1 //===- lib/MC/MachObjectWriter.cpp - Mach-O File Writer -------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 
10 #include "llvm/MC/MCMachObjectWriter.h"
11 #include "llvm/ADT/StringMap.h"
12 #include "llvm/ADT/Twine.h"
13 #include "llvm/MC/MCAsmBackend.h"
14 #include "llvm/MC/MCAsmLayout.h"
15 #include "llvm/MC/MCAssembler.h"
16 #include "llvm/MC/MCExpr.h"
17 #include "llvm/MC/MCFixupKindInfo.h"
18 #include "llvm/MC/MCObjectWriter.h"
19 #include "llvm/MC/MCSectionMachO.h"
20 #include "llvm/MC/MCSymbolMachO.h"
21 #include "llvm/MC/MCValue.h"
22 #include "llvm/Support/Debug.h"
23 #include "llvm/Support/ErrorHandling.h"
24 #include "llvm/Support/MachO.h"
25 #include "llvm/Support/raw_ostream.h"
26 #include <vector>
27 using namespace llvm;
28 
29 #define DEBUG_TYPE "mc"
30 
reset()31 void MachObjectWriter::reset() {
32   Relocations.clear();
33   IndirectSymBase.clear();
34   StringTable.clear();
35   LocalSymbolData.clear();
36   ExternalSymbolData.clear();
37   UndefinedSymbolData.clear();
38   MCObjectWriter::reset();
39 }
40 
doesSymbolRequireExternRelocation(const MCSymbol & S)41 bool MachObjectWriter::doesSymbolRequireExternRelocation(const MCSymbol &S) {
42   // Undefined symbols are always extern.
43   if (S.isUndefined())
44     return true;
45 
46   // References to weak definitions require external relocation entries; the
47   // definition may not always be the one in the same object file.
48   if (cast<MCSymbolMachO>(S).isWeakDefinition())
49     return true;
50 
51   // Otherwise, we can use an internal relocation.
52   return false;
53 }
54 
55 bool MachObjectWriter::
operator <(const MachSymbolData & RHS) const56 MachSymbolData::operator<(const MachSymbolData &RHS) const {
57   return Symbol->getName() < RHS.Symbol->getName();
58 }
59 
isFixupKindPCRel(const MCAssembler & Asm,unsigned Kind)60 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
61   const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
62     (MCFixupKind) Kind);
63 
64   return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
65 }
66 
getFragmentAddress(const MCFragment * Fragment,const MCAsmLayout & Layout) const67 uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment,
68                                               const MCAsmLayout &Layout) const {
69   return getSectionAddress(Fragment->getParent()) +
70          Layout.getFragmentOffset(Fragment);
71 }
72 
getSymbolAddress(const MCSymbol & S,const MCAsmLayout & Layout) const73 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbol &S,
74                                             const MCAsmLayout &Layout) const {
75   // If this is a variable, then recursively evaluate now.
76   if (S.isVariable()) {
77     if (const MCConstantExpr *C =
78           dyn_cast<const MCConstantExpr>(S.getVariableValue()))
79       return C->getValue();
80 
81     MCValue Target;
82     if (!S.getVariableValue()->evaluateAsRelocatable(Target, &Layout, nullptr))
83       report_fatal_error("unable to evaluate offset for variable '" +
84                          S.getName() + "'");
85 
86     // Verify that any used symbols are defined.
87     if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
88       report_fatal_error("unable to evaluate offset to undefined symbol '" +
89                          Target.getSymA()->getSymbol().getName() + "'");
90     if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
91       report_fatal_error("unable to evaluate offset to undefined symbol '" +
92                          Target.getSymB()->getSymbol().getName() + "'");
93 
94     uint64_t Address = Target.getConstant();
95     if (Target.getSymA())
96       Address += getSymbolAddress(Target.getSymA()->getSymbol(), Layout);
97     if (Target.getSymB())
98       Address += getSymbolAddress(Target.getSymB()->getSymbol(), Layout);
99     return Address;
100   }
101 
102   return getSectionAddress(S.getFragment()->getParent()) +
103          Layout.getSymbolOffset(S);
104 }
105 
getPaddingSize(const MCSection * Sec,const MCAsmLayout & Layout) const106 uint64_t MachObjectWriter::getPaddingSize(const MCSection *Sec,
107                                           const MCAsmLayout &Layout) const {
108   uint64_t EndAddr = getSectionAddress(Sec) + Layout.getSectionAddressSize(Sec);
109   unsigned Next = Sec->getLayoutOrder() + 1;
110   if (Next >= Layout.getSectionOrder().size())
111     return 0;
112 
113   const MCSection &NextSec = *Layout.getSectionOrder()[Next];
114   if (NextSec.isVirtualSection())
115     return 0;
116   return OffsetToAlignment(EndAddr, NextSec.getAlignment());
117 }
118 
writeHeader(MachO::HeaderFileType Type,unsigned NumLoadCommands,unsigned LoadCommandsSize,bool SubsectionsViaSymbols)119 void MachObjectWriter::writeHeader(MachO::HeaderFileType Type,
120                                    unsigned NumLoadCommands,
121                                    unsigned LoadCommandsSize,
122                                    bool SubsectionsViaSymbols) {
123   uint32_t Flags = 0;
124 
125   if (SubsectionsViaSymbols)
126     Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
127 
128   // struct mach_header (28 bytes) or
129   // struct mach_header_64 (32 bytes)
130 
131   uint64_t Start = getStream().tell();
132   (void) Start;
133 
134   write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
135 
136   write32(TargetObjectWriter->getCPUType());
137   write32(TargetObjectWriter->getCPUSubtype());
138 
139   write32(Type);
140   write32(NumLoadCommands);
141   write32(LoadCommandsSize);
142   write32(Flags);
143   if (is64Bit())
144     write32(0); // reserved
145 
146   assert(
147       getStream().tell() - Start ==
148       (is64Bit() ? sizeof(MachO::mach_header_64) : sizeof(MachO::mach_header)));
149 }
150 
151 /// writeSegmentLoadCommand - Write a segment load command.
152 ///
153 /// \param NumSections The number of sections in this segment.
154 /// \param SectionDataSize The total size of the sections.
writeSegmentLoadCommand(StringRef Name,unsigned NumSections,uint64_t VMAddr,uint64_t VMSize,uint64_t SectionDataStartOffset,uint64_t SectionDataSize,uint32_t MaxProt,uint32_t InitProt)155 void MachObjectWriter::writeSegmentLoadCommand(
156     StringRef Name, unsigned NumSections, uint64_t VMAddr, uint64_t VMSize,
157     uint64_t SectionDataStartOffset, uint64_t SectionDataSize, uint32_t MaxProt,
158     uint32_t InitProt) {
159   // struct segment_command (56 bytes) or
160   // struct segment_command_64 (72 bytes)
161 
162   uint64_t Start = getStream().tell();
163   (void) Start;
164 
165   unsigned SegmentLoadCommandSize =
166     is64Bit() ? sizeof(MachO::segment_command_64):
167     sizeof(MachO::segment_command);
168   write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
169   write32(SegmentLoadCommandSize +
170           NumSections * (is64Bit() ? sizeof(MachO::section_64) :
171                          sizeof(MachO::section)));
172 
173   assert(Name.size() <= 16);
174   writeBytes(Name, 16);
175   if (is64Bit()) {
176     write64(VMAddr);                 // vmaddr
177     write64(VMSize); // vmsize
178     write64(SectionDataStartOffset); // file offset
179     write64(SectionDataSize); // file size
180   } else {
181     write32(VMAddr);                 // vmaddr
182     write32(VMSize); // vmsize
183     write32(SectionDataStartOffset); // file offset
184     write32(SectionDataSize); // file size
185   }
186   // maxprot
187   write32(MaxProt);
188   // initprot
189   write32(InitProt);
190   write32(NumSections);
191   write32(0); // flags
192 
193   assert(getStream().tell() - Start == SegmentLoadCommandSize);
194 }
195 
writeSection(const MCAsmLayout & Layout,const MCSection & Sec,uint64_t VMAddr,uint64_t FileOffset,unsigned Flags,uint64_t RelocationsStart,unsigned NumRelocations)196 void MachObjectWriter::writeSection(const MCAsmLayout &Layout,
197                                     const MCSection &Sec, uint64_t VMAddr,
198                                     uint64_t FileOffset, unsigned Flags,
199                                     uint64_t RelocationsStart,
200                                     unsigned NumRelocations) {
201   uint64_t SectionSize = Layout.getSectionAddressSize(&Sec);
202   const MCSectionMachO &Section = cast<MCSectionMachO>(Sec);
203 
204   // The offset is unused for virtual sections.
205   if (Section.isVirtualSection()) {
206     assert(Layout.getSectionFileSize(&Sec) == 0 && "Invalid file size!");
207     FileOffset = 0;
208   }
209 
210   // struct section (68 bytes) or
211   // struct section_64 (80 bytes)
212 
213   uint64_t Start = getStream().tell();
214   (void) Start;
215 
216   writeBytes(Section.getSectionName(), 16);
217   writeBytes(Section.getSegmentName(), 16);
218   if (is64Bit()) {
219     write64(VMAddr);      // address
220     write64(SectionSize); // size
221   } else {
222     write32(VMAddr);      // address
223     write32(SectionSize); // size
224   }
225   write32(FileOffset);
226 
227   assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!");
228   write32(Log2_32(Section.getAlignment()));
229   write32(NumRelocations ? RelocationsStart : 0);
230   write32(NumRelocations);
231   write32(Flags);
232   write32(IndirectSymBase.lookup(&Sec)); // reserved1
233   write32(Section.getStubSize()); // reserved2
234   if (is64Bit())
235     write32(0); // reserved3
236 
237   assert(getStream().tell() - Start ==
238          (is64Bit() ? sizeof(MachO::section_64) : sizeof(MachO::section)));
239 }
240 
writeSymtabLoadCommand(uint32_t SymbolOffset,uint32_t NumSymbols,uint32_t StringTableOffset,uint32_t StringTableSize)241 void MachObjectWriter::writeSymtabLoadCommand(uint32_t SymbolOffset,
242                                               uint32_t NumSymbols,
243                                               uint32_t StringTableOffset,
244                                               uint32_t StringTableSize) {
245   // struct symtab_command (24 bytes)
246 
247   uint64_t Start = getStream().tell();
248   (void) Start;
249 
250   write32(MachO::LC_SYMTAB);
251   write32(sizeof(MachO::symtab_command));
252   write32(SymbolOffset);
253   write32(NumSymbols);
254   write32(StringTableOffset);
255   write32(StringTableSize);
256 
257   assert(getStream().tell() - Start == sizeof(MachO::symtab_command));
258 }
259 
writeDysymtabLoadCommand(uint32_t FirstLocalSymbol,uint32_t NumLocalSymbols,uint32_t FirstExternalSymbol,uint32_t NumExternalSymbols,uint32_t FirstUndefinedSymbol,uint32_t NumUndefinedSymbols,uint32_t IndirectSymbolOffset,uint32_t NumIndirectSymbols)260 void MachObjectWriter::writeDysymtabLoadCommand(uint32_t FirstLocalSymbol,
261                                                 uint32_t NumLocalSymbols,
262                                                 uint32_t FirstExternalSymbol,
263                                                 uint32_t NumExternalSymbols,
264                                                 uint32_t FirstUndefinedSymbol,
265                                                 uint32_t NumUndefinedSymbols,
266                                                 uint32_t IndirectSymbolOffset,
267                                                 uint32_t NumIndirectSymbols) {
268   // struct dysymtab_command (80 bytes)
269 
270   uint64_t Start = getStream().tell();
271   (void) Start;
272 
273   write32(MachO::LC_DYSYMTAB);
274   write32(sizeof(MachO::dysymtab_command));
275   write32(FirstLocalSymbol);
276   write32(NumLocalSymbols);
277   write32(FirstExternalSymbol);
278   write32(NumExternalSymbols);
279   write32(FirstUndefinedSymbol);
280   write32(NumUndefinedSymbols);
281   write32(0); // tocoff
282   write32(0); // ntoc
283   write32(0); // modtaboff
284   write32(0); // nmodtab
285   write32(0); // extrefsymoff
286   write32(0); // nextrefsyms
287   write32(IndirectSymbolOffset);
288   write32(NumIndirectSymbols);
289   write32(0); // extreloff
290   write32(0); // nextrel
291   write32(0); // locreloff
292   write32(0); // nlocrel
293 
294   assert(getStream().tell() - Start == sizeof(MachO::dysymtab_command));
295 }
296 
297 MachObjectWriter::MachSymbolData *
findSymbolData(const MCSymbol & Sym)298 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
299   for (auto *SymbolData :
300        {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
301     for (MachSymbolData &Entry : *SymbolData)
302       if (Entry.Symbol == &Sym)
303         return &Entry;
304 
305   return nullptr;
306 }
307 
findAliasedSymbol(const MCSymbol & Sym) const308 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
309   const MCSymbol *S = &Sym;
310   while (S->isVariable()) {
311     const MCExpr *Value = S->getVariableValue();
312     const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
313     if (!Ref)
314       return *S;
315     S = &Ref->getSymbol();
316   }
317   return *S;
318 }
319 
writeNlist(MachSymbolData & MSD,const MCAsmLayout & Layout)320 void MachObjectWriter::writeNlist(MachSymbolData &MSD,
321                                   const MCAsmLayout &Layout) {
322   const MCSymbol *Symbol = MSD.Symbol;
323   const MCSymbol &Data = *Symbol;
324   const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
325   uint8_t SectionIndex = MSD.SectionIndex;
326   uint8_t Type = 0;
327   uint64_t Address = 0;
328   bool IsAlias = Symbol != AliasedSymbol;
329 
330   const MCSymbol &OrigSymbol = *Symbol;
331   MachSymbolData *AliaseeInfo;
332   if (IsAlias) {
333     AliaseeInfo = findSymbolData(*AliasedSymbol);
334     if (AliaseeInfo)
335       SectionIndex = AliaseeInfo->SectionIndex;
336     Symbol = AliasedSymbol;
337     // FIXME: Should this update Data as well?  Do we need OrigSymbol at all?
338   }
339 
340   // Set the N_TYPE bits. See <mach-o/nlist.h>.
341   //
342   // FIXME: Are the prebound or indirect fields possible here?
343   if (IsAlias && Symbol->isUndefined())
344     Type = MachO::N_INDR;
345   else if (Symbol->isUndefined())
346     Type = MachO::N_UNDF;
347   else if (Symbol->isAbsolute())
348     Type = MachO::N_ABS;
349   else
350     Type = MachO::N_SECT;
351 
352   // FIXME: Set STAB bits.
353 
354   if (Data.isPrivateExtern())
355     Type |= MachO::N_PEXT;
356 
357   // Set external bit.
358   if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
359     Type |= MachO::N_EXT;
360 
361   // Compute the symbol address.
362   if (IsAlias && Symbol->isUndefined())
363     Address = AliaseeInfo->StringIndex;
364   else if (Symbol->isDefined())
365     Address = getSymbolAddress(OrigSymbol, Layout);
366   else if (Symbol->isCommon()) {
367     // Common symbols are encoded with the size in the address
368     // field, and their alignment in the flags.
369     Address = Symbol->getCommonSize();
370   }
371 
372   // struct nlist (12 bytes)
373 
374   write32(MSD.StringIndex);
375   write8(Type);
376   write8(SectionIndex);
377 
378   // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
379   // value.
380   write16(cast<MCSymbolMachO>(Symbol)->getEncodedFlags());
381   if (is64Bit())
382     write64(Address);
383   else
384     write32(Address);
385 }
386 
writeLinkeditLoadCommand(uint32_t Type,uint32_t DataOffset,uint32_t DataSize)387 void MachObjectWriter::writeLinkeditLoadCommand(uint32_t Type,
388                                                 uint32_t DataOffset,
389                                                 uint32_t DataSize) {
390   uint64_t Start = getStream().tell();
391   (void) Start;
392 
393   write32(Type);
394   write32(sizeof(MachO::linkedit_data_command));
395   write32(DataOffset);
396   write32(DataSize);
397 
398   assert(getStream().tell() - Start == sizeof(MachO::linkedit_data_command));
399 }
400 
ComputeLinkerOptionsLoadCommandSize(const std::vector<std::string> & Options,bool is64Bit)401 static unsigned ComputeLinkerOptionsLoadCommandSize(
402   const std::vector<std::string> &Options, bool is64Bit)
403 {
404   unsigned Size = sizeof(MachO::linker_option_command);
405   for (const std::string &Option : Options)
406     Size += Option.size() + 1;
407   return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
408 }
409 
writeLinkerOptionsLoadCommand(const std::vector<std::string> & Options)410 void MachObjectWriter::writeLinkerOptionsLoadCommand(
411   const std::vector<std::string> &Options)
412 {
413   unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
414   uint64_t Start = getStream().tell();
415   (void) Start;
416 
417   write32(MachO::LC_LINKER_OPTION);
418   write32(Size);
419   write32(Options.size());
420   uint64_t BytesWritten = sizeof(MachO::linker_option_command);
421   for (const std::string &Option : Options) {
422     // Write each string, including the null byte.
423     writeBytes(Option.c_str(), Option.size() + 1);
424     BytesWritten += Option.size() + 1;
425   }
426 
427   // Pad to a multiple of the pointer size.
428   writeBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
429 
430   assert(getStream().tell() - Start == Size);
431 }
432 
recordRelocation(MCAssembler & Asm,const MCAsmLayout & Layout,const MCFragment * Fragment,const MCFixup & Fixup,MCValue Target,bool & IsPCRel,uint64_t & FixedValue)433 void MachObjectWriter::recordRelocation(MCAssembler &Asm,
434                                         const MCAsmLayout &Layout,
435                                         const MCFragment *Fragment,
436                                         const MCFixup &Fixup, MCValue Target,
437                                         bool &IsPCRel, uint64_t &FixedValue) {
438   TargetObjectWriter->recordRelocation(this, Asm, Layout, Fragment, Fixup,
439                                        Target, FixedValue);
440 }
441 
bindIndirectSymbols(MCAssembler & Asm)442 void MachObjectWriter::bindIndirectSymbols(MCAssembler &Asm) {
443   // This is the point where 'as' creates actual symbols for indirect symbols
444   // (in the following two passes). It would be easier for us to do this sooner
445   // when we see the attribute, but that makes getting the order in the symbol
446   // table much more complicated than it is worth.
447   //
448   // FIXME: Revisit this when the dust settles.
449 
450   // Report errors for use of .indirect_symbol not in a symbol pointer section
451   // or stub section.
452   for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
453          ie = Asm.indirect_symbol_end(); it != ie; ++it) {
454     const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
455 
456     if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
457         Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
458         Section.getType() != MachO::S_SYMBOL_STUBS) {
459       MCSymbol &Symbol = *it->Symbol;
460       report_fatal_error("indirect symbol '" + Symbol.getName() +
461                          "' not in a symbol pointer or stub section");
462     }
463   }
464 
465   // Bind non-lazy symbol pointers first.
466   unsigned IndirectIndex = 0;
467   for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
468          ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
469     const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
470 
471     if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
472       continue;
473 
474     // Initialize the section indirect symbol base, if necessary.
475     IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
476 
477     Asm.registerSymbol(*it->Symbol);
478   }
479 
480   // Then lazy symbol pointers and symbol stubs.
481   IndirectIndex = 0;
482   for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
483          ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
484     const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
485 
486     if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
487         Section.getType() != MachO::S_SYMBOL_STUBS)
488       continue;
489 
490     // Initialize the section indirect symbol base, if necessary.
491     IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
492 
493     // Set the symbol type to undefined lazy, but only on construction.
494     //
495     // FIXME: Do not hardcode.
496     bool Created;
497     Asm.registerSymbol(*it->Symbol, &Created);
498     if (Created)
499       cast<MCSymbolMachO>(it->Symbol)->setReferenceTypeUndefinedLazy(true);
500   }
501 }
502 
503 /// computeSymbolTable - Compute the symbol table data
computeSymbolTable(MCAssembler & Asm,std::vector<MachSymbolData> & LocalSymbolData,std::vector<MachSymbolData> & ExternalSymbolData,std::vector<MachSymbolData> & UndefinedSymbolData)504 void MachObjectWriter::computeSymbolTable(
505     MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
506     std::vector<MachSymbolData> &ExternalSymbolData,
507     std::vector<MachSymbolData> &UndefinedSymbolData) {
508   // Build section lookup table.
509   DenseMap<const MCSection*, uint8_t> SectionIndexMap;
510   unsigned Index = 1;
511   for (MCAssembler::iterator it = Asm.begin(),
512          ie = Asm.end(); it != ie; ++it, ++Index)
513     SectionIndexMap[&*it] = Index;
514   assert(Index <= 256 && "Too many sections!");
515 
516   // Build the string table.
517   for (const MCSymbol &Symbol : Asm.symbols()) {
518     if (!Asm.isSymbolLinkerVisible(Symbol))
519       continue;
520 
521     StringTable.add(Symbol.getName());
522   }
523   StringTable.finalize();
524 
525   // Build the symbol arrays but only for non-local symbols.
526   //
527   // The particular order that we collect and then sort the symbols is chosen to
528   // match 'as'. Even though it doesn't matter for correctness, this is
529   // important for letting us diff .o files.
530   for (const MCSymbol &Symbol : Asm.symbols()) {
531     // Ignore non-linker visible symbols.
532     if (!Asm.isSymbolLinkerVisible(Symbol))
533       continue;
534 
535     if (!Symbol.isExternal() && !Symbol.isUndefined())
536       continue;
537 
538     MachSymbolData MSD;
539     MSD.Symbol = &Symbol;
540     MSD.StringIndex = StringTable.getOffset(Symbol.getName());
541 
542     if (Symbol.isUndefined()) {
543       MSD.SectionIndex = 0;
544       UndefinedSymbolData.push_back(MSD);
545     } else if (Symbol.isAbsolute()) {
546       MSD.SectionIndex = 0;
547       ExternalSymbolData.push_back(MSD);
548     } else {
549       MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
550       assert(MSD.SectionIndex && "Invalid section index!");
551       ExternalSymbolData.push_back(MSD);
552     }
553   }
554 
555   // Now add the data for local symbols.
556   for (const MCSymbol &Symbol : Asm.symbols()) {
557     // Ignore non-linker visible symbols.
558     if (!Asm.isSymbolLinkerVisible(Symbol))
559       continue;
560 
561     if (Symbol.isExternal() || Symbol.isUndefined())
562       continue;
563 
564     MachSymbolData MSD;
565     MSD.Symbol = &Symbol;
566     MSD.StringIndex = StringTable.getOffset(Symbol.getName());
567 
568     if (Symbol.isAbsolute()) {
569       MSD.SectionIndex = 0;
570       LocalSymbolData.push_back(MSD);
571     } else {
572       MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
573       assert(MSD.SectionIndex && "Invalid section index!");
574       LocalSymbolData.push_back(MSD);
575     }
576   }
577 
578   // External and undefined symbols are required to be in lexicographic order.
579   std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
580   std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
581 
582   // Set the symbol indices.
583   Index = 0;
584   for (auto *SymbolData :
585        {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
586     for (MachSymbolData &Entry : *SymbolData)
587       Entry.Symbol->setIndex(Index++);
588 
589   for (const MCSection &Section : Asm) {
590     for (RelAndSymbol &Rel : Relocations[&Section]) {
591       if (!Rel.Sym)
592         continue;
593 
594       // Set the Index and the IsExtern bit.
595       unsigned Index = Rel.Sym->getIndex();
596       assert(isInt<24>(Index));
597       if (IsLittleEndian)
598         Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
599       else
600         Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
601     }
602   }
603 }
604 
computeSectionAddresses(const MCAssembler & Asm,const MCAsmLayout & Layout)605 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
606                                                const MCAsmLayout &Layout) {
607   uint64_t StartAddress = 0;
608   for (const MCSection *Sec : Layout.getSectionOrder()) {
609     StartAddress = RoundUpToAlignment(StartAddress, Sec->getAlignment());
610     SectionAddress[Sec] = StartAddress;
611     StartAddress += Layout.getSectionAddressSize(Sec);
612 
613     // Explicitly pad the section to match the alignment requirements of the
614     // following one. This is for 'gas' compatibility, it shouldn't
615     /// strictly be necessary.
616     StartAddress += getPaddingSize(Sec, Layout);
617   }
618 }
619 
executePostLayoutBinding(MCAssembler & Asm,const MCAsmLayout & Layout)620 void MachObjectWriter::executePostLayoutBinding(MCAssembler &Asm,
621                                                 const MCAsmLayout &Layout) {
622   computeSectionAddresses(Asm, Layout);
623 
624   // Create symbol data for any indirect symbols.
625   bindIndirectSymbols(Asm);
626 }
627 
isSymbolRefDifferenceFullyResolvedImpl(const MCAssembler & Asm,const MCSymbol & A,const MCSymbol & B,bool InSet) const628 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
629     const MCAssembler &Asm, const MCSymbol &A, const MCSymbol &B,
630     bool InSet) const {
631   // FIXME: We don't handle things like
632   // foo = .
633   // creating atoms.
634   if (A.isVariable() || B.isVariable())
635     return false;
636   return MCObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(Asm, A, B,
637                                                                 InSet);
638 }
639 
isSymbolRefDifferenceFullyResolvedImpl(const MCAssembler & Asm,const MCSymbol & SymA,const MCFragment & FB,bool InSet,bool IsPCRel) const640 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
641     const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
642     bool InSet, bool IsPCRel) const {
643   if (InSet)
644     return true;
645 
646   // The effective address is
647   //     addr(atom(A)) + offset(A)
648   //   - addr(atom(B)) - offset(B)
649   // and the offsets are not relocatable, so the fixup is fully resolved when
650   //  addr(atom(A)) - addr(atom(B)) == 0.
651   const MCSymbol &SA = findAliasedSymbol(SymA);
652   const MCSection &SecA = SA.getSection();
653   const MCSection &SecB = *FB.getParent();
654 
655   if (IsPCRel) {
656     // The simple (Darwin, except on x86_64) way of dealing with this was to
657     // assume that any reference to a temporary symbol *must* be a temporary
658     // symbol in the same atom, unless the sections differ. Therefore, any PCrel
659     // relocation to a temporary symbol (in the same section) is fully
660     // resolved. This also works in conjunction with absolutized .set, which
661     // requires the compiler to use .set to absolutize the differences between
662     // symbols which the compiler knows to be assembly time constants, so we
663     // don't need to worry about considering symbol differences fully resolved.
664     //
665     // If the file isn't using sub-sections-via-symbols, we can make the
666     // same assumptions about any symbol that we normally make about
667     // assembler locals.
668 
669     bool hasReliableSymbolDifference = isX86_64();
670     if (!hasReliableSymbolDifference) {
671       if (!SA.isInSection() || &SecA != &SecB ||
672           (!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() &&
673            Asm.getSubsectionsViaSymbols()))
674         return false;
675       return true;
676     }
677     // For Darwin x86_64, there is one special case when the reference IsPCRel.
678     // If the fragment with the reference does not have a base symbol but meets
679     // the simple way of dealing with this, in that it is a temporary symbol in
680     // the same atom then it is assumed to be fully resolved.  This is needed so
681     // a relocation entry is not created and so the static linker does not
682     // mess up the reference later.
683     else if(!FB.getAtom() &&
684             SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
685       return true;
686     }
687   }
688 
689   // If they are not in the same section, we can't compute the diff.
690   if (&SecA != &SecB)
691     return false;
692 
693   const MCFragment *FA = SA.getFragment();
694 
695   // Bail if the symbol has no fragment.
696   if (!FA)
697     return false;
698 
699   // If the atoms are the same, they are guaranteed to have the same address.
700   if (FA->getAtom() == FB.getAtom())
701     return true;
702 
703   // Otherwise, we can't prove this is fully resolved.
704   return false;
705 }
706 
writeObject(MCAssembler & Asm,const MCAsmLayout & Layout)707 void MachObjectWriter::writeObject(MCAssembler &Asm,
708                                    const MCAsmLayout &Layout) {
709   // Compute symbol table information and bind symbol indices.
710   computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
711                      UndefinedSymbolData);
712 
713   unsigned NumSections = Asm.size();
714   const MCAssembler::VersionMinInfoType &VersionInfo =
715     Layout.getAssembler().getVersionMinInfo();
716 
717   // The section data starts after the header, the segment load command (and
718   // section headers) and the symbol table.
719   unsigned NumLoadCommands = 1;
720   uint64_t LoadCommandsSize = is64Bit() ?
721     sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
722     sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
723 
724   // Add the deployment target version info load command size, if used.
725   if (VersionInfo.Major != 0) {
726     ++NumLoadCommands;
727     LoadCommandsSize += sizeof(MachO::version_min_command);
728   }
729 
730   // Add the data-in-code load command size, if used.
731   unsigned NumDataRegions = Asm.getDataRegions().size();
732   if (NumDataRegions) {
733     ++NumLoadCommands;
734     LoadCommandsSize += sizeof(MachO::linkedit_data_command);
735   }
736 
737   // Add the loh load command size, if used.
738   uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
739   uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
740   if (LOHSize) {
741     ++NumLoadCommands;
742     LoadCommandsSize += sizeof(MachO::linkedit_data_command);
743   }
744 
745   // Add the symbol table load command sizes, if used.
746   unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
747     UndefinedSymbolData.size();
748   if (NumSymbols) {
749     NumLoadCommands += 2;
750     LoadCommandsSize += (sizeof(MachO::symtab_command) +
751                          sizeof(MachO::dysymtab_command));
752   }
753 
754   // Add the linker option load commands sizes.
755   for (const auto &Option : Asm.getLinkerOptions()) {
756     ++NumLoadCommands;
757     LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit());
758   }
759 
760   // Compute the total size of the section data, as well as its file size and vm
761   // size.
762   uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
763                                sizeof(MachO::mach_header)) + LoadCommandsSize;
764   uint64_t SectionDataSize = 0;
765   uint64_t SectionDataFileSize = 0;
766   uint64_t VMSize = 0;
767   for (const MCSection &Sec : Asm) {
768     uint64_t Address = getSectionAddress(&Sec);
769     uint64_t Size = Layout.getSectionAddressSize(&Sec);
770     uint64_t FileSize = Layout.getSectionFileSize(&Sec);
771     FileSize += getPaddingSize(&Sec, Layout);
772 
773     VMSize = std::max(VMSize, Address + Size);
774 
775     if (Sec.isVirtualSection())
776       continue;
777 
778     SectionDataSize = std::max(SectionDataSize, Address + Size);
779     SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
780   }
781 
782   // The section data is padded to 4 bytes.
783   //
784   // FIXME: Is this machine dependent?
785   unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
786   SectionDataFileSize += SectionDataPadding;
787 
788   // Write the prolog, starting with the header and load command...
789   writeHeader(MachO::MH_OBJECT, NumLoadCommands, LoadCommandsSize,
790               Asm.getSubsectionsViaSymbols());
791   uint32_t Prot =
792       MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE;
793   writeSegmentLoadCommand("", NumSections, 0, VMSize, SectionDataStart,
794                           SectionDataSize, Prot, Prot);
795 
796   // ... and then the section headers.
797   uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
798   for (const MCSection &Section : Asm) {
799     const auto &Sec = cast<MCSectionMachO>(Section);
800     std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
801     unsigned NumRelocs = Relocs.size();
802     uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec);
803     unsigned Flags = Sec.getTypeAndAttributes();
804     if (Sec.hasInstructions())
805       Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
806     writeSection(Layout, Sec, getSectionAddress(&Sec), SectionStart, Flags,
807                  RelocTableEnd, NumRelocs);
808     RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
809   }
810 
811   // Write out the deployment target information, if it's available.
812   if (VersionInfo.Major != 0) {
813     assert(VersionInfo.Update < 256 && "unencodable update target version");
814     assert(VersionInfo.Minor < 256 && "unencodable minor target version");
815     assert(VersionInfo.Major < 65536 && "unencodable major target version");
816     uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
817       (VersionInfo.Major << 16);
818     MachO::LoadCommandType LCType;
819     switch (VersionInfo.Kind) {
820     case MCVM_OSXVersionMin:
821       LCType = MachO::LC_VERSION_MIN_MACOSX;
822       break;
823     case MCVM_IOSVersionMin:
824       LCType = MachO::LC_VERSION_MIN_IPHONEOS;
825       break;
826     case MCVM_TvOSVersionMin:
827       LCType = MachO::LC_VERSION_MIN_TVOS;
828       break;
829     case MCVM_WatchOSVersionMin:
830       LCType = MachO::LC_VERSION_MIN_WATCHOS;
831       break;
832     }
833     write32(LCType);
834     write32(sizeof(MachO::version_min_command));
835     write32(EncodedVersion);
836     write32(0);         // reserved.
837   }
838 
839   // Write the data-in-code load command, if used.
840   uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
841   if (NumDataRegions) {
842     uint64_t DataRegionsOffset = RelocTableEnd;
843     uint64_t DataRegionsSize = NumDataRegions * 8;
844     writeLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
845                              DataRegionsSize);
846   }
847 
848   // Write the loh load command, if used.
849   uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
850   if (LOHSize)
851     writeLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
852                              DataInCodeTableEnd, LOHSize);
853 
854   // Write the symbol table load command, if used.
855   if (NumSymbols) {
856     unsigned FirstLocalSymbol = 0;
857     unsigned NumLocalSymbols = LocalSymbolData.size();
858     unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
859     unsigned NumExternalSymbols = ExternalSymbolData.size();
860     unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
861     unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
862     unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
863     unsigned NumSymTabSymbols =
864       NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
865     uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
866     uint64_t IndirectSymbolOffset = 0;
867 
868     // If used, the indirect symbols are written after the section data.
869     if (NumIndirectSymbols)
870       IndirectSymbolOffset = LOHTableEnd;
871 
872     // The symbol table is written after the indirect symbol data.
873     uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
874 
875     // The string table is written after symbol table.
876     uint64_t StringTableOffset =
877       SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
878                                               sizeof(MachO::nlist_64) :
879                                               sizeof(MachO::nlist));
880     writeSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
881                            StringTableOffset, StringTable.data().size());
882 
883     writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
884                              FirstExternalSymbol, NumExternalSymbols,
885                              FirstUndefinedSymbol, NumUndefinedSymbols,
886                              IndirectSymbolOffset, NumIndirectSymbols);
887   }
888 
889   // Write the linker options load commands.
890   for (const auto &Option : Asm.getLinkerOptions())
891     writeLinkerOptionsLoadCommand(Option);
892 
893   // Write the actual section data.
894   for (const MCSection &Sec : Asm) {
895     Asm.writeSectionData(&Sec, Layout);
896 
897     uint64_t Pad = getPaddingSize(&Sec, Layout);
898     WriteZeros(Pad);
899   }
900 
901   // Write the extra padding.
902   WriteZeros(SectionDataPadding);
903 
904   // Write the relocation entries.
905   for (const MCSection &Sec : Asm) {
906     // Write the section relocation entries, in reverse order to match 'as'
907     // (approximately, the exact algorithm is more complicated than this).
908     std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
909     for (const RelAndSymbol &Rel : make_range(Relocs.rbegin(), Relocs.rend())) {
910       write32(Rel.MRE.r_word0);
911       write32(Rel.MRE.r_word1);
912     }
913   }
914 
915   // Write out the data-in-code region payload, if there is one.
916   for (MCAssembler::const_data_region_iterator
917          it = Asm.data_region_begin(), ie = Asm.data_region_end();
918          it != ie; ++it) {
919     const DataRegionData *Data = &(*it);
920     uint64_t Start = getSymbolAddress(*Data->Start, Layout);
921     uint64_t End = getSymbolAddress(*Data->End, Layout);
922     DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
923                  << "  start: " << Start << "(" << Data->Start->getName() << ")"
924                  << "  end: " << End << "(" << Data->End->getName() << ")"
925                  << "  size: " << End - Start
926                  << "\n");
927     write32(Start);
928     write16(End - Start);
929     write16(Data->Kind);
930   }
931 
932   // Write out the loh commands, if there is one.
933   if (LOHSize) {
934 #ifndef NDEBUG
935     unsigned Start = getStream().tell();
936 #endif
937     Asm.getLOHContainer().emit(*this, Layout);
938     // Pad to a multiple of the pointer size.
939     writeBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
940     assert(getStream().tell() - Start == LOHSize);
941   }
942 
943   // Write the symbol table data, if used.
944   if (NumSymbols) {
945     // Write the indirect symbol entries.
946     for (MCAssembler::const_indirect_symbol_iterator
947            it = Asm.indirect_symbol_begin(),
948            ie = Asm.indirect_symbol_end(); it != ie; ++it) {
949       // Indirect symbols in the non-lazy symbol pointer section have some
950       // special handling.
951       const MCSectionMachO &Section =
952           static_cast<const MCSectionMachO &>(*it->Section);
953       if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
954         // If this symbol is defined and internal, mark it as such.
955         if (it->Symbol->isDefined() && !it->Symbol->isExternal()) {
956           uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
957           if (it->Symbol->isAbsolute())
958             Flags |= MachO::INDIRECT_SYMBOL_ABS;
959           write32(Flags);
960           continue;
961         }
962       }
963 
964       write32(it->Symbol->getIndex());
965     }
966 
967     // FIXME: Check that offsets match computed ones.
968 
969     // Write the symbol table entries.
970     for (auto *SymbolData :
971          {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
972       for (MachSymbolData &Entry : *SymbolData)
973         writeNlist(Entry, Layout);
974 
975     // Write the string table.
976     getStream() << StringTable.data();
977   }
978 }
979 
createMachObjectWriter(MCMachObjectTargetWriter * MOTW,raw_pwrite_stream & OS,bool IsLittleEndian)980 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
981                                              raw_pwrite_stream &OS,
982                                              bool IsLittleEndian) {
983   return new MachObjectWriter(MOTW, OS, IsLittleEndian);
984 }
985