1 //===- ELFObject.cpp ------------------------------------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #include "ELFObject.h" 10 #include "llvm/ADT/ArrayRef.h" 11 #include "llvm/ADT/STLExtras.h" 12 #include "llvm/ADT/StringRef.h" 13 #include "llvm/ADT/Twine.h" 14 #include "llvm/ADT/iterator_range.h" 15 #include "llvm/BinaryFormat/ELF.h" 16 #include "llvm/MC/MCTargetOptions.h" 17 #include "llvm/Object/ELF.h" 18 #include "llvm/Object/ELFObjectFile.h" 19 #include "llvm/Support/Compression.h" 20 #include "llvm/Support/Endian.h" 21 #include "llvm/Support/ErrorHandling.h" 22 #include "llvm/Support/FileOutputBuffer.h" 23 #include "llvm/Support/Path.h" 24 #include <algorithm> 25 #include <cstddef> 26 #include <cstdint> 27 #include <iterator> 28 #include <unordered_set> 29 #include <utility> 30 #include <vector> 31 32 using namespace llvm; 33 using namespace llvm::ELF; 34 using namespace llvm::objcopy::elf; 35 using namespace llvm::object; 36 37 template <class ELFT> void ELFWriter<ELFT>::writePhdr(const Segment &Seg) { 38 uint8_t *B = reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + 39 Obj.ProgramHdrSegment.Offset + Seg.Index * sizeof(Elf_Phdr); 40 Elf_Phdr &Phdr = *reinterpret_cast<Elf_Phdr *>(B); 41 Phdr.p_type = Seg.Type; 42 Phdr.p_flags = Seg.Flags; 43 Phdr.p_offset = Seg.Offset; 44 Phdr.p_vaddr = Seg.VAddr; 45 Phdr.p_paddr = Seg.PAddr; 46 Phdr.p_filesz = Seg.FileSize; 47 Phdr.p_memsz = Seg.MemSize; 48 Phdr.p_align = Seg.Align; 49 } 50 51 Error SectionBase::removeSectionReferences( 52 bool, function_ref<bool(const SectionBase *)>) { 53 return Error::success(); 54 } 55 56 Error SectionBase::removeSymbols(function_ref<bool(const Symbol &)>) { 57 return Error::success(); 58 } 59 60 Error SectionBase::initialize(SectionTableRef) { return Error::success(); } 61 void SectionBase::finalize() {} 62 void SectionBase::markSymbols() {} 63 void SectionBase::replaceSectionReferences( 64 const DenseMap<SectionBase *, SectionBase *> &) {} 65 void SectionBase::onRemove() {} 66 67 template <class ELFT> void ELFWriter<ELFT>::writeShdr(const SectionBase &Sec) { 68 uint8_t *B = 69 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Sec.HeaderOffset; 70 Elf_Shdr &Shdr = *reinterpret_cast<Elf_Shdr *>(B); 71 Shdr.sh_name = Sec.NameIndex; 72 Shdr.sh_type = Sec.Type; 73 Shdr.sh_flags = Sec.Flags; 74 Shdr.sh_addr = Sec.Addr; 75 Shdr.sh_offset = Sec.Offset; 76 Shdr.sh_size = Sec.Size; 77 Shdr.sh_link = Sec.Link; 78 Shdr.sh_info = Sec.Info; 79 Shdr.sh_addralign = Sec.Align; 80 Shdr.sh_entsize = Sec.EntrySize; 81 } 82 83 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(Section &) { 84 return Error::success(); 85 } 86 87 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(OwnedDataSection &) { 88 return Error::success(); 89 } 90 91 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(StringTableSection &) { 92 return Error::success(); 93 } 94 95 template <class ELFT> 96 Error ELFSectionSizer<ELFT>::visit(DynamicRelocationSection &) { 97 return Error::success(); 98 } 99 100 template <class ELFT> 101 Error ELFSectionSizer<ELFT>::visit(SymbolTableSection &Sec) { 102 Sec.EntrySize = sizeof(Elf_Sym); 103 Sec.Size = Sec.Symbols.size() * Sec.EntrySize; 104 // Align to the largest field in Elf_Sym. 105 Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word); 106 return Error::success(); 107 } 108 109 template <class ELFT> 110 Error ELFSectionSizer<ELFT>::visit(RelocationSection &Sec) { 111 Sec.EntrySize = Sec.Type == SHT_REL ? sizeof(Elf_Rel) : sizeof(Elf_Rela); 112 Sec.Size = Sec.Relocations.size() * Sec.EntrySize; 113 // Align to the largest field in Elf_Rel(a). 114 Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word); 115 return Error::success(); 116 } 117 118 template <class ELFT> 119 Error ELFSectionSizer<ELFT>::visit(GnuDebugLinkSection &) { 120 return Error::success(); 121 } 122 123 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(GroupSection &Sec) { 124 Sec.Size = sizeof(Elf_Word) + Sec.GroupMembers.size() * sizeof(Elf_Word); 125 return Error::success(); 126 } 127 128 template <class ELFT> 129 Error ELFSectionSizer<ELFT>::visit(SectionIndexSection &) { 130 return Error::success(); 131 } 132 133 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(CompressedSection &) { 134 return Error::success(); 135 } 136 137 template <class ELFT> 138 Error ELFSectionSizer<ELFT>::visit(DecompressedSection &) { 139 return Error::success(); 140 } 141 142 Error BinarySectionWriter::visit(const SectionIndexSection &Sec) { 143 return createStringError(errc::operation_not_permitted, 144 "cannot write symbol section index table '" + 145 Sec.Name + "' "); 146 } 147 148 Error BinarySectionWriter::visit(const SymbolTableSection &Sec) { 149 return createStringError(errc::operation_not_permitted, 150 "cannot write symbol table '" + Sec.Name + 151 "' out to binary"); 152 } 153 154 Error BinarySectionWriter::visit(const RelocationSection &Sec) { 155 return createStringError(errc::operation_not_permitted, 156 "cannot write relocation section '" + Sec.Name + 157 "' out to binary"); 158 } 159 160 Error BinarySectionWriter::visit(const GnuDebugLinkSection &Sec) { 161 return createStringError(errc::operation_not_permitted, 162 "cannot write '" + Sec.Name + "' out to binary"); 163 } 164 165 Error BinarySectionWriter::visit(const GroupSection &Sec) { 166 return createStringError(errc::operation_not_permitted, 167 "cannot write '" + Sec.Name + "' out to binary"); 168 } 169 170 Error SectionWriter::visit(const Section &Sec) { 171 if (Sec.Type != SHT_NOBITS) 172 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset); 173 174 return Error::success(); 175 } 176 177 static bool addressOverflows32bit(uint64_t Addr) { 178 // Sign extended 32 bit addresses (e.g 0xFFFFFFFF80000000) are ok 179 return Addr > UINT32_MAX && Addr + 0x80000000 > UINT32_MAX; 180 } 181 182 template <class T> static T checkedGetHex(StringRef S) { 183 T Value; 184 bool Fail = S.getAsInteger(16, Value); 185 assert(!Fail); 186 (void)Fail; 187 return Value; 188 } 189 190 // Fills exactly Len bytes of buffer with hexadecimal characters 191 // representing value 'X' 192 template <class T, class Iterator> 193 static Iterator toHexStr(T X, Iterator It, size_t Len) { 194 // Fill range with '0' 195 std::fill(It, It + Len, '0'); 196 197 for (long I = Len - 1; I >= 0; --I) { 198 unsigned char Mod = static_cast<unsigned char>(X) & 15; 199 *(It + I) = hexdigit(Mod, false); 200 X >>= 4; 201 } 202 assert(X == 0); 203 return It + Len; 204 } 205 206 uint8_t IHexRecord::getChecksum(StringRef S) { 207 assert((S.size() & 1) == 0); 208 uint8_t Checksum = 0; 209 while (!S.empty()) { 210 Checksum += checkedGetHex<uint8_t>(S.take_front(2)); 211 S = S.drop_front(2); 212 } 213 return -Checksum; 214 } 215 216 IHexLineData IHexRecord::getLine(uint8_t Type, uint16_t Addr, 217 ArrayRef<uint8_t> Data) { 218 IHexLineData Line(getLineLength(Data.size())); 219 assert(Line.size()); 220 auto Iter = Line.begin(); 221 *Iter++ = ':'; 222 Iter = toHexStr(Data.size(), Iter, 2); 223 Iter = toHexStr(Addr, Iter, 4); 224 Iter = toHexStr(Type, Iter, 2); 225 for (uint8_t X : Data) 226 Iter = toHexStr(X, Iter, 2); 227 StringRef S(Line.data() + 1, std::distance(Line.begin() + 1, Iter)); 228 Iter = toHexStr(getChecksum(S), Iter, 2); 229 *Iter++ = '\r'; 230 *Iter++ = '\n'; 231 assert(Iter == Line.end()); 232 return Line; 233 } 234 235 static Error checkRecord(const IHexRecord &R) { 236 switch (R.Type) { 237 case IHexRecord::Data: 238 if (R.HexData.size() == 0) 239 return createStringError( 240 errc::invalid_argument, 241 "zero data length is not allowed for data records"); 242 break; 243 case IHexRecord::EndOfFile: 244 break; 245 case IHexRecord::SegmentAddr: 246 // 20-bit segment address. Data length must be 2 bytes 247 // (4 bytes in hex) 248 if (R.HexData.size() != 4) 249 return createStringError( 250 errc::invalid_argument, 251 "segment address data should be 2 bytes in size"); 252 break; 253 case IHexRecord::StartAddr80x86: 254 case IHexRecord::StartAddr: 255 if (R.HexData.size() != 8) 256 return createStringError(errc::invalid_argument, 257 "start address data should be 4 bytes in size"); 258 // According to Intel HEX specification '03' record 259 // only specifies the code address within the 20-bit 260 // segmented address space of the 8086/80186. This 261 // means 12 high order bits should be zeroes. 262 if (R.Type == IHexRecord::StartAddr80x86 && 263 R.HexData.take_front(3) != "000") 264 return createStringError(errc::invalid_argument, 265 "start address exceeds 20 bit for 80x86"); 266 break; 267 case IHexRecord::ExtendedAddr: 268 // 16-31 bits of linear base address 269 if (R.HexData.size() != 4) 270 return createStringError( 271 errc::invalid_argument, 272 "extended address data should be 2 bytes in size"); 273 break; 274 default: 275 // Unknown record type 276 return createStringError(errc::invalid_argument, "unknown record type: %u", 277 static_cast<unsigned>(R.Type)); 278 } 279 return Error::success(); 280 } 281 282 // Checks that IHEX line contains valid characters. 283 // This allows converting hexadecimal data to integers 284 // without extra verification. 285 static Error checkChars(StringRef Line) { 286 assert(!Line.empty()); 287 if (Line[0] != ':') 288 return createStringError(errc::invalid_argument, 289 "missing ':' in the beginning of line."); 290 291 for (size_t Pos = 1; Pos < Line.size(); ++Pos) 292 if (hexDigitValue(Line[Pos]) == -1U) 293 return createStringError(errc::invalid_argument, 294 "invalid character at position %zu.", Pos + 1); 295 return Error::success(); 296 } 297 298 Expected<IHexRecord> IHexRecord::parse(StringRef Line) { 299 assert(!Line.empty()); 300 301 // ':' + Length + Address + Type + Checksum with empty data ':LLAAAATTCC' 302 if (Line.size() < 11) 303 return createStringError(errc::invalid_argument, 304 "line is too short: %zu chars.", Line.size()); 305 306 if (Error E = checkChars(Line)) 307 return std::move(E); 308 309 IHexRecord Rec; 310 size_t DataLen = checkedGetHex<uint8_t>(Line.substr(1, 2)); 311 if (Line.size() != getLength(DataLen)) 312 return createStringError(errc::invalid_argument, 313 "invalid line length %zu (should be %zu)", 314 Line.size(), getLength(DataLen)); 315 316 Rec.Addr = checkedGetHex<uint16_t>(Line.substr(3, 4)); 317 Rec.Type = checkedGetHex<uint8_t>(Line.substr(7, 2)); 318 Rec.HexData = Line.substr(9, DataLen * 2); 319 320 if (getChecksum(Line.drop_front(1)) != 0) 321 return createStringError(errc::invalid_argument, "incorrect checksum."); 322 if (Error E = checkRecord(Rec)) 323 return std::move(E); 324 return Rec; 325 } 326 327 static uint64_t sectionPhysicalAddr(const SectionBase *Sec) { 328 Segment *Seg = Sec->ParentSegment; 329 if (Seg && Seg->Type != ELF::PT_LOAD) 330 Seg = nullptr; 331 return Seg ? Seg->PAddr + Sec->OriginalOffset - Seg->OriginalOffset 332 : Sec->Addr; 333 } 334 335 void IHexSectionWriterBase::writeSection(const SectionBase *Sec, 336 ArrayRef<uint8_t> Data) { 337 assert(Data.size() == Sec->Size); 338 const uint32_t ChunkSize = 16; 339 uint32_t Addr = sectionPhysicalAddr(Sec) & 0xFFFFFFFFU; 340 while (!Data.empty()) { 341 uint64_t DataSize = std::min<uint64_t>(Data.size(), ChunkSize); 342 if (Addr > SegmentAddr + BaseAddr + 0xFFFFU) { 343 if (Addr > 0xFFFFFU) { 344 // Write extended address record, zeroing segment address 345 // if needed. 346 if (SegmentAddr != 0) 347 SegmentAddr = writeSegmentAddr(0U); 348 BaseAddr = writeBaseAddr(Addr); 349 } else { 350 // We can still remain 16-bit 351 SegmentAddr = writeSegmentAddr(Addr); 352 } 353 } 354 uint64_t SegOffset = Addr - BaseAddr - SegmentAddr; 355 assert(SegOffset <= 0xFFFFU); 356 DataSize = std::min(DataSize, 0x10000U - SegOffset); 357 writeData(0, SegOffset, Data.take_front(DataSize)); 358 Addr += DataSize; 359 Data = Data.drop_front(DataSize); 360 } 361 } 362 363 uint64_t IHexSectionWriterBase::writeSegmentAddr(uint64_t Addr) { 364 assert(Addr <= 0xFFFFFU); 365 uint8_t Data[] = {static_cast<uint8_t>((Addr & 0xF0000U) >> 12), 0}; 366 writeData(2, 0, Data); 367 return Addr & 0xF0000U; 368 } 369 370 uint64_t IHexSectionWriterBase::writeBaseAddr(uint64_t Addr) { 371 assert(Addr <= 0xFFFFFFFFU); 372 uint64_t Base = Addr & 0xFFFF0000U; 373 uint8_t Data[] = {static_cast<uint8_t>(Base >> 24), 374 static_cast<uint8_t>((Base >> 16) & 0xFF)}; 375 writeData(4, 0, Data); 376 return Base; 377 } 378 379 void IHexSectionWriterBase::writeData(uint8_t, uint16_t, 380 ArrayRef<uint8_t> Data) { 381 Offset += IHexRecord::getLineLength(Data.size()); 382 } 383 384 Error IHexSectionWriterBase::visit(const Section &Sec) { 385 writeSection(&Sec, Sec.Contents); 386 return Error::success(); 387 } 388 389 Error IHexSectionWriterBase::visit(const OwnedDataSection &Sec) { 390 writeSection(&Sec, Sec.Data); 391 return Error::success(); 392 } 393 394 Error IHexSectionWriterBase::visit(const StringTableSection &Sec) { 395 // Check that sizer has already done its work 396 assert(Sec.Size == Sec.StrTabBuilder.getSize()); 397 // We are free to pass an invalid pointer to writeSection as long 398 // as we don't actually write any data. The real writer class has 399 // to override this method . 400 writeSection(&Sec, {nullptr, static_cast<size_t>(Sec.Size)}); 401 return Error::success(); 402 } 403 404 Error IHexSectionWriterBase::visit(const DynamicRelocationSection &Sec) { 405 writeSection(&Sec, Sec.Contents); 406 return Error::success(); 407 } 408 409 void IHexSectionWriter::writeData(uint8_t Type, uint16_t Addr, 410 ArrayRef<uint8_t> Data) { 411 IHexLineData HexData = IHexRecord::getLine(Type, Addr, Data); 412 memcpy(Out.getBufferStart() + Offset, HexData.data(), HexData.size()); 413 Offset += HexData.size(); 414 } 415 416 Error IHexSectionWriter::visit(const StringTableSection &Sec) { 417 assert(Sec.Size == Sec.StrTabBuilder.getSize()); 418 std::vector<uint8_t> Data(Sec.Size); 419 Sec.StrTabBuilder.write(Data.data()); 420 writeSection(&Sec, Data); 421 return Error::success(); 422 } 423 424 Error Section::accept(SectionVisitor &Visitor) const { 425 return Visitor.visit(*this); 426 } 427 428 Error Section::accept(MutableSectionVisitor &Visitor) { 429 return Visitor.visit(*this); 430 } 431 432 Error SectionWriter::visit(const OwnedDataSection &Sec) { 433 llvm::copy(Sec.Data, Out.getBufferStart() + Sec.Offset); 434 return Error::success(); 435 } 436 437 static constexpr std::array<uint8_t, 4> ZlibGnuMagic = {{'Z', 'L', 'I', 'B'}}; 438 439 static bool isDataGnuCompressed(ArrayRef<uint8_t> Data) { 440 return Data.size() > ZlibGnuMagic.size() && 441 std::equal(ZlibGnuMagic.begin(), ZlibGnuMagic.end(), Data.data()); 442 } 443 444 template <class ELFT> 445 static std::tuple<uint64_t, uint64_t> 446 getDecompressedSizeAndAlignment(ArrayRef<uint8_t> Data) { 447 const bool IsGnuDebug = isDataGnuCompressed(Data); 448 const uint64_t DecompressedSize = 449 IsGnuDebug 450 ? support::endian::read64be(Data.data() + ZlibGnuMagic.size()) 451 : reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data.data())->ch_size; 452 const uint64_t DecompressedAlign = 453 IsGnuDebug ? 1 454 : reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data.data()) 455 ->ch_addralign; 456 457 return std::make_tuple(DecompressedSize, DecompressedAlign); 458 } 459 460 template <class ELFT> 461 Error ELFSectionWriter<ELFT>::visit(const DecompressedSection &Sec) { 462 const size_t DataOffset = isDataGnuCompressed(Sec.OriginalData) 463 ? (ZlibGnuMagic.size() + sizeof(Sec.Size)) 464 : sizeof(Elf_Chdr_Impl<ELFT>); 465 466 StringRef CompressedContent( 467 reinterpret_cast<const char *>(Sec.OriginalData.data()) + DataOffset, 468 Sec.OriginalData.size() - DataOffset); 469 470 SmallVector<char, 128> DecompressedContent; 471 if (Error Err = zlib::uncompress(CompressedContent, DecompressedContent, 472 static_cast<size_t>(Sec.Size))) 473 return createStringError(errc::invalid_argument, 474 "'" + Sec.Name + "': " + toString(std::move(Err))); 475 476 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 477 std::copy(DecompressedContent.begin(), DecompressedContent.end(), Buf); 478 479 return Error::success(); 480 } 481 482 Error BinarySectionWriter::visit(const DecompressedSection &Sec) { 483 return createStringError(errc::operation_not_permitted, 484 "cannot write compressed section '" + Sec.Name + 485 "' "); 486 } 487 488 Error DecompressedSection::accept(SectionVisitor &Visitor) const { 489 return Visitor.visit(*this); 490 } 491 492 Error DecompressedSection::accept(MutableSectionVisitor &Visitor) { 493 return Visitor.visit(*this); 494 } 495 496 Error OwnedDataSection::accept(SectionVisitor &Visitor) const { 497 return Visitor.visit(*this); 498 } 499 500 Error OwnedDataSection::accept(MutableSectionVisitor &Visitor) { 501 return Visitor.visit(*this); 502 } 503 504 void OwnedDataSection::appendHexData(StringRef HexData) { 505 assert((HexData.size() & 1) == 0); 506 while (!HexData.empty()) { 507 Data.push_back(checkedGetHex<uint8_t>(HexData.take_front(2))); 508 HexData = HexData.drop_front(2); 509 } 510 Size = Data.size(); 511 } 512 513 Error BinarySectionWriter::visit(const CompressedSection &Sec) { 514 return createStringError(errc::operation_not_permitted, 515 "cannot write compressed section '" + Sec.Name + 516 "' "); 517 } 518 519 template <class ELFT> 520 Error ELFSectionWriter<ELFT>::visit(const CompressedSection &Sec) { 521 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 522 if (Sec.CompressionType == DebugCompressionType::None) { 523 std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(), Buf); 524 return Error::success(); 525 } 526 527 if (Sec.CompressionType == DebugCompressionType::GNU) { 528 const char *Magic = "ZLIB"; 529 memcpy(Buf, Magic, strlen(Magic)); 530 Buf += strlen(Magic); 531 const uint64_t DecompressedSize = 532 support::endian::read64be(&Sec.DecompressedSize); 533 memcpy(Buf, &DecompressedSize, sizeof(DecompressedSize)); 534 Buf += sizeof(DecompressedSize); 535 } else { 536 Elf_Chdr_Impl<ELFT> Chdr; 537 Chdr.ch_type = ELF::ELFCOMPRESS_ZLIB; 538 Chdr.ch_size = Sec.DecompressedSize; 539 Chdr.ch_addralign = Sec.DecompressedAlign; 540 memcpy(Buf, &Chdr, sizeof(Chdr)); 541 Buf += sizeof(Chdr); 542 } 543 544 std::copy(Sec.CompressedData.begin(), Sec.CompressedData.end(), Buf); 545 return Error::success(); 546 } 547 548 CompressedSection::CompressedSection(const SectionBase &Sec, 549 DebugCompressionType CompressionType) 550 : SectionBase(Sec), CompressionType(CompressionType), 551 DecompressedSize(Sec.OriginalData.size()), DecompressedAlign(Sec.Align) { 552 zlib::compress(StringRef(reinterpret_cast<const char *>(OriginalData.data()), 553 OriginalData.size()), 554 CompressedData); 555 556 size_t ChdrSize; 557 if (CompressionType == DebugCompressionType::GNU) { 558 Name = ".z" + Sec.Name.substr(1); 559 ChdrSize = sizeof("ZLIB") - 1 + sizeof(uint64_t); 560 } else { 561 Flags |= ELF::SHF_COMPRESSED; 562 ChdrSize = 563 std::max(std::max(sizeof(object::Elf_Chdr_Impl<object::ELF64LE>), 564 sizeof(object::Elf_Chdr_Impl<object::ELF64BE>)), 565 std::max(sizeof(object::Elf_Chdr_Impl<object::ELF32LE>), 566 sizeof(object::Elf_Chdr_Impl<object::ELF32BE>))); 567 } 568 Size = ChdrSize + CompressedData.size(); 569 Align = 8; 570 } 571 572 CompressedSection::CompressedSection(ArrayRef<uint8_t> CompressedData, 573 uint64_t DecompressedSize, 574 uint64_t DecompressedAlign) 575 : CompressionType(DebugCompressionType::None), 576 DecompressedSize(DecompressedSize), DecompressedAlign(DecompressedAlign) { 577 OriginalData = CompressedData; 578 } 579 580 Error CompressedSection::accept(SectionVisitor &Visitor) const { 581 return Visitor.visit(*this); 582 } 583 584 Error CompressedSection::accept(MutableSectionVisitor &Visitor) { 585 return Visitor.visit(*this); 586 } 587 588 void StringTableSection::addString(StringRef Name) { StrTabBuilder.add(Name); } 589 590 uint32_t StringTableSection::findIndex(StringRef Name) const { 591 return StrTabBuilder.getOffset(Name); 592 } 593 594 void StringTableSection::prepareForLayout() { 595 StrTabBuilder.finalize(); 596 Size = StrTabBuilder.getSize(); 597 } 598 599 Error SectionWriter::visit(const StringTableSection &Sec) { 600 Sec.StrTabBuilder.write(reinterpret_cast<uint8_t *>(Out.getBufferStart()) + 601 Sec.Offset); 602 return Error::success(); 603 } 604 605 Error StringTableSection::accept(SectionVisitor &Visitor) const { 606 return Visitor.visit(*this); 607 } 608 609 Error StringTableSection::accept(MutableSectionVisitor &Visitor) { 610 return Visitor.visit(*this); 611 } 612 613 template <class ELFT> 614 Error ELFSectionWriter<ELFT>::visit(const SectionIndexSection &Sec) { 615 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 616 llvm::copy(Sec.Indexes, reinterpret_cast<Elf_Word *>(Buf)); 617 return Error::success(); 618 } 619 620 Error SectionIndexSection::initialize(SectionTableRef SecTable) { 621 Size = 0; 622 Expected<SymbolTableSection *> Sec = 623 SecTable.getSectionOfType<SymbolTableSection>( 624 Link, 625 "Link field value " + Twine(Link) + " in section " + Name + 626 " is invalid", 627 "Link field value " + Twine(Link) + " in section " + Name + 628 " is not a symbol table"); 629 if (!Sec) 630 return Sec.takeError(); 631 632 setSymTab(*Sec); 633 Symbols->setShndxTable(this); 634 return Error::success(); 635 } 636 637 void SectionIndexSection::finalize() { Link = Symbols->Index; } 638 639 Error SectionIndexSection::accept(SectionVisitor &Visitor) const { 640 return Visitor.visit(*this); 641 } 642 643 Error SectionIndexSection::accept(MutableSectionVisitor &Visitor) { 644 return Visitor.visit(*this); 645 } 646 647 static bool isValidReservedSectionIndex(uint16_t Index, uint16_t Machine) { 648 switch (Index) { 649 case SHN_ABS: 650 case SHN_COMMON: 651 return true; 652 } 653 654 if (Machine == EM_AMDGPU) { 655 return Index == SHN_AMDGPU_LDS; 656 } 657 658 if (Machine == EM_HEXAGON) { 659 switch (Index) { 660 case SHN_HEXAGON_SCOMMON: 661 case SHN_HEXAGON_SCOMMON_1: 662 case SHN_HEXAGON_SCOMMON_2: 663 case SHN_HEXAGON_SCOMMON_4: 664 case SHN_HEXAGON_SCOMMON_8: 665 return true; 666 } 667 } 668 return false; 669 } 670 671 // Large indexes force us to clarify exactly what this function should do. This 672 // function should return the value that will appear in st_shndx when written 673 // out. 674 uint16_t Symbol::getShndx() const { 675 if (DefinedIn != nullptr) { 676 if (DefinedIn->Index >= SHN_LORESERVE) 677 return SHN_XINDEX; 678 return DefinedIn->Index; 679 } 680 681 if (ShndxType == SYMBOL_SIMPLE_INDEX) { 682 // This means that we don't have a defined section but we do need to 683 // output a legitimate section index. 684 return SHN_UNDEF; 685 } 686 687 assert(ShndxType == SYMBOL_ABS || ShndxType == SYMBOL_COMMON || 688 (ShndxType >= SYMBOL_LOPROC && ShndxType <= SYMBOL_HIPROC) || 689 (ShndxType >= SYMBOL_LOOS && ShndxType <= SYMBOL_HIOS)); 690 return static_cast<uint16_t>(ShndxType); 691 } 692 693 bool Symbol::isCommon() const { return getShndx() == SHN_COMMON; } 694 695 void SymbolTableSection::assignIndices() { 696 uint32_t Index = 0; 697 for (auto &Sym : Symbols) 698 Sym->Index = Index++; 699 } 700 701 void SymbolTableSection::addSymbol(Twine Name, uint8_t Bind, uint8_t Type, 702 SectionBase *DefinedIn, uint64_t Value, 703 uint8_t Visibility, uint16_t Shndx, 704 uint64_t SymbolSize) { 705 Symbol Sym; 706 Sym.Name = Name.str(); 707 Sym.Binding = Bind; 708 Sym.Type = Type; 709 Sym.DefinedIn = DefinedIn; 710 if (DefinedIn != nullptr) 711 DefinedIn->HasSymbol = true; 712 if (DefinedIn == nullptr) { 713 if (Shndx >= SHN_LORESERVE) 714 Sym.ShndxType = static_cast<SymbolShndxType>(Shndx); 715 else 716 Sym.ShndxType = SYMBOL_SIMPLE_INDEX; 717 } 718 Sym.Value = Value; 719 Sym.Visibility = Visibility; 720 Sym.Size = SymbolSize; 721 Sym.Index = Symbols.size(); 722 Symbols.emplace_back(std::make_unique<Symbol>(Sym)); 723 Size += this->EntrySize; 724 } 725 726 Error SymbolTableSection::removeSectionReferences( 727 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 728 if (ToRemove(SectionIndexTable)) 729 SectionIndexTable = nullptr; 730 if (ToRemove(SymbolNames)) { 731 if (!AllowBrokenLinks) 732 return createStringError( 733 llvm::errc::invalid_argument, 734 "string table '%s' cannot be removed because it is " 735 "referenced by the symbol table '%s'", 736 SymbolNames->Name.data(), this->Name.data()); 737 SymbolNames = nullptr; 738 } 739 return removeSymbols( 740 [ToRemove](const Symbol &Sym) { return ToRemove(Sym.DefinedIn); }); 741 } 742 743 void SymbolTableSection::updateSymbols(function_ref<void(Symbol &)> Callable) { 744 std::for_each(std::begin(Symbols) + 1, std::end(Symbols), 745 [Callable](SymPtr &Sym) { Callable(*Sym); }); 746 std::stable_partition( 747 std::begin(Symbols), std::end(Symbols), 748 [](const SymPtr &Sym) { return Sym->Binding == STB_LOCAL; }); 749 assignIndices(); 750 } 751 752 Error SymbolTableSection::removeSymbols( 753 function_ref<bool(const Symbol &)> ToRemove) { 754 Symbols.erase( 755 std::remove_if(std::begin(Symbols) + 1, std::end(Symbols), 756 [ToRemove](const SymPtr &Sym) { return ToRemove(*Sym); }), 757 std::end(Symbols)); 758 Size = Symbols.size() * EntrySize; 759 assignIndices(); 760 return Error::success(); 761 } 762 763 void SymbolTableSection::replaceSectionReferences( 764 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 765 for (std::unique_ptr<Symbol> &Sym : Symbols) 766 if (SectionBase *To = FromTo.lookup(Sym->DefinedIn)) 767 Sym->DefinedIn = To; 768 } 769 770 Error SymbolTableSection::initialize(SectionTableRef SecTable) { 771 Size = 0; 772 Expected<StringTableSection *> Sec = 773 SecTable.getSectionOfType<StringTableSection>( 774 Link, 775 "Symbol table has link index of " + Twine(Link) + 776 " which is not a valid index", 777 "Symbol table has link index of " + Twine(Link) + 778 " which is not a string table"); 779 if (!Sec) 780 return Sec.takeError(); 781 782 setStrTab(*Sec); 783 return Error::success(); 784 } 785 786 void SymbolTableSection::finalize() { 787 uint32_t MaxLocalIndex = 0; 788 for (std::unique_ptr<Symbol> &Sym : Symbols) { 789 Sym->NameIndex = 790 SymbolNames == nullptr ? 0 : SymbolNames->findIndex(Sym->Name); 791 if (Sym->Binding == STB_LOCAL) 792 MaxLocalIndex = std::max(MaxLocalIndex, Sym->Index); 793 } 794 // Now we need to set the Link and Info fields. 795 Link = SymbolNames == nullptr ? 0 : SymbolNames->Index; 796 Info = MaxLocalIndex + 1; 797 } 798 799 void SymbolTableSection::prepareForLayout() { 800 // Reserve proper amount of space in section index table, so we can 801 // layout sections correctly. We will fill the table with correct 802 // indexes later in fillShdnxTable. 803 if (SectionIndexTable) 804 SectionIndexTable->reserve(Symbols.size()); 805 806 // Add all of our strings to SymbolNames so that SymbolNames has the right 807 // size before layout is decided. 808 // If the symbol names section has been removed, don't try to add strings to 809 // the table. 810 if (SymbolNames != nullptr) 811 for (std::unique_ptr<Symbol> &Sym : Symbols) 812 SymbolNames->addString(Sym->Name); 813 } 814 815 void SymbolTableSection::fillShndxTable() { 816 if (SectionIndexTable == nullptr) 817 return; 818 // Fill section index table with real section indexes. This function must 819 // be called after assignOffsets. 820 for (const std::unique_ptr<Symbol> &Sym : Symbols) { 821 if (Sym->DefinedIn != nullptr && Sym->DefinedIn->Index >= SHN_LORESERVE) 822 SectionIndexTable->addIndex(Sym->DefinedIn->Index); 823 else 824 SectionIndexTable->addIndex(SHN_UNDEF); 825 } 826 } 827 828 Expected<const Symbol *> 829 SymbolTableSection::getSymbolByIndex(uint32_t Index) const { 830 if (Symbols.size() <= Index) 831 return createStringError(errc::invalid_argument, 832 "invalid symbol index: " + Twine(Index)); 833 return Symbols[Index].get(); 834 } 835 836 Expected<Symbol *> SymbolTableSection::getSymbolByIndex(uint32_t Index) { 837 Expected<const Symbol *> Sym = 838 static_cast<const SymbolTableSection *>(this)->getSymbolByIndex(Index); 839 if (!Sym) 840 return Sym.takeError(); 841 842 return const_cast<Symbol *>(*Sym); 843 } 844 845 template <class ELFT> 846 Error ELFSectionWriter<ELFT>::visit(const SymbolTableSection &Sec) { 847 Elf_Sym *Sym = reinterpret_cast<Elf_Sym *>(Out.getBufferStart() + Sec.Offset); 848 // Loop though symbols setting each entry of the symbol table. 849 for (const std::unique_ptr<Symbol> &Symbol : Sec.Symbols) { 850 Sym->st_name = Symbol->NameIndex; 851 Sym->st_value = Symbol->Value; 852 Sym->st_size = Symbol->Size; 853 Sym->st_other = Symbol->Visibility; 854 Sym->setBinding(Symbol->Binding); 855 Sym->setType(Symbol->Type); 856 Sym->st_shndx = Symbol->getShndx(); 857 ++Sym; 858 } 859 return Error::success(); 860 } 861 862 Error SymbolTableSection::accept(SectionVisitor &Visitor) const { 863 return Visitor.visit(*this); 864 } 865 866 Error SymbolTableSection::accept(MutableSectionVisitor &Visitor) { 867 return Visitor.visit(*this); 868 } 869 870 StringRef RelocationSectionBase::getNamePrefix() const { 871 switch (Type) { 872 case SHT_REL: 873 return ".rel"; 874 case SHT_RELA: 875 return ".rela"; 876 default: 877 llvm_unreachable("not a relocation section"); 878 } 879 } 880 881 Error RelocationSection::removeSectionReferences( 882 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 883 if (ToRemove(Symbols)) { 884 if (!AllowBrokenLinks) 885 return createStringError( 886 llvm::errc::invalid_argument, 887 "symbol table '%s' cannot be removed because it is " 888 "referenced by the relocation section '%s'", 889 Symbols->Name.data(), this->Name.data()); 890 Symbols = nullptr; 891 } 892 893 for (const Relocation &R : Relocations) { 894 if (!R.RelocSymbol || !R.RelocSymbol->DefinedIn || 895 !ToRemove(R.RelocSymbol->DefinedIn)) 896 continue; 897 return createStringError(llvm::errc::invalid_argument, 898 "section '%s' cannot be removed: (%s+0x%" PRIx64 899 ") has relocation against symbol '%s'", 900 R.RelocSymbol->DefinedIn->Name.data(), 901 SecToApplyRel->Name.data(), R.Offset, 902 R.RelocSymbol->Name.c_str()); 903 } 904 905 return Error::success(); 906 } 907 908 template <class SymTabType> 909 Error RelocSectionWithSymtabBase<SymTabType>::initialize( 910 SectionTableRef SecTable) { 911 if (Link != SHN_UNDEF) { 912 Expected<SymTabType *> Sec = SecTable.getSectionOfType<SymTabType>( 913 Link, 914 "Link field value " + Twine(Link) + " in section " + Name + 915 " is invalid", 916 "Link field value " + Twine(Link) + " in section " + Name + 917 " is not a symbol table"); 918 if (!Sec) 919 return Sec.takeError(); 920 921 setSymTab(*Sec); 922 } 923 924 if (Info != SHN_UNDEF) { 925 Expected<SectionBase *> Sec = 926 SecTable.getSection(Info, "Info field value " + Twine(Info) + 927 " in section " + Name + " is invalid"); 928 if (!Sec) 929 return Sec.takeError(); 930 931 setSection(*Sec); 932 } else 933 setSection(nullptr); 934 935 return Error::success(); 936 } 937 938 template <class SymTabType> 939 void RelocSectionWithSymtabBase<SymTabType>::finalize() { 940 this->Link = Symbols ? Symbols->Index : 0; 941 942 if (SecToApplyRel != nullptr) 943 this->Info = SecToApplyRel->Index; 944 } 945 946 template <class ELFT> 947 static void setAddend(Elf_Rel_Impl<ELFT, false> &, uint64_t) {} 948 949 template <class ELFT> 950 static void setAddend(Elf_Rel_Impl<ELFT, true> &Rela, uint64_t Addend) { 951 Rela.r_addend = Addend; 952 } 953 954 template <class RelRange, class T> 955 static void writeRel(const RelRange &Relocations, T *Buf, bool IsMips64EL) { 956 for (const auto &Reloc : Relocations) { 957 Buf->r_offset = Reloc.Offset; 958 setAddend(*Buf, Reloc.Addend); 959 Buf->setSymbolAndType(Reloc.RelocSymbol ? Reloc.RelocSymbol->Index : 0, 960 Reloc.Type, IsMips64EL); 961 ++Buf; 962 } 963 } 964 965 template <class ELFT> 966 Error ELFSectionWriter<ELFT>::visit(const RelocationSection &Sec) { 967 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 968 if (Sec.Type == SHT_REL) 969 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rel *>(Buf), 970 Sec.getObject().IsMips64EL); 971 else 972 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rela *>(Buf), 973 Sec.getObject().IsMips64EL); 974 return Error::success(); 975 } 976 977 Error RelocationSection::accept(SectionVisitor &Visitor) const { 978 return Visitor.visit(*this); 979 } 980 981 Error RelocationSection::accept(MutableSectionVisitor &Visitor) { 982 return Visitor.visit(*this); 983 } 984 985 Error RelocationSection::removeSymbols( 986 function_ref<bool(const Symbol &)> ToRemove) { 987 for (const Relocation &Reloc : Relocations) 988 if (Reloc.RelocSymbol && ToRemove(*Reloc.RelocSymbol)) 989 return createStringError( 990 llvm::errc::invalid_argument, 991 "not stripping symbol '%s' because it is named in a relocation", 992 Reloc.RelocSymbol->Name.data()); 993 return Error::success(); 994 } 995 996 void RelocationSection::markSymbols() { 997 for (const Relocation &Reloc : Relocations) 998 if (Reloc.RelocSymbol) 999 Reloc.RelocSymbol->Referenced = true; 1000 } 1001 1002 void RelocationSection::replaceSectionReferences( 1003 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 1004 // Update the target section if it was replaced. 1005 if (SectionBase *To = FromTo.lookup(SecToApplyRel)) 1006 SecToApplyRel = To; 1007 } 1008 1009 Error SectionWriter::visit(const DynamicRelocationSection &Sec) { 1010 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset); 1011 return Error::success(); 1012 } 1013 1014 Error DynamicRelocationSection::accept(SectionVisitor &Visitor) const { 1015 return Visitor.visit(*this); 1016 } 1017 1018 Error DynamicRelocationSection::accept(MutableSectionVisitor &Visitor) { 1019 return Visitor.visit(*this); 1020 } 1021 1022 Error DynamicRelocationSection::removeSectionReferences( 1023 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 1024 if (ToRemove(Symbols)) { 1025 if (!AllowBrokenLinks) 1026 return createStringError( 1027 llvm::errc::invalid_argument, 1028 "symbol table '%s' cannot be removed because it is " 1029 "referenced by the relocation section '%s'", 1030 Symbols->Name.data(), this->Name.data()); 1031 Symbols = nullptr; 1032 } 1033 1034 // SecToApplyRel contains a section referenced by sh_info field. It keeps 1035 // a section to which the relocation section applies. When we remove any 1036 // sections we also remove their relocation sections. Since we do that much 1037 // earlier, this assert should never be triggered. 1038 assert(!SecToApplyRel || !ToRemove(SecToApplyRel)); 1039 return Error::success(); 1040 } 1041 1042 Error Section::removeSectionReferences( 1043 bool AllowBrokenDependency, 1044 function_ref<bool(const SectionBase *)> ToRemove) { 1045 if (ToRemove(LinkSection)) { 1046 if (!AllowBrokenDependency) 1047 return createStringError(llvm::errc::invalid_argument, 1048 "section '%s' cannot be removed because it is " 1049 "referenced by the section '%s'", 1050 LinkSection->Name.data(), this->Name.data()); 1051 LinkSection = nullptr; 1052 } 1053 return Error::success(); 1054 } 1055 1056 void GroupSection::finalize() { 1057 this->Info = Sym ? Sym->Index : 0; 1058 this->Link = SymTab ? SymTab->Index : 0; 1059 // Linker deduplication for GRP_COMDAT is based on Sym->Name. The local/global 1060 // status is not part of the equation. If Sym is localized, the intention is 1061 // likely to make the group fully localized. Drop GRP_COMDAT to suppress 1062 // deduplication. See https://groups.google.com/g/generic-abi/c/2X6mR-s2zoc 1063 if ((FlagWord & GRP_COMDAT) && Sym && Sym->Binding == STB_LOCAL) 1064 this->FlagWord &= ~GRP_COMDAT; 1065 } 1066 1067 Error GroupSection::removeSectionReferences( 1068 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { 1069 if (ToRemove(SymTab)) { 1070 if (!AllowBrokenLinks) 1071 return createStringError( 1072 llvm::errc::invalid_argument, 1073 "section '.symtab' cannot be removed because it is " 1074 "referenced by the group section '%s'", 1075 this->Name.data()); 1076 SymTab = nullptr; 1077 Sym = nullptr; 1078 } 1079 llvm::erase_if(GroupMembers, ToRemove); 1080 return Error::success(); 1081 } 1082 1083 Error GroupSection::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) { 1084 if (ToRemove(*Sym)) 1085 return createStringError(llvm::errc::invalid_argument, 1086 "symbol '%s' cannot be removed because it is " 1087 "referenced by the section '%s[%d]'", 1088 Sym->Name.data(), this->Name.data(), this->Index); 1089 return Error::success(); 1090 } 1091 1092 void GroupSection::markSymbols() { 1093 if (Sym) 1094 Sym->Referenced = true; 1095 } 1096 1097 void GroupSection::replaceSectionReferences( 1098 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 1099 for (SectionBase *&Sec : GroupMembers) 1100 if (SectionBase *To = FromTo.lookup(Sec)) 1101 Sec = To; 1102 } 1103 1104 void GroupSection::onRemove() { 1105 // As the header section of the group is removed, drop the Group flag in its 1106 // former members. 1107 for (SectionBase *Sec : GroupMembers) 1108 Sec->Flags &= ~SHF_GROUP; 1109 } 1110 1111 Error Section::initialize(SectionTableRef SecTable) { 1112 if (Link == ELF::SHN_UNDEF) 1113 return Error::success(); 1114 1115 Expected<SectionBase *> Sec = 1116 SecTable.getSection(Link, "Link field value " + Twine(Link) + 1117 " in section " + Name + " is invalid"); 1118 if (!Sec) 1119 return Sec.takeError(); 1120 1121 LinkSection = *Sec; 1122 1123 if (LinkSection->Type == ELF::SHT_SYMTAB) 1124 LinkSection = nullptr; 1125 1126 return Error::success(); 1127 } 1128 1129 void Section::finalize() { this->Link = LinkSection ? LinkSection->Index : 0; } 1130 1131 void GnuDebugLinkSection::init(StringRef File) { 1132 FileName = sys::path::filename(File); 1133 // The format for the .gnu_debuglink starts with the file name and is 1134 // followed by a null terminator and then the CRC32 of the file. The CRC32 1135 // should be 4 byte aligned. So we add the FileName size, a 1 for the null 1136 // byte, and then finally push the size to alignment and add 4. 1137 Size = alignTo(FileName.size() + 1, 4) + 4; 1138 // The CRC32 will only be aligned if we align the whole section. 1139 Align = 4; 1140 Type = OriginalType = ELF::SHT_PROGBITS; 1141 Name = ".gnu_debuglink"; 1142 // For sections not found in segments, OriginalOffset is only used to 1143 // establish the order that sections should go in. By using the maximum 1144 // possible offset we cause this section to wind up at the end. 1145 OriginalOffset = std::numeric_limits<uint64_t>::max(); 1146 } 1147 1148 GnuDebugLinkSection::GnuDebugLinkSection(StringRef File, 1149 uint32_t PrecomputedCRC) 1150 : FileName(File), CRC32(PrecomputedCRC) { 1151 init(File); 1152 } 1153 1154 template <class ELFT> 1155 Error ELFSectionWriter<ELFT>::visit(const GnuDebugLinkSection &Sec) { 1156 unsigned char *Buf = 1157 reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; 1158 Elf_Word *CRC = 1159 reinterpret_cast<Elf_Word *>(Buf + Sec.Size - sizeof(Elf_Word)); 1160 *CRC = Sec.CRC32; 1161 llvm::copy(Sec.FileName, Buf); 1162 return Error::success(); 1163 } 1164 1165 Error GnuDebugLinkSection::accept(SectionVisitor &Visitor) const { 1166 return Visitor.visit(*this); 1167 } 1168 1169 Error GnuDebugLinkSection::accept(MutableSectionVisitor &Visitor) { 1170 return Visitor.visit(*this); 1171 } 1172 1173 template <class ELFT> 1174 Error ELFSectionWriter<ELFT>::visit(const GroupSection &Sec) { 1175 ELF::Elf32_Word *Buf = 1176 reinterpret_cast<ELF::Elf32_Word *>(Out.getBufferStart() + Sec.Offset); 1177 support::endian::write32<ELFT::TargetEndianness>(Buf++, Sec.FlagWord); 1178 for (SectionBase *S : Sec.GroupMembers) 1179 support::endian::write32<ELFT::TargetEndianness>(Buf++, S->Index); 1180 return Error::success(); 1181 } 1182 1183 Error GroupSection::accept(SectionVisitor &Visitor) const { 1184 return Visitor.visit(*this); 1185 } 1186 1187 Error GroupSection::accept(MutableSectionVisitor &Visitor) { 1188 return Visitor.visit(*this); 1189 } 1190 1191 // Returns true IFF a section is wholly inside the range of a segment 1192 static bool sectionWithinSegment(const SectionBase &Sec, const Segment &Seg) { 1193 // If a section is empty it should be treated like it has a size of 1. This is 1194 // to clarify the case when an empty section lies on a boundary between two 1195 // segments and ensures that the section "belongs" to the second segment and 1196 // not the first. 1197 uint64_t SecSize = Sec.Size ? Sec.Size : 1; 1198 1199 // Ignore just added sections. 1200 if (Sec.OriginalOffset == std::numeric_limits<uint64_t>::max()) 1201 return false; 1202 1203 if (Sec.Type == SHT_NOBITS) { 1204 if (!(Sec.Flags & SHF_ALLOC)) 1205 return false; 1206 1207 bool SectionIsTLS = Sec.Flags & SHF_TLS; 1208 bool SegmentIsTLS = Seg.Type == PT_TLS; 1209 if (SectionIsTLS != SegmentIsTLS) 1210 return false; 1211 1212 return Seg.VAddr <= Sec.Addr && 1213 Seg.VAddr + Seg.MemSize >= Sec.Addr + SecSize; 1214 } 1215 1216 return Seg.Offset <= Sec.OriginalOffset && 1217 Seg.Offset + Seg.FileSize >= Sec.OriginalOffset + SecSize; 1218 } 1219 1220 // Returns true IFF a segment's original offset is inside of another segment's 1221 // range. 1222 static bool segmentOverlapsSegment(const Segment &Child, 1223 const Segment &Parent) { 1224 1225 return Parent.OriginalOffset <= Child.OriginalOffset && 1226 Parent.OriginalOffset + Parent.FileSize > Child.OriginalOffset; 1227 } 1228 1229 static bool compareSegmentsByOffset(const Segment *A, const Segment *B) { 1230 // Any segment without a parent segment should come before a segment 1231 // that has a parent segment. 1232 if (A->OriginalOffset < B->OriginalOffset) 1233 return true; 1234 if (A->OriginalOffset > B->OriginalOffset) 1235 return false; 1236 return A->Index < B->Index; 1237 } 1238 1239 void BasicELFBuilder::initFileHeader() { 1240 Obj->Flags = 0x0; 1241 Obj->Type = ET_REL; 1242 Obj->OSABI = ELFOSABI_NONE; 1243 Obj->ABIVersion = 0; 1244 Obj->Entry = 0x0; 1245 Obj->Machine = EM_NONE; 1246 Obj->Version = 1; 1247 } 1248 1249 void BasicELFBuilder::initHeaderSegment() { Obj->ElfHdrSegment.Index = 0; } 1250 1251 StringTableSection *BasicELFBuilder::addStrTab() { 1252 auto &StrTab = Obj->addSection<StringTableSection>(); 1253 StrTab.Name = ".strtab"; 1254 1255 Obj->SectionNames = &StrTab; 1256 return &StrTab; 1257 } 1258 1259 SymbolTableSection *BasicELFBuilder::addSymTab(StringTableSection *StrTab) { 1260 auto &SymTab = Obj->addSection<SymbolTableSection>(); 1261 1262 SymTab.Name = ".symtab"; 1263 SymTab.Link = StrTab->Index; 1264 1265 // The symbol table always needs a null symbol 1266 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0); 1267 1268 Obj->SymbolTable = &SymTab; 1269 return &SymTab; 1270 } 1271 1272 Error BasicELFBuilder::initSections() { 1273 for (SectionBase &Sec : Obj->sections()) 1274 if (Error Err = Sec.initialize(Obj->sections())) 1275 return Err; 1276 1277 return Error::success(); 1278 } 1279 1280 void BinaryELFBuilder::addData(SymbolTableSection *SymTab) { 1281 auto Data = ArrayRef<uint8_t>( 1282 reinterpret_cast<const uint8_t *>(MemBuf->getBufferStart()), 1283 MemBuf->getBufferSize()); 1284 auto &DataSection = Obj->addSection<Section>(Data); 1285 DataSection.Name = ".data"; 1286 DataSection.Type = ELF::SHT_PROGBITS; 1287 DataSection.Size = Data.size(); 1288 DataSection.Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE; 1289 1290 std::string SanitizedFilename = MemBuf->getBufferIdentifier().str(); 1291 std::replace_if( 1292 std::begin(SanitizedFilename), std::end(SanitizedFilename), 1293 [](char C) { return !isAlnum(C); }, '_'); 1294 Twine Prefix = Twine("_binary_") + SanitizedFilename; 1295 1296 SymTab->addSymbol(Prefix + "_start", STB_GLOBAL, STT_NOTYPE, &DataSection, 1297 /*Value=*/0, NewSymbolVisibility, 0, 0); 1298 SymTab->addSymbol(Prefix + "_end", STB_GLOBAL, STT_NOTYPE, &DataSection, 1299 /*Value=*/DataSection.Size, NewSymbolVisibility, 0, 0); 1300 SymTab->addSymbol(Prefix + "_size", STB_GLOBAL, STT_NOTYPE, nullptr, 1301 /*Value=*/DataSection.Size, NewSymbolVisibility, SHN_ABS, 1302 0); 1303 } 1304 1305 Expected<std::unique_ptr<Object>> BinaryELFBuilder::build() { 1306 initFileHeader(); 1307 initHeaderSegment(); 1308 1309 SymbolTableSection *SymTab = addSymTab(addStrTab()); 1310 if (Error Err = initSections()) 1311 return std::move(Err); 1312 addData(SymTab); 1313 1314 return std::move(Obj); 1315 } 1316 1317 // Adds sections from IHEX data file. Data should have been 1318 // fully validated by this time. 1319 void IHexELFBuilder::addDataSections() { 1320 OwnedDataSection *Section = nullptr; 1321 uint64_t SegmentAddr = 0, BaseAddr = 0; 1322 uint32_t SecNo = 1; 1323 1324 for (const IHexRecord &R : Records) { 1325 uint64_t RecAddr; 1326 switch (R.Type) { 1327 case IHexRecord::Data: 1328 // Ignore empty data records 1329 if (R.HexData.empty()) 1330 continue; 1331 RecAddr = R.Addr + SegmentAddr + BaseAddr; 1332 if (!Section || Section->Addr + Section->Size != RecAddr) { 1333 // OriginalOffset field is only used to sort sections before layout, so 1334 // instead of keeping track of real offsets in IHEX file, and as 1335 // layoutSections() and layoutSectionsForOnlyKeepDebug() use 1336 // llvm::stable_sort(), we can just set it to a constant (zero). 1337 Section = &Obj->addSection<OwnedDataSection>( 1338 ".sec" + std::to_string(SecNo), RecAddr, 1339 ELF::SHF_ALLOC | ELF::SHF_WRITE, 0); 1340 SecNo++; 1341 } 1342 Section->appendHexData(R.HexData); 1343 break; 1344 case IHexRecord::EndOfFile: 1345 break; 1346 case IHexRecord::SegmentAddr: 1347 // 20-bit segment address. 1348 SegmentAddr = checkedGetHex<uint16_t>(R.HexData) << 4; 1349 break; 1350 case IHexRecord::StartAddr80x86: 1351 case IHexRecord::StartAddr: 1352 Obj->Entry = checkedGetHex<uint32_t>(R.HexData); 1353 assert(Obj->Entry <= 0xFFFFFU); 1354 break; 1355 case IHexRecord::ExtendedAddr: 1356 // 16-31 bits of linear base address 1357 BaseAddr = checkedGetHex<uint16_t>(R.HexData) << 16; 1358 break; 1359 default: 1360 llvm_unreachable("unknown record type"); 1361 } 1362 } 1363 } 1364 1365 Expected<std::unique_ptr<Object>> IHexELFBuilder::build() { 1366 initFileHeader(); 1367 initHeaderSegment(); 1368 StringTableSection *StrTab = addStrTab(); 1369 addSymTab(StrTab); 1370 if (Error Err = initSections()) 1371 return std::move(Err); 1372 addDataSections(); 1373 1374 return std::move(Obj); 1375 } 1376 1377 template <class ELFT> 1378 ELFBuilder<ELFT>::ELFBuilder(const ELFObjectFile<ELFT> &ElfObj, Object &Obj, 1379 Optional<StringRef> ExtractPartition) 1380 : ElfFile(ElfObj.getELFFile()), Obj(Obj), 1381 ExtractPartition(ExtractPartition) { 1382 Obj.IsMips64EL = ElfFile.isMips64EL(); 1383 } 1384 1385 template <class ELFT> void ELFBuilder<ELFT>::setParentSegment(Segment &Child) { 1386 for (Segment &Parent : Obj.segments()) { 1387 // Every segment will overlap with itself but we don't want a segment to 1388 // be its own parent so we avoid that situation. 1389 if (&Child != &Parent && segmentOverlapsSegment(Child, Parent)) { 1390 // We want a canonical "most parental" segment but this requires 1391 // inspecting the ParentSegment. 1392 if (compareSegmentsByOffset(&Parent, &Child)) 1393 if (Child.ParentSegment == nullptr || 1394 compareSegmentsByOffset(&Parent, Child.ParentSegment)) { 1395 Child.ParentSegment = &Parent; 1396 } 1397 } 1398 } 1399 } 1400 1401 template <class ELFT> Error ELFBuilder<ELFT>::findEhdrOffset() { 1402 if (!ExtractPartition) 1403 return Error::success(); 1404 1405 for (const SectionBase &Sec : Obj.sections()) { 1406 if (Sec.Type == SHT_LLVM_PART_EHDR && Sec.Name == *ExtractPartition) { 1407 EhdrOffset = Sec.Offset; 1408 return Error::success(); 1409 } 1410 } 1411 return createStringError(errc::invalid_argument, 1412 "could not find partition named '" + 1413 *ExtractPartition + "'"); 1414 } 1415 1416 template <class ELFT> 1417 Error ELFBuilder<ELFT>::readProgramHeaders(const ELFFile<ELFT> &HeadersFile) { 1418 uint32_t Index = 0; 1419 1420 Expected<typename ELFFile<ELFT>::Elf_Phdr_Range> Headers = 1421 HeadersFile.program_headers(); 1422 if (!Headers) 1423 return Headers.takeError(); 1424 1425 for (const typename ELFFile<ELFT>::Elf_Phdr &Phdr : *Headers) { 1426 if (Phdr.p_offset + Phdr.p_filesz > HeadersFile.getBufSize()) 1427 return createStringError( 1428 errc::invalid_argument, 1429 "program header with offset 0x" + Twine::utohexstr(Phdr.p_offset) + 1430 " and file size 0x" + Twine::utohexstr(Phdr.p_filesz) + 1431 " goes past the end of the file"); 1432 1433 ArrayRef<uint8_t> Data{HeadersFile.base() + Phdr.p_offset, 1434 (size_t)Phdr.p_filesz}; 1435 Segment &Seg = Obj.addSegment(Data); 1436 Seg.Type = Phdr.p_type; 1437 Seg.Flags = Phdr.p_flags; 1438 Seg.OriginalOffset = Phdr.p_offset + EhdrOffset; 1439 Seg.Offset = Phdr.p_offset + EhdrOffset; 1440 Seg.VAddr = Phdr.p_vaddr; 1441 Seg.PAddr = Phdr.p_paddr; 1442 Seg.FileSize = Phdr.p_filesz; 1443 Seg.MemSize = Phdr.p_memsz; 1444 Seg.Align = Phdr.p_align; 1445 Seg.Index = Index++; 1446 for (SectionBase &Sec : Obj.sections()) 1447 if (sectionWithinSegment(Sec, Seg)) { 1448 Seg.addSection(&Sec); 1449 if (!Sec.ParentSegment || Sec.ParentSegment->Offset > Seg.Offset) 1450 Sec.ParentSegment = &Seg; 1451 } 1452 } 1453 1454 auto &ElfHdr = Obj.ElfHdrSegment; 1455 ElfHdr.Index = Index++; 1456 ElfHdr.OriginalOffset = ElfHdr.Offset = EhdrOffset; 1457 1458 const typename ELFT::Ehdr &Ehdr = HeadersFile.getHeader(); 1459 auto &PrHdr = Obj.ProgramHdrSegment; 1460 PrHdr.Type = PT_PHDR; 1461 PrHdr.Flags = 0; 1462 // The spec requires us to have p_vaddr % p_align == p_offset % p_align. 1463 // Whereas this works automatically for ElfHdr, here OriginalOffset is 1464 // always non-zero and to ensure the equation we assign the same value to 1465 // VAddr as well. 1466 PrHdr.OriginalOffset = PrHdr.Offset = PrHdr.VAddr = EhdrOffset + Ehdr.e_phoff; 1467 PrHdr.PAddr = 0; 1468 PrHdr.FileSize = PrHdr.MemSize = Ehdr.e_phentsize * Ehdr.e_phnum; 1469 // The spec requires us to naturally align all the fields. 1470 PrHdr.Align = sizeof(Elf_Addr); 1471 PrHdr.Index = Index++; 1472 1473 // Now we do an O(n^2) loop through the segments in order to match up 1474 // segments. 1475 for (Segment &Child : Obj.segments()) 1476 setParentSegment(Child); 1477 setParentSegment(ElfHdr); 1478 setParentSegment(PrHdr); 1479 1480 return Error::success(); 1481 } 1482 1483 template <class ELFT> 1484 Error ELFBuilder<ELFT>::initGroupSection(GroupSection *GroupSec) { 1485 if (GroupSec->Align % sizeof(ELF::Elf32_Word) != 0) 1486 return createStringError(errc::invalid_argument, 1487 "invalid alignment " + Twine(GroupSec->Align) + 1488 " of group section '" + GroupSec->Name + "'"); 1489 SectionTableRef SecTable = Obj.sections(); 1490 if (GroupSec->Link != SHN_UNDEF) { 1491 auto SymTab = SecTable.template getSectionOfType<SymbolTableSection>( 1492 GroupSec->Link, 1493 "link field value '" + Twine(GroupSec->Link) + "' in section '" + 1494 GroupSec->Name + "' is invalid", 1495 "link field value '" + Twine(GroupSec->Link) + "' in section '" + 1496 GroupSec->Name + "' is not a symbol table"); 1497 if (!SymTab) 1498 return SymTab.takeError(); 1499 1500 Expected<Symbol *> Sym = (*SymTab)->getSymbolByIndex(GroupSec->Info); 1501 if (!Sym) 1502 return createStringError(errc::invalid_argument, 1503 "info field value '" + Twine(GroupSec->Info) + 1504 "' in section '" + GroupSec->Name + 1505 "' is not a valid symbol index"); 1506 GroupSec->setSymTab(*SymTab); 1507 GroupSec->setSymbol(*Sym); 1508 } 1509 if (GroupSec->Contents.size() % sizeof(ELF::Elf32_Word) || 1510 GroupSec->Contents.empty()) 1511 return createStringError(errc::invalid_argument, 1512 "the content of the section " + GroupSec->Name + 1513 " is malformed"); 1514 const ELF::Elf32_Word *Word = 1515 reinterpret_cast<const ELF::Elf32_Word *>(GroupSec->Contents.data()); 1516 const ELF::Elf32_Word *End = 1517 Word + GroupSec->Contents.size() / sizeof(ELF::Elf32_Word); 1518 GroupSec->setFlagWord( 1519 support::endian::read32<ELFT::TargetEndianness>(Word++)); 1520 for (; Word != End; ++Word) { 1521 uint32_t Index = support::endian::read32<ELFT::TargetEndianness>(Word); 1522 Expected<SectionBase *> Sec = SecTable.getSection( 1523 Index, "group member index " + Twine(Index) + " in section '" + 1524 GroupSec->Name + "' is invalid"); 1525 if (!Sec) 1526 return Sec.takeError(); 1527 1528 GroupSec->addMember(*Sec); 1529 } 1530 1531 return Error::success(); 1532 } 1533 1534 template <class ELFT> 1535 Error ELFBuilder<ELFT>::initSymbolTable(SymbolTableSection *SymTab) { 1536 Expected<const Elf_Shdr *> Shdr = ElfFile.getSection(SymTab->Index); 1537 if (!Shdr) 1538 return Shdr.takeError(); 1539 1540 Expected<StringRef> StrTabData = ElfFile.getStringTableForSymtab(**Shdr); 1541 if (!StrTabData) 1542 return StrTabData.takeError(); 1543 1544 ArrayRef<Elf_Word> ShndxData; 1545 1546 Expected<typename ELFFile<ELFT>::Elf_Sym_Range> Symbols = 1547 ElfFile.symbols(*Shdr); 1548 if (!Symbols) 1549 return Symbols.takeError(); 1550 1551 for (const typename ELFFile<ELFT>::Elf_Sym &Sym : *Symbols) { 1552 SectionBase *DefSection = nullptr; 1553 1554 Expected<StringRef> Name = Sym.getName(*StrTabData); 1555 if (!Name) 1556 return Name.takeError(); 1557 1558 if (Sym.st_shndx == SHN_XINDEX) { 1559 if (SymTab->getShndxTable() == nullptr) 1560 return createStringError(errc::invalid_argument, 1561 "symbol '" + *Name + 1562 "' has index SHN_XINDEX but no " 1563 "SHT_SYMTAB_SHNDX section exists"); 1564 if (ShndxData.data() == nullptr) { 1565 Expected<const Elf_Shdr *> ShndxSec = 1566 ElfFile.getSection(SymTab->getShndxTable()->Index); 1567 if (!ShndxSec) 1568 return ShndxSec.takeError(); 1569 1570 Expected<ArrayRef<Elf_Word>> Data = 1571 ElfFile.template getSectionContentsAsArray<Elf_Word>(**ShndxSec); 1572 if (!Data) 1573 return Data.takeError(); 1574 1575 ShndxData = *Data; 1576 if (ShndxData.size() != Symbols->size()) 1577 return createStringError( 1578 errc::invalid_argument, 1579 "symbol section index table does not have the same number of " 1580 "entries as the symbol table"); 1581 } 1582 Elf_Word Index = ShndxData[&Sym - Symbols->begin()]; 1583 Expected<SectionBase *> Sec = Obj.sections().getSection( 1584 Index, 1585 "symbol '" + *Name + "' has invalid section index " + Twine(Index)); 1586 if (!Sec) 1587 return Sec.takeError(); 1588 1589 DefSection = *Sec; 1590 } else if (Sym.st_shndx >= SHN_LORESERVE) { 1591 if (!isValidReservedSectionIndex(Sym.st_shndx, Obj.Machine)) { 1592 return createStringError( 1593 errc::invalid_argument, 1594 "symbol '" + *Name + 1595 "' has unsupported value greater than or equal " 1596 "to SHN_LORESERVE: " + 1597 Twine(Sym.st_shndx)); 1598 } 1599 } else if (Sym.st_shndx != SHN_UNDEF) { 1600 Expected<SectionBase *> Sec = Obj.sections().getSection( 1601 Sym.st_shndx, "symbol '" + *Name + 1602 "' is defined has invalid section index " + 1603 Twine(Sym.st_shndx)); 1604 if (!Sec) 1605 return Sec.takeError(); 1606 1607 DefSection = *Sec; 1608 } 1609 1610 SymTab->addSymbol(*Name, Sym.getBinding(), Sym.getType(), DefSection, 1611 Sym.getValue(), Sym.st_other, Sym.st_shndx, Sym.st_size); 1612 } 1613 1614 return Error::success(); 1615 } 1616 1617 template <class ELFT> 1618 static void getAddend(uint64_t &, const Elf_Rel_Impl<ELFT, false> &) {} 1619 1620 template <class ELFT> 1621 static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, true> &Rela) { 1622 ToSet = Rela.r_addend; 1623 } 1624 1625 template <class T> 1626 static Error initRelocations(RelocationSection *Relocs, T RelRange) { 1627 for (const auto &Rel : RelRange) { 1628 Relocation ToAdd; 1629 ToAdd.Offset = Rel.r_offset; 1630 getAddend(ToAdd.Addend, Rel); 1631 ToAdd.Type = Rel.getType(Relocs->getObject().IsMips64EL); 1632 1633 if (uint32_t Sym = Rel.getSymbol(Relocs->getObject().IsMips64EL)) { 1634 if (!Relocs->getObject().SymbolTable) 1635 return createStringError( 1636 errc::invalid_argument, 1637 "'" + Relocs->Name + "': relocation references symbol with index " + 1638 Twine(Sym) + ", but there is no symbol table"); 1639 Expected<Symbol *> SymByIndex = 1640 Relocs->getObject().SymbolTable->getSymbolByIndex(Sym); 1641 if (!SymByIndex) 1642 return SymByIndex.takeError(); 1643 1644 ToAdd.RelocSymbol = *SymByIndex; 1645 } 1646 1647 Relocs->addRelocation(ToAdd); 1648 } 1649 1650 return Error::success(); 1651 } 1652 1653 Expected<SectionBase *> SectionTableRef::getSection(uint32_t Index, 1654 Twine ErrMsg) { 1655 if (Index == SHN_UNDEF || Index > Sections.size()) 1656 return createStringError(errc::invalid_argument, ErrMsg); 1657 return Sections[Index - 1].get(); 1658 } 1659 1660 template <class T> 1661 Expected<T *> SectionTableRef::getSectionOfType(uint32_t Index, 1662 Twine IndexErrMsg, 1663 Twine TypeErrMsg) { 1664 Expected<SectionBase *> BaseSec = getSection(Index, IndexErrMsg); 1665 if (!BaseSec) 1666 return BaseSec.takeError(); 1667 1668 if (T *Sec = dyn_cast<T>(*BaseSec)) 1669 return Sec; 1670 1671 return createStringError(errc::invalid_argument, TypeErrMsg); 1672 } 1673 1674 template <class ELFT> 1675 Expected<SectionBase &> ELFBuilder<ELFT>::makeSection(const Elf_Shdr &Shdr) { 1676 switch (Shdr.sh_type) { 1677 case SHT_REL: 1678 case SHT_RELA: 1679 if (Shdr.sh_flags & SHF_ALLOC) { 1680 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1681 return Obj.addSection<DynamicRelocationSection>(*Data); 1682 else 1683 return Data.takeError(); 1684 } 1685 return Obj.addSection<RelocationSection>(Obj); 1686 case SHT_STRTAB: 1687 // If a string table is allocated we don't want to mess with it. That would 1688 // mean altering the memory image. There are no special link types or 1689 // anything so we can just use a Section. 1690 if (Shdr.sh_flags & SHF_ALLOC) { 1691 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1692 return Obj.addSection<Section>(*Data); 1693 else 1694 return Data.takeError(); 1695 } 1696 return Obj.addSection<StringTableSection>(); 1697 case SHT_HASH: 1698 case SHT_GNU_HASH: 1699 // Hash tables should refer to SHT_DYNSYM which we're not going to change. 1700 // Because of this we don't need to mess with the hash tables either. 1701 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1702 return Obj.addSection<Section>(*Data); 1703 else 1704 return Data.takeError(); 1705 case SHT_GROUP: 1706 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1707 return Obj.addSection<GroupSection>(*Data); 1708 else 1709 return Data.takeError(); 1710 case SHT_DYNSYM: 1711 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1712 return Obj.addSection<DynamicSymbolTableSection>(*Data); 1713 else 1714 return Data.takeError(); 1715 case SHT_DYNAMIC: 1716 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) 1717 return Obj.addSection<DynamicSection>(*Data); 1718 else 1719 return Data.takeError(); 1720 case SHT_SYMTAB: { 1721 auto &SymTab = Obj.addSection<SymbolTableSection>(); 1722 Obj.SymbolTable = &SymTab; 1723 return SymTab; 1724 } 1725 case SHT_SYMTAB_SHNDX: { 1726 auto &ShndxSection = Obj.addSection<SectionIndexSection>(); 1727 Obj.SectionIndexTable = &ShndxSection; 1728 return ShndxSection; 1729 } 1730 case SHT_NOBITS: 1731 return Obj.addSection<Section>(ArrayRef<uint8_t>()); 1732 default: { 1733 Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr); 1734 if (!Data) 1735 return Data.takeError(); 1736 1737 Expected<StringRef> Name = ElfFile.getSectionName(Shdr); 1738 if (!Name) 1739 return Name.takeError(); 1740 1741 if (Name->startswith(".zdebug") || (Shdr.sh_flags & ELF::SHF_COMPRESSED)) { 1742 uint64_t DecompressedSize, DecompressedAlign; 1743 std::tie(DecompressedSize, DecompressedAlign) = 1744 getDecompressedSizeAndAlignment<ELFT>(*Data); 1745 return Obj.addSection<CompressedSection>( 1746 CompressedSection(*Data, DecompressedSize, DecompressedAlign)); 1747 } 1748 1749 return Obj.addSection<Section>(*Data); 1750 } 1751 } 1752 } 1753 1754 template <class ELFT> Error ELFBuilder<ELFT>::readSectionHeaders() { 1755 uint32_t Index = 0; 1756 Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections = 1757 ElfFile.sections(); 1758 if (!Sections) 1759 return Sections.takeError(); 1760 1761 for (const typename ELFFile<ELFT>::Elf_Shdr &Shdr : *Sections) { 1762 if (Index == 0) { 1763 ++Index; 1764 continue; 1765 } 1766 Expected<SectionBase &> Sec = makeSection(Shdr); 1767 if (!Sec) 1768 return Sec.takeError(); 1769 1770 Expected<StringRef> SecName = ElfFile.getSectionName(Shdr); 1771 if (!SecName) 1772 return SecName.takeError(); 1773 Sec->Name = SecName->str(); 1774 Sec->Type = Sec->OriginalType = Shdr.sh_type; 1775 Sec->Flags = Sec->OriginalFlags = Shdr.sh_flags; 1776 Sec->Addr = Shdr.sh_addr; 1777 Sec->Offset = Shdr.sh_offset; 1778 Sec->OriginalOffset = Shdr.sh_offset; 1779 Sec->Size = Shdr.sh_size; 1780 Sec->Link = Shdr.sh_link; 1781 Sec->Info = Shdr.sh_info; 1782 Sec->Align = Shdr.sh_addralign; 1783 Sec->EntrySize = Shdr.sh_entsize; 1784 Sec->Index = Index++; 1785 Sec->OriginalIndex = Sec->Index; 1786 Sec->OriginalData = ArrayRef<uint8_t>( 1787 ElfFile.base() + Shdr.sh_offset, 1788 (Shdr.sh_type == SHT_NOBITS) ? (size_t)0 : Shdr.sh_size); 1789 } 1790 1791 return Error::success(); 1792 } 1793 1794 template <class ELFT> Error ELFBuilder<ELFT>::readSections(bool EnsureSymtab) { 1795 uint32_t ShstrIndex = ElfFile.getHeader().e_shstrndx; 1796 if (ShstrIndex == SHN_XINDEX) { 1797 Expected<const Elf_Shdr *> Sec = ElfFile.getSection(0); 1798 if (!Sec) 1799 return Sec.takeError(); 1800 1801 ShstrIndex = (*Sec)->sh_link; 1802 } 1803 1804 if (ShstrIndex == SHN_UNDEF) 1805 Obj.HadShdrs = false; 1806 else { 1807 Expected<StringTableSection *> Sec = 1808 Obj.sections().template getSectionOfType<StringTableSection>( 1809 ShstrIndex, 1810 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " + 1811 " is invalid", 1812 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " + 1813 " does not reference a string table"); 1814 if (!Sec) 1815 return Sec.takeError(); 1816 1817 Obj.SectionNames = *Sec; 1818 } 1819 1820 // If a section index table exists we'll need to initialize it before we 1821 // initialize the symbol table because the symbol table might need to 1822 // reference it. 1823 if (Obj.SectionIndexTable) 1824 if (Error Err = Obj.SectionIndexTable->initialize(Obj.sections())) 1825 return Err; 1826 1827 // Now that all of the sections have been added we can fill out some extra 1828 // details about symbol tables. We need the symbol table filled out before 1829 // any relocations. 1830 if (Obj.SymbolTable) { 1831 if (Error Err = Obj.SymbolTable->initialize(Obj.sections())) 1832 return Err; 1833 if (Error Err = initSymbolTable(Obj.SymbolTable)) 1834 return Err; 1835 } else if (EnsureSymtab) { 1836 if (Error Err = Obj.addNewSymbolTable()) 1837 return Err; 1838 } 1839 1840 // Now that all sections and symbols have been added we can add 1841 // relocations that reference symbols and set the link and info fields for 1842 // relocation sections. 1843 for (SectionBase &Sec : Obj.sections()) { 1844 if (&Sec == Obj.SymbolTable) 1845 continue; 1846 if (Error Err = Sec.initialize(Obj.sections())) 1847 return Err; 1848 if (auto RelSec = dyn_cast<RelocationSection>(&Sec)) { 1849 Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections = 1850 ElfFile.sections(); 1851 if (!Sections) 1852 return Sections.takeError(); 1853 1854 const typename ELFFile<ELFT>::Elf_Shdr *Shdr = 1855 Sections->begin() + RelSec->Index; 1856 if (RelSec->Type == SHT_REL) { 1857 Expected<typename ELFFile<ELFT>::Elf_Rel_Range> Rels = 1858 ElfFile.rels(*Shdr); 1859 if (!Rels) 1860 return Rels.takeError(); 1861 1862 if (Error Err = initRelocations(RelSec, *Rels)) 1863 return Err; 1864 } else { 1865 Expected<typename ELFFile<ELFT>::Elf_Rela_Range> Relas = 1866 ElfFile.relas(*Shdr); 1867 if (!Relas) 1868 return Relas.takeError(); 1869 1870 if (Error Err = initRelocations(RelSec, *Relas)) 1871 return Err; 1872 } 1873 } else if (auto GroupSec = dyn_cast<GroupSection>(&Sec)) { 1874 if (Error Err = initGroupSection(GroupSec)) 1875 return Err; 1876 } 1877 } 1878 1879 return Error::success(); 1880 } 1881 1882 template <class ELFT> Error ELFBuilder<ELFT>::build(bool EnsureSymtab) { 1883 if (Error E = readSectionHeaders()) 1884 return E; 1885 if (Error E = findEhdrOffset()) 1886 return E; 1887 1888 // The ELFFile whose ELF headers and program headers are copied into the 1889 // output file. Normally the same as ElfFile, but if we're extracting a 1890 // loadable partition it will point to the partition's headers. 1891 Expected<ELFFile<ELFT>> HeadersFile = ELFFile<ELFT>::create(toStringRef( 1892 {ElfFile.base() + EhdrOffset, ElfFile.getBufSize() - EhdrOffset})); 1893 if (!HeadersFile) 1894 return HeadersFile.takeError(); 1895 1896 const typename ELFFile<ELFT>::Elf_Ehdr &Ehdr = HeadersFile->getHeader(); 1897 Obj.OSABI = Ehdr.e_ident[EI_OSABI]; 1898 Obj.ABIVersion = Ehdr.e_ident[EI_ABIVERSION]; 1899 Obj.Type = Ehdr.e_type; 1900 Obj.Machine = Ehdr.e_machine; 1901 Obj.Version = Ehdr.e_version; 1902 Obj.Entry = Ehdr.e_entry; 1903 Obj.Flags = Ehdr.e_flags; 1904 1905 if (Error E = readSections(EnsureSymtab)) 1906 return E; 1907 return readProgramHeaders(*HeadersFile); 1908 } 1909 1910 Writer::~Writer() = default; 1911 1912 Reader::~Reader() = default; 1913 1914 Expected<std::unique_ptr<Object>> 1915 BinaryReader::create(bool /*EnsureSymtab*/) const { 1916 return BinaryELFBuilder(MemBuf, NewSymbolVisibility).build(); 1917 } 1918 1919 Expected<std::vector<IHexRecord>> IHexReader::parse() const { 1920 SmallVector<StringRef, 16> Lines; 1921 std::vector<IHexRecord> Records; 1922 bool HasSections = false; 1923 1924 MemBuf->getBuffer().split(Lines, '\n'); 1925 Records.reserve(Lines.size()); 1926 for (size_t LineNo = 1; LineNo <= Lines.size(); ++LineNo) { 1927 StringRef Line = Lines[LineNo - 1].trim(); 1928 if (Line.empty()) 1929 continue; 1930 1931 Expected<IHexRecord> R = IHexRecord::parse(Line); 1932 if (!R) 1933 return parseError(LineNo, R.takeError()); 1934 if (R->Type == IHexRecord::EndOfFile) 1935 break; 1936 HasSections |= (R->Type == IHexRecord::Data); 1937 Records.push_back(*R); 1938 } 1939 if (!HasSections) 1940 return parseError(-1U, "no sections"); 1941 1942 return std::move(Records); 1943 } 1944 1945 Expected<std::unique_ptr<Object>> 1946 IHexReader::create(bool /*EnsureSymtab*/) const { 1947 Expected<std::vector<IHexRecord>> Records = parse(); 1948 if (!Records) 1949 return Records.takeError(); 1950 1951 return IHexELFBuilder(*Records).build(); 1952 } 1953 1954 Expected<std::unique_ptr<Object>> ELFReader::create(bool EnsureSymtab) const { 1955 auto Obj = std::make_unique<Object>(); 1956 if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Bin)) { 1957 ELFBuilder<ELF32LE> Builder(*O, *Obj, ExtractPartition); 1958 if (Error Err = Builder.build(EnsureSymtab)) 1959 return std::move(Err); 1960 return std::move(Obj); 1961 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Bin)) { 1962 ELFBuilder<ELF64LE> Builder(*O, *Obj, ExtractPartition); 1963 if (Error Err = Builder.build(EnsureSymtab)) 1964 return std::move(Err); 1965 return std::move(Obj); 1966 } else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Bin)) { 1967 ELFBuilder<ELF32BE> Builder(*O, *Obj, ExtractPartition); 1968 if (Error Err = Builder.build(EnsureSymtab)) 1969 return std::move(Err); 1970 return std::move(Obj); 1971 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Bin)) { 1972 ELFBuilder<ELF64BE> Builder(*O, *Obj, ExtractPartition); 1973 if (Error Err = Builder.build(EnsureSymtab)) 1974 return std::move(Err); 1975 return std::move(Obj); 1976 } 1977 return createStringError(errc::invalid_argument, "invalid file type"); 1978 } 1979 1980 template <class ELFT> void ELFWriter<ELFT>::writeEhdr() { 1981 Elf_Ehdr &Ehdr = *reinterpret_cast<Elf_Ehdr *>(Buf->getBufferStart()); 1982 std::fill(Ehdr.e_ident, Ehdr.e_ident + 16, 0); 1983 Ehdr.e_ident[EI_MAG0] = 0x7f; 1984 Ehdr.e_ident[EI_MAG1] = 'E'; 1985 Ehdr.e_ident[EI_MAG2] = 'L'; 1986 Ehdr.e_ident[EI_MAG3] = 'F'; 1987 Ehdr.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32; 1988 Ehdr.e_ident[EI_DATA] = 1989 ELFT::TargetEndianness == support::big ? ELFDATA2MSB : ELFDATA2LSB; 1990 Ehdr.e_ident[EI_VERSION] = EV_CURRENT; 1991 Ehdr.e_ident[EI_OSABI] = Obj.OSABI; 1992 Ehdr.e_ident[EI_ABIVERSION] = Obj.ABIVersion; 1993 1994 Ehdr.e_type = Obj.Type; 1995 Ehdr.e_machine = Obj.Machine; 1996 Ehdr.e_version = Obj.Version; 1997 Ehdr.e_entry = Obj.Entry; 1998 // We have to use the fully-qualified name llvm::size 1999 // since some compilers complain on ambiguous resolution. 2000 Ehdr.e_phnum = llvm::size(Obj.segments()); 2001 Ehdr.e_phoff = (Ehdr.e_phnum != 0) ? Obj.ProgramHdrSegment.Offset : 0; 2002 Ehdr.e_phentsize = (Ehdr.e_phnum != 0) ? sizeof(Elf_Phdr) : 0; 2003 Ehdr.e_flags = Obj.Flags; 2004 Ehdr.e_ehsize = sizeof(Elf_Ehdr); 2005 if (WriteSectionHeaders && Obj.sections().size() != 0) { 2006 Ehdr.e_shentsize = sizeof(Elf_Shdr); 2007 Ehdr.e_shoff = Obj.SHOff; 2008 // """ 2009 // If the number of sections is greater than or equal to 2010 // SHN_LORESERVE (0xff00), this member has the value zero and the actual 2011 // number of section header table entries is contained in the sh_size field 2012 // of the section header at index 0. 2013 // """ 2014 auto Shnum = Obj.sections().size() + 1; 2015 if (Shnum >= SHN_LORESERVE) 2016 Ehdr.e_shnum = 0; 2017 else 2018 Ehdr.e_shnum = Shnum; 2019 // """ 2020 // If the section name string table section index is greater than or equal 2021 // to SHN_LORESERVE (0xff00), this member has the value SHN_XINDEX (0xffff) 2022 // and the actual index of the section name string table section is 2023 // contained in the sh_link field of the section header at index 0. 2024 // """ 2025 if (Obj.SectionNames->Index >= SHN_LORESERVE) 2026 Ehdr.e_shstrndx = SHN_XINDEX; 2027 else 2028 Ehdr.e_shstrndx = Obj.SectionNames->Index; 2029 } else { 2030 Ehdr.e_shentsize = 0; 2031 Ehdr.e_shoff = 0; 2032 Ehdr.e_shnum = 0; 2033 Ehdr.e_shstrndx = 0; 2034 } 2035 } 2036 2037 template <class ELFT> void ELFWriter<ELFT>::writePhdrs() { 2038 for (auto &Seg : Obj.segments()) 2039 writePhdr(Seg); 2040 } 2041 2042 template <class ELFT> void ELFWriter<ELFT>::writeShdrs() { 2043 // This reference serves to write the dummy section header at the begining 2044 // of the file. It is not used for anything else 2045 Elf_Shdr &Shdr = 2046 *reinterpret_cast<Elf_Shdr *>(Buf->getBufferStart() + Obj.SHOff); 2047 Shdr.sh_name = 0; 2048 Shdr.sh_type = SHT_NULL; 2049 Shdr.sh_flags = 0; 2050 Shdr.sh_addr = 0; 2051 Shdr.sh_offset = 0; 2052 // See writeEhdr for why we do this. 2053 uint64_t Shnum = Obj.sections().size() + 1; 2054 if (Shnum >= SHN_LORESERVE) 2055 Shdr.sh_size = Shnum; 2056 else 2057 Shdr.sh_size = 0; 2058 // See writeEhdr for why we do this. 2059 if (Obj.SectionNames != nullptr && Obj.SectionNames->Index >= SHN_LORESERVE) 2060 Shdr.sh_link = Obj.SectionNames->Index; 2061 else 2062 Shdr.sh_link = 0; 2063 Shdr.sh_info = 0; 2064 Shdr.sh_addralign = 0; 2065 Shdr.sh_entsize = 0; 2066 2067 for (SectionBase &Sec : Obj.sections()) 2068 writeShdr(Sec); 2069 } 2070 2071 template <class ELFT> Error ELFWriter<ELFT>::writeSectionData() { 2072 for (SectionBase &Sec : Obj.sections()) 2073 // Segments are responsible for writing their contents, so only write the 2074 // section data if the section is not in a segment. Note that this renders 2075 // sections in segments effectively immutable. 2076 if (Sec.ParentSegment == nullptr) 2077 if (Error Err = Sec.accept(*SecWriter)) 2078 return Err; 2079 2080 return Error::success(); 2081 } 2082 2083 template <class ELFT> void ELFWriter<ELFT>::writeSegmentData() { 2084 for (Segment &Seg : Obj.segments()) { 2085 size_t Size = std::min<size_t>(Seg.FileSize, Seg.getContents().size()); 2086 std::memcpy(Buf->getBufferStart() + Seg.Offset, Seg.getContents().data(), 2087 Size); 2088 } 2089 2090 for (auto it : Obj.getUpdatedSections()) { 2091 SectionBase *Sec = it.first; 2092 ArrayRef<uint8_t> Data = it.second; 2093 2094 auto *Parent = Sec->ParentSegment; 2095 assert(Parent && "This section should've been part of a segment."); 2096 uint64_t Offset = 2097 Sec->OriginalOffset - Parent->OriginalOffset + Parent->Offset; 2098 llvm::copy(Data, Buf->getBufferStart() + Offset); 2099 } 2100 2101 // Iterate over removed sections and overwrite their old data with zeroes. 2102 for (auto &Sec : Obj.removedSections()) { 2103 Segment *Parent = Sec.ParentSegment; 2104 if (Parent == nullptr || Sec.Type == SHT_NOBITS || Sec.Size == 0) 2105 continue; 2106 uint64_t Offset = 2107 Sec.OriginalOffset - Parent->OriginalOffset + Parent->Offset; 2108 std::memset(Buf->getBufferStart() + Offset, 0, Sec.Size); 2109 } 2110 } 2111 2112 template <class ELFT> 2113 ELFWriter<ELFT>::ELFWriter(Object &Obj, raw_ostream &Buf, bool WSH, 2114 bool OnlyKeepDebug) 2115 : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs), 2116 OnlyKeepDebug(OnlyKeepDebug) {} 2117 2118 Error Object::updateSection(StringRef Name, ArrayRef<uint8_t> Data) { 2119 auto It = llvm::find_if(Sections, 2120 [&](const SecPtr &Sec) { return Sec->Name == Name; }); 2121 if (It == Sections.end()) 2122 return createStringError(errc::invalid_argument, "section '%s' not found", 2123 Name.str().c_str()); 2124 2125 auto *OldSec = It->get(); 2126 if (!OldSec->hasContents()) 2127 return createStringError( 2128 errc::invalid_argument, 2129 "section '%s' cannot be updated because it does not have contents", 2130 Name.str().c_str()); 2131 2132 if (Data.size() > OldSec->Size && OldSec->ParentSegment) 2133 return createStringError(errc::invalid_argument, 2134 "cannot fit data of size %zu into section '%s' " 2135 "with size %zu that is part of a segment", 2136 Data.size(), Name.str().c_str(), OldSec->Size); 2137 2138 if (!OldSec->ParentSegment) { 2139 *It = std::make_unique<OwnedDataSection>(*OldSec, Data); 2140 } else { 2141 // The segment writer will be in charge of updating these contents. 2142 OldSec->Size = Data.size(); 2143 UpdatedSections[OldSec] = Data; 2144 } 2145 2146 return Error::success(); 2147 } 2148 2149 Error Object::removeSections( 2150 bool AllowBrokenLinks, std::function<bool(const SectionBase &)> ToRemove) { 2151 2152 auto Iter = std::stable_partition( 2153 std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) { 2154 if (ToRemove(*Sec)) 2155 return false; 2156 if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) { 2157 if (auto ToRelSec = RelSec->getSection()) 2158 return !ToRemove(*ToRelSec); 2159 } 2160 return true; 2161 }); 2162 if (SymbolTable != nullptr && ToRemove(*SymbolTable)) 2163 SymbolTable = nullptr; 2164 if (SectionNames != nullptr && ToRemove(*SectionNames)) 2165 SectionNames = nullptr; 2166 if (SectionIndexTable != nullptr && ToRemove(*SectionIndexTable)) 2167 SectionIndexTable = nullptr; 2168 // Now make sure there are no remaining references to the sections that will 2169 // be removed. Sometimes it is impossible to remove a reference so we emit 2170 // an error here instead. 2171 std::unordered_set<const SectionBase *> RemoveSections; 2172 RemoveSections.reserve(std::distance(Iter, std::end(Sections))); 2173 for (auto &RemoveSec : make_range(Iter, std::end(Sections))) { 2174 for (auto &Segment : Segments) 2175 Segment->removeSection(RemoveSec.get()); 2176 RemoveSec->onRemove(); 2177 RemoveSections.insert(RemoveSec.get()); 2178 } 2179 2180 // For each section that remains alive, we want to remove the dead references. 2181 // This either might update the content of the section (e.g. remove symbols 2182 // from symbol table that belongs to removed section) or trigger an error if 2183 // a live section critically depends on a section being removed somehow 2184 // (e.g. the removed section is referenced by a relocation). 2185 for (auto &KeepSec : make_range(std::begin(Sections), Iter)) { 2186 if (Error E = KeepSec->removeSectionReferences( 2187 AllowBrokenLinks, [&RemoveSections](const SectionBase *Sec) { 2188 return RemoveSections.find(Sec) != RemoveSections.end(); 2189 })) 2190 return E; 2191 } 2192 2193 // Transfer removed sections into the Object RemovedSections container for use 2194 // later. 2195 std::move(Iter, Sections.end(), std::back_inserter(RemovedSections)); 2196 // Now finally get rid of them all together. 2197 Sections.erase(Iter, std::end(Sections)); 2198 return Error::success(); 2199 } 2200 2201 Error Object::replaceSections( 2202 const DenseMap<SectionBase *, SectionBase *> &FromTo) { 2203 auto SectionIndexLess = [](const SecPtr &Lhs, const SecPtr &Rhs) { 2204 return Lhs->Index < Rhs->Index; 2205 }; 2206 assert(llvm::is_sorted(Sections, SectionIndexLess) && 2207 "Sections are expected to be sorted by Index"); 2208 // Set indices of new sections so that they can be later sorted into positions 2209 // of removed ones. 2210 for (auto &I : FromTo) 2211 I.second->Index = I.first->Index; 2212 2213 // Notify all sections about the replacement. 2214 for (auto &Sec : Sections) 2215 Sec->replaceSectionReferences(FromTo); 2216 2217 if (Error E = removeSections( 2218 /*AllowBrokenLinks=*/false, 2219 [=](const SectionBase &Sec) { return FromTo.count(&Sec) > 0; })) 2220 return E; 2221 llvm::sort(Sections, SectionIndexLess); 2222 return Error::success(); 2223 } 2224 2225 Error Object::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) { 2226 if (SymbolTable) 2227 for (const SecPtr &Sec : Sections) 2228 if (Error E = Sec->removeSymbols(ToRemove)) 2229 return E; 2230 return Error::success(); 2231 } 2232 2233 Error Object::addNewSymbolTable() { 2234 assert(!SymbolTable && "Object must not has a SymbolTable."); 2235 2236 // Reuse an existing SHT_STRTAB section if it exists. 2237 StringTableSection *StrTab = nullptr; 2238 for (SectionBase &Sec : sections()) { 2239 if (Sec.Type == ELF::SHT_STRTAB && !(Sec.Flags & SHF_ALLOC)) { 2240 StrTab = static_cast<StringTableSection *>(&Sec); 2241 2242 // Prefer a string table that is not the section header string table, if 2243 // such a table exists. 2244 if (SectionNames != &Sec) 2245 break; 2246 } 2247 } 2248 if (!StrTab) 2249 StrTab = &addSection<StringTableSection>(); 2250 2251 SymbolTableSection &SymTab = addSection<SymbolTableSection>(); 2252 SymTab.Name = ".symtab"; 2253 SymTab.Link = StrTab->Index; 2254 if (Error Err = SymTab.initialize(sections())) 2255 return Err; 2256 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0); 2257 2258 SymbolTable = &SymTab; 2259 2260 return Error::success(); 2261 } 2262 2263 // Orders segments such that if x = y->ParentSegment then y comes before x. 2264 static void orderSegments(std::vector<Segment *> &Segments) { 2265 llvm::stable_sort(Segments, compareSegmentsByOffset); 2266 } 2267 2268 // This function finds a consistent layout for a list of segments starting from 2269 // an Offset. It assumes that Segments have been sorted by orderSegments and 2270 // returns an Offset one past the end of the last segment. 2271 static uint64_t layoutSegments(std::vector<Segment *> &Segments, 2272 uint64_t Offset) { 2273 assert(llvm::is_sorted(Segments, compareSegmentsByOffset)); 2274 // The only way a segment should move is if a section was between two 2275 // segments and that section was removed. If that section isn't in a segment 2276 // then it's acceptable, but not ideal, to simply move it to after the 2277 // segments. So we can simply layout segments one after the other accounting 2278 // for alignment. 2279 for (Segment *Seg : Segments) { 2280 // We assume that segments have been ordered by OriginalOffset and Index 2281 // such that a parent segment will always come before a child segment in 2282 // OrderedSegments. This means that the Offset of the ParentSegment should 2283 // already be set and we can set our offset relative to it. 2284 if (Seg->ParentSegment != nullptr) { 2285 Segment *Parent = Seg->ParentSegment; 2286 Seg->Offset = 2287 Parent->Offset + Seg->OriginalOffset - Parent->OriginalOffset; 2288 } else { 2289 Seg->Offset = 2290 alignTo(Offset, std::max<uint64_t>(Seg->Align, 1), Seg->VAddr); 2291 } 2292 Offset = std::max(Offset, Seg->Offset + Seg->FileSize); 2293 } 2294 return Offset; 2295 } 2296 2297 // This function finds a consistent layout for a list of sections. It assumes 2298 // that the ->ParentSegment of each section has already been laid out. The 2299 // supplied starting Offset is used for the starting offset of any section that 2300 // does not have a ParentSegment. It returns either the offset given if all 2301 // sections had a ParentSegment or an offset one past the last section if there 2302 // was a section that didn't have a ParentSegment. 2303 template <class Range> 2304 static uint64_t layoutSections(Range Sections, uint64_t Offset) { 2305 // Now the offset of every segment has been set we can assign the offsets 2306 // of each section. For sections that are covered by a segment we should use 2307 // the segment's original offset and the section's original offset to compute 2308 // the offset from the start of the segment. Using the offset from the start 2309 // of the segment we can assign a new offset to the section. For sections not 2310 // covered by segments we can just bump Offset to the next valid location. 2311 // While it is not necessary, layout the sections in the order based on their 2312 // original offsets to resemble the input file as close as possible. 2313 std::vector<SectionBase *> OutOfSegmentSections; 2314 uint32_t Index = 1; 2315 for (auto &Sec : Sections) { 2316 Sec.Index = Index++; 2317 if (Sec.ParentSegment != nullptr) { 2318 auto Segment = *Sec.ParentSegment; 2319 Sec.Offset = 2320 Segment.Offset + (Sec.OriginalOffset - Segment.OriginalOffset); 2321 } else 2322 OutOfSegmentSections.push_back(&Sec); 2323 } 2324 2325 llvm::stable_sort(OutOfSegmentSections, 2326 [](const SectionBase *Lhs, const SectionBase *Rhs) { 2327 return Lhs->OriginalOffset < Rhs->OriginalOffset; 2328 }); 2329 for (auto *Sec : OutOfSegmentSections) { 2330 Offset = alignTo(Offset, Sec->Align == 0 ? 1 : Sec->Align); 2331 Sec->Offset = Offset; 2332 if (Sec->Type != SHT_NOBITS) 2333 Offset += Sec->Size; 2334 } 2335 return Offset; 2336 } 2337 2338 // Rewrite sh_offset after some sections are changed to SHT_NOBITS and thus 2339 // occupy no space in the file. 2340 static uint64_t layoutSectionsForOnlyKeepDebug(Object &Obj, uint64_t Off) { 2341 // The layout algorithm requires the sections to be handled in the order of 2342 // their offsets in the input file, at least inside segments. 2343 std::vector<SectionBase *> Sections; 2344 Sections.reserve(Obj.sections().size()); 2345 uint32_t Index = 1; 2346 for (auto &Sec : Obj.sections()) { 2347 Sec.Index = Index++; 2348 Sections.push_back(&Sec); 2349 } 2350 llvm::stable_sort(Sections, 2351 [](const SectionBase *Lhs, const SectionBase *Rhs) { 2352 return Lhs->OriginalOffset < Rhs->OriginalOffset; 2353 }); 2354 2355 for (auto *Sec : Sections) { 2356 auto *FirstSec = Sec->ParentSegment && Sec->ParentSegment->Type == PT_LOAD 2357 ? Sec->ParentSegment->firstSection() 2358 : nullptr; 2359 2360 // The first section in a PT_LOAD has to have congruent offset and address 2361 // modulo the alignment, which usually equals the maximum page size. 2362 if (FirstSec && FirstSec == Sec) 2363 Off = alignTo(Off, Sec->ParentSegment->Align, Sec->Addr); 2364 2365 // sh_offset is not significant for SHT_NOBITS sections, but the congruence 2366 // rule must be followed if it is the first section in a PT_LOAD. Do not 2367 // advance Off. 2368 if (Sec->Type == SHT_NOBITS) { 2369 Sec->Offset = Off; 2370 continue; 2371 } 2372 2373 if (!FirstSec) { 2374 // FirstSec being nullptr generally means that Sec does not have the 2375 // SHF_ALLOC flag. 2376 Off = Sec->Align ? alignTo(Off, Sec->Align) : Off; 2377 } else if (FirstSec != Sec) { 2378 // The offset is relative to the first section in the PT_LOAD segment. Use 2379 // sh_offset for non-SHF_ALLOC sections. 2380 Off = Sec->OriginalOffset - FirstSec->OriginalOffset + FirstSec->Offset; 2381 } 2382 Sec->Offset = Off; 2383 Off += Sec->Size; 2384 } 2385 return Off; 2386 } 2387 2388 // Rewrite p_offset and p_filesz of non-PT_PHDR segments after sh_offset values 2389 // have been updated. 2390 static uint64_t layoutSegmentsForOnlyKeepDebug(std::vector<Segment *> &Segments, 2391 uint64_t HdrEnd) { 2392 uint64_t MaxOffset = 0; 2393 for (Segment *Seg : Segments) { 2394 if (Seg->Type == PT_PHDR) 2395 continue; 2396 2397 // The segment offset is generally the offset of the first section. 2398 // 2399 // For a segment containing no section (see sectionWithinSegment), if it has 2400 // a parent segment, copy the parent segment's offset field. This works for 2401 // empty PT_TLS. If no parent segment, use 0: the segment is not useful for 2402 // debugging anyway. 2403 const SectionBase *FirstSec = Seg->firstSection(); 2404 uint64_t Offset = 2405 FirstSec ? FirstSec->Offset 2406 : (Seg->ParentSegment ? Seg->ParentSegment->Offset : 0); 2407 uint64_t FileSize = 0; 2408 for (const SectionBase *Sec : Seg->Sections) { 2409 uint64_t Size = Sec->Type == SHT_NOBITS ? 0 : Sec->Size; 2410 if (Sec->Offset + Size > Offset) 2411 FileSize = std::max(FileSize, Sec->Offset + Size - Offset); 2412 } 2413 2414 // If the segment includes EHDR and program headers, don't make it smaller 2415 // than the headers. 2416 if (Seg->Offset < HdrEnd && HdrEnd <= Seg->Offset + Seg->FileSize) { 2417 FileSize += Offset - Seg->Offset; 2418 Offset = Seg->Offset; 2419 FileSize = std::max(FileSize, HdrEnd - Offset); 2420 } 2421 2422 Seg->Offset = Offset; 2423 Seg->FileSize = FileSize; 2424 MaxOffset = std::max(MaxOffset, Offset + FileSize); 2425 } 2426 return MaxOffset; 2427 } 2428 2429 template <class ELFT> void ELFWriter<ELFT>::initEhdrSegment() { 2430 Segment &ElfHdr = Obj.ElfHdrSegment; 2431 ElfHdr.Type = PT_PHDR; 2432 ElfHdr.Flags = 0; 2433 ElfHdr.VAddr = 0; 2434 ElfHdr.PAddr = 0; 2435 ElfHdr.FileSize = ElfHdr.MemSize = sizeof(Elf_Ehdr); 2436 ElfHdr.Align = 0; 2437 } 2438 2439 template <class ELFT> void ELFWriter<ELFT>::assignOffsets() { 2440 // We need a temporary list of segments that has a special order to it 2441 // so that we know that anytime ->ParentSegment is set that segment has 2442 // already had its offset properly set. 2443 std::vector<Segment *> OrderedSegments; 2444 for (Segment &Segment : Obj.segments()) 2445 OrderedSegments.push_back(&Segment); 2446 OrderedSegments.push_back(&Obj.ElfHdrSegment); 2447 OrderedSegments.push_back(&Obj.ProgramHdrSegment); 2448 orderSegments(OrderedSegments); 2449 2450 uint64_t Offset; 2451 if (OnlyKeepDebug) { 2452 // For --only-keep-debug, the sections that did not preserve contents were 2453 // changed to SHT_NOBITS. We now rewrite sh_offset fields of sections, and 2454 // then rewrite p_offset/p_filesz of program headers. 2455 uint64_t HdrEnd = 2456 sizeof(Elf_Ehdr) + llvm::size(Obj.segments()) * sizeof(Elf_Phdr); 2457 Offset = layoutSectionsForOnlyKeepDebug(Obj, HdrEnd); 2458 Offset = std::max(Offset, 2459 layoutSegmentsForOnlyKeepDebug(OrderedSegments, HdrEnd)); 2460 } else { 2461 // Offset is used as the start offset of the first segment to be laid out. 2462 // Since the ELF Header (ElfHdrSegment) must be at the start of the file, 2463 // we start at offset 0. 2464 Offset = layoutSegments(OrderedSegments, 0); 2465 Offset = layoutSections(Obj.sections(), Offset); 2466 } 2467 // If we need to write the section header table out then we need to align the 2468 // Offset so that SHOffset is valid. 2469 if (WriteSectionHeaders) 2470 Offset = alignTo(Offset, sizeof(Elf_Addr)); 2471 Obj.SHOff = Offset; 2472 } 2473 2474 template <class ELFT> size_t ELFWriter<ELFT>::totalSize() const { 2475 // We already have the section header offset so we can calculate the total 2476 // size by just adding up the size of each section header. 2477 if (!WriteSectionHeaders) 2478 return Obj.SHOff; 2479 size_t ShdrCount = Obj.sections().size() + 1; // Includes null shdr. 2480 return Obj.SHOff + ShdrCount * sizeof(Elf_Shdr); 2481 } 2482 2483 template <class ELFT> Error ELFWriter<ELFT>::write() { 2484 // Segment data must be written first, so that the ELF header and program 2485 // header tables can overwrite it, if covered by a segment. 2486 writeSegmentData(); 2487 writeEhdr(); 2488 writePhdrs(); 2489 if (Error E = writeSectionData()) 2490 return E; 2491 if (WriteSectionHeaders) 2492 writeShdrs(); 2493 2494 // TODO: Implement direct writing to the output stream (without intermediate 2495 // memory buffer Buf). 2496 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2497 return Error::success(); 2498 } 2499 2500 static Error removeUnneededSections(Object &Obj) { 2501 // We can remove an empty symbol table from non-relocatable objects. 2502 // Relocatable objects typically have relocation sections whose 2503 // sh_link field points to .symtab, so we can't remove .symtab 2504 // even if it is empty. 2505 if (Obj.isRelocatable() || Obj.SymbolTable == nullptr || 2506 !Obj.SymbolTable->empty()) 2507 return Error::success(); 2508 2509 // .strtab can be used for section names. In such a case we shouldn't 2510 // remove it. 2511 auto *StrTab = Obj.SymbolTable->getStrTab() == Obj.SectionNames 2512 ? nullptr 2513 : Obj.SymbolTable->getStrTab(); 2514 return Obj.removeSections(false, [&](const SectionBase &Sec) { 2515 return &Sec == Obj.SymbolTable || &Sec == StrTab; 2516 }); 2517 } 2518 2519 template <class ELFT> Error ELFWriter<ELFT>::finalize() { 2520 // It could happen that SectionNames has been removed and yet the user wants 2521 // a section header table output. We need to throw an error if a user tries 2522 // to do that. 2523 if (Obj.SectionNames == nullptr && WriteSectionHeaders) 2524 return createStringError(llvm::errc::invalid_argument, 2525 "cannot write section header table because " 2526 "section header string table was removed"); 2527 2528 if (Error E = removeUnneededSections(Obj)) 2529 return E; 2530 2531 // We need to assign indexes before we perform layout because we need to know 2532 // if we need large indexes or not. We can assign indexes first and check as 2533 // we go to see if we will actully need large indexes. 2534 bool NeedsLargeIndexes = false; 2535 if (Obj.sections().size() >= SHN_LORESERVE) { 2536 SectionTableRef Sections = Obj.sections(); 2537 // Sections doesn't include the null section header, so account for this 2538 // when skipping the first N sections. 2539 NeedsLargeIndexes = 2540 any_of(drop_begin(Sections, SHN_LORESERVE - 1), 2541 [](const SectionBase &Sec) { return Sec.HasSymbol; }); 2542 // TODO: handle case where only one section needs the large index table but 2543 // only needs it because the large index table hasn't been removed yet. 2544 } 2545 2546 if (NeedsLargeIndexes) { 2547 // This means we definitely need to have a section index table but if we 2548 // already have one then we should use it instead of making a new one. 2549 if (Obj.SymbolTable != nullptr && Obj.SectionIndexTable == nullptr) { 2550 // Addition of a section to the end does not invalidate the indexes of 2551 // other sections and assigns the correct index to the new section. 2552 auto &Shndx = Obj.addSection<SectionIndexSection>(); 2553 Obj.SymbolTable->setShndxTable(&Shndx); 2554 Shndx.setSymTab(Obj.SymbolTable); 2555 } 2556 } else { 2557 // Since we don't need SectionIndexTable we should remove it and all 2558 // references to it. 2559 if (Obj.SectionIndexTable != nullptr) { 2560 // We do not support sections referring to the section index table. 2561 if (Error E = Obj.removeSections(false /*AllowBrokenLinks*/, 2562 [this](const SectionBase &Sec) { 2563 return &Sec == Obj.SectionIndexTable; 2564 })) 2565 return E; 2566 } 2567 } 2568 2569 // Make sure we add the names of all the sections. Importantly this must be 2570 // done after we decide to add or remove SectionIndexes. 2571 if (Obj.SectionNames != nullptr) 2572 for (const SectionBase &Sec : Obj.sections()) 2573 Obj.SectionNames->addString(Sec.Name); 2574 2575 initEhdrSegment(); 2576 2577 // Before we can prepare for layout the indexes need to be finalized. 2578 // Also, the output arch may not be the same as the input arch, so fix up 2579 // size-related fields before doing layout calculations. 2580 uint64_t Index = 0; 2581 auto SecSizer = std::make_unique<ELFSectionSizer<ELFT>>(); 2582 for (SectionBase &Sec : Obj.sections()) { 2583 Sec.Index = Index++; 2584 if (Error Err = Sec.accept(*SecSizer)) 2585 return Err; 2586 } 2587 2588 // The symbol table does not update all other sections on update. For 2589 // instance, symbol names are not added as new symbols are added. This means 2590 // that some sections, like .strtab, don't yet have their final size. 2591 if (Obj.SymbolTable != nullptr) 2592 Obj.SymbolTable->prepareForLayout(); 2593 2594 // Now that all strings are added we want to finalize string table builders, 2595 // because that affects section sizes which in turn affects section offsets. 2596 for (SectionBase &Sec : Obj.sections()) 2597 if (auto StrTab = dyn_cast<StringTableSection>(&Sec)) 2598 StrTab->prepareForLayout(); 2599 2600 assignOffsets(); 2601 2602 // layoutSections could have modified section indexes, so we need 2603 // to fill the index table after assignOffsets. 2604 if (Obj.SymbolTable != nullptr) 2605 Obj.SymbolTable->fillShndxTable(); 2606 2607 // Finally now that all offsets and indexes have been set we can finalize any 2608 // remaining issues. 2609 uint64_t Offset = Obj.SHOff + sizeof(Elf_Shdr); 2610 for (SectionBase &Sec : Obj.sections()) { 2611 Sec.HeaderOffset = Offset; 2612 Offset += sizeof(Elf_Shdr); 2613 if (WriteSectionHeaders) 2614 Sec.NameIndex = Obj.SectionNames->findIndex(Sec.Name); 2615 Sec.finalize(); 2616 } 2617 2618 size_t TotalSize = totalSize(); 2619 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2620 if (!Buf) 2621 return createStringError(errc::not_enough_memory, 2622 "failed to allocate memory buffer of " + 2623 Twine::utohexstr(TotalSize) + " bytes"); 2624 2625 SecWriter = std::make_unique<ELFSectionWriter<ELFT>>(*Buf); 2626 return Error::success(); 2627 } 2628 2629 Error BinaryWriter::write() { 2630 for (const SectionBase &Sec : Obj.allocSections()) 2631 if (Error Err = Sec.accept(*SecWriter)) 2632 return Err; 2633 2634 // TODO: Implement direct writing to the output stream (without intermediate 2635 // memory buffer Buf). 2636 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2637 return Error::success(); 2638 } 2639 2640 Error BinaryWriter::finalize() { 2641 // Compute the section LMA based on its sh_offset and the containing segment's 2642 // p_offset and p_paddr. Also compute the minimum LMA of all non-empty 2643 // sections as MinAddr. In the output, the contents between address 0 and 2644 // MinAddr will be skipped. 2645 uint64_t MinAddr = UINT64_MAX; 2646 for (SectionBase &Sec : Obj.allocSections()) { 2647 if (Sec.ParentSegment != nullptr) 2648 Sec.Addr = 2649 Sec.Offset - Sec.ParentSegment->Offset + Sec.ParentSegment->PAddr; 2650 if (Sec.Type != SHT_NOBITS && Sec.Size > 0) 2651 MinAddr = std::min(MinAddr, Sec.Addr); 2652 } 2653 2654 // Now that every section has been laid out we just need to compute the total 2655 // file size. This might not be the same as the offset returned by 2656 // layoutSections, because we want to truncate the last segment to the end of 2657 // its last non-empty section, to match GNU objcopy's behaviour. 2658 TotalSize = 0; 2659 for (SectionBase &Sec : Obj.allocSections()) 2660 if (Sec.Type != SHT_NOBITS && Sec.Size > 0) { 2661 Sec.Offset = Sec.Addr - MinAddr; 2662 TotalSize = std::max(TotalSize, Sec.Offset + Sec.Size); 2663 } 2664 2665 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2666 if (!Buf) 2667 return createStringError(errc::not_enough_memory, 2668 "failed to allocate memory buffer of " + 2669 Twine::utohexstr(TotalSize) + " bytes"); 2670 SecWriter = std::make_unique<BinarySectionWriter>(*Buf); 2671 return Error::success(); 2672 } 2673 2674 bool IHexWriter::SectionCompare::operator()(const SectionBase *Lhs, 2675 const SectionBase *Rhs) const { 2676 return (sectionPhysicalAddr(Lhs) & 0xFFFFFFFFU) < 2677 (sectionPhysicalAddr(Rhs) & 0xFFFFFFFFU); 2678 } 2679 2680 uint64_t IHexWriter::writeEntryPointRecord(uint8_t *Buf) { 2681 IHexLineData HexData; 2682 uint8_t Data[4] = {}; 2683 // We don't write entry point record if entry is zero. 2684 if (Obj.Entry == 0) 2685 return 0; 2686 2687 if (Obj.Entry <= 0xFFFFFU) { 2688 Data[0] = ((Obj.Entry & 0xF0000U) >> 12) & 0xFF; 2689 support::endian::write(&Data[2], static_cast<uint16_t>(Obj.Entry), 2690 support::big); 2691 HexData = IHexRecord::getLine(IHexRecord::StartAddr80x86, 0, Data); 2692 } else { 2693 support::endian::write(Data, static_cast<uint32_t>(Obj.Entry), 2694 support::big); 2695 HexData = IHexRecord::getLine(IHexRecord::StartAddr, 0, Data); 2696 } 2697 memcpy(Buf, HexData.data(), HexData.size()); 2698 return HexData.size(); 2699 } 2700 2701 uint64_t IHexWriter::writeEndOfFileRecord(uint8_t *Buf) { 2702 IHexLineData HexData = IHexRecord::getLine(IHexRecord::EndOfFile, 0, {}); 2703 memcpy(Buf, HexData.data(), HexData.size()); 2704 return HexData.size(); 2705 } 2706 2707 Error IHexWriter::write() { 2708 IHexSectionWriter Writer(*Buf); 2709 // Write sections. 2710 for (const SectionBase *Sec : Sections) 2711 if (Error Err = Sec->accept(Writer)) 2712 return Err; 2713 2714 uint64_t Offset = Writer.getBufferOffset(); 2715 // Write entry point address. 2716 Offset += writeEntryPointRecord( 2717 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset); 2718 // Write EOF. 2719 Offset += writeEndOfFileRecord( 2720 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset); 2721 assert(Offset == TotalSize); 2722 2723 // TODO: Implement direct writing to the output stream (without intermediate 2724 // memory buffer Buf). 2725 Out.write(Buf->getBufferStart(), Buf->getBufferSize()); 2726 return Error::success(); 2727 } 2728 2729 Error IHexWriter::checkSection(const SectionBase &Sec) { 2730 uint64_t Addr = sectionPhysicalAddr(&Sec); 2731 if (addressOverflows32bit(Addr) || addressOverflows32bit(Addr + Sec.Size - 1)) 2732 return createStringError( 2733 errc::invalid_argument, 2734 "Section '%s' address range [0x%llx, 0x%llx] is not 32 bit", 2735 Sec.Name.c_str(), Addr, Addr + Sec.Size - 1); 2736 return Error::success(); 2737 } 2738 2739 Error IHexWriter::finalize() { 2740 // We can't write 64-bit addresses. 2741 if (addressOverflows32bit(Obj.Entry)) 2742 return createStringError(errc::invalid_argument, 2743 "Entry point address 0x%llx overflows 32 bits", 2744 Obj.Entry); 2745 2746 for (const SectionBase &Sec : Obj.sections()) 2747 if ((Sec.Flags & ELF::SHF_ALLOC) && Sec.Type != ELF::SHT_NOBITS && 2748 Sec.Size > 0) { 2749 if (Error E = checkSection(Sec)) 2750 return E; 2751 Sections.insert(&Sec); 2752 } 2753 2754 std::unique_ptr<WritableMemoryBuffer> EmptyBuffer = 2755 WritableMemoryBuffer::getNewMemBuffer(0); 2756 if (!EmptyBuffer) 2757 return createStringError(errc::not_enough_memory, 2758 "failed to allocate memory buffer of 0 bytes"); 2759 2760 IHexSectionWriterBase LengthCalc(*EmptyBuffer); 2761 for (const SectionBase *Sec : Sections) 2762 if (Error Err = Sec->accept(LengthCalc)) 2763 return Err; 2764 2765 // We need space to write section records + StartAddress record 2766 // (if start adress is not zero) + EndOfFile record. 2767 TotalSize = LengthCalc.getBufferOffset() + 2768 (Obj.Entry ? IHexRecord::getLineLength(4) : 0) + 2769 IHexRecord::getLineLength(0); 2770 2771 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); 2772 if (!Buf) 2773 return createStringError(errc::not_enough_memory, 2774 "failed to allocate memory buffer of " + 2775 Twine::utohexstr(TotalSize) + " bytes"); 2776 2777 return Error::success(); 2778 } 2779 2780 namespace llvm { 2781 namespace objcopy { 2782 namespace elf { 2783 2784 template class ELFBuilder<ELF64LE>; 2785 template class ELFBuilder<ELF64BE>; 2786 template class ELFBuilder<ELF32LE>; 2787 template class ELFBuilder<ELF32BE>; 2788 2789 template class ELFWriter<ELF64LE>; 2790 template class ELFWriter<ELF64BE>; 2791 template class ELFWriter<ELF32LE>; 2792 template class ELFWriter<ELF32BE>; 2793 2794 } // end namespace elf 2795 } // end namespace objcopy 2796 } // end namespace llvm 2797