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 = 472 compression::zlib::uncompress(CompressedContent, DecompressedContent, 473 static_cast<size_t>(Sec.Size))) 474 return createStringError(errc::invalid_argument, 475 "'" + Sec.Name + "': " + toString(std::move(Err))); 476 477 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 478 std::copy(DecompressedContent.begin(), DecompressedContent.end(), Buf); 479 480 return Error::success(); 481 } 482 483 Error BinarySectionWriter::visit(const DecompressedSection &Sec) { 484 return createStringError(errc::operation_not_permitted, 485 "cannot write compressed section '" + Sec.Name + 486 "' "); 487 } 488 489 Error DecompressedSection::accept(SectionVisitor &Visitor) const { 490 return Visitor.visit(*this); 491 } 492 493 Error DecompressedSection::accept(MutableSectionVisitor &Visitor) { 494 return Visitor.visit(*this); 495 } 496 497 Error OwnedDataSection::accept(SectionVisitor &Visitor) const { 498 return Visitor.visit(*this); 499 } 500 501 Error OwnedDataSection::accept(MutableSectionVisitor &Visitor) { 502 return Visitor.visit(*this); 503 } 504 505 void OwnedDataSection::appendHexData(StringRef HexData) { 506 assert((HexData.size() & 1) == 0); 507 while (!HexData.empty()) { 508 Data.push_back(checkedGetHex<uint8_t>(HexData.take_front(2))); 509 HexData = HexData.drop_front(2); 510 } 511 Size = Data.size(); 512 } 513 514 Error BinarySectionWriter::visit(const CompressedSection &Sec) { 515 return createStringError(errc::operation_not_permitted, 516 "cannot write compressed section '" + Sec.Name + 517 "' "); 518 } 519 520 template <class ELFT> 521 Error ELFSectionWriter<ELFT>::visit(const CompressedSection &Sec) { 522 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 523 Elf_Chdr_Impl<ELFT> Chdr; 524 switch (Sec.CompressionType) { 525 case DebugCompressionType::None: 526 std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(), Buf); 527 return Error::success(); 528 case DebugCompressionType::GNU: 529 llvm_unreachable("unexpected zlib-gnu"); 530 break; 531 case DebugCompressionType::Z: 532 Chdr.ch_type = ELF::ELFCOMPRESS_ZLIB; 533 break; 534 } 535 Chdr.ch_size = Sec.DecompressedSize; 536 Chdr.ch_addralign = Sec.DecompressedAlign; 537 memcpy(Buf, &Chdr, sizeof(Chdr)); 538 Buf += sizeof(Chdr); 539 540 std::copy(Sec.CompressedData.begin(), Sec.CompressedData.end(), Buf); 541 return Error::success(); 542 } 543 544 CompressedSection::CompressedSection(const SectionBase &Sec, 545 DebugCompressionType CompressionType) 546 : SectionBase(Sec), CompressionType(CompressionType), 547 DecompressedSize(Sec.OriginalData.size()), DecompressedAlign(Sec.Align) { 548 compression::zlib::compress( 549 StringRef(reinterpret_cast<const char *>(OriginalData.data()), 550 OriginalData.size()), 551 CompressedData); 552 553 assert(CompressionType != DebugCompressionType::None); 554 Flags |= ELF::SHF_COMPRESSED; 555 size_t ChdrSize = 556 std::max(std::max(sizeof(object::Elf_Chdr_Impl<object::ELF64LE>), 557 sizeof(object::Elf_Chdr_Impl<object::ELF64BE>)), 558 std::max(sizeof(object::Elf_Chdr_Impl<object::ELF32LE>), 559 sizeof(object::Elf_Chdr_Impl<object::ELF32BE>))); 560 Size = ChdrSize + CompressedData.size(); 561 Align = 8; 562 } 563 564 CompressedSection::CompressedSection(ArrayRef<uint8_t> CompressedData, 565 uint64_t DecompressedSize, 566 uint64_t DecompressedAlign) 567 : CompressionType(DebugCompressionType::None), 568 DecompressedSize(DecompressedSize), DecompressedAlign(DecompressedAlign) { 569 OriginalData = CompressedData; 570 } 571 572 Error CompressedSection::accept(SectionVisitor &Visitor) const { 573 return Visitor.visit(*this); 574 } 575 576 Error CompressedSection::accept(MutableSectionVisitor &Visitor) { 577 return Visitor.visit(*this); 578 } 579 580 void StringTableSection::addString(StringRef Name) { StrTabBuilder.add(Name); } 581 582 uint32_t StringTableSection::findIndex(StringRef Name) const { 583 return StrTabBuilder.getOffset(Name); 584 } 585 586 void StringTableSection::prepareForLayout() { 587 StrTabBuilder.finalize(); 588 Size = StrTabBuilder.getSize(); 589 } 590 591 Error SectionWriter::visit(const StringTableSection &Sec) { 592 Sec.StrTabBuilder.write(reinterpret_cast<uint8_t *>(Out.getBufferStart()) + 593 Sec.Offset); 594 return Error::success(); 595 } 596 597 Error StringTableSection::accept(SectionVisitor &Visitor) const { 598 return Visitor.visit(*this); 599 } 600 601 Error StringTableSection::accept(MutableSectionVisitor &Visitor) { 602 return Visitor.visit(*this); 603 } 604 605 template <class ELFT> 606 Error ELFSectionWriter<ELFT>::visit(const SectionIndexSection &Sec) { 607 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 608 llvm::copy(Sec.Indexes, reinterpret_cast<Elf_Word *>(Buf)); 609 return Error::success(); 610 } 611 612 Error SectionIndexSection::initialize(SectionTableRef SecTable) { 613 Size = 0; 614 Expected<SymbolTableSection *> Sec = 615 SecTable.getSectionOfType<SymbolTableSection>( 616 Link, 617 "Link field value " + Twine(Link) + " in section " + Name + 618 " is invalid", 619 "Link field value " + Twine(Link) + " in section " + Name + 620 " is not a symbol table"); 621 if (!Sec) 622 return Sec.takeError(); 623 624 setSymTab(*Sec); 625 Symbols->setShndxTable(this); 626 return Error::success(); 627 } 628 629 void SectionIndexSection::finalize() { Link = Symbols->Index; } 630 631 Error SectionIndexSection::accept(SectionVisitor &Visitor) const { 632 return Visitor.visit(*this); 633 } 634 635 Error SectionIndexSection::accept(MutableSectionVisitor &Visitor) { 636 return Visitor.visit(*this); 637 } 638 639 static bool isValidReservedSectionIndex(uint16_t Index, uint16_t Machine) { 640 switch (Index) { 641 case SHN_ABS: 642 case SHN_COMMON: 643 return true; 644 } 645 646 if (Machine == EM_AMDGPU) { 647 return Index == SHN_AMDGPU_LDS; 648 } 649 650 if (Machine == EM_MIPS) { 651 switch (Index) { 652 case SHN_MIPS_ACOMMON: 653 case SHN_MIPS_SCOMMON: 654 case SHN_MIPS_SUNDEFINED: 655 return true; 656 } 657 } 658 659 if (Machine == EM_HEXAGON) { 660 switch (Index) { 661 case SHN_HEXAGON_SCOMMON: 662 case SHN_HEXAGON_SCOMMON_1: 663 case SHN_HEXAGON_SCOMMON_2: 664 case SHN_HEXAGON_SCOMMON_4: 665 case SHN_HEXAGON_SCOMMON_8: 666 return true; 667 } 668 } 669 return false; 670 } 671 672 // Large indexes force us to clarify exactly what this function should do. This 673 // function should return the value that will appear in st_shndx when written 674 // out. 675 uint16_t Symbol::getShndx() const { 676 if (DefinedIn != nullptr) { 677 if (DefinedIn->Index >= SHN_LORESERVE) 678 return SHN_XINDEX; 679 return DefinedIn->Index; 680 } 681 682 if (ShndxType == SYMBOL_SIMPLE_INDEX) { 683 // This means that we don't have a defined section but we do need to 684 // output a legitimate section index. 685 return SHN_UNDEF; 686 } 687 688 assert(ShndxType == SYMBOL_ABS || ShndxType == SYMBOL_COMMON || 689 (ShndxType >= SYMBOL_LOPROC && ShndxType <= SYMBOL_HIPROC) || 690 (ShndxType >= SYMBOL_LOOS && ShndxType <= SYMBOL_HIOS)); 691 return static_cast<uint16_t>(ShndxType); 692 } 693 694 bool Symbol::isCommon() const { return getShndx() == SHN_COMMON; } 695 696 void SymbolTableSection::assignIndices() { 697 uint32_t Index = 0; 698 for (auto &Sym : Symbols) 699 Sym->Index = Index++; 700 } 701 702 void SymbolTableSection::addSymbol(Twine Name, uint8_t Bind, uint8_t Type, 703 SectionBase *DefinedIn, uint64_t Value, 704 uint8_t Visibility, uint16_t Shndx, 705 uint64_t SymbolSize) { 706 Symbol Sym; 707 Sym.Name = Name.str(); 708 Sym.Binding = Bind; 709 Sym.Type = Type; 710 Sym.DefinedIn = DefinedIn; 711 if (DefinedIn != nullptr) 712 DefinedIn->HasSymbol = true; 713 if (DefinedIn == nullptr) { 714 if (Shndx >= SHN_LORESERVE) 715 Sym.ShndxType = static_cast<SymbolShndxType>(Shndx); 716 else 717 Sym.ShndxType = SYMBOL_SIMPLE_INDEX; 718 } 719 Sym.Value = Value; 720 Sym.Visibility = Visibility; 721 Sym.Size = SymbolSize; 722 Sym.Index = Symbols.size(); 723 Symbols.emplace_back(std::make_unique<Symbol>(Sym)); 724 Size += this->EntrySize; 725 } 726 727 Error SymbolTableSection::removeSectionReferences( 728 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 729 if (ToRemove(SectionIndexTable)) 730 SectionIndexTable = nullptr; 731 if (ToRemove(SymbolNames)) { 732 if (!AllowBrokenLinks) 733 return createStringError( 734 llvm::errc::invalid_argument, 735 "string table '%s' cannot be removed because it is " 736 "referenced by the symbol table '%s'", 737 SymbolNames->Name.data(), this->Name.data()); 738 SymbolNames = nullptr; 739 } 740 return removeSymbols( 741 [ToRemove](const Symbol &Sym) { return ToRemove(Sym.DefinedIn); }); 742 } 743 744 void SymbolTableSection::updateSymbols(function_ref<void(Symbol &)> Callable) { 745 for (SymPtr &Sym : llvm::drop_begin(Symbols)) 746 Callable(*Sym); 747 std::stable_partition( 748 std::begin(Symbols), std::end(Symbols), 749 [](const SymPtr &Sym) { return Sym->Binding == STB_LOCAL; }); 750 assignIndices(); 751 } 752 753 Error SymbolTableSection::removeSymbols( 754 function_ref<bool(const Symbol &)> ToRemove) { 755 Symbols.erase( 756 std::remove_if(std::begin(Symbols) + 1, std::end(Symbols), 757 [ToRemove](const SymPtr &Sym) { return ToRemove(*Sym); }), 758 std::end(Symbols)); 759 Size = Symbols.size() * EntrySize; 760 assignIndices(); 761 return Error::success(); 762 } 763 764 void SymbolTableSection::replaceSectionReferences( 765 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 766 for (std::unique_ptr<Symbol> &Sym : Symbols) 767 if (SectionBase *To = FromTo.lookup(Sym->DefinedIn)) 768 Sym->DefinedIn = To; 769 } 770 771 Error SymbolTableSection::initialize(SectionTableRef SecTable) { 772 Size = 0; 773 Expected<StringTableSection *> Sec = 774 SecTable.getSectionOfType<StringTableSection>( 775 Link, 776 "Symbol table has link index of " + Twine(Link) + 777 " which is not a valid index", 778 "Symbol table has link index of " + Twine(Link) + 779 " which is not a string table"); 780 if (!Sec) 781 return Sec.takeError(); 782 783 setStrTab(*Sec); 784 return Error::success(); 785 } 786 787 void SymbolTableSection::finalize() { 788 uint32_t MaxLocalIndex = 0; 789 for (std::unique_ptr<Symbol> &Sym : Symbols) { 790 Sym->NameIndex = 791 SymbolNames == nullptr ? 0 : SymbolNames->findIndex(Sym->Name); 792 if (Sym->Binding == STB_LOCAL) 793 MaxLocalIndex = std::max(MaxLocalIndex, Sym->Index); 794 } 795 // Now we need to set the Link and Info fields. 796 Link = SymbolNames == nullptr ? 0 : SymbolNames->Index; 797 Info = MaxLocalIndex + 1; 798 } 799 800 void SymbolTableSection::prepareForLayout() { 801 // Reserve proper amount of space in section index table, so we can 802 // layout sections correctly. We will fill the table with correct 803 // indexes later in fillShdnxTable. 804 if (SectionIndexTable) 805 SectionIndexTable->reserve(Symbols.size()); 806 807 // Add all of our strings to SymbolNames so that SymbolNames has the right 808 // size before layout is decided. 809 // If the symbol names section has been removed, don't try to add strings to 810 // the table. 811 if (SymbolNames != nullptr) 812 for (std::unique_ptr<Symbol> &Sym : Symbols) 813 SymbolNames->addString(Sym->Name); 814 } 815 816 void SymbolTableSection::fillShndxTable() { 817 if (SectionIndexTable == nullptr) 818 return; 819 // Fill section index table with real section indexes. This function must 820 // be called after assignOffsets. 821 for (const std::unique_ptr<Symbol> &Sym : Symbols) { 822 if (Sym->DefinedIn != nullptr && Sym->DefinedIn->Index >= SHN_LORESERVE) 823 SectionIndexTable->addIndex(Sym->DefinedIn->Index); 824 else 825 SectionIndexTable->addIndex(SHN_UNDEF); 826 } 827 } 828 829 Expected<const Symbol *> 830 SymbolTableSection::getSymbolByIndex(uint32_t Index) const { 831 if (Symbols.size() <= Index) 832 return createStringError(errc::invalid_argument, 833 "invalid symbol index: " + Twine(Index)); 834 return Symbols[Index].get(); 835 } 836 837 Expected<Symbol *> SymbolTableSection::getSymbolByIndex(uint32_t Index) { 838 Expected<const Symbol *> Sym = 839 static_cast<const SymbolTableSection *>(this)->getSymbolByIndex(Index); 840 if (!Sym) 841 return Sym.takeError(); 842 843 return const_cast<Symbol *>(*Sym); 844 } 845 846 template <class ELFT> 847 Error ELFSectionWriter<ELFT>::visit(const SymbolTableSection &Sec) { 848 Elf_Sym *Sym = reinterpret_cast<Elf_Sym *>(Out.getBufferStart() + Sec.Offset); 849 // Loop though symbols setting each entry of the symbol table. 850 for (const std::unique_ptr<Symbol> &Symbol : Sec.Symbols) { 851 Sym->st_name = Symbol->NameIndex; 852 Sym->st_value = Symbol->Value; 853 Sym->st_size = Symbol->Size; 854 Sym->st_other = Symbol->Visibility; 855 Sym->setBinding(Symbol->Binding); 856 Sym->setType(Symbol->Type); 857 Sym->st_shndx = Symbol->getShndx(); 858 ++Sym; 859 } 860 return Error::success(); 861 } 862 863 Error SymbolTableSection::accept(SectionVisitor &Visitor) const { 864 return Visitor.visit(*this); 865 } 866 867 Error SymbolTableSection::accept(MutableSectionVisitor &Visitor) { 868 return Visitor.visit(*this); 869 } 870 871 StringRef RelocationSectionBase::getNamePrefix() const { 872 switch (Type) { 873 case SHT_REL: 874 return ".rel"; 875 case SHT_RELA: 876 return ".rela"; 877 default: 878 llvm_unreachable("not a relocation section"); 879 } 880 } 881 882 Error RelocationSection::removeSectionReferences( 883 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 884 if (ToRemove(Symbols)) { 885 if (!AllowBrokenLinks) 886 return createStringError( 887 llvm::errc::invalid_argument, 888 "symbol table '%s' cannot be removed because it is " 889 "referenced by the relocation section '%s'", 890 Symbols->Name.data(), this->Name.data()); 891 Symbols = nullptr; 892 } 893 894 for (const Relocation &R : Relocations) { 895 if (!R.RelocSymbol || !R.RelocSymbol->DefinedIn || 896 !ToRemove(R.RelocSymbol->DefinedIn)) 897 continue; 898 return createStringError(llvm::errc::invalid_argument, 899 "section '%s' cannot be removed: (%s+0x%" PRIx64 900 ") has relocation against symbol '%s'", 901 R.RelocSymbol->DefinedIn->Name.data(), 902 SecToApplyRel->Name.data(), R.Offset, 903 R.RelocSymbol->Name.c_str()); 904 } 905 906 return Error::success(); 907 } 908 909 template <class SymTabType> 910 Error RelocSectionWithSymtabBase<SymTabType>::initialize( 911 SectionTableRef SecTable) { 912 if (Link != SHN_UNDEF) { 913 Expected<SymTabType *> Sec = SecTable.getSectionOfType<SymTabType>( 914 Link, 915 "Link field value " + Twine(Link) + " in section " + Name + 916 " is invalid", 917 "Link field value " + Twine(Link) + " in section " + Name + 918 " is not a symbol table"); 919 if (!Sec) 920 return Sec.takeError(); 921 922 setSymTab(*Sec); 923 } 924 925 if (Info != SHN_UNDEF) { 926 Expected<SectionBase *> Sec = 927 SecTable.getSection(Info, "Info field value " + Twine(Info) + 928 " in section " + Name + " is invalid"); 929 if (!Sec) 930 return Sec.takeError(); 931 932 setSection(*Sec); 933 } else 934 setSection(nullptr); 935 936 return Error::success(); 937 } 938 939 template <class SymTabType> 940 void RelocSectionWithSymtabBase<SymTabType>::finalize() { 941 this->Link = Symbols ? Symbols->Index : 0; 942 943 if (SecToApplyRel != nullptr) 944 this->Info = SecToApplyRel->Index; 945 } 946 947 template <class ELFT> 948 static void setAddend(Elf_Rel_Impl<ELFT, false> &, uint64_t) {} 949 950 template <class ELFT> 951 static void setAddend(Elf_Rel_Impl<ELFT, true> &Rela, uint64_t Addend) { 952 Rela.r_addend = Addend; 953 } 954 955 template <class RelRange, class T> 956 static void writeRel(const RelRange &Relocations, T *Buf, bool IsMips64EL) { 957 for (const auto &Reloc : Relocations) { 958 Buf->r_offset = Reloc.Offset; 959 setAddend(*Buf, Reloc.Addend); 960 Buf->setSymbolAndType(Reloc.RelocSymbol ? Reloc.RelocSymbol->Index : 0, 961 Reloc.Type, IsMips64EL); 962 ++Buf; 963 } 964 } 965 966 template <class ELFT> 967 Error ELFSectionWriter<ELFT>::visit(const RelocationSection &Sec) { 968 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 969 if (Sec.Type == SHT_REL) 970 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rel *>(Buf), 971 Sec.getObject().IsMips64EL); 972 else 973 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rela *>(Buf), 974 Sec.getObject().IsMips64EL); 975 return Error::success(); 976 } 977 978 Error RelocationSection::accept(SectionVisitor &Visitor) const { 979 return Visitor.visit(*this); 980 } 981 982 Error RelocationSection::accept(MutableSectionVisitor &Visitor) { 983 return Visitor.visit(*this); 984 } 985 986 Error RelocationSection::removeSymbols( 987 function_ref<bool(const Symbol &)> ToRemove) { 988 for (const Relocation &Reloc : Relocations) 989 if (Reloc.RelocSymbol && ToRemove(*Reloc.RelocSymbol)) 990 return createStringError( 991 llvm::errc::invalid_argument, 992 "not stripping symbol '%s' because it is named in a relocation", 993 Reloc.RelocSymbol->Name.data()); 994 return Error::success(); 995 } 996 997 void RelocationSection::markSymbols() { 998 for (const Relocation &Reloc : Relocations) 999 if (Reloc.RelocSymbol) 1000 Reloc.RelocSymbol->Referenced = true; 1001 } 1002 1003 void RelocationSection::replaceSectionReferences( 1004 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 1005 // Update the target section if it was replaced. 1006 if (SectionBase *To = FromTo.lookup(SecToApplyRel)) 1007 SecToApplyRel = To; 1008 } 1009 1010 Error SectionWriter::visit(const DynamicRelocationSection &Sec) { 1011 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset); 1012 return Error::success(); 1013 } 1014 1015 Error DynamicRelocationSection::accept(SectionVisitor &Visitor) const { 1016 return Visitor.visit(*this); 1017 } 1018 1019 Error DynamicRelocationSection::accept(MutableSectionVisitor &Visitor) { 1020 return Visitor.visit(*this); 1021 } 1022 1023 Error DynamicRelocationSection::removeSectionReferences( 1024 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 1025 if (ToRemove(Symbols)) { 1026 if (!AllowBrokenLinks) 1027 return createStringError( 1028 llvm::errc::invalid_argument, 1029 "symbol table '%s' cannot be removed because it is " 1030 "referenced by the relocation section '%s'", 1031 Symbols->Name.data(), this->Name.data()); 1032 Symbols = nullptr; 1033 } 1034 1035 // SecToApplyRel contains a section referenced by sh_info field. It keeps 1036 // a section to which the relocation section applies. When we remove any 1037 // sections we also remove their relocation sections. Since we do that much 1038 // earlier, this assert should never be triggered. 1039 assert(!SecToApplyRel || !ToRemove(SecToApplyRel)); 1040 return Error::success(); 1041 } 1042 1043 Error Section::removeSectionReferences( 1044 bool AllowBrokenDependency, 1045 function_ref<bool(const SectionBase *)> ToRemove) { 1046 if (ToRemove(LinkSection)) { 1047 if (!AllowBrokenDependency) 1048 return createStringError(llvm::errc::invalid_argument, 1049 "section '%s' cannot be removed because it is " 1050 "referenced by the section '%s'", 1051 LinkSection->Name.data(), this->Name.data()); 1052 LinkSection = nullptr; 1053 } 1054 return Error::success(); 1055 } 1056 1057 void GroupSection::finalize() { 1058 this->Info = Sym ? Sym->Index : 0; 1059 this->Link = SymTab ? SymTab->Index : 0; 1060 // Linker deduplication for GRP_COMDAT is based on Sym->Name. The local/global 1061 // status is not part of the equation. If Sym is localized, the intention is 1062 // likely to make the group fully localized. Drop GRP_COMDAT to suppress 1063 // deduplication. See https://groups.google.com/g/generic-abi/c/2X6mR-s2zoc 1064 if ((FlagWord & GRP_COMDAT) && Sym && Sym->Binding == STB_LOCAL) 1065 this->FlagWord &= ~GRP_COMDAT; 1066 } 1067 1068 Error GroupSection::removeSectionReferences( 1069 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 1070 if (ToRemove(SymTab)) { 1071 if (!AllowBrokenLinks) 1072 return createStringError( 1073 llvm::errc::invalid_argument, 1074 "section '.symtab' cannot be removed because it is " 1075 "referenced by the group section '%s'", 1076 this->Name.data()); 1077 SymTab = nullptr; 1078 Sym = nullptr; 1079 } 1080 llvm::erase_if(GroupMembers, ToRemove); 1081 return Error::success(); 1082 } 1083 1084 Error GroupSection::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) { 1085 if (ToRemove(*Sym)) 1086 return createStringError(llvm::errc::invalid_argument, 1087 "symbol '%s' cannot be removed because it is " 1088 "referenced by the section '%s[%d]'", 1089 Sym->Name.data(), this->Name.data(), this->Index); 1090 return Error::success(); 1091 } 1092 1093 void GroupSection::markSymbols() { 1094 if (Sym) 1095 Sym->Referenced = true; 1096 } 1097 1098 void GroupSection::replaceSectionReferences( 1099 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 1100 for (SectionBase *&Sec : GroupMembers) 1101 if (SectionBase *To = FromTo.lookup(Sec)) 1102 Sec = To; 1103 } 1104 1105 void GroupSection::onRemove() { 1106 // As the header section of the group is removed, drop the Group flag in its 1107 // former members. 1108 for (SectionBase *Sec : GroupMembers) 1109 Sec->Flags &= ~SHF_GROUP; 1110 } 1111 1112 Error Section::initialize(SectionTableRef SecTable) { 1113 if (Link == ELF::SHN_UNDEF) 1114 return Error::success(); 1115 1116 Expected<SectionBase *> Sec = 1117 SecTable.getSection(Link, "Link field value " + Twine(Link) + 1118 " in section " + Name + " is invalid"); 1119 if (!Sec) 1120 return Sec.takeError(); 1121 1122 LinkSection = *Sec; 1123 1124 if (LinkSection->Type == ELF::SHT_SYMTAB) 1125 LinkSection = nullptr; 1126 1127 return Error::success(); 1128 } 1129 1130 void Section::finalize() { this->Link = LinkSection ? LinkSection->Index : 0; } 1131 1132 void GnuDebugLinkSection::init(StringRef File) { 1133 FileName = sys::path::filename(File); 1134 // The format for the .gnu_debuglink starts with the file name and is 1135 // followed by a null terminator and then the CRC32 of the file. The CRC32 1136 // should be 4 byte aligned. So we add the FileName size, a 1 for the null 1137 // byte, and then finally push the size to alignment and add 4. 1138 Size = alignTo(FileName.size() + 1, 4) + 4; 1139 // The CRC32 will only be aligned if we align the whole section. 1140 Align = 4; 1141 Type = OriginalType = ELF::SHT_PROGBITS; 1142 Name = ".gnu_debuglink"; 1143 // For sections not found in segments, OriginalOffset is only used to 1144 // establish the order that sections should go in. By using the maximum 1145 // possible offset we cause this section to wind up at the end. 1146 OriginalOffset = std::numeric_limits<uint64_t>::max(); 1147 } 1148 1149 GnuDebugLinkSection::GnuDebugLinkSection(StringRef File, 1150 uint32_t PrecomputedCRC) 1151 : FileName(File), CRC32(PrecomputedCRC) { 1152 init(File); 1153 } 1154 1155 template <class ELFT> 1156 Error ELFSectionWriter<ELFT>::visit(const GnuDebugLinkSection &Sec) { 1157 unsigned char *Buf = 1158 reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 1159 Elf_Word *CRC = 1160 reinterpret_cast<Elf_Word *>(Buf + Sec.Size - sizeof(Elf_Word)); 1161 *CRC = Sec.CRC32; 1162 llvm::copy(Sec.FileName, Buf); 1163 return Error::success(); 1164 } 1165 1166 Error GnuDebugLinkSection::accept(SectionVisitor &Visitor) const { 1167 return Visitor.visit(*this); 1168 } 1169 1170 Error GnuDebugLinkSection::accept(MutableSectionVisitor &Visitor) { 1171 return Visitor.visit(*this); 1172 } 1173 1174 template <class ELFT> 1175 Error ELFSectionWriter<ELFT>::visit(const GroupSection &Sec) { 1176 ELF::Elf32_Word *Buf = 1177 reinterpret_cast<ELF::Elf32_Word *>(Out.getBufferStart() + Sec.Offset); 1178 support::endian::write32<ELFT::TargetEndianness>(Buf++, Sec.FlagWord); 1179 for (SectionBase *S : Sec.GroupMembers) 1180 support::endian::write32<ELFT::TargetEndianness>(Buf++, S->Index); 1181 return Error::success(); 1182 } 1183 1184 Error GroupSection::accept(SectionVisitor &Visitor) const { 1185 return Visitor.visit(*this); 1186 } 1187 1188 Error GroupSection::accept(MutableSectionVisitor &Visitor) { 1189 return Visitor.visit(*this); 1190 } 1191 1192 // Returns true IFF a section is wholly inside the range of a segment 1193 static bool sectionWithinSegment(const SectionBase &Sec, const Segment &Seg) { 1194 // If a section is empty it should be treated like it has a size of 1. This is 1195 // to clarify the case when an empty section lies on a boundary between two 1196 // segments and ensures that the section "belongs" to the second segment and 1197 // not the first. 1198 uint64_t SecSize = Sec.Size ? Sec.Size : 1; 1199 1200 // Ignore just added sections. 1201 if (Sec.OriginalOffset == std::numeric_limits<uint64_t>::max()) 1202 return false; 1203 1204 if (Sec.Type == SHT_NOBITS) { 1205 if (!(Sec.Flags & SHF_ALLOC)) 1206 return false; 1207 1208 bool SectionIsTLS = Sec.Flags & SHF_TLS; 1209 bool SegmentIsTLS = Seg.Type == PT_TLS; 1210 if (SectionIsTLS != SegmentIsTLS) 1211 return false; 1212 1213 return Seg.VAddr <= Sec.Addr && 1214 Seg.VAddr + Seg.MemSize >= Sec.Addr + SecSize; 1215 } 1216 1217 return Seg.Offset <= Sec.OriginalOffset && 1218 Seg.Offset + Seg.FileSize >= Sec.OriginalOffset + SecSize; 1219 } 1220 1221 // Returns true IFF a segment's original offset is inside of another segment's 1222 // range. 1223 static bool segmentOverlapsSegment(const Segment &Child, 1224 const Segment &Parent) { 1225 1226 return Parent.OriginalOffset <= Child.OriginalOffset && 1227 Parent.OriginalOffset + Parent.FileSize > Child.OriginalOffset; 1228 } 1229 1230 static bool compareSegmentsByOffset(const Segment *A, const Segment *B) { 1231 // Any segment without a parent segment should come before a segment 1232 // that has a parent segment. 1233 if (A->OriginalOffset < B->OriginalOffset) 1234 return true; 1235 if (A->OriginalOffset > B->OriginalOffset) 1236 return false; 1237 return A->Index < B->Index; 1238 } 1239 1240 void BasicELFBuilder::initFileHeader() { 1241 Obj->Flags = 0x0; 1242 Obj->Type = ET_REL; 1243 Obj->OSABI = ELFOSABI_NONE; 1244 Obj->ABIVersion = 0; 1245 Obj->Entry = 0x0; 1246 Obj->Machine = EM_NONE; 1247 Obj->Version = 1; 1248 } 1249 1250 void BasicELFBuilder::initHeaderSegment() { Obj->ElfHdrSegment.Index = 0; } 1251 1252 StringTableSection *BasicELFBuilder::addStrTab() { 1253 auto &StrTab = Obj->addSection<StringTableSection>(); 1254 StrTab.Name = ".strtab"; 1255 1256 Obj->SectionNames = &StrTab; 1257 return &StrTab; 1258 } 1259 1260 SymbolTableSection *BasicELFBuilder::addSymTab(StringTableSection *StrTab) { 1261 auto &SymTab = Obj->addSection<SymbolTableSection>(); 1262 1263 SymTab.Name = ".symtab"; 1264 SymTab.Link = StrTab->Index; 1265 1266 // The symbol table always needs a null symbol 1267 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0); 1268 1269 Obj->SymbolTable = &SymTab; 1270 return &SymTab; 1271 } 1272 1273 Error BasicELFBuilder::initSections() { 1274 for (SectionBase &Sec : Obj->sections()) 1275 if (Error Err = Sec.initialize(Obj->sections())) 1276 return Err; 1277 1278 return Error::success(); 1279 } 1280 1281 void BinaryELFBuilder::addData(SymbolTableSection *SymTab) { 1282 auto Data = ArrayRef<uint8_t>( 1283 reinterpret_cast<const uint8_t *>(MemBuf->getBufferStart()), 1284 MemBuf->getBufferSize()); 1285 auto &DataSection = Obj->addSection<Section>(Data); 1286 DataSection.Name = ".data"; 1287 DataSection.Type = ELF::SHT_PROGBITS; 1288 DataSection.Size = Data.size(); 1289 DataSection.Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE; 1290 1291 std::string SanitizedFilename = MemBuf->getBufferIdentifier().str(); 1292 std::replace_if( 1293 std::begin(SanitizedFilename), std::end(SanitizedFilename), 1294 [](char C) { return !isAlnum(C); }, '_'); 1295 Twine Prefix = Twine("_binary_") + SanitizedFilename; 1296 1297 SymTab->addSymbol(Prefix + "_start", STB_GLOBAL, STT_NOTYPE, &DataSection, 1298 /*Value=*/0, NewSymbolVisibility, 0, 0); 1299 SymTab->addSymbol(Prefix + "_end", STB_GLOBAL, STT_NOTYPE, &DataSection, 1300 /*Value=*/DataSection.Size, NewSymbolVisibility, 0, 0); 1301 SymTab->addSymbol(Prefix + "_size", STB_GLOBAL, STT_NOTYPE, nullptr, 1302 /*Value=*/DataSection.Size, NewSymbolVisibility, SHN_ABS, 1303 0); 1304 } 1305 1306 Expected<std::unique_ptr<Object>> BinaryELFBuilder::build() { 1307 initFileHeader(); 1308 initHeaderSegment(); 1309 1310 SymbolTableSection *SymTab = addSymTab(addStrTab()); 1311 if (Error Err = initSections()) 1312 return std::move(Err); 1313 addData(SymTab); 1314 1315 return std::move(Obj); 1316 } 1317 1318 // Adds sections from IHEX data file. Data should have been 1319 // fully validated by this time. 1320 void IHexELFBuilder::addDataSections() { 1321 OwnedDataSection *Section = nullptr; 1322 uint64_t SegmentAddr = 0, BaseAddr = 0; 1323 uint32_t SecNo = 1; 1324 1325 for (const IHexRecord &R : Records) { 1326 uint64_t RecAddr; 1327 switch (R.Type) { 1328 case IHexRecord::Data: 1329 // Ignore empty data records 1330 if (R.HexData.empty()) 1331 continue; 1332 RecAddr = R.Addr + SegmentAddr + BaseAddr; 1333 if (!Section || Section->Addr + Section->Size != RecAddr) { 1334 // OriginalOffset field is only used to sort sections before layout, so 1335 // instead of keeping track of real offsets in IHEX file, and as 1336 // layoutSections() and layoutSectionsForOnlyKeepDebug() use 1337 // llvm::stable_sort(), we can just set it to a constant (zero). 1338 Section = &Obj->addSection<OwnedDataSection>( 1339 ".sec" + std::to_string(SecNo), RecAddr, 1340 ELF::SHF_ALLOC | ELF::SHF_WRITE, 0); 1341 SecNo++; 1342 } 1343 Section->appendHexData(R.HexData); 1344 break; 1345 case IHexRecord::EndOfFile: 1346 break; 1347 case IHexRecord::SegmentAddr: 1348 // 20-bit segment address. 1349 SegmentAddr = checkedGetHex<uint16_t>(R.HexData) << 4; 1350 break; 1351 case IHexRecord::StartAddr80x86: 1352 case IHexRecord::StartAddr: 1353 Obj->Entry = checkedGetHex<uint32_t>(R.HexData); 1354 assert(Obj->Entry <= 0xFFFFFU); 1355 break; 1356 case IHexRecord::ExtendedAddr: 1357 // 16-31 bits of linear base address 1358 BaseAddr = checkedGetHex<uint16_t>(R.HexData) << 16; 1359 break; 1360 default: 1361 llvm_unreachable("unknown record type"); 1362 } 1363 } 1364 } 1365 1366 Expected<std::unique_ptr<Object>> IHexELFBuilder::build() { 1367 initFileHeader(); 1368 initHeaderSegment(); 1369 StringTableSection *StrTab = addStrTab(); 1370 addSymTab(StrTab); 1371 if (Error Err = initSections()) 1372 return std::move(Err); 1373 addDataSections(); 1374 1375 return std::move(Obj); 1376 } 1377 1378 template <class ELFT> 1379 ELFBuilder<ELFT>::ELFBuilder(const ELFObjectFile<ELFT> &ElfObj, Object &Obj, 1380 Optional<StringRef> ExtractPartition) 1381 : ElfFile(ElfObj.getELFFile()), Obj(Obj), 1382 ExtractPartition(ExtractPartition) { 1383 Obj.IsMips64EL = ElfFile.isMips64EL(); 1384 } 1385 1386 template <class ELFT> void ELFBuilder<ELFT>::setParentSegment(Segment &Child) { 1387 for (Segment &Parent : Obj.segments()) { 1388 // Every segment will overlap with itself but we don't want a segment to 1389 // be its own parent so we avoid that situation. 1390 if (&Child != &Parent && segmentOverlapsSegment(Child, Parent)) { 1391 // We want a canonical "most parental" segment but this requires 1392 // inspecting the ParentSegment. 1393 if (compareSegmentsByOffset(&Parent, &Child)) 1394 if (Child.ParentSegment == nullptr || 1395 compareSegmentsByOffset(&Parent, Child.ParentSegment)) { 1396 Child.ParentSegment = &Parent; 1397 } 1398 } 1399 } 1400 } 1401 1402 template <class ELFT> Error ELFBuilder<ELFT>::findEhdrOffset() { 1403 if (!ExtractPartition) 1404 return Error::success(); 1405 1406 for (const SectionBase &Sec : Obj.sections()) { 1407 if (Sec.Type == SHT_LLVM_PART_EHDR && Sec.Name == *ExtractPartition) { 1408 EhdrOffset = Sec.Offset; 1409 return Error::success(); 1410 } 1411 } 1412 return createStringError(errc::invalid_argument, 1413 "could not find partition named '" + 1414 *ExtractPartition + "'"); 1415 } 1416 1417 template <class ELFT> 1418 Error ELFBuilder<ELFT>::readProgramHeaders(const ELFFile<ELFT> &HeadersFile) { 1419 uint32_t Index = 0; 1420 1421 Expected<typename ELFFile<ELFT>::Elf_Phdr_Range> Headers = 1422 HeadersFile.program_headers(); 1423 if (!Headers) 1424 return Headers.takeError(); 1425 1426 for (const typename ELFFile<ELFT>::Elf_Phdr &Phdr : *Headers) { 1427 if (Phdr.p_offset + Phdr.p_filesz > HeadersFile.getBufSize()) 1428 return createStringError( 1429 errc::invalid_argument, 1430 "program header with offset 0x" + Twine::utohexstr(Phdr.p_offset) + 1431 " and file size 0x" + Twine::utohexstr(Phdr.p_filesz) + 1432 " goes past the end of the file"); 1433 1434 ArrayRef<uint8_t> Data{HeadersFile.base() + Phdr.p_offset, 1435 (size_t)Phdr.p_filesz}; 1436 Segment &Seg = Obj.addSegment(Data); 1437 Seg.Type = Phdr.p_type; 1438 Seg.Flags = Phdr.p_flags; 1439 Seg.OriginalOffset = Phdr.p_offset + EhdrOffset; 1440 Seg.Offset = Phdr.p_offset + EhdrOffset; 1441 Seg.VAddr = Phdr.p_vaddr; 1442 Seg.PAddr = Phdr.p_paddr; 1443 Seg.FileSize = Phdr.p_filesz; 1444 Seg.MemSize = Phdr.p_memsz; 1445 Seg.Align = Phdr.p_align; 1446 Seg.Index = Index++; 1447 for (SectionBase &Sec : Obj.sections()) 1448 if (sectionWithinSegment(Sec, Seg)) { 1449 Seg.addSection(&Sec); 1450 if (!Sec.ParentSegment || Sec.ParentSegment->Offset > Seg.Offset) 1451 Sec.ParentSegment = &Seg; 1452 } 1453 } 1454 1455 auto &ElfHdr = Obj.ElfHdrSegment; 1456 ElfHdr.Index = Index++; 1457 ElfHdr.OriginalOffset = ElfHdr.Offset = EhdrOffset; 1458 1459 const typename ELFT::Ehdr &Ehdr = HeadersFile.getHeader(); 1460 auto &PrHdr = Obj.ProgramHdrSegment; 1461 PrHdr.Type = PT_PHDR; 1462 PrHdr.Flags = 0; 1463 // The spec requires us to have p_vaddr % p_align == p_offset % p_align. 1464 // Whereas this works automatically for ElfHdr, here OriginalOffset is 1465 // always non-zero and to ensure the equation we assign the same value to 1466 // VAddr as well. 1467 PrHdr.OriginalOffset = PrHdr.Offset = PrHdr.VAddr = EhdrOffset + Ehdr.e_phoff; 1468 PrHdr.PAddr = 0; 1469 PrHdr.FileSize = PrHdr.MemSize = Ehdr.e_phentsize * Ehdr.e_phnum; 1470 // The spec requires us to naturally align all the fields. 1471 PrHdr.Align = sizeof(Elf_Addr); 1472 PrHdr.Index = Index++; 1473 1474 // Now we do an O(n^2) loop through the segments in order to match up 1475 // segments. 1476 for (Segment &Child : Obj.segments()) 1477 setParentSegment(Child); 1478 setParentSegment(ElfHdr); 1479 setParentSegment(PrHdr); 1480 1481 return Error::success(); 1482 } 1483 1484 template <class ELFT> 1485 Error ELFBuilder<ELFT>::initGroupSection(GroupSection *GroupSec) { 1486 if (GroupSec->Align % sizeof(ELF::Elf32_Word) != 0) 1487 return createStringError(errc::invalid_argument, 1488 "invalid alignment " + Twine(GroupSec->Align) + 1489 " of group section '" + GroupSec->Name + "'"); 1490 SectionTableRef SecTable = Obj.sections(); 1491 if (GroupSec->Link != SHN_UNDEF) { 1492 auto SymTab = SecTable.template getSectionOfType<SymbolTableSection>( 1493 GroupSec->Link, 1494 "link field value '" + Twine(GroupSec->Link) + "' in section '" + 1495 GroupSec->Name + "' is invalid", 1496 "link field value '" + Twine(GroupSec->Link) + "' in section '" + 1497 GroupSec->Name + "' is not a symbol table"); 1498 if (!SymTab) 1499 return SymTab.takeError(); 1500 1501 Expected<Symbol *> Sym = (*SymTab)->getSymbolByIndex(GroupSec->Info); 1502 if (!Sym) 1503 return createStringError(errc::invalid_argument, 1504 "info field value '" + Twine(GroupSec->Info) + 1505 "' in section '" + GroupSec->Name + 1506 "' is not a valid symbol index"); 1507 GroupSec->setSymTab(*SymTab); 1508 GroupSec->setSymbol(*Sym); 1509 } 1510 if (GroupSec->Contents.size() % sizeof(ELF::Elf32_Word) || 1511 GroupSec->Contents.empty()) 1512 return createStringError(errc::invalid_argument, 1513 "the content of the section " + GroupSec->Name + 1514 " is malformed"); 1515 const ELF::Elf32_Word *Word = 1516 reinterpret_cast<const ELF::Elf32_Word *>(GroupSec->Contents.data()); 1517 const ELF::Elf32_Word *End = 1518 Word + GroupSec->Contents.size() / sizeof(ELF::Elf32_Word); 1519 GroupSec->setFlagWord( 1520 support::endian::read32<ELFT::TargetEndianness>(Word++)); 1521 for (; Word != End; ++Word) { 1522 uint32_t Index = support::endian::read32<ELFT::TargetEndianness>(Word); 1523 Expected<SectionBase *> Sec = SecTable.getSection( 1524 Index, "group member index " + Twine(Index) + " in section '" + 1525 GroupSec->Name + "' is invalid"); 1526 if (!Sec) 1527 return Sec.takeError(); 1528 1529 GroupSec->addMember(*Sec); 1530 } 1531 1532 return Error::success(); 1533 } 1534 1535 template <class ELFT> 1536 Error ELFBuilder<ELFT>::initSymbolTable(SymbolTableSection *SymTab) { 1537 Expected<const Elf_Shdr *> Shdr = ElfFile.getSection(SymTab->Index); 1538 if (!Shdr) 1539 return Shdr.takeError(); 1540 1541 Expected<StringRef> StrTabData = ElfFile.getStringTableForSymtab(**Shdr); 1542 if (!StrTabData) 1543 return StrTabData.takeError(); 1544 1545 ArrayRef<Elf_Word> ShndxData; 1546 1547 Expected<typename ELFFile<ELFT>::Elf_Sym_Range> Symbols = 1548 ElfFile.symbols(*Shdr); 1549 if (!Symbols) 1550 return Symbols.takeError(); 1551 1552 for (const typename ELFFile<ELFT>::Elf_Sym &Sym : *Symbols) { 1553 SectionBase *DefSection = nullptr; 1554 1555 Expected<StringRef> Name = Sym.getName(*StrTabData); 1556 if (!Name) 1557 return Name.takeError(); 1558 1559 if (Sym.st_shndx == SHN_XINDEX) { 1560 if (SymTab->getShndxTable() == nullptr) 1561 return createStringError(errc::invalid_argument, 1562 "symbol '" + *Name + 1563 "' has index SHN_XINDEX but no " 1564 "SHT_SYMTAB_SHNDX section exists"); 1565 if (ShndxData.data() == nullptr) { 1566 Expected<const Elf_Shdr *> ShndxSec = 1567 ElfFile.getSection(SymTab->getShndxTable()->Index); 1568 if (!ShndxSec) 1569 return ShndxSec.takeError(); 1570 1571 Expected<ArrayRef<Elf_Word>> Data = 1572 ElfFile.template getSectionContentsAsArray<Elf_Word>(**ShndxSec); 1573 if (!Data) 1574 return Data.takeError(); 1575 1576 ShndxData = *Data; 1577 if (ShndxData.size() != Symbols->size()) 1578 return createStringError( 1579 errc::invalid_argument, 1580 "symbol section index table does not have the same number of " 1581 "entries as the symbol table"); 1582 } 1583 Elf_Word Index = ShndxData[&Sym - Symbols->begin()]; 1584 Expected<SectionBase *> Sec = Obj.sections().getSection( 1585 Index, 1586 "symbol '" + *Name + "' has invalid section index " + Twine(Index)); 1587 if (!Sec) 1588 return Sec.takeError(); 1589 1590 DefSection = *Sec; 1591 } else if (Sym.st_shndx >= SHN_LORESERVE) { 1592 if (!isValidReservedSectionIndex(Sym.st_shndx, Obj.Machine)) { 1593 return createStringError( 1594 errc::invalid_argument, 1595 "symbol '" + *Name + 1596 "' has unsupported value greater than or equal " 1597 "to SHN_LORESERVE: " + 1598 Twine(Sym.st_shndx)); 1599 } 1600 } else if (Sym.st_shndx != SHN_UNDEF) { 1601 Expected<SectionBase *> Sec = Obj.sections().getSection( 1602 Sym.st_shndx, "symbol '" + *Name + 1603 "' is defined has invalid section index " + 1604 Twine(Sym.st_shndx)); 1605 if (!Sec) 1606 return Sec.takeError(); 1607 1608 DefSection = *Sec; 1609 } 1610 1611 SymTab->addSymbol(*Name, Sym.getBinding(), Sym.getType(), DefSection, 1612 Sym.getValue(), Sym.st_other, Sym.st_shndx, Sym.st_size); 1613 } 1614 1615 return Error::success(); 1616 } 1617 1618 template <class ELFT> 1619 static void getAddend(uint64_t &, const Elf_Rel_Impl<ELFT, false> &) {} 1620 1621 template <class ELFT> 1622 static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, true> &Rela) { 1623 ToSet = Rela.r_addend; 1624 } 1625 1626 template <class T> 1627 static Error initRelocations(RelocationSection *Relocs, T RelRange) { 1628 for (const auto &Rel : RelRange) { 1629 Relocation ToAdd; 1630 ToAdd.Offset = Rel.r_offset; 1631 getAddend(ToAdd.Addend, Rel); 1632 ToAdd.Type = Rel.getType(Relocs->getObject().IsMips64EL); 1633 1634 if (uint32_t Sym = Rel.getSymbol(Relocs->getObject().IsMips64EL)) { 1635 if (!Relocs->getObject().SymbolTable) 1636 return createStringError( 1637 errc::invalid_argument, 1638 "'" + Relocs->Name + "': relocation references symbol with index " + 1639 Twine(Sym) + ", but there is no symbol table"); 1640 Expected<Symbol *> SymByIndex = 1641 Relocs->getObject().SymbolTable->getSymbolByIndex(Sym); 1642 if (!SymByIndex) 1643 return SymByIndex.takeError(); 1644 1645 ToAdd.RelocSymbol = *SymByIndex; 1646 } 1647 1648 Relocs->addRelocation(ToAdd); 1649 } 1650 1651 return Error::success(); 1652 } 1653 1654 Expected<SectionBase *> SectionTableRef::getSection(uint32_t Index, 1655 Twine ErrMsg) { 1656 if (Index == SHN_UNDEF || Index > Sections.size()) 1657 return createStringError(errc::invalid_argument, ErrMsg); 1658 return Sections[Index - 1].get(); 1659 } 1660 1661 template <class T> 1662 Expected<T *> SectionTableRef::getSectionOfType(uint32_t Index, 1663 Twine IndexErrMsg, 1664 Twine TypeErrMsg) { 1665 Expected<SectionBase *> BaseSec = getSection(Index, IndexErrMsg); 1666 if (!BaseSec) 1667 return BaseSec.takeError(); 1668 1669 if (T *Sec = dyn_cast<T>(*BaseSec)) 1670 return Sec; 1671 1672 return createStringError(errc::invalid_argument, TypeErrMsg); 1673 } 1674 1675 template <class ELFT> 1676 Expected<SectionBase &> ELFBuilder<ELFT>::makeSection(const Elf_Shdr &Shdr) { 1677 switch (Shdr.sh_type) { 1678 case SHT_REL: 1679 case SHT_RELA: 1680 if (Shdr.sh_flags & SHF_ALLOC) { 1681 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1682 return Obj.addSection<DynamicRelocationSection>(*Data); 1683 else 1684 return Data.takeError(); 1685 } 1686 return Obj.addSection<RelocationSection>(Obj); 1687 case SHT_STRTAB: 1688 // If a string table is allocated we don't want to mess with it. That would 1689 // mean altering the memory image. There are no special link types or 1690 // anything so we can just use a Section. 1691 if (Shdr.sh_flags & SHF_ALLOC) { 1692 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1693 return Obj.addSection<Section>(*Data); 1694 else 1695 return Data.takeError(); 1696 } 1697 return Obj.addSection<StringTableSection>(); 1698 case SHT_HASH: 1699 case SHT_GNU_HASH: 1700 // Hash tables should refer to SHT_DYNSYM which we're not going to change. 1701 // Because of this we don't need to mess with the hash tables either. 1702 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1703 return Obj.addSection<Section>(*Data); 1704 else 1705 return Data.takeError(); 1706 case SHT_GROUP: 1707 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1708 return Obj.addSection<GroupSection>(*Data); 1709 else 1710 return Data.takeError(); 1711 case SHT_DYNSYM: 1712 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1713 return Obj.addSection<DynamicSymbolTableSection>(*Data); 1714 else 1715 return Data.takeError(); 1716 case SHT_DYNAMIC: 1717 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1718 return Obj.addSection<DynamicSection>(*Data); 1719 else 1720 return Data.takeError(); 1721 case SHT_SYMTAB: { 1722 auto &SymTab = Obj.addSection<SymbolTableSection>(); 1723 Obj.SymbolTable = &SymTab; 1724 return SymTab; 1725 } 1726 case SHT_SYMTAB_SHNDX: { 1727 auto &ShndxSection = Obj.addSection<SectionIndexSection>(); 1728 Obj.SectionIndexTable = &ShndxSection; 1729 return ShndxSection; 1730 } 1731 case SHT_NOBITS: 1732 return Obj.addSection<Section>(ArrayRef<uint8_t>()); 1733 default: { 1734 Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr); 1735 if (!Data) 1736 return Data.takeError(); 1737 1738 Expected<StringRef> Name = ElfFile.getSectionName(Shdr); 1739 if (!Name) 1740 return Name.takeError(); 1741 1742 if (Name->startswith(".zdebug") || (Shdr.sh_flags & ELF::SHF_COMPRESSED)) { 1743 uint64_t DecompressedSize, DecompressedAlign; 1744 std::tie(DecompressedSize, DecompressedAlign) = 1745 getDecompressedSizeAndAlignment<ELFT>(*Data); 1746 return Obj.addSection<CompressedSection>( 1747 CompressedSection(*Data, DecompressedSize, DecompressedAlign)); 1748 } 1749 1750 return Obj.addSection<Section>(*Data); 1751 } 1752 } 1753 } 1754 1755 template <class ELFT> Error ELFBuilder<ELFT>::readSectionHeaders() { 1756 uint32_t Index = 0; 1757 Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections = 1758 ElfFile.sections(); 1759 if (!Sections) 1760 return Sections.takeError(); 1761 1762 for (const typename ELFFile<ELFT>::Elf_Shdr &Shdr : *Sections) { 1763 if (Index == 0) { 1764 ++Index; 1765 continue; 1766 } 1767 Expected<SectionBase &> Sec = makeSection(Shdr); 1768 if (!Sec) 1769 return Sec.takeError(); 1770 1771 Expected<StringRef> SecName = ElfFile.getSectionName(Shdr); 1772 if (!SecName) 1773 return SecName.takeError(); 1774 Sec->Name = SecName->str(); 1775 Sec->Type = Sec->OriginalType = Shdr.sh_type; 1776 Sec->Flags = Sec->OriginalFlags = Shdr.sh_flags; 1777 Sec->Addr = Shdr.sh_addr; 1778 Sec->Offset = Shdr.sh_offset; 1779 Sec->OriginalOffset = Shdr.sh_offset; 1780 Sec->Size = Shdr.sh_size; 1781 Sec->Link = Shdr.sh_link; 1782 Sec->Info = Shdr.sh_info; 1783 Sec->Align = Shdr.sh_addralign; 1784 Sec->EntrySize = Shdr.sh_entsize; 1785 Sec->Index = Index++; 1786 Sec->OriginalIndex = Sec->Index; 1787 Sec->OriginalData = ArrayRef<uint8_t>( 1788 ElfFile.base() + Shdr.sh_offset, 1789 (Shdr.sh_type == SHT_NOBITS) ? (size_t)0 : Shdr.sh_size); 1790 } 1791 1792 return Error::success(); 1793 } 1794 1795 template <class ELFT> Error ELFBuilder<ELFT>::readSections(bool EnsureSymtab) { 1796 uint32_t ShstrIndex = ElfFile.getHeader().e_shstrndx; 1797 if (ShstrIndex == SHN_XINDEX) { 1798 Expected<const Elf_Shdr *> Sec = ElfFile.getSection(0); 1799 if (!Sec) 1800 return Sec.takeError(); 1801 1802 ShstrIndex = (*Sec)->sh_link; 1803 } 1804 1805 if (ShstrIndex == SHN_UNDEF) 1806 Obj.HadShdrs = false; 1807 else { 1808 Expected<StringTableSection *> Sec = 1809 Obj.sections().template getSectionOfType<StringTableSection>( 1810 ShstrIndex, 1811 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " + 1812 " is invalid", 1813 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " + 1814 " does not reference a string table"); 1815 if (!Sec) 1816 return Sec.takeError(); 1817 1818 Obj.SectionNames = *Sec; 1819 } 1820 1821 // If a section index table exists we'll need to initialize it before we 1822 // initialize the symbol table because the symbol table might need to 1823 // reference it. 1824 if (Obj.SectionIndexTable) 1825 if (Error Err = Obj.SectionIndexTable->initialize(Obj.sections())) 1826 return Err; 1827 1828 // Now that all of the sections have been added we can fill out some extra 1829 // details about symbol tables. We need the symbol table filled out before 1830 // any relocations. 1831 if (Obj.SymbolTable) { 1832 if (Error Err = Obj.SymbolTable->initialize(Obj.sections())) 1833 return Err; 1834 if (Error Err = initSymbolTable(Obj.SymbolTable)) 1835 return Err; 1836 } else if (EnsureSymtab) { 1837 if (Error Err = Obj.addNewSymbolTable()) 1838 return Err; 1839 } 1840 1841 // Now that all sections and symbols have been added we can add 1842 // relocations that reference symbols and set the link and info fields for 1843 // relocation sections. 1844 for (SectionBase &Sec : Obj.sections()) { 1845 if (&Sec == Obj.SymbolTable) 1846 continue; 1847 if (Error Err = Sec.initialize(Obj.sections())) 1848 return Err; 1849 if (auto RelSec = dyn_cast<RelocationSection>(&Sec)) { 1850 Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections = 1851 ElfFile.sections(); 1852 if (!Sections) 1853 return Sections.takeError(); 1854 1855 const typename ELFFile<ELFT>::Elf_Shdr *Shdr = 1856 Sections->begin() + RelSec->Index; 1857 if (RelSec->Type == SHT_REL) { 1858 Expected<typename ELFFile<ELFT>::Elf_Rel_Range> Rels = 1859 ElfFile.rels(*Shdr); 1860 if (!Rels) 1861 return Rels.takeError(); 1862 1863 if (Error Err = initRelocations(RelSec, *Rels)) 1864 return Err; 1865 } else { 1866 Expected<typename ELFFile<ELFT>::Elf_Rela_Range> Relas = 1867 ElfFile.relas(*Shdr); 1868 if (!Relas) 1869 return Relas.takeError(); 1870 1871 if (Error Err = initRelocations(RelSec, *Relas)) 1872 return Err; 1873 } 1874 } else if (auto GroupSec = dyn_cast<GroupSection>(&Sec)) { 1875 if (Error Err = initGroupSection(GroupSec)) 1876 return Err; 1877 } 1878 } 1879 1880 return Error::success(); 1881 } 1882 1883 template <class ELFT> Error ELFBuilder<ELFT>::build(bool EnsureSymtab) { 1884 if (Error E = readSectionHeaders()) 1885 return E; 1886 if (Error E = findEhdrOffset()) 1887 return E; 1888 1889 // The ELFFile whose ELF headers and program headers are copied into the 1890 // output file. Normally the same as ElfFile, but if we're extracting a 1891 // loadable partition it will point to the partition's headers. 1892 Expected<ELFFile<ELFT>> HeadersFile = ELFFile<ELFT>::create(toStringRef( 1893 {ElfFile.base() + EhdrOffset, ElfFile.getBufSize() - EhdrOffset})); 1894 if (!HeadersFile) 1895 return HeadersFile.takeError(); 1896 1897 const typename ELFFile<ELFT>::Elf_Ehdr &Ehdr = HeadersFile->getHeader(); 1898 Obj.OSABI = Ehdr.e_ident[EI_OSABI]; 1899 Obj.ABIVersion = Ehdr.e_ident[EI_ABIVERSION]; 1900 Obj.Type = Ehdr.e_type; 1901 Obj.Machine = Ehdr.e_machine; 1902 Obj.Version = Ehdr.e_version; 1903 Obj.Entry = Ehdr.e_entry; 1904 Obj.Flags = Ehdr.e_flags; 1905 1906 if (Error E = readSections(EnsureSymtab)) 1907 return E; 1908 return readProgramHeaders(*HeadersFile); 1909 } 1910 1911 Writer::~Writer() = default; 1912 1913 Reader::~Reader() = default; 1914 1915 Expected<std::unique_ptr<Object>> 1916 BinaryReader::create(bool /*EnsureSymtab*/) const { 1917 return BinaryELFBuilder(MemBuf, NewSymbolVisibility).build(); 1918 } 1919 1920 Expected<std::vector<IHexRecord>> IHexReader::parse() const { 1921 SmallVector<StringRef, 16> Lines; 1922 std::vector<IHexRecord> Records; 1923 bool HasSections = false; 1924 1925 MemBuf->getBuffer().split(Lines, '\n'); 1926 Records.reserve(Lines.size()); 1927 for (size_t LineNo = 1; LineNo <= Lines.size(); ++LineNo) { 1928 StringRef Line = Lines[LineNo - 1].trim(); 1929 if (Line.empty()) 1930 continue; 1931 1932 Expected<IHexRecord> R = IHexRecord::parse(Line); 1933 if (!R) 1934 return parseError(LineNo, R.takeError()); 1935 if (R->Type == IHexRecord::EndOfFile) 1936 break; 1937 HasSections |= (R->Type == IHexRecord::Data); 1938 Records.push_back(*R); 1939 } 1940 if (!HasSections) 1941 return parseError(-1U, "no sections"); 1942 1943 return std::move(Records); 1944 } 1945 1946 Expected<std::unique_ptr<Object>> 1947 IHexReader::create(bool /*EnsureSymtab*/) const { 1948 Expected<std::vector<IHexRecord>> Records = parse(); 1949 if (!Records) 1950 return Records.takeError(); 1951 1952 return IHexELFBuilder(*Records).build(); 1953 } 1954 1955 Expected<std::unique_ptr<Object>> ELFReader::create(bool EnsureSymtab) const { 1956 auto Obj = std::make_unique<Object>(); 1957 if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Bin)) { 1958 ELFBuilder<ELF32LE> Builder(*O, *Obj, ExtractPartition); 1959 if (Error Err = Builder.build(EnsureSymtab)) 1960 return std::move(Err); 1961 return std::move(Obj); 1962 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Bin)) { 1963 ELFBuilder<ELF64LE> Builder(*O, *Obj, ExtractPartition); 1964 if (Error Err = Builder.build(EnsureSymtab)) 1965 return std::move(Err); 1966 return std::move(Obj); 1967 } else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Bin)) { 1968 ELFBuilder<ELF32BE> Builder(*O, *Obj, ExtractPartition); 1969 if (Error Err = Builder.build(EnsureSymtab)) 1970 return std::move(Err); 1971 return std::move(Obj); 1972 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Bin)) { 1973 ELFBuilder<ELF64BE> Builder(*O, *Obj, ExtractPartition); 1974 if (Error Err = Builder.build(EnsureSymtab)) 1975 return std::move(Err); 1976 return std::move(Obj); 1977 } 1978 return createStringError(errc::invalid_argument, "invalid file type"); 1979 } 1980 1981 template <class ELFT> void ELFWriter<ELFT>::writeEhdr() { 1982 Elf_Ehdr &Ehdr = *reinterpret_cast<Elf_Ehdr *>(Buf->getBufferStart()); 1983 std::fill(Ehdr.e_ident, Ehdr.e_ident + 16, 0); 1984 Ehdr.e_ident[EI_MAG0] = 0x7f; 1985 Ehdr.e_ident[EI_MAG1] = 'E'; 1986 Ehdr.e_ident[EI_MAG2] = 'L'; 1987 Ehdr.e_ident[EI_MAG3] = 'F'; 1988 Ehdr.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32; 1989 Ehdr.e_ident[EI_DATA] = 1990 ELFT::TargetEndianness == support::big ? ELFDATA2MSB : ELFDATA2LSB; 1991 Ehdr.e_ident[EI_VERSION] = EV_CURRENT; 1992 Ehdr.e_ident[EI_OSABI] = Obj.OSABI; 1993 Ehdr.e_ident[EI_ABIVERSION] = Obj.ABIVersion; 1994 1995 Ehdr.e_type = Obj.Type; 1996 Ehdr.e_machine = Obj.Machine; 1997 Ehdr.e_version = Obj.Version; 1998 Ehdr.e_entry = Obj.Entry; 1999 // We have to use the fully-qualified name llvm::size 2000 // since some compilers complain on ambiguous resolution. 2001 Ehdr.e_phnum = llvm::size(Obj.segments()); 2002 Ehdr.e_phoff = (Ehdr.e_phnum != 0) ? Obj.ProgramHdrSegment.Offset : 0; 2003 Ehdr.e_phentsize = (Ehdr.e_phnum != 0) ? sizeof(Elf_Phdr) : 0; 2004 Ehdr.e_flags = Obj.Flags; 2005 Ehdr.e_ehsize = sizeof(Elf_Ehdr); 2006 if (WriteSectionHeaders && Obj.sections().size() != 0) { 2007 Ehdr.e_shentsize = sizeof(Elf_Shdr); 2008 Ehdr.e_shoff = Obj.SHOff; 2009 // """ 2010 // If the number of sections is greater than or equal to 2011 // SHN_LORESERVE (0xff00), this member has the value zero and the actual 2012 // number of section header table entries is contained in the sh_size field 2013 // of the section header at index 0. 2014 // """ 2015 auto Shnum = Obj.sections().size() + 1; 2016 if (Shnum >= SHN_LORESERVE) 2017 Ehdr.e_shnum = 0; 2018 else 2019 Ehdr.e_shnum = Shnum; 2020 // """ 2021 // If the section name string table section index is greater than or equal 2022 // to SHN_LORESERVE (0xff00), this member has the value SHN_XINDEX (0xffff) 2023 // and the actual index of the section name string table section is 2024 // contained in the sh_link field of the section header at index 0. 2025 // """ 2026 if (Obj.SectionNames->Index >= SHN_LORESERVE) 2027 Ehdr.e_shstrndx = SHN_XINDEX; 2028 else 2029 Ehdr.e_shstrndx = Obj.SectionNames->Index; 2030 } else { 2031 Ehdr.e_shentsize = 0; 2032 Ehdr.e_shoff = 0; 2033 Ehdr.e_shnum = 0; 2034 Ehdr.e_shstrndx = 0; 2035 } 2036 } 2037 2038 template <class ELFT> void ELFWriter<ELFT>::writePhdrs() { 2039 for (auto &Seg : Obj.segments()) 2040 writePhdr(Seg); 2041 } 2042 2043 template <class ELFT> void ELFWriter<ELFT>::writeShdrs() { 2044 // This reference serves to write the dummy section header at the begining 2045 // of the file. It is not used for anything else 2046 Elf_Shdr &Shdr = 2047 *reinterpret_cast<Elf_Shdr *>(Buf->getBufferStart() + Obj.SHOff); 2048 Shdr.sh_name = 0; 2049 Shdr.sh_type = SHT_NULL; 2050 Shdr.sh_flags = 0; 2051 Shdr.sh_addr = 0; 2052 Shdr.sh_offset = 0; 2053 // See writeEhdr for why we do this. 2054 uint64_t Shnum = Obj.sections().size() + 1; 2055 if (Shnum >= SHN_LORESERVE) 2056 Shdr.sh_size = Shnum; 2057 else 2058 Shdr.sh_size = 0; 2059 // See writeEhdr for why we do this. 2060 if (Obj.SectionNames != nullptr && Obj.SectionNames->Index >= SHN_LORESERVE) 2061 Shdr.sh_link = Obj.SectionNames->Index; 2062 else 2063 Shdr.sh_link = 0; 2064 Shdr.sh_info = 0; 2065 Shdr.sh_addralign = 0; 2066 Shdr.sh_entsize = 0; 2067 2068 for (SectionBase &Sec : Obj.sections()) 2069 writeShdr(Sec); 2070 } 2071 2072 template <class ELFT> Error ELFWriter<ELFT>::writeSectionData() { 2073 for (SectionBase &Sec : Obj.sections()) 2074 // Segments are responsible for writing their contents, so only write the 2075 // section data if the section is not in a segment. Note that this renders 2076 // sections in segments effectively immutable. 2077 if (Sec.ParentSegment == nullptr) 2078 if (Error Err = Sec.accept(*SecWriter)) 2079 return Err; 2080 2081 return Error::success(); 2082 } 2083 2084 template <class ELFT> void ELFWriter<ELFT>::writeSegmentData() { 2085 for (Segment &Seg : Obj.segments()) { 2086 size_t Size = std::min<size_t>(Seg.FileSize, Seg.getContents().size()); 2087 std::memcpy(Buf->getBufferStart() + Seg.Offset, Seg.getContents().data(), 2088 Size); 2089 } 2090 2091 for (auto it : Obj.getUpdatedSections()) { 2092 SectionBase *Sec = it.first; 2093 ArrayRef<uint8_t> Data = it.second; 2094 2095 auto *Parent = Sec->ParentSegment; 2096 assert(Parent && "This section should've been part of a segment."); 2097 uint64_t Offset = 2098 Sec->OriginalOffset - Parent->OriginalOffset + Parent->Offset; 2099 llvm::copy(Data, Buf->getBufferStart() + Offset); 2100 } 2101 2102 // Iterate over removed sections and overwrite their old data with zeroes. 2103 for (auto &Sec : Obj.removedSections()) { 2104 Segment *Parent = Sec.ParentSegment; 2105 if (Parent == nullptr || Sec.Type == SHT_NOBITS || Sec.Size == 0) 2106 continue; 2107 uint64_t Offset = 2108 Sec.OriginalOffset - Parent->OriginalOffset + Parent->Offset; 2109 std::memset(Buf->getBufferStart() + Offset, 0, Sec.Size); 2110 } 2111 } 2112 2113 template <class ELFT> 2114 ELFWriter<ELFT>::ELFWriter(Object &Obj, raw_ostream &Buf, bool WSH, 2115 bool OnlyKeepDebug) 2116 : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs), 2117 OnlyKeepDebug(OnlyKeepDebug) {} 2118 2119 Error Object::updateSection(StringRef Name, ArrayRef<uint8_t> Data) { 2120 auto It = llvm::find_if(Sections, 2121 [&](const SecPtr &Sec) { return Sec->Name == Name; }); 2122 if (It == Sections.end()) 2123 return createStringError(errc::invalid_argument, "section '%s' not found", 2124 Name.str().c_str()); 2125 2126 auto *OldSec = It->get(); 2127 if (!OldSec->hasContents()) 2128 return createStringError( 2129 errc::invalid_argument, 2130 "section '%s' cannot be updated because it does not have contents", 2131 Name.str().c_str()); 2132 2133 if (Data.size() > OldSec->Size && OldSec->ParentSegment) 2134 return createStringError(errc::invalid_argument, 2135 "cannot fit data of size %zu into section '%s' " 2136 "with size %zu that is part of a segment", 2137 Data.size(), Name.str().c_str(), OldSec->Size); 2138 2139 if (!OldSec->ParentSegment) { 2140 *It = std::make_unique<OwnedDataSection>(*OldSec, Data); 2141 } else { 2142 // The segment writer will be in charge of updating these contents. 2143 OldSec->Size = Data.size(); 2144 UpdatedSections[OldSec] = Data; 2145 } 2146 2147 return Error::success(); 2148 } 2149 2150 Error Object::removeSections( 2151 bool AllowBrokenLinks, std::function<bool(const SectionBase &)> ToRemove) { 2152 2153 auto Iter = std::stable_partition( 2154 std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) { 2155 if (ToRemove(*Sec)) 2156 return false; 2157 if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) { 2158 if (auto ToRelSec = RelSec->getSection()) 2159 return !ToRemove(*ToRelSec); 2160 } 2161 return true; 2162 }); 2163 if (SymbolTable != nullptr && ToRemove(*SymbolTable)) 2164 SymbolTable = nullptr; 2165 if (SectionNames != nullptr && ToRemove(*SectionNames)) 2166 SectionNames = nullptr; 2167 if (SectionIndexTable != nullptr && ToRemove(*SectionIndexTable)) 2168 SectionIndexTable = nullptr; 2169 // Now make sure there are no remaining references to the sections that will 2170 // be removed. Sometimes it is impossible to remove a reference so we emit 2171 // an error here instead. 2172 std::unordered_set<const SectionBase *> RemoveSections; 2173 RemoveSections.reserve(std::distance(Iter, std::end(Sections))); 2174 for (auto &RemoveSec : make_range(Iter, std::end(Sections))) { 2175 for (auto &Segment : Segments) 2176 Segment->removeSection(RemoveSec.get()); 2177 RemoveSec->onRemove(); 2178 RemoveSections.insert(RemoveSec.get()); 2179 } 2180 2181 // For each section that remains alive, we want to remove the dead references. 2182 // This either might update the content of the section (e.g. remove symbols 2183 // from symbol table that belongs to removed section) or trigger an error if 2184 // a live section critically depends on a section being removed somehow 2185 // (e.g. the removed section is referenced by a relocation). 2186 for (auto &KeepSec : make_range(std::begin(Sections), Iter)) { 2187 if (Error E = KeepSec->removeSectionReferences( 2188 AllowBrokenLinks, [&RemoveSections](const SectionBase *Sec) { 2189 return RemoveSections.find(Sec) != RemoveSections.end(); 2190 })) 2191 return E; 2192 } 2193 2194 // Transfer removed sections into the Object RemovedSections container for use 2195 // later. 2196 std::move(Iter, Sections.end(), std::back_inserter(RemovedSections)); 2197 // Now finally get rid of them all together. 2198 Sections.erase(Iter, std::end(Sections)); 2199 return Error::success(); 2200 } 2201 2202 Error Object::replaceSections( 2203 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 2204 auto SectionIndexLess = [](const SecPtr &Lhs, const SecPtr &Rhs) { 2205 return Lhs->Index < Rhs->Index; 2206 }; 2207 assert(llvm::is_sorted(Sections, SectionIndexLess) && 2208 "Sections are expected to be sorted by Index"); 2209 // Set indices of new sections so that they can be later sorted into positions 2210 // of removed ones. 2211 for (auto &I : FromTo) 2212 I.second->Index = I.first->Index; 2213 2214 // Notify all sections about the replacement. 2215 for (auto &Sec : Sections) 2216 Sec->replaceSectionReferences(FromTo); 2217 2218 if (Error E = removeSections( 2219 /*AllowBrokenLinks=*/false, 2220 [=](const SectionBase &Sec) { return FromTo.count(&Sec) > 0; })) 2221 return E; 2222 llvm::sort(Sections, SectionIndexLess); 2223 return Error::success(); 2224 } 2225 2226 Error Object::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) { 2227 if (SymbolTable) 2228 for (const SecPtr &Sec : Sections) 2229 if (Error E = Sec->removeSymbols(ToRemove)) 2230 return E; 2231 return Error::success(); 2232 } 2233 2234 Error Object::addNewSymbolTable() { 2235 assert(!SymbolTable && "Object must not has a SymbolTable."); 2236 2237 // Reuse an existing SHT_STRTAB section if it exists. 2238 StringTableSection *StrTab = nullptr; 2239 for (SectionBase &Sec : sections()) { 2240 if (Sec.Type == ELF::SHT_STRTAB && !(Sec.Flags & SHF_ALLOC)) { 2241 StrTab = static_cast<StringTableSection *>(&Sec); 2242 2243 // Prefer a string table that is not the section header string table, if 2244 // such a table exists. 2245 if (SectionNames != &Sec) 2246 break; 2247 } 2248 } 2249 if (!StrTab) 2250 StrTab = &addSection<StringTableSection>(); 2251 2252 SymbolTableSection &SymTab = addSection<SymbolTableSection>(); 2253 SymTab.Name = ".symtab"; 2254 SymTab.Link = StrTab->Index; 2255 if (Error Err = SymTab.initialize(sections())) 2256 return Err; 2257 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0); 2258 2259 SymbolTable = &SymTab; 2260 2261 return Error::success(); 2262 } 2263 2264 // Orders segments such that if x = y->ParentSegment then y comes before x. 2265 static void orderSegments(std::vector<Segment *> &Segments) { 2266 llvm::stable_sort(Segments, compareSegmentsByOffset); 2267 } 2268 2269 // This function finds a consistent layout for a list of segments starting from 2270 // an Offset. It assumes that Segments have been sorted by orderSegments and 2271 // returns an Offset one past the end of the last segment. 2272 static uint64_t layoutSegments(std::vector<Segment *> &Segments, 2273 uint64_t Offset) { 2274 assert(llvm::is_sorted(Segments, compareSegmentsByOffset)); 2275 // The only way a segment should move is if a section was between two 2276 // segments and that section was removed. If that section isn't in a segment 2277 // then it's acceptable, but not ideal, to simply move it to after the 2278 // segments. So we can simply layout segments one after the other accounting 2279 // for alignment. 2280 for (Segment *Seg : Segments) { 2281 // We assume that segments have been ordered by OriginalOffset and Index 2282 // such that a parent segment will always come before a child segment in 2283 // OrderedSegments. This means that the Offset of the ParentSegment should 2284 // already be set and we can set our offset relative to it. 2285 if (Seg->ParentSegment != nullptr) { 2286 Segment *Parent = Seg->ParentSegment; 2287 Seg->Offset = 2288 Parent->Offset + Seg->OriginalOffset - Parent->OriginalOffset; 2289 } else { 2290 Seg->Offset = 2291 alignTo(Offset, std::max<uint64_t>(Seg->Align, 1), Seg->VAddr); 2292 } 2293 Offset = std::max(Offset, Seg->Offset + Seg->FileSize); 2294 } 2295 return Offset; 2296 } 2297 2298 // This function finds a consistent layout for a list of sections. It assumes 2299 // that the ->ParentSegment of each section has already been laid out. The 2300 // supplied starting Offset is used for the starting offset of any section that 2301 // does not have a ParentSegment. It returns either the offset given if all 2302 // sections had a ParentSegment or an offset one past the last section if there 2303 // was a section that didn't have a ParentSegment. 2304 template <class Range> 2305 static uint64_t layoutSections(Range Sections, uint64_t Offset) { 2306 // Now the offset of every segment has been set we can assign the offsets 2307 // of each section. For sections that are covered by a segment we should use 2308 // the segment's original offset and the section's original offset to compute 2309 // the offset from the start of the segment. Using the offset from the start 2310 // of the segment we can assign a new offset to the section. For sections not 2311 // covered by segments we can just bump Offset to the next valid location. 2312 // While it is not necessary, layout the sections in the order based on their 2313 // original offsets to resemble the input file as close as possible. 2314 std::vector<SectionBase *> OutOfSegmentSections; 2315 uint32_t Index = 1; 2316 for (auto &Sec : Sections) { 2317 Sec.Index = Index++; 2318 if (Sec.ParentSegment != nullptr) { 2319 auto Segment = *Sec.ParentSegment; 2320 Sec.Offset = 2321 Segment.Offset + (Sec.OriginalOffset - Segment.OriginalOffset); 2322 } else 2323 OutOfSegmentSections.push_back(&Sec); 2324 } 2325 2326 llvm::stable_sort(OutOfSegmentSections, 2327 [](const SectionBase *Lhs, const SectionBase *Rhs) { 2328 return Lhs->OriginalOffset < Rhs->OriginalOffset; 2329 }); 2330 for (auto *Sec : OutOfSegmentSections) { 2331 Offset = alignTo(Offset, Sec->Align == 0 ? 1 : Sec->Align); 2332 Sec->Offset = Offset; 2333 if (Sec->Type != SHT_NOBITS) 2334 Offset += Sec->Size; 2335 } 2336 return Offset; 2337 } 2338 2339 // Rewrite sh_offset after some sections are changed to SHT_NOBITS and thus 2340 // occupy no space in the file. 2341 static uint64_t layoutSectionsForOnlyKeepDebug(Object &Obj, uint64_t Off) { 2342 // The layout algorithm requires the sections to be handled in the order of 2343 // their offsets in the input file, at least inside segments. 2344 std::vector<SectionBase *> Sections; 2345 Sections.reserve(Obj.sections().size()); 2346 uint32_t Index = 1; 2347 for (auto &Sec : Obj.sections()) { 2348 Sec.Index = Index++; 2349 Sections.push_back(&Sec); 2350 } 2351 llvm::stable_sort(Sections, 2352 [](const SectionBase *Lhs, const SectionBase *Rhs) { 2353 return Lhs->OriginalOffset < Rhs->OriginalOffset; 2354 }); 2355 2356 for (auto *Sec : Sections) { 2357 auto *FirstSec = Sec->ParentSegment && Sec->ParentSegment->Type == PT_LOAD 2358 ? Sec->ParentSegment->firstSection() 2359 : nullptr; 2360 2361 // The first section in a PT_LOAD has to have congruent offset and address 2362 // modulo the alignment, which usually equals the maximum page size. 2363 if (FirstSec && FirstSec == Sec) 2364 Off = alignTo(Off, Sec->ParentSegment->Align, Sec->Addr); 2365 2366 // sh_offset is not significant for SHT_NOBITS sections, but the congruence 2367 // rule must be followed if it is the first section in a PT_LOAD. Do not 2368 // advance Off. 2369 if (Sec->Type == SHT_NOBITS) { 2370 Sec->Offset = Off; 2371 continue; 2372 } 2373 2374 if (!FirstSec) { 2375 // FirstSec being nullptr generally means that Sec does not have the 2376 // SHF_ALLOC flag. 2377 Off = Sec->Align ? alignTo(Off, Sec->Align) : Off; 2378 } else if (FirstSec != Sec) { 2379 // The offset is relative to the first section in the PT_LOAD segment. Use 2380 // sh_offset for non-SHF_ALLOC sections. 2381 Off = Sec->OriginalOffset - FirstSec->OriginalOffset + FirstSec->Offset; 2382 } 2383 Sec->Offset = Off; 2384 Off += Sec->Size; 2385 } 2386 return Off; 2387 } 2388 2389 // Rewrite p_offset and p_filesz of non-PT_PHDR segments after sh_offset values 2390 // have been updated. 2391 static uint64_t layoutSegmentsForOnlyKeepDebug(std::vector<Segment *> &Segments, 2392 uint64_t HdrEnd) { 2393 uint64_t MaxOffset = 0; 2394 for (Segment *Seg : Segments) { 2395 if (Seg->Type == PT_PHDR) 2396 continue; 2397 2398 // The segment offset is generally the offset of the first section. 2399 // 2400 // For a segment containing no section (see sectionWithinSegment), if it has 2401 // a parent segment, copy the parent segment's offset field. This works for 2402 // empty PT_TLS. If no parent segment, use 0: the segment is not useful for 2403 // debugging anyway. 2404 const SectionBase *FirstSec = Seg->firstSection(); 2405 uint64_t Offset = 2406 FirstSec ? FirstSec->Offset 2407 : (Seg->ParentSegment ? Seg->ParentSegment->Offset : 0); 2408 uint64_t FileSize = 0; 2409 for (const SectionBase *Sec : Seg->Sections) { 2410 uint64_t Size = Sec->Type == SHT_NOBITS ? 0 : Sec->Size; 2411 if (Sec->Offset + Size > Offset) 2412 FileSize = std::max(FileSize, Sec->Offset + Size - Offset); 2413 } 2414 2415 // If the segment includes EHDR and program headers, don't make it smaller 2416 // than the headers. 2417 if (Seg->Offset < HdrEnd && HdrEnd <= Seg->Offset + Seg->FileSize) { 2418 FileSize += Offset - Seg->Offset; 2419 Offset = Seg->Offset; 2420 FileSize = std::max(FileSize, HdrEnd - Offset); 2421 } 2422 2423 Seg->Offset = Offset; 2424 Seg->FileSize = FileSize; 2425 MaxOffset = std::max(MaxOffset, Offset + FileSize); 2426 } 2427 return MaxOffset; 2428 } 2429 2430 template <class ELFT> void ELFWriter<ELFT>::initEhdrSegment() { 2431 Segment &ElfHdr = Obj.ElfHdrSegment; 2432 ElfHdr.Type = PT_PHDR; 2433 ElfHdr.Flags = 0; 2434 ElfHdr.VAddr = 0; 2435 ElfHdr.PAddr = 0; 2436 ElfHdr.FileSize = ElfHdr.MemSize = sizeof(Elf_Ehdr); 2437 ElfHdr.Align = 0; 2438 } 2439 2440 template <class ELFT> void ELFWriter<ELFT>::assignOffsets() { 2441 // We need a temporary list of segments that has a special order to it 2442 // so that we know that anytime ->ParentSegment is set that segment has 2443 // already had its offset properly set. 2444 std::vector<Segment *> OrderedSegments; 2445 for (Segment &Segment : Obj.segments()) 2446 OrderedSegments.push_back(&Segment); 2447 OrderedSegments.push_back(&Obj.ElfHdrSegment); 2448 OrderedSegments.push_back(&Obj.ProgramHdrSegment); 2449 orderSegments(OrderedSegments); 2450 2451 uint64_t Offset; 2452 if (OnlyKeepDebug) { 2453 // For --only-keep-debug, the sections that did not preserve contents were 2454 // changed to SHT_NOBITS. We now rewrite sh_offset fields of sections, and 2455 // then rewrite p_offset/p_filesz of program headers. 2456 uint64_t HdrEnd = 2457 sizeof(Elf_Ehdr) + llvm::size(Obj.segments()) * sizeof(Elf_Phdr); 2458 Offset = layoutSectionsForOnlyKeepDebug(Obj, HdrEnd); 2459 Offset = std::max(Offset, 2460 layoutSegmentsForOnlyKeepDebug(OrderedSegments, HdrEnd)); 2461 } else { 2462 // Offset is used as the start offset of the first segment to be laid out. 2463 // Since the ELF Header (ElfHdrSegment) must be at the start of the file, 2464 // we start at offset 0. 2465 Offset = layoutSegments(OrderedSegments, 0); 2466 Offset = layoutSections(Obj.sections(), Offset); 2467 } 2468 // If we need to write the section header table out then we need to align the 2469 // Offset so that SHOffset is valid. 2470 if (WriteSectionHeaders) 2471 Offset = alignTo(Offset, sizeof(Elf_Addr)); 2472 Obj.SHOff = Offset; 2473 } 2474 2475 template <class ELFT> size_t ELFWriter<ELFT>::totalSize() const { 2476 // We already have the section header offset so we can calculate the total 2477 // size by just adding up the size of each section header. 2478 if (!WriteSectionHeaders) 2479 return Obj.SHOff; 2480 size_t ShdrCount = Obj.sections().size() + 1; // Includes null shdr. 2481 return Obj.SHOff + ShdrCount * sizeof(Elf_Shdr); 2482 } 2483 2484 template <class ELFT> Error ELFWriter<ELFT>::write() { 2485 // Segment data must be written first, so that the ELF header and program 2486 // header tables can overwrite it, if covered by a segment. 2487 writeSegmentData(); 2488 writeEhdr(); 2489 writePhdrs(); 2490 if (Error E = writeSectionData()) 2491 return E; 2492 if (WriteSectionHeaders) 2493 writeShdrs(); 2494 2495 // TODO: Implement direct writing to the output stream (without intermediate 2496 // memory buffer Buf). 2497 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2498 return Error::success(); 2499 } 2500 2501 static Error removeUnneededSections(Object &Obj) { 2502 // We can remove an empty symbol table from non-relocatable objects. 2503 // Relocatable objects typically have relocation sections whose 2504 // sh_link field points to .symtab, so we can't remove .symtab 2505 // even if it is empty. 2506 if (Obj.isRelocatable() || Obj.SymbolTable == nullptr || 2507 !Obj.SymbolTable->empty()) 2508 return Error::success(); 2509 2510 // .strtab can be used for section names. In such a case we shouldn't 2511 // remove it. 2512 auto *StrTab = Obj.SymbolTable->getStrTab() == Obj.SectionNames 2513 ? nullptr 2514 : Obj.SymbolTable->getStrTab(); 2515 return Obj.removeSections(false, [&](const SectionBase &Sec) { 2516 return &Sec == Obj.SymbolTable || &Sec == StrTab; 2517 }); 2518 } 2519 2520 template <class ELFT> Error ELFWriter<ELFT>::finalize() { 2521 // It could happen that SectionNames has been removed and yet the user wants 2522 // a section header table output. We need to throw an error if a user tries 2523 // to do that. 2524 if (Obj.SectionNames == nullptr && WriteSectionHeaders) 2525 return createStringError(llvm::errc::invalid_argument, 2526 "cannot write section header table because " 2527 "section header string table was removed"); 2528 2529 if (Error E = removeUnneededSections(Obj)) 2530 return E; 2531 2532 // We need to assign indexes before we perform layout because we need to know 2533 // if we need large indexes or not. We can assign indexes first and check as 2534 // we go to see if we will actully need large indexes. 2535 bool NeedsLargeIndexes = false; 2536 if (Obj.sections().size() >= SHN_LORESERVE) { 2537 SectionTableRef Sections = Obj.sections(); 2538 // Sections doesn't include the null section header, so account for this 2539 // when skipping the first N sections. 2540 NeedsLargeIndexes = 2541 any_of(drop_begin(Sections, SHN_LORESERVE - 1), 2542 [](const SectionBase &Sec) { return Sec.HasSymbol; }); 2543 // TODO: handle case where only one section needs the large index table but 2544 // only needs it because the large index table hasn't been removed yet. 2545 } 2546 2547 if (NeedsLargeIndexes) { 2548 // This means we definitely need to have a section index table but if we 2549 // already have one then we should use it instead of making a new one. 2550 if (Obj.SymbolTable != nullptr && Obj.SectionIndexTable == nullptr) { 2551 // Addition of a section to the end does not invalidate the indexes of 2552 // other sections and assigns the correct index to the new section. 2553 auto &Shndx = Obj.addSection<SectionIndexSection>(); 2554 Obj.SymbolTable->setShndxTable(&Shndx); 2555 Shndx.setSymTab(Obj.SymbolTable); 2556 } 2557 } else { 2558 // Since we don't need SectionIndexTable we should remove it and all 2559 // references to it. 2560 if (Obj.SectionIndexTable != nullptr) { 2561 // We do not support sections referring to the section index table. 2562 if (Error E = Obj.removeSections(false /*AllowBrokenLinks*/, 2563 [this](const SectionBase &Sec) { 2564 return &Sec == Obj.SectionIndexTable; 2565 })) 2566 return E; 2567 } 2568 } 2569 2570 // Make sure we add the names of all the sections. Importantly this must be 2571 // done after we decide to add or remove SectionIndexes. 2572 if (Obj.SectionNames != nullptr) 2573 for (const SectionBase &Sec : Obj.sections()) 2574 Obj.SectionNames->addString(Sec.Name); 2575 2576 initEhdrSegment(); 2577 2578 // Before we can prepare for layout the indexes need to be finalized. 2579 // Also, the output arch may not be the same as the input arch, so fix up 2580 // size-related fields before doing layout calculations. 2581 uint64_t Index = 0; 2582 auto SecSizer = std::make_unique<ELFSectionSizer<ELFT>>(); 2583 for (SectionBase &Sec : Obj.sections()) { 2584 Sec.Index = Index++; 2585 if (Error Err = Sec.accept(*SecSizer)) 2586 return Err; 2587 } 2588 2589 // The symbol table does not update all other sections on update. For 2590 // instance, symbol names are not added as new symbols are added. This means 2591 // that some sections, like .strtab, don't yet have their final size. 2592 if (Obj.SymbolTable != nullptr) 2593 Obj.SymbolTable->prepareForLayout(); 2594 2595 // Now that all strings are added we want to finalize string table builders, 2596 // because that affects section sizes which in turn affects section offsets. 2597 for (SectionBase &Sec : Obj.sections()) 2598 if (auto StrTab = dyn_cast<StringTableSection>(&Sec)) 2599 StrTab->prepareForLayout(); 2600 2601 assignOffsets(); 2602 2603 // layoutSections could have modified section indexes, so we need 2604 // to fill the index table after assignOffsets. 2605 if (Obj.SymbolTable != nullptr) 2606 Obj.SymbolTable->fillShndxTable(); 2607 2608 // Finally now that all offsets and indexes have been set we can finalize any 2609 // remaining issues. 2610 uint64_t Offset = Obj.SHOff + sizeof(Elf_Shdr); 2611 for (SectionBase &Sec : Obj.sections()) { 2612 Sec.HeaderOffset = Offset; 2613 Offset += sizeof(Elf_Shdr); 2614 if (WriteSectionHeaders) 2615 Sec.NameIndex = Obj.SectionNames->findIndex(Sec.Name); 2616 Sec.finalize(); 2617 } 2618 2619 size_t TotalSize = totalSize(); 2620 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2621 if (!Buf) 2622 return createStringError(errc::not_enough_memory, 2623 "failed to allocate memory buffer of " + 2624 Twine::utohexstr(TotalSize) + " bytes"); 2625 2626 SecWriter = std::make_unique<ELFSectionWriter<ELFT>>(*Buf); 2627 return Error::success(); 2628 } 2629 2630 Error BinaryWriter::write() { 2631 for (const SectionBase &Sec : Obj.allocSections()) 2632 if (Error Err = Sec.accept(*SecWriter)) 2633 return Err; 2634 2635 // TODO: Implement direct writing to the output stream (without intermediate 2636 // memory buffer Buf). 2637 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2638 return Error::success(); 2639 } 2640 2641 Error BinaryWriter::finalize() { 2642 // Compute the section LMA based on its sh_offset and the containing segment's 2643 // p_offset and p_paddr. Also compute the minimum LMA of all non-empty 2644 // sections as MinAddr. In the output, the contents between address 0 and 2645 // MinAddr will be skipped. 2646 uint64_t MinAddr = UINT64_MAX; 2647 for (SectionBase &Sec : Obj.allocSections()) { 2648 // If Sec's type is changed from SHT_NOBITS due to --set-section-flags, 2649 // Offset may not be aligned. Align it to max(Align, 1). 2650 if (Sec.ParentSegment != nullptr) 2651 Sec.Addr = alignTo(Sec.Offset - Sec.ParentSegment->Offset + 2652 Sec.ParentSegment->PAddr, 2653 std::max(Sec.Align, uint64_t(1))); 2654 if (Sec.Type != SHT_NOBITS && Sec.Size > 0) 2655 MinAddr = std::min(MinAddr, Sec.Addr); 2656 } 2657 2658 // Now that every section has been laid out we just need to compute the total 2659 // file size. This might not be the same as the offset returned by 2660 // layoutSections, because we want to truncate the last segment to the end of 2661 // its last non-empty section, to match GNU objcopy's behaviour. 2662 TotalSize = 0; 2663 for (SectionBase &Sec : Obj.allocSections()) 2664 if (Sec.Type != SHT_NOBITS && Sec.Size > 0) { 2665 Sec.Offset = Sec.Addr - MinAddr; 2666 TotalSize = std::max(TotalSize, Sec.Offset + Sec.Size); 2667 } 2668 2669 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2670 if (!Buf) 2671 return createStringError(errc::not_enough_memory, 2672 "failed to allocate memory buffer of " + 2673 Twine::utohexstr(TotalSize) + " bytes"); 2674 SecWriter = std::make_unique<BinarySectionWriter>(*Buf); 2675 return Error::success(); 2676 } 2677 2678 bool IHexWriter::SectionCompare::operator()(const SectionBase *Lhs, 2679 const SectionBase *Rhs) const { 2680 return (sectionPhysicalAddr(Lhs) & 0xFFFFFFFFU) < 2681 (sectionPhysicalAddr(Rhs) & 0xFFFFFFFFU); 2682 } 2683 2684 uint64_t IHexWriter::writeEntryPointRecord(uint8_t *Buf) { 2685 IHexLineData HexData; 2686 uint8_t Data[4] = {}; 2687 // We don't write entry point record if entry is zero. 2688 if (Obj.Entry == 0) 2689 return 0; 2690 2691 if (Obj.Entry <= 0xFFFFFU) { 2692 Data[0] = ((Obj.Entry & 0xF0000U) >> 12) & 0xFF; 2693 support::endian::write(&Data[2], static_cast<uint16_t>(Obj.Entry), 2694 support::big); 2695 HexData = IHexRecord::getLine(IHexRecord::StartAddr80x86, 0, Data); 2696 } else { 2697 support::endian::write(Data, static_cast<uint32_t>(Obj.Entry), 2698 support::big); 2699 HexData = IHexRecord::getLine(IHexRecord::StartAddr, 0, Data); 2700 } 2701 memcpy(Buf, HexData.data(), HexData.size()); 2702 return HexData.size(); 2703 } 2704 2705 uint64_t IHexWriter::writeEndOfFileRecord(uint8_t *Buf) { 2706 IHexLineData HexData = IHexRecord::getLine(IHexRecord::EndOfFile, 0, {}); 2707 memcpy(Buf, HexData.data(), HexData.size()); 2708 return HexData.size(); 2709 } 2710 2711 Error IHexWriter::write() { 2712 IHexSectionWriter Writer(*Buf); 2713 // Write sections. 2714 for (const SectionBase *Sec : Sections) 2715 if (Error Err = Sec->accept(Writer)) 2716 return Err; 2717 2718 uint64_t Offset = Writer.getBufferOffset(); 2719 // Write entry point address. 2720 Offset += writeEntryPointRecord( 2721 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset); 2722 // Write EOF. 2723 Offset += writeEndOfFileRecord( 2724 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset); 2725 assert(Offset == TotalSize); 2726 2727 // TODO: Implement direct writing to the output stream (without intermediate 2728 // memory buffer Buf). 2729 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2730 return Error::success(); 2731 } 2732 2733 Error IHexWriter::checkSection(const SectionBase &Sec) { 2734 uint64_t Addr = sectionPhysicalAddr(&Sec); 2735 if (addressOverflows32bit(Addr) || addressOverflows32bit(Addr + Sec.Size - 1)) 2736 return createStringError( 2737 errc::invalid_argument, 2738 "Section '%s' address range [0x%llx, 0x%llx] is not 32 bit", 2739 Sec.Name.c_str(), Addr, Addr + Sec.Size - 1); 2740 return Error::success(); 2741 } 2742 2743 Error IHexWriter::finalize() { 2744 // We can't write 64-bit addresses. 2745 if (addressOverflows32bit(Obj.Entry)) 2746 return createStringError(errc::invalid_argument, 2747 "Entry point address 0x%llx overflows 32 bits", 2748 Obj.Entry); 2749 2750 for (const SectionBase &Sec : Obj.sections()) 2751 if ((Sec.Flags & ELF::SHF_ALLOC) && Sec.Type != ELF::SHT_NOBITS && 2752 Sec.Size > 0) { 2753 if (Error E = checkSection(Sec)) 2754 return E; 2755 Sections.insert(&Sec); 2756 } 2757 2758 std::unique_ptr<WritableMemoryBuffer> EmptyBuffer = 2759 WritableMemoryBuffer::getNewMemBuffer(0); 2760 if (!EmptyBuffer) 2761 return createStringError(errc::not_enough_memory, 2762 "failed to allocate memory buffer of 0 bytes"); 2763 2764 IHexSectionWriterBase LengthCalc(*EmptyBuffer); 2765 for (const SectionBase *Sec : Sections) 2766 if (Error Err = Sec->accept(LengthCalc)) 2767 return Err; 2768 2769 // We need space to write section records + StartAddress record 2770 // (if start adress is not zero) + EndOfFile record. 2771 TotalSize = LengthCalc.getBufferOffset() + 2772 (Obj.Entry ? IHexRecord::getLineLength(4) : 0) + 2773 IHexRecord::getLineLength(0); 2774 2775 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2776 if (!Buf) 2777 return createStringError(errc::not_enough_memory, 2778 "failed to allocate memory buffer of " + 2779 Twine::utohexstr(TotalSize) + " bytes"); 2780 2781 return Error::success(); 2782 } 2783 2784 namespace llvm { 2785 namespace objcopy { 2786 namespace elf { 2787 2788 template class ELFBuilder<ELF64LE>; 2789 template class ELFBuilder<ELF64BE>; 2790 template class ELFBuilder<ELF32LE>; 2791 template class ELFBuilder<ELF32BE>; 2792 2793 template class ELFWriter<ELF64LE>; 2794 template class ELFWriter<ELF64BE>; 2795 template class ELFWriter<ELF32LE>; 2796 template class ELFWriter<ELF32BE>; 2797 2798 } // end namespace elf 2799 } // end namespace objcopy 2800 } // end namespace llvm 2801