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 "OutputSections.h" 16 #include "Relocations.h" 17 #include "Strings.h" 18 #include "SymbolTable.h" 19 #include "SyntheticSections.h" 20 #include "Target.h" 21 #include "lld/Common/Memory.h" 22 #include "lld/Common/Threads.h" 23 #include "llvm/ADT/StringMap.h" 24 #include "llvm/ADT/StringSwitch.h" 25 #include <climits> 26 27 using namespace llvm; 28 using namespace llvm::ELF; 29 using namespace llvm::object; 30 using namespace llvm::support; 31 using namespace llvm::support::endian; 32 33 using namespace lld; 34 using namespace lld::elf; 35 36 namespace { 37 // The writer writes a SymbolTable result to a file. 38 template <class ELFT> class Writer { 39 public: 40 Writer() : Buffer(errorHandler().OutputBuffer) {} 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 forEachRelSec(std::function<void(InputSectionBase &)> Fn); 53 void sortSections(); 54 void sortInputSections(); 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 void addRelIpltSymbols(); 76 void addStartEndSymbols(); 77 void addStartStopSymbols(OutputSection *Sec); 78 uint64_t getEntryAddr(); 79 OutputSection *findSection(StringRef Name); 80 81 std::vector<PhdrEntry *> Phdrs; 82 83 uint64_t FileSize; 84 uint64_t SectionHeaderOff; 85 86 bool HasGotBaseSym = false; 87 }; 88 } // anonymous namespace 89 90 StringRef elf::getOutputSectionName(StringRef Name) { 91 // ".zdebug_" is a prefix for ZLIB-compressed sections. 92 // Because we decompressed input sections, we want to remove 'z'. 93 if (Name.startswith(".zdebug_")) 94 return Saver.save("." + Name.substr(2)); 95 96 if (Config->Relocatable) 97 return Name; 98 99 // This is for --emit-relocs. If .text.foo is emitted as .text, we want to 100 // emit .rela.text.foo as .rel.text for consistency (this is not technically 101 // required, but not doing it is odd). This code guarantees that. 102 if (Name.startswith(".rel.")) 103 return Saver.save(".rel" + getOutputSectionName(Name.substr(4))); 104 if (Name.startswith(".rela.")) 105 return Saver.save(".rela" + getOutputSectionName(Name.substr(5))); 106 107 for (StringRef V : 108 {".text.", ".rodata.", ".data.rel.ro.", ".data.", ".bss.rel.ro.", 109 ".bss.", ".init_array.", ".fini_array.", ".ctors.", ".dtors.", ".tbss.", 110 ".gcc_except_table.", ".tdata.", ".ARM.exidx.", ".ARM.extab."}) { 111 StringRef Prefix = V.drop_back(); 112 if (Name.startswith(V) || Name == Prefix) 113 return Prefix; 114 } 115 116 // CommonSection is identified as "COMMON" in linker scripts. 117 // By default, it should go to .bss section. 118 if (Name == "COMMON") 119 return ".bss"; 120 121 return Name; 122 } 123 124 static bool needsInterpSection() { 125 return !SharedFiles.empty() && !Config->DynamicLinker.empty() && 126 Script->needsInterpSection(); 127 } 128 129 template <class ELFT> void elf::writeResult() { Writer<ELFT>().run(); } 130 131 template <class ELFT> void Writer<ELFT>::removeEmptyPTLoad() { 132 llvm::erase_if(Phdrs, [&](const PhdrEntry *P) { 133 if (P->p_type != PT_LOAD) 134 return false; 135 if (!P->FirstSec) 136 return true; 137 uint64_t Size = P->LastSec->Addr + P->LastSec->Size - P->FirstSec->Addr; 138 return Size == 0; 139 }); 140 } 141 142 template <class ELFT> static void combineEhFrameSections() { 143 for (InputSectionBase *&S : InputSections) { 144 EhInputSection *ES = dyn_cast<EhInputSection>(S); 145 if (!ES || !ES->Live) 146 continue; 147 148 InX::EhFrame->addSection<ELFT>(ES); 149 S = nullptr; 150 } 151 152 std::vector<InputSectionBase *> &V = InputSections; 153 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end()); 154 } 155 156 // The main function of the writer. 157 template <class ELFT> void Writer<ELFT>::run() { 158 // Create linker-synthesized sections such as .got or .plt. 159 // Such sections are of type input section. 160 createSyntheticSections(); 161 162 if (!Config->Relocatable) 163 combineEhFrameSections<ELFT>(); 164 165 // We need to create some reserved symbols such as _end. Create them. 166 if (!Config->Relocatable) 167 addReservedSymbols(); 168 169 // We want to process linker script commands. When SECTIONS command 170 // is given we let it create sections. 171 Script->processSectionCommands(); 172 173 // Linker scripts controls how input sections are assigned to output sections. 174 // Input sections that were not handled by scripts are called "orphans", and 175 // they are assigned to output sections by the default rule. Process that. 176 Script->addOrphanSections(); 177 178 if (Config->Discard != DiscardPolicy::All) 179 copyLocalSymbols(); 180 181 if (Config->CopyRelocs) 182 addSectionSymbols(); 183 184 // Now that we have a complete set of output sections. This function 185 // completes section contents. For example, we need to add strings 186 // to the string table, and add entries to .got and .plt. 187 // finalizeSections does that. 188 finalizeSections(); 189 if (errorCount()) 190 return; 191 192 // If -compressed-debug-sections is specified, we need to compress 193 // .debug_* sections. Do it right now because it changes the size of 194 // output sections. 195 parallelForEach(OutputSections, 196 [](OutputSection *Sec) { Sec->maybeCompress<ELFT>(); }); 197 198 Script->assignAddresses(); 199 Script->allocateHeaders(Phdrs); 200 201 // Remove empty PT_LOAD to avoid causing the dynamic linker to try to mmap a 202 // 0 sized region. This has to be done late since only after assignAddresses 203 // we know the size of the sections. 204 removeEmptyPTLoad(); 205 206 if (!Config->OFormatBinary) 207 assignFileOffsets(); 208 else 209 assignFileOffsetsBinary(); 210 211 setPhdrs(); 212 213 if (Config->Relocatable) { 214 for (OutputSection *Sec : OutputSections) 215 Sec->Addr = 0; 216 } 217 218 // It does not make sense try to open the file if we have error already. 219 if (errorCount()) 220 return; 221 // Write the result down to a file. 222 openFile(); 223 if (errorCount()) 224 return; 225 226 if (!Config->OFormatBinary) { 227 writeTrapInstr(); 228 writeHeader(); 229 writeSections(); 230 } else { 231 writeSectionsBinary(); 232 } 233 234 // Backfill .note.gnu.build-id section content. This is done at last 235 // because the content is usually a hash value of the entire output file. 236 writeBuildId(); 237 if (errorCount()) 238 return; 239 240 // Handle -Map option. 241 writeMapFile(); 242 if (errorCount()) 243 return; 244 245 if (auto E = Buffer->commit()) 246 error("failed to write to the output file: " + toString(std::move(E))); 247 } 248 249 // Initialize Out members. 250 template <class ELFT> void Writer<ELFT>::createSyntheticSections() { 251 // Initialize all pointers with NULL. This is needed because 252 // you can call lld::elf::main more than once as a library. 253 memset(&Out::First, 0, sizeof(Out)); 254 255 auto Add = [](InputSectionBase *Sec) { InputSections.push_back(Sec); }; 256 257 InX::DynStrTab = make<StringTableSection>(".dynstr", true); 258 InX::Dynamic = make<DynamicSection<ELFT>>(); 259 if (Config->AndroidPackDynRelocs) { 260 In<ELFT>::RelaDyn = make<AndroidPackedRelocationSection<ELFT>>( 261 Config->IsRela ? ".rela.dyn" : ".rel.dyn"); 262 } else { 263 In<ELFT>::RelaDyn = make<RelocationSection<ELFT>>( 264 Config->IsRela ? ".rela.dyn" : ".rel.dyn", Config->ZCombreloc); 265 } 266 InX::ShStrTab = make<StringTableSection>(".shstrtab", false); 267 268 Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC); 269 Out::ElfHeader->Size = sizeof(Elf_Ehdr); 270 Out::ProgramHeaders = make<OutputSection>("", 0, SHF_ALLOC); 271 Out::ProgramHeaders->Alignment = Config->Wordsize; 272 273 if (needsInterpSection()) { 274 InX::Interp = createInterpSection(); 275 Add(InX::Interp); 276 } else { 277 InX::Interp = nullptr; 278 } 279 280 if (Config->Strip != StripPolicy::All) { 281 InX::StrTab = make<StringTableSection>(".strtab", false); 282 InX::SymTab = make<SymbolTableSection<ELFT>>(*InX::StrTab); 283 } 284 285 if (Config->BuildId != BuildIdKind::None) { 286 InX::BuildId = make<BuildIdSection>(); 287 Add(InX::BuildId); 288 } 289 290 InX::Bss = make<BssSection>(".bss", 0, 1); 291 Add(InX::Bss); 292 293 // If there is a SECTIONS command and a .data.rel.ro section name use name 294 // .data.rel.ro.bss so that we match in the .data.rel.ro output section. 295 // This makes sure our relro is contiguous. 296 bool HasDataRelRo = 297 Script->HasSectionsCommand && findSection(".data.rel.ro"); 298 InX::BssRelRo = make<BssSection>( 299 HasDataRelRo ? ".data.rel.ro.bss" : ".bss.rel.ro", 0, 1); 300 Add(InX::BssRelRo); 301 302 // Add MIPS-specific sections. 303 if (Config->EMachine == EM_MIPS) { 304 if (!Config->Shared && Config->HasDynSymTab) { 305 InX::MipsRldMap = make<MipsRldMapSection>(); 306 Add(InX::MipsRldMap); 307 } 308 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create()) 309 Add(Sec); 310 if (auto *Sec = MipsOptionsSection<ELFT>::create()) 311 Add(Sec); 312 if (auto *Sec = MipsReginfoSection<ELFT>::create()) 313 Add(Sec); 314 } 315 316 if (Config->HasDynSymTab) { 317 InX::DynSymTab = make<SymbolTableSection<ELFT>>(*InX::DynStrTab); 318 Add(InX::DynSymTab); 319 320 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>(); 321 Add(In<ELFT>::VerSym); 322 323 if (!Config->VersionDefinitions.empty()) { 324 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>(); 325 Add(In<ELFT>::VerDef); 326 } 327 328 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>(); 329 Add(In<ELFT>::VerNeed); 330 331 if (Config->GnuHash) { 332 InX::GnuHashTab = make<GnuHashTableSection>(); 333 Add(InX::GnuHashTab); 334 } 335 336 if (Config->SysvHash) { 337 InX::HashTab = make<HashTableSection>(); 338 Add(InX::HashTab); 339 } 340 341 Add(InX::Dynamic); 342 Add(InX::DynStrTab); 343 Add(In<ELFT>::RelaDyn); 344 } 345 346 // Add .got. MIPS' .got is so different from the other archs, 347 // it has its own class. 348 if (Config->EMachine == EM_MIPS) { 349 InX::MipsGot = make<MipsGotSection>(); 350 Add(InX::MipsGot); 351 } else { 352 InX::Got = make<GotSection>(); 353 Add(InX::Got); 354 } 355 356 InX::GotPlt = make<GotPltSection>(); 357 Add(InX::GotPlt); 358 InX::IgotPlt = make<IgotPltSection>(); 359 Add(InX::IgotPlt); 360 361 if (Config->GdbIndex) { 362 InX::GdbIndex = createGdbIndex<ELFT>(); 363 Add(InX::GdbIndex); 364 } 365 366 // We always need to add rel[a].plt to output if it has entries. 367 // Even for static linking it can contain R_[*]_IRELATIVE relocations. 368 In<ELFT>::RelaPlt = make<RelocationSection<ELFT>>( 369 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/); 370 Add(In<ELFT>::RelaPlt); 371 372 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure 373 // that the IRelative relocations are processed last by the dynamic loader. 374 // We cannot place the iplt section in .rel.dyn when Android relocation 375 // packing is enabled because that would cause a section type mismatch. 376 // However, because the Android dynamic loader reads .rel.plt after .rel.dyn, 377 // we can get the desired behaviour by placing the iplt section in .rel.plt. 378 In<ELFT>::RelaIplt = make<RelocationSection<ELFT>>( 379 (Config->EMachine == EM_ARM && !Config->AndroidPackDynRelocs) 380 ? ".rel.dyn" 381 : In<ELFT>::RelaPlt->Name, 382 false /*Sort*/); 383 Add(In<ELFT>::RelaIplt); 384 385 InX::Plt = make<PltSection>(Target->PltHeaderSize); 386 Add(InX::Plt); 387 InX::Iplt = make<PltSection>(0); 388 Add(InX::Iplt); 389 390 if (!Config->Relocatable) { 391 if (Config->EhFrameHdr) { 392 InX::EhFrameHdr = make<EhFrameHeader>(); 393 Add(InX::EhFrameHdr); 394 } 395 InX::EhFrame = make<EhFrameSection>(); 396 Add(InX::EhFrame); 397 } 398 399 if (InX::SymTab) 400 Add(InX::SymTab); 401 Add(InX::ShStrTab); 402 if (InX::StrTab) 403 Add(InX::StrTab); 404 } 405 406 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName, 407 const Symbol &B) { 408 if (B.isFile() || B.isSection()) 409 return false; 410 411 // If sym references a section in a discarded group, don't keep it. 412 if (Sec == &InputSection::Discarded) 413 return false; 414 415 if (Config->Discard == DiscardPolicy::None) 416 return true; 417 418 // In ELF assembly .L symbols are normally discarded by the assembler. 419 // If the assembler fails to do so, the linker discards them if 420 // * --discard-locals is used. 421 // * The symbol is in a SHF_MERGE section, which is normally the reason for 422 // the assembler keeping the .L symbol. 423 if (!SymName.startswith(".L") && !SymName.empty()) 424 return true; 425 426 if (Config->Discard == DiscardPolicy::Locals) 427 return false; 428 429 return !Sec || !(Sec->Flags & SHF_MERGE); 430 } 431 432 static bool includeInSymtab(const Symbol &B) { 433 if (!B.isLocal() && !B.IsUsedInRegularObj) 434 return false; 435 436 if (auto *D = dyn_cast<Defined>(&B)) { 437 // Always include absolute symbols. 438 SectionBase *Sec = D->Section; 439 if (!Sec) 440 return true; 441 if (auto *IS = dyn_cast<InputSectionBase>(Sec)) { 442 Sec = IS->Repl; 443 IS = cast<InputSectionBase>(Sec); 444 // Exclude symbols pointing to garbage-collected sections. 445 if (!IS->Live) 446 return false; 447 } 448 if (auto *S = dyn_cast<MergeInputSection>(Sec)) 449 if (!S->getSectionPiece(D->Value)->Live) 450 return false; 451 return true; 452 } 453 return B.Used; 454 } 455 456 // Local symbols are not in the linker's symbol table. This function scans 457 // each object file's symbol table to copy local symbols to the output. 458 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() { 459 if (!InX::SymTab) 460 return; 461 for (InputFile *File : ObjectFiles) { 462 ObjFile<ELFT> *F = cast<ObjFile<ELFT>>(File); 463 for (Symbol *B : F->getLocalSymbols()) { 464 if (!B->isLocal()) 465 fatal(toString(F) + 466 ": broken object: getLocalSymbols returns a non-local symbol"); 467 auto *DR = dyn_cast<Defined>(B); 468 469 // No reason to keep local undefined symbol in symtab. 470 if (!DR) 471 continue; 472 if (!includeInSymtab(*B)) 473 continue; 474 475 SectionBase *Sec = DR->Section; 476 if (!shouldKeepInSymtab(Sec, B->getName(), *B)) 477 continue; 478 InX::SymTab->addSymbol(B); 479 } 480 } 481 } 482 483 template <class ELFT> void Writer<ELFT>::addSectionSymbols() { 484 // Create a section symbol for each output section so that we can represent 485 // relocations that point to the section. If we know that no relocation is 486 // referring to a section (that happens if the section is a synthetic one), we 487 // don't create a section symbol for that section. 488 for (BaseCommand *Base : Script->SectionCommands) { 489 auto *Sec = dyn_cast<OutputSection>(Base); 490 if (!Sec) 491 continue; 492 auto I = llvm::find_if(Sec->SectionCommands, [](BaseCommand *Base) { 493 if (auto *ISD = dyn_cast<InputSectionDescription>(Base)) 494 return !ISD->Sections.empty(); 495 return false; 496 }); 497 if (I == Sec->SectionCommands.end()) 498 continue; 499 InputSection *IS = cast<InputSectionDescription>(*I)->Sections[0]; 500 501 // Relocations are not using REL[A] section symbols. 502 if (IS->Type == SHT_REL || IS->Type == SHT_RELA) 503 continue; 504 505 // Unlike other synthetic sections, mergeable output sections contain data 506 // copied from input sections, and there may be a relocation pointing to its 507 // contents if -r or -emit-reloc are given. 508 if (isa<SyntheticSection>(IS) && !(IS->Flags & SHF_MERGE)) 509 continue; 510 511 auto *Sym = 512 make<Defined>(IS->File, "", STB_LOCAL, /*StOther=*/0, STT_SECTION, 513 /*Value=*/0, /*Size=*/0, IS); 514 InX::SymTab->addSymbol(Sym); 515 } 516 } 517 518 // Today's loaders have a feature to make segments read-only after 519 // processing dynamic relocations to enhance security. PT_GNU_RELRO 520 // is defined for that. 521 // 522 // This function returns true if a section needs to be put into a 523 // PT_GNU_RELRO segment. 524 static bool isRelroSection(const OutputSection *Sec) { 525 if (!Config->ZRelro) 526 return false; 527 528 uint64_t Flags = Sec->Flags; 529 530 // Non-allocatable or non-writable sections don't need RELRO because 531 // they are not writable or not even mapped to memory in the first place. 532 // RELRO is for sections that are essentially read-only but need to 533 // be writable only at process startup to allow dynamic linker to 534 // apply relocations. 535 if (!(Flags & SHF_ALLOC) || !(Flags & SHF_WRITE)) 536 return false; 537 538 // Once initialized, TLS data segments are used as data templates 539 // for a thread-local storage. For each new thread, runtime 540 // allocates memory for a TLS and copy templates there. No thread 541 // are supposed to use templates directly. Thus, it can be in RELRO. 542 if (Flags & SHF_TLS) 543 return true; 544 545 // .init_array, .preinit_array and .fini_array contain pointers to 546 // functions that are executed on process startup or exit. These 547 // pointers are set by the static linker, and they are not expected 548 // to change at runtime. But if you are an attacker, you could do 549 // interesting things by manipulating pointers in .fini_array, for 550 // example. So they are put into RELRO. 551 uint32_t Type = Sec->Type; 552 if (Type == SHT_INIT_ARRAY || Type == SHT_FINI_ARRAY || 553 Type == SHT_PREINIT_ARRAY) 554 return true; 555 556 // .got contains pointers to external symbols. They are resolved by 557 // the dynamic linker when a module is loaded into memory, and after 558 // that they are not expected to change. So, it can be in RELRO. 559 if (InX::Got && Sec == InX::Got->getParent()) 560 return true; 561 562 // .got.plt contains pointers to external function symbols. They are 563 // by default resolved lazily, so we usually cannot put it into RELRO. 564 // However, if "-z now" is given, the lazy symbol resolution is 565 // disabled, which enables us to put it into RELRO. 566 if (Sec == InX::GotPlt->getParent()) 567 return Config->ZNow; 568 569 // .dynamic section contains data for the dynamic linker, and 570 // there's no need to write to it at runtime, so it's better to put 571 // it into RELRO. 572 if (Sec == InX::Dynamic->getParent()) 573 return true; 574 575 // .bss.rel.ro is used for copy relocations for read-only symbols. 576 // Since the dynamic linker needs to process copy relocations, the 577 // section cannot be read-only, but once initialized, they shouldn't 578 // change. 579 if (Sec == InX::BssRelRo->getParent()) 580 return true; 581 582 // Sections with some special names are put into RELRO. This is a 583 // bit unfortunate because section names shouldn't be significant in 584 // ELF in spirit. But in reality many linker features depend on 585 // magic section names. 586 StringRef S = Sec->Name; 587 return S == ".data.rel.ro" || S == ".ctors" || S == ".dtors" || S == ".jcr" || 588 S == ".eh_frame" || S == ".openbsd.randomdata"; 589 } 590 591 // We compute a rank for each section. The rank indicates where the 592 // section should be placed in the file. Instead of using simple 593 // numbers (0,1,2...), we use a series of flags. One for each decision 594 // point when placing the section. 595 // Using flags has two key properties: 596 // * It is easy to check if a give branch was taken. 597 // * It is easy two see how similar two ranks are (see getRankProximity). 598 enum RankFlags { 599 RF_NOT_ADDR_SET = 1 << 16, 600 RF_NOT_INTERP = 1 << 15, 601 RF_NOT_ALLOC = 1 << 14, 602 RF_WRITE = 1 << 13, 603 RF_EXEC_WRITE = 1 << 12, 604 RF_EXEC = 1 << 11, 605 RF_NON_TLS_BSS = 1 << 10, 606 RF_NON_TLS_BSS_RO = 1 << 9, 607 RF_NOT_TLS = 1 << 8, 608 RF_BSS = 1 << 7, 609 RF_PPC_NOT_TOCBSS = 1 << 6, 610 RF_PPC_OPD = 1 << 5, 611 RF_PPC_TOCL = 1 << 4, 612 RF_PPC_TOC = 1 << 3, 613 RF_PPC_BRANCH_LT = 1 << 2, 614 RF_MIPS_GPREL = 1 << 1, 615 RF_MIPS_NOT_GOT = 1 << 0 616 }; 617 618 static unsigned getSectionRank(const OutputSection *Sec) { 619 unsigned Rank = 0; 620 621 // We want to put section specified by -T option first, so we 622 // can start assigning VA starting from them later. 623 if (Config->SectionStartMap.count(Sec->Name)) 624 return Rank; 625 Rank |= RF_NOT_ADDR_SET; 626 627 // Put .interp first because some loaders want to see that section 628 // on the first page of the executable file when loaded into memory. 629 if (Sec->Name == ".interp") 630 return Rank; 631 Rank |= RF_NOT_INTERP; 632 633 // Allocatable sections go first to reduce the total PT_LOAD size and 634 // so debug info doesn't change addresses in actual code. 635 if (!(Sec->Flags & SHF_ALLOC)) 636 return Rank | RF_NOT_ALLOC; 637 638 // Sort sections based on their access permission in the following 639 // order: R, RX, RWX, RW. This order is based on the following 640 // considerations: 641 // * Read-only sections come first such that they go in the 642 // PT_LOAD covering the program headers at the start of the file. 643 // * Read-only, executable sections come next, unless the 644 // -no-rosegment option is used. 645 // * Writable, executable sections follow such that .plt on 646 // architectures where it needs to be writable will be placed 647 // between .text and .data. 648 // * Writable sections come last, such that .bss lands at the very 649 // end of the last PT_LOAD. 650 bool IsExec = Sec->Flags & SHF_EXECINSTR; 651 bool IsWrite = Sec->Flags & SHF_WRITE; 652 653 if (IsExec) { 654 if (IsWrite) 655 Rank |= RF_EXEC_WRITE; 656 else if (!Config->SingleRoRx) 657 Rank |= RF_EXEC; 658 } else { 659 if (IsWrite) 660 Rank |= RF_WRITE; 661 } 662 663 // If we got here we know that both A and B are in the same PT_LOAD. 664 665 bool IsTls = Sec->Flags & SHF_TLS; 666 bool IsNoBits = Sec->Type == SHT_NOBITS; 667 668 // The first requirement we have is to put (non-TLS) nobits sections last. The 669 // reason is that the only thing the dynamic linker will see about them is a 670 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the 671 // PT_LOAD, so that has to correspond to the nobits sections. 672 bool IsNonTlsNoBits = IsNoBits && !IsTls; 673 if (IsNonTlsNoBits) 674 Rank |= RF_NON_TLS_BSS; 675 676 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo 677 // sections after r/w ones, so that the RelRo sections are contiguous. 678 bool IsRelRo = isRelroSection(Sec); 679 if (IsNonTlsNoBits && !IsRelRo) 680 Rank |= RF_NON_TLS_BSS_RO; 681 if (!IsNonTlsNoBits && IsRelRo) 682 Rank |= RF_NON_TLS_BSS_RO; 683 684 // The TLS initialization block needs to be a single contiguous block in a R/W 685 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W 686 // sections. The TLS NOBITS sections are placed here as they don't take up 687 // virtual address space in the PT_LOAD. 688 if (!IsTls) 689 Rank |= RF_NOT_TLS; 690 691 // Within the TLS initialization block, the non-nobits sections need to appear 692 // first. 693 if (IsNoBits) 694 Rank |= RF_BSS; 695 696 // Some architectures have additional ordering restrictions for sections 697 // within the same PT_LOAD. 698 if (Config->EMachine == EM_PPC64) { 699 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections 700 // that we would like to make sure appear is a specific order to maximize 701 // their coverage by a single signed 16-bit offset from the TOC base 702 // pointer. Conversely, the special .tocbss section should be first among 703 // all SHT_NOBITS sections. This will put it next to the loaded special 704 // PPC64 sections (and, thus, within reach of the TOC base pointer). 705 StringRef Name = Sec->Name; 706 if (Name != ".tocbss") 707 Rank |= RF_PPC_NOT_TOCBSS; 708 709 if (Name == ".opd") 710 Rank |= RF_PPC_OPD; 711 712 if (Name == ".toc1") 713 Rank |= RF_PPC_TOCL; 714 715 if (Name == ".toc") 716 Rank |= RF_PPC_TOC; 717 718 if (Name == ".branch_lt") 719 Rank |= RF_PPC_BRANCH_LT; 720 } 721 if (Config->EMachine == EM_MIPS) { 722 // All sections with SHF_MIPS_GPREL flag should be grouped together 723 // because data in these sections is addressable with a gp relative address. 724 if (Sec->Flags & SHF_MIPS_GPREL) 725 Rank |= RF_MIPS_GPREL; 726 727 if (Sec->Name != ".got") 728 Rank |= RF_MIPS_NOT_GOT; 729 } 730 731 return Rank; 732 } 733 734 static bool compareSections(const BaseCommand *ACmd, const BaseCommand *BCmd) { 735 const OutputSection *A = cast<OutputSection>(ACmd); 736 const OutputSection *B = cast<OutputSection>(BCmd); 737 if (A->SortRank != B->SortRank) 738 return A->SortRank < B->SortRank; 739 if (!(A->SortRank & RF_NOT_ADDR_SET)) 740 return Config->SectionStartMap.lookup(A->Name) < 741 Config->SectionStartMap.lookup(B->Name); 742 return false; 743 } 744 745 void PhdrEntry::add(OutputSection *Sec) { 746 LastSec = Sec; 747 if (!FirstSec) 748 FirstSec = Sec; 749 p_align = std::max(p_align, Sec->Alignment); 750 if (p_type == PT_LOAD) 751 Sec->PtLoad = this; 752 } 753 754 template <class ELFT> 755 static Defined *addOptionalRegular(StringRef Name, SectionBase *Sec, 756 uint64_t Val, uint8_t StOther = STV_HIDDEN, 757 uint8_t Binding = STB_GLOBAL) { 758 Symbol *S = Symtab->find(Name); 759 if (!S || S->isDefined()) 760 return nullptr; 761 Symbol *Sym = Symtab->addRegular<ELFT>(Name, StOther, STT_NOTYPE, Val, 762 /*Size=*/0, Binding, Sec, 763 /*File=*/nullptr); 764 return cast<Defined>(Sym); 765 } 766 767 // The beginning and the ending of .rel[a].plt section are marked 768 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked 769 // executable. The runtime needs these symbols in order to resolve 770 // all IRELATIVE relocs on startup. For dynamic executables, we don't 771 // need these symbols, since IRELATIVE relocs are resolved through GOT 772 // and PLT. For details, see http://www.airs.com/blog/archives/403. 773 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() { 774 if (!Config->Static) 775 return; 776 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start"; 777 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, 0, STV_HIDDEN, STB_WEAK); 778 779 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end"; 780 addOptionalRegular<ELFT>(S, In<ELFT>::RelaIplt, -1, STV_HIDDEN, STB_WEAK); 781 } 782 783 // The linker is expected to define some symbols depending on 784 // the linking result. This function defines such symbols. 785 template <class ELFT> void Writer<ELFT>::addReservedSymbols() { 786 if (Config->EMachine == EM_MIPS) { 787 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer 788 // so that it points to an absolute address which by default is relative 789 // to GOT. Default offset is 0x7ff0. 790 // See "Global Data Symbols" in Chapter 6 in the following document: 791 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf 792 ElfSym::MipsGp = Symtab->addAbsolute<ELFT>("_gp", STV_HIDDEN, STB_GLOBAL); 793 794 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between 795 // start of function and 'gp' pointer into GOT. 796 if (Symtab->find("_gp_disp")) 797 ElfSym::MipsGpDisp = 798 Symtab->addAbsolute<ELFT>("_gp_disp", STV_HIDDEN, STB_GLOBAL); 799 800 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp' 801 // pointer. This symbol is used in the code generated by .cpload pseudo-op 802 // in case of using -mno-shared option. 803 // https://sourceware.org/ml/binutils/2004-12/msg00094.html 804 if (Symtab->find("__gnu_local_gp")) 805 ElfSym::MipsLocalGp = 806 Symtab->addAbsolute<ELFT>("__gnu_local_gp", STV_HIDDEN, STB_GLOBAL); 807 } 808 809 // The _GLOBAL_OFFSET_TABLE_ symbol is defined by target convention to 810 // be at some offset from the base of the .got section, usually 0 or the end 811 // of the .got 812 InputSection *GotSection = InX::MipsGot ? cast<InputSection>(InX::MipsGot) 813 : cast<InputSection>(InX::Got); 814 ElfSym::GlobalOffsetTable = addOptionalRegular<ELFT>( 815 "_GLOBAL_OFFSET_TABLE_", GotSection, Target->GotBaseSymOff); 816 817 // __ehdr_start is the location of ELF file headers. Note that we define 818 // this symbol unconditionally even when using a linker script, which 819 // differs from the behavior implemented by GNU linker which only define 820 // this symbol if ELF headers are in the memory mapped segment. 821 // __executable_start is not documented, but the expectation of at 822 // least the android libc is that it points to the elf header too. 823 // __dso_handle symbol is passed to cxa_finalize as a marker to identify 824 // each DSO. The address of the symbol doesn't matter as long as they are 825 // different in different DSOs, so we chose the start address of the DSO. 826 for (const char *Name : 827 {"__ehdr_start", "__executable_start", "__dso_handle"}) 828 addOptionalRegular<ELFT>(Name, Out::ElfHeader, 0, STV_HIDDEN); 829 830 // If linker script do layout we do not need to create any standart symbols. 831 if (Script->HasSectionsCommand) 832 return; 833 834 auto Add = [](StringRef S, int64_t Pos) { 835 return addOptionalRegular<ELFT>(S, Out::ElfHeader, Pos, STV_DEFAULT); 836 }; 837 838 ElfSym::Bss = Add("__bss_start", 0); 839 ElfSym::End1 = Add("end", -1); 840 ElfSym::End2 = Add("_end", -1); 841 ElfSym::Etext1 = Add("etext", -1); 842 ElfSym::Etext2 = Add("_etext", -1); 843 ElfSym::Edata1 = Add("edata", -1); 844 ElfSym::Edata2 = Add("_edata", -1); 845 } 846 847 template <class ELFT> 848 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) { 849 // Scan all relocations. Each relocation goes through a series 850 // of tests to determine if it needs special treatment, such as 851 // creating GOT, PLT, copy relocations, etc. 852 // Note that relocations for non-alloc sections are directly 853 // processed by InputSection::relocateNonAlloc. 854 for (InputSectionBase *IS : InputSections) 855 if (IS->Live && isa<InputSection>(IS) && (IS->Flags & SHF_ALLOC)) 856 Fn(*IS); 857 for (EhInputSection *ES : InX::EhFrame->Sections) 858 Fn(*ES); 859 } 860 861 // This function generates assignments for predefined symbols (e.g. _end or 862 // _etext) and inserts them into the commands sequence to be processed at the 863 // appropriate time. This ensures that the value is going to be correct by the 864 // time any references to these symbols are processed and is equivalent to 865 // defining these symbols explicitly in the linker script. 866 template <class ELFT> void Writer<ELFT>::setReservedSymbolSections() { 867 PhdrEntry *Last = nullptr; 868 PhdrEntry *LastRO = nullptr; 869 870 for (PhdrEntry *P : Phdrs) { 871 if (P->p_type != PT_LOAD) 872 continue; 873 Last = P; 874 if (!(P->p_flags & PF_W)) 875 LastRO = P; 876 } 877 878 if (LastRO) { 879 // _etext is the first location after the last read-only loadable segment. 880 if (ElfSym::Etext1) 881 ElfSym::Etext1->Section = LastRO->LastSec; 882 if (ElfSym::Etext2) 883 ElfSym::Etext2->Section = LastRO->LastSec; 884 } 885 886 if (Last) { 887 // _edata points to the end of the last mapped initialized section. 888 OutputSection *Edata = nullptr; 889 for (OutputSection *OS : OutputSections) { 890 if (OS->Type != SHT_NOBITS) 891 Edata = OS; 892 if (OS == Last->LastSec) 893 break; 894 } 895 896 if (ElfSym::Edata1) 897 ElfSym::Edata1->Section = Edata; 898 if (ElfSym::Edata2) 899 ElfSym::Edata2->Section = Edata; 900 901 // _end is the first location after the uninitialized data region. 902 if (ElfSym::End1) 903 ElfSym::End1->Section = Last->LastSec; 904 if (ElfSym::End2) 905 ElfSym::End2->Section = Last->LastSec; 906 } 907 908 if (ElfSym::Bss) 909 ElfSym::Bss->Section = findSection(".bss"); 910 911 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should 912 // be equal to the _gp symbol's value. 913 if (ElfSym::MipsGp) { 914 // Find GP-relative section with the lowest address 915 // and use this address to calculate default _gp value. 916 for (OutputSection *OS : OutputSections) { 917 if (OS->Flags & SHF_MIPS_GPREL) { 918 ElfSym::MipsGp->Section = OS; 919 ElfSym::MipsGp->Value = 0x7ff0; 920 break; 921 } 922 } 923 } 924 } 925 926 // We want to find how similar two ranks are. 927 // The more branches in getSectionRank that match, the more similar they are. 928 // Since each branch corresponds to a bit flag, we can just use 929 // countLeadingZeros. 930 static int getRankProximityAux(OutputSection *A, OutputSection *B) { 931 return countLeadingZeros(A->SortRank ^ B->SortRank); 932 } 933 934 static int getRankProximity(OutputSection *A, BaseCommand *B) { 935 if (auto *Sec = dyn_cast<OutputSection>(B)) 936 if (Sec->Live) 937 return getRankProximityAux(A, Sec); 938 return -1; 939 } 940 941 // When placing orphan sections, we want to place them after symbol assignments 942 // so that an orphan after 943 // begin_foo = .; 944 // foo : { *(foo) } 945 // end_foo = .; 946 // doesn't break the intended meaning of the begin/end symbols. 947 // We don't want to go over sections since findOrphanPos is the 948 // one in charge of deciding the order of the sections. 949 // We don't want to go over changes to '.', since doing so in 950 // rx_sec : { *(rx_sec) } 951 // . = ALIGN(0x1000); 952 // /* The RW PT_LOAD starts here*/ 953 // rw_sec : { *(rw_sec) } 954 // would mean that the RW PT_LOAD would become unaligned. 955 static bool shouldSkip(BaseCommand *Cmd) { 956 if (isa<OutputSection>(Cmd)) 957 return false; 958 if (auto *Assign = dyn_cast<SymbolAssignment>(Cmd)) 959 return Assign->Name != "."; 960 return true; 961 } 962 963 // We want to place orphan sections so that they share as much 964 // characteristics with their neighbors as possible. For example, if 965 // both are rw, or both are tls. 966 template <typename ELFT> 967 static std::vector<BaseCommand *>::iterator 968 findOrphanPos(std::vector<BaseCommand *>::iterator B, 969 std::vector<BaseCommand *>::iterator E) { 970 OutputSection *Sec = cast<OutputSection>(*E); 971 972 // Find the first element that has as close a rank as possible. 973 auto I = std::max_element(B, E, [=](BaseCommand *A, BaseCommand *B) { 974 return getRankProximity(Sec, A) < getRankProximity(Sec, B); 975 }); 976 if (I == E) 977 return E; 978 979 // Consider all existing sections with the same proximity. 980 int Proximity = getRankProximity(Sec, *I); 981 for (; I != E; ++I) { 982 auto *CurSec = dyn_cast<OutputSection>(*I); 983 if (!CurSec || !CurSec->Live) 984 continue; 985 if (getRankProximity(Sec, CurSec) != Proximity || 986 Sec->SortRank < CurSec->SortRank) 987 break; 988 } 989 990 auto IsLiveSection = [](BaseCommand *Cmd) { 991 auto *OS = dyn_cast<OutputSection>(Cmd); 992 return OS && OS->Live; 993 }; 994 995 auto J = std::find_if(llvm::make_reverse_iterator(I), 996 llvm::make_reverse_iterator(B), IsLiveSection); 997 I = J.base(); 998 999 // As a special case, if the orphan section is the last section, put 1000 // it at the very end, past any other commands. 1001 // This matches bfd's behavior and is convenient when the linker script fully 1002 // specifies the start of the file, but doesn't care about the end (the non 1003 // alloc sections for example). 1004 auto NextSec = std::find_if(I, E, IsLiveSection); 1005 if (NextSec == E) 1006 return E; 1007 1008 while (I != E && shouldSkip(*I)) 1009 ++I; 1010 return I; 1011 } 1012 1013 // If no layout was provided by linker script, we want to apply default 1014 // sorting for special input sections and handle --symbol-ordering-file. 1015 template <class ELFT> void Writer<ELFT>::sortInputSections() { 1016 assert(!Script->HasSectionsCommand); 1017 1018 // Sort input sections by priority using the list provided 1019 // by --symbol-ordering-file. 1020 DenseMap<SectionBase *, int> Order = buildSectionOrder(); 1021 if (!Order.empty()) 1022 for (BaseCommand *Base : Script->SectionCommands) 1023 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1024 if (Sec->Live) 1025 Sec->sort([&](InputSectionBase *S) { return Order.lookup(S); }); 1026 1027 // Sort input sections by section name suffixes for 1028 // __attribute__((init_priority(N))). 1029 if (OutputSection *Sec = findSection(".init_array")) 1030 Sec->sortInitFini(); 1031 if (OutputSection *Sec = findSection(".fini_array")) 1032 Sec->sortInitFini(); 1033 1034 // Sort input sections by the special rule for .ctors and .dtors. 1035 if (OutputSection *Sec = findSection(".ctors")) 1036 Sec->sortCtorsDtors(); 1037 if (OutputSection *Sec = findSection(".dtors")) 1038 Sec->sortCtorsDtors(); 1039 } 1040 1041 template <class ELFT> void Writer<ELFT>::sortSections() { 1042 Script->adjustSectionsBeforeSorting(); 1043 1044 // Don't sort if using -r. It is not necessary and we want to preserve the 1045 // relative order for SHF_LINK_ORDER sections. 1046 if (Config->Relocatable) 1047 return; 1048 1049 for (BaseCommand *Base : Script->SectionCommands) 1050 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1051 Sec->SortRank = getSectionRank(Sec); 1052 1053 if (!Script->HasSectionsCommand) { 1054 sortInputSections(); 1055 1056 // We know that all the OutputSections are contiguous in this case. 1057 auto E = Script->SectionCommands.end(); 1058 auto I = Script->SectionCommands.begin(); 1059 auto IsSection = [](BaseCommand *Base) { return isa<OutputSection>(Base); }; 1060 I = std::find_if(I, E, IsSection); 1061 E = std::find_if(llvm::make_reverse_iterator(E), 1062 llvm::make_reverse_iterator(I), IsSection) 1063 .base(); 1064 std::stable_sort(I, E, compareSections); 1065 return; 1066 } 1067 1068 // Orphan sections are sections present in the input files which are 1069 // not explicitly placed into the output file by the linker script. 1070 // 1071 // The sections in the linker script are already in the correct 1072 // order. We have to figuere out where to insert the orphan 1073 // sections. 1074 // 1075 // The order of the sections in the script is arbitrary and may not agree with 1076 // compareSections. This means that we cannot easily define a strict weak 1077 // ordering. To see why, consider a comparison of a section in the script and 1078 // one not in the script. We have a two simple options: 1079 // * Make them equivalent (a is not less than b, and b is not less than a). 1080 // The problem is then that equivalence has to be transitive and we can 1081 // have sections a, b and c with only b in a script and a less than c 1082 // which breaks this property. 1083 // * Use compareSectionsNonScript. Given that the script order doesn't have 1084 // to match, we can end up with sections a, b, c, d where b and c are in the 1085 // script and c is compareSectionsNonScript less than b. In which case d 1086 // can be equivalent to c, a to b and d < a. As a concrete example: 1087 // .a (rx) # not in script 1088 // .b (rx) # in script 1089 // .c (ro) # in script 1090 // .d (ro) # not in script 1091 // 1092 // The way we define an order then is: 1093 // * Sort only the orphan sections. They are in the end right now. 1094 // * Move each orphan section to its preferred position. We try 1095 // to put each section in the last position where it it can share 1096 // a PT_LOAD. 1097 // 1098 // There is some ambiguity as to where exactly a new entry should be 1099 // inserted, because Commands contains not only output section 1100 // commands but also other types of commands such as symbol assignment 1101 // expressions. There's no correct answer here due to the lack of the 1102 // formal specification of the linker script. We use heuristics to 1103 // determine whether a new output command should be added before or 1104 // after another commands. For the details, look at shouldSkip 1105 // function. 1106 1107 auto I = Script->SectionCommands.begin(); 1108 auto E = Script->SectionCommands.end(); 1109 auto NonScriptI = std::find_if(I, E, [](BaseCommand *Base) { 1110 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1111 return Sec->Live && Sec->SectionIndex == INT_MAX; 1112 return false; 1113 }); 1114 1115 // Sort the orphan sections. 1116 std::stable_sort(NonScriptI, E, compareSections); 1117 1118 // As a horrible special case, skip the first . assignment if it is before any 1119 // section. We do this because it is common to set a load address by starting 1120 // the script with ". = 0xabcd" and the expectation is that every section is 1121 // after that. 1122 auto FirstSectionOrDotAssignment = 1123 std::find_if(I, E, [](BaseCommand *Cmd) { return !shouldSkip(Cmd); }); 1124 if (FirstSectionOrDotAssignment != E && 1125 isa<SymbolAssignment>(**FirstSectionOrDotAssignment)) 1126 ++FirstSectionOrDotAssignment; 1127 I = FirstSectionOrDotAssignment; 1128 1129 while (NonScriptI != E) { 1130 auto Pos = findOrphanPos<ELFT>(I, NonScriptI); 1131 OutputSection *Orphan = cast<OutputSection>(*NonScriptI); 1132 1133 // As an optimization, find all sections with the same sort rank 1134 // and insert them with one rotate. 1135 unsigned Rank = Orphan->SortRank; 1136 auto End = std::find_if(NonScriptI + 1, E, [=](BaseCommand *Cmd) { 1137 return cast<OutputSection>(Cmd)->SortRank != Rank; 1138 }); 1139 std::rotate(Pos, NonScriptI, End); 1140 NonScriptI = End; 1141 } 1142 1143 Script->adjustSectionsAfterSorting(); 1144 } 1145 1146 static void applySynthetic(const std::vector<SyntheticSection *> &Sections, 1147 std::function<void(SyntheticSection *)> Fn) { 1148 for (SyntheticSection *SS : Sections) 1149 if (SS && SS->getParent() && !SS->empty()) 1150 Fn(SS); 1151 } 1152 1153 // In order to allow users to manipulate linker-synthesized sections, 1154 // we had to add synthetic sections to the input section list early, 1155 // even before we make decisions whether they are needed. This allows 1156 // users to write scripts like this: ".mygot : { .got }". 1157 // 1158 // Doing it has an unintended side effects. If it turns out that we 1159 // don't need a .got (for example) at all because there's no 1160 // relocation that needs a .got, we don't want to emit .got. 1161 // 1162 // To deal with the above problem, this function is called after 1163 // scanRelocations is called to remove synthetic sections that turn 1164 // out to be empty. 1165 static void removeUnusedSyntheticSections() { 1166 // All input synthetic sections that can be empty are placed after 1167 // all regular ones. We iterate over them all and exit at first 1168 // non-synthetic. 1169 for (InputSectionBase *S : llvm::reverse(InputSections)) { 1170 SyntheticSection *SS = dyn_cast<SyntheticSection>(S); 1171 if (!SS) 1172 return; 1173 OutputSection *OS = SS->getParent(); 1174 if (!SS->empty() || !OS) 1175 continue; 1176 1177 std::vector<BaseCommand *>::iterator Empty = OS->SectionCommands.end(); 1178 for (auto I = OS->SectionCommands.begin(), E = OS->SectionCommands.end(); 1179 I != E; ++I) { 1180 BaseCommand *B = *I; 1181 if (auto *ISD = dyn_cast<InputSectionDescription>(B)) { 1182 llvm::erase_if(ISD->Sections, 1183 [=](InputSection *IS) { return IS == SS; }); 1184 if (ISD->Sections.empty()) 1185 Empty = I; 1186 } 1187 } 1188 if (Empty != OS->SectionCommands.end()) 1189 OS->SectionCommands.erase(Empty); 1190 1191 // If there are no other sections in the output section, remove it from the 1192 // output. 1193 if (OS->SectionCommands.empty()) 1194 OS->Live = false; 1195 } 1196 } 1197 1198 // Returns true if a symbol can be replaced at load-time by a symbol 1199 // with the same name defined in other ELF executable or DSO. 1200 static bool computeIsPreemptible(const Symbol &B) { 1201 assert(!B.isLocal()); 1202 // Only symbols that appear in dynsym can be preempted. 1203 if (!B.includeInDynsym()) 1204 return false; 1205 1206 // Only default visibility symbols can be preempted. 1207 if (B.Visibility != STV_DEFAULT) 1208 return false; 1209 1210 // At this point copy relocations have not been created yet, so any 1211 // symbol that is not defined locally is preemptible. 1212 if (!B.isDefined()) 1213 return true; 1214 1215 // If we have a dynamic list it specifies which local symbols are preemptible. 1216 if (Config->HasDynamicList) 1217 return false; 1218 1219 if (!Config->Shared) 1220 return false; 1221 1222 // -Bsymbolic means that definitions are not preempted. 1223 if (Config->Bsymbolic || (Config->BsymbolicFunctions && B.isFunc())) 1224 return false; 1225 return true; 1226 } 1227 1228 // Create output section objects and add them to OutputSections. 1229 template <class ELFT> void Writer<ELFT>::finalizeSections() { 1230 Out::DebugInfo = findSection(".debug_info"); 1231 Out::PreinitArray = findSection(".preinit_array"); 1232 Out::InitArray = findSection(".init_array"); 1233 Out::FiniArray = findSection(".fini_array"); 1234 1235 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop 1236 // symbols for sections, so that the runtime can get the start and end 1237 // addresses of each section by section name. Add such symbols. 1238 if (!Config->Relocatable) { 1239 addStartEndSymbols(); 1240 for (BaseCommand *Base : Script->SectionCommands) 1241 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1242 addStartStopSymbols(Sec); 1243 } 1244 1245 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type. 1246 // It should be okay as no one seems to care about the type. 1247 // Even the author of gold doesn't remember why gold behaves that way. 1248 // https://sourceware.org/ml/binutils/2002-03/msg00360.html 1249 if (InX::DynSymTab) 1250 Symtab->addRegular<ELFT>("_DYNAMIC", STV_HIDDEN, STT_NOTYPE, 0 /*Value*/, 1251 /*Size=*/0, STB_WEAK, InX::Dynamic, 1252 /*File=*/nullptr); 1253 1254 // Define __rel[a]_iplt_{start,end} symbols if needed. 1255 addRelIpltSymbols(); 1256 1257 // This responsible for splitting up .eh_frame section into 1258 // pieces. The relocation scan uses those pieces, so this has to be 1259 // earlier. 1260 applySynthetic({InX::EhFrame}, 1261 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1262 1263 for (Symbol *S : Symtab->getSymbols()) 1264 S->IsPreemptible |= computeIsPreemptible(*S); 1265 1266 // Scan relocations. This must be done after every symbol is declared so that 1267 // we can correctly decide if a dynamic relocation is needed. 1268 if (!Config->Relocatable) 1269 forEachRelSec(scanRelocations<ELFT>); 1270 1271 if (InX::Plt && !InX::Plt->empty()) 1272 InX::Plt->addSymbols(); 1273 if (InX::Iplt && !InX::Iplt->empty()) 1274 InX::Iplt->addSymbols(); 1275 1276 // Now that we have defined all possible global symbols including linker- 1277 // synthesized ones. Visit all symbols to give the finishing touches. 1278 for (Symbol *Sym : Symtab->getSymbols()) { 1279 if (!includeInSymtab(*Sym)) 1280 continue; 1281 if (InX::SymTab) 1282 InX::SymTab->addSymbol(Sym); 1283 1284 if (InX::DynSymTab && Sym->includeInDynsym()) { 1285 InX::DynSymTab->addSymbol(Sym); 1286 if (auto *SS = dyn_cast<SharedSymbol>(Sym)) 1287 if (cast<SharedFile<ELFT>>(Sym->File)->IsNeeded) 1288 In<ELFT>::VerNeed->addSymbol(SS); 1289 } 1290 } 1291 1292 // Do not proceed if there was an undefined symbol. 1293 if (errorCount()) 1294 return; 1295 1296 addPredefinedSections(); 1297 removeUnusedSyntheticSections(); 1298 1299 sortSections(); 1300 Script->removeEmptyCommands(); 1301 1302 // Now that we have the final list, create a list of all the 1303 // OutputSections for convenience. 1304 for (BaseCommand *Base : Script->SectionCommands) 1305 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1306 OutputSections.push_back(Sec); 1307 1308 // Prefer command line supplied address over other constraints. 1309 for (OutputSection *Sec : OutputSections) { 1310 auto I = Config->SectionStartMap.find(Sec->Name); 1311 if (I != Config->SectionStartMap.end()) 1312 Sec->AddrExpr = [=] { return I->second; }; 1313 } 1314 1315 // This is a bit of a hack. A value of 0 means undef, so we set it 1316 // to 1 t make __ehdr_start defined. The section number is not 1317 // particularly relevant. 1318 Out::ElfHeader->SectionIndex = 1; 1319 1320 unsigned I = 1; 1321 for (OutputSection *Sec : OutputSections) { 1322 Sec->SectionIndex = I++; 1323 Sec->ShName = InX::ShStrTab->addString(Sec->Name); 1324 } 1325 1326 // Binary and relocatable output does not have PHDRS. 1327 // The headers have to be created before finalize as that can influence the 1328 // image base and the dynamic section on mips includes the image base. 1329 if (!Config->Relocatable && !Config->OFormatBinary) { 1330 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs(); 1331 addPtArmExid(Phdrs); 1332 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size(); 1333 } 1334 1335 // Some symbols are defined in term of program headers. Now that we 1336 // have the headers, we can find out which sections they point to. 1337 setReservedSymbolSections(); 1338 1339 // Dynamic section must be the last one in this list and dynamic 1340 // symbol table section (DynSymTab) must be the first one. 1341 applySynthetic({InX::DynSymTab, InX::Bss, InX::BssRelRo, 1342 InX::GnuHashTab, InX::HashTab, InX::SymTab, 1343 InX::ShStrTab, InX::StrTab, In<ELFT>::VerDef, 1344 InX::DynStrTab, InX::Got, InX::MipsGot, 1345 InX::IgotPlt, InX::GotPlt, In<ELFT>::RelaDyn, 1346 In<ELFT>::RelaIplt, In<ELFT>::RelaPlt, InX::Plt, 1347 InX::Iplt, InX::EhFrameHdr, In<ELFT>::VerSym, 1348 In<ELFT>::VerNeed, InX::Dynamic}, 1349 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1350 1351 if (!Script->HasSectionsCommand && !Config->Relocatable) 1352 fixSectionAlignments(); 1353 1354 // Some architectures use small displacements for jump instructions. 1355 // It is linker's responsibility to create thunks containing long 1356 // jump instructions if jump targets are too far. Create thunks. 1357 if (Target->NeedsThunks || Config->AndroidPackDynRelocs) { 1358 ThunkCreator TC; 1359 bool Changed; 1360 do { 1361 Script->assignAddresses(); 1362 Changed = false; 1363 if (Target->NeedsThunks) 1364 Changed |= TC.createThunks(OutputSections); 1365 if (InX::MipsGot) 1366 InX::MipsGot->updateAllocSize(); 1367 Changed |= In<ELFT>::RelaDyn->updateAllocSize(); 1368 } while (Changed); 1369 } 1370 1371 // Fill other section headers. The dynamic table is finalized 1372 // at the end because some tags like RELSZ depend on result 1373 // of finalizing other sections. 1374 for (OutputSection *Sec : OutputSections) 1375 Sec->finalize<ELFT>(); 1376 1377 // createThunks may have added local symbols to the static symbol table 1378 applySynthetic({InX::SymTab}, 1379 [](SyntheticSection *SS) { SS->postThunkContents(); }); 1380 } 1381 1382 template <class ELFT> void Writer<ELFT>::addPredefinedSections() { 1383 // ARM ABI requires .ARM.exidx to be terminated by some piece of data. 1384 // We have the terminater synthetic section class. Add that at the end. 1385 OutputSection *Cmd = findSection(".ARM.exidx"); 1386 if (!Cmd || !Cmd->Live || Config->Relocatable) 1387 return; 1388 1389 auto *Sentinel = make<ARMExidxSentinelSection>(); 1390 Cmd->addSection(Sentinel); 1391 } 1392 1393 // The linker is expected to define SECNAME_start and SECNAME_end 1394 // symbols for a few sections. This function defines them. 1395 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() { 1396 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) { 1397 // These symbols resolve to the image base if the section does not exist. 1398 // A special value -1 indicates end of the section. 1399 if (OS) { 1400 addOptionalRegular<ELFT>(Start, OS, 0); 1401 addOptionalRegular<ELFT>(End, OS, -1); 1402 } else { 1403 if (Config->Pic) 1404 OS = Out::ElfHeader; 1405 addOptionalRegular<ELFT>(Start, OS, 0); 1406 addOptionalRegular<ELFT>(End, OS, 0); 1407 } 1408 }; 1409 1410 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray); 1411 Define("__init_array_start", "__init_array_end", Out::InitArray); 1412 Define("__fini_array_start", "__fini_array_end", Out::FiniArray); 1413 1414 if (OutputSection *Sec = findSection(".ARM.exidx")) 1415 Define("__exidx_start", "__exidx_end", Sec); 1416 } 1417 1418 // If a section name is valid as a C identifier (which is rare because of 1419 // the leading '.'), linkers are expected to define __start_<secname> and 1420 // __stop_<secname> symbols. They are at beginning and end of the section, 1421 // respectively. This is not requested by the ELF standard, but GNU ld and 1422 // gold provide the feature, and used by many programs. 1423 template <class ELFT> 1424 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) { 1425 StringRef S = Sec->Name; 1426 if (!isValidCIdentifier(S)) 1427 return; 1428 addOptionalRegular<ELFT>(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT); 1429 addOptionalRegular<ELFT>(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT); 1430 } 1431 1432 template <class ELFT> OutputSection *Writer<ELFT>::findSection(StringRef Name) { 1433 for (BaseCommand *Base : Script->SectionCommands) 1434 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1435 if (Sec->Name == Name) 1436 return Sec; 1437 return nullptr; 1438 } 1439 1440 static bool needsPtLoad(OutputSection *Sec) { 1441 if (!(Sec->Flags & SHF_ALLOC)) 1442 return false; 1443 1444 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is 1445 // responsible for allocating space for them, not the PT_LOAD that 1446 // contains the TLS initialization image. 1447 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS) 1448 return false; 1449 return true; 1450 } 1451 1452 // Linker scripts are responsible for aligning addresses. Unfortunately, most 1453 // linker scripts are designed for creating two PT_LOADs only, one RX and one 1454 // RW. This means that there is no alignment in the RO to RX transition and we 1455 // cannot create a PT_LOAD there. 1456 static uint64_t computeFlags(uint64_t Flags) { 1457 if (Config->Omagic) 1458 return PF_R | PF_W | PF_X; 1459 if (Config->SingleRoRx && !(Flags & PF_W)) 1460 return Flags | PF_X; 1461 return Flags; 1462 } 1463 1464 // Prior to finalizeContents() an OutputSection containing SyntheticSections 1465 // may have 0 Size, but contain SyntheticSections that haven't had their size 1466 // calculated yet. We must use SyntheticSection->empty() for these sections. 1467 static bool isOutputSectionZeroSize(const OutputSection* Sec) { 1468 if (Sec->Size > 0) 1469 return false; 1470 for (BaseCommand *BC : Sec->SectionCommands) { 1471 if (auto *ISD = dyn_cast<InputSectionDescription>(BC)) 1472 for (InputSection *IS : ISD->Sections) 1473 if (SyntheticSection *SS = dyn_cast<SyntheticSection>(IS)) 1474 if (!SS->empty()) 1475 return false; 1476 } 1477 return true; 1478 } 1479 1480 // Decide which program headers to create and which sections to include in each 1481 // one. 1482 template <class ELFT> std::vector<PhdrEntry *> Writer<ELFT>::createPhdrs() { 1483 std::vector<PhdrEntry *> Ret; 1484 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * { 1485 Ret.push_back(make<PhdrEntry>(Type, Flags)); 1486 return Ret.back(); 1487 }; 1488 1489 // The first phdr entry is PT_PHDR which describes the program header itself. 1490 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders); 1491 1492 // PT_INTERP must be the second entry if exists. 1493 if (OutputSection *Cmd = findSection(".interp")) 1494 AddHdr(PT_INTERP, Cmd->getPhdrFlags())->add(Cmd); 1495 1496 // Add the first PT_LOAD segment for regular output sections. 1497 uint64_t Flags = computeFlags(PF_R); 1498 PhdrEntry *Load = AddHdr(PT_LOAD, Flags); 1499 1500 // Add the headers. We will remove them if they don't fit. 1501 Load->add(Out::ElfHeader); 1502 Load->add(Out::ProgramHeaders); 1503 1504 for (OutputSection *Sec : OutputSections) { 1505 if (!(Sec->Flags & SHF_ALLOC)) 1506 break; 1507 if (!needsPtLoad(Sec)) 1508 continue; 1509 1510 // Segments are contiguous memory regions that has the same attributes 1511 // (e.g. executable or writable). There is one phdr for each segment. 1512 // Therefore, we need to create a new phdr when the next section has 1513 // different flags or is loaded at a discontiguous address using AT linker 1514 // script command. 1515 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags()); 1516 if (Sec->LMAExpr || Flags != NewFlags) { 1517 Load = AddHdr(PT_LOAD, NewFlags); 1518 Flags = NewFlags; 1519 } 1520 1521 Load->add(Sec); 1522 } 1523 1524 // Add a TLS segment if any. 1525 PhdrEntry *TlsHdr = make<PhdrEntry>(PT_TLS, PF_R); 1526 for (OutputSection *Sec : OutputSections) 1527 if (Sec->Flags & SHF_TLS) 1528 TlsHdr->add(Sec); 1529 if (TlsHdr->FirstSec) 1530 Ret.push_back(TlsHdr); 1531 1532 // Add an entry for .dynamic. 1533 if (InX::DynSymTab) 1534 AddHdr(PT_DYNAMIC, InX::Dynamic->getParent()->getPhdrFlags()) 1535 ->add(InX::Dynamic->getParent()); 1536 1537 // PT_GNU_RELRO includes all sections that should be marked as 1538 // read-only by dynamic linker after proccessing relocations. 1539 // Current dynamic loaders only support one PT_GNU_RELRO PHDR, give 1540 // an error message if more than one PT_GNU_RELRO PHDR is required. 1541 PhdrEntry *RelRo = make<PhdrEntry>(PT_GNU_RELRO, PF_R); 1542 bool InRelroPhdr = false; 1543 bool IsRelroFinished = false; 1544 for (OutputSection *Sec : OutputSections) { 1545 if (!needsPtLoad(Sec) || isOutputSectionZeroSize(Sec)) 1546 continue; 1547 if (isRelroSection(Sec)) { 1548 InRelroPhdr = true; 1549 if (!IsRelroFinished) 1550 RelRo->add(Sec); 1551 else 1552 error("section: " + Sec->Name + " is not contiguous with other relro" + 1553 " sections"); 1554 } else if (InRelroPhdr) { 1555 InRelroPhdr = false; 1556 IsRelroFinished = true; 1557 } 1558 } 1559 if (RelRo->FirstSec) 1560 Ret.push_back(RelRo); 1561 1562 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr. 1563 if (!InX::EhFrame->empty() && InX::EhFrameHdr && InX::EhFrame->getParent() && 1564 InX::EhFrameHdr->getParent()) 1565 AddHdr(PT_GNU_EH_FRAME, InX::EhFrameHdr->getParent()->getPhdrFlags()) 1566 ->add(InX::EhFrameHdr->getParent()); 1567 1568 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes 1569 // the dynamic linker fill the segment with random data. 1570 if (OutputSection *Cmd = findSection(".openbsd.randomdata")) 1571 AddHdr(PT_OPENBSD_RANDOMIZE, Cmd->getPhdrFlags())->add(Cmd); 1572 1573 // PT_GNU_STACK is a special section to tell the loader to make the 1574 // pages for the stack non-executable. If you really want an executable 1575 // stack, you can pass -z execstack, but that's not recommended for 1576 // security reasons. 1577 unsigned Perm; 1578 if (Config->ZExecstack) 1579 Perm = PF_R | PF_W | PF_X; 1580 else 1581 Perm = PF_R | PF_W; 1582 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize; 1583 1584 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable 1585 // is expected to perform W^X violations, such as calling mprotect(2) or 1586 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on 1587 // OpenBSD. 1588 if (Config->ZWxneeded) 1589 AddHdr(PT_OPENBSD_WXNEEDED, PF_X); 1590 1591 // Create one PT_NOTE per a group of contiguous .note sections. 1592 PhdrEntry *Note = nullptr; 1593 for (OutputSection *Sec : OutputSections) { 1594 if (Sec->Type == SHT_NOTE) { 1595 if (!Note || Sec->LMAExpr) 1596 Note = AddHdr(PT_NOTE, PF_R); 1597 Note->add(Sec); 1598 } else { 1599 Note = nullptr; 1600 } 1601 } 1602 return Ret; 1603 } 1604 1605 template <class ELFT> 1606 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry *> &Phdrs) { 1607 if (Config->EMachine != EM_ARM) 1608 return; 1609 auto I = llvm::find_if(OutputSections, [](OutputSection *Cmd) { 1610 return Cmd->Type == SHT_ARM_EXIDX; 1611 }); 1612 if (I == OutputSections.end()) 1613 return; 1614 1615 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME 1616 PhdrEntry *ARMExidx = make<PhdrEntry>(PT_ARM_EXIDX, PF_R); 1617 ARMExidx->add(*I); 1618 Phdrs.push_back(ARMExidx); 1619 } 1620 1621 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the 1622 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic 1623 // linker can set the permissions. 1624 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() { 1625 auto PageAlign = [](OutputSection *Cmd) { 1626 if (Cmd && !Cmd->AddrExpr) 1627 Cmd->AddrExpr = [=] { 1628 return alignTo(Script->getDot(), Config->MaxPageSize); 1629 }; 1630 }; 1631 1632 for (const PhdrEntry *P : Phdrs) 1633 if (P->p_type == PT_LOAD && P->FirstSec) 1634 PageAlign(P->FirstSec); 1635 1636 for (const PhdrEntry *P : Phdrs) { 1637 if (P->p_type != PT_GNU_RELRO) 1638 continue; 1639 if (P->FirstSec) 1640 PageAlign(P->FirstSec); 1641 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we 1642 // have to align it to a page. 1643 auto End = OutputSections.end(); 1644 auto I = std::find(OutputSections.begin(), End, P->LastSec); 1645 if (I == End || (I + 1) == End) 1646 continue; 1647 OutputSection *Cmd = (*(I + 1)); 1648 if (needsPtLoad(Cmd)) 1649 PageAlign(Cmd); 1650 } 1651 } 1652 1653 // Adjusts the file alignment for a given output section and returns 1654 // its new file offset. The file offset must be the same with its 1655 // virtual address (modulo the page size) so that the loader can load 1656 // executables without any address adjustment. 1657 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Cmd) { 1658 // If the section is not in a PT_LOAD, we just have to align it. 1659 if (!Cmd->PtLoad) 1660 return alignTo(Off, Cmd->Alignment); 1661 1662 OutputSection *First = Cmd->PtLoad->FirstSec; 1663 // The first section in a PT_LOAD has to have congruent offset and address 1664 // module the page size. 1665 if (Cmd == First) 1666 return alignTo(Off, std::max<uint64_t>(Cmd->Alignment, Config->MaxPageSize), 1667 Cmd->Addr); 1668 1669 // If two sections share the same PT_LOAD the file offset is calculated 1670 // using this formula: Off2 = Off1 + (VA2 - VA1). 1671 return First->Offset + Cmd->Addr - First->Addr; 1672 } 1673 1674 static uint64_t setOffset(OutputSection *Cmd, uint64_t Off) { 1675 if (Cmd->Type == SHT_NOBITS) { 1676 Cmd->Offset = Off; 1677 return Off; 1678 } 1679 1680 Off = getFileAlignment(Off, Cmd); 1681 Cmd->Offset = Off; 1682 return Off + Cmd->Size; 1683 } 1684 1685 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() { 1686 uint64_t Off = 0; 1687 for (OutputSection *Sec : OutputSections) 1688 if (Sec->Flags & SHF_ALLOC) 1689 Off = setOffset(Sec, Off); 1690 FileSize = alignTo(Off, Config->Wordsize); 1691 } 1692 1693 // Assign file offsets to output sections. 1694 template <class ELFT> void Writer<ELFT>::assignFileOffsets() { 1695 uint64_t Off = 0; 1696 Off = setOffset(Out::ElfHeader, Off); 1697 Off = setOffset(Out::ProgramHeaders, Off); 1698 1699 PhdrEntry *LastRX = nullptr; 1700 for (PhdrEntry *P : Phdrs) 1701 if (P->p_type == PT_LOAD && (P->p_flags & PF_X)) 1702 LastRX = P; 1703 1704 for (OutputSection *Sec : OutputSections) { 1705 Off = setOffset(Sec, Off); 1706 if (Script->HasSectionsCommand) 1707 continue; 1708 // If this is a last section of the last executable segment and that 1709 // segment is the last loadable segment, align the offset of the 1710 // following section to avoid loading non-segments parts of the file. 1711 if (LastRX && LastRX->LastSec == Sec) 1712 Off = alignTo(Off, Target->PageSize); 1713 } 1714 1715 SectionHeaderOff = alignTo(Off, Config->Wordsize); 1716 FileSize = SectionHeaderOff + (OutputSections.size() + 1) * sizeof(Elf_Shdr); 1717 } 1718 1719 // Finalize the program headers. We call this function after we assign 1720 // file offsets and VAs to all sections. 1721 template <class ELFT> void Writer<ELFT>::setPhdrs() { 1722 for (PhdrEntry *P : Phdrs) { 1723 OutputSection *First = P->FirstSec; 1724 OutputSection *Last = P->LastSec; 1725 if (First) { 1726 P->p_filesz = Last->Offset - First->Offset; 1727 if (Last->Type != SHT_NOBITS) 1728 P->p_filesz += Last->Size; 1729 P->p_memsz = Last->Addr + Last->Size - First->Addr; 1730 P->p_offset = First->Offset; 1731 P->p_vaddr = First->Addr; 1732 if (!P->HasLMA) 1733 P->p_paddr = First->getLMA(); 1734 } 1735 if (P->p_type == PT_LOAD) 1736 P->p_align = std::max<uint64_t>(P->p_align, Config->MaxPageSize); 1737 else if (P->p_type == PT_GNU_RELRO) { 1738 P->p_align = 1; 1739 // The glibc dynamic loader rounds the size down, so we need to round up 1740 // to protect the last page. This is a no-op on FreeBSD which always 1741 // rounds up. 1742 P->p_memsz = alignTo(P->p_memsz, Target->PageSize); 1743 } 1744 1745 // The TLS pointer goes after PT_TLS. At least glibc will align it, 1746 // so round up the size to make sure the offsets are correct. 1747 if (P->p_type == PT_TLS) { 1748 Out::TlsPhdr = P; 1749 if (P->p_memsz) 1750 P->p_memsz = alignTo(P->p_memsz, P->p_align); 1751 } 1752 } 1753 } 1754 1755 // The entry point address is chosen in the following ways. 1756 // 1757 // 1. the '-e' entry command-line option; 1758 // 2. the ENTRY(symbol) command in a linker control script; 1759 // 3. the value of the symbol _start, if present; 1760 // 4. the number represented by the entry symbol, if it is a number; 1761 // 5. the address of the first byte of the .text section, if present; 1762 // 6. the address 0. 1763 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() { 1764 // Case 1, 2 or 3 1765 if (Symbol *B = Symtab->find(Config->Entry)) 1766 return B->getVA(); 1767 1768 // Case 4 1769 uint64_t Addr; 1770 if (to_integer(Config->Entry, Addr)) 1771 return Addr; 1772 1773 // Case 5 1774 if (OutputSection *Sec = findSection(".text")) { 1775 if (Config->WarnMissingEntry) 1776 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" + 1777 utohexstr(Sec->Addr)); 1778 return Sec->Addr; 1779 } 1780 1781 // Case 6 1782 if (Config->WarnMissingEntry) 1783 warn("cannot find entry symbol " + Config->Entry + 1784 "; not setting start address"); 1785 return 0; 1786 } 1787 1788 static uint16_t getELFType() { 1789 if (Config->Pic) 1790 return ET_DYN; 1791 if (Config->Relocatable) 1792 return ET_REL; 1793 return ET_EXEC; 1794 } 1795 1796 template <class ELFT> void Writer<ELFT>::writeHeader() { 1797 uint8_t *Buf = Buffer->getBufferStart(); 1798 memcpy(Buf, "\177ELF", 4); 1799 1800 // Write the ELF header. 1801 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf); 1802 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32; 1803 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB; 1804 EHdr->e_ident[EI_VERSION] = EV_CURRENT; 1805 EHdr->e_ident[EI_OSABI] = Config->OSABI; 1806 EHdr->e_type = getELFType(); 1807 EHdr->e_machine = Config->EMachine; 1808 EHdr->e_version = EV_CURRENT; 1809 EHdr->e_entry = getEntryAddr(); 1810 EHdr->e_shoff = SectionHeaderOff; 1811 EHdr->e_flags = Config->EFlags; 1812 EHdr->e_ehsize = sizeof(Elf_Ehdr); 1813 EHdr->e_phnum = Phdrs.size(); 1814 EHdr->e_shentsize = sizeof(Elf_Shdr); 1815 EHdr->e_shnum = OutputSections.size() + 1; 1816 EHdr->e_shstrndx = InX::ShStrTab->getParent()->SectionIndex; 1817 1818 if (!Config->Relocatable) { 1819 EHdr->e_phoff = sizeof(Elf_Ehdr); 1820 EHdr->e_phentsize = sizeof(Elf_Phdr); 1821 } 1822 1823 // Write the program header table. 1824 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff); 1825 for (PhdrEntry *P : Phdrs) { 1826 HBuf->p_type = P->p_type; 1827 HBuf->p_flags = P->p_flags; 1828 HBuf->p_offset = P->p_offset; 1829 HBuf->p_vaddr = P->p_vaddr; 1830 HBuf->p_paddr = P->p_paddr; 1831 HBuf->p_filesz = P->p_filesz; 1832 HBuf->p_memsz = P->p_memsz; 1833 HBuf->p_align = P->p_align; 1834 ++HBuf; 1835 } 1836 1837 // Write the section header table. Note that the first table entry is null. 1838 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff); 1839 for (OutputSection *Sec : OutputSections) 1840 Sec->writeHeaderTo<ELFT>(++SHdrs); 1841 } 1842 1843 // Open a result file. 1844 template <class ELFT> void Writer<ELFT>::openFile() { 1845 if (!Config->Is64 && FileSize > UINT32_MAX) { 1846 error("output file too large: " + Twine(FileSize) + " bytes"); 1847 return; 1848 } 1849 1850 unlinkAsync(Config->OutputFile); 1851 Expected<std::unique_ptr<FileOutputBuffer>> BufferOrErr = 1852 FileOutputBuffer::create(Config->OutputFile, FileSize, 1853 FileOutputBuffer::F_executable); 1854 1855 if (!BufferOrErr) 1856 error("failed to open " + Config->OutputFile + ": " + 1857 llvm::toString(BufferOrErr.takeError())); 1858 else 1859 Buffer = std::move(*BufferOrErr); 1860 } 1861 1862 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() { 1863 uint8_t *Buf = Buffer->getBufferStart(); 1864 for (OutputSection *Sec : OutputSections) 1865 if (Sec->Flags & SHF_ALLOC) 1866 Sec->writeTo<ELFT>(Buf + Sec->Offset); 1867 } 1868 1869 static void fillTrap(uint8_t *I, uint8_t *End) { 1870 for (; I + 4 <= End; I += 4) 1871 memcpy(I, &Target->TrapInstr, 4); 1872 } 1873 1874 // Fill the last page of executable segments with trap instructions 1875 // instead of leaving them as zero. Even though it is not required by any 1876 // standard, it is in general a good thing to do for security reasons. 1877 // 1878 // We'll leave other pages in segments as-is because the rest will be 1879 // overwritten by output sections. 1880 template <class ELFT> void Writer<ELFT>::writeTrapInstr() { 1881 if (Script->HasSectionsCommand) 1882 return; 1883 1884 // Fill the last page. 1885 uint8_t *Buf = Buffer->getBufferStart(); 1886 for (PhdrEntry *P : Phdrs) 1887 if (P->p_type == PT_LOAD && (P->p_flags & PF_X)) 1888 fillTrap(Buf + alignDown(P->p_offset + P->p_filesz, Target->PageSize), 1889 Buf + alignTo(P->p_offset + P->p_filesz, Target->PageSize)); 1890 1891 // Round up the file size of the last segment to the page boundary iff it is 1892 // an executable segment to ensure that other tools don't accidentally 1893 // trim the instruction padding (e.g. when stripping the file). 1894 PhdrEntry *Last = nullptr; 1895 for (PhdrEntry *P : Phdrs) 1896 if (P->p_type == PT_LOAD) 1897 Last = P; 1898 1899 if (Last && (Last->p_flags & PF_X)) 1900 Last->p_memsz = Last->p_filesz = alignTo(Last->p_filesz, Target->PageSize); 1901 } 1902 1903 // Write section contents to a mmap'ed file. 1904 template <class ELFT> void Writer<ELFT>::writeSections() { 1905 uint8_t *Buf = Buffer->getBufferStart(); 1906 1907 // PPC64 needs to process relocations in the .opd section 1908 // before processing relocations in code-containing sections. 1909 if (auto *OpdCmd = findSection(".opd")) { 1910 Out::Opd = OpdCmd; 1911 Out::OpdBuf = Buf + Out::Opd->Offset; 1912 OpdCmd->template writeTo<ELFT>(Buf + Out::Opd->Offset); 1913 } 1914 1915 OutputSection *EhFrameHdr = nullptr; 1916 if (InX::EhFrameHdr && !InX::EhFrameHdr->empty()) 1917 EhFrameHdr = InX::EhFrameHdr->getParent(); 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