1 //===- MachOReader.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 "MachOReader.h"
10 #include "MachOObject.h"
11 #include "llvm/BinaryFormat/MachO.h"
12 #include "llvm/Object/MachO.h"
13 #include "llvm/Support/Errc.h"
14 #include <memory>
15 
16 using namespace llvm;
17 using namespace llvm::objcopy;
18 using namespace llvm::objcopy::macho;
19 
20 void MachOReader::readHeader(Object &O) const {
21   O.Header.Magic = MachOObj.getHeader().magic;
22   O.Header.CPUType = MachOObj.getHeader().cputype;
23   O.Header.CPUSubType = MachOObj.getHeader().cpusubtype;
24   O.Header.FileType = MachOObj.getHeader().filetype;
25   O.Header.NCmds = MachOObj.getHeader().ncmds;
26   O.Header.SizeOfCmds = MachOObj.getHeader().sizeofcmds;
27   O.Header.Flags = MachOObj.getHeader().flags;
28 }
29 
30 template <typename SectionType>
31 static Section constructSectionCommon(const SectionType &Sec, uint32_t Index) {
32   StringRef SegName(Sec.segname, strnlen(Sec.segname, sizeof(Sec.segname)));
33   StringRef SectName(Sec.sectname, strnlen(Sec.sectname, sizeof(Sec.sectname)));
34   Section S(SegName, SectName);
35   S.Index = Index;
36   S.Addr = Sec.addr;
37   S.Size = Sec.size;
38   S.OriginalOffset = Sec.offset;
39   S.Align = Sec.align;
40   S.RelOff = Sec.reloff;
41   S.NReloc = Sec.nreloc;
42   S.Flags = Sec.flags;
43   S.Reserved1 = Sec.reserved1;
44   S.Reserved2 = Sec.reserved2;
45   S.Reserved3 = 0;
46   return S;
47 }
48 
49 Section constructSection(const MachO::section &Sec, uint32_t Index) {
50   return constructSectionCommon(Sec, Index);
51 }
52 
53 Section constructSection(const MachO::section_64 &Sec, uint32_t Index) {
54   Section S = constructSectionCommon(Sec, Index);
55   S.Reserved3 = Sec.reserved3;
56   return S;
57 }
58 
59 template <typename SectionType, typename SegmentType>
60 Expected<std::vector<std::unique_ptr<Section>>> static extractSections(
61     const object::MachOObjectFile::LoadCommandInfo &LoadCmd,
62     const object::MachOObjectFile &MachOObj, uint32_t &NextSectionIndex) {
63   std::vector<std::unique_ptr<Section>> Sections;
64   for (auto Curr = reinterpret_cast<const SectionType *>(LoadCmd.Ptr +
65                                                          sizeof(SegmentType)),
66             End = reinterpret_cast<const SectionType *>(LoadCmd.Ptr +
67                                                         LoadCmd.C.cmdsize);
68        Curr < End; ++Curr) {
69     SectionType Sec;
70     memcpy((void *)&Sec, Curr, sizeof(SectionType));
71 
72     if (MachOObj.isLittleEndian() != sys::IsLittleEndianHost)
73       MachO::swapStruct(Sec);
74 
75     Sections.push_back(
76         std::make_unique<Section>(constructSection(Sec, NextSectionIndex)));
77 
78     Section &S = *Sections.back();
79 
80     Expected<object::SectionRef> SecRef =
81         MachOObj.getSection(NextSectionIndex++);
82     if (!SecRef)
83       return SecRef.takeError();
84 
85     Expected<ArrayRef<uint8_t>> Data =
86         MachOObj.getSectionContents(SecRef->getRawDataRefImpl());
87     if (!Data)
88       return Data.takeError();
89 
90     S.Content =
91         StringRef(reinterpret_cast<const char *>(Data->data()), Data->size());
92 
93     const uint32_t CPUType = MachOObj.getHeader().cputype;
94     S.Relocations.reserve(S.NReloc);
95     for (auto RI = MachOObj.section_rel_begin(SecRef->getRawDataRefImpl()),
96               RE = MachOObj.section_rel_end(SecRef->getRawDataRefImpl());
97          RI != RE; ++RI) {
98       RelocationInfo R;
99       R.Symbol = nullptr; // We'll fill this field later.
100       R.Info = MachOObj.getRelocation(RI->getRawDataRefImpl());
101       R.Scattered = MachOObj.isRelocationScattered(R.Info);
102       unsigned Type = MachOObj.getAnyRelocationType(R.Info);
103       // TODO Support CPU_TYPE_ARM.
104       R.IsAddend = !R.Scattered && (CPUType == MachO::CPU_TYPE_ARM64 &&
105                                     Type == MachO::ARM64_RELOC_ADDEND);
106       R.Extern = !R.Scattered && MachOObj.getPlainRelocationExternal(R.Info);
107       S.Relocations.push_back(R);
108     }
109 
110     assert(S.NReloc == S.Relocations.size() &&
111            "Incorrect number of relocations");
112   }
113   return std::move(Sections);
114 }
115 
116 Error MachOReader::readLoadCommands(Object &O) const {
117   // For MachO sections indices start from 1.
118   uint32_t NextSectionIndex = 1;
119   static constexpr char TextSegmentName[] = "__TEXT";
120   for (auto LoadCmd : MachOObj.load_commands()) {
121     LoadCommand LC;
122     switch (LoadCmd.C.cmd) {
123     case MachO::LC_CODE_SIGNATURE:
124       O.CodeSignatureCommandIndex = O.LoadCommands.size();
125       break;
126     case MachO::LC_SEGMENT:
127       // LoadCmd.Ptr might not be aligned temporarily as
128       // MachO::segment_command requires, but the segname char pointer do not
129       // have alignment restrictions.
130       if (StringRef(reinterpret_cast<const char *>(
131               LoadCmd.Ptr + offsetof(MachO::segment_command, segname))) ==
132           TextSegmentName)
133         O.TextSegmentCommandIndex = O.LoadCommands.size();
134 
135       if (Expected<std::vector<std::unique_ptr<Section>>> Sections =
136               extractSections<MachO::section, MachO::segment_command>(
137                   LoadCmd, MachOObj, NextSectionIndex))
138         LC.Sections = std::move(*Sections);
139       else
140         return Sections.takeError();
141       break;
142     case MachO::LC_SEGMENT_64:
143       // LoadCmd.Ptr might not be aligned temporarily as
144       // MachO::segment_command_64 requires, but the segname char pointer do
145       // not have alignment restrictions.
146       if (StringRef(reinterpret_cast<const char *>(
147               LoadCmd.Ptr + offsetof(MachO::segment_command_64, segname))) ==
148           TextSegmentName)
149         O.TextSegmentCommandIndex = O.LoadCommands.size();
150 
151       if (Expected<std::vector<std::unique_ptr<Section>>> Sections =
152               extractSections<MachO::section_64, MachO::segment_command_64>(
153                   LoadCmd, MachOObj, NextSectionIndex))
154         LC.Sections = std::move(*Sections);
155       else
156         return Sections.takeError();
157       break;
158     case MachO::LC_SYMTAB:
159       O.SymTabCommandIndex = O.LoadCommands.size();
160       break;
161     case MachO::LC_DYSYMTAB:
162       O.DySymTabCommandIndex = O.LoadCommands.size();
163       break;
164     case MachO::LC_DYLD_INFO:
165     case MachO::LC_DYLD_INFO_ONLY:
166       O.DyLdInfoCommandIndex = O.LoadCommands.size();
167       break;
168     case MachO::LC_DATA_IN_CODE:
169       O.DataInCodeCommandIndex = O.LoadCommands.size();
170       break;
171     case MachO::LC_LINKER_OPTIMIZATION_HINT:
172       O.LinkerOptimizationHintCommandIndex = O.LoadCommands.size();
173       break;
174     case MachO::LC_FUNCTION_STARTS:
175       O.FunctionStartsCommandIndex = O.LoadCommands.size();
176       break;
177     case MachO::LC_DYLIB_CODE_SIGN_DRS:
178       O.DylibCodeSignDRsIndex = O.LoadCommands.size();
179       break;
180     case MachO::LC_DYLD_EXPORTS_TRIE:
181       O.ExportsTrieCommandIndex = O.LoadCommands.size();
182       break;
183     case MachO::LC_DYLD_CHAINED_FIXUPS:
184       O.ChainedFixupsCommandIndex = O.LoadCommands.size();
185       break;
186     }
187 #define HANDLE_LOAD_COMMAND(LCName, LCValue, LCStruct)                         \
188   case MachO::LCName:                                                          \
189     memcpy((void *)&(LC.MachOLoadCommand.LCStruct##_data), LoadCmd.Ptr,        \
190            sizeof(MachO::LCStruct));                                           \
191     if (MachOObj.isLittleEndian() != sys::IsLittleEndianHost)                  \
192       MachO::swapStruct(LC.MachOLoadCommand.LCStruct##_data);                  \
193     if (LoadCmd.C.cmdsize > sizeof(MachO::LCStruct))                           \
194       LC.Payload = ArrayRef<uint8_t>(                                          \
195           reinterpret_cast<uint8_t *>(const_cast<char *>(LoadCmd.Ptr)) +       \
196               sizeof(MachO::LCStruct),                                         \
197           LoadCmd.C.cmdsize - sizeof(MachO::LCStruct));                        \
198     break;
199 
200     switch (LoadCmd.C.cmd) {
201     default:
202       memcpy((void *)&(LC.MachOLoadCommand.load_command_data), LoadCmd.Ptr,
203              sizeof(MachO::load_command));
204       if (MachOObj.isLittleEndian() != sys::IsLittleEndianHost)
205         MachO::swapStruct(LC.MachOLoadCommand.load_command_data);
206       if (LoadCmd.C.cmdsize > sizeof(MachO::load_command))
207         LC.Payload = ArrayRef<uint8_t>(
208             reinterpret_cast<uint8_t *>(const_cast<char *>(LoadCmd.Ptr)) +
209                 sizeof(MachO::load_command),
210             LoadCmd.C.cmdsize - sizeof(MachO::load_command));
211       break;
212 #include "llvm/BinaryFormat/MachO.def"
213     }
214     O.LoadCommands.push_back(std::move(LC));
215   }
216   return Error::success();
217 }
218 
219 template <typename nlist_t>
220 SymbolEntry constructSymbolEntry(StringRef StrTable, const nlist_t &nlist) {
221   assert(nlist.n_strx < StrTable.size() &&
222          "n_strx exceeds the size of the string table");
223   SymbolEntry SE;
224   SE.Name = StringRef(StrTable.data() + nlist.n_strx).str();
225   SE.n_type = nlist.n_type;
226   SE.n_sect = nlist.n_sect;
227   SE.n_desc = nlist.n_desc;
228   SE.n_value = nlist.n_value;
229   return SE;
230 }
231 
232 void MachOReader::readSymbolTable(Object &O) const {
233   StringRef StrTable = MachOObj.getStringTableData();
234   for (auto Symbol : MachOObj.symbols()) {
235     SymbolEntry SE =
236         (MachOObj.is64Bit()
237              ? constructSymbolEntry(StrTable, MachOObj.getSymbol64TableEntry(
238                                                   Symbol.getRawDataRefImpl()))
239              : constructSymbolEntry(StrTable, MachOObj.getSymbolTableEntry(
240                                                   Symbol.getRawDataRefImpl())));
241 
242     O.SymTable.Symbols.push_back(std::make_unique<SymbolEntry>(SE));
243   }
244 }
245 
246 void MachOReader::setSymbolInRelocationInfo(Object &O) const {
247   std::vector<const Section *> Sections;
248   for (auto &LC : O.LoadCommands)
249     for (std::unique_ptr<Section> &Sec : LC.Sections)
250       Sections.push_back(Sec.get());
251 
252   for (LoadCommand &LC : O.LoadCommands)
253     for (std::unique_ptr<Section> &Sec : LC.Sections)
254       for (auto &Reloc : Sec->Relocations)
255         if (!Reloc.Scattered && !Reloc.IsAddend) {
256           const uint32_t SymbolNum =
257               Reloc.getPlainRelocationSymbolNum(MachOObj.isLittleEndian());
258           if (Reloc.Extern) {
259             Reloc.Symbol = O.SymTable.getSymbolByIndex(SymbolNum);
260           } else {
261             // FIXME: Refactor error handling in MachOReader and report an error
262             // if we encounter an invalid relocation.
263             assert(SymbolNum >= 1 && SymbolNum <= Sections.size() &&
264                    "Invalid section index.");
265             Reloc.Sec = Sections[SymbolNum - 1];
266           }
267         }
268 }
269 
270 void MachOReader::readRebaseInfo(Object &O) const {
271   O.Rebases.Opcodes = MachOObj.getDyldInfoRebaseOpcodes();
272 }
273 
274 void MachOReader::readBindInfo(Object &O) const {
275   O.Binds.Opcodes = MachOObj.getDyldInfoBindOpcodes();
276 }
277 
278 void MachOReader::readWeakBindInfo(Object &O) const {
279   O.WeakBinds.Opcodes = MachOObj.getDyldInfoWeakBindOpcodes();
280 }
281 
282 void MachOReader::readLazyBindInfo(Object &O) const {
283   O.LazyBinds.Opcodes = MachOObj.getDyldInfoLazyBindOpcodes();
284 }
285 
286 void MachOReader::readExportInfo(Object &O) const {
287   // This information can be in LC_DYLD_INFO or in LC_DYLD_EXPORTS_TRIE
288   ArrayRef<uint8_t> Trie = MachOObj.getDyldInfoExportsTrie();
289   if (Trie.empty())
290     Trie = MachOObj.getDyldExportsTrie();
291   O.Exports.Trie = Trie;
292 }
293 
294 void MachOReader::readLinkData(Object &O, std::optional<size_t> LCIndex,
295                                LinkData &LD) const {
296   if (!LCIndex)
297     return;
298   const MachO::linkedit_data_command &LC =
299       O.LoadCommands[*LCIndex].MachOLoadCommand.linkedit_data_command_data;
300   LD.Data =
301       arrayRefFromStringRef(MachOObj.getData().substr(LC.dataoff, LC.datasize));
302 }
303 
304 void MachOReader::readDataInCodeData(Object &O) const {
305   return readLinkData(O, O.DataInCodeCommandIndex, O.DataInCode);
306 }
307 
308 void MachOReader::readLinkerOptimizationHint(Object &O) const {
309   return readLinkData(O, O.LinkerOptimizationHintCommandIndex,
310                       O.LinkerOptimizationHint);
311 }
312 
313 void MachOReader::readFunctionStartsData(Object &O) const {
314   return readLinkData(O, O.FunctionStartsCommandIndex, O.FunctionStarts);
315 }
316 
317 void MachOReader::readDylibCodeSignDRs(Object &O) const {
318   return readLinkData(O, O.DylibCodeSignDRsIndex, O.DylibCodeSignDRs);
319 }
320 
321 void MachOReader::readExportsTrie(Object &O) const {
322   return readLinkData(O, O.ExportsTrieCommandIndex, O.ExportsTrie);
323 }
324 
325 void MachOReader::readChainedFixups(Object &O) const {
326   return readLinkData(O, O.ChainedFixupsCommandIndex, O.ChainedFixups);
327 }
328 
329 void MachOReader::readIndirectSymbolTable(Object &O) const {
330   MachO::dysymtab_command DySymTab = MachOObj.getDysymtabLoadCommand();
331   constexpr uint32_t AbsOrLocalMask =
332       MachO::INDIRECT_SYMBOL_LOCAL | MachO::INDIRECT_SYMBOL_ABS;
333   for (uint32_t i = 0; i < DySymTab.nindirectsyms; ++i) {
334     uint32_t Index = MachOObj.getIndirectSymbolTableEntry(DySymTab, i);
335     if ((Index & AbsOrLocalMask) != 0)
336       O.IndirectSymTable.Symbols.emplace_back(Index, std::nullopt);
337     else
338       O.IndirectSymTable.Symbols.emplace_back(
339           Index, O.SymTable.getSymbolByIndex(Index));
340   }
341 }
342 
343 void MachOReader::readSwiftVersion(Object &O) const {
344   struct ObjCImageInfo {
345     uint32_t Version;
346     uint32_t Flags;
347   } ImageInfo;
348 
349   for (const LoadCommand &LC : O.LoadCommands)
350     for (const std::unique_ptr<Section> &Sec : LC.Sections)
351       if (Sec->Sectname == "__objc_imageinfo" &&
352           (Sec->Segname == "__DATA" || Sec->Segname == "__DATA_CONST" ||
353            Sec->Segname == "__DATA_DIRTY") &&
354           Sec->Content.size() >= sizeof(ObjCImageInfo)) {
355         memcpy(&ImageInfo, Sec->Content.data(), sizeof(ObjCImageInfo));
356         if (MachOObj.isLittleEndian() != sys::IsLittleEndianHost) {
357           sys::swapByteOrder(ImageInfo.Version);
358           sys::swapByteOrder(ImageInfo.Flags);
359         }
360         O.SwiftVersion = (ImageInfo.Flags >> 8) & 0xff;
361         return;
362       }
363 }
364 
365 Expected<std::unique_ptr<Object>> MachOReader::create() const {
366   auto Obj = std::make_unique<Object>();
367   readHeader(*Obj);
368   if (Error E = readLoadCommands(*Obj))
369     return std::move(E);
370   readSymbolTable(*Obj);
371   setSymbolInRelocationInfo(*Obj);
372   readRebaseInfo(*Obj);
373   readBindInfo(*Obj);
374   readWeakBindInfo(*Obj);
375   readLazyBindInfo(*Obj);
376   readExportInfo(*Obj);
377   readDataInCodeData(*Obj);
378   readLinkerOptimizationHint(*Obj);
379   readFunctionStartsData(*Obj);
380   readDylibCodeSignDRs(*Obj);
381   readExportsTrie(*Obj);
382   readChainedFixups(*Obj);
383   readIndirectSymbolTable(*Obj);
384   readSwiftVersion(*Obj);
385   return std::move(Obj);
386 }
387