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 */
115 /** FUSE inode state bits */
117 /** Advise readdirplus */
118 FUSE_I_ADVISE_RDPLUS,
119 /** Initialized with readdirplus */
121 /** An operation changing file size is in progress */
122 FUSE_I_SIZE_UNSTABLE,
127 /** FUSE specific file data */
129 /** Fuse connection for this file */
130 struct fuse_conn *fc;
132 /** Request reserved for flush and release */
133 struct fuse_req *reserved_req;
135 /** Kernel file handle guaranteed to be unique */
138 /** File handle used by userspace */
141 /** Node id of this file */
147 /** FOPEN_* flags returned by open */
150 /** Entry on inode's write_files list */
151 struct list_head write_entry;
153 /** RB node to be linked on fuse_conn->polled_files */
154 struct rb_node polled_node;
156 /** Wait queue head for poll */
157 wait_queue_head_t poll_wait;
159 /** Has flock been performed on this file? */
163 /** One input argument of a request */
169 /** The request input */
171 /** The request header */
172 struct fuse_in_header h;
174 /** True if the data for the last argument is in req->pages */
177 /** Number of arguments */
180 /** Array of arguments */
181 struct fuse_in_arg args[3];
184 /** One output argument of a request */
190 /** The request output */
192 /** Header returned from userspace */
193 struct fuse_out_header h;
196 * The following bitfields are not changed during the request
200 /** Last argument is variable length (can be shorter than
204 /** Last argument is a list of pages to copy data to */
207 /** Zero partially or not copied pages */
208 unsigned page_zeroing:1;
210 /** Pages may be replaced with new ones */
211 unsigned page_replace:1;
213 /** Number or arguments */
216 /** Array of arguments */
217 struct fuse_arg args[2];
220 /** FUSE page descriptor */
221 struct fuse_page_desc {
233 struct fuse_in_arg args[3];
239 struct fuse_arg args[2];
243 #define FUSE_ARGS(args) struct fuse_args args = {}
245 /** The request IO state (for asynchronous processing) */
246 struct fuse_io_priv {
259 struct completion *done;
262 #define FUSE_IO_PRIV_SYNC(f) \
264 .refcnt = { ATOMIC_INIT(1) }, \
272 * FR_ISREPLY: set if the request has reply
273 * FR_FORCE: force sending of the request even if interrupted
274 * FR_BACKGROUND: request is sent in the background
275 * FR_WAITING: request is counted as "waiting"
276 * FR_ABORTED: the request was aborted
277 * FR_INTERRUPTED: the request has been interrupted
278 * FR_LOCKED: data is being copied to/from the request
279 * FR_PENDING: request is not yet in userspace
280 * FR_SENT: request is in userspace, waiting for an answer
281 * FR_FINISHED: request is finished
282 * FR_PRIVATE: request is on private list
299 * A request to the client
301 * .waitq.lock protects the following fields:
303 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
306 /** This can be on either pending processing or io lists in
308 struct list_head list;
310 /** Entry on the interrupts list */
311 struct list_head intr_entry;
316 /** Unique ID for the interrupt request */
319 /* Request flags, updated with test/set/clear_bit() */
322 /** The request input */
325 /** The request output */
328 /** Used to wake up the task waiting for completion of request*/
329 wait_queue_head_t waitq;
331 /** Data for asynchronous requests */
334 struct fuse_release_in in;
337 struct fuse_init_in init_in;
338 struct fuse_init_out init_out;
339 struct cuse_init_in cuse_init_in;
341 struct fuse_read_in in;
345 struct fuse_write_in in;
346 struct fuse_write_out out;
347 struct fuse_req *next;
349 struct fuse_notify_retrieve_in retrieve_in;
355 /** page-descriptor vector */
356 struct fuse_page_desc *page_descs;
358 /** size of the 'pages' array */
361 /** inline page vector */
362 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
364 /** inline page-descriptor vector */
365 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
367 /** number of pages in vector */
370 /** File used in the request (or NULL) */
371 struct fuse_file *ff;
373 /** Inode used in the request or NULL */
376 /** Path used for completing d_canonical_path */
377 struct path *canonical_path;
379 /** AIO control block */
380 struct fuse_io_priv *io;
382 /** Link on fi->writepages */
383 struct list_head writepages_entry;
385 /** Request completion callback */
386 void (*end)(struct fuse_conn *, struct fuse_req *);
388 /** Request is stolen from fuse_file->reserved_req */
389 struct file *stolen_file;
393 /** Connection established */
396 /** Readers of the connection are waiting on this */
397 wait_queue_head_t waitq;
399 /** The next unique request id */
402 /** The list of pending requests */
403 struct list_head pending;
405 /** Pending interrupts */
406 struct list_head interrupts;
408 /** Queue of pending forgets */
409 struct fuse_forget_link forget_list_head;
410 struct fuse_forget_link *forget_list_tail;
412 /** Batching of FORGET requests (positive indicates FORGET batch) */
415 /** O_ASYNC requests */
416 struct fasync_struct *fasync;
420 /** Connection established */
423 /** Lock protecting accessess to members of this structure */
426 /** The list of requests being processed */
427 struct list_head processing;
429 /** The list of requests under I/O */
434 * Fuse device instance
437 /** Fuse connection for this device */
438 struct fuse_conn *fc;
440 /** Processing queue */
441 struct fuse_pqueue pq;
443 /** list entry on fc->devices */
444 struct list_head entry;
450 * This structure is created, when the filesystem is mounted, and is
451 * destroyed, when the client device is closed and the filesystem is
455 /** Lock protecting accessess to members of this structure */
461 /** Number of fuse_dev's */
466 /** The user id for this mount */
469 /** The group id for this mount */
472 /** The fuse mount flags for this mount */
475 /** Maximum read size */
478 /** Maximum write size */
482 struct fuse_iqueue iq;
484 /** The next unique kernel file handle */
487 /** rbtree of fuse_files waiting for poll events indexed by ph */
488 struct rb_root polled_files;
490 /** Maximum number of outstanding background requests */
491 unsigned max_background;
493 /** Number of background requests at which congestion starts */
494 unsigned congestion_threshold;
496 /** Number of requests currently in the background */
497 unsigned num_background;
499 /** Number of background requests currently queued for userspace */
500 unsigned active_background;
502 /** The list of background requests set aside for later queuing */
503 struct list_head bg_queue;
505 /** Flag indicating that INIT reply has been received. Allocating
506 * any fuse request will be suspended until the flag is set */
509 /** Flag indicating if connection is blocked. This will be
510 the case before the INIT reply is received, and if there
511 are too many outstading backgrounds requests */
514 /** waitq for blocked connection */
515 wait_queue_head_t blocked_waitq;
517 /** waitq for reserved requests */
518 wait_queue_head_t reserved_req_waitq;
520 /** Connection established, cleared on umount, connection
521 abort and device release */
524 /** Connection failed (version mismatch). Cannot race with
525 setting other bitfields since it is only set once in INIT
526 reply, before any other request, and never cleared */
527 unsigned conn_error:1;
529 /** Connection successful. Only set in INIT */
530 unsigned conn_init:1;
532 /** Do readpages asynchronously? Only set in INIT */
533 unsigned async_read:1;
535 /** Do not send separate SETATTR request before open(O_TRUNC) */
536 unsigned atomic_o_trunc:1;
538 /** Filesystem supports NFS exporting. Only set in INIT */
539 unsigned export_support:1;
541 /** Set if bdi is valid */
542 unsigned bdi_initialized:1;
544 /** write-back cache policy (default is write-through) */
545 unsigned writeback_cache:1;
548 * The following bitfields are only for optimization purposes
549 * and hence races in setting them will not cause malfunction
552 /** Is open/release not implemented by fs? */
555 /** Is fsync not implemented by fs? */
558 /** Is fsyncdir not implemented by fs? */
559 unsigned no_fsyncdir:1;
561 /** Is flush not implemented by fs? */
564 /** Is setxattr not implemented by fs? */
565 unsigned no_setxattr:1;
567 /** Is getxattr not implemented by fs? */
568 unsigned no_getxattr:1;
570 /** Is listxattr not implemented by fs? */
571 unsigned no_listxattr:1;
573 /** Is removexattr not implemented by fs? */
574 unsigned no_removexattr:1;
576 /** Are posix file locking primitives not implemented by fs? */
579 /** Is access not implemented by fs? */
580 unsigned no_access:1;
582 /** Is create not implemented by fs? */
583 unsigned no_create:1;
585 /** Is interrupt not implemented by fs? */
586 unsigned no_interrupt:1;
588 /** Is bmap not implemented by fs? */
591 /** Is poll not implemented by fs? */
594 /** Do multi-page cached writes */
595 unsigned big_writes:1;
597 /** Don't apply umask to creation modes */
598 unsigned dont_mask:1;
600 /** Are BSD file locking primitives not implemented by fs? */
603 /** Is fallocate not implemented by fs? */
604 unsigned no_fallocate:1;
606 /** Is rename with flags implemented by fs? */
607 unsigned no_rename2:1;
609 /** Use enhanced/automatic page cache invalidation. */
610 unsigned auto_inval_data:1;
612 /** Does the filesystem support readdirplus? */
613 unsigned do_readdirplus:1;
615 /** Does the filesystem want adaptive readdirplus? */
616 unsigned readdirplus_auto:1;
618 /** Does the filesystem support asynchronous direct-IO submission? */
619 unsigned async_dio:1;
621 /** The number of requests waiting for completion */
622 atomic_t num_waiting;
624 /** Negotiated minor version */
627 /** Backing dev info */
628 struct backing_dev_info bdi;
630 /** Entry on the fuse_conn_list */
631 struct list_head entry;
633 /** Device ID from super block */
636 /** Dentries in the control filesystem */
637 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
639 /** number of dentries used in the above array */
642 /** Key for lock owner ID scrambling */
645 /** Reserved request for the DESTROY message */
646 struct fuse_req *destroy_req;
648 /** Version counter for attribute changes */
651 /** Called on final put */
652 void (*release)(struct fuse_conn *);
654 /** Super block for this connection. */
655 struct super_block *sb;
657 /** Read/write semaphore to hold when accessing sb. */
658 struct rw_semaphore killsb;
660 /** List of device instances belonging to this connection */
661 struct list_head devices;
664 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
666 return sb->s_fs_info;
669 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
671 return get_fuse_conn_super(inode->i_sb);
674 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
676 return container_of(inode, struct fuse_inode, inode);
679 static inline u64 get_node_id(struct inode *inode)
681 return get_fuse_inode(inode)->nodeid;
684 /** Device operations */
685 extern const struct file_operations fuse_dev_operations;
687 extern const struct dentry_operations fuse_dentry_operations;
690 * Inode to nodeid comparison.
692 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
695 * Get a filled in inode
697 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
698 int generation, struct fuse_attr *attr,
699 u64 attr_valid, u64 attr_version);
701 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
702 struct fuse_entry_out *outarg, struct inode **inode);
705 * Send FORGET command
707 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
708 u64 nodeid, u64 nlookup);
710 struct fuse_forget_link *fuse_alloc_forget(void);
712 /* Used by READDIRPLUS */
713 void fuse_force_forget(struct file *file, u64 nodeid);
716 * Initialize READ or READDIR request
718 void fuse_read_fill(struct fuse_req *req, struct file *file,
719 loff_t pos, size_t count, int opcode);
722 * Send OPEN or OPENDIR request
724 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
726 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
727 struct fuse_file *fuse_file_get(struct fuse_file *ff);
728 void fuse_file_free(struct fuse_file *ff);
729 void fuse_finish_open(struct inode *inode, struct file *file);
731 void fuse_sync_release(struct fuse_file *ff, int flags);
734 * Send RELEASE or RELEASEDIR request
736 void fuse_release_common(struct file *file, int opcode);
739 * Send FSYNC or FSYNCDIR request
741 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
742 int datasync, int isdir);
747 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
748 struct fuse_notify_poll_wakeup_out *outarg);
751 * Initialize file operations on a regular file
753 void fuse_init_file_inode(struct inode *inode);
756 * Initialize inode operations on regular files and special files
758 void fuse_init_common(struct inode *inode);
761 * Initialize inode and file operations on a directory
763 void fuse_init_dir(struct inode *inode);
766 * Initialize inode operations on a symlink
768 void fuse_init_symlink(struct inode *inode);
771 * Change attributes of an inode
773 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
774 u64 attr_valid, u64 attr_version);
776 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
780 * Initialize the client device
782 int fuse_dev_init(void);
785 * Cleanup the client device
787 void fuse_dev_cleanup(void);
789 int fuse_ctl_init(void);
790 void __exit fuse_ctl_cleanup(void);
795 struct fuse_req *fuse_request_alloc(unsigned npages);
797 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
802 void fuse_request_free(struct fuse_req *req);
805 * Get a request, may fail with -ENOMEM,
806 * caller should specify # elements in req->pages[] explicitly
808 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
809 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
813 * Increment reference count on request
815 void __fuse_get_request(struct fuse_req *req);
818 * Gets a requests for a file operation, always succeeds
820 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
824 * Decrement reference count of a request. If count goes to zero free
827 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
830 * Send a request (synchronous)
832 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
835 * Simple request sending that does request allocation and freeing
837 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
840 * Send a request in the background
842 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
844 void fuse_request_send_background_locked(struct fuse_conn *fc,
845 struct fuse_req *req);
847 /* Abort all requests */
848 void fuse_abort_conn(struct fuse_conn *fc);
849 void fuse_wait_aborted(struct fuse_conn *fc);
852 * Invalidate inode attributes
854 void fuse_invalidate_attr(struct inode *inode);
856 void fuse_invalidate_entry_cache(struct dentry *entry);
858 void fuse_invalidate_atime(struct inode *inode);
861 * Acquire reference to fuse_conn
863 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
866 * Initialize fuse_conn
868 void fuse_conn_init(struct fuse_conn *fc);
871 * Release reference to fuse_conn
873 void fuse_conn_put(struct fuse_conn *fc);
875 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc);
876 void fuse_dev_free(struct fuse_dev *fud);
879 * Add connection to control filesystem
881 int fuse_ctl_add_conn(struct fuse_conn *fc);
884 * Remove connection from control filesystem
886 void fuse_ctl_remove_conn(struct fuse_conn *fc);
889 * Is file type valid?
891 int fuse_valid_type(int m);
894 * Is current process allowed to perform filesystem operation?
896 int fuse_allow_current_process(struct fuse_conn *fc);
898 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
900 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
901 struct file *file, bool *refreshed);
903 void fuse_flush_writepages(struct inode *inode);
905 void fuse_set_nowrite(struct inode *inode);
906 void fuse_release_nowrite(struct inode *inode);
908 u64 fuse_get_attr_version(struct fuse_conn *fc);
911 * File-system tells the kernel to invalidate cache for the given node id.
913 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
914 loff_t offset, loff_t len);
917 * File-system tells the kernel to invalidate parent attributes and
918 * the dentry matching parent/name.
920 * If the child_nodeid is non-zero and:
921 * - matches the inode number for the dentry matching parent/name,
922 * - is not a mount point
923 * - is a file or oan empty directory
924 * then the dentry is unhashed (d_delete()).
926 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
927 u64 child_nodeid, struct qstr *name);
929 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
933 * fuse_direct_io() flags
936 /** If set, it is WRITE; otherwise - READ */
937 #define FUSE_DIO_WRITE (1 << 0)
939 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
940 #define FUSE_DIO_CUSE (1 << 1)
942 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
943 loff_t *ppos, int flags);
944 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
946 long fuse_ioctl_common(struct file *file, unsigned int cmd,
947 unsigned long arg, unsigned int flags);
948 unsigned fuse_file_poll(struct file *file, poll_table *wait);
949 int fuse_dev_release(struct inode *inode, struct file *file);
951 bool fuse_write_update_size(struct inode *inode, loff_t pos);
953 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
954 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
956 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
959 void fuse_set_initialized(struct fuse_conn *fc);
961 #endif /* _FS_FUSE_I_H */