1 //===- MachOWriter.cpp ------------------------------------------*- C++ -*-===//
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 "MachOWriter.h"
10 #include "MachOLayoutBuilder.h"
11 #include "MachOObject.h"
12 #include "llvm/ADT/STLExtras.h"
13 #include "llvm/BinaryFormat/MachO.h"
14 #include "llvm/Object/MachO.h"
15 #include "llvm/Support/Errc.h"
16 #include "llvm/Support/ErrorHandling.h"
17 #include "llvm/Support/SHA256.h"
18 #include <memory>
19 
20 #if defined(__APPLE__)
21 #include <sys/mman.h>
22 #endif
23 
24 using namespace llvm;
25 using namespace llvm::objcopy::macho;
26 using namespace llvm::support::endian;
27 
headerSize() const28 size_t MachOWriter::headerSize() const {
29   return Is64Bit ? sizeof(MachO::mach_header_64) : sizeof(MachO::mach_header);
30 }
31 
loadCommandsSize() const32 size_t MachOWriter::loadCommandsSize() const { return O.Header.SizeOfCmds; }
33 
symTableSize() const34 size_t MachOWriter::symTableSize() const {
35   return O.SymTable.Symbols.size() *
36          (Is64Bit ? sizeof(MachO::nlist_64) : sizeof(MachO::nlist));
37 }
38 
totalSize() const39 size_t MachOWriter::totalSize() const {
40   // Going from tail to head and looking for an appropriate "anchor" to
41   // calculate the total size assuming that all the offsets are either valid
42   // ("true") or 0 (0 indicates that the corresponding part is missing).
43 
44   SmallVector<size_t, 7> Ends;
45   if (O.SymTabCommandIndex) {
46     const MachO::symtab_command &SymTabCommand =
47         O.LoadCommands[*O.SymTabCommandIndex]
48             .MachOLoadCommand.symtab_command_data;
49     if (SymTabCommand.symoff)
50       Ends.push_back(SymTabCommand.symoff + symTableSize());
51     if (SymTabCommand.stroff)
52       Ends.push_back(SymTabCommand.stroff + SymTabCommand.strsize);
53   }
54   if (O.DyLdInfoCommandIndex) {
55     const MachO::dyld_info_command &DyLdInfoCommand =
56         O.LoadCommands[*O.DyLdInfoCommandIndex]
57             .MachOLoadCommand.dyld_info_command_data;
58     if (DyLdInfoCommand.rebase_off) {
59       assert((DyLdInfoCommand.rebase_size == O.Rebases.Opcodes.size()) &&
60              "Incorrect rebase opcodes size");
61       Ends.push_back(DyLdInfoCommand.rebase_off + DyLdInfoCommand.rebase_size);
62     }
63     if (DyLdInfoCommand.bind_off) {
64       assert((DyLdInfoCommand.bind_size == O.Binds.Opcodes.size()) &&
65              "Incorrect bind opcodes size");
66       Ends.push_back(DyLdInfoCommand.bind_off + DyLdInfoCommand.bind_size);
67     }
68     if (DyLdInfoCommand.weak_bind_off) {
69       assert((DyLdInfoCommand.weak_bind_size == O.WeakBinds.Opcodes.size()) &&
70              "Incorrect weak bind opcodes size");
71       Ends.push_back(DyLdInfoCommand.weak_bind_off +
72                      DyLdInfoCommand.weak_bind_size);
73     }
74     if (DyLdInfoCommand.lazy_bind_off) {
75       assert((DyLdInfoCommand.lazy_bind_size == O.LazyBinds.Opcodes.size()) &&
76              "Incorrect lazy bind opcodes size");
77       Ends.push_back(DyLdInfoCommand.lazy_bind_off +
78                      DyLdInfoCommand.lazy_bind_size);
79     }
80     if (DyLdInfoCommand.export_off) {
81       assert((DyLdInfoCommand.export_size == O.Exports.Trie.size()) &&
82              "Incorrect trie size");
83       Ends.push_back(DyLdInfoCommand.export_off + DyLdInfoCommand.export_size);
84     }
85   }
86 
87   if (O.DySymTabCommandIndex) {
88     const MachO::dysymtab_command &DySymTabCommand =
89         O.LoadCommands[*O.DySymTabCommandIndex]
90             .MachOLoadCommand.dysymtab_command_data;
91 
92     if (DySymTabCommand.indirectsymoff)
93       Ends.push_back(DySymTabCommand.indirectsymoff +
94                      sizeof(uint32_t) * O.IndirectSymTable.Symbols.size());
95   }
96 
97   for (std::optional<size_t> LinkEditDataCommandIndex :
98        {O.CodeSignatureCommandIndex, O.DylibCodeSignDRsIndex,
99         O.DataInCodeCommandIndex, O.LinkerOptimizationHintCommandIndex,
100         O.FunctionStartsCommandIndex, O.ChainedFixupsCommandIndex,
101         O.ExportsTrieCommandIndex})
102     if (LinkEditDataCommandIndex) {
103       const MachO::linkedit_data_command &LinkEditDataCommand =
104           O.LoadCommands[*LinkEditDataCommandIndex]
105               .MachOLoadCommand.linkedit_data_command_data;
106       if (LinkEditDataCommand.dataoff)
107         Ends.push_back(LinkEditDataCommand.dataoff +
108                        LinkEditDataCommand.datasize);
109     }
110 
111   // Otherwise, use the last section / reloction.
112   for (const LoadCommand &LC : O.LoadCommands)
113     for (const std::unique_ptr<Section> &S : LC.Sections) {
114       if (!S->hasValidOffset()) {
115         assert((S->Offset == 0) && "Skipped section's offset must be zero");
116         assert((S->isVirtualSection() || S->Size == 0) &&
117                "Non-zero-fill sections with zero offset must have zero size");
118         continue;
119       }
120       assert((S->Offset != 0) &&
121              "Non-zero-fill section's offset cannot be zero");
122       Ends.push_back(S->Offset + S->Size);
123       if (S->RelOff)
124         Ends.push_back(S->RelOff +
125                        S->NReloc * sizeof(MachO::any_relocation_info));
126     }
127 
128   if (!Ends.empty())
129     return *std::max_element(Ends.begin(), Ends.end());
130 
131   // Otherwise, we have only Mach header and load commands.
132   return headerSize() + loadCommandsSize();
133 }
134 
writeHeader()135 void MachOWriter::writeHeader() {
136   MachO::mach_header_64 Header;
137 
138   Header.magic = O.Header.Magic;
139   Header.cputype = O.Header.CPUType;
140   Header.cpusubtype = O.Header.CPUSubType;
141   Header.filetype = O.Header.FileType;
142   Header.ncmds = O.Header.NCmds;
143   Header.sizeofcmds = O.Header.SizeOfCmds;
144   Header.flags = O.Header.Flags;
145   Header.reserved = O.Header.Reserved;
146 
147   if (IsLittleEndian != sys::IsLittleEndianHost)
148     MachO::swapStruct(Header);
149 
150   auto HeaderSize =
151       Is64Bit ? sizeof(MachO::mach_header_64) : sizeof(MachO::mach_header);
152   memcpy(Buf->getBufferStart(), &Header, HeaderSize);
153 }
154 
writeLoadCommands()155 void MachOWriter::writeLoadCommands() {
156   uint8_t *Begin =
157       reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + headerSize();
158   for (const LoadCommand &LC : O.LoadCommands) {
159     // Construct a load command.
160     MachO::macho_load_command MLC = LC.MachOLoadCommand;
161     switch (MLC.load_command_data.cmd) {
162     case MachO::LC_SEGMENT:
163       if (IsLittleEndian != sys::IsLittleEndianHost)
164         MachO::swapStruct(MLC.segment_command_data);
165       memcpy(Begin, &MLC.segment_command_data, sizeof(MachO::segment_command));
166       Begin += sizeof(MachO::segment_command);
167 
168       for (const std::unique_ptr<Section> &Sec : LC.Sections)
169         writeSectionInLoadCommand<MachO::section>(*Sec, Begin);
170       continue;
171     case MachO::LC_SEGMENT_64:
172       if (IsLittleEndian != sys::IsLittleEndianHost)
173         MachO::swapStruct(MLC.segment_command_64_data);
174       memcpy(Begin, &MLC.segment_command_64_data,
175              sizeof(MachO::segment_command_64));
176       Begin += sizeof(MachO::segment_command_64);
177 
178       for (const std::unique_ptr<Section> &Sec : LC.Sections)
179         writeSectionInLoadCommand<MachO::section_64>(*Sec, Begin);
180       continue;
181     }
182 
183 #define HANDLE_LOAD_COMMAND(LCName, LCValue, LCStruct)                         \
184   case MachO::LCName:                                                          \
185     assert(sizeof(MachO::LCStruct) + LC.Payload.size() ==                      \
186            MLC.load_command_data.cmdsize);                                     \
187     if (IsLittleEndian != sys::IsLittleEndianHost)                             \
188       MachO::swapStruct(MLC.LCStruct##_data);                                  \
189     memcpy(Begin, &MLC.LCStruct##_data, sizeof(MachO::LCStruct));              \
190     Begin += sizeof(MachO::LCStruct);                                          \
191     if (!LC.Payload.empty())                                                   \
192       memcpy(Begin, LC.Payload.data(), LC.Payload.size());                     \
193     Begin += LC.Payload.size();                                                \
194     break;
195 
196     // Copy the load command as it is.
197     switch (MLC.load_command_data.cmd) {
198     default:
199       assert(sizeof(MachO::load_command) + LC.Payload.size() ==
200              MLC.load_command_data.cmdsize);
201       if (IsLittleEndian != sys::IsLittleEndianHost)
202         MachO::swapStruct(MLC.load_command_data);
203       memcpy(Begin, &MLC.load_command_data, sizeof(MachO::load_command));
204       Begin += sizeof(MachO::load_command);
205       if (!LC.Payload.empty())
206         memcpy(Begin, LC.Payload.data(), LC.Payload.size());
207       Begin += LC.Payload.size();
208       break;
209 #include "llvm/BinaryFormat/MachO.def"
210     }
211   }
212 }
213 
214 template <typename StructType>
writeSectionInLoadCommand(const Section & Sec,uint8_t * & Out)215 void MachOWriter::writeSectionInLoadCommand(const Section &Sec, uint8_t *&Out) {
216   StructType Temp;
217   assert(Sec.Segname.size() <= sizeof(Temp.segname) && "too long segment name");
218   assert(Sec.Sectname.size() <= sizeof(Temp.sectname) &&
219          "too long section name");
220   memset(&Temp, 0, sizeof(StructType));
221   memcpy(Temp.segname, Sec.Segname.data(), Sec.Segname.size());
222   memcpy(Temp.sectname, Sec.Sectname.data(), Sec.Sectname.size());
223   Temp.addr = Sec.Addr;
224   Temp.size = Sec.Size;
225   Temp.offset = Sec.Offset;
226   Temp.align = Sec.Align;
227   Temp.reloff = Sec.RelOff;
228   Temp.nreloc = Sec.NReloc;
229   Temp.flags = Sec.Flags;
230   Temp.reserved1 = Sec.Reserved1;
231   Temp.reserved2 = Sec.Reserved2;
232 
233   if (IsLittleEndian != sys::IsLittleEndianHost)
234     MachO::swapStruct(Temp);
235   memcpy(Out, &Temp, sizeof(StructType));
236   Out += sizeof(StructType);
237 }
238 
writeSections()239 void MachOWriter::writeSections() {
240   for (const LoadCommand &LC : O.LoadCommands)
241     for (const std::unique_ptr<Section> &Sec : LC.Sections) {
242       if (!Sec->hasValidOffset()) {
243         assert((Sec->Offset == 0) && "Skipped section's offset must be zero");
244         assert((Sec->isVirtualSection() || Sec->Size == 0) &&
245                "Non-zero-fill sections with zero offset must have zero size");
246         continue;
247       }
248 
249       assert(Sec->Offset && "Section offset can not be zero");
250       assert((Sec->Size == Sec->Content.size()) && "Incorrect section size");
251       memcpy(Buf->getBufferStart() + Sec->Offset, Sec->Content.data(),
252              Sec->Content.size());
253       for (size_t Index = 0; Index < Sec->Relocations.size(); ++Index) {
254         RelocationInfo RelocInfo = Sec->Relocations[Index];
255         if (!RelocInfo.Scattered && !RelocInfo.IsAddend) {
256           const uint32_t SymbolNum = RelocInfo.Extern
257                                          ? (*RelocInfo.Symbol)->Index
258                                          : (*RelocInfo.Sec)->Index;
259           RelocInfo.setPlainRelocationSymbolNum(SymbolNum, IsLittleEndian);
260         }
261         if (IsLittleEndian != sys::IsLittleEndianHost)
262           MachO::swapStruct(
263               reinterpret_cast<MachO::any_relocation_info &>(RelocInfo.Info));
264         memcpy(Buf->getBufferStart() + Sec->RelOff +
265                    Index * sizeof(MachO::any_relocation_info),
266                &RelocInfo.Info, sizeof(RelocInfo.Info));
267       }
268     }
269 }
270 
271 template <typename NListType>
writeNListEntry(const SymbolEntry & SE,bool IsLittleEndian,char * & Out,uint32_t Nstrx)272 void writeNListEntry(const SymbolEntry &SE, bool IsLittleEndian, char *&Out,
273                      uint32_t Nstrx) {
274   NListType ListEntry;
275   ListEntry.n_strx = Nstrx;
276   ListEntry.n_type = SE.n_type;
277   ListEntry.n_sect = SE.n_sect;
278   ListEntry.n_desc = SE.n_desc;
279   ListEntry.n_value = SE.n_value;
280 
281   if (IsLittleEndian != sys::IsLittleEndianHost)
282     MachO::swapStruct(ListEntry);
283   memcpy(Out, reinterpret_cast<const char *>(&ListEntry), sizeof(NListType));
284   Out += sizeof(NListType);
285 }
286 
writeStringTable()287 void MachOWriter::writeStringTable() {
288   if (!O.SymTabCommandIndex)
289     return;
290   const MachO::symtab_command &SymTabCommand =
291       O.LoadCommands[*O.SymTabCommandIndex]
292           .MachOLoadCommand.symtab_command_data;
293 
294   uint8_t *StrTable = (uint8_t *)Buf->getBufferStart() + SymTabCommand.stroff;
295   LayoutBuilder.getStringTableBuilder().write(StrTable);
296 }
297 
writeSymbolTable()298 void MachOWriter::writeSymbolTable() {
299   if (!O.SymTabCommandIndex)
300     return;
301   const MachO::symtab_command &SymTabCommand =
302       O.LoadCommands[*O.SymTabCommandIndex]
303           .MachOLoadCommand.symtab_command_data;
304 
305   char *SymTable = (char *)Buf->getBufferStart() + SymTabCommand.symoff;
306   for (auto &Symbol : O.SymTable.Symbols) {
307     SymbolEntry *Sym = Symbol.get();
308     uint32_t Nstrx = LayoutBuilder.getStringTableBuilder().getOffset(Sym->Name);
309 
310     if (Is64Bit)
311       writeNListEntry<MachO::nlist_64>(*Sym, IsLittleEndian, SymTable, Nstrx);
312     else
313       writeNListEntry<MachO::nlist>(*Sym, IsLittleEndian, SymTable, Nstrx);
314   }
315 }
316 
writeRebaseInfo()317 void MachOWriter::writeRebaseInfo() {
318   if (!O.DyLdInfoCommandIndex)
319     return;
320   const MachO::dyld_info_command &DyLdInfoCommand =
321       O.LoadCommands[*O.DyLdInfoCommandIndex]
322           .MachOLoadCommand.dyld_info_command_data;
323   char *Out = (char *)Buf->getBufferStart() + DyLdInfoCommand.rebase_off;
324   assert((DyLdInfoCommand.rebase_size == O.Rebases.Opcodes.size()) &&
325          "Incorrect rebase opcodes size");
326   memcpy(Out, O.Rebases.Opcodes.data(), O.Rebases.Opcodes.size());
327 }
328 
writeBindInfo()329 void MachOWriter::writeBindInfo() {
330   if (!O.DyLdInfoCommandIndex)
331     return;
332   const MachO::dyld_info_command &DyLdInfoCommand =
333       O.LoadCommands[*O.DyLdInfoCommandIndex]
334           .MachOLoadCommand.dyld_info_command_data;
335   char *Out = (char *)Buf->getBufferStart() + DyLdInfoCommand.bind_off;
336   assert((DyLdInfoCommand.bind_size == O.Binds.Opcodes.size()) &&
337          "Incorrect bind opcodes size");
338   memcpy(Out, O.Binds.Opcodes.data(), O.Binds.Opcodes.size());
339 }
340 
writeWeakBindInfo()341 void MachOWriter::writeWeakBindInfo() {
342   if (!O.DyLdInfoCommandIndex)
343     return;
344   const MachO::dyld_info_command &DyLdInfoCommand =
345       O.LoadCommands[*O.DyLdInfoCommandIndex]
346           .MachOLoadCommand.dyld_info_command_data;
347   char *Out = (char *)Buf->getBufferStart() + DyLdInfoCommand.weak_bind_off;
348   assert((DyLdInfoCommand.weak_bind_size == O.WeakBinds.Opcodes.size()) &&
349          "Incorrect weak bind opcodes size");
350   memcpy(Out, O.WeakBinds.Opcodes.data(), O.WeakBinds.Opcodes.size());
351 }
352 
writeLazyBindInfo()353 void MachOWriter::writeLazyBindInfo() {
354   if (!O.DyLdInfoCommandIndex)
355     return;
356   const MachO::dyld_info_command &DyLdInfoCommand =
357       O.LoadCommands[*O.DyLdInfoCommandIndex]
358           .MachOLoadCommand.dyld_info_command_data;
359   char *Out = (char *)Buf->getBufferStart() + DyLdInfoCommand.lazy_bind_off;
360   assert((DyLdInfoCommand.lazy_bind_size == O.LazyBinds.Opcodes.size()) &&
361          "Incorrect lazy bind opcodes size");
362   memcpy(Out, O.LazyBinds.Opcodes.data(), O.LazyBinds.Opcodes.size());
363 }
364 
writeExportInfo()365 void MachOWriter::writeExportInfo() {
366   if (!O.DyLdInfoCommandIndex)
367     return;
368   const MachO::dyld_info_command &DyLdInfoCommand =
369       O.LoadCommands[*O.DyLdInfoCommandIndex]
370           .MachOLoadCommand.dyld_info_command_data;
371   char *Out = (char *)Buf->getBufferStart() + DyLdInfoCommand.export_off;
372   assert((DyLdInfoCommand.export_size == O.Exports.Trie.size()) &&
373          "Incorrect export trie size");
374   memcpy(Out, O.Exports.Trie.data(), O.Exports.Trie.size());
375 }
376 
writeIndirectSymbolTable()377 void MachOWriter::writeIndirectSymbolTable() {
378   if (!O.DySymTabCommandIndex)
379     return;
380 
381   const MachO::dysymtab_command &DySymTabCommand =
382       O.LoadCommands[*O.DySymTabCommandIndex]
383           .MachOLoadCommand.dysymtab_command_data;
384 
385   uint32_t *Out =
386       (uint32_t *)(Buf->getBufferStart() + DySymTabCommand.indirectsymoff);
387   for (const IndirectSymbolEntry &Sym : O.IndirectSymTable.Symbols) {
388     uint32_t Entry = (Sym.Symbol) ? (*Sym.Symbol)->Index : Sym.OriginalIndex;
389     if (IsLittleEndian != sys::IsLittleEndianHost)
390       sys::swapByteOrder(Entry);
391     *Out++ = Entry;
392   }
393 }
394 
writeLinkData(std::optional<size_t> LCIndex,const LinkData & LD)395 void MachOWriter::writeLinkData(std::optional<size_t> LCIndex,
396                                 const LinkData &LD) {
397   if (!LCIndex)
398     return;
399   const MachO::linkedit_data_command &LinkEditDataCommand =
400       O.LoadCommands[*LCIndex].MachOLoadCommand.linkedit_data_command_data;
401   char *Out = (char *)Buf->getBufferStart() + LinkEditDataCommand.dataoff;
402   assert((LinkEditDataCommand.datasize == LD.Data.size()) &&
403          "Incorrect data size");
404   memcpy(Out, LD.Data.data(), LD.Data.size());
405 }
406 
407 static uint64_t
getSegmentFileOffset(const LoadCommand & TextSegmentLoadCommand)408 getSegmentFileOffset(const LoadCommand &TextSegmentLoadCommand) {
409   const MachO::macho_load_command &MLC =
410       TextSegmentLoadCommand.MachOLoadCommand;
411   switch (MLC.load_command_data.cmd) {
412   case MachO::LC_SEGMENT:
413     return MLC.segment_command_data.fileoff;
414   case MachO::LC_SEGMENT_64:
415     return MLC.segment_command_64_data.fileoff;
416   default:
417     return 0;
418   }
419 }
420 
getSegmentFileSize(const LoadCommand & TextSegmentLoadCommand)421 static uint64_t getSegmentFileSize(const LoadCommand &TextSegmentLoadCommand) {
422   const MachO::macho_load_command &MLC =
423       TextSegmentLoadCommand.MachOLoadCommand;
424   switch (MLC.load_command_data.cmd) {
425   case MachO::LC_SEGMENT:
426     return MLC.segment_command_data.filesize;
427   case MachO::LC_SEGMENT_64:
428     return MLC.segment_command_64_data.filesize;
429   default:
430     return 0;
431   }
432 }
433 
writeCodeSignatureData()434 void MachOWriter::writeCodeSignatureData() {
435   // NOTE: This CodeSignature section behaviour must be kept in sync with that
436   // performed in LLD's CodeSignatureSection::write /
437   // CodeSignatureSection::writeHashes. Furthermore, this call must occur only
438   // after the rest of the binary has already been written to the buffer. This
439   // is because the buffer is read from to perform the necessary hashing.
440 
441   // The CodeSignature section is the last section in the MachO binary and
442   // contains a hash of all content in the binary before it. Since llvm-objcopy
443   // has likely modified the target binary, the hash must be regenerated
444   // entirely. To generate this hash, we must read from the start of the binary
445   // (HashReadStart) to just before the start of the CodeSignature section
446   // (HashReadEnd).
447 
448   const CodeSignatureInfo &CodeSignature = LayoutBuilder.getCodeSignature();
449 
450   uint8_t *BufferStart = reinterpret_cast<uint8_t *>(Buf->getBufferStart());
451   uint8_t *HashReadStart = BufferStart;
452   uint8_t *HashReadEnd = BufferStart + CodeSignature.StartOffset;
453 
454   // The CodeSignature section begins with a header, after which the hashes
455   // of each page of the binary are written.
456   uint8_t *HashWriteStart = HashReadEnd + CodeSignature.AllHeadersSize;
457 
458   uint32_t TextSegmentFileOff = 0;
459   uint32_t TextSegmentFileSize = 0;
460   if (O.TextSegmentCommandIndex) {
461     const LoadCommand &TextSegmentLoadCommand =
462         O.LoadCommands[*O.TextSegmentCommandIndex];
463     assert(TextSegmentLoadCommand.MachOLoadCommand.load_command_data.cmd ==
464                MachO::LC_SEGMENT ||
465            TextSegmentLoadCommand.MachOLoadCommand.load_command_data.cmd ==
466                MachO::LC_SEGMENT_64);
467     assert(StringRef(TextSegmentLoadCommand.MachOLoadCommand
468                          .segment_command_data.segname) == "__TEXT");
469     TextSegmentFileOff = getSegmentFileOffset(TextSegmentLoadCommand);
470     TextSegmentFileSize = getSegmentFileSize(TextSegmentLoadCommand);
471   }
472 
473   const uint32_t FileNamePad = CodeSignature.AllHeadersSize -
474                                CodeSignature.FixedHeadersSize -
475                                CodeSignature.OutputFileName.size();
476 
477   // Write code section header.
478   auto *SuperBlob = reinterpret_cast<MachO::CS_SuperBlob *>(HashReadEnd);
479   write32be(&SuperBlob->magic, MachO::CSMAGIC_EMBEDDED_SIGNATURE);
480   write32be(&SuperBlob->length, CodeSignature.Size);
481   write32be(&SuperBlob->count, 1);
482   auto *BlobIndex = reinterpret_cast<MachO::CS_BlobIndex *>(&SuperBlob[1]);
483   write32be(&BlobIndex->type, MachO::CSSLOT_CODEDIRECTORY);
484   write32be(&BlobIndex->offset, CodeSignature.BlobHeadersSize);
485   auto *CodeDirectory = reinterpret_cast<MachO::CS_CodeDirectory *>(
486       HashReadEnd + CodeSignature.BlobHeadersSize);
487   write32be(&CodeDirectory->magic, MachO::CSMAGIC_CODEDIRECTORY);
488   write32be(&CodeDirectory->length,
489             CodeSignature.Size - CodeSignature.BlobHeadersSize);
490   write32be(&CodeDirectory->version, MachO::CS_SUPPORTSEXECSEG);
491   write32be(&CodeDirectory->flags, MachO::CS_ADHOC | MachO::CS_LINKER_SIGNED);
492   write32be(&CodeDirectory->hashOffset,
493             sizeof(MachO::CS_CodeDirectory) +
494                 CodeSignature.OutputFileName.size() + FileNamePad);
495   write32be(&CodeDirectory->identOffset, sizeof(MachO::CS_CodeDirectory));
496   CodeDirectory->nSpecialSlots = 0;
497   write32be(&CodeDirectory->nCodeSlots, CodeSignature.BlockCount);
498   write32be(&CodeDirectory->codeLimit, CodeSignature.StartOffset);
499   CodeDirectory->hashSize = static_cast<uint8_t>(CodeSignature.HashSize);
500   CodeDirectory->hashType = MachO::kSecCodeSignatureHashSHA256;
501   CodeDirectory->platform = 0;
502   CodeDirectory->pageSize = CodeSignature.BlockSizeShift;
503   CodeDirectory->spare2 = 0;
504   CodeDirectory->scatterOffset = 0;
505   CodeDirectory->teamOffset = 0;
506   CodeDirectory->spare3 = 0;
507   CodeDirectory->codeLimit64 = 0;
508   write64be(&CodeDirectory->execSegBase, TextSegmentFileOff);
509   write64be(&CodeDirectory->execSegLimit, TextSegmentFileSize);
510   write64be(&CodeDirectory->execSegFlags, O.Header.FileType == MachO::MH_EXECUTE
511                                               ? MachO::CS_EXECSEG_MAIN_BINARY
512                                               : 0);
513 
514   auto *Id = reinterpret_cast<char *>(&CodeDirectory[1]);
515   memcpy(Id, CodeSignature.OutputFileName.begin(),
516          CodeSignature.OutputFileName.size());
517   memset(Id + CodeSignature.OutputFileName.size(), 0, FileNamePad);
518 
519   // Write the hashes.
520   uint8_t *CurrHashReadPosition = HashReadStart;
521   uint8_t *CurrHashWritePosition = HashWriteStart;
522   while (CurrHashReadPosition < HashReadEnd) {
523     StringRef Block(reinterpret_cast<char *>(CurrHashReadPosition),
524                     std::min(static_cast<size_t>(HashReadEnd
525                              - CurrHashReadPosition),
526                              static_cast<size_t>(CodeSignature.BlockSize)));
527     SHA256 Hasher;
528     Hasher.update(Block);
529     std::array<uint8_t, 32> Hash = Hasher.final();
530     assert(Hash.size() == CodeSignature.HashSize);
531     memcpy(CurrHashWritePosition, Hash.data(), CodeSignature.HashSize);
532     CurrHashReadPosition += CodeSignature.BlockSize;
533     CurrHashWritePosition += CodeSignature.HashSize;
534   }
535 #if defined(__APPLE__)
536   // This is macOS-specific work-around and makes no sense for any
537   // other host OS. See https://openradar.appspot.com/FB8914231
538   //
539   // The macOS kernel maintains a signature-verification cache to
540   // quickly validate applications at time of execve(2).  The trouble
541   // is that for the kernel creates the cache entry at the time of the
542   // mmap(2) call, before we have a chance to write either the code to
543   // sign or the signature header+hashes.  The fix is to invalidate
544   // all cached data associated with the output file, thus discarding
545   // the bogus prematurely-cached signature.
546   msync(BufferStart, CodeSignature.StartOffset + CodeSignature.Size,
547         MS_INVALIDATE);
548 #endif
549 }
550 
writeDataInCodeData()551 void MachOWriter::writeDataInCodeData() {
552   return writeLinkData(O.DataInCodeCommandIndex, O.DataInCode);
553 }
554 
writeLinkerOptimizationHint()555 void MachOWriter::writeLinkerOptimizationHint() {
556   return writeLinkData(O.LinkerOptimizationHintCommandIndex,
557                        O.LinkerOptimizationHint);
558 }
559 
writeFunctionStartsData()560 void MachOWriter::writeFunctionStartsData() {
561   return writeLinkData(O.FunctionStartsCommandIndex, O.FunctionStarts);
562 }
563 
writeDylibCodeSignDRsData()564 void MachOWriter::writeDylibCodeSignDRsData() {
565   return writeLinkData(O.DylibCodeSignDRsIndex, O.DylibCodeSignDRs);
566 }
567 
writeChainedFixupsData()568 void MachOWriter::writeChainedFixupsData() {
569   return writeLinkData(O.ChainedFixupsCommandIndex, O.ChainedFixups);
570 }
571 
writeExportsTrieData()572 void MachOWriter::writeExportsTrieData() {
573   if (!O.ExportsTrieCommandIndex)
574     return;
575   const MachO::linkedit_data_command &ExportsTrieCmd =
576       O.LoadCommands[*O.ExportsTrieCommandIndex]
577           .MachOLoadCommand.linkedit_data_command_data;
578   char *Out = (char *)Buf->getBufferStart() + ExportsTrieCmd.dataoff;
579   assert((ExportsTrieCmd.datasize == O.Exports.Trie.size()) &&
580          "Incorrect export trie size");
581   memcpy(Out, O.Exports.Trie.data(), O.Exports.Trie.size());
582 }
583 
writeTail()584 void MachOWriter::writeTail() {
585   typedef void (MachOWriter::*WriteHandlerType)();
586   typedef std::pair<uint64_t, WriteHandlerType> WriteOperation;
587   SmallVector<WriteOperation, 7> Queue;
588 
589   if (O.SymTabCommandIndex) {
590     const MachO::symtab_command &SymTabCommand =
591         O.LoadCommands[*O.SymTabCommandIndex]
592             .MachOLoadCommand.symtab_command_data;
593     if (SymTabCommand.symoff)
594       Queue.push_back({SymTabCommand.symoff, &MachOWriter::writeSymbolTable});
595     if (SymTabCommand.stroff)
596       Queue.push_back({SymTabCommand.stroff, &MachOWriter::writeStringTable});
597   }
598 
599   if (O.DyLdInfoCommandIndex) {
600     const MachO::dyld_info_command &DyLdInfoCommand =
601         O.LoadCommands[*O.DyLdInfoCommandIndex]
602             .MachOLoadCommand.dyld_info_command_data;
603     if (DyLdInfoCommand.rebase_off)
604       Queue.push_back(
605           {DyLdInfoCommand.rebase_off, &MachOWriter::writeRebaseInfo});
606     if (DyLdInfoCommand.bind_off)
607       Queue.push_back({DyLdInfoCommand.bind_off, &MachOWriter::writeBindInfo});
608     if (DyLdInfoCommand.weak_bind_off)
609       Queue.push_back(
610           {DyLdInfoCommand.weak_bind_off, &MachOWriter::writeWeakBindInfo});
611     if (DyLdInfoCommand.lazy_bind_off)
612       Queue.push_back(
613           {DyLdInfoCommand.lazy_bind_off, &MachOWriter::writeLazyBindInfo});
614     if (DyLdInfoCommand.export_off)
615       Queue.push_back(
616           {DyLdInfoCommand.export_off, &MachOWriter::writeExportInfo});
617   }
618 
619   if (O.DySymTabCommandIndex) {
620     const MachO::dysymtab_command &DySymTabCommand =
621         O.LoadCommands[*O.DySymTabCommandIndex]
622             .MachOLoadCommand.dysymtab_command_data;
623 
624     if (DySymTabCommand.indirectsymoff)
625       Queue.emplace_back(DySymTabCommand.indirectsymoff,
626                          &MachOWriter::writeIndirectSymbolTable);
627   }
628 
629   std::initializer_list<std::pair<std::optional<size_t>, WriteHandlerType>>
630       LinkEditDataCommandWriters = {
631           {O.CodeSignatureCommandIndex, &MachOWriter::writeCodeSignatureData},
632           {O.DylibCodeSignDRsIndex, &MachOWriter::writeDylibCodeSignDRsData},
633           {O.DataInCodeCommandIndex, &MachOWriter::writeDataInCodeData},
634           {O.LinkerOptimizationHintCommandIndex,
635            &MachOWriter::writeLinkerOptimizationHint},
636           {O.FunctionStartsCommandIndex, &MachOWriter::writeFunctionStartsData},
637           {O.ChainedFixupsCommandIndex, &MachOWriter::writeChainedFixupsData},
638           {O.ExportsTrieCommandIndex, &MachOWriter::writeExportsTrieData}};
639   for (const auto &W : LinkEditDataCommandWriters) {
640     std::optional<size_t> LinkEditDataCommandIndex;
641     WriteHandlerType WriteHandler;
642     std::tie(LinkEditDataCommandIndex, WriteHandler) = W;
643     if (LinkEditDataCommandIndex) {
644       const MachO::linkedit_data_command &LinkEditDataCommand =
645           O.LoadCommands[*LinkEditDataCommandIndex]
646               .MachOLoadCommand.linkedit_data_command_data;
647       if (LinkEditDataCommand.dataoff)
648         Queue.emplace_back(LinkEditDataCommand.dataoff, WriteHandler);
649     }
650   }
651 
652   llvm::sort(Queue, llvm::less_first());
653 
654   for (auto WriteOp : Queue)
655     (this->*WriteOp.second)();
656 }
657 
finalize()658 Error MachOWriter::finalize() { return LayoutBuilder.layout(); }
659 
write()660 Error MachOWriter::write() {
661   size_t TotalSize = totalSize();
662   Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
663   if (!Buf)
664     return createStringError(errc::not_enough_memory,
665                              "failed to allocate memory buffer of " +
666                                  Twine::utohexstr(TotalSize) + " bytes");
667   writeHeader();
668   writeLoadCommands();
669   writeSections();
670   writeTail();
671 
672   // TODO: Implement direct writing to the output stream (without intermediate
673   // memory buffer Buf).
674   Out.write(Buf->getBufferStart(), Buf->getBufferSize());
675   return Error::success();
676 }
677