1 //===- InputFiles.cpp -----------------------------------------------------===// 2 // 3 // The LLVM Linker 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 10 #include "InputFiles.h" 11 #include "InputSection.h" 12 #include "LinkerScript.h" 13 #include "SymbolTable.h" 14 #include "Symbols.h" 15 #include "SyntheticSections.h" 16 #include "lld/Common/ErrorHandler.h" 17 #include "lld/Common/Memory.h" 18 #include "llvm/ADT/STLExtras.h" 19 #include "llvm/CodeGen/Analysis.h" 20 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 21 #include "llvm/IR/LLVMContext.h" 22 #include "llvm/IR/Module.h" 23 #include "llvm/LTO/LTO.h" 24 #include "llvm/MC/StringTableBuilder.h" 25 #include "llvm/Object/ELFObjectFile.h" 26 #include "llvm/Support/ARMAttributeParser.h" 27 #include "llvm/Support/ARMBuildAttributes.h" 28 #include "llvm/Support/Path.h" 29 #include "llvm/Support/TarWriter.h" 30 #include "llvm/Support/raw_ostream.h" 31 32 using namespace llvm; 33 using namespace llvm::ELF; 34 using namespace llvm::object; 35 using namespace llvm::sys; 36 using namespace llvm::sys::fs; 37 38 using namespace lld; 39 using namespace lld::elf; 40 41 std::vector<BinaryFile *> elf::BinaryFiles; 42 std::vector<BitcodeFile *> elf::BitcodeFiles; 43 std::vector<InputFile *> elf::ObjectFiles; 44 std::vector<InputFile *> elf::SharedFiles; 45 46 TarWriter *elf::Tar; 47 48 InputFile::InputFile(Kind K, MemoryBufferRef M) : MB(M), FileKind(K) {} 49 50 Optional<MemoryBufferRef> elf::readFile(StringRef Path) { 51 // The --chroot option changes our virtual root directory. 52 // This is useful when you are dealing with files created by --reproduce. 53 if (!Config->Chroot.empty() && Path.startswith("/")) 54 Path = Saver.save(Config->Chroot + Path); 55 56 log(Path); 57 58 auto MBOrErr = MemoryBuffer::getFile(Path); 59 if (auto EC = MBOrErr.getError()) { 60 error("cannot open " + Path + ": " + EC.message()); 61 return None; 62 } 63 64 std::unique_ptr<MemoryBuffer> &MB = *MBOrErr; 65 MemoryBufferRef MBRef = MB->getMemBufferRef(); 66 make<std::unique_ptr<MemoryBuffer>>(std::move(MB)); // take MB ownership 67 68 if (Tar) 69 Tar->append(relativeToRoot(Path), MBRef.getBuffer()); 70 return MBRef; 71 } 72 73 // Concatenates arguments to construct a string representing an error location. 74 static std::string createFileLineMsg(StringRef Path, unsigned Line) { 75 std::string Filename = path::filename(Path); 76 std::string Lineno = ":" + std::to_string(Line); 77 if (Filename == Path) 78 return Filename + Lineno; 79 return Filename + Lineno + " (" + Path.str() + Lineno + ")"; 80 } 81 82 template <class ELFT> 83 static std::string getSrcMsgAux(ObjFile<ELFT> &File, const Symbol &Sym, 84 InputSectionBase &Sec, uint64_t Offset) { 85 // In DWARF, functions and variables are stored to different places. 86 // First, lookup a function for a given offset. 87 if (Optional<DILineInfo> Info = File.getDILineInfo(&Sec, Offset)) 88 return createFileLineMsg(Info->FileName, Info->Line); 89 90 // If it failed, lookup again as a variable. 91 if (Optional<std::pair<std::string, unsigned>> FileLine = 92 File.getVariableLoc(Sym.getName())) 93 return createFileLineMsg(FileLine->first, FileLine->second); 94 95 // File.SourceFile contains STT_FILE symbol, and that is a last resort. 96 return File.SourceFile; 97 } 98 99 std::string InputFile::getSrcMsg(const Symbol &Sym, InputSectionBase &Sec, 100 uint64_t Offset) { 101 if (kind() != ObjKind) 102 return ""; 103 switch (Config->EKind) { 104 default: 105 llvm_unreachable("Invalid kind"); 106 case ELF32LEKind: 107 return getSrcMsgAux(cast<ObjFile<ELF32LE>>(*this), Sym, Sec, Offset); 108 case ELF32BEKind: 109 return getSrcMsgAux(cast<ObjFile<ELF32BE>>(*this), Sym, Sec, Offset); 110 case ELF64LEKind: 111 return getSrcMsgAux(cast<ObjFile<ELF64LE>>(*this), Sym, Sec, Offset); 112 case ELF64BEKind: 113 return getSrcMsgAux(cast<ObjFile<ELF64BE>>(*this), Sym, Sec, Offset); 114 } 115 } 116 117 template <class ELFT> void ObjFile<ELFT>::initializeDwarf() { 118 Dwarf = llvm::make_unique<DWARFContext>(make_unique<LLDDwarfObj<ELFT>>(this)); 119 const DWARFObject &Obj = Dwarf->getDWARFObj(); 120 DwarfLine.reset(new DWARFDebugLine); 121 DWARFDataExtractor LineData(Obj, Obj.getLineSection(), Config->IsLE, 122 Config->Wordsize); 123 124 for (std::unique_ptr<DWARFCompileUnit> &CU : Dwarf->compile_units()) { 125 const DWARFDebugLine::LineTable *LT = Dwarf->getLineTableForUnit(CU.get()); 126 if (!LT) 127 continue; 128 LineTables.push_back(LT); 129 130 // Loop over variable records and insert them to VariableLoc. 131 for (const auto &Entry : CU->dies()) { 132 DWARFDie Die(CU.get(), &Entry); 133 // Skip all tags that are not variables. 134 if (Die.getTag() != dwarf::DW_TAG_variable) 135 continue; 136 137 // Skip if a local variable because we don't need them for generating 138 // error messages. In general, only non-local symbols can fail to be 139 // linked. 140 if (!dwarf::toUnsigned(Die.find(dwarf::DW_AT_external), 0)) 141 continue; 142 143 // Get the source filename index for the variable. 144 unsigned File = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_file), 0); 145 if (!LT->hasFileAtIndex(File)) 146 continue; 147 148 // Get the line number on which the variable is declared. 149 unsigned Line = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_line), 0); 150 151 // Get the name of the variable and add the collected information to 152 // VariableLoc. Usually Name is non-empty, but it can be empty if the 153 // input object file lacks some debug info. 154 StringRef Name = dwarf::toString(Die.find(dwarf::DW_AT_name), ""); 155 if (!Name.empty()) 156 VariableLoc.insert({Name, {LT, File, Line}}); 157 } 158 } 159 } 160 161 // Returns the pair of file name and line number describing location of data 162 // object (variable, array, etc) definition. 163 template <class ELFT> 164 Optional<std::pair<std::string, unsigned>> 165 ObjFile<ELFT>::getVariableLoc(StringRef Name) { 166 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 167 168 // Return if we have no debug information about data object. 169 auto It = VariableLoc.find(Name); 170 if (It == VariableLoc.end()) 171 return None; 172 173 // Take file name string from line table. 174 std::string FileName; 175 if (!It->second.LT->getFileNameByIndex( 176 It->second.File, nullptr, 177 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FileName)) 178 return None; 179 180 return std::make_pair(FileName, It->second.Line); 181 } 182 183 // Returns source line information for a given offset 184 // using DWARF debug info. 185 template <class ELFT> 186 Optional<DILineInfo> ObjFile<ELFT>::getDILineInfo(InputSectionBase *S, 187 uint64_t Offset) { 188 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 189 190 // Use fake address calcuated by adding section file offset and offset in 191 // section. See comments for ObjectInfo class. 192 DILineInfo Info; 193 for (const llvm::DWARFDebugLine::LineTable *LT : LineTables) 194 if (LT->getFileLineInfoForAddress( 195 S->getOffsetInFile() + Offset, nullptr, 196 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, Info)) 197 return Info; 198 return None; 199 } 200 201 // Returns source line information for a given offset using DWARF debug info. 202 template <class ELFT> 203 std::string ObjFile<ELFT>::getLineInfo(InputSectionBase *S, uint64_t Offset) { 204 if (Optional<DILineInfo> Info = getDILineInfo(S, Offset)) 205 return Info->FileName + ":" + std::to_string(Info->Line); 206 return ""; 207 } 208 209 // Returns "<internal>", "foo.a(bar.o)" or "baz.o". 210 std::string lld::toString(const InputFile *F) { 211 if (!F) 212 return "<internal>"; 213 214 if (F->ToStringCache.empty()) { 215 if (F->ArchiveName.empty()) 216 F->ToStringCache = F->getName(); 217 else 218 F->ToStringCache = (F->ArchiveName + "(" + F->getName() + ")").str(); 219 } 220 return F->ToStringCache; 221 } 222 223 template <class ELFT> 224 ELFFileBase<ELFT>::ELFFileBase(Kind K, MemoryBufferRef MB) : InputFile(K, MB) { 225 if (ELFT::TargetEndianness == support::little) 226 EKind = ELFT::Is64Bits ? ELF64LEKind : ELF32LEKind; 227 else 228 EKind = ELFT::Is64Bits ? ELF64BEKind : ELF32BEKind; 229 230 EMachine = getObj().getHeader()->e_machine; 231 OSABI = getObj().getHeader()->e_ident[llvm::ELF::EI_OSABI]; 232 } 233 234 template <class ELFT> 235 typename ELFT::SymRange ELFFileBase<ELFT>::getGlobalELFSyms() { 236 return makeArrayRef(ELFSyms.begin() + FirstNonLocal, ELFSyms.end()); 237 } 238 239 template <class ELFT> 240 uint32_t ELFFileBase<ELFT>::getSectionIndex(const Elf_Sym &Sym) const { 241 return CHECK(getObj().getSectionIndex(&Sym, ELFSyms, SymtabSHNDX), this); 242 } 243 244 template <class ELFT> 245 void ELFFileBase<ELFT>::initSymtab(ArrayRef<Elf_Shdr> Sections, 246 const Elf_Shdr *Symtab) { 247 FirstNonLocal = Symtab->sh_info; 248 ELFSyms = CHECK(getObj().symbols(Symtab), this); 249 if (FirstNonLocal == 0 || FirstNonLocal > ELFSyms.size()) 250 fatal(toString(this) + ": invalid sh_info in symbol table"); 251 252 StringTable = 253 CHECK(getObj().getStringTableForSymtab(*Symtab, Sections), this); 254 } 255 256 template <class ELFT> 257 ObjFile<ELFT>::ObjFile(MemoryBufferRef M, StringRef ArchiveName) 258 : ELFFileBase<ELFT>(Base::ObjKind, M) { 259 this->ArchiveName = ArchiveName; 260 } 261 262 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() { 263 if (this->Symbols.empty()) 264 return {}; 265 return makeArrayRef(this->Symbols).slice(1, this->FirstNonLocal - 1); 266 } 267 268 template <class ELFT> 269 void ObjFile<ELFT>::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 270 // Read section and symbol tables. 271 initializeSections(ComdatGroups); 272 initializeSymbols(); 273 } 274 275 // Sections with SHT_GROUP and comdat bits define comdat section groups. 276 // They are identified and deduplicated by group name. This function 277 // returns a group name. 278 template <class ELFT> 279 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> Sections, 280 const Elf_Shdr &Sec) { 281 // Group signatures are stored as symbol names in object files. 282 // sh_info contains a symbol index, so we fetch a symbol and read its name. 283 if (this->ELFSyms.empty()) 284 this->initSymtab( 285 Sections, CHECK(object::getSection<ELFT>(Sections, Sec.sh_link), this)); 286 287 const Elf_Sym *Sym = 288 CHECK(object::getSymbol<ELFT>(this->ELFSyms, Sec.sh_info), this); 289 StringRef Signature = CHECK(Sym->getName(this->StringTable), this); 290 291 // As a special case, if a symbol is a section symbol and has no name, 292 // we use a section name as a signature. 293 // 294 // Such SHT_GROUP sections are invalid from the perspective of the ELF 295 // standard, but GNU gold 1.14 (the newest version as of July 2017) or 296 // older produce such sections as outputs for the -r option, so we need 297 // a bug-compatibility. 298 if (Signature.empty() && Sym->getType() == STT_SECTION) 299 return getSectionName(Sec); 300 return Signature; 301 } 302 303 template <class ELFT> 304 ArrayRef<typename ObjFile<ELFT>::Elf_Word> 305 ObjFile<ELFT>::getShtGroupEntries(const Elf_Shdr &Sec) { 306 const ELFFile<ELFT> &Obj = this->getObj(); 307 ArrayRef<Elf_Word> Entries = 308 CHECK(Obj.template getSectionContentsAsArray<Elf_Word>(&Sec), this); 309 if (Entries.empty() || Entries[0] != GRP_COMDAT) 310 fatal(toString(this) + ": unsupported SHT_GROUP format"); 311 return Entries.slice(1); 312 } 313 314 template <class ELFT> bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &Sec) { 315 // We don't merge sections if -O0 (default is -O1). This makes sometimes 316 // the linker significantly faster, although the output will be bigger. 317 if (Config->Optimize == 0) 318 return false; 319 320 // A mergeable section with size 0 is useless because they don't have 321 // any data to merge. A mergeable string section with size 0 can be 322 // argued as invalid because it doesn't end with a null character. 323 // We'll avoid a mess by handling them as if they were non-mergeable. 324 if (Sec.sh_size == 0) 325 return false; 326 327 // Check for sh_entsize. The ELF spec is not clear about the zero 328 // sh_entsize. It says that "the member [sh_entsize] contains 0 if 329 // the section does not hold a table of fixed-size entries". We know 330 // that Rust 1.13 produces a string mergeable section with a zero 331 // sh_entsize. Here we just accept it rather than being picky about it. 332 uint64_t EntSize = Sec.sh_entsize; 333 if (EntSize == 0) 334 return false; 335 if (Sec.sh_size % EntSize) 336 fatal(toString(this) + 337 ": SHF_MERGE section size must be a multiple of sh_entsize"); 338 339 uint64_t Flags = Sec.sh_flags; 340 if (!(Flags & SHF_MERGE)) 341 return false; 342 if (Flags & SHF_WRITE) 343 fatal(toString(this) + ": writable SHF_MERGE section is not supported"); 344 345 return true; 346 } 347 348 template <class ELFT> 349 void ObjFile<ELFT>::initializeSections( 350 DenseSet<CachedHashStringRef> &ComdatGroups) { 351 const ELFFile<ELFT> &Obj = this->getObj(); 352 353 ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this); 354 uint64_t Size = ObjSections.size(); 355 this->Sections.resize(Size); 356 this->SectionStringTable = 357 CHECK(Obj.getSectionStringTable(ObjSections), this); 358 359 for (size_t I = 0, E = ObjSections.size(); I < E; I++) { 360 if (this->Sections[I] == &InputSection::Discarded) 361 continue; 362 const Elf_Shdr &Sec = ObjSections[I]; 363 364 // SHF_EXCLUDE'ed sections are discarded by the linker. However, 365 // if -r is given, we'll let the final link discard such sections. 366 // This is compatible with GNU. 367 if ((Sec.sh_flags & SHF_EXCLUDE) && !Config->Relocatable) { 368 this->Sections[I] = &InputSection::Discarded; 369 continue; 370 } 371 372 switch (Sec.sh_type) { 373 case SHT_GROUP: { 374 // De-duplicate section groups by their signatures. 375 StringRef Signature = getShtGroupSignature(ObjSections, Sec); 376 bool IsNew = ComdatGroups.insert(CachedHashStringRef(Signature)).second; 377 this->Sections[I] = &InputSection::Discarded; 378 379 // If it is a new section group, we want to keep group members. 380 // Group leader sections, which contain indices of group members, are 381 // discarded because they are useless beyond this point. The only 382 // exception is the -r option because in order to produce re-linkable 383 // object files, we want to pass through basically everything. 384 if (IsNew) { 385 if (Config->Relocatable) 386 this->Sections[I] = createInputSection(Sec); 387 continue; 388 } 389 390 // Otherwise, discard group members. 391 for (uint32_t SecIndex : getShtGroupEntries(Sec)) { 392 if (SecIndex >= Size) 393 fatal(toString(this) + 394 ": invalid section index in group: " + Twine(SecIndex)); 395 this->Sections[SecIndex] = &InputSection::Discarded; 396 } 397 break; 398 } 399 case SHT_SYMTAB: 400 this->initSymtab(ObjSections, &Sec); 401 break; 402 case SHT_SYMTAB_SHNDX: 403 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, ObjSections), this); 404 break; 405 case SHT_STRTAB: 406 case SHT_NULL: 407 break; 408 default: 409 this->Sections[I] = createInputSection(Sec); 410 } 411 412 // .ARM.exidx sections have a reverse dependency on the InputSection they 413 // have a SHF_LINK_ORDER dependency, this is identified by the sh_link. 414 if (Sec.sh_flags & SHF_LINK_ORDER) { 415 if (Sec.sh_link >= this->Sections.size()) 416 fatal(toString(this) + 417 ": invalid sh_link index: " + Twine(Sec.sh_link)); 418 419 InputSectionBase *LinkSec = this->Sections[Sec.sh_link]; 420 InputSection *IS = cast<InputSection>(this->Sections[I]); 421 LinkSec->DependentSections.push_back(IS); 422 if (!isa<InputSection>(LinkSec)) 423 error("a section " + IS->Name + 424 " with SHF_LINK_ORDER should not refer a non-regular " 425 "section: " + 426 toString(LinkSec)); 427 } 428 } 429 } 430 431 // The ARM support in lld makes some use of instructions that are not available 432 // on all ARM architectures. Namely: 433 // - Use of BLX instruction for interworking between ARM and Thumb state. 434 // - Use of the extended Thumb branch encoding in relocation. 435 // - Use of the MOVT/MOVW instructions in Thumb Thunks. 436 // The ARM Attributes section contains information about the architecture chosen 437 // at compile time. We follow the convention that if at least one input object 438 // is compiled with an architecture that supports these features then lld is 439 // permitted to use them. 440 static void updateSupportedARMFeatures(const ARMAttributeParser &Attributes) { 441 if (!Attributes.hasAttribute(ARMBuildAttrs::CPU_arch)) 442 return; 443 auto Arch = Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch); 444 switch (Arch) { 445 case ARMBuildAttrs::Pre_v4: 446 case ARMBuildAttrs::v4: 447 case ARMBuildAttrs::v4T: 448 // Architectures prior to v5 do not support BLX instruction 449 break; 450 case ARMBuildAttrs::v5T: 451 case ARMBuildAttrs::v5TE: 452 case ARMBuildAttrs::v5TEJ: 453 case ARMBuildAttrs::v6: 454 case ARMBuildAttrs::v6KZ: 455 case ARMBuildAttrs::v6K: 456 Config->ARMHasBlx = true; 457 // Architectures used in pre-Cortex processors do not support 458 // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception 459 // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do. 460 break; 461 default: 462 // All other Architectures have BLX and extended branch encoding 463 Config->ARMHasBlx = true; 464 Config->ARMJ1J2BranchEncoding = true; 465 if (Arch != ARMBuildAttrs::v6_M && Arch != ARMBuildAttrs::v6S_M) 466 // All Architectures used in Cortex processors with the exception 467 // of v6-M and v6S-M have the MOVT and MOVW instructions. 468 Config->ARMHasMovtMovw = true; 469 break; 470 } 471 } 472 473 template <class ELFT> 474 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &Sec) { 475 uint32_t Idx = Sec.sh_info; 476 if (Idx >= this->Sections.size()) 477 fatal(toString(this) + ": invalid relocated section index: " + Twine(Idx)); 478 InputSectionBase *Target = this->Sections[Idx]; 479 480 // Strictly speaking, a relocation section must be included in the 481 // group of the section it relocates. However, LLVM 3.3 and earlier 482 // would fail to do so, so we gracefully handle that case. 483 if (Target == &InputSection::Discarded) 484 return nullptr; 485 486 if (!Target) 487 fatal(toString(this) + ": unsupported relocation reference"); 488 return Target; 489 } 490 491 // Create a regular InputSection class that has the same contents 492 // as a given section. 493 static InputSection *toRegularSection(MergeInputSection *Sec) { 494 return make<InputSection>(Sec->File, Sec->Flags, Sec->Type, Sec->Alignment, 495 Sec->Data, Sec->Name); 496 } 497 498 template <class ELFT> 499 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &Sec) { 500 StringRef Name = getSectionName(Sec); 501 502 switch (Sec.sh_type) { 503 case SHT_ARM_ATTRIBUTES: { 504 if (Config->EMachine != EM_ARM) 505 break; 506 ARMAttributeParser Attributes; 507 ArrayRef<uint8_t> Contents = check(this->getObj().getSectionContents(&Sec)); 508 Attributes.Parse(Contents, /*isLittle*/ Config->EKind == ELF32LEKind); 509 updateSupportedARMFeatures(Attributes); 510 // FIXME: Retain the first attribute section we see. The eglibc ARM 511 // dynamic loaders require the presence of an attribute section for dlopen 512 // to work. In a full implementation we would merge all attribute sections. 513 if (InX::ARMAttributes == nullptr) { 514 InX::ARMAttributes = make<InputSection>(*this, Sec, Name); 515 return InX::ARMAttributes; 516 } 517 return &InputSection::Discarded; 518 } 519 case SHT_RELA: 520 case SHT_REL: { 521 // Find the relocation target section and associate this 522 // section with it. Target can be discarded, for example 523 // if it is a duplicated member of SHT_GROUP section, we 524 // do not create or proccess relocatable sections then. 525 InputSectionBase *Target = getRelocTarget(Sec); 526 if (!Target) 527 return nullptr; 528 529 // This section contains relocation information. 530 // If -r is given, we do not interpret or apply relocation 531 // but just copy relocation sections to output. 532 if (Config->Relocatable) 533 return make<InputSection>(*this, Sec, Name); 534 535 if (Target->FirstRelocation) 536 fatal(toString(this) + 537 ": multiple relocation sections to one section are not supported"); 538 539 // Mergeable sections with relocations are tricky because relocations 540 // need to be taken into account when comparing section contents for 541 // merging. It's not worth supporting such mergeable sections because 542 // they are rare and it'd complicates the internal design (we usually 543 // have to determine if two sections are mergeable early in the link 544 // process much before applying relocations). We simply handle mergeable 545 // sections with relocations as non-mergeable. 546 if (auto *MS = dyn_cast<MergeInputSection>(Target)) { 547 Target = toRegularSection(MS); 548 this->Sections[Sec.sh_info] = Target; 549 } 550 551 size_t NumRelocations; 552 if (Sec.sh_type == SHT_RELA) { 553 ArrayRef<Elf_Rela> Rels = CHECK(this->getObj().relas(&Sec), this); 554 Target->FirstRelocation = Rels.begin(); 555 NumRelocations = Rels.size(); 556 Target->AreRelocsRela = true; 557 } else { 558 ArrayRef<Elf_Rel> Rels = CHECK(this->getObj().rels(&Sec), this); 559 Target->FirstRelocation = Rels.begin(); 560 NumRelocations = Rels.size(); 561 Target->AreRelocsRela = false; 562 } 563 assert(isUInt<31>(NumRelocations)); 564 Target->NumRelocations = NumRelocations; 565 566 // Relocation sections processed by the linker are usually removed 567 // from the output, so returning `nullptr` for the normal case. 568 // However, if -emit-relocs is given, we need to leave them in the output. 569 // (Some post link analysis tools need this information.) 570 if (Config->EmitRelocs) { 571 InputSection *RelocSec = make<InputSection>(*this, Sec, Name); 572 // We will not emit relocation section if target was discarded. 573 Target->DependentSections.push_back(RelocSec); 574 return RelocSec; 575 } 576 return nullptr; 577 } 578 } 579 580 // The GNU linker uses .note.GNU-stack section as a marker indicating 581 // that the code in the object file does not expect that the stack is 582 // executable (in terms of NX bit). If all input files have the marker, 583 // the GNU linker adds a PT_GNU_STACK segment to tells the loader to 584 // make the stack non-executable. Most object files have this section as 585 // of 2017. 586 // 587 // But making the stack non-executable is a norm today for security 588 // reasons. Failure to do so may result in a serious security issue. 589 // Therefore, we make LLD always add PT_GNU_STACK unless it is 590 // explicitly told to do otherwise (by -z execstack). Because the stack 591 // executable-ness is controlled solely by command line options, 592 // .note.GNU-stack sections are simply ignored. 593 if (Name == ".note.GNU-stack") 594 return &InputSection::Discarded; 595 596 // Split stacks is a feature to support a discontiguous stack. At least 597 // as of 2017, it seems that the feature is not being used widely. 598 // Only GNU gold supports that. We don't. For the details about that, 599 // see https://gcc.gnu.org/wiki/SplitStacks 600 if (Name == ".note.GNU-split-stack") { 601 error(toString(this) + 602 ": object file compiled with -fsplit-stack is not supported"); 603 return &InputSection::Discarded; 604 } 605 606 // The linkonce feature is a sort of proto-comdat. Some glibc i386 object 607 // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce 608 // sections. Drop those sections to avoid duplicate symbol errors. 609 // FIXME: This is glibc PR20543, we should remove this hack once that has been 610 // fixed for a while. 611 if (Name.startswith(".gnu.linkonce.")) 612 return &InputSection::Discarded; 613 614 // If we are creating a new .build-id section, strip existing .build-id 615 // sections so that the output won't have more than one .build-id. 616 // This is not usually a problem because input object files normally don't 617 // have .build-id sections, but you can create such files by 618 // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it. 619 if (Name == ".note.gnu.build-id" && Config->BuildId != BuildIdKind::None) 620 return &InputSection::Discarded; 621 622 // The linker merges EH (exception handling) frames and creates a 623 // .eh_frame_hdr section for runtime. So we handle them with a special 624 // class. For relocatable outputs, they are just passed through. 625 if (Name == ".eh_frame" && !Config->Relocatable) 626 return make<EhInputSection>(*this, Sec, Name); 627 628 if (shouldMerge(Sec)) 629 return make<MergeInputSection>(*this, Sec, Name); 630 return make<InputSection>(*this, Sec, Name); 631 } 632 633 template <class ELFT> 634 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &Sec) { 635 return CHECK(this->getObj().getSectionName(&Sec, SectionStringTable), this); 636 } 637 638 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() { 639 this->Symbols.reserve(this->ELFSyms.size()); 640 for (const Elf_Sym &Sym : this->ELFSyms) 641 this->Symbols.push_back(createSymbol(&Sym)); 642 } 643 644 template <class ELFT> Symbol *ObjFile<ELFT>::createSymbol(const Elf_Sym *Sym) { 645 int Binding = Sym->getBinding(); 646 647 uint32_t SecIdx = this->getSectionIndex(*Sym); 648 if (SecIdx >= this->Sections.size()) 649 fatal(toString(this) + ": invalid section index: " + Twine(SecIdx)); 650 651 InputSectionBase *Sec = this->Sections[SecIdx]; 652 uint8_t StOther = Sym->st_other; 653 uint8_t Type = Sym->getType(); 654 uint64_t Value = Sym->st_value; 655 uint64_t Size = Sym->st_size; 656 657 if (Binding == STB_LOCAL) { 658 if (Sym->getType() == STT_FILE) 659 SourceFile = CHECK(Sym->getName(this->StringTable), this); 660 661 if (this->StringTable.size() <= Sym->st_name) 662 fatal(toString(this) + ": invalid symbol name offset"); 663 664 StringRefZ Name = this->StringTable.data() + Sym->st_name; 665 if (Sym->st_shndx == SHN_UNDEF) 666 return make<Undefined>(this, Name, Binding, StOther, Type); 667 668 return make<Defined>(this, Name, Binding, StOther, Type, Value, Size, Sec); 669 } 670 671 StringRef Name = CHECK(Sym->getName(this->StringTable), this); 672 673 switch (Sym->st_shndx) { 674 case SHN_UNDEF: 675 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 676 /*CanOmitFromDynSym=*/false, this); 677 case SHN_COMMON: 678 if (Value == 0 || Value >= UINT32_MAX) 679 fatal(toString(this) + ": common symbol '" + Name + 680 "' has invalid alignment: " + Twine(Value)); 681 return Symtab->addCommon(Name, Size, Value, Binding, StOther, Type, *this); 682 } 683 684 switch (Binding) { 685 default: 686 fatal(toString(this) + ": unexpected binding: " + Twine(Binding)); 687 case STB_GLOBAL: 688 case STB_WEAK: 689 case STB_GNU_UNIQUE: 690 if (Sec == &InputSection::Discarded) 691 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 692 /*CanOmitFromDynSym=*/false, this); 693 return Symtab->addRegular(Name, StOther, Type, Value, Size, Binding, Sec, 694 this); 695 } 696 } 697 698 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&File) 699 : InputFile(ArchiveKind, File->getMemoryBufferRef()), 700 File(std::move(File)) {} 701 702 template <class ELFT> void ArchiveFile::parse() { 703 for (const Archive::Symbol &Sym : File->symbols()) 704 Symtab->addLazyArchive<ELFT>(Sym.getName(), *this, Sym); 705 } 706 707 // Returns a buffer pointing to a member file containing a given symbol. 708 std::pair<MemoryBufferRef, uint64_t> 709 ArchiveFile::getMember(const Archive::Symbol *Sym) { 710 Archive::Child C = 711 CHECK(Sym->getMember(), toString(this) + 712 ": could not get the member for symbol " + 713 Sym->getName()); 714 715 if (!Seen.insert(C.getChildOffset()).second) 716 return {MemoryBufferRef(), 0}; 717 718 MemoryBufferRef Ret = 719 CHECK(C.getMemoryBufferRef(), 720 toString(this) + 721 ": could not get the buffer for the member defining symbol " + 722 Sym->getName()); 723 724 if (C.getParent()->isThin() && Tar) 725 Tar->append(relativeToRoot(CHECK(C.getFullName(), this)), Ret.getBuffer()); 726 if (C.getParent()->isThin()) 727 return {Ret, 0}; 728 return {Ret, C.getChildOffset()}; 729 } 730 731 template <class ELFT> 732 SharedFile<ELFT>::SharedFile(MemoryBufferRef M, StringRef DefaultSoName) 733 : ELFFileBase<ELFT>(Base::SharedKind, M), SoName(DefaultSoName), 734 IsNeeded(!Config->AsNeeded) {} 735 736 // Partially parse the shared object file so that we can call 737 // getSoName on this object. 738 template <class ELFT> void SharedFile<ELFT>::parseSoName() { 739 const Elf_Shdr *DynamicSec = nullptr; 740 const ELFFile<ELFT> Obj = this->getObj(); 741 ArrayRef<Elf_Shdr> Sections = CHECK(Obj.sections(), this); 742 743 // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d. 744 for (const Elf_Shdr &Sec : Sections) { 745 switch (Sec.sh_type) { 746 default: 747 continue; 748 case SHT_DYNSYM: 749 this->initSymtab(Sections, &Sec); 750 break; 751 case SHT_DYNAMIC: 752 DynamicSec = &Sec; 753 break; 754 case SHT_SYMTAB_SHNDX: 755 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, Sections), this); 756 break; 757 case SHT_GNU_versym: 758 this->VersymSec = &Sec; 759 break; 760 case SHT_GNU_verdef: 761 this->VerdefSec = &Sec; 762 break; 763 } 764 } 765 766 if (this->VersymSec && this->ELFSyms.empty()) 767 error("SHT_GNU_versym should be associated with symbol table"); 768 769 // Search for a DT_SONAME tag to initialize this->SoName. 770 if (!DynamicSec) 771 return; 772 ArrayRef<Elf_Dyn> Arr = 773 CHECK(Obj.template getSectionContentsAsArray<Elf_Dyn>(DynamicSec), this); 774 for (const Elf_Dyn &Dyn : Arr) { 775 if (Dyn.d_tag == DT_SONAME) { 776 uint64_t Val = Dyn.getVal(); 777 if (Val >= this->StringTable.size()) 778 fatal(toString(this) + ": invalid DT_SONAME entry"); 779 SoName = this->StringTable.data() + Val; 780 return; 781 } 782 } 783 } 784 785 // Parse the version definitions in the object file if present. Returns a vector 786 // whose nth element contains a pointer to the Elf_Verdef for version identifier 787 // n. Version identifiers that are not definitions map to nullptr. 788 template <class ELFT> 789 std::vector<const typename ELFT::Verdef *> 790 SharedFile<ELFT>::parseVerdefs(const Elf_Versym *&Versym) { 791 // We only need to process symbol versions for this DSO if it has both a 792 // versym and a verdef section, which indicates that the DSO contains symbol 793 // version definitions. 794 if (!VersymSec || !VerdefSec) 795 return {}; 796 797 // The location of the first global versym entry. 798 const char *Base = this->MB.getBuffer().data(); 799 Versym = reinterpret_cast<const Elf_Versym *>(Base + VersymSec->sh_offset) + 800 this->FirstNonLocal; 801 802 // We cannot determine the largest verdef identifier without inspecting 803 // every Elf_Verdef, but both bfd and gold assign verdef identifiers 804 // sequentially starting from 1, so we predict that the largest identifier 805 // will be VerdefCount. 806 unsigned VerdefCount = VerdefSec->sh_info; 807 std::vector<const Elf_Verdef *> Verdefs(VerdefCount + 1); 808 809 // Build the Verdefs array by following the chain of Elf_Verdef objects 810 // from the start of the .gnu.version_d section. 811 const char *Verdef = Base + VerdefSec->sh_offset; 812 for (unsigned I = 0; I != VerdefCount; ++I) { 813 auto *CurVerdef = reinterpret_cast<const Elf_Verdef *>(Verdef); 814 Verdef += CurVerdef->vd_next; 815 unsigned VerdefIndex = CurVerdef->vd_ndx; 816 if (Verdefs.size() <= VerdefIndex) 817 Verdefs.resize(VerdefIndex + 1); 818 Verdefs[VerdefIndex] = CurVerdef; 819 } 820 821 return Verdefs; 822 } 823 824 // Fully parse the shared object file. This must be called after parseSoName(). 825 template <class ELFT> void SharedFile<ELFT>::parseRest() { 826 // Create mapping from version identifiers to Elf_Verdef entries. 827 const Elf_Versym *Versym = nullptr; 828 Verdefs = parseVerdefs(Versym); 829 830 ArrayRef<Elf_Shdr> Sections = CHECK(this->getObj().sections(), this); 831 832 // Add symbols to the symbol table. 833 Elf_Sym_Range Syms = this->getGlobalELFSyms(); 834 for (const Elf_Sym &Sym : Syms) { 835 unsigned VersymIndex = VER_NDX_GLOBAL; 836 if (Versym) { 837 VersymIndex = Versym->vs_index; 838 ++Versym; 839 } 840 bool Hidden = VersymIndex & VERSYM_HIDDEN; 841 VersymIndex = VersymIndex & ~VERSYM_HIDDEN; 842 843 StringRef Name = CHECK(Sym.getName(this->StringTable), this); 844 if (Sym.isUndefined()) { 845 Symbol *S = Symtab->addUndefined<ELFT>(Name, Sym.getBinding(), 846 Sym.st_other, Sym.getType(), 847 /*CanOmitFromDynSym=*/false, this); 848 S->ExportDynamic = true; 849 continue; 850 } 851 852 // ELF spec requires that all local symbols precede weak or global 853 // symbols in each symbol table, and the index of first non-local symbol 854 // is stored to sh_info. If a local symbol appears after some non-local 855 // symbol, that's a violation of the spec. 856 if (Sym.getBinding() == STB_LOCAL) { 857 warn("found local symbol '" + Name + 858 "' in global part of symbol table in file " + toString(this)); 859 continue; 860 } 861 862 // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly 863 // assigns VER_NDX_LOCAL to this section global symbol. Here is a 864 // workaround for this bug. 865 if (Config->EMachine == EM_MIPS && VersymIndex == VER_NDX_LOCAL && 866 Name == "_gp_disp") 867 continue; 868 869 const Elf_Verdef *Ver = nullptr; 870 if (VersymIndex != VER_NDX_GLOBAL) { 871 if (VersymIndex >= Verdefs.size() || VersymIndex == VER_NDX_LOCAL) { 872 error("corrupt input file: version definition index " + 873 Twine(VersymIndex) + " for symbol " + Name + 874 " is out of bounds\n>>> defined in " + toString(this)); 875 continue; 876 } 877 Ver = Verdefs[VersymIndex]; 878 } 879 880 // We do not usually care about alignments of data in shared object 881 // files because the loader takes care of it. However, if we promote a 882 // DSO symbol to point to .bss due to copy relocation, we need to keep 883 // the original alignment requirements. We infer it here. 884 uint64_t Alignment = 1; 885 if (Sym.st_value) 886 Alignment = 1ULL << countTrailingZeros((uint64_t)Sym.st_value); 887 if (0 < Sym.st_shndx && Sym.st_shndx < Sections.size()) { 888 uint64_t SecAlign = Sections[Sym.st_shndx].sh_addralign; 889 Alignment = std::min(Alignment, SecAlign); 890 } 891 if (Alignment > UINT32_MAX) 892 error(toString(this) + ": alignment too large: " + Name); 893 894 if (!Hidden) 895 Symtab->addShared(Name, *this, Sym, Alignment, VersymIndex); 896 897 // Also add the symbol with the versioned name to handle undefined symbols 898 // with explicit versions. 899 if (Ver) { 900 StringRef VerName = this->StringTable.data() + Ver->getAux()->vda_name; 901 Name = Saver.save(Name + "@" + VerName); 902 Symtab->addShared(Name, *this, Sym, Alignment, VersymIndex); 903 } 904 } 905 } 906 907 static ELFKind getBitcodeELFKind(const Triple &T) { 908 if (T.isLittleEndian()) 909 return T.isArch64Bit() ? ELF64LEKind : ELF32LEKind; 910 return T.isArch64Bit() ? ELF64BEKind : ELF32BEKind; 911 } 912 913 static uint8_t getBitcodeMachineKind(StringRef Path, const Triple &T) { 914 switch (T.getArch()) { 915 case Triple::aarch64: 916 return EM_AARCH64; 917 case Triple::arm: 918 case Triple::thumb: 919 return EM_ARM; 920 case Triple::avr: 921 return EM_AVR; 922 case Triple::mips: 923 case Triple::mipsel: 924 case Triple::mips64: 925 case Triple::mips64el: 926 return EM_MIPS; 927 case Triple::ppc: 928 return EM_PPC; 929 case Triple::ppc64: 930 return EM_PPC64; 931 case Triple::x86: 932 return T.isOSIAMCU() ? EM_IAMCU : EM_386; 933 case Triple::x86_64: 934 return EM_X86_64; 935 default: 936 fatal(Path + ": could not infer e_machine from bitcode target triple " + 937 T.str()); 938 } 939 } 940 941 BitcodeFile::BitcodeFile(MemoryBufferRef MB, StringRef ArchiveName, 942 uint64_t OffsetInArchive) 943 : InputFile(BitcodeKind, MB) { 944 this->ArchiveName = ArchiveName; 945 946 // Here we pass a new MemoryBufferRef which is identified by ArchiveName 947 // (the fully resolved path of the archive) + member name + offset of the 948 // member in the archive. 949 // ThinLTO uses the MemoryBufferRef identifier to access its internal 950 // data structures and if two archives define two members with the same name, 951 // this causes a collision which result in only one of the objects being 952 // taken into consideration at LTO time (which very likely causes undefined 953 // symbols later in the link stage). 954 MemoryBufferRef MBRef(MB.getBuffer(), 955 Saver.save(ArchiveName + MB.getBufferIdentifier() + 956 utostr(OffsetInArchive))); 957 Obj = CHECK(lto::InputFile::create(MBRef), this); 958 959 Triple T(Obj->getTargetTriple()); 960 EKind = getBitcodeELFKind(T); 961 EMachine = getBitcodeMachineKind(MB.getBufferIdentifier(), T); 962 } 963 964 static uint8_t mapVisibility(GlobalValue::VisibilityTypes GvVisibility) { 965 switch (GvVisibility) { 966 case GlobalValue::DefaultVisibility: 967 return STV_DEFAULT; 968 case GlobalValue::HiddenVisibility: 969 return STV_HIDDEN; 970 case GlobalValue::ProtectedVisibility: 971 return STV_PROTECTED; 972 } 973 llvm_unreachable("unknown visibility"); 974 } 975 976 template <class ELFT> 977 static Symbol *createBitcodeSymbol(const std::vector<bool> &KeptComdats, 978 const lto::InputFile::Symbol &ObjSym, 979 BitcodeFile &F) { 980 StringRef NameRef = Saver.save(ObjSym.getName()); 981 uint32_t Binding = ObjSym.isWeak() ? STB_WEAK : STB_GLOBAL; 982 983 uint8_t Type = ObjSym.isTLS() ? STT_TLS : STT_NOTYPE; 984 uint8_t Visibility = mapVisibility(ObjSym.getVisibility()); 985 bool CanOmitFromDynSym = ObjSym.canBeOmittedFromSymbolTable(); 986 987 int C = ObjSym.getComdatIndex(); 988 if (C != -1 && !KeptComdats[C]) 989 return Symtab->addUndefined<ELFT>(NameRef, Binding, Visibility, Type, 990 CanOmitFromDynSym, &F); 991 992 if (ObjSym.isUndefined()) 993 return Symtab->addUndefined<ELFT>(NameRef, Binding, Visibility, Type, 994 CanOmitFromDynSym, &F); 995 996 if (ObjSym.isCommon()) 997 return Symtab->addCommon(NameRef, ObjSym.getCommonSize(), 998 ObjSym.getCommonAlignment(), Binding, Visibility, 999 STT_OBJECT, F); 1000 1001 return Symtab->addBitcode(NameRef, Binding, Visibility, Type, 1002 CanOmitFromDynSym, F); 1003 } 1004 1005 template <class ELFT> 1006 void BitcodeFile::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 1007 std::vector<bool> KeptComdats; 1008 for (StringRef S : Obj->getComdatTable()) 1009 KeptComdats.push_back(ComdatGroups.insert(CachedHashStringRef(S)).second); 1010 1011 for (const lto::InputFile::Symbol &ObjSym : Obj->symbols()) 1012 Symbols.push_back(createBitcodeSymbol<ELFT>(KeptComdats, ObjSym, *this)); 1013 } 1014 1015 static ELFKind getELFKind(MemoryBufferRef MB) { 1016 unsigned char Size; 1017 unsigned char Endian; 1018 std::tie(Size, Endian) = getElfArchType(MB.getBuffer()); 1019 1020 if (Endian != ELFDATA2LSB && Endian != ELFDATA2MSB) 1021 fatal(MB.getBufferIdentifier() + ": invalid data encoding"); 1022 if (Size != ELFCLASS32 && Size != ELFCLASS64) 1023 fatal(MB.getBufferIdentifier() + ": invalid file class"); 1024 1025 size_t BufSize = MB.getBuffer().size(); 1026 if ((Size == ELFCLASS32 && BufSize < sizeof(Elf32_Ehdr)) || 1027 (Size == ELFCLASS64 && BufSize < sizeof(Elf64_Ehdr))) 1028 fatal(MB.getBufferIdentifier() + ": file is too short"); 1029 1030 if (Size == ELFCLASS32) 1031 return (Endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind; 1032 return (Endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind; 1033 } 1034 1035 void BinaryFile::parse() { 1036 ArrayRef<uint8_t> Data = toArrayRef(MB.getBuffer()); 1037 auto *Section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS, 1038 8, Data, ".data"); 1039 Sections.push_back(Section); 1040 1041 // For each input file foo that is embedded to a result as a binary 1042 // blob, we define _binary_foo_{start,end,size} symbols, so that 1043 // user programs can access blobs by name. Non-alphanumeric 1044 // characters in a filename are replaced with underscore. 1045 std::string S = "_binary_" + MB.getBufferIdentifier().str(); 1046 for (size_t I = 0; I < S.size(); ++I) 1047 if (!isAlnum(S[I])) 1048 S[I] = '_'; 1049 1050 Symtab->addRegular(Saver.save(S + "_start"), STV_DEFAULT, STT_OBJECT, 0, 0, 1051 STB_GLOBAL, Section, nullptr); 1052 Symtab->addRegular(Saver.save(S + "_end"), STV_DEFAULT, STT_OBJECT, 1053 Data.size(), 0, STB_GLOBAL, Section, nullptr); 1054 Symtab->addRegular(Saver.save(S + "_size"), STV_DEFAULT, STT_OBJECT, 1055 Data.size(), 0, STB_GLOBAL, nullptr, nullptr); 1056 } 1057 1058 static bool isBitcode(MemoryBufferRef MB) { 1059 using namespace sys::fs; 1060 return identify_magic(MB.getBuffer()) == file_magic::bitcode; 1061 } 1062 1063 InputFile *elf::createObjectFile(MemoryBufferRef MB, StringRef ArchiveName, 1064 uint64_t OffsetInArchive) { 1065 if (isBitcode(MB)) 1066 return make<BitcodeFile>(MB, ArchiveName, OffsetInArchive); 1067 1068 switch (getELFKind(MB)) { 1069 case ELF32LEKind: 1070 return make<ObjFile<ELF32LE>>(MB, ArchiveName); 1071 case ELF32BEKind: 1072 return make<ObjFile<ELF32BE>>(MB, ArchiveName); 1073 case ELF64LEKind: 1074 return make<ObjFile<ELF64LE>>(MB, ArchiveName); 1075 case ELF64BEKind: 1076 return make<ObjFile<ELF64BE>>(MB, ArchiveName); 1077 default: 1078 llvm_unreachable("getELFKind"); 1079 } 1080 } 1081 1082 InputFile *elf::createSharedFile(MemoryBufferRef MB, StringRef DefaultSoName) { 1083 switch (getELFKind(MB)) { 1084 case ELF32LEKind: 1085 return make<SharedFile<ELF32LE>>(MB, DefaultSoName); 1086 case ELF32BEKind: 1087 return make<SharedFile<ELF32BE>>(MB, DefaultSoName); 1088 case ELF64LEKind: 1089 return make<SharedFile<ELF64LE>>(MB, DefaultSoName); 1090 case ELF64BEKind: 1091 return make<SharedFile<ELF64BE>>(MB, DefaultSoName); 1092 default: 1093 llvm_unreachable("getELFKind"); 1094 } 1095 } 1096 1097 MemoryBufferRef LazyObjFile::getBuffer() { 1098 if (Seen) 1099 return MemoryBufferRef(); 1100 Seen = true; 1101 return MB; 1102 } 1103 1104 InputFile *LazyObjFile::fetch() { 1105 MemoryBufferRef MBRef = getBuffer(); 1106 if (MBRef.getBuffer().empty()) 1107 return nullptr; 1108 return createObjectFile(MBRef, ArchiveName, OffsetInArchive); 1109 } 1110 1111 template <class ELFT> void LazyObjFile::parse() { 1112 // A lazy object file wraps either a bitcode file or an ELF file. 1113 if (isBitcode(this->MB)) { 1114 std::unique_ptr<lto::InputFile> Obj = 1115 CHECK(lto::InputFile::create(this->MB), this); 1116 for (const lto::InputFile::Symbol &Sym : Obj->symbols()) 1117 if (!Sym.isUndefined()) 1118 Symtab->addLazyObject<ELFT>(Saver.save(Sym.getName()), *this); 1119 return; 1120 } 1121 1122 switch (getELFKind(this->MB)) { 1123 case ELF32LEKind: 1124 addElfSymbols<ELF32LE>(); 1125 return; 1126 case ELF32BEKind: 1127 addElfSymbols<ELF32BE>(); 1128 return; 1129 case ELF64LEKind: 1130 addElfSymbols<ELF64LE>(); 1131 return; 1132 case ELF64BEKind: 1133 addElfSymbols<ELF64BE>(); 1134 return; 1135 default: 1136 llvm_unreachable("getELFKind"); 1137 } 1138 } 1139 1140 template <class ELFT> void LazyObjFile::addElfSymbols() { 1141 ELFFile<ELFT> Obj = check(ELFFile<ELFT>::create(MB.getBuffer())); 1142 ArrayRef<typename ELFT::Shdr> Sections = CHECK(Obj.sections(), this); 1143 1144 for (const typename ELFT::Shdr &Sec : Sections) { 1145 if (Sec.sh_type != SHT_SYMTAB) 1146 continue; 1147 1148 typename ELFT::SymRange Syms = CHECK(Obj.symbols(&Sec), this); 1149 uint32_t FirstNonLocal = Sec.sh_info; 1150 StringRef StringTable = 1151 CHECK(Obj.getStringTableForSymtab(Sec, Sections), this); 1152 1153 for (const typename ELFT::Sym &Sym : Syms.slice(FirstNonLocal)) 1154 if (Sym.st_shndx != SHN_UNDEF) 1155 Symtab->addLazyObject<ELFT>(CHECK(Sym.getName(StringTable), this), 1156 *this); 1157 return; 1158 } 1159 } 1160 1161 // This is for --just-symbols. 1162 // 1163 // This option allows you to link your output against other existing 1164 // program, so that if you load both your program and the other program 1165 // into memory, your output can use program's symbols. 1166 // 1167 // What we are doing here is to read defined symbols from a given ELF 1168 // file and add them as absolute symbols. 1169 template <class ELFT> void elf::readJustSymbolsFile(MemoryBufferRef MB) { 1170 typedef typename ELFT::Shdr Elf_Shdr; 1171 typedef typename ELFT::Sym Elf_Sym; 1172 typedef typename ELFT::SymRange Elf_Sym_Range; 1173 1174 StringRef ObjName = MB.getBufferIdentifier(); 1175 ELFFile<ELFT> Obj = check(ELFFile<ELFT>::create(MB.getBuffer())); 1176 ArrayRef<Elf_Shdr> Sections = CHECK(Obj.sections(), ObjName); 1177 1178 for (const Elf_Shdr &Sec : Sections) { 1179 if (Sec.sh_type != SHT_SYMTAB) 1180 continue; 1181 1182 Elf_Sym_Range Syms = CHECK(Obj.symbols(&Sec), ObjName); 1183 uint32_t FirstNonLocal = Sec.sh_info; 1184 StringRef StringTable = 1185 CHECK(Obj.getStringTableForSymtab(Sec, Sections), ObjName); 1186 1187 for (const Elf_Sym &Sym : Syms.slice(FirstNonLocal)) 1188 if (Sym.st_shndx != SHN_UNDEF) 1189 Symtab->addRegular(CHECK(Sym.getName(StringTable), ObjName), 1190 Sym.st_other, Sym.getType(), Sym.st_value, 1191 Sym.st_size, Sym.getBinding(), nullptr, nullptr); 1192 return; 1193 } 1194 } 1195 1196 template void ArchiveFile::parse<ELF32LE>(); 1197 template void ArchiveFile::parse<ELF32BE>(); 1198 template void ArchiveFile::parse<ELF64LE>(); 1199 template void ArchiveFile::parse<ELF64BE>(); 1200 1201 template void BitcodeFile::parse<ELF32LE>(DenseSet<CachedHashStringRef> &); 1202 template void BitcodeFile::parse<ELF32BE>(DenseSet<CachedHashStringRef> &); 1203 template void BitcodeFile::parse<ELF64LE>(DenseSet<CachedHashStringRef> &); 1204 template void BitcodeFile::parse<ELF64BE>(DenseSet<CachedHashStringRef> &); 1205 1206 template void LazyObjFile::parse<ELF32LE>(); 1207 template void LazyObjFile::parse<ELF32BE>(); 1208 template void LazyObjFile::parse<ELF64LE>(); 1209 template void LazyObjFile::parse<ELF64BE>(); 1210 1211 template class elf::ELFFileBase<ELF32LE>; 1212 template class elf::ELFFileBase<ELF32BE>; 1213 template class elf::ELFFileBase<ELF64LE>; 1214 template class elf::ELFFileBase<ELF64BE>; 1215 1216 template class elf::ObjFile<ELF32LE>; 1217 template class elf::ObjFile<ELF32BE>; 1218 template class elf::ObjFile<ELF64LE>; 1219 template class elf::ObjFile<ELF64BE>; 1220 1221 template class elf::SharedFile<ELF32LE>; 1222 template class elf::SharedFile<ELF32BE>; 1223 template class elf::SharedFile<ELF64LE>; 1224 template class elf::SharedFile<ELF64BE>; 1225 1226 template void elf::readJustSymbolsFile<ELF32LE>(MemoryBufferRef); 1227 template void elf::readJustSymbolsFile<ELF32BE>(MemoryBufferRef); 1228 template void elf::readJustSymbolsFile<ELF64LE>(MemoryBufferRef); 1229 template void elf::readJustSymbolsFile<ELF64BE>(MemoryBufferRef); 1230