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/MCMachOSymbolFlags.h"
19 #include "llvm/MC/MCObjectWriter.h"
20 #include "llvm/MC/MCSectionMachO.h"
21 #include "llvm/MC/MCSymbol.h"
22 #include "llvm/MC/MCValue.h"
23 #include "llvm/Support/Debug.h"
24 #include "llvm/Support/ErrorHandling.h"
25 #include "llvm/Support/MachO.h"
26 #include "llvm/Support/raw_ostream.h"
27 #include <vector>
28 using namespace llvm;
29 
30 #define DEBUG_TYPE "mc"
31 
reset()32 void MachObjectWriter::reset() {
33   Relocations.clear();
34   IndirectSymBase.clear();
35   StringTable.clear();
36   LocalSymbolData.clear();
37   ExternalSymbolData.clear();
38   UndefinedSymbolData.clear();
39   MCObjectWriter::reset();
40 }
41 
42 bool MachObjectWriter::
doesSymbolRequireExternRelocation(const MCSymbolData * SD)43 doesSymbolRequireExternRelocation(const MCSymbolData *SD) {
44   // Undefined symbols are always extern.
45   if (SD->getSymbol().isUndefined())
46     return true;
47 
48   // References to weak definitions require external relocation entries; the
49   // definition may not always be the one in the same object file.
50   if (SD->getFlags() & SF_WeakDefinition)
51     return true;
52 
53   // Otherwise, we can use an internal relocation.
54   return false;
55 }
56 
57 bool MachObjectWriter::
operator <(const MachSymbolData & RHS) const58 MachSymbolData::operator<(const MachSymbolData &RHS) const {
59   return SymbolData->getSymbol().getName() <
60     RHS.SymbolData->getSymbol().getName();
61 }
62 
isFixupKindPCRel(const MCAssembler & Asm,unsigned Kind)63 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
64   const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
65     (MCFixupKind) Kind);
66 
67   return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
68 }
69 
getFragmentAddress(const MCFragment * Fragment,const MCAsmLayout & Layout) const70 uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment,
71                                               const MCAsmLayout &Layout) const {
72   return getSectionAddress(Fragment->getParent()) +
73     Layout.getFragmentOffset(Fragment);
74 }
75 
getSymbolAddress(const MCSymbolData * SD,const MCAsmLayout & Layout) const76 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbolData* SD,
77                                             const MCAsmLayout &Layout) const {
78   const MCSymbol &S = SD->getSymbol();
79 
80   // If this is a variable, then recursively evaluate now.
81   if (S.isVariable()) {
82     if (const MCConstantExpr *C =
83           dyn_cast<const MCConstantExpr>(S.getVariableValue()))
84       return C->getValue();
85 
86 
87     MCValue Target;
88     if (!S.getVariableValue()->EvaluateAsRelocatable(Target, &Layout, nullptr))
89       report_fatal_error("unable to evaluate offset for variable '" +
90                          S.getName() + "'");
91 
92     // Verify that any used symbols are defined.
93     if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
94       report_fatal_error("unable to evaluate offset to undefined symbol '" +
95                          Target.getSymA()->getSymbol().getName() + "'");
96     if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
97       report_fatal_error("unable to evaluate offset to undefined symbol '" +
98                          Target.getSymB()->getSymbol().getName() + "'");
99 
100     uint64_t Address = Target.getConstant();
101     if (Target.getSymA())
102       Address += getSymbolAddress(&Layout.getAssembler().getSymbolData(
103                                     Target.getSymA()->getSymbol()), Layout);
104     if (Target.getSymB())
105       Address += getSymbolAddress(&Layout.getAssembler().getSymbolData(
106                                     Target.getSymB()->getSymbol()), Layout);
107     return Address;
108   }
109 
110   return getSectionAddress(SD->getFragment()->getParent()) +
111     Layout.getSymbolOffset(SD);
112 }
113 
getPaddingSize(const MCSectionData * SD,const MCAsmLayout & Layout) const114 uint64_t MachObjectWriter::getPaddingSize(const MCSectionData *SD,
115                                           const MCAsmLayout &Layout) const {
116   uint64_t EndAddr = getSectionAddress(SD) + Layout.getSectionAddressSize(SD);
117   unsigned Next = SD->getLayoutOrder() + 1;
118   if (Next >= Layout.getSectionOrder().size())
119     return 0;
120 
121   const MCSectionData &NextSD = *Layout.getSectionOrder()[Next];
122   if (NextSD.getSection().isVirtualSection())
123     return 0;
124   return OffsetToAlignment(EndAddr, NextSD.getAlignment());
125 }
126 
WriteHeader(unsigned NumLoadCommands,unsigned LoadCommandsSize,bool SubsectionsViaSymbols)127 void MachObjectWriter::WriteHeader(unsigned NumLoadCommands,
128                                    unsigned LoadCommandsSize,
129                                    bool SubsectionsViaSymbols) {
130   uint32_t Flags = 0;
131 
132   if (SubsectionsViaSymbols)
133     Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
134 
135   // struct mach_header (28 bytes) or
136   // struct mach_header_64 (32 bytes)
137 
138   uint64_t Start = OS.tell();
139   (void) Start;
140 
141   Write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
142 
143   Write32(TargetObjectWriter->getCPUType());
144   Write32(TargetObjectWriter->getCPUSubtype());
145 
146   Write32(MachO::MH_OBJECT);
147   Write32(NumLoadCommands);
148   Write32(LoadCommandsSize);
149   Write32(Flags);
150   if (is64Bit())
151     Write32(0); // reserved
152 
153   assert(OS.tell() - Start ==
154          (is64Bit()?sizeof(MachO::mach_header_64): sizeof(MachO::mach_header)));
155 }
156 
157 /// WriteSegmentLoadCommand - Write a segment load command.
158 ///
159 /// \param NumSections The number of sections in this segment.
160 /// \param SectionDataSize The total size of the sections.
WriteSegmentLoadCommand(unsigned NumSections,uint64_t VMSize,uint64_t SectionDataStartOffset,uint64_t SectionDataSize)161 void MachObjectWriter::WriteSegmentLoadCommand(unsigned NumSections,
162                                                uint64_t VMSize,
163                                                uint64_t SectionDataStartOffset,
164                                                uint64_t SectionDataSize) {
165   // struct segment_command (56 bytes) or
166   // struct segment_command_64 (72 bytes)
167 
168   uint64_t Start = OS.tell();
169   (void) Start;
170 
171   unsigned SegmentLoadCommandSize =
172     is64Bit() ? sizeof(MachO::segment_command_64):
173     sizeof(MachO::segment_command);
174   Write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
175   Write32(SegmentLoadCommandSize +
176           NumSections * (is64Bit() ? sizeof(MachO::section_64) :
177                          sizeof(MachO::section)));
178 
179   WriteBytes("", 16);
180   if (is64Bit()) {
181     Write64(0); // vmaddr
182     Write64(VMSize); // vmsize
183     Write64(SectionDataStartOffset); // file offset
184     Write64(SectionDataSize); // file size
185   } else {
186     Write32(0); // vmaddr
187     Write32(VMSize); // vmsize
188     Write32(SectionDataStartOffset); // file offset
189     Write32(SectionDataSize); // file size
190   }
191   // maxprot
192   Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
193   // initprot
194   Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
195   Write32(NumSections);
196   Write32(0); // flags
197 
198   assert(OS.tell() - Start == SegmentLoadCommandSize);
199 }
200 
WriteSection(const MCAssembler & Asm,const MCAsmLayout & Layout,const MCSectionData & SD,uint64_t FileOffset,uint64_t RelocationsStart,unsigned NumRelocations)201 void MachObjectWriter::WriteSection(const MCAssembler &Asm,
202                                     const MCAsmLayout &Layout,
203                                     const MCSectionData &SD,
204                                     uint64_t FileOffset,
205                                     uint64_t RelocationsStart,
206                                     unsigned NumRelocations) {
207   uint64_t SectionSize = Layout.getSectionAddressSize(&SD);
208 
209   // The offset is unused for virtual sections.
210   if (SD.getSection().isVirtualSection()) {
211     assert(Layout.getSectionFileSize(&SD) == 0 && "Invalid file size!");
212     FileOffset = 0;
213   }
214 
215   // struct section (68 bytes) or
216   // struct section_64 (80 bytes)
217 
218   uint64_t Start = OS.tell();
219   (void) Start;
220 
221   const MCSectionMachO &Section = cast<MCSectionMachO>(SD.getSection());
222   WriteBytes(Section.getSectionName(), 16);
223   WriteBytes(Section.getSegmentName(), 16);
224   if (is64Bit()) {
225     Write64(getSectionAddress(&SD)); // address
226     Write64(SectionSize); // size
227   } else {
228     Write32(getSectionAddress(&SD)); // address
229     Write32(SectionSize); // size
230   }
231   Write32(FileOffset);
232 
233   unsigned Flags = Section.getTypeAndAttributes();
234   if (SD.hasInstructions())
235     Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
236 
237   assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
238   Write32(Log2_32(SD.getAlignment()));
239   Write32(NumRelocations ? RelocationsStart : 0);
240   Write32(NumRelocations);
241   Write32(Flags);
242   Write32(IndirectSymBase.lookup(&SD)); // reserved1
243   Write32(Section.getStubSize()); // reserved2
244   if (is64Bit())
245     Write32(0); // reserved3
246 
247   assert(OS.tell() - Start == (is64Bit() ? sizeof(MachO::section_64) :
248                                sizeof(MachO::section)));
249 }
250 
WriteSymtabLoadCommand(uint32_t SymbolOffset,uint32_t NumSymbols,uint32_t StringTableOffset,uint32_t StringTableSize)251 void MachObjectWriter::WriteSymtabLoadCommand(uint32_t SymbolOffset,
252                                               uint32_t NumSymbols,
253                                               uint32_t StringTableOffset,
254                                               uint32_t StringTableSize) {
255   // struct symtab_command (24 bytes)
256 
257   uint64_t Start = OS.tell();
258   (void) Start;
259 
260   Write32(MachO::LC_SYMTAB);
261   Write32(sizeof(MachO::symtab_command));
262   Write32(SymbolOffset);
263   Write32(NumSymbols);
264   Write32(StringTableOffset);
265   Write32(StringTableSize);
266 
267   assert(OS.tell() - Start == sizeof(MachO::symtab_command));
268 }
269 
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)270 void MachObjectWriter::WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
271                                                 uint32_t NumLocalSymbols,
272                                                 uint32_t FirstExternalSymbol,
273                                                 uint32_t NumExternalSymbols,
274                                                 uint32_t FirstUndefinedSymbol,
275                                                 uint32_t NumUndefinedSymbols,
276                                                 uint32_t IndirectSymbolOffset,
277                                                 uint32_t NumIndirectSymbols) {
278   // struct dysymtab_command (80 bytes)
279 
280   uint64_t Start = OS.tell();
281   (void) Start;
282 
283   Write32(MachO::LC_DYSYMTAB);
284   Write32(sizeof(MachO::dysymtab_command));
285   Write32(FirstLocalSymbol);
286   Write32(NumLocalSymbols);
287   Write32(FirstExternalSymbol);
288   Write32(NumExternalSymbols);
289   Write32(FirstUndefinedSymbol);
290   Write32(NumUndefinedSymbols);
291   Write32(0); // tocoff
292   Write32(0); // ntoc
293   Write32(0); // modtaboff
294   Write32(0); // nmodtab
295   Write32(0); // extrefsymoff
296   Write32(0); // nextrefsyms
297   Write32(IndirectSymbolOffset);
298   Write32(NumIndirectSymbols);
299   Write32(0); // extreloff
300   Write32(0); // nextrel
301   Write32(0); // locreloff
302   Write32(0); // nlocrel
303 
304   assert(OS.tell() - Start == sizeof(MachO::dysymtab_command));
305 }
306 
307 MachObjectWriter::MachSymbolData *
findSymbolData(const MCSymbol & Sym)308 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
309   for (auto &Entry : LocalSymbolData)
310     if (&Entry.SymbolData->getSymbol() == &Sym)
311       return &Entry;
312 
313   for (auto &Entry : ExternalSymbolData)
314     if (&Entry.SymbolData->getSymbol() == &Sym)
315       return &Entry;
316 
317   for (auto &Entry : UndefinedSymbolData)
318     if (&Entry.SymbolData->getSymbol() == &Sym)
319       return &Entry;
320 
321   return nullptr;
322 }
323 
WriteNlist(MachSymbolData & MSD,const MCAsmLayout & Layout)324 void MachObjectWriter::WriteNlist(MachSymbolData &MSD,
325                                   const MCAsmLayout &Layout) {
326   MCSymbolData &Data = *MSD.SymbolData;
327   const MCSymbol *Symbol = &Data.getSymbol();
328   const MCSymbol *AliasedSymbol = &Symbol->AliasedSymbol();
329   uint8_t SectionIndex = MSD.SectionIndex;
330   uint8_t Type = 0;
331   uint16_t Flags = Data.getFlags();
332   uint64_t Address = 0;
333   bool IsAlias = Symbol != AliasedSymbol;
334 
335   MachSymbolData *AliaseeInfo;
336   if (IsAlias) {
337     AliaseeInfo = findSymbolData(*AliasedSymbol);
338     if (AliaseeInfo)
339       SectionIndex = AliaseeInfo->SectionIndex;
340     Symbol = AliasedSymbol;
341   }
342 
343   // Set the N_TYPE bits. See <mach-o/nlist.h>.
344   //
345   // FIXME: Are the prebound or indirect fields possible here?
346   if (IsAlias && Symbol->isUndefined())
347     Type = MachO::N_INDR;
348   else if (Symbol->isUndefined())
349     Type = MachO::N_UNDF;
350   else if (Symbol->isAbsolute())
351     Type = MachO::N_ABS;
352   else
353     Type = MachO::N_SECT;
354 
355   // FIXME: Set STAB bits.
356 
357   if (Data.isPrivateExtern())
358     Type |= MachO::N_PEXT;
359 
360   // Set external bit.
361   if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
362     Type |= MachO::N_EXT;
363 
364   // Compute the symbol address.
365   if (IsAlias && Symbol->isUndefined())
366     Address = AliaseeInfo->StringIndex;
367   else if (Symbol->isDefined())
368     Address = getSymbolAddress(&Data, Layout);
369   else if (Data.isCommon()) {
370     // Common symbols are encoded with the size in the address
371     // field, and their alignment in the flags.
372     Address = Data.getCommonSize();
373 
374     // Common alignment is packed into the 'desc' bits.
375     if (unsigned Align = Data.getCommonAlignment()) {
376       unsigned Log2Size = Log2_32(Align);
377       assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
378       if (Log2Size > 15)
379         report_fatal_error("invalid 'common' alignment '" +
380                            Twine(Align) + "' for '" + Symbol->getName() + "'",
381                            false);
382       // FIXME: Keep this mask with the SymbolFlags enumeration.
383       Flags = (Flags & 0xF0FF) | (Log2Size << 8);
384     }
385   }
386 
387   if (Layout.getAssembler().isThumbFunc(Symbol))
388     Flags |= SF_ThumbFunc;
389 
390   // struct nlist (12 bytes)
391 
392   Write32(MSD.StringIndex);
393   Write8(Type);
394   Write8(SectionIndex);
395 
396   // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
397   // value.
398   Write16(Flags);
399   if (is64Bit())
400     Write64(Address);
401   else
402     Write32(Address);
403 }
404 
WriteLinkeditLoadCommand(uint32_t Type,uint32_t DataOffset,uint32_t DataSize)405 void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type,
406                                                 uint32_t DataOffset,
407                                                 uint32_t DataSize) {
408   uint64_t Start = OS.tell();
409   (void) Start;
410 
411   Write32(Type);
412   Write32(sizeof(MachO::linkedit_data_command));
413   Write32(DataOffset);
414   Write32(DataSize);
415 
416   assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
417 }
418 
ComputeLinkerOptionsLoadCommandSize(const std::vector<std::string> & Options,bool is64Bit)419 static unsigned ComputeLinkerOptionsLoadCommandSize(
420   const std::vector<std::string> &Options, bool is64Bit)
421 {
422   unsigned Size = sizeof(MachO::linker_option_command);
423   for (unsigned i = 0, e = Options.size(); i != e; ++i)
424     Size += Options[i].size() + 1;
425   return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
426 }
427 
WriteLinkerOptionsLoadCommand(const std::vector<std::string> & Options)428 void MachObjectWriter::WriteLinkerOptionsLoadCommand(
429   const std::vector<std::string> &Options)
430 {
431   unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
432   uint64_t Start = OS.tell();
433   (void) Start;
434 
435   Write32(MachO::LC_LINKER_OPTION);
436   Write32(Size);
437   Write32(Options.size());
438   uint64_t BytesWritten = sizeof(MachO::linker_option_command);
439   for (unsigned i = 0, e = Options.size(); i != e; ++i) {
440     // Write each string, including the null byte.
441     const std::string &Option = Options[i];
442     WriteBytes(Option.c_str(), Option.size() + 1);
443     BytesWritten += Option.size() + 1;
444   }
445 
446   // Pad to a multiple of the pointer size.
447   WriteBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
448 
449   assert(OS.tell() - Start == Size);
450 }
451 
RecordRelocation(MCAssembler & Asm,const MCAsmLayout & Layout,const MCFragment * Fragment,const MCFixup & Fixup,MCValue Target,bool & IsPCRel,uint64_t & FixedValue)452 void MachObjectWriter::RecordRelocation(MCAssembler &Asm,
453                                         const MCAsmLayout &Layout,
454                                         const MCFragment *Fragment,
455                                         const MCFixup &Fixup, MCValue Target,
456                                         bool &IsPCRel, uint64_t &FixedValue) {
457   TargetObjectWriter->RecordRelocation(this, Asm, Layout, Fragment, Fixup,
458                                        Target, FixedValue);
459 }
460 
BindIndirectSymbols(MCAssembler & Asm)461 void MachObjectWriter::BindIndirectSymbols(MCAssembler &Asm) {
462   // This is the point where 'as' creates actual symbols for indirect symbols
463   // (in the following two passes). It would be easier for us to do this sooner
464   // when we see the attribute, but that makes getting the order in the symbol
465   // table much more complicated than it is worth.
466   //
467   // FIXME: Revisit this when the dust settles.
468 
469   // Report errors for use of .indirect_symbol not in a symbol pointer section
470   // or stub section.
471   for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
472          ie = Asm.indirect_symbol_end(); it != ie; ++it) {
473     const MCSectionMachO &Section =
474       cast<MCSectionMachO>(it->SectionData->getSection());
475 
476     if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
477         Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
478         Section.getType() != MachO::S_SYMBOL_STUBS) {
479 	MCSymbol &Symbol = *it->Symbol;
480 	report_fatal_error("indirect symbol '" + Symbol.getName() +
481                            "' not in a symbol pointer or stub section");
482     }
483   }
484 
485   // Bind non-lazy symbol pointers first.
486   unsigned IndirectIndex = 0;
487   for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
488          ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
489     const MCSectionMachO &Section =
490       cast<MCSectionMachO>(it->SectionData->getSection());
491 
492     if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
493       continue;
494 
495     // Initialize the section indirect symbol base, if necessary.
496     IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
497 
498     Asm.getOrCreateSymbolData(*it->Symbol);
499   }
500 
501   // Then lazy symbol pointers and symbol stubs.
502   IndirectIndex = 0;
503   for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
504          ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
505     const MCSectionMachO &Section =
506       cast<MCSectionMachO>(it->SectionData->getSection());
507 
508     if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
509         Section.getType() != MachO::S_SYMBOL_STUBS)
510       continue;
511 
512     // Initialize the section indirect symbol base, if necessary.
513     IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
514 
515     // Set the symbol type to undefined lazy, but only on construction.
516     //
517     // FIXME: Do not hardcode.
518     bool Created;
519     MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created);
520     if (Created)
521       Entry.setFlags(Entry.getFlags() | 0x0001);
522   }
523 }
524 
525 /// ComputeSymbolTable - Compute the symbol table data
ComputeSymbolTable(MCAssembler & Asm,std::vector<MachSymbolData> & LocalSymbolData,std::vector<MachSymbolData> & ExternalSymbolData,std::vector<MachSymbolData> & UndefinedSymbolData)526 void MachObjectWriter::ComputeSymbolTable(
527     MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
528     std::vector<MachSymbolData> &ExternalSymbolData,
529     std::vector<MachSymbolData> &UndefinedSymbolData) {
530   // Build section lookup table.
531   DenseMap<const MCSection*, uint8_t> SectionIndexMap;
532   unsigned Index = 1;
533   for (MCAssembler::iterator it = Asm.begin(),
534          ie = Asm.end(); it != ie; ++it, ++Index)
535     SectionIndexMap[&it->getSection()] = Index;
536   assert(Index <= 256 && "Too many sections!");
537 
538   // Build the string table.
539   for (MCSymbolData &SD : Asm.symbols()) {
540     const MCSymbol &Symbol = SD.getSymbol();
541     if (!Asm.isSymbolLinkerVisible(Symbol))
542       continue;
543 
544     StringTable.add(Symbol.getName());
545   }
546   StringTable.finalize(StringTableBuilder::MachO);
547 
548   // Build the symbol arrays but only for non-local symbols.
549   //
550   // The particular order that we collect and then sort the symbols is chosen to
551   // match 'as'. Even though it doesn't matter for correctness, this is
552   // important for letting us diff .o files.
553   for (MCSymbolData &SD : Asm.symbols()) {
554     const MCSymbol &Symbol = SD.getSymbol();
555 
556     // Ignore non-linker visible symbols.
557     if (!Asm.isSymbolLinkerVisible(Symbol))
558       continue;
559 
560     if (!SD.isExternal() && !Symbol.isUndefined())
561       continue;
562 
563     MachSymbolData MSD;
564     MSD.SymbolData = &SD;
565     MSD.StringIndex = StringTable.getOffset(Symbol.getName());
566 
567     if (Symbol.isUndefined()) {
568       MSD.SectionIndex = 0;
569       UndefinedSymbolData.push_back(MSD);
570     } else if (Symbol.isAbsolute()) {
571       MSD.SectionIndex = 0;
572       ExternalSymbolData.push_back(MSD);
573     } else {
574       MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
575       assert(MSD.SectionIndex && "Invalid section index!");
576       ExternalSymbolData.push_back(MSD);
577     }
578   }
579 
580   // Now add the data for local symbols.
581   for (MCSymbolData &SD : Asm.symbols()) {
582     const MCSymbol &Symbol = SD.getSymbol();
583 
584     // Ignore non-linker visible symbols.
585     if (!Asm.isSymbolLinkerVisible(Symbol))
586       continue;
587 
588     if (SD.isExternal() || Symbol.isUndefined())
589       continue;
590 
591     MachSymbolData MSD;
592     MSD.SymbolData = &SD;
593     MSD.StringIndex = StringTable.getOffset(Symbol.getName());
594 
595     if (Symbol.isAbsolute()) {
596       MSD.SectionIndex = 0;
597       LocalSymbolData.push_back(MSD);
598     } else {
599       MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
600       assert(MSD.SectionIndex && "Invalid section index!");
601       LocalSymbolData.push_back(MSD);
602     }
603   }
604 
605   // External and undefined symbols are required to be in lexicographic order.
606   std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
607   std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
608 
609   // Set the symbol indices.
610   Index = 0;
611   for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
612     LocalSymbolData[i].SymbolData->setIndex(Index++);
613   for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
614     ExternalSymbolData[i].SymbolData->setIndex(Index++);
615   for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
616     UndefinedSymbolData[i].SymbolData->setIndex(Index++);
617 
618   for (const MCSectionData &SD : Asm) {
619     std::vector<RelAndSymbol> &Relocs = Relocations[&SD];
620     for (RelAndSymbol &Rel : Relocs) {
621       if (!Rel.Sym)
622         continue;
623 
624       // Set the Index and the IsExtern bit.
625       unsigned Index = Rel.Sym->getIndex();
626       assert(isInt<24>(Index));
627       if (IsLittleEndian)
628         Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (-1 << 24)) | Index | (1 << 27);
629       else
630         Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
631     }
632   }
633 }
634 
computeSectionAddresses(const MCAssembler & Asm,const MCAsmLayout & Layout)635 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
636                                                const MCAsmLayout &Layout) {
637   uint64_t StartAddress = 0;
638   const SmallVectorImpl<MCSectionData*> &Order = Layout.getSectionOrder();
639   for (int i = 0, n = Order.size(); i != n ; ++i) {
640     const MCSectionData *SD = Order[i];
641     StartAddress = RoundUpToAlignment(StartAddress, SD->getAlignment());
642     SectionAddress[SD] = StartAddress;
643     StartAddress += Layout.getSectionAddressSize(SD);
644 
645     // Explicitly pad the section to match the alignment requirements of the
646     // following one. This is for 'gas' compatibility, it shouldn't
647     /// strictly be necessary.
648     StartAddress += getPaddingSize(SD, Layout);
649   }
650 }
651 
ExecutePostLayoutBinding(MCAssembler & Asm,const MCAsmLayout & Layout)652 void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm,
653                                                 const MCAsmLayout &Layout) {
654   computeSectionAddresses(Asm, Layout);
655 
656   // Create symbol data for any indirect symbols.
657   BindIndirectSymbols(Asm);
658 }
659 
660 bool MachObjectWriter::
IsSymbolRefDifferenceFullyResolvedImpl(const MCAssembler & Asm,const MCSymbolData & DataA,const MCSymbolData * DataB,const MCFragment & FB,bool InSet,bool IsPCRel) const661 IsSymbolRefDifferenceFullyResolvedImpl(const MCAssembler &Asm,
662                                        const MCSymbolData &DataA,
663                                        const MCSymbolData *DataB,
664                                        const MCFragment &FB,
665                                        bool InSet,
666                                        bool IsPCRel) const {
667   if (InSet)
668     return true;
669 
670   // The effective address is
671   //     addr(atom(A)) + offset(A)
672   //   - addr(atom(B)) - offset(B)
673   // and the offsets are not relocatable, so the fixup is fully resolved when
674   //  addr(atom(A)) - addr(atom(B)) == 0.
675   const MCSymbolData *A_Base = nullptr, *B_Base = nullptr;
676 
677   const MCSymbol &SA = DataA.getSymbol().AliasedSymbol();
678   const MCSection &SecA = SA.getSection();
679   const MCSection &SecB = FB.getParent()->getSection();
680 
681   if (IsPCRel) {
682     // The simple (Darwin, except on x86_64) way of dealing with this was to
683     // assume that any reference to a temporary symbol *must* be a temporary
684     // symbol in the same atom, unless the sections differ. Therefore, any PCrel
685     // relocation to a temporary symbol (in the same section) is fully
686     // resolved. This also works in conjunction with absolutized .set, which
687     // requires the compiler to use .set to absolutize the differences between
688     // symbols which the compiler knows to be assembly time constants, so we
689     // don't need to worry about considering symbol differences fully resolved.
690     //
691     // If the file isn't using sub-sections-via-symbols, we can make the
692     // same assumptions about any symbol that we normally make about
693     // assembler locals.
694 
695     bool hasReliableSymbolDifference = isX86_64();
696     if (!hasReliableSymbolDifference) {
697       if (!SA.isInSection() || &SecA != &SecB ||
698           (!SA.isTemporary() &&
699            FB.getAtom() != Asm.getSymbolData(SA).getFragment()->getAtom() &&
700            Asm.getSubsectionsViaSymbols()))
701         return false;
702       return true;
703     }
704     // For Darwin x86_64, there is one special case when the reference IsPCRel.
705     // If the fragment with the reference does not have a base symbol but meets
706     // the simple way of dealing with this, in that it is a temporary symbol in
707     // the same atom then it is assumed to be fully resolved.  This is needed so
708     // a relocation entry is not created and so the static linker does not
709     // mess up the reference later.
710     else if(!FB.getAtom() &&
711             SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
712       return true;
713     }
714   } else {
715     if (!TargetObjectWriter->useAggressiveSymbolFolding())
716       return false;
717   }
718 
719   // If they are not in the same section, we can't compute the diff.
720   if (&SecA != &SecB)
721     return false;
722 
723   const MCFragment *FA = Asm.getSymbolData(SA).getFragment();
724 
725   // Bail if the symbol has no fragment.
726   if (!FA)
727     return false;
728 
729   A_Base = FA->getAtom();
730   B_Base = FB.getAtom();
731 
732   // If the atoms are the same, they are guaranteed to have the same address.
733   if (A_Base == B_Base)
734     return true;
735 
736   // Otherwise, we can't prove this is fully resolved.
737   return false;
738 }
739 
WriteObject(MCAssembler & Asm,const MCAsmLayout & Layout)740 void MachObjectWriter::WriteObject(MCAssembler &Asm,
741                                    const MCAsmLayout &Layout) {
742   // Compute symbol table information and bind symbol indices.
743   ComputeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
744                      UndefinedSymbolData);
745 
746   unsigned NumSections = Asm.size();
747   const MCAssembler::VersionMinInfoType &VersionInfo =
748     Layout.getAssembler().getVersionMinInfo();
749 
750   // The section data starts after the header, the segment load command (and
751   // section headers) and the symbol table.
752   unsigned NumLoadCommands = 1;
753   uint64_t LoadCommandsSize = is64Bit() ?
754     sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
755     sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
756 
757   // Add the deployment target version info load command size, if used.
758   if (VersionInfo.Major != 0) {
759     ++NumLoadCommands;
760     LoadCommandsSize += sizeof(MachO::version_min_command);
761   }
762 
763   // Add the data-in-code load command size, if used.
764   unsigned NumDataRegions = Asm.getDataRegions().size();
765   if (NumDataRegions) {
766     ++NumLoadCommands;
767     LoadCommandsSize += sizeof(MachO::linkedit_data_command);
768   }
769 
770   // Add the loh load command size, if used.
771   uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
772   uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
773   if (LOHSize) {
774     ++NumLoadCommands;
775     LoadCommandsSize += sizeof(MachO::linkedit_data_command);
776   }
777 
778   // Add the symbol table load command sizes, if used.
779   unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
780     UndefinedSymbolData.size();
781   if (NumSymbols) {
782     NumLoadCommands += 2;
783     LoadCommandsSize += (sizeof(MachO::symtab_command) +
784                          sizeof(MachO::dysymtab_command));
785   }
786 
787   // Add the linker option load commands sizes.
788   const std::vector<std::vector<std::string> > &LinkerOptions =
789     Asm.getLinkerOptions();
790   for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
791     ++NumLoadCommands;
792     LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i],
793                                                             is64Bit());
794   }
795 
796   // Compute the total size of the section data, as well as its file size and vm
797   // size.
798   uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
799                                sizeof(MachO::mach_header)) + LoadCommandsSize;
800   uint64_t SectionDataSize = 0;
801   uint64_t SectionDataFileSize = 0;
802   uint64_t VMSize = 0;
803   for (MCAssembler::const_iterator it = Asm.begin(),
804          ie = Asm.end(); it != ie; ++it) {
805     const MCSectionData &SD = *it;
806     uint64_t Address = getSectionAddress(&SD);
807     uint64_t Size = Layout.getSectionAddressSize(&SD);
808     uint64_t FileSize = Layout.getSectionFileSize(&SD);
809     FileSize += getPaddingSize(&SD, Layout);
810 
811     VMSize = std::max(VMSize, Address + Size);
812 
813     if (SD.getSection().isVirtualSection())
814       continue;
815 
816     SectionDataSize = std::max(SectionDataSize, Address + Size);
817     SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
818   }
819 
820   // The section data is padded to 4 bytes.
821   //
822   // FIXME: Is this machine dependent?
823   unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
824   SectionDataFileSize += SectionDataPadding;
825 
826   // Write the prolog, starting with the header and load command...
827   WriteHeader(NumLoadCommands, LoadCommandsSize,
828               Asm.getSubsectionsViaSymbols());
829   WriteSegmentLoadCommand(NumSections, VMSize,
830                           SectionDataStart, SectionDataSize);
831 
832   // ... and then the section headers.
833   uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
834   for (MCAssembler::const_iterator it = Asm.begin(),
835          ie = Asm.end(); it != ie; ++it) {
836     std::vector<RelAndSymbol> &Relocs = Relocations[it];
837     unsigned NumRelocs = Relocs.size();
838     uint64_t SectionStart = SectionDataStart + getSectionAddress(it);
839     WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
840     RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
841   }
842 
843   // Write out the deployment target information, if it's available.
844   if (VersionInfo.Major != 0) {
845     assert(VersionInfo.Update < 256 && "unencodable update target version");
846     assert(VersionInfo.Minor < 256 && "unencodable minor target version");
847     assert(VersionInfo.Major < 65536 && "unencodable major target version");
848     uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
849       (VersionInfo.Major << 16);
850     Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
851             MachO::LC_VERSION_MIN_IPHONEOS);
852     Write32(sizeof(MachO::version_min_command));
853     Write32(EncodedVersion);
854     Write32(0);         // reserved.
855   }
856 
857   // Write the data-in-code load command, if used.
858   uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
859   if (NumDataRegions) {
860     uint64_t DataRegionsOffset = RelocTableEnd;
861     uint64_t DataRegionsSize = NumDataRegions * 8;
862     WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
863                              DataRegionsSize);
864   }
865 
866   // Write the loh load command, if used.
867   uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
868   if (LOHSize)
869     WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
870                              DataInCodeTableEnd, LOHSize);
871 
872   // Write the symbol table load command, if used.
873   if (NumSymbols) {
874     unsigned FirstLocalSymbol = 0;
875     unsigned NumLocalSymbols = LocalSymbolData.size();
876     unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
877     unsigned NumExternalSymbols = ExternalSymbolData.size();
878     unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
879     unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
880     unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
881     unsigned NumSymTabSymbols =
882       NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
883     uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
884     uint64_t IndirectSymbolOffset = 0;
885 
886     // If used, the indirect symbols are written after the section data.
887     if (NumIndirectSymbols)
888       IndirectSymbolOffset = LOHTableEnd;
889 
890     // The symbol table is written after the indirect symbol data.
891     uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
892 
893     // The string table is written after symbol table.
894     uint64_t StringTableOffset =
895       SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
896                                               sizeof(MachO::nlist_64) :
897                                               sizeof(MachO::nlist));
898     WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
899                            StringTableOffset, StringTable.data().size());
900 
901     WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
902                              FirstExternalSymbol, NumExternalSymbols,
903                              FirstUndefinedSymbol, NumUndefinedSymbols,
904                              IndirectSymbolOffset, NumIndirectSymbols);
905   }
906 
907   // Write the linker options load commands.
908   for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
909     WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
910   }
911 
912   // Write the actual section data.
913   for (MCAssembler::const_iterator it = Asm.begin(),
914          ie = Asm.end(); it != ie; ++it) {
915     Asm.writeSectionData(it, Layout);
916 
917     uint64_t Pad = getPaddingSize(it, Layout);
918     for (unsigned int i = 0; i < Pad; ++i)
919       Write8(0);
920   }
921 
922   // Write the extra padding.
923   WriteZeros(SectionDataPadding);
924 
925   // Write the relocation entries.
926   for (MCAssembler::const_iterator it = Asm.begin(),
927          ie = Asm.end(); it != ie; ++it) {
928     // Write the section relocation entries, in reverse order to match 'as'
929     // (approximately, the exact algorithm is more complicated than this).
930     std::vector<RelAndSymbol> &Relocs = Relocations[it];
931     for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
932       Write32(Relocs[e - i - 1].MRE.r_word0);
933       Write32(Relocs[e - i - 1].MRE.r_word1);
934     }
935   }
936 
937   // Write out the data-in-code region payload, if there is one.
938   for (MCAssembler::const_data_region_iterator
939          it = Asm.data_region_begin(), ie = Asm.data_region_end();
940          it != ie; ++it) {
941     const DataRegionData *Data = &(*it);
942     uint64_t Start =
943       getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->Start),
944                        Layout);
945     uint64_t End =
946       getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->End),
947                        Layout);
948     DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
949                  << "  start: " << Start << "(" << Data->Start->getName() << ")"
950                  << "  end: " << End << "(" << Data->End->getName() << ")"
951                  << "  size: " << End - Start
952                  << "\n");
953     Write32(Start);
954     Write16(End - Start);
955     Write16(Data->Kind);
956   }
957 
958   // Write out the loh commands, if there is one.
959   if (LOHSize) {
960 #ifndef NDEBUG
961     unsigned Start = OS.tell();
962 #endif
963     Asm.getLOHContainer().Emit(*this, Layout);
964     // Pad to a multiple of the pointer size.
965     WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
966     assert(OS.tell() - Start == LOHSize);
967   }
968 
969   // Write the symbol table data, if used.
970   if (NumSymbols) {
971     // Write the indirect symbol entries.
972     for (MCAssembler::const_indirect_symbol_iterator
973            it = Asm.indirect_symbol_begin(),
974            ie = Asm.indirect_symbol_end(); it != ie; ++it) {
975       // Indirect symbols in the non-lazy symbol pointer section have some
976       // special handling.
977       const MCSectionMachO &Section =
978         static_cast<const MCSectionMachO&>(it->SectionData->getSection());
979       if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
980         // If this symbol is defined and internal, mark it as such.
981         if (it->Symbol->isDefined() &&
982             !Asm.getSymbolData(*it->Symbol).isExternal()) {
983           uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
984           if (it->Symbol->isAbsolute())
985             Flags |= MachO::INDIRECT_SYMBOL_ABS;
986           Write32(Flags);
987           continue;
988         }
989       }
990 
991       Write32(Asm.getSymbolData(*it->Symbol).getIndex());
992     }
993 
994     // FIXME: Check that offsets match computed ones.
995 
996     // Write the symbol table entries.
997     for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
998       WriteNlist(LocalSymbolData[i], Layout);
999     for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
1000       WriteNlist(ExternalSymbolData[i], Layout);
1001     for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
1002       WriteNlist(UndefinedSymbolData[i], Layout);
1003 
1004     // Write the string table.
1005     OS << StringTable.data();
1006   }
1007 }
1008 
createMachObjectWriter(MCMachObjectTargetWriter * MOTW,raw_pwrite_stream & OS,bool IsLittleEndian)1009 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
1010                                              raw_pwrite_stream &OS,
1011                                              bool IsLittleEndian) {
1012   return new MachObjectWriter(MOTW, OS, IsLittleEndian);
1013 }
1014