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 */
114 /** Lock for serializing lookup and readdir for back compatibility*/
118 /** FUSE inode state bits */
120 /** Advise readdirplus */
121 FUSE_I_ADVISE_RDPLUS
,
122 /** Initialized with readdirplus */
124 /** An operation changing file size is in progress */
125 FUSE_I_SIZE_UNSTABLE
,
130 /** FUSE specific file data */
132 /** Fuse connection for this file */
133 struct fuse_conn
*fc
;
135 /** Request reserved for flush and release */
136 struct fuse_req
*reserved_req
;
138 /** Kernel file handle guaranteed to be unique */
141 /** File handle used by userspace */
144 /** Node id of this file */
150 /** FOPEN_* flags returned by open */
153 /** Entry on inode's write_files list */
154 struct list_head write_entry
;
156 /** RB node to be linked on fuse_conn->polled_files */
157 struct rb_node polled_node
;
159 /** Wait queue head for poll */
160 wait_queue_head_t poll_wait
;
162 /** Has flock been performed on this file? */
166 /** One input argument of a request */
172 /** The request input */
174 /** The request header */
175 struct fuse_in_header h
;
177 /** True if the data for the last argument is in req->pages */
180 /** Number of arguments */
183 /** Array of arguments */
184 struct fuse_in_arg args
[3];
187 /** One output argument of a request */
193 /** The request output */
195 /** Header returned from userspace */
196 struct fuse_out_header h
;
199 * The following bitfields are not changed during the request
203 /** Last argument is variable length (can be shorter than
207 /** Last argument is a list of pages to copy data to */
210 /** Zero partially or not copied pages */
211 unsigned page_zeroing
:1;
213 /** Pages may be replaced with new ones */
214 unsigned page_replace
:1;
216 /** Number or arguments */
219 /** Array of arguments */
220 struct fuse_arg args
[2];
223 /** FUSE page descriptor */
224 struct fuse_page_desc
{
236 struct fuse_in_arg args
[3];
242 struct fuse_arg args
[2];
246 #define FUSE_ARGS(args) struct fuse_args args = {}
248 /** The request IO state (for asynchronous processing) */
249 struct fuse_io_priv
{
261 struct completion
*done
;
265 #define FUSE_IO_PRIV_SYNC(f) \
267 .refcnt = { ATOMIC_INIT(1) }, \
275 * FR_ISREPLY: set if the request has reply
276 * FR_FORCE: force sending of the request even if interrupted
277 * FR_BACKGROUND: request is sent in the background
278 * FR_WAITING: request is counted as "waiting"
279 * FR_ABORTED: the request was aborted
280 * FR_INTERRUPTED: the request has been interrupted
281 * FR_LOCKED: data is being copied to/from the request
282 * FR_PENDING: request is not yet in userspace
283 * FR_SENT: request is in userspace, waiting for an answer
284 * FR_FINISHED: request is finished
285 * FR_PRIVATE: request is on private list
302 * A request to the client
304 * .waitq.lock protects the following fields:
306 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
309 /** This can be on either pending processing or io lists in
311 struct list_head list
;
313 /** Entry on the interrupts list */
314 struct list_head intr_entry
;
319 /** Unique ID for the interrupt request */
322 /* Request flags, updated with test/set/clear_bit() */
325 /** The request input */
328 /** The request output */
331 /** Used to wake up the task waiting for completion of request*/
332 wait_queue_head_t waitq
;
334 /** Data for asynchronous requests */
337 struct fuse_release_in in
;
340 struct fuse_init_in init_in
;
341 struct fuse_init_out init_out
;
342 struct cuse_init_in cuse_init_in
;
344 struct fuse_read_in in
;
348 struct fuse_write_in in
;
349 struct fuse_write_out out
;
350 struct fuse_req
*next
;
352 struct fuse_notify_retrieve_in retrieve_in
;
358 /** page-descriptor vector */
359 struct fuse_page_desc
*page_descs
;
361 /** size of the 'pages' array */
364 /** inline page vector */
365 struct page
*inline_pages
[FUSE_REQ_INLINE_PAGES
];
367 /** inline page-descriptor vector */
368 struct fuse_page_desc inline_page_descs
[FUSE_REQ_INLINE_PAGES
];
370 /** number of pages in vector */
373 /** File used in the request (or NULL) */
374 struct fuse_file
*ff
;
376 /** Inode used in the request or NULL */
379 /** AIO control block */
380 struct fuse_io_priv
*io
;
382 /** Link on fi->writepages */
383 struct list_head writepages_entry
;
385 /** Request completion callback */
386 void (*end
)(struct fuse_conn
*, struct fuse_req
*);
388 /** Request is stolen from fuse_file->reserved_req */
389 struct file
*stolen_file
;
393 /** Connection established */
396 /** Readers of the connection are waiting on this */
397 wait_queue_head_t waitq
;
399 /** The next unique request id */
402 /** The list of pending requests */
403 struct list_head pending
;
405 /** Pending interrupts */
406 struct list_head interrupts
;
408 /** Queue of pending forgets */
409 struct fuse_forget_link forget_list_head
;
410 struct fuse_forget_link
*forget_list_tail
;
412 /** Batching of FORGET requests (positive indicates FORGET batch) */
415 /** O_ASYNC requests */
416 struct fasync_struct
*fasync
;
420 /** Connection established */
423 /** Lock protecting accessess to members of this structure */
426 /** The list of requests being processed */
427 struct list_head processing
;
429 /** The list of requests under I/O */
434 * Fuse device instance
437 /** Fuse connection for this device */
438 struct fuse_conn
*fc
;
440 /** Processing queue */
441 struct fuse_pqueue pq
;
443 /** list entry on fc->devices */
444 struct list_head entry
;
450 * This structure is created, when the filesystem is mounted, and is
451 * destroyed, when the client device is closed and the filesystem is
455 /** Lock protecting accessess to members of this structure */
461 /** Number of fuse_dev's */
466 /** The user id for this mount */
469 /** The group id for this mount */
472 /** The fuse mount flags for this mount */
475 /** Maximum read size */
478 /** Maximum write size */
482 struct fuse_iqueue iq
;
484 /** The next unique kernel file handle */
487 /** rbtree of fuse_files waiting for poll events indexed by ph */
488 struct rb_root polled_files
;
490 /** Maximum number of outstanding background requests */
491 unsigned max_background
;
493 /** Number of background requests at which congestion starts */
494 unsigned congestion_threshold
;
496 /** Number of requests currently in the background */
497 unsigned num_background
;
499 /** Number of background requests currently queued for userspace */
500 unsigned active_background
;
502 /** The list of background requests set aside for later queuing */
503 struct list_head bg_queue
;
505 /** Flag indicating that INIT reply has been received. Allocating
506 * any fuse request will be suspended until the flag is set */
509 /** Flag indicating if connection is blocked. This will be
510 the case before the INIT reply is received, and if there
511 are too many outstading backgrounds requests */
514 /** waitq for blocked connection */
515 wait_queue_head_t blocked_waitq
;
517 /** waitq for reserved requests */
518 wait_queue_head_t reserved_req_waitq
;
520 /** Connection established, cleared on umount, connection
521 abort and device release */
524 /** Connection failed (version mismatch). Cannot race with
525 setting other bitfields since it is only set once in INIT
526 reply, before any other request, and never cleared */
527 unsigned conn_error
:1;
529 /** Connection successful. Only set in INIT */
530 unsigned conn_init
:1;
532 /** Do readpages asynchronously? Only set in INIT */
533 unsigned async_read
:1;
535 /** Do not send separate SETATTR request before open(O_TRUNC) */
536 unsigned atomic_o_trunc
:1;
538 /** Filesystem supports NFS exporting. Only set in INIT */
539 unsigned export_support
:1;
541 /** Set if bdi is valid */
542 unsigned bdi_initialized
:1;
544 /** write-back cache policy (default is write-through) */
545 unsigned writeback_cache
:1;
547 /** allow parallel lookups and readdir (default is serialized) */
548 unsigned parallel_dirops
:1;
551 * The following bitfields are only for optimization purposes
552 * and hence races in setting them will not cause malfunction
555 /** Is open/release not implemented by fs? */
558 /** Is fsync not implemented by fs? */
561 /** Is fsyncdir not implemented by fs? */
562 unsigned no_fsyncdir
:1;
564 /** Is flush not implemented by fs? */
567 /** Is setxattr not implemented by fs? */
568 unsigned no_setxattr
:1;
570 /** Is getxattr not implemented by fs? */
571 unsigned no_getxattr
:1;
573 /** Is listxattr not implemented by fs? */
574 unsigned no_listxattr
:1;
576 /** Is removexattr not implemented by fs? */
577 unsigned no_removexattr
:1;
579 /** Are posix file locking primitives not implemented by fs? */
582 /** Is access not implemented by fs? */
583 unsigned no_access
:1;
585 /** Is create not implemented by fs? */
586 unsigned no_create
:1;
588 /** Is interrupt not implemented by fs? */
589 unsigned no_interrupt
:1;
591 /** Is bmap not implemented by fs? */
594 /** Is poll not implemented by fs? */
597 /** Do multi-page cached writes */
598 unsigned big_writes
:1;
600 /** Don't apply umask to creation modes */
601 unsigned dont_mask
:1;
603 /** Are BSD file locking primitives not implemented by fs? */
606 /** Is fallocate not implemented by fs? */
607 unsigned no_fallocate
:1;
609 /** Is rename with flags implemented by fs? */
610 unsigned no_rename2
:1;
612 /** Use enhanced/automatic page cache invalidation. */
613 unsigned auto_inval_data
:1;
615 /** Does the filesystem support readdirplus? */
616 unsigned do_readdirplus
:1;
618 /** Does the filesystem want adaptive readdirplus? */
619 unsigned readdirplus_auto
:1;
621 /** Does the filesystem support asynchronous direct-IO submission? */
622 unsigned async_dio
:1;
624 /** Is lseek not implemented by fs? */
627 /** The number of requests waiting for completion */
628 atomic_t num_waiting
;
630 /** Negotiated minor version */
633 /** Backing dev info */
634 struct backing_dev_info bdi
;
636 /** Entry on the fuse_conn_list */
637 struct list_head entry
;
639 /** Device ID from super block */
642 /** Dentries in the control filesystem */
643 struct dentry
*ctl_dentry
[FUSE_CTL_NUM_DENTRIES
];
645 /** number of dentries used in the above array */
648 /** Key for lock owner ID scrambling */
651 /** Reserved request for the DESTROY message */
652 struct fuse_req
*destroy_req
;
654 /** Version counter for attribute changes */
657 /** Called on final put */
658 void (*release
)(struct fuse_conn
*);
660 /** Super block for this connection. */
661 struct super_block
*sb
;
663 /** Read/write semaphore to hold when accessing sb. */
664 struct rw_semaphore killsb
;
666 /** List of device instances belonging to this connection */
667 struct list_head devices
;
670 static inline struct fuse_conn
*get_fuse_conn_super(struct super_block
*sb
)
672 return sb
->s_fs_info
;
675 static inline struct fuse_conn
*get_fuse_conn(struct inode
*inode
)
677 return get_fuse_conn_super(inode
->i_sb
);
680 static inline struct fuse_inode
*get_fuse_inode(struct inode
*inode
)
682 return container_of(inode
, struct fuse_inode
, inode
);
685 static inline u64
get_node_id(struct inode
*inode
)
687 return get_fuse_inode(inode
)->nodeid
;
690 /** Device operations */
691 extern const struct file_operations fuse_dev_operations
;
693 extern const struct dentry_operations fuse_dentry_operations
;
696 * Inode to nodeid comparison.
698 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
);
701 * Get a filled in inode
703 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
704 int generation
, struct fuse_attr
*attr
,
705 u64 attr_valid
, u64 attr_version
);
707 int fuse_lookup_name(struct super_block
*sb
, u64 nodeid
, const struct qstr
*name
,
708 struct fuse_entry_out
*outarg
, struct inode
**inode
);
711 * Send FORGET command
713 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
714 u64 nodeid
, u64 nlookup
);
716 struct fuse_forget_link
*fuse_alloc_forget(void);
718 /* Used by READDIRPLUS */
719 void fuse_force_forget(struct file
*file
, u64 nodeid
);
722 * Initialize READ or READDIR request
724 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
725 loff_t pos
, size_t count
, int opcode
);
728 * Send OPEN or OPENDIR request
730 int fuse_open_common(struct inode
*inode
, struct file
*file
, bool isdir
);
732 struct fuse_file
*fuse_file_alloc(struct fuse_conn
*fc
);
733 struct fuse_file
*fuse_file_get(struct fuse_file
*ff
);
734 void fuse_file_free(struct fuse_file
*ff
);
735 void fuse_finish_open(struct inode
*inode
, struct file
*file
);
737 void fuse_sync_release(struct fuse_file
*ff
, int flags
);
740 * Send RELEASE or RELEASEDIR request
742 void fuse_release_common(struct file
*file
, int opcode
);
745 * Send FSYNC or FSYNCDIR request
747 int fuse_fsync_common(struct file
*file
, loff_t start
, loff_t end
,
748 int datasync
, int isdir
);
753 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
754 struct fuse_notify_poll_wakeup_out
*outarg
);
757 * Initialize file operations on a regular file
759 void fuse_init_file_inode(struct inode
*inode
);
762 * Initialize inode operations on regular files and special files
764 void fuse_init_common(struct inode
*inode
);
767 * Initialize inode and file operations on a directory
769 void fuse_init_dir(struct inode
*inode
);
772 * Initialize inode operations on a symlink
774 void fuse_init_symlink(struct inode
*inode
);
777 * Change attributes of an inode
779 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
780 u64 attr_valid
, u64 attr_version
);
782 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
786 * Initialize the client device
788 int fuse_dev_init(void);
791 * Cleanup the client device
793 void fuse_dev_cleanup(void);
795 int fuse_ctl_init(void);
796 void __exit
fuse_ctl_cleanup(void);
801 struct fuse_req
*fuse_request_alloc(unsigned npages
);
803 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
);
808 void fuse_request_free(struct fuse_req
*req
);
811 * Get a request, may fail with -ENOMEM,
812 * caller should specify # elements in req->pages[] explicitly
814 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
);
815 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
819 * Increment reference count on request
821 void __fuse_get_request(struct fuse_req
*req
);
824 * Gets a requests for a file operation, always succeeds
826 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
830 * Decrement reference count of a request. If count goes to zero free
833 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
);
836 * Send a request (synchronous)
838 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
);
841 * Simple request sending that does request allocation and freeing
843 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
);
846 * Send a request in the background
848 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
);
850 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
851 struct fuse_req
*req
);
853 /* Abort all requests */
854 void fuse_abort_conn(struct fuse_conn
*fc
);
857 * Invalidate inode attributes
859 void fuse_invalidate_attr(struct inode
*inode
);
861 void fuse_invalidate_entry_cache(struct dentry
*entry
);
863 void fuse_invalidate_atime(struct inode
*inode
);
866 * Acquire reference to fuse_conn
868 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
);
871 * Initialize fuse_conn
873 void fuse_conn_init(struct fuse_conn
*fc
);
876 * Release reference to fuse_conn
878 void fuse_conn_put(struct fuse_conn
*fc
);
880 struct fuse_dev
*fuse_dev_alloc(struct fuse_conn
*fc
);
881 void fuse_dev_free(struct fuse_dev
*fud
);
884 * Add connection to control filesystem
886 int fuse_ctl_add_conn(struct fuse_conn
*fc
);
889 * Remove connection from control filesystem
891 void fuse_ctl_remove_conn(struct fuse_conn
*fc
);
894 * Is file type valid?
896 int fuse_valid_type(int m
);
899 * Is current process allowed to perform filesystem operation?
901 int fuse_allow_current_process(struct fuse_conn
*fc
);
903 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
);
905 int fuse_update_attributes(struct inode
*inode
, struct kstat
*stat
,
906 struct file
*file
, bool *refreshed
);
908 void fuse_flush_writepages(struct inode
*inode
);
910 void fuse_set_nowrite(struct inode
*inode
);
911 void fuse_release_nowrite(struct inode
*inode
);
913 u64
fuse_get_attr_version(struct fuse_conn
*fc
);
916 * File-system tells the kernel to invalidate cache for the given node id.
918 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
919 loff_t offset
, loff_t len
);
922 * File-system tells the kernel to invalidate parent attributes and
923 * the dentry matching parent/name.
925 * If the child_nodeid is non-zero and:
926 * - matches the inode number for the dentry matching parent/name,
927 * - is not a mount point
928 * - is a file or oan empty directory
929 * then the dentry is unhashed (d_delete()).
931 int fuse_reverse_inval_entry(struct super_block
*sb
, u64 parent_nodeid
,
932 u64 child_nodeid
, struct qstr
*name
);
934 int fuse_do_open(struct fuse_conn
*fc
, u64 nodeid
, struct file
*file
,
938 * fuse_direct_io() flags
941 /** If set, it is WRITE; otherwise - READ */
942 #define FUSE_DIO_WRITE (1 << 0)
944 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
945 #define FUSE_DIO_CUSE (1 << 1)
947 ssize_t
fuse_direct_io(struct fuse_io_priv
*io
, struct iov_iter
*iter
,
948 loff_t
*ppos
, int flags
);
949 long fuse_do_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
,
951 long fuse_ioctl_common(struct file
*file
, unsigned int cmd
,
952 unsigned long arg
, unsigned int flags
);
953 unsigned fuse_file_poll(struct file
*file
, poll_table
*wait
);
954 int fuse_dev_release(struct inode
*inode
, struct file
*file
);
956 bool fuse_write_update_size(struct inode
*inode
, loff_t pos
);
958 int fuse_flush_times(struct inode
*inode
, struct fuse_file
*ff
);
959 int fuse_write_inode(struct inode
*inode
, struct writeback_control
*wbc
);
961 int fuse_do_setattr(struct inode
*inode
, struct iattr
*attr
,
964 void fuse_set_initialized(struct fuse_conn
*fc
);
966 void fuse_unlock_inode(struct inode
*inode
);
967 void fuse_lock_inode(struct inode
*inode
);
969 #endif /* _FS_FUSE_I_H */