| File: | src/gnu/usr.bin/clang/llvm-objcopy/obj/ELFObject.cpp |
| Warning: | line 2653, column 3 Value stored to 'Offset' is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===- Object.cpp ---------------------------------------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | #include "Object.h" |
| 10 | #include "llvm/ADT/ArrayRef.h" |
| 11 | #include "llvm/ADT/STLExtras.h" |
| 12 | #include "llvm/ADT/StringRef.h" |
| 13 | #include "llvm/ADT/Twine.h" |
| 14 | #include "llvm/ADT/iterator_range.h" |
| 15 | #include "llvm/BinaryFormat/ELF.h" |
| 16 | #include "llvm/MC/MCTargetOptions.h" |
| 17 | #include "llvm/Object/ELF.h" |
| 18 | #include "llvm/Object/ELFObjectFile.h" |
| 19 | #include "llvm/Support/Compression.h" |
| 20 | #include "llvm/Support/Endian.h" |
| 21 | #include "llvm/Support/ErrorHandling.h" |
| 22 | #include "llvm/Support/FileOutputBuffer.h" |
| 23 | #include "llvm/Support/Path.h" |
| 24 | #include <algorithm> |
| 25 | #include <cstddef> |
| 26 | #include <cstdint> |
| 27 | #include <iterator> |
| 28 | #include <unordered_set> |
| 29 | #include <utility> |
| 30 | #include <vector> |
| 31 | |
| 32 | 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_MAX0xffffffffU && Addr + 0x80000000 > UINT32_MAX0xffffffffU; |
| 180 | } |
| 181 | |
| 182 | template <class T> static T checkedGetHex(StringRef S) { |
| 183 | T Value; |
| 184 | bool Fail = S.getAsInteger(16, Value); |
| 185 | assert(!Fail)((void)0); |
| 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)((void)0); |
| 203 | return It + Len; |
| 204 | } |
| 205 | |
| 206 | uint8_t IHexRecord::getChecksum(StringRef S) { |
| 207 | assert((S.size() & 1) == 0)((void)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())((void)0); |
| 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())((void)0); |
| 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())((void)0); |
| 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())((void)0); |
| 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)((void)0); |
| 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)((void)0); |
| 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)((void)0); |
| 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)((void)0); |
| 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())((void)0); |
| 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())((void)0); |
| 418 | std::vector<uint8_t> Data(Sec.Size); |
| 419 | Sec.StrTabBuilder.write(Data.data()); |
| 420 | writeSection(&Sec, Data); |
| 421 | return Error::success(); |
| 422 | } |
| 423 | |
| 424 | Error Section::accept(SectionVisitor &Visitor) const { |
| 425 | return Visitor.visit(*this); |
| 426 | } |
| 427 | |
| 428 | Error Section::accept(MutableSectionVisitor &Visitor) { |
| 429 | return Visitor.visit(*this); |
| 430 | } |
| 431 | |
| 432 | Error SectionWriter::visit(const OwnedDataSection &Sec) { |
| 433 | llvm::copy(Sec.Data, Out.getBufferStart() + Sec.Offset); |
| 434 | return Error::success(); |
| 435 | } |
| 436 | |
| 437 | static constexpr std::array<uint8_t, 4> ZlibGnuMagic = {{'Z', 'L', 'I', 'B'}}; |
| 438 | |
| 439 | static bool isDataGnuCompressed(ArrayRef<uint8_t> Data) { |
| 440 | return Data.size() > ZlibGnuMagic.size() && |
| 441 | std::equal(ZlibGnuMagic.begin(), ZlibGnuMagic.end(), Data.data()); |
| 442 | } |
| 443 | |
| 444 | template <class ELFT> |
| 445 | static std::tuple<uint64_t, uint64_t> |
| 446 | getDecompressedSizeAndAlignment(ArrayRef<uint8_t> Data) { |
| 447 | const bool IsGnuDebug = isDataGnuCompressed(Data); |
| 448 | const uint64_t DecompressedSize = |
| 449 | IsGnuDebug |
| 450 | ? support::endian::read64be(Data.data() + ZlibGnuMagic.size()) |
| 451 | : reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data.data())->ch_size; |
| 452 | const uint64_t DecompressedAlign = |
| 453 | IsGnuDebug ? 1 |
| 454 | : reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data.data()) |
| 455 | ->ch_addralign; |
| 456 | |
| 457 | return std::make_tuple(DecompressedSize, DecompressedAlign); |
| 458 | } |
| 459 | |
| 460 | template <class ELFT> |
| 461 | Error ELFSectionWriter<ELFT>::visit(const DecompressedSection &Sec) { |
| 462 | const size_t DataOffset = isDataGnuCompressed(Sec.OriginalData) |
| 463 | ? (ZlibGnuMagic.size() + sizeof(Sec.Size)) |
| 464 | : sizeof(Elf_Chdr_Impl<ELFT>); |
| 465 | |
| 466 | StringRef CompressedContent( |
| 467 | reinterpret_cast<const char *>(Sec.OriginalData.data()) + DataOffset, |
| 468 | Sec.OriginalData.size() - DataOffset); |
| 469 | |
| 470 | SmallVector<char, 128> DecompressedContent; |
| 471 | if (Error Err = zlib::uncompress(CompressedContent, DecompressedContent, |
| 472 | static_cast<size_t>(Sec.Size))) |
| 473 | return createStringError(errc::invalid_argument, |
| 474 | "'" + Sec.Name + "': " + toString(std::move(Err))); |
| 475 | |
| 476 | uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; |
| 477 | std::copy(DecompressedContent.begin(), DecompressedContent.end(), Buf); |
| 478 | |
| 479 | return Error::success(); |
| 480 | } |
| 481 | |
| 482 | Error BinarySectionWriter::visit(const DecompressedSection &Sec) { |
| 483 | return createStringError(errc::operation_not_permitted, |
| 484 | "cannot write compressed section '" + Sec.Name + |
| 485 | "' "); |
| 486 | } |
| 487 | |
| 488 | Error DecompressedSection::accept(SectionVisitor &Visitor) const { |
| 489 | return Visitor.visit(*this); |
| 490 | } |
| 491 | |
| 492 | Error DecompressedSection::accept(MutableSectionVisitor &Visitor) { |
| 493 | return Visitor.visit(*this); |
| 494 | } |
| 495 | |
| 496 | Error OwnedDataSection::accept(SectionVisitor &Visitor) const { |
| 497 | return Visitor.visit(*this); |
| 498 | } |
| 499 | |
| 500 | Error OwnedDataSection::accept(MutableSectionVisitor &Visitor) { |
| 501 | return Visitor.visit(*this); |
| 502 | } |
| 503 | |
| 504 | void OwnedDataSection::appendHexData(StringRef HexData) { |
| 505 | assert((HexData.size() & 1) == 0)((void)0); |
| 506 | while (!HexData.empty()) { |
| 507 | Data.push_back(checkedGetHex<uint8_t>(HexData.take_front(2))); |
| 508 | HexData = HexData.drop_front(2); |
| 509 | } |
| 510 | Size = Data.size(); |
| 511 | } |
| 512 | |
| 513 | Error BinarySectionWriter::visit(const CompressedSection &Sec) { |
| 514 | return createStringError(errc::operation_not_permitted, |
| 515 | "cannot write compressed section '" + Sec.Name + |
| 516 | "' "); |
| 517 | } |
| 518 | |
| 519 | template <class ELFT> |
| 520 | Error ELFSectionWriter<ELFT>::visit(const CompressedSection &Sec) { |
| 521 | uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; |
| 522 | if (Sec.CompressionType == DebugCompressionType::None) { |
| 523 | std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(), Buf); |
| 524 | return Error::success(); |
| 525 | } |
| 526 | |
| 527 | if (Sec.CompressionType == DebugCompressionType::GNU) { |
| 528 | const char *Magic = "ZLIB"; |
| 529 | memcpy(Buf, Magic, strlen(Magic)); |
| 530 | Buf += strlen(Magic); |
| 531 | const uint64_t DecompressedSize = |
| 532 | support::endian::read64be(&Sec.DecompressedSize); |
| 533 | memcpy(Buf, &DecompressedSize, sizeof(DecompressedSize)); |
| 534 | Buf += sizeof(DecompressedSize); |
| 535 | } else { |
| 536 | Elf_Chdr_Impl<ELFT> Chdr; |
| 537 | Chdr.ch_type = ELF::ELFCOMPRESS_ZLIB; |
| 538 | Chdr.ch_size = Sec.DecompressedSize; |
| 539 | Chdr.ch_addralign = Sec.DecompressedAlign; |
| 540 | memcpy(Buf, &Chdr, sizeof(Chdr)); |
| 541 | Buf += sizeof(Chdr); |
| 542 | } |
| 543 | |
| 544 | std::copy(Sec.CompressedData.begin(), Sec.CompressedData.end(), Buf); |
| 545 | return Error::success(); |
| 546 | } |
| 547 | |
| 548 | Expected<CompressedSection> |
| 549 | CompressedSection::create(const SectionBase &Sec, |
| 550 | DebugCompressionType CompressionType) { |
| 551 | Error Err = Error::success(); |
| 552 | CompressedSection Section(Sec, CompressionType, Err); |
| 553 | |
| 554 | if (Err) |
| 555 | return std::move(Err); |
| 556 | |
| 557 | return Section; |
| 558 | } |
| 559 | Expected<CompressedSection> |
| 560 | CompressedSection::create(ArrayRef<uint8_t> CompressedData, |
| 561 | uint64_t DecompressedSize, |
| 562 | uint64_t DecompressedAlign) { |
| 563 | return CompressedSection(CompressedData, DecompressedSize, DecompressedAlign); |
| 564 | } |
| 565 | |
| 566 | CompressedSection::CompressedSection(const SectionBase &Sec, |
| 567 | DebugCompressionType CompressionType, |
| 568 | Error &OutErr) |
| 569 | : SectionBase(Sec), CompressionType(CompressionType), |
| 570 | DecompressedSize(Sec.OriginalData.size()), DecompressedAlign(Sec.Align) { |
| 571 | ErrorAsOutParameter EAO(&OutErr); |
| 572 | |
| 573 | if (Error Err = zlib::compress( |
| 574 | StringRef(reinterpret_cast<const char *>(OriginalData.data()), |
| 575 | OriginalData.size()), |
| 576 | CompressedData)) { |
| 577 | OutErr = createStringError(llvm::errc::invalid_argument, |
| 578 | "'" + Name + "': " + toString(std::move(Err))); |
| 579 | return; |
| 580 | } |
| 581 | |
| 582 | size_t ChdrSize; |
| 583 | if (CompressionType == DebugCompressionType::GNU) { |
| 584 | Name = ".z" + Sec.Name.substr(1); |
| 585 | ChdrSize = sizeof("ZLIB") - 1 + sizeof(uint64_t); |
| 586 | } else { |
| 587 | Flags |= ELF::SHF_COMPRESSED; |
| 588 | ChdrSize = |
| 589 | std::max(std::max(sizeof(object::Elf_Chdr_Impl<object::ELF64LE>), |
| 590 | sizeof(object::Elf_Chdr_Impl<object::ELF64BE>)), |
| 591 | std::max(sizeof(object::Elf_Chdr_Impl<object::ELF32LE>), |
| 592 | sizeof(object::Elf_Chdr_Impl<object::ELF32BE>))); |
| 593 | } |
| 594 | Size = ChdrSize + CompressedData.size(); |
| 595 | Align = 8; |
| 596 | } |
| 597 | |
| 598 | CompressedSection::CompressedSection(ArrayRef<uint8_t> CompressedData, |
| 599 | uint64_t DecompressedSize, |
| 600 | uint64_t DecompressedAlign) |
| 601 | : CompressionType(DebugCompressionType::None), |
| 602 | DecompressedSize(DecompressedSize), DecompressedAlign(DecompressedAlign) { |
| 603 | OriginalData = CompressedData; |
| 604 | } |
| 605 | |
| 606 | Error CompressedSection::accept(SectionVisitor &Visitor) const { |
| 607 | return Visitor.visit(*this); |
| 608 | } |
| 609 | |
| 610 | Error CompressedSection::accept(MutableSectionVisitor &Visitor) { |
| 611 | return Visitor.visit(*this); |
| 612 | } |
| 613 | |
| 614 | void StringTableSection::addString(StringRef Name) { StrTabBuilder.add(Name); } |
| 615 | |
| 616 | uint32_t StringTableSection::findIndex(StringRef Name) const { |
| 617 | return StrTabBuilder.getOffset(Name); |
| 618 | } |
| 619 | |
| 620 | void StringTableSection::prepareForLayout() { |
| 621 | StrTabBuilder.finalize(); |
| 622 | Size = StrTabBuilder.getSize(); |
| 623 | } |
| 624 | |
| 625 | Error SectionWriter::visit(const StringTableSection &Sec) { |
| 626 | Sec.StrTabBuilder.write(reinterpret_cast<uint8_t *>(Out.getBufferStart()) + |
| 627 | Sec.Offset); |
| 628 | return Error::success(); |
| 629 | } |
| 630 | |
| 631 | Error StringTableSection::accept(SectionVisitor &Visitor) const { |
| 632 | return Visitor.visit(*this); |
| 633 | } |
| 634 | |
| 635 | Error StringTableSection::accept(MutableSectionVisitor &Visitor) { |
| 636 | return Visitor.visit(*this); |
| 637 | } |
| 638 | |
| 639 | template <class ELFT> |
| 640 | Error ELFSectionWriter<ELFT>::visit(const SectionIndexSection &Sec) { |
| 641 | uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; |
| 642 | llvm::copy(Sec.Indexes, reinterpret_cast<Elf_Word *>(Buf)); |
| 643 | return Error::success(); |
| 644 | } |
| 645 | |
| 646 | Error SectionIndexSection::initialize(SectionTableRef SecTable) { |
| 647 | Size = 0; |
| 648 | Expected<SymbolTableSection *> Sec = |
| 649 | SecTable.getSectionOfType<SymbolTableSection>( |
| 650 | Link, |
| 651 | "Link field value " + Twine(Link) + " in section " + Name + |
| 652 | " is invalid", |
| 653 | "Link field value " + Twine(Link) + " in section " + Name + |
| 654 | " is not a symbol table"); |
| 655 | if (!Sec) |
| 656 | return Sec.takeError(); |
| 657 | |
| 658 | setSymTab(*Sec); |
| 659 | Symbols->setShndxTable(this); |
| 660 | return Error::success(); |
| 661 | } |
| 662 | |
| 663 | void SectionIndexSection::finalize() { Link = Symbols->Index; } |
| 664 | |
| 665 | Error SectionIndexSection::accept(SectionVisitor &Visitor) const { |
| 666 | return Visitor.visit(*this); |
| 667 | } |
| 668 | |
| 669 | Error SectionIndexSection::accept(MutableSectionVisitor &Visitor) { |
| 670 | return Visitor.visit(*this); |
| 671 | } |
| 672 | |
| 673 | static bool isValidReservedSectionIndex(uint16_t Index, uint16_t Machine) { |
| 674 | switch (Index) { |
| 675 | case SHN_ABS: |
| 676 | case SHN_COMMON: |
| 677 | return true; |
| 678 | } |
| 679 | |
| 680 | if (Machine == EM_AMDGPU) { |
| 681 | return Index == SHN_AMDGPU_LDS; |
| 682 | } |
| 683 | |
| 684 | if (Machine == EM_HEXAGON) { |
| 685 | switch (Index) { |
| 686 | case SHN_HEXAGON_SCOMMON: |
| 687 | case SHN_HEXAGON_SCOMMON_1: |
| 688 | case SHN_HEXAGON_SCOMMON_2: |
| 689 | case SHN_HEXAGON_SCOMMON_4: |
| 690 | case SHN_HEXAGON_SCOMMON_8: |
| 691 | return true; |
| 692 | } |
| 693 | } |
| 694 | return false; |
| 695 | } |
| 696 | |
| 697 | // Large indexes force us to clarify exactly what this function should do. This |
| 698 | // function should return the value that will appear in st_shndx when written |
| 699 | // out. |
| 700 | uint16_t Symbol::getShndx() const { |
| 701 | if (DefinedIn != nullptr) { |
| 702 | if (DefinedIn->Index >= SHN_LORESERVE) |
| 703 | return SHN_XINDEX; |
| 704 | return DefinedIn->Index; |
| 705 | } |
| 706 | |
| 707 | if (ShndxType == SYMBOL_SIMPLE_INDEX) { |
| 708 | // This means that we don't have a defined section but we do need to |
| 709 | // output a legitimate section index. |
| 710 | return SHN_UNDEF; |
| 711 | } |
| 712 | |
| 713 | assert(ShndxType == SYMBOL_ABS || ShndxType == SYMBOL_COMMON ||((void)0) |
| 714 | (ShndxType >= SYMBOL_LOPROC && ShndxType <= SYMBOL_HIPROC) ||((void)0) |
| 715 | (ShndxType >= SYMBOL_LOOS && ShndxType <= SYMBOL_HIOS))((void)0); |
| 716 | return static_cast<uint16_t>(ShndxType); |
| 717 | } |
| 718 | |
| 719 | bool Symbol::isCommon() const { return getShndx() == SHN_COMMON; } |
| 720 | |
| 721 | void SymbolTableSection::assignIndices() { |
| 722 | uint32_t Index = 0; |
| 723 | for (auto &Sym : Symbols) |
| 724 | Sym->Index = Index++; |
| 725 | } |
| 726 | |
| 727 | void SymbolTableSection::addSymbol(Twine Name, uint8_t Bind, uint8_t Type, |
| 728 | SectionBase *DefinedIn, uint64_t Value, |
| 729 | uint8_t Visibility, uint16_t Shndx, |
| 730 | uint64_t SymbolSize) { |
| 731 | Symbol Sym; |
| 732 | Sym.Name = Name.str(); |
| 733 | Sym.Binding = Bind; |
| 734 | Sym.Type = Type; |
| 735 | Sym.DefinedIn = DefinedIn; |
| 736 | if (DefinedIn != nullptr) |
| 737 | DefinedIn->HasSymbol = true; |
| 738 | if (DefinedIn == nullptr) { |
| 739 | if (Shndx >= SHN_LORESERVE) |
| 740 | Sym.ShndxType = static_cast<SymbolShndxType>(Shndx); |
| 741 | else |
| 742 | Sym.ShndxType = SYMBOL_SIMPLE_INDEX; |
| 743 | } |
| 744 | Sym.Value = Value; |
| 745 | Sym.Visibility = Visibility; |
| 746 | Sym.Size = SymbolSize; |
| 747 | Sym.Index = Symbols.size(); |
| 748 | Symbols.emplace_back(std::make_unique<Symbol>(Sym)); |
| 749 | Size += this->EntrySize; |
| 750 | } |
| 751 | |
| 752 | Error SymbolTableSection::removeSectionReferences( |
| 753 | bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { |
| 754 | if (ToRemove(SectionIndexTable)) |
| 755 | SectionIndexTable = nullptr; |
| 756 | if (ToRemove(SymbolNames)) { |
| 757 | if (!AllowBrokenLinks) |
| 758 | return createStringError( |
| 759 | llvm::errc::invalid_argument, |
| 760 | "string table '%s' cannot be removed because it is " |
| 761 | "referenced by the symbol table '%s'", |
| 762 | SymbolNames->Name.data(), this->Name.data()); |
| 763 | SymbolNames = nullptr; |
| 764 | } |
| 765 | return removeSymbols( |
| 766 | [ToRemove](const Symbol &Sym) { return ToRemove(Sym.DefinedIn); }); |
| 767 | } |
| 768 | |
| 769 | void SymbolTableSection::updateSymbols(function_ref<void(Symbol &)> Callable) { |
| 770 | std::for_each(std::begin(Symbols) + 1, std::end(Symbols), |
| 771 | [Callable](SymPtr &Sym) { Callable(*Sym); }); |
| 772 | std::stable_partition( |
| 773 | std::begin(Symbols), std::end(Symbols), |
| 774 | [](const SymPtr &Sym) { return Sym->Binding == STB_LOCAL; }); |
| 775 | assignIndices(); |
| 776 | } |
| 777 | |
| 778 | Error SymbolTableSection::removeSymbols( |
| 779 | function_ref<bool(const Symbol &)> ToRemove) { |
| 780 | Symbols.erase( |
| 781 | std::remove_if(std::begin(Symbols) + 1, std::end(Symbols), |
| 782 | [ToRemove](const SymPtr &Sym) { return ToRemove(*Sym); }), |
| 783 | std::end(Symbols)); |
| 784 | Size = Symbols.size() * EntrySize; |
| 785 | assignIndices(); |
| 786 | return Error::success(); |
| 787 | } |
| 788 | |
| 789 | void SymbolTableSection::replaceSectionReferences( |
| 790 | const DenseMap<SectionBase *, SectionBase *> &FromTo) { |
| 791 | for (std::unique_ptr<Symbol> &Sym : Symbols) |
| 792 | if (SectionBase *To = FromTo.lookup(Sym->DefinedIn)) |
| 793 | Sym->DefinedIn = To; |
| 794 | } |
| 795 | |
| 796 | Error SymbolTableSection::initialize(SectionTableRef SecTable) { |
| 797 | Size = 0; |
| 798 | Expected<StringTableSection *> Sec = |
| 799 | SecTable.getSectionOfType<StringTableSection>( |
| 800 | Link, |
| 801 | "Symbol table has link index of " + Twine(Link) + |
| 802 | " which is not a valid index", |
| 803 | "Symbol table has link index of " + Twine(Link) + |
| 804 | " which is not a string table"); |
| 805 | if (!Sec) |
| 806 | return Sec.takeError(); |
| 807 | |
| 808 | setStrTab(*Sec); |
| 809 | return Error::success(); |
| 810 | } |
| 811 | |
| 812 | void SymbolTableSection::finalize() { |
| 813 | uint32_t MaxLocalIndex = 0; |
| 814 | for (std::unique_ptr<Symbol> &Sym : Symbols) { |
| 815 | Sym->NameIndex = |
| 816 | SymbolNames == nullptr ? 0 : SymbolNames->findIndex(Sym->Name); |
| 817 | if (Sym->Binding == STB_LOCAL) |
| 818 | MaxLocalIndex = std::max(MaxLocalIndex, Sym->Index); |
| 819 | } |
| 820 | // Now we need to set the Link and Info fields. |
| 821 | Link = SymbolNames == nullptr ? 0 : SymbolNames->Index; |
| 822 | Info = MaxLocalIndex + 1; |
| 823 | } |
| 824 | |
| 825 | void SymbolTableSection::prepareForLayout() { |
| 826 | // Reserve proper amount of space in section index table, so we can |
| 827 | // layout sections correctly. We will fill the table with correct |
| 828 | // indexes later in fillShdnxTable. |
| 829 | if (SectionIndexTable) |
| 830 | SectionIndexTable->reserve(Symbols.size()); |
| 831 | |
| 832 | // Add all of our strings to SymbolNames so that SymbolNames has the right |
| 833 | // size before layout is decided. |
| 834 | // If the symbol names section has been removed, don't try to add strings to |
| 835 | // the table. |
| 836 | if (SymbolNames != nullptr) |
| 837 | for (std::unique_ptr<Symbol> &Sym : Symbols) |
| 838 | SymbolNames->addString(Sym->Name); |
| 839 | } |
| 840 | |
| 841 | void SymbolTableSection::fillShndxTable() { |
| 842 | if (SectionIndexTable == nullptr) |
| 843 | return; |
| 844 | // Fill section index table with real section indexes. This function must |
| 845 | // be called after assignOffsets. |
| 846 | for (const std::unique_ptr<Symbol> &Sym : Symbols) { |
| 847 | if (Sym->DefinedIn != nullptr && Sym->DefinedIn->Index >= SHN_LORESERVE) |
| 848 | SectionIndexTable->addIndex(Sym->DefinedIn->Index); |
| 849 | else |
| 850 | SectionIndexTable->addIndex(SHN_UNDEF); |
| 851 | } |
| 852 | } |
| 853 | |
| 854 | Expected<const Symbol *> |
| 855 | SymbolTableSection::getSymbolByIndex(uint32_t Index) const { |
| 856 | if (Symbols.size() <= Index) |
| 857 | return createStringError(errc::invalid_argument, |
| 858 | "invalid symbol index: " + Twine(Index)); |
| 859 | return Symbols[Index].get(); |
| 860 | } |
| 861 | |
| 862 | Expected<Symbol *> SymbolTableSection::getSymbolByIndex(uint32_t Index) { |
| 863 | Expected<const Symbol *> Sym = |
| 864 | static_cast<const SymbolTableSection *>(this)->getSymbolByIndex(Index); |
| 865 | if (!Sym) |
| 866 | return Sym.takeError(); |
| 867 | |
| 868 | return const_cast<Symbol *>(*Sym); |
| 869 | } |
| 870 | |
| 871 | template <class ELFT> |
| 872 | Error ELFSectionWriter<ELFT>::visit(const SymbolTableSection &Sec) { |
| 873 | Elf_Sym *Sym = reinterpret_cast<Elf_Sym *>(Out.getBufferStart() + Sec.Offset); |
| 874 | // Loop though symbols setting each entry of the symbol table. |
| 875 | for (const std::unique_ptr<Symbol> &Symbol : Sec.Symbols) { |
| 876 | Sym->st_name = Symbol->NameIndex; |
| 877 | Sym->st_value = Symbol->Value; |
| 878 | Sym->st_size = Symbol->Size; |
| 879 | Sym->st_other = Symbol->Visibility; |
| 880 | Sym->setBinding(Symbol->Binding); |
| 881 | Sym->setType(Symbol->Type); |
| 882 | Sym->st_shndx = Symbol->getShndx(); |
| 883 | ++Sym; |
| 884 | } |
| 885 | return Error::success(); |
| 886 | } |
| 887 | |
| 888 | Error SymbolTableSection::accept(SectionVisitor &Visitor) const { |
| 889 | return Visitor.visit(*this); |
| 890 | } |
| 891 | |
| 892 | Error SymbolTableSection::accept(MutableSectionVisitor &Visitor) { |
| 893 | return Visitor.visit(*this); |
| 894 | } |
| 895 | |
| 896 | Error RelocationSection::removeSectionReferences( |
| 897 | bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { |
| 898 | if (ToRemove(Symbols)) { |
| 899 | if (!AllowBrokenLinks) |
| 900 | return createStringError( |
| 901 | llvm::errc::invalid_argument, |
| 902 | "symbol table '%s' cannot be removed because it is " |
| 903 | "referenced by the relocation section '%s'", |
| 904 | Symbols->Name.data(), this->Name.data()); |
| 905 | Symbols = nullptr; |
| 906 | } |
| 907 | |
| 908 | for (const Relocation &R : Relocations) { |
| 909 | if (!R.RelocSymbol || !R.RelocSymbol->DefinedIn || |
| 910 | !ToRemove(R.RelocSymbol->DefinedIn)) |
| 911 | continue; |
| 912 | return createStringError(llvm::errc::invalid_argument, |
| 913 | "section '%s' cannot be removed: (%s+0x%" PRIx64"llx" |
| 914 | ") has relocation against symbol '%s'", |
| 915 | R.RelocSymbol->DefinedIn->Name.data(), |
| 916 | SecToApplyRel->Name.data(), R.Offset, |
| 917 | R.RelocSymbol->Name.c_str()); |
| 918 | } |
| 919 | |
| 920 | return Error::success(); |
| 921 | } |
| 922 | |
| 923 | template <class SymTabType> |
| 924 | Error RelocSectionWithSymtabBase<SymTabType>::initialize( |
| 925 | SectionTableRef SecTable) { |
| 926 | if (Link != SHN_UNDEF) { |
| 927 | Expected<SymTabType *> Sec = SecTable.getSectionOfType<SymTabType>( |
| 928 | Link, |
| 929 | "Link field value " + Twine(Link) + " in section " + Name + |
| 930 | " is invalid", |
| 931 | "Link field value " + Twine(Link) + " in section " + Name + |
| 932 | " is not a symbol table"); |
| 933 | if (!Sec) |
| 934 | return Sec.takeError(); |
| 935 | |
| 936 | setSymTab(*Sec); |
| 937 | } |
| 938 | |
| 939 | if (Info != SHN_UNDEF) { |
| 940 | Expected<SectionBase *> Sec = |
| 941 | SecTable.getSection(Info, "Info field value " + Twine(Info) + |
| 942 | " in section " + Name + " is invalid"); |
| 943 | if (!Sec) |
| 944 | return Sec.takeError(); |
| 945 | |
| 946 | setSection(*Sec); |
| 947 | } else |
| 948 | setSection(nullptr); |
| 949 | |
| 950 | return Error::success(); |
| 951 | } |
| 952 | |
| 953 | template <class SymTabType> |
| 954 | void RelocSectionWithSymtabBase<SymTabType>::finalize() { |
| 955 | this->Link = Symbols ? Symbols->Index : 0; |
| 956 | |
| 957 | if (SecToApplyRel != nullptr) |
| 958 | this->Info = SecToApplyRel->Index; |
| 959 | } |
| 960 | |
| 961 | template <class ELFT> |
| 962 | static void setAddend(Elf_Rel_Impl<ELFT, false> &, uint64_t) {} |
| 963 | |
| 964 | template <class ELFT> |
| 965 | static void setAddend(Elf_Rel_Impl<ELFT, true> &Rela, uint64_t Addend) { |
| 966 | Rela.r_addend = Addend; |
| 967 | } |
| 968 | |
| 969 | template <class RelRange, class T> |
| 970 | static void writeRel(const RelRange &Relocations, T *Buf) { |
| 971 | for (const auto &Reloc : Relocations) { |
| 972 | Buf->r_offset = Reloc.Offset; |
| 973 | setAddend(*Buf, Reloc.Addend); |
| 974 | Buf->setSymbolAndType(Reloc.RelocSymbol ? Reloc.RelocSymbol->Index : 0, |
| 975 | Reloc.Type, false); |
| 976 | ++Buf; |
| 977 | } |
| 978 | } |
| 979 | |
| 980 | template <class ELFT> |
| 981 | Error ELFSectionWriter<ELFT>::visit(const RelocationSection &Sec) { |
| 982 | uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; |
| 983 | if (Sec.Type == SHT_REL) |
| 984 | writeRel(Sec.Relocations, reinterpret_cast<Elf_Rel *>(Buf)); |
| 985 | else |
| 986 | writeRel(Sec.Relocations, reinterpret_cast<Elf_Rela *>(Buf)); |
| 987 | return Error::success(); |
| 988 | } |
| 989 | |
| 990 | Error RelocationSection::accept(SectionVisitor &Visitor) const { |
| 991 | return Visitor.visit(*this); |
| 992 | } |
| 993 | |
| 994 | Error RelocationSection::accept(MutableSectionVisitor &Visitor) { |
| 995 | return Visitor.visit(*this); |
| 996 | } |
| 997 | |
| 998 | Error RelocationSection::removeSymbols( |
| 999 | function_ref<bool(const Symbol &)> ToRemove) { |
| 1000 | for (const Relocation &Reloc : Relocations) |
| 1001 | if (Reloc.RelocSymbol && ToRemove(*Reloc.RelocSymbol)) |
| 1002 | return createStringError( |
| 1003 | llvm::errc::invalid_argument, |
| 1004 | "not stripping symbol '%s' because it is named in a relocation", |
| 1005 | Reloc.RelocSymbol->Name.data()); |
| 1006 | return Error::success(); |
| 1007 | } |
| 1008 | |
| 1009 | void RelocationSection::markSymbols() { |
| 1010 | for (const Relocation &Reloc : Relocations) |
| 1011 | if (Reloc.RelocSymbol) |
| 1012 | Reloc.RelocSymbol->Referenced = true; |
| 1013 | } |
| 1014 | |
| 1015 | void RelocationSection::replaceSectionReferences( |
| 1016 | const DenseMap<SectionBase *, SectionBase *> &FromTo) { |
| 1017 | // Update the target section if it was replaced. |
| 1018 | if (SectionBase *To = FromTo.lookup(SecToApplyRel)) |
| 1019 | SecToApplyRel = To; |
| 1020 | } |
| 1021 | |
| 1022 | Error SectionWriter::visit(const DynamicRelocationSection &Sec) { |
| 1023 | llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset); |
| 1024 | return Error::success(); |
| 1025 | } |
| 1026 | |
| 1027 | Error DynamicRelocationSection::accept(SectionVisitor &Visitor) const { |
| 1028 | return Visitor.visit(*this); |
| 1029 | } |
| 1030 | |
| 1031 | Error DynamicRelocationSection::accept(MutableSectionVisitor &Visitor) { |
| 1032 | return Visitor.visit(*this); |
| 1033 | } |
| 1034 | |
| 1035 | Error DynamicRelocationSection::removeSectionReferences( |
| 1036 | bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { |
| 1037 | if (ToRemove(Symbols)) { |
| 1038 | if (!AllowBrokenLinks) |
| 1039 | return createStringError( |
| 1040 | llvm::errc::invalid_argument, |
| 1041 | "symbol table '%s' cannot be removed because it is " |
| 1042 | "referenced by the relocation section '%s'", |
| 1043 | Symbols->Name.data(), this->Name.data()); |
| 1044 | Symbols = nullptr; |
| 1045 | } |
| 1046 | |
| 1047 | // SecToApplyRel contains a section referenced by sh_info field. It keeps |
| 1048 | // a section to which the relocation section applies. When we remove any |
| 1049 | // sections we also remove their relocation sections. Since we do that much |
| 1050 | // earlier, this assert should never be triggered. |
| 1051 | assert(!SecToApplyRel || !ToRemove(SecToApplyRel))((void)0); |
| 1052 | return Error::success(); |
| 1053 | } |
| 1054 | |
| 1055 | Error Section::removeSectionReferences( |
| 1056 | bool AllowBrokenDependency, |
| 1057 | function_ref<bool(const SectionBase *)> ToRemove) { |
| 1058 | if (ToRemove(LinkSection)) { |
| 1059 | if (!AllowBrokenDependency) |
| 1060 | return createStringError(llvm::errc::invalid_argument, |
| 1061 | "section '%s' cannot be removed because it is " |
| 1062 | "referenced by the section '%s'", |
| 1063 | LinkSection->Name.data(), this->Name.data()); |
| 1064 | LinkSection = nullptr; |
| 1065 | } |
| 1066 | return Error::success(); |
| 1067 | } |
| 1068 | |
| 1069 | void GroupSection::finalize() { |
| 1070 | this->Info = Sym ? Sym->Index : 0; |
| 1071 | this->Link = SymTab ? SymTab->Index : 0; |
| 1072 | // Linker deduplication for GRP_COMDAT is based on Sym->Name. The local/global |
| 1073 | // status is not part of the equation. If Sym is localized, the intention is |
| 1074 | // likely to make the group fully localized. Drop GRP_COMDAT to suppress |
| 1075 | // deduplication. See https://groups.google.com/g/generic-abi/c/2X6mR-s2zoc |
| 1076 | if ((FlagWord & GRP_COMDAT) && Sym && Sym->Binding == STB_LOCAL) |
| 1077 | this->FlagWord &= ~GRP_COMDAT; |
| 1078 | } |
| 1079 | |
| 1080 | Error GroupSection::removeSectionReferences( |
| 1081 | bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) { |
| 1082 | if (ToRemove(SymTab)) { |
| 1083 | if (!AllowBrokenLinks) |
| 1084 | return createStringError( |
| 1085 | llvm::errc::invalid_argument, |
| 1086 | "section '.symtab' cannot be removed because it is " |
| 1087 | "referenced by the group section '%s'", |
| 1088 | this->Name.data()); |
| 1089 | SymTab = nullptr; |
| 1090 | Sym = nullptr; |
| 1091 | } |
| 1092 | llvm::erase_if(GroupMembers, ToRemove); |
| 1093 | return Error::success(); |
| 1094 | } |
| 1095 | |
| 1096 | Error GroupSection::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) { |
| 1097 | if (ToRemove(*Sym)) |
| 1098 | return createStringError(llvm::errc::invalid_argument, |
| 1099 | "symbol '%s' cannot be removed because it is " |
| 1100 | "referenced by the section '%s[%d]'", |
| 1101 | Sym->Name.data(), this->Name.data(), this->Index); |
| 1102 | return Error::success(); |
| 1103 | } |
| 1104 | |
| 1105 | void GroupSection::markSymbols() { |
| 1106 | if (Sym) |
| 1107 | Sym->Referenced = true; |
| 1108 | } |
| 1109 | |
| 1110 | void GroupSection::replaceSectionReferences( |
| 1111 | const DenseMap<SectionBase *, SectionBase *> &FromTo) { |
| 1112 | for (SectionBase *&Sec : GroupMembers) |
| 1113 | if (SectionBase *To = FromTo.lookup(Sec)) |
| 1114 | Sec = To; |
| 1115 | } |
| 1116 | |
| 1117 | void GroupSection::onRemove() { |
| 1118 | // As the header section of the group is removed, drop the Group flag in its |
| 1119 | // former members. |
| 1120 | for (SectionBase *Sec : GroupMembers) |
| 1121 | Sec->Flags &= ~SHF_GROUP; |
| 1122 | } |
| 1123 | |
| 1124 | Error Section::initialize(SectionTableRef SecTable) { |
| 1125 | if (Link == ELF::SHN_UNDEF) |
| 1126 | return Error::success(); |
| 1127 | |
| 1128 | Expected<SectionBase *> Sec = |
| 1129 | SecTable.getSection(Link, "Link field value " + Twine(Link) + |
| 1130 | " in section " + Name + " is invalid"); |
| 1131 | if (!Sec) |
| 1132 | return Sec.takeError(); |
| 1133 | |
| 1134 | LinkSection = *Sec; |
| 1135 | |
| 1136 | if (LinkSection->Type == ELF::SHT_SYMTAB) |
| 1137 | LinkSection = nullptr; |
| 1138 | |
| 1139 | return Error::success(); |
| 1140 | } |
| 1141 | |
| 1142 | void Section::finalize() { this->Link = LinkSection ? LinkSection->Index : 0; } |
| 1143 | |
| 1144 | void GnuDebugLinkSection::init(StringRef File) { |
| 1145 | FileName = sys::path::filename(File); |
| 1146 | // The format for the .gnu_debuglink starts with the file name and is |
| 1147 | // followed by a null terminator and then the CRC32 of the file. The CRC32 |
| 1148 | // should be 4 byte aligned. So we add the FileName size, a 1 for the null |
| 1149 | // byte, and then finally push the size to alignment and add 4. |
| 1150 | Size = alignTo(FileName.size() + 1, 4) + 4; |
| 1151 | // The CRC32 will only be aligned if we align the whole section. |
| 1152 | Align = 4; |
| 1153 | Type = OriginalType = ELF::SHT_PROGBITS; |
| 1154 | Name = ".gnu_debuglink"; |
| 1155 | // For sections not found in segments, OriginalOffset is only used to |
| 1156 | // establish the order that sections should go in. By using the maximum |
| 1157 | // possible offset we cause this section to wind up at the end. |
| 1158 | OriginalOffset = std::numeric_limits<uint64_t>::max(); |
| 1159 | } |
| 1160 | |
| 1161 | GnuDebugLinkSection::GnuDebugLinkSection(StringRef File, |
| 1162 | uint32_t PrecomputedCRC) |
| 1163 | : FileName(File), CRC32(PrecomputedCRC) { |
| 1164 | init(File); |
| 1165 | } |
| 1166 | |
| 1167 | template <class ELFT> |
| 1168 | Error ELFSectionWriter<ELFT>::visit(const GnuDebugLinkSection &Sec) { |
| 1169 | unsigned char *Buf = |
| 1170 | reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset; |
| 1171 | Elf_Word *CRC = |
| 1172 | reinterpret_cast<Elf_Word *>(Buf + Sec.Size - sizeof(Elf_Word)); |
| 1173 | *CRC = Sec.CRC32; |
| 1174 | llvm::copy(Sec.FileName, Buf); |
| 1175 | return Error::success(); |
| 1176 | } |
| 1177 | |
| 1178 | Error GnuDebugLinkSection::accept(SectionVisitor &Visitor) const { |
| 1179 | return Visitor.visit(*this); |
| 1180 | } |
| 1181 | |
| 1182 | Error GnuDebugLinkSection::accept(MutableSectionVisitor &Visitor) { |
| 1183 | return Visitor.visit(*this); |
| 1184 | } |
| 1185 | |
| 1186 | template <class ELFT> |
| 1187 | Error ELFSectionWriter<ELFT>::visit(const GroupSection &Sec) { |
| 1188 | ELF::Elf32_Word *Buf = |
| 1189 | reinterpret_cast<ELF::Elf32_Word *>(Out.getBufferStart() + Sec.Offset); |
| 1190 | support::endian::write32<ELFT::TargetEndianness>(Buf++, Sec.FlagWord); |
| 1191 | for (SectionBase *S : Sec.GroupMembers) |
| 1192 | support::endian::write32<ELFT::TargetEndianness>(Buf++, S->Index); |
| 1193 | return Error::success(); |
| 1194 | } |
| 1195 | |
| 1196 | Error GroupSection::accept(SectionVisitor &Visitor) const { |
| 1197 | return Visitor.visit(*this); |
| 1198 | } |
| 1199 | |
| 1200 | Error GroupSection::accept(MutableSectionVisitor &Visitor) { |
| 1201 | return Visitor.visit(*this); |
| 1202 | } |
| 1203 | |
| 1204 | // Returns true IFF a section is wholly inside the range of a segment |
| 1205 | static bool sectionWithinSegment(const SectionBase &Sec, const Segment &Seg) { |
| 1206 | // If a section is empty it should be treated like it has a size of 1. This is |
| 1207 | // to clarify the case when an empty section lies on a boundary between two |
| 1208 | // segments and ensures that the section "belongs" to the second segment and |
| 1209 | // not the first. |
| 1210 | uint64_t SecSize = Sec.Size ? Sec.Size : 1; |
| 1211 | |
| 1212 | // Ignore just added sections. |
| 1213 | if (Sec.OriginalOffset == std::numeric_limits<uint64_t>::max()) |
| 1214 | return false; |
| 1215 | |
| 1216 | if (Sec.Type == SHT_NOBITS) { |
| 1217 | if (!(Sec.Flags & SHF_ALLOC)) |
| 1218 | return false; |
| 1219 | |
| 1220 | bool SectionIsTLS = Sec.Flags & SHF_TLS; |
| 1221 | bool SegmentIsTLS = Seg.Type == PT_TLS; |
| 1222 | if (SectionIsTLS != SegmentIsTLS) |
| 1223 | return false; |
| 1224 | |
| 1225 | return Seg.VAddr <= Sec.Addr && |
| 1226 | Seg.VAddr + Seg.MemSize >= Sec.Addr + SecSize; |
| 1227 | } |
| 1228 | |
| 1229 | return Seg.Offset <= Sec.OriginalOffset && |
| 1230 | Seg.Offset + Seg.FileSize >= Sec.OriginalOffset + SecSize; |
| 1231 | } |
| 1232 | |
| 1233 | // Returns true IFF a segment's original offset is inside of another segment's |
| 1234 | // range. |
| 1235 | static bool segmentOverlapsSegment(const Segment &Child, |
| 1236 | const Segment &Parent) { |
| 1237 | |
| 1238 | return Parent.OriginalOffset <= Child.OriginalOffset && |
| 1239 | Parent.OriginalOffset + Parent.FileSize > Child.OriginalOffset; |
| 1240 | } |
| 1241 | |
| 1242 | static bool compareSegmentsByOffset(const Segment *A, const Segment *B) { |
| 1243 | // Any segment without a parent segment should come before a segment |
| 1244 | // that has a parent segment. |
| 1245 | if (A->OriginalOffset < B->OriginalOffset) |
| 1246 | return true; |
| 1247 | if (A->OriginalOffset > B->OriginalOffset) |
| 1248 | return false; |
| 1249 | return A->Index < B->Index; |
| 1250 | } |
| 1251 | |
| 1252 | void BasicELFBuilder::initFileHeader() { |
| 1253 | Obj->Flags = 0x0; |
| 1254 | Obj->Type = ET_REL; |
| 1255 | Obj->OSABI = ELFOSABI_NONE; |
| 1256 | Obj->ABIVersion = 0; |
| 1257 | Obj->Entry = 0x0; |
| 1258 | Obj->Machine = EM_NONE; |
| 1259 | Obj->Version = 1; |
| 1260 | } |
| 1261 | |
| 1262 | void BasicELFBuilder::initHeaderSegment() { Obj->ElfHdrSegment.Index = 0; } |
| 1263 | |
| 1264 | StringTableSection *BasicELFBuilder::addStrTab() { |
| 1265 | auto &StrTab = Obj->addSection<StringTableSection>(); |
| 1266 | StrTab.Name = ".strtab"; |
| 1267 | |
| 1268 | Obj->SectionNames = &StrTab; |
| 1269 | return &StrTab; |
| 1270 | } |
| 1271 | |
| 1272 | SymbolTableSection *BasicELFBuilder::addSymTab(StringTableSection *StrTab) { |
| 1273 | auto &SymTab = Obj->addSection<SymbolTableSection>(); |
| 1274 | |
| 1275 | SymTab.Name = ".symtab"; |
| 1276 | SymTab.Link = StrTab->Index; |
| 1277 | |
| 1278 | // The symbol table always needs a null symbol |
| 1279 | SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0); |
| 1280 | |
| 1281 | Obj->SymbolTable = &SymTab; |
| 1282 | return &SymTab; |
| 1283 | } |
| 1284 | |
| 1285 | Error BasicELFBuilder::initSections() { |
| 1286 | for (SectionBase &Sec : Obj->sections()) |
| 1287 | if (Error Err = Sec.initialize(Obj->sections())) |
| 1288 | return Err; |
| 1289 | |
| 1290 | return Error::success(); |
| 1291 | } |
| 1292 | |
| 1293 | void BinaryELFBuilder::addData(SymbolTableSection *SymTab) { |
| 1294 | auto Data = ArrayRef<uint8_t>( |
| 1295 | reinterpret_cast<const uint8_t *>(MemBuf->getBufferStart()), |
| 1296 | MemBuf->getBufferSize()); |
| 1297 | auto &DataSection = Obj->addSection<Section>(Data); |
| 1298 | DataSection.Name = ".data"; |
| 1299 | DataSection.Type = ELF::SHT_PROGBITS; |
| 1300 | DataSection.Size = Data.size(); |
| 1301 | DataSection.Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE; |
| 1302 | |
| 1303 | std::string SanitizedFilename = MemBuf->getBufferIdentifier().str(); |
| 1304 | std::replace_if( |
| 1305 | std::begin(SanitizedFilename), std::end(SanitizedFilename), |
| 1306 | [](char C) { return !isAlnum(C); }, '_'); |
| 1307 | Twine Prefix = Twine("_binary_") + SanitizedFilename; |
| 1308 | |
| 1309 | SymTab->addSymbol(Prefix + "_start", STB_GLOBAL, STT_NOTYPE, &DataSection, |
| 1310 | /*Value=*/0, NewSymbolVisibility, 0, 0); |
| 1311 | SymTab->addSymbol(Prefix + "_end", STB_GLOBAL, STT_NOTYPE, &DataSection, |
| 1312 | /*Value=*/DataSection.Size, NewSymbolVisibility, 0, 0); |
| 1313 | SymTab->addSymbol(Prefix + "_size", STB_GLOBAL, STT_NOTYPE, nullptr, |
| 1314 | /*Value=*/DataSection.Size, NewSymbolVisibility, SHN_ABS, |
| 1315 | 0); |
| 1316 | } |
| 1317 | |
| 1318 | Expected<std::unique_ptr<Object>> BinaryELFBuilder::build() { |
| 1319 | initFileHeader(); |
| 1320 | initHeaderSegment(); |
| 1321 | |
| 1322 | SymbolTableSection *SymTab = addSymTab(addStrTab()); |
| 1323 | if (Error Err = initSections()) |
| 1324 | return std::move(Err); |
| 1325 | addData(SymTab); |
| 1326 | |
| 1327 | return std::move(Obj); |
| 1328 | } |
| 1329 | |
| 1330 | // Adds sections from IHEX data file. Data should have been |
| 1331 | // fully validated by this time. |
| 1332 | void IHexELFBuilder::addDataSections() { |
| 1333 | OwnedDataSection *Section = nullptr; |
| 1334 | uint64_t SegmentAddr = 0, BaseAddr = 0; |
| 1335 | uint32_t SecNo = 1; |
| 1336 | |
| 1337 | for (const IHexRecord &R : Records) { |
| 1338 | uint64_t RecAddr; |
| 1339 | switch (R.Type) { |
| 1340 | case IHexRecord::Data: |
| 1341 | // Ignore empty data records |
| 1342 | if (R.HexData.empty()) |
| 1343 | continue; |
| 1344 | RecAddr = R.Addr + SegmentAddr + BaseAddr; |
| 1345 | if (!Section || Section->Addr + Section->Size != RecAddr) |
| 1346 | // OriginalOffset field is only used to sort section properly, so |
| 1347 | // instead of keeping track of real offset in IHEX file, we use |
| 1348 | // section number. |
| 1349 | Section = &Obj->addSection<OwnedDataSection>( |
| 1350 | ".sec" + std::to_string(SecNo++), RecAddr, |
| 1351 | ELF::SHF_ALLOC | ELF::SHF_WRITE, SecNo); |
| 1352 | Section->appendHexData(R.HexData); |
| 1353 | break; |
| 1354 | case IHexRecord::EndOfFile: |
| 1355 | break; |
| 1356 | case IHexRecord::SegmentAddr: |
| 1357 | // 20-bit segment address. |
| 1358 | SegmentAddr = checkedGetHex<uint16_t>(R.HexData) << 4; |
| 1359 | break; |
| 1360 | case IHexRecord::StartAddr80x86: |
| 1361 | case IHexRecord::StartAddr: |
| 1362 | Obj->Entry = checkedGetHex<uint32_t>(R.HexData); |
| 1363 | assert(Obj->Entry <= 0xFFFFFU)((void)0); |
| 1364 | break; |
| 1365 | case IHexRecord::ExtendedAddr: |
| 1366 | // 16-31 bits of linear base address |
| 1367 | BaseAddr = checkedGetHex<uint16_t>(R.HexData) << 16; |
| 1368 | break; |
| 1369 | default: |
| 1370 | llvm_unreachable("unknown record type")__builtin_unreachable(); |
| 1371 | } |
| 1372 | } |
| 1373 | } |
| 1374 | |
| 1375 | Expected<std::unique_ptr<Object>> IHexELFBuilder::build() { |
| 1376 | initFileHeader(); |
| 1377 | initHeaderSegment(); |
| 1378 | StringTableSection *StrTab = addStrTab(); |
| 1379 | addSymTab(StrTab); |
| 1380 | if (Error Err = initSections()) |
| 1381 | return std::move(Err); |
| 1382 | addDataSections(); |
| 1383 | |
| 1384 | return std::move(Obj); |
| 1385 | } |
| 1386 | |
| 1387 | template <class ELFT> void ELFBuilder<ELFT>::setParentSegment(Segment &Child) { |
| 1388 | for (Segment &Parent : Obj.segments()) { |
| 1389 | // Every segment will overlap with itself but we don't want a segment to |
| 1390 | // be its own parent so we avoid that situation. |
| 1391 | if (&Child != &Parent && segmentOverlapsSegment(Child, Parent)) { |
| 1392 | // We want a canonical "most parental" segment but this requires |
| 1393 | // inspecting the ParentSegment. |
| 1394 | if (compareSegmentsByOffset(&Parent, &Child)) |
| 1395 | if (Child.ParentSegment == nullptr || |
| 1396 | compareSegmentsByOffset(&Parent, Child.ParentSegment)) { |
| 1397 | Child.ParentSegment = &Parent; |
| 1398 | } |
| 1399 | } |
| 1400 | } |
| 1401 | } |
| 1402 | |
| 1403 | template <class ELFT> Error ELFBuilder<ELFT>::findEhdrOffset() { |
| 1404 | if (!ExtractPartition) |
| 1405 | return Error::success(); |
| 1406 | |
| 1407 | for (const SectionBase &Sec : Obj.sections()) { |
| 1408 | if (Sec.Type == SHT_LLVM_PART_EHDR && Sec.Name == *ExtractPartition) { |
| 1409 | EhdrOffset = Sec.Offset; |
| 1410 | return Error::success(); |
| 1411 | } |
| 1412 | } |
| 1413 | return createStringError(errc::invalid_argument, |
| 1414 | "could not find partition named '" + |
| 1415 | *ExtractPartition + "'"); |
| 1416 | } |
| 1417 | |
| 1418 | template <class ELFT> |
| 1419 | Error ELFBuilder<ELFT>::readProgramHeaders(const ELFFile<ELFT> &HeadersFile) { |
| 1420 | uint32_t Index = 0; |
| 1421 | |
| 1422 | Expected<typename ELFFile<ELFT>::Elf_Phdr_Range> Headers = |
| 1423 | HeadersFile.program_headers(); |
| 1424 | if (!Headers) |
| 1425 | return Headers.takeError(); |
| 1426 | |
| 1427 | for (const typename ELFFile<ELFT>::Elf_Phdr &Phdr : *Headers) { |
| 1428 | if (Phdr.p_offset + Phdr.p_filesz > HeadersFile.getBufSize()) |
| 1429 | return createStringError( |
| 1430 | errc::invalid_argument, |
| 1431 | "program header with offset 0x" + Twine::utohexstr(Phdr.p_offset) + |
| 1432 | " and file size 0x" + Twine::utohexstr(Phdr.p_filesz) + |
| 1433 | " goes past the end of the file"); |
| 1434 | |
| 1435 | ArrayRef<uint8_t> Data{HeadersFile.base() + Phdr.p_offset, |
| 1436 | (size_t)Phdr.p_filesz}; |
| 1437 | Segment &Seg = Obj.addSegment(Data); |
| 1438 | Seg.Type = Phdr.p_type; |
| 1439 | Seg.Flags = Phdr.p_flags; |
| 1440 | Seg.OriginalOffset = Phdr.p_offset + EhdrOffset; |
| 1441 | Seg.Offset = Phdr.p_offset + EhdrOffset; |
| 1442 | Seg.VAddr = Phdr.p_vaddr; |
| 1443 | Seg.PAddr = Phdr.p_paddr; |
| 1444 | Seg.FileSize = Phdr.p_filesz; |
| 1445 | Seg.MemSize = Phdr.p_memsz; |
| 1446 | Seg.Align = Phdr.p_align; |
| 1447 | Seg.Index = Index++; |
| 1448 | for (SectionBase &Sec : Obj.sections()) |
| 1449 | if (sectionWithinSegment(Sec, Seg)) { |
| 1450 | Seg.addSection(&Sec); |
| 1451 | if (!Sec.ParentSegment || Sec.ParentSegment->Offset > Seg.Offset) |
| 1452 | Sec.ParentSegment = &Seg; |
| 1453 | } |
| 1454 | } |
| 1455 | |
| 1456 | auto &ElfHdr = Obj.ElfHdrSegment; |
| 1457 | ElfHdr.Index = Index++; |
| 1458 | ElfHdr.OriginalOffset = ElfHdr.Offset = EhdrOffset; |
| 1459 | |
| 1460 | const typename ELFT::Ehdr &Ehdr = HeadersFile.getHeader(); |
| 1461 | auto &PrHdr = Obj.ProgramHdrSegment; |
| 1462 | PrHdr.Type = PT_PHDR; |
| 1463 | PrHdr.Flags = 0; |
| 1464 | // The spec requires us to have p_vaddr % p_align == p_offset % p_align. |
| 1465 | // Whereas this works automatically for ElfHdr, here OriginalOffset is |
| 1466 | // always non-zero and to ensure the equation we assign the same value to |
| 1467 | // VAddr as well. |
| 1468 | PrHdr.OriginalOffset = PrHdr.Offset = PrHdr.VAddr = EhdrOffset + Ehdr.e_phoff; |
| 1469 | PrHdr.PAddr = 0; |
| 1470 | PrHdr.FileSize = PrHdr.MemSize = Ehdr.e_phentsize * Ehdr.e_phnum; |
| 1471 | // The spec requires us to naturally align all the fields. |
| 1472 | PrHdr.Align = sizeof(Elf_Addr); |
| 1473 | PrHdr.Index = Index++; |
| 1474 | |
| 1475 | // Now we do an O(n^2) loop through the segments in order to match up |
| 1476 | // segments. |
| 1477 | for (Segment &Child : Obj.segments()) |
| 1478 | setParentSegment(Child); |
| 1479 | setParentSegment(ElfHdr); |
| 1480 | setParentSegment(PrHdr); |
| 1481 | |
| 1482 | return Error::success(); |
| 1483 | } |
| 1484 | |
| 1485 | template <class ELFT> |
| 1486 | Error ELFBuilder<ELFT>::initGroupSection(GroupSection *GroupSec) { |
| 1487 | if (GroupSec->Align % sizeof(ELF::Elf32_Word) != 0) |
| 1488 | return createStringError(errc::invalid_argument, |
| 1489 | "invalid alignment " + Twine(GroupSec->Align) + |
| 1490 | " of group section '" + GroupSec->Name + "'"); |
| 1491 | SectionTableRef SecTable = Obj.sections(); |
| 1492 | if (GroupSec->Link != SHN_UNDEF) { |
| 1493 | auto SymTab = SecTable.template getSectionOfType<SymbolTableSection>( |
| 1494 | GroupSec->Link, |
| 1495 | "link field value '" + Twine(GroupSec->Link) + "' in section '" + |
| 1496 | GroupSec->Name + "' is invalid", |
| 1497 | "link field value '" + Twine(GroupSec->Link) + "' in section '" + |
| 1498 | GroupSec->Name + "' is not a symbol table"); |
| 1499 | if (!SymTab) |
| 1500 | return SymTab.takeError(); |
| 1501 | |
| 1502 | Expected<Symbol *> Sym = (*SymTab)->getSymbolByIndex(GroupSec->Info); |
| 1503 | if (!Sym) |
| 1504 | return createStringError(errc::invalid_argument, |
| 1505 | "info field value '" + Twine(GroupSec->Info) + |
| 1506 | "' in section '" + GroupSec->Name + |
| 1507 | "' is not a valid symbol index"); |
| 1508 | GroupSec->setSymTab(*SymTab); |
| 1509 | GroupSec->setSymbol(*Sym); |
| 1510 | } |
| 1511 | if (GroupSec->Contents.size() % sizeof(ELF::Elf32_Word) || |
| 1512 | GroupSec->Contents.empty()) |
| 1513 | return createStringError(errc::invalid_argument, |
| 1514 | "the content of the section " + GroupSec->Name + |
| 1515 | " is malformed"); |
| 1516 | const ELF::Elf32_Word *Word = |
| 1517 | reinterpret_cast<const ELF::Elf32_Word *>(GroupSec->Contents.data()); |
| 1518 | const ELF::Elf32_Word *End = |
| 1519 | Word + GroupSec->Contents.size() / sizeof(ELF::Elf32_Word); |
| 1520 | GroupSec->setFlagWord( |
| 1521 | support::endian::read32<ELFT::TargetEndianness>(Word++)); |
| 1522 | for (; Word != End; ++Word) { |
| 1523 | uint32_t Index = support::endian::read32<ELFT::TargetEndianness>(Word); |
| 1524 | Expected<SectionBase *> Sec = SecTable.getSection( |
| 1525 | Index, "group member index " + Twine(Index) + " in section '" + |
| 1526 | GroupSec->Name + "' is invalid"); |
| 1527 | if (!Sec) |
| 1528 | return Sec.takeError(); |
| 1529 | |
| 1530 | GroupSec->addMember(*Sec); |
| 1531 | } |
| 1532 | |
| 1533 | return Error::success(); |
| 1534 | } |
| 1535 | |
| 1536 | template <class ELFT> |
| 1537 | Error ELFBuilder<ELFT>::initSymbolTable(SymbolTableSection *SymTab) { |
| 1538 | Expected<const Elf_Shdr *> Shdr = ElfFile.getSection(SymTab->Index); |
| 1539 | if (!Shdr) |
| 1540 | return Shdr.takeError(); |
| 1541 | |
| 1542 | Expected<StringRef> StrTabData = ElfFile.getStringTableForSymtab(**Shdr); |
| 1543 | if (!StrTabData) |
| 1544 | return StrTabData.takeError(); |
| 1545 | |
| 1546 | ArrayRef<Elf_Word> ShndxData; |
| 1547 | |
| 1548 | Expected<typename ELFFile<ELFT>::Elf_Sym_Range> Symbols = |
| 1549 | ElfFile.symbols(*Shdr); |
| 1550 | if (!Symbols) |
| 1551 | return Symbols.takeError(); |
| 1552 | |
| 1553 | for (const typename ELFFile<ELFT>::Elf_Sym &Sym : *Symbols) { |
| 1554 | SectionBase *DefSection = nullptr; |
| 1555 | |
| 1556 | Expected<StringRef> Name = Sym.getName(*StrTabData); |
| 1557 | if (!Name) |
| 1558 | return Name.takeError(); |
| 1559 | |
| 1560 | if (Sym.st_shndx == SHN_XINDEX) { |
| 1561 | if (SymTab->getShndxTable() == nullptr) |
| 1562 | return createStringError(errc::invalid_argument, |
| 1563 | "symbol '" + *Name + |
| 1564 | "' has index SHN_XINDEX but no " |
| 1565 | "SHT_SYMTAB_SHNDX section exists"); |
| 1566 | if (ShndxData.data() == nullptr) { |
| 1567 | Expected<const Elf_Shdr *> ShndxSec = |
| 1568 | ElfFile.getSection(SymTab->getShndxTable()->Index); |
| 1569 | if (!ShndxSec) |
| 1570 | return ShndxSec.takeError(); |
| 1571 | |
| 1572 | Expected<ArrayRef<Elf_Word>> Data = |
| 1573 | ElfFile.template getSectionContentsAsArray<Elf_Word>(**ShndxSec); |
| 1574 | if (!Data) |
| 1575 | return Data.takeError(); |
| 1576 | |
| 1577 | ShndxData = *Data; |
| 1578 | if (ShndxData.size() != Symbols->size()) |
| 1579 | return createStringError( |
| 1580 | errc::invalid_argument, |
| 1581 | "symbol section index table does not have the same number of " |
| 1582 | "entries as the symbol table"); |
| 1583 | } |
| 1584 | Elf_Word Index = ShndxData[&Sym - Symbols->begin()]; |
| 1585 | Expected<SectionBase *> Sec = Obj.sections().getSection( |
| 1586 | Index, |
| 1587 | "symbol '" + *Name + "' has invalid section index " + Twine(Index)); |
| 1588 | if (!Sec) |
| 1589 | return Sec.takeError(); |
| 1590 | |
| 1591 | DefSection = *Sec; |
| 1592 | } else if (Sym.st_shndx >= SHN_LORESERVE) { |
| 1593 | if (!isValidReservedSectionIndex(Sym.st_shndx, Obj.Machine)) { |
| 1594 | return createStringError( |
| 1595 | errc::invalid_argument, |
| 1596 | "symbol '" + *Name + |
| 1597 | "' has unsupported value greater than or equal " |
| 1598 | "to SHN_LORESERVE: " + |
| 1599 | Twine(Sym.st_shndx)); |
| 1600 | } |
| 1601 | } else if (Sym.st_shndx != SHN_UNDEF) { |
| 1602 | Expected<SectionBase *> Sec = Obj.sections().getSection( |
| 1603 | Sym.st_shndx, "symbol '" + *Name + |
| 1604 | "' is defined has invalid section index " + |
| 1605 | Twine(Sym.st_shndx)); |
| 1606 | if (!Sec) |
| 1607 | return Sec.takeError(); |
| 1608 | |
| 1609 | DefSection = *Sec; |
| 1610 | } |
| 1611 | |
| 1612 | SymTab->addSymbol(*Name, Sym.getBinding(), Sym.getType(), DefSection, |
| 1613 | Sym.getValue(), Sym.st_other, Sym.st_shndx, Sym.st_size); |
| 1614 | } |
| 1615 | |
| 1616 | return Error::success(); |
| 1617 | } |
| 1618 | |
| 1619 | template <class ELFT> |
| 1620 | static void getAddend(uint64_t &, const Elf_Rel_Impl<ELFT, false> &) {} |
| 1621 | |
| 1622 | template <class ELFT> |
| 1623 | static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, true> &Rela) { |
| 1624 | ToSet = Rela.r_addend; |
| 1625 | } |
| 1626 | |
| 1627 | template <class T> |
| 1628 | static Error initRelocations(RelocationSection *Relocs, |
| 1629 | SymbolTableSection *SymbolTable, T RelRange) { |
| 1630 | for (const auto &Rel : RelRange) { |
| 1631 | Relocation ToAdd; |
| 1632 | ToAdd.Offset = Rel.r_offset; |
| 1633 | getAddend(ToAdd.Addend, Rel); |
| 1634 | ToAdd.Type = Rel.getType(false); |
| 1635 | |
| 1636 | if (uint32_t Sym = Rel.getSymbol(false)) { |
| 1637 | if (!SymbolTable) |
| 1638 | return createStringError( |
| 1639 | errc::invalid_argument, |
| 1640 | "'" + Relocs->Name + "': relocation references symbol with index " + |
| 1641 | Twine(Sym) + ", but there is no symbol table"); |
| 1642 | Expected<Symbol *> SymByIndex = SymbolTable->getSymbolByIndex(Sym); |
| 1643 | if (!SymByIndex) |
| 1644 | return SymByIndex.takeError(); |
| 1645 | |
| 1646 | ToAdd.RelocSymbol = *SymByIndex; |
| 1647 | } |
| 1648 | |
| 1649 | Relocs->addRelocation(ToAdd); |
| 1650 | } |
| 1651 | |
| 1652 | return Error::success(); |
| 1653 | } |
| 1654 | |
| 1655 | Expected<SectionBase *> SectionTableRef::getSection(uint32_t Index, |
| 1656 | Twine ErrMsg) { |
| 1657 | if (Index == SHN_UNDEF || Index > Sections.size()) |
| 1658 | return createStringError(errc::invalid_argument, ErrMsg); |
| 1659 | return Sections[Index - 1].get(); |
| 1660 | } |
| 1661 | |
| 1662 | template <class T> |
| 1663 | Expected<T *> SectionTableRef::getSectionOfType(uint32_t Index, |
| 1664 | Twine IndexErrMsg, |
| 1665 | Twine TypeErrMsg) { |
| 1666 | Expected<SectionBase *> BaseSec = getSection(Index, IndexErrMsg); |
| 1667 | if (!BaseSec) |
| 1668 | return BaseSec.takeError(); |
| 1669 | |
| 1670 | if (T *Sec = dyn_cast<T>(*BaseSec)) |
| 1671 | return Sec; |
| 1672 | |
| 1673 | return createStringError(errc::invalid_argument, TypeErrMsg); |
| 1674 | } |
| 1675 | |
| 1676 | template <class ELFT> |
| 1677 | Expected<SectionBase &> ELFBuilder<ELFT>::makeSection(const Elf_Shdr &Shdr) { |
| 1678 | switch (Shdr.sh_type) { |
| 1679 | case SHT_REL: |
| 1680 | case SHT_RELA: |
| 1681 | if (Shdr.sh_flags & SHF_ALLOC) { |
| 1682 | if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) |
| 1683 | return Obj.addSection<DynamicRelocationSection>(*Data); |
| 1684 | else |
| 1685 | return Data.takeError(); |
| 1686 | } |
| 1687 | return Obj.addSection<RelocationSection>(); |
| 1688 | case SHT_STRTAB: |
| 1689 | // If a string table is allocated we don't want to mess with it. That would |
| 1690 | // mean altering the memory image. There are no special link types or |
| 1691 | // anything so we can just use a Section. |
| 1692 | if (Shdr.sh_flags & SHF_ALLOC) { |
| 1693 | if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) |
| 1694 | return Obj.addSection<Section>(*Data); |
| 1695 | else |
| 1696 | return Data.takeError(); |
| 1697 | } |
| 1698 | return Obj.addSection<StringTableSection>(); |
| 1699 | case SHT_HASH: |
| 1700 | case SHT_GNU_HASH: |
| 1701 | // Hash tables should refer to SHT_DYNSYM which we're not going to change. |
| 1702 | // Because of this we don't need to mess with the hash tables either. |
| 1703 | if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) |
| 1704 | return Obj.addSection<Section>(*Data); |
| 1705 | else |
| 1706 | return Data.takeError(); |
| 1707 | case SHT_GROUP: |
| 1708 | if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) |
| 1709 | return Obj.addSection<GroupSection>(*Data); |
| 1710 | else |
| 1711 | return Data.takeError(); |
| 1712 | case SHT_DYNSYM: |
| 1713 | if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) |
| 1714 | return Obj.addSection<DynamicSymbolTableSection>(*Data); |
| 1715 | else |
| 1716 | return Data.takeError(); |
| 1717 | case SHT_DYNAMIC: |
| 1718 | if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr)) |
| 1719 | return Obj.addSection<DynamicSection>(*Data); |
| 1720 | else |
| 1721 | return Data.takeError(); |
| 1722 | case SHT_SYMTAB: { |
| 1723 | auto &SymTab = Obj.addSection<SymbolTableSection>(); |
| 1724 | Obj.SymbolTable = &SymTab; |
| 1725 | return SymTab; |
| 1726 | } |
| 1727 | case SHT_SYMTAB_SHNDX: { |
| 1728 | auto &ShndxSection = Obj.addSection<SectionIndexSection>(); |
| 1729 | Obj.SectionIndexTable = &ShndxSection; |
| 1730 | return ShndxSection; |
| 1731 | } |
| 1732 | case SHT_NOBITS: |
| 1733 | return Obj.addSection<Section>(ArrayRef<uint8_t>()); |
| 1734 | default: { |
| 1735 | Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr); |
| 1736 | if (!Data) |
| 1737 | return Data.takeError(); |
| 1738 | |
| 1739 | Expected<StringRef> Name = ElfFile.getSectionName(Shdr); |
| 1740 | if (!Name) |
| 1741 | return Name.takeError(); |
| 1742 | |
| 1743 | if (Name->startswith(".zdebug") || (Shdr.sh_flags & ELF::SHF_COMPRESSED)) { |
| 1744 | uint64_t DecompressedSize, DecompressedAlign; |
| 1745 | std::tie(DecompressedSize, DecompressedAlign) = |
| 1746 | getDecompressedSizeAndAlignment<ELFT>(*Data); |
| 1747 | Expected<CompressedSection> NewSection = |
| 1748 | CompressedSection::create(*Data, DecompressedSize, DecompressedAlign); |
| 1749 | if (!NewSection) |
| 1750 | return NewSection.takeError(); |
| 1751 | |
| 1752 | return Obj.addSection<CompressedSection>(std::move(*NewSection)); |
| 1753 | } |
| 1754 | |
| 1755 | return Obj.addSection<Section>(*Data); |
| 1756 | } |
| 1757 | } |
| 1758 | } |
| 1759 | |
| 1760 | template <class ELFT> Error ELFBuilder<ELFT>::readSectionHeaders() { |
| 1761 | uint32_t Index = 0; |
| 1762 | Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections = |
| 1763 | ElfFile.sections(); |
| 1764 | if (!Sections) |
| 1765 | return Sections.takeError(); |
| 1766 | |
| 1767 | for (const typename ELFFile<ELFT>::Elf_Shdr &Shdr : *Sections) { |
| 1768 | if (Index == 0) { |
| 1769 | ++Index; |
| 1770 | continue; |
| 1771 | } |
| 1772 | Expected<SectionBase &> Sec = makeSection(Shdr); |
| 1773 | if (!Sec) |
| 1774 | return Sec.takeError(); |
| 1775 | |
| 1776 | Expected<StringRef> SecName = ElfFile.getSectionName(Shdr); |
| 1777 | if (!SecName) |
| 1778 | return SecName.takeError(); |
| 1779 | Sec->Name = SecName->str(); |
| 1780 | Sec->Type = Sec->OriginalType = Shdr.sh_type; |
| 1781 | Sec->Flags = Sec->OriginalFlags = Shdr.sh_flags; |
| 1782 | Sec->Addr = Shdr.sh_addr; |
| 1783 | Sec->Offset = Shdr.sh_offset; |
| 1784 | Sec->OriginalOffset = Shdr.sh_offset; |
| 1785 | Sec->Size = Shdr.sh_size; |
| 1786 | Sec->Link = Shdr.sh_link; |
| 1787 | Sec->Info = Shdr.sh_info; |
| 1788 | Sec->Align = Shdr.sh_addralign; |
| 1789 | Sec->EntrySize = Shdr.sh_entsize; |
| 1790 | Sec->Index = Index++; |
| 1791 | Sec->OriginalIndex = Sec->Index; |
| 1792 | Sec->OriginalData = |
| 1793 | ArrayRef<uint8_t>(ElfFile.base() + Shdr.sh_offset, |
| 1794 | (Shdr.sh_type == SHT_NOBITS) ? (size_t)0 : Shdr.sh_size); |
| 1795 | } |
| 1796 | |
| 1797 | return Error::success(); |
| 1798 | } |
| 1799 | |
| 1800 | template <class ELFT> Error ELFBuilder<ELFT>::readSections(bool EnsureSymtab) { |
| 1801 | uint32_t ShstrIndex = ElfFile.getHeader().e_shstrndx; |
| 1802 | if (ShstrIndex == SHN_XINDEX) { |
| 1803 | Expected<const Elf_Shdr *> Sec = ElfFile.getSection(0); |
| 1804 | if (!Sec) |
| 1805 | return Sec.takeError(); |
| 1806 | |
| 1807 | ShstrIndex = (*Sec)->sh_link; |
| 1808 | } |
| 1809 | |
| 1810 | if (ShstrIndex == SHN_UNDEF) |
| 1811 | Obj.HadShdrs = false; |
| 1812 | else { |
| 1813 | Expected<StringTableSection *> Sec = |
| 1814 | Obj.sections().template getSectionOfType<StringTableSection>( |
| 1815 | ShstrIndex, |
| 1816 | "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " + |
| 1817 | " is invalid", |
| 1818 | "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " + |
| 1819 | " does not reference a string table"); |
| 1820 | if (!Sec) |
| 1821 | return Sec.takeError(); |
| 1822 | |
| 1823 | Obj.SectionNames = *Sec; |
| 1824 | } |
| 1825 | |
| 1826 | // If a section index table exists we'll need to initialize it before we |
| 1827 | // initialize the symbol table because the symbol table might need to |
| 1828 | // reference it. |
| 1829 | if (Obj.SectionIndexTable) |
| 1830 | if (Error Err = Obj.SectionIndexTable->initialize(Obj.sections())) |
| 1831 | return Err; |
| 1832 | |
| 1833 | // Now that all of the sections have been added we can fill out some extra |
| 1834 | // details about symbol tables. We need the symbol table filled out before |
| 1835 | // any relocations. |
| 1836 | if (Obj.SymbolTable) { |
| 1837 | if (Error Err = Obj.SymbolTable->initialize(Obj.sections())) |
| 1838 | return Err; |
| 1839 | if (Error Err = initSymbolTable(Obj.SymbolTable)) |
| 1840 | return Err; |
| 1841 | } else if (EnsureSymtab) { |
| 1842 | if (Error Err = Obj.addNewSymbolTable()) |
| 1843 | return Err; |
| 1844 | } |
| 1845 | |
| 1846 | // Now that all sections and symbols have been added we can add |
| 1847 | // relocations that reference symbols and set the link and info fields for |
| 1848 | // relocation sections. |
| 1849 | for (SectionBase &Sec : Obj.sections()) { |
| 1850 | if (&Sec == Obj.SymbolTable) |
| 1851 | continue; |
| 1852 | if (Error Err = Sec.initialize(Obj.sections())) |
| 1853 | return Err; |
| 1854 | if (auto RelSec = dyn_cast<RelocationSection>(&Sec)) { |
| 1855 | Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections = |
| 1856 | ElfFile.sections(); |
| 1857 | if (!Sections) |
| 1858 | return Sections.takeError(); |
| 1859 | |
| 1860 | const typename ELFFile<ELFT>::Elf_Shdr *Shdr = |
| 1861 | Sections->begin() + RelSec->Index; |
| 1862 | if (RelSec->Type == SHT_REL) { |
| 1863 | Expected<typename ELFFile<ELFT>::Elf_Rel_Range> Rels = |
| 1864 | ElfFile.rels(*Shdr); |
| 1865 | if (!Rels) |
| 1866 | return Rels.takeError(); |
| 1867 | |
| 1868 | if (Error Err = initRelocations(RelSec, Obj.SymbolTable, *Rels)) |
| 1869 | return Err; |
| 1870 | } else { |
| 1871 | Expected<typename ELFFile<ELFT>::Elf_Rela_Range> Relas = |
| 1872 | ElfFile.relas(*Shdr); |
| 1873 | if (!Relas) |
| 1874 | return Relas.takeError(); |
| 1875 | |
| 1876 | if (Error Err = initRelocations(RelSec, Obj.SymbolTable, *Relas)) |
| 1877 | return Err; |
| 1878 | } |
| 1879 | } else if (auto GroupSec = dyn_cast<GroupSection>(&Sec)) { |
| 1880 | if (Error Err = initGroupSection(GroupSec)) |
| 1881 | return Err; |
| 1882 | } |
| 1883 | } |
| 1884 | |
| 1885 | return Error::success(); |
| 1886 | } |
| 1887 | |
| 1888 | template <class ELFT> Error ELFBuilder<ELFT>::build(bool EnsureSymtab) { |
| 1889 | if (Error E = readSectionHeaders()) |
| 1890 | return E; |
| 1891 | if (Error E = findEhdrOffset()) |
| 1892 | return E; |
| 1893 | |
| 1894 | // The ELFFile whose ELF headers and program headers are copied into the |
| 1895 | // output file. Normally the same as ElfFile, but if we're extracting a |
| 1896 | // loadable partition it will point to the partition's headers. |
| 1897 | Expected<ELFFile<ELFT>> HeadersFile = ELFFile<ELFT>::create(toStringRef( |
| 1898 | {ElfFile.base() + EhdrOffset, ElfFile.getBufSize() - EhdrOffset})); |
| 1899 | if (!HeadersFile) |
| 1900 | return HeadersFile.takeError(); |
| 1901 | |
| 1902 | const typename ELFFile<ELFT>::Elf_Ehdr &Ehdr = HeadersFile->getHeader(); |
| 1903 | Obj.OSABI = Ehdr.e_ident[EI_OSABI]; |
| 1904 | Obj.ABIVersion = Ehdr.e_ident[EI_ABIVERSION]; |
| 1905 | Obj.Type = Ehdr.e_type; |
| 1906 | Obj.Machine = Ehdr.e_machine; |
| 1907 | Obj.Version = Ehdr.e_version; |
| 1908 | Obj.Entry = Ehdr.e_entry; |
| 1909 | Obj.Flags = Ehdr.e_flags; |
| 1910 | |
| 1911 | if (Error E = readSections(EnsureSymtab)) |
| 1912 | return E; |
| 1913 | return readProgramHeaders(*HeadersFile); |
| 1914 | } |
| 1915 | |
| 1916 | Writer::~Writer() {} |
| 1917 | |
| 1918 | Reader::~Reader() {} |
| 1919 | |
| 1920 | Expected<std::unique_ptr<Object>> |
| 1921 | BinaryReader::create(bool /*EnsureSymtab*/) const { |
| 1922 | return BinaryELFBuilder(MemBuf, NewSymbolVisibility).build(); |
| 1923 | } |
| 1924 | |
| 1925 | Expected<std::vector<IHexRecord>> IHexReader::parse() const { |
| 1926 | SmallVector<StringRef, 16> Lines; |
| 1927 | std::vector<IHexRecord> Records; |
| 1928 | bool HasSections = false; |
| 1929 | |
| 1930 | MemBuf->getBuffer().split(Lines, '\n'); |
| 1931 | Records.reserve(Lines.size()); |
| 1932 | for (size_t LineNo = 1; LineNo <= Lines.size(); ++LineNo) { |
| 1933 | StringRef Line = Lines[LineNo - 1].trim(); |
| 1934 | if (Line.empty()) |
| 1935 | continue; |
| 1936 | |
| 1937 | Expected<IHexRecord> R = IHexRecord::parse(Line); |
| 1938 | if (!R) |
| 1939 | return parseError(LineNo, R.takeError()); |
| 1940 | if (R->Type == IHexRecord::EndOfFile) |
| 1941 | break; |
| 1942 | HasSections |= (R->Type == IHexRecord::Data); |
| 1943 | Records.push_back(*R); |
| 1944 | } |
| 1945 | if (!HasSections) |
| 1946 | return parseError(-1U, "no sections"); |
| 1947 | |
| 1948 | return std::move(Records); |
| 1949 | } |
| 1950 | |
| 1951 | Expected<std::unique_ptr<Object>> |
| 1952 | IHexReader::create(bool /*EnsureSymtab*/) const { |
| 1953 | Expected<std::vector<IHexRecord>> Records = parse(); |
| 1954 | if (!Records) |
| 1955 | return Records.takeError(); |
| 1956 | |
| 1957 | return IHexELFBuilder(*Records).build(); |
| 1958 | } |
| 1959 | |
| 1960 | Expected<std::unique_ptr<Object>> ELFReader::create(bool EnsureSymtab) const { |
| 1961 | auto Obj = std::make_unique<Object>(); |
| 1962 | if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Bin)) { |
| 1963 | ELFBuilder<ELF32LE> Builder(*O, *Obj, ExtractPartition); |
| 1964 | if (Error Err = Builder.build(EnsureSymtab)) |
| 1965 | return std::move(Err); |
| 1966 | return std::move(Obj); |
| 1967 | } else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Bin)) { |
| 1968 | ELFBuilder<ELF64LE> Builder(*O, *Obj, ExtractPartition); |
| 1969 | if (Error Err = Builder.build(EnsureSymtab)) |
| 1970 | return std::move(Err); |
| 1971 | return std::move(Obj); |
| 1972 | } else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Bin)) { |
| 1973 | ELFBuilder<ELF32BE> Builder(*O, *Obj, ExtractPartition); |
| 1974 | if (Error Err = Builder.build(EnsureSymtab)) |
| 1975 | return std::move(Err); |
| 1976 | return std::move(Obj); |
| 1977 | } else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Bin)) { |
| 1978 | ELFBuilder<ELF64BE> Builder(*O, *Obj, ExtractPartition); |
| 1979 | if (Error Err = Builder.build(EnsureSymtab)) |
| 1980 | return std::move(Err); |
| 1981 | return std::move(Obj); |
| 1982 | } |
| 1983 | return createStringError(errc::invalid_argument, "invalid file type"); |
| 1984 | } |
| 1985 | |
| 1986 | template <class ELFT> void ELFWriter<ELFT>::writeEhdr() { |
| 1987 | Elf_Ehdr &Ehdr = *reinterpret_cast<Elf_Ehdr *>(Buf->getBufferStart()); |
| 1988 | std::fill(Ehdr.e_ident, Ehdr.e_ident + 16, 0); |
| 1989 | Ehdr.e_ident[EI_MAG0] = 0x7f; |
| 1990 | Ehdr.e_ident[EI_MAG1] = 'E'; |
| 1991 | Ehdr.e_ident[EI_MAG2] = 'L'; |
| 1992 | Ehdr.e_ident[EI_MAG3] = 'F'; |
| 1993 | Ehdr.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32; |
| 1994 | Ehdr.e_ident[EI_DATA] = |
| 1995 | ELFT::TargetEndianness == support::big ? ELFDATA2MSB : ELFDATA2LSB; |
| 1996 | Ehdr.e_ident[EI_VERSION] = EV_CURRENT; |
| 1997 | Ehdr.e_ident[EI_OSABI] = Obj.OSABI; |
| 1998 | Ehdr.e_ident[EI_ABIVERSION] = Obj.ABIVersion; |
| 1999 | |
| 2000 | Ehdr.e_type = Obj.Type; |
| 2001 | Ehdr.e_machine = Obj.Machine; |
| 2002 | Ehdr.e_version = Obj.Version; |
| 2003 | Ehdr.e_entry = Obj.Entry; |
| 2004 | // We have to use the fully-qualified name llvm::size |
| 2005 | // since some compilers complain on ambiguous resolution. |
| 2006 | Ehdr.e_phnum = llvm::size(Obj.segments()); |
| 2007 | Ehdr.e_phoff = (Ehdr.e_phnum != 0) ? Obj.ProgramHdrSegment.Offset : 0; |
| 2008 | Ehdr.e_phentsize = (Ehdr.e_phnum != 0) ? sizeof(Elf_Phdr) : 0; |
| 2009 | Ehdr.e_flags = Obj.Flags; |
| 2010 | Ehdr.e_ehsize = sizeof(Elf_Ehdr); |
| 2011 | if (WriteSectionHeaders && Obj.sections().size() != 0) { |
| 2012 | Ehdr.e_shentsize = sizeof(Elf_Shdr); |
| 2013 | Ehdr.e_shoff = Obj.SHOff; |
| 2014 | // """ |
| 2015 | // If the number of sections is greater than or equal to |
| 2016 | // SHN_LORESERVE (0xff00), this member has the value zero and the actual |
| 2017 | // number of section header table entries is contained in the sh_size field |
| 2018 | // of the section header at index 0. |
| 2019 | // """ |
| 2020 | auto Shnum = Obj.sections().size() + 1; |
| 2021 | if (Shnum >= SHN_LORESERVE) |
| 2022 | Ehdr.e_shnum = 0; |
| 2023 | else |
| 2024 | Ehdr.e_shnum = Shnum; |
| 2025 | // """ |
| 2026 | // If the section name string table section index is greater than or equal |
| 2027 | // to SHN_LORESERVE (0xff00), this member has the value SHN_XINDEX (0xffff) |
| 2028 | // and the actual index of the section name string table section is |
| 2029 | // contained in the sh_link field of the section header at index 0. |
| 2030 | // """ |
| 2031 | if (Obj.SectionNames->Index >= SHN_LORESERVE) |
| 2032 | Ehdr.e_shstrndx = SHN_XINDEX; |
| 2033 | else |
| 2034 | Ehdr.e_shstrndx = Obj.SectionNames->Index; |
| 2035 | } else { |
| 2036 | Ehdr.e_shentsize = 0; |
| 2037 | Ehdr.e_shoff = 0; |
| 2038 | Ehdr.e_shnum = 0; |
| 2039 | Ehdr.e_shstrndx = 0; |
| 2040 | } |
| 2041 | } |
| 2042 | |
| 2043 | template <class ELFT> void ELFWriter<ELFT>::writePhdrs() { |
| 2044 | for (auto &Seg : Obj.segments()) |
| 2045 | writePhdr(Seg); |
| 2046 | } |
| 2047 | |
| 2048 | template <class ELFT> void ELFWriter<ELFT>::writeShdrs() { |
| 2049 | // This reference serves to write the dummy section header at the begining |
| 2050 | // of the file. It is not used for anything else |
| 2051 | Elf_Shdr &Shdr = |
| 2052 | *reinterpret_cast<Elf_Shdr *>(Buf->getBufferStart() + Obj.SHOff); |
| 2053 | Shdr.sh_name = 0; |
| 2054 | Shdr.sh_type = SHT_NULL; |
| 2055 | Shdr.sh_flags = 0; |
| 2056 | Shdr.sh_addr = 0; |
| 2057 | Shdr.sh_offset = 0; |
| 2058 | // See writeEhdr for why we do this. |
| 2059 | uint64_t Shnum = Obj.sections().size() + 1; |
| 2060 | if (Shnum >= SHN_LORESERVE) |
| 2061 | Shdr.sh_size = Shnum; |
| 2062 | else |
| 2063 | Shdr.sh_size = 0; |
| 2064 | // See writeEhdr for why we do this. |
| 2065 | if (Obj.SectionNames != nullptr && Obj.SectionNames->Index >= SHN_LORESERVE) |
| 2066 | Shdr.sh_link = Obj.SectionNames->Index; |
| 2067 | else |
| 2068 | Shdr.sh_link = 0; |
| 2069 | Shdr.sh_info = 0; |
| 2070 | Shdr.sh_addralign = 0; |
| 2071 | Shdr.sh_entsize = 0; |
| 2072 | |
| 2073 | for (SectionBase &Sec : Obj.sections()) |
| 2074 | writeShdr(Sec); |
| 2075 | } |
| 2076 | |
| 2077 | template <class ELFT> Error ELFWriter<ELFT>::writeSectionData() { |
| 2078 | for (SectionBase &Sec : Obj.sections()) |
| 2079 | // Segments are responsible for writing their contents, so only write the |
| 2080 | // section data if the section is not in a segment. Note that this renders |
| 2081 | // sections in segments effectively immutable. |
| 2082 | if (Sec.ParentSegment == nullptr) |
| 2083 | if (Error Err = Sec.accept(*SecWriter)) |
| 2084 | return Err; |
| 2085 | |
| 2086 | return Error::success(); |
| 2087 | } |
| 2088 | |
| 2089 | template <class ELFT> void ELFWriter<ELFT>::writeSegmentData() { |
| 2090 | for (Segment &Seg : Obj.segments()) { |
| 2091 | size_t Size = std::min<size_t>(Seg.FileSize, Seg.getContents().size()); |
| 2092 | std::memcpy(Buf->getBufferStart() + Seg.Offset, Seg.getContents().data(), |
| 2093 | Size); |
| 2094 | } |
| 2095 | |
| 2096 | // Iterate over removed sections and overwrite their old data with zeroes. |
| 2097 | for (auto &Sec : Obj.removedSections()) { |
| 2098 | Segment *Parent = Sec.ParentSegment; |
| 2099 | if (Parent == nullptr || Sec.Type == SHT_NOBITS || Sec.Size == 0) |
| 2100 | continue; |
| 2101 | uint64_t Offset = |
| 2102 | Sec.OriginalOffset - Parent->OriginalOffset + Parent->Offset; |
| 2103 | std::memset(Buf->getBufferStart() + Offset, 0, Sec.Size); |
| 2104 | } |
| 2105 | } |
| 2106 | |
| 2107 | template <class ELFT> |
| 2108 | ELFWriter<ELFT>::ELFWriter(Object &Obj, raw_ostream &Buf, bool WSH, |
| 2109 | bool OnlyKeepDebug) |
| 2110 | : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs), |
| 2111 | OnlyKeepDebug(OnlyKeepDebug) {} |
| 2112 | |
| 2113 | Error Object::removeSections( |
| 2114 | bool AllowBrokenLinks, std::function<bool(const SectionBase &)> ToRemove) { |
| 2115 | |
| 2116 | auto Iter = std::stable_partition( |
| 2117 | std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) { |
| 2118 | if (ToRemove(*Sec)) |
| 2119 | return false; |
| 2120 | if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) { |
| 2121 | if (auto ToRelSec = RelSec->getSection()) |
| 2122 | return !ToRemove(*ToRelSec); |
| 2123 | } |
| 2124 | return true; |
| 2125 | }); |
| 2126 | if (SymbolTable != nullptr && ToRemove(*SymbolTable)) |
| 2127 | SymbolTable = nullptr; |
| 2128 | if (SectionNames != nullptr && ToRemove(*SectionNames)) |
| 2129 | SectionNames = nullptr; |
| 2130 | if (SectionIndexTable != nullptr && ToRemove(*SectionIndexTable)) |
| 2131 | SectionIndexTable = nullptr; |
| 2132 | // Now make sure there are no remaining references to the sections that will |
| 2133 | // be removed. Sometimes it is impossible to remove a reference so we emit |
| 2134 | // an error here instead. |
| 2135 | std::unordered_set<const SectionBase *> RemoveSections; |
| 2136 | RemoveSections.reserve(std::distance(Iter, std::end(Sections))); |
| 2137 | for (auto &RemoveSec : make_range(Iter, std::end(Sections))) { |
| 2138 | for (auto &Segment : Segments) |
| 2139 | Segment->removeSection(RemoveSec.get()); |
| 2140 | RemoveSec->onRemove(); |
| 2141 | RemoveSections.insert(RemoveSec.get()); |
| 2142 | } |
| 2143 | |
| 2144 | // For each section that remains alive, we want to remove the dead references. |
| 2145 | // This either might update the content of the section (e.g. remove symbols |
| 2146 | // from symbol table that belongs to removed section) or trigger an error if |
| 2147 | // a live section critically depends on a section being removed somehow |
| 2148 | // (e.g. the removed section is referenced by a relocation). |
| 2149 | for (auto &KeepSec : make_range(std::begin(Sections), Iter)) { |
| 2150 | if (Error E = KeepSec->removeSectionReferences( |
| 2151 | AllowBrokenLinks, [&RemoveSections](const SectionBase *Sec) { |
| 2152 | return RemoveSections.find(Sec) != RemoveSections.end(); |
| 2153 | })) |
| 2154 | return E; |
| 2155 | } |
| 2156 | |
| 2157 | // Transfer removed sections into the Object RemovedSections container for use |
| 2158 | // later. |
| 2159 | std::move(Iter, Sections.end(), std::back_inserter(RemovedSections)); |
| 2160 | // Now finally get rid of them all together. |
| 2161 | Sections.erase(Iter, std::end(Sections)); |
| 2162 | return Error::success(); |
| 2163 | } |
| 2164 | |
| 2165 | Error Object::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) { |
| 2166 | if (SymbolTable) |
| 2167 | for (const SecPtr &Sec : Sections) |
| 2168 | if (Error E = Sec->removeSymbols(ToRemove)) |
| 2169 | return E; |
| 2170 | return Error::success(); |
| 2171 | } |
| 2172 | |
| 2173 | Error Object::addNewSymbolTable() { |
| 2174 | assert(!SymbolTable && "Object must not has a SymbolTable.")((void)0); |
| 2175 | |
| 2176 | // Reuse an existing SHT_STRTAB section if it exists. |
| 2177 | StringTableSection *StrTab = nullptr; |
| 2178 | for (SectionBase &Sec : sections()) { |
| 2179 | if (Sec.Type == ELF::SHT_STRTAB && !(Sec.Flags & SHF_ALLOC)) { |
| 2180 | StrTab = static_cast<StringTableSection *>(&Sec); |
| 2181 | |
| 2182 | // Prefer a string table that is not the section header string table, if |
| 2183 | // such a table exists. |
| 2184 | if (SectionNames != &Sec) |
| 2185 | break; |
| 2186 | } |
| 2187 | } |
| 2188 | if (!StrTab) |
| 2189 | StrTab = &addSection<StringTableSection>(); |
| 2190 | |
| 2191 | SymbolTableSection &SymTab = addSection<SymbolTableSection>(); |
| 2192 | SymTab.Name = ".symtab"; |
| 2193 | SymTab.Link = StrTab->Index; |
| 2194 | if (Error Err = SymTab.initialize(sections())) |
| 2195 | return Err; |
| 2196 | SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0); |
| 2197 | |
| 2198 | SymbolTable = &SymTab; |
| 2199 | |
| 2200 | return Error::success(); |
| 2201 | } |
| 2202 | |
| 2203 | void Object::sortSections() { |
| 2204 | // Use stable_sort to maintain the original ordering as closely as possible. |
| 2205 | llvm::stable_sort(Sections, [](const SecPtr &A, const SecPtr &B) { |
| 2206 | // Put SHT_GROUP sections first, since group section headers must come |
| 2207 | // before the sections they contain. This also matches what GNU objcopy |
| 2208 | // does. |
| 2209 | if (A->Type != B->Type && |
| 2210 | (A->Type == ELF::SHT_GROUP || B->Type == ELF::SHT_GROUP)) |
| 2211 | return A->Type == ELF::SHT_GROUP; |
| 2212 | // For all other sections, sort by offset order. |
| 2213 | return A->OriginalOffset < B->OriginalOffset; |
| 2214 | }); |
| 2215 | } |
| 2216 | |
| 2217 | // Orders segments such that if x = y->ParentSegment then y comes before x. |
| 2218 | static void orderSegments(std::vector<Segment *> &Segments) { |
| 2219 | llvm::stable_sort(Segments, compareSegmentsByOffset); |
| 2220 | } |
| 2221 | |
| 2222 | // This function finds a consistent layout for a list of segments starting from |
| 2223 | // an Offset. It assumes that Segments have been sorted by orderSegments and |
| 2224 | // returns an Offset one past the end of the last segment. |
| 2225 | static uint64_t layoutSegments(std::vector<Segment *> &Segments, |
| 2226 | uint64_t Offset) { |
| 2227 | assert(llvm::is_sorted(Segments, compareSegmentsByOffset))((void)0); |
| 2228 | // The only way a segment should move is if a section was between two |
| 2229 | // segments and that section was removed. If that section isn't in a segment |
| 2230 | // then it's acceptable, but not ideal, to simply move it to after the |
| 2231 | // segments. So we can simply layout segments one after the other accounting |
| 2232 | // for alignment. |
| 2233 | for (Segment *Seg : Segments) { |
| 2234 | // We assume that segments have been ordered by OriginalOffset and Index |
| 2235 | // such that a parent segment will always come before a child segment in |
| 2236 | // OrderedSegments. This means that the Offset of the ParentSegment should |
| 2237 | // already be set and we can set our offset relative to it. |
| 2238 | if (Seg->ParentSegment != nullptr) { |
| 2239 | Segment *Parent = Seg->ParentSegment; |
| 2240 | Seg->Offset = |
| 2241 | Parent->Offset + Seg->OriginalOffset - Parent->OriginalOffset; |
| 2242 | } else { |
| 2243 | Seg->Offset = |
| 2244 | alignTo(Offset, std::max<uint64_t>(Seg->Align, 1), Seg->VAddr); |
| 2245 | } |
| 2246 | Offset = std::max(Offset, Seg->Offset + Seg->FileSize); |
| 2247 | } |
| 2248 | return Offset; |
| 2249 | } |
| 2250 | |
| 2251 | // This function finds a consistent layout for a list of sections. It assumes |
| 2252 | // that the ->ParentSegment of each section has already been laid out. The |
| 2253 | // supplied starting Offset is used for the starting offset of any section that |
| 2254 | // does not have a ParentSegment. It returns either the offset given if all |
| 2255 | // sections had a ParentSegment or an offset one past the last section if there |
| 2256 | // was a section that didn't have a ParentSegment. |
| 2257 | template <class Range> |
| 2258 | static uint64_t layoutSections(Range Sections, uint64_t Offset) { |
| 2259 | // Now the offset of every segment has been set we can assign the offsets |
| 2260 | // of each section. For sections that are covered by a segment we should use |
| 2261 | // the segment's original offset and the section's original offset to compute |
| 2262 | // the offset from the start of the segment. Using the offset from the start |
| 2263 | // of the segment we can assign a new offset to the section. For sections not |
| 2264 | // covered by segments we can just bump Offset to the next valid location. |
| 2265 | uint32_t Index = 1; |
| 2266 | for (auto &Sec : Sections) { |
| 2267 | Sec.Index = Index++; |
| 2268 | if (Sec.ParentSegment != nullptr) { |
| 2269 | auto Segment = *Sec.ParentSegment; |
| 2270 | Sec.Offset = |
| 2271 | Segment.Offset + (Sec.OriginalOffset - Segment.OriginalOffset); |
| 2272 | } else { |
| 2273 | Offset = alignTo(Offset, Sec.Align == 0 ? 1 : Sec.Align); |
| 2274 | Sec.Offset = Offset; |
| 2275 | if (Sec.Type != SHT_NOBITS) |
| 2276 | Offset += Sec.Size; |
| 2277 | } |
| 2278 | } |
| 2279 | return Offset; |
| 2280 | } |
| 2281 | |
| 2282 | // Rewrite sh_offset after some sections are changed to SHT_NOBITS and thus |
| 2283 | // occupy no space in the file. |
| 2284 | static uint64_t layoutSectionsForOnlyKeepDebug(Object &Obj, uint64_t Off) { |
| 2285 | uint32_t Index = 1; |
| 2286 | for (auto &Sec : Obj.sections()) { |
| 2287 | Sec.Index = Index++; |
| 2288 | |
| 2289 | auto *FirstSec = Sec.ParentSegment && Sec.ParentSegment->Type == PT_LOAD |
| 2290 | ? Sec.ParentSegment->firstSection() |
| 2291 | : nullptr; |
| 2292 | |
| 2293 | // The first section in a PT_LOAD has to have congruent offset and address |
| 2294 | // modulo the alignment, which usually equals the maximum page size. |
| 2295 | if (FirstSec && FirstSec == &Sec) |
| 2296 | Off = alignTo(Off, Sec.ParentSegment->Align, Sec.Addr); |
| 2297 | |
| 2298 | // sh_offset is not significant for SHT_NOBITS sections, but the congruence |
| 2299 | // rule must be followed if it is the first section in a PT_LOAD. Do not |
| 2300 | // advance Off. |
| 2301 | if (Sec.Type == SHT_NOBITS) { |
| 2302 | Sec.Offset = Off; |
| 2303 | continue; |
| 2304 | } |
| 2305 | |
| 2306 | if (!FirstSec) { |
| 2307 | // FirstSec being nullptr generally means that Sec does not have the |
| 2308 | // SHF_ALLOC flag. |
| 2309 | Off = Sec.Align ? alignTo(Off, Sec.Align) : Off; |
| 2310 | } else if (FirstSec != &Sec) { |
| 2311 | // The offset is relative to the first section in the PT_LOAD segment. Use |
| 2312 | // sh_offset for non-SHF_ALLOC sections. |
| 2313 | Off = Sec.OriginalOffset - FirstSec->OriginalOffset + FirstSec->Offset; |
| 2314 | } |
| 2315 | Sec.Offset = Off; |
| 2316 | Off += Sec.Size; |
| 2317 | } |
| 2318 | return Off; |
| 2319 | } |
| 2320 | |
| 2321 | // Rewrite p_offset and p_filesz of non-PT_PHDR segments after sh_offset values |
| 2322 | // have been updated. |
| 2323 | static uint64_t layoutSegmentsForOnlyKeepDebug(std::vector<Segment *> &Segments, |
| 2324 | uint64_t HdrEnd) { |
| 2325 | uint64_t MaxOffset = 0; |
| 2326 | for (Segment *Seg : Segments) { |
| 2327 | if (Seg->Type == PT_PHDR) |
| 2328 | continue; |
| 2329 | |
| 2330 | // The segment offset is generally the offset of the first section. |
| 2331 | // |
| 2332 | // For a segment containing no section (see sectionWithinSegment), if it has |
| 2333 | // a parent segment, copy the parent segment's offset field. This works for |
| 2334 | // empty PT_TLS. If no parent segment, use 0: the segment is not useful for |
| 2335 | // debugging anyway. |
| 2336 | const SectionBase *FirstSec = Seg->firstSection(); |
| 2337 | uint64_t Offset = |
| 2338 | FirstSec ? FirstSec->Offset |
| 2339 | : (Seg->ParentSegment ? Seg->ParentSegment->Offset : 0); |
| 2340 | uint64_t FileSize = 0; |
| 2341 | for (const SectionBase *Sec : Seg->Sections) { |
| 2342 | uint64_t Size = Sec->Type == SHT_NOBITS ? 0 : Sec->Size; |
| 2343 | if (Sec->Offset + Size > Offset) |
| 2344 | FileSize = std::max(FileSize, Sec->Offset + Size - Offset); |
| 2345 | } |
| 2346 | |
| 2347 | // If the segment includes EHDR and program headers, don't make it smaller |
| 2348 | // than the headers. |
| 2349 | if (Seg->Offset < HdrEnd && HdrEnd <= Seg->Offset + Seg->FileSize) { |
| 2350 | FileSize += Offset - Seg->Offset; |
| 2351 | Offset = Seg->Offset; |
| 2352 | FileSize = std::max(FileSize, HdrEnd - Offset); |
| 2353 | } |
| 2354 | |
| 2355 | Seg->Offset = Offset; |
| 2356 | Seg->FileSize = FileSize; |
| 2357 | MaxOffset = std::max(MaxOffset, Offset + FileSize); |
| 2358 | } |
| 2359 | return MaxOffset; |
| 2360 | } |
| 2361 | |
| 2362 | template <class ELFT> void ELFWriter<ELFT>::initEhdrSegment() { |
| 2363 | Segment &ElfHdr = Obj.ElfHdrSegment; |
| 2364 | ElfHdr.Type = PT_PHDR; |
| 2365 | ElfHdr.Flags = 0; |
| 2366 | ElfHdr.VAddr = 0; |
| 2367 | ElfHdr.PAddr = 0; |
| 2368 | ElfHdr.FileSize = ElfHdr.MemSize = sizeof(Elf_Ehdr); |
| 2369 | ElfHdr.Align = 0; |
| 2370 | } |
| 2371 | |
| 2372 | template <class ELFT> void ELFWriter<ELFT>::assignOffsets() { |
| 2373 | // We need a temporary list of segments that has a special order to it |
| 2374 | // so that we know that anytime ->ParentSegment is set that segment has |
| 2375 | // already had its offset properly set. |
| 2376 | std::vector<Segment *> OrderedSegments; |
| 2377 | for (Segment &Segment : Obj.segments()) |
| 2378 | OrderedSegments.push_back(&Segment); |
| 2379 | OrderedSegments.push_back(&Obj.ElfHdrSegment); |
| 2380 | OrderedSegments.push_back(&Obj.ProgramHdrSegment); |
| 2381 | orderSegments(OrderedSegments); |
| 2382 | |
| 2383 | uint64_t Offset; |
| 2384 | if (OnlyKeepDebug) { |
| 2385 | // For --only-keep-debug, the sections that did not preserve contents were |
| 2386 | // changed to SHT_NOBITS. We now rewrite sh_offset fields of sections, and |
| 2387 | // then rewrite p_offset/p_filesz of program headers. |
| 2388 | uint64_t HdrEnd = |
| 2389 | sizeof(Elf_Ehdr) + llvm::size(Obj.segments()) * sizeof(Elf_Phdr); |
| 2390 | Offset = layoutSectionsForOnlyKeepDebug(Obj, HdrEnd); |
| 2391 | Offset = std::max(Offset, |
| 2392 | layoutSegmentsForOnlyKeepDebug(OrderedSegments, HdrEnd)); |
| 2393 | } else { |
| 2394 | // Offset is used as the start offset of the first segment to be laid out. |
| 2395 | // Since the ELF Header (ElfHdrSegment) must be at the start of the file, |
| 2396 | // we start at offset 0. |
| 2397 | Offset = layoutSegments(OrderedSegments, 0); |
| 2398 | Offset = layoutSections(Obj.sections(), Offset); |
| 2399 | } |
| 2400 | // If we need to write the section header table out then we need to align the |
| 2401 | // Offset so that SHOffset is valid. |
| 2402 | if (WriteSectionHeaders) |
| 2403 | Offset = alignTo(Offset, sizeof(Elf_Addr)); |
| 2404 | Obj.SHOff = Offset; |
| 2405 | } |
| 2406 | |
| 2407 | template <class ELFT> size_t ELFWriter<ELFT>::totalSize() const { |
| 2408 | // We already have the section header offset so we can calculate the total |
| 2409 | // size by just adding up the size of each section header. |
| 2410 | if (!WriteSectionHeaders) |
| 2411 | return Obj.SHOff; |
| 2412 | size_t ShdrCount = Obj.sections().size() + 1; // Includes null shdr. |
| 2413 | return Obj.SHOff + ShdrCount * sizeof(Elf_Shdr); |
| 2414 | } |
| 2415 | |
| 2416 | template <class ELFT> Error ELFWriter<ELFT>::write() { |
| 2417 | // Segment data must be written first, so that the ELF header and program |
| 2418 | // header tables can overwrite it, if covered by a segment. |
| 2419 | writeSegmentData(); |
| 2420 | writeEhdr(); |
| 2421 | writePhdrs(); |
| 2422 | if (Error E = writeSectionData()) |
| 2423 | return E; |
| 2424 | if (WriteSectionHeaders) |
| 2425 | writeShdrs(); |
| 2426 | |
| 2427 | // TODO: Implement direct writing to the output stream (without intermediate |
| 2428 | // memory buffer Buf). |
| 2429 | Out.write(Buf->getBufferStart(), Buf->getBufferSize()); |
| 2430 | return Error::success(); |
| 2431 | } |
| 2432 | |
| 2433 | static Error removeUnneededSections(Object &Obj) { |
| 2434 | // We can remove an empty symbol table from non-relocatable objects. |
| 2435 | // Relocatable objects typically have relocation sections whose |
| 2436 | // sh_link field points to .symtab, so we can't remove .symtab |
| 2437 | // even if it is empty. |
| 2438 | if (Obj.isRelocatable() || Obj.SymbolTable == nullptr || |
| 2439 | !Obj.SymbolTable->empty()) |
| 2440 | return Error::success(); |
| 2441 | |
| 2442 | // .strtab can be used for section names. In such a case we shouldn't |
| 2443 | // remove it. |
| 2444 | auto *StrTab = Obj.SymbolTable->getStrTab() == Obj.SectionNames |
| 2445 | ? nullptr |
| 2446 | : Obj.SymbolTable->getStrTab(); |
| 2447 | return Obj.removeSections(false, [&](const SectionBase &Sec) { |
| 2448 | return &Sec == Obj.SymbolTable || &Sec == StrTab; |
| 2449 | }); |
| 2450 | } |
| 2451 | |
| 2452 | template <class ELFT> Error ELFWriter<ELFT>::finalize() { |
| 2453 | // It could happen that SectionNames has been removed and yet the user wants |
| 2454 | // a section header table output. We need to throw an error if a user tries |
| 2455 | // to do that. |
| 2456 | if (Obj.SectionNames == nullptr && WriteSectionHeaders) |
| 2457 | return createStringError(llvm::errc::invalid_argument, |
| 2458 | "cannot write section header table because " |
| 2459 | "section header string table was removed"); |
| 2460 | |
| 2461 | if (Error E = removeUnneededSections(Obj)) |
| 2462 | return E; |
| 2463 | Obj.sortSections(); |
| 2464 | |
| 2465 | // We need to assign indexes before we perform layout because we need to know |
| 2466 | // if we need large indexes or not. We can assign indexes first and check as |
| 2467 | // we go to see if we will actully need large indexes. |
| 2468 | bool NeedsLargeIndexes = false; |
| 2469 | if (Obj.sections().size() >= SHN_LORESERVE) { |
| 2470 | SectionTableRef Sections = Obj.sections(); |
| 2471 | // Sections doesn't include the null section header, so account for this |
| 2472 | // when skipping the first N sections. |
| 2473 | NeedsLargeIndexes = |
| 2474 | any_of(drop_begin(Sections, SHN_LORESERVE - 1), |
| 2475 | [](const SectionBase &Sec) { return Sec.HasSymbol; }); |
| 2476 | // TODO: handle case where only one section needs the large index table but |
| 2477 | // only needs it because the large index table hasn't been removed yet. |
| 2478 | } |
| 2479 | |
| 2480 | if (NeedsLargeIndexes) { |
| 2481 | // This means we definitely need to have a section index table but if we |
| 2482 | // already have one then we should use it instead of making a new one. |
| 2483 | if (Obj.SymbolTable != nullptr && Obj.SectionIndexTable == nullptr) { |
| 2484 | // Addition of a section to the end does not invalidate the indexes of |
| 2485 | // other sections and assigns the correct index to the new section. |
| 2486 | auto &Shndx = Obj.addSection<SectionIndexSection>(); |
| 2487 | Obj.SymbolTable->setShndxTable(&Shndx); |
| 2488 | Shndx.setSymTab(Obj.SymbolTable); |
| 2489 | } |
| 2490 | } else { |
| 2491 | // Since we don't need SectionIndexTable we should remove it and all |
| 2492 | // references to it. |
| 2493 | if (Obj.SectionIndexTable != nullptr) { |
| 2494 | // We do not support sections referring to the section index table. |
| 2495 | if (Error E = Obj.removeSections(false /*AllowBrokenLinks*/, |
| 2496 | [this](const SectionBase &Sec) { |
| 2497 | return &Sec == Obj.SectionIndexTable; |
| 2498 | })) |
| 2499 | return E; |
| 2500 | } |
| 2501 | } |
| 2502 | |
| 2503 | // Make sure we add the names of all the sections. Importantly this must be |
| 2504 | // done after we decide to add or remove SectionIndexes. |
| 2505 | if (Obj.SectionNames != nullptr) |
| 2506 | for (const SectionBase &Sec : Obj.sections()) |
| 2507 | Obj.SectionNames->addString(Sec.Name); |
| 2508 | |
| 2509 | initEhdrSegment(); |
| 2510 | |
| 2511 | // Before we can prepare for layout the indexes need to be finalized. |
| 2512 | // Also, the output arch may not be the same as the input arch, so fix up |
| 2513 | // size-related fields before doing layout calculations. |
| 2514 | uint64_t Index = 0; |
| 2515 | auto SecSizer = std::make_unique<ELFSectionSizer<ELFT>>(); |
| 2516 | for (SectionBase &Sec : Obj.sections()) { |
| 2517 | Sec.Index = Index++; |
| 2518 | if (Error Err = Sec.accept(*SecSizer)) |
| 2519 | return Err; |
| 2520 | } |
| 2521 | |
| 2522 | // The symbol table does not update all other sections on update. For |
| 2523 | // instance, symbol names are not added as new symbols are added. This means |
| 2524 | // that some sections, like .strtab, don't yet have their final size. |
| 2525 | if (Obj.SymbolTable != nullptr) |
| 2526 | Obj.SymbolTable->prepareForLayout(); |
| 2527 | |
| 2528 | // Now that all strings are added we want to finalize string table builders, |
| 2529 | // because that affects section sizes which in turn affects section offsets. |
| 2530 | for (SectionBase &Sec : Obj.sections()) |
| 2531 | if (auto StrTab = dyn_cast<StringTableSection>(&Sec)) |
| 2532 | StrTab->prepareForLayout(); |
| 2533 | |
| 2534 | assignOffsets(); |
| 2535 | |
| 2536 | // layoutSections could have modified section indexes, so we need |
| 2537 | // to fill the index table after assignOffsets. |
| 2538 | if (Obj.SymbolTable != nullptr) |
| 2539 | Obj.SymbolTable->fillShndxTable(); |
| 2540 | |
| 2541 | // Finally now that all offsets and indexes have been set we can finalize any |
| 2542 | // remaining issues. |
| 2543 | uint64_t Offset = Obj.SHOff + sizeof(Elf_Shdr); |
| 2544 | for (SectionBase &Sec : Obj.sections()) { |
| 2545 | Sec.HeaderOffset = Offset; |
| 2546 | Offset += sizeof(Elf_Shdr); |
| 2547 | if (WriteSectionHeaders) |
| 2548 | Sec.NameIndex = Obj.SectionNames->findIndex(Sec.Name); |
| 2549 | Sec.finalize(); |
| 2550 | } |
| 2551 | |
| 2552 | size_t TotalSize = totalSize(); |
| 2553 | Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); |
| 2554 | if (!Buf) |
| 2555 | return createStringError(errc::not_enough_memory, |
| 2556 | "failed to allocate memory buffer of " + |
| 2557 | Twine::utohexstr(TotalSize) + " bytes"); |
| 2558 | |
| 2559 | SecWriter = std::make_unique<ELFSectionWriter<ELFT>>(*Buf); |
| 2560 | return Error::success(); |
| 2561 | } |
| 2562 | |
| 2563 | Error BinaryWriter::write() { |
| 2564 | for (const SectionBase &Sec : Obj.allocSections()) |
| 2565 | if (Error Err = Sec.accept(*SecWriter)) |
| 2566 | return Err; |
| 2567 | |
| 2568 | // TODO: Implement direct writing to the output stream (without intermediate |
| 2569 | // memory buffer Buf). |
| 2570 | Out.write(Buf->getBufferStart(), Buf->getBufferSize()); |
| 2571 | return Error::success(); |
| 2572 | } |
| 2573 | |
| 2574 | Error BinaryWriter::finalize() { |
| 2575 | // Compute the section LMA based on its sh_offset and the containing segment's |
| 2576 | // p_offset and p_paddr. Also compute the minimum LMA of all non-empty |
| 2577 | // sections as MinAddr. In the output, the contents between address 0 and |
| 2578 | // MinAddr will be skipped. |
| 2579 | uint64_t MinAddr = UINT64_MAX0xffffffffffffffffULL; |
| 2580 | for (SectionBase &Sec : Obj.allocSections()) { |
| 2581 | if (Sec.ParentSegment != nullptr) |
| 2582 | Sec.Addr = |
| 2583 | Sec.Offset - Sec.ParentSegment->Offset + Sec.ParentSegment->PAddr; |
| 2584 | if (Sec.Type != SHT_NOBITS && Sec.Size > 0) |
| 2585 | MinAddr = std::min(MinAddr, Sec.Addr); |
| 2586 | } |
| 2587 | |
| 2588 | // Now that every section has been laid out we just need to compute the total |
| 2589 | // file size. This might not be the same as the offset returned by |
| 2590 | // layoutSections, because we want to truncate the last segment to the end of |
| 2591 | // its last non-empty section, to match GNU objcopy's behaviour. |
| 2592 | TotalSize = 0; |
| 2593 | for (SectionBase &Sec : Obj.allocSections()) |
| 2594 | if (Sec.Type != SHT_NOBITS && Sec.Size > 0) { |
| 2595 | Sec.Offset = Sec.Addr - MinAddr; |
| 2596 | TotalSize = std::max(TotalSize, Sec.Offset + Sec.Size); |
| 2597 | } |
| 2598 | |
| 2599 | Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); |
| 2600 | if (!Buf) |
| 2601 | return createStringError(errc::not_enough_memory, |
| 2602 | "failed to allocate memory buffer of " + |
| 2603 | Twine::utohexstr(TotalSize) + " bytes"); |
| 2604 | SecWriter = std::make_unique<BinarySectionWriter>(*Buf); |
| 2605 | return Error::success(); |
| 2606 | } |
| 2607 | |
| 2608 | bool IHexWriter::SectionCompare::operator()(const SectionBase *Lhs, |
| 2609 | const SectionBase *Rhs) const { |
| 2610 | return (sectionPhysicalAddr(Lhs) & 0xFFFFFFFFU) < |
| 2611 | (sectionPhysicalAddr(Rhs) & 0xFFFFFFFFU); |
| 2612 | } |
| 2613 | |
| 2614 | uint64_t IHexWriter::writeEntryPointRecord(uint8_t *Buf) { |
| 2615 | IHexLineData HexData; |
| 2616 | uint8_t Data[4] = {}; |
| 2617 | // We don't write entry point record if entry is zero. |
| 2618 | if (Obj.Entry == 0) |
| 2619 | return 0; |
| 2620 | |
| 2621 | if (Obj.Entry <= 0xFFFFFU) { |
| 2622 | Data[0] = ((Obj.Entry & 0xF0000U) >> 12) & 0xFF; |
| 2623 | support::endian::write(&Data[2], static_cast<uint16_t>(Obj.Entry), |
| 2624 | support::big); |
| 2625 | HexData = IHexRecord::getLine(IHexRecord::StartAddr80x86, 0, Data); |
| 2626 | } else { |
| 2627 | support::endian::write(Data, static_cast<uint32_t>(Obj.Entry), |
| 2628 | support::big); |
| 2629 | HexData = IHexRecord::getLine(IHexRecord::StartAddr, 0, Data); |
| 2630 | } |
| 2631 | memcpy(Buf, HexData.data(), HexData.size()); |
| 2632 | return HexData.size(); |
| 2633 | } |
| 2634 | |
| 2635 | uint64_t IHexWriter::writeEndOfFileRecord(uint8_t *Buf) { |
| 2636 | IHexLineData HexData = IHexRecord::getLine(IHexRecord::EndOfFile, 0, {}); |
| 2637 | memcpy(Buf, HexData.data(), HexData.size()); |
| 2638 | return HexData.size(); |
| 2639 | } |
| 2640 | |
| 2641 | Error IHexWriter::write() { |
| 2642 | IHexSectionWriter Writer(*Buf); |
| 2643 | // Write sections. |
| 2644 | for (const SectionBase *Sec : Sections) |
| 2645 | if (Error Err = Sec->accept(Writer)) |
| 2646 | return Err; |
| 2647 | |
| 2648 | uint64_t Offset = Writer.getBufferOffset(); |
| 2649 | // Write entry point address. |
| 2650 | Offset += writeEntryPointRecord( |
| 2651 | reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset); |
| 2652 | // Write EOF. |
| 2653 | Offset += writeEndOfFileRecord( |
Value stored to 'Offset' is never read | |
| 2654 | reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset); |
| 2655 | assert(Offset == TotalSize)((void)0); |
| 2656 | |
| 2657 | // TODO: Implement direct writing to the output stream (without intermediate |
| 2658 | // memory buffer Buf). |
| 2659 | Out.write(Buf->getBufferStart(), Buf->getBufferSize()); |
| 2660 | return Error::success(); |
| 2661 | } |
| 2662 | |
| 2663 | Error IHexWriter::checkSection(const SectionBase &Sec) { |
| 2664 | uint64_t Addr = sectionPhysicalAddr(&Sec); |
| 2665 | if (addressOverflows32bit(Addr) || addressOverflows32bit(Addr + Sec.Size - 1)) |
| 2666 | return createStringError( |
| 2667 | errc::invalid_argument, |
| 2668 | "Section '%s' address range [0x%llx, 0x%llx] is not 32 bit", |
| 2669 | Sec.Name.c_str(), Addr, Addr + Sec.Size - 1); |
| 2670 | return Error::success(); |
| 2671 | } |
| 2672 | |
| 2673 | Error IHexWriter::finalize() { |
| 2674 | // We can't write 64-bit addresses. |
| 2675 | if (addressOverflows32bit(Obj.Entry)) |
| 2676 | return createStringError(errc::invalid_argument, |
| 2677 | "Entry point address 0x%llx overflows 32 bits", |
| 2678 | Obj.Entry); |
| 2679 | |
| 2680 | for (const SectionBase &Sec : Obj.sections()) |
| 2681 | if ((Sec.Flags & ELF::SHF_ALLOC) && Sec.Type != ELF::SHT_NOBITS && |
| 2682 | Sec.Size > 0) { |
| 2683 | if (Error E = checkSection(Sec)) |
| 2684 | return E; |
| 2685 | Sections.insert(&Sec); |
| 2686 | } |
| 2687 | |
| 2688 | std::unique_ptr<WritableMemoryBuffer> EmptyBuffer = |
| 2689 | WritableMemoryBuffer::getNewMemBuffer(0); |
| 2690 | if (!EmptyBuffer) |
| 2691 | return createStringError(errc::not_enough_memory, |
| 2692 | "failed to allocate memory buffer of 0 bytes"); |
| 2693 | |
| 2694 | IHexSectionWriterBase LengthCalc(*EmptyBuffer); |
| 2695 | for (const SectionBase *Sec : Sections) |
| 2696 | if (Error Err = Sec->accept(LengthCalc)) |
| 2697 | return Err; |
| 2698 | |
| 2699 | // We need space to write section records + StartAddress record |
| 2700 | // (if start adress is not zero) + EndOfFile record. |
| 2701 | TotalSize = LengthCalc.getBufferOffset() + |
| 2702 | (Obj.Entry ? IHexRecord::getLineLength(4) : 0) + |
| 2703 | IHexRecord::getLineLength(0); |
| 2704 | |
| 2705 | Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize); |
| 2706 | if (!Buf) |
| 2707 | return createStringError(errc::not_enough_memory, |
| 2708 | "failed to allocate memory buffer of " + |
| 2709 | Twine::utohexstr(TotalSize) + " bytes"); |
| 2710 | |
| 2711 | return Error::success(); |
| 2712 | } |
| 2713 | |
| 2714 | namespace llvm { |
| 2715 | namespace objcopy { |
| 2716 | namespace elf { |
| 2717 | |
| 2718 | template class ELFBuilder<ELF64LE>; |
| 2719 | template class ELFBuilder<ELF64BE>; |
| 2720 | template class ELFBuilder<ELF32LE>; |
| 2721 | template class ELFBuilder<ELF32BE>; |
| 2722 | |
| 2723 | template class ELFWriter<ELF64LE>; |
| 2724 | template class ELFWriter<ELF64BE>; |
| 2725 | template class ELFWriter<ELF32LE>; |
| 2726 | template class ELFWriter<ELF32BE>; |
| 2727 | |
| 2728 | } // end namespace elf |
| 2729 | } // end namespace objcopy |
| 2730 | } // end namespace llvm |