2 FUSE: Filesystem in Userspace
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,
126 /** FUSE specific file data */
128 /** Fuse connection for this file */
129 struct fuse_conn *fc;
131 /** Request reserved for flush and release */
132 struct fuse_req *reserved_req;
134 /** Kernel file handle guaranteed to be unique */
137 /** File handle used by userspace */
140 /** Node id of this file */
146 /** FOPEN_* flags returned by open */
149 /** Entry on inode's write_files list */
150 struct list_head write_entry;
152 /** RB node to be linked on fuse_conn->polled_files */
153 struct rb_node polled_node;
155 /** Wait queue head for poll */
156 wait_queue_head_t poll_wait;
158 /** Has flock been performed on this file? */
162 /** One input argument of a request */
168 /** The request input */
170 /** The request header */
171 struct fuse_in_header h;
173 /** True if the data for the last argument is in req->pages */
176 /** Number of arguments */
179 /** Array of arguments */
180 struct fuse_in_arg args[3];
183 /** One output argument of a request */
189 /** The request output */
191 /** Header returned from userspace */
192 struct fuse_out_header h;
195 * The following bitfields are not changed during the request
199 /** Last argument is variable length (can be shorter than
203 /** Last argument is a list of pages to copy data to */
206 /** Zero partially or not copied pages */
207 unsigned page_zeroing:1;
209 /** Pages may be replaced with new ones */
210 unsigned page_replace:1;
212 /** Number or arguments */
215 /** Array of arguments */
216 struct fuse_arg args[2];
219 /** FUSE page descriptor */
220 struct fuse_page_desc {
232 struct fuse_in_arg args[3];
238 struct fuse_arg args[2];
242 #define FUSE_ARGS(args) struct fuse_args args = {}
244 /** The request IO state (for asynchronous processing) */
245 struct fuse_io_priv {
256 struct completion *done;
262 * FR_ISREPLY: set if the request has reply
263 * FR_FORCE: force sending of the request even if interrupted
264 * FR_BACKGROUND: request is sent in the background
265 * FR_WAITING: request is counted as "waiting"
266 * FR_ABORTED: the request was aborted
267 * FR_INTERRUPTED: the request has been interrupted
268 * FR_LOCKED: data is being copied to/from the request
269 * FR_PENDING: request is not yet in userspace
270 * FR_SENT: request is in userspace, waiting for an answer
271 * FR_FINISHED: request is finished
287 * A request to the client
289 * .waitq.lock protects the following fields:
291 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
294 /** This can be on either pending processing or io lists in
296 struct list_head list;
298 /** Entry on the interrupts list */
299 struct list_head intr_entry;
304 /** Unique ID for the interrupt request */
307 /* Request flags, updated with test/set/clear_bit() */
310 /** The request input */
313 /** The request output */
316 /** Used to wake up the task waiting for completion of request*/
317 wait_queue_head_t waitq;
319 /** Data for asynchronous requests */
322 struct fuse_release_in in;
325 struct fuse_init_in init_in;
326 struct fuse_init_out init_out;
327 struct cuse_init_in cuse_init_in;
329 struct fuse_read_in in;
333 struct fuse_write_in in;
334 struct fuse_write_out out;
335 struct fuse_req *next;
337 struct fuse_notify_retrieve_in retrieve_in;
343 /** page-descriptor vector */
344 struct fuse_page_desc *page_descs;
346 /** size of the 'pages' array */
349 /** inline page vector */
350 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
352 /** inline page-descriptor vector */
353 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
355 /** number of pages in vector */
358 /** File used in the request (or NULL) */
359 struct fuse_file *ff;
361 /** Inode used in the request or NULL */
364 /** AIO control block */
365 struct fuse_io_priv *io;
367 /** Link on fi->writepages */
368 struct list_head writepages_entry;
370 /** Request completion callback */
371 void (*end)(struct fuse_conn *, struct fuse_req *);
373 /** Request is stolen from fuse_file->reserved_req */
374 struct file *stolen_file;
378 /** Connection established */
381 /** Readers of the connection are waiting on this */
382 wait_queue_head_t waitq;
384 /** The next unique request id */
387 /** The list of pending requests */
388 struct list_head pending;
390 /** Pending interrupts */
391 struct list_head interrupts;
393 /** Queue of pending forgets */
394 struct fuse_forget_link forget_list_head;
395 struct fuse_forget_link *forget_list_tail;
397 /** Batching of FORGET requests (positive indicates FORGET batch) */
400 /** O_ASYNC requests */
401 struct fasync_struct *fasync;
405 /** Connection established */
408 /** Lock protecting accessess to members of this structure */
411 /** The list of requests being processed */
412 struct list_head processing;
414 /** The list of requests under I/O */
421 * This structure is created, when the filesystem is mounted, and is
422 * destroyed, when the client device is closed and the filesystem is
426 /** Lock protecting accessess to members of this structure */
434 /** The user id for this mount */
437 /** The group id for this mount */
440 /** The fuse mount flags for this mount */
443 /** Maximum read size */
446 /** Maximum write size */
450 struct fuse_iqueue iq;
452 /** Processing queue */
453 struct fuse_pqueue pq;
455 /** The next unique kernel file handle */
458 /** rbtree of fuse_files waiting for poll events indexed by ph */
459 struct rb_root polled_files;
461 /** Maximum number of outstanding background requests */
462 unsigned max_background;
464 /** Number of background requests at which congestion starts */
465 unsigned congestion_threshold;
467 /** Number of requests currently in the background */
468 unsigned num_background;
470 /** Number of background requests currently queued for userspace */
471 unsigned active_background;
473 /** The list of background requests set aside for later queuing */
474 struct list_head bg_queue;
476 /** Flag indicating that INIT reply has been received. Allocating
477 * any fuse request will be suspended until the flag is set */
480 /** Flag indicating if connection is blocked. This will be
481 the case before the INIT reply is received, and if there
482 are too many outstading backgrounds requests */
485 /** waitq for blocked connection */
486 wait_queue_head_t blocked_waitq;
488 /** waitq for reserved requests */
489 wait_queue_head_t reserved_req_waitq;
491 /** Connection established, cleared on umount, connection
492 abort and device release */
495 /** Connection failed (version mismatch). Cannot race with
496 setting other bitfields since it is only set once in INIT
497 reply, before any other request, and never cleared */
498 unsigned conn_error:1;
500 /** Connection successful. Only set in INIT */
501 unsigned conn_init:1;
503 /** Do readpages asynchronously? Only set in INIT */
504 unsigned async_read:1;
506 /** Do not send separate SETATTR request before open(O_TRUNC) */
507 unsigned atomic_o_trunc:1;
509 /** Filesystem supports NFS exporting. Only set in INIT */
510 unsigned export_support:1;
512 /** Set if bdi is valid */
513 unsigned bdi_initialized:1;
515 /** write-back cache policy (default is write-through) */
516 unsigned writeback_cache:1;
519 * The following bitfields are only for optimization purposes
520 * and hence races in setting them will not cause malfunction
523 /** Is open/release not implemented by fs? */
526 /** Is fsync not implemented by fs? */
529 /** Is fsyncdir not implemented by fs? */
530 unsigned no_fsyncdir:1;
532 /** Is flush not implemented by fs? */
535 /** Is setxattr not implemented by fs? */
536 unsigned no_setxattr:1;
538 /** Is getxattr not implemented by fs? */
539 unsigned no_getxattr:1;
541 /** Is listxattr not implemented by fs? */
542 unsigned no_listxattr:1;
544 /** Is removexattr not implemented by fs? */
545 unsigned no_removexattr:1;
547 /** Are posix file locking primitives not implemented by fs? */
550 /** Is access not implemented by fs? */
551 unsigned no_access:1;
553 /** Is create not implemented by fs? */
554 unsigned no_create:1;
556 /** Is interrupt not implemented by fs? */
557 unsigned no_interrupt:1;
559 /** Is bmap not implemented by fs? */
562 /** Is poll not implemented by fs? */
565 /** Do multi-page cached writes */
566 unsigned big_writes:1;
568 /** Don't apply umask to creation modes */
569 unsigned dont_mask:1;
571 /** Are BSD file locking primitives not implemented by fs? */
574 /** Is fallocate not implemented by fs? */
575 unsigned no_fallocate:1;
577 /** Is rename with flags implemented by fs? */
578 unsigned no_rename2:1;
580 /** Use enhanced/automatic page cache invalidation. */
581 unsigned auto_inval_data:1;
583 /** Does the filesystem support readdirplus? */
584 unsigned do_readdirplus:1;
586 /** Does the filesystem want adaptive readdirplus? */
587 unsigned readdirplus_auto:1;
589 /** Does the filesystem support asynchronous direct-IO submission? */
590 unsigned async_dio:1;
592 /** The number of requests waiting for completion */
593 atomic_t num_waiting;
595 /** Negotiated minor version */
598 /** Backing dev info */
599 struct backing_dev_info bdi;
601 /** Entry on the fuse_conn_list */
602 struct list_head entry;
604 /** Device ID from super block */
607 /** Dentries in the control filesystem */
608 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
610 /** number of dentries used in the above array */
613 /** Key for lock owner ID scrambling */
616 /** Reserved request for the DESTROY message */
617 struct fuse_req *destroy_req;
619 /** Version counter for attribute changes */
622 /** Called on final put */
623 void (*release)(struct fuse_conn *);
625 /** Super block for this connection. */
626 struct super_block *sb;
628 /** Read/write semaphore to hold when accessing sb. */
629 struct rw_semaphore killsb;
632 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
634 return sb->s_fs_info;
637 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
639 return get_fuse_conn_super(inode->i_sb);
642 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
644 return container_of(inode, struct fuse_inode, inode);
647 static inline u64 get_node_id(struct inode *inode)
649 return get_fuse_inode(inode)->nodeid;
652 /** Device operations */
653 extern const struct file_operations fuse_dev_operations;
655 extern const struct dentry_operations fuse_dentry_operations;
658 * Inode to nodeid comparison.
660 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
663 * Get a filled in inode
665 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
666 int generation, struct fuse_attr *attr,
667 u64 attr_valid, u64 attr_version);
669 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
670 struct fuse_entry_out *outarg, struct inode **inode);
673 * Send FORGET command
675 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
676 u64 nodeid, u64 nlookup);
678 struct fuse_forget_link *fuse_alloc_forget(void);
680 /* Used by READDIRPLUS */
681 void fuse_force_forget(struct file *file, u64 nodeid);
684 * Initialize READ or READDIR request
686 void fuse_read_fill(struct fuse_req *req, struct file *file,
687 loff_t pos, size_t count, int opcode);
690 * Send OPEN or OPENDIR request
692 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
694 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
695 struct fuse_file *fuse_file_get(struct fuse_file *ff);
696 void fuse_file_free(struct fuse_file *ff);
697 void fuse_finish_open(struct inode *inode, struct file *file);
699 void fuse_sync_release(struct fuse_file *ff, int flags);
702 * Send RELEASE or RELEASEDIR request
704 void fuse_release_common(struct file *file, int opcode);
707 * Send FSYNC or FSYNCDIR request
709 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
710 int datasync, int isdir);
715 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
716 struct fuse_notify_poll_wakeup_out *outarg);
719 * Initialize file operations on a regular file
721 void fuse_init_file_inode(struct inode *inode);
724 * Initialize inode operations on regular files and special files
726 void fuse_init_common(struct inode *inode);
729 * Initialize inode and file operations on a directory
731 void fuse_init_dir(struct inode *inode);
734 * Initialize inode operations on a symlink
736 void fuse_init_symlink(struct inode *inode);
739 * Change attributes of an inode
741 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
742 u64 attr_valid, u64 attr_version);
744 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
748 * Initialize the client device
750 int fuse_dev_init(void);
753 * Cleanup the client device
755 void fuse_dev_cleanup(void);
757 int fuse_ctl_init(void);
758 void __exit fuse_ctl_cleanup(void);
763 struct fuse_req *fuse_request_alloc(unsigned npages);
765 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
770 void fuse_request_free(struct fuse_req *req);
773 * Get a request, may fail with -ENOMEM,
774 * caller should specify # elements in req->pages[] explicitly
776 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
777 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
781 * Increment reference count on request
783 void __fuse_get_request(struct fuse_req *req);
786 * Gets a requests for a file operation, always succeeds
788 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
792 * Decrement reference count of a request. If count goes to zero free
795 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
798 * Send a request (synchronous)
800 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
803 * Simple request sending that does request allocation and freeing
805 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
808 * Send a request in the background
810 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
812 void fuse_request_send_background_locked(struct fuse_conn *fc,
813 struct fuse_req *req);
815 /* Abort all requests */
816 void fuse_abort_conn(struct fuse_conn *fc);
819 * Invalidate inode attributes
821 void fuse_invalidate_attr(struct inode *inode);
823 void fuse_invalidate_entry_cache(struct dentry *entry);
825 void fuse_invalidate_atime(struct inode *inode);
828 * Acquire reference to fuse_conn
830 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
833 * Initialize fuse_conn
835 void fuse_conn_init(struct fuse_conn *fc);
838 * Release reference to fuse_conn
840 void fuse_conn_put(struct fuse_conn *fc);
843 * Add connection to control filesystem
845 int fuse_ctl_add_conn(struct fuse_conn *fc);
848 * Remove connection from control filesystem
850 void fuse_ctl_remove_conn(struct fuse_conn *fc);
853 * Is file type valid?
855 int fuse_valid_type(int m);
858 * Is current process allowed to perform filesystem operation?
860 int fuse_allow_current_process(struct fuse_conn *fc);
862 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
864 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
865 struct file *file, bool *refreshed);
867 void fuse_flush_writepages(struct inode *inode);
869 void fuse_set_nowrite(struct inode *inode);
870 void fuse_release_nowrite(struct inode *inode);
872 u64 fuse_get_attr_version(struct fuse_conn *fc);
875 * File-system tells the kernel to invalidate cache for the given node id.
877 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
878 loff_t offset, loff_t len);
881 * File-system tells the kernel to invalidate parent attributes and
882 * the dentry matching parent/name.
884 * If the child_nodeid is non-zero and:
885 * - matches the inode number for the dentry matching parent/name,
886 * - is not a mount point
887 * - is a file or oan empty directory
888 * then the dentry is unhashed (d_delete()).
890 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
891 u64 child_nodeid, struct qstr *name);
893 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
897 * fuse_direct_io() flags
900 /** If set, it is WRITE; otherwise - READ */
901 #define FUSE_DIO_WRITE (1 << 0)
903 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
904 #define FUSE_DIO_CUSE (1 << 1)
906 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
907 loff_t *ppos, int flags);
908 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
910 long fuse_ioctl_common(struct file *file, unsigned int cmd,
911 unsigned long arg, unsigned int flags);
912 unsigned fuse_file_poll(struct file *file, poll_table *wait);
913 int fuse_dev_release(struct inode *inode, struct file *file);
915 bool fuse_write_update_size(struct inode *inode, loff_t pos);
917 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
918 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
920 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
923 void fuse_set_initialized(struct fuse_conn *fc);
925 #endif /* _FS_FUSE_I_H */