1//===--- FileManager.h - File System Probing and Caching --------*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8///
9/// \file
10/// Defines the clang::FileManager interface and associated types.
11///
12//===----------------------------------------------------------------------===//
13
14#ifndef LLVM_CLANG_BASIC_FILEMANAGER_H
15#define LLVM_CLANG_BASIC_FILEMANAGER_H
16
17#include "clang/Basic/FileSystemOptions.h"
18#include "clang/Basic/LLVM.h"
19#include "llvm/ADT/DenseMap.h"
20#include "llvm/ADT/IntrusiveRefCntPtr.h"
21#include "llvm/ADT/PointerUnion.h"
22#include "llvm/ADT/SmallVector.h"
23#include "llvm/ADT/StringMap.h"
24#include "llvm/ADT/StringRef.h"
25#include "llvm/Support/Allocator.h"
26#include "llvm/Support/ErrorOr.h"
27#include "llvm/Support/FileSystem.h"
28#include "llvm/Support/VirtualFileSystem.h"
29#include <ctime>
30#include <map>
31#include <memory>
32#include <string>
33
34namespace llvm {
35
36class MemoryBuffer;
37
38} // end namespace llvm
39
40namespace clang {
41
42class FileSystemStatCache;
43
44/// Cached information about one directory (either on disk or in
45/// the virtual file system).
46class DirectoryEntry {
47 friend class FileManager;
48
49 // FIXME: We should not be storing a directory entry name here.
50 StringRef Name; // Name of the directory.
51
52public:
53 StringRef getName() const { return Name; }
54};
55
56/// A reference to a \c DirectoryEntry that includes the name of the directory
57/// as it was accessed by the FileManager's client.
58class DirectoryEntryRef {
59public:
60 const DirectoryEntry &getDirEntry() const { return *Entry->getValue(); }
61
62 StringRef getName() const { return Entry->getKey(); }
63
64private:
65 friend class FileManager;
66
67 DirectoryEntryRef(
68 llvm::StringMapEntry<llvm::ErrorOr<DirectoryEntry &>> *Entry)
69 : Entry(Entry) {}
70
71 const llvm::StringMapEntry<llvm::ErrorOr<DirectoryEntry &>> *Entry;
72};
73
74/// Cached information about one file (either on disk
75/// or in the virtual file system).
76///
77/// If the 'File' member is valid, then this FileEntry has an open file
78/// descriptor for the file.
79class FileEntry {
80 friend class FileManager;
81
82 StringRef Name; // Name of the file.
83 std::string RealPathName; // Real path to the file; could be empty.
84 off_t Size; // File size in bytes.
85 time_t ModTime; // Modification time of file.
86 const DirectoryEntry *Dir; // Directory file lives in.
87 llvm::sys::fs::UniqueID UniqueID;
88 unsigned UID; // A unique (small) ID for the file.
89 bool IsNamedPipe;
90 bool IsValid; // Is this \c FileEntry initialized and valid?
91
92 /// The open file, if it is owned by the \p FileEntry.
93 mutable std::unique_ptr<llvm::vfs::File> File;
94
95public:
96 FileEntry()
97 : UniqueID(0, 0), IsNamedPipe(false), IsValid(false)
98 {}
99
100 FileEntry(const FileEntry &) = delete;
101 FileEntry &operator=(const FileEntry &) = delete;
102
103 StringRef getName() const { return Name; }
104 StringRef tryGetRealPathName() const { return RealPathName; }
105 bool isValid() const { return IsValid; }
106 off_t getSize() const { return Size; }
107 unsigned getUID() const { return UID; }
108 const llvm::sys::fs::UniqueID &getUniqueID() const { return UniqueID; }
109 time_t getModificationTime() const { return ModTime; }
110
111 /// Return the directory the file lives in.
112 const DirectoryEntry *getDir() const { return Dir; }
113
114 bool operator<(const FileEntry &RHS) const { return UniqueID < RHS.UniqueID; }
115
116 /// Check whether the file is a named pipe (and thus can't be opened by
117 /// the native FileManager methods).
118 bool isNamedPipe() const { return IsNamedPipe; }
119
120 void closeFile() const {
121 File.reset(); // rely on destructor to close File
122 }
123
124 // Only for use in tests to see if deferred opens are happening, rather than
125 // relying on RealPathName being empty.
126 bool isOpenForTests() const { return File != nullptr; }
127};
128
129/// A reference to a \c FileEntry that includes the name of the file as it was
130/// accessed by the FileManager's client.
131class FileEntryRef {
132public:
133 FileEntryRef() = delete;
134 FileEntryRef(StringRef Name, const FileEntry &Entry)
135 : Name(Name), Entry(&Entry) {}
136
137 const StringRef getName() const { return Name; }
138
139 bool isValid() const { return Entry->isValid(); }
140
141 const FileEntry &getFileEntry() const { return *Entry; }
142
143 off_t getSize() const { return Entry->getSize(); }
144
145 unsigned getUID() const { return Entry->getUID(); }
146
147 const llvm::sys::fs::UniqueID &getUniqueID() const {
148 return Entry->getUniqueID();
149 }
150
151 time_t getModificationTime() const { return Entry->getModificationTime(); }
152
153 friend bool operator==(const FileEntryRef &LHS, const FileEntryRef &RHS) {
154 return LHS.Entry == RHS.Entry && LHS.Name == RHS.Name;
155 }
156 friend bool operator!=(const FileEntryRef &LHS, const FileEntryRef &RHS) {
157 return !(LHS == RHS);
158 }
159
160private:
161 StringRef Name;
162 const FileEntry *Entry;
163};
164
165/// Implements support for file system lookup, file system caching,
166/// and directory search management.
167///
168/// This also handles more advanced properties, such as uniquing files based
169/// on "inode", so that a file with two names (e.g. symlinked) will be treated
170/// as a single file.
171///
172class FileManager : public RefCountedBase<FileManager> {
173 IntrusiveRefCntPtr<llvm::vfs::FileSystem> FS;
174 FileSystemOptions FileSystemOpts;
175
176 /// Cache for existing real directories.
177 std::map<llvm::sys::fs::UniqueID, DirectoryEntry> UniqueRealDirs;
178
179 /// Cache for existing real files.
180 std::map<llvm::sys::fs::UniqueID, FileEntry> UniqueRealFiles;
181
182 /// The virtual directories that we have allocated.
183 ///
184 /// For each virtual file (e.g. foo/bar/baz.cpp), we add all of its parent
185 /// directories (foo/ and foo/bar/) here.
186 SmallVector<std::unique_ptr<DirectoryEntry>, 4> VirtualDirectoryEntries;
187 /// The virtual files that we have allocated.
188 SmallVector<std::unique_ptr<FileEntry>, 4> VirtualFileEntries;
189
190 /// A set of files that bypass the maps and uniquing. They can have
191 /// conflicting filenames.
192 SmallVector<std::unique_ptr<FileEntry>, 0> BypassFileEntries;
193
194 /// A cache that maps paths to directory entries (either real or
195 /// virtual) we have looked up, or an error that occurred when we looked up
196 /// the directory.
197 ///
198 /// The actual Entries for real directories/files are
199 /// owned by UniqueRealDirs/UniqueRealFiles above, while the Entries
200 /// for virtual directories/files are owned by
201 /// VirtualDirectoryEntries/VirtualFileEntries above.
202 ///
203 llvm::StringMap<llvm::ErrorOr<DirectoryEntry &>, llvm::BumpPtrAllocator>
204 SeenDirEntries;
205
206 /// A reference to the file entry that is associated with a particular
207 /// filename, or a reference to another filename that should be looked up
208 /// instead of the accessed filename.
209 ///
210 /// The reference to another filename is specifically useful for Redirecting
211 /// VFSs that use external names. In that case, the \c FileEntryRef returned
212 /// by the \c FileManager will have the external name, and not the name that
213 /// was used to lookup the file.
214 using SeenFileEntryOrRedirect =
215 llvm::PointerUnion<FileEntry *, const StringRef *>;
216
217 /// A cache that maps paths to file entries (either real or
218 /// virtual) we have looked up, or an error that occurred when we looked up
219 /// the file.
220 ///
221 /// \see SeenDirEntries
222 llvm::StringMap<llvm::ErrorOr<SeenFileEntryOrRedirect>,
223 llvm::BumpPtrAllocator>
224 SeenFileEntries;
225
226 /// The canonical names of files and directories .
227 llvm::DenseMap<const void *, llvm::StringRef> CanonicalNames;
228
229 /// Storage for canonical names that we have computed.
230 llvm::BumpPtrAllocator CanonicalNameStorage;
231
232 /// Each FileEntry we create is assigned a unique ID #.
233 ///
234 unsigned NextFileUID;
235
236 // Caching.
237 std::unique_ptr<FileSystemStatCache> StatCache;
238
239 std::error_code getStatValue(StringRef Path, llvm::vfs::Status &Status,
240 bool isFile,
241 std::unique_ptr<llvm::vfs::File> *F);
242
243 /// Add all ancestors of the given path (pointing to either a file
244 /// or a directory) as virtual directories.
245 void addAncestorsAsVirtualDirs(StringRef Path);
246
247 /// Fills the RealPathName in file entry.
248 void fillRealPathName(FileEntry *UFE, llvm::StringRef FileName);
249
250public:
251 /// Construct a file manager, optionally with a custom VFS.
252 ///
253 /// \param FS if non-null, the VFS to use. Otherwise uses
254 /// llvm::vfs::getRealFileSystem().
255 FileManager(const FileSystemOptions &FileSystemOpts,
256 IntrusiveRefCntPtr<llvm::vfs::FileSystem> FS = nullptr);
257 ~FileManager();
258
259 /// Installs the provided FileSystemStatCache object within
260 /// the FileManager.
261 ///
262 /// Ownership of this object is transferred to the FileManager.
263 ///
264 /// \param statCache the new stat cache to install. Ownership of this
265 /// object is transferred to the FileManager.
266 void setStatCache(std::unique_ptr<FileSystemStatCache> statCache);
267
268 /// Removes the FileSystemStatCache object from the manager.
269 void clearStatCache();
270
271 /// Returns the number of unique real file entries cached by the file manager.
272 size_t getNumUniqueRealFiles() const { return UniqueRealFiles.size(); }
273
274 /// Lookup, cache, and verify the specified directory (real or
275 /// virtual).
276 ///
277 /// This returns a \c std::error_code if there was an error reading the
278 /// directory. On success, returns the reference to the directory entry
279 /// together with the exact path that was used to access a file by a
280 /// particular call to getDirectoryRef.
281 ///
282 /// \param CacheFailure If true and the file does not exist, we'll cache
283 /// the failure to find this file.
284 llvm::Expected<DirectoryEntryRef> getDirectoryRef(StringRef DirName,
285 bool CacheFailure = true);
286
287 /// Get a \c DirectoryEntryRef if it exists, without doing anything on error.
288 llvm::Optional<DirectoryEntryRef>
289 getOptionalDirectoryRef(StringRef DirName, bool CacheFailure = true) {
290 return llvm::expectedToOptional(getDirectoryRef(DirName, CacheFailure));
291 }
292
293 /// Lookup, cache, and verify the specified directory (real or
294 /// virtual).
295 ///
296 /// This function is deprecated and will be removed at some point in the
297 /// future, new clients should use
298 /// \c getDirectoryRef.
299 ///
300 /// This returns a \c std::error_code if there was an error reading the
301 /// directory. If there is no error, the DirectoryEntry is guaranteed to be
302 /// non-NULL.
303 ///
304 /// \param CacheFailure If true and the file does not exist, we'll cache
305 /// the failure to find this file.
306 llvm::ErrorOr<const DirectoryEntry *>
307 getDirectory(StringRef DirName, bool CacheFailure = true);
308
309 /// Lookup, cache, and verify the specified file (real or
310 /// virtual).
311 ///
312 /// This function is deprecated and will be removed at some point in the
313 /// future, new clients should use
314 /// \c getFileRef.
315 ///
316 /// This returns a \c std::error_code if there was an error loading the file.
317 /// If there is no error, the FileEntry is guaranteed to be non-NULL.
318 ///
319 /// \param OpenFile if true and the file exists, it will be opened.
320 ///
321 /// \param CacheFailure If true and the file does not exist, we'll cache
322 /// the failure to find this file.
323 llvm::ErrorOr<const FileEntry *>
324 getFile(StringRef Filename, bool OpenFile = false, bool CacheFailure = true);
325
326 /// Lookup, cache, and verify the specified file (real or virtual). Return the
327 /// reference to the file entry together with the exact path that was used to
328 /// access a file by a particular call to getFileRef. If the underlying VFS is
329 /// a redirecting VFS that uses external file names, the returned FileEntryRef
330 /// will use the external name instead of the filename that was passed to this
331 /// method.
332 ///
333 /// This returns a \c std::error_code if there was an error loading the file,
334 /// or a \c FileEntryRef otherwise.
335 ///
336 /// \param OpenFile if true and the file exists, it will be opened.
337 ///
338 /// \param CacheFailure If true and the file does not exist, we'll cache
339 /// the failure to find this file.
340 llvm::Expected<FileEntryRef> getFileRef(StringRef Filename,
341 bool OpenFile = false,
342 bool CacheFailure = true);
343
344 /// Get a FileEntryRef if it exists, without doing anything on error.
345 llvm::Optional<FileEntryRef> getOptionalFileRef(StringRef Filename,
346 bool OpenFile = false,
347 bool CacheFailure = true) {
348 return llvm::expectedToOptional(
349 getFileRef(Filename, OpenFile, CacheFailure));
350 }
351
352 /// Returns the current file system options
353 FileSystemOptions &getFileSystemOpts() { return FileSystemOpts; }
354 const FileSystemOptions &getFileSystemOpts() const { return FileSystemOpts; }
355
356 llvm::vfs::FileSystem &getVirtualFileSystem() const { return *FS; }
357
358 void setVirtualFileSystem(IntrusiveRefCntPtr<llvm::vfs::FileSystem> FS) {
359 this->FS = std::move(FS);
360 }
361
362 /// Retrieve a file entry for a "virtual" file that acts as
363 /// if there were a file with the given name on disk.
364 ///
365 /// The file itself is not accessed.
366 const FileEntry *getVirtualFile(StringRef Filename, off_t Size,
367 time_t ModificationTime);
368
369 /// Retrieve a FileEntry that bypasses VFE, which is expected to be a virtual
370 /// file entry, to access the real file. The returned FileEntry will have
371 /// the same filename as FE but a different identity and its own stat.
372 ///
373 /// This should be used only for rare error recovery paths because it
374 /// bypasses all mapping and uniquing, blindly creating a new FileEntry.
375 /// There is no attempt to deduplicate these; if you bypass the same file
376 /// twice, you get two new file entries.
377 llvm::Optional<FileEntryRef> getBypassFile(FileEntryRef VFE);
378
379 /// Open the specified file as a MemoryBuffer, returning a new
380 /// MemoryBuffer if successful, otherwise returning null.
381 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
382 getBufferForFile(const FileEntry *Entry, bool isVolatile = false,
383 bool RequiresNullTerminator = true);
384 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
385 getBufferForFile(StringRef Filename, bool isVolatile = false,
386 bool RequiresNullTerminator = true) {
387 return getBufferForFileImpl(Filename, /*FileSize=*/-1, isVolatile,
388 RequiresNullTerminator);
389 }
390
391private:
392 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
393 getBufferForFileImpl(StringRef Filename, int64_t FileSize, bool isVolatile,
394 bool RequiresNullTerminator);
395
396public:
397 /// Get the 'stat' information for the given \p Path.
398 ///
399 /// If the path is relative, it will be resolved against the WorkingDir of the
400 /// FileManager's FileSystemOptions.
401 ///
402 /// \returns a \c std::error_code describing an error, if there was one
403 std::error_code getNoncachedStatValue(StringRef Path,
404 llvm::vfs::Status &Result);
405
406 /// If path is not absolute and FileSystemOptions set the working
407 /// directory, the path is modified to be relative to the given
408 /// working directory.
409 /// \returns true if \c path changed.
410 bool FixupRelativePath(SmallVectorImpl<char> &path) const;
411
412 /// Makes \c Path absolute taking into account FileSystemOptions and the
413 /// working directory option.
414 /// \returns true if \c Path changed to absolute.
415 bool makeAbsolutePath(SmallVectorImpl<char> &Path) const;
416
417 /// Produce an array mapping from the unique IDs assigned to each
418 /// file to the corresponding FileEntry pointer.
419 void GetUniqueIDMapping(
420 SmallVectorImpl<const FileEntry *> &UIDToFiles) const;
421
422 /// Retrieve the canonical name for a given directory.
423 ///
424 /// This is a very expensive operation, despite its results being cached,
425 /// and should only be used when the physical layout of the file system is
426 /// required, which is (almost) never.
427 StringRef getCanonicalName(const DirectoryEntry *Dir);
428
429 /// Retrieve the canonical name for a given file.
430 ///
431 /// This is a very expensive operation, despite its results being cached,
432 /// and should only be used when the physical layout of the file system is
433 /// required, which is (almost) never.
434 StringRef getCanonicalName(const FileEntry *File);
435
436 void PrintStats() const;
437};
438
439} // end namespace clang
440
441#endif // LLVM_CLANG_BASIC_FILEMANAGER_H
442