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 "AArch64ErrataFix.h" 12 #include "Config.h" 13 #include "Filesystem.h" 14 #include "LinkerScript.h" 15 #include "MapFile.h" 16 #include "OutputSections.h" 17 #include "Relocations.h" 18 #include "Strings.h" 19 #include "SymbolTable.h" 20 #include "Symbols.h" 21 #include "SyntheticSections.h" 22 #include "Target.h" 23 #include "lld/Common/Memory.h" 24 #include "lld/Common/Threads.h" 25 #include "llvm/ADT/StringMap.h" 26 #include "llvm/ADT/StringSwitch.h" 27 #include <climits> 28 29 using namespace llvm; 30 using namespace llvm::ELF; 31 using namespace llvm::object; 32 using namespace llvm::support; 33 using namespace llvm::support::endian; 34 35 using namespace lld; 36 using namespace lld::elf; 37 38 namespace { 39 // The writer writes a SymbolTable result to a file. 40 template <class ELFT> class Writer { 41 public: 42 Writer() : Buffer(errorHandler().OutputBuffer) {} 43 typedef typename ELFT::Shdr Elf_Shdr; 44 typedef typename ELFT::Ehdr Elf_Ehdr; 45 typedef typename ELFT::Phdr Elf_Phdr; 46 47 void run(); 48 49 private: 50 void copyLocalSymbols(); 51 void addSectionSymbols(); 52 void forEachRelSec(std::function<void(InputSectionBase &)> Fn); 53 void sortSections(); 54 void resolveShfLinkOrder(); 55 void sortInputSections(); 56 void finalizeSections(); 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 80 std::vector<PhdrEntry *> Phdrs; 81 82 uint64_t FileSize; 83 uint64_t SectionHeaderOff; 84 85 bool HasGotBaseSym = false; 86 }; 87 } // anonymous namespace 88 89 StringRef elf::getOutputSectionName(InputSectionBase *S) { 90 // ".zdebug_" is a prefix for ZLIB-compressed sections. 91 // Because we decompressed input sections, we want to remove 'z'. 92 if (S->Name.startswith(".zdebug_")) 93 return Saver.save("." + S->Name.substr(2)); 94 95 if (Config->Relocatable) 96 return S->Name; 97 98 // This is for --emit-relocs. If .text.foo is emitted as .text.bar, we want 99 // to emit .rela.text.foo as .rela.text.bar for consistency (this is not 100 // technically required, but not doing it is odd). This code guarantees that. 101 if ((S->Type == SHT_REL || S->Type == SHT_RELA) && 102 !isa<SyntheticSection>(S)) { 103 OutputSection *Out = 104 cast<InputSection>(S)->getRelocatedSection()->getOutputSection(); 105 if (S->Type == SHT_RELA) 106 return Saver.save(".rela" + Out->Name); 107 return Saver.save(".rel" + Out->Name); 108 } 109 110 for (StringRef V : 111 {".text.", ".rodata.", ".data.rel.ro.", ".data.", ".bss.rel.ro.", 112 ".bss.", ".init_array.", ".fini_array.", ".ctors.", ".dtors.", ".tbss.", 113 ".gcc_except_table.", ".tdata.", ".ARM.exidx.", ".ARM.extab."}) { 114 StringRef Prefix = V.drop_back(); 115 if (S->Name.startswith(V) || S->Name == Prefix) 116 return Prefix; 117 } 118 119 // CommonSection is identified as "COMMON" in linker scripts. 120 // By default, it should go to .bss section. 121 if (S->Name == "COMMON") 122 return ".bss"; 123 124 return S->Name; 125 } 126 127 static bool needsInterpSection() { 128 return !SharedFiles.empty() && !Config->DynamicLinker.empty() && 129 Script->needsInterpSection(); 130 } 131 132 template <class ELFT> void elf::writeResult() { Writer<ELFT>().run(); } 133 134 template <class ELFT> void Writer<ELFT>::removeEmptyPTLoad() { 135 llvm::erase_if(Phdrs, [&](const PhdrEntry *P) { 136 if (P->p_type != PT_LOAD) 137 return false; 138 if (!P->FirstSec) 139 return true; 140 uint64_t Size = P->LastSec->Addr + P->LastSec->Size - P->FirstSec->Addr; 141 return Size == 0; 142 }); 143 } 144 145 template <class ELFT> static void combineEhFrameSections() { 146 for (InputSectionBase *&S : InputSections) { 147 EhInputSection *ES = dyn_cast<EhInputSection>(S); 148 if (!ES || !ES->Live) 149 continue; 150 151 InX::EhFrame->addSection<ELFT>(ES); 152 S = nullptr; 153 } 154 155 std::vector<InputSectionBase *> &V = InputSections; 156 V.erase(std::remove(V.begin(), V.end(), nullptr), V.end()); 157 } 158 159 static Defined *addOptionalRegular(StringRef Name, SectionBase *Sec, 160 uint64_t Val, uint8_t StOther = STV_HIDDEN, 161 uint8_t Binding = STB_GLOBAL) { 162 Symbol *S = Symtab->find(Name); 163 if (!S || S->isDefined()) 164 return nullptr; 165 Symbol *Sym = Symtab->addRegular(Name, StOther, STT_NOTYPE, Val, 166 /*Size=*/0, Binding, Sec, 167 /*File=*/nullptr); 168 return cast<Defined>(Sym); 169 } 170 171 // The linker is expected to define some symbols depending on 172 // the linking result. This function defines such symbols. 173 void elf::addReservedSymbols() { 174 if (Config->EMachine == EM_MIPS) { 175 // Define _gp for MIPS. st_value of _gp symbol will be updated by Writer 176 // so that it points to an absolute address which by default is relative 177 // to GOT. Default offset is 0x7ff0. 178 // See "Global Data Symbols" in Chapter 6 in the following document: 179 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf 180 ElfSym::MipsGp = Symtab->addAbsolute("_gp", STV_HIDDEN, STB_GLOBAL); 181 182 // On MIPS O32 ABI, _gp_disp is a magic symbol designates offset between 183 // start of function and 'gp' pointer into GOT. 184 if (Symtab->find("_gp_disp")) 185 ElfSym::MipsGpDisp = 186 Symtab->addAbsolute("_gp_disp", STV_HIDDEN, STB_GLOBAL); 187 188 // The __gnu_local_gp is a magic symbol equal to the current value of 'gp' 189 // pointer. This symbol is used in the code generated by .cpload pseudo-op 190 // in case of using -mno-shared option. 191 // https://sourceware.org/ml/binutils/2004-12/msg00094.html 192 if (Symtab->find("__gnu_local_gp")) 193 ElfSym::MipsLocalGp = 194 Symtab->addAbsolute("__gnu_local_gp", STV_HIDDEN, STB_GLOBAL); 195 } 196 197 ElfSym::GlobalOffsetTable = addOptionalRegular( 198 "_GLOBAL_OFFSET_TABLE_", Out::ElfHeader, Target->GotBaseSymOff); 199 200 // __ehdr_start is the location of ELF file headers. Note that we define 201 // this symbol unconditionally even when using a linker script, which 202 // differs from the behavior implemented by GNU linker which only define 203 // this symbol if ELF headers are in the memory mapped segment. 204 // __executable_start is not documented, but the expectation of at 205 // least the android libc is that it points to the elf header too. 206 // __dso_handle symbol is passed to cxa_finalize as a marker to identify 207 // each DSO. The address of the symbol doesn't matter as long as they are 208 // different in different DSOs, so we chose the start address of the DSO. 209 for (const char *Name : 210 {"__ehdr_start", "__executable_start", "__dso_handle"}) 211 addOptionalRegular(Name, Out::ElfHeader, 0, STV_HIDDEN); 212 213 // If linker script do layout we do not need to create any standart symbols. 214 if (Script->HasSectionsCommand) 215 return; 216 217 auto Add = [](StringRef S, int64_t Pos) { 218 return addOptionalRegular(S, Out::ElfHeader, Pos, STV_DEFAULT); 219 }; 220 221 ElfSym::Bss = Add("__bss_start", 0); 222 ElfSym::End1 = Add("end", -1); 223 ElfSym::End2 = Add("_end", -1); 224 ElfSym::Etext1 = Add("etext", -1); 225 ElfSym::Etext2 = Add("_etext", -1); 226 ElfSym::Edata1 = Add("edata", -1); 227 ElfSym::Edata2 = Add("_edata", -1); 228 } 229 230 static OutputSection *findSection(StringRef Name) { 231 for (BaseCommand *Base : Script->SectionCommands) 232 if (auto *Sec = dyn_cast<OutputSection>(Base)) 233 if (Sec->Name == Name) 234 return Sec; 235 return nullptr; 236 } 237 238 // Initialize Out members. 239 template <class ELFT> static void createSyntheticSections() { 240 // Initialize all pointers with NULL. This is needed because 241 // you can call lld::elf::main more than once as a library. 242 memset(&Out::First, 0, sizeof(Out)); 243 244 auto Add = [](InputSectionBase *Sec) { InputSections.push_back(Sec); }; 245 246 InX::DynStrTab = make<StringTableSection>(".dynstr", true); 247 InX::Dynamic = make<DynamicSection<ELFT>>(); 248 if (Config->AndroidPackDynRelocs) { 249 InX::RelaDyn = make<AndroidPackedRelocationSection<ELFT>>( 250 Config->IsRela ? ".rela.dyn" : ".rel.dyn"); 251 } else { 252 InX::RelaDyn = make<RelocationSection<ELFT>>( 253 Config->IsRela ? ".rela.dyn" : ".rel.dyn", Config->ZCombreloc); 254 } 255 InX::ShStrTab = make<StringTableSection>(".shstrtab", false); 256 257 Out::ProgramHeaders = make<OutputSection>("", 0, SHF_ALLOC); 258 Out::ProgramHeaders->Alignment = Config->Wordsize; 259 260 if (needsInterpSection()) { 261 InX::Interp = createInterpSection(); 262 Add(InX::Interp); 263 } else { 264 InX::Interp = nullptr; 265 } 266 267 if (Config->Strip != StripPolicy::All) { 268 InX::StrTab = make<StringTableSection>(".strtab", false); 269 InX::SymTab = make<SymbolTableSection<ELFT>>(*InX::StrTab); 270 } 271 272 if (Config->BuildId != BuildIdKind::None) { 273 InX::BuildId = make<BuildIdSection>(); 274 Add(InX::BuildId); 275 } 276 277 InX::Bss = make<BssSection>(".bss", 0, 1); 278 Add(InX::Bss); 279 280 // If there is a SECTIONS command and a .data.rel.ro section name use name 281 // .data.rel.ro.bss so that we match in the .data.rel.ro output section. 282 // This makes sure our relro is contiguous. 283 bool HasDataRelRo = 284 Script->HasSectionsCommand && findSection(".data.rel.ro"); 285 InX::BssRelRo = make<BssSection>( 286 HasDataRelRo ? ".data.rel.ro.bss" : ".bss.rel.ro", 0, 1); 287 Add(InX::BssRelRo); 288 289 // Add MIPS-specific sections. 290 if (Config->EMachine == EM_MIPS) { 291 if (!Config->Shared && Config->HasDynSymTab) { 292 InX::MipsRldMap = make<MipsRldMapSection>(); 293 Add(InX::MipsRldMap); 294 } 295 if (auto *Sec = MipsAbiFlagsSection<ELFT>::create()) 296 Add(Sec); 297 if (auto *Sec = MipsOptionsSection<ELFT>::create()) 298 Add(Sec); 299 if (auto *Sec = MipsReginfoSection<ELFT>::create()) 300 Add(Sec); 301 } 302 303 if (Config->HasDynSymTab) { 304 InX::DynSymTab = make<SymbolTableSection<ELFT>>(*InX::DynStrTab); 305 Add(InX::DynSymTab); 306 307 In<ELFT>::VerSym = make<VersionTableSection<ELFT>>(); 308 Add(In<ELFT>::VerSym); 309 310 if (!Config->VersionDefinitions.empty()) { 311 In<ELFT>::VerDef = make<VersionDefinitionSection<ELFT>>(); 312 Add(In<ELFT>::VerDef); 313 } 314 315 In<ELFT>::VerNeed = make<VersionNeedSection<ELFT>>(); 316 Add(In<ELFT>::VerNeed); 317 318 if (Config->GnuHash) { 319 InX::GnuHashTab = make<GnuHashTableSection>(); 320 Add(InX::GnuHashTab); 321 } 322 323 if (Config->SysvHash) { 324 InX::HashTab = make<HashTableSection>(); 325 Add(InX::HashTab); 326 } 327 328 Add(InX::Dynamic); 329 Add(InX::DynStrTab); 330 Add(InX::RelaDyn); 331 } 332 333 // Add .got. MIPS' .got is so different from the other archs, 334 // it has its own class. 335 if (Config->EMachine == EM_MIPS) { 336 InX::MipsGot = make<MipsGotSection>(); 337 Add(InX::MipsGot); 338 } else { 339 InX::Got = make<GotSection>(); 340 Add(InX::Got); 341 } 342 343 InX::GotPlt = make<GotPltSection>(); 344 Add(InX::GotPlt); 345 InX::IgotPlt = make<IgotPltSection>(); 346 Add(InX::IgotPlt); 347 348 if (Config->GdbIndex) { 349 InX::GdbIndex = createGdbIndex<ELFT>(); 350 Add(InX::GdbIndex); 351 } 352 353 // We always need to add rel[a].plt to output if it has entries. 354 // Even for static linking it can contain R_[*]_IRELATIVE relocations. 355 InX::RelaPlt = make<RelocationSection<ELFT>>( 356 Config->IsRela ? ".rela.plt" : ".rel.plt", false /*Sort*/); 357 Add(InX::RelaPlt); 358 359 // The RelaIplt immediately follows .rel.plt (.rel.dyn for ARM) to ensure 360 // that the IRelative relocations are processed last by the dynamic loader. 361 // We cannot place the iplt section in .rel.dyn when Android relocation 362 // packing is enabled because that would cause a section type mismatch. 363 // However, because the Android dynamic loader reads .rel.plt after .rel.dyn, 364 // we can get the desired behaviour by placing the iplt section in .rel.plt. 365 InX::RelaIplt = make<RelocationSection<ELFT>>( 366 (Config->EMachine == EM_ARM && !Config->AndroidPackDynRelocs) 367 ? ".rel.dyn" 368 : InX::RelaPlt->Name, 369 false /*Sort*/); 370 Add(InX::RelaIplt); 371 372 InX::Plt = make<PltSection>(false); 373 Add(InX::Plt); 374 InX::Iplt = make<PltSection>(true); 375 Add(InX::Iplt); 376 377 if (!Config->Relocatable) { 378 if (Config->EhFrameHdr) { 379 InX::EhFrameHdr = make<EhFrameHeader>(); 380 Add(InX::EhFrameHdr); 381 } 382 InX::EhFrame = make<EhFrameSection>(); 383 Add(InX::EhFrame); 384 } 385 386 if (InX::SymTab) 387 Add(InX::SymTab); 388 Add(InX::ShStrTab); 389 if (InX::StrTab) 390 Add(InX::StrTab); 391 392 if (Config->EMachine == EM_ARM && !Config->Relocatable) 393 // Add a sentinel to terminate .ARM.exidx. It helps an unwinder 394 // to find the exact address range of the last entry. 395 Add(make<ARMExidxSentinelSection>()); 396 } 397 398 // The main function of the writer. 399 template <class ELFT> void Writer<ELFT>::run() { 400 // Create linker-synthesized sections such as .got or .plt. 401 // Such sections are of type input section. 402 createSyntheticSections<ELFT>(); 403 404 if (!Config->Relocatable) 405 combineEhFrameSections<ELFT>(); 406 407 // We want to process linker script commands. When SECTIONS command 408 // is given we let it create sections. 409 Script->processSectionCommands(); 410 411 // Linker scripts controls how input sections are assigned to output sections. 412 // Input sections that were not handled by scripts are called "orphans", and 413 // they are assigned to output sections by the default rule. Process that. 414 Script->addOrphanSections(); 415 416 if (Config->Discard != DiscardPolicy::All) 417 copyLocalSymbols(); 418 419 if (Config->CopyRelocs) 420 addSectionSymbols(); 421 422 // Now that we have a complete set of output sections. This function 423 // completes section contents. For example, we need to add strings 424 // to the string table, and add entries to .got and .plt. 425 // finalizeSections does that. 426 finalizeSections(); 427 if (errorCount()) 428 return; 429 430 Script->assignAddresses(); 431 432 // If -compressed-debug-sections is specified, we need to compress 433 // .debug_* sections. Do it right now because it changes the size of 434 // output sections. 435 for (OutputSection *Sec : OutputSections) 436 Sec->maybeCompress<ELFT>(); 437 438 Script->allocateHeaders(Phdrs); 439 440 // Remove empty PT_LOAD to avoid causing the dynamic linker to try to mmap a 441 // 0 sized region. This has to be done late since only after assignAddresses 442 // we know the size of the sections. 443 removeEmptyPTLoad(); 444 445 if (!Config->OFormatBinary) 446 assignFileOffsets(); 447 else 448 assignFileOffsetsBinary(); 449 450 setPhdrs(); 451 452 if (Config->Relocatable) { 453 for (OutputSection *Sec : OutputSections) 454 Sec->Addr = 0; 455 } 456 457 // It does not make sense try to open the file if we have error already. 458 if (errorCount()) 459 return; 460 // Write the result down to a file. 461 openFile(); 462 if (errorCount()) 463 return; 464 465 if (!Config->OFormatBinary) { 466 writeTrapInstr(); 467 writeHeader(); 468 writeSections(); 469 } else { 470 writeSectionsBinary(); 471 } 472 473 // Backfill .note.gnu.build-id section content. This is done at last 474 // because the content is usually a hash value of the entire output file. 475 writeBuildId(); 476 if (errorCount()) 477 return; 478 479 // Handle -Map option. 480 writeMapFile(); 481 if (errorCount()) 482 return; 483 484 if (auto E = Buffer->commit()) 485 error("failed to write to the output file: " + toString(std::move(E))); 486 } 487 488 static bool shouldKeepInSymtab(SectionBase *Sec, StringRef SymName, 489 const Symbol &B) { 490 if (B.isFile() || B.isSection()) 491 return false; 492 493 // If sym references a section in a discarded group, don't keep it. 494 if (Sec == &InputSection::Discarded) 495 return false; 496 497 if (Config->Discard == DiscardPolicy::None) 498 return true; 499 500 // In ELF assembly .L symbols are normally discarded by the assembler. 501 // If the assembler fails to do so, the linker discards them if 502 // * --discard-locals is used. 503 // * The symbol is in a SHF_MERGE section, which is normally the reason for 504 // the assembler keeping the .L symbol. 505 if (!SymName.startswith(".L") && !SymName.empty()) 506 return true; 507 508 if (Config->Discard == DiscardPolicy::Locals) 509 return false; 510 511 return !Sec || !(Sec->Flags & SHF_MERGE); 512 } 513 514 static bool includeInSymtab(const Symbol &B) { 515 if (!B.isLocal() && !B.IsUsedInRegularObj) 516 return false; 517 518 if (auto *D = dyn_cast<Defined>(&B)) { 519 // Always include absolute symbols. 520 SectionBase *Sec = D->Section; 521 if (!Sec) 522 return true; 523 Sec = Sec->Repl; 524 // Exclude symbols pointing to garbage-collected sections. 525 if (isa<InputSectionBase>(Sec) && !Sec->Live) 526 return false; 527 if (auto *S = dyn_cast<MergeInputSection>(Sec)) 528 if (!S->getSectionPiece(D->Value)->Live) 529 return false; 530 return true; 531 } 532 return B.Used; 533 } 534 535 // Local symbols are not in the linker's symbol table. This function scans 536 // each object file's symbol table to copy local symbols to the output. 537 template <class ELFT> void Writer<ELFT>::copyLocalSymbols() { 538 if (!InX::SymTab) 539 return; 540 for (InputFile *File : ObjectFiles) { 541 ObjFile<ELFT> *F = cast<ObjFile<ELFT>>(File); 542 for (Symbol *B : F->getLocalSymbols()) { 543 if (!B->isLocal()) 544 fatal(toString(F) + 545 ": broken object: getLocalSymbols returns a non-local symbol"); 546 auto *DR = dyn_cast<Defined>(B); 547 548 // No reason to keep local undefined symbol in symtab. 549 if (!DR) 550 continue; 551 if (!includeInSymtab(*B)) 552 continue; 553 554 SectionBase *Sec = DR->Section; 555 if (!shouldKeepInSymtab(Sec, B->getName(), *B)) 556 continue; 557 InX::SymTab->addSymbol(B); 558 } 559 } 560 } 561 562 template <class ELFT> void Writer<ELFT>::addSectionSymbols() { 563 // Create a section symbol for each output section so that we can represent 564 // relocations that point to the section. If we know that no relocation is 565 // referring to a section (that happens if the section is a synthetic one), we 566 // don't create a section symbol for that section. 567 for (BaseCommand *Base : Script->SectionCommands) { 568 auto *Sec = dyn_cast<OutputSection>(Base); 569 if (!Sec) 570 continue; 571 auto I = llvm::find_if(Sec->SectionCommands, [](BaseCommand *Base) { 572 if (auto *ISD = dyn_cast<InputSectionDescription>(Base)) 573 return !ISD->Sections.empty(); 574 return false; 575 }); 576 if (I == Sec->SectionCommands.end()) 577 continue; 578 InputSection *IS = cast<InputSectionDescription>(*I)->Sections[0]; 579 580 // Relocations are not using REL[A] section symbols. 581 if (IS->Type == SHT_REL || IS->Type == SHT_RELA) 582 continue; 583 584 // Unlike other synthetic sections, mergeable output sections contain data 585 // copied from input sections, and there may be a relocation pointing to its 586 // contents if -r or -emit-reloc are given. 587 if (isa<SyntheticSection>(IS) && !(IS->Flags & SHF_MERGE)) 588 continue; 589 590 auto *Sym = 591 make<Defined>(IS->File, "", STB_LOCAL, /*StOther=*/0, STT_SECTION, 592 /*Value=*/0, /*Size=*/0, IS); 593 InX::SymTab->addSymbol(Sym); 594 } 595 } 596 597 // Today's loaders have a feature to make segments read-only after 598 // processing dynamic relocations to enhance security. PT_GNU_RELRO 599 // is defined for that. 600 // 601 // This function returns true if a section needs to be put into a 602 // PT_GNU_RELRO segment. 603 static bool isRelroSection(const OutputSection *Sec) { 604 if (!Config->ZRelro) 605 return false; 606 607 uint64_t Flags = Sec->Flags; 608 609 // Non-allocatable or non-writable sections don't need RELRO because 610 // they are not writable or not even mapped to memory in the first place. 611 // RELRO is for sections that are essentially read-only but need to 612 // be writable only at process startup to allow dynamic linker to 613 // apply relocations. 614 if (!(Flags & SHF_ALLOC) || !(Flags & SHF_WRITE)) 615 return false; 616 617 // Once initialized, TLS data segments are used as data templates 618 // for a thread-local storage. For each new thread, runtime 619 // allocates memory for a TLS and copy templates there. No thread 620 // are supposed to use templates directly. Thus, it can be in RELRO. 621 if (Flags & SHF_TLS) 622 return true; 623 624 // .init_array, .preinit_array and .fini_array contain pointers to 625 // functions that are executed on process startup or exit. These 626 // pointers are set by the static linker, and they are not expected 627 // to change at runtime. But if you are an attacker, you could do 628 // interesting things by manipulating pointers in .fini_array, for 629 // example. So they are put into RELRO. 630 uint32_t Type = Sec->Type; 631 if (Type == SHT_INIT_ARRAY || Type == SHT_FINI_ARRAY || 632 Type == SHT_PREINIT_ARRAY) 633 return true; 634 635 // .got contains pointers to external symbols. They are resolved by 636 // the dynamic linker when a module is loaded into memory, and after 637 // that they are not expected to change. So, it can be in RELRO. 638 if (InX::Got && Sec == InX::Got->getParent()) 639 return true; 640 641 // .got.plt contains pointers to external function symbols. They are 642 // by default resolved lazily, so we usually cannot put it into RELRO. 643 // However, if "-z now" is given, the lazy symbol resolution is 644 // disabled, which enables us to put it into RELRO. 645 if (Sec == InX::GotPlt->getParent()) 646 return Config->ZNow; 647 648 // .dynamic section contains data for the dynamic linker, and 649 // there's no need to write to it at runtime, so it's better to put 650 // it into RELRO. 651 if (Sec == InX::Dynamic->getParent()) 652 return true; 653 654 // Sections with some special names are put into RELRO. This is a 655 // bit unfortunate because section names shouldn't be significant in 656 // ELF in spirit. But in reality many linker features depend on 657 // magic section names. 658 StringRef S = Sec->Name; 659 return S == ".data.rel.ro" || S == ".bss.rel.ro" || S == ".ctors" || 660 S == ".dtors" || S == ".jcr" || S == ".eh_frame" || 661 S == ".openbsd.randomdata"; 662 } 663 664 // We compute a rank for each section. The rank indicates where the 665 // section should be placed in the file. Instead of using simple 666 // numbers (0,1,2...), we use a series of flags. One for each decision 667 // point when placing the section. 668 // Using flags has two key properties: 669 // * It is easy to check if a give branch was taken. 670 // * It is easy two see how similar two ranks are (see getRankProximity). 671 enum RankFlags { 672 RF_NOT_ADDR_SET = 1 << 16, 673 RF_NOT_INTERP = 1 << 15, 674 RF_NOT_ALLOC = 1 << 14, 675 RF_WRITE = 1 << 13, 676 RF_EXEC_WRITE = 1 << 12, 677 RF_EXEC = 1 << 11, 678 RF_NON_TLS_BSS = 1 << 10, 679 RF_NON_TLS_BSS_RO = 1 << 9, 680 RF_NOT_TLS = 1 << 8, 681 RF_BSS = 1 << 7, 682 RF_PPC_NOT_TOCBSS = 1 << 6, 683 RF_PPC_OPD = 1 << 5, 684 RF_PPC_TOCL = 1 << 4, 685 RF_PPC_TOC = 1 << 3, 686 RF_PPC_BRANCH_LT = 1 << 2, 687 RF_MIPS_GPREL = 1 << 1, 688 RF_MIPS_NOT_GOT = 1 << 0 689 }; 690 691 static unsigned getSectionRank(const OutputSection *Sec) { 692 unsigned Rank = 0; 693 694 // We want to put section specified by -T option first, so we 695 // can start assigning VA starting from them later. 696 if (Config->SectionStartMap.count(Sec->Name)) 697 return Rank; 698 Rank |= RF_NOT_ADDR_SET; 699 700 // Put .interp first because some loaders want to see that section 701 // on the first page of the executable file when loaded into memory. 702 if (Sec->Name == ".interp") 703 return Rank; 704 Rank |= RF_NOT_INTERP; 705 706 // Allocatable sections go first to reduce the total PT_LOAD size and 707 // so debug info doesn't change addresses in actual code. 708 if (!(Sec->Flags & SHF_ALLOC)) 709 return Rank | RF_NOT_ALLOC; 710 711 // Sort sections based on their access permission in the following 712 // order: R, RX, RWX, RW. This order is based on the following 713 // considerations: 714 // * Read-only sections come first such that they go in the 715 // PT_LOAD covering the program headers at the start of the file. 716 // * Read-only, executable sections come next, unless the 717 // -no-rosegment option is used. 718 // * Writable, executable sections follow such that .plt on 719 // architectures where it needs to be writable will be placed 720 // between .text and .data. 721 // * Writable sections come last, such that .bss lands at the very 722 // end of the last PT_LOAD. 723 bool IsExec = Sec->Flags & SHF_EXECINSTR; 724 bool IsWrite = Sec->Flags & SHF_WRITE; 725 726 if (IsExec) { 727 if (IsWrite) 728 Rank |= RF_EXEC_WRITE; 729 else if (!Config->SingleRoRx) 730 Rank |= RF_EXEC; 731 } else { 732 if (IsWrite) 733 Rank |= RF_WRITE; 734 } 735 736 // If we got here we know that both A and B are in the same PT_LOAD. 737 738 bool IsTls = Sec->Flags & SHF_TLS; 739 bool IsNoBits = Sec->Type == SHT_NOBITS; 740 741 // The first requirement we have is to put (non-TLS) nobits sections last. The 742 // reason is that the only thing the dynamic linker will see about them is a 743 // p_memsz that is larger than p_filesz. Seeing that it zeros the end of the 744 // PT_LOAD, so that has to correspond to the nobits sections. 745 bool IsNonTlsNoBits = IsNoBits && !IsTls; 746 if (IsNonTlsNoBits) 747 Rank |= RF_NON_TLS_BSS; 748 749 // We place nobits RelRo sections before plain r/w ones, and non-nobits RelRo 750 // sections after r/w ones, so that the RelRo sections are contiguous. 751 bool IsRelRo = isRelroSection(Sec); 752 if (IsNonTlsNoBits && !IsRelRo) 753 Rank |= RF_NON_TLS_BSS_RO; 754 if (!IsNonTlsNoBits && IsRelRo) 755 Rank |= RF_NON_TLS_BSS_RO; 756 757 // The TLS initialization block needs to be a single contiguous block in a R/W 758 // PT_LOAD, so stick TLS sections directly before the other RelRo R/W 759 // sections. The TLS NOBITS sections are placed here as they don't take up 760 // virtual address space in the PT_LOAD. 761 if (!IsTls) 762 Rank |= RF_NOT_TLS; 763 764 // Within the TLS initialization block, the non-nobits sections need to appear 765 // first. 766 if (IsNoBits) 767 Rank |= RF_BSS; 768 769 // Some architectures have additional ordering restrictions for sections 770 // within the same PT_LOAD. 771 if (Config->EMachine == EM_PPC64) { 772 // PPC64 has a number of special SHT_PROGBITS+SHF_ALLOC+SHF_WRITE sections 773 // that we would like to make sure appear is a specific order to maximize 774 // their coverage by a single signed 16-bit offset from the TOC base 775 // pointer. Conversely, the special .tocbss section should be first among 776 // all SHT_NOBITS sections. This will put it next to the loaded special 777 // PPC64 sections (and, thus, within reach of the TOC base pointer). 778 StringRef Name = Sec->Name; 779 if (Name != ".tocbss") 780 Rank |= RF_PPC_NOT_TOCBSS; 781 782 if (Name == ".opd") 783 Rank |= RF_PPC_OPD; 784 785 if (Name == ".toc1") 786 Rank |= RF_PPC_TOCL; 787 788 if (Name == ".toc") 789 Rank |= RF_PPC_TOC; 790 791 if (Name == ".branch_lt") 792 Rank |= RF_PPC_BRANCH_LT; 793 } 794 if (Config->EMachine == EM_MIPS) { 795 // All sections with SHF_MIPS_GPREL flag should be grouped together 796 // because data in these sections is addressable with a gp relative address. 797 if (Sec->Flags & SHF_MIPS_GPREL) 798 Rank |= RF_MIPS_GPREL; 799 800 if (Sec->Name != ".got") 801 Rank |= RF_MIPS_NOT_GOT; 802 } 803 804 return Rank; 805 } 806 807 static bool compareSections(const BaseCommand *ACmd, const BaseCommand *BCmd) { 808 const OutputSection *A = cast<OutputSection>(ACmd); 809 const OutputSection *B = cast<OutputSection>(BCmd); 810 if (A->SortRank != B->SortRank) 811 return A->SortRank < B->SortRank; 812 if (!(A->SortRank & RF_NOT_ADDR_SET)) 813 return Config->SectionStartMap.lookup(A->Name) < 814 Config->SectionStartMap.lookup(B->Name); 815 return false; 816 } 817 818 void PhdrEntry::add(OutputSection *Sec) { 819 LastSec = Sec; 820 if (!FirstSec) 821 FirstSec = Sec; 822 p_align = std::max(p_align, Sec->Alignment); 823 if (p_type == PT_LOAD) 824 Sec->PtLoad = this; 825 } 826 827 // The beginning and the ending of .rel[a].plt section are marked 828 // with __rel[a]_iplt_{start,end} symbols if it is a statically linked 829 // executable. The runtime needs these symbols in order to resolve 830 // all IRELATIVE relocs on startup. For dynamic executables, we don't 831 // need these symbols, since IRELATIVE relocs are resolved through GOT 832 // and PLT. For details, see http://www.airs.com/blog/archives/403. 833 template <class ELFT> void Writer<ELFT>::addRelIpltSymbols() { 834 if (!Config->Static) 835 return; 836 StringRef S = Config->IsRela ? "__rela_iplt_start" : "__rel_iplt_start"; 837 addOptionalRegular(S, InX::RelaIplt, 0, STV_HIDDEN, STB_WEAK); 838 839 S = Config->IsRela ? "__rela_iplt_end" : "__rel_iplt_end"; 840 addOptionalRegular(S, InX::RelaIplt, -1, STV_HIDDEN, STB_WEAK); 841 } 842 843 template <class ELFT> 844 void Writer<ELFT>::forEachRelSec(std::function<void(InputSectionBase &)> Fn) { 845 // Scan all relocations. Each relocation goes through a series 846 // of tests to determine if it needs special treatment, such as 847 // creating GOT, PLT, copy relocations, etc. 848 // Note that relocations for non-alloc sections are directly 849 // processed by InputSection::relocateNonAlloc. 850 for (InputSectionBase *IS : InputSections) 851 if (IS->Live && isa<InputSection>(IS) && (IS->Flags & SHF_ALLOC)) 852 Fn(*IS); 853 for (EhInputSection *ES : InX::EhFrame->Sections) 854 Fn(*ES); 855 } 856 857 // This function generates assignments for predefined symbols (e.g. _end or 858 // _etext) and inserts them into the commands sequence to be processed at the 859 // appropriate time. This ensures that the value is going to be correct by the 860 // time any references to these symbols are processed and is equivalent to 861 // defining these symbols explicitly in the linker script. 862 template <class ELFT> void Writer<ELFT>::setReservedSymbolSections() { 863 if (ElfSym::GlobalOffsetTable) { 864 // The _GLOBAL_OFFSET_TABLE_ symbol is defined by target convention to 865 // be at some offset from the base of the .got section, usually 0 or the end 866 // of the .got 867 InputSection *GotSection = InX::MipsGot ? cast<InputSection>(InX::MipsGot) 868 : cast<InputSection>(InX::Got); 869 ElfSym::GlobalOffsetTable->Section = GotSection; 870 } 871 872 PhdrEntry *Last = nullptr; 873 PhdrEntry *LastRO = nullptr; 874 875 for (PhdrEntry *P : Phdrs) { 876 if (P->p_type != PT_LOAD) 877 continue; 878 Last = P; 879 if (!(P->p_flags & PF_W)) 880 LastRO = P; 881 } 882 883 if (LastRO) { 884 // _etext is the first location after the last read-only loadable segment. 885 if (ElfSym::Etext1) 886 ElfSym::Etext1->Section = LastRO->LastSec; 887 if (ElfSym::Etext2) 888 ElfSym::Etext2->Section = LastRO->LastSec; 889 } 890 891 if (Last) { 892 // _edata points to the end of the last mapped initialized section. 893 OutputSection *Edata = nullptr; 894 for (OutputSection *OS : OutputSections) { 895 if (OS->Type != SHT_NOBITS) 896 Edata = OS; 897 if (OS == Last->LastSec) 898 break; 899 } 900 901 if (ElfSym::Edata1) 902 ElfSym::Edata1->Section = Edata; 903 if (ElfSym::Edata2) 904 ElfSym::Edata2->Section = Edata; 905 906 // _end is the first location after the uninitialized data region. 907 if (ElfSym::End1) 908 ElfSym::End1->Section = Last->LastSec; 909 if (ElfSym::End2) 910 ElfSym::End2->Section = Last->LastSec; 911 } 912 913 if (ElfSym::Bss) 914 ElfSym::Bss->Section = findSection(".bss"); 915 916 // Setup MIPS _gp_disp/__gnu_local_gp symbols which should 917 // be equal to the _gp symbol's value. 918 if (ElfSym::MipsGp) { 919 // Find GP-relative section with the lowest address 920 // and use this address to calculate default _gp value. 921 for (OutputSection *OS : OutputSections) { 922 if (OS->Flags & SHF_MIPS_GPREL) { 923 ElfSym::MipsGp->Section = OS; 924 ElfSym::MipsGp->Value = 0x7ff0; 925 break; 926 } 927 } 928 } 929 } 930 931 // We want to find how similar two ranks are. 932 // The more branches in getSectionRank that match, the more similar they are. 933 // Since each branch corresponds to a bit flag, we can just use 934 // countLeadingZeros. 935 static int getRankProximityAux(OutputSection *A, OutputSection *B) { 936 return countLeadingZeros(A->SortRank ^ B->SortRank); 937 } 938 939 static int getRankProximity(OutputSection *A, BaseCommand *B) { 940 if (auto *Sec = dyn_cast<OutputSection>(B)) 941 if (Sec->Live) 942 return getRankProximityAux(A, Sec); 943 return -1; 944 } 945 946 // When placing orphan sections, we want to place them after symbol assignments 947 // so that an orphan after 948 // begin_foo = .; 949 // foo : { *(foo) } 950 // end_foo = .; 951 // doesn't break the intended meaning of the begin/end symbols. 952 // We don't want to go over sections since findOrphanPos is the 953 // one in charge of deciding the order of the sections. 954 // We don't want to go over changes to '.', since doing so in 955 // rx_sec : { *(rx_sec) } 956 // . = ALIGN(0x1000); 957 // /* The RW PT_LOAD starts here*/ 958 // rw_sec : { *(rw_sec) } 959 // would mean that the RW PT_LOAD would become unaligned. 960 static bool shouldSkip(BaseCommand *Cmd) { 961 if (isa<OutputSection>(Cmd)) 962 return false; 963 if (auto *Assign = dyn_cast<SymbolAssignment>(Cmd)) 964 return Assign->Name != "."; 965 return true; 966 } 967 968 // We want to place orphan sections so that they share as much 969 // characteristics with their neighbors as possible. For example, if 970 // both are rw, or both are tls. 971 template <typename ELFT> 972 static std::vector<BaseCommand *>::iterator 973 findOrphanPos(std::vector<BaseCommand *>::iterator B, 974 std::vector<BaseCommand *>::iterator E) { 975 OutputSection *Sec = cast<OutputSection>(*E); 976 977 // Find the first element that has as close a rank as possible. 978 auto I = std::max_element(B, E, [=](BaseCommand *A, BaseCommand *B) { 979 return getRankProximity(Sec, A) < getRankProximity(Sec, B); 980 }); 981 if (I == E) 982 return E; 983 984 // Consider all existing sections with the same proximity. 985 int Proximity = getRankProximity(Sec, *I); 986 for (; I != E; ++I) { 987 auto *CurSec = dyn_cast<OutputSection>(*I); 988 if (!CurSec || !CurSec->Live) 989 continue; 990 if (getRankProximity(Sec, CurSec) != Proximity || 991 Sec->SortRank < CurSec->SortRank) 992 break; 993 } 994 995 auto IsLiveSection = [](BaseCommand *Cmd) { 996 auto *OS = dyn_cast<OutputSection>(Cmd); 997 return OS && OS->Live; 998 }; 999 1000 auto J = std::find_if(llvm::make_reverse_iterator(I), 1001 llvm::make_reverse_iterator(B), IsLiveSection); 1002 I = J.base(); 1003 1004 // As a special case, if the orphan section is the last section, put 1005 // it at the very end, past any other commands. 1006 // This matches bfd's behavior and is convenient when the linker script fully 1007 // specifies the start of the file, but doesn't care about the end (the non 1008 // alloc sections for example). 1009 auto NextSec = std::find_if(I, E, IsLiveSection); 1010 if (NextSec == E) 1011 return E; 1012 1013 while (I != E && shouldSkip(*I)) 1014 ++I; 1015 return I; 1016 } 1017 1018 // If no layout was provided by linker script, we want to apply default 1019 // sorting for special input sections and handle --symbol-ordering-file. 1020 template <class ELFT> void Writer<ELFT>::sortInputSections() { 1021 assert(!Script->HasSectionsCommand); 1022 1023 // Sort input sections by priority using the list provided 1024 // by --symbol-ordering-file. 1025 DenseMap<SectionBase *, int> Order = buildSectionOrder(); 1026 if (!Order.empty()) 1027 for (BaseCommand *Base : Script->SectionCommands) 1028 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1029 if (Sec->Live) 1030 Sec->sort([&](InputSectionBase *S) { return Order.lookup(S); }); 1031 1032 // Sort input sections by section name suffixes for 1033 // __attribute__((init_priority(N))). 1034 if (OutputSection *Sec = findSection(".init_array")) 1035 Sec->sortInitFini(); 1036 if (OutputSection *Sec = findSection(".fini_array")) 1037 Sec->sortInitFini(); 1038 1039 // Sort input sections by the special rule for .ctors and .dtors. 1040 if (OutputSection *Sec = findSection(".ctors")) 1041 Sec->sortCtorsDtors(); 1042 if (OutputSection *Sec = findSection(".dtors")) 1043 Sec->sortCtorsDtors(); 1044 } 1045 1046 template <class ELFT> void Writer<ELFT>::sortSections() { 1047 Script->adjustSectionsBeforeSorting(); 1048 1049 // Don't sort if using -r. It is not necessary and we want to preserve the 1050 // relative order for SHF_LINK_ORDER sections. 1051 if (Config->Relocatable) 1052 return; 1053 1054 for (BaseCommand *Base : Script->SectionCommands) 1055 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1056 Sec->SortRank = getSectionRank(Sec); 1057 1058 if (!Script->HasSectionsCommand) { 1059 sortInputSections(); 1060 1061 // We know that all the OutputSections are contiguous in this case. 1062 auto E = Script->SectionCommands.end(); 1063 auto I = Script->SectionCommands.begin(); 1064 auto IsSection = [](BaseCommand *Base) { return isa<OutputSection>(Base); }; 1065 I = std::find_if(I, E, IsSection); 1066 E = std::find_if(llvm::make_reverse_iterator(E), 1067 llvm::make_reverse_iterator(I), IsSection) 1068 .base(); 1069 std::stable_sort(I, E, compareSections); 1070 return; 1071 } 1072 1073 // Orphan sections are sections present in the input files which are 1074 // not explicitly placed into the output file by the linker script. 1075 // 1076 // The sections in the linker script are already in the correct 1077 // order. We have to figuere out where to insert the orphan 1078 // sections. 1079 // 1080 // The order of the sections in the script is arbitrary and may not agree with 1081 // compareSections. This means that we cannot easily define a strict weak 1082 // ordering. To see why, consider a comparison of a section in the script and 1083 // one not in the script. We have a two simple options: 1084 // * Make them equivalent (a is not less than b, and b is not less than a). 1085 // The problem is then that equivalence has to be transitive and we can 1086 // have sections a, b and c with only b in a script and a less than c 1087 // which breaks this property. 1088 // * Use compareSectionsNonScript. Given that the script order doesn't have 1089 // to match, we can end up with sections a, b, c, d where b and c are in the 1090 // script and c is compareSectionsNonScript less than b. In which case d 1091 // can be equivalent to c, a to b and d < a. As a concrete example: 1092 // .a (rx) # not in script 1093 // .b (rx) # in script 1094 // .c (ro) # in script 1095 // .d (ro) # not in script 1096 // 1097 // The way we define an order then is: 1098 // * Sort only the orphan sections. They are in the end right now. 1099 // * Move each orphan section to its preferred position. We try 1100 // to put each section in the last position where it it can share 1101 // a PT_LOAD. 1102 // 1103 // There is some ambiguity as to where exactly a new entry should be 1104 // inserted, because Commands contains not only output section 1105 // commands but also other types of commands such as symbol assignment 1106 // expressions. There's no correct answer here due to the lack of the 1107 // formal specification of the linker script. We use heuristics to 1108 // determine whether a new output command should be added before or 1109 // after another commands. For the details, look at shouldSkip 1110 // function. 1111 1112 auto I = Script->SectionCommands.begin(); 1113 auto E = Script->SectionCommands.end(); 1114 auto NonScriptI = std::find_if(I, E, [](BaseCommand *Base) { 1115 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1116 return Sec->Live && Sec->SectionIndex == INT_MAX; 1117 return false; 1118 }); 1119 1120 // Sort the orphan sections. 1121 std::stable_sort(NonScriptI, E, compareSections); 1122 1123 // As a horrible special case, skip the first . assignment if it is before any 1124 // section. We do this because it is common to set a load address by starting 1125 // the script with ". = 0xabcd" and the expectation is that every section is 1126 // after that. 1127 auto FirstSectionOrDotAssignment = 1128 std::find_if(I, E, [](BaseCommand *Cmd) { return !shouldSkip(Cmd); }); 1129 if (FirstSectionOrDotAssignment != E && 1130 isa<SymbolAssignment>(**FirstSectionOrDotAssignment)) 1131 ++FirstSectionOrDotAssignment; 1132 I = FirstSectionOrDotAssignment; 1133 1134 while (NonScriptI != E) { 1135 auto Pos = findOrphanPos<ELFT>(I, NonScriptI); 1136 OutputSection *Orphan = cast<OutputSection>(*NonScriptI); 1137 1138 // As an optimization, find all sections with the same sort rank 1139 // and insert them with one rotate. 1140 unsigned Rank = Orphan->SortRank; 1141 auto End = std::find_if(NonScriptI + 1, E, [=](BaseCommand *Cmd) { 1142 return cast<OutputSection>(Cmd)->SortRank != Rank; 1143 }); 1144 std::rotate(Pos, NonScriptI, End); 1145 NonScriptI = End; 1146 } 1147 1148 Script->adjustSectionsAfterSorting(); 1149 } 1150 1151 static bool compareByFilePosition(InputSection *A, InputSection *B) { 1152 // Synthetic, i. e. a sentinel section, should go last. 1153 if (A->kind() == InputSectionBase::Synthetic || 1154 B->kind() == InputSectionBase::Synthetic) 1155 return A->kind() != InputSectionBase::Synthetic; 1156 InputSection *LA = A->getLinkOrderDep(); 1157 InputSection *LB = B->getLinkOrderDep(); 1158 OutputSection *AOut = LA->getParent(); 1159 OutputSection *BOut = LB->getParent(); 1160 if (AOut != BOut) 1161 return AOut->SectionIndex < BOut->SectionIndex; 1162 return LA->OutSecOff < LB->OutSecOff; 1163 } 1164 1165 // This function is used by the --merge-exidx-entries to detect duplicate 1166 // .ARM.exidx sections. It is Arm only. 1167 // 1168 // The .ARM.exidx section is of the form: 1169 // | PREL31 offset to function | Unwind instructions for function | 1170 // where the unwind instructions are either a small number of unwind 1171 // instructions inlined into the table entry, the special CANT_UNWIND value of 1172 // 0x1 or a PREL31 offset into a .ARM.extab Section that contains unwind 1173 // instructions. 1174 // 1175 // We return true if all the unwind instructions in the .ARM.exidx entries of 1176 // Cur can be merged into the last entry of Prev. 1177 static bool isDuplicateArmExidxSec(InputSection *Prev, InputSection *Cur) { 1178 1179 // References to .ARM.Extab Sections have bit 31 clear and are not the 1180 // special EXIDX_CANTUNWIND bit-pattern. 1181 auto IsExtabRef = [](uint32_t Unwind) { 1182 return (Unwind & 0x80000000) == 0 && Unwind != 0x1; 1183 }; 1184 1185 struct ExidxEntry { 1186 ulittle32_t Fn; 1187 ulittle32_t Unwind; 1188 }; 1189 1190 // Get the last table Entry from the previous .ARM.exidx section. 1191 const ExidxEntry &PrevEntry = *reinterpret_cast<const ExidxEntry *>( 1192 Prev->Data.data() + Prev->getSize() - sizeof(ExidxEntry)); 1193 if (IsExtabRef(PrevEntry.Unwind)) 1194 return false; 1195 1196 // We consider the unwind instructions of an .ARM.exidx table entry 1197 // a duplicate if the previous unwind instructions if: 1198 // - Both are the special EXIDX_CANTUNWIND. 1199 // - Both are the same inline unwind instructions. 1200 // We do not attempt to follow and check links into .ARM.extab tables as 1201 // consecutive identical entries are rare and the effort to check that they 1202 // are identical is high. 1203 1204 if (isa<SyntheticSection>(Cur)) 1205 // Exidx sentinel section has implicit EXIDX_CANTUNWIND; 1206 return PrevEntry.Unwind == 0x1; 1207 1208 ArrayRef<const ExidxEntry> Entries( 1209 reinterpret_cast<const ExidxEntry *>(Cur->Data.data()), 1210 Cur->getSize() / sizeof(ExidxEntry)); 1211 for (const ExidxEntry &Entry : Entries) 1212 if (IsExtabRef(Entry.Unwind) || Entry.Unwind != PrevEntry.Unwind) 1213 return false; 1214 // All table entries in this .ARM.exidx Section can be merged into the 1215 // previous Section. 1216 return true; 1217 } 1218 1219 template <class ELFT> void Writer<ELFT>::resolveShfLinkOrder() { 1220 for (OutputSection *Sec : OutputSections) { 1221 if (!(Sec->Flags & SHF_LINK_ORDER)) 1222 continue; 1223 1224 // Link order may be distributed across several InputSectionDescriptions 1225 // but sort must consider them all at once. 1226 std::vector<InputSection **> ScriptSections; 1227 std::vector<InputSection *> Sections; 1228 for (BaseCommand *Base : Sec->SectionCommands) { 1229 if (auto *ISD = dyn_cast<InputSectionDescription>(Base)) { 1230 for (InputSection *&IS : ISD->Sections) { 1231 ScriptSections.push_back(&IS); 1232 Sections.push_back(IS); 1233 } 1234 } 1235 } 1236 std::stable_sort(Sections.begin(), Sections.end(), compareByFilePosition); 1237 1238 if (!Config->Relocatable && Config->EMachine == EM_ARM && 1239 Sec->Type == SHT_ARM_EXIDX) { 1240 1241 if (!Sections.empty() && isa<ARMExidxSentinelSection>(Sections.back())) { 1242 assert(Sections.size() >= 2 && 1243 "We should create a sentinel section only if there are " 1244 "alive regular exidx sections."); 1245 // The last executable section is required to fill the sentinel. 1246 // Remember it here so that we don't have to find it again. 1247 auto *Sentinel = cast<ARMExidxSentinelSection>(Sections.back()); 1248 Sentinel->Highest = Sections[Sections.size() - 2]->getLinkOrderDep(); 1249 } 1250 1251 if (Config->MergeArmExidx) { 1252 // The EHABI for the Arm Architecture permits consecutive identical 1253 // table entries to be merged. We use a simple implementation that 1254 // removes a .ARM.exidx Input Section if it can be merged into the 1255 // previous one. This does not require any rewriting of InputSection 1256 // contents but misses opportunities for fine grained deduplication 1257 // where only a subset of the InputSection contents can be merged. 1258 int Cur = 1; 1259 int Prev = 0; 1260 // The last one is a sentinel entry which should not be removed. 1261 int N = Sections.size() - 1; 1262 while (Cur < N) { 1263 if (isDuplicateArmExidxSec(Sections[Prev], Sections[Cur])) 1264 Sections[Cur] = nullptr; 1265 else 1266 Prev = Cur; 1267 ++Cur; 1268 } 1269 } 1270 } 1271 1272 for (int I = 0, N = Sections.size(); I < N; ++I) 1273 *ScriptSections[I] = Sections[I]; 1274 1275 // Remove the Sections we marked as duplicate earlier. 1276 for (BaseCommand *Base : Sec->SectionCommands) 1277 if (auto *ISD = dyn_cast<InputSectionDescription>(Base)) 1278 ISD->Sections.erase( 1279 std::remove(ISD->Sections.begin(), ISD->Sections.end(), nullptr), 1280 ISD->Sections.end()); 1281 } 1282 } 1283 1284 static void applySynthetic(const std::vector<SyntheticSection *> &Sections, 1285 std::function<void(SyntheticSection *)> Fn) { 1286 for (SyntheticSection *SS : Sections) 1287 if (SS && SS->getParent() && !SS->empty()) 1288 Fn(SS); 1289 } 1290 1291 // In order to allow users to manipulate linker-synthesized sections, 1292 // we had to add synthetic sections to the input section list early, 1293 // even before we make decisions whether they are needed. This allows 1294 // users to write scripts like this: ".mygot : { .got }". 1295 // 1296 // Doing it has an unintended side effects. If it turns out that we 1297 // don't need a .got (for example) at all because there's no 1298 // relocation that needs a .got, we don't want to emit .got. 1299 // 1300 // To deal with the above problem, this function is called after 1301 // scanRelocations is called to remove synthetic sections that turn 1302 // out to be empty. 1303 static void removeUnusedSyntheticSections() { 1304 // All input synthetic sections that can be empty are placed after 1305 // all regular ones. We iterate over them all and exit at first 1306 // non-synthetic. 1307 for (InputSectionBase *S : llvm::reverse(InputSections)) { 1308 SyntheticSection *SS = dyn_cast<SyntheticSection>(S); 1309 if (!SS) 1310 return; 1311 OutputSection *OS = SS->getParent(); 1312 if (!OS || !SS->empty()) 1313 continue; 1314 1315 std::vector<BaseCommand *>::iterator Empty = OS->SectionCommands.end(); 1316 for (auto I = OS->SectionCommands.begin(), E = OS->SectionCommands.end(); 1317 I != E; ++I) { 1318 BaseCommand *B = *I; 1319 if (auto *ISD = dyn_cast<InputSectionDescription>(B)) { 1320 llvm::erase_if(ISD->Sections, 1321 [=](InputSection *IS) { return IS == SS; }); 1322 if (ISD->Sections.empty()) 1323 Empty = I; 1324 } 1325 } 1326 if (Empty != OS->SectionCommands.end()) 1327 OS->SectionCommands.erase(Empty); 1328 1329 // If there are no other sections in the output section, remove it from the 1330 // output. 1331 if (OS->SectionCommands.empty()) 1332 OS->Live = false; 1333 } 1334 } 1335 1336 // Returns true if a symbol can be replaced at load-time by a symbol 1337 // with the same name defined in other ELF executable or DSO. 1338 static bool computeIsPreemptible(const Symbol &B) { 1339 assert(!B.isLocal()); 1340 // Only symbols that appear in dynsym can be preempted. 1341 if (!B.includeInDynsym()) 1342 return false; 1343 1344 // Only default visibility symbols can be preempted. 1345 if (B.Visibility != STV_DEFAULT) 1346 return false; 1347 1348 // At this point copy relocations have not been created yet, so any 1349 // symbol that is not defined locally is preemptible. 1350 if (!B.isDefined()) 1351 return true; 1352 1353 // If we have a dynamic list it specifies which local symbols are preemptible. 1354 if (Config->HasDynamicList) 1355 return false; 1356 1357 if (!Config->Shared) 1358 return false; 1359 1360 // -Bsymbolic means that definitions are not preempted. 1361 if (Config->Bsymbolic || (Config->BsymbolicFunctions && B.isFunc())) 1362 return false; 1363 return true; 1364 } 1365 1366 // Create output section objects and add them to OutputSections. 1367 template <class ELFT> void Writer<ELFT>::finalizeSections() { 1368 Out::DebugInfo = findSection(".debug_info"); 1369 Out::PreinitArray = findSection(".preinit_array"); 1370 Out::InitArray = findSection(".init_array"); 1371 Out::FiniArray = findSection(".fini_array"); 1372 1373 // The linker needs to define SECNAME_start, SECNAME_end and SECNAME_stop 1374 // symbols for sections, so that the runtime can get the start and end 1375 // addresses of each section by section name. Add such symbols. 1376 if (!Config->Relocatable) { 1377 addStartEndSymbols(); 1378 for (BaseCommand *Base : Script->SectionCommands) 1379 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1380 addStartStopSymbols(Sec); 1381 } 1382 1383 // Add _DYNAMIC symbol. Unlike GNU gold, our _DYNAMIC symbol has no type. 1384 // It should be okay as no one seems to care about the type. 1385 // Even the author of gold doesn't remember why gold behaves that way. 1386 // https://sourceware.org/ml/binutils/2002-03/msg00360.html 1387 if (InX::DynSymTab) 1388 Symtab->addRegular("_DYNAMIC", STV_HIDDEN, STT_NOTYPE, 0 /*Value*/, 1389 /*Size=*/0, STB_WEAK, InX::Dynamic, 1390 /*File=*/nullptr); 1391 1392 // Define __rel[a]_iplt_{start,end} symbols if needed. 1393 addRelIpltSymbols(); 1394 1395 // This responsible for splitting up .eh_frame section into 1396 // pieces. The relocation scan uses those pieces, so this has to be 1397 // earlier. 1398 applySynthetic({InX::EhFrame}, 1399 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1400 1401 for (Symbol *S : Symtab->getSymbols()) 1402 S->IsPreemptible |= computeIsPreemptible(*S); 1403 1404 // Scan relocations. This must be done after every symbol is declared so that 1405 // we can correctly decide if a dynamic relocation is needed. 1406 if (!Config->Relocatable) 1407 forEachRelSec(scanRelocations<ELFT>); 1408 1409 if (InX::Plt && !InX::Plt->empty()) 1410 InX::Plt->addSymbols(); 1411 if (InX::Iplt && !InX::Iplt->empty()) 1412 InX::Iplt->addSymbols(); 1413 1414 // Now that we have defined all possible global symbols including linker- 1415 // synthesized ones. Visit all symbols to give the finishing touches. 1416 for (Symbol *Sym : Symtab->getSymbols()) { 1417 if (!includeInSymtab(*Sym)) 1418 continue; 1419 if (InX::SymTab) 1420 InX::SymTab->addSymbol(Sym); 1421 1422 if (InX::DynSymTab && Sym->includeInDynsym()) { 1423 InX::DynSymTab->addSymbol(Sym); 1424 if (auto *SS = dyn_cast<SharedSymbol>(Sym)) 1425 if (cast<SharedFile<ELFT>>(Sym->File)->IsNeeded) 1426 In<ELFT>::VerNeed->addSymbol(SS); 1427 } 1428 } 1429 1430 // Do not proceed if there was an undefined symbol. 1431 if (errorCount()) 1432 return; 1433 1434 removeUnusedSyntheticSections(); 1435 1436 sortSections(); 1437 Script->removeEmptyCommands(); 1438 1439 // Now that we have the final list, create a list of all the 1440 // OutputSections for convenience. 1441 for (BaseCommand *Base : Script->SectionCommands) 1442 if (auto *Sec = dyn_cast<OutputSection>(Base)) 1443 OutputSections.push_back(Sec); 1444 1445 // Prefer command line supplied address over other constraints. 1446 for (OutputSection *Sec : OutputSections) { 1447 auto I = Config->SectionStartMap.find(Sec->Name); 1448 if (I != Config->SectionStartMap.end()) 1449 Sec->AddrExpr = [=] { return I->second; }; 1450 } 1451 1452 // This is a bit of a hack. A value of 0 means undef, so we set it 1453 // to 1 t make __ehdr_start defined. The section number is not 1454 // particularly relevant. 1455 Out::ElfHeader->SectionIndex = 1; 1456 1457 unsigned I = 1; 1458 for (OutputSection *Sec : OutputSections) { 1459 Sec->SectionIndex = I++; 1460 Sec->ShName = InX::ShStrTab->addString(Sec->Name); 1461 } 1462 1463 // Binary and relocatable output does not have PHDRS. 1464 // The headers have to be created before finalize as that can influence the 1465 // image base and the dynamic section on mips includes the image base. 1466 if (!Config->Relocatable && !Config->OFormatBinary) { 1467 Phdrs = Script->hasPhdrsCommands() ? Script->createPhdrs() : createPhdrs(); 1468 addPtArmExid(Phdrs); 1469 Out::ProgramHeaders->Size = sizeof(Elf_Phdr) * Phdrs.size(); 1470 } 1471 1472 // Some symbols are defined in term of program headers. Now that we 1473 // have the headers, we can find out which sections they point to. 1474 setReservedSymbolSections(); 1475 1476 // Dynamic section must be the last one in this list and dynamic 1477 // symbol table section (DynSymTab) must be the first one. 1478 applySynthetic( 1479 {InX::DynSymTab, InX::Bss, InX::BssRelRo, InX::GnuHashTab, 1480 InX::HashTab, InX::SymTab, InX::ShStrTab, InX::StrTab, 1481 In<ELFT>::VerDef, InX::DynStrTab, InX::Got, InX::MipsGot, 1482 InX::IgotPlt, InX::GotPlt, InX::RelaDyn, InX::RelaIplt, 1483 InX::RelaPlt, InX::Plt, InX::Iplt, InX::EhFrameHdr, 1484 In<ELFT>::VerSym, In<ELFT>::VerNeed, InX::Dynamic}, 1485 [](SyntheticSection *SS) { SS->finalizeContents(); }); 1486 1487 if (!Script->HasSectionsCommand && !Config->Relocatable) 1488 fixSectionAlignments(); 1489 1490 // After link order processing .ARM.exidx sections can be deduplicated, which 1491 // needs to be resolved before any other address dependent operation. 1492 resolveShfLinkOrder(); 1493 1494 // Some architectures need to generate content that depends on the address 1495 // of InputSections. For example some architectures use small displacements 1496 // for jump instructions that is is the linker's responsibility for creating 1497 // range extension thunks for. As the generation of the content may also 1498 // alter InputSection addresses we must converge to a fixed point. 1499 if (Target->NeedsThunks || Config->AndroidPackDynRelocs) { 1500 ThunkCreator TC; 1501 AArch64Err843419Patcher A64P; 1502 bool Changed; 1503 do { 1504 Script->assignAddresses(); 1505 Changed = false; 1506 if (Target->NeedsThunks) 1507 Changed |= TC.createThunks(OutputSections); 1508 if (Config->FixCortexA53Errata843419) { 1509 if (Changed) 1510 Script->assignAddresses(); 1511 Changed |= A64P.createFixes(); 1512 } 1513 if (InX::MipsGot) 1514 InX::MipsGot->updateAllocSize(); 1515 Changed |= InX::RelaDyn->updateAllocSize(); 1516 } while (Changed); 1517 } 1518 1519 // Fill other section headers. The dynamic table is finalized 1520 // at the end because some tags like RELSZ depend on result 1521 // of finalizing other sections. 1522 for (OutputSection *Sec : OutputSections) 1523 Sec->finalize<ELFT>(); 1524 1525 // createThunks may have added local symbols to the static symbol table 1526 applySynthetic({InX::SymTab}, 1527 [](SyntheticSection *SS) { SS->postThunkContents(); }); 1528 } 1529 1530 // The linker is expected to define SECNAME_start and SECNAME_end 1531 // symbols for a few sections. This function defines them. 1532 template <class ELFT> void Writer<ELFT>::addStartEndSymbols() { 1533 auto Define = [&](StringRef Start, StringRef End, OutputSection *OS) { 1534 // These symbols resolve to the image base if the section does not exist. 1535 // A special value -1 indicates end of the section. 1536 if (OS) { 1537 addOptionalRegular(Start, OS, 0); 1538 addOptionalRegular(End, OS, -1); 1539 } else { 1540 if (Config->Pic) 1541 OS = Out::ElfHeader; 1542 addOptionalRegular(Start, OS, 0); 1543 addOptionalRegular(End, OS, 0); 1544 } 1545 }; 1546 1547 Define("__preinit_array_start", "__preinit_array_end", Out::PreinitArray); 1548 Define("__init_array_start", "__init_array_end", Out::InitArray); 1549 Define("__fini_array_start", "__fini_array_end", Out::FiniArray); 1550 1551 if (OutputSection *Sec = findSection(".ARM.exidx")) 1552 Define("__exidx_start", "__exidx_end", Sec); 1553 } 1554 1555 // If a section name is valid as a C identifier (which is rare because of 1556 // the leading '.'), linkers are expected to define __start_<secname> and 1557 // __stop_<secname> symbols. They are at beginning and end of the section, 1558 // respectively. This is not requested by the ELF standard, but GNU ld and 1559 // gold provide the feature, and used by many programs. 1560 template <class ELFT> 1561 void Writer<ELFT>::addStartStopSymbols(OutputSection *Sec) { 1562 StringRef S = Sec->Name; 1563 if (!isValidCIdentifier(S)) 1564 return; 1565 addOptionalRegular(Saver.save("__start_" + S), Sec, 0, STV_DEFAULT); 1566 addOptionalRegular(Saver.save("__stop_" + S), Sec, -1, STV_DEFAULT); 1567 } 1568 1569 static bool needsPtLoad(OutputSection *Sec) { 1570 if (!(Sec->Flags & SHF_ALLOC)) 1571 return false; 1572 1573 // Don't allocate VA space for TLS NOBITS sections. The PT_TLS PHDR is 1574 // responsible for allocating space for them, not the PT_LOAD that 1575 // contains the TLS initialization image. 1576 if (Sec->Flags & SHF_TLS && Sec->Type == SHT_NOBITS) 1577 return false; 1578 return true; 1579 } 1580 1581 // Linker scripts are responsible for aligning addresses. Unfortunately, most 1582 // linker scripts are designed for creating two PT_LOADs only, one RX and one 1583 // RW. This means that there is no alignment in the RO to RX transition and we 1584 // cannot create a PT_LOAD there. 1585 static uint64_t computeFlags(uint64_t Flags) { 1586 if (Config->Omagic) 1587 return PF_R | PF_W | PF_X; 1588 if (Config->SingleRoRx && !(Flags & PF_W)) 1589 return Flags | PF_X; 1590 return Flags; 1591 } 1592 1593 // Decide which program headers to create and which sections to include in each 1594 // one. 1595 template <class ELFT> std::vector<PhdrEntry *> Writer<ELFT>::createPhdrs() { 1596 std::vector<PhdrEntry *> Ret; 1597 auto AddHdr = [&](unsigned Type, unsigned Flags) -> PhdrEntry * { 1598 Ret.push_back(make<PhdrEntry>(Type, Flags)); 1599 return Ret.back(); 1600 }; 1601 1602 // The first phdr entry is PT_PHDR which describes the program header itself. 1603 AddHdr(PT_PHDR, PF_R)->add(Out::ProgramHeaders); 1604 1605 // PT_INTERP must be the second entry if exists. 1606 if (OutputSection *Cmd = findSection(".interp")) 1607 AddHdr(PT_INTERP, Cmd->getPhdrFlags())->add(Cmd); 1608 1609 // Add the first PT_LOAD segment for regular output sections. 1610 uint64_t Flags = computeFlags(PF_R); 1611 PhdrEntry *Load = AddHdr(PT_LOAD, Flags); 1612 1613 // Add the headers. We will remove them if they don't fit. 1614 Load->add(Out::ElfHeader); 1615 Load->add(Out::ProgramHeaders); 1616 1617 for (OutputSection *Sec : OutputSections) { 1618 if (!(Sec->Flags & SHF_ALLOC)) 1619 break; 1620 if (!needsPtLoad(Sec)) 1621 continue; 1622 1623 // Segments are contiguous memory regions that has the same attributes 1624 // (e.g. executable or writable). There is one phdr for each segment. 1625 // Therefore, we need to create a new phdr when the next section has 1626 // different flags or is loaded at a discontiguous address using AT linker 1627 // script command. 1628 uint64_t NewFlags = computeFlags(Sec->getPhdrFlags()); 1629 if (Sec->LMAExpr || Flags != NewFlags) { 1630 Load = AddHdr(PT_LOAD, NewFlags); 1631 Flags = NewFlags; 1632 } 1633 1634 Load->add(Sec); 1635 } 1636 1637 // Add a TLS segment if any. 1638 PhdrEntry *TlsHdr = make<PhdrEntry>(PT_TLS, PF_R); 1639 for (OutputSection *Sec : OutputSections) 1640 if (Sec->Flags & SHF_TLS) 1641 TlsHdr->add(Sec); 1642 if (TlsHdr->FirstSec) 1643 Ret.push_back(TlsHdr); 1644 1645 // Add an entry for .dynamic. 1646 if (InX::DynSymTab) 1647 AddHdr(PT_DYNAMIC, InX::Dynamic->getParent()->getPhdrFlags()) 1648 ->add(InX::Dynamic->getParent()); 1649 1650 // PT_GNU_RELRO includes all sections that should be marked as 1651 // read-only by dynamic linker after proccessing relocations. 1652 // Current dynamic loaders only support one PT_GNU_RELRO PHDR, give 1653 // an error message if more than one PT_GNU_RELRO PHDR is required. 1654 PhdrEntry *RelRo = make<PhdrEntry>(PT_GNU_RELRO, PF_R); 1655 bool InRelroPhdr = false; 1656 bool IsRelroFinished = false; 1657 for (OutputSection *Sec : OutputSections) { 1658 if (!needsPtLoad(Sec)) 1659 continue; 1660 if (isRelroSection(Sec)) { 1661 InRelroPhdr = true; 1662 if (!IsRelroFinished) 1663 RelRo->add(Sec); 1664 else 1665 error("section: " + Sec->Name + " is not contiguous with other relro" + 1666 " sections"); 1667 } else if (InRelroPhdr) { 1668 InRelroPhdr = false; 1669 IsRelroFinished = true; 1670 } 1671 } 1672 if (RelRo->FirstSec) 1673 Ret.push_back(RelRo); 1674 1675 // PT_GNU_EH_FRAME is a special section pointing on .eh_frame_hdr. 1676 if (!InX::EhFrame->empty() && InX::EhFrameHdr && InX::EhFrame->getParent() && 1677 InX::EhFrameHdr->getParent()) 1678 AddHdr(PT_GNU_EH_FRAME, InX::EhFrameHdr->getParent()->getPhdrFlags()) 1679 ->add(InX::EhFrameHdr->getParent()); 1680 1681 // PT_OPENBSD_RANDOMIZE is an OpenBSD-specific feature. That makes 1682 // the dynamic linker fill the segment with random data. 1683 if (OutputSection *Cmd = findSection(".openbsd.randomdata")) 1684 AddHdr(PT_OPENBSD_RANDOMIZE, Cmd->getPhdrFlags())->add(Cmd); 1685 1686 // PT_GNU_STACK is a special section to tell the loader to make the 1687 // pages for the stack non-executable. If you really want an executable 1688 // stack, you can pass -z execstack, but that's not recommended for 1689 // security reasons. 1690 unsigned Perm; 1691 if (Config->ZExecstack) 1692 Perm = PF_R | PF_W | PF_X; 1693 else 1694 Perm = PF_R | PF_W; 1695 AddHdr(PT_GNU_STACK, Perm)->p_memsz = Config->ZStackSize; 1696 1697 // PT_OPENBSD_WXNEEDED is a OpenBSD-specific header to mark the executable 1698 // is expected to perform W^X violations, such as calling mprotect(2) or 1699 // mmap(2) with PROT_WRITE | PROT_EXEC, which is prohibited by default on 1700 // OpenBSD. 1701 if (Config->ZWxneeded) 1702 AddHdr(PT_OPENBSD_WXNEEDED, PF_X); 1703 1704 // Create one PT_NOTE per a group of contiguous .note sections. 1705 PhdrEntry *Note = nullptr; 1706 for (OutputSection *Sec : OutputSections) { 1707 if (Sec->Type == SHT_NOTE) { 1708 if (!Note || Sec->LMAExpr) 1709 Note = AddHdr(PT_NOTE, PF_R); 1710 Note->add(Sec); 1711 } else { 1712 Note = nullptr; 1713 } 1714 } 1715 return Ret; 1716 } 1717 1718 template <class ELFT> 1719 void Writer<ELFT>::addPtArmExid(std::vector<PhdrEntry *> &Phdrs) { 1720 if (Config->EMachine != EM_ARM) 1721 return; 1722 auto I = llvm::find_if(OutputSections, [](OutputSection *Cmd) { 1723 return Cmd->Type == SHT_ARM_EXIDX; 1724 }); 1725 if (I == OutputSections.end()) 1726 return; 1727 1728 // PT_ARM_EXIDX is the ARM EHABI equivalent of PT_GNU_EH_FRAME 1729 PhdrEntry *ARMExidx = make<PhdrEntry>(PT_ARM_EXIDX, PF_R); 1730 ARMExidx->add(*I); 1731 Phdrs.push_back(ARMExidx); 1732 } 1733 1734 // The first section of each PT_LOAD, the first section in PT_GNU_RELRO and the 1735 // first section after PT_GNU_RELRO have to be page aligned so that the dynamic 1736 // linker can set the permissions. 1737 template <class ELFT> void Writer<ELFT>::fixSectionAlignments() { 1738 auto PageAlign = [](OutputSection *Cmd) { 1739 if (Cmd && !Cmd->AddrExpr) 1740 Cmd->AddrExpr = [=] { 1741 return alignTo(Script->getDot(), Config->MaxPageSize); 1742 }; 1743 }; 1744 1745 for (const PhdrEntry *P : Phdrs) 1746 if (P->p_type == PT_LOAD && P->FirstSec) 1747 PageAlign(P->FirstSec); 1748 1749 for (const PhdrEntry *P : Phdrs) { 1750 if (P->p_type != PT_GNU_RELRO) 1751 continue; 1752 if (P->FirstSec) 1753 PageAlign(P->FirstSec); 1754 // Find the first section after PT_GNU_RELRO. If it is in a PT_LOAD we 1755 // have to align it to a page. 1756 auto End = OutputSections.end(); 1757 auto I = std::find(OutputSections.begin(), End, P->LastSec); 1758 if (I == End || (I + 1) == End) 1759 continue; 1760 OutputSection *Cmd = (*(I + 1)); 1761 if (needsPtLoad(Cmd)) 1762 PageAlign(Cmd); 1763 } 1764 } 1765 1766 // Adjusts the file alignment for a given output section and returns 1767 // its new file offset. The file offset must be the same with its 1768 // virtual address (modulo the page size) so that the loader can load 1769 // executables without any address adjustment. 1770 static uint64_t getFileAlignment(uint64_t Off, OutputSection *Cmd) { 1771 OutputSection *First = Cmd->PtLoad ? Cmd->PtLoad->FirstSec : nullptr; 1772 // The first section in a PT_LOAD has to have congruent offset and address 1773 // module the page size. 1774 if (Cmd == First) 1775 return alignTo(Off, std::max<uint64_t>(Cmd->Alignment, Config->MaxPageSize), 1776 Cmd->Addr); 1777 1778 // For SHT_NOBITS we don't want the alignment of the section to impact the 1779 // offset of the sections that follow. Since nothing seems to care about the 1780 // sh_offset of the SHT_NOBITS section itself, just ignore it. 1781 if (Cmd->Type == SHT_NOBITS) 1782 return Off; 1783 1784 // If the section is not in a PT_LOAD, we just have to align it. 1785 if (!Cmd->PtLoad) 1786 return alignTo(Off, Cmd->Alignment); 1787 1788 // If two sections share the same PT_LOAD the file offset is calculated 1789 // using this formula: Off2 = Off1 + (VA2 - VA1). 1790 return First->Offset + Cmd->Addr - First->Addr; 1791 } 1792 1793 static uint64_t setOffset(OutputSection *Cmd, uint64_t Off) { 1794 Off = getFileAlignment(Off, Cmd); 1795 Cmd->Offset = Off; 1796 1797 // For SHT_NOBITS we should not count the size. 1798 if (Cmd->Type == SHT_NOBITS) 1799 return Off; 1800 1801 return Off + Cmd->Size; 1802 } 1803 1804 template <class ELFT> void Writer<ELFT>::assignFileOffsetsBinary() { 1805 uint64_t Off = 0; 1806 for (OutputSection *Sec : OutputSections) 1807 if (Sec->Flags & SHF_ALLOC) 1808 Off = setOffset(Sec, Off); 1809 FileSize = alignTo(Off, Config->Wordsize); 1810 } 1811 1812 // Assign file offsets to output sections. 1813 template <class ELFT> void Writer<ELFT>::assignFileOffsets() { 1814 uint64_t Off = 0; 1815 Off = setOffset(Out::ElfHeader, Off); 1816 Off = setOffset(Out::ProgramHeaders, Off); 1817 1818 PhdrEntry *LastRX = nullptr; 1819 for (PhdrEntry *P : Phdrs) 1820 if (P->p_type == PT_LOAD && (P->p_flags & PF_X)) 1821 LastRX = P; 1822 1823 for (OutputSection *Sec : OutputSections) { 1824 Off = setOffset(Sec, Off); 1825 if (Script->HasSectionsCommand) 1826 continue; 1827 // If this is a last section of the last executable segment and that 1828 // segment is the last loadable segment, align the offset of the 1829 // following section to avoid loading non-segments parts of the file. 1830 if (LastRX && LastRX->LastSec == Sec) 1831 Off = alignTo(Off, Target->PageSize); 1832 } 1833 1834 SectionHeaderOff = alignTo(Off, Config->Wordsize); 1835 FileSize = SectionHeaderOff + (OutputSections.size() + 1) * sizeof(Elf_Shdr); 1836 } 1837 1838 // Finalize the program headers. We call this function after we assign 1839 // file offsets and VAs to all sections. 1840 template <class ELFT> void Writer<ELFT>::setPhdrs() { 1841 for (PhdrEntry *P : Phdrs) { 1842 OutputSection *First = P->FirstSec; 1843 OutputSection *Last = P->LastSec; 1844 if (First) { 1845 P->p_filesz = Last->Offset - First->Offset; 1846 if (Last->Type != SHT_NOBITS) 1847 P->p_filesz += Last->Size; 1848 P->p_memsz = Last->Addr + Last->Size - First->Addr; 1849 P->p_offset = First->Offset; 1850 P->p_vaddr = First->Addr; 1851 if (!P->HasLMA) 1852 P->p_paddr = First->getLMA(); 1853 } 1854 if (P->p_type == PT_LOAD) 1855 P->p_align = std::max<uint64_t>(P->p_align, Config->MaxPageSize); 1856 else if (P->p_type == PT_GNU_RELRO) { 1857 P->p_align = 1; 1858 // The glibc dynamic loader rounds the size down, so we need to round up 1859 // to protect the last page. This is a no-op on FreeBSD which always 1860 // rounds up. 1861 P->p_memsz = alignTo(P->p_memsz, Target->PageSize); 1862 } 1863 1864 // The TLS pointer goes after PT_TLS. At least glibc will align it, 1865 // so round up the size to make sure the offsets are correct. 1866 if (P->p_type == PT_TLS) { 1867 Out::TlsPhdr = P; 1868 if (P->p_memsz) 1869 P->p_memsz = alignTo(P->p_memsz, P->p_align); 1870 } 1871 } 1872 } 1873 1874 // The entry point address is chosen in the following ways. 1875 // 1876 // 1. the '-e' entry command-line option; 1877 // 2. the ENTRY(symbol) command in a linker control script; 1878 // 3. the value of the symbol _start, if present; 1879 // 4. the number represented by the entry symbol, if it is a number; 1880 // 5. the address of the first byte of the .text section, if present; 1881 // 6. the address 0. 1882 template <class ELFT> uint64_t Writer<ELFT>::getEntryAddr() { 1883 // Case 1, 2 or 3 1884 if (Symbol *B = Symtab->find(Config->Entry)) 1885 return B->getVA(); 1886 1887 // Case 4 1888 uint64_t Addr; 1889 if (to_integer(Config->Entry, Addr)) 1890 return Addr; 1891 1892 // Case 5 1893 if (OutputSection *Sec = findSection(".text")) { 1894 if (Config->WarnMissingEntry) 1895 warn("cannot find entry symbol " + Config->Entry + "; defaulting to 0x" + 1896 utohexstr(Sec->Addr)); 1897 return Sec->Addr; 1898 } 1899 1900 // Case 6 1901 if (Config->WarnMissingEntry) 1902 warn("cannot find entry symbol " + Config->Entry + 1903 "; not setting start address"); 1904 return 0; 1905 } 1906 1907 static uint16_t getELFType() { 1908 if (Config->Pic) 1909 return ET_DYN; 1910 if (Config->Relocatable) 1911 return ET_REL; 1912 return ET_EXEC; 1913 } 1914 1915 template <class ELFT> void Writer<ELFT>::writeHeader() { 1916 uint8_t *Buf = Buffer->getBufferStart(); 1917 memcpy(Buf, "\177ELF", 4); 1918 1919 // Write the ELF header. 1920 auto *EHdr = reinterpret_cast<Elf_Ehdr *>(Buf); 1921 EHdr->e_ident[EI_CLASS] = Config->Is64 ? ELFCLASS64 : ELFCLASS32; 1922 EHdr->e_ident[EI_DATA] = Config->IsLE ? ELFDATA2LSB : ELFDATA2MSB; 1923 EHdr->e_ident[EI_VERSION] = EV_CURRENT; 1924 EHdr->e_ident[EI_OSABI] = Config->OSABI; 1925 EHdr->e_type = getELFType(); 1926 EHdr->e_machine = Config->EMachine; 1927 EHdr->e_version = EV_CURRENT; 1928 EHdr->e_entry = getEntryAddr(); 1929 EHdr->e_shoff = SectionHeaderOff; 1930 EHdr->e_flags = Config->EFlags; 1931 EHdr->e_ehsize = sizeof(Elf_Ehdr); 1932 EHdr->e_phnum = Phdrs.size(); 1933 EHdr->e_shentsize = sizeof(Elf_Shdr); 1934 EHdr->e_shnum = OutputSections.size() + 1; 1935 EHdr->e_shstrndx = InX::ShStrTab->getParent()->SectionIndex; 1936 1937 if (!Config->Relocatable) { 1938 EHdr->e_phoff = sizeof(Elf_Ehdr); 1939 EHdr->e_phentsize = sizeof(Elf_Phdr); 1940 } 1941 1942 // Write the program header table. 1943 auto *HBuf = reinterpret_cast<Elf_Phdr *>(Buf + EHdr->e_phoff); 1944 for (PhdrEntry *P : Phdrs) { 1945 HBuf->p_type = P->p_type; 1946 HBuf->p_flags = P->p_flags; 1947 HBuf->p_offset = P->p_offset; 1948 HBuf->p_vaddr = P->p_vaddr; 1949 HBuf->p_paddr = P->p_paddr; 1950 HBuf->p_filesz = P->p_filesz; 1951 HBuf->p_memsz = P->p_memsz; 1952 HBuf->p_align = P->p_align; 1953 ++HBuf; 1954 } 1955 1956 // Write the section header table. Note that the first table entry is null. 1957 auto *SHdrs = reinterpret_cast<Elf_Shdr *>(Buf + EHdr->e_shoff); 1958 for (OutputSection *Sec : OutputSections) 1959 Sec->writeHeaderTo<ELFT>(++SHdrs); 1960 } 1961 1962 // Open a result file. 1963 template <class ELFT> void Writer<ELFT>::openFile() { 1964 if (!Config->Is64 && FileSize > UINT32_MAX) { 1965 error("output file too large: " + Twine(FileSize) + " bytes"); 1966 return; 1967 } 1968 1969 unlinkAsync(Config->OutputFile); 1970 unsigned Flags = 0; 1971 if (!Config->Relocatable) 1972 Flags = FileOutputBuffer::F_executable; 1973 Expected<std::unique_ptr<FileOutputBuffer>> BufferOrErr = 1974 FileOutputBuffer::create(Config->OutputFile, FileSize, Flags); 1975 1976 if (!BufferOrErr) 1977 error("failed to open " + Config->OutputFile + ": " + 1978 llvm::toString(BufferOrErr.takeError())); 1979 else 1980 Buffer = std::move(*BufferOrErr); 1981 } 1982 1983 template <class ELFT> void Writer<ELFT>::writeSectionsBinary() { 1984 uint8_t *Buf = Buffer->getBufferStart(); 1985 for (OutputSection *Sec : OutputSections) 1986 if (Sec->Flags & SHF_ALLOC) 1987 Sec->writeTo<ELFT>(Buf + Sec->Offset); 1988 } 1989 1990 static void fillTrap(uint8_t *I, uint8_t *End) { 1991 for (; I + 4 <= End; I += 4) 1992 memcpy(I, &Target->TrapInstr, 4); 1993 } 1994 1995 // Fill the last page of executable segments with trap instructions 1996 // instead of leaving them as zero. Even though it is not required by any 1997 // standard, it is in general a good thing to do for security reasons. 1998 // 1999 // We'll leave other pages in segments as-is because the rest will be 2000 // overwritten by output sections. 2001 template <class ELFT> void Writer<ELFT>::writeTrapInstr() { 2002 if (Script->HasSectionsCommand) 2003 return; 2004 2005 // Fill the last page. 2006 uint8_t *Buf = Buffer->getBufferStart(); 2007 for (PhdrEntry *P : Phdrs) 2008 if (P->p_type == PT_LOAD && (P->p_flags & PF_X)) 2009 fillTrap(Buf + alignDown(P->p_offset + P->p_filesz, Target->PageSize), 2010 Buf + alignTo(P->p_offset + P->p_filesz, Target->PageSize)); 2011 2012 // Round up the file size of the last segment to the page boundary iff it is 2013 // an executable segment to ensure that other tools don't accidentally 2014 // trim the instruction padding (e.g. when stripping the file). 2015 PhdrEntry *Last = nullptr; 2016 for (PhdrEntry *P : Phdrs) 2017 if (P->p_type == PT_LOAD) 2018 Last = P; 2019 2020 if (Last && (Last->p_flags & PF_X)) 2021 Last->p_memsz = Last->p_filesz = alignTo(Last->p_filesz, Target->PageSize); 2022 } 2023 2024 // Write section contents to a mmap'ed file. 2025 template <class ELFT> void Writer<ELFT>::writeSections() { 2026 uint8_t *Buf = Buffer->getBufferStart(); 2027 2028 // PPC64 needs to process relocations in the .opd section 2029 // before processing relocations in code-containing sections. 2030 if (auto *OpdCmd = findSection(".opd")) { 2031 Out::Opd = OpdCmd; 2032 Out::OpdBuf = Buf + Out::Opd->Offset; 2033 OpdCmd->template writeTo<ELFT>(Buf + Out::Opd->Offset); 2034 } 2035 2036 OutputSection *EhFrameHdr = nullptr; 2037 if (InX::EhFrameHdr && !InX::EhFrameHdr->empty()) 2038 EhFrameHdr = InX::EhFrameHdr->getParent(); 2039 2040 // In -r or -emit-relocs mode, write the relocation sections first as in 2041 // ELf_Rel targets we might find out that we need to modify the relocated 2042 // section while doing it. 2043 for (OutputSection *Sec : OutputSections) 2044 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA) 2045 Sec->writeTo<ELFT>(Buf + Sec->Offset); 2046 2047 for (OutputSection *Sec : OutputSections) 2048 if (Sec != Out::Opd && Sec != EhFrameHdr && Sec->Type != SHT_REL && 2049 Sec->Type != SHT_RELA) 2050 Sec->writeTo<ELFT>(Buf + Sec->Offset); 2051 2052 // The .eh_frame_hdr depends on .eh_frame section contents, therefore 2053 // it should be written after .eh_frame is written. 2054 if (EhFrameHdr) 2055 EhFrameHdr->writeTo<ELFT>(Buf + EhFrameHdr->Offset); 2056 } 2057 2058 template <class ELFT> void Writer<ELFT>::writeBuildId() { 2059 if (!InX::BuildId || !InX::BuildId->getParent()) 2060 return; 2061 2062 // Compute a hash of all sections of the output file. 2063 uint8_t *Start = Buffer->getBufferStart(); 2064 uint8_t *End = Start + FileSize; 2065 InX::BuildId->writeBuildId({Start, End}); 2066 } 2067 2068 template void elf::writeResult<ELF32LE>(); 2069 template void elf::writeResult<ELF32BE>(); 2070 template void elf::writeResult<ELF64LE>(); 2071 template void elf::writeResult<ELF64BE>(); 2072