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 OutputSectionFactory Factory{OutputSections};
78 void addRelIpltSymbols();
79 void addStartEndSymbols();
80 void addStartStopSymbols(OutputSection *Sec);
81 uint64_t getEntryAddr();
82 OutputSection *findSection(StringRef Name);
83 OutputSection *findSectionInScript(StringRef Name);
84 OutputSectionCommand *findSectionCommand(StringRef Name);
86 std::vector<PhdrEntry> Phdrs;
89 uint64_t SectionHeaderOff;
91 } // anonymous namespace
93 StringRef elf::getOutputSectionName(StringRef Name) {
94 // ".zdebug_" is a prefix for ZLIB-compressed sections.
95 // Because we decompressed input sections, we want to remove 'z'.
96 if (Name.startswith(".zdebug_"))
97 return Saver.save("." + Name.substr(2));
99 if (Config->Relocatable)
103 {".text.", ".rodata.", ".data.rel.ro.", ".data.", ".bss.rel.ro.",
104 ".bss.", ".init_array.", ".fini_array.", ".ctors.", ".dtors.", ".tbss.",
105 ".gcc_except_table.", ".tdata.", ".ARM.exidx."}) {
106 StringRef Prefix = V.drop_back();
107 if (Name.startswith(V) || Name == Prefix)
111 // CommonSection is identified as "COMMON" in linker scripts.
112 // By default, it should go to .bss section.
113 if (Name == "COMMON")
119 template <class ELFT> static bool needsInterpSection() {
120 return !Symtab<ELFT>::X->getSharedFiles().empty() &&
121 !Config->DynamicLinker.empty() && !Script->ignoreInterpSection();
124 template <class ELFT> void elf::writeResult() { Writer<ELFT>().run(); }
126 template <class ELFT> void Writer<ELFT>::removeEmptyPTLoad() {
127 auto I = std::remove_if(Phdrs.begin(), Phdrs.end(), [&](const PhdrEntry &P) {
128 if (P.p_type != PT_LOAD)
132 uint64_t Size = P.Last->Addr + P.Last->Size - P.First->Addr;
135 Phdrs.erase(I, Phdrs.end());
138 template <class ELFT> static void combineEhFrameSections() {
139 for (InputSectionBase *&S : InputSections) {
140 EhInputSection *ES = dyn_cast<EhInputSection>(S);
141 if (!ES || !ES->Live)
144 In<ELFT>::EhFrame->addSection(ES);
148 std::vector<InputSectionBase *> &V = InputSections;
149 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end());
152 template <class ELFT> void Writer<ELFT>::clearOutputSections() {
153 if (Script->Opt.HasSections)
154 Script->createOrphanCommands();
156 Script->fabricateDefaultCommands();
157 // Clear the OutputSections to make sure it is not used anymore. Any
158 // code from this point on should be using the linker script
160 for (OutputSection *Sec : OutputSections)
161 Sec->Sections.clear();
162 OutputSections.clear();
165 // The main function of the writer.
166 template <class ELFT> void Writer<ELFT>::run() {
167 // Create linker-synthesized sections such as .got or .plt.
168 // Such sections are of type input section.
169 createSyntheticSections();
171 if (!Config->Relocatable)
172 combineEhFrameSections<ELFT>();
174 // We need to create some reserved symbols such as _end. Create them.
175 if (!Config->Relocatable)
176 addReservedSymbols();
178 // Create output sections.
179 if (Script->Opt.HasSections) {
180 // If linker script contains SECTIONS commands, let it create sections.
181 Script->processCommands(Factory);
183 // Linker scripts may have left some input sections unassigned.
184 // Assign such sections using the default rule.
185 Script->addOrphanSections(Factory);
187 // If linker script does not contain SECTIONS commands, create
188 // output sections by default rules. We still need to give the
189 // linker script a chance to run, because it might contain
190 // non-SECTIONS commands such as ASSERT.
192 Script->processCommands(Factory);
195 if (Config->Discard != DiscardPolicy::All)
198 if (Config->CopyRelocs)
201 // Now that we have a complete set of output sections. This function
202 // completes section contents. For example, we need to add strings
203 // to the string table, and add entries to .got and .plt.
204 // finalizeSections does that.
209 if (!Script->Opt.HasSections && !Config->Relocatable)
210 fixSectionAlignments();
212 // If -compressed-debug-sections is specified, we need to compress
213 // .debug_* sections. Do it right now because it changes the size of
216 OutputSectionCommands.begin(), OutputSectionCommands.end(),
217 [](OutputSectionCommand *Cmd) { Cmd->maybeCompress<ELFT>(); });
219 Script->assignAddresses(Phdrs);
221 // Remove empty PT_LOAD to avoid causing the dynamic linker to try to mmap a
222 // 0 sized region. This has to be done late since only after assignAddresses
223 // we know the size of the sections.
226 if (!Config->OFormatBinary)
229 assignFileOffsetsBinary();
233 if (Config->Relocatable) {
234 for (OutputSectionCommand *Cmd : OutputSectionCommands)
237 fixPredefinedSymbols();
240 // It does not make sense try to open the file if we have error already.
243 // Write the result down to a file.
248 if (!Config->OFormatBinary) {
252 writeSectionsBinary();
255 // Backfill .note.gnu.build-id section content. This is done at last
256 // because the content is usually a hash value of the entire output file.
262 // Handle -Map option.
263 writeMapFile<ELFT>(OutputSectionCommands);
267 if (auto EC = Buffer->commit())
268 error("failed to write to the output file: " + EC.message());
270 // Flush the output streams and exit immediately. A full shutdown
271 // is a good test that we are keeping track of all allocated memory,
272 // but actually freeing it is a waste of time in a regular linker run.
273 if (Config->ExitEarly)
277 // Initialize Out members.
278 template <class ELFT> void Writer<ELFT>::createSyntheticSections() {
279 // Initialize all pointers with NULL. This is needed because
280 // you can call lld::elf::main more than once as a library.
281 memset(&Out::First, 0, sizeof(Out));
283 auto Add = [](InputSectionBase *Sec) { InputSections.push_back(Sec); };
285 InX::DynStrTab = make<StringTableSection>(".dynstr", true);
286 InX::Dynamic = make<DynamicSection<ELFT>>();
287 In<ELFT>::RelaDyn = make<RelocationSection<ELFT>>(
288 Config->IsRela ? ".rela.dyn" : ".rel.dyn", Config->ZCombreloc);
289 InX::ShStrTab = make<StringTableSection>(".shstrtab", false);
291 Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC);
292 Out::ElfHeader->Size = sizeof(Elf_Ehdr);
293 Out::ProgramHeaders = make<OutputSection>("", 0, SHF_ALLOC);
294 Out::ProgramHeaders->updateAlignment(Config->Wordsize);
296 if (needsInterpSection<ELFT>()) {
297 InX::Interp = createInterpSection();
300 InX::Interp = nullptr;
303 if (Config->Strip != StripPolicy::All) {
304 InX::StrTab = make<StringTableSection>(".strtab", false);
305 InX::SymTab = make<SymbolTableSection<ELFT>>(*InX::StrTab);
308 if (Config->BuildId != BuildIdKind::None) {
309 InX::BuildId = make<BuildIdSection>();
313 InX::Common = createCommonSection<ELFT>();
317 InX::Bss = make<BssSection>(".bss");
319 InX::BssRelRo = make<BssSection>(".bss.rel.ro");
322 // Add MIPS-specific sections.
323 bool HasDynSymTab = !Symtab<ELFT>::X->getSharedFiles().empty() ||
324 Config->Pic || Config->ExportDynamic;
325 if (Config->EMachine == EM_MIPS) {
326 if (!Config->Shared && HasDynSymTab) {
327 InX::MipsRldMap = make<MipsRldMapSection>();
328 Add(InX::MipsRldMap);
330 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create())
332 if (auto *Sec = MipsOptionsSection<ELFT>::create())
334 if (auto *Sec = MipsReginfoSection<ELFT>::create())
339 InX::DynSymTab = make<SymbolTableSection<ELFT>>(*InX::DynStrTab);
342 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>();
343 Add(In<ELFT>::VerSym);
345 if (!Config->VersionDefinitions.empty()) {
346 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>();
347 Add(In<ELFT>::VerDef);
350 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>();
351 Add(In<ELFT>::VerNeed);
353 if (Config->GnuHash) {
354 InX::GnuHashTab = make<GnuHashTableSection>();
355 Add(InX::GnuHashTab);
358 if (Config->SysvHash) {
359 In<ELFT>::HashTab = make<HashTableSection<ELFT>>();
360 Add(In<ELFT>::HashTab);
365 Add(In<ELFT>::RelaDyn);
368 // Add .got. MIPS' .got is so different from the other archs,
369 // it has its own class.
370 if (Config->EMachine == EM_MIPS) {
371 InX::MipsGot = make<MipsGotSection>();
374 InX::Got = make<GotSection>();
378 InX::GotPlt = make<GotPltSection>();
380 InX::IgotPlt = make<IgotPltSection>();
383 if (Config->GdbIndex) {
384 InX::GdbIndex = make<GdbIndexSection>();
388 // We always need to add rel[a].plt to output if it has entries.
389 // Even for static linking it can contain R_[*]_IRELATIVE relocations.
390 In<ELFT>::RelaPlt = make<RelocationSection<ELFT>>(
391 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/);
392 Add(In<ELFT>::RelaPlt);
394 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure
395 // that the IRelative relocations are processed last by the dynamic loader
396 In<ELFT>::RelaIplt = make<RelocationSection<ELFT>>(
397 (Config->EMachine == EM_ARM) ? ".rel.dyn" : In<ELFT>::RelaPlt->Name,
399 Add(In<ELFT>::RelaIplt);
401 InX::Plt = make<PltSection>(Target->PltHeaderSize);
403 InX::Iplt = make<PltSection>(0);
406 if (!Config->Relocatable) {
407 if (Config->EhFrameHdr) {
408 In<ELFT>::EhFrameHdr = make<EhFrameHeader<ELFT>>();
409 Add(In<ELFT>::EhFrameHdr);
411 In<ELFT>::EhFrame = make<EhFrameSection<ELFT>>();
412 Add(In<ELFT>::EhFrame);
422 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName,
423 const SymbolBody &B) {
424 if (B.isFile() || B.isSection())
427 // If sym references a section in a discarded group, don't keep it.
428 if (Sec == &InputSection::Discarded)
431 if (Config->Discard == DiscardPolicy::None)
434 // In ELF assembly .L symbols are normally discarded by the assembler.
435 // If the assembler fails to do so, the linker discards them if
436 // * --discard-locals is used.
437 // * The symbol is in a SHF_MERGE section, which is normally the reason for
438 // the assembler keeping the .L symbol.
439 if (!SymName.startswith(".L") && !SymName.empty())
442 if (Config->Discard == DiscardPolicy::Locals)
445 return !Sec || !(Sec->Flags & SHF_MERGE);
448 static bool includeInSymtab(const SymbolBody &B) {
449 if (!B.isLocal() && !B.symbol()->IsUsedInRegularObj)
452 if (auto *D = dyn_cast<DefinedRegular>(&B)) {
453 // Always include absolute symbols.
454 SectionBase *Sec = D->Section;
457 if (auto *IS = dyn_cast<InputSectionBase>(Sec)) {
459 IS = cast<InputSectionBase>(Sec);
460 // Exclude symbols pointing to garbage-collected sections.
464 if (auto *S = dyn_cast<MergeInputSection>(Sec))
465 if (!S->getSectionPiece(D->Value)->Live)
471 // Local symbols are not in the linker's symbol table. This function scans
472 // each object file's symbol table to copy local symbols to the output.
473 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() {
476 for (elf::ObjectFile<ELFT> *F : Symtab<ELFT>::X->getObjectFiles()) {
477 for (SymbolBody *B : F->getLocalSymbols()) {
480 ": broken object: getLocalSymbols returns a non-local symbol");
481 auto *DR = dyn_cast<DefinedRegular>(B);
483 // No reason to keep local undefined symbol in symtab.
486 if (!includeInSymtab(*B))
489 SectionBase *Sec = DR->Section;
490 if (!shouldKeepInSymtab(Sec, B->getName(), *B))
492 InX::SymTab->addSymbol(B);
497 template <class ELFT> void Writer<ELFT>::addSectionSymbols() {
498 // Create one STT_SECTION symbol for each output section we might
499 // have a relocation with.
500 for (OutputSection *Sec : OutputSections) {
501 if (Sec->Sections.empty())
504 InputSection *IS = Sec->Sections[0];
505 if (isa<SyntheticSection>(IS) || IS->Type == SHT_REL ||
506 IS->Type == SHT_RELA)
510 make<DefinedRegular>("", /*IsLocal=*/true, /*StOther=*/0, STT_SECTION,
511 /*Value=*/0, /*Size=*/0, IS, nullptr);
512 InX::SymTab->addSymbol(Sym);
516 // Today's loaders have a feature to make segments read-only after
517 // processing dynamic relocations to enhance security. PT_GNU_RELRO
518 // is defined for that.
520 // This function returns true if a section needs to be put into a
521 // PT_GNU_RELRO segment.
522 bool elf::isRelroSection(const OutputSection *Sec) {
526 uint64_t Flags = Sec->Flags;
528 // Non-allocatable or non-writable sections don't need RELRO because
529 // they are not writable or not even mapped to memory in the first place.
530 // RELRO is for sections that are essentially read-only but need to
531 // be writable only at process startup to allow dynamic linker to
532 // apply relocations.
533 if (!(Flags & SHF_ALLOC) || !(Flags & SHF_WRITE))
536 // Once initialized, TLS data segments are used as data templates
537 // for a thread-local storage. For each new thread, runtime
538 // allocates memory for a TLS and copy templates there. No thread
539 // are supposed to use templates directly. Thus, it can be in RELRO.
543 // .init_array, .preinit_array and .fini_array contain pointers to
544 // functions that are executed on process startup or exit. These
545 // pointers are set by the static linker, and they are not expected
546 // to change at runtime. But if you are an attacker, you could do
547 // interesting things by manipulating pointers in .fini_array, for
548 // example. So they are put into RELRO.
549 uint32_t Type = Sec->Type;
550 if (Type == SHT_INIT_ARRAY || Type == SHT_FINI_ARRAY ||
551 Type == SHT_PREINIT_ARRAY)
554 // .got contains pointers to external symbols. They are resolved by
555 // the dynamic linker when a module is loaded into memory, and after
556 // that they are not expected to change. So, it can be in RELRO.
557 if (InX::Got && Sec == InX::Got->getParent())
560 // .got.plt contains pointers to external function symbols. They are
561 // by default resolved lazily, so we usually cannot put it into RELRO.
562 // However, if "-z now" is given, the lazy symbol resolution is
563 // disabled, which enables us to put it into RELRO.
564 if (Sec == InX::GotPlt->getParent())
567 // .dynamic section contains data for the dynamic linker, and
568 // there's no need to write to it at runtime, so it's better to put
570 if (Sec == InX::Dynamic->getParent())
573 // .bss.rel.ro is used for copy relocations for read-only symbols.
574 // Since the dynamic linker needs to process copy relocations, the
575 // section cannot be read-only, but once initialized, they shouldn't
577 if (Sec == InX::BssRelRo->getParent())
580 // Sections with some special names are put into RELRO. This is a
581 // bit unfortunate because section names shouldn't be significant in
582 // ELF in spirit. But in reality many linker features depend on
583 // magic section names.
584 StringRef S = Sec->Name;
585 return S == ".data.rel.ro" || S == ".ctors" || S == ".dtors" || S == ".jcr" ||
586 S == ".eh_frame" || S == ".openbsd.randomdata";
589 // We compute a rank for each section. The rank indicates where the
590 // section should be placed in the file. Instead of using simple
591 // numbers (0,1,2...), we use a series of flags. One for each decision
592 // point when placing the section.
593 // Using flags has two key properties:
594 // * It is easy to check if a give branch was taken.
595 // * It is easy two see how similar two ranks are (see getRankProximity).
597 RF_NOT_ADDR_SET = 1 << 16,
598 RF_NOT_INTERP = 1 << 15,
599 RF_NOT_ALLOC = 1 << 14,
601 RF_EXEC_WRITE = 1 << 12,
603 RF_NON_TLS_BSS = 1 << 10,
604 RF_NON_TLS_BSS_RO = 1 << 9,
607 RF_PPC_NOT_TOCBSS = 1 << 6,
609 RF_PPC_TOCL = 1 << 4,
611 RF_PPC_BRANCH_LT = 1 << 2,
612 RF_MIPS_GPREL = 1 << 1,
613 RF_MIPS_NOT_GOT = 1 << 0
616 static unsigned getSectionRank(const OutputSection *Sec) {
619 // We want to put section specified by -T option first, so we
620 // can start assigning VA starting from them later.
621 if (Config->SectionStartMap.count(Sec->Name))
623 Rank |= RF_NOT_ADDR_SET;
625 // Put .interp first because some loaders want to see that section
626 // on the first page of the executable file when loaded into memory.
627 if (Sec->Name == ".interp")
629 Rank |= RF_NOT_INTERP;
631 // Allocatable sections go first to reduce the total PT_LOAD size and
632 // so debug info doesn't change addresses in actual code.
633 if (!(Sec->Flags & SHF_ALLOC))
634 return Rank | RF_NOT_ALLOC;
636 // Sort sections based on their access permission in the following
637 // order: R, RX, RWX, RW. This order is based on the following
639 // * Read-only sections come first such that they go in the
640 // PT_LOAD covering the program headers at the start of the file.
641 // * Read-only, executable sections come next, unless the
642 // -no-rosegment option is used.
643 // * Writable, executable sections follow such that .plt on
644 // architectures where it needs to be writable will be placed
645 // between .text and .data.
646 // * Writable sections come last, such that .bss lands at the very
647 // end of the last PT_LOAD.
648 bool IsExec = Sec->Flags & SHF_EXECINSTR;
649 bool IsWrite = Sec->Flags & SHF_WRITE;
653 Rank |= RF_EXEC_WRITE;
654 else if (!Config->SingleRoRx)
661 // If we got here we know that both A and B are in the same PT_LOAD.
663 bool IsTls = Sec->Flags & SHF_TLS;
664 bool IsNoBits = Sec->Type == SHT_NOBITS;
666 // The first requirement we have is to put (non-TLS) nobits sections last. The
667 // reason is that the only thing the dynamic linker will see about them is a
668 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the
669 // PT_LOAD, so that has to correspond to the nobits sections.
670 bool IsNonTlsNoBits = IsNoBits && !IsTls;
672 Rank |= RF_NON_TLS_BSS;
674 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo
675 // sections after r/w ones, so that the RelRo sections are contiguous.
676 bool IsRelRo = isRelroSection(Sec);
677 if (IsNonTlsNoBits && !IsRelRo)
678 Rank |= RF_NON_TLS_BSS_RO;
679 if (!IsNonTlsNoBits && IsRelRo)
680 Rank |= RF_NON_TLS_BSS_RO;
682 // The TLS initialization block needs to be a single contiguous block in a R/W
683 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W
684 // sections. The TLS NOBITS sections are placed here as they don't take up
685 // virtual address space in the PT_LOAD.
689 // Within the TLS initialization block, the non-nobits sections need to appear
694 // // Some architectures have additional ordering restrictions for sections
695 // // within the same PT_LOAD.
696 if (Config->EMachine == EM_PPC64) {
697 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections
698 // that we would like to make sure appear is a specific order to maximize
699 // their coverage by a single signed 16-bit offset from the TOC base
700 // pointer. Conversely, the special .tocbss section should be first among
701 // all SHT_NOBITS sections. This will put it next to the loaded special
702 // PPC64 sections (and, thus, within reach of the TOC base pointer).
703 StringRef Name = Sec->Name;
704 if (Name != ".tocbss")
705 Rank |= RF_PPC_NOT_TOCBSS;
716 if (Name == ".branch_lt")
717 Rank |= RF_PPC_BRANCH_LT;
719 if (Config->EMachine == EM_MIPS) {
720 // All sections with SHF_MIPS_GPREL flag should be grouped together
721 // because data in these sections is addressable with a gp relative address.
722 if (Sec->Flags & SHF_MIPS_GPREL)
723 Rank |= RF_MIPS_GPREL;
725 if (Sec->Name != ".got")
726 Rank |= RF_MIPS_NOT_GOT;
732 static bool compareSections(const BaseCommand *ACmd, const BaseCommand *BCmd) {
733 const OutputSection *A = cast<OutputSectionCommand>(ACmd)->Sec;
734 const OutputSection *B = cast<OutputSectionCommand>(BCmd)->Sec;
735 if (A->SortRank != B->SortRank)
736 return A->SortRank < B->SortRank;
737 if (!(A->SortRank & RF_NOT_ADDR_SET))
738 return Config->SectionStartMap.lookup(A->Name) <
739 Config->SectionStartMap.lookup(B->Name);
743 void PhdrEntry::add(OutputSection *Sec) {
747 p_align = std::max(p_align, Sec->Alignment);
748 if (p_type == PT_LOAD)
749 Sec->FirstInPtLoad = First;
752 template <class ELFT>
753 static Symbol *addRegular(StringRef Name, SectionBase *Sec, uint64_t Value,
754 uint8_t StOther = STV_HIDDEN,
755 uint8_t Binding = STB_WEAK) {
756 // The linker generated symbols are added as STB_WEAK to allow user defined
757 // ones to override them.
758 return Symtab<ELFT>::X->addRegular(Name, StOther, STT_NOTYPE, Value,
759 /*Size=*/0, Binding, Sec,
763 template <class ELFT>
764 static DefinedRegular *
765 addOptionalRegular(StringRef Name, SectionBase *Sec, uint64_t Val,
766 uint8_t StOther = STV_HIDDEN, uint8_t Binding = STB_GLOBAL) {
767 SymbolBody *S = Symtab<ELFT>::X->find(Name);
770 if (S->isInCurrentDSO())
772 return cast<DefinedRegular>(
773 addRegular<ELFT>(Name, Sec, Val, StOther, Binding)->body());
776 // The beginning and the ending of .rel[a].plt section are marked
777 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked
778 // executable. The runtime needs these symbols in order to resolve
779 // all IRELATIVE relocs on startup. For dynamic executables, we don't
780 // need these symbols, since IRELATIVE relocs are resolved through GOT
781 // and PLT. For details, see http://www.airs.com/blog/archives/403.
782 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() {
785 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start";
786 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, 0, STV_HIDDEN, STB_WEAK);
788 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end";
789 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, -1, STV_HIDDEN, STB_WEAK);
792 // The linker is expected to define some symbols depending on
793 // the linking result. This function defines such symbols.
794 template <class ELFT> void Writer<ELFT>::addReservedSymbols() {
795 if (Config->EMachine == EM_MIPS) {
796 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer
797 // so that it points to an absolute address which by default is relative
798 // to GOT. Default offset is 0x7ff0.
799 // See "Global Data Symbols" in Chapter 6 in the following document:
800 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf
801 ElfSym::MipsGp = Symtab<ELFT>::X->addAbsolute("_gp", STV_HIDDEN, STB_LOCAL);
803 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between
804 // start of function and 'gp' pointer into GOT.
805 if (Symtab<ELFT>::X->find("_gp_disp"))
807 Symtab<ELFT>::X->addAbsolute("_gp_disp", STV_HIDDEN, STB_LOCAL);
809 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp'
810 // pointer. This symbol is used in the code generated by .cpload pseudo-op
811 // in case of using -mno-shared option.
812 // https://sourceware.org/ml/binutils/2004-12/msg00094.html
813 if (Symtab<ELFT>::X->find("__gnu_local_gp"))
814 ElfSym::MipsLocalGp =
815 Symtab<ELFT>::X->addAbsolute("__gnu_local_gp", STV_HIDDEN, STB_LOCAL);
818 // In the assembly for 32 bit x86 the _GLOBAL_OFFSET_TABLE_ symbol
819 // is magical and is used to produce a R_386_GOTPC relocation.
820 // The R_386_GOTPC relocation value doesn't actually depend on the
821 // symbol value, so it could use an index of STN_UNDEF which, according
822 // to the spec, means the symbol value is 0.
823 // Unfortunately both gas and MC keep the _GLOBAL_OFFSET_TABLE_ symbol in
825 // The situation is even stranger on x86_64 where the assembly doesn't
826 // need the magical symbol, but gas still puts _GLOBAL_OFFSET_TABLE_ as
827 // an undefined symbol in the .o files.
828 // Given that the symbol is effectively unused, we just create a dummy
829 // hidden one to avoid the undefined symbol error.
830 Symtab<ELFT>::X->addIgnored("_GLOBAL_OFFSET_TABLE_");
832 // __tls_get_addr is defined by the dynamic linker for dynamic ELFs. For
833 // static linking the linker is required to optimize away any references to
834 // __tls_get_addr, so it's not defined anywhere. Create a hidden definition
835 // to avoid the undefined symbol error.
837 Symtab<ELFT>::X->addIgnored("__tls_get_addr");
839 // __ehdr_start is the location of ELF file headers. Note that we define
840 // this symbol unconditionally even when using a linker script, which
841 // differs from the behavior implemented by GNU linker which only define
842 // this symbol if ELF headers are in the memory mapped segment.
843 // __executable_start is not documented, but the expectation of at
844 // least the android libc is that it points to the elf header too.
845 // __dso_handle symbol is passed to cxa_finalize as a marker to identify
846 // each DSO. The address of the symbol doesn't matter as long as they are
847 // different in different DSOs, so we chose the start address of the DSO.
848 for (const char *Name :
849 {"__ehdr_start", "__executable_start", "__dso_handle"})
850 addOptionalRegular<ELFT>(Name, Out::ElfHeader, 0, STV_HIDDEN);
852 // If linker script do layout we do not need to create any standart symbols.
853 if (Script->Opt.HasSections)
856 auto Add = [](StringRef S) {
857 return addOptionalRegular<ELFT>(S, Out::ElfHeader, 0, STV_DEFAULT);
860 ElfSym::Bss = Add("__bss_start");
861 ElfSym::End1 = Add("end");
862 ElfSym::End2 = Add("_end");
863 ElfSym::Etext1 = Add("etext");
864 ElfSym::Etext2 = Add("_etext");
865 ElfSym::Edata1 = Add("edata");
866 ElfSym::Edata2 = Add("_edata");
869 // Sort input sections by section name suffixes for
870 // __attribute__((init_priority(N))).
871 static void sortInitFini(OutputSection *S) {
873 reinterpret_cast<OutputSection *>(S)->sortInitFini();
876 // Sort input sections by the special rule for .ctors and .dtors.
877 static void sortCtorsDtors(OutputSection *S) {
879 reinterpret_cast<OutputSection *>(S)->sortCtorsDtors();
882 // Sort input sections using the list provided by --symbol-ordering-file.
883 template <class ELFT>
884 static void sortBySymbolsOrder(ArrayRef<OutputSection *> OutputSections) {
885 if (Config->SymbolOrderingFile.empty())
888 // Build a map from symbols to their priorities. Symbols that didn't
889 // appear in the symbol ordering file have the lowest priority 0.
890 // All explicitly mentioned symbols have negative (higher) priorities.
891 DenseMap<StringRef, int> SymbolOrder;
892 int Priority = -Config->SymbolOrderingFile.size();
893 for (StringRef S : Config->SymbolOrderingFile)
894 SymbolOrder.insert({S, Priority++});
896 // Build a map from sections to their priorities.
897 DenseMap<SectionBase *, int> SectionOrder;
898 for (elf::ObjectFile<ELFT> *File : Symtab<ELFT>::X->getObjectFiles()) {
899 for (SymbolBody *Body : File->getSymbols()) {
900 auto *D = dyn_cast<DefinedRegular>(Body);
901 if (!D || !D->Section)
903 int &Priority = SectionOrder[D->Section];
904 Priority = std::min(Priority, SymbolOrder.lookup(D->getName()));
908 // Sort sections by priority.
909 for (OutputSection *Base : OutputSections)
910 if (auto *Sec = dyn_cast<OutputSection>(Base))
911 Sec->sort([&](InputSectionBase *S) { return SectionOrder.lookup(S); });
914 template <class ELFT>
915 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) {
916 for (InputSectionBase *IS : InputSections) {
919 // Scan all relocations. Each relocation goes through a series
920 // of tests to determine if it needs special treatment, such as
921 // creating GOT, PLT, copy relocations, etc.
922 // Note that relocations for non-alloc sections are directly
923 // processed by InputSection::relocateNonAlloc.
924 if (!(IS->Flags & SHF_ALLOC))
926 if (isa<InputSection>(IS) || isa<EhInputSection>(IS))
930 if (!Config->Relocatable) {
931 for (EhInputSection *ES : In<ELFT>::EhFrame->Sections)
936 template <class ELFT> void Writer<ELFT>::createSections() {
937 for (InputSectionBase *IS : InputSections)
939 Factory.addInputSec(IS, getOutputSectionName(IS->Name));
941 sortBySymbolsOrder<ELFT>(OutputSections);
942 sortInitFini(findSection(".init_array"));
943 sortInitFini(findSection(".fini_array"));
944 sortCtorsDtors(findSection(".ctors"));
945 sortCtorsDtors(findSection(".dtors"));
948 // We want to find how similar two ranks are.
949 // The more branches in getSectionRank that match, the more similar they are.
950 // Since each branch corresponds to a bit flag, we can just use
951 // countLeadingZeros.
952 static int getRankProximity(OutputSection *A, OutputSection *B) {
953 return countLeadingZeros(A->SortRank ^ B->SortRank);
956 static int getRankProximity(OutputSection *A, BaseCommand *B) {
957 if (auto *Cmd = dyn_cast<OutputSectionCommand>(B))
959 return getRankProximity(A, Cmd->Sec);
963 // When placing orphan sections, we want to place them after symbol assignments
964 // so that an orphan after
968 // doesn't break the intended meaning of the begin/end symbols.
969 // We don't want to go over sections since findOrphanPos is the
970 // one in charge of deciding the order of the sections.
971 // We don't want to go over changes to '.', since doing so in
972 // rx_sec : { *(rx_sec) }
973 // . = ALIGN(0x1000);
974 // /* The RW PT_LOAD starts here*/
975 // rw_sec : { *(rw_sec) }
976 // would mean that the RW PT_LOAD would become unaligned.
977 static bool shouldSkip(BaseCommand *Cmd) {
978 if (isa<OutputSectionCommand>(Cmd))
980 if (auto *Assign = dyn_cast<SymbolAssignment>(Cmd))
981 return Assign->Name != ".";
985 // We want to place orphan sections so that they share as much
986 // characteristics with their neighbors as possible. For example, if
987 // both are rw, or both are tls.
988 template <typename ELFT>
989 static std::vector<BaseCommand *>::iterator
990 findOrphanPos(std::vector<BaseCommand *>::iterator B,
991 std::vector<BaseCommand *>::iterator E) {
992 OutputSection *Sec = cast<OutputSectionCommand>(*E)->Sec;
994 // Find the first element that has as close a rank as possible.
995 auto I = std::max_element(B, E, [=](BaseCommand *A, BaseCommand *B) {
996 return getRankProximity(Sec, A) < getRankProximity(Sec, B);
1001 // Consider all existing sections with the same proximity.
1002 int Proximity = getRankProximity(Sec, *I);
1003 for (; I != E; ++I) {
1004 auto *Cmd = dyn_cast<OutputSectionCommand>(*I);
1005 if (!Cmd || !Cmd->Sec)
1007 if (getRankProximity(Sec, Cmd->Sec) != Proximity ||
1008 Sec->SortRank < Cmd->Sec->SortRank)
1011 auto J = std::find_if(
1012 llvm::make_reverse_iterator(I), llvm::make_reverse_iterator(B),
1013 [](BaseCommand *Cmd) { return isa<OutputSectionCommand>(Cmd); });
1015 while (I != E && shouldSkip(*I))
1020 template <class ELFT> void Writer<ELFT>::sortSections() {
1021 // Don't sort if using -r. It is not necessary and we want to preserve the
1022 // relative order for SHF_LINK_ORDER sections.
1023 if (Config->Relocatable)
1026 if (Script->Opt.HasSections)
1027 Script->adjustSectionsBeforeSorting();
1029 for (BaseCommand *Base : Script->Opt.Commands)
1030 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
1031 if (OutputSection *Sec = Cmd->Sec)
1032 Sec->SortRank = getSectionRank(Sec);
1034 if (!Script->Opt.HasSections) {
1035 // We know that all the OutputSectionCommands are contiguous in
1037 auto E = Script->Opt.Commands.end();
1038 auto I = Script->Opt.Commands.begin();
1039 auto IsSection = [](BaseCommand *Base) {
1040 return isa<OutputSectionCommand>(Base);
1042 I = std::find_if(I, E, IsSection);
1043 E = std::find_if(llvm::make_reverse_iterator(E),
1044 llvm::make_reverse_iterator(I), IsSection)
1046 std::stable_sort(I, E, compareSections);
1050 // Orphan sections are sections present in the input files which are
1051 // not explicitly placed into the output file by the linker script.
1053 // The sections in the linker script are already in the correct
1054 // order. We have to figuere out where to insert the orphan
1057 // The order of the sections in the script is arbitrary and may not agree with
1058 // compareSections. This means that we cannot easily define a strict weak
1059 // ordering. To see why, consider a comparison of a section in the script and
1060 // one not in the script. We have a two simple options:
1061 // * Make them equivalent (a is not less than b, and b is not less than a).
1062 // The problem is then that equivalence has to be transitive and we can
1063 // have sections a, b and c with only b in a script and a less than c
1064 // which breaks this property.
1065 // * Use compareSectionsNonScript. Given that the script order doesn't have
1066 // to match, we can end up with sections a, b, c, d where b and c are in the
1067 // script and c is compareSectionsNonScript less than b. In which case d
1068 // can be equivalent to c, a to b and d < a. As a concrete example:
1069 // .a (rx) # not in script
1070 // .b (rx) # in script
1071 // .c (ro) # in script
1072 // .d (ro) # not in script
1074 // The way we define an order then is:
1075 // * Sort only the orphan sections. They are in the end right now.
1076 // * Move each orphan section to its preferred position. We try
1077 // to put each section in the last position where it it can share
1080 // There is some ambiguity as to where exactly a new entry should be
1081 // inserted, because Opt.Commands contains not only output section
1082 // commands but also other types of commands such as symbol assignment
1083 // expressions. There's no correct answer here due to the lack of the
1084 // formal specification of the linker script. We use heuristics to
1085 // determine whether a new output command should be added before or
1086 // after another commands. For the details, look at shouldSkip
1089 auto I = Script->Opt.Commands.begin();
1090 auto E = Script->Opt.Commands.end();
1091 auto NonScriptI = std::find_if(I, E, [](BaseCommand *Base) {
1092 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
1093 return Cmd->Sec && Cmd->Sec->SectionIndex == INT_MAX;
1097 // Sort the orphan sections.
1098 std::stable_sort(NonScriptI, E, compareSections);
1100 // As a horrible special case, skip the first . assignment if it is before any
1101 // section. We do this because it is common to set a load address by starting
1102 // the script with ". = 0xabcd" and the expectation is that every section is
1104 auto FirstSectionOrDotAssignment =
1105 std::find_if(I, E, [](BaseCommand *Cmd) { return !shouldSkip(Cmd); });
1106 if (FirstSectionOrDotAssignment != E &&
1107 isa<SymbolAssignment>(**FirstSectionOrDotAssignment))
1108 ++FirstSectionOrDotAssignment;
1109 I = FirstSectionOrDotAssignment;
1111 while (NonScriptI != E) {
1112 auto Pos = findOrphanPos<ELFT>(I, NonScriptI);
1113 OutputSection *Orphan = cast<OutputSectionCommand>(*NonScriptI)->Sec;
1115 // As an optimization, find all sections with the same sort rank
1116 // and insert them with one rotate.
1117 unsigned Rank = Orphan->SortRank;
1118 auto End = std::find_if(NonScriptI + 1, E, [=](BaseCommand *Cmd) {
1119 return cast<OutputSectionCommand>(Cmd)->Sec->SortRank != Rank;
1121 std::rotate(Pos, NonScriptI, End);
1125 Script->adjustSectionsAfterSorting();
1128 static void applySynthetic(const std::vector<SyntheticSection *> &Sections,
1129 std::function<void(SyntheticSection *)> Fn) {
1130 for (SyntheticSection *SS : Sections)
1131 if (SS && SS->getParent() && !SS->empty())
1135 // We need to add input synthetic sections early in createSyntheticSections()
1136 // to make them visible from linkescript side. But not all sections are always
1137 // required to be in output. For example we don't need dynamic section content
1138 // sometimes. This function filters out such unused sections from the output.
1139 static void removeUnusedSyntheticSections(std::vector<OutputSection *> &V) {
1140 // All input synthetic sections that can be empty are placed after
1141 // all regular ones. We iterate over them all and exit at first
1143 for (InputSectionBase *S : llvm::reverse(InputSections)) {
1144 SyntheticSection *SS = dyn_cast<SyntheticSection>(S);
1147 OutputSection *OS = SS->getParent();
1148 if (!SS->empty() || !OS)
1150 OS->Sections.erase(std::find(OS->Sections.begin(), OS->Sections.end(), SS));
1152 // If there are no other sections in the output section, remove it from the
1154 if (OS->Sections.empty())
1155 V.erase(std::find(V.begin(), V.end(), OS));
1159 // Create output section objects and add them to OutputSections.
1160 template <class ELFT> void Writer<ELFT>::finalizeSections() {
1161 Out::DebugInfo = findSection(".debug_info");
1162 Out::PreinitArray = findSection(".preinit_array");
1163 Out::InitArray = findSection(".init_array");
1164 Out::FiniArray = findSection(".fini_array");
1166 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop
1167 // symbols for sections, so that the runtime can get the start and end
1168 // addresses of each section by section name. Add such symbols.
1169 if (!Config->Relocatable) {
1170 addStartEndSymbols();
1171 for (OutputSection *Sec : OutputSections)
1172 addStartStopSymbols(Sec);
1175 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type.
1176 // It should be okay as no one seems to care about the type.
1177 // Even the author of gold doesn't remember why gold behaves that way.
1178 // https://sourceware.org/ml/binutils/2002-03/msg00360.html
1180 addRegular<ELFT>("_DYNAMIC", InX::Dynamic, 0);
1182 // Define __rel[a]_iplt_{start,end} symbols if needed.
1183 addRelIpltSymbols();
1185 // This responsible for splitting up .eh_frame section into
1186 // pieces. The relocation scan uses those pieces, so this has to be
1188 applySynthetic({In<ELFT>::EhFrame},
1189 [](SyntheticSection *SS) { SS->finalizeContents(); });
1191 // Scan relocations. This must be done after every symbol is declared so that
1192 // we can correctly decide if a dynamic relocation is needed.
1193 forEachRelSec(scanRelocations<ELFT>);
1195 if (InX::Plt && !InX::Plt->empty())
1196 InX::Plt->addSymbols();
1197 if (InX::Iplt && !InX::Iplt->empty())
1198 InX::Iplt->addSymbols();
1200 // Now that we have defined all possible global symbols including linker-
1201 // synthesized ones. Visit all symbols to give the finishing touches.
1202 for (Symbol *S : Symtab<ELFT>::X->getSymbols()) {
1203 SymbolBody *Body = S->body();
1205 if (!includeInSymtab(*Body))
1208 InX::SymTab->addSymbol(Body);
1210 if (InX::DynSymTab && S->includeInDynsym()) {
1211 InX::DynSymTab->addSymbol(Body);
1212 if (auto *SS = dyn_cast<SharedSymbol>(Body))
1213 if (cast<SharedFile<ELFT>>(SS->File)->isNeeded())
1214 In<ELFT>::VerNeed->addSymbol(SS);
1218 // Do not proceed if there was an undefined symbol.
1222 addPredefinedSections();
1223 removeUnusedSyntheticSections(OutputSections);
1225 clearOutputSections();
1228 // Now that we have the final list, create a list of all the
1229 // OutputSectionCommands for convenience.
1230 for (BaseCommand *Base : Script->Opt.Commands)
1231 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
1232 OutputSectionCommands.push_back(Cmd);
1234 // This is a bit of a hack. A value of 0 means undef, so we set it
1235 // to 1 t make __ehdr_start defined. The section number is not
1236 // particularly relevant.
1237 Out::ElfHeader->SectionIndex = 1;
1240 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1241 OutputSection *Sec = Cmd->Sec;
1242 Sec->SectionIndex = I++;
1243 Sec->ShName = InX::ShStrTab->addString(Sec->Name);
1246 // Binary and relocatable output does not have PHDRS.
1247 // The headers have to be created before finalize as that can influence the
1248 // image base and the dynamic section on mips includes the image base.
1249 if (!Config->Relocatable && !Config->OFormatBinary) {
1250 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs();
1251 addPtArmExid(Phdrs);
1252 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size();
1255 // Compute the size of .rela.dyn and .rela.plt early since we need
1256 // them to populate .dynamic.
1257 for (SyntheticSection *SS : {In<ELFT>::RelaDyn, In<ELFT>::RelaPlt})
1258 if (SS->getParent() && !SS->empty())
1259 SS->getParent()->assignOffsets();
1261 // Dynamic section must be the last one in this list and dynamic
1262 // symbol table section (DynSymTab) must be the first one.
1263 applySynthetic({InX::DynSymTab, InX::Bss, InX::BssRelRo,
1264 InX::GnuHashTab, In<ELFT>::HashTab, InX::SymTab,
1265 InX::ShStrTab, InX::StrTab, In<ELFT>::VerDef,
1266 InX::DynStrTab, InX::GdbIndex, InX::Got,
1267 InX::MipsGot, InX::IgotPlt, InX::GotPlt,
1268 In<ELFT>::RelaDyn, In<ELFT>::RelaIplt, In<ELFT>::RelaPlt,
1269 InX::Plt, InX::Iplt, In<ELFT>::EhFrameHdr,
1270 In<ELFT>::VerSym, In<ELFT>::VerNeed, InX::Dynamic},
1271 [](SyntheticSection *SS) { SS->finalizeContents(); });
1273 // Some architectures use small displacements for jump instructions.
1274 // It is linker's responsibility to create thunks containing long
1275 // jump instructions if jump targets are too far. Create thunks.
1276 if (Target->NeedsThunks) {
1277 // FIXME: only ARM Interworking and Mips LA25 Thunks are implemented,
1279 // do not require address information. To support range extension Thunks
1280 // we need to assign addresses so that we can tell if jump instructions
1281 // are out of range. This will need to turn into a loop that converges
1282 // when no more Thunks are added
1284 if (TC.createThunks(OutputSectionCommands)) {
1285 applySynthetic({InX::MipsGot},
1286 [](SyntheticSection *SS) { SS->updateAllocSize(); });
1287 if (TC.createThunks(OutputSectionCommands))
1288 fatal("All non-range thunks should be created in first call");
1292 // Fill other section headers. The dynamic table is finalized
1293 // at the end because some tags like RELSZ depend on result
1294 // of finalizing other sections.
1295 for (OutputSectionCommand *Cmd : OutputSectionCommands)
1296 Cmd->finalize<ELFT>();
1298 // createThunks may have added local symbols to the static symbol table
1299 applySynthetic({InX::SymTab, InX::ShStrTab, InX::StrTab},
1300 [](SyntheticSection *SS) { SS->postThunkContents(); });
1303 template <class ELFT> void Writer<ELFT>::addPredefinedSections() {
1304 // ARM ABI requires .ARM.exidx to be terminated by some piece of data.
1305 // We have the terminater synthetic section class. Add that at the end.
1306 auto *OS = dyn_cast_or_null<OutputSection>(findSection(".ARM.exidx"));
1307 if (!OS || OS->Sections.empty() || Config->Relocatable)
1310 auto *Sentinel = make<ARMExidxSentinelSection>();
1311 OS->addSection(Sentinel);
1312 // If there are linker script commands existing at this point then add the
1313 // sentinel to the last of these too.
1314 if (OutputSectionCommand *C = Script->getCmd(OS)) {
1315 auto ISD = std::find_if(C->Commands.rbegin(), C->Commands.rend(),
1316 [](const BaseCommand *Base) {
1317 return isa<InputSectionDescription>(Base);
1319 cast<InputSectionDescription>(*ISD)->Sections.push_back(Sentinel);
1323 // The linker is expected to define SECNAME_start and SECNAME_end
1324 // symbols for a few sections. This function defines them.
1325 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() {
1326 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) {
1327 // These symbols resolve to the image base if the section does not exist.
1328 // A special value -1 indicates end of the section.
1330 addOptionalRegular<ELFT>(Start, OS, 0);
1331 addOptionalRegular<ELFT>(End, OS, -1);
1334 OS = Out::ElfHeader;
1335 addOptionalRegular<ELFT>(Start, OS, 0);
1336 addOptionalRegular<ELFT>(End, OS, 0);
1340 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray);
1341 Define("__init_array_start", "__init_array_end", Out::InitArray);
1342 Define("__fini_array_start", "__fini_array_end", Out::FiniArray);
1344 if (OutputSection *Sec = findSection(".ARM.exidx"))
1345 Define("__exidx_start", "__exidx_end", Sec);
1348 // If a section name is valid as a C identifier (which is rare because of
1349 // the leading '.'), linkers are expected to define __start_<secname> and
1350 // __stop_<secname> symbols. They are at beginning and end of the section,
1351 // respectively. This is not requested by the ELF standard, but GNU ld and
1352 // gold provide the feature, and used by many programs.
1353 template <class ELFT>
1354 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) {
1355 StringRef S = Sec->Name;
1356 if (!isValidCIdentifier(S))
1358 addOptionalRegular<ELFT>(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT);
1359 addOptionalRegular<ELFT>(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT);
1362 template <class ELFT>
1363 OutputSectionCommand *Writer<ELFT>::findSectionCommand(StringRef Name) {
1364 for (OutputSectionCommand *Cmd : OutputSectionCommands)
1365 if (Cmd->Name == Name)
1370 template <class ELFT> OutputSection *Writer<ELFT>::findSectionInScript(StringRef Name) {
1371 if (OutputSectionCommand *Cmd = findSectionCommand(Name))
1376 template <class ELFT> OutputSection *Writer<ELFT>::findSection(StringRef Name) {
1377 for (OutputSection *Sec : OutputSections)
1378 if (Sec->Name == Name)
1383 static bool needsPtLoad(OutputSection *Sec) {
1384 if (!(Sec->Flags & SHF_ALLOC))
1387 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is
1388 // responsible for allocating space for them, not the PT_LOAD that
1389 // contains the TLS initialization image.
1390 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS)
1395 // Linker scripts are responsible for aligning addresses. Unfortunately, most
1396 // linker scripts are designed for creating two PT_LOADs only, one RX and one
1397 // RW. This means that there is no alignment in the RO to RX transition and we
1398 // cannot create a PT_LOAD there.
1399 static uint64_t computeFlags(uint64_t Flags) {
1401 return PF_R | PF_W | PF_X;
1402 if (Config->SingleRoRx && !(Flags & PF_W))
1403 return Flags | PF_X;
1407 // Decide which program headers to create and which sections to include in each
1409 template <class ELFT> std::vector<PhdrEntry> Writer<ELFT>::createPhdrs() {
1410 std::vector<PhdrEntry> Ret;
1411 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * {
1412 Ret.emplace_back(Type, Flags);
1416 // The first phdr entry is PT_PHDR which describes the program header itself.
1417 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders);
1419 // PT_INTERP must be the second entry if exists.
1420 if (OutputSection *Sec = findSectionInScript(".interp"))
1421 AddHdr(PT_INTERP, Sec->getPhdrFlags())->add(Sec);
1423 // Add the first PT_LOAD segment for regular output sections.
1424 uint64_t Flags = computeFlags(PF_R);
1425 PhdrEntry *Load = AddHdr(PT_LOAD, Flags);
1427 // Add the headers. We will remove them if they don't fit.
1428 Load->add(Out::ElfHeader);
1429 Load->add(Out::ProgramHeaders);
1431 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1432 OutputSection *Sec = Cmd->Sec;
1433 if (!(Sec->Flags & SHF_ALLOC))
1435 if (!needsPtLoad(Sec))
1438 // Segments are contiguous memory regions that has the same attributes
1439 // (e.g. executable or writable). There is one phdr for each segment.
1440 // Therefore, we need to create a new phdr when the next section has
1441 // different flags or is loaded at a discontiguous address using AT linker
1443 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags());
1444 if (Script->hasLMA(Sec) || Flags != NewFlags) {
1445 Load = AddHdr(PT_LOAD, NewFlags);
1452 // Add a TLS segment if any.
1453 PhdrEntry TlsHdr(PT_TLS, PF_R);
1454 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1455 OutputSection *Sec = Cmd->Sec;
1456 if (Sec->Flags & SHF_TLS)
1460 Ret.push_back(std::move(TlsHdr));
1462 // Add an entry for .dynamic.
1464 AddHdr(PT_DYNAMIC, InX::Dynamic->getParent()->getPhdrFlags())
1465 ->add(InX::Dynamic->getParent());
1467 // PT_GNU_RELRO includes all sections that should be marked as
1468 // read-only by dynamic linker after proccessing relocations.
1469 PhdrEntry RelRo(PT_GNU_RELRO, PF_R);
1470 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1471 OutputSection *Sec = Cmd->Sec;
1472 if (needsPtLoad(Sec) && isRelroSection(Sec))
1476 Ret.push_back(std::move(RelRo));
1478 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr.
1479 if (!In<ELFT>::EhFrame->empty() && In<ELFT>::EhFrameHdr &&
1480 In<ELFT>::EhFrame->getParent() && In<ELFT>::EhFrameHdr->getParent())
1481 AddHdr(PT_GNU_EH_FRAME, In<ELFT>::EhFrameHdr->getParent()->getPhdrFlags())
1482 ->add(In<ELFT>::EhFrameHdr->getParent());
1484 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes
1485 // the dynamic linker fill the segment with random data.
1486 if (OutputSection *Sec = findSectionInScript(".openbsd.randomdata"))
1487 AddHdr(PT_OPENBSD_RANDOMIZE, Sec->getPhdrFlags())->add(Sec);
1489 // PT_GNU_STACK is a special section to tell the loader to make the
1490 // pages for the stack non-executable. If you really want an executable
1491 // stack, you can pass -z execstack, but that's not recommended for
1492 // security reasons.
1494 if (Config->ZExecstack)
1495 Perm = PF_R | PF_W | PF_X;
1498 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize;
1500 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable
1501 // is expected to perform W^X violations, such as calling mprotect(2) or
1502 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on
1504 if (Config->ZWxneeded)
1505 AddHdr(PT_OPENBSD_WXNEEDED, PF_X);
1507 // Create one PT_NOTE per a group of contiguous .note sections.
1508 PhdrEntry *Note = nullptr;
1509 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1510 OutputSection *Sec = Cmd->Sec;
1511 if (Sec->Type == SHT_NOTE) {
1512 if (!Note || Script->hasLMA(Sec))
1513 Note = AddHdr(PT_NOTE, PF_R);
1522 template <class ELFT>
1523 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry> &Phdrs) {
1524 if (Config->EMachine != EM_ARM)
1527 std::find_if(OutputSectionCommands.begin(), OutputSectionCommands.end(),
1528 [](OutputSectionCommand *Cmd) {
1529 return Cmd->Sec->Type == SHT_ARM_EXIDX;
1531 if (I == OutputSectionCommands.end())
1534 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME
1535 PhdrEntry ARMExidx(PT_ARM_EXIDX, PF_R);
1536 ARMExidx.add((*I)->Sec);
1537 Phdrs.push_back(ARMExidx);
1540 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the
1541 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic
1542 // linker can set the permissions.
1543 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() {
1544 auto PageAlign = [](OutputSection *Sec) {
1545 OutputSectionCommand *Cmd = Script->getCmd(Sec);
1546 if (Cmd && !Cmd->AddrExpr)
1547 Cmd->AddrExpr = [=] {
1548 return alignTo(Script->getDot(), Config->MaxPageSize);
1552 for (const PhdrEntry &P : Phdrs)
1553 if (P.p_type == PT_LOAD && P.First)
1556 for (const PhdrEntry &P : Phdrs) {
1557 if (P.p_type != PT_GNU_RELRO)
1561 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we
1562 // have to align it to a page.
1563 auto End = OutputSectionCommands.end();
1565 std::find(OutputSectionCommands.begin(), End, Script->getCmd(P.Last));
1566 if (I == End || (I + 1) == End)
1568 OutputSection *Sec = (*(I + 1))->Sec;
1569 if (needsPtLoad(Sec))
1574 // Adjusts the file alignment for a given output section and returns
1575 // its new file offset. The file offset must be the same with its
1576 // virtual address (modulo the page size) so that the loader can load
1577 // executables without any address adjustment.
1578 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Sec) {
1579 OutputSection *First = Sec->FirstInPtLoad;
1580 // If the section is not in a PT_LOAD, we just have to align it.
1582 return alignTo(Off, Sec->Alignment);
1584 // The first section in a PT_LOAD has to have congruent offset and address
1585 // module the page size.
1587 return alignTo(Off, Config->MaxPageSize, Sec->Addr);
1589 // If two sections share the same PT_LOAD the file offset is calculated
1590 // using this formula: Off2 = Off1 + (VA2 - VA1).
1591 return First->Offset + Sec->Addr - First->Addr;
1594 static uint64_t setOffset(OutputSection *Sec, uint64_t Off) {
1595 if (Sec->Type == SHT_NOBITS) {
1600 Off = getFileAlignment(Off, Sec);
1602 return Off + Sec->Size;
1605 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() {
1607 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1608 OutputSection *Sec = Cmd->Sec;
1609 if (Sec->Flags & SHF_ALLOC)
1610 Off = setOffset(Sec, Off);
1612 FileSize = alignTo(Off, Config->Wordsize);
1615 // Assign file offsets to output sections.
1616 template <class ELFT> void Writer<ELFT>::assignFileOffsets() {
1618 Off = setOffset(Out::ElfHeader, Off);
1619 Off = setOffset(Out::ProgramHeaders, Off);
1621 for (OutputSectionCommand *Cmd : OutputSectionCommands)
1622 Off = setOffset(Cmd->Sec, Off);
1624 SectionHeaderOff = alignTo(Off, Config->Wordsize);
1626 SectionHeaderOff + (OutputSectionCommands.size() + 1) * sizeof(Elf_Shdr);
1629 // Finalize the program headers. We call this function after we assign
1630 // file offsets and VAs to all sections.
1631 template <class ELFT> void Writer<ELFT>::setPhdrs() {
1632 for (PhdrEntry &P : Phdrs) {
1633 OutputSection *First = P.First;
1634 OutputSection *Last = P.Last;
1636 P.p_filesz = Last->Offset - First->Offset;
1637 if (Last->Type != SHT_NOBITS)
1638 P.p_filesz += Last->Size;
1639 P.p_memsz = Last->Addr + Last->Size - First->Addr;
1640 P.p_offset = First->Offset;
1641 P.p_vaddr = First->Addr;
1643 P.p_paddr = First->getLMA();
1645 if (P.p_type == PT_LOAD)
1646 P.p_align = Config->MaxPageSize;
1647 else if (P.p_type == PT_GNU_RELRO)
1650 // The TLS pointer goes after PT_TLS. At least glibc will align it,
1651 // so round up the size to make sure the offsets are correct.
1652 if (P.p_type == PT_TLS) {
1655 P.p_memsz = alignTo(P.p_memsz, P.p_align);
1660 // The entry point address is chosen in the following ways.
1662 // 1. the '-e' entry command-line option;
1663 // 2. the ENTRY(symbol) command in a linker control script;
1664 // 3. the value of the symbol start, if present;
1665 // 4. the address of the first byte of the .text section, if present;
1666 // 5. the address 0.
1667 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() {
1668 // Case 1, 2 or 3. As a special case, if the symbol is actually
1669 // a number, we'll use that number as an address.
1670 if (SymbolBody *B = Symtab<ELFT>::X->find(Config->Entry))
1673 if (to_integer(Config->Entry, Addr))
1677 if (OutputSection *Sec = findSectionInScript(".text")) {
1678 if (Config->WarnMissingEntry)
1679 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" +
1680 utohexstr(Sec->Addr));
1685 if (Config->WarnMissingEntry)
1686 warn("cannot find entry symbol " + Config->Entry +
1687 "; not setting start address");
1691 static uint16_t getELFType() {
1694 if (Config->Relocatable)
1699 // This function is called after we have assigned address and size
1700 // to each section. This function fixes some predefined
1701 // symbol values that depend on section address and size.
1702 template <class ELFT> void Writer<ELFT>::fixPredefinedSymbols() {
1703 // _etext is the first location after the last read-only loadable segment.
1704 // _edata is the first location after the last read-write loadable segment.
1705 // _end is the first location after the uninitialized data region.
1706 PhdrEntry *Last = nullptr;
1707 PhdrEntry *LastRO = nullptr;
1708 PhdrEntry *LastRW = nullptr;
1709 for (PhdrEntry &P : Phdrs) {
1710 if (P.p_type != PT_LOAD)
1713 if (P.p_flags & PF_W)
1719 auto Set = [](DefinedRegular *S, OutputSection *Sec, uint64_t Value) {
1727 Set(ElfSym::End1, Last->First, Last->p_memsz);
1728 Set(ElfSym::End2, Last->First, Last->p_memsz);
1731 Set(ElfSym::Etext1, LastRO->First, LastRO->p_filesz);
1732 Set(ElfSym::Etext2, LastRO->First, LastRO->p_filesz);
1735 Set(ElfSym::Edata1, LastRW->First, LastRW->p_filesz);
1736 Set(ElfSym::Edata2, LastRW->First, LastRW->p_filesz);
1740 ElfSym::Bss->Section = findSectionInScript(".bss");
1742 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should
1743 // be equal to the _gp symbol's value.
1744 if (Config->EMachine == EM_MIPS && !ElfSym::MipsGp->Value) {
1745 // Find GP-relative section with the lowest address
1746 // and use this address to calculate default _gp value.
1747 for (const OutputSectionCommand *Cmd : OutputSectionCommands) {
1748 OutputSection *OS = Cmd->Sec;
1749 if (OS->Flags & SHF_MIPS_GPREL) {
1750 ElfSym::MipsGp->Value = OS->Addr + 0x7ff0;
1757 template <class ELFT> void Writer<ELFT>::writeHeader() {
1758 uint8_t *Buf = Buffer->getBufferStart();
1759 memcpy(Buf, "\177ELF", 4);
1761 // Write the ELF header.
1762 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf);
1763 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32;
1764 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB;
1765 EHdr->e_ident[EI_VERSION] = EV_CURRENT;
1766 EHdr->e_ident[EI_OSABI] = Config->OSABI;
1767 EHdr->e_type = getELFType();
1768 EHdr->e_machine = Config->EMachine;
1769 EHdr->e_version = EV_CURRENT;
1770 EHdr->e_entry = getEntryAddr();
1771 EHdr->e_shoff = SectionHeaderOff;
1772 EHdr->e_ehsize = sizeof(Elf_Ehdr);
1773 EHdr->e_phnum = Phdrs.size();
1774 EHdr->e_shentsize = sizeof(Elf_Shdr);
1775 EHdr->e_shnum = OutputSectionCommands.size() + 1;
1776 EHdr->e_shstrndx = InX::ShStrTab->getParent()->SectionIndex;
1778 if (Config->EMachine == EM_ARM)
1779 // We don't currently use any features incompatible with EF_ARM_EABI_VER5,
1780 // but we don't have any firm guarantees of conformance. Linux AArch64
1781 // kernels (as of 2016) require an EABI version to be set.
1782 EHdr->e_flags = EF_ARM_EABI_VER5;
1783 else if (Config->EMachine == EM_MIPS)
1784 EHdr->e_flags = getMipsEFlags<ELFT>();
1786 if (!Config->Relocatable) {
1787 EHdr->e_phoff = sizeof(Elf_Ehdr);
1788 EHdr->e_phentsize = sizeof(Elf_Phdr);
1791 // Write the program header table.
1792 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff);
1793 for (PhdrEntry &P : Phdrs) {
1794 HBuf->p_type = P.p_type;
1795 HBuf->p_flags = P.p_flags;
1796 HBuf->p_offset = P.p_offset;
1797 HBuf->p_vaddr = P.p_vaddr;
1798 HBuf->p_paddr = P.p_paddr;
1799 HBuf->p_filesz = P.p_filesz;
1800 HBuf->p_memsz = P.p_memsz;
1801 HBuf->p_align = P.p_align;
1805 // Write the section header table. Note that the first table entry is null.
1806 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff);
1807 for (OutputSectionCommand *Cmd : OutputSectionCommands)
1808 Cmd->Sec->writeHeaderTo<ELFT>(++SHdrs);
1811 // Open a result file.
1812 template <class ELFT> void Writer<ELFT>::openFile() {
1813 if (!Config->Is64 && FileSize > UINT32_MAX) {
1814 error("output file too large: " + Twine(FileSize) + " bytes");
1818 unlinkAsync(Config->OutputFile);
1819 ErrorOr<std::unique_ptr<FileOutputBuffer>> BufferOrErr =
1820 FileOutputBuffer::create(Config->OutputFile, FileSize,
1821 FileOutputBuffer::F_executable);
1823 if (auto EC = BufferOrErr.getError())
1824 error("failed to open " + Config->OutputFile + ": " + EC.message());
1826 Buffer = std::move(*BufferOrErr);
1829 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() {
1830 uint8_t *Buf = Buffer->getBufferStart();
1831 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1832 OutputSection *Sec = Cmd->Sec;
1833 if (Sec->Flags & SHF_ALLOC)
1834 Cmd->writeTo<ELFT>(Buf + Sec->Offset);
1838 // Write section contents to a mmap'ed file.
1839 template <class ELFT> void Writer<ELFT>::writeSections() {
1840 uint8_t *Buf = Buffer->getBufferStart();
1842 // PPC64 needs to process relocations in the .opd section
1843 // before processing relocations in code-containing sections.
1844 if (auto *OpdCmd = findSectionCommand(".opd")) {
1845 Out::Opd = OpdCmd->Sec;
1846 Out::OpdBuf = Buf + Out::Opd->Offset;
1847 OpdCmd->template writeTo<ELFT>(Buf + Out::Opd->Offset);
1850 OutputSection *EhFrameHdr =
1851 (In<ELFT>::EhFrameHdr && !In<ELFT>::EhFrameHdr->empty())
1852 ? In<ELFT>::EhFrameHdr->getParent()
1855 // In -r or -emit-relocs mode, write the relocation sections first as in
1856 // ELf_Rel targets we might find out that we need to modify the relocated
1857 // section while doing it.
1858 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1859 OutputSection *Sec = Cmd->Sec;
1860 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA)
1861 Cmd->writeTo<ELFT>(Buf + Sec->Offset);
1864 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
1865 OutputSection *Sec = Cmd->Sec;
1866 if (Sec != Out::Opd && Sec != EhFrameHdr && Sec->Type != SHT_REL &&
1867 Sec->Type != SHT_RELA)
1868 Cmd->writeTo<ELFT>(Buf + Sec->Offset);
1871 // The .eh_frame_hdr depends on .eh_frame section contents, therefore
1872 // it should be written after .eh_frame is written.
1874 OutputSectionCommand *Cmd = Script->getCmd(EhFrameHdr);
1875 Cmd->writeTo<ELFT>(Buf + EhFrameHdr->Offset);
1879 template <class ELFT> void Writer<ELFT>::writeBuildId() {
1880 if (!InX::BuildId || !InX::BuildId->getParent())
1883 // Compute a hash of all sections of the output file.
1884 uint8_t *Start = Buffer->getBufferStart();
1885 uint8_t *End = Start + FileSize;
1886 InX::BuildId->writeBuildId({Start, End});
1889 template void elf::writeResult<ELF32LE>();
1890 template void elf::writeResult<ELF32BE>();
1891 template void elf::writeResult<ELF64LE>();
1892 template void elf::writeResult<ELF64BE>();