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.
13 # define pr_fmt(fmt) "fuse: " fmt
16 #include <linux/fuse.h>
18 #include <linux/mount.h>
19 #include <linux/wait.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
23 #include <linux/backing-dev.h>
24 #include <linux/mutex.h>
25 #include <linux/rwsem.h>
26 #include <linux/rbtree.h>
27 #include <linux/poll.h>
28 #include <linux/workqueue.h>
29 #include <linux/kref.h>
30 #include <linux/xattr.h>
31 #include <linux/pid_namespace.h>
32 #include <linux/refcount.h>
33 #include <linux/user_namespace.h>
35 /** Default max number of pages that can be used in a single read request */
36 #define FUSE_DEFAULT_MAX_PAGES_PER_REQ 32
38 /** Maximum of max_pages received in init_out */
39 #define FUSE_MAX_MAX_PAGES 256
41 /** Bias for fi->writectr, meaning new writepages must not be sent */
42 #define FUSE_NOWRITE INT_MIN
44 /** It could be as large as PATH_MAX, but would that have any uses? */
45 #define FUSE_NAME_MAX 1024
47 /** Number of dentries for each connection in the control filesystem */
48 #define FUSE_CTL_NUM_DENTRIES 5
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 /* Which attributes are invalid */
87 /** The sticky bit in inode->i_mode may have been removed, so
88 preserve the original mode */
91 /** 64 bit inode number */
94 /** Version of last attribute change */
98 /* Write related fields (regular file only) */
100 /* Files usable in writepage. Protected by fi->lock */
101 struct list_head write_files;
103 /* Writepages pending on truncate or fsync */
104 struct list_head queued_writes;
106 /* Number of sent writes, a negative bias
107 * (FUSE_NOWRITE) means more writes are blocked */
110 /* Waitq for writepage completion */
111 wait_queue_head_t page_waitq;
113 /* List of writepage requestst (pending or sent) */
114 struct list_head writepages;
117 /* readdir cache (directory only) */
119 /* true if fully cached */
125 /* position at end of cache (position of next entry) */
128 /* version of the cache */
131 /* modification time of directory when cache was
133 struct timespec64 mtime;
135 /* iversion of directory when cache was started */
138 /* protects above fields */
143 /** Miscellaneous bits describing inode state */
146 /** Lock for serializing lookup and readdir for back compatibility*/
149 /** Lock to protect write related fields */
153 /** FUSE inode state bits */
155 /** Advise readdirplus */
156 FUSE_I_ADVISE_RDPLUS,
157 /** Initialized with readdirplus */
159 /** An operation changing file size is in progress */
160 FUSE_I_SIZE_UNSTABLE,
164 struct fuse_release_args;
166 /** FUSE specific file data */
168 /** Fuse connection for this file */
169 struct fuse_conn *fc;
171 /* Argument space reserved for release */
172 struct fuse_release_args *release_args;
174 /** Kernel file handle guaranteed to be unique */
177 /** File handle used by userspace */
180 /** Node id of this file */
186 /** FOPEN_* flags returned by open */
189 /** Entry on inode's write_files list */
190 struct list_head write_entry;
192 /* Readdir related */
195 * Protects below fields against (crazy) parallel readdir on
196 * same open file. Uncontended in the normal case.
200 /* Dir stream position */
203 /* Offset in cache */
206 /* Version of cache we are reading */
211 /** RB node to be linked on fuse_conn->polled_files */
212 struct rb_node polled_node;
214 /** Wait queue head for poll */
215 wait_queue_head_t poll_wait;
217 /** Has flock been performed on this file? */
221 /** One input argument of a request */
227 /** One output argument of a request */
233 /** FUSE page descriptor */
234 struct fuse_page_desc {
242 unsigned short in_numargs;
243 unsigned short out_numargs;
252 struct fuse_in_arg in_args[3];
253 struct fuse_arg out_args[2];
254 void (*end)(struct fuse_conn *fc, struct fuse_args *args, int error);
257 struct fuse_args_pages {
258 struct fuse_args args;
260 struct fuse_page_desc *descs;
261 unsigned int num_pages;
264 #define FUSE_ARGS(args) struct fuse_args args = {}
266 /** The request IO state (for asynchronous processing) */
267 struct fuse_io_priv {
279 struct completion *done;
283 #define FUSE_IO_PRIV_SYNC(i) \
285 .refcnt = KREF_INIT(1), \
293 * FR_ISREPLY: set if the request has reply
294 * FR_FORCE: force sending of the request even if interrupted
295 * FR_BACKGROUND: request is sent in the background
296 * FR_WAITING: request is counted as "waiting"
297 * FR_ABORTED: the request was aborted
298 * FR_INTERRUPTED: the request has been interrupted
299 * FR_LOCKED: data is being copied to/from the request
300 * FR_PENDING: request is not yet in userspace
301 * FR_SENT: request is in userspace, waiting for an answer
302 * FR_FINISHED: request is finished
303 * FR_PRIVATE: request is on private list
320 * A request to the client
322 * .waitq.lock protects the following fields:
324 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
327 /** This can be on either pending processing or io lists in
329 struct list_head list;
331 /** Entry on the interrupts list */
332 struct list_head intr_entry;
334 /* Input/output arguments */
335 struct fuse_args *args;
340 /* Request flags, updated with test/set/clear_bit() */
343 /* The request input header */
345 struct fuse_in_header h;
348 /* The request output header */
350 struct fuse_out_header h;
353 /** Used to wake up the task waiting for completion of request*/
354 wait_queue_head_t waitq;
359 /** Connection established */
362 /** Lock protecting accesses to members of this structure */
365 /** Readers of the connection are waiting on this */
366 wait_queue_head_t waitq;
368 /** The next unique request id */
371 /** The list of pending requests */
372 struct list_head pending;
374 /** Pending interrupts */
375 struct list_head interrupts;
377 /** Queue of pending forgets */
378 struct fuse_forget_link forget_list_head;
379 struct fuse_forget_link *forget_list_tail;
381 /** Batching of FORGET requests (positive indicates FORGET batch) */
384 /** O_ASYNC requests */
385 struct fasync_struct *fasync;
388 #define FUSE_PQ_HASH_BITS 8
389 #define FUSE_PQ_HASH_SIZE (1 << FUSE_PQ_HASH_BITS)
392 /** Connection established */
395 /** Lock protecting accessess to members of this structure */
398 /** Hash table of requests being processed */
399 struct list_head *processing;
401 /** The list of requests under I/O */
406 * Fuse device instance
409 /** Fuse connection for this device */
410 struct fuse_conn *fc;
412 /** Processing queue */
413 struct fuse_pqueue pq;
415 /** list entry on fc->devices */
416 struct list_head entry;
422 * This structure is created, when the filesystem is mounted, and is
423 * destroyed, when the client device is closed and the filesystem is
427 /** Lock protecting accessess to members of this structure */
433 /** Number of fuse_dev's */
438 /** The user id for this mount */
441 /** The group id for this mount */
444 /** The pid namespace for this mount */
445 struct pid_namespace *pid_ns;
447 /** The user namespace for this mount */
448 struct user_namespace *user_ns;
450 /** Maximum read size */
453 /** Maximum write size */
456 /** Maxmum number of pages that can be used in a single request */
457 unsigned int max_pages;
460 struct fuse_iqueue iq;
462 /** The next unique kernel file handle */
465 /** rbtree of fuse_files waiting for poll events indexed by ph */
466 struct rb_root polled_files;
468 /** Maximum number of outstanding background requests */
469 unsigned max_background;
471 /** Number of background requests at which congestion starts */
472 unsigned congestion_threshold;
474 /** Number of requests currently in the background */
475 unsigned num_background;
477 /** Number of background requests currently queued for userspace */
478 unsigned active_background;
480 /** The list of background requests set aside for later queuing */
481 struct list_head bg_queue;
483 /** Protects: max_background, congestion_threshold, num_background,
484 * active_background, bg_queue, blocked */
487 /** Flag indicating that INIT reply has been received. Allocating
488 * any fuse request will be suspended until the flag is set */
491 /** Flag indicating if connection is blocked. This will be
492 the case before the INIT reply is received, and if there
493 are too many outstading backgrounds requests */
496 /** waitq for blocked connection */
497 wait_queue_head_t blocked_waitq;
499 /** Connection established, cleared on umount, connection
500 abort and device release */
503 /** Connection aborted via sysfs */
506 /** Connection failed (version mismatch). Cannot race with
507 setting other bitfields since it is only set once in INIT
508 reply, before any other request, and never cleared */
509 unsigned conn_error:1;
511 /** Connection successful. Only set in INIT */
512 unsigned conn_init:1;
514 /** Do readpages asynchronously? Only set in INIT */
515 unsigned async_read:1;
517 /** Return an unique read error after abort. Only set in INIT */
518 unsigned abort_err:1;
520 /** Do not send separate SETATTR request before open(O_TRUNC) */
521 unsigned atomic_o_trunc:1;
523 /** Filesystem supports NFS exporting. Only set in INIT */
524 unsigned export_support:1;
526 /** write-back cache policy (default is write-through) */
527 unsigned writeback_cache:1;
529 /** allow parallel lookups and readdir (default is serialized) */
530 unsigned parallel_dirops:1;
532 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
533 unsigned handle_killpriv:1;
535 /** cache READLINK responses in page cache */
536 unsigned cache_symlinks:1;
539 * The following bitfields are only for optimization purposes
540 * and hence races in setting them will not cause malfunction
543 /** Is open/release not implemented by fs? */
546 /** Is opendir/releasedir not implemented by fs? */
547 unsigned no_opendir:1;
549 /** Is fsync not implemented by fs? */
552 /** Is fsyncdir not implemented by fs? */
553 unsigned no_fsyncdir:1;
555 /** Is flush not implemented by fs? */
558 /** Is setxattr not implemented by fs? */
559 unsigned no_setxattr:1;
561 /** Is getxattr not implemented by fs? */
562 unsigned no_getxattr:1;
564 /** Is listxattr not implemented by fs? */
565 unsigned no_listxattr:1;
567 /** Is removexattr not implemented by fs? */
568 unsigned no_removexattr:1;
570 /** Are posix file locking primitives not implemented by fs? */
573 /** Is access not implemented by fs? */
574 unsigned no_access:1;
576 /** Is create not implemented by fs? */
577 unsigned no_create:1;
579 /** Is interrupt not implemented by fs? */
580 unsigned no_interrupt:1;
582 /** Is bmap not implemented by fs? */
585 /** Is poll not implemented by fs? */
588 /** Do multi-page cached writes */
589 unsigned big_writes:1;
591 /** Don't apply umask to creation modes */
592 unsigned dont_mask:1;
594 /** Are BSD file locking primitives not implemented by fs? */
597 /** Is fallocate not implemented by fs? */
598 unsigned no_fallocate:1;
600 /** Is rename with flags implemented by fs? */
601 unsigned no_rename2:1;
603 /** Use enhanced/automatic page cache invalidation. */
604 unsigned auto_inval_data:1;
606 /** Filesystem is fully reponsible for page cache invalidation. */
607 unsigned explicit_inval_data:1;
609 /** Does the filesystem support readdirplus? */
610 unsigned do_readdirplus:1;
612 /** Does the filesystem want adaptive readdirplus? */
613 unsigned readdirplus_auto:1;
615 /** Does the filesystem support asynchronous direct-IO submission? */
616 unsigned async_dio:1;
618 /** Is lseek not implemented by fs? */
621 /** Does the filesystem support posix acls? */
622 unsigned posix_acl:1;
624 /** Check permissions based on the file mode or not? */
625 unsigned default_permissions:1;
627 /** Allow other than the mounter user to access the filesystem ? */
628 unsigned allow_other:1;
630 /** Does the filesystem support copy_file_range? */
631 unsigned no_copy_file_range:1;
633 /* Send DESTROY request */
634 unsigned int destroy:1;
636 /** The number of requests waiting for completion */
637 atomic_t num_waiting;
639 /** Negotiated minor version */
642 /** Entry on the fuse_conn_list */
643 struct list_head entry;
645 /** Device ID from super block */
648 /** Dentries in the control filesystem */
649 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
651 /** number of dentries used in the above array */
654 /** Key for lock owner ID scrambling */
657 /** Version counter for attribute changes */
658 atomic64_t attr_version;
660 /** Called on final put */
661 void (*release)(struct fuse_conn *);
663 /** Super block for this connection. */
664 struct super_block *sb;
666 /** Read/write semaphore to hold when accessing sb. */
667 struct rw_semaphore killsb;
669 /** List of device instances belonging to this connection */
670 struct list_head devices;
673 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
675 return sb->s_fs_info;
678 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
680 return get_fuse_conn_super(inode->i_sb);
683 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
685 return container_of(inode, struct fuse_inode, inode);
688 static inline u64 get_node_id(struct inode *inode)
690 return get_fuse_inode(inode)->nodeid;
693 static inline int invalid_nodeid(u64 nodeid)
695 return !nodeid || nodeid == FUSE_ROOT_ID;
698 static inline u64 fuse_get_attr_version(struct fuse_conn *fc)
700 return atomic64_read(&fc->attr_version);
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);
733 * Initialize READ or READDIR request
735 struct fuse_io_args {
738 struct fuse_read_in in;
742 struct fuse_write_in in;
743 struct fuse_write_out out;
746 struct fuse_args_pages ap;
747 struct fuse_io_priv *io;
748 struct fuse_file *ff;
751 void fuse_read_args_fill(struct fuse_io_args *ia, struct file *file, loff_t pos,
752 size_t count, int opcode);
756 * Send OPEN or OPENDIR request
758 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
760 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
761 void fuse_file_free(struct fuse_file *ff);
762 void fuse_finish_open(struct inode *inode, struct file *file);
764 void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff, int flags);
767 * Send RELEASE or RELEASEDIR request
769 void fuse_release_common(struct file *file, bool isdir);
772 * Send FSYNC or FSYNCDIR request
774 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
775 int datasync, int opcode);
780 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
781 struct fuse_notify_poll_wakeup_out *outarg);
784 * Initialize file operations on a regular file
786 void fuse_init_file_inode(struct inode *inode);
789 * Initialize inode operations on regular files and special files
791 void fuse_init_common(struct inode *inode);
794 * Initialize inode and file operations on a directory
796 void fuse_init_dir(struct inode *inode);
799 * Initialize inode operations on a symlink
801 void fuse_init_symlink(struct inode *inode);
804 * Change attributes of an inode
806 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
807 u64 attr_valid, u64 attr_version);
809 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
813 * Initialize the client device
815 int fuse_dev_init(void);
818 * Cleanup the client device
820 void fuse_dev_cleanup(void);
822 int fuse_ctl_init(void);
823 void __exit fuse_ctl_cleanup(void);
826 * Simple request sending that does request allocation and freeing
828 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
829 int fuse_simple_background(struct fuse_conn *fc, struct fuse_args *args,
833 * End a finished request
835 void fuse_request_end(struct fuse_conn *fc, struct fuse_req *req);
837 /* Abort all requests */
838 void fuse_abort_conn(struct fuse_conn *fc);
839 void fuse_wait_aborted(struct fuse_conn *fc);
842 * Invalidate inode attributes
844 void fuse_invalidate_attr(struct inode *inode);
846 void fuse_invalidate_entry_cache(struct dentry *entry);
848 void fuse_invalidate_atime(struct inode *inode);
850 u64 entry_attr_timeout(struct fuse_entry_out *o);
851 void fuse_change_entry_timeout(struct dentry *entry, struct fuse_entry_out *o);
854 * Acquire reference to fuse_conn
856 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
859 * Initialize fuse_conn
861 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns);
864 * Release reference to fuse_conn
866 void fuse_conn_put(struct fuse_conn *fc);
868 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc);
869 void fuse_dev_free(struct fuse_dev *fud);
870 void fuse_send_init(struct fuse_conn *fc);
873 * Add connection to control filesystem
875 int fuse_ctl_add_conn(struct fuse_conn *fc);
878 * Remove connection from control filesystem
880 void fuse_ctl_remove_conn(struct fuse_conn *fc);
883 * Is file type valid?
885 int fuse_valid_type(int m);
888 * Is current process allowed to perform filesystem operation?
890 int fuse_allow_current_process(struct fuse_conn *fc);
892 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
894 void fuse_update_ctime(struct inode *inode);
896 int fuse_update_attributes(struct inode *inode, struct file *file);
898 void fuse_flush_writepages(struct inode *inode);
900 void fuse_set_nowrite(struct inode *inode);
901 void fuse_release_nowrite(struct inode *inode);
904 * File-system tells the kernel to invalidate cache for the given node id.
906 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
907 loff_t offset, loff_t len);
910 * File-system tells the kernel to invalidate parent attributes and
911 * the dentry matching parent/name.
913 * If the child_nodeid is non-zero and:
914 * - matches the inode number for the dentry matching parent/name,
915 * - is not a mount point
916 * - is a file or oan empty directory
917 * then the dentry is unhashed (d_delete()).
919 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
920 u64 child_nodeid, struct qstr *name);
922 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
926 * fuse_direct_io() flags
929 /** If set, it is WRITE; otherwise - READ */
930 #define FUSE_DIO_WRITE (1 << 0)
932 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
933 #define FUSE_DIO_CUSE (1 << 1)
935 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
936 loff_t *ppos, int flags);
937 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
939 long fuse_ioctl_common(struct file *file, unsigned int cmd,
940 unsigned long arg, unsigned int flags);
941 __poll_t fuse_file_poll(struct file *file, poll_table *wait);
942 int fuse_dev_release(struct inode *inode, struct file *file);
944 bool fuse_write_update_size(struct inode *inode, loff_t pos);
946 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
947 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
949 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
952 void fuse_set_initialized(struct fuse_conn *fc);
954 void fuse_unlock_inode(struct inode *inode, bool locked);
955 bool fuse_lock_inode(struct inode *inode);
957 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
958 size_t size, int flags);
959 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
961 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
962 int fuse_removexattr(struct inode *inode, const char *name);
963 extern const struct xattr_handler *fuse_xattr_handlers[];
964 extern const struct xattr_handler *fuse_acl_xattr_handlers[];
965 extern const struct xattr_handler *fuse_no_acl_xattr_handlers[];
968 struct posix_acl *fuse_get_acl(struct inode *inode, int type);
969 int fuse_set_acl(struct inode *inode, struct posix_acl *acl, int type);
973 int fuse_readdir(struct file *file, struct dir_context *ctx);
976 * Return the number of bytes in an arguments list
978 unsigned int fuse_len_args(unsigned int numargs, struct fuse_arg *args);
980 #endif /* _FS_FUSE_I_H */