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 /** Default max number of pages that can be used in a single read request */
32 #define FUSE_DEFAULT_MAX_PAGES_PER_REQ 32
34 /** Maximum of max_pages received in init_out */
35 #define FUSE_MAX_MAX_PAGES 256
37 /** Bias for fi->writectr, meaning new writepages must not be sent */
38 #define FUSE_NOWRITE INT_MIN
40 /** It could be as large as PATH_MAX, but would that have any uses? */
41 #define FUSE_NAME_MAX 1024
43 /** Number of dentries for each connection in the control filesystem */
44 #define FUSE_CTL_NUM_DENTRIES 5
46 /** Number of page pointers embedded in fuse_req */
47 #define FUSE_REQ_INLINE_PAGES 1
49 /** List of active connections */
50 extern struct list_head fuse_conn_list;
52 /** Global mutex protecting fuse_conn_list and the control filesystem */
53 extern struct mutex fuse_mutex;
55 /** Module parameters */
56 extern unsigned max_user_bgreq;
57 extern unsigned max_user_congthresh;
59 /* One forget request */
60 struct fuse_forget_link {
61 struct fuse_forget_one forget_one;
62 struct fuse_forget_link *next;
70 /** Unique ID, which identifies the inode between userspace
74 /** Number of lookups on this inode */
77 /** The request used for sending the FORGET message */
78 struct fuse_forget_link *forget;
80 /** Time in jiffies until the file attributes are valid */
83 /* Which attributes are invalid */
86 /** The sticky bit in inode->i_mode may have been removed, so
87 preserve the original mode */
90 /** 64 bit inode number */
93 /** Version of last attribute change */
97 /* Write related fields (regular file only) */
99 /* Files usable in writepage. Protected by fc->lock */
100 struct list_head write_files;
102 /* Writepages pending on truncate or fsync */
103 struct list_head queued_writes;
105 /* Number of sent writes, a negative bias
106 * (FUSE_NOWRITE) means more writes are blocked */
109 /* Waitq for writepage completion */
110 wait_queue_head_t page_waitq;
112 /* List of writepage requestst (pending or sent) */
113 struct list_head writepages;
116 /* readdir cache (directory only) */
118 /* true if fully cached */
124 /* position at end of cache (position of next entry) */
127 /* version of the cache */
130 /* modification time of directory when cache was
132 struct timespec64 mtime;
134 /* iversion of directory when cache was started */
137 /* protects above fields */
142 /** Miscellaneous bits describing inode state */
145 /** Lock for serializing lookup and readdir for back compatibility*/
149 /** FUSE inode state bits */
151 /** Advise readdirplus */
152 FUSE_I_ADVISE_RDPLUS,
153 /** Initialized with readdirplus */
155 /** An operation changing file size is in progress */
156 FUSE_I_SIZE_UNSTABLE,
161 /** FUSE specific file data */
163 /** Fuse connection for this file */
164 struct fuse_conn *fc;
166 /** Request reserved for flush and release */
167 struct fuse_req *reserved_req;
169 /** Kernel file handle guaranteed to be unique */
172 /** File handle used by userspace */
175 /** Node id of this file */
181 /** FOPEN_* flags returned by open */
184 /** Entry on inode's write_files list */
185 struct list_head write_entry;
187 /* Readdir related */
190 * Protects below fields against (crazy) parallel readdir on
191 * same open file. Uncontended in the normal case.
195 /* Dir stream position */
198 /* Offset in cache */
201 /* Version of cache we are reading */
206 /** RB node to be linked on fuse_conn->polled_files */
207 struct rb_node polled_node;
209 /** Wait queue head for poll */
210 wait_queue_head_t poll_wait;
212 /** Has flock been performed on this file? */
216 /** One input argument of a request */
222 /** The request input */
224 /** The request header */
225 struct fuse_in_header h;
227 /** True if the data for the last argument is in req->pages */
230 /** Number of arguments */
233 /** Array of arguments */
234 struct fuse_in_arg args[3];
237 /** One output argument of a request */
243 /** The request output */
245 /** Header returned from userspace */
246 struct fuse_out_header h;
249 * The following bitfields are not changed during the request
253 /** Last argument is variable length (can be shorter than
257 /** Last argument is a list of pages to copy data to */
260 /** Zero partially or not copied pages */
261 unsigned page_zeroing:1;
263 /** Pages may be replaced with new ones */
264 unsigned page_replace:1;
266 /** Number or arguments */
269 /** Array of arguments */
270 struct fuse_arg args[2];
273 /** FUSE page descriptor */
274 struct fuse_page_desc {
286 struct fuse_in_arg args[3];
292 struct fuse_arg args[2];
296 #define FUSE_ARGS(args) struct fuse_args args = {}
298 /** The request IO state (for asynchronous processing) */
299 struct fuse_io_priv {
311 struct completion *done;
315 #define FUSE_IO_PRIV_SYNC(i) \
317 .refcnt = KREF_INIT(1), \
325 * FR_ISREPLY: set if the request has reply
326 * FR_FORCE: force sending of the request even if interrupted
327 * FR_BACKGROUND: request is sent in the background
328 * FR_WAITING: request is counted as "waiting"
329 * FR_ABORTED: the request was aborted
330 * FR_INTERRUPTED: the request has been interrupted
331 * FR_LOCKED: data is being copied to/from the request
332 * FR_PENDING: request is not yet in userspace
333 * FR_SENT: request is in userspace, waiting for an answer
334 * FR_FINISHED: request is finished
335 * FR_PRIVATE: request is on private list
352 * A request to the client
354 * .waitq.lock protects the following fields:
356 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
359 /** This can be on either pending processing or io lists in
361 struct list_head list;
363 /** Entry on the interrupts list */
364 struct list_head intr_entry;
369 /* Request flags, updated with test/set/clear_bit() */
372 /** The request input */
375 /** The request output */
378 /** Used to wake up the task waiting for completion of request*/
379 wait_queue_head_t waitq;
381 /** Data for asynchronous requests */
384 struct fuse_release_in in;
387 struct fuse_init_in init_in;
388 struct fuse_init_out init_out;
389 struct cuse_init_in cuse_init_in;
391 struct fuse_read_in in;
395 struct fuse_write_in in;
396 struct fuse_write_out out;
397 struct fuse_req *next;
399 struct fuse_notify_retrieve_in retrieve_in;
405 /** page-descriptor vector */
406 struct fuse_page_desc *page_descs;
408 /** size of the 'pages' array */
411 /** inline page vector */
412 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
414 /** inline page-descriptor vector */
415 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
417 /** number of pages in vector */
420 /** File used in the request (or NULL) */
421 struct fuse_file *ff;
423 /** Inode used in the request or NULL */
426 /** AIO control block */
427 struct fuse_io_priv *io;
429 /** Link on fi->writepages */
430 struct list_head writepages_entry;
432 /** Request completion callback */
433 void (*end)(struct fuse_conn *, struct fuse_req *);
435 /** Request is stolen from fuse_file->reserved_req */
436 struct file *stolen_file;
440 /** Connection established */
443 /** Readers of the connection are waiting on this */
444 wait_queue_head_t waitq;
446 /** The next unique request id */
449 /** The list of pending requests */
450 struct list_head pending;
452 /** Pending interrupts */
453 struct list_head interrupts;
455 /** Queue of pending forgets */
456 struct fuse_forget_link forget_list_head;
457 struct fuse_forget_link *forget_list_tail;
459 /** Batching of FORGET requests (positive indicates FORGET batch) */
462 /** O_ASYNC requests */
463 struct fasync_struct *fasync;
466 #define FUSE_PQ_HASH_BITS 8
467 #define FUSE_PQ_HASH_SIZE (1 << FUSE_PQ_HASH_BITS)
470 /** Connection established */
473 /** Lock protecting accessess to members of this structure */
476 /** Hash table of requests being processed */
477 struct list_head *processing;
479 /** The list of requests under I/O */
484 * Fuse device instance
487 /** Fuse connection for this device */
488 struct fuse_conn *fc;
490 /** Processing queue */
491 struct fuse_pqueue pq;
493 /** list entry on fc->devices */
494 struct list_head entry;
500 * This structure is created, when the filesystem is mounted, and is
501 * destroyed, when the client device is closed and the filesystem is
505 /** Lock protecting accessess to members of this structure */
511 /** Number of fuse_dev's */
516 /** The user id for this mount */
519 /** The group id for this mount */
522 /** The pid namespace for this mount */
523 struct pid_namespace *pid_ns;
525 /** The user namespace for this mount */
526 struct user_namespace *user_ns;
528 /** Maximum read size */
531 /** Maximum write size */
534 /** Maxmum number of pages that can be used in a single request */
535 unsigned int max_pages;
538 struct fuse_iqueue iq;
540 /** The next unique kernel file handle */
543 /** rbtree of fuse_files waiting for poll events indexed by ph */
544 struct rb_root polled_files;
546 /** Maximum number of outstanding background requests */
547 unsigned max_background;
549 /** Number of background requests at which congestion starts */
550 unsigned congestion_threshold;
552 /** Number of requests currently in the background */
553 unsigned num_background;
555 /** Number of background requests currently queued for userspace */
556 unsigned active_background;
558 /** The list of background requests set aside for later queuing */
559 struct list_head bg_queue;
561 /** Protects: max_background, congestion_threshold, num_background,
562 * active_background, bg_queue, blocked */
565 /** Flag indicating that INIT reply has been received. Allocating
566 * any fuse request will be suspended until the flag is set */
569 /** Flag indicating if connection is blocked. This will be
570 the case before the INIT reply is received, and if there
571 are too many outstading backgrounds requests */
574 /** waitq for blocked connection */
575 wait_queue_head_t blocked_waitq;
577 /** waitq for reserved requests */
578 wait_queue_head_t reserved_req_waitq;
580 /** Connection established, cleared on umount, connection
581 abort and device release */
584 /** Connection aborted via sysfs */
587 /** Connection failed (version mismatch). Cannot race with
588 setting other bitfields since it is only set once in INIT
589 reply, before any other request, and never cleared */
590 unsigned conn_error:1;
592 /** Connection successful. Only set in INIT */
593 unsigned conn_init:1;
595 /** Do readpages asynchronously? Only set in INIT */
596 unsigned async_read:1;
598 /** Return an unique read error after abort. Only set in INIT */
599 unsigned abort_err:1;
601 /** Do not send separate SETATTR request before open(O_TRUNC) */
602 unsigned atomic_o_trunc:1;
604 /** Filesystem supports NFS exporting. Only set in INIT */
605 unsigned export_support:1;
607 /** write-back cache policy (default is write-through) */
608 unsigned writeback_cache:1;
610 /** allow parallel lookups and readdir (default is serialized) */
611 unsigned parallel_dirops:1;
613 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
614 unsigned handle_killpriv:1;
616 /** cache READLINK responses in page cache */
617 unsigned cache_symlinks:1;
620 * The following bitfields are only for optimization purposes
621 * and hence races in setting them will not cause malfunction
624 /** Is open/release not implemented by fs? */
627 /** Is fsync not implemented by fs? */
630 /** Is fsyncdir not implemented by fs? */
631 unsigned no_fsyncdir:1;
633 /** Is flush not implemented by fs? */
636 /** Is setxattr not implemented by fs? */
637 unsigned no_setxattr:1;
639 /** Is getxattr not implemented by fs? */
640 unsigned no_getxattr:1;
642 /** Is listxattr not implemented by fs? */
643 unsigned no_listxattr:1;
645 /** Is removexattr not implemented by fs? */
646 unsigned no_removexattr:1;
648 /** Are posix file locking primitives not implemented by fs? */
651 /** Is access not implemented by fs? */
652 unsigned no_access:1;
654 /** Is create not implemented by fs? */
655 unsigned no_create:1;
657 /** Is interrupt not implemented by fs? */
658 unsigned no_interrupt:1;
660 /** Is bmap not implemented by fs? */
663 /** Is poll not implemented by fs? */
666 /** Do multi-page cached writes */
667 unsigned big_writes:1;
669 /** Don't apply umask to creation modes */
670 unsigned dont_mask:1;
672 /** Are BSD file locking primitives not implemented by fs? */
675 /** Is fallocate not implemented by fs? */
676 unsigned no_fallocate:1;
678 /** Is rename with flags implemented by fs? */
679 unsigned no_rename2:1;
681 /** Use enhanced/automatic page cache invalidation. */
682 unsigned auto_inval_data:1;
684 /** Does the filesystem support readdirplus? */
685 unsigned do_readdirplus:1;
687 /** Does the filesystem want adaptive readdirplus? */
688 unsigned readdirplus_auto:1;
690 /** Does the filesystem support asynchronous direct-IO submission? */
691 unsigned async_dio:1;
693 /** Is lseek not implemented by fs? */
696 /** Does the filesystem support posix acls? */
697 unsigned posix_acl:1;
699 /** Check permissions based on the file mode or not? */
700 unsigned default_permissions:1;
702 /** Allow other than the mounter user to access the filesystem ? */
703 unsigned allow_other:1;
705 /** Does the filesystem support copy_file_range? */
706 unsigned no_copy_file_range:1;
708 /** The number of requests waiting for completion */
709 atomic_t num_waiting;
711 /** Negotiated minor version */
714 /** Entry on the fuse_conn_list */
715 struct list_head entry;
717 /** Device ID from super block */
720 /** Dentries in the control filesystem */
721 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
723 /** number of dentries used in the above array */
726 /** Key for lock owner ID scrambling */
729 /** Reserved request for the DESTROY message */
730 struct fuse_req *destroy_req;
732 /** Version counter for attribute changes */
735 /** Called on final put */
736 void (*release)(struct fuse_conn *);
738 /** Super block for this connection. */
739 struct super_block *sb;
741 /** Read/write semaphore to hold when accessing sb. */
742 struct rw_semaphore killsb;
744 /** List of device instances belonging to this connection */
745 struct list_head devices;
748 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
750 return sb->s_fs_info;
753 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
755 return get_fuse_conn_super(inode->i_sb);
758 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
760 return container_of(inode, struct fuse_inode, inode);
763 static inline u64 get_node_id(struct inode *inode)
765 return get_fuse_inode(inode)->nodeid;
768 static inline int invalid_nodeid(u64 nodeid)
770 return !nodeid || nodeid == FUSE_ROOT_ID;
773 /** Device operations */
774 extern const struct file_operations fuse_dev_operations;
776 extern const struct dentry_operations fuse_dentry_operations;
777 extern const struct dentry_operations fuse_root_dentry_operations;
780 * Inode to nodeid comparison.
782 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
785 * Get a filled in inode
787 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
788 int generation, struct fuse_attr *attr,
789 u64 attr_valid, u64 attr_version);
791 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
792 struct fuse_entry_out *outarg, struct inode **inode);
795 * Send FORGET command
797 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
798 u64 nodeid, u64 nlookup);
800 struct fuse_forget_link *fuse_alloc_forget(void);
802 /* Used by READDIRPLUS */
803 void fuse_force_forget(struct file *file, u64 nodeid);
806 * Initialize READ or READDIR request
808 void fuse_read_fill(struct fuse_req *req, struct file *file,
809 loff_t pos, size_t count, int opcode);
812 * Send OPEN or OPENDIR request
814 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
816 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
817 void fuse_file_free(struct fuse_file *ff);
818 void fuse_finish_open(struct inode *inode, struct file *file);
820 void fuse_sync_release(struct fuse_file *ff, int flags);
823 * Send RELEASE or RELEASEDIR request
825 void fuse_release_common(struct file *file, int opcode);
828 * Send FSYNC or FSYNCDIR request
830 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
831 int datasync, int isdir);
836 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
837 struct fuse_notify_poll_wakeup_out *outarg);
840 * Initialize file operations on a regular file
842 void fuse_init_file_inode(struct inode *inode);
845 * Initialize inode operations on regular files and special files
847 void fuse_init_common(struct inode *inode);
850 * Initialize inode and file operations on a directory
852 void fuse_init_dir(struct inode *inode);
855 * Initialize inode operations on a symlink
857 void fuse_init_symlink(struct inode *inode);
860 * Change attributes of an inode
862 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
863 u64 attr_valid, u64 attr_version);
865 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
869 * Initialize the client device
871 int fuse_dev_init(void);
874 * Cleanup the client device
876 void fuse_dev_cleanup(void);
878 int fuse_ctl_init(void);
879 void __exit fuse_ctl_cleanup(void);
884 struct fuse_req *fuse_request_alloc(unsigned npages);
886 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
888 bool fuse_req_realloc_pages(struct fuse_conn *fc, struct fuse_req *req,
895 void fuse_request_free(struct fuse_req *req);
898 * Get a request, may fail with -ENOMEM,
899 * caller should specify # elements in req->pages[] explicitly
901 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
902 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
906 * Increment reference count on request
908 void __fuse_get_request(struct fuse_req *req);
911 * Gets a requests for a file operation, always succeeds
913 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
917 * Decrement reference count of a request. If count goes to zero free
920 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
923 * Send a request (synchronous)
925 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
928 * Simple request sending that does request allocation and freeing
930 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
933 * Send a request in the background
935 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
936 bool fuse_request_queue_background(struct fuse_conn *fc, struct fuse_req *req);
938 /* Abort all requests */
939 void fuse_abort_conn(struct fuse_conn *fc, bool is_abort);
940 void fuse_wait_aborted(struct fuse_conn *fc);
943 * Invalidate inode attributes
945 void fuse_invalidate_attr(struct inode *inode);
947 void fuse_invalidate_entry_cache(struct dentry *entry);
949 void fuse_invalidate_atime(struct inode *inode);
951 u64 entry_attr_timeout(struct fuse_entry_out *o);
952 void fuse_change_entry_timeout(struct dentry *entry, struct fuse_entry_out *o);
955 * Acquire reference to fuse_conn
957 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
960 * Initialize fuse_conn
962 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns);
965 * Release reference to fuse_conn
967 void fuse_conn_put(struct fuse_conn *fc);
969 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc);
970 void fuse_dev_free(struct fuse_dev *fud);
973 * Add connection to control filesystem
975 int fuse_ctl_add_conn(struct fuse_conn *fc);
978 * Remove connection from control filesystem
980 void fuse_ctl_remove_conn(struct fuse_conn *fc);
983 * Is file type valid?
985 int fuse_valid_type(int m);
988 * Is current process allowed to perform filesystem operation?
990 int fuse_allow_current_process(struct fuse_conn *fc);
992 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
994 void fuse_update_ctime(struct inode *inode);
996 int fuse_update_attributes(struct inode *inode, struct file *file);
998 void fuse_flush_writepages(struct inode *inode);
1000 void fuse_set_nowrite(struct inode *inode);
1001 void fuse_release_nowrite(struct inode *inode);
1003 u64 fuse_get_attr_version(struct fuse_conn *fc);
1006 * File-system tells the kernel to invalidate cache for the given node id.
1008 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
1009 loff_t offset, loff_t len);
1012 * File-system tells the kernel to invalidate parent attributes and
1013 * the dentry matching parent/name.
1015 * If the child_nodeid is non-zero and:
1016 * - matches the inode number for the dentry matching parent/name,
1017 * - is not a mount point
1018 * - is a file or oan empty directory
1019 * then the dentry is unhashed (d_delete()).
1021 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
1022 u64 child_nodeid, struct qstr *name);
1024 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
1028 * fuse_direct_io() flags
1031 /** If set, it is WRITE; otherwise - READ */
1032 #define FUSE_DIO_WRITE (1 << 0)
1034 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
1035 #define FUSE_DIO_CUSE (1 << 1)
1037 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1038 loff_t *ppos, int flags);
1039 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1040 unsigned int flags);
1041 long fuse_ioctl_common(struct file *file, unsigned int cmd,
1042 unsigned long arg, unsigned int flags);
1043 __poll_t fuse_file_poll(struct file *file, poll_table *wait);
1044 int fuse_dev_release(struct inode *inode, struct file *file);
1046 bool fuse_write_update_size(struct inode *inode, loff_t pos);
1048 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
1049 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
1051 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1054 void fuse_set_initialized(struct fuse_conn *fc);
1056 void fuse_unlock_inode(struct inode *inode, bool locked);
1057 bool fuse_lock_inode(struct inode *inode);
1059 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
1060 size_t size, int flags);
1061 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
1063 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
1064 int fuse_removexattr(struct inode *inode, const char *name);
1065 extern const struct xattr_handler *fuse_xattr_handlers[];
1066 extern const struct xattr_handler *fuse_acl_xattr_handlers[];
1067 extern const struct xattr_handler *fuse_no_acl_xattr_handlers[];
1070 struct posix_acl *fuse_get_acl(struct inode *inode, int type);
1071 int fuse_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1075 int fuse_readdir(struct file *file, struct dir_context *ctx);
1077 #endif /* _FS_FUSE_I_H */