//===- Writer.cpp ---------------------------------------------------------===// // // The LLVM Linker // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// #include "Writer.h" #include "Chunks.h" #include "Config.h" #include "Error.h" #include "Symbols.h" #include "SymbolTable.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/STLExtras.h" #include "llvm/MC/StringTableBuilder.h" #include "llvm/Support/FileOutputBuffer.h" #include "llvm/Support/raw_ostream.h" using namespace llvm; using namespace llvm::ELF; using namespace llvm::object; using namespace lld; using namespace lld::elf2; static const int PageSize = 4096; // On freebsd x86_64 the first page cannot be mmaped. // On linux that is controled by vm.mmap_min_addr. At least on some x86_64 // installs that is 65536, so the first 15 pages cannot be used. // Given that, the smallest value that can be used in here is 0x10000. // If using 2MB pages, the smallest page aligned address that works is // 0x200000, but it looks like every OS uses 4k pages for executables. // FIXME: This is architecture and OS dependent. static const int VAStart = 0x10000; namespace { // OutputSection represents a section in an output file. It's a // container of chunks. OutputSection and Chunk are 1:N relationship. // Chunks cannot belong to more than one OutputSections. The writer // creates multiple OutputSections and assign them unique, // non-overlapping file offsets and VAs. template class OutputSectionBase { public: typedef typename std::conditional::type Elf_Dyn; typedef typename std::conditional::type uintX_t; typedef typename std::conditional::type HeaderT; OutputSectionBase(StringRef Name, uint32_t sh_type, uintX_t sh_flags) : Name(Name) { memset(&Header, 0, sizeof(HeaderT)); Header.sh_type = sh_type; Header.sh_flags = sh_flags; } void setVA(uintX_t VA) { Header.sh_addr = VA; } uintX_t getVA() const { return Header.sh_addr; } void setFileOffset(uintX_t Off) { Header.sh_offset = Off; } template void writeHeaderTo(typename ELFFile>::Elf_Shdr *SHdr); StringRef getName() { return Name; } void setNameOffset(uintX_t Offset) { Header.sh_name = Offset; } unsigned getSectionIndex() const { return SectionIndex; } void setSectionIndex(unsigned I) { SectionIndex = I; } // Returns the size of the section in the output file. uintX_t getSize() { return Header.sh_size; } void setSize(uintX_t Val) { Header.sh_size = Val; } uintX_t getFlags() { return Header.sh_flags; } uintX_t getFileOff() { return Header.sh_offset; } uintX_t getAlign() { // The ELF spec states that a value of 0 means the section has no alignment // constraits. return std::max(Header.sh_addralign, 1); } uint32_t getType() { return Header.sh_type; } static unsigned getAddrSize() { return Is64Bits ? 8 : 4; } virtual void finalize() {} virtual void writeTo(uint8_t *Buf) = 0; protected: StringRef Name; HeaderT Header; unsigned SectionIndex; ~OutputSectionBase() = default; }; template class SymbolTableSection; template struct DynamicReloc { typedef typename ELFFile::Elf_Rel Elf_Rel; const InputSection &C; const Elf_Rel &RI; }; static bool relocNeedsPLT(uint32_t Type) { switch (Type) { default: return false; case R_X86_64_PLT32: return true; } } static bool relocNeedsGOT(uint32_t Type) { if (relocNeedsPLT(Type)) return true; switch (Type) { default: return false; case R_X86_64_GOTPCREL: return true; } } template class GotSection final : public OutputSectionBase { typedef OutputSectionBase Base; typedef typename Base::uintX_t uintX_t; public: GotSection() : OutputSectionBase(".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE) { this->Header.sh_addralign = this->getAddrSize(); } void finalize() override { this->Header.sh_size = Entries.size() * this->getAddrSize(); } void writeTo(uint8_t *Buf) override {} void addEntry(SymbolBody *Sym) { Sym->setGotIndex(Entries.size()); Entries.push_back(Sym); } bool empty() const { return Entries.empty(); } uintX_t getEntryAddr(const SymbolBody &B) const { return this->getVA() + B.getGotIndex() * this->getAddrSize(); } private: std::vector Entries; }; template class PltSection final : public OutputSectionBase { typedef OutputSectionBase Base; typedef typename Base::uintX_t uintX_t; public: PltSection(const GotSection &GotSec) : OutputSectionBase(".plt", SHT_PROGBITS, SHF_ALLOC | SHF_EXECINSTR), GotSec(GotSec) { this->Header.sh_addralign = 16; } void finalize() override { this->Header.sh_size = Entries.size() * EntrySize; } void writeTo(uint8_t *Buf) override { uintptr_t Start = reinterpret_cast(Buf); ArrayRef Jmp = {0xff, 0x25}; // jmpq *val(%rip) for (const SymbolBody *E : Entries) { uintptr_t InstPos = reinterpret_cast(Buf); memcpy(Buf, Jmp.data(), Jmp.size()); Buf += Jmp.size(); uintptr_t OffsetInPLT = (InstPos + 6) - Start; uintptr_t Delta = GotSec.getEntryAddr(*E) - (this->getVA() + OffsetInPLT); assert(isInt<32>(Delta)); support::endian::write32le(Buf, Delta); Buf += 4; *Buf = 0x90; // nop ++Buf; *Buf = 0x90; // nop ++Buf; } } void addEntry(SymbolBody *Sym) { Sym->setPltIndex(Entries.size()); Entries.push_back(Sym); } bool empty() const { return Entries.empty(); } uintX_t getEntryAddr(const SymbolBody &B) const { return this->getVA() + B.getPltIndex() * EntrySize; } static const unsigned EntrySize = 8; private: std::vector Entries; const GotSection &GotSec; }; template class RelocationSection final : public OutputSectionBase { typedef typename ELFFile::Elf_Rel Elf_Rel; typedef typename ELFFile::Elf_Rela Elf_Rela; public: RelocationSection(SymbolTableSection &DynSymSec, const GotSection &GotSec, bool IsRela) : OutputSectionBase(IsRela ? ".rela.dyn" : ".rel.dyn", IsRela ? SHT_RELA : SHT_REL, SHF_ALLOC), DynSymSec(DynSymSec), GotSec(GotSec), IsRela(IsRela) { this->Header.sh_entsize = IsRela ? sizeof(Elf_Rela) : sizeof(Elf_Rel); this->Header.sh_addralign = ELFT::Is64Bits ? 8 : 4; } void addReloc(const DynamicReloc &Reloc) { Relocs.push_back(Reloc); } void finalize() override { this->Header.sh_link = DynSymSec.getSectionIndex(); this->Header.sh_size = Relocs.size() * this->Header.sh_entsize; } void writeTo(uint8_t *Buf) override { auto *P = reinterpret_cast(Buf); bool IsMips64EL = Relocs[0].C.getFile()->getObj()->isMips64EL(); for (const DynamicReloc &Rel : Relocs) { const InputSection &C = Rel.C; const Elf_Rel &RI = Rel.RI; OutputSection *Out = C.getOutputSection(); uint32_t SymIndex = RI.getSymbol(IsMips64EL); const SymbolBody *Body = C.getFile()->getSymbolBody(SymIndex); uint32_t Type = RI.getType(IsMips64EL); if (relocNeedsGOT(Type)) { P->r_offset = GotSec.getEntryAddr(*Body); P->setSymbolAndType(Body->getDynamicSymbolTableIndex(), R_X86_64_GLOB_DAT, IsMips64EL); } else { P->r_offset = RI.r_offset + C.getOutputSectionOff() + Out->getVA(); P->setSymbolAndType(Body->getDynamicSymbolTableIndex(), Type, IsMips64EL); if (IsRela) P->r_addend = static_cast(RI).r_addend; } ++P; } } bool hasRelocs() const { return !Relocs.empty(); } bool isRela() const { return IsRela; } private: std::vector> Relocs; SymbolTableSection &DynSymSec; const GotSection &GotSec; const bool IsRela; }; } template class lld::elf2::OutputSection final : public OutputSectionBase { public: typedef typename OutputSectionBase::uintX_t uintX_t; typedef typename ELFFile::Elf_Shdr Elf_Shdr; typedef typename ELFFile::Elf_Sym Elf_Sym; typedef typename ELFFile::Elf_Rel Elf_Rel; typedef typename ELFFile::Elf_Rela Elf_Rela; OutputSection(const PltSection &PltSec, const GotSection &GotSec, StringRef Name, uint32_t sh_type, uintX_t sh_flags) : OutputSectionBase(Name, sh_type, sh_flags), PltSec(PltSec), GotSec(GotSec) {} void addChunk(InputSection *C); void writeTo(uint8_t *Buf) override; template void relocate(uint8_t *Buf, iterator_range *> Rels, const ObjectFile &File, uintX_t BaseAddr); void relocateOne(uint8_t *Buf, const Elf_Rela &Rel, uint32_t Type, uintX_t BaseAddr, uintX_t SymVA); void relocateOne(uint8_t *Buf, const Elf_Rel &Rel, uint32_t Type, uintX_t BaseAddr, uintX_t SymVA); private: std::vector *> Chunks; const PltSection &PltSec; const GotSection &GotSec; }; namespace { template class InterpSection final : public OutputSectionBase { public: InterpSection() : OutputSectionBase(".interp", SHT_PROGBITS, SHF_ALLOC) { this->Header.sh_size = Config->DynamicLinker.size() + 1; this->Header.sh_addralign = 1; } void writeTo(uint8_t *Buf) override { memcpy(Buf, Config->DynamicLinker.data(), Config->DynamicLinker.size()); } }; template class StringTableSection final : public OutputSectionBase { public: typedef typename OutputSectionBase::uintX_t uintX_t; StringTableSection(bool Dynamic) : OutputSectionBase(Dynamic ? ".dynstr" : ".strtab", SHT_STRTAB, Dynamic ? (uintX_t)SHF_ALLOC : 0), Dynamic(Dynamic) { this->Header.sh_addralign = 1; } void add(StringRef S) { StrTabBuilder.add(S); } size_t getFileOff(StringRef S) const { return StrTabBuilder.getOffset(S); } StringRef data() const { return StrTabBuilder.data(); } void writeTo(uint8_t *Buf) override; void finalize() override { StrTabBuilder.finalize(StringTableBuilder::ELF); this->Header.sh_size = StrTabBuilder.data().size(); } bool isDynamic() const { return Dynamic; } private: const bool Dynamic; llvm::StringTableBuilder StrTabBuilder; }; template class Writer; template class SymbolTableSection final : public OutputSectionBase { public: typedef typename ELFFile::Elf_Shdr Elf_Shdr; typedef typename ELFFile::Elf_Sym Elf_Sym; typedef typename ELFFile::Elf_Sym_Range Elf_Sym_Range; typedef typename OutputSectionBase::uintX_t uintX_t; SymbolTableSection(Writer &W, SymbolTable &Table, StringTableSection &StrTabSec) : OutputSectionBase( StrTabSec.isDynamic() ? ".dynsym" : ".symtab", StrTabSec.isDynamic() ? SHT_DYNSYM : SHT_SYMTAB, StrTabSec.isDynamic() ? (uintX_t)SHF_ALLOC : 0), Table(Table), StrTabSec(StrTabSec), W(W) { typedef OutputSectionBase Base; typename Base::HeaderT &Header = this->Header; Header.sh_entsize = sizeof(Elf_Sym); Header.sh_addralign = ELFT::Is64Bits ? 8 : 4; } void finalize() override { this->Header.sh_size = getNumSymbols() * sizeof(Elf_Sym); this->Header.sh_link = StrTabSec.getSectionIndex(); this->Header.sh_info = NumLocals + 1; } void writeTo(uint8_t *Buf) override; const SymbolTable &getSymTable() const { return Table; } void addSymbol(StringRef Name, bool isLocal = false) { StrTabSec.add(Name); ++NumVisible; if (isLocal) ++NumLocals; } StringTableSection &getStrTabSec() { return StrTabSec; } unsigned getNumSymbols() const { return NumVisible + 1; } private: SymbolTable &Table; StringTableSection &StrTabSec; unsigned NumVisible = 0; unsigned NumLocals = 0; const Writer &W; }; template class HashTableSection final : public OutputSectionBase { typedef typename ELFFile::Elf_Word Elf_Word; public: HashTableSection(SymbolTableSection &DynSymSec) : OutputSectionBase(".hash", SHT_HASH, SHF_ALLOC), DynSymSec(DynSymSec) { this->Header.sh_entsize = sizeof(Elf_Word); this->Header.sh_addralign = sizeof(Elf_Word); } void addSymbol(SymbolBody *S) { StringRef Name = S->getName(); DynSymSec.addSymbol(Name); Hashes.push_back(hash(Name)); S->setDynamicSymbolTableIndex(Hashes.size()); } void finalize() override { this->Header.sh_link = DynSymSec.getSectionIndex(); assert(DynSymSec.getNumSymbols() == Hashes.size() + 1); unsigned NumEntries = 2; // nbucket and nchain. NumEntries += DynSymSec.getNumSymbols(); // The chain entries. // Create as many buckets as there are symbols. // FIXME: This is simplistic. We can try to optimize it, but implementing // support for SHT_GNU_HASH is probably even more profitable. NumEntries += DynSymSec.getNumSymbols(); this->Header.sh_size = NumEntries * sizeof(Elf_Word); } void writeTo(uint8_t *Buf) override { unsigned NumSymbols = DynSymSec.getNumSymbols(); auto *P = reinterpret_cast(Buf); *P++ = NumSymbols; // nbucket *P++ = NumSymbols; // nchain Elf_Word *Buckets = P; Elf_Word *Chains = P + NumSymbols; for (unsigned I = 1; I < NumSymbols; ++I) { uint32_t Hash = Hashes[I - 1] % NumSymbols; Chains[I] = Buckets[Hash]; Buckets[Hash] = I; } } SymbolTableSection &getDynSymSec() { return DynSymSec; } private: uint32_t hash(StringRef Name) { uint32_t H = 0; for (char C : Name) { H = (H << 4) + C; uint32_t G = H & 0xf0000000; if (G) H ^= G >> 24; H &= ~G; } return H; } SymbolTableSection &DynSymSec; std::vector Hashes; }; template class DynamicSection final : public OutputSectionBase { typedef OutputSectionBase Base; typedef typename Base::HeaderT HeaderT; typedef typename Base::Elf_Dyn Elf_Dyn; public: DynamicSection(SymbolTable &SymTab, HashTableSection &HashSec, RelocationSection &RelaDynSec) : OutputSectionBase(".dynamic", SHT_DYNAMIC, SHF_ALLOC | SHF_WRITE), HashSec(HashSec), DynSymSec(HashSec.getDynSymSec()), DynStrSec(DynSymSec.getStrTabSec()), RelaDynSec(RelaDynSec), SymTab(SymTab) { typename Base::HeaderT &Header = this->Header; Header.sh_addralign = ELFT::Is64Bits ? 8 : 4; Header.sh_entsize = ELFT::Is64Bits ? 16 : 8; } void finalize() override { typename Base::HeaderT &Header = this->Header; Header.sh_link = DynStrSec.getSectionIndex(); unsigned NumEntries = 0; if (RelaDynSec.hasRelocs()) { ++NumEntries; // DT_RELA / DT_REL ++NumEntries; // DT_RELASZ / DTRELSZ } ++NumEntries; // DT_SYMTAB ++NumEntries; // DT_STRTAB ++NumEntries; // DT_STRSZ ++NumEntries; // DT_HASH StringRef RPath = Config->RPath; if (!RPath.empty()) { ++NumEntries; // DT_RUNPATH DynStrSec.add(RPath); } const std::vector> &SharedFiles = SymTab.getSharedFiles(); for (const std::unique_ptr &File : SharedFiles) DynStrSec.add(File->getName()); NumEntries += SharedFiles.size(); ++NumEntries; // DT_NULL Header.sh_size = NumEntries * Header.sh_entsize; } void writeTo(uint8_t *Buf) override { auto *P = reinterpret_cast(Buf); if (RelaDynSec.hasRelocs()) { bool IsRela = RelaDynSec.isRela(); P->d_tag = IsRela ? DT_RELA : DT_REL; P->d_un.d_ptr = RelaDynSec.getVA(); ++P; P->d_tag = IsRela ? DT_RELASZ : DT_RELSZ; P->d_un.d_val = RelaDynSec.getSize(); ++P; } P->d_tag = DT_SYMTAB; P->d_un.d_ptr = DynSymSec.getVA(); ++P; P->d_tag = DT_STRTAB; P->d_un.d_ptr = DynStrSec.getVA(); ++P; P->d_tag = DT_STRSZ; P->d_un.d_val = DynStrSec.data().size(); ++P; P->d_tag = DT_HASH; P->d_un.d_ptr = HashSec.getVA(); ++P; StringRef RPath = Config->RPath; if (!RPath.empty()) { P->d_tag = DT_RUNPATH; P->d_un.d_val = DynStrSec.getFileOff(RPath); ++P; } const std::vector> &SharedFiles = SymTab.getSharedFiles(); for (const std::unique_ptr &File : SharedFiles) { P->d_tag = DT_NEEDED; P->d_un.d_val = DynStrSec.getFileOff(File->getName()); ++P; } P->d_tag = DT_NULL; P->d_un.d_val = 0; ++P; } private: HashTableSection &HashSec; SymbolTableSection &DynSymSec; StringTableSection &DynStrSec; RelocationSection &RelaDynSec; SymbolTable &SymTab; }; static uint32_t convertSectionFlagsToPHDRFlags(uint64_t Flags) { uint32_t Ret = PF_R; if (Flags & SHF_WRITE) Ret |= PF_W; if (Flags & SHF_EXECINSTR) Ret |= PF_X; return Ret; } template class ProgramHeader { public: typedef typename std::conditional::type uintX_t; typedef typename std::conditional::type HeaderT; ProgramHeader(uintX_t p_type, uintX_t p_flags) { std::memset(&Header, 0, sizeof(HeaderT)); Header.p_type = p_type; Header.p_flags = p_flags; Header.p_align = PageSize; } void setValuesFromSection(OutputSectionBase &Sec) { Header.p_flags = convertSectionFlagsToPHDRFlags(Sec.getFlags()); Header.p_offset = Sec.getFileOff(); Header.p_vaddr = Sec.getVA(); Header.p_paddr = Header.p_vaddr; Header.p_filesz = Sec.getSize(); Header.p_memsz = Header.p_filesz; Header.p_align = Sec.getAlign(); } template void writeHeaderTo(typename ELFFile>::Elf_Phdr *PHDR) { PHDR->p_type = Header.p_type; PHDR->p_flags = Header.p_flags; PHDR->p_offset = Header.p_offset; PHDR->p_vaddr = Header.p_vaddr; PHDR->p_paddr = Header.p_paddr; PHDR->p_filesz = Header.p_filesz; PHDR->p_memsz = Header.p_memsz; PHDR->p_align = Header.p_align; } HeaderT Header; bool Closed = false; }; // The writer writes a SymbolTable result to a file. template class Writer { public: typedef typename ELFFile::uintX_t uintX_t; typedef typename ELFFile::Elf_Shdr Elf_Shdr; typedef typename ELFFile::Elf_Ehdr Elf_Ehdr; typedef typename ELFFile::Elf_Phdr Elf_Phdr; typedef typename ELFFile::Elf_Sym Elf_Sym; typedef typename ELFFile::Elf_Sym_Range Elf_Sym_Range; typedef typename ELFFile::Elf_Rela Elf_Rela; Writer(SymbolTable *T) : SymTabSec(*this, *T, StrTabSec), DynSymSec(*this, *T, DynStrSec), RelaDynSec(DynSymSec, GotSec, T->shouldUseRela()), PltSec(GotSec), HashSec(DynSymSec), DynamicSec(*T, HashSec, RelaDynSec) {} void run(); const OutputSection &getBSS() const { assert(BSSSec); return *BSSSec; } private: void createSections(); template void scanRelocs(const InputSection &C, iterator_range *> Rels); void scanRelocs(const InputSection &C); void assignAddresses(); void openFile(StringRef OutputPath); void writeHeader(); void writeSections(); bool needsInterpSection() const { return !SymTabSec.getSymTable().getSharedFiles().empty() && !Config->DynamicLinker.empty(); } bool needsDynamicSections() const { return !SymTabSec.getSymTable().getSharedFiles().empty() || Config->Shared; } unsigned getVAStart() const { return Config->Shared ? 0 : VAStart; } std::unique_ptr Buffer; llvm::SpecificBumpPtrAllocator> CAlloc; std::vector *> OutputSections; unsigned getNumSections() const { return OutputSections.size() + 1; } llvm::BumpPtrAllocator PAlloc; std::vector *> PHDRs; ProgramHeader FileHeaderPHDR{PT_LOAD, PF_R}; ProgramHeader InterpPHDR{PT_INTERP, 0}; ProgramHeader DynamicPHDR{PT_DYNAMIC, 0}; uintX_t FileSize; uintX_t ProgramHeaderOff; uintX_t SectionHeaderOff; StringTableSection StrTabSec = { /*dynamic=*/false }; StringTableSection DynStrSec = { /*dynamic=*/true }; SymbolTableSection SymTabSec; SymbolTableSection DynSymSec; RelocationSection RelaDynSec; GotSection GotSec; PltSection PltSec; HashTableSection HashSec; DynamicSection DynamicSec; InterpSection InterpSec; OutputSection *BSSSec = nullptr; }; } // anonymous namespace namespace lld { namespace elf2 { template void writeResult(SymbolTable *Symtab) { Writer(Symtab).run(); } template void writeResult(SymbolTable *); template void writeResult(SymbolTable *); template void writeResult(SymbolTable *); template void writeResult(SymbolTable *); } // namespace elf2 } // namespace lld // The main function of the writer. template void Writer::run() { createSections(); assignAddresses(); openFile(Config->OutputFile); writeHeader(); writeSections(); error(Buffer->commit()); } template void OutputSection::addChunk(InputSection *C) { Chunks.push_back(C); C->setOutputSection(this); uint32_t Align = C->getAlign(); if (Align > this->Header.sh_addralign) this->Header.sh_addralign = Align; uintX_t Off = this->Header.sh_size; Off = RoundUpToAlignment(Off, Align); C->setOutputSectionOff(Off); Off += C->getSize(); this->Header.sh_size = Off; } template static typename ELFFile::uintX_t getSymVA(const DefinedRegular *DR) { const InputSection *SC = &DR->Section; OutputSection *OS = SC->getOutputSection(); return OS->getVA() + SC->getOutputSectionOff() + DR->Sym.st_value; } template static typename ELFFile::uintX_t getLocalSymVA(const typename ELFFile::Elf_Sym *Sym, const ObjectFile &File) { uint32_t SecIndex = Sym->st_shndx; if (SecIndex == SHN_XINDEX) SecIndex = File.getObj()->getExtendedSymbolTableIndex( Sym, File.getSymbolTable(), File.getSymbolTableShndx()); ArrayRef *> Chunks = File.getChunks(); InputSection *Section = Chunks[SecIndex]; OutputSection *Out = Section->getOutputSection(); return Out->getVA() + Section->getOutputSectionOff() + Sym->st_value; } template void OutputSection::relocateOne(uint8_t *Buf, const Elf_Rel &Rel, uint32_t Type, uintX_t BaseAddr, uintX_t SymVA) { uintX_t Offset = Rel.r_offset; uint8_t *Location = Buf + Offset; switch (Type) { case R_386_32: support::endian::write32le(Location, SymVA); break; default: llvm::errs() << Twine("unrecognized reloc ") + Twine(Type) << '\n'; break; } } template void OutputSection::relocateOne(uint8_t *Buf, const Elf_Rela &Rel, uint32_t Type, uintX_t BaseAddr, uintX_t SymVA) { uintX_t Offset = Rel.r_offset; uint8_t *Location = Buf + Offset; switch (Type) { case R_X86_64_PC32: support::endian::write32le(Location, SymVA + (Rel.r_addend - (BaseAddr + Offset))); break; case R_X86_64_64: support::endian::write64le(Location, SymVA + Rel.r_addend); break; case R_X86_64_32: { case R_X86_64_32S: uint64_t VA = SymVA + Rel.r_addend; if (Type == R_X86_64_32 && !isUInt<32>(VA)) error("R_X86_64_32 out of range"); else if (!isInt<32>(VA)) error("R_X86_64_32S out of range"); support::endian::write32le(Location, VA); break; } default: llvm::errs() << Twine("unrecognized reloc ") + Twine(Type) << '\n'; break; } } template template void OutputSection::relocate( uint8_t *Buf, iterator_range *> Rels, const ObjectFile &File, uintX_t BaseAddr) { typedef Elf_Rel_Impl RelType; bool IsMips64EL = File.getObj()->isMips64EL(); for (const RelType &RI : Rels) { uint32_t SymIndex = RI.getSymbol(IsMips64EL); uint32_t Type = RI.getType(IsMips64EL); uintX_t SymVA; // Handle relocations for local symbols -- they never get // resolved so we don't allocate a SymbolBody. const Elf_Shdr *SymTab = File.getSymbolTable(); if (SymIndex < SymTab->sh_info) { const Elf_Sym *Sym = File.getObj()->getRelocationSymbol(&RI, SymTab); if (!Sym) continue; SymVA = getLocalSymVA(Sym, File); } else { const SymbolBody *Body = File.getSymbolBody(SymIndex); if (!Body) continue; switch (Body->kind()) { case SymbolBody::DefinedRegularKind: SymVA = getSymVA(cast>(Body)); break; case SymbolBody::DefinedAbsoluteKind: SymVA = cast>(Body)->Sym.st_value; break; case SymbolBody::DefinedCommonKind: { auto *DC = cast>(Body); SymVA = DC->OutputSec->getVA() + DC->OffsetInBSS; break; } case SymbolBody::SharedKind: if (relocNeedsPLT(Type)) { SymVA = PltSec.getEntryAddr(*Body); Type = R_X86_64_PC32; } else if (relocNeedsGOT(Type)) { SymVA = GotSec.getEntryAddr(*Body); Type = R_X86_64_PC32; } else { continue; } break; case SymbolBody::UndefinedKind: assert(Body->isWeak() && "Undefined symbol reached writer"); SymVA = 0; break; case SymbolBody::LazyKind: llvm_unreachable("Lazy symbol reached writer"); } } relocateOne(Buf, RI, Type, BaseAddr, SymVA); } } template void OutputSection::writeTo(uint8_t *Buf) { for (InputSection *C : Chunks) { C->writeTo(Buf); const ObjectFile *File = C->getFile(); ELFFile *EObj = File->getObj(); uint8_t *Base = Buf + C->getOutputSectionOff(); uintX_t BaseAddr = this->getVA() + C->getOutputSectionOff(); // Iterate over all relocation sections that apply to this section. for (const Elf_Shdr *RelSec : C->RelocSections) { if (RelSec->sh_type == SHT_RELA) relocate(Base, EObj->relas(RelSec), *File, BaseAddr); else relocate(Base, EObj->rels(RelSec), *File, BaseAddr); } } } template void StringTableSection::writeTo(uint8_t *Buf) { StringRef Data = StrTabBuilder.data(); memcpy(Buf, Data.data(), Data.size()); } template static int compareSym(const typename ELFFile::Elf_Sym *A, const typename ELFFile::Elf_Sym *B) { uint32_t AN = A->st_name; uint32_t BN = B->st_name; assert(AN != BN); return AN - BN; } static bool includeInSymtab(const SymbolBody &B) { if (B.isLazy()) return false; if (!B.isUsedInRegularObj()) return false; uint8_t V = B.getMostConstrainingVisibility(); if (V != STV_DEFAULT && V != STV_PROTECTED) return false; return true; } template void SymbolTableSection::writeTo(uint8_t *Buf) { const OutputSection *Out = nullptr; const InputSection *Section = nullptr; Buf += sizeof(Elf_Sym); // All symbols with STB_LOCAL binding precede the weak and global symbols. // .dynsym only contains global symbols. if (!Config->DiscardAll && !StrTabSec.isDynamic()) { for (const std::unique_ptr &FileB : Table.getObjectFiles()) { auto &File = cast>(*FileB); Elf_Sym_Range Syms = File.getLocalSymbols(); for (const Elf_Sym &Sym : Syms) { auto *ESym = reinterpret_cast(Buf); uint32_t SecIndex = Sym.st_shndx; ErrorOr SymName = Sym.getName(File.getStringTable()); if (Config->DiscardLocals && SymName->startswith(".L")) continue; ESym->st_name = (SymName) ? StrTabSec.getFileOff(*SymName) : 0; ESym->st_size = Sym.st_size; ESym->setBindingAndType(Sym.getBinding(), Sym.getType()); if (SecIndex == SHN_XINDEX) SecIndex = File.getObj()->getExtendedSymbolTableIndex( &Sym, File.getSymbolTable(), File.getSymbolTableShndx()); ArrayRef *> Chunks = File.getChunks(); Section = Chunks[SecIndex]; assert(Section != nullptr); Out = Section->getOutputSection(); assert(Out != nullptr); ESym->st_shndx = Out->getSectionIndex(); ESym->st_value = Out->getVA() + Section->getOutputSectionOff() + Sym.st_value; Buf += sizeof(Elf_Sym); } } } for (auto &P : Table.getSymbols()) { StringRef Name = P.first; Symbol *Sym = P.second; SymbolBody *Body = Sym->Body; if (!includeInSymtab(*Body)) continue; const Elf_Sym &InputSym = cast>(Body)->Sym; auto *ESym = reinterpret_cast(Buf); ESym->st_name = StrTabSec.getFileOff(Name); Out = nullptr; Section = nullptr; switch (Body->kind()) { case SymbolBody::DefinedRegularKind: Section = &cast>(Body)->Section; break; case SymbolBody::DefinedCommonKind: Out = &W.getBSS(); break; case SymbolBody::UndefinedKind: case SymbolBody::DefinedAbsoluteKind: case SymbolBody::SharedKind: break; case SymbolBody::LazyKind: llvm_unreachable("Lazy symbol got to output symbol table!"); } ESym->setBindingAndType(InputSym.getBinding(), InputSym.getType()); ESym->st_size = InputSym.st_size; ESym->setVisibility(Body->getMostConstrainingVisibility()); if (InputSym.isAbsolute()) { ESym->st_shndx = SHN_ABS; ESym->st_value = InputSym.st_value; } if (Section) Out = Section->getOutputSection(); if (Out) { ESym->st_shndx = Out->getSectionIndex(); uintX_t VA = Out->getVA(); if (Section) VA += Section->getOutputSectionOff(); if (auto *C = dyn_cast>(Body)) VA += C->OffsetInBSS; else VA += InputSym.st_value; ESym->st_value = VA; } Buf += sizeof(Elf_Sym); } } template template void OutputSectionBase::writeHeaderTo( typename ELFFile>::Elf_Shdr *SHdr) { SHdr->sh_name = Header.sh_name; SHdr->sh_type = Header.sh_type; SHdr->sh_flags = Header.sh_flags; SHdr->sh_addr = Header.sh_addr; SHdr->sh_offset = Header.sh_offset; SHdr->sh_size = Header.sh_size; SHdr->sh_link = Header.sh_link; SHdr->sh_info = Header.sh_info; SHdr->sh_addralign = Header.sh_addralign; SHdr->sh_entsize = Header.sh_entsize; } namespace { template struct SectionKey { typedef typename std::conditional::type uintX_t; StringRef Name; uint32_t sh_type; uintX_t sh_flags; }; } namespace llvm { template struct DenseMapInfo> { static SectionKey getEmptyKey() { return SectionKey{DenseMapInfo::getEmptyKey(), 0, 0}; } static SectionKey getTombstoneKey() { return SectionKey{DenseMapInfo::getTombstoneKey(), 0, 0}; } static unsigned getHashValue(const SectionKey &Val) { return hash_combine(Val.Name, Val.sh_type, Val.sh_flags); } static bool isEqual(const SectionKey &LHS, const SectionKey &RHS) { return DenseMapInfo::isEqual(LHS.Name, RHS.Name) && LHS.sh_type == RHS.sh_type && LHS.sh_flags == RHS.sh_flags; } }; } template static bool cmpAlign(const DefinedCommon *A, const DefinedCommon *B) { return A->MaxAlignment > B->MaxAlignment; } template static bool compSec(OutputSectionBase *A, OutputSectionBase *B) { // Place SHF_ALLOC sections first. return (A->getFlags() & SHF_ALLOC) && !(B->getFlags() & SHF_ALLOC); } // The reason we have to do this early scan is as follows // * To mmap the output file, we need to know the size // * For that, we need to know how many dynamic relocs we will have. // It might be possible to avoid this by outputting the file with write: // * Write the allocated output sections, computing addresses. // * Apply relocations, recording which ones require a dynamic reloc. // * Write the dynamic relocations. // * Write the rest of the file. template template void Writer::scanRelocs( const InputSection &C, iterator_range *> Rels) { typedef Elf_Rel_Impl RelType; const ObjectFile &File = *C.getFile(); bool IsMips64EL = File.getObj()->isMips64EL(); for (const RelType &RI : Rels) { uint32_t SymIndex = RI.getSymbol(IsMips64EL); SymbolBody *Body = File.getSymbolBody(SymIndex); if (!Body) continue; auto *S = dyn_cast>(Body); if (!S) continue; uint32_t Type = RI.getType(IsMips64EL); if (relocNeedsPLT(Type)) { if (Body->isInPlt()) continue; PltSec.addEntry(Body); } if (relocNeedsGOT(Type)) { if (Body->isInGot()) continue; GotSec.addEntry(Body); } RelaDynSec.addReloc({C, RI}); } } template void Writer::scanRelocs(const InputSection &C) { const ObjectFile *File = C.getFile(); ELFFile *EObj = File->getObj(); if (!(C.getSectionHdr()->sh_flags & SHF_ALLOC)) return; for (const Elf_Shdr *RelSec : C.RelocSections) { if (RelSec->sh_type == SHT_RELA) scanRelocs(C, EObj->relas(RelSec)); else scanRelocs(C, EObj->rels(RelSec)); } } // Create output section objects and add them to OutputSections. template void Writer::createSections() { SmallDenseMap, OutputSection *> Map; auto getSection = [&](StringRef Name, uint32_t sh_type, uintX_t sh_flags) -> OutputSection * { SectionKey Key{Name, sh_type, sh_flags}; OutputSection *&Sec = Map[Key]; if (!Sec) { Sec = new (CAlloc.Allocate()) OutputSection( PltSec, GotSec, Key.Name, Key.sh_type, Key.sh_flags); OutputSections.push_back(Sec); } return Sec; }; // FIXME: Try to avoid the extra walk over all global symbols. const SymbolTable &Symtab = SymTabSec.getSymTable(); std::vector *> CommonSymbols; for (auto &P : Symtab.getSymbols()) { StringRef Name = P.first; SymbolBody *Body = P.second->Body; if (Body->isStrongUndefined()) error(Twine("undefined symbol: ") + Name); if (auto *C = dyn_cast>(Body)) CommonSymbols.push_back(C); if (!includeInSymtab(*Body)) continue; SymTabSec.addSymbol(Name); // FIXME: This adds way too much to the dynamic symbol table. We only // need to add the symbols use by dynamic relocations when producing // an executable (ignoring --export-dynamic). if (needsDynamicSections()) HashSec.addSymbol(Body); } for (const std::unique_ptr &FileB : Symtab.getObjectFiles()) { auto &File = cast>(*FileB); if (!Config->DiscardAll) { Elf_Sym_Range Syms = File.getLocalSymbols(); for (const Elf_Sym &Sym : Syms) { ErrorOr SymName = Sym.getName(File.getStringTable()); if (SymName && !(Config->DiscardLocals && SymName->startswith(".L"))) SymTabSec.addSymbol(*SymName, true); } } for (InputSection *C : File.getChunks()) { if (!C) continue; const Elf_Shdr *H = C->getSectionHdr(); OutputSection *Sec = getSection(C->getSectionName(), H->sh_type, H->sh_flags); Sec->addChunk(C); scanRelocs(*C); } } BSSSec = getSection(".bss", SHT_NOBITS, SHF_ALLOC | SHF_WRITE); // Sort the common symbols by alignment as an heuristic to pack them better. std::stable_sort(CommonSymbols.begin(), CommonSymbols.end(), cmpAlign); uintX_t Off = BSSSec->getSize(); for (DefinedCommon *C : CommonSymbols) { const Elf_Sym &Sym = C->Sym; uintX_t Align = C->MaxAlignment; Off = RoundUpToAlignment(Off, Align); C->OffsetInBSS = Off; C->OutputSec = BSSSec; Off += Sym.st_size; } BSSSec->setSize(Off); OutputSections.push_back(&SymTabSec); OutputSections.push_back(&StrTabSec); if (needsDynamicSections()) { if (needsInterpSection()) OutputSections.push_back(&InterpSec); OutputSections.push_back(&DynSymSec); OutputSections.push_back(&HashSec); OutputSections.push_back(&DynamicSec); OutputSections.push_back(&DynStrSec); if (RelaDynSec.hasRelocs()) OutputSections.push_back(&RelaDynSec); if (!GotSec.empty()) OutputSections.push_back(&GotSec); if (!PltSec.empty()) OutputSections.push_back(&PltSec); } std::stable_sort(OutputSections.begin(), OutputSections.end(), compSec); for (unsigned I = 0, N = OutputSections.size(); I < N; ++I) OutputSections[I]->setSectionIndex(I + 1); } template static bool outputSectionHasPHDR(OutputSectionBase *Sec) { return Sec->getFlags() & SHF_ALLOC; } // Visits all sections to assign incremental, non-overlapping RVAs and // file offsets. template void Writer::assignAddresses() { assert(!OutputSections.empty() && "No output sections to layout!"); uintX_t VA = getVAStart(); uintX_t FileOff = 0; FileOff += sizeof(Elf_Ehdr); VA += sizeof(Elf_Ehdr); // Reserve space for PHDRs. ProgramHeaderOff = FileOff; FileOff = RoundUpToAlignment(FileOff, PageSize); VA = RoundUpToAlignment(VA, PageSize); if (needsInterpSection()) PHDRs.push_back(&InterpPHDR); ProgramHeader *LastPHDR = &FileHeaderPHDR; // Create a PHDR for the file header. PHDRs.push_back(&FileHeaderPHDR); FileHeaderPHDR.Header.p_vaddr = getVAStart(); FileHeaderPHDR.Header.p_paddr = getVAStart(); FileHeaderPHDR.Header.p_align = PageSize; for (OutputSectionBase *Sec : OutputSections) { StrTabSec.add(Sec->getName()); Sec->finalize(); if (Sec->getSize()) { uintX_t Flags = convertSectionFlagsToPHDRFlags(Sec->getFlags()); if (LastPHDR->Header.p_flags != Flags || !outputSectionHasPHDR(Sec)) { // Flags changed. End current PHDR and potentially create a new one. if (!LastPHDR->Closed) { LastPHDR->Header.p_filesz = FileOff - LastPHDR->Header.p_offset; LastPHDR->Header.p_memsz = VA - LastPHDR->Header.p_vaddr; LastPHDR->Closed = true; } if (outputSectionHasPHDR(Sec)) { LastPHDR = new (PAlloc) ProgramHeader(PT_LOAD, Flags); PHDRs.push_back(LastPHDR); VA = RoundUpToAlignment(VA, PageSize); FileOff = RoundUpToAlignment(FileOff, PageSize); LastPHDR->Header.p_offset = FileOff; LastPHDR->Header.p_vaddr = VA; LastPHDR->Header.p_paddr = VA; } } } uintX_t Align = Sec->getAlign(); uintX_t Size = Sec->getSize(); if (Sec->getFlags() & SHF_ALLOC) { VA = RoundUpToAlignment(VA, Align); Sec->setVA(VA); VA += Size; } FileOff = RoundUpToAlignment(FileOff, Align); Sec->setFileOffset(FileOff); if (Sec->getType() != SHT_NOBITS) FileOff += Size; } // Add a PHDR for the dynamic table. if (needsDynamicSections()) PHDRs.push_back(&DynamicPHDR); FileOff += OffsetToAlignment(FileOff, ELFT::Is64Bits ? 8 : 4); // Add space for section headers. SectionHeaderOff = FileOff; FileOff += getNumSections() * sizeof(Elf_Shdr); FileSize = FileOff; } template void Writer::writeHeader() { uint8_t *Buf = Buffer->getBufferStart(); auto *EHdr = reinterpret_cast(Buf); EHdr->e_ident[EI_MAG0] = 0x7F; EHdr->e_ident[EI_MAG1] = 0x45; EHdr->e_ident[EI_MAG2] = 0x4C; EHdr->e_ident[EI_MAG3] = 0x46; EHdr->e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32; EHdr->e_ident[EI_DATA] = ELFT::TargetEndianness == llvm::support::little ? ELFDATA2LSB : ELFDATA2MSB; EHdr->e_ident[EI_VERSION] = EV_CURRENT; EHdr->e_ident[EI_OSABI] = ELFOSABI_NONE; // FIXME: Generalize the segment construction similar to how we create // output sections. const SymbolTable &Symtab = SymTabSec.getSymTable(); EHdr->e_type = Config->Shared ? ET_DYN : ET_EXEC; auto &FirstObj = cast>(*Symtab.getFirstELF()); EHdr->e_machine = FirstObj.getEMachine(); EHdr->e_version = EV_CURRENT; SymbolBody *Entry = Symtab.getEntrySym(); EHdr->e_entry = Entry ? getSymVA(cast>(Entry)) : 0; EHdr->e_phoff = ProgramHeaderOff; EHdr->e_shoff = SectionHeaderOff; EHdr->e_ehsize = sizeof(Elf_Ehdr); EHdr->e_phentsize = sizeof(Elf_Phdr); EHdr->e_phnum = PHDRs.size(); EHdr->e_shentsize = sizeof(Elf_Shdr); EHdr->e_shnum = getNumSections(); EHdr->e_shstrndx = StrTabSec.getSectionIndex(); // If nothing was merged into the file header PT_LOAD, set the size correctly. if (FileHeaderPHDR.Header.p_filesz == PageSize) FileHeaderPHDR.Header.p_filesz = FileHeaderPHDR.Header.p_memsz = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * PHDRs.size(); if (needsInterpSection()) InterpPHDR.setValuesFromSection(InterpSec); if (needsDynamicSections()) DynamicPHDR.setValuesFromSection(DynamicSec); auto PHdrs = reinterpret_cast(Buf + EHdr->e_phoff); for (ProgramHeader *PHDR : PHDRs) PHDR->template writeHeaderTo(PHdrs++); auto SHdrs = reinterpret_cast(Buf + EHdr->e_shoff); // First entry is null. ++SHdrs; for (OutputSectionBase *Sec : OutputSections) { Sec->setNameOffset(StrTabSec.getFileOff(Sec->getName())); Sec->template writeHeaderTo(SHdrs++); } } template void Writer::openFile(StringRef Path) { ErrorOr> BufferOrErr = FileOutputBuffer::create(Path, FileSize, FileOutputBuffer::F_executable); error(BufferOrErr, Twine("failed to open ") + Path); Buffer = std::move(*BufferOrErr); } // Write section contents to a mmap'ed file. template void Writer::writeSections() { uint8_t *Buf = Buffer->getBufferStart(); for (OutputSectionBase *Sec : OutputSections) Sec->writeTo(Buf + Sec->getFileOff()); }