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>
26 /** Max number of pages that can be used in a single read request */
27 #define FUSE_MAX_PAGES_PER_REQ 32
29 /** Bias for fi->writectr, meaning new writepages must not be sent */
30 #define FUSE_NOWRITE INT_MIN
32 /** It could be as large as PATH_MAX, but would that have any uses? */
33 #define FUSE_NAME_MAX 1024
35 /** Number of dentries for each connection in the control filesystem */
36 #define FUSE_CTL_NUM_DENTRIES 5
38 /** If the FUSE_DEFAULT_PERMISSIONS flag is given, the filesystem
39 module will check permissions based on the file mode. Otherwise no
40 permission checking is done in the kernel */
41 #define FUSE_DEFAULT_PERMISSIONS (1 << 0)
43 /** If the FUSE_ALLOW_OTHER flag is given, then not only the user
44 doing the mount will be allowed to access the filesystem */
45 #define FUSE_ALLOW_OTHER (1 << 1)
47 /** Number of page pointers embedded in fuse_req */
48 #define FUSE_REQ_INLINE_PAGES 1
50 /** List of active connections */
51 extern struct list_head fuse_conn_list;
53 /** Global mutex protecting fuse_conn_list and the control filesystem */
54 extern struct mutex fuse_mutex;
56 /** Module parameters */
57 extern unsigned max_user_bgreq;
58 extern unsigned max_user_congthresh;
60 /* One forget request */
61 struct fuse_forget_link {
62 struct fuse_forget_one forget_one;
63 struct fuse_forget_link *next;
71 /** Unique ID, which identifies the inode between userspace
75 /** Number of lookups on this inode */
78 /** The request used for sending the FORGET message */
79 struct fuse_forget_link *forget;
81 /** Time in jiffies until the file attributes are valid */
84 /** The sticky bit in inode->i_mode may have been removed, so
85 preserve the original mode */
88 /** 64 bit inode number */
91 /** Version of last attribute change */
94 /** Files usable in writepage. Protected by fc->lock */
95 struct list_head write_files;
97 /** Writepages pending on truncate or fsync */
98 struct list_head queued_writes;
100 /** Number of sent writes, a negative bias (FUSE_NOWRITE)
101 * means more writes are blocked */
104 /** Waitq for writepage completion */
105 wait_queue_head_t page_waitq;
107 /** List of writepage requestst (pending or sent) */
108 struct list_head writepages;
110 /** Miscellaneous bits describing inode state */
114 /** FUSE inode state bits */
116 /** Advise readdirplus */
117 FUSE_I_ADVISE_RDPLUS,
118 /** Initialized with readdirplus */
120 /** An operation changing file size is in progress */
121 FUSE_I_SIZE_UNSTABLE,
122 /** i_mtime has been updated locally; a flush to userspace needed */
128 /** FUSE specific file data */
130 /** Fuse connection for this file */
131 struct fuse_conn *fc;
133 /** Request reserved for flush and release */
134 struct fuse_req *reserved_req;
136 /** Kernel file handle guaranteed to be unique */
139 /** File handle used by userspace */
142 /** Node id of this file */
148 /** FOPEN_* flags returned by open */
151 /** Entry on inode's write_files list */
152 struct list_head write_entry;
154 /** RB node to be linked on fuse_conn->polled_files */
155 struct rb_node polled_node;
157 /** Wait queue head for poll */
158 wait_queue_head_t poll_wait;
160 /** Has flock been performed on this file? */
164 /** One input argument of a request */
170 /** The request input */
172 /** The request header */
173 struct fuse_in_header h;
175 /** True if the data for the last argument is in req->pages */
178 /** Number of arguments */
181 /** Array of arguments */
182 struct fuse_in_arg args[3];
185 /** One output argument of a request */
191 /** The request output */
193 /** Header returned from userspace */
194 struct fuse_out_header h;
197 * The following bitfields are not changed during the request
201 /** Last argument is variable length (can be shorter than
205 /** Last argument is a list of pages to copy data to */
208 /** Zero partially or not copied pages */
209 unsigned page_zeroing:1;
211 /** Pages may be replaced with new ones */
212 unsigned page_replace:1;
214 /** Number or arguments */
217 /** Array of arguments */
218 struct fuse_arg args[3];
221 /** FUSE page descriptor */
222 struct fuse_page_desc {
227 /** The request state */
228 enum fuse_req_state {
237 /** The request IO state (for asynchronous processing) */
238 struct fuse_io_priv {
252 * A request to the client
255 /** This can be on either pending processing or io lists in
257 struct list_head list;
259 /** Entry on the interrupts list */
260 struct list_head intr_entry;
265 /** Unique ID for the interrupt request */
269 * The following bitfields are either set once before the
270 * request is queued or setting/clearing them is protected by
274 /** True if the request has reply */
277 /** Force sending of the request even if interrupted */
280 /** The request was aborted */
283 /** Request is sent in the background */
284 unsigned background:1;
286 /** The request has been interrupted */
287 unsigned interrupted:1;
289 /** Data is being copied to/from the request */
292 /** Request is counted as "waiting" */
295 /** State of the request */
296 enum fuse_req_state state;
298 /** The request input */
301 /** The request output */
304 /** Used to wake up the task waiting for completion of request*/
305 wait_queue_head_t waitq;
307 /** Data for asynchronous requests */
311 struct fuse_release_in in;
312 struct work_struct work;
316 struct fuse_init_in init_in;
317 struct fuse_init_out init_out;
318 struct cuse_init_in cuse_init_in;
320 struct fuse_read_in in;
324 struct fuse_write_in in;
325 struct fuse_write_out out;
326 struct fuse_req *next;
328 struct fuse_notify_retrieve_in retrieve_in;
329 struct fuse_lk_in lk_in;
335 /** page-descriptor vector */
336 struct fuse_page_desc *page_descs;
338 /** size of the 'pages' array */
341 /** inline page vector */
342 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
344 /** inline page-descriptor vector */
345 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
347 /** number of pages in vector */
350 /** File used in the request (or NULL) */
351 struct fuse_file *ff;
353 /** Inode used in the request or NULL */
356 /** AIO control block */
357 struct fuse_io_priv *io;
359 /** Link on fi->writepages */
360 struct list_head writepages_entry;
362 /** Request completion callback */
363 void (*end)(struct fuse_conn *, struct fuse_req *);
365 /** Request is stolen from fuse_file->reserved_req */
366 struct file *stolen_file;
372 * This structure is created, when the filesystem is mounted, and is
373 * destroyed, when the client device is closed and the filesystem is
377 /** Lock protecting accessess to members of this structure */
385 /** The user id for this mount */
388 /** The group id for this mount */
391 /** The fuse mount flags for this mount */
394 /** Maximum read size */
397 /** Maximum write size */
400 /** Readers of the connection are waiting on this */
401 wait_queue_head_t waitq;
403 /** The list of pending requests */
404 struct list_head pending;
406 /** The list of requests being processed */
407 struct list_head processing;
409 /** The list of requests under I/O */
412 /** The next unique kernel file handle */
415 /** rbtree of fuse_files waiting for poll events indexed by ph */
416 struct rb_root polled_files;
418 /** Maximum number of outstanding background requests */
419 unsigned max_background;
421 /** Number of background requests at which congestion starts */
422 unsigned congestion_threshold;
424 /** Number of requests currently in the background */
425 unsigned num_background;
427 /** Number of background requests currently queued for userspace */
428 unsigned active_background;
430 /** The list of background requests set aside for later queuing */
431 struct list_head bg_queue;
433 /** Pending interrupts */
434 struct list_head interrupts;
436 /** Queue of pending forgets */
437 struct fuse_forget_link forget_list_head;
438 struct fuse_forget_link *forget_list_tail;
440 /** Batching of FORGET requests (positive indicates FORGET batch) */
443 /** Flag indicating that INIT reply has been received. Allocating
444 * any fuse request will be suspended until the flag is set */
447 /** Flag indicating if connection is blocked. This will be
448 the case before the INIT reply is received, and if there
449 are too many outstading backgrounds requests */
452 /** waitq for blocked connection */
453 wait_queue_head_t blocked_waitq;
455 /** waitq for reserved requests */
456 wait_queue_head_t reserved_req_waitq;
458 /** The next unique request id */
461 /** Connection established, cleared on umount, connection
462 abort and device release */
465 /** Connection failed (version mismatch). Cannot race with
466 setting other bitfields since it is only set once in INIT
467 reply, before any other request, and never cleared */
468 unsigned conn_error:1;
470 /** Connection successful. Only set in INIT */
471 unsigned conn_init:1;
473 /** Do readpages asynchronously? Only set in INIT */
474 unsigned async_read:1;
476 /** Do not send separate SETATTR request before open(O_TRUNC) */
477 unsigned atomic_o_trunc:1;
479 /** Filesystem supports NFS exporting. Only set in INIT */
480 unsigned export_support:1;
482 /** Set if bdi is valid */
483 unsigned bdi_initialized:1;
485 /** write-back cache policy (default is write-through) */
486 unsigned writeback_cache:1;
489 * The following bitfields are only for optimization purposes
490 * and hence races in setting them will not cause malfunction
493 /** Is open/release not implemented by fs? */
496 /** Is fsync not implemented by fs? */
499 /** Is fsyncdir not implemented by fs? */
500 unsigned no_fsyncdir:1;
502 /** Is flush not implemented by fs? */
505 /** Is setxattr not implemented by fs? */
506 unsigned no_setxattr:1;
508 /** Is getxattr not implemented by fs? */
509 unsigned no_getxattr:1;
511 /** Is listxattr not implemented by fs? */
512 unsigned no_listxattr:1;
514 /** Is removexattr not implemented by fs? */
515 unsigned no_removexattr:1;
517 /** Are posix file locking primitives not implemented by fs? */
520 /** Is access not implemented by fs? */
521 unsigned no_access:1;
523 /** Is create not implemented by fs? */
524 unsigned no_create:1;
526 /** Is interrupt not implemented by fs? */
527 unsigned no_interrupt:1;
529 /** Is bmap not implemented by fs? */
532 /** Is poll not implemented by fs? */
535 /** Do multi-page cached writes */
536 unsigned big_writes:1;
538 /** Don't apply umask to creation modes */
539 unsigned dont_mask:1;
541 /** Are BSD file locking primitives not implemented by fs? */
544 /** Is fallocate not implemented by fs? */
545 unsigned no_fallocate:1;
547 /** Use enhanced/automatic page cache invalidation. */
548 unsigned auto_inval_data:1;
550 /** Does the filesystem support readdirplus? */
551 unsigned do_readdirplus:1;
553 /** Does the filesystem want adaptive readdirplus? */
554 unsigned readdirplus_auto:1;
556 /** Does the filesystem support asynchronous direct-IO submission? */
557 unsigned async_dio:1;
559 /** The number of requests waiting for completion */
560 atomic_t num_waiting;
562 /** Negotiated minor version */
565 /** Backing dev info */
566 struct backing_dev_info bdi;
568 /** Entry on the fuse_conn_list */
569 struct list_head entry;
571 /** Device ID from super block */
574 /** Dentries in the control filesystem */
575 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
577 /** number of dentries used in the above array */
580 /** O_ASYNC requests */
581 struct fasync_struct *fasync;
583 /** Key for lock owner ID scrambling */
586 /** Reserved request for the DESTROY message */
587 struct fuse_req *destroy_req;
589 /** Version counter for attribute changes */
592 /** Called on final put */
593 void (*release)(struct fuse_conn *);
595 /** Super block for this connection. */
596 struct super_block *sb;
598 /** Read/write semaphore to hold when accessing sb. */
599 struct rw_semaphore killsb;
602 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
604 return sb->s_fs_info;
607 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
609 return get_fuse_conn_super(inode->i_sb);
612 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
614 return container_of(inode, struct fuse_inode, inode);
617 static inline u64 get_node_id(struct inode *inode)
619 return get_fuse_inode(inode)->nodeid;
622 /** Device operations */
623 extern const struct file_operations fuse_dev_operations;
625 extern const struct dentry_operations fuse_dentry_operations;
628 * Inode to nodeid comparison.
630 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
633 * Get a filled in inode
635 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
636 int generation, struct fuse_attr *attr,
637 u64 attr_valid, u64 attr_version);
639 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
640 struct fuse_entry_out *outarg, struct inode **inode);
643 * Send FORGET command
645 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
646 u64 nodeid, u64 nlookup);
648 struct fuse_forget_link *fuse_alloc_forget(void);
650 /* Used by READDIRPLUS */
651 void fuse_force_forget(struct file *file, u64 nodeid);
654 * Initialize READ or READDIR request
656 void fuse_read_fill(struct fuse_req *req, struct file *file,
657 loff_t pos, size_t count, int opcode);
660 * Send OPEN or OPENDIR request
662 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
664 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
665 struct fuse_file *fuse_file_get(struct fuse_file *ff);
666 void fuse_file_free(struct fuse_file *ff);
667 void fuse_finish_open(struct inode *inode, struct file *file);
669 void fuse_sync_release(struct fuse_file *ff, int flags);
672 * Send RELEASE or RELEASEDIR request
674 void fuse_release_common(struct file *file, int opcode);
677 * Send FSYNC or FSYNCDIR request
679 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
680 int datasync, int isdir);
685 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
686 struct fuse_notify_poll_wakeup_out *outarg);
689 * Initialize file operations on a regular file
691 void fuse_init_file_inode(struct inode *inode);
694 * Initialize inode operations on regular files and special files
696 void fuse_init_common(struct inode *inode);
699 * Initialize inode and file operations on a directory
701 void fuse_init_dir(struct inode *inode);
704 * Initialize inode operations on a symlink
706 void fuse_init_symlink(struct inode *inode);
709 * Change attributes of an inode
711 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
712 u64 attr_valid, u64 attr_version);
714 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
718 * Initialize the client device
720 int fuse_dev_init(void);
723 * Cleanup the client device
725 void fuse_dev_cleanup(void);
727 int fuse_ctl_init(void);
728 void fuse_ctl_cleanup(void);
733 struct fuse_req *fuse_request_alloc(unsigned npages);
735 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
740 void fuse_request_free(struct fuse_req *req);
743 * Get a request, may fail with -ENOMEM,
744 * caller should specify # elements in req->pages[] explicitly
746 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
747 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
751 * Increment reference count on request
753 void __fuse_get_request(struct fuse_req *req);
756 * Get a request, may fail with -ENOMEM,
757 * useful for callers who doesn't use req->pages[]
759 static inline struct fuse_req *fuse_get_req_nopages(struct fuse_conn *fc)
761 return fuse_get_req(fc, 0);
765 * Gets a requests for a file operation, always succeeds
767 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
771 * Decrement reference count of a request. If count goes to zero free
774 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
777 * Send a request (synchronous)
779 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
782 * Send a request in the background
784 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
786 void fuse_request_send_background_locked(struct fuse_conn *fc,
787 struct fuse_req *req);
789 /* Abort all requests */
790 void fuse_abort_conn(struct fuse_conn *fc);
793 * Invalidate inode attributes
795 void fuse_invalidate_attr(struct inode *inode);
797 void fuse_invalidate_entry_cache(struct dentry *entry);
799 void fuse_invalidate_atime(struct inode *inode);
802 * Acquire reference to fuse_conn
804 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
806 void fuse_conn_kill(struct fuse_conn *fc);
809 * Initialize fuse_conn
811 void fuse_conn_init(struct fuse_conn *fc);
814 * Release reference to fuse_conn
816 void fuse_conn_put(struct fuse_conn *fc);
819 * Add connection to control filesystem
821 int fuse_ctl_add_conn(struct fuse_conn *fc);
824 * Remove connection from control filesystem
826 void fuse_ctl_remove_conn(struct fuse_conn *fc);
829 * Is file type valid?
831 int fuse_valid_type(int m);
834 * Is current process allowed to perform filesystem operation?
836 int fuse_allow_current_process(struct fuse_conn *fc);
838 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
840 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
841 struct file *file, bool *refreshed);
843 void fuse_flush_writepages(struct inode *inode);
845 void fuse_set_nowrite(struct inode *inode);
846 void fuse_release_nowrite(struct inode *inode);
848 u64 fuse_get_attr_version(struct fuse_conn *fc);
851 * File-system tells the kernel to invalidate cache for the given node id.
853 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
854 loff_t offset, loff_t len);
857 * File-system tells the kernel to invalidate parent attributes and
858 * the dentry matching parent/name.
860 * If the child_nodeid is non-zero and:
861 * - matches the inode number for the dentry matching parent/name,
862 * - is not a mount point
863 * - is a file or oan empty directory
864 * then the dentry is unhashed (d_delete()).
866 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
867 u64 child_nodeid, struct qstr *name);
869 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
873 * fuse_direct_io() flags
876 /** If set, it is WRITE; otherwise - READ */
877 #define FUSE_DIO_WRITE (1 << 0)
879 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
880 #define FUSE_DIO_CUSE (1 << 1)
882 ssize_t fuse_direct_io(struct fuse_io_priv *io, const struct iovec *iov,
883 unsigned long nr_segs, size_t count, loff_t *ppos,
885 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
887 long fuse_ioctl_common(struct file *file, unsigned int cmd,
888 unsigned long arg, unsigned int flags);
889 unsigned fuse_file_poll(struct file *file, poll_table *wait);
890 int fuse_dev_release(struct inode *inode, struct file *file);
892 bool fuse_write_update_size(struct inode *inode, loff_t pos);
894 int fuse_flush_mtime(struct file *file, bool nofail);
896 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
899 #endif /* _FS_FUSE_I_H */