1 //===--- FileManager.h - File System Probing and Caching --------*- C++ -*-===//
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 /// \file
11 /// \brief Defines the clang::FileManager interface and associated types.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #ifndef LLVM_CLANG_BASIC_FILEMANAGER_H
16 #define LLVM_CLANG_BASIC_FILEMANAGER_H
17 
18 #include "clang/Basic/FileSystemOptions.h"
19 #include "clang/Basic/LLVM.h"
20 #include "clang/Basic/VirtualFileSystem.h"
21 #include "llvm/ADT/DenseMap.h"
22 #include "llvm/ADT/IntrusiveRefCntPtr.h"
23 #include "llvm/ADT/SmallVector.h"
24 #include "llvm/ADT/StringMap.h"
25 #include "llvm/ADT/StringRef.h"
26 #include "llvm/Support/Allocator.h"
27 #include <memory>
28 #include <map>
29 
30 namespace llvm {
31 class MemoryBuffer;
32 }
33 
34 namespace clang {
35 class FileManager;
36 class FileSystemStatCache;
37 
38 /// \brief Cached information about one directory (either on disk or in
39 /// the virtual file system).
40 class DirectoryEntry {
41   const char *Name;   // Name of the directory.
42   friend class FileManager;
43 public:
DirectoryEntry()44   DirectoryEntry() : Name(nullptr) {}
getName()45   const char *getName() const { return Name; }
46 };
47 
48 /// \brief Cached information about one file (either on disk
49 /// or in the virtual file system).
50 ///
51 /// If the 'File' member is valid, then this FileEntry has an open file
52 /// descriptor for the file.
53 class FileEntry {
54   const char *Name;           // Name of the file.
55   off_t Size;                 // File size in bytes.
56   time_t ModTime;             // Modification time of file.
57   const DirectoryEntry *Dir;  // Directory file lives in.
58   unsigned UID;               // A unique (small) ID for the file.
59   llvm::sys::fs::UniqueID UniqueID;
60   bool IsNamedPipe;
61   bool InPCH;
62   bool IsValid;               // Is this \c FileEntry initialized and valid?
63 
64   /// \brief The open file, if it is owned by the \p FileEntry.
65   mutable std::unique_ptr<vfs::File> File;
66   friend class FileManager;
67 
68   void operator=(const FileEntry &) = delete;
69 
70 public:
FileEntry()71   FileEntry()
72       : UniqueID(0, 0), IsNamedPipe(false), InPCH(false), IsValid(false)
73   {}
74 
75   // FIXME: this is here to allow putting FileEntry in std::map.  Once we have
76   // emplace, we shouldn't need a copy constructor anymore.
77   /// Intentionally does not copy fields that are not set in an uninitialized
78   /// \c FileEntry.
FileEntry(const FileEntry & FE)79   FileEntry(const FileEntry &FE) : UniqueID(FE.UniqueID),
80       IsNamedPipe(FE.IsNamedPipe), InPCH(FE.InPCH), IsValid(FE.IsValid) {
81     assert(!isValid() && "Cannot copy an initialized FileEntry");
82   }
83 
getName()84   const char *getName() const { return Name; }
isValid()85   bool isValid() const { return IsValid; }
getSize()86   off_t getSize() const { return Size; }
getUID()87   unsigned getUID() const { return UID; }
getUniqueID()88   const llvm::sys::fs::UniqueID &getUniqueID() const { return UniqueID; }
isInPCH()89   bool isInPCH() const { return InPCH; }
getModificationTime()90   time_t getModificationTime() const { return ModTime; }
91 
92   /// \brief Return the directory the file lives in.
getDir()93   const DirectoryEntry *getDir() const { return Dir; }
94 
95   bool operator<(const FileEntry &RHS) const { return UniqueID < RHS.UniqueID; }
96 
97   /// \brief Check whether the file is a named pipe (and thus can't be opened by
98   /// the native FileManager methods).
isNamedPipe()99   bool isNamedPipe() const { return IsNamedPipe; }
100 
closeFile()101   void closeFile() const {
102     File.reset(); // rely on destructor to close File
103   }
104 };
105 
106 struct FileData;
107 
108 /// \brief Implements support for file system lookup, file system caching,
109 /// and directory search management.
110 ///
111 /// This also handles more advanced properties, such as uniquing files based
112 /// on "inode", so that a file with two names (e.g. symlinked) will be treated
113 /// as a single file.
114 ///
115 class FileManager : public RefCountedBase<FileManager> {
116   IntrusiveRefCntPtr<vfs::FileSystem> FS;
117   FileSystemOptions FileSystemOpts;
118 
119   /// \brief Cache for existing real directories.
120   std::map<llvm::sys::fs::UniqueID, DirectoryEntry> UniqueRealDirs;
121 
122   /// \brief Cache for existing real files.
123   std::map<llvm::sys::fs::UniqueID, FileEntry> UniqueRealFiles;
124 
125   /// \brief The virtual directories that we have allocated.
126   ///
127   /// For each virtual file (e.g. foo/bar/baz.cpp), we add all of its parent
128   /// directories (foo/ and foo/bar/) here.
129   SmallVector<std::unique_ptr<DirectoryEntry>, 4> VirtualDirectoryEntries;
130   /// \brief The virtual files that we have allocated.
131   SmallVector<std::unique_ptr<FileEntry>, 4> VirtualFileEntries;
132 
133   /// \brief A cache that maps paths to directory entries (either real or
134   /// virtual) we have looked up
135   ///
136   /// The actual Entries for real directories/files are
137   /// owned by UniqueRealDirs/UniqueRealFiles above, while the Entries
138   /// for virtual directories/files are owned by
139   /// VirtualDirectoryEntries/VirtualFileEntries above.
140   ///
141   llvm::StringMap<DirectoryEntry*, llvm::BumpPtrAllocator> SeenDirEntries;
142 
143   /// \brief A cache that maps paths to file entries (either real or
144   /// virtual) we have looked up.
145   ///
146   /// \see SeenDirEntries
147   llvm::StringMap<FileEntry*, llvm::BumpPtrAllocator> SeenFileEntries;
148 
149   /// \brief The canonical names of directories.
150   llvm::DenseMap<const DirectoryEntry *, llvm::StringRef> CanonicalDirNames;
151 
152   /// \brief Storage for canonical names that we have computed.
153   llvm::BumpPtrAllocator CanonicalNameStorage;
154 
155   /// \brief Each FileEntry we create is assigned a unique ID #.
156   ///
157   unsigned NextFileUID;
158 
159   // Statistics.
160   unsigned NumDirLookups, NumFileLookups;
161   unsigned NumDirCacheMisses, NumFileCacheMisses;
162 
163   // Caching.
164   std::unique_ptr<FileSystemStatCache> StatCache;
165 
166   bool getStatValue(const char *Path, FileData &Data, bool isFile,
167                     std::unique_ptr<vfs::File> *F);
168 
169   /// Add all ancestors of the given path (pointing to either a file
170   /// or a directory) as virtual directories.
171   void addAncestorsAsVirtualDirs(StringRef Path);
172 
173 public:
174   FileManager(const FileSystemOptions &FileSystemOpts,
175               IntrusiveRefCntPtr<vfs::FileSystem> FS = nullptr);
176   ~FileManager();
177 
178   /// \brief Installs the provided FileSystemStatCache object within
179   /// the FileManager.
180   ///
181   /// Ownership of this object is transferred to the FileManager.
182   ///
183   /// \param statCache the new stat cache to install. Ownership of this
184   /// object is transferred to the FileManager.
185   ///
186   /// \param AtBeginning whether this new stat cache must be installed at the
187   /// beginning of the chain of stat caches. Otherwise, it will be added to
188   /// the end of the chain.
189   void addStatCache(std::unique_ptr<FileSystemStatCache> statCache,
190                     bool AtBeginning = false);
191 
192   /// \brief Removes the specified FileSystemStatCache object from the manager.
193   void removeStatCache(FileSystemStatCache *statCache);
194 
195   /// \brief Removes all FileSystemStatCache objects from the manager.
196   void clearStatCaches();
197 
198   /// \brief Lookup, cache, and verify the specified directory (real or
199   /// virtual).
200   ///
201   /// This returns NULL if the directory doesn't exist.
202   ///
203   /// \param CacheFailure If true and the file does not exist, we'll cache
204   /// the failure to find this file.
205   const DirectoryEntry *getDirectory(StringRef DirName,
206                                      bool CacheFailure = true);
207 
208   /// \brief Lookup, cache, and verify the specified file (real or
209   /// virtual).
210   ///
211   /// This returns NULL if the file doesn't exist.
212   ///
213   /// \param OpenFile if true and the file exists, it will be opened.
214   ///
215   /// \param CacheFailure If true and the file does not exist, we'll cache
216   /// the failure to find this file.
217   const FileEntry *getFile(StringRef Filename, bool OpenFile = false,
218                            bool CacheFailure = true);
219 
220   /// \brief Returns the current file system options
getFileSystemOpts()221   FileSystemOptions &getFileSystemOpts() { return FileSystemOpts; }
getFileSystemOpts()222   const FileSystemOptions &getFileSystemOpts() const { return FileSystemOpts; }
223 
getVirtualFileSystem()224   IntrusiveRefCntPtr<vfs::FileSystem> getVirtualFileSystem() const {
225     return FS;
226   }
227 
228   /// \brief Retrieve a file entry for a "virtual" file that acts as
229   /// if there were a file with the given name on disk.
230   ///
231   /// The file itself is not accessed.
232   const FileEntry *getVirtualFile(StringRef Filename, off_t Size,
233                                   time_t ModificationTime);
234 
235   /// \brief Open the specified file as a MemoryBuffer, returning a new
236   /// MemoryBuffer if successful, otherwise returning null.
237   llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
238   getBufferForFile(const FileEntry *Entry, bool isVolatile = false,
239                    bool ShouldCloseOpenFile = true);
240   llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
241   getBufferForFile(StringRef Filename);
242 
243   /// \brief Get the 'stat' information for the given \p Path.
244   ///
245   /// If the path is relative, it will be resolved against the WorkingDir of the
246   /// FileManager's FileSystemOptions.
247   ///
248   /// \returns false on success, true on error.
249   bool getNoncachedStatValue(StringRef Path,
250                              vfs::Status &Result);
251 
252   /// \brief Remove the real file \p Entry from the cache.
253   void invalidateCache(const FileEntry *Entry);
254 
255   /// \brief If path is not absolute and FileSystemOptions set the working
256   /// directory, the path is modified to be relative to the given
257   /// working directory.
258   /// \returns true if \c path changed.
259   bool FixupRelativePath(SmallVectorImpl<char> &path) const;
260 
261   /// Makes \c Path absolute taking into account FileSystemOptions and the
262   /// working directory option.
263   /// \returns true if \c Path changed to absolute.
264   bool makeAbsolutePath(SmallVectorImpl<char> &Path) const;
265 
266   /// \brief Produce an array mapping from the unique IDs assigned to each
267   /// file to the corresponding FileEntry pointer.
268   void GetUniqueIDMapping(
269                     SmallVectorImpl<const FileEntry *> &UIDToFiles) const;
270 
271   /// \brief Modifies the size and modification time of a previously created
272   /// FileEntry. Use with caution.
273   static void modifyFileEntry(FileEntry *File, off_t Size,
274                               time_t ModificationTime);
275 
276   /// \brief Retrieve the canonical name for a given directory.
277   ///
278   /// This is a very expensive operation, despite its results being cached,
279   /// and should only be used when the physical layout of the file system is
280   /// required, which is (almost) never.
281   StringRef getCanonicalName(const DirectoryEntry *Dir);
282 
283   void PrintStats() const;
284 };
285 
286 }  // end namespace clang
287 
288 #endif
289