1 //===- Archive.cpp - ar File Format implementation --------------*- C++ -*-===// 2 // 3 // The LLVM Compiler Infrastructure 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 defines the ArchiveObjectFile class. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/Object/Archive.h" 15 #include "llvm/ADT/APInt.h" 16 #include "llvm/ADT/SmallString.h" 17 #include "llvm/ADT/Twine.h" 18 #include "llvm/Support/Endian.h" 19 #include "llvm/Support/MemoryBuffer.h" 20 #include "llvm/Support/Path.h" 21 22 using namespace llvm; 23 using namespace object; 24 using namespace llvm::support::endian; 25 26 static const char *const Magic = "!<arch>\n"; 27 static const char *const ThinMagic = "!<thin>\n"; 28 29 void Archive::anchor() { } 30 31 StringRef ArchiveMemberHeader::getName() const { 32 char EndCond; 33 if (Name[0] == '/' || Name[0] == '#') 34 EndCond = ' '; 35 else 36 EndCond = '/'; 37 llvm::StringRef::size_type end = 38 llvm::StringRef(Name, sizeof(Name)).find(EndCond); 39 if (end == llvm::StringRef::npos) 40 end = sizeof(Name); 41 assert(end <= sizeof(Name) && end > 0); 42 // Don't include the EndCond if there is one. 43 return llvm::StringRef(Name, end); 44 } 45 46 ErrorOr<uint32_t> ArchiveMemberHeader::getSize() const { 47 uint32_t Ret; 48 if (llvm::StringRef(Size, sizeof(Size)).rtrim(" ").getAsInteger(10, Ret)) 49 return object_error::parse_failed; 50 return Ret; 51 } 52 53 sys::fs::perms ArchiveMemberHeader::getAccessMode() const { 54 unsigned Ret; 55 if (StringRef(AccessMode, sizeof(AccessMode)).rtrim(" ").getAsInteger(8, Ret)) 56 llvm_unreachable("Access mode is not an octal number."); 57 return static_cast<sys::fs::perms>(Ret); 58 } 59 60 sys::TimeValue ArchiveMemberHeader::getLastModified() const { 61 unsigned Seconds; 62 if (StringRef(LastModified, sizeof(LastModified)).rtrim(" ") 63 .getAsInteger(10, Seconds)) 64 llvm_unreachable("Last modified time not a decimal number."); 65 66 sys::TimeValue Ret; 67 Ret.fromEpochTime(Seconds); 68 return Ret; 69 } 70 71 unsigned ArchiveMemberHeader::getUID() const { 72 unsigned Ret; 73 if (StringRef(UID, sizeof(UID)).rtrim(" ").getAsInteger(10, Ret)) 74 llvm_unreachable("UID time not a decimal number."); 75 return Ret; 76 } 77 78 unsigned ArchiveMemberHeader::getGID() const { 79 unsigned Ret; 80 if (StringRef(GID, sizeof(GID)).rtrim(" ").getAsInteger(10, Ret)) 81 llvm_unreachable("GID time not a decimal number."); 82 return Ret; 83 } 84 85 Archive::Child::Child(const Archive *Parent, const char *Start) 86 : Parent(Parent) { 87 if (!Start) 88 return; 89 90 uint64_t Size = sizeof(ArchiveMemberHeader); 91 Data = StringRef(Start, Size); 92 if (!isThinMember()) { 93 Size += getRawSize(); 94 Data = StringRef(Start, Size); 95 } 96 97 // Setup StartOfFile and PaddingBytes. 98 StartOfFile = sizeof(ArchiveMemberHeader); 99 // Don't include attached name. 100 StringRef Name = getRawName(); 101 if (Name.startswith("#1/")) { 102 uint64_t NameSize; 103 if (Name.substr(3).rtrim(" ").getAsInteger(10, NameSize)) 104 llvm_unreachable("Long name length is not an integer"); 105 StartOfFile += NameSize; 106 } 107 } 108 109 uint64_t Archive::Child::getSize() const { 110 if (Parent->IsThin) { 111 ErrorOr<uint32_t> Size = getHeader()->getSize(); 112 if (Size.getError()) 113 return 0; 114 return Size.get(); 115 } 116 return Data.size() - StartOfFile; 117 } 118 119 uint64_t Archive::Child::getRawSize() const { 120 ErrorOr<uint32_t> Size = getHeader()->getSize(); 121 if (Size.getError()) 122 return 0; 123 return Size.get(); 124 } 125 126 bool Archive::Child::isThinMember() const { 127 StringRef Name = getHeader()->getName(); 128 return Parent->IsThin && Name != "/" && Name != "//"; 129 } 130 131 ErrorOr<StringRef> Archive::Child::getBuffer() const { 132 if (!isThinMember()) 133 return StringRef(Data.data() + StartOfFile, getSize()); 134 ErrorOr<StringRef> Name = getName(); 135 if (std::error_code EC = Name.getError()) 136 return EC; 137 SmallString<128> FullName = sys::path::parent_path( 138 Parent->getMemoryBufferRef().getBufferIdentifier()); 139 sys::path::append(FullName, *Name); 140 ErrorOr<std::unique_ptr<MemoryBuffer>> Buf = MemoryBuffer::getFile(FullName); 141 if (std::error_code EC = Buf.getError()) 142 return EC; 143 Parent->ThinBuffers.push_back(std::move(*Buf)); 144 return Parent->ThinBuffers.back()->getBuffer(); 145 } 146 147 Archive::Child Archive::Child::getNext() const { 148 size_t SpaceToSkip = Data.size(); 149 // If it's odd, add 1 to make it even. 150 if (SpaceToSkip & 1) 151 ++SpaceToSkip; 152 153 const char *NextLoc = Data.data() + SpaceToSkip; 154 155 // Check to see if this is past the end of the archive. 156 if (NextLoc >= Parent->Data.getBufferEnd()) 157 return Child(Parent, nullptr); 158 159 return Child(Parent, NextLoc); 160 } 161 162 uint64_t Archive::Child::getChildOffset() const { 163 const char *a = Parent->Data.getBuffer().data(); 164 const char *c = Data.data(); 165 uint64_t offset = c - a; 166 return offset; 167 } 168 169 ErrorOr<StringRef> Archive::Child::getName() const { 170 StringRef name = getRawName(); 171 // Check if it's a special name. 172 if (name[0] == '/') { 173 if (name.size() == 1) // Linker member. 174 return name; 175 if (name.size() == 2 && name[1] == '/') // String table. 176 return name; 177 // It's a long name. 178 // Get the offset. 179 std::size_t offset; 180 if (name.substr(1).rtrim(" ").getAsInteger(10, offset)) 181 llvm_unreachable("Long name offset is not an integer"); 182 183 // Verify it. 184 if (offset >= Parent->StringTable.size()) 185 return object_error::parse_failed; 186 const char *addr = Parent->StringTable.begin() + offset; 187 188 // GNU long file names end with a "/\n". 189 if (Parent->kind() == K_GNU || Parent->kind() == K_MIPS64) { 190 StringRef::size_type End = StringRef(addr).find('\n'); 191 return StringRef(addr, End - 1); 192 } 193 return StringRef(addr); 194 } else if (name.startswith("#1/")) { 195 uint64_t name_size; 196 if (name.substr(3).rtrim(" ").getAsInteger(10, name_size)) 197 llvm_unreachable("Long name length is not an ingeter"); 198 return Data.substr(sizeof(ArchiveMemberHeader), name_size) 199 .rtrim(StringRef("\0", 1)); 200 } 201 // It's a simple name. 202 if (name[name.size() - 1] == '/') 203 return name.substr(0, name.size() - 1); 204 return name; 205 } 206 207 ErrorOr<MemoryBufferRef> Archive::Child::getMemoryBufferRef() const { 208 ErrorOr<StringRef> NameOrErr = getName(); 209 if (std::error_code EC = NameOrErr.getError()) 210 return EC; 211 StringRef Name = NameOrErr.get(); 212 ErrorOr<StringRef> Buf = getBuffer(); 213 if (std::error_code EC = Buf.getError()) 214 return EC; 215 return MemoryBufferRef(*Buf, Name); 216 } 217 218 ErrorOr<std::unique_ptr<Binary>> 219 Archive::Child::getAsBinary(LLVMContext *Context) const { 220 ErrorOr<MemoryBufferRef> BuffOrErr = getMemoryBufferRef(); 221 if (std::error_code EC = BuffOrErr.getError()) 222 return EC; 223 224 return createBinary(BuffOrErr.get(), Context); 225 } 226 227 ErrorOr<std::unique_ptr<Archive>> Archive::create(MemoryBufferRef Source) { 228 std::error_code EC; 229 std::unique_ptr<Archive> Ret(new Archive(Source, EC)); 230 if (EC) 231 return EC; 232 return std::move(Ret); 233 } 234 235 Archive::Archive(MemoryBufferRef Source, std::error_code &ec) 236 : Binary(Binary::ID_Archive, Source), SymbolTable(child_end()), 237 FirstRegular(child_end()) { 238 StringRef Buffer = Data.getBuffer(); 239 // Check for sufficient magic. 240 if (Buffer.startswith(ThinMagic)) { 241 IsThin = true; 242 } else if (Buffer.startswith(Magic)) { 243 IsThin = false; 244 } else { 245 ec = object_error::invalid_file_type; 246 return; 247 } 248 249 // Get the special members. 250 child_iterator i = child_begin(false); 251 child_iterator e = child_end(); 252 253 if (i == e) { 254 ec = std::error_code(); 255 return; 256 } 257 258 StringRef Name = i->getRawName(); 259 260 // Below is the pattern that is used to figure out the archive format 261 // GNU archive format 262 // First member : / (may exist, if it exists, points to the symbol table ) 263 // Second member : // (may exist, if it exists, points to the string table) 264 // Note : The string table is used if the filename exceeds 15 characters 265 // BSD archive format 266 // First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table) 267 // There is no string table, if the filename exceeds 15 characters or has a 268 // embedded space, the filename has #1/<size>, The size represents the size 269 // of the filename that needs to be read after the archive header 270 // COFF archive format 271 // First member : / 272 // Second member : / (provides a directory of symbols) 273 // Third member : // (may exist, if it exists, contains the string table) 274 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present 275 // even if the string table is empty. However, lib.exe does not in fact 276 // seem to create the third member if there's no member whose filename 277 // exceeds 15 characters. So the third member is optional. 278 279 if (Name == "__.SYMDEF") { 280 Format = K_BSD; 281 SymbolTable = i; 282 ++i; 283 FirstRegular = i; 284 ec = std::error_code(); 285 return; 286 } 287 288 if (Name.startswith("#1/")) { 289 Format = K_BSD; 290 // We know this is BSD, so getName will work since there is no string table. 291 ErrorOr<StringRef> NameOrErr = i->getName(); 292 ec = NameOrErr.getError(); 293 if (ec) 294 return; 295 Name = NameOrErr.get(); 296 if (Name == "__.SYMDEF SORTED" || Name == "__.SYMDEF") { 297 SymbolTable = i; 298 ++i; 299 } 300 FirstRegular = i; 301 return; 302 } 303 304 // MIPS 64-bit ELF archives use a special format of a symbol table. 305 // This format is marked by `ar_name` field equals to "/SYM64/". 306 // For detailed description see page 96 in the following document: 307 // http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf 308 309 bool has64SymTable = false; 310 if (Name == "/" || Name == "/SYM64/") { 311 SymbolTable = i; 312 if (Name == "/SYM64/") 313 has64SymTable = true; 314 315 ++i; 316 if (i == e) { 317 ec = std::error_code(); 318 return; 319 } 320 Name = i->getRawName(); 321 } 322 323 if (Name == "//") { 324 Format = has64SymTable ? K_MIPS64 : K_GNU; 325 // The string table is never an external member, so we just assert on the 326 // ErrorOr. 327 StringTable = *i->getBuffer(); 328 ++i; 329 FirstRegular = i; 330 ec = std::error_code(); 331 return; 332 } 333 334 if (Name[0] != '/') { 335 Format = has64SymTable ? K_MIPS64 : K_GNU; 336 FirstRegular = i; 337 ec = std::error_code(); 338 return; 339 } 340 341 if (Name != "/") { 342 ec = object_error::parse_failed; 343 return; 344 } 345 346 Format = K_COFF; 347 SymbolTable = i; 348 349 ++i; 350 if (i == e) { 351 FirstRegular = i; 352 ec = std::error_code(); 353 return; 354 } 355 356 Name = i->getRawName(); 357 358 if (Name == "//") { 359 // The string table is never an external member, so we just assert on the 360 // ErrorOr. 361 StringTable = *i->getBuffer(); 362 ++i; 363 } 364 365 FirstRegular = i; 366 ec = std::error_code(); 367 } 368 369 Archive::child_iterator Archive::child_begin(bool SkipInternal) const { 370 if (Data.getBufferSize() == 8) // empty archive. 371 return child_end(); 372 373 if (SkipInternal) 374 return FirstRegular; 375 376 const char *Loc = Data.getBufferStart() + strlen(Magic); 377 Child c(this, Loc); 378 return c; 379 } 380 381 Archive::child_iterator Archive::child_end() const { 382 return Child(this, nullptr); 383 } 384 385 StringRef Archive::Symbol::getName() const { 386 return Parent->getSymbolTable().begin() + StringIndex; 387 } 388 389 ErrorOr<Archive::child_iterator> Archive::Symbol::getMember() const { 390 const char *Buf = Parent->getSymbolTable().begin(); 391 const char *Offsets = Buf; 392 if (Parent->kind() == K_MIPS64) 393 Offsets += sizeof(uint64_t); 394 else 395 Offsets += sizeof(uint32_t); 396 uint32_t Offset = 0; 397 if (Parent->kind() == K_GNU) { 398 Offset = read32be(Offsets + SymbolIndex * 4); 399 } else if (Parent->kind() == K_MIPS64) { 400 Offset = read64be(Offsets + SymbolIndex * 8); 401 } else if (Parent->kind() == K_BSD) { 402 // The SymbolIndex is an index into the ranlib structs that start at 403 // Offsets (the first uint32_t is the number of bytes of the ranlib 404 // structs). The ranlib structs are a pair of uint32_t's the first 405 // being a string table offset and the second being the offset into 406 // the archive of the member that defines the symbol. Which is what 407 // is needed here. 408 Offset = read32le(Offsets + SymbolIndex * 8 + 4); 409 } else { 410 // Skip offsets. 411 uint32_t MemberCount = read32le(Buf); 412 Buf += MemberCount * 4 + 4; 413 414 uint32_t SymbolCount = read32le(Buf); 415 if (SymbolIndex >= SymbolCount) 416 return object_error::parse_failed; 417 418 // Skip SymbolCount to get to the indices table. 419 const char *Indices = Buf + 4; 420 421 // Get the index of the offset in the file member offset table for this 422 // symbol. 423 uint16_t OffsetIndex = read16le(Indices + SymbolIndex * 2); 424 // Subtract 1 since OffsetIndex is 1 based. 425 --OffsetIndex; 426 427 if (OffsetIndex >= MemberCount) 428 return object_error::parse_failed; 429 430 Offset = read32le(Offsets + OffsetIndex * 4); 431 } 432 433 const char *Loc = Parent->getData().begin() + Offset; 434 child_iterator Iter(Child(Parent, Loc)); 435 return Iter; 436 } 437 438 Archive::Symbol Archive::Symbol::getNext() const { 439 Symbol t(*this); 440 if (Parent->kind() == K_BSD) { 441 // t.StringIndex is an offset from the start of the __.SYMDEF or 442 // "__.SYMDEF SORTED" member into the string table for the ranlib 443 // struct indexed by t.SymbolIndex . To change t.StringIndex to the 444 // offset in the string table for t.SymbolIndex+1 we subtract the 445 // its offset from the start of the string table for t.SymbolIndex 446 // and add the offset of the string table for t.SymbolIndex+1. 447 448 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t 449 // which is the number of bytes of ranlib structs that follow. The ranlib 450 // structs are a pair of uint32_t's the first being a string table offset 451 // and the second being the offset into the archive of the member that 452 // define the symbol. After that the next uint32_t is the byte count of 453 // the string table followed by the string table. 454 const char *Buf = Parent->getSymbolTable().begin(); 455 uint32_t RanlibCount = 0; 456 RanlibCount = read32le(Buf) / 8; 457 // If t.SymbolIndex + 1 will be past the count of symbols (the RanlibCount) 458 // don't change the t.StringIndex as we don't want to reference a ranlib 459 // past RanlibCount. 460 if (t.SymbolIndex + 1 < RanlibCount) { 461 const char *Ranlibs = Buf + 4; 462 uint32_t CurRanStrx = 0; 463 uint32_t NextRanStrx = 0; 464 CurRanStrx = read32le(Ranlibs + t.SymbolIndex * 8); 465 NextRanStrx = read32le(Ranlibs + (t.SymbolIndex + 1) * 8); 466 t.StringIndex -= CurRanStrx; 467 t.StringIndex += NextRanStrx; 468 } 469 } else { 470 // Go to one past next null. 471 t.StringIndex = Parent->getSymbolTable().find('\0', t.StringIndex) + 1; 472 } 473 ++t.SymbolIndex; 474 return t; 475 } 476 477 Archive::symbol_iterator Archive::symbol_begin() const { 478 if (!hasSymbolTable()) 479 return symbol_iterator(Symbol(this, 0, 0)); 480 481 const char *buf = getSymbolTable().begin(); 482 if (kind() == K_GNU) { 483 uint32_t symbol_count = 0; 484 symbol_count = read32be(buf); 485 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t))); 486 } else if (kind() == K_MIPS64) { 487 uint64_t symbol_count = read64be(buf); 488 buf += sizeof(uint64_t) + (symbol_count * (sizeof(uint64_t))); 489 } else if (kind() == K_BSD) { 490 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t 491 // which is the number of bytes of ranlib structs that follow. The ranlib 492 // structs are a pair of uint32_t's the first being a string table offset 493 // and the second being the offset into the archive of the member that 494 // define the symbol. After that the next uint32_t is the byte count of 495 // the string table followed by the string table. 496 uint32_t ranlib_count = 0; 497 ranlib_count = read32le(buf) / 8; 498 const char *ranlibs = buf + 4; 499 uint32_t ran_strx = 0; 500 ran_strx = read32le(ranlibs); 501 buf += sizeof(uint32_t) + (ranlib_count * (2 * (sizeof(uint32_t)))); 502 // Skip the byte count of the string table. 503 buf += sizeof(uint32_t); 504 buf += ran_strx; 505 } else { 506 uint32_t member_count = 0; 507 uint32_t symbol_count = 0; 508 member_count = read32le(buf); 509 buf += 4 + (member_count * 4); // Skip offsets. 510 symbol_count = read32le(buf); 511 buf += 4 + (symbol_count * 2); // Skip indices. 512 } 513 uint32_t string_start_offset = buf - getSymbolTable().begin(); 514 return symbol_iterator(Symbol(this, 0, string_start_offset)); 515 } 516 517 Archive::symbol_iterator Archive::symbol_end() const { 518 return symbol_iterator(Symbol(this, getNumberOfSymbols(), 0)); 519 } 520 521 uint32_t Archive::getNumberOfSymbols() const { 522 if (!hasSymbolTable()) 523 return 0; 524 const char *buf = getSymbolTable().begin(); 525 if (kind() == K_GNU) 526 return read32be(buf); 527 if (kind() == K_MIPS64) 528 return read64be(buf); 529 if (kind() == K_BSD) 530 return read32le(buf) / 8; 531 uint32_t member_count = 0; 532 member_count = read32le(buf); 533 buf += 4 + (member_count * 4); // Skip offsets. 534 return read32le(buf); 535 } 536 537 Archive::child_iterator Archive::findSym(StringRef name) const { 538 Archive::symbol_iterator bs = symbol_begin(); 539 Archive::symbol_iterator es = symbol_end(); 540 541 for (; bs != es; ++bs) { 542 StringRef SymName = bs->getName(); 543 if (SymName == name) { 544 ErrorOr<Archive::child_iterator> ResultOrErr = bs->getMember(); 545 // FIXME: Should we really eat the error? 546 if (ResultOrErr.getError()) 547 return child_end(); 548 return ResultOrErr.get(); 549 } 550 } 551 return child_end(); 552 } 553 554 bool Archive::hasSymbolTable() const { 555 return SymbolTable != child_end(); 556 } 557