xref: /freebsd/contrib/llvm-project/lld/ELF/SymbolTable.cpp (revision 0eae32dcef82f6f06de6419a0d623d7def0cc8f6)
10b57cec5SDimitry Andric //===- SymbolTable.cpp ----------------------------------------------------===//
20b57cec5SDimitry Andric //
30b57cec5SDimitry Andric // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
40b57cec5SDimitry Andric // See https://llvm.org/LICENSE.txt for license information.
50b57cec5SDimitry Andric // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
60b57cec5SDimitry Andric //
70b57cec5SDimitry Andric //===----------------------------------------------------------------------===//
80b57cec5SDimitry Andric //
90b57cec5SDimitry Andric // Symbol table is a bag of all known symbols. We put all symbols of
100b57cec5SDimitry Andric // all input files to the symbol table. The symbol table is basically
110b57cec5SDimitry Andric // a hash table with the logic to resolve symbol name conflicts using
120b57cec5SDimitry Andric // the symbol types.
130b57cec5SDimitry Andric //
140b57cec5SDimitry Andric //===----------------------------------------------------------------------===//
150b57cec5SDimitry Andric 
160b57cec5SDimitry Andric #include "SymbolTable.h"
170b57cec5SDimitry Andric #include "Config.h"
180b57cec5SDimitry Andric #include "LinkerScript.h"
190b57cec5SDimitry Andric #include "Symbols.h"
200b57cec5SDimitry Andric #include "SyntheticSections.h"
210b57cec5SDimitry Andric #include "lld/Common/ErrorHandler.h"
220b57cec5SDimitry Andric #include "lld/Common/Memory.h"
230b57cec5SDimitry Andric #include "lld/Common/Strings.h"
240b57cec5SDimitry Andric #include "llvm/ADT/STLExtras.h"
250b57cec5SDimitry Andric 
260b57cec5SDimitry Andric using namespace llvm;
270b57cec5SDimitry Andric using namespace llvm::object;
280b57cec5SDimitry Andric using namespace llvm::ELF;
295ffd83dbSDimitry Andric using namespace lld;
305ffd83dbSDimitry Andric using namespace lld::elf;
310b57cec5SDimitry Andric 
32*0eae32dcSDimitry Andric std::unique_ptr<SymbolTable> elf::symtab;
330b57cec5SDimitry Andric 
340b57cec5SDimitry Andric void SymbolTable::wrap(Symbol *sym, Symbol *real, Symbol *wrap) {
35349cc55cSDimitry Andric   // Redirect __real_foo to the original foo and foo to the original __wrap_foo.
360b57cec5SDimitry Andric   int &idx1 = symMap[CachedHashStringRef(sym->getName())];
370b57cec5SDimitry Andric   int &idx2 = symMap[CachedHashStringRef(real->getName())];
380b57cec5SDimitry Andric   int &idx3 = symMap[CachedHashStringRef(wrap->getName())];
390b57cec5SDimitry Andric 
400b57cec5SDimitry Andric   idx2 = idx1;
410b57cec5SDimitry Andric   idx1 = idx3;
420b57cec5SDimitry Andric 
435ffd83dbSDimitry Andric   if (real->exportDynamic)
445ffd83dbSDimitry Andric     sym->exportDynamic = true;
45fe6060f1SDimitry Andric   if (!real->isUsedInRegularObj && sym->isUndefined())
46e8d8bef9SDimitry Andric     sym->isUsedInRegularObj = false;
475ffd83dbSDimitry Andric 
485ffd83dbSDimitry Andric   // Now renaming is complete, and no one refers to real. We drop real from
495ffd83dbSDimitry Andric   // .symtab and .dynsym. If real is undefined, it is important that we don't
505ffd83dbSDimitry Andric   // leave it in .dynsym, because otherwise it might lead to an undefined symbol
515ffd83dbSDimitry Andric   // error in a subsequent link. If real is defined, we could emit real as an
525ffd83dbSDimitry Andric   // alias for sym, but that could degrade the user experience of some tools
535ffd83dbSDimitry Andric   // that can print out only one symbol for each location: sym is a preferred
545ffd83dbSDimitry Andric   // name than real, but they might print out real instead.
550b57cec5SDimitry Andric   memcpy(real, sym, sizeof(SymbolUnion));
565ffd83dbSDimitry Andric   real->isUsedInRegularObj = false;
570b57cec5SDimitry Andric }
580b57cec5SDimitry Andric 
590b57cec5SDimitry Andric // Find an existing symbol or create a new one.
600b57cec5SDimitry Andric Symbol *SymbolTable::insert(StringRef name) {
610b57cec5SDimitry Andric   // <name>@@<version> means the symbol is the default version. In that
620b57cec5SDimitry Andric   // case <name>@@<version> will be used to resolve references to <name>.
630b57cec5SDimitry Andric   //
640b57cec5SDimitry Andric   // Since this is a hot path, the following string search code is
650b57cec5SDimitry Andric   // optimized for speed. StringRef::find(char) is much faster than
660b57cec5SDimitry Andric   // StringRef::find(StringRef).
67*0eae32dcSDimitry Andric   StringRef stem = name;
680b57cec5SDimitry Andric   size_t pos = name.find('@');
690b57cec5SDimitry Andric   if (pos != StringRef::npos && pos + 1 < name.size() && name[pos + 1] == '@')
70*0eae32dcSDimitry Andric     stem = name.take_front(pos);
710b57cec5SDimitry Andric 
72*0eae32dcSDimitry Andric   auto p = symMap.insert({CachedHashStringRef(stem), (int)symVector.size()});
73*0eae32dcSDimitry Andric   if (!p.second) {
74*0eae32dcSDimitry Andric     Symbol *sym = symVector[p.first->second];
75*0eae32dcSDimitry Andric     if (stem.size() != name.size())
76*0eae32dcSDimitry Andric       sym->setName(name);
77*0eae32dcSDimitry Andric     return sym;
78*0eae32dcSDimitry Andric   }
790b57cec5SDimitry Andric 
800b57cec5SDimitry Andric   Symbol *sym = reinterpret_cast<Symbol *>(make<SymbolUnion>());
810b57cec5SDimitry Andric   symVector.push_back(sym);
820b57cec5SDimitry Andric 
8385868e8aSDimitry Andric   // *sym was not initialized by a constructor. Fields that may get referenced
8485868e8aSDimitry Andric   // when it is a placeholder must be initialized here.
850b57cec5SDimitry Andric   sym->setName(name);
860b57cec5SDimitry Andric   sym->symbolKind = Symbol::PlaceholderKind;
8785868e8aSDimitry Andric   sym->versionId = VER_NDX_GLOBAL;
880b57cec5SDimitry Andric   sym->visibility = STV_DEFAULT;
890b57cec5SDimitry Andric   sym->isUsedInRegularObj = false;
900b57cec5SDimitry Andric   sym->exportDynamic = false;
9185868e8aSDimitry Andric   sym->inDynamicList = false;
920b57cec5SDimitry Andric   sym->canInline = true;
9385868e8aSDimitry Andric   sym->referenced = false;
9485868e8aSDimitry Andric   sym->traced = false;
950b57cec5SDimitry Andric   sym->scriptDefined = false;
960b57cec5SDimitry Andric   sym->partition = 1;
970b57cec5SDimitry Andric   return sym;
980b57cec5SDimitry Andric }
990b57cec5SDimitry Andric 
10085868e8aSDimitry Andric Symbol *SymbolTable::addSymbol(const Symbol &newSym) {
1015ffd83dbSDimitry Andric   Symbol *sym = insert(newSym.getName());
10285868e8aSDimitry Andric   sym->resolve(newSym);
1030b57cec5SDimitry Andric   return sym;
1040b57cec5SDimitry Andric }
1050b57cec5SDimitry Andric 
1060b57cec5SDimitry Andric Symbol *SymbolTable::find(StringRef name) {
1070b57cec5SDimitry Andric   auto it = symMap.find(CachedHashStringRef(name));
1080b57cec5SDimitry Andric   if (it == symMap.end())
1090b57cec5SDimitry Andric     return nullptr;
110*0eae32dcSDimitry Andric   return symVector[it->second];
1110b57cec5SDimitry Andric }
1120b57cec5SDimitry Andric 
1135ffd83dbSDimitry Andric // A version script/dynamic list is only meaningful for a Defined symbol.
1145ffd83dbSDimitry Andric // A CommonSymbol will be converted to a Defined in replaceCommonSymbols().
1154824e7fdSDimitry Andric // A lazy symbol may be made Defined if an LTO libcall extracts it.
1165ffd83dbSDimitry Andric static bool canBeVersioned(const Symbol &sym) {
1175ffd83dbSDimitry Andric   return sym.isDefined() || sym.isCommon() || sym.isLazy();
1185ffd83dbSDimitry Andric }
1195ffd83dbSDimitry Andric 
1200b57cec5SDimitry Andric // Initialize demangledSyms with a map from demangled symbols to symbol
1210b57cec5SDimitry Andric // objects. Used to handle "extern C++" directive in version scripts.
1220b57cec5SDimitry Andric //
1230b57cec5SDimitry Andric // The map will contain all demangled symbols. That can be very large,
1240b57cec5SDimitry Andric // and in LLD we generally want to avoid do anything for each symbol.
1250b57cec5SDimitry Andric // Then, why are we doing this? Here's why.
1260b57cec5SDimitry Andric //
1270b57cec5SDimitry Andric // Users can use "extern C++ {}" directive to match against demangled
1280b57cec5SDimitry Andric // C++ symbols. For example, you can write a pattern such as
1290b57cec5SDimitry Andric // "llvm::*::foo(int, ?)". Obviously, there's no way to handle this
1300b57cec5SDimitry Andric // other than trying to match a pattern against all demangled symbols.
1310b57cec5SDimitry Andric // So, if "extern C++" feature is used, we need to demangle all known
1320b57cec5SDimitry Andric // symbols.
133*0eae32dcSDimitry Andric StringMap<SmallVector<Symbol *, 0>> &SymbolTable::getDemangledSyms() {
1340b57cec5SDimitry Andric   if (!demangledSyms) {
1350b57cec5SDimitry Andric     demangledSyms.emplace();
1366e75b2fbSDimitry Andric     std::string demangled;
1375ffd83dbSDimitry Andric     for (Symbol *sym : symVector)
1386e75b2fbSDimitry Andric       if (canBeVersioned(*sym)) {
1396e75b2fbSDimitry Andric         StringRef name = sym->getName();
1406e75b2fbSDimitry Andric         size_t pos = name.find('@');
1416e75b2fbSDimitry Andric         if (pos == std::string::npos)
1426e75b2fbSDimitry Andric           demangled = demangleItanium(name);
1436e75b2fbSDimitry Andric         else if (pos + 1 == name.size() || name[pos + 1] == '@')
1446e75b2fbSDimitry Andric           demangled = demangleItanium(name.substr(0, pos));
1456e75b2fbSDimitry Andric         else
1466e75b2fbSDimitry Andric           demangled =
1476e75b2fbSDimitry Andric               (demangleItanium(name.substr(0, pos)) + name.substr(pos)).str();
1486e75b2fbSDimitry Andric         (*demangledSyms)[demangled].push_back(sym);
1496e75b2fbSDimitry Andric       }
1500b57cec5SDimitry Andric   }
1510b57cec5SDimitry Andric   return *demangledSyms;
1520b57cec5SDimitry Andric }
1530b57cec5SDimitry Andric 
154*0eae32dcSDimitry Andric SmallVector<Symbol *, 0> SymbolTable::findByVersion(SymbolVersion ver) {
1550b57cec5SDimitry Andric   if (ver.isExternCpp)
1560b57cec5SDimitry Andric     return getDemangledSyms().lookup(ver.name);
1575ffd83dbSDimitry Andric   if (Symbol *sym = find(ver.name))
1585ffd83dbSDimitry Andric     if (canBeVersioned(*sym))
1595ffd83dbSDimitry Andric       return {sym};
1600b57cec5SDimitry Andric   return {};
1610b57cec5SDimitry Andric }
1620b57cec5SDimitry Andric 
163*0eae32dcSDimitry Andric SmallVector<Symbol *, 0> SymbolTable::findAllByVersion(SymbolVersion ver,
1646e75b2fbSDimitry Andric                                                        bool includeNonDefault) {
165*0eae32dcSDimitry Andric   SmallVector<Symbol *, 0> res;
1665ffd83dbSDimitry Andric   SingleStringMatcher m(ver.name);
1676e75b2fbSDimitry Andric   auto check = [&](StringRef name) {
1686e75b2fbSDimitry Andric     size_t pos = name.find('@');
1696e75b2fbSDimitry Andric     if (!includeNonDefault)
1706e75b2fbSDimitry Andric       return pos == StringRef::npos;
1716e75b2fbSDimitry Andric     return !(pos + 1 < name.size() && name[pos + 1] == '@');
1726e75b2fbSDimitry Andric   };
1730b57cec5SDimitry Andric 
1740b57cec5SDimitry Andric   if (ver.isExternCpp) {
1750b57cec5SDimitry Andric     for (auto &p : getDemangledSyms())
1760b57cec5SDimitry Andric       if (m.match(p.first()))
1776e75b2fbSDimitry Andric         for (Symbol *sym : p.second)
1786e75b2fbSDimitry Andric           if (check(sym->getName()))
1796e75b2fbSDimitry Andric             res.push_back(sym);
1800b57cec5SDimitry Andric     return res;
1810b57cec5SDimitry Andric   }
1820b57cec5SDimitry Andric 
1830b57cec5SDimitry Andric   for (Symbol *sym : symVector)
1846e75b2fbSDimitry Andric     if (canBeVersioned(*sym) && check(sym->getName()) &&
1856e75b2fbSDimitry Andric         m.match(sym->getName()))
1860b57cec5SDimitry Andric       res.push_back(sym);
1870b57cec5SDimitry Andric   return res;
1880b57cec5SDimitry Andric }
1890b57cec5SDimitry Andric 
1900b57cec5SDimitry Andric void SymbolTable::handleDynamicList() {
191*0eae32dcSDimitry Andric   SmallVector<Symbol *, 0> syms;
1920b57cec5SDimitry Andric   for (SymbolVersion &ver : config->dynamicList) {
1930b57cec5SDimitry Andric     if (ver.hasWildcard)
1946e75b2fbSDimitry Andric       syms = findAllByVersion(ver, /*includeNonDefault=*/true);
1950b57cec5SDimitry Andric     else
1960b57cec5SDimitry Andric       syms = findByVersion(ver);
1970b57cec5SDimitry Andric 
19885868e8aSDimitry Andric     for (Symbol *sym : syms)
19985868e8aSDimitry Andric       sym->inDynamicList = true;
2000b57cec5SDimitry Andric   }
2010b57cec5SDimitry Andric }
2020b57cec5SDimitry Andric 
2036e75b2fbSDimitry Andric // Set symbol versions to symbols. This function handles patterns containing no
2046e75b2fbSDimitry Andric // wildcard characters. Return false if no symbol definition matches ver.
2056e75b2fbSDimitry Andric bool SymbolTable::assignExactVersion(SymbolVersion ver, uint16_t versionId,
2066e75b2fbSDimitry Andric                                      StringRef versionName,
2076e75b2fbSDimitry Andric                                      bool includeNonDefault) {
2080b57cec5SDimitry Andric   // Get a list of symbols which we need to assign the version to.
209*0eae32dcSDimitry Andric   SmallVector<Symbol *, 0> syms = findByVersion(ver);
2100b57cec5SDimitry Andric 
2110b57cec5SDimitry Andric   auto getName = [](uint16_t ver) -> std::string {
2120b57cec5SDimitry Andric     if (ver == VER_NDX_LOCAL)
2130b57cec5SDimitry Andric       return "VER_NDX_LOCAL";
2140b57cec5SDimitry Andric     if (ver == VER_NDX_GLOBAL)
2150b57cec5SDimitry Andric       return "VER_NDX_GLOBAL";
21685868e8aSDimitry Andric     return ("version '" + config->versionDefinitions[ver].name + "'").str();
2170b57cec5SDimitry Andric   };
2180b57cec5SDimitry Andric 
2190b57cec5SDimitry Andric   // Assign the version.
2200b57cec5SDimitry Andric   for (Symbol *sym : syms) {
2216e75b2fbSDimitry Andric     // For a non-local versionId, skip symbols containing version info because
2226e75b2fbSDimitry Andric     // symbol versions specified by symbol names take precedence over version
2236e75b2fbSDimitry Andric     // scripts. See parseSymbolVersion().
2246e75b2fbSDimitry Andric     if (!includeNonDefault && versionId != VER_NDX_LOCAL &&
2256e75b2fbSDimitry Andric         sym->getName().contains('@'))
2260b57cec5SDimitry Andric       continue;
2270b57cec5SDimitry Andric 
22885868e8aSDimitry Andric     // If the version has not been assigned, verdefIndex is -1. Use an arbitrary
22985868e8aSDimitry Andric     // number (0) to indicate the version has been assigned.
230*0eae32dcSDimitry Andric     if (sym->verdefIndex == uint16_t(-1)) {
23185868e8aSDimitry Andric       sym->verdefIndex = 0;
2320b57cec5SDimitry Andric       sym->versionId = versionId;
23385868e8aSDimitry Andric     }
2340b57cec5SDimitry Andric     if (sym->versionId == versionId)
2350b57cec5SDimitry Andric       continue;
2360b57cec5SDimitry Andric 
2370b57cec5SDimitry Andric     warn("attempt to reassign symbol '" + ver.name + "' of " +
2380b57cec5SDimitry Andric          getName(sym->versionId) + " to " + getName(versionId));
2390b57cec5SDimitry Andric   }
2406e75b2fbSDimitry Andric   return !syms.empty();
2410b57cec5SDimitry Andric }
2420b57cec5SDimitry Andric 
2436e75b2fbSDimitry Andric void SymbolTable::assignWildcardVersion(SymbolVersion ver, uint16_t versionId,
2446e75b2fbSDimitry Andric                                         bool includeNonDefault) {
245480093f4SDimitry Andric   // Exact matching takes precedence over fuzzy matching,
2460b57cec5SDimitry Andric   // so we set a version to a symbol only if no version has been assigned
2470b57cec5SDimitry Andric   // to the symbol. This behavior is compatible with GNU.
2486e75b2fbSDimitry Andric   for (Symbol *sym : findAllByVersion(ver, includeNonDefault))
249*0eae32dcSDimitry Andric     if (sym->verdefIndex == uint16_t(-1)) {
25085868e8aSDimitry Andric       sym->verdefIndex = 0;
25185868e8aSDimitry Andric       sym->versionId = versionId;
25285868e8aSDimitry Andric     }
2530b57cec5SDimitry Andric }
2540b57cec5SDimitry Andric 
2550b57cec5SDimitry Andric // This function processes version scripts by updating the versionId
2560b57cec5SDimitry Andric // member of symbols.
2570b57cec5SDimitry Andric // If there's only one anonymous version definition in a version
2580b57cec5SDimitry Andric // script file, the script does not actually define any symbol version,
2590b57cec5SDimitry Andric // but just specifies symbols visibilities.
2600b57cec5SDimitry Andric void SymbolTable::scanVersionScript() {
2616e75b2fbSDimitry Andric   SmallString<128> buf;
2620b57cec5SDimitry Andric   // First, we assign versions to exact matching symbols,
2630b57cec5SDimitry Andric   // i.e. version definitions not containing any glob meta-characters.
2646e75b2fbSDimitry Andric   for (VersionDefinition &v : config->versionDefinitions) {
2656e75b2fbSDimitry Andric     auto assignExact = [&](SymbolVersion pat, uint16_t id, StringRef ver) {
2666e75b2fbSDimitry Andric       bool found =
2676e75b2fbSDimitry Andric           assignExactVersion(pat, id, ver, /*includeNonDefault=*/false);
2686e75b2fbSDimitry Andric       buf.clear();
2696e75b2fbSDimitry Andric       found |= assignExactVersion({(pat.name + "@" + v.name).toStringRef(buf),
2706e75b2fbSDimitry Andric                                    pat.isExternCpp, /*hasWildCard=*/false},
2716e75b2fbSDimitry Andric                                   id, ver, /*includeNonDefault=*/true);
2726e75b2fbSDimitry Andric       if (!found && !config->undefinedVersion)
2736e75b2fbSDimitry Andric         errorOrWarn("version script assignment of '" + ver + "' to symbol '" +
2746e75b2fbSDimitry Andric                     pat.name + "' failed: symbol not defined");
2756e75b2fbSDimitry Andric     };
2766e75b2fbSDimitry Andric     for (SymbolVersion &pat : v.nonLocalPatterns)
2776e75b2fbSDimitry Andric       if (!pat.hasWildcard)
2786e75b2fbSDimitry Andric         assignExact(pat, v.id, v.name);
2796e75b2fbSDimitry Andric     for (SymbolVersion pat : v.localPatterns)
2806e75b2fbSDimitry Andric       if (!pat.hasWildcard)
2816e75b2fbSDimitry Andric         assignExact(pat, VER_NDX_LOCAL, "local");
2826e75b2fbSDimitry Andric   }
2830b57cec5SDimitry Andric 
28485868e8aSDimitry Andric   // Next, assign versions to wildcards that are not "*". Note that because the
28585868e8aSDimitry Andric   // last match takes precedence over previous matches, we iterate over the
28685868e8aSDimitry Andric   // definitions in the reverse order.
2876e75b2fbSDimitry Andric   auto assignWildcard = [&](SymbolVersion pat, uint16_t id, StringRef ver) {
2886e75b2fbSDimitry Andric     assignWildcardVersion(pat, id, /*includeNonDefault=*/false);
2896e75b2fbSDimitry Andric     buf.clear();
2906e75b2fbSDimitry Andric     assignWildcardVersion({(pat.name + "@" + ver).toStringRef(buf),
2916e75b2fbSDimitry Andric                            pat.isExternCpp, /*hasWildCard=*/true},
2926e75b2fbSDimitry Andric                           id,
2936e75b2fbSDimitry Andric                           /*includeNonDefault=*/true);
2946e75b2fbSDimitry Andric   };
2956e75b2fbSDimitry Andric   for (VersionDefinition &v : llvm::reverse(config->versionDefinitions)) {
2966e75b2fbSDimitry Andric     for (SymbolVersion &pat : v.nonLocalPatterns)
29785868e8aSDimitry Andric       if (pat.hasWildcard && pat.name != "*")
2986e75b2fbSDimitry Andric         assignWildcard(pat, v.id, v.name);
2996e75b2fbSDimitry Andric     for (SymbolVersion &pat : v.localPatterns)
3006e75b2fbSDimitry Andric       if (pat.hasWildcard && pat.name != "*")
3016e75b2fbSDimitry Andric         assignWildcard(pat, VER_NDX_LOCAL, v.name);
3026e75b2fbSDimitry Andric   }
30385868e8aSDimitry Andric 
30485868e8aSDimitry Andric   // Then, assign versions to "*". In GNU linkers they have lower priority than
30585868e8aSDimitry Andric   // other wildcards.
3066e75b2fbSDimitry Andric   for (VersionDefinition &v : config->versionDefinitions) {
3076e75b2fbSDimitry Andric     for (SymbolVersion &pat : v.nonLocalPatterns)
30885868e8aSDimitry Andric       if (pat.hasWildcard && pat.name == "*")
3096e75b2fbSDimitry Andric         assignWildcard(pat, v.id, v.name);
3106e75b2fbSDimitry Andric     for (SymbolVersion &pat : v.localPatterns)
3116e75b2fbSDimitry Andric       if (pat.hasWildcard && pat.name == "*")
3126e75b2fbSDimitry Andric         assignWildcard(pat, VER_NDX_LOCAL, v.name);
3136e75b2fbSDimitry Andric   }
3140b57cec5SDimitry Andric 
3150b57cec5SDimitry Andric   // Symbol themselves might know their versions because symbols
3160b57cec5SDimitry Andric   // can contain versions in the form of <name>@<version>.
3170b57cec5SDimitry Andric   // Let them parse and update their names to exclude version suffix.
3180b57cec5SDimitry Andric   for (Symbol *sym : symVector)
3190b57cec5SDimitry Andric     sym->parseSymbolVersion();
3200b57cec5SDimitry Andric 
3210b57cec5SDimitry Andric   // isPreemptible is false at this point. To correctly compute the binding of a
3220b57cec5SDimitry Andric   // Defined (which is used by includeInDynsym()), we need to know if it is
32385868e8aSDimitry Andric   // VER_NDX_LOCAL or not. Compute symbol versions before handling
32485868e8aSDimitry Andric   // --dynamic-list.
3250b57cec5SDimitry Andric   handleDynamicList();
3260b57cec5SDimitry Andric }
327