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,
127 /** FUSE specific file data */
129 /** Fuse connection for this file */
130 struct fuse_conn *fc;
132 /** Request reserved for flush and release */
133 struct fuse_req *reserved_req;
135 /** Kernel file handle guaranteed to be unique */
138 /** File handle used by userspace */
141 /** Node id of this file */
147 /** FOPEN_* flags returned by open */
150 /** Entry on inode's write_files list */
151 struct list_head write_entry;
153 /** RB node to be linked on fuse_conn->polled_files */
154 struct rb_node polled_node;
156 /** Wait queue head for poll */
157 wait_queue_head_t poll_wait;
159 /** Has flock been performed on this file? */
163 /** One input argument of a request */
169 /** The request input */
171 /** The request header */
172 struct fuse_in_header h;
174 /** True if the data for the last argument is in req->pages */
177 /** Number of arguments */
180 /** Array of arguments */
181 struct fuse_in_arg args[3];
184 /** One output argument of a request */
190 /** The request output */
192 /** Header returned from userspace */
193 struct fuse_out_header h;
196 * The following bitfields are not changed during the request
200 /** Last argument is variable length (can be shorter than
204 /** Last argument is a list of pages to copy data to */
207 /** Zero partially or not copied pages */
208 unsigned page_zeroing:1;
210 /** Pages may be replaced with new ones */
211 unsigned page_replace:1;
213 /** Number or arguments */
216 /** Array of arguments */
217 struct fuse_arg args[2];
220 /** FUSE page descriptor */
221 struct fuse_page_desc {
233 struct fuse_in_arg args[3];
239 struct fuse_arg args[2];
243 #define FUSE_ARGS(args) struct fuse_args args = {}
245 /** The request IO state (for asynchronous processing) */
246 struct fuse_io_priv {
258 struct completion *done;
262 #define FUSE_IO_PRIV_SYNC(i) \
264 .refcnt = KREF_INIT(1), \
272 * FR_ISREPLY: set if the request has reply
273 * FR_FORCE: force sending of the request even if interrupted
274 * FR_BACKGROUND: request is sent in the background
275 * FR_WAITING: request is counted as "waiting"
276 * FR_ABORTED: the request was aborted
277 * FR_INTERRUPTED: the request has been interrupted
278 * FR_LOCKED: data is being copied to/from the request
279 * FR_PENDING: request is not yet in userspace
280 * FR_SENT: request is in userspace, waiting for an answer
281 * FR_FINISHED: request is finished
282 * FR_PRIVATE: request is on private list
299 * A request to the client
301 * .waitq.lock protects the following fields:
303 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
306 /** This can be on either pending processing or io lists in
308 struct list_head list;
310 /** Entry on the interrupts list */
311 struct list_head intr_entry;
318 /** Unique ID for the interrupt request */
321 /* Request flags, updated with test/set/clear_bit() */
324 /** The request input */
327 /** The request output */
330 /** Used to wake up the task waiting for completion of request*/
331 wait_queue_head_t waitq;
333 /** Data for asynchronous requests */
336 struct fuse_release_in in;
339 struct fuse_init_in init_in;
340 struct fuse_init_out init_out;
341 struct cuse_init_in cuse_init_in;
343 struct fuse_read_in in;
347 struct fuse_write_in in;
348 struct fuse_write_out out;
349 struct fuse_req *next;
351 struct fuse_notify_retrieve_in retrieve_in;
357 /** page-descriptor vector */
358 struct fuse_page_desc *page_descs;
360 /** size of the 'pages' array */
363 /** inline page vector */
364 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
366 /** inline page-descriptor vector */
367 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
369 /** number of pages in vector */
372 /** File used in the request (or NULL) */
373 struct fuse_file *ff;
375 /** Inode used in the request or NULL */
378 /** AIO control block */
379 struct fuse_io_priv *io;
381 /** Link on fi->writepages */
382 struct list_head writepages_entry;
384 /** Request completion callback */
385 void (*end)(struct fuse_conn *, struct fuse_req *);
387 /** Request is stolen from fuse_file->reserved_req */
388 struct file *stolen_file;
392 /** Connection established */
395 /** Lock protecting accesses to members of this structure */
398 /** Readers of the connection are waiting on this */
399 wait_queue_head_t waitq;
401 /** The next unique request id */
404 /** The list of pending requests */
405 struct list_head pending;
407 /** Pending interrupts */
408 struct list_head interrupts;
410 /** Queue of pending forgets */
411 struct fuse_forget_link forget_list_head;
412 struct fuse_forget_link *forget_list_tail;
414 /** Batching of FORGET requests (positive indicates FORGET batch) */
417 /** O_ASYNC requests */
418 struct fasync_struct *fasync;
422 /** Connection established */
425 /** Lock protecting accessess to members of this structure */
428 /** The list of requests being processed */
429 struct list_head processing;
431 /** The list of requests under I/O */
436 * Fuse device instance
439 /** Fuse connection for this device */
440 struct fuse_conn *fc;
442 /** Processing queue */
443 struct fuse_pqueue pq;
445 /** list entry on fc->devices */
446 struct list_head entry;
452 * This structure is created, when the filesystem is mounted, and is
453 * destroyed, when the client device is closed and the filesystem is
457 /** Lock protecting accessess to members of this structure */
463 /** Number of fuse_dev's */
468 /** The user id for this mount */
471 /** The group id for this mount */
474 /** The pid namespace for this mount */
475 struct pid_namespace *pid_ns;
477 /** The user namespace for this mount */
478 struct user_namespace *user_ns;
480 /** Maximum read size */
483 /** Maximum write size */
487 struct fuse_iqueue iq;
489 /** The next unique kernel file handle */
492 /** rbtree of fuse_files waiting for poll events indexed by ph */
493 struct rb_root polled_files;
495 /** Maximum number of outstanding background requests */
496 unsigned max_background;
498 /** Number of background requests at which congestion starts */
499 unsigned congestion_threshold;
501 /** Number of requests currently in the background */
502 unsigned num_background;
504 /** Number of background requests currently queued for userspace */
505 unsigned active_background;
507 /** The list of background requests set aside for later queuing */
508 struct list_head bg_queue;
510 /** Flag indicating that INIT reply has been received. Allocating
511 * any fuse request will be suspended until the flag is set */
514 /** Flag indicating if connection is blocked. This will be
515 the case before the INIT reply is received, and if there
516 are too many outstading backgrounds requests */
519 /** waitq for blocked connection */
520 wait_queue_head_t blocked_waitq;
522 /** waitq for reserved requests */
523 wait_queue_head_t reserved_req_waitq;
525 /** Connection established, cleared on umount, connection
526 abort and device release */
529 /** Connection aborted via sysfs */
532 /** Connection failed (version mismatch). Cannot race with
533 setting other bitfields since it is only set once in INIT
534 reply, before any other request, and never cleared */
535 unsigned conn_error:1;
537 /** Connection successful. Only set in INIT */
538 unsigned conn_init:1;
540 /** Do readpages asynchronously? Only set in INIT */
541 unsigned async_read:1;
543 /** Return an unique read error after abort. Only set in INIT */
544 unsigned abort_err:1;
546 /** Do not send separate SETATTR request before open(O_TRUNC) */
547 unsigned atomic_o_trunc:1;
549 /** Filesystem supports NFS exporting. Only set in INIT */
550 unsigned export_support:1;
552 /** write-back cache policy (default is write-through) */
553 unsigned writeback_cache:1;
555 /** allow parallel lookups and readdir (default is serialized) */
556 unsigned parallel_dirops:1;
558 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
559 unsigned handle_killpriv:1;
562 * The following bitfields are only for optimization purposes
563 * and hence races in setting them will not cause malfunction
566 /** Is open/release not implemented by fs? */
569 /** Is fsync not implemented by fs? */
572 /** Is fsyncdir not implemented by fs? */
573 unsigned no_fsyncdir:1;
575 /** Is flush not implemented by fs? */
578 /** Is setxattr not implemented by fs? */
579 unsigned no_setxattr:1;
581 /** Is getxattr not implemented by fs? */
582 unsigned no_getxattr:1;
584 /** Is listxattr not implemented by fs? */
585 unsigned no_listxattr:1;
587 /** Is removexattr not implemented by fs? */
588 unsigned no_removexattr:1;
590 /** Are posix file locking primitives not implemented by fs? */
593 /** Is access not implemented by fs? */
594 unsigned no_access:1;
596 /** Is create not implemented by fs? */
597 unsigned no_create:1;
599 /** Is interrupt not implemented by fs? */
600 unsigned no_interrupt:1;
602 /** Is bmap not implemented by fs? */
605 /** Is poll not implemented by fs? */
608 /** Do multi-page cached writes */
609 unsigned big_writes:1;
611 /** Don't apply umask to creation modes */
612 unsigned dont_mask:1;
614 /** Are BSD file locking primitives not implemented by fs? */
617 /** Is fallocate not implemented by fs? */
618 unsigned no_fallocate:1;
620 /** Is rename with flags implemented by fs? */
621 unsigned no_rename2:1;
623 /** Use enhanced/automatic page cache invalidation. */
624 unsigned auto_inval_data:1;
626 /** Does the filesystem support readdirplus? */
627 unsigned do_readdirplus:1;
629 /** Does the filesystem want adaptive readdirplus? */
630 unsigned readdirplus_auto:1;
632 /** Does the filesystem support asynchronous direct-IO submission? */
633 unsigned async_dio:1;
635 /** Is lseek not implemented by fs? */
638 /** Does the filesystem support posix acls? */
639 unsigned posix_acl:1;
641 /** Check permissions based on the file mode or not? */
642 unsigned default_permissions:1;
644 /** Allow other than the mounter user to access the filesystem ? */
645 unsigned allow_other:1;
647 /** The number of requests waiting for completion */
648 atomic_t num_waiting;
650 /** Negotiated minor version */
653 /** Entry on the fuse_conn_list */
654 struct list_head entry;
656 /** Device ID from super block */
659 /** Dentries in the control filesystem */
660 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
662 /** number of dentries used in the above array */
665 /** Key for lock owner ID scrambling */
668 /** Reserved request for the DESTROY message */
669 struct fuse_req *destroy_req;
671 /** Version counter for attribute changes */
674 /** Called on final put */
675 void (*release)(struct fuse_conn *);
677 /** Super block for this connection. */
678 struct super_block *sb;
680 /** Read/write semaphore to hold when accessing sb. */
681 struct rw_semaphore killsb;
683 /** List of device instances belonging to this connection */
684 struct list_head devices;
687 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
689 return sb->s_fs_info;
692 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
694 return get_fuse_conn_super(inode->i_sb);
697 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
699 return container_of(inode, struct fuse_inode, inode);
702 static inline u64 get_node_id(struct inode *inode)
704 return get_fuse_inode(inode)->nodeid;
707 static inline void fuse_make_bad(struct inode *inode)
709 remove_inode_hash(inode);
710 set_bit(FUSE_I_BAD, &get_fuse_inode(inode)->state);
713 static inline bool fuse_is_bad(struct inode *inode)
715 return unlikely(test_bit(FUSE_I_BAD, &get_fuse_inode(inode)->state));
718 /** Device operations */
719 extern const struct file_operations fuse_dev_operations;
721 extern const struct dentry_operations fuse_dentry_operations;
722 extern const struct dentry_operations fuse_root_dentry_operations;
725 * Inode to nodeid comparison.
727 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
730 * Get a filled in inode
732 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
733 int generation, struct fuse_attr *attr,
734 u64 attr_valid, u64 attr_version);
736 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
737 struct fuse_entry_out *outarg, struct inode **inode);
740 * Send FORGET command
742 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
743 u64 nodeid, u64 nlookup);
745 struct fuse_forget_link *fuse_alloc_forget(void);
747 /* Used by READDIRPLUS */
748 void fuse_force_forget(struct file *file, u64 nodeid);
751 * Initialize READ or READDIR request
753 void fuse_read_fill(struct fuse_req *req, struct file *file,
754 loff_t pos, size_t count, int opcode);
757 * Send OPEN or OPENDIR request
759 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
761 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
762 void fuse_file_free(struct fuse_file *ff);
763 void fuse_finish_open(struct inode *inode, struct file *file);
765 void fuse_sync_release(struct fuse_file *ff, int flags);
768 * Send RELEASE or RELEASEDIR request
770 void fuse_release_common(struct file *file, bool isdir);
773 * Send FSYNC or FSYNCDIR request
775 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
776 int datasync, int isdir);
781 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
782 struct fuse_notify_poll_wakeup_out *outarg);
785 * Initialize file operations on a regular file
787 void fuse_init_file_inode(struct inode *inode);
790 * Initialize inode operations on regular files and special files
792 void fuse_init_common(struct inode *inode);
795 * Initialize inode and file operations on a directory
797 void fuse_init_dir(struct inode *inode);
800 * Initialize inode operations on a symlink
802 void fuse_init_symlink(struct inode *inode);
805 * Change attributes of an inode
807 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
808 u64 attr_valid, u64 attr_version);
810 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
814 * Initialize the client device
816 int fuse_dev_init(void);
819 * Cleanup the client device
821 void fuse_dev_cleanup(void);
823 int fuse_ctl_init(void);
824 void __exit fuse_ctl_cleanup(void);
829 struct fuse_req *fuse_request_alloc(unsigned npages);
831 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
836 void fuse_request_free(struct fuse_req *req);
839 * Get a request, may fail with -ENOMEM,
840 * caller should specify # elements in req->pages[] explicitly
842 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
843 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
847 * Increment reference count on request
849 void __fuse_get_request(struct fuse_req *req);
852 * Gets a requests for a file operation, always succeeds
854 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
858 * Decrement reference count of a request. If count goes to zero free
861 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
864 * Send a request (synchronous)
866 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
869 * Simple request sending that does request allocation and freeing
871 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
874 * Send a request in the background
876 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
878 void fuse_request_send_background_locked(struct fuse_conn *fc,
879 struct fuse_req *req);
881 /* Abort all requests */
882 void fuse_abort_conn(struct fuse_conn *fc, bool is_abort);
883 void fuse_wait_aborted(struct fuse_conn *fc);
886 * Invalidate inode attributes
888 void fuse_invalidate_attr(struct inode *inode);
890 void fuse_invalidate_entry_cache(struct dentry *entry);
892 void fuse_invalidate_atime(struct inode *inode);
895 * Acquire reference to fuse_conn
897 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
900 * Initialize fuse_conn
902 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns);
905 * Release reference to fuse_conn
907 void fuse_conn_put(struct fuse_conn *fc);
909 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc);
910 void fuse_dev_free(struct fuse_dev *fud);
913 * Add connection to control filesystem
915 int fuse_ctl_add_conn(struct fuse_conn *fc);
918 * Remove connection from control filesystem
920 void fuse_ctl_remove_conn(struct fuse_conn *fc);
923 * Is file type valid?
925 int fuse_valid_type(int m);
927 bool fuse_invalid_attr(struct fuse_attr *attr);
930 * Is current process allowed to perform filesystem operation?
932 int fuse_allow_current_process(struct fuse_conn *fc);
934 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
936 void fuse_update_ctime(struct inode *inode);
938 int fuse_update_attributes(struct inode *inode, struct file *file);
940 void fuse_flush_writepages(struct inode *inode);
942 void fuse_set_nowrite(struct inode *inode);
943 void fuse_release_nowrite(struct inode *inode);
945 u64 fuse_get_attr_version(struct fuse_conn *fc);
948 * File-system tells the kernel to invalidate cache for the given node id.
950 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
951 loff_t offset, loff_t len);
954 * File-system tells the kernel to invalidate parent attributes and
955 * the dentry matching parent/name.
957 * If the child_nodeid is non-zero and:
958 * - matches the inode number for the dentry matching parent/name,
959 * - is not a mount point
960 * - is a file or oan empty directory
961 * then the dentry is unhashed (d_delete()).
963 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
964 u64 child_nodeid, struct qstr *name);
966 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
970 * fuse_direct_io() flags
973 /** If set, it is WRITE; otherwise - READ */
974 #define FUSE_DIO_WRITE (1 << 0)
976 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
977 #define FUSE_DIO_CUSE (1 << 1)
979 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
980 loff_t *ppos, int flags);
981 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
983 long fuse_ioctl_common(struct file *file, unsigned int cmd,
984 unsigned long arg, unsigned int flags);
985 __poll_t fuse_file_poll(struct file *file, poll_table *wait);
986 int fuse_dev_release(struct inode *inode, struct file *file);
988 bool fuse_write_update_size(struct inode *inode, loff_t pos);
990 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
991 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
993 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
996 void fuse_set_initialized(struct fuse_conn *fc);
998 void fuse_unlock_inode(struct inode *inode, bool locked);
999 bool fuse_lock_inode(struct inode *inode);
1001 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
1002 size_t size, int flags);
1003 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
1005 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
1006 int fuse_removexattr(struct inode *inode, const char *name);
1007 extern const struct xattr_handler *fuse_xattr_handlers[];
1008 extern const struct xattr_handler *fuse_acl_xattr_handlers[];
1009 extern const struct xattr_handler *fuse_no_acl_xattr_handlers[];
1012 struct posix_acl *fuse_get_acl(struct inode *inode, int type);
1013 int fuse_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1015 #endif /* _FS_FUSE_I_H */