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
25 using 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`).
30 namespace {
31 class ContiguousBlobAccumulator {
32 const uint64_t InitialOffset;
33 SmallVector<char, 128> Buf;
34 raw_svector_ostream OS;
35
36 /// \returns The new offset.
padToAlignment(unsigned Align)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
45 public:
ContiguousBlobAccumulator(uint64_t InitialOffset_)46 ContiguousBlobAccumulator(uint64_t InitialOffset_)
47 : InitialOffset(InitialOffset_), Buf(), OS(Buf) {}
48 template <class Integer>
getOSAndAlignedOffset(Integer & Offset,unsigned Align=16)49 raw_ostream &getOSAndAlignedOffset(Integer &Offset, unsigned Align = 16) {
50 Offset = padToAlignment(Align);
51 return OS;
52 }
writeBlobToStream(raw_ostream & Out)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.
59 namespace {
60 class NameToIdxMap {
61 StringMap<int> Map;
62 public:
63 /// \returns true if name is already present in the map.
addName(StringRef Name,unsigned i)64 bool addName(StringRef Name, unsigned i) {
65 StringMapEntry<int> &Entry = Map.GetOrCreateValue(Name, -1);
66 if (Entry.getValue() != -1)
67 return true;
68 Entry.setValue((int)i);
69 return false;
70 }
71 /// \returns true if name is not present in the map
lookup(StringRef Name,unsigned & Idx) const72 bool lookup(StringRef Name, unsigned &Idx) const {
73 StringMap<int>::const_iterator I = Map.find(Name);
74 if (I == Map.end())
75 return true;
76 Idx = I->getValue();
77 return false;
78 }
79 };
80 } // end anonymous namespace
81
82 template <class T>
arrayDataSize(ArrayRef<T> A)83 static size_t arrayDataSize(ArrayRef<T> A) {
84 return A.size() * sizeof(T);
85 }
86
87 template <class T>
writeArrayData(raw_ostream & OS,ArrayRef<T> A)88 static void writeArrayData(raw_ostream &OS, ArrayRef<T> A) {
89 OS.write((const char *)A.data(), arrayDataSize(A));
90 }
91
92 template <class T>
zero(T & Obj)93 static void zero(T &Obj) {
94 memset(&Obj, 0, sizeof(Obj));
95 }
96
97 namespace {
98 /// \brief "Single point of truth" for the ELF file construction.
99 /// TODO: This class still has a ways to go before it is truly a "single
100 /// point of truth".
101 template <class ELFT>
102 class ELFState {
103 typedef typename object::ELFFile<ELFT>::Elf_Ehdr Elf_Ehdr;
104 typedef typename object::ELFFile<ELFT>::Elf_Shdr Elf_Shdr;
105 typedef typename object::ELFFile<ELFT>::Elf_Sym Elf_Sym;
106 typedef typename object::ELFFile<ELFT>::Elf_Rel Elf_Rel;
107 typedef typename object::ELFFile<ELFT>::Elf_Rela Elf_Rela;
108
109 /// \brief The future ".strtab" section.
110 StringTableBuilder DotStrtab;
111
112 /// \brief The future ".shstrtab" section.
113 StringTableBuilder DotShStrtab;
114
115 NameToIdxMap SN2I;
116 NameToIdxMap SymN2I;
117 const ELFYAML::Object &Doc;
118
119 bool buildSectionIndex();
120 bool buildSymbolIndex(std::size_t &StartIndex,
121 const std::vector<ELFYAML::Symbol> &Symbols);
122 void initELFHeader(Elf_Ehdr &Header);
123 bool initSectionHeaders(std::vector<Elf_Shdr> &SHeaders,
124 ContiguousBlobAccumulator &CBA);
125 void initSymtabSectionHeader(Elf_Shdr &SHeader,
126 ContiguousBlobAccumulator &CBA);
127 void initStrtabSectionHeader(Elf_Shdr &SHeader, StringRef Name,
128 StringTableBuilder &STB,
129 ContiguousBlobAccumulator &CBA);
130 void addSymbols(const std::vector<ELFYAML::Symbol> &Symbols,
131 std::vector<Elf_Sym> &Syms, unsigned SymbolBinding);
132 void writeSectionContent(Elf_Shdr &SHeader,
133 const ELFYAML::RawContentSection &Section,
134 ContiguousBlobAccumulator &CBA);
135 bool writeSectionContent(Elf_Shdr &SHeader,
136 const ELFYAML::RelocationSection &Section,
137 ContiguousBlobAccumulator &CBA);
138
139 // - SHT_NULL entry (placed first, i.e. 0'th entry)
140 // - symbol table (.symtab) (placed third to last)
141 // - string table (.strtab) (placed second to last)
142 // - section header string table (.shstrtab) (placed last)
getDotSymTabSecNo() const143 unsigned getDotSymTabSecNo() const { return Doc.Sections.size() + 1; }
getDotStrTabSecNo() const144 unsigned getDotStrTabSecNo() const { return Doc.Sections.size() + 2; }
getDotShStrTabSecNo() const145 unsigned getDotShStrTabSecNo() const { return Doc.Sections.size() + 3; }
getSectionCount() const146 unsigned getSectionCount() const { return Doc.Sections.size() + 4; }
147
ELFState(const ELFYAML::Object & D)148 ELFState(const ELFYAML::Object &D) : Doc(D) {}
149
150 public:
151 static int writeELF(raw_ostream &OS, const ELFYAML::Object &Doc);
152 };
153 } // end anonymous namespace
154
155 template <class ELFT>
initELFHeader(Elf_Ehdr & Header)156 void ELFState<ELFT>::initELFHeader(Elf_Ehdr &Header) {
157 using namespace llvm::ELF;
158 zero(Header);
159 Header.e_ident[EI_MAG0] = 0x7f;
160 Header.e_ident[EI_MAG1] = 'E';
161 Header.e_ident[EI_MAG2] = 'L';
162 Header.e_ident[EI_MAG3] = 'F';
163 Header.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32;
164 bool IsLittleEndian = ELFT::TargetEndianness == support::little;
165 Header.e_ident[EI_DATA] = IsLittleEndian ? ELFDATA2LSB : ELFDATA2MSB;
166 Header.e_ident[EI_VERSION] = EV_CURRENT;
167 Header.e_ident[EI_OSABI] = Doc.Header.OSABI;
168 Header.e_ident[EI_ABIVERSION] = 0;
169 Header.e_type = Doc.Header.Type;
170 Header.e_machine = Doc.Header.Machine;
171 Header.e_version = EV_CURRENT;
172 Header.e_entry = Doc.Header.Entry;
173 Header.e_flags = Doc.Header.Flags;
174 Header.e_ehsize = sizeof(Elf_Ehdr);
175 Header.e_shentsize = sizeof(Elf_Shdr);
176 // Immediately following the ELF header.
177 Header.e_shoff = sizeof(Header);
178 Header.e_shnum = getSectionCount();
179 Header.e_shstrndx = getDotShStrTabSecNo();
180 }
181
182 template <class ELFT>
initSectionHeaders(std::vector<Elf_Shdr> & SHeaders,ContiguousBlobAccumulator & CBA)183 bool ELFState<ELFT>::initSectionHeaders(std::vector<Elf_Shdr> &SHeaders,
184 ContiguousBlobAccumulator &CBA) {
185 // Ensure SHN_UNDEF entry is present. An all-zero section header is a
186 // valid SHN_UNDEF entry since SHT_NULL == 0.
187 Elf_Shdr SHeader;
188 zero(SHeader);
189 SHeaders.push_back(SHeader);
190
191 for (const auto &Sec : Doc.Sections)
192 DotShStrtab.add(Sec->Name);
193 DotShStrtab.finalize();
194
195 for (const auto &Sec : Doc.Sections) {
196 zero(SHeader);
197 SHeader.sh_name = DotShStrtab.getOffset(Sec->Name);
198 SHeader.sh_type = Sec->Type;
199 SHeader.sh_flags = Sec->Flags;
200 SHeader.sh_addr = Sec->Address;
201 SHeader.sh_addralign = Sec->AddressAlign;
202
203 if (!Sec->Link.empty()) {
204 unsigned Index;
205 if (SN2I.lookup(Sec->Link, Index)) {
206 errs() << "error: Unknown section referenced: '" << Sec->Link
207 << "' at YAML section '" << Sec->Name << "'.\n";
208 return false;
209 }
210 SHeader.sh_link = Index;
211 }
212
213 if (auto S = dyn_cast<ELFYAML::RawContentSection>(Sec.get()))
214 writeSectionContent(SHeader, *S, CBA);
215 else if (auto S = dyn_cast<ELFYAML::RelocationSection>(Sec.get())) {
216 if (S->Link.empty())
217 // For relocation section set link to .symtab by default.
218 SHeader.sh_link = getDotSymTabSecNo();
219
220 unsigned Index;
221 if (SN2I.lookup(S->Info, Index)) {
222 errs() << "error: Unknown section referenced: '" << S->Info
223 << "' at YAML section '" << S->Name << "'.\n";
224 return false;
225 }
226 SHeader.sh_info = Index;
227
228 if (!writeSectionContent(SHeader, *S, CBA))
229 return false;
230 } else
231 llvm_unreachable("Unknown section type");
232
233 SHeaders.push_back(SHeader);
234 }
235 return true;
236 }
237
238 template <class ELFT>
initSymtabSectionHeader(Elf_Shdr & SHeader,ContiguousBlobAccumulator & CBA)239 void ELFState<ELFT>::initSymtabSectionHeader(Elf_Shdr &SHeader,
240 ContiguousBlobAccumulator &CBA) {
241 zero(SHeader);
242 SHeader.sh_name = DotShStrtab.getOffset(".symtab");
243 SHeader.sh_type = ELF::SHT_SYMTAB;
244 SHeader.sh_link = getDotStrTabSecNo();
245 // One greater than symbol table index of the last local symbol.
246 SHeader.sh_info = Doc.Symbols.Local.size() + 1;
247 SHeader.sh_entsize = sizeof(Elf_Sym);
248
249 std::vector<Elf_Sym> Syms;
250 {
251 // Ensure STN_UNDEF is present
252 Elf_Sym Sym;
253 zero(Sym);
254 Syms.push_back(Sym);
255 }
256
257 // Add symbol names to .strtab.
258 for (const auto &Sym : Doc.Symbols.Local)
259 DotStrtab.add(Sym.Name);
260 for (const auto &Sym : Doc.Symbols.Global)
261 DotStrtab.add(Sym.Name);
262 for (const auto &Sym : Doc.Symbols.Weak)
263 DotStrtab.add(Sym.Name);
264 DotStrtab.finalize();
265
266 addSymbols(Doc.Symbols.Local, Syms, ELF::STB_LOCAL);
267 addSymbols(Doc.Symbols.Global, Syms, ELF::STB_GLOBAL);
268 addSymbols(Doc.Symbols.Weak, Syms, ELF::STB_WEAK);
269
270 writeArrayData(CBA.getOSAndAlignedOffset(SHeader.sh_offset),
271 makeArrayRef(Syms));
272 SHeader.sh_size = arrayDataSize(makeArrayRef(Syms));
273 }
274
275 template <class ELFT>
initStrtabSectionHeader(Elf_Shdr & SHeader,StringRef Name,StringTableBuilder & STB,ContiguousBlobAccumulator & CBA)276 void ELFState<ELFT>::initStrtabSectionHeader(Elf_Shdr &SHeader, StringRef Name,
277 StringTableBuilder &STB,
278 ContiguousBlobAccumulator &CBA) {
279 zero(SHeader);
280 SHeader.sh_name = DotShStrtab.getOffset(Name);
281 SHeader.sh_type = ELF::SHT_STRTAB;
282 CBA.getOSAndAlignedOffset(SHeader.sh_offset) << STB.data();
283 SHeader.sh_size = STB.data().size();
284 SHeader.sh_addralign = 1;
285 }
286
287 template <class ELFT>
addSymbols(const std::vector<ELFYAML::Symbol> & Symbols,std::vector<Elf_Sym> & Syms,unsigned SymbolBinding)288 void ELFState<ELFT>::addSymbols(const std::vector<ELFYAML::Symbol> &Symbols,
289 std::vector<Elf_Sym> &Syms,
290 unsigned SymbolBinding) {
291 for (const auto &Sym : Symbols) {
292 Elf_Sym Symbol;
293 zero(Symbol);
294 if (!Sym.Name.empty())
295 Symbol.st_name = DotStrtab.getOffset(Sym.Name);
296 Symbol.setBindingAndType(SymbolBinding, Sym.Type);
297 if (!Sym.Section.empty()) {
298 unsigned Index;
299 if (SN2I.lookup(Sym.Section, Index)) {
300 errs() << "error: Unknown section referenced: '" << Sym.Section
301 << "' by YAML symbol " << Sym.Name << ".\n";
302 exit(1);
303 }
304 Symbol.st_shndx = Index;
305 } // else Symbol.st_shndex == SHN_UNDEF (== 0), since it was zero'd earlier.
306 Symbol.st_value = Sym.Value;
307 Symbol.st_other = Sym.Visibility;
308 Symbol.st_size = Sym.Size;
309 Syms.push_back(Symbol);
310 }
311 }
312
313 template <class ELFT>
314 void
writeSectionContent(Elf_Shdr & SHeader,const ELFYAML::RawContentSection & Section,ContiguousBlobAccumulator & CBA)315 ELFState<ELFT>::writeSectionContent(Elf_Shdr &SHeader,
316 const ELFYAML::RawContentSection &Section,
317 ContiguousBlobAccumulator &CBA) {
318 assert(Section.Size >= Section.Content.binary_size() &&
319 "Section size and section content are inconsistent");
320 raw_ostream &OS = CBA.getOSAndAlignedOffset(SHeader.sh_offset);
321 Section.Content.writeAsBinary(OS);
322 for (auto i = Section.Content.binary_size(); i < Section.Size; ++i)
323 OS.write(0);
324 SHeader.sh_entsize = 0;
325 SHeader.sh_size = Section.Size;
326 }
327
328 template <class ELFT>
329 bool
writeSectionContent(Elf_Shdr & SHeader,const ELFYAML::RelocationSection & Section,ContiguousBlobAccumulator & CBA)330 ELFState<ELFT>::writeSectionContent(Elf_Shdr &SHeader,
331 const ELFYAML::RelocationSection &Section,
332 ContiguousBlobAccumulator &CBA) {
333 if (Section.Type != llvm::ELF::SHT_REL &&
334 Section.Type != llvm::ELF::SHT_RELA) {
335 errs() << "error: Invalid relocation section type.\n";
336 return false;
337 }
338
339 bool IsRela = Section.Type == llvm::ELF::SHT_RELA;
340 SHeader.sh_entsize = IsRela ? sizeof(Elf_Rela) : sizeof(Elf_Rel);
341 SHeader.sh_size = SHeader.sh_entsize * Section.Relocations.size();
342
343 auto &OS = CBA.getOSAndAlignedOffset(SHeader.sh_offset);
344
345 for (const auto &Rel : Section.Relocations) {
346 unsigned SymIdx;
347 if (SymN2I.lookup(Rel.Symbol, SymIdx)) {
348 errs() << "error: Unknown symbol referenced: '" << Rel.Symbol
349 << "' at YAML relocation.\n";
350 return false;
351 }
352
353 if (IsRela) {
354 Elf_Rela REntry;
355 zero(REntry);
356 REntry.r_offset = Rel.Offset;
357 REntry.r_addend = Rel.Addend;
358 REntry.setSymbolAndType(SymIdx, Rel.Type);
359 OS.write((const char *)&REntry, sizeof(REntry));
360 } else {
361 Elf_Rel REntry;
362 zero(REntry);
363 REntry.r_offset = Rel.Offset;
364 REntry.setSymbolAndType(SymIdx, Rel.Type);
365 OS.write((const char *)&REntry, sizeof(REntry));
366 }
367 }
368 return true;
369 }
370
buildSectionIndex()371 template <class ELFT> bool ELFState<ELFT>::buildSectionIndex() {
372 SN2I.addName(".symtab", getDotSymTabSecNo());
373 SN2I.addName(".strtab", getDotStrTabSecNo());
374 SN2I.addName(".shstrtab", getDotShStrTabSecNo());
375
376 for (unsigned i = 0, e = Doc.Sections.size(); i != e; ++i) {
377 StringRef Name = Doc.Sections[i]->Name;
378 if (Name.empty())
379 continue;
380 // "+ 1" to take into account the SHT_NULL entry.
381 if (SN2I.addName(Name, i + 1)) {
382 errs() << "error: Repeated section name: '" << Name
383 << "' at YAML section number " << i << ".\n";
384 return false;
385 }
386 }
387 return true;
388 }
389
390 template <class ELFT>
391 bool
buildSymbolIndex(std::size_t & StartIndex,const std::vector<ELFYAML::Symbol> & Symbols)392 ELFState<ELFT>::buildSymbolIndex(std::size_t &StartIndex,
393 const std::vector<ELFYAML::Symbol> &Symbols) {
394 for (const auto &Sym : Symbols) {
395 ++StartIndex;
396 if (Sym.Name.empty())
397 continue;
398 if (SymN2I.addName(Sym.Name, StartIndex)) {
399 errs() << "error: Repeated symbol name: '" << Sym.Name << "'.\n";
400 return false;
401 }
402 }
403 return true;
404 }
405
406 template <class ELFT>
writeELF(raw_ostream & OS,const ELFYAML::Object & Doc)407 int ELFState<ELFT>::writeELF(raw_ostream &OS, const ELFYAML::Object &Doc) {
408 ELFState<ELFT> State(Doc);
409 if (!State.buildSectionIndex())
410 return 1;
411
412 std::size_t StartSymIndex = 0;
413 if (!State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Local) ||
414 !State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Global) ||
415 !State.buildSymbolIndex(StartSymIndex, Doc.Symbols.Weak))
416 return 1;
417
418 Elf_Ehdr Header;
419 State.initELFHeader(Header);
420
421 // TODO: Flesh out section header support.
422 // TODO: Program headers.
423
424 // XXX: This offset is tightly coupled with the order that we write
425 // things to `OS`.
426 const size_t SectionContentBeginOffset =
427 Header.e_ehsize + Header.e_shentsize * Header.e_shnum;
428 ContiguousBlobAccumulator CBA(SectionContentBeginOffset);
429
430 // Doc might not contain .symtab, .strtab and .shstrtab sections,
431 // but we will emit them, so make sure to add them to ShStrTabSHeader.
432 State.DotShStrtab.add(".symtab");
433 State.DotShStrtab.add(".strtab");
434 State.DotShStrtab.add(".shstrtab");
435
436 std::vector<Elf_Shdr> SHeaders;
437 if(!State.initSectionHeaders(SHeaders, CBA))
438 return 1;
439
440 // .symtab section.
441 Elf_Shdr SymtabSHeader;
442 State.initSymtabSectionHeader(SymtabSHeader, CBA);
443 SHeaders.push_back(SymtabSHeader);
444
445 // .strtab string table header.
446 Elf_Shdr DotStrTabSHeader;
447 State.initStrtabSectionHeader(DotStrTabSHeader, ".strtab", State.DotStrtab,
448 CBA);
449 SHeaders.push_back(DotStrTabSHeader);
450
451 // .shstrtab string table header.
452 Elf_Shdr ShStrTabSHeader;
453 State.initStrtabSectionHeader(ShStrTabSHeader, ".shstrtab", State.DotShStrtab,
454 CBA);
455 SHeaders.push_back(ShStrTabSHeader);
456
457 OS.write((const char *)&Header, sizeof(Header));
458 writeArrayData(OS, makeArrayRef(SHeaders));
459 CBA.writeBlobToStream(OS);
460 return 0;
461 }
462
is64Bit(const ELFYAML::Object & Doc)463 static bool is64Bit(const ELFYAML::Object &Doc) {
464 return Doc.Header.Class == ELFYAML::ELF_ELFCLASS(ELF::ELFCLASS64);
465 }
466
isLittleEndian(const ELFYAML::Object & Doc)467 static bool isLittleEndian(const ELFYAML::Object &Doc) {
468 return Doc.Header.Data == ELFYAML::ELF_ELFDATA(ELF::ELFDATA2LSB);
469 }
470
yaml2elf(yaml::Input & YIn,raw_ostream & Out)471 int yaml2elf(yaml::Input &YIn, raw_ostream &Out) {
472 ELFYAML::Object Doc;
473 YIn >> Doc;
474 if (YIn.error()) {
475 errs() << "yaml2obj: Failed to parse YAML file!\n";
476 return 1;
477 }
478 using object::ELFType;
479 typedef ELFType<support::little, 8, true> LE64;
480 typedef ELFType<support::big, 8, true> BE64;
481 typedef ELFType<support::little, 4, false> LE32;
482 typedef ELFType<support::big, 4, false> BE32;
483 if (is64Bit(Doc)) {
484 if (isLittleEndian(Doc))
485 return ELFState<LE64>::writeELF(Out, Doc);
486 else
487 return ELFState<BE64>::writeELF(Out, Doc);
488 } else {
489 if (isLittleEndian(Doc))
490 return ELFState<LE32>::writeELF(Out, Doc);
491 else
492 return ELFState<BE32>::writeELF(Out, Doc);
493 }
494 }
495