1 //===- InputSection.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 "InputSection.h" 11 #include "Config.h" 12 #include "EhFrame.h" 13 #include "Error.h" 14 #include "InputFiles.h" 15 #include "LinkerScript.h" 16 #include "Memory.h" 17 #include "OutputSections.h" 18 #include "Relocations.h" 19 #include "SyntheticSections.h" 20 #include "Target.h" 21 #include "Thunks.h" 22 #include "llvm/Object/Decompressor.h" 23 #include "llvm/Support/Compression.h" 24 #include "llvm/Support/Endian.h" 25 #include "llvm/Support/Path.h" 26 #include <mutex> 27 28 using namespace llvm; 29 using namespace llvm::ELF; 30 using namespace llvm::object; 31 using namespace llvm::support; 32 using namespace llvm::support::endian; 33 using namespace llvm::sys; 34 35 using namespace lld; 36 using namespace lld::elf; 37 38 std::vector<InputSectionBase *> elf::InputSections; 39 40 // Returns a string to construct an error message. 41 std::string lld::toString(const InputSectionBase *Sec) { 42 return (toString(Sec->File) + ":(" + Sec->Name + ")").str(); 43 } 44 45 template <class ELFT> 46 static ArrayRef<uint8_t> getSectionContents(elf::ObjectFile<ELFT> *File, 47 const typename ELFT::Shdr *Hdr) { 48 if (!File || Hdr->sh_type == SHT_NOBITS) 49 return makeArrayRef<uint8_t>(nullptr, Hdr->sh_size); 50 return check(File->getObj().getSectionContents(Hdr)); 51 } 52 53 InputSectionBase::InputSectionBase(InputFile *File, uint64_t Flags, 54 uint32_t Type, uint64_t Entsize, 55 uint32_t Link, uint32_t Info, 56 uint32_t Alignment, ArrayRef<uint8_t> Data, 57 StringRef Name, Kind SectionKind) 58 : SectionBase(SectionKind, Name, Flags, Entsize, Alignment, Type, Info, 59 Link), 60 File(File), Data(Data), Repl(this) { 61 Live = !Config->GcSections || !(Flags & SHF_ALLOC); 62 Assigned = false; 63 NumRelocations = 0; 64 AreRelocsRela = false; 65 66 // The ELF spec states that a value of 0 means the section has 67 // no alignment constraits. 68 uint32_t V = std::max<uint64_t>(Alignment, 1); 69 if (!isPowerOf2_64(V)) 70 fatal(toString(File) + ": section sh_addralign is not a power of 2"); 71 this->Alignment = V; 72 } 73 74 // GNU assembler 2.24 and LLVM 4.0.0's MC (the newest release as of 75 // March 2017) fail to infer section types for sections starting with 76 // ".init_array." or ".fini_array.". They set SHT_PROGBITS instead of 77 // SHF_INIT_ARRAY. As a result, the following assembler directive 78 // creates ".init_array.100" with SHT_PROGBITS, for example. 79 // 80 // .section .init_array.100, "aw" 81 // 82 // This function forces SHT_{INIT,FINI}_ARRAY so that we can handle 83 // incorrect inputs as if they were correct from the beginning. 84 static uint64_t getType(uint64_t Type, StringRef Name) { 85 if (Type == SHT_PROGBITS && Name.startswith(".init_array.")) 86 return SHT_INIT_ARRAY; 87 if (Type == SHT_PROGBITS && Name.startswith(".fini_array.")) 88 return SHT_FINI_ARRAY; 89 return Type; 90 } 91 92 template <class ELFT> 93 InputSectionBase::InputSectionBase(elf::ObjectFile<ELFT> *File, 94 const typename ELFT::Shdr *Hdr, 95 StringRef Name, Kind SectionKind) 96 : InputSectionBase(File, Hdr->sh_flags & ~SHF_INFO_LINK, 97 getType(Hdr->sh_type, Name), Hdr->sh_entsize, 98 Hdr->sh_link, Hdr->sh_info, Hdr->sh_addralign, 99 getSectionContents(File, Hdr), Name, SectionKind) { 100 // We reject object files having insanely large alignments even though 101 // they are allowed by the spec. I think 4GB is a reasonable limitation. 102 // We might want to relax this in the future. 103 if (Hdr->sh_addralign > UINT32_MAX) 104 fatal(toString(File) + ": section sh_addralign is too large"); 105 } 106 107 size_t InputSectionBase::getSize() const { 108 if (auto *S = dyn_cast<SyntheticSection>(this)) 109 return S->getSize(); 110 111 return Data.size(); 112 } 113 114 uint64_t InputSectionBase::getOffsetInFile() const { 115 const uint8_t *FileStart = (const uint8_t *)File->MB.getBufferStart(); 116 const uint8_t *SecStart = Data.begin(); 117 return SecStart - FileStart; 118 } 119 120 uint64_t SectionBase::getOffset(uint64_t Offset) const { 121 switch (kind()) { 122 case Output: { 123 auto *OS = cast<OutputSection>(this); 124 // For output sections we treat offset -1 as the end of the section. 125 return Offset == uint64_t(-1) ? OS->Size : Offset; 126 } 127 case Regular: 128 return cast<InputSection>(this)->OutSecOff + Offset; 129 case Synthetic: { 130 auto *IS = cast<InputSection>(this); 131 // For synthetic sections we treat offset -1 as the end of the section. 132 return IS->OutSecOff + (Offset == uint64_t(-1) ? IS->getSize() : Offset); 133 } 134 case EHFrame: 135 // The file crtbeginT.o has relocations pointing to the start of an empty 136 // .eh_frame that is known to be the first in the link. It does that to 137 // identify the start of the output .eh_frame. 138 return Offset; 139 case Merge: 140 const MergeInputSection *MS = cast<MergeInputSection>(this); 141 if (MS->MergeSec) 142 return MS->MergeSec->OutSecOff + MS->getOffset(Offset); 143 return MS->getOffset(Offset); 144 } 145 llvm_unreachable("invalid section kind"); 146 } 147 148 OutputSection *SectionBase::getOutputSection() { 149 if (auto *IS = dyn_cast<InputSection>(this)) 150 return IS->OutSec; 151 if (auto *MS = dyn_cast<MergeInputSection>(this)) 152 return MS->MergeSec ? MS->MergeSec->OutSec : nullptr; 153 if (auto *EH = dyn_cast<EhInputSection>(this)) 154 return EH->EHSec->OutSec; 155 return cast<OutputSection>(this); 156 } 157 158 // Uncompress section contents. Note that this function is called 159 // from parallel_for_each, so it must be thread-safe. 160 void InputSectionBase::uncompress() { 161 Decompressor Dec = check(Decompressor::create(Name, toStringRef(Data), 162 Config->IsLE, Config->Is64)); 163 164 size_t Size = Dec.getDecompressedSize(); 165 char *OutputBuf; 166 { 167 static std::mutex Mu; 168 std::lock_guard<std::mutex> Lock(Mu); 169 OutputBuf = BAlloc.Allocate<char>(Size); 170 } 171 172 if (Error E = Dec.decompress({OutputBuf, Size})) 173 fatal(toString(this) + 174 ": decompress failed: " + llvm::toString(std::move(E))); 175 Data = ArrayRef<uint8_t>((uint8_t *)OutputBuf, Size); 176 } 177 178 uint64_t SectionBase::getOffset(const DefinedRegular &Sym) const { 179 return getOffset(Sym.Value); 180 } 181 182 InputSectionBase *InputSectionBase::getLinkOrderDep() const { 183 if ((Flags & SHF_LINK_ORDER) && Link != 0) 184 return File->getSections()[Link]; 185 return nullptr; 186 } 187 188 // Returns a source location string. Used to construct an error message. 189 template <class ELFT> 190 std::string InputSectionBase::getLocation(uint64_t Offset) { 191 // We don't have file for synthetic sections. 192 if (getFile<ELFT>() == nullptr) 193 return (Config->OutputFile + ":(" + Name + "+0x" + utohexstr(Offset) + ")") 194 .str(); 195 196 // First check if we can get desired values from debugging information. 197 std::string LineInfo = getFile<ELFT>()->getLineInfo(this, Offset); 198 if (!LineInfo.empty()) 199 return LineInfo; 200 201 // File->SourceFile contains STT_FILE symbol that contains a 202 // source file name. If it's missing, we use an object file name. 203 std::string SrcFile = getFile<ELFT>()->SourceFile; 204 if (SrcFile.empty()) 205 SrcFile = toString(File); 206 207 // Find a function symbol that encloses a given location. 208 for (SymbolBody *B : getFile<ELFT>()->getSymbols()) 209 if (auto *D = dyn_cast<DefinedRegular>(B)) 210 if (D->Section == this && D->Type == STT_FUNC) 211 if (D->Value <= Offset && Offset < D->Value + D->Size) 212 return SrcFile + ":(function " + toString(*D) + ")"; 213 214 // If there's no symbol, print out the offset in the section. 215 return (SrcFile + ":(" + Name + "+0x" + utohexstr(Offset) + ")").str(); 216 } 217 218 // Returns a source location string. This function is intended to be 219 // used for constructing an error message. The returned message looks 220 // like this: 221 // 222 // foo.c:42 (/home/alice/possibly/very/long/path/foo.c:42) 223 // 224 // Returns an empty string if there's no way to get line info. 225 template <class ELFT> std::string InputSectionBase::getSrcMsg(uint64_t Offset) { 226 // Synthetic sections don't have input files. 227 elf::ObjectFile<ELFT> *File = getFile<ELFT>(); 228 if (!File) 229 return ""; 230 231 Optional<DILineInfo> Info = File->getDILineInfo(this, Offset); 232 233 // File->SourceFile contains STT_FILE symbol, and that is a last resort. 234 if (!Info) 235 return File->SourceFile; 236 237 std::string Path = Info->FileName; 238 std::string Filename = path::filename(Path); 239 std::string Lineno = ":" + std::to_string(Info->Line); 240 if (Filename == Path) 241 return Filename + Lineno; 242 return Filename + Lineno + " (" + Path + Lineno + ")"; 243 } 244 245 // Returns a filename string along with an optional section name. This 246 // function is intended to be used for constructing an error 247 // message. The returned message looks like this: 248 // 249 // path/to/foo.o:(function bar) 250 // 251 // or 252 // 253 // path/to/foo.o:(function bar) in archive path/to/bar.a 254 template <class ELFT> std::string InputSectionBase::getObjMsg(uint64_t Off) { 255 // Synthetic sections don't have input files. 256 elf::ObjectFile<ELFT> *File = getFile<ELFT>(); 257 std::string Filename = File ? File->getName() : "(internal)"; 258 259 std::string Archive; 260 if (!File->ArchiveName.empty()) 261 Archive = (" in archive " + File->ArchiveName).str(); 262 263 // Find a symbol that encloses a given location. 264 for (SymbolBody *B : getFile<ELFT>()->getSymbols()) 265 if (auto *D = dyn_cast<DefinedRegular>(B)) 266 if (D->Section == this && D->Value <= Off && Off < D->Value + D->Size) 267 return Filename + ":(" + toString(*D) + ")" + Archive; 268 269 // If there's no symbol, print out the offset in the section. 270 return (Filename + ":(" + Name + "+0x" + utohexstr(Off) + ")" + Archive) 271 .str(); 272 } 273 274 InputSectionBase InputSectionBase::Discarded; 275 276 InputSection::InputSection(uint64_t Flags, uint32_t Type, uint32_t Alignment, 277 ArrayRef<uint8_t> Data, StringRef Name, Kind K) 278 : InputSectionBase(nullptr, Flags, Type, 279 /*Entsize*/ 0, /*Link*/ 0, /*Info*/ 0, Alignment, Data, 280 Name, K) {} 281 282 template <class ELFT> 283 InputSection::InputSection(elf::ObjectFile<ELFT> *F, 284 const typename ELFT::Shdr *Header, StringRef Name) 285 : InputSectionBase(F, Header, Name, InputSectionBase::Regular) {} 286 287 bool InputSection::classof(const SectionBase *S) { 288 return S->kind() == SectionBase::Regular || 289 S->kind() == SectionBase::Synthetic; 290 } 291 292 bool InputSectionBase::classof(const SectionBase *S) { 293 return S->kind() != Output; 294 } 295 296 InputSectionBase *InputSection::getRelocatedSection() { 297 assert(this->Type == SHT_RELA || this->Type == SHT_REL); 298 ArrayRef<InputSectionBase *> Sections = this->File->getSections(); 299 return Sections[this->Info]; 300 } 301 302 // This is used for -r and --emit-relocs. We can't use memcpy to copy 303 // relocations because we need to update symbol table offset and section index 304 // for each relocation. So we copy relocations one by one. 305 template <class ELFT, class RelTy> 306 void InputSection::copyRelocations(uint8_t *Buf, ArrayRef<RelTy> Rels) { 307 InputSectionBase *RelocatedSection = getRelocatedSection(); 308 309 // Loop is slow and have complexity O(N*M), where N - amount of 310 // relocations and M - amount of symbols in symbol table. 311 // That happens because getSymbolIndex(...) call below performs 312 // simple linear search. 313 for (const RelTy &Rel : Rels) { 314 uint32_t Type = Rel.getType(Config->IsMips64EL); 315 SymbolBody &Body = this->getFile<ELFT>()->getRelocTargetSym(Rel); 316 317 auto *P = reinterpret_cast<typename ELFT::Rela *>(Buf); 318 Buf += sizeof(RelTy); 319 320 if (Config->IsRela) 321 P->r_addend = getAddend<ELFT>(Rel); 322 323 // Output section VA is zero for -r, so r_offset is an offset within the 324 // section, but for --emit-relocs it is an virtual address. 325 P->r_offset = RelocatedSection->OutSec->Addr + 326 RelocatedSection->getOffset(Rel.r_offset); 327 P->setSymbolAndType(In<ELFT>::SymTab->getSymbolIndex(&Body), Type, 328 Config->IsMips64EL); 329 330 if (Body.Type == STT_SECTION) { 331 // We combine multiple section symbols into only one per 332 // section. This means we have to update the addend. That is 333 // trivial for Elf_Rela, but for Elf_Rel we have to write to the 334 // section data. We do that by adding to the Relocation vector. 335 336 // .eh_frame is horribly special and can reference discarded sections. To 337 // avoid having to parse and recreate .eh_frame, we just replace any 338 // relocation in it pointing to discarded sections with R_*_NONE, which 339 // hopefully creates a frame that is ignored at runtime. 340 SectionBase *Section = cast<DefinedRegular>(Body).Section; 341 if (Section == &InputSection::Discarded) { 342 P->setSymbolAndType(0, 0, false); 343 continue; 344 } 345 346 if (Config->IsRela) { 347 P->r_addend += Body.getVA() - Section->getOutputSection()->Addr; 348 } else if (Config->Relocatable) { 349 const uint8_t *BufLoc = RelocatedSection->Data.begin() + Rel.r_offset; 350 RelocatedSection->Relocations.push_back( 351 {R_ABS, Type, Rel.r_offset, Target->getImplicitAddend(BufLoc, Type), 352 &Body}); 353 } 354 } 355 356 } 357 } 358 359 static uint32_t getARMUndefinedRelativeWeakVA(uint32_t Type, uint32_t A, 360 uint32_t P) { 361 switch (Type) { 362 case R_ARM_THM_JUMP11: 363 return P + 2; 364 case R_ARM_CALL: 365 case R_ARM_JUMP24: 366 case R_ARM_PC24: 367 case R_ARM_PLT32: 368 case R_ARM_PREL31: 369 case R_ARM_THM_JUMP19: 370 case R_ARM_THM_JUMP24: 371 return P + 4; 372 case R_ARM_THM_CALL: 373 // We don't want an interworking BLX to ARM 374 return P + 5; 375 default: 376 return A; 377 } 378 } 379 380 static uint64_t getAArch64UndefinedRelativeWeakVA(uint64_t Type, uint64_t A, 381 uint64_t P) { 382 switch (Type) { 383 case R_AARCH64_CALL26: 384 case R_AARCH64_CONDBR19: 385 case R_AARCH64_JUMP26: 386 case R_AARCH64_TSTBR14: 387 return P + 4; 388 default: 389 return A; 390 } 391 } 392 393 template <class ELFT> 394 static typename ELFT::uint 395 getRelocTargetVA(uint32_t Type, int64_t A, typename ELFT::uint P, 396 const SymbolBody &Body, RelExpr Expr) { 397 switch (Expr) { 398 case R_ABS: 399 case R_RELAX_GOT_PC_NOPIC: 400 return Body.getVA(A); 401 case R_GOT: 402 case R_RELAX_TLS_GD_TO_IE_ABS: 403 return Body.getGotVA<ELFT>() + A; 404 case R_GOTONLY_PC: 405 return In<ELFT>::Got->getVA() + A - P; 406 case R_GOTONLY_PC_FROM_END: 407 return In<ELFT>::Got->getVA() + A - P + In<ELFT>::Got->getSize(); 408 case R_GOTREL: 409 return Body.getVA(A) - In<ELFT>::Got->getVA(); 410 case R_GOTREL_FROM_END: 411 return Body.getVA(A) - In<ELFT>::Got->getVA() - In<ELFT>::Got->getSize(); 412 case R_GOT_FROM_END: 413 case R_RELAX_TLS_GD_TO_IE_END: 414 return Body.getGotOffset() + A - In<ELFT>::Got->getSize(); 415 case R_GOT_OFF: 416 return Body.getGotOffset() + A; 417 case R_GOT_PAGE_PC: 418 case R_RELAX_TLS_GD_TO_IE_PAGE_PC: 419 return getAArch64Page(Body.getGotVA<ELFT>() + A) - getAArch64Page(P); 420 case R_GOT_PC: 421 case R_RELAX_TLS_GD_TO_IE: 422 return Body.getGotVA<ELFT>() + A - P; 423 case R_HINT: 424 case R_NONE: 425 case R_TLSDESC_CALL: 426 llvm_unreachable("cannot relocate hint relocs"); 427 case R_MIPS_GOTREL: 428 return Body.getVA(A) - InX::MipsGot->getGp(); 429 case R_MIPS_GOT_GP: 430 return InX::MipsGot->getGp() + A; 431 case R_MIPS_GOT_GP_PC: { 432 // R_MIPS_LO16 expression has R_MIPS_GOT_GP_PC type iif the target 433 // is _gp_disp symbol. In that case we should use the following 434 // formula for calculation "AHL + GP - P + 4". For details see p. 4-19 at 435 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf 436 uint64_t V = InX::MipsGot->getGp() + A - P; 437 if (Type == R_MIPS_LO16) 438 V += 4; 439 return V; 440 } 441 case R_MIPS_GOT_LOCAL_PAGE: 442 // If relocation against MIPS local symbol requires GOT entry, this entry 443 // should be initialized by 'page address'. This address is high 16-bits 444 // of sum the symbol's value and the addend. 445 return InX::MipsGot->getVA() + InX::MipsGot->getPageEntryOffset(Body, A) - 446 InX::MipsGot->getGp(); 447 case R_MIPS_GOT_OFF: 448 case R_MIPS_GOT_OFF32: 449 // In case of MIPS if a GOT relocation has non-zero addend this addend 450 // should be applied to the GOT entry content not to the GOT entry offset. 451 // That is why we use separate expression type. 452 return InX::MipsGot->getVA() + InX::MipsGot->getBodyEntryOffset(Body, A) - 453 InX::MipsGot->getGp(); 454 case R_MIPS_TLSGD: 455 return InX::MipsGot->getVA() + InX::MipsGot->getTlsOffset() + 456 InX::MipsGot->getGlobalDynOffset(Body) - InX::MipsGot->getGp(); 457 case R_MIPS_TLSLD: 458 return InX::MipsGot->getVA() + InX::MipsGot->getTlsOffset() + 459 InX::MipsGot->getTlsIndexOff() - InX::MipsGot->getGp(); 460 case R_PAGE_PC: 461 case R_PLT_PAGE_PC: 462 if (Body.isUndefined() && !Body.isLocal() && Body.symbol()->isWeak()) 463 return getAArch64Page(A); 464 return getAArch64Page(Body.getVA(A)) - getAArch64Page(P); 465 case R_PC: 466 if (Body.isUndefined() && !Body.isLocal() && Body.symbol()->isWeak()) { 467 // On ARM and AArch64 a branch to an undefined weak resolves to the 468 // next instruction, otherwise the place. 469 if (Config->EMachine == EM_ARM) 470 return getARMUndefinedRelativeWeakVA(Type, A, P); 471 if (Config->EMachine == EM_AARCH64) 472 return getAArch64UndefinedRelativeWeakVA(Type, A, P); 473 } 474 return Body.getVA(A) - P; 475 case R_PLT: 476 return Body.getPltVA() + A; 477 case R_PLT_PC: 478 case R_PPC_PLT_OPD: 479 return Body.getPltVA() + A - P; 480 case R_PPC_OPD: { 481 uint64_t SymVA = Body.getVA(A); 482 // If we have an undefined weak symbol, we might get here with a symbol 483 // address of zero. That could overflow, but the code must be unreachable, 484 // so don't bother doing anything at all. 485 if (!SymVA) 486 return 0; 487 if (Out::Opd) { 488 // If this is a local call, and we currently have the address of a 489 // function-descriptor, get the underlying code address instead. 490 uint64_t OpdStart = Out::Opd->Addr; 491 uint64_t OpdEnd = OpdStart + Out::Opd->Size; 492 bool InOpd = OpdStart <= SymVA && SymVA < OpdEnd; 493 if (InOpd) 494 SymVA = read64be(&Out::OpdBuf[SymVA - OpdStart]); 495 } 496 return SymVA - P; 497 } 498 case R_PPC_TOC: 499 return getPPC64TocBase() + A; 500 case R_RELAX_GOT_PC: 501 return Body.getVA(A) - P; 502 case R_RELAX_TLS_GD_TO_LE: 503 case R_RELAX_TLS_IE_TO_LE: 504 case R_RELAX_TLS_LD_TO_LE: 505 case R_TLS: 506 // A weak undefined TLS symbol resolves to the base of the TLS 507 // block, i.e. gets a value of zero. If we pass --gc-sections to 508 // lld and .tbss is not referenced, it gets reclaimed and we don't 509 // create a TLS program header. Therefore, we resolve this 510 // statically to zero. 511 if (Body.isTls() && (Body.isLazy() || Body.isUndefined()) && 512 Body.symbol()->isWeak()) 513 return 0; 514 if (Target->TcbSize) 515 return Body.getVA(A) + alignTo(Target->TcbSize, Out::TlsPhdr->p_align); 516 return Body.getVA(A) - Out::TlsPhdr->p_memsz; 517 case R_RELAX_TLS_GD_TO_LE_NEG: 518 case R_NEG_TLS: 519 return Out::TlsPhdr->p_memsz - Body.getVA(A); 520 case R_SIZE: 521 return Body.getSize<ELFT>() + A; 522 case R_TLSDESC: 523 return In<ELFT>::Got->getGlobalDynAddr(Body) + A; 524 case R_TLSDESC_PAGE: 525 return getAArch64Page(In<ELFT>::Got->getGlobalDynAddr(Body) + A) - 526 getAArch64Page(P); 527 case R_TLSGD: 528 return In<ELFT>::Got->getGlobalDynOffset(Body) + A - 529 In<ELFT>::Got->getSize(); 530 case R_TLSGD_PC: 531 return In<ELFT>::Got->getGlobalDynAddr(Body) + A - P; 532 case R_TLSLD: 533 return In<ELFT>::Got->getTlsIndexOff() + A - In<ELFT>::Got->getSize(); 534 case R_TLSLD_PC: 535 return In<ELFT>::Got->getTlsIndexVA() + A - P; 536 } 537 llvm_unreachable("Invalid expression"); 538 } 539 540 // This function applies relocations to sections without SHF_ALLOC bit. 541 // Such sections are never mapped to memory at runtime. Debug sections are 542 // an example. Relocations in non-alloc sections are much easier to 543 // handle than in allocated sections because it will never need complex 544 // treatement such as GOT or PLT (because at runtime no one refers them). 545 // So, we handle relocations for non-alloc sections directly in this 546 // function as a performance optimization. 547 template <class ELFT, class RelTy> 548 void InputSection::relocateNonAlloc(uint8_t *Buf, ArrayRef<RelTy> Rels) { 549 for (const RelTy &Rel : Rels) { 550 uint32_t Type = Rel.getType(Config->IsMips64EL); 551 uint64_t Offset = getOffset(Rel.r_offset); 552 uint8_t *BufLoc = Buf + Offset; 553 int64_t Addend = getAddend<ELFT>(Rel); 554 if (!RelTy::IsRela) 555 Addend += Target->getImplicitAddend(BufLoc, Type); 556 557 SymbolBody &Sym = this->getFile<ELFT>()->getRelocTargetSym(Rel); 558 RelExpr Expr = Target->getRelExpr(Type, Sym, BufLoc); 559 if (Expr == R_NONE) 560 continue; 561 if (Expr != R_ABS) { 562 error(this->getLocation<ELFT>(Offset) + ": has non-ABS reloc"); 563 return; 564 } 565 566 uint64_t AddrLoc = this->OutSec->Addr + Offset; 567 uint64_t SymVA = 0; 568 if (!Sym.isTls() || Out::TlsPhdr) 569 SymVA = SignExtend64<sizeof(typename ELFT::uint) * 8>( 570 getRelocTargetVA<ELFT>(Type, Addend, AddrLoc, Sym, R_ABS)); 571 Target->relocateOne(BufLoc, Type, SymVA); 572 } 573 } 574 575 template <class ELFT> elf::ObjectFile<ELFT> *InputSectionBase::getFile() const { 576 return cast_or_null<elf::ObjectFile<ELFT>>(File); 577 } 578 579 template <class ELFT> 580 void InputSectionBase::relocate(uint8_t *Buf, uint8_t *BufEnd) { 581 // scanReloc function in Writer.cpp constructs Relocations 582 // vector only for SHF_ALLOC'ed sections. For other sections, 583 // we handle relocations directly here. 584 auto *IS = dyn_cast<InputSection>(this); 585 if (IS && !(IS->Flags & SHF_ALLOC)) { 586 if (IS->AreRelocsRela) 587 IS->relocateNonAlloc<ELFT>(Buf, IS->template relas<ELFT>()); 588 else 589 IS->relocateNonAlloc<ELFT>(Buf, IS->template rels<ELFT>()); 590 return; 591 } 592 593 const unsigned Bits = sizeof(typename ELFT::uint) * 8; 594 for (const Relocation &Rel : Relocations) { 595 uint64_t Offset = getOffset(Rel.Offset); 596 uint8_t *BufLoc = Buf + Offset; 597 uint32_t Type = Rel.Type; 598 599 uint64_t AddrLoc = getOutputSection()->Addr + Offset; 600 RelExpr Expr = Rel.Expr; 601 uint64_t TargetVA = SignExtend64<Bits>( 602 getRelocTargetVA<ELFT>(Type, Rel.Addend, AddrLoc, *Rel.Sym, Expr)); 603 604 switch (Expr) { 605 case R_RELAX_GOT_PC: 606 case R_RELAX_GOT_PC_NOPIC: 607 Target->relaxGot(BufLoc, TargetVA); 608 break; 609 case R_RELAX_TLS_IE_TO_LE: 610 Target->relaxTlsIeToLe(BufLoc, Type, TargetVA); 611 break; 612 case R_RELAX_TLS_LD_TO_LE: 613 Target->relaxTlsLdToLe(BufLoc, Type, TargetVA); 614 break; 615 case R_RELAX_TLS_GD_TO_LE: 616 case R_RELAX_TLS_GD_TO_LE_NEG: 617 Target->relaxTlsGdToLe(BufLoc, Type, TargetVA); 618 break; 619 case R_RELAX_TLS_GD_TO_IE: 620 case R_RELAX_TLS_GD_TO_IE_ABS: 621 case R_RELAX_TLS_GD_TO_IE_PAGE_PC: 622 case R_RELAX_TLS_GD_TO_IE_END: 623 Target->relaxTlsGdToIe(BufLoc, Type, TargetVA); 624 break; 625 case R_PPC_PLT_OPD: 626 // Patch a nop (0x60000000) to a ld. 627 if (BufLoc + 8 <= BufEnd && read32be(BufLoc + 4) == 0x60000000) 628 write32be(BufLoc + 4, 0xe8410028); // ld %r2, 40(%r1) 629 // fallthrough 630 default: 631 Target->relocateOne(BufLoc, Type, TargetVA); 632 break; 633 } 634 } 635 } 636 637 template <class ELFT> void InputSection::writeTo(uint8_t *Buf) { 638 if (this->Type == SHT_NOBITS) 639 return; 640 641 if (auto *S = dyn_cast<SyntheticSection>(this)) { 642 S->writeTo(Buf + OutSecOff); 643 return; 644 } 645 646 // If -r or --emit-relocs is given, then an InputSection 647 // may be a relocation section. 648 if (this->Type == SHT_RELA) { 649 copyRelocations<ELFT>(Buf + OutSecOff, 650 this->template getDataAs<typename ELFT::Rela>()); 651 return; 652 } 653 if (this->Type == SHT_REL) { 654 copyRelocations<ELFT>(Buf + OutSecOff, 655 this->template getDataAs<typename ELFT::Rel>()); 656 return; 657 } 658 659 // Copy section contents from source object file to output file 660 // and then apply relocations. 661 memcpy(Buf + OutSecOff, Data.data(), Data.size()); 662 uint8_t *BufEnd = Buf + OutSecOff + Data.size(); 663 this->relocate<ELFT>(Buf, BufEnd); 664 } 665 666 void InputSection::replace(InputSection *Other) { 667 this->Alignment = std::max(this->Alignment, Other->Alignment); 668 Other->Repl = this->Repl; 669 Other->Live = false; 670 } 671 672 template <class ELFT> 673 EhInputSection::EhInputSection(elf::ObjectFile<ELFT> *F, 674 const typename ELFT::Shdr *Header, 675 StringRef Name) 676 : InputSectionBase(F, Header, Name, InputSectionBase::EHFrame) { 677 // Mark .eh_frame sections as live by default because there are 678 // usually no relocations that point to .eh_frames. Otherwise, 679 // the garbage collector would drop all .eh_frame sections. 680 this->Live = true; 681 } 682 683 bool EhInputSection::classof(const SectionBase *S) { 684 return S->kind() == InputSectionBase::EHFrame; 685 } 686 687 // Returns the index of the first relocation that points to a region between 688 // Begin and Begin+Size. 689 template <class IntTy, class RelTy> 690 static unsigned getReloc(IntTy Begin, IntTy Size, const ArrayRef<RelTy> &Rels, 691 unsigned &RelocI) { 692 // Start search from RelocI for fast access. That works because the 693 // relocations are sorted in .eh_frame. 694 for (unsigned N = Rels.size(); RelocI < N; ++RelocI) { 695 const RelTy &Rel = Rels[RelocI]; 696 if (Rel.r_offset < Begin) 697 continue; 698 699 if (Rel.r_offset < Begin + Size) 700 return RelocI; 701 return -1; 702 } 703 return -1; 704 } 705 706 // .eh_frame is a sequence of CIE or FDE records. 707 // This function splits an input section into records and returns them. 708 template <class ELFT> void EhInputSection::split() { 709 // Early exit if already split. 710 if (!this->Pieces.empty()) 711 return; 712 713 if (this->NumRelocations) { 714 if (this->AreRelocsRela) 715 split<ELFT>(this->relas<ELFT>()); 716 else 717 split<ELFT>(this->rels<ELFT>()); 718 return; 719 } 720 split<ELFT>(makeArrayRef<typename ELFT::Rela>(nullptr, nullptr)); 721 } 722 723 template <class ELFT, class RelTy> 724 void EhInputSection::split(ArrayRef<RelTy> Rels) { 725 ArrayRef<uint8_t> Data = this->Data; 726 unsigned RelI = 0; 727 for (size_t Off = 0, End = Data.size(); Off != End;) { 728 size_t Size = readEhRecordSize<ELFT>(this, Off); 729 this->Pieces.emplace_back(Off, this, Size, getReloc(Off, Size, Rels, RelI)); 730 // The empty record is the end marker. 731 if (Size == 4) 732 break; 733 Off += Size; 734 } 735 } 736 737 static size_t findNull(ArrayRef<uint8_t> A, size_t EntSize) { 738 // Optimize the common case. 739 StringRef S((const char *)A.data(), A.size()); 740 if (EntSize == 1) 741 return S.find(0); 742 743 for (unsigned I = 0, N = S.size(); I != N; I += EntSize) { 744 const char *B = S.begin() + I; 745 if (std::all_of(B, B + EntSize, [](char C) { return C == 0; })) 746 return I; 747 } 748 return StringRef::npos; 749 } 750 751 // Split SHF_STRINGS section. Such section is a sequence of 752 // null-terminated strings. 753 void MergeInputSection::splitStrings(ArrayRef<uint8_t> Data, size_t EntSize) { 754 size_t Off = 0; 755 bool IsAlloc = this->Flags & SHF_ALLOC; 756 while (!Data.empty()) { 757 size_t End = findNull(Data, EntSize); 758 if (End == StringRef::npos) 759 fatal(toString(this) + ": string is not null terminated"); 760 size_t Size = End + EntSize; 761 Pieces.emplace_back(Off, !IsAlloc); 762 Hashes.push_back(hash_value(toStringRef(Data.slice(0, Size)))); 763 Data = Data.slice(Size); 764 Off += Size; 765 } 766 } 767 768 // Split non-SHF_STRINGS section. Such section is a sequence of 769 // fixed size records. 770 void MergeInputSection::splitNonStrings(ArrayRef<uint8_t> Data, 771 size_t EntSize) { 772 size_t Size = Data.size(); 773 assert((Size % EntSize) == 0); 774 bool IsAlloc = this->Flags & SHF_ALLOC; 775 for (unsigned I = 0, N = Size; I != N; I += EntSize) { 776 Hashes.push_back(hash_value(toStringRef(Data.slice(I, EntSize)))); 777 Pieces.emplace_back(I, !IsAlloc); 778 } 779 } 780 781 template <class ELFT> 782 MergeInputSection::MergeInputSection(elf::ObjectFile<ELFT> *F, 783 const typename ELFT::Shdr *Header, 784 StringRef Name) 785 : InputSectionBase(F, Header, Name, InputSectionBase::Merge) {} 786 787 // This function is called after we obtain a complete list of input sections 788 // that need to be linked. This is responsible to split section contents 789 // into small chunks for further processing. 790 // 791 // Note that this function is called from parallel_for_each. This must be 792 // thread-safe (i.e. no memory allocation from the pools). 793 void MergeInputSection::splitIntoPieces() { 794 ArrayRef<uint8_t> Data = this->Data; 795 uint64_t EntSize = this->Entsize; 796 if (this->Flags & SHF_STRINGS) 797 splitStrings(Data, EntSize); 798 else 799 splitNonStrings(Data, EntSize); 800 801 if (Config->GcSections && (this->Flags & SHF_ALLOC)) 802 for (uint64_t Off : LiveOffsets) 803 this->getSectionPiece(Off)->Live = true; 804 } 805 806 bool MergeInputSection::classof(const SectionBase *S) { 807 return S->kind() == InputSectionBase::Merge; 808 } 809 810 // Do binary search to get a section piece at a given input offset. 811 SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) { 812 auto *This = static_cast<const MergeInputSection *>(this); 813 return const_cast<SectionPiece *>(This->getSectionPiece(Offset)); 814 } 815 816 template <class It, class T, class Compare> 817 static It fastUpperBound(It First, It Last, const T &Value, Compare Comp) { 818 size_t Size = std::distance(First, Last); 819 assert(Size != 0); 820 while (Size != 1) { 821 size_t H = Size / 2; 822 const It MI = First + H; 823 Size -= H; 824 First = Comp(Value, *MI) ? First : First + H; 825 } 826 return Comp(Value, *First) ? First : First + 1; 827 } 828 829 const SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) const { 830 uint64_t Size = this->Data.size(); 831 if (Offset >= Size) 832 fatal(toString(this) + ": entry is past the end of the section"); 833 834 // Find the element this offset points to. 835 auto I = fastUpperBound( 836 Pieces.begin(), Pieces.end(), Offset, 837 [](const uint64_t &A, const SectionPiece &B) { return A < B.InputOff; }); 838 --I; 839 return &*I; 840 } 841 842 // Returns the offset in an output section for a given input offset. 843 // Because contents of a mergeable section is not contiguous in output, 844 // it is not just an addition to a base output offset. 845 uint64_t MergeInputSection::getOffset(uint64_t Offset) const { 846 // Initialize OffsetMap lazily. 847 std::call_once(InitOffsetMap, [&] { 848 OffsetMap.reserve(Pieces.size()); 849 for (const SectionPiece &Piece : Pieces) 850 OffsetMap[Piece.InputOff] = Piece.OutputOff; 851 }); 852 853 // Find a string starting at a given offset. 854 auto It = OffsetMap.find(Offset); 855 if (It != OffsetMap.end()) 856 return It->second; 857 858 if (!this->Live) 859 return 0; 860 861 // If Offset is not at beginning of a section piece, it is not in the map. 862 // In that case we need to search from the original section piece vector. 863 const SectionPiece &Piece = *this->getSectionPiece(Offset); 864 if (!Piece.Live) 865 return 0; 866 867 uint64_t Addend = Offset - Piece.InputOff; 868 return Piece.OutputOff + Addend; 869 } 870 871 template InputSection::InputSection(elf::ObjectFile<ELF32LE> *, 872 const ELF32LE::Shdr *, StringRef); 873 template InputSection::InputSection(elf::ObjectFile<ELF32BE> *, 874 const ELF32BE::Shdr *, StringRef); 875 template InputSection::InputSection(elf::ObjectFile<ELF64LE> *, 876 const ELF64LE::Shdr *, StringRef); 877 template InputSection::InputSection(elf::ObjectFile<ELF64BE> *, 878 const ELF64BE::Shdr *, StringRef); 879 880 template std::string InputSectionBase::getLocation<ELF32LE>(uint64_t); 881 template std::string InputSectionBase::getLocation<ELF32BE>(uint64_t); 882 template std::string InputSectionBase::getLocation<ELF64LE>(uint64_t); 883 template std::string InputSectionBase::getLocation<ELF64BE>(uint64_t); 884 885 template std::string InputSectionBase::getSrcMsg<ELF32LE>(uint64_t); 886 template std::string InputSectionBase::getSrcMsg<ELF32BE>(uint64_t); 887 template std::string InputSectionBase::getSrcMsg<ELF64LE>(uint64_t); 888 template std::string InputSectionBase::getSrcMsg<ELF64BE>(uint64_t); 889 890 template std::string InputSectionBase::getObjMsg<ELF32LE>(uint64_t); 891 template std::string InputSectionBase::getObjMsg<ELF32BE>(uint64_t); 892 template std::string InputSectionBase::getObjMsg<ELF64LE>(uint64_t); 893 template std::string InputSectionBase::getObjMsg<ELF64BE>(uint64_t); 894 895 template void InputSection::writeTo<ELF32LE>(uint8_t *); 896 template void InputSection::writeTo<ELF32BE>(uint8_t *); 897 template void InputSection::writeTo<ELF64LE>(uint8_t *); 898 template void InputSection::writeTo<ELF64BE>(uint8_t *); 899 900 template elf::ObjectFile<ELF32LE> *InputSectionBase::getFile<ELF32LE>() const; 901 template elf::ObjectFile<ELF32BE> *InputSectionBase::getFile<ELF32BE>() const; 902 template elf::ObjectFile<ELF64LE> *InputSectionBase::getFile<ELF64LE>() const; 903 template elf::ObjectFile<ELF64BE> *InputSectionBase::getFile<ELF64BE>() const; 904 905 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF32LE> *, 906 const ELF32LE::Shdr *, StringRef); 907 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF32BE> *, 908 const ELF32BE::Shdr *, StringRef); 909 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF64LE> *, 910 const ELF64LE::Shdr *, StringRef); 911 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF64BE> *, 912 const ELF64BE::Shdr *, StringRef); 913 914 template EhInputSection::EhInputSection(elf::ObjectFile<ELF32LE> *, 915 const ELF32LE::Shdr *, StringRef); 916 template EhInputSection::EhInputSection(elf::ObjectFile<ELF32BE> *, 917 const ELF32BE::Shdr *, StringRef); 918 template EhInputSection::EhInputSection(elf::ObjectFile<ELF64LE> *, 919 const ELF64LE::Shdr *, StringRef); 920 template EhInputSection::EhInputSection(elf::ObjectFile<ELF64BE> *, 921 const ELF64BE::Shdr *, StringRef); 922 923 template void EhInputSection::split<ELF32LE>(); 924 template void EhInputSection::split<ELF32BE>(); 925 template void EhInputSection::split<ELF64LE>(); 926 template void EhInputSection::split<ELF64BE>(); 927