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>
27 #include <linux/pid_namespace.h>
28 #include <linux/user_namespace.h>
30 /** Max number of pages that can be used in a single read request */
31 #define FUSE_MAX_PAGES_PER_REQ 32
33 /** Bias for fi->writectr, meaning new writepages must not be sent */
34 #define FUSE_NOWRITE INT_MIN
36 /** It could be as large as PATH_MAX, but would that have any uses? */
37 #define FUSE_NAME_MAX 1024
39 /** Number of dentries for each connection in the control filesystem */
40 #define FUSE_CTL_NUM_DENTRIES 5
42 /** Number of page pointers embedded in fuse_req */
43 #define FUSE_REQ_INLINE_PAGES 1
45 /** List of active connections */
46 extern struct list_head fuse_conn_list
;
48 /** Global mutex protecting fuse_conn_list and the control filesystem */
49 extern struct mutex fuse_mutex
;
51 /** Module parameters */
52 extern unsigned max_user_bgreq
;
53 extern unsigned max_user_congthresh
;
55 /* One forget request */
56 struct fuse_forget_link
{
57 struct fuse_forget_one forget_one
;
58 struct fuse_forget_link
*next
;
66 /** Unique ID, which identifies the inode between userspace
70 /** Number of lookups on this inode */
73 /** The request used for sending the FORGET message */
74 struct fuse_forget_link
*forget
;
76 /** Time in jiffies until the file attributes are valid */
79 /** The sticky bit in inode->i_mode may have been removed, so
80 preserve the original mode */
83 /** 64 bit inode number */
86 /** Version of last attribute change */
89 /** Files usable in writepage. Protected by fc->lock */
90 struct list_head write_files
;
92 /** Writepages pending on truncate or fsync */
93 struct list_head queued_writes
;
95 /** Number of sent writes, a negative bias (FUSE_NOWRITE)
96 * means more writes are blocked */
99 /** Waitq for writepage completion */
100 wait_queue_head_t page_waitq
;
102 /** List of writepage requestst (pending or sent) */
103 struct list_head writepages
;
105 /** Miscellaneous bits describing inode state */
108 /** Lock for serializing lookup and readdir for back compatibility*/
112 /** FUSE inode state bits */
114 /** Advise readdirplus */
115 FUSE_I_ADVISE_RDPLUS
,
116 /** Initialized with readdirplus */
118 /** An operation changing file size is in progress */
119 FUSE_I_SIZE_UNSTABLE
,
124 /** FUSE specific file data */
126 /** Fuse connection for this file */
127 struct fuse_conn
*fc
;
129 /** Request reserved for flush and release */
130 struct fuse_req
*reserved_req
;
132 /** Kernel file handle guaranteed to be unique */
135 /** File handle used by userspace */
138 /** Node id of this file */
144 /** FOPEN_* flags returned by open */
147 /** Entry on inode's write_files list */
148 struct list_head write_entry
;
150 /** RB node to be linked on fuse_conn->polled_files */
151 struct rb_node polled_node
;
153 /** Wait queue head for poll */
154 wait_queue_head_t poll_wait
;
156 /** Has flock been performed on this file? */
160 /** One input argument of a request */
166 /** The request input */
168 /** The request header */
169 struct fuse_in_header h
;
171 /** True if the data for the last argument is in req->pages */
174 /** Number of arguments */
177 /** Array of arguments */
178 struct fuse_in_arg args
[3];
181 /** One output argument of a request */
187 /** The request output */
189 /** Header returned from userspace */
190 struct fuse_out_header h
;
193 * The following bitfields are not changed during the request
197 /** Last argument is variable length (can be shorter than
201 /** Last argument is a list of pages to copy data to */
204 /** Zero partially or not copied pages */
205 unsigned page_zeroing
:1;
207 /** Pages may be replaced with new ones */
208 unsigned page_replace
:1;
210 /** Number or arguments */
213 /** Array of arguments */
214 struct fuse_arg args
[2];
217 /** FUSE page descriptor */
218 struct fuse_page_desc
{
230 struct fuse_in_arg args
[3];
236 struct fuse_arg args
[2];
240 #define FUSE_ARGS(args) struct fuse_args args = {}
242 /** The request IO state (for asynchronous processing) */
243 struct fuse_io_priv
{
255 struct completion
*done
;
259 #define FUSE_IO_PRIV_SYNC(f) \
261 .refcnt = { ATOMIC_INIT(1) }, \
269 * FR_ISREPLY: set if the request has reply
270 * FR_FORCE: force sending of the request even if interrupted
271 * FR_BACKGROUND: request is sent in the background
272 * FR_WAITING: request is counted as "waiting"
273 * FR_ABORTED: the request was aborted
274 * FR_INTERRUPTED: the request has been interrupted
275 * FR_LOCKED: data is being copied to/from the request
276 * FR_PENDING: request is not yet in userspace
277 * FR_SENT: request is in userspace, waiting for an answer
278 * FR_FINISHED: request is finished
279 * FR_PRIVATE: request is on private list
296 * A request to the client
298 * .waitq.lock protects the following fields:
300 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
303 /** This can be on either pending processing or io lists in
305 struct list_head list
;
307 /** Entry on the interrupts list */
308 struct list_head intr_entry
;
313 /** Unique ID for the interrupt request */
316 /* Request flags, updated with test/set/clear_bit() */
319 /** The request input */
322 /** The request output */
325 /** Used to wake up the task waiting for completion of request*/
326 wait_queue_head_t waitq
;
328 /** Data for asynchronous requests */
331 struct fuse_release_in in
;
334 struct fuse_init_in init_in
;
335 struct fuse_init_out init_out
;
336 struct cuse_init_in cuse_init_in
;
338 struct fuse_read_in in
;
342 struct fuse_write_in in
;
343 struct fuse_write_out out
;
344 struct fuse_req
*next
;
346 struct fuse_notify_retrieve_in retrieve_in
;
352 /** page-descriptor vector */
353 struct fuse_page_desc
*page_descs
;
355 /** size of the 'pages' array */
358 /** inline page vector */
359 struct page
*inline_pages
[FUSE_REQ_INLINE_PAGES
];
361 /** inline page-descriptor vector */
362 struct fuse_page_desc inline_page_descs
[FUSE_REQ_INLINE_PAGES
];
364 /** number of pages in vector */
367 /** File used in the request (or NULL) */
368 struct fuse_file
*ff
;
370 /** Inode used in the request or NULL */
373 /** AIO control block */
374 struct fuse_io_priv
*io
;
376 /** Link on fi->writepages */
377 struct list_head writepages_entry
;
379 /** Request completion callback */
380 void (*end
)(struct fuse_conn
*, struct fuse_req
*);
382 /** Request is stolen from fuse_file->reserved_req */
383 struct file
*stolen_file
;
387 /** Connection established */
390 /** Readers of the connection are waiting on this */
391 wait_queue_head_t waitq
;
393 /** The next unique request id */
396 /** The list of pending requests */
397 struct list_head pending
;
399 /** Pending interrupts */
400 struct list_head interrupts
;
402 /** Queue of pending forgets */
403 struct fuse_forget_link forget_list_head
;
404 struct fuse_forget_link
*forget_list_tail
;
406 /** Batching of FORGET requests (positive indicates FORGET batch) */
409 /** O_ASYNC requests */
410 struct fasync_struct
*fasync
;
414 /** Connection established */
417 /** Lock protecting accessess to members of this structure */
420 /** The list of requests being processed */
421 struct list_head processing
;
423 /** The list of requests under I/O */
428 * Fuse device instance
431 /** Fuse connection for this device */
432 struct fuse_conn
*fc
;
434 /** Processing queue */
435 struct fuse_pqueue pq
;
437 /** list entry on fc->devices */
438 struct list_head entry
;
444 * This structure is created, when the filesystem is mounted, and is
445 * destroyed, when the client device is closed and the filesystem is
449 /** Lock protecting accessess to members of this structure */
455 /** Number of fuse_dev's */
460 /** The user id for this mount */
463 /** The group id for this mount */
466 /** The pid namespace for this mount */
467 struct pid_namespace
*pid_ns
;
469 /** The user namespace for this mount */
470 struct user_namespace
*user_ns
;
472 /** Maximum read size */
475 /** Maximum write size */
479 struct fuse_iqueue iq
;
481 /** The next unique kernel file handle */
484 /** rbtree of fuse_files waiting for poll events indexed by ph */
485 struct rb_root polled_files
;
487 /** Maximum number of outstanding background requests */
488 unsigned max_background
;
490 /** Number of background requests at which congestion starts */
491 unsigned congestion_threshold
;
493 /** Number of requests currently in the background */
494 unsigned num_background
;
496 /** Number of background requests currently queued for userspace */
497 unsigned active_background
;
499 /** The list of background requests set aside for later queuing */
500 struct list_head bg_queue
;
502 /** Flag indicating that INIT reply has been received. Allocating
503 * any fuse request will be suspended until the flag is set */
506 /** Flag indicating if connection is blocked. This will be
507 the case before the INIT reply is received, and if there
508 are too many outstading backgrounds requests */
511 /** waitq for blocked connection */
512 wait_queue_head_t blocked_waitq
;
514 /** waitq for reserved requests */
515 wait_queue_head_t reserved_req_waitq
;
517 /** Connection established, cleared on umount, connection
518 abort and device release */
521 /** Connection failed (version mismatch). Cannot race with
522 setting other bitfields since it is only set once in INIT
523 reply, before any other request, and never cleared */
524 unsigned conn_error
:1;
526 /** Connection successful. Only set in INIT */
527 unsigned conn_init
:1;
529 /** Do readpages asynchronously? Only set in INIT */
530 unsigned async_read
:1;
532 /** Do not send separate SETATTR request before open(O_TRUNC) */
533 unsigned atomic_o_trunc
:1;
535 /** Filesystem supports NFS exporting. Only set in INIT */
536 unsigned export_support
:1;
538 /** Set if bdi is valid */
539 unsigned bdi_initialized
:1;
541 /** write-back cache policy (default is write-through) */
542 unsigned writeback_cache
:1;
544 /** allow parallel lookups and readdir (default is serialized) */
545 unsigned parallel_dirops
:1;
547 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
548 unsigned handle_killpriv
: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 /** Does the filesystem support posix acls? */
628 unsigned posix_acl
:1;
630 /** Check permissions based on the file mode or not? */
631 unsigned default_permissions
:1;
633 /** Allow other than the mounter user to access the filesystem ? */
634 unsigned allow_other
:1;
636 /** The number of requests waiting for completion */
637 atomic_t num_waiting
;
639 /** Negotiated minor version */
642 /** Backing dev info */
643 struct backing_dev_info bdi
;
645 /** Entry on the fuse_conn_list */
646 struct list_head entry
;
648 /** Device ID from super block */
651 /** Dentries in the control filesystem */
652 struct dentry
*ctl_dentry
[FUSE_CTL_NUM_DENTRIES
];
654 /** number of dentries used in the above array */
657 /** Key for lock owner ID scrambling */
660 /** Reserved request for the DESTROY message */
661 struct fuse_req
*destroy_req
;
663 /** Version counter for attribute changes */
666 /** Called on final put */
667 void (*release
)(struct fuse_conn
*);
669 /** Super block for this connection. */
670 struct super_block
*sb
;
672 /** Read/write semaphore to hold when accessing sb. */
673 struct rw_semaphore killsb
;
675 /** List of device instances belonging to this connection */
676 struct list_head devices
;
679 static inline struct fuse_conn
*get_fuse_conn_super(struct super_block
*sb
)
681 return sb
->s_fs_info
;
684 static inline struct fuse_conn
*get_fuse_conn(struct inode
*inode
)
686 return get_fuse_conn_super(inode
->i_sb
);
689 static inline struct fuse_inode
*get_fuse_inode(struct inode
*inode
)
691 return container_of(inode
, struct fuse_inode
, inode
);
694 static inline u64
get_node_id(struct inode
*inode
)
696 return get_fuse_inode(inode
)->nodeid
;
699 /** Device operations */
700 extern const struct file_operations fuse_dev_operations
;
702 extern const struct dentry_operations fuse_dentry_operations
;
703 extern const struct dentry_operations fuse_root_dentry_operations
;
706 * Inode to nodeid comparison.
708 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
);
711 * Get a filled in inode
713 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
714 int generation
, struct fuse_attr
*attr
,
715 u64 attr_valid
, u64 attr_version
);
717 int fuse_lookup_name(struct super_block
*sb
, u64 nodeid
, const struct qstr
*name
,
718 struct fuse_entry_out
*outarg
, struct inode
**inode
);
721 * Send FORGET command
723 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
724 u64 nodeid
, u64 nlookup
);
726 struct fuse_forget_link
*fuse_alloc_forget(void);
728 /* Used by READDIRPLUS */
729 void fuse_force_forget(struct file
*file
, u64 nodeid
);
732 * Initialize READ or READDIR request
734 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
735 loff_t pos
, size_t count
, int opcode
);
738 * Send OPEN or OPENDIR request
740 int fuse_open_common(struct inode
*inode
, struct file
*file
, bool isdir
);
742 struct fuse_file
*fuse_file_alloc(struct fuse_conn
*fc
);
743 struct fuse_file
*fuse_file_get(struct fuse_file
*ff
);
744 void fuse_file_free(struct fuse_file
*ff
);
745 void fuse_finish_open(struct inode
*inode
, struct file
*file
);
747 void fuse_sync_release(struct fuse_file
*ff
, int flags
);
750 * Send RELEASE or RELEASEDIR request
752 void fuse_release_common(struct file
*file
, int opcode
);
755 * Send FSYNC or FSYNCDIR request
757 int fuse_fsync_common(struct file
*file
, loff_t start
, loff_t end
,
758 int datasync
, int isdir
);
763 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
764 struct fuse_notify_poll_wakeup_out
*outarg
);
767 * Initialize file operations on a regular file
769 void fuse_init_file_inode(struct inode
*inode
);
772 * Initialize inode operations on regular files and special files
774 void fuse_init_common(struct inode
*inode
);
777 * Initialize inode and file operations on a directory
779 void fuse_init_dir(struct inode
*inode
);
782 * Initialize inode operations on a symlink
784 void fuse_init_symlink(struct inode
*inode
);
787 * Change attributes of an inode
789 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
790 u64 attr_valid
, u64 attr_version
);
792 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
796 * Initialize the client device
798 int fuse_dev_init(void);
801 * Cleanup the client device
803 void fuse_dev_cleanup(void);
805 int fuse_ctl_init(void);
806 void __exit
fuse_ctl_cleanup(void);
811 struct fuse_req
*fuse_request_alloc(unsigned npages
);
813 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
);
818 void fuse_request_free(struct fuse_req
*req
);
821 * Get a request, may fail with -ENOMEM,
822 * caller should specify # elements in req->pages[] explicitly
824 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
);
825 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
829 * Increment reference count on request
831 void __fuse_get_request(struct fuse_req
*req
);
834 * Gets a requests for a file operation, always succeeds
836 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
840 * Decrement reference count of a request. If count goes to zero free
843 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
);
846 * Send a request (synchronous)
848 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
);
851 * Simple request sending that does request allocation and freeing
853 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
);
856 * Send a request in the background
858 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
);
860 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
861 struct fuse_req
*req
);
863 /* Abort all requests */
864 void fuse_abort_conn(struct fuse_conn
*fc
);
867 * Invalidate inode attributes
869 void fuse_invalidate_attr(struct inode
*inode
);
871 void fuse_invalidate_entry_cache(struct dentry
*entry
);
873 void fuse_invalidate_atime(struct inode
*inode
);
876 * Acquire reference to fuse_conn
878 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
);
881 * Initialize fuse_conn
883 void fuse_conn_init(struct fuse_conn
*fc
, struct user_namespace
*user_ns
);
886 * Release reference to fuse_conn
888 void fuse_conn_put(struct fuse_conn
*fc
);
890 struct fuse_dev
*fuse_dev_alloc(struct fuse_conn
*fc
);
891 void fuse_dev_free(struct fuse_dev
*fud
);
894 * Add connection to control filesystem
896 int fuse_ctl_add_conn(struct fuse_conn
*fc
);
899 * Remove connection from control filesystem
901 void fuse_ctl_remove_conn(struct fuse_conn
*fc
);
904 * Is file type valid?
906 int fuse_valid_type(int m
);
909 * Is current process allowed to perform filesystem operation?
911 int fuse_allow_current_process(struct fuse_conn
*fc
);
913 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
);
915 void fuse_update_ctime(struct inode
*inode
);
917 int fuse_update_attributes(struct inode
*inode
, struct kstat
*stat
,
918 struct file
*file
, bool *refreshed
);
920 void fuse_flush_writepages(struct inode
*inode
);
922 void fuse_set_nowrite(struct inode
*inode
);
923 void fuse_release_nowrite(struct inode
*inode
);
925 u64
fuse_get_attr_version(struct fuse_conn
*fc
);
928 * File-system tells the kernel to invalidate cache for the given node id.
930 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
931 loff_t offset
, loff_t len
);
934 * File-system tells the kernel to invalidate parent attributes and
935 * the dentry matching parent/name.
937 * If the child_nodeid is non-zero and:
938 * - matches the inode number for the dentry matching parent/name,
939 * - is not a mount point
940 * - is a file or oan empty directory
941 * then the dentry is unhashed (d_delete()).
943 int fuse_reverse_inval_entry(struct super_block
*sb
, u64 parent_nodeid
,
944 u64 child_nodeid
, struct qstr
*name
);
946 int fuse_do_open(struct fuse_conn
*fc
, u64 nodeid
, struct file
*file
,
950 * fuse_direct_io() flags
953 /** If set, it is WRITE; otherwise - READ */
954 #define FUSE_DIO_WRITE (1 << 0)
956 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
957 #define FUSE_DIO_CUSE (1 << 1)
959 ssize_t
fuse_direct_io(struct fuse_io_priv
*io
, struct iov_iter
*iter
,
960 loff_t
*ppos
, int flags
);
961 long fuse_do_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
,
963 long fuse_ioctl_common(struct file
*file
, unsigned int cmd
,
964 unsigned long arg
, unsigned int flags
);
965 unsigned fuse_file_poll(struct file
*file
, poll_table
*wait
);
966 int fuse_dev_release(struct inode
*inode
, struct file
*file
);
968 bool fuse_write_update_size(struct inode
*inode
, loff_t pos
);
970 int fuse_flush_times(struct inode
*inode
, struct fuse_file
*ff
);
971 int fuse_write_inode(struct inode
*inode
, struct writeback_control
*wbc
);
973 int fuse_do_setattr(struct dentry
*dentry
, struct iattr
*attr
,
976 void fuse_set_initialized(struct fuse_conn
*fc
);
978 void fuse_unlock_inode(struct inode
*inode
);
979 void fuse_lock_inode(struct inode
*inode
);
981 int fuse_setxattr(struct inode
*inode
, const char *name
, const void *value
,
982 size_t size
, int flags
);
983 ssize_t
fuse_getxattr(struct inode
*inode
, const char *name
, void *value
,
985 ssize_t
fuse_listxattr(struct dentry
*entry
, char *list
, size_t size
);
986 int fuse_removexattr(struct inode
*inode
, const char *name
);
987 extern const struct xattr_handler
*fuse_xattr_handlers
[];
988 extern const struct xattr_handler
*fuse_acl_xattr_handlers
[];
991 struct posix_acl
*fuse_get_acl(struct inode
*inode
, int type
);
992 int fuse_set_acl(struct inode
*inode
, struct posix_acl
*acl
, int type
);
994 #endif /* _FS_FUSE_I_H */