3 * @defgroup FileBackend File backend
6 * File backend is used to interact with file storage systems,
7 * such as the local file system, NFS, or cloud storage systems.
11 * Base class for all file backends.
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
23 * You should have received a copy of the GNU General Public License along
24 * with this program; if not, write to the Free Software Foundation, Inc.,
25 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
26 * http://www.gnu.org/copyleft/gpl.html
29 * @ingroup FileBackend
30 * @author Aaron Schulz
34 * @brief Base class for all file backend classes (including multi-write backends).
36 * This class defines the methods as abstract that subclasses must implement.
37 * Outside callers can assume that all backends will have these functions.
39 * All "storage paths" are of the format "mwstore://<backend>/<container>/<path>".
40 * The <path> portion is a relative path that uses UNIX file system (FS) notation,
41 * though any particular backend may not actually be using a local filesystem.
42 * Therefore, the relative paths are only virtual.
44 * Backend contents are stored under wiki-specific container names by default.
45 * For legacy reasons, this has no effect for the FS backend class, and per-wiki
46 * segregation must be done by setting the container paths appropriately.
48 * FS-based backends are somewhat more restrictive due to the existence of real
49 * directory files; a regular file cannot have the same name as a directory. Other
50 * backends with virtual directories may not have this limitation. Callers should
51 * store files in such a way that no files and directories are under the same path.
53 * Methods should avoid throwing exceptions at all costs.
54 * As a corollary, external dependencies should be kept to a minimum.
56 * @ingroup FileBackend
59 abstract class FileBackend
{
60 protected $name; // string; unique backend name
61 protected $wikiId; // string; unique wiki name
62 protected $readOnly; // string; read-only explanation message
63 protected $parallelize; // string; when to do operations in parallel
64 protected $concurrency; // integer; how many operations can be done in parallel
66 /** @var LockManager */
67 protected $lockManager;
68 /** @var FileJournal */
69 protected $fileJournal;
72 * Create a new backend instance from configuration.
73 * This should only be called from within FileBackendGroup.
76 * 'name' : The unique name of this backend.
77 * This should consist of alphanumberic, '-', and '_' characters.
78 * This name should not be changed after use.
79 * 'wikiId' : Prefix to container names that is unique to this wiki.
80 * It should only consist of alphanumberic, '-', and '_' characters.
81 * 'lockManager' : Registered name of a file lock manager to use.
82 * 'fileJournal' : File journal configuration; see FileJournal::factory().
83 * Journals simply log changes to files stored in the backend.
84 * 'readOnly' : Write operations are disallowed if this is a non-empty string.
85 * It should be an explanation for the backend being read-only.
86 * 'parallelize' : When to do file operations in parallel (when possible).
87 * Allowed values are "implicit", "explicit" and "off".
88 * 'concurrency' : How many file operations can be done in parallel.
90 * @param $config Array
92 public function __construct( array $config ) {
93 $this->name
= $config['name'];
94 if ( !preg_match( '!^[a-zA-Z0-9-_]{1,255}$!', $this->name
) ) {
95 throw new MWException( "Backend name `{$this->name}` is invalid." );
97 $this->wikiId
= isset( $config['wikiId'] )
99 : wfWikiID(); // e.g. "my_wiki-en_"
100 $this->lockManager
= ( $config['lockManager'] instanceof LockManager
)
101 ?
$config['lockManager']
102 : LockManagerGroup
::singleton()->get( $config['lockManager'] );
103 $this->fileJournal
= isset( $config['fileJournal'] )
104 ? FileJournal
::factory( $config['fileJournal'], $this->name
)
105 : FileJournal
::factory( array( 'class' => 'NullFileJournal' ), $this->name
);
106 $this->readOnly
= isset( $config['readOnly'] )
107 ?
(string)$config['readOnly']
109 $this->parallelize
= isset( $config['parallelize'] )
110 ?
(string)$config['parallelize']
112 $this->concurrency
= isset( $config['concurrency'] )
113 ?
(int)$config['concurrency']
118 * Get the unique backend name.
119 * We may have multiple different backends of the same type.
120 * For example, we can have two Swift backends using different proxies.
124 final public function getName() {
129 * Check if this backend is read-only
133 final public function isReadOnly() {
134 return ( $this->readOnly
!= '' );
138 * Get an explanatory message if this backend is read-only
140 * @return string|bool Returns false if the backend is not read-only
142 final public function getReadOnlyReason() {
143 return ( $this->readOnly
!= '' ) ?
$this->readOnly
: false;
147 * This is the main entry point into the backend for write operations.
148 * Callers supply an ordered list of operations to perform as a transaction.
149 * Files will be locked, the stat cache cleared, and then the operations attempted.
150 * If any serious errors occur, all attempted operations will be rolled back.
152 * $ops is an array of arrays. The outer array holds a list of operations.
153 * Each inner array is a set of key value pairs that specify an operation.
155 * Supported operations and their parameters:
156 * a) Create a new file in storage with the contents of a string
159 * 'dst' => <storage path>,
160 * 'content' => <string of new file contents>,
161 * 'overwrite' => <boolean>,
162 * 'overwriteSame' => <boolean>
164 * b) Copy a file system file into storage
167 * 'src' => <file system path>,
168 * 'dst' => <storage path>,
169 * 'overwrite' => <boolean>,
170 * 'overwriteSame' => <boolean>
172 * c) Copy a file within storage
175 * 'src' => <storage path>,
176 * 'dst' => <storage path>,
177 * 'overwrite' => <boolean>,
178 * 'overwriteSame' => <boolean>
180 * d) Move a file within storage
183 * 'src' => <storage path>,
184 * 'dst' => <storage path>,
185 * 'overwrite' => <boolean>,
186 * 'overwriteSame' => <boolean>
188 * e) Delete a file within storage
191 * 'src' => <storage path>,
192 * 'ignoreMissingSource' => <boolean>
194 * f) Do nothing (no-op)
199 * Boolean flags for operations (operation-specific):
200 * 'ignoreMissingSource' : The operation will simply succeed and do
201 * nothing if the source file does not exist.
202 * 'overwrite' : Any destination file will be overwritten.
203 * 'overwriteSame' : An error will not be given if a file already
204 * exists at the destination that has the same
205 * contents as the new contents to be written there.
207 * $opts is an associative of boolean flags, including:
208 * 'force' : Operation precondition errors no longer trigger an abort.
209 * Any remaining operations are still attempted. Unexpected
210 * failures may still cause remaning operations to be aborted.
211 * 'nonLocking' : No locks are acquired for the operations.
212 * This can increase performance for non-critical writes.
213 * This has no effect unless the 'force' flag is set.
214 * 'allowStale' : Don't require the latest available data.
215 * This can increase performance for non-critical writes.
216 * This has no effect unless the 'force' flag is set.
217 * 'nonJournaled' : Don't log this operation batch in the file journal.
218 * This limits the ability of recovery scripts.
219 * 'parallelize' : Try to do operations in parallel when possible.
221 * Remarks on locking:
222 * File system paths given to operations should refer to files that are
223 * already locked or otherwise safe from modification from other processes.
224 * Normally these files will be new temp files, which should be adequate.
227 * This returns a Status, which contains all warnings and fatals that occured
228 * during the operation. The 'failCount', 'successCount', and 'success' members
229 * will reflect each operation attempted. The status will be "OK" unless:
230 * a) unexpected operation errors occurred (network partitions, disk full...)
231 * b) significant operation errors occured and 'force' was not set
233 * @param $ops Array List of operations to execute in order
234 * @param $opts Array Batch operation options
237 final public function doOperations( array $ops, array $opts = array() ) {
238 if ( $this->isReadOnly() ) {
239 return Status
::newFatal( 'backend-fail-readonly', $this->name
, $this->readOnly
);
241 if ( empty( $opts['force'] ) ) { // sanity
242 unset( $opts['nonLocking'] );
243 unset( $opts['allowStale'] );
245 $opts['concurrency'] = 1; // off
246 if ( $this->parallelize
=== 'implicit' ) {
247 if ( !isset( $opts['parallelize'] ) ||
$opts['parallelize'] ) {
248 $opts['concurrency'] = $this->concurrency
;
250 } elseif ( $this->parallelize
=== 'explicit' ) {
251 if ( !empty( $opts['parallelize'] ) ) {
252 $opts['concurrency'] = $this->concurrency
;
255 return $this->doOperationsInternal( $ops, $opts );
259 * @see FileBackend::doOperations()
261 abstract protected function doOperationsInternal( array $ops, array $opts );
264 * Same as doOperations() except it takes a single operation.
265 * If you are doing a batch of operations that should either
266 * all succeed or all fail, then use that function instead.
268 * @see FileBackend::doOperations()
270 * @param $op Array Operation
271 * @param $opts Array Operation options
274 final public function doOperation( array $op, array $opts = array() ) {
275 return $this->doOperations( array( $op ), $opts );
279 * Performs a single create operation.
280 * This sets $params['op'] to 'create' and passes it to doOperation().
282 * @see FileBackend::doOperation()
284 * @param $params Array Operation parameters
285 * @param $opts Array Operation options
288 final public function create( array $params, array $opts = array() ) {
289 $params['op'] = 'create';
290 return $this->doOperation( $params, $opts );
294 * Performs a single store operation.
295 * This sets $params['op'] to 'store' and passes it to doOperation().
297 * @see FileBackend::doOperation()
299 * @param $params Array Operation parameters
300 * @param $opts Array Operation options
303 final public function store( array $params, array $opts = array() ) {
304 $params['op'] = 'store';
305 return $this->doOperation( $params, $opts );
309 * Performs a single copy operation.
310 * This sets $params['op'] to 'copy' and passes it to doOperation().
312 * @see FileBackend::doOperation()
314 * @param $params Array Operation parameters
315 * @param $opts Array Operation options
318 final public function copy( array $params, array $opts = array() ) {
319 $params['op'] = 'copy';
320 return $this->doOperation( $params, $opts );
324 * Performs a single move operation.
325 * This sets $params['op'] to 'move' and passes it to doOperation().
327 * @see FileBackend::doOperation()
329 * @param $params Array Operation parameters
330 * @param $opts Array Operation options
333 final public function move( array $params, array $opts = array() ) {
334 $params['op'] = 'move';
335 return $this->doOperation( $params, $opts );
339 * Performs a single delete operation.
340 * This sets $params['op'] to 'delete' and passes it to doOperation().
342 * @see FileBackend::doOperation()
344 * @param $params Array Operation parameters
345 * @param $opts Array Operation options
348 final public function delete( array $params, array $opts = array() ) {
349 $params['op'] = 'delete';
350 return $this->doOperation( $params, $opts );
354 * Concatenate a list of storage files into a single file system file.
355 * The target path should refer to a file that is already locked or
356 * otherwise safe from modification from other processes. Normally,
357 * the file will be a new temp file, which should be adequate.
359 * srcs : ordered source storage paths (e.g. chunk1, chunk2, ...)
360 * dst : file system path to 0-byte temp file
362 * @param $params Array Operation parameters
365 abstract public function concatenate( array $params );
368 * Prepare a storage directory for usage.
369 * This will create any required containers and parent directories.
370 * Backends using key/value stores only need to create the container.
373 * dir : storage directory
375 * @param $params Array
378 final public function prepare( array $params ) {
379 if ( $this->isReadOnly() ) {
380 return Status
::newFatal( 'backend-fail-readonly', $this->name
, $this->readOnly
);
382 return $this->doPrepare( $params );
386 * @see FileBackend::prepare()
388 abstract protected function doPrepare( array $params );
391 * Take measures to block web access to a storage directory and
392 * the container it belongs to. FS backends might add .htaccess
393 * files whereas key/value store backends might restrict container
394 * access to the auth user that represents end-users in web request.
395 * This is not guaranteed to actually do anything.
398 * dir : storage directory
399 * noAccess : try to deny file access
400 * noListing : try to deny file listing
402 * @param $params Array
405 final public function secure( array $params ) {
406 if ( $this->isReadOnly() ) {
407 return Status
::newFatal( 'backend-fail-readonly', $this->name
, $this->readOnly
);
409 $status = $this->doPrepare( $params ); // dir must exist to restrict it
410 if ( $status->isOK() ) {
411 $status->merge( $this->doSecure( $params ) );
417 * @see FileBackend::secure()
419 abstract protected function doSecure( array $params );
422 * Delete a storage directory if it is empty.
423 * Backends using key/value stores may do nothing unless the directory
424 * is that of an empty container, in which case it should be deleted.
427 * dir : storage directory
428 * recursive : recursively delete empty subdirectories first (@since 1.20)
430 * @param $params Array
433 final public function clean( array $params ) {
434 if ( $this->isReadOnly() ) {
435 return Status
::newFatal( 'backend-fail-readonly', $this->name
, $this->readOnly
);
437 return $this->doClean( $params );
441 * @see FileBackend::clean()
443 abstract protected function doClean( array $params );
446 * Check if a file exists at a storage path in the backend.
447 * This returns false if only a directory exists at the path.
450 * src : source storage path
451 * latest : use the latest available data
453 * @param $params Array
454 * @return bool|null Returns null on failure
456 abstract public function fileExists( array $params );
459 * Get the last-modified timestamp of the file at a storage path.
462 * src : source storage path
463 * latest : use the latest available data
465 * @param $params Array
466 * @return string|bool TS_MW timestamp or false on failure
468 abstract public function getFileTimestamp( array $params );
471 * Get the contents of a file at a storage path in the backend.
472 * This should be avoided for potentially large files.
475 * src : source storage path
476 * latest : use the latest available data
478 * @param $params Array
479 * @return string|bool Returns false on failure
481 abstract public function getFileContents( array $params );
484 * Get the size (bytes) of a file at a storage path in the backend.
487 * src : source storage path
488 * latest : use the latest available data
490 * @param $params Array
491 * @return integer|bool Returns false on failure
493 abstract public function getFileSize( array $params );
496 * Get quick information about a file at a storage path in the backend.
497 * If the file does not exist, then this returns false.
498 * Otherwise, the result is an associative array that includes:
499 * mtime : the last-modified timestamp (TS_MW)
500 * size : the file size (bytes)
501 * Additional values may be included for internal use only.
504 * src : source storage path
505 * latest : use the latest available data
507 * @param $params Array
508 * @return Array|bool|null Returns null on failure
510 abstract public function getFileStat( array $params );
513 * Get a SHA-1 hash of the file at a storage path in the backend.
516 * src : source storage path
517 * latest : use the latest available data
519 * @param $params Array
520 * @return string|bool Hash string or false on failure
522 abstract public function getFileSha1Base36( array $params );
525 * Get the properties of the file at a storage path in the backend.
526 * Returns FSFile::placeholderProps() on failure.
529 * src : source storage path
530 * latest : use the latest available data
532 * @param $params Array
535 abstract public function getFileProps( array $params );
538 * Stream the file at a storage path in the backend.
539 * If the file does not exists, a 404 error will be given.
540 * Appropriate HTTP headers (Status, Content-Type, Content-Length)
541 * must be sent if streaming began, while none should be sent otherwise.
542 * Implementations should flush the output buffer before sending data.
545 * src : source storage path
546 * headers : additional HTTP headers to send on success
547 * latest : use the latest available data
549 * @param $params Array
552 abstract public function streamFile( array $params );
555 * Returns a file system file, identical to the file at a storage path.
556 * The file returned is either:
557 * a) A local copy of the file at a storage path in the backend.
558 * The temporary copy will have the same extension as the source.
559 * b) An original of the file at a storage path in the backend.
560 * Temporary files may be purged when the file object falls out of scope.
562 * Write operations should *never* be done on this file as some backends
563 * may do internal tracking or may be instances of FileBackendMultiWrite.
564 * In that later case, there are copies of the file that must stay in sync.
565 * Additionally, further calls to this function may return the same file.
568 * src : source storage path
569 * latest : use the latest available data
571 * @param $params Array
572 * @return FSFile|null Returns null on failure
574 abstract public function getLocalReference( array $params );
577 * Get a local copy on disk of the file at a storage path in the backend.
578 * The temporary copy will have the same file extension as the source.
579 * Temporary files may be purged when the file object falls out of scope.
582 * src : source storage path
583 * latest : use the latest available data
585 * @param $params Array
586 * @return TempFSFile|null Returns null on failure
588 abstract public function getLocalCopy( array $params );
591 * Check if a directory exists at a given storage path.
592 * Backends using key/value stores will check if the path is a
593 * virtual directory, meaning there are files under the given directory.
595 * Storage backends with eventual consistency might return stale data.
598 * dir : storage directory
600 * @return bool|null Returns null on failure
603 abstract public function directoryExists( array $params );
606 * Get an iterator to list *all* directories under a storage directory.
607 * If the directory is of the form "mwstore://backend/container",
608 * then all directories in the container should be listed.
609 * If the directory is of form "mwstore://backend/container/dir",
610 * then all directories directly under that directory should be listed.
611 * Results should be storage directories relative to the given directory.
613 * Storage backends with eventual consistency might return stale data.
616 * dir : storage directory
617 * topOnly : only return direct child dirs of the directory
619 * @return Traversable|Array|null Returns null on failure
622 abstract public function getDirectoryList( array $params );
625 * Same as FileBackend::getDirectoryList() except only lists
626 * directories that are immediately under the given directory.
628 * Storage backends with eventual consistency might return stale data.
631 * dir : storage directory
633 * @return Traversable|Array|null Returns null on failure
636 final public function getTopDirectoryList( array $params ) {
637 return $this->getDirectoryList( array( 'topOnly' => true ) +
$params );
641 * Get an iterator to list *all* stored files under a storage directory.
642 * If the directory is of the form "mwstore://backend/container",
643 * then all files in the container should be listed.
644 * If the directory is of form "mwstore://backend/container/dir",
645 * then all files under that directory should be listed.
646 * Results should be storage paths relative to the given directory.
648 * Storage backends with eventual consistency might return stale data.
651 * dir : storage directory
652 * topOnly : only return direct child files of the directory (@since 1.20)
654 * @return Traversable|Array|null Returns null on failure
656 abstract public function getFileList( array $params );
659 * Same as FileBackend::getFileList() except only lists
660 * files that are immediately under the given directory.
662 * Storage backends with eventual consistency might return stale data.
665 * dir : storage directory
667 * @return Traversable|Array|null Returns null on failure
670 final public function getTopFileList( array $params ) {
671 return $this->getFileList( array( 'topOnly' => true ) +
$params );
675 * Invalidate any in-process file existence and property cache.
676 * If $paths is given, then only the cache for those files will be cleared.
678 * @param $paths Array Storage paths (optional)
681 public function clearCache( array $paths = null ) {}
684 * Lock the files at the given storage paths in the backend.
685 * This will either lock all the files or none (on failure).
687 * Callers should consider using getScopedFileLocks() instead.
689 * @param $paths Array Storage paths
690 * @param $type integer LockManager::LOCK_* constant
693 final public function lockFiles( array $paths, $type ) {
694 return $this->lockManager
->lock( $paths, $type );
698 * Unlock the files at the given storage paths in the backend.
700 * @param $paths Array Storage paths
701 * @param $type integer LockManager::LOCK_* constant
704 final public function unlockFiles( array $paths, $type ) {
705 return $this->lockManager
->unlock( $paths, $type );
709 * Lock the files at the given storage paths in the backend.
710 * This will either lock all the files or none (on failure).
711 * On failure, the status object will be updated with errors.
713 * Once the return value goes out scope, the locks will be released and
714 * the status updated. Unlock fatals will not change the status "OK" value.
716 * @param $paths Array Storage paths
717 * @param $type integer LockManager::LOCK_* constant
718 * @param $status Status Status to update on lock/unlock
719 * @return ScopedLock|null Returns null on failure
721 final public function getScopedFileLocks( array $paths, $type, Status
$status ) {
722 return ScopedLock
::factory( $this->lockManager
, $paths, $type, $status );
726 * Get the root storage path of this backend.
727 * All container paths are "subdirectories" of this path.
729 * @return string Storage path
732 final public function getRootStoragePath() {
733 return "mwstore://{$this->name}";
737 * Get the file journal object for this backend
739 * @return FileJournal
741 final public function getJournal() {
742 return $this->fileJournal
;
746 * Check if a given path is a "mwstore://" path.
747 * This does not do any further validation or any existence checks.
749 * @param $path string
752 final public static function isStoragePath( $path ) {
753 return ( strpos( $path, 'mwstore://' ) === 0 );
757 * Split a storage path into a backend name, a container name,
758 * and a relative file path. The relative path may be the empty string.
759 * This does not do any path normalization or traversal checks.
761 * @param $storagePath string
762 * @return Array (backend, container, rel object) or (null, null, null)
764 final public static function splitStoragePath( $storagePath ) {
765 if ( self
::isStoragePath( $storagePath ) ) {
766 // Remove the "mwstore://" prefix and split the path
767 $parts = explode( '/', substr( $storagePath, 10 ), 3 );
768 if ( count( $parts ) >= 2 && $parts[0] != '' && $parts[1] != '' ) {
769 if ( count( $parts ) == 3 ) {
770 return $parts; // e.g. "backend/container/path"
772 return array( $parts[0], $parts[1], '' ); // e.g. "backend/container"
776 return array( null, null, null );
780 * Normalize a storage path by cleaning up directory separators.
781 * Returns null if the path is not of the format of a valid storage path.
783 * @param $storagePath string
784 * @return string|null
786 final public static function normalizeStoragePath( $storagePath ) {
787 list( $backend, $container, $relPath ) = self
::splitStoragePath( $storagePath );
788 if ( $relPath !== null ) { // must be for this backend
789 $relPath = self
::normalizeContainerPath( $relPath );
790 if ( $relPath !== null ) {
791 return ( $relPath != '' )
792 ?
"mwstore://{$backend}/{$container}/{$relPath}"
793 : "mwstore://{$backend}/{$container}";
800 * Get the parent storage directory of a storage path.
801 * This returns a path like "mwstore://backend/container",
802 * "mwstore://backend/container/...", or null if there is no parent.
804 * @param $storagePath string
805 * @return string|null
807 final public static function parentStoragePath( $storagePath ) {
808 $storagePath = dirname( $storagePath );
809 list( $b, $cont, $rel ) = self
::splitStoragePath( $storagePath );
810 return ( $rel === null ) ?
null : $storagePath;
814 * Get the final extension from a storage or FS path
816 * @param $path string
819 final public static function extensionFromPath( $path ) {
820 $i = strrpos( $path, '.' );
821 return strtolower( $i ?
substr( $path, $i +
1 ) : '' );
825 * Check if a relative path has no directory traversals
827 * @param $path string
831 final public static function isPathTraversalFree( $path ) {
832 return ( self
::normalizeContainerPath( $path ) !== null );
836 * Validate and normalize a relative storage path.
837 * Null is returned if the path involves directory traversal.
838 * Traversal is insecure for FS backends and broken for others.
840 * This uses the same traversal protection as Title::secureAndSplit().
842 * @param $path string Storage path relative to a container
843 * @return string|null
845 final protected static function normalizeContainerPath( $path ) {
846 // Normalize directory separators
847 $path = strtr( $path, '\\', '/' );
848 // Collapse any consecutive directory separators
849 $path = preg_replace( '![/]{2,}!', '/', $path );
850 // Remove any leading directory separator
851 $path = ltrim( $path, '/' );
852 // Use the same traversal protection as Title::secureAndSplit()
853 if ( strpos( $path, '.' ) !== false ) {
857 strpos( $path, './' ) === 0 ||
858 strpos( $path, '../' ) === 0 ||
859 strpos( $path, '/./' ) !== false ||
860 strpos( $path, '/../' ) !== false