1 //===- MachOObjcopy.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 "MachOObjcopy.h"
10 #include "../llvm-objcopy.h"
11 #include "CommonConfig.h"
12 #include "MachOReader.h"
13 #include "MachOWriter.h"
14 #include "MultiFormatConfig.h"
15 #include "llvm/ADT/DenseSet.h"
16 #include "llvm/Object/ArchiveWriter.h"
17 #include "llvm/Object/MachOUniversal.h"
18 #include "llvm/Object/MachOUniversalWriter.h"
19 #include "llvm/Support/Errc.h"
20 #include "llvm/Support/Error.h"
21 #include "llvm/Support/FileOutputBuffer.h"
22 #include "llvm/Support/SmallVectorMemoryBuffer.h"
23
24 namespace llvm {
25 namespace objcopy {
26 namespace macho {
27
28 using namespace object;
29 using SectionPred = std::function<bool(const std::unique_ptr<Section> &Sec)>;
30 using LoadCommandPred = std::function<bool(const LoadCommand &LC)>;
31
32 #ifndef NDEBUG
isLoadCommandWithPayloadString(const LoadCommand & LC)33 static bool isLoadCommandWithPayloadString(const LoadCommand &LC) {
34 // TODO: Add support for LC_REEXPORT_DYLIB, LC_LOAD_UPWARD_DYLIB and
35 // LC_LAZY_LOAD_DYLIB
36 return LC.MachOLoadCommand.load_command_data.cmd == MachO::LC_RPATH ||
37 LC.MachOLoadCommand.load_command_data.cmd == MachO::LC_ID_DYLIB ||
38 LC.MachOLoadCommand.load_command_data.cmd == MachO::LC_LOAD_DYLIB ||
39 LC.MachOLoadCommand.load_command_data.cmd == MachO::LC_LOAD_WEAK_DYLIB;
40 }
41 #endif
42
getPayloadString(const LoadCommand & LC)43 static StringRef getPayloadString(const LoadCommand &LC) {
44 assert(isLoadCommandWithPayloadString(LC) &&
45 "unsupported load command encountered");
46
47 return StringRef(reinterpret_cast<const char *>(LC.Payload.data()),
48 LC.Payload.size())
49 .rtrim('\0');
50 }
51
removeSections(const CommonConfig & Config,Object & Obj)52 static Error removeSections(const CommonConfig &Config, Object &Obj) {
53 SectionPred RemovePred = [](const std::unique_ptr<Section> &) {
54 return false;
55 };
56
57 if (!Config.ToRemove.empty()) {
58 RemovePred = [&Config, RemovePred](const std::unique_ptr<Section> &Sec) {
59 return Config.ToRemove.matches(Sec->CanonicalName);
60 };
61 }
62
63 if (Config.StripAll || Config.StripDebug) {
64 // Remove all debug sections.
65 RemovePred = [RemovePred](const std::unique_ptr<Section> &Sec) {
66 if (Sec->Segname == "__DWARF")
67 return true;
68
69 return RemovePred(Sec);
70 };
71 }
72
73 if (!Config.OnlySection.empty()) {
74 // Overwrite RemovePred because --only-section takes priority.
75 RemovePred = [&Config](const std::unique_ptr<Section> &Sec) {
76 return !Config.OnlySection.matches(Sec->CanonicalName);
77 };
78 }
79
80 return Obj.removeSections(RemovePred);
81 }
82
markSymbols(const CommonConfig &,Object & Obj)83 static void markSymbols(const CommonConfig &, Object &Obj) {
84 // Symbols referenced from the indirect symbol table must not be removed.
85 for (IndirectSymbolEntry &ISE : Obj.IndirectSymTable.Symbols)
86 if (ISE.Symbol)
87 (*ISE.Symbol)->Referenced = true;
88 }
89
updateAndRemoveSymbols(const CommonConfig & Config,Object & Obj)90 static void updateAndRemoveSymbols(const CommonConfig &Config, Object &Obj) {
91 for (SymbolEntry &Sym : Obj.SymTable) {
92 auto I = Config.SymbolsToRename.find(Sym.Name);
93 if (I != Config.SymbolsToRename.end())
94 Sym.Name = std::string(I->getValue());
95 }
96
97 auto RemovePred = [Config, &Obj](const std::unique_ptr<SymbolEntry> &N) {
98 if (N->Referenced)
99 return false;
100 if (Config.KeepUndefined && N->isUndefinedSymbol())
101 return false;
102 if (Config.StripAll)
103 return true;
104 if (Config.DiscardMode == DiscardType::All && !(N->n_type & MachO::N_EXT))
105 return true;
106 // This behavior is consistent with cctools' strip.
107 if (Config.StripSwiftSymbols && (Obj.Header.Flags & MachO::MH_DYLDLINK) &&
108 Obj.SwiftVersion && *Obj.SwiftVersion && N->isSwiftSymbol())
109 return true;
110 return false;
111 };
112
113 Obj.SymTable.removeSymbols(RemovePred);
114 }
115
116 template <typename LCType>
updateLoadCommandPayloadString(LoadCommand & LC,StringRef S)117 static void updateLoadCommandPayloadString(LoadCommand &LC, StringRef S) {
118 assert(isLoadCommandWithPayloadString(LC) &&
119 "unsupported load command encountered");
120
121 uint32_t NewCmdsize = alignTo(sizeof(LCType) + S.size() + 1, 8);
122
123 LC.MachOLoadCommand.load_command_data.cmdsize = NewCmdsize;
124 LC.Payload.assign(NewCmdsize - sizeof(LCType), 0);
125 std::copy(S.begin(), S.end(), LC.Payload.begin());
126 }
127
buildRPathLoadCommand(StringRef Path)128 static LoadCommand buildRPathLoadCommand(StringRef Path) {
129 LoadCommand LC;
130 MachO::rpath_command RPathLC;
131 RPathLC.cmd = MachO::LC_RPATH;
132 RPathLC.path = sizeof(MachO::rpath_command);
133 RPathLC.cmdsize = alignTo(sizeof(MachO::rpath_command) + Path.size() + 1, 8);
134 LC.MachOLoadCommand.rpath_command_data = RPathLC;
135 LC.Payload.assign(RPathLC.cmdsize - sizeof(MachO::rpath_command), 0);
136 std::copy(Path.begin(), Path.end(), LC.Payload.begin());
137 return LC;
138 }
139
processLoadCommands(const CommonConfig & Config,Object & Obj)140 static Error processLoadCommands(const CommonConfig &Config, Object &Obj) {
141 // Remove RPaths.
142 DenseSet<StringRef> RPathsToRemove(Config.RPathsToRemove.begin(),
143 Config.RPathsToRemove.end());
144
145 LoadCommandPred RemovePred = [&RPathsToRemove,
146 &Config](const LoadCommand &LC) {
147 if (LC.MachOLoadCommand.load_command_data.cmd == MachO::LC_RPATH) {
148 // When removing all RPaths we don't need to care
149 // about what it contains
150 if (Config.RemoveAllRpaths)
151 return true;
152
153 StringRef RPath = getPayloadString(LC);
154 if (RPathsToRemove.count(RPath)) {
155 RPathsToRemove.erase(RPath);
156 return true;
157 }
158 }
159 return false;
160 };
161
162 if (Error E = Obj.removeLoadCommands(RemovePred))
163 return E;
164
165 // Emit an error if the Mach-O binary does not contain an rpath path name
166 // specified in -delete_rpath.
167 for (StringRef RPath : Config.RPathsToRemove) {
168 if (RPathsToRemove.count(RPath))
169 return createStringError(errc::invalid_argument,
170 "no LC_RPATH load command with path: %s",
171 RPath.str().c_str());
172 }
173
174 DenseSet<StringRef> RPaths;
175
176 // Get all existing RPaths.
177 for (LoadCommand &LC : Obj.LoadCommands) {
178 if (LC.MachOLoadCommand.load_command_data.cmd == MachO::LC_RPATH)
179 RPaths.insert(getPayloadString(LC));
180 }
181
182 // Throw errors for invalid RPaths.
183 for (const auto &OldNew : Config.RPathsToUpdate) {
184 StringRef Old = OldNew.getFirst();
185 StringRef New = OldNew.getSecond();
186 if (!RPaths.contains(Old))
187 return createStringError(errc::invalid_argument,
188 "no LC_RPATH load command with path: " + Old);
189 if (RPaths.contains(New))
190 return createStringError(errc::invalid_argument,
191 "rpath '" + New +
192 "' would create a duplicate load command");
193 }
194
195 // Update load commands.
196 for (LoadCommand &LC : Obj.LoadCommands) {
197 switch (LC.MachOLoadCommand.load_command_data.cmd) {
198 case MachO::LC_ID_DYLIB:
199 if (Config.SharedLibId)
200 updateLoadCommandPayloadString<MachO::dylib_command>(
201 LC, *Config.SharedLibId);
202 break;
203
204 case MachO::LC_RPATH: {
205 StringRef RPath = getPayloadString(LC);
206 StringRef NewRPath = Config.RPathsToUpdate.lookup(RPath);
207 if (!NewRPath.empty())
208 updateLoadCommandPayloadString<MachO::rpath_command>(LC, NewRPath);
209 break;
210 }
211
212 // TODO: Add LC_REEXPORT_DYLIB, LC_LAZY_LOAD_DYLIB, and LC_LOAD_UPWARD_DYLIB
213 // here once llvm-objcopy supports them.
214 case MachO::LC_LOAD_DYLIB:
215 case MachO::LC_LOAD_WEAK_DYLIB:
216 StringRef InstallName = getPayloadString(LC);
217 StringRef NewInstallName =
218 Config.InstallNamesToUpdate.lookup(InstallName);
219 if (!NewInstallName.empty())
220 updateLoadCommandPayloadString<MachO::dylib_command>(LC,
221 NewInstallName);
222 break;
223 }
224 }
225
226 // Add new RPaths.
227 for (StringRef RPath : Config.RPathToAdd) {
228 if (RPaths.contains(RPath))
229 return createStringError(errc::invalid_argument,
230 "rpath '" + RPath +
231 "' would create a duplicate load command");
232 RPaths.insert(RPath);
233 Obj.LoadCommands.push_back(buildRPathLoadCommand(RPath));
234 }
235
236 for (StringRef RPath : Config.RPathToPrepend) {
237 if (RPaths.contains(RPath))
238 return createStringError(errc::invalid_argument,
239 "rpath '" + RPath +
240 "' would create a duplicate load command");
241
242 RPaths.insert(RPath);
243 Obj.LoadCommands.insert(Obj.LoadCommands.begin(),
244 buildRPathLoadCommand(RPath));
245 }
246
247 // Unlike appending rpaths, the indexes of subsequent load commands must
248 // be recalculated after prepending one.
249 if (!Config.RPathToPrepend.empty())
250 Obj.updateLoadCommandIndexes();
251
252 return Error::success();
253 }
254
dumpSectionToFile(StringRef SecName,StringRef Filename,Object & Obj)255 static Error dumpSectionToFile(StringRef SecName, StringRef Filename,
256 Object &Obj) {
257 for (LoadCommand &LC : Obj.LoadCommands)
258 for (const std::unique_ptr<Section> &Sec : LC.Sections) {
259 if (Sec->CanonicalName == SecName) {
260 Expected<std::unique_ptr<FileOutputBuffer>> BufferOrErr =
261 FileOutputBuffer::create(Filename, Sec->Content.size());
262 if (!BufferOrErr)
263 return BufferOrErr.takeError();
264 std::unique_ptr<FileOutputBuffer> Buf = std::move(*BufferOrErr);
265 llvm::copy(Sec->Content, Buf->getBufferStart());
266
267 if (Error E = Buf->commit())
268 return E;
269 return Error::success();
270 }
271 }
272
273 return createStringError(object_error::parse_failed, "section '%s' not found",
274 SecName.str().c_str());
275 }
276
addSection(StringRef SecName,StringRef Filename,Object & Obj)277 static Error addSection(StringRef SecName, StringRef Filename, Object &Obj) {
278 ErrorOr<std::unique_ptr<MemoryBuffer>> BufOrErr =
279 MemoryBuffer::getFile(Filename);
280 if (!BufOrErr)
281 return createFileError(Filename, errorCodeToError(BufOrErr.getError()));
282 std::unique_ptr<MemoryBuffer> Buf = std::move(*BufOrErr);
283
284 std::pair<StringRef, StringRef> Pair = SecName.split(',');
285 StringRef TargetSegName = Pair.first;
286 Section Sec(TargetSegName, Pair.second);
287 Sec.Content = Obj.NewSectionsContents.save(Buf->getBuffer());
288 Sec.Size = Sec.Content.size();
289
290 // Add the a section into an existing segment.
291 for (LoadCommand &LC : Obj.LoadCommands) {
292 Optional<StringRef> SegName = LC.getSegmentName();
293 if (SegName && SegName == TargetSegName) {
294 uint64_t Addr = *LC.getSegmentVMAddr();
295 for (const std::unique_ptr<Section> &S : LC.Sections)
296 Addr = std::max(Addr, S->Addr + S->Size);
297 LC.Sections.push_back(std::make_unique<Section>(Sec));
298 LC.Sections.back()->Addr = Addr;
299 return Error::success();
300 }
301 }
302
303 // There's no segment named TargetSegName. Create a new load command and
304 // Insert a new section into it.
305 LoadCommand &NewSegment =
306 Obj.addSegment(TargetSegName, alignTo(Sec.Size, 16384));
307 NewSegment.Sections.push_back(std::make_unique<Section>(Sec));
308 NewSegment.Sections.back()->Addr = *NewSegment.getSegmentVMAddr();
309 return Error::success();
310 }
311
312 // isValidMachOCannonicalName returns success if Name is a MachO cannonical name
313 // ("<segment>,<section>") and lengths of both segment and section names are
314 // valid.
isValidMachOCannonicalName(StringRef Name)315 static Error isValidMachOCannonicalName(StringRef Name) {
316 if (Name.count(',') != 1)
317 return createStringError(errc::invalid_argument,
318 "invalid section name '%s' (should be formatted "
319 "as '<segment name>,<section name>')",
320 Name.str().c_str());
321
322 std::pair<StringRef, StringRef> Pair = Name.split(',');
323 if (Pair.first.size() > 16)
324 return createStringError(errc::invalid_argument,
325 "too long segment name: '%s'",
326 Pair.first.str().c_str());
327 if (Pair.second.size() > 16)
328 return createStringError(errc::invalid_argument,
329 "too long section name: '%s'",
330 Pair.second.str().c_str());
331 return Error::success();
332 }
333
handleArgs(const CommonConfig & Config,Object & Obj)334 static Error handleArgs(const CommonConfig &Config, Object &Obj) {
335 // Dump sections before add/remove for compatibility with GNU objcopy.
336 for (StringRef Flag : Config.DumpSection) {
337 StringRef SectionName;
338 StringRef FileName;
339 std::tie(SectionName, FileName) = Flag.split('=');
340 if (Error E = dumpSectionToFile(SectionName, FileName, Obj))
341 return E;
342 }
343
344 if (Error E = removeSections(Config, Obj))
345 return E;
346
347 // Mark symbols to determine which symbols are still needed.
348 if (Config.StripAll)
349 markSymbols(Config, Obj);
350
351 updateAndRemoveSymbols(Config, Obj);
352
353 if (Config.StripAll)
354 for (LoadCommand &LC : Obj.LoadCommands)
355 for (std::unique_ptr<Section> &Sec : LC.Sections)
356 Sec->Relocations.clear();
357
358 for (const auto &Flag : Config.AddSection) {
359 std::pair<StringRef, StringRef> SecPair = Flag.split("=");
360 StringRef SecName = SecPair.first;
361 StringRef File = SecPair.second;
362 if (Error E = isValidMachOCannonicalName(SecName))
363 return E;
364 if (Error E = addSection(SecName, File, Obj))
365 return E;
366 }
367
368 if (Error E = processLoadCommands(Config, Obj))
369 return E;
370
371 return Error::success();
372 }
373
executeObjcopyOnBinary(const CommonConfig & Config,const MachOConfig &,object::MachOObjectFile & In,raw_ostream & Out)374 Error executeObjcopyOnBinary(const CommonConfig &Config, const MachOConfig &,
375 object::MachOObjectFile &In, raw_ostream &Out) {
376 MachOReader Reader(In);
377 Expected<std::unique_ptr<Object>> O = Reader.create();
378 if (!O)
379 return createFileError(Config.InputFilename, O.takeError());
380
381 if (Error E = handleArgs(Config, **O))
382 return createFileError(Config.InputFilename, std::move(E));
383
384 // Page size used for alignment of segment sizes in Mach-O executables and
385 // dynamic libraries.
386 uint64_t PageSize;
387 switch (In.getArch()) {
388 case Triple::ArchType::arm:
389 case Triple::ArchType::aarch64:
390 case Triple::ArchType::aarch64_32:
391 PageSize = 16384;
392 break;
393 default:
394 PageSize = 4096;
395 }
396
397 MachOWriter Writer(**O, In.is64Bit(), In.isLittleEndian(), PageSize, Out);
398 if (auto E = Writer.finalize())
399 return E;
400 return Writer.write();
401 }
402
executeObjcopyOnMachOUniversalBinary(const MultiFormatConfig & Config,const MachOUniversalBinary & In,raw_ostream & Out)403 Error executeObjcopyOnMachOUniversalBinary(const MultiFormatConfig &Config,
404 const MachOUniversalBinary &In,
405 raw_ostream &Out) {
406 SmallVector<OwningBinary<Binary>, 2> Binaries;
407 SmallVector<Slice, 2> Slices;
408 for (const auto &O : In.objects()) {
409 Expected<std::unique_ptr<Archive>> ArOrErr = O.getAsArchive();
410 if (ArOrErr) {
411 Expected<std::vector<NewArchiveMember>> NewArchiveMembersOrErr =
412 createNewArchiveMembers(Config, **ArOrErr);
413 if (!NewArchiveMembersOrErr)
414 return NewArchiveMembersOrErr.takeError();
415 Expected<std::unique_ptr<MemoryBuffer>> OutputBufferOrErr =
416 writeArchiveToBuffer(*NewArchiveMembersOrErr,
417 (*ArOrErr)->hasSymbolTable(), (*ArOrErr)->kind(),
418 Config.getCommonConfig().DeterministicArchives,
419 (*ArOrErr)->isThin());
420 if (!OutputBufferOrErr)
421 return OutputBufferOrErr.takeError();
422 Expected<std::unique_ptr<Binary>> BinaryOrErr =
423 object::createBinary(**OutputBufferOrErr);
424 if (!BinaryOrErr)
425 return BinaryOrErr.takeError();
426 Binaries.emplace_back(std::move(*BinaryOrErr),
427 std::move(*OutputBufferOrErr));
428 Slices.emplace_back(*cast<Archive>(Binaries.back().getBinary()),
429 O.getCPUType(), O.getCPUSubType(),
430 O.getArchFlagName(), O.getAlign());
431 continue;
432 }
433 // The methods getAsArchive, getAsObjectFile, getAsIRObject of the class
434 // ObjectForArch return an Error in case of the type mismatch. We need to
435 // check each in turn to see what kind of slice this is, so ignore errors
436 // produced along the way.
437 consumeError(ArOrErr.takeError());
438
439 Expected<std::unique_ptr<MachOObjectFile>> ObjOrErr = O.getAsObjectFile();
440 if (!ObjOrErr) {
441 consumeError(ObjOrErr.takeError());
442 return createStringError(
443 std::errc::invalid_argument,
444 "slice for '%s' of the universal Mach-O binary "
445 "'%s' is not a Mach-O object or an archive",
446 O.getArchFlagName().c_str(),
447 Config.getCommonConfig().InputFilename.str().c_str());
448 }
449 std::string ArchFlagName = O.getArchFlagName();
450
451 SmallVector<char, 0> Buffer;
452 raw_svector_ostream MemStream(Buffer);
453
454 Expected<const MachOConfig &> MachO = Config.getMachOConfig();
455 if (!MachO)
456 return MachO.takeError();
457
458 if (Error E = executeObjcopyOnBinary(Config.getCommonConfig(), *MachO,
459 **ObjOrErr, MemStream))
460 return E;
461
462 std::unique_ptr<MemoryBuffer> MB =
463 std::make_unique<SmallVectorMemoryBuffer>(std::move(Buffer),
464 ArchFlagName);
465 Expected<std::unique_ptr<Binary>> BinaryOrErr = object::createBinary(*MB);
466 if (!BinaryOrErr)
467 return BinaryOrErr.takeError();
468 Binaries.emplace_back(std::move(*BinaryOrErr), std::move(MB));
469 Slices.emplace_back(*cast<MachOObjectFile>(Binaries.back().getBinary()),
470 O.getAlign());
471 }
472
473 if (Error Err = writeUniversalBinaryToStream(Slices, Out))
474 return Err;
475
476 return Error::success();
477 }
478
479 } // end namespace macho
480 } // end namespace objcopy
481 } // end namespace llvm
482