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?
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 bool EncodeAsAltEntry =
381 IsAlias && cast<MCSymbolMachO>(OrigSymbol).isAltEntry();
382 write16(cast<MCSymbolMachO>(Symbol)->getEncodedFlags(EncodeAsAltEntry));
383 if (is64Bit())
384 write64(Address);
385 else
386 write32(Address);
387 }
388
writeLinkeditLoadCommand(uint32_t Type,uint32_t DataOffset,uint32_t DataSize)389 void MachObjectWriter::writeLinkeditLoadCommand(uint32_t Type,
390 uint32_t DataOffset,
391 uint32_t DataSize) {
392 uint64_t Start = getStream().tell();
393 (void) Start;
394
395 write32(Type);
396 write32(sizeof(MachO::linkedit_data_command));
397 write32(DataOffset);
398 write32(DataSize);
399
400 assert(getStream().tell() - Start == sizeof(MachO::linkedit_data_command));
401 }
402
ComputeLinkerOptionsLoadCommandSize(const std::vector<std::string> & Options,bool is64Bit)403 static unsigned ComputeLinkerOptionsLoadCommandSize(
404 const std::vector<std::string> &Options, bool is64Bit)
405 {
406 unsigned Size = sizeof(MachO::linker_option_command);
407 for (const std::string &Option : Options)
408 Size += Option.size() + 1;
409 return alignTo(Size, is64Bit ? 8 : 4);
410 }
411
writeLinkerOptionsLoadCommand(const std::vector<std::string> & Options)412 void MachObjectWriter::writeLinkerOptionsLoadCommand(
413 const std::vector<std::string> &Options)
414 {
415 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
416 uint64_t Start = getStream().tell();
417 (void) Start;
418
419 write32(MachO::LC_LINKER_OPTION);
420 write32(Size);
421 write32(Options.size());
422 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
423 for (const std::string &Option : Options) {
424 // Write each string, including the null byte.
425 writeBytes(Option.c_str(), Option.size() + 1);
426 BytesWritten += Option.size() + 1;
427 }
428
429 // Pad to a multiple of the pointer size.
430 writeBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
431
432 assert(getStream().tell() - Start == Size);
433 }
434
recordRelocation(MCAssembler & Asm,const MCAsmLayout & Layout,const MCFragment * Fragment,const MCFixup & Fixup,MCValue Target,bool & IsPCRel,uint64_t & FixedValue)435 void MachObjectWriter::recordRelocation(MCAssembler &Asm,
436 const MCAsmLayout &Layout,
437 const MCFragment *Fragment,
438 const MCFixup &Fixup, MCValue Target,
439 bool &IsPCRel, uint64_t &FixedValue) {
440 TargetObjectWriter->recordRelocation(this, Asm, Layout, Fragment, Fixup,
441 Target, FixedValue);
442 }
443
bindIndirectSymbols(MCAssembler & Asm)444 void MachObjectWriter::bindIndirectSymbols(MCAssembler &Asm) {
445 // This is the point where 'as' creates actual symbols for indirect symbols
446 // (in the following two passes). It would be easier for us to do this sooner
447 // when we see the attribute, but that makes getting the order in the symbol
448 // table much more complicated than it is worth.
449 //
450 // FIXME: Revisit this when the dust settles.
451
452 // Report errors for use of .indirect_symbol not in a symbol pointer section
453 // or stub section.
454 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
455 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
456 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
457
458 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
459 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
460 Section.getType() != MachO::S_THREAD_LOCAL_VARIABLE_POINTERS &&
461 Section.getType() != MachO::S_SYMBOL_STUBS) {
462 MCSymbol &Symbol = *it->Symbol;
463 report_fatal_error("indirect symbol '" + Symbol.getName() +
464 "' not in a symbol pointer or stub section");
465 }
466 }
467
468 // Bind non-lazy symbol pointers first.
469 unsigned IndirectIndex = 0;
470 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
471 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
472 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
473
474 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
475 Section.getType() != MachO::S_THREAD_LOCAL_VARIABLE_POINTERS)
476 continue;
477
478 // Initialize the section indirect symbol base, if necessary.
479 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
480
481 Asm.registerSymbol(*it->Symbol);
482 }
483
484 // Then lazy symbol pointers and symbol stubs.
485 IndirectIndex = 0;
486 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
487 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
488 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
489
490 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
491 Section.getType() != MachO::S_SYMBOL_STUBS)
492 continue;
493
494 // Initialize the section indirect symbol base, if necessary.
495 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
496
497 // Set the symbol type to undefined lazy, but only on construction.
498 //
499 // FIXME: Do not hardcode.
500 bool Created;
501 Asm.registerSymbol(*it->Symbol, &Created);
502 if (Created)
503 cast<MCSymbolMachO>(it->Symbol)->setReferenceTypeUndefinedLazy(true);
504 }
505 }
506
507 /// computeSymbolTable - Compute the symbol table data
computeSymbolTable(MCAssembler & Asm,std::vector<MachSymbolData> & LocalSymbolData,std::vector<MachSymbolData> & ExternalSymbolData,std::vector<MachSymbolData> & UndefinedSymbolData)508 void MachObjectWriter::computeSymbolTable(
509 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
510 std::vector<MachSymbolData> &ExternalSymbolData,
511 std::vector<MachSymbolData> &UndefinedSymbolData) {
512 // Build section lookup table.
513 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
514 unsigned Index = 1;
515 for (MCAssembler::iterator it = Asm.begin(),
516 ie = Asm.end(); it != ie; ++it, ++Index)
517 SectionIndexMap[&*it] = Index;
518 assert(Index <= 256 && "Too many sections!");
519
520 // Build the string table.
521 for (const MCSymbol &Symbol : Asm.symbols()) {
522 if (!Asm.isSymbolLinkerVisible(Symbol))
523 continue;
524
525 StringTable.add(Symbol.getName());
526 }
527 StringTable.finalize();
528
529 // Build the symbol arrays but only for non-local symbols.
530 //
531 // The particular order that we collect and then sort the symbols is chosen to
532 // match 'as'. Even though it doesn't matter for correctness, this is
533 // important for letting us diff .o files.
534 for (const MCSymbol &Symbol : Asm.symbols()) {
535 // Ignore non-linker visible symbols.
536 if (!Asm.isSymbolLinkerVisible(Symbol))
537 continue;
538
539 if (!Symbol.isExternal() && !Symbol.isUndefined())
540 continue;
541
542 MachSymbolData MSD;
543 MSD.Symbol = &Symbol;
544 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
545
546 if (Symbol.isUndefined()) {
547 MSD.SectionIndex = 0;
548 UndefinedSymbolData.push_back(MSD);
549 } else if (Symbol.isAbsolute()) {
550 MSD.SectionIndex = 0;
551 ExternalSymbolData.push_back(MSD);
552 } else {
553 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
554 assert(MSD.SectionIndex && "Invalid section index!");
555 ExternalSymbolData.push_back(MSD);
556 }
557 }
558
559 // Now add the data for local symbols.
560 for (const MCSymbol &Symbol : Asm.symbols()) {
561 // Ignore non-linker visible symbols.
562 if (!Asm.isSymbolLinkerVisible(Symbol))
563 continue;
564
565 if (Symbol.isExternal() || Symbol.isUndefined())
566 continue;
567
568 MachSymbolData MSD;
569 MSD.Symbol = &Symbol;
570 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
571
572 if (Symbol.isAbsolute()) {
573 MSD.SectionIndex = 0;
574 LocalSymbolData.push_back(MSD);
575 } else {
576 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
577 assert(MSD.SectionIndex && "Invalid section index!");
578 LocalSymbolData.push_back(MSD);
579 }
580 }
581
582 // External and undefined symbols are required to be in lexicographic order.
583 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
584 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
585
586 // Set the symbol indices.
587 Index = 0;
588 for (auto *SymbolData :
589 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
590 for (MachSymbolData &Entry : *SymbolData)
591 Entry.Symbol->setIndex(Index++);
592
593 for (const MCSection &Section : Asm) {
594 for (RelAndSymbol &Rel : Relocations[&Section]) {
595 if (!Rel.Sym)
596 continue;
597
598 // Set the Index and the IsExtern bit.
599 unsigned Index = Rel.Sym->getIndex();
600 assert(isInt<24>(Index));
601 if (IsLittleEndian)
602 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
603 else
604 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
605 }
606 }
607 }
608
computeSectionAddresses(const MCAssembler & Asm,const MCAsmLayout & Layout)609 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
610 const MCAsmLayout &Layout) {
611 uint64_t StartAddress = 0;
612 for (const MCSection *Sec : Layout.getSectionOrder()) {
613 StartAddress = alignTo(StartAddress, Sec->getAlignment());
614 SectionAddress[Sec] = StartAddress;
615 StartAddress += Layout.getSectionAddressSize(Sec);
616
617 // Explicitly pad the section to match the alignment requirements of the
618 // following one. This is for 'gas' compatibility, it shouldn't
619 /// strictly be necessary.
620 StartAddress += getPaddingSize(Sec, Layout);
621 }
622 }
623
executePostLayoutBinding(MCAssembler & Asm,const MCAsmLayout & Layout)624 void MachObjectWriter::executePostLayoutBinding(MCAssembler &Asm,
625 const MCAsmLayout &Layout) {
626 computeSectionAddresses(Asm, Layout);
627
628 // Create symbol data for any indirect symbols.
629 bindIndirectSymbols(Asm);
630 }
631
isSymbolRefDifferenceFullyResolvedImpl(const MCAssembler & Asm,const MCSymbol & A,const MCSymbol & B,bool InSet) const632 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
633 const MCAssembler &Asm, const MCSymbol &A, const MCSymbol &B,
634 bool InSet) const {
635 // FIXME: We don't handle things like
636 // foo = .
637 // creating atoms.
638 if (A.isVariable() || B.isVariable())
639 return false;
640 return MCObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(Asm, A, B,
641 InSet);
642 }
643
isSymbolRefDifferenceFullyResolvedImpl(const MCAssembler & Asm,const MCSymbol & SymA,const MCFragment & FB,bool InSet,bool IsPCRel) const644 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
645 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
646 bool InSet, bool IsPCRel) const {
647 if (InSet)
648 return true;
649
650 // The effective address is
651 // addr(atom(A)) + offset(A)
652 // - addr(atom(B)) - offset(B)
653 // and the offsets are not relocatable, so the fixup is fully resolved when
654 // addr(atom(A)) - addr(atom(B)) == 0.
655 const MCSymbol &SA = findAliasedSymbol(SymA);
656 const MCSection &SecA = SA.getSection();
657 const MCSection &SecB = *FB.getParent();
658
659 if (IsPCRel) {
660 // The simple (Darwin, except on x86_64) way of dealing with this was to
661 // assume that any reference to a temporary symbol *must* be a temporary
662 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
663 // relocation to a temporary symbol (in the same section) is fully
664 // resolved. This also works in conjunction with absolutized .set, which
665 // requires the compiler to use .set to absolutize the differences between
666 // symbols which the compiler knows to be assembly time constants, so we
667 // don't need to worry about considering symbol differences fully resolved.
668 //
669 // If the file isn't using sub-sections-via-symbols, we can make the
670 // same assumptions about any symbol that we normally make about
671 // assembler locals.
672
673 bool hasReliableSymbolDifference = isX86_64();
674 if (!hasReliableSymbolDifference) {
675 if (!SA.isInSection() || &SecA != &SecB ||
676 (!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() &&
677 Asm.getSubsectionsViaSymbols()))
678 return false;
679 return true;
680 }
681 // For Darwin x86_64, there is one special case when the reference IsPCRel.
682 // If the fragment with the reference does not have a base symbol but meets
683 // the simple way of dealing with this, in that it is a temporary symbol in
684 // the same atom then it is assumed to be fully resolved. This is needed so
685 // a relocation entry is not created and so the static linker does not
686 // mess up the reference later.
687 else if(!FB.getAtom() &&
688 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
689 return true;
690 }
691 }
692
693 // If they are not in the same section, we can't compute the diff.
694 if (&SecA != &SecB)
695 return false;
696
697 const MCFragment *FA = SA.getFragment();
698
699 // Bail if the symbol has no fragment.
700 if (!FA)
701 return false;
702
703 // If the atoms are the same, they are guaranteed to have the same address.
704 if (FA->getAtom() == FB.getAtom())
705 return true;
706
707 // Otherwise, we can't prove this is fully resolved.
708 return false;
709 }
710
writeObject(MCAssembler & Asm,const MCAsmLayout & Layout)711 void MachObjectWriter::writeObject(MCAssembler &Asm,
712 const MCAsmLayout &Layout) {
713 // Compute symbol table information and bind symbol indices.
714 computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
715 UndefinedSymbolData);
716
717 unsigned NumSections = Asm.size();
718 const MCAssembler::VersionMinInfoType &VersionInfo =
719 Layout.getAssembler().getVersionMinInfo();
720
721 // The section data starts after the header, the segment load command (and
722 // section headers) and the symbol table.
723 unsigned NumLoadCommands = 1;
724 uint64_t LoadCommandsSize = is64Bit() ?
725 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
726 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
727
728 // Add the deployment target version info load command size, if used.
729 if (VersionInfo.Major != 0) {
730 ++NumLoadCommands;
731 LoadCommandsSize += sizeof(MachO::version_min_command);
732 }
733
734 // Add the data-in-code load command size, if used.
735 unsigned NumDataRegions = Asm.getDataRegions().size();
736 if (NumDataRegions) {
737 ++NumLoadCommands;
738 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
739 }
740
741 // Add the loh load command size, if used.
742 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
743 uint64_t LOHSize = alignTo(LOHRawSize, is64Bit() ? 8 : 4);
744 if (LOHSize) {
745 ++NumLoadCommands;
746 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
747 }
748
749 // Add the symbol table load command sizes, if used.
750 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
751 UndefinedSymbolData.size();
752 if (NumSymbols) {
753 NumLoadCommands += 2;
754 LoadCommandsSize += (sizeof(MachO::symtab_command) +
755 sizeof(MachO::dysymtab_command));
756 }
757
758 // Add the linker option load commands sizes.
759 for (const auto &Option : Asm.getLinkerOptions()) {
760 ++NumLoadCommands;
761 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit());
762 }
763
764 // Compute the total size of the section data, as well as its file size and vm
765 // size.
766 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
767 sizeof(MachO::mach_header)) + LoadCommandsSize;
768 uint64_t SectionDataSize = 0;
769 uint64_t SectionDataFileSize = 0;
770 uint64_t VMSize = 0;
771 for (const MCSection &Sec : Asm) {
772 uint64_t Address = getSectionAddress(&Sec);
773 uint64_t Size = Layout.getSectionAddressSize(&Sec);
774 uint64_t FileSize = Layout.getSectionFileSize(&Sec);
775 FileSize += getPaddingSize(&Sec, Layout);
776
777 VMSize = std::max(VMSize, Address + Size);
778
779 if (Sec.isVirtualSection())
780 continue;
781
782 SectionDataSize = std::max(SectionDataSize, Address + Size);
783 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
784 }
785
786 // The section data is padded to 4 bytes.
787 //
788 // FIXME: Is this machine dependent?
789 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
790 SectionDataFileSize += SectionDataPadding;
791
792 // Write the prolog, starting with the header and load command...
793 writeHeader(MachO::MH_OBJECT, NumLoadCommands, LoadCommandsSize,
794 Asm.getSubsectionsViaSymbols());
795 uint32_t Prot =
796 MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE;
797 writeSegmentLoadCommand("", NumSections, 0, VMSize, SectionDataStart,
798 SectionDataSize, Prot, Prot);
799
800 // ... and then the section headers.
801 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
802 for (const MCSection &Section : Asm) {
803 const auto &Sec = cast<MCSectionMachO>(Section);
804 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
805 unsigned NumRelocs = Relocs.size();
806 uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec);
807 unsigned Flags = Sec.getTypeAndAttributes();
808 if (Sec.hasInstructions())
809 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
810 writeSection(Layout, Sec, getSectionAddress(&Sec), SectionStart, Flags,
811 RelocTableEnd, NumRelocs);
812 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
813 }
814
815 // Write out the deployment target information, if it's available.
816 if (VersionInfo.Major != 0) {
817 assert(VersionInfo.Update < 256 && "unencodable update target version");
818 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
819 assert(VersionInfo.Major < 65536 && "unencodable major target version");
820 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
821 (VersionInfo.Major << 16);
822 MachO::LoadCommandType LCType;
823 switch (VersionInfo.Kind) {
824 case MCVM_OSXVersionMin:
825 LCType = MachO::LC_VERSION_MIN_MACOSX;
826 break;
827 case MCVM_IOSVersionMin:
828 LCType = MachO::LC_VERSION_MIN_IPHONEOS;
829 break;
830 case MCVM_TvOSVersionMin:
831 LCType = MachO::LC_VERSION_MIN_TVOS;
832 break;
833 case MCVM_WatchOSVersionMin:
834 LCType = MachO::LC_VERSION_MIN_WATCHOS;
835 break;
836 }
837 write32(LCType);
838 write32(sizeof(MachO::version_min_command));
839 write32(EncodedVersion);
840 write32(0); // reserved.
841 }
842
843 // Write the data-in-code load command, if used.
844 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
845 if (NumDataRegions) {
846 uint64_t DataRegionsOffset = RelocTableEnd;
847 uint64_t DataRegionsSize = NumDataRegions * 8;
848 writeLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
849 DataRegionsSize);
850 }
851
852 // Write the loh load command, if used.
853 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
854 if (LOHSize)
855 writeLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
856 DataInCodeTableEnd, LOHSize);
857
858 // Write the symbol table load command, if used.
859 if (NumSymbols) {
860 unsigned FirstLocalSymbol = 0;
861 unsigned NumLocalSymbols = LocalSymbolData.size();
862 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
863 unsigned NumExternalSymbols = ExternalSymbolData.size();
864 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
865 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
866 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
867 unsigned NumSymTabSymbols =
868 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
869 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
870 uint64_t IndirectSymbolOffset = 0;
871
872 // If used, the indirect symbols are written after the section data.
873 if (NumIndirectSymbols)
874 IndirectSymbolOffset = LOHTableEnd;
875
876 // The symbol table is written after the indirect symbol data.
877 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
878
879 // The string table is written after symbol table.
880 uint64_t StringTableOffset =
881 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
882 sizeof(MachO::nlist_64) :
883 sizeof(MachO::nlist));
884 writeSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
885 StringTableOffset, StringTable.data().size());
886
887 writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
888 FirstExternalSymbol, NumExternalSymbols,
889 FirstUndefinedSymbol, NumUndefinedSymbols,
890 IndirectSymbolOffset, NumIndirectSymbols);
891 }
892
893 // Write the linker options load commands.
894 for (const auto &Option : Asm.getLinkerOptions())
895 writeLinkerOptionsLoadCommand(Option);
896
897 // Write the actual section data.
898 for (const MCSection &Sec : Asm) {
899 Asm.writeSectionData(&Sec, Layout);
900
901 uint64_t Pad = getPaddingSize(&Sec, Layout);
902 WriteZeros(Pad);
903 }
904
905 // Write the extra padding.
906 WriteZeros(SectionDataPadding);
907
908 // Write the relocation entries.
909 for (const MCSection &Sec : Asm) {
910 // Write the section relocation entries, in reverse order to match 'as'
911 // (approximately, the exact algorithm is more complicated than this).
912 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
913 for (const RelAndSymbol &Rel : make_range(Relocs.rbegin(), Relocs.rend())) {
914 write32(Rel.MRE.r_word0);
915 write32(Rel.MRE.r_word1);
916 }
917 }
918
919 // Write out the data-in-code region payload, if there is one.
920 for (MCAssembler::const_data_region_iterator
921 it = Asm.data_region_begin(), ie = Asm.data_region_end();
922 it != ie; ++it) {
923 const DataRegionData *Data = &(*it);
924 uint64_t Start = getSymbolAddress(*Data->Start, Layout);
925 uint64_t End = getSymbolAddress(*Data->End, Layout);
926 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
927 << " start: " << Start << "(" << Data->Start->getName() << ")"
928 << " end: " << End << "(" << Data->End->getName() << ")"
929 << " size: " << End - Start
930 << "\n");
931 write32(Start);
932 write16(End - Start);
933 write16(Data->Kind);
934 }
935
936 // Write out the loh commands, if there is one.
937 if (LOHSize) {
938 #ifndef NDEBUG
939 unsigned Start = getStream().tell();
940 #endif
941 Asm.getLOHContainer().emit(*this, Layout);
942 // Pad to a multiple of the pointer size.
943 writeBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
944 assert(getStream().tell() - Start == LOHSize);
945 }
946
947 // Write the symbol table data, if used.
948 if (NumSymbols) {
949 // Write the indirect symbol entries.
950 for (MCAssembler::const_indirect_symbol_iterator
951 it = Asm.indirect_symbol_begin(),
952 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
953 // Indirect symbols in the non-lazy symbol pointer section have some
954 // special handling.
955 const MCSectionMachO &Section =
956 static_cast<const MCSectionMachO &>(*it->Section);
957 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
958 // If this symbol is defined and internal, mark it as such.
959 if (it->Symbol->isDefined() && !it->Symbol->isExternal()) {
960 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
961 if (it->Symbol->isAbsolute())
962 Flags |= MachO::INDIRECT_SYMBOL_ABS;
963 write32(Flags);
964 continue;
965 }
966 }
967
968 write32(it->Symbol->getIndex());
969 }
970
971 // FIXME: Check that offsets match computed ones.
972
973 // Write the symbol table entries.
974 for (auto *SymbolData :
975 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
976 for (MachSymbolData &Entry : *SymbolData)
977 writeNlist(Entry, Layout);
978
979 // Write the string table.
980 getStream() << StringTable.data();
981 }
982 }
983
createMachObjectWriter(MCMachObjectTargetWriter * MOTW,raw_pwrite_stream & OS,bool IsLittleEndian)984 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
985 raw_pwrite_stream &OS,
986 bool IsLittleEndian) {
987 return new MachObjectWriter(MOTW, OS, IsLittleEndian);
988 }
989