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