1 //===- ELFObjcopy.cpp -----------------------------------------------------===//
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
7 //===----------------------------------------------------------------------===//
9 #include "ELFObjcopy.h"
11 #include "CopyConfig.h"
13 #include "llvm-objcopy.h"
15 #include "llvm/ADT/BitmaskEnum.h"
16 #include "llvm/ADT/Optional.h"
17 #include "llvm/ADT/STLExtras.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/ADT/StringRef.h"
20 #include "llvm/ADT/Twine.h"
21 #include "llvm/BinaryFormat/ELF.h"
22 #include "llvm/MC/MCTargetOptions.h"
23 #include "llvm/Object/Binary.h"
24 #include "llvm/Object/ELFObjectFile.h"
25 #include "llvm/Object/ELFTypes.h"
26 #include "llvm/Object/Error.h"
27 #include "llvm/Option/Option.h"
28 #include "llvm/Support/Casting.h"
29 #include "llvm/Support/Compression.h"
30 #include "llvm/Support/Errc.h"
31 #include "llvm/Support/Error.h"
32 #include "llvm/Support/ErrorHandling.h"
33 #include "llvm/Support/ErrorOr.h"
34 #include "llvm/Support/Memory.h"
35 #include "llvm/Support/Path.h"
36 #include "llvm/Support/raw_ostream.h"
44 #include <system_error>
51 using namespace object;
53 using SectionPred = std::function<bool(const SectionBase &Sec)>;
55 static bool isDebugSection(const SectionBase &Sec) {
56 return StringRef(Sec.Name).startswith(".debug") ||
57 StringRef(Sec.Name).startswith(".zdebug") || Sec.Name == ".gdb_index";
60 static bool isDWOSection(const SectionBase &Sec) {
61 return StringRef(Sec.Name).endswith(".dwo");
64 static bool onlyKeepDWOPred(const Object &Obj, const SectionBase &Sec) {
65 // We can't remove the section header string table.
66 if (&Sec == Obj.SectionNames)
68 // Short of keeping the string table we want to keep everything that is a DWO
69 // section and remove everything else.
70 return !isDWOSection(Sec);
73 static uint64_t setSectionFlagsPreserveMask(uint64_t OldFlags,
75 // Preserve some flags which should not be dropped when setting flags.
76 // Also, preserve anything OS/processor dependant.
77 const uint64_t PreserveMask = ELF::SHF_COMPRESSED | ELF::SHF_EXCLUDE |
78 ELF::SHF_GROUP | ELF::SHF_LINK_ORDER |
79 ELF::SHF_MASKOS | ELF::SHF_MASKPROC |
80 ELF::SHF_TLS | ELF::SHF_INFO_LINK;
81 return (OldFlags & PreserveMask) | (NewFlags & ~PreserveMask);
84 static ElfType getOutputElfType(const Binary &Bin) {
85 // Infer output ELF type from the input ELF object
86 if (isa<ELFObjectFile<ELF32LE>>(Bin))
88 if (isa<ELFObjectFile<ELF64LE>>(Bin))
90 if (isa<ELFObjectFile<ELF32BE>>(Bin))
92 if (isa<ELFObjectFile<ELF64BE>>(Bin))
94 llvm_unreachable("Invalid ELFType");
97 static ElfType getOutputElfType(const MachineInfo &MI) {
98 // Infer output ELF type from the binary arch specified
100 return MI.IsLittleEndian ? ELFT_ELF64LE : ELFT_ELF64BE;
102 return MI.IsLittleEndian ? ELFT_ELF32LE : ELFT_ELF32BE;
105 static std::unique_ptr<Writer> createWriter(const CopyConfig &Config,
106 Object &Obj, Buffer &Buf,
107 ElfType OutputElfType) {
108 if (Config.OutputFormat == "binary") {
109 return llvm::make_unique<BinaryWriter>(Obj, Buf);
111 // Depending on the initial ELFT and OutputFormat we need a different Writer.
112 switch (OutputElfType) {
114 return llvm::make_unique<ELFWriter<ELF32LE>>(Obj, Buf,
115 !Config.StripSections);
117 return llvm::make_unique<ELFWriter<ELF64LE>>(Obj, Buf,
118 !Config.StripSections);
120 return llvm::make_unique<ELFWriter<ELF32BE>>(Obj, Buf,
121 !Config.StripSections);
123 return llvm::make_unique<ELFWriter<ELF64BE>>(Obj, Buf,
124 !Config.StripSections);
126 llvm_unreachable("Invalid output format");
129 template <class ELFT>
130 static Expected<ArrayRef<uint8_t>>
131 findBuildID(const object::ELFFile<ELFT> &In) {
132 for (const auto &Phdr : unwrapOrError(In.program_headers())) {
133 if (Phdr.p_type != PT_NOTE)
135 Error Err = Error::success();
136 for (const auto &Note : In.notes(Phdr, Err))
137 if (Note.getType() == NT_GNU_BUILD_ID && Note.getName() == ELF_NOTE_GNU)
138 return Note.getDesc();
140 return std::move(Err);
142 return createStringError(llvm::errc::invalid_argument,
143 "Could not find build ID.");
146 static Expected<ArrayRef<uint8_t>>
147 findBuildID(const object::ELFObjectFileBase &In) {
148 if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(&In))
149 return findBuildID(*O->getELFFile());
150 else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(&In))
151 return findBuildID(*O->getELFFile());
152 else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(&In))
153 return findBuildID(*O->getELFFile());
154 else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(&In))
155 return findBuildID(*O->getELFFile());
157 llvm_unreachable("Bad file format");
160 static Error linkToBuildIdDir(const CopyConfig &Config, StringRef ToLink,
162 ArrayRef<uint8_t> BuildIdBytes) {
163 SmallString<128> Path = Config.BuildIdLinkDir;
164 sys::path::append(Path, llvm::toHex(BuildIdBytes[0], /*LowerCase*/ true));
165 if (auto EC = sys::fs::create_directories(Path))
166 return createFileError(
168 createStringError(EC, "cannot create build ID link directory"));
170 sys::path::append(Path,
171 llvm::toHex(BuildIdBytes.slice(1), /*LowerCase*/ true));
173 if (auto EC = sys::fs::create_hard_link(ToLink, Path)) {
174 // Hard linking failed, try to remove the file first if it exists.
175 if (sys::fs::exists(Path))
176 sys::fs::remove(Path);
177 EC = sys::fs::create_hard_link(ToLink, Path);
179 return createStringError(EC, "cannot link %s to %s", ToLink.data(),
182 return Error::success();
185 static Error splitDWOToFile(const CopyConfig &Config, const Reader &Reader,
186 StringRef File, ElfType OutputElfType) {
187 auto DWOFile = Reader.create();
188 auto OnlyKeepDWOPred = [&DWOFile](const SectionBase &Sec) {
189 return onlyKeepDWOPred(*DWOFile, Sec);
191 if (Error E = DWOFile->removeSections(OnlyKeepDWOPred))
193 if (Config.OutputArch)
194 DWOFile->Machine = Config.OutputArch.getValue().EMachine;
196 auto Writer = createWriter(Config, *DWOFile, FB, OutputElfType);
197 if (Error E = Writer->finalize())
199 return Writer->write();
202 static Error dumpSectionToFile(StringRef SecName, StringRef Filename,
204 for (auto &Sec : Obj.sections()) {
205 if (Sec.Name == SecName) {
206 if (Sec.OriginalData.empty())
207 return createStringError(
208 object_error::parse_failed,
209 "Can't dump section \"%s\": it has no contents",
210 SecName.str().c_str());
211 Expected<std::unique_ptr<FileOutputBuffer>> BufferOrErr =
212 FileOutputBuffer::create(Filename, Sec.OriginalData.size());
214 return BufferOrErr.takeError();
215 std::unique_ptr<FileOutputBuffer> Buf = std::move(*BufferOrErr);
216 std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(),
217 Buf->getBufferStart());
218 if (Error E = Buf->commit())
220 return Error::success();
223 return createStringError(object_error::parse_failed, "Section not found");
226 static bool isCompressable(const SectionBase &Section) {
227 return !(Section.Flags & ELF::SHF_COMPRESSED) &&
228 StringRef(Section.Name).startswith(".debug");
231 static void replaceDebugSections(
232 Object &Obj, SectionPred &RemovePred,
233 function_ref<bool(const SectionBase &)> shouldReplace,
234 function_ref<SectionBase *(const SectionBase *)> addSection) {
235 // Build a list of the debug sections we are going to replace.
236 // We can't call `addSection` while iterating over sections,
237 // because it would mutate the sections array.
238 SmallVector<SectionBase *, 13> ToReplace;
239 for (auto &Sec : Obj.sections())
240 if (shouldReplace(Sec))
241 ToReplace.push_back(&Sec);
243 // Build a mapping from original section to a new one.
244 DenseMap<SectionBase *, SectionBase *> FromTo;
245 for (SectionBase *S : ToReplace)
246 FromTo[S] = addSection(S);
248 // Now we want to update the target sections of relocation
249 // sections. Also we will update the relocations themselves
250 // to update the symbol references.
251 for (auto &Sec : Obj.sections())
252 Sec.replaceSectionReferences(FromTo);
254 RemovePred = [shouldReplace, RemovePred](const SectionBase &Sec) {
255 return shouldReplace(Sec) || RemovePred(Sec);
259 static bool isUnneededSymbol(const Symbol &Sym) {
260 return !Sym.Referenced &&
261 (Sym.Binding == STB_LOCAL || Sym.getShndx() == SHN_UNDEF) &&
262 Sym.Type != STT_FILE && Sym.Type != STT_SECTION;
265 // This function handles the high level operations of GNU objcopy including
266 // handling command line options. It's important to outline certain properties
267 // we expect to hold of the command line operations. Any operation that "keeps"
268 // should keep regardless of a remove. Additionally any removal should respect
269 // any previous removals. Lastly whether or not something is removed shouldn't
270 // depend a) on the order the options occur in or b) on some opaque priority
271 // system. The only priority is that keeps/copies overrule removes.
272 static Error handleArgs(const CopyConfig &Config, Object &Obj,
273 const Reader &Reader, ElfType OutputElfType) {
275 if (!Config.SplitDWO.empty())
277 splitDWOToFile(Config, Reader, Config.SplitDWO, OutputElfType))
280 if (Config.OutputArch)
281 Obj.Machine = Config.OutputArch.getValue().EMachine;
283 // TODO: update or remove symbols only if there is an option that affects
285 if (Obj.SymbolTable) {
286 Obj.SymbolTable->updateSymbols([&](Symbol &Sym) {
287 // Common and undefined symbols don't make sense as local symbols, and can
288 // even cause crashes if we localize those, so skip them.
289 if (!Sym.isCommon() && Sym.getShndx() != SHN_UNDEF &&
290 ((Config.LocalizeHidden &&
291 (Sym.Visibility == STV_HIDDEN || Sym.Visibility == STV_INTERNAL)) ||
292 is_contained(Config.SymbolsToLocalize, Sym.Name)))
293 Sym.Binding = STB_LOCAL;
295 // Note: these two globalize flags have very similar names but different
298 // --globalize-symbol: promote a symbol to global
299 // --keep-global-symbol: all symbols except for these should be made local
301 // If --globalize-symbol is specified for a given symbol, it will be
302 // global in the output file even if it is not included via
303 // --keep-global-symbol. Because of that, make sure to check
304 // --globalize-symbol second.
305 if (!Config.SymbolsToKeepGlobal.empty() &&
306 !is_contained(Config.SymbolsToKeepGlobal, Sym.Name) &&
307 Sym.getShndx() != SHN_UNDEF)
308 Sym.Binding = STB_LOCAL;
310 if (is_contained(Config.SymbolsToGlobalize, Sym.Name) &&
311 Sym.getShndx() != SHN_UNDEF)
312 Sym.Binding = STB_GLOBAL;
314 if (is_contained(Config.SymbolsToWeaken, Sym.Name) &&
315 Sym.Binding == STB_GLOBAL)
316 Sym.Binding = STB_WEAK;
318 if (Config.Weaken && Sym.Binding == STB_GLOBAL &&
319 Sym.getShndx() != SHN_UNDEF)
320 Sym.Binding = STB_WEAK;
322 const auto I = Config.SymbolsToRename.find(Sym.Name);
323 if (I != Config.SymbolsToRename.end())
324 Sym.Name = I->getValue();
326 if (!Config.SymbolsPrefix.empty() && Sym.Type != STT_SECTION)
327 Sym.Name = (Config.SymbolsPrefix + Sym.Name).str();
330 // The purpose of this loop is to mark symbols referenced by sections
331 // (like GroupSection or RelocationSection). This way, we know which
332 // symbols are still 'needed' and which are not.
333 if (Config.StripUnneeded || !Config.UnneededSymbolsToRemove.empty()) {
334 for (auto &Section : Obj.sections())
335 Section.markSymbols();
338 auto RemoveSymbolsPred = [&](const Symbol &Sym) {
339 if (is_contained(Config.SymbolsToKeep, Sym.Name) ||
340 (Config.KeepFileSymbols && Sym.Type == STT_FILE))
343 if ((Config.DiscardMode == DiscardType::All ||
344 (Config.DiscardMode == DiscardType::Locals &&
345 StringRef(Sym.Name).startswith(".L"))) &&
346 Sym.Binding == STB_LOCAL && Sym.getShndx() != SHN_UNDEF &&
347 Sym.Type != STT_FILE && Sym.Type != STT_SECTION)
350 if (Config.StripAll || Config.StripAllGNU)
353 if (is_contained(Config.SymbolsToRemove, Sym.Name))
356 if ((Config.StripUnneeded ||
357 is_contained(Config.UnneededSymbolsToRemove, Sym.Name)) &&
358 isUnneededSymbol(Sym))
363 if (Error E = Obj.removeSymbols(RemoveSymbolsPred))
367 SectionPred RemovePred = [](const SectionBase &) { return false; };
370 if (!Config.ToRemove.empty()) {
371 RemovePred = [&Config](const SectionBase &Sec) {
372 return is_contained(Config.ToRemove, Sec.Name);
376 if (Config.StripDWO || !Config.SplitDWO.empty())
377 RemovePred = [RemovePred](const SectionBase &Sec) {
378 return isDWOSection(Sec) || RemovePred(Sec);
381 if (Config.ExtractDWO)
382 RemovePred = [RemovePred, &Obj](const SectionBase &Sec) {
383 return onlyKeepDWOPred(Obj, Sec) || RemovePred(Sec);
386 if (Config.StripAllGNU)
387 RemovePred = [RemovePred, &Obj](const SectionBase &Sec) {
390 if ((Sec.Flags & SHF_ALLOC) != 0)
392 if (&Sec == Obj.SectionNames)
401 return isDebugSection(Sec);
404 if (Config.StripSections) {
405 RemovePred = [RemovePred](const SectionBase &Sec) {
406 return RemovePred(Sec) || Sec.ParentSegment == nullptr;
410 if (Config.StripDebug) {
411 RemovePred = [RemovePred](const SectionBase &Sec) {
412 return RemovePred(Sec) || isDebugSection(Sec);
416 if (Config.StripNonAlloc)
417 RemovePred = [RemovePred, &Obj](const SectionBase &Sec) {
420 if (&Sec == Obj.SectionNames)
422 return (Sec.Flags & SHF_ALLOC) == 0 && Sec.ParentSegment == nullptr;
426 RemovePred = [RemovePred, &Obj](const SectionBase &Sec) {
429 if (&Sec == Obj.SectionNames)
431 if (StringRef(Sec.Name).startswith(".gnu.warning"))
433 if (Sec.ParentSegment != nullptr)
435 return (Sec.Flags & SHF_ALLOC) == 0;
439 if (!Config.OnlySection.empty()) {
440 RemovePred = [&Config, RemovePred, &Obj](const SectionBase &Sec) {
441 // Explicitly keep these sections regardless of previous removes.
442 if (is_contained(Config.OnlySection, Sec.Name))
445 // Allow all implicit removes.
449 // Keep special sections.
450 if (Obj.SectionNames == &Sec)
452 if (Obj.SymbolTable == &Sec ||
453 (Obj.SymbolTable && Obj.SymbolTable->getStrTab() == &Sec))
456 // Remove everything else.
461 if (!Config.KeepSection.empty()) {
462 RemovePred = [&Config, RemovePred](const SectionBase &Sec) {
463 // Explicitly keep these sections regardless of previous removes.
464 if (is_contained(Config.KeepSection, Sec.Name))
466 // Otherwise defer to RemovePred.
467 return RemovePred(Sec);
471 // This has to be the last predicate assignment.
472 // If the option --keep-symbol has been specified
473 // and at least one of those symbols is present
474 // (equivalently, the updated symbol table is not empty)
475 // the symbol table and the string table should not be removed.
476 if ((!Config.SymbolsToKeep.empty() || Config.KeepFileSymbols) &&
477 Obj.SymbolTable && !Obj.SymbolTable->empty()) {
478 RemovePred = [&Obj, RemovePred](const SectionBase &Sec) {
479 if (&Sec == Obj.SymbolTable || &Sec == Obj.SymbolTable->getStrTab())
481 return RemovePred(Sec);
485 if (Config.CompressionType != DebugCompressionType::None)
486 replaceDebugSections(Obj, RemovePred, isCompressable,
487 [&Config, &Obj](const SectionBase *S) {
488 return &Obj.addSection<CompressedSection>(
489 *S, Config.CompressionType);
491 else if (Config.DecompressDebugSections)
492 replaceDebugSections(
494 [](const SectionBase &S) { return isa<CompressedSection>(&S); },
495 [&Obj](const SectionBase *S) {
496 auto CS = cast<CompressedSection>(S);
497 return &Obj.addSection<DecompressedSection>(*CS);
500 if (Error E = Obj.removeSections(RemovePred))
503 if (!Config.SectionsToRename.empty()) {
504 for (auto &Sec : Obj.sections()) {
505 const auto Iter = Config.SectionsToRename.find(Sec.Name);
506 if (Iter != Config.SectionsToRename.end()) {
507 const SectionRename &SR = Iter->second;
508 Sec.Name = SR.NewName;
509 if (SR.NewFlags.hasValue())
511 setSectionFlagsPreserveMask(Sec.Flags, SR.NewFlags.getValue());
516 if (!Config.SetSectionFlags.empty()) {
517 for (auto &Sec : Obj.sections()) {
518 const auto Iter = Config.SetSectionFlags.find(Sec.Name);
519 if (Iter != Config.SetSectionFlags.end()) {
520 const SectionFlagsUpdate &SFU = Iter->second;
521 Sec.Flags = setSectionFlagsPreserveMask(Sec.Flags, SFU.NewFlags);
526 for (const auto &Flag : Config.AddSection) {
527 std::pair<StringRef, StringRef> SecPair = Flag.split("=");
528 StringRef SecName = SecPair.first;
529 StringRef File = SecPair.second;
530 ErrorOr<std::unique_ptr<MemoryBuffer>> BufOrErr =
531 MemoryBuffer::getFile(File);
533 return createFileError(File, errorCodeToError(BufOrErr.getError()));
534 std::unique_ptr<MemoryBuffer> Buf = std::move(*BufOrErr);
535 ArrayRef<uint8_t> Data(
536 reinterpret_cast<const uint8_t *>(Buf->getBufferStart()),
537 Buf->getBufferSize());
538 OwnedDataSection &NewSection =
539 Obj.addSection<OwnedDataSection>(SecName, Data);
540 if (SecName.startswith(".note") && SecName != ".note.GNU-stack")
541 NewSection.Type = SHT_NOTE;
544 for (const auto &Flag : Config.DumpSection) {
545 std::pair<StringRef, StringRef> SecPair = Flag.split("=");
546 StringRef SecName = SecPair.first;
547 StringRef File = SecPair.second;
548 if (Error E = dumpSectionToFile(SecName, File, Obj))
549 return createFileError(Config.InputFilename, std::move(E));
552 if (!Config.AddGnuDebugLink.empty())
553 Obj.addSection<GnuDebugLinkSection>(Config.AddGnuDebugLink);
555 for (const NewSymbolInfo &SI : Config.SymbolsToAdd) {
556 SectionBase *Sec = Obj.findSection(SI.SectionName);
557 uint64_t Value = Sec ? Sec->Addr + SI.Value : SI.Value;
558 Obj.SymbolTable->addSymbol(
559 SI.SymbolName, SI.Bind, SI.Type, Sec, Value, SI.Visibility,
560 Sec ? (uint16_t)SYMBOL_SIMPLE_INDEX : (uint16_t)SHN_ABS, 0);
563 if (Config.EntryExpr)
564 Obj.Entry = Config.EntryExpr(Obj.Entry);
565 return Error::success();
568 Error executeObjcopyOnRawBinary(const CopyConfig &Config, MemoryBuffer &In,
570 BinaryReader Reader(Config.BinaryArch, &In);
571 std::unique_ptr<Object> Obj = Reader.create();
573 // Prefer OutputArch (-O<format>) if set, otherwise fallback to BinaryArch
575 const ElfType OutputElfType = getOutputElfType(
576 Config.OutputArch ? Config.OutputArch.getValue() : Config.BinaryArch);
577 if (Error E = handleArgs(Config, *Obj, Reader, OutputElfType))
579 std::unique_ptr<Writer> Writer =
580 createWriter(Config, *Obj, Out, OutputElfType);
581 if (Error E = Writer->finalize())
583 return Writer->write();
586 Error executeObjcopyOnBinary(const CopyConfig &Config,
587 object::ELFObjectFileBase &In, Buffer &Out) {
588 ELFReader Reader(&In);
589 std::unique_ptr<Object> Obj = Reader.create();
590 // Prefer OutputArch (-O<format>) if set, otherwise infer it from the input.
591 const ElfType OutputElfType =
592 Config.OutputArch ? getOutputElfType(Config.OutputArch.getValue())
593 : getOutputElfType(In);
594 ArrayRef<uint8_t> BuildIdBytes;
596 if (!Config.BuildIdLinkDir.empty()) {
597 BuildIdBytes = unwrapOrError(findBuildID(In));
598 if (BuildIdBytes.size() < 2)
599 return createFileError(
600 Config.InputFilename,
601 createStringError(object_error::parse_failed,
602 "build ID is smaller than two bytes."));
605 if (!Config.BuildIdLinkDir.empty() && Config.BuildIdLinkInput)
607 linkToBuildIdDir(Config, Config.InputFilename,
608 Config.BuildIdLinkInput.getValue(), BuildIdBytes))
611 if (Error E = handleArgs(Config, *Obj, Reader, OutputElfType))
613 std::unique_ptr<Writer> Writer =
614 createWriter(Config, *Obj, Out, OutputElfType);
615 if (Error E = Writer->finalize())
617 if (Error E = Writer->write())
619 if (!Config.BuildIdLinkDir.empty() && Config.BuildIdLinkOutput)
621 linkToBuildIdDir(Config, Config.OutputFilename,
622 Config.BuildIdLinkOutput.getValue(), BuildIdBytes))
625 return Error::success();
628 } // end namespace elf
629 } // end namespace objcopy
630 } // end namespace llvm