ArchiveReader.cpp revision dd95e8d71ec04cc91796c008773cd237a4ed107e
1//===-- ArchiveReader.cpp - Read LLVM archive files -------------*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file was developed by Reid Spencer and is distributed under the 6// University of Illinois Open Source License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// Builds up standard unix archive files (.a) containing LLVM bytecode. 11// 12//===----------------------------------------------------------------------===// 13 14#include "ArchiveInternals.h" 15#include "llvm/Bytecode/Reader.h" 16 17using namespace llvm; 18 19/// Read a variable-bit-rate encoded unsigned integer 20inline unsigned readInteger(const char*&At, const char*End) { 21 unsigned Shift = 0; 22 unsigned Result = 0; 23 24 do { 25 if (At == End) 26 throw std::string("Ran out of data reading vbr_uint!"); 27 Result |= (unsigned)((*At++) & 0x7F) << Shift; 28 Shift += 7; 29 } while (At[-1] & 0x80); 30 return Result; 31} 32 33// Completely parse the Archive's symbol table and populate symTab member var. 34void 35Archive::parseSymbolTable(const void* data, unsigned size) { 36 const char* At = (const char*) data; 37 const char* End = At + size; 38 while (At < End) { 39 unsigned offset = readInteger(At, End); 40 unsigned length = readInteger(At, End); 41 if (At + length > End) 42 throw std::string("malformed symbol table"); 43 // we don't care if it can't be inserted (duplicate entry) 44 symTab.insert(std::make_pair(std::string(At,length),offset)); 45 At += length; 46 } 47 symTabSize = size; 48} 49 50// This member parses an ArchiveMemberHeader that is presumed to be pointed to 51// by At. The At pointer is updated to the byte just after the header, which 52// can be variable in size. 53ArchiveMember* 54Archive::parseMemberHeader(const char*& At, const char* End) { 55 assert(At + sizeof(ArchiveMemberHeader) < End && "Not enough data"); 56 57 // Cast archive member header 58 ArchiveMemberHeader* Hdr = (ArchiveMemberHeader*)At; 59 At += sizeof(ArchiveMemberHeader); 60 61 // Instantiate the ArchiveMember to be filled 62 ArchiveMember* member = new ArchiveMember(this); 63 64 // Extract the size and determine if the file is 65 // compressed or not (negative length). 66 int flags = 0; 67 int MemberSize = atoi(Hdr->size); 68 if (MemberSize < 0) { 69 flags |= ArchiveMember::CompressedFlag; 70 MemberSize = -MemberSize; 71 } 72 73 // Check the size of the member for sanity 74 if (At + MemberSize > End) 75 throw std::string("invalid member length in archive file"); 76 77 // Check the member signature 78 if (!Hdr->checkSignature()) 79 throw std::string("invalid file member signature"); 80 81 // Convert and check the member name 82 // The empty name ( '/' and 15 blanks) is for a foreign (non-LLVM) symbol 83 // table. The special name "//" and 14 blanks is for a string table, used 84 // for long file names. This library doesn't generate either of those but 85 // it will accept them. If the name starts with #1/ and the remainder is 86 // digits, then those digits specify the length of the name that is 87 // stored immediately following the header. The special name 88 // __LLVM_SYM_TAB__ identifies the symbol table for LLVM bytecode. 89 // Anything else is a regular, short filename that is terminated with 90 // a '/' and blanks. 91 92 std::string pathname; 93 unsigned index; 94 switch (Hdr->name[0]) { 95 case '#': 96 if (Hdr->name[1] == '1' && Hdr->name[2] == '/') { 97 if (isdigit(Hdr->name[3])) { 98 unsigned len = atoi(&Hdr->name[3]); 99 pathname.assign(At,len); 100 At += len; 101 MemberSize -= len; 102 flags |= ArchiveMember::HasLongFilenameFlag; 103 } else 104 throw std::string("invalid long filename"); 105 } else if (Hdr->name[1] == '_' && 106 (0==memcmp(Hdr->name,ARFILE_LLVM_SYMTAB_NAME,16))) { 107 // The member is using a long file name (>15 chars) format. 108 // This format is standard for 4.4BSD and Mac OSX operating 109 // systems. LLVM uses it similarly. In this format, the 110 // remainder of the name field (after #1/) specifies the 111 // length of the file name which occupy the first bytes of 112 // the member's data. The pathname already has the #1/ stripped. 113 pathname.assign(ARFILE_LLVM_SYMTAB_NAME); 114 flags |= ArchiveMember::LLVMSymbolTableFlag; 115 } 116 break; 117 case '/': 118 if (Hdr->name[1]== '/') { 119 if (0==memcmp(Hdr->name,ARFILE_STRTAB_NAME,16)) { 120 pathname.assign(ARFILE_STRTAB_NAME); 121 flags |= ArchiveMember::StringTableFlag; 122 } else { 123 throw std::string("invalid string table name"); 124 } 125 } else if (Hdr->name[1] == ' ') { 126 if (0==memcmp(Hdr->name,ARFILE_SYMTAB_NAME,16)) { 127 pathname.assign(ARFILE_SYMTAB_NAME); 128 flags |= ArchiveMember::ForeignSymbolTableFlag; 129 } else { 130 throw std::string("invalid foreign symbol table name"); 131 } 132 } else if (isdigit(Hdr->name[1])) { 133 unsigned index = atoi(&Hdr->name[1]); 134 if (index < strtab.length()) { 135 const char* namep = strtab.c_str() + index; 136 const char* endp = strtab.c_str() + strtab.length(); 137 const char* p = namep; 138 const char* last_p = p; 139 while (p < endp) { 140 if (*p == '\n' && *last_p == '/') { 141 pathname.assign(namep,last_p-namep); 142 flags |= ArchiveMember::HasLongFilenameFlag; 143 break; 144 } 145 last_p = p; 146 p++; 147 } 148 if (p >= endp) 149 throw std::string("missing name termiantor in string table"); 150 } else { 151 throw std::string("name index beyond string table"); 152 } 153 } 154 break; 155 156 default: 157 char* slash = (char*) memchr(Hdr->name,'/',16); 158 if (slash == 0) 159 slash = Hdr->name + 16; 160 pathname.assign(Hdr->name,slash-Hdr->name); 161 break; 162 } 163 164 // Determine if this is a bytecode file 165 switch (sys::IdentifyFileType(At,4)) { 166 case sys::BytecodeFileType: 167 flags |= ArchiveMember::BytecodeFlag; 168 break; 169 case sys::CompressedBytecodeFileType: 170 flags |= ArchiveMember::CompressedBytecodeFlag; 171 flags &= ~ArchiveMember::CompressedFlag; 172 break; 173 default: 174 flags &= ~(ArchiveMember::BytecodeFlag| 175 ArchiveMember::CompressedBytecodeFlag); 176 break; 177 } 178 179 // Fill in fields of the ArchiveMember 180 member->next = 0; 181 member->prev = 0; 182 member->parent = this; 183 member->path.setFile(pathname); 184 member->info.fileSize = MemberSize; 185 member->info.modTime.fromEpochTime(atoi(Hdr->date)); 186 sscanf(Hdr->mode,"%o",&(member->info.mode)); 187 member->info.user = atoi(Hdr->uid); 188 member->info.group = atoi(Hdr->gid); 189 member->flags = flags; 190 member->data = At; 191 192 return member; 193} 194 195void 196Archive::checkSignature() { 197 // Check the magic string at file's header 198 if (mapfile->size() < 8 || memcmp(base, ARFILE_MAGIC,8)) 199 throw std::string("invalid signature for an archive file"); 200} 201 202// This function loads the entire archive and fully populates its ilist with 203// the members of the archive file. This is typically used in preparation for 204// editing the contents of the archive. 205void 206Archive::loadArchive() { 207 208 // Set up parsing 209 members.clear(); 210 symTab.clear(); 211 const char *At = base; 212 const char *End = base + mapfile->size(); 213 214 checkSignature(); 215 At += 8; // Skip the magic string. 216 217 bool seenSymbolTable = false; 218 bool foundFirstFile = false; 219 while (At < End) { 220 // parse the member header 221 const char* Save = At; 222 ArchiveMember* mbr = parseMemberHeader(At, End); 223 224 // check if this is the foreign symbol table 225 if (mbr->isForeignSymbolTable()) { 226 // We just save this but don't do anything special 227 // with it. It doesn't count as the "first file". 228 foreignST = mbr; 229 At += mbr->getSize(); 230 if ((mbr->getSize() & 1) == 1) 231 At++; 232 } else if (mbr->isStringTable()) { 233 // Simply suck the entire string table into a string 234 // variable. This will be used to get the names of the 235 // members that use the "/ddd" format for their names 236 // (SVR4 style long names). 237 strtab.assign(At,mbr->getSize()); 238 At += mbr->getSize(); 239 if ((mbr->getSize() & 1) == 1) 240 At++; 241 delete mbr; 242 } else if (mbr->isLLVMSymbolTable()) { 243 // This is the LLVM symbol table for the archive. If we've seen it 244 // already, its an error. Otherwise, parse the symbol table and move on. 245 if (seenSymbolTable) 246 throw std::string("invalid archive: multiple symbol tables"); 247 parseSymbolTable(mbr->getData(),mbr->getSize()); 248 seenSymbolTable = true; 249 At += mbr->getSize(); 250 if ((mbr->getSize() & 1) == 1) 251 At++; 252 delete mbr; // We don't need this member in the list of members. 253 } else { 254 // This is just a regular file. If its the first one, save its offset. 255 // Otherwise just push it on the list and move on to the next file. 256 if (!foundFirstFile) { 257 firstFileOffset = Save - base; 258 foundFirstFile = true; 259 } 260 members.push_back(mbr); 261 At += mbr->getSize(); 262 if ((mbr->getSize() & 1) == 1) 263 At++; 264 } 265 } 266} 267 268// Open and completely load the archive file. 269Archive* 270Archive::OpenAndLoad(const sys::Path& file) { 271 272 Archive* result = new Archive(file,true); 273 274 result->loadArchive(); 275 276 return result; 277} 278 279// Get all the bytecode modules from the archive 280bool 281Archive::getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage) { 282 283 for (iterator I=begin(), E=end(); I != E; ++I) { 284 if (I->isBytecode() || I->isCompressedBytecode()) { 285 Module* M = ParseBytecodeBuffer((const unsigned char*)I->getData(), 286 I->getSize(), I->getPath().get(), ErrMessage); 287 if (!M) 288 return true; 289 290 Modules.push_back(M); 291 } 292 } 293 return false; 294} 295 296// Load just the symbol table from the archive file 297void 298Archive::loadSymbolTable() { 299 300 // Set up parsing 301 members.clear(); 302 symTab.clear(); 303 const char *At = base; 304 const char *End = base + mapfile->size(); 305 306 // Make sure we're dealing with an archive 307 checkSignature(); 308 309 At += 8; // Skip signature 310 311 // Parse the first file member header 312 const char* FirstFile = At; 313 ArchiveMember* mbr = parseMemberHeader(At, End); 314 315 if (mbr->isForeignSymbolTable()) { 316 // Skip the foreign symbol table, we don't do anything with it 317 At += mbr->getSize(); 318 if ((mbr->getSize() & 1) == 1) 319 At++; 320 delete mbr; 321 322 // Read the next one 323 FirstFile = At; 324 mbr = parseMemberHeader(At,End); 325 } 326 327 if (mbr->isStringTable()) { 328 // Process the string table entry 329 strtab.assign((const char*)mbr->getData(),mbr->getSize()); 330 At += mbr->getSize(); 331 if ((mbr->getSize() & 1) == 1) 332 At++; 333 delete mbr; 334 // Get the next one 335 FirstFile = At; 336 mbr = parseMemberHeader(At,End); 337 } 338 339 // See if its the symbol table 340 if (mbr->isLLVMSymbolTable()) { 341 parseSymbolTable(mbr->getData(),mbr->getSize()); 342 FirstFile = At + mbr->getSize(); 343 if ((mbr->getSize() & 1) == 1) 344 FirstFile++; 345 } else { 346 // There's no symbol table in the file. We have to rebuild it from scratch 347 // because the intent of this method is to get the symbol table loaded so 348 // it can be searched efficiently. 349 // Add the member to the members list 350 members.push_back(mbr); 351 } 352 353 firstFileOffset = FirstFile - base; 354} 355 356// Open the archive and load just the symbol tables 357Archive* 358Archive::OpenAndLoadSymbols(const sys::Path& file) { 359 Archive* result = new Archive(file,true); 360 361 result->loadSymbolTable(); 362 363 return result; 364} 365 366// Look up one symbol in the symbol table and return a ModuleProvider for the 367// module that defines that symbol. 368ModuleProvider* 369Archive::findModuleDefiningSymbol(const std::string& symbol) { 370 SymTabType::iterator SI = symTab.find(symbol); 371 if (SI == symTab.end()) 372 return 0; 373 374 // The symbol table was previously constructed assuming that the members were 375 // written without the symbol table header. Because VBR encoding is used, the 376 // values could not be adjusted to account for the offset of the symbol table 377 // because that could affect the size of the symbol table due to VBR encoding. 378 // We now have to account for this by adjusting the offset by the size of the 379 // symbol table and its header. 380 unsigned fileOffset = 381 SI->second + // offset in symbol-table-less file 382 firstFileOffset; // add offset to first "real" file in archive 383 384 // See if the module is already loaded 385 ModuleMap::iterator MI = modules.find(fileOffset); 386 if (MI != modules.end()) 387 return MI->second.first; 388 389 // Module hasn't been loaded yet, we need to load it 390 const char* modptr = base + fileOffset; 391 ArchiveMember* mbr = parseMemberHeader(modptr, base + mapfile->size()); 392 393 // Now, load the bytecode module to get the ModuleProvider 394 ModuleProvider* mp = getBytecodeBufferModuleProvider( 395 (const unsigned char*) mbr->getData(), mbr->getSize(), 396 mbr->getPath().get(), 0); 397 398 modules.insert(std::make_pair(fileOffset,std::make_pair(mp,mbr))); 399 400 return mp; 401} 402 403// Look up multiple symbols in the symbol table and return a set of 404// ModuleProviders that define those symbols. 405void 406Archive::findModulesDefiningSymbols(const std::set<std::string>& symbols, 407 std::set<ModuleProvider*>& result) 408{ 409 assert(mapfile && base && "Can't findModulesDefiningSymbols on new archive"); 410 if (symTab.empty()) { 411 // We don't have a symbol table, so we must build it now but lets also 412 // make sure that we populate the modules table as we do this to ensure 413 // that we don't load them twice when findModuleDefiningSymbol is called 414 // below. 415 416 // Get a pointer to the first file 417 const char* At = ((const char*)base) + firstFileOffset; 418 const char* End = ((const char*)base) + mapfile->size(); 419 420 while ( At < End) { 421 // Compute the offset to be put in the symbol table 422 unsigned offset = At - base - firstFileOffset; 423 424 // Parse the file's header 425 ArchiveMember* mbr = parseMemberHeader(At, End); 426 427 // If it contains symbols 428 if (mbr->isBytecode() || mbr->isCompressedBytecode()) { 429 // Get the symbols 430 std::vector<std::string> symbols; 431 ModuleProvider* MP = GetBytecodeSymbols((const unsigned char*)At, 432 mbr->getSize(), mbr->getPath().get(),symbols); 433 434 if (MP) { 435 // Insert the module's symbols into the symbol table 436 for (std::vector<std::string>::iterator I = symbols.begin(), 437 E=symbols.end(); I != E; ++I ) { 438 symTab.insert(std::make_pair(*I,offset)); 439 } 440 // Insert the ModuleProvider and the ArchiveMember into the table of 441 // modules. 442 modules.insert(std::make_pair(offset,std::make_pair(MP,mbr))); 443 } else { 444 throw std::string("Can't parse bytecode member: ") + 445 mbr->getPath().get(); 446 } 447 } 448 449 // Go to the next file location 450 At += mbr->getSize(); 451 if ((mbr->getSize() & 1) == 1) 452 At++; 453 } 454 } 455 456 // At this point we have a valid symbol table (one way or another) so we 457 // just use it to quickly find the symbols requested. 458 459 for (std::set<std::string>::const_iterator I=symbols.begin(), 460 E=symbols.end(); I != E; ++I) { 461 ModuleProvider* mp = findModuleDefiningSymbol(*I); 462 if (mp) { 463 result.insert(mp); 464 } 465 } 466} 467