1 //===- yaml2elf - Convert YAML to a ELF object file -----------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
11 /// \brief The ELF component of yaml2obj.
13 //===----------------------------------------------------------------------===//
16 #include "llvm/ADT/ArrayRef.h"
17 #include "llvm/MC/StringTableBuilder.h"
18 #include "llvm/Object/ELFObjectFile.h"
19 #include "llvm/Object/ELFYAML.h"
20 #include "llvm/Support/ELF.h"
21 #include "llvm/Support/MemoryBuffer.h"
22 #include "llvm/Support/YAMLTraits.h"
23 #include "llvm/Support/raw_ostream.h"
27 // This class is used to build up a contiguous binary blob while keeping
28 // track of an offset in the output (which notionally begins at
31 class ContiguousBlobAccumulator {
32 const uint64_t InitialOffset;
33 SmallVector<char, 128> Buf;
34 raw_svector_ostream OS;
36 /// \returns The new offset.
37 uint64_t padToAlignment(unsigned Align) {
38 uint64_t CurrentOffset = InitialOffset + OS.tell();
39 uint64_t AlignedOffset = RoundUpToAlignment(CurrentOffset, Align);
40 for (; CurrentOffset != AlignedOffset; ++CurrentOffset)
42 return AlignedOffset; // == CurrentOffset;
46 ContiguousBlobAccumulator(uint64_t InitialOffset_)
47 : InitialOffset(InitialOffset_), Buf(), OS(Buf) {}
48 template <class Integer>
49 raw_ostream &getOSAndAlignedOffset(Integer &Offset, unsigned Align = 16) {
50 Offset = padToAlignment(Align);
53 void writeBlobToStream(raw_ostream &Out) { Out << OS.str(); }
55 } // end anonymous namespace
57 // Used to keep track of section and symbol names, so that in the YAML file
58 // sections and symbols can be referenced by name instead of by index.
63 /// \returns true if name is already present in the map.
64 bool addName(StringRef Name, unsigned i) {
65 return !Map.insert(std::make_pair(Name, (int)i)).second;
67 /// \returns true if name is not present in the map
68 bool lookup(StringRef Name, unsigned &Idx) const {
69 StringMap<int>::const_iterator I = Map.find(Name);
76 } // end anonymous namespace
79 static size_t arrayDataSize(ArrayRef<T> A) {
80 return A.size() * sizeof(T);
84 static void writeArrayData(raw_ostream &OS, ArrayRef<T> A) {
85 OS.write((const char *)A.data(), arrayDataSize(A));
89 static void zero(T &Obj) {
90 memset(&Obj, 0, sizeof(Obj));
94 /// \brief "Single point of truth" for the ELF file construction.
95 /// TODO: This class still has a ways to go before it is truly a "single
99 typedef typename object::ELFFile<ELFT>::Elf_Ehdr Elf_Ehdr;
100 typedef typename object::ELFFile<ELFT>::Elf_Shdr Elf_Shdr;
101 typedef typename object::ELFFile<ELFT>::Elf_Sym Elf_Sym;
102 typedef typename object::ELFFile<ELFT>::Elf_Rel Elf_Rel;
103 typedef typename object::ELFFile<ELFT>::Elf_Rela Elf_Rela;
105 /// \brief The future ".strtab" section.
106 StringTableBuilder DotStrtab;
108 /// \brief The future ".shstrtab" section.
109 StringTableBuilder DotShStrtab;
113 const ELFYAML::Object &Doc;
115 bool buildSectionIndex();
116 bool buildSymbolIndex(std::size_t &StartIndex,
117 const std::vector<ELFYAML::Symbol> &Symbols);
118 void initELFHeader(Elf_Ehdr &Header);
119 bool initSectionHeaders(std::vector<Elf_Shdr> &SHeaders,
120 ContiguousBlobAccumulator &CBA);
121 void initSymtabSectionHeader(Elf_Shdr &SHeader,
122 ContiguousBlobAccumulator &CBA);
123 void initStrtabSectionHeader(Elf_Shdr &SHeader, StringRef Name,
124 StringTableBuilder &STB,
125 ContiguousBlobAccumulator &CBA);
126 void addSymbols(const std::vector<ELFYAML::Symbol> &Symbols,
127 std::vector<Elf_Sym> &Syms, unsigned SymbolBinding);
128 void writeSectionContent(Elf_Shdr &SHeader,
129 const ELFYAML::RawContentSection &Section,
130 ContiguousBlobAccumulator &CBA);
131 bool writeSectionContent(Elf_Shdr &SHeader,
132 const ELFYAML::RelocationSection &Section,
133 ContiguousBlobAccumulator &CBA);
135 // - SHT_NULL entry (placed first, i.e. 0'th entry)
136 // - symbol table (.symtab) (placed third to last)
137 // - string table (.strtab) (placed second to last)
138 // - section header string table (.shstrtab) (placed last)
139 unsigned getDotSymTabSecNo() const { return Doc.Sections.size() + 1; }
140 unsigned getDotStrTabSecNo() const { return Doc.Sections.size() + 2; }
141 unsigned getDotShStrTabSecNo() const { return Doc.Sections.size() + 3; }
142 unsigned getSectionCount() const { return Doc.Sections.size() + 4; }
144 ELFState(const ELFYAML::Object &D) : Doc(D) {}
147 static int writeELF(raw_ostream &OS, const ELFYAML::Object &Doc);
149 } // end anonymous namespace
151 template <class ELFT>
152 void ELFState<ELFT>::initELFHeader(Elf_Ehdr &Header) {
153 using namespace llvm::ELF;
155 Header.e_ident[EI_MAG0] = 0x7f;
156 Header.e_ident[EI_MAG1] = 'E';
157 Header.e_ident[EI_MAG2] = 'L';
158 Header.e_ident[EI_MAG3] = 'F';
159 Header.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32;
160 bool IsLittleEndian = ELFT::TargetEndianness == support::little;
161 Header.e_ident[EI_DATA] = IsLittleEndian ? ELFDATA2LSB : ELFDATA2MSB;
162 Header.e_ident[EI_VERSION] = EV_CURRENT;
163 Header.e_ident[EI_OSABI] = Doc.Header.OSABI;
164 Header.e_ident[EI_ABIVERSION] = 0;
165 Header.e_type = Doc.Header.Type;
166 Header.e_machine = Doc.Header.Machine;
167 Header.e_version = EV_CURRENT;
168 Header.e_entry = Doc.Header.Entry;
169 Header.e_flags = Doc.Header.Flags;
170 Header.e_ehsize = sizeof(Elf_Ehdr);
171 Header.e_shentsize = sizeof(Elf_Shdr);
172 // Immediately following the ELF header.
173 Header.e_shoff = sizeof(Header);
174 Header.e_shnum = getSectionCount();
175 Header.e_shstrndx = getDotShStrTabSecNo();
178 template <class ELFT>
179 bool ELFState<ELFT>::initSectionHeaders(std::vector<Elf_Shdr> &SHeaders,
180 ContiguousBlobAccumulator &CBA) {
181 // Ensure SHN_UNDEF entry is present. An all-zero section header is a
182 // valid SHN_UNDEF entry since SHT_NULL == 0.
185 SHeaders.push_back(SHeader);
187 for (const auto &Sec : Doc.Sections)
188 DotShStrtab.add(Sec->Name);
189 DotShStrtab.finalize(StringTableBuilder::ELF);
191 for (const auto &Sec : Doc.Sections) {
193 SHeader.sh_name = DotShStrtab.getOffset(Sec->Name);
194 SHeader.sh_type = Sec->Type;
195 SHeader.sh_flags = Sec->Flags;
196 SHeader.sh_addr = Sec->Address;
197 SHeader.sh_addralign = Sec->AddressAlign;
199 if (!Sec->Link.empty()) {
201 if (SN2I.lookup(Sec->Link, Index)) {
202 errs() << "error: Unknown section referenced: '" << Sec->Link
203 << "' at YAML section '" << Sec->Name << "'.\n";
206 SHeader.sh_link = Index;
209 if (auto S = dyn_cast<ELFYAML::RawContentSection>(Sec.get()))
210 writeSectionContent(SHeader, *S, CBA);
211 else if (auto S = dyn_cast<ELFYAML::RelocationSection>(Sec.get())) {
213 // For relocation section set link to .symtab by default.
214 SHeader.sh_link = getDotSymTabSecNo();
217 if (SN2I.lookup(S->Info, Index)) {
218 errs() << "error: Unknown section referenced: '" << S->Info
219 << "' at YAML section '" << S->Name << "'.\n";
222 SHeader.sh_info = Index;
224 if (!writeSectionContent(SHeader, *S, CBA))
227 llvm_unreachable("Unknown section type");
229 SHeaders.push_back(SHeader);
234 template <class ELFT>
235 void ELFState<ELFT>::initSymtabSectionHeader(Elf_Shdr &SHeader,
236 ContiguousBlobAccumulator &CBA) {
238 SHeader.sh_name = DotShStrtab.getOffset(".symtab");
239 SHeader.sh_type = ELF::SHT_SYMTAB;
240 SHeader.sh_link = getDotStrTabSecNo();
241 // One greater than symbol table index of the last local symbol.
242 SHeader.sh_info = Doc.Symbols.Local.size() + 1;
243 SHeader.sh_entsize = sizeof(Elf_Sym);
245 std::vector<Elf_Sym> Syms;
247 // Ensure STN_UNDEF is present
253 // Add symbol names to .strtab.
254 for (const auto &Sym : Doc.Symbols.Local)
255 DotStrtab.add(Sym.Name);
256 for (const auto &Sym : Doc.Symbols.Global)
257 DotStrtab.add(Sym.Name);
258 for (const auto &Sym : Doc.Symbols.Weak)
259 DotStrtab.add(Sym.Name);
260 DotStrtab.finalize(StringTableBuilder::ELF);
262 addSymbols(Doc.Symbols.Local, Syms, ELF::STB_LOCAL);
263 addSymbols(Doc.Symbols.Global, Syms, ELF::STB_GLOBAL);
264 addSymbols(Doc.Symbols.Weak, Syms, ELF::STB_WEAK);
266 writeArrayData(CBA.getOSAndAlignedOffset(SHeader.sh_offset),
268 SHeader.sh_size = arrayDataSize(makeArrayRef(Syms));
271 template <class ELFT>
272 void ELFState<ELFT>::initStrtabSectionHeader(Elf_Shdr &SHeader, StringRef Name,
273 StringTableBuilder &STB,
274 ContiguousBlobAccumulator &CBA) {
276 SHeader.sh_name = DotShStrtab.getOffset(Name);
277 SHeader.sh_type = ELF::SHT_STRTAB;
278 CBA.getOSAndAlignedOffset(SHeader.sh_offset) << STB.data();
279 SHeader.sh_size = STB.data().size();
280 SHeader.sh_addralign = 1;
283 template <class ELFT>
284 void ELFState<ELFT>::addSymbols(const std::vector<ELFYAML::Symbol> &Symbols,
285 std::vector<Elf_Sym> &Syms,
286 unsigned SymbolBinding) {
287 for (const auto &Sym : Symbols) {
290 if (!Sym.Name.empty())
291 Symbol.st_name = DotStrtab.getOffset(Sym.Name);
292 Symbol.setBindingAndType(SymbolBinding, Sym.Type);
293 if (!Sym.Section.empty()) {
295 if (SN2I.lookup(Sym.Section, Index)) {
296 errs() << "error: Unknown section referenced: '" << Sym.Section
297 << "' by YAML symbol " << Sym.Name << ".\n";
300 Symbol.st_shndx = Index;
301 } // else Symbol.st_shndex == SHN_UNDEF (== 0), since it was zero'd earlier.
302 Symbol.st_value = Sym.Value;
303 Symbol.st_other = Sym.Other;
304 Symbol.st_size = Sym.Size;
305 Syms.push_back(Symbol);
309 template <class ELFT>
311 ELFState<ELFT>::writeSectionContent(Elf_Shdr &SHeader,
312 const ELFYAML::RawContentSection &Section,
313 ContiguousBlobAccumulator &CBA) {
314 assert(Section.Size >= Section.Content.binary_size() &&
315 "Section size and section content are inconsistent");
316 raw_ostream &OS = CBA.getOSAndAlignedOffset(SHeader.sh_offset);
317 Section.Content.writeAsBinary(OS);
318 for (auto i = Section.Content.binary_size(); i < Section.Size; ++i)
320 SHeader.sh_entsize = 0;
321 SHeader.sh_size = Section.Size;
324 static bool isMips64EL(const ELFYAML::Object &Doc) {
325 return Doc.Header.Machine == ELFYAML::ELF_EM(llvm::ELF::EM_MIPS) &&
326 Doc.Header.Class == ELFYAML::ELF_ELFCLASS(ELF::ELFCLASS64) &&
327 Doc.Header.Data == ELFYAML::ELF_ELFDATA(ELF::ELFDATA2LSB);
330 template <class ELFT>
332 ELFState<ELFT>::writeSectionContent(Elf_Shdr &SHeader,
333 const ELFYAML::RelocationSection &Section,
334 ContiguousBlobAccumulator &CBA) {
335 if (Section.Type != llvm::ELF::SHT_REL &&
336 Section.Type != llvm::ELF::SHT_RELA) {
337 errs() << "error: Invalid relocation section type.\n";
341 bool IsRela = Section.Type == llvm::ELF::SHT_RELA;
342 SHeader.sh_entsize = IsRela ? sizeof(Elf_Rela) : sizeof(Elf_Rel);
343 SHeader.sh_size = SHeader.sh_entsize * Section.Relocations.size();
345 auto &OS = CBA.getOSAndAlignedOffset(SHeader.sh_offset);
347 for (const auto &Rel : Section.Relocations) {
349 if (SymN2I.lookup(Rel.Symbol, SymIdx)) {
350 errs() << "error: Unknown symbol referenced: '" << Rel.Symbol
351 << "' at YAML relocation.\n";
358 REntry.r_offset = Rel.Offset;
359 REntry.r_addend = Rel.Addend;
360 REntry.setSymbolAndType(SymIdx, Rel.Type, isMips64EL(Doc));
361 OS.write((const char *)&REntry, sizeof(REntry));
365 REntry.r_offset = Rel.Offset;
366 REntry.setSymbolAndType(SymIdx, Rel.Type, isMips64EL(Doc));
367 OS.write((const char *)&REntry, sizeof(REntry));
373 template <class ELFT> bool ELFState<ELFT>::buildSectionIndex() {
374 SN2I.addName(".symtab", getDotSymTabSecNo());
375 SN2I.addName(".strtab", getDotStrTabSecNo());
376 SN2I.addName(".shstrtab", getDotShStrTabSecNo());
378 for (unsigned i = 0, e = Doc.Sections.size(); i != e; ++i) {
379 StringRef Name = Doc.Sections[i]->Name;
382 // "+ 1" to take into account the SHT_NULL entry.
383 if (SN2I.addName(Name, i + 1)) {
384 errs() << "error: Repeated section name: '" << Name
385 << "' at YAML section number " << i << ".\n";
392 template <class ELFT>
394 ELFState<ELFT>::buildSymbolIndex(std::size_t &StartIndex,
395 const std::vector<ELFYAML::Symbol> &Symbols) {
396 for (const auto &Sym : Symbols) {
398 if (Sym.Name.empty())
400 if (SymN2I.addName(Sym.Name, StartIndex)) {
401 errs() << "error: Repeated symbol name: '" << Sym.Name << "'.\n";
408 template <class ELFT>
409 int ELFState<ELFT>::writeELF(raw_ostream &OS, const ELFYAML::Object &Doc) {
410 ELFState<ELFT> State(Doc);
411 if (!State.buildSectionIndex())
414 std::size_t StartSymIndex = 0;
415 if (!State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Local) ||
416 !State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Global) ||
417 !State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Weak))
421 State.initELFHeader(Header);
423 // TODO: Flesh out section header support.
424 // TODO: Program headers.
426 // XXX: This offset is tightly coupled with the order that we write
428 const size_t SectionContentBeginOffset =
429 Header.e_ehsize + Header.e_shentsize * Header.e_shnum;
430 ContiguousBlobAccumulator CBA(SectionContentBeginOffset);
432 // Doc might not contain .symtab, .strtab and .shstrtab sections,
433 // but we will emit them, so make sure to add them to ShStrTabSHeader.
434 State.DotShStrtab.add(".symtab");
435 State.DotShStrtab.add(".strtab");
436 State.DotShStrtab.add(".shstrtab");
438 std::vector<Elf_Shdr> SHeaders;
439 if(!State.initSectionHeaders(SHeaders, CBA))
443 Elf_Shdr SymtabSHeader;
444 State.initSymtabSectionHeader(SymtabSHeader, CBA);
445 SHeaders.push_back(SymtabSHeader);
447 // .strtab string table header.
448 Elf_Shdr DotStrTabSHeader;
449 State.initStrtabSectionHeader(DotStrTabSHeader, ".strtab", State.DotStrtab,
451 SHeaders.push_back(DotStrTabSHeader);
453 // .shstrtab string table header.
454 Elf_Shdr ShStrTabSHeader;
455 State.initStrtabSectionHeader(ShStrTabSHeader, ".shstrtab", State.DotShStrtab,
457 SHeaders.push_back(ShStrTabSHeader);
459 OS.write((const char *)&Header, sizeof(Header));
460 writeArrayData(OS, makeArrayRef(SHeaders));
461 CBA.writeBlobToStream(OS);
465 static bool is64Bit(const ELFYAML::Object &Doc) {
466 return Doc.Header.Class == ELFYAML::ELF_ELFCLASS(ELF::ELFCLASS64);
469 static bool isLittleEndian(const ELFYAML::Object &Doc) {
470 return Doc.Header.Data == ELFYAML::ELF_ELFDATA(ELF::ELFDATA2LSB);
473 int yaml2elf(yaml::Input &YIn, raw_ostream &Out) {
477 errs() << "yaml2obj: Failed to parse YAML file!\n";
480 using object::ELFType;
481 typedef ELFType<support::little, 8, true> LE64;
482 typedef ELFType<support::big, 8, true> BE64;
483 typedef ELFType<support::little, 4, false> LE32;
484 typedef ELFType<support::big, 4, false> BE32;
486 if (isLittleEndian(Doc))
487 return ELFState<LE64>::writeELF(Out, Doc);
489 return ELFState<BE64>::writeELF(Out, Doc);
491 if (isLittleEndian(Doc))
492 return ELFState<LE32>::writeELF(Out, Doc);
494 return ELFState<BE32>::writeELF(Out, Doc);