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
;
264 #define FUSE_IO_PRIV_SYNC(f) \
266 .refcnt = { ATOMIC_INIT(1) }, \
274 * FR_ISREPLY: set if the request has reply
275 * FR_FORCE: force sending of the request even if interrupted
276 * FR_BACKGROUND: request is sent in the background
277 * FR_WAITING: request is counted as "waiting"
278 * FR_ABORTED: the request was aborted
279 * FR_INTERRUPTED: the request has been interrupted
280 * FR_LOCKED: data is being copied to/from the request
281 * FR_PENDING: request is not yet in userspace
282 * FR_SENT: request is in userspace, waiting for an answer
283 * FR_FINISHED: request is finished
284 * FR_PRIVATE: request is on private list
301 * A request to the client
303 * .waitq.lock protects the following fields:
305 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
308 /** This can be on either pending processing or io lists in
310 struct list_head list
;
312 /** Entry on the interrupts list */
313 struct list_head intr_entry
;
318 /** Unique ID for the interrupt request */
321 /* Request flags, updated with test/set/clear_bit() */
324 /** The request input */
327 /** The request output */
330 /** Used to wake up the task waiting for completion of request*/
331 wait_queue_head_t waitq
;
333 /** Data for asynchronous requests */
336 struct fuse_release_in in
;
339 struct fuse_init_in init_in
;
340 struct fuse_init_out init_out
;
341 struct cuse_init_in cuse_init_in
;
343 struct fuse_read_in in
;
347 struct fuse_write_in in
;
348 struct fuse_write_out out
;
349 struct fuse_req
*next
;
351 struct fuse_notify_retrieve_in retrieve_in
;
357 /** page-descriptor vector */
358 struct fuse_page_desc
*page_descs
;
360 /** size of the 'pages' array */
363 /** inline page vector */
364 struct page
*inline_pages
[FUSE_REQ_INLINE_PAGES
];
366 /** inline page-descriptor vector */
367 struct fuse_page_desc inline_page_descs
[FUSE_REQ_INLINE_PAGES
];
369 /** number of pages in vector */
372 /** File used in the request (or NULL) */
373 struct fuse_file
*ff
;
375 /** Inode used in the request or NULL */
378 /** AIO control block */
379 struct fuse_io_priv
*io
;
381 /** Link on fi->writepages */
382 struct list_head writepages_entry
;
384 /** Request completion callback */
385 void (*end
)(struct fuse_conn
*, struct fuse_req
*);
387 /** Request is stolen from fuse_file->reserved_req */
388 struct file
*stolen_file
;
392 /** Connection established */
395 /** Readers of the connection are waiting on this */
396 wait_queue_head_t waitq
;
398 /** The next unique request id */
401 /** The list of pending requests */
402 struct list_head pending
;
404 /** Pending interrupts */
405 struct list_head interrupts
;
407 /** Queue of pending forgets */
408 struct fuse_forget_link forget_list_head
;
409 struct fuse_forget_link
*forget_list_tail
;
411 /** Batching of FORGET requests (positive indicates FORGET batch) */
414 /** O_ASYNC requests */
415 struct fasync_struct
*fasync
;
419 /** Connection established */
422 /** Lock protecting accessess to members of this structure */
425 /** The list of requests being processed */
426 struct list_head processing
;
428 /** The list of requests under I/O */
433 * Fuse device instance
436 /** Fuse connection for this device */
437 struct fuse_conn
*fc
;
439 /** Processing queue */
440 struct fuse_pqueue pq
;
442 /** list entry on fc->devices */
443 struct list_head entry
;
449 * This structure is created, when the filesystem is mounted, and is
450 * destroyed, when the client device is closed and the filesystem is
454 /** Lock protecting accessess to members of this structure */
460 /** Number of fuse_dev's */
465 /** The user id for this mount */
468 /** The group id for this mount */
471 /** The fuse mount flags for this mount */
474 /** Maximum read size */
477 /** Maximum write size */
481 struct fuse_iqueue iq
;
483 /** The next unique kernel file handle */
486 /** rbtree of fuse_files waiting for poll events indexed by ph */
487 struct rb_root polled_files
;
489 /** Maximum number of outstanding background requests */
490 unsigned max_background
;
492 /** Number of background requests at which congestion starts */
493 unsigned congestion_threshold
;
495 /** Number of requests currently in the background */
496 unsigned num_background
;
498 /** Number of background requests currently queued for userspace */
499 unsigned active_background
;
501 /** The list of background requests set aside for later queuing */
502 struct list_head bg_queue
;
504 /** Flag indicating that INIT reply has been received. Allocating
505 * any fuse request will be suspended until the flag is set */
508 /** Flag indicating if connection is blocked. This will be
509 the case before the INIT reply is received, and if there
510 are too many outstading backgrounds requests */
513 /** waitq for blocked connection */
514 wait_queue_head_t blocked_waitq
;
516 /** waitq for reserved requests */
517 wait_queue_head_t reserved_req_waitq
;
519 /** Connection established, cleared on umount, connection
520 abort and device release */
523 /** Connection failed (version mismatch). Cannot race with
524 setting other bitfields since it is only set once in INIT
525 reply, before any other request, and never cleared */
526 unsigned conn_error
:1;
528 /** Connection successful. Only set in INIT */
529 unsigned conn_init
:1;
531 /** Do readpages asynchronously? Only set in INIT */
532 unsigned async_read
:1;
534 /** Do not send separate SETATTR request before open(O_TRUNC) */
535 unsigned atomic_o_trunc
:1;
537 /** Filesystem supports NFS exporting. Only set in INIT */
538 unsigned export_support
:1;
540 /** Set if bdi is valid */
541 unsigned bdi_initialized
:1;
543 /** write-back cache policy (default is write-through) */
544 unsigned writeback_cache
:1;
546 /** allow parallel lookups and readdir (default is serialized) */
547 unsigned parallel_dirops
:1;
550 * The following bitfields are only for optimization purposes
551 * and hence races in setting them will not cause malfunction
554 /** Is open/release not implemented by fs? */
557 /** Is fsync not implemented by fs? */
560 /** Is fsyncdir not implemented by fs? */
561 unsigned no_fsyncdir
:1;
563 /** Is flush not implemented by fs? */
566 /** Is setxattr not implemented by fs? */
567 unsigned no_setxattr
:1;
569 /** Is getxattr not implemented by fs? */
570 unsigned no_getxattr
:1;
572 /** Is listxattr not implemented by fs? */
573 unsigned no_listxattr
:1;
575 /** Is removexattr not implemented by fs? */
576 unsigned no_removexattr
:1;
578 /** Are posix file locking primitives not implemented by fs? */
581 /** Is access not implemented by fs? */
582 unsigned no_access
:1;
584 /** Is create not implemented by fs? */
585 unsigned no_create
:1;
587 /** Is interrupt not implemented by fs? */
588 unsigned no_interrupt
:1;
590 /** Is bmap not implemented by fs? */
593 /** Is poll not implemented by fs? */
596 /** Do multi-page cached writes */
597 unsigned big_writes
:1;
599 /** Don't apply umask to creation modes */
600 unsigned dont_mask
:1;
602 /** Are BSD file locking primitives not implemented by fs? */
605 /** Is fallocate not implemented by fs? */
606 unsigned no_fallocate
:1;
608 /** Is rename with flags implemented by fs? */
609 unsigned no_rename2
:1;
611 /** Use enhanced/automatic page cache invalidation. */
612 unsigned auto_inval_data
:1;
614 /** Does the filesystem support readdirplus? */
615 unsigned do_readdirplus
:1;
617 /** Does the filesystem want adaptive readdirplus? */
618 unsigned readdirplus_auto
:1;
620 /** Does the filesystem support asynchronous direct-IO submission? */
621 unsigned async_dio
:1;
623 /** Is lseek not implemented by fs? */
626 /** The number of requests waiting for completion */
627 atomic_t num_waiting
;
629 /** Negotiated minor version */
632 /** Backing dev info */
633 struct backing_dev_info bdi
;
635 /** Entry on the fuse_conn_list */
636 struct list_head entry
;
638 /** Device ID from super block */
641 /** Dentries in the control filesystem */
642 struct dentry
*ctl_dentry
[FUSE_CTL_NUM_DENTRIES
];
644 /** number of dentries used in the above array */
647 /** Key for lock owner ID scrambling */
650 /** Reserved request for the DESTROY message */
651 struct fuse_req
*destroy_req
;
653 /** Version counter for attribute changes */
656 /** Called on final put */
657 void (*release
)(struct fuse_conn
*);
659 /** Super block for this connection. */
660 struct super_block
*sb
;
662 /** Read/write semaphore to hold when accessing sb. */
663 struct rw_semaphore killsb
;
665 /** List of device instances belonging to this connection */
666 struct list_head devices
;
669 static inline struct fuse_conn
*get_fuse_conn_super(struct super_block
*sb
)
671 return sb
->s_fs_info
;
674 static inline struct fuse_conn
*get_fuse_conn(struct inode
*inode
)
676 return get_fuse_conn_super(inode
->i_sb
);
679 static inline struct fuse_inode
*get_fuse_inode(struct inode
*inode
)
681 return container_of(inode
, struct fuse_inode
, inode
);
684 static inline u64
get_node_id(struct inode
*inode
)
686 return get_fuse_inode(inode
)->nodeid
;
689 /** Device operations */
690 extern const struct file_operations fuse_dev_operations
;
692 extern const struct dentry_operations fuse_dentry_operations
;
695 * Inode to nodeid comparison.
697 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
);
700 * Get a filled in inode
702 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
703 int generation
, struct fuse_attr
*attr
,
704 u64 attr_valid
, u64 attr_version
);
706 int fuse_lookup_name(struct super_block
*sb
, u64 nodeid
, struct qstr
*name
,
707 struct fuse_entry_out
*outarg
, struct inode
**inode
);
710 * Send FORGET command
712 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
713 u64 nodeid
, u64 nlookup
);
715 struct fuse_forget_link
*fuse_alloc_forget(void);
717 /* Used by READDIRPLUS */
718 void fuse_force_forget(struct file
*file
, u64 nodeid
);
721 * Initialize READ or READDIR request
723 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
724 loff_t pos
, size_t count
, int opcode
);
727 * Send OPEN or OPENDIR request
729 int fuse_open_common(struct inode
*inode
, struct file
*file
, bool isdir
);
731 struct fuse_file
*fuse_file_alloc(struct fuse_conn
*fc
);
732 struct fuse_file
*fuse_file_get(struct fuse_file
*ff
);
733 void fuse_file_free(struct fuse_file
*ff
);
734 void fuse_finish_open(struct inode
*inode
, struct file
*file
);
736 void fuse_sync_release(struct fuse_file
*ff
, int flags
);
739 * Send RELEASE or RELEASEDIR request
741 void fuse_release_common(struct file
*file
, int opcode
);
744 * Send FSYNC or FSYNCDIR request
746 int fuse_fsync_common(struct file
*file
, loff_t start
, loff_t end
,
747 int datasync
, int isdir
);
752 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
753 struct fuse_notify_poll_wakeup_out
*outarg
);
756 * Initialize file operations on a regular file
758 void fuse_init_file_inode(struct inode
*inode
);
761 * Initialize inode operations on regular files and special files
763 void fuse_init_common(struct inode
*inode
);
766 * Initialize inode and file operations on a directory
768 void fuse_init_dir(struct inode
*inode
);
771 * Initialize inode operations on a symlink
773 void fuse_init_symlink(struct inode
*inode
);
776 * Change attributes of an inode
778 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
779 u64 attr_valid
, u64 attr_version
);
781 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
785 * Initialize the client device
787 int fuse_dev_init(void);
790 * Cleanup the client device
792 void fuse_dev_cleanup(void);
794 int fuse_ctl_init(void);
795 void __exit
fuse_ctl_cleanup(void);
800 struct fuse_req
*fuse_request_alloc(unsigned npages
);
802 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
);
807 void fuse_request_free(struct fuse_req
*req
);
810 * Get a request, may fail with -ENOMEM,
811 * caller should specify # elements in req->pages[] explicitly
813 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
);
814 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
818 * Increment reference count on request
820 void __fuse_get_request(struct fuse_req
*req
);
823 * Gets a requests for a file operation, always succeeds
825 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
829 * Decrement reference count of a request. If count goes to zero free
832 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
);
835 * Send a request (synchronous)
837 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
);
840 * Simple request sending that does request allocation and freeing
842 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
);
845 * Send a request in the background
847 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
);
849 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
850 struct fuse_req
*req
);
852 /* Abort all requests */
853 void fuse_abort_conn(struct fuse_conn
*fc
);
856 * Invalidate inode attributes
858 void fuse_invalidate_attr(struct inode
*inode
);
860 void fuse_invalidate_entry_cache(struct dentry
*entry
);
862 void fuse_invalidate_atime(struct inode
*inode
);
865 * Acquire reference to fuse_conn
867 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
);
870 * Initialize fuse_conn
872 void fuse_conn_init(struct fuse_conn
*fc
);
875 * Release reference to fuse_conn
877 void fuse_conn_put(struct fuse_conn
*fc
);
879 struct fuse_dev
*fuse_dev_alloc(struct fuse_conn
*fc
);
880 void fuse_dev_free(struct fuse_dev
*fud
);
883 * Add connection to control filesystem
885 int fuse_ctl_add_conn(struct fuse_conn
*fc
);
888 * Remove connection from control filesystem
890 void fuse_ctl_remove_conn(struct fuse_conn
*fc
);
893 * Is file type valid?
895 int fuse_valid_type(int m
);
898 * Is current process allowed to perform filesystem operation?
900 int fuse_allow_current_process(struct fuse_conn
*fc
);
902 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
);
904 int fuse_update_attributes(struct inode
*inode
, struct kstat
*stat
,
905 struct file
*file
, bool *refreshed
);
907 void fuse_flush_writepages(struct inode
*inode
);
909 void fuse_set_nowrite(struct inode
*inode
);
910 void fuse_release_nowrite(struct inode
*inode
);
912 u64
fuse_get_attr_version(struct fuse_conn
*fc
);
915 * File-system tells the kernel to invalidate cache for the given node id.
917 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
918 loff_t offset
, loff_t len
);
921 * File-system tells the kernel to invalidate parent attributes and
922 * the dentry matching parent/name.
924 * If the child_nodeid is non-zero and:
925 * - matches the inode number for the dentry matching parent/name,
926 * - is not a mount point
927 * - is a file or oan empty directory
928 * then the dentry is unhashed (d_delete()).
930 int fuse_reverse_inval_entry(struct super_block
*sb
, u64 parent_nodeid
,
931 u64 child_nodeid
, struct qstr
*name
);
933 int fuse_do_open(struct fuse_conn
*fc
, u64 nodeid
, struct file
*file
,
937 * fuse_direct_io() flags
940 /** If set, it is WRITE; otherwise - READ */
941 #define FUSE_DIO_WRITE (1 << 0)
943 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
944 #define FUSE_DIO_CUSE (1 << 1)
946 ssize_t
fuse_direct_io(struct fuse_io_priv
*io
, struct iov_iter
*iter
,
947 loff_t
*ppos
, int flags
);
948 long fuse_do_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
,
950 long fuse_ioctl_common(struct file
*file
, unsigned int cmd
,
951 unsigned long arg
, unsigned int flags
);
952 unsigned fuse_file_poll(struct file
*file
, poll_table
*wait
);
953 int fuse_dev_release(struct inode
*inode
, struct file
*file
);
955 bool fuse_write_update_size(struct inode
*inode
, loff_t pos
);
957 int fuse_flush_times(struct inode
*inode
, struct fuse_file
*ff
);
958 int fuse_write_inode(struct inode
*inode
, struct writeback_control
*wbc
);
960 int fuse_do_setattr(struct inode
*inode
, struct iattr
*attr
,
963 void fuse_set_initialized(struct fuse_conn
*fc
);
965 void fuse_unlock_inode(struct inode
*inode
);
966 void fuse_lock_inode(struct inode
*inode
);
968 #endif /* _FS_FUSE_I_H */