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 <climits> 27 28 using namespace llvm; 29 using namespace llvm::ELF; 30 using namespace llvm::object; 31 using namespace llvm::support; 32 using namespace llvm::support::endian; 33 34 using namespace lld; 35 using namespace lld::elf; 36 37 namespace { 38 // The writer writes a SymbolTable result to a file. 39 template <class ELFT> class Writer { 40 public: 41 typedef typename ELFT::Shdr Elf_Shdr; 42 typedef typename ELFT::Ehdr Elf_Ehdr; 43 typedef typename ELFT::Phdr Elf_Phdr; 44 45 void run(); 46 47 private: 48 void createSyntheticSections(); 49 void copyLocalSymbols(); 50 void addSectionSymbols(); 51 void addReservedSymbols(); 52 void createSections(); 53 void forEachRelSec(std::function<void(InputSectionBase &)> Fn); 54 void sortSections(); 55 void finalizeSections(); 56 void addPredefinedSections(); 57 void setReservedSymbolSections(); 58 59 std::vector<PhdrEntry *> createPhdrs(); 60 void removeEmptyPTLoad(); 61 void addPtArmExid(std::vector<PhdrEntry *> &Phdrs); 62 void assignFileOffsets(); 63 void assignFileOffsetsBinary(); 64 void setPhdrs(); 65 void fixSectionAlignments(); 66 void openFile(); 67 void writeTrapInstr(); 68 void writeHeader(); 69 void writeSections(); 70 void writeSectionsBinary(); 71 void writeBuildId(); 72 73 std::unique_ptr<FileOutputBuffer> Buffer; 74 75 OutputSectionFactory Factory; 76 77 void addRelIpltSymbols(); 78 void addStartEndSymbols(); 79 void addStartStopSymbols(OutputSection *Sec); 80 uint64_t getEntryAddr(); 81 OutputSection *findSection(StringRef Name); 82 83 std::vector<PhdrEntry *> Phdrs; 84 85 uint64_t FileSize; 86 uint64_t SectionHeaderOff; 87 88 bool HasGotBaseSym = false; 89 }; 90 } // anonymous namespace 91 92 StringRef elf::getOutputSectionName(StringRef Name) { 93 // ".zdebug_" is a prefix for ZLIB-compressed sections. 94 // Because we decompressed input sections, we want to remove 'z'. 95 if (Name.startswith(".zdebug_")) 96 return Saver.save("." + Name.substr(2)); 97 98 if (Config->Relocatable) 99 return Name; 100 101 for (StringRef V : 102 {".text.", ".rodata.", ".data.rel.ro.", ".data.", ".bss.rel.ro.", 103 ".bss.", ".init_array.", ".fini_array.", ".ctors.", ".dtors.", ".tbss.", 104 ".gcc_except_table.", ".tdata.", ".ARM.exidx.", ".ARM.extab."}) { 105 StringRef Prefix = V.drop_back(); 106 if (Name.startswith(V) || Name == Prefix) 107 return Prefix; 108 } 109 110 // CommonSection is identified as "COMMON" in linker scripts. 111 // By default, it should go to .bss section. 112 if (Name == "COMMON") 113 return ".bss"; 114 115 return Name; 116 } 117 118 template <class ELFT> static bool needsInterpSection() { 119 return !SharedFile<ELFT>::Instances.empty() && 120 !Config->DynamicLinker.empty() && !Script->ignoreInterpSection(); 121 } 122 123 template <class ELFT> void elf::writeResult() { Writer<ELFT>().run(); } 124 125 template <class ELFT> void Writer<ELFT>::removeEmptyPTLoad() { 126 llvm::erase_if(Phdrs, [&](const PhdrEntry *P) { 127 if (P->p_type != PT_LOAD) 128 return false; 129 if (!P->FirstSec) 130 return true; 131 uint64_t Size = P->LastSec->Addr + P->LastSec->Size - P->FirstSec->Addr; 132 return Size == 0; 133 }); 134 } 135 136 template <class ELFT> static void combineEhFrameSections() { 137 for (InputSectionBase *&S : InputSections) { 138 EhInputSection *ES = dyn_cast<EhInputSection>(S); 139 if (!ES || !ES->Live) 140 continue; 141 142 In<ELFT>::EhFrame->addSection(ES); 143 S = nullptr; 144 } 145 146 std::vector<InputSectionBase *> &V = InputSections; 147 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end()); 148 } 149 150 // The main function of the writer. 151 template <class ELFT> void Writer<ELFT>::run() { 152 // Create linker-synthesized sections such as .got or .plt. 153 // Such sections are of type input section. 154 createSyntheticSections(); 155 156 if (!Config->Relocatable) 157 combineEhFrameSections<ELFT>(); 158 159 // We need to create some reserved symbols such as _end. Create them. 160 if (!Config->Relocatable) 161 addReservedSymbols(); 162 163 // Create output sections. 164 if (Script->Opt.HasSections) { 165 // If linker script contains SECTIONS commands, let it create sections. 166 Script->processCommands(Factory); 167 168 // Linker scripts may have left some input sections unassigned. 169 // Assign such sections using the default rule. 170 Script->addOrphanSections(Factory); 171 } else { 172 // If linker script does not contain SECTIONS commands, create 173 // output sections by default rules. We still need to give the 174 // linker script a chance to run, because it might contain 175 // non-SECTIONS commands such as ASSERT. 176 Script->processCommands(Factory); 177 createSections(); 178 } 179 180 if (Config->Discard != DiscardPolicy::All) 181 copyLocalSymbols(); 182 183 if (Config->CopyRelocs) 184 addSectionSymbols(); 185 186 // Now that we have a complete set of output sections. This function 187 // completes section contents. For example, we need to add strings 188 // to the string table, and add entries to .got and .plt. 189 // finalizeSections does that. 190 finalizeSections(); 191 if (ErrorCount) 192 return; 193 194 // If -compressed-debug-sections is specified, we need to compress 195 // .debug_* sections. Do it right now because it changes the size of 196 // output sections. 197 parallelForEach(OutputSections, 198 [](OutputSection *Sec) { Sec->maybeCompress<ELFT>(); }); 199 200 Script->assignAddresses(); 201 Script->allocateHeaders(Phdrs); 202 203 // Remove empty PT_LOAD to avoid causing the dynamic linker to try to mmap a 204 // 0 sized region. This has to be done late since only after assignAddresses 205 // we know the size of the sections. 206 removeEmptyPTLoad(); 207 208 if (!Config->OFormatBinary) 209 assignFileOffsets(); 210 else 211 assignFileOffsetsBinary(); 212 213 setPhdrs(); 214 215 if (Config->Relocatable) { 216 for (OutputSection *Sec : OutputSections) 217 Sec->Addr = 0; 218 } 219 220 // It does not make sense try to open the file if we have error already. 221 if (ErrorCount) 222 return; 223 // Write the result down to a file. 224 openFile(); 225 if (ErrorCount) 226 return; 227 228 if (!Config->OFormatBinary) { 229 writeTrapInstr(); 230 writeHeader(); 231 writeSections(); 232 } else { 233 writeSectionsBinary(); 234 } 235 236 // Backfill .note.gnu.build-id section content. This is done at last 237 // because the content is usually a hash value of the entire output file. 238 writeBuildId(); 239 if (ErrorCount) 240 return; 241 242 // Handle -Map option. 243 writeMapFile<ELFT>(); 244 if (ErrorCount) 245 return; 246 247 if (auto EC = Buffer->commit()) 248 error("failed to write to the output file: " + EC.message()); 249 250 // Flush the output streams and exit immediately. A full shutdown 251 // is a good test that we are keeping track of all allocated memory, 252 // but actually freeing it is a waste of time in a regular linker run. 253 if (Config->ExitEarly) 254 exitLld(0); 255 } 256 257 // Initialize Out members. 258 template <class ELFT> void Writer<ELFT>::createSyntheticSections() { 259 // Initialize all pointers with NULL. This is needed because 260 // you can call lld::elf::main more than once as a library. 261 memset(&Out::First, 0, sizeof(Out)); 262 263 auto Add = [](InputSectionBase *Sec) { InputSections.push_back(Sec); }; 264 265 InX::DynStrTab = make<StringTableSection>(".dynstr", true); 266 InX::Dynamic = make<DynamicSection<ELFT>>(); 267 In<ELFT>::RelaDyn = make<RelocationSection<ELFT>>( 268 Config->IsRela ? ".rela.dyn" : ".rel.dyn", Config->ZCombreloc); 269 InX::ShStrTab = make<StringTableSection>(".shstrtab", false); 270 271 Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC); 272 Out::ElfHeader->Size = sizeof(Elf_Ehdr); 273 Out::ProgramHeaders = make<OutputSection>("", 0, SHF_ALLOC); 274 Out::ProgramHeaders->updateAlignment(Config->Wordsize); 275 276 if (needsInterpSection<ELFT>()) { 277 InX::Interp = createInterpSection(); 278 Add(InX::Interp); 279 } else { 280 InX::Interp = nullptr; 281 } 282 283 if (Config->Strip != StripPolicy::All) { 284 InX::StrTab = make<StringTableSection>(".strtab", false); 285 InX::SymTab = make<SymbolTableSection<ELFT>>(*InX::StrTab); 286 } 287 288 if (Config->BuildId != BuildIdKind::None) { 289 InX::BuildId = make<BuildIdSection>(); 290 Add(InX::BuildId); 291 } 292 293 for (InputSection *S : createCommonSections()) 294 Add(S); 295 296 InX::Bss = make<BssSection>(".bss"); 297 Add(InX::Bss); 298 InX::BssRelRo = make<BssSection>(".bss.rel.ro"); 299 Add(InX::BssRelRo); 300 301 // Add MIPS-specific sections. 302 bool HasDynSymTab = !SharedFile<ELFT>::Instances.empty() || Config->Pic || 303 Config->ExportDynamic; 304 if (Config->EMachine == EM_MIPS) { 305 if (!Config->Shared && HasDynSymTab) { 306 InX::MipsRldMap = make<MipsRldMapSection>(); 307 Add(InX::MipsRldMap); 308 } 309 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create()) 310 Add(Sec); 311 if (auto *Sec = MipsOptionsSection<ELFT>::create()) 312 Add(Sec); 313 if (auto *Sec = MipsReginfoSection<ELFT>::create()) 314 Add(Sec); 315 } 316 317 if (HasDynSymTab) { 318 InX::DynSymTab = make<SymbolTableSection<ELFT>>(*InX::DynStrTab); 319 Add(InX::DynSymTab); 320 321 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>(); 322 Add(In<ELFT>::VerSym); 323 324 if (!Config->VersionDefinitions.empty()) { 325 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>(); 326 Add(In<ELFT>::VerDef); 327 } 328 329 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>(); 330 Add(In<ELFT>::VerNeed); 331 332 if (Config->GnuHash) { 333 InX::GnuHashTab = make<GnuHashTableSection>(); 334 Add(InX::GnuHashTab); 335 } 336 337 if (Config->SysvHash) { 338 In<ELFT>::HashTab = make<HashTableSection<ELFT>>(); 339 Add(In<ELFT>::HashTab); 340 } 341 342 Add(InX::Dynamic); 343 Add(InX::DynStrTab); 344 Add(In<ELFT>::RelaDyn); 345 } 346 347 // Add .got. MIPS' .got is so different from the other archs, 348 // it has its own class. 349 if (Config->EMachine == EM_MIPS) { 350 InX::MipsGot = make<MipsGotSection>(); 351 Add(InX::MipsGot); 352 } else { 353 InX::Got = make<GotSection>(); 354 Add(InX::Got); 355 } 356 357 InX::GotPlt = make<GotPltSection>(); 358 Add(InX::GotPlt); 359 InX::IgotPlt = make<IgotPltSection>(); 360 Add(InX::IgotPlt); 361 362 if (Config->GdbIndex) { 363 InX::GdbIndex = createGdbIndex<ELFT>(); 364 Add(InX::GdbIndex); 365 } 366 367 // We always need to add rel[a].plt to output if it has entries. 368 // Even for static linking it can contain R_[*]_IRELATIVE relocations. 369 In<ELFT>::RelaPlt = make<RelocationSection<ELFT>>( 370 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/); 371 Add(In<ELFT>::RelaPlt); 372 373 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure 374 // that the IRelative relocations are processed last by the dynamic loader 375 In<ELFT>::RelaIplt = make<RelocationSection<ELFT>>( 376 (Config->EMachine == EM_ARM) ? ".rel.dyn" : In<ELFT>::RelaPlt->Name, 377 false /*Sort*/); 378 Add(In<ELFT>::RelaIplt); 379 380 InX::Plt = make<PltSection>(Target->PltHeaderSize); 381 Add(InX::Plt); 382 InX::Iplt = make<PltSection>(0); 383 Add(InX::Iplt); 384 385 if (!Config->Relocatable) { 386 if (Config->EhFrameHdr) { 387 In<ELFT>::EhFrameHdr = make<EhFrameHeader<ELFT>>(); 388 Add(In<ELFT>::EhFrameHdr); 389 } 390 In<ELFT>::EhFrame = make<EhFrameSection<ELFT>>(); 391 Add(In<ELFT>::EhFrame); 392 } 393 394 if (InX::SymTab) 395 Add(InX::SymTab); 396 Add(InX::ShStrTab); 397 if (InX::StrTab) 398 Add(InX::StrTab); 399 } 400 401 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName, 402 const SymbolBody &B) { 403 if (B.isFile() || B.isSection()) 404 return false; 405 406 // If sym references a section in a discarded group, don't keep it. 407 if (Sec == &InputSection::Discarded) 408 return false; 409 410 if (Config->Discard == DiscardPolicy::None) 411 return true; 412 413 // In ELF assembly .L symbols are normally discarded by the assembler. 414 // If the assembler fails to do so, the linker discards them if 415 // * --discard-locals is used. 416 // * The symbol is in a SHF_MERGE section, which is normally the reason for 417 // the assembler keeping the .L symbol. 418 if (!SymName.startswith(".L") && !SymName.empty()) 419 return true; 420 421 if (Config->Discard == DiscardPolicy::Locals) 422 return false; 423 424 return !Sec || !(Sec->Flags & SHF_MERGE); 425 } 426 427 static bool includeInSymtab(const SymbolBody &B) { 428 if (!B.isLocal() && !B.symbol()->IsUsedInRegularObj) 429 return false; 430 431 if (auto *D = dyn_cast<DefinedRegular>(&B)) { 432 // Always include absolute symbols. 433 SectionBase *Sec = D->Section; 434 if (!Sec) 435 return true; 436 if (auto *IS = dyn_cast<InputSectionBase>(Sec)) { 437 Sec = IS->Repl; 438 IS = cast<InputSectionBase>(Sec); 439 // Exclude symbols pointing to garbage-collected sections. 440 if (!IS->Live) 441 return false; 442 } 443 if (auto *S = dyn_cast<MergeInputSection>(Sec)) 444 if (!S->getSectionPiece(D->Value)->Live) 445 return false; 446 return true; 447 } 448 449 if (auto *Sym = dyn_cast<DefinedCommon>(&B)) 450 return Sym->Live; 451 return true; 452 } 453 454 // Local symbols are not in the linker's symbol table. This function scans 455 // each object file's symbol table to copy local symbols to the output. 456 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() { 457 if (!InX::SymTab) 458 return; 459 for (ObjFile<ELFT> *F : ObjFile<ELFT>::Instances) { 460 for (SymbolBody *B : F->getLocalSymbols()) { 461 if (!B->IsLocal) 462 fatal(toString(F) + 463 ": broken object: getLocalSymbols returns a non-local symbol"); 464 auto *DR = dyn_cast<DefinedRegular>(B); 465 466 // No reason to keep local undefined symbol in symtab. 467 if (!DR) 468 continue; 469 if (!includeInSymtab(*B)) 470 continue; 471 472 SectionBase *Sec = DR->Section; 473 if (!shouldKeepInSymtab(Sec, B->getName(), *B)) 474 continue; 475 InX::SymTab->addSymbol(B); 476 } 477 } 478 } 479 480 template <class ELFT> void Writer<ELFT>::addSectionSymbols() { 481 // Create one STT_SECTION symbol for each output section we might 482 // have a relocation with. 483 for (BaseCommand *Base : Script->Opt.Commands) { 484 auto *Sec = dyn_cast<OutputSection>(Base); 485 if (!Sec) 486 continue; 487 auto I = llvm::find_if(Sec->Commands, [](BaseCommand *Base) { 488 if (auto *ISD = dyn_cast<InputSectionDescription>(Base)) 489 return !ISD->Sections.empty(); 490 return false; 491 }); 492 if (I == Sec->Commands.end()) 493 continue; 494 InputSection *IS = cast<InputSectionDescription>(*I)->Sections[0]; 495 if (isa<SyntheticSection>(IS) || IS->Type == SHT_REL || 496 IS->Type == SHT_RELA) 497 continue; 498 499 auto *Sym = 500 make<DefinedRegular>("", /*IsLocal=*/true, /*StOther=*/0, STT_SECTION, 501 /*Value=*/0, /*Size=*/0, IS); 502 InX::SymTab->addSymbol(Sym); 503 } 504 } 505 506 // Today's loaders have a feature to make segments read-only after 507 // processing dynamic relocations to enhance security. PT_GNU_RELRO 508 // is defined for that. 509 // 510 // This function returns true if a section needs to be put into a 511 // PT_GNU_RELRO segment. 512 static bool isRelroSection(const OutputSection *Sec) { 513 if (!Config->ZRelro) 514 return false; 515 516 uint64_t Flags = Sec->Flags; 517 518 // Non-allocatable or non-writable sections don't need RELRO because 519 // they are not writable or not even mapped to memory in the first place. 520 // RELRO is for sections that are essentially read-only but need to 521 // be writable only at process startup to allow dynamic linker to 522 // apply relocations. 523 if (!(Flags & SHF_ALLOC) || !(Flags & SHF_WRITE)) 524 return false; 525 526 // Once initialized, TLS data segments are used as data templates 527 // for a thread-local storage. For each new thread, runtime 528 // allocates memory for a TLS and copy templates there. No thread 529 // are supposed to use templates directly. Thus, it can be in RELRO. 530 if (Flags & SHF_TLS) 531 return true; 532 533 // .init_array, .preinit_array and .fini_array contain pointers to 534 // functions that are executed on process startup or exit. These 535 // pointers are set by the static linker, and they are not expected 536 // to change at runtime. But if you are an attacker, you could do 537 // interesting things by manipulating pointers in .fini_array, for 538 // example. So they are put into RELRO. 539 uint32_t Type = Sec->Type; 540 if (Type == SHT_INIT_ARRAY || Type == SHT_FINI_ARRAY || 541 Type == SHT_PREINIT_ARRAY) 542 return true; 543 544 // .got contains pointers to external symbols. They are resolved by 545 // the dynamic linker when a module is loaded into memory, and after 546 // that they are not expected to change. So, it can be in RELRO. 547 if (InX::Got && Sec == InX::Got->getParent()) 548 return true; 549 550 // .got.plt contains pointers to external function symbols. They are 551 // by default resolved lazily, so we usually cannot put it into RELRO. 552 // However, if "-z now" is given, the lazy symbol resolution is 553 // disabled, which enables us to put it into RELRO. 554 if (Sec == InX::GotPlt->getParent()) 555 return Config->ZNow; 556 557 // .dynamic section contains data for the dynamic linker, and 558 // there's no need to write to it at runtime, so it's better to put 559 // it into RELRO. 560 if (Sec == InX::Dynamic->getParent()) 561 return true; 562 563 // .bss.rel.ro is used for copy relocations for read-only symbols. 564 // Since the dynamic linker needs to process copy relocations, the 565 // section cannot be read-only, but once initialized, they shouldn't 566 // change. 567 if (Sec == InX::BssRelRo->getParent()) 568 return true; 569 570 // Sections with some special names are put into RELRO. This is a 571 // bit unfortunate because section names shouldn't be significant in 572 // ELF in spirit. But in reality many linker features depend on 573 // magic section names. 574 StringRef S = Sec->Name; 575 return S == ".data.rel.ro" || S == ".ctors" || S == ".dtors" || S == ".jcr" || 576 S == ".eh_frame" || S == ".openbsd.randomdata"; 577 } 578 579 // We compute a rank for each section. The rank indicates where the 580 // section should be placed in the file. Instead of using simple 581 // numbers (0,1,2...), we use a series of flags. One for each decision 582 // point when placing the section. 583 // Using flags has two key properties: 584 // * It is easy to check if a give branch was taken. 585 // * It is easy two see how similar two ranks are (see getRankProximity). 586 enum RankFlags { 587 RF_NOT_ADDR_SET = 1 << 16, 588 RF_NOT_INTERP = 1 << 15, 589 RF_NOT_ALLOC = 1 << 14, 590 RF_WRITE = 1 << 13, 591 RF_EXEC_WRITE = 1 << 12, 592 RF_EXEC = 1 << 11, 593 RF_NON_TLS_BSS = 1 << 10, 594 RF_NON_TLS_BSS_RO = 1 << 9, 595 RF_NOT_TLS = 1 << 8, 596 RF_BSS = 1 << 7, 597 RF_PPC_NOT_TOCBSS = 1 << 6, 598 RF_PPC_OPD = 1 << 5, 599 RF_PPC_TOCL = 1 << 4, 600 RF_PPC_TOC = 1 << 3, 601 RF_PPC_BRANCH_LT = 1 << 2, 602 RF_MIPS_GPREL = 1 << 1, 603 RF_MIPS_NOT_GOT = 1 << 0 604 }; 605 606 static unsigned getSectionRank(const OutputSection *Sec) { 607 unsigned Rank = 0; 608 609 // We want to put section specified by -T option first, so we 610 // can start assigning VA starting from them later. 611 if (Config->SectionStartMap.count(Sec->Name)) 612 return Rank; 613 Rank |= RF_NOT_ADDR_SET; 614 615 // Put .interp first because some loaders want to see that section 616 // on the first page of the executable file when loaded into memory. 617 if (Sec->Name == ".interp") 618 return Rank; 619 Rank |= RF_NOT_INTERP; 620 621 // Allocatable sections go first to reduce the total PT_LOAD size and 622 // so debug info doesn't change addresses in actual code. 623 if (!(Sec->Flags & SHF_ALLOC)) 624 return Rank | RF_NOT_ALLOC; 625 626 // Sort sections based on their access permission in the following 627 // order: R, RX, RWX, RW. This order is based on the following 628 // considerations: 629 // * Read-only sections come first such that they go in the 630 // PT_LOAD covering the program headers at the start of the file. 631 // * Read-only, executable sections come next, unless the 632 // -no-rosegment option is used. 633 // * Writable, executable sections follow such that .plt on 634 // architectures where it needs to be writable will be placed 635 // between .text and .data. 636 // * Writable sections come last, such that .bss lands at the very 637 // end of the last PT_LOAD. 638 bool IsExec = Sec->Flags & SHF_EXECINSTR; 639 bool IsWrite = Sec->Flags & SHF_WRITE; 640 641 if (IsExec) { 642 if (IsWrite) 643 Rank |= RF_EXEC_WRITE; 644 else if (!Config->SingleRoRx) 645 Rank |= RF_EXEC; 646 } else { 647 if (IsWrite) 648 Rank |= RF_WRITE; 649 } 650 651 // If we got here we know that both A and B are in the same PT_LOAD. 652 653 bool IsTls = Sec->Flags & SHF_TLS; 654 bool IsNoBits = Sec->Type == SHT_NOBITS; 655 656 // The first requirement we have is to put (non-TLS) nobits sections last. The 657 // reason is that the only thing the dynamic linker will see about them is a 658 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the 659 // PT_LOAD, so that has to correspond to the nobits sections. 660 bool IsNonTlsNoBits = IsNoBits && !IsTls; 661 if (IsNonTlsNoBits) 662 Rank |= RF_NON_TLS_BSS; 663 664 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo 665 // sections after r/w ones, so that the RelRo sections are contiguous. 666 bool IsRelRo = isRelroSection(Sec); 667 if (IsNonTlsNoBits && !IsRelRo) 668 Rank |= RF_NON_TLS_BSS_RO; 669 if (!IsNonTlsNoBits && IsRelRo) 670 Rank |= RF_NON_TLS_BSS_RO; 671 672 // The TLS initialization block needs to be a single contiguous block in a R/W 673 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W 674 // sections. The TLS NOBITS sections are placed here as they don't take up 675 // virtual address space in the PT_LOAD. 676 if (!IsTls) 677 Rank |= RF_NOT_TLS; 678 679 // Within the TLS initialization block, the non-nobits sections need to appear 680 // first. 681 if (IsNoBits) 682 Rank |= RF_BSS; 683 684 // Some architectures have additional ordering restrictions for sections 685 // within the same PT_LOAD. 686 if (Config->EMachine == EM_PPC64) { 687 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections 688 // that we would like to make sure appear is a specific order to maximize 689 // their coverage by a single signed 16-bit offset from the TOC base 690 // pointer. Conversely, the special .tocbss section should be first among 691 // all SHT_NOBITS sections. This will put it next to the loaded special 692 // PPC64 sections (and, thus, within reach of the TOC base pointer). 693 StringRef Name = Sec->Name; 694 if (Name != ".tocbss") 695 Rank |= RF_PPC_NOT_TOCBSS; 696 697 if (Name == ".opd") 698 Rank |= RF_PPC_OPD; 699 700 if (Name == ".toc1") 701 Rank |= RF_PPC_TOCL; 702 703 if (Name == ".toc") 704 Rank |= RF_PPC_TOC; 705 706 if (Name == ".branch_lt") 707 Rank |= RF_PPC_BRANCH_LT; 708 } 709 if (Config->EMachine == EM_MIPS) { 710 // All sections with SHF_MIPS_GPREL flag should be grouped together 711 // because data in these sections is addressable with a gp relative address. 712 if (Sec->Flags & SHF_MIPS_GPREL) 713 Rank |= RF_MIPS_GPREL; 714 715 if (Sec->Name != ".got") 716 Rank |= RF_MIPS_NOT_GOT; 717 } 718 719 return Rank; 720 } 721 722 static bool compareSections(const BaseCommand *ACmd, const BaseCommand *BCmd) { 723 const OutputSection *A = cast<OutputSection>(ACmd); 724 const OutputSection *B = cast<OutputSection>(BCmd); 725 if (A->SortRank != B->SortRank) 726 return A->SortRank < B->SortRank; 727 if (!(A->SortRank & RF_NOT_ADDR_SET)) 728 return Config->SectionStartMap.lookup(A->Name) < 729 Config->SectionStartMap.lookup(B->Name); 730 return false; 731 } 732 733 void PhdrEntry::add(OutputSection *Sec) { 734 LastSec = Sec; 735 if (!FirstSec) 736 FirstSec = Sec; 737 p_align = std::max(p_align, Sec->Alignment); 738 if (p_type == PT_LOAD) 739 Sec->PtLoad = this; 740 } 741 742 template <class ELFT> 743 static Symbol *addRegular(StringRef Name, SectionBase *Sec, uint64_t Value, 744 uint8_t StOther = STV_HIDDEN, 745 uint8_t Binding = STB_WEAK) { 746 // The linker generated symbols are added as STB_WEAK to allow user defined 747 // ones to override them. 748 return Symtab->addRegular<ELFT>(Name, StOther, STT_NOTYPE, Value, 749 /*Size=*/0, Binding, Sec, 750 /*File=*/nullptr); 751 } 752 753 template <class ELFT> 754 static DefinedRegular * 755 addOptionalRegular(StringRef Name, SectionBase *Sec, uint64_t Val, 756 uint8_t StOther = STV_HIDDEN, uint8_t Binding = STB_GLOBAL) { 757 SymbolBody *S = Symtab->find(Name); 758 if (!S) 759 return nullptr; 760 if (S->isInCurrentDSO()) 761 return nullptr; 762 return cast<DefinedRegular>( 763 addRegular<ELFT>(Name, Sec, Val, StOther, Binding)->body()); 764 } 765 766 // The beginning and the ending of .rel[a].plt section are marked 767 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked 768 // executable. The runtime needs these symbols in order to resolve 769 // all IRELATIVE relocs on startup. For dynamic executables, we don't 770 // need these symbols, since IRELATIVE relocs are resolved through GOT 771 // and PLT. For details, see http://www.airs.com/blog/archives/403. 772 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() { 773 if (!Config->Static) 774 return; 775 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start"; 776 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, 0, STV_HIDDEN, STB_WEAK); 777 778 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end"; 779 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, -1, STV_HIDDEN, STB_WEAK); 780 } 781 782 // The linker is expected to define some symbols depending on 783 // the linking result. This function defines such symbols. 784 template <class ELFT> void Writer<ELFT>::addReservedSymbols() { 785 if (Config->EMachine == EM_MIPS) { 786 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer 787 // so that it points to an absolute address which by default is relative 788 // to GOT. Default offset is 0x7ff0. 789 // See "Global Data Symbols" in Chapter 6 in the following document: 790 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf 791 ElfSym::MipsGp = Symtab->addAbsolute<ELFT>("_gp", STV_HIDDEN, STB_LOCAL); 792 793 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between 794 // start of function and 'gp' pointer into GOT. 795 if (Symtab->find("_gp_disp")) 796 ElfSym::MipsGpDisp = 797 Symtab->addAbsolute<ELFT>("_gp_disp", STV_HIDDEN, STB_LOCAL); 798 799 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp' 800 // pointer. This symbol is used in the code generated by .cpload pseudo-op 801 // in case of using -mno-shared option. 802 // https://sourceware.org/ml/binutils/2004-12/msg00094.html 803 if (Symtab->find("__gnu_local_gp")) 804 ElfSym::MipsLocalGp = 805 Symtab->addAbsolute<ELFT>("__gnu_local_gp", STV_HIDDEN, STB_LOCAL); 806 } 807 808 // The _GLOBAL_OFFSET_TABLE_ symbol is defined by target convention to 809 // be at some offset from the base of the .got section, usually 0 or the end 810 // of the .got 811 InputSection *GotSection = InX::MipsGot ? cast<InputSection>(InX::MipsGot) 812 : cast<InputSection>(InX::Got); 813 ElfSym::GlobalOffsetTable = addOptionalRegular<ELFT>( 814 "_GLOBAL_OFFSET_TABLE_", GotSection, Target->GotBaseSymOff); 815 816 // __tls_get_addr is defined by the dynamic linker for dynamic ELFs. For 817 // static linking the linker is required to optimize away any references to 818 // __tls_get_addr, so it's not defined anywhere. Create a hidden definition 819 // to avoid the undefined symbol error. 820 if (!InX::DynSymTab) 821 Symtab->addIgnored<ELFT>("__tls_get_addr"); 822 823 // __ehdr_start is the location of ELF file headers. Note that we define 824 // this symbol unconditionally even when using a linker script, which 825 // differs from the behavior implemented by GNU linker which only define 826 // this symbol if ELF headers are in the memory mapped segment. 827 // __executable_start is not documented, but the expectation of at 828 // least the android libc is that it points to the elf header too. 829 // __dso_handle symbol is passed to cxa_finalize as a marker to identify 830 // each DSO. The address of the symbol doesn't matter as long as they are 831 // different in different DSOs, so we chose the start address of the DSO. 832 for (const char *Name : 833 {"__ehdr_start", "__executable_start", "__dso_handle"}) 834 addOptionalRegular<ELFT>(Name, Out::ElfHeader, 0, STV_HIDDEN); 835 836 // If linker script do layout we do not need to create any standart symbols. 837 if (Script->Opt.HasSections) 838 return; 839 840 auto Add = [](StringRef S, int64_t Pos = -1) { 841 return addOptionalRegular<ELFT>(S, Out::ElfHeader, Pos, STV_DEFAULT); 842 }; 843 844 ElfSym::Bss = Add("__bss_start", 0); 845 ElfSym::End1 = Add("end"); 846 ElfSym::End2 = Add("_end"); 847 ElfSym::Etext1 = Add("etext"); 848 ElfSym::Etext2 = Add("_etext"); 849 ElfSym::Edata1 = Add("edata"); 850 ElfSym::Edata2 = Add("_edata"); 851 } 852 853 // Sort input sections by section name suffixes for 854 // __attribute__((init_priority(N))). 855 static void sortInitFini(OutputSection *Cmd) { 856 if (Cmd) 857 Cmd->sortInitFini(); 858 } 859 860 // Sort input sections by the special rule for .ctors and .dtors. 861 static void sortCtorsDtors(OutputSection *Cmd) { 862 if (Cmd) 863 Cmd->sortCtorsDtors(); 864 } 865 866 // Sort input sections using the list provided by --symbol-ordering-file. 867 template <class ELFT> static void sortBySymbolsOrder() { 868 if (Config->SymbolOrderingFile.empty()) 869 return; 870 871 // Sort sections by priority. 872 DenseMap<SectionBase *, int> SectionOrder = buildSectionOrder<ELFT>(); 873 for (BaseCommand *Base : Script->Opt.Commands) 874 if (auto *Sec = dyn_cast<OutputSection>(Base)) 875 Sec->sort([&](InputSectionBase *S) { return SectionOrder.lookup(S); }); 876 } 877 878 template <class ELFT> 879 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) { 880 // Scan all relocations. Each relocation goes through a series 881 // of tests to determine if it needs special treatment, such as 882 // creating GOT, PLT, copy relocations, etc. 883 // Note that relocations for non-alloc sections are directly 884 // processed by InputSection::relocateNonAlloc. 885 for (InputSectionBase *IS : InputSections) 886 if (IS->Live && isa<InputSection>(IS) && (IS->Flags & SHF_ALLOC)) 887 Fn(*IS); 888 for (EhInputSection *ES : In<ELFT>::EhFrame->Sections) 889 Fn(*ES); 890 } 891 892 template <class ELFT> void Writer<ELFT>::createSections() { 893 std::vector<BaseCommand *> Old = Script->Opt.Commands; 894 Script->Opt.Commands.clear(); 895 for (InputSectionBase *IS : InputSections) 896 if (IS) 897 Factory.addInputSec(IS, getOutputSectionName(IS->Name)); 898 Script->Opt.Commands.insert(Script->Opt.Commands.end(), Old.begin(), 899 Old.end()); 900 901 Script->fabricateDefaultCommands(); 902 sortBySymbolsOrder<ELFT>(); 903 sortInitFini(findSection(".init_array")); 904 sortInitFini(findSection(".fini_array")); 905 sortCtorsDtors(findSection(".ctors")); 906 sortCtorsDtors(findSection(".dtors")); 907 } 908 909 // This function generates assignments for predefined symbols (e.g. _end or 910 // _etext) and inserts them into the commands sequence to be processed at the 911 // appropriate time. This ensures that the value is going to be correct by the 912 // time any references to these symbols are processed and is equivalent to 913 // defining these symbols explicitly in the linker script. 914 template <class ELFT> void Writer<ELFT>::setReservedSymbolSections() { 915 PhdrEntry *Last = nullptr; 916 PhdrEntry *LastRO = nullptr; 917 PhdrEntry *LastRW = nullptr; 918 919 for (PhdrEntry *P : Phdrs) { 920 if (P->p_type != PT_LOAD) 921 continue; 922 Last = P; 923 if (P->p_flags & PF_W) 924 LastRW = P; 925 else 926 LastRO = P; 927 } 928 929 // _end is the first location after the uninitialized data region. 930 if (Last) { 931 if (ElfSym::End1) 932 ElfSym::End1->Section = Last->LastSec; 933 if (ElfSym::End2) 934 ElfSym::End2->Section = Last->LastSec; 935 } 936 937 // _etext is the first location after the last read-only loadable segment. 938 if (LastRO) { 939 if (ElfSym::Etext1) 940 ElfSym::Etext1->Section = LastRO->LastSec; 941 if (ElfSym::Etext2) 942 ElfSym::Etext2->Section = LastRO->LastSec; 943 } 944 945 // _edata points to the end of the last non SHT_NOBITS section. 946 if (LastRW) { 947 size_t I = 0; 948 for (; I < OutputSections.size(); ++I) 949 if (OutputSections[I] == LastRW->FirstSec) 950 break; 951 952 for (; I < OutputSections.size(); ++I) { 953 if (OutputSections[I]->Type != SHT_NOBITS) 954 continue; 955 break; 956 } 957 if (ElfSym::Edata1) 958 ElfSym::Edata1->Section = OutputSections[I - 1]; 959 if (ElfSym::Edata2) 960 ElfSym::Edata2->Section = OutputSections[I - 1]; 961 } 962 963 if (ElfSym::Bss) 964 ElfSym::Bss->Section = findSection(".bss"); 965 966 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should 967 // be equal to the _gp symbol's value. 968 if (ElfSym::MipsGp && !ElfSym::MipsGp->Value) { 969 // Find GP-relative section with the lowest address 970 // and use this address to calculate default _gp value. 971 for (OutputSection *OS : OutputSections) { 972 if (OS->Flags & SHF_MIPS_GPREL) { 973 ElfSym::MipsGp->Section = OS; 974 ElfSym::MipsGp->Value = 0x7ff0; 975 break; 976 } 977 } 978 } 979 } 980 981 // We want to find how similar two ranks are. 982 // The more branches in getSectionRank that match, the more similar they are. 983 // Since each branch corresponds to a bit flag, we can just use 984 // countLeadingZeros. 985 static int getRankProximityAux(OutputSection *A, OutputSection *B) { 986 return countLeadingZeros(A->SortRank ^ B->SortRank); 987 } 988 989 static int getRankProximity(OutputSection *A, BaseCommand *B) { 990 if (auto *Sec = dyn_cast<OutputSection>(B)) 991 if (Sec->Live) 992 return getRankProximityAux(A, Sec); 993 return -1; 994 } 995 996 // When placing orphan sections, we want to place them after symbol assignments 997 // so that an orphan after 998 // begin_foo = .; 999 // foo : { *(foo) } 1000 // end_foo = .; 1001 // doesn't break the intended meaning of the begin/end symbols. 1002 // We don't want to go over sections since findOrphanPos is the 1003 // one in charge of deciding the order of the sections. 1004 // We don't want to go over changes to '.', since doing so in 1005 // rx_sec : { *(rx_sec) } 1006 // . = ALIGN(0x1000); 1007 // /* The RW PT_LOAD starts here*/ 1008 // rw_sec : { *(rw_sec) } 1009 // would mean that the RW PT_LOAD would become unaligned. 1010 static bool shouldSkip(BaseCommand *Cmd) { 1011 if (isa<OutputSection>(Cmd)) 1012 return false; 1013 if (auto *Assign = dyn_cast<SymbolAssignment>(Cmd)) 1014 return Assign->Name != "."; 1015 return true; 1016 } 1017 1018 // We want to place orphan sections so that they share as much 1019 // characteristics with their neighbors as possible. For example, if 1020 // both are rw, or both are tls. 1021 template <typename ELFT> 1022 static std::vector<BaseCommand *>::iterator 1023 findOrphanPos(std::vector<BaseCommand *>::iterator B, 1024 std::vector<BaseCommand *>::iterator E) { 1025 OutputSection *Sec = cast<OutputSection>(*E); 1026 1027 // Find the first element that has as close a rank as possible. 1028 auto I = std::max_element(B, E, [=](BaseCommand *A, BaseCommand *B) { 1029 return getRankProximity(Sec, A) < getRankProximity(Sec, B); 1030 }); 1031 if (I == E) 1032 return E; 1033 1034 // Consider all existing sections with the same proximity. 1035 int Proximity = getRankProximity(Sec, *I); 1036 for (; I != E; ++I) { 1037 auto *CurSec = dyn_cast<OutputSection>(*I); 1038 if (!CurSec || !CurSec->Live) 1039 continue; 1040 if (getRankProximity(Sec, CurSec) != Proximity || 1041 Sec->SortRank < CurSec->SortRank) 1042 break; 1043 } 1044 auto J = std::find_if( 1045 llvm::make_reverse_iterator(I), llvm::make_reverse_iterator(B), 1046 [](BaseCommand *Cmd) { return isa<OutputSection>(Cmd); }); 1047 I = J.base(); 1048 while (I != E && shouldSkip(*I)) 1049 ++I; 1050 return I; 1051 } 1052 1053 template <class ELFT> void Writer<ELFT>::sortSections() { 1054 if (Script->Opt.HasSections) 1055 Script->adjustSectionsBeforeSorting(); 1056 1057 // Don't sort if using -r. It is not necessary and we want to preserve the 1058 // relative order for SHF_LINK_ORDER sections. 1059 if (Config->Relocatable) 1060 return; 1061 1062 for (BaseCommand *Base : Script->Opt.Commands) 1063 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1064 Sec->SortRank = getSectionRank(Sec); 1065 1066 if (!Script->Opt.HasSections) { 1067 // We know that all the OutputSections are contiguous in 1068 // this case. 1069 auto E = Script->Opt.Commands.end(); 1070 auto I = Script->Opt.Commands.begin(); 1071 auto IsSection = [](BaseCommand *Base) { return isa<OutputSection>(Base); }; 1072 I = std::find_if(I, E, IsSection); 1073 E = std::find_if(llvm::make_reverse_iterator(E), 1074 llvm::make_reverse_iterator(I), IsSection) 1075 .base(); 1076 std::stable_sort(I, E, compareSections); 1077 return; 1078 } 1079 1080 // Orphan sections are sections present in the input files which are 1081 // not explicitly placed into the output file by the linker script. 1082 // 1083 // The sections in the linker script are already in the correct 1084 // order. We have to figuere out where to insert the orphan 1085 // sections. 1086 // 1087 // The order of the sections in the script is arbitrary and may not agree with 1088 // compareSections. This means that we cannot easily define a strict weak 1089 // ordering. To see why, consider a comparison of a section in the script and 1090 // one not in the script. We have a two simple options: 1091 // * Make them equivalent (a is not less than b, and b is not less than a). 1092 // The problem is then that equivalence has to be transitive and we can 1093 // have sections a, b and c with only b in a script and a less than c 1094 // which breaks this property. 1095 // * Use compareSectionsNonScript. Given that the script order doesn't have 1096 // to match, we can end up with sections a, b, c, d where b and c are in the 1097 // script and c is compareSectionsNonScript less than b. In which case d 1098 // can be equivalent to c, a to b and d < a. As a concrete example: 1099 // .a (rx) # not in script 1100 // .b (rx) # in script 1101 // .c (ro) # in script 1102 // .d (ro) # not in script 1103 // 1104 // The way we define an order then is: 1105 // * Sort only the orphan sections. They are in the end right now. 1106 // * Move each orphan section to its preferred position. We try 1107 // to put each section in the last position where it it can share 1108 // a PT_LOAD. 1109 // 1110 // There is some ambiguity as to where exactly a new entry should be 1111 // inserted, because Opt.Commands contains not only output section 1112 // commands but also other types of commands such as symbol assignment 1113 // expressions. There's no correct answer here due to the lack of the 1114 // formal specification of the linker script. We use heuristics to 1115 // determine whether a new output command should be added before or 1116 // after another commands. For the details, look at shouldSkip 1117 // function. 1118 1119 auto I = Script->Opt.Commands.begin(); 1120 auto E = Script->Opt.Commands.end(); 1121 auto NonScriptI = std::find_if(I, E, [](BaseCommand *Base) { 1122 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1123 return Sec->Live && Sec->SectionIndex == INT_MAX; 1124 return false; 1125 }); 1126 1127 // Sort the orphan sections. 1128 std::stable_sort(NonScriptI, E, compareSections); 1129 1130 // As a horrible special case, skip the first . assignment if it is before any 1131 // section. We do this because it is common to set a load address by starting 1132 // the script with ". = 0xabcd" and the expectation is that every section is 1133 // after that. 1134 auto FirstSectionOrDotAssignment = 1135 std::find_if(I, E, [](BaseCommand *Cmd) { return !shouldSkip(Cmd); }); 1136 if (FirstSectionOrDotAssignment != E && 1137 isa<SymbolAssignment>(**FirstSectionOrDotAssignment)) 1138 ++FirstSectionOrDotAssignment; 1139 I = FirstSectionOrDotAssignment; 1140 1141 while (NonScriptI != E) { 1142 auto Pos = findOrphanPos<ELFT>(I, NonScriptI); 1143 OutputSection *Orphan = cast<OutputSection>(*NonScriptI); 1144 1145 // As an optimization, find all sections with the same sort rank 1146 // and insert them with one rotate. 1147 unsigned Rank = Orphan->SortRank; 1148 auto End = std::find_if(NonScriptI + 1, E, [=](BaseCommand *Cmd) { 1149 return cast<OutputSection>(Cmd)->SortRank != Rank; 1150 }); 1151 std::rotate(Pos, NonScriptI, End); 1152 NonScriptI = End; 1153 } 1154 1155 Script->adjustSectionsAfterSorting(); 1156 } 1157 1158 static void applySynthetic(const std::vector<SyntheticSection *> &Sections, 1159 std::function<void(SyntheticSection *)> Fn) { 1160 for (SyntheticSection *SS : Sections) 1161 if (SS && SS->getParent() && !SS->empty()) 1162 Fn(SS); 1163 } 1164 1165 // We need to add input synthetic sections early in createSyntheticSections() 1166 // to make them visible from linkescript side. But not all sections are always 1167 // required to be in output. For example we don't need dynamic section content 1168 // sometimes. This function filters out such unused sections from the output. 1169 static void removeUnusedSyntheticSections() { 1170 // All input synthetic sections that can be empty are placed after 1171 // all regular ones. We iterate over them all and exit at first 1172 // non-synthetic. 1173 for (InputSectionBase *S : llvm::reverse(InputSections)) { 1174 SyntheticSection *SS = dyn_cast<SyntheticSection>(S); 1175 if (!SS) 1176 return; 1177 OutputSection *OS = SS->getParent(); 1178 if (!SS->empty() || !OS) 1179 continue; 1180 if ((SS == InX::Got || SS == InX::MipsGot) && ElfSym::GlobalOffsetTable) 1181 continue; 1182 1183 std::vector<BaseCommand *>::iterator Empty = OS->Commands.end(); 1184 for (auto I = OS->Commands.begin(), E = OS->Commands.end(); I != E; ++I) { 1185 BaseCommand *B = *I; 1186 if (auto *ISD = dyn_cast<InputSectionDescription>(B)) { 1187 llvm::erase_if(ISD->Sections, 1188 [=](InputSection *IS) { return IS == SS; }); 1189 if (ISD->Sections.empty()) 1190 Empty = I; 1191 } 1192 } 1193 if (Empty != OS->Commands.end()) 1194 OS->Commands.erase(Empty); 1195 1196 // If there are no other sections in the output section, remove it from the 1197 // output. 1198 if (OS->Commands.empty()) 1199 llvm::erase_if(Script->Opt.Commands, 1200 [&](BaseCommand *Cmd) { return Cmd == OS; }); 1201 } 1202 } 1203 1204 // Returns true if a symbol can be replaced at load-time by a symbol 1205 // with the same name defined in other ELF executable or DSO. 1206 static bool computeIsPreemptible(const SymbolBody &B) { 1207 assert(!B.isLocal()); 1208 // Only symbols that appear in dynsym can be preempted. 1209 if (!B.symbol()->includeInDynsym()) 1210 return false; 1211 1212 // Only default visibility symbols can be preempted. 1213 if (B.symbol()->Visibility != STV_DEFAULT) 1214 return false; 1215 1216 // At this point copy relocations have not been created yet, so any 1217 // symbol that is not defined locally is preemptible. 1218 if (!B.isInCurrentDSO()) 1219 return true; 1220 1221 // If we have a dynamic list it specifies which local symbols are preemptible. 1222 if (Config->HasDynamicList) 1223 return false; 1224 1225 if (!Config->Shared) 1226 return false; 1227 1228 // -Bsymbolic means that definitions are not preempted. 1229 if (Config->Bsymbolic || (Config->BsymbolicFunctions && B.isFunc())) 1230 return false; 1231 return true; 1232 } 1233 1234 // Create output section objects and add them to OutputSections. 1235 template <class ELFT> void Writer<ELFT>::finalizeSections() { 1236 Out::DebugInfo = findSection(".debug_info"); 1237 Out::PreinitArray = findSection(".preinit_array"); 1238 Out::InitArray = findSection(".init_array"); 1239 Out::FiniArray = findSection(".fini_array"); 1240 1241 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop 1242 // symbols for sections, so that the runtime can get the start and end 1243 // addresses of each section by section name. Add such symbols. 1244 if (!Config->Relocatable) { 1245 addStartEndSymbols(); 1246 for (BaseCommand *Base : Script->Opt.Commands) 1247 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1248 addStartStopSymbols(Sec); 1249 } 1250 1251 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type. 1252 // It should be okay as no one seems to care about the type. 1253 // Even the author of gold doesn't remember why gold behaves that way. 1254 // https://sourceware.org/ml/binutils/2002-03/msg00360.html 1255 if (InX::DynSymTab) 1256 addRegular<ELFT>("_DYNAMIC", InX::Dynamic, 0); 1257 1258 // Define __rel[a]_iplt_{start,end} symbols if needed. 1259 addRelIpltSymbols(); 1260 1261 // This responsible for splitting up .eh_frame section into 1262 // pieces. The relocation scan uses those pieces, so this has to be 1263 // earlier. 1264 applySynthetic({In<ELFT>::EhFrame}, 1265 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1266 1267 for (Symbol *S : Symtab->getSymbols()) 1268 S->body()->IsPreemptible |= computeIsPreemptible(*S->body()); 1269 1270 // Scan relocations. This must be done after every symbol is declared so that 1271 // we can correctly decide if a dynamic relocation is needed. 1272 if (!Config->Relocatable) 1273 forEachRelSec(scanRelocations<ELFT>); 1274 1275 if (InX::Plt && !InX::Plt->empty()) 1276 InX::Plt->addSymbols(); 1277 if (InX::Iplt && !InX::Iplt->empty()) 1278 InX::Iplt->addSymbols(); 1279 1280 // Now that we have defined all possible global symbols including linker- 1281 // synthesized ones. Visit all symbols to give the finishing touches. 1282 for (Symbol *S : Symtab->getSymbols()) { 1283 SymbolBody *Body = S->body(); 1284 1285 if (!includeInSymtab(*Body)) 1286 continue; 1287 if (InX::SymTab) 1288 InX::SymTab->addSymbol(Body); 1289 1290 if (InX::DynSymTab && S->includeInDynsym()) { 1291 InX::DynSymTab->addSymbol(Body); 1292 if (auto *SS = dyn_cast<SharedSymbol>(Body)) 1293 if (cast<SharedFile<ELFT>>(S->File)->isNeeded()) 1294 In<ELFT>::VerNeed->addSymbol(SS); 1295 } 1296 } 1297 1298 // Do not proceed if there was an undefined symbol. 1299 if (ErrorCount) 1300 return; 1301 1302 addPredefinedSections(); 1303 removeUnusedSyntheticSections(); 1304 1305 sortSections(); 1306 1307 // Now that we have the final list, create a list of all the 1308 // OutputSections for convenience. 1309 for (BaseCommand *Base : Script->Opt.Commands) 1310 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1311 OutputSections.push_back(Sec); 1312 1313 // Prefer command line supplied address over other constraints. 1314 for (OutputSection *Sec : OutputSections) { 1315 auto I = Config->SectionStartMap.find(Sec->Name); 1316 if (I != Config->SectionStartMap.end()) 1317 Sec->AddrExpr = [=] { return I->second; }; 1318 } 1319 1320 // This is a bit of a hack. A value of 0 means undef, so we set it 1321 // to 1 t make __ehdr_start defined. The section number is not 1322 // particularly relevant. 1323 Out::ElfHeader->SectionIndex = 1; 1324 1325 unsigned I = 1; 1326 for (OutputSection *Sec : OutputSections) { 1327 Sec->SectionIndex = I++; 1328 Sec->ShName = InX::ShStrTab->addString(Sec->Name); 1329 } 1330 1331 // Binary and relocatable output does not have PHDRS. 1332 // The headers have to be created before finalize as that can influence the 1333 // image base and the dynamic section on mips includes the image base. 1334 if (!Config->Relocatable && !Config->OFormatBinary) { 1335 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs(); 1336 addPtArmExid(Phdrs); 1337 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size(); 1338 } 1339 1340 // Some symbols are defined in term of program headers. Now that we 1341 // have the headers, we can find out which sections they point to. 1342 setReservedSymbolSections(); 1343 1344 // Dynamic section must be the last one in this list and dynamic 1345 // symbol table section (DynSymTab) must be the first one. 1346 applySynthetic({InX::DynSymTab, InX::Bss, 1347 InX::BssRelRo, InX::GnuHashTab, 1348 In<ELFT>::HashTab, InX::SymTab, 1349 InX::ShStrTab, InX::StrTab, 1350 In<ELFT>::VerDef, InX::DynStrTab, 1351 InX::Got, InX::MipsGot, 1352 InX::IgotPlt, InX::GotPlt, 1353 In<ELFT>::RelaDyn, In<ELFT>::RelaIplt, 1354 In<ELFT>::RelaPlt, InX::Plt, 1355 InX::Iplt, In<ELFT>::EhFrameHdr, 1356 In<ELFT>::VerSym, In<ELFT>::VerNeed, 1357 InX::Dynamic}, 1358 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1359 1360 if (!Script->Opt.HasSections && !Config->Relocatable) 1361 fixSectionAlignments(); 1362 1363 // Some architectures use small displacements for jump instructions. 1364 // It is linker's responsibility to create thunks containing long 1365 // jump instructions if jump targets are too far. Create thunks. 1366 if (Target->NeedsThunks) { 1367 // FIXME: only ARM Interworking and Mips LA25 Thunks are implemented, 1368 // these 1369 // do not require address information. To support range extension Thunks 1370 // we need to assign addresses so that we can tell if jump instructions 1371 // are out of range. This will need to turn into a loop that converges 1372 // when no more Thunks are added 1373 ThunkCreator TC; 1374 Script->assignAddresses(); 1375 if (TC.createThunks(OutputSections)) { 1376 applySynthetic({InX::MipsGot}, 1377 [](SyntheticSection *SS) { SS->updateAllocSize(); }); 1378 if (TC.createThunks(OutputSections)) 1379 fatal("All non-range thunks should be created in first call"); 1380 } 1381 } 1382 1383 // Fill other section headers. The dynamic table is finalized 1384 // at the end because some tags like RELSZ depend on result 1385 // of finalizing other sections. 1386 for (OutputSection *Sec : OutputSections) 1387 Sec->finalize<ELFT>(); 1388 1389 // createThunks may have added local symbols to the static symbol table 1390 applySynthetic({InX::SymTab, InX::ShStrTab, InX::StrTab}, 1391 [](SyntheticSection *SS) { SS->postThunkContents(); }); 1392 } 1393 1394 template <class ELFT> void Writer<ELFT>::addPredefinedSections() { 1395 // ARM ABI requires .ARM.exidx to be terminated by some piece of data. 1396 // We have the terminater synthetic section class. Add that at the end. 1397 OutputSection *Cmd = findSection(".ARM.exidx"); 1398 if (!Cmd || !Cmd->Live || Config->Relocatable) 1399 return; 1400 1401 auto *Sentinel = make<ARMExidxSentinelSection>(); 1402 Cmd->addSection(Sentinel); 1403 } 1404 1405 // The linker is expected to define SECNAME_start and SECNAME_end 1406 // symbols for a few sections. This function defines them. 1407 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() { 1408 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) { 1409 // These symbols resolve to the image base if the section does not exist. 1410 // A special value -1 indicates end of the section. 1411 if (OS) { 1412 addOptionalRegular<ELFT>(Start, OS, 0); 1413 addOptionalRegular<ELFT>(End, OS, -1); 1414 } else { 1415 if (Config->Pic) 1416 OS = Out::ElfHeader; 1417 addOptionalRegular<ELFT>(Start, OS, 0); 1418 addOptionalRegular<ELFT>(End, OS, 0); 1419 } 1420 }; 1421 1422 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray); 1423 Define("__init_array_start", "__init_array_end", Out::InitArray); 1424 Define("__fini_array_start", "__fini_array_end", Out::FiniArray); 1425 1426 if (OutputSection *Sec = findSection(".ARM.exidx")) 1427 Define("__exidx_start", "__exidx_end", Sec); 1428 } 1429 1430 // If a section name is valid as a C identifier (which is rare because of 1431 // the leading '.'), linkers are expected to define __start_<secname> and 1432 // __stop_<secname> symbols. They are at beginning and end of the section, 1433 // respectively. This is not requested by the ELF standard, but GNU ld and 1434 // gold provide the feature, and used by many programs. 1435 template <class ELFT> 1436 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) { 1437 StringRef S = Sec->Name; 1438 if (!isValidCIdentifier(S)) 1439 return; 1440 addOptionalRegular<ELFT>(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT); 1441 addOptionalRegular<ELFT>(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT); 1442 } 1443 1444 template <class ELFT> OutputSection *Writer<ELFT>::findSection(StringRef Name) { 1445 for (BaseCommand *Base : Script->Opt.Commands) 1446 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1447 if (Sec->Name == Name) 1448 return Sec; 1449 return nullptr; 1450 } 1451 1452 static bool needsPtLoad(OutputSection *Sec) { 1453 if (!(Sec->Flags & SHF_ALLOC)) 1454 return false; 1455 1456 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is 1457 // responsible for allocating space for them, not the PT_LOAD that 1458 // contains the TLS initialization image. 1459 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS) 1460 return false; 1461 return true; 1462 } 1463 1464 // Linker scripts are responsible for aligning addresses. Unfortunately, most 1465 // linker scripts are designed for creating two PT_LOADs only, one RX and one 1466 // RW. This means that there is no alignment in the RO to RX transition and we 1467 // cannot create a PT_LOAD there. 1468 static uint64_t computeFlags(uint64_t Flags) { 1469 if (Config->Omagic) 1470 return PF_R | PF_W | PF_X; 1471 if (Config->SingleRoRx && !(Flags & PF_W)) 1472 return Flags | PF_X; 1473 return Flags; 1474 } 1475 1476 // Decide which program headers to create and which sections to include in each 1477 // one. 1478 template <class ELFT> std::vector<PhdrEntry *> Writer<ELFT>::createPhdrs() { 1479 std::vector<PhdrEntry *> Ret; 1480 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * { 1481 Ret.push_back(make<PhdrEntry>(Type, Flags)); 1482 return Ret.back(); 1483 }; 1484 1485 // The first phdr entry is PT_PHDR which describes the program header itself. 1486 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders); 1487 1488 // PT_INTERP must be the second entry if exists. 1489 if (OutputSection *Cmd = findSection(".interp")) 1490 AddHdr(PT_INTERP, Cmd->getPhdrFlags())->add(Cmd); 1491 1492 // Add the first PT_LOAD segment for regular output sections. 1493 uint64_t Flags = computeFlags(PF_R); 1494 PhdrEntry *Load = AddHdr(PT_LOAD, Flags); 1495 1496 // Add the headers. We will remove them if they don't fit. 1497 Load->add(Out::ElfHeader); 1498 Load->add(Out::ProgramHeaders); 1499 1500 for (OutputSection *Sec : OutputSections) { 1501 if (!(Sec->Flags & SHF_ALLOC)) 1502 break; 1503 if (!needsPtLoad(Sec)) 1504 continue; 1505 1506 // Segments are contiguous memory regions that has the same attributes 1507 // (e.g. executable or writable). There is one phdr for each segment. 1508 // Therefore, we need to create a new phdr when the next section has 1509 // different flags or is loaded at a discontiguous address using AT linker 1510 // script command. 1511 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags()); 1512 if (Sec->LMAExpr || Flags != NewFlags) { 1513 Load = AddHdr(PT_LOAD, NewFlags); 1514 Flags = NewFlags; 1515 } 1516 1517 Load->add(Sec); 1518 } 1519 1520 // Add a TLS segment if any. 1521 PhdrEntry *TlsHdr = make<PhdrEntry>(PT_TLS, PF_R); 1522 for (OutputSection *Sec : OutputSections) 1523 if (Sec->Flags & SHF_TLS) 1524 TlsHdr->add(Sec); 1525 if (TlsHdr->FirstSec) 1526 Ret.push_back(TlsHdr); 1527 1528 // Add an entry for .dynamic. 1529 if (InX::DynSymTab) 1530 AddHdr(PT_DYNAMIC, InX::Dynamic->getParent()->getPhdrFlags()) 1531 ->add(InX::Dynamic->getParent()); 1532 1533 // PT_GNU_RELRO includes all sections that should be marked as 1534 // read-only by dynamic linker after proccessing relocations. 1535 PhdrEntry *RelRo = make<PhdrEntry>(PT_GNU_RELRO, PF_R); 1536 for (OutputSection *Sec : OutputSections) 1537 if (needsPtLoad(Sec) && isRelroSection(Sec)) 1538 RelRo->add(Sec); 1539 if (RelRo->FirstSec) 1540 Ret.push_back(RelRo); 1541 1542 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr. 1543 if (!In<ELFT>::EhFrame->empty() && In<ELFT>::EhFrameHdr && 1544 In<ELFT>::EhFrame->getParent() && In<ELFT>::EhFrameHdr->getParent()) 1545 AddHdr(PT_GNU_EH_FRAME, In<ELFT>::EhFrameHdr->getParent()->getPhdrFlags()) 1546 ->add(In<ELFT>::EhFrameHdr->getParent()); 1547 1548 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes 1549 // the dynamic linker fill the segment with random data. 1550 if (OutputSection *Cmd = findSection(".openbsd.randomdata")) 1551 AddHdr(PT_OPENBSD_RANDOMIZE, Cmd->getPhdrFlags())->add(Cmd); 1552 1553 // PT_GNU_STACK is a special section to tell the loader to make the 1554 // pages for the stack non-executable. If you really want an executable 1555 // stack, you can pass -z execstack, but that's not recommended for 1556 // security reasons. 1557 unsigned Perm; 1558 if (Config->ZExecstack) 1559 Perm = PF_R | PF_W | PF_X; 1560 else 1561 Perm = PF_R | PF_W; 1562 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize; 1563 1564 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable 1565 // is expected to perform W^X violations, such as calling mprotect(2) or 1566 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on 1567 // OpenBSD. 1568 if (Config->ZWxneeded) 1569 AddHdr(PT_OPENBSD_WXNEEDED, PF_X); 1570 1571 // Create one PT_NOTE per a group of contiguous .note sections. 1572 PhdrEntry *Note = nullptr; 1573 for (OutputSection *Sec : OutputSections) { 1574 if (Sec->Type == SHT_NOTE) { 1575 if (!Note || Sec->LMAExpr) 1576 Note = AddHdr(PT_NOTE, PF_R); 1577 Note->add(Sec); 1578 } else { 1579 Note = nullptr; 1580 } 1581 } 1582 return Ret; 1583 } 1584 1585 template <class ELFT> 1586 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry *> &Phdrs) { 1587 if (Config->EMachine != EM_ARM) 1588 return; 1589 auto I = llvm::find_if(OutputSections, [](OutputSection *Cmd) { 1590 return Cmd->Type == SHT_ARM_EXIDX; 1591 }); 1592 if (I == OutputSections.end()) 1593 return; 1594 1595 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME 1596 PhdrEntry *ARMExidx = make<PhdrEntry>(PT_ARM_EXIDX, PF_R); 1597 ARMExidx->add(*I); 1598 Phdrs.push_back(ARMExidx); 1599 } 1600 1601 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the 1602 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic 1603 // linker can set the permissions. 1604 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() { 1605 auto PageAlign = [](OutputSection *Cmd) { 1606 if (Cmd && !Cmd->AddrExpr) 1607 Cmd->AddrExpr = [=] { 1608 return alignTo(Script->getDot(), Config->MaxPageSize); 1609 }; 1610 }; 1611 1612 for (const PhdrEntry *P : Phdrs) 1613 if (P->p_type == PT_LOAD && P->FirstSec) 1614 PageAlign(P->FirstSec); 1615 1616 for (const PhdrEntry *P : Phdrs) { 1617 if (P->p_type != PT_GNU_RELRO) 1618 continue; 1619 if (P->FirstSec) 1620 PageAlign(P->FirstSec); 1621 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we 1622 // have to align it to a page. 1623 auto End = OutputSections.end(); 1624 auto I = std::find(OutputSections.begin(), End, P->LastSec); 1625 if (I == End || (I + 1) == End) 1626 continue; 1627 OutputSection *Cmd = (*(I + 1)); 1628 if (needsPtLoad(Cmd)) 1629 PageAlign(Cmd); 1630 } 1631 } 1632 1633 // Adjusts the file alignment for a given output section and returns 1634 // its new file offset. The file offset must be the same with its 1635 // virtual address (modulo the page size) so that the loader can load 1636 // executables without any address adjustment. 1637 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Cmd) { 1638 // If the section is not in a PT_LOAD, we just have to align it. 1639 if (!Cmd->PtLoad) 1640 return alignTo(Off, Cmd->Alignment); 1641 1642 OutputSection *First = Cmd->PtLoad->FirstSec; 1643 // The first section in a PT_LOAD has to have congruent offset and address 1644 // module the page size. 1645 if (Cmd == First) 1646 return alignTo(Off, std::max<uint64_t>(Cmd->Alignment, Config->MaxPageSize), 1647 Cmd->Addr); 1648 1649 // If two sections share the same PT_LOAD the file offset is calculated 1650 // using this formula: Off2 = Off1 + (VA2 - VA1). 1651 return First->Offset + Cmd->Addr - First->Addr; 1652 } 1653 1654 static uint64_t setOffset(OutputSection *Cmd, uint64_t Off) { 1655 if (Cmd->Type == SHT_NOBITS) { 1656 Cmd->Offset = Off; 1657 return Off; 1658 } 1659 1660 Off = getFileAlignment(Off, Cmd); 1661 Cmd->Offset = Off; 1662 return Off + Cmd->Size; 1663 } 1664 1665 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() { 1666 uint64_t Off = 0; 1667 for (OutputSection *Sec : OutputSections) 1668 if (Sec->Flags & SHF_ALLOC) 1669 Off = setOffset(Sec, Off); 1670 FileSize = alignTo(Off, Config->Wordsize); 1671 } 1672 1673 // Assign file offsets to output sections. 1674 template <class ELFT> void Writer<ELFT>::assignFileOffsets() { 1675 uint64_t Off = 0; 1676 Off = setOffset(Out::ElfHeader, Off); 1677 Off = setOffset(Out::ProgramHeaders, Off); 1678 1679 PhdrEntry *LastRX = nullptr; 1680 for (PhdrEntry *P : Phdrs) 1681 if (P->p_type == PT_LOAD && (P->p_flags & PF_X)) 1682 LastRX = P; 1683 1684 for (OutputSection *Sec : OutputSections) { 1685 Off = setOffset(Sec, Off); 1686 if (Script->Opt.HasSections) 1687 continue; 1688 // If this is a last section of the last executable segment and that 1689 // segment is the last loadable segment, align the offset of the 1690 // following section to avoid loading non-segments parts of the file. 1691 if (LastRX && LastRX->LastSec == Sec) 1692 Off = alignTo(Off, Target->PageSize); 1693 } 1694 1695 SectionHeaderOff = alignTo(Off, Config->Wordsize); 1696 FileSize = SectionHeaderOff + (OutputSections.size() + 1) * sizeof(Elf_Shdr); 1697 } 1698 1699 // Finalize the program headers. We call this function after we assign 1700 // file offsets and VAs to all sections. 1701 template <class ELFT> void Writer<ELFT>::setPhdrs() { 1702 for (PhdrEntry *P : Phdrs) { 1703 OutputSection *First = P->FirstSec; 1704 OutputSection *Last = P->LastSec; 1705 if (First) { 1706 P->p_filesz = Last->Offset - First->Offset; 1707 if (Last->Type != SHT_NOBITS) 1708 P->p_filesz += Last->Size; 1709 P->p_memsz = Last->Addr + Last->Size - First->Addr; 1710 P->p_offset = First->Offset; 1711 P->p_vaddr = First->Addr; 1712 if (!P->HasLMA) 1713 P->p_paddr = First->getLMA(); 1714 } 1715 if (P->p_type == PT_LOAD) 1716 P->p_align = std::max<uint64_t>(P->p_align, Config->MaxPageSize); 1717 else if (P->p_type == PT_GNU_RELRO) { 1718 P->p_align = 1; 1719 // The glibc dynamic loader rounds the size down, so we need to round up 1720 // to protect the last page. This is a no-op on FreeBSD which always 1721 // rounds up. 1722 P->p_memsz = alignTo(P->p_memsz, Target->PageSize); 1723 } 1724 1725 // The TLS pointer goes after PT_TLS. At least glibc will align it, 1726 // so round up the size to make sure the offsets are correct. 1727 if (P->p_type == PT_TLS) { 1728 Out::TlsPhdr = P; 1729 if (P->p_memsz) 1730 P->p_memsz = alignTo(P->p_memsz, P->p_align); 1731 } 1732 } 1733 } 1734 1735 // The entry point address is chosen in the following ways. 1736 // 1737 // 1. the '-e' entry command-line option; 1738 // 2. the ENTRY(symbol) command in a linker control script; 1739 // 3. the value of the symbol start, if present; 1740 // 4. the address of the first byte of the .text section, if present; 1741 // 5. the address 0. 1742 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() { 1743 // Case 1, 2 or 3. As a special case, if the symbol is actually 1744 // a number, we'll use that number as an address. 1745 if (SymbolBody *B = Symtab->find(Config->Entry)) 1746 return B->getVA(); 1747 uint64_t Addr; 1748 if (to_integer(Config->Entry, Addr)) 1749 return Addr; 1750 1751 // Case 4 1752 if (OutputSection *Sec = findSection(".text")) { 1753 if (Config->WarnMissingEntry) 1754 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" + 1755 utohexstr(Sec->Addr)); 1756 return Sec->Addr; 1757 } 1758 1759 // Case 5 1760 if (Config->WarnMissingEntry) 1761 warn("cannot find entry symbol " + Config->Entry + 1762 "; not setting start address"); 1763 return 0; 1764 } 1765 1766 static uint16_t getELFType() { 1767 if (Config->Pic) 1768 return ET_DYN; 1769 if (Config->Relocatable) 1770 return ET_REL; 1771 return ET_EXEC; 1772 } 1773 1774 template <class ELFT> void Writer<ELFT>::writeHeader() { 1775 uint8_t *Buf = Buffer->getBufferStart(); 1776 memcpy(Buf, "\177ELF", 4); 1777 1778 // Write the ELF header. 1779 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf); 1780 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32; 1781 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB; 1782 EHdr->e_ident[EI_VERSION] = EV_CURRENT; 1783 EHdr->e_ident[EI_OSABI] = Config->OSABI; 1784 EHdr->e_type = getELFType(); 1785 EHdr->e_machine = Config->EMachine; 1786 EHdr->e_version = EV_CURRENT; 1787 EHdr->e_entry = getEntryAddr(); 1788 EHdr->e_shoff = SectionHeaderOff; 1789 EHdr->e_ehsize = sizeof(Elf_Ehdr); 1790 EHdr->e_phnum = Phdrs.size(); 1791 EHdr->e_shentsize = sizeof(Elf_Shdr); 1792 EHdr->e_shnum = OutputSections.size() + 1; 1793 EHdr->e_shstrndx = InX::ShStrTab->getParent()->SectionIndex; 1794 1795 if (Config->EMachine == EM_ARM) 1796 // We don't currently use any features incompatible with EF_ARM_EABI_VER5, 1797 // but we don't have any firm guarantees of conformance. Linux AArch64 1798 // kernels (as of 2016) require an EABI version to be set. 1799 EHdr->e_flags = EF_ARM_EABI_VER5; 1800 else if (Config->EMachine == EM_MIPS) 1801 EHdr->e_flags = getMipsEFlags<ELFT>(); 1802 1803 if (!Config->Relocatable) { 1804 EHdr->e_phoff = sizeof(Elf_Ehdr); 1805 EHdr->e_phentsize = sizeof(Elf_Phdr); 1806 } 1807 1808 // Write the program header table. 1809 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff); 1810 for (PhdrEntry *P : Phdrs) { 1811 HBuf->p_type = P->p_type; 1812 HBuf->p_flags = P->p_flags; 1813 HBuf->p_offset = P->p_offset; 1814 HBuf->p_vaddr = P->p_vaddr; 1815 HBuf->p_paddr = P->p_paddr; 1816 HBuf->p_filesz = P->p_filesz; 1817 HBuf->p_memsz = P->p_memsz; 1818 HBuf->p_align = P->p_align; 1819 ++HBuf; 1820 } 1821 1822 // Write the section header table. Note that the first table entry is null. 1823 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff); 1824 for (OutputSection *Sec : OutputSections) 1825 Sec->writeHeaderTo<ELFT>(++SHdrs); 1826 } 1827 1828 // Open a result file. 1829 template <class ELFT> void Writer<ELFT>::openFile() { 1830 if (!Config->Is64 && FileSize > UINT32_MAX) { 1831 error("output file too large: " + Twine(FileSize) + " bytes"); 1832 return; 1833 } 1834 1835 unlinkAsync(Config->OutputFile); 1836 ErrorOr<std::unique_ptr<FileOutputBuffer>> BufferOrErr = 1837 FileOutputBuffer::create(Config->OutputFile, FileSize, 1838 FileOutputBuffer::F_executable); 1839 1840 if (auto EC = BufferOrErr.getError()) 1841 error("failed to open " + Config->OutputFile + ": " + EC.message()); 1842 else 1843 Buffer = std::move(*BufferOrErr); 1844 } 1845 1846 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() { 1847 uint8_t *Buf = Buffer->getBufferStart(); 1848 for (OutputSection *Sec : OutputSections) 1849 if (Sec->Flags & SHF_ALLOC) 1850 Sec->writeTo<ELFT>(Buf + Sec->Offset); 1851 } 1852 1853 static void fillTrap(uint8_t *I, uint8_t *End) { 1854 for (; I + 4 <= End; I += 4) 1855 memcpy(I, &Target->TrapInstr, 4); 1856 } 1857 1858 // Fill the last page of executable segments with trap instructions 1859 // instead of leaving them as zero. Even though it is not required by any 1860 // standard, it is in general a good thing to do for security reasons. 1861 // 1862 // We'll leave other pages in segments as-is because the rest will be 1863 // overwritten by output sections. 1864 template <class ELFT> void Writer<ELFT>::writeTrapInstr() { 1865 if (Script->Opt.HasSections) 1866 return; 1867 1868 // Fill the last page. 1869 uint8_t *Buf = Buffer->getBufferStart(); 1870 for (PhdrEntry *P : Phdrs) 1871 if (P->p_type == PT_LOAD && (P->p_flags & PF_X)) 1872 fillTrap(Buf + alignDown(P->p_offset + P->p_filesz, Target->PageSize), 1873 Buf + alignTo(P->p_offset + P->p_filesz, Target->PageSize)); 1874 1875 // Round up the file size of the last segment to the page boundary iff it is 1876 // an executable segment to ensure that other other tools don't accidentally 1877 // trim the instruction padding (e.g. when stripping the file). 1878 PhdrEntry *LastRX = nullptr; 1879 for (PhdrEntry *P : Phdrs) { 1880 if (P->p_type != PT_LOAD) 1881 continue; 1882 if (P->p_flags & PF_X) 1883 LastRX = P; 1884 else 1885 LastRX = nullptr; 1886 } 1887 if (LastRX) 1888 LastRX->p_memsz = LastRX->p_filesz = 1889 alignTo(LastRX->p_filesz, Target->PageSize); 1890 } 1891 1892 // Write section contents to a mmap'ed file. 1893 template <class ELFT> void Writer<ELFT>::writeSections() { 1894 uint8_t *Buf = Buffer->getBufferStart(); 1895 1896 // PPC64 needs to process relocations in the .opd section 1897 // before processing relocations in code-containing sections. 1898 if (auto *OpdCmd = findSection(".opd")) { 1899 Out::Opd = OpdCmd; 1900 Out::OpdBuf = Buf + Out::Opd->Offset; 1901 OpdCmd->template writeTo<ELFT>(Buf + Out::Opd->Offset); 1902 } 1903 1904 OutputSection *EhFrameHdr = 1905 (In<ELFT>::EhFrameHdr && !In<ELFT>::EhFrameHdr->empty()) 1906 ? In<ELFT>::EhFrameHdr->getParent() 1907 : nullptr; 1908 1909 // In -r or -emit-relocs mode, write the relocation sections first as in 1910 // ELf_Rel targets we might find out that we need to modify the relocated 1911 // section while doing it. 1912 for (OutputSection *Sec : OutputSections) 1913 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA) 1914 Sec->writeTo<ELFT>(Buf + Sec->Offset); 1915 1916 for (OutputSection *Sec : OutputSections) 1917 if (Sec != Out::Opd && Sec != EhFrameHdr && Sec->Type != SHT_REL && 1918 Sec->Type != SHT_RELA) 1919 Sec->writeTo<ELFT>(Buf + Sec->Offset); 1920 1921 // The .eh_frame_hdr depends on .eh_frame section contents, therefore 1922 // it should be written after .eh_frame is written. 1923 if (EhFrameHdr) 1924 EhFrameHdr->writeTo<ELFT>(Buf + EhFrameHdr->Offset); 1925 } 1926 1927 template <class ELFT> void Writer<ELFT>::writeBuildId() { 1928 if (!InX::BuildId || !InX::BuildId->getParent()) 1929 return; 1930 1931 // Compute a hash of all sections of the output file. 1932 uint8_t *Start = Buffer->getBufferStart(); 1933 uint8_t *End = Start + FileSize; 1934 InX::BuildId->writeBuildId({Start, End}); 1935 } 1936 1937 template void elf::writeResult<ELF32LE>(); 1938 template void elf::writeResult<ELF32BE>(); 1939 template void elf::writeResult<ELF64LE>(); 1940 template void elf::writeResult<ELF64BE>(); 1941