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 clearOutputSections();
50 void createSyntheticSections();
51 void copyLocalSymbols();
52 void addSectionSymbols();
53 void addReservedSymbols();
54 void createSections();
55 void forEachRelSec(std::function<void(InputSectionBase &)> Fn);
57 void finalizeSections();
58 void addPredefinedSections();
60 std::vector<PhdrEntry> createPhdrs();
61 void removeEmptyPTLoad();
62 void addPtArmExid(std::vector<PhdrEntry> &Phdrs);
63 void assignFileOffsets();
64 void assignFileOffsetsBinary();
66 void fixSectionAlignments();
67 void fixPredefinedSymbols();
71 void writeSectionsBinary();
74 std::unique_ptr<FileOutputBuffer> Buffer;
76 std::vector<OutputSection *> OutputSections;
77 std::vector<OutputSectionCommand *> OutputSectionCommands;
78 OutputSectionFactory Factory{OutputSections};
80 void addRelIpltSymbols();
81 void addStartEndSymbols();
82 void addStartStopSymbols(OutputSection *Sec);
83 uint64_t getEntryAddr();
84 OutputSection *findSection(StringRef Name);
85 OutputSection *findSectionInScript(StringRef Name);
86 OutputSectionCommand *findSectionCommand(StringRef Name);
88 std::vector<PhdrEntry> Phdrs;
91 uint64_t SectionHeaderOff;
93 } // anonymous namespace
95 StringRef elf::getOutputSectionName(StringRef Name) {
96 if (Config->Relocatable)
99 // If -emit-relocs is given (which is rare), we need to copy
100 // relocation sections to the output. If input section .foo is
101 // output as .bar, we want to rename .rel.foo .rel.bar as well.
102 if (Config->EmitRelocs) {
103 for (StringRef V : {".rel.", ".rela."}) {
104 if (Name.startswith(V)) {
105 StringRef Inner = getOutputSectionName(Name.substr(V.size() - 1));
106 return Saver.save(V.drop_back() + Inner);
112 {".text.", ".rodata.", ".data.rel.ro.", ".data.", ".bss.rel.ro.",
113 ".bss.", ".init_array.", ".fini_array.", ".ctors.", ".dtors.", ".tbss.",
114 ".gcc_except_table.", ".tdata.", ".ARM.exidx."}) {
115 StringRef Prefix = V.drop_back();
116 if (Name.startswith(V) || Name == Prefix)
120 // CommonSection is identified as "COMMON" in linker scripts.
121 // By default, it should go to .bss section.
122 if (Name == "COMMON")
125 // ".zdebug_" is a prefix for ZLIB-compressed sections.
126 // Because we decompressed input sections, we want to remove 'z'.
127 if (Name.startswith(".zdebug_"))
128 return Saver.save("." + Name.substr(2));
132 template <class ELFT> static bool needsInterpSection() {
133 return !Symtab<ELFT>::X->getSharedFiles().empty() &&
134 !Config->DynamicLinker.empty() && !Script->ignoreInterpSection();
137 template <class ELFT> void elf::writeResult() { Writer<ELFT>().run(); }
139 template <class ELFT> void Writer<ELFT>::removeEmptyPTLoad() {
140 auto I = std::remove_if(Phdrs.begin(), Phdrs.end(), [&](const PhdrEntry &P) {
141 if (P.p_type != PT_LOAD)
145 uint64_t Size = P.Last->Addr + P.Last->Size - P.First->Addr;
148 Phdrs.erase(I, Phdrs.end());
151 // This function scans over the input sections and creates mergeable
152 // synthetic sections. It removes MergeInputSections from array and
153 // adds new synthetic ones. Each synthetic section is added to the
154 // location of the first input section it replaces.
155 static void combineMergableSections() {
156 std::vector<MergeSyntheticSection *> MergeSections;
157 for (InputSectionBase *&S : InputSections) {
158 MergeInputSection *MS = dyn_cast<MergeInputSection>(S);
162 // We do not want to handle sections that are not alive, so just remove
163 // them instead of trying to merge.
167 StringRef OutsecName = getOutputSectionName(MS->Name);
168 uint64_t Flags = MS->Flags & ~(uint64_t)SHF_GROUP;
169 uint32_t Alignment = std::max<uint32_t>(MS->Alignment, MS->Entsize);
171 auto I = llvm::find_if(MergeSections, [=](MergeSyntheticSection *Sec) {
172 return Sec->Name == OutsecName && Sec->Flags == Flags &&
173 Sec->Alignment == Alignment;
175 if (I == MergeSections.end()) {
176 MergeSyntheticSection *Syn =
177 make<MergeSyntheticSection>(OutsecName, MS->Type, Flags, Alignment);
178 MergeSections.push_back(Syn);
179 I = std::prev(MergeSections.end());
184 (*I)->addSection(MS);
187 std::vector<InputSectionBase *> &V = InputSections;
188 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end());
191 template <class ELFT> static void combineEhFrameSections() {
192 for (InputSectionBase *&S : InputSections) {
193 EhInputSection *ES = dyn_cast<EhInputSection>(S);
194 if (!ES || !ES->Live)
197 In<ELFT>::EhFrame->addSection(ES);
201 std::vector<InputSectionBase *> &V = InputSections;
202 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end());
205 template <class ELFT> void Writer<ELFT>::clearOutputSections() {
206 // Clear the OutputSections to make sure it is not used anymore. Any
207 // code from this point on should be using the linker script
209 for (OutputSection *Sec : OutputSections)
210 Sec->Sections.clear();
211 OutputSections.clear();
214 // The main function of the writer.
215 template <class ELFT> void Writer<ELFT>::run() {
216 // Create linker-synthesized sections such as .got or .plt.
217 // Such sections are of type input section.
218 createSyntheticSections();
219 combineMergableSections();
221 if (!Config->Relocatable)
222 combineEhFrameSections<ELFT>();
224 // We need to create some reserved symbols such as _end. Create them.
225 if (!Config->Relocatable)
226 addReservedSymbols();
228 // Create output sections.
229 Script->OutputSections = &OutputSections;
230 if (Script->Opt.HasSections) {
231 // If linker script contains SECTIONS commands, let it create sections.
232 Script->processCommands(Factory);
234 // Linker scripts may have left some input sections unassigned.
235 // Assign such sections using the default rule.
236 Script->addOrphanSections(Factory);
238 // If linker script does not contain SECTIONS commands, create
239 // output sections by default rules. We still need to give the
240 // linker script a chance to run, because it might contain
241 // non-SECTIONS commands such as ASSERT.
243 Script->processCommands(Factory);
246 if (Config->Discard != DiscardPolicy::All)
249 if (Config->CopyRelocs)
252 // Now that we have a complete set of output sections. This function
253 // completes section contents. For example, we need to add strings
254 // to the string table, and add entries to .got and .plt.
255 // finalizeSections does that.
260 if (!Script->Opt.HasSections) {
261 if (!Config->Relocatable)
262 fixSectionAlignments();
263 Script->fabricateDefaultCommands();
265 Script->synchronize();
268 for (BaseCommand *Base : Script->Opt.Commands)
269 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
270 OutputSectionCommands.push_back(Cmd);
272 clearOutputSections();
273 // If -compressed-debug-sections is specified, we need to compress
274 // .debug_* sections. Do it right now because it changes the size of
277 OutputSectionCommands.begin(), OutputSectionCommands.end(),
278 [](OutputSectionCommand *Cmd) { Cmd->maybeCompress<ELFT>(); });
280 if (Config->Relocatable) {
283 Script->assignAddresses(Phdrs, OutputSectionCommands);
285 // Remove empty PT_LOAD to avoid causing the dynamic linker to try to mmap a
286 // 0 sized region. This has to be done late since only after assignAddresses
287 // we know the size of the sections.
290 if (!Config->OFormatBinary)
293 assignFileOffsetsBinary();
296 fixPredefinedSymbols();
299 // It does not make sense try to open the file if we have error already.
302 // Write the result down to a file.
307 if (!Config->OFormatBinary) {
311 writeSectionsBinary();
314 // Backfill .note.gnu.build-id section content. This is done at last
315 // because the content is usually a hash value of the entire output file.
321 // Handle -Map option.
322 writeMapFile<ELFT>(OutputSectionCommands);
326 if (auto EC = Buffer->commit())
327 error("failed to write to the output file: " + EC.message());
329 // Flush the output streams and exit immediately. A full shutdown
330 // is a good test that we are keeping track of all allocated memory,
331 // but actually freeing it is a waste of time in a regular linker run.
332 if (Config->ExitEarly)
336 // Initialize Out members.
337 template <class ELFT> void Writer<ELFT>::createSyntheticSections() {
338 // Initialize all pointers with NULL. This is needed because
339 // you can call lld::elf::main more than once as a library.
340 memset(&Out::First, 0, sizeof(Out));
342 auto Add = [](InputSectionBase *Sec) { InputSections.push_back(Sec); };
344 InX::DynStrTab = make<StringTableSection>(".dynstr", true);
345 InX::Dynamic = make<DynamicSection<ELFT>>();
346 In<ELFT>::RelaDyn = make<RelocationSection<ELFT>>(
347 Config->IsRela ? ".rela.dyn" : ".rel.dyn", Config->ZCombreloc);
348 InX::ShStrTab = make<StringTableSection>(".shstrtab", false);
350 Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC);
351 Out::ElfHeader->Size = sizeof(Elf_Ehdr);
352 Out::ProgramHeaders = make<OutputSection>("", 0, SHF_ALLOC);
353 Out::ProgramHeaders->updateAlignment(Config->Wordsize);
355 if (needsInterpSection<ELFT>()) {
356 InX::Interp = createInterpSection();
359 InX::Interp = nullptr;
362 if (!Config->Relocatable)
363 Add(createCommentSection<ELFT>());
365 if (Config->Strip != StripPolicy::All) {
366 InX::StrTab = make<StringTableSection>(".strtab", false);
367 InX::SymTab = make<SymbolTableSection<ELFT>>(*InX::StrTab);
370 if (Config->BuildId != BuildIdKind::None) {
371 InX::BuildId = make<BuildIdSection>();
375 InX::Common = createCommonSection<ELFT>();
379 InX::Bss = make<BssSection>(".bss");
381 InX::BssRelRo = make<BssSection>(".bss.rel.ro");
384 // Add MIPS-specific sections.
385 bool HasDynSymTab = !Symtab<ELFT>::X->getSharedFiles().empty() ||
386 Config->Pic || Config->ExportDynamic;
387 if (Config->EMachine == EM_MIPS) {
388 if (!Config->Shared && HasDynSymTab) {
389 InX::MipsRldMap = make<MipsRldMapSection>();
390 Add(InX::MipsRldMap);
392 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create())
394 if (auto *Sec = MipsOptionsSection<ELFT>::create())
396 if (auto *Sec = MipsReginfoSection<ELFT>::create())
401 InX::DynSymTab = make<SymbolTableSection<ELFT>>(*InX::DynStrTab);
404 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>();
405 Add(In<ELFT>::VerSym);
407 if (!Config->VersionDefinitions.empty()) {
408 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>();
409 Add(In<ELFT>::VerDef);
412 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>();
413 Add(In<ELFT>::VerNeed);
415 if (Config->GnuHash) {
416 InX::GnuHashTab = make<GnuHashTableSection>();
417 Add(InX::GnuHashTab);
420 if (Config->SysvHash) {
421 In<ELFT>::HashTab = make<HashTableSection<ELFT>>();
422 Add(In<ELFT>::HashTab);
427 Add(In<ELFT>::RelaDyn);
430 // Add .got. MIPS' .got is so different from the other archs,
431 // it has its own class.
432 if (Config->EMachine == EM_MIPS) {
433 InX::MipsGot = make<MipsGotSection>();
436 InX::Got = make<GotSection>();
440 InX::GotPlt = make<GotPltSection>();
442 InX::IgotPlt = make<IgotPltSection>();
445 if (Config->GdbIndex) {
446 InX::GdbIndex = make<GdbIndexSection>();
450 // We always need to add rel[a].plt to output if it has entries.
451 // Even for static linking it can contain R_[*]_IRELATIVE relocations.
452 In<ELFT>::RelaPlt = make<RelocationSection<ELFT>>(
453 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/);
454 Add(In<ELFT>::RelaPlt);
456 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure
457 // that the IRelative relocations are processed last by the dynamic loader
458 In<ELFT>::RelaIplt = make<RelocationSection<ELFT>>(
459 (Config->EMachine == EM_ARM) ? ".rel.dyn" : In<ELFT>::RelaPlt->Name,
461 Add(In<ELFT>::RelaIplt);
463 InX::Plt = make<PltSection>(Target->PltHeaderSize);
465 InX::Iplt = make<PltSection>(0);
468 if (!Config->Relocatable) {
469 if (Config->EhFrameHdr) {
470 In<ELFT>::EhFrameHdr = make<EhFrameHeader<ELFT>>();
471 Add(In<ELFT>::EhFrameHdr);
473 In<ELFT>::EhFrame = make<EhFrameSection<ELFT>>();
474 Add(In<ELFT>::EhFrame);
484 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName,
485 const SymbolBody &B) {
486 if (B.isFile() || B.isSection())
489 // If sym references a section in a discarded group, don't keep it.
490 if (Sec == &InputSection::Discarded)
493 if (Config->Discard == DiscardPolicy::None)
496 // In ELF assembly .L symbols are normally discarded by the assembler.
497 // If the assembler fails to do so, the linker discards them if
498 // * --discard-locals is used.
499 // * The symbol is in a SHF_MERGE section, which is normally the reason for
500 // the assembler keeping the .L symbol.
501 if (!SymName.startswith(".L") && !SymName.empty())
504 if (Config->Discard == DiscardPolicy::Locals)
507 return !Sec || !(Sec->Flags & SHF_MERGE);
510 static bool includeInSymtab(const SymbolBody &B) {
511 if (!B.isLocal() && !B.symbol()->IsUsedInRegularObj)
514 if (auto *D = dyn_cast<DefinedRegular>(&B)) {
515 // Always include absolute symbols.
516 SectionBase *Sec = D->Section;
519 if (auto *IS = dyn_cast<InputSectionBase>(Sec)) {
521 IS = cast<InputSectionBase>(Sec);
522 // Exclude symbols pointing to garbage-collected sections.
526 if (auto *S = dyn_cast<MergeInputSection>(Sec))
527 if (!S->getSectionPiece(D->Value)->Live)
533 // Local symbols are not in the linker's symbol table. This function scans
534 // each object file's symbol table to copy local symbols to the output.
535 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() {
538 for (elf::ObjectFile<ELFT> *F : Symtab<ELFT>::X->getObjectFiles()) {
539 for (SymbolBody *B : F->getLocalSymbols()) {
542 ": broken object: getLocalSymbols returns a non-local symbol");
543 auto *DR = dyn_cast<DefinedRegular>(B);
545 // No reason to keep local undefined symbol in symtab.
548 if (!includeInSymtab(*B))
551 SectionBase *Sec = DR->Section;
552 if (!shouldKeepInSymtab(Sec, B->getName(), *B))
554 InX::SymTab->addSymbol(B);
559 template <class ELFT> void Writer<ELFT>::addSectionSymbols() {
560 // Create one STT_SECTION symbol for each output section we might
561 // have a relocation with.
562 for (OutputSection *Sec : OutputSections) {
563 if (Sec->Sections.empty())
566 InputSection *IS = Sec->Sections[0];
567 if (isa<SyntheticSection>(IS) || IS->Type == SHT_REL ||
568 IS->Type == SHT_RELA)
572 make<DefinedRegular>("", /*IsLocal=*/true, /*StOther=*/0, STT_SECTION,
573 /*Value=*/0, /*Size=*/0, IS, nullptr);
574 InX::SymTab->addSymbol(Sym);
578 // Today's loaders have a feature to make segments read-only after
579 // processing dynamic relocations to enhance security. PT_GNU_RELRO
580 // is defined for that.
582 // This function returns true if a section needs to be put into a
583 // PT_GNU_RELRO segment.
584 bool elf::isRelroSection(const OutputSection *Sec) {
588 uint64_t Flags = Sec->Flags;
590 // Non-allocatable or non-writable sections don't need RELRO because
591 // they are not writable or not even mapped to memory in the first place.
592 // RELRO is for sections that are essentially read-only but need to
593 // be writable only at process startup to allow dynamic linker to
594 // apply relocations.
595 if (!(Flags & SHF_ALLOC) || !(Flags & SHF_WRITE))
598 // Once initialized, TLS data segments are used as data templates
599 // for a thread-local storage. For each new thread, runtime
600 // allocates memory for a TLS and copy templates there. No thread
601 // are supposed to use templates directly. Thus, it can be in RELRO.
605 // .init_array, .preinit_array and .fini_array contain pointers to
606 // functions that are executed on process startup or exit. These
607 // pointers are set by the static linker, and they are not expected
608 // to change at runtime. But if you are an attacker, you could do
609 // interesting things by manipulating pointers in .fini_array, for
610 // example. So they are put into RELRO.
611 uint32_t Type = Sec->Type;
612 if (Type == SHT_INIT_ARRAY || Type == SHT_FINI_ARRAY ||
613 Type == SHT_PREINIT_ARRAY)
616 // .got contains pointers to external symbols. They are resolved by
617 // the dynamic linker when a module is loaded into memory, and after
618 // that they are not expected to change. So, it can be in RELRO.
619 if (InX::Got && Sec == InX::Got->getParent())
622 // .got.plt contains pointers to external function symbols. They are
623 // by default resolved lazily, so we usually cannot put it into RELRO.
624 // However, if "-z now" is given, the lazy symbol resolution is
625 // disabled, which enables us to put it into RELRO.
626 if (Sec == InX::GotPlt->getParent())
629 // .dynamic section contains data for the dynamic linker, and
630 // there's no need to write to it at runtime, so it's better to put
632 if (Sec == InX::Dynamic->getParent())
635 // .bss.rel.ro is used for copy relocations for read-only symbols.
636 // Since the dynamic linker needs to process copy relocations, the
637 // section cannot be read-only, but once initialized, they shouldn't
639 if (Sec == InX::BssRelRo->getParent())
642 // Sections with some special names are put into RELRO. This is a
643 // bit unfortunate because section names shouldn't be significant in
644 // ELF in spirit. But in reality many linker features depend on
645 // magic section names.
646 StringRef S = Sec->Name;
647 return S == ".data.rel.ro" || S == ".ctors" || S == ".dtors" || S == ".jcr" ||
648 S == ".eh_frame" || S == ".openbsd.randomdata";
651 // We compute a rank for each section. The rank indicates where the
652 // section should be placed in the file. Instead of using simple
653 // numbers (0,1,2...), we use a series of flags. One for each decision
654 // point when placing the section.
655 // Using flags has two key properties:
656 // * It is easy to check if a give branch was taken.
657 // * It is easy two see how similar two ranks are (see getRankProximity).
659 RF_NOT_ADDR_SET = 1 << 16,
660 RF_NOT_INTERP = 1 << 15,
661 RF_NOT_ALLOC = 1 << 14,
663 RF_EXEC_WRITE = 1 << 12,
665 RF_NON_TLS_BSS = 1 << 10,
666 RF_NON_TLS_BSS_RO = 1 << 9,
669 RF_PPC_NOT_TOCBSS = 1 << 6,
671 RF_PPC_TOCL = 1 << 4,
673 RF_PPC_BRANCH_LT = 1 << 2,
674 RF_MIPS_GPREL = 1 << 1,
675 RF_MIPS_NOT_GOT = 1 << 0
678 static unsigned getSectionRank(const OutputSection *Sec) {
681 // We want to put section specified by -T option first, so we
682 // can start assigning VA starting from them later.
683 if (Config->SectionStartMap.count(Sec->Name))
685 Rank |= RF_NOT_ADDR_SET;
687 // Put .interp first because some loaders want to see that section
688 // on the first page of the executable file when loaded into memory.
689 if (Sec->Name == ".interp")
691 Rank |= RF_NOT_INTERP;
693 // Allocatable sections go first to reduce the total PT_LOAD size and
694 // so debug info doesn't change addresses in actual code.
695 if (!(Sec->Flags & SHF_ALLOC))
696 return Rank | RF_NOT_ALLOC;
698 // Sort sections based on their access permission in the following
699 // order: R, RX, RWX, RW. This order is based on the following
701 // * Read-only sections come first such that they go in the
702 // PT_LOAD covering the program headers at the start of the file.
703 // * Read-only, executable sections come next, unless the
704 // -no-rosegment option is used.
705 // * Writable, executable sections follow such that .plt on
706 // architectures where it needs to be writable will be placed
707 // between .text and .data.
708 // * Writable sections come last, such that .bss lands at the very
709 // end of the last PT_LOAD.
710 bool IsExec = Sec->Flags & SHF_EXECINSTR;
711 bool IsWrite = Sec->Flags & SHF_WRITE;
715 Rank |= RF_EXEC_WRITE;
716 else if (!Config->SingleRoRx)
723 // If we got here we know that both A and B are in the same PT_LOAD.
725 bool IsTls = Sec->Flags & SHF_TLS;
726 bool IsNoBits = Sec->Type == SHT_NOBITS;
728 // The first requirement we have is to put (non-TLS) nobits sections last. The
729 // reason is that the only thing the dynamic linker will see about them is a
730 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the
731 // PT_LOAD, so that has to correspond to the nobits sections.
732 bool IsNonTlsNoBits = IsNoBits && !IsTls;
734 Rank |= RF_NON_TLS_BSS;
736 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo
737 // sections after r/w ones, so that the RelRo sections are contiguous.
738 bool IsRelRo = isRelroSection(Sec);
739 if (IsNonTlsNoBits && !IsRelRo)
740 Rank |= RF_NON_TLS_BSS_RO;
741 if (!IsNonTlsNoBits && IsRelRo)
742 Rank |= RF_NON_TLS_BSS_RO;
744 // The TLS initialization block needs to be a single contiguous block in a R/W
745 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W
746 // sections. The TLS NOBITS sections are placed here as they don't take up
747 // virtual address space in the PT_LOAD.
751 // Within the TLS initialization block, the non-nobits sections need to appear
756 // // Some architectures have additional ordering restrictions for sections
757 // // within the same PT_LOAD.
758 if (Config->EMachine == EM_PPC64) {
759 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections
760 // that we would like to make sure appear is a specific order to maximize
761 // their coverage by a single signed 16-bit offset from the TOC base
762 // pointer. Conversely, the special .tocbss section should be first among
763 // all SHT_NOBITS sections. This will put it next to the loaded special
764 // PPC64 sections (and, thus, within reach of the TOC base pointer).
765 StringRef Name = Sec->Name;
766 if (Name != ".tocbss")
767 Rank |= RF_PPC_NOT_TOCBSS;
778 if (Name == ".branch_lt")
779 Rank |= RF_PPC_BRANCH_LT;
781 if (Config->EMachine == EM_MIPS) {
782 // All sections with SHF_MIPS_GPREL flag should be grouped together
783 // because data in these sections is addressable with a gp relative address.
784 if (Sec->Flags & SHF_MIPS_GPREL)
785 Rank |= RF_MIPS_GPREL;
787 if (Sec->Name != ".got")
788 Rank |= RF_MIPS_NOT_GOT;
794 static bool compareSectionsNonScript(const OutputSection *A,
795 const OutputSection *B) {
796 if (A->SortRank != B->SortRank)
797 return A->SortRank < B->SortRank;
798 if (!(A->SortRank & RF_NOT_ADDR_SET))
799 return Config->SectionStartMap.lookup(A->Name) <
800 Config->SectionStartMap.lookup(B->Name);
804 // Output section ordering is determined by this function.
805 static bool compareSections(const OutputSection *A, const OutputSection *B) {
806 // For now, put sections mentioned in a linker script
807 // first. Sections not on linker script will have a SectionIndex of
809 int AIndex = A->SectionIndex;
810 int BIndex = B->SectionIndex;
811 if (AIndex != BIndex)
812 return AIndex < BIndex;
814 return compareSectionsNonScript(A, B);
817 void PhdrEntry::add(OutputSection *Sec) {
821 p_align = std::max(p_align, Sec->Alignment);
822 if (p_type == PT_LOAD)
823 Sec->FirstInPtLoad = First;
826 template <class ELFT>
827 static Symbol *addRegular(StringRef Name, SectionBase *Sec, uint64_t Value,
828 uint8_t StOther = STV_HIDDEN,
829 uint8_t Binding = STB_WEAK) {
830 // The linker generated symbols are added as STB_WEAK to allow user defined
831 // ones to override them.
832 return Symtab<ELFT>::X->addRegular(Name, StOther, STT_NOTYPE, Value,
833 /*Size=*/0, Binding, Sec,
837 template <class ELFT>
838 static DefinedRegular *
839 addOptionalRegular(StringRef Name, SectionBase *Sec, uint64_t Val,
840 uint8_t StOther = STV_HIDDEN, uint8_t Binding = STB_GLOBAL) {
841 SymbolBody *S = Symtab<ELFT>::X->find(Name);
844 if (S->isInCurrentDSO())
846 return cast<DefinedRegular>(
847 addRegular<ELFT>(Name, Sec, Val, StOther, Binding)->body());
850 // The beginning and the ending of .rel[a].plt section are marked
851 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked
852 // executable. The runtime needs these symbols in order to resolve
853 // all IRELATIVE relocs on startup. For dynamic executables, we don't
854 // need these symbols, since IRELATIVE relocs are resolved through GOT
855 // and PLT. For details, see http://www.airs.com/blog/archives/403.
856 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() {
859 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start";
860 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, 0, STV_HIDDEN, STB_WEAK);
862 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end";
863 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, -1, STV_HIDDEN, STB_WEAK);
866 // The linker is expected to define some symbols depending on
867 // the linking result. This function defines such symbols.
868 template <class ELFT> void Writer<ELFT>::addReservedSymbols() {
869 if (Config->EMachine == EM_MIPS) {
870 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer
871 // so that it points to an absolute address which by default is relative
872 // to GOT. Default offset is 0x7ff0.
873 // See "Global Data Symbols" in Chapter 6 in the following document:
874 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf
875 ElfSym::MipsGp = Symtab<ELFT>::X->addAbsolute("_gp", STV_HIDDEN, STB_LOCAL);
877 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between
878 // start of function and 'gp' pointer into GOT.
879 if (Symtab<ELFT>::X->find("_gp_disp"))
881 Symtab<ELFT>::X->addAbsolute("_gp_disp", STV_HIDDEN, STB_LOCAL);
883 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp'
884 // pointer. This symbol is used in the code generated by .cpload pseudo-op
885 // in case of using -mno-shared option.
886 // https://sourceware.org/ml/binutils/2004-12/msg00094.html
887 if (Symtab<ELFT>::X->find("__gnu_local_gp"))
888 ElfSym::MipsLocalGp =
889 Symtab<ELFT>::X->addAbsolute("__gnu_local_gp", STV_HIDDEN, STB_LOCAL);
892 // In the assembly for 32 bit x86 the _GLOBAL_OFFSET_TABLE_ symbol
893 // is magical and is used to produce a R_386_GOTPC relocation.
894 // The R_386_GOTPC relocation value doesn't actually depend on the
895 // symbol value, so it could use an index of STN_UNDEF which, according
896 // to the spec, means the symbol value is 0.
897 // Unfortunately both gas and MC keep the _GLOBAL_OFFSET_TABLE_ symbol in
899 // The situation is even stranger on x86_64 where the assembly doesn't
900 // need the magical symbol, but gas still puts _GLOBAL_OFFSET_TABLE_ as
901 // an undefined symbol in the .o files.
902 // Given that the symbol is effectively unused, we just create a dummy
903 // hidden one to avoid the undefined symbol error.
904 Symtab<ELFT>::X->addIgnored("_GLOBAL_OFFSET_TABLE_");
906 // __tls_get_addr is defined by the dynamic linker for dynamic ELFs. For
907 // static linking the linker is required to optimize away any references to
908 // __tls_get_addr, so it's not defined anywhere. Create a hidden definition
909 // to avoid the undefined symbol error.
911 Symtab<ELFT>::X->addIgnored("__tls_get_addr");
913 // __ehdr_start is the location of ELF file headers. Note that we define
914 // this symbol unconditionally even when using a linker script, which
915 // differs from the behavior implemented by GNU linker which only define
916 // this symbol if ELF headers are in the memory mapped segment.
917 addOptionalRegular<ELFT>("__ehdr_start", Out::ElfHeader, 0, STV_HIDDEN);
919 // If linker script do layout we do not need to create any standart symbols.
920 if (Script->Opt.HasSections)
923 auto Add = [](StringRef S) {
924 return addOptionalRegular<ELFT>(S, Out::ElfHeader, 0, STV_DEFAULT);
927 ElfSym::Bss = Add("__bss_start");
928 ElfSym::End1 = Add("end");
929 ElfSym::End2 = Add("_end");
930 ElfSym::Etext1 = Add("etext");
931 ElfSym::Etext2 = Add("_etext");
932 ElfSym::Edata1 = Add("edata");
933 ElfSym::Edata2 = Add("_edata");
936 // Sort input sections by section name suffixes for
937 // __attribute__((init_priority(N))).
938 static void sortInitFini(OutputSection *S) {
940 reinterpret_cast<OutputSection *>(S)->sortInitFini();
943 // Sort input sections by the special rule for .ctors and .dtors.
944 static void sortCtorsDtors(OutputSection *S) {
946 reinterpret_cast<OutputSection *>(S)->sortCtorsDtors();
949 // Sort input sections using the list provided by --symbol-ordering-file.
950 template <class ELFT>
951 static void sortBySymbolsOrder(ArrayRef<OutputSection *> OutputSections) {
952 if (Config->SymbolOrderingFile.empty())
955 // Build a map from symbols to their priorities. Symbols that didn't
956 // appear in the symbol ordering file have the lowest priority 0.
957 // All explicitly mentioned symbols have negative (higher) priorities.
958 DenseMap<StringRef, int> SymbolOrder;
959 int Priority = -Config->SymbolOrderingFile.size();
960 for (StringRef S : Config->SymbolOrderingFile)
961 SymbolOrder.insert({S, Priority++});
963 // Build a map from sections to their priorities.
964 DenseMap<SectionBase *, int> SectionOrder;
965 for (elf::ObjectFile<ELFT> *File : Symtab<ELFT>::X->getObjectFiles()) {
966 for (SymbolBody *Body : File->getSymbols()) {
967 auto *D = dyn_cast<DefinedRegular>(Body);
968 if (!D || !D->Section)
970 int &Priority = SectionOrder[D->Section];
971 Priority = std::min(Priority, SymbolOrder.lookup(D->getName()));
975 // Sort sections by priority.
976 for (OutputSection *Base : OutputSections)
977 if (auto *Sec = dyn_cast<OutputSection>(Base))
978 Sec->sort([&](InputSectionBase *S) { return SectionOrder.lookup(S); });
981 template <class ELFT>
982 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) {
983 for (InputSectionBase *IS : InputSections) {
986 // Scan all relocations. Each relocation goes through a series
987 // of tests to determine if it needs special treatment, such as
988 // creating GOT, PLT, copy relocations, etc.
989 // Note that relocations for non-alloc sections are directly
990 // processed by InputSection::relocateNonAlloc.
991 if (!(IS->Flags & SHF_ALLOC))
993 if (isa<InputSection>(IS) || isa<EhInputSection>(IS))
997 if (!Config->Relocatable) {
998 for (EhInputSection *ES : In<ELFT>::EhFrame->Sections)
1003 template <class ELFT> void Writer<ELFT>::createSections() {
1004 for (InputSectionBase *IS : InputSections)
1006 Factory.addInputSec(IS, getOutputSectionName(IS->Name));
1008 sortBySymbolsOrder<ELFT>(OutputSections);
1009 sortInitFini(findSection(".init_array"));
1010 sortInitFini(findSection(".fini_array"));
1011 sortCtorsDtors(findSection(".ctors"));
1012 sortCtorsDtors(findSection(".dtors"));
1014 for (OutputSection *Sec : OutputSections)
1015 Sec->assignOffsets();
1018 // We want to find how similar two ranks are.
1019 // The more branches in getSectionRank that match, the more similar they are.
1020 // Since each branch corresponds to a bit flag, we can just use
1021 // countLeadingZeros.
1022 static unsigned getRankProximity(OutputSection *A, OutputSection *B) {
1023 return countLeadingZeros(A->SortRank ^ B->SortRank);
1026 // We want to place orphan sections so that they share as much
1027 // characteristics with their neighbors as possible. For example, if
1028 // both are rw, or both are tls.
1029 template <typename ELFT>
1030 static std::vector<OutputSection *>::iterator
1031 findOrphanPos(std::vector<OutputSection *>::iterator B,
1032 std::vector<OutputSection *>::iterator E) {
1033 OutputSection *Sec = *E;
1035 // Find the first element that has as close a rank as possible.
1036 auto I = std::max_element(B, E, [=](OutputSection *A, OutputSection *B) {
1037 return getRankProximity(Sec, A) < getRankProximity(Sec, B);
1042 // Consider all existing sections with the same proximity.
1043 unsigned Proximity = getRankProximity(Sec, *I);
1044 while (I != E && getRankProximity(Sec, *I) == Proximity &&
1045 Sec->SortRank >= (*I)->SortRank)
1050 template <class ELFT> void Writer<ELFT>::sortSections() {
1051 // Don't sort if using -r. It is not necessary and we want to preserve the
1052 // relative order for SHF_LINK_ORDER sections.
1053 if (Config->Relocatable)
1056 if (Script->Opt.HasSections)
1057 Script->adjustSectionsBeforeSorting();
1059 for (OutputSection *Sec : OutputSections)
1060 Sec->SortRank = getSectionRank(Sec);
1062 if (!Script->Opt.HasSections) {
1063 std::stable_sort(OutputSections.begin(), OutputSections.end(),
1064 compareSectionsNonScript);
1068 // The order of the sections in the script is arbitrary and may not agree with
1069 // compareSectionsNonScript. This means that we cannot easily define a
1070 // strict weak ordering. To see why, consider a comparison of a section in the
1071 // script and one not in the script. We have a two simple options:
1072 // * Make them equivalent (a is not less than b, and b is not less than a).
1073 // The problem is then that equivalence has to be transitive and we can
1074 // have sections a, b and c with only b in a script and a less than c
1075 // which breaks this property.
1076 // * Use compareSectionsNonScript. Given that the script order doesn't have
1077 // to match, we can end up with sections a, b, c, d where b and c are in the
1078 // script and c is compareSectionsNonScript less than b. In which case d
1079 // can be equivalent to c, a to b and d < a. As a concrete example:
1080 // .a (rx) # not in script
1081 // .b (rx) # in script
1082 // .c (ro) # in script
1083 // .d (ro) # not in script
1085 // The way we define an order then is:
1086 // * First put script sections at the start and sort the script sections.
1087 // * Move each non-script section to its preferred position. We try
1088 // to put each section in the last position where it it can share
1091 std::stable_sort(OutputSections.begin(), OutputSections.end(),
1094 auto I = OutputSections.begin();
1095 auto E = OutputSections.end();
1097 std::find_if(OutputSections.begin(), E,
1098 [](OutputSection *S) { return S->SectionIndex == INT_MAX; });
1099 while (NonScriptI != E) {
1100 auto Pos = findOrphanPos<ELFT>(I, NonScriptI);
1102 // As an optimization, find all sections with the same sort rank
1103 // and insert them with one rotate.
1104 unsigned Rank = (*NonScriptI)->SortRank;
1105 auto End = std::find_if(NonScriptI + 1, E, [=](OutputSection *Sec) {
1106 return Sec->SortRank != Rank;
1108 std::rotate(Pos, NonScriptI, End);
1112 Script->adjustSectionsAfterSorting();
1115 static void applySynthetic(const std::vector<SyntheticSection *> &Sections,
1116 std::function<void(SyntheticSection *)> Fn) {
1117 for (SyntheticSection *SS : Sections)
1118 if (SS && SS->getParent() && !SS->empty()) {
1120 SS->getParent()->assignOffsets();
1124 // We need to add input synthetic sections early in createSyntheticSections()
1125 // to make them visible from linkescript side. But not all sections are always
1126 // required to be in output. For example we don't need dynamic section content
1127 // sometimes. This function filters out such unused sections from the output.
1128 static void removeUnusedSyntheticSections(std::vector<OutputSection *> &V) {
1129 // All input synthetic sections that can be empty are placed after
1130 // all regular ones. We iterate over them all and exit at first
1132 for (InputSectionBase *S : llvm::reverse(InputSections)) {
1133 SyntheticSection *SS = dyn_cast<SyntheticSection>(S);
1136 OutputSection *OS = SS->getParent();
1137 if (!SS->empty() || !OS)
1139 OS->Sections.erase(std::find(OS->Sections.begin(), OS->Sections.end(), SS));
1141 // If there are no other sections in the output section, remove it from the
1143 if (OS->Sections.empty())
1144 V.erase(std::find(V.begin(), V.end(), OS));
1148 // Create output section objects and add them to OutputSections.
1149 template <class ELFT> void Writer<ELFT>::finalizeSections() {
1150 Out::DebugInfo = findSection(".debug_info");
1151 Out::PreinitArray = findSection(".preinit_array");
1152 Out::InitArray = findSection(".init_array");
1153 Out::FiniArray = findSection(".fini_array");
1155 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop
1156 // symbols for sections, so that the runtime can get the start and end
1157 // addresses of each section by section name. Add such symbols.
1158 if (!Config->Relocatable) {
1159 addStartEndSymbols();
1160 for (OutputSection *Sec : OutputSections)
1161 addStartStopSymbols(Sec);
1164 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type.
1165 // It should be okay as no one seems to care about the type.
1166 // Even the author of gold doesn't remember why gold behaves that way.
1167 // https://sourceware.org/ml/binutils/2002-03/msg00360.html
1169 addRegular<ELFT>("_DYNAMIC", InX::Dynamic, 0);
1171 // Define __rel[a]_iplt_{start,end} symbols if needed.
1172 addRelIpltSymbols();
1174 // This responsible for splitting up .eh_frame section into
1175 // pieces. The relocation scan uses those pieces, so this has to be
1177 applySynthetic({In<ELFT>::EhFrame},
1178 [](SyntheticSection *SS) { SS->finalizeContents(); });
1180 // Scan relocations. This must be done after every symbol is declared so that
1181 // we can correctly decide if a dynamic relocation is needed.
1182 forEachRelSec(scanRelocations<ELFT>);
1184 if (InX::Plt && !InX::Plt->empty())
1185 InX::Plt->addSymbols();
1186 if (InX::Iplt && !InX::Iplt->empty())
1187 InX::Iplt->addSymbols();
1189 // Now that we have defined all possible global symbols including linker-
1190 // synthesized ones. Visit all symbols to give the finishing touches.
1191 for (Symbol *S : Symtab<ELFT>::X->getSymbols()) {
1192 SymbolBody *Body = S->body();
1194 if (!includeInSymtab(*Body))
1197 InX::SymTab->addSymbol(Body);
1199 if (InX::DynSymTab && S->includeInDynsym()) {
1200 InX::DynSymTab->addSymbol(Body);
1201 if (auto *SS = dyn_cast<SharedSymbol>(Body))
1202 if (cast<SharedFile<ELFT>>(SS->File)->isNeeded())
1203 In<ELFT>::VerNeed->addSymbol(SS);
1207 // Do not proceed if there was an undefined symbol.
1211 addPredefinedSections();
1212 removeUnusedSyntheticSections(OutputSections);
1216 // This is a bit of a hack. A value of 0 means undef, so we set it
1217 // to 1 t make __ehdr_start defined. The section number is not
1218 // particularly relevant.
1219 Out::ElfHeader->SectionIndex = 1;
1222 for (OutputSection *Sec : OutputSections) {
1223 Sec->SectionIndex = I++;
1224 Sec->ShName = InX::ShStrTab->addString(Sec->Name);
1227 // Binary and relocatable output does not have PHDRS.
1228 // The headers have to be created before finalize as that can influence the
1229 // image base and the dynamic section on mips includes the image base.
1230 if (!Config->Relocatable && !Config->OFormatBinary) {
1231 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs();
1232 addPtArmExid(Phdrs);
1233 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size();
1236 // Dynamic section must be the last one in this list and dynamic
1237 // symbol table section (DynSymTab) must be the first one.
1238 applySynthetic({InX::DynSymTab, InX::Bss, InX::BssRelRo,
1239 InX::GnuHashTab, In<ELFT>::HashTab, InX::SymTab,
1240 InX::ShStrTab, InX::StrTab, In<ELFT>::VerDef,
1241 InX::DynStrTab, InX::GdbIndex, InX::Got,
1242 InX::MipsGot, InX::IgotPlt, InX::GotPlt,
1243 In<ELFT>::RelaDyn, In<ELFT>::RelaIplt, In<ELFT>::RelaPlt,
1244 InX::Plt, InX::Iplt, In<ELFT>::EhFrameHdr,
1245 In<ELFT>::VerSym, In<ELFT>::VerNeed, InX::Dynamic},
1246 [](SyntheticSection *SS) { SS->finalizeContents(); });
1248 // Some architectures use small displacements for jump instructions.
1249 // It is linker's responsibility to create thunks containing long
1250 // jump instructions if jump targets are too far. Create thunks.
1251 if (Target->NeedsThunks) {
1252 // FIXME: only ARM Interworking and Mips LA25 Thunks are implemented,
1254 // do not require address information. To support range extension Thunks
1255 // we need to assign addresses so that we can tell if jump instructions
1256 // are out of range. This will need to turn into a loop that converges
1257 // when no more Thunks are added
1259 if (TC.createThunks(OutputSections))
1260 applySynthetic({InX::MipsGot},
1261 [](SyntheticSection *SS) { SS->updateAllocSize(); });
1263 // Fill other section headers. The dynamic table is finalized
1264 // at the end because some tags like RELSZ depend on result
1265 // of finalizing other sections.
1266 for (OutputSection *Sec : OutputSections)
1267 Sec->finalize<ELFT>();
1269 // createThunks may have added local symbols to the static symbol table
1270 applySynthetic({InX::SymTab, InX::ShStrTab, InX::StrTab},
1271 [](SyntheticSection *SS) { SS->postThunkContents(); });
1274 template <class ELFT> void Writer<ELFT>::addPredefinedSections() {
1275 // ARM ABI requires .ARM.exidx to be terminated by some piece of data.
1276 // We have the terminater synthetic section class. Add that at the end.
1277 auto *OS = dyn_cast_or_null<OutputSection>(findSection(".ARM.exidx"));
1278 if (!OS || OS->Sections.empty() || Config->Relocatable)
1281 auto *Sentinel = make<ARMExidxSentinelSection>();
1282 OS->addSection(Sentinel);
1283 // If there are linker script commands existing at this point then add the
1284 // sentinel to the last of these too.
1285 if (OutputSectionCommand *C = Script->getCmd(OS)) {
1286 auto ISD = std::find_if(C->Commands.rbegin(), C->Commands.rend(),
1287 [](const BaseCommand *Base) {
1288 return isa<InputSectionDescription>(Base);
1290 cast<InputSectionDescription>(*ISD)->Sections.push_back(Sentinel);
1294 // The linker is expected to define SECNAME_start and SECNAME_end
1295 // symbols for a few sections. This function defines them.
1296 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() {
1297 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) {
1298 // These symbols resolve to the image base if the section does not exist.
1299 // A special value -1 indicates end of the section.
1301 addOptionalRegular<ELFT>(Start, OS, 0);
1302 addOptionalRegular<ELFT>(End, OS, -1);
1305 OS = Out::ElfHeader;
1306 addOptionalRegular<ELFT>(Start, OS, 0);
1307 addOptionalRegular<ELFT>(End, OS, 0);
1311 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray);
1312 Define("__init_array_start", "__init_array_end", Out::InitArray);
1313 Define("__fini_array_start", "__fini_array_end", Out::FiniArray);
1315 if (OutputSection *Sec = findSection(".ARM.exidx"))
1316 Define("__exidx_start", "__exidx_end", Sec);
1319 // If a section name is valid as a C identifier (which is rare because of
1320 // the leading '.'), linkers are expected to define __start_<secname> and
1321 // __stop_<secname> symbols. They are at beginning and end of the section,
1322 // respectively. This is not requested by the ELF standard, but GNU ld and
1323 // gold provide the feature, and used by many programs.
1324 template <class ELFT>
1325 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) {
1326 StringRef S = Sec->Name;
1327 if (!isValidCIdentifier(S))
1329 addOptionalRegular<ELFT>(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT);
1330 addOptionalRegular<ELFT>(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT);
1333 template <class ELFT>
1334 OutputSectionCommand *Writer<ELFT>::findSectionCommand(StringRef Name) {
1335 for (OutputSectionCommand *Cmd : OutputSectionCommands)
1336 if (Cmd->Name == Name)
1341 template <class ELFT> OutputSection *Writer<ELFT>::findSectionInScript(StringRef Name) {
1342 if (OutputSectionCommand *Cmd = findSectionCommand(Name))
1347 template <class ELFT> OutputSection *Writer<ELFT>::findSection(StringRef Name) {
1348 for (OutputSection *Sec : OutputSections)
1349 if (Sec->Name == Name)
1354 static bool needsPtLoad(OutputSection *Sec) {
1355 if (!(Sec->Flags & SHF_ALLOC))
1358 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is
1359 // responsible for allocating space for them, not the PT_LOAD that
1360 // contains the TLS initialization image.
1361 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS)
1366 // Linker scripts are responsible for aligning addresses. Unfortunately, most
1367 // linker scripts are designed for creating two PT_LOADs only, one RX and one
1368 // RW. This means that there is no alignment in the RO to RX transition and we
1369 // cannot create a PT_LOAD there.
1370 static uint64_t computeFlags(uint64_t Flags) {
1372 return PF_R | PF_W | PF_X;
1373 if (Config->SingleRoRx && !(Flags & PF_W))
1374 return Flags | PF_X;
1378 // Decide which program headers to create and which sections to include in each
1380 template <class ELFT> std::vector<PhdrEntry> Writer<ELFT>::createPhdrs() {
1381 std::vector<PhdrEntry> Ret;
1382 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * {
1383 Ret.emplace_back(Type, Flags);
1387 // The first phdr entry is PT_PHDR which describes the program header itself.
1388 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders);
1390 // PT_INTERP must be the second entry if exists.
1391 if (OutputSection *Sec = findSection(".interp"))
1392 AddHdr(PT_INTERP, Sec->getPhdrFlags())->add(Sec);
1394 // Add the first PT_LOAD segment for regular output sections.
1395 uint64_t Flags = computeFlags(PF_R);
1396 PhdrEntry *Load = AddHdr(PT_LOAD, Flags);
1398 // Add the headers. We will remove them if they don't fit.
1399 Load->add(Out::ElfHeader);
1400 Load->add(Out::ProgramHeaders);
1402 for (OutputSection *Sec : OutputSections) {
1403 if (!(Sec->Flags & SHF_ALLOC))
1405 if (!needsPtLoad(Sec))
1408 // Segments are contiguous memory regions that has the same attributes
1409 // (e.g. executable or writable). There is one phdr for each segment.
1410 // Therefore, we need to create a new phdr when the next section has
1411 // different flags or is loaded at a discontiguous address using AT linker
1413 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags());
1414 if (Script->hasLMA(Sec) || Flags != NewFlags) {
1415 Load = AddHdr(PT_LOAD, NewFlags);
1422 // Add a TLS segment if any.
1423 PhdrEntry TlsHdr(PT_TLS, PF_R);
1424 for (OutputSection *Sec : OutputSections)
1425 if (Sec->Flags & SHF_TLS)
1428 Ret.push_back(std::move(TlsHdr));
1430 // Add an entry for .dynamic.
1432 AddHdr(PT_DYNAMIC, InX::Dynamic->getParent()->getPhdrFlags())
1433 ->add(InX::Dynamic->getParent());
1435 // PT_GNU_RELRO includes all sections that should be marked as
1436 // read-only by dynamic linker after proccessing relocations.
1437 PhdrEntry RelRo(PT_GNU_RELRO, PF_R);
1438 for (OutputSection *Sec : OutputSections)
1439 if (needsPtLoad(Sec) && isRelroSection(Sec))
1442 Ret.push_back(std::move(RelRo));
1444 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr.
1445 if (!In<ELFT>::EhFrame->empty() && In<ELFT>::EhFrameHdr &&
1446 In<ELFT>::EhFrame->getParent() && In<ELFT>::EhFrameHdr->getParent())
1447 AddHdr(PT_GNU_EH_FRAME, In<ELFT>::EhFrameHdr->getParent()->getPhdrFlags())
1448 ->add(In<ELFT>::EhFrameHdr->getParent());
1450 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes
1451 // the dynamic linker fill the segment with random data.
1452 if (OutputSection *Sec = findSection(".openbsd.randomdata"))
1453 AddHdr(PT_OPENBSD_RANDOMIZE, Sec->getPhdrFlags())->add(Sec);
1455 // PT_GNU_STACK is a special section to tell the loader to make the
1456 // pages for the stack non-executable. If you really want an executable
1457 // stack, you can pass -z execstack, but that's not recommended for
1458 // security reasons.
1460 if (Config->ZExecstack)
1461 Perm = PF_R | PF_W | PF_X;
1464 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize;
1466 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable
1467 // is expected to perform W^X violations, such as calling mprotect(2) or
1468 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on
1470 if (Config->ZWxneeded)
1471 AddHdr(PT_OPENBSD_WXNEEDED, PF_X);
1473 // Create one PT_NOTE per a group of contiguous .note sections.
1474 PhdrEntry *Note = nullptr;
1475 for (OutputSection *Sec : OutputSections) {
1476 if (Sec->Type == SHT_NOTE) {
1477 if (!Note || Script->hasLMA(Sec))
1478 Note = AddHdr(PT_NOTE, PF_R);
1487 template <class ELFT>
1488 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry> &Phdrs) {
1489 if (Config->EMachine != EM_ARM)
1491 auto I = std::find_if(
1492 OutputSections.begin(), OutputSections.end(),
1493 [](OutputSection *Sec) { return Sec->Type == SHT_ARM_EXIDX; });
1494 if (I == OutputSections.end())
1497 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME
1498 PhdrEntry ARMExidx(PT_ARM_EXIDX, PF_R);
1500 Phdrs.push_back(ARMExidx);
1503 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the
1504 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic
1505 // linker can set the permissions.
1506 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() {
1507 for (const PhdrEntry &P : Phdrs)
1508 if (P.p_type == PT_LOAD && P.First)
1509 P.First->PageAlign = true;
1511 for (const PhdrEntry &P : Phdrs) {
1512 if (P.p_type != PT_GNU_RELRO)
1515 P.First->PageAlign = true;
1516 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we
1517 // have to align it to a page.
1518 auto End = OutputSections.end();
1519 auto I = std::find(OutputSections.begin(), End, P.Last);
1520 if (I == End || (I + 1) == End)
1522 OutputSection *Sec = *(I + 1);
1523 if (needsPtLoad(Sec))
1524 Sec->PageAlign = true;
1528 // Adjusts the file alignment for a given output section and returns
1529 // its new file offset. The file offset must be the same with its
1530 // virtual address (modulo the page size) so that the loader can load
1531 // executables without any address adjustment.
1532 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Sec) {
1533 OutputSection *First = Sec->FirstInPtLoad;
1534 // If the section is not in a PT_LOAD, we just have to align it.
1536 return alignTo(Off, Sec->Alignment);
1538 // The first section in a PT_LOAD has to have congruent offset and address
1539 // module the page size.
1541 return alignTo(Off, Config->MaxPageSize, Sec->Addr);
1543 // If two sections share the same PT_LOAD the file offset is calculated
1544 // using this formula: Off2 = Off1 + (VA2 - VA1).
1545 return First->Offset + Sec->Addr - First->Addr;
1548 static uint64_t setOffset(OutputSection *Sec, uint64_t Off) {
1549 if (Sec->Type == SHT_NOBITS) {
1554 Off = getFileAlignment(Off, Sec);
1556 return Off + Sec->Size;
1559 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() {
1561 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1562 OutputSection *Sec = Cmd->Sec;
1563 if (Sec->Flags & SHF_ALLOC)
1564 Off = setOffset(Sec, Off);
1566 FileSize = alignTo(Off, Config->Wordsize);
1569 // Assign file offsets to output sections.
1570 template <class ELFT> void Writer<ELFT>::assignFileOffsets() {
1572 Off = setOffset(Out::ElfHeader, Off);
1573 Off = setOffset(Out::ProgramHeaders, Off);
1575 for (OutputSectionCommand *Cmd : OutputSectionCommands)
1576 Off = setOffset(Cmd->Sec, Off);
1578 SectionHeaderOff = alignTo(Off, Config->Wordsize);
1580 SectionHeaderOff + (OutputSectionCommands.size() + 1) * sizeof(Elf_Shdr);
1583 // Finalize the program headers. We call this function after we assign
1584 // file offsets and VAs to all sections.
1585 template <class ELFT> void Writer<ELFT>::setPhdrs() {
1586 for (PhdrEntry &P : Phdrs) {
1587 OutputSection *First = P.First;
1588 OutputSection *Last = P.Last;
1590 P.p_filesz = Last->Offset - First->Offset;
1591 if (Last->Type != SHT_NOBITS)
1592 P.p_filesz += Last->Size;
1593 P.p_memsz = Last->Addr + Last->Size - First->Addr;
1594 P.p_offset = First->Offset;
1595 P.p_vaddr = First->Addr;
1597 P.p_paddr = First->getLMA();
1599 if (P.p_type == PT_LOAD)
1600 P.p_align = Config->MaxPageSize;
1601 else if (P.p_type == PT_GNU_RELRO)
1604 // The TLS pointer goes after PT_TLS. At least glibc will align it,
1605 // so round up the size to make sure the offsets are correct.
1606 if (P.p_type == PT_TLS) {
1609 P.p_memsz = alignTo(P.p_memsz, P.p_align);
1614 // The entry point address is chosen in the following ways.
1616 // 1. the '-e' entry command-line option;
1617 // 2. the ENTRY(symbol) command in a linker control script;
1618 // 3. the value of the symbol start, if present;
1619 // 4. the address of the first byte of the .text section, if present;
1620 // 5. the address 0.
1621 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() {
1622 // Case 1, 2 or 3. As a special case, if the symbol is actually
1623 // a number, we'll use that number as an address.
1624 if (SymbolBody *B = Symtab<ELFT>::X->find(Config->Entry))
1627 if (to_integer(Config->Entry, Addr))
1631 if (OutputSection *Sec = findSectionInScript(".text")) {
1632 if (Config->WarnMissingEntry)
1633 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" +
1634 utohexstr(Sec->Addr));
1639 if (Config->WarnMissingEntry)
1640 warn("cannot find entry symbol " + Config->Entry +
1641 "; not setting start address");
1645 static uint16_t getELFType() {
1648 if (Config->Relocatable)
1653 // This function is called after we have assigned address and size
1654 // to each section. This function fixes some predefined
1655 // symbol values that depend on section address and size.
1656 template <class ELFT> void Writer<ELFT>::fixPredefinedSymbols() {
1657 // _etext is the first location after the last read-only loadable segment.
1658 // _edata is the first location after the last read-write loadable segment.
1659 // _end is the first location after the uninitialized data region.
1660 PhdrEntry *Last = nullptr;
1661 PhdrEntry *LastRO = nullptr;
1662 PhdrEntry *LastRW = nullptr;
1663 for (PhdrEntry &P : Phdrs) {
1664 if (P.p_type != PT_LOAD)
1667 if (P.p_flags & PF_W)
1673 auto Set = [](DefinedRegular *S, OutputSection *Sec, uint64_t Value) {
1681 Set(ElfSym::End1, Last->First, Last->p_memsz);
1682 Set(ElfSym::End2, Last->First, Last->p_memsz);
1685 Set(ElfSym::Etext1, LastRO->First, LastRO->p_filesz);
1686 Set(ElfSym::Etext2, LastRO->First, LastRO->p_filesz);
1689 Set(ElfSym::Edata1, LastRW->First, LastRW->p_filesz);
1690 Set(ElfSym::Edata2, LastRW->First, LastRW->p_filesz);
1694 ElfSym::Bss->Section = findSectionInScript(".bss");
1696 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should
1697 // be equal to the _gp symbol's value.
1698 if (Config->EMachine == EM_MIPS && !ElfSym::MipsGp->Value) {
1699 // Find GP-relative section with the lowest address
1700 // and use this address to calculate default _gp value.
1701 for (const OutputSectionCommand *Cmd : OutputSectionCommands) {
1702 OutputSection *OS = Cmd->Sec;
1703 if (OS->Flags & SHF_MIPS_GPREL) {
1704 ElfSym::MipsGp->Value = OS->Addr + 0x7ff0;
1711 template <class ELFT> void Writer<ELFT>::writeHeader() {
1712 uint8_t *Buf = Buffer->getBufferStart();
1713 memcpy(Buf, "\177ELF", 4);
1715 // Write the ELF header.
1716 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf);
1717 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32;
1718 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB;
1719 EHdr->e_ident[EI_VERSION] = EV_CURRENT;
1720 EHdr->e_ident[EI_OSABI] = Config->OSABI;
1721 EHdr->e_type = getELFType();
1722 EHdr->e_machine = Config->EMachine;
1723 EHdr->e_version = EV_CURRENT;
1724 EHdr->e_entry = getEntryAddr();
1725 EHdr->e_shoff = SectionHeaderOff;
1726 EHdr->e_ehsize = sizeof(Elf_Ehdr);
1727 EHdr->e_phnum = Phdrs.size();
1728 EHdr->e_shentsize = sizeof(Elf_Shdr);
1729 EHdr->e_shnum = OutputSectionCommands.size() + 1;
1730 EHdr->e_shstrndx = InX::ShStrTab->getParent()->SectionIndex;
1732 if (Config->EMachine == EM_ARM)
1733 // We don't currently use any features incompatible with EF_ARM_EABI_VER5,
1734 // but we don't have any firm guarantees of conformance. Linux AArch64
1735 // kernels (as of 2016) require an EABI version to be set.
1736 EHdr->e_flags = EF_ARM_EABI_VER5;
1737 else if (Config->EMachine == EM_MIPS)
1738 EHdr->e_flags = getMipsEFlags<ELFT>();
1740 if (!Config->Relocatable) {
1741 EHdr->e_phoff = sizeof(Elf_Ehdr);
1742 EHdr->e_phentsize = sizeof(Elf_Phdr);
1745 // Write the program header table.
1746 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff);
1747 for (PhdrEntry &P : Phdrs) {
1748 HBuf->p_type = P.p_type;
1749 HBuf->p_flags = P.p_flags;
1750 HBuf->p_offset = P.p_offset;
1751 HBuf->p_vaddr = P.p_vaddr;
1752 HBuf->p_paddr = P.p_paddr;
1753 HBuf->p_filesz = P.p_filesz;
1754 HBuf->p_memsz = P.p_memsz;
1755 HBuf->p_align = P.p_align;
1759 // Write the section header table. Note that the first table entry is null.
1760 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff);
1761 for (OutputSectionCommand *Cmd : OutputSectionCommands)
1762 Cmd->Sec->writeHeaderTo<ELFT>(++SHdrs);
1765 // Open a result file.
1766 template <class ELFT> void Writer<ELFT>::openFile() {
1767 if (!Config->Is64 && FileSize > UINT32_MAX) {
1768 error("output file too large: " + Twine(FileSize) + " bytes");
1772 unlinkAsync(Config->OutputFile);
1773 ErrorOr<std::unique_ptr<FileOutputBuffer>> BufferOrErr =
1774 FileOutputBuffer::create(Config->OutputFile, FileSize,
1775 FileOutputBuffer::F_executable);
1777 if (auto EC = BufferOrErr.getError())
1778 error("failed to open " + Config->OutputFile + ": " + EC.message());
1780 Buffer = std::move(*BufferOrErr);
1783 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() {
1784 uint8_t *Buf = Buffer->getBufferStart();
1785 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1786 OutputSection *Sec = Cmd->Sec;
1787 if (Sec->Flags & SHF_ALLOC)
1788 Cmd->writeTo<ELFT>(Buf + Sec->Offset);
1792 // Write section contents to a mmap'ed file.
1793 template <class ELFT> void Writer<ELFT>::writeSections() {
1794 uint8_t *Buf = Buffer->getBufferStart();
1796 // PPC64 needs to process relocations in the .opd section
1797 // before processing relocations in code-containing sections.
1798 if (auto *OpdCmd = findSectionCommand(".opd")) {
1799 Out::Opd = OpdCmd->Sec;
1800 Out::OpdBuf = Buf + Out::Opd->Offset;
1801 OpdCmd->template writeTo<ELFT>(Buf + Out::Opd->Offset);
1804 OutputSection *EhFrameHdr =
1805 (In<ELFT>::EhFrameHdr && !In<ELFT>::EhFrameHdr->empty())
1806 ? In<ELFT>::EhFrameHdr->getParent()
1809 // In -r or -emit-relocs mode, write the relocation sections first as in
1810 // ELf_Rel targets we might find out that we need to modify the relocated
1811 // section while doing it.
1812 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1813 OutputSection *Sec = Cmd->Sec;
1814 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA)
1815 Cmd->writeTo<ELFT>(Buf + Sec->Offset);
1818 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1819 OutputSection *Sec = Cmd->Sec;
1820 if (Sec != Out::Opd && Sec != EhFrameHdr && Sec->Type != SHT_REL &&
1821 Sec->Type != SHT_RELA)
1822 Cmd->writeTo<ELFT>(Buf + Sec->Offset);
1825 // The .eh_frame_hdr depends on .eh_frame section contents, therefore
1826 // it should be written after .eh_frame is written.
1828 OutputSectionCommand *Cmd = Script->getCmd(EhFrameHdr);
1829 Cmd->writeTo<ELFT>(Buf + EhFrameHdr->Offset);
1833 template <class ELFT> void Writer<ELFT>::writeBuildId() {
1834 if (!InX::BuildId || !InX::BuildId->getParent())
1837 // Compute a hash of all sections of the output file.
1838 uint8_t *Start = Buffer->getBufferStart();
1839 uint8_t *End = Start + FileSize;
1840 InX::BuildId->writeBuildId({Start, End});
1843 template void elf::writeResult<ELF32LE>();
1844 template void elf::writeResult<ELF32BE>();
1845 template void elf::writeResult<ELF64LE>();
1846 template void elf::writeResult<ELF64BE>();