1//===- yaml2elf - Convert YAML to a ELF object file -----------------------===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9///
10/// \file
11/// \brief The ELF component of yaml2obj.
12///
13//===----------------------------------------------------------------------===//
14
15#include "yaml2obj.h"
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"
24
25using namespace llvm;
26
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
29// `InitialOffset`).
30namespace {
31class ContiguousBlobAccumulator {
32  const uint64_t InitialOffset;
33  SmallVector<char, 128> Buf;
34  raw_svector_ostream OS;
35
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)
41      OS.write('\0');
42    return AlignedOffset; // == CurrentOffset;
43  }
44
45public:
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);
51    return OS;
52  }
53  void writeBlobToStream(raw_ostream &Out) { Out << OS.str(); }
54};
55} // end anonymous namespace
56
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.
59namespace {
60class NameToIdxMap {
61  StringMap<int> Map;
62public:
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;
66  }
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);
70    if (I == Map.end())
71      return true;
72    Idx = I->getValue();
73    return false;
74  }
75};
76} // end anonymous namespace
77
78template <class T>
79static size_t arrayDataSize(ArrayRef<T> A) {
80  return A.size() * sizeof(T);
81}
82
83template <class T>
84static void writeArrayData(raw_ostream &OS, ArrayRef<T> A) {
85  OS.write((const char *)A.data(), arrayDataSize(A));
86}
87
88template <class T>
89static void zero(T &Obj) {
90  memset(&Obj, 0, sizeof(Obj));
91}
92
93namespace {
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
96/// point of truth".
97template <class ELFT>
98class ELFState {
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;
104
105  /// \brief The future ".strtab" section.
106  StringTableBuilder DotStrtab;
107
108  /// \brief The future ".shstrtab" section.
109  StringTableBuilder DotShStrtab;
110
111  NameToIdxMap SN2I;
112  NameToIdxMap SymN2I;
113  const ELFYAML::Object &Doc;
114
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);
134  bool writeSectionContent(Elf_Shdr &SHeader, const ELFYAML::Group &Group,
135                           ContiguousBlobAccumulator &CBA);
136
137  // - SHT_NULL entry (placed first, i.e. 0'th entry)
138  // - symbol table (.symtab) (placed third to last)
139  // - string table (.strtab) (placed second to last)
140  // - section header string table (.shstrtab) (placed last)
141  unsigned getDotSymTabSecNo() const { return Doc.Sections.size() + 1; }
142  unsigned getDotStrTabSecNo() const { return Doc.Sections.size() + 2; }
143  unsigned getDotShStrTabSecNo() const { return Doc.Sections.size() + 3; }
144  unsigned getSectionCount() const { return Doc.Sections.size() + 4; }
145
146  ELFState(const ELFYAML::Object &D) : Doc(D) {}
147
148public:
149  static int writeELF(raw_ostream &OS, const ELFYAML::Object &Doc);
150};
151} // end anonymous namespace
152
153template <class ELFT>
154void ELFState<ELFT>::initELFHeader(Elf_Ehdr &Header) {
155  using namespace llvm::ELF;
156  zero(Header);
157  Header.e_ident[EI_MAG0] = 0x7f;
158  Header.e_ident[EI_MAG1] = 'E';
159  Header.e_ident[EI_MAG2] = 'L';
160  Header.e_ident[EI_MAG3] = 'F';
161  Header.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32;
162  bool IsLittleEndian = ELFT::TargetEndianness == support::little;
163  Header.e_ident[EI_DATA] = IsLittleEndian ? ELFDATA2LSB : ELFDATA2MSB;
164  Header.e_ident[EI_VERSION] = EV_CURRENT;
165  Header.e_ident[EI_OSABI] = Doc.Header.OSABI;
166  Header.e_ident[EI_ABIVERSION] = 0;
167  Header.e_type = Doc.Header.Type;
168  Header.e_machine = Doc.Header.Machine;
169  Header.e_version = EV_CURRENT;
170  Header.e_entry = Doc.Header.Entry;
171  Header.e_flags = Doc.Header.Flags;
172  Header.e_ehsize = sizeof(Elf_Ehdr);
173  Header.e_shentsize = sizeof(Elf_Shdr);
174  // Immediately following the ELF header.
175  Header.e_shoff = sizeof(Header);
176  Header.e_shnum = getSectionCount();
177  Header.e_shstrndx = getDotShStrTabSecNo();
178}
179
180template <class ELFT>
181bool ELFState<ELFT>::initSectionHeaders(std::vector<Elf_Shdr> &SHeaders,
182                                        ContiguousBlobAccumulator &CBA) {
183  // Ensure SHN_UNDEF entry is present. An all-zero section header is a
184  // valid SHN_UNDEF entry since SHT_NULL == 0.
185  Elf_Shdr SHeader;
186  zero(SHeader);
187  SHeaders.push_back(SHeader);
188
189  for (const auto &Sec : Doc.Sections)
190    DotShStrtab.add(Sec->Name);
191  DotShStrtab.finalize(StringTableBuilder::ELF);
192
193  for (const auto &Sec : Doc.Sections) {
194    zero(SHeader);
195    SHeader.sh_name = DotShStrtab.getOffset(Sec->Name);
196    SHeader.sh_type = Sec->Type;
197    SHeader.sh_flags = Sec->Flags;
198    SHeader.sh_addr = Sec->Address;
199    SHeader.sh_addralign = Sec->AddressAlign;
200
201    if (!Sec->Link.empty()) {
202      unsigned Index;
203      if (SN2I.lookup(Sec->Link, Index)) {
204        errs() << "error: Unknown section referenced: '" << Sec->Link
205               << "' at YAML section '" << Sec->Name << "'.\n";
206        return false;
207      }
208      SHeader.sh_link = Index;
209    }
210
211    if (auto S = dyn_cast<ELFYAML::RawContentSection>(Sec.get()))
212      writeSectionContent(SHeader, *S, CBA);
213    else if (auto S = dyn_cast<ELFYAML::RelocationSection>(Sec.get())) {
214      if (S->Link.empty())
215        // For relocation section set link to .symtab by default.
216        SHeader.sh_link = getDotSymTabSecNo();
217
218      unsigned Index;
219      if (SN2I.lookup(S->Info, Index)) {
220        errs() << "error: Unknown section referenced: '" << S->Info
221               << "' at YAML section '" << S->Name << "'.\n";
222        return false;
223      }
224      SHeader.sh_info = Index;
225
226      if (!writeSectionContent(SHeader, *S, CBA))
227        return false;
228    } else if (auto S = dyn_cast<ELFYAML::Group>(Sec.get())) {
229      unsigned SymIdx;
230      if (SymN2I.lookup(S->Info, SymIdx)) {
231        errs() << "error: Unknown symbol referenced: '" << S->Info
232               << "' at YAML section '" << S->Name << "'.\n";
233        return false;
234      }
235      SHeader.sh_info = SymIdx;
236      if (!writeSectionContent(SHeader, *S, CBA))
237        return false;
238    } else
239      llvm_unreachable("Unknown section type");
240
241    SHeaders.push_back(SHeader);
242  }
243  return true;
244}
245
246template <class ELFT>
247void ELFState<ELFT>::initSymtabSectionHeader(Elf_Shdr &SHeader,
248                                             ContiguousBlobAccumulator &CBA) {
249  zero(SHeader);
250  SHeader.sh_name = DotShStrtab.getOffset(".symtab");
251  SHeader.sh_type = ELF::SHT_SYMTAB;
252  SHeader.sh_link = getDotStrTabSecNo();
253  // One greater than symbol table index of the last local symbol.
254  SHeader.sh_info = Doc.Symbols.Local.size() + 1;
255  SHeader.sh_entsize = sizeof(Elf_Sym);
256
257  std::vector<Elf_Sym> Syms;
258  {
259    // Ensure STN_UNDEF is present
260    Elf_Sym Sym;
261    zero(Sym);
262    Syms.push_back(Sym);
263  }
264
265  // Add symbol names to .strtab.
266  for (const auto &Sym : Doc.Symbols.Local)
267    DotStrtab.add(Sym.Name);
268  for (const auto &Sym : Doc.Symbols.Global)
269    DotStrtab.add(Sym.Name);
270  for (const auto &Sym : Doc.Symbols.Weak)
271    DotStrtab.add(Sym.Name);
272  DotStrtab.finalize(StringTableBuilder::ELF);
273
274  addSymbols(Doc.Symbols.Local, Syms, ELF::STB_LOCAL);
275  addSymbols(Doc.Symbols.Global, Syms, ELF::STB_GLOBAL);
276  addSymbols(Doc.Symbols.Weak, Syms, ELF::STB_WEAK);
277
278  writeArrayData(CBA.getOSAndAlignedOffset(SHeader.sh_offset),
279                 makeArrayRef(Syms));
280  SHeader.sh_size = arrayDataSize(makeArrayRef(Syms));
281}
282
283template <class ELFT>
284void ELFState<ELFT>::initStrtabSectionHeader(Elf_Shdr &SHeader, StringRef Name,
285                                             StringTableBuilder &STB,
286                                             ContiguousBlobAccumulator &CBA) {
287  zero(SHeader);
288  SHeader.sh_name = DotShStrtab.getOffset(Name);
289  SHeader.sh_type = ELF::SHT_STRTAB;
290  CBA.getOSAndAlignedOffset(SHeader.sh_offset) << STB.data();
291  SHeader.sh_size = STB.data().size();
292  SHeader.sh_addralign = 1;
293}
294
295template <class ELFT>
296void ELFState<ELFT>::addSymbols(const std::vector<ELFYAML::Symbol> &Symbols,
297                                std::vector<Elf_Sym> &Syms,
298                                unsigned SymbolBinding) {
299  for (const auto &Sym : Symbols) {
300    Elf_Sym Symbol;
301    zero(Symbol);
302    if (!Sym.Name.empty())
303      Symbol.st_name = DotStrtab.getOffset(Sym.Name);
304    Symbol.setBindingAndType(SymbolBinding, Sym.Type);
305    if (!Sym.Section.empty()) {
306      unsigned Index;
307      if (SN2I.lookup(Sym.Section, Index)) {
308        errs() << "error: Unknown section referenced: '" << Sym.Section
309               << "' by YAML symbol " << Sym.Name << ".\n";
310        exit(1);
311      }
312      Symbol.st_shndx = Index;
313    } // else Symbol.st_shndex == SHN_UNDEF (== 0), since it was zero'd earlier.
314    Symbol.st_value = Sym.Value;
315    Symbol.st_other = Sym.Other;
316    Symbol.st_size = Sym.Size;
317    Syms.push_back(Symbol);
318  }
319}
320
321template <class ELFT>
322void
323ELFState<ELFT>::writeSectionContent(Elf_Shdr &SHeader,
324                                    const ELFYAML::RawContentSection &Section,
325                                    ContiguousBlobAccumulator &CBA) {
326  assert(Section.Size >= Section.Content.binary_size() &&
327         "Section size and section content are inconsistent");
328  raw_ostream &OS = CBA.getOSAndAlignedOffset(SHeader.sh_offset);
329  Section.Content.writeAsBinary(OS);
330  for (auto i = Section.Content.binary_size(); i < Section.Size; ++i)
331    OS.write(0);
332  SHeader.sh_entsize = 0;
333  SHeader.sh_size = Section.Size;
334}
335
336static bool isMips64EL(const ELFYAML::Object &Doc) {
337  return Doc.Header.Machine == ELFYAML::ELF_EM(llvm::ELF::EM_MIPS) &&
338         Doc.Header.Class == ELFYAML::ELF_ELFCLASS(ELF::ELFCLASS64) &&
339         Doc.Header.Data == ELFYAML::ELF_ELFDATA(ELF::ELFDATA2LSB);
340}
341
342template <class ELFT>
343bool
344ELFState<ELFT>::writeSectionContent(Elf_Shdr &SHeader,
345                                    const ELFYAML::RelocationSection &Section,
346                                    ContiguousBlobAccumulator &CBA) {
347  if (Section.Type != llvm::ELF::SHT_REL &&
348      Section.Type != llvm::ELF::SHT_RELA) {
349    errs() << "error: Invalid relocation section type.\n";
350    return false;
351  }
352
353  bool IsRela = Section.Type == llvm::ELF::SHT_RELA;
354  SHeader.sh_entsize = IsRela ? sizeof(Elf_Rela) : sizeof(Elf_Rel);
355  SHeader.sh_size = SHeader.sh_entsize * Section.Relocations.size();
356
357  auto &OS = CBA.getOSAndAlignedOffset(SHeader.sh_offset);
358
359  for (const auto &Rel : Section.Relocations) {
360    unsigned SymIdx;
361    if (SymN2I.lookup(Rel.Symbol, SymIdx)) {
362      errs() << "error: Unknown symbol referenced: '" << Rel.Symbol
363             << "' at YAML relocation.\n";
364      return false;
365    }
366
367    if (IsRela) {
368      Elf_Rela REntry;
369      zero(REntry);
370      REntry.r_offset = Rel.Offset;
371      REntry.r_addend = Rel.Addend;
372      REntry.setSymbolAndType(SymIdx, Rel.Type, isMips64EL(Doc));
373      OS.write((const char *)&REntry, sizeof(REntry));
374    } else {
375      Elf_Rel REntry;
376      zero(REntry);
377      REntry.r_offset = Rel.Offset;
378      REntry.setSymbolAndType(SymIdx, Rel.Type, isMips64EL(Doc));
379      OS.write((const char *)&REntry, sizeof(REntry));
380    }
381  }
382  return true;
383}
384
385template <class ELFT>
386bool ELFState<ELFT>::writeSectionContent(Elf_Shdr &SHeader,
387                                         const ELFYAML::Group &Section,
388                                         ContiguousBlobAccumulator &CBA) {
389  typedef typename object::ELFFile<ELFT>::Elf_Word Elf_Word;
390  if (Section.Type != llvm::ELF::SHT_GROUP) {
391    errs() << "error: Invalid section type.\n";
392    return false;
393  }
394
395  SHeader.sh_entsize = sizeof(Elf_Word);
396  SHeader.sh_size = SHeader.sh_entsize * Section.Members.size();
397
398  auto &OS = CBA.getOSAndAlignedOffset(SHeader.sh_offset);
399
400  for (auto member : Section.Members) {
401    Elf_Word SIdx;
402    unsigned int sectionIndex = 0;
403    if (member.sectionNameOrType == "GRP_COMDAT")
404      sectionIndex = llvm::ELF::GRP_COMDAT;
405    else if (SN2I.lookup(member.sectionNameOrType, sectionIndex)) {
406      errs() << "error: Unknown section referenced: '"
407             << member.sectionNameOrType << "' at YAML section' "
408             << Section.Name << "\n";
409      return false;
410    }
411    SIdx = sectionIndex;
412    OS.write((const char *)&SIdx, sizeof(SIdx));
413  }
414  return true;
415}
416
417template <class ELFT> bool ELFState<ELFT>::buildSectionIndex() {
418  SN2I.addName(".symtab", getDotSymTabSecNo());
419  SN2I.addName(".strtab", getDotStrTabSecNo());
420  SN2I.addName(".shstrtab", getDotShStrTabSecNo());
421
422  for (unsigned i = 0, e = Doc.Sections.size(); i != e; ++i) {
423    StringRef Name = Doc.Sections[i]->Name;
424    if (Name.empty())
425      continue;
426    // "+ 1" to take into account the SHT_NULL entry.
427    if (SN2I.addName(Name, i + 1)) {
428      errs() << "error: Repeated section name: '" << Name
429             << "' at YAML section number " << i << ".\n";
430      return false;
431    }
432  }
433  return true;
434}
435
436template <class ELFT>
437bool
438ELFState<ELFT>::buildSymbolIndex(std::size_t &StartIndex,
439                                 const std::vector<ELFYAML::Symbol> &Symbols) {
440  for (const auto &Sym : Symbols) {
441    ++StartIndex;
442    if (Sym.Name.empty())
443      continue;
444    if (SymN2I.addName(Sym.Name, StartIndex)) {
445      errs() << "error: Repeated symbol name: '" << Sym.Name << "'.\n";
446      return false;
447    }
448  }
449  return true;
450}
451
452template <class ELFT>
453int ELFState<ELFT>::writeELF(raw_ostream &OS, const ELFYAML::Object &Doc) {
454  ELFState<ELFT> State(Doc);
455  if (!State.buildSectionIndex())
456    return 1;
457
458  std::size_t StartSymIndex = 0;
459  if (!State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Local) ||
460      !State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Global) ||
461      !State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Weak))
462    return 1;
463
464  Elf_Ehdr Header;
465  State.initELFHeader(Header);
466
467  // TODO: Flesh out section header support.
468  // TODO: Program headers.
469
470  // XXX: This offset is tightly coupled with the order that we write
471  // things to `OS`.
472  const size_t SectionContentBeginOffset =
473      Header.e_ehsize + Header.e_shentsize * Header.e_shnum;
474  ContiguousBlobAccumulator CBA(SectionContentBeginOffset);
475
476  // Doc might not contain .symtab, .strtab and .shstrtab sections,
477  // but we will emit them, so make sure to add them to ShStrTabSHeader.
478  State.DotShStrtab.add(".symtab");
479  State.DotShStrtab.add(".strtab");
480  State.DotShStrtab.add(".shstrtab");
481
482  std::vector<Elf_Shdr> SHeaders;
483  if(!State.initSectionHeaders(SHeaders, CBA))
484    return 1;
485
486  // .symtab section.
487  Elf_Shdr SymtabSHeader;
488  State.initSymtabSectionHeader(SymtabSHeader, CBA);
489  SHeaders.push_back(SymtabSHeader);
490
491  // .strtab string table header.
492  Elf_Shdr DotStrTabSHeader;
493  State.initStrtabSectionHeader(DotStrTabSHeader, ".strtab", State.DotStrtab,
494                                CBA);
495  SHeaders.push_back(DotStrTabSHeader);
496
497  // .shstrtab string table header.
498  Elf_Shdr ShStrTabSHeader;
499  State.initStrtabSectionHeader(ShStrTabSHeader, ".shstrtab", State.DotShStrtab,
500                                CBA);
501  SHeaders.push_back(ShStrTabSHeader);
502
503  OS.write((const char *)&Header, sizeof(Header));
504  writeArrayData(OS, makeArrayRef(SHeaders));
505  CBA.writeBlobToStream(OS);
506  return 0;
507}
508
509static bool is64Bit(const ELFYAML::Object &Doc) {
510  return Doc.Header.Class == ELFYAML::ELF_ELFCLASS(ELF::ELFCLASS64);
511}
512
513static bool isLittleEndian(const ELFYAML::Object &Doc) {
514  return Doc.Header.Data == ELFYAML::ELF_ELFDATA(ELF::ELFDATA2LSB);
515}
516
517int yaml2elf(yaml::Input &YIn, raw_ostream &Out) {
518  ELFYAML::Object Doc;
519  YIn >> Doc;
520  if (YIn.error()) {
521    errs() << "yaml2obj: Failed to parse YAML file!\n";
522    return 1;
523  }
524  using object::ELFType;
525  typedef ELFType<support::little, 8, true> LE64;
526  typedef ELFType<support::big, 8, true> BE64;
527  typedef ELFType<support::little, 4, false> LE32;
528  typedef ELFType<support::big, 4, false> BE32;
529  if (is64Bit(Doc)) {
530    if (isLittleEndian(Doc))
531      return ELFState<LE64>::writeELF(Out, Doc);
532    else
533      return ELFState<BE64>::writeELF(Out, Doc);
534  } else {
535    if (isLittleEndian(Doc))
536      return ELFState<LE32>::writeELF(Out, Doc);
537    else
538      return ELFState<BE32>::writeELF(Out, Doc);
539  }
540}
541