1//===- Symbols.h ------------------------------------------------*- C++ -*-===//
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#ifndef LLD_COFF_SYMBOLS_H
10#define LLD_COFF_SYMBOLS_H
11
12#include "Chunks.h"
13#include "Config.h"
14#include "lld/Common/LLVM.h"
15#include "lld/Common/Memory.h"
16#include "llvm/ADT/ArrayRef.h"
17#include "llvm/Object/Archive.h"
18#include "llvm/Object/COFF.h"
19#include <atomic>
20#include <memory>
21#include <vector>
22
23namespace lld {
24
25namespace coff {
26
27using llvm::object::Archive;
28using llvm::object::COFFSymbolRef;
29using llvm::object::coff_import_header;
30using llvm::object::coff_symbol_generic;
31
32class ArchiveFile;
33class COFFLinkerContext;
34class InputFile;
35class ObjFile;
36class Symbol;
37class SymbolTable;
38
39const COFFSyncStream &operator<<(const COFFSyncStream &,
40 const llvm::object::Archive::Symbol *);
41
42// The base class for real symbol classes.
43class Symbol {
44public:
45 enum Kind {
46 // The order of these is significant. We start with the regular defined
47 // symbols as those are the most prevalent and the zero tag is the cheapest
48 // to set. Among the defined kinds, the lower the kind is preferred over
49 // the higher kind when testing whether one symbol should take precedence
50 // over another.
51 DefinedRegularKind = 0,
52 DefinedCommonKind,
53 DefinedLocalImportKind,
54 DefinedImportThunkKind,
55 DefinedImportDataKind,
56 DefinedAbsoluteKind,
57 DefinedSyntheticKind,
58
59 UndefinedKind,
60 LazyArchiveKind,
61 LazyObjectKind,
62 LazyDLLSymbolKind,
63
64 LastDefinedCOFFKind = DefinedCommonKind,
65 LastDefinedKind = DefinedSyntheticKind,
66 };
67
68 Kind kind() const { return static_cast<Kind>(symbolKind); }
69
70 // Returns the symbol name.
71 StringRef getName() {
72 // COFF symbol names are read lazily for a performance reason.
73 // Non-external symbol names are never used by the linker except for logging
74 // or debugging. Their internal references are resolved not by name but by
75 // symbol index. And because they are not external, no one can refer them by
76 // name. Object files contain lots of non-external symbols, and creating
77 // StringRefs for them (which involves lots of strlen() on the string table)
78 // is a waste of time.
79 if (nameData == nullptr)
80 computeName();
81 return StringRef(nameData, nameSize);
82 }
83
84 void replaceKeepingName(Symbol *other, size_t size);
85
86 // Returns the file from which this symbol was created.
87 InputFile *getFile();
88
89 // Indicates that this symbol will be included in the final image. Only valid
90 // after calling markLive.
91 bool isLive() const;
92
93 bool isLazy() const {
94 return symbolKind == LazyArchiveKind || symbolKind == LazyObjectKind ||
95 symbolKind == LazyDLLSymbolKind;
96 }
97
98private:
99 void computeName();
100
101protected:
102 friend SymbolTable;
103 explicit Symbol(Kind k, StringRef n = "")
104 : symbolKind(k), isExternal(true), isCOMDAT(false),
105 writtenToSymtab(false), isUsedInRegularObj(false),
106 pendingArchiveLoad(false), isGCRoot(false), isRuntimePseudoReloc(false),
107 deferUndefined(false), canInline(true), isWeak(false), isAntiDep(false),
108 nameSize(n.size()), nameData(n.empty() ? nullptr : n.data()) {
109 assert((!n.empty() || k <= LastDefinedCOFFKind) &&
110 "If the name is empty, the Symbol must be a DefinedCOFF.");
111 }
112
113 unsigned symbolKind : 8;
114 unsigned isExternal : 1;
115
116public:
117 // This bit is used by the \c DefinedRegular subclass.
118 unsigned isCOMDAT : 1;
119
120 // This bit is used by Writer::createSymbolAndStringTable() to prevent
121 // symbols from being written to the symbol table more than once.
122 unsigned writtenToSymtab : 1;
123
124 // True if this symbol was referenced by a regular (non-bitcode) object.
125 unsigned isUsedInRegularObj : 1;
126
127 // True if we've seen both a lazy and an undefined symbol with this symbol
128 // name, which means that we have enqueued an archive member load and should
129 // not load any more archive members to resolve the same symbol.
130 unsigned pendingArchiveLoad : 1;
131
132 /// True if we've already added this symbol to the list of GC roots.
133 unsigned isGCRoot : 1;
134
135 unsigned isRuntimePseudoReloc : 1;
136
137 // True if we want to allow this symbol to be undefined in the early
138 // undefined check pass in SymbolTable::reportUnresolvable(), as it
139 // might be fixed up later.
140 unsigned deferUndefined : 1;
141
142 // False if LTO shouldn't inline whatever this symbol points to. If a symbol
143 // is overwritten after LTO, LTO shouldn't inline the symbol because it
144 // doesn't know the final contents of the symbol.
145 unsigned canInline : 1;
146
147 // True if the symbol is weak. This is only tracked for bitcode/LTO symbols.
148 // This information isn't written to the output; rather, it's used for
149 // managing weak symbol overrides.
150 unsigned isWeak : 1;
151
152 // True if the symbol is an anti-dependency.
153 unsigned isAntiDep : 1;
154
155protected:
156 // Symbol name length. Assume symbol lengths fit in a 32-bit integer.
157 uint32_t nameSize;
158
159 const char *nameData;
160};
161
162// The base class for any defined symbols, including absolute symbols,
163// etc.
164class Defined : public Symbol {
165public:
166 Defined(Kind k, StringRef n) : Symbol(k, n) {}
167
168 static bool classof(const Symbol *s) { return s->kind() <= LastDefinedKind; }
169
170 // Returns the RVA (relative virtual address) of this symbol. The
171 // writer sets and uses RVAs.
172 uint64_t getRVA();
173
174 // Returns the chunk containing this symbol. Absolute symbols and __ImageBase
175 // do not have chunks, so this may return null.
176 Chunk *getChunk();
177};
178
179// Symbols defined via a COFF object file or bitcode file. For COFF files, this
180// stores a coff_symbol_generic*, and names of internal symbols are lazily
181// loaded through that. For bitcode files, Sym is nullptr and the name is stored
182// as a decomposed StringRef.
183class DefinedCOFF : public Defined {
184 friend Symbol;
185
186public:
187 DefinedCOFF(Kind k, InputFile *f, StringRef n, const coff_symbol_generic *s)
188 : Defined(k, n), file(f), sym(s) {}
189
190 static bool classof(const Symbol *s) {
191 return s->kind() <= LastDefinedCOFFKind;
192 }
193
194 InputFile *getFile() { return file; }
195
196 COFFSymbolRef getCOFFSymbol();
197
198 InputFile *file;
199
200protected:
201 const coff_symbol_generic *sym;
202};
203
204// Regular defined symbols read from object file symbol tables.
205class DefinedRegular : public DefinedCOFF {
206public:
207 DefinedRegular(InputFile *f, StringRef n, bool isCOMDAT,
208 bool isExternal = false,
209 const coff_symbol_generic *s = nullptr,
210 SectionChunk *c = nullptr, bool isWeak = false)
211 : DefinedCOFF(DefinedRegularKind, f, n, s), data(c ? &c->repl : nullptr) {
212 this->isExternal = isExternal;
213 this->isCOMDAT = isCOMDAT;
214 this->isWeak = isWeak;
215 }
216
217 static bool classof(const Symbol *s) {
218 return s->kind() == DefinedRegularKind;
219 }
220
221 uint64_t getRVA() const { return (*data)->getRVA() + sym->Value; }
222 SectionChunk *getChunk() const { return *data; }
223 uint32_t getValue() const { return sym->Value; }
224
225 SectionChunk **data;
226};
227
228class DefinedCommon : public DefinedCOFF {
229public:
230 DefinedCommon(InputFile *f, StringRef n, uint64_t size,
231 const coff_symbol_generic *s = nullptr,
232 CommonChunk *c = nullptr)
233 : DefinedCOFF(DefinedCommonKind, f, n, s), data(c), size(size) {
234 this->isExternal = true;
235 }
236
237 static bool classof(const Symbol *s) {
238 return s->kind() == DefinedCommonKind;
239 }
240
241 uint64_t getRVA() { return data->getRVA(); }
242 CommonChunk *getChunk() { return data; }
243
244private:
245 friend SymbolTable;
246 uint64_t getSize() const { return size; }
247 CommonChunk *data;
248 uint64_t size;
249};
250
251// Absolute symbols.
252class DefinedAbsolute : public Defined {
253public:
254 DefinedAbsolute(const COFFLinkerContext &c, StringRef n, COFFSymbolRef s)
255 : Defined(DefinedAbsoluteKind, n), va(s.getValue()), ctx(c) {
256 isExternal = s.isExternal();
257 }
258
259 DefinedAbsolute(const COFFLinkerContext &c, StringRef n, uint64_t v)
260 : Defined(DefinedAbsoluteKind, n), va(v), ctx(c) {}
261
262 static bool classof(const Symbol *s) {
263 return s->kind() == DefinedAbsoluteKind;
264 }
265
266 uint64_t getRVA();
267 void setVA(uint64_t v) { va = v; }
268 uint64_t getVA() const { return va; }
269
270private:
271 uint64_t va;
272 const COFFLinkerContext &ctx;
273};
274
275// This symbol is used for linker-synthesized symbols like __ImageBase and
276// __safe_se_handler_table.
277class DefinedSynthetic : public Defined {
278public:
279 explicit DefinedSynthetic(StringRef name, Chunk *c, uint32_t offset = 0)
280 : Defined(DefinedSyntheticKind, name), c(c), offset(offset) {}
281
282 static bool classof(const Symbol *s) {
283 return s->kind() == DefinedSyntheticKind;
284 }
285
286 // A null chunk indicates that this is __ImageBase. Otherwise, this is some
287 // other synthesized chunk, like SEHTableChunk.
288 uint32_t getRVA() { return c ? c->getRVA() + offset : 0; }
289 Chunk *getChunk() { return c; }
290
291private:
292 Chunk *c;
293 uint32_t offset;
294};
295
296// This class represents a symbol defined in an archive file. It is
297// created from an archive file header, and it knows how to load an
298// object file from an archive to replace itself with a defined
299// symbol. If the resolver finds both Undefined and LazyArchive for
300// the same name, it will ask the LazyArchive to load a file.
301class LazyArchive : public Symbol {
302public:
303 LazyArchive(ArchiveFile *f, const Archive::Symbol s)
304 : Symbol(LazyArchiveKind, s.getName()), file(f), sym(s) {}
305
306 static bool classof(const Symbol *s) { return s->kind() == LazyArchiveKind; }
307
308 MemoryBufferRef getMemberBuffer();
309
310 ArchiveFile *file;
311 const Archive::Symbol sym;
312};
313
314class LazyObject : public Symbol {
315public:
316 LazyObject(InputFile *f, StringRef n) : Symbol(LazyObjectKind, n), file(f) {}
317 static bool classof(const Symbol *s) { return s->kind() == LazyObjectKind; }
318 InputFile *file;
319};
320
321// MinGW only.
322class LazyDLLSymbol : public Symbol {
323public:
324 LazyDLLSymbol(DLLFile *f, DLLFile::Symbol *s, StringRef n)
325 : Symbol(LazyDLLSymbolKind, n), file(f), sym(s) {}
326 static bool classof(const Symbol *s) {
327 return s->kind() == LazyDLLSymbolKind;
328 }
329
330 DLLFile *file;
331 DLLFile::Symbol *sym;
332};
333
334// Undefined symbols.
335class Undefined : public Symbol {
336public:
337 explicit Undefined(StringRef n) : Symbol(UndefinedKind, n) {}
338
339 static bool classof(const Symbol *s) { return s->kind() == UndefinedKind; }
340
341 // An undefined symbol can have a fallback symbol which gives an
342 // undefined symbol a second chance if it would remain undefined.
343 // If it remains undefined, it'll be replaced with whatever the
344 // Alias pointer points to.
345 Symbol *weakAlias = nullptr;
346
347 // If this symbol is external weak, try to resolve it to a defined
348 // symbol by searching the chain of fallback symbols. Returns the symbol if
349 // successful, otherwise returns null.
350 Symbol *getWeakAlias();
351 Defined *getDefinedWeakAlias() {
352 return dyn_cast_or_null<Defined>(Val: getWeakAlias());
353 }
354
355 void setWeakAlias(Symbol *sym, bool antiDep = false) {
356 weakAlias = sym;
357 isAntiDep = antiDep;
358 }
359
360 bool isECAlias(MachineTypes machine) const {
361 return weakAlias && isAntiDep && isArm64EC(Machine: machine);
362 }
363
364 // If this symbol is external weak, replace this object with aliased symbol.
365 bool resolveWeakAlias();
366};
367
368// Windows-specific classes.
369
370// This class represents a symbol imported from a DLL. This has two
371// names for internal use and external use. The former is used for
372// name resolution, and the latter is used for the import descriptor
373// table in an output. The former has "__imp_" prefix.
374class DefinedImportData : public Defined {
375public:
376 DefinedImportData(StringRef n, ImportFile *file, Chunk *&location)
377 : Defined(DefinedImportDataKind, n), file(file), location(location) {}
378
379 static bool classof(const Symbol *s) {
380 return s->kind() == DefinedImportDataKind;
381 }
382
383 uint64_t getRVA() { return getChunk()->getRVA(); }
384 Chunk *getChunk() { return location; }
385 void setLocation(Chunk *addressTable) { location = addressTable; }
386
387 StringRef getDLLName() { return file->dllName; }
388 StringRef getExternalName() { return file->externalName; }
389 uint16_t getOrdinal() { return file->hdr->OrdinalHint; }
390
391 ImportFile *file;
392 Chunk *&location;
393
394 // This is a pointer to the synthetic symbol associated with the load thunk
395 // for this symbol that will be called if the DLL is delay-loaded. This is
396 // needed for Control Flow Guard because if this DefinedImportData symbol is a
397 // valid call target, the corresponding load thunk must also be marked as a
398 // valid call target.
399 DefinedSynthetic *loadThunkSym = nullptr;
400};
401
402// This class represents a symbol for a jump table entry which jumps
403// to a function in a DLL. Linker are supposed to create such symbols
404// without "__imp_" prefix for all function symbols exported from
405// DLLs, so that you can call DLL functions as regular functions with
406// a regular name. A function pointer is given as a DefinedImportData.
407class DefinedImportThunk : public Defined {
408public:
409 DefinedImportThunk(COFFLinkerContext &ctx, StringRef name,
410 DefinedImportData *s, ImportThunkChunk *chunk);
411
412 static bool classof(const Symbol *s) {
413 return s->kind() == DefinedImportThunkKind;
414 }
415
416 uint64_t getRVA() { return data->getRVA(); }
417 ImportThunkChunk *getChunk() const { return data; }
418
419 DefinedImportData *wrappedSym;
420
421private:
422 ImportThunkChunk *data;
423};
424
425// If you have a symbol "foo" in your object file, a symbol name
426// "__imp_foo" becomes automatically available as a pointer to "foo".
427// This class is for such automatically-created symbols.
428// Yes, this is an odd feature. We didn't intend to implement that.
429// This is here just for compatibility with MSVC.
430class DefinedLocalImport : public Defined {
431public:
432 DefinedLocalImport(COFFLinkerContext &ctx, StringRef n, Defined *s)
433 : Defined(DefinedLocalImportKind, n),
434 data(make<LocalImportChunk>(args&: ctx, args&: s)) {}
435
436 static bool classof(const Symbol *s) {
437 return s->kind() == DefinedLocalImportKind;
438 }
439
440 uint64_t getRVA() { return data->getRVA(); }
441 Chunk *getChunk() { return data; }
442
443private:
444 LocalImportChunk *data;
445};
446
447inline uint64_t Defined::getRVA() {
448 switch (kind()) {
449 case DefinedAbsoluteKind:
450 return cast<DefinedAbsolute>(Val: this)->getRVA();
451 case DefinedSyntheticKind:
452 return cast<DefinedSynthetic>(Val: this)->getRVA();
453 case DefinedImportDataKind:
454 return cast<DefinedImportData>(Val: this)->getRVA();
455 case DefinedImportThunkKind:
456 return cast<DefinedImportThunk>(Val: this)->getRVA();
457 case DefinedLocalImportKind:
458 return cast<DefinedLocalImport>(Val: this)->getRVA();
459 case DefinedCommonKind:
460 return cast<DefinedCommon>(Val: this)->getRVA();
461 case DefinedRegularKind:
462 return cast<DefinedRegular>(Val: this)->getRVA();
463 case LazyArchiveKind:
464 case LazyObjectKind:
465 case LazyDLLSymbolKind:
466 case UndefinedKind:
467 llvm_unreachable("Cannot get the address for an undefined symbol.");
468 }
469 llvm_unreachable("unknown symbol kind");
470}
471
472inline Chunk *Defined::getChunk() {
473 switch (kind()) {
474 case DefinedRegularKind:
475 return cast<DefinedRegular>(Val: this)->getChunk();
476 case DefinedAbsoluteKind:
477 return nullptr;
478 case DefinedSyntheticKind:
479 return cast<DefinedSynthetic>(Val: this)->getChunk();
480 case DefinedImportDataKind:
481 return cast<DefinedImportData>(Val: this)->getChunk();
482 case DefinedImportThunkKind:
483 return cast<DefinedImportThunk>(Val: this)->getChunk();
484 case DefinedLocalImportKind:
485 return cast<DefinedLocalImport>(Val: this)->getChunk();
486 case DefinedCommonKind:
487 return cast<DefinedCommon>(Val: this)->getChunk();
488 case LazyArchiveKind:
489 case LazyObjectKind:
490 case LazyDLLSymbolKind:
491 case UndefinedKind:
492 llvm_unreachable("Cannot get the chunk of an undefined symbol.");
493 }
494 llvm_unreachable("unknown symbol kind");
495}
496
497// A buffer class that is large enough to hold any Symbol-derived
498// object. We allocate memory using this class and instantiate a symbol
499// using the placement new.
500union SymbolUnion {
501 alignas(DefinedRegular) char a[sizeof(DefinedRegular)];
502 alignas(DefinedCommon) char b[sizeof(DefinedCommon)];
503 alignas(DefinedAbsolute) char c[sizeof(DefinedAbsolute)];
504 alignas(DefinedSynthetic) char d[sizeof(DefinedSynthetic)];
505 alignas(LazyArchive) char e[sizeof(LazyArchive)];
506 alignas(Undefined) char f[sizeof(Undefined)];
507 alignas(DefinedImportData) char g[sizeof(DefinedImportData)];
508 alignas(DefinedImportThunk) char h[sizeof(DefinedImportThunk)];
509 alignas(DefinedLocalImport) char i[sizeof(DefinedLocalImport)];
510 alignas(LazyObject) char j[sizeof(LazyObject)];
511 alignas(LazyDLLSymbol) char k[sizeof(LazyDLLSymbol)];
512};
513
514template <typename T, typename... ArgT>
515void replaceSymbol(Symbol *s, ArgT &&... arg) {
516 static_assert(std::is_trivially_destructible<T>(),
517 "Symbol types must be trivially destructible");
518 static_assert(sizeof(T) <= sizeof(SymbolUnion), "Symbol too small");
519 static_assert(alignof(T) <= alignof(SymbolUnion),
520 "SymbolUnion not aligned enough");
521 assert(static_cast<Symbol *>(static_cast<T *>(nullptr)) == nullptr &&
522 "Not a Symbol");
523 bool canInline = s->canInline;
524 bool isUsedInRegularObj = s->isUsedInRegularObj;
525 new (s) T(std::forward<ArgT>(arg)...);
526 s->canInline = canInline;
527 s->isUsedInRegularObj = isUsedInRegularObj;
528}
529} // namespace coff
530
531std::string toString(const coff::COFFLinkerContext &ctx, coff::Symbol &b);
532std::string toCOFFString(const coff::COFFLinkerContext &ctx,
533 const llvm::object::Archive::Symbol &b);
534
535// Returns a symbol name for an error message.
536std::string maybeDemangleSymbol(const coff::COFFLinkerContext &ctx,
537 StringRef symName);
538
539} // namespace lld
540
541#endif
542