1 //===- ELFObject.cpp ------------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "ELFObject.h"
10 #include "llvm/ADT/ArrayRef.h"
11 #include "llvm/ADT/STLExtras.h"
12 #include "llvm/ADT/StringRef.h"
13 #include "llvm/ADT/Twine.h"
14 #include "llvm/ADT/iterator_range.h"
15 #include "llvm/BinaryFormat/ELF.h"
16 #include "llvm/MC/MCTargetOptions.h"
17 #include "llvm/Object/ELF.h"
18 #include "llvm/Object/ELFObjectFile.h"
19 #include "llvm/Support/Compression.h"
20 #include "llvm/Support/Endian.h"
21 #include "llvm/Support/ErrorHandling.h"
22 #include "llvm/Support/FileOutputBuffer.h"
23 #include "llvm/Support/Path.h"
24 #include <algorithm>
25 #include <cstddef>
26 #include <cstdint>
27 #include <iterator>
28 #include <unordered_set>
29 #include <utility>
30 #include <vector>
31 
32 using namespace llvm;
33 using namespace llvm::ELF;
34 using namespace llvm::objcopy::elf;
35 using namespace llvm::object;
36 
37 template <class ELFT> void ELFWriter<ELFT>::writePhdr(const Segment &Seg) {
38   uint8_t *B = reinterpret_cast<uint8_t *>(Buf->getBufferStart()) +
39                Obj.ProgramHdrSegment.Offset + Seg.Index * sizeof(Elf_Phdr);
40   Elf_Phdr &Phdr = *reinterpret_cast<Elf_Phdr *>(B);
41   Phdr.p_type = Seg.Type;
42   Phdr.p_flags = Seg.Flags;
43   Phdr.p_offset = Seg.Offset;
44   Phdr.p_vaddr = Seg.VAddr;
45   Phdr.p_paddr = Seg.PAddr;
46   Phdr.p_filesz = Seg.FileSize;
47   Phdr.p_memsz = Seg.MemSize;
48   Phdr.p_align = Seg.Align;
49 }
50 
51 Error SectionBase::removeSectionReferences(
52     bool, function_ref<bool(const SectionBase *)>) {
53   return Error::success();
54 }
55 
56 Error SectionBase::removeSymbols(function_ref<bool(const Symbol &)>) {
57   return Error::success();
58 }
59 
60 Error SectionBase::initialize(SectionTableRef) { return Error::success(); }
61 void SectionBase::finalize() {}
62 void SectionBase::markSymbols() {}
63 void SectionBase::replaceSectionReferences(
64     const DenseMap<SectionBase *, SectionBase *> &) {}
65 void SectionBase::onRemove() {}
66 
67 template <class ELFT> void ELFWriter<ELFT>::writeShdr(const SectionBase &Sec) {
68   uint8_t *B =
69       reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Sec.HeaderOffset;
70   Elf_Shdr &Shdr = *reinterpret_cast<Elf_Shdr *>(B);
71   Shdr.sh_name = Sec.NameIndex;
72   Shdr.sh_type = Sec.Type;
73   Shdr.sh_flags = Sec.Flags;
74   Shdr.sh_addr = Sec.Addr;
75   Shdr.sh_offset = Sec.Offset;
76   Shdr.sh_size = Sec.Size;
77   Shdr.sh_link = Sec.Link;
78   Shdr.sh_info = Sec.Info;
79   Shdr.sh_addralign = Sec.Align;
80   Shdr.sh_entsize = Sec.EntrySize;
81 }
82 
83 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(Section &) {
84   return Error::success();
85 }
86 
87 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(OwnedDataSection &) {
88   return Error::success();
89 }
90 
91 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(StringTableSection &) {
92   return Error::success();
93 }
94 
95 template <class ELFT>
96 Error ELFSectionSizer<ELFT>::visit(DynamicRelocationSection &) {
97   return Error::success();
98 }
99 
100 template <class ELFT>
101 Error ELFSectionSizer<ELFT>::visit(SymbolTableSection &Sec) {
102   Sec.EntrySize = sizeof(Elf_Sym);
103   Sec.Size = Sec.Symbols.size() * Sec.EntrySize;
104   // Align to the largest field in Elf_Sym.
105   Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word);
106   return Error::success();
107 }
108 
109 template <class ELFT>
110 Error ELFSectionSizer<ELFT>::visit(RelocationSection &Sec) {
111   Sec.EntrySize = Sec.Type == SHT_REL ? sizeof(Elf_Rel) : sizeof(Elf_Rela);
112   Sec.Size = Sec.Relocations.size() * Sec.EntrySize;
113   // Align to the largest field in Elf_Rel(a).
114   Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word);
115   return Error::success();
116 }
117 
118 template <class ELFT>
119 Error ELFSectionSizer<ELFT>::visit(GnuDebugLinkSection &) {
120   return Error::success();
121 }
122 
123 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(GroupSection &Sec) {
124   Sec.Size = sizeof(Elf_Word) + Sec.GroupMembers.size() * sizeof(Elf_Word);
125   return Error::success();
126 }
127 
128 template <class ELFT>
129 Error ELFSectionSizer<ELFT>::visit(SectionIndexSection &) {
130   return Error::success();
131 }
132 
133 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(CompressedSection &) {
134   return Error::success();
135 }
136 
137 template <class ELFT>
138 Error ELFSectionSizer<ELFT>::visit(DecompressedSection &) {
139   return Error::success();
140 }
141 
142 Error BinarySectionWriter::visit(const SectionIndexSection &Sec) {
143   return createStringError(errc::operation_not_permitted,
144                            "cannot write symbol section index table '" +
145                                Sec.Name + "' ");
146 }
147 
148 Error BinarySectionWriter::visit(const SymbolTableSection &Sec) {
149   return createStringError(errc::operation_not_permitted,
150                            "cannot write symbol table '" + Sec.Name +
151                                "' out to binary");
152 }
153 
154 Error BinarySectionWriter::visit(const RelocationSection &Sec) {
155   return createStringError(errc::operation_not_permitted,
156                            "cannot write relocation section '" + Sec.Name +
157                                "' out to binary");
158 }
159 
160 Error BinarySectionWriter::visit(const GnuDebugLinkSection &Sec) {
161   return createStringError(errc::operation_not_permitted,
162                            "cannot write '" + Sec.Name + "' out to binary");
163 }
164 
165 Error BinarySectionWriter::visit(const GroupSection &Sec) {
166   return createStringError(errc::operation_not_permitted,
167                            "cannot write '" + Sec.Name + "' out to binary");
168 }
169 
170 Error SectionWriter::visit(const Section &Sec) {
171   if (Sec.Type != SHT_NOBITS)
172     llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset);
173 
174   return Error::success();
175 }
176 
177 static bool addressOverflows32bit(uint64_t Addr) {
178   // Sign extended 32 bit addresses (e.g 0xFFFFFFFF80000000) are ok
179   return Addr > UINT32_MAX && Addr + 0x80000000 > UINT32_MAX;
180 }
181 
182 template <class T> static T checkedGetHex(StringRef S) {
183   T Value;
184   bool Fail = S.getAsInteger(16, Value);
185   assert(!Fail);
186   (void)Fail;
187   return Value;
188 }
189 
190 // Fills exactly Len bytes of buffer with hexadecimal characters
191 // representing value 'X'
192 template <class T, class Iterator>
193 static Iterator toHexStr(T X, Iterator It, size_t Len) {
194   // Fill range with '0'
195   std::fill(It, It + Len, '0');
196 
197   for (long I = Len - 1; I >= 0; --I) {
198     unsigned char Mod = static_cast<unsigned char>(X) & 15;
199     *(It + I) = hexdigit(Mod, false);
200     X >>= 4;
201   }
202   assert(X == 0);
203   return It + Len;
204 }
205 
206 uint8_t IHexRecord::getChecksum(StringRef S) {
207   assert((S.size() & 1) == 0);
208   uint8_t Checksum = 0;
209   while (!S.empty()) {
210     Checksum += checkedGetHex<uint8_t>(S.take_front(2));
211     S = S.drop_front(2);
212   }
213   return -Checksum;
214 }
215 
216 IHexLineData IHexRecord::getLine(uint8_t Type, uint16_t Addr,
217                                  ArrayRef<uint8_t> Data) {
218   IHexLineData Line(getLineLength(Data.size()));
219   assert(Line.size());
220   auto Iter = Line.begin();
221   *Iter++ = ':';
222   Iter = toHexStr(Data.size(), Iter, 2);
223   Iter = toHexStr(Addr, Iter, 4);
224   Iter = toHexStr(Type, Iter, 2);
225   for (uint8_t X : Data)
226     Iter = toHexStr(X, Iter, 2);
227   StringRef S(Line.data() + 1, std::distance(Line.begin() + 1, Iter));
228   Iter = toHexStr(getChecksum(S), Iter, 2);
229   *Iter++ = '\r';
230   *Iter++ = '\n';
231   assert(Iter == Line.end());
232   return Line;
233 }
234 
235 static Error checkRecord(const IHexRecord &R) {
236   switch (R.Type) {
237   case IHexRecord::Data:
238     if (R.HexData.size() == 0)
239       return createStringError(
240           errc::invalid_argument,
241           "zero data length is not allowed for data records");
242     break;
243   case IHexRecord::EndOfFile:
244     break;
245   case IHexRecord::SegmentAddr:
246     // 20-bit segment address. Data length must be 2 bytes
247     // (4 bytes in hex)
248     if (R.HexData.size() != 4)
249       return createStringError(
250           errc::invalid_argument,
251           "segment address data should be 2 bytes in size");
252     break;
253   case IHexRecord::StartAddr80x86:
254   case IHexRecord::StartAddr:
255     if (R.HexData.size() != 8)
256       return createStringError(errc::invalid_argument,
257                                "start address data should be 4 bytes in size");
258     // According to Intel HEX specification '03' record
259     // only specifies the code address within the 20-bit
260     // segmented address space of the 8086/80186. This
261     // means 12 high order bits should be zeroes.
262     if (R.Type == IHexRecord::StartAddr80x86 &&
263         R.HexData.take_front(3) != "000")
264       return createStringError(errc::invalid_argument,
265                                "start address exceeds 20 bit for 80x86");
266     break;
267   case IHexRecord::ExtendedAddr:
268     // 16-31 bits of linear base address
269     if (R.HexData.size() != 4)
270       return createStringError(
271           errc::invalid_argument,
272           "extended address data should be 2 bytes in size");
273     break;
274   default:
275     // Unknown record type
276     return createStringError(errc::invalid_argument, "unknown record type: %u",
277                              static_cast<unsigned>(R.Type));
278   }
279   return Error::success();
280 }
281 
282 // Checks that IHEX line contains valid characters.
283 // This allows converting hexadecimal data to integers
284 // without extra verification.
285 static Error checkChars(StringRef Line) {
286   assert(!Line.empty());
287   if (Line[0] != ':')
288     return createStringError(errc::invalid_argument,
289                              "missing ':' in the beginning of line.");
290 
291   for (size_t Pos = 1; Pos < Line.size(); ++Pos)
292     if (hexDigitValue(Line[Pos]) == -1U)
293       return createStringError(errc::invalid_argument,
294                                "invalid character at position %zu.", Pos + 1);
295   return Error::success();
296 }
297 
298 Expected<IHexRecord> IHexRecord::parse(StringRef Line) {
299   assert(!Line.empty());
300 
301   // ':' + Length + Address + Type + Checksum with empty data ':LLAAAATTCC'
302   if (Line.size() < 11)
303     return createStringError(errc::invalid_argument,
304                              "line is too short: %zu chars.", Line.size());
305 
306   if (Error E = checkChars(Line))
307     return std::move(E);
308 
309   IHexRecord Rec;
310   size_t DataLen = checkedGetHex<uint8_t>(Line.substr(1, 2));
311   if (Line.size() != getLength(DataLen))
312     return createStringError(errc::invalid_argument,
313                              "invalid line length %zu (should be %zu)",
314                              Line.size(), getLength(DataLen));
315 
316   Rec.Addr = checkedGetHex<uint16_t>(Line.substr(3, 4));
317   Rec.Type = checkedGetHex<uint8_t>(Line.substr(7, 2));
318   Rec.HexData = Line.substr(9, DataLen * 2);
319 
320   if (getChecksum(Line.drop_front(1)) != 0)
321     return createStringError(errc::invalid_argument, "incorrect checksum.");
322   if (Error E = checkRecord(Rec))
323     return std::move(E);
324   return Rec;
325 }
326 
327 static uint64_t sectionPhysicalAddr(const SectionBase *Sec) {
328   Segment *Seg = Sec->ParentSegment;
329   if (Seg && Seg->Type != ELF::PT_LOAD)
330     Seg = nullptr;
331   return Seg ? Seg->PAddr + Sec->OriginalOffset - Seg->OriginalOffset
332              : Sec->Addr;
333 }
334 
335 void IHexSectionWriterBase::writeSection(const SectionBase *Sec,
336                                          ArrayRef<uint8_t> Data) {
337   assert(Data.size() == Sec->Size);
338   const uint32_t ChunkSize = 16;
339   uint32_t Addr = sectionPhysicalAddr(Sec) & 0xFFFFFFFFU;
340   while (!Data.empty()) {
341     uint64_t DataSize = std::min<uint64_t>(Data.size(), ChunkSize);
342     if (Addr > SegmentAddr + BaseAddr + 0xFFFFU) {
343       if (Addr > 0xFFFFFU) {
344         // Write extended address record, zeroing segment address
345         // if needed.
346         if (SegmentAddr != 0)
347           SegmentAddr = writeSegmentAddr(0U);
348         BaseAddr = writeBaseAddr(Addr);
349       } else {
350         // We can still remain 16-bit
351         SegmentAddr = writeSegmentAddr(Addr);
352       }
353     }
354     uint64_t SegOffset = Addr - BaseAddr - SegmentAddr;
355     assert(SegOffset <= 0xFFFFU);
356     DataSize = std::min(DataSize, 0x10000U - SegOffset);
357     writeData(0, SegOffset, Data.take_front(DataSize));
358     Addr += DataSize;
359     Data = Data.drop_front(DataSize);
360   }
361 }
362 
363 uint64_t IHexSectionWriterBase::writeSegmentAddr(uint64_t Addr) {
364   assert(Addr <= 0xFFFFFU);
365   uint8_t Data[] = {static_cast<uint8_t>((Addr & 0xF0000U) >> 12), 0};
366   writeData(2, 0, Data);
367   return Addr & 0xF0000U;
368 }
369 
370 uint64_t IHexSectionWriterBase::writeBaseAddr(uint64_t Addr) {
371   assert(Addr <= 0xFFFFFFFFU);
372   uint64_t Base = Addr & 0xFFFF0000U;
373   uint8_t Data[] = {static_cast<uint8_t>(Base >> 24),
374                     static_cast<uint8_t>((Base >> 16) & 0xFF)};
375   writeData(4, 0, Data);
376   return Base;
377 }
378 
379 void IHexSectionWriterBase::writeData(uint8_t, uint16_t,
380                                       ArrayRef<uint8_t> Data) {
381   Offset += IHexRecord::getLineLength(Data.size());
382 }
383 
384 Error IHexSectionWriterBase::visit(const Section &Sec) {
385   writeSection(&Sec, Sec.Contents);
386   return Error::success();
387 }
388 
389 Error IHexSectionWriterBase::visit(const OwnedDataSection &Sec) {
390   writeSection(&Sec, Sec.Data);
391   return Error::success();
392 }
393 
394 Error IHexSectionWriterBase::visit(const StringTableSection &Sec) {
395   // Check that sizer has already done its work
396   assert(Sec.Size == Sec.StrTabBuilder.getSize());
397   // We are free to pass an invalid pointer to writeSection as long
398   // as we don't actually write any data. The real writer class has
399   // to override this method .
400   writeSection(&Sec, {nullptr, static_cast<size_t>(Sec.Size)});
401   return Error::success();
402 }
403 
404 Error IHexSectionWriterBase::visit(const DynamicRelocationSection &Sec) {
405   writeSection(&Sec, Sec.Contents);
406   return Error::success();
407 }
408 
409 void IHexSectionWriter::writeData(uint8_t Type, uint16_t Addr,
410                                   ArrayRef<uint8_t> Data) {
411   IHexLineData HexData = IHexRecord::getLine(Type, Addr, Data);
412   memcpy(Out.getBufferStart() + Offset, HexData.data(), HexData.size());
413   Offset += HexData.size();
414 }
415 
416 Error IHexSectionWriter::visit(const StringTableSection &Sec) {
417   assert(Sec.Size == Sec.StrTabBuilder.getSize());
418   std::vector<uint8_t> Data(Sec.Size);
419   Sec.StrTabBuilder.write(Data.data());
420   writeSection(&Sec, Data);
421   return Error::success();
422 }
423 
424 Error Section::accept(SectionVisitor &Visitor) const {
425   return Visitor.visit(*this);
426 }
427 
428 Error Section::accept(MutableSectionVisitor &Visitor) {
429   return Visitor.visit(*this);
430 }
431 
432 Error SectionWriter::visit(const OwnedDataSection &Sec) {
433   llvm::copy(Sec.Data, Out.getBufferStart() + Sec.Offset);
434   return Error::success();
435 }
436 
437 template <class ELFT>
438 Error ELFSectionWriter<ELFT>::visit(const DecompressedSection &Sec) {
439   ArrayRef<uint8_t> Compressed =
440       Sec.OriginalData.slice(sizeof(Elf_Chdr_Impl<ELFT>));
441   SmallVector<uint8_t, 128> Decompressed;
442   DebugCompressionType Type;
443   switch (Sec.ChType) {
444   case ELFCOMPRESS_ZLIB:
445     Type = DebugCompressionType::Zlib;
446     break;
447   case ELFCOMPRESS_ZSTD:
448     Type = DebugCompressionType::Zstd;
449     break;
450   default:
451     return createStringError(errc::invalid_argument,
452                              "--decompress-debug-sections: ch_type (" +
453                                  Twine(Sec.ChType) + ") of section '" +
454                                  Sec.Name + "' is unsupported");
455   }
456   if (auto *Reason =
457           compression::getReasonIfUnsupported(compression::formatFor(Type)))
458     return createStringError(errc::invalid_argument,
459                              "failed to decompress section '" + Sec.Name +
460                                  "': " + Reason);
461   if (Error E = compression::decompress(Type, Compressed, Decompressed,
462                                         static_cast<size_t>(Sec.Size)))
463     return createStringError(errc::invalid_argument,
464                              "failed to decompress section '" + Sec.Name +
465                                  "': " + toString(std::move(E)));
466 
467   uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
468   std::copy(Decompressed.begin(), Decompressed.end(), Buf);
469 
470   return Error::success();
471 }
472 
473 Error BinarySectionWriter::visit(const DecompressedSection &Sec) {
474   return createStringError(errc::operation_not_permitted,
475                            "cannot write compressed section '" + Sec.Name +
476                                "' ");
477 }
478 
479 Error DecompressedSection::accept(SectionVisitor &Visitor) const {
480   return Visitor.visit(*this);
481 }
482 
483 Error DecompressedSection::accept(MutableSectionVisitor &Visitor) {
484   return Visitor.visit(*this);
485 }
486 
487 Error OwnedDataSection::accept(SectionVisitor &Visitor) const {
488   return Visitor.visit(*this);
489 }
490 
491 Error OwnedDataSection::accept(MutableSectionVisitor &Visitor) {
492   return Visitor.visit(*this);
493 }
494 
495 void OwnedDataSection::appendHexData(StringRef HexData) {
496   assert((HexData.size() & 1) == 0);
497   while (!HexData.empty()) {
498     Data.push_back(checkedGetHex<uint8_t>(HexData.take_front(2)));
499     HexData = HexData.drop_front(2);
500   }
501   Size = Data.size();
502 }
503 
504 Error BinarySectionWriter::visit(const CompressedSection &Sec) {
505   return createStringError(errc::operation_not_permitted,
506                            "cannot write compressed section '" + Sec.Name +
507                                "' ");
508 }
509 
510 template <class ELFT>
511 Error ELFSectionWriter<ELFT>::visit(const CompressedSection &Sec) {
512   uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
513   Elf_Chdr_Impl<ELFT> Chdr = {};
514   switch (Sec.CompressionType) {
515   case DebugCompressionType::None:
516     std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(), Buf);
517     return Error::success();
518   case DebugCompressionType::Zlib:
519     Chdr.ch_type = ELF::ELFCOMPRESS_ZLIB;
520     break;
521   case DebugCompressionType::Zstd:
522     Chdr.ch_type = ELF::ELFCOMPRESS_ZSTD;
523     break;
524   }
525   Chdr.ch_size = Sec.DecompressedSize;
526   Chdr.ch_addralign = Sec.DecompressedAlign;
527   memcpy(Buf, &Chdr, sizeof(Chdr));
528   Buf += sizeof(Chdr);
529 
530   std::copy(Sec.CompressedData.begin(), Sec.CompressedData.end(), Buf);
531   return Error::success();
532 }
533 
534 CompressedSection::CompressedSection(const SectionBase &Sec,
535                                      DebugCompressionType CompressionType,
536                                      bool Is64Bits)
537     : SectionBase(Sec), CompressionType(CompressionType),
538       DecompressedSize(Sec.OriginalData.size()), DecompressedAlign(Sec.Align) {
539   compression::compress(compression::Params(CompressionType), OriginalData,
540                         CompressedData);
541 
542   Flags |= ELF::SHF_COMPRESSED;
543   size_t ChdrSize = Is64Bits ? sizeof(object::Elf_Chdr_Impl<object::ELF64LE>)
544                              : sizeof(object::Elf_Chdr_Impl<object::ELF32LE>);
545   Size = ChdrSize + CompressedData.size();
546   Align = 8;
547 }
548 
549 CompressedSection::CompressedSection(ArrayRef<uint8_t> CompressedData,
550                                      uint32_t ChType, uint64_t DecompressedSize,
551                                      uint64_t DecompressedAlign)
552     : ChType(ChType), CompressionType(DebugCompressionType::None),
553       DecompressedSize(DecompressedSize), DecompressedAlign(DecompressedAlign) {
554   OriginalData = CompressedData;
555 }
556 
557 Error CompressedSection::accept(SectionVisitor &Visitor) const {
558   return Visitor.visit(*this);
559 }
560 
561 Error CompressedSection::accept(MutableSectionVisitor &Visitor) {
562   return Visitor.visit(*this);
563 }
564 
565 void StringTableSection::addString(StringRef Name) { StrTabBuilder.add(Name); }
566 
567 uint32_t StringTableSection::findIndex(StringRef Name) const {
568   return StrTabBuilder.getOffset(Name);
569 }
570 
571 void StringTableSection::prepareForLayout() {
572   StrTabBuilder.finalize();
573   Size = StrTabBuilder.getSize();
574 }
575 
576 Error SectionWriter::visit(const StringTableSection &Sec) {
577   Sec.StrTabBuilder.write(reinterpret_cast<uint8_t *>(Out.getBufferStart()) +
578                           Sec.Offset);
579   return Error::success();
580 }
581 
582 Error StringTableSection::accept(SectionVisitor &Visitor) const {
583   return Visitor.visit(*this);
584 }
585 
586 Error StringTableSection::accept(MutableSectionVisitor &Visitor) {
587   return Visitor.visit(*this);
588 }
589 
590 template <class ELFT>
591 Error ELFSectionWriter<ELFT>::visit(const SectionIndexSection &Sec) {
592   uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
593   llvm::copy(Sec.Indexes, reinterpret_cast<Elf_Word *>(Buf));
594   return Error::success();
595 }
596 
597 Error SectionIndexSection::initialize(SectionTableRef SecTable) {
598   Size = 0;
599   Expected<SymbolTableSection *> Sec =
600       SecTable.getSectionOfType<SymbolTableSection>(
601           Link,
602           "Link field value " + Twine(Link) + " in section " + Name +
603               " is invalid",
604           "Link field value " + Twine(Link) + " in section " + Name +
605               " is not a symbol table");
606   if (!Sec)
607     return Sec.takeError();
608 
609   setSymTab(*Sec);
610   Symbols->setShndxTable(this);
611   return Error::success();
612 }
613 
614 void SectionIndexSection::finalize() { Link = Symbols->Index; }
615 
616 Error SectionIndexSection::accept(SectionVisitor &Visitor) const {
617   return Visitor.visit(*this);
618 }
619 
620 Error SectionIndexSection::accept(MutableSectionVisitor &Visitor) {
621   return Visitor.visit(*this);
622 }
623 
624 static bool isValidReservedSectionIndex(uint16_t Index, uint16_t Machine) {
625   switch (Index) {
626   case SHN_ABS:
627   case SHN_COMMON:
628     return true;
629   }
630 
631   if (Machine == EM_AMDGPU) {
632     return Index == SHN_AMDGPU_LDS;
633   }
634 
635   if (Machine == EM_MIPS) {
636     switch (Index) {
637     case SHN_MIPS_ACOMMON:
638     case SHN_MIPS_SCOMMON:
639     case SHN_MIPS_SUNDEFINED:
640       return true;
641     }
642   }
643 
644   if (Machine == EM_HEXAGON) {
645     switch (Index) {
646     case SHN_HEXAGON_SCOMMON:
647     case SHN_HEXAGON_SCOMMON_1:
648     case SHN_HEXAGON_SCOMMON_2:
649     case SHN_HEXAGON_SCOMMON_4:
650     case SHN_HEXAGON_SCOMMON_8:
651       return true;
652     }
653   }
654   return false;
655 }
656 
657 // Large indexes force us to clarify exactly what this function should do. This
658 // function should return the value that will appear in st_shndx when written
659 // out.
660 uint16_t Symbol::getShndx() const {
661   if (DefinedIn != nullptr) {
662     if (DefinedIn->Index >= SHN_LORESERVE)
663       return SHN_XINDEX;
664     return DefinedIn->Index;
665   }
666 
667   if (ShndxType == SYMBOL_SIMPLE_INDEX) {
668     // This means that we don't have a defined section but we do need to
669     // output a legitimate section index.
670     return SHN_UNDEF;
671   }
672 
673   assert(ShndxType == SYMBOL_ABS || ShndxType == SYMBOL_COMMON ||
674          (ShndxType >= SYMBOL_LOPROC && ShndxType <= SYMBOL_HIPROC) ||
675          (ShndxType >= SYMBOL_LOOS && ShndxType <= SYMBOL_HIOS));
676   return static_cast<uint16_t>(ShndxType);
677 }
678 
679 bool Symbol::isCommon() const { return getShndx() == SHN_COMMON; }
680 
681 void SymbolTableSection::assignIndices() {
682   uint32_t Index = 0;
683   for (auto &Sym : Symbols)
684     Sym->Index = Index++;
685 }
686 
687 void SymbolTableSection::addSymbol(Twine Name, uint8_t Bind, uint8_t Type,
688                                    SectionBase *DefinedIn, uint64_t Value,
689                                    uint8_t Visibility, uint16_t Shndx,
690                                    uint64_t SymbolSize) {
691   Symbol Sym;
692   Sym.Name = Name.str();
693   Sym.Binding = Bind;
694   Sym.Type = Type;
695   Sym.DefinedIn = DefinedIn;
696   if (DefinedIn != nullptr)
697     DefinedIn->HasSymbol = true;
698   if (DefinedIn == nullptr) {
699     if (Shndx >= SHN_LORESERVE)
700       Sym.ShndxType = static_cast<SymbolShndxType>(Shndx);
701     else
702       Sym.ShndxType = SYMBOL_SIMPLE_INDEX;
703   }
704   Sym.Value = Value;
705   Sym.Visibility = Visibility;
706   Sym.Size = SymbolSize;
707   Sym.Index = Symbols.size();
708   Symbols.emplace_back(std::make_unique<Symbol>(Sym));
709   Size += this->EntrySize;
710 }
711 
712 Error SymbolTableSection::removeSectionReferences(
713     bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
714   if (ToRemove(SectionIndexTable))
715     SectionIndexTable = nullptr;
716   if (ToRemove(SymbolNames)) {
717     if (!AllowBrokenLinks)
718       return createStringError(
719           llvm::errc::invalid_argument,
720           "string table '%s' cannot be removed because it is "
721           "referenced by the symbol table '%s'",
722           SymbolNames->Name.data(), this->Name.data());
723     SymbolNames = nullptr;
724   }
725   return removeSymbols(
726       [ToRemove](const Symbol &Sym) { return ToRemove(Sym.DefinedIn); });
727 }
728 
729 void SymbolTableSection::updateSymbols(function_ref<void(Symbol &)> Callable) {
730   for (SymPtr &Sym : llvm::drop_begin(Symbols))
731     Callable(*Sym);
732   std::stable_partition(
733       std::begin(Symbols), std::end(Symbols),
734       [](const SymPtr &Sym) { return Sym->Binding == STB_LOCAL; });
735   assignIndices();
736 }
737 
738 Error SymbolTableSection::removeSymbols(
739     function_ref<bool(const Symbol &)> ToRemove) {
740   Symbols.erase(
741       std::remove_if(std::begin(Symbols) + 1, std::end(Symbols),
742                      [ToRemove](const SymPtr &Sym) { return ToRemove(*Sym); }),
743       std::end(Symbols));
744   Size = Symbols.size() * EntrySize;
745   assignIndices();
746   return Error::success();
747 }
748 
749 void SymbolTableSection::replaceSectionReferences(
750     const DenseMap<SectionBase *, SectionBase *> &FromTo) {
751   for (std::unique_ptr<Symbol> &Sym : Symbols)
752     if (SectionBase *To = FromTo.lookup(Sym->DefinedIn))
753       Sym->DefinedIn = To;
754 }
755 
756 Error SymbolTableSection::initialize(SectionTableRef SecTable) {
757   Size = 0;
758   Expected<StringTableSection *> Sec =
759       SecTable.getSectionOfType<StringTableSection>(
760           Link,
761           "Symbol table has link index of " + Twine(Link) +
762               " which is not a valid index",
763           "Symbol table has link index of " + Twine(Link) +
764               " which is not a string table");
765   if (!Sec)
766     return Sec.takeError();
767 
768   setStrTab(*Sec);
769   return Error::success();
770 }
771 
772 void SymbolTableSection::finalize() {
773   uint32_t MaxLocalIndex = 0;
774   for (std::unique_ptr<Symbol> &Sym : Symbols) {
775     Sym->NameIndex =
776         SymbolNames == nullptr ? 0 : SymbolNames->findIndex(Sym->Name);
777     if (Sym->Binding == STB_LOCAL)
778       MaxLocalIndex = std::max(MaxLocalIndex, Sym->Index);
779   }
780   // Now we need to set the Link and Info fields.
781   Link = SymbolNames == nullptr ? 0 : SymbolNames->Index;
782   Info = MaxLocalIndex + 1;
783 }
784 
785 void SymbolTableSection::prepareForLayout() {
786   // Reserve proper amount of space in section index table, so we can
787   // layout sections correctly. We will fill the table with correct
788   // indexes later in fillShdnxTable.
789   if (SectionIndexTable)
790     SectionIndexTable->reserve(Symbols.size());
791 
792   // Add all of our strings to SymbolNames so that SymbolNames has the right
793   // size before layout is decided.
794   // If the symbol names section has been removed, don't try to add strings to
795   // the table.
796   if (SymbolNames != nullptr)
797     for (std::unique_ptr<Symbol> &Sym : Symbols)
798       SymbolNames->addString(Sym->Name);
799 }
800 
801 void SymbolTableSection::fillShndxTable() {
802   if (SectionIndexTable == nullptr)
803     return;
804   // Fill section index table with real section indexes. This function must
805   // be called after assignOffsets.
806   for (const std::unique_ptr<Symbol> &Sym : Symbols) {
807     if (Sym->DefinedIn != nullptr && Sym->DefinedIn->Index >= SHN_LORESERVE)
808       SectionIndexTable->addIndex(Sym->DefinedIn->Index);
809     else
810       SectionIndexTable->addIndex(SHN_UNDEF);
811   }
812 }
813 
814 Expected<const Symbol *>
815 SymbolTableSection::getSymbolByIndex(uint32_t Index) const {
816   if (Symbols.size() <= Index)
817     return createStringError(errc::invalid_argument,
818                              "invalid symbol index: " + Twine(Index));
819   return Symbols[Index].get();
820 }
821 
822 Expected<Symbol *> SymbolTableSection::getSymbolByIndex(uint32_t Index) {
823   Expected<const Symbol *> Sym =
824       static_cast<const SymbolTableSection *>(this)->getSymbolByIndex(Index);
825   if (!Sym)
826     return Sym.takeError();
827 
828   return const_cast<Symbol *>(*Sym);
829 }
830 
831 template <class ELFT>
832 Error ELFSectionWriter<ELFT>::visit(const SymbolTableSection &Sec) {
833   Elf_Sym *Sym = reinterpret_cast<Elf_Sym *>(Out.getBufferStart() + Sec.Offset);
834   // Loop though symbols setting each entry of the symbol table.
835   for (const std::unique_ptr<Symbol> &Symbol : Sec.Symbols) {
836     Sym->st_name = Symbol->NameIndex;
837     Sym->st_value = Symbol->Value;
838     Sym->st_size = Symbol->Size;
839     Sym->st_other = Symbol->Visibility;
840     Sym->setBinding(Symbol->Binding);
841     Sym->setType(Symbol->Type);
842     Sym->st_shndx = Symbol->getShndx();
843     ++Sym;
844   }
845   return Error::success();
846 }
847 
848 Error SymbolTableSection::accept(SectionVisitor &Visitor) const {
849   return Visitor.visit(*this);
850 }
851 
852 Error SymbolTableSection::accept(MutableSectionVisitor &Visitor) {
853   return Visitor.visit(*this);
854 }
855 
856 StringRef RelocationSectionBase::getNamePrefix() const {
857   switch (Type) {
858   case SHT_REL:
859     return ".rel";
860   case SHT_RELA:
861     return ".rela";
862   default:
863     llvm_unreachable("not a relocation section");
864   }
865 }
866 
867 Error RelocationSection::removeSectionReferences(
868     bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
869   if (ToRemove(Symbols)) {
870     if (!AllowBrokenLinks)
871       return createStringError(
872           llvm::errc::invalid_argument,
873           "symbol table '%s' cannot be removed because it is "
874           "referenced by the relocation section '%s'",
875           Symbols->Name.data(), this->Name.data());
876     Symbols = nullptr;
877   }
878 
879   for (const Relocation &R : Relocations) {
880     if (!R.RelocSymbol || !R.RelocSymbol->DefinedIn ||
881         !ToRemove(R.RelocSymbol->DefinedIn))
882       continue;
883     return createStringError(llvm::errc::invalid_argument,
884                              "section '%s' cannot be removed: (%s+0x%" PRIx64
885                              ") has relocation against symbol '%s'",
886                              R.RelocSymbol->DefinedIn->Name.data(),
887                              SecToApplyRel->Name.data(), R.Offset,
888                              R.RelocSymbol->Name.c_str());
889   }
890 
891   return Error::success();
892 }
893 
894 template <class SymTabType>
895 Error RelocSectionWithSymtabBase<SymTabType>::initialize(
896     SectionTableRef SecTable) {
897   if (Link != SHN_UNDEF) {
898     Expected<SymTabType *> Sec = SecTable.getSectionOfType<SymTabType>(
899         Link,
900         "Link field value " + Twine(Link) + " in section " + Name +
901             " is invalid",
902         "Link field value " + Twine(Link) + " in section " + Name +
903             " is not a symbol table");
904     if (!Sec)
905       return Sec.takeError();
906 
907     setSymTab(*Sec);
908   }
909 
910   if (Info != SHN_UNDEF) {
911     Expected<SectionBase *> Sec =
912         SecTable.getSection(Info, "Info field value " + Twine(Info) +
913                                       " in section " + Name + " is invalid");
914     if (!Sec)
915       return Sec.takeError();
916 
917     setSection(*Sec);
918   } else
919     setSection(nullptr);
920 
921   return Error::success();
922 }
923 
924 template <class SymTabType>
925 void RelocSectionWithSymtabBase<SymTabType>::finalize() {
926   this->Link = Symbols ? Symbols->Index : 0;
927 
928   if (SecToApplyRel != nullptr)
929     this->Info = SecToApplyRel->Index;
930 }
931 
932 template <class ELFT>
933 static void setAddend(Elf_Rel_Impl<ELFT, false> &, uint64_t) {}
934 
935 template <class ELFT>
936 static void setAddend(Elf_Rel_Impl<ELFT, true> &Rela, uint64_t Addend) {
937   Rela.r_addend = Addend;
938 }
939 
940 template <class RelRange, class T>
941 static void writeRel(const RelRange &Relocations, T *Buf, bool IsMips64EL) {
942   for (const auto &Reloc : Relocations) {
943     Buf->r_offset = Reloc.Offset;
944     setAddend(*Buf, Reloc.Addend);
945     Buf->setSymbolAndType(Reloc.RelocSymbol ? Reloc.RelocSymbol->Index : 0,
946                           Reloc.Type, IsMips64EL);
947     ++Buf;
948   }
949 }
950 
951 template <class ELFT>
952 Error ELFSectionWriter<ELFT>::visit(const RelocationSection &Sec) {
953   uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
954   if (Sec.Type == SHT_REL)
955     writeRel(Sec.Relocations, reinterpret_cast<Elf_Rel *>(Buf),
956              Sec.getObject().IsMips64EL);
957   else
958     writeRel(Sec.Relocations, reinterpret_cast<Elf_Rela *>(Buf),
959              Sec.getObject().IsMips64EL);
960   return Error::success();
961 }
962 
963 Error RelocationSection::accept(SectionVisitor &Visitor) const {
964   return Visitor.visit(*this);
965 }
966 
967 Error RelocationSection::accept(MutableSectionVisitor &Visitor) {
968   return Visitor.visit(*this);
969 }
970 
971 Error RelocationSection::removeSymbols(
972     function_ref<bool(const Symbol &)> ToRemove) {
973   for (const Relocation &Reloc : Relocations)
974     if (Reloc.RelocSymbol && ToRemove(*Reloc.RelocSymbol))
975       return createStringError(
976           llvm::errc::invalid_argument,
977           "not stripping symbol '%s' because it is named in a relocation",
978           Reloc.RelocSymbol->Name.data());
979   return Error::success();
980 }
981 
982 void RelocationSection::markSymbols() {
983   for (const Relocation &Reloc : Relocations)
984     if (Reloc.RelocSymbol)
985       Reloc.RelocSymbol->Referenced = true;
986 }
987 
988 void RelocationSection::replaceSectionReferences(
989     const DenseMap<SectionBase *, SectionBase *> &FromTo) {
990   // Update the target section if it was replaced.
991   if (SectionBase *To = FromTo.lookup(SecToApplyRel))
992     SecToApplyRel = To;
993 }
994 
995 Error SectionWriter::visit(const DynamicRelocationSection &Sec) {
996   llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset);
997   return Error::success();
998 }
999 
1000 Error DynamicRelocationSection::accept(SectionVisitor &Visitor) const {
1001   return Visitor.visit(*this);
1002 }
1003 
1004 Error DynamicRelocationSection::accept(MutableSectionVisitor &Visitor) {
1005   return Visitor.visit(*this);
1006 }
1007 
1008 Error DynamicRelocationSection::removeSectionReferences(
1009     bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
1010   if (ToRemove(Symbols)) {
1011     if (!AllowBrokenLinks)
1012       return createStringError(
1013           llvm::errc::invalid_argument,
1014           "symbol table '%s' cannot be removed because it is "
1015           "referenced by the relocation section '%s'",
1016           Symbols->Name.data(), this->Name.data());
1017     Symbols = nullptr;
1018   }
1019 
1020   // SecToApplyRel contains a section referenced by sh_info field. It keeps
1021   // a section to which the relocation section applies. When we remove any
1022   // sections we also remove their relocation sections. Since we do that much
1023   // earlier, this assert should never be triggered.
1024   assert(!SecToApplyRel || !ToRemove(SecToApplyRel));
1025   return Error::success();
1026 }
1027 
1028 Error Section::removeSectionReferences(
1029     bool AllowBrokenDependency,
1030     function_ref<bool(const SectionBase *)> ToRemove) {
1031   if (ToRemove(LinkSection)) {
1032     if (!AllowBrokenDependency)
1033       return createStringError(llvm::errc::invalid_argument,
1034                                "section '%s' cannot be removed because it is "
1035                                "referenced by the section '%s'",
1036                                LinkSection->Name.data(), this->Name.data());
1037     LinkSection = nullptr;
1038   }
1039   return Error::success();
1040 }
1041 
1042 void GroupSection::finalize() {
1043   this->Info = Sym ? Sym->Index : 0;
1044   this->Link = SymTab ? SymTab->Index : 0;
1045   // Linker deduplication for GRP_COMDAT is based on Sym->Name. The local/global
1046   // status is not part of the equation. If Sym is localized, the intention is
1047   // likely to make the group fully localized. Drop GRP_COMDAT to suppress
1048   // deduplication. See https://groups.google.com/g/generic-abi/c/2X6mR-s2zoc
1049   if ((FlagWord & GRP_COMDAT) && Sym && Sym->Binding == STB_LOCAL)
1050     this->FlagWord &= ~GRP_COMDAT;
1051 }
1052 
1053 Error GroupSection::removeSectionReferences(
1054     bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
1055   if (ToRemove(SymTab)) {
1056     if (!AllowBrokenLinks)
1057       return createStringError(
1058           llvm::errc::invalid_argument,
1059           "section '.symtab' cannot be removed because it is "
1060           "referenced by the group section '%s'",
1061           this->Name.data());
1062     SymTab = nullptr;
1063     Sym = nullptr;
1064   }
1065   llvm::erase_if(GroupMembers, ToRemove);
1066   return Error::success();
1067 }
1068 
1069 Error GroupSection::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
1070   if (ToRemove(*Sym))
1071     return createStringError(llvm::errc::invalid_argument,
1072                              "symbol '%s' cannot be removed because it is "
1073                              "referenced by the section '%s[%d]'",
1074                              Sym->Name.data(), this->Name.data(), this->Index);
1075   return Error::success();
1076 }
1077 
1078 void GroupSection::markSymbols() {
1079   if (Sym)
1080     Sym->Referenced = true;
1081 }
1082 
1083 void GroupSection::replaceSectionReferences(
1084     const DenseMap<SectionBase *, SectionBase *> &FromTo) {
1085   for (SectionBase *&Sec : GroupMembers)
1086     if (SectionBase *To = FromTo.lookup(Sec))
1087       Sec = To;
1088 }
1089 
1090 void GroupSection::onRemove() {
1091   // As the header section of the group is removed, drop the Group flag in its
1092   // former members.
1093   for (SectionBase *Sec : GroupMembers)
1094     Sec->Flags &= ~SHF_GROUP;
1095 }
1096 
1097 Error Section::initialize(SectionTableRef SecTable) {
1098   if (Link == ELF::SHN_UNDEF)
1099     return Error::success();
1100 
1101   Expected<SectionBase *> Sec =
1102       SecTable.getSection(Link, "Link field value " + Twine(Link) +
1103                                     " in section " + Name + " is invalid");
1104   if (!Sec)
1105     return Sec.takeError();
1106 
1107   LinkSection = *Sec;
1108 
1109   if (LinkSection->Type == ELF::SHT_SYMTAB)
1110     LinkSection = nullptr;
1111 
1112   return Error::success();
1113 }
1114 
1115 void Section::finalize() { this->Link = LinkSection ? LinkSection->Index : 0; }
1116 
1117 void GnuDebugLinkSection::init(StringRef File) {
1118   FileName = sys::path::filename(File);
1119   // The format for the .gnu_debuglink starts with the file name and is
1120   // followed by a null terminator and then the CRC32 of the file. The CRC32
1121   // should be 4 byte aligned. So we add the FileName size, a 1 for the null
1122   // byte, and then finally push the size to alignment and add 4.
1123   Size = alignTo(FileName.size() + 1, 4) + 4;
1124   // The CRC32 will only be aligned if we align the whole section.
1125   Align = 4;
1126   Type = OriginalType = ELF::SHT_PROGBITS;
1127   Name = ".gnu_debuglink";
1128   // For sections not found in segments, OriginalOffset is only used to
1129   // establish the order that sections should go in. By using the maximum
1130   // possible offset we cause this section to wind up at the end.
1131   OriginalOffset = std::numeric_limits<uint64_t>::max();
1132 }
1133 
1134 GnuDebugLinkSection::GnuDebugLinkSection(StringRef File,
1135                                          uint32_t PrecomputedCRC)
1136     : FileName(File), CRC32(PrecomputedCRC) {
1137   init(File);
1138 }
1139 
1140 template <class ELFT>
1141 Error ELFSectionWriter<ELFT>::visit(const GnuDebugLinkSection &Sec) {
1142   unsigned char *Buf =
1143       reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
1144   Elf_Word *CRC =
1145       reinterpret_cast<Elf_Word *>(Buf + Sec.Size - sizeof(Elf_Word));
1146   *CRC = Sec.CRC32;
1147   llvm::copy(Sec.FileName, Buf);
1148   return Error::success();
1149 }
1150 
1151 Error GnuDebugLinkSection::accept(SectionVisitor &Visitor) const {
1152   return Visitor.visit(*this);
1153 }
1154 
1155 Error GnuDebugLinkSection::accept(MutableSectionVisitor &Visitor) {
1156   return Visitor.visit(*this);
1157 }
1158 
1159 template <class ELFT>
1160 Error ELFSectionWriter<ELFT>::visit(const GroupSection &Sec) {
1161   ELF::Elf32_Word *Buf =
1162       reinterpret_cast<ELF::Elf32_Word *>(Out.getBufferStart() + Sec.Offset);
1163   support::endian::write32<ELFT::TargetEndianness>(Buf++, Sec.FlagWord);
1164   for (SectionBase *S : Sec.GroupMembers)
1165     support::endian::write32<ELFT::TargetEndianness>(Buf++, S->Index);
1166   return Error::success();
1167 }
1168 
1169 Error GroupSection::accept(SectionVisitor &Visitor) const {
1170   return Visitor.visit(*this);
1171 }
1172 
1173 Error GroupSection::accept(MutableSectionVisitor &Visitor) {
1174   return Visitor.visit(*this);
1175 }
1176 
1177 // Returns true IFF a section is wholly inside the range of a segment
1178 static bool sectionWithinSegment(const SectionBase &Sec, const Segment &Seg) {
1179   // If a section is empty it should be treated like it has a size of 1. This is
1180   // to clarify the case when an empty section lies on a boundary between two
1181   // segments and ensures that the section "belongs" to the second segment and
1182   // not the first.
1183   uint64_t SecSize = Sec.Size ? Sec.Size : 1;
1184 
1185   // Ignore just added sections.
1186   if (Sec.OriginalOffset == std::numeric_limits<uint64_t>::max())
1187     return false;
1188 
1189   if (Sec.Type == SHT_NOBITS) {
1190     if (!(Sec.Flags & SHF_ALLOC))
1191       return false;
1192 
1193     bool SectionIsTLS = Sec.Flags & SHF_TLS;
1194     bool SegmentIsTLS = Seg.Type == PT_TLS;
1195     if (SectionIsTLS != SegmentIsTLS)
1196       return false;
1197 
1198     return Seg.VAddr <= Sec.Addr &&
1199            Seg.VAddr + Seg.MemSize >= Sec.Addr + SecSize;
1200   }
1201 
1202   return Seg.Offset <= Sec.OriginalOffset &&
1203          Seg.Offset + Seg.FileSize >= Sec.OriginalOffset + SecSize;
1204 }
1205 
1206 // Returns true IFF a segment's original offset is inside of another segment's
1207 // range.
1208 static bool segmentOverlapsSegment(const Segment &Child,
1209                                    const Segment &Parent) {
1210 
1211   return Parent.OriginalOffset <= Child.OriginalOffset &&
1212          Parent.OriginalOffset + Parent.FileSize > Child.OriginalOffset;
1213 }
1214 
1215 static bool compareSegmentsByOffset(const Segment *A, const Segment *B) {
1216   // Any segment without a parent segment should come before a segment
1217   // that has a parent segment.
1218   if (A->OriginalOffset < B->OriginalOffset)
1219     return true;
1220   if (A->OriginalOffset > B->OriginalOffset)
1221     return false;
1222   return A->Index < B->Index;
1223 }
1224 
1225 void BasicELFBuilder::initFileHeader() {
1226   Obj->Flags = 0x0;
1227   Obj->Type = ET_REL;
1228   Obj->OSABI = ELFOSABI_NONE;
1229   Obj->ABIVersion = 0;
1230   Obj->Entry = 0x0;
1231   Obj->Machine = EM_NONE;
1232   Obj->Version = 1;
1233 }
1234 
1235 void BasicELFBuilder::initHeaderSegment() { Obj->ElfHdrSegment.Index = 0; }
1236 
1237 StringTableSection *BasicELFBuilder::addStrTab() {
1238   auto &StrTab = Obj->addSection<StringTableSection>();
1239   StrTab.Name = ".strtab";
1240 
1241   Obj->SectionNames = &StrTab;
1242   return &StrTab;
1243 }
1244 
1245 SymbolTableSection *BasicELFBuilder::addSymTab(StringTableSection *StrTab) {
1246   auto &SymTab = Obj->addSection<SymbolTableSection>();
1247 
1248   SymTab.Name = ".symtab";
1249   SymTab.Link = StrTab->Index;
1250 
1251   // The symbol table always needs a null symbol
1252   SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0);
1253 
1254   Obj->SymbolTable = &SymTab;
1255   return &SymTab;
1256 }
1257 
1258 Error BasicELFBuilder::initSections() {
1259   for (SectionBase &Sec : Obj->sections())
1260     if (Error Err = Sec.initialize(Obj->sections()))
1261       return Err;
1262 
1263   return Error::success();
1264 }
1265 
1266 void BinaryELFBuilder::addData(SymbolTableSection *SymTab) {
1267   auto Data = ArrayRef<uint8_t>(
1268       reinterpret_cast<const uint8_t *>(MemBuf->getBufferStart()),
1269       MemBuf->getBufferSize());
1270   auto &DataSection = Obj->addSection<Section>(Data);
1271   DataSection.Name = ".data";
1272   DataSection.Type = ELF::SHT_PROGBITS;
1273   DataSection.Size = Data.size();
1274   DataSection.Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE;
1275 
1276   std::string SanitizedFilename = MemBuf->getBufferIdentifier().str();
1277   std::replace_if(
1278       std::begin(SanitizedFilename), std::end(SanitizedFilename),
1279       [](char C) { return !isAlnum(C); }, '_');
1280   Twine Prefix = Twine("_binary_") + SanitizedFilename;
1281 
1282   SymTab->addSymbol(Prefix + "_start", STB_GLOBAL, STT_NOTYPE, &DataSection,
1283                     /*Value=*/0, NewSymbolVisibility, 0, 0);
1284   SymTab->addSymbol(Prefix + "_end", STB_GLOBAL, STT_NOTYPE, &DataSection,
1285                     /*Value=*/DataSection.Size, NewSymbolVisibility, 0, 0);
1286   SymTab->addSymbol(Prefix + "_size", STB_GLOBAL, STT_NOTYPE, nullptr,
1287                     /*Value=*/DataSection.Size, NewSymbolVisibility, SHN_ABS,
1288                     0);
1289 }
1290 
1291 Expected<std::unique_ptr<Object>> BinaryELFBuilder::build() {
1292   initFileHeader();
1293   initHeaderSegment();
1294 
1295   SymbolTableSection *SymTab = addSymTab(addStrTab());
1296   if (Error Err = initSections())
1297     return std::move(Err);
1298   addData(SymTab);
1299 
1300   return std::move(Obj);
1301 }
1302 
1303 // Adds sections from IHEX data file. Data should have been
1304 // fully validated by this time.
1305 void IHexELFBuilder::addDataSections() {
1306   OwnedDataSection *Section = nullptr;
1307   uint64_t SegmentAddr = 0, BaseAddr = 0;
1308   uint32_t SecNo = 1;
1309 
1310   for (const IHexRecord &R : Records) {
1311     uint64_t RecAddr;
1312     switch (R.Type) {
1313     case IHexRecord::Data:
1314       // Ignore empty data records
1315       if (R.HexData.empty())
1316         continue;
1317       RecAddr = R.Addr + SegmentAddr + BaseAddr;
1318       if (!Section || Section->Addr + Section->Size != RecAddr) {
1319         // OriginalOffset field is only used to sort sections before layout, so
1320         // instead of keeping track of real offsets in IHEX file, and as
1321         // layoutSections() and layoutSectionsForOnlyKeepDebug() use
1322         // llvm::stable_sort(), we can just set it to a constant (zero).
1323         Section = &Obj->addSection<OwnedDataSection>(
1324             ".sec" + std::to_string(SecNo), RecAddr,
1325             ELF::SHF_ALLOC | ELF::SHF_WRITE, 0);
1326         SecNo++;
1327       }
1328       Section->appendHexData(R.HexData);
1329       break;
1330     case IHexRecord::EndOfFile:
1331       break;
1332     case IHexRecord::SegmentAddr:
1333       // 20-bit segment address.
1334       SegmentAddr = checkedGetHex<uint16_t>(R.HexData) << 4;
1335       break;
1336     case IHexRecord::StartAddr80x86:
1337     case IHexRecord::StartAddr:
1338       Obj->Entry = checkedGetHex<uint32_t>(R.HexData);
1339       assert(Obj->Entry <= 0xFFFFFU);
1340       break;
1341     case IHexRecord::ExtendedAddr:
1342       // 16-31 bits of linear base address
1343       BaseAddr = checkedGetHex<uint16_t>(R.HexData) << 16;
1344       break;
1345     default:
1346       llvm_unreachable("unknown record type");
1347     }
1348   }
1349 }
1350 
1351 Expected<std::unique_ptr<Object>> IHexELFBuilder::build() {
1352   initFileHeader();
1353   initHeaderSegment();
1354   StringTableSection *StrTab = addStrTab();
1355   addSymTab(StrTab);
1356   if (Error Err = initSections())
1357     return std::move(Err);
1358   addDataSections();
1359 
1360   return std::move(Obj);
1361 }
1362 
1363 template <class ELFT>
1364 ELFBuilder<ELFT>::ELFBuilder(const ELFObjectFile<ELFT> &ElfObj, Object &Obj,
1365                              std::optional<StringRef> ExtractPartition)
1366     : ElfFile(ElfObj.getELFFile()), Obj(Obj),
1367       ExtractPartition(ExtractPartition) {
1368   Obj.IsMips64EL = ElfFile.isMips64EL();
1369 }
1370 
1371 template <class ELFT> void ELFBuilder<ELFT>::setParentSegment(Segment &Child) {
1372   for (Segment &Parent : Obj.segments()) {
1373     // Every segment will overlap with itself but we don't want a segment to
1374     // be its own parent so we avoid that situation.
1375     if (&Child != &Parent && segmentOverlapsSegment(Child, Parent)) {
1376       // We want a canonical "most parental" segment but this requires
1377       // inspecting the ParentSegment.
1378       if (compareSegmentsByOffset(&Parent, &Child))
1379         if (Child.ParentSegment == nullptr ||
1380             compareSegmentsByOffset(&Parent, Child.ParentSegment)) {
1381           Child.ParentSegment = &Parent;
1382         }
1383     }
1384   }
1385 }
1386 
1387 template <class ELFT> Error ELFBuilder<ELFT>::findEhdrOffset() {
1388   if (!ExtractPartition)
1389     return Error::success();
1390 
1391   for (const SectionBase &Sec : Obj.sections()) {
1392     if (Sec.Type == SHT_LLVM_PART_EHDR && Sec.Name == *ExtractPartition) {
1393       EhdrOffset = Sec.Offset;
1394       return Error::success();
1395     }
1396   }
1397   return createStringError(errc::invalid_argument,
1398                            "could not find partition named '" +
1399                                *ExtractPartition + "'");
1400 }
1401 
1402 template <class ELFT>
1403 Error ELFBuilder<ELFT>::readProgramHeaders(const ELFFile<ELFT> &HeadersFile) {
1404   uint32_t Index = 0;
1405 
1406   Expected<typename ELFFile<ELFT>::Elf_Phdr_Range> Headers =
1407       HeadersFile.program_headers();
1408   if (!Headers)
1409     return Headers.takeError();
1410 
1411   for (const typename ELFFile<ELFT>::Elf_Phdr &Phdr : *Headers) {
1412     if (Phdr.p_offset + Phdr.p_filesz > HeadersFile.getBufSize())
1413       return createStringError(
1414           errc::invalid_argument,
1415           "program header with offset 0x" + Twine::utohexstr(Phdr.p_offset) +
1416               " and file size 0x" + Twine::utohexstr(Phdr.p_filesz) +
1417               " goes past the end of the file");
1418 
1419     ArrayRef<uint8_t> Data{HeadersFile.base() + Phdr.p_offset,
1420                            (size_t)Phdr.p_filesz};
1421     Segment &Seg = Obj.addSegment(Data);
1422     Seg.Type = Phdr.p_type;
1423     Seg.Flags = Phdr.p_flags;
1424     Seg.OriginalOffset = Phdr.p_offset + EhdrOffset;
1425     Seg.Offset = Phdr.p_offset + EhdrOffset;
1426     Seg.VAddr = Phdr.p_vaddr;
1427     Seg.PAddr = Phdr.p_paddr;
1428     Seg.FileSize = Phdr.p_filesz;
1429     Seg.MemSize = Phdr.p_memsz;
1430     Seg.Align = Phdr.p_align;
1431     Seg.Index = Index++;
1432     for (SectionBase &Sec : Obj.sections())
1433       if (sectionWithinSegment(Sec, Seg)) {
1434         Seg.addSection(&Sec);
1435         if (!Sec.ParentSegment || Sec.ParentSegment->Offset > Seg.Offset)
1436           Sec.ParentSegment = &Seg;
1437       }
1438   }
1439 
1440   auto &ElfHdr = Obj.ElfHdrSegment;
1441   ElfHdr.Index = Index++;
1442   ElfHdr.OriginalOffset = ElfHdr.Offset = EhdrOffset;
1443 
1444   const typename ELFT::Ehdr &Ehdr = HeadersFile.getHeader();
1445   auto &PrHdr = Obj.ProgramHdrSegment;
1446   PrHdr.Type = PT_PHDR;
1447   PrHdr.Flags = 0;
1448   // The spec requires us to have p_vaddr % p_align == p_offset % p_align.
1449   // Whereas this works automatically for ElfHdr, here OriginalOffset is
1450   // always non-zero and to ensure the equation we assign the same value to
1451   // VAddr as well.
1452   PrHdr.OriginalOffset = PrHdr.Offset = PrHdr.VAddr = EhdrOffset + Ehdr.e_phoff;
1453   PrHdr.PAddr = 0;
1454   PrHdr.FileSize = PrHdr.MemSize = Ehdr.e_phentsize * Ehdr.e_phnum;
1455   // The spec requires us to naturally align all the fields.
1456   PrHdr.Align = sizeof(Elf_Addr);
1457   PrHdr.Index = Index++;
1458 
1459   // Now we do an O(n^2) loop through the segments in order to match up
1460   // segments.
1461   for (Segment &Child : Obj.segments())
1462     setParentSegment(Child);
1463   setParentSegment(ElfHdr);
1464   setParentSegment(PrHdr);
1465 
1466   return Error::success();
1467 }
1468 
1469 template <class ELFT>
1470 Error ELFBuilder<ELFT>::initGroupSection(GroupSection *GroupSec) {
1471   if (GroupSec->Align % sizeof(ELF::Elf32_Word) != 0)
1472     return createStringError(errc::invalid_argument,
1473                              "invalid alignment " + Twine(GroupSec->Align) +
1474                                  " of group section '" + GroupSec->Name + "'");
1475   SectionTableRef SecTable = Obj.sections();
1476   if (GroupSec->Link != SHN_UNDEF) {
1477     auto SymTab = SecTable.template getSectionOfType<SymbolTableSection>(
1478         GroupSec->Link,
1479         "link field value '" + Twine(GroupSec->Link) + "' in section '" +
1480             GroupSec->Name + "' is invalid",
1481         "link field value '" + Twine(GroupSec->Link) + "' in section '" +
1482             GroupSec->Name + "' is not a symbol table");
1483     if (!SymTab)
1484       return SymTab.takeError();
1485 
1486     Expected<Symbol *> Sym = (*SymTab)->getSymbolByIndex(GroupSec->Info);
1487     if (!Sym)
1488       return createStringError(errc::invalid_argument,
1489                                "info field value '" + Twine(GroupSec->Info) +
1490                                    "' in section '" + GroupSec->Name +
1491                                    "' is not a valid symbol index");
1492     GroupSec->setSymTab(*SymTab);
1493     GroupSec->setSymbol(*Sym);
1494   }
1495   if (GroupSec->Contents.size() % sizeof(ELF::Elf32_Word) ||
1496       GroupSec->Contents.empty())
1497     return createStringError(errc::invalid_argument,
1498                              "the content of the section " + GroupSec->Name +
1499                                  " is malformed");
1500   const ELF::Elf32_Word *Word =
1501       reinterpret_cast<const ELF::Elf32_Word *>(GroupSec->Contents.data());
1502   const ELF::Elf32_Word *End =
1503       Word + GroupSec->Contents.size() / sizeof(ELF::Elf32_Word);
1504   GroupSec->setFlagWord(
1505       support::endian::read32<ELFT::TargetEndianness>(Word++));
1506   for (; Word != End; ++Word) {
1507     uint32_t Index = support::endian::read32<ELFT::TargetEndianness>(Word);
1508     Expected<SectionBase *> Sec = SecTable.getSection(
1509         Index, "group member index " + Twine(Index) + " in section '" +
1510                    GroupSec->Name + "' is invalid");
1511     if (!Sec)
1512       return Sec.takeError();
1513 
1514     GroupSec->addMember(*Sec);
1515   }
1516 
1517   return Error::success();
1518 }
1519 
1520 template <class ELFT>
1521 Error ELFBuilder<ELFT>::initSymbolTable(SymbolTableSection *SymTab) {
1522   Expected<const Elf_Shdr *> Shdr = ElfFile.getSection(SymTab->Index);
1523   if (!Shdr)
1524     return Shdr.takeError();
1525 
1526   Expected<StringRef> StrTabData = ElfFile.getStringTableForSymtab(**Shdr);
1527   if (!StrTabData)
1528     return StrTabData.takeError();
1529 
1530   ArrayRef<Elf_Word> ShndxData;
1531 
1532   Expected<typename ELFFile<ELFT>::Elf_Sym_Range> Symbols =
1533       ElfFile.symbols(*Shdr);
1534   if (!Symbols)
1535     return Symbols.takeError();
1536 
1537   for (const typename ELFFile<ELFT>::Elf_Sym &Sym : *Symbols) {
1538     SectionBase *DefSection = nullptr;
1539 
1540     Expected<StringRef> Name = Sym.getName(*StrTabData);
1541     if (!Name)
1542       return Name.takeError();
1543 
1544     if (Sym.st_shndx == SHN_XINDEX) {
1545       if (SymTab->getShndxTable() == nullptr)
1546         return createStringError(errc::invalid_argument,
1547                                  "symbol '" + *Name +
1548                                      "' has index SHN_XINDEX but no "
1549                                      "SHT_SYMTAB_SHNDX section exists");
1550       if (ShndxData.data() == nullptr) {
1551         Expected<const Elf_Shdr *> ShndxSec =
1552             ElfFile.getSection(SymTab->getShndxTable()->Index);
1553         if (!ShndxSec)
1554           return ShndxSec.takeError();
1555 
1556         Expected<ArrayRef<Elf_Word>> Data =
1557             ElfFile.template getSectionContentsAsArray<Elf_Word>(**ShndxSec);
1558         if (!Data)
1559           return Data.takeError();
1560 
1561         ShndxData = *Data;
1562         if (ShndxData.size() != Symbols->size())
1563           return createStringError(
1564               errc::invalid_argument,
1565               "symbol section index table does not have the same number of "
1566               "entries as the symbol table");
1567       }
1568       Elf_Word Index = ShndxData[&Sym - Symbols->begin()];
1569       Expected<SectionBase *> Sec = Obj.sections().getSection(
1570           Index,
1571           "symbol '" + *Name + "' has invalid section index " + Twine(Index));
1572       if (!Sec)
1573         return Sec.takeError();
1574 
1575       DefSection = *Sec;
1576     } else if (Sym.st_shndx >= SHN_LORESERVE) {
1577       if (!isValidReservedSectionIndex(Sym.st_shndx, Obj.Machine)) {
1578         return createStringError(
1579             errc::invalid_argument,
1580             "symbol '" + *Name +
1581                 "' has unsupported value greater than or equal "
1582                 "to SHN_LORESERVE: " +
1583                 Twine(Sym.st_shndx));
1584       }
1585     } else if (Sym.st_shndx != SHN_UNDEF) {
1586       Expected<SectionBase *> Sec = Obj.sections().getSection(
1587           Sym.st_shndx, "symbol '" + *Name +
1588                             "' is defined has invalid section index " +
1589                             Twine(Sym.st_shndx));
1590       if (!Sec)
1591         return Sec.takeError();
1592 
1593       DefSection = *Sec;
1594     }
1595 
1596     SymTab->addSymbol(*Name, Sym.getBinding(), Sym.getType(), DefSection,
1597                       Sym.getValue(), Sym.st_other, Sym.st_shndx, Sym.st_size);
1598   }
1599 
1600   return Error::success();
1601 }
1602 
1603 template <class ELFT>
1604 static void getAddend(uint64_t &, const Elf_Rel_Impl<ELFT, false> &) {}
1605 
1606 template <class ELFT>
1607 static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, true> &Rela) {
1608   ToSet = Rela.r_addend;
1609 }
1610 
1611 template <class T>
1612 static Error initRelocations(RelocationSection *Relocs, T RelRange) {
1613   for (const auto &Rel : RelRange) {
1614     Relocation ToAdd;
1615     ToAdd.Offset = Rel.r_offset;
1616     getAddend(ToAdd.Addend, Rel);
1617     ToAdd.Type = Rel.getType(Relocs->getObject().IsMips64EL);
1618 
1619     if (uint32_t Sym = Rel.getSymbol(Relocs->getObject().IsMips64EL)) {
1620       if (!Relocs->getObject().SymbolTable)
1621         return createStringError(
1622             errc::invalid_argument,
1623             "'" + Relocs->Name + "': relocation references symbol with index " +
1624                 Twine(Sym) + ", but there is no symbol table");
1625       Expected<Symbol *> SymByIndex =
1626           Relocs->getObject().SymbolTable->getSymbolByIndex(Sym);
1627       if (!SymByIndex)
1628         return SymByIndex.takeError();
1629 
1630       ToAdd.RelocSymbol = *SymByIndex;
1631     }
1632 
1633     Relocs->addRelocation(ToAdd);
1634   }
1635 
1636   return Error::success();
1637 }
1638 
1639 Expected<SectionBase *> SectionTableRef::getSection(uint32_t Index,
1640                                                     Twine ErrMsg) {
1641   if (Index == SHN_UNDEF || Index > Sections.size())
1642     return createStringError(errc::invalid_argument, ErrMsg);
1643   return Sections[Index - 1].get();
1644 }
1645 
1646 template <class T>
1647 Expected<T *> SectionTableRef::getSectionOfType(uint32_t Index,
1648                                                 Twine IndexErrMsg,
1649                                                 Twine TypeErrMsg) {
1650   Expected<SectionBase *> BaseSec = getSection(Index, IndexErrMsg);
1651   if (!BaseSec)
1652     return BaseSec.takeError();
1653 
1654   if (T *Sec = dyn_cast<T>(*BaseSec))
1655     return Sec;
1656 
1657   return createStringError(errc::invalid_argument, TypeErrMsg);
1658 }
1659 
1660 template <class ELFT>
1661 Expected<SectionBase &> ELFBuilder<ELFT>::makeSection(const Elf_Shdr &Shdr) {
1662   switch (Shdr.sh_type) {
1663   case SHT_REL:
1664   case SHT_RELA:
1665     if (Shdr.sh_flags & SHF_ALLOC) {
1666       if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1667         return Obj.addSection<DynamicRelocationSection>(*Data);
1668       else
1669         return Data.takeError();
1670     }
1671     return Obj.addSection<RelocationSection>(Obj);
1672   case SHT_STRTAB:
1673     // If a string table is allocated we don't want to mess with it. That would
1674     // mean altering the memory image. There are no special link types or
1675     // anything so we can just use a Section.
1676     if (Shdr.sh_flags & SHF_ALLOC) {
1677       if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1678         return Obj.addSection<Section>(*Data);
1679       else
1680         return Data.takeError();
1681     }
1682     return Obj.addSection<StringTableSection>();
1683   case SHT_HASH:
1684   case SHT_GNU_HASH:
1685     // Hash tables should refer to SHT_DYNSYM which we're not going to change.
1686     // Because of this we don't need to mess with the hash tables either.
1687     if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1688       return Obj.addSection<Section>(*Data);
1689     else
1690       return Data.takeError();
1691   case SHT_GROUP:
1692     if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1693       return Obj.addSection<GroupSection>(*Data);
1694     else
1695       return Data.takeError();
1696   case SHT_DYNSYM:
1697     if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1698       return Obj.addSection<DynamicSymbolTableSection>(*Data);
1699     else
1700       return Data.takeError();
1701   case SHT_DYNAMIC:
1702     if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1703       return Obj.addSection<DynamicSection>(*Data);
1704     else
1705       return Data.takeError();
1706   case SHT_SYMTAB: {
1707     auto &SymTab = Obj.addSection<SymbolTableSection>();
1708     Obj.SymbolTable = &SymTab;
1709     return SymTab;
1710   }
1711   case SHT_SYMTAB_SHNDX: {
1712     auto &ShndxSection = Obj.addSection<SectionIndexSection>();
1713     Obj.SectionIndexTable = &ShndxSection;
1714     return ShndxSection;
1715   }
1716   case SHT_NOBITS:
1717     return Obj.addSection<Section>(ArrayRef<uint8_t>());
1718   default: {
1719     Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr);
1720     if (!Data)
1721       return Data.takeError();
1722 
1723     Expected<StringRef> Name = ElfFile.getSectionName(Shdr);
1724     if (!Name)
1725       return Name.takeError();
1726 
1727     if (!(Shdr.sh_flags & ELF::SHF_COMPRESSED))
1728       return Obj.addSection<Section>(*Data);
1729     auto *Chdr = reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data->data());
1730     return Obj.addSection<CompressedSection>(CompressedSection(
1731         *Data, Chdr->ch_type, Chdr->ch_size, Chdr->ch_addralign));
1732   }
1733   }
1734 }
1735 
1736 template <class ELFT> Error ELFBuilder<ELFT>::readSectionHeaders() {
1737   uint32_t Index = 0;
1738   Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections =
1739       ElfFile.sections();
1740   if (!Sections)
1741     return Sections.takeError();
1742 
1743   for (const typename ELFFile<ELFT>::Elf_Shdr &Shdr : *Sections) {
1744     if (Index == 0) {
1745       ++Index;
1746       continue;
1747     }
1748     Expected<SectionBase &> Sec = makeSection(Shdr);
1749     if (!Sec)
1750       return Sec.takeError();
1751 
1752     Expected<StringRef> SecName = ElfFile.getSectionName(Shdr);
1753     if (!SecName)
1754       return SecName.takeError();
1755     Sec->Name = SecName->str();
1756     Sec->Type = Sec->OriginalType = Shdr.sh_type;
1757     Sec->Flags = Sec->OriginalFlags = Shdr.sh_flags;
1758     Sec->Addr = Shdr.sh_addr;
1759     Sec->Offset = Shdr.sh_offset;
1760     Sec->OriginalOffset = Shdr.sh_offset;
1761     Sec->Size = Shdr.sh_size;
1762     Sec->Link = Shdr.sh_link;
1763     Sec->Info = Shdr.sh_info;
1764     Sec->Align = Shdr.sh_addralign;
1765     Sec->EntrySize = Shdr.sh_entsize;
1766     Sec->Index = Index++;
1767     Sec->OriginalIndex = Sec->Index;
1768     Sec->OriginalData = ArrayRef<uint8_t>(
1769         ElfFile.base() + Shdr.sh_offset,
1770         (Shdr.sh_type == SHT_NOBITS) ? (size_t)0 : Shdr.sh_size);
1771   }
1772 
1773   return Error::success();
1774 }
1775 
1776 template <class ELFT> Error ELFBuilder<ELFT>::readSections(bool EnsureSymtab) {
1777   uint32_t ShstrIndex = ElfFile.getHeader().e_shstrndx;
1778   if (ShstrIndex == SHN_XINDEX) {
1779     Expected<const Elf_Shdr *> Sec = ElfFile.getSection(0);
1780     if (!Sec)
1781       return Sec.takeError();
1782 
1783     ShstrIndex = (*Sec)->sh_link;
1784   }
1785 
1786   if (ShstrIndex == SHN_UNDEF)
1787     Obj.HadShdrs = false;
1788   else {
1789     Expected<StringTableSection *> Sec =
1790         Obj.sections().template getSectionOfType<StringTableSection>(
1791             ShstrIndex,
1792             "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " +
1793                 " is invalid",
1794             "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " +
1795                 " does not reference a string table");
1796     if (!Sec)
1797       return Sec.takeError();
1798 
1799     Obj.SectionNames = *Sec;
1800   }
1801 
1802   // If a section index table exists we'll need to initialize it before we
1803   // initialize the symbol table because the symbol table might need to
1804   // reference it.
1805   if (Obj.SectionIndexTable)
1806     if (Error Err = Obj.SectionIndexTable->initialize(Obj.sections()))
1807       return Err;
1808 
1809   // Now that all of the sections have been added we can fill out some extra
1810   // details about symbol tables. We need the symbol table filled out before
1811   // any relocations.
1812   if (Obj.SymbolTable) {
1813     if (Error Err = Obj.SymbolTable->initialize(Obj.sections()))
1814       return Err;
1815     if (Error Err = initSymbolTable(Obj.SymbolTable))
1816       return Err;
1817   } else if (EnsureSymtab) {
1818     if (Error Err = Obj.addNewSymbolTable())
1819       return Err;
1820   }
1821 
1822   // Now that all sections and symbols have been added we can add
1823   // relocations that reference symbols and set the link and info fields for
1824   // relocation sections.
1825   for (SectionBase &Sec : Obj.sections()) {
1826     if (&Sec == Obj.SymbolTable)
1827       continue;
1828     if (Error Err = Sec.initialize(Obj.sections()))
1829       return Err;
1830     if (auto RelSec = dyn_cast<RelocationSection>(&Sec)) {
1831       Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections =
1832           ElfFile.sections();
1833       if (!Sections)
1834         return Sections.takeError();
1835 
1836       const typename ELFFile<ELFT>::Elf_Shdr *Shdr =
1837           Sections->begin() + RelSec->Index;
1838       if (RelSec->Type == SHT_REL) {
1839         Expected<typename ELFFile<ELFT>::Elf_Rel_Range> Rels =
1840             ElfFile.rels(*Shdr);
1841         if (!Rels)
1842           return Rels.takeError();
1843 
1844         if (Error Err = initRelocations(RelSec, *Rels))
1845           return Err;
1846       } else {
1847         Expected<typename ELFFile<ELFT>::Elf_Rela_Range> Relas =
1848             ElfFile.relas(*Shdr);
1849         if (!Relas)
1850           return Relas.takeError();
1851 
1852         if (Error Err = initRelocations(RelSec, *Relas))
1853           return Err;
1854       }
1855     } else if (auto GroupSec = dyn_cast<GroupSection>(&Sec)) {
1856       if (Error Err = initGroupSection(GroupSec))
1857         return Err;
1858     }
1859   }
1860 
1861   return Error::success();
1862 }
1863 
1864 template <class ELFT> Error ELFBuilder<ELFT>::build(bool EnsureSymtab) {
1865   if (Error E = readSectionHeaders())
1866     return E;
1867   if (Error E = findEhdrOffset())
1868     return E;
1869 
1870   // The ELFFile whose ELF headers and program headers are copied into the
1871   // output file. Normally the same as ElfFile, but if we're extracting a
1872   // loadable partition it will point to the partition's headers.
1873   Expected<ELFFile<ELFT>> HeadersFile = ELFFile<ELFT>::create(toStringRef(
1874       {ElfFile.base() + EhdrOffset, ElfFile.getBufSize() - EhdrOffset}));
1875   if (!HeadersFile)
1876     return HeadersFile.takeError();
1877 
1878   const typename ELFFile<ELFT>::Elf_Ehdr &Ehdr = HeadersFile->getHeader();
1879   Obj.Is64Bits = Ehdr.e_ident[EI_CLASS] == ELFCLASS64;
1880   Obj.OSABI = Ehdr.e_ident[EI_OSABI];
1881   Obj.ABIVersion = Ehdr.e_ident[EI_ABIVERSION];
1882   Obj.Type = Ehdr.e_type;
1883   Obj.Machine = Ehdr.e_machine;
1884   Obj.Version = Ehdr.e_version;
1885   Obj.Entry = Ehdr.e_entry;
1886   Obj.Flags = Ehdr.e_flags;
1887 
1888   if (Error E = readSections(EnsureSymtab))
1889     return E;
1890   return readProgramHeaders(*HeadersFile);
1891 }
1892 
1893 Writer::~Writer() = default;
1894 
1895 Reader::~Reader() = default;
1896 
1897 Expected<std::unique_ptr<Object>>
1898 BinaryReader::create(bool /*EnsureSymtab*/) const {
1899   return BinaryELFBuilder(MemBuf, NewSymbolVisibility).build();
1900 }
1901 
1902 Expected<std::vector<IHexRecord>> IHexReader::parse() const {
1903   SmallVector<StringRef, 16> Lines;
1904   std::vector<IHexRecord> Records;
1905   bool HasSections = false;
1906 
1907   MemBuf->getBuffer().split(Lines, '\n');
1908   Records.reserve(Lines.size());
1909   for (size_t LineNo = 1; LineNo <= Lines.size(); ++LineNo) {
1910     StringRef Line = Lines[LineNo - 1].trim();
1911     if (Line.empty())
1912       continue;
1913 
1914     Expected<IHexRecord> R = IHexRecord::parse(Line);
1915     if (!R)
1916       return parseError(LineNo, R.takeError());
1917     if (R->Type == IHexRecord::EndOfFile)
1918       break;
1919     HasSections |= (R->Type == IHexRecord::Data);
1920     Records.push_back(*R);
1921   }
1922   if (!HasSections)
1923     return parseError(-1U, "no sections");
1924 
1925   return std::move(Records);
1926 }
1927 
1928 Expected<std::unique_ptr<Object>>
1929 IHexReader::create(bool /*EnsureSymtab*/) const {
1930   Expected<std::vector<IHexRecord>> Records = parse();
1931   if (!Records)
1932     return Records.takeError();
1933 
1934   return IHexELFBuilder(*Records).build();
1935 }
1936 
1937 Expected<std::unique_ptr<Object>> ELFReader::create(bool EnsureSymtab) const {
1938   auto Obj = std::make_unique<Object>();
1939   if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Bin)) {
1940     ELFBuilder<ELF32LE> Builder(*O, *Obj, ExtractPartition);
1941     if (Error Err = Builder.build(EnsureSymtab))
1942       return std::move(Err);
1943     return std::move(Obj);
1944   } else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Bin)) {
1945     ELFBuilder<ELF64LE> Builder(*O, *Obj, ExtractPartition);
1946     if (Error Err = Builder.build(EnsureSymtab))
1947       return std::move(Err);
1948     return std::move(Obj);
1949   } else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Bin)) {
1950     ELFBuilder<ELF32BE> Builder(*O, *Obj, ExtractPartition);
1951     if (Error Err = Builder.build(EnsureSymtab))
1952       return std::move(Err);
1953     return std::move(Obj);
1954   } else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Bin)) {
1955     ELFBuilder<ELF64BE> Builder(*O, *Obj, ExtractPartition);
1956     if (Error Err = Builder.build(EnsureSymtab))
1957       return std::move(Err);
1958     return std::move(Obj);
1959   }
1960   return createStringError(errc::invalid_argument, "invalid file type");
1961 }
1962 
1963 template <class ELFT> void ELFWriter<ELFT>::writeEhdr() {
1964   Elf_Ehdr &Ehdr = *reinterpret_cast<Elf_Ehdr *>(Buf->getBufferStart());
1965   std::fill(Ehdr.e_ident, Ehdr.e_ident + 16, 0);
1966   Ehdr.e_ident[EI_MAG0] = 0x7f;
1967   Ehdr.e_ident[EI_MAG1] = 'E';
1968   Ehdr.e_ident[EI_MAG2] = 'L';
1969   Ehdr.e_ident[EI_MAG3] = 'F';
1970   Ehdr.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32;
1971   Ehdr.e_ident[EI_DATA] =
1972       ELFT::TargetEndianness == support::big ? ELFDATA2MSB : ELFDATA2LSB;
1973   Ehdr.e_ident[EI_VERSION] = EV_CURRENT;
1974   Ehdr.e_ident[EI_OSABI] = Obj.OSABI;
1975   Ehdr.e_ident[EI_ABIVERSION] = Obj.ABIVersion;
1976 
1977   Ehdr.e_type = Obj.Type;
1978   Ehdr.e_machine = Obj.Machine;
1979   Ehdr.e_version = Obj.Version;
1980   Ehdr.e_entry = Obj.Entry;
1981   // We have to use the fully-qualified name llvm::size
1982   // since some compilers complain on ambiguous resolution.
1983   Ehdr.e_phnum = llvm::size(Obj.segments());
1984   Ehdr.e_phoff = (Ehdr.e_phnum != 0) ? Obj.ProgramHdrSegment.Offset : 0;
1985   Ehdr.e_phentsize = (Ehdr.e_phnum != 0) ? sizeof(Elf_Phdr) : 0;
1986   Ehdr.e_flags = Obj.Flags;
1987   Ehdr.e_ehsize = sizeof(Elf_Ehdr);
1988   if (WriteSectionHeaders && Obj.sections().size() != 0) {
1989     Ehdr.e_shentsize = sizeof(Elf_Shdr);
1990     Ehdr.e_shoff = Obj.SHOff;
1991     // """
1992     // If the number of sections is greater than or equal to
1993     // SHN_LORESERVE (0xff00), this member has the value zero and the actual
1994     // number of section header table entries is contained in the sh_size field
1995     // of the section header at index 0.
1996     // """
1997     auto Shnum = Obj.sections().size() + 1;
1998     if (Shnum >= SHN_LORESERVE)
1999       Ehdr.e_shnum = 0;
2000     else
2001       Ehdr.e_shnum = Shnum;
2002     // """
2003     // If the section name string table section index is greater than or equal
2004     // to SHN_LORESERVE (0xff00), this member has the value SHN_XINDEX (0xffff)
2005     // and the actual index of the section name string table section is
2006     // contained in the sh_link field of the section header at index 0.
2007     // """
2008     if (Obj.SectionNames->Index >= SHN_LORESERVE)
2009       Ehdr.e_shstrndx = SHN_XINDEX;
2010     else
2011       Ehdr.e_shstrndx = Obj.SectionNames->Index;
2012   } else {
2013     Ehdr.e_shentsize = 0;
2014     Ehdr.e_shoff = 0;
2015     Ehdr.e_shnum = 0;
2016     Ehdr.e_shstrndx = 0;
2017   }
2018 }
2019 
2020 template <class ELFT> void ELFWriter<ELFT>::writePhdrs() {
2021   for (auto &Seg : Obj.segments())
2022     writePhdr(Seg);
2023 }
2024 
2025 template <class ELFT> void ELFWriter<ELFT>::writeShdrs() {
2026   // This reference serves to write the dummy section header at the begining
2027   // of the file. It is not used for anything else
2028   Elf_Shdr &Shdr =
2029       *reinterpret_cast<Elf_Shdr *>(Buf->getBufferStart() + Obj.SHOff);
2030   Shdr.sh_name = 0;
2031   Shdr.sh_type = SHT_NULL;
2032   Shdr.sh_flags = 0;
2033   Shdr.sh_addr = 0;
2034   Shdr.sh_offset = 0;
2035   // See writeEhdr for why we do this.
2036   uint64_t Shnum = Obj.sections().size() + 1;
2037   if (Shnum >= SHN_LORESERVE)
2038     Shdr.sh_size = Shnum;
2039   else
2040     Shdr.sh_size = 0;
2041   // See writeEhdr for why we do this.
2042   if (Obj.SectionNames != nullptr && Obj.SectionNames->Index >= SHN_LORESERVE)
2043     Shdr.sh_link = Obj.SectionNames->Index;
2044   else
2045     Shdr.sh_link = 0;
2046   Shdr.sh_info = 0;
2047   Shdr.sh_addralign = 0;
2048   Shdr.sh_entsize = 0;
2049 
2050   for (SectionBase &Sec : Obj.sections())
2051     writeShdr(Sec);
2052 }
2053 
2054 template <class ELFT> Error ELFWriter<ELFT>::writeSectionData() {
2055   for (SectionBase &Sec : Obj.sections())
2056     // Segments are responsible for writing their contents, so only write the
2057     // section data if the section is not in a segment. Note that this renders
2058     // sections in segments effectively immutable.
2059     if (Sec.ParentSegment == nullptr)
2060       if (Error Err = Sec.accept(*SecWriter))
2061         return Err;
2062 
2063   return Error::success();
2064 }
2065 
2066 template <class ELFT> void ELFWriter<ELFT>::writeSegmentData() {
2067   for (Segment &Seg : Obj.segments()) {
2068     size_t Size = std::min<size_t>(Seg.FileSize, Seg.getContents().size());
2069     std::memcpy(Buf->getBufferStart() + Seg.Offset, Seg.getContents().data(),
2070                 Size);
2071   }
2072 
2073   for (auto it : Obj.getUpdatedSections()) {
2074     SectionBase *Sec = it.first;
2075     ArrayRef<uint8_t> Data = it.second;
2076 
2077     auto *Parent = Sec->ParentSegment;
2078     assert(Parent && "This section should've been part of a segment.");
2079     uint64_t Offset =
2080         Sec->OriginalOffset - Parent->OriginalOffset + Parent->Offset;
2081     llvm::copy(Data, Buf->getBufferStart() + Offset);
2082   }
2083 
2084   // Iterate over removed sections and overwrite their old data with zeroes.
2085   for (auto &Sec : Obj.removedSections()) {
2086     Segment *Parent = Sec.ParentSegment;
2087     if (Parent == nullptr || Sec.Type == SHT_NOBITS || Sec.Size == 0)
2088       continue;
2089     uint64_t Offset =
2090         Sec.OriginalOffset - Parent->OriginalOffset + Parent->Offset;
2091     std::memset(Buf->getBufferStart() + Offset, 0, Sec.Size);
2092   }
2093 }
2094 
2095 template <class ELFT>
2096 ELFWriter<ELFT>::ELFWriter(Object &Obj, raw_ostream &Buf, bool WSH,
2097                            bool OnlyKeepDebug)
2098     : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs),
2099       OnlyKeepDebug(OnlyKeepDebug) {}
2100 
2101 Error Object::updateSection(StringRef Name, ArrayRef<uint8_t> Data) {
2102   auto It = llvm::find_if(Sections,
2103                           [&](const SecPtr &Sec) { return Sec->Name == Name; });
2104   if (It == Sections.end())
2105     return createStringError(errc::invalid_argument, "section '%s' not found",
2106                              Name.str().c_str());
2107 
2108   auto *OldSec = It->get();
2109   if (!OldSec->hasContents())
2110     return createStringError(
2111         errc::invalid_argument,
2112         "section '%s' cannot be updated because it does not have contents",
2113         Name.str().c_str());
2114 
2115   if (Data.size() > OldSec->Size && OldSec->ParentSegment)
2116     return createStringError(errc::invalid_argument,
2117                              "cannot fit data of size %zu into section '%s' "
2118                              "with size %" PRIu64 " that is part of a segment",
2119                              Data.size(), Name.str().c_str(), OldSec->Size);
2120 
2121   if (!OldSec->ParentSegment) {
2122     *It = std::make_unique<OwnedDataSection>(*OldSec, Data);
2123   } else {
2124     // The segment writer will be in charge of updating these contents.
2125     OldSec->Size = Data.size();
2126     UpdatedSections[OldSec] = Data;
2127   }
2128 
2129   return Error::success();
2130 }
2131 
2132 Error Object::removeSections(
2133     bool AllowBrokenLinks, std::function<bool(const SectionBase &)> ToRemove) {
2134 
2135   auto Iter = std::stable_partition(
2136       std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) {
2137         if (ToRemove(*Sec))
2138           return false;
2139         if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) {
2140           if (auto ToRelSec = RelSec->getSection())
2141             return !ToRemove(*ToRelSec);
2142         }
2143         return true;
2144       });
2145   if (SymbolTable != nullptr && ToRemove(*SymbolTable))
2146     SymbolTable = nullptr;
2147   if (SectionNames != nullptr && ToRemove(*SectionNames))
2148     SectionNames = nullptr;
2149   if (SectionIndexTable != nullptr && ToRemove(*SectionIndexTable))
2150     SectionIndexTable = nullptr;
2151   // Now make sure there are no remaining references to the sections that will
2152   // be removed. Sometimes it is impossible to remove a reference so we emit
2153   // an error here instead.
2154   std::unordered_set<const SectionBase *> RemoveSections;
2155   RemoveSections.reserve(std::distance(Iter, std::end(Sections)));
2156   for (auto &RemoveSec : make_range(Iter, std::end(Sections))) {
2157     for (auto &Segment : Segments)
2158       Segment->removeSection(RemoveSec.get());
2159     RemoveSec->onRemove();
2160     RemoveSections.insert(RemoveSec.get());
2161   }
2162 
2163   // For each section that remains alive, we want to remove the dead references.
2164   // This either might update the content of the section (e.g. remove symbols
2165   // from symbol table that belongs to removed section) or trigger an error if
2166   // a live section critically depends on a section being removed somehow
2167   // (e.g. the removed section is referenced by a relocation).
2168   for (auto &KeepSec : make_range(std::begin(Sections), Iter)) {
2169     if (Error E = KeepSec->removeSectionReferences(
2170             AllowBrokenLinks, [&RemoveSections](const SectionBase *Sec) {
2171               return RemoveSections.find(Sec) != RemoveSections.end();
2172             }))
2173       return E;
2174   }
2175 
2176   // Transfer removed sections into the Object RemovedSections container for use
2177   // later.
2178   std::move(Iter, Sections.end(), std::back_inserter(RemovedSections));
2179   // Now finally get rid of them all together.
2180   Sections.erase(Iter, std::end(Sections));
2181   return Error::success();
2182 }
2183 
2184 Error Object::replaceSections(
2185     const DenseMap<SectionBase *, SectionBase *> &FromTo) {
2186   auto SectionIndexLess = [](const SecPtr &Lhs, const SecPtr &Rhs) {
2187     return Lhs->Index < Rhs->Index;
2188   };
2189   assert(llvm::is_sorted(Sections, SectionIndexLess) &&
2190          "Sections are expected to be sorted by Index");
2191   // Set indices of new sections so that they can be later sorted into positions
2192   // of removed ones.
2193   for (auto &I : FromTo)
2194     I.second->Index = I.first->Index;
2195 
2196   // Notify all sections about the replacement.
2197   for (auto &Sec : Sections)
2198     Sec->replaceSectionReferences(FromTo);
2199 
2200   if (Error E = removeSections(
2201           /*AllowBrokenLinks=*/false,
2202           [=](const SectionBase &Sec) { return FromTo.count(&Sec) > 0; }))
2203     return E;
2204   llvm::sort(Sections, SectionIndexLess);
2205   return Error::success();
2206 }
2207 
2208 Error Object::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
2209   if (SymbolTable)
2210     for (const SecPtr &Sec : Sections)
2211       if (Error E = Sec->removeSymbols(ToRemove))
2212         return E;
2213   return Error::success();
2214 }
2215 
2216 Error Object::addNewSymbolTable() {
2217   assert(!SymbolTable && "Object must not has a SymbolTable.");
2218 
2219   // Reuse an existing SHT_STRTAB section if it exists.
2220   StringTableSection *StrTab = nullptr;
2221   for (SectionBase &Sec : sections()) {
2222     if (Sec.Type == ELF::SHT_STRTAB && !(Sec.Flags & SHF_ALLOC)) {
2223       StrTab = static_cast<StringTableSection *>(&Sec);
2224 
2225       // Prefer a string table that is not the section header string table, if
2226       // such a table exists.
2227       if (SectionNames != &Sec)
2228         break;
2229     }
2230   }
2231   if (!StrTab)
2232     StrTab = &addSection<StringTableSection>();
2233 
2234   SymbolTableSection &SymTab = addSection<SymbolTableSection>();
2235   SymTab.Name = ".symtab";
2236   SymTab.Link = StrTab->Index;
2237   if (Error Err = SymTab.initialize(sections()))
2238     return Err;
2239   SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0);
2240 
2241   SymbolTable = &SymTab;
2242 
2243   return Error::success();
2244 }
2245 
2246 // Orders segments such that if x = y->ParentSegment then y comes before x.
2247 static void orderSegments(std::vector<Segment *> &Segments) {
2248   llvm::stable_sort(Segments, compareSegmentsByOffset);
2249 }
2250 
2251 // This function finds a consistent layout for a list of segments starting from
2252 // an Offset. It assumes that Segments have been sorted by orderSegments and
2253 // returns an Offset one past the end of the last segment.
2254 static uint64_t layoutSegments(std::vector<Segment *> &Segments,
2255                                uint64_t Offset) {
2256   assert(llvm::is_sorted(Segments, compareSegmentsByOffset));
2257   // The only way a segment should move is if a section was between two
2258   // segments and that section was removed. If that section isn't in a segment
2259   // then it's acceptable, but not ideal, to simply move it to after the
2260   // segments. So we can simply layout segments one after the other accounting
2261   // for alignment.
2262   for (Segment *Seg : Segments) {
2263     // We assume that segments have been ordered by OriginalOffset and Index
2264     // such that a parent segment will always come before a child segment in
2265     // OrderedSegments. This means that the Offset of the ParentSegment should
2266     // already be set and we can set our offset relative to it.
2267     if (Seg->ParentSegment != nullptr) {
2268       Segment *Parent = Seg->ParentSegment;
2269       Seg->Offset =
2270           Parent->Offset + Seg->OriginalOffset - Parent->OriginalOffset;
2271     } else {
2272       Seg->Offset =
2273           alignTo(Offset, std::max<uint64_t>(Seg->Align, 1), Seg->VAddr);
2274     }
2275     Offset = std::max(Offset, Seg->Offset + Seg->FileSize);
2276   }
2277   return Offset;
2278 }
2279 
2280 // This function finds a consistent layout for a list of sections. It assumes
2281 // that the ->ParentSegment of each section has already been laid out. The
2282 // supplied starting Offset is used for the starting offset of any section that
2283 // does not have a ParentSegment. It returns either the offset given if all
2284 // sections had a ParentSegment or an offset one past the last section if there
2285 // was a section that didn't have a ParentSegment.
2286 template <class Range>
2287 static uint64_t layoutSections(Range Sections, uint64_t Offset) {
2288   // Now the offset of every segment has been set we can assign the offsets
2289   // of each section. For sections that are covered by a segment we should use
2290   // the segment's original offset and the section's original offset to compute
2291   // the offset from the start of the segment. Using the offset from the start
2292   // of the segment we can assign a new offset to the section. For sections not
2293   // covered by segments we can just bump Offset to the next valid location.
2294   // While it is not necessary, layout the sections in the order based on their
2295   // original offsets to resemble the input file as close as possible.
2296   std::vector<SectionBase *> OutOfSegmentSections;
2297   uint32_t Index = 1;
2298   for (auto &Sec : Sections) {
2299     Sec.Index = Index++;
2300     if (Sec.ParentSegment != nullptr) {
2301       auto Segment = *Sec.ParentSegment;
2302       Sec.Offset =
2303           Segment.Offset + (Sec.OriginalOffset - Segment.OriginalOffset);
2304     } else
2305       OutOfSegmentSections.push_back(&Sec);
2306   }
2307 
2308   llvm::stable_sort(OutOfSegmentSections,
2309                     [](const SectionBase *Lhs, const SectionBase *Rhs) {
2310                       return Lhs->OriginalOffset < Rhs->OriginalOffset;
2311                     });
2312   for (auto *Sec : OutOfSegmentSections) {
2313     Offset = alignTo(Offset, Sec->Align == 0 ? 1 : Sec->Align);
2314     Sec->Offset = Offset;
2315     if (Sec->Type != SHT_NOBITS)
2316       Offset += Sec->Size;
2317   }
2318   return Offset;
2319 }
2320 
2321 // Rewrite sh_offset after some sections are changed to SHT_NOBITS and thus
2322 // occupy no space in the file.
2323 static uint64_t layoutSectionsForOnlyKeepDebug(Object &Obj, uint64_t Off) {
2324   // The layout algorithm requires the sections to be handled in the order of
2325   // their offsets in the input file, at least inside segments.
2326   std::vector<SectionBase *> Sections;
2327   Sections.reserve(Obj.sections().size());
2328   uint32_t Index = 1;
2329   for (auto &Sec : Obj.sections()) {
2330     Sec.Index = Index++;
2331     Sections.push_back(&Sec);
2332   }
2333   llvm::stable_sort(Sections,
2334                     [](const SectionBase *Lhs, const SectionBase *Rhs) {
2335                       return Lhs->OriginalOffset < Rhs->OriginalOffset;
2336                     });
2337 
2338   for (auto *Sec : Sections) {
2339     auto *FirstSec = Sec->ParentSegment && Sec->ParentSegment->Type == PT_LOAD
2340                          ? Sec->ParentSegment->firstSection()
2341                          : nullptr;
2342 
2343     // The first section in a PT_LOAD has to have congruent offset and address
2344     // modulo the alignment, which usually equals the maximum page size.
2345     if (FirstSec && FirstSec == Sec)
2346       Off = alignTo(Off, Sec->ParentSegment->Align, Sec->Addr);
2347 
2348     // sh_offset is not significant for SHT_NOBITS sections, but the congruence
2349     // rule must be followed if it is the first section in a PT_LOAD. Do not
2350     // advance Off.
2351     if (Sec->Type == SHT_NOBITS) {
2352       Sec->Offset = Off;
2353       continue;
2354     }
2355 
2356     if (!FirstSec) {
2357       // FirstSec being nullptr generally means that Sec does not have the
2358       // SHF_ALLOC flag.
2359       Off = Sec->Align ? alignTo(Off, Sec->Align) : Off;
2360     } else if (FirstSec != Sec) {
2361       // The offset is relative to the first section in the PT_LOAD segment. Use
2362       // sh_offset for non-SHF_ALLOC sections.
2363       Off = Sec->OriginalOffset - FirstSec->OriginalOffset + FirstSec->Offset;
2364     }
2365     Sec->Offset = Off;
2366     Off += Sec->Size;
2367   }
2368   return Off;
2369 }
2370 
2371 // Rewrite p_offset and p_filesz of non-PT_PHDR segments after sh_offset values
2372 // have been updated.
2373 static uint64_t layoutSegmentsForOnlyKeepDebug(std::vector<Segment *> &Segments,
2374                                                uint64_t HdrEnd) {
2375   uint64_t MaxOffset = 0;
2376   for (Segment *Seg : Segments) {
2377     if (Seg->Type == PT_PHDR)
2378       continue;
2379 
2380     // The segment offset is generally the offset of the first section.
2381     //
2382     // For a segment containing no section (see sectionWithinSegment), if it has
2383     // a parent segment, copy the parent segment's offset field. This works for
2384     // empty PT_TLS. If no parent segment, use 0: the segment is not useful for
2385     // debugging anyway.
2386     const SectionBase *FirstSec = Seg->firstSection();
2387     uint64_t Offset =
2388         FirstSec ? FirstSec->Offset
2389                  : (Seg->ParentSegment ? Seg->ParentSegment->Offset : 0);
2390     uint64_t FileSize = 0;
2391     for (const SectionBase *Sec : Seg->Sections) {
2392       uint64_t Size = Sec->Type == SHT_NOBITS ? 0 : Sec->Size;
2393       if (Sec->Offset + Size > Offset)
2394         FileSize = std::max(FileSize, Sec->Offset + Size - Offset);
2395     }
2396 
2397     // If the segment includes EHDR and program headers, don't make it smaller
2398     // than the headers.
2399     if (Seg->Offset < HdrEnd && HdrEnd <= Seg->Offset + Seg->FileSize) {
2400       FileSize += Offset - Seg->Offset;
2401       Offset = Seg->Offset;
2402       FileSize = std::max(FileSize, HdrEnd - Offset);
2403     }
2404 
2405     Seg->Offset = Offset;
2406     Seg->FileSize = FileSize;
2407     MaxOffset = std::max(MaxOffset, Offset + FileSize);
2408   }
2409   return MaxOffset;
2410 }
2411 
2412 template <class ELFT> void ELFWriter<ELFT>::initEhdrSegment() {
2413   Segment &ElfHdr = Obj.ElfHdrSegment;
2414   ElfHdr.Type = PT_PHDR;
2415   ElfHdr.Flags = 0;
2416   ElfHdr.VAddr = 0;
2417   ElfHdr.PAddr = 0;
2418   ElfHdr.FileSize = ElfHdr.MemSize = sizeof(Elf_Ehdr);
2419   ElfHdr.Align = 0;
2420 }
2421 
2422 template <class ELFT> void ELFWriter<ELFT>::assignOffsets() {
2423   // We need a temporary list of segments that has a special order to it
2424   // so that we know that anytime ->ParentSegment is set that segment has
2425   // already had its offset properly set.
2426   std::vector<Segment *> OrderedSegments;
2427   for (Segment &Segment : Obj.segments())
2428     OrderedSegments.push_back(&Segment);
2429   OrderedSegments.push_back(&Obj.ElfHdrSegment);
2430   OrderedSegments.push_back(&Obj.ProgramHdrSegment);
2431   orderSegments(OrderedSegments);
2432 
2433   uint64_t Offset;
2434   if (OnlyKeepDebug) {
2435     // For --only-keep-debug, the sections that did not preserve contents were
2436     // changed to SHT_NOBITS. We now rewrite sh_offset fields of sections, and
2437     // then rewrite p_offset/p_filesz of program headers.
2438     uint64_t HdrEnd =
2439         sizeof(Elf_Ehdr) + llvm::size(Obj.segments()) * sizeof(Elf_Phdr);
2440     Offset = layoutSectionsForOnlyKeepDebug(Obj, HdrEnd);
2441     Offset = std::max(Offset,
2442                       layoutSegmentsForOnlyKeepDebug(OrderedSegments, HdrEnd));
2443   } else {
2444     // Offset is used as the start offset of the first segment to be laid out.
2445     // Since the ELF Header (ElfHdrSegment) must be at the start of the file,
2446     // we start at offset 0.
2447     Offset = layoutSegments(OrderedSegments, 0);
2448     Offset = layoutSections(Obj.sections(), Offset);
2449   }
2450   // If we need to write the section header table out then we need to align the
2451   // Offset so that SHOffset is valid.
2452   if (WriteSectionHeaders)
2453     Offset = alignTo(Offset, sizeof(Elf_Addr));
2454   Obj.SHOff = Offset;
2455 }
2456 
2457 template <class ELFT> size_t ELFWriter<ELFT>::totalSize() const {
2458   // We already have the section header offset so we can calculate the total
2459   // size by just adding up the size of each section header.
2460   if (!WriteSectionHeaders)
2461     return Obj.SHOff;
2462   size_t ShdrCount = Obj.sections().size() + 1; // Includes null shdr.
2463   return Obj.SHOff + ShdrCount * sizeof(Elf_Shdr);
2464 }
2465 
2466 template <class ELFT> Error ELFWriter<ELFT>::write() {
2467   // Segment data must be written first, so that the ELF header and program
2468   // header tables can overwrite it, if covered by a segment.
2469   writeSegmentData();
2470   writeEhdr();
2471   writePhdrs();
2472   if (Error E = writeSectionData())
2473     return E;
2474   if (WriteSectionHeaders)
2475     writeShdrs();
2476 
2477   // TODO: Implement direct writing to the output stream (without intermediate
2478   // memory buffer Buf).
2479   Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2480   return Error::success();
2481 }
2482 
2483 static Error removeUnneededSections(Object &Obj) {
2484   // We can remove an empty symbol table from non-relocatable objects.
2485   // Relocatable objects typically have relocation sections whose
2486   // sh_link field points to .symtab, so we can't remove .symtab
2487   // even if it is empty.
2488   if (Obj.isRelocatable() || Obj.SymbolTable == nullptr ||
2489       !Obj.SymbolTable->empty())
2490     return Error::success();
2491 
2492   // .strtab can be used for section names. In such a case we shouldn't
2493   // remove it.
2494   auto *StrTab = Obj.SymbolTable->getStrTab() == Obj.SectionNames
2495                      ? nullptr
2496                      : Obj.SymbolTable->getStrTab();
2497   return Obj.removeSections(false, [&](const SectionBase &Sec) {
2498     return &Sec == Obj.SymbolTable || &Sec == StrTab;
2499   });
2500 }
2501 
2502 template <class ELFT> Error ELFWriter<ELFT>::finalize() {
2503   // It could happen that SectionNames has been removed and yet the user wants
2504   // a section header table output. We need to throw an error if a user tries
2505   // to do that.
2506   if (Obj.SectionNames == nullptr && WriteSectionHeaders)
2507     return createStringError(llvm::errc::invalid_argument,
2508                              "cannot write section header table because "
2509                              "section header string table was removed");
2510 
2511   if (Error E = removeUnneededSections(Obj))
2512     return E;
2513 
2514   // We need to assign indexes before we perform layout because we need to know
2515   // if we need large indexes or not. We can assign indexes first and check as
2516   // we go to see if we will actully need large indexes.
2517   bool NeedsLargeIndexes = false;
2518   if (Obj.sections().size() >= SHN_LORESERVE) {
2519     SectionTableRef Sections = Obj.sections();
2520     // Sections doesn't include the null section header, so account for this
2521     // when skipping the first N sections.
2522     NeedsLargeIndexes =
2523         any_of(drop_begin(Sections, SHN_LORESERVE - 1),
2524                [](const SectionBase &Sec) { return Sec.HasSymbol; });
2525     // TODO: handle case where only one section needs the large index table but
2526     // only needs it because the large index table hasn't been removed yet.
2527   }
2528 
2529   if (NeedsLargeIndexes) {
2530     // This means we definitely need to have a section index table but if we
2531     // already have one then we should use it instead of making a new one.
2532     if (Obj.SymbolTable != nullptr && Obj.SectionIndexTable == nullptr) {
2533       // Addition of a section to the end does not invalidate the indexes of
2534       // other sections and assigns the correct index to the new section.
2535       auto &Shndx = Obj.addSection<SectionIndexSection>();
2536       Obj.SymbolTable->setShndxTable(&Shndx);
2537       Shndx.setSymTab(Obj.SymbolTable);
2538     }
2539   } else {
2540     // Since we don't need SectionIndexTable we should remove it and all
2541     // references to it.
2542     if (Obj.SectionIndexTable != nullptr) {
2543       // We do not support sections referring to the section index table.
2544       if (Error E = Obj.removeSections(false /*AllowBrokenLinks*/,
2545                                        [this](const SectionBase &Sec) {
2546                                          return &Sec == Obj.SectionIndexTable;
2547                                        }))
2548         return E;
2549     }
2550   }
2551 
2552   // Make sure we add the names of all the sections. Importantly this must be
2553   // done after we decide to add or remove SectionIndexes.
2554   if (Obj.SectionNames != nullptr)
2555     for (const SectionBase &Sec : Obj.sections())
2556       Obj.SectionNames->addString(Sec.Name);
2557 
2558   initEhdrSegment();
2559 
2560   // Before we can prepare for layout the indexes need to be finalized.
2561   // Also, the output arch may not be the same as the input arch, so fix up
2562   // size-related fields before doing layout calculations.
2563   uint64_t Index = 0;
2564   auto SecSizer = std::make_unique<ELFSectionSizer<ELFT>>();
2565   for (SectionBase &Sec : Obj.sections()) {
2566     Sec.Index = Index++;
2567     if (Error Err = Sec.accept(*SecSizer))
2568       return Err;
2569   }
2570 
2571   // The symbol table does not update all other sections on update. For
2572   // instance, symbol names are not added as new symbols are added. This means
2573   // that some sections, like .strtab, don't yet have their final size.
2574   if (Obj.SymbolTable != nullptr)
2575     Obj.SymbolTable->prepareForLayout();
2576 
2577   // Now that all strings are added we want to finalize string table builders,
2578   // because that affects section sizes which in turn affects section offsets.
2579   for (SectionBase &Sec : Obj.sections())
2580     if (auto StrTab = dyn_cast<StringTableSection>(&Sec))
2581       StrTab->prepareForLayout();
2582 
2583   assignOffsets();
2584 
2585   // layoutSections could have modified section indexes, so we need
2586   // to fill the index table after assignOffsets.
2587   if (Obj.SymbolTable != nullptr)
2588     Obj.SymbolTable->fillShndxTable();
2589 
2590   // Finally now that all offsets and indexes have been set we can finalize any
2591   // remaining issues.
2592   uint64_t Offset = Obj.SHOff + sizeof(Elf_Shdr);
2593   for (SectionBase &Sec : Obj.sections()) {
2594     Sec.HeaderOffset = Offset;
2595     Offset += sizeof(Elf_Shdr);
2596     if (WriteSectionHeaders)
2597       Sec.NameIndex = Obj.SectionNames->findIndex(Sec.Name);
2598     Sec.finalize();
2599   }
2600 
2601   size_t TotalSize = totalSize();
2602   Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2603   if (!Buf)
2604     return createStringError(errc::not_enough_memory,
2605                              "failed to allocate memory buffer of " +
2606                                  Twine::utohexstr(TotalSize) + " bytes");
2607 
2608   SecWriter = std::make_unique<ELFSectionWriter<ELFT>>(*Buf);
2609   return Error::success();
2610 }
2611 
2612 Error BinaryWriter::write() {
2613   for (const SectionBase &Sec : Obj.allocSections())
2614     if (Error Err = Sec.accept(*SecWriter))
2615       return Err;
2616 
2617   // TODO: Implement direct writing to the output stream (without intermediate
2618   // memory buffer Buf).
2619   Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2620   return Error::success();
2621 }
2622 
2623 Error BinaryWriter::finalize() {
2624   // Compute the section LMA based on its sh_offset and the containing segment's
2625   // p_offset and p_paddr. Also compute the minimum LMA of all non-empty
2626   // sections as MinAddr. In the output, the contents between address 0 and
2627   // MinAddr will be skipped.
2628   uint64_t MinAddr = UINT64_MAX;
2629   for (SectionBase &Sec : Obj.allocSections()) {
2630     // If Sec's type is changed from SHT_NOBITS due to --set-section-flags,
2631     // Offset may not be aligned. Align it to max(Align, 1).
2632     if (Sec.ParentSegment != nullptr)
2633       Sec.Addr = alignTo(Sec.Offset - Sec.ParentSegment->Offset +
2634                              Sec.ParentSegment->PAddr,
2635                          std::max(Sec.Align, uint64_t(1)));
2636     if (Sec.Type != SHT_NOBITS && Sec.Size > 0)
2637       MinAddr = std::min(MinAddr, Sec.Addr);
2638   }
2639 
2640   // Now that every section has been laid out we just need to compute the total
2641   // file size. This might not be the same as the offset returned by
2642   // layoutSections, because we want to truncate the last segment to the end of
2643   // its last non-empty section, to match GNU objcopy's behaviour.
2644   TotalSize = 0;
2645   for (SectionBase &Sec : Obj.allocSections())
2646     if (Sec.Type != SHT_NOBITS && Sec.Size > 0) {
2647       Sec.Offset = Sec.Addr - MinAddr;
2648       TotalSize = std::max(TotalSize, Sec.Offset + Sec.Size);
2649     }
2650 
2651   Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2652   if (!Buf)
2653     return createStringError(errc::not_enough_memory,
2654                              "failed to allocate memory buffer of " +
2655                                  Twine::utohexstr(TotalSize) + " bytes");
2656   SecWriter = std::make_unique<BinarySectionWriter>(*Buf);
2657   return Error::success();
2658 }
2659 
2660 bool IHexWriter::SectionCompare::operator()(const SectionBase *Lhs,
2661                                             const SectionBase *Rhs) const {
2662   return (sectionPhysicalAddr(Lhs) & 0xFFFFFFFFU) <
2663          (sectionPhysicalAddr(Rhs) & 0xFFFFFFFFU);
2664 }
2665 
2666 uint64_t IHexWriter::writeEntryPointRecord(uint8_t *Buf) {
2667   IHexLineData HexData;
2668   uint8_t Data[4] = {};
2669   // We don't write entry point record if entry is zero.
2670   if (Obj.Entry == 0)
2671     return 0;
2672 
2673   if (Obj.Entry <= 0xFFFFFU) {
2674     Data[0] = ((Obj.Entry & 0xF0000U) >> 12) & 0xFF;
2675     support::endian::write(&Data[2], static_cast<uint16_t>(Obj.Entry),
2676                            support::big);
2677     HexData = IHexRecord::getLine(IHexRecord::StartAddr80x86, 0, Data);
2678   } else {
2679     support::endian::write(Data, static_cast<uint32_t>(Obj.Entry),
2680                            support::big);
2681     HexData = IHexRecord::getLine(IHexRecord::StartAddr, 0, Data);
2682   }
2683   memcpy(Buf, HexData.data(), HexData.size());
2684   return HexData.size();
2685 }
2686 
2687 uint64_t IHexWriter::writeEndOfFileRecord(uint8_t *Buf) {
2688   IHexLineData HexData = IHexRecord::getLine(IHexRecord::EndOfFile, 0, {});
2689   memcpy(Buf, HexData.data(), HexData.size());
2690   return HexData.size();
2691 }
2692 
2693 Error IHexWriter::write() {
2694   IHexSectionWriter Writer(*Buf);
2695   // Write sections.
2696   for (const SectionBase *Sec : Sections)
2697     if (Error Err = Sec->accept(Writer))
2698       return Err;
2699 
2700   uint64_t Offset = Writer.getBufferOffset();
2701   // Write entry point address.
2702   Offset += writeEntryPointRecord(
2703       reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset);
2704   // Write EOF.
2705   Offset += writeEndOfFileRecord(
2706       reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset);
2707   assert(Offset == TotalSize);
2708 
2709   // TODO: Implement direct writing to the output stream (without intermediate
2710   // memory buffer Buf).
2711   Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2712   return Error::success();
2713 }
2714 
2715 Error IHexWriter::checkSection(const SectionBase &Sec) {
2716   uint64_t Addr = sectionPhysicalAddr(&Sec);
2717   if (addressOverflows32bit(Addr) || addressOverflows32bit(Addr + Sec.Size - 1))
2718     return createStringError(
2719         errc::invalid_argument,
2720         "Section '%s' address range [0x%llx, 0x%llx] is not 32 bit",
2721         Sec.Name.c_str(), Addr, Addr + Sec.Size - 1);
2722   return Error::success();
2723 }
2724 
2725 Error IHexWriter::finalize() {
2726   // We can't write 64-bit addresses.
2727   if (addressOverflows32bit(Obj.Entry))
2728     return createStringError(errc::invalid_argument,
2729                              "Entry point address 0x%llx overflows 32 bits",
2730                              Obj.Entry);
2731 
2732   for (const SectionBase &Sec : Obj.sections())
2733     if ((Sec.Flags & ELF::SHF_ALLOC) && Sec.Type != ELF::SHT_NOBITS &&
2734         Sec.Size > 0) {
2735       if (Error E = checkSection(Sec))
2736         return E;
2737       Sections.insert(&Sec);
2738     }
2739 
2740   std::unique_ptr<WritableMemoryBuffer> EmptyBuffer =
2741       WritableMemoryBuffer::getNewMemBuffer(0);
2742   if (!EmptyBuffer)
2743     return createStringError(errc::not_enough_memory,
2744                              "failed to allocate memory buffer of 0 bytes");
2745 
2746   IHexSectionWriterBase LengthCalc(*EmptyBuffer);
2747   for (const SectionBase *Sec : Sections)
2748     if (Error Err = Sec->accept(LengthCalc))
2749       return Err;
2750 
2751   // We need space to write section records + StartAddress record
2752   // (if start adress is not zero) + EndOfFile record.
2753   TotalSize = LengthCalc.getBufferOffset() +
2754               (Obj.Entry ? IHexRecord::getLineLength(4) : 0) +
2755               IHexRecord::getLineLength(0);
2756 
2757   Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2758   if (!Buf)
2759     return createStringError(errc::not_enough_memory,
2760                              "failed to allocate memory buffer of " +
2761                                  Twine::utohexstr(TotalSize) + " bytes");
2762 
2763   return Error::success();
2764 }
2765 
2766 namespace llvm {
2767 namespace objcopy {
2768 namespace elf {
2769 
2770 template class ELFBuilder<ELF64LE>;
2771 template class ELFBuilder<ELF64BE>;
2772 template class ELFBuilder<ELF32LE>;
2773 template class ELFBuilder<ELF32BE>;
2774 
2775 template class ELFWriter<ELF64LE>;
2776 template class ELFWriter<ELF64BE>;
2777 template class ELFWriter<ELF32LE>;
2778 template class ELFWriter<ELF32BE>;
2779 
2780 } // end namespace elf
2781 } // end namespace objcopy
2782 } // end namespace llvm
2783