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 {
269 * A request to the client
272 /** This can be on either pending processing or io lists in
274 struct list_head list;
276 /** Entry on the interrupts list */
277 struct list_head intr_entry;
282 /** Unique ID for the interrupt request */
286 * The following bitfields are either set once before the
287 * request is queued or setting/clearing them is protected by
291 /** True if the request has reply */
294 /** Force sending of the request even if interrupted */
297 /** The request was aborted */
300 /** Request is sent in the background */
301 unsigned background:1;
303 /** The request has been interrupted */
304 unsigned interrupted:1;
306 /** Data is being copied to/from the request */
309 /** Request is counted as "waiting" */
312 /** State of the request */
313 enum fuse_req_state state;
315 /** The request input */
318 /** The request output */
321 /** Used to wake up the task waiting for completion of request*/
322 wait_queue_head_t waitq;
324 /** Data for asynchronous requests */
327 struct fuse_release_in in;
330 struct fuse_init_in init_in;
331 struct fuse_init_out init_out;
332 struct cuse_init_in cuse_init_in;
334 struct fuse_read_in in;
338 struct fuse_write_in in;
339 struct fuse_write_out out;
340 struct fuse_req *next;
342 struct fuse_notify_retrieve_in retrieve_in;
348 /** page-descriptor vector */
349 struct fuse_page_desc *page_descs;
351 /** size of the 'pages' array */
354 /** inline page vector */
355 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
357 /** inline page-descriptor vector */
358 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
360 /** number of pages in vector */
363 /** File used in the request (or NULL) */
364 struct fuse_file *ff;
366 /** Inode used in the request or NULL */
369 /** AIO control block */
370 struct fuse_io_priv *io;
372 /** Link on fi->writepages */
373 struct list_head writepages_entry;
375 /** Request completion callback */
376 void (*end)(struct fuse_conn *, struct fuse_req *);
378 /** Request is stolen from fuse_file->reserved_req */
379 struct file *stolen_file;
385 * This structure is created, when the filesystem is mounted, and is
386 * destroyed, when the client device is closed and the filesystem is
390 /** Lock protecting accessess to members of this structure */
398 /** The user id for this mount */
401 /** The group id for this mount */
404 /** The fuse mount flags for this mount */
407 /** Maximum read size */
410 /** Maximum write size */
413 /** Readers of the connection are waiting on this */
414 wait_queue_head_t waitq;
416 /** The list of pending requests */
417 struct list_head pending;
419 /** The list of requests being processed */
420 struct list_head processing;
422 /** The list of requests under I/O */
425 /** The next unique kernel file handle */
428 /** rbtree of fuse_files waiting for poll events indexed by ph */
429 struct rb_root polled_files;
431 /** Maximum number of outstanding background requests */
432 unsigned max_background;
434 /** Number of background requests at which congestion starts */
435 unsigned congestion_threshold;
437 /** Number of requests currently in the background */
438 unsigned num_background;
440 /** Number of background requests currently queued for userspace */
441 unsigned active_background;
443 /** The list of background requests set aside for later queuing */
444 struct list_head bg_queue;
446 /** Pending interrupts */
447 struct list_head interrupts;
449 /** Queue of pending forgets */
450 struct fuse_forget_link forget_list_head;
451 struct fuse_forget_link *forget_list_tail;
453 /** Batching of FORGET requests (positive indicates FORGET batch) */
456 /** Flag indicating that INIT reply has been received. Allocating
457 * any fuse request will be suspended until the flag is set */
460 /** Flag indicating if connection is blocked. This will be
461 the case before the INIT reply is received, and if there
462 are too many outstading backgrounds requests */
465 /** waitq for blocked connection */
466 wait_queue_head_t blocked_waitq;
468 /** waitq for reserved requests */
469 wait_queue_head_t reserved_req_waitq;
471 /** The next unique request id */
474 /** Connection established, cleared on umount, connection
475 abort and device release */
478 /** Connection failed (version mismatch). Cannot race with
479 setting other bitfields since it is only set once in INIT
480 reply, before any other request, and never cleared */
481 unsigned conn_error:1;
483 /** Connection successful. Only set in INIT */
484 unsigned conn_init:1;
486 /** Do readpages asynchronously? Only set in INIT */
487 unsigned async_read:1;
489 /** Do not send separate SETATTR request before open(O_TRUNC) */
490 unsigned atomic_o_trunc:1;
492 /** Filesystem supports NFS exporting. Only set in INIT */
493 unsigned export_support:1;
495 /** Set if bdi is valid */
496 unsigned bdi_initialized:1;
498 /** write-back cache policy (default is write-through) */
499 unsigned writeback_cache:1;
502 * The following bitfields are only for optimization purposes
503 * and hence races in setting them will not cause malfunction
506 /** Is open/release not implemented by fs? */
509 /** Is fsync not implemented by fs? */
512 /** Is fsyncdir not implemented by fs? */
513 unsigned no_fsyncdir:1;
515 /** Is flush not implemented by fs? */
518 /** Is setxattr not implemented by fs? */
519 unsigned no_setxattr:1;
521 /** Is getxattr not implemented by fs? */
522 unsigned no_getxattr:1;
524 /** Is listxattr not implemented by fs? */
525 unsigned no_listxattr:1;
527 /** Is removexattr not implemented by fs? */
528 unsigned no_removexattr:1;
530 /** Are posix file locking primitives not implemented by fs? */
533 /** Is access not implemented by fs? */
534 unsigned no_access:1;
536 /** Is create not implemented by fs? */
537 unsigned no_create:1;
539 /** Is interrupt not implemented by fs? */
540 unsigned no_interrupt:1;
542 /** Is bmap not implemented by fs? */
545 /** Is poll not implemented by fs? */
548 /** Do multi-page cached writes */
549 unsigned big_writes:1;
551 /** Don't apply umask to creation modes */
552 unsigned dont_mask:1;
554 /** Are BSD file locking primitives not implemented by fs? */
557 /** Is fallocate not implemented by fs? */
558 unsigned no_fallocate:1;
560 /** Is rename with flags implemented by fs? */
561 unsigned no_rename2:1;
563 /** Use enhanced/automatic page cache invalidation. */
564 unsigned auto_inval_data:1;
566 /** Does the filesystem support readdirplus? */
567 unsigned do_readdirplus:1;
569 /** Does the filesystem want adaptive readdirplus? */
570 unsigned readdirplus_auto:1;
572 /** Does the filesystem support asynchronous direct-IO submission? */
573 unsigned async_dio:1;
575 /** The number of requests waiting for completion */
576 atomic_t num_waiting;
578 /** Negotiated minor version */
581 /** Backing dev info */
582 struct backing_dev_info bdi;
584 /** Entry on the fuse_conn_list */
585 struct list_head entry;
587 /** Device ID from super block */
590 /** Dentries in the control filesystem */
591 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
593 /** number of dentries used in the above array */
596 /** O_ASYNC requests */
597 struct fasync_struct *fasync;
599 /** Key for lock owner ID scrambling */
602 /** Reserved request for the DESTROY message */
603 struct fuse_req *destroy_req;
605 /** Version counter for attribute changes */
608 /** Called on final put */
609 void (*release)(struct fuse_conn *);
611 /** Super block for this connection. */
612 struct super_block *sb;
614 /** Read/write semaphore to hold when accessing sb. */
615 struct rw_semaphore killsb;
618 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
620 return sb->s_fs_info;
623 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
625 return get_fuse_conn_super(inode->i_sb);
628 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
630 return container_of(inode, struct fuse_inode, inode);
633 static inline u64 get_node_id(struct inode *inode)
635 return get_fuse_inode(inode)->nodeid;
638 /** Device operations */
639 extern const struct file_operations fuse_dev_operations;
641 extern const struct dentry_operations fuse_dentry_operations;
644 * Inode to nodeid comparison.
646 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
649 * Get a filled in inode
651 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
652 int generation, struct fuse_attr *attr,
653 u64 attr_valid, u64 attr_version);
655 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
656 struct fuse_entry_out *outarg, struct inode **inode);
659 * Send FORGET command
661 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
662 u64 nodeid, u64 nlookup);
664 struct fuse_forget_link *fuse_alloc_forget(void);
666 /* Used by READDIRPLUS */
667 void fuse_force_forget(struct file *file, u64 nodeid);
670 * Initialize READ or READDIR request
672 void fuse_read_fill(struct fuse_req *req, struct file *file,
673 loff_t pos, size_t count, int opcode);
676 * Send OPEN or OPENDIR request
678 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
680 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
681 struct fuse_file *fuse_file_get(struct fuse_file *ff);
682 void fuse_file_free(struct fuse_file *ff);
683 void fuse_finish_open(struct inode *inode, struct file *file);
685 void fuse_sync_release(struct fuse_file *ff, int flags);
688 * Send RELEASE or RELEASEDIR request
690 void fuse_release_common(struct file *file, int opcode);
693 * Send FSYNC or FSYNCDIR request
695 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
696 int datasync, int isdir);
701 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
702 struct fuse_notify_poll_wakeup_out *outarg);
705 * Initialize file operations on a regular file
707 void fuse_init_file_inode(struct inode *inode);
710 * Initialize inode operations on regular files and special files
712 void fuse_init_common(struct inode *inode);
715 * Initialize inode and file operations on a directory
717 void fuse_init_dir(struct inode *inode);
720 * Initialize inode operations on a symlink
722 void fuse_init_symlink(struct inode *inode);
725 * Change attributes of an inode
727 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
728 u64 attr_valid, u64 attr_version);
730 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
734 * Initialize the client device
736 int fuse_dev_init(void);
739 * Cleanup the client device
741 void fuse_dev_cleanup(void);
743 int fuse_ctl_init(void);
744 void __exit fuse_ctl_cleanup(void);
749 struct fuse_req *fuse_request_alloc(unsigned npages);
751 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
756 void fuse_request_free(struct fuse_req *req);
759 * Get a request, may fail with -ENOMEM,
760 * caller should specify # elements in req->pages[] explicitly
762 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
763 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
767 * Increment reference count on request
769 void __fuse_get_request(struct fuse_req *req);
772 * Gets a requests for a file operation, always succeeds
774 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
778 * Decrement reference count of a request. If count goes to zero free
781 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
784 * Send a request (synchronous)
786 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
789 * Simple request sending that does request allocation and freeing
791 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
794 * Send a request in the background
796 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
798 void fuse_request_send_background_locked(struct fuse_conn *fc,
799 struct fuse_req *req);
801 /* Abort all requests */
802 void fuse_abort_conn(struct fuse_conn *fc);
805 * Invalidate inode attributes
807 void fuse_invalidate_attr(struct inode *inode);
809 void fuse_invalidate_entry_cache(struct dentry *entry);
811 void fuse_invalidate_atime(struct inode *inode);
814 * Acquire reference to fuse_conn
816 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
819 * Initialize fuse_conn
821 void fuse_conn_init(struct fuse_conn *fc);
824 * Release reference to fuse_conn
826 void fuse_conn_put(struct fuse_conn *fc);
829 * Add connection to control filesystem
831 int fuse_ctl_add_conn(struct fuse_conn *fc);
834 * Remove connection from control filesystem
836 void fuse_ctl_remove_conn(struct fuse_conn *fc);
839 * Is file type valid?
841 int fuse_valid_type(int m);
844 * Is current process allowed to perform filesystem operation?
846 int fuse_allow_current_process(struct fuse_conn *fc);
848 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
850 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
851 struct file *file, bool *refreshed);
853 void fuse_flush_writepages(struct inode *inode);
855 void fuse_set_nowrite(struct inode *inode);
856 void fuse_release_nowrite(struct inode *inode);
858 u64 fuse_get_attr_version(struct fuse_conn *fc);
861 * File-system tells the kernel to invalidate cache for the given node id.
863 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
864 loff_t offset, loff_t len);
867 * File-system tells the kernel to invalidate parent attributes and
868 * the dentry matching parent/name.
870 * If the child_nodeid is non-zero and:
871 * - matches the inode number for the dentry matching parent/name,
872 * - is not a mount point
873 * - is a file or oan empty directory
874 * then the dentry is unhashed (d_delete()).
876 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
877 u64 child_nodeid, struct qstr *name);
879 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
883 * fuse_direct_io() flags
886 /** If set, it is WRITE; otherwise - READ */
887 #define FUSE_DIO_WRITE (1 << 0)
889 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
890 #define FUSE_DIO_CUSE (1 << 1)
892 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
893 loff_t *ppos, int flags);
894 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
896 long fuse_ioctl_common(struct file *file, unsigned int cmd,
897 unsigned long arg, unsigned int flags);
898 unsigned fuse_file_poll(struct file *file, poll_table *wait);
899 int fuse_dev_release(struct inode *inode, struct file *file);
901 bool fuse_write_update_size(struct inode *inode, loff_t pos);
903 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
904 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
906 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
909 void fuse_set_initialized(struct fuse_conn *fc);
911 #endif /* _FS_FUSE_I_H */