1 //===- Writer.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 "Writer.h" 11 #include "Config.h" 12 #include "Filesystem.h" 13 #include "LinkerScript.h" 14 #include "MapFile.h" 15 #include "Memory.h" 16 #include "OutputSections.h" 17 #include "Relocations.h" 18 #include "Strings.h" 19 #include "SymbolTable.h" 20 #include "SyntheticSections.h" 21 #include "Target.h" 22 #include "Threads.h" 23 #include "llvm/ADT/StringMap.h" 24 #include "llvm/ADT/StringSwitch.h" 25 #include "llvm/Support/FileOutputBuffer.h" 26 #include "llvm/Support/raw_ostream.h" 27 #include <climits> 28 29 using namespace llvm; 30 using namespace llvm::ELF; 31 using namespace llvm::object; 32 using namespace llvm::support; 33 using namespace llvm::support::endian; 34 35 using namespace lld; 36 using namespace lld::elf; 37 38 namespace { 39 // The writer writes a SymbolTable result to a file. 40 template <class ELFT> class Writer { 41 public: 42 typedef typename ELFT::Shdr Elf_Shdr; 43 typedef typename ELFT::Ehdr Elf_Ehdr; 44 typedef typename ELFT::Phdr Elf_Phdr; 45 46 void run(); 47 48 private: 49 void clearOutputSections(); 50 void createSyntheticSections(); 51 void copyLocalSymbols(); 52 void addSectionSymbols(); 53 void addReservedSymbols(); 54 void createSections(); 55 void forEachRelSec(std::function<void(InputSectionBase &)> Fn); 56 void sortSections(); 57 void finalizeSections(); 58 void addPredefinedSections(); 59 60 std::vector<PhdrEntry> createPhdrs(); 61 void removeEmptyPTLoad(); 62 void addPtArmExid(std::vector<PhdrEntry> &Phdrs); 63 void assignFileOffsets(); 64 void assignFileOffsetsBinary(); 65 void setPhdrs(); 66 void fixSectionAlignments(); 67 void fixPredefinedSymbols(); 68 void openFile(); 69 void writeHeader(); 70 void writeSections(); 71 void writeSectionsBinary(); 72 void writeBuildId(); 73 74 std::unique_ptr<FileOutputBuffer> Buffer; 75 76 OutputSectionFactory Factory; 77 78 void addRelIpltSymbols(); 79 void addStartEndSymbols(); 80 void addStartStopSymbols(OutputSection *Sec); 81 uint64_t getEntryAddr(); 82 OutputSection *findSectionInScript(StringRef Name); 83 OutputSectionCommand *findSectionCommand(StringRef Name); 84 85 std::vector<PhdrEntry> Phdrs; 86 87 uint64_t FileSize; 88 uint64_t SectionHeaderOff; 89 90 bool HasGotBaseSym = false; 91 }; 92 } // anonymous namespace 93 94 StringRef elf::getOutputSectionName(StringRef Name) { 95 // ".zdebug_" is a prefix for ZLIB-compressed sections. 96 // Because we decompressed input sections, we want to remove 'z'. 97 if (Name.startswith(".zdebug_")) 98 return Saver.save("." + Name.substr(2)); 99 100 if (Config->Relocatable) 101 return Name; 102 103 for (StringRef V : 104 {".text.", ".rodata.", ".data.rel.ro.", ".data.", ".bss.rel.ro.", 105 ".bss.", ".init_array.", ".fini_array.", ".ctors.", ".dtors.", ".tbss.", 106 ".gcc_except_table.", ".tdata.", ".ARM.exidx.", ".ARM.extab."}) { 107 StringRef Prefix = V.drop_back(); 108 if (Name.startswith(V) || Name == Prefix) 109 return Prefix; 110 } 111 112 // CommonSection is identified as "COMMON" in linker scripts. 113 // By default, it should go to .bss section. 114 if (Name == "COMMON") 115 return ".bss"; 116 117 return Name; 118 } 119 120 template <class ELFT> static bool needsInterpSection() { 121 return !Symtab<ELFT>::X->getSharedFiles().empty() && 122 !Config->DynamicLinker.empty() && !Script->ignoreInterpSection(); 123 } 124 125 template <class ELFT> void elf::writeResult() { Writer<ELFT>().run(); } 126 127 template <class ELFT> void Writer<ELFT>::removeEmptyPTLoad() { 128 auto I = std::remove_if(Phdrs.begin(), Phdrs.end(), [&](const PhdrEntry &P) { 129 if (P.p_type != PT_LOAD) 130 return false; 131 if (!P.First) 132 return true; 133 uint64_t Size = P.Last->Addr + P.Last->Size - P.First->Addr; 134 return Size == 0; 135 }); 136 Phdrs.erase(I, Phdrs.end()); 137 } 138 139 template <class ELFT> static void combineEhFrameSections() { 140 for (InputSectionBase *&S : InputSections) { 141 EhInputSection *ES = dyn_cast<EhInputSection>(S); 142 if (!ES || !ES->Live) 143 continue; 144 145 In<ELFT>::EhFrame->addSection(ES); 146 S = nullptr; 147 } 148 149 std::vector<InputSectionBase *> &V = InputSections; 150 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end()); 151 } 152 153 template <class ELFT> void Writer<ELFT>::clearOutputSections() { 154 // Clear the OutputSections to make sure it is not used anymore. Any 155 // code from this point on should be using the linker script 156 // commands. 157 for (OutputSection *Sec : OutputSections) 158 Sec->Sections.clear(); 159 OutputSections.clear(); 160 } 161 162 // The main function of the writer. 163 template <class ELFT> void Writer<ELFT>::run() { 164 // Create linker-synthesized sections such as .got or .plt. 165 // Such sections are of type input section. 166 createSyntheticSections(); 167 168 if (!Config->Relocatable) 169 combineEhFrameSections<ELFT>(); 170 171 // We need to create some reserved symbols such as _end. Create them. 172 if (!Config->Relocatable) 173 addReservedSymbols(); 174 175 // Create output sections. 176 if (Script->Opt.HasSections) { 177 // If linker script contains SECTIONS commands, let it create sections. 178 Script->processCommands(Factory); 179 180 // Linker scripts may have left some input sections unassigned. 181 // Assign such sections using the default rule. 182 Script->addOrphanSections(Factory); 183 } else { 184 // If linker script does not contain SECTIONS commands, create 185 // output sections by default rules. We still need to give the 186 // linker script a chance to run, because it might contain 187 // non-SECTIONS commands such as ASSERT. 188 Script->processCommands(Factory); 189 createSections(); 190 } 191 clearOutputSections(); 192 193 if (Config->Discard != DiscardPolicy::All) 194 copyLocalSymbols(); 195 196 if (Config->CopyRelocs) 197 addSectionSymbols(); 198 199 // Now that we have a complete set of output sections. This function 200 // completes section contents. For example, we need to add strings 201 // to the string table, and add entries to .got and .plt. 202 // finalizeSections does that. 203 finalizeSections(); 204 if (ErrorCount) 205 return; 206 207 if (!Script->Opt.HasSections && !Config->Relocatable) 208 fixSectionAlignments(); 209 210 // If -compressed-debug-sections is specified, we need to compress 211 // .debug_* sections. Do it right now because it changes the size of 212 // output sections. 213 parallelForEach( 214 OutputSectionCommands.begin(), OutputSectionCommands.end(), 215 [](OutputSectionCommand *Cmd) { Cmd->maybeCompress<ELFT>(); }); 216 217 Script->assignAddresses(); 218 Script->allocateHeaders(Phdrs); 219 220 // Remove empty PT_LOAD to avoid causing the dynamic linker to try to mmap a 221 // 0 sized region. This has to be done late since only after assignAddresses 222 // we know the size of the sections. 223 removeEmptyPTLoad(); 224 225 if (!Config->OFormatBinary) 226 assignFileOffsets(); 227 else 228 assignFileOffsetsBinary(); 229 230 setPhdrs(); 231 232 if (Config->Relocatable) { 233 for (OutputSectionCommand *Cmd : OutputSectionCommands) 234 Cmd->Sec->Addr = 0; 235 } else { 236 fixPredefinedSymbols(); 237 } 238 239 // It does not make sense try to open the file if we have error already. 240 if (ErrorCount) 241 return; 242 // Write the result down to a file. 243 openFile(); 244 if (ErrorCount) 245 return; 246 247 if (!Config->OFormatBinary) { 248 writeHeader(); 249 writeSections(); 250 } else { 251 writeSectionsBinary(); 252 } 253 254 // Backfill .note.gnu.build-id section content. This is done at last 255 // because the content is usually a hash value of the entire output file. 256 writeBuildId(); 257 if (ErrorCount) 258 return; 259 260 261 // Handle -Map option. 262 writeMapFile<ELFT>(OutputSectionCommands); 263 if (ErrorCount) 264 return; 265 266 if (auto EC = Buffer->commit()) 267 error("failed to write to the output file: " + EC.message()); 268 269 // Flush the output streams and exit immediately. A full shutdown 270 // is a good test that we are keeping track of all allocated memory, 271 // but actually freeing it is a waste of time in a regular linker run. 272 if (Config->ExitEarly) 273 exitLld(0); 274 } 275 276 // Initialize Out members. 277 template <class ELFT> void Writer<ELFT>::createSyntheticSections() { 278 // Initialize all pointers with NULL. This is needed because 279 // you can call lld::elf::main more than once as a library. 280 memset(&Out::First, 0, sizeof(Out)); 281 282 auto Add = [](InputSectionBase *Sec) { InputSections.push_back(Sec); }; 283 284 InX::DynStrTab = make<StringTableSection>(".dynstr", true); 285 InX::Dynamic = make<DynamicSection<ELFT>>(); 286 In<ELFT>::RelaDyn = make<RelocationSection<ELFT>>( 287 Config->IsRela ? ".rela.dyn" : ".rel.dyn", Config->ZCombreloc); 288 InX::ShStrTab = make<StringTableSection>(".shstrtab", false); 289 290 Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC); 291 Out::ElfHeader->Size = sizeof(Elf_Ehdr); 292 Out::ProgramHeaders = make<OutputSection>("", 0, SHF_ALLOC); 293 Out::ProgramHeaders->updateAlignment(Config->Wordsize); 294 295 if (needsInterpSection<ELFT>()) { 296 InX::Interp = createInterpSection(); 297 Add(InX::Interp); 298 } else { 299 InX::Interp = nullptr; 300 } 301 302 if (Config->Strip != StripPolicy::All) { 303 InX::StrTab = make<StringTableSection>(".strtab", false); 304 InX::SymTab = make<SymbolTableSection<ELFT>>(*InX::StrTab); 305 } 306 307 if (Config->BuildId != BuildIdKind::None) { 308 InX::BuildId = make<BuildIdSection>(); 309 Add(InX::BuildId); 310 } 311 312 InX::Common = createCommonSection<ELFT>(); 313 if (InX::Common) 314 Add(InX::Common); 315 316 InX::Bss = make<BssSection>(".bss"); 317 Add(InX::Bss); 318 InX::BssRelRo = make<BssSection>(".bss.rel.ro"); 319 Add(InX::BssRelRo); 320 321 // Add MIPS-specific sections. 322 bool HasDynSymTab = !Symtab<ELFT>::X->getSharedFiles().empty() || 323 Config->Pic || Config->ExportDynamic; 324 if (Config->EMachine == EM_MIPS) { 325 if (!Config->Shared && HasDynSymTab) { 326 InX::MipsRldMap = make<MipsRldMapSection>(); 327 Add(InX::MipsRldMap); 328 } 329 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create()) 330 Add(Sec); 331 if (auto *Sec = MipsOptionsSection<ELFT>::create()) 332 Add(Sec); 333 if (auto *Sec = MipsReginfoSection<ELFT>::create()) 334 Add(Sec); 335 } 336 337 if (HasDynSymTab) { 338 InX::DynSymTab = make<SymbolTableSection<ELFT>>(*InX::DynStrTab); 339 Add(InX::DynSymTab); 340 341 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>(); 342 Add(In<ELFT>::VerSym); 343 344 if (!Config->VersionDefinitions.empty()) { 345 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>(); 346 Add(In<ELFT>::VerDef); 347 } 348 349 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>(); 350 Add(In<ELFT>::VerNeed); 351 352 if (Config->GnuHash) { 353 InX::GnuHashTab = make<GnuHashTableSection>(); 354 Add(InX::GnuHashTab); 355 } 356 357 if (Config->SysvHash) { 358 In<ELFT>::HashTab = make<HashTableSection<ELFT>>(); 359 Add(In<ELFT>::HashTab); 360 } 361 362 Add(InX::Dynamic); 363 Add(InX::DynStrTab); 364 Add(In<ELFT>::RelaDyn); 365 } 366 367 // Add .got. MIPS' .got is so different from the other archs, 368 // it has its own class. 369 if (Config->EMachine == EM_MIPS) { 370 InX::MipsGot = make<MipsGotSection>(); 371 Add(InX::MipsGot); 372 } else { 373 InX::Got = make<GotSection>(); 374 Add(InX::Got); 375 } 376 377 InX::GotPlt = make<GotPltSection>(); 378 Add(InX::GotPlt); 379 InX::IgotPlt = make<IgotPltSection>(); 380 Add(InX::IgotPlt); 381 382 if (Config->GdbIndex) { 383 InX::GdbIndex = make<GdbIndexSection>(); 384 Add(InX::GdbIndex); 385 } 386 387 // We always need to add rel[a].plt to output if it has entries. 388 // Even for static linking it can contain R_[*]_IRELATIVE relocations. 389 In<ELFT>::RelaPlt = make<RelocationSection<ELFT>>( 390 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/); 391 Add(In<ELFT>::RelaPlt); 392 393 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure 394 // that the IRelative relocations are processed last by the dynamic loader 395 In<ELFT>::RelaIplt = make<RelocationSection<ELFT>>( 396 (Config->EMachine == EM_ARM) ? ".rel.dyn" : In<ELFT>::RelaPlt->Name, 397 false /*Sort*/); 398 Add(In<ELFT>::RelaIplt); 399 400 InX::Plt = make<PltSection>(Target->PltHeaderSize); 401 Add(InX::Plt); 402 InX::Iplt = make<PltSection>(0); 403 Add(InX::Iplt); 404 405 if (!Config->Relocatable) { 406 if (Config->EhFrameHdr) { 407 In<ELFT>::EhFrameHdr = make<EhFrameHeader<ELFT>>(); 408 Add(In<ELFT>::EhFrameHdr); 409 } 410 In<ELFT>::EhFrame = make<EhFrameSection<ELFT>>(); 411 Add(In<ELFT>::EhFrame); 412 } 413 414 if (InX::SymTab) 415 Add(InX::SymTab); 416 Add(InX::ShStrTab); 417 if (InX::StrTab) 418 Add(InX::StrTab); 419 } 420 421 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName, 422 const SymbolBody &B) { 423 if (B.isFile() || B.isSection()) 424 return false; 425 426 // If sym references a section in a discarded group, don't keep it. 427 if (Sec == &InputSection::Discarded) 428 return false; 429 430 if (Config->Discard == DiscardPolicy::None) 431 return true; 432 433 // In ELF assembly .L symbols are normally discarded by the assembler. 434 // If the assembler fails to do so, the linker discards them if 435 // * --discard-locals is used. 436 // * The symbol is in a SHF_MERGE section, which is normally the reason for 437 // the assembler keeping the .L symbol. 438 if (!SymName.startswith(".L") && !SymName.empty()) 439 return true; 440 441 if (Config->Discard == DiscardPolicy::Locals) 442 return false; 443 444 return !Sec || !(Sec->Flags & SHF_MERGE); 445 } 446 447 static bool includeInSymtab(const SymbolBody &B) { 448 if (!B.isLocal() && !B.symbol()->IsUsedInRegularObj) 449 return false; 450 451 if (auto *D = dyn_cast<DefinedRegular>(&B)) { 452 // Always include absolute symbols. 453 SectionBase *Sec = D->Section; 454 if (!Sec) 455 return true; 456 if (auto *IS = dyn_cast<InputSectionBase>(Sec)) { 457 Sec = IS->Repl; 458 IS = cast<InputSectionBase>(Sec); 459 // Exclude symbols pointing to garbage-collected sections. 460 if (!IS->Live) 461 return false; 462 } 463 if (auto *S = dyn_cast<MergeInputSection>(Sec)) 464 if (!S->getSectionPiece(D->Value)->Live) 465 return false; 466 } 467 return true; 468 } 469 470 // Local symbols are not in the linker's symbol table. This function scans 471 // each object file's symbol table to copy local symbols to the output. 472 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() { 473 if (!InX::SymTab) 474 return; 475 for (elf::ObjectFile<ELFT> *F : Symtab<ELFT>::X->getObjectFiles()) { 476 for (SymbolBody *B : F->getLocalSymbols()) { 477 if (!B->IsLocal) 478 fatal(toString(F) + 479 ": broken object: getLocalSymbols returns a non-local symbol"); 480 auto *DR = dyn_cast<DefinedRegular>(B); 481 482 // No reason to keep local undefined symbol in symtab. 483 if (!DR) 484 continue; 485 if (!includeInSymtab(*B)) 486 continue; 487 488 SectionBase *Sec = DR->Section; 489 if (!shouldKeepInSymtab(Sec, B->getName(), *B)) 490 continue; 491 InX::SymTab->addSymbol(B); 492 } 493 } 494 } 495 496 template <class ELFT> void Writer<ELFT>::addSectionSymbols() { 497 // Create one STT_SECTION symbol for each output section we might 498 // have a relocation with. 499 for (BaseCommand *Base : Script->Opt.Commands) { 500 auto *Cmd = dyn_cast<OutputSectionCommand>(Base); 501 if (!Cmd) 502 continue; 503 auto I = llvm::find_if(Cmd->Commands, [](BaseCommand *Base) { 504 if (auto *ISD = dyn_cast<InputSectionDescription>(Base)) 505 return !ISD->Sections.empty(); 506 return false; 507 }); 508 if (I == Cmd->Commands.end()) 509 continue; 510 InputSection *IS = cast<InputSectionDescription>(*I)->Sections[0]; 511 if (isa<SyntheticSection>(IS) || IS->Type == SHT_REL || 512 IS->Type == SHT_RELA) 513 continue; 514 515 auto *Sym = 516 make<DefinedRegular>("", /*IsLocal=*/true, /*StOther=*/0, STT_SECTION, 517 /*Value=*/0, /*Size=*/0, IS, nullptr); 518 InX::SymTab->addSymbol(Sym); 519 } 520 } 521 522 // Today's loaders have a feature to make segments read-only after 523 // processing dynamic relocations to enhance security. PT_GNU_RELRO 524 // is defined for that. 525 // 526 // This function returns true if a section needs to be put into a 527 // PT_GNU_RELRO segment. 528 bool elf::isRelroSection(const OutputSection *Sec) { 529 if (!Config->ZRelro) 530 return false; 531 532 uint64_t Flags = Sec->Flags; 533 534 // Non-allocatable or non-writable sections don't need RELRO because 535 // they are not writable or not even mapped to memory in the first place. 536 // RELRO is for sections that are essentially read-only but need to 537 // be writable only at process startup to allow dynamic linker to 538 // apply relocations. 539 if (!(Flags & SHF_ALLOC) || !(Flags & SHF_WRITE)) 540 return false; 541 542 // Once initialized, TLS data segments are used as data templates 543 // for a thread-local storage. For each new thread, runtime 544 // allocates memory for a TLS and copy templates there. No thread 545 // are supposed to use templates directly. Thus, it can be in RELRO. 546 if (Flags & SHF_TLS) 547 return true; 548 549 // .init_array, .preinit_array and .fini_array contain pointers to 550 // functions that are executed on process startup or exit. These 551 // pointers are set by the static linker, and they are not expected 552 // to change at runtime. But if you are an attacker, you could do 553 // interesting things by manipulating pointers in .fini_array, for 554 // example. So they are put into RELRO. 555 uint32_t Type = Sec->Type; 556 if (Type == SHT_INIT_ARRAY || Type == SHT_FINI_ARRAY || 557 Type == SHT_PREINIT_ARRAY) 558 return true; 559 560 // .got contains pointers to external symbols. They are resolved by 561 // the dynamic linker when a module is loaded into memory, and after 562 // that they are not expected to change. So, it can be in RELRO. 563 if (InX::Got && Sec == InX::Got->getParent()) 564 return true; 565 566 // .got.plt contains pointers to external function symbols. They are 567 // by default resolved lazily, so we usually cannot put it into RELRO. 568 // However, if "-z now" is given, the lazy symbol resolution is 569 // disabled, which enables us to put it into RELRO. 570 if (Sec == InX::GotPlt->getParent()) 571 return Config->ZNow; 572 573 // .dynamic section contains data for the dynamic linker, and 574 // there's no need to write to it at runtime, so it's better to put 575 // it into RELRO. 576 if (Sec == InX::Dynamic->getParent()) 577 return true; 578 579 // .bss.rel.ro is used for copy relocations for read-only symbols. 580 // Since the dynamic linker needs to process copy relocations, the 581 // section cannot be read-only, but once initialized, they shouldn't 582 // change. 583 if (Sec == InX::BssRelRo->getParent()) 584 return true; 585 586 // Sections with some special names are put into RELRO. This is a 587 // bit unfortunate because section names shouldn't be significant in 588 // ELF in spirit. But in reality many linker features depend on 589 // magic section names. 590 StringRef S = Sec->Name; 591 return S == ".data.rel.ro" || S == ".ctors" || S == ".dtors" || S == ".jcr" || 592 S == ".eh_frame" || S == ".openbsd.randomdata"; 593 } 594 595 // We compute a rank for each section. The rank indicates where the 596 // section should be placed in the file. Instead of using simple 597 // numbers (0,1,2...), we use a series of flags. One for each decision 598 // point when placing the section. 599 // Using flags has two key properties: 600 // * It is easy to check if a give branch was taken. 601 // * It is easy two see how similar two ranks are (see getRankProximity). 602 enum RankFlags { 603 RF_NOT_ADDR_SET = 1 << 16, 604 RF_NOT_INTERP = 1 << 15, 605 RF_NOT_ALLOC = 1 << 14, 606 RF_WRITE = 1 << 13, 607 RF_EXEC_WRITE = 1 << 12, 608 RF_EXEC = 1 << 11, 609 RF_NON_TLS_BSS = 1 << 10, 610 RF_NON_TLS_BSS_RO = 1 << 9, 611 RF_NOT_TLS = 1 << 8, 612 RF_BSS = 1 << 7, 613 RF_PPC_NOT_TOCBSS = 1 << 6, 614 RF_PPC_OPD = 1 << 5, 615 RF_PPC_TOCL = 1 << 4, 616 RF_PPC_TOC = 1 << 3, 617 RF_PPC_BRANCH_LT = 1 << 2, 618 RF_MIPS_GPREL = 1 << 1, 619 RF_MIPS_NOT_GOT = 1 << 0 620 }; 621 622 static unsigned getSectionRank(const OutputSection *Sec) { 623 unsigned Rank = 0; 624 625 // We want to put section specified by -T option first, so we 626 // can start assigning VA starting from them later. 627 if (Config->SectionStartMap.count(Sec->Name)) 628 return Rank; 629 Rank |= RF_NOT_ADDR_SET; 630 631 // Put .interp first because some loaders want to see that section 632 // on the first page of the executable file when loaded into memory. 633 if (Sec->Name == ".interp") 634 return Rank; 635 Rank |= RF_NOT_INTERP; 636 637 // Allocatable sections go first to reduce the total PT_LOAD size and 638 // so debug info doesn't change addresses in actual code. 639 if (!(Sec->Flags & SHF_ALLOC)) 640 return Rank | RF_NOT_ALLOC; 641 642 // Sort sections based on their access permission in the following 643 // order: R, RX, RWX, RW. This order is based on the following 644 // considerations: 645 // * Read-only sections come first such that they go in the 646 // PT_LOAD covering the program headers at the start of the file. 647 // * Read-only, executable sections come next, unless the 648 // -no-rosegment option is used. 649 // * Writable, executable sections follow such that .plt on 650 // architectures where it needs to be writable will be placed 651 // between .text and .data. 652 // * Writable sections come last, such that .bss lands at the very 653 // end of the last PT_LOAD. 654 bool IsExec = Sec->Flags & SHF_EXECINSTR; 655 bool IsWrite = Sec->Flags & SHF_WRITE; 656 657 if (IsExec) { 658 if (IsWrite) 659 Rank |= RF_EXEC_WRITE; 660 else if (!Config->SingleRoRx) 661 Rank |= RF_EXEC; 662 } else { 663 if (IsWrite) 664 Rank |= RF_WRITE; 665 } 666 667 // If we got here we know that both A and B are in the same PT_LOAD. 668 669 bool IsTls = Sec->Flags & SHF_TLS; 670 bool IsNoBits = Sec->Type == SHT_NOBITS; 671 672 // The first requirement we have is to put (non-TLS) nobits sections last. The 673 // reason is that the only thing the dynamic linker will see about them is a 674 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the 675 // PT_LOAD, so that has to correspond to the nobits sections. 676 bool IsNonTlsNoBits = IsNoBits && !IsTls; 677 if (IsNonTlsNoBits) 678 Rank |= RF_NON_TLS_BSS; 679 680 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo 681 // sections after r/w ones, so that the RelRo sections are contiguous. 682 bool IsRelRo = isRelroSection(Sec); 683 if (IsNonTlsNoBits && !IsRelRo) 684 Rank |= RF_NON_TLS_BSS_RO; 685 if (!IsNonTlsNoBits && IsRelRo) 686 Rank |= RF_NON_TLS_BSS_RO; 687 688 // The TLS initialization block needs to be a single contiguous block in a R/W 689 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W 690 // sections. The TLS NOBITS sections are placed here as they don't take up 691 // virtual address space in the PT_LOAD. 692 if (!IsTls) 693 Rank |= RF_NOT_TLS; 694 695 // Within the TLS initialization block, the non-nobits sections need to appear 696 // first. 697 if (IsNoBits) 698 Rank |= RF_BSS; 699 700 // // Some architectures have additional ordering restrictions for sections 701 // // within the same PT_LOAD. 702 if (Config->EMachine == EM_PPC64) { 703 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections 704 // that we would like to make sure appear is a specific order to maximize 705 // their coverage by a single signed 16-bit offset from the TOC base 706 // pointer. Conversely, the special .tocbss section should be first among 707 // all SHT_NOBITS sections. This will put it next to the loaded special 708 // PPC64 sections (and, thus, within reach of the TOC base pointer). 709 StringRef Name = Sec->Name; 710 if (Name != ".tocbss") 711 Rank |= RF_PPC_NOT_TOCBSS; 712 713 if (Name == ".opd") 714 Rank |= RF_PPC_OPD; 715 716 if (Name == ".toc1") 717 Rank |= RF_PPC_TOCL; 718 719 if (Name == ".toc") 720 Rank |= RF_PPC_TOC; 721 722 if (Name == ".branch_lt") 723 Rank |= RF_PPC_BRANCH_LT; 724 } 725 if (Config->EMachine == EM_MIPS) { 726 // All sections with SHF_MIPS_GPREL flag should be grouped together 727 // because data in these sections is addressable with a gp relative address. 728 if (Sec->Flags & SHF_MIPS_GPREL) 729 Rank |= RF_MIPS_GPREL; 730 731 if (Sec->Name != ".got") 732 Rank |= RF_MIPS_NOT_GOT; 733 } 734 735 return Rank; 736 } 737 738 static bool compareSections(const BaseCommand *ACmd, const BaseCommand *BCmd) { 739 const OutputSection *A = cast<OutputSectionCommand>(ACmd)->Sec; 740 const OutputSection *B = cast<OutputSectionCommand>(BCmd)->Sec; 741 if (A->SortRank != B->SortRank) 742 return A->SortRank < B->SortRank; 743 if (!(A->SortRank & RF_NOT_ADDR_SET)) 744 return Config->SectionStartMap.lookup(A->Name) < 745 Config->SectionStartMap.lookup(B->Name); 746 return false; 747 } 748 749 void PhdrEntry::add(OutputSection *Sec) { 750 Last = Sec; 751 if (!First) 752 First = Sec; 753 p_align = std::max(p_align, Sec->Alignment); 754 if (p_type == PT_LOAD) 755 Sec->FirstInPtLoad = First; 756 } 757 758 template <class ELFT> 759 static Symbol *addRegular(StringRef Name, SectionBase *Sec, uint64_t Value, 760 uint8_t StOther = STV_HIDDEN, 761 uint8_t Binding = STB_WEAK) { 762 // The linker generated symbols are added as STB_WEAK to allow user defined 763 // ones to override them. 764 return Symtab<ELFT>::X->addRegular(Name, StOther, STT_NOTYPE, Value, 765 /*Size=*/0, Binding, Sec, 766 /*File=*/nullptr); 767 } 768 769 template <class ELFT> 770 static DefinedRegular * 771 addOptionalRegular(StringRef Name, SectionBase *Sec, uint64_t Val, 772 uint8_t StOther = STV_HIDDEN, uint8_t Binding = STB_GLOBAL) { 773 SymbolBody *S = Symtab<ELFT>::X->find(Name); 774 if (!S) 775 return nullptr; 776 if (S->isInCurrentDSO()) 777 return nullptr; 778 return cast<DefinedRegular>( 779 addRegular<ELFT>(Name, Sec, Val, StOther, Binding)->body()); 780 } 781 782 // The beginning and the ending of .rel[a].plt section are marked 783 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked 784 // executable. The runtime needs these symbols in order to resolve 785 // all IRELATIVE relocs on startup. For dynamic executables, we don't 786 // need these symbols, since IRELATIVE relocs are resolved through GOT 787 // and PLT. For details, see http://www.airs.com/blog/archives/403. 788 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() { 789 if (InX::DynSymTab) 790 return; 791 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start"; 792 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, 0, STV_HIDDEN, STB_WEAK); 793 794 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end"; 795 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, -1, STV_HIDDEN, STB_WEAK); 796 } 797 798 // The linker is expected to define some symbols depending on 799 // the linking result. This function defines such symbols. 800 template <class ELFT> void Writer<ELFT>::addReservedSymbols() { 801 if (Config->EMachine == EM_MIPS) { 802 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer 803 // so that it points to an absolute address which by default is relative 804 // to GOT. Default offset is 0x7ff0. 805 // See "Global Data Symbols" in Chapter 6 in the following document: 806 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf 807 ElfSym::MipsGp = Symtab<ELFT>::X->addAbsolute("_gp", STV_HIDDEN, STB_LOCAL); 808 809 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between 810 // start of function and 'gp' pointer into GOT. 811 if (Symtab<ELFT>::X->find("_gp_disp")) 812 ElfSym::MipsGpDisp = 813 Symtab<ELFT>::X->addAbsolute("_gp_disp", STV_HIDDEN, STB_LOCAL); 814 815 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp' 816 // pointer. This symbol is used in the code generated by .cpload pseudo-op 817 // in case of using -mno-shared option. 818 // https://sourceware.org/ml/binutils/2004-12/msg00094.html 819 if (Symtab<ELFT>::X->find("__gnu_local_gp")) 820 ElfSym::MipsLocalGp = 821 Symtab<ELFT>::X->addAbsolute("__gnu_local_gp", STV_HIDDEN, STB_LOCAL); 822 } 823 824 // The _GLOBAL_OFFSET_TABLE_ symbol is defined by target convention to 825 // be at some offset from the base of the .got section, usually 0 or the end 826 // of the .got 827 InputSection *GotSection = InX::MipsGot ? cast<InputSection>(InX::MipsGot) 828 : cast<InputSection>(InX::Got); 829 ElfSym::GlobalOffsetTable = addOptionalRegular<ELFT>( 830 "_GLOBAL_OFFSET_TABLE_", GotSection, Target->GotBaseSymOff); 831 832 // __tls_get_addr is defined by the dynamic linker for dynamic ELFs. For 833 // static linking the linker is required to optimize away any references to 834 // __tls_get_addr, so it's not defined anywhere. Create a hidden definition 835 // to avoid the undefined symbol error. 836 if (!InX::DynSymTab) 837 Symtab<ELFT>::X->addIgnored("__tls_get_addr"); 838 839 // __ehdr_start is the location of ELF file headers. Note that we define 840 // this symbol unconditionally even when using a linker script, which 841 // differs from the behavior implemented by GNU linker which only define 842 // this symbol if ELF headers are in the memory mapped segment. 843 // __executable_start is not documented, but the expectation of at 844 // least the android libc is that it points to the elf header too. 845 // __dso_handle symbol is passed to cxa_finalize as a marker to identify 846 // each DSO. The address of the symbol doesn't matter as long as they are 847 // different in different DSOs, so we chose the start address of the DSO. 848 for (const char *Name : 849 {"__ehdr_start", "__executable_start", "__dso_handle"}) 850 addOptionalRegular<ELFT>(Name, Out::ElfHeader, 0, STV_HIDDEN); 851 852 // If linker script do layout we do not need to create any standart symbols. 853 if (Script->Opt.HasSections) 854 return; 855 856 auto Add = [](StringRef S) { 857 return addOptionalRegular<ELFT>(S, Out::ElfHeader, 0, STV_DEFAULT); 858 }; 859 860 ElfSym::Bss = Add("__bss_start"); 861 ElfSym::End1 = Add("end"); 862 ElfSym::End2 = Add("_end"); 863 ElfSym::Etext1 = Add("etext"); 864 ElfSym::Etext2 = Add("_etext"); 865 ElfSym::Edata1 = Add("edata"); 866 ElfSym::Edata2 = Add("_edata"); 867 } 868 869 // Sort input sections by section name suffixes for 870 // __attribute__((init_priority(N))). 871 static void sortInitFini(OutputSectionCommand *Cmd) { 872 if (Cmd) 873 Cmd->sortInitFini(); 874 } 875 876 // Sort input sections by the special rule for .ctors and .dtors. 877 static void sortCtorsDtors(OutputSectionCommand *Cmd) { 878 if (Cmd) 879 Cmd->sortCtorsDtors(); 880 } 881 882 // Sort input sections using the list provided by --symbol-ordering-file. 883 template <class ELFT> static void sortBySymbolsOrder() { 884 if (Config->SymbolOrderingFile.empty()) 885 return; 886 887 // Build a map from symbols to their priorities. Symbols that didn't 888 // appear in the symbol ordering file have the lowest priority 0. 889 // All explicitly mentioned symbols have negative (higher) priorities. 890 DenseMap<StringRef, int> SymbolOrder; 891 int Priority = -Config->SymbolOrderingFile.size(); 892 for (StringRef S : Config->SymbolOrderingFile) 893 SymbolOrder.insert({S, Priority++}); 894 895 // Build a map from sections to their priorities. 896 DenseMap<SectionBase *, int> SectionOrder; 897 for (elf::ObjectFile<ELFT> *File : Symtab<ELFT>::X->getObjectFiles()) { 898 for (SymbolBody *Body : File->getSymbols()) { 899 auto *D = dyn_cast<DefinedRegular>(Body); 900 if (!D || !D->Section) 901 continue; 902 int &Priority = SectionOrder[D->Section]; 903 Priority = std::min(Priority, SymbolOrder.lookup(D->getName())); 904 } 905 } 906 907 // Sort sections by priority. 908 for (BaseCommand *Base : Script->Opt.Commands) 909 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) 910 Cmd->sort([&](InputSectionBase *S) { return SectionOrder.lookup(S); }); 911 } 912 913 template <class ELFT> 914 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) { 915 for (InputSectionBase *IS : InputSections) { 916 if (!IS->Live) 917 continue; 918 // Scan all relocations. Each relocation goes through a series 919 // of tests to determine if it needs special treatment, such as 920 // creating GOT, PLT, copy relocations, etc. 921 // Note that relocations for non-alloc sections are directly 922 // processed by InputSection::relocateNonAlloc. 923 if (!(IS->Flags & SHF_ALLOC)) 924 continue; 925 if (isa<InputSection>(IS) || isa<EhInputSection>(IS)) 926 Fn(*IS); 927 } 928 929 if (!Config->Relocatable) { 930 for (EhInputSection *ES : In<ELFT>::EhFrame->Sections) 931 Fn(*ES); 932 } 933 } 934 935 template <class ELFT> void Writer<ELFT>::createSections() { 936 for (InputSectionBase *IS : InputSections) 937 if (IS) 938 Factory.addInputSec(IS, getOutputSectionName(IS->Name)); 939 940 Script->fabricateDefaultCommands(); 941 sortBySymbolsOrder<ELFT>(); 942 sortInitFini(findSectionCommand(".init_array")); 943 sortInitFini(findSectionCommand(".fini_array")); 944 sortCtorsDtors(findSectionCommand(".ctors")); 945 sortCtorsDtors(findSectionCommand(".dtors")); 946 } 947 948 // We want to find how similar two ranks are. 949 // The more branches in getSectionRank that match, the more similar they are. 950 // Since each branch corresponds to a bit flag, we can just use 951 // countLeadingZeros. 952 static int getRankProximity(OutputSection *A, OutputSection *B) { 953 return countLeadingZeros(A->SortRank ^ B->SortRank); 954 } 955 956 static int getRankProximity(OutputSection *A, BaseCommand *B) { 957 if (auto *Cmd = dyn_cast<OutputSectionCommand>(B)) 958 if (Cmd->Sec) 959 return getRankProximity(A, Cmd->Sec); 960 return -1; 961 } 962 963 // When placing orphan sections, we want to place them after symbol assignments 964 // so that an orphan after 965 // begin_foo = .; 966 // foo : { *(foo) } 967 // end_foo = .; 968 // doesn't break the intended meaning of the begin/end symbols. 969 // We don't want to go over sections since findOrphanPos is the 970 // one in charge of deciding the order of the sections. 971 // We don't want to go over changes to '.', since doing so in 972 // rx_sec : { *(rx_sec) } 973 // . = ALIGN(0x1000); 974 // /* The RW PT_LOAD starts here*/ 975 // rw_sec : { *(rw_sec) } 976 // would mean that the RW PT_LOAD would become unaligned. 977 static bool shouldSkip(BaseCommand *Cmd) { 978 if (isa<OutputSectionCommand>(Cmd)) 979 return false; 980 if (auto *Assign = dyn_cast<SymbolAssignment>(Cmd)) 981 return Assign->Name != "."; 982 return true; 983 } 984 985 // We want to place orphan sections so that they share as much 986 // characteristics with their neighbors as possible. For example, if 987 // both are rw, or both are tls. 988 template <typename ELFT> 989 static std::vector<BaseCommand *>::iterator 990 findOrphanPos(std::vector<BaseCommand *>::iterator B, 991 std::vector<BaseCommand *>::iterator E) { 992 OutputSection *Sec = cast<OutputSectionCommand>(*E)->Sec; 993 994 // Find the first element that has as close a rank as possible. 995 auto I = std::max_element(B, E, [=](BaseCommand *A, BaseCommand *B) { 996 return getRankProximity(Sec, A) < getRankProximity(Sec, B); 997 }); 998 if (I == E) 999 return E; 1000 1001 // Consider all existing sections with the same proximity. 1002 int Proximity = getRankProximity(Sec, *I); 1003 for (; I != E; ++I) { 1004 auto *Cmd = dyn_cast<OutputSectionCommand>(*I); 1005 if (!Cmd || !Cmd->Sec) 1006 continue; 1007 if (getRankProximity(Sec, Cmd->Sec) != Proximity || 1008 Sec->SortRank < Cmd->Sec->SortRank) 1009 break; 1010 } 1011 auto J = std::find_if( 1012 llvm::make_reverse_iterator(I), llvm::make_reverse_iterator(B), 1013 [](BaseCommand *Cmd) { return isa<OutputSectionCommand>(Cmd); }); 1014 I = J.base(); 1015 while (I != E && shouldSkip(*I)) 1016 ++I; 1017 return I; 1018 } 1019 1020 template <class ELFT> void Writer<ELFT>::sortSections() { 1021 if (Script->Opt.HasSections) 1022 Script->adjustSectionsBeforeSorting(); 1023 1024 // Don't sort if using -r. It is not necessary and we want to preserve the 1025 // relative order for SHF_LINK_ORDER sections. 1026 if (Config->Relocatable) 1027 return; 1028 1029 for (BaseCommand *Base : Script->Opt.Commands) 1030 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) 1031 if (OutputSection *Sec = Cmd->Sec) 1032 Sec->SortRank = getSectionRank(Sec); 1033 1034 if (!Script->Opt.HasSections) { 1035 // We know that all the OutputSectionCommands are contiguous in 1036 // this case. 1037 auto E = Script->Opt.Commands.end(); 1038 auto I = Script->Opt.Commands.begin(); 1039 auto IsSection = [](BaseCommand *Base) { 1040 return isa<OutputSectionCommand>(Base); 1041 }; 1042 I = std::find_if(I, E, IsSection); 1043 E = std::find_if(llvm::make_reverse_iterator(E), 1044 llvm::make_reverse_iterator(I), IsSection) 1045 .base(); 1046 std::stable_sort(I, E, compareSections); 1047 return; 1048 } 1049 1050 // Orphan sections are sections present in the input files which are 1051 // not explicitly placed into the output file by the linker script. 1052 // 1053 // The sections in the linker script are already in the correct 1054 // order. We have to figuere out where to insert the orphan 1055 // sections. 1056 // 1057 // The order of the sections in the script is arbitrary and may not agree with 1058 // compareSections. This means that we cannot easily define a strict weak 1059 // ordering. To see why, consider a comparison of a section in the script and 1060 // one not in the script. We have a two simple options: 1061 // * Make them equivalent (a is not less than b, and b is not less than a). 1062 // The problem is then that equivalence has to be transitive and we can 1063 // have sections a, b and c with only b in a script and a less than c 1064 // which breaks this property. 1065 // * Use compareSectionsNonScript. Given that the script order doesn't have 1066 // to match, we can end up with sections a, b, c, d where b and c are in the 1067 // script and c is compareSectionsNonScript less than b. In which case d 1068 // can be equivalent to c, a to b and d < a. As a concrete example: 1069 // .a (rx) # not in script 1070 // .b (rx) # in script 1071 // .c (ro) # in script 1072 // .d (ro) # not in script 1073 // 1074 // The way we define an order then is: 1075 // * Sort only the orphan sections. They are in the end right now. 1076 // * Move each orphan section to its preferred position. We try 1077 // to put each section in the last position where it it can share 1078 // a PT_LOAD. 1079 // 1080 // There is some ambiguity as to where exactly a new entry should be 1081 // inserted, because Opt.Commands contains not only output section 1082 // commands but also other types of commands such as symbol assignment 1083 // expressions. There's no correct answer here due to the lack of the 1084 // formal specification of the linker script. We use heuristics to 1085 // determine whether a new output command should be added before or 1086 // after another commands. For the details, look at shouldSkip 1087 // function. 1088 1089 auto I = Script->Opt.Commands.begin(); 1090 auto E = Script->Opt.Commands.end(); 1091 auto NonScriptI = std::find_if(I, E, [](BaseCommand *Base) { 1092 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) 1093 return Cmd->Sec && Cmd->Sec->SectionIndex == INT_MAX; 1094 return false; 1095 }); 1096 1097 // Sort the orphan sections. 1098 std::stable_sort(NonScriptI, E, compareSections); 1099 1100 // As a horrible special case, skip the first . assignment if it is before any 1101 // section. We do this because it is common to set a load address by starting 1102 // the script with ". = 0xabcd" and the expectation is that every section is 1103 // after that. 1104 auto FirstSectionOrDotAssignment = 1105 std::find_if(I, E, [](BaseCommand *Cmd) { return !shouldSkip(Cmd); }); 1106 if (FirstSectionOrDotAssignment != E && 1107 isa<SymbolAssignment>(**FirstSectionOrDotAssignment)) 1108 ++FirstSectionOrDotAssignment; 1109 I = FirstSectionOrDotAssignment; 1110 1111 while (NonScriptI != E) { 1112 auto Pos = findOrphanPos<ELFT>(I, NonScriptI); 1113 OutputSection *Orphan = cast<OutputSectionCommand>(*NonScriptI)->Sec; 1114 1115 // As an optimization, find all sections with the same sort rank 1116 // and insert them with one rotate. 1117 unsigned Rank = Orphan->SortRank; 1118 auto End = std::find_if(NonScriptI + 1, E, [=](BaseCommand *Cmd) { 1119 return cast<OutputSectionCommand>(Cmd)->Sec->SortRank != Rank; 1120 }); 1121 std::rotate(Pos, NonScriptI, End); 1122 NonScriptI = End; 1123 } 1124 1125 Script->adjustSectionsAfterSorting(); 1126 } 1127 1128 static void applySynthetic(const std::vector<SyntheticSection *> &Sections, 1129 std::function<void(SyntheticSection *)> Fn) { 1130 for (SyntheticSection *SS : Sections) 1131 if (SS && SS->getParent() && !SS->empty()) 1132 Fn(SS); 1133 } 1134 1135 // We need to add input synthetic sections early in createSyntheticSections() 1136 // to make them visible from linkescript side. But not all sections are always 1137 // required to be in output. For example we don't need dynamic section content 1138 // sometimes. This function filters out such unused sections from the output. 1139 static void removeUnusedSyntheticSections() { 1140 // All input synthetic sections that can be empty are placed after 1141 // all regular ones. We iterate over them all and exit at first 1142 // non-synthetic. 1143 for (InputSectionBase *S : llvm::reverse(InputSections)) { 1144 SyntheticSection *SS = dyn_cast<SyntheticSection>(S); 1145 if (!SS) 1146 return; 1147 OutputSection *OS = SS->getParent(); 1148 if (!SS->empty() || !OS) 1149 continue; 1150 if ((SS == InX::Got || SS == InX::MipsGot) && ElfSym::GlobalOffsetTable) 1151 continue; 1152 1153 OutputSectionCommand *Cmd = Script->getCmd(OS); 1154 std::vector<BaseCommand *>::iterator Empty = Cmd->Commands.end(); 1155 for (auto I = Cmd->Commands.begin(), E = Cmd->Commands.end(); I != E; ++I) { 1156 BaseCommand *B = *I; 1157 if (auto *ISD = dyn_cast<InputSectionDescription>(B)) { 1158 auto P = std::find(ISD->Sections.begin(), ISD->Sections.end(), SS); 1159 if (P != ISD->Sections.end()) 1160 ISD->Sections.erase(P); 1161 if (ISD->Sections.empty()) 1162 Empty = I; 1163 } 1164 } 1165 if (Empty != Cmd->Commands.end()) 1166 Cmd->Commands.erase(Empty); 1167 1168 // If there are no other sections in the output section, remove it from the 1169 // output. 1170 if (Cmd->Commands.empty()) { 1171 // Also remove script commands matching the output section. 1172 auto &Cmds = Script->Opt.Commands; 1173 auto I = std::remove_if(Cmds.begin(), Cmds.end(), [&](BaseCommand *Cmd) { 1174 if (auto *OSCmd = dyn_cast<OutputSectionCommand>(Cmd)) 1175 return OSCmd->Sec == OS; 1176 return false; 1177 }); 1178 Cmds.erase(I, Cmds.end()); 1179 } 1180 } 1181 } 1182 1183 // Create output section objects and add them to OutputSections. 1184 template <class ELFT> void Writer<ELFT>::finalizeSections() { 1185 Out::DebugInfo = findSectionInScript(".debug_info"); 1186 Out::PreinitArray = findSectionInScript(".preinit_array"); 1187 Out::InitArray = findSectionInScript(".init_array"); 1188 Out::FiniArray = findSectionInScript(".fini_array"); 1189 1190 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop 1191 // symbols for sections, so that the runtime can get the start and end 1192 // addresses of each section by section name. Add such symbols. 1193 if (!Config->Relocatable) { 1194 addStartEndSymbols(); 1195 for (BaseCommand *Base : Script->Opt.Commands) 1196 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) 1197 if (Cmd->Sec) 1198 addStartStopSymbols(Cmd->Sec); 1199 } 1200 1201 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type. 1202 // It should be okay as no one seems to care about the type. 1203 // Even the author of gold doesn't remember why gold behaves that way. 1204 // https://sourceware.org/ml/binutils/2002-03/msg00360.html 1205 if (InX::DynSymTab) 1206 addRegular<ELFT>("_DYNAMIC", InX::Dynamic, 0); 1207 1208 // Define __rel[a]_iplt_{start,end} symbols if needed. 1209 addRelIpltSymbols(); 1210 1211 // This responsible for splitting up .eh_frame section into 1212 // pieces. The relocation scan uses those pieces, so this has to be 1213 // earlier. 1214 applySynthetic({In<ELFT>::EhFrame}, 1215 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1216 1217 // Scan relocations. This must be done after every symbol is declared so that 1218 // we can correctly decide if a dynamic relocation is needed. 1219 forEachRelSec(scanRelocations<ELFT>); 1220 1221 if (InX::Plt && !InX::Plt->empty()) 1222 InX::Plt->addSymbols(); 1223 if (InX::Iplt && !InX::Iplt->empty()) 1224 InX::Iplt->addSymbols(); 1225 1226 // Now that we have defined all possible global symbols including linker- 1227 // synthesized ones. Visit all symbols to give the finishing touches. 1228 for (Symbol *S : Symtab<ELFT>::X->getSymbols()) { 1229 SymbolBody *Body = S->body(); 1230 1231 if (!includeInSymtab(*Body)) 1232 continue; 1233 if (InX::SymTab) 1234 InX::SymTab->addSymbol(Body); 1235 1236 if (InX::DynSymTab && S->includeInDynsym()) { 1237 InX::DynSymTab->addSymbol(Body); 1238 if (auto *SS = dyn_cast<SharedSymbol>(Body)) 1239 if (cast<SharedFile<ELFT>>(SS->File)->isNeeded()) 1240 In<ELFT>::VerNeed->addSymbol(SS); 1241 } 1242 } 1243 1244 // Do not proceed if there was an undefined symbol. 1245 if (ErrorCount) 1246 return; 1247 1248 addPredefinedSections(); 1249 removeUnusedSyntheticSections(); 1250 1251 sortSections(); 1252 1253 // Now that we have the final list, create a list of all the 1254 // OutputSectionCommands for convenience. 1255 for (BaseCommand *Base : Script->Opt.Commands) 1256 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) 1257 OutputSectionCommands.push_back(Cmd); 1258 1259 // Prefer command line supplied address over other constraints. 1260 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1261 auto I = Config->SectionStartMap.find(Cmd->Name); 1262 if (I != Config->SectionStartMap.end()) 1263 Cmd->AddrExpr = [=] { return I->second; }; 1264 } 1265 1266 // This is a bit of a hack. A value of 0 means undef, so we set it 1267 // to 1 t make __ehdr_start defined. The section number is not 1268 // particularly relevant. 1269 Out::ElfHeader->SectionIndex = 1; 1270 1271 unsigned I = 1; 1272 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1273 OutputSection *Sec = Cmd->Sec; 1274 Sec->SectionIndex = I++; 1275 Sec->ShName = InX::ShStrTab->addString(Sec->Name); 1276 } 1277 1278 // Binary and relocatable output does not have PHDRS. 1279 // The headers have to be created before finalize as that can influence the 1280 // image base and the dynamic section on mips includes the image base. 1281 if (!Config->Relocatable && !Config->OFormatBinary) { 1282 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs(); 1283 addPtArmExid(Phdrs); 1284 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size(); 1285 } 1286 1287 // Dynamic section must be the last one in this list and dynamic 1288 // symbol table section (DynSymTab) must be the first one. 1289 applySynthetic({InX::DynSymTab, InX::Bss, InX::BssRelRo, 1290 InX::GnuHashTab, In<ELFT>::HashTab, InX::SymTab, 1291 InX::ShStrTab, InX::StrTab, In<ELFT>::VerDef, 1292 InX::DynStrTab, InX::GdbIndex, InX::Got, 1293 InX::MipsGot, InX::IgotPlt, InX::GotPlt, 1294 In<ELFT>::RelaDyn, In<ELFT>::RelaIplt, In<ELFT>::RelaPlt, 1295 InX::Plt, InX::Iplt, In<ELFT>::EhFrameHdr, 1296 In<ELFT>::VerSym, In<ELFT>::VerNeed, InX::Dynamic}, 1297 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1298 1299 // Some architectures use small displacements for jump instructions. 1300 // It is linker's responsibility to create thunks containing long 1301 // jump instructions if jump targets are too far. Create thunks. 1302 if (Target->NeedsThunks) { 1303 // FIXME: only ARM Interworking and Mips LA25 Thunks are implemented, 1304 // these 1305 // do not require address information. To support range extension Thunks 1306 // we need to assign addresses so that we can tell if jump instructions 1307 // are out of range. This will need to turn into a loop that converges 1308 // when no more Thunks are added 1309 ThunkCreator TC; 1310 Script->assignAddresses(); 1311 if (TC.createThunks(OutputSectionCommands)) { 1312 applySynthetic({InX::MipsGot}, 1313 [](SyntheticSection *SS) { SS->updateAllocSize(); }); 1314 if (TC.createThunks(OutputSectionCommands)) 1315 fatal("All non-range thunks should be created in first call"); 1316 } 1317 } 1318 1319 // Fill other section headers. The dynamic table is finalized 1320 // at the end because some tags like RELSZ depend on result 1321 // of finalizing other sections. 1322 for (OutputSectionCommand *Cmd : OutputSectionCommands) 1323 Cmd->finalize<ELFT>(); 1324 1325 // createThunks may have added local symbols to the static symbol table 1326 applySynthetic({InX::SymTab, InX::ShStrTab, InX::StrTab}, 1327 [](SyntheticSection *SS) { SS->postThunkContents(); }); 1328 } 1329 1330 template <class ELFT> void Writer<ELFT>::addPredefinedSections() { 1331 // ARM ABI requires .ARM.exidx to be terminated by some piece of data. 1332 // We have the terminater synthetic section class. Add that at the end. 1333 OutputSectionCommand *Cmd = findSectionCommand(".ARM.exidx"); 1334 if (!Cmd || Cmd->Commands.empty() || Config->Relocatable) 1335 return; 1336 1337 auto *Sentinel = make<ARMExidxSentinelSection>(); 1338 Cmd->Sec->addSection(Sentinel); 1339 // Add the sentinel to the last of these too. 1340 auto ISD = std::find_if(Cmd->Commands.rbegin(), Cmd->Commands.rend(), 1341 [](const BaseCommand *Base) { 1342 return isa<InputSectionDescription>(Base); 1343 }); 1344 cast<InputSectionDescription>(*ISD)->Sections.push_back(Sentinel); 1345 } 1346 1347 // The linker is expected to define SECNAME_start and SECNAME_end 1348 // symbols for a few sections. This function defines them. 1349 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() { 1350 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) { 1351 // These symbols resolve to the image base if the section does not exist. 1352 // A special value -1 indicates end of the section. 1353 if (OS) { 1354 addOptionalRegular<ELFT>(Start, OS, 0); 1355 addOptionalRegular<ELFT>(End, OS, -1); 1356 } else { 1357 if (Config->Pic) 1358 OS = Out::ElfHeader; 1359 addOptionalRegular<ELFT>(Start, OS, 0); 1360 addOptionalRegular<ELFT>(End, OS, 0); 1361 } 1362 }; 1363 1364 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray); 1365 Define("__init_array_start", "__init_array_end", Out::InitArray); 1366 Define("__fini_array_start", "__fini_array_end", Out::FiniArray); 1367 1368 if (OutputSection *Sec = findSectionInScript(".ARM.exidx")) 1369 Define("__exidx_start", "__exidx_end", Sec); 1370 } 1371 1372 // If a section name is valid as a C identifier (which is rare because of 1373 // the leading '.'), linkers are expected to define __start_<secname> and 1374 // __stop_<secname> symbols. They are at beginning and end of the section, 1375 // respectively. This is not requested by the ELF standard, but GNU ld and 1376 // gold provide the feature, and used by many programs. 1377 template <class ELFT> 1378 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) { 1379 StringRef S = Sec->Name; 1380 if (!isValidCIdentifier(S)) 1381 return; 1382 addOptionalRegular<ELFT>(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT); 1383 addOptionalRegular<ELFT>(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT); 1384 } 1385 1386 template <class ELFT> 1387 OutputSectionCommand *Writer<ELFT>::findSectionCommand(StringRef Name) { 1388 for (BaseCommand *Base : Script->Opt.Commands) 1389 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) 1390 if (Cmd->Name == Name) 1391 return Cmd; 1392 return nullptr; 1393 } 1394 1395 template <class ELFT> OutputSection *Writer<ELFT>::findSectionInScript(StringRef Name) { 1396 if (OutputSectionCommand *Cmd = findSectionCommand(Name)) 1397 return Cmd->Sec; 1398 return nullptr; 1399 } 1400 1401 static bool needsPtLoad(OutputSection *Sec) { 1402 if (!(Sec->Flags & SHF_ALLOC)) 1403 return false; 1404 1405 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is 1406 // responsible for allocating space for them, not the PT_LOAD that 1407 // contains the TLS initialization image. 1408 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS) 1409 return false; 1410 return true; 1411 } 1412 1413 // Linker scripts are responsible for aligning addresses. Unfortunately, most 1414 // linker scripts are designed for creating two PT_LOADs only, one RX and one 1415 // RW. This means that there is no alignment in the RO to RX transition and we 1416 // cannot create a PT_LOAD there. 1417 static uint64_t computeFlags(uint64_t Flags) { 1418 if (Config->Omagic) 1419 return PF_R | PF_W | PF_X; 1420 if (Config->SingleRoRx && !(Flags & PF_W)) 1421 return Flags | PF_X; 1422 return Flags; 1423 } 1424 1425 // Decide which program headers to create and which sections to include in each 1426 // one. 1427 template <class ELFT> std::vector<PhdrEntry> Writer<ELFT>::createPhdrs() { 1428 std::vector<PhdrEntry> Ret; 1429 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * { 1430 Ret.emplace_back(Type, Flags); 1431 return &Ret.back(); 1432 }; 1433 1434 // The first phdr entry is PT_PHDR which describes the program header itself. 1435 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders); 1436 1437 // PT_INTERP must be the second entry if exists. 1438 if (OutputSection *Sec = findSectionInScript(".interp")) 1439 AddHdr(PT_INTERP, Sec->getPhdrFlags())->add(Sec); 1440 1441 // Add the first PT_LOAD segment for regular output sections. 1442 uint64_t Flags = computeFlags(PF_R); 1443 PhdrEntry *Load = AddHdr(PT_LOAD, Flags); 1444 1445 // Add the headers. We will remove them if they don't fit. 1446 Load->add(Out::ElfHeader); 1447 Load->add(Out::ProgramHeaders); 1448 1449 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1450 OutputSection *Sec = Cmd->Sec; 1451 if (!(Sec->Flags & SHF_ALLOC)) 1452 break; 1453 if (!needsPtLoad(Sec)) 1454 continue; 1455 1456 // Segments are contiguous memory regions that has the same attributes 1457 // (e.g. executable or writable). There is one phdr for each segment. 1458 // Therefore, we need to create a new phdr when the next section has 1459 // different flags or is loaded at a discontiguous address using AT linker 1460 // script command. 1461 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags()); 1462 if (Cmd->LMAExpr || Flags != NewFlags) { 1463 Load = AddHdr(PT_LOAD, NewFlags); 1464 Flags = NewFlags; 1465 } 1466 1467 Load->add(Sec); 1468 } 1469 1470 // Add a TLS segment if any. 1471 PhdrEntry TlsHdr(PT_TLS, PF_R); 1472 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1473 OutputSection *Sec = Cmd->Sec; 1474 if (Sec->Flags & SHF_TLS) 1475 TlsHdr.add(Sec); 1476 } 1477 if (TlsHdr.First) 1478 Ret.push_back(std::move(TlsHdr)); 1479 1480 // Add an entry for .dynamic. 1481 if (InX::DynSymTab) 1482 AddHdr(PT_DYNAMIC, InX::Dynamic->getParent()->getPhdrFlags()) 1483 ->add(InX::Dynamic->getParent()); 1484 1485 // PT_GNU_RELRO includes all sections that should be marked as 1486 // read-only by dynamic linker after proccessing relocations. 1487 PhdrEntry RelRo(PT_GNU_RELRO, PF_R); 1488 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1489 OutputSection *Sec = Cmd->Sec; 1490 if (needsPtLoad(Sec) && isRelroSection(Sec)) 1491 RelRo.add(Sec); 1492 } 1493 if (RelRo.First) 1494 Ret.push_back(std::move(RelRo)); 1495 1496 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr. 1497 if (!In<ELFT>::EhFrame->empty() && In<ELFT>::EhFrameHdr && 1498 In<ELFT>::EhFrame->getParent() && In<ELFT>::EhFrameHdr->getParent()) 1499 AddHdr(PT_GNU_EH_FRAME, In<ELFT>::EhFrameHdr->getParent()->getPhdrFlags()) 1500 ->add(In<ELFT>::EhFrameHdr->getParent()); 1501 1502 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes 1503 // the dynamic linker fill the segment with random data. 1504 if (OutputSection *Sec = findSectionInScript(".openbsd.randomdata")) 1505 AddHdr(PT_OPENBSD_RANDOMIZE, Sec->getPhdrFlags())->add(Sec); 1506 1507 // PT_GNU_STACK is a special section to tell the loader to make the 1508 // pages for the stack non-executable. If you really want an executable 1509 // stack, you can pass -z execstack, but that's not recommended for 1510 // security reasons. 1511 unsigned Perm; 1512 if (Config->ZExecstack) 1513 Perm = PF_R | PF_W | PF_X; 1514 else 1515 Perm = PF_R | PF_W; 1516 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize; 1517 1518 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable 1519 // is expected to perform W^X violations, such as calling mprotect(2) or 1520 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on 1521 // OpenBSD. 1522 if (Config->ZWxneeded) 1523 AddHdr(PT_OPENBSD_WXNEEDED, PF_X); 1524 1525 // Create one PT_NOTE per a group of contiguous .note sections. 1526 PhdrEntry *Note = nullptr; 1527 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1528 OutputSection *Sec = Cmd->Sec; 1529 if (Sec->Type == SHT_NOTE) { 1530 if (!Note || Cmd->LMAExpr) 1531 Note = AddHdr(PT_NOTE, PF_R); 1532 Note->add(Sec); 1533 } else { 1534 Note = nullptr; 1535 } 1536 } 1537 return Ret; 1538 } 1539 1540 template <class ELFT> 1541 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry> &Phdrs) { 1542 if (Config->EMachine != EM_ARM) 1543 return; 1544 auto I = llvm::find_if(OutputSectionCommands, [](OutputSectionCommand *Cmd) { 1545 return Cmd->Sec->Type == SHT_ARM_EXIDX; 1546 }); 1547 if (I == OutputSectionCommands.end()) 1548 return; 1549 1550 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME 1551 PhdrEntry ARMExidx(PT_ARM_EXIDX, PF_R); 1552 ARMExidx.add((*I)->Sec); 1553 Phdrs.push_back(ARMExidx); 1554 } 1555 1556 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the 1557 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic 1558 // linker can set the permissions. 1559 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() { 1560 auto PageAlign = [](OutputSection *Sec) { 1561 OutputSectionCommand *Cmd = Script->getCmd(Sec); 1562 if (Cmd && !Cmd->AddrExpr) 1563 Cmd->AddrExpr = [=] { 1564 return alignTo(Script->getDot(), Config->MaxPageSize); 1565 }; 1566 }; 1567 1568 for (const PhdrEntry &P : Phdrs) 1569 if (P.p_type == PT_LOAD && P.First) 1570 PageAlign(P.First); 1571 1572 for (const PhdrEntry &P : Phdrs) { 1573 if (P.p_type != PT_GNU_RELRO) 1574 continue; 1575 if (P.First) 1576 PageAlign(P.First); 1577 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we 1578 // have to align it to a page. 1579 auto End = OutputSectionCommands.end(); 1580 auto I = 1581 std::find(OutputSectionCommands.begin(), End, Script->getCmd(P.Last)); 1582 if (I == End || (I + 1) == End) 1583 continue; 1584 OutputSection *Sec = (*(I + 1))->Sec; 1585 if (needsPtLoad(Sec)) 1586 PageAlign(Sec); 1587 } 1588 } 1589 1590 // Adjusts the file alignment for a given output section and returns 1591 // its new file offset. The file offset must be the same with its 1592 // virtual address (modulo the page size) so that the loader can load 1593 // executables without any address adjustment. 1594 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Sec) { 1595 OutputSection *First = Sec->FirstInPtLoad; 1596 // If the section is not in a PT_LOAD, we just have to align it. 1597 if (!First) 1598 return alignTo(Off, Sec->Alignment); 1599 1600 // The first section in a PT_LOAD has to have congruent offset and address 1601 // module the page size. 1602 if (Sec == First) 1603 return alignTo(Off, Config->MaxPageSize, Sec->Addr); 1604 1605 // If two sections share the same PT_LOAD the file offset is calculated 1606 // using this formula: Off2 = Off1 + (VA2 - VA1). 1607 return First->Offset + Sec->Addr - First->Addr; 1608 } 1609 1610 static uint64_t setOffset(OutputSection *Sec, uint64_t Off) { 1611 if (Sec->Type == SHT_NOBITS) { 1612 Sec->Offset = Off; 1613 return Off; 1614 } 1615 1616 Off = getFileAlignment(Off, Sec); 1617 Sec->Offset = Off; 1618 return Off + Sec->Size; 1619 } 1620 1621 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() { 1622 uint64_t Off = 0; 1623 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1624 OutputSection *Sec = Cmd->Sec; 1625 if (Sec->Flags & SHF_ALLOC) 1626 Off = setOffset(Sec, Off); 1627 } 1628 FileSize = alignTo(Off, Config->Wordsize); 1629 } 1630 1631 // Assign file offsets to output sections. 1632 template <class ELFT> void Writer<ELFT>::assignFileOffsets() { 1633 uint64_t Off = 0; 1634 Off = setOffset(Out::ElfHeader, Off); 1635 Off = setOffset(Out::ProgramHeaders, Off); 1636 1637 for (OutputSectionCommand *Cmd : OutputSectionCommands) 1638 Off = setOffset(Cmd->Sec, Off); 1639 1640 SectionHeaderOff = alignTo(Off, Config->Wordsize); 1641 FileSize = 1642 SectionHeaderOff + (OutputSectionCommands.size() + 1) * sizeof(Elf_Shdr); 1643 } 1644 1645 // Finalize the program headers. We call this function after we assign 1646 // file offsets and VAs to all sections. 1647 template <class ELFT> void Writer<ELFT>::setPhdrs() { 1648 for (PhdrEntry &P : Phdrs) { 1649 OutputSection *First = P.First; 1650 OutputSection *Last = P.Last; 1651 if (First) { 1652 P.p_filesz = Last->Offset - First->Offset; 1653 if (Last->Type != SHT_NOBITS) 1654 P.p_filesz += Last->Size; 1655 P.p_memsz = Last->Addr + Last->Size - First->Addr; 1656 P.p_offset = First->Offset; 1657 P.p_vaddr = First->Addr; 1658 if (!P.HasLMA) 1659 P.p_paddr = First->getLMA(); 1660 } 1661 if (P.p_type == PT_LOAD) 1662 P.p_align = Config->MaxPageSize; 1663 else if (P.p_type == PT_GNU_RELRO) { 1664 P.p_align = 1; 1665 // The glibc dynamic loader rounds the size down, so we need to round up 1666 // to protect the last page. This is a no-op on FreeBSD which always 1667 // rounds up. 1668 P.p_memsz = alignTo(P.p_memsz, Target->PageSize); 1669 } 1670 1671 // The TLS pointer goes after PT_TLS. At least glibc will align it, 1672 // so round up the size to make sure the offsets are correct. 1673 if (P.p_type == PT_TLS) { 1674 Out::TlsPhdr = &P; 1675 if (P.p_memsz) 1676 P.p_memsz = alignTo(P.p_memsz, P.p_align); 1677 } 1678 } 1679 } 1680 1681 // The entry point address is chosen in the following ways. 1682 // 1683 // 1. the '-e' entry command-line option; 1684 // 2. the ENTRY(symbol) command in a linker control script; 1685 // 3. the value of the symbol start, if present; 1686 // 4. the address of the first byte of the .text section, if present; 1687 // 5. the address 0. 1688 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() { 1689 // Case 1, 2 or 3. As a special case, if the symbol is actually 1690 // a number, we'll use that number as an address. 1691 if (SymbolBody *B = Symtab<ELFT>::X->find(Config->Entry)) 1692 return B->getVA(); 1693 uint64_t Addr; 1694 if (to_integer(Config->Entry, Addr)) 1695 return Addr; 1696 1697 // Case 4 1698 if (OutputSection *Sec = findSectionInScript(".text")) { 1699 if (Config->WarnMissingEntry) 1700 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" + 1701 utohexstr(Sec->Addr)); 1702 return Sec->Addr; 1703 } 1704 1705 // Case 5 1706 if (Config->WarnMissingEntry) 1707 warn("cannot find entry symbol " + Config->Entry + 1708 "; not setting start address"); 1709 return 0; 1710 } 1711 1712 static uint16_t getELFType() { 1713 if (Config->Pic) 1714 return ET_DYN; 1715 if (Config->Relocatable) 1716 return ET_REL; 1717 return ET_EXEC; 1718 } 1719 1720 // This function is called after we have assigned address and size 1721 // to each section. This function fixes some predefined 1722 // symbol values that depend on section address and size. 1723 template <class ELFT> void Writer<ELFT>::fixPredefinedSymbols() { 1724 // _etext is the first location after the last read-only loadable segment. 1725 // _edata is the first location after the last read-write loadable segment. 1726 // _end is the first location after the uninitialized data region. 1727 PhdrEntry *Last = nullptr; 1728 PhdrEntry *LastRO = nullptr; 1729 PhdrEntry *LastRW = nullptr; 1730 for (PhdrEntry &P : Phdrs) { 1731 if (P.p_type != PT_LOAD) 1732 continue; 1733 Last = &P; 1734 if (P.p_flags & PF_W) 1735 LastRW = &P; 1736 else 1737 LastRO = &P; 1738 } 1739 1740 auto Set = [](DefinedRegular *S, OutputSection *Sec, uint64_t Value) { 1741 if (S) { 1742 S->Section = Sec; 1743 S->Value = Value; 1744 } 1745 }; 1746 1747 if (Last) { 1748 Set(ElfSym::End1, Last->First, Last->p_memsz); 1749 Set(ElfSym::End2, Last->First, Last->p_memsz); 1750 } 1751 if (LastRO) { 1752 Set(ElfSym::Etext1, LastRO->First, LastRO->p_filesz); 1753 Set(ElfSym::Etext2, LastRO->First, LastRO->p_filesz); 1754 } 1755 if (LastRW) { 1756 Set(ElfSym::Edata1, LastRW->First, LastRW->p_filesz); 1757 Set(ElfSym::Edata2, LastRW->First, LastRW->p_filesz); 1758 } 1759 1760 if (ElfSym::Bss) 1761 ElfSym::Bss->Section = findSectionInScript(".bss"); 1762 1763 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should 1764 // be equal to the _gp symbol's value. 1765 if (Config->EMachine == EM_MIPS && !ElfSym::MipsGp->Value) { 1766 // Find GP-relative section with the lowest address 1767 // and use this address to calculate default _gp value. 1768 for (const OutputSectionCommand *Cmd : OutputSectionCommands) { 1769 OutputSection *OS = Cmd->Sec; 1770 if (OS->Flags & SHF_MIPS_GPREL) { 1771 ElfSym::MipsGp->Value = OS->Addr + 0x7ff0; 1772 break; 1773 } 1774 } 1775 } 1776 } 1777 1778 template <class ELFT> void Writer<ELFT>::writeHeader() { 1779 uint8_t *Buf = Buffer->getBufferStart(); 1780 memcpy(Buf, "\177ELF", 4); 1781 1782 // Write the ELF header. 1783 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf); 1784 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32; 1785 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB; 1786 EHdr->e_ident[EI_VERSION] = EV_CURRENT; 1787 EHdr->e_ident[EI_OSABI] = Config->OSABI; 1788 EHdr->e_type = getELFType(); 1789 EHdr->e_machine = Config->EMachine; 1790 EHdr->e_version = EV_CURRENT; 1791 EHdr->e_entry = getEntryAddr(); 1792 EHdr->e_shoff = SectionHeaderOff; 1793 EHdr->e_ehsize = sizeof(Elf_Ehdr); 1794 EHdr->e_phnum = Phdrs.size(); 1795 EHdr->e_shentsize = sizeof(Elf_Shdr); 1796 EHdr->e_shnum = OutputSectionCommands.size() + 1; 1797 EHdr->e_shstrndx = InX::ShStrTab->getParent()->SectionIndex; 1798 1799 if (Config->EMachine == EM_ARM) 1800 // We don't currently use any features incompatible with EF_ARM_EABI_VER5, 1801 // but we don't have any firm guarantees of conformance. Linux AArch64 1802 // kernels (as of 2016) require an EABI version to be set. 1803 EHdr->e_flags = EF_ARM_EABI_VER5; 1804 else if (Config->EMachine == EM_MIPS) 1805 EHdr->e_flags = getMipsEFlags<ELFT>(); 1806 1807 if (!Config->Relocatable) { 1808 EHdr->e_phoff = sizeof(Elf_Ehdr); 1809 EHdr->e_phentsize = sizeof(Elf_Phdr); 1810 } 1811 1812 // Write the program header table. 1813 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff); 1814 for (PhdrEntry &P : Phdrs) { 1815 HBuf->p_type = P.p_type; 1816 HBuf->p_flags = P.p_flags; 1817 HBuf->p_offset = P.p_offset; 1818 HBuf->p_vaddr = P.p_vaddr; 1819 HBuf->p_paddr = P.p_paddr; 1820 HBuf->p_filesz = P.p_filesz; 1821 HBuf->p_memsz = P.p_memsz; 1822 HBuf->p_align = P.p_align; 1823 ++HBuf; 1824 } 1825 1826 // Write the section header table. Note that the first table entry is null. 1827 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff); 1828 for (OutputSectionCommand *Cmd : OutputSectionCommands) 1829 Cmd->Sec->writeHeaderTo<ELFT>(++SHdrs); 1830 } 1831 1832 // Open a result file. 1833 template <class ELFT> void Writer<ELFT>::openFile() { 1834 if (!Config->Is64 && FileSize > UINT32_MAX) { 1835 error("output file too large: " + Twine(FileSize) + " bytes"); 1836 return; 1837 } 1838 1839 unlinkAsync(Config->OutputFile); 1840 ErrorOr<std::unique_ptr<FileOutputBuffer>> BufferOrErr = 1841 FileOutputBuffer::create(Config->OutputFile, FileSize, 1842 FileOutputBuffer::F_executable); 1843 1844 if (auto EC = BufferOrErr.getError()) 1845 error("failed to open " + Config->OutputFile + ": " + EC.message()); 1846 else 1847 Buffer = std::move(*BufferOrErr); 1848 } 1849 1850 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() { 1851 uint8_t *Buf = Buffer->getBufferStart(); 1852 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1853 OutputSection *Sec = Cmd->Sec; 1854 if (Sec->Flags & SHF_ALLOC) 1855 Cmd->writeTo<ELFT>(Buf + Sec->Offset); 1856 } 1857 } 1858 1859 // Write section contents to a mmap'ed file. 1860 template <class ELFT> void Writer<ELFT>::writeSections() { 1861 uint8_t *Buf = Buffer->getBufferStart(); 1862 1863 // PPC64 needs to process relocations in the .opd section 1864 // before processing relocations in code-containing sections. 1865 if (auto *OpdCmd = findSectionCommand(".opd")) { 1866 Out::Opd = OpdCmd->Sec; 1867 Out::OpdBuf = Buf + Out::Opd->Offset; 1868 OpdCmd->template writeTo<ELFT>(Buf + Out::Opd->Offset); 1869 } 1870 1871 OutputSection *EhFrameHdr = 1872 (In<ELFT>::EhFrameHdr && !In<ELFT>::EhFrameHdr->empty()) 1873 ? In<ELFT>::EhFrameHdr->getParent() 1874 : nullptr; 1875 1876 // In -r or -emit-relocs mode, write the relocation sections first as in 1877 // ELf_Rel targets we might find out that we need to modify the relocated 1878 // section while doing it. 1879 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1880 OutputSection *Sec = Cmd->Sec; 1881 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA) 1882 Cmd->writeTo<ELFT>(Buf + Sec->Offset); 1883 } 1884 1885 for (OutputSectionCommand *Cmd : OutputSectionCommands) { 1886 OutputSection *Sec = Cmd->Sec; 1887 if (Sec != Out::Opd && Sec != EhFrameHdr && Sec->Type != SHT_REL && 1888 Sec->Type != SHT_RELA) 1889 Cmd->writeTo<ELFT>(Buf + Sec->Offset); 1890 } 1891 1892 // The .eh_frame_hdr depends on .eh_frame section contents, therefore 1893 // it should be written after .eh_frame is written. 1894 if (EhFrameHdr) { 1895 OutputSectionCommand *Cmd = Script->getCmd(EhFrameHdr); 1896 Cmd->writeTo<ELFT>(Buf + EhFrameHdr->Offset); 1897 } 1898 } 1899 1900 template <class ELFT> void Writer<ELFT>::writeBuildId() { 1901 if (!InX::BuildId || !InX::BuildId->getParent()) 1902 return; 1903 1904 // Compute a hash of all sections of the output file. 1905 uint8_t *Start = Buffer->getBufferStart(); 1906 uint8_t *End = Start + FileSize; 1907 InX::BuildId->writeBuildId({Start, End}); 1908 } 1909 1910 template void elf::writeResult<ELF32LE>(); 1911 template void elf::writeResult<ELF32BE>(); 1912 template void elf::writeResult<ELF64LE>(); 1913 template void elf::writeResult<ELF64BE>(); 1914