1 //===--- FileManager.cpp - File System Probing and Caching ----------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the FileManager interface.
12 //===----------------------------------------------------------------------===//
14 // TODO: This should index all interesting directories with dirent calls.
16 // opendir/readdir_r/closedir ?
18 //===----------------------------------------------------------------------===//
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/ADT/STLExtras.h"
25 #include "llvm/Support/FileSystem.h"
26 #include "llvm/Support/MemoryBuffer.h"
27 #include "llvm/Support/Path.h"
28 #include "llvm/Support/raw_ostream.h"
37 using namespace clang;
39 /// NON_EXISTENT_DIR - A special value distinct from null that is used to
40 /// represent a dir name that doesn't exist on the disk.
41 #define NON_EXISTENT_DIR reinterpret_cast<DirectoryEntry*>((intptr_t)-1)
43 /// NON_EXISTENT_FILE - A special value distinct from null that is used to
44 /// represent a filename that doesn't exist on the disk.
45 #define NON_EXISTENT_FILE reinterpret_cast<FileEntry*>((intptr_t)-1)
47 //===----------------------------------------------------------------------===//
49 //===----------------------------------------------------------------------===//
51 FileManager::FileManager(const FileSystemOptions &FSO,
52 IntrusiveRefCntPtr<vfs::FileSystem> FS)
53 : FS(std::move(FS)), FileSystemOpts(FSO), SeenDirEntries(64),
54 SeenFileEntries(64), NextFileUID(0) {
55 NumDirLookups = NumFileLookups = 0;
56 NumDirCacheMisses = NumFileCacheMisses = 0;
58 // If the caller doesn't provide a virtual file system, just grab the real
61 this->FS = vfs::getRealFileSystem();
64 FileManager::~FileManager() = default;
66 void FileManager::addStatCache(std::unique_ptr<FileSystemStatCache> statCache,
68 assert(statCache && "No stat cache provided?");
69 if (AtBeginning || !StatCache.get()) {
70 statCache->setNextStatCache(std::move(StatCache));
71 StatCache = std::move(statCache);
75 FileSystemStatCache *LastCache = StatCache.get();
76 while (LastCache->getNextStatCache())
77 LastCache = LastCache->getNextStatCache();
79 LastCache->setNextStatCache(std::move(statCache));
82 void FileManager::removeStatCache(FileSystemStatCache *statCache) {
86 if (StatCache.get() == statCache) {
87 // This is the first stat cache.
88 StatCache = StatCache->takeNextStatCache();
92 // Find the stat cache in the list.
93 FileSystemStatCache *PrevCache = StatCache.get();
94 while (PrevCache && PrevCache->getNextStatCache() != statCache)
95 PrevCache = PrevCache->getNextStatCache();
97 assert(PrevCache && "Stat cache not found for removal");
98 PrevCache->setNextStatCache(statCache->takeNextStatCache());
101 void FileManager::clearStatCaches() {
105 /// \brief Retrieve the directory that the given file name resides in.
106 /// Filename can point to either a real file or a virtual file.
107 static const DirectoryEntry *getDirectoryFromFile(FileManager &FileMgr,
110 if (Filename.empty())
113 if (llvm::sys::path::is_separator(Filename[Filename.size() - 1]))
114 return nullptr; // If Filename is a directory.
116 StringRef DirName = llvm::sys::path::parent_path(Filename);
117 // Use the current directory if file has no path component.
121 return FileMgr.getDirectory(DirName, CacheFailure);
124 /// Add all ancestors of the given path (pointing to either a file or
125 /// a directory) as virtual directories.
126 void FileManager::addAncestorsAsVirtualDirs(StringRef Path) {
127 StringRef DirName = llvm::sys::path::parent_path(Path);
132 *SeenDirEntries.insert(std::make_pair(DirName, nullptr)).first;
134 // When caching a virtual directory, we always cache its ancestors
135 // at the same time. Therefore, if DirName is already in the cache,
136 // we don't need to recurse as its ancestors must also already be in
138 if (NamedDirEnt.second && NamedDirEnt.second != NON_EXISTENT_DIR)
141 // Add the virtual directory to the cache.
142 auto UDE = llvm::make_unique<DirectoryEntry>();
143 UDE->Name = NamedDirEnt.first();
144 NamedDirEnt.second = UDE.get();
145 VirtualDirectoryEntries.push_back(std::move(UDE));
147 // Recursively add the other ancestors.
148 addAncestorsAsVirtualDirs(DirName);
151 const DirectoryEntry *FileManager::getDirectory(StringRef DirName,
153 // stat doesn't like trailing separators except for root directory.
154 // At least, on Win32 MSVCRT, stat() cannot strip trailing '/'.
155 // (though it can strip '\\')
156 if (DirName.size() > 1 &&
157 DirName != llvm::sys::path::root_path(DirName) &&
158 llvm::sys::path::is_separator(DirName.back()))
159 DirName = DirName.substr(0, DirName.size()-1);
161 // Fixing a problem with "clang C:test.c" on Windows.
162 // Stat("C:") does not recognize "C:" as a valid directory
163 std::string DirNameStr;
164 if (DirName.size() > 1 && DirName.back() == ':' &&
165 DirName.equals_lower(llvm::sys::path::root_name(DirName))) {
166 DirNameStr = DirName.str() + '.';
167 DirName = DirNameStr;
173 *SeenDirEntries.insert(std::make_pair(DirName, nullptr)).first;
175 // See if there was already an entry in the map. Note that the map
176 // contains both virtual and real directories.
177 if (NamedDirEnt.second)
178 return NamedDirEnt.second == NON_EXISTENT_DIR ? nullptr
179 : NamedDirEnt.second;
183 // By default, initialize it to invalid.
184 NamedDirEnt.second = NON_EXISTENT_DIR;
186 // Get the null-terminated directory name as stored as the key of the
187 // SeenDirEntries map.
188 StringRef InterndDirName = NamedDirEnt.first();
190 // Check to see if the directory exists.
192 if (getStatValue(InterndDirName, Data, false, nullptr /*directory lookup*/)) {
193 // There's no real directory at the given path.
195 SeenDirEntries.erase(DirName);
199 // It exists. See if we have already opened a directory with the
200 // same inode (this occurs on Unix-like systems when one dir is
201 // symlinked to another, for example) or the same path (on
203 DirectoryEntry &UDE = UniqueRealDirs[Data.UniqueID];
205 NamedDirEnt.second = &UDE;
206 if (UDE.getName().empty()) {
207 // We don't have this directory yet, add it. We use the string
208 // key from the SeenDirEntries map as the string.
209 UDE.Name = InterndDirName;
215 const FileEntry *FileManager::getFile(StringRef Filename, bool openFile,
219 // See if there is already an entry in the map.
221 *SeenFileEntries.insert(std::make_pair(Filename, nullptr)).first;
223 // See if there is already an entry in the map.
224 if (NamedFileEnt.second)
225 return NamedFileEnt.second == NON_EXISTENT_FILE ? nullptr
226 : NamedFileEnt.second;
228 ++NumFileCacheMisses;
230 // By default, initialize it to invalid.
231 NamedFileEnt.second = NON_EXISTENT_FILE;
233 // Get the null-terminated file name as stored as the key of the
234 // SeenFileEntries map.
235 StringRef InterndFileName = NamedFileEnt.first();
237 // Look up the directory for the file. When looking up something like
238 // sys/foo.h we'll discover all of the search directories that have a 'sys'
239 // subdirectory. This will let us avoid having to waste time on known-to-fail
240 // searches when we go to find sys/bar.h, because all the search directories
241 // without a 'sys' subdir will get a cached failure result.
242 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename,
244 if (DirInfo == nullptr) { // Directory doesn't exist, file can't exist.
246 SeenFileEntries.erase(Filename);
251 // FIXME: Use the directory info to prune this, before doing the stat syscall.
252 // FIXME: This will reduce the # syscalls.
254 // Nope, there isn't. Check to see if the file exists.
255 std::unique_ptr<vfs::File> F;
257 if (getStatValue(InterndFileName, Data, true, openFile ? &F : nullptr)) {
258 // There's no real file at the given path.
260 SeenFileEntries.erase(Filename);
265 assert((openFile || !F) && "undesired open file");
267 // It exists. See if we have already opened a file with the same inode.
268 // This occurs when one dir is symlinked to another, for example.
269 FileEntry &UFE = UniqueRealFiles[Data.UniqueID];
271 NamedFileEnt.second = &UFE;
273 // If the name returned by getStatValue is different than Filename, re-intern
275 if (Data.Name != Filename) {
277 *SeenFileEntries.insert(std::make_pair(Data.Name, nullptr)).first;
278 if (!NamedFileEnt.second)
279 NamedFileEnt.second = &UFE;
281 assert(NamedFileEnt.second == &UFE &&
282 "filename from getStatValue() refers to wrong file");
283 InterndFileName = NamedFileEnt.first().data();
286 if (UFE.isValid()) { // Already have an entry with this inode, return it.
288 // FIXME: this hack ensures that if we look up a file by a virtual path in
289 // the VFS that the getDir() will have the virtual path, even if we found
290 // the file by a 'real' path first. This is required in order to find a
291 // module's structure when its headers/module map are mapped in the VFS.
292 // We should remove this as soon as we can properly support a file having
294 if (DirInfo != UFE.Dir && Data.IsVFSMapped)
297 // Always update the name to use the last name by which a file was accessed.
298 // FIXME: Neither this nor always using the first name is correct; we want
299 // to switch towards a design where we return a FileName object that
300 // encapsulates both the name by which the file was accessed and the
301 // corresponding FileEntry.
302 UFE.Name = InterndFileName;
307 // Otherwise, we don't have this file yet, add it.
308 UFE.Name = InterndFileName;
309 UFE.Size = Data.Size;
310 UFE.ModTime = Data.ModTime;
312 UFE.UID = NextFileUID++;
313 UFE.UniqueID = Data.UniqueID;
314 UFE.IsNamedPipe = Data.IsNamedPipe;
315 UFE.InPCH = Data.InPCH;
316 UFE.File = std::move(F);
319 if (auto RealPathName = UFE.File->getName())
320 UFE.RealPathName = *RealPathName;
325 FileManager::getVirtualFile(StringRef Filename, off_t Size,
326 time_t ModificationTime) {
329 // See if there is already an entry in the map.
331 *SeenFileEntries.insert(std::make_pair(Filename, nullptr)).first;
333 // See if there is already an entry in the map.
334 if (NamedFileEnt.second && NamedFileEnt.second != NON_EXISTENT_FILE)
335 return NamedFileEnt.second;
337 ++NumFileCacheMisses;
339 // By default, initialize it to invalid.
340 NamedFileEnt.second = NON_EXISTENT_FILE;
342 addAncestorsAsVirtualDirs(Filename);
343 FileEntry *UFE = nullptr;
345 // Now that all ancestors of Filename are in the cache, the
346 // following call is guaranteed to find the DirectoryEntry from the
348 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename,
349 /*CacheFailure=*/true);
351 "The directory of a virtual file should already be in the cache.");
353 // Check to see if the file exists. If so, drop the virtual file
355 const char *InterndFileName = NamedFileEnt.first().data();
356 if (getStatValue(InterndFileName, Data, true, nullptr) == 0) {
358 Data.ModTime = ModificationTime;
359 UFE = &UniqueRealFiles[Data.UniqueID];
361 NamedFileEnt.second = UFE;
363 // If we had already opened this file, close it now so we don't
364 // leak the descriptor. We're not going to use the file
365 // descriptor anyway, since this is a virtual file.
369 // If we already have an entry with this inode, return it.
373 UFE->UniqueID = Data.UniqueID;
374 UFE->IsNamedPipe = Data.IsNamedPipe;
375 UFE->InPCH = Data.InPCH;
379 VirtualFileEntries.push_back(llvm::make_unique<FileEntry>());
380 UFE = VirtualFileEntries.back().get();
381 NamedFileEnt.second = UFE;
384 UFE->Name = InterndFileName;
386 UFE->ModTime = ModificationTime;
388 UFE->UID = NextFileUID++;
394 bool FileManager::FixupRelativePath(SmallVectorImpl<char> &path) const {
395 StringRef pathRef(path.data(), path.size());
397 if (FileSystemOpts.WorkingDir.empty()
398 || llvm::sys::path::is_absolute(pathRef))
401 SmallString<128> NewPath(FileSystemOpts.WorkingDir);
402 llvm::sys::path::append(NewPath, pathRef);
407 bool FileManager::makeAbsolutePath(SmallVectorImpl<char> &Path) const {
408 bool Changed = FixupRelativePath(Path);
410 if (!llvm::sys::path::is_absolute(StringRef(Path.data(), Path.size()))) {
411 FS->makeAbsolute(Path);
418 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
419 FileManager::getBufferForFile(const FileEntry *Entry, bool isVolatile,
420 bool ShouldCloseOpenFile) {
421 uint64_t FileSize = Entry->getSize();
422 // If there's a high enough chance that the file have changed since we
423 // got its size, force a stat before opening it.
427 StringRef Filename = Entry->getName();
428 // If the file is already open, use the open file descriptor.
431 Entry->File->getBuffer(Filename, FileSize,
432 /*RequiresNullTerminator=*/true, isVolatile);
433 // FIXME: we need a set of APIs that can make guarantees about whether a
434 // FileEntry is open or not.
435 if (ShouldCloseOpenFile)
440 // Otherwise, open the file.
442 if (FileSystemOpts.WorkingDir.empty())
443 return FS->getBufferForFile(Filename, FileSize,
444 /*RequiresNullTerminator=*/true, isVolatile);
446 SmallString<128> FilePath(Entry->getName());
447 FixupRelativePath(FilePath);
448 return FS->getBufferForFile(FilePath, FileSize,
449 /*RequiresNullTerminator=*/true, isVolatile);
452 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
453 FileManager::getBufferForFile(StringRef Filename) {
454 if (FileSystemOpts.WorkingDir.empty())
455 return FS->getBufferForFile(Filename);
457 SmallString<128> FilePath(Filename);
458 FixupRelativePath(FilePath);
459 return FS->getBufferForFile(FilePath.c_str());
462 /// getStatValue - Get the 'stat' information for the specified path,
463 /// using the cache to accelerate it if possible. This returns true
464 /// if the path points to a virtual file or does not exist, or returns
465 /// false if it's an existent real file. If FileDescriptor is NULL,
466 /// do directory look-up instead of file look-up.
467 bool FileManager::getStatValue(StringRef Path, FileData &Data, bool isFile,
468 std::unique_ptr<vfs::File> *F) {
469 // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be
471 if (FileSystemOpts.WorkingDir.empty())
472 return FileSystemStatCache::get(Path, Data, isFile, F,StatCache.get(), *FS);
474 SmallString<128> FilePath(Path);
475 FixupRelativePath(FilePath);
477 return FileSystemStatCache::get(FilePath.c_str(), Data, isFile, F,
478 StatCache.get(), *FS);
481 bool FileManager::getNoncachedStatValue(StringRef Path,
482 vfs::Status &Result) {
483 SmallString<128> FilePath(Path);
484 FixupRelativePath(FilePath);
486 llvm::ErrorOr<vfs::Status> S = FS->status(FilePath.c_str());
493 void FileManager::invalidateCache(const FileEntry *Entry) {
494 assert(Entry && "Cannot invalidate a NULL FileEntry");
496 SeenFileEntries.erase(Entry->getName());
498 // FileEntry invalidation should not block future optimizations in the file
499 // caches. Possible alternatives are cache truncation (invalidate last N) or
500 // invalidation of the whole cache.
501 UniqueRealFiles.erase(Entry->getUniqueID());
504 void FileManager::GetUniqueIDMapping(
505 SmallVectorImpl<const FileEntry *> &UIDToFiles) const {
507 UIDToFiles.resize(NextFileUID);
510 for (llvm::StringMap<FileEntry*, llvm::BumpPtrAllocator>::const_iterator
511 FE = SeenFileEntries.begin(), FEEnd = SeenFileEntries.end();
513 if (FE->getValue() && FE->getValue() != NON_EXISTENT_FILE)
514 UIDToFiles[FE->getValue()->getUID()] = FE->getValue();
516 // Map virtual file entries
517 for (const auto &VFE : VirtualFileEntries)
518 if (VFE && VFE.get() != NON_EXISTENT_FILE)
519 UIDToFiles[VFE->getUID()] = VFE.get();
522 void FileManager::modifyFileEntry(FileEntry *File,
523 off_t Size, time_t ModificationTime) {
525 File->ModTime = ModificationTime;
528 StringRef FileManager::getCanonicalName(const DirectoryEntry *Dir) {
529 // FIXME: use llvm::sys::fs::canonical() when it gets implemented
530 llvm::DenseMap<const DirectoryEntry *, llvm::StringRef>::iterator Known
531 = CanonicalDirNames.find(Dir);
532 if (Known != CanonicalDirNames.end())
533 return Known->second;
535 StringRef CanonicalName(Dir->getName());
538 char CanonicalNameBuf[PATH_MAX];
539 if (realpath(Dir->getName().str().c_str(), CanonicalNameBuf))
540 CanonicalName = StringRef(CanonicalNameBuf).copy(CanonicalNameStorage);
542 SmallString<256> CanonicalNameBuf(CanonicalName);
543 llvm::sys::fs::make_absolute(CanonicalNameBuf);
544 llvm::sys::path::native(CanonicalNameBuf);
545 // We've run into needing to remove '..' here in the wild though, so
547 // On Windows, symlinks are significantly less prevalent, so removing
548 // '..' is pretty safe.
549 // Ideally we'd have an equivalent of `realpath` and could implement
550 // sys::fs::canonical across all the platforms.
551 llvm::sys::path::remove_dots(CanonicalNameBuf, /* remove_dot_dot */ true);
552 CanonicalName = StringRef(CanonicalNameBuf).copy(CanonicalNameStorage);
555 CanonicalDirNames.insert(std::make_pair(Dir, CanonicalName));
556 return CanonicalName;
559 void FileManager::PrintStats() const {
560 llvm::errs() << "\n*** File Manager Stats:\n";
561 llvm::errs() << UniqueRealFiles.size() << " real files found, "
562 << UniqueRealDirs.size() << " real dirs found.\n";
563 llvm::errs() << VirtualFileEntries.size() << " virtual files found, "
564 << VirtualDirectoryEntries.size() << " virtual dirs found.\n";
565 llvm::errs() << NumDirLookups << " dir lookups, "
566 << NumDirCacheMisses << " dir cache misses.\n";
567 llvm::errs() << NumFileLookups << " file lookups, "
568 << NumFileCacheMisses << " file cache misses.\n";
570 //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups;