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/Parallel.h"
29 #include "llvm/Support/Path.h"
30 #include "llvm/Support/TimeProfiler.h"
31 #include "llvm/Support/thread.h"
32 #include "llvm/Support/xxhash.h"
33
34 #include <algorithm>
35
36 using namespace llvm;
37 using namespace llvm::MachO;
38 using namespace llvm::sys;
39 using namespace lld;
40 using namespace lld::macho;
41
42 namespace {
43 class LCUuid;
44
45 class Writer {
46 public:
Writer()47 Writer() : buffer(errorHandler().outputBuffer) {}
48
49 void treatSpecialUndefineds();
50 void scanRelocations();
51 void scanSymbols();
52 template <class LP> void createOutputSections();
53 template <class LP> void createLoadCommands();
54 void finalizeAddresses();
55 void finalizeLinkEditSegment();
56 void assignAddresses(OutputSegment *);
57
58 void openFile();
59 void writeSections();
60 void applyOptimizationHints();
61 void buildFixupChains();
62 void writeUuid();
63 void writeCodeSignature();
64 void writeOutputFile();
65
66 template <class LP> void run();
67
68 std::unique_ptr<FileOutputBuffer> &buffer;
69 uint64_t addr = 0;
70 uint64_t fileOff = 0;
71 MachHeaderSection *header = nullptr;
72 StringTableSection *stringTableSection = nullptr;
73 SymtabSection *symtabSection = nullptr;
74 IndirectSymtabSection *indirectSymtabSection = nullptr;
75 CodeSignatureSection *codeSignatureSection = nullptr;
76 DataInCodeSection *dataInCodeSection = nullptr;
77 FunctionStartsSection *functionStartsSection = nullptr;
78
79 LCUuid *uuidCommand = nullptr;
80 OutputSegment *linkEditSegment = nullptr;
81 };
82
83 // LC_DYLD_INFO_ONLY stores the offsets of symbol import/export information.
84 class LCDyldInfo final : public LoadCommand {
85 public:
LCDyldInfo(RebaseSection * rebaseSection,BindingSection * bindingSection,WeakBindingSection * weakBindingSection,LazyBindingSection * lazyBindingSection,ExportSection * exportSection)86 LCDyldInfo(RebaseSection *rebaseSection, BindingSection *bindingSection,
87 WeakBindingSection *weakBindingSection,
88 LazyBindingSection *lazyBindingSection,
89 ExportSection *exportSection)
90 : rebaseSection(rebaseSection), bindingSection(bindingSection),
91 weakBindingSection(weakBindingSection),
92 lazyBindingSection(lazyBindingSection), exportSection(exportSection) {}
93
getSize() const94 uint32_t getSize() const override { return sizeof(dyld_info_command); }
95
writeTo(uint8_t * buf) const96 void writeTo(uint8_t *buf) const override {
97 auto *c = reinterpret_cast<dyld_info_command *>(buf);
98 c->cmd = LC_DYLD_INFO_ONLY;
99 c->cmdsize = getSize();
100 if (rebaseSection->isNeeded()) {
101 c->rebase_off = rebaseSection->fileOff;
102 c->rebase_size = rebaseSection->getFileSize();
103 }
104 if (bindingSection->isNeeded()) {
105 c->bind_off = bindingSection->fileOff;
106 c->bind_size = bindingSection->getFileSize();
107 }
108 if (weakBindingSection->isNeeded()) {
109 c->weak_bind_off = weakBindingSection->fileOff;
110 c->weak_bind_size = weakBindingSection->getFileSize();
111 }
112 if (lazyBindingSection->isNeeded()) {
113 c->lazy_bind_off = lazyBindingSection->fileOff;
114 c->lazy_bind_size = lazyBindingSection->getFileSize();
115 }
116 if (exportSection->isNeeded()) {
117 c->export_off = exportSection->fileOff;
118 c->export_size = exportSection->getFileSize();
119 }
120 }
121
122 RebaseSection *rebaseSection;
123 BindingSection *bindingSection;
124 WeakBindingSection *weakBindingSection;
125 LazyBindingSection *lazyBindingSection;
126 ExportSection *exportSection;
127 };
128
129 class LCSubFramework final : public LoadCommand {
130 public:
LCSubFramework(StringRef umbrella)131 LCSubFramework(StringRef umbrella) : umbrella(umbrella) {}
132
getSize() const133 uint32_t getSize() const override {
134 return alignToPowerOf2(sizeof(sub_framework_command) + umbrella.size() + 1,
135 target->wordSize);
136 }
137
writeTo(uint8_t * buf) const138 void writeTo(uint8_t *buf) const override {
139 auto *c = reinterpret_cast<sub_framework_command *>(buf);
140 buf += sizeof(sub_framework_command);
141
142 c->cmd = LC_SUB_FRAMEWORK;
143 c->cmdsize = getSize();
144 c->umbrella = sizeof(sub_framework_command);
145
146 memcpy(buf, umbrella.data(), umbrella.size());
147 buf[umbrella.size()] = '\0';
148 }
149
150 private:
151 const StringRef umbrella;
152 };
153
154 class LCFunctionStarts final : public LoadCommand {
155 public:
LCFunctionStarts(FunctionStartsSection * functionStartsSection)156 explicit LCFunctionStarts(FunctionStartsSection *functionStartsSection)
157 : functionStartsSection(functionStartsSection) {}
158
getSize() const159 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
160
writeTo(uint8_t * buf) const161 void writeTo(uint8_t *buf) const override {
162 auto *c = reinterpret_cast<linkedit_data_command *>(buf);
163 c->cmd = LC_FUNCTION_STARTS;
164 c->cmdsize = getSize();
165 c->dataoff = functionStartsSection->fileOff;
166 c->datasize = functionStartsSection->getFileSize();
167 }
168
169 private:
170 FunctionStartsSection *functionStartsSection;
171 };
172
173 class LCDataInCode final : public LoadCommand {
174 public:
LCDataInCode(DataInCodeSection * dataInCodeSection)175 explicit LCDataInCode(DataInCodeSection *dataInCodeSection)
176 : dataInCodeSection(dataInCodeSection) {}
177
getSize() const178 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
179
writeTo(uint8_t * buf) const180 void writeTo(uint8_t *buf) const override {
181 auto *c = reinterpret_cast<linkedit_data_command *>(buf);
182 c->cmd = LC_DATA_IN_CODE;
183 c->cmdsize = getSize();
184 c->dataoff = dataInCodeSection->fileOff;
185 c->datasize = dataInCodeSection->getFileSize();
186 }
187
188 private:
189 DataInCodeSection *dataInCodeSection;
190 };
191
192 class LCDysymtab final : public LoadCommand {
193 public:
LCDysymtab(SymtabSection * symtabSection,IndirectSymtabSection * indirectSymtabSection)194 LCDysymtab(SymtabSection *symtabSection,
195 IndirectSymtabSection *indirectSymtabSection)
196 : symtabSection(symtabSection),
197 indirectSymtabSection(indirectSymtabSection) {}
198
getSize() const199 uint32_t getSize() const override { return sizeof(dysymtab_command); }
200
writeTo(uint8_t * buf) const201 void writeTo(uint8_t *buf) const override {
202 auto *c = reinterpret_cast<dysymtab_command *>(buf);
203 c->cmd = LC_DYSYMTAB;
204 c->cmdsize = getSize();
205
206 c->ilocalsym = 0;
207 c->iextdefsym = c->nlocalsym = symtabSection->getNumLocalSymbols();
208 c->nextdefsym = symtabSection->getNumExternalSymbols();
209 c->iundefsym = c->iextdefsym + c->nextdefsym;
210 c->nundefsym = symtabSection->getNumUndefinedSymbols();
211
212 c->indirectsymoff = indirectSymtabSection->fileOff;
213 c->nindirectsyms = indirectSymtabSection->getNumSymbols();
214 }
215
216 SymtabSection *symtabSection;
217 IndirectSymtabSection *indirectSymtabSection;
218 };
219
220 template <class LP> class LCSegment final : public LoadCommand {
221 public:
LCSegment(StringRef name,OutputSegment * seg)222 LCSegment(StringRef name, OutputSegment *seg) : name(name), seg(seg) {}
223
getSize() const224 uint32_t getSize() const override {
225 return sizeof(typename LP::segment_command) +
226 seg->numNonHiddenSections() * sizeof(typename LP::section);
227 }
228
writeTo(uint8_t * buf) const229 void writeTo(uint8_t *buf) const override {
230 using SegmentCommand = typename LP::segment_command;
231 using SectionHeader = typename LP::section;
232
233 auto *c = reinterpret_cast<SegmentCommand *>(buf);
234 buf += sizeof(SegmentCommand);
235
236 c->cmd = LP::segmentLCType;
237 c->cmdsize = getSize();
238 memcpy(c->segname, name.data(), name.size());
239 c->fileoff = seg->fileOff;
240 c->maxprot = seg->maxProt;
241 c->initprot = seg->initProt;
242
243 c->vmaddr = seg->addr;
244 c->vmsize = seg->vmSize;
245 c->filesize = seg->fileSize;
246 c->nsects = seg->numNonHiddenSections();
247 c->flags = seg->flags;
248
249 for (const OutputSection *osec : seg->getSections()) {
250 if (osec->isHidden())
251 continue;
252
253 auto *sectHdr = reinterpret_cast<SectionHeader *>(buf);
254 buf += sizeof(SectionHeader);
255
256 memcpy(sectHdr->sectname, osec->name.data(), osec->name.size());
257 memcpy(sectHdr->segname, name.data(), name.size());
258
259 sectHdr->addr = osec->addr;
260 sectHdr->offset = osec->fileOff;
261 sectHdr->align = Log2_32(osec->align);
262 sectHdr->flags = osec->flags;
263 sectHdr->size = osec->getSize();
264 sectHdr->reserved1 = osec->reserved1;
265 sectHdr->reserved2 = osec->reserved2;
266 }
267 }
268
269 private:
270 StringRef name;
271 OutputSegment *seg;
272 };
273
274 class LCMain final : public LoadCommand {
getSize() const275 uint32_t getSize() const override {
276 return sizeof(structs::entry_point_command);
277 }
278
writeTo(uint8_t * buf) const279 void writeTo(uint8_t *buf) const override {
280 auto *c = reinterpret_cast<structs::entry_point_command *>(buf);
281 c->cmd = LC_MAIN;
282 c->cmdsize = getSize();
283
284 if (config->entry->isInStubs())
285 c->entryoff =
286 in.stubs->fileOff + config->entry->stubsIndex * target->stubSize;
287 else
288 c->entryoff = config->entry->getVA() - in.header->addr;
289
290 c->stacksize = 0;
291 }
292 };
293
294 class LCSymtab final : public LoadCommand {
295 public:
LCSymtab(SymtabSection * symtabSection,StringTableSection * stringTableSection)296 LCSymtab(SymtabSection *symtabSection, StringTableSection *stringTableSection)
297 : symtabSection(symtabSection), stringTableSection(stringTableSection) {}
298
getSize() const299 uint32_t getSize() const override { return sizeof(symtab_command); }
300
writeTo(uint8_t * buf) const301 void writeTo(uint8_t *buf) const override {
302 auto *c = reinterpret_cast<symtab_command *>(buf);
303 c->cmd = LC_SYMTAB;
304 c->cmdsize = getSize();
305 c->symoff = symtabSection->fileOff;
306 c->nsyms = symtabSection->getNumSymbols();
307 c->stroff = stringTableSection->fileOff;
308 c->strsize = stringTableSection->getFileSize();
309 }
310
311 SymtabSection *symtabSection = nullptr;
312 StringTableSection *stringTableSection = nullptr;
313 };
314
315 // There are several dylib load commands that share the same structure:
316 // * LC_LOAD_DYLIB
317 // * LC_ID_DYLIB
318 // * LC_REEXPORT_DYLIB
319 class LCDylib final : public LoadCommand {
320 public:
LCDylib(LoadCommandType type,StringRef path,uint32_t compatibilityVersion=0,uint32_t currentVersion=0)321 LCDylib(LoadCommandType type, StringRef path,
322 uint32_t compatibilityVersion = 0, uint32_t currentVersion = 0)
323 : type(type), path(path), compatibilityVersion(compatibilityVersion),
324 currentVersion(currentVersion) {
325 instanceCount++;
326 }
327
getSize() const328 uint32_t getSize() const override {
329 return alignToPowerOf2(sizeof(dylib_command) + path.size() + 1,
330 target->wordSize);
331 }
332
writeTo(uint8_t * buf) const333 void writeTo(uint8_t *buf) const override {
334 auto *c = reinterpret_cast<dylib_command *>(buf);
335 buf += sizeof(dylib_command);
336
337 c->cmd = type;
338 c->cmdsize = getSize();
339 c->dylib.name = sizeof(dylib_command);
340 c->dylib.timestamp = 0;
341 c->dylib.compatibility_version = compatibilityVersion;
342 c->dylib.current_version = currentVersion;
343
344 memcpy(buf, path.data(), path.size());
345 buf[path.size()] = '\0';
346 }
347
getInstanceCount()348 static uint32_t getInstanceCount() { return instanceCount; }
resetInstanceCount()349 static void resetInstanceCount() { instanceCount = 0; }
350
351 private:
352 LoadCommandType type;
353 StringRef path;
354 uint32_t compatibilityVersion;
355 uint32_t currentVersion;
356 static uint32_t instanceCount;
357 };
358
359 uint32_t LCDylib::instanceCount = 0;
360
361 class LCLoadDylinker final : public LoadCommand {
362 public:
getSize() const363 uint32_t getSize() const override {
364 return alignToPowerOf2(sizeof(dylinker_command) + path.size() + 1,
365 target->wordSize);
366 }
367
writeTo(uint8_t * buf) const368 void writeTo(uint8_t *buf) const override {
369 auto *c = reinterpret_cast<dylinker_command *>(buf);
370 buf += sizeof(dylinker_command);
371
372 c->cmd = LC_LOAD_DYLINKER;
373 c->cmdsize = getSize();
374 c->name = sizeof(dylinker_command);
375
376 memcpy(buf, path.data(), path.size());
377 buf[path.size()] = '\0';
378 }
379
380 private:
381 // Recent versions of Darwin won't run any binary that has dyld at a
382 // different location.
383 const StringRef path = "/usr/lib/dyld";
384 };
385
386 class LCRPath final : public LoadCommand {
387 public:
LCRPath(StringRef path)388 explicit LCRPath(StringRef path) : path(path) {}
389
getSize() const390 uint32_t getSize() const override {
391 return alignToPowerOf2(sizeof(rpath_command) + path.size() + 1,
392 target->wordSize);
393 }
394
writeTo(uint8_t * buf) const395 void writeTo(uint8_t *buf) const override {
396 auto *c = reinterpret_cast<rpath_command *>(buf);
397 buf += sizeof(rpath_command);
398
399 c->cmd = LC_RPATH;
400 c->cmdsize = getSize();
401 c->path = sizeof(rpath_command);
402
403 memcpy(buf, path.data(), path.size());
404 buf[path.size()] = '\0';
405 }
406
407 private:
408 StringRef path;
409 };
410
411 class LCSubClient final : public LoadCommand {
412 public:
LCSubClient(StringRef client)413 explicit LCSubClient(StringRef client) : client(client) {}
414
getSize() const415 uint32_t getSize() const override {
416 return alignToPowerOf2(sizeof(sub_client_command) + client.size() + 1,
417 target->wordSize);
418 }
419
writeTo(uint8_t * buf) const420 void writeTo(uint8_t *buf) const override {
421 auto *c = reinterpret_cast<sub_client_command *>(buf);
422 buf += sizeof(sub_client_command);
423
424 c->cmd = LC_SUB_CLIENT;
425 c->cmdsize = getSize();
426 c->client = sizeof(sub_client_command);
427
428 memcpy(buf, client.data(), client.size());
429 buf[client.size()] = '\0';
430 }
431
432 private:
433 StringRef client;
434 };
435
436 class LCDyldEnv final : public LoadCommand {
437 public:
LCDyldEnv(StringRef name)438 explicit LCDyldEnv(StringRef name) : name(name) {}
439
getSize() const440 uint32_t getSize() const override {
441 return alignToPowerOf2(sizeof(dyld_env_command) + name.size() + 1,
442 target->wordSize);
443 }
444
writeTo(uint8_t * buf) const445 void writeTo(uint8_t *buf) const override {
446 auto *c = reinterpret_cast<dyld_env_command *>(buf);
447 buf += sizeof(dyld_env_command);
448
449 c->cmd = LC_DYLD_ENVIRONMENT;
450 c->cmdsize = getSize();
451 c->name = sizeof(dyld_env_command);
452
453 memcpy(buf, name.data(), name.size());
454 buf[name.size()] = '\0';
455 }
456
457 private:
458 StringRef name;
459 };
460
461 class LCMinVersion final : public LoadCommand {
462 public:
LCMinVersion(const PlatformInfo & platformInfo)463 explicit LCMinVersion(const PlatformInfo &platformInfo)
464 : platformInfo(platformInfo) {}
465
getSize() const466 uint32_t getSize() const override { return sizeof(version_min_command); }
467
writeTo(uint8_t * buf) const468 void writeTo(uint8_t *buf) const override {
469 auto *c = reinterpret_cast<version_min_command *>(buf);
470 switch (platformInfo.target.Platform) {
471 case PLATFORM_MACOS:
472 c->cmd = LC_VERSION_MIN_MACOSX;
473 break;
474 case PLATFORM_IOS:
475 case PLATFORM_IOSSIMULATOR:
476 c->cmd = LC_VERSION_MIN_IPHONEOS;
477 break;
478 case PLATFORM_TVOS:
479 case PLATFORM_TVOSSIMULATOR:
480 c->cmd = LC_VERSION_MIN_TVOS;
481 break;
482 case PLATFORM_WATCHOS:
483 case PLATFORM_WATCHOSSIMULATOR:
484 c->cmd = LC_VERSION_MIN_WATCHOS;
485 break;
486 default:
487 llvm_unreachable("invalid platform");
488 break;
489 }
490 c->cmdsize = getSize();
491 c->version = encodeVersion(platformInfo.target.MinDeployment);
492 c->sdk = encodeVersion(platformInfo.sdk);
493 }
494
495 private:
496 const PlatformInfo &platformInfo;
497 };
498
499 class LCBuildVersion final : public LoadCommand {
500 public:
LCBuildVersion(const PlatformInfo & platformInfo)501 explicit LCBuildVersion(const PlatformInfo &platformInfo)
502 : platformInfo(platformInfo) {}
503
504 const int ntools = 1;
505
getSize() const506 uint32_t getSize() const override {
507 return sizeof(build_version_command) + ntools * sizeof(build_tool_version);
508 }
509
writeTo(uint8_t * buf) const510 void writeTo(uint8_t *buf) const override {
511 auto *c = reinterpret_cast<build_version_command *>(buf);
512 c->cmd = LC_BUILD_VERSION;
513 c->cmdsize = getSize();
514
515 c->platform = static_cast<uint32_t>(platformInfo.target.Platform);
516 c->minos = encodeVersion(platformInfo.target.MinDeployment);
517 c->sdk = encodeVersion(platformInfo.sdk);
518
519 c->ntools = ntools;
520 auto *t = reinterpret_cast<build_tool_version *>(&c[1]);
521 t->tool = TOOL_LLD;
522 t->version = encodeVersion(VersionTuple(
523 LLVM_VERSION_MAJOR, LLVM_VERSION_MINOR, LLVM_VERSION_PATCH));
524 }
525
526 private:
527 const PlatformInfo &platformInfo;
528 };
529
530 // Stores a unique identifier for the output file based on an MD5 hash of its
531 // contents. In order to hash the contents, we must first write them, but
532 // LC_UUID itself must be part of the written contents in order for all the
533 // offsets to be calculated correctly. We resolve this circular paradox by
534 // first writing an LC_UUID with an all-zero UUID, then updating the UUID with
535 // its real value later.
536 class LCUuid final : public LoadCommand {
537 public:
getSize() const538 uint32_t getSize() const override { return sizeof(uuid_command); }
539
writeTo(uint8_t * buf) const540 void writeTo(uint8_t *buf) const override {
541 auto *c = reinterpret_cast<uuid_command *>(buf);
542 c->cmd = LC_UUID;
543 c->cmdsize = getSize();
544 uuidBuf = c->uuid;
545 }
546
writeUuid(uint64_t digest) const547 void writeUuid(uint64_t digest) const {
548 // xxhash only gives us 8 bytes, so put some fixed data in the other half.
549 static_assert(sizeof(uuid_command::uuid) == 16, "unexpected uuid size");
550 memcpy(uuidBuf, "LLD\xa1UU1D", 8);
551 memcpy(uuidBuf + 8, &digest, 8);
552
553 // RFC 4122 conformance. We need to fix 4 bits in byte 6 and 2 bits in
554 // byte 8. Byte 6 is already fine due to the fixed data we put in. We don't
555 // want to lose bits of the digest in byte 8, so swap that with a byte of
556 // fixed data that happens to have the right bits set.
557 std::swap(uuidBuf[3], uuidBuf[8]);
558
559 // Claim that this is an MD5-based hash. It isn't, but this signals that
560 // this is not a time-based and not a random hash. MD5 seems like the least
561 // bad lie we can put here.
562 assert((uuidBuf[6] & 0xf0) == 0x30 && "See RFC 4122 Sections 4.2.2, 4.1.3");
563 assert((uuidBuf[8] & 0xc0) == 0x80 && "See RFC 4122 Section 4.2.2");
564 }
565
566 mutable uint8_t *uuidBuf;
567 };
568
569 template <class LP> class LCEncryptionInfo final : public LoadCommand {
570 public:
getSize() const571 uint32_t getSize() const override {
572 return sizeof(typename LP::encryption_info_command);
573 }
574
writeTo(uint8_t * buf) const575 void writeTo(uint8_t *buf) const override {
576 using EncryptionInfo = typename LP::encryption_info_command;
577 auto *c = reinterpret_cast<EncryptionInfo *>(buf);
578 buf += sizeof(EncryptionInfo);
579 c->cmd = LP::encryptionInfoLCType;
580 c->cmdsize = getSize();
581 c->cryptoff = in.header->getSize();
582 auto it = find_if(outputSegments, [](const OutputSegment *seg) {
583 return seg->name == segment_names::text;
584 });
585 assert(it != outputSegments.end());
586 c->cryptsize = (*it)->fileSize - c->cryptoff;
587 }
588 };
589
590 class LCCodeSignature final : public LoadCommand {
591 public:
LCCodeSignature(CodeSignatureSection * section)592 LCCodeSignature(CodeSignatureSection *section) : section(section) {}
593
getSize() const594 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
595
writeTo(uint8_t * buf) const596 void writeTo(uint8_t *buf) const override {
597 auto *c = reinterpret_cast<linkedit_data_command *>(buf);
598 c->cmd = LC_CODE_SIGNATURE;
599 c->cmdsize = getSize();
600 c->dataoff = static_cast<uint32_t>(section->fileOff);
601 c->datasize = section->getSize();
602 }
603
604 CodeSignatureSection *section;
605 };
606
607 class LCExportsTrie final : public LoadCommand {
608 public:
LCExportsTrie(ExportSection * section)609 LCExportsTrie(ExportSection *section) : section(section) {}
610
getSize() const611 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
612
writeTo(uint8_t * buf) const613 void writeTo(uint8_t *buf) const override {
614 auto *c = reinterpret_cast<linkedit_data_command *>(buf);
615 c->cmd = LC_DYLD_EXPORTS_TRIE;
616 c->cmdsize = getSize();
617 c->dataoff = section->fileOff;
618 c->datasize = section->getSize();
619 }
620
621 ExportSection *section;
622 };
623
624 class LCChainedFixups final : public LoadCommand {
625 public:
LCChainedFixups(ChainedFixupsSection * section)626 LCChainedFixups(ChainedFixupsSection *section) : section(section) {}
627
getSize() const628 uint32_t getSize() const override { return sizeof(linkedit_data_command); }
629
writeTo(uint8_t * buf) const630 void writeTo(uint8_t *buf) const override {
631 auto *c = reinterpret_cast<linkedit_data_command *>(buf);
632 c->cmd = LC_DYLD_CHAINED_FIXUPS;
633 c->cmdsize = getSize();
634 c->dataoff = section->fileOff;
635 c->datasize = section->getSize();
636 }
637
638 ChainedFixupsSection *section;
639 };
640
641 } // namespace
642
treatSpecialUndefineds()643 void Writer::treatSpecialUndefineds() {
644 if (config->entry)
645 if (auto *undefined = dyn_cast<Undefined>(config->entry))
646 treatUndefinedSymbol(*undefined, "the entry point");
647
648 // FIXME: This prints symbols that are undefined both in input files and
649 // via -u flag twice.
650 for (const Symbol *sym : config->explicitUndefineds) {
651 if (const auto *undefined = dyn_cast<Undefined>(sym))
652 treatUndefinedSymbol(*undefined, "-u");
653 }
654 // Literal exported-symbol names must be defined, but glob
655 // patterns need not match.
656 for (const CachedHashStringRef &cachedName :
657 config->exportedSymbols.literals) {
658 if (const Symbol *sym = symtab->find(cachedName))
659 if (const auto *undefined = dyn_cast<Undefined>(sym))
660 treatUndefinedSymbol(*undefined, "-exported_symbol(s_list)");
661 }
662 }
663
prepareSymbolRelocation(Symbol * sym,const InputSection * isec,const lld::macho::Reloc & r)664 static void prepareSymbolRelocation(Symbol *sym, const InputSection *isec,
665 const lld::macho::Reloc &r) {
666 if (!sym->isLive()) {
667 if (Defined *defined = dyn_cast<Defined>(sym)) {
668 if (config->emitInitOffsets &&
669 defined->isec()->getName() == section_names::moduleInitFunc)
670 fatal(isec->getLocation(r.offset) + ": cannot reference " +
671 sym->getName() +
672 " defined in __mod_init_func when -init_offsets is used");
673 }
674 assert(false && "referenced symbol must be live");
675 }
676
677 const RelocAttrs &relocAttrs = target->getRelocAttrs(r.type);
678
679 if (relocAttrs.hasAttr(RelocAttrBits::BRANCH)) {
680 if (needsBinding(sym))
681 in.stubs->addEntry(sym);
682 } else if (relocAttrs.hasAttr(RelocAttrBits::GOT)) {
683 if (relocAttrs.hasAttr(RelocAttrBits::POINTER) || needsBinding(sym))
684 in.got->addEntry(sym);
685 } else if (relocAttrs.hasAttr(RelocAttrBits::TLV)) {
686 if (needsBinding(sym))
687 in.tlvPointers->addEntry(sym);
688 } else if (relocAttrs.hasAttr(RelocAttrBits::UNSIGNED)) {
689 // References from thread-local variable sections are treated as offsets
690 // relative to the start of the referent section, and therefore have no
691 // need of rebase opcodes.
692 if (!(isThreadLocalVariables(isec->getFlags()) && isa<Defined>(sym)))
693 addNonLazyBindingEntries(sym, isec, r.offset, r.addend);
694 }
695 }
696
scanRelocations()697 void Writer::scanRelocations() {
698 TimeTraceScope timeScope("Scan relocations");
699
700 // This can't use a for-each loop: It calls treatUndefinedSymbol(), which can
701 // add to inputSections, which invalidates inputSections's iterators.
702 for (size_t i = 0; i < inputSections.size(); ++i) {
703 ConcatInputSection *isec = inputSections[i];
704
705 if (isec->shouldOmitFromOutput())
706 continue;
707
708 for (auto it = isec->relocs.begin(); it != isec->relocs.end(); ++it) {
709 lld::macho::Reloc &r = *it;
710
711 // Canonicalize the referent so that later accesses in Writer won't
712 // have to worry about it.
713 if (auto *referentIsec = dyn_cast_if_present<InputSection *>(r.referent))
714 r.referent = referentIsec->canonical();
715
716 if (target->hasAttr(r.type, RelocAttrBits::SUBTRAHEND)) {
717 // Skip over the following UNSIGNED relocation -- it's just there as the
718 // minuend, and doesn't have the usual UNSIGNED semantics. We don't want
719 // to emit rebase opcodes for it.
720 ++it;
721 // Canonicalize the referent so that later accesses in Writer won't
722 // have to worry about it.
723 if (auto *referentIsec = it->referent.dyn_cast<InputSection *>())
724 it->referent = referentIsec->canonical();
725 continue;
726 }
727 if (auto *sym = dyn_cast_if_present<Symbol *>(r.referent)) {
728 if (auto *undefined = dyn_cast<Undefined>(sym))
729 treatUndefinedSymbol(*undefined, isec, r.offset);
730 // treatUndefinedSymbol() can replace sym with a DylibSymbol; re-check.
731 if (!isa<Undefined>(sym) && validateSymbolRelocation(sym, isec, r))
732 prepareSymbolRelocation(sym, isec, r);
733 } else {
734 if (!r.pcrel) {
735 if (config->emitChainedFixups)
736 in.chainedFixups->addRebase(isec, r.offset);
737 else
738 in.rebase->addEntry(isec, r.offset);
739 }
740 }
741 }
742 }
743
744 in.unwindInfo->prepare();
745 }
746
addNonWeakDefinition(const Defined * defined)747 static void addNonWeakDefinition(const Defined *defined) {
748 if (config->emitChainedFixups)
749 in.chainedFixups->setHasNonWeakDefinition();
750 else
751 in.weakBinding->addNonWeakDefinition(defined);
752 }
753
scanSymbols()754 void Writer::scanSymbols() {
755 TimeTraceScope timeScope("Scan symbols");
756 ObjCSelRefsHelper::initialize();
757 for (Symbol *sym : symtab->getSymbols()) {
758 if (auto *defined = dyn_cast<Defined>(sym)) {
759 if (!defined->isLive())
760 continue;
761 if (defined->overridesWeakDef)
762 addNonWeakDefinition(defined);
763 if (!defined->isAbsolute() && isCodeSection(defined->isec()))
764 in.unwindInfo->addSymbol(defined);
765 } else if (const auto *dysym = dyn_cast<DylibSymbol>(sym)) {
766 // This branch intentionally doesn't check isLive().
767 if (dysym->isDynamicLookup())
768 continue;
769 dysym->getFile()->refState =
770 std::max(dysym->getFile()->refState, dysym->getRefState());
771 } else if (isa<Undefined>(sym)) {
772 if (ObjCStubsSection::isObjCStubSymbol(sym)) {
773 // When -dead_strip is enabled, we don't want to emit any dead stubs.
774 // Although this stub symbol is yet undefined, addSym() was called
775 // during MarkLive.
776 if (config->deadStrip) {
777 if (!sym->isLive())
778 continue;
779 }
780 in.objcStubs->addEntry(sym);
781 }
782 }
783 }
784
785 for (const InputFile *file : inputFiles) {
786 if (auto *objFile = dyn_cast<ObjFile>(file))
787 for (Symbol *sym : objFile->symbols) {
788 if (auto *defined = dyn_cast_or_null<Defined>(sym)) {
789 if (!defined->isLive())
790 continue;
791 if (!defined->isExternal() && !defined->isAbsolute() &&
792 isCodeSection(defined->isec()))
793 in.unwindInfo->addSymbol(defined);
794 }
795 }
796 }
797 }
798
799 // TODO: ld64 enforces the old load commands in a few other cases.
useLCBuildVersion(const PlatformInfo & platformInfo)800 static bool useLCBuildVersion(const PlatformInfo &platformInfo) {
801 static const std::array<std::pair<PlatformType, VersionTuple>, 7> minVersion =
802 {{{PLATFORM_MACOS, VersionTuple(10, 14)},
803 {PLATFORM_IOS, VersionTuple(12, 0)},
804 {PLATFORM_IOSSIMULATOR, VersionTuple(13, 0)},
805 {PLATFORM_TVOS, VersionTuple(12, 0)},
806 {PLATFORM_TVOSSIMULATOR, VersionTuple(13, 0)},
807 {PLATFORM_WATCHOS, VersionTuple(5, 0)},
808 {PLATFORM_WATCHOSSIMULATOR, VersionTuple(6, 0)}}};
809 auto it = llvm::find_if(minVersion, [&](const auto &p) {
810 return p.first == platformInfo.target.Platform;
811 });
812 return it == minVersion.end()
813 ? true
814 : platformInfo.target.MinDeployment >= it->second;
815 }
816
createLoadCommands()817 template <class LP> void Writer::createLoadCommands() {
818 uint8_t segIndex = 0;
819 for (OutputSegment *seg : outputSegments) {
820 in.header->addLoadCommand(make<LCSegment<LP>>(seg->name, seg));
821 seg->index = segIndex++;
822 }
823
824 if (config->emitChainedFixups) {
825 in.header->addLoadCommand(make<LCChainedFixups>(in.chainedFixups));
826 in.header->addLoadCommand(make<LCExportsTrie>(in.exports));
827 } else {
828 in.header->addLoadCommand(make<LCDyldInfo>(
829 in.rebase, in.binding, in.weakBinding, in.lazyBinding, in.exports));
830 }
831 in.header->addLoadCommand(make<LCSymtab>(symtabSection, stringTableSection));
832 in.header->addLoadCommand(
833 make<LCDysymtab>(symtabSection, indirectSymtabSection));
834 if (!config->umbrella.empty())
835 in.header->addLoadCommand(make<LCSubFramework>(config->umbrella));
836 if (config->emitEncryptionInfo)
837 in.header->addLoadCommand(make<LCEncryptionInfo<LP>>());
838 for (StringRef path : config->runtimePaths)
839 in.header->addLoadCommand(make<LCRPath>(path));
840
841 switch (config->outputType) {
842 case MH_EXECUTE:
843 in.header->addLoadCommand(make<LCLoadDylinker>());
844 break;
845 case MH_DYLIB:
846 in.header->addLoadCommand(make<LCDylib>(LC_ID_DYLIB, config->installName,
847 config->dylibCompatibilityVersion,
848 config->dylibCurrentVersion));
849 for (StringRef client : config->allowableClients)
850 in.header->addLoadCommand(make<LCSubClient>(client));
851 break;
852 case MH_BUNDLE:
853 break;
854 default:
855 llvm_unreachable("unhandled output file type");
856 }
857
858 if (config->generateUuid) {
859 uuidCommand = make<LCUuid>();
860 in.header->addLoadCommand(uuidCommand);
861 }
862
863 if (useLCBuildVersion(config->platformInfo))
864 in.header->addLoadCommand(make<LCBuildVersion>(config->platformInfo));
865 else
866 in.header->addLoadCommand(make<LCMinVersion>(config->platformInfo));
867
868 if (config->secondaryPlatformInfo) {
869 in.header->addLoadCommand(
870 make<LCBuildVersion>(*config->secondaryPlatformInfo));
871 }
872
873 // This is down here to match ld64's load command order.
874 if (config->outputType == MH_EXECUTE)
875 in.header->addLoadCommand(make<LCMain>());
876
877 // See ld64's OutputFile::buildDylibOrdinalMapping for the corresponding
878 // library ordinal computation code in ld64.
879 int64_t dylibOrdinal = 1;
880 DenseMap<StringRef, int64_t> ordinalForInstallName;
881
882 std::vector<DylibFile *> dylibFiles;
883 for (InputFile *file : inputFiles) {
884 if (auto *dylibFile = dyn_cast<DylibFile>(file))
885 dylibFiles.push_back(dylibFile);
886 }
887 for (size_t i = 0; i < dylibFiles.size(); ++i)
888 dylibFiles.insert(dylibFiles.end(), dylibFiles[i]->extraDylibs.begin(),
889 dylibFiles[i]->extraDylibs.end());
890
891 for (DylibFile *dylibFile : dylibFiles) {
892 if (dylibFile->isBundleLoader) {
893 dylibFile->ordinal = BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE;
894 // Shortcut since bundle-loader does not re-export the symbols.
895
896 dylibFile->reexport = false;
897 continue;
898 }
899
900 // Don't emit load commands for a dylib that is not referenced if:
901 // - it was added implicitly (via a reexport, an LC_LOAD_DYLINKER --
902 // if it's on the linker command line, it's explicit)
903 // - or it's marked MH_DEAD_STRIPPABLE_DYLIB
904 // - or the flag -dead_strip_dylibs is used
905 // FIXME: `isReferenced()` is currently computed before dead code
906 // stripping, so references from dead code keep a dylib alive. This
907 // matches ld64, but it's something we should do better.
908 if (!dylibFile->isReferenced() && !dylibFile->forceNeeded &&
909 (!dylibFile->isExplicitlyLinked() || dylibFile->deadStrippable ||
910 config->deadStripDylibs))
911 continue;
912
913 // Several DylibFiles can have the same installName. Only emit a single
914 // load command for that installName and give all these DylibFiles the
915 // same ordinal.
916 // This can happen in several cases:
917 // - a new framework could change its installName to an older
918 // framework name via an $ld$ symbol depending on platform_version
919 // - symlinks (for example, libpthread.tbd is a symlink to libSystem.tbd;
920 // Foo.framework/Foo.tbd is usually a symlink to
921 // Foo.framework/Versions/Current/Foo.tbd, where
922 // Foo.framework/Versions/Current is usually a symlink to
923 // Foo.framework/Versions/A)
924 // - a framework can be linked both explicitly on the linker
925 // command line and implicitly as a reexport from a different
926 // framework. The re-export will usually point to the tbd file
927 // in Foo.framework/Versions/A/Foo.tbd, while the explicit link will
928 // usually find Foo.framework/Foo.tbd. These are usually symlinks,
929 // but in a --reproduce archive they will be identical but distinct
930 // files.
931 // In the first case, *semantically distinct* DylibFiles will have the
932 // same installName.
933 int64_t &ordinal = ordinalForInstallName[dylibFile->installName];
934 if (ordinal) {
935 dylibFile->ordinal = ordinal;
936 continue;
937 }
938
939 ordinal = dylibFile->ordinal = dylibOrdinal++;
940 LoadCommandType lcType = LC_LOAD_DYLIB;
941 if (dylibFile->reexport) {
942 if (dylibFile->forceWeakImport)
943 warn(path::filename(dylibFile->getName()) +
944 " is re-exported so cannot be weak-linked");
945
946 lcType = LC_REEXPORT_DYLIB;
947 } else if (dylibFile->forceWeakImport ||
948 dylibFile->refState == RefState::Weak) {
949 lcType = LC_LOAD_WEAK_DYLIB;
950 }
951 in.header->addLoadCommand(make<LCDylib>(lcType, dylibFile->installName,
952 dylibFile->compatibilityVersion,
953 dylibFile->currentVersion));
954 }
955
956 for (const auto &dyldEnv : config->dyldEnvs)
957 in.header->addLoadCommand(make<LCDyldEnv>(dyldEnv));
958
959 if (functionStartsSection)
960 in.header->addLoadCommand(make<LCFunctionStarts>(functionStartsSection));
961 if (dataInCodeSection)
962 in.header->addLoadCommand(make<LCDataInCode>(dataInCodeSection));
963 if (codeSignatureSection)
964 in.header->addLoadCommand(make<LCCodeSignature>(codeSignatureSection));
965
966 const uint32_t MACOS_MAXPATHLEN = 1024;
967 config->headerPad = std::max(
968 config->headerPad, (config->headerPadMaxInstallNames
969 ? LCDylib::getInstanceCount() * MACOS_MAXPATHLEN
970 : 0));
971 }
972
973 // Sorting only can happen once all outputs have been collected. Here we sort
974 // segments, output sections within each segment, and input sections within each
975 // output segment.
sortSegmentsAndSections()976 static void sortSegmentsAndSections() {
977 TimeTraceScope timeScope("Sort segments and sections");
978 sortOutputSegments();
979
980 DenseMap<const InputSection *, int> isecPriorities =
981 priorityBuilder.buildInputSectionPriorities();
982
983 uint32_t sectionIndex = 0;
984 for (OutputSegment *seg : outputSegments) {
985 seg->sortOutputSections();
986 // References from thread-local variable sections are treated as offsets
987 // relative to the start of the thread-local data memory area, which
988 // is initialized via copying all the TLV data sections (which are all
989 // contiguous). If later data sections require a greater alignment than
990 // earlier ones, the offsets of data within those sections won't be
991 // guaranteed to aligned unless we normalize alignments. We therefore use
992 // the largest alignment for all TLV data sections.
993 uint32_t tlvAlign = 0;
994 for (const OutputSection *osec : seg->getSections())
995 if (isThreadLocalData(osec->flags) && osec->align > tlvAlign)
996 tlvAlign = osec->align;
997
998 for (OutputSection *osec : seg->getSections()) {
999 // Now that the output sections are sorted, assign the final
1000 // output section indices.
1001 if (!osec->isHidden())
1002 osec->index = ++sectionIndex;
1003 if (isThreadLocalData(osec->flags)) {
1004 if (!firstTLVDataSection)
1005 firstTLVDataSection = osec;
1006 osec->align = tlvAlign;
1007 }
1008
1009 if (!isecPriorities.empty()) {
1010 if (auto *merged = dyn_cast<ConcatOutputSection>(osec)) {
1011 llvm::stable_sort(
1012 merged->inputs, [&](InputSection *a, InputSection *b) {
1013 return isecPriorities.lookup(a) < isecPriorities.lookup(b);
1014 });
1015 }
1016 }
1017 }
1018 }
1019 }
1020
createOutputSections()1021 template <class LP> void Writer::createOutputSections() {
1022 TimeTraceScope timeScope("Create output sections");
1023 // First, create hidden sections
1024 stringTableSection = make<StringTableSection>();
1025 symtabSection = makeSymtabSection<LP>(*stringTableSection);
1026 indirectSymtabSection = make<IndirectSymtabSection>();
1027 if (config->adhocCodesign)
1028 codeSignatureSection = make<CodeSignatureSection>();
1029 if (config->emitDataInCodeInfo)
1030 dataInCodeSection = make<DataInCodeSection>();
1031 if (config->emitFunctionStarts)
1032 functionStartsSection = make<FunctionStartsSection>();
1033
1034 switch (config->outputType) {
1035 case MH_EXECUTE:
1036 make<PageZeroSection>();
1037 break;
1038 case MH_DYLIB:
1039 case MH_BUNDLE:
1040 break;
1041 default:
1042 llvm_unreachable("unhandled output file type");
1043 }
1044
1045 // Then add input sections to output sections.
1046 for (ConcatInputSection *isec : inputSections) {
1047 if (isec->shouldOmitFromOutput())
1048 continue;
1049 ConcatOutputSection *osec = cast<ConcatOutputSection>(isec->parent);
1050 osec->addInput(isec);
1051 osec->inputOrder =
1052 std::min(osec->inputOrder, static_cast<int>(isec->outSecOff));
1053 }
1054
1055 // Once all the inputs are added, we can finalize the output section
1056 // properties and create the corresponding output segments.
1057 for (const auto &it : concatOutputSections) {
1058 StringRef segname = it.first.first;
1059 ConcatOutputSection *osec = it.second;
1060 assert(segname != segment_names::ld);
1061 if (osec->isNeeded()) {
1062 // See comment in ObjFile::splitEhFrames()
1063 if (osec->name == section_names::ehFrame &&
1064 segname == segment_names::text)
1065 osec->align = target->wordSize;
1066
1067 // MC keeps the default 1-byte alignment for __thread_vars, even though it
1068 // contains pointers that are fixed up by dyld, which requires proper
1069 // alignment.
1070 if (isThreadLocalVariables(osec->flags))
1071 osec->align = std::max<uint32_t>(osec->align, target->wordSize);
1072
1073 getOrCreateOutputSegment(segname)->addOutputSection(osec);
1074 }
1075 }
1076
1077 for (SyntheticSection *ssec : syntheticSections) {
1078 auto it = concatOutputSections.find({ssec->segname, ssec->name});
1079 // We add all LinkEdit sections here because we don't know if they are
1080 // needed until their finalizeContents() methods get called later. While
1081 // this means that we add some redundant sections to __LINKEDIT, there is
1082 // is no redundancy in the output, as we do not emit section headers for
1083 // any LinkEdit sections.
1084 if (ssec->isNeeded() || ssec->segname == segment_names::linkEdit) {
1085 if (it == concatOutputSections.end()) {
1086 getOrCreateOutputSegment(ssec->segname)->addOutputSection(ssec);
1087 } else {
1088 fatal("section from " +
1089 toString(it->second->firstSection()->getFile()) +
1090 " conflicts with synthetic section " + ssec->segname + "," +
1091 ssec->name);
1092 }
1093 }
1094 }
1095
1096 // dyld requires __LINKEDIT segment to always exist (even if empty).
1097 linkEditSegment = getOrCreateOutputSegment(segment_names::linkEdit);
1098 }
1099
finalizeAddresses()1100 void Writer::finalizeAddresses() {
1101 TimeTraceScope timeScope("Finalize addresses");
1102 uint64_t pageSize = target->getPageSize();
1103
1104 // We could parallelize this loop, but local benchmarking indicates it is
1105 // faster to do it all in the main thread.
1106 for (OutputSegment *seg : outputSegments) {
1107 if (seg == linkEditSegment)
1108 continue;
1109 for (OutputSection *osec : seg->getSections()) {
1110 if (!osec->isNeeded())
1111 continue;
1112 // Other kinds of OutputSections have already been finalized.
1113 if (auto *concatOsec = dyn_cast<ConcatOutputSection>(osec))
1114 concatOsec->finalizeContents();
1115 }
1116 }
1117
1118 // Ensure that segments (and the sections they contain) are allocated
1119 // addresses in ascending order, which dyld requires.
1120 //
1121 // Note that at this point, __LINKEDIT sections are empty, but we need to
1122 // determine addresses of other segments/sections before generating its
1123 // contents.
1124 for (OutputSegment *seg : outputSegments) {
1125 if (seg == linkEditSegment)
1126 continue;
1127 seg->addr = addr;
1128 assignAddresses(seg);
1129 // codesign / libstuff checks for segment ordering by verifying that
1130 // `fileOff + fileSize == next segment fileOff`. So we call
1131 // alignToPowerOf2() before (instead of after) computing fileSize to ensure
1132 // that the segments are contiguous. We handle addr / vmSize similarly for
1133 // the same reason.
1134 fileOff = alignToPowerOf2(fileOff, pageSize);
1135 addr = alignToPowerOf2(addr, pageSize);
1136 seg->vmSize = addr - seg->addr;
1137 seg->fileSize = fileOff - seg->fileOff;
1138 seg->assignAddressesToStartEndSymbols();
1139 }
1140 }
1141
finalizeLinkEditSegment()1142 void Writer::finalizeLinkEditSegment() {
1143 TimeTraceScope timeScope("Finalize __LINKEDIT segment");
1144 // Fill __LINKEDIT contents.
1145 std::array<LinkEditSection *, 10> linkEditSections{
1146 in.rebase, in.binding,
1147 in.weakBinding, in.lazyBinding,
1148 in.exports, in.chainedFixups,
1149 symtabSection, indirectSymtabSection,
1150 dataInCodeSection, functionStartsSection,
1151 };
1152
1153 parallelForEach(linkEditSections.begin(), linkEditSections.end(),
1154 [](LinkEditSection *osec) {
1155 if (osec)
1156 osec->finalizeContents();
1157 });
1158
1159 // Now that __LINKEDIT is filled out, do a proper calculation of its
1160 // addresses and offsets.
1161 linkEditSegment->addr = addr;
1162 assignAddresses(linkEditSegment);
1163 // No need to page-align fileOff / addr here since this is the last segment.
1164 linkEditSegment->vmSize = addr - linkEditSegment->addr;
1165 linkEditSegment->fileSize = fileOff - linkEditSegment->fileOff;
1166 }
1167
assignAddresses(OutputSegment * seg)1168 void Writer::assignAddresses(OutputSegment *seg) {
1169 seg->fileOff = fileOff;
1170
1171 for (OutputSection *osec : seg->getSections()) {
1172 if (!osec->isNeeded())
1173 continue;
1174 addr = alignToPowerOf2(addr, osec->align);
1175 fileOff = alignToPowerOf2(fileOff, osec->align);
1176 osec->addr = addr;
1177 osec->fileOff = isZeroFill(osec->flags) ? 0 : fileOff;
1178 osec->finalize();
1179 osec->assignAddressesToStartEndSymbols();
1180
1181 addr += osec->getSize();
1182 fileOff += osec->getFileSize();
1183 }
1184 }
1185
openFile()1186 void Writer::openFile() {
1187 Expected<std::unique_ptr<FileOutputBuffer>> bufferOrErr =
1188 FileOutputBuffer::create(config->outputFile, fileOff,
1189 FileOutputBuffer::F_executable);
1190
1191 if (!bufferOrErr)
1192 fatal("failed to open " + config->outputFile + ": " +
1193 llvm::toString(bufferOrErr.takeError()));
1194 buffer = std::move(*bufferOrErr);
1195 in.bufferStart = buffer->getBufferStart();
1196 }
1197
writeSections()1198 void Writer::writeSections() {
1199 TimeTraceScope timeScope("Write output sections");
1200
1201 uint8_t *buf = buffer->getBufferStart();
1202 std::vector<const OutputSection *> osecs;
1203 for (const OutputSegment *seg : outputSegments)
1204 append_range(osecs, seg->getSections());
1205
1206 parallelForEach(osecs.begin(), osecs.end(), [&](const OutputSection *osec) {
1207 osec->writeTo(buf + osec->fileOff);
1208 });
1209 }
1210
applyOptimizationHints()1211 void Writer::applyOptimizationHints() {
1212 if (config->arch() != AK_arm64 || config->ignoreOptimizationHints)
1213 return;
1214
1215 uint8_t *buf = buffer->getBufferStart();
1216 TimeTraceScope timeScope("Apply linker optimization hints");
1217 parallelForEach(inputFiles, [buf](const InputFile *file) {
1218 if (const auto *objFile = dyn_cast<ObjFile>(file))
1219 target->applyOptimizationHints(buf, *objFile);
1220 });
1221 }
1222
1223 // In order to utilize multiple cores, we first split the buffer into chunks,
1224 // compute a hash for each chunk, and then compute a hash value of the hash
1225 // values.
writeUuid()1226 void Writer::writeUuid() {
1227 TimeTraceScope timeScope("Computing UUID");
1228
1229 ArrayRef<uint8_t> data{buffer->getBufferStart(), buffer->getBufferEnd()};
1230 std::vector<ArrayRef<uint8_t>> chunks = split(data, 1024 * 1024);
1231
1232 // Leave one slot for filename
1233 std::vector<uint64_t> hashes(chunks.size() + 1);
1234 parallelFor(0, chunks.size(),
1235 [&](size_t i) { hashes[i] = xxh3_64bits(chunks[i]); });
1236 // Append the output filename so that identical binaries with different names
1237 // don't get the same UUID.
1238 hashes[chunks.size()] = xxh3_64bits(sys::path::filename(config->finalOutput));
1239
1240 uint64_t digest = xxh3_64bits({reinterpret_cast<uint8_t *>(hashes.data()),
1241 hashes.size() * sizeof(uint64_t)});
1242 uuidCommand->writeUuid(digest);
1243 }
1244
1245 // This is step 5 of the algorithm described in the class comment of
1246 // ChainedFixupsSection.
buildFixupChains()1247 void Writer::buildFixupChains() {
1248 if (!config->emitChainedFixups)
1249 return;
1250
1251 const std::vector<Location> &loc = in.chainedFixups->getLocations();
1252 if (loc.empty())
1253 return;
1254
1255 TimeTraceScope timeScope("Build fixup chains");
1256
1257 const uint64_t pageSize = target->getPageSize();
1258 constexpr uint32_t stride = 4; // for DYLD_CHAINED_PTR_64
1259
1260 for (size_t i = 0, count = loc.size(); i < count;) {
1261 const OutputSegment *oseg = loc[i].isec->parent->parent;
1262 uint8_t *buf = buffer->getBufferStart() + oseg->fileOff;
1263 uint64_t pageIdx = loc[i].offset / pageSize;
1264 ++i;
1265
1266 while (i < count && loc[i].isec->parent->parent == oseg &&
1267 (loc[i].offset / pageSize) == pageIdx) {
1268 uint64_t offset = loc[i].offset - loc[i - 1].offset;
1269
1270 auto fail = [&](Twine message) {
1271 error(loc[i].isec->getSegName() + "," + loc[i].isec->getName() +
1272 ", offset " +
1273 Twine(loc[i].offset - loc[i].isec->parent->getSegmentOffset()) +
1274 ": " + message);
1275 };
1276
1277 if (offset < target->wordSize)
1278 return fail("fixups overlap");
1279 if (offset % stride != 0)
1280 return fail(
1281 "fixups are unaligned (offset " + Twine(offset) +
1282 " is not a multiple of the stride). Re-link with -no_fixup_chains");
1283
1284 // The "next" field is in the same location for bind and rebase entries.
1285 reinterpret_cast<dyld_chained_ptr_64_bind *>(buf + loc[i - 1].offset)
1286 ->next = offset / stride;
1287 ++i;
1288 }
1289 }
1290 }
1291
writeCodeSignature()1292 void Writer::writeCodeSignature() {
1293 if (codeSignatureSection) {
1294 TimeTraceScope timeScope("Write code signature");
1295 codeSignatureSection->writeHashes(buffer->getBufferStart());
1296 }
1297 }
1298
writeOutputFile()1299 void Writer::writeOutputFile() {
1300 TimeTraceScope timeScope("Write output file");
1301 openFile();
1302 reportPendingUndefinedSymbols();
1303 if (errorCount())
1304 return;
1305 writeSections();
1306 applyOptimizationHints();
1307 buildFixupChains();
1308 if (config->generateUuid)
1309 writeUuid();
1310 writeCodeSignature();
1311
1312 if (auto e = buffer->commit())
1313 fatal("failed to write output '" + buffer->getPath() +
1314 "': " + toString(std::move(e)));
1315 }
1316
run()1317 template <class LP> void Writer::run() {
1318 treatSpecialUndefineds();
1319 if (config->entry && needsBinding(config->entry))
1320 in.stubs->addEntry(config->entry);
1321
1322 // Canonicalization of all pointers to InputSections should be handled by
1323 // these two scan* methods. I.e. from this point onward, for all live
1324 // InputSections, we should have `isec->canonical() == isec`.
1325 scanSymbols();
1326 if (in.objcStubs->isNeeded())
1327 in.objcStubs->setUp();
1328 if (in.objcMethList->isNeeded())
1329 in.objcMethList->setUp();
1330 scanRelocations();
1331 if (in.initOffsets->isNeeded())
1332 in.initOffsets->setUp();
1333
1334 // Do not proceed if there were undefined or duplicate symbols.
1335 reportPendingUndefinedSymbols();
1336 reportPendingDuplicateSymbols();
1337 if (errorCount())
1338 return;
1339
1340 if (in.stubHelper && in.stubHelper->isNeeded())
1341 in.stubHelper->setUp();
1342
1343 if (in.objCImageInfo->isNeeded())
1344 in.objCImageInfo->finalizeContents();
1345
1346 // At this point, we should know exactly which output sections are needed,
1347 // courtesy of scanSymbols() and scanRelocations().
1348 createOutputSections<LP>();
1349
1350 // After this point, we create no new segments; HOWEVER, we might
1351 // yet create branch-range extension thunks for architectures whose
1352 // hardware call instructions have limited range, e.g., ARM(64).
1353 // The thunks are created as InputSections interspersed among
1354 // the ordinary __TEXT,_text InputSections.
1355 sortSegmentsAndSections();
1356 createLoadCommands<LP>();
1357 finalizeAddresses();
1358
1359 llvm::thread mapFileWriter([&] {
1360 if (LLVM_ENABLE_THREADS && config->timeTraceEnabled)
1361 timeTraceProfilerInitialize(config->timeTraceGranularity, "writeMapFile");
1362 writeMapFile();
1363 if (LLVM_ENABLE_THREADS && config->timeTraceEnabled)
1364 timeTraceProfilerFinishThread();
1365 });
1366
1367 finalizeLinkEditSegment();
1368 writeOutputFile();
1369 mapFileWriter.join();
1370 }
1371
writeResult()1372 template <class LP> void macho::writeResult() { Writer().run<LP>(); }
1373
resetWriter()1374 void macho::resetWriter() { LCDylib::resetInstanceCount(); }
1375
createSyntheticSections()1376 void macho::createSyntheticSections() {
1377 in.header = make<MachHeaderSection>();
1378 if (config->dedupStrings)
1379 in.cStringSection =
1380 make<DeduplicatedCStringSection>(section_names::cString);
1381 else
1382 in.cStringSection = make<CStringSection>(section_names::cString);
1383 in.objcMethnameSection =
1384 make<DeduplicatedCStringSection>(section_names::objcMethname);
1385 in.wordLiteralSection = make<WordLiteralSection>();
1386 if (config->emitChainedFixups) {
1387 in.chainedFixups = make<ChainedFixupsSection>();
1388 } else {
1389 in.rebase = make<RebaseSection>();
1390 in.binding = make<BindingSection>();
1391 in.weakBinding = make<WeakBindingSection>();
1392 in.lazyBinding = make<LazyBindingSection>();
1393 in.lazyPointers = make<LazyPointerSection>();
1394 in.stubHelper = make<StubHelperSection>();
1395 }
1396 in.exports = make<ExportSection>();
1397 in.got = make<GotSection>();
1398 in.tlvPointers = make<TlvPointerSection>();
1399 in.stubs = make<StubsSection>();
1400 in.objcStubs = make<ObjCStubsSection>();
1401 in.unwindInfo = makeUnwindInfoSection();
1402 in.objCImageInfo = make<ObjCImageInfoSection>();
1403 in.initOffsets = make<InitOffsetsSection>();
1404 in.objcMethList = make<ObjCMethListSection>();
1405
1406 // This section contains space for just a single word, and will be used by
1407 // dyld to cache an address to the image loader it uses.
1408 uint8_t *arr = bAlloc().Allocate<uint8_t>(target->wordSize);
1409 memset(arr, 0, target->wordSize);
1410 in.imageLoaderCache = makeSyntheticInputSection(
1411 segment_names::data, section_names::data, S_REGULAR,
1412 ArrayRef<uint8_t>{arr, target->wordSize},
1413 /*align=*/target->wordSize);
1414 assert(in.imageLoaderCache->live);
1415 }
1416
1417 OutputSection *macho::firstTLVDataSection = nullptr;
1418
1419 template void macho::writeResult<LP64>();
1420 template void macho::writeResult<ILP32>();
1421