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 state */
245 enum fuse_req_state {
254 /** The request IO state (for asynchronous processing) */
255 struct fuse_io_priv {
266 struct completion *done;
270 * A request to the client
273 /** This can be on either pending processing or io lists in
275 struct list_head list;
277 /** Entry on the interrupts list */
278 struct list_head intr_entry;
283 /** Unique ID for the interrupt request */
287 * The following bitfields are either set once before the
288 * request is queued or setting/clearing them is protected by
292 /** True if the request has reply */
295 /** Force sending of the request even if interrupted */
298 /** The request was aborted */
301 /** Request is sent in the background */
302 unsigned background:1;
304 /** The request has been interrupted */
305 unsigned interrupted:1;
307 /** Data is being copied to/from the request */
310 /** Request is counted as "waiting" */
313 /** State of the request */
314 enum fuse_req_state state;
316 /** The request input */
319 /** The request output */
322 /** Used to wake up the task waiting for completion of request*/
323 wait_queue_head_t waitq;
325 /** Data for asynchronous requests */
328 struct fuse_release_in in;
331 struct fuse_init_in init_in;
332 struct fuse_init_out init_out;
333 struct cuse_init_in cuse_init_in;
335 struct fuse_read_in in;
339 struct fuse_write_in in;
340 struct fuse_write_out out;
341 struct fuse_req *next;
343 struct fuse_notify_retrieve_in retrieve_in;
349 /** page-descriptor vector */
350 struct fuse_page_desc *page_descs;
352 /** size of the 'pages' array */
355 /** inline page vector */
356 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
358 /** inline page-descriptor vector */
359 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
361 /** number of pages in vector */
364 /** File used in the request (or NULL) */
365 struct fuse_file *ff;
367 /** Inode used in the request or NULL */
370 /** AIO control block */
371 struct fuse_io_priv *io;
373 /** Link on fi->writepages */
374 struct list_head writepages_entry;
376 /** Request completion callback */
377 void (*end)(struct fuse_conn *, struct fuse_req *);
379 /** Request is stolen from fuse_file->reserved_req */
380 struct file *stolen_file;
386 * This structure is created, when the filesystem is mounted, and is
387 * destroyed, when the client device is closed and the filesystem is
391 /** Lock protecting accessess to members of this structure */
399 /** The user id for this mount */
402 /** The group id for this mount */
405 /** The fuse mount flags for this mount */
408 /** Maximum read size */
411 /** Maximum write size */
414 /** Readers of the connection are waiting on this */
415 wait_queue_head_t waitq;
417 /** The list of pending requests */
418 struct list_head pending;
420 /** The list of requests being processed */
421 struct list_head processing;
423 /** The list of requests under I/O */
426 /** The next unique kernel file handle */
429 /** rbtree of fuse_files waiting for poll events indexed by ph */
430 struct rb_root polled_files;
432 /** Maximum number of outstanding background requests */
433 unsigned max_background;
435 /** Number of background requests at which congestion starts */
436 unsigned congestion_threshold;
438 /** Number of requests currently in the background */
439 unsigned num_background;
441 /** Number of background requests currently queued for userspace */
442 unsigned active_background;
444 /** The list of background requests set aside for later queuing */
445 struct list_head bg_queue;
447 /** Pending interrupts */
448 struct list_head interrupts;
450 /** Queue of pending forgets */
451 struct fuse_forget_link forget_list_head;
452 struct fuse_forget_link *forget_list_tail;
454 /** Batching of FORGET requests (positive indicates FORGET batch) */
457 /** Flag indicating that INIT reply has been received. Allocating
458 * any fuse request will be suspended until the flag is set */
461 /** Flag indicating if connection is blocked. This will be
462 the case before the INIT reply is received, and if there
463 are too many outstading backgrounds requests */
466 /** waitq for blocked connection */
467 wait_queue_head_t blocked_waitq;
469 /** waitq for reserved requests */
470 wait_queue_head_t reserved_req_waitq;
472 /** The next unique request id */
475 /** Connection established, cleared on umount, connection
476 abort and device release */
479 /** Connection failed (version mismatch). Cannot race with
480 setting other bitfields since it is only set once in INIT
481 reply, before any other request, and never cleared */
482 unsigned conn_error:1;
484 /** Connection successful. Only set in INIT */
485 unsigned conn_init:1;
487 /** Do readpages asynchronously? Only set in INIT */
488 unsigned async_read:1;
490 /** Do not send separate SETATTR request before open(O_TRUNC) */
491 unsigned atomic_o_trunc:1;
493 /** Filesystem supports NFS exporting. Only set in INIT */
494 unsigned export_support:1;
496 /** Set if bdi is valid */
497 unsigned bdi_initialized:1;
499 /** write-back cache policy (default is write-through) */
500 unsigned writeback_cache:1;
503 * The following bitfields are only for optimization purposes
504 * and hence races in setting them will not cause malfunction
507 /** Is open/release not implemented by fs? */
510 /** Is fsync not implemented by fs? */
513 /** Is fsyncdir not implemented by fs? */
514 unsigned no_fsyncdir:1;
516 /** Is flush not implemented by fs? */
519 /** Is setxattr not implemented by fs? */
520 unsigned no_setxattr:1;
522 /** Is getxattr not implemented by fs? */
523 unsigned no_getxattr:1;
525 /** Is listxattr not implemented by fs? */
526 unsigned no_listxattr:1;
528 /** Is removexattr not implemented by fs? */
529 unsigned no_removexattr:1;
531 /** Are posix file locking primitives not implemented by fs? */
534 /** Is access not implemented by fs? */
535 unsigned no_access:1;
537 /** Is create not implemented by fs? */
538 unsigned no_create:1;
540 /** Is interrupt not implemented by fs? */
541 unsigned no_interrupt:1;
543 /** Is bmap not implemented by fs? */
546 /** Is poll not implemented by fs? */
549 /** Do multi-page cached writes */
550 unsigned big_writes:1;
552 /** Don't apply umask to creation modes */
553 unsigned dont_mask:1;
555 /** Are BSD file locking primitives not implemented by fs? */
558 /** Is fallocate not implemented by fs? */
559 unsigned no_fallocate:1;
561 /** Is rename with flags implemented by fs? */
562 unsigned no_rename2:1;
564 /** Use enhanced/automatic page cache invalidation. */
565 unsigned auto_inval_data:1;
567 /** Does the filesystem support readdirplus? */
568 unsigned do_readdirplus:1;
570 /** Does the filesystem want adaptive readdirplus? */
571 unsigned readdirplus_auto:1;
573 /** Does the filesystem support asynchronous direct-IO submission? */
574 unsigned async_dio:1;
576 /** The number of requests waiting for completion */
577 atomic_t num_waiting;
579 /** Negotiated minor version */
582 /** Backing dev info */
583 struct backing_dev_info bdi;
585 /** Entry on the fuse_conn_list */
586 struct list_head entry;
588 /** Device ID from super block */
591 /** Dentries in the control filesystem */
592 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
594 /** number of dentries used in the above array */
597 /** O_ASYNC requests */
598 struct fasync_struct *fasync;
600 /** Key for lock owner ID scrambling */
603 /** Reserved request for the DESTROY message */
604 struct fuse_req *destroy_req;
606 /** Version counter for attribute changes */
609 /** Called on final put */
610 void (*release)(struct fuse_conn *);
612 /** Super block for this connection. */
613 struct super_block *sb;
615 /** Read/write semaphore to hold when accessing sb. */
616 struct rw_semaphore killsb;
619 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
621 return sb->s_fs_info;
624 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
626 return get_fuse_conn_super(inode->i_sb);
629 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
631 return container_of(inode, struct fuse_inode, inode);
634 static inline u64 get_node_id(struct inode *inode)
636 return get_fuse_inode(inode)->nodeid;
639 /** Device operations */
640 extern const struct file_operations fuse_dev_operations;
642 extern const struct dentry_operations fuse_dentry_operations;
645 * Inode to nodeid comparison.
647 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
650 * Get a filled in inode
652 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
653 int generation, struct fuse_attr *attr,
654 u64 attr_valid, u64 attr_version);
656 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
657 struct fuse_entry_out *outarg, struct inode **inode);
660 * Send FORGET command
662 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
663 u64 nodeid, u64 nlookup);
665 struct fuse_forget_link *fuse_alloc_forget(void);
667 /* Used by READDIRPLUS */
668 void fuse_force_forget(struct file *file, u64 nodeid);
671 * Initialize READ or READDIR request
673 void fuse_read_fill(struct fuse_req *req, struct file *file,
674 loff_t pos, size_t count, int opcode);
677 * Send OPEN or OPENDIR request
679 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
681 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
682 struct fuse_file *fuse_file_get(struct fuse_file *ff);
683 void fuse_file_free(struct fuse_file *ff);
684 void fuse_finish_open(struct inode *inode, struct file *file);
686 void fuse_sync_release(struct fuse_file *ff, int flags);
689 * Send RELEASE or RELEASEDIR request
691 void fuse_release_common(struct file *file, int opcode);
694 * Send FSYNC or FSYNCDIR request
696 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
697 int datasync, int isdir);
702 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
703 struct fuse_notify_poll_wakeup_out *outarg);
706 * Initialize file operations on a regular file
708 void fuse_init_file_inode(struct inode *inode);
711 * Initialize inode operations on regular files and special files
713 void fuse_init_common(struct inode *inode);
716 * Initialize inode and file operations on a directory
718 void fuse_init_dir(struct inode *inode);
721 * Initialize inode operations on a symlink
723 void fuse_init_symlink(struct inode *inode);
726 * Change attributes of an inode
728 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
729 u64 attr_valid, u64 attr_version);
731 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
735 * Initialize the client device
737 int fuse_dev_init(void);
740 * Cleanup the client device
742 void fuse_dev_cleanup(void);
744 int fuse_ctl_init(void);
745 void __exit fuse_ctl_cleanup(void);
750 struct fuse_req *fuse_request_alloc(unsigned npages);
752 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
757 void fuse_request_free(struct fuse_req *req);
760 * Get a request, may fail with -ENOMEM,
761 * caller should specify # elements in req->pages[] explicitly
763 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
764 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
768 * Increment reference count on request
770 void __fuse_get_request(struct fuse_req *req);
773 * Gets a requests for a file operation, always succeeds
775 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
779 * Decrement reference count of a request. If count goes to zero free
782 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
785 * Send a request (synchronous)
787 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
790 * Simple request sending that does request allocation and freeing
792 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
795 * Send a request in the background
797 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
799 void fuse_request_send_background_locked(struct fuse_conn *fc,
800 struct fuse_req *req);
802 /* Abort all requests */
803 void fuse_abort_conn(struct fuse_conn *fc);
806 * Invalidate inode attributes
808 void fuse_invalidate_attr(struct inode *inode);
810 void fuse_invalidate_entry_cache(struct dentry *entry);
812 void fuse_invalidate_atime(struct inode *inode);
815 * Acquire reference to fuse_conn
817 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
820 * Initialize fuse_conn
822 void fuse_conn_init(struct fuse_conn *fc);
825 * Release reference to fuse_conn
827 void fuse_conn_put(struct fuse_conn *fc);
830 * Add connection to control filesystem
832 int fuse_ctl_add_conn(struct fuse_conn *fc);
835 * Remove connection from control filesystem
837 void fuse_ctl_remove_conn(struct fuse_conn *fc);
840 * Is file type valid?
842 int fuse_valid_type(int m);
845 * Is current process allowed to perform filesystem operation?
847 int fuse_allow_current_process(struct fuse_conn *fc);
849 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
851 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
852 struct file *file, bool *refreshed);
854 void fuse_flush_writepages(struct inode *inode);
856 void fuse_set_nowrite(struct inode *inode);
857 void fuse_release_nowrite(struct inode *inode);
859 u64 fuse_get_attr_version(struct fuse_conn *fc);
862 * File-system tells the kernel to invalidate cache for the given node id.
864 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
865 loff_t offset, loff_t len);
868 * File-system tells the kernel to invalidate parent attributes and
869 * the dentry matching parent/name.
871 * If the child_nodeid is non-zero and:
872 * - matches the inode number for the dentry matching parent/name,
873 * - is not a mount point
874 * - is a file or oan empty directory
875 * then the dentry is unhashed (d_delete()).
877 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
878 u64 child_nodeid, struct qstr *name);
880 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
884 * fuse_direct_io() flags
887 /** If set, it is WRITE; otherwise - READ */
888 #define FUSE_DIO_WRITE (1 << 0)
890 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
891 #define FUSE_DIO_CUSE (1 << 1)
893 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
894 loff_t *ppos, int flags);
895 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
897 long fuse_ioctl_common(struct file *file, unsigned int cmd,
898 unsigned long arg, unsigned int flags);
899 unsigned fuse_file_poll(struct file *file, poll_table *wait);
900 int fuse_dev_release(struct inode *inode, struct file *file);
902 bool fuse_write_update_size(struct inode *inode, loff_t pos);
904 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
905 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
907 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
910 void fuse_set_initialized(struct fuse_conn *fc);
912 #endif /* _FS_FUSE_I_H */