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 */
91 /* Write related fields (regular file only) */
93 /* Files usable in writepage. Protected by fc->lock */
94 struct list_head write_files;
96 /* Writepages pending on truncate or fsync */
97 struct list_head queued_writes;
99 /* Number of sent writes, a negative bias
100 * (FUSE_NOWRITE) means more writes are blocked */
103 /* Waitq for writepage completion */
104 wait_queue_head_t page_waitq;
106 /* List of writepage requestst (pending or sent) */
107 struct list_head writepages;
110 /* readdir cache (directory only) */
112 /* true if fully cached */
118 /* position at end of cache (position of next entry) */
121 /* version of the cache */
124 /* modification time of directory when cache was
126 struct timespec64 mtime;
128 /* iversion of directory when cache was started */
131 /* protects above fields */
136 /** Miscellaneous bits describing inode state */
139 /** Lock for serializing lookup and readdir for back compatibility*/
143 /** FUSE inode state bits */
145 /** Advise readdirplus */
146 FUSE_I_ADVISE_RDPLUS,
147 /** Initialized with readdirplus */
149 /** An operation changing file size is in progress */
150 FUSE_I_SIZE_UNSTABLE,
155 /** FUSE specific file data */
157 /** Fuse connection for this file */
158 struct fuse_conn *fc;
160 /** Request reserved for flush and release */
161 struct fuse_req *reserved_req;
163 /** Kernel file handle guaranteed to be unique */
166 /** File handle used by userspace */
169 /** Node id of this file */
175 /** FOPEN_* flags returned by open */
178 /** Entry on inode's write_files list */
179 struct list_head write_entry;
181 /* Readdir related */
184 * Protects below fields against (crazy) parallel readdir on
185 * same open file. Uncontended in the normal case.
189 /* Dir stream position */
192 /* Offset in cache */
195 /* Version of cache we are reading */
200 /** RB node to be linked on fuse_conn->polled_files */
201 struct rb_node polled_node;
203 /** Wait queue head for poll */
204 wait_queue_head_t poll_wait;
206 /** Has flock been performed on this file? */
210 /** One input argument of a request */
216 /** The request input */
218 /** The request header */
219 struct fuse_in_header h;
221 /** True if the data for the last argument is in req->pages */
224 /** Number of arguments */
227 /** Array of arguments */
228 struct fuse_in_arg args[3];
231 /** One output argument of a request */
237 /** The request output */
239 /** Header returned from userspace */
240 struct fuse_out_header h;
243 * The following bitfields are not changed during the request
247 /** Last argument is variable length (can be shorter than
251 /** Last argument is a list of pages to copy data to */
254 /** Zero partially or not copied pages */
255 unsigned page_zeroing:1;
257 /** Pages may be replaced with new ones */
258 unsigned page_replace:1;
260 /** Number or arguments */
263 /** Array of arguments */
264 struct fuse_arg args[2];
267 /** FUSE page descriptor */
268 struct fuse_page_desc {
280 struct fuse_in_arg args[3];
286 struct fuse_arg args[2];
290 #define FUSE_ARGS(args) struct fuse_args args = {}
292 /** The request IO state (for asynchronous processing) */
293 struct fuse_io_priv {
305 struct completion *done;
309 #define FUSE_IO_PRIV_SYNC(i) \
311 .refcnt = KREF_INIT(1), \
319 * FR_ISREPLY: set if the request has reply
320 * FR_FORCE: force sending of the request even if interrupted
321 * FR_BACKGROUND: request is sent in the background
322 * FR_WAITING: request is counted as "waiting"
323 * FR_ABORTED: the request was aborted
324 * FR_INTERRUPTED: the request has been interrupted
325 * FR_LOCKED: data is being copied to/from the request
326 * FR_PENDING: request is not yet in userspace
327 * FR_SENT: request is in userspace, waiting for an answer
328 * FR_FINISHED: request is finished
329 * FR_PRIVATE: request is on private list
346 * A request to the client
348 * .waitq.lock protects the following fields:
350 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
353 /** This can be on either pending processing or io lists in
355 struct list_head list;
357 /** Entry on the interrupts list */
358 struct list_head intr_entry;
363 /* Request flags, updated with test/set/clear_bit() */
366 /** The request input */
369 /** The request output */
372 /** Used to wake up the task waiting for completion of request*/
373 wait_queue_head_t waitq;
375 /** Data for asynchronous requests */
378 struct fuse_release_in in;
381 struct fuse_init_in init_in;
382 struct fuse_init_out init_out;
383 struct cuse_init_in cuse_init_in;
385 struct fuse_read_in in;
389 struct fuse_write_in in;
390 struct fuse_write_out out;
391 struct fuse_req *next;
393 struct fuse_notify_retrieve_in retrieve_in;
399 /** page-descriptor vector */
400 struct fuse_page_desc *page_descs;
402 /** size of the 'pages' array */
405 /** inline page vector */
406 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
408 /** inline page-descriptor vector */
409 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
411 /** number of pages in vector */
414 /** File used in the request (or NULL) */
415 struct fuse_file *ff;
417 /** Inode used in the request or NULL */
420 /** AIO control block */
421 struct fuse_io_priv *io;
423 /** Link on fi->writepages */
424 struct list_head writepages_entry;
426 /** Request completion callback */
427 void (*end)(struct fuse_conn *, struct fuse_req *);
429 /** Request is stolen from fuse_file->reserved_req */
430 struct file *stolen_file;
434 /** Connection established */
437 /** Readers of the connection are waiting on this */
438 wait_queue_head_t waitq;
440 /** The next unique request id */
443 /** The list of pending requests */
444 struct list_head pending;
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 /** O_ASYNC requests */
457 struct fasync_struct *fasync;
460 #define FUSE_PQ_HASH_BITS 8
461 #define FUSE_PQ_HASH_SIZE (1 << FUSE_PQ_HASH_BITS)
464 /** Connection established */
467 /** Lock protecting accessess to members of this structure */
470 /** Hash table of requests being processed */
471 struct list_head *processing;
473 /** The list of requests under I/O */
478 * Fuse device instance
481 /** Fuse connection for this device */
482 struct fuse_conn *fc;
484 /** Processing queue */
485 struct fuse_pqueue pq;
487 /** list entry on fc->devices */
488 struct list_head entry;
494 * This structure is created, when the filesystem is mounted, and is
495 * destroyed, when the client device is closed and the filesystem is
499 /** Lock protecting accessess to members of this structure */
505 /** Number of fuse_dev's */
510 /** The user id for this mount */
513 /** The group id for this mount */
516 /** The pid namespace for this mount */
517 struct pid_namespace *pid_ns;
519 /** The user namespace for this mount */
520 struct user_namespace *user_ns;
522 /** Maximum read size */
525 /** Maximum write size */
529 struct fuse_iqueue iq;
531 /** The next unique kernel file handle */
534 /** rbtree of fuse_files waiting for poll events indexed by ph */
535 struct rb_root polled_files;
537 /** Maximum number of outstanding background requests */
538 unsigned max_background;
540 /** Number of background requests at which congestion starts */
541 unsigned congestion_threshold;
543 /** Number of requests currently in the background */
544 unsigned num_background;
546 /** Number of background requests currently queued for userspace */
547 unsigned active_background;
549 /** The list of background requests set aside for later queuing */
550 struct list_head bg_queue;
552 /** Protects: max_background, congestion_threshold, num_background,
553 * active_background, bg_queue, blocked */
556 /** Flag indicating that INIT reply has been received. Allocating
557 * any fuse request will be suspended until the flag is set */
560 /** Flag indicating if connection is blocked. This will be
561 the case before the INIT reply is received, and if there
562 are too many outstading backgrounds requests */
565 /** waitq for blocked connection */
566 wait_queue_head_t blocked_waitq;
568 /** waitq for reserved requests */
569 wait_queue_head_t reserved_req_waitq;
571 /** Connection established, cleared on umount, connection
572 abort and device release */
575 /** Connection aborted via sysfs */
578 /** Connection failed (version mismatch). Cannot race with
579 setting other bitfields since it is only set once in INIT
580 reply, before any other request, and never cleared */
581 unsigned conn_error:1;
583 /** Connection successful. Only set in INIT */
584 unsigned conn_init:1;
586 /** Do readpages asynchronously? Only set in INIT */
587 unsigned async_read:1;
589 /** Return an unique read error after abort. Only set in INIT */
590 unsigned abort_err:1;
592 /** Do not send separate SETATTR request before open(O_TRUNC) */
593 unsigned atomic_o_trunc:1;
595 /** Filesystem supports NFS exporting. Only set in INIT */
596 unsigned export_support:1;
598 /** write-back cache policy (default is write-through) */
599 unsigned writeback_cache:1;
601 /** allow parallel lookups and readdir (default is serialized) */
602 unsigned parallel_dirops:1;
604 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
605 unsigned handle_killpriv:1;
608 * The following bitfields are only for optimization purposes
609 * and hence races in setting them will not cause malfunction
612 /** Is open/release not implemented by fs? */
615 /** Is fsync not implemented by fs? */
618 /** Is fsyncdir not implemented by fs? */
619 unsigned no_fsyncdir:1;
621 /** Is flush not implemented by fs? */
624 /** Is setxattr not implemented by fs? */
625 unsigned no_setxattr:1;
627 /** Is getxattr not implemented by fs? */
628 unsigned no_getxattr:1;
630 /** Is listxattr not implemented by fs? */
631 unsigned no_listxattr:1;
633 /** Is removexattr not implemented by fs? */
634 unsigned no_removexattr:1;
636 /** Are posix file locking primitives not implemented by fs? */
639 /** Is access not implemented by fs? */
640 unsigned no_access:1;
642 /** Is create not implemented by fs? */
643 unsigned no_create:1;
645 /** Is interrupt not implemented by fs? */
646 unsigned no_interrupt:1;
648 /** Is bmap not implemented by fs? */
651 /** Is poll not implemented by fs? */
654 /** Do multi-page cached writes */
655 unsigned big_writes:1;
657 /** Don't apply umask to creation modes */
658 unsigned dont_mask:1;
660 /** Are BSD file locking primitives not implemented by fs? */
663 /** Is fallocate not implemented by fs? */
664 unsigned no_fallocate:1;
666 /** Is rename with flags implemented by fs? */
667 unsigned no_rename2:1;
669 /** Use enhanced/automatic page cache invalidation. */
670 unsigned auto_inval_data:1;
672 /** Does the filesystem support readdirplus? */
673 unsigned do_readdirplus:1;
675 /** Does the filesystem want adaptive readdirplus? */
676 unsigned readdirplus_auto:1;
678 /** Does the filesystem support asynchronous direct-IO submission? */
679 unsigned async_dio:1;
681 /** Is lseek not implemented by fs? */
684 /** Does the filesystem support posix acls? */
685 unsigned posix_acl:1;
687 /** Check permissions based on the file mode or not? */
688 unsigned default_permissions:1;
690 /** Allow other than the mounter user to access the filesystem ? */
691 unsigned allow_other:1;
693 /** Does the filesystem support copy_file_range? */
694 unsigned no_copy_file_range:1;
696 /** The number of requests waiting for completion */
697 atomic_t num_waiting;
699 /** Negotiated minor version */
702 /** Entry on the fuse_conn_list */
703 struct list_head entry;
705 /** Device ID from super block */
708 /** Dentries in the control filesystem */
709 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
711 /** number of dentries used in the above array */
714 /** Key for lock owner ID scrambling */
717 /** Reserved request for the DESTROY message */
718 struct fuse_req *destroy_req;
720 /** Version counter for attribute changes */
723 /** Called on final put */
724 void (*release)(struct fuse_conn *);
726 /** Super block for this connection. */
727 struct super_block *sb;
729 /** Read/write semaphore to hold when accessing sb. */
730 struct rw_semaphore killsb;
732 /** List of device instances belonging to this connection */
733 struct list_head devices;
736 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
738 return sb->s_fs_info;
741 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
743 return get_fuse_conn_super(inode->i_sb);
746 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
748 return container_of(inode, struct fuse_inode, inode);
751 static inline u64 get_node_id(struct inode *inode)
753 return get_fuse_inode(inode)->nodeid;
756 static inline int invalid_nodeid(u64 nodeid)
758 return !nodeid || nodeid == FUSE_ROOT_ID;
761 /** Device operations */
762 extern const struct file_operations fuse_dev_operations;
764 extern const struct dentry_operations fuse_dentry_operations;
765 extern const struct dentry_operations fuse_root_dentry_operations;
768 * Inode to nodeid comparison.
770 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
773 * Get a filled in inode
775 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
776 int generation, struct fuse_attr *attr,
777 u64 attr_valid, u64 attr_version);
779 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
780 struct fuse_entry_out *outarg, struct inode **inode);
783 * Send FORGET command
785 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
786 u64 nodeid, u64 nlookup);
788 struct fuse_forget_link *fuse_alloc_forget(void);
790 /* Used by READDIRPLUS */
791 void fuse_force_forget(struct file *file, u64 nodeid);
794 * Initialize READ or READDIR request
796 void fuse_read_fill(struct fuse_req *req, struct file *file,
797 loff_t pos, size_t count, int opcode);
800 * Send OPEN or OPENDIR request
802 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
804 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
805 void fuse_file_free(struct fuse_file *ff);
806 void fuse_finish_open(struct inode *inode, struct file *file);
808 void fuse_sync_release(struct fuse_file *ff, int flags);
811 * Send RELEASE or RELEASEDIR request
813 void fuse_release_common(struct file *file, int opcode);
816 * Send FSYNC or FSYNCDIR request
818 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
819 int datasync, int isdir);
824 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
825 struct fuse_notify_poll_wakeup_out *outarg);
828 * Initialize file operations on a regular file
830 void fuse_init_file_inode(struct inode *inode);
833 * Initialize inode operations on regular files and special files
835 void fuse_init_common(struct inode *inode);
838 * Initialize inode and file operations on a directory
840 void fuse_init_dir(struct inode *inode);
843 * Initialize inode operations on a symlink
845 void fuse_init_symlink(struct inode *inode);
848 * Change attributes of an inode
850 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
851 u64 attr_valid, u64 attr_version);
853 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
857 * Initialize the client device
859 int fuse_dev_init(void);
862 * Cleanup the client device
864 void fuse_dev_cleanup(void);
866 int fuse_ctl_init(void);
867 void __exit fuse_ctl_cleanup(void);
872 struct fuse_req *fuse_request_alloc(unsigned npages);
874 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
879 void fuse_request_free(struct fuse_req *req);
882 * Get a request, may fail with -ENOMEM,
883 * caller should specify # elements in req->pages[] explicitly
885 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
886 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
890 * Increment reference count on request
892 void __fuse_get_request(struct fuse_req *req);
895 * Gets a requests for a file operation, always succeeds
897 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
901 * Decrement reference count of a request. If count goes to zero free
904 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
907 * Send a request (synchronous)
909 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
912 * Simple request sending that does request allocation and freeing
914 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
917 * Send a request in the background
919 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
920 bool fuse_request_queue_background(struct fuse_conn *fc, struct fuse_req *req);
922 /* Abort all requests */
923 void fuse_abort_conn(struct fuse_conn *fc, bool is_abort);
924 void fuse_wait_aborted(struct fuse_conn *fc);
927 * Invalidate inode attributes
929 void fuse_invalidate_attr(struct inode *inode);
931 void fuse_invalidate_entry_cache(struct dentry *entry);
933 void fuse_invalidate_atime(struct inode *inode);
935 u64 entry_attr_timeout(struct fuse_entry_out *o);
936 void fuse_change_entry_timeout(struct dentry *entry, struct fuse_entry_out *o);
939 * Acquire reference to fuse_conn
941 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
944 * Initialize fuse_conn
946 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns);
949 * Release reference to fuse_conn
951 void fuse_conn_put(struct fuse_conn *fc);
953 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc);
954 void fuse_dev_free(struct fuse_dev *fud);
957 * Add connection to control filesystem
959 int fuse_ctl_add_conn(struct fuse_conn *fc);
962 * Remove connection from control filesystem
964 void fuse_ctl_remove_conn(struct fuse_conn *fc);
967 * Is file type valid?
969 int fuse_valid_type(int m);
972 * Is current process allowed to perform filesystem operation?
974 int fuse_allow_current_process(struct fuse_conn *fc);
976 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
978 void fuse_update_ctime(struct inode *inode);
980 int fuse_update_attributes(struct inode *inode, struct file *file);
982 void fuse_flush_writepages(struct inode *inode);
984 void fuse_set_nowrite(struct inode *inode);
985 void fuse_release_nowrite(struct inode *inode);
987 u64 fuse_get_attr_version(struct fuse_conn *fc);
990 * File-system tells the kernel to invalidate cache for the given node id.
992 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
993 loff_t offset, loff_t len);
996 * File-system tells the kernel to invalidate parent attributes and
997 * the dentry matching parent/name.
999 * If the child_nodeid is non-zero and:
1000 * - matches the inode number for the dentry matching parent/name,
1001 * - is not a mount point
1002 * - is a file or oan empty directory
1003 * then the dentry is unhashed (d_delete()).
1005 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
1006 u64 child_nodeid, struct qstr *name);
1008 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
1012 * fuse_direct_io() flags
1015 /** If set, it is WRITE; otherwise - READ */
1016 #define FUSE_DIO_WRITE (1 << 0)
1018 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
1019 #define FUSE_DIO_CUSE (1 << 1)
1021 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1022 loff_t *ppos, int flags);
1023 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1024 unsigned int flags);
1025 long fuse_ioctl_common(struct file *file, unsigned int cmd,
1026 unsigned long arg, unsigned int flags);
1027 __poll_t fuse_file_poll(struct file *file, poll_table *wait);
1028 int fuse_dev_release(struct inode *inode, struct file *file);
1030 bool fuse_write_update_size(struct inode *inode, loff_t pos);
1032 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
1033 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
1035 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1038 void fuse_set_initialized(struct fuse_conn *fc);
1040 void fuse_unlock_inode(struct inode *inode, bool locked);
1041 bool fuse_lock_inode(struct inode *inode);
1043 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
1044 size_t size, int flags);
1045 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
1047 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
1048 int fuse_removexattr(struct inode *inode, const char *name);
1049 extern const struct xattr_handler *fuse_xattr_handlers[];
1050 extern const struct xattr_handler *fuse_acl_xattr_handlers[];
1051 extern const struct xattr_handler *fuse_no_acl_xattr_handlers[];
1054 struct posix_acl *fuse_get_acl(struct inode *inode, int type);
1055 int fuse_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1059 int fuse_readdir(struct file *file, struct dir_context *ctx);
1061 #endif /* _FS_FUSE_I_H */