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>
26 #include <linux/xattr.h>
28 /** Max number of pages that can be used in a single read request */
29 #define FUSE_MAX_PAGES_PER_REQ 32
31 /** Bias for fi->writectr, meaning new writepages must not be sent */
32 #define FUSE_NOWRITE INT_MIN
34 /** It could be as large as PATH_MAX, but would that have any uses? */
35 #define FUSE_NAME_MAX 1024
37 /** Number of dentries for each connection in the control filesystem */
38 #define FUSE_CTL_NUM_DENTRIES 5
40 /** Number of page pointers embedded in fuse_req */
41 #define FUSE_REQ_INLINE_PAGES 1
43 /** List of active connections */
44 extern struct list_head fuse_conn_list
;
46 /** Global mutex protecting fuse_conn_list and the control filesystem */
47 extern struct mutex fuse_mutex
;
49 /** Module parameters */
50 extern unsigned max_user_bgreq
;
51 extern unsigned max_user_congthresh
;
53 /* One forget request */
54 struct fuse_forget_link
{
55 struct fuse_forget_one forget_one
;
56 struct fuse_forget_link
*next
;
64 /** Unique ID, which identifies the inode between userspace
68 /** Number of lookups on this inode */
71 /** The request used for sending the FORGET message */
72 struct fuse_forget_link
*forget
;
74 /** Time in jiffies until the file attributes are valid */
77 /** The sticky bit in inode->i_mode may have been removed, so
78 preserve the original mode */
81 /** 64 bit inode number */
84 /** Version of last attribute change */
87 /** Files usable in writepage. Protected by fc->lock */
88 struct list_head write_files
;
90 /** Writepages pending on truncate or fsync */
91 struct list_head queued_writes
;
93 /** Number of sent writes, a negative bias (FUSE_NOWRITE)
94 * means more writes are blocked */
97 /** Waitq for writepage completion */
98 wait_queue_head_t page_waitq
;
100 /** List of writepage requestst (pending or sent) */
101 struct list_head writepages
;
103 /** Miscellaneous bits describing inode state */
106 /** Lock for serializing lookup and readdir for back compatibility*/
110 /** FUSE inode state bits */
112 /** Advise readdirplus */
113 FUSE_I_ADVISE_RDPLUS
,
114 /** Initialized with readdirplus */
116 /** An operation changing file size is in progress */
117 FUSE_I_SIZE_UNSTABLE
,
122 /** FUSE specific file data */
124 /** Fuse connection for this file */
125 struct fuse_conn
*fc
;
127 /** Request reserved for flush and release */
128 struct fuse_req
*reserved_req
;
130 /** Kernel file handle guaranteed to be unique */
133 /** File handle used by userspace */
136 /** Node id of this file */
142 /** FOPEN_* flags returned by open */
145 /** Entry on inode's write_files list */
146 struct list_head write_entry
;
148 /** RB node to be linked on fuse_conn->polled_files */
149 struct rb_node polled_node
;
151 /** Wait queue head for poll */
152 wait_queue_head_t poll_wait
;
154 /** Has flock been performed on this file? */
158 /** One input argument of a request */
164 /** The request input */
166 /** The request header */
167 struct fuse_in_header h
;
169 /** True if the data for the last argument is in req->pages */
172 /** Number of arguments */
175 /** Array of arguments */
176 struct fuse_in_arg args
[3];
179 /** One output argument of a request */
185 /** The request output */
187 /** Header returned from userspace */
188 struct fuse_out_header h
;
191 * The following bitfields are not changed during the request
195 /** Last argument is variable length (can be shorter than
199 /** Last argument is a list of pages to copy data to */
202 /** Zero partially or not copied pages */
203 unsigned page_zeroing
:1;
205 /** Pages may be replaced with new ones */
206 unsigned page_replace
:1;
208 /** Number or arguments */
211 /** Array of arguments */
212 struct fuse_arg args
[2];
215 /** FUSE page descriptor */
216 struct fuse_page_desc
{
228 struct fuse_in_arg args
[3];
234 struct fuse_arg args
[2];
238 #define FUSE_ARGS(args) struct fuse_args args = {}
240 /** The request IO state (for asynchronous processing) */
241 struct fuse_io_priv
{
253 struct completion
*done
;
257 #define FUSE_IO_PRIV_SYNC(f) \
259 .refcnt = { ATOMIC_INIT(1) }, \
267 * FR_ISREPLY: set if the request has reply
268 * FR_FORCE: force sending of the request even if interrupted
269 * FR_BACKGROUND: request is sent in the background
270 * FR_WAITING: request is counted as "waiting"
271 * FR_ABORTED: the request was aborted
272 * FR_INTERRUPTED: the request has been interrupted
273 * FR_LOCKED: data is being copied to/from the request
274 * FR_PENDING: request is not yet in userspace
275 * FR_SENT: request is in userspace, waiting for an answer
276 * FR_FINISHED: request is finished
277 * FR_PRIVATE: request is on private list
294 * A request to the client
296 * .waitq.lock protects the following fields:
298 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
301 /** This can be on either pending processing or io lists in
303 struct list_head list
;
305 /** Entry on the interrupts list */
306 struct list_head intr_entry
;
311 /** Unique ID for the interrupt request */
314 /* Request flags, updated with test/set/clear_bit() */
317 /** The request input */
320 /** The request output */
323 /** Used to wake up the task waiting for completion of request*/
324 wait_queue_head_t waitq
;
326 /** Data for asynchronous requests */
329 struct fuse_release_in in
;
332 struct fuse_init_in init_in
;
333 struct fuse_init_out init_out
;
334 struct cuse_init_in cuse_init_in
;
336 struct fuse_read_in in
;
340 struct fuse_write_in in
;
341 struct fuse_write_out out
;
342 struct fuse_req
*next
;
344 struct fuse_notify_retrieve_in retrieve_in
;
350 /** page-descriptor vector */
351 struct fuse_page_desc
*page_descs
;
353 /** size of the 'pages' array */
356 /** inline page vector */
357 struct page
*inline_pages
[FUSE_REQ_INLINE_PAGES
];
359 /** inline page-descriptor vector */
360 struct fuse_page_desc inline_page_descs
[FUSE_REQ_INLINE_PAGES
];
362 /** number of pages in vector */
365 /** File used in the request (or NULL) */
366 struct fuse_file
*ff
;
368 /** Inode used in the request or NULL */
371 /** AIO control block */
372 struct fuse_io_priv
*io
;
374 /** Link on fi->writepages */
375 struct list_head writepages_entry
;
377 /** Request completion callback */
378 void (*end
)(struct fuse_conn
*, struct fuse_req
*);
380 /** Request is stolen from fuse_file->reserved_req */
381 struct file
*stolen_file
;
385 /** Connection established */
388 /** Readers of the connection are waiting on this */
389 wait_queue_head_t waitq
;
391 /** The next unique request id */
394 /** The list of pending requests */
395 struct list_head pending
;
397 /** Pending interrupts */
398 struct list_head interrupts
;
400 /** Queue of pending forgets */
401 struct fuse_forget_link forget_list_head
;
402 struct fuse_forget_link
*forget_list_tail
;
404 /** Batching of FORGET requests (positive indicates FORGET batch) */
407 /** O_ASYNC requests */
408 struct fasync_struct
*fasync
;
412 /** Connection established */
415 /** Lock protecting accessess to members of this structure */
418 /** The list of requests being processed */
419 struct list_head processing
;
421 /** The list of requests under I/O */
426 * Fuse device instance
429 /** Fuse connection for this device */
430 struct fuse_conn
*fc
;
432 /** Processing queue */
433 struct fuse_pqueue pq
;
435 /** list entry on fc->devices */
436 struct list_head entry
;
442 * This structure is created, when the filesystem is mounted, and is
443 * destroyed, when the client device is closed and the filesystem is
447 /** Lock protecting accessess to members of this structure */
453 /** Number of fuse_dev's */
458 /** The user id for this mount */
461 /** The group id for this mount */
464 /** Maximum read size */
467 /** Maximum write size */
471 struct fuse_iqueue iq
;
473 /** The next unique kernel file handle */
476 /** rbtree of fuse_files waiting for poll events indexed by ph */
477 struct rb_root polled_files
;
479 /** Maximum number of outstanding background requests */
480 unsigned max_background
;
482 /** Number of background requests at which congestion starts */
483 unsigned congestion_threshold
;
485 /** Number of requests currently in the background */
486 unsigned num_background
;
488 /** Number of background requests currently queued for userspace */
489 unsigned active_background
;
491 /** The list of background requests set aside for later queuing */
492 struct list_head bg_queue
;
494 /** Flag indicating that INIT reply has been received. Allocating
495 * any fuse request will be suspended until the flag is set */
498 /** Flag indicating if connection is blocked. This will be
499 the case before the INIT reply is received, and if there
500 are too many outstading backgrounds requests */
503 /** waitq for blocked connection */
504 wait_queue_head_t blocked_waitq
;
506 /** waitq for reserved requests */
507 wait_queue_head_t reserved_req_waitq
;
509 /** Connection established, cleared on umount, connection
510 abort and device release */
513 /** Connection failed (version mismatch). Cannot race with
514 setting other bitfields since it is only set once in INIT
515 reply, before any other request, and never cleared */
516 unsigned conn_error
:1;
518 /** Connection successful. Only set in INIT */
519 unsigned conn_init
:1;
521 /** Do readpages asynchronously? Only set in INIT */
522 unsigned async_read
:1;
524 /** Do not send separate SETATTR request before open(O_TRUNC) */
525 unsigned atomic_o_trunc
:1;
527 /** Filesystem supports NFS exporting. Only set in INIT */
528 unsigned export_support
:1;
530 /** Set if bdi is valid */
531 unsigned bdi_initialized
:1;
533 /** write-back cache policy (default is write-through) */
534 unsigned writeback_cache
:1;
536 /** allow parallel lookups and readdir (default is serialized) */
537 unsigned parallel_dirops
:1;
539 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
540 unsigned handle_killpriv
:1;
543 * The following bitfields are only for optimization purposes
544 * and hence races in setting them will not cause malfunction
547 /** Is open/release not implemented by fs? */
550 /** Is fsync not implemented by fs? */
553 /** Is fsyncdir not implemented by fs? */
554 unsigned no_fsyncdir
:1;
556 /** Is flush not implemented by fs? */
559 /** Is setxattr not implemented by fs? */
560 unsigned no_setxattr
:1;
562 /** Is getxattr not implemented by fs? */
563 unsigned no_getxattr
:1;
565 /** Is listxattr not implemented by fs? */
566 unsigned no_listxattr
:1;
568 /** Is removexattr not implemented by fs? */
569 unsigned no_removexattr
:1;
571 /** Are posix file locking primitives not implemented by fs? */
574 /** Is access not implemented by fs? */
575 unsigned no_access
:1;
577 /** Is create not implemented by fs? */
578 unsigned no_create
:1;
580 /** Is interrupt not implemented by fs? */
581 unsigned no_interrupt
:1;
583 /** Is bmap not implemented by fs? */
586 /** Is poll not implemented by fs? */
589 /** Do multi-page cached writes */
590 unsigned big_writes
:1;
592 /** Don't apply umask to creation modes */
593 unsigned dont_mask
:1;
595 /** Are BSD file locking primitives not implemented by fs? */
598 /** Is fallocate not implemented by fs? */
599 unsigned no_fallocate
:1;
601 /** Is rename with flags implemented by fs? */
602 unsigned no_rename2
:1;
604 /** Use enhanced/automatic page cache invalidation. */
605 unsigned auto_inval_data
:1;
607 /** Does the filesystem support readdirplus? */
608 unsigned do_readdirplus
:1;
610 /** Does the filesystem want adaptive readdirplus? */
611 unsigned readdirplus_auto
:1;
613 /** Does the filesystem support asynchronous direct-IO submission? */
614 unsigned async_dio
:1;
616 /** Is lseek not implemented by fs? */
619 /** Does the filesystem support posix acls? */
620 unsigned posix_acl
:1;
622 /** Check permissions based on the file mode or not? */
623 unsigned default_permissions
:1;
625 /** Allow other than the mounter user to access the filesystem ? */
626 unsigned allow_other
:1;
628 /** The number of requests waiting for completion */
629 atomic_t num_waiting
;
631 /** Negotiated minor version */
634 /** Backing dev info */
635 struct backing_dev_info bdi
;
637 /** Entry on the fuse_conn_list */
638 struct list_head entry
;
640 /** Device ID from super block */
643 /** Dentries in the control filesystem */
644 struct dentry
*ctl_dentry
[FUSE_CTL_NUM_DENTRIES
];
646 /** number of dentries used in the above array */
649 /** Key for lock owner ID scrambling */
652 /** Reserved request for the DESTROY message */
653 struct fuse_req
*destroy_req
;
655 /** Version counter for attribute changes */
658 /** Called on final put */
659 void (*release
)(struct fuse_conn
*);
661 /** Super block for this connection. */
662 struct super_block
*sb
;
664 /** Read/write semaphore to hold when accessing sb. */
665 struct rw_semaphore killsb
;
667 /** List of device instances belonging to this connection */
668 struct list_head devices
;
671 static inline struct fuse_conn
*get_fuse_conn_super(struct super_block
*sb
)
673 return sb
->s_fs_info
;
676 static inline struct fuse_conn
*get_fuse_conn(struct inode
*inode
)
678 return get_fuse_conn_super(inode
->i_sb
);
681 static inline struct fuse_inode
*get_fuse_inode(struct inode
*inode
)
683 return container_of(inode
, struct fuse_inode
, inode
);
686 static inline u64
get_node_id(struct inode
*inode
)
688 return get_fuse_inode(inode
)->nodeid
;
691 /** Device operations */
692 extern const struct file_operations fuse_dev_operations
;
694 extern const struct dentry_operations fuse_dentry_operations
;
695 extern const struct dentry_operations fuse_root_dentry_operations
;
698 * Inode to nodeid comparison.
700 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
);
703 * Get a filled in inode
705 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
706 int generation
, struct fuse_attr
*attr
,
707 u64 attr_valid
, u64 attr_version
);
709 int fuse_lookup_name(struct super_block
*sb
, u64 nodeid
, const struct qstr
*name
,
710 struct fuse_entry_out
*outarg
, struct inode
**inode
);
713 * Send FORGET command
715 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
716 u64 nodeid
, u64 nlookup
);
718 struct fuse_forget_link
*fuse_alloc_forget(void);
720 /* Used by READDIRPLUS */
721 void fuse_force_forget(struct file
*file
, u64 nodeid
);
724 * Initialize READ or READDIR request
726 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
727 loff_t pos
, size_t count
, int opcode
);
730 * Send OPEN or OPENDIR request
732 int fuse_open_common(struct inode
*inode
, struct file
*file
, bool isdir
);
734 struct fuse_file
*fuse_file_alloc(struct fuse_conn
*fc
);
735 struct fuse_file
*fuse_file_get(struct fuse_file
*ff
);
736 void fuse_file_free(struct fuse_file
*ff
);
737 void fuse_finish_open(struct inode
*inode
, struct file
*file
);
739 void fuse_sync_release(struct fuse_file
*ff
, int flags
);
742 * Send RELEASE or RELEASEDIR request
744 void fuse_release_common(struct file
*file
, int opcode
);
747 * Send FSYNC or FSYNCDIR request
749 int fuse_fsync_common(struct file
*file
, loff_t start
, loff_t end
,
750 int datasync
, int isdir
);
755 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
756 struct fuse_notify_poll_wakeup_out
*outarg
);
759 * Initialize file operations on a regular file
761 void fuse_init_file_inode(struct inode
*inode
);
764 * Initialize inode operations on regular files and special files
766 void fuse_init_common(struct inode
*inode
);
769 * Initialize inode and file operations on a directory
771 void fuse_init_dir(struct inode
*inode
);
774 * Initialize inode operations on a symlink
776 void fuse_init_symlink(struct inode
*inode
);
779 * Change attributes of an inode
781 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
782 u64 attr_valid
, u64 attr_version
);
784 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
788 * Initialize the client device
790 int fuse_dev_init(void);
793 * Cleanup the client device
795 void fuse_dev_cleanup(void);
797 int fuse_ctl_init(void);
798 void __exit
fuse_ctl_cleanup(void);
803 struct fuse_req
*fuse_request_alloc(unsigned npages
);
805 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
);
810 void fuse_request_free(struct fuse_req
*req
);
813 * Get a request, may fail with -ENOMEM,
814 * caller should specify # elements in req->pages[] explicitly
816 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
);
817 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
821 * Increment reference count on request
823 void __fuse_get_request(struct fuse_req
*req
);
826 * Gets a requests for a file operation, always succeeds
828 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
832 * Decrement reference count of a request. If count goes to zero free
835 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
);
838 * Send a request (synchronous)
840 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
);
843 * Simple request sending that does request allocation and freeing
845 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
);
848 * Send a request in the background
850 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
);
852 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
853 struct fuse_req
*req
);
855 /* Abort all requests */
856 void fuse_abort_conn(struct fuse_conn
*fc
);
859 * Invalidate inode attributes
861 void fuse_invalidate_attr(struct inode
*inode
);
863 void fuse_invalidate_entry_cache(struct dentry
*entry
);
865 void fuse_invalidate_atime(struct inode
*inode
);
868 * Acquire reference to fuse_conn
870 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
);
873 * Initialize fuse_conn
875 void fuse_conn_init(struct fuse_conn
*fc
);
878 * Release reference to fuse_conn
880 void fuse_conn_put(struct fuse_conn
*fc
);
882 struct fuse_dev
*fuse_dev_alloc(struct fuse_conn
*fc
);
883 void fuse_dev_free(struct fuse_dev
*fud
);
886 * Add connection to control filesystem
888 int fuse_ctl_add_conn(struct fuse_conn
*fc
);
891 * Remove connection from control filesystem
893 void fuse_ctl_remove_conn(struct fuse_conn
*fc
);
896 * Is file type valid?
898 int fuse_valid_type(int m
);
901 * Is current process allowed to perform filesystem operation?
903 int fuse_allow_current_process(struct fuse_conn
*fc
);
905 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
);
907 void fuse_update_ctime(struct inode
*inode
);
909 int fuse_update_attributes(struct inode
*inode
, struct kstat
*stat
,
910 struct file
*file
, bool *refreshed
);
912 void fuse_flush_writepages(struct inode
*inode
);
914 void fuse_set_nowrite(struct inode
*inode
);
915 void fuse_release_nowrite(struct inode
*inode
);
917 u64
fuse_get_attr_version(struct fuse_conn
*fc
);
920 * File-system tells the kernel to invalidate cache for the given node id.
922 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
923 loff_t offset
, loff_t len
);
926 * File-system tells the kernel to invalidate parent attributes and
927 * the dentry matching parent/name.
929 * If the child_nodeid is non-zero and:
930 * - matches the inode number for the dentry matching parent/name,
931 * - is not a mount point
932 * - is a file or oan empty directory
933 * then the dentry is unhashed (d_delete()).
935 int fuse_reverse_inval_entry(struct super_block
*sb
, u64 parent_nodeid
,
936 u64 child_nodeid
, struct qstr
*name
);
938 int fuse_do_open(struct fuse_conn
*fc
, u64 nodeid
, struct file
*file
,
942 * fuse_direct_io() flags
945 /** If set, it is WRITE; otherwise - READ */
946 #define FUSE_DIO_WRITE (1 << 0)
948 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
949 #define FUSE_DIO_CUSE (1 << 1)
951 ssize_t
fuse_direct_io(struct fuse_io_priv
*io
, struct iov_iter
*iter
,
952 loff_t
*ppos
, int flags
);
953 long fuse_do_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
,
955 long fuse_ioctl_common(struct file
*file
, unsigned int cmd
,
956 unsigned long arg
, unsigned int flags
);
957 unsigned fuse_file_poll(struct file
*file
, poll_table
*wait
);
958 int fuse_dev_release(struct inode
*inode
, struct file
*file
);
960 bool fuse_write_update_size(struct inode
*inode
, loff_t pos
);
962 int fuse_flush_times(struct inode
*inode
, struct fuse_file
*ff
);
963 int fuse_write_inode(struct inode
*inode
, struct writeback_control
*wbc
);
965 int fuse_do_setattr(struct dentry
*dentry
, struct iattr
*attr
,
968 void fuse_set_initialized(struct fuse_conn
*fc
);
970 void fuse_unlock_inode(struct inode
*inode
);
971 void fuse_lock_inode(struct inode
*inode
);
973 int fuse_setxattr(struct inode
*inode
, const char *name
, const void *value
,
974 size_t size
, int flags
);
975 ssize_t
fuse_getxattr(struct inode
*inode
, const char *name
, void *value
,
977 ssize_t
fuse_listxattr(struct dentry
*entry
, char *list
, size_t size
);
978 int fuse_removexattr(struct inode
*inode
, const char *name
);
979 extern const struct xattr_handler
*fuse_xattr_handlers
[];
980 extern const struct xattr_handler
*fuse_acl_xattr_handlers
[];
983 struct posix_acl
*fuse_get_acl(struct inode
*inode
, int type
);
984 int fuse_set_acl(struct inode
*inode
, struct posix_acl
*acl
, int type
);
986 #endif /* _FS_FUSE_I_H */