xref: /freebsd/contrib/llvm-project/llvm/lib/ObjCopy/ELF/ELFObject.cpp (revision 0fca6ea1d4eea4c934cfff25ac9ee8ad6fe95583)
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/MCELFExtras.h"
17 #include "llvm/MC/MCTargetOptions.h"
18 #include "llvm/Object/ELF.h"
19 #include "llvm/Object/ELFObjectFile.h"
20 #include "llvm/Support/Compression.h"
21 #include "llvm/Support/Endian.h"
22 #include "llvm/Support/ErrorHandling.h"
23 #include "llvm/Support/FileOutputBuffer.h"
24 #include "llvm/Support/Path.h"
25 #include <algorithm>
26 #include <cstddef>
27 #include <cstdint>
28 #include <iterator>
29 #include <unordered_set>
30 #include <utility>
31 #include <vector>
32 
33 using namespace llvm;
34 using namespace llvm::ELF;
35 using namespace llvm::objcopy::elf;
36 using namespace llvm::object;
37 using namespace llvm::support;
38 
writePhdr(const Segment & Seg)39 template <class ELFT> void ELFWriter<ELFT>::writePhdr(const Segment &Seg) {
40   uint8_t *B = reinterpret_cast<uint8_t *>(Buf->getBufferStart()) +
41                Obj.ProgramHdrSegment.Offset + Seg.Index * sizeof(Elf_Phdr);
42   Elf_Phdr &Phdr = *reinterpret_cast<Elf_Phdr *>(B);
43   Phdr.p_type = Seg.Type;
44   Phdr.p_flags = Seg.Flags;
45   Phdr.p_offset = Seg.Offset;
46   Phdr.p_vaddr = Seg.VAddr;
47   Phdr.p_paddr = Seg.PAddr;
48   Phdr.p_filesz = Seg.FileSize;
49   Phdr.p_memsz = Seg.MemSize;
50   Phdr.p_align = Seg.Align;
51 }
52 
removeSectionReferences(bool,function_ref<bool (const SectionBase *)>)53 Error SectionBase::removeSectionReferences(
54     bool, function_ref<bool(const SectionBase *)>) {
55   return Error::success();
56 }
57 
removeSymbols(function_ref<bool (const Symbol &)>)58 Error SectionBase::removeSymbols(function_ref<bool(const Symbol &)>) {
59   return Error::success();
60 }
61 
initialize(SectionTableRef)62 Error SectionBase::initialize(SectionTableRef) { return Error::success(); }
finalize()63 void SectionBase::finalize() {}
markSymbols()64 void SectionBase::markSymbols() {}
replaceSectionReferences(const DenseMap<SectionBase *,SectionBase * > &)65 void SectionBase::replaceSectionReferences(
66     const DenseMap<SectionBase *, SectionBase *> &) {}
onRemove()67 void SectionBase::onRemove() {}
68 
writeShdr(const SectionBase & Sec)69 template <class ELFT> void ELFWriter<ELFT>::writeShdr(const SectionBase &Sec) {
70   uint8_t *B =
71       reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Sec.HeaderOffset;
72   Elf_Shdr &Shdr = *reinterpret_cast<Elf_Shdr *>(B);
73   Shdr.sh_name = Sec.NameIndex;
74   Shdr.sh_type = Sec.Type;
75   Shdr.sh_flags = Sec.Flags;
76   Shdr.sh_addr = Sec.Addr;
77   Shdr.sh_offset = Sec.Offset;
78   Shdr.sh_size = Sec.Size;
79   Shdr.sh_link = Sec.Link;
80   Shdr.sh_info = Sec.Info;
81   Shdr.sh_addralign = Sec.Align;
82   Shdr.sh_entsize = Sec.EntrySize;
83 }
84 
visit(Section &)85 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(Section &) {
86   return Error::success();
87 }
88 
visit(OwnedDataSection &)89 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(OwnedDataSection &) {
90   return Error::success();
91 }
92 
visit(StringTableSection &)93 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(StringTableSection &) {
94   return Error::success();
95 }
96 
97 template <class ELFT>
visit(DynamicRelocationSection &)98 Error ELFSectionSizer<ELFT>::visit(DynamicRelocationSection &) {
99   return Error::success();
100 }
101 
102 template <class ELFT>
visit(SymbolTableSection & Sec)103 Error ELFSectionSizer<ELFT>::visit(SymbolTableSection &Sec) {
104   Sec.EntrySize = sizeof(Elf_Sym);
105   Sec.Size = Sec.Symbols.size() * Sec.EntrySize;
106   // Align to the largest field in Elf_Sym.
107   Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word);
108   return Error::success();
109 }
110 
111 template <bool Is64>
encodeCrel(ArrayRef<Relocation> Relocations)112 static SmallVector<char, 0> encodeCrel(ArrayRef<Relocation> Relocations) {
113   using uint = std::conditional_t<Is64, uint64_t, uint32_t>;
114   SmallVector<char, 0> Content;
115   raw_svector_ostream OS(Content);
116   ELF::encodeCrel<Is64>(OS, Relocations, [&](const Relocation &R) {
117     uint32_t CurSymIdx = R.RelocSymbol ? R.RelocSymbol->Index : 0;
118     return ELF::Elf_Crel<Is64>{static_cast<uint>(R.Offset), CurSymIdx, R.Type,
119                                std::make_signed_t<uint>(R.Addend)};
120   });
121   return Content;
122 }
123 
124 template <class ELFT>
visit(RelocationSection & Sec)125 Error ELFSectionSizer<ELFT>::visit(RelocationSection &Sec) {
126   if (Sec.Type == SHT_CREL) {
127     Sec.Size = encodeCrel<ELFT::Is64Bits>(Sec.Relocations).size();
128   } else {
129     Sec.EntrySize = Sec.Type == SHT_REL ? sizeof(Elf_Rel) : sizeof(Elf_Rela);
130     Sec.Size = Sec.Relocations.size() * Sec.EntrySize;
131     // Align to the largest field in Elf_Rel(a).
132     Sec.Align = ELFT::Is64Bits ? sizeof(Elf_Xword) : sizeof(Elf_Word);
133   }
134   return Error::success();
135 }
136 
137 template <class ELFT>
visit(GnuDebugLinkSection &)138 Error ELFSectionSizer<ELFT>::visit(GnuDebugLinkSection &) {
139   return Error::success();
140 }
141 
visit(GroupSection & Sec)142 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(GroupSection &Sec) {
143   Sec.Size = sizeof(Elf_Word) + Sec.GroupMembers.size() * sizeof(Elf_Word);
144   return Error::success();
145 }
146 
147 template <class ELFT>
visit(SectionIndexSection &)148 Error ELFSectionSizer<ELFT>::visit(SectionIndexSection &) {
149   return Error::success();
150 }
151 
visit(CompressedSection &)152 template <class ELFT> Error ELFSectionSizer<ELFT>::visit(CompressedSection &) {
153   return Error::success();
154 }
155 
156 template <class ELFT>
visit(DecompressedSection &)157 Error ELFSectionSizer<ELFT>::visit(DecompressedSection &) {
158   return Error::success();
159 }
160 
visit(const SectionIndexSection & Sec)161 Error BinarySectionWriter::visit(const SectionIndexSection &Sec) {
162   return createStringError(errc::operation_not_permitted,
163                            "cannot write symbol section index table '" +
164                                Sec.Name + "' ");
165 }
166 
visit(const SymbolTableSection & Sec)167 Error BinarySectionWriter::visit(const SymbolTableSection &Sec) {
168   return createStringError(errc::operation_not_permitted,
169                            "cannot write symbol table '" + Sec.Name +
170                                "' out to binary");
171 }
172 
visit(const RelocationSection & Sec)173 Error BinarySectionWriter::visit(const RelocationSection &Sec) {
174   return createStringError(errc::operation_not_permitted,
175                            "cannot write relocation section '" + Sec.Name +
176                                "' out to binary");
177 }
178 
visit(const GnuDebugLinkSection & Sec)179 Error BinarySectionWriter::visit(const GnuDebugLinkSection &Sec) {
180   return createStringError(errc::operation_not_permitted,
181                            "cannot write '" + Sec.Name + "' out to binary");
182 }
183 
visit(const GroupSection & Sec)184 Error BinarySectionWriter::visit(const GroupSection &Sec) {
185   return createStringError(errc::operation_not_permitted,
186                            "cannot write '" + Sec.Name + "' out to binary");
187 }
188 
visit(const Section & Sec)189 Error SectionWriter::visit(const Section &Sec) {
190   if (Sec.Type != SHT_NOBITS)
191     llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset);
192 
193   return Error::success();
194 }
195 
addressOverflows32bit(uint64_t Addr)196 static bool addressOverflows32bit(uint64_t Addr) {
197   // Sign extended 32 bit addresses (e.g 0xFFFFFFFF80000000) are ok
198   return Addr > UINT32_MAX && Addr + 0x80000000 > UINT32_MAX;
199 }
200 
checkedGetHex(StringRef S)201 template <class T> static T checkedGetHex(StringRef S) {
202   T Value;
203   bool Fail = S.getAsInteger(16, Value);
204   assert(!Fail);
205   (void)Fail;
206   return Value;
207 }
208 
209 // Fills exactly Len bytes of buffer with hexadecimal characters
210 // representing value 'X'
211 template <class T, class Iterator>
toHexStr(T X,Iterator It,size_t Len)212 static Iterator toHexStr(T X, Iterator It, size_t Len) {
213   // Fill range with '0'
214   std::fill(It, It + Len, '0');
215 
216   for (long I = Len - 1; I >= 0; --I) {
217     unsigned char Mod = static_cast<unsigned char>(X) & 15;
218     *(It + I) = hexdigit(Mod, false);
219     X >>= 4;
220   }
221   assert(X == 0);
222   return It + Len;
223 }
224 
getChecksum(StringRef S)225 uint8_t IHexRecord::getChecksum(StringRef S) {
226   assert((S.size() & 1) == 0);
227   uint8_t Checksum = 0;
228   while (!S.empty()) {
229     Checksum += checkedGetHex<uint8_t>(S.take_front(2));
230     S = S.drop_front(2);
231   }
232   return -Checksum;
233 }
234 
getLine(uint8_t Type,uint16_t Addr,ArrayRef<uint8_t> Data)235 IHexLineData IHexRecord::getLine(uint8_t Type, uint16_t Addr,
236                                  ArrayRef<uint8_t> Data) {
237   IHexLineData Line(getLineLength(Data.size()));
238   assert(Line.size());
239   auto Iter = Line.begin();
240   *Iter++ = ':';
241   Iter = toHexStr(Data.size(), Iter, 2);
242   Iter = toHexStr(Addr, Iter, 4);
243   Iter = toHexStr(Type, Iter, 2);
244   for (uint8_t X : Data)
245     Iter = toHexStr(X, Iter, 2);
246   StringRef S(Line.data() + 1, std::distance(Line.begin() + 1, Iter));
247   Iter = toHexStr(getChecksum(S), Iter, 2);
248   *Iter++ = '\r';
249   *Iter++ = '\n';
250   assert(Iter == Line.end());
251   return Line;
252 }
253 
checkRecord(const IHexRecord & R)254 static Error checkRecord(const IHexRecord &R) {
255   switch (R.Type) {
256   case IHexRecord::Data:
257     if (R.HexData.size() == 0)
258       return createStringError(
259           errc::invalid_argument,
260           "zero data length is not allowed for data records");
261     break;
262   case IHexRecord::EndOfFile:
263     break;
264   case IHexRecord::SegmentAddr:
265     // 20-bit segment address. Data length must be 2 bytes
266     // (4 bytes in hex)
267     if (R.HexData.size() != 4)
268       return createStringError(
269           errc::invalid_argument,
270           "segment address data should be 2 bytes in size");
271     break;
272   case IHexRecord::StartAddr80x86:
273   case IHexRecord::StartAddr:
274     if (R.HexData.size() != 8)
275       return createStringError(errc::invalid_argument,
276                                "start address data should be 4 bytes in size");
277     // According to Intel HEX specification '03' record
278     // only specifies the code address within the 20-bit
279     // segmented address space of the 8086/80186. This
280     // means 12 high order bits should be zeroes.
281     if (R.Type == IHexRecord::StartAddr80x86 &&
282         R.HexData.take_front(3) != "000")
283       return createStringError(errc::invalid_argument,
284                                "start address exceeds 20 bit for 80x86");
285     break;
286   case IHexRecord::ExtendedAddr:
287     // 16-31 bits of linear base address
288     if (R.HexData.size() != 4)
289       return createStringError(
290           errc::invalid_argument,
291           "extended address data should be 2 bytes in size");
292     break;
293   default:
294     // Unknown record type
295     return createStringError(errc::invalid_argument, "unknown record type: %u",
296                              static_cast<unsigned>(R.Type));
297   }
298   return Error::success();
299 }
300 
301 // Checks that IHEX line contains valid characters.
302 // This allows converting hexadecimal data to integers
303 // without extra verification.
checkChars(StringRef Line)304 static Error checkChars(StringRef Line) {
305   assert(!Line.empty());
306   if (Line[0] != ':')
307     return createStringError(errc::invalid_argument,
308                              "missing ':' in the beginning of line.");
309 
310   for (size_t Pos = 1; Pos < Line.size(); ++Pos)
311     if (hexDigitValue(Line[Pos]) == -1U)
312       return createStringError(errc::invalid_argument,
313                                "invalid character at position %zu.", Pos + 1);
314   return Error::success();
315 }
316 
parse(StringRef Line)317 Expected<IHexRecord> IHexRecord::parse(StringRef Line) {
318   assert(!Line.empty());
319 
320   // ':' + Length + Address + Type + Checksum with empty data ':LLAAAATTCC'
321   if (Line.size() < 11)
322     return createStringError(errc::invalid_argument,
323                              "line is too short: %zu chars.", Line.size());
324 
325   if (Error E = checkChars(Line))
326     return std::move(E);
327 
328   IHexRecord Rec;
329   size_t DataLen = checkedGetHex<uint8_t>(Line.substr(1, 2));
330   if (Line.size() != getLength(DataLen))
331     return createStringError(errc::invalid_argument,
332                              "invalid line length %zu (should be %zu)",
333                              Line.size(), getLength(DataLen));
334 
335   Rec.Addr = checkedGetHex<uint16_t>(Line.substr(3, 4));
336   Rec.Type = checkedGetHex<uint8_t>(Line.substr(7, 2));
337   Rec.HexData = Line.substr(9, DataLen * 2);
338 
339   if (getChecksum(Line.drop_front(1)) != 0)
340     return createStringError(errc::invalid_argument, "incorrect checksum.");
341   if (Error E = checkRecord(Rec))
342     return std::move(E);
343   return Rec;
344 }
345 
sectionPhysicalAddr(const SectionBase * Sec)346 static uint64_t sectionPhysicalAddr(const SectionBase *Sec) {
347   Segment *Seg = Sec->ParentSegment;
348   if (Seg && Seg->Type != ELF::PT_LOAD)
349     Seg = nullptr;
350   return Seg ? Seg->PAddr + Sec->OriginalOffset - Seg->OriginalOffset
351              : Sec->Addr;
352 }
353 
writeSection(const SectionBase * Sec,ArrayRef<uint8_t> Data)354 void IHexSectionWriterBase::writeSection(const SectionBase *Sec,
355                                          ArrayRef<uint8_t> Data) {
356   assert(Data.size() == Sec->Size);
357   const uint32_t ChunkSize = 16;
358   uint32_t Addr = sectionPhysicalAddr(Sec) & 0xFFFFFFFFU;
359   while (!Data.empty()) {
360     uint64_t DataSize = std::min<uint64_t>(Data.size(), ChunkSize);
361     if (Addr > SegmentAddr + BaseAddr + 0xFFFFU) {
362       if (Addr > 0xFFFFFU) {
363         // Write extended address record, zeroing segment address
364         // if needed.
365         if (SegmentAddr != 0)
366           SegmentAddr = writeSegmentAddr(0U);
367         BaseAddr = writeBaseAddr(Addr);
368       } else {
369         // We can still remain 16-bit
370         SegmentAddr = writeSegmentAddr(Addr);
371       }
372     }
373     uint64_t SegOffset = Addr - BaseAddr - SegmentAddr;
374     assert(SegOffset <= 0xFFFFU);
375     DataSize = std::min(DataSize, 0x10000U - SegOffset);
376     writeData(0, SegOffset, Data.take_front(DataSize));
377     Addr += DataSize;
378     Data = Data.drop_front(DataSize);
379   }
380 }
381 
writeSegmentAddr(uint64_t Addr)382 uint64_t IHexSectionWriterBase::writeSegmentAddr(uint64_t Addr) {
383   assert(Addr <= 0xFFFFFU);
384   uint8_t Data[] = {static_cast<uint8_t>((Addr & 0xF0000U) >> 12), 0};
385   writeData(2, 0, Data);
386   return Addr & 0xF0000U;
387 }
388 
writeBaseAddr(uint64_t Addr)389 uint64_t IHexSectionWriterBase::writeBaseAddr(uint64_t Addr) {
390   assert(Addr <= 0xFFFFFFFFU);
391   uint64_t Base = Addr & 0xFFFF0000U;
392   uint8_t Data[] = {static_cast<uint8_t>(Base >> 24),
393                     static_cast<uint8_t>((Base >> 16) & 0xFF)};
394   writeData(4, 0, Data);
395   return Base;
396 }
397 
writeData(uint8_t,uint16_t,ArrayRef<uint8_t> Data)398 void IHexSectionWriterBase::writeData(uint8_t, uint16_t,
399                                       ArrayRef<uint8_t> Data) {
400   Offset += IHexRecord::getLineLength(Data.size());
401 }
402 
visit(const Section & Sec)403 Error IHexSectionWriterBase::visit(const Section &Sec) {
404   writeSection(&Sec, Sec.Contents);
405   return Error::success();
406 }
407 
visit(const OwnedDataSection & Sec)408 Error IHexSectionWriterBase::visit(const OwnedDataSection &Sec) {
409   writeSection(&Sec, Sec.Data);
410   return Error::success();
411 }
412 
visit(const StringTableSection & Sec)413 Error IHexSectionWriterBase::visit(const StringTableSection &Sec) {
414   // Check that sizer has already done its work
415   assert(Sec.Size == Sec.StrTabBuilder.getSize());
416   // We are free to pass an invalid pointer to writeSection as long
417   // as we don't actually write any data. The real writer class has
418   // to override this method .
419   writeSection(&Sec, {nullptr, static_cast<size_t>(Sec.Size)});
420   return Error::success();
421 }
422 
visit(const DynamicRelocationSection & Sec)423 Error IHexSectionWriterBase::visit(const DynamicRelocationSection &Sec) {
424   writeSection(&Sec, Sec.Contents);
425   return Error::success();
426 }
427 
writeData(uint8_t Type,uint16_t Addr,ArrayRef<uint8_t> Data)428 void IHexSectionWriter::writeData(uint8_t Type, uint16_t Addr,
429                                   ArrayRef<uint8_t> Data) {
430   IHexLineData HexData = IHexRecord::getLine(Type, Addr, Data);
431   memcpy(Out.getBufferStart() + Offset, HexData.data(), HexData.size());
432   Offset += HexData.size();
433 }
434 
visit(const StringTableSection & Sec)435 Error IHexSectionWriter::visit(const StringTableSection &Sec) {
436   assert(Sec.Size == Sec.StrTabBuilder.getSize());
437   std::vector<uint8_t> Data(Sec.Size);
438   Sec.StrTabBuilder.write(Data.data());
439   writeSection(&Sec, Data);
440   return Error::success();
441 }
442 
accept(SectionVisitor & Visitor) const443 Error Section::accept(SectionVisitor &Visitor) const {
444   return Visitor.visit(*this);
445 }
446 
accept(MutableSectionVisitor & Visitor)447 Error Section::accept(MutableSectionVisitor &Visitor) {
448   return Visitor.visit(*this);
449 }
450 
restoreSymTabLink(SymbolTableSection & SymTab)451 void Section::restoreSymTabLink(SymbolTableSection &SymTab) {
452   if (HasSymTabLink) {
453     assert(LinkSection == nullptr);
454     LinkSection = &SymTab;
455   }
456 }
457 
visit(const OwnedDataSection & Sec)458 Error SectionWriter::visit(const OwnedDataSection &Sec) {
459   llvm::copy(Sec.Data, Out.getBufferStart() + Sec.Offset);
460   return Error::success();
461 }
462 
463 template <class ELFT>
visit(const DecompressedSection & Sec)464 Error ELFSectionWriter<ELFT>::visit(const DecompressedSection &Sec) {
465   ArrayRef<uint8_t> Compressed =
466       Sec.OriginalData.slice(sizeof(Elf_Chdr_Impl<ELFT>));
467   SmallVector<uint8_t, 128> Decompressed;
468   DebugCompressionType Type;
469   switch (Sec.ChType) {
470   case ELFCOMPRESS_ZLIB:
471     Type = DebugCompressionType::Zlib;
472     break;
473   case ELFCOMPRESS_ZSTD:
474     Type = DebugCompressionType::Zstd;
475     break;
476   default:
477     return createStringError(errc::invalid_argument,
478                              "--decompress-debug-sections: ch_type (" +
479                                  Twine(Sec.ChType) + ") of section '" +
480                                  Sec.Name + "' is unsupported");
481   }
482   if (auto *Reason =
483           compression::getReasonIfUnsupported(compression::formatFor(Type)))
484     return createStringError(errc::invalid_argument,
485                              "failed to decompress section '" + Sec.Name +
486                                  "': " + Reason);
487   if (Error E = compression::decompress(Type, Compressed, Decompressed,
488                                         static_cast<size_t>(Sec.Size)))
489     return createStringError(errc::invalid_argument,
490                              "failed to decompress section '" + Sec.Name +
491                                  "': " + toString(std::move(E)));
492 
493   uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
494   std::copy(Decompressed.begin(), Decompressed.end(), Buf);
495 
496   return Error::success();
497 }
498 
visit(const DecompressedSection & Sec)499 Error BinarySectionWriter::visit(const DecompressedSection &Sec) {
500   return createStringError(errc::operation_not_permitted,
501                            "cannot write compressed section '" + Sec.Name +
502                                "' ");
503 }
504 
accept(SectionVisitor & Visitor) const505 Error DecompressedSection::accept(SectionVisitor &Visitor) const {
506   return Visitor.visit(*this);
507 }
508 
accept(MutableSectionVisitor & Visitor)509 Error DecompressedSection::accept(MutableSectionVisitor &Visitor) {
510   return Visitor.visit(*this);
511 }
512 
accept(SectionVisitor & Visitor) const513 Error OwnedDataSection::accept(SectionVisitor &Visitor) const {
514   return Visitor.visit(*this);
515 }
516 
accept(MutableSectionVisitor & Visitor)517 Error OwnedDataSection::accept(MutableSectionVisitor &Visitor) {
518   return Visitor.visit(*this);
519 }
520 
appendHexData(StringRef HexData)521 void OwnedDataSection::appendHexData(StringRef HexData) {
522   assert((HexData.size() & 1) == 0);
523   while (!HexData.empty()) {
524     Data.push_back(checkedGetHex<uint8_t>(HexData.take_front(2)));
525     HexData = HexData.drop_front(2);
526   }
527   Size = Data.size();
528 }
529 
visit(const CompressedSection & Sec)530 Error BinarySectionWriter::visit(const CompressedSection &Sec) {
531   return createStringError(errc::operation_not_permitted,
532                            "cannot write compressed section '" + Sec.Name +
533                                "' ");
534 }
535 
536 template <class ELFT>
visit(const CompressedSection & Sec)537 Error ELFSectionWriter<ELFT>::visit(const CompressedSection &Sec) {
538   uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
539   Elf_Chdr_Impl<ELFT> Chdr = {};
540   switch (Sec.CompressionType) {
541   case DebugCompressionType::None:
542     std::copy(Sec.OriginalData.begin(), Sec.OriginalData.end(), Buf);
543     return Error::success();
544   case DebugCompressionType::Zlib:
545     Chdr.ch_type = ELF::ELFCOMPRESS_ZLIB;
546     break;
547   case DebugCompressionType::Zstd:
548     Chdr.ch_type = ELF::ELFCOMPRESS_ZSTD;
549     break;
550   }
551   Chdr.ch_size = Sec.DecompressedSize;
552   Chdr.ch_addralign = Sec.DecompressedAlign;
553   memcpy(Buf, &Chdr, sizeof(Chdr));
554   Buf += sizeof(Chdr);
555 
556   std::copy(Sec.CompressedData.begin(), Sec.CompressedData.end(), Buf);
557   return Error::success();
558 }
559 
CompressedSection(const SectionBase & Sec,DebugCompressionType CompressionType,bool Is64Bits)560 CompressedSection::CompressedSection(const SectionBase &Sec,
561                                      DebugCompressionType CompressionType,
562                                      bool Is64Bits)
563     : SectionBase(Sec), CompressionType(CompressionType),
564       DecompressedSize(Sec.OriginalData.size()), DecompressedAlign(Sec.Align) {
565   compression::compress(compression::Params(CompressionType), OriginalData,
566                         CompressedData);
567 
568   Flags |= ELF::SHF_COMPRESSED;
569   OriginalFlags |= ELF::SHF_COMPRESSED;
570   size_t ChdrSize = Is64Bits ? sizeof(object::Elf_Chdr_Impl<object::ELF64LE>)
571                              : sizeof(object::Elf_Chdr_Impl<object::ELF32LE>);
572   Size = ChdrSize + CompressedData.size();
573   Align = 8;
574 }
575 
CompressedSection(ArrayRef<uint8_t> CompressedData,uint32_t ChType,uint64_t DecompressedSize,uint64_t DecompressedAlign)576 CompressedSection::CompressedSection(ArrayRef<uint8_t> CompressedData,
577                                      uint32_t ChType, uint64_t DecompressedSize,
578                                      uint64_t DecompressedAlign)
579     : ChType(ChType), CompressionType(DebugCompressionType::None),
580       DecompressedSize(DecompressedSize), DecompressedAlign(DecompressedAlign) {
581   OriginalData = CompressedData;
582 }
583 
accept(SectionVisitor & Visitor) const584 Error CompressedSection::accept(SectionVisitor &Visitor) const {
585   return Visitor.visit(*this);
586 }
587 
accept(MutableSectionVisitor & Visitor)588 Error CompressedSection::accept(MutableSectionVisitor &Visitor) {
589   return Visitor.visit(*this);
590 }
591 
addString(StringRef Name)592 void StringTableSection::addString(StringRef Name) { StrTabBuilder.add(Name); }
593 
findIndex(StringRef Name) const594 uint32_t StringTableSection::findIndex(StringRef Name) const {
595   return StrTabBuilder.getOffset(Name);
596 }
597 
prepareForLayout()598 void StringTableSection::prepareForLayout() {
599   StrTabBuilder.finalize();
600   Size = StrTabBuilder.getSize();
601 }
602 
visit(const StringTableSection & Sec)603 Error SectionWriter::visit(const StringTableSection &Sec) {
604   Sec.StrTabBuilder.write(reinterpret_cast<uint8_t *>(Out.getBufferStart()) +
605                           Sec.Offset);
606   return Error::success();
607 }
608 
accept(SectionVisitor & Visitor) const609 Error StringTableSection::accept(SectionVisitor &Visitor) const {
610   return Visitor.visit(*this);
611 }
612 
accept(MutableSectionVisitor & Visitor)613 Error StringTableSection::accept(MutableSectionVisitor &Visitor) {
614   return Visitor.visit(*this);
615 }
616 
617 template <class ELFT>
visit(const SectionIndexSection & Sec)618 Error ELFSectionWriter<ELFT>::visit(const SectionIndexSection &Sec) {
619   uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
620   llvm::copy(Sec.Indexes, reinterpret_cast<Elf_Word *>(Buf));
621   return Error::success();
622 }
623 
initialize(SectionTableRef SecTable)624 Error SectionIndexSection::initialize(SectionTableRef SecTable) {
625   Size = 0;
626   Expected<SymbolTableSection *> Sec =
627       SecTable.getSectionOfType<SymbolTableSection>(
628           Link,
629           "Link field value " + Twine(Link) + " in section " + Name +
630               " is invalid",
631           "Link field value " + Twine(Link) + " in section " + Name +
632               " is not a symbol table");
633   if (!Sec)
634     return Sec.takeError();
635 
636   setSymTab(*Sec);
637   Symbols->setShndxTable(this);
638   return Error::success();
639 }
640 
finalize()641 void SectionIndexSection::finalize() { Link = Symbols->Index; }
642 
accept(SectionVisitor & Visitor) const643 Error SectionIndexSection::accept(SectionVisitor &Visitor) const {
644   return Visitor.visit(*this);
645 }
646 
accept(MutableSectionVisitor & Visitor)647 Error SectionIndexSection::accept(MutableSectionVisitor &Visitor) {
648   return Visitor.visit(*this);
649 }
650 
isValidReservedSectionIndex(uint16_t Index,uint16_t Machine)651 static bool isValidReservedSectionIndex(uint16_t Index, uint16_t Machine) {
652   switch (Index) {
653   case SHN_ABS:
654   case SHN_COMMON:
655     return true;
656   }
657 
658   if (Machine == EM_AMDGPU) {
659     return Index == SHN_AMDGPU_LDS;
660   }
661 
662   if (Machine == EM_MIPS) {
663     switch (Index) {
664     case SHN_MIPS_ACOMMON:
665     case SHN_MIPS_SCOMMON:
666     case SHN_MIPS_SUNDEFINED:
667       return true;
668     }
669   }
670 
671   if (Machine == EM_HEXAGON) {
672     switch (Index) {
673     case SHN_HEXAGON_SCOMMON:
674     case SHN_HEXAGON_SCOMMON_1:
675     case SHN_HEXAGON_SCOMMON_2:
676     case SHN_HEXAGON_SCOMMON_4:
677     case SHN_HEXAGON_SCOMMON_8:
678       return true;
679     }
680   }
681   return false;
682 }
683 
684 // Large indexes force us to clarify exactly what this function should do. This
685 // function should return the value that will appear in st_shndx when written
686 // out.
getShndx() const687 uint16_t Symbol::getShndx() const {
688   if (DefinedIn != nullptr) {
689     if (DefinedIn->Index >= SHN_LORESERVE)
690       return SHN_XINDEX;
691     return DefinedIn->Index;
692   }
693 
694   if (ShndxType == SYMBOL_SIMPLE_INDEX) {
695     // This means that we don't have a defined section but we do need to
696     // output a legitimate section index.
697     return SHN_UNDEF;
698   }
699 
700   assert(ShndxType == SYMBOL_ABS || ShndxType == SYMBOL_COMMON ||
701          (ShndxType >= SYMBOL_LOPROC && ShndxType <= SYMBOL_HIPROC) ||
702          (ShndxType >= SYMBOL_LOOS && ShndxType <= SYMBOL_HIOS));
703   return static_cast<uint16_t>(ShndxType);
704 }
705 
isCommon() const706 bool Symbol::isCommon() const { return getShndx() == SHN_COMMON; }
707 
assignIndices()708 void SymbolTableSection::assignIndices() {
709   uint32_t Index = 0;
710   for (auto &Sym : Symbols) {
711     if (Sym->Index != Index)
712       IndicesChanged = true;
713     Sym->Index = Index++;
714   }
715 }
716 
addSymbol(Twine Name,uint8_t Bind,uint8_t Type,SectionBase * DefinedIn,uint64_t Value,uint8_t Visibility,uint16_t Shndx,uint64_t SymbolSize)717 void SymbolTableSection::addSymbol(Twine Name, uint8_t Bind, uint8_t Type,
718                                    SectionBase *DefinedIn, uint64_t Value,
719                                    uint8_t Visibility, uint16_t Shndx,
720                                    uint64_t SymbolSize) {
721   Symbol Sym;
722   Sym.Name = Name.str();
723   Sym.Binding = Bind;
724   Sym.Type = Type;
725   Sym.DefinedIn = DefinedIn;
726   if (DefinedIn != nullptr)
727     DefinedIn->HasSymbol = true;
728   if (DefinedIn == nullptr) {
729     if (Shndx >= SHN_LORESERVE)
730       Sym.ShndxType = static_cast<SymbolShndxType>(Shndx);
731     else
732       Sym.ShndxType = SYMBOL_SIMPLE_INDEX;
733   }
734   Sym.Value = Value;
735   Sym.Visibility = Visibility;
736   Sym.Size = SymbolSize;
737   Sym.Index = Symbols.size();
738   Symbols.emplace_back(std::make_unique<Symbol>(Sym));
739   Size += this->EntrySize;
740 }
741 
removeSectionReferences(bool AllowBrokenLinks,function_ref<bool (const SectionBase *)> ToRemove)742 Error SymbolTableSection::removeSectionReferences(
743     bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
744   if (ToRemove(SectionIndexTable))
745     SectionIndexTable = nullptr;
746   if (ToRemove(SymbolNames)) {
747     if (!AllowBrokenLinks)
748       return createStringError(
749           llvm::errc::invalid_argument,
750           "string table '%s' cannot be removed because it is "
751           "referenced by the symbol table '%s'",
752           SymbolNames->Name.data(), this->Name.data());
753     SymbolNames = nullptr;
754   }
755   return removeSymbols(
756       [ToRemove](const Symbol &Sym) { return ToRemove(Sym.DefinedIn); });
757 }
758 
updateSymbols(function_ref<void (Symbol &)> Callable)759 void SymbolTableSection::updateSymbols(function_ref<void(Symbol &)> Callable) {
760   for (SymPtr &Sym : llvm::drop_begin(Symbols))
761     Callable(*Sym);
762   std::stable_partition(
763       std::begin(Symbols), std::end(Symbols),
764       [](const SymPtr &Sym) { return Sym->Binding == STB_LOCAL; });
765   assignIndices();
766 }
767 
removeSymbols(function_ref<bool (const Symbol &)> ToRemove)768 Error SymbolTableSection::removeSymbols(
769     function_ref<bool(const Symbol &)> ToRemove) {
770   Symbols.erase(
771       std::remove_if(std::begin(Symbols) + 1, std::end(Symbols),
772                      [ToRemove](const SymPtr &Sym) { return ToRemove(*Sym); }),
773       std::end(Symbols));
774   auto PrevSize = Size;
775   Size = Symbols.size() * EntrySize;
776   if (Size < PrevSize)
777     IndicesChanged = true;
778   assignIndices();
779   return Error::success();
780 }
781 
replaceSectionReferences(const DenseMap<SectionBase *,SectionBase * > & FromTo)782 void SymbolTableSection::replaceSectionReferences(
783     const DenseMap<SectionBase *, SectionBase *> &FromTo) {
784   for (std::unique_ptr<Symbol> &Sym : Symbols)
785     if (SectionBase *To = FromTo.lookup(Sym->DefinedIn))
786       Sym->DefinedIn = To;
787 }
788 
initialize(SectionTableRef SecTable)789 Error SymbolTableSection::initialize(SectionTableRef SecTable) {
790   Size = 0;
791   Expected<StringTableSection *> Sec =
792       SecTable.getSectionOfType<StringTableSection>(
793           Link,
794           "Symbol table has link index of " + Twine(Link) +
795               " which is not a valid index",
796           "Symbol table has link index of " + Twine(Link) +
797               " which is not a string table");
798   if (!Sec)
799     return Sec.takeError();
800 
801   setStrTab(*Sec);
802   return Error::success();
803 }
804 
finalize()805 void SymbolTableSection::finalize() {
806   uint32_t MaxLocalIndex = 0;
807   for (std::unique_ptr<Symbol> &Sym : Symbols) {
808     Sym->NameIndex =
809         SymbolNames == nullptr ? 0 : SymbolNames->findIndex(Sym->Name);
810     if (Sym->Binding == STB_LOCAL)
811       MaxLocalIndex = std::max(MaxLocalIndex, Sym->Index);
812   }
813   // Now we need to set the Link and Info fields.
814   Link = SymbolNames == nullptr ? 0 : SymbolNames->Index;
815   Info = MaxLocalIndex + 1;
816 }
817 
prepareForLayout()818 void SymbolTableSection::prepareForLayout() {
819   // Reserve proper amount of space in section index table, so we can
820   // layout sections correctly. We will fill the table with correct
821   // indexes later in fillShdnxTable.
822   if (SectionIndexTable)
823     SectionIndexTable->reserve(Symbols.size());
824 
825   // Add all of our strings to SymbolNames so that SymbolNames has the right
826   // size before layout is decided.
827   // If the symbol names section has been removed, don't try to add strings to
828   // the table.
829   if (SymbolNames != nullptr)
830     for (std::unique_ptr<Symbol> &Sym : Symbols)
831       SymbolNames->addString(Sym->Name);
832 }
833 
fillShndxTable()834 void SymbolTableSection::fillShndxTable() {
835   if (SectionIndexTable == nullptr)
836     return;
837   // Fill section index table with real section indexes. This function must
838   // be called after assignOffsets.
839   for (const std::unique_ptr<Symbol> &Sym : Symbols) {
840     if (Sym->DefinedIn != nullptr && Sym->DefinedIn->Index >= SHN_LORESERVE)
841       SectionIndexTable->addIndex(Sym->DefinedIn->Index);
842     else
843       SectionIndexTable->addIndex(SHN_UNDEF);
844   }
845 }
846 
847 Expected<const Symbol *>
getSymbolByIndex(uint32_t Index) const848 SymbolTableSection::getSymbolByIndex(uint32_t Index) const {
849   if (Symbols.size() <= Index)
850     return createStringError(errc::invalid_argument,
851                              "invalid symbol index: " + Twine(Index));
852   return Symbols[Index].get();
853 }
854 
getSymbolByIndex(uint32_t Index)855 Expected<Symbol *> SymbolTableSection::getSymbolByIndex(uint32_t Index) {
856   Expected<const Symbol *> Sym =
857       static_cast<const SymbolTableSection *>(this)->getSymbolByIndex(Index);
858   if (!Sym)
859     return Sym.takeError();
860 
861   return const_cast<Symbol *>(*Sym);
862 }
863 
864 template <class ELFT>
visit(const SymbolTableSection & Sec)865 Error ELFSectionWriter<ELFT>::visit(const SymbolTableSection &Sec) {
866   Elf_Sym *Sym = reinterpret_cast<Elf_Sym *>(Out.getBufferStart() + Sec.Offset);
867   // Loop though symbols setting each entry of the symbol table.
868   for (const std::unique_ptr<Symbol> &Symbol : Sec.Symbols) {
869     Sym->st_name = Symbol->NameIndex;
870     Sym->st_value = Symbol->Value;
871     Sym->st_size = Symbol->Size;
872     Sym->st_other = Symbol->Visibility;
873     Sym->setBinding(Symbol->Binding);
874     Sym->setType(Symbol->Type);
875     Sym->st_shndx = Symbol->getShndx();
876     ++Sym;
877   }
878   return Error::success();
879 }
880 
accept(SectionVisitor & Visitor) const881 Error SymbolTableSection::accept(SectionVisitor &Visitor) const {
882   return Visitor.visit(*this);
883 }
884 
accept(MutableSectionVisitor & Visitor)885 Error SymbolTableSection::accept(MutableSectionVisitor &Visitor) {
886   return Visitor.visit(*this);
887 }
888 
getNamePrefix() const889 StringRef RelocationSectionBase::getNamePrefix() const {
890   switch (Type) {
891   case SHT_REL:
892     return ".rel";
893   case SHT_RELA:
894     return ".rela";
895   case SHT_CREL:
896     return ".crel";
897   default:
898     llvm_unreachable("not a relocation section");
899   }
900 }
901 
removeSectionReferences(bool AllowBrokenLinks,function_ref<bool (const SectionBase *)> ToRemove)902 Error RelocationSection::removeSectionReferences(
903     bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
904   if (ToRemove(Symbols)) {
905     if (!AllowBrokenLinks)
906       return createStringError(
907           llvm::errc::invalid_argument,
908           "symbol table '%s' cannot be removed because it is "
909           "referenced by the relocation section '%s'",
910           Symbols->Name.data(), this->Name.data());
911     Symbols = nullptr;
912   }
913 
914   for (const Relocation &R : Relocations) {
915     if (!R.RelocSymbol || !R.RelocSymbol->DefinedIn ||
916         !ToRemove(R.RelocSymbol->DefinedIn))
917       continue;
918     return createStringError(llvm::errc::invalid_argument,
919                              "section '%s' cannot be removed: (%s+0x%" PRIx64
920                              ") has relocation against symbol '%s'",
921                              R.RelocSymbol->DefinedIn->Name.data(),
922                              SecToApplyRel->Name.data(), R.Offset,
923                              R.RelocSymbol->Name.c_str());
924   }
925 
926   return Error::success();
927 }
928 
929 template <class SymTabType>
initialize(SectionTableRef SecTable)930 Error RelocSectionWithSymtabBase<SymTabType>::initialize(
931     SectionTableRef SecTable) {
932   if (Link != SHN_UNDEF) {
933     Expected<SymTabType *> Sec = SecTable.getSectionOfType<SymTabType>(
934         Link,
935         "Link field value " + Twine(Link) + " in section " + Name +
936             " is invalid",
937         "Link field value " + Twine(Link) + " in section " + Name +
938             " is not a symbol table");
939     if (!Sec)
940       return Sec.takeError();
941 
942     setSymTab(*Sec);
943   }
944 
945   if (Info != SHN_UNDEF) {
946     Expected<SectionBase *> Sec =
947         SecTable.getSection(Info, "Info field value " + Twine(Info) +
948                                       " in section " + Name + " is invalid");
949     if (!Sec)
950       return Sec.takeError();
951 
952     setSection(*Sec);
953   } else
954     setSection(nullptr);
955 
956   return Error::success();
957 }
958 
959 template <class SymTabType>
finalize()960 void RelocSectionWithSymtabBase<SymTabType>::finalize() {
961   this->Link = Symbols ? Symbols->Index : 0;
962 
963   if (SecToApplyRel != nullptr)
964     this->Info = SecToApplyRel->Index;
965 }
966 
967 template <class ELFT>
setAddend(Elf_Rel_Impl<ELFT,false> &,uint64_t)968 static void setAddend(Elf_Rel_Impl<ELFT, false> &, uint64_t) {}
969 
970 template <class ELFT>
setAddend(Elf_Rel_Impl<ELFT,true> & Rela,uint64_t Addend)971 static void setAddend(Elf_Rel_Impl<ELFT, true> &Rela, uint64_t Addend) {
972   Rela.r_addend = Addend;
973 }
974 
975 template <class RelRange, class T>
writeRel(const RelRange & Relocations,T * Buf,bool IsMips64EL)976 static void writeRel(const RelRange &Relocations, T *Buf, bool IsMips64EL) {
977   for (const auto &Reloc : Relocations) {
978     Buf->r_offset = Reloc.Offset;
979     setAddend(*Buf, Reloc.Addend);
980     Buf->setSymbolAndType(Reloc.RelocSymbol ? Reloc.RelocSymbol->Index : 0,
981                           Reloc.Type, IsMips64EL);
982     ++Buf;
983   }
984 }
985 
986 template <class ELFT>
visit(const RelocationSection & Sec)987 Error ELFSectionWriter<ELFT>::visit(const RelocationSection &Sec) {
988   uint8_t *Buf = reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
989   if (Sec.Type == SHT_CREL) {
990     auto Content = encodeCrel<ELFT::Is64Bits>(Sec.Relocations);
991     memcpy(Buf, Content.data(), Content.size());
992   } else if (Sec.Type == SHT_REL) {
993     writeRel(Sec.Relocations, reinterpret_cast<Elf_Rel *>(Buf),
994              Sec.getObject().IsMips64EL);
995   } else {
996     writeRel(Sec.Relocations, reinterpret_cast<Elf_Rela *>(Buf),
997              Sec.getObject().IsMips64EL);
998   }
999   return Error::success();
1000 }
1001 
accept(SectionVisitor & Visitor) const1002 Error RelocationSection::accept(SectionVisitor &Visitor) const {
1003   return Visitor.visit(*this);
1004 }
1005 
accept(MutableSectionVisitor & Visitor)1006 Error RelocationSection::accept(MutableSectionVisitor &Visitor) {
1007   return Visitor.visit(*this);
1008 }
1009 
removeSymbols(function_ref<bool (const Symbol &)> ToRemove)1010 Error RelocationSection::removeSymbols(
1011     function_ref<bool(const Symbol &)> ToRemove) {
1012   for (const Relocation &Reloc : Relocations)
1013     if (Reloc.RelocSymbol && ToRemove(*Reloc.RelocSymbol))
1014       return createStringError(
1015           llvm::errc::invalid_argument,
1016           "not stripping symbol '%s' because it is named in a relocation",
1017           Reloc.RelocSymbol->Name.data());
1018   return Error::success();
1019 }
1020 
markSymbols()1021 void RelocationSection::markSymbols() {
1022   for (const Relocation &Reloc : Relocations)
1023     if (Reloc.RelocSymbol)
1024       Reloc.RelocSymbol->Referenced = true;
1025 }
1026 
replaceSectionReferences(const DenseMap<SectionBase *,SectionBase * > & FromTo)1027 void RelocationSection::replaceSectionReferences(
1028     const DenseMap<SectionBase *, SectionBase *> &FromTo) {
1029   // Update the target section if it was replaced.
1030   if (SectionBase *To = FromTo.lookup(SecToApplyRel))
1031     SecToApplyRel = To;
1032 }
1033 
visit(const DynamicRelocationSection & Sec)1034 Error SectionWriter::visit(const DynamicRelocationSection &Sec) {
1035   llvm::copy(Sec.Contents, Out.getBufferStart() + Sec.Offset);
1036   return Error::success();
1037 }
1038 
accept(SectionVisitor & Visitor) const1039 Error DynamicRelocationSection::accept(SectionVisitor &Visitor) const {
1040   return Visitor.visit(*this);
1041 }
1042 
accept(MutableSectionVisitor & Visitor)1043 Error DynamicRelocationSection::accept(MutableSectionVisitor &Visitor) {
1044   return Visitor.visit(*this);
1045 }
1046 
removeSectionReferences(bool AllowBrokenLinks,function_ref<bool (const SectionBase *)> ToRemove)1047 Error DynamicRelocationSection::removeSectionReferences(
1048     bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
1049   if (ToRemove(Symbols)) {
1050     if (!AllowBrokenLinks)
1051       return createStringError(
1052           llvm::errc::invalid_argument,
1053           "symbol table '%s' cannot be removed because it is "
1054           "referenced by the relocation section '%s'",
1055           Symbols->Name.data(), this->Name.data());
1056     Symbols = nullptr;
1057   }
1058 
1059   // SecToApplyRel contains a section referenced by sh_info field. It keeps
1060   // a section to which the relocation section applies. When we remove any
1061   // sections we also remove their relocation sections. Since we do that much
1062   // earlier, this assert should never be triggered.
1063   assert(!SecToApplyRel || !ToRemove(SecToApplyRel));
1064   return Error::success();
1065 }
1066 
removeSectionReferences(bool AllowBrokenDependency,function_ref<bool (const SectionBase *)> ToRemove)1067 Error Section::removeSectionReferences(
1068     bool AllowBrokenDependency,
1069     function_ref<bool(const SectionBase *)> ToRemove) {
1070   if (ToRemove(LinkSection)) {
1071     if (!AllowBrokenDependency)
1072       return createStringError(llvm::errc::invalid_argument,
1073                                "section '%s' cannot be removed because it is "
1074                                "referenced by the section '%s'",
1075                                LinkSection->Name.data(), this->Name.data());
1076     LinkSection = nullptr;
1077   }
1078   return Error::success();
1079 }
1080 
finalize()1081 void GroupSection::finalize() {
1082   this->Info = Sym ? Sym->Index : 0;
1083   this->Link = SymTab ? SymTab->Index : 0;
1084   // Linker deduplication for GRP_COMDAT is based on Sym->Name. The local/global
1085   // status is not part of the equation. If Sym is localized, the intention is
1086   // likely to make the group fully localized. Drop GRP_COMDAT to suppress
1087   // deduplication. See https://groups.google.com/g/generic-abi/c/2X6mR-s2zoc
1088   if ((FlagWord & GRP_COMDAT) && Sym && Sym->Binding == STB_LOCAL)
1089     this->FlagWord &= ~GRP_COMDAT;
1090 }
1091 
removeSectionReferences(bool AllowBrokenLinks,function_ref<bool (const SectionBase *)> ToRemove)1092 Error GroupSection::removeSectionReferences(
1093     bool AllowBrokenLinks, function_ref<bool(const SectionBase *)> ToRemove) {
1094   if (ToRemove(SymTab)) {
1095     if (!AllowBrokenLinks)
1096       return createStringError(
1097           llvm::errc::invalid_argument,
1098           "section '.symtab' cannot be removed because it is "
1099           "referenced by the group section '%s'",
1100           this->Name.data());
1101     SymTab = nullptr;
1102     Sym = nullptr;
1103   }
1104   llvm::erase_if(GroupMembers, ToRemove);
1105   return Error::success();
1106 }
1107 
removeSymbols(function_ref<bool (const Symbol &)> ToRemove)1108 Error GroupSection::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
1109   if (ToRemove(*Sym))
1110     return createStringError(llvm::errc::invalid_argument,
1111                              "symbol '%s' cannot be removed because it is "
1112                              "referenced by the section '%s[%d]'",
1113                              Sym->Name.data(), this->Name.data(), this->Index);
1114   return Error::success();
1115 }
1116 
markSymbols()1117 void GroupSection::markSymbols() {
1118   if (Sym)
1119     Sym->Referenced = true;
1120 }
1121 
replaceSectionReferences(const DenseMap<SectionBase *,SectionBase * > & FromTo)1122 void GroupSection::replaceSectionReferences(
1123     const DenseMap<SectionBase *, SectionBase *> &FromTo) {
1124   for (SectionBase *&Sec : GroupMembers)
1125     if (SectionBase *To = FromTo.lookup(Sec))
1126       Sec = To;
1127 }
1128 
onRemove()1129 void GroupSection::onRemove() {
1130   // As the header section of the group is removed, drop the Group flag in its
1131   // former members.
1132   for (SectionBase *Sec : GroupMembers)
1133     Sec->Flags &= ~SHF_GROUP;
1134 }
1135 
initialize(SectionTableRef SecTable)1136 Error Section::initialize(SectionTableRef SecTable) {
1137   if (Link == ELF::SHN_UNDEF)
1138     return Error::success();
1139 
1140   Expected<SectionBase *> Sec =
1141       SecTable.getSection(Link, "Link field value " + Twine(Link) +
1142                                     " in section " + Name + " is invalid");
1143   if (!Sec)
1144     return Sec.takeError();
1145 
1146   LinkSection = *Sec;
1147 
1148   if (LinkSection->Type == ELF::SHT_SYMTAB) {
1149     HasSymTabLink = true;
1150     LinkSection = nullptr;
1151   }
1152 
1153   return Error::success();
1154 }
1155 
finalize()1156 void Section::finalize() { this->Link = LinkSection ? LinkSection->Index : 0; }
1157 
init(StringRef File)1158 void GnuDebugLinkSection::init(StringRef File) {
1159   FileName = sys::path::filename(File);
1160   // The format for the .gnu_debuglink starts with the file name and is
1161   // followed by a null terminator and then the CRC32 of the file. The CRC32
1162   // should be 4 byte aligned. So we add the FileName size, a 1 for the null
1163   // byte, and then finally push the size to alignment and add 4.
1164   Size = alignTo(FileName.size() + 1, 4) + 4;
1165   // The CRC32 will only be aligned if we align the whole section.
1166   Align = 4;
1167   Type = OriginalType = ELF::SHT_PROGBITS;
1168   Name = ".gnu_debuglink";
1169   // For sections not found in segments, OriginalOffset is only used to
1170   // establish the order that sections should go in. By using the maximum
1171   // possible offset we cause this section to wind up at the end.
1172   OriginalOffset = std::numeric_limits<uint64_t>::max();
1173 }
1174 
GnuDebugLinkSection(StringRef File,uint32_t PrecomputedCRC)1175 GnuDebugLinkSection::GnuDebugLinkSection(StringRef File,
1176                                          uint32_t PrecomputedCRC)
1177     : FileName(File), CRC32(PrecomputedCRC) {
1178   init(File);
1179 }
1180 
1181 template <class ELFT>
visit(const GnuDebugLinkSection & Sec)1182 Error ELFSectionWriter<ELFT>::visit(const GnuDebugLinkSection &Sec) {
1183   unsigned char *Buf =
1184       reinterpret_cast<uint8_t *>(Out.getBufferStart()) + Sec.Offset;
1185   Elf_Word *CRC =
1186       reinterpret_cast<Elf_Word *>(Buf + Sec.Size - sizeof(Elf_Word));
1187   *CRC = Sec.CRC32;
1188   llvm::copy(Sec.FileName, Buf);
1189   return Error::success();
1190 }
1191 
accept(SectionVisitor & Visitor) const1192 Error GnuDebugLinkSection::accept(SectionVisitor &Visitor) const {
1193   return Visitor.visit(*this);
1194 }
1195 
accept(MutableSectionVisitor & Visitor)1196 Error GnuDebugLinkSection::accept(MutableSectionVisitor &Visitor) {
1197   return Visitor.visit(*this);
1198 }
1199 
1200 template <class ELFT>
visit(const GroupSection & Sec)1201 Error ELFSectionWriter<ELFT>::visit(const GroupSection &Sec) {
1202   ELF::Elf32_Word *Buf =
1203       reinterpret_cast<ELF::Elf32_Word *>(Out.getBufferStart() + Sec.Offset);
1204   endian::write32<ELFT::Endianness>(Buf++, Sec.FlagWord);
1205   for (SectionBase *S : Sec.GroupMembers)
1206     endian::write32<ELFT::Endianness>(Buf++, S->Index);
1207   return Error::success();
1208 }
1209 
accept(SectionVisitor & Visitor) const1210 Error GroupSection::accept(SectionVisitor &Visitor) const {
1211   return Visitor.visit(*this);
1212 }
1213 
accept(MutableSectionVisitor & Visitor)1214 Error GroupSection::accept(MutableSectionVisitor &Visitor) {
1215   return Visitor.visit(*this);
1216 }
1217 
1218 // Returns true IFF a section is wholly inside the range of a segment
sectionWithinSegment(const SectionBase & Sec,const Segment & Seg)1219 static bool sectionWithinSegment(const SectionBase &Sec, const Segment &Seg) {
1220   // If a section is empty it should be treated like it has a size of 1. This is
1221   // to clarify the case when an empty section lies on a boundary between two
1222   // segments and ensures that the section "belongs" to the second segment and
1223   // not the first.
1224   uint64_t SecSize = Sec.Size ? Sec.Size : 1;
1225 
1226   // Ignore just added sections.
1227   if (Sec.OriginalOffset == std::numeric_limits<uint64_t>::max())
1228     return false;
1229 
1230   if (Sec.Type == SHT_NOBITS) {
1231     if (!(Sec.Flags & SHF_ALLOC))
1232       return false;
1233 
1234     bool SectionIsTLS = Sec.Flags & SHF_TLS;
1235     bool SegmentIsTLS = Seg.Type == PT_TLS;
1236     if (SectionIsTLS != SegmentIsTLS)
1237       return false;
1238 
1239     return Seg.VAddr <= Sec.Addr &&
1240            Seg.VAddr + Seg.MemSize >= Sec.Addr + SecSize;
1241   }
1242 
1243   return Seg.Offset <= Sec.OriginalOffset &&
1244          Seg.Offset + Seg.FileSize >= Sec.OriginalOffset + SecSize;
1245 }
1246 
1247 // Returns true IFF a segment's original offset is inside of another segment's
1248 // range.
segmentOverlapsSegment(const Segment & Child,const Segment & Parent)1249 static bool segmentOverlapsSegment(const Segment &Child,
1250                                    const Segment &Parent) {
1251 
1252   return Parent.OriginalOffset <= Child.OriginalOffset &&
1253          Parent.OriginalOffset + Parent.FileSize > Child.OriginalOffset;
1254 }
1255 
compareSegmentsByOffset(const Segment * A,const Segment * B)1256 static bool compareSegmentsByOffset(const Segment *A, const Segment *B) {
1257   // Any segment without a parent segment should come before a segment
1258   // that has a parent segment.
1259   if (A->OriginalOffset < B->OriginalOffset)
1260     return true;
1261   if (A->OriginalOffset > B->OriginalOffset)
1262     return false;
1263   // If alignments are different, the one with a smaller alignment cannot be the
1264   // parent; otherwise, layoutSegments will not respect the larger alignment
1265   // requirement. This rule ensures that PT_LOAD/PT_INTERP/PT_GNU_RELRO/PT_TLS
1266   // segments at the same offset will be aligned correctly.
1267   if (A->Align != B->Align)
1268     return A->Align > B->Align;
1269   return A->Index < B->Index;
1270 }
1271 
initFileHeader()1272 void BasicELFBuilder::initFileHeader() {
1273   Obj->Flags = 0x0;
1274   Obj->Type = ET_REL;
1275   Obj->OSABI = ELFOSABI_NONE;
1276   Obj->ABIVersion = 0;
1277   Obj->Entry = 0x0;
1278   Obj->Machine = EM_NONE;
1279   Obj->Version = 1;
1280 }
1281 
initHeaderSegment()1282 void BasicELFBuilder::initHeaderSegment() { Obj->ElfHdrSegment.Index = 0; }
1283 
addStrTab()1284 StringTableSection *BasicELFBuilder::addStrTab() {
1285   auto &StrTab = Obj->addSection<StringTableSection>();
1286   StrTab.Name = ".strtab";
1287 
1288   Obj->SectionNames = &StrTab;
1289   return &StrTab;
1290 }
1291 
addSymTab(StringTableSection * StrTab)1292 SymbolTableSection *BasicELFBuilder::addSymTab(StringTableSection *StrTab) {
1293   auto &SymTab = Obj->addSection<SymbolTableSection>();
1294 
1295   SymTab.Name = ".symtab";
1296   SymTab.Link = StrTab->Index;
1297 
1298   // The symbol table always needs a null symbol
1299   SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0);
1300 
1301   Obj->SymbolTable = &SymTab;
1302   return &SymTab;
1303 }
1304 
initSections()1305 Error BasicELFBuilder::initSections() {
1306   for (SectionBase &Sec : Obj->sections())
1307     if (Error Err = Sec.initialize(Obj->sections()))
1308       return Err;
1309 
1310   return Error::success();
1311 }
1312 
addData(SymbolTableSection * SymTab)1313 void BinaryELFBuilder::addData(SymbolTableSection *SymTab) {
1314   auto Data = ArrayRef<uint8_t>(
1315       reinterpret_cast<const uint8_t *>(MemBuf->getBufferStart()),
1316       MemBuf->getBufferSize());
1317   auto &DataSection = Obj->addSection<Section>(Data);
1318   DataSection.Name = ".data";
1319   DataSection.Type = ELF::SHT_PROGBITS;
1320   DataSection.Size = Data.size();
1321   DataSection.Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE;
1322 
1323   std::string SanitizedFilename = MemBuf->getBufferIdentifier().str();
1324   std::replace_if(
1325       std::begin(SanitizedFilename), std::end(SanitizedFilename),
1326       [](char C) { return !isAlnum(C); }, '_');
1327   Twine Prefix = Twine("_binary_") + SanitizedFilename;
1328 
1329   SymTab->addSymbol(Prefix + "_start", STB_GLOBAL, STT_NOTYPE, &DataSection,
1330                     /*Value=*/0, NewSymbolVisibility, 0, 0);
1331   SymTab->addSymbol(Prefix + "_end", STB_GLOBAL, STT_NOTYPE, &DataSection,
1332                     /*Value=*/DataSection.Size, NewSymbolVisibility, 0, 0);
1333   SymTab->addSymbol(Prefix + "_size", STB_GLOBAL, STT_NOTYPE, nullptr,
1334                     /*Value=*/DataSection.Size, NewSymbolVisibility, SHN_ABS,
1335                     0);
1336 }
1337 
build()1338 Expected<std::unique_ptr<Object>> BinaryELFBuilder::build() {
1339   initFileHeader();
1340   initHeaderSegment();
1341 
1342   SymbolTableSection *SymTab = addSymTab(addStrTab());
1343   if (Error Err = initSections())
1344     return std::move(Err);
1345   addData(SymTab);
1346 
1347   return std::move(Obj);
1348 }
1349 
1350 // Adds sections from IHEX data file. Data should have been
1351 // fully validated by this time.
addDataSections()1352 void IHexELFBuilder::addDataSections() {
1353   OwnedDataSection *Section = nullptr;
1354   uint64_t SegmentAddr = 0, BaseAddr = 0;
1355   uint32_t SecNo = 1;
1356 
1357   for (const IHexRecord &R : Records) {
1358     uint64_t RecAddr;
1359     switch (R.Type) {
1360     case IHexRecord::Data:
1361       // Ignore empty data records
1362       if (R.HexData.empty())
1363         continue;
1364       RecAddr = R.Addr + SegmentAddr + BaseAddr;
1365       if (!Section || Section->Addr + Section->Size != RecAddr) {
1366         // OriginalOffset field is only used to sort sections before layout, so
1367         // instead of keeping track of real offsets in IHEX file, and as
1368         // layoutSections() and layoutSectionsForOnlyKeepDebug() use
1369         // llvm::stable_sort(), we can just set it to a constant (zero).
1370         Section = &Obj->addSection<OwnedDataSection>(
1371             ".sec" + std::to_string(SecNo), RecAddr,
1372             ELF::SHF_ALLOC | ELF::SHF_WRITE, 0);
1373         SecNo++;
1374       }
1375       Section->appendHexData(R.HexData);
1376       break;
1377     case IHexRecord::EndOfFile:
1378       break;
1379     case IHexRecord::SegmentAddr:
1380       // 20-bit segment address.
1381       SegmentAddr = checkedGetHex<uint16_t>(R.HexData) << 4;
1382       break;
1383     case IHexRecord::StartAddr80x86:
1384     case IHexRecord::StartAddr:
1385       Obj->Entry = checkedGetHex<uint32_t>(R.HexData);
1386       assert(Obj->Entry <= 0xFFFFFU);
1387       break;
1388     case IHexRecord::ExtendedAddr:
1389       // 16-31 bits of linear base address
1390       BaseAddr = checkedGetHex<uint16_t>(R.HexData) << 16;
1391       break;
1392     default:
1393       llvm_unreachable("unknown record type");
1394     }
1395   }
1396 }
1397 
build()1398 Expected<std::unique_ptr<Object>> IHexELFBuilder::build() {
1399   initFileHeader();
1400   initHeaderSegment();
1401   StringTableSection *StrTab = addStrTab();
1402   addSymTab(StrTab);
1403   if (Error Err = initSections())
1404     return std::move(Err);
1405   addDataSections();
1406 
1407   return std::move(Obj);
1408 }
1409 
1410 template <class ELFT>
ELFBuilder(const ELFObjectFile<ELFT> & ElfObj,Object & Obj,std::optional<StringRef> ExtractPartition)1411 ELFBuilder<ELFT>::ELFBuilder(const ELFObjectFile<ELFT> &ElfObj, Object &Obj,
1412                              std::optional<StringRef> ExtractPartition)
1413     : ElfFile(ElfObj.getELFFile()), Obj(Obj),
1414       ExtractPartition(ExtractPartition) {
1415   Obj.IsMips64EL = ElfFile.isMips64EL();
1416 }
1417 
setParentSegment(Segment & Child)1418 template <class ELFT> void ELFBuilder<ELFT>::setParentSegment(Segment &Child) {
1419   for (Segment &Parent : Obj.segments()) {
1420     // Every segment will overlap with itself but we don't want a segment to
1421     // be its own parent so we avoid that situation.
1422     if (&Child != &Parent && segmentOverlapsSegment(Child, Parent)) {
1423       // We want a canonical "most parental" segment but this requires
1424       // inspecting the ParentSegment.
1425       if (compareSegmentsByOffset(&Parent, &Child))
1426         if (Child.ParentSegment == nullptr ||
1427             compareSegmentsByOffset(&Parent, Child.ParentSegment)) {
1428           Child.ParentSegment = &Parent;
1429         }
1430     }
1431   }
1432 }
1433 
findEhdrOffset()1434 template <class ELFT> Error ELFBuilder<ELFT>::findEhdrOffset() {
1435   if (!ExtractPartition)
1436     return Error::success();
1437 
1438   for (const SectionBase &Sec : Obj.sections()) {
1439     if (Sec.Type == SHT_LLVM_PART_EHDR && Sec.Name == *ExtractPartition) {
1440       EhdrOffset = Sec.Offset;
1441       return Error::success();
1442     }
1443   }
1444   return createStringError(errc::invalid_argument,
1445                            "could not find partition named '" +
1446                                *ExtractPartition + "'");
1447 }
1448 
1449 template <class ELFT>
readProgramHeaders(const ELFFile<ELFT> & HeadersFile)1450 Error ELFBuilder<ELFT>::readProgramHeaders(const ELFFile<ELFT> &HeadersFile) {
1451   uint32_t Index = 0;
1452 
1453   Expected<typename ELFFile<ELFT>::Elf_Phdr_Range> Headers =
1454       HeadersFile.program_headers();
1455   if (!Headers)
1456     return Headers.takeError();
1457 
1458   for (const typename ELFFile<ELFT>::Elf_Phdr &Phdr : *Headers) {
1459     if (Phdr.p_offset + Phdr.p_filesz > HeadersFile.getBufSize())
1460       return createStringError(
1461           errc::invalid_argument,
1462           "program header with offset 0x" + Twine::utohexstr(Phdr.p_offset) +
1463               " and file size 0x" + Twine::utohexstr(Phdr.p_filesz) +
1464               " goes past the end of the file");
1465 
1466     ArrayRef<uint8_t> Data{HeadersFile.base() + Phdr.p_offset,
1467                            (size_t)Phdr.p_filesz};
1468     Segment &Seg = Obj.addSegment(Data);
1469     Seg.Type = Phdr.p_type;
1470     Seg.Flags = Phdr.p_flags;
1471     Seg.OriginalOffset = Phdr.p_offset + EhdrOffset;
1472     Seg.Offset = Phdr.p_offset + EhdrOffset;
1473     Seg.VAddr = Phdr.p_vaddr;
1474     Seg.PAddr = Phdr.p_paddr;
1475     Seg.FileSize = Phdr.p_filesz;
1476     Seg.MemSize = Phdr.p_memsz;
1477     Seg.Align = Phdr.p_align;
1478     Seg.Index = Index++;
1479     for (SectionBase &Sec : Obj.sections())
1480       if (sectionWithinSegment(Sec, Seg)) {
1481         Seg.addSection(&Sec);
1482         if (!Sec.ParentSegment || Sec.ParentSegment->Offset > Seg.Offset)
1483           Sec.ParentSegment = &Seg;
1484       }
1485   }
1486 
1487   auto &ElfHdr = Obj.ElfHdrSegment;
1488   ElfHdr.Index = Index++;
1489   ElfHdr.OriginalOffset = ElfHdr.Offset = EhdrOffset;
1490 
1491   const typename ELFT::Ehdr &Ehdr = HeadersFile.getHeader();
1492   auto &PrHdr = Obj.ProgramHdrSegment;
1493   PrHdr.Type = PT_PHDR;
1494   PrHdr.Flags = 0;
1495   // The spec requires us to have p_vaddr % p_align == p_offset % p_align.
1496   // Whereas this works automatically for ElfHdr, here OriginalOffset is
1497   // always non-zero and to ensure the equation we assign the same value to
1498   // VAddr as well.
1499   PrHdr.OriginalOffset = PrHdr.Offset = PrHdr.VAddr = EhdrOffset + Ehdr.e_phoff;
1500   PrHdr.PAddr = 0;
1501   PrHdr.FileSize = PrHdr.MemSize = Ehdr.e_phentsize * Ehdr.e_phnum;
1502   // The spec requires us to naturally align all the fields.
1503   PrHdr.Align = sizeof(Elf_Addr);
1504   PrHdr.Index = Index++;
1505 
1506   // Now we do an O(n^2) loop through the segments in order to match up
1507   // segments.
1508   for (Segment &Child : Obj.segments())
1509     setParentSegment(Child);
1510   setParentSegment(ElfHdr);
1511   setParentSegment(PrHdr);
1512 
1513   return Error::success();
1514 }
1515 
1516 template <class ELFT>
initGroupSection(GroupSection * GroupSec)1517 Error ELFBuilder<ELFT>::initGroupSection(GroupSection *GroupSec) {
1518   if (GroupSec->Align % sizeof(ELF::Elf32_Word) != 0)
1519     return createStringError(errc::invalid_argument,
1520                              "invalid alignment " + Twine(GroupSec->Align) +
1521                                  " of group section '" + GroupSec->Name + "'");
1522   SectionTableRef SecTable = Obj.sections();
1523   if (GroupSec->Link != SHN_UNDEF) {
1524     auto SymTab = SecTable.template getSectionOfType<SymbolTableSection>(
1525         GroupSec->Link,
1526         "link field value '" + Twine(GroupSec->Link) + "' in section '" +
1527             GroupSec->Name + "' is invalid",
1528         "link field value '" + Twine(GroupSec->Link) + "' in section '" +
1529             GroupSec->Name + "' is not a symbol table");
1530     if (!SymTab)
1531       return SymTab.takeError();
1532 
1533     Expected<Symbol *> Sym = (*SymTab)->getSymbolByIndex(GroupSec->Info);
1534     if (!Sym)
1535       return createStringError(errc::invalid_argument,
1536                                "info field value '" + Twine(GroupSec->Info) +
1537                                    "' in section '" + GroupSec->Name +
1538                                    "' is not a valid symbol index");
1539     GroupSec->setSymTab(*SymTab);
1540     GroupSec->setSymbol(*Sym);
1541   }
1542   if (GroupSec->Contents.size() % sizeof(ELF::Elf32_Word) ||
1543       GroupSec->Contents.empty())
1544     return createStringError(errc::invalid_argument,
1545                              "the content of the section " + GroupSec->Name +
1546                                  " is malformed");
1547   const ELF::Elf32_Word *Word =
1548       reinterpret_cast<const ELF::Elf32_Word *>(GroupSec->Contents.data());
1549   const ELF::Elf32_Word *End =
1550       Word + GroupSec->Contents.size() / sizeof(ELF::Elf32_Word);
1551   GroupSec->setFlagWord(endian::read32<ELFT::Endianness>(Word++));
1552   for (; Word != End; ++Word) {
1553     uint32_t Index = support::endian::read32<ELFT::Endianness>(Word);
1554     Expected<SectionBase *> Sec = SecTable.getSection(
1555         Index, "group member index " + Twine(Index) + " in section '" +
1556                    GroupSec->Name + "' is invalid");
1557     if (!Sec)
1558       return Sec.takeError();
1559 
1560     GroupSec->addMember(*Sec);
1561   }
1562 
1563   return Error::success();
1564 }
1565 
1566 template <class ELFT>
initSymbolTable(SymbolTableSection * SymTab)1567 Error ELFBuilder<ELFT>::initSymbolTable(SymbolTableSection *SymTab) {
1568   Expected<const Elf_Shdr *> Shdr = ElfFile.getSection(SymTab->Index);
1569   if (!Shdr)
1570     return Shdr.takeError();
1571 
1572   Expected<StringRef> StrTabData = ElfFile.getStringTableForSymtab(**Shdr);
1573   if (!StrTabData)
1574     return StrTabData.takeError();
1575 
1576   ArrayRef<Elf_Word> ShndxData;
1577 
1578   Expected<typename ELFFile<ELFT>::Elf_Sym_Range> Symbols =
1579       ElfFile.symbols(*Shdr);
1580   if (!Symbols)
1581     return Symbols.takeError();
1582 
1583   for (const typename ELFFile<ELFT>::Elf_Sym &Sym : *Symbols) {
1584     SectionBase *DefSection = nullptr;
1585 
1586     Expected<StringRef> Name = Sym.getName(*StrTabData);
1587     if (!Name)
1588       return Name.takeError();
1589 
1590     if (Sym.st_shndx == SHN_XINDEX) {
1591       if (SymTab->getShndxTable() == nullptr)
1592         return createStringError(errc::invalid_argument,
1593                                  "symbol '" + *Name +
1594                                      "' has index SHN_XINDEX but no "
1595                                      "SHT_SYMTAB_SHNDX section exists");
1596       if (ShndxData.data() == nullptr) {
1597         Expected<const Elf_Shdr *> ShndxSec =
1598             ElfFile.getSection(SymTab->getShndxTable()->Index);
1599         if (!ShndxSec)
1600           return ShndxSec.takeError();
1601 
1602         Expected<ArrayRef<Elf_Word>> Data =
1603             ElfFile.template getSectionContentsAsArray<Elf_Word>(**ShndxSec);
1604         if (!Data)
1605           return Data.takeError();
1606 
1607         ShndxData = *Data;
1608         if (ShndxData.size() != Symbols->size())
1609           return createStringError(
1610               errc::invalid_argument,
1611               "symbol section index table does not have the same number of "
1612               "entries as the symbol table");
1613       }
1614       Elf_Word Index = ShndxData[&Sym - Symbols->begin()];
1615       Expected<SectionBase *> Sec = Obj.sections().getSection(
1616           Index,
1617           "symbol '" + *Name + "' has invalid section index " + Twine(Index));
1618       if (!Sec)
1619         return Sec.takeError();
1620 
1621       DefSection = *Sec;
1622     } else if (Sym.st_shndx >= SHN_LORESERVE) {
1623       if (!isValidReservedSectionIndex(Sym.st_shndx, Obj.Machine)) {
1624         return createStringError(
1625             errc::invalid_argument,
1626             "symbol '" + *Name +
1627                 "' has unsupported value greater than or equal "
1628                 "to SHN_LORESERVE: " +
1629                 Twine(Sym.st_shndx));
1630       }
1631     } else if (Sym.st_shndx != SHN_UNDEF) {
1632       Expected<SectionBase *> Sec = Obj.sections().getSection(
1633           Sym.st_shndx, "symbol '" + *Name +
1634                             "' is defined has invalid section index " +
1635                             Twine(Sym.st_shndx));
1636       if (!Sec)
1637         return Sec.takeError();
1638 
1639       DefSection = *Sec;
1640     }
1641 
1642     SymTab->addSymbol(*Name, Sym.getBinding(), Sym.getType(), DefSection,
1643                       Sym.getValue(), Sym.st_other, Sym.st_shndx, Sym.st_size);
1644   }
1645 
1646   return Error::success();
1647 }
1648 
1649 template <class ELFT>
getAddend(uint64_t &,const Elf_Rel_Impl<ELFT,false> &)1650 static void getAddend(uint64_t &, const Elf_Rel_Impl<ELFT, false> &) {}
1651 
1652 template <class ELFT>
getAddend(uint64_t & ToSet,const Elf_Rel_Impl<ELFT,true> & Rela)1653 static void getAddend(uint64_t &ToSet, const Elf_Rel_Impl<ELFT, true> &Rela) {
1654   ToSet = Rela.r_addend;
1655 }
1656 
1657 template <class T>
initRelocations(RelocationSection * Relocs,T RelRange)1658 static Error initRelocations(RelocationSection *Relocs, T RelRange) {
1659   for (const auto &Rel : RelRange) {
1660     Relocation ToAdd;
1661     ToAdd.Offset = Rel.r_offset;
1662     getAddend(ToAdd.Addend, Rel);
1663     ToAdd.Type = Rel.getType(Relocs->getObject().IsMips64EL);
1664 
1665     if (uint32_t Sym = Rel.getSymbol(Relocs->getObject().IsMips64EL)) {
1666       if (!Relocs->getObject().SymbolTable)
1667         return createStringError(
1668             errc::invalid_argument,
1669             "'" + Relocs->Name + "': relocation references symbol with index " +
1670                 Twine(Sym) + ", but there is no symbol table");
1671       Expected<Symbol *> SymByIndex =
1672           Relocs->getObject().SymbolTable->getSymbolByIndex(Sym);
1673       if (!SymByIndex)
1674         return SymByIndex.takeError();
1675 
1676       ToAdd.RelocSymbol = *SymByIndex;
1677     }
1678 
1679     Relocs->addRelocation(ToAdd);
1680   }
1681 
1682   return Error::success();
1683 }
1684 
getSection(uint32_t Index,Twine ErrMsg)1685 Expected<SectionBase *> SectionTableRef::getSection(uint32_t Index,
1686                                                     Twine ErrMsg) {
1687   if (Index == SHN_UNDEF || Index > Sections.size())
1688     return createStringError(errc::invalid_argument, ErrMsg);
1689   return Sections[Index - 1].get();
1690 }
1691 
1692 template <class T>
getSectionOfType(uint32_t Index,Twine IndexErrMsg,Twine TypeErrMsg)1693 Expected<T *> SectionTableRef::getSectionOfType(uint32_t Index,
1694                                                 Twine IndexErrMsg,
1695                                                 Twine TypeErrMsg) {
1696   Expected<SectionBase *> BaseSec = getSection(Index, IndexErrMsg);
1697   if (!BaseSec)
1698     return BaseSec.takeError();
1699 
1700   if (T *Sec = dyn_cast<T>(*BaseSec))
1701     return Sec;
1702 
1703   return createStringError(errc::invalid_argument, TypeErrMsg);
1704 }
1705 
1706 template <class ELFT>
makeSection(const Elf_Shdr & Shdr)1707 Expected<SectionBase &> ELFBuilder<ELFT>::makeSection(const Elf_Shdr &Shdr) {
1708   switch (Shdr.sh_type) {
1709   case SHT_REL:
1710   case SHT_RELA:
1711   case SHT_CREL:
1712     if (Shdr.sh_flags & SHF_ALLOC) {
1713       if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1714         return Obj.addSection<DynamicRelocationSection>(*Data);
1715       else
1716         return Data.takeError();
1717     }
1718     return Obj.addSection<RelocationSection>(Obj);
1719   case SHT_STRTAB:
1720     // If a string table is allocated we don't want to mess with it. That would
1721     // mean altering the memory image. There are no special link types or
1722     // anything so we can just use a Section.
1723     if (Shdr.sh_flags & SHF_ALLOC) {
1724       if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1725         return Obj.addSection<Section>(*Data);
1726       else
1727         return Data.takeError();
1728     }
1729     return Obj.addSection<StringTableSection>();
1730   case SHT_HASH:
1731   case SHT_GNU_HASH:
1732     // Hash tables should refer to SHT_DYNSYM which we're not going to change.
1733     // Because of this we don't need to mess with the hash tables either.
1734     if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1735       return Obj.addSection<Section>(*Data);
1736     else
1737       return Data.takeError();
1738   case SHT_GROUP:
1739     if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1740       return Obj.addSection<GroupSection>(*Data);
1741     else
1742       return Data.takeError();
1743   case SHT_DYNSYM:
1744     if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1745       return Obj.addSection<DynamicSymbolTableSection>(*Data);
1746     else
1747       return Data.takeError();
1748   case SHT_DYNAMIC:
1749     if (Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr))
1750       return Obj.addSection<DynamicSection>(*Data);
1751     else
1752       return Data.takeError();
1753   case SHT_SYMTAB: {
1754     // Multiple SHT_SYMTAB sections are forbidden by the ELF gABI.
1755     if (Obj.SymbolTable != nullptr)
1756       return createStringError(llvm::errc::invalid_argument,
1757                                "found multiple SHT_SYMTAB sections");
1758     auto &SymTab = Obj.addSection<SymbolTableSection>();
1759     Obj.SymbolTable = &SymTab;
1760     return SymTab;
1761   }
1762   case SHT_SYMTAB_SHNDX: {
1763     auto &ShndxSection = Obj.addSection<SectionIndexSection>();
1764     Obj.SectionIndexTable = &ShndxSection;
1765     return ShndxSection;
1766   }
1767   case SHT_NOBITS:
1768     return Obj.addSection<Section>(ArrayRef<uint8_t>());
1769   default: {
1770     Expected<ArrayRef<uint8_t>> Data = ElfFile.getSectionContents(Shdr);
1771     if (!Data)
1772       return Data.takeError();
1773 
1774     Expected<StringRef> Name = ElfFile.getSectionName(Shdr);
1775     if (!Name)
1776       return Name.takeError();
1777 
1778     if (!(Shdr.sh_flags & ELF::SHF_COMPRESSED))
1779       return Obj.addSection<Section>(*Data);
1780     auto *Chdr = reinterpret_cast<const Elf_Chdr_Impl<ELFT> *>(Data->data());
1781     return Obj.addSection<CompressedSection>(CompressedSection(
1782         *Data, Chdr->ch_type, Chdr->ch_size, Chdr->ch_addralign));
1783   }
1784   }
1785 }
1786 
readSectionHeaders()1787 template <class ELFT> Error ELFBuilder<ELFT>::readSectionHeaders() {
1788   uint32_t Index = 0;
1789   Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections =
1790       ElfFile.sections();
1791   if (!Sections)
1792     return Sections.takeError();
1793 
1794   for (const typename ELFFile<ELFT>::Elf_Shdr &Shdr : *Sections) {
1795     if (Index == 0) {
1796       ++Index;
1797       continue;
1798     }
1799     Expected<SectionBase &> Sec = makeSection(Shdr);
1800     if (!Sec)
1801       return Sec.takeError();
1802 
1803     Expected<StringRef> SecName = ElfFile.getSectionName(Shdr);
1804     if (!SecName)
1805       return SecName.takeError();
1806     Sec->Name = SecName->str();
1807     Sec->Type = Sec->OriginalType = Shdr.sh_type;
1808     Sec->Flags = Sec->OriginalFlags = Shdr.sh_flags;
1809     Sec->Addr = Shdr.sh_addr;
1810     Sec->Offset = Shdr.sh_offset;
1811     Sec->OriginalOffset = Shdr.sh_offset;
1812     Sec->Size = Shdr.sh_size;
1813     Sec->Link = Shdr.sh_link;
1814     Sec->Info = Shdr.sh_info;
1815     Sec->Align = Shdr.sh_addralign;
1816     Sec->EntrySize = Shdr.sh_entsize;
1817     Sec->Index = Index++;
1818     Sec->OriginalIndex = Sec->Index;
1819     Sec->OriginalData = ArrayRef<uint8_t>(
1820         ElfFile.base() + Shdr.sh_offset,
1821         (Shdr.sh_type == SHT_NOBITS) ? (size_t)0 : Shdr.sh_size);
1822   }
1823 
1824   return Error::success();
1825 }
1826 
readSections(bool EnsureSymtab)1827 template <class ELFT> Error ELFBuilder<ELFT>::readSections(bool EnsureSymtab) {
1828   uint32_t ShstrIndex = ElfFile.getHeader().e_shstrndx;
1829   if (ShstrIndex == SHN_XINDEX) {
1830     Expected<const Elf_Shdr *> Sec = ElfFile.getSection(0);
1831     if (!Sec)
1832       return Sec.takeError();
1833 
1834     ShstrIndex = (*Sec)->sh_link;
1835   }
1836 
1837   if (ShstrIndex == SHN_UNDEF)
1838     Obj.HadShdrs = false;
1839   else {
1840     Expected<StringTableSection *> Sec =
1841         Obj.sections().template getSectionOfType<StringTableSection>(
1842             ShstrIndex,
1843             "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " +
1844                 " is invalid",
1845             "e_shstrndx field value " + Twine(ShstrIndex) + " in elf header " +
1846                 " does not reference a string table");
1847     if (!Sec)
1848       return Sec.takeError();
1849 
1850     Obj.SectionNames = *Sec;
1851   }
1852 
1853   // If a section index table exists we'll need to initialize it before we
1854   // initialize the symbol table because the symbol table might need to
1855   // reference it.
1856   if (Obj.SectionIndexTable)
1857     if (Error Err = Obj.SectionIndexTable->initialize(Obj.sections()))
1858       return Err;
1859 
1860   // Now that all of the sections have been added we can fill out some extra
1861   // details about symbol tables. We need the symbol table filled out before
1862   // any relocations.
1863   if (Obj.SymbolTable) {
1864     if (Error Err = Obj.SymbolTable->initialize(Obj.sections()))
1865       return Err;
1866     if (Error Err = initSymbolTable(Obj.SymbolTable))
1867       return Err;
1868   } else if (EnsureSymtab) {
1869     if (Error Err = Obj.addNewSymbolTable())
1870       return Err;
1871   }
1872 
1873   // Now that all sections and symbols have been added we can add
1874   // relocations that reference symbols and set the link and info fields for
1875   // relocation sections.
1876   for (SectionBase &Sec : Obj.sections()) {
1877     if (&Sec == Obj.SymbolTable)
1878       continue;
1879     if (Error Err = Sec.initialize(Obj.sections()))
1880       return Err;
1881     if (auto RelSec = dyn_cast<RelocationSection>(&Sec)) {
1882       Expected<typename ELFFile<ELFT>::Elf_Shdr_Range> Sections =
1883           ElfFile.sections();
1884       if (!Sections)
1885         return Sections.takeError();
1886 
1887       const typename ELFFile<ELFT>::Elf_Shdr *Shdr =
1888           Sections->begin() + RelSec->Index;
1889       if (RelSec->Type == SHT_CREL) {
1890         auto RelsOrRelas = ElfFile.crels(*Shdr);
1891         if (!RelsOrRelas)
1892           return RelsOrRelas.takeError();
1893         if (Error Err = initRelocations(RelSec, RelsOrRelas->first))
1894           return Err;
1895         if (Error Err = initRelocations(RelSec, RelsOrRelas->second))
1896           return Err;
1897       } else if (RelSec->Type == SHT_REL) {
1898         Expected<typename ELFFile<ELFT>::Elf_Rel_Range> Rels =
1899             ElfFile.rels(*Shdr);
1900         if (!Rels)
1901           return Rels.takeError();
1902 
1903         if (Error Err = initRelocations(RelSec, *Rels))
1904           return Err;
1905       } else {
1906         Expected<typename ELFFile<ELFT>::Elf_Rela_Range> Relas =
1907             ElfFile.relas(*Shdr);
1908         if (!Relas)
1909           return Relas.takeError();
1910 
1911         if (Error Err = initRelocations(RelSec, *Relas))
1912           return Err;
1913       }
1914     } else if (auto GroupSec = dyn_cast<GroupSection>(&Sec)) {
1915       if (Error Err = initGroupSection(GroupSec))
1916         return Err;
1917     }
1918   }
1919 
1920   return Error::success();
1921 }
1922 
build(bool EnsureSymtab)1923 template <class ELFT> Error ELFBuilder<ELFT>::build(bool EnsureSymtab) {
1924   if (Error E = readSectionHeaders())
1925     return E;
1926   if (Error E = findEhdrOffset())
1927     return E;
1928 
1929   // The ELFFile whose ELF headers and program headers are copied into the
1930   // output file. Normally the same as ElfFile, but if we're extracting a
1931   // loadable partition it will point to the partition's headers.
1932   Expected<ELFFile<ELFT>> HeadersFile = ELFFile<ELFT>::create(toStringRef(
1933       {ElfFile.base() + EhdrOffset, ElfFile.getBufSize() - EhdrOffset}));
1934   if (!HeadersFile)
1935     return HeadersFile.takeError();
1936 
1937   const typename ELFFile<ELFT>::Elf_Ehdr &Ehdr = HeadersFile->getHeader();
1938   Obj.Is64Bits = Ehdr.e_ident[EI_CLASS] == ELFCLASS64;
1939   Obj.OSABI = Ehdr.e_ident[EI_OSABI];
1940   Obj.ABIVersion = Ehdr.e_ident[EI_ABIVERSION];
1941   Obj.Type = Ehdr.e_type;
1942   Obj.Machine = Ehdr.e_machine;
1943   Obj.Version = Ehdr.e_version;
1944   Obj.Entry = Ehdr.e_entry;
1945   Obj.Flags = Ehdr.e_flags;
1946 
1947   if (Error E = readSections(EnsureSymtab))
1948     return E;
1949   return readProgramHeaders(*HeadersFile);
1950 }
1951 
1952 Writer::~Writer() = default;
1953 
1954 Reader::~Reader() = default;
1955 
1956 Expected<std::unique_ptr<Object>>
create(bool) const1957 BinaryReader::create(bool /*EnsureSymtab*/) const {
1958   return BinaryELFBuilder(MemBuf, NewSymbolVisibility).build();
1959 }
1960 
parse() const1961 Expected<std::vector<IHexRecord>> IHexReader::parse() const {
1962   SmallVector<StringRef, 16> Lines;
1963   std::vector<IHexRecord> Records;
1964   bool HasSections = false;
1965 
1966   MemBuf->getBuffer().split(Lines, '\n');
1967   Records.reserve(Lines.size());
1968   for (size_t LineNo = 1; LineNo <= Lines.size(); ++LineNo) {
1969     StringRef Line = Lines[LineNo - 1].trim();
1970     if (Line.empty())
1971       continue;
1972 
1973     Expected<IHexRecord> R = IHexRecord::parse(Line);
1974     if (!R)
1975       return parseError(LineNo, R.takeError());
1976     if (R->Type == IHexRecord::EndOfFile)
1977       break;
1978     HasSections |= (R->Type == IHexRecord::Data);
1979     Records.push_back(*R);
1980   }
1981   if (!HasSections)
1982     return parseError(-1U, "no sections");
1983 
1984   return std::move(Records);
1985 }
1986 
1987 Expected<std::unique_ptr<Object>>
create(bool) const1988 IHexReader::create(bool /*EnsureSymtab*/) const {
1989   Expected<std::vector<IHexRecord>> Records = parse();
1990   if (!Records)
1991     return Records.takeError();
1992 
1993   return IHexELFBuilder(*Records).build();
1994 }
1995 
create(bool EnsureSymtab) const1996 Expected<std::unique_ptr<Object>> ELFReader::create(bool EnsureSymtab) const {
1997   auto Obj = std::make_unique<Object>();
1998   if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Bin)) {
1999     ELFBuilder<ELF32LE> Builder(*O, *Obj, ExtractPartition);
2000     if (Error Err = Builder.build(EnsureSymtab))
2001       return std::move(Err);
2002     return std::move(Obj);
2003   } else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Bin)) {
2004     ELFBuilder<ELF64LE> Builder(*O, *Obj, ExtractPartition);
2005     if (Error Err = Builder.build(EnsureSymtab))
2006       return std::move(Err);
2007     return std::move(Obj);
2008   } else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Bin)) {
2009     ELFBuilder<ELF32BE> Builder(*O, *Obj, ExtractPartition);
2010     if (Error Err = Builder.build(EnsureSymtab))
2011       return std::move(Err);
2012     return std::move(Obj);
2013   } else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Bin)) {
2014     ELFBuilder<ELF64BE> Builder(*O, *Obj, ExtractPartition);
2015     if (Error Err = Builder.build(EnsureSymtab))
2016       return std::move(Err);
2017     return std::move(Obj);
2018   }
2019   return createStringError(errc::invalid_argument, "invalid file type");
2020 }
2021 
writeEhdr()2022 template <class ELFT> void ELFWriter<ELFT>::writeEhdr() {
2023   Elf_Ehdr &Ehdr = *reinterpret_cast<Elf_Ehdr *>(Buf->getBufferStart());
2024   std::fill(Ehdr.e_ident, Ehdr.e_ident + 16, 0);
2025   Ehdr.e_ident[EI_MAG0] = 0x7f;
2026   Ehdr.e_ident[EI_MAG1] = 'E';
2027   Ehdr.e_ident[EI_MAG2] = 'L';
2028   Ehdr.e_ident[EI_MAG3] = 'F';
2029   Ehdr.e_ident[EI_CLASS] = ELFT::Is64Bits ? ELFCLASS64 : ELFCLASS32;
2030   Ehdr.e_ident[EI_DATA] =
2031       ELFT::Endianness == llvm::endianness::big ? ELFDATA2MSB : ELFDATA2LSB;
2032   Ehdr.e_ident[EI_VERSION] = EV_CURRENT;
2033   Ehdr.e_ident[EI_OSABI] = Obj.OSABI;
2034   Ehdr.e_ident[EI_ABIVERSION] = Obj.ABIVersion;
2035 
2036   Ehdr.e_type = Obj.Type;
2037   Ehdr.e_machine = Obj.Machine;
2038   Ehdr.e_version = Obj.Version;
2039   Ehdr.e_entry = Obj.Entry;
2040   // We have to use the fully-qualified name llvm::size
2041   // since some compilers complain on ambiguous resolution.
2042   Ehdr.e_phnum = llvm::size(Obj.segments());
2043   Ehdr.e_phoff = (Ehdr.e_phnum != 0) ? Obj.ProgramHdrSegment.Offset : 0;
2044   Ehdr.e_phentsize = (Ehdr.e_phnum != 0) ? sizeof(Elf_Phdr) : 0;
2045   Ehdr.e_flags = Obj.Flags;
2046   Ehdr.e_ehsize = sizeof(Elf_Ehdr);
2047   if (WriteSectionHeaders && Obj.sections().size() != 0) {
2048     Ehdr.e_shentsize = sizeof(Elf_Shdr);
2049     Ehdr.e_shoff = Obj.SHOff;
2050     // """
2051     // If the number of sections is greater than or equal to
2052     // SHN_LORESERVE (0xff00), this member has the value zero and the actual
2053     // number of section header table entries is contained in the sh_size field
2054     // of the section header at index 0.
2055     // """
2056     auto Shnum = Obj.sections().size() + 1;
2057     if (Shnum >= SHN_LORESERVE)
2058       Ehdr.e_shnum = 0;
2059     else
2060       Ehdr.e_shnum = Shnum;
2061     // """
2062     // If the section name string table section index is greater than or equal
2063     // to SHN_LORESERVE (0xff00), this member has the value SHN_XINDEX (0xffff)
2064     // and the actual index of the section name string table section is
2065     // contained in the sh_link field of the section header at index 0.
2066     // """
2067     if (Obj.SectionNames->Index >= SHN_LORESERVE)
2068       Ehdr.e_shstrndx = SHN_XINDEX;
2069     else
2070       Ehdr.e_shstrndx = Obj.SectionNames->Index;
2071   } else {
2072     Ehdr.e_shentsize = 0;
2073     Ehdr.e_shoff = 0;
2074     Ehdr.e_shnum = 0;
2075     Ehdr.e_shstrndx = 0;
2076   }
2077 }
2078 
writePhdrs()2079 template <class ELFT> void ELFWriter<ELFT>::writePhdrs() {
2080   for (auto &Seg : Obj.segments())
2081     writePhdr(Seg);
2082 }
2083 
writeShdrs()2084 template <class ELFT> void ELFWriter<ELFT>::writeShdrs() {
2085   // This reference serves to write the dummy section header at the begining
2086   // of the file. It is not used for anything else
2087   Elf_Shdr &Shdr =
2088       *reinterpret_cast<Elf_Shdr *>(Buf->getBufferStart() + Obj.SHOff);
2089   Shdr.sh_name = 0;
2090   Shdr.sh_type = SHT_NULL;
2091   Shdr.sh_flags = 0;
2092   Shdr.sh_addr = 0;
2093   Shdr.sh_offset = 0;
2094   // See writeEhdr for why we do this.
2095   uint64_t Shnum = Obj.sections().size() + 1;
2096   if (Shnum >= SHN_LORESERVE)
2097     Shdr.sh_size = Shnum;
2098   else
2099     Shdr.sh_size = 0;
2100   // See writeEhdr for why we do this.
2101   if (Obj.SectionNames != nullptr && Obj.SectionNames->Index >= SHN_LORESERVE)
2102     Shdr.sh_link = Obj.SectionNames->Index;
2103   else
2104     Shdr.sh_link = 0;
2105   Shdr.sh_info = 0;
2106   Shdr.sh_addralign = 0;
2107   Shdr.sh_entsize = 0;
2108 
2109   for (SectionBase &Sec : Obj.sections())
2110     writeShdr(Sec);
2111 }
2112 
writeSectionData()2113 template <class ELFT> Error ELFWriter<ELFT>::writeSectionData() {
2114   for (SectionBase &Sec : Obj.sections())
2115     // Segments are responsible for writing their contents, so only write the
2116     // section data if the section is not in a segment. Note that this renders
2117     // sections in segments effectively immutable.
2118     if (Sec.ParentSegment == nullptr)
2119       if (Error Err = Sec.accept(*SecWriter))
2120         return Err;
2121 
2122   return Error::success();
2123 }
2124 
writeSegmentData()2125 template <class ELFT> void ELFWriter<ELFT>::writeSegmentData() {
2126   for (Segment &Seg : Obj.segments()) {
2127     size_t Size = std::min<size_t>(Seg.FileSize, Seg.getContents().size());
2128     std::memcpy(Buf->getBufferStart() + Seg.Offset, Seg.getContents().data(),
2129                 Size);
2130   }
2131 
2132   for (const auto &it : Obj.getUpdatedSections()) {
2133     SectionBase *Sec = it.first;
2134     ArrayRef<uint8_t> Data = it.second;
2135 
2136     auto *Parent = Sec->ParentSegment;
2137     assert(Parent && "This section should've been part of a segment.");
2138     uint64_t Offset =
2139         Sec->OriginalOffset - Parent->OriginalOffset + Parent->Offset;
2140     llvm::copy(Data, Buf->getBufferStart() + Offset);
2141   }
2142 
2143   // Iterate over removed sections and overwrite their old data with zeroes.
2144   for (auto &Sec : Obj.removedSections()) {
2145     Segment *Parent = Sec.ParentSegment;
2146     if (Parent == nullptr || Sec.Type == SHT_NOBITS || Sec.Size == 0)
2147       continue;
2148     uint64_t Offset =
2149         Sec.OriginalOffset - Parent->OriginalOffset + Parent->Offset;
2150     std::memset(Buf->getBufferStart() + Offset, 0, Sec.Size);
2151   }
2152 }
2153 
2154 template <class ELFT>
ELFWriter(Object & Obj,raw_ostream & Buf,bool WSH,bool OnlyKeepDebug)2155 ELFWriter<ELFT>::ELFWriter(Object &Obj, raw_ostream &Buf, bool WSH,
2156                            bool OnlyKeepDebug)
2157     : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs),
2158       OnlyKeepDebug(OnlyKeepDebug) {}
2159 
updateSection(StringRef Name,ArrayRef<uint8_t> Data)2160 Error Object::updateSection(StringRef Name, ArrayRef<uint8_t> Data) {
2161   auto It = llvm::find_if(Sections,
2162                           [&](const SecPtr &Sec) { return Sec->Name == Name; });
2163   if (It == Sections.end())
2164     return createStringError(errc::invalid_argument, "section '%s' not found",
2165                              Name.str().c_str());
2166 
2167   auto *OldSec = It->get();
2168   if (!OldSec->hasContents())
2169     return createStringError(
2170         errc::invalid_argument,
2171         "section '%s' cannot be updated because it does not have contents",
2172         Name.str().c_str());
2173 
2174   if (Data.size() > OldSec->Size && OldSec->ParentSegment)
2175     return createStringError(errc::invalid_argument,
2176                              "cannot fit data of size %zu into section '%s' "
2177                              "with size %" PRIu64 " that is part of a segment",
2178                              Data.size(), Name.str().c_str(), OldSec->Size);
2179 
2180   if (!OldSec->ParentSegment) {
2181     *It = std::make_unique<OwnedDataSection>(*OldSec, Data);
2182   } else {
2183     // The segment writer will be in charge of updating these contents.
2184     OldSec->Size = Data.size();
2185     UpdatedSections[OldSec] = Data;
2186   }
2187 
2188   return Error::success();
2189 }
2190 
removeSections(bool AllowBrokenLinks,std::function<bool (const SectionBase &)> ToRemove)2191 Error Object::removeSections(
2192     bool AllowBrokenLinks, std::function<bool(const SectionBase &)> ToRemove) {
2193 
2194   auto Iter = std::stable_partition(
2195       std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) {
2196         if (ToRemove(*Sec))
2197           return false;
2198         // TODO: A compressed relocation section may be recognized as
2199         // RelocationSectionBase. We don't want such a section to be removed.
2200         if (isa<CompressedSection>(Sec))
2201           return true;
2202         if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) {
2203           if (auto ToRelSec = RelSec->getSection())
2204             return !ToRemove(*ToRelSec);
2205         }
2206         // Remove empty group sections.
2207         if (Sec->Type == ELF::SHT_GROUP) {
2208           auto GroupSec = cast<GroupSection>(Sec.get());
2209           return !llvm::all_of(GroupSec->members(), ToRemove);
2210         }
2211         return true;
2212       });
2213   if (SymbolTable != nullptr && ToRemove(*SymbolTable))
2214     SymbolTable = nullptr;
2215   if (SectionNames != nullptr && ToRemove(*SectionNames))
2216     SectionNames = nullptr;
2217   if (SectionIndexTable != nullptr && ToRemove(*SectionIndexTable))
2218     SectionIndexTable = nullptr;
2219   // Now make sure there are no remaining references to the sections that will
2220   // be removed. Sometimes it is impossible to remove a reference so we emit
2221   // an error here instead.
2222   std::unordered_set<const SectionBase *> RemoveSections;
2223   RemoveSections.reserve(std::distance(Iter, std::end(Sections)));
2224   for (auto &RemoveSec : make_range(Iter, std::end(Sections))) {
2225     for (auto &Segment : Segments)
2226       Segment->removeSection(RemoveSec.get());
2227     RemoveSec->onRemove();
2228     RemoveSections.insert(RemoveSec.get());
2229   }
2230 
2231   // For each section that remains alive, we want to remove the dead references.
2232   // This either might update the content of the section (e.g. remove symbols
2233   // from symbol table that belongs to removed section) or trigger an error if
2234   // a live section critically depends on a section being removed somehow
2235   // (e.g. the removed section is referenced by a relocation).
2236   for (auto &KeepSec : make_range(std::begin(Sections), Iter)) {
2237     if (Error E = KeepSec->removeSectionReferences(
2238             AllowBrokenLinks, [&RemoveSections](const SectionBase *Sec) {
2239               return RemoveSections.find(Sec) != RemoveSections.end();
2240             }))
2241       return E;
2242   }
2243 
2244   // Transfer removed sections into the Object RemovedSections container for use
2245   // later.
2246   std::move(Iter, Sections.end(), std::back_inserter(RemovedSections));
2247   // Now finally get rid of them all together.
2248   Sections.erase(Iter, std::end(Sections));
2249   return Error::success();
2250 }
2251 
replaceSections(const DenseMap<SectionBase *,SectionBase * > & FromTo)2252 Error Object::replaceSections(
2253     const DenseMap<SectionBase *, SectionBase *> &FromTo) {
2254   auto SectionIndexLess = [](const SecPtr &Lhs, const SecPtr &Rhs) {
2255     return Lhs->Index < Rhs->Index;
2256   };
2257   assert(llvm::is_sorted(Sections, SectionIndexLess) &&
2258          "Sections are expected to be sorted by Index");
2259   // Set indices of new sections so that they can be later sorted into positions
2260   // of removed ones.
2261   for (auto &I : FromTo)
2262     I.second->Index = I.first->Index;
2263 
2264   // Notify all sections about the replacement.
2265   for (auto &Sec : Sections)
2266     Sec->replaceSectionReferences(FromTo);
2267 
2268   if (Error E = removeSections(
2269           /*AllowBrokenLinks=*/false,
2270           [=](const SectionBase &Sec) { return FromTo.count(&Sec) > 0; }))
2271     return E;
2272   llvm::sort(Sections, SectionIndexLess);
2273   return Error::success();
2274 }
2275 
removeSymbols(function_ref<bool (const Symbol &)> ToRemove)2276 Error Object::removeSymbols(function_ref<bool(const Symbol &)> ToRemove) {
2277   if (SymbolTable)
2278     for (const SecPtr &Sec : Sections)
2279       if (Error E = Sec->removeSymbols(ToRemove))
2280         return E;
2281   return Error::success();
2282 }
2283 
addNewSymbolTable()2284 Error Object::addNewSymbolTable() {
2285   assert(!SymbolTable && "Object must not has a SymbolTable.");
2286 
2287   // Reuse an existing SHT_STRTAB section if it exists.
2288   StringTableSection *StrTab = nullptr;
2289   for (SectionBase &Sec : sections()) {
2290     if (Sec.Type == ELF::SHT_STRTAB && !(Sec.Flags & SHF_ALLOC)) {
2291       StrTab = static_cast<StringTableSection *>(&Sec);
2292 
2293       // Prefer a string table that is not the section header string table, if
2294       // such a table exists.
2295       if (SectionNames != &Sec)
2296         break;
2297     }
2298   }
2299   if (!StrTab)
2300     StrTab = &addSection<StringTableSection>();
2301 
2302   SymbolTableSection &SymTab = addSection<SymbolTableSection>();
2303   SymTab.Name = ".symtab";
2304   SymTab.Link = StrTab->Index;
2305   if (Error Err = SymTab.initialize(sections()))
2306     return Err;
2307   SymTab.addSymbol("", 0, 0, nullptr, 0, 0, 0, 0);
2308 
2309   SymbolTable = &SymTab;
2310 
2311   return Error::success();
2312 }
2313 
2314 // Orders segments such that if x = y->ParentSegment then y comes before x.
orderSegments(std::vector<Segment * > & Segments)2315 static void orderSegments(std::vector<Segment *> &Segments) {
2316   llvm::stable_sort(Segments, compareSegmentsByOffset);
2317 }
2318 
2319 // This function finds a consistent layout for a list of segments starting from
2320 // an Offset. It assumes that Segments have been sorted by orderSegments and
2321 // returns an Offset one past the end of the last segment.
layoutSegments(std::vector<Segment * > & Segments,uint64_t Offset)2322 static uint64_t layoutSegments(std::vector<Segment *> &Segments,
2323                                uint64_t Offset) {
2324   assert(llvm::is_sorted(Segments, compareSegmentsByOffset));
2325   // The only way a segment should move is if a section was between two
2326   // segments and that section was removed. If that section isn't in a segment
2327   // then it's acceptable, but not ideal, to simply move it to after the
2328   // segments. So we can simply layout segments one after the other accounting
2329   // for alignment.
2330   for (Segment *Seg : Segments) {
2331     // We assume that segments have been ordered by OriginalOffset and Index
2332     // such that a parent segment will always come before a child segment in
2333     // OrderedSegments. This means that the Offset of the ParentSegment should
2334     // already be set and we can set our offset relative to it.
2335     if (Seg->ParentSegment != nullptr) {
2336       Segment *Parent = Seg->ParentSegment;
2337       Seg->Offset =
2338           Parent->Offset + Seg->OriginalOffset - Parent->OriginalOffset;
2339     } else {
2340       Seg->Offset =
2341           alignTo(Offset, std::max<uint64_t>(Seg->Align, 1), Seg->VAddr);
2342     }
2343     Offset = std::max(Offset, Seg->Offset + Seg->FileSize);
2344   }
2345   return Offset;
2346 }
2347 
2348 // This function finds a consistent layout for a list of sections. It assumes
2349 // that the ->ParentSegment of each section has already been laid out. The
2350 // supplied starting Offset is used for the starting offset of any section that
2351 // does not have a ParentSegment. It returns either the offset given if all
2352 // sections had a ParentSegment or an offset one past the last section if there
2353 // was a section that didn't have a ParentSegment.
2354 template <class Range>
layoutSections(Range Sections,uint64_t Offset)2355 static uint64_t layoutSections(Range Sections, uint64_t Offset) {
2356   // Now the offset of every segment has been set we can assign the offsets
2357   // of each section. For sections that are covered by a segment we should use
2358   // the segment's original offset and the section's original offset to compute
2359   // the offset from the start of the segment. Using the offset from the start
2360   // of the segment we can assign a new offset to the section. For sections not
2361   // covered by segments we can just bump Offset to the next valid location.
2362   // While it is not necessary, layout the sections in the order based on their
2363   // original offsets to resemble the input file as close as possible.
2364   std::vector<SectionBase *> OutOfSegmentSections;
2365   uint32_t Index = 1;
2366   for (auto &Sec : Sections) {
2367     Sec.Index = Index++;
2368     if (Sec.ParentSegment != nullptr) {
2369       const Segment &Segment = *Sec.ParentSegment;
2370       Sec.Offset =
2371           Segment.Offset + (Sec.OriginalOffset - Segment.OriginalOffset);
2372     } else
2373       OutOfSegmentSections.push_back(&Sec);
2374   }
2375 
2376   llvm::stable_sort(OutOfSegmentSections,
2377                     [](const SectionBase *Lhs, const SectionBase *Rhs) {
2378                       return Lhs->OriginalOffset < Rhs->OriginalOffset;
2379                     });
2380   for (auto *Sec : OutOfSegmentSections) {
2381     Offset = alignTo(Offset, Sec->Align == 0 ? 1 : Sec->Align);
2382     Sec->Offset = Offset;
2383     if (Sec->Type != SHT_NOBITS)
2384       Offset += Sec->Size;
2385   }
2386   return Offset;
2387 }
2388 
2389 // Rewrite sh_offset after some sections are changed to SHT_NOBITS and thus
2390 // occupy no space in the file.
layoutSectionsForOnlyKeepDebug(Object & Obj,uint64_t Off)2391 static uint64_t layoutSectionsForOnlyKeepDebug(Object &Obj, uint64_t Off) {
2392   // The layout algorithm requires the sections to be handled in the order of
2393   // their offsets in the input file, at least inside segments.
2394   std::vector<SectionBase *> Sections;
2395   Sections.reserve(Obj.sections().size());
2396   uint32_t Index = 1;
2397   for (auto &Sec : Obj.sections()) {
2398     Sec.Index = Index++;
2399     Sections.push_back(&Sec);
2400   }
2401   llvm::stable_sort(Sections,
2402                     [](const SectionBase *Lhs, const SectionBase *Rhs) {
2403                       return Lhs->OriginalOffset < Rhs->OriginalOffset;
2404                     });
2405 
2406   for (auto *Sec : Sections) {
2407     auto *FirstSec = Sec->ParentSegment && Sec->ParentSegment->Type == PT_LOAD
2408                          ? Sec->ParentSegment->firstSection()
2409                          : nullptr;
2410 
2411     // The first section in a PT_LOAD has to have congruent offset and address
2412     // modulo the alignment, which usually equals the maximum page size.
2413     if (FirstSec && FirstSec == Sec)
2414       Off = alignTo(Off, Sec->ParentSegment->Align, Sec->Addr);
2415 
2416     // sh_offset is not significant for SHT_NOBITS sections, but the congruence
2417     // rule must be followed if it is the first section in a PT_LOAD. Do not
2418     // advance Off.
2419     if (Sec->Type == SHT_NOBITS) {
2420       Sec->Offset = Off;
2421       continue;
2422     }
2423 
2424     if (!FirstSec) {
2425       // FirstSec being nullptr generally means that Sec does not have the
2426       // SHF_ALLOC flag.
2427       Off = Sec->Align ? alignTo(Off, Sec->Align) : Off;
2428     } else if (FirstSec != Sec) {
2429       // The offset is relative to the first section in the PT_LOAD segment. Use
2430       // sh_offset for non-SHF_ALLOC sections.
2431       Off = Sec->OriginalOffset - FirstSec->OriginalOffset + FirstSec->Offset;
2432     }
2433     Sec->Offset = Off;
2434     Off += Sec->Size;
2435   }
2436   return Off;
2437 }
2438 
2439 // Rewrite p_offset and p_filesz of non-PT_PHDR segments after sh_offset values
2440 // have been updated.
layoutSegmentsForOnlyKeepDebug(std::vector<Segment * > & Segments,uint64_t HdrEnd)2441 static uint64_t layoutSegmentsForOnlyKeepDebug(std::vector<Segment *> &Segments,
2442                                                uint64_t HdrEnd) {
2443   uint64_t MaxOffset = 0;
2444   for (Segment *Seg : Segments) {
2445     if (Seg->Type == PT_PHDR)
2446       continue;
2447 
2448     // The segment offset is generally the offset of the first section.
2449     //
2450     // For a segment containing no section (see sectionWithinSegment), if it has
2451     // a parent segment, copy the parent segment's offset field. This works for
2452     // empty PT_TLS. If no parent segment, use 0: the segment is not useful for
2453     // debugging anyway.
2454     const SectionBase *FirstSec = Seg->firstSection();
2455     uint64_t Offset =
2456         FirstSec ? FirstSec->Offset
2457                  : (Seg->ParentSegment ? Seg->ParentSegment->Offset : 0);
2458     uint64_t FileSize = 0;
2459     for (const SectionBase *Sec : Seg->Sections) {
2460       uint64_t Size = Sec->Type == SHT_NOBITS ? 0 : Sec->Size;
2461       if (Sec->Offset + Size > Offset)
2462         FileSize = std::max(FileSize, Sec->Offset + Size - Offset);
2463     }
2464 
2465     // If the segment includes EHDR and program headers, don't make it smaller
2466     // than the headers.
2467     if (Seg->Offset < HdrEnd && HdrEnd <= Seg->Offset + Seg->FileSize) {
2468       FileSize += Offset - Seg->Offset;
2469       Offset = Seg->Offset;
2470       FileSize = std::max(FileSize, HdrEnd - Offset);
2471     }
2472 
2473     Seg->Offset = Offset;
2474     Seg->FileSize = FileSize;
2475     MaxOffset = std::max(MaxOffset, Offset + FileSize);
2476   }
2477   return MaxOffset;
2478 }
2479 
initEhdrSegment()2480 template <class ELFT> void ELFWriter<ELFT>::initEhdrSegment() {
2481   Segment &ElfHdr = Obj.ElfHdrSegment;
2482   ElfHdr.Type = PT_PHDR;
2483   ElfHdr.Flags = 0;
2484   ElfHdr.VAddr = 0;
2485   ElfHdr.PAddr = 0;
2486   ElfHdr.FileSize = ElfHdr.MemSize = sizeof(Elf_Ehdr);
2487   ElfHdr.Align = 0;
2488 }
2489 
assignOffsets()2490 template <class ELFT> void ELFWriter<ELFT>::assignOffsets() {
2491   // We need a temporary list of segments that has a special order to it
2492   // so that we know that anytime ->ParentSegment is set that segment has
2493   // already had its offset properly set.
2494   std::vector<Segment *> OrderedSegments;
2495   for (Segment &Segment : Obj.segments())
2496     OrderedSegments.push_back(&Segment);
2497   OrderedSegments.push_back(&Obj.ElfHdrSegment);
2498   OrderedSegments.push_back(&Obj.ProgramHdrSegment);
2499   orderSegments(OrderedSegments);
2500 
2501   uint64_t Offset;
2502   if (OnlyKeepDebug) {
2503     // For --only-keep-debug, the sections that did not preserve contents were
2504     // changed to SHT_NOBITS. We now rewrite sh_offset fields of sections, and
2505     // then rewrite p_offset/p_filesz of program headers.
2506     uint64_t HdrEnd =
2507         sizeof(Elf_Ehdr) + llvm::size(Obj.segments()) * sizeof(Elf_Phdr);
2508     Offset = layoutSectionsForOnlyKeepDebug(Obj, HdrEnd);
2509     Offset = std::max(Offset,
2510                       layoutSegmentsForOnlyKeepDebug(OrderedSegments, HdrEnd));
2511   } else {
2512     // Offset is used as the start offset of the first segment to be laid out.
2513     // Since the ELF Header (ElfHdrSegment) must be at the start of the file,
2514     // we start at offset 0.
2515     Offset = layoutSegments(OrderedSegments, 0);
2516     Offset = layoutSections(Obj.sections(), Offset);
2517   }
2518   // If we need to write the section header table out then we need to align the
2519   // Offset so that SHOffset is valid.
2520   if (WriteSectionHeaders)
2521     Offset = alignTo(Offset, sizeof(Elf_Addr));
2522   Obj.SHOff = Offset;
2523 }
2524 
totalSize() const2525 template <class ELFT> size_t ELFWriter<ELFT>::totalSize() const {
2526   // We already have the section header offset so we can calculate the total
2527   // size by just adding up the size of each section header.
2528   if (!WriteSectionHeaders)
2529     return Obj.SHOff;
2530   size_t ShdrCount = Obj.sections().size() + 1; // Includes null shdr.
2531   return Obj.SHOff + ShdrCount * sizeof(Elf_Shdr);
2532 }
2533 
write()2534 template <class ELFT> Error ELFWriter<ELFT>::write() {
2535   // Segment data must be written first, so that the ELF header and program
2536   // header tables can overwrite it, if covered by a segment.
2537   writeSegmentData();
2538   writeEhdr();
2539   writePhdrs();
2540   if (Error E = writeSectionData())
2541     return E;
2542   if (WriteSectionHeaders)
2543     writeShdrs();
2544 
2545   // TODO: Implement direct writing to the output stream (without intermediate
2546   // memory buffer Buf).
2547   Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2548   return Error::success();
2549 }
2550 
removeUnneededSections(Object & Obj)2551 static Error removeUnneededSections(Object &Obj) {
2552   // We can remove an empty symbol table from non-relocatable objects.
2553   // Relocatable objects typically have relocation sections whose
2554   // sh_link field points to .symtab, so we can't remove .symtab
2555   // even if it is empty.
2556   if (Obj.isRelocatable() || Obj.SymbolTable == nullptr ||
2557       !Obj.SymbolTable->empty())
2558     return Error::success();
2559 
2560   // .strtab can be used for section names. In such a case we shouldn't
2561   // remove it.
2562   auto *StrTab = Obj.SymbolTable->getStrTab() == Obj.SectionNames
2563                      ? nullptr
2564                      : Obj.SymbolTable->getStrTab();
2565   return Obj.removeSections(false, [&](const SectionBase &Sec) {
2566     return &Sec == Obj.SymbolTable || &Sec == StrTab;
2567   });
2568 }
2569 
finalize()2570 template <class ELFT> Error ELFWriter<ELFT>::finalize() {
2571   // It could happen that SectionNames has been removed and yet the user wants
2572   // a section header table output. We need to throw an error if a user tries
2573   // to do that.
2574   if (Obj.SectionNames == nullptr && WriteSectionHeaders)
2575     return createStringError(llvm::errc::invalid_argument,
2576                              "cannot write section header table because "
2577                              "section header string table was removed");
2578 
2579   if (Error E = removeUnneededSections(Obj))
2580     return E;
2581 
2582   // If the .symtab indices have not been changed, restore the sh_link to
2583   // .symtab for sections that were linked to .symtab.
2584   if (Obj.SymbolTable && !Obj.SymbolTable->indicesChanged())
2585     for (SectionBase &Sec : Obj.sections())
2586       Sec.restoreSymTabLink(*Obj.SymbolTable);
2587 
2588   // We need to assign indexes before we perform layout because we need to know
2589   // if we need large indexes or not. We can assign indexes first and check as
2590   // we go to see if we will actully need large indexes.
2591   bool NeedsLargeIndexes = false;
2592   if (Obj.sections().size() >= SHN_LORESERVE) {
2593     SectionTableRef Sections = Obj.sections();
2594     // Sections doesn't include the null section header, so account for this
2595     // when skipping the first N sections.
2596     NeedsLargeIndexes =
2597         any_of(drop_begin(Sections, SHN_LORESERVE - 1),
2598                [](const SectionBase &Sec) { return Sec.HasSymbol; });
2599     // TODO: handle case where only one section needs the large index table but
2600     // only needs it because the large index table hasn't been removed yet.
2601   }
2602 
2603   if (NeedsLargeIndexes) {
2604     // This means we definitely need to have a section index table but if we
2605     // already have one then we should use it instead of making a new one.
2606     if (Obj.SymbolTable != nullptr && Obj.SectionIndexTable == nullptr) {
2607       // Addition of a section to the end does not invalidate the indexes of
2608       // other sections and assigns the correct index to the new section.
2609       auto &Shndx = Obj.addSection<SectionIndexSection>();
2610       Obj.SymbolTable->setShndxTable(&Shndx);
2611       Shndx.setSymTab(Obj.SymbolTable);
2612     }
2613   } else {
2614     // Since we don't need SectionIndexTable we should remove it and all
2615     // references to it.
2616     if (Obj.SectionIndexTable != nullptr) {
2617       // We do not support sections referring to the section index table.
2618       if (Error E = Obj.removeSections(false /*AllowBrokenLinks*/,
2619                                        [this](const SectionBase &Sec) {
2620                                          return &Sec == Obj.SectionIndexTable;
2621                                        }))
2622         return E;
2623     }
2624   }
2625 
2626   // Make sure we add the names of all the sections. Importantly this must be
2627   // done after we decide to add or remove SectionIndexes.
2628   if (Obj.SectionNames != nullptr)
2629     for (const SectionBase &Sec : Obj.sections())
2630       Obj.SectionNames->addString(Sec.Name);
2631 
2632   initEhdrSegment();
2633 
2634   // Before we can prepare for layout the indexes need to be finalized.
2635   // Also, the output arch may not be the same as the input arch, so fix up
2636   // size-related fields before doing layout calculations.
2637   uint64_t Index = 0;
2638   auto SecSizer = std::make_unique<ELFSectionSizer<ELFT>>();
2639   for (SectionBase &Sec : Obj.sections()) {
2640     Sec.Index = Index++;
2641     if (Error Err = Sec.accept(*SecSizer))
2642       return Err;
2643   }
2644 
2645   // The symbol table does not update all other sections on update. For
2646   // instance, symbol names are not added as new symbols are added. This means
2647   // that some sections, like .strtab, don't yet have their final size.
2648   if (Obj.SymbolTable != nullptr)
2649     Obj.SymbolTable->prepareForLayout();
2650 
2651   // Now that all strings are added we want to finalize string table builders,
2652   // because that affects section sizes which in turn affects section offsets.
2653   for (SectionBase &Sec : Obj.sections())
2654     if (auto StrTab = dyn_cast<StringTableSection>(&Sec))
2655       StrTab->prepareForLayout();
2656 
2657   assignOffsets();
2658 
2659   // layoutSections could have modified section indexes, so we need
2660   // to fill the index table after assignOffsets.
2661   if (Obj.SymbolTable != nullptr)
2662     Obj.SymbolTable->fillShndxTable();
2663 
2664   // Finally now that all offsets and indexes have been set we can finalize any
2665   // remaining issues.
2666   uint64_t Offset = Obj.SHOff + sizeof(Elf_Shdr);
2667   for (SectionBase &Sec : Obj.sections()) {
2668     Sec.HeaderOffset = Offset;
2669     Offset += sizeof(Elf_Shdr);
2670     if (WriteSectionHeaders)
2671       Sec.NameIndex = Obj.SectionNames->findIndex(Sec.Name);
2672     Sec.finalize();
2673   }
2674 
2675   size_t TotalSize = totalSize();
2676   Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2677   if (!Buf)
2678     return createStringError(errc::not_enough_memory,
2679                              "failed to allocate memory buffer of " +
2680                                  Twine::utohexstr(TotalSize) + " bytes");
2681 
2682   SecWriter = std::make_unique<ELFSectionWriter<ELFT>>(*Buf);
2683   return Error::success();
2684 }
2685 
write()2686 Error BinaryWriter::write() {
2687   SmallVector<const SectionBase *, 30> SectionsToWrite;
2688   for (const SectionBase &Sec : Obj.allocSections()) {
2689     if (Sec.Type != SHT_NOBITS && Sec.Size > 0)
2690       SectionsToWrite.push_back(&Sec);
2691   }
2692 
2693   if (SectionsToWrite.empty())
2694     return Error::success();
2695 
2696   llvm::stable_sort(SectionsToWrite,
2697                     [](const SectionBase *LHS, const SectionBase *RHS) {
2698                       return LHS->Offset < RHS->Offset;
2699                     });
2700 
2701   assert(SectionsToWrite.front()->Offset == 0);
2702 
2703   for (size_t i = 0; i != SectionsToWrite.size(); ++i) {
2704     const SectionBase &Sec = *SectionsToWrite[i];
2705     if (Error Err = Sec.accept(*SecWriter))
2706       return Err;
2707     if (GapFill == 0)
2708       continue;
2709     uint64_t PadOffset = (i < SectionsToWrite.size() - 1)
2710                              ? SectionsToWrite[i + 1]->Offset
2711                              : Buf->getBufferSize();
2712     assert(PadOffset <= Buf->getBufferSize());
2713     assert(Sec.Offset + Sec.Size <= PadOffset);
2714     std::fill(Buf->getBufferStart() + Sec.Offset + Sec.Size,
2715               Buf->getBufferStart() + PadOffset, GapFill);
2716   }
2717 
2718   // TODO: Implement direct writing to the output stream (without intermediate
2719   // memory buffer Buf).
2720   Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2721   return Error::success();
2722 }
2723 
finalize()2724 Error BinaryWriter::finalize() {
2725   // Compute the section LMA based on its sh_offset and the containing segment's
2726   // p_offset and p_paddr. Also compute the minimum LMA of all non-empty
2727   // sections as MinAddr. In the output, the contents between address 0 and
2728   // MinAddr will be skipped.
2729   uint64_t MinAddr = UINT64_MAX;
2730   for (SectionBase &Sec : Obj.allocSections()) {
2731     if (Sec.ParentSegment != nullptr)
2732       Sec.Addr =
2733           Sec.Offset - Sec.ParentSegment->Offset + Sec.ParentSegment->PAddr;
2734     if (Sec.Type != SHT_NOBITS && Sec.Size > 0)
2735       MinAddr = std::min(MinAddr, Sec.Addr);
2736   }
2737 
2738   // Now that every section has been laid out we just need to compute the total
2739   // file size. This might not be the same as the offset returned by
2740   // layoutSections, because we want to truncate the last segment to the end of
2741   // its last non-empty section, to match GNU objcopy's behaviour.
2742   TotalSize = PadTo > MinAddr ? PadTo - MinAddr : 0;
2743   for (SectionBase &Sec : Obj.allocSections())
2744     if (Sec.Type != SHT_NOBITS && Sec.Size > 0) {
2745       Sec.Offset = Sec.Addr - MinAddr;
2746       TotalSize = std::max(TotalSize, Sec.Offset + Sec.Size);
2747     }
2748 
2749   Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2750   if (!Buf)
2751     return createStringError(errc::not_enough_memory,
2752                              "failed to allocate memory buffer of " +
2753                                  Twine::utohexstr(TotalSize) + " bytes");
2754   SecWriter = std::make_unique<BinarySectionWriter>(*Buf);
2755   return Error::success();
2756 }
2757 
checkSection(const SectionBase & S) const2758 Error ASCIIHexWriter::checkSection(const SectionBase &S) const {
2759   if (addressOverflows32bit(S.Addr) ||
2760       addressOverflows32bit(S.Addr + S.Size - 1))
2761     return createStringError(
2762         errc::invalid_argument,
2763         "section '%s' address range [0x%llx, 0x%llx] is not 32 bit",
2764         S.Name.c_str(), S.Addr, S.Addr + S.Size - 1);
2765   return Error::success();
2766 }
2767 
finalize()2768 Error ASCIIHexWriter::finalize() {
2769   // We can't write 64-bit addresses.
2770   if (addressOverflows32bit(Obj.Entry))
2771     return createStringError(errc::invalid_argument,
2772                              "entry point address 0x%llx overflows 32 bits",
2773                              Obj.Entry);
2774 
2775   for (const SectionBase &S : Obj.sections()) {
2776     if ((S.Flags & ELF::SHF_ALLOC) && S.Type != ELF::SHT_NOBITS && S.Size > 0) {
2777       if (Error E = checkSection(S))
2778         return E;
2779       Sections.push_back(&S);
2780     }
2781   }
2782 
2783   llvm::sort(Sections, [](const SectionBase *A, const SectionBase *B) {
2784     return sectionPhysicalAddr(A) < sectionPhysicalAddr(B);
2785   });
2786 
2787   std::unique_ptr<WritableMemoryBuffer> EmptyBuffer =
2788       WritableMemoryBuffer::getNewMemBuffer(0);
2789   if (!EmptyBuffer)
2790     return createStringError(errc::not_enough_memory,
2791                              "failed to allocate memory buffer of 0 bytes");
2792 
2793   Expected<size_t> ExpTotalSize = getTotalSize(*EmptyBuffer);
2794   if (!ExpTotalSize)
2795     return ExpTotalSize.takeError();
2796   TotalSize = *ExpTotalSize;
2797 
2798   Buf = WritableMemoryBuffer::getNewMemBuffer(TotalSize);
2799   if (!Buf)
2800     return createStringError(errc::not_enough_memory,
2801                              "failed to allocate memory buffer of 0x" +
2802                                  Twine::utohexstr(TotalSize) + " bytes");
2803   return Error::success();
2804 }
2805 
writeEntryPointRecord(uint8_t * Buf)2806 uint64_t IHexWriter::writeEntryPointRecord(uint8_t *Buf) {
2807   IHexLineData HexData;
2808   uint8_t Data[4] = {};
2809   // We don't write entry point record if entry is zero.
2810   if (Obj.Entry == 0)
2811     return 0;
2812 
2813   if (Obj.Entry <= 0xFFFFFU) {
2814     Data[0] = ((Obj.Entry & 0xF0000U) >> 12) & 0xFF;
2815     support::endian::write(&Data[2], static_cast<uint16_t>(Obj.Entry),
2816                            llvm::endianness::big);
2817     HexData = IHexRecord::getLine(IHexRecord::StartAddr80x86, 0, Data);
2818   } else {
2819     support::endian::write(Data, static_cast<uint32_t>(Obj.Entry),
2820                            llvm::endianness::big);
2821     HexData = IHexRecord::getLine(IHexRecord::StartAddr, 0, Data);
2822   }
2823   memcpy(Buf, HexData.data(), HexData.size());
2824   return HexData.size();
2825 }
2826 
writeEndOfFileRecord(uint8_t * Buf)2827 uint64_t IHexWriter::writeEndOfFileRecord(uint8_t *Buf) {
2828   IHexLineData HexData = IHexRecord::getLine(IHexRecord::EndOfFile, 0, {});
2829   memcpy(Buf, HexData.data(), HexData.size());
2830   return HexData.size();
2831 }
2832 
2833 Expected<size_t>
getTotalSize(WritableMemoryBuffer & EmptyBuffer) const2834 IHexWriter::getTotalSize(WritableMemoryBuffer &EmptyBuffer) const {
2835   IHexSectionWriterBase LengthCalc(EmptyBuffer);
2836   for (const SectionBase *Sec : Sections)
2837     if (Error Err = Sec->accept(LengthCalc))
2838       return std::move(Err);
2839 
2840   // We need space to write section records + StartAddress record
2841   // (if start adress is not zero) + EndOfFile record.
2842   return LengthCalc.getBufferOffset() +
2843          (Obj.Entry ? IHexRecord::getLineLength(4) : 0) +
2844          IHexRecord::getLineLength(0);
2845 }
2846 
write()2847 Error IHexWriter::write() {
2848   IHexSectionWriter Writer(*Buf);
2849   // Write sections.
2850   for (const SectionBase *Sec : Sections)
2851     if (Error Err = Sec->accept(Writer))
2852       return Err;
2853 
2854   uint64_t Offset = Writer.getBufferOffset();
2855   // Write entry point address.
2856   Offset += writeEntryPointRecord(
2857       reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset);
2858   // Write EOF.
2859   Offset += writeEndOfFileRecord(
2860       reinterpret_cast<uint8_t *>(Buf->getBufferStart()) + Offset);
2861   assert(Offset == TotalSize);
2862 
2863   // TODO: Implement direct writing to the output stream (without intermediate
2864   // memory buffer Buf).
2865   Out.write(Buf->getBufferStart(), Buf->getBufferSize());
2866   return Error::success();
2867 }
2868 
visit(const StringTableSection & Sec)2869 Error SRECSectionWriterBase::visit(const StringTableSection &Sec) {
2870   // Check that the sizer has already done its work.
2871   assert(Sec.Size == Sec.StrTabBuilder.getSize() &&
2872          "Expected section size to have been finalized");
2873   // We don't need to write anything here because the real writer has already
2874   // done it.
2875   return Error::success();
2876 }
2877 
visit(const Section & Sec)2878 Error SRECSectionWriterBase::visit(const Section &Sec) {
2879   writeSection(Sec, Sec.Contents);
2880   return Error::success();
2881 }
2882 
visit(const OwnedDataSection & Sec)2883 Error SRECSectionWriterBase::visit(const OwnedDataSection &Sec) {
2884   writeSection(Sec, Sec.Data);
2885   return Error::success();
2886 }
2887 
visit(const DynamicRelocationSection & Sec)2888 Error SRECSectionWriterBase::visit(const DynamicRelocationSection &Sec) {
2889   writeSection(Sec, Sec.Contents);
2890   return Error::success();
2891 }
2892 
writeRecord(SRecord & Record,uint64_t Off)2893 void SRECSectionWriter::writeRecord(SRecord &Record, uint64_t Off) {
2894   SRecLineData Data = Record.toString();
2895   memcpy(Out.getBufferStart() + Off, Data.data(), Data.size());
2896 }
2897 
writeRecords(uint32_t Entry)2898 void SRECSectionWriterBase::writeRecords(uint32_t Entry) {
2899   // The ELF header could contain an entry point outside of the sections we have
2900   // seen that does not fit the current record Type.
2901   Type = std::max(Type, SRecord::getType(Entry));
2902   uint64_t Off = HeaderSize;
2903   for (SRecord &Record : Records) {
2904     Record.Type = Type;
2905     writeRecord(Record, Off);
2906     Off += Record.getSize();
2907   }
2908   Offset = Off;
2909 }
2910 
writeSection(const SectionBase & S,ArrayRef<uint8_t> Data)2911 void SRECSectionWriterBase::writeSection(const SectionBase &S,
2912                                          ArrayRef<uint8_t> Data) {
2913   const uint32_t ChunkSize = 16;
2914   uint32_t Address = sectionPhysicalAddr(&S);
2915   uint32_t EndAddr = Address + S.Size - 1;
2916   Type = std::max(SRecord::getType(EndAddr), Type);
2917   while (!Data.empty()) {
2918     uint64_t DataSize = std::min<uint64_t>(Data.size(), ChunkSize);
2919     SRecord Record{Type, Address, Data.take_front(DataSize)};
2920     Records.push_back(Record);
2921     Data = Data.drop_front(DataSize);
2922     Address += DataSize;
2923   }
2924 }
2925 
visit(const StringTableSection & Sec)2926 Error SRECSectionWriter::visit(const StringTableSection &Sec) {
2927   assert(Sec.Size == Sec.StrTabBuilder.getSize() &&
2928          "Section size does not match the section's string table builder size");
2929   std::vector<uint8_t> Data(Sec.Size);
2930   Sec.StrTabBuilder.write(Data.data());
2931   writeSection(Sec, Data);
2932   return Error::success();
2933 }
2934 
toString() const2935 SRecLineData SRecord::toString() const {
2936   SRecLineData Line(getSize());
2937   auto *Iter = Line.begin();
2938   *Iter++ = 'S';
2939   *Iter++ = '0' + Type;
2940   // Write 1 byte (2 hex characters) record count.
2941   Iter = toHexStr(getCount(), Iter, 2);
2942   // Write the address field with length depending on record type.
2943   Iter = toHexStr(Address, Iter, getAddressSize());
2944   // Write data byte by byte.
2945   for (uint8_t X : Data)
2946     Iter = toHexStr(X, Iter, 2);
2947   // Write the 1 byte checksum.
2948   Iter = toHexStr(getChecksum(), Iter, 2);
2949   *Iter++ = '\r';
2950   *Iter++ = '\n';
2951   assert(Iter == Line.end());
2952   return Line;
2953 }
2954 
getChecksum() const2955 uint8_t SRecord::getChecksum() const {
2956   uint32_t Sum = getCount();
2957   Sum += (Address >> 24) & 0xFF;
2958   Sum += (Address >> 16) & 0xFF;
2959   Sum += (Address >> 8) & 0xFF;
2960   Sum += Address & 0xFF;
2961   for (uint8_t Byte : Data)
2962     Sum += Byte;
2963   return 0xFF - (Sum & 0xFF);
2964 }
2965 
getSize() const2966 size_t SRecord::getSize() const {
2967   // Type, Count, Checksum, and CRLF are two characters each.
2968   return 2 + 2 + getAddressSize() + Data.size() * 2 + 2 + 2;
2969 }
2970 
getAddressSize() const2971 uint8_t SRecord::getAddressSize() const {
2972   switch (Type) {
2973   case Type::S2:
2974     return 6;
2975   case Type::S3:
2976     return 8;
2977   case Type::S7:
2978     return 8;
2979   case Type::S8:
2980     return 6;
2981   default:
2982     return 4;
2983   }
2984 }
2985 
getCount() const2986 uint8_t SRecord::getCount() const {
2987   uint8_t DataSize = Data.size();
2988   uint8_t ChecksumSize = 1;
2989   return getAddressSize() / 2 + DataSize + ChecksumSize;
2990 }
2991 
getType(uint32_t Address)2992 uint8_t SRecord::getType(uint32_t Address) {
2993   if (isUInt<16>(Address))
2994     return SRecord::S1;
2995   if (isUInt<24>(Address))
2996     return SRecord::S2;
2997   return SRecord::S3;
2998 }
2999 
getHeader(StringRef FileName)3000 SRecord SRecord::getHeader(StringRef FileName) {
3001   // Header is a record with Type S0, Address 0, and Data that is a
3002   // vendor-specific text comment. For the comment we will use the output file
3003   // name truncated to 40 characters to match the behavior of GNU objcopy.
3004   StringRef HeaderContents = FileName.slice(0, 40);
3005   ArrayRef<uint8_t> Data(
3006       reinterpret_cast<const uint8_t *>(HeaderContents.data()),
3007       HeaderContents.size());
3008   return {SRecord::S0, 0, Data};
3009 }
3010 
writeHeader(uint8_t * Buf)3011 size_t SRECWriter::writeHeader(uint8_t *Buf) {
3012   SRecLineData Record = SRecord::getHeader(OutputFileName).toString();
3013   memcpy(Buf, Record.data(), Record.size());
3014   return Record.size();
3015 }
3016 
writeTerminator(uint8_t * Buf,uint8_t Type)3017 size_t SRECWriter::writeTerminator(uint8_t *Buf, uint8_t Type) {
3018   assert(Type >= SRecord::S7 && Type <= SRecord::S9 &&
3019          "Invalid record type for terminator");
3020   uint32_t Entry = Obj.Entry;
3021   SRecLineData Data = SRecord{Type, Entry, {}}.toString();
3022   memcpy(Buf, Data.data(), Data.size());
3023   return Data.size();
3024 }
3025 
3026 Expected<size_t>
getTotalSize(WritableMemoryBuffer & EmptyBuffer) const3027 SRECWriter::getTotalSize(WritableMemoryBuffer &EmptyBuffer) const {
3028   SRECSizeCalculator SizeCalc(EmptyBuffer, 0);
3029   for (const SectionBase *Sec : Sections)
3030     if (Error Err = Sec->accept(SizeCalc))
3031       return std::move(Err);
3032 
3033   SizeCalc.writeRecords(Obj.Entry);
3034   // We need to add the size of the Header and Terminator records.
3035   SRecord Header = SRecord::getHeader(OutputFileName);
3036   uint8_t TerminatorType = 10 - SizeCalc.getType();
3037   SRecord Terminator = {TerminatorType, static_cast<uint32_t>(Obj.Entry), {}};
3038   return Header.getSize() + SizeCalc.getBufferOffset() + Terminator.getSize();
3039 }
3040 
write()3041 Error SRECWriter::write() {
3042   uint32_t HeaderSize =
3043       writeHeader(reinterpret_cast<uint8_t *>(Buf->getBufferStart()));
3044   SRECSectionWriter Writer(*Buf, HeaderSize);
3045   for (const SectionBase *S : Sections) {
3046     if (Error E = S->accept(Writer))
3047       return E;
3048   }
3049   Writer.writeRecords(Obj.Entry);
3050   uint64_t Offset = Writer.getBufferOffset();
3051 
3052   // An S1 record terminates with an S9 record, S2 with S8, and S3 with S7.
3053   uint8_t TerminatorType = 10 - Writer.getType();
3054   Offset += writeTerminator(
3055       reinterpret_cast<uint8_t *>(Buf->getBufferStart() + Offset),
3056       TerminatorType);
3057   assert(Offset == TotalSize);
3058   Out.write(Buf->getBufferStart(), Buf->getBufferSize());
3059   return Error::success();
3060 }
3061 
3062 namespace llvm {
3063 namespace objcopy {
3064 namespace elf {
3065 
3066 template class ELFBuilder<ELF64LE>;
3067 template class ELFBuilder<ELF64BE>;
3068 template class ELFBuilder<ELF32LE>;
3069 template class ELFBuilder<ELF32BE>;
3070 
3071 template class ELFWriter<ELF64LE>;
3072 template class ELFWriter<ELF64BE>;
3073 template class ELFWriter<ELF32LE>;
3074 template class ELFWriter<ELF32BE>;
3075 
3076 } // end namespace elf
3077 } // end namespace objcopy
3078 } // end namespace llvm
3079