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 for (std::unique_ptr<DWARFUnit> &CU : Dwarf->compile_units()) { 129 auto Report = [](Error Err) { 130 handleAllErrors(std::move(Err), 131 [](ErrorInfoBase &Info) { warn(Info.message()); }); 132 }; 133 Expected<const DWARFDebugLine::LineTable *> ExpectedLT = 134 Dwarf->getLineTableForUnit(CU.get(), Report); 135 const DWARFDebugLine::LineTable *LT = nullptr; 136 if (ExpectedLT) 137 LT = *ExpectedLT; 138 else 139 Report(ExpectedLT.takeError()); 140 if (!LT) 141 continue; 142 LineTables.push_back(LT); 143 144 // Loop over variable records and insert them to VariableLoc. 145 for (const auto &Entry : CU->dies()) { 146 DWARFDie Die(CU.get(), &Entry); 147 // Skip all tags that are not variables. 148 if (Die.getTag() != dwarf::DW_TAG_variable) 149 continue; 150 151 // Skip if a local variable because we don't need them for generating 152 // error messages. In general, only non-local symbols can fail to be 153 // linked. 154 if (!dwarf::toUnsigned(Die.find(dwarf::DW_AT_external), 0)) 155 continue; 156 157 // Get the source filename index for the variable. 158 unsigned File = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_file), 0); 159 if (!LT->hasFileAtIndex(File)) 160 continue; 161 162 // Get the line number on which the variable is declared. 163 unsigned Line = dwarf::toUnsigned(Die.find(dwarf::DW_AT_decl_line), 0); 164 165 // Here we want to take the variable name to add it into VariableLoc. 166 // Variable can have regular and linkage name associated. At first, we try 167 // to get linkage name as it can be different, for example when we have 168 // two variables in different namespaces of the same object. Use common 169 // name otherwise, but handle the case when it also absent in case if the 170 // input object file lacks some debug info. 171 StringRef Name = 172 dwarf::toString(Die.find(dwarf::DW_AT_linkage_name), 173 dwarf::toString(Die.find(dwarf::DW_AT_name), "")); 174 if (!Name.empty()) 175 VariableLoc.insert({Name, {LT, File, Line}}); 176 } 177 } 178 } 179 180 // Returns the pair of file name and line number describing location of data 181 // object (variable, array, etc) definition. 182 template <class ELFT> 183 Optional<std::pair<std::string, unsigned>> 184 ObjFile<ELFT>::getVariableLoc(StringRef Name) { 185 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 186 187 // Return if we have no debug information about data object. 188 auto It = VariableLoc.find(Name); 189 if (It == VariableLoc.end()) 190 return None; 191 192 // Take file name string from line table. 193 std::string FileName; 194 if (!It->second.LT->getFileNameByIndex( 195 It->second.File, nullptr, 196 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FileName)) 197 return None; 198 199 return std::make_pair(FileName, It->second.Line); 200 } 201 202 // Returns source line information for a given offset 203 // using DWARF debug info. 204 template <class ELFT> 205 Optional<DILineInfo> ObjFile<ELFT>::getDILineInfo(InputSectionBase *S, 206 uint64_t Offset) { 207 llvm::call_once(InitDwarfLine, [this]() { initializeDwarf(); }); 208 209 // Use fake address calcuated by adding section file offset and offset in 210 // section. See comments for ObjectInfo class. 211 DILineInfo Info; 212 for (const llvm::DWARFDebugLine::LineTable *LT : LineTables) 213 if (LT->getFileLineInfoForAddress( 214 S->getOffsetInFile() + Offset, nullptr, 215 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, Info)) 216 return Info; 217 return None; 218 } 219 220 // Returns "<internal>", "foo.a(bar.o)" or "baz.o". 221 std::string lld::toString(const InputFile *F) { 222 if (!F) 223 return "<internal>"; 224 225 if (F->ToStringCache.empty()) { 226 if (F->ArchiveName.empty()) 227 F->ToStringCache = F->getName(); 228 else 229 F->ToStringCache = (F->ArchiveName + "(" + F->getName() + ")").str(); 230 } 231 return F->ToStringCache; 232 } 233 234 template <class ELFT> 235 ELFFileBase<ELFT>::ELFFileBase(Kind K, MemoryBufferRef MB) : InputFile(K, MB) { 236 if (ELFT::TargetEndianness == support::little) 237 EKind = ELFT::Is64Bits ? ELF64LEKind : ELF32LEKind; 238 else 239 EKind = ELFT::Is64Bits ? ELF64BEKind : ELF32BEKind; 240 241 EMachine = getObj().getHeader()->e_machine; 242 OSABI = getObj().getHeader()->e_ident[llvm::ELF::EI_OSABI]; 243 } 244 245 template <class ELFT> 246 typename ELFT::SymRange ELFFileBase<ELFT>::getGlobalELFSyms() { 247 return makeArrayRef(ELFSyms.begin() + FirstGlobal, ELFSyms.end()); 248 } 249 250 template <class ELFT> 251 uint32_t ELFFileBase<ELFT>::getSectionIndex(const Elf_Sym &Sym) const { 252 return CHECK(getObj().getSectionIndex(&Sym, ELFSyms, SymtabSHNDX), this); 253 } 254 255 template <class ELFT> 256 void ELFFileBase<ELFT>::initSymtab(ArrayRef<Elf_Shdr> Sections, 257 const Elf_Shdr *Symtab) { 258 FirstGlobal = Symtab->sh_info; 259 ELFSyms = CHECK(getObj().symbols(Symtab), this); 260 if (FirstGlobal == 0 || FirstGlobal > ELFSyms.size()) 261 fatal(toString(this) + ": invalid sh_info in symbol table"); 262 263 StringTable = 264 CHECK(getObj().getStringTableForSymtab(*Symtab, Sections), this); 265 } 266 267 template <class ELFT> 268 ObjFile<ELFT>::ObjFile(MemoryBufferRef M, StringRef ArchiveName) 269 : ELFFileBase<ELFT>(Base::ObjKind, M) { 270 this->ArchiveName = ArchiveName; 271 } 272 273 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getLocalSymbols() { 274 if (this->Symbols.empty()) 275 return {}; 276 return makeArrayRef(this->Symbols).slice(1, this->FirstGlobal - 1); 277 } 278 279 template <class ELFT> ArrayRef<Symbol *> ObjFile<ELFT>::getGlobalSymbols() { 280 return makeArrayRef(this->Symbols).slice(this->FirstGlobal); 281 } 282 283 template <class ELFT> 284 void ObjFile<ELFT>::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 285 // Read a section table. JustSymbols is usually false. 286 if (this->JustSymbols) 287 initializeJustSymbols(); 288 else 289 initializeSections(ComdatGroups); 290 291 // Read a symbol table. 292 initializeSymbols(); 293 } 294 295 // Sections with SHT_GROUP and comdat bits define comdat section groups. 296 // They are identified and deduplicated by group name. This function 297 // returns a group name. 298 template <class ELFT> 299 StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> Sections, 300 const Elf_Shdr &Sec) { 301 // Group signatures are stored as symbol names in object files. 302 // sh_info contains a symbol index, so we fetch a symbol and read its name. 303 if (this->ELFSyms.empty()) 304 this->initSymtab( 305 Sections, CHECK(object::getSection<ELFT>(Sections, Sec.sh_link), this)); 306 307 const Elf_Sym *Sym = 308 CHECK(object::getSymbol<ELFT>(this->ELFSyms, Sec.sh_info), this); 309 StringRef Signature = CHECK(Sym->getName(this->StringTable), this); 310 311 // As a special case, if a symbol is a section symbol and has no name, 312 // we use a section name as a signature. 313 // 314 // Such SHT_GROUP sections are invalid from the perspective of the ELF 315 // standard, but GNU gold 1.14 (the newest version as of July 2017) or 316 // older produce such sections as outputs for the -r option, so we need 317 // a bug-compatibility. 318 if (Signature.empty() && Sym->getType() == STT_SECTION) 319 return getSectionName(Sec); 320 return Signature; 321 } 322 323 template <class ELFT> 324 ArrayRef<typename ObjFile<ELFT>::Elf_Word> 325 ObjFile<ELFT>::getShtGroupEntries(const Elf_Shdr &Sec) { 326 const ELFFile<ELFT> &Obj = this->getObj(); 327 ArrayRef<Elf_Word> Entries = 328 CHECK(Obj.template getSectionContentsAsArray<Elf_Word>(&Sec), this); 329 if (Entries.empty() || Entries[0] != GRP_COMDAT) 330 fatal(toString(this) + ": unsupported SHT_GROUP format"); 331 return Entries.slice(1); 332 } 333 334 template <class ELFT> bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &Sec) { 335 // On a regular link we don't merge sections if -O0 (default is -O1). This 336 // sometimes makes the linker significantly faster, although the output will 337 // be bigger. 338 // 339 // Doing the same for -r would create a problem as it would combine sections 340 // with different sh_entsize. One option would be to just copy every SHF_MERGE 341 // section as is to the output. While this would produce a valid ELF file with 342 // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when 343 // they see two .debug_str. We could have separate logic for combining 344 // SHF_MERGE sections based both on their name and sh_entsize, but that seems 345 // to be more trouble than it is worth. Instead, we just use the regular (-O1) 346 // logic for -r. 347 if (Config->Optimize == 0 && !Config->Relocatable) 348 return false; 349 350 // A mergeable section with size 0 is useless because they don't have 351 // any data to merge. A mergeable string section with size 0 can be 352 // argued as invalid because it doesn't end with a null character. 353 // We'll avoid a mess by handling them as if they were non-mergeable. 354 if (Sec.sh_size == 0) 355 return false; 356 357 // Check for sh_entsize. The ELF spec is not clear about the zero 358 // sh_entsize. It says that "the member [sh_entsize] contains 0 if 359 // the section does not hold a table of fixed-size entries". We know 360 // that Rust 1.13 produces a string mergeable section with a zero 361 // sh_entsize. Here we just accept it rather than being picky about it. 362 uint64_t EntSize = Sec.sh_entsize; 363 if (EntSize == 0) 364 return false; 365 if (Sec.sh_size % EntSize) 366 fatal(toString(this) + 367 ": SHF_MERGE section size must be a multiple of sh_entsize"); 368 369 uint64_t Flags = Sec.sh_flags; 370 if (!(Flags & SHF_MERGE)) 371 return false; 372 if (Flags & SHF_WRITE) 373 fatal(toString(this) + ": writable SHF_MERGE section is not supported"); 374 375 return true; 376 } 377 378 // This is for --just-symbols. 379 // 380 // --just-symbols is a very minor feature that allows you to link your 381 // output against other existing program, so that if you load both your 382 // program and the other program into memory, your output can refer the 383 // other program's symbols. 384 // 385 // When the option is given, we link "just symbols". The section table is 386 // initialized with null pointers. 387 template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() { 388 ArrayRef<Elf_Shdr> ObjSections = CHECK(this->getObj().sections(), this); 389 this->Sections.resize(ObjSections.size()); 390 391 for (const Elf_Shdr &Sec : ObjSections) { 392 if (Sec.sh_type != SHT_SYMTAB) 393 continue; 394 this->initSymtab(ObjSections, &Sec); 395 return; 396 } 397 } 398 399 template <class ELFT> 400 void ObjFile<ELFT>::initializeSections( 401 DenseSet<CachedHashStringRef> &ComdatGroups) { 402 const ELFFile<ELFT> &Obj = this->getObj(); 403 404 ArrayRef<Elf_Shdr> ObjSections = CHECK(Obj.sections(), this); 405 uint64_t Size = ObjSections.size(); 406 this->Sections.resize(Size); 407 this->SectionStringTable = 408 CHECK(Obj.getSectionStringTable(ObjSections), this); 409 410 for (size_t I = 0, E = ObjSections.size(); I < E; I++) { 411 if (this->Sections[I] == &InputSection::Discarded) 412 continue; 413 const Elf_Shdr &Sec = ObjSections[I]; 414 415 if (Sec.sh_type == ELF::SHT_LLVM_CALL_GRAPH_PROFILE) 416 CGProfile = check( 417 this->getObj().template getSectionContentsAsArray<Elf_CGProfile>( 418 &Sec)); 419 420 // SHF_EXCLUDE'ed sections are discarded by the linker. However, 421 // if -r is given, we'll let the final link discard such sections. 422 // This is compatible with GNU. 423 if ((Sec.sh_flags & SHF_EXCLUDE) && !Config->Relocatable) { 424 if (Sec.sh_type == SHT_LLVM_ADDRSIG) { 425 // We ignore the address-significance table if we know that the object 426 // file was created by objcopy or ld -r. This is because these tools 427 // will reorder the symbols in the symbol table, invalidating the data 428 // in the address-significance table, which refers to symbols by index. 429 if (Sec.sh_link != 0) 430 this->AddrsigSec = &Sec; 431 else if (Config->ICF == ICFLevel::Safe) 432 warn(toString(this) + ": --icf=safe is incompatible with object " 433 "files created using objcopy or ld -r"); 434 } 435 this->Sections[I] = &InputSection::Discarded; 436 continue; 437 } 438 439 switch (Sec.sh_type) { 440 case SHT_GROUP: { 441 // De-duplicate section groups by their signatures. 442 StringRef Signature = getShtGroupSignature(ObjSections, Sec); 443 bool IsNew = ComdatGroups.insert(CachedHashStringRef(Signature)).second; 444 this->Sections[I] = &InputSection::Discarded; 445 446 // We only support GRP_COMDAT type of group. Get the all entries of the 447 // section here to let getShtGroupEntries to check the type early for us. 448 ArrayRef<Elf_Word> Entries = getShtGroupEntries(Sec); 449 450 // If it is a new section group, we want to keep group members. 451 // Group leader sections, which contain indices of group members, are 452 // discarded because they are useless beyond this point. The only 453 // exception is the -r option because in order to produce re-linkable 454 // object files, we want to pass through basically everything. 455 if (IsNew) { 456 if (Config->Relocatable) 457 this->Sections[I] = createInputSection(Sec); 458 continue; 459 } 460 461 // Otherwise, discard group members. 462 for (uint32_t SecIndex : Entries) { 463 if (SecIndex >= Size) 464 fatal(toString(this) + 465 ": invalid section index in group: " + Twine(SecIndex)); 466 this->Sections[SecIndex] = &InputSection::Discarded; 467 } 468 break; 469 } 470 case SHT_SYMTAB: 471 this->initSymtab(ObjSections, &Sec); 472 break; 473 case SHT_SYMTAB_SHNDX: 474 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, ObjSections), this); 475 break; 476 case SHT_STRTAB: 477 case SHT_NULL: 478 break; 479 default: 480 this->Sections[I] = createInputSection(Sec); 481 } 482 483 // .ARM.exidx sections have a reverse dependency on the InputSection they 484 // have a SHF_LINK_ORDER dependency, this is identified by the sh_link. 485 if (Sec.sh_flags & SHF_LINK_ORDER) { 486 InputSectionBase *LinkSec = nullptr; 487 if (Sec.sh_link < this->Sections.size()) 488 LinkSec = this->Sections[Sec.sh_link]; 489 if (!LinkSec) 490 fatal(toString(this) + 491 ": invalid sh_link index: " + Twine(Sec.sh_link)); 492 493 InputSection *IS = cast<InputSection>(this->Sections[I]); 494 LinkSec->DependentSections.push_back(IS); 495 if (!isa<InputSection>(LinkSec)) 496 error("a section " + IS->Name + 497 " with SHF_LINK_ORDER should not refer a non-regular " 498 "section: " + 499 toString(LinkSec)); 500 } 501 } 502 } 503 504 // For ARM only, to set the EF_ARM_ABI_FLOAT_SOFT or EF_ARM_ABI_FLOAT_HARD 505 // flag in the ELF Header we need to look at Tag_ABI_VFP_args to find out how 506 // the input objects have been compiled. 507 static void updateARMVFPArgs(const ARMAttributeParser &Attributes, 508 const InputFile *F) { 509 if (!Attributes.hasAttribute(ARMBuildAttrs::ABI_VFP_args)) 510 // If an ABI tag isn't present then it is implicitly given the value of 0 511 // which maps to ARMBuildAttrs::BaseAAPCS. However many assembler files, 512 // including some in glibc that don't use FP args (and should have value 3) 513 // don't have the attribute so we do not consider an implicit value of 0 514 // as a clash. 515 return; 516 517 unsigned VFPArgs = Attributes.getAttributeValue(ARMBuildAttrs::ABI_VFP_args); 518 ARMVFPArgKind Arg; 519 switch (VFPArgs) { 520 case ARMBuildAttrs::BaseAAPCS: 521 Arg = ARMVFPArgKind::Base; 522 break; 523 case ARMBuildAttrs::HardFPAAPCS: 524 Arg = ARMVFPArgKind::VFP; 525 break; 526 case ARMBuildAttrs::ToolChainFPPCS: 527 // Tool chain specific convention that conforms to neither AAPCS variant. 528 Arg = ARMVFPArgKind::ToolChain; 529 break; 530 case ARMBuildAttrs::CompatibleFPAAPCS: 531 // Object compatible with all conventions. 532 return; 533 default: 534 error(toString(F) + ": unknown Tag_ABI_VFP_args value: " + Twine(VFPArgs)); 535 return; 536 } 537 // Follow ld.bfd and error if there is a mix of calling conventions. 538 if (Config->ARMVFPArgs != Arg && Config->ARMVFPArgs != ARMVFPArgKind::Default) 539 error(toString(F) + ": incompatible Tag_ABI_VFP_args"); 540 else 541 Config->ARMVFPArgs = Arg; 542 } 543 544 // The ARM support in lld makes some use of instructions that are not available 545 // on all ARM architectures. Namely: 546 // - Use of BLX instruction for interworking between ARM and Thumb state. 547 // - Use of the extended Thumb branch encoding in relocation. 548 // - Use of the MOVT/MOVW instructions in Thumb Thunks. 549 // The ARM Attributes section contains information about the architecture chosen 550 // at compile time. We follow the convention that if at least one input object 551 // is compiled with an architecture that supports these features then lld is 552 // permitted to use them. 553 static void updateSupportedARMFeatures(const ARMAttributeParser &Attributes) { 554 if (!Attributes.hasAttribute(ARMBuildAttrs::CPU_arch)) 555 return; 556 auto Arch = Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch); 557 switch (Arch) { 558 case ARMBuildAttrs::Pre_v4: 559 case ARMBuildAttrs::v4: 560 case ARMBuildAttrs::v4T: 561 // Architectures prior to v5 do not support BLX instruction 562 break; 563 case ARMBuildAttrs::v5T: 564 case ARMBuildAttrs::v5TE: 565 case ARMBuildAttrs::v5TEJ: 566 case ARMBuildAttrs::v6: 567 case ARMBuildAttrs::v6KZ: 568 case ARMBuildAttrs::v6K: 569 Config->ARMHasBlx = true; 570 // Architectures used in pre-Cortex processors do not support 571 // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception 572 // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do. 573 break; 574 default: 575 // All other Architectures have BLX and extended branch encoding 576 Config->ARMHasBlx = true; 577 Config->ARMJ1J2BranchEncoding = true; 578 if (Arch != ARMBuildAttrs::v6_M && Arch != ARMBuildAttrs::v6S_M) 579 // All Architectures used in Cortex processors with the exception 580 // of v6-M and v6S-M have the MOVT and MOVW instructions. 581 Config->ARMHasMovtMovw = true; 582 break; 583 } 584 } 585 586 template <class ELFT> 587 InputSectionBase *ObjFile<ELFT>::getRelocTarget(const Elf_Shdr &Sec) { 588 uint32_t Idx = Sec.sh_info; 589 if (Idx >= this->Sections.size()) 590 fatal(toString(this) + ": invalid relocated section index: " + Twine(Idx)); 591 InputSectionBase *Target = this->Sections[Idx]; 592 593 // Strictly speaking, a relocation section must be included in the 594 // group of the section it relocates. However, LLVM 3.3 and earlier 595 // would fail to do so, so we gracefully handle that case. 596 if (Target == &InputSection::Discarded) 597 return nullptr; 598 599 if (!Target) 600 fatal(toString(this) + ": unsupported relocation reference"); 601 return Target; 602 } 603 604 // Create a regular InputSection class that has the same contents 605 // as a given section. 606 static InputSection *toRegularSection(MergeInputSection *Sec) { 607 return make<InputSection>(Sec->File, Sec->Flags, Sec->Type, Sec->Alignment, 608 Sec->data(), Sec->Name); 609 } 610 611 template <class ELFT> 612 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &Sec) { 613 StringRef Name = getSectionName(Sec); 614 615 switch (Sec.sh_type) { 616 case SHT_ARM_ATTRIBUTES: { 617 if (Config->EMachine != EM_ARM) 618 break; 619 ARMAttributeParser Attributes; 620 ArrayRef<uint8_t> Contents = check(this->getObj().getSectionContents(&Sec)); 621 Attributes.Parse(Contents, /*isLittle*/ Config->EKind == ELF32LEKind); 622 updateSupportedARMFeatures(Attributes); 623 updateARMVFPArgs(Attributes, this); 624 625 // FIXME: Retain the first attribute section we see. The eglibc ARM 626 // dynamic loaders require the presence of an attribute section for dlopen 627 // to work. In a full implementation we would merge all attribute sections. 628 if (In.ARMAttributes == nullptr) { 629 In.ARMAttributes = make<InputSection>(*this, Sec, Name); 630 return In.ARMAttributes; 631 } 632 return &InputSection::Discarded; 633 } 634 case SHT_RELA: 635 case SHT_REL: { 636 // Find a relocation target section and associate this section with that. 637 // Target may have been discarded if it is in a different section group 638 // and the group is discarded, even though it's a violation of the 639 // spec. We handle that situation gracefully by discarding dangling 640 // relocation sections. 641 InputSectionBase *Target = getRelocTarget(Sec); 642 if (!Target) 643 return nullptr; 644 645 // This section contains relocation information. 646 // If -r is given, we do not interpret or apply relocation 647 // but just copy relocation sections to output. 648 if (Config->Relocatable) 649 return make<InputSection>(*this, Sec, Name); 650 651 if (Target->FirstRelocation) 652 fatal(toString(this) + 653 ": multiple relocation sections to one section are not supported"); 654 655 // ELF spec allows mergeable sections with relocations, but they are 656 // rare, and it is in practice hard to merge such sections by contents, 657 // because applying relocations at end of linking changes section 658 // contents. So, we simply handle such sections as non-mergeable ones. 659 // Degrading like this is acceptable because section merging is optional. 660 if (auto *MS = dyn_cast<MergeInputSection>(Target)) { 661 Target = toRegularSection(MS); 662 this->Sections[Sec.sh_info] = Target; 663 } 664 665 if (Sec.sh_type == SHT_RELA) { 666 ArrayRef<Elf_Rela> Rels = CHECK(this->getObj().relas(&Sec), this); 667 Target->FirstRelocation = Rels.begin(); 668 Target->NumRelocations = Rels.size(); 669 Target->AreRelocsRela = true; 670 } else { 671 ArrayRef<Elf_Rel> Rels = CHECK(this->getObj().rels(&Sec), this); 672 Target->FirstRelocation = Rels.begin(); 673 Target->NumRelocations = Rels.size(); 674 Target->AreRelocsRela = false; 675 } 676 assert(isUInt<31>(Target->NumRelocations)); 677 678 // Relocation sections processed by the linker are usually removed 679 // from the output, so returning `nullptr` for the normal case. 680 // However, if -emit-relocs is given, we need to leave them in the output. 681 // (Some post link analysis tools need this information.) 682 if (Config->EmitRelocs) { 683 InputSection *RelocSec = make<InputSection>(*this, Sec, Name); 684 // We will not emit relocation section if target was discarded. 685 Target->DependentSections.push_back(RelocSec); 686 return RelocSec; 687 } 688 return nullptr; 689 } 690 } 691 692 // The GNU linker uses .note.GNU-stack section as a marker indicating 693 // that the code in the object file does not expect that the stack is 694 // executable (in terms of NX bit). If all input files have the marker, 695 // the GNU linker adds a PT_GNU_STACK segment to tells the loader to 696 // make the stack non-executable. Most object files have this section as 697 // of 2017. 698 // 699 // But making the stack non-executable is a norm today for security 700 // reasons. Failure to do so may result in a serious security issue. 701 // Therefore, we make LLD always add PT_GNU_STACK unless it is 702 // explicitly told to do otherwise (by -z execstack). Because the stack 703 // executable-ness is controlled solely by command line options, 704 // .note.GNU-stack sections are simply ignored. 705 if (Name == ".note.GNU-stack") 706 return &InputSection::Discarded; 707 708 // Split stacks is a feature to support a discontiguous stack, 709 // commonly used in the programming language Go. For the details, 710 // see https://gcc.gnu.org/wiki/SplitStacks. An object file compiled 711 // for split stack will include a .note.GNU-split-stack section. 712 if (Name == ".note.GNU-split-stack") { 713 if (Config->Relocatable) { 714 error("cannot mix split-stack and non-split-stack in a relocatable link"); 715 return &InputSection::Discarded; 716 } 717 this->SplitStack = true; 718 return &InputSection::Discarded; 719 } 720 721 // An object file cmpiled for split stack, but where some of the 722 // functions were compiled with the no_split_stack_attribute will 723 // include a .note.GNU-no-split-stack section. 724 if (Name == ".note.GNU-no-split-stack") { 725 this->SomeNoSplitStack = true; 726 return &InputSection::Discarded; 727 } 728 729 // The linkonce feature is a sort of proto-comdat. Some glibc i386 object 730 // files contain definitions of symbol "__x86.get_pc_thunk.bx" in linkonce 731 // sections. Drop those sections to avoid duplicate symbol errors. 732 // FIXME: This is glibc PR20543, we should remove this hack once that has been 733 // fixed for a while. 734 if (Name.startswith(".gnu.linkonce.")) 735 return &InputSection::Discarded; 736 737 // If we are creating a new .build-id section, strip existing .build-id 738 // sections so that the output won't have more than one .build-id. 739 // This is not usually a problem because input object files normally don't 740 // have .build-id sections, but you can create such files by 741 // "ld.{bfd,gold,lld} -r --build-id", and we want to guard against it. 742 if (Name == ".note.gnu.build-id" && Config->BuildId != BuildIdKind::None) 743 return &InputSection::Discarded; 744 745 // The linker merges EH (exception handling) frames and creates a 746 // .eh_frame_hdr section for runtime. So we handle them with a special 747 // class. For relocatable outputs, they are just passed through. 748 if (Name == ".eh_frame" && !Config->Relocatable) 749 return make<EhInputSection>(*this, Sec, Name); 750 751 if (shouldMerge(Sec)) 752 return make<MergeInputSection>(*this, Sec, Name); 753 return make<InputSection>(*this, Sec, Name); 754 } 755 756 template <class ELFT> 757 StringRef ObjFile<ELFT>::getSectionName(const Elf_Shdr &Sec) { 758 return CHECK(this->getObj().getSectionName(&Sec, SectionStringTable), this); 759 } 760 761 template <class ELFT> void ObjFile<ELFT>::initializeSymbols() { 762 this->Symbols.reserve(this->ELFSyms.size()); 763 for (const Elf_Sym &Sym : this->ELFSyms) 764 this->Symbols.push_back(createSymbol(&Sym)); 765 } 766 767 template <class ELFT> Symbol *ObjFile<ELFT>::createSymbol(const Elf_Sym *Sym) { 768 int Binding = Sym->getBinding(); 769 770 uint32_t SecIdx = this->getSectionIndex(*Sym); 771 if (SecIdx >= this->Sections.size()) 772 fatal(toString(this) + ": invalid section index: " + Twine(SecIdx)); 773 774 InputSectionBase *Sec = this->Sections[SecIdx]; 775 uint8_t StOther = Sym->st_other; 776 uint8_t Type = Sym->getType(); 777 uint64_t Value = Sym->st_value; 778 uint64_t Size = Sym->st_size; 779 780 if (Binding == STB_LOCAL) { 781 if (Sym->getType() == STT_FILE) 782 SourceFile = CHECK(Sym->getName(this->StringTable), this); 783 784 if (this->StringTable.size() <= Sym->st_name) 785 fatal(toString(this) + ": invalid symbol name offset"); 786 787 StringRefZ Name = this->StringTable.data() + Sym->st_name; 788 if (Sym->st_shndx == SHN_UNDEF) 789 return make<Undefined>(this, Name, Binding, StOther, Type); 790 791 return make<Defined>(this, Name, Binding, StOther, Type, Value, Size, Sec); 792 } 793 794 StringRef Name = CHECK(Sym->getName(this->StringTable), this); 795 796 switch (Sym->st_shndx) { 797 case SHN_UNDEF: 798 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 799 /*CanOmitFromDynSym=*/false, this); 800 case SHN_COMMON: 801 if (Value == 0 || Value >= UINT32_MAX) 802 fatal(toString(this) + ": common symbol '" + Name + 803 "' has invalid alignment: " + Twine(Value)); 804 return Symtab->addCommon(Name, Size, Value, Binding, StOther, Type, *this); 805 } 806 807 switch (Binding) { 808 default: 809 fatal(toString(this) + ": unexpected binding: " + Twine(Binding)); 810 case STB_GLOBAL: 811 case STB_WEAK: 812 case STB_GNU_UNIQUE: 813 if (Sec == &InputSection::Discarded) 814 return Symtab->addUndefined<ELFT>(Name, Binding, StOther, Type, 815 /*CanOmitFromDynSym=*/false, this); 816 return Symtab->addDefined(Name, StOther, Type, Value, Size, Binding, Sec, 817 this); 818 } 819 } 820 821 ArchiveFile::ArchiveFile(std::unique_ptr<Archive> &&File) 822 : InputFile(ArchiveKind, File->getMemoryBufferRef()), 823 File(std::move(File)) {} 824 825 template <class ELFT> void ArchiveFile::parse() { 826 for (const Archive::Symbol &Sym : File->symbols()) 827 Symtab->addLazyArchive<ELFT>(Sym.getName(), *this, Sym); 828 } 829 830 // Returns a buffer pointing to a member file containing a given symbol. 831 InputFile *ArchiveFile::fetch(const Archive::Symbol &Sym) { 832 Archive::Child C = 833 CHECK(Sym.getMember(), toString(this) + 834 ": could not get the member for symbol " + 835 Sym.getName()); 836 837 if (!Seen.insert(C.getChildOffset()).second) 838 return nullptr; 839 840 MemoryBufferRef MB = 841 CHECK(C.getMemoryBufferRef(), 842 toString(this) + 843 ": could not get the buffer for the member defining symbol " + 844 Sym.getName()); 845 846 if (Tar && C.getParent()->isThin()) 847 Tar->append(relativeToRoot(CHECK(C.getFullName(), this)), MB.getBuffer()); 848 849 InputFile *File = createObjectFile( 850 MB, getName(), C.getParent()->isThin() ? 0 : C.getChildOffset()); 851 File->GroupId = GroupId; 852 return File; 853 } 854 855 template <class ELFT> 856 SharedFile<ELFT>::SharedFile(MemoryBufferRef M, StringRef DefaultSoName) 857 : ELFFileBase<ELFT>(Base::SharedKind, M), SoName(DefaultSoName), 858 IsNeeded(!Config->AsNeeded) {} 859 860 // Partially parse the shared object file so that we can call 861 // getSoName on this object. 862 template <class ELFT> void SharedFile<ELFT>::parseSoName() { 863 const Elf_Shdr *DynamicSec = nullptr; 864 const ELFFile<ELFT> Obj = this->getObj(); 865 ArrayRef<Elf_Shdr> Sections = CHECK(Obj.sections(), this); 866 867 // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d. 868 for (const Elf_Shdr &Sec : Sections) { 869 switch (Sec.sh_type) { 870 default: 871 continue; 872 case SHT_DYNSYM: 873 this->initSymtab(Sections, &Sec); 874 break; 875 case SHT_DYNAMIC: 876 DynamicSec = &Sec; 877 break; 878 case SHT_SYMTAB_SHNDX: 879 this->SymtabSHNDX = CHECK(Obj.getSHNDXTable(Sec, Sections), this); 880 break; 881 case SHT_GNU_versym: 882 this->VersymSec = &Sec; 883 break; 884 case SHT_GNU_verdef: 885 this->VerdefSec = &Sec; 886 break; 887 } 888 } 889 890 if (this->VersymSec && this->ELFSyms.empty()) 891 error("SHT_GNU_versym should be associated with symbol table"); 892 893 // Search for a DT_SONAME tag to initialize this->SoName. 894 if (!DynamicSec) 895 return; 896 ArrayRef<Elf_Dyn> Arr = 897 CHECK(Obj.template getSectionContentsAsArray<Elf_Dyn>(DynamicSec), this); 898 for (const Elf_Dyn &Dyn : Arr) { 899 if (Dyn.d_tag == DT_SONAME) { 900 uint64_t Val = Dyn.getVal(); 901 if (Val >= this->StringTable.size()) 902 fatal(toString(this) + ": invalid DT_SONAME entry"); 903 SoName = this->StringTable.data() + Val; 904 return; 905 } 906 } 907 } 908 909 // Parses ".gnu.version" section which is a parallel array for the symbol table. 910 // If a given file doesn't have ".gnu.version" section, returns VER_NDX_GLOBAL. 911 template <class ELFT> std::vector<uint32_t> SharedFile<ELFT>::parseVersyms() { 912 size_t Size = this->ELFSyms.size() - this->FirstGlobal; 913 if (!VersymSec) 914 return std::vector<uint32_t>(Size, VER_NDX_GLOBAL); 915 916 const char *Base = this->MB.getBuffer().data(); 917 const Elf_Versym *Versym = 918 reinterpret_cast<const Elf_Versym *>(Base + VersymSec->sh_offset) + 919 this->FirstGlobal; 920 921 std::vector<uint32_t> Ret(Size); 922 for (size_t I = 0; I < Size; ++I) 923 Ret[I] = Versym[I].vs_index; 924 return Ret; 925 } 926 927 // Parse the version definitions in the object file if present. Returns a vector 928 // whose nth element contains a pointer to the Elf_Verdef for version identifier 929 // n. Version identifiers that are not definitions map to nullptr. 930 template <class ELFT> 931 std::vector<const typename ELFT::Verdef *> SharedFile<ELFT>::parseVerdefs() { 932 if (!VerdefSec) 933 return {}; 934 935 // We cannot determine the largest verdef identifier without inspecting 936 // every Elf_Verdef, but both bfd and gold assign verdef identifiers 937 // sequentially starting from 1, so we predict that the largest identifier 938 // will be VerdefCount. 939 unsigned VerdefCount = VerdefSec->sh_info; 940 std::vector<const Elf_Verdef *> Verdefs(VerdefCount + 1); 941 942 // Build the Verdefs array by following the chain of Elf_Verdef objects 943 // from the start of the .gnu.version_d section. 944 const char *Base = this->MB.getBuffer().data(); 945 const char *Verdef = Base + VerdefSec->sh_offset; 946 for (unsigned I = 0; I != VerdefCount; ++I) { 947 auto *CurVerdef = reinterpret_cast<const Elf_Verdef *>(Verdef); 948 Verdef += CurVerdef->vd_next; 949 unsigned VerdefIndex = CurVerdef->vd_ndx; 950 Verdefs.resize(VerdefIndex + 1); 951 Verdefs[VerdefIndex] = CurVerdef; 952 } 953 954 return Verdefs; 955 } 956 957 // We do not usually care about alignments of data in shared object 958 // files because the loader takes care of it. However, if we promote a 959 // DSO symbol to point to .bss due to copy relocation, we need to keep 960 // the original alignment requirements. We infer it in this function. 961 template <class ELFT> 962 uint32_t SharedFile<ELFT>::getAlignment(ArrayRef<Elf_Shdr> Sections, 963 const Elf_Sym &Sym) { 964 uint64_t Ret = UINT64_MAX; 965 if (Sym.st_value) 966 Ret = 1ULL << countTrailingZeros((uint64_t)Sym.st_value); 967 if (0 < Sym.st_shndx && Sym.st_shndx < Sections.size()) 968 Ret = std::min<uint64_t>(Ret, Sections[Sym.st_shndx].sh_addralign); 969 return (Ret > UINT32_MAX) ? 0 : Ret; 970 } 971 972 // Fully parse the shared object file. This must be called after parseSoName(). 973 // 974 // This function parses symbol versions. If a DSO has version information, 975 // the file has a ".gnu.version_d" section which contains symbol version 976 // definitions. Each symbol is associated to one version through a table in 977 // ".gnu.version" section. That table is a parallel array for the symbol 978 // table, and each table entry contains an index in ".gnu.version_d". 979 // 980 // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for 981 // VER_NDX_GLOBAL. There's no table entry for these special versions in 982 // ".gnu.version_d". 983 // 984 // The file format for symbol versioning is perhaps a bit more complicated 985 // than necessary, but you can easily understand the code if you wrap your 986 // head around the data structure described above. 987 template <class ELFT> void SharedFile<ELFT>::parseRest() { 988 Verdefs = parseVerdefs(); // parse .gnu.version_d 989 std::vector<uint32_t> Versyms = parseVersyms(); // parse .gnu.version 990 ArrayRef<Elf_Shdr> Sections = CHECK(this->getObj().sections(), this); 991 992 // System libraries can have a lot of symbols with versions. Using a 993 // fixed buffer for computing the versions name (foo@ver) can save a 994 // lot of allocations. 995 SmallString<0> VersionedNameBuffer; 996 997 // Add symbols to the symbol table. 998 ArrayRef<Elf_Sym> Syms = this->getGlobalELFSyms(); 999 for (size_t I = 0; I < Syms.size(); ++I) { 1000 const Elf_Sym &Sym = Syms[I]; 1001 1002 // ELF spec requires that all local symbols precede weak or global 1003 // symbols in each symbol table, and the index of first non-local symbol 1004 // is stored to sh_info. If a local symbol appears after some non-local 1005 // symbol, that's a violation of the spec. 1006 StringRef Name = CHECK(Sym.getName(this->StringTable), this); 1007 if (Sym.getBinding() == STB_LOCAL) { 1008 warn("found local symbol '" + Name + 1009 "' in global part of symbol table in file " + toString(this)); 1010 continue; 1011 } 1012 1013 if (Sym.isUndefined()) { 1014 Symbol *S = Symtab->addUndefined<ELFT>(Name, Sym.getBinding(), 1015 Sym.st_other, Sym.getType(), 1016 /*CanOmitFromDynSym=*/false, this); 1017 S->ExportDynamic = true; 1018 continue; 1019 } 1020 1021 // MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly 1022 // assigns VER_NDX_LOCAL to this section global symbol. Here is a 1023 // workaround for this bug. 1024 uint32_t Idx = Versyms[I] & ~VERSYM_HIDDEN; 1025 if (Config->EMachine == EM_MIPS && Idx == VER_NDX_LOCAL && 1026 Name == "_gp_disp") 1027 continue; 1028 1029 uint64_t Alignment = getAlignment(Sections, Sym); 1030 if (!(Versyms[I] & VERSYM_HIDDEN)) 1031 Symtab->addShared(Name, *this, Sym, Alignment, Idx); 1032 1033 // Also add the symbol with the versioned name to handle undefined symbols 1034 // with explicit versions. 1035 if (Idx == VER_NDX_GLOBAL) 1036 continue; 1037 1038 if (Idx >= Verdefs.size() || Idx == VER_NDX_LOCAL) { 1039 error("corrupt input file: version definition index " + Twine(Idx) + 1040 " for symbol " + Name + " is out of bounds\n>>> defined in " + 1041 toString(this)); 1042 continue; 1043 } 1044 1045 StringRef VerName = 1046 this->StringTable.data() + Verdefs[Idx]->getAux()->vda_name; 1047 VersionedNameBuffer.clear(); 1048 Name = (Name + "@" + VerName).toStringRef(VersionedNameBuffer); 1049 Symtab->addShared(Saver.save(Name), *this, Sym, Alignment, Idx); 1050 } 1051 } 1052 1053 static ELFKind getBitcodeELFKind(const Triple &T) { 1054 if (T.isLittleEndian()) 1055 return T.isArch64Bit() ? ELF64LEKind : ELF32LEKind; 1056 return T.isArch64Bit() ? ELF64BEKind : ELF32BEKind; 1057 } 1058 1059 static uint8_t getBitcodeMachineKind(StringRef Path, const Triple &T) { 1060 switch (T.getArch()) { 1061 case Triple::aarch64: 1062 return EM_AARCH64; 1063 case Triple::amdgcn: 1064 case Triple::r600: 1065 return EM_AMDGPU; 1066 case Triple::arm: 1067 case Triple::thumb: 1068 return EM_ARM; 1069 case Triple::avr: 1070 return EM_AVR; 1071 case Triple::mips: 1072 case Triple::mipsel: 1073 case Triple::mips64: 1074 case Triple::mips64el: 1075 return EM_MIPS; 1076 case Triple::ppc: 1077 return EM_PPC; 1078 case Triple::ppc64: 1079 case Triple::ppc64le: 1080 return EM_PPC64; 1081 case Triple::x86: 1082 return T.isOSIAMCU() ? EM_IAMCU : EM_386; 1083 case Triple::x86_64: 1084 return EM_X86_64; 1085 default: 1086 error(Path + ": could not infer e_machine from bitcode target triple " + 1087 T.str()); 1088 return EM_NONE; 1089 } 1090 } 1091 1092 BitcodeFile::BitcodeFile(MemoryBufferRef MB, StringRef ArchiveName, 1093 uint64_t OffsetInArchive) 1094 : InputFile(BitcodeKind, MB) { 1095 this->ArchiveName = ArchiveName; 1096 1097 std::string Path = MB.getBufferIdentifier().str(); 1098 if (Config->ThinLTOIndexOnly) 1099 Path = replaceThinLTOSuffix(MB.getBufferIdentifier()); 1100 1101 // ThinLTO assumes that all MemoryBufferRefs given to it have a unique 1102 // name. If two archives define two members with the same name, this 1103 // causes a collision which result in only one of the objects being taken 1104 // into consideration at LTO time (which very likely causes undefined 1105 // symbols later in the link stage). So we append file offset to make 1106 // filename unique. 1107 MemoryBufferRef MBRef( 1108 MB.getBuffer(), 1109 Saver.save(ArchiveName + Path + 1110 (ArchiveName.empty() ? "" : utostr(OffsetInArchive)))); 1111 1112 Obj = CHECK(lto::InputFile::create(MBRef), this); 1113 1114 Triple T(Obj->getTargetTriple()); 1115 EKind = getBitcodeELFKind(T); 1116 EMachine = getBitcodeMachineKind(MB.getBufferIdentifier(), T); 1117 } 1118 1119 static uint8_t mapVisibility(GlobalValue::VisibilityTypes GvVisibility) { 1120 switch (GvVisibility) { 1121 case GlobalValue::DefaultVisibility: 1122 return STV_DEFAULT; 1123 case GlobalValue::HiddenVisibility: 1124 return STV_HIDDEN; 1125 case GlobalValue::ProtectedVisibility: 1126 return STV_PROTECTED; 1127 } 1128 llvm_unreachable("unknown visibility"); 1129 } 1130 1131 template <class ELFT> 1132 static Symbol *createBitcodeSymbol(const std::vector<bool> &KeptComdats, 1133 const lto::InputFile::Symbol &ObjSym, 1134 BitcodeFile &F) { 1135 StringRef Name = Saver.save(ObjSym.getName()); 1136 uint32_t Binding = ObjSym.isWeak() ? STB_WEAK : STB_GLOBAL; 1137 1138 uint8_t Type = ObjSym.isTLS() ? STT_TLS : STT_NOTYPE; 1139 uint8_t Visibility = mapVisibility(ObjSym.getVisibility()); 1140 bool CanOmitFromDynSym = ObjSym.canBeOmittedFromSymbolTable(); 1141 1142 int C = ObjSym.getComdatIndex(); 1143 if (C != -1 && !KeptComdats[C]) 1144 return Symtab->addUndefined<ELFT>(Name, Binding, Visibility, Type, 1145 CanOmitFromDynSym, &F); 1146 1147 if (ObjSym.isUndefined()) 1148 return Symtab->addUndefined<ELFT>(Name, Binding, Visibility, Type, 1149 CanOmitFromDynSym, &F); 1150 1151 if (ObjSym.isCommon()) 1152 return Symtab->addCommon(Name, ObjSym.getCommonSize(), 1153 ObjSym.getCommonAlignment(), Binding, Visibility, 1154 STT_OBJECT, F); 1155 1156 return Symtab->addBitcode(Name, Binding, Visibility, Type, CanOmitFromDynSym, 1157 F); 1158 } 1159 1160 template <class ELFT> 1161 void BitcodeFile::parse(DenseSet<CachedHashStringRef> &ComdatGroups) { 1162 std::vector<bool> KeptComdats; 1163 for (StringRef S : Obj->getComdatTable()) 1164 KeptComdats.push_back(ComdatGroups.insert(CachedHashStringRef(S)).second); 1165 1166 for (const lto::InputFile::Symbol &ObjSym : Obj->symbols()) 1167 Symbols.push_back(createBitcodeSymbol<ELFT>(KeptComdats, ObjSym, *this)); 1168 } 1169 1170 static ELFKind getELFKind(MemoryBufferRef MB) { 1171 unsigned char Size; 1172 unsigned char Endian; 1173 std::tie(Size, Endian) = getElfArchType(MB.getBuffer()); 1174 1175 if (Endian != ELFDATA2LSB && Endian != ELFDATA2MSB) 1176 fatal(MB.getBufferIdentifier() + ": invalid data encoding"); 1177 if (Size != ELFCLASS32 && Size != ELFCLASS64) 1178 fatal(MB.getBufferIdentifier() + ": invalid file class"); 1179 1180 size_t BufSize = MB.getBuffer().size(); 1181 if ((Size == ELFCLASS32 && BufSize < sizeof(Elf32_Ehdr)) || 1182 (Size == ELFCLASS64 && BufSize < sizeof(Elf64_Ehdr))) 1183 fatal(MB.getBufferIdentifier() + ": file is too short"); 1184 1185 if (Size == ELFCLASS32) 1186 return (Endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind; 1187 return (Endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind; 1188 } 1189 1190 void BinaryFile::parse() { 1191 ArrayRef<uint8_t> Data = toArrayRef(MB.getBuffer()); 1192 auto *Section = make<InputSection>(this, SHF_ALLOC | SHF_WRITE, SHT_PROGBITS, 1193 8, Data, ".data"); 1194 Sections.push_back(Section); 1195 1196 // For each input file foo that is embedded to a result as a binary 1197 // blob, we define _binary_foo_{start,end,size} symbols, so that 1198 // user programs can access blobs by name. Non-alphanumeric 1199 // characters in a filename are replaced with underscore. 1200 std::string S = "_binary_" + MB.getBufferIdentifier().str(); 1201 for (size_t I = 0; I < S.size(); ++I) 1202 if (!isAlnum(S[I])) 1203 S[I] = '_'; 1204 1205 Symtab->addDefined(Saver.save(S + "_start"), STV_DEFAULT, STT_OBJECT, 0, 0, 1206 STB_GLOBAL, Section, nullptr); 1207 Symtab->addDefined(Saver.save(S + "_end"), STV_DEFAULT, STT_OBJECT, 1208 Data.size(), 0, STB_GLOBAL, Section, nullptr); 1209 Symtab->addDefined(Saver.save(S + "_size"), STV_DEFAULT, STT_OBJECT, 1210 Data.size(), 0, STB_GLOBAL, nullptr, nullptr); 1211 } 1212 1213 InputFile *elf::createObjectFile(MemoryBufferRef MB, StringRef ArchiveName, 1214 uint64_t OffsetInArchive) { 1215 if (isBitcode(MB)) 1216 return make<BitcodeFile>(MB, ArchiveName, OffsetInArchive); 1217 1218 switch (getELFKind(MB)) { 1219 case ELF32LEKind: 1220 return make<ObjFile<ELF32LE>>(MB, ArchiveName); 1221 case ELF32BEKind: 1222 return make<ObjFile<ELF32BE>>(MB, ArchiveName); 1223 case ELF64LEKind: 1224 return make<ObjFile<ELF64LE>>(MB, ArchiveName); 1225 case ELF64BEKind: 1226 return make<ObjFile<ELF64BE>>(MB, ArchiveName); 1227 default: 1228 llvm_unreachable("getELFKind"); 1229 } 1230 } 1231 1232 InputFile *elf::createSharedFile(MemoryBufferRef MB, StringRef DefaultSoName) { 1233 switch (getELFKind(MB)) { 1234 case ELF32LEKind: 1235 return make<SharedFile<ELF32LE>>(MB, DefaultSoName); 1236 case ELF32BEKind: 1237 return make<SharedFile<ELF32BE>>(MB, DefaultSoName); 1238 case ELF64LEKind: 1239 return make<SharedFile<ELF64LE>>(MB, DefaultSoName); 1240 case ELF64BEKind: 1241 return make<SharedFile<ELF64BE>>(MB, DefaultSoName); 1242 default: 1243 llvm_unreachable("getELFKind"); 1244 } 1245 } 1246 1247 MemoryBufferRef LazyObjFile::getBuffer() { 1248 if (AddedToLink) 1249 return MemoryBufferRef(); 1250 AddedToLink = true; 1251 return MB; 1252 } 1253 1254 InputFile *LazyObjFile::fetch() { 1255 MemoryBufferRef MBRef = getBuffer(); 1256 if (MBRef.getBuffer().empty()) 1257 return nullptr; 1258 1259 InputFile *File = createObjectFile(MBRef, ArchiveName, OffsetInArchive); 1260 File->GroupId = GroupId; 1261 return File; 1262 } 1263 1264 template <class ELFT> void LazyObjFile::parse() { 1265 // A lazy object file wraps either a bitcode file or an ELF file. 1266 if (isBitcode(this->MB)) { 1267 std::unique_ptr<lto::InputFile> Obj = 1268 CHECK(lto::InputFile::create(this->MB), this); 1269 for (const lto::InputFile::Symbol &Sym : Obj->symbols()) 1270 if (!Sym.isUndefined()) 1271 Symtab->addLazyObject<ELFT>(Saver.save(Sym.getName()), *this); 1272 return; 1273 } 1274 1275 if (getELFKind(this->MB) != Config->EKind) { 1276 error("incompatible file: " + this->MB.getBufferIdentifier()); 1277 return; 1278 } 1279 1280 ELFFile<ELFT> Obj = check(ELFFile<ELFT>::create(MB.getBuffer())); 1281 ArrayRef<typename ELFT::Shdr> Sections = CHECK(Obj.sections(), this); 1282 1283 for (const typename ELFT::Shdr &Sec : Sections) { 1284 if (Sec.sh_type != SHT_SYMTAB) 1285 continue; 1286 1287 typename ELFT::SymRange Syms = CHECK(Obj.symbols(&Sec), this); 1288 uint32_t FirstGlobal = Sec.sh_info; 1289 StringRef StringTable = 1290 CHECK(Obj.getStringTableForSymtab(Sec, Sections), this); 1291 1292 for (const typename ELFT::Sym &Sym : Syms.slice(FirstGlobal)) 1293 if (Sym.st_shndx != SHN_UNDEF) 1294 Symtab->addLazyObject<ELFT>(CHECK(Sym.getName(StringTable), this), 1295 *this); 1296 return; 1297 } 1298 } 1299 1300 std::string elf::replaceThinLTOSuffix(StringRef Path) { 1301 StringRef Suffix = Config->ThinLTOObjectSuffixReplace.first; 1302 StringRef Repl = Config->ThinLTOObjectSuffixReplace.second; 1303 1304 if (Path.consume_back(Suffix)) 1305 return (Path + Repl).str(); 1306 return Path; 1307 } 1308 1309 template void ArchiveFile::parse<ELF32LE>(); 1310 template void ArchiveFile::parse<ELF32BE>(); 1311 template void ArchiveFile::parse<ELF64LE>(); 1312 template void ArchiveFile::parse<ELF64BE>(); 1313 1314 template void BitcodeFile::parse<ELF32LE>(DenseSet<CachedHashStringRef> &); 1315 template void BitcodeFile::parse<ELF32BE>(DenseSet<CachedHashStringRef> &); 1316 template void BitcodeFile::parse<ELF64LE>(DenseSet<CachedHashStringRef> &); 1317 template void BitcodeFile::parse<ELF64BE>(DenseSet<CachedHashStringRef> &); 1318 1319 template void LazyObjFile::parse<ELF32LE>(); 1320 template void LazyObjFile::parse<ELF32BE>(); 1321 template void LazyObjFile::parse<ELF64LE>(); 1322 template void LazyObjFile::parse<ELF64BE>(); 1323 1324 template class elf::ELFFileBase<ELF32LE>; 1325 template class elf::ELFFileBase<ELF32BE>; 1326 template class elf::ELFFileBase<ELF64LE>; 1327 template class elf::ELFFileBase<ELF64BE>; 1328 1329 template class elf::ObjFile<ELF32LE>; 1330 template class elf::ObjFile<ELF32BE>; 1331 template class elf::ObjFile<ELF64LE>; 1332 template class elf::ObjFile<ELF64BE>; 1333 1334 template class elf::SharedFile<ELF32LE>; 1335 template class elf::SharedFile<ELF32BE>; 1336 template class elf::SharedFile<ELF64LE>; 1337 template class elf::SharedFile<ELF64BE>; 1338