1 //===- ELFObject.cpp ------------------------------------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #include "ELFObject.h" 10 #include "llvm/ADT/ArrayRef.h" 11 #include "llvm/ADT/STLExtras.h" 12 #include "llvm/ADT/StringRef.h" 13 #include "llvm/ADT/Twine.h" 14 #include "llvm/ADT/iterator_range.h" 15 #include "llvm/BinaryFormat/ELF.h" 16 #include "llvm/MC/MCTargetOptions.h" 17 #include "llvm/Object/ELF.h" 18 #include "llvm/Object/ELFObjectFile.h" 19 #include "llvm/Support/Compression.h" 20 #include "llvm/Support/Endian.h" 21 #include "llvm/Support/ErrorHandling.h" 22 #include "llvm/Support/FileOutputBuffer.h" 23 #include "llvm/Support/Path.h" 24 #include <algorithm> 25 #include <cstddef> 26 #include <cstdint> 27 #include <iterator> 28 #include <unordered_set> 29 #include <utility> 30 #include <vector> 31 32 using namespace llvm; 33 using namespace llvm::ELF; 34 using namespace llvm::objcopy::elf; 35 using namespace llvm::object; 36 37 template <class ELFT> void ELFWriter<ELFT>::writePhdr(const Segment &Seg) { 38 uint8_t *B = reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + 39 Obj.ProgramHdrSegment.Offset + Seg.Index * sizeof(Elf_Phdr); 40 Elf_Phdr &Phdr = *reinterpret_cast<Elf_Phdr *>(B); 41 Phdr.p_type = Seg.Type; 42 Phdr.p_flags = Seg.Flags; 43 Phdr.p_offset = Seg.Offset; 44 Phdr.p_vaddr = Seg.VAddr; 45 Phdr.p_paddr = Seg.PAddr; 46 Phdr.p_filesz = Seg.FileSize; 47 Phdr.p_memsz = Seg.MemSize; 48 Phdr.p_align = Seg.Align; 49 } 50 51 Error SectionBase::removeSectionReferences( 52 bool, function_ref<bool(const SectionBase *)>) { 53 return Error::success(); 54 } 55 56 Error SectionBase::removeSymbols(function_ref<bool(const Symbol &)>) { 57 return Error::success(); 58 } 59 60 Error SectionBase::initialize(SectionTableRef) { return Error::success(); } 61 void SectionBase::finalize() {} 62 void SectionBase::markSymbols() {} 63 void SectionBase::replaceSectionReferences( 64 const DenseMap<SectionBase *, SectionBase *> &) {} 65 void SectionBase::onRemove() {} 66 67 template <class ELFT> void ELFWriter<ELFT>::writeShdr(const SectionBase &Sec) { 68 uint8_t *B = 69 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Sec.HeaderOffset; 70 Elf_Shdr &Shdr = *reinterpret_cast<Elf_Shdr *>(B); 71 Shdr.sh_name = Sec.NameIndex; 72 Shdr.sh_type = Sec.Type; 73 Shdr.sh_flags = Sec.Flags; 74 Shdr.sh_addr = Sec.Addr; 75 Shdr.sh_offset = Sec.Offset; 76 Shdr.sh_size = Sec.Size; 77 Shdr.sh_link = Sec.Link; 78 Shdr.sh_info = Sec.Info; 79 Shdr.sh_addralign = Sec.Align; 80 Shdr.sh_entsize = Sec.EntrySize; 81 } 82 83 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(Section &) { 84 return Error::success(); 85 } 86 87 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(OwnedDataSection &) { 88 return Error::success(); 89 } 90 91 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(StringTableSection &) { 92 return Error::success(); 93 } 94 95 template <class ELFT> 96 Error ELFSectionSizer<ELFT>::visit(DynamicRelocationSection &) { 97 return Error::success(); 98 } 99 100 template <class ELFT> 101 Error ELFSectionSizer<ELFT>::visit(SymbolTableSection &Sec) { 102 Sec.EntrySize = sizeof(Elf_Sym); 103 Sec.Size = Sec.Symbols.size() * Sec.EntrySize; 104 // Align to the largest field in Elf_Sym. 105 Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word); 106 return Error::success(); 107 } 108 109 template <class ELFT> 110 Error ELFSectionSizer<ELFT>::visit(RelocationSection &Sec) { 111 Sec.EntrySize = Sec.Type == SHT_REL ? sizeof(Elf_Rel) : sizeof(Elf_Rela); 112 Sec.Size = Sec.Relocations.size() * Sec.EntrySize; 113 // Align to the largest field in Elf_Rel(a). 114 Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word); 115 return Error::success(); 116 } 117 118 template <class ELFT> 119 Error ELFSectionSizer<ELFT>::visit(GnuDebugLinkSection &) { 120 return Error::success(); 121 } 122 123 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(GroupSection &Sec) { 124 Sec.Size = sizeof(Elf_Word) + Sec.GroupMembers.size() * sizeof(Elf_Word); 125 return Error::success(); 126 } 127 128 template <class ELFT> 129 Error ELFSectionSizer<ELFT>::visit(SectionIndexSection &) { 130 return Error::success(); 131 } 132 133 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(CompressedSection &) { 134 return Error::success(); 135 } 136 137 template <class ELFT> 138 Error ELFSectionSizer<ELFT>::visit(DecompressedSection &) { 139 return Error::success(); 140 } 141 142 Error BinarySectionWriter::visit(const SectionIndexSection &Sec) { 143 return createStringError(errc::operation_not_permitted, 144 "cannot write symbol section index table '" + 145 Sec.Name + "' "); 146 } 147 148 Error BinarySectionWriter::visit(const SymbolTableSection &Sec) { 149 return createStringError(errc::operation_not_permitted, 150 "cannot write symbol table '" + Sec.Name + 151 "' out to binary"); 152 } 153 154 Error BinarySectionWriter::visit(const RelocationSection &Sec) { 155 return createStringError(errc::operation_not_permitted, 156 "cannot write relocation section '" + Sec.Name + 157 "' out to binary"); 158 } 159 160 Error BinarySectionWriter::visit(const GnuDebugLinkSection &Sec) { 161 return createStringError(errc::operation_not_permitted, 162 "cannot write '" + Sec.Name + "' out to binary"); 163 } 164 165 Error BinarySectionWriter::visit(const GroupSection &Sec) { 166 return createStringError(errc::operation_not_permitted, 167 "cannot write '" + Sec.Name + "' out to binary"); 168 } 169 170 Error SectionWriter::visit(const Section &Sec) { 171 if (Sec.Type != SHT_NOBITS) 172 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset); 173 174 return Error::success(); 175 } 176 177 static bool addressOverflows32bit(uint64_t Addr) { 178 // Sign extended 32 bit addresses (e.g 0xFFFFFFFF80000000) are ok 179 return Addr > UINT32_MAX && Addr + 0x80000000 > UINT32_MAX; 180 } 181 182 template <class T> static T checkedGetHex(StringRef S) { 183 T Value; 184 bool Fail = S.getAsInteger(16, Value); 185 assert(!Fail); 186 (void)Fail; 187 return Value; 188 } 189 190 // Fills exactly Len bytes of buffer with hexadecimal characters 191 // representing value 'X' 192 template <class T, class Iterator> 193 static Iterator toHexStr(T X, Iterator It, size_t Len) { 194 // Fill range with '0' 195 std::fill(It, It + Len, '0'); 196 197 for (long I = Len - 1; I >= 0; --I) { 198 unsigned char Mod = static_cast<unsigned char>(X) & 15; 199 *(It + I) = hexdigit(Mod, false); 200 X >>= 4; 201 } 202 assert(X == 0); 203 return It + Len; 204 } 205 206 uint8_t IHexRecord::getChecksum(StringRef S) { 207 assert((S.size() & 1) == 0); 208 uint8_t Checksum = 0; 209 while (!S.empty()) { 210 Checksum += checkedGetHex<uint8_t>(S.take_front(2)); 211 S = S.drop_front(2); 212 } 213 return -Checksum; 214 } 215 216 IHexLineData IHexRecord::getLine(uint8_t Type, uint16_t Addr, 217 ArrayRef<uint8_t> Data) { 218 IHexLineData Line(getLineLength(Data.size())); 219 assert(Line.size()); 220 auto Iter = Line.begin(); 221 *Iter++ = ':'; 222 Iter = toHexStr(Data.size(), Iter, 2); 223 Iter = toHexStr(Addr, Iter, 4); 224 Iter = toHexStr(Type, Iter, 2); 225 for (uint8_t X : Data) 226 Iter = toHexStr(X, Iter, 2); 227 StringRef S(Line.data() + 1, std::distance(Line.begin() + 1, Iter)); 228 Iter = toHexStr(getChecksum(S), Iter, 2); 229 *Iter++ = '\r'; 230 *Iter++ = '\n'; 231 assert(Iter == Line.end()); 232 return Line; 233 } 234 235 static Error checkRecord(const IHexRecord &R) { 236 switch (R.Type) { 237 case IHexRecord::Data: 238 if (R.HexData.size() == 0) 239 return createStringError( 240 errc::invalid_argument, 241 "zero data length is not allowed for data records"); 242 break; 243 case IHexRecord::EndOfFile: 244 break; 245 case IHexRecord::SegmentAddr: 246 // 20-bit segment address. Data length must be 2 bytes 247 // (4 bytes in hex) 248 if (R.HexData.size() != 4) 249 return createStringError( 250 errc::invalid_argument, 251 "segment address data should be 2 bytes in size"); 252 break; 253 case IHexRecord::StartAddr80x86: 254 case IHexRecord::StartAddr: 255 if (R.HexData.size() != 8) 256 return createStringError(errc::invalid_argument, 257 "start address data should be 4 bytes in size"); 258 // According to Intel HEX specification '03' record 259 // only specifies the code address within the 20-bit 260 // segmented address space of the 8086/80186. This 261 // means 12 high order bits should be zeroes. 262 if (R.Type == IHexRecord::StartAddr80x86 && 263 R.HexData.take_front(3) != "000") 264 return createStringError(errc::invalid_argument, 265 "start address exceeds 20 bit for 80x86"); 266 break; 267 case IHexRecord::ExtendedAddr: 268 // 16-31 bits of linear base address 269 if (R.HexData.size() != 4) 270 return createStringError( 271 errc::invalid_argument, 272 "extended address data should be 2 bytes in size"); 273 break; 274 default: 275 // Unknown record type 276 return createStringError(errc::invalid_argument, "unknown record type: %u", 277 static_cast<unsigned>(R.Type)); 278 } 279 return Error::success(); 280 } 281 282 // Checks that IHEX line contains valid characters. 283 // This allows converting hexadecimal data to integers 284 // without extra verification. 285 static Error checkChars(StringRef Line) { 286 assert(!Line.empty()); 287 if (Line[0] != ':') 288 return createStringError(errc::invalid_argument, 289 "missing ':' in the beginning of line."); 290 291 for (size_t Pos = 1; Pos < Line.size(); ++Pos) 292 if (hexDigitValue(Line[Pos]) == -1U) 293 return createStringError(errc::invalid_argument, 294 "invalid character at position %zu.", Pos + 1); 295 return Error::success(); 296 } 297 298 Expected<IHexRecord> IHexRecord::parse(StringRef Line) { 299 assert(!Line.empty()); 300 301 // ':' + Length + Address + Type + Checksum with empty data ':LLAAAATTCC' 302 if (Line.size() < 11) 303 return createStringError(errc::invalid_argument, 304 "line is too short: %zu chars.", Line.size()); 305 306 if (Error E = checkChars(Line)) 307 return std::move(E); 308 309 IHexRecord Rec; 310 size_t DataLen = checkedGetHex<uint8_t>(Line.substr(1, 2)); 311 if (Line.size() != getLength(DataLen)) 312 return createStringError(errc::invalid_argument, 313 "invalid line length %zu (should be %zu)", 314 Line.size(), getLength(DataLen)); 315 316 Rec.Addr = checkedGetHex<uint16_t>(Line.substr(3, 4)); 317 Rec.Type = checkedGetHex<uint8_t>(Line.substr(7, 2)); 318 Rec.HexData = Line.substr(9, DataLen * 2); 319 320 if (getChecksum(Line.drop_front(1)) != 0) 321 return createStringError(errc::invalid_argument, "incorrect checksum."); 322 if (Error E = checkRecord(Rec)) 323 return std::move(E); 324 return Rec; 325 } 326 327 static uint64_t sectionPhysicalAddr(const SectionBase *Sec) { 328 Segment *Seg = Sec->ParentSegment; 329 if (Seg && Seg->Type != ELF::PT_LOAD) 330 Seg = nullptr; 331 return Seg ? Seg->PAddr + Sec->OriginalOffset - Seg->OriginalOffset 332 : Sec->Addr; 333 } 334 335 void IHexSectionWriterBase::writeSection(const SectionBase *Sec, 336 ArrayRef<uint8_t> Data) { 337 assert(Data.size() == Sec->Size); 338 const uint32_t ChunkSize = 16; 339 uint32_t Addr = sectionPhysicalAddr(Sec) & 0xFFFFFFFFU; 340 while (!Data.empty()) { 341 uint64_t DataSize = std::min<uint64_t>(Data.size(), ChunkSize); 342 if (Addr > SegmentAddr + BaseAddr + 0xFFFFU) { 343 if (Addr > 0xFFFFFU) { 344 // Write extended address record, zeroing segment address 345 // if needed. 346 if (SegmentAddr != 0) 347 SegmentAddr = writeSegmentAddr(0U); 348 BaseAddr = writeBaseAddr(Addr); 349 } else { 350 // We can still remain 16-bit 351 SegmentAddr = writeSegmentAddr(Addr); 352 } 353 } 354 uint64_t SegOffset = Addr - BaseAddr - SegmentAddr; 355 assert(SegOffset <= 0xFFFFU); 356 DataSize = std::min(DataSize, 0x10000U - SegOffset); 357 writeData(0, SegOffset, Data.take_front(DataSize)); 358 Addr += DataSize; 359 Data = Data.drop_front(DataSize); 360 } 361 } 362 363 uint64_t IHexSectionWriterBase::writeSegmentAddr(uint64_t Addr) { 364 assert(Addr <= 0xFFFFFU); 365 uint8_t Data[] = {static_cast<uint8_t>((Addr & 0xF0000U) >> 12), 0}; 366 writeData(2, 0, Data); 367 return Addr & 0xF0000U; 368 } 369 370 uint64_t IHexSectionWriterBase::writeBaseAddr(uint64_t Addr) { 371 assert(Addr <= 0xFFFFFFFFU); 372 uint64_t Base = Addr & 0xFFFF0000U; 373 uint8_t Data[] = {static_cast<uint8_t>(Base >> 24), 374 static_cast<uint8_t>((Base >> 16) & 0xFF)}; 375 writeData(4, 0, Data); 376 return Base; 377 } 378 379 void IHexSectionWriterBase::writeData(uint8_t, uint16_t, 380 ArrayRef<uint8_t> Data) { 381 Offset += IHexRecord::getLineLength(Data.size()); 382 } 383 384 Error IHexSectionWriterBase::visit(const Section &Sec) { 385 writeSection(&Sec, Sec.Contents); 386 return Error::success(); 387 } 388 389 Error IHexSectionWriterBase::visit(const OwnedDataSection &Sec) { 390 writeSection(&Sec, Sec.Data); 391 return Error::success(); 392 } 393 394 Error IHexSectionWriterBase::visit(const StringTableSection &Sec) { 395 // Check that sizer has already done its work 396 assert(Sec.Size == Sec.StrTabBuilder.getSize()); 397 // We are free to pass an invalid pointer to writeSection as long 398 // as we don't actually write any data. The real writer class has 399 // to override this method . 400 writeSection(&Sec, {nullptr, static_cast<size_t>(Sec.Size)}); 401 return Error::success(); 402 } 403 404 Error IHexSectionWriterBase::visit(const DynamicRelocationSection &Sec) { 405 writeSection(&Sec, Sec.Contents); 406 return Error::success(); 407 } 408 409 void IHexSectionWriter::writeData(uint8_t Type, uint16_t Addr, 410 ArrayRef<uint8_t> Data) { 411 IHexLineData HexData = IHexRecord::getLine(Type, Addr, Data); 412 memcpy(Out.getBufferStart() + Offset, HexData.data(), HexData.size()); 413 Offset += HexData.size(); 414 } 415 416 Error IHexSectionWriter::visit(const StringTableSection &Sec) { 417 assert(Sec.Size == Sec.StrTabBuilder.getSize()); 418 std::vector<uint8_t> Data(Sec.Size); 419 Sec.StrTabBuilder.write(Data.data()); 420 writeSection(&Sec, Data); 421 return Error::success(); 422 } 423 424 Error Section::accept(SectionVisitor &Visitor) const { 425 return Visitor.visit(*this); 426 } 427 428 Error Section::accept(MutableSectionVisitor &Visitor) { 429 return Visitor.visit(*this); 430 } 431 432 Error SectionWriter::visit(const OwnedDataSection &Sec) { 433 llvm::copy(Sec.Data, Out.getBufferStart() + Sec.Offset); 434 return Error::success(); 435 } 436 437 static constexpr std::array<uint8_t, 4> ZlibGnuMagic = {{'Z', 'L', 'I', 'B'}}; 438 439 static bool isDataGnuCompressed(ArrayRef<uint8_t> Data) { 440 return Data.size() > ZlibGnuMagic.size() && 441 std::equal(ZlibGnuMagic.begin(), ZlibGnuMagic.end(), Data.data()); 442 } 443 444 template <class ELFT> 445 static std::tuple<uint64_t, uint64_t> 446 getDecompressedSizeAndAlignment(ArrayRef<uint8_t> Data) { 447 const bool IsGnuDebug = isDataGnuCompressed(Data); 448 const uint64_t DecompressedSize = 449 IsGnuDebug 450 ? support::endian::read64be(Data.data() + ZlibGnuMagic.size()) 451 : reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data.data())->ch_size; 452 const uint64_t DecompressedAlign = 453 IsGnuDebug ? 1 454 : reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data.data()) 455 ->ch_addralign; 456 457 return std::make_tuple(DecompressedSize, DecompressedAlign); 458 } 459 460 template <class ELFT> 461 Error ELFSectionWriter<ELFT>::visit(const DecompressedSection &Sec) { 462 const size_t DataOffset = isDataGnuCompressed(Sec.OriginalData) 463 ? (ZlibGnuMagic.size() + sizeof(Sec.Size)) 464 : sizeof(Elf_Chdr_Impl<ELFT>); 465 466 StringRef CompressedContent( 467 reinterpret_cast<const char *>(Sec.OriginalData.data()) + DataOffset, 468 Sec.OriginalData.size() - DataOffset); 469 470 SmallVector<char, 128> DecompressedContent; 471 if (Error Err = zlib::uncompress(CompressedContent, DecompressedContent, 472 static_cast<size_t>(Sec.Size))) 473 return createStringError(errc::invalid_argument, 474 "'" + Sec.Name + "': " + toString(std::move(Err))); 475 476 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 477 std::copy(DecompressedContent.begin(), DecompressedContent.end(), Buf); 478 479 return Error::success(); 480 } 481 482 Error BinarySectionWriter::visit(const DecompressedSection &Sec) { 483 return createStringError(errc::operation_not_permitted, 484 "cannot write compressed section '" + Sec.Name + 485 "' "); 486 } 487 488 Error DecompressedSection::accept(SectionVisitor &Visitor) const { 489 return Visitor.visit(*this); 490 } 491 492 Error DecompressedSection::accept(MutableSectionVisitor &Visitor) { 493 return Visitor.visit(*this); 494 } 495 496 Error OwnedDataSection::accept(SectionVisitor &Visitor) const { 497 return Visitor.visit(*this); 498 } 499 500 Error OwnedDataSection::accept(MutableSectionVisitor &Visitor) { 501 return Visitor.visit(*this); 502 } 503 504 void OwnedDataSection::appendHexData(StringRef HexData) { 505 assert((HexData.size() & 1) == 0); 506 while (!HexData.empty()) { 507 Data.push_back(checkedGetHex<uint8_t>(HexData.take_front(2))); 508 HexData = HexData.drop_front(2); 509 } 510 Size = Data.size(); 511 } 512 513 Error BinarySectionWriter::visit(const CompressedSection &Sec) { 514 return createStringError(errc::operation_not_permitted, 515 "cannot write compressed section '" + Sec.Name + 516 "' "); 517 } 518 519 template <class ELFT> 520 Error ELFSectionWriter<ELFT>::visit(const CompressedSection &Sec) { 521 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 522 Elf_Chdr_Impl<ELFT> Chdr; 523 switch (Sec.CompressionType) { 524 case DebugCompressionType::None: 525 std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(), Buf); 526 return Error::success(); 527 case DebugCompressionType::GNU: 528 llvm_unreachable("unexpected zlib-gnu"); 529 break; 530 case DebugCompressionType::Z: 531 Chdr.ch_type = ELF::ELFCOMPRESS_ZLIB; 532 break; 533 } 534 Chdr.ch_size = Sec.DecompressedSize; 535 Chdr.ch_addralign = Sec.DecompressedAlign; 536 memcpy(Buf, &Chdr, sizeof(Chdr)); 537 Buf += sizeof(Chdr); 538 539 std::copy(Sec.CompressedData.begin(), Sec.CompressedData.end(), Buf); 540 return Error::success(); 541 } 542 543 CompressedSection::CompressedSection(const SectionBase &Sec, 544 DebugCompressionType CompressionType) 545 : SectionBase(Sec), CompressionType(CompressionType), 546 DecompressedSize(Sec.OriginalData.size()), DecompressedAlign(Sec.Align) { 547 zlib::compress(StringRef(reinterpret_cast<const char *>(OriginalData.data()), 548 OriginalData.size()), 549 CompressedData); 550 551 assert(CompressionType != DebugCompressionType::None); 552 Flags |= ELF::SHF_COMPRESSED; 553 size_t ChdrSize = 554 std::max(std::max(sizeof(object::Elf_Chdr_Impl<object::ELF64LE>), 555 sizeof(object::Elf_Chdr_Impl<object::ELF64BE>)), 556 std::max(sizeof(object::Elf_Chdr_Impl<object::ELF32LE>), 557 sizeof(object::Elf_Chdr_Impl<object::ELF32BE>))); 558 Size = ChdrSize + CompressedData.size(); 559 Align = 8; 560 } 561 562 CompressedSection::CompressedSection(ArrayRef<uint8_t> CompressedData, 563 uint64_t DecompressedSize, 564 uint64_t DecompressedAlign) 565 : CompressionType(DebugCompressionType::None), 566 DecompressedSize(DecompressedSize), DecompressedAlign(DecompressedAlign) { 567 OriginalData = CompressedData; 568 } 569 570 Error CompressedSection::accept(SectionVisitor &Visitor) const { 571 return Visitor.visit(*this); 572 } 573 574 Error CompressedSection::accept(MutableSectionVisitor &Visitor) { 575 return Visitor.visit(*this); 576 } 577 578 void StringTableSection::addString(StringRef Name) { StrTabBuilder.add(Name); } 579 580 uint32_t StringTableSection::findIndex(StringRef Name) const { 581 return StrTabBuilder.getOffset(Name); 582 } 583 584 void StringTableSection::prepareForLayout() { 585 StrTabBuilder.finalize(); 586 Size = StrTabBuilder.getSize(); 587 } 588 589 Error SectionWriter::visit(const StringTableSection &Sec) { 590 Sec.StrTabBuilder.write(reinterpret_cast<uint8_t *>(Out.getBufferStart()) + 591 Sec.Offset); 592 return Error::success(); 593 } 594 595 Error StringTableSection::accept(SectionVisitor &Visitor) const { 596 return Visitor.visit(*this); 597 } 598 599 Error StringTableSection::accept(MutableSectionVisitor &Visitor) { 600 return Visitor.visit(*this); 601 } 602 603 template <class ELFT> 604 Error ELFSectionWriter<ELFT>::visit(const SectionIndexSection &Sec) { 605 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 606 llvm::copy(Sec.Indexes, reinterpret_cast<Elf_Word *>(Buf)); 607 return Error::success(); 608 } 609 610 Error SectionIndexSection::initialize(SectionTableRef SecTable) { 611 Size = 0; 612 Expected<SymbolTableSection *> Sec = 613 SecTable.getSectionOfType<SymbolTableSection>( 614 Link, 615 "Link field value " + Twine(Link) + " in section " + Name + 616 " is invalid", 617 "Link field value " + Twine(Link) + " in section " + Name + 618 " is not a symbol table"); 619 if (!Sec) 620 return Sec.takeError(); 621 622 setSymTab(*Sec); 623 Symbols->setShndxTable(this); 624 return Error::success(); 625 } 626 627 void SectionIndexSection::finalize() { Link = Symbols->Index; } 628 629 Error SectionIndexSection::accept(SectionVisitor &Visitor) const { 630 return Visitor.visit(*this); 631 } 632 633 Error SectionIndexSection::accept(MutableSectionVisitor &Visitor) { 634 return Visitor.visit(*this); 635 } 636 637 static bool isValidReservedSectionIndex(uint16_t Index, uint16_t Machine) { 638 switch (Index) { 639 case SHN_ABS: 640 case SHN_COMMON: 641 return true; 642 } 643 644 if (Machine == EM_AMDGPU) { 645 return Index == SHN_AMDGPU_LDS; 646 } 647 648 if (Machine == EM_MIPS) { 649 switch (Index) { 650 case SHN_MIPS_ACOMMON: 651 case SHN_MIPS_SCOMMON: 652 case SHN_MIPS_SUNDEFINED: 653 return true; 654 } 655 } 656 657 if (Machine == EM_HEXAGON) { 658 switch (Index) { 659 case SHN_HEXAGON_SCOMMON: 660 case SHN_HEXAGON_SCOMMON_1: 661 case SHN_HEXAGON_SCOMMON_2: 662 case SHN_HEXAGON_SCOMMON_4: 663 case SHN_HEXAGON_SCOMMON_8: 664 return true; 665 } 666 } 667 return false; 668 } 669 670 // Large indexes force us to clarify exactly what this function should do. This 671 // function should return the value that will appear in st_shndx when written 672 // out. 673 uint16_t Symbol::getShndx() const { 674 if (DefinedIn != nullptr) { 675 if (DefinedIn->Index >= SHN_LORESERVE) 676 return SHN_XINDEX; 677 return DefinedIn->Index; 678 } 679 680 if (ShndxType == SYMBOL_SIMPLE_INDEX) { 681 // This means that we don't have a defined section but we do need to 682 // output a legitimate section index. 683 return SHN_UNDEF; 684 } 685 686 assert(ShndxType == SYMBOL_ABS || ShndxType == SYMBOL_COMMON || 687 (ShndxType >= SYMBOL_LOPROC && ShndxType <= SYMBOL_HIPROC) || 688 (ShndxType >= SYMBOL_LOOS && ShndxType <= SYMBOL_HIOS)); 689 return static_cast<uint16_t>(ShndxType); 690 } 691 692 bool Symbol::isCommon() const { return getShndx() == SHN_COMMON; } 693 694 void SymbolTableSection::assignIndices() { 695 uint32_t Index = 0; 696 for (auto &Sym : Symbols) 697 Sym->Index = Index++; 698 } 699 700 void SymbolTableSection::addSymbol(Twine Name, uint8_t Bind, uint8_t Type, 701 SectionBase *DefinedIn, uint64_t Value, 702 uint8_t Visibility, uint16_t Shndx, 703 uint64_t SymbolSize) { 704 Symbol Sym; 705 Sym.Name = Name.str(); 706 Sym.Binding = Bind; 707 Sym.Type = Type; 708 Sym.DefinedIn = DefinedIn; 709 if (DefinedIn != nullptr) 710 DefinedIn->HasSymbol = true; 711 if (DefinedIn == nullptr) { 712 if (Shndx >= SHN_LORESERVE) 713 Sym.ShndxType = static_cast<SymbolShndxType>(Shndx); 714 else 715 Sym.ShndxType = SYMBOL_SIMPLE_INDEX; 716 } 717 Sym.Value = Value; 718 Sym.Visibility = Visibility; 719 Sym.Size = SymbolSize; 720 Sym.Index = Symbols.size(); 721 Symbols.emplace_back(std::make_unique<Symbol>(Sym)); 722 Size += this->EntrySize; 723 } 724 725 Error SymbolTableSection::removeSectionReferences( 726 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 727 if (ToRemove(SectionIndexTable)) 728 SectionIndexTable = nullptr; 729 if (ToRemove(SymbolNames)) { 730 if (!AllowBrokenLinks) 731 return createStringError( 732 llvm::errc::invalid_argument, 733 "string table '%s' cannot be removed because it is " 734 "referenced by the symbol table '%s'", 735 SymbolNames->Name.data(), this->Name.data()); 736 SymbolNames = nullptr; 737 } 738 return removeSymbols( 739 [ToRemove](const Symbol &Sym) { return ToRemove(Sym.DefinedIn); }); 740 } 741 742 void SymbolTableSection::updateSymbols(function_ref<void(Symbol &)> Callable) { 743 for (SymPtr &Sym : llvm::drop_begin(Symbols)) 744 Callable(*Sym); 745 std::stable_partition( 746 std::begin(Symbols), std::end(Symbols), 747 [](const SymPtr &Sym) { return Sym->Binding == STB_LOCAL; }); 748 assignIndices(); 749 } 750 751 Error SymbolTableSection::removeSymbols( 752 function_ref<bool(const Symbol &)> ToRemove) { 753 Symbols.erase( 754 std::remove_if(std::begin(Symbols) + 1, std::end(Symbols), 755 [ToRemove](const SymPtr &Sym) { return ToRemove(*Sym); }), 756 std::end(Symbols)); 757 Size = Symbols.size() * EntrySize; 758 assignIndices(); 759 return Error::success(); 760 } 761 762 void SymbolTableSection::replaceSectionReferences( 763 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 764 for (std::unique_ptr<Symbol> &Sym : Symbols) 765 if (SectionBase *To = FromTo.lookup(Sym->DefinedIn)) 766 Sym->DefinedIn = To; 767 } 768 769 Error SymbolTableSection::initialize(SectionTableRef SecTable) { 770 Size = 0; 771 Expected<StringTableSection *> Sec = 772 SecTable.getSectionOfType<StringTableSection>( 773 Link, 774 "Symbol table has link index of " + Twine(Link) + 775 " which is not a valid index", 776 "Symbol table has link index of " + Twine(Link) + 777 " which is not a string table"); 778 if (!Sec) 779 return Sec.takeError(); 780 781 setStrTab(*Sec); 782 return Error::success(); 783 } 784 785 void SymbolTableSection::finalize() { 786 uint32_t MaxLocalIndex = 0; 787 for (std::unique_ptr<Symbol> &Sym : Symbols) { 788 Sym->NameIndex = 789 SymbolNames == nullptr ? 0 : SymbolNames->findIndex(Sym->Name); 790 if (Sym->Binding == STB_LOCAL) 791 MaxLocalIndex = std::max(MaxLocalIndex, Sym->Index); 792 } 793 // Now we need to set the Link and Info fields. 794 Link = SymbolNames == nullptr ? 0 : SymbolNames->Index; 795 Info = MaxLocalIndex + 1; 796 } 797 798 void SymbolTableSection::prepareForLayout() { 799 // Reserve proper amount of space in section index table, so we can 800 // layout sections correctly. We will fill the table with correct 801 // indexes later in fillShdnxTable. 802 if (SectionIndexTable) 803 SectionIndexTable->reserve(Symbols.size()); 804 805 // Add all of our strings to SymbolNames so that SymbolNames has the right 806 // size before layout is decided. 807 // If the symbol names section has been removed, don't try to add strings to 808 // the table. 809 if (SymbolNames != nullptr) 810 for (std::unique_ptr<Symbol> &Sym : Symbols) 811 SymbolNames->addString(Sym->Name); 812 } 813 814 void SymbolTableSection::fillShndxTable() { 815 if (SectionIndexTable == nullptr) 816 return; 817 // Fill section index table with real section indexes. This function must 818 // be called after assignOffsets. 819 for (const std::unique_ptr<Symbol> &Sym : Symbols) { 820 if (Sym->DefinedIn != nullptr && Sym->DefinedIn->Index >= SHN_LORESERVE) 821 SectionIndexTable->addIndex(Sym->DefinedIn->Index); 822 else 823 SectionIndexTable->addIndex(SHN_UNDEF); 824 } 825 } 826 827 Expected<const Symbol *> 828 SymbolTableSection::getSymbolByIndex(uint32_t Index) const { 829 if (Symbols.size() <= Index) 830 return createStringError(errc::invalid_argument, 831 "invalid symbol index: " + Twine(Index)); 832 return Symbols[Index].get(); 833 } 834 835 Expected<Symbol *> SymbolTableSection::getSymbolByIndex(uint32_t Index) { 836 Expected<const Symbol *> Sym = 837 static_cast<const SymbolTableSection *>(this)->getSymbolByIndex(Index); 838 if (!Sym) 839 return Sym.takeError(); 840 841 return const_cast<Symbol *>(*Sym); 842 } 843 844 template <class ELFT> 845 Error ELFSectionWriter<ELFT>::visit(const SymbolTableSection &Sec) { 846 Elf_Sym *Sym = reinterpret_cast<Elf_Sym *>(Out.getBufferStart() + Sec.Offset); 847 // Loop though symbols setting each entry of the symbol table. 848 for (const std::unique_ptr<Symbol> &Symbol : Sec.Symbols) { 849 Sym->st_name = Symbol->NameIndex; 850 Sym->st_value = Symbol->Value; 851 Sym->st_size = Symbol->Size; 852 Sym->st_other = Symbol->Visibility; 853 Sym->setBinding(Symbol->Binding); 854 Sym->setType(Symbol->Type); 855 Sym->st_shndx = Symbol->getShndx(); 856 ++Sym; 857 } 858 return Error::success(); 859 } 860 861 Error SymbolTableSection::accept(SectionVisitor &Visitor) const { 862 return Visitor.visit(*this); 863 } 864 865 Error SymbolTableSection::accept(MutableSectionVisitor &Visitor) { 866 return Visitor.visit(*this); 867 } 868 869 StringRef RelocationSectionBase::getNamePrefix() const { 870 switch (Type) { 871 case SHT_REL: 872 return ".rel"; 873 case SHT_RELA: 874 return ".rela"; 875 default: 876 llvm_unreachable("not a relocation section"); 877 } 878 } 879 880 Error RelocationSection::removeSectionReferences( 881 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 882 if (ToRemove(Symbols)) { 883 if (!AllowBrokenLinks) 884 return createStringError( 885 llvm::errc::invalid_argument, 886 "symbol table '%s' cannot be removed because it is " 887 "referenced by the relocation section '%s'", 888 Symbols->Name.data(), this->Name.data()); 889 Symbols = nullptr; 890 } 891 892 for (const Relocation &R : Relocations) { 893 if (!R.RelocSymbol || !R.RelocSymbol->DefinedIn || 894 !ToRemove(R.RelocSymbol->DefinedIn)) 895 continue; 896 return createStringError(llvm::errc::invalid_argument, 897 "section '%s' cannot be removed: (%s+0x%" PRIx64 898 ") has relocation against symbol '%s'", 899 R.RelocSymbol->DefinedIn->Name.data(), 900 SecToApplyRel->Name.data(), R.Offset, 901 R.RelocSymbol->Name.c_str()); 902 } 903 904 return Error::success(); 905 } 906 907 template <class SymTabType> 908 Error RelocSectionWithSymtabBase<SymTabType>::initialize( 909 SectionTableRef SecTable) { 910 if (Link != SHN_UNDEF) { 911 Expected<SymTabType *> Sec = SecTable.getSectionOfType<SymTabType>( 912 Link, 913 "Link field value " + Twine(Link) + " in section " + Name + 914 " is invalid", 915 "Link field value " + Twine(Link) + " in section " + Name + 916 " is not a symbol table"); 917 if (!Sec) 918 return Sec.takeError(); 919 920 setSymTab(*Sec); 921 } 922 923 if (Info != SHN_UNDEF) { 924 Expected<SectionBase *> Sec = 925 SecTable.getSection(Info, "Info field value " + Twine(Info) + 926 " in section " + Name + " is invalid"); 927 if (!Sec) 928 return Sec.takeError(); 929 930 setSection(*Sec); 931 } else 932 setSection(nullptr); 933 934 return Error::success(); 935 } 936 937 template <class SymTabType> 938 void RelocSectionWithSymtabBase<SymTabType>::finalize() { 939 this->Link = Symbols ? Symbols->Index : 0; 940 941 if (SecToApplyRel != nullptr) 942 this->Info = SecToApplyRel->Index; 943 } 944 945 template <class ELFT> 946 static void setAddend(Elf_Rel_Impl<ELFT, false> &, uint64_t) {} 947 948 template <class ELFT> 949 static void setAddend(Elf_Rel_Impl<ELFT, true> &Rela, uint64_t Addend) { 950 Rela.r_addend = Addend; 951 } 952 953 template <class RelRange, class T> 954 static void writeRel(const RelRange &Relocations, T *Buf, bool IsMips64EL) { 955 for (const auto &Reloc : Relocations) { 956 Buf->r_offset = Reloc.Offset; 957 setAddend(*Buf, Reloc.Addend); 958 Buf->setSymbolAndType(Reloc.RelocSymbol ? Reloc.RelocSymbol->Index : 0, 959 Reloc.Type, IsMips64EL); 960 ++Buf; 961 } 962 } 963 964 template <class ELFT> 965 Error ELFSectionWriter<ELFT>::visit(const RelocationSection &Sec) { 966 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 967 if (Sec.Type == SHT_REL) 968 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rel *>(Buf), 969 Sec.getObject().IsMips64EL); 970 else 971 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rela *>(Buf), 972 Sec.getObject().IsMips64EL); 973 return Error::success(); 974 } 975 976 Error RelocationSection::accept(SectionVisitor &Visitor) const { 977 return Visitor.visit(*this); 978 } 979 980 Error RelocationSection::accept(MutableSectionVisitor &Visitor) { 981 return Visitor.visit(*this); 982 } 983 984 Error RelocationSection::removeSymbols( 985 function_ref<bool(const Symbol &)> ToRemove) { 986 for (const Relocation &Reloc : Relocations) 987 if (Reloc.RelocSymbol && ToRemove(*Reloc.RelocSymbol)) 988 return createStringError( 989 llvm::errc::invalid_argument, 990 "not stripping symbol '%s' because it is named in a relocation", 991 Reloc.RelocSymbol->Name.data()); 992 return Error::success(); 993 } 994 995 void RelocationSection::markSymbols() { 996 for (const Relocation &Reloc : Relocations) 997 if (Reloc.RelocSymbol) 998 Reloc.RelocSymbol->Referenced = true; 999 } 1000 1001 void RelocationSection::replaceSectionReferences( 1002 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 1003 // Update the target section if it was replaced. 1004 if (SectionBase *To = FromTo.lookup(SecToApplyRel)) 1005 SecToApplyRel = To; 1006 } 1007 1008 Error SectionWriter::visit(const DynamicRelocationSection &Sec) { 1009 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset); 1010 return Error::success(); 1011 } 1012 1013 Error DynamicRelocationSection::accept(SectionVisitor &Visitor) const { 1014 return Visitor.visit(*this); 1015 } 1016 1017 Error DynamicRelocationSection::accept(MutableSectionVisitor &Visitor) { 1018 return Visitor.visit(*this); 1019 } 1020 1021 Error DynamicRelocationSection::removeSectionReferences( 1022 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 1023 if (ToRemove(Symbols)) { 1024 if (!AllowBrokenLinks) 1025 return createStringError( 1026 llvm::errc::invalid_argument, 1027 "symbol table '%s' cannot be removed because it is " 1028 "referenced by the relocation section '%s'", 1029 Symbols->Name.data(), this->Name.data()); 1030 Symbols = nullptr; 1031 } 1032 1033 // SecToApplyRel contains a section referenced by sh_info field. It keeps 1034 // a section to which the relocation section applies. When we remove any 1035 // sections we also remove their relocation sections. Since we do that much 1036 // earlier, this assert should never be triggered. 1037 assert(!SecToApplyRel || !ToRemove(SecToApplyRel)); 1038 return Error::success(); 1039 } 1040 1041 Error Section::removeSectionReferences( 1042 bool AllowBrokenDependency, 1043 function_ref<bool(const SectionBase *)> ToRemove) { 1044 if (ToRemove(LinkSection)) { 1045 if (!AllowBrokenDependency) 1046 return createStringError(llvm::errc::invalid_argument, 1047 "section '%s' cannot be removed because it is " 1048 "referenced by the section '%s'", 1049 LinkSection->Name.data(), this->Name.data()); 1050 LinkSection = nullptr; 1051 } 1052 return Error::success(); 1053 } 1054 1055 void GroupSection::finalize() { 1056 this->Info = Sym ? Sym->Index : 0; 1057 this->Link = SymTab ? SymTab->Index : 0; 1058 // Linker deduplication for GRP_COMDAT is based on Sym->Name. The local/global 1059 // status is not part of the equation. If Sym is localized, the intention is 1060 // likely to make the group fully localized. Drop GRP_COMDAT to suppress 1061 // deduplication. See https://groups.google.com/g/generic-abi/c/2X6mR-s2zoc 1062 if ((FlagWord & GRP_COMDAT) && Sym && Sym->Binding == STB_LOCAL) 1063 this->FlagWord &= ~GRP_COMDAT; 1064 } 1065 1066 Error GroupSection::removeSectionReferences( 1067 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 1068 if (ToRemove(SymTab)) { 1069 if (!AllowBrokenLinks) 1070 return createStringError( 1071 llvm::errc::invalid_argument, 1072 "section '.symtab' cannot be removed because it is " 1073 "referenced by the group section '%s'", 1074 this->Name.data()); 1075 SymTab = nullptr; 1076 Sym = nullptr; 1077 } 1078 llvm::erase_if(GroupMembers, ToRemove); 1079 return Error::success(); 1080 } 1081 1082 Error GroupSection::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) { 1083 if (ToRemove(*Sym)) 1084 return createStringError(llvm::errc::invalid_argument, 1085 "symbol '%s' cannot be removed because it is " 1086 "referenced by the section '%s[%d]'", 1087 Sym->Name.data(), this->Name.data(), this->Index); 1088 return Error::success(); 1089 } 1090 1091 void GroupSection::markSymbols() { 1092 if (Sym) 1093 Sym->Referenced = true; 1094 } 1095 1096 void GroupSection::replaceSectionReferences( 1097 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 1098 for (SectionBase *&Sec : GroupMembers) 1099 if (SectionBase *To = FromTo.lookup(Sec)) 1100 Sec = To; 1101 } 1102 1103 void GroupSection::onRemove() { 1104 // As the header section of the group is removed, drop the Group flag in its 1105 // former members. 1106 for (SectionBase *Sec : GroupMembers) 1107 Sec->Flags &= ~SHF_GROUP; 1108 } 1109 1110 Error Section::initialize(SectionTableRef SecTable) { 1111 if (Link == ELF::SHN_UNDEF) 1112 return Error::success(); 1113 1114 Expected<SectionBase *> Sec = 1115 SecTable.getSection(Link, "Link field value " + Twine(Link) + 1116 " in section " + Name + " is invalid"); 1117 if (!Sec) 1118 return Sec.takeError(); 1119 1120 LinkSection = *Sec; 1121 1122 if (LinkSection->Type == ELF::SHT_SYMTAB) 1123 LinkSection = nullptr; 1124 1125 return Error::success(); 1126 } 1127 1128 void Section::finalize() { this->Link = LinkSection ? LinkSection->Index : 0; } 1129 1130 void GnuDebugLinkSection::init(StringRef File) { 1131 FileName = sys::path::filename(File); 1132 // The format for the .gnu_debuglink starts with the file name and is 1133 // followed by a null terminator and then the CRC32 of the file. The CRC32 1134 // should be 4 byte aligned. So we add the FileName size, a 1 for the null 1135 // byte, and then finally push the size to alignment and add 4. 1136 Size = alignTo(FileName.size() + 1, 4) + 4; 1137 // The CRC32 will only be aligned if we align the whole section. 1138 Align = 4; 1139 Type = OriginalType = ELF::SHT_PROGBITS; 1140 Name = ".gnu_debuglink"; 1141 // For sections not found in segments, OriginalOffset is only used to 1142 // establish the order that sections should go in. By using the maximum 1143 // possible offset we cause this section to wind up at the end. 1144 OriginalOffset = std::numeric_limits<uint64_t>::max(); 1145 } 1146 1147 GnuDebugLinkSection::GnuDebugLinkSection(StringRef File, 1148 uint32_t PrecomputedCRC) 1149 : FileName(File), CRC32(PrecomputedCRC) { 1150 init(File); 1151 } 1152 1153 template <class ELFT> 1154 Error ELFSectionWriter<ELFT>::visit(const GnuDebugLinkSection &Sec) { 1155 unsigned char *Buf = 1156 reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 1157 Elf_Word *CRC = 1158 reinterpret_cast<Elf_Word *>(Buf + Sec.Size - sizeof(Elf_Word)); 1159 *CRC = Sec.CRC32; 1160 llvm::copy(Sec.FileName, Buf); 1161 return Error::success(); 1162 } 1163 1164 Error GnuDebugLinkSection::accept(SectionVisitor &Visitor) const { 1165 return Visitor.visit(*this); 1166 } 1167 1168 Error GnuDebugLinkSection::accept(MutableSectionVisitor &Visitor) { 1169 return Visitor.visit(*this); 1170 } 1171 1172 template <class ELFT> 1173 Error ELFSectionWriter<ELFT>::visit(const GroupSection &Sec) { 1174 ELF::Elf32_Word *Buf = 1175 reinterpret_cast<ELF::Elf32_Word *>(Out.getBufferStart() + Sec.Offset); 1176 support::endian::write32<ELFT::TargetEndianness>(Buf++, Sec.FlagWord); 1177 for (SectionBase *S : Sec.GroupMembers) 1178 support::endian::write32<ELFT::TargetEndianness>(Buf++, S->Index); 1179 return Error::success(); 1180 } 1181 1182 Error GroupSection::accept(SectionVisitor &Visitor) const { 1183 return Visitor.visit(*this); 1184 } 1185 1186 Error GroupSection::accept(MutableSectionVisitor &Visitor) { 1187 return Visitor.visit(*this); 1188 } 1189 1190 // Returns true IFF a section is wholly inside the range of a segment 1191 static bool sectionWithinSegment(const SectionBase &Sec, const Segment &Seg) { 1192 // If a section is empty it should be treated like it has a size of 1. This is 1193 // to clarify the case when an empty section lies on a boundary between two 1194 // segments and ensures that the section "belongs" to the second segment and 1195 // not the first. 1196 uint64_t SecSize = Sec.Size ? Sec.Size : 1; 1197 1198 // Ignore just added sections. 1199 if (Sec.OriginalOffset == std::numeric_limits<uint64_t>::max()) 1200 return false; 1201 1202 if (Sec.Type == SHT_NOBITS) { 1203 if (!(Sec.Flags & SHF_ALLOC)) 1204 return false; 1205 1206 bool SectionIsTLS = Sec.Flags & SHF_TLS; 1207 bool SegmentIsTLS = Seg.Type == PT_TLS; 1208 if (SectionIsTLS != SegmentIsTLS) 1209 return false; 1210 1211 return Seg.VAddr <= Sec.Addr && 1212 Seg.VAddr + Seg.MemSize >= Sec.Addr + SecSize; 1213 } 1214 1215 return Seg.Offset <= Sec.OriginalOffset && 1216 Seg.Offset + Seg.FileSize >= Sec.OriginalOffset + SecSize; 1217 } 1218 1219 // Returns true IFF a segment's original offset is inside of another segment's 1220 // range. 1221 static bool segmentOverlapsSegment(const Segment &Child, 1222 const Segment &Parent) { 1223 1224 return Parent.OriginalOffset <= Child.OriginalOffset && 1225 Parent.OriginalOffset + Parent.FileSize > Child.OriginalOffset; 1226 } 1227 1228 static bool compareSegmentsByOffset(const Segment *A, const Segment *B) { 1229 // Any segment without a parent segment should come before a segment 1230 // that has a parent segment. 1231 if (A->OriginalOffset < B->OriginalOffset) 1232 return true; 1233 if (A->OriginalOffset > B->OriginalOffset) 1234 return false; 1235 return A->Index < B->Index; 1236 } 1237 1238 void BasicELFBuilder::initFileHeader() { 1239 Obj->Flags = 0x0; 1240 Obj->Type = ET_REL; 1241 Obj->OSABI = ELFOSABI_NONE; 1242 Obj->ABIVersion = 0; 1243 Obj->Entry = 0x0; 1244 Obj->Machine = EM_NONE; 1245 Obj->Version = 1; 1246 } 1247 1248 void BasicELFBuilder::initHeaderSegment() { Obj->ElfHdrSegment.Index = 0; } 1249 1250 StringTableSection *BasicELFBuilder::addStrTab() { 1251 auto &StrTab = Obj->addSection<StringTableSection>(); 1252 StrTab.Name = ".strtab"; 1253 1254 Obj->SectionNames = &StrTab; 1255 return &StrTab; 1256 } 1257 1258 SymbolTableSection *BasicELFBuilder::addSymTab(StringTableSection *StrTab) { 1259 auto &SymTab = Obj->addSection<SymbolTableSection>(); 1260 1261 SymTab.Name = ".symtab"; 1262 SymTab.Link = StrTab->Index; 1263 1264 // The symbol table always needs a null symbol 1265 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0); 1266 1267 Obj->SymbolTable = &SymTab; 1268 return &SymTab; 1269 } 1270 1271 Error BasicELFBuilder::initSections() { 1272 for (SectionBase &Sec : Obj->sections()) 1273 if (Error Err = Sec.initialize(Obj->sections())) 1274 return Err; 1275 1276 return Error::success(); 1277 } 1278 1279 void BinaryELFBuilder::addData(SymbolTableSection *SymTab) { 1280 auto Data = ArrayRef<uint8_t>( 1281 reinterpret_cast<const uint8_t *>(MemBuf->getBufferStart()), 1282 MemBuf->getBufferSize()); 1283 auto &DataSection = Obj->addSection<Section>(Data); 1284 DataSection.Name = ".data"; 1285 DataSection.Type = ELF::SHT_PROGBITS; 1286 DataSection.Size = Data.size(); 1287 DataSection.Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE; 1288 1289 std::string SanitizedFilename = MemBuf->getBufferIdentifier().str(); 1290 std::replace_if( 1291 std::begin(SanitizedFilename), std::end(SanitizedFilename), 1292 [](char C) { return !isAlnum(C); }, '_'); 1293 Twine Prefix = Twine("_binary_") + SanitizedFilename; 1294 1295 SymTab->addSymbol(Prefix + "_start", STB_GLOBAL, STT_NOTYPE, &DataSection, 1296 /*Value=*/0, NewSymbolVisibility, 0, 0); 1297 SymTab->addSymbol(Prefix + "_end", STB_GLOBAL, STT_NOTYPE, &DataSection, 1298 /*Value=*/DataSection.Size, NewSymbolVisibility, 0, 0); 1299 SymTab->addSymbol(Prefix + "_size", STB_GLOBAL, STT_NOTYPE, nullptr, 1300 /*Value=*/DataSection.Size, NewSymbolVisibility, SHN_ABS, 1301 0); 1302 } 1303 1304 Expected<std::unique_ptr<Object>> BinaryELFBuilder::build() { 1305 initFileHeader(); 1306 initHeaderSegment(); 1307 1308 SymbolTableSection *SymTab = addSymTab(addStrTab()); 1309 if (Error Err = initSections()) 1310 return std::move(Err); 1311 addData(SymTab); 1312 1313 return std::move(Obj); 1314 } 1315 1316 // Adds sections from IHEX data file. Data should have been 1317 // fully validated by this time. 1318 void IHexELFBuilder::addDataSections() { 1319 OwnedDataSection *Section = nullptr; 1320 uint64_t SegmentAddr = 0, BaseAddr = 0; 1321 uint32_t SecNo = 1; 1322 1323 for (const IHexRecord &R : Records) { 1324 uint64_t RecAddr; 1325 switch (R.Type) { 1326 case IHexRecord::Data: 1327 // Ignore empty data records 1328 if (R.HexData.empty()) 1329 continue; 1330 RecAddr = R.Addr + SegmentAddr + BaseAddr; 1331 if (!Section || Section->Addr + Section->Size != RecAddr) { 1332 // OriginalOffset field is only used to sort sections before layout, so 1333 // instead of keeping track of real offsets in IHEX file, and as 1334 // layoutSections() and layoutSectionsForOnlyKeepDebug() use 1335 // llvm::stable_sort(), we can just set it to a constant (zero). 1336 Section = &Obj->addSection<OwnedDataSection>( 1337 ".sec" + std::to_string(SecNo), RecAddr, 1338 ELF::SHF_ALLOC | ELF::SHF_WRITE, 0); 1339 SecNo++; 1340 } 1341 Section->appendHexData(R.HexData); 1342 break; 1343 case IHexRecord::EndOfFile: 1344 break; 1345 case IHexRecord::SegmentAddr: 1346 // 20-bit segment address. 1347 SegmentAddr = checkedGetHex<uint16_t>(R.HexData) << 4; 1348 break; 1349 case IHexRecord::StartAddr80x86: 1350 case IHexRecord::StartAddr: 1351 Obj->Entry = checkedGetHex<uint32_t>(R.HexData); 1352 assert(Obj->Entry <= 0xFFFFFU); 1353 break; 1354 case IHexRecord::ExtendedAddr: 1355 // 16-31 bits of linear base address 1356 BaseAddr = checkedGetHex<uint16_t>(R.HexData) << 16; 1357 break; 1358 default: 1359 llvm_unreachable("unknown record type"); 1360 } 1361 } 1362 } 1363 1364 Expected<std::unique_ptr<Object>> IHexELFBuilder::build() { 1365 initFileHeader(); 1366 initHeaderSegment(); 1367 StringTableSection *StrTab = addStrTab(); 1368 addSymTab(StrTab); 1369 if (Error Err = initSections()) 1370 return std::move(Err); 1371 addDataSections(); 1372 1373 return std::move(Obj); 1374 } 1375 1376 template <class ELFT> 1377 ELFBuilder<ELFT>::ELFBuilder(const ELFObjectFile<ELFT> &ElfObj, Object &Obj, 1378 Optional<StringRef> ExtractPartition) 1379 : ElfFile(ElfObj.getELFFile()), Obj(Obj), 1380 ExtractPartition(ExtractPartition) { 1381 Obj.IsMips64EL = ElfFile.isMips64EL(); 1382 } 1383 1384 template <class ELFT> void ELFBuilder<ELFT>::setParentSegment(Segment &Child) { 1385 for (Segment &Parent : Obj.segments()) { 1386 // Every segment will overlap with itself but we don't want a segment to 1387 // be its own parent so we avoid that situation. 1388 if (&Child != &Parent && segmentOverlapsSegment(Child, Parent)) { 1389 // We want a canonical "most parental" segment but this requires 1390 // inspecting the ParentSegment. 1391 if (compareSegmentsByOffset(&Parent, &Child)) 1392 if (Child.ParentSegment == nullptr || 1393 compareSegmentsByOffset(&Parent, Child.ParentSegment)) { 1394 Child.ParentSegment = &Parent; 1395 } 1396 } 1397 } 1398 } 1399 1400 template <class ELFT> Error ELFBuilder<ELFT>::findEhdrOffset() { 1401 if (!ExtractPartition) 1402 return Error::success(); 1403 1404 for (const SectionBase &Sec : Obj.sections()) { 1405 if (Sec.Type == SHT_LLVM_PART_EHDR && Sec.Name == *ExtractPartition) { 1406 EhdrOffset = Sec.Offset; 1407 return Error::success(); 1408 } 1409 } 1410 return createStringError(errc::invalid_argument, 1411 "could not find partition named '" + 1412 *ExtractPartition + "'"); 1413 } 1414 1415 template <class ELFT> 1416 Error ELFBuilder<ELFT>::readProgramHeaders(const ELFFile<ELFT> &HeadersFile) { 1417 uint32_t Index = 0; 1418 1419 Expected<typename ELFFile<ELFT>::Elf_Phdr_Range> Headers = 1420 HeadersFile.program_headers(); 1421 if (!Headers) 1422 return Headers.takeError(); 1423 1424 for (const typename ELFFile<ELFT>::Elf_Phdr &Phdr : *Headers) { 1425 if (Phdr.p_offset + Phdr.p_filesz > HeadersFile.getBufSize()) 1426 return createStringError( 1427 errc::invalid_argument, 1428 "program header with offset 0x" + Twine::utohexstr(Phdr.p_offset) + 1429 " and file size 0x" + Twine::utohexstr(Phdr.p_filesz) + 1430 " goes past the end of the file"); 1431 1432 ArrayRef<uint8_t> Data{HeadersFile.base() + Phdr.p_offset, 1433 (size_t)Phdr.p_filesz}; 1434 Segment &Seg = Obj.addSegment(Data); 1435 Seg.Type = Phdr.p_type; 1436 Seg.Flags = Phdr.p_flags; 1437 Seg.OriginalOffset = Phdr.p_offset + EhdrOffset; 1438 Seg.Offset = Phdr.p_offset + EhdrOffset; 1439 Seg.VAddr = Phdr.p_vaddr; 1440 Seg.PAddr = Phdr.p_paddr; 1441 Seg.FileSize = Phdr.p_filesz; 1442 Seg.MemSize = Phdr.p_memsz; 1443 Seg.Align = Phdr.p_align; 1444 Seg.Index = Index++; 1445 for (SectionBase &Sec : Obj.sections()) 1446 if (sectionWithinSegment(Sec, Seg)) { 1447 Seg.addSection(&Sec); 1448 if (!Sec.ParentSegment || Sec.ParentSegment->Offset > Seg.Offset) 1449 Sec.ParentSegment = &Seg; 1450 } 1451 } 1452 1453 auto &ElfHdr = Obj.ElfHdrSegment; 1454 ElfHdr.Index = Index++; 1455 ElfHdr.OriginalOffset = ElfHdr.Offset = EhdrOffset; 1456 1457 const typename ELFT::Ehdr &Ehdr = HeadersFile.getHeader(); 1458 auto &PrHdr = Obj.ProgramHdrSegment; 1459 PrHdr.Type = PT_PHDR; 1460 PrHdr.Flags = 0; 1461 // The spec requires us to have p_vaddr % p_align == p_offset % p_align. 1462 // Whereas this works automatically for ElfHdr, here OriginalOffset is 1463 // always non-zero and to ensure the equation we assign the same value to 1464 // VAddr as well. 1465 PrHdr.OriginalOffset = PrHdr.Offset = PrHdr.VAddr = EhdrOffset + Ehdr.e_phoff; 1466 PrHdr.PAddr = 0; 1467 PrHdr.FileSize = PrHdr.MemSize = Ehdr.e_phentsize * Ehdr.e_phnum; 1468 // The spec requires us to naturally align all the fields. 1469 PrHdr.Align = sizeof(Elf_Addr); 1470 PrHdr.Index = Index++; 1471 1472 // Now we do an O(n^2) loop through the segments in order to match up 1473 // segments. 1474 for (Segment &Child : Obj.segments()) 1475 setParentSegment(Child); 1476 setParentSegment(ElfHdr); 1477 setParentSegment(PrHdr); 1478 1479 return Error::success(); 1480 } 1481 1482 template <class ELFT> 1483 Error ELFBuilder<ELFT>::initGroupSection(GroupSection *GroupSec) { 1484 if (GroupSec->Align % sizeof(ELF::Elf32_Word) != 0) 1485 return createStringError(errc::invalid_argument, 1486 "invalid alignment " + Twine(GroupSec->Align) + 1487 " of group section '" + GroupSec->Name + "'"); 1488 SectionTableRef SecTable = Obj.sections(); 1489 if (GroupSec->Link != SHN_UNDEF) { 1490 auto SymTab = SecTable.template getSectionOfType<SymbolTableSection>( 1491 GroupSec->Link, 1492 "link field value '" + Twine(GroupSec->Link) + "' in section '" + 1493 GroupSec->Name + "' is invalid", 1494 "link field value '" + Twine(GroupSec->Link) + "' in section '" + 1495 GroupSec->Name + "' is not a symbol table"); 1496 if (!SymTab) 1497 return SymTab.takeError(); 1498 1499 Expected<Symbol *> Sym = (*SymTab)->getSymbolByIndex(GroupSec->Info); 1500 if (!Sym) 1501 return createStringError(errc::invalid_argument, 1502 "info field value '" + Twine(GroupSec->Info) + 1503 "' in section '" + GroupSec->Name + 1504 "' is not a valid symbol index"); 1505 GroupSec->setSymTab(*SymTab); 1506 GroupSec->setSymbol(*Sym); 1507 } 1508 if (GroupSec->Contents.size() % sizeof(ELF::Elf32_Word) || 1509 GroupSec->Contents.empty()) 1510 return createStringError(errc::invalid_argument, 1511 "the content of the section " + GroupSec->Name + 1512 " is malformed"); 1513 const ELF::Elf32_Word *Word = 1514 reinterpret_cast<const ELF::Elf32_Word *>(GroupSec->Contents.data()); 1515 const ELF::Elf32_Word *End = 1516 Word + GroupSec->Contents.size() / sizeof(ELF::Elf32_Word); 1517 GroupSec->setFlagWord( 1518 support::endian::read32<ELFT::TargetEndianness>(Word++)); 1519 for (; Word != End; ++Word) { 1520 uint32_t Index = support::endian::read32<ELFT::TargetEndianness>(Word); 1521 Expected<SectionBase *> Sec = SecTable.getSection( 1522 Index, "group member index " + Twine(Index) + " in section '" + 1523 GroupSec->Name + "' is invalid"); 1524 if (!Sec) 1525 return Sec.takeError(); 1526 1527 GroupSec->addMember(*Sec); 1528 } 1529 1530 return Error::success(); 1531 } 1532 1533 template <class ELFT> 1534 Error ELFBuilder<ELFT>::initSymbolTable(SymbolTableSection *SymTab) { 1535 Expected<const Elf_Shdr *> Shdr = ElfFile.getSection(SymTab->Index); 1536 if (!Shdr) 1537 return Shdr.takeError(); 1538 1539 Expected<StringRef> StrTabData = ElfFile.getStringTableForSymtab(**Shdr); 1540 if (!StrTabData) 1541 return StrTabData.takeError(); 1542 1543 ArrayRef<Elf_Word> ShndxData; 1544 1545 Expected<typename ELFFile<ELFT>::Elf_Sym_Range> Symbols = 1546 ElfFile.symbols(*Shdr); 1547 if (!Symbols) 1548 return Symbols.takeError(); 1549 1550 for (const typename ELFFile<ELFT>::Elf_Sym &Sym : *Symbols) { 1551 SectionBase *DefSection = nullptr; 1552 1553 Expected<StringRef> Name = Sym.getName(*StrTabData); 1554 if (!Name) 1555 return Name.takeError(); 1556 1557 if (Sym.st_shndx == SHN_XINDEX) { 1558 if (SymTab->getShndxTable() == nullptr) 1559 return createStringError(errc::invalid_argument, 1560 "symbol '" + *Name + 1561 "' has index SHN_XINDEX but no " 1562 "SHT_SYMTAB_SHNDX section exists"); 1563 if (ShndxData.data() == nullptr) { 1564 Expected<const Elf_Shdr *> ShndxSec = 1565 ElfFile.getSection(SymTab->getShndxTable()->Index); 1566 if (!ShndxSec) 1567 return ShndxSec.takeError(); 1568 1569 Expected<ArrayRef<Elf_Word>> Data = 1570 ElfFile.template getSectionContentsAsArray<Elf_Word>(**ShndxSec); 1571 if (!Data) 1572 return Data.takeError(); 1573 1574 ShndxData = *Data; 1575 if (ShndxData.size() != Symbols->size()) 1576 return createStringError( 1577 errc::invalid_argument, 1578 "symbol section index table does not have the same number of " 1579 "entries as the symbol table"); 1580 } 1581 Elf_Word Index = ShndxData[&Sym - Symbols->begin()]; 1582 Expected<SectionBase *> Sec = Obj.sections().getSection( 1583 Index, 1584 "symbol '" + *Name + "' has invalid section index " + Twine(Index)); 1585 if (!Sec) 1586 return Sec.takeError(); 1587 1588 DefSection = *Sec; 1589 } else if (Sym.st_shndx >= SHN_LORESERVE) { 1590 if (!isValidReservedSectionIndex(Sym.st_shndx, Obj.Machine)) { 1591 return createStringError( 1592 errc::invalid_argument, 1593 "symbol '" + *Name + 1594 "' has unsupported value greater than or equal " 1595 "to SHN_LORESERVE: " + 1596 Twine(Sym.st_shndx)); 1597 } 1598 } else if (Sym.st_shndx != SHN_UNDEF) { 1599 Expected<SectionBase *> Sec = Obj.sections().getSection( 1600 Sym.st_shndx, "symbol '" + *Name + 1601 "' is defined has invalid section index " + 1602 Twine(Sym.st_shndx)); 1603 if (!Sec) 1604 return Sec.takeError(); 1605 1606 DefSection = *Sec; 1607 } 1608 1609 SymTab->addSymbol(*Name, Sym.getBinding(), Sym.getType(), DefSection, 1610 Sym.getValue(), Sym.st_other, Sym.st_shndx, Sym.st_size); 1611 } 1612 1613 return Error::success(); 1614 } 1615 1616 template <class ELFT> 1617 static void getAddend(uint64_t &, const Elf_Rel_Impl<ELFT, false> &) {} 1618 1619 template <class ELFT> 1620 static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, true> &Rela) { 1621 ToSet = Rela.r_addend; 1622 } 1623 1624 template <class T> 1625 static Error initRelocations(RelocationSection *Relocs, T RelRange) { 1626 for (const auto &Rel : RelRange) { 1627 Relocation ToAdd; 1628 ToAdd.Offset = Rel.r_offset; 1629 getAddend(ToAdd.Addend, Rel); 1630 ToAdd.Type = Rel.getType(Relocs->getObject().IsMips64EL); 1631 1632 if (uint32_t Sym = Rel.getSymbol(Relocs->getObject().IsMips64EL)) { 1633 if (!Relocs->getObject().SymbolTable) 1634 return createStringError( 1635 errc::invalid_argument, 1636 "'" + Relocs->Name + "': relocation references symbol with index " + 1637 Twine(Sym) + ", but there is no symbol table"); 1638 Expected<Symbol *> SymByIndex = 1639 Relocs->getObject().SymbolTable->getSymbolByIndex(Sym); 1640 if (!SymByIndex) 1641 return SymByIndex.takeError(); 1642 1643 ToAdd.RelocSymbol = *SymByIndex; 1644 } 1645 1646 Relocs->addRelocation(ToAdd); 1647 } 1648 1649 return Error::success(); 1650 } 1651 1652 Expected<SectionBase *> SectionTableRef::getSection(uint32_t Index, 1653 Twine ErrMsg) { 1654 if (Index == SHN_UNDEF || Index > Sections.size()) 1655 return createStringError(errc::invalid_argument, ErrMsg); 1656 return Sections[Index - 1].get(); 1657 } 1658 1659 template <class T> 1660 Expected<T *> SectionTableRef::getSectionOfType(uint32_t Index, 1661 Twine IndexErrMsg, 1662 Twine TypeErrMsg) { 1663 Expected<SectionBase *> BaseSec = getSection(Index, IndexErrMsg); 1664 if (!BaseSec) 1665 return BaseSec.takeError(); 1666 1667 if (T *Sec = dyn_cast<T>(*BaseSec)) 1668 return Sec; 1669 1670 return createStringError(errc::invalid_argument, TypeErrMsg); 1671 } 1672 1673 template <class ELFT> 1674 Expected<SectionBase &> ELFBuilder<ELFT>::makeSection(const Elf_Shdr &Shdr) { 1675 switch (Shdr.sh_type) { 1676 case SHT_REL: 1677 case SHT_RELA: 1678 if (Shdr.sh_flags & SHF_ALLOC) { 1679 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1680 return Obj.addSection<DynamicRelocationSection>(*Data); 1681 else 1682 return Data.takeError(); 1683 } 1684 return Obj.addSection<RelocationSection>(Obj); 1685 case SHT_STRTAB: 1686 // If a string table is allocated we don't want to mess with it. That would 1687 // mean altering the memory image. There are no special link types or 1688 // anything so we can just use a Section. 1689 if (Shdr.sh_flags & SHF_ALLOC) { 1690 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1691 return Obj.addSection<Section>(*Data); 1692 else 1693 return Data.takeError(); 1694 } 1695 return Obj.addSection<StringTableSection>(); 1696 case SHT_HASH: 1697 case SHT_GNU_HASH: 1698 // Hash tables should refer to SHT_DYNSYM which we're not going to change. 1699 // Because of this we don't need to mess with the hash tables either. 1700 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1701 return Obj.addSection<Section>(*Data); 1702 else 1703 return Data.takeError(); 1704 case SHT_GROUP: 1705 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1706 return Obj.addSection<GroupSection>(*Data); 1707 else 1708 return Data.takeError(); 1709 case SHT_DYNSYM: 1710 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1711 return Obj.addSection<DynamicSymbolTableSection>(*Data); 1712 else 1713 return Data.takeError(); 1714 case SHT_DYNAMIC: 1715 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1716 return Obj.addSection<DynamicSection>(*Data); 1717 else 1718 return Data.takeError(); 1719 case SHT_SYMTAB: { 1720 auto &SymTab = Obj.addSection<SymbolTableSection>(); 1721 Obj.SymbolTable = &SymTab; 1722 return SymTab; 1723 } 1724 case SHT_SYMTAB_SHNDX: { 1725 auto &ShndxSection = Obj.addSection<SectionIndexSection>(); 1726 Obj.SectionIndexTable = &ShndxSection; 1727 return ShndxSection; 1728 } 1729 case SHT_NOBITS: 1730 return Obj.addSection<Section>(ArrayRef<uint8_t>()); 1731 default: { 1732 Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr); 1733 if (!Data) 1734 return Data.takeError(); 1735 1736 Expected<StringRef> Name = ElfFile.getSectionName(Shdr); 1737 if (!Name) 1738 return Name.takeError(); 1739 1740 if (Name->startswith(".zdebug") || (Shdr.sh_flags & ELF::SHF_COMPRESSED)) { 1741 uint64_t DecompressedSize, DecompressedAlign; 1742 std::tie(DecompressedSize, DecompressedAlign) = 1743 getDecompressedSizeAndAlignment<ELFT>(*Data); 1744 return Obj.addSection<CompressedSection>( 1745 CompressedSection(*Data, DecompressedSize, DecompressedAlign)); 1746 } 1747 1748 return Obj.addSection<Section>(*Data); 1749 } 1750 } 1751 } 1752 1753 template <class ELFT> Error ELFBuilder<ELFT>::readSectionHeaders() { 1754 uint32_t Index = 0; 1755 Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections = 1756 ElfFile.sections(); 1757 if (!Sections) 1758 return Sections.takeError(); 1759 1760 for (const typename ELFFile<ELFT>::Elf_Shdr &Shdr : *Sections) { 1761 if (Index == 0) { 1762 ++Index; 1763 continue; 1764 } 1765 Expected<SectionBase &> Sec = makeSection(Shdr); 1766 if (!Sec) 1767 return Sec.takeError(); 1768 1769 Expected<StringRef> SecName = ElfFile.getSectionName(Shdr); 1770 if (!SecName) 1771 return SecName.takeError(); 1772 Sec->Name = SecName->str(); 1773 Sec->Type = Sec->OriginalType = Shdr.sh_type; 1774 Sec->Flags = Sec->OriginalFlags = Shdr.sh_flags; 1775 Sec->Addr = Shdr.sh_addr; 1776 Sec->Offset = Shdr.sh_offset; 1777 Sec->OriginalOffset = Shdr.sh_offset; 1778 Sec->Size = Shdr.sh_size; 1779 Sec->Link = Shdr.sh_link; 1780 Sec->Info = Shdr.sh_info; 1781 Sec->Align = Shdr.sh_addralign; 1782 Sec->EntrySize = Shdr.sh_entsize; 1783 Sec->Index = Index++; 1784 Sec->OriginalIndex = Sec->Index; 1785 Sec->OriginalData = ArrayRef<uint8_t>( 1786 ElfFile.base() + Shdr.sh_offset, 1787 (Shdr.sh_type == SHT_NOBITS) ? (size_t)0 : Shdr.sh_size); 1788 } 1789 1790 return Error::success(); 1791 } 1792 1793 template <class ELFT> Error ELFBuilder<ELFT>::readSections(bool EnsureSymtab) { 1794 uint32_t ShstrIndex = ElfFile.getHeader().e_shstrndx; 1795 if (ShstrIndex == SHN_XINDEX) { 1796 Expected<const Elf_Shdr *> Sec = ElfFile.getSection(0); 1797 if (!Sec) 1798 return Sec.takeError(); 1799 1800 ShstrIndex = (*Sec)->sh_link; 1801 } 1802 1803 if (ShstrIndex == SHN_UNDEF) 1804 Obj.HadShdrs = false; 1805 else { 1806 Expected<StringTableSection *> Sec = 1807 Obj.sections().template getSectionOfType<StringTableSection>( 1808 ShstrIndex, 1809 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " + 1810 " is invalid", 1811 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " + 1812 " does not reference a string table"); 1813 if (!Sec) 1814 return Sec.takeError(); 1815 1816 Obj.SectionNames = *Sec; 1817 } 1818 1819 // If a section index table exists we'll need to initialize it before we 1820 // initialize the symbol table because the symbol table might need to 1821 // reference it. 1822 if (Obj.SectionIndexTable) 1823 if (Error Err = Obj.SectionIndexTable->initialize(Obj.sections())) 1824 return Err; 1825 1826 // Now that all of the sections have been added we can fill out some extra 1827 // details about symbol tables. We need the symbol table filled out before 1828 // any relocations. 1829 if (Obj.SymbolTable) { 1830 if (Error Err = Obj.SymbolTable->initialize(Obj.sections())) 1831 return Err; 1832 if (Error Err = initSymbolTable(Obj.SymbolTable)) 1833 return Err; 1834 } else if (EnsureSymtab) { 1835 if (Error Err = Obj.addNewSymbolTable()) 1836 return Err; 1837 } 1838 1839 // Now that all sections and symbols have been added we can add 1840 // relocations that reference symbols and set the link and info fields for 1841 // relocation sections. 1842 for (SectionBase &Sec : Obj.sections()) { 1843 if (&Sec == Obj.SymbolTable) 1844 continue; 1845 if (Error Err = Sec.initialize(Obj.sections())) 1846 return Err; 1847 if (auto RelSec = dyn_cast<RelocationSection>(&Sec)) { 1848 Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections = 1849 ElfFile.sections(); 1850 if (!Sections) 1851 return Sections.takeError(); 1852 1853 const typename ELFFile<ELFT>::Elf_Shdr *Shdr = 1854 Sections->begin() + RelSec->Index; 1855 if (RelSec->Type == SHT_REL) { 1856 Expected<typename ELFFile<ELFT>::Elf_Rel_Range> Rels = 1857 ElfFile.rels(*Shdr); 1858 if (!Rels) 1859 return Rels.takeError(); 1860 1861 if (Error Err = initRelocations(RelSec, *Rels)) 1862 return Err; 1863 } else { 1864 Expected<typename ELFFile<ELFT>::Elf_Rela_Range> Relas = 1865 ElfFile.relas(*Shdr); 1866 if (!Relas) 1867 return Relas.takeError(); 1868 1869 if (Error Err = initRelocations(RelSec, *Relas)) 1870 return Err; 1871 } 1872 } else if (auto GroupSec = dyn_cast<GroupSection>(&Sec)) { 1873 if (Error Err = initGroupSection(GroupSec)) 1874 return Err; 1875 } 1876 } 1877 1878 return Error::success(); 1879 } 1880 1881 template <class ELFT> Error ELFBuilder<ELFT>::build(bool EnsureSymtab) { 1882 if (Error E = readSectionHeaders()) 1883 return E; 1884 if (Error E = findEhdrOffset()) 1885 return E; 1886 1887 // The ELFFile whose ELF headers and program headers are copied into the 1888 // output file. Normally the same as ElfFile, but if we're extracting a 1889 // loadable partition it will point to the partition's headers. 1890 Expected<ELFFile<ELFT>> HeadersFile = ELFFile<ELFT>::create(toStringRef( 1891 {ElfFile.base() + EhdrOffset, ElfFile.getBufSize() - EhdrOffset})); 1892 if (!HeadersFile) 1893 return HeadersFile.takeError(); 1894 1895 const typename ELFFile<ELFT>::Elf_Ehdr &Ehdr = HeadersFile->getHeader(); 1896 Obj.OSABI = Ehdr.e_ident[EI_OSABI]; 1897 Obj.ABIVersion = Ehdr.e_ident[EI_ABIVERSION]; 1898 Obj.Type = Ehdr.e_type; 1899 Obj.Machine = Ehdr.e_machine; 1900 Obj.Version = Ehdr.e_version; 1901 Obj.Entry = Ehdr.e_entry; 1902 Obj.Flags = Ehdr.e_flags; 1903 1904 if (Error E = readSections(EnsureSymtab)) 1905 return E; 1906 return readProgramHeaders(*HeadersFile); 1907 } 1908 1909 Writer::~Writer() = default; 1910 1911 Reader::~Reader() = default; 1912 1913 Expected<std::unique_ptr<Object>> 1914 BinaryReader::create(bool /*EnsureSymtab*/) const { 1915 return BinaryELFBuilder(MemBuf, NewSymbolVisibility).build(); 1916 } 1917 1918 Expected<std::vector<IHexRecord>> IHexReader::parse() const { 1919 SmallVector<StringRef, 16> Lines; 1920 std::vector<IHexRecord> Records; 1921 bool HasSections = false; 1922 1923 MemBuf->getBuffer().split(Lines, '\n'); 1924 Records.reserve(Lines.size()); 1925 for (size_t LineNo = 1; LineNo <= Lines.size(); ++LineNo) { 1926 StringRef Line = Lines[LineNo - 1].trim(); 1927 if (Line.empty()) 1928 continue; 1929 1930 Expected<IHexRecord> R = IHexRecord::parse(Line); 1931 if (!R) 1932 return parseError(LineNo, R.takeError()); 1933 if (R->Type == IHexRecord::EndOfFile) 1934 break; 1935 HasSections |= (R->Type == IHexRecord::Data); 1936 Records.push_back(*R); 1937 } 1938 if (!HasSections) 1939 return parseError(-1U, "no sections"); 1940 1941 return std::move(Records); 1942 } 1943 1944 Expected<std::unique_ptr<Object>> 1945 IHexReader::create(bool /*EnsureSymtab*/) const { 1946 Expected<std::vector<IHexRecord>> Records = parse(); 1947 if (!Records) 1948 return Records.takeError(); 1949 1950 return IHexELFBuilder(*Records).build(); 1951 } 1952 1953 Expected<std::unique_ptr<Object>> ELFReader::create(bool EnsureSymtab) const { 1954 auto Obj = std::make_unique<Object>(); 1955 if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Bin)) { 1956 ELFBuilder<ELF32LE> Builder(*O, *Obj, ExtractPartition); 1957 if (Error Err = Builder.build(EnsureSymtab)) 1958 return std::move(Err); 1959 return std::move(Obj); 1960 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Bin)) { 1961 ELFBuilder<ELF64LE> Builder(*O, *Obj, ExtractPartition); 1962 if (Error Err = Builder.build(EnsureSymtab)) 1963 return std::move(Err); 1964 return std::move(Obj); 1965 } else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Bin)) { 1966 ELFBuilder<ELF32BE> Builder(*O, *Obj, ExtractPartition); 1967 if (Error Err = Builder.build(EnsureSymtab)) 1968 return std::move(Err); 1969 return std::move(Obj); 1970 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Bin)) { 1971 ELFBuilder<ELF64BE> Builder(*O, *Obj, ExtractPartition); 1972 if (Error Err = Builder.build(EnsureSymtab)) 1973 return std::move(Err); 1974 return std::move(Obj); 1975 } 1976 return createStringError(errc::invalid_argument, "invalid file type"); 1977 } 1978 1979 template <class ELFT> void ELFWriter<ELFT>::writeEhdr() { 1980 Elf_Ehdr &Ehdr = *reinterpret_cast<Elf_Ehdr *>(Buf->getBufferStart()); 1981 std::fill(Ehdr.e_ident, Ehdr.e_ident + 16, 0); 1982 Ehdr.e_ident[EI_MAG0] = 0x7f; 1983 Ehdr.e_ident[EI_MAG1] = 'E'; 1984 Ehdr.e_ident[EI_MAG2] = 'L'; 1985 Ehdr.e_ident[EI_MAG3] = 'F'; 1986 Ehdr.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32; 1987 Ehdr.e_ident[EI_DATA] = 1988 ELFT::TargetEndianness == support::big ? ELFDATA2MSB : ELFDATA2LSB; 1989 Ehdr.e_ident[EI_VERSION] = EV_CURRENT; 1990 Ehdr.e_ident[EI_OSABI] = Obj.OSABI; 1991 Ehdr.e_ident[EI_ABIVERSION] = Obj.ABIVersion; 1992 1993 Ehdr.e_type = Obj.Type; 1994 Ehdr.e_machine = Obj.Machine; 1995 Ehdr.e_version = Obj.Version; 1996 Ehdr.e_entry = Obj.Entry; 1997 // We have to use the fully-qualified name llvm::size 1998 // since some compilers complain on ambiguous resolution. 1999 Ehdr.e_phnum = llvm::size(Obj.segments()); 2000 Ehdr.e_phoff = (Ehdr.e_phnum != 0) ? Obj.ProgramHdrSegment.Offset : 0; 2001 Ehdr.e_phentsize = (Ehdr.e_phnum != 0) ? sizeof(Elf_Phdr) : 0; 2002 Ehdr.e_flags = Obj.Flags; 2003 Ehdr.e_ehsize = sizeof(Elf_Ehdr); 2004 if (WriteSectionHeaders && Obj.sections().size() != 0) { 2005 Ehdr.e_shentsize = sizeof(Elf_Shdr); 2006 Ehdr.e_shoff = Obj.SHOff; 2007 // """ 2008 // If the number of sections is greater than or equal to 2009 // SHN_LORESERVE (0xff00), this member has the value zero and the actual 2010 // number of section header table entries is contained in the sh_size field 2011 // of the section header at index 0. 2012 // """ 2013 auto Shnum = Obj.sections().size() + 1; 2014 if (Shnum >= SHN_LORESERVE) 2015 Ehdr.e_shnum = 0; 2016 else 2017 Ehdr.e_shnum = Shnum; 2018 // """ 2019 // If the section name string table section index is greater than or equal 2020 // to SHN_LORESERVE (0xff00), this member has the value SHN_XINDEX (0xffff) 2021 // and the actual index of the section name string table section is 2022 // contained in the sh_link field of the section header at index 0. 2023 // """ 2024 if (Obj.SectionNames->Index >= SHN_LORESERVE) 2025 Ehdr.e_shstrndx = SHN_XINDEX; 2026 else 2027 Ehdr.e_shstrndx = Obj.SectionNames->Index; 2028 } else { 2029 Ehdr.e_shentsize = 0; 2030 Ehdr.e_shoff = 0; 2031 Ehdr.e_shnum = 0; 2032 Ehdr.e_shstrndx = 0; 2033 } 2034 } 2035 2036 template <class ELFT> void ELFWriter<ELFT>::writePhdrs() { 2037 for (auto &Seg : Obj.segments()) 2038 writePhdr(Seg); 2039 } 2040 2041 template <class ELFT> void ELFWriter<ELFT>::writeShdrs() { 2042 // This reference serves to write the dummy section header at the begining 2043 // of the file. It is not used for anything else 2044 Elf_Shdr &Shdr = 2045 *reinterpret_cast<Elf_Shdr *>(Buf->getBufferStart() + Obj.SHOff); 2046 Shdr.sh_name = 0; 2047 Shdr.sh_type = SHT_NULL; 2048 Shdr.sh_flags = 0; 2049 Shdr.sh_addr = 0; 2050 Shdr.sh_offset = 0; 2051 // See writeEhdr for why we do this. 2052 uint64_t Shnum = Obj.sections().size() + 1; 2053 if (Shnum >= SHN_LORESERVE) 2054 Shdr.sh_size = Shnum; 2055 else 2056 Shdr.sh_size = 0; 2057 // See writeEhdr for why we do this. 2058 if (Obj.SectionNames != nullptr && Obj.SectionNames->Index >= SHN_LORESERVE) 2059 Shdr.sh_link = Obj.SectionNames->Index; 2060 else 2061 Shdr.sh_link = 0; 2062 Shdr.sh_info = 0; 2063 Shdr.sh_addralign = 0; 2064 Shdr.sh_entsize = 0; 2065 2066 for (SectionBase &Sec : Obj.sections()) 2067 writeShdr(Sec); 2068 } 2069 2070 template <class ELFT> Error ELFWriter<ELFT>::writeSectionData() { 2071 for (SectionBase &Sec : Obj.sections()) 2072 // Segments are responsible for writing their contents, so only write the 2073 // section data if the section is not in a segment. Note that this renders 2074 // sections in segments effectively immutable. 2075 if (Sec.ParentSegment == nullptr) 2076 if (Error Err = Sec.accept(*SecWriter)) 2077 return Err; 2078 2079 return Error::success(); 2080 } 2081 2082 template <class ELFT> void ELFWriter<ELFT>::writeSegmentData() { 2083 for (Segment &Seg : Obj.segments()) { 2084 size_t Size = std::min<size_t>(Seg.FileSize, Seg.getContents().size()); 2085 std::memcpy(Buf->getBufferStart() + Seg.Offset, Seg.getContents().data(), 2086 Size); 2087 } 2088 2089 for (auto it : Obj.getUpdatedSections()) { 2090 SectionBase *Sec = it.first; 2091 ArrayRef<uint8_t> Data = it.second; 2092 2093 auto *Parent = Sec->ParentSegment; 2094 assert(Parent && "This section should've been part of a segment."); 2095 uint64_t Offset = 2096 Sec->OriginalOffset - Parent->OriginalOffset + Parent->Offset; 2097 llvm::copy(Data, Buf->getBufferStart() + Offset); 2098 } 2099 2100 // Iterate over removed sections and overwrite their old data with zeroes. 2101 for (auto &Sec : Obj.removedSections()) { 2102 Segment *Parent = Sec.ParentSegment; 2103 if (Parent == nullptr || Sec.Type == SHT_NOBITS || Sec.Size == 0) 2104 continue; 2105 uint64_t Offset = 2106 Sec.OriginalOffset - Parent->OriginalOffset + Parent->Offset; 2107 std::memset(Buf->getBufferStart() + Offset, 0, Sec.Size); 2108 } 2109 } 2110 2111 template <class ELFT> 2112 ELFWriter<ELFT>::ELFWriter(Object &Obj, raw_ostream &Buf, bool WSH, 2113 bool OnlyKeepDebug) 2114 : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs), 2115 OnlyKeepDebug(OnlyKeepDebug) {} 2116 2117 Error Object::updateSection(StringRef Name, ArrayRef<uint8_t> Data) { 2118 auto It = llvm::find_if(Sections, 2119 [&](const SecPtr &Sec) { return Sec->Name == Name; }); 2120 if (It == Sections.end()) 2121 return createStringError(errc::invalid_argument, "section '%s' not found", 2122 Name.str().c_str()); 2123 2124 auto *OldSec = It->get(); 2125 if (!OldSec->hasContents()) 2126 return createStringError( 2127 errc::invalid_argument, 2128 "section '%s' cannot be updated because it does not have contents", 2129 Name.str().c_str()); 2130 2131 if (Data.size() > OldSec->Size && OldSec->ParentSegment) 2132 return createStringError(errc::invalid_argument, 2133 "cannot fit data of size %zu into section '%s' " 2134 "with size %zu that is part of a segment", 2135 Data.size(), Name.str().c_str(), OldSec->Size); 2136 2137 if (!OldSec->ParentSegment) { 2138 *It = std::make_unique<OwnedDataSection>(*OldSec, Data); 2139 } else { 2140 // The segment writer will be in charge of updating these contents. 2141 OldSec->Size = Data.size(); 2142 UpdatedSections[OldSec] = Data; 2143 } 2144 2145 return Error::success(); 2146 } 2147 2148 Error Object::removeSections( 2149 bool AllowBrokenLinks, std::function<bool(const SectionBase &)> ToRemove) { 2150 2151 auto Iter = std::stable_partition( 2152 std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) { 2153 if (ToRemove(*Sec)) 2154 return false; 2155 if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) { 2156 if (auto ToRelSec = RelSec->getSection()) 2157 return !ToRemove(*ToRelSec); 2158 } 2159 return true; 2160 }); 2161 if (SymbolTable != nullptr && ToRemove(*SymbolTable)) 2162 SymbolTable = nullptr; 2163 if (SectionNames != nullptr && ToRemove(*SectionNames)) 2164 SectionNames = nullptr; 2165 if (SectionIndexTable != nullptr && ToRemove(*SectionIndexTable)) 2166 SectionIndexTable = nullptr; 2167 // Now make sure there are no remaining references to the sections that will 2168 // be removed. Sometimes it is impossible to remove a reference so we emit 2169 // an error here instead. 2170 std::unordered_set<const SectionBase *> RemoveSections; 2171 RemoveSections.reserve(std::distance(Iter, std::end(Sections))); 2172 for (auto &RemoveSec : make_range(Iter, std::end(Sections))) { 2173 for (auto &Segment : Segments) 2174 Segment->removeSection(RemoveSec.get()); 2175 RemoveSec->onRemove(); 2176 RemoveSections.insert(RemoveSec.get()); 2177 } 2178 2179 // For each section that remains alive, we want to remove the dead references. 2180 // This either might update the content of the section (e.g. remove symbols 2181 // from symbol table that belongs to removed section) or trigger an error if 2182 // a live section critically depends on a section being removed somehow 2183 // (e.g. the removed section is referenced by a relocation). 2184 for (auto &KeepSec : make_range(std::begin(Sections), Iter)) { 2185 if (Error E = KeepSec->removeSectionReferences( 2186 AllowBrokenLinks, [&RemoveSections](const SectionBase *Sec) { 2187 return RemoveSections.find(Sec) != RemoveSections.end(); 2188 })) 2189 return E; 2190 } 2191 2192 // Transfer removed sections into the Object RemovedSections container for use 2193 // later. 2194 std::move(Iter, Sections.end(), std::back_inserter(RemovedSections)); 2195 // Now finally get rid of them all together. 2196 Sections.erase(Iter, std::end(Sections)); 2197 return Error::success(); 2198 } 2199 2200 Error Object::replaceSections( 2201 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 2202 auto SectionIndexLess = [](const SecPtr &Lhs, const SecPtr &Rhs) { 2203 return Lhs->Index < Rhs->Index; 2204 }; 2205 assert(llvm::is_sorted(Sections, SectionIndexLess) && 2206 "Sections are expected to be sorted by Index"); 2207 // Set indices of new sections so that they can be later sorted into positions 2208 // of removed ones. 2209 for (auto &I : FromTo) 2210 I.second->Index = I.first->Index; 2211 2212 // Notify all sections about the replacement. 2213 for (auto &Sec : Sections) 2214 Sec->replaceSectionReferences(FromTo); 2215 2216 if (Error E = removeSections( 2217 /*AllowBrokenLinks=*/false, 2218 [=](const SectionBase &Sec) { return FromTo.count(&Sec) > 0; })) 2219 return E; 2220 llvm::sort(Sections, SectionIndexLess); 2221 return Error::success(); 2222 } 2223 2224 Error Object::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) { 2225 if (SymbolTable) 2226 for (const SecPtr &Sec : Sections) 2227 if (Error E = Sec->removeSymbols(ToRemove)) 2228 return E; 2229 return Error::success(); 2230 } 2231 2232 Error Object::addNewSymbolTable() { 2233 assert(!SymbolTable && "Object must not has a SymbolTable."); 2234 2235 // Reuse an existing SHT_STRTAB section if it exists. 2236 StringTableSection *StrTab = nullptr; 2237 for (SectionBase &Sec : sections()) { 2238 if (Sec.Type == ELF::SHT_STRTAB && !(Sec.Flags & SHF_ALLOC)) { 2239 StrTab = static_cast<StringTableSection *>(&Sec); 2240 2241 // Prefer a string table that is not the section header string table, if 2242 // such a table exists. 2243 if (SectionNames != &Sec) 2244 break; 2245 } 2246 } 2247 if (!StrTab) 2248 StrTab = &addSection<StringTableSection>(); 2249 2250 SymbolTableSection &SymTab = addSection<SymbolTableSection>(); 2251 SymTab.Name = ".symtab"; 2252 SymTab.Link = StrTab->Index; 2253 if (Error Err = SymTab.initialize(sections())) 2254 return Err; 2255 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0); 2256 2257 SymbolTable = &SymTab; 2258 2259 return Error::success(); 2260 } 2261 2262 // Orders segments such that if x = y->ParentSegment then y comes before x. 2263 static void orderSegments(std::vector<Segment *> &Segments) { 2264 llvm::stable_sort(Segments, compareSegmentsByOffset); 2265 } 2266 2267 // This function finds a consistent layout for a list of segments starting from 2268 // an Offset. It assumes that Segments have been sorted by orderSegments and 2269 // returns an Offset one past the end of the last segment. 2270 static uint64_t layoutSegments(std::vector<Segment *> &Segments, 2271 uint64_t Offset) { 2272 assert(llvm::is_sorted(Segments, compareSegmentsByOffset)); 2273 // The only way a segment should move is if a section was between two 2274 // segments and that section was removed. If that section isn't in a segment 2275 // then it's acceptable, but not ideal, to simply move it to after the 2276 // segments. So we can simply layout segments one after the other accounting 2277 // for alignment. 2278 for (Segment *Seg : Segments) { 2279 // We assume that segments have been ordered by OriginalOffset and Index 2280 // such that a parent segment will always come before a child segment in 2281 // OrderedSegments. This means that the Offset of the ParentSegment should 2282 // already be set and we can set our offset relative to it. 2283 if (Seg->ParentSegment != nullptr) { 2284 Segment *Parent = Seg->ParentSegment; 2285 Seg->Offset = 2286 Parent->Offset + Seg->OriginalOffset - Parent->OriginalOffset; 2287 } else { 2288 Seg->Offset = 2289 alignTo(Offset, std::max<uint64_t>(Seg->Align, 1), Seg->VAddr); 2290 } 2291 Offset = std::max(Offset, Seg->Offset + Seg->FileSize); 2292 } 2293 return Offset; 2294 } 2295 2296 // This function finds a consistent layout for a list of sections. It assumes 2297 // that the ->ParentSegment of each section has already been laid out. The 2298 // supplied starting Offset is used for the starting offset of any section that 2299 // does not have a ParentSegment. It returns either the offset given if all 2300 // sections had a ParentSegment or an offset one past the last section if there 2301 // was a section that didn't have a ParentSegment. 2302 template <class Range> 2303 static uint64_t layoutSections(Range Sections, uint64_t Offset) { 2304 // Now the offset of every segment has been set we can assign the offsets 2305 // of each section. For sections that are covered by a segment we should use 2306 // the segment's original offset and the section's original offset to compute 2307 // the offset from the start of the segment. Using the offset from the start 2308 // of the segment we can assign a new offset to the section. For sections not 2309 // covered by segments we can just bump Offset to the next valid location. 2310 // While it is not necessary, layout the sections in the order based on their 2311 // original offsets to resemble the input file as close as possible. 2312 std::vector<SectionBase *> OutOfSegmentSections; 2313 uint32_t Index = 1; 2314 for (auto &Sec : Sections) { 2315 Sec.Index = Index++; 2316 if (Sec.ParentSegment != nullptr) { 2317 auto Segment = *Sec.ParentSegment; 2318 Sec.Offset = 2319 Segment.Offset + (Sec.OriginalOffset - Segment.OriginalOffset); 2320 } else 2321 OutOfSegmentSections.push_back(&Sec); 2322 } 2323 2324 llvm::stable_sort(OutOfSegmentSections, 2325 [](const SectionBase *Lhs, const SectionBase *Rhs) { 2326 return Lhs->OriginalOffset < Rhs->OriginalOffset; 2327 }); 2328 for (auto *Sec : OutOfSegmentSections) { 2329 Offset = alignTo(Offset, Sec->Align == 0 ? 1 : Sec->Align); 2330 Sec->Offset = Offset; 2331 if (Sec->Type != SHT_NOBITS) 2332 Offset += Sec->Size; 2333 } 2334 return Offset; 2335 } 2336 2337 // Rewrite sh_offset after some sections are changed to SHT_NOBITS and thus 2338 // occupy no space in the file. 2339 static uint64_t layoutSectionsForOnlyKeepDebug(Object &Obj, uint64_t Off) { 2340 // The layout algorithm requires the sections to be handled in the order of 2341 // their offsets in the input file, at least inside segments. 2342 std::vector<SectionBase *> Sections; 2343 Sections.reserve(Obj.sections().size()); 2344 uint32_t Index = 1; 2345 for (auto &Sec : Obj.sections()) { 2346 Sec.Index = Index++; 2347 Sections.push_back(&Sec); 2348 } 2349 llvm::stable_sort(Sections, 2350 [](const SectionBase *Lhs, const SectionBase *Rhs) { 2351 return Lhs->OriginalOffset < Rhs->OriginalOffset; 2352 }); 2353 2354 for (auto *Sec : Sections) { 2355 auto *FirstSec = Sec->ParentSegment && Sec->ParentSegment->Type == PT_LOAD 2356 ? Sec->ParentSegment->firstSection() 2357 : nullptr; 2358 2359 // The first section in a PT_LOAD has to have congruent offset and address 2360 // modulo the alignment, which usually equals the maximum page size. 2361 if (FirstSec && FirstSec == Sec) 2362 Off = alignTo(Off, Sec->ParentSegment->Align, Sec->Addr); 2363 2364 // sh_offset is not significant for SHT_NOBITS sections, but the congruence 2365 // rule must be followed if it is the first section in a PT_LOAD. Do not 2366 // advance Off. 2367 if (Sec->Type == SHT_NOBITS) { 2368 Sec->Offset = Off; 2369 continue; 2370 } 2371 2372 if (!FirstSec) { 2373 // FirstSec being nullptr generally means that Sec does not have the 2374 // SHF_ALLOC flag. 2375 Off = Sec->Align ? alignTo(Off, Sec->Align) : Off; 2376 } else if (FirstSec != Sec) { 2377 // The offset is relative to the first section in the PT_LOAD segment. Use 2378 // sh_offset for non-SHF_ALLOC sections. 2379 Off = Sec->OriginalOffset - FirstSec->OriginalOffset + FirstSec->Offset; 2380 } 2381 Sec->Offset = Off; 2382 Off += Sec->Size; 2383 } 2384 return Off; 2385 } 2386 2387 // Rewrite p_offset and p_filesz of non-PT_PHDR segments after sh_offset values 2388 // have been updated. 2389 static uint64_t layoutSegmentsForOnlyKeepDebug(std::vector<Segment *> &Segments, 2390 uint64_t HdrEnd) { 2391 uint64_t MaxOffset = 0; 2392 for (Segment *Seg : Segments) { 2393 if (Seg->Type == PT_PHDR) 2394 continue; 2395 2396 // The segment offset is generally the offset of the first section. 2397 // 2398 // For a segment containing no section (see sectionWithinSegment), if it has 2399 // a parent segment, copy the parent segment's offset field. This works for 2400 // empty PT_TLS. If no parent segment, use 0: the segment is not useful for 2401 // debugging anyway. 2402 const SectionBase *FirstSec = Seg->firstSection(); 2403 uint64_t Offset = 2404 FirstSec ? FirstSec->Offset 2405 : (Seg->ParentSegment ? Seg->ParentSegment->Offset : 0); 2406 uint64_t FileSize = 0; 2407 for (const SectionBase *Sec : Seg->Sections) { 2408 uint64_t Size = Sec->Type == SHT_NOBITS ? 0 : Sec->Size; 2409 if (Sec->Offset + Size > Offset) 2410 FileSize = std::max(FileSize, Sec->Offset + Size - Offset); 2411 } 2412 2413 // If the segment includes EHDR and program headers, don't make it smaller 2414 // than the headers. 2415 if (Seg->Offset < HdrEnd && HdrEnd <= Seg->Offset + Seg->FileSize) { 2416 FileSize += Offset - Seg->Offset; 2417 Offset = Seg->Offset; 2418 FileSize = std::max(FileSize, HdrEnd - Offset); 2419 } 2420 2421 Seg->Offset = Offset; 2422 Seg->FileSize = FileSize; 2423 MaxOffset = std::max(MaxOffset, Offset + FileSize); 2424 } 2425 return MaxOffset; 2426 } 2427 2428 template <class ELFT> void ELFWriter<ELFT>::initEhdrSegment() { 2429 Segment &ElfHdr = Obj.ElfHdrSegment; 2430 ElfHdr.Type = PT_PHDR; 2431 ElfHdr.Flags = 0; 2432 ElfHdr.VAddr = 0; 2433 ElfHdr.PAddr = 0; 2434 ElfHdr.FileSize = ElfHdr.MemSize = sizeof(Elf_Ehdr); 2435 ElfHdr.Align = 0; 2436 } 2437 2438 template <class ELFT> void ELFWriter<ELFT>::assignOffsets() { 2439 // We need a temporary list of segments that has a special order to it 2440 // so that we know that anytime ->ParentSegment is set that segment has 2441 // already had its offset properly set. 2442 std::vector<Segment *> OrderedSegments; 2443 for (Segment &Segment : Obj.segments()) 2444 OrderedSegments.push_back(&Segment); 2445 OrderedSegments.push_back(&Obj.ElfHdrSegment); 2446 OrderedSegments.push_back(&Obj.ProgramHdrSegment); 2447 orderSegments(OrderedSegments); 2448 2449 uint64_t Offset; 2450 if (OnlyKeepDebug) { 2451 // For --only-keep-debug, the sections that did not preserve contents were 2452 // changed to SHT_NOBITS. We now rewrite sh_offset fields of sections, and 2453 // then rewrite p_offset/p_filesz of program headers. 2454 uint64_t HdrEnd = 2455 sizeof(Elf_Ehdr) + llvm::size(Obj.segments()) * sizeof(Elf_Phdr); 2456 Offset = layoutSectionsForOnlyKeepDebug(Obj, HdrEnd); 2457 Offset = std::max(Offset, 2458 layoutSegmentsForOnlyKeepDebug(OrderedSegments, HdrEnd)); 2459 } else { 2460 // Offset is used as the start offset of the first segment to be laid out. 2461 // Since the ELF Header (ElfHdrSegment) must be at the start of the file, 2462 // we start at offset 0. 2463 Offset = layoutSegments(OrderedSegments, 0); 2464 Offset = layoutSections(Obj.sections(), Offset); 2465 } 2466 // If we need to write the section header table out then we need to align the 2467 // Offset so that SHOffset is valid. 2468 if (WriteSectionHeaders) 2469 Offset = alignTo(Offset, sizeof(Elf_Addr)); 2470 Obj.SHOff = Offset; 2471 } 2472 2473 template <class ELFT> size_t ELFWriter<ELFT>::totalSize() const { 2474 // We already have the section header offset so we can calculate the total 2475 // size by just adding up the size of each section header. 2476 if (!WriteSectionHeaders) 2477 return Obj.SHOff; 2478 size_t ShdrCount = Obj.sections().size() + 1; // Includes null shdr. 2479 return Obj.SHOff + ShdrCount * sizeof(Elf_Shdr); 2480 } 2481 2482 template <class ELFT> Error ELFWriter<ELFT>::write() { 2483 // Segment data must be written first, so that the ELF header and program 2484 // header tables can overwrite it, if covered by a segment. 2485 writeSegmentData(); 2486 writeEhdr(); 2487 writePhdrs(); 2488 if (Error E = writeSectionData()) 2489 return E; 2490 if (WriteSectionHeaders) 2491 writeShdrs(); 2492 2493 // TODO: Implement direct writing to the output stream (without intermediate 2494 // memory buffer Buf). 2495 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2496 return Error::success(); 2497 } 2498 2499 static Error removeUnneededSections(Object &Obj) { 2500 // We can remove an empty symbol table from non-relocatable objects. 2501 // Relocatable objects typically have relocation sections whose 2502 // sh_link field points to .symtab, so we can't remove .symtab 2503 // even if it is empty. 2504 if (Obj.isRelocatable() || Obj.SymbolTable == nullptr || 2505 !Obj.SymbolTable->empty()) 2506 return Error::success(); 2507 2508 // .strtab can be used for section names. In such a case we shouldn't 2509 // remove it. 2510 auto *StrTab = Obj.SymbolTable->getStrTab() == Obj.SectionNames 2511 ? nullptr 2512 : Obj.SymbolTable->getStrTab(); 2513 return Obj.removeSections(false, [&](const SectionBase &Sec) { 2514 return &Sec == Obj.SymbolTable || &Sec == StrTab; 2515 }); 2516 } 2517 2518 template <class ELFT> Error ELFWriter<ELFT>::finalize() { 2519 // It could happen that SectionNames has been removed and yet the user wants 2520 // a section header table output. We need to throw an error if a user tries 2521 // to do that. 2522 if (Obj.SectionNames == nullptr && WriteSectionHeaders) 2523 return createStringError(llvm::errc::invalid_argument, 2524 "cannot write section header table because " 2525 "section header string table was removed"); 2526 2527 if (Error E = removeUnneededSections(Obj)) 2528 return E; 2529 2530 // We need to assign indexes before we perform layout because we need to know 2531 // if we need large indexes or not. We can assign indexes first and check as 2532 // we go to see if we will actully need large indexes. 2533 bool NeedsLargeIndexes = false; 2534 if (Obj.sections().size() >= SHN_LORESERVE) { 2535 SectionTableRef Sections = Obj.sections(); 2536 // Sections doesn't include the null section header, so account for this 2537 // when skipping the first N sections. 2538 NeedsLargeIndexes = 2539 any_of(drop_begin(Sections, SHN_LORESERVE - 1), 2540 [](const SectionBase &Sec) { return Sec.HasSymbol; }); 2541 // TODO: handle case where only one section needs the large index table but 2542 // only needs it because the large index table hasn't been removed yet. 2543 } 2544 2545 if (NeedsLargeIndexes) { 2546 // This means we definitely need to have a section index table but if we 2547 // already have one then we should use it instead of making a new one. 2548 if (Obj.SymbolTable != nullptr && Obj.SectionIndexTable == nullptr) { 2549 // Addition of a section to the end does not invalidate the indexes of 2550 // other sections and assigns the correct index to the new section. 2551 auto &Shndx = Obj.addSection<SectionIndexSection>(); 2552 Obj.SymbolTable->setShndxTable(&Shndx); 2553 Shndx.setSymTab(Obj.SymbolTable); 2554 } 2555 } else { 2556 // Since we don't need SectionIndexTable we should remove it and all 2557 // references to it. 2558 if (Obj.SectionIndexTable != nullptr) { 2559 // We do not support sections referring to the section index table. 2560 if (Error E = Obj.removeSections(false /*AllowBrokenLinks*/, 2561 [this](const SectionBase &Sec) { 2562 return &Sec == Obj.SectionIndexTable; 2563 })) 2564 return E; 2565 } 2566 } 2567 2568 // Make sure we add the names of all the sections. Importantly this must be 2569 // done after we decide to add or remove SectionIndexes. 2570 if (Obj.SectionNames != nullptr) 2571 for (const SectionBase &Sec : Obj.sections()) 2572 Obj.SectionNames->addString(Sec.Name); 2573 2574 initEhdrSegment(); 2575 2576 // Before we can prepare for layout the indexes need to be finalized. 2577 // Also, the output arch may not be the same as the input arch, so fix up 2578 // size-related fields before doing layout calculations. 2579 uint64_t Index = 0; 2580 auto SecSizer = std::make_unique<ELFSectionSizer<ELFT>>(); 2581 for (SectionBase &Sec : Obj.sections()) { 2582 Sec.Index = Index++; 2583 if (Error Err = Sec.accept(*SecSizer)) 2584 return Err; 2585 } 2586 2587 // The symbol table does not update all other sections on update. For 2588 // instance, symbol names are not added as new symbols are added. This means 2589 // that some sections, like .strtab, don't yet have their final size. 2590 if (Obj.SymbolTable != nullptr) 2591 Obj.SymbolTable->prepareForLayout(); 2592 2593 // Now that all strings are added we want to finalize string table builders, 2594 // because that affects section sizes which in turn affects section offsets. 2595 for (SectionBase &Sec : Obj.sections()) 2596 if (auto StrTab = dyn_cast<StringTableSection>(&Sec)) 2597 StrTab->prepareForLayout(); 2598 2599 assignOffsets(); 2600 2601 // layoutSections could have modified section indexes, so we need 2602 // to fill the index table after assignOffsets. 2603 if (Obj.SymbolTable != nullptr) 2604 Obj.SymbolTable->fillShndxTable(); 2605 2606 // Finally now that all offsets and indexes have been set we can finalize any 2607 // remaining issues. 2608 uint64_t Offset = Obj.SHOff + sizeof(Elf_Shdr); 2609 for (SectionBase &Sec : Obj.sections()) { 2610 Sec.HeaderOffset = Offset; 2611 Offset += sizeof(Elf_Shdr); 2612 if (WriteSectionHeaders) 2613 Sec.NameIndex = Obj.SectionNames->findIndex(Sec.Name); 2614 Sec.finalize(); 2615 } 2616 2617 size_t TotalSize = totalSize(); 2618 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2619 if (!Buf) 2620 return createStringError(errc::not_enough_memory, 2621 "failed to allocate memory buffer of " + 2622 Twine::utohexstr(TotalSize) + " bytes"); 2623 2624 SecWriter = std::make_unique<ELFSectionWriter<ELFT>>(*Buf); 2625 return Error::success(); 2626 } 2627 2628 Error BinaryWriter::write() { 2629 for (const SectionBase &Sec : Obj.allocSections()) 2630 if (Error Err = Sec.accept(*SecWriter)) 2631 return Err; 2632 2633 // TODO: Implement direct writing to the output stream (without intermediate 2634 // memory buffer Buf). 2635 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2636 return Error::success(); 2637 } 2638 2639 Error BinaryWriter::finalize() { 2640 // Compute the section LMA based on its sh_offset and the containing segment's 2641 // p_offset and p_paddr. Also compute the minimum LMA of all non-empty 2642 // sections as MinAddr. In the output, the contents between address 0 and 2643 // MinAddr will be skipped. 2644 uint64_t MinAddr = UINT64_MAX; 2645 for (SectionBase &Sec : Obj.allocSections()) { 2646 if (Sec.ParentSegment != nullptr) 2647 Sec.Addr = 2648 Sec.Offset - Sec.ParentSegment->Offset + Sec.ParentSegment->PAddr; 2649 if (Sec.Type != SHT_NOBITS && Sec.Size > 0) 2650 MinAddr = std::min(MinAddr, Sec.Addr); 2651 } 2652 2653 // Now that every section has been laid out we just need to compute the total 2654 // file size. This might not be the same as the offset returned by 2655 // layoutSections, because we want to truncate the last segment to the end of 2656 // its last non-empty section, to match GNU objcopy's behaviour. 2657 TotalSize = 0; 2658 for (SectionBase &Sec : Obj.allocSections()) 2659 if (Sec.Type != SHT_NOBITS && Sec.Size > 0) { 2660 Sec.Offset = Sec.Addr - MinAddr; 2661 TotalSize = std::max(TotalSize, Sec.Offset + Sec.Size); 2662 } 2663 2664 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2665 if (!Buf) 2666 return createStringError(errc::not_enough_memory, 2667 "failed to allocate memory buffer of " + 2668 Twine::utohexstr(TotalSize) + " bytes"); 2669 SecWriter = std::make_unique<BinarySectionWriter>(*Buf); 2670 return Error::success(); 2671 } 2672 2673 bool IHexWriter::SectionCompare::operator()(const SectionBase *Lhs, 2674 const SectionBase *Rhs) const { 2675 return (sectionPhysicalAddr(Lhs) & 0xFFFFFFFFU) < 2676 (sectionPhysicalAddr(Rhs) & 0xFFFFFFFFU); 2677 } 2678 2679 uint64_t IHexWriter::writeEntryPointRecord(uint8_t *Buf) { 2680 IHexLineData HexData; 2681 uint8_t Data[4] = {}; 2682 // We don't write entry point record if entry is zero. 2683 if (Obj.Entry == 0) 2684 return 0; 2685 2686 if (Obj.Entry <= 0xFFFFFU) { 2687 Data[0] = ((Obj.Entry & 0xF0000U) >> 12) & 0xFF; 2688 support::endian::write(&Data[2], static_cast<uint16_t>(Obj.Entry), 2689 support::big); 2690 HexData = IHexRecord::getLine(IHexRecord::StartAddr80x86, 0, Data); 2691 } else { 2692 support::endian::write(Data, static_cast<uint32_t>(Obj.Entry), 2693 support::big); 2694 HexData = IHexRecord::getLine(IHexRecord::StartAddr, 0, Data); 2695 } 2696 memcpy(Buf, HexData.data(), HexData.size()); 2697 return HexData.size(); 2698 } 2699 2700 uint64_t IHexWriter::writeEndOfFileRecord(uint8_t *Buf) { 2701 IHexLineData HexData = IHexRecord::getLine(IHexRecord::EndOfFile, 0, {}); 2702 memcpy(Buf, HexData.data(), HexData.size()); 2703 return HexData.size(); 2704 } 2705 2706 Error IHexWriter::write() { 2707 IHexSectionWriter Writer(*Buf); 2708 // Write sections. 2709 for (const SectionBase *Sec : Sections) 2710 if (Error Err = Sec->accept(Writer)) 2711 return Err; 2712 2713 uint64_t Offset = Writer.getBufferOffset(); 2714 // Write entry point address. 2715 Offset += writeEntryPointRecord( 2716 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset); 2717 // Write EOF. 2718 Offset += writeEndOfFileRecord( 2719 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset); 2720 assert(Offset == TotalSize); 2721 2722 // TODO: Implement direct writing to the output stream (without intermediate 2723 // memory buffer Buf). 2724 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2725 return Error::success(); 2726 } 2727 2728 Error IHexWriter::checkSection(const SectionBase &Sec) { 2729 uint64_t Addr = sectionPhysicalAddr(&Sec); 2730 if (addressOverflows32bit(Addr) || addressOverflows32bit(Addr + Sec.Size - 1)) 2731 return createStringError( 2732 errc::invalid_argument, 2733 "Section '%s' address range [0x%llx, 0x%llx] is not 32 bit", 2734 Sec.Name.c_str(), Addr, Addr + Sec.Size - 1); 2735 return Error::success(); 2736 } 2737 2738 Error IHexWriter::finalize() { 2739 // We can't write 64-bit addresses. 2740 if (addressOverflows32bit(Obj.Entry)) 2741 return createStringError(errc::invalid_argument, 2742 "Entry point address 0x%llx overflows 32 bits", 2743 Obj.Entry); 2744 2745 for (const SectionBase &Sec : Obj.sections()) 2746 if ((Sec.Flags & ELF::SHF_ALLOC) && Sec.Type != ELF::SHT_NOBITS && 2747 Sec.Size > 0) { 2748 if (Error E = checkSection(Sec)) 2749 return E; 2750 Sections.insert(&Sec); 2751 } 2752 2753 std::unique_ptr<WritableMemoryBuffer> EmptyBuffer = 2754 WritableMemoryBuffer::getNewMemBuffer(0); 2755 if (!EmptyBuffer) 2756 return createStringError(errc::not_enough_memory, 2757 "failed to allocate memory buffer of 0 bytes"); 2758 2759 IHexSectionWriterBase LengthCalc(*EmptyBuffer); 2760 for (const SectionBase *Sec : Sections) 2761 if (Error Err = Sec->accept(LengthCalc)) 2762 return Err; 2763 2764 // We need space to write section records + StartAddress record 2765 // (if start adress is not zero) + EndOfFile record. 2766 TotalSize = LengthCalc.getBufferOffset() + 2767 (Obj.Entry ? IHexRecord::getLineLength(4) : 0) + 2768 IHexRecord::getLineLength(0); 2769 2770 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2771 if (!Buf) 2772 return createStringError(errc::not_enough_memory, 2773 "failed to allocate memory buffer of " + 2774 Twine::utohexstr(TotalSize) + " bytes"); 2775 2776 return Error::success(); 2777 } 2778 2779 namespace llvm { 2780 namespace objcopy { 2781 namespace elf { 2782 2783 template class ELFBuilder<ELF64LE>; 2784 template class ELFBuilder<ELF64BE>; 2785 template class ELFBuilder<ELF32LE>; 2786 template class ELFBuilder<ELF32BE>; 2787 2788 template class ELFWriter<ELF64LE>; 2789 template class ELFWriter<ELF64BE>; 2790 template class ELFWriter<ELF32LE>; 2791 template class ELFWriter<ELF32BE>; 2792 2793 } // end namespace elf 2794 } // end namespace objcopy 2795 } // end namespace llvm 2796