xref: /linux/fs/fuse/fuse_i.h (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 #include <linux/fuse.h>
13 #include <linux/fs.h>
14 #include <linux/mount.h>
15 #include <linux/wait.h>
16 #include <linux/list.h>
17 #include <linux/spinlock.h>
18 #include <linux/mm.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 
26 /** Max number of pages that can be used in a single read request */
27 #define FUSE_MAX_PAGES_PER_REQ 32
28 
29 /** Bias for fi->writectr, meaning new writepages must not be sent */
30 #define FUSE_NOWRITE INT_MIN
31 
32 /** It could be as large as PATH_MAX, but would that have any uses? */
33 #define FUSE_NAME_MAX 1024
34 
35 /** Number of dentries for each connection in the control filesystem */
36 #define FUSE_CTL_NUM_DENTRIES 5
37 
38 /** If the FUSE_DEFAULT_PERMISSIONS flag is given, the filesystem
39     module will check permissions based on the file mode.  Otherwise no
40     permission checking is done in the kernel */
41 #define FUSE_DEFAULT_PERMISSIONS (1 << 0)
42 
43 /** If the FUSE_ALLOW_OTHER flag is given, then not only the user
44     doing the mount will be allowed to access the filesystem */
45 #define FUSE_ALLOW_OTHER         (1 << 1)
46 
47 /** List of active connections */
48 extern struct list_head fuse_conn_list;
49 
50 /** Global mutex protecting fuse_conn_list and the control filesystem */
51 extern struct mutex fuse_mutex;
52 
53 /** Module parameters */
54 extern unsigned max_user_bgreq;
55 extern unsigned max_user_congthresh;
56 
57 /* One forget request */
58 struct fuse_forget_link {
59 	struct fuse_forget_one forget_one;
60 	struct fuse_forget_link *next;
61 };
62 
63 /** FUSE inode */
64 struct fuse_inode {
65 	/** Inode data */
66 	struct inode inode;
67 
68 	/** Unique ID, which identifies the inode between userspace
69 	 * and kernel */
70 	u64 nodeid;
71 
72 	/** Number of lookups on this inode */
73 	u64 nlookup;
74 
75 	/** The request used for sending the FORGET message */
76 	struct fuse_forget_link *forget;
77 
78 	/** Time in jiffies until the file attributes are valid */
79 	u64 i_time;
80 
81 	/** The sticky bit in inode->i_mode may have been removed, so
82 	    preserve the original mode */
83 	umode_t orig_i_mode;
84 
85 	/** 64 bit inode number */
86 	u64 orig_ino;
87 
88 	/** Version of last attribute change */
89 	u64 attr_version;
90 
91 	/** Files usable in writepage.  Protected by fc->lock */
92 	struct list_head write_files;
93 
94 	/** Writepages pending on truncate or fsync */
95 	struct list_head queued_writes;
96 
97 	/** Number of sent writes, a negative bias (FUSE_NOWRITE)
98 	 * means more writes are blocked */
99 	int writectr;
100 
101 	/** Waitq for writepage completion */
102 	wait_queue_head_t page_waitq;
103 
104 	/** List of writepage requestst (pending or sent) */
105 	struct list_head writepages;
106 };
107 
108 struct fuse_conn;
109 
110 /** FUSE specific file data */
111 struct fuse_file {
112 	/** Fuse connection for this file */
113 	struct fuse_conn *fc;
114 
115 	/** Request reserved for flush and release */
116 	struct fuse_req *reserved_req;
117 
118 	/** Kernel file handle guaranteed to be unique */
119 	u64 kh;
120 
121 	/** File handle used by userspace */
122 	u64 fh;
123 
124 	/** Node id of this file */
125 	u64 nodeid;
126 
127 	/** Refcount */
128 	atomic_t count;
129 
130 	/** FOPEN_* flags returned by open */
131 	u32 open_flags;
132 
133 	/** Entry on inode's write_files list */
134 	struct list_head write_entry;
135 
136 	/** RB node to be linked on fuse_conn->polled_files */
137 	struct rb_node polled_node;
138 
139 	/** Wait queue head for poll */
140 	wait_queue_head_t poll_wait;
141 
142 	/** Has flock been performed on this file? */
143 	bool flock:1;
144 };
145 
146 /** One input argument of a request */
147 struct fuse_in_arg {
148 	unsigned size;
149 	const void *value;
150 };
151 
152 /** The request input */
153 struct fuse_in {
154 	/** The request header */
155 	struct fuse_in_header h;
156 
157 	/** True if the data for the last argument is in req->pages */
158 	unsigned argpages:1;
159 
160 	/** Number of arguments */
161 	unsigned numargs;
162 
163 	/** Array of arguments */
164 	struct fuse_in_arg args[3];
165 };
166 
167 /** One output argument of a request */
168 struct fuse_arg {
169 	unsigned size;
170 	void *value;
171 };
172 
173 /** The request output */
174 struct fuse_out {
175 	/** Header returned from userspace */
176 	struct fuse_out_header h;
177 
178 	/*
179 	 * The following bitfields are not changed during the request
180 	 * processing
181 	 */
182 
183 	/** Last argument is variable length (can be shorter than
184 	    arg->size) */
185 	unsigned argvar:1;
186 
187 	/** Last argument is a list of pages to copy data to */
188 	unsigned argpages:1;
189 
190 	/** Zero partially or not copied pages */
191 	unsigned page_zeroing:1;
192 
193 	/** Pages may be replaced with new ones */
194 	unsigned page_replace:1;
195 
196 	/** Number or arguments */
197 	unsigned numargs;
198 
199 	/** Array of arguments */
200 	struct fuse_arg args[3];
201 };
202 
203 /** The request state */
204 enum fuse_req_state {
205 	FUSE_REQ_INIT = 0,
206 	FUSE_REQ_PENDING,
207 	FUSE_REQ_READING,
208 	FUSE_REQ_SENT,
209 	FUSE_REQ_WRITING,
210 	FUSE_REQ_FINISHED
211 };
212 
213 /**
214  * A request to the client
215  */
216 struct fuse_req {
217 	/** This can be on either pending processing or io lists in
218 	    fuse_conn */
219 	struct list_head list;
220 
221 	/** Entry on the interrupts list  */
222 	struct list_head intr_entry;
223 
224 	/** refcount */
225 	atomic_t count;
226 
227 	/** Unique ID for the interrupt request */
228 	u64 intr_unique;
229 
230 	/*
231 	 * The following bitfields are either set once before the
232 	 * request is queued or setting/clearing them is protected by
233 	 * fuse_conn->lock
234 	 */
235 
236 	/** True if the request has reply */
237 	unsigned isreply:1;
238 
239 	/** Force sending of the request even if interrupted */
240 	unsigned force:1;
241 
242 	/** The request was aborted */
243 	unsigned aborted:1;
244 
245 	/** Request is sent in the background */
246 	unsigned background:1;
247 
248 	/** The request has been interrupted */
249 	unsigned interrupted:1;
250 
251 	/** Data is being copied to/from the request */
252 	unsigned locked:1;
253 
254 	/** Request is counted as "waiting" */
255 	unsigned waiting:1;
256 
257 	/** State of the request */
258 	enum fuse_req_state state;
259 
260 	/** The request input */
261 	struct fuse_in in;
262 
263 	/** The request output */
264 	struct fuse_out out;
265 
266 	/** Used to wake up the task waiting for completion of request*/
267 	wait_queue_head_t waitq;
268 
269 	/** Data for asynchronous requests */
270 	union {
271 		struct {
272 			union {
273 				struct fuse_release_in in;
274 				struct work_struct work;
275 			};
276 			struct path path;
277 		} release;
278 		struct fuse_init_in init_in;
279 		struct fuse_init_out init_out;
280 		struct cuse_init_in cuse_init_in;
281 		struct {
282 			struct fuse_read_in in;
283 			u64 attr_ver;
284 		} read;
285 		struct {
286 			struct fuse_write_in in;
287 			struct fuse_write_out out;
288 		} write;
289 		struct fuse_notify_retrieve_in retrieve_in;
290 		struct fuse_lk_in lk_in;
291 	} misc;
292 
293 	/** page vector */
294 	struct page *pages[FUSE_MAX_PAGES_PER_REQ];
295 
296 	/** number of pages in vector */
297 	unsigned num_pages;
298 
299 	/** offset of data on first page */
300 	unsigned page_offset;
301 
302 	/** File used in the request (or NULL) */
303 	struct fuse_file *ff;
304 
305 	/** Inode used in the request or NULL */
306 	struct inode *inode;
307 
308 	/** Link on fi->writepages */
309 	struct list_head writepages_entry;
310 
311 	/** Request completion callback */
312 	void (*end)(struct fuse_conn *, struct fuse_req *);
313 
314 	/** Request is stolen from fuse_file->reserved_req */
315 	struct file *stolen_file;
316 };
317 
318 /**
319  * A Fuse connection.
320  *
321  * This structure is created, when the filesystem is mounted, and is
322  * destroyed, when the client device is closed and the filesystem is
323  * unmounted.
324  */
325 struct fuse_conn {
326 	/** Lock protecting accessess to  members of this structure */
327 	spinlock_t lock;
328 
329 	/** Mutex protecting against directory alias creation */
330 	struct mutex inst_mutex;
331 
332 	/** Refcount */
333 	atomic_t count;
334 
335 	/** The user id for this mount */
336 	kuid_t user_id;
337 
338 	/** The group id for this mount */
339 	kgid_t group_id;
340 
341 	/** The fuse mount flags for this mount */
342 	unsigned flags;
343 
344 	/** Maximum read size */
345 	unsigned max_read;
346 
347 	/** Maximum write size */
348 	unsigned max_write;
349 
350 	/** Readers of the connection are waiting on this */
351 	wait_queue_head_t waitq;
352 
353 	/** The list of pending requests */
354 	struct list_head pending;
355 
356 	/** The list of requests being processed */
357 	struct list_head processing;
358 
359 	/** The list of requests under I/O */
360 	struct list_head io;
361 
362 	/** The next unique kernel file handle */
363 	u64 khctr;
364 
365 	/** rbtree of fuse_files waiting for poll events indexed by ph */
366 	struct rb_root polled_files;
367 
368 	/** Maximum number of outstanding background requests */
369 	unsigned max_background;
370 
371 	/** Number of background requests at which congestion starts */
372 	unsigned congestion_threshold;
373 
374 	/** Number of requests currently in the background */
375 	unsigned num_background;
376 
377 	/** Number of background requests currently queued for userspace */
378 	unsigned active_background;
379 
380 	/** The list of background requests set aside for later queuing */
381 	struct list_head bg_queue;
382 
383 	/** Pending interrupts */
384 	struct list_head interrupts;
385 
386 	/** Queue of pending forgets */
387 	struct fuse_forget_link forget_list_head;
388 	struct fuse_forget_link *forget_list_tail;
389 
390 	/** Batching of FORGET requests (positive indicates FORGET batch) */
391 	int forget_batch;
392 
393 	/** Flag indicating if connection is blocked.  This will be
394 	    the case before the INIT reply is received, and if there
395 	    are too many outstading backgrounds requests */
396 	int blocked;
397 
398 	/** waitq for blocked connection */
399 	wait_queue_head_t blocked_waitq;
400 
401 	/** waitq for reserved requests */
402 	wait_queue_head_t reserved_req_waitq;
403 
404 	/** The next unique request id */
405 	u64 reqctr;
406 
407 	/** Connection established, cleared on umount, connection
408 	    abort and device release */
409 	unsigned connected;
410 
411 	/** Connection failed (version mismatch).  Cannot race with
412 	    setting other bitfields since it is only set once in INIT
413 	    reply, before any other request, and never cleared */
414 	unsigned conn_error:1;
415 
416 	/** Connection successful.  Only set in INIT */
417 	unsigned conn_init:1;
418 
419 	/** Do readpages asynchronously?  Only set in INIT */
420 	unsigned async_read:1;
421 
422 	/** Do not send separate SETATTR request before open(O_TRUNC)  */
423 	unsigned atomic_o_trunc:1;
424 
425 	/** Filesystem supports NFS exporting.  Only set in INIT */
426 	unsigned export_support:1;
427 
428 	/** Set if bdi is valid */
429 	unsigned bdi_initialized:1;
430 
431 	/*
432 	 * The following bitfields are only for optimization purposes
433 	 * and hence races in setting them will not cause malfunction
434 	 */
435 
436 	/** Is fsync not implemented by fs? */
437 	unsigned no_fsync:1;
438 
439 	/** Is fsyncdir not implemented by fs? */
440 	unsigned no_fsyncdir:1;
441 
442 	/** Is flush not implemented by fs? */
443 	unsigned no_flush:1;
444 
445 	/** Is setxattr not implemented by fs? */
446 	unsigned no_setxattr:1;
447 
448 	/** Is getxattr not implemented by fs? */
449 	unsigned no_getxattr:1;
450 
451 	/** Is listxattr not implemented by fs? */
452 	unsigned no_listxattr:1;
453 
454 	/** Is removexattr not implemented by fs? */
455 	unsigned no_removexattr:1;
456 
457 	/** Are posix file locking primitives not implemented by fs? */
458 	unsigned no_lock:1;
459 
460 	/** Is access not implemented by fs? */
461 	unsigned no_access:1;
462 
463 	/** Is create not implemented by fs? */
464 	unsigned no_create:1;
465 
466 	/** Is interrupt not implemented by fs? */
467 	unsigned no_interrupt:1;
468 
469 	/** Is bmap not implemented by fs? */
470 	unsigned no_bmap:1;
471 
472 	/** Is poll not implemented by fs? */
473 	unsigned no_poll:1;
474 
475 	/** Do multi-page cached writes */
476 	unsigned big_writes:1;
477 
478 	/** Don't apply umask to creation modes */
479 	unsigned dont_mask:1;
480 
481 	/** Are BSD file locking primitives not implemented by fs? */
482 	unsigned no_flock:1;
483 
484 	/** Is fallocate not implemented by fs? */
485 	unsigned no_fallocate:1;
486 
487 	/** Use enhanced/automatic page cache invalidation. */
488 	unsigned auto_inval_data:1;
489 
490 	/** The number of requests waiting for completion */
491 	atomic_t num_waiting;
492 
493 	/** Negotiated minor version */
494 	unsigned minor;
495 
496 	/** Backing dev info */
497 	struct backing_dev_info bdi;
498 
499 	/** Entry on the fuse_conn_list */
500 	struct list_head entry;
501 
502 	/** Device ID from super block */
503 	dev_t dev;
504 
505 	/** Dentries in the control filesystem */
506 	struct dentry *ctl_dentry[FUSE_CTL_NUM_DENTRIES];
507 
508 	/** number of dentries used in the above array */
509 	int ctl_ndents;
510 
511 	/** O_ASYNC requests */
512 	struct fasync_struct *fasync;
513 
514 	/** Key for lock owner ID scrambling */
515 	u32 scramble_key[4];
516 
517 	/** Reserved request for the DESTROY message */
518 	struct fuse_req *destroy_req;
519 
520 	/** Version counter for attribute changes */
521 	u64 attr_version;
522 
523 	/** Called on final put */
524 	void (*release)(struct fuse_conn *);
525 
526 	/** Super block for this connection. */
527 	struct super_block *sb;
528 
529 	/** Read/write semaphore to hold when accessing sb. */
530 	struct rw_semaphore killsb;
531 };
532 
533 static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
534 {
535 	return sb->s_fs_info;
536 }
537 
538 static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
539 {
540 	return get_fuse_conn_super(inode->i_sb);
541 }
542 
543 static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
544 {
545 	return container_of(inode, struct fuse_inode, inode);
546 }
547 
548 static inline u64 get_node_id(struct inode *inode)
549 {
550 	return get_fuse_inode(inode)->nodeid;
551 }
552 
553 /** Device operations */
554 extern const struct file_operations fuse_dev_operations;
555 
556 extern const struct dentry_operations fuse_dentry_operations;
557 
558 /**
559  * Inode to nodeid comparison.
560  */
561 int fuse_inode_eq(struct inode *inode, void *_nodeidp);
562 
563 /**
564  * Get a filled in inode
565  */
566 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
567 			int generation, struct fuse_attr *attr,
568 			u64 attr_valid, u64 attr_version);
569 
570 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
571 		     struct fuse_entry_out *outarg, struct inode **inode);
572 
573 /**
574  * Send FORGET command
575  */
576 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
577 		       u64 nodeid, u64 nlookup);
578 
579 struct fuse_forget_link *fuse_alloc_forget(void);
580 
581 /**
582  * Initialize READ or READDIR request
583  */
584 void fuse_read_fill(struct fuse_req *req, struct file *file,
585 		    loff_t pos, size_t count, int opcode);
586 
587 /**
588  * Send OPEN or OPENDIR request
589  */
590 int fuse_open_common(struct inode *inode, struct file *file, bool isdir);
591 
592 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc);
593 struct fuse_file *fuse_file_get(struct fuse_file *ff);
594 void fuse_file_free(struct fuse_file *ff);
595 void fuse_finish_open(struct inode *inode, struct file *file);
596 
597 void fuse_sync_release(struct fuse_file *ff, int flags);
598 
599 /**
600  * Send RELEASE or RELEASEDIR request
601  */
602 void fuse_release_common(struct file *file, int opcode);
603 
604 /**
605  * Send FSYNC or FSYNCDIR request
606  */
607 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
608 		      int datasync, int isdir);
609 
610 /**
611  * Notify poll wakeup
612  */
613 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
614 			    struct fuse_notify_poll_wakeup_out *outarg);
615 
616 /**
617  * Initialize file operations on a regular file
618  */
619 void fuse_init_file_inode(struct inode *inode);
620 
621 /**
622  * Initialize inode operations on regular files and special files
623  */
624 void fuse_init_common(struct inode *inode);
625 
626 /**
627  * Initialize inode and file operations on a directory
628  */
629 void fuse_init_dir(struct inode *inode);
630 
631 /**
632  * Initialize inode operations on a symlink
633  */
634 void fuse_init_symlink(struct inode *inode);
635 
636 /**
637  * Change attributes of an inode
638  */
639 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
640 			    u64 attr_valid, u64 attr_version);
641 
642 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
643 				   u64 attr_valid);
644 
645 /**
646  * Initialize the client device
647  */
648 int fuse_dev_init(void);
649 
650 /**
651  * Cleanup the client device
652  */
653 void fuse_dev_cleanup(void);
654 
655 int fuse_ctl_init(void);
656 void fuse_ctl_cleanup(void);
657 
658 /**
659  * Allocate a request
660  */
661 struct fuse_req *fuse_request_alloc(void);
662 
663 struct fuse_req *fuse_request_alloc_nofs(void);
664 
665 /**
666  * Free a request
667  */
668 void fuse_request_free(struct fuse_req *req);
669 
670 /**
671  * Get a request, may fail with -ENOMEM
672  */
673 struct fuse_req *fuse_get_req(struct fuse_conn *fc);
674 
675 /**
676  * Gets a requests for a file operation, always succeeds
677  */
678 struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file);
679 
680 /**
681  * Decrement reference count of a request.  If count goes to zero free
682  * the request.
683  */
684 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
685 
686 /**
687  * Send a request (synchronous)
688  */
689 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req);
690 
691 /**
692  * Send a request in the background
693  */
694 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req);
695 
696 void fuse_request_send_background_locked(struct fuse_conn *fc,
697 					 struct fuse_req *req);
698 
699 /* Abort all requests */
700 void fuse_abort_conn(struct fuse_conn *fc);
701 
702 /**
703  * Invalidate inode attributes
704  */
705 void fuse_invalidate_attr(struct inode *inode);
706 
707 void fuse_invalidate_entry_cache(struct dentry *entry);
708 
709 /**
710  * Acquire reference to fuse_conn
711  */
712 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc);
713 
714 void fuse_conn_kill(struct fuse_conn *fc);
715 
716 /**
717  * Initialize fuse_conn
718  */
719 void fuse_conn_init(struct fuse_conn *fc);
720 
721 /**
722  * Release reference to fuse_conn
723  */
724 void fuse_conn_put(struct fuse_conn *fc);
725 
726 /**
727  * Add connection to control filesystem
728  */
729 int fuse_ctl_add_conn(struct fuse_conn *fc);
730 
731 /**
732  * Remove connection from control filesystem
733  */
734 void fuse_ctl_remove_conn(struct fuse_conn *fc);
735 
736 /**
737  * Is file type valid?
738  */
739 int fuse_valid_type(int m);
740 
741 /**
742  * Is task allowed to perform filesystem operation?
743  */
744 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task);
745 
746 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
747 
748 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
749 			   struct file *file, bool *refreshed);
750 
751 void fuse_flush_writepages(struct inode *inode);
752 
753 void fuse_set_nowrite(struct inode *inode);
754 void fuse_release_nowrite(struct inode *inode);
755 
756 u64 fuse_get_attr_version(struct fuse_conn *fc);
757 
758 /**
759  * File-system tells the kernel to invalidate cache for the given node id.
760  */
761 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
762 			     loff_t offset, loff_t len);
763 
764 /**
765  * File-system tells the kernel to invalidate parent attributes and
766  * the dentry matching parent/name.
767  *
768  * If the child_nodeid is non-zero and:
769  *    - matches the inode number for the dentry matching parent/name,
770  *    - is not a mount point
771  *    - is a file or oan empty directory
772  * then the dentry is unhashed (d_delete()).
773  */
774 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
775 			     u64 child_nodeid, struct qstr *name);
776 
777 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
778 		 bool isdir);
779 ssize_t fuse_direct_io(struct file *file, const char __user *buf,
780 		       size_t count, loff_t *ppos, int write);
781 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
782 		   unsigned int flags);
783 long fuse_ioctl_common(struct file *file, unsigned int cmd,
784 		       unsigned long arg, unsigned int flags);
785 unsigned fuse_file_poll(struct file *file, poll_table *wait);
786 int fuse_dev_release(struct inode *inode, struct file *file);
787 
788 void fuse_write_update_size(struct inode *inode, loff_t pos);
789 
790 #endif /* _FS_FUSE_I_H */
791