[RISCV] Fix mgather -> riscv.masked.strided.load combine not extending indices (...
[llvm-project.git] / llvm / lib / ObjCopy / ELF / ELFObject.cpp
blobc8b66d6fcb5ebfae71881182827759b5f3516388
1 //===- ELFObject.cpp ------------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
9 #include "ELFObject.h"
10 #include "llvm/ADT/ArrayRef.h"
11 #include "llvm/ADT/STLExtras.h"
12 #include "llvm/ADT/StringRef.h"
13 #include "llvm/ADT/Twine.h"
14 #include "llvm/ADT/iterator_range.h"
15 #include "llvm/BinaryFormat/ELF.h"
16 #include "llvm/MC/MCTargetOptions.h"
17 #include "llvm/Object/ELF.h"
18 #include "llvm/Object/ELFObjectFile.h"
19 #include "llvm/Support/Compression.h"
20 #include "llvm/Support/Endian.h"
21 #include "llvm/Support/ErrorHandling.h"
22 #include "llvm/Support/FileOutputBuffer.h"
23 #include "llvm/Support/Path.h"
24 #include <algorithm>
25 #include <cstddef>
26 #include <cstdint>
27 #include <iterator>
28 #include <unordered_set>
29 #include <utility>
30 #include <vector>
32 using namespace llvm;
33 using namespace llvm::ELF;
34 using namespace llvm::objcopy::elf;
35 using namespace llvm::object;
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;
51 Error SectionBase::removeSectionReferences(
52 bool, function_ref<bool(const SectionBase *)>) {
53 return Error::success();
56 Error SectionBase::removeSymbols(function_ref<bool(const Symbol &)>) {
57 return Error::success();
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() {}
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;
83 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(Section &) {
84 return Error::success();
87 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(OwnedDataSection &) {
88 return Error::success();
91 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(StringTableSection &) {
92 return Error::success();
95 template <class ELFT>
96 Error ELFSectionSizer<ELFT>::visit(DynamicRelocationSection &) {
97 return Error::success();
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();
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();
118 template <class ELFT>
119 Error ELFSectionSizer<ELFT>::visit(GnuDebugLinkSection &) {
120 return Error::success();
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();
128 template <class ELFT>
129 Error ELFSectionSizer<ELFT>::visit(SectionIndexSection &) {
130 return Error::success();
133 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(CompressedSection &) {
134 return Error::success();
137 template <class ELFT>
138 Error ELFSectionSizer<ELFT>::visit(DecompressedSection &) {
139 return Error::success();
142 Error BinarySectionWriter::visit(const SectionIndexSection &Sec) {
143 return createStringError(errc::operation_not_permitted,
144 "cannot write symbol section index table '" +
145 Sec.Name + "' ");
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");
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");
160 Error BinarySectionWriter::visit(const GnuDebugLinkSection &Sec) {
161 return createStringError(errc::operation_not_permitted,
162 "cannot write '" + Sec.Name + "' out to binary");
165 Error BinarySectionWriter::visit(const GroupSection &Sec) {
166 return createStringError(errc::operation_not_permitted,
167 "cannot write '" + Sec.Name + "' out to binary");
170 Error SectionWriter::visit(const Section &Sec) {
171 if (Sec.Type != SHT_NOBITS)
172 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset);
174 return Error::success();
177 static bool addressOverflows32bit(uint64_t Addr) {
178 // Sign extended 32 bit addresses (e.g 0xFFFFFFFF80000000) are ok
179 return Addr > UINT32_MAX && Addr + 0x80000000 > UINT32_MAX;
182 template <class T> static T checkedGetHex(StringRef S) {
183 T Value;
184 bool Fail = S.getAsInteger(16, Value);
185 assert(!Fail);
186 (void)Fail;
187 return Value;
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');
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;
202 assert(X == 0);
203 return It + Len;
206 uint8_t IHexRecord::getChecksum(StringRef S) {
207 assert((S.size() & 1) == 0);
208 uint8_t Checksum = 0;
209 while (!S.empty()) {
210 Checksum += checkedGetHex<uint8_t>(S.take_front(2));
211 S = S.drop_front(2);
213 return -Checksum;
216 IHexLineData IHexRecord::getLine(uint8_t Type, uint16_t Addr,
217 ArrayRef<uint8_t> Data) {
218 IHexLineData Line(getLineLength(Data.size()));
219 assert(Line.size());
220 auto Iter = Line.begin();
221 *Iter++ = ':';
222 Iter = toHexStr(Data.size(), Iter, 2);
223 Iter = toHexStr(Addr, Iter, 4);
224 Iter = toHexStr(Type, Iter, 2);
225 for (uint8_t X : Data)
226 Iter = toHexStr(X, Iter, 2);
227 StringRef S(Line.data() + 1, std::distance(Line.begin() + 1, Iter));
228 Iter = toHexStr(getChecksum(S), Iter, 2);
229 *Iter++ = '\r';
230 *Iter++ = '\n';
231 assert(Iter == Line.end());
232 return Line;
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));
279 return Error::success();
282 // Checks that IHEX line contains valid characters.
283 // This allows converting hexadecimal data to integers
284 // without extra verification.
285 static Error checkChars(StringRef Line) {
286 assert(!Line.empty());
287 if (Line[0] != ':')
288 return createStringError(errc::invalid_argument,
289 "missing ':' in the beginning of line.");
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();
298 Expected<IHexRecord> IHexRecord::parse(StringRef Line) {
299 assert(!Line.empty());
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());
306 if (Error E = checkChars(Line))
307 return std::move(E);
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));
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);
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;
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;
335 void IHexSectionWriterBase::writeSection(const SectionBase *Sec,
336 ArrayRef<uint8_t> Data) {
337 assert(Data.size() == Sec->Size);
338 const uint32_t ChunkSize = 16;
339 uint32_t Addr = sectionPhysicalAddr(Sec) & 0xFFFFFFFFU;
340 while (!Data.empty()) {
341 uint64_t DataSize = std::min<uint64_t>(Data.size(), ChunkSize);
342 if (Addr > SegmentAddr + BaseAddr + 0xFFFFU) {
343 if (Addr > 0xFFFFFU) {
344 // Write extended address record, zeroing segment address
345 // if needed.
346 if (SegmentAddr != 0)
347 SegmentAddr = writeSegmentAddr(0U);
348 BaseAddr = writeBaseAddr(Addr);
349 } else {
350 // We can still remain 16-bit
351 SegmentAddr = writeSegmentAddr(Addr);
354 uint64_t SegOffset = Addr - BaseAddr - SegmentAddr;
355 assert(SegOffset <= 0xFFFFU);
356 DataSize = std::min(DataSize, 0x10000U - SegOffset);
357 writeData(0, SegOffset, Data.take_front(DataSize));
358 Addr += DataSize;
359 Data = Data.drop_front(DataSize);
363 uint64_t IHexSectionWriterBase::writeSegmentAddr(uint64_t Addr) {
364 assert(Addr <= 0xFFFFFU);
365 uint8_t Data[] = {static_cast<uint8_t>((Addr & 0xF0000U) >> 12), 0};
366 writeData(2, 0, Data);
367 return Addr & 0xF0000U;
370 uint64_t IHexSectionWriterBase::writeBaseAddr(uint64_t Addr) {
371 assert(Addr <= 0xFFFFFFFFU);
372 uint64_t Base = Addr & 0xFFFF0000U;
373 uint8_t Data[] = {static_cast<uint8_t>(Base >> 24),
374 static_cast<uint8_t>((Base >> 16) & 0xFF)};
375 writeData(4, 0, Data);
376 return Base;
379 void IHexSectionWriterBase::writeData(uint8_t, uint16_t,
380 ArrayRef<uint8_t> Data) {
381 Offset += IHexRecord::getLineLength(Data.size());
384 Error IHexSectionWriterBase::visit(const Section &Sec) {
385 writeSection(&Sec, Sec.Contents);
386 return Error::success();
389 Error IHexSectionWriterBase::visit(const OwnedDataSection &Sec) {
390 writeSection(&Sec, Sec.Data);
391 return Error::success();
394 Error IHexSectionWriterBase::visit(const StringTableSection &Sec) {
395 // Check that sizer has already done its work
396 assert(Sec.Size == Sec.StrTabBuilder.getSize());
397 // We are free to pass an invalid pointer to writeSection as long
398 // as we don't actually write any data. The real writer class has
399 // to override this method .
400 writeSection(&Sec, {nullptr, static_cast<size_t>(Sec.Size)});
401 return Error::success();
404 Error IHexSectionWriterBase::visit(const DynamicRelocationSection &Sec) {
405 writeSection(&Sec, Sec.Contents);
406 return Error::success();
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();
416 Error IHexSectionWriter::visit(const StringTableSection &Sec) {
417 assert(Sec.Size == Sec.StrTabBuilder.getSize());
418 std::vector<uint8_t> Data(Sec.Size);
419 Sec.StrTabBuilder.write(Data.data());
420 writeSection(&Sec, Data);
421 return Error::success();
424 Error Section::accept(SectionVisitor &Visitor) const {
425 return Visitor.visit(*this);
428 Error Section::accept(MutableSectionVisitor &Visitor) {
429 return Visitor.visit(*this);
432 void Section::restoreSymTabLink(SymbolTableSection &SymTab) {
433 if (HasSymTabLink) {
434 assert(LinkSection == nullptr);
435 LinkSection = &SymTab;
439 Error SectionWriter::visit(const OwnedDataSection &Sec) {
440 llvm::copy(Sec.Data, Out.getBufferStart() + Sec.Offset);
441 return Error::success();
444 template <class ELFT>
445 Error ELFSectionWriter<ELFT>::visit(const DecompressedSection &Sec) {
446 ArrayRef<uint8_t> Compressed =
447 Sec.OriginalData.slice(sizeof(Elf_Chdr_Impl<ELFT>));
448 SmallVector<uint8_t, 128> Decompressed;
449 DebugCompressionType Type;
450 switch (Sec.ChType) {
451 case ELFCOMPRESS_ZLIB:
452 Type = DebugCompressionType::Zlib;
453 break;
454 case ELFCOMPRESS_ZSTD:
455 Type = DebugCompressionType::Zstd;
456 break;
457 default:
458 return createStringError(errc::invalid_argument,
459 "--decompress-debug-sections: ch_type (" +
460 Twine(Sec.ChType) + ") of section '" +
461 Sec.Name + "' is unsupported");
463 if (auto *Reason =
464 compression::getReasonIfUnsupported(compression::formatFor(Type)))
465 return createStringError(errc::invalid_argument,
466 "failed to decompress section '" + Sec.Name +
467 "': " + Reason);
468 if (Error E = compression::decompress(Type, Compressed, Decompressed,
469 static_cast<size_t>(Sec.Size)))
470 return createStringError(errc::invalid_argument,
471 "failed to decompress section '" + Sec.Name +
472 "': " + toString(std::move(E)));
474 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
475 std::copy(Decompressed.begin(), Decompressed.end(), Buf);
477 return Error::success();
480 Error BinarySectionWriter::visit(const DecompressedSection &Sec) {
481 return createStringError(errc::operation_not_permitted,
482 "cannot write compressed section '" + Sec.Name +
483 "' ");
486 Error DecompressedSection::accept(SectionVisitor &Visitor) const {
487 return Visitor.visit(*this);
490 Error DecompressedSection::accept(MutableSectionVisitor &Visitor) {
491 return Visitor.visit(*this);
494 Error OwnedDataSection::accept(SectionVisitor &Visitor) const {
495 return Visitor.visit(*this);
498 Error OwnedDataSection::accept(MutableSectionVisitor &Visitor) {
499 return Visitor.visit(*this);
502 void OwnedDataSection::appendHexData(StringRef HexData) {
503 assert((HexData.size() & 1) == 0);
504 while (!HexData.empty()) {
505 Data.push_back(checkedGetHex<uint8_t>(HexData.take_front(2)));
506 HexData = HexData.drop_front(2);
508 Size = Data.size();
511 Error BinarySectionWriter::visit(const CompressedSection &Sec) {
512 return createStringError(errc::operation_not_permitted,
513 "cannot write compressed section '" + Sec.Name +
514 "' ");
517 template <class ELFT>
518 Error ELFSectionWriter<ELFT>::visit(const CompressedSection &Sec) {
519 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
520 Elf_Chdr_Impl<ELFT> Chdr = {};
521 switch (Sec.CompressionType) {
522 case DebugCompressionType::None:
523 std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(), Buf);
524 return Error::success();
525 case DebugCompressionType::Zlib:
526 Chdr.ch_type = ELF::ELFCOMPRESS_ZLIB;
527 break;
528 case DebugCompressionType::Zstd:
529 Chdr.ch_type = ELF::ELFCOMPRESS_ZSTD;
530 break;
532 Chdr.ch_size = Sec.DecompressedSize;
533 Chdr.ch_addralign = Sec.DecompressedAlign;
534 memcpy(Buf, &Chdr, sizeof(Chdr));
535 Buf += sizeof(Chdr);
537 std::copy(Sec.CompressedData.begin(), Sec.CompressedData.end(), Buf);
538 return Error::success();
541 CompressedSection::CompressedSection(const SectionBase &Sec,
542 DebugCompressionType CompressionType,
543 bool Is64Bits)
544 : SectionBase(Sec), CompressionType(CompressionType),
545 DecompressedSize(Sec.OriginalData.size()), DecompressedAlign(Sec.Align) {
546 compression::compress(compression::Params(CompressionType), OriginalData,
547 CompressedData);
549 Flags |= ELF::SHF_COMPRESSED;
550 size_t ChdrSize = Is64Bits ? sizeof(object::Elf_Chdr_Impl<object::ELF64LE>)
551 : sizeof(object::Elf_Chdr_Impl<object::ELF32LE>);
552 Size = ChdrSize + CompressedData.size();
553 Align = 8;
556 CompressedSection::CompressedSection(ArrayRef<uint8_t> CompressedData,
557 uint32_t ChType, uint64_t DecompressedSize,
558 uint64_t DecompressedAlign)
559 : ChType(ChType), CompressionType(DebugCompressionType::None),
560 DecompressedSize(DecompressedSize), DecompressedAlign(DecompressedAlign) {
561 OriginalData = CompressedData;
564 Error CompressedSection::accept(SectionVisitor &Visitor) const {
565 return Visitor.visit(*this);
568 Error CompressedSection::accept(MutableSectionVisitor &Visitor) {
569 return Visitor.visit(*this);
572 void StringTableSection::addString(StringRef Name) { StrTabBuilder.add(Name); }
574 uint32_t StringTableSection::findIndex(StringRef Name) const {
575 return StrTabBuilder.getOffset(Name);
578 void StringTableSection::prepareForLayout() {
579 StrTabBuilder.finalize();
580 Size = StrTabBuilder.getSize();
583 Error SectionWriter::visit(const StringTableSection &Sec) {
584 Sec.StrTabBuilder.write(reinterpret_cast<uint8_t *>(Out.getBufferStart()) +
585 Sec.Offset);
586 return Error::success();
589 Error StringTableSection::accept(SectionVisitor &Visitor) const {
590 return Visitor.visit(*this);
593 Error StringTableSection::accept(MutableSectionVisitor &Visitor) {
594 return Visitor.visit(*this);
597 template <class ELFT>
598 Error ELFSectionWriter<ELFT>::visit(const SectionIndexSection &Sec) {
599 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
600 llvm::copy(Sec.Indexes, reinterpret_cast<Elf_Word *>(Buf));
601 return Error::success();
604 Error SectionIndexSection::initialize(SectionTableRef SecTable) {
605 Size = 0;
606 Expected<SymbolTableSection *> Sec =
607 SecTable.getSectionOfType<SymbolTableSection>(
608 Link,
609 "Link field value " + Twine(Link) + " in section " + Name +
610 " is invalid",
611 "Link field value " + Twine(Link) + " in section " + Name +
612 " is not a symbol table");
613 if (!Sec)
614 return Sec.takeError();
616 setSymTab(*Sec);
617 Symbols->setShndxTable(this);
618 return Error::success();
621 void SectionIndexSection::finalize() { Link = Symbols->Index; }
623 Error SectionIndexSection::accept(SectionVisitor &Visitor) const {
624 return Visitor.visit(*this);
627 Error SectionIndexSection::accept(MutableSectionVisitor &Visitor) {
628 return Visitor.visit(*this);
631 static bool isValidReservedSectionIndex(uint16_t Index, uint16_t Machine) {
632 switch (Index) {
633 case SHN_ABS:
634 case SHN_COMMON:
635 return true;
638 if (Machine == EM_AMDGPU) {
639 return Index == SHN_AMDGPU_LDS;
642 if (Machine == EM_MIPS) {
643 switch (Index) {
644 case SHN_MIPS_ACOMMON:
645 case SHN_MIPS_SCOMMON:
646 case SHN_MIPS_SUNDEFINED:
647 return true;
651 if (Machine == EM_HEXAGON) {
652 switch (Index) {
653 case SHN_HEXAGON_SCOMMON:
654 case SHN_HEXAGON_SCOMMON_1:
655 case SHN_HEXAGON_SCOMMON_2:
656 case SHN_HEXAGON_SCOMMON_4:
657 case SHN_HEXAGON_SCOMMON_8:
658 return true;
661 return false;
664 // Large indexes force us to clarify exactly what this function should do. This
665 // function should return the value that will appear in st_shndx when written
666 // out.
667 uint16_t Symbol::getShndx() const {
668 if (DefinedIn != nullptr) {
669 if (DefinedIn->Index >= SHN_LORESERVE)
670 return SHN_XINDEX;
671 return DefinedIn->Index;
674 if (ShndxType == SYMBOL_SIMPLE_INDEX) {
675 // This means that we don't have a defined section but we do need to
676 // output a legitimate section index.
677 return SHN_UNDEF;
680 assert(ShndxType == SYMBOL_ABS || ShndxType == SYMBOL_COMMON ||
681 (ShndxType >= SYMBOL_LOPROC && ShndxType <= SYMBOL_HIPROC) ||
682 (ShndxType >= SYMBOL_LOOS && ShndxType <= SYMBOL_HIOS));
683 return static_cast<uint16_t>(ShndxType);
686 bool Symbol::isCommon() const { return getShndx() == SHN_COMMON; }
688 void SymbolTableSection::assignIndices() {
689 uint32_t Index = 0;
690 for (auto &Sym : Symbols) {
691 if (Sym->Index != Index)
692 IndicesChanged = true;
693 Sym->Index = Index++;
697 void SymbolTableSection::addSymbol(Twine Name, uint8_t Bind, uint8_t Type,
698 SectionBase *DefinedIn, uint64_t Value,
699 uint8_t Visibility, uint16_t Shndx,
700 uint64_t SymbolSize) {
701 Symbol Sym;
702 Sym.Name = Name.str();
703 Sym.Binding = Bind;
704 Sym.Type = Type;
705 Sym.DefinedIn = DefinedIn;
706 if (DefinedIn != nullptr)
707 DefinedIn->HasSymbol = true;
708 if (DefinedIn == nullptr) {
709 if (Shndx >= SHN_LORESERVE)
710 Sym.ShndxType = static_cast<SymbolShndxType>(Shndx);
711 else
712 Sym.ShndxType = SYMBOL_SIMPLE_INDEX;
714 Sym.Value = Value;
715 Sym.Visibility = Visibility;
716 Sym.Size = SymbolSize;
717 Sym.Index = Symbols.size();
718 Symbols.emplace_back(std::make_unique<Symbol>(Sym));
719 Size += this->EntrySize;
722 Error SymbolTableSection::removeSectionReferences(
723 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
724 if (ToRemove(SectionIndexTable))
725 SectionIndexTable = nullptr;
726 if (ToRemove(SymbolNames)) {
727 if (!AllowBrokenLinks)
728 return createStringError(
729 llvm::errc::invalid_argument,
730 "string table '%s' cannot be removed because it is "
731 "referenced by the symbol table '%s'",
732 SymbolNames->Name.data(), this->Name.data());
733 SymbolNames = nullptr;
735 return removeSymbols(
736 [ToRemove](const Symbol &Sym) { return ToRemove(Sym.DefinedIn); });
739 void SymbolTableSection::updateSymbols(function_ref<void(Symbol &)> Callable) {
740 for (SymPtr &Sym : llvm::drop_begin(Symbols))
741 Callable(*Sym);
742 std::stable_partition(
743 std::begin(Symbols), std::end(Symbols),
744 [](const SymPtr &Sym) { return Sym->Binding == STB_LOCAL; });
745 assignIndices();
748 Error SymbolTableSection::removeSymbols(
749 function_ref<bool(const Symbol &)> ToRemove) {
750 Symbols.erase(
751 std::remove_if(std::begin(Symbols) + 1, std::end(Symbols),
752 [ToRemove](const SymPtr &Sym) { return ToRemove(*Sym); }),
753 std::end(Symbols));
754 auto PrevSize = Size;
755 Size = Symbols.size() * EntrySize;
756 if (Size < PrevSize)
757 IndicesChanged = true;
758 assignIndices();
759 return Error::success();
762 void SymbolTableSection::replaceSectionReferences(
763 const DenseMap<SectionBase *, SectionBase *> &FromTo) {
764 for (std::unique_ptr<Symbol> &Sym : Symbols)
765 if (SectionBase *To = FromTo.lookup(Sym->DefinedIn))
766 Sym->DefinedIn = To;
769 Error SymbolTableSection::initialize(SectionTableRef SecTable) {
770 Size = 0;
771 Expected<StringTableSection *> Sec =
772 SecTable.getSectionOfType<StringTableSection>(
773 Link,
774 "Symbol table has link index of " + Twine(Link) +
775 " which is not a valid index",
776 "Symbol table has link index of " + Twine(Link) +
777 " which is not a string table");
778 if (!Sec)
779 return Sec.takeError();
781 setStrTab(*Sec);
782 return Error::success();
785 void SymbolTableSection::finalize() {
786 uint32_t MaxLocalIndex = 0;
787 for (std::unique_ptr<Symbol> &Sym : Symbols) {
788 Sym->NameIndex =
789 SymbolNames == nullptr ? 0 : SymbolNames->findIndex(Sym->Name);
790 if (Sym->Binding == STB_LOCAL)
791 MaxLocalIndex = std::max(MaxLocalIndex, Sym->Index);
793 // Now we need to set the Link and Info fields.
794 Link = SymbolNames == nullptr ? 0 : SymbolNames->Index;
795 Info = MaxLocalIndex + 1;
798 void SymbolTableSection::prepareForLayout() {
799 // Reserve proper amount of space in section index table, so we can
800 // layout sections correctly. We will fill the table with correct
801 // indexes later in fillShdnxTable.
802 if (SectionIndexTable)
803 SectionIndexTable->reserve(Symbols.size());
805 // Add all of our strings to SymbolNames so that SymbolNames has the right
806 // size before layout is decided.
807 // If the symbol names section has been removed, don't try to add strings to
808 // the table.
809 if (SymbolNames != nullptr)
810 for (std::unique_ptr<Symbol> &Sym : Symbols)
811 SymbolNames->addString(Sym->Name);
814 void SymbolTableSection::fillShndxTable() {
815 if (SectionIndexTable == nullptr)
816 return;
817 // Fill section index table with real section indexes. This function must
818 // be called after assignOffsets.
819 for (const std::unique_ptr<Symbol> &Sym : Symbols) {
820 if (Sym->DefinedIn != nullptr && Sym->DefinedIn->Index >= SHN_LORESERVE)
821 SectionIndexTable->addIndex(Sym->DefinedIn->Index);
822 else
823 SectionIndexTable->addIndex(SHN_UNDEF);
827 Expected<const Symbol *>
828 SymbolTableSection::getSymbolByIndex(uint32_t Index) const {
829 if (Symbols.size() <= Index)
830 return createStringError(errc::invalid_argument,
831 "invalid symbol index: " + Twine(Index));
832 return Symbols[Index].get();
835 Expected<Symbol *> SymbolTableSection::getSymbolByIndex(uint32_t Index) {
836 Expected<const Symbol *> Sym =
837 static_cast<const SymbolTableSection *>(this)->getSymbolByIndex(Index);
838 if (!Sym)
839 return Sym.takeError();
841 return const_cast<Symbol *>(*Sym);
844 template <class ELFT>
845 Error ELFSectionWriter<ELFT>::visit(const SymbolTableSection &Sec) {
846 Elf_Sym *Sym = reinterpret_cast<Elf_Sym *>(Out.getBufferStart() + Sec.Offset);
847 // Loop though symbols setting each entry of the symbol table.
848 for (const std::unique_ptr<Symbol> &Symbol : Sec.Symbols) {
849 Sym->st_name = Symbol->NameIndex;
850 Sym->st_value = Symbol->Value;
851 Sym->st_size = Symbol->Size;
852 Sym->st_other = Symbol->Visibility;
853 Sym->setBinding(Symbol->Binding);
854 Sym->setType(Symbol->Type);
855 Sym->st_shndx = Symbol->getShndx();
856 ++Sym;
858 return Error::success();
861 Error SymbolTableSection::accept(SectionVisitor &Visitor) const {
862 return Visitor.visit(*this);
865 Error SymbolTableSection::accept(MutableSectionVisitor &Visitor) {
866 return Visitor.visit(*this);
869 StringRef RelocationSectionBase::getNamePrefix() const {
870 switch (Type) {
871 case SHT_REL:
872 return ".rel";
873 case SHT_RELA:
874 return ".rela";
875 default:
876 llvm_unreachable("not a relocation section");
880 Error RelocationSection::removeSectionReferences(
881 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
882 if (ToRemove(Symbols)) {
883 if (!AllowBrokenLinks)
884 return createStringError(
885 llvm::errc::invalid_argument,
886 "symbol table '%s' cannot be removed because it is "
887 "referenced by the relocation section '%s'",
888 Symbols->Name.data(), this->Name.data());
889 Symbols = nullptr;
892 for (const Relocation &R : Relocations) {
893 if (!R.RelocSymbol || !R.RelocSymbol->DefinedIn ||
894 !ToRemove(R.RelocSymbol->DefinedIn))
895 continue;
896 return createStringError(llvm::errc::invalid_argument,
897 "section '%s' cannot be removed: (%s+0x%" PRIx64
898 ") has relocation against symbol '%s'",
899 R.RelocSymbol->DefinedIn->Name.data(),
900 SecToApplyRel->Name.data(), R.Offset,
901 R.RelocSymbol->Name.c_str());
904 return Error::success();
907 template <class SymTabType>
908 Error RelocSectionWithSymtabBase<SymTabType>::initialize(
909 SectionTableRef SecTable) {
910 if (Link != SHN_UNDEF) {
911 Expected<SymTabType *> Sec = SecTable.getSectionOfType<SymTabType>(
912 Link,
913 "Link field value " + Twine(Link) + " in section " + Name +
914 " is invalid",
915 "Link field value " + Twine(Link) + " in section " + Name +
916 " is not a symbol table");
917 if (!Sec)
918 return Sec.takeError();
920 setSymTab(*Sec);
923 if (Info != SHN_UNDEF) {
924 Expected<SectionBase *> Sec =
925 SecTable.getSection(Info, "Info field value " + Twine(Info) +
926 " in section " + Name + " is invalid");
927 if (!Sec)
928 return Sec.takeError();
930 setSection(*Sec);
931 } else
932 setSection(nullptr);
934 return Error::success();
937 template <class SymTabType>
938 void RelocSectionWithSymtabBase<SymTabType>::finalize() {
939 this->Link = Symbols ? Symbols->Index : 0;
941 if (SecToApplyRel != nullptr)
942 this->Info = SecToApplyRel->Index;
945 template <class ELFT>
946 static void setAddend(Elf_Rel_Impl<ELFT, false> &, uint64_t) {}
948 template <class ELFT>
949 static void setAddend(Elf_Rel_Impl<ELFT, true> &Rela, uint64_t Addend) {
950 Rela.r_addend = Addend;
953 template <class RelRange, class T>
954 static void writeRel(const RelRange &Relocations, T *Buf, bool IsMips64EL) {
955 for (const auto &Reloc : Relocations) {
956 Buf->r_offset = Reloc.Offset;
957 setAddend(*Buf, Reloc.Addend);
958 Buf->setSymbolAndType(Reloc.RelocSymbol ? Reloc.RelocSymbol->Index : 0,
959 Reloc.Type, IsMips64EL);
960 ++Buf;
964 template <class ELFT>
965 Error ELFSectionWriter<ELFT>::visit(const RelocationSection &Sec) {
966 uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
967 if (Sec.Type == SHT_REL)
968 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rel *>(Buf),
969 Sec.getObject().IsMips64EL);
970 else
971 writeRel(Sec.Relocations, reinterpret_cast<Elf_Rela *>(Buf),
972 Sec.getObject().IsMips64EL);
973 return Error::success();
976 Error RelocationSection::accept(SectionVisitor &Visitor) const {
977 return Visitor.visit(*this);
980 Error RelocationSection::accept(MutableSectionVisitor &Visitor) {
981 return Visitor.visit(*this);
984 Error RelocationSection::removeSymbols(
985 function_ref<bool(const Symbol &)> ToRemove) {
986 for (const Relocation &Reloc : Relocations)
987 if (Reloc.RelocSymbol && ToRemove(*Reloc.RelocSymbol))
988 return createStringError(
989 llvm::errc::invalid_argument,
990 "not stripping symbol '%s' because it is named in a relocation",
991 Reloc.RelocSymbol->Name.data());
992 return Error::success();
995 void RelocationSection::markSymbols() {
996 for (const Relocation &Reloc : Relocations)
997 if (Reloc.RelocSymbol)
998 Reloc.RelocSymbol->Referenced = true;
1001 void RelocationSection::replaceSectionReferences(
1002 const DenseMap<SectionBase *, SectionBase *> &FromTo) {
1003 // Update the target section if it was replaced.
1004 if (SectionBase *To = FromTo.lookup(SecToApplyRel))
1005 SecToApplyRel = To;
1008 Error SectionWriter::visit(const DynamicRelocationSection &Sec) {
1009 llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset);
1010 return Error::success();
1013 Error DynamicRelocationSection::accept(SectionVisitor &Visitor) const {
1014 return Visitor.visit(*this);
1017 Error DynamicRelocationSection::accept(MutableSectionVisitor &Visitor) {
1018 return Visitor.visit(*this);
1021 Error DynamicRelocationSection::removeSectionReferences(
1022 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
1023 if (ToRemove(Symbols)) {
1024 if (!AllowBrokenLinks)
1025 return createStringError(
1026 llvm::errc::invalid_argument,
1027 "symbol table '%s' cannot be removed because it is "
1028 "referenced by the relocation section '%s'",
1029 Symbols->Name.data(), this->Name.data());
1030 Symbols = nullptr;
1033 // SecToApplyRel contains a section referenced by sh_info field. It keeps
1034 // a section to which the relocation section applies. When we remove any
1035 // sections we also remove their relocation sections. Since we do that much
1036 // earlier, this assert should never be triggered.
1037 assert(!SecToApplyRel || !ToRemove(SecToApplyRel));
1038 return Error::success();
1041 Error Section::removeSectionReferences(
1042 bool AllowBrokenDependency,
1043 function_ref<bool(const SectionBase *)> ToRemove) {
1044 if (ToRemove(LinkSection)) {
1045 if (!AllowBrokenDependency)
1046 return createStringError(llvm::errc::invalid_argument,
1047 "section '%s' cannot be removed because it is "
1048 "referenced by the section '%s'",
1049 LinkSection->Name.data(), this->Name.data());
1050 LinkSection = nullptr;
1052 return Error::success();
1055 void GroupSection::finalize() {
1056 this->Info = Sym ? Sym->Index : 0;
1057 this->Link = SymTab ? SymTab->Index : 0;
1058 // Linker deduplication for GRP_COMDAT is based on Sym->Name. The local/global
1059 // status is not part of the equation. If Sym is localized, the intention is
1060 // likely to make the group fully localized. Drop GRP_COMDAT to suppress
1061 // deduplication. See https://groups.google.com/g/generic-abi/c/2X6mR-s2zoc
1062 if ((FlagWord & GRP_COMDAT) && Sym && Sym->Binding == STB_LOCAL)
1063 this->FlagWord &= ~GRP_COMDAT;
1066 Error GroupSection::removeSectionReferences(
1067 bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
1068 if (ToRemove(SymTab)) {
1069 if (!AllowBrokenLinks)
1070 return createStringError(
1071 llvm::errc::invalid_argument,
1072 "section '.symtab' cannot be removed because it is "
1073 "referenced by the group section '%s'",
1074 this->Name.data());
1075 SymTab = nullptr;
1076 Sym = nullptr;
1078 llvm::erase_if(GroupMembers, ToRemove);
1079 return Error::success();
1082 Error GroupSection::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
1083 if (ToRemove(*Sym))
1084 return createStringError(llvm::errc::invalid_argument,
1085 "symbol '%s' cannot be removed because it is "
1086 "referenced by the section '%s[%d]'",
1087 Sym->Name.data(), this->Name.data(), this->Index);
1088 return Error::success();
1091 void GroupSection::markSymbols() {
1092 if (Sym)
1093 Sym->Referenced = true;
1096 void GroupSection::replaceSectionReferences(
1097 const DenseMap<SectionBase *, SectionBase *> &FromTo) {
1098 for (SectionBase *&Sec : GroupMembers)
1099 if (SectionBase *To = FromTo.lookup(Sec))
1100 Sec = To;
1103 void GroupSection::onRemove() {
1104 // As the header section of the group is removed, drop the Group flag in its
1105 // former members.
1106 for (SectionBase *Sec : GroupMembers)
1107 Sec->Flags &= ~SHF_GROUP;
1110 Error Section::initialize(SectionTableRef SecTable) {
1111 if (Link == ELF::SHN_UNDEF)
1112 return Error::success();
1114 Expected<SectionBase *> Sec =
1115 SecTable.getSection(Link, "Link field value " + Twine(Link) +
1116 " in section " + Name + " is invalid");
1117 if (!Sec)
1118 return Sec.takeError();
1120 LinkSection = *Sec;
1122 if (LinkSection->Type == ELF::SHT_SYMTAB) {
1123 HasSymTabLink = true;
1124 LinkSection = nullptr;
1127 return Error::success();
1130 void Section::finalize() { this->Link = LinkSection ? LinkSection->Index : 0; }
1132 void GnuDebugLinkSection::init(StringRef File) {
1133 FileName = sys::path::filename(File);
1134 // The format for the .gnu_debuglink starts with the file name and is
1135 // followed by a null terminator and then the CRC32 of the file. The CRC32
1136 // should be 4 byte aligned. So we add the FileName size, a 1 for the null
1137 // byte, and then finally push the size to alignment and add 4.
1138 Size = alignTo(FileName.size() + 1, 4) + 4;
1139 // The CRC32 will only be aligned if we align the whole section.
1140 Align = 4;
1141 Type = OriginalType = ELF::SHT_PROGBITS;
1142 Name = ".gnu_debuglink";
1143 // For sections not found in segments, OriginalOffset is only used to
1144 // establish the order that sections should go in. By using the maximum
1145 // possible offset we cause this section to wind up at the end.
1146 OriginalOffset = std::numeric_limits<uint64_t>::max();
1149 GnuDebugLinkSection::GnuDebugLinkSection(StringRef File,
1150 uint32_t PrecomputedCRC)
1151 : FileName(File), CRC32(PrecomputedCRC) {
1152 init(File);
1155 template <class ELFT>
1156 Error ELFSectionWriter<ELFT>::visit(const GnuDebugLinkSection &Sec) {
1157 unsigned char *Buf =
1158 reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
1159 Elf_Word *CRC =
1160 reinterpret_cast<Elf_Word *>(Buf + Sec.Size - sizeof(Elf_Word));
1161 *CRC = Sec.CRC32;
1162 llvm::copy(Sec.FileName, Buf);
1163 return Error::success();
1166 Error GnuDebugLinkSection::accept(SectionVisitor &Visitor) const {
1167 return Visitor.visit(*this);
1170 Error GnuDebugLinkSection::accept(MutableSectionVisitor &Visitor) {
1171 return Visitor.visit(*this);
1174 template <class ELFT>
1175 Error ELFSectionWriter<ELFT>::visit(const GroupSection &Sec) {
1176 ELF::Elf32_Word *Buf =
1177 reinterpret_cast<ELF::Elf32_Word *>(Out.getBufferStart() + Sec.Offset);
1178 support::endian::write32<ELFT::TargetEndianness>(Buf++, Sec.FlagWord);
1179 for (SectionBase *S : Sec.GroupMembers)
1180 support::endian::write32<ELFT::TargetEndianness>(Buf++, S->Index);
1181 return Error::success();
1184 Error GroupSection::accept(SectionVisitor &Visitor) const {
1185 return Visitor.visit(*this);
1188 Error GroupSection::accept(MutableSectionVisitor &Visitor) {
1189 return Visitor.visit(*this);
1192 // Returns true IFF a section is wholly inside the range of a segment
1193 static bool sectionWithinSegment(const SectionBase &Sec, const Segment &Seg) {
1194 // If a section is empty it should be treated like it has a size of 1. This is
1195 // to clarify the case when an empty section lies on a boundary between two
1196 // segments and ensures that the section "belongs" to the second segment and
1197 // not the first.
1198 uint64_t SecSize = Sec.Size ? Sec.Size : 1;
1200 // Ignore just added sections.
1201 if (Sec.OriginalOffset == std::numeric_limits<uint64_t>::max())
1202 return false;
1204 if (Sec.Type == SHT_NOBITS) {
1205 if (!(Sec.Flags & SHF_ALLOC))
1206 return false;
1208 bool SectionIsTLS = Sec.Flags & SHF_TLS;
1209 bool SegmentIsTLS = Seg.Type == PT_TLS;
1210 if (SectionIsTLS != SegmentIsTLS)
1211 return false;
1213 return Seg.VAddr <= Sec.Addr &&
1214 Seg.VAddr + Seg.MemSize >= Sec.Addr + SecSize;
1217 return Seg.Offset <= Sec.OriginalOffset &&
1218 Seg.Offset + Seg.FileSize >= Sec.OriginalOffset + SecSize;
1221 // Returns true IFF a segment's original offset is inside of another segment's
1222 // range.
1223 static bool segmentOverlapsSegment(const Segment &Child,
1224 const Segment &Parent) {
1226 return Parent.OriginalOffset <= Child.OriginalOffset &&
1227 Parent.OriginalOffset + Parent.FileSize > Child.OriginalOffset;
1230 static bool compareSegmentsByOffset(const Segment *A, const Segment *B) {
1231 // Any segment without a parent segment should come before a segment
1232 // that has a parent segment.
1233 if (A->OriginalOffset < B->OriginalOffset)
1234 return true;
1235 if (A->OriginalOffset > B->OriginalOffset)
1236 return false;
1237 return A->Index < B->Index;
1240 void BasicELFBuilder::initFileHeader() {
1241 Obj->Flags = 0x0;
1242 Obj->Type = ET_REL;
1243 Obj->OSABI = ELFOSABI_NONE;
1244 Obj->ABIVersion = 0;
1245 Obj->Entry = 0x0;
1246 Obj->Machine = EM_NONE;
1247 Obj->Version = 1;
1250 void BasicELFBuilder::initHeaderSegment() { Obj->ElfHdrSegment.Index = 0; }
1252 StringTableSection *BasicELFBuilder::addStrTab() {
1253 auto &StrTab = Obj->addSection<StringTableSection>();
1254 StrTab.Name = ".strtab";
1256 Obj->SectionNames = &StrTab;
1257 return &StrTab;
1260 SymbolTableSection *BasicELFBuilder::addSymTab(StringTableSection *StrTab) {
1261 auto &SymTab = Obj->addSection<SymbolTableSection>();
1263 SymTab.Name = ".symtab";
1264 SymTab.Link = StrTab->Index;
1266 // The symbol table always needs a null symbol
1267 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0);
1269 Obj->SymbolTable = &SymTab;
1270 return &SymTab;
1273 Error BasicELFBuilder::initSections() {
1274 for (SectionBase &Sec : Obj->sections())
1275 if (Error Err = Sec.initialize(Obj->sections()))
1276 return Err;
1278 return Error::success();
1281 void BinaryELFBuilder::addData(SymbolTableSection *SymTab) {
1282 auto Data = ArrayRef<uint8_t>(
1283 reinterpret_cast<const uint8_t *>(MemBuf->getBufferStart()),
1284 MemBuf->getBufferSize());
1285 auto &DataSection = Obj->addSection<Section>(Data);
1286 DataSection.Name = ".data";
1287 DataSection.Type = ELF::SHT_PROGBITS;
1288 DataSection.Size = Data.size();
1289 DataSection.Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE;
1291 std::string SanitizedFilename = MemBuf->getBufferIdentifier().str();
1292 std::replace_if(
1293 std::begin(SanitizedFilename), std::end(SanitizedFilename),
1294 [](char C) { return !isAlnum(C); }, '_');
1295 Twine Prefix = Twine("_binary_") + SanitizedFilename;
1297 SymTab->addSymbol(Prefix + "_start", STB_GLOBAL, STT_NOTYPE, &DataSection,
1298 /*Value=*/0, NewSymbolVisibility, 0, 0);
1299 SymTab->addSymbol(Prefix + "_end", STB_GLOBAL, STT_NOTYPE, &DataSection,
1300 /*Value=*/DataSection.Size, NewSymbolVisibility, 0, 0);
1301 SymTab->addSymbol(Prefix + "_size", STB_GLOBAL, STT_NOTYPE, nullptr,
1302 /*Value=*/DataSection.Size, NewSymbolVisibility, SHN_ABS,
1306 Expected<std::unique_ptr<Object>> BinaryELFBuilder::build() {
1307 initFileHeader();
1308 initHeaderSegment();
1310 SymbolTableSection *SymTab = addSymTab(addStrTab());
1311 if (Error Err = initSections())
1312 return std::move(Err);
1313 addData(SymTab);
1315 return std::move(Obj);
1318 // Adds sections from IHEX data file. Data should have been
1319 // fully validated by this time.
1320 void IHexELFBuilder::addDataSections() {
1321 OwnedDataSection *Section = nullptr;
1322 uint64_t SegmentAddr = 0, BaseAddr = 0;
1323 uint32_t SecNo = 1;
1325 for (const IHexRecord &R : Records) {
1326 uint64_t RecAddr;
1327 switch (R.Type) {
1328 case IHexRecord::Data:
1329 // Ignore empty data records
1330 if (R.HexData.empty())
1331 continue;
1332 RecAddr = R.Addr + SegmentAddr + BaseAddr;
1333 if (!Section || Section->Addr + Section->Size != RecAddr) {
1334 // OriginalOffset field is only used to sort sections before layout, so
1335 // instead of keeping track of real offsets in IHEX file, and as
1336 // layoutSections() and layoutSectionsForOnlyKeepDebug() use
1337 // llvm::stable_sort(), we can just set it to a constant (zero).
1338 Section = &Obj->addSection<OwnedDataSection>(
1339 ".sec" + std::to_string(SecNo), RecAddr,
1340 ELF::SHF_ALLOC | ELF::SHF_WRITE, 0);
1341 SecNo++;
1343 Section->appendHexData(R.HexData);
1344 break;
1345 case IHexRecord::EndOfFile:
1346 break;
1347 case IHexRecord::SegmentAddr:
1348 // 20-bit segment address.
1349 SegmentAddr = checkedGetHex<uint16_t>(R.HexData) << 4;
1350 break;
1351 case IHexRecord::StartAddr80x86:
1352 case IHexRecord::StartAddr:
1353 Obj->Entry = checkedGetHex<uint32_t>(R.HexData);
1354 assert(Obj->Entry <= 0xFFFFFU);
1355 break;
1356 case IHexRecord::ExtendedAddr:
1357 // 16-31 bits of linear base address
1358 BaseAddr = checkedGetHex<uint16_t>(R.HexData) << 16;
1359 break;
1360 default:
1361 llvm_unreachable("unknown record type");
1366 Expected<std::unique_ptr<Object>> IHexELFBuilder::build() {
1367 initFileHeader();
1368 initHeaderSegment();
1369 StringTableSection *StrTab = addStrTab();
1370 addSymTab(StrTab);
1371 if (Error Err = initSections())
1372 return std::move(Err);
1373 addDataSections();
1375 return std::move(Obj);
1378 template <class ELFT>
1379 ELFBuilder<ELFT>::ELFBuilder(const ELFObjectFile<ELFT> &ElfObj, Object &Obj,
1380 std::optional<StringRef> ExtractPartition)
1381 : ElfFile(ElfObj.getELFFile()), Obj(Obj),
1382 ExtractPartition(ExtractPartition) {
1383 Obj.IsMips64EL = ElfFile.isMips64EL();
1386 template <class ELFT> void ELFBuilder<ELFT>::setParentSegment(Segment &Child) {
1387 for (Segment &Parent : Obj.segments()) {
1388 // Every segment will overlap with itself but we don't want a segment to
1389 // be its own parent so we avoid that situation.
1390 if (&Child != &Parent && segmentOverlapsSegment(Child, Parent)) {
1391 // We want a canonical "most parental" segment but this requires
1392 // inspecting the ParentSegment.
1393 if (compareSegmentsByOffset(&Parent, &Child))
1394 if (Child.ParentSegment == nullptr ||
1395 compareSegmentsByOffset(&Parent, Child.ParentSegment)) {
1396 Child.ParentSegment = &Parent;
1402 template <class ELFT> Error ELFBuilder<ELFT>::findEhdrOffset() {
1403 if (!ExtractPartition)
1404 return Error::success();
1406 for (const SectionBase &Sec : Obj.sections()) {
1407 if (Sec.Type == SHT_LLVM_PART_EHDR && Sec.Name == *ExtractPartition) {
1408 EhdrOffset = Sec.Offset;
1409 return Error::success();
1412 return createStringError(errc::invalid_argument,
1413 "could not find partition named '" +
1414 *ExtractPartition + "'");
1417 template <class ELFT>
1418 Error ELFBuilder<ELFT>::readProgramHeaders(const ELFFile<ELFT> &HeadersFile) {
1419 uint32_t Index = 0;
1421 Expected<typename ELFFile<ELFT>::Elf_Phdr_Range> Headers =
1422 HeadersFile.program_headers();
1423 if (!Headers)
1424 return Headers.takeError();
1426 for (const typename ELFFile<ELFT>::Elf_Phdr &Phdr : *Headers) {
1427 if (Phdr.p_offset + Phdr.p_filesz > HeadersFile.getBufSize())
1428 return createStringError(
1429 errc::invalid_argument,
1430 "program header with offset 0x" + Twine::utohexstr(Phdr.p_offset) +
1431 " and file size 0x" + Twine::utohexstr(Phdr.p_filesz) +
1432 " goes past the end of the file");
1434 ArrayRef<uint8_t> Data{HeadersFile.base() + Phdr.p_offset,
1435 (size_t)Phdr.p_filesz};
1436 Segment &Seg = Obj.addSegment(Data);
1437 Seg.Type = Phdr.p_type;
1438 Seg.Flags = Phdr.p_flags;
1439 Seg.OriginalOffset = Phdr.p_offset + EhdrOffset;
1440 Seg.Offset = Phdr.p_offset + EhdrOffset;
1441 Seg.VAddr = Phdr.p_vaddr;
1442 Seg.PAddr = Phdr.p_paddr;
1443 Seg.FileSize = Phdr.p_filesz;
1444 Seg.MemSize = Phdr.p_memsz;
1445 Seg.Align = Phdr.p_align;
1446 Seg.Index = Index++;
1447 for (SectionBase &Sec : Obj.sections())
1448 if (sectionWithinSegment(Sec, Seg)) {
1449 Seg.addSection(&Sec);
1450 if (!Sec.ParentSegment || Sec.ParentSegment->Offset > Seg.Offset)
1451 Sec.ParentSegment = &Seg;
1455 auto &ElfHdr = Obj.ElfHdrSegment;
1456 ElfHdr.Index = Index++;
1457 ElfHdr.OriginalOffset = ElfHdr.Offset = EhdrOffset;
1459 const typename ELFT::Ehdr &Ehdr = HeadersFile.getHeader();
1460 auto &PrHdr = Obj.ProgramHdrSegment;
1461 PrHdr.Type = PT_PHDR;
1462 PrHdr.Flags = 0;
1463 // The spec requires us to have p_vaddr % p_align == p_offset % p_align.
1464 // Whereas this works automatically for ElfHdr, here OriginalOffset is
1465 // always non-zero and to ensure the equation we assign the same value to
1466 // VAddr as well.
1467 PrHdr.OriginalOffset = PrHdr.Offset = PrHdr.VAddr = EhdrOffset + Ehdr.e_phoff;
1468 PrHdr.PAddr = 0;
1469 PrHdr.FileSize = PrHdr.MemSize = Ehdr.e_phentsize * Ehdr.e_phnum;
1470 // The spec requires us to naturally align all the fields.
1471 PrHdr.Align = sizeof(Elf_Addr);
1472 PrHdr.Index = Index++;
1474 // Now we do an O(n^2) loop through the segments in order to match up
1475 // segments.
1476 for (Segment &Child : Obj.segments())
1477 setParentSegment(Child);
1478 setParentSegment(ElfHdr);
1479 setParentSegment(PrHdr);
1481 return Error::success();
1484 template <class ELFT>
1485 Error ELFBuilder<ELFT>::initGroupSection(GroupSection *GroupSec) {
1486 if (GroupSec->Align % sizeof(ELF::Elf32_Word) != 0)
1487 return createStringError(errc::invalid_argument,
1488 "invalid alignment " + Twine(GroupSec->Align) +
1489 " of group section '" + GroupSec->Name + "'");
1490 SectionTableRef SecTable = Obj.sections();
1491 if (GroupSec->Link != SHN_UNDEF) {
1492 auto SymTab = SecTable.template getSectionOfType<SymbolTableSection>(
1493 GroupSec->Link,
1494 "link field value '" + Twine(GroupSec->Link) + "' in section '" +
1495 GroupSec->Name + "' is invalid",
1496 "link field value '" + Twine(GroupSec->Link) + "' in section '" +
1497 GroupSec->Name + "' is not a symbol table");
1498 if (!SymTab)
1499 return SymTab.takeError();
1501 Expected<Symbol *> Sym = (*SymTab)->getSymbolByIndex(GroupSec->Info);
1502 if (!Sym)
1503 return createStringError(errc::invalid_argument,
1504 "info field value '" + Twine(GroupSec->Info) +
1505 "' in section '" + GroupSec->Name +
1506 "' is not a valid symbol index");
1507 GroupSec->setSymTab(*SymTab);
1508 GroupSec->setSymbol(*Sym);
1510 if (GroupSec->Contents.size() % sizeof(ELF::Elf32_Word) ||
1511 GroupSec->Contents.empty())
1512 return createStringError(errc::invalid_argument,
1513 "the content of the section " + GroupSec->Name +
1514 " is malformed");
1515 const ELF::Elf32_Word *Word =
1516 reinterpret_cast<const ELF::Elf32_Word *>(GroupSec->Contents.data());
1517 const ELF::Elf32_Word *End =
1518 Word + GroupSec->Contents.size() / sizeof(ELF::Elf32_Word);
1519 GroupSec->setFlagWord(
1520 support::endian::read32<ELFT::TargetEndianness>(Word++));
1521 for (; Word != End; ++Word) {
1522 uint32_t Index = support::endian::read32<ELFT::TargetEndianness>(Word);
1523 Expected<SectionBase *> Sec = SecTable.getSection(
1524 Index, "group member index " + Twine(Index) + " in section '" +
1525 GroupSec->Name + "' is invalid");
1526 if (!Sec)
1527 return Sec.takeError();
1529 GroupSec->addMember(*Sec);
1532 return Error::success();
1535 template <class ELFT>
1536 Error ELFBuilder<ELFT>::initSymbolTable(SymbolTableSection *SymTab) {
1537 Expected<const Elf_Shdr *> Shdr = ElfFile.getSection(SymTab->Index);
1538 if (!Shdr)
1539 return Shdr.takeError();
1541 Expected<StringRef> StrTabData = ElfFile.getStringTableForSymtab(**Shdr);
1542 if (!StrTabData)
1543 return StrTabData.takeError();
1545 ArrayRef<Elf_Word> ShndxData;
1547 Expected<typename ELFFile<ELFT>::Elf_Sym_Range> Symbols =
1548 ElfFile.symbols(*Shdr);
1549 if (!Symbols)
1550 return Symbols.takeError();
1552 for (const typename ELFFile<ELFT>::Elf_Sym &Sym : *Symbols) {
1553 SectionBase *DefSection = nullptr;
1555 Expected<StringRef> Name = Sym.getName(*StrTabData);
1556 if (!Name)
1557 return Name.takeError();
1559 if (Sym.st_shndx == SHN_XINDEX) {
1560 if (SymTab->getShndxTable() == nullptr)
1561 return createStringError(errc::invalid_argument,
1562 "symbol '" + *Name +
1563 "' has index SHN_XINDEX but no "
1564 "SHT_SYMTAB_SHNDX section exists");
1565 if (ShndxData.data() == nullptr) {
1566 Expected<const Elf_Shdr *> ShndxSec =
1567 ElfFile.getSection(SymTab->getShndxTable()->Index);
1568 if (!ShndxSec)
1569 return ShndxSec.takeError();
1571 Expected<ArrayRef<Elf_Word>> Data =
1572 ElfFile.template getSectionContentsAsArray<Elf_Word>(**ShndxSec);
1573 if (!Data)
1574 return Data.takeError();
1576 ShndxData = *Data;
1577 if (ShndxData.size() != Symbols->size())
1578 return createStringError(
1579 errc::invalid_argument,
1580 "symbol section index table does not have the same number of "
1581 "entries as the symbol table");
1583 Elf_Word Index = ShndxData[&Sym - Symbols->begin()];
1584 Expected<SectionBase *> Sec = Obj.sections().getSection(
1585 Index,
1586 "symbol '" + *Name + "' has invalid section index " + Twine(Index));
1587 if (!Sec)
1588 return Sec.takeError();
1590 DefSection = *Sec;
1591 } else if (Sym.st_shndx >= SHN_LORESERVE) {
1592 if (!isValidReservedSectionIndex(Sym.st_shndx, Obj.Machine)) {
1593 return createStringError(
1594 errc::invalid_argument,
1595 "symbol '" + *Name +
1596 "' has unsupported value greater than or equal "
1597 "to SHN_LORESERVE: " +
1598 Twine(Sym.st_shndx));
1600 } else if (Sym.st_shndx != SHN_UNDEF) {
1601 Expected<SectionBase *> Sec = Obj.sections().getSection(
1602 Sym.st_shndx, "symbol '" + *Name +
1603 "' is defined has invalid section index " +
1604 Twine(Sym.st_shndx));
1605 if (!Sec)
1606 return Sec.takeError();
1608 DefSection = *Sec;
1611 SymTab->addSymbol(*Name, Sym.getBinding(), Sym.getType(), DefSection,
1612 Sym.getValue(), Sym.st_other, Sym.st_shndx, Sym.st_size);
1615 return Error::success();
1618 template <class ELFT>
1619 static void getAddend(uint64_t &, const Elf_Rel_Impl<ELFT, false> &) {}
1621 template <class ELFT>
1622 static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, true> &Rela) {
1623 ToSet = Rela.r_addend;
1626 template <class T>
1627 static Error initRelocations(RelocationSection *Relocs, T RelRange) {
1628 for (const auto &Rel : RelRange) {
1629 Relocation ToAdd;
1630 ToAdd.Offset = Rel.r_offset;
1631 getAddend(ToAdd.Addend, Rel);
1632 ToAdd.Type = Rel.getType(Relocs->getObject().IsMips64EL);
1634 if (uint32_t Sym = Rel.getSymbol(Relocs->getObject().IsMips64EL)) {
1635 if (!Relocs->getObject().SymbolTable)
1636 return createStringError(
1637 errc::invalid_argument,
1638 "'" + Relocs->Name + "': relocation references symbol with index " +
1639 Twine(Sym) + ", but there is no symbol table");
1640 Expected<Symbol *> SymByIndex =
1641 Relocs->getObject().SymbolTable->getSymbolByIndex(Sym);
1642 if (!SymByIndex)
1643 return SymByIndex.takeError();
1645 ToAdd.RelocSymbol = *SymByIndex;
1648 Relocs->addRelocation(ToAdd);
1651 return Error::success();
1654 Expected<SectionBase *> SectionTableRef::getSection(uint32_t Index,
1655 Twine ErrMsg) {
1656 if (Index == SHN_UNDEF || Index > Sections.size())
1657 return createStringError(errc::invalid_argument, ErrMsg);
1658 return Sections[Index - 1].get();
1661 template <class T>
1662 Expected<T *> SectionTableRef::getSectionOfType(uint32_t Index,
1663 Twine IndexErrMsg,
1664 Twine TypeErrMsg) {
1665 Expected<SectionBase *> BaseSec = getSection(Index, IndexErrMsg);
1666 if (!BaseSec)
1667 return BaseSec.takeError();
1669 if (T *Sec = dyn_cast<T>(*BaseSec))
1670 return Sec;
1672 return createStringError(errc::invalid_argument, TypeErrMsg);
1675 template <class ELFT>
1676 Expected<SectionBase &> ELFBuilder<ELFT>::makeSection(const Elf_Shdr &Shdr) {
1677 switch (Shdr.sh_type) {
1678 case SHT_REL:
1679 case SHT_RELA:
1680 if (Shdr.sh_flags & SHF_ALLOC) {
1681 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1682 return Obj.addSection<DynamicRelocationSection>(*Data);
1683 else
1684 return Data.takeError();
1686 return Obj.addSection<RelocationSection>(Obj);
1687 case SHT_STRTAB:
1688 // If a string table is allocated we don't want to mess with it. That would
1689 // mean altering the memory image. There are no special link types or
1690 // anything so we can just use a Section.
1691 if (Shdr.sh_flags & SHF_ALLOC) {
1692 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1693 return Obj.addSection<Section>(*Data);
1694 else
1695 return Data.takeError();
1697 return Obj.addSection<StringTableSection>();
1698 case SHT_HASH:
1699 case SHT_GNU_HASH:
1700 // Hash tables should refer to SHT_DYNSYM which we're not going to change.
1701 // Because of this we don't need to mess with the hash tables either.
1702 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1703 return Obj.addSection<Section>(*Data);
1704 else
1705 return Data.takeError();
1706 case SHT_GROUP:
1707 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1708 return Obj.addSection<GroupSection>(*Data);
1709 else
1710 return Data.takeError();
1711 case SHT_DYNSYM:
1712 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1713 return Obj.addSection<DynamicSymbolTableSection>(*Data);
1714 else
1715 return Data.takeError();
1716 case SHT_DYNAMIC:
1717 if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1718 return Obj.addSection<DynamicSection>(*Data);
1719 else
1720 return Data.takeError();
1721 case SHT_SYMTAB: {
1722 // Multiple SHT_SYMTAB sections are forbidden by the ELF gABI.
1723 if (Obj.SymbolTable != nullptr)
1724 return createStringError(llvm::errc::invalid_argument,
1725 "found multiple SHT_SYMTAB sections");
1726 auto &SymTab = Obj.addSection<SymbolTableSection>();
1727 Obj.SymbolTable = &SymTab;
1728 return SymTab;
1730 case SHT_SYMTAB_SHNDX: {
1731 auto &ShndxSection = Obj.addSection<SectionIndexSection>();
1732 Obj.SectionIndexTable = &ShndxSection;
1733 return ShndxSection;
1735 case SHT_NOBITS:
1736 return Obj.addSection<Section>(ArrayRef<uint8_t>());
1737 default: {
1738 Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr);
1739 if (!Data)
1740 return Data.takeError();
1742 Expected<StringRef> Name = ElfFile.getSectionName(Shdr);
1743 if (!Name)
1744 return Name.takeError();
1746 if (!(Shdr.sh_flags & ELF::SHF_COMPRESSED))
1747 return Obj.addSection<Section>(*Data);
1748 auto *Chdr = reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data->data());
1749 return Obj.addSection<CompressedSection>(CompressedSection(
1750 *Data, Chdr->ch_type, Chdr->ch_size, Chdr->ch_addralign));
1755 template <class ELFT> Error ELFBuilder<ELFT>::readSectionHeaders() {
1756 uint32_t Index = 0;
1757 Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections =
1758 ElfFile.sections();
1759 if (!Sections)
1760 return Sections.takeError();
1762 for (const typename ELFFile<ELFT>::Elf_Shdr &Shdr : *Sections) {
1763 if (Index == 0) {
1764 ++Index;
1765 continue;
1767 Expected<SectionBase &> Sec = makeSection(Shdr);
1768 if (!Sec)
1769 return Sec.takeError();
1771 Expected<StringRef> SecName = ElfFile.getSectionName(Shdr);
1772 if (!SecName)
1773 return SecName.takeError();
1774 Sec->Name = SecName->str();
1775 Sec->Type = Sec->OriginalType = Shdr.sh_type;
1776 Sec->Flags = Sec->OriginalFlags = Shdr.sh_flags;
1777 Sec->Addr = Shdr.sh_addr;
1778 Sec->Offset = Shdr.sh_offset;
1779 Sec->OriginalOffset = Shdr.sh_offset;
1780 Sec->Size = Shdr.sh_size;
1781 Sec->Link = Shdr.sh_link;
1782 Sec->Info = Shdr.sh_info;
1783 Sec->Align = Shdr.sh_addralign;
1784 Sec->EntrySize = Shdr.sh_entsize;
1785 Sec->Index = Index++;
1786 Sec->OriginalIndex = Sec->Index;
1787 Sec->OriginalData = ArrayRef<uint8_t>(
1788 ElfFile.base() + Shdr.sh_offset,
1789 (Shdr.sh_type == SHT_NOBITS) ? (size_t)0 : Shdr.sh_size);
1792 return Error::success();
1795 template <class ELFT> Error ELFBuilder<ELFT>::readSections(bool EnsureSymtab) {
1796 uint32_t ShstrIndex = ElfFile.getHeader().e_shstrndx;
1797 if (ShstrIndex == SHN_XINDEX) {
1798 Expected<const Elf_Shdr *> Sec = ElfFile.getSection(0);
1799 if (!Sec)
1800 return Sec.takeError();
1802 ShstrIndex = (*Sec)->sh_link;
1805 if (ShstrIndex == SHN_UNDEF)
1806 Obj.HadShdrs = false;
1807 else {
1808 Expected<StringTableSection *> Sec =
1809 Obj.sections().template getSectionOfType<StringTableSection>(
1810 ShstrIndex,
1811 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " +
1812 " is invalid",
1813 "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " +
1814 " does not reference a string table");
1815 if (!Sec)
1816 return Sec.takeError();
1818 Obj.SectionNames = *Sec;
1821 // If a section index table exists we'll need to initialize it before we
1822 // initialize the symbol table because the symbol table might need to
1823 // reference it.
1824 if (Obj.SectionIndexTable)
1825 if (Error Err = Obj.SectionIndexTable->initialize(Obj.sections()))
1826 return Err;
1828 // Now that all of the sections have been added we can fill out some extra
1829 // details about symbol tables. We need the symbol table filled out before
1830 // any relocations.
1831 if (Obj.SymbolTable) {
1832 if (Error Err = Obj.SymbolTable->initialize(Obj.sections()))
1833 return Err;
1834 if (Error Err = initSymbolTable(Obj.SymbolTable))
1835 return Err;
1836 } else if (EnsureSymtab) {
1837 if (Error Err = Obj.addNewSymbolTable())
1838 return Err;
1841 // Now that all sections and symbols have been added we can add
1842 // relocations that reference symbols and set the link and info fields for
1843 // relocation sections.
1844 for (SectionBase &Sec : Obj.sections()) {
1845 if (&Sec == Obj.SymbolTable)
1846 continue;
1847 if (Error Err = Sec.initialize(Obj.sections()))
1848 return Err;
1849 if (auto RelSec = dyn_cast<RelocationSection>(&Sec)) {
1850 Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections =
1851 ElfFile.sections();
1852 if (!Sections)
1853 return Sections.takeError();
1855 const typename ELFFile<ELFT>::Elf_Shdr *Shdr =
1856 Sections->begin() + RelSec->Index;
1857 if (RelSec->Type == SHT_REL) {
1858 Expected<typename ELFFile<ELFT>::Elf_Rel_Range> Rels =
1859 ElfFile.rels(*Shdr);
1860 if (!Rels)
1861 return Rels.takeError();
1863 if (Error Err = initRelocations(RelSec, *Rels))
1864 return Err;
1865 } else {
1866 Expected<typename ELFFile<ELFT>::Elf_Rela_Range> Relas =
1867 ElfFile.relas(*Shdr);
1868 if (!Relas)
1869 return Relas.takeError();
1871 if (Error Err = initRelocations(RelSec, *Relas))
1872 return Err;
1874 } else if (auto GroupSec = dyn_cast<GroupSection>(&Sec)) {
1875 if (Error Err = initGroupSection(GroupSec))
1876 return Err;
1880 return Error::success();
1883 template <class ELFT> Error ELFBuilder<ELFT>::build(bool EnsureSymtab) {
1884 if (Error E = readSectionHeaders())
1885 return E;
1886 if (Error E = findEhdrOffset())
1887 return E;
1889 // The ELFFile whose ELF headers and program headers are copied into the
1890 // output file. Normally the same as ElfFile, but if we're extracting a
1891 // loadable partition it will point to the partition's headers.
1892 Expected<ELFFile<ELFT>> HeadersFile = ELFFile<ELFT>::create(toStringRef(
1893 {ElfFile.base() + EhdrOffset, ElfFile.getBufSize() - EhdrOffset}));
1894 if (!HeadersFile)
1895 return HeadersFile.takeError();
1897 const typename ELFFile<ELFT>::Elf_Ehdr &Ehdr = HeadersFile->getHeader();
1898 Obj.Is64Bits = Ehdr.e_ident[EI_CLASS] == ELFCLASS64;
1899 Obj.OSABI = Ehdr.e_ident[EI_OSABI];
1900 Obj.ABIVersion = Ehdr.e_ident[EI_ABIVERSION];
1901 Obj.Type = Ehdr.e_type;
1902 Obj.Machine = Ehdr.e_machine;
1903 Obj.Version = Ehdr.e_version;
1904 Obj.Entry = Ehdr.e_entry;
1905 Obj.Flags = Ehdr.e_flags;
1907 if (Error E = readSections(EnsureSymtab))
1908 return E;
1909 return readProgramHeaders(*HeadersFile);
1912 Writer::~Writer() = default;
1914 Reader::~Reader() = default;
1916 Expected<std::unique_ptr<Object>>
1917 BinaryReader::create(bool /*EnsureSymtab*/) const {
1918 return BinaryELFBuilder(MemBuf, NewSymbolVisibility).build();
1921 Expected<std::vector<IHexRecord>> IHexReader::parse() const {
1922 SmallVector<StringRef, 16> Lines;
1923 std::vector<IHexRecord> Records;
1924 bool HasSections = false;
1926 MemBuf->getBuffer().split(Lines, '\n');
1927 Records.reserve(Lines.size());
1928 for (size_t LineNo = 1; LineNo <= Lines.size(); ++LineNo) {
1929 StringRef Line = Lines[LineNo - 1].trim();
1930 if (Line.empty())
1931 continue;
1933 Expected<IHexRecord> R = IHexRecord::parse(Line);
1934 if (!R)
1935 return parseError(LineNo, R.takeError());
1936 if (R->Type == IHexRecord::EndOfFile)
1937 break;
1938 HasSections |= (R->Type == IHexRecord::Data);
1939 Records.push_back(*R);
1941 if (!HasSections)
1942 return parseError(-1U, "no sections");
1944 return std::move(Records);
1947 Expected<std::unique_ptr<Object>>
1948 IHexReader::create(bool /*EnsureSymtab*/) const {
1949 Expected<std::vector<IHexRecord>> Records = parse();
1950 if (!Records)
1951 return Records.takeError();
1953 return IHexELFBuilder(*Records).build();
1956 Expected<std::unique_ptr<Object>> ELFReader::create(bool EnsureSymtab) const {
1957 auto Obj = std::make_unique<Object>();
1958 if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Bin)) {
1959 ELFBuilder<ELF32LE> Builder(*O, *Obj, ExtractPartition);
1960 if (Error Err = Builder.build(EnsureSymtab))
1961 return std::move(Err);
1962 return std::move(Obj);
1963 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Bin)) {
1964 ELFBuilder<ELF64LE> Builder(*O, *Obj, ExtractPartition);
1965 if (Error Err = Builder.build(EnsureSymtab))
1966 return std::move(Err);
1967 return std::move(Obj);
1968 } else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Bin)) {
1969 ELFBuilder<ELF32BE> Builder(*O, *Obj, ExtractPartition);
1970 if (Error Err = Builder.build(EnsureSymtab))
1971 return std::move(Err);
1972 return std::move(Obj);
1973 } else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Bin)) {
1974 ELFBuilder<ELF64BE> Builder(*O, *Obj, ExtractPartition);
1975 if (Error Err = Builder.build(EnsureSymtab))
1976 return std::move(Err);
1977 return std::move(Obj);
1979 return createStringError(errc::invalid_argument, "invalid file type");
1982 template <class ELFT> void ELFWriter<ELFT>::writeEhdr() {
1983 Elf_Ehdr &Ehdr = *reinterpret_cast<Elf_Ehdr *>(Buf->getBufferStart());
1984 std::fill(Ehdr.e_ident, Ehdr.e_ident + 16, 0);
1985 Ehdr.e_ident[EI_MAG0] = 0x7f;
1986 Ehdr.e_ident[EI_MAG1] = 'E';
1987 Ehdr.e_ident[EI_MAG2] = 'L';
1988 Ehdr.e_ident[EI_MAG3] = 'F';
1989 Ehdr.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32;
1990 Ehdr.e_ident[EI_DATA] = ELFT::TargetEndianness == llvm::endianness::big
1991 ? ELFDATA2MSB
1992 : ELFDATA2LSB;
1993 Ehdr.e_ident[EI_VERSION] = EV_CURRENT;
1994 Ehdr.e_ident[EI_OSABI] = Obj.OSABI;
1995 Ehdr.e_ident[EI_ABIVERSION] = Obj.ABIVersion;
1997 Ehdr.e_type = Obj.Type;
1998 Ehdr.e_machine = Obj.Machine;
1999 Ehdr.e_version = Obj.Version;
2000 Ehdr.e_entry = Obj.Entry;
2001 // We have to use the fully-qualified name llvm::size
2002 // since some compilers complain on ambiguous resolution.
2003 Ehdr.e_phnum = llvm::size(Obj.segments());
2004 Ehdr.e_phoff = (Ehdr.e_phnum != 0) ? Obj.ProgramHdrSegment.Offset : 0;
2005 Ehdr.e_phentsize = (Ehdr.e_phnum != 0) ? sizeof(Elf_Phdr) : 0;
2006 Ehdr.e_flags = Obj.Flags;
2007 Ehdr.e_ehsize = sizeof(Elf_Ehdr);
2008 if (WriteSectionHeaders && Obj.sections().size() != 0) {
2009 Ehdr.e_shentsize = sizeof(Elf_Shdr);
2010 Ehdr.e_shoff = Obj.SHOff;
2011 // """
2012 // If the number of sections is greater than or equal to
2013 // SHN_LORESERVE (0xff00), this member has the value zero and the actual
2014 // number of section header table entries is contained in the sh_size field
2015 // of the section header at index 0.
2016 // """
2017 auto Shnum = Obj.sections().size() + 1;
2018 if (Shnum >= SHN_LORESERVE)
2019 Ehdr.e_shnum = 0;
2020 else
2021 Ehdr.e_shnum = Shnum;
2022 // """
2023 // If the section name string table section index is greater than or equal
2024 // to SHN_LORESERVE (0xff00), this member has the value SHN_XINDEX (0xffff)
2025 // and the actual index of the section name string table section is
2026 // contained in the sh_link field of the section header at index 0.
2027 // """
2028 if (Obj.SectionNames->Index >= SHN_LORESERVE)
2029 Ehdr.e_shstrndx = SHN_XINDEX;
2030 else
2031 Ehdr.e_shstrndx = Obj.SectionNames->Index;
2032 } else {
2033 Ehdr.e_shentsize = 0;
2034 Ehdr.e_shoff = 0;
2035 Ehdr.e_shnum = 0;
2036 Ehdr.e_shstrndx = 0;
2040 template <class ELFT> void ELFWriter<ELFT>::writePhdrs() {
2041 for (auto &Seg : Obj.segments())
2042 writePhdr(Seg);
2045 template <class ELFT> void ELFWriter<ELFT>::writeShdrs() {
2046 // This reference serves to write the dummy section header at the begining
2047 // of the file. It is not used for anything else
2048 Elf_Shdr &Shdr =
2049 *reinterpret_cast<Elf_Shdr *>(Buf->getBufferStart() + Obj.SHOff);
2050 Shdr.sh_name = 0;
2051 Shdr.sh_type = SHT_NULL;
2052 Shdr.sh_flags = 0;
2053 Shdr.sh_addr = 0;
2054 Shdr.sh_offset = 0;
2055 // See writeEhdr for why we do this.
2056 uint64_t Shnum = Obj.sections().size() + 1;
2057 if (Shnum >= SHN_LORESERVE)
2058 Shdr.sh_size = Shnum;
2059 else
2060 Shdr.sh_size = 0;
2061 // See writeEhdr for why we do this.
2062 if (Obj.SectionNames != nullptr && Obj.SectionNames->Index >= SHN_LORESERVE)
2063 Shdr.sh_link = Obj.SectionNames->Index;
2064 else
2065 Shdr.sh_link = 0;
2066 Shdr.sh_info = 0;
2067 Shdr.sh_addralign = 0;
2068 Shdr.sh_entsize = 0;
2070 for (SectionBase &Sec : Obj.sections())
2071 writeShdr(Sec);
2074 template <class ELFT> Error ELFWriter<ELFT>::writeSectionData() {
2075 for (SectionBase &Sec : Obj.sections())
2076 // Segments are responsible for writing their contents, so only write the
2077 // section data if the section is not in a segment. Note that this renders
2078 // sections in segments effectively immutable.
2079 if (Sec.ParentSegment == nullptr)
2080 if (Error Err = Sec.accept(*SecWriter))
2081 return Err;
2083 return Error::success();
2086 template <class ELFT> void ELFWriter<ELFT>::writeSegmentData() {
2087 for (Segment &Seg : Obj.segments()) {
2088 size_t Size = std::min<size_t>(Seg.FileSize, Seg.getContents().size());
2089 std::memcpy(Buf->getBufferStart() + Seg.Offset, Seg.getContents().data(),
2090 Size);
2093 for (const auto &it : Obj.getUpdatedSections()) {
2094 SectionBase *Sec = it.first;
2095 ArrayRef<uint8_t> Data = it.second;
2097 auto *Parent = Sec->ParentSegment;
2098 assert(Parent && "This section should've been part of a segment.");
2099 uint64_t Offset =
2100 Sec->OriginalOffset - Parent->OriginalOffset + Parent->Offset;
2101 llvm::copy(Data, Buf->getBufferStart() + Offset);
2104 // Iterate over removed sections and overwrite their old data with zeroes.
2105 for (auto &Sec : Obj.removedSections()) {
2106 Segment *Parent = Sec.ParentSegment;
2107 if (Parent == nullptr || Sec.Type == SHT_NOBITS || Sec.Size == 0)
2108 continue;
2109 uint64_t Offset =
2110 Sec.OriginalOffset - Parent->OriginalOffset + Parent->Offset;
2111 std::memset(Buf->getBufferStart() + Offset, 0, Sec.Size);
2115 template <class ELFT>
2116 ELFWriter<ELFT>::ELFWriter(Object &Obj, raw_ostream &Buf, bool WSH,
2117 bool OnlyKeepDebug)
2118 : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs),
2119 OnlyKeepDebug(OnlyKeepDebug) {}
2121 Error Object::updateSection(StringRef Name, ArrayRef<uint8_t> Data) {
2122 auto It = llvm::find_if(Sections,
2123 [&](const SecPtr &Sec) { return Sec->Name == Name; });
2124 if (It == Sections.end())
2125 return createStringError(errc::invalid_argument, "section '%s' not found",
2126 Name.str().c_str());
2128 auto *OldSec = It->get();
2129 if (!OldSec->hasContents())
2130 return createStringError(
2131 errc::invalid_argument,
2132 "section '%s' cannot be updated because it does not have contents",
2133 Name.str().c_str());
2135 if (Data.size() > OldSec->Size && OldSec->ParentSegment)
2136 return createStringError(errc::invalid_argument,
2137 "cannot fit data of size %zu into section '%s' "
2138 "with size %" PRIu64 " that is part of a segment",
2139 Data.size(), Name.str().c_str(), OldSec->Size);
2141 if (!OldSec->ParentSegment) {
2142 *It = std::make_unique<OwnedDataSection>(*OldSec, Data);
2143 } else {
2144 // The segment writer will be in charge of updating these contents.
2145 OldSec->Size = Data.size();
2146 UpdatedSections[OldSec] = Data;
2149 return Error::success();
2152 Error Object::removeSections(
2153 bool AllowBrokenLinks, std::function<bool(const SectionBase &)> ToRemove) {
2155 auto Iter = std::stable_partition(
2156 std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) {
2157 if (ToRemove(*Sec))
2158 return false;
2159 if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) {
2160 if (auto ToRelSec = RelSec->getSection())
2161 return !ToRemove(*ToRelSec);
2163 return true;
2165 if (SymbolTable != nullptr && ToRemove(*SymbolTable))
2166 SymbolTable = nullptr;
2167 if (SectionNames != nullptr && ToRemove(*SectionNames))
2168 SectionNames = nullptr;
2169 if (SectionIndexTable != nullptr && ToRemove(*SectionIndexTable))
2170 SectionIndexTable = nullptr;
2171 // Now make sure there are no remaining references to the sections that will
2172 // be removed. Sometimes it is impossible to remove a reference so we emit
2173 // an error here instead.
2174 std::unordered_set<const SectionBase *> RemoveSections;
2175 RemoveSections.reserve(std::distance(Iter, std::end(Sections)));
2176 for (auto &RemoveSec : make_range(Iter, std::end(Sections))) {
2177 for (auto &Segment : Segments)
2178 Segment->removeSection(RemoveSec.get());
2179 RemoveSec->onRemove();
2180 RemoveSections.insert(RemoveSec.get());
2183 // For each section that remains alive, we want to remove the dead references.
2184 // This either might update the content of the section (e.g. remove symbols
2185 // from symbol table that belongs to removed section) or trigger an error if
2186 // a live section critically depends on a section being removed somehow
2187 // (e.g. the removed section is referenced by a relocation).
2188 for (auto &KeepSec : make_range(std::begin(Sections), Iter)) {
2189 if (Error E = KeepSec->removeSectionReferences(
2190 AllowBrokenLinks, [&RemoveSections](const SectionBase *Sec) {
2191 return RemoveSections.find(Sec) != RemoveSections.end();
2193 return E;
2196 // Transfer removed sections into the Object RemovedSections container for use
2197 // later.
2198 std::move(Iter, Sections.end(), std::back_inserter(RemovedSections));
2199 // Now finally get rid of them all together.
2200 Sections.erase(Iter, std::end(Sections));
2201 return Error::success();
2204 Error Object::replaceSections(
2205 const DenseMap<SectionBase *, SectionBase *> &FromTo) {
2206 auto SectionIndexLess = [](const SecPtr &Lhs, const SecPtr &Rhs) {
2207 return Lhs->Index < Rhs->Index;
2209 assert(llvm::is_sorted(Sections, SectionIndexLess) &&
2210 "Sections are expected to be sorted by Index");
2211 // Set indices of new sections so that they can be later sorted into positions
2212 // of removed ones.
2213 for (auto &I : FromTo)
2214 I.second->Index = I.first->Index;
2216 // Notify all sections about the replacement.
2217 for (auto &Sec : Sections)
2218 Sec->replaceSectionReferences(FromTo);
2220 if (Error E = removeSections(
2221 /*AllowBrokenLinks=*/false,
2222 [=](const SectionBase &Sec) { return FromTo.count(&Sec) > 0; }))
2223 return E;
2224 llvm::sort(Sections, SectionIndexLess);
2225 return Error::success();
2228 Error Object::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
2229 if (SymbolTable)
2230 for (const SecPtr &Sec : Sections)
2231 if (Error E = Sec->removeSymbols(ToRemove))
2232 return E;
2233 return Error::success();
2236 Error Object::addNewSymbolTable() {
2237 assert(!SymbolTable && "Object must not has a SymbolTable.");
2239 // Reuse an existing SHT_STRTAB section if it exists.
2240 StringTableSection *StrTab = nullptr;
2241 for (SectionBase &Sec : sections()) {
2242 if (Sec.Type == ELF::SHT_STRTAB && !(Sec.Flags & SHF_ALLOC)) {
2243 StrTab = static_cast<StringTableSection *>(&Sec);
2245 // Prefer a string table that is not the section header string table, if
2246 // such a table exists.
2247 if (SectionNames != &Sec)
2248 break;
2251 if (!StrTab)
2252 StrTab = &addSection<StringTableSection>();
2254 SymbolTableSection &SymTab = addSection<SymbolTableSection>();
2255 SymTab.Name = ".symtab";
2256 SymTab.Link = StrTab->Index;
2257 if (Error Err = SymTab.initialize(sections()))
2258 return Err;
2259 SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0);
2261 SymbolTable = &SymTab;
2263 return Error::success();
2266 // Orders segments such that if x = y->ParentSegment then y comes before x.
2267 static void orderSegments(std::vector<Segment *> &Segments) {
2268 llvm::stable_sort(Segments, compareSegmentsByOffset);
2271 // This function finds a consistent layout for a list of segments starting from
2272 // an Offset. It assumes that Segments have been sorted by orderSegments and
2273 // returns an Offset one past the end of the last segment.
2274 static uint64_t layoutSegments(std::vector<Segment *> &Segments,
2275 uint64_t Offset) {
2276 assert(llvm::is_sorted(Segments, compareSegmentsByOffset));
2277 // The only way a segment should move is if a section was between two
2278 // segments and that section was removed. If that section isn't in a segment
2279 // then it's acceptable, but not ideal, to simply move it to after the
2280 // segments. So we can simply layout segments one after the other accounting
2281 // for alignment.
2282 for (Segment *Seg : Segments) {
2283 // We assume that segments have been ordered by OriginalOffset and Index
2284 // such that a parent segment will always come before a child segment in
2285 // OrderedSegments. This means that the Offset of the ParentSegment should
2286 // already be set and we can set our offset relative to it.
2287 if (Seg->ParentSegment != nullptr) {
2288 Segment *Parent = Seg->ParentSegment;
2289 Seg->Offset =
2290 Parent->Offset + Seg->OriginalOffset - Parent->OriginalOffset;
2291 } else {
2292 Seg->Offset =
2293 alignTo(Offset, std::max<uint64_t>(Seg->Align, 1), Seg->VAddr);
2295 Offset = std::max(Offset, Seg->Offset + Seg->FileSize);
2297 return Offset;
2300 // This function finds a consistent layout for a list of sections. It assumes
2301 // that the ->ParentSegment of each section has already been laid out. The
2302 // supplied starting Offset is used for the starting offset of any section that
2303 // does not have a ParentSegment. It returns either the offset given if all
2304 // sections had a ParentSegment or an offset one past the last section if there
2305 // was a section that didn't have a ParentSegment.
2306 template <class Range>
2307 static uint64_t layoutSections(Range Sections, uint64_t Offset) {
2308 // Now the offset of every segment has been set we can assign the offsets
2309 // of each section. For sections that are covered by a segment we should use
2310 // the segment's original offset and the section's original offset to compute
2311 // the offset from the start of the segment. Using the offset from the start
2312 // of the segment we can assign a new offset to the section. For sections not
2313 // covered by segments we can just bump Offset to the next valid location.
2314 // While it is not necessary, layout the sections in the order based on their
2315 // original offsets to resemble the input file as close as possible.
2316 std::vector<SectionBase *> OutOfSegmentSections;
2317 uint32_t Index = 1;
2318 for (auto &Sec : Sections) {
2319 Sec.Index = Index++;
2320 if (Sec.ParentSegment != nullptr) {
2321 const Segment &Segment = *Sec.ParentSegment;
2322 Sec.Offset =
2323 Segment.Offset + (Sec.OriginalOffset - Segment.OriginalOffset);
2324 } else
2325 OutOfSegmentSections.push_back(&Sec);
2328 llvm::stable_sort(OutOfSegmentSections,
2329 [](const SectionBase *Lhs, const SectionBase *Rhs) {
2330 return Lhs->OriginalOffset < Rhs->OriginalOffset;
2332 for (auto *Sec : OutOfSegmentSections) {
2333 Offset = alignTo(Offset, Sec->Align == 0 ? 1 : Sec->Align);
2334 Sec->Offset = Offset;
2335 if (Sec->Type != SHT_NOBITS)
2336 Offset += Sec->Size;
2338 return Offset;
2341 // Rewrite sh_offset after some sections are changed to SHT_NOBITS and thus
2342 // occupy no space in the file.
2343 static uint64_t layoutSectionsForOnlyKeepDebug(Object &Obj, uint64_t Off) {
2344 // The layout algorithm requires the sections to be handled in the order of
2345 // their offsets in the input file, at least inside segments.
2346 std::vector<SectionBase *> Sections;
2347 Sections.reserve(Obj.sections().size());
2348 uint32_t Index = 1;
2349 for (auto &Sec : Obj.sections()) {
2350 Sec.Index = Index++;
2351 Sections.push_back(&Sec);
2353 llvm::stable_sort(Sections,
2354 [](const SectionBase *Lhs, const SectionBase *Rhs) {
2355 return Lhs->OriginalOffset < Rhs->OriginalOffset;
2358 for (auto *Sec : Sections) {
2359 auto *FirstSec = Sec->ParentSegment && Sec->ParentSegment->Type == PT_LOAD
2360 ? Sec->ParentSegment->firstSection()
2361 : nullptr;
2363 // The first section in a PT_LOAD has to have congruent offset and address
2364 // modulo the alignment, which usually equals the maximum page size.
2365 if (FirstSec && FirstSec == Sec)
2366 Off = alignTo(Off, Sec->ParentSegment->Align, Sec->Addr);
2368 // sh_offset is not significant for SHT_NOBITS sections, but the congruence
2369 // rule must be followed if it is the first section in a PT_LOAD. Do not
2370 // advance Off.
2371 if (Sec->Type == SHT_NOBITS) {
2372 Sec->Offset = Off;
2373 continue;
2376 if (!FirstSec) {
2377 // FirstSec being nullptr generally means that Sec does not have the
2378 // SHF_ALLOC flag.
2379 Off = Sec->Align ? alignTo(Off, Sec->Align) : Off;
2380 } else if (FirstSec != Sec) {
2381 // The offset is relative to the first section in the PT_LOAD segment. Use
2382 // sh_offset for non-SHF_ALLOC sections.
2383 Off = Sec->OriginalOffset - FirstSec->OriginalOffset + FirstSec->Offset;
2385 Sec->Offset = Off;
2386 Off += Sec->Size;
2388 return Off;
2391 // Rewrite p_offset and p_filesz of non-PT_PHDR segments after sh_offset values
2392 // have been updated.
2393 static uint64_t layoutSegmentsForOnlyKeepDebug(std::vector<Segment *> &Segments,
2394 uint64_t HdrEnd) {
2395 uint64_t MaxOffset = 0;
2396 for (Segment *Seg : Segments) {
2397 if (Seg->Type == PT_PHDR)
2398 continue;
2400 // The segment offset is generally the offset of the first section.
2402 // For a segment containing no section (see sectionWithinSegment), if it has
2403 // a parent segment, copy the parent segment's offset field. This works for
2404 // empty PT_TLS. If no parent segment, use 0: the segment is not useful for
2405 // debugging anyway.
2406 const SectionBase *FirstSec = Seg->firstSection();
2407 uint64_t Offset =
2408 FirstSec ? FirstSec->Offset
2409 : (Seg->ParentSegment ? Seg->ParentSegment->Offset : 0);
2410 uint64_t FileSize = 0;
2411 for (const SectionBase *Sec : Seg->Sections) {
2412 uint64_t Size = Sec->Type == SHT_NOBITS ? 0 : Sec->Size;
2413 if (Sec->Offset + Size > Offset)
2414 FileSize = std::max(FileSize, Sec->Offset + Size - Offset);
2417 // If the segment includes EHDR and program headers, don't make it smaller
2418 // than the headers.
2419 if (Seg->Offset < HdrEnd && HdrEnd <= Seg->Offset + Seg->FileSize) {
2420 FileSize += Offset - Seg->Offset;
2421 Offset = Seg->Offset;
2422 FileSize = std::max(FileSize, HdrEnd - Offset);
2425 Seg->Offset = Offset;
2426 Seg->FileSize = FileSize;
2427 MaxOffset = std::max(MaxOffset, Offset + FileSize);
2429 return MaxOffset;
2432 template <class ELFT> void ELFWriter<ELFT>::initEhdrSegment() {
2433 Segment &ElfHdr = Obj.ElfHdrSegment;
2434 ElfHdr.Type = PT_PHDR;
2435 ElfHdr.Flags = 0;
2436 ElfHdr.VAddr = 0;
2437 ElfHdr.PAddr = 0;
2438 ElfHdr.FileSize = ElfHdr.MemSize = sizeof(Elf_Ehdr);
2439 ElfHdr.Align = 0;
2442 template <class ELFT> void ELFWriter<ELFT>::assignOffsets() {
2443 // We need a temporary list of segments that has a special order to it
2444 // so that we know that anytime ->ParentSegment is set that segment has
2445 // already had its offset properly set.
2446 std::vector<Segment *> OrderedSegments;
2447 for (Segment &Segment : Obj.segments())
2448 OrderedSegments.push_back(&Segment);
2449 OrderedSegments.push_back(&Obj.ElfHdrSegment);
2450 OrderedSegments.push_back(&Obj.ProgramHdrSegment);
2451 orderSegments(OrderedSegments);
2453 uint64_t Offset;
2454 if (OnlyKeepDebug) {
2455 // For --only-keep-debug, the sections that did not preserve contents were
2456 // changed to SHT_NOBITS. We now rewrite sh_offset fields of sections, and
2457 // then rewrite p_offset/p_filesz of program headers.
2458 uint64_t HdrEnd =
2459 sizeof(Elf_Ehdr) + llvm::size(Obj.segments()) * sizeof(Elf_Phdr);
2460 Offset = layoutSectionsForOnlyKeepDebug(Obj, HdrEnd);
2461 Offset = std::max(Offset,
2462 layoutSegmentsForOnlyKeepDebug(OrderedSegments, HdrEnd));
2463 } else {
2464 // Offset is used as the start offset of the first segment to be laid out.
2465 // Since the ELF Header (ElfHdrSegment) must be at the start of the file,
2466 // we start at offset 0.
2467 Offset = layoutSegments(OrderedSegments, 0);
2468 Offset = layoutSections(Obj.sections(), Offset);
2470 // If we need to write the section header table out then we need to align the
2471 // Offset so that SHOffset is valid.
2472 if (WriteSectionHeaders)
2473 Offset = alignTo(Offset, sizeof(Elf_Addr));
2474 Obj.SHOff = Offset;
2477 template <class ELFT> size_t ELFWriter<ELFT>::totalSize() const {
2478 // We already have the section header offset so we can calculate the total
2479 // size by just adding up the size of each section header.
2480 if (!WriteSectionHeaders)
2481 return Obj.SHOff;
2482 size_t ShdrCount = Obj.sections().size() + 1; // Includes null shdr.
2483 return Obj.SHOff + ShdrCount * sizeof(Elf_Shdr);
2486 template <class ELFT> Error ELFWriter<ELFT>::write() {
2487 // Segment data must be written first, so that the ELF header and program
2488 // header tables can overwrite it, if covered by a segment.
2489 writeSegmentData();
2490 writeEhdr();
2491 writePhdrs();
2492 if (Error E = writeSectionData())
2493 return E;
2494 if (WriteSectionHeaders)
2495 writeShdrs();
2497 // TODO: Implement direct writing to the output stream (without intermediate
2498 // memory buffer Buf).
2499 Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2500 return Error::success();
2503 static Error removeUnneededSections(Object &Obj) {
2504 // We can remove an empty symbol table from non-relocatable objects.
2505 // Relocatable objects typically have relocation sections whose
2506 // sh_link field points to .symtab, so we can't remove .symtab
2507 // even if it is empty.
2508 if (Obj.isRelocatable() || Obj.SymbolTable == nullptr ||
2509 !Obj.SymbolTable->empty())
2510 return Error::success();
2512 // .strtab can be used for section names. In such a case we shouldn't
2513 // remove it.
2514 auto *StrTab = Obj.SymbolTable->getStrTab() == Obj.SectionNames
2515 ? nullptr
2516 : Obj.SymbolTable->getStrTab();
2517 return Obj.removeSections(false, [&](const SectionBase &Sec) {
2518 return &Sec == Obj.SymbolTable || &Sec == StrTab;
2522 template <class ELFT> Error ELFWriter<ELFT>::finalize() {
2523 // It could happen that SectionNames has been removed and yet the user wants
2524 // a section header table output. We need to throw an error if a user tries
2525 // to do that.
2526 if (Obj.SectionNames == nullptr && WriteSectionHeaders)
2527 return createStringError(llvm::errc::invalid_argument,
2528 "cannot write section header table because "
2529 "section header string table was removed");
2531 if (Error E = removeUnneededSections(Obj))
2532 return E;
2534 // If the .symtab indices have not been changed, restore the sh_link to
2535 // .symtab for sections that were linked to .symtab.
2536 if (Obj.SymbolTable && !Obj.SymbolTable->indicesChanged())
2537 for (SectionBase &Sec : Obj.sections())
2538 Sec.restoreSymTabLink(*Obj.SymbolTable);
2540 // We need to assign indexes before we perform layout because we need to know
2541 // if we need large indexes or not. We can assign indexes first and check as
2542 // we go to see if we will actully need large indexes.
2543 bool NeedsLargeIndexes = false;
2544 if (Obj.sections().size() >= SHN_LORESERVE) {
2545 SectionTableRef Sections = Obj.sections();
2546 // Sections doesn't include the null section header, so account for this
2547 // when skipping the first N sections.
2548 NeedsLargeIndexes =
2549 any_of(drop_begin(Sections, SHN_LORESERVE - 1),
2550 [](const SectionBase &Sec) { return Sec.HasSymbol; });
2551 // TODO: handle case where only one section needs the large index table but
2552 // only needs it because the large index table hasn't been removed yet.
2555 if (NeedsLargeIndexes) {
2556 // This means we definitely need to have a section index table but if we
2557 // already have one then we should use it instead of making a new one.
2558 if (Obj.SymbolTable != nullptr && Obj.SectionIndexTable == nullptr) {
2559 // Addition of a section to the end does not invalidate the indexes of
2560 // other sections and assigns the correct index to the new section.
2561 auto &Shndx = Obj.addSection<SectionIndexSection>();
2562 Obj.SymbolTable->setShndxTable(&Shndx);
2563 Shndx.setSymTab(Obj.SymbolTable);
2565 } else {
2566 // Since we don't need SectionIndexTable we should remove it and all
2567 // references to it.
2568 if (Obj.SectionIndexTable != nullptr) {
2569 // We do not support sections referring to the section index table.
2570 if (Error E = Obj.removeSections(false /*AllowBrokenLinks*/,
2571 [this](const SectionBase &Sec) {
2572 return &Sec == Obj.SectionIndexTable;
2574 return E;
2578 // Make sure we add the names of all the sections. Importantly this must be
2579 // done after we decide to add or remove SectionIndexes.
2580 if (Obj.SectionNames != nullptr)
2581 for (const SectionBase &Sec : Obj.sections())
2582 Obj.SectionNames->addString(Sec.Name);
2584 initEhdrSegment();
2586 // Before we can prepare for layout the indexes need to be finalized.
2587 // Also, the output arch may not be the same as the input arch, so fix up
2588 // size-related fields before doing layout calculations.
2589 uint64_t Index = 0;
2590 auto SecSizer = std::make_unique<ELFSectionSizer<ELFT>>();
2591 for (SectionBase &Sec : Obj.sections()) {
2592 Sec.Index = Index++;
2593 if (Error Err = Sec.accept(*SecSizer))
2594 return Err;
2597 // The symbol table does not update all other sections on update. For
2598 // instance, symbol names are not added as new symbols are added. This means
2599 // that some sections, like .strtab, don't yet have their final size.
2600 if (Obj.SymbolTable != nullptr)
2601 Obj.SymbolTable->prepareForLayout();
2603 // Now that all strings are added we want to finalize string table builders,
2604 // because that affects section sizes which in turn affects section offsets.
2605 for (SectionBase &Sec : Obj.sections())
2606 if (auto StrTab = dyn_cast<StringTableSection>(&Sec))
2607 StrTab->prepareForLayout();
2609 assignOffsets();
2611 // layoutSections could have modified section indexes, so we need
2612 // to fill the index table after assignOffsets.
2613 if (Obj.SymbolTable != nullptr)
2614 Obj.SymbolTable->fillShndxTable();
2616 // Finally now that all offsets and indexes have been set we can finalize any
2617 // remaining issues.
2618 uint64_t Offset = Obj.SHOff + sizeof(Elf_Shdr);
2619 for (SectionBase &Sec : Obj.sections()) {
2620 Sec.HeaderOffset = Offset;
2621 Offset += sizeof(Elf_Shdr);
2622 if (WriteSectionHeaders)
2623 Sec.NameIndex = Obj.SectionNames->findIndex(Sec.Name);
2624 Sec.finalize();
2627 size_t TotalSize = totalSize();
2628 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2629 if (!Buf)
2630 return createStringError(errc::not_enough_memory,
2631 "failed to allocate memory buffer of " +
2632 Twine::utohexstr(TotalSize) + " bytes");
2634 SecWriter = std::make_unique<ELFSectionWriter<ELFT>>(*Buf);
2635 return Error::success();
2638 Error BinaryWriter::write() {
2639 SmallVector<const SectionBase *, 30> SectionsToWrite;
2640 for (const SectionBase &Sec : Obj.allocSections()) {
2641 if (Sec.Type != SHT_NOBITS && Sec.Size > 0)
2642 SectionsToWrite.push_back(&Sec);
2645 if (SectionsToWrite.empty())
2646 return Error::success();
2648 llvm::stable_sort(SectionsToWrite,
2649 [](const SectionBase *LHS, const SectionBase *RHS) {
2650 return LHS->Offset < RHS->Offset;
2653 assert(SectionsToWrite.front()->Offset == 0);
2655 for (size_t i = 0; i != SectionsToWrite.size(); ++i) {
2656 const SectionBase &Sec = *SectionsToWrite[i];
2657 if (Error Err = Sec.accept(*SecWriter))
2658 return Err;
2659 if (GapFill == 0)
2660 continue;
2661 uint64_t PadOffset = (i < SectionsToWrite.size() - 1)
2662 ? SectionsToWrite[i + 1]->Offset
2663 : Buf->getBufferSize();
2664 assert(PadOffset <= Buf->getBufferSize());
2665 assert(Sec.Offset + Sec.Size <= PadOffset);
2666 std::fill(Buf->getBufferStart() + Sec.Offset + Sec.Size,
2667 Buf->getBufferStart() + PadOffset, GapFill);
2670 // TODO: Implement direct writing to the output stream (without intermediate
2671 // memory buffer Buf).
2672 Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2673 return Error::success();
2676 Error BinaryWriter::finalize() {
2677 // Compute the section LMA based on its sh_offset and the containing segment's
2678 // p_offset and p_paddr. Also compute the minimum LMA of all non-empty
2679 // sections as MinAddr. In the output, the contents between address 0 and
2680 // MinAddr will be skipped.
2681 uint64_t MinAddr = UINT64_MAX;
2682 for (SectionBase &Sec : Obj.allocSections()) {
2683 if (Sec.ParentSegment != nullptr)
2684 Sec.Addr =
2685 Sec.Offset - Sec.ParentSegment->Offset + Sec.ParentSegment->PAddr;
2686 if (Sec.Type != SHT_NOBITS && Sec.Size > 0)
2687 MinAddr = std::min(MinAddr, Sec.Addr);
2690 // Now that every section has been laid out we just need to compute the total
2691 // file size. This might not be the same as the offset returned by
2692 // layoutSections, because we want to truncate the last segment to the end of
2693 // its last non-empty section, to match GNU objcopy's behaviour.
2694 TotalSize = PadTo > MinAddr ? PadTo - MinAddr : 0;
2695 for (SectionBase &Sec : Obj.allocSections())
2696 if (Sec.Type != SHT_NOBITS && Sec.Size > 0) {
2697 Sec.Offset = Sec.Addr - MinAddr;
2698 TotalSize = std::max(TotalSize, Sec.Offset + Sec.Size);
2701 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2702 if (!Buf)
2703 return createStringError(errc::not_enough_memory,
2704 "failed to allocate memory buffer of " +
2705 Twine::utohexstr(TotalSize) + " bytes");
2706 SecWriter = std::make_unique<BinarySectionWriter>(*Buf);
2707 return Error::success();
2710 bool IHexWriter::SectionCompare::operator()(const SectionBase *Lhs,
2711 const SectionBase *Rhs) const {
2712 return (sectionPhysicalAddr(Lhs) & 0xFFFFFFFFU) <
2713 (sectionPhysicalAddr(Rhs) & 0xFFFFFFFFU);
2716 uint64_t IHexWriter::writeEntryPointRecord(uint8_t *Buf) {
2717 IHexLineData HexData;
2718 uint8_t Data[4] = {};
2719 // We don't write entry point record if entry is zero.
2720 if (Obj.Entry == 0)
2721 return 0;
2723 if (Obj.Entry <= 0xFFFFFU) {
2724 Data[0] = ((Obj.Entry & 0xF0000U) >> 12) & 0xFF;
2725 support::endian::write(&Data[2], static_cast<uint16_t>(Obj.Entry),
2726 llvm::endianness::big);
2727 HexData = IHexRecord::getLine(IHexRecord::StartAddr80x86, 0, Data);
2728 } else {
2729 support::endian::write(Data, static_cast<uint32_t>(Obj.Entry),
2730 llvm::endianness::big);
2731 HexData = IHexRecord::getLine(IHexRecord::StartAddr, 0, Data);
2733 memcpy(Buf, HexData.data(), HexData.size());
2734 return HexData.size();
2737 uint64_t IHexWriter::writeEndOfFileRecord(uint8_t *Buf) {
2738 IHexLineData HexData = IHexRecord::getLine(IHexRecord::EndOfFile, 0, {});
2739 memcpy(Buf, HexData.data(), HexData.size());
2740 return HexData.size();
2743 Error IHexWriter::write() {
2744 IHexSectionWriter Writer(*Buf);
2745 // Write sections.
2746 for (const SectionBase *Sec : Sections)
2747 if (Error Err = Sec->accept(Writer))
2748 return Err;
2750 uint64_t Offset = Writer.getBufferOffset();
2751 // Write entry point address.
2752 Offset += writeEntryPointRecord(
2753 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset);
2754 // Write EOF.
2755 Offset += writeEndOfFileRecord(
2756 reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset);
2757 assert(Offset == TotalSize);
2759 // TODO: Implement direct writing to the output stream (without intermediate
2760 // memory buffer Buf).
2761 Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2762 return Error::success();
2765 Error IHexWriter::checkSection(const SectionBase &Sec) {
2766 uint64_t Addr = sectionPhysicalAddr(&Sec);
2767 if (addressOverflows32bit(Addr) || addressOverflows32bit(Addr + Sec.Size - 1))
2768 return createStringError(
2769 errc::invalid_argument,
2770 "Section '%s' address range [0x%llx, 0x%llx] is not 32 bit",
2771 Sec.Name.c_str(), Addr, Addr + Sec.Size - 1);
2772 return Error::success();
2775 Error IHexWriter::finalize() {
2776 // We can't write 64-bit addresses.
2777 if (addressOverflows32bit(Obj.Entry))
2778 return createStringError(errc::invalid_argument,
2779 "Entry point address 0x%llx overflows 32 bits",
2780 Obj.Entry);
2782 for (const SectionBase &Sec : Obj.sections())
2783 if ((Sec.Flags & ELF::SHF_ALLOC) && Sec.Type != ELF::SHT_NOBITS &&
2784 Sec.Size > 0) {
2785 if (Error E = checkSection(Sec))
2786 return E;
2787 Sections.insert(&Sec);
2790 std::unique_ptr<WritableMemoryBuffer> EmptyBuffer =
2791 WritableMemoryBuffer::getNewMemBuffer(0);
2792 if (!EmptyBuffer)
2793 return createStringError(errc::not_enough_memory,
2794 "failed to allocate memory buffer of 0 bytes");
2796 IHexSectionWriterBase LengthCalc(*EmptyBuffer);
2797 for (const SectionBase *Sec : Sections)
2798 if (Error Err = Sec->accept(LengthCalc))
2799 return Err;
2801 // We need space to write section records + StartAddress record
2802 // (if start adress is not zero) + EndOfFile record.
2803 TotalSize = LengthCalc.getBufferOffset() +
2804 (Obj.Entry ? IHexRecord::getLineLength(4) : 0) +
2805 IHexRecord::getLineLength(0);
2807 Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2808 if (!Buf)
2809 return createStringError(errc::not_enough_memory,
2810 "failed to allocate memory buffer of " +
2811 Twine::utohexstr(TotalSize) + " bytes");
2813 return Error::success();
2816 namespace llvm {
2817 namespace objcopy {
2818 namespace elf {
2820 template class ELFBuilder<ELF64LE>;
2821 template class ELFBuilder<ELF64BE>;
2822 template class ELFBuilder<ELF32LE>;
2823 template class ELFBuilder<ELF32BE>;
2825 template class ELFWriter<ELF64LE>;
2826 template class ELFWriter<ELF64BE>;
2827 template class ELFWriter<ELF32LE>;
2828 template class ELFWriter<ELF32BE>;
2830 } // end namespace elf
2831 } // end namespace objcopy
2832 } // end namespace llvm