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 static bool needsInterpSection() { 119 return !SharedFiles.empty() && !Config->DynamicLinker.empty() && 120 !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()) { 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 if (Config->EMachine == EM_MIPS) { 303 if (!Config->Shared && Config->HasDynSymTab) { 304 InX::MipsRldMap = make<MipsRldMapSection>(); 305 Add(InX::MipsRldMap); 306 } 307 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create()) 308 Add(Sec); 309 if (auto *Sec = MipsOptionsSection<ELFT>::create()) 310 Add(Sec); 311 if (auto *Sec = MipsReginfoSection<ELFT>::create()) 312 Add(Sec); 313 } 314 315 if (Config->HasDynSymTab) { 316 InX::DynSymTab = make<SymbolTableSection<ELFT>>(*InX::DynStrTab); 317 Add(InX::DynSymTab); 318 319 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>(); 320 Add(In<ELFT>::VerSym); 321 322 if (!Config->VersionDefinitions.empty()) { 323 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>(); 324 Add(In<ELFT>::VerDef); 325 } 326 327 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>(); 328 Add(In<ELFT>::VerNeed); 329 330 if (Config->GnuHash) { 331 InX::GnuHashTab = make<GnuHashTableSection>(); 332 Add(InX::GnuHashTab); 333 } 334 335 if (Config->SysvHash) { 336 In<ELFT>::HashTab = make<HashTableSection<ELFT>>(); 337 Add(In<ELFT>::HashTab); 338 } 339 340 Add(InX::Dynamic); 341 Add(InX::DynStrTab); 342 Add(In<ELFT>::RelaDyn); 343 } 344 345 // Add .got. MIPS' .got is so different from the other archs, 346 // it has its own class. 347 if (Config->EMachine == EM_MIPS) { 348 InX::MipsGot = make<MipsGotSection>(); 349 Add(InX::MipsGot); 350 } else { 351 InX::Got = make<GotSection>(); 352 Add(InX::Got); 353 } 354 355 InX::GotPlt = make<GotPltSection>(); 356 Add(InX::GotPlt); 357 InX::IgotPlt = make<IgotPltSection>(); 358 Add(InX::IgotPlt); 359 360 if (Config->GdbIndex) { 361 InX::GdbIndex = createGdbIndex<ELFT>(); 362 Add(InX::GdbIndex); 363 } 364 365 // We always need to add rel[a].plt to output if it has entries. 366 // Even for static linking it can contain R_[*]_IRELATIVE relocations. 367 In<ELFT>::RelaPlt = make<RelocationSection<ELFT>>( 368 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/); 369 Add(In<ELFT>::RelaPlt); 370 371 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure 372 // that the IRelative relocations are processed last by the dynamic loader 373 In<ELFT>::RelaIplt = make<RelocationSection<ELFT>>( 374 (Config->EMachine == EM_ARM) ? ".rel.dyn" : In<ELFT>::RelaPlt->Name, 375 false /*Sort*/); 376 Add(In<ELFT>::RelaIplt); 377 378 InX::Plt = make<PltSection>(Target->PltHeaderSize); 379 Add(InX::Plt); 380 InX::Iplt = make<PltSection>(0); 381 Add(InX::Iplt); 382 383 if (!Config->Relocatable) { 384 if (Config->EhFrameHdr) { 385 In<ELFT>::EhFrameHdr = make<EhFrameHeader<ELFT>>(); 386 Add(In<ELFT>::EhFrameHdr); 387 } 388 In<ELFT>::EhFrame = make<EhFrameSection<ELFT>>(); 389 Add(In<ELFT>::EhFrame); 390 } 391 392 if (InX::SymTab) 393 Add(InX::SymTab); 394 Add(InX::ShStrTab); 395 if (InX::StrTab) 396 Add(InX::StrTab); 397 } 398 399 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName, 400 const SymbolBody &B) { 401 if (B.isFile() || B.isSection()) 402 return false; 403 404 // If sym references a section in a discarded group, don't keep it. 405 if (Sec == &InputSection::Discarded) 406 return false; 407 408 if (Config->Discard == DiscardPolicy::None) 409 return true; 410 411 // In ELF assembly .L symbols are normally discarded by the assembler. 412 // If the assembler fails to do so, the linker discards them if 413 // * --discard-locals is used. 414 // * The symbol is in a SHF_MERGE section, which is normally the reason for 415 // the assembler keeping the .L symbol. 416 if (!SymName.startswith(".L") && !SymName.empty()) 417 return true; 418 419 if (Config->Discard == DiscardPolicy::Locals) 420 return false; 421 422 return !Sec || !(Sec->Flags & SHF_MERGE); 423 } 424 425 static bool includeInSymtab(const SymbolBody &B) { 426 if (!B.isLocal() && !B.symbol()->IsUsedInRegularObj) 427 return false; 428 429 if (auto *D = dyn_cast<DefinedRegular>(&B)) { 430 // Always include absolute symbols. 431 SectionBase *Sec = D->Section; 432 if (!Sec) 433 return true; 434 if (auto *IS = dyn_cast<InputSectionBase>(Sec)) { 435 Sec = IS->Repl; 436 IS = cast<InputSectionBase>(Sec); 437 // Exclude symbols pointing to garbage-collected sections. 438 if (!IS->Live) 439 return false; 440 } 441 if (auto *S = dyn_cast<MergeInputSection>(Sec)) 442 if (!S->getSectionPiece(D->Value)->Live) 443 return false; 444 return true; 445 } 446 447 if (auto *Sym = dyn_cast<DefinedCommon>(&B)) 448 return Sym->Live; 449 return true; 450 } 451 452 // Local symbols are not in the linker's symbol table. This function scans 453 // each object file's symbol table to copy local symbols to the output. 454 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() { 455 if (!InX::SymTab) 456 return; 457 for (InputFile *File : ObjectFiles) { 458 ObjFile<ELFT> *F = cast<ObjFile<ELFT>>(File); 459 for (SymbolBody *B : F->getLocalSymbols()) { 460 if (!B->IsLocal) 461 fatal(toString(F) + 462 ": broken object: getLocalSymbols returns a non-local symbol"); 463 auto *DR = dyn_cast<DefinedRegular>(B); 464 465 // No reason to keep local undefined symbol in symtab. 466 if (!DR) 467 continue; 468 if (!includeInSymtab(*B)) 469 continue; 470 471 SectionBase *Sec = DR->Section; 472 if (!shouldKeepInSymtab(Sec, B->getName(), *B)) 473 continue; 474 InX::SymTab->addSymbol(B); 475 } 476 } 477 } 478 479 template <class ELFT> void Writer<ELFT>::addSectionSymbols() { 480 // Create one STT_SECTION symbol for each output section we might 481 // have a relocation with. 482 for (BaseCommand *Base : Script->Opt.Commands) { 483 auto *Sec = dyn_cast<OutputSection>(Base); 484 if (!Sec) 485 continue; 486 auto I = llvm::find_if(Sec->Commands, [](BaseCommand *Base) { 487 if (auto *ISD = dyn_cast<InputSectionDescription>(Base)) 488 return !ISD->Sections.empty(); 489 return false; 490 }); 491 if (I == Sec->Commands.end()) 492 continue; 493 InputSection *IS = cast<InputSectionDescription>(*I)->Sections[0]; 494 if (isa<SyntheticSection>(IS) || IS->Type == SHT_REL || 495 IS->Type == SHT_RELA) 496 continue; 497 498 auto *Sym = 499 make<DefinedRegular>("", /*IsLocal=*/true, /*StOther=*/0, STT_SECTION, 500 /*Value=*/0, /*Size=*/0, IS); 501 InX::SymTab->addSymbol(Sym); 502 } 503 } 504 505 // Today's loaders have a feature to make segments read-only after 506 // processing dynamic relocations to enhance security. PT_GNU_RELRO 507 // is defined for that. 508 // 509 // This function returns true if a section needs to be put into a 510 // PT_GNU_RELRO segment. 511 static bool isRelroSection(const OutputSection *Sec) { 512 if (!Config->ZRelro) 513 return false; 514 515 uint64_t Flags = Sec->Flags; 516 517 // Non-allocatable or non-writable sections don't need RELRO because 518 // they are not writable or not even mapped to memory in the first place. 519 // RELRO is for sections that are essentially read-only but need to 520 // be writable only at process startup to allow dynamic linker to 521 // apply relocations. 522 if (!(Flags & SHF_ALLOC) || !(Flags & SHF_WRITE)) 523 return false; 524 525 // Once initialized, TLS data segments are used as data templates 526 // for a thread-local storage. For each new thread, runtime 527 // allocates memory for a TLS and copy templates there. No thread 528 // are supposed to use templates directly. Thus, it can be in RELRO. 529 if (Flags & SHF_TLS) 530 return true; 531 532 // .init_array, .preinit_array and .fini_array contain pointers to 533 // functions that are executed on process startup or exit. These 534 // pointers are set by the static linker, and they are not expected 535 // to change at runtime. But if you are an attacker, you could do 536 // interesting things by manipulating pointers in .fini_array, for 537 // example. So they are put into RELRO. 538 uint32_t Type = Sec->Type; 539 if (Type == SHT_INIT_ARRAY || Type == SHT_FINI_ARRAY || 540 Type == SHT_PREINIT_ARRAY) 541 return true; 542 543 // .got contains pointers to external symbols. They are resolved by 544 // the dynamic linker when a module is loaded into memory, and after 545 // that they are not expected to change. So, it can be in RELRO. 546 if (InX::Got && Sec == InX::Got->getParent()) 547 return true; 548 549 // .got.plt contains pointers to external function symbols. They are 550 // by default resolved lazily, so we usually cannot put it into RELRO. 551 // However, if "-z now" is given, the lazy symbol resolution is 552 // disabled, which enables us to put it into RELRO. 553 if (Sec == InX::GotPlt->getParent()) 554 return Config->ZNow; 555 556 // .dynamic section contains data for the dynamic linker, and 557 // there's no need to write to it at runtime, so it's better to put 558 // it into RELRO. 559 if (Sec == InX::Dynamic->getParent()) 560 return true; 561 562 // .bss.rel.ro is used for copy relocations for read-only symbols. 563 // Since the dynamic linker needs to process copy relocations, the 564 // section cannot be read-only, but once initialized, they shouldn't 565 // change. 566 if (Sec == InX::BssRelRo->getParent()) 567 return true; 568 569 // Sections with some special names are put into RELRO. This is a 570 // bit unfortunate because section names shouldn't be significant in 571 // ELF in spirit. But in reality many linker features depend on 572 // magic section names. 573 StringRef S = Sec->Name; 574 return S == ".data.rel.ro" || S == ".ctors" || S == ".dtors" || S == ".jcr" || 575 S == ".eh_frame" || S == ".openbsd.randomdata"; 576 } 577 578 // We compute a rank for each section. The rank indicates where the 579 // section should be placed in the file. Instead of using simple 580 // numbers (0,1,2...), we use a series of flags. One for each decision 581 // point when placing the section. 582 // Using flags has two key properties: 583 // * It is easy to check if a give branch was taken. 584 // * It is easy two see how similar two ranks are (see getRankProximity). 585 enum RankFlags { 586 RF_NOT_ADDR_SET = 1 << 16, 587 RF_NOT_INTERP = 1 << 15, 588 RF_NOT_ALLOC = 1 << 14, 589 RF_WRITE = 1 << 13, 590 RF_EXEC_WRITE = 1 << 12, 591 RF_EXEC = 1 << 11, 592 RF_NON_TLS_BSS = 1 << 10, 593 RF_NON_TLS_BSS_RO = 1 << 9, 594 RF_NOT_TLS = 1 << 8, 595 RF_BSS = 1 << 7, 596 RF_PPC_NOT_TOCBSS = 1 << 6, 597 RF_PPC_OPD = 1 << 5, 598 RF_PPC_TOCL = 1 << 4, 599 RF_PPC_TOC = 1 << 3, 600 RF_PPC_BRANCH_LT = 1 << 2, 601 RF_MIPS_GPREL = 1 << 1, 602 RF_MIPS_NOT_GOT = 1 << 0 603 }; 604 605 static unsigned getSectionRank(const OutputSection *Sec) { 606 unsigned Rank = 0; 607 608 // We want to put section specified by -T option first, so we 609 // can start assigning VA starting from them later. 610 if (Config->SectionStartMap.count(Sec->Name)) 611 return Rank; 612 Rank |= RF_NOT_ADDR_SET; 613 614 // Put .interp first because some loaders want to see that section 615 // on the first page of the executable file when loaded into memory. 616 if (Sec->Name == ".interp") 617 return Rank; 618 Rank |= RF_NOT_INTERP; 619 620 // Allocatable sections go first to reduce the total PT_LOAD size and 621 // so debug info doesn't change addresses in actual code. 622 if (!(Sec->Flags & SHF_ALLOC)) 623 return Rank | RF_NOT_ALLOC; 624 625 // Sort sections based on their access permission in the following 626 // order: R, RX, RWX, RW. This order is based on the following 627 // considerations: 628 // * Read-only sections come first such that they go in the 629 // PT_LOAD covering the program headers at the start of the file. 630 // * Read-only, executable sections come next, unless the 631 // -no-rosegment option is used. 632 // * Writable, executable sections follow such that .plt on 633 // architectures where it needs to be writable will be placed 634 // between .text and .data. 635 // * Writable sections come last, such that .bss lands at the very 636 // end of the last PT_LOAD. 637 bool IsExec = Sec->Flags & SHF_EXECINSTR; 638 bool IsWrite = Sec->Flags & SHF_WRITE; 639 640 if (IsExec) { 641 if (IsWrite) 642 Rank |= RF_EXEC_WRITE; 643 else if (!Config->SingleRoRx) 644 Rank |= RF_EXEC; 645 } else { 646 if (IsWrite) 647 Rank |= RF_WRITE; 648 } 649 650 // If we got here we know that both A and B are in the same PT_LOAD. 651 652 bool IsTls = Sec->Flags & SHF_TLS; 653 bool IsNoBits = Sec->Type == SHT_NOBITS; 654 655 // The first requirement we have is to put (non-TLS) nobits sections last. The 656 // reason is that the only thing the dynamic linker will see about them is a 657 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the 658 // PT_LOAD, so that has to correspond to the nobits sections. 659 bool IsNonTlsNoBits = IsNoBits && !IsTls; 660 if (IsNonTlsNoBits) 661 Rank |= RF_NON_TLS_BSS; 662 663 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo 664 // sections after r/w ones, so that the RelRo sections are contiguous. 665 bool IsRelRo = isRelroSection(Sec); 666 if (IsNonTlsNoBits && !IsRelRo) 667 Rank |= RF_NON_TLS_BSS_RO; 668 if (!IsNonTlsNoBits && IsRelRo) 669 Rank |= RF_NON_TLS_BSS_RO; 670 671 // The TLS initialization block needs to be a single contiguous block in a R/W 672 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W 673 // sections. The TLS NOBITS sections are placed here as they don't take up 674 // virtual address space in the PT_LOAD. 675 if (!IsTls) 676 Rank |= RF_NOT_TLS; 677 678 // Within the TLS initialization block, the non-nobits sections need to appear 679 // first. 680 if (IsNoBits) 681 Rank |= RF_BSS; 682 683 // Some architectures have additional ordering restrictions for sections 684 // within the same PT_LOAD. 685 if (Config->EMachine == EM_PPC64) { 686 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections 687 // that we would like to make sure appear is a specific order to maximize 688 // their coverage by a single signed 16-bit offset from the TOC base 689 // pointer. Conversely, the special .tocbss section should be first among 690 // all SHT_NOBITS sections. This will put it next to the loaded special 691 // PPC64 sections (and, thus, within reach of the TOC base pointer). 692 StringRef Name = Sec->Name; 693 if (Name != ".tocbss") 694 Rank |= RF_PPC_NOT_TOCBSS; 695 696 if (Name == ".opd") 697 Rank |= RF_PPC_OPD; 698 699 if (Name == ".toc1") 700 Rank |= RF_PPC_TOCL; 701 702 if (Name == ".toc") 703 Rank |= RF_PPC_TOC; 704 705 if (Name == ".branch_lt") 706 Rank |= RF_PPC_BRANCH_LT; 707 } 708 if (Config->EMachine == EM_MIPS) { 709 // All sections with SHF_MIPS_GPREL flag should be grouped together 710 // because data in these sections is addressable with a gp relative address. 711 if (Sec->Flags & SHF_MIPS_GPREL) 712 Rank |= RF_MIPS_GPREL; 713 714 if (Sec->Name != ".got") 715 Rank |= RF_MIPS_NOT_GOT; 716 } 717 718 return Rank; 719 } 720 721 static bool compareSections(const BaseCommand *ACmd, const BaseCommand *BCmd) { 722 const OutputSection *A = cast<OutputSection>(ACmd); 723 const OutputSection *B = cast<OutputSection>(BCmd); 724 if (A->SortRank != B->SortRank) 725 return A->SortRank < B->SortRank; 726 if (!(A->SortRank & RF_NOT_ADDR_SET)) 727 return Config->SectionStartMap.lookup(A->Name) < 728 Config->SectionStartMap.lookup(B->Name); 729 return false; 730 } 731 732 void PhdrEntry::add(OutputSection *Sec) { 733 LastSec = Sec; 734 if (!FirstSec) 735 FirstSec = Sec; 736 p_align = std::max(p_align, Sec->Alignment); 737 if (p_type == PT_LOAD) 738 Sec->PtLoad = this; 739 } 740 741 template <class ELFT> 742 static Symbol *addRegular(StringRef Name, SectionBase *Sec, uint64_t Value, 743 uint8_t StOther = STV_HIDDEN, 744 uint8_t Binding = STB_WEAK) { 745 // The linker generated symbols are added as STB_WEAK to allow user defined 746 // ones to override them. 747 return Symtab->addRegular<ELFT>(Name, StOther, STT_NOTYPE, Value, 748 /*Size=*/0, Binding, Sec, 749 /*File=*/nullptr); 750 } 751 752 template <class ELFT> 753 static DefinedRegular * 754 addOptionalRegular(StringRef Name, SectionBase *Sec, uint64_t Val, 755 uint8_t StOther = STV_HIDDEN, uint8_t Binding = STB_GLOBAL) { 756 SymbolBody *S = Symtab->find(Name); 757 if (!S) 758 return nullptr; 759 if (S->isInCurrentDSO()) 760 return nullptr; 761 return cast<DefinedRegular>( 762 addRegular<ELFT>(Name, Sec, Val, StOther, Binding)->body()); 763 } 764 765 // The beginning and the ending of .rel[a].plt section are marked 766 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked 767 // executable. The runtime needs these symbols in order to resolve 768 // all IRELATIVE relocs on startup. For dynamic executables, we don't 769 // need these symbols, since IRELATIVE relocs are resolved through GOT 770 // and PLT. For details, see http://www.airs.com/blog/archives/403. 771 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() { 772 if (!Config->Static) 773 return; 774 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start"; 775 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, 0, STV_HIDDEN, STB_WEAK); 776 777 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end"; 778 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, -1, STV_HIDDEN, STB_WEAK); 779 } 780 781 // The linker is expected to define some symbols depending on 782 // the linking result. This function defines such symbols. 783 template <class ELFT> void Writer<ELFT>::addReservedSymbols() { 784 if (Config->EMachine == EM_MIPS) { 785 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer 786 // so that it points to an absolute address which by default is relative 787 // to GOT. Default offset is 0x7ff0. 788 // See "Global Data Symbols" in Chapter 6 in the following document: 789 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf 790 ElfSym::MipsGp = Symtab->addAbsolute<ELFT>("_gp", STV_HIDDEN, STB_LOCAL); 791 792 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between 793 // start of function and 'gp' pointer into GOT. 794 if (Symtab->find("_gp_disp")) 795 ElfSym::MipsGpDisp = 796 Symtab->addAbsolute<ELFT>("_gp_disp", STV_HIDDEN, STB_LOCAL); 797 798 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp' 799 // pointer. This symbol is used in the code generated by .cpload pseudo-op 800 // in case of using -mno-shared option. 801 // https://sourceware.org/ml/binutils/2004-12/msg00094.html 802 if (Symtab->find("__gnu_local_gp")) 803 ElfSym::MipsLocalGp = 804 Symtab->addAbsolute<ELFT>("__gnu_local_gp", STV_HIDDEN, STB_LOCAL); 805 } 806 807 // The _GLOBAL_OFFSET_TABLE_ symbol is defined by target convention to 808 // be at some offset from the base of the .got section, usually 0 or the end 809 // of the .got 810 InputSection *GotSection = InX::MipsGot ? cast<InputSection>(InX::MipsGot) 811 : cast<InputSection>(InX::Got); 812 ElfSym::GlobalOffsetTable = addOptionalRegular<ELFT>( 813 "_GLOBAL_OFFSET_TABLE_", GotSection, Target->GotBaseSymOff); 814 815 // __tls_get_addr is defined by the dynamic linker for dynamic ELFs. For 816 // static linking the linker is required to optimize away any references to 817 // __tls_get_addr, so it's not defined anywhere. Create a hidden definition 818 // to avoid the undefined symbol error. 819 if (!InX::DynSymTab) 820 Symtab->addIgnored<ELFT>("__tls_get_addr"); 821 822 // __ehdr_start is the location of ELF file headers. Note that we define 823 // this symbol unconditionally even when using a linker script, which 824 // differs from the behavior implemented by GNU linker which only define 825 // this symbol if ELF headers are in the memory mapped segment. 826 // __executable_start is not documented, but the expectation of at 827 // least the android libc is that it points to the elf header too. 828 // __dso_handle symbol is passed to cxa_finalize as a marker to identify 829 // each DSO. The address of the symbol doesn't matter as long as they are 830 // different in different DSOs, so we chose the start address of the DSO. 831 for (const char *Name : 832 {"__ehdr_start", "__executable_start", "__dso_handle"}) 833 addOptionalRegular<ELFT>(Name, Out::ElfHeader, 0, STV_HIDDEN); 834 835 // If linker script do layout we do not need to create any standart symbols. 836 if (Script->Opt.HasSections) 837 return; 838 839 auto Add = [](StringRef S, int64_t Pos) { 840 return addOptionalRegular<ELFT>(S, Out::ElfHeader, Pos, STV_DEFAULT); 841 }; 842 843 ElfSym::Bss = Add("__bss_start", 0); 844 ElfSym::End1 = Add("end", -1); 845 ElfSym::End2 = Add("_end", -1); 846 ElfSym::Etext1 = Add("etext", -1); 847 ElfSym::Etext2 = Add("_etext", -1); 848 ElfSym::Edata1 = Add("edata", -1); 849 ElfSym::Edata2 = Add("_edata", -1); 850 } 851 852 // Sort input sections by section name suffixes for 853 // __attribute__((init_priority(N))). 854 static void sortInitFini(OutputSection *Cmd) { 855 if (Cmd) 856 Cmd->sortInitFini(); 857 } 858 859 // Sort input sections by the special rule for .ctors and .dtors. 860 static void sortCtorsDtors(OutputSection *Cmd) { 861 if (Cmd) 862 Cmd->sortCtorsDtors(); 863 } 864 865 // Sort input sections using the list provided by --symbol-ordering-file. 866 static void sortBySymbolsOrder() { 867 if (Config->SymbolOrderingFile.empty()) 868 return; 869 870 // Sort sections by priority. 871 DenseMap<SectionBase *, int> SectionOrder = buildSectionOrder(); 872 for (BaseCommand *Base : Script->Opt.Commands) 873 if (auto *Sec = dyn_cast<OutputSection>(Base)) 874 Sec->sort([&](InputSectionBase *S) { return SectionOrder.lookup(S); }); 875 } 876 877 template <class ELFT> 878 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) { 879 // Scan all relocations. Each relocation goes through a series 880 // of tests to determine if it needs special treatment, such as 881 // creating GOT, PLT, copy relocations, etc. 882 // Note that relocations for non-alloc sections are directly 883 // processed by InputSection::relocateNonAlloc. 884 for (InputSectionBase *IS : InputSections) 885 if (IS->Live && isa<InputSection>(IS) && (IS->Flags & SHF_ALLOC)) 886 Fn(*IS); 887 for (EhInputSection *ES : In<ELFT>::EhFrame->Sections) 888 Fn(*ES); 889 } 890 891 template <class ELFT> void Writer<ELFT>::createSections() { 892 std::vector<BaseCommand *> Old = Script->Opt.Commands; 893 Script->Opt.Commands.clear(); 894 for (InputSectionBase *IS : InputSections) 895 if (IS) 896 Factory.addInputSec(IS, getOutputSectionName(IS->Name)); 897 Script->Opt.Commands.insert(Script->Opt.Commands.end(), Old.begin(), 898 Old.end()); 899 900 Script->fabricateDefaultCommands(); 901 sortBySymbolsOrder(); 902 sortInitFini(findSection(".init_array")); 903 sortInitFini(findSection(".fini_array")); 904 sortCtorsDtors(findSection(".ctors")); 905 sortCtorsDtors(findSection(".dtors")); 906 } 907 908 // This function generates assignments for predefined symbols (e.g. _end or 909 // _etext) and inserts them into the commands sequence to be processed at the 910 // appropriate time. This ensures that the value is going to be correct by the 911 // time any references to these symbols are processed and is equivalent to 912 // defining these symbols explicitly in the linker script. 913 template <class ELFT> void Writer<ELFT>::setReservedSymbolSections() { 914 PhdrEntry *Last = nullptr; 915 PhdrEntry *LastRO = nullptr; 916 PhdrEntry *LastRW = nullptr; 917 918 for (PhdrEntry *P : Phdrs) { 919 if (P->p_type != PT_LOAD) 920 continue; 921 Last = P; 922 if (P->p_flags & PF_W) 923 LastRW = P; 924 else 925 LastRO = P; 926 } 927 928 // _end is the first location after the uninitialized data region. 929 if (Last) { 930 if (ElfSym::End1) 931 ElfSym::End1->Section = Last->LastSec; 932 if (ElfSym::End2) 933 ElfSym::End2->Section = Last->LastSec; 934 } 935 936 // _etext is the first location after the last read-only loadable segment. 937 if (LastRO) { 938 if (ElfSym::Etext1) 939 ElfSym::Etext1->Section = LastRO->LastSec; 940 if (ElfSym::Etext2) 941 ElfSym::Etext2->Section = LastRO->LastSec; 942 } 943 944 // _edata points to the end of the last non SHT_NOBITS section. 945 if (LastRW) { 946 size_t I = 0; 947 for (; I < OutputSections.size(); ++I) 948 if (OutputSections[I] == LastRW->FirstSec) 949 break; 950 951 for (; I < OutputSections.size(); ++I) { 952 if (OutputSections[I]->Type != SHT_NOBITS) 953 continue; 954 break; 955 } 956 if (ElfSym::Edata1) 957 ElfSym::Edata1->Section = OutputSections[I - 1]; 958 if (ElfSym::Edata2) 959 ElfSym::Edata2->Section = OutputSections[I - 1]; 960 } 961 962 if (ElfSym::Bss) 963 ElfSym::Bss->Section = findSection(".bss"); 964 965 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should 966 // be equal to the _gp symbol's value. 967 if (ElfSym::MipsGp && !ElfSym::MipsGp->Value) { 968 // Find GP-relative section with the lowest address 969 // and use this address to calculate default _gp value. 970 for (OutputSection *OS : OutputSections) { 971 if (OS->Flags & SHF_MIPS_GPREL) { 972 ElfSym::MipsGp->Section = OS; 973 ElfSym::MipsGp->Value = 0x7ff0; 974 break; 975 } 976 } 977 } 978 } 979 980 // We want to find how similar two ranks are. 981 // The more branches in getSectionRank that match, the more similar they are. 982 // Since each branch corresponds to a bit flag, we can just use 983 // countLeadingZeros. 984 static int getRankProximityAux(OutputSection *A, OutputSection *B) { 985 return countLeadingZeros(A->SortRank ^ B->SortRank); 986 } 987 988 static int getRankProximity(OutputSection *A, BaseCommand *B) { 989 if (auto *Sec = dyn_cast<OutputSection>(B)) 990 if (Sec->Live) 991 return getRankProximityAux(A, Sec); 992 return -1; 993 } 994 995 // When placing orphan sections, we want to place them after symbol assignments 996 // so that an orphan after 997 // begin_foo = .; 998 // foo : { *(foo) } 999 // end_foo = .; 1000 // doesn't break the intended meaning of the begin/end symbols. 1001 // We don't want to go over sections since findOrphanPos is the 1002 // one in charge of deciding the order of the sections. 1003 // We don't want to go over changes to '.', since doing so in 1004 // rx_sec : { *(rx_sec) } 1005 // . = ALIGN(0x1000); 1006 // /* The RW PT_LOAD starts here*/ 1007 // rw_sec : { *(rw_sec) } 1008 // would mean that the RW PT_LOAD would become unaligned. 1009 static bool shouldSkip(BaseCommand *Cmd) { 1010 if (isa<OutputSection>(Cmd)) 1011 return false; 1012 if (auto *Assign = dyn_cast<SymbolAssignment>(Cmd)) 1013 return Assign->Name != "."; 1014 return true; 1015 } 1016 1017 // We want to place orphan sections so that they share as much 1018 // characteristics with their neighbors as possible. For example, if 1019 // both are rw, or both are tls. 1020 template <typename ELFT> 1021 static std::vector<BaseCommand *>::iterator 1022 findOrphanPos(std::vector<BaseCommand *>::iterator B, 1023 std::vector<BaseCommand *>::iterator E) { 1024 OutputSection *Sec = cast<OutputSection>(*E); 1025 1026 // Find the first element that has as close a rank as possible. 1027 auto I = std::max_element(B, E, [=](BaseCommand *A, BaseCommand *B) { 1028 return getRankProximity(Sec, A) < getRankProximity(Sec, B); 1029 }); 1030 if (I == E) 1031 return E; 1032 1033 // Consider all existing sections with the same proximity. 1034 int Proximity = getRankProximity(Sec, *I); 1035 for (; I != E; ++I) { 1036 auto *CurSec = dyn_cast<OutputSection>(*I); 1037 if (!CurSec || !CurSec->Live) 1038 continue; 1039 if (getRankProximity(Sec, CurSec) != Proximity || 1040 Sec->SortRank < CurSec->SortRank) 1041 break; 1042 } 1043 auto J = std::find_if( 1044 llvm::make_reverse_iterator(I), llvm::make_reverse_iterator(B), 1045 [](BaseCommand *Cmd) { return isa<OutputSection>(Cmd); }); 1046 I = J.base(); 1047 1048 // As a special case, if the orphan section is the last section, put 1049 // it at the very end, past any other commands. 1050 // This matches bfd's behavior and is convenient when the linker script fully 1051 // specifies the start of the file, but doesn't care about the end (the non 1052 // alloc sections for example). 1053 auto NextSec = std::find_if( 1054 I, E, [](BaseCommand *Cmd) { return isa<OutputSection>(Cmd); }); 1055 if (NextSec == E) 1056 return E; 1057 1058 while (I != E && shouldSkip(*I)) 1059 ++I; 1060 return I; 1061 } 1062 1063 template <class ELFT> void Writer<ELFT>::sortSections() { 1064 if (Script->Opt.HasSections) 1065 Script->adjustSectionsBeforeSorting(); 1066 1067 // Don't sort if using -r. It is not necessary and we want to preserve the 1068 // relative order for SHF_LINK_ORDER sections. 1069 if (Config->Relocatable) 1070 return; 1071 1072 for (BaseCommand *Base : Script->Opt.Commands) 1073 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1074 Sec->SortRank = getSectionRank(Sec); 1075 1076 if (!Script->Opt.HasSections) { 1077 // We know that all the OutputSections are contiguous in 1078 // this case. 1079 auto E = Script->Opt.Commands.end(); 1080 auto I = Script->Opt.Commands.begin(); 1081 auto IsSection = [](BaseCommand *Base) { return isa<OutputSection>(Base); }; 1082 I = std::find_if(I, E, IsSection); 1083 E = std::find_if(llvm::make_reverse_iterator(E), 1084 llvm::make_reverse_iterator(I), IsSection) 1085 .base(); 1086 std::stable_sort(I, E, compareSections); 1087 return; 1088 } 1089 1090 // Orphan sections are sections present in the input files which are 1091 // not explicitly placed into the output file by the linker script. 1092 // 1093 // The sections in the linker script are already in the correct 1094 // order. We have to figuere out where to insert the orphan 1095 // sections. 1096 // 1097 // The order of the sections in the script is arbitrary and may not agree with 1098 // compareSections. This means that we cannot easily define a strict weak 1099 // ordering. To see why, consider a comparison of a section in the script and 1100 // one not in the script. We have a two simple options: 1101 // * Make them equivalent (a is not less than b, and b is not less than a). 1102 // The problem is then that equivalence has to be transitive and we can 1103 // have sections a, b and c with only b in a script and a less than c 1104 // which breaks this property. 1105 // * Use compareSectionsNonScript. Given that the script order doesn't have 1106 // to match, we can end up with sections a, b, c, d where b and c are in the 1107 // script and c is compareSectionsNonScript less than b. In which case d 1108 // can be equivalent to c, a to b and d < a. As a concrete example: 1109 // .a (rx) # not in script 1110 // .b (rx) # in script 1111 // .c (ro) # in script 1112 // .d (ro) # not in script 1113 // 1114 // The way we define an order then is: 1115 // * Sort only the orphan sections. They are in the end right now. 1116 // * Move each orphan section to its preferred position. We try 1117 // to put each section in the last position where it it can share 1118 // a PT_LOAD. 1119 // 1120 // There is some ambiguity as to where exactly a new entry should be 1121 // inserted, because Opt.Commands contains not only output section 1122 // commands but also other types of commands such as symbol assignment 1123 // expressions. There's no correct answer here due to the lack of the 1124 // formal specification of the linker script. We use heuristics to 1125 // determine whether a new output command should be added before or 1126 // after another commands. For the details, look at shouldSkip 1127 // function. 1128 1129 auto I = Script->Opt.Commands.begin(); 1130 auto E = Script->Opt.Commands.end(); 1131 auto NonScriptI = std::find_if(I, E, [](BaseCommand *Base) { 1132 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1133 return Sec->Live && Sec->SectionIndex == INT_MAX; 1134 return false; 1135 }); 1136 1137 // Sort the orphan sections. 1138 std::stable_sort(NonScriptI, E, compareSections); 1139 1140 // As a horrible special case, skip the first . assignment if it is before any 1141 // section. We do this because it is common to set a load address by starting 1142 // the script with ". = 0xabcd" and the expectation is that every section is 1143 // after that. 1144 auto FirstSectionOrDotAssignment = 1145 std::find_if(I, E, [](BaseCommand *Cmd) { return !shouldSkip(Cmd); }); 1146 if (FirstSectionOrDotAssignment != E && 1147 isa<SymbolAssignment>(**FirstSectionOrDotAssignment)) 1148 ++FirstSectionOrDotAssignment; 1149 I = FirstSectionOrDotAssignment; 1150 1151 while (NonScriptI != E) { 1152 auto Pos = findOrphanPos<ELFT>(I, NonScriptI); 1153 OutputSection *Orphan = cast<OutputSection>(*NonScriptI); 1154 1155 // As an optimization, find all sections with the same sort rank 1156 // and insert them with one rotate. 1157 unsigned Rank = Orphan->SortRank; 1158 auto End = std::find_if(NonScriptI + 1, E, [=](BaseCommand *Cmd) { 1159 return cast<OutputSection>(Cmd)->SortRank != Rank; 1160 }); 1161 std::rotate(Pos, NonScriptI, End); 1162 NonScriptI = End; 1163 } 1164 1165 Script->adjustSectionsAfterSorting(); 1166 } 1167 1168 static void applySynthetic(const std::vector<SyntheticSection *> &Sections, 1169 std::function<void(SyntheticSection *)> Fn) { 1170 for (SyntheticSection *SS : Sections) 1171 if (SS && SS->getParent() && !SS->empty()) 1172 Fn(SS); 1173 } 1174 1175 // We need to add input synthetic sections early in createSyntheticSections() 1176 // to make them visible from linkescript side. But not all sections are always 1177 // required to be in output. For example we don't need dynamic section content 1178 // sometimes. This function filters out such unused sections from the output. 1179 static void removeUnusedSyntheticSections() { 1180 // All input synthetic sections that can be empty are placed after 1181 // all regular ones. We iterate over them all and exit at first 1182 // non-synthetic. 1183 for (InputSectionBase *S : llvm::reverse(InputSections)) { 1184 SyntheticSection *SS = dyn_cast<SyntheticSection>(S); 1185 if (!SS) 1186 return; 1187 OutputSection *OS = SS->getParent(); 1188 if (!SS->empty() || !OS) 1189 continue; 1190 if ((SS == InX::Got || SS == InX::MipsGot) && ElfSym::GlobalOffsetTable) 1191 continue; 1192 1193 std::vector<BaseCommand *>::iterator Empty = OS->Commands.end(); 1194 for (auto I = OS->Commands.begin(), E = OS->Commands.end(); I != E; ++I) { 1195 BaseCommand *B = *I; 1196 if (auto *ISD = dyn_cast<InputSectionDescription>(B)) { 1197 llvm::erase_if(ISD->Sections, 1198 [=](InputSection *IS) { return IS == SS; }); 1199 if (ISD->Sections.empty()) 1200 Empty = I; 1201 } 1202 } 1203 if (Empty != OS->Commands.end()) 1204 OS->Commands.erase(Empty); 1205 1206 // If there are no other sections in the output section, remove it from the 1207 // output. 1208 if (OS->Commands.empty()) 1209 llvm::erase_if(Script->Opt.Commands, 1210 [&](BaseCommand *Cmd) { return Cmd == OS; }); 1211 } 1212 } 1213 1214 // Returns true if a symbol can be replaced at load-time by a symbol 1215 // with the same name defined in other ELF executable or DSO. 1216 static bool computeIsPreemptible(const SymbolBody &B) { 1217 assert(!B.isLocal()); 1218 // Only symbols that appear in dynsym can be preempted. 1219 if (!B.symbol()->includeInDynsym()) 1220 return false; 1221 1222 // Only default visibility symbols can be preempted. 1223 if (B.symbol()->Visibility != STV_DEFAULT) 1224 return false; 1225 1226 // At this point copy relocations have not been created yet, so any 1227 // symbol that is not defined locally is preemptible. 1228 if (!B.isInCurrentDSO()) 1229 return true; 1230 1231 // If we have a dynamic list it specifies which local symbols are preemptible. 1232 if (Config->HasDynamicList) 1233 return false; 1234 1235 if (!Config->Shared) 1236 return false; 1237 1238 // -Bsymbolic means that definitions are not preempted. 1239 if (Config->Bsymbolic || (Config->BsymbolicFunctions && B.isFunc())) 1240 return false; 1241 return true; 1242 } 1243 1244 // Create output section objects and add them to OutputSections. 1245 template <class ELFT> void Writer<ELFT>::finalizeSections() { 1246 Out::DebugInfo = findSection(".debug_info"); 1247 Out::PreinitArray = findSection(".preinit_array"); 1248 Out::InitArray = findSection(".init_array"); 1249 Out::FiniArray = findSection(".fini_array"); 1250 1251 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop 1252 // symbols for sections, so that the runtime can get the start and end 1253 // addresses of each section by section name. Add such symbols. 1254 if (!Config->Relocatable) { 1255 addStartEndSymbols(); 1256 for (BaseCommand *Base : Script->Opt.Commands) 1257 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1258 addStartStopSymbols(Sec); 1259 } 1260 1261 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type. 1262 // It should be okay as no one seems to care about the type. 1263 // Even the author of gold doesn't remember why gold behaves that way. 1264 // https://sourceware.org/ml/binutils/2002-03/msg00360.html 1265 if (InX::DynSymTab) 1266 addRegular<ELFT>("_DYNAMIC", InX::Dynamic, 0); 1267 1268 // Define __rel[a]_iplt_{start,end} symbols if needed. 1269 addRelIpltSymbols(); 1270 1271 // This responsible for splitting up .eh_frame section into 1272 // pieces. The relocation scan uses those pieces, so this has to be 1273 // earlier. 1274 applySynthetic({In<ELFT>::EhFrame}, 1275 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1276 1277 for (Symbol *S : Symtab->getSymbols()) 1278 S->body()->IsPreemptible |= computeIsPreemptible(*S->body()); 1279 1280 // Scan relocations. This must be done after every symbol is declared so that 1281 // we can correctly decide if a dynamic relocation is needed. 1282 if (!Config->Relocatable) 1283 forEachRelSec(scanRelocations<ELFT>); 1284 1285 if (InX::Plt && !InX::Plt->empty()) 1286 InX::Plt->addSymbols(); 1287 if (InX::Iplt && !InX::Iplt->empty()) 1288 InX::Iplt->addSymbols(); 1289 1290 // Now that we have defined all possible global symbols including linker- 1291 // synthesized ones. Visit all symbols to give the finishing touches. 1292 for (Symbol *S : Symtab->getSymbols()) { 1293 SymbolBody *Body = S->body(); 1294 1295 if (!includeInSymtab(*Body)) 1296 continue; 1297 if (InX::SymTab) 1298 InX::SymTab->addSymbol(Body); 1299 1300 if (InX::DynSymTab && S->includeInDynsym()) { 1301 InX::DynSymTab->addSymbol(Body); 1302 if (auto *SS = dyn_cast<SharedSymbol>(Body)) 1303 if (cast<SharedFile<ELFT>>(S->File)->isNeeded()) 1304 In<ELFT>::VerNeed->addSymbol(SS); 1305 } 1306 } 1307 1308 // Do not proceed if there was an undefined symbol. 1309 if (ErrorCount) 1310 return; 1311 1312 addPredefinedSections(); 1313 removeUnusedSyntheticSections(); 1314 1315 sortSections(); 1316 1317 // Now that we have the final list, create a list of all the 1318 // OutputSections for convenience. 1319 for (BaseCommand *Base : Script->Opt.Commands) 1320 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1321 OutputSections.push_back(Sec); 1322 1323 // Prefer command line supplied address over other constraints. 1324 for (OutputSection *Sec : OutputSections) { 1325 auto I = Config->SectionStartMap.find(Sec->Name); 1326 if (I != Config->SectionStartMap.end()) 1327 Sec->AddrExpr = [=] { return I->second; }; 1328 } 1329 1330 // This is a bit of a hack. A value of 0 means undef, so we set it 1331 // to 1 t make __ehdr_start defined. The section number is not 1332 // particularly relevant. 1333 Out::ElfHeader->SectionIndex = 1; 1334 1335 unsigned I = 1; 1336 for (OutputSection *Sec : OutputSections) { 1337 Sec->SectionIndex = I++; 1338 Sec->ShName = InX::ShStrTab->addString(Sec->Name); 1339 } 1340 1341 // Binary and relocatable output does not have PHDRS. 1342 // The headers have to be created before finalize as that can influence the 1343 // image base and the dynamic section on mips includes the image base. 1344 if (!Config->Relocatable && !Config->OFormatBinary) { 1345 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs(); 1346 addPtArmExid(Phdrs); 1347 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size(); 1348 } 1349 1350 // Some symbols are defined in term of program headers. Now that we 1351 // have the headers, we can find out which sections they point to. 1352 setReservedSymbolSections(); 1353 1354 // Dynamic section must be the last one in this list and dynamic 1355 // symbol table section (DynSymTab) must be the first one. 1356 applySynthetic({InX::DynSymTab, InX::Bss, 1357 InX::BssRelRo, InX::GnuHashTab, 1358 In<ELFT>::HashTab, InX::SymTab, 1359 InX::ShStrTab, InX::StrTab, 1360 In<ELFT>::VerDef, InX::DynStrTab, 1361 InX::Got, InX::MipsGot, 1362 InX::IgotPlt, InX::GotPlt, 1363 In<ELFT>::RelaDyn, In<ELFT>::RelaIplt, 1364 In<ELFT>::RelaPlt, InX::Plt, 1365 InX::Iplt, In<ELFT>::EhFrameHdr, 1366 In<ELFT>::VerSym, In<ELFT>::VerNeed, 1367 InX::Dynamic}, 1368 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1369 1370 if (!Script->Opt.HasSections && !Config->Relocatable) 1371 fixSectionAlignments(); 1372 1373 // Some architectures use small displacements for jump instructions. 1374 // It is linker's responsibility to create thunks containing long 1375 // jump instructions if jump targets are too far. Create thunks. 1376 if (Target->NeedsThunks) { 1377 // FIXME: only ARM Interworking and Mips LA25 Thunks are implemented, 1378 // these 1379 // do not require address information. To support range extension Thunks 1380 // we need to assign addresses so that we can tell if jump instructions 1381 // are out of range. This will need to turn into a loop that converges 1382 // when no more Thunks are added 1383 ThunkCreator TC; 1384 Script->assignAddresses(); 1385 if (TC.createThunks(OutputSections)) { 1386 applySynthetic({InX::MipsGot}, 1387 [](SyntheticSection *SS) { SS->updateAllocSize(); }); 1388 if (TC.createThunks(OutputSections)) 1389 fatal("All non-range thunks should be created in first call"); 1390 } 1391 } 1392 1393 // Fill other section headers. The dynamic table is finalized 1394 // at the end because some tags like RELSZ depend on result 1395 // of finalizing other sections. 1396 for (OutputSection *Sec : OutputSections) 1397 Sec->finalize<ELFT>(); 1398 1399 // createThunks may have added local symbols to the static symbol table 1400 applySynthetic({InX::SymTab, InX::ShStrTab, InX::StrTab}, 1401 [](SyntheticSection *SS) { SS->postThunkContents(); }); 1402 } 1403 1404 template <class ELFT> void Writer<ELFT>::addPredefinedSections() { 1405 // ARM ABI requires .ARM.exidx to be terminated by some piece of data. 1406 // We have the terminater synthetic section class. Add that at the end. 1407 OutputSection *Cmd = findSection(".ARM.exidx"); 1408 if (!Cmd || !Cmd->Live || Config->Relocatable) 1409 return; 1410 1411 auto *Sentinel = make<ARMExidxSentinelSection>(); 1412 Cmd->addSection(Sentinel); 1413 } 1414 1415 // The linker is expected to define SECNAME_start and SECNAME_end 1416 // symbols for a few sections. This function defines them. 1417 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() { 1418 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) { 1419 // These symbols resolve to the image base if the section does not exist. 1420 // A special value -1 indicates end of the section. 1421 if (OS) { 1422 addOptionalRegular<ELFT>(Start, OS, 0); 1423 addOptionalRegular<ELFT>(End, OS, -1); 1424 } else { 1425 if (Config->Pic) 1426 OS = Out::ElfHeader; 1427 addOptionalRegular<ELFT>(Start, OS, 0); 1428 addOptionalRegular<ELFT>(End, OS, 0); 1429 } 1430 }; 1431 1432 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray); 1433 Define("__init_array_start", "__init_array_end", Out::InitArray); 1434 Define("__fini_array_start", "__fini_array_end", Out::FiniArray); 1435 1436 if (OutputSection *Sec = findSection(".ARM.exidx")) 1437 Define("__exidx_start", "__exidx_end", Sec); 1438 } 1439 1440 // If a section name is valid as a C identifier (which is rare because of 1441 // the leading '.'), linkers are expected to define __start_<secname> and 1442 // __stop_<secname> symbols. They are at beginning and end of the section, 1443 // respectively. This is not requested by the ELF standard, but GNU ld and 1444 // gold provide the feature, and used by many programs. 1445 template <class ELFT> 1446 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) { 1447 StringRef S = Sec->Name; 1448 if (!isValidCIdentifier(S)) 1449 return; 1450 addOptionalRegular<ELFT>(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT); 1451 addOptionalRegular<ELFT>(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT); 1452 } 1453 1454 template <class ELFT> OutputSection *Writer<ELFT>::findSection(StringRef Name) { 1455 for (BaseCommand *Base : Script->Opt.Commands) 1456 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1457 if (Sec->Name == Name) 1458 return Sec; 1459 return nullptr; 1460 } 1461 1462 static bool needsPtLoad(OutputSection *Sec) { 1463 if (!(Sec->Flags & SHF_ALLOC)) 1464 return false; 1465 1466 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is 1467 // responsible for allocating space for them, not the PT_LOAD that 1468 // contains the TLS initialization image. 1469 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS) 1470 return false; 1471 return true; 1472 } 1473 1474 // Linker scripts are responsible for aligning addresses. Unfortunately, most 1475 // linker scripts are designed for creating two PT_LOADs only, one RX and one 1476 // RW. This means that there is no alignment in the RO to RX transition and we 1477 // cannot create a PT_LOAD there. 1478 static uint64_t computeFlags(uint64_t Flags) { 1479 if (Config->Omagic) 1480 return PF_R | PF_W | PF_X; 1481 if (Config->SingleRoRx && !(Flags & PF_W)) 1482 return Flags | PF_X; 1483 return Flags; 1484 } 1485 1486 // Decide which program headers to create and which sections to include in each 1487 // one. 1488 template <class ELFT> std::vector<PhdrEntry *> Writer<ELFT>::createPhdrs() { 1489 std::vector<PhdrEntry *> Ret; 1490 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * { 1491 Ret.push_back(make<PhdrEntry>(Type, Flags)); 1492 return Ret.back(); 1493 }; 1494 1495 // The first phdr entry is PT_PHDR which describes the program header itself. 1496 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders); 1497 1498 // PT_INTERP must be the second entry if exists. 1499 if (OutputSection *Cmd = findSection(".interp")) 1500 AddHdr(PT_INTERP, Cmd->getPhdrFlags())->add(Cmd); 1501 1502 // Add the first PT_LOAD segment for regular output sections. 1503 uint64_t Flags = computeFlags(PF_R); 1504 PhdrEntry *Load = AddHdr(PT_LOAD, Flags); 1505 1506 // Add the headers. We will remove them if they don't fit. 1507 Load->add(Out::ElfHeader); 1508 Load->add(Out::ProgramHeaders); 1509 1510 for (OutputSection *Sec : OutputSections) { 1511 if (!(Sec->Flags & SHF_ALLOC)) 1512 break; 1513 if (!needsPtLoad(Sec)) 1514 continue; 1515 1516 // Segments are contiguous memory regions that has the same attributes 1517 // (e.g. executable or writable). There is one phdr for each segment. 1518 // Therefore, we need to create a new phdr when the next section has 1519 // different flags or is loaded at a discontiguous address using AT linker 1520 // script command. 1521 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags()); 1522 if (Sec->LMAExpr || Flags != NewFlags) { 1523 Load = AddHdr(PT_LOAD, NewFlags); 1524 Flags = NewFlags; 1525 } 1526 1527 Load->add(Sec); 1528 } 1529 1530 // Add a TLS segment if any. 1531 PhdrEntry *TlsHdr = make<PhdrEntry>(PT_TLS, PF_R); 1532 for (OutputSection *Sec : OutputSections) 1533 if (Sec->Flags & SHF_TLS) 1534 TlsHdr->add(Sec); 1535 if (TlsHdr->FirstSec) 1536 Ret.push_back(TlsHdr); 1537 1538 // Add an entry for .dynamic. 1539 if (InX::DynSymTab) 1540 AddHdr(PT_DYNAMIC, InX::Dynamic->getParent()->getPhdrFlags()) 1541 ->add(InX::Dynamic->getParent()); 1542 1543 // PT_GNU_RELRO includes all sections that should be marked as 1544 // read-only by dynamic linker after proccessing relocations. 1545 PhdrEntry *RelRo = make<PhdrEntry>(PT_GNU_RELRO, PF_R); 1546 for (OutputSection *Sec : OutputSections) 1547 if (needsPtLoad(Sec) && isRelroSection(Sec)) 1548 RelRo->add(Sec); 1549 if (RelRo->FirstSec) 1550 Ret.push_back(RelRo); 1551 1552 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr. 1553 if (!In<ELFT>::EhFrame->empty() && In<ELFT>::EhFrameHdr && 1554 In<ELFT>::EhFrame->getParent() && In<ELFT>::EhFrameHdr->getParent()) 1555 AddHdr(PT_GNU_EH_FRAME, In<ELFT>::EhFrameHdr->getParent()->getPhdrFlags()) 1556 ->add(In<ELFT>::EhFrameHdr->getParent()); 1557 1558 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes 1559 // the dynamic linker fill the segment with random data. 1560 if (OutputSection *Cmd = findSection(".openbsd.randomdata")) 1561 AddHdr(PT_OPENBSD_RANDOMIZE, Cmd->getPhdrFlags())->add(Cmd); 1562 1563 // PT_GNU_STACK is a special section to tell the loader to make the 1564 // pages for the stack non-executable. If you really want an executable 1565 // stack, you can pass -z execstack, but that's not recommended for 1566 // security reasons. 1567 unsigned Perm; 1568 if (Config->ZExecstack) 1569 Perm = PF_R | PF_W | PF_X; 1570 else 1571 Perm = PF_R | PF_W; 1572 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize; 1573 1574 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable 1575 // is expected to perform W^X violations, such as calling mprotect(2) or 1576 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on 1577 // OpenBSD. 1578 if (Config->ZWxneeded) 1579 AddHdr(PT_OPENBSD_WXNEEDED, PF_X); 1580 1581 // Create one PT_NOTE per a group of contiguous .note sections. 1582 PhdrEntry *Note = nullptr; 1583 for (OutputSection *Sec : OutputSections) { 1584 if (Sec->Type == SHT_NOTE) { 1585 if (!Note || Sec->LMAExpr) 1586 Note = AddHdr(PT_NOTE, PF_R); 1587 Note->add(Sec); 1588 } else { 1589 Note = nullptr; 1590 } 1591 } 1592 return Ret; 1593 } 1594 1595 template <class ELFT> 1596 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry *> &Phdrs) { 1597 if (Config->EMachine != EM_ARM) 1598 return; 1599 auto I = llvm::find_if(OutputSections, [](OutputSection *Cmd) { 1600 return Cmd->Type == SHT_ARM_EXIDX; 1601 }); 1602 if (I == OutputSections.end()) 1603 return; 1604 1605 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME 1606 PhdrEntry *ARMExidx = make<PhdrEntry>(PT_ARM_EXIDX, PF_R); 1607 ARMExidx->add(*I); 1608 Phdrs.push_back(ARMExidx); 1609 } 1610 1611 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the 1612 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic 1613 // linker can set the permissions. 1614 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() { 1615 auto PageAlign = [](OutputSection *Cmd) { 1616 if (Cmd && !Cmd->AddrExpr) 1617 Cmd->AddrExpr = [=] { 1618 return alignTo(Script->getDot(), Config->MaxPageSize); 1619 }; 1620 }; 1621 1622 for (const PhdrEntry *P : Phdrs) 1623 if (P->p_type == PT_LOAD && P->FirstSec) 1624 PageAlign(P->FirstSec); 1625 1626 for (const PhdrEntry *P : Phdrs) { 1627 if (P->p_type != PT_GNU_RELRO) 1628 continue; 1629 if (P->FirstSec) 1630 PageAlign(P->FirstSec); 1631 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we 1632 // have to align it to a page. 1633 auto End = OutputSections.end(); 1634 auto I = std::find(OutputSections.begin(), End, P->LastSec); 1635 if (I == End || (I + 1) == End) 1636 continue; 1637 OutputSection *Cmd = (*(I + 1)); 1638 if (needsPtLoad(Cmd)) 1639 PageAlign(Cmd); 1640 } 1641 } 1642 1643 // Adjusts the file alignment for a given output section and returns 1644 // its new file offset. The file offset must be the same with its 1645 // virtual address (modulo the page size) so that the loader can load 1646 // executables without any address adjustment. 1647 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Cmd) { 1648 // If the section is not in a PT_LOAD, we just have to align it. 1649 if (!Cmd->PtLoad) 1650 return alignTo(Off, Cmd->Alignment); 1651 1652 OutputSection *First = Cmd->PtLoad->FirstSec; 1653 // The first section in a PT_LOAD has to have congruent offset and address 1654 // module the page size. 1655 if (Cmd == First) 1656 return alignTo(Off, std::max<uint64_t>(Cmd->Alignment, Config->MaxPageSize), 1657 Cmd->Addr); 1658 1659 // If two sections share the same PT_LOAD the file offset is calculated 1660 // using this formula: Off2 = Off1 + (VA2 - VA1). 1661 return First->Offset + Cmd->Addr - First->Addr; 1662 } 1663 1664 static uint64_t setOffset(OutputSection *Cmd, uint64_t Off) { 1665 if (Cmd->Type == SHT_NOBITS) { 1666 Cmd->Offset = Off; 1667 return Off; 1668 } 1669 1670 Off = getFileAlignment(Off, Cmd); 1671 Cmd->Offset = Off; 1672 return Off + Cmd->Size; 1673 } 1674 1675 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() { 1676 uint64_t Off = 0; 1677 for (OutputSection *Sec : OutputSections) 1678 if (Sec->Flags & SHF_ALLOC) 1679 Off = setOffset(Sec, Off); 1680 FileSize = alignTo(Off, Config->Wordsize); 1681 } 1682 1683 // Assign file offsets to output sections. 1684 template <class ELFT> void Writer<ELFT>::assignFileOffsets() { 1685 uint64_t Off = 0; 1686 Off = setOffset(Out::ElfHeader, Off); 1687 Off = setOffset(Out::ProgramHeaders, Off); 1688 1689 PhdrEntry *LastRX = nullptr; 1690 for (PhdrEntry *P : Phdrs) 1691 if (P->p_type == PT_LOAD && (P->p_flags & PF_X)) 1692 LastRX = P; 1693 1694 for (OutputSection *Sec : OutputSections) { 1695 Off = setOffset(Sec, Off); 1696 if (Script->Opt.HasSections) 1697 continue; 1698 // If this is a last section of the last executable segment and that 1699 // segment is the last loadable segment, align the offset of the 1700 // following section to avoid loading non-segments parts of the file. 1701 if (LastRX && LastRX->LastSec == Sec) 1702 Off = alignTo(Off, Target->PageSize); 1703 } 1704 1705 SectionHeaderOff = alignTo(Off, Config->Wordsize); 1706 FileSize = SectionHeaderOff + (OutputSections.size() + 1) * sizeof(Elf_Shdr); 1707 } 1708 1709 // Finalize the program headers. We call this function after we assign 1710 // file offsets and VAs to all sections. 1711 template <class ELFT> void Writer<ELFT>::setPhdrs() { 1712 for (PhdrEntry *P : Phdrs) { 1713 OutputSection *First = P->FirstSec; 1714 OutputSection *Last = P->LastSec; 1715 if (First) { 1716 P->p_filesz = Last->Offset - First->Offset; 1717 if (Last->Type != SHT_NOBITS) 1718 P->p_filesz += Last->Size; 1719 P->p_memsz = Last->Addr + Last->Size - First->Addr; 1720 P->p_offset = First->Offset; 1721 P->p_vaddr = First->Addr; 1722 if (!P->HasLMA) 1723 P->p_paddr = First->getLMA(); 1724 } 1725 if (P->p_type == PT_LOAD) 1726 P->p_align = std::max<uint64_t>(P->p_align, Config->MaxPageSize); 1727 else if (P->p_type == PT_GNU_RELRO) { 1728 P->p_align = 1; 1729 // The glibc dynamic loader rounds the size down, so we need to round up 1730 // to protect the last page. This is a no-op on FreeBSD which always 1731 // rounds up. 1732 P->p_memsz = alignTo(P->p_memsz, Target->PageSize); 1733 } 1734 1735 // The TLS pointer goes after PT_TLS. At least glibc will align it, 1736 // so round up the size to make sure the offsets are correct. 1737 if (P->p_type == PT_TLS) { 1738 Out::TlsPhdr = P; 1739 if (P->p_memsz) 1740 P->p_memsz = alignTo(P->p_memsz, P->p_align); 1741 } 1742 } 1743 } 1744 1745 // The entry point address is chosen in the following ways. 1746 // 1747 // 1. the '-e' entry command-line option; 1748 // 2. the ENTRY(symbol) command in a linker control script; 1749 // 3. the value of the symbol start, if present; 1750 // 4. the address of the first byte of the .text section, if present; 1751 // 5. the address 0. 1752 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() { 1753 // Case 1, 2 or 3. As a special case, if the symbol is actually 1754 // a number, we'll use that number as an address. 1755 if (SymbolBody *B = Symtab->find(Config->Entry)) 1756 return B->getVA(); 1757 uint64_t Addr; 1758 if (to_integer(Config->Entry, Addr)) 1759 return Addr; 1760 1761 // Case 4 1762 if (OutputSection *Sec = findSection(".text")) { 1763 if (Config->WarnMissingEntry) 1764 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" + 1765 utohexstr(Sec->Addr)); 1766 return Sec->Addr; 1767 } 1768 1769 // Case 5 1770 if (Config->WarnMissingEntry) 1771 warn("cannot find entry symbol " + Config->Entry + 1772 "; not setting start address"); 1773 return 0; 1774 } 1775 1776 static uint16_t getELFType() { 1777 if (Config->Pic) 1778 return ET_DYN; 1779 if (Config->Relocatable) 1780 return ET_REL; 1781 return ET_EXEC; 1782 } 1783 1784 template <class ELFT> void Writer<ELFT>::writeHeader() { 1785 uint8_t *Buf = Buffer->getBufferStart(); 1786 memcpy(Buf, "\177ELF", 4); 1787 1788 // Write the ELF header. 1789 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf); 1790 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32; 1791 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB; 1792 EHdr->e_ident[EI_VERSION] = EV_CURRENT; 1793 EHdr->e_ident[EI_OSABI] = Config->OSABI; 1794 EHdr->e_type = getELFType(); 1795 EHdr->e_machine = Config->EMachine; 1796 EHdr->e_version = EV_CURRENT; 1797 EHdr->e_entry = getEntryAddr(); 1798 EHdr->e_shoff = SectionHeaderOff; 1799 EHdr->e_ehsize = sizeof(Elf_Ehdr); 1800 EHdr->e_phnum = Phdrs.size(); 1801 EHdr->e_shentsize = sizeof(Elf_Shdr); 1802 EHdr->e_shnum = OutputSections.size() + 1; 1803 EHdr->e_shstrndx = InX::ShStrTab->getParent()->SectionIndex; 1804 1805 if (Config->EMachine == EM_ARM) 1806 // We don't currently use any features incompatible with EF_ARM_EABI_VER5, 1807 // but we don't have any firm guarantees of conformance. Linux AArch64 1808 // kernels (as of 2016) require an EABI version to be set. 1809 EHdr->e_flags = EF_ARM_EABI_VER5; 1810 else if (Config->EMachine == EM_MIPS) 1811 EHdr->e_flags = getMipsEFlags<ELFT>(); 1812 1813 if (!Config->Relocatable) { 1814 EHdr->e_phoff = sizeof(Elf_Ehdr); 1815 EHdr->e_phentsize = sizeof(Elf_Phdr); 1816 } 1817 1818 // Write the program header table. 1819 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff); 1820 for (PhdrEntry *P : Phdrs) { 1821 HBuf->p_type = P->p_type; 1822 HBuf->p_flags = P->p_flags; 1823 HBuf->p_offset = P->p_offset; 1824 HBuf->p_vaddr = P->p_vaddr; 1825 HBuf->p_paddr = P->p_paddr; 1826 HBuf->p_filesz = P->p_filesz; 1827 HBuf->p_memsz = P->p_memsz; 1828 HBuf->p_align = P->p_align; 1829 ++HBuf; 1830 } 1831 1832 // Write the section header table. Note that the first table entry is null. 1833 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff); 1834 for (OutputSection *Sec : OutputSections) 1835 Sec->writeHeaderTo<ELFT>(++SHdrs); 1836 } 1837 1838 // Open a result file. 1839 template <class ELFT> void Writer<ELFT>::openFile() { 1840 if (!Config->Is64 && FileSize > UINT32_MAX) { 1841 error("output file too large: " + Twine(FileSize) + " bytes"); 1842 return; 1843 } 1844 1845 unlinkAsync(Config->OutputFile); 1846 ErrorOr<std::unique_ptr<FileOutputBuffer>> BufferOrErr = 1847 FileOutputBuffer::create(Config->OutputFile, FileSize, 1848 FileOutputBuffer::F_executable); 1849 1850 if (auto EC = BufferOrErr.getError()) 1851 error("failed to open " + Config->OutputFile + ": " + EC.message()); 1852 else 1853 Buffer = std::move(*BufferOrErr); 1854 } 1855 1856 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() { 1857 uint8_t *Buf = Buffer->getBufferStart(); 1858 for (OutputSection *Sec : OutputSections) 1859 if (Sec->Flags & SHF_ALLOC) 1860 Sec->writeTo<ELFT>(Buf + Sec->Offset); 1861 } 1862 1863 static void fillTrap(uint8_t *I, uint8_t *End) { 1864 for (; I + 4 <= End; I += 4) 1865 memcpy(I, &Target->TrapInstr, 4); 1866 } 1867 1868 // Fill the last page of executable segments with trap instructions 1869 // instead of leaving them as zero. Even though it is not required by any 1870 // standard, it is in general a good thing to do for security reasons. 1871 // 1872 // We'll leave other pages in segments as-is because the rest will be 1873 // overwritten by output sections. 1874 template <class ELFT> void Writer<ELFT>::writeTrapInstr() { 1875 if (Script->Opt.HasSections) 1876 return; 1877 1878 // Fill the last page. 1879 uint8_t *Buf = Buffer->getBufferStart(); 1880 for (PhdrEntry *P : Phdrs) 1881 if (P->p_type == PT_LOAD && (P->p_flags & PF_X)) 1882 fillTrap(Buf + alignDown(P->p_offset + P->p_filesz, Target->PageSize), 1883 Buf + alignTo(P->p_offset + P->p_filesz, Target->PageSize)); 1884 1885 // Round up the file size of the last segment to the page boundary iff it is 1886 // an executable segment to ensure that other other tools don't accidentally 1887 // trim the instruction padding (e.g. when stripping the file). 1888 PhdrEntry *LastRX = nullptr; 1889 for (PhdrEntry *P : Phdrs) { 1890 if (P->p_type != PT_LOAD) 1891 continue; 1892 if (P->p_flags & PF_X) 1893 LastRX = P; 1894 else 1895 LastRX = nullptr; 1896 } 1897 if (LastRX) 1898 LastRX->p_memsz = LastRX->p_filesz = 1899 alignTo(LastRX->p_filesz, Target->PageSize); 1900 } 1901 1902 // Write section contents to a mmap'ed file. 1903 template <class ELFT> void Writer<ELFT>::writeSections() { 1904 uint8_t *Buf = Buffer->getBufferStart(); 1905 1906 // PPC64 needs to process relocations in the .opd section 1907 // before processing relocations in code-containing sections. 1908 if (auto *OpdCmd = findSection(".opd")) { 1909 Out::Opd = OpdCmd; 1910 Out::OpdBuf = Buf + Out::Opd->Offset; 1911 OpdCmd->template writeTo<ELFT>(Buf + Out::Opd->Offset); 1912 } 1913 1914 OutputSection *EhFrameHdr = 1915 (In<ELFT>::EhFrameHdr && !In<ELFT>::EhFrameHdr->empty()) 1916 ? In<ELFT>::EhFrameHdr->getParent() 1917 : nullptr; 1918 1919 // In -r or -emit-relocs mode, write the relocation sections first as in 1920 // ELf_Rel targets we might find out that we need to modify the relocated 1921 // section while doing it. 1922 for (OutputSection *Sec : OutputSections) 1923 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA) 1924 Sec->writeTo<ELFT>(Buf + Sec->Offset); 1925 1926 for (OutputSection *Sec : OutputSections) 1927 if (Sec != Out::Opd && Sec != EhFrameHdr && Sec->Type != SHT_REL && 1928 Sec->Type != SHT_RELA) 1929 Sec->writeTo<ELFT>(Buf + Sec->Offset); 1930 1931 // The .eh_frame_hdr depends on .eh_frame section contents, therefore 1932 // it should be written after .eh_frame is written. 1933 if (EhFrameHdr) 1934 EhFrameHdr->writeTo<ELFT>(Buf + EhFrameHdr->Offset); 1935 } 1936 1937 template <class ELFT> void Writer<ELFT>::writeBuildId() { 1938 if (!InX::BuildId || !InX::BuildId->getParent()) 1939 return; 1940 1941 // Compute a hash of all sections of the output file. 1942 uint8_t *Start = Buffer->getBufferStart(); 1943 uint8_t *End = Start + FileSize; 1944 InX::BuildId->writeBuildId({Start, End}); 1945 } 1946 1947 template void elf::writeResult<ELF32LE>(); 1948 template void elf::writeResult<ELF32BE>(); 1949 template void elf::writeResult<ELF64LE>(); 1950 template void elf::writeResult<ELF64BE>(); 1951