xref: /openbmc/linux/fs/fuse/fuse_i.h (revision 71de0a05)
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
4 
5   This program can be distributed under the terms of the GNU GPL.
6   See the file COPYING.
7 */
8 
9 #ifndef _FS_FUSE_I_H
10 #define _FS_FUSE_I_H
11 
12 #ifndef pr_fmt
13 # define pr_fmt(fmt) "fuse: " fmt
14 #endif
15 
16 #include <linux/fuse.h>
17 #include <linux/fs.h>
18 #include <linux/mount.h>
19 #include <linux/wait.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/mm.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>
34 
35 /** Default max number of pages that can be used in a single read request */
36 #define FUSE_DEFAULT_MAX_PAGES_PER_REQ 32
37 
38 /** Maximum of max_pages received in init_out */
39 #define FUSE_MAX_MAX_PAGES 256
40 
41 /** Bias for fi->writectr, meaning new writepages must not be sent */
42 #define FUSE_NOWRITE INT_MIN
43 
44 /** It could be as large as PATH_MAX, but would that have any uses? */
45 #define FUSE_NAME_MAX 1024
46 
47 /** Number of dentries for each connection in the control filesystem */
48 #define FUSE_CTL_NUM_DENTRIES 5
49 
50 /** List of active connections */
51 extern struct list_head fuse_conn_list;
52 
53 /** Global mutex protecting fuse_conn_list and the control filesystem */
54 extern struct mutex fuse_mutex;
55 
56 /** Module parameters */
57 extern unsigned max_user_bgreq;
58 extern unsigned max_user_congthresh;
59 
60 /* One forget request */
61 struct fuse_forget_link {
62 	struct fuse_forget_one forget_one;
63 	struct fuse_forget_link *next;
64 };
65 
66 /** FUSE inode */
67 struct fuse_inode {
68 	/** Inode data */
69 	struct inode inode;
70 
71 	/** Unique ID, which identifies the inode between userspace
72 	 * and kernel */
73 	u64 nodeid;
74 
75 	/** Number of lookups on this inode */
76 	u64 nlookup;
77 
78 	/** The request used for sending the FORGET message */
79 	struct fuse_forget_link *forget;
80 
81 	/** Time in jiffies until the file attributes are valid */
82 	u64 i_time;
83 
84 	/* Which attributes are invalid */
85 	u32 inval_mask;
86 
87 	/** The sticky bit in inode->i_mode may have been removed, so
88 	    preserve the original mode */
89 	umode_t orig_i_mode;
90 
91 	/** 64 bit inode number */
92 	u64 orig_ino;
93 
94 	/** Version of last attribute change */
95 	u64 attr_version;
96 
97 	union {
98 		/* Write related fields (regular file only) */
99 		struct {
100 			/* Files usable in writepage.  Protected by fi->lock */
101 			struct list_head write_files;
102 
103 			/* Writepages pending on truncate or fsync */
104 			struct list_head queued_writes;
105 
106 			/* Number of sent writes, a negative bias
107 			 * (FUSE_NOWRITE) means more writes are blocked */
108 			int writectr;
109 
110 			/* Waitq for writepage completion */
111 			wait_queue_head_t page_waitq;
112 
113 			/* List of writepage requestst (pending or sent) */
114 			struct rb_root writepages;
115 		};
116 
117 		/* readdir cache (directory only) */
118 		struct {
119 			/* true if fully cached */
120 			bool cached;
121 
122 			/* size of cache */
123 			loff_t size;
124 
125 			/* position at end of cache (position of next entry) */
126 			loff_t pos;
127 
128 			/* version of the cache */
129 			u64 version;
130 
131 			/* modification time of directory when cache was
132 			 * started */
133 			struct timespec64 mtime;
134 
135 			/* iversion of directory when cache was started */
136 			u64 iversion;
137 
138 			/* protects above fields */
139 			spinlock_t lock;
140 		} rdc;
141 	};
142 
143 	/** Miscellaneous bits describing inode state */
144 	unsigned long state;
145 
146 	/** Lock for serializing lookup and readdir for back compatibility*/
147 	struct mutex mutex;
148 
149 	/** Lock to protect write related fields */
150 	spinlock_t lock;
151 
152 #ifdef CONFIG_FUSE_DAX
153 	/*
154 	 * Dax specific inode data
155 	 */
156 	struct fuse_inode_dax *dax;
157 #endif
158 };
159 
160 /** FUSE inode state bits */
161 enum {
162 	/** Advise readdirplus  */
163 	FUSE_I_ADVISE_RDPLUS,
164 	/** Initialized with readdirplus */
165 	FUSE_I_INIT_RDPLUS,
166 	/** An operation changing file size is in progress  */
167 	FUSE_I_SIZE_UNSTABLE,
168 	/* Bad inode */
169 	FUSE_I_BAD,
170 };
171 
172 struct fuse_conn;
173 struct fuse_mount;
174 struct fuse_release_args;
175 
176 /** FUSE specific file data */
177 struct fuse_file {
178 	/** Fuse connection for this file */
179 	struct fuse_mount *fm;
180 
181 	/* Argument space reserved for release */
182 	struct fuse_release_args *release_args;
183 
184 	/** Kernel file handle guaranteed to be unique */
185 	u64 kh;
186 
187 	/** File handle used by userspace */
188 	u64 fh;
189 
190 	/** Node id of this file */
191 	u64 nodeid;
192 
193 	/** Refcount */
194 	refcount_t count;
195 
196 	/** FOPEN_* flags returned by open */
197 	u32 open_flags;
198 
199 	/** Entry on inode's write_files list */
200 	struct list_head write_entry;
201 
202 	/* Readdir related */
203 	struct {
204 		/*
205 		 * Protects below fields against (crazy) parallel readdir on
206 		 * same open file.  Uncontended in the normal case.
207 		 */
208 		struct mutex lock;
209 
210 		/* Dir stream position */
211 		loff_t pos;
212 
213 		/* Offset in cache */
214 		loff_t cache_off;
215 
216 		/* Version of cache we are reading */
217 		u64 version;
218 
219 	} readdir;
220 
221 	/** RB node to be linked on fuse_conn->polled_files */
222 	struct rb_node polled_node;
223 
224 	/** Wait queue head for poll */
225 	wait_queue_head_t poll_wait;
226 
227 	/** Has flock been performed on this file? */
228 	bool flock:1;
229 };
230 
231 /** One input argument of a request */
232 struct fuse_in_arg {
233 	unsigned size;
234 	const void *value;
235 };
236 
237 /** One output argument of a request */
238 struct fuse_arg {
239 	unsigned size;
240 	void *value;
241 };
242 
243 /** FUSE page descriptor */
244 struct fuse_page_desc {
245 	unsigned int length;
246 	unsigned int offset;
247 };
248 
249 struct fuse_args {
250 	uint64_t nodeid;
251 	uint32_t opcode;
252 	uint8_t in_numargs;
253 	uint8_t out_numargs;
254 	uint8_t ext_idx;
255 	bool force:1;
256 	bool noreply:1;
257 	bool nocreds:1;
258 	bool in_pages:1;
259 	bool out_pages:1;
260 	bool user_pages:1;
261 	bool out_argvar:1;
262 	bool page_zeroing:1;
263 	bool page_replace:1;
264 	bool may_block:1;
265 	bool is_ext:1;
266 	struct fuse_in_arg in_args[3];
267 	struct fuse_arg out_args[2];
268 	void (*end)(struct fuse_mount *fm, struct fuse_args *args, int error);
269 };
270 
271 struct fuse_args_pages {
272 	struct fuse_args args;
273 	struct page **pages;
274 	struct fuse_page_desc *descs;
275 	unsigned int num_pages;
276 };
277 
278 #define FUSE_ARGS(args) struct fuse_args args = {}
279 
280 /** The request IO state (for asynchronous processing) */
281 struct fuse_io_priv {
282 	struct kref refcnt;
283 	int async;
284 	spinlock_t lock;
285 	unsigned reqs;
286 	ssize_t bytes;
287 	size_t size;
288 	__u64 offset;
289 	bool write;
290 	bool should_dirty;
291 	int err;
292 	struct kiocb *iocb;
293 	struct completion *done;
294 	bool blocking;
295 };
296 
297 #define FUSE_IO_PRIV_SYNC(i) \
298 {					\
299 	.refcnt = KREF_INIT(1),		\
300 	.async = 0,			\
301 	.iocb = i,			\
302 }
303 
304 /**
305  * Request flags
306  *
307  * FR_ISREPLY:		set if the request has reply
308  * FR_FORCE:		force sending of the request even if interrupted
309  * FR_BACKGROUND:	request is sent in the background
310  * FR_WAITING:		request is counted as "waiting"
311  * FR_ABORTED:		the request was aborted
312  * FR_INTERRUPTED:	the request has been interrupted
313  * FR_LOCKED:		data is being copied to/from the request
314  * FR_PENDING:		request is not yet in userspace
315  * FR_SENT:		request is in userspace, waiting for an answer
316  * FR_FINISHED:		request is finished
317  * FR_PRIVATE:		request is on private list
318  * FR_ASYNC:		request is asynchronous
319  */
320 enum fuse_req_flag {
321 	FR_ISREPLY,
322 	FR_FORCE,
323 	FR_BACKGROUND,
324 	FR_WAITING,
325 	FR_ABORTED,
326 	FR_INTERRUPTED,
327 	FR_LOCKED,
328 	FR_PENDING,
329 	FR_SENT,
330 	FR_FINISHED,
331 	FR_PRIVATE,
332 	FR_ASYNC,
333 };
334 
335 /**
336  * A request to the client
337  *
338  * .waitq.lock protects the following fields:
339  *   - FR_ABORTED
340  *   - FR_LOCKED (may also be modified under fc->lock, tested under both)
341  */
342 struct fuse_req {
343 	/** This can be on either pending processing or io lists in
344 	    fuse_conn */
345 	struct list_head list;
346 
347 	/** Entry on the interrupts list  */
348 	struct list_head intr_entry;
349 
350 	/* Input/output arguments */
351 	struct fuse_args *args;
352 
353 	/** refcount */
354 	refcount_t count;
355 
356 	/* Request flags, updated with test/set/clear_bit() */
357 	unsigned long flags;
358 
359 	/* The request input header */
360 	struct {
361 		struct fuse_in_header h;
362 	} in;
363 
364 	/* The request output header */
365 	struct {
366 		struct fuse_out_header h;
367 	} out;
368 
369 	/** Used to wake up the task waiting for completion of request*/
370 	wait_queue_head_t waitq;
371 
372 #if IS_ENABLED(CONFIG_VIRTIO_FS)
373 	/** virtio-fs's physically contiguous buffer for in and out args */
374 	void *argbuf;
375 #endif
376 
377 	/** fuse_mount this request belongs to */
378 	struct fuse_mount *fm;
379 };
380 
381 struct fuse_iqueue;
382 
383 /**
384  * Input queue callbacks
385  *
386  * Input queue signalling is device-specific.  For example, the /dev/fuse file
387  * uses fiq->waitq and fasync to wake processes that are waiting on queue
388  * readiness.  These callbacks allow other device types to respond to input
389  * queue activity.
390  */
391 struct fuse_iqueue_ops {
392 	/**
393 	 * Signal that a forget has been queued
394 	 */
395 	void (*wake_forget_and_unlock)(struct fuse_iqueue *fiq)
396 		__releases(fiq->lock);
397 
398 	/**
399 	 * Signal that an INTERRUPT request has been queued
400 	 */
401 	void (*wake_interrupt_and_unlock)(struct fuse_iqueue *fiq)
402 		__releases(fiq->lock);
403 
404 	/**
405 	 * Signal that a request has been queued
406 	 */
407 	void (*wake_pending_and_unlock)(struct fuse_iqueue *fiq)
408 		__releases(fiq->lock);
409 
410 	/**
411 	 * Clean up when fuse_iqueue is destroyed
412 	 */
413 	void (*release)(struct fuse_iqueue *fiq);
414 };
415 
416 /** /dev/fuse input queue operations */
417 extern const struct fuse_iqueue_ops fuse_dev_fiq_ops;
418 
419 struct fuse_iqueue {
420 	/** Connection established */
421 	unsigned connected;
422 
423 	/** Lock protecting accesses to members of this structure */
424 	spinlock_t lock;
425 
426 	/** Readers of the connection are waiting on this */
427 	wait_queue_head_t waitq;
428 
429 	/** The next unique request id */
430 	u64 reqctr;
431 
432 	/** The list of pending requests */
433 	struct list_head pending;
434 
435 	/** Pending interrupts */
436 	struct list_head interrupts;
437 
438 	/** Queue of pending forgets */
439 	struct fuse_forget_link forget_list_head;
440 	struct fuse_forget_link *forget_list_tail;
441 
442 	/** Batching of FORGET requests (positive indicates FORGET batch) */
443 	int forget_batch;
444 
445 	/** O_ASYNC requests */
446 	struct fasync_struct *fasync;
447 
448 	/** Device-specific callbacks */
449 	const struct fuse_iqueue_ops *ops;
450 
451 	/** Device-specific state */
452 	void *priv;
453 };
454 
455 #define FUSE_PQ_HASH_BITS 8
456 #define FUSE_PQ_HASH_SIZE (1 << FUSE_PQ_HASH_BITS)
457 
458 struct fuse_pqueue {
459 	/** Connection established */
460 	unsigned connected;
461 
462 	/** Lock protecting accessess to  members of this structure */
463 	spinlock_t lock;
464 
465 	/** Hash table of requests being processed */
466 	struct list_head *processing;
467 
468 	/** The list of requests under I/O */
469 	struct list_head io;
470 };
471 
472 /**
473  * Fuse device instance
474  */
475 struct fuse_dev {
476 	/** Fuse connection for this device */
477 	struct fuse_conn *fc;
478 
479 	/** Processing queue */
480 	struct fuse_pqueue pq;
481 
482 	/** list entry on fc->devices */
483 	struct list_head entry;
484 };
485 
486 enum fuse_dax_mode {
487 	FUSE_DAX_INODE_DEFAULT,	/* default */
488 	FUSE_DAX_ALWAYS,	/* "-o dax=always" */
489 	FUSE_DAX_NEVER,		/* "-o dax=never" */
490 	FUSE_DAX_INODE_USER,	/* "-o dax=inode" */
491 };
492 
493 static inline bool fuse_is_inode_dax_mode(enum fuse_dax_mode mode)
494 {
495 	return mode == FUSE_DAX_INODE_DEFAULT || mode == FUSE_DAX_INODE_USER;
496 }
497 
498 struct fuse_fs_context {
499 	int fd;
500 	struct file *file;
501 	unsigned int rootmode;
502 	kuid_t user_id;
503 	kgid_t group_id;
504 	bool is_bdev:1;
505 	bool fd_present:1;
506 	bool rootmode_present:1;
507 	bool user_id_present:1;
508 	bool group_id_present:1;
509 	bool default_permissions:1;
510 	bool allow_other:1;
511 	bool destroy:1;
512 	bool no_control:1;
513 	bool no_force_umount:1;
514 	bool legacy_opts_show:1;
515 	enum fuse_dax_mode dax_mode;
516 	unsigned int max_read;
517 	unsigned int blksize;
518 	const char *subtype;
519 
520 	/* DAX device, may be NULL */
521 	struct dax_device *dax_dev;
522 
523 	/* fuse_dev pointer to fill in, should contain NULL on entry */
524 	void **fudptr;
525 };
526 
527 struct fuse_sync_bucket {
528 	/* count is a possible scalability bottleneck */
529 	atomic_t count;
530 	wait_queue_head_t waitq;
531 	struct rcu_head rcu;
532 };
533 
534 /**
535  * A Fuse connection.
536  *
537  * This structure is created, when the root filesystem is mounted, and
538  * is destroyed, when the client device is closed and the last
539  * fuse_mount is destroyed.
540  */
541 struct fuse_conn {
542 	/** Lock protecting accessess to  members of this structure */
543 	spinlock_t lock;
544 
545 	/** Refcount */
546 	refcount_t count;
547 
548 	/** Number of fuse_dev's */
549 	atomic_t dev_count;
550 
551 	struct rcu_head rcu;
552 
553 	/** The user id for this mount */
554 	kuid_t user_id;
555 
556 	/** The group id for this mount */
557 	kgid_t group_id;
558 
559 	/** The pid namespace for this mount */
560 	struct pid_namespace *pid_ns;
561 
562 	/** The user namespace for this mount */
563 	struct user_namespace *user_ns;
564 
565 	/** Maximum read size */
566 	unsigned max_read;
567 
568 	/** Maximum write size */
569 	unsigned max_write;
570 
571 	/** Maximum number of pages that can be used in a single request */
572 	unsigned int max_pages;
573 
574 	/** Constrain ->max_pages to this value during feature negotiation */
575 	unsigned int max_pages_limit;
576 
577 	/** Input queue */
578 	struct fuse_iqueue iq;
579 
580 	/** The next unique kernel file handle */
581 	atomic64_t khctr;
582 
583 	/** rbtree of fuse_files waiting for poll events indexed by ph */
584 	struct rb_root polled_files;
585 
586 	/** Maximum number of outstanding background requests */
587 	unsigned max_background;
588 
589 	/** Number of background requests at which congestion starts */
590 	unsigned congestion_threshold;
591 
592 	/** Number of requests currently in the background */
593 	unsigned num_background;
594 
595 	/** Number of background requests currently queued for userspace */
596 	unsigned active_background;
597 
598 	/** The list of background requests set aside for later queuing */
599 	struct list_head bg_queue;
600 
601 	/** Protects: max_background, congestion_threshold, num_background,
602 	 * active_background, bg_queue, blocked */
603 	spinlock_t bg_lock;
604 
605 	/** Flag indicating that INIT reply has been received. Allocating
606 	 * any fuse request will be suspended until the flag is set */
607 	int initialized;
608 
609 	/** Flag indicating if connection is blocked.  This will be
610 	    the case before the INIT reply is received, and if there
611 	    are too many outstading backgrounds requests */
612 	int blocked;
613 
614 	/** waitq for blocked connection */
615 	wait_queue_head_t blocked_waitq;
616 
617 	/** Connection established, cleared on umount, connection
618 	    abort and device release */
619 	unsigned connected;
620 
621 	/** Connection aborted via sysfs */
622 	bool aborted;
623 
624 	/** Connection failed (version mismatch).  Cannot race with
625 	    setting other bitfields since it is only set once in INIT
626 	    reply, before any other request, and never cleared */
627 	unsigned conn_error:1;
628 
629 	/** Connection successful.  Only set in INIT */
630 	unsigned conn_init:1;
631 
632 	/** Do readahead asynchronously?  Only set in INIT */
633 	unsigned async_read:1;
634 
635 	/** Return an unique read error after abort.  Only set in INIT */
636 	unsigned abort_err:1;
637 
638 	/** Do not send separate SETATTR request before open(O_TRUNC)  */
639 	unsigned atomic_o_trunc:1;
640 
641 	/** Filesystem supports NFS exporting.  Only set in INIT */
642 	unsigned export_support:1;
643 
644 	/** write-back cache policy (default is write-through) */
645 	unsigned writeback_cache:1;
646 
647 	/** allow parallel lookups and readdir (default is serialized) */
648 	unsigned parallel_dirops:1;
649 
650 	/** handle fs handles killing suid/sgid/cap on write/chown/trunc */
651 	unsigned handle_killpriv:1;
652 
653 	/** cache READLINK responses in page cache */
654 	unsigned cache_symlinks:1;
655 
656 	/* show legacy mount options */
657 	unsigned int legacy_opts_show:1;
658 
659 	/*
660 	 * fs kills suid/sgid/cap on write/chown/trunc. suid is killed on
661 	 * write/trunc only if caller did not have CAP_FSETID.  sgid is killed
662 	 * on write/truncate only if caller did not have CAP_FSETID as well as
663 	 * file has group execute permission.
664 	 */
665 	unsigned handle_killpriv_v2:1;
666 
667 	/*
668 	 * The following bitfields are only for optimization purposes
669 	 * and hence races in setting them will not cause malfunction
670 	 */
671 
672 	/** Is open/release not implemented by fs? */
673 	unsigned no_open:1;
674 
675 	/** Is opendir/releasedir not implemented by fs? */
676 	unsigned no_opendir:1;
677 
678 	/** Is fsync not implemented by fs? */
679 	unsigned no_fsync:1;
680 
681 	/** Is fsyncdir not implemented by fs? */
682 	unsigned no_fsyncdir:1;
683 
684 	/** Is flush not implemented by fs? */
685 	unsigned no_flush:1;
686 
687 	/** Is setxattr not implemented by fs? */
688 	unsigned no_setxattr:1;
689 
690 	/** Does file server support extended setxattr */
691 	unsigned setxattr_ext:1;
692 
693 	/** Is getxattr not implemented by fs? */
694 	unsigned no_getxattr:1;
695 
696 	/** Is listxattr not implemented by fs? */
697 	unsigned no_listxattr:1;
698 
699 	/** Is removexattr not implemented by fs? */
700 	unsigned no_removexattr:1;
701 
702 	/** Are posix file locking primitives not implemented by fs? */
703 	unsigned no_lock:1;
704 
705 	/** Is access not implemented by fs? */
706 	unsigned no_access:1;
707 
708 	/** Is create not implemented by fs? */
709 	unsigned no_create:1;
710 
711 	/** Is interrupt not implemented by fs? */
712 	unsigned no_interrupt:1;
713 
714 	/** Is bmap not implemented by fs? */
715 	unsigned no_bmap:1;
716 
717 	/** Is poll not implemented by fs? */
718 	unsigned no_poll:1;
719 
720 	/** Do multi-page cached writes */
721 	unsigned big_writes:1;
722 
723 	/** Don't apply umask to creation modes */
724 	unsigned dont_mask:1;
725 
726 	/** Are BSD file locking primitives not implemented by fs? */
727 	unsigned no_flock:1;
728 
729 	/** Is fallocate not implemented by fs? */
730 	unsigned no_fallocate:1;
731 
732 	/** Is rename with flags implemented by fs? */
733 	unsigned no_rename2:1;
734 
735 	/** Use enhanced/automatic page cache invalidation. */
736 	unsigned auto_inval_data:1;
737 
738 	/** Filesystem is fully responsible for page cache invalidation. */
739 	unsigned explicit_inval_data:1;
740 
741 	/** Does the filesystem support readdirplus? */
742 	unsigned do_readdirplus:1;
743 
744 	/** Does the filesystem want adaptive readdirplus? */
745 	unsigned readdirplus_auto:1;
746 
747 	/** Does the filesystem support asynchronous direct-IO submission? */
748 	unsigned async_dio:1;
749 
750 	/** Is lseek not implemented by fs? */
751 	unsigned no_lseek:1;
752 
753 	/** Does the filesystem support posix acls? */
754 	unsigned posix_acl:1;
755 
756 	/** Check permissions based on the file mode or not? */
757 	unsigned default_permissions:1;
758 
759 	/** Allow other than the mounter user to access the filesystem ? */
760 	unsigned allow_other:1;
761 
762 	/** Does the filesystem support copy_file_range? */
763 	unsigned no_copy_file_range:1;
764 
765 	/* Send DESTROY request */
766 	unsigned int destroy:1;
767 
768 	/* Delete dentries that have gone stale */
769 	unsigned int delete_stale:1;
770 
771 	/** Do not create entry in fusectl fs */
772 	unsigned int no_control:1;
773 
774 	/** Do not allow MNT_FORCE umount */
775 	unsigned int no_force_umount:1;
776 
777 	/* Auto-mount submounts announced by the server */
778 	unsigned int auto_submounts:1;
779 
780 	/* Propagate syncfs() to server */
781 	unsigned int sync_fs:1;
782 
783 	/* Initialize security xattrs when creating a new inode */
784 	unsigned int init_security:1;
785 
786 	/* Add supplementary group info when creating a new inode */
787 	unsigned int create_supp_group:1;
788 
789 	/* Does the filesystem support per inode DAX? */
790 	unsigned int inode_dax:1;
791 
792 	/* Is tmpfile not implemented by fs? */
793 	unsigned int no_tmpfile:1;
794 
795 	/** The number of requests waiting for completion */
796 	atomic_t num_waiting;
797 
798 	/** Negotiated minor version */
799 	unsigned minor;
800 
801 	/** Entry on the fuse_mount_list */
802 	struct list_head entry;
803 
804 	/** Device ID from the root super block */
805 	dev_t dev;
806 
807 	/** Dentries in the control filesystem */
808 	struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
809 
810 	/** number of dentries used in the above array */
811 	int ctl_ndents;
812 
813 	/** Key for lock owner ID scrambling */
814 	u32 scramble_key[4];
815 
816 	/** Version counter for attribute changes */
817 	atomic64_t attr_version;
818 
819 	/** Called on final put */
820 	void (*release)(struct fuse_conn *);
821 
822 	/**
823 	 * Read/write semaphore to hold when accessing the sb of any
824 	 * fuse_mount belonging to this connection
825 	 */
826 	struct rw_semaphore killsb;
827 
828 	/** List of device instances belonging to this connection */
829 	struct list_head devices;
830 
831 #ifdef CONFIG_FUSE_DAX
832 	/* Dax mode */
833 	enum fuse_dax_mode dax_mode;
834 
835 	/* Dax specific conn data, non-NULL if DAX is enabled */
836 	struct fuse_conn_dax *dax;
837 #endif
838 
839 	/** List of filesystems using this connection */
840 	struct list_head mounts;
841 
842 	/* New writepages go into this bucket */
843 	struct fuse_sync_bucket __rcu *curr_bucket;
844 };
845 
846 /*
847  * Represents a mounted filesystem, potentially a submount.
848  *
849  * This object allows sharing a fuse_conn between separate mounts to
850  * allow submounts with dedicated superblocks and thus separate device
851  * IDs.
852  */
853 struct fuse_mount {
854 	/* Underlying (potentially shared) connection to the FUSE server */
855 	struct fuse_conn *fc;
856 
857 	/*
858 	 * Super block for this connection (fc->killsb must be held when
859 	 * accessing this).
860 	 */
861 	struct super_block *sb;
862 
863 	/* Entry on fc->mounts */
864 	struct list_head fc_entry;
865 };
866 
867 static inline struct fuse_mount *get_fuse_mount_super(struct super_block *sb)
868 {
869 	return sb->s_fs_info;
870 }
871 
872 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
873 {
874 	return get_fuse_mount_super(sb)->fc;
875 }
876 
877 static inline struct fuse_mount *get_fuse_mount(struct inode *inode)
878 {
879 	return get_fuse_mount_super(inode->i_sb);
880 }
881 
882 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
883 {
884 	return get_fuse_mount_super(inode->i_sb)->fc;
885 }
886 
887 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
888 {
889 	return container_of(inode, struct fuse_inode, inode);
890 }
891 
892 static inline u64 get_node_id(struct inode *inode)
893 {
894 	return get_fuse_inode(inode)->nodeid;
895 }
896 
897 static inline int invalid_nodeid(u64 nodeid)
898 {
899 	return !nodeid || nodeid == FUSE_ROOT_ID;
900 }
901 
902 static inline u64 fuse_get_attr_version(struct fuse_conn *fc)
903 {
904 	return atomic64_read(&fc->attr_version);
905 }
906 
907 static inline bool fuse_stale_inode(const struct inode *inode, int generation,
908 				    struct fuse_attr *attr)
909 {
910 	return inode->i_generation != generation ||
911 		inode_wrong_type(inode, attr->mode);
912 }
913 
914 static inline void fuse_make_bad(struct inode *inode)
915 {
916 	remove_inode_hash(inode);
917 	set_bit(FUSE_I_BAD, &get_fuse_inode(inode)->state);
918 }
919 
920 static inline bool fuse_is_bad(struct inode *inode)
921 {
922 	return unlikely(test_bit(FUSE_I_BAD, &get_fuse_inode(inode)->state));
923 }
924 
925 static inline struct page **fuse_pages_alloc(unsigned int npages, gfp_t flags,
926 					     struct fuse_page_desc **desc)
927 {
928 	struct page **pages;
929 
930 	pages = kzalloc(npages * (sizeof(struct page *) +
931 				  sizeof(struct fuse_page_desc)), flags);
932 	*desc = (void *) (pages + npages);
933 
934 	return pages;
935 }
936 
937 static inline void fuse_page_descs_length_init(struct fuse_page_desc *descs,
938 					       unsigned int index,
939 					       unsigned int nr_pages)
940 {
941 	int i;
942 
943 	for (i = index; i < index + nr_pages; i++)
944 		descs[i].length = PAGE_SIZE - descs[i].offset;
945 }
946 
947 static inline void fuse_sync_bucket_dec(struct fuse_sync_bucket *bucket)
948 {
949 	/* Need RCU protection to prevent use after free after the decrement */
950 	rcu_read_lock();
951 	if (atomic_dec_and_test(&bucket->count))
952 		wake_up(&bucket->waitq);
953 	rcu_read_unlock();
954 }
955 
956 /** Device operations */
957 extern const struct file_operations fuse_dev_operations;
958 
959 extern const struct dentry_operations fuse_dentry_operations;
960 extern const struct dentry_operations fuse_root_dentry_operations;
961 
962 /**
963  * Get a filled in inode
964  */
965 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
966 			int generation, struct fuse_attr *attr,
967 			u64 attr_valid, u64 attr_version);
968 
969 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
970 		     struct fuse_entry_out *outarg, struct inode **inode);
971 
972 /**
973  * Send FORGET command
974  */
975 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
976 		       u64 nodeid, u64 nlookup);
977 
978 struct fuse_forget_link *fuse_alloc_forget(void);
979 
980 struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq,
981 					     unsigned int max,
982 					     unsigned int *countp);
983 
984 /*
985  * Initialize READ or READDIR request
986  */
987 struct fuse_io_args {
988 	union {
989 		struct {
990 			struct fuse_read_in in;
991 			u64 attr_ver;
992 		} read;
993 		struct {
994 			struct fuse_write_in in;
995 			struct fuse_write_out out;
996 			bool page_locked;
997 		} write;
998 	};
999 	struct fuse_args_pages ap;
1000 	struct fuse_io_priv *io;
1001 	struct fuse_file *ff;
1002 };
1003 
1004 void fuse_read_args_fill(struct fuse_io_args *ia, struct file *file, loff_t pos,
1005 			 size_t count, int opcode);
1006 
1007 
1008 /**
1009  * Send OPEN or OPENDIR request
1010  */
1011 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
1012 
1013 struct fuse_file *fuse_file_alloc(struct fuse_mount *fm);
1014 void fuse_file_free(struct fuse_file *ff);
1015 void fuse_finish_open(struct inode *inode, struct file *file);
1016 
1017 void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff,
1018 		       unsigned int flags);
1019 
1020 /**
1021  * Send RELEASE or RELEASEDIR request
1022  */
1023 void fuse_release_common(struct file *file, bool isdir);
1024 
1025 /**
1026  * Send FSYNC or FSYNCDIR request
1027  */
1028 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
1029 		      int datasync, int opcode);
1030 
1031 /**
1032  * Notify poll wakeup
1033  */
1034 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
1035 			    struct fuse_notify_poll_wakeup_out *outarg);
1036 
1037 /**
1038  * Initialize file operations on a regular file
1039  */
1040 void fuse_init_file_inode(struct inode *inode, unsigned int flags);
1041 
1042 /**
1043  * Initialize inode operations on regular files and special files
1044  */
1045 void fuse_init_common(struct inode *inode);
1046 
1047 /**
1048  * Initialize inode and file operations on a directory
1049  */
1050 void fuse_init_dir(struct inode *inode);
1051 
1052 /**
1053  * Initialize inode operations on a symlink
1054  */
1055 void fuse_init_symlink(struct inode *inode);
1056 
1057 /**
1058  * Change attributes of an inode
1059  */
1060 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
1061 			    u64 attr_valid, u64 attr_version);
1062 
1063 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
1064 				   u64 attr_valid, u32 cache_mask);
1065 
1066 u32 fuse_get_cache_mask(struct inode *inode);
1067 
1068 /**
1069  * Initialize the client device
1070  */
1071 int fuse_dev_init(void);
1072 
1073 /**
1074  * Cleanup the client device
1075  */
1076 void fuse_dev_cleanup(void);
1077 
1078 int fuse_ctl_init(void);
1079 void __exit fuse_ctl_cleanup(void);
1080 
1081 /**
1082  * Simple request sending that does request allocation and freeing
1083  */
1084 ssize_t fuse_simple_request(struct fuse_mount *fm, struct fuse_args *args);
1085 int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args,
1086 			   gfp_t gfp_flags);
1087 
1088 /**
1089  * End a finished request
1090  */
1091 void fuse_request_end(struct fuse_req *req);
1092 
1093 /* Abort all requests */
1094 void fuse_abort_conn(struct fuse_conn *fc);
1095 void fuse_wait_aborted(struct fuse_conn *fc);
1096 
1097 /**
1098  * Invalidate inode attributes
1099  */
1100 
1101 /* Attributes possibly changed on data modification */
1102 #define FUSE_STATX_MODIFY	(STATX_MTIME | STATX_CTIME | STATX_BLOCKS)
1103 
1104 /* Attributes possibly changed on data and/or size modification */
1105 #define FUSE_STATX_MODSIZE	(FUSE_STATX_MODIFY | STATX_SIZE)
1106 
1107 void fuse_invalidate_attr(struct inode *inode);
1108 void fuse_invalidate_attr_mask(struct inode *inode, u32 mask);
1109 
1110 void fuse_invalidate_entry_cache(struct dentry *entry);
1111 
1112 void fuse_invalidate_atime(struct inode *inode);
1113 
1114 u64 entry_attr_timeout(struct fuse_entry_out *o);
1115 void fuse_change_entry_timeout(struct dentry *entry, struct fuse_entry_out *o);
1116 
1117 /**
1118  * Acquire reference to fuse_conn
1119  */
1120 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
1121 
1122 /**
1123  * Initialize fuse_conn
1124  */
1125 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
1126 		    struct user_namespace *user_ns,
1127 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv);
1128 
1129 /**
1130  * Release reference to fuse_conn
1131  */
1132 void fuse_conn_put(struct fuse_conn *fc);
1133 
1134 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc);
1135 struct fuse_dev *fuse_dev_alloc(void);
1136 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc);
1137 void fuse_dev_free(struct fuse_dev *fud);
1138 void fuse_send_init(struct fuse_mount *fm);
1139 
1140 /**
1141  * Fill in superblock and initialize fuse connection
1142  * @sb: partially-initialized superblock to fill in
1143  * @ctx: mount context
1144  */
1145 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx);
1146 
1147 /*
1148  * Remove the mount from the connection
1149  *
1150  * Returns whether this was the last mount
1151  */
1152 bool fuse_mount_remove(struct fuse_mount *fm);
1153 
1154 /*
1155  * Setup context ops for submounts
1156  */
1157 int fuse_init_fs_context_submount(struct fs_context *fsc);
1158 
1159 /*
1160  * Shut down the connection (possibly sending DESTROY request).
1161  */
1162 void fuse_conn_destroy(struct fuse_mount *fm);
1163 
1164 /* Drop the connection and free the fuse mount */
1165 void fuse_mount_destroy(struct fuse_mount *fm);
1166 
1167 /**
1168  * Add connection to control filesystem
1169  */
1170 int fuse_ctl_add_conn(struct fuse_conn *fc);
1171 
1172 /**
1173  * Remove connection from control filesystem
1174  */
1175 void fuse_ctl_remove_conn(struct fuse_conn *fc);
1176 
1177 /**
1178  * Is file type valid?
1179  */
1180 int fuse_valid_type(int m);
1181 
1182 bool fuse_invalid_attr(struct fuse_attr *attr);
1183 
1184 /**
1185  * Is current process allowed to perform filesystem operation?
1186  */
1187 bool fuse_allow_current_process(struct fuse_conn *fc);
1188 
1189 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
1190 
1191 void fuse_flush_time_update(struct inode *inode);
1192 void fuse_update_ctime(struct inode *inode);
1193 
1194 int fuse_update_attributes(struct inode *inode, struct file *file, u32 mask);
1195 
1196 void fuse_flush_writepages(struct inode *inode);
1197 
1198 void fuse_set_nowrite(struct inode *inode);
1199 void fuse_release_nowrite(struct inode *inode);
1200 
1201 /**
1202  * Scan all fuse_mounts belonging to fc to find the first where
1203  * ilookup5() returns a result.  Return that result and the
1204  * respective fuse_mount in *fm (unless fm is NULL).
1205  *
1206  * The caller must hold fc->killsb.
1207  */
1208 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
1209 			   struct fuse_mount **fm);
1210 
1211 /**
1212  * File-system tells the kernel to invalidate cache for the given node id.
1213  */
1214 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
1215 			     loff_t offset, loff_t len);
1216 
1217 /**
1218  * File-system tells the kernel to invalidate parent attributes and
1219  * the dentry matching parent/name.
1220  *
1221  * If the child_nodeid is non-zero and:
1222  *    - matches the inode number for the dentry matching parent/name,
1223  *    - is not a mount point
1224  *    - is a file or oan empty directory
1225  * then the dentry is unhashed (d_delete()).
1226  */
1227 int fuse_reverse_inval_entry(struct fuse_conn *fc, u64 parent_nodeid,
1228 			     u64 child_nodeid, struct qstr *name, u32 flags);
1229 
1230 int fuse_do_open(struct fuse_mount *fm, u64 nodeid, struct file *file,
1231 		 bool isdir);
1232 
1233 /**
1234  * fuse_direct_io() flags
1235  */
1236 
1237 /** If set, it is WRITE; otherwise - READ */
1238 #define FUSE_DIO_WRITE (1 << 0)
1239 
1240 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
1241 #define FUSE_DIO_CUSE  (1 << 1)
1242 
1243 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1244 		       loff_t *ppos, int flags);
1245 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1246 		   unsigned int flags);
1247 long fuse_ioctl_common(struct file *file, unsigned int cmd,
1248 		       unsigned long arg, unsigned int flags);
1249 __poll_t fuse_file_poll(struct file *file, poll_table *wait);
1250 int fuse_dev_release(struct inode *inode, struct file *file);
1251 
1252 bool fuse_write_update_attr(struct inode *inode, loff_t pos, ssize_t written);
1253 
1254 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
1255 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
1256 
1257 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1258 		    struct file *file);
1259 
1260 void fuse_set_initialized(struct fuse_conn *fc);
1261 
1262 void fuse_unlock_inode(struct inode *inode, bool locked);
1263 bool fuse_lock_inode(struct inode *inode);
1264 
1265 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
1266 		  size_t size, int flags, unsigned int extra_flags);
1267 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
1268 		      size_t size);
1269 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
1270 int fuse_removexattr(struct inode *inode, const char *name);
1271 extern const struct xattr_handler *fuse_xattr_handlers[];
1272 
1273 struct posix_acl;
1274 struct posix_acl *fuse_get_inode_acl(struct inode *inode, int type, bool rcu);
1275 struct posix_acl *fuse_get_acl(struct mnt_idmap *idmap,
1276 			       struct dentry *dentry, int type);
1277 int fuse_set_acl(struct mnt_idmap *, struct dentry *dentry,
1278 		 struct posix_acl *acl, int type);
1279 
1280 /* readdir.c */
1281 int fuse_readdir(struct file *file, struct dir_context *ctx);
1282 
1283 /**
1284  * Return the number of bytes in an arguments list
1285  */
1286 unsigned int fuse_len_args(unsigned int numargs, struct fuse_arg *args);
1287 
1288 /**
1289  * Get the next unique ID for a request
1290  */
1291 u64 fuse_get_unique(struct fuse_iqueue *fiq);
1292 void fuse_free_conn(struct fuse_conn *fc);
1293 
1294 /* dax.c */
1295 
1296 #define FUSE_IS_DAX(inode) (IS_ENABLED(CONFIG_FUSE_DAX) && IS_DAX(inode))
1297 
1298 ssize_t fuse_dax_read_iter(struct kiocb *iocb, struct iov_iter *to);
1299 ssize_t fuse_dax_write_iter(struct kiocb *iocb, struct iov_iter *from);
1300 int fuse_dax_mmap(struct file *file, struct vm_area_struct *vma);
1301 int fuse_dax_break_layouts(struct inode *inode, u64 dmap_start, u64 dmap_end);
1302 int fuse_dax_conn_alloc(struct fuse_conn *fc, enum fuse_dax_mode mode,
1303 			struct dax_device *dax_dev);
1304 void fuse_dax_conn_free(struct fuse_conn *fc);
1305 bool fuse_dax_inode_alloc(struct super_block *sb, struct fuse_inode *fi);
1306 void fuse_dax_inode_init(struct inode *inode, unsigned int flags);
1307 void fuse_dax_inode_cleanup(struct inode *inode);
1308 void fuse_dax_dontcache(struct inode *inode, unsigned int flags);
1309 bool fuse_dax_check_alignment(struct fuse_conn *fc, unsigned int map_alignment);
1310 void fuse_dax_cancel_work(struct fuse_conn *fc);
1311 
1312 /* ioctl.c */
1313 long fuse_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
1314 long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
1315 			    unsigned long arg);
1316 int fuse_fileattr_get(struct dentry *dentry, struct fileattr *fa);
1317 int fuse_fileattr_set(struct mnt_idmap *idmap,
1318 		      struct dentry *dentry, struct fileattr *fa);
1319 
1320 /* file.c */
1321 
1322 struct fuse_file *fuse_file_open(struct fuse_mount *fm, u64 nodeid,
1323 				 unsigned int open_flags, bool isdir);
1324 void fuse_file_release(struct inode *inode, struct fuse_file *ff,
1325 		       unsigned int open_flags, fl_owner_t id, bool isdir);
1326 
1327 #endif /* _FS_FUSE_I_H */
1328