1 //===- Writer.cpp ---------------------------------------------------------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
12 #include "Filesystem.h"
13 #include "LinkerScript.h"
16 #include "OutputSections.h"
17 #include "Relocations.h"
19 #include "SymbolTable.h"
20 #include "SyntheticSections.h"
23 #include "llvm/ADT/StringMap.h"
24 #include "llvm/ADT/StringSwitch.h"
25 #include "llvm/Support/FileOutputBuffer.h"
26 #include "llvm/Support/raw_ostream.h"
30 using namespace llvm::ELF;
31 using namespace llvm::object;
32 using namespace llvm::support;
33 using namespace llvm::support::endian;
36 using namespace lld::elf;
39 // The writer writes a SymbolTable result to a file.
40 template <class ELFT> class Writer {
42 typedef typename ELFT::Shdr Elf_Shdr;
43 typedef typename ELFT::Ehdr Elf_Ehdr;
44 typedef typename ELFT::Phdr Elf_Phdr;
49 void createSyntheticSections();
50 void copyLocalSymbols();
51 void addSectionSymbols();
52 void addReservedSymbols();
53 void createSections();
54 void forEachRelSec(std::function<void(InputSectionBase &)> Fn);
56 void finalizeSections();
57 void addPredefinedSections();
59 std::vector<PhdrEntry> createPhdrs();
60 void removeEmptyPTLoad();
61 void addPtArmExid(std::vector<PhdrEntry> &Phdrs);
62 void assignFileOffsets();
63 void assignFileOffsetsBinary();
65 void fixSectionAlignments();
66 void fixPredefinedSymbols();
70 void writeSectionsBinary();
73 std::unique_ptr<FileOutputBuffer> Buffer;
75 std::vector<OutputSection *> OutputSections;
76 OutputSectionFactory Factory{OutputSections};
78 void addRelIpltSymbols();
79 void addStartEndSymbols();
80 void addStartStopSymbols(OutputSection *Sec);
81 uint64_t getEntryAddr();
82 OutputSection *findSection(StringRef Name);
84 std::vector<PhdrEntry> Phdrs;
87 uint64_t SectionHeaderOff;
89 } // anonymous namespace
91 StringRef elf::getOutputSectionName(StringRef Name) {
92 if (Config->Relocatable)
95 // If -emit-relocs is given (which is rare), we need to copy
96 // relocation sections to the output. If input section .foo is
97 // output as .bar, we want to rename .rel.foo .rel.bar as well.
98 if (Config->EmitRelocs) {
99 for (StringRef V : {".rel.", ".rela."}) {
100 if (Name.startswith(V)) {
101 StringRef Inner = getOutputSectionName(Name.substr(V.size() - 1));
102 return Saver.save(V.drop_back() + Inner);
108 {".text.", ".rodata.", ".data.rel.ro.", ".data.", ".bss.rel.ro.",
109 ".bss.", ".init_array.", ".fini_array.", ".ctors.", ".dtors.", ".tbss.",
110 ".gcc_except_table.", ".tdata.", ".ARM.exidx."}) {
111 StringRef Prefix = V.drop_back();
112 if (Name.startswith(V) || Name == Prefix)
116 // CommonSection is identified as "COMMON" in linker scripts.
117 // By default, it should go to .bss section.
118 if (Name == "COMMON")
121 // ".zdebug_" is a prefix for ZLIB-compressed sections.
122 // Because we decompressed input sections, we want to remove 'z'.
123 if (Name.startswith(".zdebug_"))
124 return Saver.save("." + Name.substr(2));
128 template <class ELFT> static bool needsInterpSection() {
129 return !Symtab<ELFT>::X->getSharedFiles().empty() &&
130 !Config->DynamicLinker.empty() && !Script->ignoreInterpSection();
133 template <class ELFT> void elf::writeResult() { Writer<ELFT>().run(); }
135 template <class ELFT> void Writer<ELFT>::removeEmptyPTLoad() {
136 auto I = std::remove_if(Phdrs.begin(), Phdrs.end(), [&](const PhdrEntry &P) {
137 if (P.p_type != PT_LOAD)
141 uint64_t Size = P.Last->Addr + P.Last->Size - P.First->Addr;
144 Phdrs.erase(I, Phdrs.end());
147 // This function scans over the input sections and creates mergeable
148 // synthetic sections. It removes MergeInputSections from array and
149 // adds new synthetic ones. Each synthetic section is added to the
150 // location of the first input section it replaces.
151 static void combineMergableSections() {
152 std::vector<MergeSyntheticSection *> MergeSections;
153 for (InputSectionBase *&S : InputSections) {
154 MergeInputSection *MS = dyn_cast<MergeInputSection>(S);
158 // We do not want to handle sections that are not alive, so just remove
159 // them instead of trying to merge.
163 StringRef OutsecName = getOutputSectionName(MS->Name);
164 uint64_t Flags = MS->Flags & ~(uint64_t)(SHF_GROUP | SHF_COMPRESSED);
165 uint32_t Alignment = std::max<uint32_t>(MS->Alignment, MS->Entsize);
168 llvm::find_if(MergeSections, [=](MergeSyntheticSection *Sec) {
169 return Sec->Name == OutsecName && Sec->Flags == Flags &&
170 Sec->Alignment == Alignment;
172 if (I == MergeSections.end()) {
173 MergeSyntheticSection *Syn =
174 make<MergeSyntheticSection>(OutsecName, MS->Type, Flags, Alignment);
175 MergeSections.push_back(Syn);
176 I = std::prev(MergeSections.end());
181 (*I)->addSection(MS);
184 std::vector<InputSectionBase *> &V = InputSections;
185 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end());
188 template <class ELFT> static void combineEhFrameSections() {
189 for (InputSectionBase *&S : InputSections) {
190 EhInputSection *ES = dyn_cast<EhInputSection>(S);
191 if (!ES || !ES->Live)
194 In<ELFT>::EhFrame->addSection(ES);
198 std::vector<InputSectionBase *> &V = InputSections;
199 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end());
202 // The main function of the writer.
203 template <class ELFT> void Writer<ELFT>::run() {
204 // Create linker-synthesized sections such as .got or .plt.
205 // Such sections are of type input section.
206 createSyntheticSections();
207 combineMergableSections();
209 if (!Config->Relocatable)
210 combineEhFrameSections<ELFT>();
212 // We need to create some reserved symbols such as _end. Create them.
213 if (!Config->Relocatable)
214 addReservedSymbols();
216 // Create output sections.
217 Script->OutputSections = &OutputSections;
218 if (Script->Opt.HasSections) {
219 // If linker script contains SECTIONS commands, let it create sections.
220 Script->processCommands(Factory);
222 // Linker scripts may have left some input sections unassigned.
223 // Assign such sections using the default rule.
224 Script->addOrphanSections(Factory);
226 // If linker script does not contain SECTIONS commands, create
227 // output sections by default rules. We still need to give the
228 // linker script a chance to run, because it might contain
229 // non-SECTIONS commands such as ASSERT.
231 Script->processCommands(Factory);
234 if (Config->Discard != DiscardPolicy::All)
237 if (Config->CopyRelocs)
240 // Now that we have a complete set of output sections. This function
241 // completes section contents. For example, we need to add strings
242 // to the string table, and add entries to .got and .plt.
243 // finalizeSections does that.
248 if (Config->Relocatable) {
251 if (!Script->Opt.HasSections) {
252 fixSectionAlignments();
253 Script->fabricateDefaultCommands();
255 Script->synchronize();
256 Script->assignAddresses(Phdrs);
258 // Remove empty PT_LOAD to avoid causing the dynamic linker to try to mmap a
259 // 0 sized region. This has to be done late since only after assignAddresses
260 // we know the size of the sections.
263 if (!Config->OFormatBinary)
266 assignFileOffsetsBinary();
269 fixPredefinedSymbols();
272 // It does not make sense try to open the file if we have error already.
275 // Write the result down to a file.
279 if (!Config->OFormatBinary) {
283 writeSectionsBinary();
286 // Backfill .note.gnu.build-id section content. This is done at last
287 // because the content is usually a hash value of the entire output file.
292 // Handle -Map option.
293 writeMapFile<ELFT>(OutputSections);
297 if (auto EC = Buffer->commit())
298 error("failed to write to the output file: " + EC.message());
300 // Flush the output streams and exit immediately. A full shutdown
301 // is a good test that we are keeping track of all allocated memory,
302 // but actually freeing it is a waste of time in a regular linker run.
303 if (Config->ExitEarly)
307 // Initialize Out members.
308 template <class ELFT> void Writer<ELFT>::createSyntheticSections() {
309 // Initialize all pointers with NULL. This is needed because
310 // you can call lld::elf::main more than once as a library.
311 memset(&Out::First, 0, sizeof(Out));
313 auto Add = [](InputSectionBase *Sec) { InputSections.push_back(Sec); };
315 In<ELFT>::DynStrTab = make<StringTableSection>(".dynstr", true);
316 In<ELFT>::Dynamic = make<DynamicSection<ELFT>>();
317 In<ELFT>::RelaDyn = make<RelocationSection<ELFT>>(
318 Config->IsRela ? ".rela.dyn" : ".rel.dyn", Config->ZCombreloc);
319 In<ELFT>::ShStrTab = make<StringTableSection>(".shstrtab", false);
321 Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC);
322 Out::ElfHeader->Size = sizeof(Elf_Ehdr);
323 Out::ProgramHeaders = make<OutputSection>("", 0, SHF_ALLOC);
324 Out::ProgramHeaders->updateAlignment(Config->Wordsize);
326 if (needsInterpSection<ELFT>()) {
327 In<ELFT>::Interp = createInterpSection();
328 Add(In<ELFT>::Interp);
330 In<ELFT>::Interp = nullptr;
333 if (!Config->Relocatable)
334 Add(createCommentSection<ELFT>());
336 if (Config->Strip != StripPolicy::All) {
337 In<ELFT>::StrTab = make<StringTableSection>(".strtab", false);
338 In<ELFT>::SymTab = make<SymbolTableSection<ELFT>>(*In<ELFT>::StrTab);
341 if (Config->BuildId != BuildIdKind::None) {
342 In<ELFT>::BuildId = make<BuildIdSection>();
343 Add(In<ELFT>::BuildId);
346 In<ELFT>::Common = createCommonSection<ELFT>();
347 if (In<ELFT>::Common)
350 In<ELFT>::Bss = make<BssSection>(".bss");
352 In<ELFT>::BssRelRo = make<BssSection>(".bss.rel.ro");
353 Add(In<ELFT>::BssRelRo);
355 // Add MIPS-specific sections.
356 bool HasDynSymTab = !Symtab<ELFT>::X->getSharedFiles().empty() ||
357 Config->Pic || Config->ExportDynamic;
358 if (Config->EMachine == EM_MIPS) {
359 if (!Config->Shared && HasDynSymTab) {
360 In<ELFT>::MipsRldMap = make<MipsRldMapSection>();
361 Add(In<ELFT>::MipsRldMap);
363 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create())
365 if (auto *Sec = MipsOptionsSection<ELFT>::create())
367 if (auto *Sec = MipsReginfoSection<ELFT>::create())
372 In<ELFT>::DynSymTab = make<SymbolTableSection<ELFT>>(*In<ELFT>::DynStrTab);
373 Add(In<ELFT>::DynSymTab);
375 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>();
376 Add(In<ELFT>::VerSym);
378 if (!Config->VersionDefinitions.empty()) {
379 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>();
380 Add(In<ELFT>::VerDef);
383 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>();
384 Add(In<ELFT>::VerNeed);
386 if (Config->GnuHash) {
387 In<ELFT>::GnuHashTab = make<GnuHashTableSection<ELFT>>();
388 Add(In<ELFT>::GnuHashTab);
391 if (Config->SysvHash) {
392 In<ELFT>::HashTab = make<HashTableSection<ELFT>>();
393 Add(In<ELFT>::HashTab);
396 Add(In<ELFT>::Dynamic);
397 Add(In<ELFT>::DynStrTab);
398 Add(In<ELFT>::RelaDyn);
401 // Add .got. MIPS' .got is so different from the other archs,
402 // it has its own class.
403 if (Config->EMachine == EM_MIPS) {
404 In<ELFT>::MipsGot = make<MipsGotSection>();
405 Add(In<ELFT>::MipsGot);
407 In<ELFT>::Got = make<GotSection<ELFT>>();
411 In<ELFT>::GotPlt = make<GotPltSection>();
412 Add(In<ELFT>::GotPlt);
413 In<ELFT>::IgotPlt = make<IgotPltSection>();
414 Add(In<ELFT>::IgotPlt);
416 if (Config->GdbIndex) {
417 In<ELFT>::GdbIndex = make<GdbIndexSection>();
418 Add(In<ELFT>::GdbIndex);
421 // We always need to add rel[a].plt to output if it has entries.
422 // Even for static linking it can contain R_[*]_IRELATIVE relocations.
423 In<ELFT>::RelaPlt = make<RelocationSection<ELFT>>(
424 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/);
425 Add(In<ELFT>::RelaPlt);
427 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure
428 // that the IRelative relocations are processed last by the dynamic loader
429 In<ELFT>::RelaIplt = make<RelocationSection<ELFT>>(
430 (Config->EMachine == EM_ARM) ? ".rel.dyn" : In<ELFT>::RelaPlt->Name,
432 Add(In<ELFT>::RelaIplt);
434 In<ELFT>::Plt = make<PltSection>(Target->PltHeaderSize);
436 In<ELFT>::Iplt = make<PltSection>(0);
439 if (!Config->Relocatable) {
440 if (Config->EhFrameHdr) {
441 In<ELFT>::EhFrameHdr = make<EhFrameHeader<ELFT>>();
442 Add(In<ELFT>::EhFrameHdr);
444 In<ELFT>::EhFrame = make<EhFrameSection<ELFT>>();
445 Add(In<ELFT>::EhFrame);
448 if (In<ELFT>::SymTab)
449 Add(In<ELFT>::SymTab);
450 Add(In<ELFT>::ShStrTab);
451 if (In<ELFT>::StrTab)
452 Add(In<ELFT>::StrTab);
455 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName,
456 const SymbolBody &B) {
457 if (B.isFile() || B.isSection())
460 // If sym references a section in a discarded group, don't keep it.
461 if (Sec == &InputSection::Discarded)
464 if (Config->Discard == DiscardPolicy::None)
467 // In ELF assembly .L symbols are normally discarded by the assembler.
468 // If the assembler fails to do so, the linker discards them if
469 // * --discard-locals is used.
470 // * The symbol is in a SHF_MERGE section, which is normally the reason for
471 // the assembler keeping the .L symbol.
472 if (!SymName.startswith(".L") && !SymName.empty())
475 if (Config->Discard == DiscardPolicy::Locals)
478 return !Sec || !(Sec->Flags & SHF_MERGE);
481 static bool includeInSymtab(const SymbolBody &B) {
482 if (!B.isLocal() && !B.symbol()->IsUsedInRegularObj)
485 if (auto *D = dyn_cast<DefinedRegular>(&B)) {
486 // Always include absolute symbols.
487 SectionBase *Sec = D->Section;
490 if (auto *IS = dyn_cast<InputSectionBase>(Sec)) {
492 IS = cast<InputSectionBase>(Sec);
493 // Exclude symbols pointing to garbage-collected sections.
497 if (auto *S = dyn_cast<MergeInputSection>(Sec))
498 if (!S->getSectionPiece(D->Value)->Live)
504 // Local symbols are not in the linker's symbol table. This function scans
505 // each object file's symbol table to copy local symbols to the output.
506 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() {
507 if (!In<ELFT>::SymTab)
509 for (elf::ObjectFile<ELFT> *F : Symtab<ELFT>::X->getObjectFiles()) {
510 for (SymbolBody *B : F->getLocalSymbols()) {
513 ": broken object: getLocalSymbols returns a non-local symbol");
514 auto *DR = dyn_cast<DefinedRegular>(B);
516 // No reason to keep local undefined symbol in symtab.
519 if (!includeInSymtab(*B))
522 SectionBase *Sec = DR->Section;
523 if (!shouldKeepInSymtab(Sec, B->getName(), *B))
525 In<ELFT>::SymTab->addSymbol(B);
530 template <class ELFT> void Writer<ELFT>::addSectionSymbols() {
531 // Create one STT_SECTION symbol for each output section we might
532 // have a relocation with.
533 for (OutputSection *Sec : OutputSections) {
534 if (Sec->Sections.empty())
537 InputSection *IS = Sec->Sections[0];
538 if (isa<SyntheticSection>(IS) || IS->Type == SHT_REL ||
539 IS->Type == SHT_RELA)
543 make<DefinedRegular>("", /*IsLocal=*/true, /*StOther=*/0, STT_SECTION,
544 /*Value=*/0, /*Size=*/0, IS, nullptr);
545 In<ELFT>::SymTab->addSymbol(Sym);
549 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections that
550 // we would like to make sure appear is a specific order to maximize their
551 // coverage by a single signed 16-bit offset from the TOC base pointer.
552 // Conversely, the special .tocbss section should be first among all SHT_NOBITS
553 // sections. This will put it next to the loaded special PPC64 sections (and,
554 // thus, within reach of the TOC base pointer).
555 static int getPPC64SectionRank(StringRef SectionName) {
556 return StringSwitch<int>(SectionName)
558 .Case(".branch_lt", 2)
565 // All sections with SHF_MIPS_GPREL flag should be grouped together
566 // because data in these sections is addressable with a gp relative address.
567 static int getMipsSectionRank(const OutputSection *S) {
568 if ((S->Flags & SHF_MIPS_GPREL) == 0)
570 if (S->Name == ".got")
575 // Today's loaders have a feature to make segments read-only after
576 // processing dynamic relocations to enhance security. PT_GNU_RELRO
577 // is defined for that.
579 // This function returns true if a section needs to be put into a
580 // PT_GNU_RELRO segment.
581 template <class ELFT> bool elf::isRelroSection(const OutputSection *Sec) {
585 uint64_t Flags = Sec->Flags;
587 // Non-allocatable or non-writable sections don't need RELRO because
588 // they are not writable or not even mapped to memory in the first place.
589 // RELRO is for sections that are essentially read-only but need to
590 // be writable only at process startup to allow dynamic linker to
591 // apply relocations.
592 if (!(Flags & SHF_ALLOC) || !(Flags & SHF_WRITE))
595 // Once initialized, TLS data segments are used as data templates
596 // for a thread-local storage. For each new thread, runtime
597 // allocates memory for a TLS and copy templates there. No thread
598 // are supposed to use templates directly. Thus, it can be in RELRO.
602 // .init_array, .preinit_array and .fini_array contain pointers to
603 // functions that are executed on process startup or exit. These
604 // pointers are set by the static linker, and they are not expected
605 // to change at runtime. But if you are an attacker, you could do
606 // interesting things by manipulating pointers in .fini_array, for
607 // example. So they are put into RELRO.
608 uint32_t Type = Sec->Type;
609 if (Type == SHT_INIT_ARRAY || Type == SHT_FINI_ARRAY ||
610 Type == SHT_PREINIT_ARRAY)
613 // .got contains pointers to external symbols. They are resolved by
614 // the dynamic linker when a module is loaded into memory, and after
615 // that they are not expected to change. So, it can be in RELRO.
616 if (In<ELFT>::Got && Sec == In<ELFT>::Got->OutSec)
619 // .got.plt contains pointers to external function symbols. They are
620 // by default resolved lazily, so we usually cannot put it into RELRO.
621 // However, if "-z now" is given, the lazy symbol resolution is
622 // disabled, which enables us to put it into RELRO.
623 if (Sec == In<ELFT>::GotPlt->OutSec)
626 // .dynamic section contains data for the dynamic linker, and
627 // there's no need to write to it at runtime, so it's better to put
629 if (Sec == In<ELFT>::Dynamic->OutSec)
632 // .bss.rel.ro is used for copy relocations for read-only symbols.
633 // Since the dynamic linker needs to process copy relocations, the
634 // section cannot be read-only, but once initialized, they shouldn't
636 if (Sec == In<ELFT>::BssRelRo->OutSec)
639 // Sections with some special names are put into RELRO. This is a
640 // bit unfortunate because section names shouldn't be significant in
641 // ELF in spirit. But in reality many linker features depend on
642 // magic section names.
643 StringRef S = Sec->Name;
644 return S == ".data.rel.ro" || S == ".ctors" || S == ".dtors" || S == ".jcr" ||
645 S == ".eh_frame" || S == ".openbsd.randomdata";
648 template <class ELFT>
649 static bool compareSectionsNonScript(const OutputSection *A,
650 const OutputSection *B) {
651 // Put .interp first because some loaders want to see that section
652 // on the first page of the executable file when loaded into memory.
653 bool AIsInterp = A->Name == ".interp";
654 bool BIsInterp = B->Name == ".interp";
655 if (AIsInterp != BIsInterp)
658 // Allocatable sections go first to reduce the total PT_LOAD size and
659 // so debug info doesn't change addresses in actual code.
660 bool AIsAlloc = A->Flags & SHF_ALLOC;
661 bool BIsAlloc = B->Flags & SHF_ALLOC;
662 if (AIsAlloc != BIsAlloc)
665 // We don't have any special requirements for the relative order of two non
666 // allocatable sections.
670 // We want to put section specified by -T option first, so we
671 // can start assigning VA starting from them later.
672 auto AAddrSetI = Config->SectionStartMap.find(A->Name);
673 auto BAddrSetI = Config->SectionStartMap.find(B->Name);
674 bool AHasAddrSet = AAddrSetI != Config->SectionStartMap.end();
675 bool BHasAddrSet = BAddrSetI != Config->SectionStartMap.end();
676 if (AHasAddrSet != BHasAddrSet)
679 return AAddrSetI->second < BAddrSetI->second;
681 // We want the read only sections first so that they go in the PT_LOAD
682 // covering the program headers at the start of the file.
683 bool AIsWritable = A->Flags & SHF_WRITE;
684 bool BIsWritable = B->Flags & SHF_WRITE;
685 if (AIsWritable != BIsWritable)
688 if (!Config->SingleRoRx) {
689 // For a corresponding reason, put non exec sections first (the program
690 // header PT_LOAD is not executable).
691 // We only do that if we are not using linker scripts, since with linker
692 // scripts ro and rx sections are in the same PT_LOAD, so their relative
693 // order is not important. The same applies for -no-rosegment.
694 bool AIsExec = A->Flags & SHF_EXECINSTR;
695 bool BIsExec = B->Flags & SHF_EXECINSTR;
696 if (AIsExec != BIsExec)
700 // If we got here we know that both A and B are in the same PT_LOAD.
702 bool AIsTls = A->Flags & SHF_TLS;
703 bool BIsTls = B->Flags & SHF_TLS;
704 bool AIsNoBits = A->Type == SHT_NOBITS;
705 bool BIsNoBits = B->Type == SHT_NOBITS;
707 // The first requirement we have is to put (non-TLS) nobits sections last. The
708 // reason is that the only thing the dynamic linker will see about them is a
709 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the
710 // PT_LOAD, so that has to correspond to the nobits sections.
711 bool AIsNonTlsNoBits = AIsNoBits && !AIsTls;
712 bool BIsNonTlsNoBits = BIsNoBits && !BIsTls;
713 if (AIsNonTlsNoBits != BIsNonTlsNoBits)
714 return BIsNonTlsNoBits;
716 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo
717 // sections after r/w ones, so that the RelRo sections are contiguous.
718 bool AIsRelRo = isRelroSection<ELFT>(A);
719 bool BIsRelRo = isRelroSection<ELFT>(B);
720 if (AIsRelRo != BIsRelRo)
721 return AIsNonTlsNoBits ? AIsRelRo : BIsRelRo;
723 // The TLS initialization block needs to be a single contiguous block in a R/W
724 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W
725 // sections. The TLS NOBITS sections are placed here as they don't take up
726 // virtual address space in the PT_LOAD.
727 if (AIsTls != BIsTls)
730 // Within the TLS initialization block, the non-nobits sections need to appear
732 if (AIsNoBits != BIsNoBits)
735 // Some architectures have additional ordering restrictions for sections
736 // within the same PT_LOAD.
737 if (Config->EMachine == EM_PPC64)
738 return getPPC64SectionRank(A->Name) < getPPC64SectionRank(B->Name);
739 if (Config->EMachine == EM_MIPS)
740 return getMipsSectionRank(A) < getMipsSectionRank(B);
745 // Output section ordering is determined by this function.
746 template <class ELFT>
747 static bool compareSections(const OutputSection *A, const OutputSection *B) {
748 // For now, put sections mentioned in a linker script
749 // first. Sections not on linker script will have a SectionIndex of
751 int AIndex = A->SectionIndex;
752 int BIndex = B->SectionIndex;
753 if (AIndex != BIndex)
754 return AIndex < BIndex;
756 return compareSectionsNonScript<ELFT>(A, B);
759 // Program header entry
760 PhdrEntry::PhdrEntry(unsigned Type, unsigned Flags) {
765 void PhdrEntry::add(OutputSection *Sec) {
769 p_align = std::max(p_align, Sec->Alignment);
770 if (p_type == PT_LOAD)
771 Sec->FirstInPtLoad = First;
774 template <class ELFT>
775 static Symbol *addRegular(StringRef Name, SectionBase *Sec, uint64_t Value,
776 uint8_t StOther = STV_HIDDEN,
777 uint8_t Binding = STB_WEAK) {
778 // The linker generated symbols are added as STB_WEAK to allow user defined
779 // ones to override them.
780 return Symtab<ELFT>::X->addRegular(Name, StOther, STT_NOTYPE, Value,
781 /*Size=*/0, Binding, Sec,
785 template <class ELFT>
786 static DefinedRegular *
787 addOptionalRegular(StringRef Name, SectionBase *Sec, uint64_t Val,
788 uint8_t StOther = STV_HIDDEN, uint8_t Binding = STB_GLOBAL) {
789 SymbolBody *S = Symtab<ELFT>::X->find(Name);
792 if (S->isInCurrentDSO())
794 return cast<DefinedRegular>(
795 addRegular<ELFT>(Name, Sec, Val, StOther, Binding)->body());
798 // The beginning and the ending of .rel[a].plt section are marked
799 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked
800 // executable. The runtime needs these symbols in order to resolve
801 // all IRELATIVE relocs on startup. For dynamic executables, we don't
802 // need these symbols, since IRELATIVE relocs are resolved through GOT
803 // and PLT. For details, see http://www.airs.com/blog/archives/403.
804 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() {
805 if (In<ELFT>::DynSymTab)
807 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start";
808 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, 0, STV_HIDDEN, STB_WEAK);
810 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end";
811 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, -1, STV_HIDDEN, STB_WEAK);
814 // The linker is expected to define some symbols depending on
815 // the linking result. This function defines such symbols.
816 template <class ELFT> void Writer<ELFT>::addReservedSymbols() {
817 if (Config->EMachine == EM_MIPS) {
818 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer
819 // so that it points to an absolute address which by default is relative
820 // to GOT. Default offset is 0x7ff0.
821 // See "Global Data Symbols" in Chapter 6 in the following document:
822 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf
823 ElfSym::MipsGp = Symtab<ELFT>::X->addAbsolute("_gp", STV_HIDDEN, STB_LOCAL);
825 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between
826 // start of function and 'gp' pointer into GOT.
827 if (Symtab<ELFT>::X->find("_gp_disp"))
829 Symtab<ELFT>::X->addAbsolute("_gp_disp", STV_HIDDEN, STB_LOCAL);
831 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp'
832 // pointer. This symbol is used in the code generated by .cpload pseudo-op
833 // in case of using -mno-shared option.
834 // https://sourceware.org/ml/binutils/2004-12/msg00094.html
835 if (Symtab<ELFT>::X->find("__gnu_local_gp"))
836 ElfSym::MipsLocalGp =
837 Symtab<ELFT>::X->addAbsolute("__gnu_local_gp", STV_HIDDEN, STB_LOCAL);
840 // In the assembly for 32 bit x86 the _GLOBAL_OFFSET_TABLE_ symbol
841 // is magical and is used to produce a R_386_GOTPC relocation.
842 // The R_386_GOTPC relocation value doesn't actually depend on the
843 // symbol value, so it could use an index of STN_UNDEF which, according
844 // to the spec, means the symbol value is 0.
845 // Unfortunately both gas and MC keep the _GLOBAL_OFFSET_TABLE_ symbol in
847 // The situation is even stranger on x86_64 where the assembly doesn't
848 // need the magical symbol, but gas still puts _GLOBAL_OFFSET_TABLE_ as
849 // an undefined symbol in the .o files.
850 // Given that the symbol is effectively unused, we just create a dummy
851 // hidden one to avoid the undefined symbol error.
852 Symtab<ELFT>::X->addIgnored("_GLOBAL_OFFSET_TABLE_");
854 // __tls_get_addr is defined by the dynamic linker for dynamic ELFs. For
855 // static linking the linker is required to optimize away any references to
856 // __tls_get_addr, so it's not defined anywhere. Create a hidden definition
857 // to avoid the undefined symbol error.
858 if (!In<ELFT>::DynSymTab)
859 Symtab<ELFT>::X->addIgnored("__tls_get_addr");
861 // If linker script do layout we do not need to create any standart symbols.
862 if (Script->Opt.HasSections)
865 // __ehdr_start is the location of ELF file headers.
866 addOptionalRegular<ELFT>("__ehdr_start", Out::ElfHeader, 0, STV_HIDDEN);
868 auto Add = [](StringRef S) {
869 return addOptionalRegular<ELFT>(S, Out::ElfHeader, 0, STV_DEFAULT);
872 ElfSym::Bss = Add("__bss_start");
873 ElfSym::End1 = Add("end");
874 ElfSym::End2 = Add("_end");
875 ElfSym::Etext1 = Add("etext");
876 ElfSym::Etext2 = Add("_etext");
877 ElfSym::Edata1 = Add("edata");
878 ElfSym::Edata2 = Add("_edata");
881 // Sort input sections by section name suffixes for
882 // __attribute__((init_priority(N))).
883 static void sortInitFini(OutputSection *S) {
885 reinterpret_cast<OutputSection *>(S)->sortInitFini();
888 // Sort input sections by the special rule for .ctors and .dtors.
889 static void sortCtorsDtors(OutputSection *S) {
891 reinterpret_cast<OutputSection *>(S)->sortCtorsDtors();
894 // Sort input sections using the list provided by --symbol-ordering-file.
895 template <class ELFT>
896 static void sortBySymbolsOrder(ArrayRef<OutputSection *> OutputSections) {
897 if (Config->SymbolOrderingFile.empty())
900 // Build a map from symbols to their priorities. Symbols that didn't
901 // appear in the symbol ordering file have the lowest priority 0.
902 // All explicitly mentioned symbols have negative (higher) priorities.
903 DenseMap<StringRef, int> SymbolOrder;
904 int Priority = -Config->SymbolOrderingFile.size();
905 for (StringRef S : Config->SymbolOrderingFile)
906 SymbolOrder.insert({S, Priority++});
908 // Build a map from sections to their priorities.
909 DenseMap<SectionBase *, int> SectionOrder;
910 for (elf::ObjectFile<ELFT> *File : Symtab<ELFT>::X->getObjectFiles()) {
911 for (SymbolBody *Body : File->getSymbols()) {
912 auto *D = dyn_cast<DefinedRegular>(Body);
913 if (!D || !D->Section)
915 int &Priority = SectionOrder[D->Section];
916 Priority = std::min(Priority, SymbolOrder.lookup(D->getName()));
920 // Sort sections by priority.
921 for (OutputSection *Base : OutputSections)
922 if (auto *Sec = dyn_cast<OutputSection>(Base))
923 Sec->sort([&](InputSectionBase *S) { return SectionOrder.lookup(S); });
926 template <class ELFT>
927 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) {
928 for (InputSectionBase *IS : InputSections) {
931 // Scan all relocations. Each relocation goes through a series
932 // of tests to determine if it needs special treatment, such as
933 // creating GOT, PLT, copy relocations, etc.
934 // Note that relocations for non-alloc sections are directly
935 // processed by InputSection::relocateNonAlloc.
936 if (!(IS->Flags & SHF_ALLOC))
938 if (isa<InputSection>(IS) || isa<EhInputSection>(IS))
942 if (!Config->Relocatable) {
943 for (EhInputSection *ES : In<ELFT>::EhFrame->Sections)
948 template <class ELFT> void Writer<ELFT>::createSections() {
949 for (InputSectionBase *IS : InputSections)
951 Factory.addInputSec(IS, getOutputSectionName(IS->Name));
953 sortBySymbolsOrder<ELFT>(OutputSections);
954 sortInitFini(findSection(".init_array"));
955 sortInitFini(findSection(".fini_array"));
956 sortCtorsDtors(findSection(".ctors"));
957 sortCtorsDtors(findSection(".dtors"));
959 for (OutputSection *Sec : OutputSections)
960 Sec->assignOffsets();
963 static bool canSharePtLoad(const OutputSection &S1, const OutputSection &S2) {
964 if (!(S1.Flags & SHF_ALLOC) || !(S2.Flags & SHF_ALLOC))
967 bool S1IsWrite = S1.Flags & SHF_WRITE;
968 bool S2IsWrite = S2.Flags & SHF_WRITE;
969 if (S1IsWrite != S2IsWrite)
973 return true; // RO and RX share a PT_LOAD with linker scripts.
974 return (S1.Flags & SHF_EXECINSTR) == (S2.Flags & SHF_EXECINSTR);
977 template <class ELFT> void Writer<ELFT>::sortSections() {
978 // Don't sort if using -r. It is not necessary and we want to preserve the
979 // relative order for SHF_LINK_ORDER sections.
980 if (Config->Relocatable)
982 if (!Script->Opt.HasSections) {
983 std::stable_sort(OutputSections.begin(), OutputSections.end(),
984 compareSectionsNonScript<ELFT>);
987 Script->adjustSectionsBeforeSorting();
989 // The order of the sections in the script is arbitrary and may not agree with
990 // compareSectionsNonScript. This means that we cannot easily define a
991 // strict weak ordering. To see why, consider a comparison of a section in the
992 // script and one not in the script. We have a two simple options:
993 // * Make them equivalent (a is not less than b, and b is not less than a).
994 // The problem is then that equivalence has to be transitive and we can
995 // have sections a, b and c with only b in a script and a less than c
996 // which breaks this property.
997 // * Use compareSectionsNonScript. Given that the script order doesn't have
998 // to match, we can end up with sections a, b, c, d where b and c are in the
999 // script and c is compareSectionsNonScript less than b. In which case d
1000 // can be equivalent to c, a to b and d < a. As a concrete example:
1001 // .a (rx) # not in script
1002 // .b (rx) # in script
1003 // .c (ro) # in script
1004 // .d (ro) # not in script
1006 // The way we define an order then is:
1007 // * First put script sections at the start and sort the script and
1008 // non-script sections independently.
1009 // * Move each non-script section to its preferred position. We try
1010 // to put each section in the last position where it it can share
1013 std::stable_sort(OutputSections.begin(), OutputSections.end(),
1014 compareSections<ELFT>);
1016 auto I = OutputSections.begin();
1017 auto E = OutputSections.end();
1019 std::find_if(OutputSections.begin(), E,
1020 [](OutputSection *S) { return S->SectionIndex == INT_MAX; });
1021 while (NonScriptI != E) {
1022 auto BestPos = std::max_element(
1023 I, NonScriptI, [&](OutputSection *&A, OutputSection *&B) {
1024 bool ACanSharePtLoad = canSharePtLoad(**NonScriptI, *A);
1025 bool BCanSharePtLoad = canSharePtLoad(**NonScriptI, *B);
1026 if (ACanSharePtLoad != BCanSharePtLoad)
1027 return BCanSharePtLoad;
1029 bool ACmp = compareSectionsNonScript<ELFT>(*NonScriptI, A);
1030 bool BCmp = compareSectionsNonScript<ELFT>(*NonScriptI, B);
1032 return BCmp; // FIXME: missing test
1034 size_t PosA = &A - &OutputSections[0];
1035 size_t PosB = &B - &OutputSections[0];
1036 return ACmp ? PosA > PosB : PosA < PosB;
1039 // max_element only returns NonScriptI if the range is empty. If the range
1040 // is not empty we should consider moving the the element forward one
1042 if (BestPos != NonScriptI &&
1043 !compareSectionsNonScript<ELFT>(*NonScriptI, *BestPos))
1045 std::rotate(BestPos, NonScriptI, NonScriptI + 1);
1049 Script->adjustSectionsAfterSorting();
1052 static void applySynthetic(const std::vector<SyntheticSection *> &Sections,
1053 std::function<void(SyntheticSection *)> Fn) {
1054 for (SyntheticSection *SS : Sections)
1055 if (SS && SS->OutSec && !SS->empty()) {
1057 SS->OutSec->assignOffsets();
1061 // We need to add input synthetic sections early in createSyntheticSections()
1062 // to make them visible from linkescript side. But not all sections are always
1063 // required to be in output. For example we don't need dynamic section content
1064 // sometimes. This function filters out such unused sections from the output.
1065 static void removeUnusedSyntheticSections(std::vector<OutputSection *> &V) {
1066 // All input synthetic sections that can be empty are placed after
1067 // all regular ones. We iterate over them all and exit at first
1069 for (InputSectionBase *S : llvm::reverse(InputSections)) {
1070 SyntheticSection *SS = dyn_cast<SyntheticSection>(S);
1073 if (!SS->empty() || !SS->OutSec)
1076 SS->OutSec->Sections.erase(std::find(SS->OutSec->Sections.begin(),
1077 SS->OutSec->Sections.end(), SS));
1079 // If there are no other sections in the output section, remove it from the
1081 if (SS->OutSec->Sections.empty())
1082 V.erase(std::find(V.begin(), V.end(), SS->OutSec));
1086 // Create output section objects and add them to OutputSections.
1087 template <class ELFT> void Writer<ELFT>::finalizeSections() {
1088 Out::DebugInfo = findSection(".debug_info");
1089 Out::PreinitArray = findSection(".preinit_array");
1090 Out::InitArray = findSection(".init_array");
1091 Out::FiniArray = findSection(".fini_array");
1093 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop
1094 // symbols for sections, so that the runtime can get the start and end
1095 // addresses of each section by section name. Add such symbols.
1096 if (!Config->Relocatable) {
1097 addStartEndSymbols();
1098 for (OutputSection *Sec : OutputSections)
1099 addStartStopSymbols(Sec);
1102 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type.
1103 // It should be okay as no one seems to care about the type.
1104 // Even the author of gold doesn't remember why gold behaves that way.
1105 // https://sourceware.org/ml/binutils/2002-03/msg00360.html
1106 if (In<ELFT>::DynSymTab)
1107 addRegular<ELFT>("_DYNAMIC", In<ELFT>::Dynamic, 0);
1109 // Define __rel[a]_iplt_{start,end} symbols if needed.
1110 addRelIpltSymbols();
1112 // This responsible for splitting up .eh_frame section into
1113 // pieces. The relocation scan uses those pieces, so this has to be
1115 applySynthetic({In<ELFT>::EhFrame},
1116 [](SyntheticSection *SS) { SS->finalizeContents(); });
1118 // Scan relocations. This must be done after every symbol is declared so that
1119 // we can correctly decide if a dynamic relocation is needed.
1120 forEachRelSec(scanRelocations<ELFT>);
1122 if (In<ELFT>::Plt && !In<ELFT>::Plt->empty())
1123 In<ELFT>::Plt->addSymbols();
1124 if (In<ELFT>::Iplt && !In<ELFT>::Iplt->empty())
1125 In<ELFT>::Iplt->addSymbols();
1127 // Now that we have defined all possible global symbols including linker-
1128 // synthesized ones. Visit all symbols to give the finishing touches.
1129 for (Symbol *S : Symtab<ELFT>::X->getSymbols()) {
1130 SymbolBody *Body = S->body();
1132 if (!includeInSymtab(*Body))
1134 if (In<ELFT>::SymTab)
1135 In<ELFT>::SymTab->addSymbol(Body);
1137 if (In<ELFT>::DynSymTab && S->includeInDynsym()) {
1138 In<ELFT>::DynSymTab->addSymbol(Body);
1139 if (auto *SS = dyn_cast<SharedSymbol>(Body))
1140 if (cast<SharedFile<ELFT>>(SS->File)->isNeeded())
1141 In<ELFT>::VerNeed->addSymbol(SS);
1145 // Do not proceed if there was an undefined symbol.
1149 // So far we have added sections from input object files.
1150 // This function adds linker-created Out::* sections.
1151 addPredefinedSections();
1152 removeUnusedSyntheticSections(OutputSections);
1156 // This is a bit of a hack. A value of 0 means undef, so we set it
1157 // to 1 t make __ehdr_start defined. The section number is not
1158 // particularly relevant.
1159 Out::ElfHeader->SectionIndex = 1;
1162 for (OutputSection *Sec : OutputSections) {
1163 Sec->SectionIndex = I++;
1164 Sec->ShName = In<ELFT>::ShStrTab->addString(Sec->Name);
1167 // Binary and relocatable output does not have PHDRS.
1168 // The headers have to be created before finalize as that can influence the
1169 // image base and the dynamic section on mips includes the image base.
1170 if (!Config->Relocatable && !Config->OFormatBinary) {
1171 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs();
1172 addPtArmExid(Phdrs);
1173 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size();
1176 // Dynamic section must be the last one in this list and dynamic
1177 // symbol table section (DynSymTab) must be the first one.
1178 applySynthetic({In<ELFT>::DynSymTab, In<ELFT>::Bss, In<ELFT>::BssRelRo,
1179 In<ELFT>::GnuHashTab, In<ELFT>::HashTab, In<ELFT>::SymTab,
1180 In<ELFT>::ShStrTab, In<ELFT>::StrTab, In<ELFT>::VerDef,
1181 In<ELFT>::DynStrTab, In<ELFT>::GdbIndex, In<ELFT>::Got,
1182 In<ELFT>::MipsGot, In<ELFT>::IgotPlt, In<ELFT>::GotPlt,
1183 In<ELFT>::RelaDyn, In<ELFT>::RelaIplt, In<ELFT>::RelaPlt,
1184 In<ELFT>::Plt, In<ELFT>::Iplt, In<ELFT>::Plt,
1185 In<ELFT>::EhFrameHdr, In<ELFT>::VerSym, In<ELFT>::VerNeed,
1187 [](SyntheticSection *SS) { SS->finalizeContents(); });
1189 // Some architectures use small displacements for jump instructions.
1190 // It is linker's responsibility to create thunks containing long
1191 // jump instructions if jump targets are too far. Create thunks.
1192 if (Target->NeedsThunks) {
1193 // FIXME: only ARM Interworking and Mips LA25 Thunks are implemented,
1195 // do not require address information. To support range extension Thunks
1196 // we need to assign addresses so that we can tell if jump instructions
1197 // are out of range. This will need to turn into a loop that converges
1198 // when no more Thunks are added
1199 ThunkCreator<ELFT> TC;
1200 if (TC.createThunks(OutputSections))
1201 applySynthetic({In<ELFT>::MipsGot},
1202 [](SyntheticSection *SS) { SS->updateAllocSize(); });
1204 // Fill other section headers. The dynamic table is finalized
1205 // at the end because some tags like RELSZ depend on result
1206 // of finalizing other sections.
1207 for (OutputSection *Sec : OutputSections)
1208 Sec->finalize<ELFT>();
1210 // If -compressed-debug-sections is specified, we need to compress
1211 // .debug_* sections. Do it right now because it changes the size of
1213 parallelForEach(OutputSections.begin(), OutputSections.end(),
1214 [](OutputSection *S) { S->maybeCompress<ELFT>(); });
1216 // createThunks may have added local symbols to the static symbol table
1217 applySynthetic({In<ELFT>::SymTab, In<ELFT>::ShStrTab, In<ELFT>::StrTab},
1218 [](SyntheticSection *SS) { SS->postThunkContents(); });
1221 template <class ELFT> void Writer<ELFT>::addPredefinedSections() {
1222 // ARM ABI requires .ARM.exidx to be terminated by some piece of data.
1223 // We have the terminater synthetic section class. Add that at the end.
1224 auto *OS = dyn_cast_or_null<OutputSection>(findSection(".ARM.exidx"));
1225 if (OS && !OS->Sections.empty() && !Config->Relocatable)
1226 OS->addSection(make<ARMExidxSentinelSection>());
1229 // The linker is expected to define SECNAME_start and SECNAME_end
1230 // symbols for a few sections. This function defines them.
1231 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() {
1232 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) {
1233 // These symbols resolve to the image base if the section does not exist.
1234 // A special value -1 indicates end of the section.
1236 addOptionalRegular<ELFT>(Start, OS, 0);
1237 addOptionalRegular<ELFT>(End, OS, -1);
1240 OS = Out::ElfHeader;
1241 addOptionalRegular<ELFT>(Start, OS, 0);
1242 addOptionalRegular<ELFT>(End, OS, 0);
1246 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray);
1247 Define("__init_array_start", "__init_array_end", Out::InitArray);
1248 Define("__fini_array_start", "__fini_array_end", Out::FiniArray);
1250 if (OutputSection *Sec = findSection(".ARM.exidx"))
1251 Define("__exidx_start", "__exidx_end", Sec);
1254 // If a section name is valid as a C identifier (which is rare because of
1255 // the leading '.'), linkers are expected to define __start_<secname> and
1256 // __stop_<secname> symbols. They are at beginning and end of the section,
1257 // respectively. This is not requested by the ELF standard, but GNU ld and
1258 // gold provide the feature, and used by many programs.
1259 template <class ELFT>
1260 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) {
1261 StringRef S = Sec->Name;
1262 if (!isValidCIdentifier(S))
1264 addOptionalRegular<ELFT>(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT);
1265 addOptionalRegular<ELFT>(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT);
1268 template <class ELFT> OutputSection *Writer<ELFT>::findSection(StringRef Name) {
1269 for (OutputSection *Sec : OutputSections)
1270 if (Sec->Name == Name)
1275 static bool needsPtLoad(OutputSection *Sec) {
1276 if (!(Sec->Flags & SHF_ALLOC))
1279 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is
1280 // responsible for allocating space for them, not the PT_LOAD that
1281 // contains the TLS initialization image.
1282 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS)
1287 // Linker scripts are responsible for aligning addresses. Unfortunately, most
1288 // linker scripts are designed for creating two PT_LOADs only, one RX and one
1289 // RW. This means that there is no alignment in the RO to RX transition and we
1290 // cannot create a PT_LOAD there.
1291 static uint64_t computeFlags(uint64_t Flags) {
1293 return PF_R | PF_W | PF_X;
1294 if (Config->SingleRoRx && !(Flags & PF_W))
1295 return Flags | PF_X;
1299 // Decide which program headers to create and which sections to include in each
1301 template <class ELFT> std::vector<PhdrEntry> Writer<ELFT>::createPhdrs() {
1302 std::vector<PhdrEntry> Ret;
1303 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * {
1304 Ret.emplace_back(Type, Flags);
1308 // The first phdr entry is PT_PHDR which describes the program header itself.
1309 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders);
1311 // PT_INTERP must be the second entry if exists.
1312 if (OutputSection *Sec = findSection(".interp"))
1313 AddHdr(PT_INTERP, Sec->getPhdrFlags())->add(Sec);
1315 // Add the first PT_LOAD segment for regular output sections.
1316 uint64_t Flags = computeFlags(PF_R);
1317 PhdrEntry *Load = AddHdr(PT_LOAD, Flags);
1319 // Add the headers. We will remove them if they don't fit.
1320 Load->add(Out::ElfHeader);
1321 Load->add(Out::ProgramHeaders);
1323 for (OutputSection *Sec : OutputSections) {
1324 if (!(Sec->Flags & SHF_ALLOC))
1326 if (!needsPtLoad(Sec))
1329 // Segments are contiguous memory regions that has the same attributes
1330 // (e.g. executable or writable). There is one phdr for each segment.
1331 // Therefore, we need to create a new phdr when the next section has
1332 // different flags or is loaded at a discontiguous address using AT linker
1334 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags());
1335 if (Script->hasLMA(Sec->Name) || Flags != NewFlags) {
1336 Load = AddHdr(PT_LOAD, NewFlags);
1343 // Add a TLS segment if any.
1344 PhdrEntry TlsHdr(PT_TLS, PF_R);
1345 for (OutputSection *Sec : OutputSections)
1346 if (Sec->Flags & SHF_TLS)
1349 Ret.push_back(std::move(TlsHdr));
1351 // Add an entry for .dynamic.
1352 if (In<ELFT>::DynSymTab)
1353 AddHdr(PT_DYNAMIC, In<ELFT>::Dynamic->OutSec->getPhdrFlags())
1354 ->add(In<ELFT>::Dynamic->OutSec);
1356 // PT_GNU_RELRO includes all sections that should be marked as
1357 // read-only by dynamic linker after proccessing relocations.
1358 PhdrEntry RelRo(PT_GNU_RELRO, PF_R);
1359 for (OutputSection *Sec : OutputSections)
1360 if (needsPtLoad(Sec) && isRelroSection<ELFT>(Sec))
1363 Ret.push_back(std::move(RelRo));
1365 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr.
1366 if (!In<ELFT>::EhFrame->empty() && In<ELFT>::EhFrameHdr &&
1367 In<ELFT>::EhFrame->OutSec && In<ELFT>::EhFrameHdr->OutSec)
1368 AddHdr(PT_GNU_EH_FRAME, In<ELFT>::EhFrameHdr->OutSec->getPhdrFlags())
1369 ->add(In<ELFT>::EhFrameHdr->OutSec);
1371 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes
1372 // the dynamic linker fill the segment with random data.
1373 if (OutputSection *Sec = findSection(".openbsd.randomdata"))
1374 AddHdr(PT_OPENBSD_RANDOMIZE, Sec->getPhdrFlags())->add(Sec);
1376 // PT_GNU_STACK is a special section to tell the loader to make the
1377 // pages for the stack non-executable. If you really want an executable
1378 // stack, you can pass -z execstack, but that's not recommended for
1379 // security reasons.
1381 if (Config->ZExecstack)
1382 Perm = PF_R | PF_W | PF_X;
1385 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize;
1387 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable
1388 // is expected to perform W^X violations, such as calling mprotect(2) or
1389 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on
1391 if (Config->ZWxneeded)
1392 AddHdr(PT_OPENBSD_WXNEEDED, PF_X);
1394 // Create one PT_NOTE per a group of contiguous .note sections.
1395 PhdrEntry *Note = nullptr;
1396 for (OutputSection *Sec : OutputSections) {
1397 if (Sec->Type == SHT_NOTE) {
1398 if (!Note || Script->hasLMA(Sec->Name))
1399 Note = AddHdr(PT_NOTE, PF_R);
1408 template <class ELFT>
1409 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry> &Phdrs) {
1410 if (Config->EMachine != EM_ARM)
1412 auto I = std::find_if(
1413 OutputSections.begin(), OutputSections.end(),
1414 [](OutputSection *Sec) { return Sec->Type == SHT_ARM_EXIDX; });
1415 if (I == OutputSections.end())
1418 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME
1419 PhdrEntry ARMExidx(PT_ARM_EXIDX, PF_R);
1421 Phdrs.push_back(ARMExidx);
1424 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the
1425 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic
1426 // linker can set the permissions.
1427 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() {
1428 for (const PhdrEntry &P : Phdrs)
1429 if (P.p_type == PT_LOAD && P.First)
1430 P.First->PageAlign = true;
1432 for (const PhdrEntry &P : Phdrs) {
1433 if (P.p_type != PT_GNU_RELRO)
1436 P.First->PageAlign = true;
1437 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we
1438 // have to align it to a page.
1439 auto End = OutputSections.end();
1440 auto I = std::find(OutputSections.begin(), End, P.Last);
1441 if (I == End || (I + 1) == End)
1443 OutputSection *Sec = *(I + 1);
1444 if (needsPtLoad(Sec))
1445 Sec->PageAlign = true;
1449 // Adjusts the file alignment for a given output section and returns
1450 // its new file offset. The file offset must be the same with its
1451 // virtual address (modulo the page size) so that the loader can load
1452 // executables without any address adjustment.
1453 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Sec) {
1454 OutputSection *First = Sec->FirstInPtLoad;
1455 // If the section is not in a PT_LOAD, we just have to align it.
1457 return alignTo(Off, Sec->Alignment);
1459 // The first section in a PT_LOAD has to have congruent offset and address
1460 // module the page size.
1462 return alignTo(Off, Config->MaxPageSize, Sec->Addr);
1464 // If two sections share the same PT_LOAD the file offset is calculated
1465 // using this formula: Off2 = Off1 + (VA2 - VA1).
1466 return First->Offset + Sec->Addr - First->Addr;
1469 static uint64_t setOffset(OutputSection *Sec, uint64_t Off) {
1470 if (Sec->Type == SHT_NOBITS) {
1475 Off = getFileAlignment(Off, Sec);
1477 return Off + Sec->Size;
1480 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() {
1482 for (OutputSection *Sec : OutputSections)
1483 if (Sec->Flags & SHF_ALLOC)
1484 Off = setOffset(Sec, Off);
1485 FileSize = alignTo(Off, Config->Wordsize);
1488 // Assign file offsets to output sections.
1489 template <class ELFT> void Writer<ELFT>::assignFileOffsets() {
1491 Off = setOffset(Out::ElfHeader, Off);
1492 Off = setOffset(Out::ProgramHeaders, Off);
1494 for (OutputSection *Sec : OutputSections)
1495 Off = setOffset(Sec, Off);
1497 SectionHeaderOff = alignTo(Off, Config->Wordsize);
1498 FileSize = SectionHeaderOff + (OutputSections.size() + 1) * sizeof(Elf_Shdr);
1501 // Finalize the program headers. We call this function after we assign
1502 // file offsets and VAs to all sections.
1503 template <class ELFT> void Writer<ELFT>::setPhdrs() {
1504 for (PhdrEntry &P : Phdrs) {
1505 OutputSection *First = P.First;
1506 OutputSection *Last = P.Last;
1508 P.p_filesz = Last->Offset - First->Offset;
1509 if (Last->Type != SHT_NOBITS)
1510 P.p_filesz += Last->Size;
1511 P.p_memsz = Last->Addr + Last->Size - First->Addr;
1512 P.p_offset = First->Offset;
1513 P.p_vaddr = First->Addr;
1515 P.p_paddr = First->getLMA();
1517 if (P.p_type == PT_LOAD)
1518 P.p_align = Config->MaxPageSize;
1519 else if (P.p_type == PT_GNU_RELRO)
1522 // The TLS pointer goes after PT_TLS. At least glibc will align it,
1523 // so round up the size to make sure the offsets are correct.
1524 if (P.p_type == PT_TLS) {
1527 P.p_memsz = alignTo(P.p_memsz, P.p_align);
1532 // The entry point address is chosen in the following ways.
1534 // 1. the '-e' entry command-line option;
1535 // 2. the ENTRY(symbol) command in a linker control script;
1536 // 3. the value of the symbol start, if present;
1537 // 4. the address of the first byte of the .text section, if present;
1538 // 5. the address 0.
1539 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() {
1540 // Case 1, 2 or 3. As a special case, if the symbol is actually
1541 // a number, we'll use that number as an address.
1542 if (SymbolBody *B = Symtab<ELFT>::X->find(Config->Entry))
1545 if (!Config->Entry.getAsInteger(0, Addr))
1549 if (OutputSection *Sec = findSection(".text")) {
1550 if (Config->WarnMissingEntry)
1551 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" +
1552 utohexstr(Sec->Addr));
1557 if (Config->WarnMissingEntry)
1558 warn("cannot find entry symbol " + Config->Entry +
1559 "; not setting start address");
1563 static uint16_t getELFType() {
1566 if (Config->Relocatable)
1571 // This function is called after we have assigned address and size
1572 // to each section. This function fixes some predefined
1573 // symbol values that depend on section address and size.
1574 template <class ELFT> void Writer<ELFT>::fixPredefinedSymbols() {
1575 auto Set = [](DefinedRegular *S1, DefinedRegular *S2, OutputSection *Sec,
1587 // _etext is the first location after the last read-only loadable segment.
1588 // _edata is the first location after the last read-write loadable segment.
1589 // _end is the first location after the uninitialized data region.
1590 PhdrEntry *Last = nullptr;
1591 PhdrEntry *LastRO = nullptr;
1592 PhdrEntry *LastRW = nullptr;
1593 for (PhdrEntry &P : Phdrs) {
1594 if (P.p_type != PT_LOAD)
1597 if (P.p_flags & PF_W)
1603 Set(ElfSym::End1, ElfSym::End2, Last->First, Last->p_memsz);
1605 Set(ElfSym::Etext1, ElfSym::Etext2, LastRO->First, LastRO->p_filesz);
1607 Set(ElfSym::Edata1, ElfSym::Edata2, LastRW->First, LastRW->p_filesz);
1610 ElfSym::Bss->Section = findSection(".bss");
1612 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should
1613 // be equal to the _gp symbol's value.
1614 if (Config->EMachine == EM_MIPS) {
1615 if (!ElfSym::MipsGp->Value) {
1616 // Find GP-relative section with the lowest address
1617 // and use this address to calculate default _gp value.
1619 for (const OutputSection *OS : OutputSections)
1620 if ((OS->Flags & SHF_MIPS_GPREL) && OS->Addr < Gp)
1622 if (Gp != (uint64_t)-1)
1623 ElfSym::MipsGp->Value = Gp + 0x7ff0;
1628 template <class ELFT> void Writer<ELFT>::writeHeader() {
1629 uint8_t *Buf = Buffer->getBufferStart();
1630 memcpy(Buf, "\177ELF", 4);
1632 // Write the ELF header.
1633 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf);
1634 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32;
1635 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB;
1636 EHdr->e_ident[EI_VERSION] = EV_CURRENT;
1637 EHdr->e_ident[EI_OSABI] = Config->OSABI;
1638 EHdr->e_type = getELFType();
1639 EHdr->e_machine = Config->EMachine;
1640 EHdr->e_version = EV_CURRENT;
1641 EHdr->e_entry = getEntryAddr();
1642 EHdr->e_shoff = SectionHeaderOff;
1643 EHdr->e_ehsize = sizeof(Elf_Ehdr);
1644 EHdr->e_phnum = Phdrs.size();
1645 EHdr->e_shentsize = sizeof(Elf_Shdr);
1646 EHdr->e_shnum = OutputSections.size() + 1;
1647 EHdr->e_shstrndx = In<ELFT>::ShStrTab->OutSec->SectionIndex;
1649 if (Config->EMachine == EM_ARM)
1650 // We don't currently use any features incompatible with EF_ARM_EABI_VER5,
1651 // but we don't have any firm guarantees of conformance. Linux AArch64
1652 // kernels (as of 2016) require an EABI version to be set.
1653 EHdr->e_flags = EF_ARM_EABI_VER5;
1654 else if (Config->EMachine == EM_MIPS)
1655 EHdr->e_flags = getMipsEFlags<ELFT>();
1657 if (!Config->Relocatable) {
1658 EHdr->e_phoff = sizeof(Elf_Ehdr);
1659 EHdr->e_phentsize = sizeof(Elf_Phdr);
1662 // Write the program header table.
1663 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff);
1664 for (PhdrEntry &P : Phdrs) {
1665 HBuf->p_type = P.p_type;
1666 HBuf->p_flags = P.p_flags;
1667 HBuf->p_offset = P.p_offset;
1668 HBuf->p_vaddr = P.p_vaddr;
1669 HBuf->p_paddr = P.p_paddr;
1670 HBuf->p_filesz = P.p_filesz;
1671 HBuf->p_memsz = P.p_memsz;
1672 HBuf->p_align = P.p_align;
1676 // Write the section header table. Note that the first table entry is null.
1677 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff);
1678 for (OutputSection *Sec : OutputSections)
1679 Sec->writeHeaderTo<ELFT>(++SHdrs);
1682 // Open a result file.
1683 template <class ELFT> void Writer<ELFT>::openFile() {
1684 if (!Config->Is64 && FileSize > UINT32_MAX) {
1685 error("output file too large: " + Twine(FileSize) + " bytes");
1689 unlinkAsync(Config->OutputFile);
1690 ErrorOr<std::unique_ptr<FileOutputBuffer>> BufferOrErr =
1691 FileOutputBuffer::create(Config->OutputFile, FileSize,
1692 FileOutputBuffer::F_executable);
1694 if (auto EC = BufferOrErr.getError())
1695 error("failed to open " + Config->OutputFile + ": " + EC.message());
1697 Buffer = std::move(*BufferOrErr);
1700 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() {
1701 uint8_t *Buf = Buffer->getBufferStart();
1702 for (OutputSection *Sec : OutputSections)
1703 if (Sec->Flags & SHF_ALLOC)
1704 Sec->writeTo<ELFT>(Buf + Sec->Offset);
1707 // Write section contents to a mmap'ed file.
1708 template <class ELFT> void Writer<ELFT>::writeSections() {
1709 uint8_t *Buf = Buffer->getBufferStart();
1711 // PPC64 needs to process relocations in the .opd section
1712 // before processing relocations in code-containing sections.
1713 Out::Opd = findSection(".opd");
1715 Out::OpdBuf = Buf + Out::Opd->Offset;
1716 Out::Opd->template writeTo<ELFT>(Buf + Out::Opd->Offset);
1719 OutputSection *EhFrameHdr =
1720 In<ELFT>::EhFrameHdr ? In<ELFT>::EhFrameHdr->OutSec : nullptr;
1722 // In -r or -emit-relocs mode, write the relocation sections first as in
1723 // ELf_Rel targets we might find out that we need to modify the relocated
1724 // section while doing it.
1725 for (OutputSection *Sec : OutputSections)
1726 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA)
1727 Sec->writeTo<ELFT>(Buf + Sec->Offset);
1729 for (OutputSection *Sec : OutputSections)
1730 if (Sec != Out::Opd && Sec != EhFrameHdr && Sec->Type != SHT_REL &&
1731 Sec->Type != SHT_RELA)
1732 Sec->writeTo<ELFT>(Buf + Sec->Offset);
1734 // The .eh_frame_hdr depends on .eh_frame section contents, therefore
1735 // it should be written after .eh_frame is written.
1736 if (EhFrameHdr && !EhFrameHdr->Sections.empty())
1737 EhFrameHdr->writeTo<ELFT>(Buf + EhFrameHdr->Offset);
1740 template <class ELFT> void Writer<ELFT>::writeBuildId() {
1741 if (!In<ELFT>::BuildId || !In<ELFT>::BuildId->OutSec)
1744 // Compute a hash of all sections of the output file.
1745 uint8_t *Start = Buffer->getBufferStart();
1746 uint8_t *End = Start + FileSize;
1747 In<ELFT>::BuildId->writeBuildId({Start, End});
1750 template void elf::writeResult<ELF32LE>();
1751 template void elf::writeResult<ELF32BE>();
1752 template void elf::writeResult<ELF64LE>();
1753 template void elf::writeResult<ELF64BE>();
1755 template bool elf::isRelroSection<ELF32LE>(const OutputSection *);
1756 template bool elf::isRelroSection<ELF32BE>(const OutputSection *);
1757 template bool elf::isRelroSection<ELF64LE>(const OutputSection *);
1758 template bool elf::isRelroSection<ELF64BE>(const OutputSection *);