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/refcount.h>
29 #include <linux/user_namespace.h>
31 /** Max number of pages that can be used in a single read request */
32 #define FUSE_MAX_PAGES_PER_REQ 32
34 /** Bias for fi->writectr, meaning new writepages must not be sent */
35 #define FUSE_NOWRITE INT_MIN
37 /** It could be as large as PATH_MAX, but would that have any uses? */
38 #define FUSE_NAME_MAX 1024
40 /** Number of dentries for each connection in the control filesystem */
41 #define FUSE_CTL_NUM_DENTRIES 5
43 /** Number of page pointers embedded in fuse_req */
44 #define FUSE_REQ_INLINE_PAGES 1
46 /** List of active connections */
47 extern struct list_head fuse_conn_list
;
49 /** Global mutex protecting fuse_conn_list and the control filesystem */
50 extern struct mutex fuse_mutex
;
52 /** Module parameters */
53 extern unsigned max_user_bgreq
;
54 extern unsigned max_user_congthresh
;
56 /* One forget request */
57 struct fuse_forget_link
{
58 struct fuse_forget_one forget_one
;
59 struct fuse_forget_link
*next
;
67 /** Unique ID, which identifies the inode between userspace
71 /** Number of lookups on this inode */
74 /** The request used for sending the FORGET message */
75 struct fuse_forget_link
*forget
;
77 /** Time in jiffies until the file attributes are valid */
80 /** The sticky bit in inode->i_mode may have been removed, so
81 preserve the original mode */
84 /** 64 bit inode number */
87 /** Version of last attribute change */
90 /** Files usable in writepage. Protected by fc->lock */
91 struct list_head write_files
;
93 /** Writepages pending on truncate or fsync */
94 struct list_head queued_writes
;
96 /** Number of sent writes, a negative bias (FUSE_NOWRITE)
97 * means more writes are blocked */
100 /** Waitq for writepage completion */
101 wait_queue_head_t page_waitq
;
103 /** List of writepage requestst (pending or sent) */
104 struct list_head writepages
;
106 /** Miscellaneous bits describing inode state */
109 /** Lock for serializing lookup and readdir for back compatibility*/
113 /** FUSE inode state bits */
115 /** Advise readdirplus */
116 FUSE_I_ADVISE_RDPLUS
,
117 /** Initialized with readdirplus */
119 /** An operation changing file size is in progress */
120 FUSE_I_SIZE_UNSTABLE
,
125 /** FUSE specific file data */
127 /** Fuse connection for this file */
128 struct fuse_conn
*fc
;
130 /** Request reserved for flush and release */
131 struct fuse_req
*reserved_req
;
133 /** Kernel file handle guaranteed to be unique */
136 /** File handle used by userspace */
139 /** Node id of this file */
145 /** FOPEN_* flags returned by open */
148 /** Entry on inode's write_files list */
149 struct list_head write_entry
;
151 /** RB node to be linked on fuse_conn->polled_files */
152 struct rb_node polled_node
;
154 /** Wait queue head for poll */
155 wait_queue_head_t poll_wait
;
157 /** Has flock been performed on this file? */
161 /** One input argument of a request */
167 /** The request input */
169 /** The request header */
170 struct fuse_in_header h
;
172 /** True if the data for the last argument is in req->pages */
175 /** Number of arguments */
178 /** Array of arguments */
179 struct fuse_in_arg args
[3];
182 /** One output argument of a request */
188 /** The request output */
190 /** Header returned from userspace */
191 struct fuse_out_header h
;
194 * The following bitfields are not changed during the request
198 /** Last argument is variable length (can be shorter than
202 /** Last argument is a list of pages to copy data to */
205 /** Zero partially or not copied pages */
206 unsigned page_zeroing
:1;
208 /** Pages may be replaced with new ones */
209 unsigned page_replace
:1;
211 /** Number or arguments */
214 /** Array of arguments */
215 struct fuse_arg args
[2];
218 /** FUSE page descriptor */
219 struct fuse_page_desc
{
231 struct fuse_in_arg args
[3];
237 struct fuse_arg args
[2];
241 #define FUSE_ARGS(args) struct fuse_args args = {}
243 /** The request IO state (for asynchronous processing) */
244 struct fuse_io_priv
{
257 struct completion
*done
;
261 #define FUSE_IO_PRIV_SYNC(f) \
263 .refcnt = KREF_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 pid namespace for this mount */
469 struct pid_namespace
*pid_ns
;
471 /** The user namespace for this mount */
472 struct user_namespace
*user_ns
;
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 /** write-back cache policy (default is write-through) */
541 unsigned writeback_cache
:1;
543 /** allow parallel lookups and readdir (default is serialized) */
544 unsigned parallel_dirops
:1;
546 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
547 unsigned handle_killpriv
: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 /** Does the filesystem support posix acls? */
627 unsigned posix_acl
:1;
629 /** Check permissions based on the file mode or not? */
630 unsigned default_permissions
:1;
632 /** Allow other than the mounter user to access the filesystem ? */
633 unsigned allow_other
:1;
635 /** The number of requests waiting for completion */
636 atomic_t num_waiting
;
638 /** Negotiated minor version */
641 /** Entry on the fuse_conn_list */
642 struct list_head entry
;
644 /** Device ID from super block */
647 /** Dentries in the control filesystem */
648 struct dentry
*ctl_dentry
[FUSE_CTL_NUM_DENTRIES
];
650 /** number of dentries used in the above array */
653 /** Key for lock owner ID scrambling */
656 /** Reserved request for the DESTROY message */
657 struct fuse_req
*destroy_req
;
659 /** Version counter for attribute changes */
662 /** Called on final put */
663 void (*release
)(struct fuse_conn
*);
665 /** Super block for this connection. */
666 struct super_block
*sb
;
668 /** Read/write semaphore to hold when accessing sb. */
669 struct rw_semaphore killsb
;
671 /** List of device instances belonging to this connection */
672 struct list_head devices
;
675 static inline struct fuse_conn
*get_fuse_conn_super(struct super_block
*sb
)
677 return sb
->s_fs_info
;
680 static inline struct fuse_conn
*get_fuse_conn(struct inode
*inode
)
682 return get_fuse_conn_super(inode
->i_sb
);
685 static inline struct fuse_inode
*get_fuse_inode(struct inode
*inode
)
687 return container_of(inode
, struct fuse_inode
, inode
);
690 static inline u64
get_node_id(struct inode
*inode
)
692 return get_fuse_inode(inode
)->nodeid
;
695 /** Device operations */
696 extern const struct file_operations fuse_dev_operations
;
698 extern const struct dentry_operations fuse_dentry_operations
;
699 extern const struct dentry_operations fuse_root_dentry_operations
;
702 * Inode to nodeid comparison.
704 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
);
707 * Get a filled in inode
709 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
710 int generation
, struct fuse_attr
*attr
,
711 u64 attr_valid
, u64 attr_version
);
713 int fuse_lookup_name(struct super_block
*sb
, u64 nodeid
, const struct qstr
*name
,
714 struct fuse_entry_out
*outarg
, struct inode
**inode
);
717 * Send FORGET command
719 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
720 u64 nodeid
, u64 nlookup
);
722 struct fuse_forget_link
*fuse_alloc_forget(void);
724 /* Used by READDIRPLUS */
725 void fuse_force_forget(struct file
*file
, u64 nodeid
);
728 * Initialize READ or READDIR request
730 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
731 loff_t pos
, size_t count
, int opcode
);
734 * Send OPEN or OPENDIR request
736 int fuse_open_common(struct inode
*inode
, struct file
*file
, bool isdir
);
738 struct fuse_file
*fuse_file_alloc(struct fuse_conn
*fc
);
739 void fuse_file_free(struct fuse_file
*ff
);
740 void fuse_finish_open(struct inode
*inode
, struct file
*file
);
742 void fuse_sync_release(struct fuse_file
*ff
, int flags
);
745 * Send RELEASE or RELEASEDIR request
747 void fuse_release_common(struct file
*file
, int opcode
);
750 * Send FSYNC or FSYNCDIR request
752 int fuse_fsync_common(struct file
*file
, loff_t start
, loff_t end
,
753 int datasync
, int isdir
);
758 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
759 struct fuse_notify_poll_wakeup_out
*outarg
);
762 * Initialize file operations on a regular file
764 void fuse_init_file_inode(struct inode
*inode
);
767 * Initialize inode operations on regular files and special files
769 void fuse_init_common(struct inode
*inode
);
772 * Initialize inode and file operations on a directory
774 void fuse_init_dir(struct inode
*inode
);
777 * Initialize inode operations on a symlink
779 void fuse_init_symlink(struct inode
*inode
);
782 * Change attributes of an inode
784 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
785 u64 attr_valid
, u64 attr_version
);
787 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
791 * Initialize the client device
793 int fuse_dev_init(void);
796 * Cleanup the client device
798 void fuse_dev_cleanup(void);
800 int fuse_ctl_init(void);
801 void __exit
fuse_ctl_cleanup(void);
806 struct fuse_req
*fuse_request_alloc(unsigned npages
);
808 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
);
813 void fuse_request_free(struct fuse_req
*req
);
816 * Get a request, may fail with -ENOMEM,
817 * caller should specify # elements in req->pages[] explicitly
819 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
);
820 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
824 * Increment reference count on request
826 void __fuse_get_request(struct fuse_req
*req
);
829 * Gets a requests for a file operation, always succeeds
831 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
835 * Decrement reference count of a request. If count goes to zero free
838 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
);
841 * Send a request (synchronous)
843 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
);
846 * Simple request sending that does request allocation and freeing
848 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
);
851 * Send a request in the background
853 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
);
855 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
856 struct fuse_req
*req
);
858 /* Abort all requests */
859 void fuse_abort_conn(struct fuse_conn
*fc
);
862 * Invalidate inode attributes
864 void fuse_invalidate_attr(struct inode
*inode
);
866 void fuse_invalidate_entry_cache(struct dentry
*entry
);
868 void fuse_invalidate_atime(struct inode
*inode
);
871 * Acquire reference to fuse_conn
873 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
);
876 * Initialize fuse_conn
878 void fuse_conn_init(struct fuse_conn
*fc
, struct user_namespace
*user_ns
);
881 * Release reference to fuse_conn
883 void fuse_conn_put(struct fuse_conn
*fc
);
885 struct fuse_dev
*fuse_dev_alloc(struct fuse_conn
*fc
);
886 void fuse_dev_free(struct fuse_dev
*fud
);
889 * Add connection to control filesystem
891 int fuse_ctl_add_conn(struct fuse_conn
*fc
);
894 * Remove connection from control filesystem
896 void fuse_ctl_remove_conn(struct fuse_conn
*fc
);
899 * Is file type valid?
901 int fuse_valid_type(int m
);
904 * Is current process allowed to perform filesystem operation?
906 int fuse_allow_current_process(struct fuse_conn
*fc
);
908 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
);
910 void fuse_update_ctime(struct inode
*inode
);
912 int fuse_update_attributes(struct inode
*inode
, struct kstat
*stat
,
913 struct file
*file
, bool *refreshed
);
915 void fuse_flush_writepages(struct inode
*inode
);
917 void fuse_set_nowrite(struct inode
*inode
);
918 void fuse_release_nowrite(struct inode
*inode
);
920 u64
fuse_get_attr_version(struct fuse_conn
*fc
);
923 * File-system tells the kernel to invalidate cache for the given node id.
925 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
926 loff_t offset
, loff_t len
);
929 * File-system tells the kernel to invalidate parent attributes and
930 * the dentry matching parent/name.
932 * If the child_nodeid is non-zero and:
933 * - matches the inode number for the dentry matching parent/name,
934 * - is not a mount point
935 * - is a file or oan empty directory
936 * then the dentry is unhashed (d_delete()).
938 int fuse_reverse_inval_entry(struct super_block
*sb
, u64 parent_nodeid
,
939 u64 child_nodeid
, struct qstr
*name
);
941 int fuse_do_open(struct fuse_conn
*fc
, u64 nodeid
, struct file
*file
,
945 * fuse_direct_io() flags
948 /** If set, it is WRITE; otherwise - READ */
949 #define FUSE_DIO_WRITE (1 << 0)
951 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
952 #define FUSE_DIO_CUSE (1 << 1)
954 ssize_t
fuse_direct_io(struct fuse_io_priv
*io
, struct iov_iter
*iter
,
955 loff_t
*ppos
, int flags
);
956 long fuse_do_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
,
958 long fuse_ioctl_common(struct file
*file
, unsigned int cmd
,
959 unsigned long arg
, unsigned int flags
);
960 unsigned fuse_file_poll(struct file
*file
, poll_table
*wait
);
961 int fuse_dev_release(struct inode
*inode
, struct file
*file
);
963 bool fuse_write_update_size(struct inode
*inode
, loff_t pos
);
965 int fuse_flush_times(struct inode
*inode
, struct fuse_file
*ff
);
966 int fuse_write_inode(struct inode
*inode
, struct writeback_control
*wbc
);
968 int fuse_do_setattr(struct dentry
*dentry
, struct iattr
*attr
,
971 void fuse_set_initialized(struct fuse_conn
*fc
);
973 void fuse_unlock_inode(struct inode
*inode
);
974 void fuse_lock_inode(struct inode
*inode
);
976 int fuse_setxattr(struct inode
*inode
, const char *name
, const void *value
,
977 size_t size
, int flags
);
978 ssize_t
fuse_getxattr(struct inode
*inode
, const char *name
, void *value
,
980 ssize_t
fuse_listxattr(struct dentry
*entry
, char *list
, size_t size
);
981 int fuse_removexattr(struct inode
*inode
, const char *name
);
982 extern const struct xattr_handler
*fuse_xattr_handlers
[];
983 extern const struct xattr_handler
*fuse_acl_xattr_handlers
[];
986 struct posix_acl
*fuse_get_acl(struct inode
*inode
, int type
);
987 int fuse_set_acl(struct inode
*inode
, struct posix_acl
*acl
, int type
);
989 #endif /* _FS_FUSE_I_H */