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 <mutex> 26 27 using namespace llvm; 28 using namespace llvm::ELF; 29 using namespace llvm::object; 30 using namespace llvm::support; 31 using namespace llvm::support::endian; 32 33 using namespace lld; 34 using namespace lld::elf; 35 36 std::vector<InputSectionBase *> elf::InputSections; 37 38 // Returns a string to construct an error message. 39 std::string lld::toString(const InputSectionBase *Sec) { 40 // File can be absent if section is synthetic. 41 std::string FileName = Sec->File ? Sec->File->getName() : "<internal>"; 42 return (FileName + ":(" + 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 InputSectionBase InputSectionBase::Discarded; 219 220 InputSection::InputSection(uint64_t Flags, uint32_t Type, uint32_t Alignment, 221 ArrayRef<uint8_t> Data, StringRef Name, Kind K) 222 : InputSectionBase(nullptr, Flags, Type, 223 /*Entsize*/ 0, /*Link*/ 0, /*Info*/ 0, Alignment, Data, 224 Name, K) {} 225 226 template <class ELFT> 227 InputSection::InputSection(elf::ObjectFile<ELFT> *F, 228 const typename ELFT::Shdr *Header, StringRef Name) 229 : InputSectionBase(F, Header, Name, InputSectionBase::Regular) {} 230 231 bool InputSection::classof(const SectionBase *S) { 232 return S->kind() == SectionBase::Regular || 233 S->kind() == SectionBase::Synthetic; 234 } 235 236 bool InputSectionBase::classof(const SectionBase *S) { 237 return S->kind() != Output; 238 } 239 240 InputSectionBase *InputSection::getRelocatedSection() { 241 assert(this->Type == SHT_RELA || this->Type == SHT_REL); 242 ArrayRef<InputSectionBase *> Sections = this->File->getSections(); 243 return Sections[this->Info]; 244 } 245 246 // This is used for -r and --emit-relocs. We can't use memcpy to copy 247 // relocations because we need to update symbol table offset and section index 248 // for each relocation. So we copy relocations one by one. 249 template <class ELFT, class RelTy> 250 void InputSection::copyRelocations(uint8_t *Buf, ArrayRef<RelTy> Rels) { 251 InputSectionBase *RelocatedSection = getRelocatedSection(); 252 253 // Loop is slow and have complexity O(N*M), where N - amount of 254 // relocations and M - amount of symbols in symbol table. 255 // That happens because getSymbolIndex(...) call below performs 256 // simple linear search. 257 for (const RelTy &Rel : Rels) { 258 uint32_t Type = Rel.getType(Config->IsMips64EL); 259 SymbolBody &Body = this->getFile<ELFT>()->getRelocTargetSym(Rel); 260 261 auto *P = reinterpret_cast<typename ELFT::Rela *>(Buf); 262 Buf += sizeof(RelTy); 263 264 if (Config->IsRela) 265 P->r_addend = getAddend<ELFT>(Rel); 266 267 // Output section VA is zero for -r, so r_offset is an offset within the 268 // section, but for --emit-relocs it is an virtual address. 269 P->r_offset = RelocatedSection->OutSec->Addr + 270 RelocatedSection->getOffset(Rel.r_offset); 271 P->setSymbolAndType(In<ELFT>::SymTab->getSymbolIndex(&Body), Type, 272 Config->IsMips64EL); 273 274 if (Body.Type == STT_SECTION) { 275 // We combine multiple section symbols into only one per 276 // section. This means we have to update the addend. That is 277 // trivial for Elf_Rela, but for Elf_Rel we have to write to the 278 // section data. We do that by adding to the Relocation vector. 279 280 // .eh_frame is horribly special and can reference discarded sections. To 281 // avoid having to parse and recreate .eh_frame, we just replace any 282 // relocation in it pointing to discarded sections with R_*_NONE, which 283 // hopefully creates a frame that is ignored at runtime. 284 SectionBase *Section = cast<DefinedRegular>(Body).Section; 285 if (Section == &InputSection::Discarded) { 286 P->setSymbolAndType(0, 0, false); 287 continue; 288 } 289 290 if (Config->IsRela) { 291 P->r_addend += Body.getVA() - Section->getOutputSection()->Addr; 292 } else if (Config->Relocatable) { 293 const uint8_t *BufLoc = RelocatedSection->Data.begin() + Rel.r_offset; 294 RelocatedSection->Relocations.push_back( 295 {R_ABS, Type, Rel.r_offset, Target->getImplicitAddend(BufLoc, Type), 296 &Body}); 297 } 298 } 299 300 } 301 } 302 303 static uint32_t getARMUndefinedRelativeWeakVA(uint32_t Type, uint32_t A, 304 uint32_t P) { 305 switch (Type) { 306 case R_ARM_THM_JUMP11: 307 return P + 2; 308 case R_ARM_CALL: 309 case R_ARM_JUMP24: 310 case R_ARM_PC24: 311 case R_ARM_PLT32: 312 case R_ARM_PREL31: 313 case R_ARM_THM_JUMP19: 314 case R_ARM_THM_JUMP24: 315 return P + 4; 316 case R_ARM_THM_CALL: 317 // We don't want an interworking BLX to ARM 318 return P + 5; 319 default: 320 return A; 321 } 322 } 323 324 static uint64_t getAArch64UndefinedRelativeWeakVA(uint64_t Type, uint64_t A, 325 uint64_t P) { 326 switch (Type) { 327 case R_AARCH64_CALL26: 328 case R_AARCH64_CONDBR19: 329 case R_AARCH64_JUMP26: 330 case R_AARCH64_TSTBR14: 331 return P + 4; 332 default: 333 return A; 334 } 335 } 336 337 template <class ELFT> 338 static typename ELFT::uint 339 getRelocTargetVA(uint32_t Type, int64_t A, typename ELFT::uint P, 340 const SymbolBody &Body, RelExpr Expr) { 341 switch (Expr) { 342 case R_HINT: 343 case R_NONE: 344 case R_TLSDESC_CALL: 345 llvm_unreachable("cannot relocate hint relocs"); 346 case R_TLSLD: 347 return In<ELFT>::Got->getTlsIndexOff() + A - In<ELFT>::Got->getSize(); 348 case R_TLSLD_PC: 349 return In<ELFT>::Got->getTlsIndexVA() + A - P; 350 case R_PPC_TOC: 351 return getPPC64TocBase() + A; 352 case R_TLSGD: 353 return In<ELFT>::Got->getGlobalDynOffset(Body) + A - 354 In<ELFT>::Got->getSize(); 355 case R_TLSGD_PC: 356 return In<ELFT>::Got->getGlobalDynAddr(Body) + A - P; 357 case R_TLSDESC: 358 return In<ELFT>::Got->getGlobalDynAddr(Body) + A; 359 case R_TLSDESC_PAGE: 360 return getAArch64Page(In<ELFT>::Got->getGlobalDynAddr(Body) + A) - 361 getAArch64Page(P); 362 case R_PLT: 363 return Body.getPltVA() + A; 364 case R_PLT_PC: 365 case R_PPC_PLT_OPD: 366 return Body.getPltVA() + A - P; 367 case R_SIZE: 368 return Body.getSize<ELFT>() + A; 369 case R_GOTREL: 370 return Body.getVA(A) - In<ELFT>::Got->getVA(); 371 case R_GOTREL_FROM_END: 372 return Body.getVA(A) - In<ELFT>::Got->getVA() - In<ELFT>::Got->getSize(); 373 case R_RELAX_TLS_GD_TO_IE_END: 374 case R_GOT_FROM_END: 375 return Body.getGotOffset() + A - In<ELFT>::Got->getSize(); 376 case R_RELAX_TLS_GD_TO_IE_ABS: 377 case R_GOT: 378 return Body.getGotVA<ELFT>() + A; 379 case R_RELAX_TLS_GD_TO_IE_PAGE_PC: 380 case R_GOT_PAGE_PC: 381 return getAArch64Page(Body.getGotVA<ELFT>() + A) - getAArch64Page(P); 382 case R_RELAX_TLS_GD_TO_IE: 383 case R_GOT_PC: 384 return Body.getGotVA<ELFT>() + A - P; 385 case R_GOTONLY_PC: 386 return In<ELFT>::Got->getVA() + A - P; 387 case R_GOTONLY_PC_FROM_END: 388 return In<ELFT>::Got->getVA() + A - P + In<ELFT>::Got->getSize(); 389 case R_RELAX_TLS_LD_TO_LE: 390 case R_RELAX_TLS_IE_TO_LE: 391 case R_RELAX_TLS_GD_TO_LE: 392 case R_TLS: 393 // A weak undefined TLS symbol resolves to the base of the TLS 394 // block, i.e. gets a value of zero. If we pass --gc-sections to 395 // lld and .tbss is not referenced, it gets reclaimed and we don't 396 // create a TLS program header. Therefore, we resolve this 397 // statically to zero. 398 if (Body.isTls() && (Body.isLazy() || Body.isUndefined()) && 399 Body.symbol()->isWeak()) 400 return 0; 401 if (Target->TcbSize) 402 return Body.getVA(A) + alignTo(Target->TcbSize, Out::TlsPhdr->p_align); 403 return Body.getVA(A) - Out::TlsPhdr->p_memsz; 404 case R_RELAX_TLS_GD_TO_LE_NEG: 405 case R_NEG_TLS: 406 return Out::TlsPhdr->p_memsz - Body.getVA(A); 407 case R_ABS: 408 case R_RELAX_GOT_PC_NOPIC: 409 return Body.getVA(A); 410 case R_GOT_OFF: 411 return Body.getGotOffset() + A; 412 case R_MIPS_GOT_LOCAL_PAGE: 413 // If relocation against MIPS local symbol requires GOT entry, this entry 414 // should be initialized by 'page address'. This address is high 16-bits 415 // of sum the symbol's value and the addend. 416 return In<ELFT>::MipsGot->getVA() + 417 In<ELFT>::MipsGot->getPageEntryOffset(Body, A) - 418 In<ELFT>::MipsGot->getGp(); 419 case R_MIPS_GOT_OFF: 420 case R_MIPS_GOT_OFF32: 421 // In case of MIPS if a GOT relocation has non-zero addend this addend 422 // should be applied to the GOT entry content not to the GOT entry offset. 423 // That is why we use separate expression type. 424 return In<ELFT>::MipsGot->getVA() + 425 In<ELFT>::MipsGot->getBodyEntryOffset(Body, A) - 426 In<ELFT>::MipsGot->getGp(); 427 case R_MIPS_GOTREL: 428 return Body.getVA(A) - In<ELFT>::MipsGot->getGp(); 429 case R_MIPS_GOT_GP: 430 return In<ELFT>::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 = In<ELFT>::MipsGot->getGp() + A - P; 437 if (Type == R_MIPS_LO16) 438 V += 4; 439 return V; 440 } 441 case R_MIPS_TLSGD: 442 return In<ELFT>::MipsGot->getVA() + In<ELFT>::MipsGot->getTlsOffset() + 443 In<ELFT>::MipsGot->getGlobalDynOffset(Body) - 444 In<ELFT>::MipsGot->getGp(); 445 case R_MIPS_TLSLD: 446 return In<ELFT>::MipsGot->getVA() + In<ELFT>::MipsGot->getTlsOffset() + 447 In<ELFT>::MipsGot->getTlsIndexOff() - In<ELFT>::MipsGot->getGp(); 448 case R_PPC_OPD: { 449 uint64_t SymVA = Body.getVA(A); 450 // If we have an undefined weak symbol, we might get here with a symbol 451 // address of zero. That could overflow, but the code must be unreachable, 452 // so don't bother doing anything at all. 453 if (!SymVA) 454 return 0; 455 if (Out::Opd) { 456 // If this is a local call, and we currently have the address of a 457 // function-descriptor, get the underlying code address instead. 458 uint64_t OpdStart = Out::Opd->Addr; 459 uint64_t OpdEnd = OpdStart + Out::Opd->Size; 460 bool InOpd = OpdStart <= SymVA && SymVA < OpdEnd; 461 if (InOpd) 462 SymVA = read64be(&Out::OpdBuf[SymVA - OpdStart]); 463 } 464 return SymVA - P; 465 } 466 case R_PC: 467 if (Body.isUndefined() && !Body.isLocal() && Body.symbol()->isWeak()) { 468 // On ARM and AArch64 a branch to an undefined weak resolves to the 469 // next instruction, otherwise the place. 470 if (Config->EMachine == EM_ARM) 471 return getARMUndefinedRelativeWeakVA(Type, A, P); 472 if (Config->EMachine == EM_AARCH64) 473 return getAArch64UndefinedRelativeWeakVA(Type, A, P); 474 } 475 case R_RELAX_GOT_PC: 476 return Body.getVA(A) - P; 477 case R_PLT_PAGE_PC: 478 case R_PAGE_PC: 479 if (Body.isUndefined() && !Body.isLocal() && Body.symbol()->isWeak()) 480 return getAArch64Page(A); 481 return getAArch64Page(Body.getVA(A)) - getAArch64Page(P); 482 } 483 llvm_unreachable("Invalid expression"); 484 } 485 486 // This function applies relocations to sections without SHF_ALLOC bit. 487 // Such sections are never mapped to memory at runtime. Debug sections are 488 // an example. Relocations in non-alloc sections are much easier to 489 // handle than in allocated sections because it will never need complex 490 // treatement such as GOT or PLT (because at runtime no one refers them). 491 // So, we handle relocations for non-alloc sections directly in this 492 // function as a performance optimization. 493 template <class ELFT, class RelTy> 494 void InputSection::relocateNonAlloc(uint8_t *Buf, ArrayRef<RelTy> Rels) { 495 typedef typename ELFT::uint uintX_t; 496 for (const RelTy &Rel : Rels) { 497 uint32_t Type = Rel.getType(Config->IsMips64EL); 498 uint64_t Offset = getOffset(Rel.r_offset); 499 uint8_t *BufLoc = Buf + Offset; 500 int64_t Addend = getAddend<ELFT>(Rel); 501 if (!RelTy::IsRela) 502 Addend += Target->getImplicitAddend(BufLoc, Type); 503 504 SymbolBody &Sym = this->getFile<ELFT>()->getRelocTargetSym(Rel); 505 RelExpr Expr = Target->getRelExpr(Type, Sym); 506 if (Expr == R_NONE) 507 continue; 508 if (Expr != R_ABS) { 509 error(this->getLocation<ELFT>(Offset) + ": has non-ABS reloc"); 510 return; 511 } 512 513 uintX_t AddrLoc = this->OutSec->Addr + Offset; 514 uint64_t SymVA = 0; 515 if (!Sym.isTls() || Out::TlsPhdr) 516 SymVA = SignExtend64<sizeof(uintX_t) * 8>( 517 getRelocTargetVA<ELFT>(Type, Addend, AddrLoc, Sym, R_ABS)); 518 Target->relocateOne(BufLoc, Type, SymVA); 519 } 520 } 521 522 template <class ELFT> elf::ObjectFile<ELFT> *InputSectionBase::getFile() const { 523 return cast_or_null<elf::ObjectFile<ELFT>>(File); 524 } 525 526 template <class ELFT> 527 void InputSectionBase::relocate(uint8_t *Buf, uint8_t *BufEnd) { 528 // scanReloc function in Writer.cpp constructs Relocations 529 // vector only for SHF_ALLOC'ed sections. For other sections, 530 // we handle relocations directly here. 531 auto *IS = dyn_cast<InputSection>(this); 532 if (IS && !(IS->Flags & SHF_ALLOC)) { 533 if (IS->AreRelocsRela) 534 IS->relocateNonAlloc<ELFT>(Buf, IS->template relas<ELFT>()); 535 else 536 IS->relocateNonAlloc<ELFT>(Buf, IS->template rels<ELFT>()); 537 return; 538 } 539 540 typedef typename ELFT::uint uintX_t; 541 const unsigned Bits = sizeof(uintX_t) * 8; 542 for (const Relocation &Rel : Relocations) { 543 uint64_t Offset = getOffset(Rel.Offset); 544 uint8_t *BufLoc = Buf + Offset; 545 uint32_t Type = Rel.Type; 546 547 uintX_t AddrLoc = getOutputSection()->Addr + Offset; 548 RelExpr Expr = Rel.Expr; 549 uint64_t TargetVA = SignExtend64<Bits>( 550 getRelocTargetVA<ELFT>(Type, Rel.Addend, AddrLoc, *Rel.Sym, Expr)); 551 552 switch (Expr) { 553 case R_RELAX_GOT_PC: 554 case R_RELAX_GOT_PC_NOPIC: 555 Target->relaxGot(BufLoc, TargetVA); 556 break; 557 case R_RELAX_TLS_IE_TO_LE: 558 Target->relaxTlsIeToLe(BufLoc, Type, TargetVA); 559 break; 560 case R_RELAX_TLS_LD_TO_LE: 561 Target->relaxTlsLdToLe(BufLoc, Type, TargetVA); 562 break; 563 case R_RELAX_TLS_GD_TO_LE: 564 case R_RELAX_TLS_GD_TO_LE_NEG: 565 Target->relaxTlsGdToLe(BufLoc, Type, TargetVA); 566 break; 567 case R_RELAX_TLS_GD_TO_IE: 568 case R_RELAX_TLS_GD_TO_IE_ABS: 569 case R_RELAX_TLS_GD_TO_IE_PAGE_PC: 570 case R_RELAX_TLS_GD_TO_IE_END: 571 Target->relaxTlsGdToIe(BufLoc, Type, TargetVA); 572 break; 573 case R_PPC_PLT_OPD: 574 // Patch a nop (0x60000000) to a ld. 575 if (BufLoc + 8 <= BufEnd && read32be(BufLoc + 4) == 0x60000000) 576 write32be(BufLoc + 4, 0xe8410028); // ld %r2, 40(%r1) 577 // fallthrough 578 default: 579 Target->relocateOne(BufLoc, Type, TargetVA); 580 break; 581 } 582 } 583 } 584 585 template <class ELFT> void InputSection::writeTo(uint8_t *Buf) { 586 if (this->Type == SHT_NOBITS) 587 return; 588 589 if (auto *S = dyn_cast<SyntheticSection>(this)) { 590 S->writeTo(Buf + OutSecOff); 591 return; 592 } 593 594 // If -r or --emit-relocs is given, then an InputSection 595 // may be a relocation section. 596 if (this->Type == SHT_RELA) { 597 copyRelocations<ELFT>(Buf + OutSecOff, 598 this->template getDataAs<typename ELFT::Rela>()); 599 return; 600 } 601 if (this->Type == SHT_REL) { 602 copyRelocations<ELFT>(Buf + OutSecOff, 603 this->template getDataAs<typename ELFT::Rel>()); 604 return; 605 } 606 607 // Copy section contents from source object file to output file. 608 ArrayRef<uint8_t> Data = this->Data; 609 memcpy(Buf + OutSecOff, Data.data(), Data.size()); 610 611 // Iterate over all relocation sections that apply to this section. 612 uint8_t *BufEnd = Buf + OutSecOff + Data.size(); 613 this->relocate<ELFT>(Buf, BufEnd); 614 } 615 616 void InputSection::replace(InputSection *Other) { 617 this->Alignment = std::max(this->Alignment, Other->Alignment); 618 Other->Repl = this->Repl; 619 Other->Live = false; 620 } 621 622 template <class ELFT> 623 EhInputSection::EhInputSection(elf::ObjectFile<ELFT> *F, 624 const typename ELFT::Shdr *Header, 625 StringRef Name) 626 : InputSectionBase(F, Header, Name, InputSectionBase::EHFrame) { 627 // Mark .eh_frame sections as live by default because there are 628 // usually no relocations that point to .eh_frames. Otherwise, 629 // the garbage collector would drop all .eh_frame sections. 630 this->Live = true; 631 } 632 633 bool EhInputSection::classof(const SectionBase *S) { 634 return S->kind() == InputSectionBase::EHFrame; 635 } 636 637 // Returns the index of the first relocation that points to a region between 638 // Begin and Begin+Size. 639 template <class IntTy, class RelTy> 640 static unsigned getReloc(IntTy Begin, IntTy Size, const ArrayRef<RelTy> &Rels, 641 unsigned &RelocI) { 642 // Start search from RelocI for fast access. That works because the 643 // relocations are sorted in .eh_frame. 644 for (unsigned N = Rels.size(); RelocI < N; ++RelocI) { 645 const RelTy &Rel = Rels[RelocI]; 646 if (Rel.r_offset < Begin) 647 continue; 648 649 if (Rel.r_offset < Begin + Size) 650 return RelocI; 651 return -1; 652 } 653 return -1; 654 } 655 656 // .eh_frame is a sequence of CIE or FDE records. 657 // This function splits an input section into records and returns them. 658 template <class ELFT> void EhInputSection::split() { 659 // Early exit if already split. 660 if (!this->Pieces.empty()) 661 return; 662 663 if (this->NumRelocations) { 664 if (this->AreRelocsRela) 665 split<ELFT>(this->relas<ELFT>()); 666 else 667 split<ELFT>(this->rels<ELFT>()); 668 return; 669 } 670 split<ELFT>(makeArrayRef<typename ELFT::Rela>(nullptr, nullptr)); 671 } 672 673 template <class ELFT, class RelTy> 674 void EhInputSection::split(ArrayRef<RelTy> Rels) { 675 ArrayRef<uint8_t> Data = this->Data; 676 unsigned RelI = 0; 677 for (size_t Off = 0, End = Data.size(); Off != End;) { 678 size_t Size = readEhRecordSize<ELFT>(this, Off); 679 this->Pieces.emplace_back(Off, this, Size, getReloc(Off, Size, Rels, RelI)); 680 // The empty record is the end marker. 681 if (Size == 4) 682 break; 683 Off += Size; 684 } 685 } 686 687 static size_t findNull(ArrayRef<uint8_t> A, size_t EntSize) { 688 // Optimize the common case. 689 StringRef S((const char *)A.data(), A.size()); 690 if (EntSize == 1) 691 return S.find(0); 692 693 for (unsigned I = 0, N = S.size(); I != N; I += EntSize) { 694 const char *B = S.begin() + I; 695 if (std::all_of(B, B + EntSize, [](char C) { return C == 0; })) 696 return I; 697 } 698 return StringRef::npos; 699 } 700 701 // Split SHF_STRINGS section. Such section is a sequence of 702 // null-terminated strings. 703 void MergeInputSection::splitStrings(ArrayRef<uint8_t> Data, size_t EntSize) { 704 size_t Off = 0; 705 bool IsAlloc = this->Flags & SHF_ALLOC; 706 while (!Data.empty()) { 707 size_t End = findNull(Data, EntSize); 708 if (End == StringRef::npos) 709 fatal(toString(this) + ": string is not null terminated"); 710 size_t Size = End + EntSize; 711 Pieces.emplace_back(Off, !IsAlloc); 712 Hashes.push_back(hash_value(toStringRef(Data.slice(0, Size)))); 713 Data = Data.slice(Size); 714 Off += Size; 715 } 716 } 717 718 // Split non-SHF_STRINGS section. Such section is a sequence of 719 // fixed size records. 720 void MergeInputSection::splitNonStrings(ArrayRef<uint8_t> Data, 721 size_t EntSize) { 722 size_t Size = Data.size(); 723 assert((Size % EntSize) == 0); 724 bool IsAlloc = this->Flags & SHF_ALLOC; 725 for (unsigned I = 0, N = Size; I != N; I += EntSize) { 726 Hashes.push_back(hash_value(toStringRef(Data.slice(I, EntSize)))); 727 Pieces.emplace_back(I, !IsAlloc); 728 } 729 } 730 731 template <class ELFT> 732 MergeInputSection::MergeInputSection(elf::ObjectFile<ELFT> *F, 733 const typename ELFT::Shdr *Header, 734 StringRef Name) 735 : InputSectionBase(F, Header, Name, InputSectionBase::Merge) {} 736 737 // This function is called after we obtain a complete list of input sections 738 // that need to be linked. This is responsible to split section contents 739 // into small chunks for further processing. 740 // 741 // Note that this function is called from parallel_for_each. This must be 742 // thread-safe (i.e. no memory allocation from the pools). 743 void MergeInputSection::splitIntoPieces() { 744 ArrayRef<uint8_t> Data = this->Data; 745 uint64_t EntSize = this->Entsize; 746 if (this->Flags & SHF_STRINGS) 747 splitStrings(Data, EntSize); 748 else 749 splitNonStrings(Data, EntSize); 750 751 if (Config->GcSections && (this->Flags & SHF_ALLOC)) 752 for (uint64_t Off : LiveOffsets) 753 this->getSectionPiece(Off)->Live = true; 754 } 755 756 bool MergeInputSection::classof(const SectionBase *S) { 757 return S->kind() == InputSectionBase::Merge; 758 } 759 760 // Do binary search to get a section piece at a given input offset. 761 SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) { 762 auto *This = static_cast<const MergeInputSection *>(this); 763 return const_cast<SectionPiece *>(This->getSectionPiece(Offset)); 764 } 765 766 template <class It, class T, class Compare> 767 static It fastUpperBound(It First, It Last, const T &Value, Compare Comp) { 768 size_t Size = std::distance(First, Last); 769 assert(Size != 0); 770 while (Size != 1) { 771 size_t H = Size / 2; 772 const It MI = First + H; 773 Size -= H; 774 First = Comp(Value, *MI) ? First : First + H; 775 } 776 return Comp(Value, *First) ? First : First + 1; 777 } 778 779 const SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) const { 780 uint64_t Size = this->Data.size(); 781 if (Offset >= Size) 782 fatal(toString(this) + ": entry is past the end of the section"); 783 784 // Find the element this offset points to. 785 auto I = fastUpperBound( 786 Pieces.begin(), Pieces.end(), Offset, 787 [](const uint64_t &A, const SectionPiece &B) { return A < B.InputOff; }); 788 --I; 789 return &*I; 790 } 791 792 // Returns the offset in an output section for a given input offset. 793 // Because contents of a mergeable section is not contiguous in output, 794 // it is not just an addition to a base output offset. 795 uint64_t MergeInputSection::getOffset(uint64_t Offset) const { 796 // Initialize OffsetMap lazily. 797 std::call_once(InitOffsetMap, [&] { 798 OffsetMap.reserve(Pieces.size()); 799 for (const SectionPiece &Piece : Pieces) 800 OffsetMap[Piece.InputOff] = Piece.OutputOff; 801 }); 802 803 // Find a string starting at a given offset. 804 auto It = OffsetMap.find(Offset); 805 if (It != OffsetMap.end()) 806 return It->second; 807 808 if (!this->Live) 809 return 0; 810 811 // If Offset is not at beginning of a section piece, it is not in the map. 812 // In that case we need to search from the original section piece vector. 813 const SectionPiece &Piece = *this->getSectionPiece(Offset); 814 if (!Piece.Live) 815 return 0; 816 817 uint64_t Addend = Offset - Piece.InputOff; 818 return Piece.OutputOff + Addend; 819 } 820 821 template InputSection::InputSection(elf::ObjectFile<ELF32LE> *F, 822 const ELF32LE::Shdr *Header, 823 StringRef Name); 824 template InputSection::InputSection(elf::ObjectFile<ELF32BE> *F, 825 const ELF32BE::Shdr *Header, 826 StringRef Name); 827 template InputSection::InputSection(elf::ObjectFile<ELF64LE> *F, 828 const ELF64LE::Shdr *Header, 829 StringRef Name); 830 template InputSection::InputSection(elf::ObjectFile<ELF64BE> *F, 831 const ELF64BE::Shdr *Header, 832 StringRef Name); 833 834 template std::string InputSectionBase::getLocation<ELF32LE>(uint64_t Offset); 835 template std::string InputSectionBase::getLocation<ELF32BE>(uint64_t Offset); 836 template std::string InputSectionBase::getLocation<ELF64LE>(uint64_t Offset); 837 template std::string InputSectionBase::getLocation<ELF64BE>(uint64_t Offset); 838 839 template void InputSection::writeTo<ELF32LE>(uint8_t *Buf); 840 template void InputSection::writeTo<ELF32BE>(uint8_t *Buf); 841 template void InputSection::writeTo<ELF64LE>(uint8_t *Buf); 842 template void InputSection::writeTo<ELF64BE>(uint8_t *Buf); 843 844 template elf::ObjectFile<ELF32LE> *InputSectionBase::getFile<ELF32LE>() const; 845 template elf::ObjectFile<ELF32BE> *InputSectionBase::getFile<ELF32BE>() const; 846 template elf::ObjectFile<ELF64LE> *InputSectionBase::getFile<ELF64LE>() const; 847 template elf::ObjectFile<ELF64BE> *InputSectionBase::getFile<ELF64BE>() const; 848 849 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF32LE> *F, 850 const ELF32LE::Shdr *Header, 851 StringRef Name); 852 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF32BE> *F, 853 const ELF32BE::Shdr *Header, 854 StringRef Name); 855 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF64LE> *F, 856 const ELF64LE::Shdr *Header, 857 StringRef Name); 858 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF64BE> *F, 859 const ELF64BE::Shdr *Header, 860 StringRef Name); 861 862 template EhInputSection::EhInputSection(elf::ObjectFile<ELF32LE> *F, 863 const ELF32LE::Shdr *Header, 864 StringRef Name); 865 template EhInputSection::EhInputSection(elf::ObjectFile<ELF32BE> *F, 866 const ELF32BE::Shdr *Header, 867 StringRef Name); 868 template EhInputSection::EhInputSection(elf::ObjectFile<ELF64LE> *F, 869 const ELF64LE::Shdr *Header, 870 StringRef Name); 871 template EhInputSection::EhInputSection(elf::ObjectFile<ELF64BE> *F, 872 const ELF64BE::Shdr *Header, 873 StringRef Name); 874 875 template void EhInputSection::split<ELF32LE>(); 876 template void EhInputSection::split<ELF32BE>(); 877 template void EhInputSection::split<ELF64LE>(); 878 template void EhInputSection::split<ELF64BE>(); 879