1//===- SymbolTable.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// Symbol table is a bag of all known symbols. We put all symbols of
10// all input files to the symbol table. The symbol table is basically
11// a hash table with the logic to resolve symbol name conflicts using
12// the symbol types.
13//
14//===----------------------------------------------------------------------===//
15
16#include "SymbolTable.h"
17#include "Config.h"
18#include "InputFiles.h"
19#include "Symbols.h"
20#include "lld/Common/ErrorHandler.h"
21#include "lld/Common/Memory.h"
22#include "lld/Common/Strings.h"
23#include "llvm/ADT/STLExtras.h"
24#include "llvm/Demangle/Demangle.h"
25
26using namespace llvm;
27using namespace llvm::object;
28using namespace llvm::ELF;
29using namespace lld;
30using namespace lld::elf;
31
32SymbolTable elf::symtab;
33
34void SymbolTable::wrap(Symbol *sym, Symbol *real, Symbol *wrap) {
35 // Redirect __real_foo to the original foo and foo to the original __wrap_foo.
36 int &idx1 = symMap[CachedHashStringRef(sym->getName())];
37 int &idx2 = symMap[CachedHashStringRef(real->getName())];
38 int &idx3 = symMap[CachedHashStringRef(wrap->getName())];
39
40 idx2 = idx1;
41 idx1 = idx3;
42
43 // Propagate symbol usage information to the redirected symbols.
44 if (sym->isUsedInRegularObj)
45 wrap->isUsedInRegularObj = true;
46 if (real->isUsedInRegularObj)
47 sym->isUsedInRegularObj = true;
48 else if (!sym->isDefined())
49 // Now that all references to sym have been redirected to wrap, if there are
50 // no references to real (which has been redirected to sym), we only need to
51 // keep sym if it was defined, otherwise it's unused and can be dropped.
52 sym->isUsedInRegularObj = false;
53
54 // Now renaming is complete, and no one refers to real. We drop real from
55 // .symtab and .dynsym. If real is undefined, it is important that we don't
56 // leave it in .dynsym, because otherwise it might lead to an undefined symbol
57 // error in a subsequent link. If real is defined, we could emit real as an
58 // alias for sym, but that could degrade the user experience of some tools
59 // that can print out only one symbol for each location: sym is a preferred
60 // name than real, but they might print out real instead.
61 memcpy(dest: real, src: sym, n: sizeof(SymbolUnion));
62 real->isUsedInRegularObj = false;
63}
64
65// Find an existing symbol or create a new one.
66Symbol *SymbolTable::insert(StringRef name) {
67 // <name>@@<version> means the symbol is the default version. In that
68 // case <name>@@<version> will be used to resolve references to <name>.
69 //
70 // Since this is a hot path, the following string search code is
71 // optimized for speed. StringRef::find(char) is much faster than
72 // StringRef::find(StringRef).
73 StringRef stem = name;
74 size_t pos = name.find(C: '@');
75 if (pos != StringRef::npos && pos + 1 < name.size() && name[pos + 1] == '@')
76 stem = name.take_front(N: pos);
77
78 auto p = symMap.insert(KV: {CachedHashStringRef(stem), (int)symVector.size()});
79 if (!p.second) {
80 Symbol *sym = symVector[p.first->second];
81 if (stem.size() != name.size()) {
82 sym->setName(name);
83 sym->hasVersionSuffix = true;
84 }
85 return sym;
86 }
87
88 Symbol *sym = reinterpret_cast<Symbol *>(make<SymbolUnion>());
89 symVector.push_back(Elt: sym);
90
91 // *sym was not initialized by a constructor. Initialize all Symbol fields.
92 memset(s: sym, c: 0, n: sizeof(Symbol));
93 sym->setName(name);
94 sym->partition = 1;
95 sym->versionId = VER_NDX_GLOBAL;
96 if (pos != StringRef::npos)
97 sym->hasVersionSuffix = true;
98 return sym;
99}
100
101// This variant of addSymbol is used by BinaryFile::parse to check duplicate
102// symbol errors.
103Symbol *SymbolTable::addAndCheckDuplicate(const Defined &newSym) {
104 Symbol *sym = insert(name: newSym.getName());
105 if (sym->isDefined())
106 sym->checkDuplicate(other: newSym);
107 sym->resolve(other: newSym);
108 sym->isUsedInRegularObj = true;
109 return sym;
110}
111
112Symbol *SymbolTable::find(StringRef name) {
113 auto it = symMap.find(Val: CachedHashStringRef(name));
114 if (it == symMap.end())
115 return nullptr;
116 return symVector[it->second];
117}
118
119// A version script/dynamic list is only meaningful for a Defined symbol.
120// A CommonSymbol will be converted to a Defined in replaceCommonSymbols().
121// A lazy symbol may be made Defined if an LTO libcall extracts it.
122static bool canBeVersioned(const Symbol &sym) {
123 return sym.isDefined() || sym.isCommon() || sym.isLazy();
124}
125
126// Initialize demangledSyms with a map from demangled symbols to symbol
127// objects. Used to handle "extern C++" directive in version scripts.
128//
129// The map will contain all demangled symbols. That can be very large,
130// and in LLD we generally want to avoid do anything for each symbol.
131// Then, why are we doing this? Here's why.
132//
133// Users can use "extern C++ {}" directive to match against demangled
134// C++ symbols. For example, you can write a pattern such as
135// "llvm::*::foo(int, ?)". Obviously, there's no way to handle this
136// other than trying to match a pattern against all demangled symbols.
137// So, if "extern C++" feature is used, we need to demangle all known
138// symbols.
139StringMap<SmallVector<Symbol *, 0>> &SymbolTable::getDemangledSyms() {
140 if (!demangledSyms) {
141 demangledSyms.emplace();
142 std::string demangled;
143 for (Symbol *sym : symVector)
144 if (canBeVersioned(sym: *sym)) {
145 StringRef name = sym->getName();
146 size_t pos = name.find(C: '@');
147 std::string substr;
148 if (pos == std::string::npos)
149 demangled = demangle(MangledName: name);
150 else if (pos + 1 == name.size() || name[pos + 1] == '@') {
151 substr = name.substr(Start: 0, N: pos);
152 demangled = demangle(MangledName: substr);
153 } else {
154 substr = name.substr(Start: 0, N: pos);
155 demangled = (demangle(MangledName: substr) + name.substr(Start: pos)).str();
156 }
157 (*demangledSyms)[demangled].push_back(Elt: sym);
158 }
159 }
160 return *demangledSyms;
161}
162
163SmallVector<Symbol *, 0> SymbolTable::findByVersion(SymbolVersion ver) {
164 if (ver.isExternCpp)
165 return getDemangledSyms().lookup(Key: ver.name);
166 if (Symbol *sym = find(name: ver.name))
167 if (canBeVersioned(sym: *sym))
168 return {sym};
169 return {};
170}
171
172SmallVector<Symbol *, 0> SymbolTable::findAllByVersion(SymbolVersion ver,
173 bool includeNonDefault) {
174 SmallVector<Symbol *, 0> res;
175 SingleStringMatcher m(ver.name);
176 auto check = [&](const Symbol &sym) -> bool {
177 if (!includeNonDefault)
178 return !sym.hasVersionSuffix;
179 StringRef name = sym.getName();
180 size_t pos = name.find(C: '@');
181 return !(pos + 1 < name.size() && name[pos + 1] == '@');
182 };
183
184 if (ver.isExternCpp) {
185 for (auto &p : getDemangledSyms())
186 if (m.match(s: p.first()))
187 for (Symbol *sym : p.second)
188 if (check(*sym))
189 res.push_back(Elt: sym);
190 return res;
191 }
192
193 for (Symbol *sym : symVector)
194 if (canBeVersioned(sym: *sym) && check(*sym) && m.match(s: sym->getName()))
195 res.push_back(Elt: sym);
196 return res;
197}
198
199void SymbolTable::handleDynamicList() {
200 SmallVector<Symbol *, 0> syms;
201 for (SymbolVersion &ver : config->dynamicList) {
202 if (ver.hasWildcard)
203 syms = findAllByVersion(ver, /*includeNonDefault=*/true);
204 else
205 syms = findByVersion(ver);
206
207 for (Symbol *sym : syms)
208 sym->inDynamicList = true;
209 }
210}
211
212// Set symbol versions to symbols. This function handles patterns containing no
213// wildcard characters. Return false if no symbol definition matches ver.
214bool SymbolTable::assignExactVersion(SymbolVersion ver, uint16_t versionId,
215 StringRef versionName,
216 bool includeNonDefault) {
217 // Get a list of symbols which we need to assign the version to.
218 SmallVector<Symbol *, 0> syms = findByVersion(ver);
219
220 auto getName = [](uint16_t ver) -> std::string {
221 if (ver == VER_NDX_LOCAL)
222 return "VER_NDX_LOCAL";
223 if (ver == VER_NDX_GLOBAL)
224 return "VER_NDX_GLOBAL";
225 return ("version '" + config->versionDefinitions[ver].name + "'").str();
226 };
227
228 // Assign the version.
229 for (Symbol *sym : syms) {
230 // For a non-local versionId, skip symbols containing version info because
231 // symbol versions specified by symbol names take precedence over version
232 // scripts. See parseSymbolVersion().
233 if (!includeNonDefault && versionId != VER_NDX_LOCAL &&
234 sym->getName().contains(C: '@'))
235 continue;
236
237 // If the version has not been assigned, assign versionId to the symbol.
238 if (!sym->versionScriptAssigned) {
239 sym->versionScriptAssigned = true;
240 sym->versionId = versionId;
241 }
242 if (sym->versionId == versionId)
243 continue;
244
245 warn(msg: "attempt to reassign symbol '" + ver.name + "' of " +
246 getName(sym->versionId) + " to " + getName(versionId));
247 }
248 return !syms.empty();
249}
250
251void SymbolTable::assignWildcardVersion(SymbolVersion ver, uint16_t versionId,
252 bool includeNonDefault) {
253 // Exact matching takes precedence over fuzzy matching,
254 // so we set a version to a symbol only if no version has been assigned
255 // to the symbol. This behavior is compatible with GNU.
256 for (Symbol *sym : findAllByVersion(ver, includeNonDefault))
257 if (!sym->versionScriptAssigned) {
258 sym->versionScriptAssigned = true;
259 sym->versionId = versionId;
260 }
261}
262
263// This function processes version scripts by updating the versionId
264// member of symbols.
265// If there's only one anonymous version definition in a version
266// script file, the script does not actually define any symbol version,
267// but just specifies symbols visibilities.
268void SymbolTable::scanVersionScript() {
269 SmallString<128> buf;
270 // First, we assign versions to exact matching symbols,
271 // i.e. version definitions not containing any glob meta-characters.
272 for (VersionDefinition &v : config->versionDefinitions) {
273 auto assignExact = [&](SymbolVersion pat, uint16_t id, StringRef ver) {
274 bool found =
275 assignExactVersion(ver: pat, versionId: id, versionName: ver, /*includeNonDefault=*/false);
276 buf.clear();
277 found |= assignExactVersion(ver: {.name: (pat.name + "@" + v.name).toStringRef(Out&: buf),
278 .isExternCpp: pat.isExternCpp, /*hasWildCard=*/.hasWildcard: false},
279 versionId: id, versionName: ver, /*includeNonDefault=*/true);
280 if (!found && !config->undefinedVersion)
281 errorOrWarn(msg: "version script assignment of '" + ver + "' to symbol '" +
282 pat.name + "' failed: symbol not defined");
283 };
284 for (SymbolVersion &pat : v.nonLocalPatterns)
285 if (!pat.hasWildcard)
286 assignExact(pat, v.id, v.name);
287 for (SymbolVersion pat : v.localPatterns)
288 if (!pat.hasWildcard)
289 assignExact(pat, VER_NDX_LOCAL, "local");
290 }
291
292 // Next, assign versions to wildcards that are not "*". Note that because the
293 // last match takes precedence over previous matches, we iterate over the
294 // definitions in the reverse order.
295 auto assignWildcard = [&](SymbolVersion pat, uint16_t id, StringRef ver) {
296 assignWildcardVersion(ver: pat, versionId: id, /*includeNonDefault=*/false);
297 buf.clear();
298 assignWildcardVersion(ver: {.name: (pat.name + "@" + ver).toStringRef(Out&: buf),
299 .isExternCpp: pat.isExternCpp, /*hasWildCard=*/.hasWildcard: true},
300 versionId: id,
301 /*includeNonDefault=*/true);
302 };
303 for (VersionDefinition &v : llvm::reverse(C&: config->versionDefinitions)) {
304 for (SymbolVersion &pat : v.nonLocalPatterns)
305 if (pat.hasWildcard && pat.name != "*")
306 assignWildcard(pat, v.id, v.name);
307 for (SymbolVersion &pat : v.localPatterns)
308 if (pat.hasWildcard && pat.name != "*")
309 assignWildcard(pat, VER_NDX_LOCAL, v.name);
310 }
311
312 // Then, assign versions to "*". In GNU linkers they have lower priority than
313 // other wildcards.
314 for (VersionDefinition &v : llvm::reverse(C&: config->versionDefinitions)) {
315 for (SymbolVersion &pat : v.nonLocalPatterns)
316 if (pat.hasWildcard && pat.name == "*")
317 assignWildcard(pat, v.id, v.name);
318 for (SymbolVersion &pat : v.localPatterns)
319 if (pat.hasWildcard && pat.name == "*")
320 assignWildcard(pat, VER_NDX_LOCAL, v.name);
321 }
322
323 // Symbol themselves might know their versions because symbols
324 // can contain versions in the form of <name>@<version>.
325 // Let them parse and update their names to exclude version suffix.
326 for (Symbol *sym : symVector)
327 if (sym->hasVersionSuffix)
328 sym->parseSymbolVersion();
329
330 // isPreemptible is false at this point. To correctly compute the binding of a
331 // Defined (which is used by includeInDynsym()), we need to know if it is
332 // VER_NDX_LOCAL or not. Compute symbol versions before handling
333 // --dynamic-list.
334 handleDynamicList();
335}
336
337Symbol *SymbolTable::addUnusedUndefined(StringRef name, uint8_t binding) {
338 return addSymbol(newSym: Undefined{ctx.internalFile, name, binding, STV_DEFAULT, 0});
339}
340