2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
12 #include <linux/fuse.h>
14 #include <linux/mount.h>
15 #include <linux/wait.h>
16 #include <linux/list.h>
17 #include <linux/spinlock.h>
19 #include <linux/backing-dev.h>
20 #include <linux/mutex.h>
21 #include <linux/rwsem.h>
22 #include <linux/rbtree.h>
23 #include <linux/poll.h>
24 #include <linux/workqueue.h>
25 #include <linux/kref.h>
27 /** Max number of pages that can be used in a single read request */
28 #define FUSE_MAX_PAGES_PER_REQ 32
30 /** Bias for fi->writectr, meaning new writepages must not be sent */
31 #define FUSE_NOWRITE INT_MIN
33 /** It could be as large as PATH_MAX, but would that have any uses? */
34 #define FUSE_NAME_MAX 1024
36 /** Number of dentries for each connection in the control filesystem */
37 #define FUSE_CTL_NUM_DENTRIES 5
39 /** If the FUSE_DEFAULT_PERMISSIONS flag is given, the filesystem
40 module will check permissions based on the file mode. Otherwise no
41 permission checking is done in the kernel */
42 #define FUSE_DEFAULT_PERMISSIONS (1 << 0)
44 /** If the FUSE_ALLOW_OTHER flag is given, then not only the user
45 doing the mount will be allowed to access the filesystem */
46 #define FUSE_ALLOW_OTHER (1 << 1)
48 /** Number of page pointers embedded in fuse_req */
49 #define FUSE_REQ_INLINE_PAGES 1
51 /** List of active connections */
52 extern struct list_head fuse_conn_list
;
54 /** Global mutex protecting fuse_conn_list and the control filesystem */
55 extern struct mutex fuse_mutex
;
57 /** Module parameters */
58 extern unsigned max_user_bgreq
;
59 extern unsigned max_user_congthresh
;
61 /* One forget request */
62 struct fuse_forget_link
{
63 struct fuse_forget_one forget_one
;
64 struct fuse_forget_link
*next
;
72 /** Unique ID, which identifies the inode between userspace
76 /** Number of lookups on this inode */
79 /** The request used for sending the FORGET message */
80 struct fuse_forget_link
*forget
;
82 /** Time in jiffies until the file attributes are valid */
85 /** The sticky bit in inode->i_mode may have been removed, so
86 preserve the original mode */
89 /** 64 bit inode number */
92 /** Version of last attribute change */
95 /** Files usable in writepage. Protected by fc->lock */
96 struct list_head write_files
;
98 /** Writepages pending on truncate or fsync */
99 struct list_head queued_writes
;
101 /** Number of sent writes, a negative bias (FUSE_NOWRITE)
102 * means more writes are blocked */
105 /** Waitq for writepage completion */
106 wait_queue_head_t page_waitq
;
108 /** List of writepage requestst (pending or sent) */
109 struct list_head writepages
;
111 /** Miscellaneous bits describing inode state */
115 /** FUSE inode state bits */
117 /** Advise readdirplus */
118 FUSE_I_ADVISE_RDPLUS
,
119 /** Initialized with readdirplus */
121 /** An operation changing file size is in progress */
122 FUSE_I_SIZE_UNSTABLE
,
127 /** FUSE specific file data */
129 /** Fuse connection for this file */
130 struct fuse_conn
*fc
;
132 /** Request reserved for flush and release */
133 struct fuse_req
*reserved_req
;
135 /** Kernel file handle guaranteed to be unique */
138 /** File handle used by userspace */
141 /** Node id of this file */
147 /** FOPEN_* flags returned by open */
150 /** Entry on inode's write_files list */
151 struct list_head write_entry
;
153 /** RB node to be linked on fuse_conn->polled_files */
154 struct rb_node polled_node
;
156 /** Wait queue head for poll */
157 wait_queue_head_t poll_wait
;
159 /** Has flock been performed on this file? */
163 /** One input argument of a request */
169 /** The request input */
171 /** The request header */
172 struct fuse_in_header h
;
174 /** True if the data for the last argument is in req->pages */
177 /** Number of arguments */
180 /** Array of arguments */
181 struct fuse_in_arg args
[3];
184 /** One output argument of a request */
190 /** The request output */
192 /** Header returned from userspace */
193 struct fuse_out_header h
;
196 * The following bitfields are not changed during the request
200 /** Last argument is variable length (can be shorter than
204 /** Last argument is a list of pages to copy data to */
207 /** Zero partially or not copied pages */
208 unsigned page_zeroing
:1;
210 /** Pages may be replaced with new ones */
211 unsigned page_replace
:1;
213 /** Number or arguments */
216 /** Array of arguments */
217 struct fuse_arg args
[2];
220 /** FUSE page descriptor */
221 struct fuse_page_desc
{
233 struct fuse_in_arg args
[3];
239 struct fuse_arg args
[2];
243 #define FUSE_ARGS(args) struct fuse_args args = {}
245 /** The request IO state (for asynchronous processing) */
246 struct fuse_io_priv
{
258 struct completion
*done
;
261 #define FUSE_IO_PRIV_SYNC(f) \
263 .refcnt = { ATOMIC_INIT(1) }, \
271 * FR_ISREPLY: set if the request has reply
272 * FR_FORCE: force sending of the request even if interrupted
273 * FR_BACKGROUND: request is sent in the background
274 * FR_WAITING: request is counted as "waiting"
275 * FR_ABORTED: the request was aborted
276 * FR_INTERRUPTED: the request has been interrupted
277 * FR_LOCKED: data is being copied to/from the request
278 * FR_PENDING: request is not yet in userspace
279 * FR_SENT: request is in userspace, waiting for an answer
280 * FR_FINISHED: request is finished
281 * FR_PRIVATE: request is on private list
298 * A request to the client
300 * .waitq.lock protects the following fields:
302 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
305 /** This can be on either pending processing or io lists in
307 struct list_head list
;
309 /** Entry on the interrupts list */
310 struct list_head intr_entry
;
315 /** Unique ID for the interrupt request */
318 /* Request flags, updated with test/set/clear_bit() */
321 /** The request input */
324 /** The request output */
327 /** Used to wake up the task waiting for completion of request*/
328 wait_queue_head_t waitq
;
330 /** Data for asynchronous requests */
333 struct fuse_release_in in
;
336 struct fuse_init_in init_in
;
337 struct fuse_init_out init_out
;
338 struct cuse_init_in cuse_init_in
;
340 struct fuse_read_in in
;
344 struct fuse_write_in in
;
345 struct fuse_write_out out
;
346 struct fuse_req
*next
;
348 struct fuse_notify_retrieve_in retrieve_in
;
354 /** page-descriptor vector */
355 struct fuse_page_desc
*page_descs
;
357 /** size of the 'pages' array */
360 /** inline page vector */
361 struct page
*inline_pages
[FUSE_REQ_INLINE_PAGES
];
363 /** inline page-descriptor vector */
364 struct fuse_page_desc inline_page_descs
[FUSE_REQ_INLINE_PAGES
];
366 /** number of pages in vector */
369 /** File used in the request (or NULL) */
370 struct fuse_file
*ff
;
372 /** Inode used in the request or NULL */
375 /** AIO control block */
376 struct fuse_io_priv
*io
;
378 /** Link on fi->writepages */
379 struct list_head writepages_entry
;
381 /** Request completion callback */
382 void (*end
)(struct fuse_conn
*, struct fuse_req
*);
384 /** Request is stolen from fuse_file->reserved_req */
385 struct file
*stolen_file
;
389 /** Connection established */
392 /** Readers of the connection are waiting on this */
393 wait_queue_head_t waitq
;
395 /** The next unique request id */
398 /** The list of pending requests */
399 struct list_head pending
;
401 /** Pending interrupts */
402 struct list_head interrupts
;
404 /** Queue of pending forgets */
405 struct fuse_forget_link forget_list_head
;
406 struct fuse_forget_link
*forget_list_tail
;
408 /** Batching of FORGET requests (positive indicates FORGET batch) */
411 /** O_ASYNC requests */
412 struct fasync_struct
*fasync
;
416 /** Connection established */
419 /** Lock protecting accessess to members of this structure */
422 /** The list of requests being processed */
423 struct list_head processing
;
425 /** The list of requests under I/O */
430 * Fuse device instance
433 /** Fuse connection for this device */
434 struct fuse_conn
*fc
;
436 /** Processing queue */
437 struct fuse_pqueue pq
;
439 /** list entry on fc->devices */
440 struct list_head entry
;
446 * This structure is created, when the filesystem is mounted, and is
447 * destroyed, when the client device is closed and the filesystem is
451 /** Lock protecting accessess to members of this structure */
457 /** Number of fuse_dev's */
462 /** The user id for this mount */
465 /** The group id for this mount */
468 /** The fuse mount flags for this mount */
471 /** Maximum read size */
474 /** Maximum write size */
478 struct fuse_iqueue iq
;
480 /** The next unique kernel file handle */
483 /** rbtree of fuse_files waiting for poll events indexed by ph */
484 struct rb_root polled_files
;
486 /** Maximum number of outstanding background requests */
487 unsigned max_background
;
489 /** Number of background requests at which congestion starts */
490 unsigned congestion_threshold
;
492 /** Number of requests currently in the background */
493 unsigned num_background
;
495 /** Number of background requests currently queued for userspace */
496 unsigned active_background
;
498 /** The list of background requests set aside for later queuing */
499 struct list_head bg_queue
;
501 /** Flag indicating that INIT reply has been received. Allocating
502 * any fuse request will be suspended until the flag is set */
505 /** Flag indicating if connection is blocked. This will be
506 the case before the INIT reply is received, and if there
507 are too many outstading backgrounds requests */
510 /** waitq for blocked connection */
511 wait_queue_head_t blocked_waitq
;
513 /** waitq for reserved requests */
514 wait_queue_head_t reserved_req_waitq
;
516 /** Connection established, cleared on umount, connection
517 abort and device release */
520 /** Connection failed (version mismatch). Cannot race with
521 setting other bitfields since it is only set once in INIT
522 reply, before any other request, and never cleared */
523 unsigned conn_error
:1;
525 /** Connection successful. Only set in INIT */
526 unsigned conn_init
:1;
528 /** Do readpages asynchronously? Only set in INIT */
529 unsigned async_read
:1;
531 /** Do not send separate SETATTR request before open(O_TRUNC) */
532 unsigned atomic_o_trunc
:1;
534 /** Filesystem supports NFS exporting. Only set in INIT */
535 unsigned export_support
:1;
537 /** Set if bdi is valid */
538 unsigned bdi_initialized
:1;
540 /** write-back cache policy (default is write-through) */
541 unsigned writeback_cache
:1;
544 * The following bitfields are only for optimization purposes
545 * and hence races in setting them will not cause malfunction
548 /** Is open/release not implemented by fs? */
551 /** Is fsync not implemented by fs? */
554 /** Is fsyncdir not implemented by fs? */
555 unsigned no_fsyncdir
:1;
557 /** Is flush not implemented by fs? */
560 /** Is setxattr not implemented by fs? */
561 unsigned no_setxattr
:1;
563 /** Is getxattr not implemented by fs? */
564 unsigned no_getxattr
:1;
566 /** Is listxattr not implemented by fs? */
567 unsigned no_listxattr
:1;
569 /** Is removexattr not implemented by fs? */
570 unsigned no_removexattr
:1;
572 /** Are posix file locking primitives not implemented by fs? */
575 /** Is access not implemented by fs? */
576 unsigned no_access
:1;
578 /** Is create not implemented by fs? */
579 unsigned no_create
:1;
581 /** Is interrupt not implemented by fs? */
582 unsigned no_interrupt
:1;
584 /** Is bmap not implemented by fs? */
587 /** Is poll not implemented by fs? */
590 /** Do multi-page cached writes */
591 unsigned big_writes
:1;
593 /** Don't apply umask to creation modes */
594 unsigned dont_mask
:1;
596 /** Are BSD file locking primitives not implemented by fs? */
599 /** Is fallocate not implemented by fs? */
600 unsigned no_fallocate
:1;
602 /** Is rename with flags implemented by fs? */
603 unsigned no_rename2
:1;
605 /** Use enhanced/automatic page cache invalidation. */
606 unsigned auto_inval_data
:1;
608 /** Does the filesystem support readdirplus? */
609 unsigned do_readdirplus
:1;
611 /** Does the filesystem want adaptive readdirplus? */
612 unsigned readdirplus_auto
:1;
614 /** Does the filesystem support asynchronous direct-IO submission? */
615 unsigned async_dio
:1;
617 /** Is lseek not implemented by fs? */
620 /** The number of requests waiting for completion */
621 atomic_t num_waiting
;
623 /** Negotiated minor version */
626 /** Backing dev info */
627 struct backing_dev_info bdi
;
629 /** Entry on the fuse_conn_list */
630 struct list_head entry
;
632 /** Device ID from super block */
635 /** Dentries in the control filesystem */
636 struct dentry
*ctl_dentry
[FUSE_CTL_NUM_DENTRIES
];
638 /** number of dentries used in the above array */
641 /** Key for lock owner ID scrambling */
644 /** Reserved request for the DESTROY message */
645 struct fuse_req
*destroy_req
;
647 /** Version counter for attribute changes */
650 /** Called on final put */
651 void (*release
)(struct fuse_conn
*);
653 /** Super block for this connection. */
654 struct super_block
*sb
;
656 /** Read/write semaphore to hold when accessing sb. */
657 struct rw_semaphore killsb
;
659 /** List of device instances belonging to this connection */
660 struct list_head devices
;
663 static inline struct fuse_conn
*get_fuse_conn_super(struct super_block
*sb
)
665 return sb
->s_fs_info
;
668 static inline struct fuse_conn
*get_fuse_conn(struct inode
*inode
)
670 return get_fuse_conn_super(inode
->i_sb
);
673 static inline struct fuse_inode
*get_fuse_inode(struct inode
*inode
)
675 return container_of(inode
, struct fuse_inode
, inode
);
678 static inline u64
get_node_id(struct inode
*inode
)
680 return get_fuse_inode(inode
)->nodeid
;
683 /** Device operations */
684 extern const struct file_operations fuse_dev_operations
;
686 extern const struct dentry_operations fuse_dentry_operations
;
689 * Inode to nodeid comparison.
691 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
);
694 * Get a filled in inode
696 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
697 int generation
, struct fuse_attr
*attr
,
698 u64 attr_valid
, u64 attr_version
);
700 int fuse_lookup_name(struct super_block
*sb
, u64 nodeid
, struct qstr
*name
,
701 struct fuse_entry_out
*outarg
, struct inode
**inode
);
704 * Send FORGET command
706 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
707 u64 nodeid
, u64 nlookup
);
709 struct fuse_forget_link
*fuse_alloc_forget(void);
711 /* Used by READDIRPLUS */
712 void fuse_force_forget(struct file
*file
, u64 nodeid
);
715 * Initialize READ or READDIR request
717 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
718 loff_t pos
, size_t count
, int opcode
);
721 * Send OPEN or OPENDIR request
723 int fuse_open_common(struct inode
*inode
, struct file
*file
, bool isdir
);
725 struct fuse_file
*fuse_file_alloc(struct fuse_conn
*fc
);
726 struct fuse_file
*fuse_file_get(struct fuse_file
*ff
);
727 void fuse_file_free(struct fuse_file
*ff
);
728 void fuse_finish_open(struct inode
*inode
, struct file
*file
);
730 void fuse_sync_release(struct fuse_file
*ff
, int flags
);
733 * Send RELEASE or RELEASEDIR request
735 void fuse_release_common(struct file
*file
, int opcode
);
738 * Send FSYNC or FSYNCDIR request
740 int fuse_fsync_common(struct file
*file
, loff_t start
, loff_t end
,
741 int datasync
, int isdir
);
746 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
747 struct fuse_notify_poll_wakeup_out
*outarg
);
750 * Initialize file operations on a regular file
752 void fuse_init_file_inode(struct inode
*inode
);
755 * Initialize inode operations on regular files and special files
757 void fuse_init_common(struct inode
*inode
);
760 * Initialize inode and file operations on a directory
762 void fuse_init_dir(struct inode
*inode
);
765 * Initialize inode operations on a symlink
767 void fuse_init_symlink(struct inode
*inode
);
770 * Change attributes of an inode
772 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
773 u64 attr_valid
, u64 attr_version
);
775 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
779 * Initialize the client device
781 int fuse_dev_init(void);
784 * Cleanup the client device
786 void fuse_dev_cleanup(void);
788 int fuse_ctl_init(void);
789 void __exit
fuse_ctl_cleanup(void);
794 struct fuse_req
*fuse_request_alloc(unsigned npages
);
796 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
);
801 void fuse_request_free(struct fuse_req
*req
);
804 * Get a request, may fail with -ENOMEM,
805 * caller should specify # elements in req->pages[] explicitly
807 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
);
808 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
812 * Increment reference count on request
814 void __fuse_get_request(struct fuse_req
*req
);
817 * Gets a requests for a file operation, always succeeds
819 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
823 * Decrement reference count of a request. If count goes to zero free
826 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
);
829 * Send a request (synchronous)
831 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
);
834 * Simple request sending that does request allocation and freeing
836 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
);
839 * Send a request in the background
841 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
);
843 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
844 struct fuse_req
*req
);
846 /* Abort all requests */
847 void fuse_abort_conn(struct fuse_conn
*fc
);
850 * Invalidate inode attributes
852 void fuse_invalidate_attr(struct inode
*inode
);
854 void fuse_invalidate_entry_cache(struct dentry
*entry
);
856 void fuse_invalidate_atime(struct inode
*inode
);
859 * Acquire reference to fuse_conn
861 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
);
864 * Initialize fuse_conn
866 void fuse_conn_init(struct fuse_conn
*fc
);
869 * Release reference to fuse_conn
871 void fuse_conn_put(struct fuse_conn
*fc
);
873 struct fuse_dev
*fuse_dev_alloc(struct fuse_conn
*fc
);
874 void fuse_dev_free(struct fuse_dev
*fud
);
877 * Add connection to control filesystem
879 int fuse_ctl_add_conn(struct fuse_conn
*fc
);
882 * Remove connection from control filesystem
884 void fuse_ctl_remove_conn(struct fuse_conn
*fc
);
887 * Is file type valid?
889 int fuse_valid_type(int m
);
892 * Is current process allowed to perform filesystem operation?
894 int fuse_allow_current_process(struct fuse_conn
*fc
);
896 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
);
898 int fuse_update_attributes(struct inode
*inode
, struct kstat
*stat
,
899 struct file
*file
, bool *refreshed
);
901 void fuse_flush_writepages(struct inode
*inode
);
903 void fuse_set_nowrite(struct inode
*inode
);
904 void fuse_release_nowrite(struct inode
*inode
);
906 u64
fuse_get_attr_version(struct fuse_conn
*fc
);
909 * File-system tells the kernel to invalidate cache for the given node id.
911 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
912 loff_t offset
, loff_t len
);
915 * File-system tells the kernel to invalidate parent attributes and
916 * the dentry matching parent/name.
918 * If the child_nodeid is non-zero and:
919 * - matches the inode number for the dentry matching parent/name,
920 * - is not a mount point
921 * - is a file or oan empty directory
922 * then the dentry is unhashed (d_delete()).
924 int fuse_reverse_inval_entry(struct super_block
*sb
, u64 parent_nodeid
,
925 u64 child_nodeid
, struct qstr
*name
);
927 int fuse_do_open(struct fuse_conn
*fc
, u64 nodeid
, struct file
*file
,
931 * fuse_direct_io() flags
934 /** If set, it is WRITE; otherwise - READ */
935 #define FUSE_DIO_WRITE (1 << 0)
937 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
938 #define FUSE_DIO_CUSE (1 << 1)
940 ssize_t
fuse_direct_io(struct fuse_io_priv
*io
, struct iov_iter
*iter
,
941 loff_t
*ppos
, int flags
);
942 long fuse_do_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
,
944 long fuse_ioctl_common(struct file
*file
, unsigned int cmd
,
945 unsigned long arg
, unsigned int flags
);
946 unsigned fuse_file_poll(struct file
*file
, poll_table
*wait
);
947 int fuse_dev_release(struct inode
*inode
, struct file
*file
);
949 bool fuse_write_update_size(struct inode
*inode
, loff_t pos
);
951 int fuse_flush_times(struct inode
*inode
, struct fuse_file
*ff
);
952 int fuse_write_inode(struct inode
*inode
, struct writeback_control
*wbc
);
954 int fuse_do_setattr(struct inode
*inode
, struct iattr
*attr
,
957 void fuse_set_initialized(struct fuse_conn
*fc
);
959 #endif /* _FS_FUSE_I_H */