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/ADT/StringExtras.h"
24 #include "llvm/Support/FileSystem.h"
25 #include "llvm/Support/MemoryBuffer.h"
26 #include "llvm/Support/raw_ostream.h"
27 #include "llvm/Support/Path.h"
28 #include "llvm/Support/system_error.h"
29 #include "llvm/Config/config.h"
34 // FIXME: This is terrible, we need this for ::close.
35 #if !defined(_MSC_VER) && !defined(__MINGW32__)
41 using namespace clang
;
43 // FIXME: Enhance libsystem to support inode and other fields.
46 /// NON_EXISTENT_DIR - A special value distinct from null that is used to
47 /// represent a dir name that doesn't exist on the disk.
48 #define NON_EXISTENT_DIR reinterpret_cast<DirectoryEntry*>((intptr_t)-1)
50 /// NON_EXISTENT_FILE - A special value distinct from null that is used to
51 /// represent a filename that doesn't exist on the disk.
52 #define NON_EXISTENT_FILE reinterpret_cast<FileEntry*>((intptr_t)-1)
55 FileEntry::~FileEntry() {
56 // If this FileEntry owns an open file descriptor that never got used, close
58 if (FD
!= -1) ::close(FD
);
61 //===----------------------------------------------------------------------===//
63 //===----------------------------------------------------------------------===//
68 static std::string
GetFullPath(const char *relPath
) {
69 char *absPathStrPtr
= _fullpath(NULL
, relPath
, 0);
70 assert(absPathStrPtr
&& "_fullpath() returned NULL!");
72 std::string
absPath(absPathStrPtr
);
79 class FileManager::UniqueDirContainer
{
80 /// UniqueDirs - Cache from full path to existing directories/files.
82 llvm::StringMap
<DirectoryEntry
> UniqueDirs
;
85 /// getDirectory - Return an existing DirectoryEntry with the given
86 /// name if there is already one; otherwise create and return a
87 /// default-constructed DirectoryEntry.
88 DirectoryEntry
&getDirectory(const char *Name
,
89 const struct stat
& /*StatBuf*/) {
90 std::string
FullPath(GetFullPath(Name
));
91 return UniqueDirs
.GetOrCreateValue(FullPath
).getValue();
94 size_t size() const { return UniqueDirs
.size(); }
97 class FileManager::UniqueFileContainer
{
98 /// UniqueFiles - Cache from full path to existing directories/files.
100 llvm::StringMap
<FileEntry
, llvm::BumpPtrAllocator
> UniqueFiles
;
103 /// getFile - Return an existing FileEntry with the given name if
104 /// there is already one; otherwise create and return a
105 /// default-constructed FileEntry.
106 FileEntry
&getFile(const char *Name
, const struct stat
& /*StatBuf*/) {
107 std::string
FullPath(GetFullPath(Name
));
109 // LowercaseString because Windows filesystem is case insensitive.
110 FullPath
= llvm::LowercaseString(FullPath
);
111 return UniqueFiles
.GetOrCreateValue(FullPath
).getValue();
114 size_t size() const { return UniqueFiles
.size(); }
117 //===----------------------------------------------------------------------===//
118 // Unix-like Systems.
119 //===----------------------------------------------------------------------===//
123 class FileManager::UniqueDirContainer
{
124 /// UniqueDirs - Cache from ID's to existing directories/files.
125 std::map
<std::pair
<dev_t
, ino_t
>, DirectoryEntry
> UniqueDirs
;
128 /// getDirectory - Return an existing DirectoryEntry with the given
129 /// ID's if there is already one; otherwise create and return a
130 /// default-constructed DirectoryEntry.
131 DirectoryEntry
&getDirectory(const char * /*Name*/,
132 const struct stat
&StatBuf
) {
133 return UniqueDirs
[std::make_pair(StatBuf
.st_dev
, StatBuf
.st_ino
)];
136 size_t size() const { return UniqueDirs
.size(); }
139 class FileManager::UniqueFileContainer
{
140 /// UniqueFiles - Cache from ID's to existing directories/files.
141 std::set
<FileEntry
> UniqueFiles
;
144 /// getFile - Return an existing FileEntry with the given ID's if
145 /// there is already one; otherwise create and return a
146 /// default-constructed FileEntry.
147 FileEntry
&getFile(const char * /*Name*/, const struct stat
&StatBuf
) {
149 const_cast<FileEntry
&>(
150 *UniqueFiles
.insert(FileEntry(StatBuf
.st_dev
,
152 StatBuf
.st_mode
)).first
);
155 size_t size() const { return UniqueFiles
.size(); }
160 //===----------------------------------------------------------------------===//
162 //===----------------------------------------------------------------------===//
164 FileManager::FileManager(const FileSystemOptions
&FSO
)
165 : FileSystemOpts(FSO
),
166 UniqueRealDirs(*new UniqueDirContainer()),
167 UniqueRealFiles(*new UniqueFileContainer()),
168 SeenDirEntries(64), SeenFileEntries(64), NextFileUID(0) {
169 NumDirLookups
= NumFileLookups
= 0;
170 NumDirCacheMisses
= NumFileCacheMisses
= 0;
173 FileManager::~FileManager() {
174 delete &UniqueRealDirs
;
175 delete &UniqueRealFiles
;
176 for (unsigned i
= 0, e
= VirtualFileEntries
.size(); i
!= e
; ++i
)
177 delete VirtualFileEntries
[i
];
178 for (unsigned i
= 0, e
= VirtualDirectoryEntries
.size(); i
!= e
; ++i
)
179 delete VirtualDirectoryEntries
[i
];
182 void FileManager::addStatCache(FileSystemStatCache
*statCache
,
184 assert(statCache
&& "No stat cache provided?");
185 if (AtBeginning
|| StatCache
.get() == 0) {
186 statCache
->setNextStatCache(StatCache
.take());
187 StatCache
.reset(statCache
);
191 FileSystemStatCache
*LastCache
= StatCache
.get();
192 while (LastCache
->getNextStatCache())
193 LastCache
= LastCache
->getNextStatCache();
195 LastCache
->setNextStatCache(statCache
);
198 void FileManager::removeStatCache(FileSystemStatCache
*statCache
) {
202 if (StatCache
.get() == statCache
) {
203 // This is the first stat cache.
204 StatCache
.reset(StatCache
->takeNextStatCache());
208 // Find the stat cache in the list.
209 FileSystemStatCache
*PrevCache
= StatCache
.get();
210 while (PrevCache
&& PrevCache
->getNextStatCache() != statCache
)
211 PrevCache
= PrevCache
->getNextStatCache();
213 assert(PrevCache
&& "Stat cache not found for removal");
214 PrevCache
->setNextStatCache(statCache
->getNextStatCache());
217 /// \brief Retrieve the directory that the given file name resides in.
218 /// Filename can point to either a real file or a virtual file.
219 static const DirectoryEntry
*getDirectoryFromFile(FileManager
&FileMgr
,
220 llvm::StringRef Filename
) {
221 if (Filename
.empty())
224 if (llvm::sys::path::is_separator(Filename
[Filename
.size() - 1]))
225 return NULL
; // If Filename is a directory.
227 llvm::StringRef DirName
= llvm::sys::path::parent_path(Filename
);
228 // Use the current directory if file has no path component.
232 return FileMgr
.getDirectory(DirName
);
235 /// Add all ancestors of the given path (pointing to either a file or
236 /// a directory) as virtual directories.
237 void FileManager::addAncestorsAsVirtualDirs(llvm::StringRef Path
) {
238 llvm::StringRef DirName
= llvm::sys::path::parent_path(Path
);
242 llvm::StringMapEntry
<DirectoryEntry
*> &NamedDirEnt
=
243 SeenDirEntries
.GetOrCreateValue(DirName
);
245 // When caching a virtual directory, we always cache its ancestors
246 // at the same time. Therefore, if DirName is already in the cache,
247 // we don't need to recurse as its ancestors must also already be in
249 if (NamedDirEnt
.getValue())
252 // Add the virtual directory to the cache.
253 DirectoryEntry
*UDE
= new DirectoryEntry
;
254 UDE
->Name
= NamedDirEnt
.getKeyData();
255 NamedDirEnt
.setValue(UDE
);
256 VirtualDirectoryEntries
.push_back(UDE
);
258 // Recursively add the other ancestors.
259 addAncestorsAsVirtualDirs(DirName
);
262 /// getDirectory - Lookup, cache, and verify the specified directory
263 /// (real or virtual). This returns NULL if the directory doesn't
266 const DirectoryEntry
*FileManager::getDirectory(llvm::StringRef DirName
) {
267 // stat doesn't like trailing separators (at least on Windows).
268 if (DirName
.size() > 1 && llvm::sys::path::is_separator(DirName
.back()))
269 DirName
= DirName
.substr(0, DirName
.size()-1);
272 llvm::StringMapEntry
<DirectoryEntry
*> &NamedDirEnt
=
273 SeenDirEntries
.GetOrCreateValue(DirName
);
275 // See if there was already an entry in the map. Note that the map
276 // contains both virtual and real directories.
277 if (NamedDirEnt
.getValue())
278 return NamedDirEnt
.getValue() == NON_EXISTENT_DIR
279 ? 0 : NamedDirEnt
.getValue();
283 // By default, initialize it to invalid.
284 NamedDirEnt
.setValue(NON_EXISTENT_DIR
);
286 // Get the null-terminated directory name as stored as the key of the
287 // SeenDirEntries map.
288 const char *InterndDirName
= NamedDirEnt
.getKeyData();
290 // Check to see if the directory exists.
292 if (getStatValue(InterndDirName
, StatBuf
, 0/*directory lookup*/)) {
293 // There's no real directory at the given path.
297 // It exists. See if we have already opened a directory with the
298 // same inode (this occurs on Unix-like systems when one dir is
299 // symlinked to another, for example) or the same path (on
301 DirectoryEntry
&UDE
= UniqueRealDirs
.getDirectory(InterndDirName
, StatBuf
);
303 NamedDirEnt
.setValue(&UDE
);
304 if (!UDE
.getName()) {
305 // We don't have this directory yet, add it. We use the string
306 // key from the SeenDirEntries map as the string.
307 UDE
.Name
= InterndDirName
;
313 /// getFile - Lookup, cache, and verify the specified file (real or
314 /// virtual). This returns NULL if the file doesn't exist.
316 const FileEntry
*FileManager::getFile(llvm::StringRef Filename
) {
319 // See if there is already an entry in the map.
320 llvm::StringMapEntry
<FileEntry
*> &NamedFileEnt
=
321 SeenFileEntries
.GetOrCreateValue(Filename
);
323 // See if there is already an entry in the map.
324 if (NamedFileEnt
.getValue())
325 return NamedFileEnt
.getValue() == NON_EXISTENT_FILE
326 ? 0 : NamedFileEnt
.getValue();
328 ++NumFileCacheMisses
;
330 // By default, initialize it to invalid.
331 NamedFileEnt
.setValue(NON_EXISTENT_FILE
);
333 // Get the null-terminated file name as stored as the key of the
334 // SeenFileEntries map.
335 const char *InterndFileName
= NamedFileEnt
.getKeyData();
337 // Look up the directory for the file. When looking up something like
338 // sys/foo.h we'll discover all of the search directories that have a 'sys'
339 // subdirectory. This will let us avoid having to waste time on known-to-fail
340 // searches when we go to find sys/bar.h, because all the search directories
341 // without a 'sys' subdir will get a cached failure result.
342 const DirectoryEntry
*DirInfo
= getDirectoryFromFile(*this, Filename
);
343 if (DirInfo
== 0) // Directory doesn't exist, file can't exist.
346 // FIXME: Use the directory info to prune this, before doing the stat syscall.
347 // FIXME: This will reduce the # syscalls.
349 // Nope, there isn't. Check to see if the file exists.
350 int FileDescriptor
= -1;
352 if (getStatValue(InterndFileName
, StatBuf
, &FileDescriptor
)) {
353 // There's no real file at the given path.
357 // It exists. See if we have already opened a file with the same inode.
358 // This occurs when one dir is symlinked to another, for example.
359 FileEntry
&UFE
= UniqueRealFiles
.getFile(InterndFileName
, StatBuf
);
361 NamedFileEnt
.setValue(&UFE
);
362 if (UFE
.getName()) { // Already have an entry with this inode, return it.
363 // If the stat process opened the file, close it to avoid a FD leak.
364 if (FileDescriptor
!= -1)
365 close(FileDescriptor
);
370 // Otherwise, we don't have this directory yet, add it.
371 // FIXME: Change the name to be a char* that points back to the
372 // 'SeenFileEntries' key.
373 UFE
.Name
= InterndFileName
;
374 UFE
.Size
= StatBuf
.st_size
;
375 UFE
.ModTime
= StatBuf
.st_mtime
;
377 UFE
.UID
= NextFileUID
++;
378 UFE
.FD
= FileDescriptor
;
383 FileManager::getVirtualFile(llvm::StringRef Filename
, off_t Size
,
384 time_t ModificationTime
) {
387 // See if there is already an entry in the map.
388 llvm::StringMapEntry
<FileEntry
*> &NamedFileEnt
=
389 SeenFileEntries
.GetOrCreateValue(Filename
);
391 // See if there is already an entry in the map.
392 if (NamedFileEnt
.getValue() && NamedFileEnt
.getValue() != NON_EXISTENT_FILE
)
393 return NamedFileEnt
.getValue();
395 ++NumFileCacheMisses
;
397 // By default, initialize it to invalid.
398 NamedFileEnt
.setValue(NON_EXISTENT_FILE
);
400 addAncestorsAsVirtualDirs(Filename
);
403 // Now that all ancestors of Filename are in the cache, the
404 // following call is guaranteed to find the DirectoryEntry from the
406 const DirectoryEntry
*DirInfo
= getDirectoryFromFile(*this, Filename
);
408 "The directory of a virtual file should already be in the cache.");
410 // Check to see if the file exists. If so, drop the virtual file
411 int FileDescriptor
= -1;
413 const char *InterndFileName
= NamedFileEnt
.getKeyData();
414 if (getStatValue(InterndFileName
, StatBuf
, &FileDescriptor
) == 0) {
415 // If the stat process opened the file, close it to avoid a FD leak.
416 if (FileDescriptor
!= -1)
417 close(FileDescriptor
);
419 StatBuf
.st_size
= Size
;
420 StatBuf
.st_mtime
= ModificationTime
;
421 UFE
= &UniqueRealFiles
.getFile(InterndFileName
, StatBuf
);
423 NamedFileEnt
.setValue(UFE
);
425 // If we had already opened this file, close it now so we don't
426 // leak the descriptor. We're not going to use the file
427 // descriptor anyway, since this is a virtual file.
433 // If we already have an entry with this inode, return it.
439 UFE
= new FileEntry();
440 VirtualFileEntries
.push_back(UFE
);
441 NamedFileEnt
.setValue(UFE
);
444 UFE
->Name
= InterndFileName
;
446 UFE
->ModTime
= ModificationTime
;
448 UFE
->UID
= NextFileUID
++;
453 void FileManager::FixupRelativePath(llvm::sys::Path
&path
,
454 const FileSystemOptions
&FSOpts
) {
455 if (FSOpts
.WorkingDir
.empty() || llvm::sys::path::is_absolute(path
.str()))
458 llvm::SmallString
<128> NewPath(FSOpts
.WorkingDir
);
459 llvm::sys::path::append(NewPath
, path
.str());
463 llvm::MemoryBuffer
*FileManager::
464 getBufferForFile(const FileEntry
*Entry
, std::string
*ErrorStr
) {
465 llvm::OwningPtr
<llvm::MemoryBuffer
> Result
;
467 if (FileSystemOpts
.WorkingDir
.empty()) {
468 const char *Filename
= Entry
->getName();
469 // If the file is already open, use the open file descriptor.
470 if (Entry
->FD
!= -1) {
471 ec
= llvm::MemoryBuffer::getOpenFile(Entry
->FD
, Filename
, Result
,
474 *ErrorStr
= ec
.message();
478 return Result
.take();
481 // Otherwise, open the file.
482 ec
= llvm::MemoryBuffer::getFile(Filename
, Result
, Entry
->getSize());
484 *ErrorStr
= ec
.message();
485 return Result
.take();
488 llvm::sys::Path
FilePath(Entry
->getName());
489 FixupRelativePath(FilePath
, FileSystemOpts
);
490 ec
= llvm::MemoryBuffer::getFile(FilePath
.c_str(), Result
, Entry
->getSize());
492 *ErrorStr
= ec
.message();
493 return Result
.take();
496 llvm::MemoryBuffer
*FileManager::
497 getBufferForFile(llvm::StringRef Filename
, std::string
*ErrorStr
) {
498 llvm::OwningPtr
<llvm::MemoryBuffer
> Result
;
500 if (FileSystemOpts
.WorkingDir
.empty()) {
501 ec
= llvm::MemoryBuffer::getFile(Filename
, Result
);
503 *ErrorStr
= ec
.message();
504 return Result
.take();
507 llvm::sys::Path
FilePath(Filename
);
508 FixupRelativePath(FilePath
, FileSystemOpts
);
509 ec
= llvm::MemoryBuffer::getFile(FilePath
.c_str(), Result
);
511 *ErrorStr
= ec
.message();
512 return Result
.take();
515 /// getStatValue - Get the 'stat' information for the specified path,
516 /// using the cache to accelerate it if possible. This returns true
517 /// if the path points to a virtual file or does not exist, or returns
518 /// false if it's an existent real file. If FileDescriptor is NULL,
519 /// do directory look-up instead of file look-up.
520 bool FileManager::getStatValue(const char *Path
, struct stat
&StatBuf
,
521 int *FileDescriptor
) {
522 // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be
524 if (FileSystemOpts
.WorkingDir
.empty())
525 return FileSystemStatCache::get(Path
, StatBuf
, FileDescriptor
,
528 llvm::sys::Path
FilePath(Path
);
529 FixupRelativePath(FilePath
, FileSystemOpts
);
531 return FileSystemStatCache::get(FilePath
.c_str(), StatBuf
, FileDescriptor
,
535 void FileManager::GetUniqueIDMapping(
536 llvm::SmallVectorImpl
<const FileEntry
*> &UIDToFiles
) const {
538 UIDToFiles
.resize(NextFileUID
);
541 for (llvm::StringMap
<FileEntry
*, llvm::BumpPtrAllocator
>::const_iterator
542 FE
= SeenFileEntries
.begin(), FEEnd
= SeenFileEntries
.end();
544 if (FE
->getValue() && FE
->getValue() != NON_EXISTENT_FILE
)
545 UIDToFiles
[FE
->getValue()->getUID()] = FE
->getValue();
547 // Map virtual file entries
548 for (llvm::SmallVector
<FileEntry
*, 4>::const_iterator
549 VFE
= VirtualFileEntries
.begin(), VFEEnd
= VirtualFileEntries
.end();
550 VFE
!= VFEEnd
; ++VFE
)
551 if (*VFE
&& *VFE
!= NON_EXISTENT_FILE
)
552 UIDToFiles
[(*VFE
)->getUID()] = *VFE
;
556 void FileManager::PrintStats() const {
557 llvm::errs() << "\n*** File Manager Stats:\n";
558 llvm::errs() << UniqueRealFiles
.size() << " real files found, "
559 << UniqueRealDirs
.size() << " real dirs found.\n";
560 llvm::errs() << VirtualFileEntries
.size() << " virtual files found, "
561 << VirtualDirectoryEntries
.size() << " virtual dirs found.\n";
562 llvm::errs() << NumDirLookups
<< " dir lookups, "
563 << NumDirCacheMisses
<< " dir cache misses.\n";
564 llvm::errs() << NumFileLookups
<< " file lookups, "
565 << NumFileCacheMisses
<< " file cache misses.\n";
567 //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups;