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// This file defines various types of Symbols.
10//
11//===----------------------------------------------------------------------===//
12
13#ifndef LLD_ELF_SYMBOLS_H
14#define LLD_ELF_SYMBOLS_H
15
16#include "Config.h"
17#include "lld/Common/LLVM.h"
18#include "lld/Common/Memory.h"
19#include "llvm/ADT/DenseMap.h"
20#include "llvm/Object/ELF.h"
21#include "llvm/Support/Compiler.h"
22#include <tuple>
23
24namespace lld::elf {
25class CommonSymbol;
26class Defined;
27class OutputSection;
28class SectionBase;
29class InputSectionBase;
30class SharedSymbol;
31class Symbol;
32class Undefined;
33class LazySymbol;
34class InputFile;
35
36// Returns a string representation for a symbol for diagnostics.
37std::string toStr(Ctx &, const Symbol &);
38const ELFSyncStream &operator<<(const ELFSyncStream &, const Symbol *);
39
40void printTraceSymbol(const Symbol &sym, StringRef name);
41
42enum {
43 NEEDS_GOT = 1 << 0,
44 NEEDS_PLT = 1 << 1,
45 HAS_DIRECT_RELOC = 1 << 2,
46 // True if this symbol needs a canonical PLT entry, or (during
47 // postScanRelocations) a copy relocation.
48 NEEDS_COPY = 1 << 3,
49 NEEDS_TLSDESC = 1 << 4,
50 NEEDS_TLSGD = 1 << 5,
51 // 1 << 6 unused
52 NEEDS_GOT_DTPREL = 1 << 7,
53 NEEDS_TLSIE = 1 << 8,
54 NEEDS_GOT_AUTH = 1 << 9,
55 NEEDS_GOT_NONAUTH = 1 << 10,
56 NEEDS_TLSDESC_AUTH = 1 << 11,
57 NEEDS_TLSDESC_NONAUTH = 1 << 12,
58};
59
60// The base class for real symbol classes.
61class Symbol {
62public:
63 enum Kind {
64 PlaceholderKind,
65 DefinedKind,
66 CommonKind,
67 SharedKind,
68 UndefinedKind,
69 LazyKind,
70 };
71
72 Kind kind() const { return static_cast<Kind>(symbolKind); }
73
74 // The file from which this symbol was created.
75 InputFile *file;
76
77 // The default copy constructor is deleted due to atomic flags. Define one for
78 // places where no atomic is needed.
79 Symbol(const Symbol &o) { memcpy(dest: static_cast<void *>(this), src: &o, n: sizeof(o)); }
80
81protected:
82 const char *nameData;
83 // 32-bit size saves space.
84 uint32_t nameSize;
85
86public:
87 // The next three fields have the same meaning as the ELF symbol attributes.
88 // type and binding are placed in this order to optimize generating st_info,
89 // which is defined as (binding << 4) + (type & 0xf), on a little-endian
90 // system.
91 uint8_t type : 4; // symbol type
92
93 // Symbol binding. This is not overwritten by replace() to track
94 // changes during resolution. In particular:
95 // - An undefined weak is still weak when it resolves to a shared library.
96 // - An undefined weak will not extract archive members, but we have to
97 // remember it is weak.
98 uint8_t binding : 4;
99
100 uint8_t stOther; // st_other field value
101
102 uint8_t symbolKind;
103
104 // The partition whose dynamic symbol table contains this symbol's definition.
105 uint8_t partition;
106
107 // True if this symbol is preemptible at load time.
108 //
109 // Primarily set in two locations, (a) parseVersionAndComputeIsPreemptible and
110 // (b) demoteSymbolsAndComputeIsPreemptible.
111 LLVM_PREFERRED_TYPE(bool)
112 uint8_t isPreemptible : 1;
113
114 // True if the symbol was used for linking and thus need to be added to the
115 // output file's symbol table. This is true for all symbols except for
116 // unreferenced DSO symbols, lazy (archive) symbols, and bitcode symbols that
117 // are unreferenced except by other bitcode objects.
118 LLVM_PREFERRED_TYPE(bool)
119 uint8_t isUsedInRegularObj : 1;
120
121 // True if an undefined or shared symbol is used from a live section.
122 //
123 // NOTE: In Writer.cpp the field is used to mark local defined symbols
124 // which are referenced by relocations when -r or --emit-relocs is given.
125 LLVM_PREFERRED_TYPE(bool)
126 uint8_t used : 1;
127
128 // Used by a Defined symbol with protected or default visibility, to record
129 // whether it is required to be exported into .dynsym. This is set when any of
130 // the following conditions hold:
131 //
132 // - If there is an interposable symbol from a DSO. Note: We also do this for
133 // STV_PROTECTED symbols which can't be interposed (to match BFD behavior).
134 // - If -shared or --export-dynamic is specified, any symbol in an object
135 // file/bitcode sets this property, unless suppressed by LTO
136 // canBeOmittedFromSymbolTable().
137 LLVM_PREFERRED_TYPE(bool)
138 uint8_t isExported : 1;
139
140 LLVM_PREFERRED_TYPE(bool)
141 uint8_t ltoCanOmit : 1;
142
143 // True if this symbol is specified by --trace-symbol option.
144 LLVM_PREFERRED_TYPE(bool)
145 uint8_t traced : 1;
146
147 // True if the name contains '@'.
148 LLVM_PREFERRED_TYPE(bool)
149 uint8_t hasVersionSuffix : 1;
150
151 // Symbol visibility. This is the computed minimum visibility of all
152 // observed non-DSO symbols.
153 uint8_t visibility() const { return stOther & 3; }
154 void setVisibility(uint8_t visibility) {
155 stOther = (stOther & ~3) | visibility;
156 }
157
158 uint8_t computeBinding(Ctx &) const;
159 bool isGlobal() const { return binding == llvm::ELF::STB_GLOBAL; }
160 bool isWeak() const { return binding == llvm::ELF::STB_WEAK; }
161
162 bool isUndefined() const { return symbolKind == UndefinedKind; }
163 bool isCommon() const { return symbolKind == CommonKind; }
164 bool isDefined() const { return symbolKind == DefinedKind; }
165 bool isShared() const { return symbolKind == SharedKind; }
166 bool isPlaceholder() const { return symbolKind == PlaceholderKind; }
167
168 bool isLocal() const { return binding == llvm::ELF::STB_LOCAL; }
169
170 bool isLazy() const { return symbolKind == LazyKind; }
171
172 // True if this is an undefined weak symbol. This only works once
173 // all input files have been added.
174 bool isUndefWeak() const { return isWeak() && isUndefined(); }
175
176 StringRef getName() const { return {nameData, nameSize}; }
177
178 void setName(StringRef s) {
179 nameData = s.data();
180 nameSize = s.size();
181 }
182
183 void parseSymbolVersion(Ctx &);
184
185 // Get the NUL-terminated version suffix ("", "@...", or "@@...").
186 //
187 // For @@, the name has been truncated by insert(). For @, the name has been
188 // truncated by Symbol::parseSymbolVersion(ctx).
189 const char *getVersionSuffix() const { return nameData + nameSize; }
190
191 uint32_t getGotIdx(Ctx &ctx) const { return ctx.symAux[auxIdx].gotIdx; }
192 uint32_t getPltIdx(Ctx &ctx) const { return ctx.symAux[auxIdx].pltIdx; }
193 uint32_t getTlsDescIdx(Ctx &ctx) const {
194 return ctx.symAux[auxIdx].tlsDescIdx;
195 }
196 uint32_t getTlsGdIdx(Ctx &ctx) const { return ctx.symAux[auxIdx].tlsGdIdx; }
197
198 bool isInGot(Ctx &ctx) const { return getGotIdx(ctx) != uint32_t(-1); }
199 bool isInPlt(Ctx &ctx) const { return getPltIdx(ctx) != uint32_t(-1); }
200
201 uint64_t getVA(Ctx &, int64_t addend = 0) const;
202
203 uint64_t getGotOffset(Ctx &) const;
204 uint64_t getGotVA(Ctx &) const;
205 uint64_t getGotPltOffset(Ctx &) const;
206 uint64_t getGotPltVA(Ctx &) const;
207 uint64_t getPltVA(Ctx &) const;
208 uint64_t getSize() const;
209 OutputSection *getOutputSection() const;
210
211 // The following two functions are used for symbol resolution.
212 //
213 // You are expected to call mergeProperties for all symbols in input
214 // files so that attributes that are attached to names rather than
215 // indivisual symbol (such as visibility) are merged together.
216 //
217 // Every time you read a new symbol from an input, you are supposed
218 // to call resolve() with the new symbol. That function replaces
219 // "this" object as a result of name resolution if the new symbol is
220 // more appropriate to be included in the output.
221 //
222 // For example, if "this" is an undefined symbol and a new symbol is
223 // a defined symbol, "this" is replaced with the new symbol.
224 void mergeProperties(const Symbol &other);
225 void resolve(Ctx &, const Undefined &other);
226 void resolve(Ctx &, const CommonSymbol &other);
227 void resolve(Ctx &, const Defined &other);
228 void resolve(Ctx &, const LazySymbol &other);
229 void resolve(Ctx &, const SharedSymbol &other);
230
231 // If this is a lazy symbol, extract an input file and add the symbol
232 // in the file to the symbol table. Calling this function on
233 // non-lazy object causes a runtime error.
234 void extract(Ctx &) const;
235
236 void checkDuplicate(Ctx &, const Defined &other) const;
237
238private:
239 bool shouldReplace(Ctx &, const Defined &other) const;
240
241protected:
242 Symbol(Kind k, InputFile *file, StringRef name, uint8_t binding,
243 uint8_t stOther, uint8_t type)
244 : file(file), nameData(name.data()), nameSize(name.size()), type(type),
245 binding(binding), stOther(stOther), symbolKind(k), ltoCanOmit(false),
246 archSpecificBit(false) {}
247
248 void overwrite(Symbol &sym, Kind k) const {
249 if (sym.traced)
250 printTraceSymbol(sym: *this, name: sym.getName());
251 sym.file = file;
252 sym.type = type;
253 sym.binding = binding;
254 sym.stOther = (stOther & ~3) | sym.visibility();
255 sym.symbolKind = k;
256 }
257
258public:
259 // True if this symbol is in the Iplt sub-section of the Plt and the Igot
260 // sub-section of the .got.plt or .got.
261 LLVM_PREFERRED_TYPE(bool)
262 uint8_t isInIplt : 1;
263
264 // True if this symbol needs a GOT entry and its GOT entry is actually in
265 // Igot. This will be true only for certain non-preemptible ifuncs.
266 LLVM_PREFERRED_TYPE(bool)
267 uint8_t gotInIgot : 1;
268
269 // True if defined relative to a section discarded by ICF.
270 LLVM_PREFERRED_TYPE(bool)
271 uint8_t folded : 1;
272
273 // Allow reuse of a bit between architecture-exclusive symbol flags.
274 // - needsTocRestore(): On PPC64, true if a call to this symbol needs to be
275 // followed by a restore of the toc pointer.
276 // - isTagged(): On AArch64, true if the symbol needs special relocation and
277 // metadata semantics because it's tagged, under the AArch64 MemtagABI.
278 LLVM_PREFERRED_TYPE(bool)
279 uint8_t archSpecificBit : 1;
280 bool needsTocRestore() const { return archSpecificBit; }
281 bool isTagged() const { return archSpecificBit; }
282 void setNeedsTocRestore(bool v) { archSpecificBit = v; }
283 void setIsTagged(bool v) {
284 archSpecificBit = v;
285 }
286
287 // True if this symbol is defined by a symbol assignment or wrapped by --wrap.
288 //
289 // LTO shouldn't inline the symbol because it doesn't know the final content
290 // of the symbol.
291 LLVM_PREFERRED_TYPE(bool)
292 uint8_t scriptDefined : 1;
293
294 // True if defined in a DSO. There may also be a definition in a relocatable
295 // object file.
296 LLVM_PREFERRED_TYPE(bool)
297 uint8_t dsoDefined : 1;
298
299 // True if defined in a DSO as protected visibility.
300 LLVM_PREFERRED_TYPE(bool)
301 uint8_t dsoProtected : 1;
302
303 // Temporary flags used to communicate which symbol entries need PLT and GOT
304 // entries during postScanRelocations();
305 std::atomic<uint16_t> flags;
306
307 // A ctx.symAux index used to access GOT/PLT entry indexes. This is allocated
308 // in postScanRelocations().
309 uint32_t auxIdx;
310 uint32_t dynsymIndex;
311
312 // If `file` is SharedFile (for SharedSymbol or copy-relocated Defined), this
313 // represents the Verdef index within the input DSO, which will be converted
314 // to a Verneed index in the output. Otherwise, this represents the Verdef
315 // index (VER_NDX_LOCAL, VER_NDX_GLOBAL, or a named version).
316 // VER_NDX_LOCAL indicates a defined symbol that has been localized by a
317 // version script's local: directive or --exclude-libs.
318 uint16_t versionId;
319 LLVM_PREFERRED_TYPE(bool)
320 uint8_t versionScriptAssigned : 1;
321
322 // True if targeted by a range extension thunk.
323 LLVM_PREFERRED_TYPE(bool)
324 uint8_t thunkAccessed : 1;
325
326 // True if the symbol is in the --dynamic-list file. A Defined symbol with
327 // protected or default visibility with this property is required to be
328 // exported into .dynsym.
329 LLVM_PREFERRED_TYPE(bool)
330 uint8_t inDynamicList : 1;
331
332 // Used to track if there has been at least one undefined reference to the
333 // symbol. For Undefined and SharedSymbol, the binding may change to STB_WEAK
334 // if the first undefined reference from a non-shared object is weak.
335 LLVM_PREFERRED_TYPE(bool)
336 uint8_t referenced : 1;
337
338 // Used to track if this symbol will be referenced after wrapping is performed
339 // (i.e. this will be true for foo if __real_foo is referenced, and will be
340 // true for __wrap_foo if foo is referenced).
341 LLVM_PREFERRED_TYPE(bool)
342 uint8_t referencedAfterWrap : 1;
343
344 void setFlags(uint16_t bits) {
345 flags.fetch_or(i: bits, m: std::memory_order_relaxed);
346 }
347 bool hasFlag(uint16_t bit) const {
348 assert(llvm::has_single_bit(bit) && "bit must be a power of 2");
349 return flags.load(m: std::memory_order_relaxed) & bit;
350 }
351
352 bool needsDynReloc() const {
353 return flags.load(m: std::memory_order_relaxed) &
354 (NEEDS_COPY | NEEDS_GOT | NEEDS_PLT | NEEDS_TLSDESC | NEEDS_TLSGD |
355 NEEDS_GOT_DTPREL | NEEDS_TLSIE);
356 }
357 void allocateAux(Ctx &ctx) {
358 assert(auxIdx == 0);
359 auxIdx = ctx.symAux.size();
360 ctx.symAux.emplace_back();
361 }
362
363 bool isSection() const { return type == llvm::ELF::STT_SECTION; }
364 bool isTls() const { return type == llvm::ELF::STT_TLS; }
365 bool isFunc() const { return type == llvm::ELF::STT_FUNC; }
366 bool isGnuIFunc() const { return type == llvm::ELF::STT_GNU_IFUNC; }
367 bool isObject() const { return type == llvm::ELF::STT_OBJECT; }
368 bool isFile() const { return type == llvm::ELF::STT_FILE; }
369};
370
371// Represents a symbol that is defined in the current output file.
372class Defined : public Symbol {
373public:
374 Defined(Ctx &ctx, InputFile *file, StringRef name, uint8_t binding,
375 uint8_t stOther, uint8_t type, uint64_t value, uint64_t size,
376 SectionBase *section)
377 : Symbol(DefinedKind, file, name, binding, stOther, type), value(value),
378 size(size), section(section) {
379 }
380 void overwrite(Symbol &sym) const;
381
382 static bool classof(const Symbol *s) { return s->isDefined(); }
383
384 uint64_t value;
385 uint64_t size;
386 SectionBase *section;
387};
388
389// Represents a common symbol.
390//
391// On Unix, it is traditionally allowed to write variable definitions
392// without initialization expressions (such as "int foo;") to header
393// files. Such definition is called "tentative definition".
394//
395// Using tentative definition is usually considered a bad practice
396// because you should write only declarations (such as "extern int
397// foo;") to header files. Nevertheless, the linker and the compiler
398// have to do something to support bad code by allowing duplicate
399// definitions for this particular case.
400//
401// Common symbols represent variable definitions without initializations.
402// The compiler creates common symbols when it sees variable definitions
403// without initialization (you can suppress this behavior and let the
404// compiler create a regular defined symbol by -fno-common).
405//
406// The linker allows common symbols to be replaced by regular defined
407// symbols. If there are remaining common symbols after name resolution is
408// complete, they are converted to regular defined symbols in a .bss
409// section. (Therefore, the later passes don't see any CommonSymbols.)
410class CommonSymbol : public Symbol {
411public:
412 CommonSymbol(Ctx &ctx, InputFile *file, StringRef name, uint8_t binding,
413 uint8_t stOther, uint8_t type, uint64_t alignment, uint64_t size)
414 : Symbol(CommonKind, file, name, binding, stOther, type),
415 alignment(alignment), size(size) {
416 }
417 void overwrite(Symbol &sym) const {
418 Symbol::overwrite(sym, k: CommonKind);
419 auto &s = static_cast<CommonSymbol &>(sym);
420 s.alignment = alignment;
421 s.size = size;
422 }
423
424 static bool classof(const Symbol *s) { return s->isCommon(); }
425
426 uint32_t alignment;
427 uint64_t size;
428};
429
430class Undefined : public Symbol {
431public:
432 Undefined(InputFile *file, StringRef name, uint8_t binding, uint8_t stOther,
433 uint8_t type, uint32_t discardedSecIdx = 0)
434 : Symbol(UndefinedKind, file, name, binding, stOther, type),
435 discardedSecIdx(discardedSecIdx) {}
436 void overwrite(Symbol &sym) const {
437 Symbol::overwrite(sym, k: UndefinedKind);
438 auto &s = static_cast<Undefined &>(sym);
439 s.discardedSecIdx = discardedSecIdx;
440 s.nonPrevailing = nonPrevailing;
441 }
442
443 static bool classof(const Symbol *s) { return s->kind() == UndefinedKind; }
444
445 // The section index if in a discarded section, 0 otherwise.
446 uint32_t discardedSecIdx;
447 bool nonPrevailing = false;
448};
449
450class SharedSymbol : public Symbol {
451public:
452 static bool classof(const Symbol *s) { return s->kind() == SharedKind; }
453
454 SharedSymbol(InputFile &file, StringRef name, uint8_t binding,
455 uint8_t stOther, uint8_t type, uint64_t value, uint64_t size,
456 uint32_t alignment)
457 : Symbol(SharedKind, &file, name, binding, stOther, type), value(value),
458 size(size), alignment(alignment) {
459 dsoProtected = visibility() == llvm::ELF::STV_PROTECTED;
460 // GNU ifunc is a mechanism to allow user-supplied functions to
461 // resolve PLT slot values at load-time. This is contrary to the
462 // regular symbol resolution scheme in which symbols are resolved just
463 // by name. Using this hook, you can program how symbols are solved
464 // for you program. For example, you can make "memcpy" to be resolved
465 // to a SSE-enabled version of memcpy only when a machine running the
466 // program supports the SSE instruction set.
467 //
468 // Naturally, such symbols should always be called through their PLT
469 // slots. What GNU ifunc symbols point to are resolver functions, and
470 // calling them directly doesn't make sense (unless you are writing a
471 // loader).
472 //
473 // For DSO symbols, we always call them through PLT slots anyway.
474 // So there's no difference between GNU ifunc and regular function
475 // symbols if they are in DSOs. So we can handle GNU_IFUNC as FUNC.
476 if (this->type == llvm::ELF::STT_GNU_IFUNC)
477 this->type = llvm::ELF::STT_FUNC;
478 }
479 void overwrite(Symbol &sym) const {
480 Symbol::overwrite(sym, k: SharedKind);
481 auto &s = static_cast<SharedSymbol &>(sym);
482 s.dsoProtected = dsoProtected;
483 s.value = value;
484 s.size = size;
485 s.alignment = alignment;
486 }
487
488 uint64_t value; // st_value
489 uint64_t size; // st_size
490 uint32_t alignment;
491};
492
493// LazySymbol symbols represent symbols in object files between --start-lib and
494// --end-lib options. LLD also handles traditional archives as if all the files
495// in the archive are surrounded by --start-lib and --end-lib.
496//
497// A special complication is the handling of weak undefined symbols. They should
498// not load a file, but we have to remember we have seen both the weak undefined
499// and the lazy. We represent that with a lazy symbol with a weak binding. This
500// means that code looking for undefined symbols normally also has to take lazy
501// symbols into consideration.
502class LazySymbol : public Symbol {
503public:
504 LazySymbol(InputFile &file)
505 : Symbol(LazyKind, &file, {}, llvm::ELF::STB_GLOBAL,
506 llvm::ELF::STV_DEFAULT, llvm::ELF::STT_NOTYPE) {}
507 void overwrite(Symbol &sym) const { Symbol::overwrite(sym, k: LazyKind); }
508
509 static bool classof(const Symbol *s) { return s->kind() == LazyKind; }
510};
511
512// A buffer class that is large enough to hold any Symbol-derived
513// object. We allocate memory using this class and instantiate a symbol
514// using the placement new.
515
516// It is important to keep the size of SymbolUnion small for performance and
517// memory usage reasons. 64 bytes is a soft limit based on the size of Defined
518// on a 64-bit system. This is enforced by a static_assert in Symbols.cpp.
519union SymbolUnion {
520 alignas(Defined) char a[sizeof(Defined)];
521 alignas(CommonSymbol) char b[sizeof(CommonSymbol)];
522 alignas(Undefined) char c[sizeof(Undefined)];
523 alignas(SharedSymbol) char d[sizeof(SharedSymbol)];
524 alignas(LazySymbol) char e[sizeof(LazySymbol)];
525};
526
527template <typename... T> Defined *makeDefined(T &&...args) {
528 auto *sym = getSpecificAllocSingleton<SymbolUnion>().Allocate();
529 memset(s: sym, c: 0, n: sizeof(Symbol));
530 auto &s = *new (reinterpret_cast<Defined *>(sym)) Defined(std::forward<T>(args)...);
531 return &s;
532}
533
534void reportDuplicate(Ctx &, const Symbol &sym, const InputFile *newFile,
535 InputSectionBase *errSec, uint64_t errOffset);
536void maybeWarnUnorderableSymbol(Ctx &, const Symbol *sym);
537bool computeIsPreemptible(Ctx &, const Symbol &sym);
538void parseVersionAndComputeIsPreemptible(Ctx &);
539
540} // namespace lld::elf
541
542#endif
543