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"
22 #include "llvm/ADT/StringMap.h"
23 #include "llvm/ADT/StringSwitch.h"
24 #include "llvm/Support/FileOutputBuffer.h"
25 #include "llvm/Support/raw_ostream.h"
29 using namespace llvm::ELF;
30 using namespace llvm::object;
31 using namespace llvm::support;
32 using namespace llvm::support::endian;
35 using namespace lld::elf;
38 // The writer writes a SymbolTable result to a file.
39 template <class ELFT> class Writer {
41 typedef typename ELFT::Shdr Elf_Shdr;
42 typedef typename ELFT::Ehdr Elf_Ehdr;
43 typedef typename ELFT::Phdr Elf_Phdr;
48 void createSyntheticSections();
49 void copyLocalSymbols();
50 void addSectionSymbols();
51 void addReservedSymbols();
52 void createSections();
53 void forEachRelSec(std::function<void(InputSectionBase &)> Fn);
55 void finalizeSections();
56 void addPredefinedSections();
58 std::vector<PhdrEntry> createPhdrs();
59 void removeEmptyPTLoad();
60 void addPtArmExid(std::vector<PhdrEntry> &Phdrs);
61 void assignAddresses();
62 void assignFileOffsets();
63 void assignFileOffsetsBinary();
66 void fixSectionAlignments();
67 void fixPredefinedSymbols();
71 void writeSectionsBinary();
74 std::unique_ptr<FileOutputBuffer> Buffer;
76 std::vector<OutputSection *> OutputSections;
77 OutputSectionFactory Factory{OutputSections};
79 void addRelIpltSymbols();
80 void addStartEndSymbols();
81 void addStartStopSymbols(OutputSection *Sec);
82 uint64_t getEntryAddr();
83 OutputSection *findSection(StringRef Name);
85 std::vector<PhdrEntry> Phdrs;
88 uint64_t SectionHeaderOff;
89 bool AllocateHeader = true;
91 } // anonymous namespace
93 StringRef elf::getOutputSectionName(StringRef Name) {
94 if (Config->Relocatable)
97 // If -emit-relocs is given (which is rare), we need to copy
98 // relocation sections to the output. If input section .foo is
99 // output as .bar, we want to rename .rel.foo .rel.bar as well.
100 if (Config->EmitRelocs) {
101 for (StringRef V : {".rel.", ".rela."}) {
102 if (Name.startswith(V)) {
103 StringRef Inner = getOutputSectionName(Name.substr(V.size() - 1));
104 return Saver.save(Twine(V.drop_back()) + Inner);
110 {".text.", ".rodata.", ".data.rel.ro.", ".data.", ".bss.rel.ro.",
111 ".bss.", ".init_array.", ".fini_array.", ".ctors.", ".dtors.", ".tbss.",
112 ".gcc_except_table.", ".tdata.", ".ARM.exidx."}) {
113 StringRef Prefix = V.drop_back();
114 if (Name.startswith(V) || Name == Prefix)
118 // CommonSection is identified as "COMMON" in linker scripts.
119 // By default, it should go to .bss section.
120 if (Name == "COMMON")
123 // ".zdebug_" is a prefix for ZLIB-compressed sections.
124 // Because we decompressed input sections, we want to remove 'z'.
125 if (Name.startswith(".zdebug_"))
126 return Saver.save(Twine(".") + Name.substr(2));
130 template <class ELFT> static bool needsInterpSection() {
131 return !Symtab<ELFT>::X->getSharedFiles().empty() &&
132 !Config->DynamicLinker.empty() && !Script->ignoreInterpSection();
135 template <class ELFT> void elf::writeResult() { Writer<ELFT>().run(); }
137 template <class ELFT> void Writer<ELFT>::removeEmptyPTLoad() {
138 auto I = std::remove_if(Phdrs.begin(), Phdrs.end(), [&](const PhdrEntry &P) {
139 if (P.p_type != PT_LOAD)
143 uint64_t Size = P.Last->Addr + P.Last->Size - P.First->Addr;
146 Phdrs.erase(I, Phdrs.end());
149 // This function scans over the input sections and creates mergeable
150 // synthetic sections. It removes MergeInputSections from array and
151 // adds new synthetic ones. Each synthetic section is added to the
152 // location of the first input section it replaces.
153 static void combineMergableSections() {
154 std::vector<MergeSyntheticSection *> MergeSections;
155 for (InputSectionBase *&S : InputSections) {
156 MergeInputSection *MS = dyn_cast<MergeInputSection>(S);
160 // We do not want to handle sections that are not alive, so just remove
161 // them instead of trying to merge.
165 StringRef OutsecName = getOutputSectionName(MS->Name);
166 uint64_t Flags = MS->Flags & ~(uint64_t)(SHF_GROUP | SHF_COMPRESSED);
167 uint32_t Alignment = std::max<uint32_t>(MS->Alignment, MS->Entsize);
170 llvm::find_if(MergeSections, [=](MergeSyntheticSection *Sec) {
171 return Sec->Name == OutsecName && Sec->Flags == Flags &&
172 Sec->Alignment == Alignment;
174 if (I == MergeSections.end()) {
175 MergeSyntheticSection *Syn =
176 make<MergeSyntheticSection>(OutsecName, MS->Type, Flags, Alignment);
177 MergeSections.push_back(Syn);
178 I = std::prev(MergeSections.end());
183 (*I)->addSection(MS);
186 std::vector<InputSectionBase *> &V = InputSections;
187 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end());
190 template <class ELFT> static void combineEhFrameSections() {
191 for (InputSectionBase *&S : InputSections) {
192 EhInputSection *ES = dyn_cast<EhInputSection>(S);
193 if (!ES || !ES->Live)
196 In<ELFT>::EhFrame->addSection(ES);
200 std::vector<InputSectionBase *> &V = InputSections;
201 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end());
204 // The main function of the writer.
205 template <class ELFT> void Writer<ELFT>::run() {
206 // Create linker-synthesized sections such as .got or .plt.
207 // Such sections are of type input section.
208 createSyntheticSections();
209 combineMergableSections();
211 if (!Config->Relocatable)
212 combineEhFrameSections<ELFT>();
214 // We need to create some reserved symbols such as _end. Create them.
215 if (!Config->Relocatable)
216 addReservedSymbols();
218 // Create output sections.
219 Script->OutputSections = &OutputSections;
220 if (Script->Opt.HasSections) {
221 // If linker script contains SECTIONS commands, let it create sections.
222 Script->processCommands(Factory);
224 // Linker scripts may have left some input sections unassigned.
225 // Assign such sections using the default rule.
226 Script->addOrphanSections(Factory);
228 // If linker script does not contain SECTIONS commands, create
229 // output sections by default rules. We still need to give the
230 // linker script a chance to run, because it might contain
231 // non-SECTIONS commands such as ASSERT.
233 Script->processCommands(Factory);
236 if (Config->Discard != DiscardPolicy::All)
239 if (Config->CopyRelocs)
242 // Now that we have a complete set of output sections. This function
243 // completes section contents. For example, we need to add strings
244 // to the string table, and add entries to .got and .plt.
245 // finalizeSections does that.
250 if (Config->Relocatable) {
253 if (Script->Opt.HasSections) {
254 Script->assignAddresses(Phdrs);
256 fixSectionAlignments();
258 Script->processNonSectionCommands();
261 // Remove empty PT_LOAD to avoid causing the dynamic linker to try to mmap a
262 // 0 sized region. This has to be done late since only after assignAddresses
263 // we know the size of the sections.
266 if (!Config->OFormatBinary)
269 assignFileOffsetsBinary();
272 fixPredefinedSymbols();
275 // It does not make sense try to open the file if we have error already.
278 // Write the result down to a file.
282 if (!Config->OFormatBinary) {
286 writeSectionsBinary();
289 // Backfill .note.gnu.build-id section content. This is done at last
290 // because the content is usually a hash value of the entire output file.
295 // Handle -Map option.
296 writeMapFile<ELFT>(OutputSections);
300 if (auto EC = Buffer->commit())
301 error("failed to write to the output file: " + EC.message());
303 // Flush the output streams and exit immediately. A full shutdown
304 // is a good test that we are keeping track of all allocated memory,
305 // but actually freeing it is a waste of time in a regular linker run.
306 if (Config->ExitEarly)
310 // Initialize Out members.
311 template <class ELFT> void Writer<ELFT>::createSyntheticSections() {
312 // Initialize all pointers with NULL. This is needed because
313 // you can call lld::elf::main more than once as a library.
314 memset(&Out::First, 0, sizeof(Out));
316 auto Add = [](InputSectionBase *Sec) { InputSections.push_back(Sec); };
318 In<ELFT>::DynStrTab = make<StringTableSection>(".dynstr", true);
319 In<ELFT>::Dynamic = make<DynamicSection<ELFT>>();
320 In<ELFT>::RelaDyn = make<RelocationSection<ELFT>>(
321 Config->IsRela ? ".rela.dyn" : ".rel.dyn", Config->ZCombreloc);
322 In<ELFT>::ShStrTab = make<StringTableSection>(".shstrtab", false);
324 Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC);
325 Out::ElfHeader->Size = sizeof(Elf_Ehdr);
326 Out::ProgramHeaders = make<OutputSection>("", 0, SHF_ALLOC);
327 Out::ProgramHeaders->updateAlignment(Config->Wordsize);
329 if (needsInterpSection<ELFT>()) {
330 In<ELFT>::Interp = createInterpSection();
331 Add(In<ELFT>::Interp);
333 In<ELFT>::Interp = nullptr;
336 if (!Config->Relocatable)
337 Add(createCommentSection<ELFT>());
339 if (Config->Strip != StripPolicy::All) {
340 In<ELFT>::StrTab = make<StringTableSection>(".strtab", false);
341 In<ELFT>::SymTab = make<SymbolTableSection<ELFT>>(*In<ELFT>::StrTab);
344 if (Config->BuildId != BuildIdKind::None) {
345 In<ELFT>::BuildId = make<BuildIdSection>();
346 Add(In<ELFT>::BuildId);
349 In<ELFT>::Common = createCommonSection<ELFT>();
350 if (In<ELFT>::Common)
353 In<ELFT>::Bss = make<BssSection>(".bss");
355 In<ELFT>::BssRelRo = make<BssSection>(".bss.rel.ro");
356 Add(In<ELFT>::BssRelRo);
358 // Add MIPS-specific sections.
359 bool HasDynSymTab = !Symtab<ELFT>::X->getSharedFiles().empty() ||
360 Config->Pic || Config->ExportDynamic;
361 if (Config->EMachine == EM_MIPS) {
362 if (!Config->Shared && HasDynSymTab) {
363 In<ELFT>::MipsRldMap = make<MipsRldMapSection>();
364 Add(In<ELFT>::MipsRldMap);
366 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create())
368 if (auto *Sec = MipsOptionsSection<ELFT>::create())
370 if (auto *Sec = MipsReginfoSection<ELFT>::create())
375 In<ELFT>::DynSymTab = make<SymbolTableSection<ELFT>>(*In<ELFT>::DynStrTab);
376 Add(In<ELFT>::DynSymTab);
378 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>();
379 Add(In<ELFT>::VerSym);
381 if (!Config->VersionDefinitions.empty()) {
382 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>();
383 Add(In<ELFT>::VerDef);
386 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>();
387 Add(In<ELFT>::VerNeed);
389 if (Config->GnuHash) {
390 In<ELFT>::GnuHashTab = make<GnuHashTableSection<ELFT>>();
391 Add(In<ELFT>::GnuHashTab);
394 if (Config->SysvHash) {
395 In<ELFT>::HashTab = make<HashTableSection<ELFT>>();
396 Add(In<ELFT>::HashTab);
399 Add(In<ELFT>::Dynamic);
400 Add(In<ELFT>::DynStrTab);
401 Add(In<ELFT>::RelaDyn);
404 // Add .got. MIPS' .got is so different from the other archs,
405 // it has its own class.
406 if (Config->EMachine == EM_MIPS) {
407 In<ELFT>::MipsGot = make<MipsGotSection>();
408 Add(In<ELFT>::MipsGot);
410 In<ELFT>::Got = make<GotSection<ELFT>>();
414 In<ELFT>::GotPlt = make<GotPltSection>();
415 Add(In<ELFT>::GotPlt);
416 In<ELFT>::IgotPlt = make<IgotPltSection>();
417 Add(In<ELFT>::IgotPlt);
419 if (Config->GdbIndex) {
420 In<ELFT>::GdbIndex = make<GdbIndexSection>();
421 Add(In<ELFT>::GdbIndex);
424 // We always need to add rel[a].plt to output if it has entries.
425 // Even for static linking it can contain R_[*]_IRELATIVE relocations.
426 In<ELFT>::RelaPlt = make<RelocationSection<ELFT>>(
427 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/);
428 Add(In<ELFT>::RelaPlt);
430 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure
431 // that the IRelative relocations are processed last by the dynamic loader
432 In<ELFT>::RelaIplt = make<RelocationSection<ELFT>>(
433 (Config->EMachine == EM_ARM) ? ".rel.dyn" : In<ELFT>::RelaPlt->Name,
435 Add(In<ELFT>::RelaIplt);
437 In<ELFT>::Plt = make<PltSection>(Target->PltHeaderSize);
439 In<ELFT>::Iplt = make<PltSection>(0);
442 if (!Config->Relocatable) {
443 if (Config->EhFrameHdr) {
444 In<ELFT>::EhFrameHdr = make<EhFrameHeader<ELFT>>();
445 Add(In<ELFT>::EhFrameHdr);
447 In<ELFT>::EhFrame = make<EhFrameSection<ELFT>>();
448 Add(In<ELFT>::EhFrame);
451 if (In<ELFT>::SymTab)
452 Add(In<ELFT>::SymTab);
453 Add(In<ELFT>::ShStrTab);
454 if (In<ELFT>::StrTab)
455 Add(In<ELFT>::StrTab);
458 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName,
459 const SymbolBody &B) {
460 if (B.isFile() || B.isSection())
463 // If sym references a section in a discarded group, don't keep it.
464 if (Sec == &InputSection::Discarded)
467 if (Config->Discard == DiscardPolicy::None)
470 // In ELF assembly .L symbols are normally discarded by the assembler.
471 // If the assembler fails to do so, the linker discards them if
472 // * --discard-locals is used.
473 // * The symbol is in a SHF_MERGE section, which is normally the reason for
474 // the assembler keeping the .L symbol.
475 if (!SymName.startswith(".L") && !SymName.empty())
478 if (Config->Discard == DiscardPolicy::Locals)
481 return !Sec || !(Sec->Flags & SHF_MERGE);
484 static bool includeInSymtab(const SymbolBody &B) {
485 if (!B.isLocal() && !B.symbol()->IsUsedInRegularObj)
488 if (auto *D = dyn_cast<DefinedRegular>(&B)) {
489 // Always include absolute symbols.
490 SectionBase *Sec = D->Section;
493 if (auto *IS = dyn_cast<InputSectionBase>(Sec)) {
495 IS = cast<InputSectionBase>(Sec);
496 // Exclude symbols pointing to garbage-collected sections.
500 if (auto *S = dyn_cast<MergeInputSection>(Sec))
501 if (!S->getSectionPiece(D->Value)->Live)
507 // Local symbols are not in the linker's symbol table. This function scans
508 // each object file's symbol table to copy local symbols to the output.
509 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() {
510 if (!In<ELFT>::SymTab)
512 for (elf::ObjectFile<ELFT> *F : Symtab<ELFT>::X->getObjectFiles()) {
513 for (SymbolBody *B : F->getLocalSymbols()) {
516 ": broken object: getLocalSymbols returns a non-local symbol");
517 auto *DR = dyn_cast<DefinedRegular>(B);
519 // No reason to keep local undefined symbol in symtab.
522 if (!includeInSymtab(*B))
525 SectionBase *Sec = DR->Section;
526 if (!shouldKeepInSymtab(Sec, B->getName(), *B))
528 In<ELFT>::SymTab->addSymbol(B);
533 template <class ELFT> void Writer<ELFT>::addSectionSymbols() {
534 // Create one STT_SECTION symbol for each output section we might
535 // have a relocation with.
536 for (OutputSection *Sec : OutputSections) {
537 if (Sec->Sections.empty())
540 InputSection *IS = Sec->Sections[0];
541 if (isa<SyntheticSection>(IS) || IS->Type == SHT_REL ||
542 IS->Type == SHT_RELA)
546 make<DefinedRegular>("", /*IsLocal=*/true, /*StOther=*/0, STT_SECTION,
547 /*Value=*/0, /*Size=*/0, IS, nullptr);
548 In<ELFT>::SymTab->addSymbol(Sym);
552 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections that
553 // we would like to make sure appear is a specific order to maximize their
554 // coverage by a single signed 16-bit offset from the TOC base pointer.
555 // Conversely, the special .tocbss section should be first among all SHT_NOBITS
556 // sections. This will put it next to the loaded special PPC64 sections (and,
557 // thus, within reach of the TOC base pointer).
558 static int getPPC64SectionRank(StringRef SectionName) {
559 return StringSwitch<int>(SectionName)
561 .Case(".branch_lt", 2)
568 // All sections with SHF_MIPS_GPREL flag should be grouped together
569 // because data in these sections is addressable with a gp relative address.
570 static int getMipsSectionRank(const OutputSection *S) {
571 if ((S->Flags & SHF_MIPS_GPREL) == 0)
573 if (S->Name == ".got")
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 template <class ELFT> 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 (In<ELFT>::Got && Sec == In<ELFT>::Got->OutSec)
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 == In<ELFT>::GotPlt->OutSec)
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 == In<ELFT>::Dynamic->OutSec)
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 == In<ELFT>::BssRelRo->OutSec)
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 template <class ELFT>
652 static bool compareSectionsNonScript(const OutputSection *A,
653 const OutputSection *B) {
654 // Put .interp first because some loaders want to see that section
655 // on the first page of the executable file when loaded into memory.
656 bool AIsInterp = A->Name == ".interp";
657 bool BIsInterp = B->Name == ".interp";
658 if (AIsInterp != BIsInterp)
661 // Allocatable sections go first to reduce the total PT_LOAD size and
662 // so debug info doesn't change addresses in actual code.
663 bool AIsAlloc = A->Flags & SHF_ALLOC;
664 bool BIsAlloc = B->Flags & SHF_ALLOC;
665 if (AIsAlloc != BIsAlloc)
668 // We don't have any special requirements for the relative order of two non
669 // allocatable sections.
673 // We want to put section specified by -T option first, so we
674 // can start assigning VA starting from them later.
675 auto AAddrSetI = Config->SectionStartMap.find(A->Name);
676 auto BAddrSetI = Config->SectionStartMap.find(B->Name);
677 bool AHasAddrSet = AAddrSetI != Config->SectionStartMap.end();
678 bool BHasAddrSet = BAddrSetI != Config->SectionStartMap.end();
679 if (AHasAddrSet != BHasAddrSet)
682 return AAddrSetI->second < BAddrSetI->second;
684 // We want the read only sections first so that they go in the PT_LOAD
685 // covering the program headers at the start of the file.
686 bool AIsWritable = A->Flags & SHF_WRITE;
687 bool BIsWritable = B->Flags & SHF_WRITE;
688 if (AIsWritable != BIsWritable)
691 if (!Config->SingleRoRx) {
692 // For a corresponding reason, put non exec sections first (the program
693 // header PT_LOAD is not executable).
694 // We only do that if we are not using linker scripts, since with linker
695 // scripts ro and rx sections are in the same PT_LOAD, so their relative
696 // order is not important. The same applies for -no-rosegment.
697 bool AIsExec = A->Flags & SHF_EXECINSTR;
698 bool BIsExec = B->Flags & SHF_EXECINSTR;
699 if (AIsExec != BIsExec)
703 // If we got here we know that both A and B are in the same PT_LOAD.
705 bool AIsTls = A->Flags & SHF_TLS;
706 bool BIsTls = B->Flags & SHF_TLS;
707 bool AIsNoBits = A->Type == SHT_NOBITS;
708 bool BIsNoBits = B->Type == SHT_NOBITS;
710 // The first requirement we have is to put (non-TLS) nobits sections last. The
711 // reason is that the only thing the dynamic linker will see about them is a
712 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the
713 // PT_LOAD, so that has to correspond to the nobits sections.
714 bool AIsNonTlsNoBits = AIsNoBits && !AIsTls;
715 bool BIsNonTlsNoBits = BIsNoBits && !BIsTls;
716 if (AIsNonTlsNoBits != BIsNonTlsNoBits)
717 return BIsNonTlsNoBits;
719 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo
720 // sections after r/w ones, so that the RelRo sections are contiguous.
721 bool AIsRelRo = isRelroSection<ELFT>(A);
722 bool BIsRelRo = isRelroSection<ELFT>(B);
723 if (AIsRelRo != BIsRelRo)
724 return AIsNonTlsNoBits ? AIsRelRo : BIsRelRo;
726 // The TLS initialization block needs to be a single contiguous block in a R/W
727 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W
728 // sections. The TLS NOBITS sections are placed here as they don't take up
729 // virtual address space in the PT_LOAD.
730 if (AIsTls != BIsTls)
733 // Within the TLS initialization block, the non-nobits sections need to appear
735 if (AIsNoBits != BIsNoBits)
738 // Some architectures have additional ordering restrictions for sections
739 // within the same PT_LOAD.
740 if (Config->EMachine == EM_PPC64)
741 return getPPC64SectionRank(A->Name) < getPPC64SectionRank(B->Name);
742 if (Config->EMachine == EM_MIPS)
743 return getMipsSectionRank(A) < getMipsSectionRank(B);
748 // Output section ordering is determined by this function.
749 template <class ELFT>
750 static bool compareSections(const OutputSection *A, const OutputSection *B) {
751 // For now, put sections mentioned in a linker script first.
752 int AIndex = Script->getSectionIndex(A->Name);
753 int BIndex = Script->getSectionIndex(B->Name);
754 bool AInScript = AIndex != INT_MAX;
755 bool BInScript = BIndex != INT_MAX;
756 if (AInScript != BInScript)
758 // If both are in the script, use that order.
760 return AIndex < BIndex;
762 return compareSectionsNonScript<ELFT>(A, B);
765 // Program header entry
766 PhdrEntry::PhdrEntry(unsigned Type, unsigned Flags) {
771 void PhdrEntry::add(OutputSection *Sec) {
775 p_align = std::max(p_align, Sec->Alignment);
776 if (p_type == PT_LOAD)
777 Sec->FirstInPtLoad = First;
780 template <class ELFT>
781 static Symbol *addRegular(StringRef Name, SectionBase *Sec, uint64_t Value,
782 uint8_t StOther = STV_HIDDEN,
783 uint8_t Binding = STB_WEAK) {
784 // The linker generated symbols are added as STB_WEAK to allow user defined
785 // ones to override them.
786 return Symtab<ELFT>::X->addRegular(Name, StOther, STT_NOTYPE, Value,
787 /*Size=*/0, Binding, Sec,
791 template <class ELFT>
792 static DefinedRegular *
793 addOptionalRegular(StringRef Name, SectionBase *Sec, uint64_t Val,
794 uint8_t StOther = STV_HIDDEN, uint8_t Binding = STB_GLOBAL) {
795 SymbolBody *S = Symtab<ELFT>::X->find(Name);
798 if (S->isInCurrentDSO())
800 return cast<DefinedRegular>(
801 addRegular<ELFT>(Name, Sec, Val, StOther, Binding)->body());
804 // The beginning and the ending of .rel[a].plt section are marked
805 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked
806 // executable. The runtime needs these symbols in order to resolve
807 // all IRELATIVE relocs on startup. For dynamic executables, we don't
808 // need these symbols, since IRELATIVE relocs are resolved through GOT
809 // and PLT. For details, see http://www.airs.com/blog/archives/403.
810 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() {
811 if (In<ELFT>::DynSymTab)
813 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start";
814 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, 0, STV_HIDDEN, STB_WEAK);
816 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end";
817 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, -1, STV_HIDDEN, STB_WEAK);
820 // The linker is expected to define some symbols depending on
821 // the linking result. This function defines such symbols.
822 template <class ELFT> void Writer<ELFT>::addReservedSymbols() {
823 if (Config->EMachine == EM_MIPS) {
824 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer
825 // so that it points to an absolute address which by default is relative
826 // to GOT. Default offset is 0x7ff0.
827 // See "Global Data Symbols" in Chapter 6 in the following document:
828 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf
829 ElfSym::MipsGp = Symtab<ELFT>::X->addAbsolute("_gp", STV_HIDDEN, STB_LOCAL);
831 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between
832 // start of function and 'gp' pointer into GOT.
833 if (Symtab<ELFT>::X->find("_gp_disp"))
835 Symtab<ELFT>::X->addAbsolute("_gp_disp", STV_HIDDEN, STB_LOCAL);
837 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp'
838 // pointer. This symbol is used in the code generated by .cpload pseudo-op
839 // in case of using -mno-shared option.
840 // https://sourceware.org/ml/binutils/2004-12/msg00094.html
841 if (Symtab<ELFT>::X->find("__gnu_local_gp"))
842 ElfSym::MipsLocalGp =
843 Symtab<ELFT>::X->addAbsolute("__gnu_local_gp", STV_HIDDEN, STB_LOCAL);
846 // In the assembly for 32 bit x86 the _GLOBAL_OFFSET_TABLE_ symbol
847 // is magical and is used to produce a R_386_GOTPC relocation.
848 // The R_386_GOTPC relocation value doesn't actually depend on the
849 // symbol value, so it could use an index of STN_UNDEF which, according
850 // to the spec, means the symbol value is 0.
851 // Unfortunately both gas and MC keep the _GLOBAL_OFFSET_TABLE_ symbol in
853 // The situation is even stranger on x86_64 where the assembly doesn't
854 // need the magical symbol, but gas still puts _GLOBAL_OFFSET_TABLE_ as
855 // an undefined symbol in the .o files.
856 // Given that the symbol is effectively unused, we just create a dummy
857 // hidden one to avoid the undefined symbol error.
858 Symtab<ELFT>::X->addIgnored("_GLOBAL_OFFSET_TABLE_");
860 // __tls_get_addr is defined by the dynamic linker for dynamic ELFs. For
861 // static linking the linker is required to optimize away any references to
862 // __tls_get_addr, so it's not defined anywhere. Create a hidden definition
863 // to avoid the undefined symbol error. As usual special cases are ARM and
864 // MIPS - the libc for these targets defines __tls_get_addr itself because
865 // there are no TLS optimizations for these targets.
866 if (!In<ELFT>::DynSymTab &&
867 (Config->EMachine != EM_MIPS && Config->EMachine != EM_ARM))
868 Symtab<ELFT>::X->addIgnored("__tls_get_addr");
870 // If linker script do layout we do not need to create any standart symbols.
871 if (Script->Opt.HasSections)
874 // __ehdr_start is the location of ELF file headers.
875 addOptionalRegular<ELFT>("__ehdr_start", Out::ElfHeader, 0, STV_HIDDEN);
877 auto Add = [](StringRef S) {
878 return addOptionalRegular<ELFT>(S, Out::ElfHeader, 0, STV_DEFAULT);
881 ElfSym::Bss = Add("__bss_start");
882 ElfSym::End1 = Add("end");
883 ElfSym::End2 = Add("_end");
884 ElfSym::Etext1 = Add("etext");
885 ElfSym::Etext2 = Add("_etext");
886 ElfSym::Edata1 = Add("edata");
887 ElfSym::Edata2 = Add("_edata");
890 // Sort input sections by section name suffixes for
891 // __attribute__((init_priority(N))).
892 static void sortInitFini(OutputSection *S) {
894 reinterpret_cast<OutputSection *>(S)->sortInitFini();
897 // Sort input sections by the special rule for .ctors and .dtors.
898 static void sortCtorsDtors(OutputSection *S) {
900 reinterpret_cast<OutputSection *>(S)->sortCtorsDtors();
903 // Sort input sections using the list provided by --symbol-ordering-file.
904 template <class ELFT>
905 static void sortBySymbolsOrder(ArrayRef<OutputSection *> OutputSections) {
906 if (Config->SymbolOrderingFile.empty())
909 // Build a map from symbols to their priorities. Symbols that didn't
910 // appear in the symbol ordering file have the lowest priority 0.
911 // All explicitly mentioned symbols have negative (higher) priorities.
912 DenseMap<StringRef, int> SymbolOrder;
913 int Priority = -Config->SymbolOrderingFile.size();
914 for (StringRef S : Config->SymbolOrderingFile)
915 SymbolOrder.insert({S, Priority++});
917 // Build a map from sections to their priorities.
918 DenseMap<SectionBase *, int> SectionOrder;
919 for (elf::ObjectFile<ELFT> *File : Symtab<ELFT>::X->getObjectFiles()) {
920 for (SymbolBody *Body : File->getSymbols()) {
921 auto *D = dyn_cast<DefinedRegular>(Body);
922 if (!D || !D->Section)
924 int &Priority = SectionOrder[D->Section];
925 Priority = std::min(Priority, SymbolOrder.lookup(D->getName()));
929 // Sort sections by priority.
930 for (OutputSection *Base : OutputSections)
931 if (auto *Sec = dyn_cast<OutputSection>(Base))
932 Sec->sort([&](InputSectionBase *S) { return SectionOrder.lookup(S); });
935 template <class ELFT>
936 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) {
937 for (InputSectionBase *IS : InputSections) {
940 // Scan all relocations. Each relocation goes through a series
941 // of tests to determine if it needs special treatment, such as
942 // creating GOT, PLT, copy relocations, etc.
943 // Note that relocations for non-alloc sections are directly
944 // processed by InputSection::relocateNonAlloc.
945 if (!(IS->Flags & SHF_ALLOC))
947 if (isa<InputSection>(IS) || isa<EhInputSection>(IS))
951 if (!Config->Relocatable) {
952 for (EhInputSection *ES : In<ELFT>::EhFrame->Sections)
957 template <class ELFT> void Writer<ELFT>::createSections() {
958 for (InputSectionBase *IS : InputSections)
960 Factory.addInputSec(IS, getOutputSectionName(IS->Name));
962 sortBySymbolsOrder<ELFT>(OutputSections);
963 sortInitFini(findSection(".init_array"));
964 sortInitFini(findSection(".fini_array"));
965 sortCtorsDtors(findSection(".ctors"));
966 sortCtorsDtors(findSection(".dtors"));
968 for (OutputSection *Sec : OutputSections)
969 Sec->assignOffsets();
972 static bool canSharePtLoad(const OutputSection &S1, const OutputSection &S2) {
973 if (!(S1.Flags & SHF_ALLOC) || !(S2.Flags & SHF_ALLOC))
976 bool S1IsWrite = S1.Flags & SHF_WRITE;
977 bool S2IsWrite = S2.Flags & SHF_WRITE;
978 if (S1IsWrite != S2IsWrite)
982 return true; // RO and RX share a PT_LOAD with linker scripts.
983 return (S1.Flags & SHF_EXECINSTR) == (S2.Flags & SHF_EXECINSTR);
986 template <class ELFT> void Writer<ELFT>::sortSections() {
987 // Don't sort if using -r. It is not necessary and we want to preserve the
988 // relative order for SHF_LINK_ORDER sections.
989 if (Config->Relocatable)
991 if (!Script->Opt.HasSections) {
992 std::stable_sort(OutputSections.begin(), OutputSections.end(),
993 compareSectionsNonScript<ELFT>);
996 Script->adjustSectionsBeforeSorting();
998 // The order of the sections in the script is arbitrary and may not agree with
999 // compareSectionsNonScript. This means that we cannot easily define a
1000 // strict weak ordering. To see why, consider a comparison of a section in the
1001 // script and one not in the script. We have a two simple options:
1002 // * Make them equivalent (a is not less than b, and b is not less than a).
1003 // The problem is then that equivalence has to be transitive and we can
1004 // have sections a, b and c with only b in a script and a less than c
1005 // which breaks this property.
1006 // * Use compareSectionsNonScript. Given that the script order doesn't have
1007 // to match, we can end up with sections a, b, c, d where b and c are in the
1008 // script and c is compareSectionsNonScript less than b. In which case d
1009 // can be equivalent to c, a to b and d < a. As a concrete example:
1010 // .a (rx) # not in script
1011 // .b (rx) # in script
1012 // .c (ro) # in script
1013 // .d (ro) # not in script
1015 // The way we define an order then is:
1016 // * First put script sections at the start and sort the script and
1017 // non-script sections independently.
1018 // * Move each non-script section to its preferred position. We try
1019 // to put each section in the last position where it it can share
1022 std::stable_sort(OutputSections.begin(), OutputSections.end(),
1023 compareSections<ELFT>);
1025 auto I = OutputSections.begin();
1026 auto E = OutputSections.end();
1028 std::find_if(OutputSections.begin(), E, [](OutputSection *S) {
1029 return Script->getSectionIndex(S->Name) == INT_MAX;
1031 while (NonScriptI != E) {
1032 auto BestPos = std::max_element(
1033 I, NonScriptI, [&](OutputSection *&A, OutputSection *&B) {
1034 bool ACanSharePtLoad = canSharePtLoad(**NonScriptI, *A);
1035 bool BCanSharePtLoad = canSharePtLoad(**NonScriptI, *B);
1036 if (ACanSharePtLoad != BCanSharePtLoad)
1037 return BCanSharePtLoad;
1039 bool ACmp = compareSectionsNonScript<ELFT>(*NonScriptI, A);
1040 bool BCmp = compareSectionsNonScript<ELFT>(*NonScriptI, B);
1042 return BCmp; // FIXME: missing test
1044 size_t PosA = &A - &OutputSections[0];
1045 size_t PosB = &B - &OutputSections[0];
1046 return ACmp ? PosA > PosB : PosA < PosB;
1049 // max_element only returns NonScriptI if the range is empty. If the range
1050 // is not empty we should consider moving the the element forward one
1052 if (BestPos != NonScriptI &&
1053 !compareSectionsNonScript<ELFT>(*NonScriptI, *BestPos))
1055 std::rotate(BestPos, NonScriptI, NonScriptI + 1);
1059 Script->adjustSectionsAfterSorting();
1062 static void applySynthetic(const std::vector<SyntheticSection *> &Sections,
1063 std::function<void(SyntheticSection *)> Fn) {
1064 for (SyntheticSection *SS : Sections)
1065 if (SS && SS->OutSec && !SS->empty()) {
1067 SS->OutSec->assignOffsets();
1071 // We need to add input synthetic sections early in createSyntheticSections()
1072 // to make them visible from linkescript side. But not all sections are always
1073 // required to be in output. For example we don't need dynamic section content
1074 // sometimes. This function filters out such unused sections from the output.
1075 static void removeUnusedSyntheticSections(std::vector<OutputSection *> &V) {
1076 // All input synthetic sections that can be empty are placed after
1077 // all regular ones. We iterate over them all and exit at first
1079 for (InputSectionBase *S : llvm::reverse(InputSections)) {
1080 SyntheticSection *SS = dyn_cast<SyntheticSection>(S);
1083 if (!SS->empty() || !SS->OutSec)
1086 SS->OutSec->Sections.erase(std::find(SS->OutSec->Sections.begin(),
1087 SS->OutSec->Sections.end(), SS));
1088 // If there are no other sections in the output section, remove it from the
1090 if (SS->OutSec->Sections.empty())
1091 V.erase(std::find(V.begin(), V.end(), SS->OutSec));
1095 // Create output section objects and add them to OutputSections.
1096 template <class ELFT> void Writer<ELFT>::finalizeSections() {
1097 Out::DebugInfo = findSection(".debug_info");
1098 Out::PreinitArray = findSection(".preinit_array");
1099 Out::InitArray = findSection(".init_array");
1100 Out::FiniArray = findSection(".fini_array");
1102 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop
1103 // symbols for sections, so that the runtime can get the start and end
1104 // addresses of each section by section name. Add such symbols.
1105 if (!Config->Relocatable) {
1106 addStartEndSymbols();
1107 for (OutputSection *Sec : OutputSections)
1108 addStartStopSymbols(Sec);
1111 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type.
1112 // It should be okay as no one seems to care about the type.
1113 // Even the author of gold doesn't remember why gold behaves that way.
1114 // https://sourceware.org/ml/binutils/2002-03/msg00360.html
1115 if (In<ELFT>::DynSymTab)
1116 addRegular<ELFT>("_DYNAMIC", In<ELFT>::Dynamic, 0);
1118 // Define __rel[a]_iplt_{start,end} symbols if needed.
1119 addRelIpltSymbols();
1121 // This responsible for splitting up .eh_frame section into
1122 // pieces. The relocation scan uses those pieces, so this has to be
1124 applySynthetic({In<ELFT>::EhFrame},
1125 [](SyntheticSection *SS) { SS->finalizeContents(); });
1127 // Scan relocations. This must be done after every symbol is declared so that
1128 // we can correctly decide if a dynamic relocation is needed.
1129 forEachRelSec(scanRelocations<ELFT>);
1131 if (In<ELFT>::Plt && !In<ELFT>::Plt->empty())
1132 In<ELFT>::Plt->addSymbols();
1133 if (In<ELFT>::Iplt && !In<ELFT>::Iplt->empty())
1134 In<ELFT>::Iplt->addSymbols();
1136 // Now that we have defined all possible global symbols including linker-
1137 // synthesized ones. Visit all symbols to give the finishing touches.
1138 for (Symbol *S : Symtab<ELFT>::X->getSymbols()) {
1139 SymbolBody *Body = S->body();
1141 if (!includeInSymtab(*Body))
1143 if (In<ELFT>::SymTab)
1144 In<ELFT>::SymTab->addSymbol(Body);
1146 if (In<ELFT>::DynSymTab && S->includeInDynsym()) {
1147 In<ELFT>::DynSymTab->addSymbol(Body);
1148 if (auto *SS = dyn_cast<SharedSymbol>(Body))
1149 if (cast<SharedFile<ELFT>>(SS->File)->isNeeded())
1150 In<ELFT>::VerNeed->addSymbol(SS);
1154 // Do not proceed if there was an undefined symbol.
1158 // So far we have added sections from input object files.
1159 // This function adds linker-created Out::* sections.
1160 addPredefinedSections();
1161 removeUnusedSyntheticSections(OutputSections);
1165 // This is a bit of a hack. A value of 0 means undef, so we set it
1166 // to 1 t make __ehdr_start defined. The section number is not
1167 // particularly relevant.
1168 Out::ElfHeader->SectionIndex = 1;
1171 for (OutputSection *Sec : OutputSections) {
1172 Sec->SectionIndex = I++;
1173 Sec->ShName = In<ELFT>::ShStrTab->addString(Sec->Name);
1176 // Binary and relocatable output does not have PHDRS.
1177 // The headers have to be created before finalize as that can influence the
1178 // image base and the dynamic section on mips includes the image base.
1179 if (!Config->Relocatable && !Config->OFormatBinary) {
1180 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs();
1181 addPtArmExid(Phdrs);
1185 // Dynamic section must be the last one in this list and dynamic
1186 // symbol table section (DynSymTab) must be the first one.
1187 applySynthetic({In<ELFT>::DynSymTab, In<ELFT>::Bss, In<ELFT>::BssRelRo,
1188 In<ELFT>::GnuHashTab, In<ELFT>::HashTab, In<ELFT>::SymTab,
1189 In<ELFT>::ShStrTab, In<ELFT>::StrTab, In<ELFT>::VerDef,
1190 In<ELFT>::DynStrTab, In<ELFT>::GdbIndex, In<ELFT>::Got,
1191 In<ELFT>::MipsGot, In<ELFT>::IgotPlt, In<ELFT>::GotPlt,
1192 In<ELFT>::RelaDyn, In<ELFT>::RelaIplt, In<ELFT>::RelaPlt,
1193 In<ELFT>::Plt, In<ELFT>::Iplt, In<ELFT>::Plt,
1194 In<ELFT>::EhFrameHdr, In<ELFT>::VerSym, In<ELFT>::VerNeed,
1196 [](SyntheticSection *SS) { SS->finalizeContents(); });
1198 // Some architectures use small displacements for jump instructions.
1199 // It is linker's responsibility to create thunks containing long
1200 // jump instructions if jump targets are too far. Create thunks.
1201 if (Target->NeedsThunks) {
1202 // FIXME: only ARM Interworking and Mips LA25 Thunks are implemented,
1204 // do not require address information. To support range extension Thunks
1205 // we need to assign addresses so that we can tell if jump instructions
1206 // are out of range. This will need to turn into a loop that converges
1207 // when no more Thunks are added
1208 ThunkCreator<ELFT> TC;
1209 if (TC.createThunks(OutputSections))
1210 applySynthetic({In<ELFT>::MipsGot},
1211 [](SyntheticSection *SS) { SS->updateAllocSize(); });
1213 // Fill other section headers. The dynamic table is finalized
1214 // at the end because some tags like RELSZ depend on result
1215 // of finalizing other sections.
1216 for (OutputSection *Sec : OutputSections)
1217 Sec->finalize<ELFT>();
1219 // createThunks may have added local symbols to the static symbol table
1220 applySynthetic({In<ELFT>::SymTab, In<ELFT>::ShStrTab, In<ELFT>::StrTab},
1221 [](SyntheticSection *SS) { SS->postThunkContents(); });
1224 template <class ELFT> void Writer<ELFT>::addPredefinedSections() {
1225 // ARM ABI requires .ARM.exidx to be terminated by some piece of data.
1226 // We have the terminater synthetic section class. Add that at the end.
1227 auto *OS = dyn_cast_or_null<OutputSection>(findSection(".ARM.exidx"));
1228 if (OS && !OS->Sections.empty() && !Config->Relocatable)
1229 OS->addSection(make<ARMExidxSentinelSection>());
1232 // The linker is expected to define SECNAME_start and SECNAME_end
1233 // symbols for a few sections. This function defines them.
1234 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() {
1235 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) {
1236 // These symbols resolve to the image base if the section does not exist.
1237 // A special value -1 indicates end of the section.
1239 addOptionalRegular<ELFT>(Start, OS, 0);
1240 addOptionalRegular<ELFT>(End, OS, -1);
1243 OS = Out::ElfHeader;
1244 addOptionalRegular<ELFT>(Start, OS, 0);
1245 addOptionalRegular<ELFT>(End, OS, 0);
1249 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray);
1250 Define("__init_array_start", "__init_array_end", Out::InitArray);
1251 Define("__fini_array_start", "__fini_array_end", Out::FiniArray);
1253 if (OutputSection *Sec = findSection(".ARM.exidx"))
1254 Define("__exidx_start", "__exidx_end", Sec);
1257 // If a section name is valid as a C identifier (which is rare because of
1258 // the leading '.'), linkers are expected to define __start_<secname> and
1259 // __stop_<secname> symbols. They are at beginning and end of the section,
1260 // respectively. This is not requested by the ELF standard, but GNU ld and
1261 // gold provide the feature, and used by many programs.
1262 template <class ELFT>
1263 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) {
1264 StringRef S = Sec->Name;
1265 if (!isValidCIdentifier(S))
1267 addOptionalRegular<ELFT>(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT);
1268 addOptionalRegular<ELFT>(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT);
1271 template <class ELFT> OutputSection *Writer<ELFT>::findSection(StringRef Name) {
1272 for (OutputSection *Sec : OutputSections)
1273 if (Sec->Name == Name)
1278 static bool needsPtLoad(OutputSection *Sec) {
1279 if (!(Sec->Flags & SHF_ALLOC))
1282 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is
1283 // responsible for allocating space for them, not the PT_LOAD that
1284 // contains the TLS initialization image.
1285 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS)
1290 // Linker scripts are responsible for aligning addresses. Unfortunately, most
1291 // linker scripts are designed for creating two PT_LOADs only, one RX and one
1292 // RW. This means that there is no alignment in the RO to RX transition and we
1293 // cannot create a PT_LOAD there.
1294 static uint64_t computeFlags(uint64_t Flags) {
1296 return PF_R | PF_W | PF_X;
1297 if (Config->SingleRoRx && !(Flags & PF_W))
1298 return Flags | PF_X;
1302 // Decide which program headers to create and which sections to include in each
1304 template <class ELFT> std::vector<PhdrEntry> Writer<ELFT>::createPhdrs() {
1305 std::vector<PhdrEntry> Ret;
1306 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * {
1307 Ret.emplace_back(Type, Flags);
1311 // The first phdr entry is PT_PHDR which describes the program header itself.
1312 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders);
1314 // PT_INTERP must be the second entry if exists.
1315 if (OutputSection *Sec = findSection(".interp"))
1316 AddHdr(PT_INTERP, Sec->getPhdrFlags())->add(Sec);
1318 // Add the first PT_LOAD segment for regular output sections.
1319 uint64_t Flags = computeFlags(PF_R);
1320 PhdrEntry *Load = AddHdr(PT_LOAD, Flags);
1321 for (OutputSection *Sec : OutputSections) {
1322 if (!(Sec->Flags & SHF_ALLOC))
1324 if (!needsPtLoad(Sec))
1327 // Segments are contiguous memory regions that has the same attributes
1328 // (e.g. executable or writable). There is one phdr for each segment.
1329 // Therefore, we need to create a new phdr when the next section has
1330 // different flags or is loaded at a discontiguous address using AT linker
1332 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags());
1333 if (Script->hasLMA(Sec->Name) || Flags != NewFlags) {
1334 Load = AddHdr(PT_LOAD, NewFlags);
1341 // Add a TLS segment if any.
1342 PhdrEntry TlsHdr(PT_TLS, PF_R);
1343 for (OutputSection *Sec : OutputSections)
1344 if (Sec->Flags & SHF_TLS)
1347 Ret.push_back(std::move(TlsHdr));
1349 // Add an entry for .dynamic.
1350 if (In<ELFT>::DynSymTab)
1351 AddHdr(PT_DYNAMIC, In<ELFT>::Dynamic->OutSec->getPhdrFlags())
1352 ->add(In<ELFT>::Dynamic->OutSec);
1354 // PT_GNU_RELRO includes all sections that should be marked as
1355 // read-only by dynamic linker after proccessing relocations.
1356 PhdrEntry RelRo(PT_GNU_RELRO, PF_R);
1357 for (OutputSection *Sec : OutputSections)
1358 if (needsPtLoad(Sec) && isRelroSection<ELFT>(Sec))
1361 Ret.push_back(std::move(RelRo));
1363 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr.
1364 if (!In<ELFT>::EhFrame->empty() && In<ELFT>::EhFrameHdr &&
1365 In<ELFT>::EhFrame->OutSec && In<ELFT>::EhFrameHdr->OutSec)
1366 AddHdr(PT_GNU_EH_FRAME, In<ELFT>::EhFrameHdr->OutSec->getPhdrFlags())
1367 ->add(In<ELFT>::EhFrameHdr->OutSec);
1369 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes
1370 // the dynamic linker fill the segment with random data.
1371 if (OutputSection *Sec = findSection(".openbsd.randomdata"))
1372 AddHdr(PT_OPENBSD_RANDOMIZE, Sec->getPhdrFlags())->add(Sec);
1374 // PT_GNU_STACK is a special section to tell the loader to make the
1375 // pages for the stack non-executable. If you really want an executable
1376 // stack, you can pass -z execstack, but that's not recommended for
1377 // security reasons.
1379 if (Config->ZExecstack)
1380 Perm = PF_R | PF_W | PF_X;
1383 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize;
1385 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable
1386 // is expected to perform W^X violations, such as calling mprotect(2) or
1387 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on
1389 if (Config->ZWxneeded)
1390 AddHdr(PT_OPENBSD_WXNEEDED, PF_X);
1392 // Create one PT_NOTE per a group of contiguous .note sections.
1393 PhdrEntry *Note = nullptr;
1394 for (OutputSection *Sec : OutputSections) {
1395 if (Sec->Type == SHT_NOTE) {
1396 if (!Note || Script->hasLMA(Sec->Name))
1397 Note = AddHdr(PT_NOTE, PF_R);
1406 template <class ELFT>
1407 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry> &Phdrs) {
1408 if (Config->EMachine != EM_ARM)
1410 auto I = std::find_if(
1411 OutputSections.begin(), OutputSections.end(),
1412 [](OutputSection *Sec) { return Sec->Type == SHT_ARM_EXIDX; });
1413 if (I == OutputSections.end())
1416 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME
1417 PhdrEntry ARMExidx(PT_ARM_EXIDX, PF_R);
1419 Phdrs.push_back(ARMExidx);
1422 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the
1423 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic
1424 // linker can set the permissions.
1425 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() {
1426 for (const PhdrEntry &P : Phdrs)
1427 if (P.p_type == PT_LOAD && P.First)
1428 P.First->PageAlign = true;
1430 for (const PhdrEntry &P : Phdrs) {
1431 if (P.p_type != PT_GNU_RELRO)
1434 P.First->PageAlign = true;
1435 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we
1436 // have to align it to a page.
1437 auto End = OutputSections.end();
1438 auto I = std::find(OutputSections.begin(), End, P.Last);
1439 if (I == End || (I + 1) == End)
1441 OutputSection *Sec = *(I + 1);
1442 if (needsPtLoad(Sec))
1443 Sec->PageAlign = true;
1447 bool elf::allocateHeaders(std::vector<PhdrEntry> &Phdrs,
1448 ArrayRef<OutputSection *> OutputSections,
1451 std::find_if(Phdrs.begin(), Phdrs.end(),
1452 [](const PhdrEntry &E) { return E.p_type == PT_LOAD; });
1453 if (FirstPTLoad == Phdrs.end())
1456 uint64_t HeaderSize = getHeaderSize();
1457 if (HeaderSize > Min) {
1459 std::find_if(Phdrs.begin(), Phdrs.end(),
1460 [](const PhdrEntry &E) { return E.p_type == PT_PHDR; });
1461 if (PhdrI != Phdrs.end())
1465 Min = alignDown(Min - HeaderSize, Config->MaxPageSize);
1467 if (!Script->Opt.HasSections)
1468 Config->ImageBase = Min = std::min(Min, Config->ImageBase);
1470 Out::ElfHeader->Addr = Min;
1471 Out::ProgramHeaders->Addr = Min + Out::ElfHeader->Size;
1473 if (Script->hasPhdrsCommands())
1476 if (FirstPTLoad->First)
1477 for (OutputSection *Sec : OutputSections)
1478 if (Sec->FirstInPtLoad == FirstPTLoad->First)
1479 Sec->FirstInPtLoad = Out::ElfHeader;
1480 FirstPTLoad->First = Out::ElfHeader;
1481 if (!FirstPTLoad->Last)
1482 FirstPTLoad->Last = Out::ProgramHeaders;
1486 // We should set file offsets and VAs for elf header and program headers
1487 // sections. These are special, we do not include them into output sections
1488 // list, but have them to simplify the code.
1489 template <class ELFT> void Writer<ELFT>::fixHeaders() {
1490 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size();
1491 // If the script has SECTIONS, assignAddresses will compute the values.
1492 if (Script->Opt.HasSections)
1495 // When -T<section> option is specified, lower the base to make room for those
1498 if (!Config->SectionStartMap.empty())
1499 for (const auto &P : Config->SectionStartMap)
1500 Min = std::min(Min, P.second);
1502 AllocateHeader = allocateHeaders(Phdrs, OutputSections, Min);
1505 // Assign VAs (addresses at run-time) to output sections.
1506 template <class ELFT> void Writer<ELFT>::assignAddresses() {
1507 uint64_t VA = Config->ImageBase;
1508 uint64_t ThreadBssOffset = 0;
1511 VA += getHeaderSize();
1513 for (OutputSection *Sec : OutputSections) {
1514 uint32_t Alignment = Sec->Alignment;
1516 Alignment = std::max<uint32_t>(Alignment, Config->MaxPageSize);
1518 auto I = Config->SectionStartMap.find(Sec->Name);
1519 if (I != Config->SectionStartMap.end())
1522 // We only assign VAs to allocated sections.
1523 if (needsPtLoad(Sec)) {
1524 VA = alignTo(VA, Alignment);
1527 } else if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS) {
1528 uint64_t TVA = VA + ThreadBssOffset;
1529 TVA = alignTo(TVA, Alignment);
1531 ThreadBssOffset = TVA - VA + Sec->Size;
1536 // Adjusts the file alignment for a given output section and returns
1537 // its new file offset. The file offset must be the same with its
1538 // virtual address (modulo the page size) so that the loader can load
1539 // executables without any address adjustment.
1540 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Sec) {
1541 OutputSection *First = Sec->FirstInPtLoad;
1542 // If the section is not in a PT_LOAD, we just have to align it.
1544 return alignTo(Off, Sec->Alignment);
1546 // The first section in a PT_LOAD has to have congruent offset and address
1547 // module the page size.
1549 return alignTo(Off, Config->MaxPageSize, Sec->Addr);
1551 // If two sections share the same PT_LOAD the file offset is calculated
1552 // using this formula: Off2 = Off1 + (VA2 - VA1).
1553 return First->Offset + Sec->Addr - First->Addr;
1556 static uint64_t setOffset(OutputSection *Sec, uint64_t Off) {
1557 if (Sec->Type == SHT_NOBITS) {
1562 Off = getFileAlignment(Off, Sec);
1564 return Off + Sec->Size;
1567 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() {
1569 for (OutputSection *Sec : OutputSections)
1570 if (Sec->Flags & SHF_ALLOC)
1571 Off = setOffset(Sec, Off);
1572 FileSize = alignTo(Off, Config->Wordsize);
1575 // Assign file offsets to output sections.
1576 template <class ELFT> void Writer<ELFT>::assignFileOffsets() {
1578 Off = setOffset(Out::ElfHeader, Off);
1579 Off = setOffset(Out::ProgramHeaders, Off);
1581 for (OutputSection *Sec : OutputSections)
1582 Off = setOffset(Sec, Off);
1584 SectionHeaderOff = alignTo(Off, Config->Wordsize);
1585 FileSize = SectionHeaderOff + (OutputSections.size() + 1) * sizeof(Elf_Shdr);
1588 // Finalize the program headers. We call this function after we assign
1589 // file offsets and VAs to all sections.
1590 template <class ELFT> void Writer<ELFT>::setPhdrs() {
1591 for (PhdrEntry &P : Phdrs) {
1592 OutputSection *First = P.First;
1593 OutputSection *Last = P.Last;
1595 P.p_filesz = Last->Offset - First->Offset;
1596 if (Last->Type != SHT_NOBITS)
1597 P.p_filesz += Last->Size;
1598 P.p_memsz = Last->Addr + Last->Size - First->Addr;
1599 P.p_offset = First->Offset;
1600 P.p_vaddr = First->Addr;
1602 P.p_paddr = First->getLMA();
1604 if (P.p_type == PT_LOAD)
1605 P.p_align = Config->MaxPageSize;
1606 else if (P.p_type == PT_GNU_RELRO)
1609 // The TLS pointer goes after PT_TLS. At least glibc will align it,
1610 // so round up the size to make sure the offsets are correct.
1611 if (P.p_type == PT_TLS) {
1614 P.p_memsz = alignTo(P.p_memsz, P.p_align);
1619 // The entry point address is chosen in the following ways.
1621 // 1. the '-e' entry command-line option;
1622 // 2. the ENTRY(symbol) command in a linker control script;
1623 // 3. the value of the symbol start, if present;
1624 // 4. the address of the first byte of the .text section, if present;
1625 // 5. the address 0.
1626 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() {
1627 // Case 1, 2 or 3. As a special case, if the symbol is actually
1628 // a number, we'll use that number as an address.
1629 if (SymbolBody *B = Symtab<ELFT>::X->find(Config->Entry))
1632 if (!Config->Entry.getAsInteger(0, Addr))
1636 if (OutputSection *Sec = findSection(".text")) {
1637 if (Config->WarnMissingEntry)
1638 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" +
1639 utohexstr(Sec->Addr));
1644 if (Config->WarnMissingEntry)
1645 warn("cannot find entry symbol " + Config->Entry +
1646 "; not setting start address");
1650 static uint16_t getELFType() {
1653 if (Config->Relocatable)
1658 // This function is called after we have assigned address and size
1659 // to each section. This function fixes some predefined
1660 // symbol values that depend on section address and size.
1661 template <class ELFT> void Writer<ELFT>::fixPredefinedSymbols() {
1662 auto Set = [](DefinedRegular *S1, DefinedRegular *S2, OutputSection *Sec,
1674 // _etext is the first location after the last read-only loadable segment.
1675 // _edata is the first location after the last read-write loadable segment.
1676 // _end is the first location after the uninitialized data region.
1677 PhdrEntry *Last = nullptr;
1678 PhdrEntry *LastRO = nullptr;
1679 PhdrEntry *LastRW = nullptr;
1680 for (PhdrEntry &P : Phdrs) {
1681 if (P.p_type != PT_LOAD)
1684 if (P.p_flags & PF_W)
1690 Set(ElfSym::End1, ElfSym::End2, Last->First, Last->p_memsz);
1692 Set(ElfSym::Etext1, ElfSym::Etext2, LastRO->First, LastRO->p_filesz);
1694 Set(ElfSym::Edata1, ElfSym::Edata2, LastRW->First, LastRW->p_filesz);
1697 ElfSym::Bss->Section = findSection(".bss");
1699 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should
1700 // be equal to the _gp symbol's value.
1701 if (Config->EMachine == EM_MIPS) {
1702 if (!ElfSym::MipsGp->Value) {
1703 // Find GP-relative section with the lowest address
1704 // and use this address to calculate default _gp value.
1706 for (const OutputSection *OS : OutputSections)
1707 if ((OS->Flags & SHF_MIPS_GPREL) && OS->Addr < Gp)
1709 if (Gp != (uint64_t)-1)
1710 ElfSym::MipsGp->Value = Gp + 0x7ff0;
1715 template <class ELFT> void Writer<ELFT>::writeHeader() {
1716 uint8_t *Buf = Buffer->getBufferStart();
1717 memcpy(Buf, "\177ELF", 4);
1719 // Write the ELF header.
1720 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf);
1721 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32;
1722 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB;
1723 EHdr->e_ident[EI_VERSION] = EV_CURRENT;
1724 EHdr->e_ident[EI_OSABI] = Config->OSABI;
1725 EHdr->e_type = getELFType();
1726 EHdr->e_machine = Config->EMachine;
1727 EHdr->e_version = EV_CURRENT;
1728 EHdr->e_entry = getEntryAddr();
1729 EHdr->e_shoff = SectionHeaderOff;
1730 EHdr->e_ehsize = sizeof(Elf_Ehdr);
1731 EHdr->e_phnum = Phdrs.size();
1732 EHdr->e_shentsize = sizeof(Elf_Shdr);
1733 EHdr->e_shnum = OutputSections.size() + 1;
1734 EHdr->e_shstrndx = In<ELFT>::ShStrTab->OutSec->SectionIndex;
1736 if (Config->EMachine == EM_ARM)
1737 // We don't currently use any features incompatible with EF_ARM_EABI_VER5,
1738 // but we don't have any firm guarantees of conformance. Linux AArch64
1739 // kernels (as of 2016) require an EABI version to be set.
1740 EHdr->e_flags = EF_ARM_EABI_VER5;
1741 else if (Config->EMachine == EM_MIPS)
1742 EHdr->e_flags = getMipsEFlags<ELFT>();
1744 if (!Config->Relocatable) {
1745 EHdr->e_phoff = sizeof(Elf_Ehdr);
1746 EHdr->e_phentsize = sizeof(Elf_Phdr);
1749 // Write the program header table.
1750 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff);
1751 for (PhdrEntry &P : Phdrs) {
1752 HBuf->p_type = P.p_type;
1753 HBuf->p_flags = P.p_flags;
1754 HBuf->p_offset = P.p_offset;
1755 HBuf->p_vaddr = P.p_vaddr;
1756 HBuf->p_paddr = P.p_paddr;
1757 HBuf->p_filesz = P.p_filesz;
1758 HBuf->p_memsz = P.p_memsz;
1759 HBuf->p_align = P.p_align;
1763 // Write the section header table. Note that the first table entry is null.
1764 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff);
1765 for (OutputSection *Sec : OutputSections)
1766 Sec->writeHeaderTo<ELFT>(++SHdrs);
1769 // Open a result file.
1770 template <class ELFT> void Writer<ELFT>::openFile() {
1771 if (!Config->Is64 && FileSize > UINT32_MAX) {
1772 error("output file too large: " + Twine(FileSize) + " bytes");
1776 unlinkAsync(Config->OutputFile);
1777 ErrorOr<std::unique_ptr<FileOutputBuffer>> BufferOrErr =
1778 FileOutputBuffer::create(Config->OutputFile, FileSize,
1779 FileOutputBuffer::F_executable);
1781 if (auto EC = BufferOrErr.getError())
1782 error("failed to open " + Config->OutputFile + ": " + EC.message());
1784 Buffer = std::move(*BufferOrErr);
1787 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() {
1788 uint8_t *Buf = Buffer->getBufferStart();
1789 for (OutputSection *Sec : OutputSections)
1790 if (Sec->Flags & SHF_ALLOC)
1791 Sec->writeTo<ELFT>(Buf + Sec->Offset);
1794 // Write section contents to a mmap'ed file.
1795 template <class ELFT> void Writer<ELFT>::writeSections() {
1796 uint8_t *Buf = Buffer->getBufferStart();
1798 // PPC64 needs to process relocations in the .opd section
1799 // before processing relocations in code-containing sections.
1800 Out::Opd = findSection(".opd");
1802 Out::OpdBuf = Buf + Out::Opd->Offset;
1803 Out::Opd->template writeTo<ELFT>(Buf + Out::Opd->Offset);
1806 OutputSection *EhFrameHdr =
1807 In<ELFT>::EhFrameHdr ? In<ELFT>::EhFrameHdr->OutSec : nullptr;
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 (OutputSection *Sec : OutputSections)
1813 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA)
1814 Sec->writeTo<ELFT>(Buf + Sec->Offset);
1816 for (OutputSection *Sec : OutputSections)
1817 if (Sec != Out::Opd && Sec != EhFrameHdr && Sec->Type != SHT_REL &&
1818 Sec->Type != SHT_RELA)
1819 Sec->writeTo<ELFT>(Buf + Sec->Offset);
1821 // The .eh_frame_hdr depends on .eh_frame section contents, therefore
1822 // it should be written after .eh_frame is written.
1823 if (EhFrameHdr && !EhFrameHdr->Sections.empty())
1824 EhFrameHdr->writeTo<ELFT>(Buf + EhFrameHdr->Offset);
1827 template <class ELFT> void Writer<ELFT>::writeBuildId() {
1828 if (!In<ELFT>::BuildId || !In<ELFT>::BuildId->OutSec)
1831 // Compute a hash of all sections of the output file.
1832 uint8_t *Start = Buffer->getBufferStart();
1833 uint8_t *End = Start + FileSize;
1834 In<ELFT>::BuildId->writeBuildId({Start, End});
1837 template void elf::writeResult<ELF32LE>();
1838 template void elf::writeResult<ELF32BE>();
1839 template void elf::writeResult<ELF64LE>();
1840 template void elf::writeResult<ELF64BE>();
1842 template bool elf::isRelroSection<ELF32LE>(const OutputSection *);
1843 template bool elf::isRelroSection<ELF32BE>(const OutputSection *);
1844 template bool elf::isRelroSection<ELF64LE>(const OutputSection *);
1845 template bool elf::isRelroSection<ELF64BE>(const OutputSection *);