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