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/Object/ELF.h"
17 #include "llvm/Object/ELFYAML.h"
18 #include "llvm/Support/ELF.h"
19 #include "llvm/Support/MemoryBuffer.h"
20 #include "llvm/Support/YAMLTraits.h"
21 #include "llvm/Support/raw_ostream.h"
25 // There is similar code in yaml2coff, but with some slight COFF-specific
26 // variations like different initial state. Might be able to deduplicate
27 // some day, but also want to make sure that the Mach-O use case is served.
29 // This class has a deliberately small interface, since a lot of
30 // implementation variation is possible.
32 // TODO: Use an ordered container with a suffix-based comparison in order
33 // to deduplicate suffixes. std::map<> with a custom comparator is likely
34 // to be the simplest implementation, but a suffix trie could be more
35 // suitable for the job.
37 class StringTableBuilder {
38 /// \brief Indices of strings currently present in `Buf`.
39 StringMap<unsigned> StringIndices;
40 /// \brief The contents of the string table as we build it.
43 StringTableBuilder() {
46 /// \returns Index of string in string table.
47 unsigned addString(StringRef S) {
48 StringMapEntry<unsigned> &Entry = StringIndices.GetOrCreateValue(S);
49 unsigned &I = Entry.getValue();
53 Buf.append(S.begin(), S.end());
60 void writeToStream(raw_ostream &OS) {
61 OS.write(Buf.data(), Buf.size());
64 } // end anonymous namespace
66 // This class is used to build up a contiguous binary blob while keeping
67 // track of an offset in the output (which notionally begins at
70 class ContiguousBlobAccumulator {
71 const uint64_t InitialOffset;
72 SmallVector<char, 128> Buf;
73 raw_svector_ostream OS;
76 ContiguousBlobAccumulator(uint64_t InitialOffset_)
77 : InitialOffset(InitialOffset_), Buf(), OS(Buf) {}
78 raw_ostream &getOS() { return OS; }
79 uint64_t currentOffset() const { return InitialOffset + OS.tell(); }
80 void writeBlobToStream(raw_ostream &Out) { Out << OS.str(); }
82 } // end anonymous namespace
84 // Used to keep track of section names, so that in the YAML file sections
85 // can be referenced by name instead of by index.
87 class SectionNameToIdxMap {
90 /// \returns true if name is already present in the map.
91 bool addName(StringRef SecName, unsigned i) {
92 StringMapEntry<int> &Entry = Map.GetOrCreateValue(SecName, -1);
93 if (Entry.getValue() != -1)
95 Entry.setValue((int)i);
98 /// \returns true if name is not present in the map
99 bool lookupSection(StringRef SecName, unsigned &Idx) const {
100 StringMap<int>::const_iterator I = Map.find(SecName);
107 } // end anonymous namespace
110 static size_t vectorDataSize(const std::vector<T> &Vec) {
111 return Vec.size() * sizeof(T);
115 static void writeVectorData(raw_ostream &OS, const std::vector<T> &Vec) {
116 OS.write((const char *)Vec.data(), vectorDataSize(Vec));
120 static void zero(T &Obj) {
121 memset(&Obj, 0, sizeof(Obj));
124 /// \brief Create a string table in `SHeader`, which we assume is already
126 template <class Elf_Shdr>
127 static void createStringTableSectionHeader(Elf_Shdr &SHeader,
128 StringTableBuilder &STB,
129 ContiguousBlobAccumulator &CBA) {
130 SHeader.sh_type = ELF::SHT_STRTAB;
131 SHeader.sh_offset = CBA.currentOffset();
132 SHeader.sh_size = STB.size();
133 STB.writeToStream(CBA.getOS());
134 SHeader.sh_addralign = 1;
138 /// \brief "Single point of truth" for the ELF file construction.
139 /// TODO: This class still has a ways to go before it is truly a "single
141 template <class ELFT>
143 /// \brief The future ".strtab" section.
144 StringTableBuilder DotStrtab;
145 /// \brief The section number of the ".strtab" section.
146 unsigned DotStrtabSecNo;
147 /// \brief The accumulated contents of all sections so far.
148 ContiguousBlobAccumulator &SectionContentAccum;
149 typedef typename object::ELFObjectFile<ELFT>::Elf_Ehdr Elf_Ehdr;
150 /// \brief The ELF file header.
153 SectionNameToIdxMap &SN2I;
157 ELFState(Elf_Ehdr &Header_, ContiguousBlobAccumulator &Accum,
158 unsigned DotStrtabSecNo_, SectionNameToIdxMap &SN2I_)
159 : DotStrtab(), DotStrtabSecNo(DotStrtabSecNo_),
160 SectionContentAccum(Accum), Header(Header_), SN2I(SN2I_) {}
162 unsigned getDotStrTabSecNo() const { return DotStrtabSecNo; }
163 StringTableBuilder &getStringTable() { return DotStrtab; }
164 ContiguousBlobAccumulator &getSectionContentAccum() {
165 return SectionContentAccum;
167 SectionNameToIdxMap &getSN2I() { return SN2I; }
169 } // end anonymous namespace
171 // FIXME: At this point it is fairly clear that we need to refactor these
172 // static functions into methods of a class sharing some typedefs. These
173 // ELF type names are insane.
174 template <class ELFT>
176 addSymbols(const std::vector<ELFYAML::Symbol> &Symbols, ELFState<ELFT> &State,
177 std::vector<typename object::ELFObjectFile<ELFT>::Elf_Sym> &Syms,
178 unsigned SymbolBinding) {
179 typedef typename object::ELFObjectFile<ELFT>::Elf_Sym Elf_Sym;
180 for (unsigned i = 0, e = Symbols.size(); i != e; ++i) {
181 const ELFYAML::Symbol &Sym = Symbols[i];
184 if (!Sym.Name.empty())
185 Symbol.st_name = State.getStringTable().addString(Sym.Name);
186 Symbol.setBindingAndType(SymbolBinding, Sym.Type);
187 if (!Sym.Section.empty()) {
189 if (State.getSN2I().lookupSection(Sym.Section, Index)) {
190 errs() << "error: Unknown section referenced: '" << Sym.Section
191 << "' by YAML symbol " << Sym.Name << ".\n";
194 Symbol.st_shndx = Index;
195 } // else Symbol.st_shndex == SHN_UNDEF (== 0), since it was zero'd earlier.
196 Symbol.st_value = Sym.Value;
197 Symbol.st_size = Sym.Size;
198 Syms.push_back(Symbol);
202 template <class ELFT>
203 static void handleSymtabSectionHeader(
204 const ELFYAML::Section &Sec, ELFState<ELFT> &State,
205 typename object::ELFObjectFile<ELFT>::Elf_Shdr &SHeader) {
207 typedef typename object::ELFObjectFile<ELFT>::Elf_Sym Elf_Sym;
208 // TODO: Ensure that a manually specified `Link` field is diagnosed as an
209 // error for SHT_SYMTAB.
210 SHeader.sh_link = State.getDotStrTabSecNo();
211 // One greater than symbol table index of the last local symbol.
212 SHeader.sh_info = Sec.Symbols.Local.size() + 1;
213 SHeader.sh_entsize = sizeof(Elf_Sym);
215 std::vector<Elf_Sym> Syms;
217 // Ensure STN_UNDEF is present
222 addSymbols(Sec.Symbols.Local, State, Syms, ELF::STB_LOCAL);
223 addSymbols(Sec.Symbols.Global, State, Syms, ELF::STB_GLOBAL);
224 addSymbols(Sec.Symbols.Weak, State, Syms, ELF::STB_WEAK);
226 ContiguousBlobAccumulator &CBA = State.getSectionContentAccum();
227 SHeader.sh_offset = CBA.currentOffset();
228 SHeader.sh_size = vectorDataSize(Syms);
229 writeVectorData(CBA.getOS(), Syms);
232 template <class ELFT>
233 static int writeELF(raw_ostream &OS, const ELFYAML::Object &Doc) {
234 using namespace llvm::ELF;
235 typedef typename object::ELFObjectFile<ELFT>::Elf_Ehdr Elf_Ehdr;
236 typedef typename object::ELFObjectFile<ELFT>::Elf_Shdr Elf_Shdr;
238 const ELFYAML::FileHeader &Hdr = Doc.Header;
242 Header.e_ident[EI_MAG0] = 0x7f;
243 Header.e_ident[EI_MAG1] = 'E';
244 Header.e_ident[EI_MAG2] = 'L';
245 Header.e_ident[EI_MAG3] = 'F';
246 Header.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32;
247 bool IsLittleEndian = ELFT::TargetEndianness == support::little;
248 Header.e_ident[EI_DATA] = IsLittleEndian ? ELFDATA2LSB : ELFDATA2MSB;
249 Header.e_ident[EI_VERSION] = EV_CURRENT;
250 Header.e_ident[EI_OSABI] = Hdr.OSABI;
251 Header.e_ident[EI_ABIVERSION] = 0;
252 Header.e_type = Hdr.Type;
253 Header.e_machine = Hdr.Machine;
254 Header.e_version = EV_CURRENT;
255 Header.e_entry = Hdr.Entry;
256 Header.e_ehsize = sizeof(Elf_Ehdr);
258 // TODO: Flesh out section header support.
259 // TODO: Program headers.
261 Header.e_shentsize = sizeof(Elf_Shdr);
262 // Immediately following the ELF header.
263 Header.e_shoff = sizeof(Header);
264 const std::vector<ELFYAML::Section> &Sections = Doc.Sections;
266 // - SHT_NULL entry (placed first, i.e. 0'th entry)
267 // - string table (.strtab) (placed second to last)
268 // - section header string table. (placed last)
269 Header.e_shnum = Sections.size() + 3;
270 // Place section header string table last.
271 Header.e_shstrndx = Header.e_shnum - 1;
272 const unsigned DotStrtabSecNo = Header.e_shnum - 2;
274 // XXX: This offset is tightly coupled with the order that we write
276 const size_t SectionContentBeginOffset =
277 Header.e_ehsize + Header.e_shentsize * Header.e_shnum;
278 ContiguousBlobAccumulator CBA(SectionContentBeginOffset);
279 SectionNameToIdxMap SN2I;
280 for (unsigned i = 0, e = Sections.size(); i != e; ++i) {
281 StringRef Name = Sections[i].Name;
284 // "+ 1" to take into account the SHT_NULL entry.
285 if (SN2I.addName(Name, i + 1)) {
286 errs() << "error: Repeated section name: '" << Name
287 << "' at YAML section number " << i << ".\n";
292 ELFState<ELFT> State(Header, CBA, DotStrtabSecNo, SN2I);
294 StringTableBuilder SHStrTab;
295 std::vector<Elf_Shdr> SHeaders;
297 // Ensure SHN_UNDEF entry is present. An all-zero section header is a
298 // valid SHN_UNDEF entry since SHT_NULL == 0.
301 SHeaders.push_back(SHdr);
303 for (unsigned i = 0, e = Sections.size(); i != e; ++i) {
304 const ELFYAML::Section &Sec = Sections[i];
307 SHeader.sh_name = SHStrTab.addString(Sec.Name);
308 SHeader.sh_type = Sec.Type;
309 SHeader.sh_flags = Sec.Flags;
310 SHeader.sh_addr = Sec.Address;
312 SHeader.sh_offset = CBA.currentOffset();
313 SHeader.sh_size = Sec.Content.binary_size();
314 Sec.Content.writeAsBinary(CBA.getOS());
316 if (!Sec.Link.empty()) {
318 if (SN2I.lookupSection(Sec.Link, Index)) {
319 errs() << "error: Unknown section referenced: '" << Sec.Link
320 << "' at YAML section number " << i << ".\n";
323 SHeader.sh_link = Index;
326 SHeader.sh_addralign = Sec.AddressAlign;
327 SHeader.sh_entsize = 0;
328 // XXX: Really ugly right now. Should not be writing to `CBA` above
329 // (and setting sh_offset and sh_size) when going through this branch
331 if (Sec.Type == ELFYAML::ELF_SHT(SHT_SYMTAB))
332 handleSymtabSectionHeader<ELFT>(Sec, State, SHeader);
333 SHeaders.push_back(SHeader);
336 // .strtab string table header.
337 Elf_Shdr DotStrTabSHeader;
338 zero(DotStrTabSHeader);
339 DotStrTabSHeader.sh_name = SHStrTab.addString(StringRef(".strtab"));
340 createStringTableSectionHeader(DotStrTabSHeader, State.getStringTable(), CBA);
342 // Section header string table header.
343 Elf_Shdr SHStrTabSHeader;
344 zero(SHStrTabSHeader);
345 createStringTableSectionHeader(SHStrTabSHeader, SHStrTab, CBA);
347 OS.write((const char *)&Header, sizeof(Header));
348 writeVectorData(OS, SHeaders);
349 OS.write((const char *)&DotStrTabSHeader, sizeof(DotStrTabSHeader));
350 OS.write((const char *)&SHStrTabSHeader, sizeof(SHStrTabSHeader));
351 CBA.writeBlobToStream(OS);
355 int yaml2elf(llvm::raw_ostream &Out, llvm::MemoryBuffer *Buf) {
356 yaml::Input YIn(Buf->getBuffer());
360 errs() << "yaml2obj: Failed to parse YAML file!\n";
363 if (Doc.Header.Class == ELFYAML::ELF_ELFCLASS(ELF::ELFCLASS64)) {
364 if (Doc.Header.Data == ELFYAML::ELF_ELFDATA(ELF::ELFDATA2LSB))
365 return writeELF<object::ELFType<support::little, 8, true> >(outs(), Doc);
367 return writeELF<object::ELFType<support::big, 8, true> >(outs(), Doc);
369 if (Doc.Header.Data == ELFYAML::ELF_ELFDATA(ELF::ELFDATA2LSB))
370 return writeELF<object::ELFType<support::little, 4, false> >(outs(), Doc);
372 return writeELF<object::ELFType<support::big, 4, false> >(outs(), Doc);