1 //===- ScriptParser.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 // This file contains a recursive-descendent parser for linker scripts. 11 // Parsed results are stored to Config and Script global objects. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "ScriptParser.h" 16 #include "Config.h" 17 #include "Driver.h" 18 #include "InputSection.h" 19 #include "LinkerScript.h" 20 #include "OutputSections.h" 21 #include "ScriptLexer.h" 22 #include "Symbols.h" 23 #include "Target.h" 24 #include "lld/Common/Memory.h" 25 #include "llvm/ADT/SmallString.h" 26 #include "llvm/ADT/StringRef.h" 27 #include "llvm/ADT/StringSet.h" 28 #include "llvm/ADT/StringSwitch.h" 29 #include "llvm/BinaryFormat/ELF.h" 30 #include "llvm/Support/Casting.h" 31 #include "llvm/Support/ErrorHandling.h" 32 #include "llvm/Support/FileSystem.h" 33 #include "llvm/Support/Path.h" 34 #include <cassert> 35 #include <limits> 36 #include <vector> 37 38 using namespace llvm; 39 using namespace llvm::ELF; 40 using namespace llvm::support::endian; 41 using namespace lld; 42 using namespace lld::elf; 43 44 static bool isUnderSysroot(StringRef Path); 45 46 namespace { 47 class ScriptParser final : ScriptLexer { 48 public: 49 ScriptParser(MemoryBufferRef MB) 50 : ScriptLexer(MB), 51 IsUnderSysroot(isUnderSysroot(MB.getBufferIdentifier())) {} 52 53 void readLinkerScript(); 54 void readVersionScript(); 55 void readDynamicList(); 56 void readDefsym(StringRef Name); 57 58 private: 59 void addFile(StringRef Path); 60 61 void readAsNeeded(); 62 void readEntry(); 63 void readExtern(); 64 void readGroup(); 65 void readInclude(); 66 void readInput(); 67 void readMemory(); 68 void readOutput(); 69 void readOutputArch(); 70 void readOutputFormat(); 71 void readPhdrs(); 72 void readRegionAlias(); 73 void readSearchDir(); 74 void readSections(); 75 void readTarget(); 76 void readVersion(); 77 void readVersionScriptCommand(); 78 79 SymbolAssignment *readSymbolAssignment(StringRef Name); 80 ByteCommand *readByteCommand(StringRef Tok); 81 std::array<uint8_t, 4> readFill(); 82 std::array<uint8_t, 4> parseFill(StringRef Tok); 83 bool readSectionDirective(OutputSection *Cmd, StringRef Tok1, StringRef Tok2); 84 void readSectionAddressType(OutputSection *Cmd); 85 OutputSection *readOverlaySectionDescription(); 86 OutputSection *readOutputSectionDescription(StringRef OutSec); 87 std::vector<BaseCommand *> readOverlay(); 88 std::vector<StringRef> readOutputSectionPhdrs(); 89 InputSectionDescription *readInputSectionDescription(StringRef Tok); 90 StringMatcher readFilePatterns(); 91 std::vector<SectionPattern> readInputSectionsList(); 92 InputSectionDescription *readInputSectionRules(StringRef FilePattern); 93 unsigned readPhdrType(); 94 SortSectionPolicy readSortKind(); 95 SymbolAssignment *readProvideHidden(bool Provide, bool Hidden); 96 SymbolAssignment *readAssignment(StringRef Tok); 97 std::tuple<ELFKind, uint16_t, bool> readBfdName(); 98 void readSort(); 99 Expr readAssert(); 100 Expr readConstant(); 101 Expr getPageSize(); 102 103 uint64_t readMemoryAssignment(StringRef, StringRef, StringRef); 104 std::pair<uint32_t, uint32_t> readMemoryAttributes(); 105 106 Expr combine(StringRef Op, Expr L, Expr R); 107 Expr readExpr(); 108 Expr readExpr1(Expr Lhs, int MinPrec); 109 StringRef readParenLiteral(); 110 Expr readPrimary(); 111 Expr readTernary(Expr Cond); 112 Expr readParenExpr(); 113 114 // For parsing version script. 115 std::vector<SymbolVersion> readVersionExtern(); 116 void readAnonymousDeclaration(); 117 void readVersionDeclaration(StringRef VerStr); 118 119 std::pair<std::vector<SymbolVersion>, std::vector<SymbolVersion>> 120 readSymbols(); 121 122 // True if a script being read is in a subdirectory specified by -sysroot. 123 bool IsUnderSysroot; 124 125 // A set to detect an INCLUDE() cycle. 126 StringSet<> Seen; 127 }; 128 } // namespace 129 130 static StringRef unquote(StringRef S) { 131 if (S.startswith("\"")) 132 return S.substr(1, S.size() - 2); 133 return S; 134 } 135 136 static bool isUnderSysroot(StringRef Path) { 137 if (Config->Sysroot == "") 138 return false; 139 for (; !Path.empty(); Path = sys::path::parent_path(Path)) 140 if (sys::fs::equivalent(Config->Sysroot, Path)) 141 return true; 142 return false; 143 } 144 145 // Some operations only support one non absolute value. Move the 146 // absolute one to the right hand side for convenience. 147 static void moveAbsRight(ExprValue &A, ExprValue &B) { 148 if (A.Sec == nullptr || (A.ForceAbsolute && !B.isAbsolute())) 149 std::swap(A, B); 150 if (!B.isAbsolute()) 151 error(A.Loc + ": at least one side of the expression must be absolute"); 152 } 153 154 static ExprValue add(ExprValue A, ExprValue B) { 155 moveAbsRight(A, B); 156 return {A.Sec, A.ForceAbsolute, A.getSectionOffset() + B.getValue(), A.Loc}; 157 } 158 159 static ExprValue sub(ExprValue A, ExprValue B) { 160 // The distance between two symbols in sections is absolute. 161 if (!A.isAbsolute() && !B.isAbsolute()) 162 return A.getValue() - B.getValue(); 163 return {A.Sec, false, A.getSectionOffset() - B.getValue(), A.Loc}; 164 } 165 166 static ExprValue bitAnd(ExprValue A, ExprValue B) { 167 moveAbsRight(A, B); 168 return {A.Sec, A.ForceAbsolute, 169 (A.getValue() & B.getValue()) - A.getSecAddr(), A.Loc}; 170 } 171 172 static ExprValue bitOr(ExprValue A, ExprValue B) { 173 moveAbsRight(A, B); 174 return {A.Sec, A.ForceAbsolute, 175 (A.getValue() | B.getValue()) - A.getSecAddr(), A.Loc}; 176 } 177 178 void ScriptParser::readDynamicList() { 179 Config->HasDynamicList = true; 180 expect("{"); 181 std::vector<SymbolVersion> Locals; 182 std::vector<SymbolVersion> Globals; 183 std::tie(Locals, Globals) = readSymbols(); 184 expect(";"); 185 186 if (!atEOF()) { 187 setError("EOF expected, but got " + next()); 188 return; 189 } 190 if (!Locals.empty()) { 191 setError("\"local:\" scope not supported in --dynamic-list"); 192 return; 193 } 194 195 for (SymbolVersion V : Globals) 196 Config->DynamicList.push_back(V); 197 } 198 199 void ScriptParser::readVersionScript() { 200 readVersionScriptCommand(); 201 if (!atEOF()) 202 setError("EOF expected, but got " + next()); 203 } 204 205 void ScriptParser::readVersionScriptCommand() { 206 if (consume("{")) { 207 readAnonymousDeclaration(); 208 return; 209 } 210 211 while (!atEOF() && !errorCount() && peek() != "}") { 212 StringRef VerStr = next(); 213 if (VerStr == "{") { 214 setError("anonymous version definition is used in " 215 "combination with other version definitions"); 216 return; 217 } 218 expect("{"); 219 readVersionDeclaration(VerStr); 220 } 221 } 222 223 void ScriptParser::readVersion() { 224 expect("{"); 225 readVersionScriptCommand(); 226 expect("}"); 227 } 228 229 void ScriptParser::readLinkerScript() { 230 while (!atEOF()) { 231 StringRef Tok = next(); 232 if (Tok == ";") 233 continue; 234 235 if (Tok == "ENTRY") { 236 readEntry(); 237 } else if (Tok == "EXTERN") { 238 readExtern(); 239 } else if (Tok == "GROUP") { 240 readGroup(); 241 } else if (Tok == "INCLUDE") { 242 readInclude(); 243 } else if (Tok == "INPUT") { 244 readInput(); 245 } else if (Tok == "MEMORY") { 246 readMemory(); 247 } else if (Tok == "OUTPUT") { 248 readOutput(); 249 } else if (Tok == "OUTPUT_ARCH") { 250 readOutputArch(); 251 } else if (Tok == "OUTPUT_FORMAT") { 252 readOutputFormat(); 253 } else if (Tok == "PHDRS") { 254 readPhdrs(); 255 } else if (Tok == "REGION_ALIAS") { 256 readRegionAlias(); 257 } else if (Tok == "SEARCH_DIR") { 258 readSearchDir(); 259 } else if (Tok == "SECTIONS") { 260 readSections(); 261 } else if (Tok == "TARGET") { 262 readTarget(); 263 } else if (Tok == "VERSION") { 264 readVersion(); 265 } else if (SymbolAssignment *Cmd = readAssignment(Tok)) { 266 Script->SectionCommands.push_back(Cmd); 267 } else { 268 setError("unknown directive: " + Tok); 269 } 270 } 271 } 272 273 void ScriptParser::readDefsym(StringRef Name) { 274 if (errorCount()) 275 return; 276 Expr E = readExpr(); 277 if (!atEOF()) 278 setError("EOF expected, but got " + next()); 279 SymbolAssignment *Cmd = make<SymbolAssignment>(Name, E, getCurrentLocation()); 280 Script->SectionCommands.push_back(Cmd); 281 } 282 283 void ScriptParser::addFile(StringRef S) { 284 if (IsUnderSysroot && S.startswith("/")) { 285 SmallString<128> PathData; 286 StringRef Path = (Config->Sysroot + S).toStringRef(PathData); 287 if (sys::fs::exists(Path)) { 288 Driver->addFile(Saver.save(Path), /*WithLOption=*/false); 289 return; 290 } 291 } 292 293 if (S.startswith("/")) { 294 Driver->addFile(S, /*WithLOption=*/false); 295 } else if (S.startswith("=")) { 296 if (Config->Sysroot.empty()) 297 Driver->addFile(S.substr(1), /*WithLOption=*/false); 298 else 299 Driver->addFile(Saver.save(Config->Sysroot + "/" + S.substr(1)), 300 /*WithLOption=*/false); 301 } else if (S.startswith("-l")) { 302 Driver->addLibrary(S.substr(2)); 303 } else if (sys::fs::exists(S)) { 304 Driver->addFile(S, /*WithLOption=*/false); 305 } else { 306 if (Optional<std::string> Path = findFromSearchPaths(S)) 307 Driver->addFile(Saver.save(*Path), /*WithLOption=*/true); 308 else 309 setError("unable to find " + S); 310 } 311 } 312 313 void ScriptParser::readAsNeeded() { 314 expect("("); 315 bool Orig = Config->AsNeeded; 316 Config->AsNeeded = true; 317 while (!errorCount() && !consume(")")) 318 addFile(unquote(next())); 319 Config->AsNeeded = Orig; 320 } 321 322 void ScriptParser::readEntry() { 323 // -e <symbol> takes predecence over ENTRY(<symbol>). 324 expect("("); 325 StringRef Tok = next(); 326 if (Config->Entry.empty()) 327 Config->Entry = Tok; 328 expect(")"); 329 } 330 331 void ScriptParser::readExtern() { 332 expect("("); 333 while (!errorCount() && !consume(")")) 334 Config->Undefined.push_back(next()); 335 } 336 337 void ScriptParser::readGroup() { 338 bool Orig = InputFile::IsInGroup; 339 InputFile::IsInGroup = true; 340 readInput(); 341 InputFile::IsInGroup = Orig; 342 if (!Orig) 343 ++InputFile::NextGroupId; 344 } 345 346 void ScriptParser::readInclude() { 347 StringRef Tok = unquote(next()); 348 349 if (!Seen.insert(Tok).second) { 350 setError("there is a cycle in linker script INCLUDEs"); 351 return; 352 } 353 354 if (Optional<std::string> Path = searchScript(Tok)) { 355 if (Optional<MemoryBufferRef> MB = readFile(*Path)) 356 tokenize(*MB); 357 return; 358 } 359 setError("cannot find linker script " + Tok); 360 } 361 362 void ScriptParser::readInput() { 363 expect("("); 364 while (!errorCount() && !consume(")")) { 365 if (consume("AS_NEEDED")) 366 readAsNeeded(); 367 else 368 addFile(unquote(next())); 369 } 370 } 371 372 void ScriptParser::readOutput() { 373 // -o <file> takes predecence over OUTPUT(<file>). 374 expect("("); 375 StringRef Tok = next(); 376 if (Config->OutputFile.empty()) 377 Config->OutputFile = unquote(Tok); 378 expect(")"); 379 } 380 381 void ScriptParser::readOutputArch() { 382 // OUTPUT_ARCH is ignored for now. 383 expect("("); 384 while (!errorCount() && !consume(")")) 385 skip(); 386 } 387 388 std::tuple<ELFKind, uint16_t, bool> ScriptParser::readBfdName() { 389 StringRef S = unquote(next()); 390 if (S == "elf32-i386") 391 return std::make_tuple(ELF32LEKind, EM_386, false); 392 if (S == "elf32-iamcu") 393 return std::make_tuple(ELF32LEKind, EM_IAMCU, false); 394 if (S == "elf32-littlearm") 395 return std::make_tuple(ELF32LEKind, EM_ARM, false); 396 if (S == "elf32-x86-64") 397 return std::make_tuple(ELF32LEKind, EM_X86_64, false); 398 if (S == "elf64-littleaarch64") 399 return std::make_tuple(ELF64LEKind, EM_AARCH64, false); 400 if (S == "elf64-powerpc") 401 return std::make_tuple(ELF64BEKind, EM_PPC64, false); 402 if (S == "elf64-powerpcle") 403 return std::make_tuple(ELF64LEKind, EM_PPC64, false); 404 if (S == "elf64-x86-64") 405 return std::make_tuple(ELF64LEKind, EM_X86_64, false); 406 if (S == "elf32-tradbigmips") 407 return std::make_tuple(ELF32BEKind, EM_MIPS, false); 408 if (S == "elf32-ntradbigmips") 409 return std::make_tuple(ELF32BEKind, EM_MIPS, true); 410 if (S == "elf32-tradlittlemips") 411 return std::make_tuple(ELF32LEKind, EM_MIPS, false); 412 if (S == "elf32-ntradlittlemips") 413 return std::make_tuple(ELF32LEKind, EM_MIPS, true); 414 if (S == "elf64-tradbigmips") 415 return std::make_tuple(ELF64BEKind, EM_MIPS, false); 416 if (S == "elf64-tradlittlemips") 417 return std::make_tuple(ELF64LEKind, EM_MIPS, false); 418 419 setError("unknown output format name: " + S); 420 return std::make_tuple(ELFNoneKind, EM_NONE, false); 421 } 422 423 // Parse OUTPUT_FORMAT(bfdname) or OUTPUT_FORMAT(bfdname, big, little). 424 // Currently we ignore big and little parameters. 425 void ScriptParser::readOutputFormat() { 426 expect("("); 427 428 std::tuple<ELFKind, uint16_t, bool> BfdTuple = readBfdName(); 429 if (Config->EKind == ELFNoneKind) 430 std::tie(Config->EKind, Config->EMachine, Config->MipsN32Abi) = BfdTuple; 431 432 if (consume(")")) 433 return; 434 expect(","); 435 skip(); 436 expect(","); 437 skip(); 438 expect(")"); 439 } 440 441 void ScriptParser::readPhdrs() { 442 expect("{"); 443 444 while (!errorCount() && !consume("}")) { 445 PhdrsCommand Cmd; 446 Cmd.Name = next(); 447 Cmd.Type = readPhdrType(); 448 449 while (!errorCount() && !consume(";")) { 450 if (consume("FILEHDR")) 451 Cmd.HasFilehdr = true; 452 else if (consume("PHDRS")) 453 Cmd.HasPhdrs = true; 454 else if (consume("AT")) 455 Cmd.LMAExpr = readParenExpr(); 456 else if (consume("FLAGS")) 457 Cmd.Flags = readParenExpr()().getValue(); 458 else 459 setError("unexpected header attribute: " + next()); 460 } 461 462 Script->PhdrsCommands.push_back(Cmd); 463 } 464 } 465 466 void ScriptParser::readRegionAlias() { 467 expect("("); 468 StringRef Alias = unquote(next()); 469 expect(","); 470 StringRef Name = next(); 471 expect(")"); 472 473 if (Script->MemoryRegions.count(Alias)) 474 setError("redefinition of memory region '" + Alias + "'"); 475 if (!Script->MemoryRegions.count(Name)) 476 setError("memory region '" + Name + "' is not defined"); 477 Script->MemoryRegions.insert({Alias, Script->MemoryRegions[Name]}); 478 } 479 480 void ScriptParser::readSearchDir() { 481 expect("("); 482 StringRef Tok = next(); 483 if (!Config->Nostdlib) 484 Config->SearchPaths.push_back(unquote(Tok)); 485 expect(")"); 486 } 487 488 // This reads an overlay description. Overlays are used to describe output 489 // sections that use the same virtual memory range and normally would trigger 490 // linker's sections sanity check failures. 491 // https://sourceware.org/binutils/docs/ld/Overlay-Description.html#Overlay-Description 492 std::vector<BaseCommand *> ScriptParser::readOverlay() { 493 // VA and LMA expressions are optional, though for simplicity of 494 // implementation we assume they are not. That is what OVERLAY was designed 495 // for first of all: to allow sections with overlapping VAs at different LMAs. 496 Expr AddrExpr = readExpr(); 497 expect(":"); 498 expect("AT"); 499 Expr LMAExpr = readParenExpr(); 500 expect("{"); 501 502 std::vector<BaseCommand *> V; 503 OutputSection *Prev = nullptr; 504 while (!errorCount() && !consume("}")) { 505 // VA is the same for all sections. The LMAs are consecutive in memory 506 // starting from the base load address specified. 507 OutputSection *OS = readOverlaySectionDescription(); 508 OS->AddrExpr = AddrExpr; 509 if (Prev) 510 OS->LMAExpr = [=] { return Prev->getLMA() + Prev->Size; }; 511 else 512 OS->LMAExpr = LMAExpr; 513 V.push_back(OS); 514 Prev = OS; 515 } 516 517 // According to the specification, at the end of the overlay, the location 518 // counter should be equal to the overlay base address plus size of the 519 // largest section seen in the overlay. 520 // Here we want to create the Dot assignment command to achieve that. 521 Expr MoveDot = [=] { 522 uint64_t Max = 0; 523 for (BaseCommand *Cmd : V) 524 Max = std::max(Max, cast<OutputSection>(Cmd)->Size); 525 return AddrExpr().getValue() + Max; 526 }; 527 V.push_back(make<SymbolAssignment>(".", MoveDot, getCurrentLocation())); 528 return V; 529 } 530 531 void ScriptParser::readSections() { 532 Script->HasSectionsCommand = true; 533 534 // -no-rosegment is used to avoid placing read only non-executable sections in 535 // their own segment. We do the same if SECTIONS command is present in linker 536 // script. See comment for computeFlags(). 537 Config->SingleRoRx = true; 538 539 expect("{"); 540 std::vector<BaseCommand *> V; 541 while (!errorCount() && !consume("}")) { 542 StringRef Tok = next(); 543 if (Tok == "OVERLAY") { 544 for (BaseCommand *Cmd : readOverlay()) 545 V.push_back(Cmd); 546 continue; 547 } else if (Tok == "INCLUDE") { 548 readInclude(); 549 continue; 550 } 551 552 if (BaseCommand *Cmd = readAssignment(Tok)) 553 V.push_back(Cmd); 554 else 555 V.push_back(readOutputSectionDescription(Tok)); 556 } 557 558 if (!atEOF() && consume("INSERT")) { 559 std::vector<BaseCommand *> *Dest = nullptr; 560 if (consume("AFTER")) 561 Dest = &Script->InsertAfterCommands[next()]; 562 else if (consume("BEFORE")) 563 Dest = &Script->InsertBeforeCommands[next()]; 564 else 565 setError("expected AFTER/BEFORE, but got '" + next() + "'"); 566 if (Dest) 567 Dest->insert(Dest->end(), V.begin(), V.end()); 568 return; 569 } 570 571 Script->SectionCommands.insert(Script->SectionCommands.end(), V.begin(), 572 V.end()); 573 } 574 575 void ScriptParser::readTarget() { 576 // TARGET(foo) is an alias for "--format foo". Unlike GNU linkers, 577 // we accept only a limited set of BFD names (i.e. "elf" or "binary") 578 // for --format. We recognize only /^elf/ and "binary" in the linker 579 // script as well. 580 expect("("); 581 StringRef Tok = next(); 582 expect(")"); 583 584 if (Tok.startswith("elf")) 585 Config->FormatBinary = false; 586 else if (Tok == "binary") 587 Config->FormatBinary = true; 588 else 589 setError("unknown target: " + Tok); 590 } 591 592 static int precedence(StringRef Op) { 593 return StringSwitch<int>(Op) 594 .Cases("*", "/", "%", 8) 595 .Cases("+", "-", 7) 596 .Cases("<<", ">>", 6) 597 .Cases("<", "<=", ">", ">=", "==", "!=", 5) 598 .Case("&", 4) 599 .Case("|", 3) 600 .Case("&&", 2) 601 .Case("||", 1) 602 .Default(-1); 603 } 604 605 StringMatcher ScriptParser::readFilePatterns() { 606 std::vector<StringRef> V; 607 while (!errorCount() && !consume(")")) 608 V.push_back(next()); 609 return StringMatcher(V); 610 } 611 612 SortSectionPolicy ScriptParser::readSortKind() { 613 if (consume("SORT") || consume("SORT_BY_NAME")) 614 return SortSectionPolicy::Name; 615 if (consume("SORT_BY_ALIGNMENT")) 616 return SortSectionPolicy::Alignment; 617 if (consume("SORT_BY_INIT_PRIORITY")) 618 return SortSectionPolicy::Priority; 619 if (consume("SORT_NONE")) 620 return SortSectionPolicy::None; 621 return SortSectionPolicy::Default; 622 } 623 624 // Reads SECTIONS command contents in the following form: 625 // 626 // <contents> ::= <elem>* 627 // <elem> ::= <exclude>? <glob-pattern> 628 // <exclude> ::= "EXCLUDE_FILE" "(" <glob-pattern>+ ")" 629 // 630 // For example, 631 // 632 // *(.foo EXCLUDE_FILE (a.o) .bar EXCLUDE_FILE (b.o) .baz) 633 // 634 // is parsed as ".foo", ".bar" with "a.o", and ".baz" with "b.o". 635 // The semantics of that is section .foo in any file, section .bar in 636 // any file but a.o, and section .baz in any file but b.o. 637 std::vector<SectionPattern> ScriptParser::readInputSectionsList() { 638 std::vector<SectionPattern> Ret; 639 while (!errorCount() && peek() != ")") { 640 StringMatcher ExcludeFilePat; 641 if (consume("EXCLUDE_FILE")) { 642 expect("("); 643 ExcludeFilePat = readFilePatterns(); 644 } 645 646 std::vector<StringRef> V; 647 while (!errorCount() && peek() != ")" && peek() != "EXCLUDE_FILE") 648 V.push_back(next()); 649 650 if (!V.empty()) 651 Ret.push_back({std::move(ExcludeFilePat), StringMatcher(V)}); 652 else 653 setError("section pattern is expected"); 654 } 655 return Ret; 656 } 657 658 // Reads contents of "SECTIONS" directive. That directive contains a 659 // list of glob patterns for input sections. The grammar is as follows. 660 // 661 // <patterns> ::= <section-list> 662 // | <sort> "(" <section-list> ")" 663 // | <sort> "(" <sort> "(" <section-list> ")" ")" 664 // 665 // <sort> ::= "SORT" | "SORT_BY_NAME" | "SORT_BY_ALIGNMENT" 666 // | "SORT_BY_INIT_PRIORITY" | "SORT_NONE" 667 // 668 // <section-list> is parsed by readInputSectionsList(). 669 InputSectionDescription * 670 ScriptParser::readInputSectionRules(StringRef FilePattern) { 671 auto *Cmd = make<InputSectionDescription>(FilePattern); 672 expect("("); 673 674 while (!errorCount() && !consume(")")) { 675 SortSectionPolicy Outer = readSortKind(); 676 SortSectionPolicy Inner = SortSectionPolicy::Default; 677 std::vector<SectionPattern> V; 678 if (Outer != SortSectionPolicy::Default) { 679 expect("("); 680 Inner = readSortKind(); 681 if (Inner != SortSectionPolicy::Default) { 682 expect("("); 683 V = readInputSectionsList(); 684 expect(")"); 685 } else { 686 V = readInputSectionsList(); 687 } 688 expect(")"); 689 } else { 690 V = readInputSectionsList(); 691 } 692 693 for (SectionPattern &Pat : V) { 694 Pat.SortInner = Inner; 695 Pat.SortOuter = Outer; 696 } 697 698 std::move(V.begin(), V.end(), std::back_inserter(Cmd->SectionPatterns)); 699 } 700 return Cmd; 701 } 702 703 InputSectionDescription * 704 ScriptParser::readInputSectionDescription(StringRef Tok) { 705 // Input section wildcard can be surrounded by KEEP. 706 // https://sourceware.org/binutils/docs/ld/Input-Section-Keep.html#Input-Section-Keep 707 if (Tok == "KEEP") { 708 expect("("); 709 StringRef FilePattern = next(); 710 InputSectionDescription *Cmd = readInputSectionRules(FilePattern); 711 expect(")"); 712 Script->KeptSections.push_back(Cmd); 713 return Cmd; 714 } 715 return readInputSectionRules(Tok); 716 } 717 718 void ScriptParser::readSort() { 719 expect("("); 720 expect("CONSTRUCTORS"); 721 expect(")"); 722 } 723 724 Expr ScriptParser::readAssert() { 725 expect("("); 726 Expr E = readExpr(); 727 expect(","); 728 StringRef Msg = unquote(next()); 729 expect(")"); 730 731 return [=] { 732 if (!E().getValue()) 733 error(Msg); 734 return Script->getDot(); 735 }; 736 } 737 738 // Reads a FILL(expr) command. We handle the FILL command as an 739 // alias for =fillexp section attribute, which is different from 740 // what GNU linkers do. 741 // https://sourceware.org/binutils/docs/ld/Output-Section-Data.html 742 std::array<uint8_t, 4> ScriptParser::readFill() { 743 expect("("); 744 std::array<uint8_t, 4> V = parseFill(next()); 745 expect(")"); 746 return V; 747 } 748 749 // Tries to read the special directive for an output section definition which 750 // can be one of following: "(NOLOAD)", "(COPY)", "(INFO)" or "(OVERLAY)". 751 // Tok1 and Tok2 are next 2 tokens peeked. See comment for readSectionAddressType below. 752 bool ScriptParser::readSectionDirective(OutputSection *Cmd, StringRef Tok1, StringRef Tok2) { 753 if (Tok1 != "(") 754 return false; 755 if (Tok2 != "NOLOAD" && Tok2 != "COPY" && Tok2 != "INFO" && Tok2 != "OVERLAY") 756 return false; 757 758 expect("("); 759 if (consume("NOLOAD")) { 760 Cmd->Noload = true; 761 } else { 762 skip(); // This is "COPY", "INFO" or "OVERLAY". 763 Cmd->NonAlloc = true; 764 } 765 expect(")"); 766 return true; 767 } 768 769 // Reads an expression and/or the special directive for an output 770 // section definition. Directive is one of following: "(NOLOAD)", 771 // "(COPY)", "(INFO)" or "(OVERLAY)". 772 // 773 // An output section name can be followed by an address expression 774 // and/or directive. This grammar is not LL(1) because "(" can be 775 // interpreted as either the beginning of some expression or beginning 776 // of directive. 777 // 778 // https://sourceware.org/binutils/docs/ld/Output-Section-Address.html 779 // https://sourceware.org/binutils/docs/ld/Output-Section-Type.html 780 void ScriptParser::readSectionAddressType(OutputSection *Cmd) { 781 if (readSectionDirective(Cmd, peek(), peek2())) 782 return; 783 784 Cmd->AddrExpr = readExpr(); 785 if (peek() == "(" && !readSectionDirective(Cmd, "(", peek2())) 786 setError("unknown section directive: " + peek2()); 787 } 788 789 static Expr checkAlignment(Expr E, std::string &Loc) { 790 return [=] { 791 uint64_t Alignment = std::max((uint64_t)1, E().getValue()); 792 if (!isPowerOf2_64(Alignment)) { 793 error(Loc + ": alignment must be power of 2"); 794 return (uint64_t)1; // Return a dummy value. 795 } 796 return Alignment; 797 }; 798 } 799 800 OutputSection *ScriptParser::readOverlaySectionDescription() { 801 OutputSection *Cmd = 802 Script->createOutputSection(next(), getCurrentLocation()); 803 Cmd->InOverlay = true; 804 expect("{"); 805 while (!errorCount() && !consume("}")) 806 Cmd->SectionCommands.push_back(readInputSectionRules(next())); 807 Cmd->Phdrs = readOutputSectionPhdrs(); 808 return Cmd; 809 } 810 811 OutputSection *ScriptParser::readOutputSectionDescription(StringRef OutSec) { 812 OutputSection *Cmd = 813 Script->createOutputSection(OutSec, getCurrentLocation()); 814 815 size_t SymbolsReferenced = Script->ReferencedSymbols.size(); 816 817 if (peek() != ":") 818 readSectionAddressType(Cmd); 819 expect(":"); 820 821 std::string Location = getCurrentLocation(); 822 if (consume("AT")) 823 Cmd->LMAExpr = readParenExpr(); 824 if (consume("ALIGN")) 825 Cmd->AlignExpr = checkAlignment(readParenExpr(), Location); 826 if (consume("SUBALIGN")) 827 Cmd->SubalignExpr = checkAlignment(readParenExpr(), Location); 828 829 // Parse constraints. 830 if (consume("ONLY_IF_RO")) 831 Cmd->Constraint = ConstraintKind::ReadOnly; 832 if (consume("ONLY_IF_RW")) 833 Cmd->Constraint = ConstraintKind::ReadWrite; 834 expect("{"); 835 836 while (!errorCount() && !consume("}")) { 837 StringRef Tok = next(); 838 if (Tok == ";") { 839 // Empty commands are allowed. Do nothing here. 840 } else if (SymbolAssignment *Assign = readAssignment(Tok)) { 841 Cmd->SectionCommands.push_back(Assign); 842 } else if (ByteCommand *Data = readByteCommand(Tok)) { 843 Cmd->SectionCommands.push_back(Data); 844 } else if (Tok == "CONSTRUCTORS") { 845 // CONSTRUCTORS is a keyword to make the linker recognize C++ ctors/dtors 846 // by name. This is for very old file formats such as ECOFF/XCOFF. 847 // For ELF, we should ignore. 848 } else if (Tok == "FILL") { 849 Cmd->Filler = readFill(); 850 } else if (Tok == "SORT") { 851 readSort(); 852 } else if (Tok == "INCLUDE") { 853 readInclude(); 854 } else if (peek() == "(") { 855 Cmd->SectionCommands.push_back(readInputSectionDescription(Tok)); 856 } else { 857 // We have a file name and no input sections description. It is not a 858 // commonly used syntax, but still acceptable. In that case, all sections 859 // from the file will be included. 860 auto *ISD = make<InputSectionDescription>(Tok); 861 ISD->SectionPatterns.push_back({{}, StringMatcher({"*"})}); 862 Cmd->SectionCommands.push_back(ISD); 863 } 864 } 865 866 if (consume(">")) 867 Cmd->MemoryRegionName = next(); 868 869 if (consume("AT")) { 870 expect(">"); 871 Cmd->LMARegionName = next(); 872 } 873 874 if (Cmd->LMAExpr && !Cmd->LMARegionName.empty()) 875 error("section can't have both LMA and a load region"); 876 877 Cmd->Phdrs = readOutputSectionPhdrs(); 878 879 if (consume("=")) 880 Cmd->Filler = parseFill(next()); 881 else if (peek().startswith("=")) 882 Cmd->Filler = parseFill(next().drop_front()); 883 884 // Consume optional comma following output section command. 885 consume(","); 886 887 if (Script->ReferencedSymbols.size() > SymbolsReferenced) 888 Cmd->ExpressionsUseSymbols = true; 889 return Cmd; 890 } 891 892 // Parses a given string as a octal/decimal/hexadecimal number and 893 // returns it as a big-endian number. Used for `=<fillexp>`. 894 // https://sourceware.org/binutils/docs/ld/Output-Section-Fill.html 895 // 896 // When reading a hexstring, ld.bfd handles it as a blob of arbitrary 897 // size, while ld.gold always handles it as a 32-bit big-endian number. 898 // We are compatible with ld.gold because it's easier to implement. 899 std::array<uint8_t, 4> ScriptParser::parseFill(StringRef Tok) { 900 uint32_t V = 0; 901 if (!to_integer(Tok, V)) 902 setError("invalid filler expression: " + Tok); 903 904 std::array<uint8_t, 4> Buf; 905 write32be(Buf.data(), V); 906 return Buf; 907 } 908 909 SymbolAssignment *ScriptParser::readProvideHidden(bool Provide, bool Hidden) { 910 expect("("); 911 SymbolAssignment *Cmd = readSymbolAssignment(next()); 912 Cmd->Provide = Provide; 913 Cmd->Hidden = Hidden; 914 expect(")"); 915 return Cmd; 916 } 917 918 SymbolAssignment *ScriptParser::readAssignment(StringRef Tok) { 919 // Assert expression returns Dot, so this is equal to ".=." 920 if (Tok == "ASSERT") 921 return make<SymbolAssignment>(".", readAssert(), getCurrentLocation()); 922 923 size_t OldPos = Pos; 924 SymbolAssignment *Cmd = nullptr; 925 if (peek() == "=" || peek() == "+=") 926 Cmd = readSymbolAssignment(Tok); 927 else if (Tok == "PROVIDE") 928 Cmd = readProvideHidden(true, false); 929 else if (Tok == "HIDDEN") 930 Cmd = readProvideHidden(false, true); 931 else if (Tok == "PROVIDE_HIDDEN") 932 Cmd = readProvideHidden(true, true); 933 934 if (Cmd) { 935 Cmd->CommandString = 936 Tok.str() + " " + 937 llvm::join(Tokens.begin() + OldPos, Tokens.begin() + Pos, " "); 938 expect(";"); 939 } 940 return Cmd; 941 } 942 943 SymbolAssignment *ScriptParser::readSymbolAssignment(StringRef Name) { 944 StringRef Op = next(); 945 assert(Op == "=" || Op == "+="); 946 Expr E = readExpr(); 947 if (Op == "+=") { 948 std::string Loc = getCurrentLocation(); 949 E = [=] { return add(Script->getSymbolValue(Name, Loc), E()); }; 950 } 951 return make<SymbolAssignment>(Name, E, getCurrentLocation()); 952 } 953 954 // This is an operator-precedence parser to parse a linker 955 // script expression. 956 Expr ScriptParser::readExpr() { 957 // Our lexer is context-aware. Set the in-expression bit so that 958 // they apply different tokenization rules. 959 bool Orig = InExpr; 960 InExpr = true; 961 Expr E = readExpr1(readPrimary(), 0); 962 InExpr = Orig; 963 return E; 964 } 965 966 Expr ScriptParser::combine(StringRef Op, Expr L, Expr R) { 967 if (Op == "+") 968 return [=] { return add(L(), R()); }; 969 if (Op == "-") 970 return [=] { return sub(L(), R()); }; 971 if (Op == "*") 972 return [=] { return L().getValue() * R().getValue(); }; 973 if (Op == "/") { 974 std::string Loc = getCurrentLocation(); 975 return [=]() -> uint64_t { 976 if (uint64_t RV = R().getValue()) 977 return L().getValue() / RV; 978 error(Loc + ": division by zero"); 979 return 0; 980 }; 981 } 982 if (Op == "%") { 983 std::string Loc = getCurrentLocation(); 984 return [=]() -> uint64_t { 985 if (uint64_t RV = R().getValue()) 986 return L().getValue() % RV; 987 error(Loc + ": modulo by zero"); 988 return 0; 989 }; 990 } 991 if (Op == "<<") 992 return [=] { return L().getValue() << R().getValue(); }; 993 if (Op == ">>") 994 return [=] { return L().getValue() >> R().getValue(); }; 995 if (Op == "<") 996 return [=] { return L().getValue() < R().getValue(); }; 997 if (Op == ">") 998 return [=] { return L().getValue() > R().getValue(); }; 999 if (Op == ">=") 1000 return [=] { return L().getValue() >= R().getValue(); }; 1001 if (Op == "<=") 1002 return [=] { return L().getValue() <= R().getValue(); }; 1003 if (Op == "==") 1004 return [=] { return L().getValue() == R().getValue(); }; 1005 if (Op == "!=") 1006 return [=] { return L().getValue() != R().getValue(); }; 1007 if (Op == "||") 1008 return [=] { return L().getValue() || R().getValue(); }; 1009 if (Op == "&&") 1010 return [=] { return L().getValue() && R().getValue(); }; 1011 if (Op == "&") 1012 return [=] { return bitAnd(L(), R()); }; 1013 if (Op == "|") 1014 return [=] { return bitOr(L(), R()); }; 1015 llvm_unreachable("invalid operator"); 1016 } 1017 1018 // This is a part of the operator-precedence parser. This function 1019 // assumes that the remaining token stream starts with an operator. 1020 Expr ScriptParser::readExpr1(Expr Lhs, int MinPrec) { 1021 while (!atEOF() && !errorCount()) { 1022 // Read an operator and an expression. 1023 if (consume("?")) 1024 return readTernary(Lhs); 1025 StringRef Op1 = peek(); 1026 if (precedence(Op1) < MinPrec) 1027 break; 1028 skip(); 1029 Expr Rhs = readPrimary(); 1030 1031 // Evaluate the remaining part of the expression first if the 1032 // next operator has greater precedence than the previous one. 1033 // For example, if we have read "+" and "3", and if the next 1034 // operator is "*", then we'll evaluate 3 * ... part first. 1035 while (!atEOF()) { 1036 StringRef Op2 = peek(); 1037 if (precedence(Op2) <= precedence(Op1)) 1038 break; 1039 Rhs = readExpr1(Rhs, precedence(Op2)); 1040 } 1041 1042 Lhs = combine(Op1, Lhs, Rhs); 1043 } 1044 return Lhs; 1045 } 1046 1047 Expr ScriptParser::getPageSize() { 1048 std::string Location = getCurrentLocation(); 1049 return [=]() -> uint64_t { 1050 if (Target) 1051 return Target->PageSize; 1052 error(Location + ": unable to calculate page size"); 1053 return 4096; // Return a dummy value. 1054 }; 1055 } 1056 1057 Expr ScriptParser::readConstant() { 1058 StringRef S = readParenLiteral(); 1059 if (S == "COMMONPAGESIZE") 1060 return getPageSize(); 1061 if (S == "MAXPAGESIZE") 1062 return [] { return Config->MaxPageSize; }; 1063 setError("unknown constant: " + S); 1064 return [] { return 0; }; 1065 } 1066 1067 // Parses Tok as an integer. It recognizes hexadecimal (prefixed with 1068 // "0x" or suffixed with "H") and decimal numbers. Decimal numbers may 1069 // have "K" (Ki) or "M" (Mi) suffixes. 1070 static Optional<uint64_t> parseInt(StringRef Tok) { 1071 // Hexadecimal 1072 uint64_t Val; 1073 if (Tok.startswith_lower("0x")) { 1074 if (!to_integer(Tok.substr(2), Val, 16)) 1075 return None; 1076 return Val; 1077 } 1078 if (Tok.endswith_lower("H")) { 1079 if (!to_integer(Tok.drop_back(), Val, 16)) 1080 return None; 1081 return Val; 1082 } 1083 1084 // Decimal 1085 if (Tok.endswith_lower("K")) { 1086 if (!to_integer(Tok.drop_back(), Val, 10)) 1087 return None; 1088 return Val * 1024; 1089 } 1090 if (Tok.endswith_lower("M")) { 1091 if (!to_integer(Tok.drop_back(), Val, 10)) 1092 return None; 1093 return Val * 1024 * 1024; 1094 } 1095 if (!to_integer(Tok, Val, 10)) 1096 return None; 1097 return Val; 1098 } 1099 1100 ByteCommand *ScriptParser::readByteCommand(StringRef Tok) { 1101 int Size = StringSwitch<int>(Tok) 1102 .Case("BYTE", 1) 1103 .Case("SHORT", 2) 1104 .Case("LONG", 4) 1105 .Case("QUAD", 8) 1106 .Default(-1); 1107 if (Size == -1) 1108 return nullptr; 1109 1110 size_t OldPos = Pos; 1111 Expr E = readParenExpr(); 1112 std::string CommandString = 1113 Tok.str() + " " + 1114 llvm::join(Tokens.begin() + OldPos, Tokens.begin() + Pos, " "); 1115 return make<ByteCommand>(E, Size, CommandString); 1116 } 1117 1118 StringRef ScriptParser::readParenLiteral() { 1119 expect("("); 1120 bool Orig = InExpr; 1121 InExpr = false; 1122 StringRef Tok = next(); 1123 InExpr = Orig; 1124 expect(")"); 1125 return Tok; 1126 } 1127 1128 static void checkIfExists(OutputSection *Cmd, StringRef Location) { 1129 if (Cmd->Location.empty() && Script->ErrorOnMissingSection) 1130 error(Location + ": undefined section " + Cmd->Name); 1131 } 1132 1133 Expr ScriptParser::readPrimary() { 1134 if (peek() == "(") 1135 return readParenExpr(); 1136 1137 if (consume("~")) { 1138 Expr E = readPrimary(); 1139 return [=] { return ~E().getValue(); }; 1140 } 1141 if (consume("!")) { 1142 Expr E = readPrimary(); 1143 return [=] { return !E().getValue(); }; 1144 } 1145 if (consume("-")) { 1146 Expr E = readPrimary(); 1147 return [=] { return -E().getValue(); }; 1148 } 1149 1150 StringRef Tok = next(); 1151 std::string Location = getCurrentLocation(); 1152 1153 // Built-in functions are parsed here. 1154 // https://sourceware.org/binutils/docs/ld/Builtin-Functions.html. 1155 if (Tok == "ABSOLUTE") { 1156 Expr Inner = readParenExpr(); 1157 return [=] { 1158 ExprValue I = Inner(); 1159 I.ForceAbsolute = true; 1160 return I; 1161 }; 1162 } 1163 if (Tok == "ADDR") { 1164 StringRef Name = readParenLiteral(); 1165 OutputSection *Sec = Script->getOrCreateOutputSection(Name); 1166 return [=]() -> ExprValue { 1167 checkIfExists(Sec, Location); 1168 return {Sec, false, 0, Location}; 1169 }; 1170 } 1171 if (Tok == "ALIGN") { 1172 expect("("); 1173 Expr E = readExpr(); 1174 if (consume(")")) { 1175 E = checkAlignment(E, Location); 1176 return [=] { return alignTo(Script->getDot(), E().getValue()); }; 1177 } 1178 expect(","); 1179 Expr E2 = checkAlignment(readExpr(), Location); 1180 expect(")"); 1181 return [=] { 1182 ExprValue V = E(); 1183 V.Alignment = E2().getValue(); 1184 return V; 1185 }; 1186 } 1187 if (Tok == "ALIGNOF") { 1188 StringRef Name = readParenLiteral(); 1189 OutputSection *Cmd = Script->getOrCreateOutputSection(Name); 1190 return [=] { 1191 checkIfExists(Cmd, Location); 1192 return Cmd->Alignment; 1193 }; 1194 } 1195 if (Tok == "ASSERT") 1196 return readAssert(); 1197 if (Tok == "CONSTANT") 1198 return readConstant(); 1199 if (Tok == "DATA_SEGMENT_ALIGN") { 1200 expect("("); 1201 Expr E = readExpr(); 1202 expect(","); 1203 readExpr(); 1204 expect(")"); 1205 return [=] { 1206 return alignTo(Script->getDot(), std::max((uint64_t)1, E().getValue())); 1207 }; 1208 } 1209 if (Tok == "DATA_SEGMENT_END") { 1210 expect("("); 1211 expect("."); 1212 expect(")"); 1213 return [] { return Script->getDot(); }; 1214 } 1215 if (Tok == "DATA_SEGMENT_RELRO_END") { 1216 // GNU linkers implements more complicated logic to handle 1217 // DATA_SEGMENT_RELRO_END. We instead ignore the arguments and 1218 // just align to the next page boundary for simplicity. 1219 expect("("); 1220 readExpr(); 1221 expect(","); 1222 readExpr(); 1223 expect(")"); 1224 Expr E = getPageSize(); 1225 return [=] { return alignTo(Script->getDot(), E().getValue()); }; 1226 } 1227 if (Tok == "DEFINED") { 1228 StringRef Name = readParenLiteral(); 1229 return [=] { return Symtab->find(Name) ? 1 : 0; }; 1230 } 1231 if (Tok == "LENGTH") { 1232 StringRef Name = readParenLiteral(); 1233 if (Script->MemoryRegions.count(Name) == 0) { 1234 setError("memory region not defined: " + Name); 1235 return [] { return 0; }; 1236 } 1237 return [=] { return Script->MemoryRegions[Name]->Length; }; 1238 } 1239 if (Tok == "LOADADDR") { 1240 StringRef Name = readParenLiteral(); 1241 OutputSection *Cmd = Script->getOrCreateOutputSection(Name); 1242 return [=] { 1243 checkIfExists(Cmd, Location); 1244 return Cmd->getLMA(); 1245 }; 1246 } 1247 if (Tok == "MAX" || Tok == "MIN") { 1248 expect("("); 1249 Expr A = readExpr(); 1250 expect(","); 1251 Expr B = readExpr(); 1252 expect(")"); 1253 if (Tok == "MIN") 1254 return [=] { return std::min(A().getValue(), B().getValue()); }; 1255 return [=] { return std::max(A().getValue(), B().getValue()); }; 1256 } 1257 if (Tok == "ORIGIN") { 1258 StringRef Name = readParenLiteral(); 1259 if (Script->MemoryRegions.count(Name) == 0) { 1260 setError("memory region not defined: " + Name); 1261 return [] { return 0; }; 1262 } 1263 return [=] { return Script->MemoryRegions[Name]->Origin; }; 1264 } 1265 if (Tok == "SEGMENT_START") { 1266 expect("("); 1267 skip(); 1268 expect(","); 1269 Expr E = readExpr(); 1270 expect(")"); 1271 return [=] { return E(); }; 1272 } 1273 if (Tok == "SIZEOF") { 1274 StringRef Name = readParenLiteral(); 1275 OutputSection *Cmd = Script->getOrCreateOutputSection(Name); 1276 // Linker script does not create an output section if its content is empty. 1277 // We want to allow SIZEOF(.foo) where .foo is a section which happened to 1278 // be empty. 1279 return [=] { return Cmd->Size; }; 1280 } 1281 if (Tok == "SIZEOF_HEADERS") 1282 return [=] { return elf::getHeaderSize(); }; 1283 1284 // Tok is the dot. 1285 if (Tok == ".") 1286 return [=] { return Script->getSymbolValue(Tok, Location); }; 1287 1288 // Tok is a literal number. 1289 if (Optional<uint64_t> Val = parseInt(Tok)) 1290 return [=] { return *Val; }; 1291 1292 // Tok is a symbol name. 1293 if (!isValidCIdentifier(Tok)) 1294 setError("malformed number: " + Tok); 1295 Script->ReferencedSymbols.push_back(Tok); 1296 return [=] { return Script->getSymbolValue(Tok, Location); }; 1297 } 1298 1299 Expr ScriptParser::readTernary(Expr Cond) { 1300 Expr L = readExpr(); 1301 expect(":"); 1302 Expr R = readExpr(); 1303 return [=] { return Cond().getValue() ? L() : R(); }; 1304 } 1305 1306 Expr ScriptParser::readParenExpr() { 1307 expect("("); 1308 Expr E = readExpr(); 1309 expect(")"); 1310 return E; 1311 } 1312 1313 std::vector<StringRef> ScriptParser::readOutputSectionPhdrs() { 1314 std::vector<StringRef> Phdrs; 1315 while (!errorCount() && peek().startswith(":")) { 1316 StringRef Tok = next(); 1317 Phdrs.push_back((Tok.size() == 1) ? next() : Tok.substr(1)); 1318 } 1319 return Phdrs; 1320 } 1321 1322 // Read a program header type name. The next token must be a 1323 // name of a program header type or a constant (e.g. "0x3"). 1324 unsigned ScriptParser::readPhdrType() { 1325 StringRef Tok = next(); 1326 if (Optional<uint64_t> Val = parseInt(Tok)) 1327 return *Val; 1328 1329 unsigned Ret = StringSwitch<unsigned>(Tok) 1330 .Case("PT_NULL", PT_NULL) 1331 .Case("PT_LOAD", PT_LOAD) 1332 .Case("PT_DYNAMIC", PT_DYNAMIC) 1333 .Case("PT_INTERP", PT_INTERP) 1334 .Case("PT_NOTE", PT_NOTE) 1335 .Case("PT_SHLIB", PT_SHLIB) 1336 .Case("PT_PHDR", PT_PHDR) 1337 .Case("PT_TLS", PT_TLS) 1338 .Case("PT_GNU_EH_FRAME", PT_GNU_EH_FRAME) 1339 .Case("PT_GNU_STACK", PT_GNU_STACK) 1340 .Case("PT_GNU_RELRO", PT_GNU_RELRO) 1341 .Case("PT_OPENBSD_RANDOMIZE", PT_OPENBSD_RANDOMIZE) 1342 .Case("PT_OPENBSD_WXNEEDED", PT_OPENBSD_WXNEEDED) 1343 .Case("PT_OPENBSD_BOOTDATA", PT_OPENBSD_BOOTDATA) 1344 .Default(-1); 1345 1346 if (Ret == (unsigned)-1) { 1347 setError("invalid program header type: " + Tok); 1348 return PT_NULL; 1349 } 1350 return Ret; 1351 } 1352 1353 // Reads an anonymous version declaration. 1354 void ScriptParser::readAnonymousDeclaration() { 1355 std::vector<SymbolVersion> Locals; 1356 std::vector<SymbolVersion> Globals; 1357 std::tie(Locals, Globals) = readSymbols(); 1358 1359 for (SymbolVersion V : Locals) { 1360 if (V.Name == "*") 1361 Config->DefaultSymbolVersion = VER_NDX_LOCAL; 1362 else 1363 Config->VersionScriptLocals.push_back(V); 1364 } 1365 1366 for (SymbolVersion V : Globals) 1367 Config->VersionScriptGlobals.push_back(V); 1368 1369 expect(";"); 1370 } 1371 1372 // Reads a non-anonymous version definition, 1373 // e.g. "VerStr { global: foo; bar; local: *; };". 1374 void ScriptParser::readVersionDeclaration(StringRef VerStr) { 1375 // Read a symbol list. 1376 std::vector<SymbolVersion> Locals; 1377 std::vector<SymbolVersion> Globals; 1378 std::tie(Locals, Globals) = readSymbols(); 1379 1380 for (SymbolVersion V : Locals) { 1381 if (V.Name == "*") 1382 Config->DefaultSymbolVersion = VER_NDX_LOCAL; 1383 else 1384 Config->VersionScriptLocals.push_back(V); 1385 } 1386 1387 // Create a new version definition and add that to the global symbols. 1388 VersionDefinition Ver; 1389 Ver.Name = VerStr; 1390 Ver.Globals = Globals; 1391 1392 // User-defined version number starts from 2 because 0 and 1 are 1393 // reserved for VER_NDX_LOCAL and VER_NDX_GLOBAL, respectively. 1394 Ver.Id = Config->VersionDefinitions.size() + 2; 1395 Config->VersionDefinitions.push_back(Ver); 1396 1397 // Each version may have a parent version. For example, "Ver2" 1398 // defined as "Ver2 { global: foo; local: *; } Ver1;" has "Ver1" 1399 // as a parent. This version hierarchy is, probably against your 1400 // instinct, purely for hint; the runtime doesn't care about it 1401 // at all. In LLD, we simply ignore it. 1402 if (peek() != ";") 1403 skip(); 1404 expect(";"); 1405 } 1406 1407 static bool hasWildcard(StringRef S) { 1408 return S.find_first_of("?*[") != StringRef::npos; 1409 } 1410 1411 // Reads a list of symbols, e.g. "{ global: foo; bar; local: *; };". 1412 std::pair<std::vector<SymbolVersion>, std::vector<SymbolVersion>> 1413 ScriptParser::readSymbols() { 1414 std::vector<SymbolVersion> Locals; 1415 std::vector<SymbolVersion> Globals; 1416 std::vector<SymbolVersion> *V = &Globals; 1417 1418 while (!errorCount()) { 1419 if (consume("}")) 1420 break; 1421 if (consumeLabel("local")) { 1422 V = &Locals; 1423 continue; 1424 } 1425 if (consumeLabel("global")) { 1426 V = &Globals; 1427 continue; 1428 } 1429 1430 if (consume("extern")) { 1431 std::vector<SymbolVersion> Ext = readVersionExtern(); 1432 V->insert(V->end(), Ext.begin(), Ext.end()); 1433 } else { 1434 StringRef Tok = next(); 1435 V->push_back({unquote(Tok), false, hasWildcard(Tok)}); 1436 } 1437 expect(";"); 1438 } 1439 return {Locals, Globals}; 1440 } 1441 1442 // Reads an "extern C++" directive, e.g., 1443 // "extern "C++" { ns::*; "f(int, double)"; };" 1444 // 1445 // The last semicolon is optional. E.g. this is OK: 1446 // "extern "C++" { ns::*; "f(int, double)" };" 1447 std::vector<SymbolVersion> ScriptParser::readVersionExtern() { 1448 StringRef Tok = next(); 1449 bool IsCXX = Tok == "\"C++\""; 1450 if (!IsCXX && Tok != "\"C\"") 1451 setError("Unknown language"); 1452 expect("{"); 1453 1454 std::vector<SymbolVersion> Ret; 1455 while (!errorCount() && peek() != "}") { 1456 StringRef Tok = next(); 1457 bool HasWildcard = !Tok.startswith("\"") && hasWildcard(Tok); 1458 Ret.push_back({unquote(Tok), IsCXX, HasWildcard}); 1459 if (consume("}")) 1460 return Ret; 1461 expect(";"); 1462 } 1463 1464 expect("}"); 1465 return Ret; 1466 } 1467 1468 uint64_t ScriptParser::readMemoryAssignment(StringRef S1, StringRef S2, 1469 StringRef S3) { 1470 if (!consume(S1) && !consume(S2) && !consume(S3)) { 1471 setError("expected one of: " + S1 + ", " + S2 + ", or " + S3); 1472 return 0; 1473 } 1474 expect("="); 1475 return readExpr()().getValue(); 1476 } 1477 1478 // Parse the MEMORY command as specified in: 1479 // https://sourceware.org/binutils/docs/ld/MEMORY.html 1480 // 1481 // MEMORY { name [(attr)] : ORIGIN = origin, LENGTH = len ... } 1482 void ScriptParser::readMemory() { 1483 expect("{"); 1484 while (!errorCount() && !consume("}")) { 1485 StringRef Tok = next(); 1486 if (Tok == "INCLUDE") { 1487 readInclude(); 1488 continue; 1489 } 1490 1491 uint32_t Flags = 0; 1492 uint32_t NegFlags = 0; 1493 if (consume("(")) { 1494 std::tie(Flags, NegFlags) = readMemoryAttributes(); 1495 expect(")"); 1496 } 1497 expect(":"); 1498 1499 uint64_t Origin = readMemoryAssignment("ORIGIN", "org", "o"); 1500 expect(","); 1501 uint64_t Length = readMemoryAssignment("LENGTH", "len", "l"); 1502 1503 // Add the memory region to the region map. 1504 MemoryRegion *MR = make<MemoryRegion>(Tok, Origin, Length, Flags, NegFlags); 1505 if (!Script->MemoryRegions.insert({Tok, MR}).second) 1506 setError("region '" + Tok + "' already defined"); 1507 } 1508 } 1509 1510 // This function parses the attributes used to match against section 1511 // flags when placing output sections in a memory region. These flags 1512 // are only used when an explicit memory region name is not used. 1513 std::pair<uint32_t, uint32_t> ScriptParser::readMemoryAttributes() { 1514 uint32_t Flags = 0; 1515 uint32_t NegFlags = 0; 1516 bool Invert = false; 1517 1518 for (char C : next().lower()) { 1519 uint32_t Flag = 0; 1520 if (C == '!') 1521 Invert = !Invert; 1522 else if (C == 'w') 1523 Flag = SHF_WRITE; 1524 else if (C == 'x') 1525 Flag = SHF_EXECINSTR; 1526 else if (C == 'a') 1527 Flag = SHF_ALLOC; 1528 else if (C != 'r') 1529 setError("invalid memory region attribute"); 1530 1531 if (Invert) 1532 NegFlags |= Flag; 1533 else 1534 Flags |= Flag; 1535 } 1536 return {Flags, NegFlags}; 1537 } 1538 1539 void elf::readLinkerScript(MemoryBufferRef MB) { 1540 ScriptParser(MB).readLinkerScript(); 1541 } 1542 1543 void elf::readVersionScript(MemoryBufferRef MB) { 1544 ScriptParser(MB).readVersionScript(); 1545 } 1546 1547 void elf::readDynamicList(MemoryBufferRef MB) { 1548 ScriptParser(MB).readDynamicList(); 1549 } 1550 1551 void elf::readDefsym(StringRef Name, MemoryBufferRef MB) { 1552 ScriptParser(MB).readDefsym(Name); 1553 } 1554