1 //===- lib/ReaderWriter/MachO/MachONormalizedFileBinaryReader.cpp ---------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
11 /// \file For mach-o object files, this implementation converts from
12 /// mach-o on-disk binary format to in-memory normalized mach-o.
24 #include "MachONormalizedFile.h"
25 #include "ArchHandler.h"
26 #include "MachONormalizedFileBinaryUtils.h"
27 #include "lld/Core/Error.h"
28 #include "lld/Core/LLVM.h"
29 #include "lld/Core/SharedLibraryFile.h"
30 #include "llvm/ADT/SmallString.h"
31 #include "llvm/ADT/StringRef.h"
32 #include "llvm/ADT/StringSwitch.h"
33 #include "llvm/ADT/Twine.h"
34 #include "llvm/Object/MachO.h"
35 #include "llvm/Support/Casting.h"
36 #include "llvm/Support/Errc.h"
37 #include "llvm/Support/ErrorHandling.h"
38 #include "llvm/Support/FileOutputBuffer.h"
39 #include "llvm/Support/Host.h"
40 #include "llvm/Support/MachO.h"
41 #include "llvm/Support/MemoryBuffer.h"
42 #include "llvm/Support/raw_ostream.h"
44 #include <system_error>
46 using namespace llvm::MachO;
47 using llvm::object::ExportEntry;
48 using llvm::object::MachOObjectFile;
52 namespace normalized {
54 // Utility to call a lambda expression on each load command.
55 static std::error_code forEachLoadCommand(
56 StringRef lcRange, unsigned lcCount, bool isBig, bool is64,
57 std::function<bool(uint32_t cmd, uint32_t size, const char *lc)> func) {
58 const char* p = lcRange.begin();
59 for (unsigned i=0; i < lcCount; ++i) {
60 const load_command *lc = reinterpret_cast<const load_command*>(p);
62 const load_command *slc = lc;
63 if (isBig != llvm::sys::IsBigEndianHost) {
64 memcpy(&lcCopy, lc, sizeof(load_command));
68 if ( (p + slc->cmdsize) > lcRange.end() )
69 return make_error_code(llvm::errc::executable_format_error);
71 if (func(slc->cmd, slc->cmdsize, p))
72 return std::error_code();
77 return std::error_code();
80 static std::error_code appendRelocations(Relocations &relocs, StringRef buffer,
82 uint32_t reloff, uint32_t nreloc) {
83 if ((reloff + nreloc*8) > buffer.size())
84 return make_error_code(llvm::errc::executable_format_error);
85 const any_relocation_info* relocsArray =
86 reinterpret_cast<const any_relocation_info*>(buffer.begin()+reloff);
88 for(uint32_t i=0; i < nreloc; ++i) {
89 relocs.push_back(unpackRelocation(relocsArray[i], bigEndian));
91 return std::error_code();
94 static std::error_code
95 appendIndirectSymbols(IndirectSymbols &isyms, StringRef buffer, bool isBig,
96 uint32_t istOffset, uint32_t istCount,
97 uint32_t startIndex, uint32_t count) {
98 if ((istOffset + istCount*4) > buffer.size())
99 return make_error_code(llvm::errc::executable_format_error);
100 if (startIndex+count > istCount)
101 return make_error_code(llvm::errc::executable_format_error);
102 const uint8_t *indirectSymbolArray = (const uint8_t *)buffer.data();
104 for(uint32_t i=0; i < count; ++i) {
105 isyms.push_back(read32(
106 indirectSymbolArray + (startIndex + i) * sizeof(uint32_t), isBig));
108 return std::error_code();
112 template <typename T> static T readBigEndian(T t) {
113 if (llvm::sys::IsLittleEndianHost)
114 llvm::sys::swapByteOrder(t);
119 static bool isMachOHeader(const mach_header *mh, bool &is64, bool &isBig) {
120 switch (read32(&mh->magic, false)) {
121 case llvm::MachO::MH_MAGIC:
125 case llvm::MachO::MH_MAGIC_64:
129 case llvm::MachO::MH_CIGAM:
133 case llvm::MachO::MH_CIGAM_64:
143 bool isThinObjectFile(StringRef path, MachOLinkingContext::Arch &arch) {
144 // Try opening and mapping file at path.
145 ErrorOr<std::unique_ptr<MemoryBuffer>> b = MemoryBuffer::getFileOrSTDIN(path);
149 // If file length < 32 it is too small to be mach-o object file.
150 StringRef fileBuffer = b->get()->getBuffer();
151 if (fileBuffer.size() < 32)
154 // If file buffer does not start with MH_MAGIC (and variants), not obj file.
155 const mach_header *mh = reinterpret_cast<const mach_header *>(
158 if (!isMachOHeader(mh, is64, isBig))
161 // If not MH_OBJECT, not object file.
162 if (read32(&mh->filetype, isBig) != MH_OBJECT)
165 // Lookup up arch from cpu/subtype pair.
166 arch = MachOLinkingContext::archFromCpuType(
167 read32(&mh->cputype, isBig),
168 read32(&mh->cpusubtype, isBig));
173 bool sliceFromFatFile(const MemoryBuffer &mb, MachOLinkingContext::Arch arch,
174 uint32_t &offset, uint32_t &size) {
175 const char *start = mb.getBufferStart();
176 const llvm::MachO::fat_header *fh =
177 reinterpret_cast<const llvm::MachO::fat_header *>(start);
178 if (readBigEndian(fh->magic) != llvm::MachO::FAT_MAGIC)
180 uint32_t nfat_arch = readBigEndian(fh->nfat_arch);
181 const fat_arch *fstart =
182 reinterpret_cast<const fat_arch *>(start + sizeof(fat_header));
183 const fat_arch *fend =
184 reinterpret_cast<const fat_arch *>(start + sizeof(fat_header) +
185 sizeof(fat_arch) * nfat_arch);
186 const uint32_t reqCpuType = MachOLinkingContext::cpuTypeFromArch(arch);
187 const uint32_t reqCpuSubtype = MachOLinkingContext::cpuSubtypeFromArch(arch);
188 for (const fat_arch *fa = fstart; fa < fend; ++fa) {
189 if ((readBigEndian(fa->cputype) == reqCpuType) &&
190 (readBigEndian(fa->cpusubtype) == reqCpuSubtype)) {
191 offset = readBigEndian(fa->offset);
192 size = readBigEndian(fa->size);
193 if ((offset + size) > mb.getBufferSize())
201 /// Reads a mach-o file and produces an in-memory normalized view.
202 ErrorOr<std::unique_ptr<NormalizedFile>>
203 readBinary(std::unique_ptr<MemoryBuffer> &mb,
204 const MachOLinkingContext::Arch arch) {
205 // Make empty NormalizedFile.
206 std::unique_ptr<NormalizedFile> f(new NormalizedFile());
208 const char *start = mb->getBufferStart();
209 size_t objSize = mb->getBufferSize();
210 const mach_header *mh = reinterpret_cast<const mach_header *>(start);
212 uint32_t sliceOffset;
214 if (sliceFromFatFile(*mb, arch, sliceOffset, sliceSize)) {
215 start = &start[sliceOffset];
217 mh = reinterpret_cast<const mach_header *>(start);
220 // Determine endianness and pointer size for mach-o file.
222 if (!isMachOHeader(mh, is64, isBig))
223 return make_error_code(llvm::errc::executable_format_error);
225 // Endian swap header, if needed.
226 mach_header headerCopy;
227 const mach_header *smh = mh;
228 if (isBig != llvm::sys::IsBigEndianHost) {
229 memcpy(&headerCopy, mh, sizeof(mach_header));
230 swapStruct(headerCopy);
234 // Validate head and load commands fit in buffer.
235 const uint32_t lcCount = smh->ncmds;
236 const char *lcStart =
237 start + (is64 ? sizeof(mach_header_64) : sizeof(mach_header));
238 StringRef lcRange(lcStart, smh->sizeofcmds);
239 if (lcRange.end() > (start + objSize))
240 return make_error_code(llvm::errc::executable_format_error);
242 // Get architecture from mach_header.
243 f->arch = MachOLinkingContext::archFromCpuType(smh->cputype, smh->cpusubtype);
244 if (f->arch != arch) {
245 return make_dynamic_error_code(Twine("file is wrong architecture. Expected "
246 "(" + MachOLinkingContext::nameFromArch(arch)
248 + MachOLinkingContext::nameFromArch(f->arch)
251 // Copy file type and flags
252 f->fileType = HeaderFileType(smh->filetype);
253 f->flags = smh->flags;
256 // Pre-scan load commands looking for indirect symbol table.
257 uint32_t indirectSymbolTableOffset = 0;
258 uint32_t indirectSymbolTableCount = 0;
259 std::error_code ec = forEachLoadCommand(lcRange, lcCount, isBig, is64,
260 [&](uint32_t cmd, uint32_t size,
261 const char *lc) -> bool {
262 if (cmd == LC_DYSYMTAB) {
263 const dysymtab_command *d = reinterpret_cast<const dysymtab_command*>(lc);
264 indirectSymbolTableOffset = read32(&d->indirectsymoff, isBig);
265 indirectSymbolTableCount = read32(&d->nindirectsyms, isBig);
273 // Walk load commands looking for segments/sections and the symbol table.
274 const data_in_code_entry *dataInCode = nullptr;
275 const dyld_info_command *dyldInfo = nullptr;
276 uint32_t dataInCodeSize = 0;
277 ec = forEachLoadCommand(lcRange, lcCount, isBig, is64,
278 [&] (uint32_t cmd, uint32_t size, const char* lc) -> bool {
282 const segment_command_64 *seg =
283 reinterpret_cast<const segment_command_64*>(lc);
284 const unsigned sectionCount = read32(&seg->nsects, isBig);
285 const section_64 *sects = reinterpret_cast<const section_64*>
286 (lc + sizeof(segment_command_64));
287 const unsigned lcSize = sizeof(segment_command_64)
288 + sectionCount*sizeof(section_64);
289 // Verify sections don't extend beyond end of segment load command.
292 for (unsigned i=0; i < sectionCount; ++i) {
293 const section_64 *sect = §s[i];
295 section.segmentName = getString16(sect->segname);
296 section.sectionName = getString16(sect->sectname);
297 section.type = (SectionType)(read32(§->flags, isBig) &
299 section.attributes = read32(§->flags, isBig) & SECTION_ATTRIBUTES;
300 section.alignment = read32(§->align, isBig);
301 section.address = read64(§->addr, isBig);
302 const uint8_t *content =
303 (const uint8_t *)start + read32(§->offset, isBig);
304 size_t contentSize = read64(§->size, isBig);
305 // Note: this assign() is copying the content bytes. Ideally,
306 // we can use a custom allocator for vector to avoid the copy.
307 section.content = llvm::makeArrayRef(content, contentSize);
308 appendRelocations(section.relocations, mb->getBuffer(), isBig,
309 read32(§->reloff, isBig),
310 read32(§->nreloc, isBig));
311 if (section.type == S_NON_LAZY_SYMBOL_POINTERS) {
312 appendIndirectSymbols(section.indirectSymbols, mb->getBuffer(),
314 indirectSymbolTableOffset,
315 indirectSymbolTableCount,
316 read32(§->reserved1, isBig),
319 f->sections.push_back(section);
325 const segment_command *seg =
326 reinterpret_cast<const segment_command*>(lc);
327 const unsigned sectionCount = read32(&seg->nsects, isBig);
328 const section *sects = reinterpret_cast<const section*>
329 (lc + sizeof(segment_command));
330 const unsigned lcSize = sizeof(segment_command)
331 + sectionCount*sizeof(section);
332 // Verify sections don't extend beyond end of segment load command.
335 for (unsigned i=0; i < sectionCount; ++i) {
336 const section *sect = §s[i];
338 section.segmentName = getString16(sect->segname);
339 section.sectionName = getString16(sect->sectname);
340 section.type = (SectionType)(read32(§->flags, isBig) &
343 read32((const uint8_t *)§->flags, isBig) & SECTION_ATTRIBUTES;
344 section.alignment = read32(§->align, isBig);
345 section.address = read32(§->addr, isBig);
346 const uint8_t *content =
347 (const uint8_t *)start + read32(§->offset, isBig);
348 size_t contentSize = read32(§->size, isBig);
349 // Note: this assign() is copying the content bytes. Ideally,
350 // we can use a custom allocator for vector to avoid the copy.
351 section.content = llvm::makeArrayRef(content, contentSize);
352 appendRelocations(section.relocations, mb->getBuffer(), isBig,
353 read32(§->reloff, isBig),
354 read32(§->nreloc, isBig));
355 if (section.type == S_NON_LAZY_SYMBOL_POINTERS) {
356 appendIndirectSymbols(
357 section.indirectSymbols, mb->getBuffer(), isBig,
358 indirectSymbolTableOffset, indirectSymbolTableCount,
359 read32(§->reserved1, isBig), contentSize / 4);
361 f->sections.push_back(section);
366 const symtab_command *st = reinterpret_cast<const symtab_command*>(lc);
367 const char *strings = start + read32(&st->stroff, isBig);
368 const uint32_t strSize = read32(&st->strsize, isBig);
369 // Validate string pool and symbol table all in buffer.
370 if (read32((const uint8_t *)&st->stroff, isBig) +
371 read32((const uint8_t *)&st->strsize, isBig) >
375 const uint32_t symOffset = read32(&st->symoff, isBig);
376 const uint32_t symCount = read32(&st->nsyms, isBig);
377 if ( symOffset+(symCount*sizeof(nlist_64)) > objSize)
379 const nlist_64 *symbols =
380 reinterpret_cast<const nlist_64 *>(start + symOffset);
381 // Convert each nlist_64 to a lld::mach_o::normalized::Symbol.
382 for(uint32_t i=0; i < symCount; ++i) {
383 const nlist_64 *sin = &symbols[i];
385 if (isBig != llvm::sys::IsBigEndianHost) {
386 tempSym = *sin; swapStruct(tempSym); sin = &tempSym;
389 if (sin->n_strx > strSize)
391 sout.name = &strings[sin->n_strx];
392 sout.type = (NListType)(sin->n_type & N_TYPE);
393 sout.scope = (sin->n_type & (N_PEXT|N_EXT));
394 sout.sect = sin->n_sect;
395 sout.desc = sin->n_desc;
396 sout.value = sin->n_value;
397 if (sout.type == N_UNDF)
398 f->undefinedSymbols.push_back(sout);
399 else if (sin->n_type & N_EXT)
400 f->globalSymbols.push_back(sout);
402 f->localSymbols.push_back(sout);
405 const uint32_t symOffset = read32(&st->symoff, isBig);
406 const uint32_t symCount = read32(&st->nsyms, isBig);
407 if ( symOffset+(symCount*sizeof(nlist)) > objSize)
409 const nlist *symbols =
410 reinterpret_cast<const nlist *>(start + symOffset);
411 // Convert each nlist to a lld::mach_o::normalized::Symbol.
412 for(uint32_t i=0; i < symCount; ++i) {
413 const nlist *sin = &symbols[i];
415 if (isBig != llvm::sys::IsBigEndianHost) {
416 tempSym = *sin; swapStruct(tempSym); sin = &tempSym;
419 if (sin->n_strx > strSize)
421 sout.name = &strings[sin->n_strx];
422 sout.type = (NListType)(sin->n_type & N_TYPE);
423 sout.scope = (sin->n_type & (N_PEXT|N_EXT));
424 sout.sect = sin->n_sect;
425 sout.desc = sin->n_desc;
426 sout.value = sin->n_value;
427 if (sout.type == N_UNDF)
428 f->undefinedSymbols.push_back(sout);
429 else if (sout.scope == (SymbolScope)N_EXT)
430 f->globalSymbols.push_back(sout);
432 f->localSymbols.push_back(sout);
438 const dylib_command *dl = reinterpret_cast<const dylib_command*>(lc);
439 f->installName = lc + read32(&dl->dylib.name, isBig);
440 f->currentVersion = read32(&dl->dylib.current_version, isBig);
441 f->compatVersion = read32(&dl->dylib.compatibility_version, isBig);
444 case LC_DATA_IN_CODE: {
445 const linkedit_data_command *ldc =
446 reinterpret_cast<const linkedit_data_command*>(lc);
447 dataInCode = reinterpret_cast<const data_in_code_entry *>(
448 start + read32(&ldc->dataoff, isBig));
449 dataInCodeSize = read32(&ldc->datasize, isBig);
453 case LC_LOAD_WEAK_DYLIB:
454 case LC_REEXPORT_DYLIB:
455 case LC_LOAD_UPWARD_DYLIB: {
456 const dylib_command *dl = reinterpret_cast<const dylib_command*>(lc);
457 DependentDylib entry;
458 entry.path = lc + read32(&dl->dylib.name, isBig);
459 entry.kind = LoadCommandType(cmd);
460 entry.compatVersion = read32(&dl->dylib.compatibility_version, isBig);
461 entry.currentVersion = read32(&dl->dylib.current_version, isBig);
462 f->dependentDylibs.push_back(entry);
466 const rpath_command *rpc = reinterpret_cast<const rpath_command *>(lc);
467 f->rpaths.push_back(lc + read32(&rpc->path, isBig));
471 case LC_DYLD_INFO_ONLY:
472 dyldInfo = reinterpret_cast<const dyld_info_command*>(lc);
481 // Convert on-disk data_in_code_entry array to DataInCode vector.
482 for (unsigned i=0; i < dataInCodeSize/sizeof(data_in_code_entry); ++i) {
484 entry.offset = read32(&dataInCode[i].offset, isBig);
485 entry.length = read16(&dataInCode[i].length, isBig);
487 (DataRegionType)read16((const uint8_t *)&dataInCode[i].kind, isBig);
488 f->dataInCode.push_back(entry);
493 // If any exports, extract and add to normalized exportInfo vector.
494 if (dyldInfo->export_size) {
495 const uint8_t *trieStart = reinterpret_cast<const uint8_t*>(start +
496 dyldInfo->export_off);
497 ArrayRef<uint8_t> trie(trieStart, dyldInfo->export_size);
498 for (const ExportEntry &trieExport : MachOObjectFile::exports(trie)) {
500 normExport.name = trieExport.name().copy(f->ownedAllocations);
501 normExport.offset = trieExport.address();
502 normExport.kind = ExportSymbolKind(trieExport.flags() & EXPORT_SYMBOL_FLAGS_KIND_MASK);
503 normExport.flags = trieExport.flags() & ~EXPORT_SYMBOL_FLAGS_KIND_MASK;
504 normExport.otherOffset = trieExport.other();
505 if (!trieExport.otherName().empty())
506 normExport.otherName = trieExport.otherName().copy(f->ownedAllocations);
507 f->exportInfo.push_back(normExport);
515 class MachOObjectReader : public Reader {
517 MachOObjectReader(MachOLinkingContext &ctx) : _ctx(ctx) {}
519 bool canParse(file_magic magic, StringRef ext,
520 const MemoryBuffer &mb) const override {
522 case llvm::sys::fs::file_magic::macho_object:
523 return (mb.getBufferSize() > 32);
530 loadFile(std::unique_ptr<MemoryBuffer> mb, const Registry ®istry,
531 std::vector<std::unique_ptr<File>> &result) const override {
532 auto *file = new MachOFile(std::move(mb), &_ctx);
533 result.push_back(std::unique_ptr<MachOFile>(file));
534 return std::error_code();
538 MachOLinkingContext &_ctx;
541 class MachODylibReader : public Reader {
543 MachODylibReader(MachOLinkingContext &ctx) : _ctx(ctx) {}
545 bool canParse(file_magic magic, StringRef ext,
546 const MemoryBuffer &mb) const override {
548 case llvm::sys::fs::file_magic::macho_dynamically_linked_shared_lib:
549 case llvm::sys::fs::file_magic::macho_dynamically_linked_shared_lib_stub:
550 return (mb.getBufferSize() > 32);
557 loadFile(std::unique_ptr<MemoryBuffer> mb, const Registry ®istry,
558 std::vector<std::unique_ptr<File>> &result) const override {
559 auto *file = new MachODylibFile(std::move(mb), &_ctx);
560 result.push_back(std::unique_ptr<MachODylibFile>(file));
561 return std::error_code();
565 MachOLinkingContext &_ctx;
568 } // namespace normalized
569 } // namespace mach_o
571 void Registry::addSupportMachOObjects(MachOLinkingContext &ctx) {
572 MachOLinkingContext::Arch arch = ctx.arch();
573 add(std::unique_ptr<Reader>(new mach_o::normalized::MachOObjectReader(ctx)));
574 add(std::unique_ptr<Reader>(new mach_o::normalized::MachODylibReader(ctx)));
575 addKindTable(Reference::KindNamespace::mach_o, ctx.archHandler().kindArch(),
576 ctx.archHandler().kindStrings());
577 add(std::unique_ptr<YamlIOTaggedDocumentHandler>(
578 new mach_o::MachOYamlIOTaggedDocumentHandler(arch)));