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 Expected<const DWARFDebugLine::LineTable *> ExpectedLT = 135 Dwarf->getLineTableForUnit(CU.get(), warn); 136 const DWARFDebugLine::LineTable *LT = nullptr; 137 if (ExpectedLT) 138 LT = *ExpectedLT; 139 else 140 handleAllErrors(ExpectedLT.takeError(), 141 [](ErrorInfoBase &Err) { warn(Err.message()); }); 142 if (!LT) 143 continue; 144 LineTables.push_back(LT); 145 146 // Loop over variable records and insert them to VariableLoc. 147 for (const auto &Entry : CU->dies()) { 148 DWARFDie Die(CU.get(), &Entry); 149 // Skip all tags that are not variables. 150 if (Die.getTag() != dwarf::DW_TAG_variable) 151 continue; 152 153 // Skip if a local variable because we don't need them for generating 154 // error messages. In general, only non-local symbols can fail to be 155 // linked. 156 if (!dwarf::toUnsigned(Die.find(dwarf::DW_AT_external), 0)) 157 continue; 158 159 // Get the source filename index for the variable. 160 unsigned File = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_file), 0); 161 if (!LT->hasFileAtIndex(File)) 162 continue; 163 164 // Get the line number on which the variable is declared. 165 unsigned Line = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_line), 0); 166 167 // Get the name of the variable and add the collected information to 168 // VariableLoc. Usually Name is non-empty, but it can be empty if the 169 // input object file lacks some debug info. 170 StringRef Name = dwarf::toString(Die.find(dwarf::DW_AT_name), ""); 171 if (!Name.empty()) 172 VariableLoc.insert({Name, {LT, File, Line}}); 173 } 174 } 175 } 176 177 // Returns the pair of file name and line number describing location of data 178 // object (variable, array, etc) definition. 179 template <class ELFT> 180 Optional<std::pair<std::string, unsigned>> 181 ObjFile<ELFT>::getVariableLoc(StringRef Name) { 182 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 183 184 // Return if we have no debug information about data object. 185 auto It = VariableLoc.find(Name); 186 if (It == VariableLoc.end()) 187 return None; 188 189 // Take file name string from line table. 190 std::string FileName; 191 if (!It->second.LT->getFileNameByIndex( 192 It->second.File, nullptr, 193 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FileName)) 194 return None; 195 196 return std::make_pair(FileName, It->second.Line); 197 } 198 199 // Returns source line information for a given offset 200 // using DWARF debug info. 201 template <class ELFT> 202 Optional<DILineInfo> ObjFile<ELFT>::getDILineInfo(InputSectionBase *S, 203 uint64_t Offset) { 204 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 205 206 // Use fake address calcuated by adding section file offset and offset in 207 // section. See comments for ObjectInfo class. 208 DILineInfo Info; 209 for (const llvm::DWARFDebugLine::LineTable *LT : LineTables) 210 if (LT->getFileLineInfoForAddress( 211 S->getOffsetInFile() + Offset, nullptr, 212 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, Info)) 213 return Info; 214 return None; 215 } 216 217 // Returns source line information for a given offset using DWARF debug info. 218 template <class ELFT> 219 std::string ObjFile<ELFT>::getLineInfo(InputSectionBase *S, uint64_t Offset) { 220 if (Optional<DILineInfo> Info = getDILineInfo(S, Offset)) 221 return Info->FileName + ":" + std::to_string(Info->Line); 222 return ""; 223 } 224 225 // Returns "<internal>", "foo.a(bar.o)" or "baz.o". 226 std::string lld::toString(const InputFile *F) { 227 if (!F) 228 return "<internal>"; 229 230 if (F->ToStringCache.empty()) { 231 if (F->ArchiveName.empty()) 232 F->ToStringCache = F->getName(); 233 else 234 F->ToStringCache = (F->ArchiveName + "(" + F->getName() + ")").str(); 235 } 236 return F->ToStringCache; 237 } 238 239 template <class ELFT> 240 ELFFileBase<ELFT>::ELFFileBase(Kind K, MemoryBufferRef MB) : InputFile(K, MB) { 241 if (ELFT::TargetEndianness == support::little) 242 EKind = ELFT::Is64Bits ? ELF64LEKind : ELF32LEKind; 243 else 244 EKind = ELFT::Is64Bits ? ELF64BEKind : ELF32BEKind; 245 246 EMachine = getObj().getHeader()->e_machine; 247 OSABI = getObj().getHeader()->e_ident[llvm::ELF::EI_OSABI]; 248 } 249 250 template <class ELFT> 251 typename ELFT::SymRange ELFFileBase<ELFT>::getGlobalELFSyms() { 252 return makeArrayRef(ELFSyms.begin() + FirstGlobal, ELFSyms.end()); 253 } 254 255 template <class ELFT> 256 uint32_t ELFFileBase<ELFT>::getSectionIndex(const Elf_Sym &Sym) const { 257 return CHECK(getObj().getSectionIndex(&Sym, ELFSyms, SymtabSHNDX), this); 258 } 259 260 template <class ELFT> 261 void ELFFileBase<ELFT>::initSymtab(ArrayRef<Elf_Shdr> Sections, 262 const Elf_Shdr *Symtab) { 263 FirstGlobal = Symtab->sh_info; 264 ELFSyms = CHECK(getObj().symbols(Symtab), this); 265 if (FirstGlobal == 0 || FirstGlobal > ELFSyms.size()) 266 fatal(toString(this) + ": invalid sh_info in symbol table"); 267 268 StringTable = 269 CHECK(getObj().getStringTableForSymtab(*Symtab, Sections), this); 270 } 271 272 template <class ELFT> 273 ObjFile<ELFT>::ObjFile(MemoryBufferRef M, StringRef ArchiveName) 274 : ELFFileBase<ELFT>(Base::ObjKind, M) { 275 this->ArchiveName = ArchiveName; 276 } 277 278 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() { 279 if (this->Symbols.empty()) 280 return {}; 281 return makeArrayRef(this->Symbols).slice(1, this->FirstGlobal - 1); 282 } 283 284 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getGlobalSymbols() { 285 return makeArrayRef(this->Symbols).slice(this->FirstGlobal); 286 } 287 288 template <class ELFT> 289 void ObjFile<ELFT>::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 290 // Read a section table. JustSymbols is usually false. 291 if (this->JustSymbols) 292 initializeJustSymbols(); 293 else 294 initializeSections(ComdatGroups); 295 296 // Read a symbol table. 297 initializeSymbols(); 298 } 299 300 // Sections with SHT_GROUP and comdat bits define comdat section groups. 301 // They are identified and deduplicated by group name. This function 302 // returns a group name. 303 template <class ELFT> 304 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> Sections, 305 const Elf_Shdr &Sec) { 306 // Group signatures are stored as symbol names in object files. 307 // sh_info contains a symbol index, so we fetch a symbol and read its name. 308 if (this->ELFSyms.empty()) 309 this->initSymtab( 310 Sections, CHECK(object::getSection<ELFT>(Sections, Sec.sh_link), this)); 311 312 const Elf_Sym *Sym = 313 CHECK(object::getSymbol<ELFT>(this->ELFSyms, Sec.sh_info), this); 314 StringRef Signature = CHECK(Sym->getName(this->StringTable), this); 315 316 // As a special case, if a symbol is a section symbol and has no name, 317 // we use a section name as a signature. 318 // 319 // Such SHT_GROUP sections are invalid from the perspective of the ELF 320 // standard, but GNU gold 1.14 (the newest version as of July 2017) or 321 // older produce such sections as outputs for the -r option, so we need 322 // a bug-compatibility. 323 if (Signature.empty() && Sym->getType() == STT_SECTION) 324 return getSectionName(Sec); 325 return Signature; 326 } 327 328 template <class ELFT> 329 ArrayRef<typename ObjFile<ELFT>::Elf_Word> 330 ObjFile<ELFT>::getShtGroupEntries(const Elf_Shdr &Sec) { 331 const ELFFile<ELFT> &Obj = this->getObj(); 332 ArrayRef<Elf_Word> Entries = 333 CHECK(Obj.template getSectionContentsAsArray<Elf_Word>(&Sec), this); 334 if (Entries.empty() || Entries[0] != GRP_COMDAT) 335 fatal(toString(this) + ": unsupported SHT_GROUP format"); 336 return Entries.slice(1); 337 } 338 339 template <class ELFT> bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &Sec) { 340 // On a regular link we don't merge sections if -O0 (default is -O1). This 341 // sometimes makes the linker significantly faster, although the output will 342 // be bigger. 343 // 344 // Doing the same for -r would create a problem as it would combine sections 345 // with different sh_entsize. One option would be to just copy every SHF_MERGE 346 // section as is to the output. While this would produce a valid ELF file with 347 // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when 348 // they see two .debug_str. We could have separate logic for combining 349 // SHF_MERGE sections based both on their name and sh_entsize, but that seems 350 // to be more trouble than it is worth. Instead, we just use the regular (-O1) 351 // logic for -r. 352 if (Config->Optimize == 0 && !Config->Relocatable) 353 return false; 354 355 // A mergeable section with size 0 is useless because they don't have 356 // any data to merge. A mergeable string section with size 0 can be 357 // argued as invalid because it doesn't end with a null character. 358 // We'll avoid a mess by handling them as if they were non-mergeable. 359 if (Sec.sh_size == 0) 360 return false; 361 362 // Check for sh_entsize. The ELF spec is not clear about the zero 363 // sh_entsize. It says that "the member [sh_entsize] contains 0 if 364 // the section does not hold a table of fixed-size entries". We know 365 // that Rust 1.13 produces a string mergeable section with a zero 366 // sh_entsize. Here we just accept it rather than being picky about it. 367 uint64_t EntSize = Sec.sh_entsize; 368 if (EntSize == 0) 369 return false; 370 if (Sec.sh_size % EntSize) 371 fatal(toString(this) + 372 ": SHF_MERGE section size must be a multiple of sh_entsize"); 373 374 uint64_t Flags = Sec.sh_flags; 375 if (!(Flags & SHF_MERGE)) 376 return false; 377 if (Flags & SHF_WRITE) 378 fatal(toString(this) + ": writable SHF_MERGE section is not supported"); 379 380 return true; 381 } 382 383 // This is for --just-symbols. 384 // 385 // --just-symbols is a very minor feature that allows you to link your 386 // output against other existing program, so that if you load both your 387 // program and the other program into memory, your output can refer the 388 // other program's symbols. 389 // 390 // When the option is given, we link "just symbols". The section table is 391 // initialized with null pointers. 392 template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() { 393 ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this); 394 this->Sections.resize(ObjSections.size()); 395 396 for (const Elf_Shdr &Sec : ObjSections) { 397 if (Sec.sh_type != SHT_SYMTAB) 398 continue; 399 this->initSymtab(ObjSections, &Sec); 400 return; 401 } 402 } 403 404 template <class ELFT> 405 void ObjFile<ELFT>::initializeSections( 406 DenseSet<CachedHashStringRef> &ComdatGroups) { 407 const ELFFile<ELFT> &Obj = this->getObj(); 408 409 ArrayRef<Elf_Shdr> ObjSections = CHECK(Obj.sections(), this); 410 uint64_t Size = ObjSections.size(); 411 this->Sections.resize(Size); 412 this->SectionStringTable = 413 CHECK(Obj.getSectionStringTable(ObjSections), this); 414 415 for (size_t I = 0, E = ObjSections.size(); I < E; I++) { 416 if (this->Sections[I] == &InputSection::Discarded) 417 continue; 418 const Elf_Shdr &Sec = ObjSections[I]; 419 420 // SHF_EXCLUDE'ed sections are discarded by the linker. However, 421 // if -r is given, we'll let the final link discard such sections. 422 // This is compatible with GNU. 423 if ((Sec.sh_flags & SHF_EXCLUDE) && !Config->Relocatable) { 424 this->Sections[I] = &InputSection::Discarded; 425 continue; 426 } 427 428 switch (Sec.sh_type) { 429 case SHT_GROUP: { 430 // De-duplicate section groups by their signatures. 431 StringRef Signature = getShtGroupSignature(ObjSections, Sec); 432 bool IsNew = ComdatGroups.insert(CachedHashStringRef(Signature)).second; 433 this->Sections[I] = &InputSection::Discarded; 434 435 // If it is a new section group, we want to keep group members. 436 // Group leader sections, which contain indices of group members, are 437 // discarded because they are useless beyond this point. The only 438 // exception is the -r option because in order to produce re-linkable 439 // object files, we want to pass through basically everything. 440 if (IsNew) { 441 if (Config->Relocatable) 442 this->Sections[I] = createInputSection(Sec); 443 continue; 444 } 445 446 // Otherwise, discard group members. 447 for (uint32_t SecIndex : getShtGroupEntries(Sec)) { 448 if (SecIndex >= Size) 449 fatal(toString(this) + 450 ": invalid section index in group: " + Twine(SecIndex)); 451 this->Sections[SecIndex] = &InputSection::Discarded; 452 } 453 break; 454 } 455 case SHT_SYMTAB: 456 this->initSymtab(ObjSections, &Sec); 457 break; 458 case SHT_SYMTAB_SHNDX: 459 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, ObjSections), this); 460 break; 461 case SHT_STRTAB: 462 case SHT_NULL: 463 break; 464 default: 465 this->Sections[I] = createInputSection(Sec); 466 } 467 468 // .ARM.exidx sections have a reverse dependency on the InputSection they 469 // have a SHF_LINK_ORDER dependency, this is identified by the sh_link. 470 if (Sec.sh_flags & SHF_LINK_ORDER) { 471 if (Sec.sh_link >= this->Sections.size()) 472 fatal(toString(this) + 473 ": invalid sh_link index: " + Twine(Sec.sh_link)); 474 475 InputSectionBase *LinkSec = this->Sections[Sec.sh_link]; 476 InputSection *IS = cast<InputSection>(this->Sections[I]); 477 LinkSec->DependentSections.push_back(IS); 478 if (!isa<InputSection>(LinkSec)) 479 error("a section " + IS->Name + 480 " with SHF_LINK_ORDER should not refer a non-regular " 481 "section: " + 482 toString(LinkSec)); 483 } 484 } 485 } 486 487 // The ARM support in lld makes some use of instructions that are not available 488 // on all ARM architectures. Namely: 489 // - Use of BLX instruction for interworking between ARM and Thumb state. 490 // - Use of the extended Thumb branch encoding in relocation. 491 // - Use of the MOVT/MOVW instructions in Thumb Thunks. 492 // The ARM Attributes section contains information about the architecture chosen 493 // at compile time. We follow the convention that if at least one input object 494 // is compiled with an architecture that supports these features then lld is 495 // permitted to use them. 496 static void updateSupportedARMFeatures(const ARMAttributeParser &Attributes) { 497 if (!Attributes.hasAttribute(ARMBuildAttrs::CPU_arch)) 498 return; 499 auto Arch = Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch); 500 switch (Arch) { 501 case ARMBuildAttrs::Pre_v4: 502 case ARMBuildAttrs::v4: 503 case ARMBuildAttrs::v4T: 504 // Architectures prior to v5 do not support BLX instruction 505 break; 506 case ARMBuildAttrs::v5T: 507 case ARMBuildAttrs::v5TE: 508 case ARMBuildAttrs::v5TEJ: 509 case ARMBuildAttrs::v6: 510 case ARMBuildAttrs::v6KZ: 511 case ARMBuildAttrs::v6K: 512 Config->ARMHasBlx = true; 513 // Architectures used in pre-Cortex processors do not support 514 // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception 515 // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do. 516 break; 517 default: 518 // All other Architectures have BLX and extended branch encoding 519 Config->ARMHasBlx = true; 520 Config->ARMJ1J2BranchEncoding = true; 521 if (Arch != ARMBuildAttrs::v6_M && Arch != ARMBuildAttrs::v6S_M) 522 // All Architectures used in Cortex processors with the exception 523 // of v6-M and v6S-M have the MOVT and MOVW instructions. 524 Config->ARMHasMovtMovw = true; 525 break; 526 } 527 } 528 529 template <class ELFT> 530 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &Sec) { 531 uint32_t Idx = Sec.sh_info; 532 if (Idx >= this->Sections.size()) 533 fatal(toString(this) + ": invalid relocated section index: " + Twine(Idx)); 534 InputSectionBase *Target = this->Sections[Idx]; 535 536 // Strictly speaking, a relocation section must be included in the 537 // group of the section it relocates. However, LLVM 3.3 and earlier 538 // would fail to do so, so we gracefully handle that case. 539 if (Target == &InputSection::Discarded) 540 return nullptr; 541 542 if (!Target) 543 fatal(toString(this) + ": unsupported relocation reference"); 544 return Target; 545 } 546 547 // Create a regular InputSection class that has the same contents 548 // as a given section. 549 static InputSection *toRegularSection(MergeInputSection *Sec) { 550 return make<InputSection>(Sec->File, Sec->Flags, Sec->Type, Sec->Alignment, 551 Sec->Data, Sec->Name); 552 } 553 554 template <class ELFT> 555 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &Sec) { 556 StringRef Name = getSectionName(Sec); 557 558 switch (Sec.sh_type) { 559 case SHT_ARM_ATTRIBUTES: { 560 if (Config->EMachine != EM_ARM) 561 break; 562 ARMAttributeParser Attributes; 563 ArrayRef<uint8_t> Contents = check(this->getObj().getSectionContents(&Sec)); 564 Attributes.Parse(Contents, /*isLittle*/ Config->EKind == ELF32LEKind); 565 updateSupportedARMFeatures(Attributes); 566 // FIXME: Retain the first attribute section we see. The eglibc ARM 567 // dynamic loaders require the presence of an attribute section for dlopen 568 // to work. In a full implementation we would merge all attribute sections. 569 if (InX::ARMAttributes == nullptr) { 570 InX::ARMAttributes = make<InputSection>(*this, Sec, Name); 571 return InX::ARMAttributes; 572 } 573 return &InputSection::Discarded; 574 } 575 case SHT_RELA: 576 case SHT_REL: { 577 // Find a relocation target section and associate this section with that. 578 // Target may have been discarded if it is in a different section group 579 // and the group is discarded, even though it's a violation of the 580 // spec. We handle that situation gracefully by discarding dangling 581 // relocation sections. 582 InputSectionBase *Target = getRelocTarget(Sec); 583 if (!Target) 584 return nullptr; 585 586 // This section contains relocation information. 587 // If -r is given, we do not interpret or apply relocation 588 // but just copy relocation sections to output. 589 if (Config->Relocatable) 590 return make<InputSection>(*this, Sec, Name); 591 592 if (Target->FirstRelocation) 593 fatal(toString(this) + 594 ": multiple relocation sections to one section are not supported"); 595 596 // ELF spec allows mergeable sections with relocations, but they are 597 // rare, and it is in practice hard to merge such sections by contents, 598 // because applying relocations at end of linking changes section 599 // contents. So, we simply handle such sections as non-mergeable ones. 600 // Degrading like this is acceptable because section merging is optional. 601 if (auto *MS = dyn_cast<MergeInputSection>(Target)) { 602 Target = toRegularSection(MS); 603 this->Sections[Sec.sh_info] = Target; 604 } 605 606 if (Sec.sh_type == SHT_RELA) { 607 ArrayRef<Elf_Rela> Rels = CHECK(this->getObj().relas(&Sec), this); 608 Target->FirstRelocation = Rels.begin(); 609 Target->NumRelocations = Rels.size(); 610 Target->AreRelocsRela = true; 611 } else { 612 ArrayRef<Elf_Rel> Rels = CHECK(this->getObj().rels(&Sec), this); 613 Target->FirstRelocation = Rels.begin(); 614 Target->NumRelocations = Rels.size(); 615 Target->AreRelocsRela = false; 616 } 617 assert(isUInt<31>(Target->NumRelocations)); 618 619 // Relocation sections processed by the linker are usually removed 620 // from the output, so returning `nullptr` for the normal case. 621 // However, if -emit-relocs is given, we need to leave them in the output. 622 // (Some post link analysis tools need this information.) 623 if (Config->EmitRelocs) { 624 InputSection *RelocSec = make<InputSection>(*this, Sec, Name); 625 // We will not emit relocation section if target was discarded. 626 Target->DependentSections.push_back(RelocSec); 627 return RelocSec; 628 } 629 return nullptr; 630 } 631 } 632 633 // The GNU linker uses .note.GNU-stack section as a marker indicating 634 // that the code in the object file does not expect that the stack is 635 // executable (in terms of NX bit). If all input files have the marker, 636 // the GNU linker adds a PT_GNU_STACK segment to tells the loader to 637 // make the stack non-executable. Most object files have this section as 638 // of 2017. 639 // 640 // But making the stack non-executable is a norm today for security 641 // reasons. Failure to do so may result in a serious security issue. 642 // Therefore, we make LLD always add PT_GNU_STACK unless it is 643 // explicitly told to do otherwise (by -z execstack). Because the stack 644 // executable-ness is controlled solely by command line options, 645 // .note.GNU-stack sections are simply ignored. 646 if (Name == ".note.GNU-stack") 647 return &InputSection::Discarded; 648 649 // Split stacks is a feature to support a discontiguous stack. At least 650 // as of 2017, it seems that the feature is not being used widely. 651 // Only GNU gold supports that. We don't. For the details about that, 652 // see https://gcc.gnu.org/wiki/SplitStacks 653 if (Name == ".note.GNU-split-stack") { 654 error(toString(this) + 655 ": object file compiled with -fsplit-stack is not supported"); 656 return &InputSection::Discarded; 657 } 658 659 // The linkonce feature is a sort of proto-comdat. Some glibc i386 object 660 // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce 661 // sections. Drop those sections to avoid duplicate symbol errors. 662 // FIXME: This is glibc PR20543, we should remove this hack once that has been 663 // fixed for a while. 664 if (Name.startswith(".gnu.linkonce.")) 665 return &InputSection::Discarded; 666 667 // If we are creating a new .build-id section, strip existing .build-id 668 // sections so that the output won't have more than one .build-id. 669 // This is not usually a problem because input object files normally don't 670 // have .build-id sections, but you can create such files by 671 // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it. 672 if (Name == ".note.gnu.build-id" && Config->BuildId != BuildIdKind::None) 673 return &InputSection::Discarded; 674 675 // The linker merges EH (exception handling) frames and creates a 676 // .eh_frame_hdr section for runtime. So we handle them with a special 677 // class. For relocatable outputs, they are just passed through. 678 if (Name == ".eh_frame" && !Config->Relocatable) 679 return make<EhInputSection>(*this, Sec, Name); 680 681 if (shouldMerge(Sec)) 682 return make<MergeInputSection>(*this, Sec, Name); 683 return make<InputSection>(*this, Sec, Name); 684 } 685 686 template <class ELFT> 687 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &Sec) { 688 return CHECK(this->getObj().getSectionName(&Sec, SectionStringTable), this); 689 } 690 691 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() { 692 this->Symbols.reserve(this->ELFSyms.size()); 693 for (const Elf_Sym &Sym : this->ELFSyms) 694 this->Symbols.push_back(createSymbol(&Sym)); 695 } 696 697 template <class ELFT> Symbol *ObjFile<ELFT>::createSymbol(const Elf_Sym *Sym) { 698 int Binding = Sym->getBinding(); 699 700 uint32_t SecIdx = this->getSectionIndex(*Sym); 701 if (SecIdx >= this->Sections.size()) 702 fatal(toString(this) + ": invalid section index: " + Twine(SecIdx)); 703 704 InputSectionBase *Sec = this->Sections[SecIdx]; 705 uint8_t StOther = Sym->st_other; 706 uint8_t Type = Sym->getType(); 707 uint64_t Value = Sym->st_value; 708 uint64_t Size = Sym->st_size; 709 710 if (Binding == STB_LOCAL) { 711 if (Sym->getType() == STT_FILE) 712 SourceFile = CHECK(Sym->getName(this->StringTable), this); 713 714 if (this->StringTable.size() <= Sym->st_name) 715 fatal(toString(this) + ": invalid symbol name offset"); 716 717 StringRefZ Name = this->StringTable.data() + Sym->st_name; 718 if (Sym->st_shndx == SHN_UNDEF) 719 return make<Undefined>(this, Name, Binding, StOther, Type); 720 721 return make<Defined>(this, Name, Binding, StOther, Type, Value, Size, Sec); 722 } 723 724 StringRef Name = CHECK(Sym->getName(this->StringTable), this); 725 726 switch (Sym->st_shndx) { 727 case SHN_UNDEF: 728 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 729 /*CanOmitFromDynSym=*/false, this); 730 case SHN_COMMON: 731 if (Value == 0 || Value >= UINT32_MAX) 732 fatal(toString(this) + ": common symbol '" + Name + 733 "' has invalid alignment: " + Twine(Value)); 734 return Symtab->addCommon(Name, Size, Value, Binding, StOther, Type, *this); 735 } 736 737 switch (Binding) { 738 default: 739 fatal(toString(this) + ": unexpected binding: " + Twine(Binding)); 740 case STB_GLOBAL: 741 case STB_WEAK: 742 case STB_GNU_UNIQUE: 743 if (Sec == &InputSection::Discarded) 744 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 745 /*CanOmitFromDynSym=*/false, this); 746 return Symtab->addRegular(Name, StOther, Type, Value, Size, Binding, Sec, 747 this); 748 } 749 } 750 751 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&File) 752 : InputFile(ArchiveKind, File->getMemoryBufferRef()), 753 File(std::move(File)) {} 754 755 template <class ELFT> void ArchiveFile::parse() { 756 for (const Archive::Symbol &Sym : File->symbols()) 757 Symtab->addLazyArchive<ELFT>(Sym.getName(), *this, Sym); 758 } 759 760 // Returns a buffer pointing to a member file containing a given symbol. 761 InputFile *ArchiveFile::fetch(const Archive::Symbol &Sym) { 762 Archive::Child C = 763 CHECK(Sym.getMember(), toString(this) + 764 ": could not get the member for symbol " + 765 Sym.getName()); 766 767 if (!Seen.insert(C.getChildOffset()).second) 768 return nullptr; 769 770 MemoryBufferRef MB = 771 CHECK(C.getMemoryBufferRef(), 772 toString(this) + 773 ": could not get the buffer for the member defining symbol " + 774 Sym.getName()); 775 776 if (Tar && C.getParent()->isThin()) 777 Tar->append(relativeToRoot(CHECK(C.getFullName(), this)), MB.getBuffer()); 778 779 InputFile *File = createObjectFile( 780 MB, getName(), C.getParent()->isThin() ? 0 : C.getChildOffset()); 781 File->GroupId = GroupId; 782 return File; 783 } 784 785 template <class ELFT> 786 SharedFile<ELFT>::SharedFile(MemoryBufferRef M, StringRef DefaultSoName) 787 : ELFFileBase<ELFT>(Base::SharedKind, M), SoName(DefaultSoName), 788 IsNeeded(!Config->AsNeeded) {} 789 790 // Partially parse the shared object file so that we can call 791 // getSoName on this object. 792 template <class ELFT> void SharedFile<ELFT>::parseSoName() { 793 const Elf_Shdr *DynamicSec = nullptr; 794 const ELFFile<ELFT> Obj = this->getObj(); 795 ArrayRef<Elf_Shdr> Sections = CHECK(Obj.sections(), this); 796 797 // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d. 798 for (const Elf_Shdr &Sec : Sections) { 799 switch (Sec.sh_type) { 800 default: 801 continue; 802 case SHT_DYNSYM: 803 this->initSymtab(Sections, &Sec); 804 break; 805 case SHT_DYNAMIC: 806 DynamicSec = &Sec; 807 break; 808 case SHT_SYMTAB_SHNDX: 809 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, Sections), this); 810 break; 811 case SHT_GNU_versym: 812 this->VersymSec = &Sec; 813 break; 814 case SHT_GNU_verdef: 815 this->VerdefSec = &Sec; 816 break; 817 } 818 } 819 820 if (this->VersymSec && this->ELFSyms.empty()) 821 error("SHT_GNU_versym should be associated with symbol table"); 822 823 // Search for a DT_SONAME tag to initialize this->SoName. 824 if (!DynamicSec) 825 return; 826 ArrayRef<Elf_Dyn> Arr = 827 CHECK(Obj.template getSectionContentsAsArray<Elf_Dyn>(DynamicSec), this); 828 for (const Elf_Dyn &Dyn : Arr) { 829 if (Dyn.d_tag == DT_SONAME) { 830 uint64_t Val = Dyn.getVal(); 831 if (Val >= this->StringTable.size()) 832 fatal(toString(this) + ": invalid DT_SONAME entry"); 833 SoName = this->StringTable.data() + Val; 834 return; 835 } 836 } 837 } 838 839 // Parses ".gnu.version" section which is a parallel array for the symbol table. 840 // If a given file doesn't have ".gnu.version" section, returns VER_NDX_GLOBAL. 841 template <class ELFT> std::vector<uint32_t> SharedFile<ELFT>::parseVersyms() { 842 size_t Size = this->ELFSyms.size() - this->FirstGlobal; 843 if (!VersymSec) 844 return std::vector<uint32_t>(Size, VER_NDX_GLOBAL); 845 846 const char *Base = this->MB.getBuffer().data(); 847 const Elf_Versym *Versym = 848 reinterpret_cast<const Elf_Versym *>(Base + VersymSec->sh_offset) + 849 this->FirstGlobal; 850 851 std::vector<uint32_t> Ret(Size); 852 for (size_t I = 0; I < Size; ++I) 853 Ret[I] = Versym[I].vs_index; 854 return Ret; 855 } 856 857 // Parse the version definitions in the object file if present. Returns a vector 858 // whose nth element contains a pointer to the Elf_Verdef for version identifier 859 // n. Version identifiers that are not definitions map to nullptr. 860 template <class ELFT> 861 std::vector<const typename ELFT::Verdef *> SharedFile<ELFT>::parseVerdefs() { 862 if (!VerdefSec) 863 return {}; 864 865 // We cannot determine the largest verdef identifier without inspecting 866 // every Elf_Verdef, but both bfd and gold assign verdef identifiers 867 // sequentially starting from 1, so we predict that the largest identifier 868 // will be VerdefCount. 869 unsigned VerdefCount = VerdefSec->sh_info; 870 std::vector<const Elf_Verdef *> Verdefs(VerdefCount + 1); 871 872 // Build the Verdefs array by following the chain of Elf_Verdef objects 873 // from the start of the .gnu.version_d section. 874 const char *Base = this->MB.getBuffer().data(); 875 const char *Verdef = Base + VerdefSec->sh_offset; 876 for (unsigned I = 0; I != VerdefCount; ++I) { 877 auto *CurVerdef = reinterpret_cast<const Elf_Verdef *>(Verdef); 878 Verdef += CurVerdef->vd_next; 879 unsigned VerdefIndex = CurVerdef->vd_ndx; 880 if (Verdefs.size() <= VerdefIndex) 881 Verdefs.resize(VerdefIndex + 1); 882 Verdefs[VerdefIndex] = CurVerdef; 883 } 884 885 return Verdefs; 886 } 887 888 // We do not usually care about alignments of data in shared object 889 // files because the loader takes care of it. However, if we promote a 890 // DSO symbol to point to .bss due to copy relocation, we need to keep 891 // the original alignment requirements. We infer it in this function. 892 template <class ELFT> 893 uint32_t SharedFile<ELFT>::getAlignment(ArrayRef<Elf_Shdr> Sections, 894 const Elf_Sym &Sym) { 895 uint64_t Ret = 1; 896 if (Sym.st_value) 897 Ret = 1ULL << countTrailingZeros((uint64_t)Sym.st_value); 898 if (0 < Sym.st_shndx && Sym.st_shndx < Sections.size()) 899 Ret = std::min<uint64_t>(Ret, Sections[Sym.st_shndx].sh_addralign); 900 901 if (Ret > UINT32_MAX) 902 error(toString(this) + ": alignment too large: " + 903 CHECK(Sym.getName(this->StringTable), this)); 904 return Ret; 905 } 906 907 // Fully parse the shared object file. This must be called after parseSoName(). 908 // 909 // This function parses symbol versions. If a DSO has version information, 910 // the file has a ".gnu.version_d" section which contains symbol version 911 // definitions. Each symbol is associated to one version through a table in 912 // ".gnu.version" section. That table is a parallel array for the symbol 913 // table, and each table entry contains an index in ".gnu.version_d". 914 // 915 // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for 916 // VER_NDX_GLOBAL. There's no table entry for these special versions in 917 // ".gnu.version_d". 918 // 919 // The file format for symbol versioning is perhaps a bit more complicated 920 // than necessary, but you can easily understand the code if you wrap your 921 // head around the data structure described above. 922 template <class ELFT> void SharedFile<ELFT>::parseRest() { 923 Verdefs = parseVerdefs(); // parse .gnu.version_d 924 std::vector<uint32_t> Versyms = parseVersyms(); // parse .gnu.version 925 ArrayRef<Elf_Shdr> Sections = CHECK(this->getObj().sections(), this); 926 927 // System libraries can have a lot of symbols with versions. Using a 928 // fixed buffer for computing the versions name (foo@ver) can save a 929 // lot of allocations. 930 SmallString<0> VersionedNameBuffer; 931 932 // Add symbols to the symbol table. 933 ArrayRef<Elf_Sym> Syms = this->getGlobalELFSyms(); 934 for (size_t I = 0; I < Syms.size(); ++I) { 935 const Elf_Sym &Sym = Syms[I]; 936 937 StringRef Name = CHECK(Sym.getName(this->StringTable), this); 938 if (Sym.isUndefined()) { 939 Symbol *S = Symtab->addUndefined<ELFT>(Name, Sym.getBinding(), 940 Sym.st_other, Sym.getType(), 941 /*CanOmitFromDynSym=*/false, this); 942 S->ExportDynamic = true; 943 continue; 944 } 945 946 // ELF spec requires that all local symbols precede weak or global 947 // symbols in each symbol table, and the index of first non-local symbol 948 // is stored to sh_info. If a local symbol appears after some non-local 949 // symbol, that's a violation of the spec. 950 if (Sym.getBinding() == STB_LOCAL) { 951 warn("found local symbol '" + Name + 952 "' in global part of symbol table in file " + toString(this)); 953 continue; 954 } 955 956 // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly 957 // assigns VER_NDX_LOCAL to this section global symbol. Here is a 958 // workaround for this bug. 959 uint32_t Idx = Versyms[I] & ~VERSYM_HIDDEN; 960 if (Config->EMachine == EM_MIPS && Idx == VER_NDX_LOCAL && 961 Name == "_gp_disp") 962 continue; 963 964 uint64_t Alignment = getAlignment(Sections, Sym); 965 if (!(Versyms[I] & VERSYM_HIDDEN)) 966 Symtab->addShared(Name, *this, Sym, Alignment, Idx); 967 968 // Also add the symbol with the versioned name to handle undefined symbols 969 // with explicit versions. 970 if (Idx == VER_NDX_GLOBAL) 971 continue; 972 973 if (Idx >= Verdefs.size() || Idx == VER_NDX_LOCAL) { 974 error("corrupt input file: version definition index " + Twine(Idx) + 975 " for symbol " + Name + " is out of bounds\n>>> defined in " + 976 toString(this)); 977 continue; 978 } 979 980 StringRef VerName = 981 this->StringTable.data() + Verdefs[Idx]->getAux()->vda_name; 982 VersionedNameBuffer.clear(); 983 Name = (Name + "@" + VerName).toStringRef(VersionedNameBuffer); 984 Symtab->addShared(Saver.save(Name), *this, Sym, Alignment, Idx); 985 } 986 } 987 988 static ELFKind getBitcodeELFKind(const Triple &T) { 989 if (T.isLittleEndian()) 990 return T.isArch64Bit() ? ELF64LEKind : ELF32LEKind; 991 return T.isArch64Bit() ? ELF64BEKind : ELF32BEKind; 992 } 993 994 static uint8_t getBitcodeMachineKind(StringRef Path, const Triple &T) { 995 switch (T.getArch()) { 996 case Triple::aarch64: 997 return EM_AARCH64; 998 case Triple::arm: 999 case Triple::thumb: 1000 return EM_ARM; 1001 case Triple::avr: 1002 return EM_AVR; 1003 case Triple::mips: 1004 case Triple::mipsel: 1005 case Triple::mips64: 1006 case Triple::mips64el: 1007 return EM_MIPS; 1008 case Triple::ppc: 1009 return EM_PPC; 1010 case Triple::ppc64: 1011 return EM_PPC64; 1012 case Triple::x86: 1013 return T.isOSIAMCU() ? EM_IAMCU : EM_386; 1014 case Triple::x86_64: 1015 return EM_X86_64; 1016 default: 1017 fatal(Path + ": could not infer e_machine from bitcode target triple " + 1018 T.str()); 1019 } 1020 } 1021 1022 BitcodeFile::BitcodeFile(MemoryBufferRef MB, StringRef ArchiveName, 1023 uint64_t OffsetInArchive) 1024 : InputFile(BitcodeKind, MB) { 1025 this->ArchiveName = ArchiveName; 1026 1027 std::string Path = MB.getBufferIdentifier().str(); 1028 if (Config->ThinLTOIndexOnly) 1029 Path = replaceThinLTOSuffix(MB.getBufferIdentifier()); 1030 1031 // ThinLTO assumes that all MemoryBufferRefs given to it have a unique 1032 // name. If two archives define two members with the same name, this 1033 // causes a collision which result in only one of the objects being taken 1034 // into consideration at LTO time (which very likely causes undefined 1035 // symbols later in the link stage). So we append file offset to make 1036 // filename unique. 1037 MemoryBufferRef MBRef( 1038 MB.getBuffer(), 1039 Saver.save(ArchiveName + Path + 1040 (ArchiveName.empty() ? "" : utostr(OffsetInArchive)))); 1041 1042 Obj = CHECK(lto::InputFile::create(MBRef), this); 1043 1044 Triple T(Obj->getTargetTriple()); 1045 EKind = getBitcodeELFKind(T); 1046 EMachine = getBitcodeMachineKind(MB.getBufferIdentifier(), T); 1047 } 1048 1049 static uint8_t mapVisibility(GlobalValue::VisibilityTypes GvVisibility) { 1050 switch (GvVisibility) { 1051 case GlobalValue::DefaultVisibility: 1052 return STV_DEFAULT; 1053 case GlobalValue::HiddenVisibility: 1054 return STV_HIDDEN; 1055 case GlobalValue::ProtectedVisibility: 1056 return STV_PROTECTED; 1057 } 1058 llvm_unreachable("unknown visibility"); 1059 } 1060 1061 template <class ELFT> 1062 static Symbol *createBitcodeSymbol(const std::vector<bool> &KeptComdats, 1063 const lto::InputFile::Symbol &ObjSym, 1064 BitcodeFile &F) { 1065 StringRef NameRef = Saver.save(ObjSym.getName()); 1066 uint32_t Binding = ObjSym.isWeak() ? STB_WEAK : STB_GLOBAL; 1067 1068 uint8_t Type = ObjSym.isTLS() ? STT_TLS : STT_NOTYPE; 1069 uint8_t Visibility = mapVisibility(ObjSym.getVisibility()); 1070 bool CanOmitFromDynSym = ObjSym.canBeOmittedFromSymbolTable(); 1071 1072 int C = ObjSym.getComdatIndex(); 1073 if (C != -1 && !KeptComdats[C]) 1074 return Symtab->addUndefined<ELFT>(NameRef, Binding, Visibility, Type, 1075 CanOmitFromDynSym, &F); 1076 1077 if (ObjSym.isUndefined()) 1078 return Symtab->addUndefined<ELFT>(NameRef, Binding, Visibility, Type, 1079 CanOmitFromDynSym, &F); 1080 1081 if (ObjSym.isCommon()) 1082 return Symtab->addCommon(NameRef, ObjSym.getCommonSize(), 1083 ObjSym.getCommonAlignment(), Binding, Visibility, 1084 STT_OBJECT, F); 1085 1086 return Symtab->addBitcode(NameRef, Binding, Visibility, Type, 1087 CanOmitFromDynSym, F); 1088 } 1089 1090 template <class ELFT> 1091 void BitcodeFile::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 1092 std::vector<bool> KeptComdats; 1093 for (StringRef S : Obj->getComdatTable()) 1094 KeptComdats.push_back(ComdatGroups.insert(CachedHashStringRef(S)).second); 1095 1096 for (const lto::InputFile::Symbol &ObjSym : Obj->symbols()) 1097 Symbols.push_back(createBitcodeSymbol<ELFT>(KeptComdats, ObjSym, *this)); 1098 } 1099 1100 static ELFKind getELFKind(MemoryBufferRef MB) { 1101 unsigned char Size; 1102 unsigned char Endian; 1103 std::tie(Size, Endian) = getElfArchType(MB.getBuffer()); 1104 1105 if (Endian != ELFDATA2LSB && Endian != ELFDATA2MSB) 1106 fatal(MB.getBufferIdentifier() + ": invalid data encoding"); 1107 if (Size != ELFCLASS32 && Size != ELFCLASS64) 1108 fatal(MB.getBufferIdentifier() + ": invalid file class"); 1109 1110 size_t BufSize = MB.getBuffer().size(); 1111 if ((Size == ELFCLASS32 && BufSize < sizeof(Elf32_Ehdr)) || 1112 (Size == ELFCLASS64 && BufSize < sizeof(Elf64_Ehdr))) 1113 fatal(MB.getBufferIdentifier() + ": file is too short"); 1114 1115 if (Size == ELFCLASS32) 1116 return (Endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind; 1117 return (Endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind; 1118 } 1119 1120 void BinaryFile::parse() { 1121 ArrayRef<uint8_t> Data = toArrayRef(MB.getBuffer()); 1122 auto *Section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS, 1123 8, Data, ".data"); 1124 Sections.push_back(Section); 1125 1126 // For each input file foo that is embedded to a result as a binary 1127 // blob, we define _binary_foo_{start,end,size} symbols, so that 1128 // user programs can access blobs by name. Non-alphanumeric 1129 // characters in a filename are replaced with underscore. 1130 std::string S = "_binary_" + MB.getBufferIdentifier().str(); 1131 for (size_t I = 0; I < S.size(); ++I) 1132 if (!isAlnum(S[I])) 1133 S[I] = '_'; 1134 1135 Symtab->addRegular(Saver.save(S + "_start"), STV_DEFAULT, STT_OBJECT, 0, 0, 1136 STB_GLOBAL, Section, nullptr); 1137 Symtab->addRegular(Saver.save(S + "_end"), STV_DEFAULT, STT_OBJECT, 1138 Data.size(), 0, STB_GLOBAL, Section, nullptr); 1139 Symtab->addRegular(Saver.save(S + "_size"), STV_DEFAULT, STT_OBJECT, 1140 Data.size(), 0, STB_GLOBAL, nullptr, nullptr); 1141 } 1142 1143 InputFile *elf::createObjectFile(MemoryBufferRef MB, StringRef ArchiveName, 1144 uint64_t OffsetInArchive) { 1145 if (isBitcode(MB)) 1146 return make<BitcodeFile>(MB, ArchiveName, OffsetInArchive); 1147 1148 switch (getELFKind(MB)) { 1149 case ELF32LEKind: 1150 return make<ObjFile<ELF32LE>>(MB, ArchiveName); 1151 case ELF32BEKind: 1152 return make<ObjFile<ELF32BE>>(MB, ArchiveName); 1153 case ELF64LEKind: 1154 return make<ObjFile<ELF64LE>>(MB, ArchiveName); 1155 case ELF64BEKind: 1156 return make<ObjFile<ELF64BE>>(MB, ArchiveName); 1157 default: 1158 llvm_unreachable("getELFKind"); 1159 } 1160 } 1161 1162 InputFile *elf::createSharedFile(MemoryBufferRef MB, StringRef DefaultSoName) { 1163 switch (getELFKind(MB)) { 1164 case ELF32LEKind: 1165 return make<SharedFile<ELF32LE>>(MB, DefaultSoName); 1166 case ELF32BEKind: 1167 return make<SharedFile<ELF32BE>>(MB, DefaultSoName); 1168 case ELF64LEKind: 1169 return make<SharedFile<ELF64LE>>(MB, DefaultSoName); 1170 case ELF64BEKind: 1171 return make<SharedFile<ELF64BE>>(MB, DefaultSoName); 1172 default: 1173 llvm_unreachable("getELFKind"); 1174 } 1175 } 1176 1177 MemoryBufferRef LazyObjFile::getBuffer() { 1178 if (AddedToLink) 1179 return MemoryBufferRef(); 1180 AddedToLink = true; 1181 return MB; 1182 } 1183 1184 InputFile *LazyObjFile::fetch() { 1185 MemoryBufferRef MBRef = getBuffer(); 1186 if (MBRef.getBuffer().empty()) 1187 return nullptr; 1188 1189 InputFile *File = createObjectFile(MBRef, ArchiveName, OffsetInArchive); 1190 File->GroupId = GroupId; 1191 return File; 1192 } 1193 1194 template <class ELFT> void LazyObjFile::parse() { 1195 // A lazy object file wraps either a bitcode file or an ELF file. 1196 if (isBitcode(this->MB)) { 1197 std::unique_ptr<lto::InputFile> Obj = 1198 CHECK(lto::InputFile::create(this->MB), this); 1199 for (const lto::InputFile::Symbol &Sym : Obj->symbols()) 1200 if (!Sym.isUndefined()) 1201 Symtab->addLazyObject<ELFT>(Saver.save(Sym.getName()), *this); 1202 return; 1203 } 1204 1205 switch (getELFKind(this->MB)) { 1206 case ELF32LEKind: 1207 addElfSymbols<ELF32LE>(); 1208 return; 1209 case ELF32BEKind: 1210 addElfSymbols<ELF32BE>(); 1211 return; 1212 case ELF64LEKind: 1213 addElfSymbols<ELF64LE>(); 1214 return; 1215 case ELF64BEKind: 1216 addElfSymbols<ELF64BE>(); 1217 return; 1218 default: 1219 llvm_unreachable("getELFKind"); 1220 } 1221 } 1222 1223 template <class ELFT> void LazyObjFile::addElfSymbols() { 1224 ELFFile<ELFT> Obj = check(ELFFile<ELFT>::create(MB.getBuffer())); 1225 ArrayRef<typename ELFT::Shdr> Sections = CHECK(Obj.sections(), this); 1226 1227 for (const typename ELFT::Shdr &Sec : Sections) { 1228 if (Sec.sh_type != SHT_SYMTAB) 1229 continue; 1230 1231 typename ELFT::SymRange Syms = CHECK(Obj.symbols(&Sec), this); 1232 uint32_t FirstGlobal = Sec.sh_info; 1233 StringRef StringTable = 1234 CHECK(Obj.getStringTableForSymtab(Sec, Sections), this); 1235 1236 for (const typename ELFT::Sym &Sym : Syms.slice(FirstGlobal)) 1237 if (Sym.st_shndx != SHN_UNDEF) 1238 Symtab->addLazyObject<ELFT>(CHECK(Sym.getName(StringTable), this), 1239 *this); 1240 return; 1241 } 1242 } 1243 1244 std::string elf::replaceThinLTOSuffix(StringRef Path) { 1245 StringRef Suffix = Config->ThinLTOObjectSuffixReplace.first; 1246 StringRef Repl = Config->ThinLTOObjectSuffixReplace.second; 1247 1248 if (!Path.endswith(Suffix)) { 1249 error("-thinlto-object-suffix-replace=" + Suffix + ";" + Repl + 1250 " was given, but " + Path + " does not end with the suffix"); 1251 return ""; 1252 } 1253 return (Path.drop_back(Suffix.size()) + Repl).str(); 1254 } 1255 1256 template void ArchiveFile::parse<ELF32LE>(); 1257 template void ArchiveFile::parse<ELF32BE>(); 1258 template void ArchiveFile::parse<ELF64LE>(); 1259 template void ArchiveFile::parse<ELF64BE>(); 1260 1261 template void BitcodeFile::parse<ELF32LE>(DenseSet<CachedHashStringRef> &); 1262 template void BitcodeFile::parse<ELF32BE>(DenseSet<CachedHashStringRef> &); 1263 template void BitcodeFile::parse<ELF64LE>(DenseSet<CachedHashStringRef> &); 1264 template void BitcodeFile::parse<ELF64BE>(DenseSet<CachedHashStringRef> &); 1265 1266 template void LazyObjFile::parse<ELF32LE>(); 1267 template void LazyObjFile::parse<ELF32BE>(); 1268 template void LazyObjFile::parse<ELF64LE>(); 1269 template void LazyObjFile::parse<ELF64BE>(); 1270 1271 template class elf::ELFFileBase<ELF32LE>; 1272 template class elf::ELFFileBase<ELF32BE>; 1273 template class elf::ELFFileBase<ELF64LE>; 1274 template class elf::ELFFileBase<ELF64BE>; 1275 1276 template class elf::ObjFile<ELF32LE>; 1277 template class elf::ObjFile<ELF32BE>; 1278 template class elf::ObjFile<ELF64LE>; 1279 template class elf::ObjFile<ELF64BE>; 1280 1281 template class elf::SharedFile<ELF32LE>; 1282 template class elf::SharedFile<ELF32BE>; 1283 template class elf::SharedFile<ELF64LE>; 1284 template class elf::SharedFile<ELF64BE>; 1285