FileManager.cpp revision 3ce42c37b0ef6b10538f5040a33d19a6e6988d05
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 40using 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 54FileEntry::~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 66namespace { 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 78class FileManager::UniqueDirContainer { 79 /// UniqueDirs - Cache from full path to existing directories/files. 80 /// 81 llvm::StringMap<DirectoryEntry> UniqueDirs; 82 83public: 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 96class FileManager::UniqueFileContainer { 97 /// UniqueFiles - Cache from full path to existing directories/files. 98 /// 99 llvm::StringMap<FileEntry, llvm::BumpPtrAllocator> UniqueFiles; 100 101public: 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 friend class FileManager; 116}; 117 118//===----------------------------------------------------------------------===// 119// Unix-like Systems. 120//===----------------------------------------------------------------------===// 121 122#else 123 124class FileManager::UniqueDirContainer { 125 /// UniqueDirs - Cache from ID's to existing directories/files. 126 std::map<std::pair<dev_t, ino_t>, DirectoryEntry> UniqueDirs; 127 128public: 129 /// getDirectory - Return an existing DirectoryEntry with the given 130 /// ID's if there is already one; otherwise create and return a 131 /// default-constructed DirectoryEntry. 132 DirectoryEntry &getDirectory(const char * /*Name*/, 133 const struct stat &StatBuf) { 134 return UniqueDirs[std::make_pair(StatBuf.st_dev, StatBuf.st_ino)]; 135 } 136 137 size_t size() const { return UniqueDirs.size(); } 138}; 139 140class FileManager::UniqueFileContainer { 141 /// UniqueFiles - Cache from ID's to existing directories/files. 142 std::set<FileEntry> UniqueFiles; 143 144public: 145 /// getFile - Return an existing FileEntry with the given ID's if 146 /// there is already one; otherwise create and return a 147 /// default-constructed FileEntry. 148 FileEntry &getFile(const char * /*Name*/, const struct stat &StatBuf) { 149 return 150 const_cast<FileEntry&>( 151 *UniqueFiles.insert(FileEntry(StatBuf.st_dev, 152 StatBuf.st_ino, 153 StatBuf.st_mode)).first); 154 } 155 156 size_t size() const { return UniqueFiles.size(); } 157 158 friend class FileManager; 159}; 160 161#endif 162 163//===----------------------------------------------------------------------===// 164// Common logic. 165//===----------------------------------------------------------------------===// 166 167FileManager::FileManager(const FileSystemOptions &FSO) 168 : FileSystemOpts(FSO), 169 UniqueRealDirs(*new UniqueDirContainer()), 170 UniqueRealFiles(*new UniqueFileContainer()), 171 SeenDirEntries(64), SeenFileEntries(64), NextFileUID(0) { 172 NumDirLookups = NumFileLookups = 0; 173 NumDirCacheMisses = NumFileCacheMisses = 0; 174} 175 176FileManager::~FileManager() { 177 delete &UniqueRealDirs; 178 delete &UniqueRealFiles; 179 for (unsigned i = 0, e = VirtualFileEntries.size(); i != e; ++i) 180 delete VirtualFileEntries[i]; 181 for (unsigned i = 0, e = VirtualDirectoryEntries.size(); i != e; ++i) 182 delete VirtualDirectoryEntries[i]; 183} 184 185void FileManager::addStatCache(FileSystemStatCache *statCache, 186 bool AtBeginning) { 187 assert(statCache && "No stat cache provided?"); 188 if (AtBeginning || StatCache.get() == 0) { 189 statCache->setNextStatCache(StatCache.take()); 190 StatCache.reset(statCache); 191 return; 192 } 193 194 FileSystemStatCache *LastCache = StatCache.get(); 195 while (LastCache->getNextStatCache()) 196 LastCache = LastCache->getNextStatCache(); 197 198 LastCache->setNextStatCache(statCache); 199} 200 201void FileManager::removeStatCache(FileSystemStatCache *statCache) { 202 if (!statCache) 203 return; 204 205 if (StatCache.get() == statCache) { 206 // This is the first stat cache. 207 StatCache.reset(StatCache->takeNextStatCache()); 208 return; 209 } 210 211 // Find the stat cache in the list. 212 FileSystemStatCache *PrevCache = StatCache.get(); 213 while (PrevCache && PrevCache->getNextStatCache() != statCache) 214 PrevCache = PrevCache->getNextStatCache(); 215 216 assert(PrevCache && "Stat cache not found for removal"); 217 PrevCache->setNextStatCache(statCache->getNextStatCache()); 218} 219 220/// \brief Retrieve the directory that the given file name resides in. 221/// Filename can point to either a real file or a virtual file. 222static const DirectoryEntry *getDirectoryFromFile(FileManager &FileMgr, 223 StringRef Filename, 224 bool CacheFailure) { 225 if (Filename.empty()) 226 return NULL; 227 228 if (llvm::sys::path::is_separator(Filename[Filename.size() - 1])) 229 return NULL; // If Filename is a directory. 230 231 StringRef DirName = llvm::sys::path::parent_path(Filename); 232 // Use the current directory if file has no path component. 233 if (DirName.empty()) 234 DirName = "."; 235 236 return FileMgr.getDirectory(DirName, CacheFailure); 237} 238 239/// Add all ancestors of the given path (pointing to either a file or 240/// a directory) as virtual directories. 241void FileManager::addAncestorsAsVirtualDirs(StringRef Path) { 242 StringRef DirName = llvm::sys::path::parent_path(Path); 243 if (DirName.empty()) 244 return; 245 246 llvm::StringMapEntry<DirectoryEntry *> &NamedDirEnt = 247 SeenDirEntries.GetOrCreateValue(DirName); 248 249 // When caching a virtual directory, we always cache its ancestors 250 // at the same time. Therefore, if DirName is already in the cache, 251 // we don't need to recurse as its ancestors must also already be in 252 // the cache. 253 if (NamedDirEnt.getValue()) 254 return; 255 256 // Add the virtual directory to the cache. 257 DirectoryEntry *UDE = new DirectoryEntry; 258 UDE->Name = NamedDirEnt.getKeyData(); 259 NamedDirEnt.setValue(UDE); 260 VirtualDirectoryEntries.push_back(UDE); 261 262 // Recursively add the other ancestors. 263 addAncestorsAsVirtualDirs(DirName); 264} 265 266const DirectoryEntry *FileManager::getDirectory(StringRef DirName, 267 bool CacheFailure) { 268 // stat doesn't like trailing separators except for root directory. 269 // At least, on Win32 MSVCRT, stat() cannot strip trailing '/'. 270 // (though it can strip '\\') 271 if (DirName.size() > 1 && 272 DirName != llvm::sys::path::root_path(DirName) && 273 llvm::sys::path::is_separator(DirName.back())) 274 DirName = DirName.substr(0, DirName.size()-1); 275 276 ++NumDirLookups; 277 llvm::StringMapEntry<DirectoryEntry *> &NamedDirEnt = 278 SeenDirEntries.GetOrCreateValue(DirName); 279 280 // See if there was already an entry in the map. Note that the map 281 // contains both virtual and real directories. 282 if (NamedDirEnt.getValue()) 283 return NamedDirEnt.getValue() == NON_EXISTENT_DIR 284 ? 0 : NamedDirEnt.getValue(); 285 286 ++NumDirCacheMisses; 287 288 // By default, initialize it to invalid. 289 NamedDirEnt.setValue(NON_EXISTENT_DIR); 290 291 // Get the null-terminated directory name as stored as the key of the 292 // SeenDirEntries map. 293 const char *InterndDirName = NamedDirEnt.getKeyData(); 294 295 // Check to see if the directory exists. 296 struct stat StatBuf; 297 if (getStatValue(InterndDirName, StatBuf, 0/*directory lookup*/)) { 298 // There's no real directory at the given path. 299 if (!CacheFailure) 300 SeenDirEntries.erase(DirName); 301 return 0; 302 } 303 304 // It exists. See if we have already opened a directory with the 305 // same inode (this occurs on Unix-like systems when one dir is 306 // symlinked to another, for example) or the same path (on 307 // Windows). 308 DirectoryEntry &UDE = UniqueRealDirs.getDirectory(InterndDirName, StatBuf); 309 310 NamedDirEnt.setValue(&UDE); 311 if (!UDE.getName()) { 312 // We don't have this directory yet, add it. We use the string 313 // key from the SeenDirEntries map as the string. 314 UDE.Name = InterndDirName; 315 } 316 317 return &UDE; 318} 319 320const FileEntry *FileManager::getFile(StringRef Filename, bool openFile, 321 bool CacheFailure) { 322 ++NumFileLookups; 323 324 // See if there is already an entry in the map. 325 llvm::StringMapEntry<FileEntry *> &NamedFileEnt = 326 SeenFileEntries.GetOrCreateValue(Filename); 327 328 // See if there is already an entry in the map. 329 if (NamedFileEnt.getValue()) 330 return NamedFileEnt.getValue() == NON_EXISTENT_FILE 331 ? 0 : NamedFileEnt.getValue(); 332 333 ++NumFileCacheMisses; 334 335 // By default, initialize it to invalid. 336 NamedFileEnt.setValue(NON_EXISTENT_FILE); 337 338 // Get the null-terminated file name as stored as the key of the 339 // SeenFileEntries map. 340 const char *InterndFileName = NamedFileEnt.getKeyData(); 341 342 // Look up the directory for the file. When looking up something like 343 // sys/foo.h we'll discover all of the search directories that have a 'sys' 344 // subdirectory. This will let us avoid having to waste time on known-to-fail 345 // searches when we go to find sys/bar.h, because all the search directories 346 // without a 'sys' subdir will get a cached failure result. 347 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename, 348 CacheFailure); 349 if (DirInfo == 0) { // Directory doesn't exist, file can't exist. 350 if (!CacheFailure) 351 SeenFileEntries.erase(Filename); 352 353 return 0; 354 } 355 356 // FIXME: Use the directory info to prune this, before doing the stat syscall. 357 // FIXME: This will reduce the # syscalls. 358 359 // Nope, there isn't. Check to see if the file exists. 360 int FileDescriptor = -1; 361 struct stat StatBuf; 362 if (getStatValue(InterndFileName, StatBuf, &FileDescriptor)) { 363 // There's no real file at the given path. 364 if (!CacheFailure) 365 SeenFileEntries.erase(Filename); 366 367 return 0; 368 } 369 370 if (FileDescriptor != -1 && !openFile) { 371 close(FileDescriptor); 372 FileDescriptor = -1; 373 } 374 375 // It exists. See if we have already opened a file with the same inode. 376 // This occurs when one dir is symlinked to another, for example. 377 FileEntry &UFE = UniqueRealFiles.getFile(InterndFileName, StatBuf); 378 379 NamedFileEnt.setValue(&UFE); 380 if (UFE.getName()) { // Already have an entry with this inode, return it. 381 // If the stat process opened the file, close it to avoid a FD leak. 382 if (FileDescriptor != -1) 383 close(FileDescriptor); 384 385 return &UFE; 386 } 387 388 // Otherwise, we don't have this directory yet, add it. 389 // FIXME: Change the name to be a char* that points back to the 390 // 'SeenFileEntries' key. 391 UFE.Name = InterndFileName; 392 UFE.Size = StatBuf.st_size; 393 UFE.ModTime = StatBuf.st_mtime; 394 UFE.Dir = DirInfo; 395 UFE.UID = NextFileUID++; 396 UFE.FD = FileDescriptor; 397 return &UFE; 398} 399 400const FileEntry * 401FileManager::getVirtualFile(StringRef Filename, off_t Size, 402 time_t ModificationTime) { 403 ++NumFileLookups; 404 405 // See if there is already an entry in the map. 406 llvm::StringMapEntry<FileEntry *> &NamedFileEnt = 407 SeenFileEntries.GetOrCreateValue(Filename); 408 409 // See if there is already an entry in the map. 410 if (NamedFileEnt.getValue() && NamedFileEnt.getValue() != NON_EXISTENT_FILE) 411 return NamedFileEnt.getValue(); 412 413 ++NumFileCacheMisses; 414 415 // By default, initialize it to invalid. 416 NamedFileEnt.setValue(NON_EXISTENT_FILE); 417 418 addAncestorsAsVirtualDirs(Filename); 419 FileEntry *UFE = 0; 420 421 // Now that all ancestors of Filename are in the cache, the 422 // following call is guaranteed to find the DirectoryEntry from the 423 // cache. 424 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename, 425 /*CacheFailure=*/true); 426 assert(DirInfo && 427 "The directory of a virtual file should already be in the cache."); 428 429 // Check to see if the file exists. If so, drop the virtual file 430 int FileDescriptor = -1; 431 struct stat StatBuf; 432 const char *InterndFileName = NamedFileEnt.getKeyData(); 433 if (getStatValue(InterndFileName, StatBuf, &FileDescriptor) == 0) { 434 // If the stat process opened the file, close it to avoid a FD leak. 435 if (FileDescriptor != -1) 436 close(FileDescriptor); 437 438 StatBuf.st_size = Size; 439 StatBuf.st_mtime = ModificationTime; 440 UFE = &UniqueRealFiles.getFile(InterndFileName, StatBuf); 441 442 NamedFileEnt.setValue(UFE); 443 444 // If we had already opened this file, close it now so we don't 445 // leak the descriptor. We're not going to use the file 446 // descriptor anyway, since this is a virtual file. 447 if (UFE->FD != -1) { 448 close(UFE->FD); 449 UFE->FD = -1; 450 } 451 452 // If we already have an entry with this inode, return it. 453 if (UFE->getName()) 454 return UFE; 455 } 456 457 if (!UFE) { 458 UFE = new FileEntry(); 459 VirtualFileEntries.push_back(UFE); 460 NamedFileEnt.setValue(UFE); 461 } 462 463 UFE->Name = InterndFileName; 464 UFE->Size = Size; 465 UFE->ModTime = ModificationTime; 466 UFE->Dir = DirInfo; 467 UFE->UID = NextFileUID++; 468 UFE->FD = -1; 469 return UFE; 470} 471 472void FileManager::FixupRelativePath(SmallVectorImpl<char> &path) const { 473 StringRef pathRef(path.data(), path.size()); 474 475 if (FileSystemOpts.WorkingDir.empty() 476 || llvm::sys::path::is_absolute(pathRef)) 477 return; 478 479 SmallString<128> NewPath(FileSystemOpts.WorkingDir); 480 llvm::sys::path::append(NewPath, pathRef); 481 path = NewPath; 482} 483 484llvm::MemoryBuffer *FileManager:: 485getBufferForFile(const FileEntry *Entry, std::string *ErrorStr) { 486 OwningPtr<llvm::MemoryBuffer> Result; 487 llvm::error_code ec; 488 489 const char *Filename = Entry->getName(); 490 // If the file is already open, use the open file descriptor. 491 if (Entry->FD != -1) { 492 ec = llvm::MemoryBuffer::getOpenFile(Entry->FD, Filename, Result, 493 Entry->getSize()); 494 if (ErrorStr) 495 *ErrorStr = ec.message(); 496 497 close(Entry->FD); 498 Entry->FD = -1; 499 return Result.take(); 500 } 501 502 // Otherwise, open the file. 503 504 if (FileSystemOpts.WorkingDir.empty()) { 505 ec = llvm::MemoryBuffer::getFile(Filename, Result, Entry->getSize()); 506 if (ec && ErrorStr) 507 *ErrorStr = ec.message(); 508 return Result.take(); 509 } 510 511 SmallString<128> FilePath(Entry->getName()); 512 FixupRelativePath(FilePath); 513 ec = llvm::MemoryBuffer::getFile(FilePath.str(), Result, Entry->getSize()); 514 if (ec && ErrorStr) 515 *ErrorStr = ec.message(); 516 return Result.take(); 517} 518 519llvm::MemoryBuffer *FileManager:: 520getBufferForFile(StringRef Filename, std::string *ErrorStr) { 521 OwningPtr<llvm::MemoryBuffer> Result; 522 llvm::error_code ec; 523 if (FileSystemOpts.WorkingDir.empty()) { 524 ec = llvm::MemoryBuffer::getFile(Filename, Result); 525 if (ec && ErrorStr) 526 *ErrorStr = ec.message(); 527 return Result.take(); 528 } 529 530 SmallString<128> FilePath(Filename); 531 FixupRelativePath(FilePath); 532 ec = llvm::MemoryBuffer::getFile(FilePath.c_str(), Result); 533 if (ec && ErrorStr) 534 *ErrorStr = ec.message(); 535 return Result.take(); 536} 537 538/// getStatValue - Get the 'stat' information for the specified path, 539/// using the cache to accelerate it if possible. This returns true 540/// if the path points to a virtual file or does not exist, or returns 541/// false if it's an existent real file. If FileDescriptor is NULL, 542/// do directory look-up instead of file look-up. 543bool FileManager::getStatValue(const char *Path, struct stat &StatBuf, 544 int *FileDescriptor) { 545 // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be 546 // absolute! 547 if (FileSystemOpts.WorkingDir.empty()) 548 return FileSystemStatCache::get(Path, StatBuf, FileDescriptor, 549 StatCache.get()); 550 551 SmallString<128> FilePath(Path); 552 FixupRelativePath(FilePath); 553 554 return FileSystemStatCache::get(FilePath.c_str(), StatBuf, FileDescriptor, 555 StatCache.get()); 556} 557 558bool FileManager::getNoncachedStatValue(StringRef Path, 559 struct stat &StatBuf) { 560 SmallString<128> FilePath(Path); 561 FixupRelativePath(FilePath); 562 563 return ::stat(FilePath.c_str(), &StatBuf) != 0; 564} 565 566void FileManager::InvalidateCache(const FileEntry* Entry) { 567 if (!Entry) 568 return; 569 570 SeenFileEntries.erase(Entry->getName()); 571 UniqueRealFiles.UniqueFiles.erase(*Entry); 572} 573 574 575void FileManager::GetUniqueIDMapping( 576 SmallVectorImpl<const FileEntry *> &UIDToFiles) const { 577 UIDToFiles.clear(); 578 UIDToFiles.resize(NextFileUID); 579 580 // Map file entries 581 for (llvm::StringMap<FileEntry*, llvm::BumpPtrAllocator>::const_iterator 582 FE = SeenFileEntries.begin(), FEEnd = SeenFileEntries.end(); 583 FE != FEEnd; ++FE) 584 if (FE->getValue() && FE->getValue() != NON_EXISTENT_FILE) 585 UIDToFiles[FE->getValue()->getUID()] = FE->getValue(); 586 587 // Map virtual file entries 588 for (SmallVector<FileEntry*, 4>::const_iterator 589 VFE = VirtualFileEntries.begin(), VFEEnd = VirtualFileEntries.end(); 590 VFE != VFEEnd; ++VFE) 591 if (*VFE && *VFE != NON_EXISTENT_FILE) 592 UIDToFiles[(*VFE)->getUID()] = *VFE; 593} 594 595void FileManager::modifyFileEntry(FileEntry *File, 596 off_t Size, time_t ModificationTime) { 597 File->Size = Size; 598 File->ModTime = ModificationTime; 599} 600 601 602void FileManager::PrintStats() const { 603 llvm::errs() << "\n*** File Manager Stats:\n"; 604 llvm::errs() << UniqueRealFiles.size() << " real files found, " 605 << UniqueRealDirs.size() << " real dirs found.\n"; 606 llvm::errs() << VirtualFileEntries.size() << " virtual files found, " 607 << VirtualDirectoryEntries.size() << " virtual dirs found.\n"; 608 llvm::errs() << NumDirLookups << " dir lookups, " 609 << NumDirCacheMisses << " dir cache misses.\n"; 610 llvm::errs() << NumFileLookups << " file lookups, " 611 << NumFileCacheMisses << " file cache misses.\n"; 612 613 //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups; 614} 615