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 bool InputFile::IsInGroup; 42 uint32_t InputFile::NextGroupId; 43 std::vector<BinaryFile *> elf::BinaryFiles; 44 std::vector<BitcodeFile *> elf::BitcodeFiles; 45 std::vector<LazyObjFile *> elf::LazyObjFiles; 46 std::vector<InputFile *> elf::ObjectFiles; 47 std::vector<InputFile *> elf::SharedFiles; 48 49 TarWriter *elf::Tar; 50 51 InputFile::InputFile(Kind K, MemoryBufferRef M) 52 : MB(M), GroupId(NextGroupId), FileKind(K) { 53 // All files within the same --{start,end}-group get the same group ID. 54 // Otherwise, a new file will get a new group ID. 55 if (!IsInGroup) 56 ++NextGroupId; 57 } 58 59 Optional<MemoryBufferRef> elf::readFile(StringRef Path) { 60 // The --chroot option changes our virtual root directory. 61 // This is useful when you are dealing with files created by --reproduce. 62 if (!Config->Chroot.empty() && Path.startswith("/")) 63 Path = Saver.save(Config->Chroot + Path); 64 65 log(Path); 66 67 auto MBOrErr = MemoryBuffer::getFile(Path, -1, false); 68 if (auto EC = MBOrErr.getError()) { 69 error("cannot open " + Path + ": " + EC.message()); 70 return None; 71 } 72 73 std::unique_ptr<MemoryBuffer> &MB = *MBOrErr; 74 MemoryBufferRef MBRef = MB->getMemBufferRef(); 75 make<std::unique_ptr<MemoryBuffer>>(std::move(MB)); // take MB ownership 76 77 if (Tar) 78 Tar->append(relativeToRoot(Path), MBRef.getBuffer()); 79 return MBRef; 80 } 81 82 // Concatenates arguments to construct a string representing an error location. 83 static std::string createFileLineMsg(StringRef Path, unsigned Line) { 84 std::string Filename = path::filename(Path); 85 std::string Lineno = ":" + std::to_string(Line); 86 if (Filename == Path) 87 return Filename + Lineno; 88 return Filename + Lineno + " (" + Path.str() + Lineno + ")"; 89 } 90 91 template <class ELFT> 92 static std::string getSrcMsgAux(ObjFile<ELFT> &File, const Symbol &Sym, 93 InputSectionBase &Sec, uint64_t Offset) { 94 // In DWARF, functions and variables are stored to different places. 95 // First, lookup a function for a given offset. 96 if (Optional<DILineInfo> Info = File.getDILineInfo(&Sec, Offset)) 97 return createFileLineMsg(Info->FileName, Info->Line); 98 99 // If it failed, lookup again as a variable. 100 if (Optional<std::pair<std::string, unsigned>> FileLine = 101 File.getVariableLoc(Sym.getName())) 102 return createFileLineMsg(FileLine->first, FileLine->second); 103 104 // File.SourceFile contains STT_FILE symbol, and that is a last resort. 105 return File.SourceFile; 106 } 107 108 std::string InputFile::getSrcMsg(const Symbol &Sym, InputSectionBase &Sec, 109 uint64_t Offset) { 110 if (kind() != ObjKind) 111 return ""; 112 switch (Config->EKind) { 113 default: 114 llvm_unreachable("Invalid kind"); 115 case ELF32LEKind: 116 return getSrcMsgAux(cast<ObjFile<ELF32LE>>(*this), Sym, Sec, Offset); 117 case ELF32BEKind: 118 return getSrcMsgAux(cast<ObjFile<ELF32BE>>(*this), Sym, Sec, Offset); 119 case ELF64LEKind: 120 return getSrcMsgAux(cast<ObjFile<ELF64LE>>(*this), Sym, Sec, Offset); 121 case ELF64BEKind: 122 return getSrcMsgAux(cast<ObjFile<ELF64BE>>(*this), Sym, Sec, Offset); 123 } 124 } 125 126 template <class ELFT> void ObjFile<ELFT>::initializeDwarf() { 127 Dwarf = llvm::make_unique<DWARFContext>(make_unique<LLDDwarfObj<ELFT>>(this)); 128 const DWARFObject &Obj = Dwarf->getDWARFObj(); 129 DwarfLine.reset(new DWARFDebugLine); 130 DWARFDataExtractor LineData(Obj, Obj.getLineSection(), Config->IsLE, 131 Config->Wordsize); 132 133 for (std::unique_ptr<DWARFCompileUnit> &CU : Dwarf->compile_units()) { 134 const DWARFDebugLine::LineTable *LT = Dwarf->getLineTableForUnit(CU.get()); 135 if (!LT) 136 continue; 137 LineTables.push_back(LT); 138 139 // Loop over variable records and insert them to VariableLoc. 140 for (const auto &Entry : CU->dies()) { 141 DWARFDie Die(CU.get(), &Entry); 142 // Skip all tags that are not variables. 143 if (Die.getTag() != dwarf::DW_TAG_variable) 144 continue; 145 146 // Skip if a local variable because we don't need them for generating 147 // error messages. In general, only non-local symbols can fail to be 148 // linked. 149 if (!dwarf::toUnsigned(Die.find(dwarf::DW_AT_external), 0)) 150 continue; 151 152 // Get the source filename index for the variable. 153 unsigned File = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_file), 0); 154 if (!LT->hasFileAtIndex(File)) 155 continue; 156 157 // Get the line number on which the variable is declared. 158 unsigned Line = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_line), 0); 159 160 // Get the name of the variable and add the collected information to 161 // VariableLoc. Usually Name is non-empty, but it can be empty if the 162 // input object file lacks some debug info. 163 StringRef Name = dwarf::toString(Die.find(dwarf::DW_AT_name), ""); 164 if (!Name.empty()) 165 VariableLoc.insert({Name, {LT, File, Line}}); 166 } 167 } 168 } 169 170 // Returns the pair of file name and line number describing location of data 171 // object (variable, array, etc) definition. 172 template <class ELFT> 173 Optional<std::pair<std::string, unsigned>> 174 ObjFile<ELFT>::getVariableLoc(StringRef Name) { 175 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 176 177 // Return if we have no debug information about data object. 178 auto It = VariableLoc.find(Name); 179 if (It == VariableLoc.end()) 180 return None; 181 182 // Take file name string from line table. 183 std::string FileName; 184 if (!It->second.LT->getFileNameByIndex( 185 It->second.File, nullptr, 186 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FileName)) 187 return None; 188 189 return std::make_pair(FileName, It->second.Line); 190 } 191 192 // Returns source line information for a given offset 193 // using DWARF debug info. 194 template <class ELFT> 195 Optional<DILineInfo> ObjFile<ELFT>::getDILineInfo(InputSectionBase *S, 196 uint64_t Offset) { 197 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 198 199 // Use fake address calcuated by adding section file offset and offset in 200 // section. See comments for ObjectInfo class. 201 DILineInfo Info; 202 for (const llvm::DWARFDebugLine::LineTable *LT : LineTables) 203 if (LT->getFileLineInfoForAddress( 204 S->getOffsetInFile() + Offset, nullptr, 205 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, Info)) 206 return Info; 207 return None; 208 } 209 210 // Returns source line information for a given offset using DWARF debug info. 211 template <class ELFT> 212 std::string ObjFile<ELFT>::getLineInfo(InputSectionBase *S, uint64_t Offset) { 213 if (Optional<DILineInfo> Info = getDILineInfo(S, Offset)) 214 return Info->FileName + ":" + std::to_string(Info->Line); 215 return ""; 216 } 217 218 // Returns "<internal>", "foo.a(bar.o)" or "baz.o". 219 std::string lld::toString(const InputFile *F) { 220 if (!F) 221 return "<internal>"; 222 223 if (F->ToStringCache.empty()) { 224 if (F->ArchiveName.empty()) 225 F->ToStringCache = F->getName(); 226 else 227 F->ToStringCache = (F->ArchiveName + "(" + F->getName() + ")").str(); 228 } 229 return F->ToStringCache; 230 } 231 232 template <class ELFT> 233 ELFFileBase<ELFT>::ELFFileBase(Kind K, MemoryBufferRef MB) : InputFile(K, MB) { 234 if (ELFT::TargetEndianness == support::little) 235 EKind = ELFT::Is64Bits ? ELF64LEKind : ELF32LEKind; 236 else 237 EKind = ELFT::Is64Bits ? ELF64BEKind : ELF32BEKind; 238 239 EMachine = getObj().getHeader()->e_machine; 240 OSABI = getObj().getHeader()->e_ident[llvm::ELF::EI_OSABI]; 241 } 242 243 template <class ELFT> 244 typename ELFT::SymRange ELFFileBase<ELFT>::getGlobalELFSyms() { 245 return makeArrayRef(ELFSyms.begin() + FirstGlobal, ELFSyms.end()); 246 } 247 248 template <class ELFT> 249 uint32_t ELFFileBase<ELFT>::getSectionIndex(const Elf_Sym &Sym) const { 250 return CHECK(getObj().getSectionIndex(&Sym, ELFSyms, SymtabSHNDX), this); 251 } 252 253 template <class ELFT> 254 void ELFFileBase<ELFT>::initSymtab(ArrayRef<Elf_Shdr> Sections, 255 const Elf_Shdr *Symtab) { 256 FirstGlobal = Symtab->sh_info; 257 ELFSyms = CHECK(getObj().symbols(Symtab), this); 258 if (FirstGlobal == 0 || FirstGlobal > ELFSyms.size()) 259 fatal(toString(this) + ": invalid sh_info in symbol table"); 260 261 StringTable = 262 CHECK(getObj().getStringTableForSymtab(*Symtab, Sections), this); 263 } 264 265 template <class ELFT> 266 ObjFile<ELFT>::ObjFile(MemoryBufferRef M, StringRef ArchiveName) 267 : ELFFileBase<ELFT>(Base::ObjKind, M) { 268 this->ArchiveName = ArchiveName; 269 } 270 271 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() { 272 if (this->Symbols.empty()) 273 return {}; 274 return makeArrayRef(this->Symbols).slice(1, this->FirstGlobal - 1); 275 } 276 277 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getGlobalSymbols() { 278 return makeArrayRef(this->Symbols).slice(this->FirstGlobal); 279 } 280 281 template <class ELFT> 282 void ObjFile<ELFT>::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 283 // Read a section table. JustSymbols is usually false. 284 if (this->JustSymbols) 285 initializeJustSymbols(); 286 else 287 initializeSections(ComdatGroups); 288 289 // Read a symbol table. 290 initializeSymbols(); 291 } 292 293 // Sections with SHT_GROUP and comdat bits define comdat section groups. 294 // They are identified and deduplicated by group name. This function 295 // returns a group name. 296 template <class ELFT> 297 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> Sections, 298 const Elf_Shdr &Sec) { 299 // Group signatures are stored as symbol names in object files. 300 // sh_info contains a symbol index, so we fetch a symbol and read its name. 301 if (this->ELFSyms.empty()) 302 this->initSymtab( 303 Sections, CHECK(object::getSection<ELFT>(Sections, Sec.sh_link), this)); 304 305 const Elf_Sym *Sym = 306 CHECK(object::getSymbol<ELFT>(this->ELFSyms, Sec.sh_info), this); 307 StringRef Signature = CHECK(Sym->getName(this->StringTable), this); 308 309 // As a special case, if a symbol is a section symbol and has no name, 310 // we use a section name as a signature. 311 // 312 // Such SHT_GROUP sections are invalid from the perspective of the ELF 313 // standard, but GNU gold 1.14 (the newest version as of July 2017) or 314 // older produce such sections as outputs for the -r option, so we need 315 // a bug-compatibility. 316 if (Signature.empty() && Sym->getType() == STT_SECTION) 317 return getSectionName(Sec); 318 return Signature; 319 } 320 321 template <class ELFT> 322 ArrayRef<typename ObjFile<ELFT>::Elf_Word> 323 ObjFile<ELFT>::getShtGroupEntries(const Elf_Shdr &Sec) { 324 const ELFFile<ELFT> &Obj = this->getObj(); 325 ArrayRef<Elf_Word> Entries = 326 CHECK(Obj.template getSectionContentsAsArray<Elf_Word>(&Sec), this); 327 if (Entries.empty() || Entries[0] != GRP_COMDAT) 328 fatal(toString(this) + ": unsupported SHT_GROUP format"); 329 return Entries.slice(1); 330 } 331 332 template <class ELFT> bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &Sec) { 333 // On a regular link we don't merge sections if -O0 (default is -O1). This 334 // sometimes makes the linker significantly faster, although the output will 335 // be bigger. 336 // 337 // Doing the same for -r would create a problem as it would combine sections 338 // with different sh_entsize. One option would be to just copy every SHF_MERGE 339 // section as is to the output. While this would produce a valid ELF file with 340 // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when 341 // they see two .debug_str. We could have separate logic for combining 342 // SHF_MERGE sections based both on their name and sh_entsize, but that seems 343 // to be more trouble than it is worth. Instead, we just use the regular (-O1) 344 // logic for -r. 345 if (Config->Optimize == 0 && !Config->Relocatable) 346 return false; 347 348 // A mergeable section with size 0 is useless because they don't have 349 // any data to merge. A mergeable string section with size 0 can be 350 // argued as invalid because it doesn't end with a null character. 351 // We'll avoid a mess by handling them as if they were non-mergeable. 352 if (Sec.sh_size == 0) 353 return false; 354 355 // Check for sh_entsize. The ELF spec is not clear about the zero 356 // sh_entsize. It says that "the member [sh_entsize] contains 0 if 357 // the section does not hold a table of fixed-size entries". We know 358 // that Rust 1.13 produces a string mergeable section with a zero 359 // sh_entsize. Here we just accept it rather than being picky about it. 360 uint64_t EntSize = Sec.sh_entsize; 361 if (EntSize == 0) 362 return false; 363 if (Sec.sh_size % EntSize) 364 fatal(toString(this) + 365 ": SHF_MERGE section size must be a multiple of sh_entsize"); 366 367 uint64_t Flags = Sec.sh_flags; 368 if (!(Flags & SHF_MERGE)) 369 return false; 370 if (Flags & SHF_WRITE) 371 fatal(toString(this) + ": writable SHF_MERGE section is not supported"); 372 373 return true; 374 } 375 376 // This is for --just-symbols. 377 // 378 // --just-symbols is a very minor feature that allows you to link your 379 // output against other existing program, so that if you load both your 380 // program and the other program into memory, your output can refer the 381 // other program's symbols. 382 // 383 // When the option is given, we link "just symbols". The section table is 384 // initialized with null pointers. 385 template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() { 386 ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this); 387 this->Sections.resize(ObjSections.size()); 388 389 for (const Elf_Shdr &Sec : ObjSections) { 390 if (Sec.sh_type != SHT_SYMTAB) 391 continue; 392 this->initSymtab(ObjSections, &Sec); 393 return; 394 } 395 } 396 397 template <class ELFT> 398 void ObjFile<ELFT>::initializeSections( 399 DenseSet<CachedHashStringRef> &ComdatGroups) { 400 const ELFFile<ELFT> &Obj = this->getObj(); 401 402 ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this); 403 uint64_t Size = ObjSections.size(); 404 this->Sections.resize(Size); 405 this->SectionStringTable = 406 CHECK(Obj.getSectionStringTable(ObjSections), this); 407 408 for (size_t I = 0, E = ObjSections.size(); I < E; I++) { 409 if (this->Sections[I] == &InputSection::Discarded) 410 continue; 411 const Elf_Shdr &Sec = ObjSections[I]; 412 413 // SHF_EXCLUDE'ed sections are discarded by the linker. However, 414 // if -r is given, we'll let the final link discard such sections. 415 // This is compatible with GNU. 416 if ((Sec.sh_flags & SHF_EXCLUDE) && !Config->Relocatable) { 417 this->Sections[I] = &InputSection::Discarded; 418 continue; 419 } 420 421 switch (Sec.sh_type) { 422 case SHT_GROUP: { 423 // De-duplicate section groups by their signatures. 424 StringRef Signature = getShtGroupSignature(ObjSections, Sec); 425 bool IsNew = ComdatGroups.insert(CachedHashStringRef(Signature)).second; 426 this->Sections[I] = &InputSection::Discarded; 427 428 // If it is a new section group, we want to keep group members. 429 // Group leader sections, which contain indices of group members, are 430 // discarded because they are useless beyond this point. The only 431 // exception is the -r option because in order to produce re-linkable 432 // object files, we want to pass through basically everything. 433 if (IsNew) { 434 if (Config->Relocatable) 435 this->Sections[I] = createInputSection(Sec); 436 continue; 437 } 438 439 // Otherwise, discard group members. 440 for (uint32_t SecIndex : getShtGroupEntries(Sec)) { 441 if (SecIndex >= Size) 442 fatal(toString(this) + 443 ": invalid section index in group: " + Twine(SecIndex)); 444 this->Sections[SecIndex] = &InputSection::Discarded; 445 } 446 break; 447 } 448 case SHT_SYMTAB: 449 this->initSymtab(ObjSections, &Sec); 450 break; 451 case SHT_SYMTAB_SHNDX: 452 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, ObjSections), this); 453 break; 454 case SHT_STRTAB: 455 case SHT_NULL: 456 break; 457 default: 458 this->Sections[I] = createInputSection(Sec); 459 } 460 461 // .ARM.exidx sections have a reverse dependency on the InputSection they 462 // have a SHF_LINK_ORDER dependency, this is identified by the sh_link. 463 if (Sec.sh_flags & SHF_LINK_ORDER) { 464 if (Sec.sh_link >= this->Sections.size()) 465 fatal(toString(this) + 466 ": invalid sh_link index: " + Twine(Sec.sh_link)); 467 468 InputSectionBase *LinkSec = this->Sections[Sec.sh_link]; 469 InputSection *IS = cast<InputSection>(this->Sections[I]); 470 LinkSec->DependentSections.push_back(IS); 471 if (!isa<InputSection>(LinkSec)) 472 error("a section " + IS->Name + 473 " with SHF_LINK_ORDER should not refer a non-regular " 474 "section: " + 475 toString(LinkSec)); 476 } 477 } 478 } 479 480 // The ARM support in lld makes some use of instructions that are not available 481 // on all ARM architectures. Namely: 482 // - Use of BLX instruction for interworking between ARM and Thumb state. 483 // - Use of the extended Thumb branch encoding in relocation. 484 // - Use of the MOVT/MOVW instructions in Thumb Thunks. 485 // The ARM Attributes section contains information about the architecture chosen 486 // at compile time. We follow the convention that if at least one input object 487 // is compiled with an architecture that supports these features then lld is 488 // permitted to use them. 489 static void updateSupportedARMFeatures(const ARMAttributeParser &Attributes) { 490 if (!Attributes.hasAttribute(ARMBuildAttrs::CPU_arch)) 491 return; 492 auto Arch = Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch); 493 switch (Arch) { 494 case ARMBuildAttrs::Pre_v4: 495 case ARMBuildAttrs::v4: 496 case ARMBuildAttrs::v4T: 497 // Architectures prior to v5 do not support BLX instruction 498 break; 499 case ARMBuildAttrs::v5T: 500 case ARMBuildAttrs::v5TE: 501 case ARMBuildAttrs::v5TEJ: 502 case ARMBuildAttrs::v6: 503 case ARMBuildAttrs::v6KZ: 504 case ARMBuildAttrs::v6K: 505 Config->ARMHasBlx = true; 506 // Architectures used in pre-Cortex processors do not support 507 // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception 508 // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do. 509 break; 510 default: 511 // All other Architectures have BLX and extended branch encoding 512 Config->ARMHasBlx = true; 513 Config->ARMJ1J2BranchEncoding = true; 514 if (Arch != ARMBuildAttrs::v6_M && Arch != ARMBuildAttrs::v6S_M) 515 // All Architectures used in Cortex processors with the exception 516 // of v6-M and v6S-M have the MOVT and MOVW instructions. 517 Config->ARMHasMovtMovw = true; 518 break; 519 } 520 } 521 522 template <class ELFT> 523 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &Sec) { 524 uint32_t Idx = Sec.sh_info; 525 if (Idx >= this->Sections.size()) 526 fatal(toString(this) + ": invalid relocated section index: " + Twine(Idx)); 527 InputSectionBase *Target = this->Sections[Idx]; 528 529 // Strictly speaking, a relocation section must be included in the 530 // group of the section it relocates. However, LLVM 3.3 and earlier 531 // would fail to do so, so we gracefully handle that case. 532 if (Target == &InputSection::Discarded) 533 return nullptr; 534 535 if (!Target) 536 fatal(toString(this) + ": unsupported relocation reference"); 537 return Target; 538 } 539 540 // Create a regular InputSection class that has the same contents 541 // as a given section. 542 static InputSection *toRegularSection(MergeInputSection *Sec) { 543 return make<InputSection>(Sec->File, Sec->Flags, Sec->Type, Sec->Alignment, 544 Sec->Data, Sec->Name); 545 } 546 547 template <class ELFT> 548 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &Sec) { 549 StringRef Name = getSectionName(Sec); 550 551 switch (Sec.sh_type) { 552 case SHT_ARM_ATTRIBUTES: { 553 if (Config->EMachine != EM_ARM) 554 break; 555 ARMAttributeParser Attributes; 556 ArrayRef<uint8_t> Contents = check(this->getObj().getSectionContents(&Sec)); 557 Attributes.Parse(Contents, /*isLittle*/ Config->EKind == ELF32LEKind); 558 updateSupportedARMFeatures(Attributes); 559 // FIXME: Retain the first attribute section we see. The eglibc ARM 560 // dynamic loaders require the presence of an attribute section for dlopen 561 // to work. In a full implementation we would merge all attribute sections. 562 if (InX::ARMAttributes == nullptr) { 563 InX::ARMAttributes = make<InputSection>(*this, Sec, Name); 564 return InX::ARMAttributes; 565 } 566 return &InputSection::Discarded; 567 } 568 case SHT_RELA: 569 case SHT_REL: { 570 // Find a relocation target section and associate this section with that. 571 // Target may have been discarded if it is in a different section group 572 // and the group is discarded, even though it's a violation of the 573 // spec. We handle that situation gracefully by discarding dangling 574 // relocation sections. 575 InputSectionBase *Target = getRelocTarget(Sec); 576 if (!Target) 577 return nullptr; 578 579 // This section contains relocation information. 580 // If -r is given, we do not interpret or apply relocation 581 // but just copy relocation sections to output. 582 if (Config->Relocatable) 583 return make<InputSection>(*this, Sec, Name); 584 585 if (Target->FirstRelocation) 586 fatal(toString(this) + 587 ": multiple relocation sections to one section are not supported"); 588 589 // ELF spec allows mergeable sections with relocations, but they are 590 // rare, and it is in practice hard to merge such sections by contents, 591 // because applying relocations at end of linking changes section 592 // contents. So, we simply handle such sections as non-mergeable ones. 593 // Degrading like this is acceptable because section merging is optional. 594 if (auto *MS = dyn_cast<MergeInputSection>(Target)) { 595 Target = toRegularSection(MS); 596 this->Sections[Sec.sh_info] = Target; 597 } 598 599 if (Sec.sh_type == SHT_RELA) { 600 ArrayRef<Elf_Rela> Rels = CHECK(this->getObj().relas(&Sec), this); 601 Target->FirstRelocation = Rels.begin(); 602 Target->NumRelocations = Rels.size(); 603 Target->AreRelocsRela = true; 604 } else { 605 ArrayRef<Elf_Rel> Rels = CHECK(this->getObj().rels(&Sec), this); 606 Target->FirstRelocation = Rels.begin(); 607 Target->NumRelocations = Rels.size(); 608 Target->AreRelocsRela = false; 609 } 610 assert(isUInt<31>(Target->NumRelocations)); 611 612 // Relocation sections processed by the linker are usually removed 613 // from the output, so returning `nullptr` for the normal case. 614 // However, if -emit-relocs is given, we need to leave them in the output. 615 // (Some post link analysis tools need this information.) 616 if (Config->EmitRelocs) { 617 InputSection *RelocSec = make<InputSection>(*this, Sec, Name); 618 // We will not emit relocation section if target was discarded. 619 Target->DependentSections.push_back(RelocSec); 620 return RelocSec; 621 } 622 return nullptr; 623 } 624 } 625 626 // The GNU linker uses .note.GNU-stack section as a marker indicating 627 // that the code in the object file does not expect that the stack is 628 // executable (in terms of NX bit). If all input files have the marker, 629 // the GNU linker adds a PT_GNU_STACK segment to tells the loader to 630 // make the stack non-executable. Most object files have this section as 631 // of 2017. 632 // 633 // But making the stack non-executable is a norm today for security 634 // reasons. Failure to do so may result in a serious security issue. 635 // Therefore, we make LLD always add PT_GNU_STACK unless it is 636 // explicitly told to do otherwise (by -z execstack). Because the stack 637 // executable-ness is controlled solely by command line options, 638 // .note.GNU-stack sections are simply ignored. 639 if (Name == ".note.GNU-stack") 640 return &InputSection::Discarded; 641 642 // Split stacks is a feature to support a discontiguous stack. At least 643 // as of 2017, it seems that the feature is not being used widely. 644 // Only GNU gold supports that. We don't. For the details about that, 645 // see https://gcc.gnu.org/wiki/SplitStacks 646 if (Name == ".note.GNU-split-stack") { 647 error(toString(this) + 648 ": object file compiled with -fsplit-stack is not supported"); 649 return &InputSection::Discarded; 650 } 651 652 // The linkonce feature is a sort of proto-comdat. Some glibc i386 object 653 // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce 654 // sections. Drop those sections to avoid duplicate symbol errors. 655 // FIXME: This is glibc PR20543, we should remove this hack once that has been 656 // fixed for a while. 657 if (Name.startswith(".gnu.linkonce.")) 658 return &InputSection::Discarded; 659 660 // If we are creating a new .build-id section, strip existing .build-id 661 // sections so that the output won't have more than one .build-id. 662 // This is not usually a problem because input object files normally don't 663 // have .build-id sections, but you can create such files by 664 // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it. 665 if (Name == ".note.gnu.build-id" && Config->BuildId != BuildIdKind::None) 666 return &InputSection::Discarded; 667 668 // The linker merges EH (exception handling) frames and creates a 669 // .eh_frame_hdr section for runtime. So we handle them with a special 670 // class. For relocatable outputs, they are just passed through. 671 if (Name == ".eh_frame" && !Config->Relocatable) 672 return make<EhInputSection>(*this, Sec, Name); 673 674 if (shouldMerge(Sec)) 675 return make<MergeInputSection>(*this, Sec, Name); 676 return make<InputSection>(*this, Sec, Name); 677 } 678 679 template <class ELFT> 680 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &Sec) { 681 return CHECK(this->getObj().getSectionName(&Sec, SectionStringTable), this); 682 } 683 684 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() { 685 this->Symbols.reserve(this->ELFSyms.size()); 686 for (const Elf_Sym &Sym : this->ELFSyms) 687 this->Symbols.push_back(createSymbol(&Sym)); 688 } 689 690 template <class ELFT> Symbol *ObjFile<ELFT>::createSymbol(const Elf_Sym *Sym) { 691 int Binding = Sym->getBinding(); 692 693 uint32_t SecIdx = this->getSectionIndex(*Sym); 694 if (SecIdx >= this->Sections.size()) 695 fatal(toString(this) + ": invalid section index: " + Twine(SecIdx)); 696 697 InputSectionBase *Sec = this->Sections[SecIdx]; 698 uint8_t StOther = Sym->st_other; 699 uint8_t Type = Sym->getType(); 700 uint64_t Value = Sym->st_value; 701 uint64_t Size = Sym->st_size; 702 703 if (Binding == STB_LOCAL) { 704 if (Sym->getType() == STT_FILE) 705 SourceFile = CHECK(Sym->getName(this->StringTable), this); 706 707 if (this->StringTable.size() <= Sym->st_name) 708 fatal(toString(this) + ": invalid symbol name offset"); 709 710 StringRefZ Name = this->StringTable.data() + Sym->st_name; 711 if (Sym->st_shndx == SHN_UNDEF) 712 return make<Undefined>(this, Name, Binding, StOther, Type); 713 714 return make<Defined>(this, Name, Binding, StOther, Type, Value, Size, Sec); 715 } 716 717 StringRef Name = CHECK(Sym->getName(this->StringTable), this); 718 719 switch (Sym->st_shndx) { 720 case SHN_UNDEF: 721 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 722 /*CanOmitFromDynSym=*/false, this); 723 case SHN_COMMON: 724 if (Value == 0 || Value >= UINT32_MAX) 725 fatal(toString(this) + ": common symbol '" + Name + 726 "' has invalid alignment: " + Twine(Value)); 727 return Symtab->addCommon(Name, Size, Value, Binding, StOther, Type, *this); 728 } 729 730 switch (Binding) { 731 default: 732 fatal(toString(this) + ": unexpected binding: " + Twine(Binding)); 733 case STB_GLOBAL: 734 case STB_WEAK: 735 case STB_GNU_UNIQUE: 736 if (Sec == &InputSection::Discarded) 737 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 738 /*CanOmitFromDynSym=*/false, this); 739 return Symtab->addRegular(Name, StOther, Type, Value, Size, Binding, Sec, 740 this); 741 } 742 } 743 744 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&File) 745 : InputFile(ArchiveKind, File->getMemoryBufferRef()), 746 File(std::move(File)) {} 747 748 template <class ELFT> void ArchiveFile::parse() { 749 for (const Archive::Symbol &Sym : File->symbols()) 750 Symtab->addLazyArchive<ELFT>(Sym.getName(), *this, Sym); 751 } 752 753 // Returns a buffer pointing to a member file containing a given symbol. 754 InputFile *ArchiveFile::fetch(const Archive::Symbol &Sym) { 755 Archive::Child C = 756 CHECK(Sym.getMember(), toString(this) + 757 ": could not get the member for symbol " + 758 Sym.getName()); 759 760 if (!Seen.insert(C.getChildOffset()).second) 761 return nullptr; 762 763 MemoryBufferRef MB = 764 CHECK(C.getMemoryBufferRef(), 765 toString(this) + 766 ": could not get the buffer for the member defining symbol " + 767 Sym.getName()); 768 769 if (Tar && C.getParent()->isThin()) 770 Tar->append(relativeToRoot(CHECK(C.getFullName(), this)), MB.getBuffer()); 771 772 InputFile *File = createObjectFile( 773 MB, getName(), C.getParent()->isThin() ? 0 : C.getChildOffset()); 774 File->GroupId = GroupId; 775 return File; 776 } 777 778 template <class ELFT> 779 SharedFile<ELFT>::SharedFile(MemoryBufferRef M, StringRef DefaultSoName) 780 : ELFFileBase<ELFT>(Base::SharedKind, M), SoName(DefaultSoName), 781 IsNeeded(!Config->AsNeeded) {} 782 783 // Partially parse the shared object file so that we can call 784 // getSoName on this object. 785 template <class ELFT> void SharedFile<ELFT>::parseSoName() { 786 const Elf_Shdr *DynamicSec = nullptr; 787 const ELFFile<ELFT> Obj = this->getObj(); 788 ArrayRef<Elf_Shdr> Sections = CHECK(Obj.sections(), this); 789 790 // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d. 791 for (const Elf_Shdr &Sec : Sections) { 792 switch (Sec.sh_type) { 793 default: 794 continue; 795 case SHT_DYNSYM: 796 this->initSymtab(Sections, &Sec); 797 break; 798 case SHT_DYNAMIC: 799 DynamicSec = &Sec; 800 break; 801 case SHT_SYMTAB_SHNDX: 802 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, Sections), this); 803 break; 804 case SHT_GNU_versym: 805 this->VersymSec = &Sec; 806 break; 807 case SHT_GNU_verdef: 808 this->VerdefSec = &Sec; 809 break; 810 } 811 } 812 813 if (this->VersymSec && this->ELFSyms.empty()) 814 error("SHT_GNU_versym should be associated with symbol table"); 815 816 // Search for a DT_SONAME tag to initialize this->SoName. 817 if (!DynamicSec) 818 return; 819 ArrayRef<Elf_Dyn> Arr = 820 CHECK(Obj.template getSectionContentsAsArray<Elf_Dyn>(DynamicSec), this); 821 for (const Elf_Dyn &Dyn : Arr) { 822 if (Dyn.d_tag == DT_SONAME) { 823 uint64_t Val = Dyn.getVal(); 824 if (Val >= this->StringTable.size()) 825 fatal(toString(this) + ": invalid DT_SONAME entry"); 826 SoName = this->StringTable.data() + Val; 827 return; 828 } 829 } 830 } 831 832 // Parses ".gnu.version" section which is a parallel array for the symbol table. 833 // If a given file doesn't have ".gnu.version" section, returns VER_NDX_GLOBAL. 834 template <class ELFT> std::vector<uint32_t> SharedFile<ELFT>::parseVersyms() { 835 size_t Size = this->ELFSyms.size() - this->FirstGlobal; 836 if (!VersymSec) 837 return std::vector<uint32_t>(Size, VER_NDX_GLOBAL); 838 839 const char *Base = this->MB.getBuffer().data(); 840 const Elf_Versym *Versym = 841 reinterpret_cast<const Elf_Versym *>(Base + VersymSec->sh_offset) + 842 this->FirstGlobal; 843 844 std::vector<uint32_t> Ret(Size); 845 for (size_t I = 0; I < Size; ++I) 846 Ret[I] = Versym[I].vs_index; 847 return Ret; 848 } 849 850 // Parse the version definitions in the object file if present. Returns a vector 851 // whose nth element contains a pointer to the Elf_Verdef for version identifier 852 // n. Version identifiers that are not definitions map to nullptr. 853 template <class ELFT> 854 std::vector<const typename ELFT::Verdef *> SharedFile<ELFT>::parseVerdefs() { 855 if (!VerdefSec) 856 return {}; 857 858 // We cannot determine the largest verdef identifier without inspecting 859 // every Elf_Verdef, but both bfd and gold assign verdef identifiers 860 // sequentially starting from 1, so we predict that the largest identifier 861 // will be VerdefCount. 862 unsigned VerdefCount = VerdefSec->sh_info; 863 std::vector<const Elf_Verdef *> Verdefs(VerdefCount + 1); 864 865 // Build the Verdefs array by following the chain of Elf_Verdef objects 866 // from the start of the .gnu.version_d section. 867 const char *Base = this->MB.getBuffer().data(); 868 const char *Verdef = Base + VerdefSec->sh_offset; 869 for (unsigned I = 0; I != VerdefCount; ++I) { 870 auto *CurVerdef = reinterpret_cast<const Elf_Verdef *>(Verdef); 871 Verdef += CurVerdef->vd_next; 872 unsigned VerdefIndex = CurVerdef->vd_ndx; 873 if (Verdefs.size() <= VerdefIndex) 874 Verdefs.resize(VerdefIndex + 1); 875 Verdefs[VerdefIndex] = CurVerdef; 876 } 877 878 return Verdefs; 879 } 880 881 // We do not usually care about alignments of data in shared object 882 // files because the loader takes care of it. However, if we promote a 883 // DSO symbol to point to .bss due to copy relocation, we need to keep 884 // the original alignment requirements. We infer it in this function. 885 template <class ELFT> 886 uint32_t SharedFile<ELFT>::getAlignment(ArrayRef<Elf_Shdr> Sections, 887 const Elf_Sym &Sym) { 888 uint64_t Ret = 1; 889 if (Sym.st_value) 890 Ret = 1ULL << countTrailingZeros((uint64_t)Sym.st_value); 891 if (0 < Sym.st_shndx && Sym.st_shndx < Sections.size()) 892 Ret = std::min<uint64_t>(Ret, Sections[Sym.st_shndx].sh_addralign); 893 894 if (Ret > UINT32_MAX) 895 error(toString(this) + ": alignment too large: " + 896 CHECK(Sym.getName(this->StringTable), this)); 897 return Ret; 898 } 899 900 // Fully parse the shared object file. This must be called after parseSoName(). 901 // 902 // This function parses symbol versions. If a DSO has version information, 903 // the file has a ".gnu.version_d" section which contains symbol version 904 // definitions. Each symbol is associated to one version through a table in 905 // ".gnu.version" section. That table is a parallel array for the symbol 906 // table, and each table entry contains an index in ".gnu.version_d". 907 // 908 // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for 909 // VER_NDX_GLOBAL. There's no table entry for these special versions in 910 // ".gnu.version_d". 911 // 912 // The file format for symbol versioning is perhaps a bit more complicated 913 // than necessary, but you can easily understand the code if you wrap your 914 // head around the data structure described above. 915 template <class ELFT> void SharedFile<ELFT>::parseRest() { 916 Verdefs = parseVerdefs(); // parse .gnu.version_d 917 std::vector<uint32_t> Versyms = parseVersyms(); // parse .gnu.version 918 ArrayRef<Elf_Shdr> Sections = CHECK(this->getObj().sections(), this); 919 920 // System libraries can have a lot of symbols with versions. Using a 921 // fixed buffer for computing the versions name (foo@ver) can save a 922 // lot of allocations. 923 SmallString<0> VersionedNameBuffer; 924 925 // Add symbols to the symbol table. 926 ArrayRef<Elf_Sym> Syms = this->getGlobalELFSyms(); 927 for (size_t I = 0; I < Syms.size(); ++I) { 928 const Elf_Sym &Sym = Syms[I]; 929 930 StringRef Name = CHECK(Sym.getName(this->StringTable), this); 931 if (Sym.isUndefined()) { 932 Symbol *S = Symtab->addUndefined<ELFT>(Name, Sym.getBinding(), 933 Sym.st_other, Sym.getType(), 934 /*CanOmitFromDynSym=*/false, this); 935 S->ExportDynamic = true; 936 continue; 937 } 938 939 // ELF spec requires that all local symbols precede weak or global 940 // symbols in each symbol table, and the index of first non-local symbol 941 // is stored to sh_info. If a local symbol appears after some non-local 942 // symbol, that's a violation of the spec. 943 if (Sym.getBinding() == STB_LOCAL) { 944 warn("found local symbol '" + Name + 945 "' in global part of symbol table in file " + toString(this)); 946 continue; 947 } 948 949 // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly 950 // assigns VER_NDX_LOCAL to this section global symbol. Here is a 951 // workaround for this bug. 952 uint32_t Idx = Versyms[I] & ~VERSYM_HIDDEN; 953 if (Config->EMachine == EM_MIPS && Idx == VER_NDX_LOCAL && 954 Name == "_gp_disp") 955 continue; 956 957 uint64_t Alignment = getAlignment(Sections, Sym); 958 if (!(Versyms[I] & VERSYM_HIDDEN)) 959 Symtab->addShared(Name, *this, Sym, Alignment, Idx); 960 961 // Also add the symbol with the versioned name to handle undefined symbols 962 // with explicit versions. 963 if (Idx == VER_NDX_GLOBAL) 964 continue; 965 966 if (Idx >= Verdefs.size() || Idx == VER_NDX_LOCAL) { 967 error("corrupt input file: version definition index " + Twine(Idx) + 968 " for symbol " + Name + " is out of bounds\n>>> defined in " + 969 toString(this)); 970 continue; 971 } 972 973 StringRef VerName = 974 this->StringTable.data() + Verdefs[Idx]->getAux()->vda_name; 975 VersionedNameBuffer.clear(); 976 Name = (Name + "@" + VerName).toStringRef(VersionedNameBuffer); 977 Symtab->addShared(Saver.save(Name), *this, Sym, Alignment, Idx); 978 } 979 } 980 981 static ELFKind getBitcodeELFKind(const Triple &T) { 982 if (T.isLittleEndian()) 983 return T.isArch64Bit() ? ELF64LEKind : ELF32LEKind; 984 return T.isArch64Bit() ? ELF64BEKind : ELF32BEKind; 985 } 986 987 static uint8_t getBitcodeMachineKind(StringRef Path, const Triple &T) { 988 switch (T.getArch()) { 989 case Triple::aarch64: 990 return EM_AARCH64; 991 case Triple::arm: 992 case Triple::thumb: 993 return EM_ARM; 994 case Triple::avr: 995 return EM_AVR; 996 case Triple::mips: 997 case Triple::mipsel: 998 case Triple::mips64: 999 case Triple::mips64el: 1000 return EM_MIPS; 1001 case Triple::ppc: 1002 return EM_PPC; 1003 case Triple::ppc64: 1004 return EM_PPC64; 1005 case Triple::x86: 1006 return T.isOSIAMCU() ? EM_IAMCU : EM_386; 1007 case Triple::x86_64: 1008 return EM_X86_64; 1009 default: 1010 fatal(Path + ": could not infer e_machine from bitcode target triple " + 1011 T.str()); 1012 } 1013 } 1014 1015 BitcodeFile::BitcodeFile(MemoryBufferRef MB, StringRef ArchiveName, 1016 uint64_t OffsetInArchive) 1017 : InputFile(BitcodeKind, MB) { 1018 this->ArchiveName = ArchiveName; 1019 1020 // Here we pass a new MemoryBufferRef which is identified by ArchiveName 1021 // (the fully resolved path of the archive) + member name + offset of the 1022 // member in the archive. 1023 // ThinLTO uses the MemoryBufferRef identifier to access its internal 1024 // data structures and if two archives define two members with the same name, 1025 // this causes a collision which result in only one of the objects being 1026 // taken into consideration at LTO time (which very likely causes undefined 1027 // symbols later in the link stage). 1028 MemoryBufferRef MBRef( 1029 MB.getBuffer(), 1030 Saver.save(ArchiveName + MB.getBufferIdentifier() + 1031 (ArchiveName.empty() ? "" : utostr(OffsetInArchive)))); 1032 Obj = CHECK(lto::InputFile::create(MBRef), this); 1033 1034 Triple T(Obj->getTargetTriple()); 1035 EKind = getBitcodeELFKind(T); 1036 EMachine = getBitcodeMachineKind(MB.getBufferIdentifier(), T); 1037 } 1038 1039 static uint8_t mapVisibility(GlobalValue::VisibilityTypes GvVisibility) { 1040 switch (GvVisibility) { 1041 case GlobalValue::DefaultVisibility: 1042 return STV_DEFAULT; 1043 case GlobalValue::HiddenVisibility: 1044 return STV_HIDDEN; 1045 case GlobalValue::ProtectedVisibility: 1046 return STV_PROTECTED; 1047 } 1048 llvm_unreachable("unknown visibility"); 1049 } 1050 1051 template <class ELFT> 1052 static Symbol *createBitcodeSymbol(const std::vector<bool> &KeptComdats, 1053 const lto::InputFile::Symbol &ObjSym, 1054 BitcodeFile &F) { 1055 StringRef NameRef = Saver.save(ObjSym.getName()); 1056 uint32_t Binding = ObjSym.isWeak() ? STB_WEAK : STB_GLOBAL; 1057 1058 uint8_t Type = ObjSym.isTLS() ? STT_TLS : STT_NOTYPE; 1059 uint8_t Visibility = mapVisibility(ObjSym.getVisibility()); 1060 bool CanOmitFromDynSym = ObjSym.canBeOmittedFromSymbolTable(); 1061 1062 int C = ObjSym.getComdatIndex(); 1063 if (C != -1 && !KeptComdats[C]) 1064 return Symtab->addUndefined<ELFT>(NameRef, Binding, Visibility, Type, 1065 CanOmitFromDynSym, &F); 1066 1067 if (ObjSym.isUndefined()) 1068 return Symtab->addUndefined<ELFT>(NameRef, Binding, Visibility, Type, 1069 CanOmitFromDynSym, &F); 1070 1071 if (ObjSym.isCommon()) 1072 return Symtab->addCommon(NameRef, ObjSym.getCommonSize(), 1073 ObjSym.getCommonAlignment(), Binding, Visibility, 1074 STT_OBJECT, F); 1075 1076 return Symtab->addBitcode(NameRef, Binding, Visibility, Type, 1077 CanOmitFromDynSym, F); 1078 } 1079 1080 template <class ELFT> 1081 void BitcodeFile::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 1082 std::vector<bool> KeptComdats; 1083 for (StringRef S : Obj->getComdatTable()) 1084 KeptComdats.push_back(ComdatGroups.insert(CachedHashStringRef(S)).second); 1085 1086 for (const lto::InputFile::Symbol &ObjSym : Obj->symbols()) 1087 Symbols.push_back(createBitcodeSymbol<ELFT>(KeptComdats, ObjSym, *this)); 1088 } 1089 1090 static ELFKind getELFKind(MemoryBufferRef MB) { 1091 unsigned char Size; 1092 unsigned char Endian; 1093 std::tie(Size, Endian) = getElfArchType(MB.getBuffer()); 1094 1095 if (Endian != ELFDATA2LSB && Endian != ELFDATA2MSB) 1096 fatal(MB.getBufferIdentifier() + ": invalid data encoding"); 1097 if (Size != ELFCLASS32 && Size != ELFCLASS64) 1098 fatal(MB.getBufferIdentifier() + ": invalid file class"); 1099 1100 size_t BufSize = MB.getBuffer().size(); 1101 if ((Size == ELFCLASS32 && BufSize < sizeof(Elf32_Ehdr)) || 1102 (Size == ELFCLASS64 && BufSize < sizeof(Elf64_Ehdr))) 1103 fatal(MB.getBufferIdentifier() + ": file is too short"); 1104 1105 if (Size == ELFCLASS32) 1106 return (Endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind; 1107 return (Endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind; 1108 } 1109 1110 void BinaryFile::parse() { 1111 ArrayRef<uint8_t> Data = toArrayRef(MB.getBuffer()); 1112 auto *Section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS, 1113 8, Data, ".data"); 1114 Sections.push_back(Section); 1115 1116 // For each input file foo that is embedded to a result as a binary 1117 // blob, we define _binary_foo_{start,end,size} symbols, so that 1118 // user programs can access blobs by name. Non-alphanumeric 1119 // characters in a filename are replaced with underscore. 1120 std::string S = "_binary_" + MB.getBufferIdentifier().str(); 1121 for (size_t I = 0; I < S.size(); ++I) 1122 if (!isAlnum(S[I])) 1123 S[I] = '_'; 1124 1125 Symtab->addRegular(Saver.save(S + "_start"), STV_DEFAULT, STT_OBJECT, 0, 0, 1126 STB_GLOBAL, Section, nullptr); 1127 Symtab->addRegular(Saver.save(S + "_end"), STV_DEFAULT, STT_OBJECT, 1128 Data.size(), 0, STB_GLOBAL, Section, nullptr); 1129 Symtab->addRegular(Saver.save(S + "_size"), STV_DEFAULT, STT_OBJECT, 1130 Data.size(), 0, STB_GLOBAL, nullptr, nullptr); 1131 } 1132 1133 InputFile *elf::createObjectFile(MemoryBufferRef MB, StringRef ArchiveName, 1134 uint64_t OffsetInArchive) { 1135 if (isBitcode(MB)) 1136 return make<BitcodeFile>(MB, ArchiveName, OffsetInArchive); 1137 1138 switch (getELFKind(MB)) { 1139 case ELF32LEKind: 1140 return make<ObjFile<ELF32LE>>(MB, ArchiveName); 1141 case ELF32BEKind: 1142 return make<ObjFile<ELF32BE>>(MB, ArchiveName); 1143 case ELF64LEKind: 1144 return make<ObjFile<ELF64LE>>(MB, ArchiveName); 1145 case ELF64BEKind: 1146 return make<ObjFile<ELF64BE>>(MB, ArchiveName); 1147 default: 1148 llvm_unreachable("getELFKind"); 1149 } 1150 } 1151 1152 InputFile *elf::createSharedFile(MemoryBufferRef MB, StringRef DefaultSoName) { 1153 switch (getELFKind(MB)) { 1154 case ELF32LEKind: 1155 return make<SharedFile<ELF32LE>>(MB, DefaultSoName); 1156 case ELF32BEKind: 1157 return make<SharedFile<ELF32BE>>(MB, DefaultSoName); 1158 case ELF64LEKind: 1159 return make<SharedFile<ELF64LE>>(MB, DefaultSoName); 1160 case ELF64BEKind: 1161 return make<SharedFile<ELF64BE>>(MB, DefaultSoName); 1162 default: 1163 llvm_unreachable("getELFKind"); 1164 } 1165 } 1166 1167 MemoryBufferRef LazyObjFile::getBuffer() { 1168 if (AddedToLink) 1169 return MemoryBufferRef(); 1170 AddedToLink = true; 1171 return MB; 1172 } 1173 1174 InputFile *LazyObjFile::fetch() { 1175 MemoryBufferRef MBRef = getBuffer(); 1176 if (MBRef.getBuffer().empty()) 1177 return nullptr; 1178 1179 InputFile *File = createObjectFile(MBRef, ArchiveName, OffsetInArchive); 1180 File->GroupId = GroupId; 1181 return File; 1182 } 1183 1184 template <class ELFT> void LazyObjFile::parse() { 1185 // A lazy object file wraps either a bitcode file or an ELF file. 1186 if (isBitcode(this->MB)) { 1187 std::unique_ptr<lto::InputFile> Obj = 1188 CHECK(lto::InputFile::create(this->MB), this); 1189 for (const lto::InputFile::Symbol &Sym : Obj->symbols()) 1190 if (!Sym.isUndefined()) 1191 Symtab->addLazyObject<ELFT>(Saver.save(Sym.getName()), *this); 1192 return; 1193 } 1194 1195 switch (getELFKind(this->MB)) { 1196 case ELF32LEKind: 1197 addElfSymbols<ELF32LE>(); 1198 return; 1199 case ELF32BEKind: 1200 addElfSymbols<ELF32BE>(); 1201 return; 1202 case ELF64LEKind: 1203 addElfSymbols<ELF64LE>(); 1204 return; 1205 case ELF64BEKind: 1206 addElfSymbols<ELF64BE>(); 1207 return; 1208 default: 1209 llvm_unreachable("getELFKind"); 1210 } 1211 } 1212 1213 template <class ELFT> void LazyObjFile::addElfSymbols() { 1214 ELFFile<ELFT> Obj = check(ELFFile<ELFT>::create(MB.getBuffer())); 1215 ArrayRef<typename ELFT::Shdr> Sections = CHECK(Obj.sections(), this); 1216 1217 for (const typename ELFT::Shdr &Sec : Sections) { 1218 if (Sec.sh_type != SHT_SYMTAB) 1219 continue; 1220 1221 typename ELFT::SymRange Syms = CHECK(Obj.symbols(&Sec), this); 1222 uint32_t FirstGlobal = Sec.sh_info; 1223 StringRef StringTable = 1224 CHECK(Obj.getStringTableForSymtab(Sec, Sections), this); 1225 1226 for (const typename ELFT::Sym &Sym : Syms.slice(FirstGlobal)) 1227 if (Sym.st_shndx != SHN_UNDEF) 1228 Symtab->addLazyObject<ELFT>(CHECK(Sym.getName(StringTable), this), 1229 *this); 1230 return; 1231 } 1232 } 1233 1234 template void ArchiveFile::parse<ELF32LE>(); 1235 template void ArchiveFile::parse<ELF32BE>(); 1236 template void ArchiveFile::parse<ELF64LE>(); 1237 template void ArchiveFile::parse<ELF64BE>(); 1238 1239 template void BitcodeFile::parse<ELF32LE>(DenseSet<CachedHashStringRef> &); 1240 template void BitcodeFile::parse<ELF32BE>(DenseSet<CachedHashStringRef> &); 1241 template void BitcodeFile::parse<ELF64LE>(DenseSet<CachedHashStringRef> &); 1242 template void BitcodeFile::parse<ELF64BE>(DenseSet<CachedHashStringRef> &); 1243 1244 template void LazyObjFile::parse<ELF32LE>(); 1245 template void LazyObjFile::parse<ELF32BE>(); 1246 template void LazyObjFile::parse<ELF64LE>(); 1247 template void LazyObjFile::parse<ELF64BE>(); 1248 1249 template class elf::ELFFileBase<ELF32LE>; 1250 template class elf::ELFFileBase<ELF32BE>; 1251 template class elf::ELFFileBase<ELF64LE>; 1252 template class elf::ELFFileBase<ELF64BE>; 1253 1254 template class elf::ObjFile<ELF32LE>; 1255 template class elf::ObjFile<ELF32BE>; 1256 template class elf::ObjFile<ELF64LE>; 1257 template class elf::ObjFile<ELF64BE>; 1258 1259 template class elf::SharedFile<ELF32LE>; 1260 template class elf::SharedFile<ELF32BE>; 1261 template class elf::SharedFile<ELF64LE>; 1262 template class elf::SharedFile<ELF64BE>; 1263