's just there as the
685         // minuend, and doesn't have the usual UNSIGNED semantics. We don't want
686         // to emit rebase opcodes for it.
687         ++it;
688         // Canonicalize the referent so that later accesses in Writer won't
689         // have to worry about it.
690         if (auto *referentIsec = it->referent.dyn_cast<InputSection *>())
691           it->referent = referentIsec->canonical();
692         continue;
693       }
694       if (auto *sym = r.referent.dyn_cast<Symbol *>()) {
695         if (auto *undefined = dyn_cast<Undefined>(sym))
696           treatUndefinedSymbol(*undefined, isec, r.offset);
697         // treatUndefinedSymbol() can replace sym with a DylibSymbol; re-check.
698         if (!isa<Undefined>(sym) && validateSymbolRelocation(sym, isec, r))
699           prepareSymbolRelocation(sym, isec, r);
700       } else {
701         if (!r.pcrel) {
702           if (config->emitChainedFixups)
703             in.chainedFixups->addRebase(isec, r.offset);
704           else
705             in.rebase->addEntry(isec, r.offset);
706         }
707       }
708     }
709   }
710 
711   in.unwindInfo->prepare();
712 }
713 
addNonWeakDefinition(const Defined * defined)714 static void addNonWeakDefinition(const Defined *defined) {
715   if (config->emitChainedFixups)
716     in.chainedFixups->setHasNonWeakDefinition();
717   else
718     in.weakBinding->addNonWeakDefinition(defined);
719 }
720 
scanSymbols()721 void Writer::scanSymbols() {
722   TimeTraceScope timeScope("Scan symbols");
723   for (Symbol *sym : symtab->getSymbols()) {
724     if (auto *defined = dyn_cast<Defined>(sym)) {
725       if (!defined->isLive())
726         continue;
727       defined->canonicalize();
728       if (defined->overridesWeakDef)
729         addNonWeakDefinition(defined);
730       if (!defined->isAbsolute() && isCodeSection(defined->isec))
731         in.unwindInfo->addSymbol(defined);
732     } else if (const auto *dysym = dyn_cast<DylibSymbol>(sym)) {
733       // This branch intentionally doesn't check isLive().
734       if (dysym->isDynamicLookup())
735         continue;
736       dysym->getFile()->refState =
737           std::max(dysym->getFile()->refState, dysym->getRefState());
738     } else if (isa<Undefined>(sym)) {
739       if (sym->getName().starts_with(ObjCStubsSection::symbolPrefix))
740         in.objcStubs->addEntry(sym);
741     }
742   }
743 
744   for (const InputFile *file : inputFiles) {
745     if (auto *objFile = dyn_cast<ObjFile>(file))
746       for (Symbol *sym : objFile->symbols) {
747         if (auto *defined = dyn_cast_or_null<Defined>(sym)) {
748           if (!defined->isLive())
749             continue;
750           defined->canonicalize();
751           if (!defined->isExternal() && !defined->isAbsolute() &&
752               isCodeSection(defined->isec))
753             in.unwindInfo->addSymbol(defined);
754         }
755       }
756   }
757 }
758 
759 // TODO: ld64 enforces the old load commands in a few other cases.
useLCBuildVersion(const PlatformInfo & platformInfo)760 static bool useLCBuildVersion(const PlatformInfo &platformInfo) {
761   static const std::array<std::pair<PlatformType, VersionTuple>, 7> minVersion =
762       {{{PLATFORM_MACOS, VersionTuple(10, 14)},
763         {PLATFORM_IOS, VersionTuple(12, 0)},
764         {PLATFORM_IOSSIMULATOR, VersionTuple(13, 0)},
765         {PLATFORM_TVOS, VersionTuple(12, 0)},
766         {PLATFORM_TVOSSIMULATOR, VersionTuple(13, 0)},
767         {PLATFORM_WATCHOS, VersionTuple(5, 0)},
768         {PLATFORM_WATCHOSSIMULATOR, VersionTuple(6, 0)}}};
769   auto it = llvm::find_if(minVersion, [&](const auto &p) {
770     return p.first == platformInfo.target.Platform;
771   });
772   return it == minVersion.end()
773              ? true
774              : platformInfo.target.MinDeployment >= it->second;
775 }
776 
createLoadCommands()777 template <class LP> void Writer::createLoadCommands() {
778   uint8_t segIndex = 0;
779   for (OutputSegment *seg : outputSegments) {
780     in.header->addLoadCommand(make<LCSegment<LP>>(seg->name, seg));
781     seg->index = segIndex++;
782   }
783 
784   if (config->emitChainedFixups) {
785     in.header->addLoadCommand(make<LCChainedFixups>(in.chainedFixups));
786     in.header->addLoadCommand(make<LCExportsTrie>(in.exports));
787   } else {
788     in.header->addLoadCommand(make<LCDyldInfo>(
789         in.rebase, in.binding, in.weakBinding, in.lazyBinding, in.exports));
790   }
791   in.header->addLoadCommand(make<LCSymtab>(symtabSection, stringTableSection));
792   in.header->addLoadCommand(
793       make<LCDysymtab>(symtabSection, indirectSymtabSection));
794   if (!config->umbrella.empty())
795     in.header->addLoadCommand(make<LCSubFramework>(config->umbrella));
796   if (config->emitEncryptionInfo)
797     in.header->addLoadCommand(make<LCEncryptionInfo<LP>>());
798   for (StringRef path : config->runtimePaths)
799     in.header->addLoadCommand(make<LCRPath>(path));
800 
801   switch (config->outputType) {
802   case MH_EXECUTE:
803     in.header->addLoadCommand(make<LCLoadDylinker>());
804     break;
805   case MH_DYLIB:
806     in.header->addLoadCommand(make<LCDylib>(LC_ID_DYLIB, config->installName,
807                                             config->dylibCompatibilityVersion,
808                                             config->dylibCurrentVersion));
809     break;
810   case MH_BUNDLE:
811     break;
812   default:
813     llvm_unreachable("unhandled output file type");
814   }
815 
816   if (config->generateUuid) {
817     uuidCommand = make<LCUuid>();
818     in.header->addLoadCommand(uuidCommand);
819   }
820 
821   if (useLCBuildVersion(config->platformInfo))
822     in.header->addLoadCommand(make<LCBuildVersion>(config->platformInfo));
823   else
824     in.header->addLoadCommand(make<LCMinVersion>(config->platformInfo));
825 
826   if (config->secondaryPlatformInfo) {
827     in.header->addLoadCommand(
828         make<LCBuildVersion>(*config->secondaryPlatformInfo));
829   }
830 
831   // This is down here to match ld64's load command order.
832   if (config->outputType == MH_EXECUTE)
833     in.header->addLoadCommand(make<LCMain>());
834 
835   // See ld64's OutputFile::buildDylibOrdinalMapping for the corresponding
836   // library ordinal computation code in ld64.
837   int64_t dylibOrdinal = 1;
838   DenseMap<StringRef, int64_t> ordinalForInstallName;
839 
840   std::vector<DylibFile *> dylibFiles;
841   for (InputFile *file : inputFiles) {
842     if (auto *dylibFile = dyn_cast<DylibFile>(file))
843       dylibFiles.push_back(dylibFile);
844   }
845   for (size_t i = 0; i < dylibFiles.size(); ++i)
846     dylibFiles.insert(dylibFiles.end(), dylibFiles[i]->extraDylibs.begin(),
847                       dylibFiles[i]->extraDylibs.end());
848 
849   for (DylibFile *dylibFile : dylibFiles) {
850     if (dylibFile->isBundleLoader) {
851       dylibFile->ordinal = BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE;
852       // Shortcut since bundle-loader does not re-export the symbols.
853 
854       dylibFile->reexport = false;
855       continue;
856     }
857 
858     // Don't emit load commands for a dylib that is not referenced if:
859     // - it was added implicitly (via a reexport, an LC_LOAD_DYLINKER --
860     //   if it's on the linker command line, it's explicit)
861     // - or it's marked MH_DEAD_STRIPPABLE_DYLIB
862     // - or the flag -dead_strip_dylibs is used
863     // FIXME: `isReferenced()` is currently computed before dead code
864     // stripping, so references from dead code keep a dylib alive. This
865     // matches ld64, but it's something we should do better.
866     if (!dylibFile->isReferenced() && !dylibFile->forceNeeded &&
867         (!dylibFile->isExplicitlyLinked() || dylibFile->deadStrippable ||
868          config->deadStripDylibs))
869       continue;
870 
871     // Several DylibFiles can have the same installName. Only emit a single
872     // load command for that installName and give all these DylibFiles the
873     // same ordinal.
874     // This can happen in several cases:
875     // - a new framework could change its installName to an older
876     //   framework name via an $ld$ symbol depending on platform_version
877     // - symlinks (for example, libpthread.tbd is a symlink to libSystem.tbd;
878     //   Foo.framework/Foo.tbd is usually a symlink to
879     //   Foo.framework/Versions/Current/Foo.tbd, where
880     //   Foo.framework/Versions/Current is usually a symlink to
881     //   Foo.framework/Versions/A)
882     // - a framework can be linked both explicitly on the linker
883     //   command line and implicitly as a reexport from a different
884     //   framework. The re-export will usually point to the tbd file
885     //   in Foo.framework/Versions/A/Foo.tbd, while the explicit link will
886     //   usually find Foo.framework/Foo.tbd. These are usually symlinks,
887     //   but in a --reproduce archive they will be identical but distinct
888     //   files.
889     // In the first case, *semantically distinct* DylibFiles will have the
890     // same installName.
891     int64_t &ordinal = ordinalForInstallName[dylibFile->installName];
892     if (ordinal) {
893       dylibFile->ordinal = ordinal;
894       continue;
895     }
896 
897     ordinal = dylibFile->ordinal = dylibOrdinal++;
898     LoadCommandType lcType =
899         dylibFile->forceWeakImport || dylibFile->refState == RefState::Weak
900             ? LC_LOAD_WEAK_DYLIB
901             : LC_LOAD_DYLIB;
902     in.header->addLoadCommand(make<LCDylib>(lcType, dylibFile->installName,
903                                             dylibFile->compatibilityVersion,
904                                             dylibFile->currentVersion));
905 
906     if (dylibFile->reexport)
907       in.header->addLoadCommand(
908           make<LCDylib>(LC_REEXPORT_DYLIB, dylibFile->installName));
909   }
910 
911   for (const auto &dyldEnv : config->dyldEnvs)
912     in.header->addLoadCommand(make<LCDyldEnv>(dyldEnv));
913 
914   if (functionStartsSection)
915     in.header->addLoadCommand(make<LCFunctionStarts>(functionStartsSection));
916   if (dataInCodeSection)
917     in.header->addLoadCommand(make<LCDataInCode>(dataInCodeSection));
918   if (codeSignatureSection)
919     in.header->addLoadCommand(make<LCCodeSignature>(codeSignatureSection));
920 
921   const uint32_t MACOS_MAXPATHLEN = 1024;
922   config->headerPad = std::max(
923       config->headerPad, (config->headerPadMaxInstallNames
924                               ? LCDylib::getInstanceCount() * MACOS_MAXPATHLEN
925                               : 0));
926 }
927 
928 // Sorting only can happen once all outputs have been collected. Here we sort
929 // segments, output sections within each segment, and input sections within each
930 // output segment.
sortSegmentsAndSections()931 static void sortSegmentsAndSections() {
932   TimeTraceScope timeScope("Sort segments and sections");
933   sortOutputSegments();
934 
935   DenseMap<const InputSection *, size_t> isecPriorities =
936       priorityBuilder.buildInputSectionPriorities();
937 
938   uint32_t sectionIndex = 0;
939   for (OutputSegment *seg : outputSegments) {
940     seg->sortOutputSections();
941     // References from thread-local variable sections are treated as offsets
942     // relative to the start of the thread-local data memory area, which
943     // is initialized via copying all the TLV data sections (which are all
944     // contiguous). If later data sections require a greater alignment than
945     // earlier ones, the offsets of data within those sections won't be
946     // guaranteed to aligned unless we normalize alignments. We therefore use
947     // the largest alignment for all TLV data sections.
948     uint32_t tlvAlign = 0;
949     for (const OutputSection *osec : seg->getSections())
950       if (isThreadLocalData(osec->flags) && osec->align > tlvAlign)
951         tlvAlign = osec->align;
952 
953     for (OutputSection *osec : seg->getSections()) {
954       // Now that the output sections are sorted, assign the final
955       // output section indices.
956       if (!osec->isHidden())
957         osec->index = ++sectionIndex;
958       if (isThreadLocalData(osec->flags)) {
959         if (!firstTLVDataSection)
960           firstTLVDataSection = osec;
961         osec->align = tlvAlign;
962       }
963 
964       if (!isecPriorities.empty()) {
965         if (auto *merged = dyn_cast<ConcatOutputSection>(osec)) {
966           llvm::stable_sort(
967               merged->inputs, [&](InputSection *a, InputSection *b) {
968                 return isecPriorities.lookup(a) > isecPriorities.lookup(b);
969               });
970         }
971       }
972     }
973   }
974 }
975 
createOutputSections()976 template <class LP> void Writer::createOutputSections() {
977   TimeTraceScope timeScope("Create output sections");
978   // First, create hidden sections
979   stringTableSection = make<StringTableSection>();
980   symtabSection = makeSymtabSection<LP>(*stringTableSection);
981   indirectSymtabSection = make<IndirectSymtabSection>();
982   if (config->adhocCodesign)
983     codeSignatureSection = make<CodeSignatureSection>();
984   if (config->emitDataInCodeInfo)
985     dataInCodeSection = make<DataInCodeSection>();
986   if (config->emitFunctionStarts)
987     functionStartsSection = make<FunctionStartsSection>();
988 
989   switch (config->outputType) {
990   case MH_EXECUTE:
991     make<PageZeroSection>();
992     break;
993   case MH_DYLIB:
994   case MH_BUNDLE:
995     break;
996   default:
997     llvm_unreachable("unhandled output file type");
998   }
999 
1000   // Then add input sections to output sections.
1001   for (ConcatInputSection *isec : inputSections) {
1002     if (isec->shouldOmitFromOutput())
1003       continue;
1004     ConcatOutputSection *osec = cast<ConcatOutputSection>(isec->parent);
1005     osec->addInput(isec);
1006     osec->inputOrder =
1007         std::min(osec->inputOrder, static_cast<int>(isec->outSecOff));
1008   }
1009 
1010   // Once all the inputs are added, we can finalize the output section
1011   // properties and create the corresponding output segments.
1012   for (const auto &it : concatOutputSections) {
1013     StringRef segname = it.first.first;
1014     ConcatOutputSection *osec = it.second;
1015     assert(segname != segment_names::ld);
1016     if (osec->isNeeded()) {
1017       // See comment in ObjFile::splitEhFrames()
1018       if (osec->name == section_names::ehFrame &&
1019           segname == segment_names::text)
1020         osec->align = target->wordSize;
1021 
1022       // MC keeps the default 1-byte alignment for __thread_vars, even though it
1023       // contains pointers that are fixed up by dyld, which requires proper
1024       // alignment.
1025       if (isThreadLocalVariables(osec->flags))
1026         osec->align = std::max<uint32_t>(osec->align, target->wordSize);
1027 
1028       getOrCreateOutputSegment(segname)->addOutputSection(osec);
1029     }
1030   }
1031 
1032   for (SyntheticSection *ssec : syntheticSections) {
1033     auto it = concatOutputSections.find({ssec->segname, ssec->name});
1034     // We add all LinkEdit sections here because we don't know if they are
1035     // needed until their finalizeContents() methods get called later. While
1036     // this means that we add some redundant sections to __LINKEDIT, there is
1037     // is no redundancy in the output, as we do not emit section headers for
1038     // any LinkEdit sections.
1039     if (ssec->isNeeded() || ssec->segname == segment_names::linkEdit) {
1040       if (it == concatOutputSections.end()) {
1041         getOrCreateOutputSegment(ssec->segname)->addOutputSection(ssec);
1042       } else {
1043         fatal("section from " +
1044               toString(it->second->firstSection()->getFile()) +
1045               " conflicts with synthetic section " + ssec->segname + "," +
1046               ssec->name);
1047       }
1048     }
1049   }
1050 
1051   // dyld requires __LINKEDIT segment to always exist (even if empty).
1052   linkEditSegment = getOrCreateOutputSegment(segment_names::linkEdit);
1053 }
1054 
finalizeAddresses()1055 void Writer::finalizeAddresses() {
1056   TimeTraceScope timeScope("Finalize addresses");
1057   uint64_t pageSize = target->getPageSize();
1058 
1059   // We could parallelize this loop, but local benchmarking indicates it is
1060   // faster to do it all in the main thread.
1061   for (OutputSegment *seg : outputSegments) {
1062     if (seg == linkEditSegment)
1063       continue;
1064     for (OutputSection *osec : seg->getSections()) {
1065       if (!osec->isNeeded())
1066         continue;
1067       // Other kinds of OutputSections have already been finalized.
1068       if (auto *concatOsec = dyn_cast<ConcatOutputSection>(osec))
1069         concatOsec->finalizeContents();
1070     }
1071   }
1072 
1073   // Ensure that segments (and the sections they contain) are allocated
1074   // addresses in ascending order, which dyld requires.
1075   //
1076   // Note that at this point, __LINKEDIT sections are empty, but we need to
1077   // determine addresses of other segments/sections before generating its
1078   // contents.
1079   for (OutputSegment *seg : outputSegments) {
1080     if (seg == linkEditSegment)
1081       continue;
1082     seg->addr = addr;
1083     assignAddresses(seg);
1084     // codesign / libstuff checks for segment ordering by verifying that
1085     // `fileOff + fileSize == next segment fileOff`. So we call
1086     // alignToPowerOf2() before (instead of after) computing fileSize to ensure
1087     // that the segments are contiguous. We handle addr / vmSize similarly for
1088     // the same reason.
1089     fileOff = alignToPowerOf2(fileOff, pageSize);
1090     addr = alignToPowerOf2(addr, pageSize);
1091     seg->vmSize = addr - seg->addr;
1092     seg->fileSize = fileOff - seg->fileOff;
1093     seg->assignAddressesToStartEndSymbols();
1094   }
1095 }
1096 
finalizeLinkEditSegment()1097 void Writer::finalizeLinkEditSegment() {
1098   TimeTraceScope timeScope("Finalize __LINKEDIT segment");
1099   // Fill __LINKEDIT contents.
1100   std::array<LinkEditSection *, 10> linkEditSections{
1101       in.rebase,         in.binding,
1102       in.weakBinding,    in.lazyBinding,
1103       in.exports,        in.chainedFixups,
1104       symtabSection,     indirectSymtabSection,
1105       dataInCodeSection, functionStartsSection,
1106   };
1107   SmallVector<std::shared_future<void>> threadFutures;
1108   threadFutures.reserve(linkEditSections.size());
1109   for (LinkEditSection *osec : linkEditSections)
1110     if (osec)
1111       threadFutures.emplace_back(threadPool.async(
1112           [](LinkEditSection *osec) { osec->finalizeContents(); }, osec));
1113   for (std::shared_future<void> &future : threadFutures)
1114     future.wait();
1115 
1116   // Now that __LINKEDIT is filled out, do a proper calculation of its
1117   // addresses and offsets.
1118   linkEditSegment->addr = addr;
1119   assignAddresses(linkEditSegment);
1120   // No need to page-align fileOff / addr here since this is the last segment.
1121   linkEditSegment->vmSize = addr - linkEditSegment->addr;
1122   linkEditSegment->fileSize = fileOff - linkEditSegment->fileOff;
1123 }
1124 
assignAddresses(OutputSegment * seg)1125 void Writer::assignAddresses(OutputSegment *seg) {
1126   seg->fileOff = fileOff;
1127 
1128   for (OutputSection *osec : seg->getSections()) {
1129     if (!osec->isNeeded())
1130       continue;
1131     addr = alignToPowerOf2(addr, osec->align);
1132     fileOff = alignToPowerOf2(fileOff, osec->align);
1133     osec->addr = addr;
1134     osec->fileOff = isZeroFill(osec->flags) ? 0 : fileOff;
1135     osec->finalize();
1136     osec->assignAddressesToStartEndSymbols();
1137 
1138     addr += osec->getSize();
1139     fileOff += osec->getFileSize();
1140   }
1141 }
1142 
openFile()1143 void Writer::openFile() {
1144   Expected<std::unique_ptr<FileOutputBuffer>> bufferOrErr =
1145       FileOutputBuffer::create(config->outputFile, fileOff,
1146                                FileOutputBuffer::F_executable);
1147 
1148   if (!bufferOrErr)
1149     fatal("failed to open " + config->outputFile + ": " +
1150           llvm::toString(bufferOrErr.takeError()));
1151   buffer = std::move(*bufferOrErr);
1152   in.bufferStart = buffer->getBufferStart();
1153 }
1154 
writeSections()1155 void Writer::writeSections() {
1156   uint8_t *buf = buffer->getBufferStart();
1157   std::vector<const OutputSection *> osecs;
1158   for (const OutputSegment *seg : outputSegments)
1159     append_range(osecs, seg->getSections());
1160 
1161   parallelForEach(osecs.begin(), osecs.end(), [&](const OutputSection *osec) {
1162     osec->writeTo(buf + osec->fileOff);
1163   });
1164 }
1165 
applyOptimizationHints()1166 void Writer::applyOptimizationHints() {
1167   if (config->arch() != AK_arm64 || config->ignoreOptimizationHints)
1168     return;
1169 
1170   uint8_t *buf = buffer->getBufferStart();
1171   TimeTraceScope timeScope("Apply linker optimization hints");
1172   parallelForEach(inputFiles, [buf](const InputFile *file) {
1173     if (const auto *objFile = dyn_cast<ObjFile>(file))
1174       target->applyOptimizationHints(buf, *objFile);
1175   });
1176 }
1177 
1178 // In order to utilize multiple cores, we first split the buffer into chunks,
1179 // compute a hash for each chunk, and then compute a hash value of the hash
1180 // values.
writeUuid()1181 void Writer::writeUuid() {
1182   TimeTraceScope timeScope("Computing UUID");
1183 
1184   ArrayRef<uint8_t> data{buffer->getBufferStart(), buffer->getBufferEnd()};
1185   std::vector<ArrayRef<uint8_t>> chunks = split(data, 1024 * 1024);
1186   // Leave one slot for filename
1187   std::vector<uint64_t> hashes(chunks.size() + 1);
1188   SmallVector<std::shared_future<void>> threadFutures;
1189   threadFutures.reserve(chunks.size());
1190   for (size_t i = 0; i < chunks.size(); ++i)
1191     threadFutures.emplace_back(threadPool.async(
1192         [&](size_t j) { hashes[j] = xxh3_64bits(chunks[j]); }, i));
1193   for (std::shared_future<void> &future : threadFutures)
1194     future.wait();
1195   // Append the output filename so that identical binaries with different names
1196   // don't get the same UUID.
1197   hashes[chunks.size()] = xxh3_64bits(sys::path::filename(config->finalOutput));
1198   uint64_t digest = xxh3_64bits({reinterpret_cast<uint8_t *>(hashes.data()),
1199                                  hashes.size() * sizeof(uint64_t)});
1200   uuidCommand->writeUuid(digest);
1201 }
1202 
1203 // This is step 5 of the algorithm described in the class comment of
1204 // ChainedFixupsSection.
buildFixupChains()1205 void Writer::buildFixupChains() {
1206   if (!config->emitChainedFixups)
1207     return;
1208 
1209   const std::vector<Location> &loc = in.chainedFixups->getLocations();
1210   if (loc.empty())
1211     return;
1212 
1213   TimeTraceScope timeScope("Build fixup chains");
1214 
1215   const uint64_t pageSize = target->getPageSize();
1216   constexpr uint32_t stride = 4; // for DYLD_CHAINED_PTR_64
1217 
1218   for (size_t i = 0, count = loc.size(); i < count;) {
1219     const OutputSegment *oseg = loc[i].isec->parent->parent;
1220     uint8_t *buf = buffer->getBufferStart() + oseg->fileOff;
1221     uint64_t pageIdx = loc[i].offset / pageSize;
1222     ++i;
1223 
1224     while (i < count && loc[i].isec->parent->parent == oseg &&
1225            (loc[i].offset / pageSize) == pageIdx) {
1226       uint64_t offset = loc[i].offset - loc[i - 1].offset;
1227 
1228       auto fail = [&](Twine message) {
1229         error(loc[i].isec->getSegName() + "," + loc[i].isec->getName() +
1230               ", offset " +
1231               Twine(loc[i].offset - loc[i].isec->parent->getSegmentOffset()) +
1232               ": " + message);
1233       };
1234 
1235       if (offset < target->wordSize)
1236         return fail("fixups overlap");
1237       if (offset % stride != 0)
1238         return fail(
1239             "fixups are unaligned (offset " + Twine(offset) +
1240             " is not a multiple of the stride). Re-link with -no_fixup_chains");
1241 
1242       // The "next" field is in the same location for bind and rebase entries.
1243       reinterpret_cast<dyld_chained_ptr_64_bind *>(buf + loc[i - 1].offset)
1244           ->next = offset / stride;
1245       ++i;
1246     }
1247   }
1248 }
1249 
writeCodeSignature()1250 void Writer::writeCodeSignature() {
1251   if (codeSignatureSection) {
1252     TimeTraceScope timeScope("Write code signature");
1253     codeSignatureSection->writeHashes(buffer->getBufferStart());
1254   }
1255 }
1256 
writeOutputFile()1257 void Writer::writeOutputFile() {
1258   TimeTraceScope timeScope("Write output file");
1259   openFile();
1260   reportPendingUndefinedSymbols();
1261   if (errorCount())
1262     return;
1263   writeSections();
1264   applyOptimizationHints();
1265   buildFixupChains();
1266   if (config->generateUuid)
1267     writeUuid();
1268   writeCodeSignature();
1269 
1270   if (auto e = buffer->commit())
1271     fatal("failed to write output '" + buffer->getPath() +
1272           "': " + toString(std::move(e)));
1273 }
1274 
run()1275 template <class LP> void Writer::run() {
1276   treatSpecialUndefineds();
1277   if (config->entry && needsBinding(config->entry))
1278     in.stubs->addEntry(config->entry);
1279 
1280   // Canonicalization of all pointers to InputSections should be handled by
1281   // these two scan* methods. I.e. from this point onward, for all live
1282   // InputSections, we should have `isec->canonical() == isec`.
1283   scanSymbols();
1284   if (in.objcStubs->isNeeded())
1285     in.objcStubs->setUp();
1286   scanRelocations();
1287   if (in.initOffsets->isNeeded())
1288     in.initOffsets->setUp();
1289 
1290   // Do not proceed if there were undefined or duplicate symbols.
1291   reportPendingUndefinedSymbols();
1292   reportPendingDuplicateSymbols();
1293   if (errorCount())
1294     return;
1295 
1296   if (in.stubHelper && in.stubHelper->isNeeded())
1297     in.stubHelper->setUp();
1298 
1299   if (in.objCImageInfo->isNeeded())
1300     in.objCImageInfo->finalizeContents();
1301 
1302   // At this point, we should know exactly which output sections are needed,
1303   // courtesy of scanSymbols() and scanRelocations().
1304   createOutputSections<LP>();
1305 
1306   // After this point, we create no new segments; HOWEVER, we might
1307   // yet create branch-range extension thunks for architectures whose
1308   // hardware call instructions have limited range, e.g., ARM(64).
1309   // The thunks are created as InputSections interspersed among
1310   // the ordinary __TEXT,_text InputSections.
1311   sortSegmentsAndSections();
1312   createLoadCommands<LP>();
1313   finalizeAddresses();
1314   threadPool.async([&] {
1315     if (LLVM_ENABLE_THREADS && config->timeTraceEnabled)
1316       timeTraceProfilerInitialize(config->timeTraceGranularity, "writeMapFile");
1317     writeMapFile();
1318     if (LLVM_ENABLE_THREADS && config->timeTraceEnabled)
1319       timeTraceProfilerFinishThread();
1320   });
1321   finalizeLinkEditSegment();
1322   writeOutputFile();
1323 }
1324 
writeResult()1325 template <class LP> void macho::writeResult() { Writer().run<LP>(); }
1326 
resetWriter()1327 void macho::resetWriter() { LCDylib::resetInstanceCount(); }
1328 
createSyntheticSections()1329 void macho::createSyntheticSections() {
1330   in.header = make<MachHeaderSection>();
1331   if (config->dedupStrings)
1332     in.cStringSection =
1333         make<DeduplicatedCStringSection>(section_names::cString);
1334   else
1335     in.cStringSection = make<CStringSection>(section_names::cString);
1336   in.objcMethnameSection =
1337       make<DeduplicatedCStringSection>(section_names::objcMethname);
1338   in.wordLiteralSection = make<WordLiteralSection>();
1339   if (config->emitChainedFixups) {
1340     in.chainedFixups = make<ChainedFixupsSection>();
1341   } else {
1342     in.rebase = make<RebaseSection>();
1343     in.binding = make<BindingSection>();
1344     in.weakBinding = make<WeakBindingSection>();
1345     in.lazyBinding = make<LazyBindingSection>();
1346     in.lazyPointers = make<LazyPointerSection>();
1347     in.stubHelper = make<StubHelperSection>();
1348   }
1349   in.exports = make<ExportSection>();
1350   in.got = make<GotSection>();
1351   in.tlvPointers = make<TlvPointerSection>();
1352   in.stubs = make<StubsSection>();
1353   in.objcStubs = make<ObjCStubsSection>();
1354   in.unwindInfo = makeUnwindInfoSection();
1355   in.objCImageInfo = make<ObjCImageInfoSection>();
1356   in.initOffsets = make<InitOffsetsSection>();
1357 
1358   // This section contains space for just a single word, and will be used by
1359   // dyld to cache an address to the image loader it uses.
1360   uint8_t *arr = bAlloc().Allocate<uint8_t>(target->wordSize);
1361   memset(arr, 0, target->wordSize);
1362   in.imageLoaderCache = makeSyntheticInputSection(
1363       segment_names::data, section_names::data, S_REGULAR,
1364       ArrayRef<uint8_t>{arr, target->wordSize},
1365       /*align=*/target->wordSize);
1366   // References from dyld are not visible to us, so ensure this section is
1367   // always treated as live.
1368   in.imageLoaderCache->live = true;
1369 }
1370 
1371 OutputSection *macho::firstTLVDataSection = nullptr;
1372 
1373 template void macho::writeResult<LP64>();
1374 template void macho::writeResult<ILP32>();
1375