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 "<internal>", "foo.a(bar.o)" or "baz.o". 225 std::string lld::toString(const InputFile *F) { 226 if (!F) 227 return "<internal>"; 228 229 if (F->ToStringCache.empty()) { 230 if (F->ArchiveName.empty()) 231 F->ToStringCache = F->getName(); 232 else 233 F->ToStringCache = (F->ArchiveName + "(" + F->getName() + ")").str(); 234 } 235 return F->ToStringCache; 236 } 237 238 template <class ELFT> 239 ELFFileBase<ELFT>::ELFFileBase(Kind K, MemoryBufferRef MB) : InputFile(K, MB) { 240 if (ELFT::TargetEndianness == support::little) 241 EKind = ELFT::Is64Bits ? ELF64LEKind : ELF32LEKind; 242 else 243 EKind = ELFT::Is64Bits ? ELF64BEKind : ELF32BEKind; 244 245 EMachine = getObj().getHeader()->e_machine; 246 OSABI = getObj().getHeader()->e_ident[llvm::ELF::EI_OSABI]; 247 } 248 249 template <class ELFT> 250 typename ELFT::SymRange ELFFileBase<ELFT>::getGlobalELFSyms() { 251 return makeArrayRef(ELFSyms.begin() + FirstGlobal, ELFSyms.end()); 252 } 253 254 template <class ELFT> 255 uint32_t ELFFileBase<ELFT>::getSectionIndex(const Elf_Sym &Sym) const { 256 return CHECK(getObj().getSectionIndex(&Sym, ELFSyms, SymtabSHNDX), this); 257 } 258 259 template <class ELFT> 260 void ELFFileBase<ELFT>::initSymtab(ArrayRef<Elf_Shdr> Sections, 261 const Elf_Shdr *Symtab) { 262 FirstGlobal = Symtab->sh_info; 263 ELFSyms = CHECK(getObj().symbols(Symtab), this); 264 if (FirstGlobal == 0 || FirstGlobal > ELFSyms.size()) 265 fatal(toString(this) + ": invalid sh_info in symbol table"); 266 267 StringTable = 268 CHECK(getObj().getStringTableForSymtab(*Symtab, Sections), this); 269 } 270 271 template <class ELFT> 272 ObjFile<ELFT>::ObjFile(MemoryBufferRef M, StringRef ArchiveName) 273 : ELFFileBase<ELFT>(Base::ObjKind, M) { 274 this->ArchiveName = ArchiveName; 275 } 276 277 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() { 278 if (this->Symbols.empty()) 279 return {}; 280 return makeArrayRef(this->Symbols).slice(1, this->FirstGlobal - 1); 281 } 282 283 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getGlobalSymbols() { 284 return makeArrayRef(this->Symbols).slice(this->FirstGlobal); 285 } 286 287 template <class ELFT> 288 void ObjFile<ELFT>::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 289 // Read a section table. JustSymbols is usually false. 290 if (this->JustSymbols) 291 initializeJustSymbols(); 292 else 293 initializeSections(ComdatGroups); 294 295 // Read a symbol table. 296 initializeSymbols(); 297 } 298 299 // Sections with SHT_GROUP and comdat bits define comdat section groups. 300 // They are identified and deduplicated by group name. This function 301 // returns a group name. 302 template <class ELFT> 303 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> Sections, 304 const Elf_Shdr &Sec) { 305 // Group signatures are stored as symbol names in object files. 306 // sh_info contains a symbol index, so we fetch a symbol and read its name. 307 if (this->ELFSyms.empty()) 308 this->initSymtab( 309 Sections, CHECK(object::getSection<ELFT>(Sections, Sec.sh_link), this)); 310 311 const Elf_Sym *Sym = 312 CHECK(object::getSymbol<ELFT>(this->ELFSyms, Sec.sh_info), this); 313 StringRef Signature = CHECK(Sym->getName(this->StringTable), this); 314 315 // As a special case, if a symbol is a section symbol and has no name, 316 // we use a section name as a signature. 317 // 318 // Such SHT_GROUP sections are invalid from the perspective of the ELF 319 // standard, but GNU gold 1.14 (the newest version as of July 2017) or 320 // older produce such sections as outputs for the -r option, so we need 321 // a bug-compatibility. 322 if (Signature.empty() && Sym->getType() == STT_SECTION) 323 return getSectionName(Sec); 324 return Signature; 325 } 326 327 template <class ELFT> 328 ArrayRef<typename ObjFile<ELFT>::Elf_Word> 329 ObjFile<ELFT>::getShtGroupEntries(const Elf_Shdr &Sec) { 330 const ELFFile<ELFT> &Obj = this->getObj(); 331 ArrayRef<Elf_Word> Entries = 332 CHECK(Obj.template getSectionContentsAsArray<Elf_Word>(&Sec), this); 333 if (Entries.empty() || Entries[0] != GRP_COMDAT) 334 fatal(toString(this) + ": unsupported SHT_GROUP format"); 335 return Entries.slice(1); 336 } 337 338 template <class ELFT> bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &Sec) { 339 // On a regular link we don't merge sections if -O0 (default is -O1). This 340 // sometimes makes the linker significantly faster, although the output will 341 // be bigger. 342 // 343 // Doing the same for -r would create a problem as it would combine sections 344 // with different sh_entsize. One option would be to just copy every SHF_MERGE 345 // section as is to the output. While this would produce a valid ELF file with 346 // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when 347 // they see two .debug_str. We could have separate logic for combining 348 // SHF_MERGE sections based both on their name and sh_entsize, but that seems 349 // to be more trouble than it is worth. Instead, we just use the regular (-O1) 350 // logic for -r. 351 if (Config->Optimize == 0 && !Config->Relocatable) 352 return false; 353 354 // A mergeable section with size 0 is useless because they don't have 355 // any data to merge. A mergeable string section with size 0 can be 356 // argued as invalid because it doesn't end with a null character. 357 // We'll avoid a mess by handling them as if they were non-mergeable. 358 if (Sec.sh_size == 0) 359 return false; 360 361 // Check for sh_entsize. The ELF spec is not clear about the zero 362 // sh_entsize. It says that "the member [sh_entsize] contains 0 if 363 // the section does not hold a table of fixed-size entries". We know 364 // that Rust 1.13 produces a string mergeable section with a zero 365 // sh_entsize. Here we just accept it rather than being picky about it. 366 uint64_t EntSize = Sec.sh_entsize; 367 if (EntSize == 0) 368 return false; 369 if (Sec.sh_size % EntSize) 370 fatal(toString(this) + 371 ": SHF_MERGE section size must be a multiple of sh_entsize"); 372 373 uint64_t Flags = Sec.sh_flags; 374 if (!(Flags & SHF_MERGE)) 375 return false; 376 if (Flags & SHF_WRITE) 377 fatal(toString(this) + ": writable SHF_MERGE section is not supported"); 378 379 return true; 380 } 381 382 // This is for --just-symbols. 383 // 384 // --just-symbols is a very minor feature that allows you to link your 385 // output against other existing program, so that if you load both your 386 // program and the other program into memory, your output can refer the 387 // other program's symbols. 388 // 389 // When the option is given, we link "just symbols". The section table is 390 // initialized with null pointers. 391 template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() { 392 ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this); 393 this->Sections.resize(ObjSections.size()); 394 395 for (const Elf_Shdr &Sec : ObjSections) { 396 if (Sec.sh_type != SHT_SYMTAB) 397 continue; 398 this->initSymtab(ObjSections, &Sec); 399 return; 400 } 401 } 402 403 template <class ELFT> 404 void ObjFile<ELFT>::initializeSections( 405 DenseSet<CachedHashStringRef> &ComdatGroups) { 406 const ELFFile<ELFT> &Obj = this->getObj(); 407 408 ArrayRef<Elf_Shdr> ObjSections = CHECK(Obj.sections(), this); 409 uint64_t Size = ObjSections.size(); 410 this->Sections.resize(Size); 411 this->SectionStringTable = 412 CHECK(Obj.getSectionStringTable(ObjSections), this); 413 414 for (size_t I = 0, E = ObjSections.size(); I < E; I++) { 415 if (this->Sections[I] == &InputSection::Discarded) 416 continue; 417 const Elf_Shdr &Sec = ObjSections[I]; 418 419 // SHF_EXCLUDE'ed sections are discarded by the linker. However, 420 // if -r is given, we'll let the final link discard such sections. 421 // This is compatible with GNU. 422 if ((Sec.sh_flags & SHF_EXCLUDE) && !Config->Relocatable) { 423 if (Sec.sh_type == SHT_LLVM_ADDRSIG) { 424 // We ignore the address-significance table if we know that the object 425 // file was created by objcopy or ld -r. This is because these tools 426 // will reorder the symbols in the symbol table, invalidating the data 427 // in the address-significance table, which refers to symbols by index. 428 if (Sec.sh_link != 0) 429 this->AddrsigSec = &Sec; 430 else if (Config->ICF == ICFLevel::Safe) 431 warn(toString(this) + ": --icf=safe is incompatible with object " 432 "files created using objcopy or ld -r"); 433 } 434 this->Sections[I] = &InputSection::Discarded; 435 continue; 436 } 437 438 switch (Sec.sh_type) { 439 case SHT_GROUP: { 440 // De-duplicate section groups by their signatures. 441 StringRef Signature = getShtGroupSignature(ObjSections, Sec); 442 bool IsNew = ComdatGroups.insert(CachedHashStringRef(Signature)).second; 443 this->Sections[I] = &InputSection::Discarded; 444 445 // If it is a new section group, we want to keep group members. 446 // Group leader sections, which contain indices of group members, are 447 // discarded because they are useless beyond this point. The only 448 // exception is the -r option because in order to produce re-linkable 449 // object files, we want to pass through basically everything. 450 if (IsNew) { 451 if (Config->Relocatable) 452 this->Sections[I] = createInputSection(Sec); 453 continue; 454 } 455 456 // Otherwise, discard group members. 457 for (uint32_t SecIndex : getShtGroupEntries(Sec)) { 458 if (SecIndex >= Size) 459 fatal(toString(this) + 460 ": invalid section index in group: " + Twine(SecIndex)); 461 this->Sections[SecIndex] = &InputSection::Discarded; 462 } 463 break; 464 } 465 case SHT_SYMTAB: 466 this->initSymtab(ObjSections, &Sec); 467 break; 468 case SHT_SYMTAB_SHNDX: 469 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, ObjSections), this); 470 break; 471 case SHT_STRTAB: 472 case SHT_NULL: 473 break; 474 default: 475 this->Sections[I] = createInputSection(Sec); 476 } 477 478 // .ARM.exidx sections have a reverse dependency on the InputSection they 479 // have a SHF_LINK_ORDER dependency, this is identified by the sh_link. 480 if (Sec.sh_flags & SHF_LINK_ORDER) { 481 if (Sec.sh_link >= this->Sections.size()) 482 fatal(toString(this) + 483 ": invalid sh_link index: " + Twine(Sec.sh_link)); 484 485 InputSectionBase *LinkSec = this->Sections[Sec.sh_link]; 486 InputSection *IS = cast<InputSection>(this->Sections[I]); 487 LinkSec->DependentSections.push_back(IS); 488 if (!isa<InputSection>(LinkSec)) 489 error("a section " + IS->Name + 490 " with SHF_LINK_ORDER should not refer a non-regular " 491 "section: " + 492 toString(LinkSec)); 493 } 494 } 495 } 496 497 // The ARM support in lld makes some use of instructions that are not available 498 // on all ARM architectures. Namely: 499 // - Use of BLX instruction for interworking between ARM and Thumb state. 500 // - Use of the extended Thumb branch encoding in relocation. 501 // - Use of the MOVT/MOVW instructions in Thumb Thunks. 502 // The ARM Attributes section contains information about the architecture chosen 503 // at compile time. We follow the convention that if at least one input object 504 // is compiled with an architecture that supports these features then lld is 505 // permitted to use them. 506 static void updateSupportedARMFeatures(const ARMAttributeParser &Attributes) { 507 if (!Attributes.hasAttribute(ARMBuildAttrs::CPU_arch)) 508 return; 509 auto Arch = Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch); 510 switch (Arch) { 511 case ARMBuildAttrs::Pre_v4: 512 case ARMBuildAttrs::v4: 513 case ARMBuildAttrs::v4T: 514 // Architectures prior to v5 do not support BLX instruction 515 break; 516 case ARMBuildAttrs::v5T: 517 case ARMBuildAttrs::v5TE: 518 case ARMBuildAttrs::v5TEJ: 519 case ARMBuildAttrs::v6: 520 case ARMBuildAttrs::v6KZ: 521 case ARMBuildAttrs::v6K: 522 Config->ARMHasBlx = true; 523 // Architectures used in pre-Cortex processors do not support 524 // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception 525 // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do. 526 break; 527 default: 528 // All other Architectures have BLX and extended branch encoding 529 Config->ARMHasBlx = true; 530 Config->ARMJ1J2BranchEncoding = true; 531 if (Arch != ARMBuildAttrs::v6_M && Arch != ARMBuildAttrs::v6S_M) 532 // All Architectures used in Cortex processors with the exception 533 // of v6-M and v6S-M have the MOVT and MOVW instructions. 534 Config->ARMHasMovtMovw = true; 535 break; 536 } 537 } 538 539 template <class ELFT> 540 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &Sec) { 541 uint32_t Idx = Sec.sh_info; 542 if (Idx >= this->Sections.size()) 543 fatal(toString(this) + ": invalid relocated section index: " + Twine(Idx)); 544 InputSectionBase *Target = this->Sections[Idx]; 545 546 // Strictly speaking, a relocation section must be included in the 547 // group of the section it relocates. However, LLVM 3.3 and earlier 548 // would fail to do so, so we gracefully handle that case. 549 if (Target == &InputSection::Discarded) 550 return nullptr; 551 552 if (!Target) 553 fatal(toString(this) + ": unsupported relocation reference"); 554 return Target; 555 } 556 557 // Create a regular InputSection class that has the same contents 558 // as a given section. 559 static InputSection *toRegularSection(MergeInputSection *Sec) { 560 return make<InputSection>(Sec->File, Sec->Flags, Sec->Type, Sec->Alignment, 561 Sec->Data, Sec->Name); 562 } 563 564 template <class ELFT> 565 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &Sec) { 566 StringRef Name = getSectionName(Sec); 567 568 switch (Sec.sh_type) { 569 case SHT_ARM_ATTRIBUTES: { 570 if (Config->EMachine != EM_ARM) 571 break; 572 ARMAttributeParser Attributes; 573 ArrayRef<uint8_t> Contents = check(this->getObj().getSectionContents(&Sec)); 574 Attributes.Parse(Contents, /*isLittle*/ Config->EKind == ELF32LEKind); 575 updateSupportedARMFeatures(Attributes); 576 // FIXME: Retain the first attribute section we see. The eglibc ARM 577 // dynamic loaders require the presence of an attribute section for dlopen 578 // to work. In a full implementation we would merge all attribute sections. 579 if (InX::ARMAttributes == nullptr) { 580 InX::ARMAttributes = make<InputSection>(*this, Sec, Name); 581 return InX::ARMAttributes; 582 } 583 return &InputSection::Discarded; 584 } 585 case SHT_RELA: 586 case SHT_REL: { 587 // Find a relocation target section and associate this section with that. 588 // Target may have been discarded if it is in a different section group 589 // and the group is discarded, even though it's a violation of the 590 // spec. We handle that situation gracefully by discarding dangling 591 // relocation sections. 592 InputSectionBase *Target = getRelocTarget(Sec); 593 if (!Target) 594 return nullptr; 595 596 // This section contains relocation information. 597 // If -r is given, we do not interpret or apply relocation 598 // but just copy relocation sections to output. 599 if (Config->Relocatable) 600 return make<InputSection>(*this, Sec, Name); 601 602 if (Target->FirstRelocation) 603 fatal(toString(this) + 604 ": multiple relocation sections to one section are not supported"); 605 606 // ELF spec allows mergeable sections with relocations, but they are 607 // rare, and it is in practice hard to merge such sections by contents, 608 // because applying relocations at end of linking changes section 609 // contents. So, we simply handle such sections as non-mergeable ones. 610 // Degrading like this is acceptable because section merging is optional. 611 if (auto *MS = dyn_cast<MergeInputSection>(Target)) { 612 Target = toRegularSection(MS); 613 this->Sections[Sec.sh_info] = Target; 614 } 615 616 if (Sec.sh_type == SHT_RELA) { 617 ArrayRef<Elf_Rela> Rels = CHECK(this->getObj().relas(&Sec), this); 618 Target->FirstRelocation = Rels.begin(); 619 Target->NumRelocations = Rels.size(); 620 Target->AreRelocsRela = true; 621 } else { 622 ArrayRef<Elf_Rel> Rels = CHECK(this->getObj().rels(&Sec), this); 623 Target->FirstRelocation = Rels.begin(); 624 Target->NumRelocations = Rels.size(); 625 Target->AreRelocsRela = false; 626 } 627 assert(isUInt<31>(Target->NumRelocations)); 628 629 // Relocation sections processed by the linker are usually removed 630 // from the output, so returning `nullptr` for the normal case. 631 // However, if -emit-relocs is given, we need to leave them in the output. 632 // (Some post link analysis tools need this information.) 633 if (Config->EmitRelocs) { 634 InputSection *RelocSec = make<InputSection>(*this, Sec, Name); 635 // We will not emit relocation section if target was discarded. 636 Target->DependentSections.push_back(RelocSec); 637 return RelocSec; 638 } 639 return nullptr; 640 } 641 } 642 643 // The GNU linker uses .note.GNU-stack section as a marker indicating 644 // that the code in the object file does not expect that the stack is 645 // executable (in terms of NX bit). If all input files have the marker, 646 // the GNU linker adds a PT_GNU_STACK segment to tells the loader to 647 // make the stack non-executable. Most object files have this section as 648 // of 2017. 649 // 650 // But making the stack non-executable is a norm today for security 651 // reasons. Failure to do so may result in a serious security issue. 652 // Therefore, we make LLD always add PT_GNU_STACK unless it is 653 // explicitly told to do otherwise (by -z execstack). Because the stack 654 // executable-ness is controlled solely by command line options, 655 // .note.GNU-stack sections are simply ignored. 656 if (Name == ".note.GNU-stack") 657 return &InputSection::Discarded; 658 659 // Split stacks is a feature to support a discontiguous stack, 660 // commonly used in the programming language Go. For the details, 661 // see https://gcc.gnu.org/wiki/SplitStacks. An object file compiled 662 // for split stack will include a .note.GNU-split-stack section. 663 if (Name == ".note.GNU-split-stack") { 664 if (Config->Relocatable) { 665 error("Cannot mix split-stack and non-split-stack in a relocatable link"); 666 return &InputSection::Discarded; 667 } 668 this->SplitStack = true; 669 return &InputSection::Discarded; 670 } 671 672 // An object file cmpiled for split stack, but where some of the 673 // functions were compiled with the no_split_stack_attribute will 674 // include a .note.GNU-no-split-stack section. 675 if (Name == ".note.GNU-no-split-stack") { 676 this->SomeNoSplitStack = true; 677 return &InputSection::Discarded; 678 } 679 680 // The linkonce feature is a sort of proto-comdat. Some glibc i386 object 681 // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce 682 // sections. Drop those sections to avoid duplicate symbol errors. 683 // FIXME: This is glibc PR20543, we should remove this hack once that has been 684 // fixed for a while. 685 if (Name.startswith(".gnu.linkonce.")) 686 return &InputSection::Discarded; 687 688 // If we are creating a new .build-id section, strip existing .build-id 689 // sections so that the output won't have more than one .build-id. 690 // This is not usually a problem because input object files normally don't 691 // have .build-id sections, but you can create such files by 692 // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it. 693 if (Name == ".note.gnu.build-id" && Config->BuildId != BuildIdKind::None) 694 return &InputSection::Discarded; 695 696 // The linker merges EH (exception handling) frames and creates a 697 // .eh_frame_hdr section for runtime. So we handle them with a special 698 // class. For relocatable outputs, they are just passed through. 699 if (Name == ".eh_frame" && !Config->Relocatable) 700 return make<EhInputSection>(*this, Sec, Name); 701 702 if (shouldMerge(Sec)) 703 return make<MergeInputSection>(*this, Sec, Name); 704 return make<InputSection>(*this, Sec, Name); 705 } 706 707 template <class ELFT> 708 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &Sec) { 709 return CHECK(this->getObj().getSectionName(&Sec, SectionStringTable), this); 710 } 711 712 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() { 713 this->Symbols.reserve(this->ELFSyms.size()); 714 for (const Elf_Sym &Sym : this->ELFSyms) 715 this->Symbols.push_back(createSymbol(&Sym)); 716 } 717 718 template <class ELFT> Symbol *ObjFile<ELFT>::createSymbol(const Elf_Sym *Sym) { 719 int Binding = Sym->getBinding(); 720 721 uint32_t SecIdx = this->getSectionIndex(*Sym); 722 if (SecIdx >= this->Sections.size()) 723 fatal(toString(this) + ": invalid section index: " + Twine(SecIdx)); 724 725 InputSectionBase *Sec = this->Sections[SecIdx]; 726 uint8_t StOther = Sym->st_other; 727 uint8_t Type = Sym->getType(); 728 uint64_t Value = Sym->st_value; 729 uint64_t Size = Sym->st_size; 730 731 if (Binding == STB_LOCAL) { 732 if (Sym->getType() == STT_FILE) 733 SourceFile = CHECK(Sym->getName(this->StringTable), this); 734 735 if (this->StringTable.size() <= Sym->st_name) 736 fatal(toString(this) + ": invalid symbol name offset"); 737 738 StringRefZ Name = this->StringTable.data() + Sym->st_name; 739 if (Sym->st_shndx == SHN_UNDEF) 740 return make<Undefined>(this, Name, Binding, StOther, Type); 741 742 return make<Defined>(this, Name, Binding, StOther, Type, Value, Size, Sec); 743 } 744 745 StringRef Name = CHECK(Sym->getName(this->StringTable), this); 746 747 switch (Sym->st_shndx) { 748 case SHN_UNDEF: 749 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 750 /*CanOmitFromDynSym=*/false, this); 751 case SHN_COMMON: 752 if (Value == 0 || Value >= UINT32_MAX) 753 fatal(toString(this) + ": common symbol '" + Name + 754 "' has invalid alignment: " + Twine(Value)); 755 return Symtab->addCommon(Name, Size, Value, Binding, StOther, Type, *this); 756 } 757 758 switch (Binding) { 759 default: 760 fatal(toString(this) + ": unexpected binding: " + Twine(Binding)); 761 case STB_GLOBAL: 762 case STB_WEAK: 763 case STB_GNU_UNIQUE: 764 if (Sec == &InputSection::Discarded) 765 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 766 /*CanOmitFromDynSym=*/false, this); 767 return Symtab->addRegular(Name, StOther, Type, Value, Size, Binding, Sec, 768 this); 769 } 770 } 771 772 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&File) 773 : InputFile(ArchiveKind, File->getMemoryBufferRef()), 774 File(std::move(File)) {} 775 776 template <class ELFT> void ArchiveFile::parse() { 777 for (const Archive::Symbol &Sym : File->symbols()) 778 Symtab->addLazyArchive<ELFT>(Sym.getName(), *this, Sym); 779 } 780 781 // Returns a buffer pointing to a member file containing a given symbol. 782 InputFile *ArchiveFile::fetch(const Archive::Symbol &Sym) { 783 Archive::Child C = 784 CHECK(Sym.getMember(), toString(this) + 785 ": could not get the member for symbol " + 786 Sym.getName()); 787 788 if (!Seen.insert(C.getChildOffset()).second) 789 return nullptr; 790 791 MemoryBufferRef MB = 792 CHECK(C.getMemoryBufferRef(), 793 toString(this) + 794 ": could not get the buffer for the member defining symbol " + 795 Sym.getName()); 796 797 if (Tar && C.getParent()->isThin()) 798 Tar->append(relativeToRoot(CHECK(C.getFullName(), this)), MB.getBuffer()); 799 800 InputFile *File = createObjectFile( 801 MB, getName(), C.getParent()->isThin() ? 0 : C.getChildOffset()); 802 File->GroupId = GroupId; 803 return File; 804 } 805 806 template <class ELFT> 807 SharedFile<ELFT>::SharedFile(MemoryBufferRef M, StringRef DefaultSoName) 808 : ELFFileBase<ELFT>(Base::SharedKind, M), SoName(DefaultSoName), 809 IsNeeded(!Config->AsNeeded) {} 810 811 // Partially parse the shared object file so that we can call 812 // getSoName on this object. 813 template <class ELFT> void SharedFile<ELFT>::parseSoName() { 814 const Elf_Shdr *DynamicSec = nullptr; 815 const ELFFile<ELFT> Obj = this->getObj(); 816 ArrayRef<Elf_Shdr> Sections = CHECK(Obj.sections(), this); 817 818 // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d. 819 for (const Elf_Shdr &Sec : Sections) { 820 switch (Sec.sh_type) { 821 default: 822 continue; 823 case SHT_DYNSYM: 824 this->initSymtab(Sections, &Sec); 825 break; 826 case SHT_DYNAMIC: 827 DynamicSec = &Sec; 828 break; 829 case SHT_SYMTAB_SHNDX: 830 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, Sections), this); 831 break; 832 case SHT_GNU_versym: 833 this->VersymSec = &Sec; 834 break; 835 case SHT_GNU_verdef: 836 this->VerdefSec = &Sec; 837 break; 838 } 839 } 840 841 if (this->VersymSec && this->ELFSyms.empty()) 842 error("SHT_GNU_versym should be associated with symbol table"); 843 844 // Search for a DT_SONAME tag to initialize this->SoName. 845 if (!DynamicSec) 846 return; 847 ArrayRef<Elf_Dyn> Arr = 848 CHECK(Obj.template getSectionContentsAsArray<Elf_Dyn>(DynamicSec), this); 849 for (const Elf_Dyn &Dyn : Arr) { 850 if (Dyn.d_tag == DT_SONAME) { 851 uint64_t Val = Dyn.getVal(); 852 if (Val >= this->StringTable.size()) 853 fatal(toString(this) + ": invalid DT_SONAME entry"); 854 SoName = this->StringTable.data() + Val; 855 return; 856 } 857 } 858 } 859 860 // Parses ".gnu.version" section which is a parallel array for the symbol table. 861 // If a given file doesn't have ".gnu.version" section, returns VER_NDX_GLOBAL. 862 template <class ELFT> std::vector<uint32_t> SharedFile<ELFT>::parseVersyms() { 863 size_t Size = this->ELFSyms.size() - this->FirstGlobal; 864 if (!VersymSec) 865 return std::vector<uint32_t>(Size, VER_NDX_GLOBAL); 866 867 const char *Base = this->MB.getBuffer().data(); 868 const Elf_Versym *Versym = 869 reinterpret_cast<const Elf_Versym *>(Base + VersymSec->sh_offset) + 870 this->FirstGlobal; 871 872 std::vector<uint32_t> Ret(Size); 873 for (size_t I = 0; I < Size; ++I) 874 Ret[I] = Versym[I].vs_index; 875 return Ret; 876 } 877 878 // Parse the version definitions in the object file if present. Returns a vector 879 // whose nth element contains a pointer to the Elf_Verdef for version identifier 880 // n. Version identifiers that are not definitions map to nullptr. 881 template <class ELFT> 882 std::vector<const typename ELFT::Verdef *> SharedFile<ELFT>::parseVerdefs() { 883 if (!VerdefSec) 884 return {}; 885 886 // We cannot determine the largest verdef identifier without inspecting 887 // every Elf_Verdef, but both bfd and gold assign verdef identifiers 888 // sequentially starting from 1, so we predict that the largest identifier 889 // will be VerdefCount. 890 unsigned VerdefCount = VerdefSec->sh_info; 891 std::vector<const Elf_Verdef *> Verdefs(VerdefCount + 1); 892 893 // Build the Verdefs array by following the chain of Elf_Verdef objects 894 // from the start of the .gnu.version_d section. 895 const char *Base = this->MB.getBuffer().data(); 896 const char *Verdef = Base + VerdefSec->sh_offset; 897 for (unsigned I = 0; I != VerdefCount; ++I) { 898 auto *CurVerdef = reinterpret_cast<const Elf_Verdef *>(Verdef); 899 Verdef += CurVerdef->vd_next; 900 unsigned VerdefIndex = CurVerdef->vd_ndx; 901 if (Verdefs.size() <= VerdefIndex) 902 Verdefs.resize(VerdefIndex + 1); 903 Verdefs[VerdefIndex] = CurVerdef; 904 } 905 906 return Verdefs; 907 } 908 909 // We do not usually care about alignments of data in shared object 910 // files because the loader takes care of it. However, if we promote a 911 // DSO symbol to point to .bss due to copy relocation, we need to keep 912 // the original alignment requirements. We infer it in this function. 913 template <class ELFT> 914 uint32_t SharedFile<ELFT>::getAlignment(ArrayRef<Elf_Shdr> Sections, 915 const Elf_Sym &Sym) { 916 uint64_t Ret = UINT64_MAX; 917 if (Sym.st_value) 918 Ret = 1ULL << countTrailingZeros((uint64_t)Sym.st_value); 919 if (0 < Sym.st_shndx && Sym.st_shndx < Sections.size()) 920 Ret = std::min<uint64_t>(Ret, Sections[Sym.st_shndx].sh_addralign); 921 return (Ret > UINT32_MAX) ? 0 : Ret; 922 } 923 924 // Fully parse the shared object file. This must be called after parseSoName(). 925 // 926 // This function parses symbol versions. If a DSO has version information, 927 // the file has a ".gnu.version_d" section which contains symbol version 928 // definitions. Each symbol is associated to one version through a table in 929 // ".gnu.version" section. That table is a parallel array for the symbol 930 // table, and each table entry contains an index in ".gnu.version_d". 931 // 932 // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for 933 // VER_NDX_GLOBAL. There's no table entry for these special versions in 934 // ".gnu.version_d". 935 // 936 // The file format for symbol versioning is perhaps a bit more complicated 937 // than necessary, but you can easily understand the code if you wrap your 938 // head around the data structure described above. 939 template <class ELFT> void SharedFile<ELFT>::parseRest() { 940 Verdefs = parseVerdefs(); // parse .gnu.version_d 941 std::vector<uint32_t> Versyms = parseVersyms(); // parse .gnu.version 942 ArrayRef<Elf_Shdr> Sections = CHECK(this->getObj().sections(), this); 943 944 // System libraries can have a lot of symbols with versions. Using a 945 // fixed buffer for computing the versions name (foo@ver) can save a 946 // lot of allocations. 947 SmallString<0> VersionedNameBuffer; 948 949 // Add symbols to the symbol table. 950 ArrayRef<Elf_Sym> Syms = this->getGlobalELFSyms(); 951 for (size_t I = 0; I < Syms.size(); ++I) { 952 const Elf_Sym &Sym = Syms[I]; 953 954 StringRef Name = CHECK(Sym.getName(this->StringTable), this); 955 if (Sym.isUndefined()) { 956 Symbol *S = Symtab->addUndefined<ELFT>(Name, Sym.getBinding(), 957 Sym.st_other, Sym.getType(), 958 /*CanOmitFromDynSym=*/false, this); 959 S->ExportDynamic = true; 960 continue; 961 } 962 963 // ELF spec requires that all local symbols precede weak or global 964 // symbols in each symbol table, and the index of first non-local symbol 965 // is stored to sh_info. If a local symbol appears after some non-local 966 // symbol, that's a violation of the spec. 967 if (Sym.getBinding() == STB_LOCAL) { 968 warn("found local symbol '" + Name + 969 "' in global part of symbol table in file " + toString(this)); 970 continue; 971 } 972 973 // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly 974 // assigns VER_NDX_LOCAL to this section global symbol. Here is a 975 // workaround for this bug. 976 uint32_t Idx = Versyms[I] & ~VERSYM_HIDDEN; 977 if (Config->EMachine == EM_MIPS && Idx == VER_NDX_LOCAL && 978 Name == "_gp_disp") 979 continue; 980 981 uint64_t Alignment = getAlignment(Sections, Sym); 982 if (!(Versyms[I] & VERSYM_HIDDEN)) 983 Symtab->addShared(Name, *this, Sym, Alignment, Idx); 984 985 // Also add the symbol with the versioned name to handle undefined symbols 986 // with explicit versions. 987 if (Idx == VER_NDX_GLOBAL) 988 continue; 989 990 if (Idx >= Verdefs.size() || Idx == VER_NDX_LOCAL) { 991 error("corrupt input file: version definition index " + Twine(Idx) + 992 " for symbol " + Name + " is out of bounds\n>>> defined in " + 993 toString(this)); 994 continue; 995 } 996 997 StringRef VerName = 998 this->StringTable.data() + Verdefs[Idx]->getAux()->vda_name; 999 VersionedNameBuffer.clear(); 1000 Name = (Name + "@" + VerName).toStringRef(VersionedNameBuffer); 1001 Symtab->addShared(Saver.save(Name), *this, Sym, Alignment, Idx); 1002 } 1003 } 1004 1005 static ELFKind getBitcodeELFKind(const Triple &T) { 1006 if (T.isLittleEndian()) 1007 return T.isArch64Bit() ? ELF64LEKind : ELF32LEKind; 1008 return T.isArch64Bit() ? ELF64BEKind : ELF32BEKind; 1009 } 1010 1011 static uint8_t getBitcodeMachineKind(StringRef Path, const Triple &T) { 1012 switch (T.getArch()) { 1013 case Triple::aarch64: 1014 return EM_AARCH64; 1015 case Triple::arm: 1016 case Triple::thumb: 1017 return EM_ARM; 1018 case Triple::avr: 1019 return EM_AVR; 1020 case Triple::mips: 1021 case Triple::mipsel: 1022 case Triple::mips64: 1023 case Triple::mips64el: 1024 return EM_MIPS; 1025 case Triple::ppc: 1026 return EM_PPC; 1027 case Triple::ppc64: 1028 return EM_PPC64; 1029 case Triple::x86: 1030 return T.isOSIAMCU() ? EM_IAMCU : EM_386; 1031 case Triple::x86_64: 1032 return EM_X86_64; 1033 default: 1034 error(Path + ": could not infer e_machine from bitcode target triple " + 1035 T.str()); 1036 return EM_NONE; 1037 } 1038 } 1039 1040 BitcodeFile::BitcodeFile(MemoryBufferRef MB, StringRef ArchiveName, 1041 uint64_t OffsetInArchive) 1042 : InputFile(BitcodeKind, MB) { 1043 this->ArchiveName = ArchiveName; 1044 1045 std::string Path = MB.getBufferIdentifier().str(); 1046 if (Config->ThinLTOIndexOnly) 1047 Path = replaceThinLTOSuffix(MB.getBufferIdentifier()); 1048 1049 // ThinLTO assumes that all MemoryBufferRefs given to it have a unique 1050 // name. If two archives define two members with the same name, this 1051 // causes a collision which result in only one of the objects being taken 1052 // into consideration at LTO time (which very likely causes undefined 1053 // symbols later in the link stage). So we append file offset to make 1054 // filename unique. 1055 MemoryBufferRef MBRef( 1056 MB.getBuffer(), 1057 Saver.save(ArchiveName + Path + 1058 (ArchiveName.empty() ? "" : utostr(OffsetInArchive)))); 1059 1060 Obj = CHECK(lto::InputFile::create(MBRef), this); 1061 1062 Triple T(Obj->getTargetTriple()); 1063 EKind = getBitcodeELFKind(T); 1064 EMachine = getBitcodeMachineKind(MB.getBufferIdentifier(), T); 1065 } 1066 1067 static uint8_t mapVisibility(GlobalValue::VisibilityTypes GvVisibility) { 1068 switch (GvVisibility) { 1069 case GlobalValue::DefaultVisibility: 1070 return STV_DEFAULT; 1071 case GlobalValue::HiddenVisibility: 1072 return STV_HIDDEN; 1073 case GlobalValue::ProtectedVisibility: 1074 return STV_PROTECTED; 1075 } 1076 llvm_unreachable("unknown visibility"); 1077 } 1078 1079 template <class ELFT> 1080 static Symbol *createBitcodeSymbol(const std::vector<bool> &KeptComdats, 1081 const lto::InputFile::Symbol &ObjSym, 1082 BitcodeFile &F) { 1083 StringRef Name = Saver.save(ObjSym.getName()); 1084 uint32_t Binding = ObjSym.isWeak() ? STB_WEAK : STB_GLOBAL; 1085 1086 uint8_t Type = ObjSym.isTLS() ? STT_TLS : STT_NOTYPE; 1087 uint8_t Visibility = mapVisibility(ObjSym.getVisibility()); 1088 bool CanOmitFromDynSym = ObjSym.canBeOmittedFromSymbolTable(); 1089 1090 int C = ObjSym.getComdatIndex(); 1091 if (C != -1 && !KeptComdats[C]) 1092 return Symtab->addUndefined<ELFT>(Name, Binding, Visibility, Type, 1093 CanOmitFromDynSym, &F); 1094 1095 if (ObjSym.isUndefined()) 1096 return Symtab->addUndefined<ELFT>(Name, Binding, Visibility, Type, 1097 CanOmitFromDynSym, &F); 1098 1099 if (ObjSym.isCommon()) 1100 return Symtab->addCommon(Name, ObjSym.getCommonSize(), 1101 ObjSym.getCommonAlignment(), Binding, Visibility, 1102 STT_OBJECT, F); 1103 1104 return Symtab->addBitcode(Name, Binding, Visibility, Type, CanOmitFromDynSym, 1105 F); 1106 } 1107 1108 template <class ELFT> 1109 void BitcodeFile::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 1110 std::vector<bool> KeptComdats; 1111 for (StringRef S : Obj->getComdatTable()) 1112 KeptComdats.push_back(ComdatGroups.insert(CachedHashStringRef(S)).second); 1113 1114 for (const lto::InputFile::Symbol &ObjSym : Obj->symbols()) 1115 Symbols.push_back(createBitcodeSymbol<ELFT>(KeptComdats, ObjSym, *this)); 1116 } 1117 1118 static ELFKind getELFKind(MemoryBufferRef MB) { 1119 unsigned char Size; 1120 unsigned char Endian; 1121 std::tie(Size, Endian) = getElfArchType(MB.getBuffer()); 1122 1123 if (Endian != ELFDATA2LSB && Endian != ELFDATA2MSB) 1124 fatal(MB.getBufferIdentifier() + ": invalid data encoding"); 1125 if (Size != ELFCLASS32 && Size != ELFCLASS64) 1126 fatal(MB.getBufferIdentifier() + ": invalid file class"); 1127 1128 size_t BufSize = MB.getBuffer().size(); 1129 if ((Size == ELFCLASS32 && BufSize < sizeof(Elf32_Ehdr)) || 1130 (Size == ELFCLASS64 && BufSize < sizeof(Elf64_Ehdr))) 1131 fatal(MB.getBufferIdentifier() + ": file is too short"); 1132 1133 if (Size == ELFCLASS32) 1134 return (Endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind; 1135 return (Endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind; 1136 } 1137 1138 void BinaryFile::parse() { 1139 ArrayRef<uint8_t> Data = toArrayRef(MB.getBuffer()); 1140 auto *Section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS, 1141 8, Data, ".data"); 1142 Sections.push_back(Section); 1143 1144 // For each input file foo that is embedded to a result as a binary 1145 // blob, we define _binary_foo_{start,end,size} symbols, so that 1146 // user programs can access blobs by name. Non-alphanumeric 1147 // characters in a filename are replaced with underscore. 1148 std::string S = "_binary_" + MB.getBufferIdentifier().str(); 1149 for (size_t I = 0; I < S.size(); ++I) 1150 if (!isAlnum(S[I])) 1151 S[I] = '_'; 1152 1153 Symtab->addRegular(Saver.save(S + "_start"), STV_DEFAULT, STT_OBJECT, 0, 0, 1154 STB_GLOBAL, Section, nullptr); 1155 Symtab->addRegular(Saver.save(S + "_end"), STV_DEFAULT, STT_OBJECT, 1156 Data.size(), 0, STB_GLOBAL, Section, nullptr); 1157 Symtab->addRegular(Saver.save(S + "_size"), STV_DEFAULT, STT_OBJECT, 1158 Data.size(), 0, STB_GLOBAL, nullptr, nullptr); 1159 } 1160 1161 InputFile *elf::createObjectFile(MemoryBufferRef MB, StringRef ArchiveName, 1162 uint64_t OffsetInArchive) { 1163 if (isBitcode(MB)) 1164 return make<BitcodeFile>(MB, ArchiveName, OffsetInArchive); 1165 1166 switch (getELFKind(MB)) { 1167 case ELF32LEKind: 1168 return make<ObjFile<ELF32LE>>(MB, ArchiveName); 1169 case ELF32BEKind: 1170 return make<ObjFile<ELF32BE>>(MB, ArchiveName); 1171 case ELF64LEKind: 1172 return make<ObjFile<ELF64LE>>(MB, ArchiveName); 1173 case ELF64BEKind: 1174 return make<ObjFile<ELF64BE>>(MB, ArchiveName); 1175 default: 1176 llvm_unreachable("getELFKind"); 1177 } 1178 } 1179 1180 InputFile *elf::createSharedFile(MemoryBufferRef MB, StringRef DefaultSoName) { 1181 switch (getELFKind(MB)) { 1182 case ELF32LEKind: 1183 return make<SharedFile<ELF32LE>>(MB, DefaultSoName); 1184 case ELF32BEKind: 1185 return make<SharedFile<ELF32BE>>(MB, DefaultSoName); 1186 case ELF64LEKind: 1187 return make<SharedFile<ELF64LE>>(MB, DefaultSoName); 1188 case ELF64BEKind: 1189 return make<SharedFile<ELF64BE>>(MB, DefaultSoName); 1190 default: 1191 llvm_unreachable("getELFKind"); 1192 } 1193 } 1194 1195 MemoryBufferRef LazyObjFile::getBuffer() { 1196 if (AddedToLink) 1197 return MemoryBufferRef(); 1198 AddedToLink = true; 1199 return MB; 1200 } 1201 1202 InputFile *LazyObjFile::fetch() { 1203 MemoryBufferRef MBRef = getBuffer(); 1204 if (MBRef.getBuffer().empty()) 1205 return nullptr; 1206 1207 InputFile *File = createObjectFile(MBRef, ArchiveName, OffsetInArchive); 1208 File->GroupId = GroupId; 1209 return File; 1210 } 1211 1212 template <class ELFT> void LazyObjFile::parse() { 1213 // A lazy object file wraps either a bitcode file or an ELF file. 1214 if (isBitcode(this->MB)) { 1215 std::unique_ptr<lto::InputFile> Obj = 1216 CHECK(lto::InputFile::create(this->MB), this); 1217 for (const lto::InputFile::Symbol &Sym : Obj->symbols()) 1218 if (!Sym.isUndefined()) 1219 Symtab->addLazyObject<ELFT>(Saver.save(Sym.getName()), *this); 1220 return; 1221 } 1222 1223 switch (getELFKind(this->MB)) { 1224 case ELF32LEKind: 1225 addElfSymbols<ELF32LE>(); 1226 return; 1227 case ELF32BEKind: 1228 addElfSymbols<ELF32BE>(); 1229 return; 1230 case ELF64LEKind: 1231 addElfSymbols<ELF64LE>(); 1232 return; 1233 case ELF64BEKind: 1234 addElfSymbols<ELF64BE>(); 1235 return; 1236 default: 1237 llvm_unreachable("getELFKind"); 1238 } 1239 } 1240 1241 template <class ELFT> void LazyObjFile::addElfSymbols() { 1242 ELFFile<ELFT> Obj = check(ELFFile<ELFT>::create(MB.getBuffer())); 1243 ArrayRef<typename ELFT::Shdr> Sections = CHECK(Obj.sections(), this); 1244 1245 for (const typename ELFT::Shdr &Sec : Sections) { 1246 if (Sec.sh_type != SHT_SYMTAB) 1247 continue; 1248 1249 typename ELFT::SymRange Syms = CHECK(Obj.symbols(&Sec), this); 1250 uint32_t FirstGlobal = Sec.sh_info; 1251 StringRef StringTable = 1252 CHECK(Obj.getStringTableForSymtab(Sec, Sections), this); 1253 1254 for (const typename ELFT::Sym &Sym : Syms.slice(FirstGlobal)) 1255 if (Sym.st_shndx != SHN_UNDEF) 1256 Symtab->addLazyObject<ELFT>(CHECK(Sym.getName(StringTable), this), 1257 *this); 1258 return; 1259 } 1260 } 1261 1262 std::string elf::replaceThinLTOSuffix(StringRef Path) { 1263 StringRef Suffix = Config->ThinLTOObjectSuffixReplace.first; 1264 StringRef Repl = Config->ThinLTOObjectSuffixReplace.second; 1265 1266 if (!Path.endswith(Suffix)) { 1267 error("-thinlto-object-suffix-replace=" + Suffix + ";" + Repl + 1268 " was given, but " + Path + " does not end with the suffix"); 1269 return ""; 1270 } 1271 return (Path.drop_back(Suffix.size()) + Repl).str(); 1272 } 1273 1274 template void ArchiveFile::parse<ELF32LE>(); 1275 template void ArchiveFile::parse<ELF32BE>(); 1276 template void ArchiveFile::parse<ELF64LE>(); 1277 template void ArchiveFile::parse<ELF64BE>(); 1278 1279 template void BitcodeFile::parse<ELF32LE>(DenseSet<CachedHashStringRef> &); 1280 template void BitcodeFile::parse<ELF32BE>(DenseSet<CachedHashStringRef> &); 1281 template void BitcodeFile::parse<ELF64LE>(DenseSet<CachedHashStringRef> &); 1282 template void BitcodeFile::parse<ELF64BE>(DenseSet<CachedHashStringRef> &); 1283 1284 template void LazyObjFile::parse<ELF32LE>(); 1285 template void LazyObjFile::parse<ELF32BE>(); 1286 template void LazyObjFile::parse<ELF64LE>(); 1287 template void LazyObjFile::parse<ELF64BE>(); 1288 1289 template class elf::ELFFileBase<ELF32LE>; 1290 template class elf::ELFFileBase<ELF32BE>; 1291 template class elf::ELFFileBase<ELF64LE>; 1292 template class elf::ELFFileBase<ELF64BE>; 1293 1294 template class elf::ObjFile<ELF32LE>; 1295 template class elf::ObjFile<ELF32BE>; 1296 template class elf::ObjFile<ELF64LE>; 1297 template class elf::ObjFile<ELF64BE>; 1298 1299 template class elf::SharedFile<ELF32LE>; 1300 template class elf::SharedFile<ELF32BE>; 1301 template class elf::SharedFile<ELF64LE>; 1302 template class elf::SharedFile<ELF64BE>; 1303