xref: /linux/fs/fuse/fuse_i.h (revision 64b14a184e83eb62ea0615e31a409956049d40e7)
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 	unsigned short in_numargs;
253 	unsigned short out_numargs;
254 	bool force:1;
255 	bool noreply:1;
256 	bool nocreds:1;
257 	bool in_pages:1;
258 	bool out_pages:1;
259 	bool out_argvar:1;
260 	bool page_zeroing:1;
261 	bool page_replace:1;
262 	bool may_block:1;
263 	struct fuse_in_arg in_args[3];
264 	struct fuse_arg out_args[2];
265 	void (*end)(struct fuse_mount *fm, struct fuse_args *args, int error);
266 };
267 
268 struct fuse_args_pages {
269 	struct fuse_args args;
270 	struct page **pages;
271 	struct fuse_page_desc *descs;
272 	unsigned int num_pages;
273 };
274 
275 #define FUSE_ARGS(args) struct fuse_args args = {}
276 
277 /** The request IO state (for asynchronous processing) */
278 struct fuse_io_priv {
279 	struct kref refcnt;
280 	int async;
281 	spinlock_t lock;
282 	unsigned reqs;
283 	ssize_t bytes;
284 	size_t size;
285 	__u64 offset;
286 	bool write;
287 	bool should_dirty;
288 	int err;
289 	struct kiocb *iocb;
290 	struct completion *done;
291 	bool blocking;
292 };
293 
294 #define FUSE_IO_PRIV_SYNC(i) \
295 {					\
296 	.refcnt = KREF_INIT(1),		\
297 	.async = 0,			\
298 	.iocb = i,			\
299 }
300 
301 /**
302  * Request flags
303  *
304  * FR_ISREPLY:		set if the request has reply
305  * FR_FORCE:		force sending of the request even if interrupted
306  * FR_BACKGROUND:	request is sent in the background
307  * FR_WAITING:		request is counted as "waiting"
308  * FR_ABORTED:		the request was aborted
309  * FR_INTERRUPTED:	the request has been interrupted
310  * FR_LOCKED:		data is being copied to/from the request
311  * FR_PENDING:		request is not yet in userspace
312  * FR_SENT:		request is in userspace, waiting for an answer
313  * FR_FINISHED:		request is finished
314  * FR_PRIVATE:		request is on private list
315  * FR_ASYNC:		request is asynchronous
316  */
317 enum fuse_req_flag {
318 	FR_ISREPLY,
319 	FR_FORCE,
320 	FR_BACKGROUND,
321 	FR_WAITING,
322 	FR_ABORTED,
323 	FR_INTERRUPTED,
324 	FR_LOCKED,
325 	FR_PENDING,
326 	FR_SENT,
327 	FR_FINISHED,
328 	FR_PRIVATE,
329 	FR_ASYNC,
330 };
331 
332 /**
333  * A request to the client
334  *
335  * .waitq.lock protects the following fields:
336  *   - FR_ABORTED
337  *   - FR_LOCKED (may also be modified under fc->lock, tested under both)
338  */
339 struct fuse_req {
340 	/** This can be on either pending processing or io lists in
341 	    fuse_conn */
342 	struct list_head list;
343 
344 	/** Entry on the interrupts list  */
345 	struct list_head intr_entry;
346 
347 	/* Input/output arguments */
348 	struct fuse_args *args;
349 
350 	/** refcount */
351 	refcount_t count;
352 
353 	/* Request flags, updated with test/set/clear_bit() */
354 	unsigned long flags;
355 
356 	/* The request input header */
357 	struct {
358 		struct fuse_in_header h;
359 	} in;
360 
361 	/* The request output header */
362 	struct {
363 		struct fuse_out_header h;
364 	} out;
365 
366 	/** Used to wake up the task waiting for completion of request*/
367 	wait_queue_head_t waitq;
368 
369 #if IS_ENABLED(CONFIG_VIRTIO_FS)
370 	/** virtio-fs's physically contiguous buffer for in and out args */
371 	void *argbuf;
372 #endif
373 
374 	/** fuse_mount this request belongs to */
375 	struct fuse_mount *fm;
376 };
377 
378 struct fuse_iqueue;
379 
380 /**
381  * Input queue callbacks
382  *
383  * Input queue signalling is device-specific.  For example, the /dev/fuse file
384  * uses fiq->waitq and fasync to wake processes that are waiting on queue
385  * readiness.  These callbacks allow other device types to respond to input
386  * queue activity.
387  */
388 struct fuse_iqueue_ops {
389 	/**
390 	 * Signal that a forget has been queued
391 	 */
392 	void (*wake_forget_and_unlock)(struct fuse_iqueue *fiq)
393 		__releases(fiq->lock);
394 
395 	/**
396 	 * Signal that an INTERRUPT request has been queued
397 	 */
398 	void (*wake_interrupt_and_unlock)(struct fuse_iqueue *fiq)
399 		__releases(fiq->lock);
400 
401 	/**
402 	 * Signal that a request has been queued
403 	 */
404 	void (*wake_pending_and_unlock)(struct fuse_iqueue *fiq)
405 		__releases(fiq->lock);
406 
407 	/**
408 	 * Clean up when fuse_iqueue is destroyed
409 	 */
410 	void (*release)(struct fuse_iqueue *fiq);
411 };
412 
413 /** /dev/fuse input queue operations */
414 extern const struct fuse_iqueue_ops fuse_dev_fiq_ops;
415 
416 struct fuse_iqueue {
417 	/** Connection established */
418 	unsigned connected;
419 
420 	/** Lock protecting accesses to members of this structure */
421 	spinlock_t lock;
422 
423 	/** Readers of the connection are waiting on this */
424 	wait_queue_head_t waitq;
425 
426 	/** The next unique request id */
427 	u64 reqctr;
428 
429 	/** The list of pending requests */
430 	struct list_head pending;
431 
432 	/** Pending interrupts */
433 	struct list_head interrupts;
434 
435 	/** Queue of pending forgets */
436 	struct fuse_forget_link forget_list_head;
437 	struct fuse_forget_link *forget_list_tail;
438 
439 	/** Batching of FORGET requests (positive indicates FORGET batch) */
440 	int forget_batch;
441 
442 	/** O_ASYNC requests */
443 	struct fasync_struct *fasync;
444 
445 	/** Device-specific callbacks */
446 	const struct fuse_iqueue_ops *ops;
447 
448 	/** Device-specific state */
449 	void *priv;
450 };
451 
452 #define FUSE_PQ_HASH_BITS 8
453 #define FUSE_PQ_HASH_SIZE (1 << FUSE_PQ_HASH_BITS)
454 
455 struct fuse_pqueue {
456 	/** Connection established */
457 	unsigned connected;
458 
459 	/** Lock protecting accessess to  members of this structure */
460 	spinlock_t lock;
461 
462 	/** Hash table of requests being processed */
463 	struct list_head *processing;
464 
465 	/** The list of requests under I/O */
466 	struct list_head io;
467 };
468 
469 /**
470  * Fuse device instance
471  */
472 struct fuse_dev {
473 	/** Fuse connection for this device */
474 	struct fuse_conn *fc;
475 
476 	/** Processing queue */
477 	struct fuse_pqueue pq;
478 
479 	/** list entry on fc->devices */
480 	struct list_head entry;
481 };
482 
483 enum fuse_dax_mode {
484 	FUSE_DAX_INODE_DEFAULT,	/* default */
485 	FUSE_DAX_ALWAYS,	/* "-o dax=always" */
486 	FUSE_DAX_NEVER,		/* "-o dax=never" */
487 	FUSE_DAX_INODE_USER,	/* "-o dax=inode" */
488 };
489 
490 static inline bool fuse_is_inode_dax_mode(enum fuse_dax_mode mode)
491 {
492 	return mode == FUSE_DAX_INODE_DEFAULT || mode == FUSE_DAX_INODE_USER;
493 }
494 
495 struct fuse_fs_context {
496 	int fd;
497 	struct file *file;
498 	unsigned int rootmode;
499 	kuid_t user_id;
500 	kgid_t group_id;
501 	bool is_bdev:1;
502 	bool fd_present:1;
503 	bool rootmode_present:1;
504 	bool user_id_present:1;
505 	bool group_id_present:1;
506 	bool default_permissions:1;
507 	bool allow_other:1;
508 	bool destroy:1;
509 	bool no_control:1;
510 	bool no_force_umount:1;
511 	bool legacy_opts_show:1;
512 	enum fuse_dax_mode dax_mode;
513 	unsigned int max_read;
514 	unsigned int blksize;
515 	const char *subtype;
516 
517 	/* DAX device, may be NULL */
518 	struct dax_device *dax_dev;
519 
520 	/* fuse_dev pointer to fill in, should contain NULL on entry */
521 	void **fudptr;
522 };
523 
524 struct fuse_sync_bucket {
525 	/* count is a possible scalability bottleneck */
526 	atomic_t count;
527 	wait_queue_head_t waitq;
528 	struct rcu_head rcu;
529 };
530 
531 /**
532  * A Fuse connection.
533  *
534  * This structure is created, when the root filesystem is mounted, and
535  * is destroyed, when the client device is closed and the last
536  * fuse_mount is destroyed.
537  */
538 struct fuse_conn {
539 	/** Lock protecting accessess to  members of this structure */
540 	spinlock_t lock;
541 
542 	/** Refcount */
543 	refcount_t count;
544 
545 	/** Number of fuse_dev's */
546 	atomic_t dev_count;
547 
548 	struct rcu_head rcu;
549 
550 	/** The user id for this mount */
551 	kuid_t user_id;
552 
553 	/** The group id for this mount */
554 	kgid_t group_id;
555 
556 	/** The pid namespace for this mount */
557 	struct pid_namespace *pid_ns;
558 
559 	/** The user namespace for this mount */
560 	struct user_namespace *user_ns;
561 
562 	/** Maximum read size */
563 	unsigned max_read;
564 
565 	/** Maximum write size */
566 	unsigned max_write;
567 
568 	/** Maximum number of pages that can be used in a single request */
569 	unsigned int max_pages;
570 
571 	/** Constrain ->max_pages to this value during feature negotiation */
572 	unsigned int max_pages_limit;
573 
574 	/** Input queue */
575 	struct fuse_iqueue iq;
576 
577 	/** The next unique kernel file handle */
578 	atomic64_t khctr;
579 
580 	/** rbtree of fuse_files waiting for poll events indexed by ph */
581 	struct rb_root polled_files;
582 
583 	/** Maximum number of outstanding background requests */
584 	unsigned max_background;
585 
586 	/** Number of background requests at which congestion starts */
587 	unsigned congestion_threshold;
588 
589 	/** Number of requests currently in the background */
590 	unsigned num_background;
591 
592 	/** Number of background requests currently queued for userspace */
593 	unsigned active_background;
594 
595 	/** The list of background requests set aside for later queuing */
596 	struct list_head bg_queue;
597 
598 	/** Protects: max_background, congestion_threshold, num_background,
599 	 * active_background, bg_queue, blocked */
600 	spinlock_t bg_lock;
601 
602 	/** Flag indicating that INIT reply has been received. Allocating
603 	 * any fuse request will be suspended until the flag is set */
604 	int initialized;
605 
606 	/** Flag indicating if connection is blocked.  This will be
607 	    the case before the INIT reply is received, and if there
608 	    are too many outstading backgrounds requests */
609 	int blocked;
610 
611 	/** waitq for blocked connection */
612 	wait_queue_head_t blocked_waitq;
613 
614 	/** Connection established, cleared on umount, connection
615 	    abort and device release */
616 	unsigned connected;
617 
618 	/** Connection aborted via sysfs */
619 	bool aborted;
620 
621 	/** Connection failed (version mismatch).  Cannot race with
622 	    setting other bitfields since it is only set once in INIT
623 	    reply, before any other request, and never cleared */
624 	unsigned conn_error:1;
625 
626 	/** Connection successful.  Only set in INIT */
627 	unsigned conn_init:1;
628 
629 	/** Do readpages asynchronously?  Only set in INIT */
630 	unsigned async_read:1;
631 
632 	/** Return an unique read error after abort.  Only set in INIT */
633 	unsigned abort_err:1;
634 
635 	/** Do not send separate SETATTR request before open(O_TRUNC)  */
636 	unsigned atomic_o_trunc:1;
637 
638 	/** Filesystem supports NFS exporting.  Only set in INIT */
639 	unsigned export_support:1;
640 
641 	/** write-back cache policy (default is write-through) */
642 	unsigned writeback_cache:1;
643 
644 	/** allow parallel lookups and readdir (default is serialized) */
645 	unsigned parallel_dirops:1;
646 
647 	/** handle fs handles killing suid/sgid/cap on write/chown/trunc */
648 	unsigned handle_killpriv:1;
649 
650 	/** cache READLINK responses in page cache */
651 	unsigned cache_symlinks:1;
652 
653 	/* show legacy mount options */
654 	unsigned int legacy_opts_show:1;
655 
656 	/*
657 	 * fs kills suid/sgid/cap on write/chown/trunc. suid is killed on
658 	 * write/trunc only if caller did not have CAP_FSETID.  sgid is killed
659 	 * on write/truncate only if caller did not have CAP_FSETID as well as
660 	 * file has group execute permission.
661 	 */
662 	unsigned handle_killpriv_v2:1;
663 
664 	/*
665 	 * The following bitfields are only for optimization purposes
666 	 * and hence races in setting them will not cause malfunction
667 	 */
668 
669 	/** Is open/release not implemented by fs? */
670 	unsigned no_open:1;
671 
672 	/** Is opendir/releasedir not implemented by fs? */
673 	unsigned no_opendir:1;
674 
675 	/** Is fsync not implemented by fs? */
676 	unsigned no_fsync:1;
677 
678 	/** Is fsyncdir not implemented by fs? */
679 	unsigned no_fsyncdir:1;
680 
681 	/** Is flush not implemented by fs? */
682 	unsigned no_flush:1;
683 
684 	/** Is setxattr not implemented by fs? */
685 	unsigned no_setxattr:1;
686 
687 	/** Does file server support extended setxattr */
688 	unsigned setxattr_ext:1;
689 
690 	/** Is getxattr not implemented by fs? */
691 	unsigned no_getxattr:1;
692 
693 	/** Is listxattr not implemented by fs? */
694 	unsigned no_listxattr:1;
695 
696 	/** Is removexattr not implemented by fs? */
697 	unsigned no_removexattr:1;
698 
699 	/** Are posix file locking primitives not implemented by fs? */
700 	unsigned no_lock:1;
701 
702 	/** Is access not implemented by fs? */
703 	unsigned no_access:1;
704 
705 	/** Is create not implemented by fs? */
706 	unsigned no_create:1;
707 
708 	/** Is interrupt not implemented by fs? */
709 	unsigned no_interrupt:1;
710 
711 	/** Is bmap not implemented by fs? */
712 	unsigned no_bmap:1;
713 
714 	/** Is poll not implemented by fs? */
715 	unsigned no_poll:1;
716 
717 	/** Do multi-page cached writes */
718 	unsigned big_writes:1;
719 
720 	/** Don't apply umask to creation modes */
721 	unsigned dont_mask:1;
722 
723 	/** Are BSD file locking primitives not implemented by fs? */
724 	unsigned no_flock:1;
725 
726 	/** Is fallocate not implemented by fs? */
727 	unsigned no_fallocate:1;
728 
729 	/** Is rename with flags implemented by fs? */
730 	unsigned no_rename2:1;
731 
732 	/** Use enhanced/automatic page cache invalidation. */
733 	unsigned auto_inval_data:1;
734 
735 	/** Filesystem is fully responsible for page cache invalidation. */
736 	unsigned explicit_inval_data:1;
737 
738 	/** Does the filesystem support readdirplus? */
739 	unsigned do_readdirplus:1;
740 
741 	/** Does the filesystem want adaptive readdirplus? */
742 	unsigned readdirplus_auto:1;
743 
744 	/** Does the filesystem support asynchronous direct-IO submission? */
745 	unsigned async_dio:1;
746 
747 	/** Is lseek not implemented by fs? */
748 	unsigned no_lseek:1;
749 
750 	/** Does the filesystem support posix acls? */
751 	unsigned posix_acl:1;
752 
753 	/** Check permissions based on the file mode or not? */
754 	unsigned default_permissions:1;
755 
756 	/** Allow other than the mounter user to access the filesystem ? */
757 	unsigned allow_other:1;
758 
759 	/** Does the filesystem support copy_file_range? */
760 	unsigned no_copy_file_range:1;
761 
762 	/* Send DESTROY request */
763 	unsigned int destroy:1;
764 
765 	/* Delete dentries that have gone stale */
766 	unsigned int delete_stale:1;
767 
768 	/** Do not create entry in fusectl fs */
769 	unsigned int no_control:1;
770 
771 	/** Do not allow MNT_FORCE umount */
772 	unsigned int no_force_umount:1;
773 
774 	/* Auto-mount submounts announced by the server */
775 	unsigned int auto_submounts:1;
776 
777 	/* Propagate syncfs() to server */
778 	unsigned int sync_fs:1;
779 
780 	/* Initialize security xattrs when creating a new inode */
781 	unsigned int init_security:1;
782 
783 	/* Does the filesystem support per inode DAX? */
784 	unsigned int inode_dax:1;
785 
786 	/** The number of requests waiting for completion */
787 	atomic_t num_waiting;
788 
789 	/** Negotiated minor version */
790 	unsigned minor;
791 
792 	/** Entry on the fuse_mount_list */
793 	struct list_head entry;
794 
795 	/** Device ID from the root super block */
796 	dev_t dev;
797 
798 	/** Dentries in the control filesystem */
799 	struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
800 
801 	/** number of dentries used in the above array */
802 	int ctl_ndents;
803 
804 	/** Key for lock owner ID scrambling */
805 	u32 scramble_key[4];
806 
807 	/** Version counter for attribute changes */
808 	atomic64_t attr_version;
809 
810 	/** Called on final put */
811 	void (*release)(struct fuse_conn *);
812 
813 	/**
814 	 * Read/write semaphore to hold when accessing the sb of any
815 	 * fuse_mount belonging to this connection
816 	 */
817 	struct rw_semaphore killsb;
818 
819 	/** List of device instances belonging to this connection */
820 	struct list_head devices;
821 
822 #ifdef CONFIG_FUSE_DAX
823 	/* Dax mode */
824 	enum fuse_dax_mode dax_mode;
825 
826 	/* Dax specific conn data, non-NULL if DAX is enabled */
827 	struct fuse_conn_dax *dax;
828 #endif
829 
830 	/** List of filesystems using this connection */
831 	struct list_head mounts;
832 
833 	/* New writepages go into this bucket */
834 	struct fuse_sync_bucket __rcu *curr_bucket;
835 };
836 
837 /*
838  * Represents a mounted filesystem, potentially a submount.
839  *
840  * This object allows sharing a fuse_conn between separate mounts to
841  * allow submounts with dedicated superblocks and thus separate device
842  * IDs.
843  */
844 struct fuse_mount {
845 	/* Underlying (potentially shared) connection to the FUSE server */
846 	struct fuse_conn *fc;
847 
848 	/*
849 	 * Super block for this connection (fc->killsb must be held when
850 	 * accessing this).
851 	 */
852 	struct super_block *sb;
853 
854 	/* Entry on fc->mounts */
855 	struct list_head fc_entry;
856 };
857 
858 static inline struct fuse_mount *get_fuse_mount_super(struct super_block *sb)
859 {
860 	return sb->s_fs_info;
861 }
862 
863 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
864 {
865 	return get_fuse_mount_super(sb)->fc;
866 }
867 
868 static inline struct fuse_mount *get_fuse_mount(struct inode *inode)
869 {
870 	return get_fuse_mount_super(inode->i_sb);
871 }
872 
873 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
874 {
875 	return get_fuse_mount_super(inode->i_sb)->fc;
876 }
877 
878 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
879 {
880 	return container_of(inode, struct fuse_inode, inode);
881 }
882 
883 static inline u64 get_node_id(struct inode *inode)
884 {
885 	return get_fuse_inode(inode)->nodeid;
886 }
887 
888 static inline int invalid_nodeid(u64 nodeid)
889 {
890 	return !nodeid || nodeid == FUSE_ROOT_ID;
891 }
892 
893 static inline u64 fuse_get_attr_version(struct fuse_conn *fc)
894 {
895 	return atomic64_read(&fc->attr_version);
896 }
897 
898 static inline bool fuse_stale_inode(const struct inode *inode, int generation,
899 				    struct fuse_attr *attr)
900 {
901 	return inode->i_generation != generation ||
902 		inode_wrong_type(inode, attr->mode);
903 }
904 
905 static inline void fuse_make_bad(struct inode *inode)
906 {
907 	remove_inode_hash(inode);
908 	set_bit(FUSE_I_BAD, &get_fuse_inode(inode)->state);
909 }
910 
911 static inline bool fuse_is_bad(struct inode *inode)
912 {
913 	return unlikely(test_bit(FUSE_I_BAD, &get_fuse_inode(inode)->state));
914 }
915 
916 static inline struct page **fuse_pages_alloc(unsigned int npages, gfp_t flags,
917 					     struct fuse_page_desc **desc)
918 {
919 	struct page **pages;
920 
921 	pages = kzalloc(npages * (sizeof(struct page *) +
922 				  sizeof(struct fuse_page_desc)), flags);
923 	*desc = (void *) (pages + npages);
924 
925 	return pages;
926 }
927 
928 static inline void fuse_page_descs_length_init(struct fuse_page_desc *descs,
929 					       unsigned int index,
930 					       unsigned int nr_pages)
931 {
932 	int i;
933 
934 	for (i = index; i < index + nr_pages; i++)
935 		descs[i].length = PAGE_SIZE - descs[i].offset;
936 }
937 
938 static inline void fuse_sync_bucket_dec(struct fuse_sync_bucket *bucket)
939 {
940 	/* Need RCU protection to prevent use after free after the decrement */
941 	rcu_read_lock();
942 	if (atomic_dec_and_test(&bucket->count))
943 		wake_up(&bucket->waitq);
944 	rcu_read_unlock();
945 }
946 
947 /** Device operations */
948 extern const struct file_operations fuse_dev_operations;
949 
950 extern const struct dentry_operations fuse_dentry_operations;
951 extern const struct dentry_operations fuse_root_dentry_operations;
952 
953 /**
954  * Get a filled in inode
955  */
956 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
957 			int generation, struct fuse_attr *attr,
958 			u64 attr_valid, u64 attr_version);
959 
960 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
961 		     struct fuse_entry_out *outarg, struct inode **inode);
962 
963 /**
964  * Send FORGET command
965  */
966 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
967 		       u64 nodeid, u64 nlookup);
968 
969 struct fuse_forget_link *fuse_alloc_forget(void);
970 
971 struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq,
972 					     unsigned int max,
973 					     unsigned int *countp);
974 
975 /*
976  * Initialize READ or READDIR request
977  */
978 struct fuse_io_args {
979 	union {
980 		struct {
981 			struct fuse_read_in in;
982 			u64 attr_ver;
983 		} read;
984 		struct {
985 			struct fuse_write_in in;
986 			struct fuse_write_out out;
987 			bool page_locked;
988 		} write;
989 	};
990 	struct fuse_args_pages ap;
991 	struct fuse_io_priv *io;
992 	struct fuse_file *ff;
993 };
994 
995 void fuse_read_args_fill(struct fuse_io_args *ia, struct file *file, loff_t pos,
996 			 size_t count, int opcode);
997 
998 
999 /**
1000  * Send OPEN or OPENDIR request
1001  */
1002 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
1003 
1004 struct fuse_file *fuse_file_alloc(struct fuse_mount *fm);
1005 void fuse_file_free(struct fuse_file *ff);
1006 void fuse_finish_open(struct inode *inode, struct file *file);
1007 
1008 void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff,
1009 		       unsigned int flags);
1010 
1011 /**
1012  * Send RELEASE or RELEASEDIR request
1013  */
1014 void fuse_release_common(struct file *file, bool isdir);
1015 
1016 /**
1017  * Send FSYNC or FSYNCDIR request
1018  */
1019 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
1020 		      int datasync, int opcode);
1021 
1022 /**
1023  * Notify poll wakeup
1024  */
1025 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
1026 			    struct fuse_notify_poll_wakeup_out *outarg);
1027 
1028 /**
1029  * Initialize file operations on a regular file
1030  */
1031 void fuse_init_file_inode(struct inode *inode, unsigned int flags);
1032 
1033 /**
1034  * Initialize inode operations on regular files and special files
1035  */
1036 void fuse_init_common(struct inode *inode);
1037 
1038 /**
1039  * Initialize inode and file operations on a directory
1040  */
1041 void fuse_init_dir(struct inode *inode);
1042 
1043 /**
1044  * Initialize inode operations on a symlink
1045  */
1046 void fuse_init_symlink(struct inode *inode);
1047 
1048 /**
1049  * Change attributes of an inode
1050  */
1051 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
1052 			    u64 attr_valid, u64 attr_version);
1053 
1054 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
1055 				   u64 attr_valid, u32 cache_mask);
1056 
1057 u32 fuse_get_cache_mask(struct inode *inode);
1058 
1059 /**
1060  * Initialize the client device
1061  */
1062 int fuse_dev_init(void);
1063 
1064 /**
1065  * Cleanup the client device
1066  */
1067 void fuse_dev_cleanup(void);
1068 
1069 int fuse_ctl_init(void);
1070 void __exit fuse_ctl_cleanup(void);
1071 
1072 /**
1073  * Simple request sending that does request allocation and freeing
1074  */
1075 ssize_t fuse_simple_request(struct fuse_mount *fm, struct fuse_args *args);
1076 int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args,
1077 			   gfp_t gfp_flags);
1078 
1079 /**
1080  * End a finished request
1081  */
1082 void fuse_request_end(struct fuse_req *req);
1083 
1084 /* Abort all requests */
1085 void fuse_abort_conn(struct fuse_conn *fc);
1086 void fuse_wait_aborted(struct fuse_conn *fc);
1087 
1088 /**
1089  * Invalidate inode attributes
1090  */
1091 
1092 /* Attributes possibly changed on data modification */
1093 #define FUSE_STATX_MODIFY	(STATX_MTIME | STATX_CTIME | STATX_BLOCKS)
1094 
1095 /* Attributes possibly changed on data and/or size modification */
1096 #define FUSE_STATX_MODSIZE	(FUSE_STATX_MODIFY | STATX_SIZE)
1097 
1098 void fuse_invalidate_attr(struct inode *inode);
1099 void fuse_invalidate_attr_mask(struct inode *inode, u32 mask);
1100 
1101 void fuse_invalidate_entry_cache(struct dentry *entry);
1102 
1103 void fuse_invalidate_atime(struct inode *inode);
1104 
1105 u64 entry_attr_timeout(struct fuse_entry_out *o);
1106 void fuse_change_entry_timeout(struct dentry *entry, struct fuse_entry_out *o);
1107 
1108 /**
1109  * Acquire reference to fuse_conn
1110  */
1111 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
1112 
1113 /**
1114  * Initialize fuse_conn
1115  */
1116 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
1117 		    struct user_namespace *user_ns,
1118 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv);
1119 
1120 /**
1121  * Release reference to fuse_conn
1122  */
1123 void fuse_conn_put(struct fuse_conn *fc);
1124 
1125 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc);
1126 struct fuse_dev *fuse_dev_alloc(void);
1127 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc);
1128 void fuse_dev_free(struct fuse_dev *fud);
1129 void fuse_send_init(struct fuse_mount *fm);
1130 
1131 /**
1132  * Fill in superblock and initialize fuse connection
1133  * @sb: partially-initialized superblock to fill in
1134  * @ctx: mount context
1135  */
1136 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx);
1137 
1138 /*
1139  * Remove the mount from the connection
1140  *
1141  * Returns whether this was the last mount
1142  */
1143 bool fuse_mount_remove(struct fuse_mount *fm);
1144 
1145 /*
1146  * Setup context ops for submounts
1147  */
1148 int fuse_init_fs_context_submount(struct fs_context *fsc);
1149 
1150 /*
1151  * Shut down the connection (possibly sending DESTROY request).
1152  */
1153 void fuse_conn_destroy(struct fuse_mount *fm);
1154 
1155 /* Drop the connection and free the fuse mount */
1156 void fuse_mount_destroy(struct fuse_mount *fm);
1157 
1158 /**
1159  * Add connection to control filesystem
1160  */
1161 int fuse_ctl_add_conn(struct fuse_conn *fc);
1162 
1163 /**
1164  * Remove connection from control filesystem
1165  */
1166 void fuse_ctl_remove_conn(struct fuse_conn *fc);
1167 
1168 /**
1169  * Is file type valid?
1170  */
1171 int fuse_valid_type(int m);
1172 
1173 bool fuse_invalid_attr(struct fuse_attr *attr);
1174 
1175 /**
1176  * Is current process allowed to perform filesystem operation?
1177  */
1178 int fuse_allow_current_process(struct fuse_conn *fc);
1179 
1180 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
1181 
1182 void fuse_flush_time_update(struct inode *inode);
1183 void fuse_update_ctime(struct inode *inode);
1184 
1185 int fuse_update_attributes(struct inode *inode, struct file *file, u32 mask);
1186 
1187 void fuse_flush_writepages(struct inode *inode);
1188 
1189 void fuse_set_nowrite(struct inode *inode);
1190 void fuse_release_nowrite(struct inode *inode);
1191 
1192 /**
1193  * Scan all fuse_mounts belonging to fc to find the first where
1194  * ilookup5() returns a result.  Return that result and the
1195  * respective fuse_mount in *fm (unless fm is NULL).
1196  *
1197  * The caller must hold fc->killsb.
1198  */
1199 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
1200 			   struct fuse_mount **fm);
1201 
1202 /**
1203  * File-system tells the kernel to invalidate cache for the given node id.
1204  */
1205 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
1206 			     loff_t offset, loff_t len);
1207 
1208 /**
1209  * File-system tells the kernel to invalidate parent attributes and
1210  * the dentry matching parent/name.
1211  *
1212  * If the child_nodeid is non-zero and:
1213  *    - matches the inode number for the dentry matching parent/name,
1214  *    - is not a mount point
1215  *    - is a file or oan empty directory
1216  * then the dentry is unhashed (d_delete()).
1217  */
1218 int fuse_reverse_inval_entry(struct fuse_conn *fc, u64 parent_nodeid,
1219 			     u64 child_nodeid, struct qstr *name);
1220 
1221 int fuse_do_open(struct fuse_mount *fm, u64 nodeid, struct file *file,
1222 		 bool isdir);
1223 
1224 /**
1225  * fuse_direct_io() flags
1226  */
1227 
1228 /** If set, it is WRITE; otherwise - READ */
1229 #define FUSE_DIO_WRITE (1 << 0)
1230 
1231 /** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
1232 #define FUSE_DIO_CUSE  (1 << 1)
1233 
1234 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1235 		       loff_t *ppos, int flags);
1236 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1237 		   unsigned int flags);
1238 long fuse_ioctl_common(struct file *file, unsigned int cmd,
1239 		       unsigned long arg, unsigned int flags);
1240 __poll_t fuse_file_poll(struct file *file, poll_table *wait);
1241 int fuse_dev_release(struct inode *inode, struct file *file);
1242 
1243 bool fuse_write_update_attr(struct inode *inode, loff_t pos, ssize_t written);
1244 
1245 int fuse_flush_times(struct inode *inode, struct fuse_file *ff);
1246 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc);
1247 
1248 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1249 		    struct file *file);
1250 
1251 void fuse_set_initialized(struct fuse_conn *fc);
1252 
1253 void fuse_unlock_inode(struct inode *inode, bool locked);
1254 bool fuse_lock_inode(struct inode *inode);
1255 
1256 int fuse_setxattr(struct inode *inode, const char *name, const void *value,
1257 		  size_t size, int flags, unsigned int extra_flags);
1258 ssize_t fuse_getxattr(struct inode *inode, const char *name, void *value,
1259 		      size_t size);
1260 ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size);
1261 int fuse_removexattr(struct inode *inode, const char *name);
1262 extern const struct xattr_handler *fuse_xattr_handlers[];
1263 extern const struct xattr_handler *fuse_acl_xattr_handlers[];
1264 extern const struct xattr_handler *fuse_no_acl_xattr_handlers[];
1265 
1266 struct posix_acl;
1267 struct posix_acl *fuse_get_acl(struct inode *inode, int type, bool rcu);
1268 int fuse_set_acl(struct user_namespace *mnt_userns, struct inode *inode,
1269 		 struct posix_acl *acl, int type);
1270 
1271 /* readdir.c */
1272 int fuse_readdir(struct file *file, struct dir_context *ctx);
1273 
1274 /**
1275  * Return the number of bytes in an arguments list
1276  */
1277 unsigned int fuse_len_args(unsigned int numargs, struct fuse_arg *args);
1278 
1279 /**
1280  * Get the next unique ID for a request
1281  */
1282 u64 fuse_get_unique(struct fuse_iqueue *fiq);
1283 void fuse_free_conn(struct fuse_conn *fc);
1284 
1285 /* dax.c */
1286 
1287 #define FUSE_IS_DAX(inode) (IS_ENABLED(CONFIG_FUSE_DAX) && IS_DAX(inode))
1288 
1289 ssize_t fuse_dax_read_iter(struct kiocb *iocb, struct iov_iter *to);
1290 ssize_t fuse_dax_write_iter(struct kiocb *iocb, struct iov_iter *from);
1291 int fuse_dax_mmap(struct file *file, struct vm_area_struct *vma);
1292 int fuse_dax_break_layouts(struct inode *inode, u64 dmap_start, u64 dmap_end);
1293 int fuse_dax_conn_alloc(struct fuse_conn *fc, enum fuse_dax_mode mode,
1294 			struct dax_device *dax_dev);
1295 void fuse_dax_conn_free(struct fuse_conn *fc);
1296 bool fuse_dax_inode_alloc(struct super_block *sb, struct fuse_inode *fi);
1297 void fuse_dax_inode_init(struct inode *inode, unsigned int flags);
1298 void fuse_dax_inode_cleanup(struct inode *inode);
1299 void fuse_dax_dontcache(struct inode *inode, unsigned int flags);
1300 bool fuse_dax_check_alignment(struct fuse_conn *fc, unsigned int map_alignment);
1301 void fuse_dax_cancel_work(struct fuse_conn *fc);
1302 
1303 /* ioctl.c */
1304 long fuse_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
1305 long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
1306 			    unsigned long arg);
1307 int fuse_fileattr_get(struct dentry *dentry, struct fileattr *fa);
1308 int fuse_fileattr_set(struct user_namespace *mnt_userns,
1309 		      struct dentry *dentry, struct fileattr *fa);
1310 
1311 /* file.c */
1312 
1313 struct fuse_file *fuse_file_open(struct fuse_mount *fm, u64 nodeid,
1314 				 unsigned int open_flags, bool isdir);
1315 void fuse_file_release(struct inode *inode, struct fuse_file *ff,
1316 		       unsigned int open_flags, fl_owner_t id, bool isdir);
1317 
1318 #endif /* _FS_FUSE_I_H */
1319