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 {
256 struct completion *done;
260 #define FUSE_IO_PRIV_SYNC(i) \
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 /** Lock protecting accesses to members of this structure */
394 /** Readers of the connection are waiting on this */
395 wait_queue_head_t waitq;
397 /** The next unique request id */
400 /** The list of pending requests */
401 struct list_head pending;
403 /** Pending interrupts */
404 struct list_head interrupts;
406 /** Queue of pending forgets */
407 struct fuse_forget_link forget_list_head;
408 struct fuse_forget_link *forget_list_tail;
410 /** Batching of FORGET requests (positive indicates FORGET batch) */
413 /** O_ASYNC requests */
414 struct fasync_struct *fasync;
418 /** Connection established */
421 /** Lock protecting accessess to members of this structure */
424 /** The list of requests being processed */
425 struct list_head processing;
427 /** The list of requests under I/O */
432 * Fuse device instance
435 /** Fuse connection for this device */
436 struct fuse_conn *fc;
438 /** Processing queue */
439 struct fuse_pqueue pq;
441 /** list entry on fc->devices */
442 struct list_head entry;
448 * This structure is created, when the filesystem is mounted, and is
449 * destroyed, when the client device is closed and the filesystem is
453 /** Lock protecting accessess to members of this structure */
459 /** Number of fuse_dev's */
464 /** The user id for this mount */
467 /** The group id for this mount */
470 /** The pid namespace for this mount */
471 struct pid_namespace *pid_ns;
473 /** The user namespace for this mount */
474 struct user_namespace *user_ns;
476 /** Maximum read size */
479 /** Maximum write size */
483 struct fuse_iqueue iq;
485 /** The next unique kernel file handle */
488 /** rbtree of fuse_files waiting for poll events indexed by ph */
489 struct rb_root polled_files;
491 /** Maximum number of outstanding background requests */
492 unsigned max_background;
494 /** Number of background requests at which congestion starts */
495 unsigned congestion_threshold;
497 /** Number of requests currently in the background */
498 unsigned num_background;
500 /** Number of background requests currently queued for userspace */
501 unsigned active_background;
503 /** The list of background requests set aside for later queuing */
504 struct list_head bg_queue;
506 /** Flag indicating that INIT reply has been received. Allocating
507 * any fuse request will be suspended until the flag is set */
510 /** Flag indicating if connection is blocked. This will be
511 the case before the INIT reply is received, and if there
512 are too many outstading backgrounds requests */
515 /** waitq for blocked connection */
516 wait_queue_head_t blocked_waitq;
518 /** waitq for reserved requests */
519 wait_queue_head_t reserved_req_waitq;
521 /** Connection established, cleared on umount, connection
522 abort and device release */
525 /** Connection aborted via sysfs */
528 /** Connection failed (version mismatch). Cannot race with
529 setting other bitfields since it is only set once in INIT
530 reply, before any other request, and never cleared */
531 unsigned conn_error:1;
533 /** Connection successful. Only set in INIT */
534 unsigned conn_init:1;
536 /** Do readpages asynchronously? Only set in INIT */
537 unsigned async_read:1;
539 /** Return an unique read error after abort. Only set in INIT */
540 unsigned abort_err:1;
542 /** Do not send separate SETATTR request before open(O_TRUNC) */
543 unsigned atomic_o_trunc:1;
545 /** Filesystem supports NFS exporting. Only set in INIT */
546 unsigned export_support:1;
548 /** write-back cache policy (default is write-through) */
549 unsigned writeback_cache:1;
551 /** allow parallel lookups and readdir (default is serialized) */
552 unsigned parallel_dirops:1;
554 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
555 unsigned handle_killpriv:1;
558 * The following bitfields are only for optimization purposes
559 * and hence races in setting them will not cause malfunction
562 /** Is open/release not implemented by fs? */
565 /** Is fsync not implemented by fs? */
568 /** Is fsyncdir not implemented by fs? */
569 unsigned no_fsyncdir:1;
571 /** Is flush not implemented by fs? */
574 /** Is setxattr not implemented by fs? */
575 unsigned no_setxattr:1;
577 /** Is getxattr not implemented by fs? */
578 unsigned no_getxattr:1;
580 /** Is listxattr not implemented by fs? */
581 unsigned no_listxattr:1;
583 /** Is removexattr not implemented by fs? */
584 unsigned no_removexattr:1;
586 /** Are posix file locking primitives not implemented by fs? */
589 /** Is access not implemented by fs? */
590 unsigned no_access:1;
592 /** Is create not implemented by fs? */
593 unsigned no_create:1;
595 /** Is interrupt not implemented by fs? */
596 unsigned no_interrupt:1;
598 /** Is bmap not implemented by fs? */
601 /** Is poll not implemented by fs? */
604 /** Do multi-page cached writes */
605 unsigned big_writes:1;
607 /** Don't apply umask to creation modes */
608 unsigned dont_mask:1;
610 /** Are BSD file locking primitives not implemented by fs? */
613 /** Is fallocate not implemented by fs? */
614 unsigned no_fallocate:1;
616 /** Is rename with flags implemented by fs? */
617 unsigned no_rename2:1;
619 /** Use enhanced/automatic page cache invalidation. */
620 unsigned auto_inval_data:1;
622 /** Does the filesystem support readdirplus? */
623 unsigned do_readdirplus:1;
625 /** Does the filesystem want adaptive readdirplus? */
626 unsigned readdirplus_auto:1;
628 /** Does the filesystem support asynchronous direct-IO submission? */
629 unsigned async_dio:1;
631 /** Is lseek not implemented by fs? */
634 /** Does the filesystem support posix acls? */
635 unsigned posix_acl:1;
637 /** Check permissions based on the file mode or not? */
638 unsigned default_permissions:1;
640 /** Allow other than the mounter user to access the filesystem ? */
641 unsigned allow_other:1;
643 /** The number of requests waiting for completion */
644 atomic_t num_waiting;
646 /** Negotiated minor version */
649 /** Entry on the fuse_conn_list */
650 struct list_head entry;
652 /** Device ID from super block */
655 /** Dentries in the control filesystem */
656 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
658 /** number of dentries used in the above array */
661 /** Key for lock owner ID scrambling */
664 /** Reserved request for the DESTROY message */
665 struct fuse_req *destroy_req;
667 /** Version counter for attribute changes */
670 /** Called on final put */
671 void (*release)(struct fuse_conn *);
673 /** Super block for this connection. */
674 struct super_block *sb;
676 /** Read/write semaphore to hold when accessing sb. */
677 struct rw_semaphore killsb;
679 /** List of device instances belonging to this connection */
680 struct list_head devices;
683 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
685 return sb->s_fs_info;
688 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
690 return get_fuse_conn_super(inode->i_sb);
693 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
695 return container_of(inode, struct fuse_inode, inode);
698 static inline u64 get_node_id(struct inode *inode)
700 return get_fuse_inode(inode)->nodeid;
703 /** Device operations */
704 extern const struct file_operations fuse_dev_operations;
706 extern const struct dentry_operations fuse_dentry_operations;
707 extern const struct dentry_operations fuse_root_dentry_operations;
710 * Inode to nodeid comparison.
712 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
715 * Get a filled in inode
717 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
718 int generation, struct fuse_attr *attr,
719 u64 attr_valid, u64 attr_version);
721 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
722 struct fuse_entry_out *outarg, struct inode **inode);
725 * Send FORGET command
727 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
728 u64 nodeid, u64 nlookup);
730 struct fuse_forget_link *fuse_alloc_forget(void);
732 /* Used by READDIRPLUS */
733 void fuse_force_forget(struct file *file, u64 nodeid);
736 * Initialize READ or READDIR request
738 void fuse_read_fill(struct fuse_req *req, struct file *file,
739 loff_t pos, size_t count, int opcode);
742 * Send OPEN or OPENDIR request
744 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
746 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
747 void fuse_file_free(struct fuse_file *ff);
748 void fuse_finish_open(struct inode *inode, struct file *file);
750 void fuse_sync_release(struct fuse_file *ff, int flags);
753 * Send RELEASE or RELEASEDIR request
755 void fuse_release_common(struct file *file, bool isdir);
758 * Send FSYNC or FSYNCDIR request
760 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
761 int datasync, int isdir);
766 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
767 struct fuse_notify_poll_wakeup_out *outarg);
770 * Initialize file operations on a regular file
772 void fuse_init_file_inode(struct inode *inode);
775 * Initialize inode operations on regular files and special files
777 void fuse_init_common(struct inode *inode);
780 * Initialize inode and file operations on a directory
782 void fuse_init_dir(struct inode *inode);
785 * Initialize inode operations on a symlink
787 void fuse_init_symlink(struct inode *inode);
790 * Change attributes of an inode
792 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
793 u64 attr_valid, u64 attr_version);
795 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
799 * Initialize the client device
801 int fuse_dev_init(void);
804 * Cleanup the client device
806 void fuse_dev_cleanup(void);
808 int fuse_ctl_init(void);
809 void __exit fuse_ctl_cleanup(void);
814 struct fuse_req *fuse_request_alloc(unsigned npages);
816 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
821 void fuse_request_free(struct fuse_req *req);
824 * Get a request, may fail with -ENOMEM,
825 * caller should specify # elements in req->pages[] explicitly
827 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
828 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
832 * Increment reference count on request
834 void __fuse_get_request(struct fuse_req *req);
837 * Gets a requests for a file operation, always succeeds
839 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
843 * Decrement reference count of a request. If count goes to zero free
846 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
849 * Send a request (synchronous)
851 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
854 * Simple request sending that does request allocation and freeing
856 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
859 * Send a request in the background
861 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
863 void fuse_request_send_background_locked(struct fuse_conn *fc,
864 struct fuse_req *req);
866 /* Abort all requests */
867 void fuse_abort_conn(struct fuse_conn *fc, bool is_abort);
868 void fuse_wait_aborted(struct fuse_conn *fc);
871 * Invalidate inode attributes
873 void fuse_invalidate_attr(struct inode *inode);
875 void fuse_invalidate_entry_cache(struct dentry *entry);
877 void fuse_invalidate_atime(struct inode *inode);
880 * Acquire reference to fuse_conn
882 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
885 * Initialize fuse_conn
887 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns);
890 * Release reference to fuse_conn
892 void fuse_conn_put(struct fuse_conn *fc);
894 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc);
895 void fuse_dev_free(struct fuse_dev *fud);
898 * Add connection to control filesystem
900 int fuse_ctl_add_conn(struct fuse_conn *fc);
903 * Remove connection from control filesystem
905 void fuse_ctl_remove_conn(struct fuse_conn *fc);
908 * Is file type valid?
910 int fuse_valid_type(int m);
912 bool fuse_invalid_attr(struct fuse_attr *attr);
915 * Is current process allowed to perform filesystem operation?
917 int fuse_allow_current_process(struct fuse_conn *fc);
919 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
921 void fuse_update_ctime(struct inode *inode);
923 int fuse_update_attributes(struct inode *inode, struct file *file);
925 void fuse_flush_writepages(struct inode *inode);
927 void fuse_set_nowrite(struct inode *inode);
928 void fuse_release_nowrite(struct inode *inode);
930 u64 fuse_get_attr_version(struct fuse_conn *fc);
933 * File-system tells the kernel to invalidate cache for the given node id.
935 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
936 loff_t offset, loff_t len);
939 * File-system tells the kernel to invalidate parent attributes and
940 * the dentry matching parent/name.
942 * If the child_nodeid is non-zero and:
943 * - matches the inode number for the dentry matching parent/name,
944 * - is not a mount point
945 * - is a file or oan empty directory
946 * then the dentry is unhashed (d_delete()).
948 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
949 u64 child_nodeid, struct qstr *name);
951 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
955 * fuse_direct_io() flags
958 /** If set, it is WRITE; otherwise - READ */
959 #define FUSE_DIO_WRITE (1 << 0)
961 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
962 #define FUSE_DIO_CUSE (1 << 1)
964 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
965 loff_t *ppos, int flags);
966 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
968 long fuse_ioctl_common(struct file *file, unsigned int cmd,
969 unsigned long arg, unsigned int flags);
970 __poll_t fuse_file_poll(struct file *file, poll_table *wait);
971 int fuse_dev_release(struct inode *inode, struct file *file);
973 bool fuse_write_update_size(struct inode *inode, loff_t pos);
975 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
976 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
978 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
981 void fuse_set_initialized(struct fuse_conn *fc);
983 void fuse_unlock_inode(struct inode *inode, bool locked);
984 bool fuse_lock_inode(struct inode *inode);
986 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
987 size_t size, int flags);
988 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
990 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
991 int fuse_removexattr(struct inode *inode, const char *name);
992 extern const struct xattr_handler *fuse_xattr_handlers[];
993 extern const struct xattr_handler *fuse_acl_xattr_handlers[];
994 extern const struct xattr_handler *fuse_no_acl_xattr_handlers[];
997 struct posix_acl *fuse_get_acl(struct inode *inode, int type);
998 int fuse_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1000 #endif /* _FS_FUSE_I_H */