ArchiveReader.cpp revision 4a980d181332cd42228d7432cb5fb6bbb1f0b071
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 + 1; // terminated by \n 101 flags |= ArchiveMember::HasLongFilenameFlag; 102 } else 103 throw std::string("invalid long filename"); 104 } else if (Hdr->name[1] == '_' && 105 (0==memcmp(Hdr->name,ARFILE_LLVM_SYMTAB_NAME,16))) { 106 // The member is using a long file name (>15 chars) format. 107 // This format is standard for 4.4BSD and Mac OSX operating 108 // systems. LLVM uses it similarly. In this format, the 109 // remainder of the name field (after #1/) specifies the 110 // length of the file name which occupy the first bytes of 111 // the member's data. The pathname already has the #1/ stripped. 112 pathname.assign(ARFILE_LLVM_SYMTAB_NAME); 113 flags |= ArchiveMember::LLVMSymbolTableFlag; 114 } 115 break; 116 case '/': 117 if (Hdr->name[1]== '/') { 118 if (0==memcmp(Hdr->name,ARFILE_STRTAB_NAME,16)) { 119 pathname.assign(ARFILE_STRTAB_NAME); 120 flags |= ArchiveMember::StringTableFlag; 121 } else { 122 throw std::string("invalid string table name"); 123 } 124 } else if (Hdr->name[1] == ' ') { 125 if (0==memcmp(Hdr->name,ARFILE_SYMTAB_NAME,16)) { 126 pathname.assign(ARFILE_SYMTAB_NAME); 127 flags |= ArchiveMember::ForeignSymbolTableFlag; 128 } else { 129 throw std::string("invalid foreign symbol table name"); 130 } 131 } else if (isdigit(Hdr->name[1])) { 132 unsigned index = atoi(&Hdr->name[1]); 133 if (index < strtab.length()) { 134 const char* namep = strtab.c_str() + index; 135 const char* endp = strtab.c_str() + strtab.length(); 136 const char* p = namep; 137 const char* last_p = p; 138 while (p < endp) { 139 if (*p == '\n' && *last_p == '/') { 140 pathname.assign(namep,last_p-namep); 141 flags |= ArchiveMember::HasLongFilenameFlag; 142 break; 143 } 144 last_p = p; 145 p++; 146 } 147 if (p >= endp) 148 throw std::string("missing name termiantor in string table"); 149 } else { 150 throw std::string("name index beyond string table"); 151 } 152 } 153 break; 154 155 default: 156 char* slash = (char*) memchr(Hdr->name,'/',16); 157 if (slash == 0) 158 throw std::string("missing name terminator"); 159 pathname.assign(Hdr->name,slash-Hdr->name); 160 break; 161 } 162 163 // Determine if this is a bytecode file 164 switch (sys::IdentifyFileType(At,4)) { 165 case sys::BytecodeFileType: 166 flags |= ArchiveMember::BytecodeFlag; 167 break; 168 case sys::CompressedBytecodeFileType: 169 flags |= ArchiveMember::CompressedBytecodeFlag; 170 flags &= ~ArchiveMember::CompressedFlag; 171 break; 172 default: 173 flags &= ~(ArchiveMember::BytecodeFlag| 174 ArchiveMember::CompressedBytecodeFlag); 175 break; 176 } 177 178 // Fill in fields of the ArchiveMember 179 member->next = 0; 180 member->prev = 0; 181 member->parent = this; 182 member->path.setFile(pathname); 183 member->info.fileSize = MemberSize; 184 member->info.modTime.fromEpochTime(atoi(Hdr->date)); 185 sscanf(Hdr->mode,"%o",&(member->info.mode)); 186 member->info.user = atoi(Hdr->uid); 187 member->info.group = atoi(Hdr->gid); 188 member->flags = flags; 189 member->data = At; 190 191 return member; 192} 193 194void 195Archive::checkSignature() { 196 // Check the magic string at file's header 197 if (mapfile->size() < 8 || memcmp(base, ARFILE_MAGIC,8)) 198 throw std::string("invalid signature for an archive file"); 199} 200 201// This function loads the entire archive and fully populates its ilist with 202// the members of the archive file. This is typically used in preparation for 203// editing the contents of the archive. 204void 205Archive::loadArchive() { 206 207 // Set up parsing 208 members.clear(); 209 symTab.clear(); 210 const char *At = base; 211 const char *End = base + mapfile->size(); 212 213 checkSignature(); 214 At += 8; // Skip the magic string. 215 216 bool seenSymbolTable = false; 217 bool foundFirstFile = false; 218 while (At < End) { 219 // parse the member header 220 const char* Save = At; 221 ArchiveMember* mbr = parseMemberHeader(At, End); 222 223 // check if this is the foreign symbol table 224 if (mbr->isForeignSymbolTable()) { 225 // We just save this but don't do anything special 226 // with it. It doesn't count as the "first file". 227 foreignST = mbr; 228 At += mbr->getSize(); 229 if ((mbr->getSize() & 1) == 1) 230 At++; 231 } else if (mbr->isStringTable()) { 232 // Simply suck the entire string table into a string 233 // variable. This will be used to get the names of the 234 // members that use the "/ddd" format for their names 235 // (SVR4 style long names). 236 strtab.assign(At,mbr->getSize()); 237 At += mbr->getSize(); 238 if ((mbr->getSize() & 1) == 1) 239 At++; 240 delete mbr; 241 } else if (mbr->isLLVMSymbolTable()) { 242 // This is the LLVM symbol table for the archive. If we've seen it 243 // already, its an error. Otherwise, parse the symbol table and move on. 244 if (seenSymbolTable) 245 throw std::string("invalid archive: multiple symbol tables"); 246 parseSymbolTable(mbr->getData(),mbr->getSize()); 247 seenSymbolTable = true; 248 At += mbr->getSize(); 249 if ((mbr->getSize() & 1) == 1) 250 At++; 251 delete mbr; // We don't need this member in the list of members. 252 } else { 253 // This is just a regular file. If its the first one, save its offset. 254 // Otherwise just push it on the list and move on to the next file. 255 if (!foundFirstFile) { 256 firstFileOffset = Save - base; 257 foundFirstFile = true; 258 } 259 members.push_back(mbr); 260 At += mbr->getSize(); 261 if ((mbr->getSize() & 1) == 1) 262 At++; 263 } 264 } 265} 266 267// Open and completely load the archive file. 268Archive* 269Archive::OpenAndLoad(const sys::Path& file) { 270 271 Archive* result = new Archive(file,true); 272 273 result->loadArchive(); 274 275 return result; 276} 277 278// Get all the bytecode modules from the archive 279bool 280Archive::getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage) { 281 282 for (iterator I=begin(), E=end(); I != E; ++I) { 283 if (I->isBytecode() || I->isCompressedBytecode()) { 284 Module* M = ParseBytecodeBuffer((const unsigned char*)I->getData(), 285 I->getSize(), I->getPath().get(), ErrMessage); 286 if (!M) 287 return true; 288 289 Modules.push_back(M); 290 } 291 } 292 return false; 293} 294 295// Load just the symbol table from the archive file 296void 297Archive::loadSymbolTable() { 298 299 // Set up parsing 300 members.clear(); 301 symTab.clear(); 302 const char *At = base; 303 const char *End = base + mapfile->size(); 304 305 // Make sure we're dealing with an archive 306 checkSignature(); 307 308 At += 8; // Skip signature 309 310 // Parse the first file member header 311 const char* FirstFile = At; 312 ArchiveMember* mbr = parseMemberHeader(At, End); 313 314 if (mbr->isForeignSymbolTable()) { 315 // Skip the foreign symbol table, we don't do anything with it 316 At += mbr->getSize(); 317 if ((mbr->getSize() & 1) == 1) 318 At++; 319 delete mbr; 320 321 // Read the next one 322 FirstFile = At; 323 mbr = parseMemberHeader(At,End); 324 } 325 326 if (mbr->isStringTable()) { 327 // Process the string table entry 328 strtab.assign((const char*)mbr->getData(),mbr->getSize()); 329 At += mbr->getSize(); 330 if ((mbr->getSize() & 1) == 1) 331 At++; 332 delete mbr; 333 // Get the next one 334 FirstFile = At; 335 mbr = parseMemberHeader(At,End); 336 } 337 338 // See if its the symbol table 339 if (mbr->isLLVMSymbolTable()) { 340 parseSymbolTable(mbr->getData(),mbr->getSize()); 341 FirstFile = At + mbr->getSize(); 342 if ((mbr->getSize() & 1) == 1) 343 FirstFile++; 344 } else { 345 // There's no symbol table in the file. We have to rebuild it from scratch 346 // because the intent of this method is to get the symbol table loaded so 347 // it can be searched efficiently. 348 // Add the member to the members list 349 members.push_back(mbr); 350 } 351 352 firstFileOffset = FirstFile - base; 353} 354 355// Open the archive and load just the symbol tables 356Archive* 357Archive::OpenAndLoadSymbols(const sys::Path& file) { 358 Archive* result = new Archive(file,true); 359 360 result->loadSymbolTable(); 361 362 return result; 363} 364 365// Look up one symbol in the symbol table and return a ModuleProvider for the 366// module that defines that symbol. 367ModuleProvider* 368Archive::findModuleDefiningSymbol(const std::string& symbol) { 369 SymTabType::iterator SI = symTab.find(symbol); 370 if (SI == symTab.end()) 371 return 0; 372 373 // The symbol table was previously constructed assuming that the members were 374 // written without the symbol table header. Because VBR encoding is used, the 375 // values could not be adjusted to account for the offset of the symbol table 376 // because that could affect the size of the symbol table due to VBR encoding. 377 // We now have to account for this by adjusting the offset by the size of the 378 // symbol table and its header. 379 unsigned fileOffset = 380 SI->second + // offset in symbol-table-less file 381 firstFileOffset; // add offset to first "real" file in archive 382 383 // See if the module is already loaded 384 ModuleMap::iterator MI = modules.find(fileOffset); 385 if (MI != modules.end()) 386 return MI->second.first; 387 388 // Module hasn't been loaded yet, we need to load it 389 const char* modptr = base + fileOffset; 390 ArchiveMember* mbr = parseMemberHeader(modptr, base + mapfile->size()); 391 392 // Now, load the bytecode module to get the ModuleProvider 393 ModuleProvider* mp = getBytecodeBufferModuleProvider( 394 (const unsigned char*) mbr->getData(), mbr->getSize(), 395 mbr->getPath().get(), 0); 396 397 modules.insert(std::make_pair(fileOffset,std::make_pair(mp,mbr))); 398 399 return mp; 400} 401 402// Look up multiple symbols in the symbol table and return a set of 403// ModuleProviders that define those symbols. 404void 405Archive::findModulesDefiningSymbols(const std::set<std::string>& symbols, 406 std::set<ModuleProvider*>& result) 407{ 408 assert(mapfile && base && "Can't findModulesDefiningSymbols on new archive"); 409 if (symTab.empty()) { 410 // We don't have a symbol table, so we must build it now but lets also 411 // make sure that we populate the modules table as we do this to ensure 412 // that we don't load them twice when findModuleDefiningSymbol is called 413 // below. 414 415 // Get a pointer to the first file 416 const char* At = ((const char*)base) + firstFileOffset; 417 const char* End = ((const char*)base) + mapfile->size(); 418 419 while ( At < End) { 420 // Compute the offset to be put in the symbol table 421 unsigned offset = At - base - firstFileOffset; 422 423 // Parse the file's header 424 ArchiveMember* mbr = parseMemberHeader(At, End); 425 426 // If it contains symbols 427 if (mbr->isBytecode() || mbr->isCompressedBytecode()) { 428 // Get the symbols 429 std::vector<std::string> symbols; 430 ModuleProvider* MP = GetBytecodeSymbols((const unsigned char*)At, 431 mbr->getSize(), mbr->getPath().get(),symbols); 432 433 if (MP) { 434 // Insert the module's symbols into the symbol table 435 for (std::vector<std::string>::iterator I = symbols.begin(), 436 E=symbols.end(); I != E; ++I ) { 437 symTab.insert(std::make_pair(*I,offset)); 438 } 439 // Insert the ModuleProvider and the ArchiveMember into the table of 440 // modules. 441 modules.insert(std::make_pair(offset,std::make_pair(MP,mbr))); 442 } else { 443 throw std::string("Can't parse bytecode member: ") + 444 mbr->getPath().get(); 445 } 446 } 447 448 // Go to the next file location 449 At += mbr->getSize(); 450 if ((mbr->getSize() & 1) == 1) 451 At++; 452 } 453 } 454 455 // At this point we have a valid symbol table (one way or another) so we 456 // just use it to quickly find the symbols requested. 457 458 for (std::set<std::string>::const_iterator I=symbols.begin(), 459 E=symbols.end(); I != E; ++I) { 460 ModuleProvider* mp = findModuleDefiningSymbol(*I); 461 if (mp) { 462 result.insert(mp); 463 } 464 } 465} 466