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
37using namespace llvm;
38using namespace llvm::MachO;
39using namespace llvm::sys;
40using namespace lld;
41using namespace lld::macho;
42
43namespace {
44class LCUuid;
45
46class Writer {
47public:
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 DefaultThreadPool 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.
86class LCDyldInfo final : public LoadCommand {
87public:
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
96 uint32_t getSize() const override { return sizeof(dyld_info_command); }
97
98 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
131class LCSubFramework final : public LoadCommand {
132public:
133 LCSubFramework(StringRef umbrella) : umbrella(umbrella) {}
134
135 uint32_t getSize() const override {
136 return alignToPowerOf2(Value: sizeof(sub_framework_command) + umbrella.size() + 1,
137 Align: target->wordSize);
138 }
139
140 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(dest: buf, src: umbrella.data(), n: umbrella.size());
149 buf[umbrella.size()] = '\0';
150 }
151
152private:
153 const StringRef umbrella;
154};
155
156class LCFunctionStarts final : public LoadCommand {
157public:
158 explicit LCFunctionStarts(FunctionStartsSection *functionStartsSection)
159 : functionStartsSection(functionStartsSection) {}
160
161 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
162
163 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
171private:
172 FunctionStartsSection *functionStartsSection;
173};
174
175class LCDataInCode final : public LoadCommand {
176public:
177 explicit LCDataInCode(DataInCodeSection *dataInCodeSection)
178 : dataInCodeSection(dataInCodeSection) {}
179
180 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
181
182 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
190private:
191 DataInCodeSection *dataInCodeSection;
192};
193
194class LCDysymtab final : public LoadCommand {
195public:
196 LCDysymtab(SymtabSection *symtabSection,
197 IndirectSymtabSection *indirectSymtabSection)
198 : symtabSection(symtabSection),
199 indirectSymtabSection(indirectSymtabSection) {}
200
201 uint32_t getSize() const override { return sizeof(dysymtab_command); }
202
203 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
222template <class LP> class LCSegment final : public LoadCommand {
223public:
224 LCSegment(StringRef name, OutputSegment *seg) : name(name), seg(seg) {}
225
226 uint32_t getSize() const override {
227 return sizeof(typename LP::segment_command) +
228 seg->numNonHiddenSections() * sizeof(typename LP::section);
229 }
230
231 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(Value: 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
271private:
272 StringRef name;
273 OutputSegment *seg;
274};
275
276class LCMain final : public LoadCommand {
277 uint32_t getSize() const override {
278 return sizeof(structs::entry_point_command);
279 }
280
281 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
296class LCSymtab final : public LoadCommand {
297public:
298 LCSymtab(SymtabSection *symtabSection, StringTableSection *stringTableSection)
299 : symtabSection(symtabSection), stringTableSection(stringTableSection) {}
300
301 uint32_t getSize() const override { return sizeof(symtab_command); }
302
303 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
321class LCDylib final : public LoadCommand {
322public:
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
330 uint32_t getSize() const override {
331 return alignToPowerOf2(Value: sizeof(dylib_command) + path.size() + 1,
332 Align: target->wordSize);
333 }
334
335 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(dest: buf, src: path.data(), n: path.size());
347 buf[path.size()] = '\0';
348 }
349
350 static uint32_t getInstanceCount() { return instanceCount; }
351 static void resetInstanceCount() { instanceCount = 0; }
352
353private:
354 LoadCommandType type;
355 StringRef path;
356 uint32_t compatibilityVersion;
357 uint32_t currentVersion;
358 static uint32_t instanceCount;
359};
360
361uint32_t LCDylib::instanceCount = 0;
362
363class LCLoadDylinker final : public LoadCommand {
364public:
365 uint32_t getSize() const override {
366 return alignToPowerOf2(Value: sizeof(dylinker_command) + path.size() + 1,
367 Align: target->wordSize);
368 }
369
370 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(dest: buf, src: path.data(), n: path.size());
379 buf[path.size()] = '\0';
380 }
381
382private:
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
388class LCRPath final : public LoadCommand {
389public:
390 explicit LCRPath(StringRef path) : path(path) {}
391
392 uint32_t getSize() const override {
393 return alignToPowerOf2(Value: sizeof(rpath_command) + path.size() + 1,
394 Align: target->wordSize);
395 }
396
397 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(dest: buf, src: path.data(), n: path.size());
406 buf[path.size()] = '\0';
407 }
408
409private:
410 StringRef path;
411};
412
413class LCDyldEnv final : public LoadCommand {
414public:
415 explicit LCDyldEnv(StringRef name) : name(name) {}
416
417 uint32_t getSize() const override {
418 return alignToPowerOf2(Value: sizeof(dyld_env_command) + name.size() + 1,
419 Align: target->wordSize);
420 }
421
422 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(dest: buf, src: name.data(), n: name.size());
431 buf[name.size()] = '\0';
432 }
433
434private:
435 StringRef name;
436};
437
438class LCMinVersion final : public LoadCommand {
439public:
440 explicit LCMinVersion(const PlatformInfo &platformInfo)
441 : platformInfo(platformInfo) {}
442
443 uint32_t getSize() const override { return sizeof(version_min_command); }
444
445 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(version: platformInfo.target.MinDeployment);
469 c->sdk = encodeVersion(version: platformInfo.sdk);
470 }
471
472private:
473 const PlatformInfo &platformInfo;
474};
475
476class LCBuildVersion final : public LoadCommand {
477public:
478 explicit LCBuildVersion(const PlatformInfo &platformInfo)
479 : platformInfo(platformInfo) {}
480
481 const int ntools = 1;
482
483 uint32_t getSize() const override {
484 return sizeof(build_version_command) + ntools * sizeof(build_tool_version);
485 }
486
487 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(version: platformInfo.target.MinDeployment);
494 c->sdk = encodeVersion(version: 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(version: VersionTuple(
500 LLVM_VERSION_MAJOR, LLVM_VERSION_MINOR, LLVM_VERSION_PATCH));
501 }
502
503private:
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.
513class LCUuid final : public LoadCommand {
514public:
515 uint32_t getSize() const override { return sizeof(uuid_command); }
516
517 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
524 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(dest: uuidBuf, src: "LLD\xa1UU1D", n: 8);
528 memcpy(dest: uuidBuf + 8, src: &digest, n: 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(a&: uuidBuf[3], b&: 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
546template <class LP> class LCEncryptionInfo final : public LoadCommand {
547public:
548 uint32_t getSize() const override {
549 return sizeof(typename LP::encryption_info_command);
550 }
551
552 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
567class LCCodeSignature final : public LoadCommand {
568public:
569 LCCodeSignature(CodeSignatureSection *section) : section(section) {}
570
571 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
572
573 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
584class LCExportsTrie final : public LoadCommand {
585public:
586 LCExportsTrie(ExportSection *section) : section(section) {}
587
588 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
589
590 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
601class LCChainedFixups final : public LoadCommand {
602public:
603 LCChainedFixups(ChainedFixupsSection *section) : section(section) {}
604
605 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
606
607 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
620void Writer::treatSpecialUndefineds() {
621 if (config->entry)
622 if (auto *undefined = dyn_cast<Undefined>(Val: config->entry))
623 treatUndefinedSymbol(*undefined, source: "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>(Val: sym))
629 treatUndefinedSymbol(*undefined, source: "-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(name: cachedName))
636 if (const auto *undefined = dyn_cast<Undefined>(Val: sym))
637 treatUndefinedSymbol(*undefined, source: "-exported_symbol(s_list)");
638 }
639}
640
641static void prepareSymbolRelocation(Symbol *sym, const InputSection *isec,
642 const lld::macho::Reloc &r) {
643 assert(sym->isLive());
644 const RelocAttrs &relocAttrs = target->getRelocAttrs(type: r.type);
645
646 if (relocAttrs.hasAttr(b: RelocAttrBits::BRANCH)) {
647 if (needsBinding(sym))
648 in.stubs->addEntry(sym);
649 } else if (relocAttrs.hasAttr(b: RelocAttrBits::GOT)) {
650 if (relocAttrs.hasAttr(b: RelocAttrBits::POINTER) || needsBinding(sym))
651 in.got->addEntry(sym);
652 } else if (relocAttrs.hasAttr(b: RelocAttrBits::TLV)) {
653 if (needsBinding(sym))
654 in.tlvPointers->addEntry(sym);
655 } else if (relocAttrs.hasAttr(b: 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(flags: isec->getFlags()) && isa<Defined>(Val: sym)))
660 addNonLazyBindingEntries(sym, isec, offset: r.offset, addend: r.addend);
661 }
662}
663
664void 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(type: r.type, bit: 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>(Val: sym))
696 treatUndefinedSymbol(*undefined, isec, offset: r.offset);
697 // treatUndefinedSymbol() can replace sym with a DylibSymbol; re-check.
698 if (!isa<Undefined>(Val: 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, offset: r.offset);
704 else
705 in.rebase->addEntry(isec, offset: r.offset);
706 }
707 }
708 }
709 }
710
711 in.unwindInfo->prepare();
712}
713
714static void addNonWeakDefinition(const Defined *defined) {
715 if (config->emitChainedFixups)
716 in.chainedFixups->setHasNonWeakDefinition();
717 else
718 in.weakBinding->addNonWeakDefinition(defined);
719}
720
721void Writer::scanSymbols() {
722 TimeTraceScope timeScope("Scan symbols");
723 ObjCSelRefsHelper::initialize();
724 for (Symbol *sym : symtab->getSymbols()) {
725 if (auto *defined = dyn_cast<Defined>(Val: sym)) {
726 if (!defined->isLive())
727 continue;
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>(Val: sym)) {
733 // This branch intentionally doesn't check isLive().
734 if (dysym->isDynamicLookup())
735 continue;
736 dysym->getFile()->refState =
737 std::max(a: dysym->getFile()->refState, b: dysym->getRefState());
738 } else if (isa<Undefined>(Val: sym)) {
739 if (ObjCStubsSection::isObjCStubSymbol(sym)) {
740 // When -dead_strip is enabled, we don't want to emit any dead stubs.
741 // Although this stub symbol is yet undefined, addSym() was called
742 // during MarkLive.
743 if (config->deadStrip) {
744 if (!sym->isLive())
745 continue;
746 }
747 in.objcStubs->addEntry(sym);
748 }
749 }
750 }
751
752 for (const InputFile *file : inputFiles) {
753 if (auto *objFile = dyn_cast<ObjFile>(Val: file))
754 for (Symbol *sym : objFile->symbols) {
755 if (auto *defined = dyn_cast_or_null<Defined>(Val: sym)) {
756 if (!defined->isLive())
757 continue;
758 if (!defined->isExternal() && !defined->isAbsolute() &&
759 isCodeSection(defined->isec()))
760 in.unwindInfo->addSymbol(defined);
761 }
762 }
763 }
764}
765
766// TODO: ld64 enforces the old load commands in a few other cases.
767static bool useLCBuildVersion(const PlatformInfo &platformInfo) {
768 static const std::array<std::pair<PlatformType, VersionTuple>, 7> minVersion =
769 {._M_elems: {{PLATFORM_MACOS, VersionTuple(10, 14)},
770 {PLATFORM_IOS, VersionTuple(12, 0)},
771 {PLATFORM_IOSSIMULATOR, VersionTuple(13, 0)},
772 {PLATFORM_TVOS, VersionTuple(12, 0)},
773 {PLATFORM_TVOSSIMULATOR, VersionTuple(13, 0)},
774 {PLATFORM_WATCHOS, VersionTuple(5, 0)},
775 {PLATFORM_WATCHOSSIMULATOR, VersionTuple(6, 0)}}};
776 auto it = llvm::find_if(Range: minVersion, P: [&](const auto &p) {
777 return p.first == platformInfo.target.Platform;
778 });
779 return it == minVersion.end()
780 ? true
781 : platformInfo.target.MinDeployment >= it->second;
782}
783
784template <class LP> void Writer::createLoadCommands() {
785 uint8_t segIndex = 0;
786 for (OutputSegment *seg : outputSegments) {
787 in.header->addLoadCommand(make<LCSegment<LP>>(seg->name, seg));
788 seg->index = segIndex++;
789 }
790
791 if (config->emitChainedFixups) {
792 in.header->addLoadCommand(make<LCChainedFixups>(args&: in.chainedFixups));
793 in.header->addLoadCommand(make<LCExportsTrie>(args&: in.exports));
794 } else {
795 in.header->addLoadCommand(make<LCDyldInfo>(
796 args&: in.rebase, args&: in.binding, args&: in.weakBinding, args&: in.lazyBinding, args&: in.exports));
797 }
798 in.header->addLoadCommand(make<LCSymtab>(args&: symtabSection, args&: stringTableSection));
799 in.header->addLoadCommand(
800 make<LCDysymtab>(args&: symtabSection, args&: indirectSymtabSection));
801 if (!config->umbrella.empty())
802 in.header->addLoadCommand(make<LCSubFramework>(args&: config->umbrella));
803 if (config->emitEncryptionInfo)
804 in.header->addLoadCommand(make<LCEncryptionInfo<LP>>());
805 for (StringRef path : config->runtimePaths)
806 in.header->addLoadCommand(make<LCRPath>(args&: path));
807
808 switch (config->outputType) {
809 case MH_EXECUTE:
810 in.header->addLoadCommand(make<LCLoadDylinker>());
811 break;
812 case MH_DYLIB:
813 in.header->addLoadCommand(make<LCDylib>(args: LC_ID_DYLIB, args&: config->installName,
814 args&: config->dylibCompatibilityVersion,
815 args&: config->dylibCurrentVersion));
816 break;
817 case MH_BUNDLE:
818 break;
819 default:
820 llvm_unreachable("unhandled output file type");
821 }
822
823 if (config->generateUuid) {
824 uuidCommand = make<LCUuid>();
825 in.header->addLoadCommand(uuidCommand);
826 }
827
828 if (useLCBuildVersion(platformInfo: config->platformInfo))
829 in.header->addLoadCommand(make<LCBuildVersion>(args&: config->platformInfo));
830 else
831 in.header->addLoadCommand(make<LCMinVersion>(args&: config->platformInfo));
832
833 if (config->secondaryPlatformInfo) {
834 in.header->addLoadCommand(
835 make<LCBuildVersion>(args&: *config->secondaryPlatformInfo));
836 }
837
838 // This is down here to match ld64's load command order.
839 if (config->outputType == MH_EXECUTE)
840 in.header->addLoadCommand(make<LCMain>());
841
842 // See ld64's OutputFile::buildDylibOrdinalMapping for the corresponding
843 // library ordinal computation code in ld64.
844 int64_t dylibOrdinal = 1;
845 DenseMap<StringRef, int64_t> ordinalForInstallName;
846
847 std::vector<DylibFile *> dylibFiles;
848 for (InputFile *file : inputFiles) {
849 if (auto *dylibFile = dyn_cast<DylibFile>(Val: file))
850 dylibFiles.push_back(x: dylibFile);
851 }
852 for (size_t i = 0; i < dylibFiles.size(); ++i)
853 dylibFiles.insert(position: dylibFiles.end(), first: dylibFiles[i]->extraDylibs.begin(),
854 last: dylibFiles[i]->extraDylibs.end());
855
856 for (DylibFile *dylibFile : dylibFiles) {
857 if (dylibFile->isBundleLoader) {
858 dylibFile->ordinal = BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE;
859 // Shortcut since bundle-loader does not re-export the symbols.
860
861 dylibFile->reexport = false;
862 continue;
863 }
864
865 // Don't emit load commands for a dylib that is not referenced if:
866 // - it was added implicitly (via a reexport, an LC_LOAD_DYLINKER --
867 // if it's on the linker command line, it's explicit)
868 // - or it's marked MH_DEAD_STRIPPABLE_DYLIB
869 // - or the flag -dead_strip_dylibs is used
870 // FIXME: `isReferenced()` is currently computed before dead code
871 // stripping, so references from dead code keep a dylib alive. This
872 // matches ld64, but it's something we should do better.
873 if (!dylibFile->isReferenced() && !dylibFile->forceNeeded &&
874 (!dylibFile->isExplicitlyLinked() || dylibFile->deadStrippable ||
875 config->deadStripDylibs))
876 continue;
877
878 // Several DylibFiles can have the same installName. Only emit a single
879 // load command for that installName and give all these DylibFiles the
880 // same ordinal.
881 // This can happen in several cases:
882 // - a new framework could change its installName to an older
883 // framework name via an $ld$ symbol depending on platform_version
884 // - symlinks (for example, libpthread.tbd is a symlink to libSystem.tbd;
885 // Foo.framework/Foo.tbd is usually a symlink to
886 // Foo.framework/Versions/Current/Foo.tbd, where
887 // Foo.framework/Versions/Current is usually a symlink to
888 // Foo.framework/Versions/A)
889 // - a framework can be linked both explicitly on the linker
890 // command line and implicitly as a reexport from a different
891 // framework. The re-export will usually point to the tbd file
892 // in Foo.framework/Versions/A/Foo.tbd, while the explicit link will
893 // usually find Foo.framework/Foo.tbd. These are usually symlinks,
894 // but in a --reproduce archive they will be identical but distinct
895 // files.
896 // In the first case, *semantically distinct* DylibFiles will have the
897 // same installName.
898 int64_t &ordinal = ordinalForInstallName[dylibFile->installName];
899 if (ordinal) {
900 dylibFile->ordinal = ordinal;
901 continue;
902 }
903
904 ordinal = dylibFile->ordinal = dylibOrdinal++;
905 LoadCommandType lcType =
906 dylibFile->forceWeakImport || dylibFile->refState == RefState::Weak
907 ? LC_LOAD_WEAK_DYLIB
908 : LC_LOAD_DYLIB;
909 in.header->addLoadCommand(make<LCDylib>(args&: lcType, args&: dylibFile->installName,
910 args&: dylibFile->compatibilityVersion,
911 args&: dylibFile->currentVersion));
912
913 if (dylibFile->reexport)
914 in.header->addLoadCommand(
915 make<LCDylib>(args: LC_REEXPORT_DYLIB, args&: dylibFile->installName));
916 }
917
918 for (const auto &dyldEnv : config->dyldEnvs)
919 in.header->addLoadCommand(make<LCDyldEnv>(args: dyldEnv));
920
921 if (functionStartsSection)
922 in.header->addLoadCommand(make<LCFunctionStarts>(args&: functionStartsSection));
923 if (dataInCodeSection)
924 in.header->addLoadCommand(make<LCDataInCode>(args&: dataInCodeSection));
925 if (codeSignatureSection)
926 in.header->addLoadCommand(make<LCCodeSignature>(args&: codeSignatureSection));
927
928 const uint32_t MACOS_MAXPATHLEN = 1024;
929 config->headerPad = std::max(
930 a: config->headerPad, b: (config->headerPadMaxInstallNames
931 ? LCDylib::getInstanceCount() * MACOS_MAXPATHLEN
932 : 0));
933}
934
935// Sorting only can happen once all outputs have been collected. Here we sort
936// segments, output sections within each segment, and input sections within each
937// output segment.
938static void sortSegmentsAndSections() {
939 TimeTraceScope timeScope("Sort segments and sections");
940 sortOutputSegments();
941
942 DenseMap<const InputSection *, size_t> isecPriorities =
943 priorityBuilder.buildInputSectionPriorities();
944
945 uint32_t sectionIndex = 0;
946 for (OutputSegment *seg : outputSegments) {
947 seg->sortOutputSections();
948 // References from thread-local variable sections are treated as offsets
949 // relative to the start of the thread-local data memory area, which
950 // is initialized via copying all the TLV data sections (which are all
951 // contiguous). If later data sections require a greater alignment than
952 // earlier ones, the offsets of data within those sections won't be
953 // guaranteed to aligned unless we normalize alignments. We therefore use
954 // the largest alignment for all TLV data sections.
955 uint32_t tlvAlign = 0;
956 for (const OutputSection *osec : seg->getSections())
957 if (isThreadLocalData(flags: osec->flags) && osec->align > tlvAlign)
958 tlvAlign = osec->align;
959
960 for (OutputSection *osec : seg->getSections()) {
961 // Now that the output sections are sorted, assign the final
962 // output section indices.
963 if (!osec->isHidden())
964 osec->index = ++sectionIndex;
965 if (isThreadLocalData(flags: osec->flags)) {
966 if (!firstTLVDataSection)
967 firstTLVDataSection = osec;
968 osec->align = tlvAlign;
969 }
970
971 if (!isecPriorities.empty()) {
972 if (auto *merged = dyn_cast<ConcatOutputSection>(Val: osec)) {
973 llvm::stable_sort(
974 Range&: merged->inputs, C: [&](InputSection *a, InputSection *b) {
975 return isecPriorities.lookup(Val: a) > isecPriorities.lookup(Val: b);
976 });
977 }
978 }
979 }
980 }
981}
982
983template <class LP> void Writer::createOutputSections() {
984 TimeTraceScope timeScope("Create output sections");
985 // First, create hidden sections
986 stringTableSection = make<StringTableSection>();
987 symtabSection = makeSymtabSection<LP>(*stringTableSection);
988 indirectSymtabSection = make<IndirectSymtabSection>();
989 if (config->adhocCodesign)
990 codeSignatureSection = make<CodeSignatureSection>();
991 if (config->emitDataInCodeInfo)
992 dataInCodeSection = make<DataInCodeSection>();
993 if (config->emitFunctionStarts)
994 functionStartsSection = make<FunctionStartsSection>();
995
996 switch (config->outputType) {
997 case MH_EXECUTE:
998 make<PageZeroSection>();
999 break;
1000 case MH_DYLIB:
1001 case MH_BUNDLE:
1002 break;
1003 default:
1004 llvm_unreachable("unhandled output file type");
1005 }
1006
1007 // Then add input sections to output sections.
1008 for (ConcatInputSection *isec : inputSections) {
1009 if (isec->shouldOmitFromOutput())
1010 continue;
1011 ConcatOutputSection *osec = cast<ConcatOutputSection>(Val: isec->parent);
1012 osec->addInput(input: isec);
1013 osec->inputOrder =
1014 std::min(a: osec->inputOrder, b: static_cast<int>(isec->outSecOff));
1015 }
1016
1017 // Once all the inputs are added, we can finalize the output section
1018 // properties and create the corresponding output segments.
1019 for (const auto &it : concatOutputSections) {
1020 StringRef segname = it.first.first;
1021 ConcatOutputSection *osec = it.second;
1022 assert(segname != segment_names::ld);
1023 if (osec->isNeeded()) {
1024 // See comment in ObjFile::splitEhFrames()
1025 if (osec->name == section_names::ehFrame &&
1026 segname == segment_names::text)
1027 osec->align = target->wordSize;
1028
1029 // MC keeps the default 1-byte alignment for __thread_vars, even though it
1030 // contains pointers that are fixed up by dyld, which requires proper
1031 // alignment.
1032 if (isThreadLocalVariables(flags: osec->flags))
1033 osec->align = std::max<uint32_t>(a: osec->align, b: target->wordSize);
1034
1035 getOrCreateOutputSegment(name: segname)->addOutputSection(os: osec);
1036 }
1037 }
1038
1039 for (SyntheticSection *ssec : syntheticSections) {
1040 auto it = concatOutputSections.find(Key: {ssec->segname, ssec->name});
1041 // We add all LinkEdit sections here because we don't know if they are
1042 // needed until their finalizeContents() methods get called later. While
1043 // this means that we add some redundant sections to __LINKEDIT, there is
1044 // is no redundancy in the output, as we do not emit section headers for
1045 // any LinkEdit sections.
1046 if (ssec->isNeeded() || ssec->segname == segment_names::linkEdit) {
1047 if (it == concatOutputSections.end()) {
1048 getOrCreateOutputSegment(name: ssec->segname)->addOutputSection(os: ssec);
1049 } else {
1050 fatal(msg: "section from " +
1051 toString(file: it->second->firstSection()->getFile()) +
1052 " conflicts with synthetic section " + ssec->segname + "," +
1053 ssec->name);
1054 }
1055 }
1056 }
1057
1058 // dyld requires __LINKEDIT segment to always exist (even if empty).
1059 linkEditSegment = getOrCreateOutputSegment(name: segment_names::linkEdit);
1060}
1061
1062void Writer::finalizeAddresses() {
1063 TimeTraceScope timeScope("Finalize addresses");
1064 uint64_t pageSize = target->getPageSize();
1065
1066 // We could parallelize this loop, but local benchmarking indicates it is
1067 // faster to do it all in the main thread.
1068 for (OutputSegment *seg : outputSegments) {
1069 if (seg == linkEditSegment)
1070 continue;
1071 for (OutputSection *osec : seg->getSections()) {
1072 if (!osec->isNeeded())
1073 continue;
1074 // Other kinds of OutputSections have already been finalized.
1075 if (auto *concatOsec = dyn_cast<ConcatOutputSection>(Val: osec))
1076 concatOsec->finalizeContents();
1077 }
1078 }
1079
1080 // Ensure that segments (and the sections they contain) are allocated
1081 // addresses in ascending order, which dyld requires.
1082 //
1083 // Note that at this point, __LINKEDIT sections are empty, but we need to
1084 // determine addresses of other segments/sections before generating its
1085 // contents.
1086 for (OutputSegment *seg : outputSegments) {
1087 if (seg == linkEditSegment)
1088 continue;
1089 seg->addr = addr;
1090 assignAddresses(seg);
1091 // codesign / libstuff checks for segment ordering by verifying that
1092 // `fileOff + fileSize == next segment fileOff`. So we call
1093 // alignToPowerOf2() before (instead of after) computing fileSize to ensure
1094 // that the segments are contiguous. We handle addr / vmSize similarly for
1095 // the same reason.
1096 fileOff = alignToPowerOf2(Value: fileOff, Align: pageSize);
1097 addr = alignToPowerOf2(Value: addr, Align: pageSize);
1098 seg->vmSize = addr - seg->addr;
1099 seg->fileSize = fileOff - seg->fileOff;
1100 seg->assignAddressesToStartEndSymbols();
1101 }
1102}
1103
1104void Writer::finalizeLinkEditSegment() {
1105 TimeTraceScope timeScope("Finalize __LINKEDIT segment");
1106 // Fill __LINKEDIT contents.
1107 std::array<LinkEditSection *, 10> linkEditSections{
1108 in.rebase, in.binding,
1109 in.weakBinding, in.lazyBinding,
1110 in.exports, in.chainedFixups,
1111 symtabSection, indirectSymtabSection,
1112 dataInCodeSection, functionStartsSection,
1113 };
1114 SmallVector<std::shared_future<void>> threadFutures;
1115 threadFutures.reserve(N: linkEditSections.size());
1116 for (LinkEditSection *osec : linkEditSections)
1117 if (osec)
1118 threadFutures.emplace_back(Args: threadPool.async(
1119 F: [](LinkEditSection *osec) { osec->finalizeContents(); }, ArgList&: osec));
1120 for (std::shared_future<void> &future : threadFutures)
1121 future.wait();
1122
1123 // Now that __LINKEDIT is filled out, do a proper calculation of its
1124 // addresses and offsets.
1125 linkEditSegment->addr = addr;
1126 assignAddresses(linkEditSegment);
1127 // No need to page-align fileOff / addr here since this is the last segment.
1128 linkEditSegment->vmSize = addr - linkEditSegment->addr;
1129 linkEditSegment->fileSize = fileOff - linkEditSegment->fileOff;
1130}
1131
1132void Writer::assignAddresses(OutputSegment *seg) {
1133 seg->fileOff = fileOff;
1134
1135 for (OutputSection *osec : seg->getSections()) {
1136 if (!osec->isNeeded())
1137 continue;
1138 addr = alignToPowerOf2(Value: addr, Align: osec->align);
1139 fileOff = alignToPowerOf2(Value: fileOff, Align: osec->align);
1140 osec->addr = addr;
1141 osec->fileOff = isZeroFill(flags: osec->flags) ? 0 : fileOff;
1142 osec->finalize();
1143 osec->assignAddressesToStartEndSymbols();
1144
1145 addr += osec->getSize();
1146 fileOff += osec->getFileSize();
1147 }
1148}
1149
1150void Writer::openFile() {
1151 Expected<std::unique_ptr<FileOutputBuffer>> bufferOrErr =
1152 FileOutputBuffer::create(FilePath: config->outputFile, Size: fileOff,
1153 Flags: FileOutputBuffer::F_executable);
1154
1155 if (!bufferOrErr)
1156 fatal(msg: "failed to open " + config->outputFile + ": " +
1157 llvm::toString(E: bufferOrErr.takeError()));
1158 buffer = std::move(*bufferOrErr);
1159 in.bufferStart = buffer->getBufferStart();
1160}
1161
1162void Writer::writeSections() {
1163 uint8_t *buf = buffer->getBufferStart();
1164 std::vector<const OutputSection *> osecs;
1165 for (const OutputSegment *seg : outputSegments)
1166 append_range(C&: osecs, R: seg->getSections());
1167
1168 parallelForEach(Begin: osecs.begin(), End: osecs.end(), Fn: [&](const OutputSection *osec) {
1169 osec->writeTo(buf: buf + osec->fileOff);
1170 });
1171}
1172
1173void Writer::applyOptimizationHints() {
1174 if (config->arch() != AK_arm64 || config->ignoreOptimizationHints)
1175 return;
1176
1177 uint8_t *buf = buffer->getBufferStart();
1178 TimeTraceScope timeScope("Apply linker optimization hints");
1179 parallelForEach(R&: inputFiles, Fn: [buf](const InputFile *file) {
1180 if (const auto *objFile = dyn_cast<ObjFile>(Val: file))
1181 target->applyOptimizationHints(buf, *objFile);
1182 });
1183}
1184
1185// In order to utilize multiple cores, we first split the buffer into chunks,
1186// compute a hash for each chunk, and then compute a hash value of the hash
1187// values.
1188void Writer::writeUuid() {
1189 TimeTraceScope timeScope("Computing UUID");
1190
1191 ArrayRef<uint8_t> data{buffer->getBufferStart(), buffer->getBufferEnd()};
1192 std::vector<ArrayRef<uint8_t>> chunks = split(arr: data, chunkSize: 1024 * 1024);
1193 // Leave one slot for filename
1194 std::vector<uint64_t> hashes(chunks.size() + 1);
1195 SmallVector<std::shared_future<void>> threadFutures;
1196 threadFutures.reserve(N: chunks.size());
1197 for (size_t i = 0; i < chunks.size(); ++i)
1198 threadFutures.emplace_back(Args: threadPool.async(
1199 F: [&](size_t j) { hashes[j] = xxh3_64bits(data: chunks[j]); }, ArgList&: i));
1200 for (std::shared_future<void> &future : threadFutures)
1201 future.wait();
1202 // Append the output filename so that identical binaries with different names
1203 // don't get the same UUID.
1204 hashes[chunks.size()] = xxh3_64bits(data: sys::path::filename(path: config->finalOutput));
1205 uint64_t digest = xxh3_64bits(data: {reinterpret_cast<uint8_t *>(hashes.data()),
1206 hashes.size() * sizeof(uint64_t)});
1207 uuidCommand->writeUuid(digest);
1208}
1209
1210// This is step 5 of the algorithm described in the class comment of
1211// ChainedFixupsSection.
1212void Writer::buildFixupChains() {
1213 if (!config->emitChainedFixups)
1214 return;
1215
1216 const std::vector<Location> &loc = in.chainedFixups->getLocations();
1217 if (loc.empty())
1218 return;
1219
1220 TimeTraceScope timeScope("Build fixup chains");
1221
1222 const uint64_t pageSize = target->getPageSize();
1223 constexpr uint32_t stride = 4; // for DYLD_CHAINED_PTR_64
1224
1225 for (size_t i = 0, count = loc.size(); i < count;) {
1226 const OutputSegment *oseg = loc[i].isec->parent->parent;
1227 uint8_t *buf = buffer->getBufferStart() + oseg->fileOff;
1228 uint64_t pageIdx = loc[i].offset / pageSize;
1229 ++i;
1230
1231 while (i < count && loc[i].isec->parent->parent == oseg &&
1232 (loc[i].offset / pageSize) == pageIdx) {
1233 uint64_t offset = loc[i].offset - loc[i - 1].offset;
1234
1235 auto fail = [&](Twine message) {
1236 error(msg: loc[i].isec->getSegName() + "," + loc[i].isec->getName() +
1237 ", offset " +
1238 Twine(loc[i].offset - loc[i].isec->parent->getSegmentOffset()) +
1239 ": " + message);
1240 };
1241
1242 if (offset < target->wordSize)
1243 return fail("fixups overlap");
1244 if (offset % stride != 0)
1245 return fail(
1246 "fixups are unaligned (offset " + Twine(offset) +
1247 " is not a multiple of the stride). Re-link with -no_fixup_chains");
1248
1249 // The "next" field is in the same location for bind and rebase entries.
1250 reinterpret_cast<dyld_chained_ptr_64_bind *>(buf + loc[i - 1].offset)
1251 ->next = offset / stride;
1252 ++i;
1253 }
1254 }
1255}
1256
1257void Writer::writeCodeSignature() {
1258 if (codeSignatureSection) {
1259 TimeTraceScope timeScope("Write code signature");
1260 codeSignatureSection->writeHashes(buf: buffer->getBufferStart());
1261 }
1262}
1263
1264void Writer::writeOutputFile() {
1265 TimeTraceScope timeScope("Write output file");
1266 openFile();
1267 reportPendingUndefinedSymbols();
1268 if (errorCount())
1269 return;
1270 writeSections();
1271 applyOptimizationHints();
1272 buildFixupChains();
1273 if (config->generateUuid)
1274 writeUuid();
1275 writeCodeSignature();
1276
1277 if (auto e = buffer->commit())
1278 fatal(msg: "failed to write output '" + buffer->getPath() +
1279 "': " + toString(E: std::move(e)));
1280}
1281
1282template <class LP> void Writer::run() {
1283 treatSpecialUndefineds();
1284 if (config->entry && needsBinding(sym: config->entry))
1285 in.stubs->addEntry(config->entry);
1286
1287 // Canonicalization of all pointers to InputSections should be handled by
1288 // these two scan* methods. I.e. from this point onward, for all live
1289 // InputSections, we should have `isec->canonical() == isec`.
1290 scanSymbols();
1291 if (in.objcStubs->isNeeded())
1292 in.objcStubs->setUp();
1293 if (in.objcMethList->isNeeded())
1294 in.objcMethList->setUp();
1295 scanRelocations();
1296 if (in.initOffsets->isNeeded())
1297 in.initOffsets->setUp();
1298
1299 // Do not proceed if there were undefined or duplicate symbols.
1300 reportPendingUndefinedSymbols();
1301 reportPendingDuplicateSymbols();
1302 if (errorCount())
1303 return;
1304
1305 if (in.stubHelper && in.stubHelper->isNeeded())
1306 in.stubHelper->setUp();
1307
1308 if (in.objCImageInfo->isNeeded())
1309 in.objCImageInfo->finalizeContents();
1310
1311 // At this point, we should know exactly which output sections are needed,
1312 // courtesy of scanSymbols() and scanRelocations().
1313 createOutputSections<LP>();
1314
1315 // After this point, we create no new segments; HOWEVER, we might
1316 // yet create branch-range extension thunks for architectures whose
1317 // hardware call instructions have limited range, e.g., ARM(64).
1318 // The thunks are created as InputSections interspersed among
1319 // the ordinary __TEXT,_text InputSections.
1320 sortSegmentsAndSections();
1321 createLoadCommands<LP>();
1322 finalizeAddresses();
1323 threadPool.async([&] {
1324 if (LLVM_ENABLE_THREADS && config->timeTraceEnabled)
1325 timeTraceProfilerInitialize(TimeTraceGranularity: config->timeTraceGranularity, ProcName: "writeMapFile");
1326 writeMapFile();
1327 if (LLVM_ENABLE_THREADS && config->timeTraceEnabled)
1328 timeTraceProfilerFinishThread();
1329 });
1330 finalizeLinkEditSegment();
1331 writeOutputFile();
1332}
1333
1334template <class LP> void macho::writeResult() { Writer().run<LP>(); }
1335
1336void macho::resetWriter() { LCDylib::resetInstanceCount(); }
1337
1338void macho::createSyntheticSections() {
1339 in.header = make<MachHeaderSection>();
1340 if (config->dedupStrings)
1341 in.cStringSection =
1342 make<DeduplicatedCStringSection>(args: section_names::cString);
1343 else
1344 in.cStringSection = make<CStringSection>(args: section_names::cString);
1345 in.objcMethnameSection =
1346 make<DeduplicatedCStringSection>(args: section_names::objcMethname);
1347 in.wordLiteralSection = make<WordLiteralSection>();
1348 if (config->emitChainedFixups) {
1349 in.chainedFixups = make<ChainedFixupsSection>();
1350 } else {
1351 in.rebase = make<RebaseSection>();
1352 in.binding = make<BindingSection>();
1353 in.weakBinding = make<WeakBindingSection>();
1354 in.lazyBinding = make<LazyBindingSection>();
1355 in.lazyPointers = make<LazyPointerSection>();
1356 in.stubHelper = make<StubHelperSection>();
1357 }
1358 in.exports = make<ExportSection>();
1359 in.got = make<GotSection>();
1360 in.tlvPointers = make<TlvPointerSection>();
1361 in.stubs = make<StubsSection>();
1362 in.objcStubs = make<ObjCStubsSection>();
1363 in.unwindInfo = makeUnwindInfoSection();
1364 in.objCImageInfo = make<ObjCImageInfoSection>();
1365 in.initOffsets = make<InitOffsetsSection>();
1366 in.objcMethList = make<ObjCMethListSection>();
1367
1368 // This section contains space for just a single word, and will be used by
1369 // dyld to cache an address to the image loader it uses.
1370 uint8_t *arr = bAlloc().Allocate<uint8_t>(Num: target->wordSize);
1371 memset(s: arr, c: 0, n: target->wordSize);
1372 in.imageLoaderCache = makeSyntheticInputSection(
1373 segName: segment_names::data, sectName: section_names::data, flags: S_REGULAR,
1374 data: ArrayRef<uint8_t>{arr, target->wordSize},
1375 /*align=*/target->wordSize);
1376 assert(in.imageLoaderCache->live);
1377}
1378
1379OutputSection *macho::firstTLVDataSection = nullptr;
1380
1381template void macho::writeResult<LP64>();
1382template void macho::writeResult<ILP32>();
1383

source code of lld/MachO/Writer.cpp