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>
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 {
256 struct completion *done;
260 #define FUSE_IO_PRIV_SYNC(f) \
262 .refcnt = KREF_INIT(1), \
270 * FR_ISREPLY: set if the request has reply
271 * FR_FORCE: force sending of the request even if interrupted
272 * FR_BACKGROUND: request is sent in the background
273 * FR_WAITING: request is counted as "waiting"
274 * FR_ABORTED: the request was aborted
275 * FR_INTERRUPTED: the request has been interrupted
276 * FR_LOCKED: data is being copied to/from the request
277 * FR_PENDING: request is not yet in userspace
278 * FR_SENT: request is in userspace, waiting for an answer
279 * FR_FINISHED: request is finished
280 * FR_PRIVATE: request is on private list
297 * A request to the client
299 * .waitq.lock protects the following fields:
301 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
304 /** This can be on either pending processing or io lists in
306 struct list_head list;
308 /** Entry on the interrupts list */
309 struct list_head intr_entry;
314 /** Unique ID for the interrupt request */
317 /* Request flags, updated with test/set/clear_bit() */
320 /** The request input */
323 /** The request output */
326 /** Used to wake up the task waiting for completion of request*/
327 wait_queue_head_t waitq;
329 /** Data for asynchronous requests */
332 struct fuse_release_in in;
335 struct fuse_init_in init_in;
336 struct fuse_init_out init_out;
337 struct cuse_init_in cuse_init_in;
339 struct fuse_read_in in;
343 struct fuse_write_in in;
344 struct fuse_write_out out;
345 struct fuse_req *next;
347 struct fuse_notify_retrieve_in retrieve_in;
353 /** page-descriptor vector */
354 struct fuse_page_desc *page_descs;
356 /** size of the 'pages' array */
359 /** inline page vector */
360 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
362 /** inline page-descriptor vector */
363 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
365 /** number of pages in vector */
368 /** File used in the request (or NULL) */
369 struct fuse_file *ff;
371 /** Inode used in the request or NULL */
374 /** AIO control block */
375 struct fuse_io_priv *io;
377 /** Link on fi->writepages */
378 struct list_head writepages_entry;
380 /** Request completion callback */
381 void (*end)(struct fuse_conn *, struct fuse_req *);
383 /** Request is stolen from fuse_file->reserved_req */
384 struct file *stolen_file;
388 /** Connection established */
391 /** Readers of the connection are waiting on this */
392 wait_queue_head_t waitq;
394 /** The next unique request id */
397 /** The list of pending requests */
398 struct list_head pending;
400 /** Pending interrupts */
401 struct list_head interrupts;
403 /** Queue of pending forgets */
404 struct fuse_forget_link forget_list_head;
405 struct fuse_forget_link *forget_list_tail;
407 /** Batching of FORGET requests (positive indicates FORGET batch) */
410 /** O_ASYNC requests */
411 struct fasync_struct *fasync;
415 /** Connection established */
418 /** Lock protecting accessess to members of this structure */
421 /** The list of requests being processed */
422 struct list_head processing;
424 /** The list of requests under I/O */
429 * Fuse device instance
432 /** Fuse connection for this device */
433 struct fuse_conn *fc;
435 /** Processing queue */
436 struct fuse_pqueue pq;
438 /** list entry on fc->devices */
439 struct list_head entry;
445 * This structure is created, when the filesystem is mounted, and is
446 * destroyed, when the client device is closed and the filesystem is
450 /** Lock protecting accessess to members of this structure */
456 /** Number of fuse_dev's */
461 /** The user id for this mount */
464 /** The group id for this mount */
467 /** The pid namespace for this mount */
468 struct pid_namespace *pid_ns;
470 /** Maximum read size */
473 /** Maximum write size */
477 struct fuse_iqueue iq;
479 /** The next unique kernel file handle */
482 /** rbtree of fuse_files waiting for poll events indexed by ph */
483 struct rb_root polled_files;
485 /** Maximum number of outstanding background requests */
486 unsigned max_background;
488 /** Number of background requests at which congestion starts */
489 unsigned congestion_threshold;
491 /** Number of requests currently in the background */
492 unsigned num_background;
494 /** Number of background requests currently queued for userspace */
495 unsigned active_background;
497 /** The list of background requests set aside for later queuing */
498 struct list_head bg_queue;
500 /** Flag indicating that INIT reply has been received. Allocating
501 * any fuse request will be suspended until the flag is set */
504 /** Flag indicating if connection is blocked. This will be
505 the case before the INIT reply is received, and if there
506 are too many outstading backgrounds requests */
509 /** waitq for blocked connection */
510 wait_queue_head_t blocked_waitq;
512 /** waitq for reserved requests */
513 wait_queue_head_t reserved_req_waitq;
515 /** Connection established, cleared on umount, connection
516 abort and device release */
519 /** Connection failed (version mismatch). Cannot race with
520 setting other bitfields since it is only set once in INIT
521 reply, before any other request, and never cleared */
522 unsigned conn_error:1;
524 /** Connection successful. Only set in INIT */
525 unsigned conn_init:1;
527 /** Do readpages asynchronously? Only set in INIT */
528 unsigned async_read:1;
530 /** Do not send separate SETATTR request before open(O_TRUNC) */
531 unsigned atomic_o_trunc:1;
533 /** Filesystem supports NFS exporting. Only set in INIT */
534 unsigned export_support:1;
536 /** write-back cache policy (default is write-through) */
537 unsigned writeback_cache:1;
539 /** allow parallel lookups and readdir (default is serialized) */
540 unsigned parallel_dirops:1;
542 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
543 unsigned handle_killpriv:1;
546 * The following bitfields are only for optimization purposes
547 * and hence races in setting them will not cause malfunction
550 /** Is open/release not implemented by fs? */
553 /** Is fsync not implemented by fs? */
556 /** Is fsyncdir not implemented by fs? */
557 unsigned no_fsyncdir:1;
559 /** Is flush not implemented by fs? */
562 /** Is setxattr not implemented by fs? */
563 unsigned no_setxattr:1;
565 /** Is getxattr not implemented by fs? */
566 unsigned no_getxattr:1;
568 /** Is listxattr not implemented by fs? */
569 unsigned no_listxattr:1;
571 /** Is removexattr not implemented by fs? */
572 unsigned no_removexattr:1;
574 /** Are posix file locking primitives not implemented by fs? */
577 /** Is access not implemented by fs? */
578 unsigned no_access:1;
580 /** Is create not implemented by fs? */
581 unsigned no_create:1;
583 /** Is interrupt not implemented by fs? */
584 unsigned no_interrupt:1;
586 /** Is bmap not implemented by fs? */
589 /** Is poll not implemented by fs? */
592 /** Do multi-page cached writes */
593 unsigned big_writes:1;
595 /** Don't apply umask to creation modes */
596 unsigned dont_mask:1;
598 /** Are BSD file locking primitives not implemented by fs? */
601 /** Is fallocate not implemented by fs? */
602 unsigned no_fallocate:1;
604 /** Is rename with flags implemented by fs? */
605 unsigned no_rename2:1;
607 /** Use enhanced/automatic page cache invalidation. */
608 unsigned auto_inval_data:1;
610 /** Does the filesystem support readdirplus? */
611 unsigned do_readdirplus:1;
613 /** Does the filesystem want adaptive readdirplus? */
614 unsigned readdirplus_auto:1;
616 /** Does the filesystem support asynchronous direct-IO submission? */
617 unsigned async_dio:1;
619 /** Is lseek not implemented by fs? */
622 /** Does the filesystem support posix acls? */
623 unsigned posix_acl:1;
625 /** Check permissions based on the file mode or not? */
626 unsigned default_permissions:1;
628 /** Allow other than the mounter user to access the filesystem ? */
629 unsigned allow_other:1;
631 /** The number of requests waiting for completion */
632 atomic_t num_waiting;
634 /** Negotiated minor version */
637 /** Entry on the fuse_conn_list */
638 struct list_head entry;
640 /** Device ID from super block */
643 /** Dentries in the control filesystem */
644 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
646 /** number of dentries used in the above array */
649 /** Key for lock owner ID scrambling */
652 /** Reserved request for the DESTROY message */
653 struct fuse_req *destroy_req;
655 /** Version counter for attribute changes */
658 /** Called on final put */
659 void (*release)(struct fuse_conn *);
661 /** Super block for this connection. */
662 struct super_block *sb;
664 /** Read/write semaphore to hold when accessing sb. */
665 struct rw_semaphore killsb;
667 /** List of device instances belonging to this connection */
668 struct list_head devices;
671 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
673 return sb->s_fs_info;
676 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
678 return get_fuse_conn_super(inode->i_sb);
681 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
683 return container_of(inode, struct fuse_inode, inode);
686 static inline u64 get_node_id(struct inode *inode)
688 return get_fuse_inode(inode)->nodeid;
691 /** Device operations */
692 extern const struct file_operations fuse_dev_operations;
694 extern const struct dentry_operations fuse_dentry_operations;
695 extern const struct dentry_operations fuse_root_dentry_operations;
698 * Inode to nodeid comparison.
700 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
703 * Get a filled in inode
705 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
706 int generation, struct fuse_attr *attr,
707 u64 attr_valid, u64 attr_version);
709 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
710 struct fuse_entry_out *outarg, struct inode **inode);
713 * Send FORGET command
715 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
716 u64 nodeid, u64 nlookup);
718 struct fuse_forget_link *fuse_alloc_forget(void);
720 /* Used by READDIRPLUS */
721 void fuse_force_forget(struct file *file, u64 nodeid);
724 * Initialize READ or READDIR request
726 void fuse_read_fill(struct fuse_req *req, struct file *file,
727 loff_t pos, size_t count, int opcode);
730 * Send OPEN or OPENDIR request
732 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
734 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
735 void fuse_file_free(struct fuse_file *ff);
736 void fuse_finish_open(struct inode *inode, struct file *file);
738 void fuse_sync_release(struct fuse_file *ff, int flags);
741 * Send RELEASE or RELEASEDIR request
743 void fuse_release_common(struct file *file, int opcode);
746 * Send FSYNC or FSYNCDIR request
748 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
749 int datasync, int isdir);
754 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
755 struct fuse_notify_poll_wakeup_out *outarg);
758 * Initialize file operations on a regular file
760 void fuse_init_file_inode(struct inode *inode);
763 * Initialize inode operations on regular files and special files
765 void fuse_init_common(struct inode *inode);
768 * Initialize inode and file operations on a directory
770 void fuse_init_dir(struct inode *inode);
773 * Initialize inode operations on a symlink
775 void fuse_init_symlink(struct inode *inode);
778 * Change attributes of an inode
780 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
781 u64 attr_valid, u64 attr_version);
783 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
787 * Initialize the client device
789 int fuse_dev_init(void);
792 * Cleanup the client device
794 void fuse_dev_cleanup(void);
796 int fuse_ctl_init(void);
797 void __exit fuse_ctl_cleanup(void);
802 struct fuse_req *fuse_request_alloc(unsigned npages);
804 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
809 void fuse_request_free(struct fuse_req *req);
812 * Get a request, may fail with -ENOMEM,
813 * caller should specify # elements in req->pages[] explicitly
815 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
816 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
820 * Increment reference count on request
822 void __fuse_get_request(struct fuse_req *req);
825 * Gets a requests for a file operation, always succeeds
827 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
831 * Decrement reference count of a request. If count goes to zero free
834 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
837 * Send a request (synchronous)
839 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
842 * Simple request sending that does request allocation and freeing
844 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
847 * Send a request in the background
849 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
851 void fuse_request_send_background_locked(struct fuse_conn *fc,
852 struct fuse_req *req);
854 /* Abort all requests */
855 void fuse_abort_conn(struct fuse_conn *fc);
858 * Invalidate inode attributes
860 void fuse_invalidate_attr(struct inode *inode);
862 void fuse_invalidate_entry_cache(struct dentry *entry);
864 void fuse_invalidate_atime(struct inode *inode);
867 * Acquire reference to fuse_conn
869 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
872 * Initialize fuse_conn
874 void fuse_conn_init(struct fuse_conn *fc);
877 * Release reference to fuse_conn
879 void fuse_conn_put(struct fuse_conn *fc);
881 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc);
882 void fuse_dev_free(struct fuse_dev *fud);
885 * Add connection to control filesystem
887 int fuse_ctl_add_conn(struct fuse_conn *fc);
890 * Remove connection from control filesystem
892 void fuse_ctl_remove_conn(struct fuse_conn *fc);
895 * Is file type valid?
897 int fuse_valid_type(int m);
900 * Is current process allowed to perform filesystem operation?
902 int fuse_allow_current_process(struct fuse_conn *fc);
904 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
906 void fuse_update_ctime(struct inode *inode);
908 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
909 struct file *file, bool *refreshed);
911 void fuse_flush_writepages(struct inode *inode);
913 void fuse_set_nowrite(struct inode *inode);
914 void fuse_release_nowrite(struct inode *inode);
916 u64 fuse_get_attr_version(struct fuse_conn *fc);
919 * File-system tells the kernel to invalidate cache for the given node id.
921 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
922 loff_t offset, loff_t len);
925 * File-system tells the kernel to invalidate parent attributes and
926 * the dentry matching parent/name.
928 * If the child_nodeid is non-zero and:
929 * - matches the inode number for the dentry matching parent/name,
930 * - is not a mount point
931 * - is a file or oan empty directory
932 * then the dentry is unhashed (d_delete()).
934 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
935 u64 child_nodeid, struct qstr *name);
937 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
941 * fuse_direct_io() flags
944 /** If set, it is WRITE; otherwise - READ */
945 #define FUSE_DIO_WRITE (1 << 0)
947 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
948 #define FUSE_DIO_CUSE (1 << 1)
950 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
951 loff_t *ppos, int flags);
952 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
954 long fuse_ioctl_common(struct file *file, unsigned int cmd,
955 unsigned long arg, unsigned int flags);
956 unsigned fuse_file_poll(struct file *file, poll_table *wait);
957 int fuse_dev_release(struct inode *inode, struct file *file);
959 bool fuse_write_update_size(struct inode *inode, loff_t pos);
961 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
962 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
964 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
967 void fuse_set_initialized(struct fuse_conn *fc);
969 void fuse_unlock_inode(struct inode *inode);
970 void fuse_lock_inode(struct inode *inode);
972 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
973 size_t size, int flags);
974 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
976 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
977 int fuse_removexattr(struct inode *inode, const char *name);
978 extern const struct xattr_handler *fuse_xattr_handlers[];
979 extern const struct xattr_handler *fuse_acl_xattr_handlers[];
982 struct posix_acl *fuse_get_acl(struct inode *inode, int type);
983 int fuse_set_acl(struct inode *inode, struct posix_acl *acl, int type);
985 #endif /* _FS_FUSE_I_H */