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 /** The sticky bit in inode->i_mode may have been removed, so
84 preserve the original mode */
87 /** 64 bit inode number */
90 /** Version of last attribute change */
94 /* Write related fields (regular file only) */
96 /* Files usable in writepage. Protected by fc->lock */
97 struct list_head write_files;
99 /* Writepages pending on truncate or fsync */
100 struct list_head queued_writes;
102 /* Number of sent writes, a negative bias
103 * (FUSE_NOWRITE) means more writes are blocked */
106 /* Waitq for writepage completion */
107 wait_queue_head_t page_waitq;
109 /* List of writepage requestst (pending or sent) */
110 struct list_head writepages;
113 /* readdir cache (directory only) */
115 /* true if fully cached */
121 /* position at end of cache (position of next entry) */
124 /* version of the cache */
127 /* modification time of directory when cache was
129 struct timespec64 mtime;
131 /* iversion of directory when cache was started */
134 /* protects above fields */
139 /** Miscellaneous bits describing inode state */
142 /** Lock for serializing lookup and readdir for back compatibility*/
146 /** FUSE inode state bits */
148 /** Advise readdirplus */
149 FUSE_I_ADVISE_RDPLUS,
150 /** Initialized with readdirplus */
152 /** An operation changing file size is in progress */
153 FUSE_I_SIZE_UNSTABLE,
158 /** FUSE specific file data */
160 /** Fuse connection for this file */
161 struct fuse_conn *fc;
163 /** Request reserved for flush and release */
164 struct fuse_req *reserved_req;
166 /** Kernel file handle guaranteed to be unique */
169 /** File handle used by userspace */
172 /** Node id of this file */
178 /** FOPEN_* flags returned by open */
181 /** Entry on inode's write_files list */
182 struct list_head write_entry;
184 /* Readdir related */
187 * Protects below fields against (crazy) parallel readdir on
188 * same open file. Uncontended in the normal case.
192 /* Dir stream position */
195 /* Offset in cache */
198 /* Version of cache we are reading */
203 /** RB node to be linked on fuse_conn->polled_files */
204 struct rb_node polled_node;
206 /** Wait queue head for poll */
207 wait_queue_head_t poll_wait;
209 /** Has flock been performed on this file? */
213 /** One input argument of a request */
219 /** The request input */
221 /** The request header */
222 struct fuse_in_header h;
224 /** True if the data for the last argument is in req->pages */
227 /** Number of arguments */
230 /** Array of arguments */
231 struct fuse_in_arg args[3];
234 /** One output argument of a request */
240 /** The request output */
242 /** Header returned from userspace */
243 struct fuse_out_header h;
246 * The following bitfields are not changed during the request
250 /** Last argument is variable length (can be shorter than
254 /** Last argument is a list of pages to copy data to */
257 /** Zero partially or not copied pages */
258 unsigned page_zeroing:1;
260 /** Pages may be replaced with new ones */
261 unsigned page_replace:1;
263 /** Number or arguments */
266 /** Array of arguments */
267 struct fuse_arg args[2];
270 /** FUSE page descriptor */
271 struct fuse_page_desc {
283 struct fuse_in_arg args[3];
289 struct fuse_arg args[2];
293 #define FUSE_ARGS(args) struct fuse_args args = {}
295 /** The request IO state (for asynchronous processing) */
296 struct fuse_io_priv {
308 struct completion *done;
312 #define FUSE_IO_PRIV_SYNC(i) \
314 .refcnt = KREF_INIT(1), \
322 * FR_ISREPLY: set if the request has reply
323 * FR_FORCE: force sending of the request even if interrupted
324 * FR_BACKGROUND: request is sent in the background
325 * FR_WAITING: request is counted as "waiting"
326 * FR_ABORTED: the request was aborted
327 * FR_INTERRUPTED: the request has been interrupted
328 * FR_LOCKED: data is being copied to/from the request
329 * FR_PENDING: request is not yet in userspace
330 * FR_SENT: request is in userspace, waiting for an answer
331 * FR_FINISHED: request is finished
332 * FR_PRIVATE: request is on private list
349 * A request to the client
351 * .waitq.lock protects the following fields:
353 * - FR_LOCKED (may also be modified under fc->lock, tested under both)
356 /** This can be on either pending processing or io lists in
358 struct list_head list;
360 /** Entry on the interrupts list */
361 struct list_head intr_entry;
366 /* Request flags, updated with test/set/clear_bit() */
369 /** The request input */
372 /** The request output */
375 /** Used to wake up the task waiting for completion of request*/
376 wait_queue_head_t waitq;
378 /** Data for asynchronous requests */
381 struct fuse_release_in in;
384 struct fuse_init_in init_in;
385 struct fuse_init_out init_out;
386 struct cuse_init_in cuse_init_in;
388 struct fuse_read_in in;
392 struct fuse_write_in in;
393 struct fuse_write_out out;
394 struct fuse_req *next;
396 struct fuse_notify_retrieve_in retrieve_in;
402 /** page-descriptor vector */
403 struct fuse_page_desc *page_descs;
405 /** size of the 'pages' array */
408 /** inline page vector */
409 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
411 /** inline page-descriptor vector */
412 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
414 /** number of pages in vector */
417 /** File used in the request (or NULL) */
418 struct fuse_file *ff;
420 /** Inode used in the request or NULL */
423 /** AIO control block */
424 struct fuse_io_priv *io;
426 /** Link on fi->writepages */
427 struct list_head writepages_entry;
429 /** Request completion callback */
430 void (*end)(struct fuse_conn *, struct fuse_req *);
432 /** Request is stolen from fuse_file->reserved_req */
433 struct file *stolen_file;
437 /** Connection established */
440 /** Readers of the connection are waiting on this */
441 wait_queue_head_t waitq;
443 /** The next unique request id */
446 /** The list of pending requests */
447 struct list_head pending;
449 /** Pending interrupts */
450 struct list_head interrupts;
452 /** Queue of pending forgets */
453 struct fuse_forget_link forget_list_head;
454 struct fuse_forget_link *forget_list_tail;
456 /** Batching of FORGET requests (positive indicates FORGET batch) */
459 /** O_ASYNC requests */
460 struct fasync_struct *fasync;
463 #define FUSE_PQ_HASH_BITS 8
464 #define FUSE_PQ_HASH_SIZE (1 << FUSE_PQ_HASH_BITS)
467 /** Connection established */
470 /** Lock protecting accessess to members of this structure */
473 /** Hash table of requests being processed */
474 struct list_head *processing;
476 /** The list of requests under I/O */
481 * Fuse device instance
484 /** Fuse connection for this device */
485 struct fuse_conn *fc;
487 /** Processing queue */
488 struct fuse_pqueue pq;
490 /** list entry on fc->devices */
491 struct list_head entry;
497 * This structure is created, when the filesystem is mounted, and is
498 * destroyed, when the client device is closed and the filesystem is
502 /** Lock protecting accessess to members of this structure */
508 /** Number of fuse_dev's */
513 /** The user id for this mount */
516 /** The group id for this mount */
519 /** The pid namespace for this mount */
520 struct pid_namespace *pid_ns;
522 /** The user namespace for this mount */
523 struct user_namespace *user_ns;
525 /** Maximum read size */
528 /** Maximum write size */
531 /** Maxmum number of pages that can be used in a single request */
532 unsigned int max_pages;
535 struct fuse_iqueue iq;
537 /** The next unique kernel file handle */
540 /** rbtree of fuse_files waiting for poll events indexed by ph */
541 struct rb_root polled_files;
543 /** Maximum number of outstanding background requests */
544 unsigned max_background;
546 /** Number of background requests at which congestion starts */
547 unsigned congestion_threshold;
549 /** Number of requests currently in the background */
550 unsigned num_background;
552 /** Number of background requests currently queued for userspace */
553 unsigned active_background;
555 /** The list of background requests set aside for later queuing */
556 struct list_head bg_queue;
558 /** Protects: max_background, congestion_threshold, num_background,
559 * active_background, bg_queue, blocked */
562 /** Flag indicating that INIT reply has been received. Allocating
563 * any fuse request will be suspended until the flag is set */
566 /** Flag indicating if connection is blocked. This will be
567 the case before the INIT reply is received, and if there
568 are too many outstading backgrounds requests */
571 /** waitq for blocked connection */
572 wait_queue_head_t blocked_waitq;
574 /** waitq for reserved requests */
575 wait_queue_head_t reserved_req_waitq;
577 /** Connection established, cleared on umount, connection
578 abort and device release */
581 /** Connection aborted via sysfs */
584 /** Connection failed (version mismatch). Cannot race with
585 setting other bitfields since it is only set once in INIT
586 reply, before any other request, and never cleared */
587 unsigned conn_error:1;
589 /** Connection successful. Only set in INIT */
590 unsigned conn_init:1;
592 /** Do readpages asynchronously? Only set in INIT */
593 unsigned async_read:1;
595 /** Return an unique read error after abort. Only set in INIT */
596 unsigned abort_err:1;
598 /** Do not send separate SETATTR request before open(O_TRUNC) */
599 unsigned atomic_o_trunc:1;
601 /** Filesystem supports NFS exporting. Only set in INIT */
602 unsigned export_support:1;
604 /** write-back cache policy (default is write-through) */
605 unsigned writeback_cache:1;
607 /** allow parallel lookups and readdir (default is serialized) */
608 unsigned parallel_dirops:1;
610 /** handle fs handles killing suid/sgid/cap on write/chown/trunc */
611 unsigned handle_killpriv:1;
614 * The following bitfields are only for optimization purposes
615 * and hence races in setting them will not cause malfunction
618 /** Is open/release not implemented by fs? */
621 /** Is fsync not implemented by fs? */
624 /** Is fsyncdir not implemented by fs? */
625 unsigned no_fsyncdir:1;
627 /** Is flush not implemented by fs? */
630 /** Is setxattr not implemented by fs? */
631 unsigned no_setxattr:1;
633 /** Is getxattr not implemented by fs? */
634 unsigned no_getxattr:1;
636 /** Is listxattr not implemented by fs? */
637 unsigned no_listxattr:1;
639 /** Is removexattr not implemented by fs? */
640 unsigned no_removexattr:1;
642 /** Are posix file locking primitives not implemented by fs? */
645 /** Is access not implemented by fs? */
646 unsigned no_access:1;
648 /** Is create not implemented by fs? */
649 unsigned no_create:1;
651 /** Is interrupt not implemented by fs? */
652 unsigned no_interrupt:1;
654 /** Is bmap not implemented by fs? */
657 /** Is poll not implemented by fs? */
660 /** Do multi-page cached writes */
661 unsigned big_writes:1;
663 /** Don't apply umask to creation modes */
664 unsigned dont_mask:1;
666 /** Are BSD file locking primitives not implemented by fs? */
669 /** Is fallocate not implemented by fs? */
670 unsigned no_fallocate:1;
672 /** Is rename with flags implemented by fs? */
673 unsigned no_rename2:1;
675 /** Use enhanced/automatic page cache invalidation. */
676 unsigned auto_inval_data:1;
678 /** Does the filesystem support readdirplus? */
679 unsigned do_readdirplus:1;
681 /** Does the filesystem want adaptive readdirplus? */
682 unsigned readdirplus_auto:1;
684 /** Does the filesystem support asynchronous direct-IO submission? */
685 unsigned async_dio:1;
687 /** Is lseek not implemented by fs? */
690 /** Does the filesystem support posix acls? */
691 unsigned posix_acl:1;
693 /** Check permissions based on the file mode or not? */
694 unsigned default_permissions:1;
696 /** Allow other than the mounter user to access the filesystem ? */
697 unsigned allow_other:1;
699 /** Does the filesystem support copy_file_range? */
700 unsigned no_copy_file_range:1;
702 /** The number of requests waiting for completion */
703 atomic_t num_waiting;
705 /** Negotiated minor version */
708 /** Entry on the fuse_conn_list */
709 struct list_head entry;
711 /** Device ID from super block */
714 /** Dentries in the control filesystem */
715 struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
717 /** number of dentries used in the above array */
720 /** Key for lock owner ID scrambling */
723 /** Reserved request for the DESTROY message */
724 struct fuse_req *destroy_req;
726 /** Version counter for attribute changes */
729 /** Called on final put */
730 void (*release)(struct fuse_conn *);
732 /** Super block for this connection. */
733 struct super_block *sb;
735 /** Read/write semaphore to hold when accessing sb. */
736 struct rw_semaphore killsb;
738 /** List of device instances belonging to this connection */
739 struct list_head devices;
742 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
744 return sb->s_fs_info;
747 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
749 return get_fuse_conn_super(inode->i_sb);
752 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
754 return container_of(inode, struct fuse_inode, inode);
757 static inline u64 get_node_id(struct inode *inode)
759 return get_fuse_inode(inode)->nodeid;
762 static inline int invalid_nodeid(u64 nodeid)
764 return !nodeid || nodeid == FUSE_ROOT_ID;
767 /** Device operations */
768 extern const struct file_operations fuse_dev_operations;
770 extern const struct dentry_operations fuse_dentry_operations;
771 extern const struct dentry_operations fuse_root_dentry_operations;
774 * Inode to nodeid comparison.
776 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
779 * Get a filled in inode
781 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
782 int generation, struct fuse_attr *attr,
783 u64 attr_valid, u64 attr_version);
785 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
786 struct fuse_entry_out *outarg, struct inode **inode);
789 * Send FORGET command
791 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
792 u64 nodeid, u64 nlookup);
794 struct fuse_forget_link *fuse_alloc_forget(void);
796 /* Used by READDIRPLUS */
797 void fuse_force_forget(struct file *file, u64 nodeid);
800 * Initialize READ or READDIR request
802 void fuse_read_fill(struct fuse_req *req, struct file *file,
803 loff_t pos, size_t count, int opcode);
806 * Send OPEN or OPENDIR request
808 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
810 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
811 void fuse_file_free(struct fuse_file *ff);
812 void fuse_finish_open(struct inode *inode, struct file *file);
814 void fuse_sync_release(struct fuse_file *ff, int flags);
817 * Send RELEASE or RELEASEDIR request
819 void fuse_release_common(struct file *file, int opcode);
822 * Send FSYNC or FSYNCDIR request
824 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
825 int datasync, int isdir);
830 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
831 struct fuse_notify_poll_wakeup_out *outarg);
834 * Initialize file operations on a regular file
836 void fuse_init_file_inode(struct inode *inode);
839 * Initialize inode operations on regular files and special files
841 void fuse_init_common(struct inode *inode);
844 * Initialize inode and file operations on a directory
846 void fuse_init_dir(struct inode *inode);
849 * Initialize inode operations on a symlink
851 void fuse_init_symlink(struct inode *inode);
854 * Change attributes of an inode
856 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
857 u64 attr_valid, u64 attr_version);
859 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
863 * Initialize the client device
865 int fuse_dev_init(void);
868 * Cleanup the client device
870 void fuse_dev_cleanup(void);
872 int fuse_ctl_init(void);
873 void __exit fuse_ctl_cleanup(void);
878 struct fuse_req *fuse_request_alloc(unsigned npages);
880 struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
882 bool fuse_req_realloc_pages(struct fuse_conn *fc, struct fuse_req *req,
889 void fuse_request_free(struct fuse_req *req);
892 * Get a request, may fail with -ENOMEM,
893 * caller should specify # elements in req->pages[] explicitly
895 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
896 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
900 * Increment reference count on request
902 void __fuse_get_request(struct fuse_req *req);
905 * Gets a requests for a file operation, always succeeds
907 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
911 * Decrement reference count of a request. If count goes to zero free
914 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
917 * Send a request (synchronous)
919 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
922 * Simple request sending that does request allocation and freeing
924 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args);
927 * Send a request in the background
929 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
930 bool fuse_request_queue_background(struct fuse_conn *fc, struct fuse_req *req);
932 /* Abort all requests */
933 void fuse_abort_conn(struct fuse_conn *fc, bool is_abort);
934 void fuse_wait_aborted(struct fuse_conn *fc);
937 * Invalidate inode attributes
939 void fuse_invalidate_attr(struct inode *inode);
941 void fuse_invalidate_entry_cache(struct dentry *entry);
943 void fuse_invalidate_atime(struct inode *inode);
945 u64 entry_attr_timeout(struct fuse_entry_out *o);
946 void fuse_change_entry_timeout(struct dentry *entry, struct fuse_entry_out *o);
949 * Acquire reference to fuse_conn
951 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
954 * Initialize fuse_conn
956 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns);
959 * Release reference to fuse_conn
961 void fuse_conn_put(struct fuse_conn *fc);
963 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc);
964 void fuse_dev_free(struct fuse_dev *fud);
967 * Add connection to control filesystem
969 int fuse_ctl_add_conn(struct fuse_conn *fc);
972 * Remove connection from control filesystem
974 void fuse_ctl_remove_conn(struct fuse_conn *fc);
977 * Is file type valid?
979 int fuse_valid_type(int m);
982 * Is current process allowed to perform filesystem operation?
984 int fuse_allow_current_process(struct fuse_conn *fc);
986 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
988 void fuse_update_ctime(struct inode *inode);
990 int fuse_update_attributes(struct inode *inode, struct file *file);
992 void fuse_flush_writepages(struct inode *inode);
994 void fuse_set_nowrite(struct inode *inode);
995 void fuse_release_nowrite(struct inode *inode);
997 u64 fuse_get_attr_version(struct fuse_conn *fc);
1000 * File-system tells the kernel to invalidate cache for the given node id.
1002 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
1003 loff_t offset, loff_t len);
1006 * File-system tells the kernel to invalidate parent attributes and
1007 * the dentry matching parent/name.
1009 * If the child_nodeid is non-zero and:
1010 * - matches the inode number for the dentry matching parent/name,
1011 * - is not a mount point
1012 * - is a file or oan empty directory
1013 * then the dentry is unhashed (d_delete()).
1015 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
1016 u64 child_nodeid, struct qstr *name);
1018 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
1022 * fuse_direct_io() flags
1025 /** If set, it is WRITE; otherwise - READ */
1026 #define FUSE_DIO_WRITE (1 << 0)
1028 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
1029 #define FUSE_DIO_CUSE (1 << 1)
1031 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1032 loff_t *ppos, int flags);
1033 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1034 unsigned int flags);
1035 long fuse_ioctl_common(struct file *file, unsigned int cmd,
1036 unsigned long arg, unsigned int flags);
1037 __poll_t fuse_file_poll(struct file *file, poll_table *wait);
1038 int fuse_dev_release(struct inode *inode, struct file *file);
1040 bool fuse_write_update_size(struct inode *inode, loff_t pos);
1042 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
1043 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
1045 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1048 void fuse_set_initialized(struct fuse_conn *fc);
1050 void fuse_unlock_inode(struct inode *inode, bool locked);
1051 bool fuse_lock_inode(struct inode *inode);
1053 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
1054 size_t size, int flags);
1055 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
1057 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
1058 int fuse_removexattr(struct inode *inode, const char *name);
1059 extern const struct xattr_handler *fuse_xattr_handlers[];
1060 extern const struct xattr_handler *fuse_acl_xattr_handlers[];
1061 extern const struct xattr_handler *fuse_no_acl_xattr_handlers[];
1064 struct posix_acl *fuse_get_acl(struct inode *inode, int type);
1065 int fuse_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1069 int fuse_readdir(struct file *file, struct dir_context *ctx);
1071 #endif /* _FS_FUSE_I_H */