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" 1881ad6265SDimitry Andric #include "InputFiles.h" 190b57cec5SDimitry Andric #include "Symbols.h" 200b57cec5SDimitry Andric #include "lld/Common/ErrorHandler.h" 210b57cec5SDimitry Andric #include "lld/Common/Memory.h" 220b57cec5SDimitry Andric #include "lld/Common/Strings.h" 230b57cec5SDimitry Andric #include "llvm/ADT/STLExtras.h" 24bdd1243dSDimitry Andric #include "llvm/Demangle/Demangle.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 32bdd1243dSDimitry Andric 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 4381ad6265SDimitry Andric // Propagate symbol usage information to the redirected symbols. 4481ad6265SDimitry Andric if (sym->isUsedInRegularObj) 4581ad6265SDimitry Andric wrap->isUsedInRegularObj = true; 4681ad6265SDimitry Andric if (real->isUsedInRegularObj) 4781ad6265SDimitry Andric sym->isUsedInRegularObj = true; 4881ad6265SDimitry Andric else if (!sym->isDefined()) 4981ad6265SDimitry Andric // Now that all references to sym have been redirected to wrap, if there are 5081ad6265SDimitry Andric // no references to real (which has been redirected to sym), we only need to 5181ad6265SDimitry Andric // keep sym if it was defined, otherwise it's unused and can be dropped. 52e8d8bef9SDimitry Andric sym->isUsedInRegularObj = false; 535ffd83dbSDimitry Andric 545ffd83dbSDimitry Andric // Now renaming is complete, and no one refers to real. We drop real from 555ffd83dbSDimitry Andric // .symtab and .dynsym. If real is undefined, it is important that we don't 565ffd83dbSDimitry Andric // leave it in .dynsym, because otherwise it might lead to an undefined symbol 575ffd83dbSDimitry Andric // error in a subsequent link. If real is defined, we could emit real as an 585ffd83dbSDimitry Andric // alias for sym, but that could degrade the user experience of some tools 595ffd83dbSDimitry Andric // that can print out only one symbol for each location: sym is a preferred 605ffd83dbSDimitry Andric // name than real, but they might print out real instead. 610b57cec5SDimitry Andric memcpy(real, sym, sizeof(SymbolUnion)); 625ffd83dbSDimitry Andric real->isUsedInRegularObj = false; 630b57cec5SDimitry Andric } 640b57cec5SDimitry Andric 650b57cec5SDimitry Andric // Find an existing symbol or create a new one. 660b57cec5SDimitry Andric Symbol *SymbolTable::insert(StringRef name) { 670b57cec5SDimitry Andric // <name>@@<version> means the symbol is the default version. In that 680b57cec5SDimitry Andric // case <name>@@<version> will be used to resolve references to <name>. 690b57cec5SDimitry Andric // 700b57cec5SDimitry Andric // Since this is a hot path, the following string search code is 710b57cec5SDimitry Andric // optimized for speed. StringRef::find(char) is much faster than 720b57cec5SDimitry Andric // StringRef::find(StringRef). 730eae32dcSDimitry Andric StringRef stem = name; 740b57cec5SDimitry Andric size_t pos = name.find('@'); 750b57cec5SDimitry Andric if (pos != StringRef::npos && pos + 1 < name.size() && name[pos + 1] == '@') 760eae32dcSDimitry Andric stem = name.take_front(pos); 770b57cec5SDimitry Andric 780eae32dcSDimitry Andric auto p = symMap.insert({CachedHashStringRef(stem), (int)symVector.size()}); 790eae32dcSDimitry Andric if (!p.second) { 800eae32dcSDimitry Andric Symbol *sym = symVector[p.first->second]; 8104eeddc0SDimitry Andric if (stem.size() != name.size()) { 820eae32dcSDimitry Andric sym->setName(name); 8304eeddc0SDimitry Andric sym->hasVersionSuffix = true; 8404eeddc0SDimitry Andric } 850eae32dcSDimitry Andric return sym; 860eae32dcSDimitry Andric } 870b57cec5SDimitry Andric 880b57cec5SDimitry Andric Symbol *sym = reinterpret_cast<Symbol *>(make<SymbolUnion>()); 890b57cec5SDimitry Andric symVector.push_back(sym); 900b57cec5SDimitry Andric 91bdd1243dSDimitry Andric // *sym was not initialized by a constructor. Initialize all Symbol fields. 92bdd1243dSDimitry Andric memset(sym, 0, sizeof(Symbol)); 930b57cec5SDimitry Andric sym->setName(name); 9481ad6265SDimitry Andric sym->partition = 1; 95bdd1243dSDimitry Andric sym->verdefIndex = -1; 9681ad6265SDimitry Andric sym->versionId = VER_NDX_GLOBAL; 9704eeddc0SDimitry Andric if (pos != StringRef::npos) 9804eeddc0SDimitry Andric sym->hasVersionSuffix = true; 990b57cec5SDimitry Andric return sym; 1000b57cec5SDimitry Andric } 1010b57cec5SDimitry Andric 10281ad6265SDimitry Andric // This variant of addSymbol is used by BinaryFile::parse to check duplicate 10381ad6265SDimitry Andric // symbol errors. 10481ad6265SDimitry Andric Symbol *SymbolTable::addAndCheckDuplicate(const Defined &newSym) { 10581ad6265SDimitry Andric Symbol *sym = insert(newSym.getName()); 10681ad6265SDimitry Andric if (sym->isDefined()) 10781ad6265SDimitry Andric sym->checkDuplicate(newSym); 10881ad6265SDimitry Andric sym->resolve(newSym); 10981ad6265SDimitry Andric sym->isUsedInRegularObj = true; 11081ad6265SDimitry Andric return sym; 11181ad6265SDimitry Andric } 11281ad6265SDimitry Andric 1130b57cec5SDimitry Andric Symbol *SymbolTable::find(StringRef name) { 1140b57cec5SDimitry Andric auto it = symMap.find(CachedHashStringRef(name)); 1150b57cec5SDimitry Andric if (it == symMap.end()) 1160b57cec5SDimitry Andric return nullptr; 1170eae32dcSDimitry Andric return symVector[it->second]; 1180b57cec5SDimitry Andric } 1190b57cec5SDimitry Andric 1205ffd83dbSDimitry Andric // A version script/dynamic list is only meaningful for a Defined symbol. 1215ffd83dbSDimitry Andric // A CommonSymbol will be converted to a Defined in replaceCommonSymbols(). 1224824e7fdSDimitry Andric // A lazy symbol may be made Defined if an LTO libcall extracts it. 1235ffd83dbSDimitry Andric static bool canBeVersioned(const Symbol &sym) { 1245ffd83dbSDimitry Andric return sym.isDefined() || sym.isCommon() || sym.isLazy(); 1255ffd83dbSDimitry Andric } 1265ffd83dbSDimitry Andric 1270b57cec5SDimitry Andric // Initialize demangledSyms with a map from demangled symbols to symbol 1280b57cec5SDimitry Andric // objects. Used to handle "extern C++" directive in version scripts. 1290b57cec5SDimitry Andric // 1300b57cec5SDimitry Andric // The map will contain all demangled symbols. That can be very large, 1310b57cec5SDimitry Andric // and in LLD we generally want to avoid do anything for each symbol. 1320b57cec5SDimitry Andric // Then, why are we doing this? Here's why. 1330b57cec5SDimitry Andric // 1340b57cec5SDimitry Andric // Users can use "extern C++ {}" directive to match against demangled 1350b57cec5SDimitry Andric // C++ symbols. For example, you can write a pattern such as 1360b57cec5SDimitry Andric // "llvm::*::foo(int, ?)". Obviously, there's no way to handle this 1370b57cec5SDimitry Andric // other than trying to match a pattern against all demangled symbols. 1380b57cec5SDimitry Andric // So, if "extern C++" feature is used, we need to demangle all known 1390b57cec5SDimitry Andric // symbols. 1400eae32dcSDimitry Andric StringMap<SmallVector<Symbol *, 0>> &SymbolTable::getDemangledSyms() { 1410b57cec5SDimitry Andric if (!demangledSyms) { 1420b57cec5SDimitry Andric demangledSyms.emplace(); 1436e75b2fbSDimitry Andric std::string demangled; 1445ffd83dbSDimitry Andric for (Symbol *sym : symVector) 1456e75b2fbSDimitry Andric if (canBeVersioned(*sym)) { 1466e75b2fbSDimitry Andric StringRef name = sym->getName(); 1476e75b2fbSDimitry Andric size_t pos = name.find('@'); 148*06c3fb27SDimitry Andric std::string substr; 1496e75b2fbSDimitry Andric if (pos == std::string::npos) 150*06c3fb27SDimitry Andric demangled = demangle(name); 151*06c3fb27SDimitry Andric else if (pos + 1 == name.size() || name[pos + 1] == '@') { 152*06c3fb27SDimitry Andric substr = name.substr(0, pos); 153*06c3fb27SDimitry Andric demangled = demangle(substr); 154*06c3fb27SDimitry Andric } else { 155*06c3fb27SDimitry Andric substr = name.substr(0, pos); 156*06c3fb27SDimitry Andric demangled = (demangle(substr) + name.substr(pos)).str(); 157*06c3fb27SDimitry Andric } 1586e75b2fbSDimitry Andric (*demangledSyms)[demangled].push_back(sym); 1596e75b2fbSDimitry Andric } 1600b57cec5SDimitry Andric } 1610b57cec5SDimitry Andric return *demangledSyms; 1620b57cec5SDimitry Andric } 1630b57cec5SDimitry Andric 1640eae32dcSDimitry Andric SmallVector<Symbol *, 0> SymbolTable::findByVersion(SymbolVersion ver) { 1650b57cec5SDimitry Andric if (ver.isExternCpp) 1660b57cec5SDimitry Andric return getDemangledSyms().lookup(ver.name); 1675ffd83dbSDimitry Andric if (Symbol *sym = find(ver.name)) 1685ffd83dbSDimitry Andric if (canBeVersioned(*sym)) 1695ffd83dbSDimitry Andric return {sym}; 1700b57cec5SDimitry Andric return {}; 1710b57cec5SDimitry Andric } 1720b57cec5SDimitry Andric 1730eae32dcSDimitry Andric SmallVector<Symbol *, 0> SymbolTable::findAllByVersion(SymbolVersion ver, 1746e75b2fbSDimitry Andric bool includeNonDefault) { 1750eae32dcSDimitry Andric SmallVector<Symbol *, 0> res; 1765ffd83dbSDimitry Andric SingleStringMatcher m(ver.name); 177*06c3fb27SDimitry Andric auto check = [&](const Symbol &sym) -> bool { 1786e75b2fbSDimitry Andric if (!includeNonDefault) 179*06c3fb27SDimitry Andric return !sym.hasVersionSuffix; 180*06c3fb27SDimitry Andric StringRef name = sym.getName(); 181*06c3fb27SDimitry Andric size_t pos = name.find('@'); 1826e75b2fbSDimitry Andric return !(pos + 1 < name.size() && name[pos + 1] == '@'); 1836e75b2fbSDimitry Andric }; 1840b57cec5SDimitry Andric 1850b57cec5SDimitry Andric if (ver.isExternCpp) { 1860b57cec5SDimitry Andric for (auto &p : getDemangledSyms()) 1870b57cec5SDimitry Andric if (m.match(p.first())) 1886e75b2fbSDimitry Andric for (Symbol *sym : p.second) 189*06c3fb27SDimitry Andric if (check(*sym)) 1906e75b2fbSDimitry Andric res.push_back(sym); 1910b57cec5SDimitry Andric return res; 1920b57cec5SDimitry Andric } 1930b57cec5SDimitry Andric 1940b57cec5SDimitry Andric for (Symbol *sym : symVector) 195*06c3fb27SDimitry Andric if (canBeVersioned(*sym) && check(*sym) && m.match(sym->getName())) 1960b57cec5SDimitry Andric res.push_back(sym); 1970b57cec5SDimitry Andric return res; 1980b57cec5SDimitry Andric } 1990b57cec5SDimitry Andric 2000b57cec5SDimitry Andric void SymbolTable::handleDynamicList() { 2010eae32dcSDimitry Andric SmallVector<Symbol *, 0> syms; 2020b57cec5SDimitry Andric for (SymbolVersion &ver : config->dynamicList) { 2030b57cec5SDimitry Andric if (ver.hasWildcard) 2046e75b2fbSDimitry Andric syms = findAllByVersion(ver, /*includeNonDefault=*/true); 2050b57cec5SDimitry Andric else 2060b57cec5SDimitry Andric syms = findByVersion(ver); 2070b57cec5SDimitry Andric 20885868e8aSDimitry Andric for (Symbol *sym : syms) 20985868e8aSDimitry Andric sym->inDynamicList = true; 2100b57cec5SDimitry Andric } 2110b57cec5SDimitry Andric } 2120b57cec5SDimitry Andric 2136e75b2fbSDimitry Andric // Set symbol versions to symbols. This function handles patterns containing no 2146e75b2fbSDimitry Andric // wildcard characters. Return false if no symbol definition matches ver. 2156e75b2fbSDimitry Andric bool SymbolTable::assignExactVersion(SymbolVersion ver, uint16_t versionId, 2166e75b2fbSDimitry Andric StringRef versionName, 2176e75b2fbSDimitry Andric bool includeNonDefault) { 2180b57cec5SDimitry Andric // Get a list of symbols which we need to assign the version to. 2190eae32dcSDimitry Andric SmallVector<Symbol *, 0> syms = findByVersion(ver); 2200b57cec5SDimitry Andric 2210b57cec5SDimitry Andric auto getName = [](uint16_t ver) -> std::string { 2220b57cec5SDimitry Andric if (ver == VER_NDX_LOCAL) 2230b57cec5SDimitry Andric return "VER_NDX_LOCAL"; 2240b57cec5SDimitry Andric if (ver == VER_NDX_GLOBAL) 2250b57cec5SDimitry Andric return "VER_NDX_GLOBAL"; 22685868e8aSDimitry Andric return ("version '" + config->versionDefinitions[ver].name + "'").str(); 2270b57cec5SDimitry Andric }; 2280b57cec5SDimitry Andric 2290b57cec5SDimitry Andric // Assign the version. 2300b57cec5SDimitry Andric for (Symbol *sym : syms) { 2316e75b2fbSDimitry Andric // For a non-local versionId, skip symbols containing version info because 2326e75b2fbSDimitry Andric // symbol versions specified by symbol names take precedence over version 2336e75b2fbSDimitry Andric // scripts. See parseSymbolVersion(). 2346e75b2fbSDimitry Andric if (!includeNonDefault && versionId != VER_NDX_LOCAL && 2356e75b2fbSDimitry Andric sym->getName().contains('@')) 2360b57cec5SDimitry Andric continue; 2370b57cec5SDimitry Andric 23885868e8aSDimitry Andric // If the version has not been assigned, verdefIndex is -1. Use an arbitrary 23985868e8aSDimitry Andric // number (0) to indicate the version has been assigned. 2400eae32dcSDimitry Andric if (sym->verdefIndex == uint16_t(-1)) { 24185868e8aSDimitry Andric sym->verdefIndex = 0; 2420b57cec5SDimitry Andric sym->versionId = versionId; 24385868e8aSDimitry Andric } 2440b57cec5SDimitry Andric if (sym->versionId == versionId) 2450b57cec5SDimitry Andric continue; 2460b57cec5SDimitry Andric 2470b57cec5SDimitry Andric warn("attempt to reassign symbol '" + ver.name + "' of " + 2480b57cec5SDimitry Andric getName(sym->versionId) + " to " + getName(versionId)); 2490b57cec5SDimitry Andric } 2506e75b2fbSDimitry Andric return !syms.empty(); 2510b57cec5SDimitry Andric } 2520b57cec5SDimitry Andric 2536e75b2fbSDimitry Andric void SymbolTable::assignWildcardVersion(SymbolVersion ver, uint16_t versionId, 2546e75b2fbSDimitry Andric bool includeNonDefault) { 255480093f4SDimitry Andric // Exact matching takes precedence over fuzzy matching, 2560b57cec5SDimitry Andric // so we set a version to a symbol only if no version has been assigned 2570b57cec5SDimitry Andric // to the symbol. This behavior is compatible with GNU. 2586e75b2fbSDimitry Andric for (Symbol *sym : findAllByVersion(ver, includeNonDefault)) 2590eae32dcSDimitry Andric if (sym->verdefIndex == uint16_t(-1)) { 26085868e8aSDimitry Andric sym->verdefIndex = 0; 26185868e8aSDimitry Andric sym->versionId = versionId; 26285868e8aSDimitry Andric } 2630b57cec5SDimitry Andric } 2640b57cec5SDimitry Andric 2650b57cec5SDimitry Andric // This function processes version scripts by updating the versionId 2660b57cec5SDimitry Andric // member of symbols. 2670b57cec5SDimitry Andric // If there's only one anonymous version definition in a version 2680b57cec5SDimitry Andric // script file, the script does not actually define any symbol version, 2690b57cec5SDimitry Andric // but just specifies symbols visibilities. 2700b57cec5SDimitry Andric void SymbolTable::scanVersionScript() { 2716e75b2fbSDimitry Andric SmallString<128> buf; 2720b57cec5SDimitry Andric // First, we assign versions to exact matching symbols, 2730b57cec5SDimitry Andric // i.e. version definitions not containing any glob meta-characters. 2746e75b2fbSDimitry Andric for (VersionDefinition &v : config->versionDefinitions) { 2756e75b2fbSDimitry Andric auto assignExact = [&](SymbolVersion pat, uint16_t id, StringRef ver) { 2766e75b2fbSDimitry Andric bool found = 2776e75b2fbSDimitry Andric assignExactVersion(pat, id, ver, /*includeNonDefault=*/false); 2786e75b2fbSDimitry Andric buf.clear(); 2796e75b2fbSDimitry Andric found |= assignExactVersion({(pat.name + "@" + v.name).toStringRef(buf), 2806e75b2fbSDimitry Andric pat.isExternCpp, /*hasWildCard=*/false}, 2816e75b2fbSDimitry Andric id, ver, /*includeNonDefault=*/true); 2826e75b2fbSDimitry Andric if (!found && !config->undefinedVersion) 283*06c3fb27SDimitry Andric errorOrWarn("version script assignment of '" + ver + "' to symbol '" + 2846e75b2fbSDimitry Andric pat.name + "' failed: symbol not defined"); 2856e75b2fbSDimitry Andric }; 2866e75b2fbSDimitry Andric for (SymbolVersion &pat : v.nonLocalPatterns) 2876e75b2fbSDimitry Andric if (!pat.hasWildcard) 2886e75b2fbSDimitry Andric assignExact(pat, v.id, v.name); 2896e75b2fbSDimitry Andric for (SymbolVersion pat : v.localPatterns) 2906e75b2fbSDimitry Andric if (!pat.hasWildcard) 2916e75b2fbSDimitry Andric assignExact(pat, VER_NDX_LOCAL, "local"); 2926e75b2fbSDimitry Andric } 2930b57cec5SDimitry Andric 29485868e8aSDimitry Andric // Next, assign versions to wildcards that are not "*". Note that because the 29585868e8aSDimitry Andric // last match takes precedence over previous matches, we iterate over the 29685868e8aSDimitry Andric // definitions in the reverse order. 2976e75b2fbSDimitry Andric auto assignWildcard = [&](SymbolVersion pat, uint16_t id, StringRef ver) { 2986e75b2fbSDimitry Andric assignWildcardVersion(pat, id, /*includeNonDefault=*/false); 2996e75b2fbSDimitry Andric buf.clear(); 3006e75b2fbSDimitry Andric assignWildcardVersion({(pat.name + "@" + ver).toStringRef(buf), 3016e75b2fbSDimitry Andric pat.isExternCpp, /*hasWildCard=*/true}, 3026e75b2fbSDimitry Andric id, 3036e75b2fbSDimitry Andric /*includeNonDefault=*/true); 3046e75b2fbSDimitry Andric }; 3056e75b2fbSDimitry Andric for (VersionDefinition &v : llvm::reverse(config->versionDefinitions)) { 3066e75b2fbSDimitry Andric for (SymbolVersion &pat : v.nonLocalPatterns) 30785868e8aSDimitry Andric if (pat.hasWildcard && pat.name != "*") 3086e75b2fbSDimitry Andric assignWildcard(pat, v.id, v.name); 3096e75b2fbSDimitry Andric for (SymbolVersion &pat : v.localPatterns) 3106e75b2fbSDimitry Andric if (pat.hasWildcard && pat.name != "*") 3116e75b2fbSDimitry Andric assignWildcard(pat, VER_NDX_LOCAL, v.name); 3126e75b2fbSDimitry Andric } 31385868e8aSDimitry Andric 31485868e8aSDimitry Andric // Then, assign versions to "*". In GNU linkers they have lower priority than 31585868e8aSDimitry Andric // other wildcards. 3166e75b2fbSDimitry Andric for (VersionDefinition &v : config->versionDefinitions) { 3176e75b2fbSDimitry Andric for (SymbolVersion &pat : v.nonLocalPatterns) 31885868e8aSDimitry Andric if (pat.hasWildcard && pat.name == "*") 3196e75b2fbSDimitry Andric assignWildcard(pat, v.id, v.name); 3206e75b2fbSDimitry Andric for (SymbolVersion &pat : v.localPatterns) 3216e75b2fbSDimitry Andric if (pat.hasWildcard && pat.name == "*") 3226e75b2fbSDimitry Andric assignWildcard(pat, VER_NDX_LOCAL, v.name); 3236e75b2fbSDimitry Andric } 3240b57cec5SDimitry Andric 3250b57cec5SDimitry Andric // Symbol themselves might know their versions because symbols 3260b57cec5SDimitry Andric // can contain versions in the form of <name>@<version>. 3270b57cec5SDimitry Andric // Let them parse and update their names to exclude version suffix. 3280b57cec5SDimitry Andric for (Symbol *sym : symVector) 32904eeddc0SDimitry Andric if (sym->hasVersionSuffix) 3300b57cec5SDimitry Andric sym->parseSymbolVersion(); 3310b57cec5SDimitry Andric 3320b57cec5SDimitry Andric // isPreemptible is false at this point. To correctly compute the binding of a 3330b57cec5SDimitry Andric // Defined (which is used by includeInDynsym()), we need to know if it is 33485868e8aSDimitry Andric // VER_NDX_LOCAL or not. Compute symbol versions before handling 33585868e8aSDimitry Andric // --dynamic-list. 3360b57cec5SDimitry Andric handleDynamicList(); 3370b57cec5SDimitry Andric } 338