1 //===- IRSymtab.cpp - implementation of IR symbol tables ------------------===// 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 #include "llvm/Object/IRSymtab.h" 10 #include "llvm/ADT/ArrayRef.h" 11 #include "llvm/ADT/DenseMap.h" 12 #include "llvm/ADT/SmallPtrSet.h" 13 #include "llvm/ADT/SmallString.h" 14 #include "llvm/ADT/SmallVector.h" 15 #include "llvm/ADT/StringRef.h" 16 #include "llvm/ADT/Triple.h" 17 #include "llvm/Bitcode/BitcodeReader.h" 18 #include "llvm/Config/llvm-config.h" 19 #include "llvm/IR/Comdat.h" 20 #include "llvm/IR/DataLayout.h" 21 #include "llvm/IR/GlobalAlias.h" 22 #include "llvm/IR/GlobalObject.h" 23 #include "llvm/IR/Mangler.h" 24 #include "llvm/IR/Metadata.h" 25 #include "llvm/IR/Module.h" 26 #include "llvm/MC/StringTableBuilder.h" 27 #include "llvm/Object/IRObjectFile.h" 28 #include "llvm/Object/ModuleSymbolTable.h" 29 #include "llvm/Object/SymbolicFile.h" 30 #include "llvm/Support/Allocator.h" 31 #include "llvm/Support/Casting.h" 32 #include "llvm/Support/CommandLine.h" 33 #include "llvm/Support/Error.h" 34 #include "llvm/Support/StringSaver.h" 35 #include "llvm/Support/VCSRevision.h" 36 #include "llvm/Support/raw_ostream.h" 37 #include <cassert> 38 #include <string> 39 #include <utility> 40 #include <vector> 41 42 using namespace llvm; 43 using namespace irsymtab; 44 45 cl::opt<bool> DisableBitcodeVersionUpgrade( 46 "disable-bitcode-version-upgrade", cl::init(false), cl::Hidden, 47 cl::desc("Disable automatic bitcode upgrade for version mismatch")); 48 49 static const char *PreservedSymbols[] = { 50 #define HANDLE_LIBCALL(code, name) name, 51 #include "llvm/IR/RuntimeLibcalls.def" 52 #undef HANDLE_LIBCALL 53 // There are global variables, so put it here instead of in 54 // RuntimeLibcalls.def. 55 // TODO: Are there similar such variables? 56 "__ssp_canary_word", 57 "__stack_chk_guard", 58 }; 59 60 namespace { 61 62 const char *getExpectedProducerName() { 63 static char DefaultName[] = LLVM_VERSION_STRING 64 #ifdef LLVM_REVISION 65 " " LLVM_REVISION 66 #endif 67 ; 68 // Allows for testing of the irsymtab writer and upgrade mechanism. This 69 // environment variable should not be set by users. 70 if (char *OverrideName = getenv("LLVM_OVERRIDE_PRODUCER")) 71 return OverrideName; 72 return DefaultName; 73 } 74 75 const char *kExpectedProducerName = getExpectedProducerName(); 76 77 /// Stores the temporary state that is required to build an IR symbol table. 78 struct Builder { 79 SmallVector<char, 0> &Symtab; 80 StringTableBuilder &StrtabBuilder; 81 StringSaver Saver; 82 83 // This ctor initializes a StringSaver using the passed in BumpPtrAllocator. 84 // The StringTableBuilder does not create a copy of any strings added to it, 85 // so this provides somewhere to store any strings that we create. 86 Builder(SmallVector<char, 0> &Symtab, StringTableBuilder &StrtabBuilder, 87 BumpPtrAllocator &Alloc) 88 : Symtab(Symtab), StrtabBuilder(StrtabBuilder), Saver(Alloc) {} 89 90 DenseMap<const Comdat *, int> ComdatMap; 91 Mangler Mang; 92 Triple TT; 93 94 std::vector<storage::Comdat> Comdats; 95 std::vector<storage::Module> Mods; 96 std::vector<storage::Symbol> Syms; 97 std::vector<storage::Uncommon> Uncommons; 98 99 std::string COFFLinkerOpts; 100 raw_string_ostream COFFLinkerOptsOS{COFFLinkerOpts}; 101 102 std::vector<storage::Str> DependentLibraries; 103 104 void setStr(storage::Str &S, StringRef Value) { 105 S.Offset = StrtabBuilder.add(Value); 106 S.Size = Value.size(); 107 } 108 109 template <typename T> 110 void writeRange(storage::Range<T> &R, const std::vector<T> &Objs) { 111 R.Offset = Symtab.size(); 112 R.Size = Objs.size(); 113 Symtab.insert(Symtab.end(), reinterpret_cast<const char *>(Objs.data()), 114 reinterpret_cast<const char *>(Objs.data() + Objs.size())); 115 } 116 117 Expected<int> getComdatIndex(const Comdat *C, const Module *M); 118 119 Error addModule(Module *M); 120 Error addSymbol(const ModuleSymbolTable &Msymtab, 121 const SmallPtrSet<GlobalValue *, 4> &Used, 122 ModuleSymbolTable::Symbol Sym); 123 124 Error build(ArrayRef<Module *> Mods); 125 }; 126 127 Error Builder::addModule(Module *M) { 128 if (M->getDataLayoutStr().empty()) 129 return make_error<StringError>("input module has no datalayout", 130 inconvertibleErrorCode()); 131 132 // Symbols in the llvm.used list will get the FB_Used bit and will not be 133 // internalized. We do this for llvm.compiler.used as well: 134 // 135 // IR symbol table tracks module-level asm symbol references but not inline 136 // asm. A symbol only referenced by inline asm is not in the IR symbol table, 137 // so we may not know that the definition (in another translation unit) is 138 // referenced. That definition may have __attribute__((used)) (which lowers to 139 // llvm.compiler.used on ELF targets) to communicate to the compiler that it 140 // may be used by inline asm. The usage is perfectly fine, so we treat 141 // llvm.compiler.used conservatively as llvm.used to work around our own 142 // limitation. 143 SmallVector<GlobalValue *, 4> UsedV; 144 collectUsedGlobalVariables(*M, UsedV, /*CompilerUsed=*/false); 145 collectUsedGlobalVariables(*M, UsedV, /*CompilerUsed=*/true); 146 SmallPtrSet<GlobalValue *, 4> Used(UsedV.begin(), UsedV.end()); 147 148 ModuleSymbolTable Msymtab; 149 Msymtab.addModule(M); 150 151 storage::Module Mod; 152 Mod.Begin = Syms.size(); 153 Mod.End = Syms.size() + Msymtab.symbols().size(); 154 Mod.UncBegin = Uncommons.size(); 155 Mods.push_back(Mod); 156 157 if (TT.isOSBinFormatCOFF()) { 158 if (auto E = M->materializeMetadata()) 159 return E; 160 if (NamedMDNode *LinkerOptions = 161 M->getNamedMetadata("llvm.linker.options")) { 162 for (MDNode *MDOptions : LinkerOptions->operands()) 163 for (const MDOperand &MDOption : cast<MDNode>(MDOptions)->operands()) 164 COFFLinkerOptsOS << " " << cast<MDString>(MDOption)->getString(); 165 } 166 } 167 168 if (TT.isOSBinFormatELF()) { 169 if (auto E = M->materializeMetadata()) 170 return E; 171 if (NamedMDNode *N = M->getNamedMetadata("llvm.dependent-libraries")) { 172 for (MDNode *MDOptions : N->operands()) { 173 const auto OperandStr = 174 cast<MDString>(cast<MDNode>(MDOptions)->getOperand(0))->getString(); 175 storage::Str Specifier; 176 setStr(Specifier, OperandStr); 177 DependentLibraries.emplace_back(Specifier); 178 } 179 } 180 } 181 182 for (ModuleSymbolTable::Symbol Msym : Msymtab.symbols()) 183 if (Error Err = addSymbol(Msymtab, Used, Msym)) 184 return Err; 185 186 return Error::success(); 187 } 188 189 Expected<int> Builder::getComdatIndex(const Comdat *C, const Module *M) { 190 auto P = ComdatMap.insert(std::make_pair(C, Comdats.size())); 191 if (P.second) { 192 std::string Name; 193 if (TT.isOSBinFormatCOFF()) { 194 const GlobalValue *GV = M->getNamedValue(C->getName()); 195 if (!GV) 196 return make_error<StringError>("Could not find leader", 197 inconvertibleErrorCode()); 198 // Internal leaders do not affect symbol resolution, therefore they do not 199 // appear in the symbol table. 200 if (GV->hasLocalLinkage()) { 201 P.first->second = -1; 202 return -1; 203 } 204 llvm::raw_string_ostream OS(Name); 205 Mang.getNameWithPrefix(OS, GV, false); 206 } else { 207 Name = std::string(C->getName()); 208 } 209 210 storage::Comdat Comdat; 211 setStr(Comdat.Name, Saver.save(Name)); 212 Comdat.SelectionKind = C->getSelectionKind(); 213 Comdats.push_back(Comdat); 214 } 215 216 return P.first->second; 217 } 218 219 Error Builder::addSymbol(const ModuleSymbolTable &Msymtab, 220 const SmallPtrSet<GlobalValue *, 4> &Used, 221 ModuleSymbolTable::Symbol Msym) { 222 Syms.emplace_back(); 223 storage::Symbol &Sym = Syms.back(); 224 Sym = {}; 225 226 storage::Uncommon *Unc = nullptr; 227 auto Uncommon = [&]() -> storage::Uncommon & { 228 if (Unc) 229 return *Unc; 230 Sym.Flags |= 1 << storage::Symbol::FB_has_uncommon; 231 Uncommons.emplace_back(); 232 Unc = &Uncommons.back(); 233 *Unc = {}; 234 setStr(Unc->COFFWeakExternFallbackName, ""); 235 setStr(Unc->SectionName, ""); 236 return *Unc; 237 }; 238 239 SmallString<64> Name; 240 { 241 raw_svector_ostream OS(Name); 242 Msymtab.printSymbolName(OS, Msym); 243 } 244 setStr(Sym.Name, Saver.save(Name.str())); 245 246 auto Flags = Msymtab.getSymbolFlags(Msym); 247 if (Flags & object::BasicSymbolRef::SF_Undefined) 248 Sym.Flags |= 1 << storage::Symbol::FB_undefined; 249 if (Flags & object::BasicSymbolRef::SF_Weak) 250 Sym.Flags |= 1 << storage::Symbol::FB_weak; 251 if (Flags & object::BasicSymbolRef::SF_Common) 252 Sym.Flags |= 1 << storage::Symbol::FB_common; 253 if (Flags & object::BasicSymbolRef::SF_Indirect) 254 Sym.Flags |= 1 << storage::Symbol::FB_indirect; 255 if (Flags & object::BasicSymbolRef::SF_Global) 256 Sym.Flags |= 1 << storage::Symbol::FB_global; 257 if (Flags & object::BasicSymbolRef::SF_FormatSpecific) 258 Sym.Flags |= 1 << storage::Symbol::FB_format_specific; 259 if (Flags & object::BasicSymbolRef::SF_Executable) 260 Sym.Flags |= 1 << storage::Symbol::FB_executable; 261 262 Sym.ComdatIndex = -1; 263 auto *GV = Msym.dyn_cast<GlobalValue *>(); 264 if (!GV) { 265 // Undefined module asm symbols act as GC roots and are implicitly used. 266 if (Flags & object::BasicSymbolRef::SF_Undefined) 267 Sym.Flags |= 1 << storage::Symbol::FB_used; 268 setStr(Sym.IRName, ""); 269 return Error::success(); 270 } 271 272 setStr(Sym.IRName, GV->getName()); 273 274 bool IsPreservedSymbol = llvm::is_contained(PreservedSymbols, GV->getName()); 275 276 if (Used.count(GV) || IsPreservedSymbol) 277 Sym.Flags |= 1 << storage::Symbol::FB_used; 278 if (GV->isThreadLocal()) 279 Sym.Flags |= 1 << storage::Symbol::FB_tls; 280 if (GV->hasGlobalUnnamedAddr()) 281 Sym.Flags |= 1 << storage::Symbol::FB_unnamed_addr; 282 if (GV->canBeOmittedFromSymbolTable()) 283 Sym.Flags |= 1 << storage::Symbol::FB_may_omit; 284 Sym.Flags |= unsigned(GV->getVisibility()) << storage::Symbol::FB_visibility; 285 286 if (Flags & object::BasicSymbolRef::SF_Common) { 287 auto *GVar = dyn_cast<GlobalVariable>(GV); 288 if (!GVar) 289 return make_error<StringError>("Only variables can have common linkage!", 290 inconvertibleErrorCode()); 291 Uncommon().CommonSize = 292 GV->getParent()->getDataLayout().getTypeAllocSize(GV->getValueType()); 293 Uncommon().CommonAlign = GVar->getAlignment(); 294 } 295 296 const GlobalObject *GO = GV->getAliaseeObject(); 297 if (!GO) { 298 if (isa<GlobalIFunc>(GV)) 299 GO = cast<GlobalIFunc>(GV)->getResolverFunction(); 300 if (!GO) 301 return make_error<StringError>("Unable to determine comdat of alias!", 302 inconvertibleErrorCode()); 303 } 304 if (const Comdat *C = GO->getComdat()) { 305 Expected<int> ComdatIndexOrErr = getComdatIndex(C, GV->getParent()); 306 if (!ComdatIndexOrErr) 307 return ComdatIndexOrErr.takeError(); 308 Sym.ComdatIndex = *ComdatIndexOrErr; 309 } 310 311 if (TT.isOSBinFormatCOFF()) { 312 emitLinkerFlagsForGlobalCOFF(COFFLinkerOptsOS, GV, TT, Mang); 313 314 if ((Flags & object::BasicSymbolRef::SF_Weak) && 315 (Flags & object::BasicSymbolRef::SF_Indirect)) { 316 auto *Fallback = dyn_cast<GlobalValue>( 317 cast<GlobalAlias>(GV)->getAliasee()->stripPointerCasts()); 318 if (!Fallback) 319 return make_error<StringError>("Invalid weak external", 320 inconvertibleErrorCode()); 321 std::string FallbackName; 322 raw_string_ostream OS(FallbackName); 323 Msymtab.printSymbolName(OS, Fallback); 324 OS.flush(); 325 setStr(Uncommon().COFFWeakExternFallbackName, Saver.save(FallbackName)); 326 } 327 } 328 329 if (!GO->getSection().empty()) 330 setStr(Uncommon().SectionName, Saver.save(GO->getSection())); 331 332 return Error::success(); 333 } 334 335 Error Builder::build(ArrayRef<Module *> IRMods) { 336 storage::Header Hdr; 337 338 assert(!IRMods.empty()); 339 Hdr.Version = storage::Header::kCurrentVersion; 340 setStr(Hdr.Producer, kExpectedProducerName); 341 setStr(Hdr.TargetTriple, IRMods[0]->getTargetTriple()); 342 setStr(Hdr.SourceFileName, IRMods[0]->getSourceFileName()); 343 TT = Triple(IRMods[0]->getTargetTriple()); 344 345 for (auto *M : IRMods) 346 if (Error Err = addModule(M)) 347 return Err; 348 349 COFFLinkerOptsOS.flush(); 350 setStr(Hdr.COFFLinkerOpts, Saver.save(COFFLinkerOpts)); 351 352 // We are about to fill in the header's range fields, so reserve space for it 353 // and copy it in afterwards. 354 Symtab.resize(sizeof(storage::Header)); 355 writeRange(Hdr.Modules, Mods); 356 writeRange(Hdr.Comdats, Comdats); 357 writeRange(Hdr.Symbols, Syms); 358 writeRange(Hdr.Uncommons, Uncommons); 359 writeRange(Hdr.DependentLibraries, DependentLibraries); 360 *reinterpret_cast<storage::Header *>(Symtab.data()) = Hdr; 361 return Error::success(); 362 } 363 364 } // end anonymous namespace 365 366 Error irsymtab::build(ArrayRef<Module *> Mods, SmallVector<char, 0> &Symtab, 367 StringTableBuilder &StrtabBuilder, 368 BumpPtrAllocator &Alloc) { 369 return Builder(Symtab, StrtabBuilder, Alloc).build(Mods); 370 } 371 372 // Upgrade a vector of bitcode modules created by an old version of LLVM by 373 // creating an irsymtab for them in the current format. 374 static Expected<FileContents> upgrade(ArrayRef<BitcodeModule> BMs) { 375 FileContents FC; 376 377 LLVMContext Ctx; 378 std::vector<Module *> Mods; 379 std::vector<std::unique_ptr<Module>> OwnedMods; 380 for (auto BM : BMs) { 381 Expected<std::unique_ptr<Module>> MOrErr = 382 BM.getLazyModule(Ctx, /*ShouldLazyLoadMetadata*/ true, 383 /*IsImporting*/ false); 384 if (!MOrErr) 385 return MOrErr.takeError(); 386 387 Mods.push_back(MOrErr->get()); 388 OwnedMods.push_back(std::move(*MOrErr)); 389 } 390 391 StringTableBuilder StrtabBuilder(StringTableBuilder::RAW); 392 BumpPtrAllocator Alloc; 393 if (Error E = build(Mods, FC.Symtab, StrtabBuilder, Alloc)) 394 return std::move(E); 395 396 StrtabBuilder.finalizeInOrder(); 397 FC.Strtab.resize(StrtabBuilder.getSize()); 398 StrtabBuilder.write((uint8_t *)FC.Strtab.data()); 399 400 FC.TheReader = {{FC.Symtab.data(), FC.Symtab.size()}, 401 {FC.Strtab.data(), FC.Strtab.size()}}; 402 return std::move(FC); 403 } 404 405 Expected<FileContents> irsymtab::readBitcode(const BitcodeFileContents &BFC) { 406 if (BFC.Mods.empty()) 407 return make_error<StringError>("Bitcode file does not contain any modules", 408 inconvertibleErrorCode()); 409 410 if (!DisableBitcodeVersionUpgrade) { 411 if (BFC.StrtabForSymtab.empty() || 412 BFC.Symtab.size() < sizeof(storage::Header)) 413 return upgrade(BFC.Mods); 414 415 // We cannot use the regular reader to read the version and producer, 416 // because it will expect the header to be in the current format. The only 417 // thing we can rely on is that the version and producer will be present as 418 // the first struct elements. 419 auto *Hdr = reinterpret_cast<const storage::Header *>(BFC.Symtab.data()); 420 unsigned Version = Hdr->Version; 421 StringRef Producer = Hdr->Producer.get(BFC.StrtabForSymtab); 422 if (Version != storage::Header::kCurrentVersion || 423 Producer != kExpectedProducerName) 424 return upgrade(BFC.Mods); 425 } 426 427 FileContents FC; 428 FC.TheReader = {{BFC.Symtab.data(), BFC.Symtab.size()}, 429 {BFC.StrtabForSymtab.data(), BFC.StrtabForSymtab.size()}}; 430 431 // Finally, make sure that the number of modules in the symbol table matches 432 // the number of modules in the bitcode file. If they differ, it may mean that 433 // the bitcode file was created by binary concatenation, so we need to create 434 // a new symbol table from scratch. 435 if (FC.TheReader.getNumModules() != BFC.Mods.size()) 436 return upgrade(std::move(BFC.Mods)); 437 438 return std::move(FC); 439 } 440