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/Config/llvm-config.h" 24 #include "llvm/Support/FileSystem.h" 25 #include "llvm/Support/MemoryBuffer.h" 26 #include "llvm/Support/Path.h" 27 #include "llvm/Support/raw_ostream.h" 28 #include "llvm/Support/system_error.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 #ifndef S_ISFIFO 40 #define S_ISFIFO(x) (0) 41 #endif 42 #endif 43 #if defined(LLVM_ON_UNIX) 44 #include <limits.h> 45 #endif 46 using namespace clang; 47 48 // FIXME: Enhance libsystem to support inode and other fields. 49 #include <sys/stat.h> 50 51 /// NON_EXISTENT_DIR - A special value distinct from null that is used to 52 /// represent a dir name that doesn't exist on the disk. 53 #define NON_EXISTENT_DIR reinterpret_cast<DirectoryEntry*>((intptr_t)-1) 54 55 /// NON_EXISTENT_FILE - A special value distinct from null that is used to 56 /// represent a filename that doesn't exist on the disk. 57 #define NON_EXISTENT_FILE reinterpret_cast<FileEntry*>((intptr_t)-1) 58 59 60 FileEntry::~FileEntry() { 61 // If this FileEntry owns an open file descriptor that never got used, close 62 // it. 63 if (FD != -1) ::close(FD); 64 } 65 66 class FileManager::UniqueDirContainer { 67 /// UniqueDirs - Cache from ID's to existing directories/files. 68 std::map<llvm::sys::fs::UniqueID, DirectoryEntry> UniqueDirs; 69 70 public: 71 /// getDirectory - Return an existing DirectoryEntry with the given 72 /// ID's if there is already one; otherwise create and return a 73 /// default-constructed DirectoryEntry. 74 DirectoryEntry &getDirectory(const llvm::sys::fs::UniqueID &UniqueID) { 75 return UniqueDirs[UniqueID]; 76 } 77 78 size_t size() const { return UniqueDirs.size(); } 79 }; 80 81 class FileManager::UniqueFileContainer { 82 /// UniqueFiles - Cache from ID's to existing directories/files. 83 std::set<FileEntry> UniqueFiles; 84 85 public: 86 /// getFile - Return an existing FileEntry with the given ID's if 87 /// there is already one; otherwise create and return a 88 /// default-constructed FileEntry. 89 FileEntry &getFile(llvm::sys::fs::UniqueID UniqueID, bool IsNamedPipe, 90 bool InPCH) { 91 return const_cast<FileEntry &>( 92 *UniqueFiles.insert(FileEntry(UniqueID, IsNamedPipe, InPCH)).first); 93 } 94 95 size_t size() const { return UniqueFiles.size(); } 96 97 void erase(const FileEntry *Entry) { UniqueFiles.erase(*Entry); } 98 }; 99 100 //===----------------------------------------------------------------------===// 101 // Common logic. 102 //===----------------------------------------------------------------------===// 103 104 FileManager::FileManager(const FileSystemOptions &FSO) 105 : FileSystemOpts(FSO), 106 UniqueRealDirs(*new UniqueDirContainer()), 107 UniqueRealFiles(*new UniqueFileContainer()), 108 SeenDirEntries(64), SeenFileEntries(64), NextFileUID(0) { 109 NumDirLookups = NumFileLookups = 0; 110 NumDirCacheMisses = NumFileCacheMisses = 0; 111 } 112 113 FileManager::~FileManager() { 114 delete &UniqueRealDirs; 115 delete &UniqueRealFiles; 116 for (unsigned i = 0, e = VirtualFileEntries.size(); i != e; ++i) 117 delete VirtualFileEntries[i]; 118 for (unsigned i = 0, e = VirtualDirectoryEntries.size(); i != e; ++i) 119 delete VirtualDirectoryEntries[i]; 120 } 121 122 void FileManager::addStatCache(FileSystemStatCache *statCache, 123 bool AtBeginning) { 124 assert(statCache && "No stat cache provided?"); 125 if (AtBeginning || StatCache.get() == 0) { 126 statCache->setNextStatCache(StatCache.take()); 127 StatCache.reset(statCache); 128 return; 129 } 130 131 FileSystemStatCache *LastCache = StatCache.get(); 132 while (LastCache->getNextStatCache()) 133 LastCache = LastCache->getNextStatCache(); 134 135 LastCache->setNextStatCache(statCache); 136 } 137 138 void FileManager::removeStatCache(FileSystemStatCache *statCache) { 139 if (!statCache) 140 return; 141 142 if (StatCache.get() == statCache) { 143 // This is the first stat cache. 144 StatCache.reset(StatCache->takeNextStatCache()); 145 return; 146 } 147 148 // Find the stat cache in the list. 149 FileSystemStatCache *PrevCache = StatCache.get(); 150 while (PrevCache && PrevCache->getNextStatCache() != statCache) 151 PrevCache = PrevCache->getNextStatCache(); 152 153 assert(PrevCache && "Stat cache not found for removal"); 154 PrevCache->setNextStatCache(statCache->getNextStatCache()); 155 } 156 157 void FileManager::clearStatCaches() { 158 StatCache.reset(0); 159 } 160 161 /// \brief Retrieve the directory that the given file name resides in. 162 /// Filename can point to either a real file or a virtual file. 163 static const DirectoryEntry *getDirectoryFromFile(FileManager &FileMgr, 164 StringRef Filename, 165 bool CacheFailure) { 166 if (Filename.empty()) 167 return NULL; 168 169 if (llvm::sys::path::is_separator(Filename[Filename.size() - 1])) 170 return NULL; // If Filename is a directory. 171 172 StringRef DirName = llvm::sys::path::parent_path(Filename); 173 // Use the current directory if file has no path component. 174 if (DirName.empty()) 175 DirName = "."; 176 177 return FileMgr.getDirectory(DirName, CacheFailure); 178 } 179 180 /// Add all ancestors of the given path (pointing to either a file or 181 /// a directory) as virtual directories. 182 void FileManager::addAncestorsAsVirtualDirs(StringRef Path) { 183 StringRef DirName = llvm::sys::path::parent_path(Path); 184 if (DirName.empty()) 185 return; 186 187 llvm::StringMapEntry<DirectoryEntry *> &NamedDirEnt = 188 SeenDirEntries.GetOrCreateValue(DirName); 189 190 // When caching a virtual directory, we always cache its ancestors 191 // at the same time. Therefore, if DirName is already in the cache, 192 // we don't need to recurse as its ancestors must also already be in 193 // the cache. 194 if (NamedDirEnt.getValue()) 195 return; 196 197 // Add the virtual directory to the cache. 198 DirectoryEntry *UDE = new DirectoryEntry; 199 UDE->Name = NamedDirEnt.getKeyData(); 200 NamedDirEnt.setValue(UDE); 201 VirtualDirectoryEntries.push_back(UDE); 202 203 // Recursively add the other ancestors. 204 addAncestorsAsVirtualDirs(DirName); 205 } 206 207 const DirectoryEntry *FileManager::getDirectory(StringRef DirName, 208 bool CacheFailure) { 209 // stat doesn't like trailing separators except for root directory. 210 // At least, on Win32 MSVCRT, stat() cannot strip trailing '/'. 211 // (though it can strip '\\') 212 if (DirName.size() > 1 && 213 DirName != llvm::sys::path::root_path(DirName) && 214 llvm::sys::path::is_separator(DirName.back())) 215 DirName = DirName.substr(0, DirName.size()-1); 216 #ifdef LLVM_ON_WIN32 217 // Fixing a problem with "clang C:test.c" on Windows. 218 // Stat("C:") does not recognize "C:" as a valid directory 219 std::string DirNameStr; 220 if (DirName.size() > 1 && DirName.back() == ':' && 221 DirName.equals_lower(llvm::sys::path::root_name(DirName))) { 222 DirNameStr = DirName.str() + '.'; 223 DirName = DirNameStr; 224 } 225 #endif 226 227 ++NumDirLookups; 228 llvm::StringMapEntry<DirectoryEntry *> &NamedDirEnt = 229 SeenDirEntries.GetOrCreateValue(DirName); 230 231 // See if there was already an entry in the map. Note that the map 232 // contains both virtual and real directories. 233 if (NamedDirEnt.getValue()) 234 return NamedDirEnt.getValue() == NON_EXISTENT_DIR 235 ? 0 : NamedDirEnt.getValue(); 236 237 ++NumDirCacheMisses; 238 239 // By default, initialize it to invalid. 240 NamedDirEnt.setValue(NON_EXISTENT_DIR); 241 242 // Get the null-terminated directory name as stored as the key of the 243 // SeenDirEntries map. 244 const char *InterndDirName = NamedDirEnt.getKeyData(); 245 246 // Check to see if the directory exists. 247 FileData Data; 248 if (getStatValue(InterndDirName, Data, false, 0 /*directory lookup*/)) { 249 // There's no real directory at the given path. 250 if (!CacheFailure) 251 SeenDirEntries.erase(DirName); 252 return 0; 253 } 254 255 // It exists. See if we have already opened a directory with the 256 // same inode (this occurs on Unix-like systems when one dir is 257 // symlinked to another, for example) or the same path (on 258 // Windows). 259 DirectoryEntry &UDE = 260 UniqueRealDirs.getDirectory(Data.UniqueID); 261 262 NamedDirEnt.setValue(&UDE); 263 if (!UDE.getName()) { 264 // We don't have this directory yet, add it. We use the string 265 // key from the SeenDirEntries map as the string. 266 UDE.Name = InterndDirName; 267 } 268 269 return &UDE; 270 } 271 272 const FileEntry *FileManager::getFile(StringRef Filename, bool openFile, 273 bool CacheFailure) { 274 ++NumFileLookups; 275 276 // See if there is already an entry in the map. 277 llvm::StringMapEntry<FileEntry *> &NamedFileEnt = 278 SeenFileEntries.GetOrCreateValue(Filename); 279 280 // See if there is already an entry in the map. 281 if (NamedFileEnt.getValue()) 282 return NamedFileEnt.getValue() == NON_EXISTENT_FILE 283 ? 0 : NamedFileEnt.getValue(); 284 285 ++NumFileCacheMisses; 286 287 // By default, initialize it to invalid. 288 NamedFileEnt.setValue(NON_EXISTENT_FILE); 289 290 // Get the null-terminated file name as stored as the key of the 291 // SeenFileEntries map. 292 const char *InterndFileName = NamedFileEnt.getKeyData(); 293 294 // Look up the directory for the file. When looking up something like 295 // sys/foo.h we'll discover all of the search directories that have a 'sys' 296 // subdirectory. This will let us avoid having to waste time on known-to-fail 297 // searches when we go to find sys/bar.h, because all the search directories 298 // without a 'sys' subdir will get a cached failure result. 299 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename, 300 CacheFailure); 301 if (DirInfo == 0) { // Directory doesn't exist, file can't exist. 302 if (!CacheFailure) 303 SeenFileEntries.erase(Filename); 304 305 return 0; 306 } 307 308 // FIXME: Use the directory info to prune this, before doing the stat syscall. 309 // FIXME: This will reduce the # syscalls. 310 311 // Nope, there isn't. Check to see if the file exists. 312 int FileDescriptor = -1; 313 FileData Data; 314 if (getStatValue(InterndFileName, Data, true, 315 openFile ? &FileDescriptor : 0)) { 316 // There's no real file at the given path. 317 if (!CacheFailure) 318 SeenFileEntries.erase(Filename); 319 320 return 0; 321 } 322 323 if (FileDescriptor != -1 && !openFile) { 324 close(FileDescriptor); 325 FileDescriptor = -1; 326 } 327 328 // It exists. See if we have already opened a file with the same inode. 329 // This occurs when one dir is symlinked to another, for example. 330 FileEntry &UFE = 331 UniqueRealFiles.getFile(Data.UniqueID, Data.IsNamedPipe, Data.InPCH); 332 333 NamedFileEnt.setValue(&UFE); 334 if (UFE.getName()) { // Already have an entry with this inode, return it. 335 // If the stat process opened the file, close it to avoid a FD leak. 336 if (FileDescriptor != -1) 337 close(FileDescriptor); 338 339 return &UFE; 340 } 341 342 // Otherwise, we don't have this directory yet, add it. 343 // FIXME: Change the name to be a char* that points back to the 344 // 'SeenFileEntries' key. 345 UFE.Name = InterndFileName; 346 UFE.Size = Data.Size; 347 UFE.ModTime = Data.ModTime; 348 UFE.Dir = DirInfo; 349 UFE.UID = NextFileUID++; 350 UFE.FD = FileDescriptor; 351 return &UFE; 352 } 353 354 const FileEntry * 355 FileManager::getVirtualFile(StringRef Filename, off_t Size, 356 time_t ModificationTime) { 357 ++NumFileLookups; 358 359 // See if there is already an entry in the map. 360 llvm::StringMapEntry<FileEntry *> &NamedFileEnt = 361 SeenFileEntries.GetOrCreateValue(Filename); 362 363 // See if there is already an entry in the map. 364 if (NamedFileEnt.getValue() && NamedFileEnt.getValue() != NON_EXISTENT_FILE) 365 return NamedFileEnt.getValue(); 366 367 ++NumFileCacheMisses; 368 369 // By default, initialize it to invalid. 370 NamedFileEnt.setValue(NON_EXISTENT_FILE); 371 372 addAncestorsAsVirtualDirs(Filename); 373 FileEntry *UFE = 0; 374 375 // Now that all ancestors of Filename are in the cache, the 376 // following call is guaranteed to find the DirectoryEntry from the 377 // cache. 378 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename, 379 /*CacheFailure=*/true); 380 assert(DirInfo && 381 "The directory of a virtual file should already be in the cache."); 382 383 // Check to see if the file exists. If so, drop the virtual file 384 FileData Data; 385 const char *InterndFileName = NamedFileEnt.getKeyData(); 386 if (getStatValue(InterndFileName, Data, true, 0) == 0) { 387 Data.Size = Size; 388 Data.ModTime = ModificationTime; 389 UFE = &UniqueRealFiles.getFile(Data.UniqueID, Data.IsNamedPipe, Data.InPCH); 390 391 NamedFileEnt.setValue(UFE); 392 393 // If we had already opened this file, close it now so we don't 394 // leak the descriptor. We're not going to use the file 395 // descriptor anyway, since this is a virtual file. 396 if (UFE->FD != -1) { 397 close(UFE->FD); 398 UFE->FD = -1; 399 } 400 401 // If we already have an entry with this inode, return it. 402 if (UFE->getName()) 403 return UFE; 404 } 405 406 if (!UFE) { 407 UFE = new FileEntry(); 408 VirtualFileEntries.push_back(UFE); 409 NamedFileEnt.setValue(UFE); 410 } 411 412 UFE->Name = InterndFileName; 413 UFE->Size = Size; 414 UFE->ModTime = ModificationTime; 415 UFE->Dir = DirInfo; 416 UFE->UID = NextFileUID++; 417 UFE->FD = -1; 418 return UFE; 419 } 420 421 void FileManager::FixupRelativePath(SmallVectorImpl<char> &path) const { 422 StringRef pathRef(path.data(), path.size()); 423 424 if (FileSystemOpts.WorkingDir.empty() 425 || llvm::sys::path::is_absolute(pathRef)) 426 return; 427 428 SmallString<128> NewPath(FileSystemOpts.WorkingDir); 429 llvm::sys::path::append(NewPath, pathRef); 430 path = NewPath; 431 } 432 433 llvm::MemoryBuffer *FileManager:: 434 getBufferForFile(const FileEntry *Entry, std::string *ErrorStr, 435 bool isVolatile) { 436 OwningPtr<llvm::MemoryBuffer> Result; 437 llvm::error_code ec; 438 439 uint64_t FileSize = Entry->getSize(); 440 // If there's a high enough chance that the file have changed since we 441 // got its size, force a stat before opening it. 442 if (isVolatile) 443 FileSize = -1; 444 445 const char *Filename = Entry->getName(); 446 // If the file is already open, use the open file descriptor. 447 if (Entry->FD != -1) { 448 ec = llvm::MemoryBuffer::getOpenFile(Entry->FD, Filename, Result, FileSize); 449 if (ErrorStr) 450 *ErrorStr = ec.message(); 451 452 close(Entry->FD); 453 Entry->FD = -1; 454 return Result.take(); 455 } 456 457 // Otherwise, open the file. 458 459 if (FileSystemOpts.WorkingDir.empty()) { 460 ec = llvm::MemoryBuffer::getFile(Filename, Result, FileSize); 461 if (ec && ErrorStr) 462 *ErrorStr = ec.message(); 463 return Result.take(); 464 } 465 466 SmallString<128> FilePath(Entry->getName()); 467 FixupRelativePath(FilePath); 468 ec = llvm::MemoryBuffer::getFile(FilePath.str(), Result, FileSize); 469 if (ec && ErrorStr) 470 *ErrorStr = ec.message(); 471 return Result.take(); 472 } 473 474 llvm::MemoryBuffer *FileManager:: 475 getBufferForFile(StringRef Filename, std::string *ErrorStr) { 476 OwningPtr<llvm::MemoryBuffer> Result; 477 llvm::error_code ec; 478 if (FileSystemOpts.WorkingDir.empty()) { 479 ec = llvm::MemoryBuffer::getFile(Filename, Result); 480 if (ec && ErrorStr) 481 *ErrorStr = ec.message(); 482 return Result.take(); 483 } 484 485 SmallString<128> FilePath(Filename); 486 FixupRelativePath(FilePath); 487 ec = llvm::MemoryBuffer::getFile(FilePath.c_str(), Result); 488 if (ec && ErrorStr) 489 *ErrorStr = ec.message(); 490 return Result.take(); 491 } 492 493 /// getStatValue - Get the 'stat' information for the specified path, 494 /// using the cache to accelerate it if possible. This returns true 495 /// if the path points to a virtual file or does not exist, or returns 496 /// false if it's an existent real file. If FileDescriptor is NULL, 497 /// do directory look-up instead of file look-up. 498 bool FileManager::getStatValue(const char *Path, FileData &Data, bool isFile, 499 int *FileDescriptor) { 500 // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be 501 // absolute! 502 if (FileSystemOpts.WorkingDir.empty()) 503 return FileSystemStatCache::get(Path, Data, isFile, FileDescriptor, 504 StatCache.get()); 505 506 SmallString<128> FilePath(Path); 507 FixupRelativePath(FilePath); 508 509 return FileSystemStatCache::get(FilePath.c_str(), Data, isFile, 510 FileDescriptor, StatCache.get()); 511 } 512 513 bool FileManager::getNoncachedStatValue(StringRef Path, 514 llvm::sys::fs::file_status &Result) { 515 SmallString<128> FilePath(Path); 516 FixupRelativePath(FilePath); 517 518 return llvm::sys::fs::status(FilePath.c_str(), Result); 519 } 520 521 void FileManager::invalidateCache(const FileEntry *Entry) { 522 assert(Entry && "Cannot invalidate a NULL FileEntry"); 523 524 SeenFileEntries.erase(Entry->getName()); 525 526 // FileEntry invalidation should not block future optimizations in the file 527 // caches. Possible alternatives are cache truncation (invalidate last N) or 528 // invalidation of the whole cache. 529 UniqueRealFiles.erase(Entry); 530 } 531 532 533 void FileManager::GetUniqueIDMapping( 534 SmallVectorImpl<const FileEntry *> &UIDToFiles) const { 535 UIDToFiles.clear(); 536 UIDToFiles.resize(NextFileUID); 537 538 // Map file entries 539 for (llvm::StringMap<FileEntry*, llvm::BumpPtrAllocator>::const_iterator 540 FE = SeenFileEntries.begin(), FEEnd = SeenFileEntries.end(); 541 FE != FEEnd; ++FE) 542 if (FE->getValue() && FE->getValue() != NON_EXISTENT_FILE) 543 UIDToFiles[FE->getValue()->getUID()] = FE->getValue(); 544 545 // Map virtual file entries 546 for (SmallVectorImpl<FileEntry *>::const_iterator 547 VFE = VirtualFileEntries.begin(), VFEEnd = VirtualFileEntries.end(); 548 VFE != VFEEnd; ++VFE) 549 if (*VFE && *VFE != NON_EXISTENT_FILE) 550 UIDToFiles[(*VFE)->getUID()] = *VFE; 551 } 552 553 void FileManager::modifyFileEntry(FileEntry *File, 554 off_t Size, time_t ModificationTime) { 555 File->Size = Size; 556 File->ModTime = ModificationTime; 557 } 558 559 StringRef FileManager::getCanonicalName(const DirectoryEntry *Dir) { 560 // FIXME: use llvm::sys::fs::canonical() when it gets implemented 561 #ifdef LLVM_ON_UNIX 562 llvm::DenseMap<const DirectoryEntry *, llvm::StringRef>::iterator Known 563 = CanonicalDirNames.find(Dir); 564 if (Known != CanonicalDirNames.end()) 565 return Known->second; 566 567 StringRef CanonicalName(Dir->getName()); 568 char CanonicalNameBuf[PATH_MAX]; 569 if (realpath(Dir->getName(), CanonicalNameBuf)) { 570 unsigned Len = strlen(CanonicalNameBuf); 571 char *Mem = static_cast<char *>(CanonicalNameStorage.Allocate(Len, 1)); 572 memcpy(Mem, CanonicalNameBuf, Len); 573 CanonicalName = StringRef(Mem, Len); 574 } 575 576 CanonicalDirNames.insert(std::make_pair(Dir, CanonicalName)); 577 return CanonicalName; 578 #else 579 return StringRef(Dir->getName()); 580 #endif 581 } 582 583 void FileManager::PrintStats() const { 584 llvm::errs() << "\n*** File Manager Stats:\n"; 585 llvm::errs() << UniqueRealFiles.size() << " real files found, " 586 << UniqueRealDirs.size() << " real dirs found.\n"; 587 llvm::errs() << VirtualFileEntries.size() << " virtual files found, " 588 << VirtualDirectoryEntries.size() << " virtual dirs found.\n"; 589 llvm::errs() << NumDirLookups << " dir lookups, " 590 << NumDirCacheMisses << " dir cache misses.\n"; 591 llvm::errs() << NumFileLookups << " file lookups, " 592 << NumFileCacheMisses << " file cache misses.\n"; 593 594 //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups; 595 } 596