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 DWARFDataExtractor LineData(Obj, Obj.getLineSection(), Config->IsLE, 130 Config->Wordsize); 131 132 for (std::unique_ptr<DWARFCompileUnit> &CU : Dwarf->compile_units()) { 133 auto Report = [](Error Err) { 134 handleAllErrors(std::move(Err), 135 [](ErrorInfoBase &Info) { warn(Info.message()); }); 136 }; 137 Expected<const DWARFDebugLine::LineTable *> ExpectedLT = 138 Dwarf->getLineTableForUnit(CU.get(), Report); 139 const DWARFDebugLine::LineTable *LT = nullptr; 140 if (ExpectedLT) 141 LT = *ExpectedLT; 142 else 143 Report(ExpectedLT.takeError()); 144 if (!LT) 145 continue; 146 LineTables.push_back(LT); 147 148 // Loop over variable records and insert them to VariableLoc. 149 for (const auto &Entry : CU->dies()) { 150 DWARFDie Die(CU.get(), &Entry); 151 // Skip all tags that are not variables. 152 if (Die.getTag() != dwarf::DW_TAG_variable) 153 continue; 154 155 // Skip if a local variable because we don't need them for generating 156 // error messages. In general, only non-local symbols can fail to be 157 // linked. 158 if (!dwarf::toUnsigned(Die.find(dwarf::DW_AT_external), 0)) 159 continue; 160 161 // Get the source filename index for the variable. 162 unsigned File = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_file), 0); 163 if (!LT->hasFileAtIndex(File)) 164 continue; 165 166 // Get the line number on which the variable is declared. 167 unsigned Line = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_line), 0); 168 169 // Here we want to take the variable name to add it into VariableLoc. 170 // Variable can have regular and linkage name associated. At first, we try 171 // to get linkage name as it can be different, for example when we have 172 // two variables in different namespaces of the same object. Use common 173 // name otherwise, but handle the case when it also absent in case if the 174 // input object file lacks some debug info. 175 StringRef Name = 176 dwarf::toString(Die.find(dwarf::DW_AT_linkage_name), 177 dwarf::toString(Die.find(dwarf::DW_AT_name), "")); 178 if (!Name.empty()) 179 VariableLoc.insert({Name, {LT, File, Line}}); 180 } 181 } 182 } 183 184 // Returns the pair of file name and line number describing location of data 185 // object (variable, array, etc) definition. 186 template <class ELFT> 187 Optional<std::pair<std::string, unsigned>> 188 ObjFile<ELFT>::getVariableLoc(StringRef Name) { 189 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 190 191 // Return if we have no debug information about data object. 192 auto It = VariableLoc.find(Name); 193 if (It == VariableLoc.end()) 194 return None; 195 196 // Take file name string from line table. 197 std::string FileName; 198 if (!It->second.LT->getFileNameByIndex( 199 It->second.File, nullptr, 200 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FileName)) 201 return None; 202 203 return std::make_pair(FileName, It->second.Line); 204 } 205 206 // Returns source line information for a given offset 207 // using DWARF debug info. 208 template <class ELFT> 209 Optional<DILineInfo> ObjFile<ELFT>::getDILineInfo(InputSectionBase *S, 210 uint64_t Offset) { 211 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 212 213 // Use fake address calcuated by adding section file offset and offset in 214 // section. See comments for ObjectInfo class. 215 DILineInfo Info; 216 for (const llvm::DWARFDebugLine::LineTable *LT : LineTables) 217 if (LT->getFileLineInfoForAddress( 218 S->getOffsetInFile() + Offset, nullptr, 219 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, Info)) 220 return Info; 221 return None; 222 } 223 224 // Returns source line information for a given offset using DWARF debug info. 225 template <class ELFT> 226 std::string ObjFile<ELFT>::getLineInfo(InputSectionBase *S, uint64_t Offset) { 227 if (Optional<DILineInfo> Info = getDILineInfo(S, Offset)) 228 return Info->FileName + ":" + std::to_string(Info->Line); 229 return ""; 230 } 231 232 // Returns "<internal>", "foo.a(bar.o)" or "baz.o". 233 std::string lld::toString(const InputFile *F) { 234 if (!F) 235 return "<internal>"; 236 237 if (F->ToStringCache.empty()) { 238 if (F->ArchiveName.empty()) 239 F->ToStringCache = F->getName(); 240 else 241 F->ToStringCache = (F->ArchiveName + "(" + F->getName() + ")").str(); 242 } 243 return F->ToStringCache; 244 } 245 246 template <class ELFT> 247 ELFFileBase<ELFT>::ELFFileBase(Kind K, MemoryBufferRef MB) : InputFile(K, MB) { 248 if (ELFT::TargetEndianness == support::little) 249 EKind = ELFT::Is64Bits ? ELF64LEKind : ELF32LEKind; 250 else 251 EKind = ELFT::Is64Bits ? ELF64BEKind : ELF32BEKind; 252 253 EMachine = getObj().getHeader()->e_machine; 254 OSABI = getObj().getHeader()->e_ident[llvm::ELF::EI_OSABI]; 255 } 256 257 template <class ELFT> 258 typename ELFT::SymRange ELFFileBase<ELFT>::getGlobalELFSyms() { 259 return makeArrayRef(ELFSyms.begin() + FirstGlobal, ELFSyms.end()); 260 } 261 262 template <class ELFT> 263 uint32_t ELFFileBase<ELFT>::getSectionIndex(const Elf_Sym &Sym) const { 264 return CHECK(getObj().getSectionIndex(&Sym, ELFSyms, SymtabSHNDX), this); 265 } 266 267 template <class ELFT> 268 void ELFFileBase<ELFT>::initSymtab(ArrayRef<Elf_Shdr> Sections, 269 const Elf_Shdr *Symtab) { 270 FirstGlobal = Symtab->sh_info; 271 ELFSyms = CHECK(getObj().symbols(Symtab), this); 272 if (FirstGlobal == 0 || FirstGlobal > ELFSyms.size()) 273 fatal(toString(this) + ": invalid sh_info in symbol table"); 274 275 StringTable = 276 CHECK(getObj().getStringTableForSymtab(*Symtab, Sections), this); 277 } 278 279 template <class ELFT> 280 ObjFile<ELFT>::ObjFile(MemoryBufferRef M, StringRef ArchiveName) 281 : ELFFileBase<ELFT>(Base::ObjKind, M) { 282 this->ArchiveName = ArchiveName; 283 } 284 285 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() { 286 if (this->Symbols.empty()) 287 return {}; 288 return makeArrayRef(this->Symbols).slice(1, this->FirstGlobal - 1); 289 } 290 291 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getGlobalSymbols() { 292 return makeArrayRef(this->Symbols).slice(this->FirstGlobal); 293 } 294 295 template <class ELFT> 296 void ObjFile<ELFT>::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 297 // Read a section table. JustSymbols is usually false. 298 if (this->JustSymbols) 299 initializeJustSymbols(); 300 else 301 initializeSections(ComdatGroups); 302 303 // Read a symbol table. 304 initializeSymbols(); 305 } 306 307 // Sections with SHT_GROUP and comdat bits define comdat section groups. 308 // They are identified and deduplicated by group name. This function 309 // returns a group name. 310 template <class ELFT> 311 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> Sections, 312 const Elf_Shdr &Sec) { 313 // Group signatures are stored as symbol names in object files. 314 // sh_info contains a symbol index, so we fetch a symbol and read its name. 315 if (this->ELFSyms.empty()) 316 this->initSymtab( 317 Sections, CHECK(object::getSection<ELFT>(Sections, Sec.sh_link), this)); 318 319 const Elf_Sym *Sym = 320 CHECK(object::getSymbol<ELFT>(this->ELFSyms, Sec.sh_info), this); 321 StringRef Signature = CHECK(Sym->getName(this->StringTable), this); 322 323 // As a special case, if a symbol is a section symbol and has no name, 324 // we use a section name as a signature. 325 // 326 // Such SHT_GROUP sections are invalid from the perspective of the ELF 327 // standard, but GNU gold 1.14 (the newest version as of July 2017) or 328 // older produce such sections as outputs for the -r option, so we need 329 // a bug-compatibility. 330 if (Signature.empty() && Sym->getType() == STT_SECTION) 331 return getSectionName(Sec); 332 return Signature; 333 } 334 335 template <class ELFT> 336 ArrayRef<typename ObjFile<ELFT>::Elf_Word> 337 ObjFile<ELFT>::getShtGroupEntries(const Elf_Shdr &Sec) { 338 const ELFFile<ELFT> &Obj = this->getObj(); 339 ArrayRef<Elf_Word> Entries = 340 CHECK(Obj.template getSectionContentsAsArray<Elf_Word>(&Sec), this); 341 if (Entries.empty() || Entries[0] != GRP_COMDAT) 342 fatal(toString(this) + ": unsupported SHT_GROUP format"); 343 return Entries.slice(1); 344 } 345 346 template <class ELFT> bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &Sec) { 347 // On a regular link we don't merge sections if -O0 (default is -O1). This 348 // sometimes makes the linker significantly faster, although the output will 349 // be bigger. 350 // 351 // Doing the same for -r would create a problem as it would combine sections 352 // with different sh_entsize. One option would be to just copy every SHF_MERGE 353 // section as is to the output. While this would produce a valid ELF file with 354 // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when 355 // they see two .debug_str. We could have separate logic for combining 356 // SHF_MERGE sections based both on their name and sh_entsize, but that seems 357 // to be more trouble than it is worth. Instead, we just use the regular (-O1) 358 // logic for -r. 359 if (Config->Optimize == 0 && !Config->Relocatable) 360 return false; 361 362 // A mergeable section with size 0 is useless because they don't have 363 // any data to merge. A mergeable string section with size 0 can be 364 // argued as invalid because it doesn't end with a null character. 365 // We'll avoid a mess by handling them as if they were non-mergeable. 366 if (Sec.sh_size == 0) 367 return false; 368 369 // Check for sh_entsize. The ELF spec is not clear about the zero 370 // sh_entsize. It says that "the member [sh_entsize] contains 0 if 371 // the section does not hold a table of fixed-size entries". We know 372 // that Rust 1.13 produces a string mergeable section with a zero 373 // sh_entsize. Here we just accept it rather than being picky about it. 374 uint64_t EntSize = Sec.sh_entsize; 375 if (EntSize == 0) 376 return false; 377 if (Sec.sh_size % EntSize) 378 fatal(toString(this) + 379 ": SHF_MERGE section size must be a multiple of sh_entsize"); 380 381 uint64_t Flags = Sec.sh_flags; 382 if (!(Flags & SHF_MERGE)) 383 return false; 384 if (Flags & SHF_WRITE) 385 fatal(toString(this) + ": writable SHF_MERGE section is not supported"); 386 387 return true; 388 } 389 390 // This is for --just-symbols. 391 // 392 // --just-symbols is a very minor feature that allows you to link your 393 // output against other existing program, so that if you load both your 394 // program and the other program into memory, your output can refer the 395 // other program's symbols. 396 // 397 // When the option is given, we link "just symbols". The section table is 398 // initialized with null pointers. 399 template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() { 400 ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this); 401 this->Sections.resize(ObjSections.size()); 402 403 for (const Elf_Shdr &Sec : ObjSections) { 404 if (Sec.sh_type != SHT_SYMTAB) 405 continue; 406 this->initSymtab(ObjSections, &Sec); 407 return; 408 } 409 } 410 411 template <class ELFT> 412 void ObjFile<ELFT>::initializeSections( 413 DenseSet<CachedHashStringRef> &ComdatGroups) { 414 const ELFFile<ELFT> &Obj = this->getObj(); 415 416 ArrayRef<Elf_Shdr> ObjSections = CHECK(Obj.sections(), this); 417 uint64_t Size = ObjSections.size(); 418 this->Sections.resize(Size); 419 this->SectionStringTable = 420 CHECK(Obj.getSectionStringTable(ObjSections), this); 421 422 for (size_t I = 0, E = ObjSections.size(); I < E; I++) { 423 if (this->Sections[I] == &InputSection::Discarded) 424 continue; 425 const Elf_Shdr &Sec = ObjSections[I]; 426 427 // SHF_EXCLUDE'ed sections are discarded by the linker. However, 428 // if -r is given, we'll let the final link discard such sections. 429 // This is compatible with GNU. 430 if ((Sec.sh_flags & SHF_EXCLUDE) && !Config->Relocatable) { 431 this->Sections[I] = &InputSection::Discarded; 432 continue; 433 } 434 435 switch (Sec.sh_type) { 436 case SHT_GROUP: { 437 // De-duplicate section groups by their signatures. 438 StringRef Signature = getShtGroupSignature(ObjSections, Sec); 439 bool IsNew = ComdatGroups.insert(CachedHashStringRef(Signature)).second; 440 this->Sections[I] = &InputSection::Discarded; 441 442 // If it is a new section group, we want to keep group members. 443 // Group leader sections, which contain indices of group members, are 444 // discarded because they are useless beyond this point. The only 445 // exception is the -r option because in order to produce re-linkable 446 // object files, we want to pass through basically everything. 447 if (IsNew) { 448 if (Config->Relocatable) 449 this->Sections[I] = createInputSection(Sec); 450 continue; 451 } 452 453 // Otherwise, discard group members. 454 for (uint32_t SecIndex : getShtGroupEntries(Sec)) { 455 if (SecIndex >= Size) 456 fatal(toString(this) + 457 ": invalid section index in group: " + Twine(SecIndex)); 458 this->Sections[SecIndex] = &InputSection::Discarded; 459 } 460 break; 461 } 462 case SHT_SYMTAB: 463 this->initSymtab(ObjSections, &Sec); 464 break; 465 case SHT_SYMTAB_SHNDX: 466 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, ObjSections), this); 467 break; 468 case SHT_STRTAB: 469 case SHT_NULL: 470 break; 471 default: 472 this->Sections[I] = createInputSection(Sec); 473 } 474 475 // .ARM.exidx sections have a reverse dependency on the InputSection they 476 // have a SHF_LINK_ORDER dependency, this is identified by the sh_link. 477 if (Sec.sh_flags & SHF_LINK_ORDER) { 478 if (Sec.sh_link >= this->Sections.size()) 479 fatal(toString(this) + 480 ": invalid sh_link index: " + Twine(Sec.sh_link)); 481 482 InputSectionBase *LinkSec = this->Sections[Sec.sh_link]; 483 InputSection *IS = cast<InputSection>(this->Sections[I]); 484 LinkSec->DependentSections.push_back(IS); 485 if (!isa<InputSection>(LinkSec)) 486 error("a section " + IS->Name + 487 " with SHF_LINK_ORDER should not refer a non-regular " 488 "section: " + 489 toString(LinkSec)); 490 } 491 } 492 } 493 494 // The ARM support in lld makes some use of instructions that are not available 495 // on all ARM architectures. Namely: 496 // - Use of BLX instruction for interworking between ARM and Thumb state. 497 // - Use of the extended Thumb branch encoding in relocation. 498 // - Use of the MOVT/MOVW instructions in Thumb Thunks. 499 // The ARM Attributes section contains information about the architecture chosen 500 // at compile time. We follow the convention that if at least one input object 501 // is compiled with an architecture that supports these features then lld is 502 // permitted to use them. 503 static void updateSupportedARMFeatures(const ARMAttributeParser &Attributes) { 504 if (!Attributes.hasAttribute(ARMBuildAttrs::CPU_arch)) 505 return; 506 auto Arch = Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch); 507 switch (Arch) { 508 case ARMBuildAttrs::Pre_v4: 509 case ARMBuildAttrs::v4: 510 case ARMBuildAttrs::v4T: 511 // Architectures prior to v5 do not support BLX instruction 512 break; 513 case ARMBuildAttrs::v5T: 514 case ARMBuildAttrs::v5TE: 515 case ARMBuildAttrs::v5TEJ: 516 case ARMBuildAttrs::v6: 517 case ARMBuildAttrs::v6KZ: 518 case ARMBuildAttrs::v6K: 519 Config->ARMHasBlx = true; 520 // Architectures used in pre-Cortex processors do not support 521 // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception 522 // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do. 523 break; 524 default: 525 // All other Architectures have BLX and extended branch encoding 526 Config->ARMHasBlx = true; 527 Config->ARMJ1J2BranchEncoding = true; 528 if (Arch != ARMBuildAttrs::v6_M && Arch != ARMBuildAttrs::v6S_M) 529 // All Architectures used in Cortex processors with the exception 530 // of v6-M and v6S-M have the MOVT and MOVW instructions. 531 Config->ARMHasMovtMovw = true; 532 break; 533 } 534 } 535 536 template <class ELFT> 537 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &Sec) { 538 uint32_t Idx = Sec.sh_info; 539 if (Idx >= this->Sections.size()) 540 fatal(toString(this) + ": invalid relocated section index: " + Twine(Idx)); 541 InputSectionBase *Target = this->Sections[Idx]; 542 543 // Strictly speaking, a relocation section must be included in the 544 // group of the section it relocates. However, LLVM 3.3 and earlier 545 // would fail to do so, so we gracefully handle that case. 546 if (Target == &InputSection::Discarded) 547 return nullptr; 548 549 if (!Target) 550 fatal(toString(this) + ": unsupported relocation reference"); 551 return Target; 552 } 553 554 // Create a regular InputSection class that has the same contents 555 // as a given section. 556 static InputSection *toRegularSection(MergeInputSection *Sec) { 557 return make<InputSection>(Sec->File, Sec->Flags, Sec->Type, Sec->Alignment, 558 Sec->Data, Sec->Name); 559 } 560 561 template <class ELFT> 562 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &Sec) { 563 StringRef Name = getSectionName(Sec); 564 565 switch (Sec.sh_type) { 566 case SHT_ARM_ATTRIBUTES: { 567 if (Config->EMachine != EM_ARM) 568 break; 569 ARMAttributeParser Attributes; 570 ArrayRef<uint8_t> Contents = check(this->getObj().getSectionContents(&Sec)); 571 Attributes.Parse(Contents, /*isLittle*/ Config->EKind == ELF32LEKind); 572 updateSupportedARMFeatures(Attributes); 573 // FIXME: Retain the first attribute section we see. The eglibc ARM 574 // dynamic loaders require the presence of an attribute section for dlopen 575 // to work. In a full implementation we would merge all attribute sections. 576 if (InX::ARMAttributes == nullptr) { 577 InX::ARMAttributes = make<InputSection>(*this, Sec, Name); 578 return InX::ARMAttributes; 579 } 580 return &InputSection::Discarded; 581 } 582 case SHT_RELA: 583 case SHT_REL: { 584 // Find a relocation target section and associate this section with that. 585 // Target may have been discarded if it is in a different section group 586 // and the group is discarded, even though it's a violation of the 587 // spec. We handle that situation gracefully by discarding dangling 588 // relocation sections. 589 InputSectionBase *Target = getRelocTarget(Sec); 590 if (!Target) 591 return nullptr; 592 593 // This section contains relocation information. 594 // If -r is given, we do not interpret or apply relocation 595 // but just copy relocation sections to output. 596 if (Config->Relocatable) 597 return make<InputSection>(*this, Sec, Name); 598 599 if (Target->FirstRelocation) 600 fatal(toString(this) + 601 ": multiple relocation sections to one section are not supported"); 602 603 // ELF spec allows mergeable sections with relocations, but they are 604 // rare, and it is in practice hard to merge such sections by contents, 605 // because applying relocations at end of linking changes section 606 // contents. So, we simply handle such sections as non-mergeable ones. 607 // Degrading like this is acceptable because section merging is optional. 608 if (auto *MS = dyn_cast<MergeInputSection>(Target)) { 609 Target = toRegularSection(MS); 610 this->Sections[Sec.sh_info] = Target; 611 } 612 613 if (Sec.sh_type == SHT_RELA) { 614 ArrayRef<Elf_Rela> Rels = CHECK(this->getObj().relas(&Sec), this); 615 Target->FirstRelocation = Rels.begin(); 616 Target->NumRelocations = Rels.size(); 617 Target->AreRelocsRela = true; 618 } else { 619 ArrayRef<Elf_Rel> Rels = CHECK(this->getObj().rels(&Sec), this); 620 Target->FirstRelocation = Rels.begin(); 621 Target->NumRelocations = Rels.size(); 622 Target->AreRelocsRela = false; 623 } 624 assert(isUInt<31>(Target->NumRelocations)); 625 626 // Relocation sections processed by the linker are usually removed 627 // from the output, so returning `nullptr` for the normal case. 628 // However, if -emit-relocs is given, we need to leave them in the output. 629 // (Some post link analysis tools need this information.) 630 if (Config->EmitRelocs) { 631 InputSection *RelocSec = make<InputSection>(*this, Sec, Name); 632 // We will not emit relocation section if target was discarded. 633 Target->DependentSections.push_back(RelocSec); 634 return RelocSec; 635 } 636 return nullptr; 637 } 638 } 639 640 // The GNU linker uses .note.GNU-stack section as a marker indicating 641 // that the code in the object file does not expect that the stack is 642 // executable (in terms of NX bit). If all input files have the marker, 643 // the GNU linker adds a PT_GNU_STACK segment to tells the loader to 644 // make the stack non-executable. Most object files have this section as 645 // of 2017. 646 // 647 // But making the stack non-executable is a norm today for security 648 // reasons. Failure to do so may result in a serious security issue. 649 // Therefore, we make LLD always add PT_GNU_STACK unless it is 650 // explicitly told to do otherwise (by -z execstack). Because the stack 651 // executable-ness is controlled solely by command line options, 652 // .note.GNU-stack sections are simply ignored. 653 if (Name == ".note.GNU-stack") 654 return &InputSection::Discarded; 655 656 // Split stacks is a feature to support a discontiguous stack. At least 657 // as of 2017, it seems that the feature is not being used widely. 658 // Only GNU gold supports that. We don't. For the details about that, 659 // see https://gcc.gnu.org/wiki/SplitStacks 660 if (Name == ".note.GNU-split-stack") { 661 error(toString(this) + 662 ": object file compiled with -fsplit-stack is not supported"); 663 return &InputSection::Discarded; 664 } 665 666 // The linkonce feature is a sort of proto-comdat. Some glibc i386 object 667 // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce 668 // sections. Drop those sections to avoid duplicate symbol errors. 669 // FIXME: This is glibc PR20543, we should remove this hack once that has been 670 // fixed for a while. 671 if (Name.startswith(".gnu.linkonce.")) 672 return &InputSection::Discarded; 673 674 // If we are creating a new .build-id section, strip existing .build-id 675 // sections so that the output won't have more than one .build-id. 676 // This is not usually a problem because input object files normally don't 677 // have .build-id sections, but you can create such files by 678 // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it. 679 if (Name == ".note.gnu.build-id" && Config->BuildId != BuildIdKind::None) 680 return &InputSection::Discarded; 681 682 // The linker merges EH (exception handling) frames and creates a 683 // .eh_frame_hdr section for runtime. So we handle them with a special 684 // class. For relocatable outputs, they are just passed through. 685 if (Name == ".eh_frame" && !Config->Relocatable) 686 return make<EhInputSection>(*this, Sec, Name); 687 688 if (shouldMerge(Sec)) 689 return make<MergeInputSection>(*this, Sec, Name); 690 return make<InputSection>(*this, Sec, Name); 691 } 692 693 template <class ELFT> 694 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &Sec) { 695 return CHECK(this->getObj().getSectionName(&Sec, SectionStringTable), this); 696 } 697 698 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() { 699 this->Symbols.reserve(this->ELFSyms.size()); 700 for (const Elf_Sym &Sym : this->ELFSyms) 701 this->Symbols.push_back(createSymbol(&Sym)); 702 } 703 704 template <class ELFT> Symbol *ObjFile<ELFT>::createSymbol(const Elf_Sym *Sym) { 705 int Binding = Sym->getBinding(); 706 707 uint32_t SecIdx = this->getSectionIndex(*Sym); 708 if (SecIdx >= this->Sections.size()) 709 fatal(toString(this) + ": invalid section index: " + Twine(SecIdx)); 710 711 InputSectionBase *Sec = this->Sections[SecIdx]; 712 uint8_t StOther = Sym->st_other; 713 uint8_t Type = Sym->getType(); 714 uint64_t Value = Sym->st_value; 715 uint64_t Size = Sym->st_size; 716 717 if (Binding == STB_LOCAL) { 718 if (Sym->getType() == STT_FILE) 719 SourceFile = CHECK(Sym->getName(this->StringTable), this); 720 721 if (this->StringTable.size() <= Sym->st_name) 722 fatal(toString(this) + ": invalid symbol name offset"); 723 724 StringRefZ Name = this->StringTable.data() + Sym->st_name; 725 if (Sym->st_shndx == SHN_UNDEF) 726 return make<Undefined>(this, Name, Binding, StOther, Type); 727 728 return make<Defined>(this, Name, Binding, StOther, Type, Value, Size, Sec); 729 } 730 731 StringRef Name = CHECK(Sym->getName(this->StringTable), this); 732 733 switch (Sym->st_shndx) { 734 case SHN_UNDEF: 735 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 736 /*CanOmitFromDynSym=*/false, this); 737 case SHN_COMMON: 738 if (Value == 0 || Value >= UINT32_MAX) 739 fatal(toString(this) + ": common symbol '" + Name + 740 "' has invalid alignment: " + Twine(Value)); 741 return Symtab->addCommon(Name, Size, Value, Binding, StOther, Type, *this); 742 } 743 744 switch (Binding) { 745 default: 746 fatal(toString(this) + ": unexpected binding: " + Twine(Binding)); 747 case STB_GLOBAL: 748 case STB_WEAK: 749 case STB_GNU_UNIQUE: 750 if (Sec == &InputSection::Discarded) 751 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 752 /*CanOmitFromDynSym=*/false, this); 753 return Symtab->addRegular(Name, StOther, Type, Value, Size, Binding, Sec, 754 this); 755 } 756 } 757 758 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&File) 759 : InputFile(ArchiveKind, File->getMemoryBufferRef()), 760 File(std::move(File)) {} 761 762 template <class ELFT> void ArchiveFile::parse() { 763 for (const Archive::Symbol &Sym : File->symbols()) 764 Symtab->addLazyArchive<ELFT>(Sym.getName(), *this, Sym); 765 } 766 767 // Returns a buffer pointing to a member file containing a given symbol. 768 InputFile *ArchiveFile::fetch(const Archive::Symbol &Sym) { 769 Archive::Child C = 770 CHECK(Sym.getMember(), toString(this) + 771 ": could not get the member for symbol " + 772 Sym.getName()); 773 774 if (!Seen.insert(C.getChildOffset()).second) 775 return nullptr; 776 777 MemoryBufferRef MB = 778 CHECK(C.getMemoryBufferRef(), 779 toString(this) + 780 ": could not get the buffer for the member defining symbol " + 781 Sym.getName()); 782 783 if (Tar && C.getParent()->isThin()) 784 Tar->append(relativeToRoot(CHECK(C.getFullName(), this)), MB.getBuffer()); 785 786 InputFile *File = createObjectFile( 787 MB, getName(), C.getParent()->isThin() ? 0 : C.getChildOffset()); 788 File->GroupId = GroupId; 789 return File; 790 } 791 792 template <class ELFT> 793 SharedFile<ELFT>::SharedFile(MemoryBufferRef M, StringRef DefaultSoName) 794 : ELFFileBase<ELFT>(Base::SharedKind, M), SoName(DefaultSoName), 795 IsNeeded(!Config->AsNeeded) {} 796 797 // Partially parse the shared object file so that we can call 798 // getSoName on this object. 799 template <class ELFT> void SharedFile<ELFT>::parseSoName() { 800 const Elf_Shdr *DynamicSec = nullptr; 801 const ELFFile<ELFT> Obj = this->getObj(); 802 ArrayRef<Elf_Shdr> Sections = CHECK(Obj.sections(), this); 803 804 // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d. 805 for (const Elf_Shdr &Sec : Sections) { 806 switch (Sec.sh_type) { 807 default: 808 continue; 809 case SHT_DYNSYM: 810 this->initSymtab(Sections, &Sec); 811 break; 812 case SHT_DYNAMIC: 813 DynamicSec = &Sec; 814 break; 815 case SHT_SYMTAB_SHNDX: 816 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, Sections), this); 817 break; 818 case SHT_GNU_versym: 819 this->VersymSec = &Sec; 820 break; 821 case SHT_GNU_verdef: 822 this->VerdefSec = &Sec; 823 break; 824 } 825 } 826 827 if (this->VersymSec && this->ELFSyms.empty()) 828 error("SHT_GNU_versym should be associated with symbol table"); 829 830 // Search for a DT_SONAME tag to initialize this->SoName. 831 if (!DynamicSec) 832 return; 833 ArrayRef<Elf_Dyn> Arr = 834 CHECK(Obj.template getSectionContentsAsArray<Elf_Dyn>(DynamicSec), this); 835 for (const Elf_Dyn &Dyn : Arr) { 836 if (Dyn.d_tag == DT_SONAME) { 837 uint64_t Val = Dyn.getVal(); 838 if (Val >= this->StringTable.size()) 839 fatal(toString(this) + ": invalid DT_SONAME entry"); 840 SoName = this->StringTable.data() + Val; 841 return; 842 } 843 } 844 } 845 846 // Parses ".gnu.version" section which is a parallel array for the symbol table. 847 // If a given file doesn't have ".gnu.version" section, returns VER_NDX_GLOBAL. 848 template <class ELFT> std::vector<uint32_t> SharedFile<ELFT>::parseVersyms() { 849 size_t Size = this->ELFSyms.size() - this->FirstGlobal; 850 if (!VersymSec) 851 return std::vector<uint32_t>(Size, VER_NDX_GLOBAL); 852 853 const char *Base = this->MB.getBuffer().data(); 854 const Elf_Versym *Versym = 855 reinterpret_cast<const Elf_Versym *>(Base + VersymSec->sh_offset) + 856 this->FirstGlobal; 857 858 std::vector<uint32_t> Ret(Size); 859 for (size_t I = 0; I < Size; ++I) 860 Ret[I] = Versym[I].vs_index; 861 return Ret; 862 } 863 864 // Parse the version definitions in the object file if present. Returns a vector 865 // whose nth element contains a pointer to the Elf_Verdef for version identifier 866 // n. Version identifiers that are not definitions map to nullptr. 867 template <class ELFT> 868 std::vector<const typename ELFT::Verdef *> SharedFile<ELFT>::parseVerdefs() { 869 if (!VerdefSec) 870 return {}; 871 872 // We cannot determine the largest verdef identifier without inspecting 873 // every Elf_Verdef, but both bfd and gold assign verdef identifiers 874 // sequentially starting from 1, so we predict that the largest identifier 875 // will be VerdefCount. 876 unsigned VerdefCount = VerdefSec->sh_info; 877 std::vector<const Elf_Verdef *> Verdefs(VerdefCount + 1); 878 879 // Build the Verdefs array by following the chain of Elf_Verdef objects 880 // from the start of the .gnu.version_d section. 881 const char *Base = this->MB.getBuffer().data(); 882 const char *Verdef = Base + VerdefSec->sh_offset; 883 for (unsigned I = 0; I != VerdefCount; ++I) { 884 auto *CurVerdef = reinterpret_cast<const Elf_Verdef *>(Verdef); 885 Verdef += CurVerdef->vd_next; 886 unsigned VerdefIndex = CurVerdef->vd_ndx; 887 if (Verdefs.size() <= VerdefIndex) 888 Verdefs.resize(VerdefIndex + 1); 889 Verdefs[VerdefIndex] = CurVerdef; 890 } 891 892 return Verdefs; 893 } 894 895 // We do not usually care about alignments of data in shared object 896 // files because the loader takes care of it. However, if we promote a 897 // DSO symbol to point to .bss due to copy relocation, we need to keep 898 // the original alignment requirements. We infer it in this function. 899 template <class ELFT> 900 uint32_t SharedFile<ELFT>::getAlignment(ArrayRef<Elf_Shdr> Sections, 901 const Elf_Sym &Sym) { 902 uint64_t Ret = UINT64_MAX; 903 if (Sym.st_value) 904 Ret = 1ULL << countTrailingZeros((uint64_t)Sym.st_value); 905 if (0 < Sym.st_shndx && Sym.st_shndx < Sections.size()) 906 Ret = std::min<uint64_t>(Ret, Sections[Sym.st_shndx].sh_addralign); 907 return (Ret > UINT32_MAX) ? 0 : Ret; 908 } 909 910 // Fully parse the shared object file. This must be called after parseSoName(). 911 // 912 // This function parses symbol versions. If a DSO has version information, 913 // the file has a ".gnu.version_d" section which contains symbol version 914 // definitions. Each symbol is associated to one version through a table in 915 // ".gnu.version" section. That table is a parallel array for the symbol 916 // table, and each table entry contains an index in ".gnu.version_d". 917 // 918 // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for 919 // VER_NDX_GLOBAL. There's no table entry for these special versions in 920 // ".gnu.version_d". 921 // 922 // The file format for symbol versioning is perhaps a bit more complicated 923 // than necessary, but you can easily understand the code if you wrap your 924 // head around the data structure described above. 925 template <class ELFT> void SharedFile<ELFT>::parseRest() { 926 Verdefs = parseVerdefs(); // parse .gnu.version_d 927 std::vector<uint32_t> Versyms = parseVersyms(); // parse .gnu.version 928 ArrayRef<Elf_Shdr> Sections = CHECK(this->getObj().sections(), this); 929 930 // System libraries can have a lot of symbols with versions. Using a 931 // fixed buffer for computing the versions name (foo@ver) can save a 932 // lot of allocations. 933 SmallString<0> VersionedNameBuffer; 934 935 // Add symbols to the symbol table. 936 ArrayRef<Elf_Sym> Syms = this->getGlobalELFSyms(); 937 for (size_t I = 0; I < Syms.size(); ++I) { 938 const Elf_Sym &Sym = Syms[I]; 939 940 StringRef Name = CHECK(Sym.getName(this->StringTable), this); 941 if (Sym.isUndefined()) { 942 Symbol *S = Symtab->addUndefined<ELFT>(Name, Sym.getBinding(), 943 Sym.st_other, Sym.getType(), 944 /*CanOmitFromDynSym=*/false, this); 945 S->ExportDynamic = true; 946 continue; 947 } 948 949 // ELF spec requires that all local symbols precede weak or global 950 // symbols in each symbol table, and the index of first non-local symbol 951 // is stored to sh_info. If a local symbol appears after some non-local 952 // symbol, that's a violation of the spec. 953 if (Sym.getBinding() == STB_LOCAL) { 954 warn("found local symbol '" + Name + 955 "' in global part of symbol table in file " + toString(this)); 956 continue; 957 } 958 959 // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly 960 // assigns VER_NDX_LOCAL to this section global symbol. Here is a 961 // workaround for this bug. 962 uint32_t Idx = Versyms[I] & ~VERSYM_HIDDEN; 963 if (Config->EMachine == EM_MIPS && Idx == VER_NDX_LOCAL && 964 Name == "_gp_disp") 965 continue; 966 967 uint64_t Alignment = getAlignment(Sections, Sym); 968 if (!(Versyms[I] & VERSYM_HIDDEN)) 969 Symtab->addShared(Name, *this, Sym, Alignment, Idx); 970 971 // Also add the symbol with the versioned name to handle undefined symbols 972 // with explicit versions. 973 if (Idx == VER_NDX_GLOBAL) 974 continue; 975 976 if (Idx >= Verdefs.size() || Idx == VER_NDX_LOCAL) { 977 error("corrupt input file: version definition index " + Twine(Idx) + 978 " for symbol " + Name + " is out of bounds\n>>> defined in " + 979 toString(this)); 980 continue; 981 } 982 983 StringRef VerName = 984 this->StringTable.data() + Verdefs[Idx]->getAux()->vda_name; 985 VersionedNameBuffer.clear(); 986 Name = (Name + "@" + VerName).toStringRef(VersionedNameBuffer); 987 Symtab->addShared(Saver.save(Name), *this, Sym, Alignment, Idx); 988 } 989 } 990 991 static ELFKind getBitcodeELFKind(const Triple &T) { 992 if (T.isLittleEndian()) 993 return T.isArch64Bit() ? ELF64LEKind : ELF32LEKind; 994 return T.isArch64Bit() ? ELF64BEKind : ELF32BEKind; 995 } 996 997 static uint8_t getBitcodeMachineKind(StringRef Path, const Triple &T) { 998 switch (T.getArch()) { 999 case Triple::aarch64: 1000 return EM_AARCH64; 1001 case Triple::arm: 1002 case Triple::thumb: 1003 return EM_ARM; 1004 case Triple::avr: 1005 return EM_AVR; 1006 case Triple::mips: 1007 case Triple::mipsel: 1008 case Triple::mips64: 1009 case Triple::mips64el: 1010 return EM_MIPS; 1011 case Triple::ppc: 1012 return EM_PPC; 1013 case Triple::ppc64: 1014 return EM_PPC64; 1015 case Triple::x86: 1016 return T.isOSIAMCU() ? EM_IAMCU : EM_386; 1017 case Triple::x86_64: 1018 return EM_X86_64; 1019 default: 1020 error(Path + ": could not infer e_machine from bitcode target triple " + 1021 T.str()); 1022 return EM_NONE; 1023 } 1024 } 1025 1026 BitcodeFile::BitcodeFile(MemoryBufferRef MB, StringRef ArchiveName, 1027 uint64_t OffsetInArchive) 1028 : InputFile(BitcodeKind, MB) { 1029 this->ArchiveName = ArchiveName; 1030 1031 std::string Path = MB.getBufferIdentifier().str(); 1032 if (Config->ThinLTOIndexOnly) 1033 Path = replaceThinLTOSuffix(MB.getBufferIdentifier()); 1034 1035 // ThinLTO assumes that all MemoryBufferRefs given to it have a unique 1036 // name. If two archives define two members with the same name, this 1037 // causes a collision which result in only one of the objects being taken 1038 // into consideration at LTO time (which very likely causes undefined 1039 // symbols later in the link stage). So we append file offset to make 1040 // filename unique. 1041 MemoryBufferRef MBRef( 1042 MB.getBuffer(), 1043 Saver.save(ArchiveName + Path + 1044 (ArchiveName.empty() ? "" : utostr(OffsetInArchive)))); 1045 1046 Obj = CHECK(lto::InputFile::create(MBRef), this); 1047 1048 Triple T(Obj->getTargetTriple()); 1049 EKind = getBitcodeELFKind(T); 1050 EMachine = getBitcodeMachineKind(MB.getBufferIdentifier(), T); 1051 } 1052 1053 static uint8_t mapVisibility(GlobalValue::VisibilityTypes GvVisibility) { 1054 switch (GvVisibility) { 1055 case GlobalValue::DefaultVisibility: 1056 return STV_DEFAULT; 1057 case GlobalValue::HiddenVisibility: 1058 return STV_HIDDEN; 1059 case GlobalValue::ProtectedVisibility: 1060 return STV_PROTECTED; 1061 } 1062 llvm_unreachable("unknown visibility"); 1063 } 1064 1065 template <class ELFT> 1066 static Symbol *createBitcodeSymbol(const std::vector<bool> &KeptComdats, 1067 const lto::InputFile::Symbol &ObjSym, 1068 BitcodeFile &F) { 1069 StringRef Name = Saver.save(ObjSym.getName()); 1070 uint32_t Binding = ObjSym.isWeak() ? STB_WEAK : STB_GLOBAL; 1071 1072 uint8_t Type = ObjSym.isTLS() ? STT_TLS : STT_NOTYPE; 1073 uint8_t Visibility = mapVisibility(ObjSym.getVisibility()); 1074 bool CanOmitFromDynSym = ObjSym.canBeOmittedFromSymbolTable(); 1075 1076 int C = ObjSym.getComdatIndex(); 1077 if (C != -1 && !KeptComdats[C]) 1078 return Symtab->addUndefined<ELFT>(Name, Binding, Visibility, Type, 1079 CanOmitFromDynSym, &F); 1080 1081 if (ObjSym.isUndefined()) 1082 return Symtab->addUndefined<ELFT>(Name, Binding, Visibility, Type, 1083 CanOmitFromDynSym, &F); 1084 1085 if (ObjSym.isCommon()) 1086 return Symtab->addCommon(Name, ObjSym.getCommonSize(), 1087 ObjSym.getCommonAlignment(), Binding, Visibility, 1088 STT_OBJECT, F); 1089 1090 return Symtab->addBitcode(Name, Binding, Visibility, Type, CanOmitFromDynSym, 1091 F); 1092 } 1093 1094 template <class ELFT> 1095 void BitcodeFile::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 1096 std::vector<bool> KeptComdats; 1097 for (StringRef S : Obj->getComdatTable()) 1098 KeptComdats.push_back(ComdatGroups.insert(CachedHashStringRef(S)).second); 1099 1100 for (const lto::InputFile::Symbol &ObjSym : Obj->symbols()) 1101 Symbols.push_back(createBitcodeSymbol<ELFT>(KeptComdats, ObjSym, *this)); 1102 } 1103 1104 static ELFKind getELFKind(MemoryBufferRef MB) { 1105 unsigned char Size; 1106 unsigned char Endian; 1107 std::tie(Size, Endian) = getElfArchType(MB.getBuffer()); 1108 1109 if (Endian != ELFDATA2LSB && Endian != ELFDATA2MSB) 1110 fatal(MB.getBufferIdentifier() + ": invalid data encoding"); 1111 if (Size != ELFCLASS32 && Size != ELFCLASS64) 1112 fatal(MB.getBufferIdentifier() + ": invalid file class"); 1113 1114 size_t BufSize = MB.getBuffer().size(); 1115 if ((Size == ELFCLASS32 && BufSize < sizeof(Elf32_Ehdr)) || 1116 (Size == ELFCLASS64 && BufSize < sizeof(Elf64_Ehdr))) 1117 fatal(MB.getBufferIdentifier() + ": file is too short"); 1118 1119 if (Size == ELFCLASS32) 1120 return (Endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind; 1121 return (Endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind; 1122 } 1123 1124 void BinaryFile::parse() { 1125 ArrayRef<uint8_t> Data = toArrayRef(MB.getBuffer()); 1126 auto *Section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS, 1127 8, Data, ".data"); 1128 Sections.push_back(Section); 1129 1130 // For each input file foo that is embedded to a result as a binary 1131 // blob, we define _binary_foo_{start,end,size} symbols, so that 1132 // user programs can access blobs by name. Non-alphanumeric 1133 // characters in a filename are replaced with underscore. 1134 std::string S = "_binary_" + MB.getBufferIdentifier().str(); 1135 for (size_t I = 0; I < S.size(); ++I) 1136 if (!isAlnum(S[I])) 1137 S[I] = '_'; 1138 1139 Symtab->addRegular(Saver.save(S + "_start"), STV_DEFAULT, STT_OBJECT, 0, 0, 1140 STB_GLOBAL, Section, nullptr); 1141 Symtab->addRegular(Saver.save(S + "_end"), STV_DEFAULT, STT_OBJECT, 1142 Data.size(), 0, STB_GLOBAL, Section, nullptr); 1143 Symtab->addRegular(Saver.save(S + "_size"), STV_DEFAULT, STT_OBJECT, 1144 Data.size(), 0, STB_GLOBAL, nullptr, nullptr); 1145 } 1146 1147 InputFile *elf::createObjectFile(MemoryBufferRef MB, StringRef ArchiveName, 1148 uint64_t OffsetInArchive) { 1149 if (isBitcode(MB)) 1150 return make<BitcodeFile>(MB, ArchiveName, OffsetInArchive); 1151 1152 switch (getELFKind(MB)) { 1153 case ELF32LEKind: 1154 return make<ObjFile<ELF32LE>>(MB, ArchiveName); 1155 case ELF32BEKind: 1156 return make<ObjFile<ELF32BE>>(MB, ArchiveName); 1157 case ELF64LEKind: 1158 return make<ObjFile<ELF64LE>>(MB, ArchiveName); 1159 case ELF64BEKind: 1160 return make<ObjFile<ELF64BE>>(MB, ArchiveName); 1161 default: 1162 llvm_unreachable("getELFKind"); 1163 } 1164 } 1165 1166 InputFile *elf::createSharedFile(MemoryBufferRef MB, StringRef DefaultSoName) { 1167 switch (getELFKind(MB)) { 1168 case ELF32LEKind: 1169 return make<SharedFile<ELF32LE>>(MB, DefaultSoName); 1170 case ELF32BEKind: 1171 return make<SharedFile<ELF32BE>>(MB, DefaultSoName); 1172 case ELF64LEKind: 1173 return make<SharedFile<ELF64LE>>(MB, DefaultSoName); 1174 case ELF64BEKind: 1175 return make<SharedFile<ELF64BE>>(MB, DefaultSoName); 1176 default: 1177 llvm_unreachable("getELFKind"); 1178 } 1179 } 1180 1181 MemoryBufferRef LazyObjFile::getBuffer() { 1182 if (AddedToLink) 1183 return MemoryBufferRef(); 1184 AddedToLink = true; 1185 return MB; 1186 } 1187 1188 InputFile *LazyObjFile::fetch() { 1189 MemoryBufferRef MBRef = getBuffer(); 1190 if (MBRef.getBuffer().empty()) 1191 return nullptr; 1192 1193 InputFile *File = createObjectFile(MBRef, ArchiveName, OffsetInArchive); 1194 File->GroupId = GroupId; 1195 return File; 1196 } 1197 1198 template <class ELFT> void LazyObjFile::parse() { 1199 // A lazy object file wraps either a bitcode file or an ELF file. 1200 if (isBitcode(this->MB)) { 1201 std::unique_ptr<lto::InputFile> Obj = 1202 CHECK(lto::InputFile::create(this->MB), this); 1203 for (const lto::InputFile::Symbol &Sym : Obj->symbols()) 1204 if (!Sym.isUndefined()) 1205 Symtab->addLazyObject<ELFT>(Saver.save(Sym.getName()), *this); 1206 return; 1207 } 1208 1209 switch (getELFKind(this->MB)) { 1210 case ELF32LEKind: 1211 addElfSymbols<ELF32LE>(); 1212 return; 1213 case ELF32BEKind: 1214 addElfSymbols<ELF32BE>(); 1215 return; 1216 case ELF64LEKind: 1217 addElfSymbols<ELF64LE>(); 1218 return; 1219 case ELF64BEKind: 1220 addElfSymbols<ELF64BE>(); 1221 return; 1222 default: 1223 llvm_unreachable("getELFKind"); 1224 } 1225 } 1226 1227 template <class ELFT> void LazyObjFile::addElfSymbols() { 1228 ELFFile<ELFT> Obj = check(ELFFile<ELFT>::create(MB.getBuffer())); 1229 ArrayRef<typename ELFT::Shdr> Sections = CHECK(Obj.sections(), this); 1230 1231 for (const typename ELFT::Shdr &Sec : Sections) { 1232 if (Sec.sh_type != SHT_SYMTAB) 1233 continue; 1234 1235 typename ELFT::SymRange Syms = CHECK(Obj.symbols(&Sec), this); 1236 uint32_t FirstGlobal = Sec.sh_info; 1237 StringRef StringTable = 1238 CHECK(Obj.getStringTableForSymtab(Sec, Sections), this); 1239 1240 for (const typename ELFT::Sym &Sym : Syms.slice(FirstGlobal)) 1241 if (Sym.st_shndx != SHN_UNDEF) 1242 Symtab->addLazyObject<ELFT>(CHECK(Sym.getName(StringTable), this), 1243 *this); 1244 return; 1245 } 1246 } 1247 1248 std::string elf::replaceThinLTOSuffix(StringRef Path) { 1249 StringRef Suffix = Config->ThinLTOObjectSuffixReplace.first; 1250 StringRef Repl = Config->ThinLTOObjectSuffixReplace.second; 1251 1252 if (!Path.endswith(Suffix)) { 1253 error("-thinlto-object-suffix-replace=" + Suffix + ";" + Repl + 1254 " was given, but " + Path + " does not end with the suffix"); 1255 return ""; 1256 } 1257 return (Path.drop_back(Suffix.size()) + Repl).str(); 1258 } 1259 1260 template void ArchiveFile::parse<ELF32LE>(); 1261 template void ArchiveFile::parse<ELF32BE>(); 1262 template void ArchiveFile::parse<ELF64LE>(); 1263 template void ArchiveFile::parse<ELF64BE>(); 1264 1265 template void BitcodeFile::parse<ELF32LE>(DenseSet<CachedHashStringRef> &); 1266 template void BitcodeFile::parse<ELF32BE>(DenseSet<CachedHashStringRef> &); 1267 template void BitcodeFile::parse<ELF64LE>(DenseSet<CachedHashStringRef> &); 1268 template void BitcodeFile::parse<ELF64BE>(DenseSet<CachedHashStringRef> &); 1269 1270 template void LazyObjFile::parse<ELF32LE>(); 1271 template void LazyObjFile::parse<ELF32BE>(); 1272 template void LazyObjFile::parse<ELF64LE>(); 1273 template void LazyObjFile::parse<ELF64BE>(); 1274 1275 template class elf::ELFFileBase<ELF32LE>; 1276 template class elf::ELFFileBase<ELF32BE>; 1277 template class elf::ELFFileBase<ELF64LE>; 1278 template class elf::ELFFileBase<ELF64BE>; 1279 1280 template class elf::ObjFile<ELF32LE>; 1281 template class elf::ObjFile<ELF32BE>; 1282 template class elf::ObjFile<ELF64LE>; 1283 template class elf::ObjFile<ELF64BE>; 1284 1285 template class elf::SharedFile<ELF32LE>; 1286 template class elf::SharedFile<ELF32BE>; 1287 template class elf::SharedFile<ELF64LE>; 1288 template class elf::SharedFile<ELF64BE>; 1289