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,
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 {
257 struct completion *done;
261 #define FUSE_IO_PRIV_SYNC(i) \
263 .refcnt = KREF_INIT(1), \
271 * FR_ISREPLY: set if the request has reply
272 * FR_FORCE: force sending of the request even if interrupted
273 * FR_BACKGROUND: request is sent in the background
274 * FR_WAITING: request is counted as "waiting"
275 * FR_ABORTED: the request was aborted
276 * FR_INTERRUPTED: the request has been interrupted
277 * FR_LOCKED: data is being copied to/from the request
278 * FR_PENDING: request is not yet in userspace
279 * FR_SENT: request is in userspace, waiting for an answer
280 * FR_FINISHED: request is finished
281 * FR_PRIVATE: request is on private list
298 * A request to the client
300 * .waitq.lock protects the following fields:
302 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
305 /** This can be on either pending processing or io lists in
307 struct list_head list;
309 /** Entry on the interrupts list */
310 struct list_head intr_entry;
315 /** Unique ID for the interrupt request */
318 /* Request flags, updated with test/set/clear_bit() */
321 /** The request input */
324 /** The request output */
327 /** Used to wake up the task waiting for completion of request*/
328 wait_queue_head_t waitq;
330 /** Data for asynchronous requests */
333 struct fuse_release_in in;
336 struct fuse_init_in init_in;
337 struct fuse_init_out init_out;
338 struct cuse_init_in cuse_init_in;
340 struct fuse_read_in in;
344 struct fuse_write_in in;
345 struct fuse_write_out out;
346 struct fuse_req *next;
348 struct fuse_notify_retrieve_in retrieve_in;
354 /** page-descriptor vector */
355 struct fuse_page_desc *page_descs;
357 /** size of the 'pages' array */
360 /** inline page vector */
361 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
363 /** inline page-descriptor vector */
364 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
366 /** number of pages in vector */
369 /** File used in the request (or NULL) */
370 struct fuse_file *ff;
372 /** Inode used in the request or NULL */
375 /** AIO control block */
376 struct fuse_io_priv *io;
378 /** Link on fi->writepages */
379 struct list_head writepages_entry;
381 /** Request completion callback */
382 void (*end)(struct fuse_conn *, struct fuse_req *);
384 /** Request is stolen from fuse_file->reserved_req */
385 struct file *stolen_file;
389 /** Connection established */
392 /** Readers of the connection are waiting on this */
393 wait_queue_head_t waitq;
395 /** The next unique request id */
398 /** The list of pending requests */
399 struct list_head pending;
401 /** Pending interrupts */
402 struct list_head interrupts;
404 /** Queue of pending forgets */
405 struct fuse_forget_link forget_list_head;
406 struct fuse_forget_link *forget_list_tail;
408 /** Batching of FORGET requests (positive indicates FORGET batch) */
411 /** O_ASYNC requests */
412 struct fasync_struct *fasync;
416 /** Connection established */
419 /** Lock protecting accessess to members of this structure */
422 /** The list of requests being processed */
423 struct list_head processing;
425 /** The list of requests under I/O */
430 * Fuse device instance
433 /** Fuse connection for this device */
434 struct fuse_conn *fc;
436 /** Processing queue */
437 struct fuse_pqueue pq;
439 /** list entry on fc->devices */
440 struct list_head entry;
446 * This structure is created, when the filesystem is mounted, and is
447 * destroyed, when the client device is closed and the filesystem is
451 /** Lock protecting accessess to members of this structure */
457 /** Number of fuse_dev's */
462 /** The user id for this mount */
465 /** The group id for this mount */
468 /** The pid namespace for this mount */
469 struct pid_namespace *pid_ns;
471 /** Maximum read size */
474 /** Maximum write size */
478 struct fuse_iqueue iq;
480 /** The next unique kernel file handle */
483 /** rbtree of fuse_files waiting for poll events indexed by ph */
484 struct rb_root polled_files;
486 /** Maximum number of outstanding background requests */
487 unsigned max_background;
489 /** Number of background requests at which congestion starts */
490 unsigned congestion_threshold;
492 /** Number of requests currently in the background */
493 unsigned num_background;
495 /** Number of background requests currently queued for userspace */
496 unsigned active_background;
498 /** The list of background requests set aside for later queuing */
499 struct list_head bg_queue;
501 /** Flag indicating that INIT reply has been received. Allocating
502 * any fuse request will be suspended until the flag is set */
505 /** Flag indicating if connection is blocked. This will be
506 the case before the INIT reply is received, and if there
507 are too many outstading backgrounds requests */
510 /** waitq for blocked connection */
511 wait_queue_head_t blocked_waitq;
513 /** waitq for reserved requests */
514 wait_queue_head_t reserved_req_waitq;
516 /** Connection established, cleared on umount, connection
517 abort and device release */
520 /** Connection failed (version mismatch). Cannot race with
521 setting other bitfields since it is only set once in INIT
522 reply, before any other request, and never cleared */
523 unsigned conn_error:1;
525 /** Connection successful. Only set in INIT */
526 unsigned conn_init:1;
528 /** Do readpages asynchronously? Only set in INIT */
529 unsigned async_read:1;
531 /** Do not send separate SETATTR request before open(O_TRUNC) */
532 unsigned atomic_o_trunc:1;
534 /** Filesystem supports NFS exporting. Only set in INIT */
535 unsigned export_support:1;
537 /** write-back cache policy (default is write-through) */
538 unsigned writeback_cache:1;
540 /** allow parallel lookups and readdir (default is serialized) */
541 unsigned parallel_dirops:1;
543 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
544 unsigned handle_killpriv:1;
547 * The following bitfields are only for optimization purposes
548 * and hence races in setting them will not cause malfunction
551 /** Is open/release not implemented by fs? */
554 /** Is fsync not implemented by fs? */
557 /** Is fsyncdir not implemented by fs? */
558 unsigned no_fsyncdir:1;
560 /** Is flush not implemented by fs? */
563 /** Is setxattr not implemented by fs? */
564 unsigned no_setxattr:1;
566 /** Is getxattr not implemented by fs? */
567 unsigned no_getxattr:1;
569 /** Is listxattr not implemented by fs? */
570 unsigned no_listxattr:1;
572 /** Is removexattr not implemented by fs? */
573 unsigned no_removexattr:1;
575 /** Are posix file locking primitives not implemented by fs? */
578 /** Is access not implemented by fs? */
579 unsigned no_access:1;
581 /** Is create not implemented by fs? */
582 unsigned no_create:1;
584 /** Is interrupt not implemented by fs? */
585 unsigned no_interrupt:1;
587 /** Is bmap not implemented by fs? */
590 /** Is poll not implemented by fs? */
593 /** Do multi-page cached writes */
594 unsigned big_writes:1;
596 /** Don't apply umask to creation modes */
597 unsigned dont_mask:1;
599 /** Are BSD file locking primitives not implemented by fs? */
602 /** Is fallocate not implemented by fs? */
603 unsigned no_fallocate:1;
605 /** Is rename with flags implemented by fs? */
606 unsigned no_rename2:1;
608 /** Use enhanced/automatic page cache invalidation. */
609 unsigned auto_inval_data:1;
611 /** Does the filesystem support readdirplus? */
612 unsigned do_readdirplus:1;
614 /** Does the filesystem want adaptive readdirplus? */
615 unsigned readdirplus_auto:1;
617 /** Does the filesystem support asynchronous direct-IO submission? */
618 unsigned async_dio:1;
620 /** Is lseek not implemented by fs? */
623 /** Does the filesystem support posix acls? */
624 unsigned posix_acl:1;
626 /** Check permissions based on the file mode or not? */
627 unsigned default_permissions:1;
629 /** Allow other than the mounter user to access the filesystem ? */
630 unsigned allow_other:1;
632 /** The number of requests waiting for completion */
633 atomic_t num_waiting;
635 /** Negotiated minor version */
638 /** Entry on the fuse_conn_list */
639 struct list_head entry;
641 /** Device ID from super block */
644 /** Dentries in the control filesystem */
645 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
647 /** number of dentries used in the above array */
650 /** Key for lock owner ID scrambling */
653 /** Reserved request for the DESTROY message */
654 struct fuse_req *destroy_req;
656 /** Version counter for attribute changes */
659 /** Called on final put */
660 void (*release)(struct fuse_conn *);
662 /** Super block for this connection. */
663 struct super_block *sb;
665 /** Read/write semaphore to hold when accessing sb. */
666 struct rw_semaphore killsb;
668 /** List of device instances belonging to this connection */
669 struct list_head devices;
672 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
674 return sb->s_fs_info;
677 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
679 return get_fuse_conn_super(inode->i_sb);
682 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
684 return container_of(inode, struct fuse_inode, inode);
687 static inline u64 get_node_id(struct inode *inode)
689 return get_fuse_inode(inode)->nodeid;
692 static inline void fuse_make_bad(struct inode *inode)
694 remove_inode_hash(inode);
695 set_bit(FUSE_I_BAD, &get_fuse_inode(inode)->state);
698 static inline bool fuse_is_bad(struct inode *inode)
700 return unlikely(test_bit(FUSE_I_BAD, &get_fuse_inode(inode)->state));
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);
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);
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 unsigned 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[];
996 struct posix_acl *fuse_get_acl(struct inode *inode, int type);
997 int fuse_set_acl(struct inode *inode, struct posix_acl *acl, int type);
999 #endif /* _FS_FUSE_I_H */