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