1 //===--- FileManager.cpp - File System Probing and Caching ----------------===// 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 implements the FileManager interface. 11 // 12 //===----------------------------------------------------------------------===// 13 // 14 // TODO: This should index all interesting directories with dirent calls. 15 // getdirentries ? 16 // opendir/readdir_r/closedir ? 17 // 18 //===----------------------------------------------------------------------===// 19 20 #include "clang/Basic/FileManager.h" 21 #include "clang/Basic/FileSystemStatCache.h" 22 #include "llvm/ADT/SmallString.h" 23 #include "llvm/Support/FileSystem.h" 24 #include "llvm/Support/MemoryBuffer.h" 25 #include "llvm/Support/raw_ostream.h" 26 #include "llvm/Support/Path.h" 27 #include "llvm/Support/system_error.h" 28 #include "llvm/Config/llvm-config.h" 29 #include <map> 30 #include <set> 31 #include <string> 32 33 // FIXME: This is terrible, we need this for ::close. 34 #if !defined(_MSC_VER) && !defined(__MINGW32__) 35 #include <unistd.h> 36 #include <sys/uio.h> 37 #else 38 #include <io.h> 39 #endif 40 using namespace clang; 41 42 // FIXME: Enhance libsystem to support inode and other fields. 43 #include <sys/stat.h> 44 45 /// NON_EXISTENT_DIR - A special value distinct from null that is used to 46 /// represent a dir name that doesn't exist on the disk. 47 #define NON_EXISTENT_DIR reinterpret_cast<DirectoryEntry*>((intptr_t)-1) 48 49 /// NON_EXISTENT_FILE - A special value distinct from null that is used to 50 /// represent a filename that doesn't exist on the disk. 51 #define NON_EXISTENT_FILE reinterpret_cast<FileEntry*>((intptr_t)-1) 52 53 54 FileEntry::~FileEntry() { 55 // If this FileEntry owns an open file descriptor that never got used, close 56 // it. 57 if (FD != -1) ::close(FD); 58 } 59 60 //===----------------------------------------------------------------------===// 61 // Windows. 62 //===----------------------------------------------------------------------===// 63 64 #ifdef LLVM_ON_WIN32 65 66 namespace { 67 static std::string GetFullPath(const char *relPath) { 68 char *absPathStrPtr = _fullpath(NULL, relPath, 0); 69 assert(absPathStrPtr && "_fullpath() returned NULL!"); 70 71 std::string absPath(absPathStrPtr); 72 73 free(absPathStrPtr); 74 return absPath; 75 } 76 } 77 78 class FileManager::UniqueDirContainer { 79 /// UniqueDirs - Cache from full path to existing directories/files. 80 /// 81 llvm::StringMap<DirectoryEntry> UniqueDirs; 82 83 public: 84 /// getDirectory - Return an existing DirectoryEntry with the given 85 /// name if there is already one; otherwise create and return a 86 /// default-constructed DirectoryEntry. 87 DirectoryEntry &getDirectory(const char *Name, 88 const struct stat & /*StatBuf*/) { 89 std::string FullPath(GetFullPath(Name)); 90 return UniqueDirs.GetOrCreateValue(FullPath).getValue(); 91 } 92 93 size_t size() const { return UniqueDirs.size(); } 94 }; 95 96 class FileManager::UniqueFileContainer { 97 /// UniqueFiles - Cache from full path to existing directories/files. 98 /// 99 llvm::StringMap<FileEntry, llvm::BumpPtrAllocator> UniqueFiles; 100 101 public: 102 /// getFile - Return an existing FileEntry with the given name if 103 /// there is already one; otherwise create and return a 104 /// default-constructed FileEntry. 105 FileEntry &getFile(const char *Name, const struct stat & /*StatBuf*/) { 106 std::string FullPath(GetFullPath(Name)); 107 108 // Lowercase string because Windows filesystem is case insensitive. 109 FullPath = StringRef(FullPath).lower(); 110 return UniqueFiles.GetOrCreateValue(FullPath).getValue(); 111 } 112 113 size_t size() const { return UniqueFiles.size(); } 114 }; 115 116 //===----------------------------------------------------------------------===// 117 // Unix-like Systems. 118 //===----------------------------------------------------------------------===// 119 120 #else 121 122 class FileManager::UniqueDirContainer { 123 /// UniqueDirs - Cache from ID's to existing directories/files. 124 std::map<std::pair<dev_t, ino_t>, DirectoryEntry> UniqueDirs; 125 126 public: 127 /// getDirectory - Return an existing DirectoryEntry with the given 128 /// ID's if there is already one; otherwise create and return a 129 /// default-constructed DirectoryEntry. 130 DirectoryEntry &getDirectory(const char * /*Name*/, 131 const struct stat &StatBuf) { 132 return UniqueDirs[std::make_pair(StatBuf.st_dev, StatBuf.st_ino)]; 133 } 134 135 size_t size() const { return UniqueDirs.size(); } 136 }; 137 138 class FileManager::UniqueFileContainer { 139 /// UniqueFiles - Cache from ID's to existing directories/files. 140 std::set<FileEntry> UniqueFiles; 141 142 public: 143 /// getFile - Return an existing FileEntry with the given ID's if 144 /// there is already one; otherwise create and return a 145 /// default-constructed FileEntry. 146 FileEntry &getFile(const char * /*Name*/, const struct stat &StatBuf) { 147 return 148 const_cast<FileEntry&>( 149 *UniqueFiles.insert(FileEntry(StatBuf.st_dev, 150 StatBuf.st_ino, 151 StatBuf.st_mode)).first); 152 } 153 154 size_t size() const { return UniqueFiles.size(); } 155 }; 156 157 #endif 158 159 //===----------------------------------------------------------------------===// 160 // Common logic. 161 //===----------------------------------------------------------------------===// 162 163 FileManager::FileManager(const FileSystemOptions &FSO) 164 : FileSystemOpts(FSO), 165 UniqueRealDirs(*new UniqueDirContainer()), 166 UniqueRealFiles(*new UniqueFileContainer()), 167 SeenDirEntries(64), SeenFileEntries(64), NextFileUID(0) { 168 NumDirLookups = NumFileLookups = 0; 169 NumDirCacheMisses = NumFileCacheMisses = 0; 170 } 171 172 FileManager::~FileManager() { 173 delete &UniqueRealDirs; 174 delete &UniqueRealFiles; 175 for (unsigned i = 0, e = VirtualFileEntries.size(); i != e; ++i) 176 delete VirtualFileEntries[i]; 177 for (unsigned i = 0, e = VirtualDirectoryEntries.size(); i != e; ++i) 178 delete VirtualDirectoryEntries[i]; 179 } 180 181 void FileManager::addStatCache(FileSystemStatCache *statCache, 182 bool AtBeginning) { 183 assert(statCache && "No stat cache provided?"); 184 if (AtBeginning || StatCache.get() == 0) { 185 statCache->setNextStatCache(StatCache.take()); 186 StatCache.reset(statCache); 187 return; 188 } 189 190 FileSystemStatCache *LastCache = StatCache.get(); 191 while (LastCache->getNextStatCache()) 192 LastCache = LastCache->getNextStatCache(); 193 194 LastCache->setNextStatCache(statCache); 195 } 196 197 void FileManager::removeStatCache(FileSystemStatCache *statCache) { 198 if (!statCache) 199 return; 200 201 if (StatCache.get() == statCache) { 202 // This is the first stat cache. 203 StatCache.reset(StatCache->takeNextStatCache()); 204 return; 205 } 206 207 // Find the stat cache in the list. 208 FileSystemStatCache *PrevCache = StatCache.get(); 209 while (PrevCache && PrevCache->getNextStatCache() != statCache) 210 PrevCache = PrevCache->getNextStatCache(); 211 212 assert(PrevCache && "Stat cache not found for removal"); 213 PrevCache->setNextStatCache(statCache->getNextStatCache()); 214 } 215 216 /// \brief Retrieve the directory that the given file name resides in. 217 /// Filename can point to either a real file or a virtual file. 218 static const DirectoryEntry *getDirectoryFromFile(FileManager &FileMgr, 219 StringRef Filename, 220 bool CacheFailure) { 221 if (Filename.empty()) 222 return NULL; 223 224 if (llvm::sys::path::is_separator(Filename[Filename.size() - 1])) 225 return NULL; // If Filename is a directory. 226 227 StringRef DirName = llvm::sys::path::parent_path(Filename); 228 // Use the current directory if file has no path component. 229 if (DirName.empty()) 230 DirName = "."; 231 232 return FileMgr.getDirectory(DirName, CacheFailure); 233 } 234 235 /// Add all ancestors of the given path (pointing to either a file or 236 /// a directory) as virtual directories. 237 void FileManager::addAncestorsAsVirtualDirs(StringRef Path) { 238 StringRef DirName = llvm::sys::path::parent_path(Path); 239 if (DirName.empty()) 240 return; 241 242 llvm::StringMapEntry<DirectoryEntry *> &NamedDirEnt = 243 SeenDirEntries.GetOrCreateValue(DirName); 244 245 // When caching a virtual directory, we always cache its ancestors 246 // at the same time. Therefore, if DirName is already in the cache, 247 // we don't need to recurse as its ancestors must also already be in 248 // the cache. 249 if (NamedDirEnt.getValue()) 250 return; 251 252 // Add the virtual directory to the cache. 253 DirectoryEntry *UDE = new DirectoryEntry; 254 UDE->Name = NamedDirEnt.getKeyData(); 255 NamedDirEnt.setValue(UDE); 256 VirtualDirectoryEntries.push_back(UDE); 257 258 // Recursively add the other ancestors. 259 addAncestorsAsVirtualDirs(DirName); 260 } 261 262 /// getDirectory - Lookup, cache, and verify the specified directory 263 /// (real or virtual). This returns NULL if the directory doesn't 264 /// exist. 265 /// 266 const DirectoryEntry *FileManager::getDirectory(StringRef DirName, 267 bool CacheFailure) { 268 // stat doesn't like trailing separators. 269 // At least, on Win32 MSVCRT, stat() cannot strip trailing '/'. 270 // (though it can strip '\\') 271 if (DirName.size() > 1 && llvm::sys::path::is_separator(DirName.back())) 272 DirName = DirName.substr(0, DirName.size()-1); 273 274 ++NumDirLookups; 275 llvm::StringMapEntry<DirectoryEntry *> &NamedDirEnt = 276 SeenDirEntries.GetOrCreateValue(DirName); 277 278 // See if there was already an entry in the map. Note that the map 279 // contains both virtual and real directories. 280 if (NamedDirEnt.getValue()) 281 return NamedDirEnt.getValue() == NON_EXISTENT_DIR 282 ? 0 : NamedDirEnt.getValue(); 283 284 ++NumDirCacheMisses; 285 286 // By default, initialize it to invalid. 287 NamedDirEnt.setValue(NON_EXISTENT_DIR); 288 289 // Get the null-terminated directory name as stored as the key of the 290 // SeenDirEntries map. 291 const char *InterndDirName = NamedDirEnt.getKeyData(); 292 293 // Check to see if the directory exists. 294 struct stat StatBuf; 295 if (getStatValue(InterndDirName, StatBuf, 0/*directory lookup*/)) { 296 // There's no real directory at the given path. 297 if (!CacheFailure) 298 SeenDirEntries.erase(DirName); 299 return 0; 300 } 301 302 // It exists. See if we have already opened a directory with the 303 // same inode (this occurs on Unix-like systems when one dir is 304 // symlinked to another, for example) or the same path (on 305 // Windows). 306 DirectoryEntry &UDE = UniqueRealDirs.getDirectory(InterndDirName, StatBuf); 307 308 NamedDirEnt.setValue(&UDE); 309 if (!UDE.getName()) { 310 // We don't have this directory yet, add it. We use the string 311 // key from the SeenDirEntries map as the string. 312 UDE.Name = InterndDirName; 313 } 314 315 return &UDE; 316 } 317 318 /// getFile - Lookup, cache, and verify the specified file (real or 319 /// virtual). This returns NULL if the file doesn't exist. 320 /// 321 const FileEntry *FileManager::getFile(StringRef Filename, bool openFile, 322 bool CacheFailure) { 323 ++NumFileLookups; 324 325 // See if there is already an entry in the map. 326 llvm::StringMapEntry<FileEntry *> &NamedFileEnt = 327 SeenFileEntries.GetOrCreateValue(Filename); 328 329 // See if there is already an entry in the map. 330 if (NamedFileEnt.getValue()) 331 return NamedFileEnt.getValue() == NON_EXISTENT_FILE 332 ? 0 : NamedFileEnt.getValue(); 333 334 ++NumFileCacheMisses; 335 336 // By default, initialize it to invalid. 337 NamedFileEnt.setValue(NON_EXISTENT_FILE); 338 339 // Get the null-terminated file name as stored as the key of the 340 // SeenFileEntries map. 341 const char *InterndFileName = NamedFileEnt.getKeyData(); 342 343 // Look up the directory for the file. When looking up something like 344 // sys/foo.h we'll discover all of the search directories that have a 'sys' 345 // subdirectory. This will let us avoid having to waste time on known-to-fail 346 // searches when we go to find sys/bar.h, because all the search directories 347 // without a 'sys' subdir will get a cached failure result. 348 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename, 349 CacheFailure); 350 if (DirInfo == 0) { // Directory doesn't exist, file can't exist. 351 if (!CacheFailure) 352 SeenFileEntries.erase(Filename); 353 354 return 0; 355 } 356 357 // FIXME: Use the directory info to prune this, before doing the stat syscall. 358 // FIXME: This will reduce the # syscalls. 359 360 // Nope, there isn't. Check to see if the file exists. 361 int FileDescriptor = -1; 362 struct stat StatBuf; 363 if (getStatValue(InterndFileName, StatBuf, &FileDescriptor)) { 364 // There's no real file at the given path. 365 if (!CacheFailure) 366 SeenFileEntries.erase(Filename); 367 368 return 0; 369 } 370 371 if (FileDescriptor != -1 && !openFile) { 372 close(FileDescriptor); 373 FileDescriptor = -1; 374 } 375 376 // It exists. See if we have already opened a file with the same inode. 377 // This occurs when one dir is symlinked to another, for example. 378 FileEntry &UFE = UniqueRealFiles.getFile(InterndFileName, StatBuf); 379 380 NamedFileEnt.setValue(&UFE); 381 if (UFE.getName()) { // Already have an entry with this inode, return it. 382 // If the stat process opened the file, close it to avoid a FD leak. 383 if (FileDescriptor != -1) 384 close(FileDescriptor); 385 386 return &UFE; 387 } 388 389 // Otherwise, we don't have this directory yet, add it. 390 // FIXME: Change the name to be a char* that points back to the 391 // 'SeenFileEntries' key. 392 UFE.Name = InterndFileName; 393 UFE.Size = StatBuf.st_size; 394 UFE.ModTime = StatBuf.st_mtime; 395 UFE.Dir = DirInfo; 396 UFE.UID = NextFileUID++; 397 UFE.FD = FileDescriptor; 398 return &UFE; 399 } 400 401 const FileEntry * 402 FileManager::getVirtualFile(StringRef Filename, off_t Size, 403 time_t ModificationTime) { 404 ++NumFileLookups; 405 406 // See if there is already an entry in the map. 407 llvm::StringMapEntry<FileEntry *> &NamedFileEnt = 408 SeenFileEntries.GetOrCreateValue(Filename); 409 410 // See if there is already an entry in the map. 411 if (NamedFileEnt.getValue() && NamedFileEnt.getValue() != NON_EXISTENT_FILE) 412 return NamedFileEnt.getValue(); 413 414 ++NumFileCacheMisses; 415 416 // By default, initialize it to invalid. 417 NamedFileEnt.setValue(NON_EXISTENT_FILE); 418 419 addAncestorsAsVirtualDirs(Filename); 420 FileEntry *UFE = 0; 421 422 // Now that all ancestors of Filename are in the cache, the 423 // following call is guaranteed to find the DirectoryEntry from the 424 // cache. 425 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename, 426 /*CacheFailure=*/true); 427 assert(DirInfo && 428 "The directory of a virtual file should already be in the cache."); 429 430 // Check to see if the file exists. If so, drop the virtual file 431 int FileDescriptor = -1; 432 struct stat StatBuf; 433 const char *InterndFileName = NamedFileEnt.getKeyData(); 434 if (getStatValue(InterndFileName, StatBuf, &FileDescriptor) == 0) { 435 // If the stat process opened the file, close it to avoid a FD leak. 436 if (FileDescriptor != -1) 437 close(FileDescriptor); 438 439 StatBuf.st_size = Size; 440 StatBuf.st_mtime = ModificationTime; 441 UFE = &UniqueRealFiles.getFile(InterndFileName, StatBuf); 442 443 NamedFileEnt.setValue(UFE); 444 445 // If we had already opened this file, close it now so we don't 446 // leak the descriptor. We're not going to use the file 447 // descriptor anyway, since this is a virtual file. 448 if (UFE->FD != -1) { 449 close(UFE->FD); 450 UFE->FD = -1; 451 } 452 453 // If we already have an entry with this inode, return it. 454 if (UFE->getName()) 455 return UFE; 456 } 457 458 if (!UFE) { 459 UFE = new FileEntry(); 460 VirtualFileEntries.push_back(UFE); 461 NamedFileEnt.setValue(UFE); 462 } 463 464 UFE->Name = InterndFileName; 465 UFE->Size = Size; 466 UFE->ModTime = ModificationTime; 467 UFE->Dir = DirInfo; 468 UFE->UID = NextFileUID++; 469 UFE->FD = -1; 470 return UFE; 471 } 472 473 void FileManager::FixupRelativePath(SmallVectorImpl<char> &path) const { 474 StringRef pathRef(path.data(), path.size()); 475 476 if (FileSystemOpts.WorkingDir.empty() 477 || llvm::sys::path::is_absolute(pathRef)) 478 return; 479 480 SmallString<128> NewPath(FileSystemOpts.WorkingDir); 481 llvm::sys::path::append(NewPath, pathRef); 482 path = NewPath; 483 } 484 485 llvm::MemoryBuffer *FileManager:: 486 getBufferForFile(const FileEntry *Entry, std::string *ErrorStr) { 487 OwningPtr<llvm::MemoryBuffer> Result; 488 llvm::error_code ec; 489 490 const char *Filename = Entry->getName(); 491 // If the file is already open, use the open file descriptor. 492 if (Entry->FD != -1) { 493 ec = llvm::MemoryBuffer::getOpenFile(Entry->FD, Filename, Result, 494 Entry->getSize()); 495 if (ErrorStr) 496 *ErrorStr = ec.message(); 497 498 close(Entry->FD); 499 Entry->FD = -1; 500 return Result.take(); 501 } 502 503 // Otherwise, open the file. 504 505 if (FileSystemOpts.WorkingDir.empty()) { 506 ec = llvm::MemoryBuffer::getFile(Filename, Result, Entry->getSize()); 507 if (ec && ErrorStr) 508 *ErrorStr = ec.message(); 509 return Result.take(); 510 } 511 512 SmallString<128> FilePath(Entry->getName()); 513 FixupRelativePath(FilePath); 514 ec = llvm::MemoryBuffer::getFile(FilePath.str(), Result, Entry->getSize()); 515 if (ec && ErrorStr) 516 *ErrorStr = ec.message(); 517 return Result.take(); 518 } 519 520 llvm::MemoryBuffer *FileManager:: 521 getBufferForFile(StringRef Filename, std::string *ErrorStr) { 522 OwningPtr<llvm::MemoryBuffer> Result; 523 llvm::error_code ec; 524 if (FileSystemOpts.WorkingDir.empty()) { 525 ec = llvm::MemoryBuffer::getFile(Filename, Result); 526 if (ec && ErrorStr) 527 *ErrorStr = ec.message(); 528 return Result.take(); 529 } 530 531 SmallString<128> FilePath(Filename); 532 FixupRelativePath(FilePath); 533 ec = llvm::MemoryBuffer::getFile(FilePath.c_str(), Result); 534 if (ec && ErrorStr) 535 *ErrorStr = ec.message(); 536 return Result.take(); 537 } 538 539 /// getStatValue - Get the 'stat' information for the specified path, 540 /// using the cache to accelerate it if possible. This returns true 541 /// if the path points to a virtual file or does not exist, or returns 542 /// false if it's an existent real file. If FileDescriptor is NULL, 543 /// do directory look-up instead of file look-up. 544 bool FileManager::getStatValue(const char *Path, struct stat &StatBuf, 545 int *FileDescriptor) { 546 // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be 547 // absolute! 548 if (FileSystemOpts.WorkingDir.empty()) 549 return FileSystemStatCache::get(Path, StatBuf, FileDescriptor, 550 StatCache.get()); 551 552 SmallString<128> FilePath(Path); 553 FixupRelativePath(FilePath); 554 555 return FileSystemStatCache::get(FilePath.c_str(), StatBuf, FileDescriptor, 556 StatCache.get()); 557 } 558 559 bool FileManager::getNoncachedStatValue(StringRef Path, 560 struct stat &StatBuf) { 561 SmallString<128> FilePath(Path); 562 FixupRelativePath(FilePath); 563 564 return ::stat(FilePath.c_str(), &StatBuf) != 0; 565 } 566 567 void FileManager::GetUniqueIDMapping( 568 SmallVectorImpl<const FileEntry *> &UIDToFiles) const { 569 UIDToFiles.clear(); 570 UIDToFiles.resize(NextFileUID); 571 572 // Map file entries 573 for (llvm::StringMap<FileEntry*, llvm::BumpPtrAllocator>::const_iterator 574 FE = SeenFileEntries.begin(), FEEnd = SeenFileEntries.end(); 575 FE != FEEnd; ++FE) 576 if (FE->getValue() && FE->getValue() != NON_EXISTENT_FILE) 577 UIDToFiles[FE->getValue()->getUID()] = FE->getValue(); 578 579 // Map virtual file entries 580 for (SmallVector<FileEntry*, 4>::const_iterator 581 VFE = VirtualFileEntries.begin(), VFEEnd = VirtualFileEntries.end(); 582 VFE != VFEEnd; ++VFE) 583 if (*VFE && *VFE != NON_EXISTENT_FILE) 584 UIDToFiles[(*VFE)->getUID()] = *VFE; 585 } 586 587 588 void FileManager::PrintStats() const { 589 llvm::errs() << "\n*** File Manager Stats:\n"; 590 llvm::errs() << UniqueRealFiles.size() << " real files found, " 591 << UniqueRealDirs.size() << " real dirs found.\n"; 592 llvm::errs() << VirtualFileEntries.size() << " virtual files found, " 593 << VirtualDirectoryEntries.size() << " virtual dirs found.\n"; 594 llvm::errs() << NumDirLookups << " dir lookups, " 595 << NumDirCacheMisses << " dir cache misses.\n"; 596 llvm::errs() << NumFileLookups << " file lookups, " 597 << NumFileCacheMisses << " file cache misses.\n"; 598 599 //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups; 600 } 601