1 //===- InputFiles.cpp -----------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "InputFiles.h"
10 #include "Config.h"
11 #include "InputChunks.h"
12 #include "InputEvent.h"
13 #include "InputGlobal.h"
14 #include "OutputSegment.h"
15 #include "SymbolTable.h"
16 #include "lld/Common/ErrorHandler.h"
17 #include "lld/Common/Memory.h"
18 #include "lld/Common/Reproduce.h"
19 #include "llvm/Object/Binary.h"
20 #include "llvm/Object/Wasm.h"
21 #include "llvm/Support/TarWriter.h"
22 #include "llvm/Support/raw_ostream.h"
23 
24 #define DEBUG_TYPE "lld"
25 
26 using namespace llvm;
27 using namespace llvm::object;
28 using namespace llvm::wasm;
29 
30 namespace lld {
31 
32 // Returns a string in the format of "foo.o" or "foo.a(bar.o)".
33 std::string toString(const wasm::InputFile *file) {
34   if (!file)
35     return "<internal>";
36 
37   if (file->archiveName.empty())
38     return std::string(file->getName());
39 
40   return (file->archiveName + "(" + file->getName() + ")").str();
41 }
42 
43 namespace wasm {
44 
45 void InputFile::checkArch(Triple::ArchType arch) const {
46   bool is64 = arch == Triple::wasm64;
47   if (is64 && !config->is64.hasValue()) {
48     fatal(toString(this) +
49           ": must specify -mwasm64 to process wasm64 object files");
50   } else if (config->is64.getValueOr(false) != is64) {
51     fatal(toString(this) +
52           ": wasm32 object file can't be linked in wasm64 mode");
53   }
54 }
55 
56 std::unique_ptr<llvm::TarWriter> tar;
57 
58 Optional<MemoryBufferRef> readFile(StringRef path) {
59   log("Loading: " + path);
60 
61   auto mbOrErr = MemoryBuffer::getFile(path);
62   if (auto ec = mbOrErr.getError()) {
63     error("cannot open " + path + ": " + ec.message());
64     return None;
65   }
66   std::unique_ptr<MemoryBuffer> &mb = *mbOrErr;
67   MemoryBufferRef mbref = mb->getMemBufferRef();
68   make<std::unique_ptr<MemoryBuffer>>(std::move(mb)); // take MB ownership
69 
70   if (tar)
71     tar->append(relativeToRoot(path), mbref.getBuffer());
72   return mbref;
73 }
74 
75 InputFile *createObjectFile(MemoryBufferRef mb, StringRef archiveName) {
76   file_magic magic = identify_magic(mb.getBuffer());
77   if (magic == file_magic::wasm_object) {
78     std::unique_ptr<Binary> bin =
79         CHECK(createBinary(mb), mb.getBufferIdentifier());
80     auto *obj = cast<WasmObjectFile>(bin.get());
81     if (obj->isSharedObject())
82       return make<SharedFile>(mb);
83     return make<ObjFile>(mb, archiveName);
84   }
85 
86   if (magic == file_magic::bitcode)
87     return make<BitcodeFile>(mb, archiveName);
88 
89   fatal("unknown file type: " + mb.getBufferIdentifier());
90 }
91 
92 void ObjFile::dumpInfo() const {
93   log("info for: " + toString(this) +
94       "\n              Symbols : " + Twine(symbols.size()) +
95       "\n     Function Imports : " + Twine(wasmObj->getNumImportedFunctions()) +
96       "\n       Global Imports : " + Twine(wasmObj->getNumImportedGlobals()) +
97       "\n        Event Imports : " + Twine(wasmObj->getNumImportedEvents()));
98 }
99 
100 // Relocations contain either symbol or type indices.  This function takes a
101 // relocation and returns relocated index (i.e. translates from the input
102 // symbol/type space to the output symbol/type space).
103 uint32_t ObjFile::calcNewIndex(const WasmRelocation &reloc) const {
104   if (reloc.Type == R_WASM_TYPE_INDEX_LEB) {
105     assert(typeIsUsed[reloc.Index]);
106     return typeMap[reloc.Index];
107   }
108   const Symbol *sym = symbols[reloc.Index];
109   if (auto *ss = dyn_cast<SectionSymbol>(sym))
110     sym = ss->getOutputSectionSymbol();
111   return sym->getOutputSymbolIndex();
112 }
113 
114 // Relocations can contain addend for combined sections. This function takes a
115 // relocation and returns updated addend by offset in the output section.
116 uint64_t ObjFile::calcNewAddend(const WasmRelocation &reloc) const {
117   switch (reloc.Type) {
118   case R_WASM_MEMORY_ADDR_LEB:
119   case R_WASM_MEMORY_ADDR_LEB64:
120   case R_WASM_MEMORY_ADDR_SLEB64:
121   case R_WASM_MEMORY_ADDR_SLEB:
122   case R_WASM_MEMORY_ADDR_REL_SLEB:
123   case R_WASM_MEMORY_ADDR_REL_SLEB64:
124   case R_WASM_MEMORY_ADDR_I32:
125   case R_WASM_MEMORY_ADDR_I64:
126   case R_WASM_FUNCTION_OFFSET_I32:
127   case R_WASM_FUNCTION_OFFSET_I64:
128     return reloc.Addend;
129   case R_WASM_SECTION_OFFSET_I32:
130     return getSectionSymbol(reloc.Index)->section->outputOffset + reloc.Addend;
131   default:
132     llvm_unreachable("unexpected relocation type");
133   }
134 }
135 
136 // Calculate the value we expect to find at the relocation location.
137 // This is used as a sanity check before applying a relocation to a given
138 // location.  It is useful for catching bugs in the compiler and linker.
139 uint64_t ObjFile::calcExpectedValue(const WasmRelocation &reloc) const {
140   switch (reloc.Type) {
141   case R_WASM_TABLE_INDEX_I32:
142   case R_WASM_TABLE_INDEX_I64:
143   case R_WASM_TABLE_INDEX_SLEB:
144   case R_WASM_TABLE_INDEX_SLEB64: {
145     const WasmSymbol &sym = wasmObj->syms()[reloc.Index];
146     return tableEntries[sym.Info.ElementIndex];
147   }
148   case R_WASM_TABLE_INDEX_REL_SLEB: {
149     const WasmSymbol &sym = wasmObj->syms()[reloc.Index];
150     return tableEntriesRel[sym.Info.ElementIndex];
151   }
152   case R_WASM_MEMORY_ADDR_LEB:
153   case R_WASM_MEMORY_ADDR_LEB64:
154   case R_WASM_MEMORY_ADDR_SLEB:
155   case R_WASM_MEMORY_ADDR_SLEB64:
156   case R_WASM_MEMORY_ADDR_REL_SLEB:
157   case R_WASM_MEMORY_ADDR_REL_SLEB64:
158   case R_WASM_MEMORY_ADDR_I32:
159   case R_WASM_MEMORY_ADDR_I64:
160   case R_WASM_MEMORY_ADDR_TLS_SLEB: {
161     const WasmSymbol &sym = wasmObj->syms()[reloc.Index];
162     if (sym.isUndefined())
163       return 0;
164     const WasmSegment &segment =
165         wasmObj->dataSegments()[sym.Info.DataRef.Segment];
166     if (segment.Data.Offset.Opcode == WASM_OPCODE_I32_CONST)
167       return segment.Data.Offset.Value.Int32 + sym.Info.DataRef.Offset +
168              reloc.Addend;
169     else if (segment.Data.Offset.Opcode == WASM_OPCODE_I64_CONST)
170       return segment.Data.Offset.Value.Int64 + sym.Info.DataRef.Offset +
171              reloc.Addend;
172     else
173       llvm_unreachable("unknown init expr opcode");
174   }
175   case R_WASM_FUNCTION_OFFSET_I32:
176   case R_WASM_FUNCTION_OFFSET_I64: {
177     const WasmSymbol &sym = wasmObj->syms()[reloc.Index];
178     InputFunction *f =
179         functions[sym.Info.ElementIndex - wasmObj->getNumImportedFunctions()];
180     return f->getFunctionInputOffset() + f->getFunctionCodeOffset() +
181            reloc.Addend;
182   }
183   case R_WASM_SECTION_OFFSET_I32:
184     return reloc.Addend;
185   case R_WASM_TYPE_INDEX_LEB:
186     return reloc.Index;
187   case R_WASM_FUNCTION_INDEX_LEB:
188   case R_WASM_GLOBAL_INDEX_LEB:
189   case R_WASM_GLOBAL_INDEX_I32:
190   case R_WASM_EVENT_INDEX_LEB: {
191     const WasmSymbol &sym = wasmObj->syms()[reloc.Index];
192     return sym.Info.ElementIndex;
193   }
194   default:
195     llvm_unreachable("unknown relocation type");
196   }
197 }
198 
199 // Translate from the relocation's index into the final linked output value.
200 uint64_t ObjFile::calcNewValue(const WasmRelocation &reloc) const {
201   const Symbol* sym = nullptr;
202   if (reloc.Type != R_WASM_TYPE_INDEX_LEB) {
203     sym = symbols[reloc.Index];
204 
205     // We can end up with relocations against non-live symbols.  For example
206     // in debug sections. We return reloc.Addend because always returning zero
207     // causes the generation of spurious range-list terminators in the
208     // .debug_ranges section.
209     if ((isa<FunctionSymbol>(sym) || isa<DataSymbol>(sym)) && !sym->isLive())
210       return reloc.Addend;
211   }
212 
213   switch (reloc.Type) {
214   case R_WASM_TABLE_INDEX_I32:
215   case R_WASM_TABLE_INDEX_I64:
216   case R_WASM_TABLE_INDEX_SLEB:
217   case R_WASM_TABLE_INDEX_SLEB64:
218   case R_WASM_TABLE_INDEX_REL_SLEB: {
219     if (!getFunctionSymbol(reloc.Index)->hasTableIndex())
220       return 0;
221     uint32_t index = getFunctionSymbol(reloc.Index)->getTableIndex();
222     if (reloc.Type == R_WASM_TABLE_INDEX_REL_SLEB)
223       index -= config->tableBase;
224     return index;
225 
226   }
227   case R_WASM_MEMORY_ADDR_LEB:
228   case R_WASM_MEMORY_ADDR_LEB64:
229   case R_WASM_MEMORY_ADDR_SLEB:
230   case R_WASM_MEMORY_ADDR_SLEB64:
231   case R_WASM_MEMORY_ADDR_REL_SLEB:
232   case R_WASM_MEMORY_ADDR_REL_SLEB64:
233   case R_WASM_MEMORY_ADDR_I32:
234   case R_WASM_MEMORY_ADDR_I64: {
235     if (isa<UndefinedData>(sym) || sym->isUndefWeak())
236       return 0;
237     auto D = cast<DefinedData>(sym);
238     // Treat non-TLS relocation against symbols that live in the TLS segment
239     // like TLS relocations.  This beaviour exists to support older object
240     // files created before we introduced TLS relocations.
241     // TODO(sbc): Remove this legacy behaviour one day.  This will break
242     // backward compat with old object files built with `-fPIC`.
243     if (D->segment && D->segment->outputSeg->name == ".tdata")
244       return D->getOutputSegmentOffset() + reloc.Addend;
245     return D->getVirtualAddress() + reloc.Addend;
246   }
247   case R_WASM_MEMORY_ADDR_TLS_SLEB:
248     if (isa<UndefinedData>(sym) || sym->isUndefWeak())
249       return 0;
250     // TLS relocations are relative to the start of the TLS output segment
251     return cast<DefinedData>(sym)->getOutputSegmentOffset() + reloc.Addend;
252   case R_WASM_TYPE_INDEX_LEB:
253     return typeMap[reloc.Index];
254   case R_WASM_FUNCTION_INDEX_LEB:
255     return getFunctionSymbol(reloc.Index)->getFunctionIndex();
256   case R_WASM_GLOBAL_INDEX_LEB:
257   case R_WASM_GLOBAL_INDEX_I32:
258     if (auto gs = dyn_cast<GlobalSymbol>(sym))
259       return gs->getGlobalIndex();
260     return sym->getGOTIndex();
261   case R_WASM_EVENT_INDEX_LEB:
262     return getEventSymbol(reloc.Index)->getEventIndex();
263   case R_WASM_FUNCTION_OFFSET_I32:
264   case R_WASM_FUNCTION_OFFSET_I64: {
265     auto *f = cast<DefinedFunction>(sym);
266     return f->function->outputOffset +
267            (f->function->getFunctionCodeOffset() + reloc.Addend);
268   }
269   case R_WASM_SECTION_OFFSET_I32:
270     return getSectionSymbol(reloc.Index)->section->outputOffset + reloc.Addend;
271   default:
272     llvm_unreachable("unknown relocation type");
273   }
274 }
275 
276 template <class T>
277 static void setRelocs(const std::vector<T *> &chunks,
278                       const WasmSection *section) {
279   if (!section)
280     return;
281 
282   ArrayRef<WasmRelocation> relocs = section->Relocations;
283   assert(llvm::is_sorted(
284       relocs, [](const WasmRelocation &r1, const WasmRelocation &r2) {
285         return r1.Offset < r2.Offset;
286       }));
287   assert(llvm::is_sorted(chunks, [](InputChunk *c1, InputChunk *c2) {
288     return c1->getInputSectionOffset() < c2->getInputSectionOffset();
289   }));
290 
291   auto relocsNext = relocs.begin();
292   auto relocsEnd = relocs.end();
293   auto relocLess = [](const WasmRelocation &r, uint32_t val) {
294     return r.Offset < val;
295   };
296   for (InputChunk *c : chunks) {
297     auto relocsStart = std::lower_bound(relocsNext, relocsEnd,
298                                         c->getInputSectionOffset(), relocLess);
299     relocsNext = std::lower_bound(
300         relocsStart, relocsEnd, c->getInputSectionOffset() + c->getInputSize(),
301         relocLess);
302     c->setRelocations(ArrayRef<WasmRelocation>(relocsStart, relocsNext));
303   }
304 }
305 
306 void ObjFile::parse(bool ignoreComdats) {
307   // Parse a memory buffer as a wasm file.
308   LLVM_DEBUG(dbgs() << "Parsing object: " << toString(this) << "\n");
309   std::unique_ptr<Binary> bin = CHECK(createBinary(mb), toString(this));
310 
311   auto *obj = dyn_cast<WasmObjectFile>(bin.get());
312   if (!obj)
313     fatal(toString(this) + ": not a wasm file");
314   if (!obj->isRelocatableObject())
315     fatal(toString(this) + ": not a relocatable wasm file");
316 
317   bin.release();
318   wasmObj.reset(obj);
319 
320   checkArch(obj->getArch());
321 
322   // Build up a map of function indices to table indices for use when
323   // verifying the existing table index relocations
324   uint32_t totalFunctions =
325       wasmObj->getNumImportedFunctions() + wasmObj->functions().size();
326   tableEntriesRel.resize(totalFunctions);
327   tableEntries.resize(totalFunctions);
328   for (const WasmElemSegment &seg : wasmObj->elements()) {
329     int64_t offset;
330     if (seg.Offset.Opcode == WASM_OPCODE_I32_CONST)
331       offset = seg.Offset.Value.Int32;
332     else if (seg.Offset.Opcode == WASM_OPCODE_I64_CONST)
333       offset = seg.Offset.Value.Int64;
334     else
335       fatal(toString(this) + ": invalid table elements");
336     for (size_t index = 0; index < seg.Functions.size(); index++) {
337       auto functionIndex = seg.Functions[index];
338       tableEntriesRel[functionIndex] = index;
339       tableEntries[functionIndex] = offset + index;
340     }
341   }
342 
343   uint32_t sectionIndex = 0;
344 
345   // Bool for each symbol, true if called directly.  This allows us to implement
346   // a weaker form of signature checking where undefined functions that are not
347   // called directly (i.e. only address taken) don't have to match the defined
348   // function's signature.  We cannot do this for directly called functions
349   // because those signatures are checked at validation times.
350   // See https://bugs.llvm.org/show_bug.cgi?id=40412
351   std::vector<bool> isCalledDirectly(wasmObj->getNumberOfSymbols(), false);
352   for (const SectionRef &sec : wasmObj->sections()) {
353     const WasmSection &section = wasmObj->getWasmSection(sec);
354     // Wasm objects can have at most one code and one data section.
355     if (section.Type == WASM_SEC_CODE) {
356       assert(!codeSection);
357       codeSection = &section;
358     } else if (section.Type == WASM_SEC_DATA) {
359       assert(!dataSection);
360       dataSection = &section;
361     } else if (section.Type == WASM_SEC_CUSTOM) {
362       customSections.emplace_back(make<InputSection>(section, this));
363       customSections.back()->setRelocations(section.Relocations);
364       customSectionsByIndex[sectionIndex] = customSections.back();
365     }
366     sectionIndex++;
367     // Scans relocations to determine if a function symbol is called directly.
368     for (const WasmRelocation &reloc : section.Relocations)
369       if (reloc.Type == R_WASM_FUNCTION_INDEX_LEB)
370         isCalledDirectly[reloc.Index] = true;
371   }
372 
373   typeMap.resize(getWasmObj()->types().size());
374   typeIsUsed.resize(getWasmObj()->types().size(), false);
375 
376   ArrayRef<StringRef> comdats = wasmObj->linkingData().Comdats;
377   for (StringRef comdat : comdats) {
378     bool isNew = ignoreComdats || symtab->addComdat(comdat);
379     keptComdats.push_back(isNew);
380   }
381 
382   // Populate `Segments`.
383   for (const WasmSegment &s : wasmObj->dataSegments()) {
384     auto* seg = make<InputSegment>(s, this);
385     seg->discarded = isExcludedByComdat(seg);
386     segments.emplace_back(seg);
387   }
388   setRelocs(segments, dataSection);
389 
390   // Populate `Functions`.
391   ArrayRef<WasmFunction> funcs = wasmObj->functions();
392   ArrayRef<uint32_t> funcTypes = wasmObj->functionTypes();
393   ArrayRef<WasmSignature> types = wasmObj->types();
394   functions.reserve(funcs.size());
395 
396   for (size_t i = 0, e = funcs.size(); i != e; ++i) {
397     auto* func = make<InputFunction>(types[funcTypes[i]], &funcs[i], this);
398     func->discarded = isExcludedByComdat(func);
399     functions.emplace_back(func);
400   }
401   setRelocs(functions, codeSection);
402 
403   // Populate `Globals`.
404   for (const WasmGlobal &g : wasmObj->globals())
405     globals.emplace_back(make<InputGlobal>(g, this));
406 
407   // Populate `Events`.
408   for (const WasmEvent &e : wasmObj->events())
409     events.emplace_back(make<InputEvent>(types[e.Type.SigIndex], e, this));
410 
411   // Populate `Symbols` based on the symbols in the object.
412   symbols.reserve(wasmObj->getNumberOfSymbols());
413   for (const SymbolRef &sym : wasmObj->symbols()) {
414     const WasmSymbol &wasmSym = wasmObj->getWasmSymbol(sym.getRawDataRefImpl());
415     if (wasmSym.isDefined()) {
416       // createDefined may fail if the symbol is comdat excluded in which case
417       // we fall back to creating an undefined symbol
418       if (Symbol *d = createDefined(wasmSym)) {
419         symbols.push_back(d);
420         continue;
421       }
422     }
423     size_t idx = symbols.size();
424     symbols.push_back(createUndefined(wasmSym, isCalledDirectly[idx]));
425   }
426 }
427 
428 bool ObjFile::isExcludedByComdat(InputChunk *chunk) const {
429   uint32_t c = chunk->getComdat();
430   if (c == UINT32_MAX)
431     return false;
432   return !keptComdats[c];
433 }
434 
435 FunctionSymbol *ObjFile::getFunctionSymbol(uint32_t index) const {
436   return cast<FunctionSymbol>(symbols[index]);
437 }
438 
439 GlobalSymbol *ObjFile::getGlobalSymbol(uint32_t index) const {
440   return cast<GlobalSymbol>(symbols[index]);
441 }
442 
443 EventSymbol *ObjFile::getEventSymbol(uint32_t index) const {
444   return cast<EventSymbol>(symbols[index]);
445 }
446 
447 SectionSymbol *ObjFile::getSectionSymbol(uint32_t index) const {
448   return cast<SectionSymbol>(symbols[index]);
449 }
450 
451 DataSymbol *ObjFile::getDataSymbol(uint32_t index) const {
452   return cast<DataSymbol>(symbols[index]);
453 }
454 
455 Symbol *ObjFile::createDefined(const WasmSymbol &sym) {
456   StringRef name = sym.Info.Name;
457   uint32_t flags = sym.Info.Flags;
458 
459   switch (sym.Info.Kind) {
460   case WASM_SYMBOL_TYPE_FUNCTION: {
461     InputFunction *func =
462         functions[sym.Info.ElementIndex - wasmObj->getNumImportedFunctions()];
463     if (sym.isBindingLocal())
464       return make<DefinedFunction>(name, flags, this, func);
465     if (func->discarded)
466       return nullptr;
467     return symtab->addDefinedFunction(name, flags, this, func);
468   }
469   case WASM_SYMBOL_TYPE_DATA: {
470     InputSegment *seg = segments[sym.Info.DataRef.Segment];
471     auto offset = sym.Info.DataRef.Offset;
472     auto size = sym.Info.DataRef.Size;
473     if (sym.isBindingLocal())
474       return make<DefinedData>(name, flags, this, seg, offset, size);
475     if (seg->discarded)
476       return nullptr;
477     return symtab->addDefinedData(name, flags, this, seg, offset, size);
478   }
479   case WASM_SYMBOL_TYPE_GLOBAL: {
480     InputGlobal *global =
481         globals[sym.Info.ElementIndex - wasmObj->getNumImportedGlobals()];
482     if (sym.isBindingLocal())
483       return make<DefinedGlobal>(name, flags, this, global);
484     return symtab->addDefinedGlobal(name, flags, this, global);
485   }
486   case WASM_SYMBOL_TYPE_SECTION: {
487     InputSection *section = customSectionsByIndex[sym.Info.ElementIndex];
488     assert(sym.isBindingLocal());
489     return make<SectionSymbol>(flags, section, this);
490   }
491   case WASM_SYMBOL_TYPE_EVENT: {
492     InputEvent *event =
493         events[sym.Info.ElementIndex - wasmObj->getNumImportedEvents()];
494     if (sym.isBindingLocal())
495       return make<DefinedEvent>(name, flags, this, event);
496     return symtab->addDefinedEvent(name, flags, this, event);
497   }
498   }
499   llvm_unreachable("unknown symbol kind");
500 }
501 
502 Symbol *ObjFile::createUndefined(const WasmSymbol &sym, bool isCalledDirectly) {
503   StringRef name = sym.Info.Name;
504   uint32_t flags = sym.Info.Flags | WASM_SYMBOL_UNDEFINED;
505 
506   switch (sym.Info.Kind) {
507   case WASM_SYMBOL_TYPE_FUNCTION:
508     if (sym.isBindingLocal())
509       return make<UndefinedFunction>(name, sym.Info.ImportName,
510                                      sym.Info.ImportModule, flags, this,
511                                      sym.Signature, isCalledDirectly);
512     return symtab->addUndefinedFunction(name, sym.Info.ImportName,
513                                         sym.Info.ImportModule, flags, this,
514                                         sym.Signature, isCalledDirectly);
515   case WASM_SYMBOL_TYPE_DATA:
516     if (sym.isBindingLocal())
517       return make<UndefinedData>(name, flags, this);
518     return symtab->addUndefinedData(name, flags, this);
519   case WASM_SYMBOL_TYPE_GLOBAL:
520     if (sym.isBindingLocal())
521       return make<UndefinedGlobal>(name, sym.Info.ImportName,
522                                    sym.Info.ImportModule, flags, this,
523                                    sym.GlobalType);
524     return symtab->addUndefinedGlobal(name, sym.Info.ImportName,
525                                       sym.Info.ImportModule, flags, this,
526                                       sym.GlobalType);
527   case WASM_SYMBOL_TYPE_SECTION:
528     llvm_unreachable("section symbols cannot be undefined");
529   }
530   llvm_unreachable("unknown symbol kind");
531 }
532 
533 void ArchiveFile::parse() {
534   // Parse a MemoryBufferRef as an archive file.
535   LLVM_DEBUG(dbgs() << "Parsing library: " << toString(this) << "\n");
536   file = CHECK(Archive::create(mb), toString(this));
537 
538   // Read the symbol table to construct Lazy symbols.
539   int count = 0;
540   for (const Archive::Symbol &sym : file->symbols()) {
541     symtab->addLazy(this, &sym);
542     ++count;
543   }
544   LLVM_DEBUG(dbgs() << "Read " << count << " symbols\n");
545 }
546 
547 void ArchiveFile::addMember(const Archive::Symbol *sym) {
548   const Archive::Child &c =
549       CHECK(sym->getMember(),
550             "could not get the member for symbol " + sym->getName());
551 
552   // Don't try to load the same member twice (this can happen when members
553   // mutually reference each other).
554   if (!seen.insert(c.getChildOffset()).second)
555     return;
556 
557   LLVM_DEBUG(dbgs() << "loading lazy: " << sym->getName() << "\n");
558   LLVM_DEBUG(dbgs() << "from archive: " << toString(this) << "\n");
559 
560   MemoryBufferRef mb =
561       CHECK(c.getMemoryBufferRef(),
562             "could not get the buffer for the member defining symbol " +
563                 sym->getName());
564 
565   InputFile *obj = createObjectFile(mb, getName());
566   symtab->addFile(obj);
567 }
568 
569 static uint8_t mapVisibility(GlobalValue::VisibilityTypes gvVisibility) {
570   switch (gvVisibility) {
571   case GlobalValue::DefaultVisibility:
572     return WASM_SYMBOL_VISIBILITY_DEFAULT;
573   case GlobalValue::HiddenVisibility:
574   case GlobalValue::ProtectedVisibility:
575     return WASM_SYMBOL_VISIBILITY_HIDDEN;
576   }
577   llvm_unreachable("unknown visibility");
578 }
579 
580 static Symbol *createBitcodeSymbol(const std::vector<bool> &keptComdats,
581                                    const lto::InputFile::Symbol &objSym,
582                                    BitcodeFile &f) {
583   StringRef name = saver.save(objSym.getName());
584 
585   uint32_t flags = objSym.isWeak() ? WASM_SYMBOL_BINDING_WEAK : 0;
586   flags |= mapVisibility(objSym.getVisibility());
587 
588   int c = objSym.getComdatIndex();
589   bool excludedByComdat = c != -1 && !keptComdats[c];
590 
591   if (objSym.isUndefined() || excludedByComdat) {
592     flags |= WASM_SYMBOL_UNDEFINED;
593     if (objSym.isExecutable())
594       return symtab->addUndefinedFunction(name, None, None, flags, &f, nullptr,
595                                           true);
596     return symtab->addUndefinedData(name, flags, &f);
597   }
598 
599   if (objSym.isExecutable())
600     return symtab->addDefinedFunction(name, flags, &f, nullptr);
601   return symtab->addDefinedData(name, flags, &f, nullptr, 0, 0);
602 }
603 
604 bool BitcodeFile::doneLTO = false;
605 
606 void BitcodeFile::parse() {
607   if (doneLTO) {
608     error(toString(this) + ": attempt to add bitcode file after LTO.");
609     return;
610   }
611 
612   obj = check(lto::InputFile::create(MemoryBufferRef(
613       mb.getBuffer(), saver.save(archiveName + mb.getBufferIdentifier()))));
614   Triple t(obj->getTargetTriple());
615   if (!t.isWasm()) {
616     error(toString(this) + ": machine type must be wasm32 or wasm64");
617     return;
618   }
619   checkArch(t.getArch());
620   std::vector<bool> keptComdats;
621   for (StringRef s : obj->getComdatTable())
622     keptComdats.push_back(symtab->addComdat(s));
623 
624   for (const lto::InputFile::Symbol &objSym : obj->symbols())
625     symbols.push_back(createBitcodeSymbol(keptComdats, objSym, *this));
626 }
627 
628 } // namespace wasm
629 } // namespace lld
630