[ARM] Add earlyclobber for cross beat MVE instructions
[llvm-complete.git] / tools / llvm-objcopy / ELF / Object.cpp
blob3d1527cc2aeef8d65e50861a9f8bf9f21ef3b2ad
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 //===----------------------------------------------------------------------===//
9 #include "Object.h"
10 #include "llvm-objcopy.h"
11 #include "llvm/ADT/ArrayRef.h"
12 #include "llvm/ADT/STLExtras.h"
13 #include "llvm/ADT/StringRef.h"
14 #include "llvm/ADT/Twine.h"
15 #include "llvm/ADT/iterator_range.h"
16 #include "llvm/BinaryFormat/ELF.h"
17 #include "llvm/MC/MCTargetOptions.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>
32 namespace llvm {
33 namespace objcopy {
34 namespace elf {
36 using namespace object;
37 using namespace ELF;
39 template <class ELFT> void ELFWriter<ELFT>::writePhdr(const Segment &Seg) {
40 uint8_t *B = Buf.getBufferStart() + Obj.ProgramHdrSegment.Offset +
41 Seg.Index * sizeof(Elf_Phdr);
42 Elf_Phdr &Phdr = *reinterpret_cast<Elf_Phdr *>(B);
43 Phdr.p_type = Seg.Type;
44 Phdr.p_flags = Seg.Flags;
45 Phdr.p_offset = Seg.Offset;
46 Phdr.p_vaddr = Seg.VAddr;
47 Phdr.p_paddr = Seg.PAddr;
48 Phdr.p_filesz = Seg.FileSize;
49 Phdr.p_memsz = Seg.MemSize;
50 Phdr.p_align = Seg.Align;
53 Error SectionBase::removeSectionReferences(
54 bool AllowBrokenLinks,
55 function_ref<bool(const SectionBase *)> ToRemove) {
56 return Error::success();
59 Error SectionBase::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
60 return Error::success();
63 void SectionBase::initialize(SectionTableRef SecTable) {}
64 void SectionBase::finalize() {}
65 void SectionBase::markSymbols() {}
66 void SectionBase::replaceSectionReferences(
67 const DenseMap<SectionBase *, SectionBase *> &) {}
69 template <class ELFT> void ELFWriter<ELFT>::writeShdr(const SectionBase &Sec) {
70 uint8_t *B = Buf.getBufferStart() + Sec.HeaderOffset;
71 Elf_Shdr &Shdr = *reinterpret_cast<Elf_Shdr *>(B);
72 Shdr.sh_name = Sec.NameIndex;
73 Shdr.sh_type = Sec.Type;
74 Shdr.sh_flags = Sec.Flags;
75 Shdr.sh_addr = Sec.Addr;
76 Shdr.sh_offset = Sec.Offset;
77 Shdr.sh_size = Sec.Size;
78 Shdr.sh_link = Sec.Link;
79 Shdr.sh_info = Sec.Info;
80 Shdr.sh_addralign = Sec.Align;
81 Shdr.sh_entsize = Sec.EntrySize;
84 template <class ELFT> void ELFSectionSizer<ELFT>::visit(Section &Sec) {}
86 template <class ELFT>
87 void ELFSectionSizer<ELFT>::visit(OwnedDataSection &Sec) {}
89 template <class ELFT>
90 void ELFSectionSizer<ELFT>::visit(StringTableSection &Sec) {}
92 template <class ELFT>
93 void ELFSectionSizer<ELFT>::visit(DynamicRelocationSection &Sec) {}
95 template <class ELFT>
96 void ELFSectionSizer<ELFT>::visit(SymbolTableSection &Sec) {
97 Sec.EntrySize = sizeof(Elf_Sym);
98 Sec.Size = Sec.Symbols.size() * Sec.EntrySize;
99 // Align to the largest field in Elf_Sym.
100 Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word);
103 template <class ELFT>
104 void ELFSectionSizer<ELFT>::visit(RelocationSection &Sec) {
105 Sec.EntrySize = Sec.Type == SHT_REL ? sizeof(Elf_Rel) : sizeof(Elf_Rela);
106 Sec.Size = Sec.Relocations.size() * Sec.EntrySize;
107 // Align to the largest field in Elf_Rel(a).
108 Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word);
111 template <class ELFT>
112 void ELFSectionSizer<ELFT>::visit(GnuDebugLinkSection &Sec) {}
114 template <class ELFT> void ELFSectionSizer<ELFT>::visit(GroupSection &Sec) {}
116 template <class ELFT>
117 void ELFSectionSizer<ELFT>::visit(SectionIndexSection &Sec) {}
119 template <class ELFT>
120 void ELFSectionSizer<ELFT>::visit(CompressedSection &Sec) {}
122 template <class ELFT>
123 void ELFSectionSizer<ELFT>::visit(DecompressedSection &Sec) {}
125 void BinarySectionWriter::visit(const SectionIndexSection &Sec) {
126 error("cannot write symbol section index table '" + Sec.Name + "' ");
129 void BinarySectionWriter::visit(const SymbolTableSection &Sec) {
130 error("cannot write symbol table '" + Sec.Name + "' out to binary");
133 void BinarySectionWriter::visit(const RelocationSection &Sec) {
134 error("cannot write relocation section '" + Sec.Name + "' out to binary");
137 void BinarySectionWriter::visit(const GnuDebugLinkSection &Sec) {
138 error("cannot write '" + Sec.Name + "' out to binary");
141 void BinarySectionWriter::visit(const GroupSection &Sec) {
142 error("cannot write '" + Sec.Name + "' out to binary");
145 void SectionWriter::visit(const Section &Sec) {
146 if (Sec.Type != SHT_NOBITS)
147 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset);
150 static bool addressOverflows32bit(uint64_t Addr) {
151 // Sign extended 32 bit addresses (e.g 0xFFFFFFFF80000000) are ok
152 return Addr > UINT32_MAX && Addr + 0x80000000 > UINT32_MAX;
155 template <class T> static T checkedGetHex(StringRef S) {
156 T Value;
157 bool Fail = S.getAsInteger(16, Value);
158 assert(!Fail);
159 (void)Fail;
160 return Value;
163 // Fills exactly Len bytes of buffer with hexadecimal characters
164 // representing value 'X'
165 template <class T, class Iterator>
166 static Iterator utohexstr(T X, Iterator It, size_t Len) {
167 // Fill range with '0'
168 std::fill(It, It + Len, '0');
170 for (long I = Len - 1; I >= 0; --I) {
171 unsigned char Mod = static_cast<unsigned char>(X) & 15;
172 *(It + I) = hexdigit(Mod, false);
173 X >>= 4;
175 assert(X == 0);
176 return It + Len;
179 uint8_t IHexRecord::getChecksum(StringRef S) {
180 assert((S.size() & 1) == 0);
181 uint8_t Checksum = 0;
182 while (!S.empty()) {
183 Checksum += checkedGetHex<uint8_t>(S.take_front(2));
184 S = S.drop_front(2);
186 return -Checksum;
189 IHexLineData IHexRecord::getLine(uint8_t Type, uint16_t Addr,
190 ArrayRef<uint8_t> Data) {
191 IHexLineData Line(getLineLength(Data.size()));
192 assert(Line.size());
193 auto Iter = Line.begin();
194 *Iter++ = ':';
195 Iter = utohexstr(Data.size(), Iter, 2);
196 Iter = utohexstr(Addr, Iter, 4);
197 Iter = utohexstr(Type, Iter, 2);
198 for (uint8_t X : Data)
199 Iter = utohexstr(X, Iter, 2);
200 StringRef S(Line.data() + 1, std::distance(Line.begin() + 1, Iter));
201 Iter = utohexstr(getChecksum(S), Iter, 2);
202 *Iter++ = '\r';
203 *Iter++ = '\n';
204 assert(Iter == Line.end());
205 return Line;
208 static Error checkRecord(const IHexRecord &R) {
209 switch (R.Type) {
210 case IHexRecord::Data:
211 if (R.HexData.size() == 0)
212 return createStringError(
213 errc::invalid_argument,
214 "zero data length is not allowed for data records");
215 break;
216 case IHexRecord::EndOfFile:
217 break;
218 case IHexRecord::SegmentAddr:
219 // 20-bit segment address. Data length must be 2 bytes
220 // (4 bytes in hex)
221 if (R.HexData.size() != 4)
222 return createStringError(
223 errc::invalid_argument,
224 "segment address data should be 2 bytes in size");
225 break;
226 case IHexRecord::StartAddr80x86:
227 case IHexRecord::StartAddr:
228 if (R.HexData.size() != 8)
229 return createStringError(errc::invalid_argument,
230 "start address data should be 4 bytes in size");
231 // According to Intel HEX specification '03' record
232 // only specifies the code address within the 20-bit
233 // segmented address space of the 8086/80186. This
234 // means 12 high order bits should be zeroes.
235 if (R.Type == IHexRecord::StartAddr80x86 &&
236 R.HexData.take_front(3) != "000")
237 return createStringError(errc::invalid_argument,
238 "start address exceeds 20 bit for 80x86");
239 break;
240 case IHexRecord::ExtendedAddr:
241 // 16-31 bits of linear base address
242 if (R.HexData.size() != 4)
243 return createStringError(
244 errc::invalid_argument,
245 "extended address data should be 2 bytes in size");
246 break;
247 default:
248 // Unknown record type
249 return createStringError(errc::invalid_argument, "unknown record type: %u",
250 static_cast<unsigned>(R.Type));
252 return Error::success();
255 // Checks that IHEX line contains valid characters.
256 // This allows converting hexadecimal data to integers
257 // without extra verification.
258 static Error checkChars(StringRef Line) {
259 assert(!Line.empty());
260 if (Line[0] != ':')
261 return createStringError(errc::invalid_argument,
262 "missing ':' in the beginning of line.");
264 for (size_t Pos = 1; Pos < Line.size(); ++Pos)
265 if (hexDigitValue(Line[Pos]) == -1U)
266 return createStringError(errc::invalid_argument,
267 "invalid character at position %zu.", Pos + 1);
268 return Error::success();
271 Expected<IHexRecord> IHexRecord::parse(StringRef Line) {
272 assert(!Line.empty());
274 // ':' + Length + Address + Type + Checksum with empty data ':LLAAAATTCC'
275 if (Line.size() < 11)
276 return createStringError(errc::invalid_argument,
277 "line is too short: %zu chars.", Line.size());
279 if (Error E = checkChars(Line))
280 return std::move(E);
282 IHexRecord Rec;
283 size_t DataLen = checkedGetHex<uint8_t>(Line.substr(1, 2));
284 if (Line.size() != getLength(DataLen))
285 return createStringError(errc::invalid_argument,
286 "invalid line length %zu (should be %zu)",
287 Line.size(), getLength(DataLen));
289 Rec.Addr = checkedGetHex<uint16_t>(Line.substr(3, 4));
290 Rec.Type = checkedGetHex<uint8_t>(Line.substr(7, 2));
291 Rec.HexData = Line.substr(9, DataLen * 2);
293 if (getChecksum(Line.drop_front(1)) != 0)
294 return createStringError(errc::invalid_argument, "incorrect checksum.");
295 if (Error E = checkRecord(Rec))
296 return std::move(E);
297 return Rec;
300 static uint64_t sectionPhysicalAddr(const SectionBase *Sec) {
301 Segment *Seg = Sec->ParentSegment;
302 if (Seg && Seg->Type != ELF::PT_LOAD)
303 Seg = nullptr;
304 return Seg ? Seg->PAddr + Sec->OriginalOffset - Seg->OriginalOffset
305 : Sec->Addr;
308 void IHexSectionWriterBase::writeSection(const SectionBase *Sec,
309 ArrayRef<uint8_t> Data) {
310 assert(Data.size() == Sec->Size);
311 const uint32_t ChunkSize = 16;
312 uint32_t Addr = sectionPhysicalAddr(Sec) & 0xFFFFFFFFU;
313 while (!Data.empty()) {
314 uint64_t DataSize = std::min<uint64_t>(Data.size(), ChunkSize);
315 if (Addr > SegmentAddr + BaseAddr + 0xFFFFU) {
316 if (Addr > 0xFFFFFU) {
317 // Write extended address record, zeroing segment address
318 // if needed.
319 if (SegmentAddr != 0)
320 SegmentAddr = writeSegmentAddr(0U);
321 BaseAddr = writeBaseAddr(Addr);
322 } else {
323 // We can still remain 16-bit
324 SegmentAddr = writeSegmentAddr(Addr);
327 uint64_t SegOffset = Addr - BaseAddr - SegmentAddr;
328 assert(SegOffset <= 0xFFFFU);
329 DataSize = std::min(DataSize, 0x10000U - SegOffset);
330 writeData(0, SegOffset, Data.take_front(DataSize));
331 Addr += DataSize;
332 Data = Data.drop_front(DataSize);
336 uint64_t IHexSectionWriterBase::writeSegmentAddr(uint64_t Addr) {
337 assert(Addr <= 0xFFFFFU);
338 uint8_t Data[] = {static_cast<uint8_t>((Addr & 0xF0000U) >> 12), 0};
339 writeData(2, 0, Data);
340 return Addr & 0xF0000U;
343 uint64_t IHexSectionWriterBase::writeBaseAddr(uint64_t Addr) {
344 assert(Addr <= 0xFFFFFFFFU);
345 uint64_t Base = Addr & 0xFFFF0000U;
346 uint8_t Data[] = {static_cast<uint8_t>(Base >> 24),
347 static_cast<uint8_t>((Base >> 16) & 0xFF)};
348 writeData(4, 0, Data);
349 return Base;
352 void IHexSectionWriterBase::writeData(uint8_t Type, uint16_t Addr,
353 ArrayRef<uint8_t> Data) {
354 Offset += IHexRecord::getLineLength(Data.size());
357 void IHexSectionWriterBase::visit(const Section &Sec) {
358 writeSection(&Sec, Sec.Contents);
361 void IHexSectionWriterBase::visit(const OwnedDataSection &Sec) {
362 writeSection(&Sec, Sec.Data);
365 void IHexSectionWriterBase::visit(const StringTableSection &Sec) {
366 // Check that sizer has already done its work
367 assert(Sec.Size == Sec.StrTabBuilder.getSize());
368 // We are free to pass an invalid pointer to writeSection as long
369 // as we don't actually write any data. The real writer class has
370 // to override this method .
371 writeSection(&Sec, {nullptr, static_cast<size_t>(Sec.Size)});
374 void IHexSectionWriterBase::visit(const DynamicRelocationSection &Sec) {
375 writeSection(&Sec, Sec.Contents);
378 void IHexSectionWriter::writeData(uint8_t Type, uint16_t Addr,
379 ArrayRef<uint8_t> Data) {
380 IHexLineData HexData = IHexRecord::getLine(Type, Addr, Data);
381 memcpy(Out.getBufferStart() + Offset, HexData.data(), HexData.size());
382 Offset += HexData.size();
385 void IHexSectionWriter::visit(const StringTableSection &Sec) {
386 assert(Sec.Size == Sec.StrTabBuilder.getSize());
387 std::vector<uint8_t> Data(Sec.Size);
388 Sec.StrTabBuilder.write(Data.data());
389 writeSection(&Sec, Data);
392 void Section::accept(SectionVisitor &Visitor) const { Visitor.visit(*this); }
394 void Section::accept(MutableSectionVisitor &Visitor) { Visitor.visit(*this); }
396 void SectionWriter::visit(const OwnedDataSection &Sec) {
397 llvm::copy(Sec.Data, Out.getBufferStart() + Sec.Offset);
400 static constexpr std::array<uint8_t, 4> ZlibGnuMagic = {'Z', 'L', 'I', 'B'};
402 static bool isDataGnuCompressed(ArrayRef<uint8_t> Data) {
403 return Data.size() > ZlibGnuMagic.size() &&
404 std::equal(ZlibGnuMagic.begin(), ZlibGnuMagic.end(), Data.data());
407 template <class ELFT>
408 static std::tuple<uint64_t, uint64_t>
409 getDecompressedSizeAndAlignment(ArrayRef<uint8_t> Data) {
410 const bool IsGnuDebug = isDataGnuCompressed(Data);
411 const uint64_t DecompressedSize =
412 IsGnuDebug
413 ? support::endian::read64be(Data.data() + ZlibGnuMagic.size())
414 : reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data.data())->ch_size;
415 const uint64_t DecompressedAlign =
416 IsGnuDebug ? 1
417 : reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data.data())
418 ->ch_addralign;
420 return std::make_tuple(DecompressedSize, DecompressedAlign);
423 template <class ELFT>
424 void ELFSectionWriter<ELFT>::visit(const DecompressedSection &Sec) {
425 const size_t DataOffset = isDataGnuCompressed(Sec.OriginalData)
426 ? (ZlibGnuMagic.size() + sizeof(Sec.Size))
427 : sizeof(Elf_Chdr_Impl<ELFT>);
429 StringRef CompressedContent(
430 reinterpret_cast<const char *>(Sec.OriginalData.data()) + DataOffset,
431 Sec.OriginalData.size() - DataOffset);
433 SmallVector<char, 128> DecompressedContent;
434 if (Error E = zlib::uncompress(CompressedContent, DecompressedContent,
435 static_cast<size_t>(Sec.Size)))
436 reportError(Sec.Name, std::move(E));
438 uint8_t *Buf = Out.getBufferStart() + Sec.Offset;
439 std::copy(DecompressedContent.begin(), DecompressedContent.end(), Buf);
442 void BinarySectionWriter::visit(const DecompressedSection &Sec) {
443 error("cannot write compressed section '" + Sec.Name + "' ");
446 void DecompressedSection::accept(SectionVisitor &Visitor) const {
447 Visitor.visit(*this);
450 void DecompressedSection::accept(MutableSectionVisitor &Visitor) {
451 Visitor.visit(*this);
454 void OwnedDataSection::accept(SectionVisitor &Visitor) const {
455 Visitor.visit(*this);
458 void OwnedDataSection::accept(MutableSectionVisitor &Visitor) {
459 Visitor.visit(*this);
462 void OwnedDataSection::appendHexData(StringRef HexData) {
463 assert((HexData.size() & 1) == 0);
464 while (!HexData.empty()) {
465 Data.push_back(checkedGetHex<uint8_t>(HexData.take_front(2)));
466 HexData = HexData.drop_front(2);
468 Size = Data.size();
471 void BinarySectionWriter::visit(const CompressedSection &Sec) {
472 error("cannot write compressed section '" + Sec.Name + "' ");
475 template <class ELFT>
476 void ELFSectionWriter<ELFT>::visit(const CompressedSection &Sec) {
477 uint8_t *Buf = Out.getBufferStart() + Sec.Offset;
478 if (Sec.CompressionType == DebugCompressionType::None) {
479 std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(), Buf);
480 return;
483 if (Sec.CompressionType == DebugCompressionType::GNU) {
484 const char *Magic = "ZLIB";
485 memcpy(Buf, Magic, strlen(Magic));
486 Buf += strlen(Magic);
487 const uint64_t DecompressedSize =
488 support::endian::read64be(&Sec.DecompressedSize);
489 memcpy(Buf, &DecompressedSize, sizeof(DecompressedSize));
490 Buf += sizeof(DecompressedSize);
491 } else {
492 Elf_Chdr_Impl<ELFT> Chdr;
493 Chdr.ch_type = ELF::ELFCOMPRESS_ZLIB;
494 Chdr.ch_size = Sec.DecompressedSize;
495 Chdr.ch_addralign = Sec.DecompressedAlign;
496 memcpy(Buf, &Chdr, sizeof(Chdr));
497 Buf += sizeof(Chdr);
500 std::copy(Sec.CompressedData.begin(), Sec.CompressedData.end(), Buf);
503 CompressedSection::CompressedSection(const SectionBase &Sec,
504 DebugCompressionType CompressionType)
505 : SectionBase(Sec), CompressionType(CompressionType),
506 DecompressedSize(Sec.OriginalData.size()), DecompressedAlign(Sec.Align) {
507 if (Error E = zlib::compress(
508 StringRef(reinterpret_cast<const char *>(OriginalData.data()),
509 OriginalData.size()),
510 CompressedData))
511 reportError(Name, std::move(E));
513 size_t ChdrSize;
514 if (CompressionType == DebugCompressionType::GNU) {
515 Name = ".z" + Sec.Name.substr(1);
516 ChdrSize = sizeof("ZLIB") - 1 + sizeof(uint64_t);
517 } else {
518 Flags |= ELF::SHF_COMPRESSED;
519 ChdrSize =
520 std::max(std::max(sizeof(object::Elf_Chdr_Impl<object::ELF64LE>),
521 sizeof(object::Elf_Chdr_Impl<object::ELF64BE>)),
522 std::max(sizeof(object::Elf_Chdr_Impl<object::ELF32LE>),
523 sizeof(object::Elf_Chdr_Impl<object::ELF32BE>)));
525 Size = ChdrSize + CompressedData.size();
526 Align = 8;
529 CompressedSection::CompressedSection(ArrayRef<uint8_t> CompressedData,
530 uint64_t DecompressedSize,
531 uint64_t DecompressedAlign)
532 : CompressionType(DebugCompressionType::None),
533 DecompressedSize(DecompressedSize), DecompressedAlign(DecompressedAlign) {
534 OriginalData = CompressedData;
537 void CompressedSection::accept(SectionVisitor &Visitor) const {
538 Visitor.visit(*this);
541 void CompressedSection::accept(MutableSectionVisitor &Visitor) {
542 Visitor.visit(*this);
545 void StringTableSection::addString(StringRef Name) { StrTabBuilder.add(Name); }
547 uint32_t StringTableSection::findIndex(StringRef Name) const {
548 return StrTabBuilder.getOffset(Name);
551 void StringTableSection::prepareForLayout() {
552 StrTabBuilder.finalize();
553 Size = StrTabBuilder.getSize();
556 void SectionWriter::visit(const StringTableSection &Sec) {
557 Sec.StrTabBuilder.write(Out.getBufferStart() + Sec.Offset);
560 void StringTableSection::accept(SectionVisitor &Visitor) const {
561 Visitor.visit(*this);
564 void StringTableSection::accept(MutableSectionVisitor &Visitor) {
565 Visitor.visit(*this);
568 template <class ELFT>
569 void ELFSectionWriter<ELFT>::visit(const SectionIndexSection &Sec) {
570 uint8_t *Buf = Out.getBufferStart() + Sec.Offset;
571 llvm::copy(Sec.Indexes, reinterpret_cast<Elf_Word *>(Buf));
574 void SectionIndexSection::initialize(SectionTableRef SecTable) {
575 Size = 0;
576 setSymTab(SecTable.getSectionOfType<SymbolTableSection>(
577 Link,
578 "Link field value " + Twine(Link) + " in section " + Name + " is invalid",
579 "Link field value " + Twine(Link) + " in section " + Name +
580 " is not a symbol table"));
581 Symbols->setShndxTable(this);
584 void SectionIndexSection::finalize() { Link = Symbols->Index; }
586 void SectionIndexSection::accept(SectionVisitor &Visitor) const {
587 Visitor.visit(*this);
590 void SectionIndexSection::accept(MutableSectionVisitor &Visitor) {
591 Visitor.visit(*this);
594 static bool isValidReservedSectionIndex(uint16_t Index, uint16_t Machine) {
595 switch (Index) {
596 case SHN_ABS:
597 case SHN_COMMON:
598 return true;
601 if (Machine == EM_AMDGPU) {
602 return Index == SHN_AMDGPU_LDS;
605 if (Machine == EM_HEXAGON) {
606 switch (Index) {
607 case SHN_HEXAGON_SCOMMON:
608 case SHN_HEXAGON_SCOMMON_2:
609 case SHN_HEXAGON_SCOMMON_4:
610 case SHN_HEXAGON_SCOMMON_8:
611 return true;
614 return false;
617 // Large indexes force us to clarify exactly what this function should do. This
618 // function should return the value that will appear in st_shndx when written
619 // out.
620 uint16_t Symbol::getShndx() const {
621 if (DefinedIn != nullptr) {
622 if (DefinedIn->Index >= SHN_LORESERVE)
623 return SHN_XINDEX;
624 return DefinedIn->Index;
627 if (ShndxType == SYMBOL_SIMPLE_INDEX) {
628 // This means that we don't have a defined section but we do need to
629 // output a legitimate section index.
630 return SHN_UNDEF;
633 assert(ShndxType == SYMBOL_ABS || ShndxType == SYMBOL_COMMON ||
634 (ShndxType >= SYMBOL_LOPROC && ShndxType <= SYMBOL_HIPROC) ||
635 (ShndxType >= SYMBOL_LOOS && ShndxType <= SYMBOL_HIOS));
636 return static_cast<uint16_t>(ShndxType);
639 bool Symbol::isCommon() const { return getShndx() == SHN_COMMON; }
641 void SymbolTableSection::assignIndices() {
642 uint32_t Index = 0;
643 for (auto &Sym : Symbols)
644 Sym->Index = Index++;
647 void SymbolTableSection::addSymbol(Twine Name, uint8_t Bind, uint8_t Type,
648 SectionBase *DefinedIn, uint64_t Value,
649 uint8_t Visibility, uint16_t Shndx,
650 uint64_t SymbolSize) {
651 Symbol Sym;
652 Sym.Name = Name.str();
653 Sym.Binding = Bind;
654 Sym.Type = Type;
655 Sym.DefinedIn = DefinedIn;
656 if (DefinedIn != nullptr)
657 DefinedIn->HasSymbol = true;
658 if (DefinedIn == nullptr) {
659 if (Shndx >= SHN_LORESERVE)
660 Sym.ShndxType = static_cast<SymbolShndxType>(Shndx);
661 else
662 Sym.ShndxType = SYMBOL_SIMPLE_INDEX;
664 Sym.Value = Value;
665 Sym.Visibility = Visibility;
666 Sym.Size = SymbolSize;
667 Sym.Index = Symbols.size();
668 Symbols.emplace_back(std::make_unique<Symbol>(Sym));
669 Size += this->EntrySize;
672 Error SymbolTableSection::removeSectionReferences(
673 bool AllowBrokenLinks,
674 function_ref<bool(const SectionBase *)> ToRemove) {
675 if (ToRemove(SectionIndexTable))
676 SectionIndexTable = nullptr;
677 if (ToRemove(SymbolNames)) {
678 if (!AllowBrokenLinks)
679 return createStringError(
680 llvm::errc::invalid_argument,
681 "string table '%s' cannot be removed because it is "
682 "referenced by the symbol table '%s'",
683 SymbolNames->Name.data(), this->Name.data());
684 SymbolNames = nullptr;
686 return removeSymbols(
687 [ToRemove](const Symbol &Sym) { return ToRemove(Sym.DefinedIn); });
690 void SymbolTableSection::updateSymbols(function_ref<void(Symbol &)> Callable) {
691 std::for_each(std::begin(Symbols) + 1, std::end(Symbols),
692 [Callable](SymPtr &Sym) { Callable(*Sym); });
693 std::stable_partition(
694 std::begin(Symbols), std::end(Symbols),
695 [](const SymPtr &Sym) { return Sym->Binding == STB_LOCAL; });
696 assignIndices();
699 Error SymbolTableSection::removeSymbols(
700 function_ref<bool(const Symbol &)> ToRemove) {
701 Symbols.erase(
702 std::remove_if(std::begin(Symbols) + 1, std::end(Symbols),
703 [ToRemove](const SymPtr &Sym) { return ToRemove(*Sym); }),
704 std::end(Symbols));
705 Size = Symbols.size() * EntrySize;
706 assignIndices();
707 return Error::success();
710 void SymbolTableSection::replaceSectionReferences(
711 const DenseMap<SectionBase *, SectionBase *> &FromTo) {
712 for (std::unique_ptr<Symbol> &Sym : Symbols)
713 if (SectionBase *To = FromTo.lookup(Sym->DefinedIn))
714 Sym->DefinedIn = To;
717 void SymbolTableSection::initialize(SectionTableRef SecTable) {
718 Size = 0;
719 setStrTab(SecTable.getSectionOfType<StringTableSection>(
720 Link,
721 "Symbol table has link index of " + Twine(Link) +
722 " which is not a valid index",
723 "Symbol table has link index of " + Twine(Link) +
724 " which is not a string table"));
727 void SymbolTableSection::finalize() {
728 uint32_t MaxLocalIndex = 0;
729 for (std::unique_ptr<Symbol> &Sym : Symbols) {
730 Sym->NameIndex =
731 SymbolNames == nullptr ? 0 : SymbolNames->findIndex(Sym->Name);
732 if (Sym->Binding == STB_LOCAL)
733 MaxLocalIndex = std::max(MaxLocalIndex, Sym->Index);
735 // Now we need to set the Link and Info fields.
736 Link = SymbolNames == nullptr ? 0 : SymbolNames->Index;
737 Info = MaxLocalIndex + 1;
740 void SymbolTableSection::prepareForLayout() {
741 // Reserve proper amount of space in section index table, so we can
742 // layout sections correctly. We will fill the table with correct
743 // indexes later in fillShdnxTable.
744 if (SectionIndexTable)
745 SectionIndexTable->reserve(Symbols.size());
747 // Add all of our strings to SymbolNames so that SymbolNames has the right
748 // size before layout is decided.
749 // If the symbol names section has been removed, don't try to add strings to
750 // the table.
751 if (SymbolNames != nullptr)
752 for (std::unique_ptr<Symbol> &Sym : Symbols)
753 SymbolNames->addString(Sym->Name);
756 void SymbolTableSection::fillShndxTable() {
757 if (SectionIndexTable == nullptr)
758 return;
759 // Fill section index table with real section indexes. This function must
760 // be called after assignOffsets.
761 for (const std::unique_ptr<Symbol> &Sym : Symbols) {
762 if (Sym->DefinedIn != nullptr && Sym->DefinedIn->Index >= SHN_LORESERVE)
763 SectionIndexTable->addIndex(Sym->DefinedIn->Index);
764 else
765 SectionIndexTable->addIndex(SHN_UNDEF);
769 const Symbol *SymbolTableSection::getSymbolByIndex(uint32_t Index) const {
770 if (Symbols.size() <= Index)
771 error("invalid symbol index: " + Twine(Index));
772 return Symbols[Index].get();
775 Symbol *SymbolTableSection::getSymbolByIndex(uint32_t Index) {
776 return const_cast<Symbol *>(
777 static_cast<const SymbolTableSection *>(this)->getSymbolByIndex(Index));
780 template <class ELFT>
781 void ELFSectionWriter<ELFT>::visit(const SymbolTableSection &Sec) {
782 Elf_Sym *Sym = reinterpret_cast<Elf_Sym *>(Out.getBufferStart() + Sec.Offset);
783 // Loop though symbols setting each entry of the symbol table.
784 for (const std::unique_ptr<Symbol> &Symbol : Sec.Symbols) {
785 Sym->st_name = Symbol->NameIndex;
786 Sym->st_value = Symbol->Value;
787 Sym->st_size = Symbol->Size;
788 Sym->st_other = Symbol->Visibility;
789 Sym->setBinding(Symbol->Binding);
790 Sym->setType(Symbol->Type);
791 Sym->st_shndx = Symbol->getShndx();
792 ++Sym;
796 void SymbolTableSection::accept(SectionVisitor &Visitor) const {
797 Visitor.visit(*this);
800 void SymbolTableSection::accept(MutableSectionVisitor &Visitor) {
801 Visitor.visit(*this);
804 Error RelocationSection::removeSectionReferences(
805 bool AllowBrokenLinks,
806 function_ref<bool(const SectionBase *)> ToRemove) {
807 if (ToRemove(Symbols)) {
808 if (!AllowBrokenLinks)
809 return createStringError(
810 llvm::errc::invalid_argument,
811 "symbol table '%s' cannot be removed because it is "
812 "referenced by the relocation section '%s'",
813 Symbols->Name.data(), this->Name.data());
814 Symbols = nullptr;
817 for (const Relocation &R : Relocations) {
818 if (!R.RelocSymbol->DefinedIn || !ToRemove(R.RelocSymbol->DefinedIn))
819 continue;
820 return createStringError(llvm::errc::invalid_argument,
821 "section '%s' cannot be removed: (%s+0x%" PRIx64
822 ") has relocation against symbol '%s'",
823 R.RelocSymbol->DefinedIn->Name.data(),
824 SecToApplyRel->Name.data(), R.Offset,
825 R.RelocSymbol->Name.c_str());
828 return Error::success();
831 template <class SymTabType>
832 void RelocSectionWithSymtabBase<SymTabType>::initialize(
833 SectionTableRef SecTable) {
834 if (Link != SHN_UNDEF)
835 setSymTab(SecTable.getSectionOfType<SymTabType>(
836 Link,
837 "Link field value " + Twine(Link) + " in section " + Name +
838 " is invalid",
839 "Link field value " + Twine(Link) + " in section " + Name +
840 " is not a symbol table"));
842 if (Info != SHN_UNDEF)
843 setSection(SecTable.getSection(Info, "Info field value " + Twine(Info) +
844 " in section " + Name +
845 " is invalid"));
846 else
847 setSection(nullptr);
850 template <class SymTabType>
851 void RelocSectionWithSymtabBase<SymTabType>::finalize() {
852 this->Link = Symbols ? Symbols->Index : 0;
854 if (SecToApplyRel != nullptr)
855 this->Info = SecToApplyRel->Index;
858 template <class ELFT>
859 static void setAddend(Elf_Rel_Impl<ELFT, false> &Rel, uint64_t Addend) {}
861 template <class ELFT>
862 static void setAddend(Elf_Rel_Impl<ELFT, true> &Rela, uint64_t Addend) {
863 Rela.r_addend = Addend;
866 template <class RelRange, class T>
867 static void writeRel(const RelRange &Relocations, T *Buf) {
868 for (const auto &Reloc : Relocations) {
869 Buf->r_offset = Reloc.Offset;
870 setAddend(*Buf, Reloc.Addend);
871 Buf->setSymbolAndType(Reloc.RelocSymbol->Index, Reloc.Type, false);
872 ++Buf;
876 template <class ELFT>
877 void ELFSectionWriter<ELFT>::visit(const RelocationSection &Sec) {
878 uint8_t *Buf = Out.getBufferStart() + Sec.Offset;
879 if (Sec.Type == SHT_REL)
880 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rel *>(Buf));
881 else
882 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rela *>(Buf));
885 void RelocationSection::accept(SectionVisitor &Visitor) const {
886 Visitor.visit(*this);
889 void RelocationSection::accept(MutableSectionVisitor &Visitor) {
890 Visitor.visit(*this);
893 Error RelocationSection::removeSymbols(
894 function_ref<bool(const Symbol &)> ToRemove) {
895 for (const Relocation &Reloc : Relocations)
896 if (ToRemove(*Reloc.RelocSymbol))
897 return createStringError(
898 llvm::errc::invalid_argument,
899 "not stripping symbol '%s' because it is named in a relocation",
900 Reloc.RelocSymbol->Name.data());
901 return Error::success();
904 void RelocationSection::markSymbols() {
905 for (const Relocation &Reloc : Relocations)
906 Reloc.RelocSymbol->Referenced = true;
909 void RelocationSection::replaceSectionReferences(
910 const DenseMap<SectionBase *, SectionBase *> &FromTo) {
911 // Update the target section if it was replaced.
912 if (SectionBase *To = FromTo.lookup(SecToApplyRel))
913 SecToApplyRel = To;
916 void SectionWriter::visit(const DynamicRelocationSection &Sec) {
917 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset);
920 void DynamicRelocationSection::accept(SectionVisitor &Visitor) const {
921 Visitor.visit(*this);
924 void DynamicRelocationSection::accept(MutableSectionVisitor &Visitor) {
925 Visitor.visit(*this);
928 Error DynamicRelocationSection::removeSectionReferences(
929 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
930 if (ToRemove(Symbols)) {
931 if (!AllowBrokenLinks)
932 return createStringError(
933 llvm::errc::invalid_argument,
934 "symbol table '%s' cannot be removed because it is "
935 "referenced by the relocation section '%s'",
936 Symbols->Name.data(), this->Name.data());
937 Symbols = nullptr;
940 // SecToApplyRel contains a section referenced by sh_info field. It keeps
941 // a section to which the relocation section applies. When we remove any
942 // sections we also remove their relocation sections. Since we do that much
943 // earlier, this assert should never be triggered.
944 assert(!SecToApplyRel || !ToRemove(SecToApplyRel));
945 return Error::success();
948 Error Section::removeSectionReferences(
949 bool AllowBrokenDependency,
950 function_ref<bool(const SectionBase *)> ToRemove) {
951 if (ToRemove(LinkSection)) {
952 if (!AllowBrokenDependency)
953 return createStringError(llvm::errc::invalid_argument,
954 "section '%s' cannot be removed because it is "
955 "referenced by the section '%s'",
956 LinkSection->Name.data(), this->Name.data());
957 LinkSection = nullptr;
959 return Error::success();
962 void GroupSection::finalize() {
963 this->Info = Sym->Index;
964 this->Link = SymTab->Index;
967 Error GroupSection::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
968 if (ToRemove(*Sym))
969 return createStringError(llvm::errc::invalid_argument,
970 "symbol '%s' cannot be removed because it is "
971 "referenced by the section '%s[%d]'",
972 Sym->Name.data(), this->Name.data(), this->Index);
973 return Error::success();
976 void GroupSection::markSymbols() {
977 if (Sym)
978 Sym->Referenced = true;
981 void GroupSection::replaceSectionReferences(
982 const DenseMap<SectionBase *, SectionBase *> &FromTo) {
983 for (SectionBase *&Sec : GroupMembers)
984 if (SectionBase *To = FromTo.lookup(Sec))
985 Sec = To;
988 void Section::initialize(SectionTableRef SecTable) {
989 if (Link == ELF::SHN_UNDEF)
990 return;
991 LinkSection =
992 SecTable.getSection(Link, "Link field value " + Twine(Link) +
993 " in section " + Name + " is invalid");
994 if (LinkSection->Type == ELF::SHT_SYMTAB)
995 LinkSection = nullptr;
998 void Section::finalize() { this->Link = LinkSection ? LinkSection->Index : 0; }
1000 void GnuDebugLinkSection::init(StringRef File) {
1001 FileName = sys::path::filename(File);
1002 // The format for the .gnu_debuglink starts with the file name and is
1003 // followed by a null terminator and then the CRC32 of the file. The CRC32
1004 // should be 4 byte aligned. So we add the FileName size, a 1 for the null
1005 // byte, and then finally push the size to alignment and add 4.
1006 Size = alignTo(FileName.size() + 1, 4) + 4;
1007 // The CRC32 will only be aligned if we align the whole section.
1008 Align = 4;
1009 Type = ELF::SHT_PROGBITS;
1010 Name = ".gnu_debuglink";
1011 // For sections not found in segments, OriginalOffset is only used to
1012 // establish the order that sections should go in. By using the maximum
1013 // possible offset we cause this section to wind up at the end.
1014 OriginalOffset = std::numeric_limits<uint64_t>::max();
1017 GnuDebugLinkSection::GnuDebugLinkSection(StringRef File,
1018 uint32_t PrecomputedCRC)
1019 : FileName(File), CRC32(PrecomputedCRC) {
1020 init(File);
1023 template <class ELFT>
1024 void ELFSectionWriter<ELFT>::visit(const GnuDebugLinkSection &Sec) {
1025 unsigned char *Buf = Out.getBufferStart() + Sec.Offset;
1026 Elf_Word *CRC =
1027 reinterpret_cast<Elf_Word *>(Buf + Sec.Size - sizeof(Elf_Word));
1028 *CRC = Sec.CRC32;
1029 llvm::copy(Sec.FileName, Buf);
1032 void GnuDebugLinkSection::accept(SectionVisitor &Visitor) const {
1033 Visitor.visit(*this);
1036 void GnuDebugLinkSection::accept(MutableSectionVisitor &Visitor) {
1037 Visitor.visit(*this);
1040 template <class ELFT>
1041 void ELFSectionWriter<ELFT>::visit(const GroupSection &Sec) {
1042 ELF::Elf32_Word *Buf =
1043 reinterpret_cast<ELF::Elf32_Word *>(Out.getBufferStart() + Sec.Offset);
1044 *Buf++ = Sec.FlagWord;
1045 for (SectionBase *S : Sec.GroupMembers)
1046 support::endian::write32<ELFT::TargetEndianness>(Buf++, S->Index);
1049 void GroupSection::accept(SectionVisitor &Visitor) const {
1050 Visitor.visit(*this);
1053 void GroupSection::accept(MutableSectionVisitor &Visitor) {
1054 Visitor.visit(*this);
1057 // Returns true IFF a section is wholly inside the range of a segment
1058 static bool sectionWithinSegment(const SectionBase &Sec, const Segment &Seg) {
1059 // If a section is empty it should be treated like it has a size of 1. This is
1060 // to clarify the case when an empty section lies on a boundary between two
1061 // segments and ensures that the section "belongs" to the second segment and
1062 // not the first.
1063 uint64_t SecSize = Sec.Size ? Sec.Size : 1;
1065 if (Sec.Type == SHT_NOBITS) {
1066 if (!(Sec.Flags & SHF_ALLOC))
1067 return false;
1069 bool SectionIsTLS = Sec.Flags & SHF_TLS;
1070 bool SegmentIsTLS = Seg.Type == PT_TLS;
1071 if (SectionIsTLS != SegmentIsTLS)
1072 return false;
1074 return Seg.VAddr <= Sec.Addr &&
1075 Seg.VAddr + Seg.MemSize >= Sec.Addr + SecSize;
1078 return Seg.Offset <= Sec.OriginalOffset &&
1079 Seg.Offset + Seg.FileSize >= Sec.OriginalOffset + SecSize;
1082 // Returns true IFF a segment's original offset is inside of another segment's
1083 // range.
1084 static bool segmentOverlapsSegment(const Segment &Child,
1085 const Segment &Parent) {
1087 return Parent.OriginalOffset <= Child.OriginalOffset &&
1088 Parent.OriginalOffset + Parent.FileSize > Child.OriginalOffset;
1091 static bool compareSegmentsByOffset(const Segment *A, const Segment *B) {
1092 // Any segment without a parent segment should come before a segment
1093 // that has a parent segment.
1094 if (A->OriginalOffset < B->OriginalOffset)
1095 return true;
1096 if (A->OriginalOffset > B->OriginalOffset)
1097 return false;
1098 return A->Index < B->Index;
1101 static bool compareSegmentsByPAddr(const Segment *A, const Segment *B) {
1102 if (A->PAddr < B->PAddr)
1103 return true;
1104 if (A->PAddr > B->PAddr)
1105 return false;
1106 return A->Index < B->Index;
1109 void BasicELFBuilder::initFileHeader() {
1110 Obj->Flags = 0x0;
1111 Obj->Type = ET_REL;
1112 Obj->OSABI = ELFOSABI_NONE;
1113 Obj->ABIVersion = 0;
1114 Obj->Entry = 0x0;
1115 Obj->Machine = EMachine;
1116 Obj->Version = 1;
1119 void BasicELFBuilder::initHeaderSegment() { Obj->ElfHdrSegment.Index = 0; }
1121 StringTableSection *BasicELFBuilder::addStrTab() {
1122 auto &StrTab = Obj->addSection<StringTableSection>();
1123 StrTab.Name = ".strtab";
1125 Obj->SectionNames = &StrTab;
1126 return &StrTab;
1129 SymbolTableSection *BasicELFBuilder::addSymTab(StringTableSection *StrTab) {
1130 auto &SymTab = Obj->addSection<SymbolTableSection>();
1132 SymTab.Name = ".symtab";
1133 SymTab.Link = StrTab->Index;
1135 // The symbol table always needs a null symbol
1136 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0);
1138 Obj->SymbolTable = &SymTab;
1139 return &SymTab;
1142 void BasicELFBuilder::initSections() {
1143 for (SectionBase &Sec : Obj->sections())
1144 Sec.initialize(Obj->sections());
1147 void BinaryELFBuilder::addData(SymbolTableSection *SymTab) {
1148 auto Data = ArrayRef<uint8_t>(
1149 reinterpret_cast<const uint8_t *>(MemBuf->getBufferStart()),
1150 MemBuf->getBufferSize());
1151 auto &DataSection = Obj->addSection<Section>(Data);
1152 DataSection.Name = ".data";
1153 DataSection.Type = ELF::SHT_PROGBITS;
1154 DataSection.Size = Data.size();
1155 DataSection.Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE;
1157 std::string SanitizedFilename = MemBuf->getBufferIdentifier().str();
1158 std::replace_if(std::begin(SanitizedFilename), std::end(SanitizedFilename),
1159 [](char C) { return !isalnum(C); }, '_');
1160 Twine Prefix = Twine("_binary_") + SanitizedFilename;
1162 SymTab->addSymbol(Prefix + "_start", STB_GLOBAL, STT_NOTYPE, &DataSection,
1163 /*Value=*/0, NewSymbolVisibility, 0, 0);
1164 SymTab->addSymbol(Prefix + "_end", STB_GLOBAL, STT_NOTYPE, &DataSection,
1165 /*Value=*/DataSection.Size, NewSymbolVisibility, 0, 0);
1166 SymTab->addSymbol(Prefix + "_size", STB_GLOBAL, STT_NOTYPE, nullptr,
1167 /*Value=*/DataSection.Size, NewSymbolVisibility, SHN_ABS,
1171 std::unique_ptr<Object> BinaryELFBuilder::build() {
1172 initFileHeader();
1173 initHeaderSegment();
1175 SymbolTableSection *SymTab = addSymTab(addStrTab());
1176 initSections();
1177 addData(SymTab);
1179 return std::move(Obj);
1182 // Adds sections from IHEX data file. Data should have been
1183 // fully validated by this time.
1184 void IHexELFBuilder::addDataSections() {
1185 OwnedDataSection *Section = nullptr;
1186 uint64_t SegmentAddr = 0, BaseAddr = 0;
1187 uint32_t SecNo = 1;
1189 for (const IHexRecord &R : Records) {
1190 uint64_t RecAddr;
1191 switch (R.Type) {
1192 case IHexRecord::Data:
1193 // Ignore empty data records
1194 if (R.HexData.empty())
1195 continue;
1196 RecAddr = R.Addr + SegmentAddr + BaseAddr;
1197 if (!Section || Section->Addr + Section->Size != RecAddr)
1198 // OriginalOffset field is only used to sort section properly, so
1199 // instead of keeping track of real offset in IHEX file, we use
1200 // section number.
1201 Section = &Obj->addSection<OwnedDataSection>(
1202 ".sec" + std::to_string(SecNo++), RecAddr,
1203 ELF::SHF_ALLOC | ELF::SHF_WRITE, SecNo);
1204 Section->appendHexData(R.HexData);
1205 break;
1206 case IHexRecord::EndOfFile:
1207 break;
1208 case IHexRecord::SegmentAddr:
1209 // 20-bit segment address.
1210 SegmentAddr = checkedGetHex<uint16_t>(R.HexData) << 4;
1211 break;
1212 case IHexRecord::StartAddr80x86:
1213 case IHexRecord::StartAddr:
1214 Obj->Entry = checkedGetHex<uint32_t>(R.HexData);
1215 assert(Obj->Entry <= 0xFFFFFU);
1216 break;
1217 case IHexRecord::ExtendedAddr:
1218 // 16-31 bits of linear base address
1219 BaseAddr = checkedGetHex<uint16_t>(R.HexData) << 16;
1220 break;
1221 default:
1222 llvm_unreachable("unknown record type");
1227 std::unique_ptr<Object> IHexELFBuilder::build() {
1228 initFileHeader();
1229 initHeaderSegment();
1230 StringTableSection *StrTab = addStrTab();
1231 addSymTab(StrTab);
1232 initSections();
1233 addDataSections();
1235 return std::move(Obj);
1238 template <class ELFT> void ELFBuilder<ELFT>::setParentSegment(Segment &Child) {
1239 for (Segment &Parent : Obj.segments()) {
1240 // Every segment will overlap with itself but we don't want a segment to
1241 // be it's own parent so we avoid that situation.
1242 if (&Child != &Parent && segmentOverlapsSegment(Child, Parent)) {
1243 // We want a canonical "most parental" segment but this requires
1244 // inspecting the ParentSegment.
1245 if (compareSegmentsByOffset(&Parent, &Child))
1246 if (Child.ParentSegment == nullptr ||
1247 compareSegmentsByOffset(&Parent, Child.ParentSegment)) {
1248 Child.ParentSegment = &Parent;
1254 template <class ELFT> void ELFBuilder<ELFT>::findEhdrOffset() {
1255 if (!ExtractPartition)
1256 return;
1258 for (const SectionBase &Sec : Obj.sections()) {
1259 if (Sec.Type == SHT_LLVM_PART_EHDR && Sec.Name == *ExtractPartition) {
1260 EhdrOffset = Sec.Offset;
1261 return;
1264 error("could not find partition named '" + *ExtractPartition + "'");
1267 template <class ELFT>
1268 void ELFBuilder<ELFT>::readProgramHeaders(const ELFFile<ELFT> &HeadersFile) {
1269 uint32_t Index = 0;
1270 for (const auto &Phdr : unwrapOrError(HeadersFile.program_headers())) {
1271 if (Phdr.p_offset + Phdr.p_filesz > HeadersFile.getBufSize())
1272 error("program header with offset 0x" + Twine::utohexstr(Phdr.p_offset) +
1273 " and file size 0x" + Twine::utohexstr(Phdr.p_filesz) +
1274 " goes past the end of the file");
1276 ArrayRef<uint8_t> Data{HeadersFile.base() + Phdr.p_offset,
1277 (size_t)Phdr.p_filesz};
1278 Segment &Seg = Obj.addSegment(Data);
1279 Seg.Type = Phdr.p_type;
1280 Seg.Flags = Phdr.p_flags;
1281 Seg.OriginalOffset = Phdr.p_offset + EhdrOffset;
1282 Seg.Offset = Phdr.p_offset + EhdrOffset;
1283 Seg.VAddr = Phdr.p_vaddr;
1284 Seg.PAddr = Phdr.p_paddr;
1285 Seg.FileSize = Phdr.p_filesz;
1286 Seg.MemSize = Phdr.p_memsz;
1287 Seg.Align = Phdr.p_align;
1288 Seg.Index = Index++;
1289 for (SectionBase &Sec : Obj.sections())
1290 if (sectionWithinSegment(Sec, Seg)) {
1291 Seg.addSection(&Sec);
1292 if (!Sec.ParentSegment || Sec.ParentSegment->Offset > Seg.Offset)
1293 Sec.ParentSegment = &Seg;
1297 auto &ElfHdr = Obj.ElfHdrSegment;
1298 ElfHdr.Index = Index++;
1299 ElfHdr.OriginalOffset = ElfHdr.Offset = EhdrOffset;
1301 const auto &Ehdr = *HeadersFile.getHeader();
1302 auto &PrHdr = Obj.ProgramHdrSegment;
1303 PrHdr.Type = PT_PHDR;
1304 PrHdr.Flags = 0;
1305 // The spec requires us to have p_vaddr % p_align == p_offset % p_align.
1306 // Whereas this works automatically for ElfHdr, here OriginalOffset is
1307 // always non-zero and to ensure the equation we assign the same value to
1308 // VAddr as well.
1309 PrHdr.OriginalOffset = PrHdr.Offset = PrHdr.VAddr = EhdrOffset + Ehdr.e_phoff;
1310 PrHdr.PAddr = 0;
1311 PrHdr.FileSize = PrHdr.MemSize = Ehdr.e_phentsize * Ehdr.e_phnum;
1312 // The spec requires us to naturally align all the fields.
1313 PrHdr.Align = sizeof(Elf_Addr);
1314 PrHdr.Index = Index++;
1316 // Now we do an O(n^2) loop through the segments in order to match up
1317 // segments.
1318 for (Segment &Child : Obj.segments())
1319 setParentSegment(Child);
1320 setParentSegment(ElfHdr);
1321 setParentSegment(PrHdr);
1324 template <class ELFT>
1325 void ELFBuilder<ELFT>::initGroupSection(GroupSection *GroupSec) {
1326 if (GroupSec->Align % sizeof(ELF::Elf32_Word) != 0)
1327 error("invalid alignment " + Twine(GroupSec->Align) + " of group section '" +
1328 GroupSec->Name + "'");
1329 SectionTableRef SecTable = Obj.sections();
1330 auto SymTab = SecTable.template getSectionOfType<SymbolTableSection>(
1331 GroupSec->Link,
1332 "link field value '" + Twine(GroupSec->Link) + "' in section '" +
1333 GroupSec->Name + "' is invalid",
1334 "link field value '" + Twine(GroupSec->Link) + "' in section '" +
1335 GroupSec->Name + "' is not a symbol table");
1336 Symbol *Sym = SymTab->getSymbolByIndex(GroupSec->Info);
1337 if (!Sym)
1338 error("info field value '" + Twine(GroupSec->Info) + "' in section '" +
1339 GroupSec->Name + "' is not a valid symbol index");
1340 GroupSec->setSymTab(SymTab);
1341 GroupSec->setSymbol(Sym);
1342 if (GroupSec->Contents.size() % sizeof(ELF::Elf32_Word) ||
1343 GroupSec->Contents.empty())
1344 error("the content of the section " + GroupSec->Name + " is malformed");
1345 const ELF::Elf32_Word *Word =
1346 reinterpret_cast<const ELF::Elf32_Word *>(GroupSec->Contents.data());
1347 const ELF::Elf32_Word *End =
1348 Word + GroupSec->Contents.size() / sizeof(ELF::Elf32_Word);
1349 GroupSec->setFlagWord(*Word++);
1350 for (; Word != End; ++Word) {
1351 uint32_t Index = support::endian::read32<ELFT::TargetEndianness>(Word);
1352 GroupSec->addMember(SecTable.getSection(
1353 Index, "group member index " + Twine(Index) + " in section '" +
1354 GroupSec->Name + "' is invalid"));
1358 template <class ELFT>
1359 void ELFBuilder<ELFT>::initSymbolTable(SymbolTableSection *SymTab) {
1360 const Elf_Shdr &Shdr = *unwrapOrError(ElfFile.getSection(SymTab->Index));
1361 StringRef StrTabData = unwrapOrError(ElfFile.getStringTableForSymtab(Shdr));
1362 ArrayRef<Elf_Word> ShndxData;
1364 auto Symbols = unwrapOrError(ElfFile.symbols(&Shdr));
1365 for (const auto &Sym : Symbols) {
1366 SectionBase *DefSection = nullptr;
1367 StringRef Name = unwrapOrError(Sym.getName(StrTabData));
1369 if (Sym.st_shndx == SHN_XINDEX) {
1370 if (SymTab->getShndxTable() == nullptr)
1371 error("symbol '" + Name +
1372 "' has index SHN_XINDEX but no SHT_SYMTAB_SHNDX section exists");
1373 if (ShndxData.data() == nullptr) {
1374 const Elf_Shdr &ShndxSec =
1375 *unwrapOrError(ElfFile.getSection(SymTab->getShndxTable()->Index));
1376 ShndxData = unwrapOrError(
1377 ElfFile.template getSectionContentsAsArray<Elf_Word>(&ShndxSec));
1378 if (ShndxData.size() != Symbols.size())
1379 error("symbol section index table does not have the same number of "
1380 "entries as the symbol table");
1382 Elf_Word Index = ShndxData[&Sym - Symbols.begin()];
1383 DefSection = Obj.sections().getSection(
1384 Index,
1385 "symbol '" + Name + "' has invalid section index " + Twine(Index));
1386 } else if (Sym.st_shndx >= SHN_LORESERVE) {
1387 if (!isValidReservedSectionIndex(Sym.st_shndx, Obj.Machine)) {
1388 error(
1389 "symbol '" + Name +
1390 "' has unsupported value greater than or equal to SHN_LORESERVE: " +
1391 Twine(Sym.st_shndx));
1393 } else if (Sym.st_shndx != SHN_UNDEF) {
1394 DefSection = Obj.sections().getSection(
1395 Sym.st_shndx, "symbol '" + Name +
1396 "' is defined has invalid section index " +
1397 Twine(Sym.st_shndx));
1400 SymTab->addSymbol(Name, Sym.getBinding(), Sym.getType(), DefSection,
1401 Sym.getValue(), Sym.st_other, Sym.st_shndx, Sym.st_size);
1405 template <class ELFT>
1406 static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, false> &Rel) {}
1408 template <class ELFT>
1409 static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, true> &Rela) {
1410 ToSet = Rela.r_addend;
1413 template <class T>
1414 static void initRelocations(RelocationSection *Relocs,
1415 SymbolTableSection *SymbolTable, T RelRange) {
1416 for (const auto &Rel : RelRange) {
1417 Relocation ToAdd;
1418 ToAdd.Offset = Rel.r_offset;
1419 getAddend(ToAdd.Addend, Rel);
1420 ToAdd.Type = Rel.getType(false);
1421 ToAdd.RelocSymbol = SymbolTable->getSymbolByIndex(Rel.getSymbol(false));
1422 Relocs->addRelocation(ToAdd);
1426 SectionBase *SectionTableRef::getSection(uint32_t Index, Twine ErrMsg) {
1427 if (Index == SHN_UNDEF || Index > Sections.size())
1428 error(ErrMsg);
1429 return Sections[Index - 1].get();
1432 template <class T>
1433 T *SectionTableRef::getSectionOfType(uint32_t Index, Twine IndexErrMsg,
1434 Twine TypeErrMsg) {
1435 if (T *Sec = dyn_cast<T>(getSection(Index, IndexErrMsg)))
1436 return Sec;
1437 error(TypeErrMsg);
1440 template <class ELFT>
1441 SectionBase &ELFBuilder<ELFT>::makeSection(const Elf_Shdr &Shdr) {
1442 ArrayRef<uint8_t> Data;
1443 switch (Shdr.sh_type) {
1444 case SHT_REL:
1445 case SHT_RELA:
1446 if (Shdr.sh_flags & SHF_ALLOC) {
1447 Data = unwrapOrError(ElfFile.getSectionContents(&Shdr));
1448 return Obj.addSection<DynamicRelocationSection>(Data);
1450 return Obj.addSection<RelocationSection>();
1451 case SHT_STRTAB:
1452 // If a string table is allocated we don't want to mess with it. That would
1453 // mean altering the memory image. There are no special link types or
1454 // anything so we can just use a Section.
1455 if (Shdr.sh_flags & SHF_ALLOC) {
1456 Data = unwrapOrError(ElfFile.getSectionContents(&Shdr));
1457 return Obj.addSection<Section>(Data);
1459 return Obj.addSection<StringTableSection>();
1460 case SHT_HASH:
1461 case SHT_GNU_HASH:
1462 // Hash tables should refer to SHT_DYNSYM which we're not going to change.
1463 // Because of this we don't need to mess with the hash tables either.
1464 Data = unwrapOrError(ElfFile.getSectionContents(&Shdr));
1465 return Obj.addSection<Section>(Data);
1466 case SHT_GROUP:
1467 Data = unwrapOrError(ElfFile.getSectionContents(&Shdr));
1468 return Obj.addSection<GroupSection>(Data);
1469 case SHT_DYNSYM:
1470 Data = unwrapOrError(ElfFile.getSectionContents(&Shdr));
1471 return Obj.addSection<DynamicSymbolTableSection>(Data);
1472 case SHT_DYNAMIC:
1473 Data = unwrapOrError(ElfFile.getSectionContents(&Shdr));
1474 return Obj.addSection<DynamicSection>(Data);
1475 case SHT_SYMTAB: {
1476 auto &SymTab = Obj.addSection<SymbolTableSection>();
1477 Obj.SymbolTable = &SymTab;
1478 return SymTab;
1480 case SHT_SYMTAB_SHNDX: {
1481 auto &ShndxSection = Obj.addSection<SectionIndexSection>();
1482 Obj.SectionIndexTable = &ShndxSection;
1483 return ShndxSection;
1485 case SHT_NOBITS:
1486 return Obj.addSection<Section>(Data);
1487 default: {
1488 Data = unwrapOrError(ElfFile.getSectionContents(&Shdr));
1490 StringRef Name = unwrapOrError(ElfFile.getSectionName(&Shdr));
1491 if (Name.startswith(".zdebug") || (Shdr.sh_flags & ELF::SHF_COMPRESSED)) {
1492 uint64_t DecompressedSize, DecompressedAlign;
1493 std::tie(DecompressedSize, DecompressedAlign) =
1494 getDecompressedSizeAndAlignment<ELFT>(Data);
1495 return Obj.addSection<CompressedSection>(Data, DecompressedSize,
1496 DecompressedAlign);
1499 return Obj.addSection<Section>(Data);
1504 template <class ELFT> void ELFBuilder<ELFT>::readSectionHeaders() {
1505 uint32_t Index = 0;
1506 for (const auto &Shdr : unwrapOrError(ElfFile.sections())) {
1507 if (Index == 0) {
1508 ++Index;
1509 continue;
1511 auto &Sec = makeSection(Shdr);
1512 Sec.Name = unwrapOrError(ElfFile.getSectionName(&Shdr));
1513 Sec.Type = Shdr.sh_type;
1514 Sec.Flags = Shdr.sh_flags;
1515 Sec.Addr = Shdr.sh_addr;
1516 Sec.Offset = Shdr.sh_offset;
1517 Sec.OriginalOffset = Shdr.sh_offset;
1518 Sec.Size = Shdr.sh_size;
1519 Sec.Link = Shdr.sh_link;
1520 Sec.Info = Shdr.sh_info;
1521 Sec.Align = Shdr.sh_addralign;
1522 Sec.EntrySize = Shdr.sh_entsize;
1523 Sec.Index = Index++;
1524 Sec.OriginalData =
1525 ArrayRef<uint8_t>(ElfFile.base() + Shdr.sh_offset,
1526 (Shdr.sh_type == SHT_NOBITS) ? 0 : Shdr.sh_size);
1530 template <class ELFT> void ELFBuilder<ELFT>::readSections() {
1531 // If a section index table exists we'll need to initialize it before we
1532 // initialize the symbol table because the symbol table might need to
1533 // reference it.
1534 if (Obj.SectionIndexTable)
1535 Obj.SectionIndexTable->initialize(Obj.sections());
1537 // Now that all of the sections have been added we can fill out some extra
1538 // details about symbol tables. We need the symbol table filled out before
1539 // any relocations.
1540 if (Obj.SymbolTable) {
1541 Obj.SymbolTable->initialize(Obj.sections());
1542 initSymbolTable(Obj.SymbolTable);
1545 // Now that all sections and symbols have been added we can add
1546 // relocations that reference symbols and set the link and info fields for
1547 // relocation sections.
1548 for (auto &Sec : Obj.sections()) {
1549 if (&Sec == Obj.SymbolTable)
1550 continue;
1551 Sec.initialize(Obj.sections());
1552 if (auto RelSec = dyn_cast<RelocationSection>(&Sec)) {
1553 auto Shdr = unwrapOrError(ElfFile.sections()).begin() + RelSec->Index;
1554 if (RelSec->Type == SHT_REL)
1555 initRelocations(RelSec, Obj.SymbolTable,
1556 unwrapOrError(ElfFile.rels(Shdr)));
1557 else
1558 initRelocations(RelSec, Obj.SymbolTable,
1559 unwrapOrError(ElfFile.relas(Shdr)));
1560 } else if (auto GroupSec = dyn_cast<GroupSection>(&Sec)) {
1561 initGroupSection(GroupSec);
1565 uint32_t ShstrIndex = ElfFile.getHeader()->e_shstrndx;
1566 if (ShstrIndex == SHN_XINDEX)
1567 ShstrIndex = unwrapOrError(ElfFile.getSection(0))->sh_link;
1569 if (ShstrIndex == SHN_UNDEF)
1570 Obj.HadShdrs = false;
1571 else
1572 Obj.SectionNames =
1573 Obj.sections().template getSectionOfType<StringTableSection>(
1574 ShstrIndex,
1575 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " +
1576 " is invalid",
1577 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " +
1578 " is not a string table");
1581 template <class ELFT> void ELFBuilder<ELFT>::build() {
1582 readSectionHeaders();
1583 findEhdrOffset();
1585 // The ELFFile whose ELF headers and program headers are copied into the
1586 // output file. Normally the same as ElfFile, but if we're extracting a
1587 // loadable partition it will point to the partition's headers.
1588 ELFFile<ELFT> HeadersFile = unwrapOrError(ELFFile<ELFT>::create(toStringRef(
1589 {ElfFile.base() + EhdrOffset, ElfFile.getBufSize() - EhdrOffset})));
1591 auto &Ehdr = *HeadersFile.getHeader();
1592 Obj.OSABI = Ehdr.e_ident[EI_OSABI];
1593 Obj.ABIVersion = Ehdr.e_ident[EI_ABIVERSION];
1594 Obj.Type = Ehdr.e_type;
1595 Obj.Machine = Ehdr.e_machine;
1596 Obj.Version = Ehdr.e_version;
1597 Obj.Entry = Ehdr.e_entry;
1598 Obj.Flags = Ehdr.e_flags;
1600 readSections();
1601 readProgramHeaders(HeadersFile);
1604 Writer::~Writer() {}
1606 Reader::~Reader() {}
1608 std::unique_ptr<Object> BinaryReader::create() const {
1609 return BinaryELFBuilder(MInfo.EMachine, MemBuf, NewSymbolVisibility).build();
1612 Expected<std::vector<IHexRecord>> IHexReader::parse() const {
1613 SmallVector<StringRef, 16> Lines;
1614 std::vector<IHexRecord> Records;
1615 bool HasSections = false;
1617 MemBuf->getBuffer().split(Lines, '\n');
1618 Records.reserve(Lines.size());
1619 for (size_t LineNo = 1; LineNo <= Lines.size(); ++LineNo) {
1620 StringRef Line = Lines[LineNo - 1].trim();
1621 if (Line.empty())
1622 continue;
1624 Expected<IHexRecord> R = IHexRecord::parse(Line);
1625 if (!R)
1626 return parseError(LineNo, R.takeError());
1627 if (R->Type == IHexRecord::EndOfFile)
1628 break;
1629 HasSections |= (R->Type == IHexRecord::Data);
1630 Records.push_back(*R);
1632 if (!HasSections)
1633 return parseError(-1U, "no sections");
1635 return std::move(Records);
1638 std::unique_ptr<Object> IHexReader::create() const {
1639 std::vector<IHexRecord> Records = unwrapOrError(parse());
1640 return IHexELFBuilder(Records).build();
1643 std::unique_ptr<Object> ELFReader::create() const {
1644 auto Obj = std::make_unique<Object>();
1645 if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Bin)) {
1646 ELFBuilder<ELF32LE> Builder(*O, *Obj, ExtractPartition);
1647 Builder.build();
1648 return Obj;
1649 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Bin)) {
1650 ELFBuilder<ELF64LE> Builder(*O, *Obj, ExtractPartition);
1651 Builder.build();
1652 return Obj;
1653 } else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Bin)) {
1654 ELFBuilder<ELF32BE> Builder(*O, *Obj, ExtractPartition);
1655 Builder.build();
1656 return Obj;
1657 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Bin)) {
1658 ELFBuilder<ELF64BE> Builder(*O, *Obj, ExtractPartition);
1659 Builder.build();
1660 return Obj;
1662 error("invalid file type");
1665 template <class ELFT> void ELFWriter<ELFT>::writeEhdr() {
1666 Elf_Ehdr &Ehdr = *reinterpret_cast<Elf_Ehdr *>(Buf.getBufferStart());
1667 std::fill(Ehdr.e_ident, Ehdr.e_ident + 16, 0);
1668 Ehdr.e_ident[EI_MAG0] = 0x7f;
1669 Ehdr.e_ident[EI_MAG1] = 'E';
1670 Ehdr.e_ident[EI_MAG2] = 'L';
1671 Ehdr.e_ident[EI_MAG3] = 'F';
1672 Ehdr.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32;
1673 Ehdr.e_ident[EI_DATA] =
1674 ELFT::TargetEndianness == support::big ? ELFDATA2MSB : ELFDATA2LSB;
1675 Ehdr.e_ident[EI_VERSION] = EV_CURRENT;
1676 Ehdr.e_ident[EI_OSABI] = Obj.OSABI;
1677 Ehdr.e_ident[EI_ABIVERSION] = Obj.ABIVersion;
1679 Ehdr.e_type = Obj.Type;
1680 Ehdr.e_machine = Obj.Machine;
1681 Ehdr.e_version = Obj.Version;
1682 Ehdr.e_entry = Obj.Entry;
1683 // We have to use the fully-qualified name llvm::size
1684 // since some compilers complain on ambiguous resolution.
1685 Ehdr.e_phnum = llvm::size(Obj.segments());
1686 Ehdr.e_phoff = (Ehdr.e_phnum != 0) ? Obj.ProgramHdrSegment.Offset : 0;
1687 Ehdr.e_phentsize = (Ehdr.e_phnum != 0) ? sizeof(Elf_Phdr) : 0;
1688 Ehdr.e_flags = Obj.Flags;
1689 Ehdr.e_ehsize = sizeof(Elf_Ehdr);
1690 if (WriteSectionHeaders && Obj.sections().size() != 0) {
1691 Ehdr.e_shentsize = sizeof(Elf_Shdr);
1692 Ehdr.e_shoff = Obj.SHOff;
1693 // """
1694 // If the number of sections is greater than or equal to
1695 // SHN_LORESERVE (0xff00), this member has the value zero and the actual
1696 // number of section header table entries is contained in the sh_size field
1697 // of the section header at index 0.
1698 // """
1699 auto Shnum = Obj.sections().size() + 1;
1700 if (Shnum >= SHN_LORESERVE)
1701 Ehdr.e_shnum = 0;
1702 else
1703 Ehdr.e_shnum = Shnum;
1704 // """
1705 // If the section name string table section index is greater than or equal
1706 // to SHN_LORESERVE (0xff00), this member has the value SHN_XINDEX (0xffff)
1707 // and the actual index of the section name string table section is
1708 // contained in the sh_link field of the section header at index 0.
1709 // """
1710 if (Obj.SectionNames->Index >= SHN_LORESERVE)
1711 Ehdr.e_shstrndx = SHN_XINDEX;
1712 else
1713 Ehdr.e_shstrndx = Obj.SectionNames->Index;
1714 } else {
1715 Ehdr.e_shentsize = 0;
1716 Ehdr.e_shoff = 0;
1717 Ehdr.e_shnum = 0;
1718 Ehdr.e_shstrndx = 0;
1722 template <class ELFT> void ELFWriter<ELFT>::writePhdrs() {
1723 for (auto &Seg : Obj.segments())
1724 writePhdr(Seg);
1727 template <class ELFT> void ELFWriter<ELFT>::writeShdrs() {
1728 // This reference serves to write the dummy section header at the begining
1729 // of the file. It is not used for anything else
1730 Elf_Shdr &Shdr =
1731 *reinterpret_cast<Elf_Shdr *>(Buf.getBufferStart() + Obj.SHOff);
1732 Shdr.sh_name = 0;
1733 Shdr.sh_type = SHT_NULL;
1734 Shdr.sh_flags = 0;
1735 Shdr.sh_addr = 0;
1736 Shdr.sh_offset = 0;
1737 // See writeEhdr for why we do this.
1738 uint64_t Shnum = Obj.sections().size() + 1;
1739 if (Shnum >= SHN_LORESERVE)
1740 Shdr.sh_size = Shnum;
1741 else
1742 Shdr.sh_size = 0;
1743 // See writeEhdr for why we do this.
1744 if (Obj.SectionNames != nullptr && Obj.SectionNames->Index >= SHN_LORESERVE)
1745 Shdr.sh_link = Obj.SectionNames->Index;
1746 else
1747 Shdr.sh_link = 0;
1748 Shdr.sh_info = 0;
1749 Shdr.sh_addralign = 0;
1750 Shdr.sh_entsize = 0;
1752 for (SectionBase &Sec : Obj.sections())
1753 writeShdr(Sec);
1756 template <class ELFT> void ELFWriter<ELFT>::writeSectionData() {
1757 for (SectionBase &Sec : Obj.sections())
1758 // Segments are responsible for writing their contents, so only write the
1759 // section data if the section is not in a segment. Note that this renders
1760 // sections in segments effectively immutable.
1761 if (Sec.ParentSegment == nullptr)
1762 Sec.accept(*SecWriter);
1765 template <class ELFT> void ELFWriter<ELFT>::writeSegmentData() {
1766 for (Segment &Seg : Obj.segments()) {
1767 uint8_t *B = Buf.getBufferStart() + Seg.Offset;
1768 assert(Seg.FileSize == Seg.getContents().size() &&
1769 "Segment size must match contents size");
1770 std::memcpy(B, Seg.getContents().data(), Seg.FileSize);
1773 // Iterate over removed sections and overwrite their old data with zeroes.
1774 for (auto &Sec : Obj.removedSections()) {
1775 Segment *Parent = Sec.ParentSegment;
1776 if (Parent == nullptr || Sec.Type == SHT_NOBITS || Sec.Size == 0)
1777 continue;
1778 uint64_t Offset =
1779 Sec.OriginalOffset - Parent->OriginalOffset + Parent->Offset;
1780 std::memset(Buf.getBufferStart() + Offset, 0, Sec.Size);
1784 template <class ELFT>
1785 ELFWriter<ELFT>::ELFWriter(Object &Obj, Buffer &Buf, bool WSH)
1786 : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs) {}
1788 Error Object::removeSections(bool AllowBrokenLinks,
1789 std::function<bool(const SectionBase &)> ToRemove) {
1791 auto Iter = std::stable_partition(
1792 std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) {
1793 if (ToRemove(*Sec))
1794 return false;
1795 if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) {
1796 if (auto ToRelSec = RelSec->getSection())
1797 return !ToRemove(*ToRelSec);
1799 return true;
1801 if (SymbolTable != nullptr && ToRemove(*SymbolTable))
1802 SymbolTable = nullptr;
1803 if (SectionNames != nullptr && ToRemove(*SectionNames))
1804 SectionNames = nullptr;
1805 if (SectionIndexTable != nullptr && ToRemove(*SectionIndexTable))
1806 SectionIndexTable = nullptr;
1807 // Now make sure there are no remaining references to the sections that will
1808 // be removed. Sometimes it is impossible to remove a reference so we emit
1809 // an error here instead.
1810 std::unordered_set<const SectionBase *> RemoveSections;
1811 RemoveSections.reserve(std::distance(Iter, std::end(Sections)));
1812 for (auto &RemoveSec : make_range(Iter, std::end(Sections))) {
1813 for (auto &Segment : Segments)
1814 Segment->removeSection(RemoveSec.get());
1815 RemoveSections.insert(RemoveSec.get());
1818 // For each section that remains alive, we want to remove the dead references.
1819 // This either might update the content of the section (e.g. remove symbols
1820 // from symbol table that belongs to removed section) or trigger an error if
1821 // a live section critically depends on a section being removed somehow
1822 // (e.g. the removed section is referenced by a relocation).
1823 for (auto &KeepSec : make_range(std::begin(Sections), Iter)) {
1824 if (Error E = KeepSec->removeSectionReferences(AllowBrokenLinks,
1825 [&RemoveSections](const SectionBase *Sec) {
1826 return RemoveSections.find(Sec) != RemoveSections.end();
1828 return E;
1831 // Transfer removed sections into the Object RemovedSections container for use
1832 // later.
1833 std::move(Iter, Sections.end(), std::back_inserter(RemovedSections));
1834 // Now finally get rid of them all together.
1835 Sections.erase(Iter, std::end(Sections));
1836 return Error::success();
1839 Error Object::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
1840 if (SymbolTable)
1841 for (const SecPtr &Sec : Sections)
1842 if (Error E = Sec->removeSymbols(ToRemove))
1843 return E;
1844 return Error::success();
1847 void Object::sortSections() {
1848 // Use stable_sort to maintain the original ordering as closely as possible.
1849 llvm::stable_sort(Sections, [](const SecPtr &A, const SecPtr &B) {
1850 // Put SHT_GROUP sections first, since group section headers must come
1851 // before the sections they contain. This also matches what GNU objcopy
1852 // does.
1853 if (A->Type != B->Type &&
1854 (A->Type == ELF::SHT_GROUP || B->Type == ELF::SHT_GROUP))
1855 return A->Type == ELF::SHT_GROUP;
1856 // For all other sections, sort by offset order.
1857 return A->OriginalOffset < B->OriginalOffset;
1861 // Orders segments such that if x = y->ParentSegment then y comes before x.
1862 static void orderSegments(std::vector<Segment *> &Segments) {
1863 llvm::stable_sort(Segments, compareSegmentsByOffset);
1866 // This function finds a consistent layout for a list of segments starting from
1867 // an Offset. It assumes that Segments have been sorted by orderSegments and
1868 // returns an Offset one past the end of the last segment.
1869 static uint64_t layoutSegments(std::vector<Segment *> &Segments,
1870 uint64_t Offset) {
1871 assert(std::is_sorted(std::begin(Segments), std::end(Segments),
1872 compareSegmentsByOffset));
1873 // The only way a segment should move is if a section was between two
1874 // segments and that section was removed. If that section isn't in a segment
1875 // then it's acceptable, but not ideal, to simply move it to after the
1876 // segments. So we can simply layout segments one after the other accounting
1877 // for alignment.
1878 for (Segment *Seg : Segments) {
1879 // We assume that segments have been ordered by OriginalOffset and Index
1880 // such that a parent segment will always come before a child segment in
1881 // OrderedSegments. This means that the Offset of the ParentSegment should
1882 // already be set and we can set our offset relative to it.
1883 if (Seg->ParentSegment != nullptr) {
1884 Segment *Parent = Seg->ParentSegment;
1885 Seg->Offset =
1886 Parent->Offset + Seg->OriginalOffset - Parent->OriginalOffset;
1887 } else {
1888 Seg->Offset =
1889 alignTo(Offset, std::max<uint64_t>(Seg->Align, 1), Seg->VAddr);
1891 Offset = std::max(Offset, Seg->Offset + Seg->FileSize);
1893 return Offset;
1896 // This function finds a consistent layout for a list of sections. It assumes
1897 // that the ->ParentSegment of each section has already been laid out. The
1898 // supplied starting Offset is used for the starting offset of any section that
1899 // does not have a ParentSegment. It returns either the offset given if all
1900 // sections had a ParentSegment or an offset one past the last section if there
1901 // was a section that didn't have a ParentSegment.
1902 template <class Range>
1903 static uint64_t layoutSections(Range Sections, uint64_t Offset) {
1904 // Now the offset of every segment has been set we can assign the offsets
1905 // of each section. For sections that are covered by a segment we should use
1906 // the segment's original offset and the section's original offset to compute
1907 // the offset from the start of the segment. Using the offset from the start
1908 // of the segment we can assign a new offset to the section. For sections not
1909 // covered by segments we can just bump Offset to the next valid location.
1910 uint32_t Index = 1;
1911 for (auto &Sec : Sections) {
1912 Sec.Index = Index++;
1913 if (Sec.ParentSegment != nullptr) {
1914 auto Segment = *Sec.ParentSegment;
1915 Sec.Offset =
1916 Segment.Offset + (Sec.OriginalOffset - Segment.OriginalOffset);
1917 } else {
1918 Offset = alignTo(Offset, Sec.Align == 0 ? 1 : Sec.Align);
1919 Sec.Offset = Offset;
1920 if (Sec.Type != SHT_NOBITS)
1921 Offset += Sec.Size;
1924 return Offset;
1927 template <class ELFT> void ELFWriter<ELFT>::initEhdrSegment() {
1928 Segment &ElfHdr = Obj.ElfHdrSegment;
1929 ElfHdr.Type = PT_PHDR;
1930 ElfHdr.Flags = 0;
1931 ElfHdr.VAddr = 0;
1932 ElfHdr.PAddr = 0;
1933 ElfHdr.FileSize = ElfHdr.MemSize = sizeof(Elf_Ehdr);
1934 ElfHdr.Align = 0;
1937 template <class ELFT> void ELFWriter<ELFT>::assignOffsets() {
1938 // We need a temporary list of segments that has a special order to it
1939 // so that we know that anytime ->ParentSegment is set that segment has
1940 // already had its offset properly set.
1941 std::vector<Segment *> OrderedSegments;
1942 for (Segment &Segment : Obj.segments())
1943 OrderedSegments.push_back(&Segment);
1944 OrderedSegments.push_back(&Obj.ElfHdrSegment);
1945 OrderedSegments.push_back(&Obj.ProgramHdrSegment);
1946 orderSegments(OrderedSegments);
1947 // Offset is used as the start offset of the first segment to be laid out.
1948 // Since the ELF Header (ElfHdrSegment) must be at the start of the file,
1949 // we start at offset 0.
1950 uint64_t Offset = 0;
1951 Offset = layoutSegments(OrderedSegments, Offset);
1952 Offset = layoutSections(Obj.sections(), Offset);
1953 // If we need to write the section header table out then we need to align the
1954 // Offset so that SHOffset is valid.
1955 if (WriteSectionHeaders)
1956 Offset = alignTo(Offset, sizeof(Elf_Addr));
1957 Obj.SHOff = Offset;
1960 template <class ELFT> size_t ELFWriter<ELFT>::totalSize() const {
1961 // We already have the section header offset so we can calculate the total
1962 // size by just adding up the size of each section header.
1963 if (!WriteSectionHeaders)
1964 return Obj.SHOff;
1965 size_t ShdrCount = Obj.sections().size() + 1; // Includes null shdr.
1966 return Obj.SHOff + ShdrCount * sizeof(Elf_Shdr);
1969 template <class ELFT> Error ELFWriter<ELFT>::write() {
1970 // Segment data must be written first, so that the ELF header and program
1971 // header tables can overwrite it, if covered by a segment.
1972 writeSegmentData();
1973 writeEhdr();
1974 writePhdrs();
1975 writeSectionData();
1976 if (WriteSectionHeaders)
1977 writeShdrs();
1978 return Buf.commit();
1981 static Error removeUnneededSections(Object &Obj) {
1982 // We can remove an empty symbol table from non-relocatable objects.
1983 // Relocatable objects typically have relocation sections whose
1984 // sh_link field points to .symtab, so we can't remove .symtab
1985 // even if it is empty.
1986 if (Obj.isRelocatable() || Obj.SymbolTable == nullptr ||
1987 !Obj.SymbolTable->empty())
1988 return Error::success();
1990 // .strtab can be used for section names. In such a case we shouldn't
1991 // remove it.
1992 auto *StrTab = Obj.SymbolTable->getStrTab() == Obj.SectionNames
1993 ? nullptr
1994 : Obj.SymbolTable->getStrTab();
1995 return Obj.removeSections(false, [&](const SectionBase &Sec) {
1996 return &Sec == Obj.SymbolTable || &Sec == StrTab;
2000 template <class ELFT> Error ELFWriter<ELFT>::finalize() {
2001 // It could happen that SectionNames has been removed and yet the user wants
2002 // a section header table output. We need to throw an error if a user tries
2003 // to do that.
2004 if (Obj.SectionNames == nullptr && WriteSectionHeaders)
2005 return createStringError(llvm::errc::invalid_argument,
2006 "cannot write section header table because "
2007 "section header string table was removed");
2009 if (Error E = removeUnneededSections(Obj))
2010 return E;
2011 Obj.sortSections();
2013 // We need to assign indexes before we perform layout because we need to know
2014 // if we need large indexes or not. We can assign indexes first and check as
2015 // we go to see if we will actully need large indexes.
2016 bool NeedsLargeIndexes = false;
2017 if (Obj.sections().size() >= SHN_LORESERVE) {
2018 SectionTableRef Sections = Obj.sections();
2019 NeedsLargeIndexes =
2020 std::any_of(Sections.begin() + SHN_LORESERVE, Sections.end(),
2021 [](const SectionBase &Sec) { return Sec.HasSymbol; });
2022 // TODO: handle case where only one section needs the large index table but
2023 // only needs it because the large index table hasn't been removed yet.
2026 if (NeedsLargeIndexes) {
2027 // This means we definitely need to have a section index table but if we
2028 // already have one then we should use it instead of making a new one.
2029 if (Obj.SymbolTable != nullptr && Obj.SectionIndexTable == nullptr) {
2030 // Addition of a section to the end does not invalidate the indexes of
2031 // other sections and assigns the correct index to the new section.
2032 auto &Shndx = Obj.addSection<SectionIndexSection>();
2033 Obj.SymbolTable->setShndxTable(&Shndx);
2034 Shndx.setSymTab(Obj.SymbolTable);
2036 } else {
2037 // Since we don't need SectionIndexTable we should remove it and all
2038 // references to it.
2039 if (Obj.SectionIndexTable != nullptr) {
2040 // We do not support sections referring to the section index table.
2041 if (Error E = Obj.removeSections(false /*AllowBrokenLinks*/,
2042 [this](const SectionBase &Sec) {
2043 return &Sec == Obj.SectionIndexTable;
2045 return E;
2049 // Make sure we add the names of all the sections. Importantly this must be
2050 // done after we decide to add or remove SectionIndexes.
2051 if (Obj.SectionNames != nullptr)
2052 for (const SectionBase &Sec : Obj.sections())
2053 Obj.SectionNames->addString(Sec.Name);
2055 initEhdrSegment();
2057 // Before we can prepare for layout the indexes need to be finalized.
2058 // Also, the output arch may not be the same as the input arch, so fix up
2059 // size-related fields before doing layout calculations.
2060 uint64_t Index = 0;
2061 auto SecSizer = std::make_unique<ELFSectionSizer<ELFT>>();
2062 for (SectionBase &Sec : Obj.sections()) {
2063 Sec.Index = Index++;
2064 Sec.accept(*SecSizer);
2067 // The symbol table does not update all other sections on update. For
2068 // instance, symbol names are not added as new symbols are added. This means
2069 // that some sections, like .strtab, don't yet have their final size.
2070 if (Obj.SymbolTable != nullptr)
2071 Obj.SymbolTable->prepareForLayout();
2073 // Now that all strings are added we want to finalize string table builders,
2074 // because that affects section sizes which in turn affects section offsets.
2075 for (SectionBase &Sec : Obj.sections())
2076 if (auto StrTab = dyn_cast<StringTableSection>(&Sec))
2077 StrTab->prepareForLayout();
2079 assignOffsets();
2081 // layoutSections could have modified section indexes, so we need
2082 // to fill the index table after assignOffsets.
2083 if (Obj.SymbolTable != nullptr)
2084 Obj.SymbolTable->fillShndxTable();
2086 // Finally now that all offsets and indexes have been set we can finalize any
2087 // remaining issues.
2088 uint64_t Offset = Obj.SHOff + sizeof(Elf_Shdr);
2089 for (SectionBase &Sec : Obj.sections()) {
2090 Sec.HeaderOffset = Offset;
2091 Offset += sizeof(Elf_Shdr);
2092 if (WriteSectionHeaders)
2093 Sec.NameIndex = Obj.SectionNames->findIndex(Sec.Name);
2094 Sec.finalize();
2097 if (Error E = Buf.allocate(totalSize()))
2098 return E;
2099 SecWriter = std::make_unique<ELFSectionWriter<ELFT>>(Buf);
2100 return Error::success();
2103 Error BinaryWriter::write() {
2104 for (const SectionBase &Sec : Obj.allocSections())
2105 Sec.accept(*SecWriter);
2106 return Buf.commit();
2109 Error BinaryWriter::finalize() {
2110 // We need a temporary list of segments that has a special order to it
2111 // so that we know that anytime ->ParentSegment is set that segment has
2112 // already had it's offset properly set. We only want to consider the segments
2113 // that will affect layout of allocated sections so we only add those.
2114 std::vector<Segment *> OrderedSegments;
2115 for (const SectionBase &Sec : Obj.allocSections())
2116 if (Sec.ParentSegment != nullptr)
2117 OrderedSegments.push_back(Sec.ParentSegment);
2119 // For binary output, we're going to use physical addresses instead of
2120 // virtual addresses, since a binary output is used for cases like ROM
2121 // loading and physical addresses are intended for ROM loading.
2122 // However, if no segment has a physical address, we'll fallback to using
2123 // virtual addresses for all.
2124 if (all_of(OrderedSegments,
2125 [](const Segment *Seg) { return Seg->PAddr == 0; }))
2126 for (Segment *Seg : OrderedSegments)
2127 Seg->PAddr = Seg->VAddr;
2129 llvm::stable_sort(OrderedSegments, compareSegmentsByPAddr);
2131 // Because we add a ParentSegment for each section we might have duplicate
2132 // segments in OrderedSegments. If there were duplicates then layoutSegments
2133 // would do very strange things.
2134 auto End =
2135 std::unique(std::begin(OrderedSegments), std::end(OrderedSegments));
2136 OrderedSegments.erase(End, std::end(OrderedSegments));
2138 uint64_t Offset = 0;
2140 // Modify the first segment so that there is no gap at the start. This allows
2141 // our layout algorithm to proceed as expected while not writing out the gap
2142 // at the start.
2143 if (!OrderedSegments.empty()) {
2144 Segment *Seg = OrderedSegments[0];
2145 const SectionBase *Sec = Seg->firstSection();
2146 auto Diff = Sec->OriginalOffset - Seg->OriginalOffset;
2147 Seg->OriginalOffset += Diff;
2148 // The size needs to be shrunk as well.
2149 Seg->FileSize -= Diff;
2150 // The PAddr needs to be increased to remove the gap before the first
2151 // section.
2152 Seg->PAddr += Diff;
2153 uint64_t LowestPAddr = Seg->PAddr;
2154 for (Segment *Segment : OrderedSegments) {
2155 Segment->Offset = Segment->PAddr - LowestPAddr;
2156 Offset = std::max(Offset, Segment->Offset + Segment->FileSize);
2160 layoutSections(Obj.allocSections(), Offset);
2162 // Now that every section has been laid out we just need to compute the total
2163 // file size. This might not be the same as the offset returned by
2164 // layoutSections, because we want to truncate the last segment to the end of
2165 // its last section, to match GNU objcopy's behaviour.
2166 TotalSize = 0;
2167 for (const SectionBase &Sec : Obj.allocSections())
2168 if (Sec.Type != SHT_NOBITS)
2169 TotalSize = std::max(TotalSize, Sec.Offset + Sec.Size);
2171 if (Error E = Buf.allocate(TotalSize))
2172 return E;
2173 SecWriter = std::make_unique<BinarySectionWriter>(Buf);
2174 return Error::success();
2177 bool IHexWriter::SectionCompare::operator()(const SectionBase *Lhs,
2178 const SectionBase *Rhs) const {
2179 return (sectionPhysicalAddr(Lhs) & 0xFFFFFFFFU) <
2180 (sectionPhysicalAddr(Rhs) & 0xFFFFFFFFU);
2183 uint64_t IHexWriter::writeEntryPointRecord(uint8_t *Buf) {
2184 IHexLineData HexData;
2185 uint8_t Data[4] = {};
2186 // We don't write entry point record if entry is zero.
2187 if (Obj.Entry == 0)
2188 return 0;
2190 if (Obj.Entry <= 0xFFFFFU) {
2191 Data[0] = ((Obj.Entry & 0xF0000U) >> 12) & 0xFF;
2192 support::endian::write(&Data[2], static_cast<uint16_t>(Obj.Entry),
2193 support::big);
2194 HexData = IHexRecord::getLine(IHexRecord::StartAddr80x86, 0, Data);
2195 } else {
2196 support::endian::write(Data, static_cast<uint32_t>(Obj.Entry),
2197 support::big);
2198 HexData = IHexRecord::getLine(IHexRecord::StartAddr, 0, Data);
2200 memcpy(Buf, HexData.data(), HexData.size());
2201 return HexData.size();
2204 uint64_t IHexWriter::writeEndOfFileRecord(uint8_t *Buf) {
2205 IHexLineData HexData = IHexRecord::getLine(IHexRecord::EndOfFile, 0, {});
2206 memcpy(Buf, HexData.data(), HexData.size());
2207 return HexData.size();
2210 Error IHexWriter::write() {
2211 IHexSectionWriter Writer(Buf);
2212 // Write sections.
2213 for (const SectionBase *Sec : Sections)
2214 Sec->accept(Writer);
2216 uint64_t Offset = Writer.getBufferOffset();
2217 // Write entry point address.
2218 Offset += writeEntryPointRecord(Buf.getBufferStart() + Offset);
2219 // Write EOF.
2220 Offset += writeEndOfFileRecord(Buf.getBufferStart() + Offset);
2221 assert(Offset == TotalSize);
2222 return Buf.commit();
2225 Error IHexWriter::checkSection(const SectionBase &Sec) {
2226 uint64_t Addr = sectionPhysicalAddr(&Sec);
2227 if (addressOverflows32bit(Addr) || addressOverflows32bit(Addr + Sec.Size - 1))
2228 return createStringError(
2229 errc::invalid_argument,
2230 "Section '%s' address range [0x%llx, 0x%llx] is not 32 bit", Sec.Name.c_str(),
2231 Addr, Addr + Sec.Size - 1);
2232 return Error::success();
2235 Error IHexWriter::finalize() {
2236 bool UseSegments = false;
2237 auto ShouldWrite = [](const SectionBase &Sec) {
2238 return (Sec.Flags & ELF::SHF_ALLOC) && (Sec.Type != ELF::SHT_NOBITS);
2240 auto IsInPtLoad = [](const SectionBase &Sec) {
2241 return Sec.ParentSegment && Sec.ParentSegment->Type == ELF::PT_LOAD;
2244 // We can't write 64-bit addresses.
2245 if (addressOverflows32bit(Obj.Entry))
2246 return createStringError(errc::invalid_argument,
2247 "Entry point address 0x%llx overflows 32 bits.",
2248 Obj.Entry);
2250 // If any section we're to write has segment then we
2251 // switch to using physical addresses. Otherwise we
2252 // use section virtual address.
2253 for (const SectionBase &Sec : Obj.sections())
2254 if (ShouldWrite(Sec) && IsInPtLoad(Sec)) {
2255 UseSegments = true;
2256 break;
2259 for (const SectionBase &Sec : Obj.sections())
2260 if (ShouldWrite(Sec) && (!UseSegments || IsInPtLoad(Sec))) {
2261 if (Error E = checkSection(Sec))
2262 return E;
2263 Sections.insert(&Sec);
2266 IHexSectionWriterBase LengthCalc(Buf);
2267 for (const SectionBase *Sec : Sections)
2268 Sec->accept(LengthCalc);
2270 // We need space to write section records + StartAddress record
2271 // (if start adress is not zero) + EndOfFile record.
2272 TotalSize = LengthCalc.getBufferOffset() +
2273 (Obj.Entry ? IHexRecord::getLineLength(4) : 0) +
2274 IHexRecord::getLineLength(0);
2275 if (Error E = Buf.allocate(TotalSize))
2276 return E;
2277 return Error::success();
2280 template class ELFBuilder<ELF64LE>;
2281 template class ELFBuilder<ELF64BE>;
2282 template class ELFBuilder<ELF32LE>;
2283 template class ELFBuilder<ELF32BE>;
2285 template class ELFWriter<ELF64LE>;
2286 template class ELFWriter<ELF64BE>;
2287 template class ELFWriter<ELF32LE>;
2288 template class ELFWriter<ELF32BE>;
2290 } // end namespace elf
2291 } // end namespace objcopy
2292 } // end namespace llvm