2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/sched/signal.h>
15 #include <linux/uio.h>
16 #include <linux/miscdevice.h>
17 #include <linux/pagemap.h>
18 #include <linux/file.h>
19 #include <linux/slab.h>
20 #include <linux/pipe_fs_i.h>
21 #include <linux/swap.h>
22 #include <linux/splice.h>
23 #include <linux/sched.h>
25 MODULE_ALIAS_MISCDEV(FUSE_MINOR
);
26 MODULE_ALIAS("devname:fuse");
28 /* Ordinary requests have even IDs, while interrupts IDs are odd */
29 #define FUSE_INT_REQ_BIT (1ULL << 0)
30 #define FUSE_REQ_ID_STEP (1ULL << 1)
32 static struct kmem_cache
*fuse_req_cachep
;
34 static struct fuse_dev
*fuse_get_dev(struct file
*file
)
37 * Lockless access is OK, because file->private data is set
38 * once during mount and is valid until the file is released.
40 return READ_ONCE(file
->private_data
);
43 static void fuse_request_init(struct fuse_req
*req
, struct page
**pages
,
44 struct fuse_page_desc
*page_descs
,
47 INIT_LIST_HEAD(&req
->list
);
48 INIT_LIST_HEAD(&req
->intr_entry
);
49 init_waitqueue_head(&req
->waitq
);
50 refcount_set(&req
->count
, 1);
52 req
->page_descs
= page_descs
;
53 req
->max_pages
= npages
;
54 __set_bit(FR_PENDING
, &req
->flags
);
57 static struct fuse_req
*__fuse_request_alloc(unsigned npages
, gfp_t flags
)
59 struct fuse_req
*req
= kmem_cache_zalloc(fuse_req_cachep
, flags
);
61 struct page
**pages
= NULL
;
62 struct fuse_page_desc
*page_descs
= NULL
;
64 WARN_ON(npages
> FUSE_MAX_MAX_PAGES
);
65 if (npages
> FUSE_REQ_INLINE_PAGES
) {
66 pages
= kzalloc(npages
* (sizeof(*pages
) +
67 sizeof(*page_descs
)), flags
);
69 kmem_cache_free(fuse_req_cachep
, req
);
72 page_descs
= (void *) pages
+ npages
* sizeof(*pages
);
74 pages
= req
->inline_pages
;
75 page_descs
= req
->inline_page_descs
;
78 fuse_request_init(req
, pages
, page_descs
, npages
);
83 struct fuse_req
*fuse_request_alloc(unsigned npages
)
85 return __fuse_request_alloc(npages
, GFP_KERNEL
);
87 EXPORT_SYMBOL_GPL(fuse_request_alloc
);
89 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
)
91 return __fuse_request_alloc(npages
, GFP_NOFS
);
94 void fuse_request_free(struct fuse_req
*req
)
96 if (req
->pages
!= req
->inline_pages
)
99 kmem_cache_free(fuse_req_cachep
, req
);
102 void __fuse_get_request(struct fuse_req
*req
)
104 refcount_inc(&req
->count
);
107 /* Must be called with > 1 refcount */
108 static void __fuse_put_request(struct fuse_req
*req
)
110 refcount_dec(&req
->count
);
113 void fuse_set_initialized(struct fuse_conn
*fc
)
115 /* Make sure stores before this are seen on another CPU */
120 static bool fuse_block_alloc(struct fuse_conn
*fc
, bool for_background
)
122 return !fc
->initialized
|| (for_background
&& fc
->blocked
);
125 static void fuse_drop_waiting(struct fuse_conn
*fc
)
128 atomic_dec(&fc
->num_waiting
);
129 } else if (atomic_dec_and_test(&fc
->num_waiting
)) {
130 /* wake up aborters */
131 wake_up_all(&fc
->blocked_waitq
);
135 static struct fuse_req
*__fuse_get_req(struct fuse_conn
*fc
, unsigned npages
,
138 struct fuse_req
*req
;
140 atomic_inc(&fc
->num_waiting
);
142 if (fuse_block_alloc(fc
, for_background
)) {
144 if (wait_event_killable_exclusive(fc
->blocked_waitq
,
145 !fuse_block_alloc(fc
, for_background
)))
148 /* Matches smp_wmb() in fuse_set_initialized() */
159 req
= fuse_request_alloc(npages
);
163 wake_up(&fc
->blocked_waitq
);
167 req
->in
.h
.uid
= from_kuid(fc
->user_ns
, current_fsuid());
168 req
->in
.h
.gid
= from_kgid(fc
->user_ns
, current_fsgid());
169 req
->in
.h
.pid
= pid_nr_ns(task_pid(current
), fc
->pid_ns
);
171 __set_bit(FR_WAITING
, &req
->flags
);
173 __set_bit(FR_BACKGROUND
, &req
->flags
);
175 if (unlikely(req
->in
.h
.uid
== ((uid_t
)-1) ||
176 req
->in
.h
.gid
== ((gid_t
)-1))) {
177 fuse_put_request(fc
, req
);
178 return ERR_PTR(-EOVERFLOW
);
183 fuse_drop_waiting(fc
);
187 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
)
189 return __fuse_get_req(fc
, npages
, false);
191 EXPORT_SYMBOL_GPL(fuse_get_req
);
193 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
196 return __fuse_get_req(fc
, npages
, true);
198 EXPORT_SYMBOL_GPL(fuse_get_req_for_background
);
201 * Return request in fuse_file->reserved_req. However that may
202 * currently be in use. If that is the case, wait for it to become
205 static struct fuse_req
*get_reserved_req(struct fuse_conn
*fc
,
208 struct fuse_req
*req
= NULL
;
209 struct fuse_file
*ff
= file
->private_data
;
212 wait_event(fc
->reserved_req_waitq
, ff
->reserved_req
);
213 spin_lock(&fc
->lock
);
214 if (ff
->reserved_req
) {
215 req
= ff
->reserved_req
;
216 ff
->reserved_req
= NULL
;
217 req
->stolen_file
= get_file(file
);
219 spin_unlock(&fc
->lock
);
226 * Put stolen request back into fuse_file->reserved_req
228 static void put_reserved_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
230 struct file
*file
= req
->stolen_file
;
231 struct fuse_file
*ff
= file
->private_data
;
233 WARN_ON(req
->max_pages
);
234 spin_lock(&fc
->lock
);
235 memset(req
, 0, sizeof(*req
));
236 fuse_request_init(req
, NULL
, NULL
, 0);
237 BUG_ON(ff
->reserved_req
);
238 ff
->reserved_req
= req
;
239 wake_up_all(&fc
->reserved_req_waitq
);
240 spin_unlock(&fc
->lock
);
245 * Gets a requests for a file operation, always succeeds
247 * This is used for sending the FLUSH request, which must get to
248 * userspace, due to POSIX locks which may need to be unlocked.
250 * If allocation fails due to OOM, use the reserved request in
253 * This is very unlikely to deadlock accidentally, since the
254 * filesystem should not have it's own file open. If deadlock is
255 * intentional, it can still be broken by "aborting" the filesystem.
257 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
260 struct fuse_req
*req
;
262 atomic_inc(&fc
->num_waiting
);
263 wait_event(fc
->blocked_waitq
, fc
->initialized
);
264 /* Matches smp_wmb() in fuse_set_initialized() */
266 req
= fuse_request_alloc(0);
268 req
= get_reserved_req(fc
, file
);
270 req
->in
.h
.uid
= from_kuid_munged(fc
->user_ns
, current_fsuid());
271 req
->in
.h
.gid
= from_kgid_munged(fc
->user_ns
, current_fsgid());
272 req
->in
.h
.pid
= pid_nr_ns(task_pid(current
), fc
->pid_ns
);
274 __set_bit(FR_WAITING
, &req
->flags
);
275 __clear_bit(FR_BACKGROUND
, &req
->flags
);
279 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
281 if (refcount_dec_and_test(&req
->count
)) {
282 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
284 * We get here in the unlikely case that a background
285 * request was allocated but not sent
287 spin_lock(&fc
->bg_lock
);
289 wake_up(&fc
->blocked_waitq
);
290 spin_unlock(&fc
->bg_lock
);
293 if (test_bit(FR_WAITING
, &req
->flags
)) {
294 __clear_bit(FR_WAITING
, &req
->flags
);
295 fuse_drop_waiting(fc
);
298 if (req
->stolen_file
)
299 put_reserved_req(fc
, req
);
301 fuse_request_free(req
);
304 EXPORT_SYMBOL_GPL(fuse_put_request
);
306 static unsigned len_args(unsigned numargs
, struct fuse_arg
*args
)
311 for (i
= 0; i
< numargs
; i
++)
312 nbytes
+= args
[i
].size
;
317 static u64
fuse_get_unique(struct fuse_iqueue
*fiq
)
319 fiq
->reqctr
+= FUSE_REQ_ID_STEP
;
323 static unsigned int fuse_req_hash(u64 unique
)
325 return hash_long(unique
& ~FUSE_INT_REQ_BIT
, FUSE_PQ_HASH_BITS
);
328 static void queue_request(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
330 req
->in
.h
.len
= sizeof(struct fuse_in_header
) +
331 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
332 list_add_tail(&req
->list
, &fiq
->pending
);
333 wake_up_locked(&fiq
->waitq
);
334 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
337 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
338 u64 nodeid
, u64 nlookup
)
340 struct fuse_iqueue
*fiq
= &fc
->iq
;
342 forget
->forget_one
.nodeid
= nodeid
;
343 forget
->forget_one
.nlookup
= nlookup
;
345 spin_lock(&fiq
->waitq
.lock
);
346 if (fiq
->connected
) {
347 fiq
->forget_list_tail
->next
= forget
;
348 fiq
->forget_list_tail
= forget
;
349 wake_up_locked(&fiq
->waitq
);
350 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
354 spin_unlock(&fiq
->waitq
.lock
);
357 static void flush_bg_queue(struct fuse_conn
*fc
)
359 struct fuse_iqueue
*fiq
= &fc
->iq
;
361 while (fc
->active_background
< fc
->max_background
&&
362 !list_empty(&fc
->bg_queue
)) {
363 struct fuse_req
*req
;
365 req
= list_first_entry(&fc
->bg_queue
, struct fuse_req
, list
);
366 list_del(&req
->list
);
367 fc
->active_background
++;
368 spin_lock(&fiq
->waitq
.lock
);
369 req
->in
.h
.unique
= fuse_get_unique(fiq
);
370 queue_request(fiq
, req
);
371 spin_unlock(&fiq
->waitq
.lock
);
376 * This function is called when a request is finished. Either a reply
377 * has arrived or it was aborted (and not yet sent) or some error
378 * occurred during communication with userspace, or the device file
379 * was closed. The requester thread is woken up (if still waiting),
380 * the 'end' callback is called if given, else the reference to the
381 * request is released
383 static void request_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
385 struct fuse_iqueue
*fiq
= &fc
->iq
;
387 if (test_and_set_bit(FR_FINISHED
, &req
->flags
))
390 spin_lock(&fiq
->waitq
.lock
);
391 list_del_init(&req
->intr_entry
);
392 spin_unlock(&fiq
->waitq
.lock
);
393 WARN_ON(test_bit(FR_PENDING
, &req
->flags
));
394 WARN_ON(test_bit(FR_SENT
, &req
->flags
));
395 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
396 spin_lock(&fc
->bg_lock
);
397 clear_bit(FR_BACKGROUND
, &req
->flags
);
398 if (fc
->num_background
== fc
->max_background
) {
400 wake_up(&fc
->blocked_waitq
);
401 } else if (!fc
->blocked
) {
403 * Wake up next waiter, if any. It's okay to use
404 * waitqueue_active(), as we've already synced up
405 * fc->blocked with waiters with the wake_up() call
408 if (waitqueue_active(&fc
->blocked_waitq
))
409 wake_up(&fc
->blocked_waitq
);
412 if (fc
->num_background
== fc
->congestion_threshold
&& fc
->sb
) {
413 clear_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_SYNC
);
414 clear_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_ASYNC
);
416 fc
->num_background
--;
417 fc
->active_background
--;
419 spin_unlock(&fc
->bg_lock
);
421 wake_up(&req
->waitq
);
425 fuse_put_request(fc
, req
);
428 static void queue_interrupt(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
430 spin_lock(&fiq
->waitq
.lock
);
431 if (test_bit(FR_FINISHED
, &req
->flags
)) {
432 spin_unlock(&fiq
->waitq
.lock
);
435 if (list_empty(&req
->intr_entry
)) {
436 list_add_tail(&req
->intr_entry
, &fiq
->interrupts
);
437 wake_up_locked(&fiq
->waitq
);
439 spin_unlock(&fiq
->waitq
.lock
);
440 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
443 static void request_wait_answer(struct fuse_conn
*fc
, struct fuse_req
*req
)
445 struct fuse_iqueue
*fiq
= &fc
->iq
;
448 if (!fc
->no_interrupt
) {
449 /* Any signal may interrupt this */
450 err
= wait_event_interruptible(req
->waitq
,
451 test_bit(FR_FINISHED
, &req
->flags
));
455 set_bit(FR_INTERRUPTED
, &req
->flags
);
456 /* matches barrier in fuse_dev_do_read() */
457 smp_mb__after_atomic();
458 if (test_bit(FR_SENT
, &req
->flags
))
459 queue_interrupt(fiq
, req
);
462 if (!test_bit(FR_FORCE
, &req
->flags
)) {
463 /* Only fatal signals may interrupt this */
464 err
= wait_event_killable(req
->waitq
,
465 test_bit(FR_FINISHED
, &req
->flags
));
469 spin_lock(&fiq
->waitq
.lock
);
470 /* Request is not yet in userspace, bail out */
471 if (test_bit(FR_PENDING
, &req
->flags
)) {
472 list_del(&req
->list
);
473 spin_unlock(&fiq
->waitq
.lock
);
474 __fuse_put_request(req
);
475 req
->out
.h
.error
= -EINTR
;
478 spin_unlock(&fiq
->waitq
.lock
);
482 * Either request is already in userspace, or it was forced.
485 wait_event(req
->waitq
, test_bit(FR_FINISHED
, &req
->flags
));
488 static void __fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
490 struct fuse_iqueue
*fiq
= &fc
->iq
;
492 BUG_ON(test_bit(FR_BACKGROUND
, &req
->flags
));
493 spin_lock(&fiq
->waitq
.lock
);
494 if (!fiq
->connected
) {
495 spin_unlock(&fiq
->waitq
.lock
);
496 req
->out
.h
.error
= -ENOTCONN
;
498 req
->in
.h
.unique
= fuse_get_unique(fiq
);
499 queue_request(fiq
, req
);
500 /* acquire extra reference, since request is still needed
501 after request_end() */
502 __fuse_get_request(req
);
503 spin_unlock(&fiq
->waitq
.lock
);
505 request_wait_answer(fc
, req
);
506 /* Pairs with smp_wmb() in request_end() */
511 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
513 __set_bit(FR_ISREPLY
, &req
->flags
);
514 if (!test_bit(FR_WAITING
, &req
->flags
)) {
515 __set_bit(FR_WAITING
, &req
->flags
);
516 atomic_inc(&fc
->num_waiting
);
518 __fuse_request_send(fc
, req
);
520 EXPORT_SYMBOL_GPL(fuse_request_send
);
522 static void fuse_adjust_compat(struct fuse_conn
*fc
, struct fuse_args
*args
)
524 if (fc
->minor
< 4 && args
->in
.h
.opcode
== FUSE_STATFS
)
525 args
->out
.args
[0].size
= FUSE_COMPAT_STATFS_SIZE
;
528 switch (args
->in
.h
.opcode
) {
535 args
->out
.args
[0].size
= FUSE_COMPAT_ENTRY_OUT_SIZE
;
539 args
->out
.args
[0].size
= FUSE_COMPAT_ATTR_OUT_SIZE
;
543 if (fc
->minor
< 12) {
544 switch (args
->in
.h
.opcode
) {
546 args
->in
.args
[0].size
= sizeof(struct fuse_open_in
);
549 args
->in
.args
[0].size
= FUSE_COMPAT_MKNOD_IN_SIZE
;
555 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
)
557 struct fuse_req
*req
;
560 req
= fuse_get_req(fc
, 0);
564 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
565 fuse_adjust_compat(fc
, args
);
567 req
->in
.h
.opcode
= args
->in
.h
.opcode
;
568 req
->in
.h
.nodeid
= args
->in
.h
.nodeid
;
569 req
->in
.numargs
= args
->in
.numargs
;
570 memcpy(req
->in
.args
, args
->in
.args
,
571 args
->in
.numargs
* sizeof(struct fuse_in_arg
));
572 req
->out
.argvar
= args
->out
.argvar
;
573 req
->out
.numargs
= args
->out
.numargs
;
574 memcpy(req
->out
.args
, args
->out
.args
,
575 args
->out
.numargs
* sizeof(struct fuse_arg
));
576 fuse_request_send(fc
, req
);
577 ret
= req
->out
.h
.error
;
578 if (!ret
&& args
->out
.argvar
) {
579 BUG_ON(args
->out
.numargs
!= 1);
580 ret
= req
->out
.args
[0].size
;
582 fuse_put_request(fc
, req
);
587 bool fuse_request_queue_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
591 WARN_ON(!test_bit(FR_BACKGROUND
, &req
->flags
));
592 if (!test_bit(FR_WAITING
, &req
->flags
)) {
593 __set_bit(FR_WAITING
, &req
->flags
);
594 atomic_inc(&fc
->num_waiting
);
596 __set_bit(FR_ISREPLY
, &req
->flags
);
597 spin_lock(&fc
->bg_lock
);
598 if (likely(fc
->connected
)) {
599 fc
->num_background
++;
600 if (fc
->num_background
== fc
->max_background
)
602 if (fc
->num_background
== fc
->congestion_threshold
&& fc
->sb
) {
603 set_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_SYNC
);
604 set_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_ASYNC
);
606 list_add_tail(&req
->list
, &fc
->bg_queue
);
610 spin_unlock(&fc
->bg_lock
);
615 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
618 if (!fuse_request_queue_background(fc
, req
)) {
619 req
->out
.h
.error
= -ENOTCONN
;
621 fuse_put_request(fc
, req
);
624 EXPORT_SYMBOL_GPL(fuse_request_send_background
);
626 static int fuse_request_send_notify_reply(struct fuse_conn
*fc
,
627 struct fuse_req
*req
, u64 unique
)
630 struct fuse_iqueue
*fiq
= &fc
->iq
;
632 __clear_bit(FR_ISREPLY
, &req
->flags
);
633 req
->in
.h
.unique
= unique
;
634 spin_lock(&fiq
->waitq
.lock
);
635 if (fiq
->connected
) {
636 queue_request(fiq
, req
);
639 spin_unlock(&fiq
->waitq
.lock
);
644 void fuse_force_forget(struct file
*file
, u64 nodeid
)
646 struct inode
*inode
= file_inode(file
);
647 struct fuse_conn
*fc
= get_fuse_conn(inode
);
648 struct fuse_req
*req
;
649 struct fuse_forget_in inarg
;
651 memset(&inarg
, 0, sizeof(inarg
));
653 req
= fuse_get_req_nofail_nopages(fc
, file
);
654 req
->in
.h
.opcode
= FUSE_FORGET
;
655 req
->in
.h
.nodeid
= nodeid
;
657 req
->in
.args
[0].size
= sizeof(inarg
);
658 req
->in
.args
[0].value
= &inarg
;
659 __clear_bit(FR_ISREPLY
, &req
->flags
);
660 __fuse_request_send(fc
, req
);
662 fuse_put_request(fc
, req
);
666 * Lock the request. Up to the next unlock_request() there mustn't be
667 * anything that could cause a page-fault. If the request was already
670 static int lock_request(struct fuse_req
*req
)
674 spin_lock(&req
->waitq
.lock
);
675 if (test_bit(FR_ABORTED
, &req
->flags
))
678 set_bit(FR_LOCKED
, &req
->flags
);
679 spin_unlock(&req
->waitq
.lock
);
685 * Unlock request. If it was aborted while locked, caller is responsible
686 * for unlocking and ending the request.
688 static int unlock_request(struct fuse_req
*req
)
692 spin_lock(&req
->waitq
.lock
);
693 if (test_bit(FR_ABORTED
, &req
->flags
))
696 clear_bit(FR_LOCKED
, &req
->flags
);
697 spin_unlock(&req
->waitq
.lock
);
702 struct fuse_copy_state
{
704 struct fuse_req
*req
;
705 struct iov_iter
*iter
;
706 struct pipe_buffer
*pipebufs
;
707 struct pipe_buffer
*currbuf
;
708 struct pipe_inode_info
*pipe
;
709 unsigned long nr_segs
;
713 unsigned move_pages
:1;
716 static void fuse_copy_init(struct fuse_copy_state
*cs
, int write
,
717 struct iov_iter
*iter
)
719 memset(cs
, 0, sizeof(*cs
));
724 /* Unmap and put previous page of userspace buffer */
725 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
728 struct pipe_buffer
*buf
= cs
->currbuf
;
731 buf
->len
= PAGE_SIZE
- cs
->len
;
735 flush_dcache_page(cs
->pg
);
736 set_page_dirty_lock(cs
->pg
);
744 * Get another pagefull of userspace buffer, and map it to kernel
745 * address space, and lock request
747 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
752 err
= unlock_request(cs
->req
);
756 fuse_copy_finish(cs
);
758 struct pipe_buffer
*buf
= cs
->pipebufs
;
761 err
= pipe_buf_confirm(cs
->pipe
, buf
);
765 BUG_ON(!cs
->nr_segs
);
768 cs
->offset
= buf
->offset
;
773 if (cs
->nr_segs
== cs
->pipe
->buffers
)
776 page
= alloc_page(GFP_HIGHUSER
);
793 err
= iov_iter_get_pages(cs
->iter
, &page
, PAGE_SIZE
, 1, &off
);
800 iov_iter_advance(cs
->iter
, err
);
803 return lock_request(cs
->req
);
806 /* Do as much copy to/from userspace buffer as we can */
807 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
809 unsigned ncpy
= min(*size
, cs
->len
);
811 void *pgaddr
= kmap_atomic(cs
->pg
);
812 void *buf
= pgaddr
+ cs
->offset
;
815 memcpy(buf
, *val
, ncpy
);
817 memcpy(*val
, buf
, ncpy
);
819 kunmap_atomic(pgaddr
);
828 static int fuse_check_page(struct page
*page
)
830 if (page_mapcount(page
) ||
831 page
->mapping
!= NULL
||
832 page_count(page
) != 1 ||
833 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
840 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
841 printk(KERN_WARNING
" page=%p index=%li flags=%08lx, count=%i, mapcount=%i, mapping=%p\n", page
, page
->index
, page
->flags
, page_count(page
), page_mapcount(page
), page
->mapping
);
847 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
850 struct page
*oldpage
= *pagep
;
851 struct page
*newpage
;
852 struct pipe_buffer
*buf
= cs
->pipebufs
;
854 err
= unlock_request(cs
->req
);
858 fuse_copy_finish(cs
);
860 err
= pipe_buf_confirm(cs
->pipe
, buf
);
864 BUG_ON(!cs
->nr_segs
);
870 if (cs
->len
!= PAGE_SIZE
)
873 if (pipe_buf_steal(cs
->pipe
, buf
) != 0)
878 if (!PageUptodate(newpage
))
879 SetPageUptodate(newpage
);
881 ClearPageMappedToDisk(newpage
);
883 if (fuse_check_page(newpage
) != 0)
884 goto out_fallback_unlock
;
887 * This is a new and locked page, it shouldn't be mapped or
888 * have any special flags on it
890 if (WARN_ON(page_mapped(oldpage
)))
891 goto out_fallback_unlock
;
892 if (WARN_ON(page_has_private(oldpage
)))
893 goto out_fallback_unlock
;
894 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
895 goto out_fallback_unlock
;
896 if (WARN_ON(PageMlocked(oldpage
)))
897 goto out_fallback_unlock
;
899 err
= replace_page_cache_page(oldpage
, newpage
, GFP_KERNEL
);
901 unlock_page(newpage
);
907 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
908 lru_cache_add_file(newpage
);
911 spin_lock(&cs
->req
->waitq
.lock
);
912 if (test_bit(FR_ABORTED
, &cs
->req
->flags
))
916 spin_unlock(&cs
->req
->waitq
.lock
);
919 unlock_page(newpage
);
924 unlock_page(oldpage
);
931 unlock_page(newpage
);
934 cs
->offset
= buf
->offset
;
936 err
= lock_request(cs
->req
);
943 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
944 unsigned offset
, unsigned count
)
946 struct pipe_buffer
*buf
;
949 if (cs
->nr_segs
== cs
->pipe
->buffers
)
952 err
= unlock_request(cs
->req
);
956 fuse_copy_finish(cs
);
961 buf
->offset
= offset
;
972 * Copy a page in the request to/from the userspace buffer. Must be
975 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
976 unsigned offset
, unsigned count
, int zeroing
)
979 struct page
*page
= *pagep
;
981 if (page
&& zeroing
&& count
< PAGE_SIZE
)
982 clear_highpage(page
);
985 if (cs
->write
&& cs
->pipebufs
&& page
) {
986 return fuse_ref_page(cs
, page
, offset
, count
);
987 } else if (!cs
->len
) {
988 if (cs
->move_pages
&& page
&&
989 offset
== 0 && count
== PAGE_SIZE
) {
990 err
= fuse_try_move_page(cs
, pagep
);
994 err
= fuse_copy_fill(cs
);
1000 void *mapaddr
= kmap_atomic(page
);
1001 void *buf
= mapaddr
+ offset
;
1002 offset
+= fuse_copy_do(cs
, &buf
, &count
);
1003 kunmap_atomic(mapaddr
);
1005 offset
+= fuse_copy_do(cs
, NULL
, &count
);
1007 if (page
&& !cs
->write
)
1008 flush_dcache_page(page
);
1012 /* Copy pages in the request to/from userspace buffer */
1013 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
1017 struct fuse_req
*req
= cs
->req
;
1019 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
1021 unsigned offset
= req
->page_descs
[i
].offset
;
1022 unsigned count
= min(nbytes
, req
->page_descs
[i
].length
);
1024 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
1034 /* Copy a single argument in the request to/from userspace buffer */
1035 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
1039 int err
= fuse_copy_fill(cs
);
1043 fuse_copy_do(cs
, &val
, &size
);
1048 /* Copy request arguments to/from userspace buffer */
1049 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
1050 unsigned argpages
, struct fuse_arg
*args
,
1056 for (i
= 0; !err
&& i
< numargs
; i
++) {
1057 struct fuse_arg
*arg
= &args
[i
];
1058 if (i
== numargs
- 1 && argpages
)
1059 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
1061 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
1066 static int forget_pending(struct fuse_iqueue
*fiq
)
1068 return fiq
->forget_list_head
.next
!= NULL
;
1071 static int request_pending(struct fuse_iqueue
*fiq
)
1073 return !list_empty(&fiq
->pending
) || !list_empty(&fiq
->interrupts
) ||
1074 forget_pending(fiq
);
1078 * Transfer an interrupt request to userspace
1080 * Unlike other requests this is assembled on demand, without a need
1081 * to allocate a separate fuse_req structure.
1083 * Called with fiq->waitq.lock held, releases it
1085 static int fuse_read_interrupt(struct fuse_iqueue
*fiq
,
1086 struct fuse_copy_state
*cs
,
1087 size_t nbytes
, struct fuse_req
*req
)
1088 __releases(fiq
->waitq
.lock
)
1090 struct fuse_in_header ih
;
1091 struct fuse_interrupt_in arg
;
1092 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
1095 list_del_init(&req
->intr_entry
);
1096 memset(&ih
, 0, sizeof(ih
));
1097 memset(&arg
, 0, sizeof(arg
));
1099 ih
.opcode
= FUSE_INTERRUPT
;
1100 ih
.unique
= (req
->in
.h
.unique
| FUSE_INT_REQ_BIT
);
1101 arg
.unique
= req
->in
.h
.unique
;
1103 spin_unlock(&fiq
->waitq
.lock
);
1104 if (nbytes
< reqsize
)
1107 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1109 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1110 fuse_copy_finish(cs
);
1112 return err
? err
: reqsize
;
1115 static struct fuse_forget_link
*dequeue_forget(struct fuse_iqueue
*fiq
,
1119 struct fuse_forget_link
*head
= fiq
->forget_list_head
.next
;
1120 struct fuse_forget_link
**newhead
= &head
;
1123 for (count
= 0; *newhead
!= NULL
&& count
< max
; count
++)
1124 newhead
= &(*newhead
)->next
;
1126 fiq
->forget_list_head
.next
= *newhead
;
1128 if (fiq
->forget_list_head
.next
== NULL
)
1129 fiq
->forget_list_tail
= &fiq
->forget_list_head
;
1137 static int fuse_read_single_forget(struct fuse_iqueue
*fiq
,
1138 struct fuse_copy_state
*cs
,
1140 __releases(fiq
->waitq
.lock
)
1143 struct fuse_forget_link
*forget
= dequeue_forget(fiq
, 1, NULL
);
1144 struct fuse_forget_in arg
= {
1145 .nlookup
= forget
->forget_one
.nlookup
,
1147 struct fuse_in_header ih
= {
1148 .opcode
= FUSE_FORGET
,
1149 .nodeid
= forget
->forget_one
.nodeid
,
1150 .unique
= fuse_get_unique(fiq
),
1151 .len
= sizeof(ih
) + sizeof(arg
),
1154 spin_unlock(&fiq
->waitq
.lock
);
1156 if (nbytes
< ih
.len
)
1159 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1161 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1162 fuse_copy_finish(cs
);
1170 static int fuse_read_batch_forget(struct fuse_iqueue
*fiq
,
1171 struct fuse_copy_state
*cs
, size_t nbytes
)
1172 __releases(fiq
->waitq
.lock
)
1175 unsigned max_forgets
;
1177 struct fuse_forget_link
*head
;
1178 struct fuse_batch_forget_in arg
= { .count
= 0 };
1179 struct fuse_in_header ih
= {
1180 .opcode
= FUSE_BATCH_FORGET
,
1181 .unique
= fuse_get_unique(fiq
),
1182 .len
= sizeof(ih
) + sizeof(arg
),
1185 if (nbytes
< ih
.len
) {
1186 spin_unlock(&fiq
->waitq
.lock
);
1190 max_forgets
= (nbytes
- ih
.len
) / sizeof(struct fuse_forget_one
);
1191 head
= dequeue_forget(fiq
, max_forgets
, &count
);
1192 spin_unlock(&fiq
->waitq
.lock
);
1195 ih
.len
+= count
* sizeof(struct fuse_forget_one
);
1196 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1198 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1201 struct fuse_forget_link
*forget
= head
;
1204 err
= fuse_copy_one(cs
, &forget
->forget_one
,
1205 sizeof(forget
->forget_one
));
1207 head
= forget
->next
;
1211 fuse_copy_finish(cs
);
1219 static int fuse_read_forget(struct fuse_conn
*fc
, struct fuse_iqueue
*fiq
,
1220 struct fuse_copy_state
*cs
,
1222 __releases(fiq
->waitq
.lock
)
1224 if (fc
->minor
< 16 || fiq
->forget_list_head
.next
->next
== NULL
)
1225 return fuse_read_single_forget(fiq
, cs
, nbytes
);
1227 return fuse_read_batch_forget(fiq
, cs
, nbytes
);
1231 * Read a single request into the userspace filesystem's buffer. This
1232 * function waits until a request is available, then removes it from
1233 * the pending list and copies request data to userspace buffer. If
1234 * no reply is needed (FORGET) or request has been aborted or there
1235 * was an error during the copying then it's finished by calling
1236 * request_end(). Otherwise add it to the processing list, and set
1239 static ssize_t
fuse_dev_do_read(struct fuse_dev
*fud
, struct file
*file
,
1240 struct fuse_copy_state
*cs
, size_t nbytes
)
1243 struct fuse_conn
*fc
= fud
->fc
;
1244 struct fuse_iqueue
*fiq
= &fc
->iq
;
1245 struct fuse_pqueue
*fpq
= &fud
->pq
;
1246 struct fuse_req
*req
;
1252 spin_lock(&fiq
->waitq
.lock
);
1254 if ((file
->f_flags
& O_NONBLOCK
) && fiq
->connected
&&
1255 !request_pending(fiq
))
1258 err
= wait_event_interruptible_exclusive_locked(fiq
->waitq
,
1259 !fiq
->connected
|| request_pending(fiq
));
1263 if (!fiq
->connected
) {
1264 err
= (fc
->aborted
&& fc
->abort_err
) ? -ECONNABORTED
: -ENODEV
;
1268 if (!list_empty(&fiq
->interrupts
)) {
1269 req
= list_entry(fiq
->interrupts
.next
, struct fuse_req
,
1271 return fuse_read_interrupt(fiq
, cs
, nbytes
, req
);
1274 if (forget_pending(fiq
)) {
1275 if (list_empty(&fiq
->pending
) || fiq
->forget_batch
-- > 0)
1276 return fuse_read_forget(fc
, fiq
, cs
, nbytes
);
1278 if (fiq
->forget_batch
<= -8)
1279 fiq
->forget_batch
= 16;
1282 req
= list_entry(fiq
->pending
.next
, struct fuse_req
, list
);
1283 clear_bit(FR_PENDING
, &req
->flags
);
1284 list_del_init(&req
->list
);
1285 spin_unlock(&fiq
->waitq
.lock
);
1288 reqsize
= in
->h
.len
;
1290 /* If request is too large, reply with an error and restart the read */
1291 if (nbytes
< reqsize
) {
1292 req
->out
.h
.error
= -EIO
;
1293 /* SETXATTR is special, since it may contain too large data */
1294 if (in
->h
.opcode
== FUSE_SETXATTR
)
1295 req
->out
.h
.error
= -E2BIG
;
1296 request_end(fc
, req
);
1299 spin_lock(&fpq
->lock
);
1300 list_add(&req
->list
, &fpq
->io
);
1301 spin_unlock(&fpq
->lock
);
1303 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1305 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1306 (struct fuse_arg
*) in
->args
, 0);
1307 fuse_copy_finish(cs
);
1308 spin_lock(&fpq
->lock
);
1309 clear_bit(FR_LOCKED
, &req
->flags
);
1310 if (!fpq
->connected
) {
1311 err
= (fc
->aborted
&& fc
->abort_err
) ? -ECONNABORTED
: -ENODEV
;
1315 req
->out
.h
.error
= -EIO
;
1318 if (!test_bit(FR_ISREPLY
, &req
->flags
)) {
1322 hash
= fuse_req_hash(req
->in
.h
.unique
);
1323 list_move_tail(&req
->list
, &fpq
->processing
[hash
]);
1324 __fuse_get_request(req
);
1325 set_bit(FR_SENT
, &req
->flags
);
1326 spin_unlock(&fpq
->lock
);
1327 /* matches barrier in request_wait_answer() */
1328 smp_mb__after_atomic();
1329 if (test_bit(FR_INTERRUPTED
, &req
->flags
))
1330 queue_interrupt(fiq
, req
);
1331 fuse_put_request(fc
, req
);
1336 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1337 list_del_init(&req
->list
);
1338 spin_unlock(&fpq
->lock
);
1339 request_end(fc
, req
);
1343 spin_unlock(&fiq
->waitq
.lock
);
1347 static int fuse_dev_open(struct inode
*inode
, struct file
*file
)
1350 * The fuse device's file's private_data is used to hold
1351 * the fuse_conn(ection) when it is mounted, and is used to
1352 * keep track of whether the file has been mounted already.
1354 file
->private_data
= NULL
;
1358 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, struct iov_iter
*to
)
1360 struct fuse_copy_state cs
;
1361 struct file
*file
= iocb
->ki_filp
;
1362 struct fuse_dev
*fud
= fuse_get_dev(file
);
1367 if (!iter_is_iovec(to
))
1370 fuse_copy_init(&cs
, 1, to
);
1372 return fuse_dev_do_read(fud
, file
, &cs
, iov_iter_count(to
));
1375 static ssize_t
fuse_dev_splice_read(struct file
*in
, loff_t
*ppos
,
1376 struct pipe_inode_info
*pipe
,
1377 size_t len
, unsigned int flags
)
1381 struct pipe_buffer
*bufs
;
1382 struct fuse_copy_state cs
;
1383 struct fuse_dev
*fud
= fuse_get_dev(in
);
1388 bufs
= kvmalloc_array(pipe
->buffers
, sizeof(struct pipe_buffer
),
1393 fuse_copy_init(&cs
, 1, NULL
);
1396 ret
= fuse_dev_do_read(fud
, in
, &cs
, len
);
1400 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1405 for (ret
= total
= 0; page_nr
< cs
.nr_segs
; total
+= ret
) {
1407 * Need to be careful about this. Having buf->ops in module
1408 * code can Oops if the buffer persists after module unload.
1410 bufs
[page_nr
].ops
= &nosteal_pipe_buf_ops
;
1411 bufs
[page_nr
].flags
= 0;
1412 ret
= add_to_pipe(pipe
, &bufs
[page_nr
++]);
1413 if (unlikely(ret
< 0))
1419 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1420 put_page(bufs
[page_nr
].page
);
1426 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1427 struct fuse_copy_state
*cs
)
1429 struct fuse_notify_poll_wakeup_out outarg
;
1432 if (size
!= sizeof(outarg
))
1435 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1439 fuse_copy_finish(cs
);
1440 return fuse_notify_poll_wakeup(fc
, &outarg
);
1443 fuse_copy_finish(cs
);
1447 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1448 struct fuse_copy_state
*cs
)
1450 struct fuse_notify_inval_inode_out outarg
;
1453 if (size
!= sizeof(outarg
))
1456 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1459 fuse_copy_finish(cs
);
1461 down_read(&fc
->killsb
);
1464 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1465 outarg
.off
, outarg
.len
);
1467 up_read(&fc
->killsb
);
1471 fuse_copy_finish(cs
);
1475 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1476 struct fuse_copy_state
*cs
)
1478 struct fuse_notify_inval_entry_out outarg
;
1483 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1488 if (size
< sizeof(outarg
))
1491 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1495 err
= -ENAMETOOLONG
;
1496 if (outarg
.namelen
> FUSE_NAME_MAX
)
1500 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1504 name
.len
= outarg
.namelen
;
1505 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1508 fuse_copy_finish(cs
);
1509 buf
[outarg
.namelen
] = 0;
1511 down_read(&fc
->killsb
);
1514 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, 0, &name
);
1515 up_read(&fc
->killsb
);
1521 fuse_copy_finish(cs
);
1525 static int fuse_notify_delete(struct fuse_conn
*fc
, unsigned int size
,
1526 struct fuse_copy_state
*cs
)
1528 struct fuse_notify_delete_out outarg
;
1533 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1538 if (size
< sizeof(outarg
))
1541 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1545 err
= -ENAMETOOLONG
;
1546 if (outarg
.namelen
> FUSE_NAME_MAX
)
1550 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1554 name
.len
= outarg
.namelen
;
1555 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1558 fuse_copy_finish(cs
);
1559 buf
[outarg
.namelen
] = 0;
1561 down_read(&fc
->killsb
);
1564 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
,
1565 outarg
.child
, &name
);
1566 up_read(&fc
->killsb
);
1572 fuse_copy_finish(cs
);
1576 static int fuse_notify_store(struct fuse_conn
*fc
, unsigned int size
,
1577 struct fuse_copy_state
*cs
)
1579 struct fuse_notify_store_out outarg
;
1580 struct inode
*inode
;
1581 struct address_space
*mapping
;
1585 unsigned int offset
;
1591 if (size
< sizeof(outarg
))
1594 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1599 if (size
- sizeof(outarg
) != outarg
.size
)
1602 nodeid
= outarg
.nodeid
;
1604 down_read(&fc
->killsb
);
1610 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1614 mapping
= inode
->i_mapping
;
1615 index
= outarg
.offset
>> PAGE_SHIFT
;
1616 offset
= outarg
.offset
& ~PAGE_MASK
;
1617 file_size
= i_size_read(inode
);
1618 end
= outarg
.offset
+ outarg
.size
;
1619 if (end
> file_size
) {
1621 fuse_write_update_size(inode
, file_size
);
1627 unsigned int this_num
;
1630 page
= find_or_create_page(mapping
, index
,
1631 mapping_gfp_mask(mapping
));
1635 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1636 err
= fuse_copy_page(cs
, &page
, offset
, this_num
, 0);
1637 if (!err
&& offset
== 0 &&
1638 (this_num
== PAGE_SIZE
|| file_size
== end
))
1639 SetPageUptodate(page
);
1656 up_read(&fc
->killsb
);
1658 fuse_copy_finish(cs
);
1662 static void fuse_retrieve_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1664 release_pages(req
->pages
, req
->num_pages
);
1667 static int fuse_retrieve(struct fuse_conn
*fc
, struct inode
*inode
,
1668 struct fuse_notify_retrieve_out
*outarg
)
1671 struct address_space
*mapping
= inode
->i_mapping
;
1672 struct fuse_req
*req
;
1676 unsigned int offset
;
1677 size_t total_len
= 0;
1678 unsigned int num_pages
;
1680 offset
= outarg
->offset
& ~PAGE_MASK
;
1681 file_size
= i_size_read(inode
);
1684 if (outarg
->offset
> file_size
)
1686 else if (outarg
->offset
+ num
> file_size
)
1687 num
= file_size
- outarg
->offset
;
1689 num_pages
= (num
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1690 num_pages
= min(num_pages
, fc
->max_pages
);
1692 req
= fuse_get_req(fc
, num_pages
);
1694 return PTR_ERR(req
);
1696 req
->in
.h
.opcode
= FUSE_NOTIFY_REPLY
;
1697 req
->in
.h
.nodeid
= outarg
->nodeid
;
1698 req
->in
.numargs
= 2;
1699 req
->in
.argpages
= 1;
1700 req
->page_descs
[0].offset
= offset
;
1701 req
->end
= fuse_retrieve_end
;
1703 index
= outarg
->offset
>> PAGE_SHIFT
;
1705 while (num
&& req
->num_pages
< num_pages
) {
1707 unsigned int this_num
;
1709 page
= find_get_page(mapping
, index
);
1713 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1714 req
->pages
[req
->num_pages
] = page
;
1715 req
->page_descs
[req
->num_pages
].length
= this_num
;
1720 total_len
+= this_num
;
1723 req
->misc
.retrieve_in
.offset
= outarg
->offset
;
1724 req
->misc
.retrieve_in
.size
= total_len
;
1725 req
->in
.args
[0].size
= sizeof(req
->misc
.retrieve_in
);
1726 req
->in
.args
[0].value
= &req
->misc
.retrieve_in
;
1727 req
->in
.args
[1].size
= total_len
;
1729 err
= fuse_request_send_notify_reply(fc
, req
, outarg
->notify_unique
);
1731 fuse_retrieve_end(fc
, req
);
1736 static int fuse_notify_retrieve(struct fuse_conn
*fc
, unsigned int size
,
1737 struct fuse_copy_state
*cs
)
1739 struct fuse_notify_retrieve_out outarg
;
1740 struct inode
*inode
;
1744 if (size
!= sizeof(outarg
))
1747 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1751 fuse_copy_finish(cs
);
1753 down_read(&fc
->killsb
);
1756 u64 nodeid
= outarg
.nodeid
;
1758 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1760 err
= fuse_retrieve(fc
, inode
, &outarg
);
1764 up_read(&fc
->killsb
);
1769 fuse_copy_finish(cs
);
1773 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1774 unsigned int size
, struct fuse_copy_state
*cs
)
1776 /* Don't try to move pages (yet) */
1780 case FUSE_NOTIFY_POLL
:
1781 return fuse_notify_poll(fc
, size
, cs
);
1783 case FUSE_NOTIFY_INVAL_INODE
:
1784 return fuse_notify_inval_inode(fc
, size
, cs
);
1786 case FUSE_NOTIFY_INVAL_ENTRY
:
1787 return fuse_notify_inval_entry(fc
, size
, cs
);
1789 case FUSE_NOTIFY_STORE
:
1790 return fuse_notify_store(fc
, size
, cs
);
1792 case FUSE_NOTIFY_RETRIEVE
:
1793 return fuse_notify_retrieve(fc
, size
, cs
);
1795 case FUSE_NOTIFY_DELETE
:
1796 return fuse_notify_delete(fc
, size
, cs
);
1799 fuse_copy_finish(cs
);
1804 /* Look up request on processing list by unique ID */
1805 static struct fuse_req
*request_find(struct fuse_pqueue
*fpq
, u64 unique
)
1807 unsigned int hash
= fuse_req_hash(unique
);
1808 struct fuse_req
*req
;
1810 list_for_each_entry(req
, &fpq
->processing
[hash
], list
) {
1811 if (req
->in
.h
.unique
== unique
)
1817 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1820 unsigned reqsize
= sizeof(struct fuse_out_header
);
1823 return nbytes
!= reqsize
? -EINVAL
: 0;
1825 reqsize
+= len_args(out
->numargs
, out
->args
);
1827 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1829 else if (reqsize
> nbytes
) {
1830 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1831 unsigned diffsize
= reqsize
- nbytes
;
1832 if (diffsize
> lastarg
->size
)
1834 lastarg
->size
-= diffsize
;
1836 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1841 * Write a single reply to a request. First the header is copied from
1842 * the write buffer. The request is then searched on the processing
1843 * list by the unique ID found in the header. If found, then remove
1844 * it from the list and copy the rest of the buffer to the request.
1845 * The request is finished by calling request_end()
1847 static ssize_t
fuse_dev_do_write(struct fuse_dev
*fud
,
1848 struct fuse_copy_state
*cs
, size_t nbytes
)
1851 struct fuse_conn
*fc
= fud
->fc
;
1852 struct fuse_pqueue
*fpq
= &fud
->pq
;
1853 struct fuse_req
*req
;
1854 struct fuse_out_header oh
;
1856 if (nbytes
< sizeof(struct fuse_out_header
))
1859 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1864 if (oh
.len
!= nbytes
)
1868 * Zero oh.unique indicates unsolicited notification message
1869 * and error contains notification code.
1872 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1873 return err
? err
: nbytes
;
1877 if (oh
.error
<= -1000 || oh
.error
> 0)
1880 spin_lock(&fpq
->lock
);
1882 if (!fpq
->connected
)
1885 req
= request_find(fpq
, oh
.unique
& ~FUSE_INT_REQ_BIT
);
1889 /* Is it an interrupt reply ID? */
1890 if (oh
.unique
& FUSE_INT_REQ_BIT
) {
1891 __fuse_get_request(req
);
1892 spin_unlock(&fpq
->lock
);
1895 if (nbytes
!= sizeof(struct fuse_out_header
)) {
1896 fuse_put_request(fc
, req
);
1900 if (oh
.error
== -ENOSYS
)
1901 fc
->no_interrupt
= 1;
1902 else if (oh
.error
== -EAGAIN
)
1903 queue_interrupt(&fc
->iq
, req
);
1904 fuse_put_request(fc
, req
);
1906 fuse_copy_finish(cs
);
1910 clear_bit(FR_SENT
, &req
->flags
);
1911 list_move(&req
->list
, &fpq
->io
);
1913 set_bit(FR_LOCKED
, &req
->flags
);
1914 spin_unlock(&fpq
->lock
);
1916 if (!req
->out
.page_replace
)
1919 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1920 fuse_copy_finish(cs
);
1922 spin_lock(&fpq
->lock
);
1923 clear_bit(FR_LOCKED
, &req
->flags
);
1924 if (!fpq
->connected
)
1927 req
->out
.h
.error
= -EIO
;
1928 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1929 list_del_init(&req
->list
);
1930 spin_unlock(&fpq
->lock
);
1932 request_end(fc
, req
);
1934 return err
? err
: nbytes
;
1937 spin_unlock(&fpq
->lock
);
1939 fuse_copy_finish(cs
);
1943 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, struct iov_iter
*from
)
1945 struct fuse_copy_state cs
;
1946 struct fuse_dev
*fud
= fuse_get_dev(iocb
->ki_filp
);
1951 if (!iter_is_iovec(from
))
1954 fuse_copy_init(&cs
, 0, from
);
1956 return fuse_dev_do_write(fud
, &cs
, iov_iter_count(from
));
1959 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1960 struct file
*out
, loff_t
*ppos
,
1961 size_t len
, unsigned int flags
)
1965 struct pipe_buffer
*bufs
;
1966 struct fuse_copy_state cs
;
1967 struct fuse_dev
*fud
;
1971 fud
= fuse_get_dev(out
);
1977 bufs
= kvmalloc_array(pipe
->nrbufs
, sizeof(struct pipe_buffer
),
1986 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
1987 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
1997 struct pipe_buffer
*ibuf
;
1998 struct pipe_buffer
*obuf
;
2000 BUG_ON(nbuf
>= pipe
->buffers
);
2001 BUG_ON(!pipe
->nrbufs
);
2002 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
2005 if (rem
>= ibuf
->len
) {
2008 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
2011 pipe_buf_get(pipe
, ibuf
);
2013 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
2015 ibuf
->offset
+= obuf
->len
;
2016 ibuf
->len
-= obuf
->len
;
2023 fuse_copy_init(&cs
, 0, NULL
);
2028 if (flags
& SPLICE_F_MOVE
)
2031 ret
= fuse_dev_do_write(fud
, &cs
, len
);
2033 for (idx
= 0; idx
< nbuf
; idx
++)
2034 pipe_buf_release(pipe
, &bufs
[idx
]);
2041 static __poll_t
fuse_dev_poll(struct file
*file
, poll_table
*wait
)
2043 __poll_t mask
= EPOLLOUT
| EPOLLWRNORM
;
2044 struct fuse_iqueue
*fiq
;
2045 struct fuse_dev
*fud
= fuse_get_dev(file
);
2051 poll_wait(file
, &fiq
->waitq
, wait
);
2053 spin_lock(&fiq
->waitq
.lock
);
2054 if (!fiq
->connected
)
2056 else if (request_pending(fiq
))
2057 mask
|= EPOLLIN
| EPOLLRDNORM
;
2058 spin_unlock(&fiq
->waitq
.lock
);
2064 * Abort all requests on the given list (pending or processing)
2066 * This function releases and reacquires fc->lock
2068 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
2070 while (!list_empty(head
)) {
2071 struct fuse_req
*req
;
2072 req
= list_entry(head
->next
, struct fuse_req
, list
);
2073 req
->out
.h
.error
= -ECONNABORTED
;
2074 clear_bit(FR_SENT
, &req
->flags
);
2075 list_del_init(&req
->list
);
2076 request_end(fc
, req
);
2080 static void end_polls(struct fuse_conn
*fc
)
2084 p
= rb_first(&fc
->polled_files
);
2087 struct fuse_file
*ff
;
2088 ff
= rb_entry(p
, struct fuse_file
, polled_node
);
2089 wake_up_interruptible_all(&ff
->poll_wait
);
2096 * Abort all requests.
2098 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2101 * The same effect is usually achievable through killing the filesystem daemon
2102 * and all users of the filesystem. The exception is the combination of an
2103 * asynchronous request and the tricky deadlock (see
2104 * Documentation/filesystems/fuse.txt).
2106 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2107 * requests, they should be finished off immediately. Locked requests will be
2108 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2109 * requests. It is possible that some request will finish before we can. This
2110 * is OK, the request will in that case be removed from the list before we touch
2113 void fuse_abort_conn(struct fuse_conn
*fc
, bool is_abort
)
2115 struct fuse_iqueue
*fiq
= &fc
->iq
;
2117 spin_lock(&fc
->lock
);
2118 if (fc
->connected
) {
2119 struct fuse_dev
*fud
;
2120 struct fuse_req
*req
, *next
;
2124 /* Background queuing checks fc->connected under bg_lock */
2125 spin_lock(&fc
->bg_lock
);
2127 spin_unlock(&fc
->bg_lock
);
2129 fc
->aborted
= is_abort
;
2130 fuse_set_initialized(fc
);
2131 list_for_each_entry(fud
, &fc
->devices
, entry
) {
2132 struct fuse_pqueue
*fpq
= &fud
->pq
;
2134 spin_lock(&fpq
->lock
);
2136 list_for_each_entry_safe(req
, next
, &fpq
->io
, list
) {
2137 req
->out
.h
.error
= -ECONNABORTED
;
2138 spin_lock(&req
->waitq
.lock
);
2139 set_bit(FR_ABORTED
, &req
->flags
);
2140 if (!test_bit(FR_LOCKED
, &req
->flags
)) {
2141 set_bit(FR_PRIVATE
, &req
->flags
);
2142 __fuse_get_request(req
);
2143 list_move(&req
->list
, &to_end
);
2145 spin_unlock(&req
->waitq
.lock
);
2147 for (i
= 0; i
< FUSE_PQ_HASH_SIZE
; i
++)
2148 list_splice_tail_init(&fpq
->processing
[i
],
2150 spin_unlock(&fpq
->lock
);
2152 spin_lock(&fc
->bg_lock
);
2154 fc
->max_background
= UINT_MAX
;
2156 spin_unlock(&fc
->bg_lock
);
2158 spin_lock(&fiq
->waitq
.lock
);
2160 list_for_each_entry(req
, &fiq
->pending
, list
)
2161 clear_bit(FR_PENDING
, &req
->flags
);
2162 list_splice_tail_init(&fiq
->pending
, &to_end
);
2163 while (forget_pending(fiq
))
2164 kfree(dequeue_forget(fiq
, 1, NULL
));
2165 wake_up_all_locked(&fiq
->waitq
);
2166 spin_unlock(&fiq
->waitq
.lock
);
2167 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
2169 wake_up_all(&fc
->blocked_waitq
);
2170 spin_unlock(&fc
->lock
);
2172 end_requests(fc
, &to_end
);
2174 spin_unlock(&fc
->lock
);
2177 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
2179 void fuse_wait_aborted(struct fuse_conn
*fc
)
2181 wait_event(fc
->blocked_waitq
, atomic_read(&fc
->num_waiting
) == 0);
2184 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
2186 struct fuse_dev
*fud
= fuse_get_dev(file
);
2189 struct fuse_conn
*fc
= fud
->fc
;
2190 struct fuse_pqueue
*fpq
= &fud
->pq
;
2194 spin_lock(&fpq
->lock
);
2195 WARN_ON(!list_empty(&fpq
->io
));
2196 for (i
= 0; i
< FUSE_PQ_HASH_SIZE
; i
++)
2197 list_splice_init(&fpq
->processing
[i
], &to_end
);
2198 spin_unlock(&fpq
->lock
);
2200 end_requests(fc
, &to_end
);
2202 /* Are we the last open device? */
2203 if (atomic_dec_and_test(&fc
->dev_count
)) {
2204 WARN_ON(fc
->iq
.fasync
!= NULL
);
2205 fuse_abort_conn(fc
, false);
2211 EXPORT_SYMBOL_GPL(fuse_dev_release
);
2213 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
2215 struct fuse_dev
*fud
= fuse_get_dev(file
);
2220 /* No locking - fasync_helper does its own locking */
2221 return fasync_helper(fd
, file
, on
, &fud
->fc
->iq
.fasync
);
2224 static int fuse_device_clone(struct fuse_conn
*fc
, struct file
*new)
2226 struct fuse_dev
*fud
;
2228 if (new->private_data
)
2231 fud
= fuse_dev_alloc(fc
);
2235 new->private_data
= fud
;
2236 atomic_inc(&fc
->dev_count
);
2241 static long fuse_dev_ioctl(struct file
*file
, unsigned int cmd
,
2246 if (cmd
== FUSE_DEV_IOC_CLONE
) {
2250 if (!get_user(oldfd
, (__u32 __user
*) arg
)) {
2251 struct file
*old
= fget(oldfd
);
2255 struct fuse_dev
*fud
= NULL
;
2258 * Check against file->f_op because CUSE
2259 * uses the same ioctl handler.
2261 if (old
->f_op
== file
->f_op
&&
2262 old
->f_cred
->user_ns
== file
->f_cred
->user_ns
)
2263 fud
= fuse_get_dev(old
);
2266 mutex_lock(&fuse_mutex
);
2267 err
= fuse_device_clone(fud
->fc
, file
);
2268 mutex_unlock(&fuse_mutex
);
2277 const struct file_operations fuse_dev_operations
= {
2278 .owner
= THIS_MODULE
,
2279 .open
= fuse_dev_open
,
2280 .llseek
= no_llseek
,
2281 .read_iter
= fuse_dev_read
,
2282 .splice_read
= fuse_dev_splice_read
,
2283 .write_iter
= fuse_dev_write
,
2284 .splice_write
= fuse_dev_splice_write
,
2285 .poll
= fuse_dev_poll
,
2286 .release
= fuse_dev_release
,
2287 .fasync
= fuse_dev_fasync
,
2288 .unlocked_ioctl
= fuse_dev_ioctl
,
2289 .compat_ioctl
= fuse_dev_ioctl
,
2291 EXPORT_SYMBOL_GPL(fuse_dev_operations
);
2293 static struct miscdevice fuse_miscdevice
= {
2294 .minor
= FUSE_MINOR
,
2296 .fops
= &fuse_dev_operations
,
2299 int __init
fuse_dev_init(void)
2302 fuse_req_cachep
= kmem_cache_create("fuse_request",
2303 sizeof(struct fuse_req
),
2305 if (!fuse_req_cachep
)
2308 err
= misc_register(&fuse_miscdevice
);
2310 goto out_cache_clean
;
2315 kmem_cache_destroy(fuse_req_cachep
);
2320 void fuse_dev_cleanup(void)
2322 misc_deregister(&fuse_miscdevice
);
2323 kmem_cache_destroy(fuse_req_cachep
);