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/namei.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/slab.h>
21 #include <linux/pipe_fs_i.h>
22 #include <linux/swap.h>
23 #include <linux/splice.h>
24 #include <linux/sched.h>
25 #include <linux/freezer.h>
27 MODULE_ALIAS_MISCDEV(FUSE_MINOR
);
28 MODULE_ALIAS("devname:fuse");
30 static struct kmem_cache
*fuse_req_cachep
;
32 static struct fuse_dev
*fuse_get_dev(struct file
*file
)
35 * Lockless access is OK, because file->private data is set
36 * once during mount and is valid until the file is released.
38 return ACCESS_ONCE(file
->private_data
);
41 static void fuse_request_init(struct fuse_req
*req
, struct page
**pages
,
42 struct fuse_page_desc
*page_descs
,
45 memset(req
, 0, sizeof(*req
));
46 memset(pages
, 0, sizeof(*pages
) * npages
);
47 memset(page_descs
, 0, sizeof(*page_descs
) * npages
);
48 INIT_LIST_HEAD(&req
->list
);
49 INIT_LIST_HEAD(&req
->intr_entry
);
50 init_waitqueue_head(&req
->waitq
);
51 refcount_set(&req
->count
, 1);
53 req
->page_descs
= page_descs
;
54 req
->max_pages
= npages
;
55 __set_bit(FR_PENDING
, &req
->flags
);
58 static struct fuse_req
*__fuse_request_alloc(unsigned npages
, gfp_t flags
)
60 struct fuse_req
*req
= kmem_cache_alloc(fuse_req_cachep
, flags
);
63 struct fuse_page_desc
*page_descs
;
65 if (npages
<= FUSE_REQ_INLINE_PAGES
) {
66 pages
= req
->inline_pages
;
67 page_descs
= req
->inline_page_descs
;
69 pages
= kmalloc(sizeof(struct page
*) * npages
, flags
);
70 page_descs
= kmalloc(sizeof(struct fuse_page_desc
) *
74 if (!pages
|| !page_descs
) {
77 kmem_cache_free(fuse_req_cachep
, req
);
81 fuse_request_init(req
, pages
, page_descs
, npages
);
86 struct fuse_req
*fuse_request_alloc(unsigned npages
)
88 return __fuse_request_alloc(npages
, GFP_KERNEL
);
90 EXPORT_SYMBOL_GPL(fuse_request_alloc
);
92 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
)
94 return __fuse_request_alloc(npages
, GFP_NOFS
);
97 void fuse_request_free(struct fuse_req
*req
)
99 if (req
->pages
!= req
->inline_pages
) {
101 kfree(req
->page_descs
);
103 kmem_cache_free(fuse_req_cachep
, req
);
106 void __fuse_get_request(struct fuse_req
*req
)
108 refcount_inc(&req
->count
);
111 /* Must be called with > 1 refcount */
112 static void __fuse_put_request(struct fuse_req
*req
)
114 refcount_dec(&req
->count
);
117 static void fuse_req_init_context(struct fuse_conn
*fc
, struct fuse_req
*req
)
119 req
->in
.h
.uid
= from_kuid_munged(&init_user_ns
, current_fsuid());
120 req
->in
.h
.gid
= from_kgid_munged(&init_user_ns
, current_fsgid());
121 req
->in
.h
.pid
= pid_nr_ns(task_pid(current
), fc
->pid_ns
);
124 void fuse_set_initialized(struct fuse_conn
*fc
)
126 /* Make sure stores before this are seen on another CPU */
131 static bool fuse_block_alloc(struct fuse_conn
*fc
, bool for_background
)
133 return !fc
->initialized
|| (for_background
&& fc
->blocked
);
136 static struct fuse_req
*__fuse_get_req(struct fuse_conn
*fc
, unsigned npages
,
139 struct fuse_req
*req
;
141 atomic_inc(&fc
->num_waiting
);
143 if (fuse_block_alloc(fc
, for_background
)) {
145 if (wait_event_killable_exclusive(fc
->blocked_waitq
,
146 !fuse_block_alloc(fc
, for_background
)))
149 /* Matches smp_wmb() in fuse_set_initialized() */
160 req
= fuse_request_alloc(npages
);
164 wake_up(&fc
->blocked_waitq
);
168 fuse_req_init_context(fc
, req
);
169 __set_bit(FR_WAITING
, &req
->flags
);
171 __set_bit(FR_BACKGROUND
, &req
->flags
);
176 atomic_dec(&fc
->num_waiting
);
180 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
)
182 return __fuse_get_req(fc
, npages
, false);
184 EXPORT_SYMBOL_GPL(fuse_get_req
);
186 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
189 return __fuse_get_req(fc
, npages
, true);
191 EXPORT_SYMBOL_GPL(fuse_get_req_for_background
);
194 * Return request in fuse_file->reserved_req. However that may
195 * currently be in use. If that is the case, wait for it to become
198 static struct fuse_req
*get_reserved_req(struct fuse_conn
*fc
,
201 struct fuse_req
*req
= NULL
;
202 struct fuse_file
*ff
= file
->private_data
;
205 wait_event(fc
->reserved_req_waitq
, ff
->reserved_req
);
206 spin_lock(&fc
->lock
);
207 if (ff
->reserved_req
) {
208 req
= ff
->reserved_req
;
209 ff
->reserved_req
= NULL
;
210 req
->stolen_file
= get_file(file
);
212 spin_unlock(&fc
->lock
);
219 * Put stolen request back into fuse_file->reserved_req
221 static void put_reserved_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
223 struct file
*file
= req
->stolen_file
;
224 struct fuse_file
*ff
= file
->private_data
;
226 spin_lock(&fc
->lock
);
227 fuse_request_init(req
, req
->pages
, req
->page_descs
, req
->max_pages
);
228 BUG_ON(ff
->reserved_req
);
229 ff
->reserved_req
= req
;
230 wake_up_all(&fc
->reserved_req_waitq
);
231 spin_unlock(&fc
->lock
);
236 * Gets a requests for a file operation, always succeeds
238 * This is used for sending the FLUSH request, which must get to
239 * userspace, due to POSIX locks which may need to be unlocked.
241 * If allocation fails due to OOM, use the reserved request in
244 * This is very unlikely to deadlock accidentally, since the
245 * filesystem should not have it's own file open. If deadlock is
246 * intentional, it can still be broken by "aborting" the filesystem.
248 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
251 struct fuse_req
*req
;
253 atomic_inc(&fc
->num_waiting
);
254 wait_event(fc
->blocked_waitq
, fc
->initialized
);
255 /* Matches smp_wmb() in fuse_set_initialized() */
257 req
= fuse_request_alloc(0);
259 req
= get_reserved_req(fc
, file
);
261 fuse_req_init_context(fc
, req
);
262 __set_bit(FR_WAITING
, &req
->flags
);
263 __clear_bit(FR_BACKGROUND
, &req
->flags
);
267 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
269 if (refcount_dec_and_test(&req
->count
)) {
270 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
272 * We get here in the unlikely case that a background
273 * request was allocated but not sent
275 spin_lock(&fc
->lock
);
277 wake_up(&fc
->blocked_waitq
);
278 spin_unlock(&fc
->lock
);
281 if (test_bit(FR_WAITING
, &req
->flags
)) {
282 __clear_bit(FR_WAITING
, &req
->flags
);
283 atomic_dec(&fc
->num_waiting
);
286 if (req
->stolen_file
)
287 put_reserved_req(fc
, req
);
289 fuse_request_free(req
);
292 EXPORT_SYMBOL_GPL(fuse_put_request
);
294 static unsigned len_args(unsigned numargs
, struct fuse_arg
*args
)
299 for (i
= 0; i
< numargs
; i
++)
300 nbytes
+= args
[i
].size
;
305 static u64
fuse_get_unique(struct fuse_iqueue
*fiq
)
307 return ++fiq
->reqctr
;
310 static void queue_request(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
312 req
->in
.h
.len
= sizeof(struct fuse_in_header
) +
313 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
314 list_add_tail(&req
->list
, &fiq
->pending
);
315 wake_up_locked(&fiq
->waitq
);
316 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
319 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
320 u64 nodeid
, u64 nlookup
)
322 struct fuse_iqueue
*fiq
= &fc
->iq
;
324 forget
->forget_one
.nodeid
= nodeid
;
325 forget
->forget_one
.nlookup
= nlookup
;
327 spin_lock(&fiq
->waitq
.lock
);
328 if (fiq
->connected
) {
329 fiq
->forget_list_tail
->next
= forget
;
330 fiq
->forget_list_tail
= forget
;
331 wake_up_locked(&fiq
->waitq
);
332 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
336 spin_unlock(&fiq
->waitq
.lock
);
339 static void flush_bg_queue(struct fuse_conn
*fc
)
341 while (fc
->active_background
< fc
->max_background
&&
342 !list_empty(&fc
->bg_queue
)) {
343 struct fuse_req
*req
;
344 struct fuse_iqueue
*fiq
= &fc
->iq
;
346 req
= list_entry(fc
->bg_queue
.next
, struct fuse_req
, list
);
347 list_del(&req
->list
);
348 fc
->active_background
++;
349 spin_lock(&fiq
->waitq
.lock
);
350 req
->in
.h
.unique
= fuse_get_unique(fiq
);
351 queue_request(fiq
, req
);
352 spin_unlock(&fiq
->waitq
.lock
);
357 * This function is called when a request is finished. Either a reply
358 * has arrived or it was aborted (and not yet sent) or some error
359 * occurred during communication with userspace, or the device file
360 * was closed. The requester thread is woken up (if still waiting),
361 * the 'end' callback is called if given, else the reference to the
362 * request is released
364 static void request_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
366 struct fuse_iqueue
*fiq
= &fc
->iq
;
368 if (test_and_set_bit(FR_FINISHED
, &req
->flags
))
371 spin_lock(&fiq
->waitq
.lock
);
372 list_del_init(&req
->intr_entry
);
373 spin_unlock(&fiq
->waitq
.lock
);
374 WARN_ON(test_bit(FR_PENDING
, &req
->flags
));
375 WARN_ON(test_bit(FR_SENT
, &req
->flags
));
376 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
377 spin_lock(&fc
->lock
);
378 clear_bit(FR_BACKGROUND
, &req
->flags
);
379 if (fc
->num_background
== fc
->max_background
)
382 /* Wake up next waiter, if any */
383 if (!fc
->blocked
&& waitqueue_active(&fc
->blocked_waitq
))
384 wake_up(&fc
->blocked_waitq
);
386 if (fc
->num_background
== fc
->congestion_threshold
&&
387 fc
->connected
&& fc
->sb
) {
388 clear_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_SYNC
);
389 clear_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_ASYNC
);
391 fc
->num_background
--;
392 fc
->active_background
--;
394 spin_unlock(&fc
->lock
);
396 wake_up(&req
->waitq
);
399 fuse_put_request(fc
, req
);
402 static void queue_interrupt(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
404 spin_lock(&fiq
->waitq
.lock
);
405 if (test_bit(FR_FINISHED
, &req
->flags
)) {
406 spin_unlock(&fiq
->waitq
.lock
);
409 if (list_empty(&req
->intr_entry
)) {
410 list_add_tail(&req
->intr_entry
, &fiq
->interrupts
);
411 wake_up_locked(&fiq
->waitq
);
413 spin_unlock(&fiq
->waitq
.lock
);
414 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
417 static void request_wait_answer(struct fuse_conn
*fc
, struct fuse_req
*req
)
419 struct fuse_iqueue
*fiq
= &fc
->iq
;
422 if (!fc
->no_interrupt
) {
423 /* Any signal may interrupt this */
424 err
= wait_event_interruptible(req
->waitq
,
425 test_bit(FR_FINISHED
, &req
->flags
));
429 set_bit(FR_INTERRUPTED
, &req
->flags
);
430 /* matches barrier in fuse_dev_do_read() */
431 smp_mb__after_atomic();
432 if (test_bit(FR_SENT
, &req
->flags
))
433 queue_interrupt(fiq
, req
);
436 if (!test_bit(FR_FORCE
, &req
->flags
)) {
437 /* Only fatal signals may interrupt this */
438 err
= wait_event_killable(req
->waitq
,
439 test_bit(FR_FINISHED
, &req
->flags
));
443 spin_lock(&fiq
->waitq
.lock
);
444 /* Request is not yet in userspace, bail out */
445 if (test_bit(FR_PENDING
, &req
->flags
)) {
446 list_del(&req
->list
);
447 spin_unlock(&fiq
->waitq
.lock
);
448 __fuse_put_request(req
);
449 req
->out
.h
.error
= -EINTR
;
452 spin_unlock(&fiq
->waitq
.lock
);
456 * Either request is already in userspace, or it was forced.
459 while (!test_bit(FR_FINISHED
, &req
->flags
))
460 wait_event_freezable(req
->waitq
,
461 test_bit(FR_FINISHED
, &req
->flags
));
464 static void __fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
466 struct fuse_iqueue
*fiq
= &fc
->iq
;
468 BUG_ON(test_bit(FR_BACKGROUND
, &req
->flags
));
469 spin_lock(&fiq
->waitq
.lock
);
470 if (!fiq
->connected
) {
471 spin_unlock(&fiq
->waitq
.lock
);
472 req
->out
.h
.error
= -ENOTCONN
;
474 req
->in
.h
.unique
= fuse_get_unique(fiq
);
475 queue_request(fiq
, req
);
476 /* acquire extra reference, since request is still needed
477 after request_end() */
478 __fuse_get_request(req
);
479 spin_unlock(&fiq
->waitq
.lock
);
481 request_wait_answer(fc
, req
);
482 /* Pairs with smp_wmb() in request_end() */
487 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
489 __set_bit(FR_ISREPLY
, &req
->flags
);
490 if (!test_bit(FR_WAITING
, &req
->flags
)) {
491 __set_bit(FR_WAITING
, &req
->flags
);
492 atomic_inc(&fc
->num_waiting
);
494 __fuse_request_send(fc
, req
);
496 EXPORT_SYMBOL_GPL(fuse_request_send
);
498 static void fuse_adjust_compat(struct fuse_conn
*fc
, struct fuse_args
*args
)
500 if (fc
->minor
< 4 && args
->in
.h
.opcode
== FUSE_STATFS
)
501 args
->out
.args
[0].size
= FUSE_COMPAT_STATFS_SIZE
;
504 switch (args
->in
.h
.opcode
) {
511 args
->out
.args
[0].size
= FUSE_COMPAT_ENTRY_OUT_SIZE
;
515 args
->out
.args
[0].size
= FUSE_COMPAT_ATTR_OUT_SIZE
;
519 if (fc
->minor
< 12) {
520 switch (args
->in
.h
.opcode
) {
522 args
->in
.args
[0].size
= sizeof(struct fuse_open_in
);
525 args
->in
.args
[0].size
= FUSE_COMPAT_MKNOD_IN_SIZE
;
531 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
)
533 struct fuse_req
*req
;
536 req
= fuse_get_req(fc
, 0);
540 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
541 fuse_adjust_compat(fc
, args
);
543 req
->in
.h
.opcode
= args
->in
.h
.opcode
;
544 req
->in
.h
.nodeid
= args
->in
.h
.nodeid
;
545 req
->in
.numargs
= args
->in
.numargs
;
546 memcpy(req
->in
.args
, args
->in
.args
,
547 args
->in
.numargs
* sizeof(struct fuse_in_arg
));
548 req
->out
.argvar
= args
->out
.argvar
;
549 req
->out
.numargs
= args
->out
.numargs
;
550 memcpy(req
->out
.args
, args
->out
.args
,
551 args
->out
.numargs
* sizeof(struct fuse_arg
));
552 fuse_request_send(fc
, req
);
553 ret
= req
->out
.h
.error
;
554 if (!ret
&& args
->out
.argvar
) {
555 BUG_ON(args
->out
.numargs
!= 1);
556 ret
= req
->out
.args
[0].size
;
558 fuse_put_request(fc
, req
);
564 * Called under fc->lock
566 * fc->connected must have been checked previously
568 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
569 struct fuse_req
*req
)
571 BUG_ON(!test_bit(FR_BACKGROUND
, &req
->flags
));
572 if (!test_bit(FR_WAITING
, &req
->flags
)) {
573 __set_bit(FR_WAITING
, &req
->flags
);
574 atomic_inc(&fc
->num_waiting
);
576 __set_bit(FR_ISREPLY
, &req
->flags
);
577 fc
->num_background
++;
578 if (fc
->num_background
== fc
->max_background
)
580 if (fc
->num_background
== fc
->congestion_threshold
&& fc
->sb
) {
581 set_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_SYNC
);
582 set_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_ASYNC
);
584 list_add_tail(&req
->list
, &fc
->bg_queue
);
588 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
591 spin_lock(&fc
->lock
);
593 fuse_request_send_background_locked(fc
, req
);
594 spin_unlock(&fc
->lock
);
596 spin_unlock(&fc
->lock
);
597 req
->out
.h
.error
= -ENOTCONN
;
599 fuse_put_request(fc
, req
);
602 EXPORT_SYMBOL_GPL(fuse_request_send_background
);
604 static int fuse_request_send_notify_reply(struct fuse_conn
*fc
,
605 struct fuse_req
*req
, u64 unique
)
608 struct fuse_iqueue
*fiq
= &fc
->iq
;
610 __clear_bit(FR_ISREPLY
, &req
->flags
);
611 req
->in
.h
.unique
= unique
;
612 spin_lock(&fiq
->waitq
.lock
);
613 if (fiq
->connected
) {
614 queue_request(fiq
, req
);
617 spin_unlock(&fiq
->waitq
.lock
);
622 void fuse_force_forget(struct file
*file
, u64 nodeid
)
624 struct inode
*inode
= file_inode(file
);
625 struct fuse_conn
*fc
= get_fuse_conn(inode
);
626 struct fuse_req
*req
;
627 struct fuse_forget_in inarg
;
629 memset(&inarg
, 0, sizeof(inarg
));
631 req
= fuse_get_req_nofail_nopages(fc
, file
);
632 req
->in
.h
.opcode
= FUSE_FORGET
;
633 req
->in
.h
.nodeid
= nodeid
;
635 req
->in
.args
[0].size
= sizeof(inarg
);
636 req
->in
.args
[0].value
= &inarg
;
637 __clear_bit(FR_ISREPLY
, &req
->flags
);
638 __fuse_request_send(fc
, req
);
640 fuse_put_request(fc
, req
);
644 * Lock the request. Up to the next unlock_request() there mustn't be
645 * anything that could cause a page-fault. If the request was already
648 static int lock_request(struct fuse_req
*req
)
652 spin_lock(&req
->waitq
.lock
);
653 if (test_bit(FR_ABORTED
, &req
->flags
))
656 set_bit(FR_LOCKED
, &req
->flags
);
657 spin_unlock(&req
->waitq
.lock
);
663 * Unlock request. If it was aborted while locked, caller is responsible
664 * for unlocking and ending the request.
666 static int unlock_request(struct fuse_req
*req
)
670 spin_lock(&req
->waitq
.lock
);
671 if (test_bit(FR_ABORTED
, &req
->flags
))
674 clear_bit(FR_LOCKED
, &req
->flags
);
675 spin_unlock(&req
->waitq
.lock
);
680 struct fuse_copy_state
{
682 struct fuse_req
*req
;
683 struct iov_iter
*iter
;
684 struct pipe_buffer
*pipebufs
;
685 struct pipe_buffer
*currbuf
;
686 struct pipe_inode_info
*pipe
;
687 unsigned long nr_segs
;
691 unsigned move_pages
:1;
694 static void fuse_copy_init(struct fuse_copy_state
*cs
, int write
,
695 struct iov_iter
*iter
)
697 memset(cs
, 0, sizeof(*cs
));
702 /* Unmap and put previous page of userspace buffer */
703 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
706 struct pipe_buffer
*buf
= cs
->currbuf
;
709 buf
->len
= PAGE_SIZE
- cs
->len
;
713 flush_dcache_page(cs
->pg
);
714 set_page_dirty_lock(cs
->pg
);
722 * Get another pagefull of userspace buffer, and map it to kernel
723 * address space, and lock request
725 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
730 err
= unlock_request(cs
->req
);
734 fuse_copy_finish(cs
);
736 struct pipe_buffer
*buf
= cs
->pipebufs
;
739 err
= pipe_buf_confirm(cs
->pipe
, buf
);
743 BUG_ON(!cs
->nr_segs
);
746 cs
->offset
= buf
->offset
;
751 if (cs
->nr_segs
== cs
->pipe
->buffers
)
754 page
= alloc_page(GFP_HIGHUSER
);
771 err
= iov_iter_get_pages(cs
->iter
, &page
, PAGE_SIZE
, 1, &off
);
778 iov_iter_advance(cs
->iter
, err
);
781 return lock_request(cs
->req
);
784 /* Do as much copy to/from userspace buffer as we can */
785 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
787 unsigned ncpy
= min(*size
, cs
->len
);
789 void *pgaddr
= kmap_atomic(cs
->pg
);
790 void *buf
= pgaddr
+ cs
->offset
;
793 memcpy(buf
, *val
, ncpy
);
795 memcpy(*val
, buf
, ncpy
);
797 kunmap_atomic(pgaddr
);
806 static int fuse_check_page(struct page
*page
)
808 if (page_mapcount(page
) ||
809 page
->mapping
!= NULL
||
810 page_count(page
) != 1 ||
811 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
818 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
819 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
);
825 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
828 struct page
*oldpage
= *pagep
;
829 struct page
*newpage
;
830 struct pipe_buffer
*buf
= cs
->pipebufs
;
832 err
= unlock_request(cs
->req
);
836 fuse_copy_finish(cs
);
838 err
= pipe_buf_confirm(cs
->pipe
, buf
);
842 BUG_ON(!cs
->nr_segs
);
848 if (cs
->len
!= PAGE_SIZE
)
851 if (pipe_buf_steal(cs
->pipe
, buf
) != 0)
856 if (!PageUptodate(newpage
))
857 SetPageUptodate(newpage
);
859 ClearPageMappedToDisk(newpage
);
861 if (fuse_check_page(newpage
) != 0)
862 goto out_fallback_unlock
;
865 * This is a new and locked page, it shouldn't be mapped or
866 * have any special flags on it
868 if (WARN_ON(page_mapped(oldpage
)))
869 goto out_fallback_unlock
;
870 if (WARN_ON(page_has_private(oldpage
)))
871 goto out_fallback_unlock
;
872 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
873 goto out_fallback_unlock
;
874 if (WARN_ON(PageMlocked(oldpage
)))
875 goto out_fallback_unlock
;
877 err
= replace_page_cache_page(oldpage
, newpage
, GFP_KERNEL
);
879 unlock_page(newpage
);
885 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
886 lru_cache_add_file(newpage
);
889 spin_lock(&cs
->req
->waitq
.lock
);
890 if (test_bit(FR_ABORTED
, &cs
->req
->flags
))
894 spin_unlock(&cs
->req
->waitq
.lock
);
897 unlock_page(newpage
);
902 unlock_page(oldpage
);
909 unlock_page(newpage
);
912 cs
->offset
= buf
->offset
;
914 err
= lock_request(cs
->req
);
921 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
922 unsigned offset
, unsigned count
)
924 struct pipe_buffer
*buf
;
927 if (cs
->nr_segs
== cs
->pipe
->buffers
)
930 err
= unlock_request(cs
->req
);
934 fuse_copy_finish(cs
);
939 buf
->offset
= offset
;
950 * Copy a page in the request to/from the userspace buffer. Must be
953 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
954 unsigned offset
, unsigned count
, int zeroing
)
957 struct page
*page
= *pagep
;
959 if (page
&& zeroing
&& count
< PAGE_SIZE
)
960 clear_highpage(page
);
963 if (cs
->write
&& cs
->pipebufs
&& page
) {
964 return fuse_ref_page(cs
, page
, offset
, count
);
965 } else if (!cs
->len
) {
966 if (cs
->move_pages
&& page
&&
967 offset
== 0 && count
== PAGE_SIZE
) {
968 err
= fuse_try_move_page(cs
, pagep
);
972 err
= fuse_copy_fill(cs
);
978 void *mapaddr
= kmap_atomic(page
);
979 void *buf
= mapaddr
+ offset
;
980 offset
+= fuse_copy_do(cs
, &buf
, &count
);
981 kunmap_atomic(mapaddr
);
983 offset
+= fuse_copy_do(cs
, NULL
, &count
);
985 if (page
&& !cs
->write
)
986 flush_dcache_page(page
);
990 /* Copy pages in the request to/from userspace buffer */
991 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
995 struct fuse_req
*req
= cs
->req
;
997 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
999 unsigned offset
= req
->page_descs
[i
].offset
;
1000 unsigned count
= min(nbytes
, req
->page_descs
[i
].length
);
1002 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
1012 /* Copy a single argument in the request to/from userspace buffer */
1013 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
1017 int err
= fuse_copy_fill(cs
);
1021 fuse_copy_do(cs
, &val
, &size
);
1026 /* Copy request arguments to/from userspace buffer */
1027 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
1028 unsigned argpages
, struct fuse_arg
*args
,
1034 for (i
= 0; !err
&& i
< numargs
; i
++) {
1035 struct fuse_arg
*arg
= &args
[i
];
1036 if (i
== numargs
- 1 && argpages
)
1037 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
1039 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
1044 static int forget_pending(struct fuse_iqueue
*fiq
)
1046 return fiq
->forget_list_head
.next
!= NULL
;
1049 static int request_pending(struct fuse_iqueue
*fiq
)
1051 return !list_empty(&fiq
->pending
) || !list_empty(&fiq
->interrupts
) ||
1052 forget_pending(fiq
);
1056 * Transfer an interrupt request to userspace
1058 * Unlike other requests this is assembled on demand, without a need
1059 * to allocate a separate fuse_req structure.
1061 * Called with fiq->waitq.lock held, releases it
1063 static int fuse_read_interrupt(struct fuse_iqueue
*fiq
,
1064 struct fuse_copy_state
*cs
,
1065 size_t nbytes
, struct fuse_req
*req
)
1066 __releases(fiq
->waitq
.lock
)
1068 struct fuse_in_header ih
;
1069 struct fuse_interrupt_in arg
;
1070 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
1073 list_del_init(&req
->intr_entry
);
1074 req
->intr_unique
= fuse_get_unique(fiq
);
1075 memset(&ih
, 0, sizeof(ih
));
1076 memset(&arg
, 0, sizeof(arg
));
1078 ih
.opcode
= FUSE_INTERRUPT
;
1079 ih
.unique
= req
->intr_unique
;
1080 arg
.unique
= req
->in
.h
.unique
;
1082 spin_unlock(&fiq
->waitq
.lock
);
1083 if (nbytes
< reqsize
)
1086 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1088 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1089 fuse_copy_finish(cs
);
1091 return err
? err
: reqsize
;
1094 static struct fuse_forget_link
*dequeue_forget(struct fuse_iqueue
*fiq
,
1098 struct fuse_forget_link
*head
= fiq
->forget_list_head
.next
;
1099 struct fuse_forget_link
**newhead
= &head
;
1102 for (count
= 0; *newhead
!= NULL
&& count
< max
; count
++)
1103 newhead
= &(*newhead
)->next
;
1105 fiq
->forget_list_head
.next
= *newhead
;
1107 if (fiq
->forget_list_head
.next
== NULL
)
1108 fiq
->forget_list_tail
= &fiq
->forget_list_head
;
1116 static int fuse_read_single_forget(struct fuse_iqueue
*fiq
,
1117 struct fuse_copy_state
*cs
,
1119 __releases(fiq
->waitq
.lock
)
1122 struct fuse_forget_link
*forget
= dequeue_forget(fiq
, 1, NULL
);
1123 struct fuse_forget_in arg
= {
1124 .nlookup
= forget
->forget_one
.nlookup
,
1126 struct fuse_in_header ih
= {
1127 .opcode
= FUSE_FORGET
,
1128 .nodeid
= forget
->forget_one
.nodeid
,
1129 .unique
= fuse_get_unique(fiq
),
1130 .len
= sizeof(ih
) + sizeof(arg
),
1133 spin_unlock(&fiq
->waitq
.lock
);
1135 if (nbytes
< ih
.len
)
1138 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1140 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1141 fuse_copy_finish(cs
);
1149 static int fuse_read_batch_forget(struct fuse_iqueue
*fiq
,
1150 struct fuse_copy_state
*cs
, size_t nbytes
)
1151 __releases(fiq
->waitq
.lock
)
1154 unsigned max_forgets
;
1156 struct fuse_forget_link
*head
;
1157 struct fuse_batch_forget_in arg
= { .count
= 0 };
1158 struct fuse_in_header ih
= {
1159 .opcode
= FUSE_BATCH_FORGET
,
1160 .unique
= fuse_get_unique(fiq
),
1161 .len
= sizeof(ih
) + sizeof(arg
),
1164 if (nbytes
< ih
.len
) {
1165 spin_unlock(&fiq
->waitq
.lock
);
1169 max_forgets
= (nbytes
- ih
.len
) / sizeof(struct fuse_forget_one
);
1170 head
= dequeue_forget(fiq
, max_forgets
, &count
);
1171 spin_unlock(&fiq
->waitq
.lock
);
1174 ih
.len
+= count
* sizeof(struct fuse_forget_one
);
1175 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1177 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1180 struct fuse_forget_link
*forget
= head
;
1183 err
= fuse_copy_one(cs
, &forget
->forget_one
,
1184 sizeof(forget
->forget_one
));
1186 head
= forget
->next
;
1190 fuse_copy_finish(cs
);
1198 static int fuse_read_forget(struct fuse_conn
*fc
, struct fuse_iqueue
*fiq
,
1199 struct fuse_copy_state
*cs
,
1201 __releases(fiq
->waitq
.lock
)
1203 if (fc
->minor
< 16 || fiq
->forget_list_head
.next
->next
== NULL
)
1204 return fuse_read_single_forget(fiq
, cs
, nbytes
);
1206 return fuse_read_batch_forget(fiq
, cs
, nbytes
);
1210 * Read a single request into the userspace filesystem's buffer. This
1211 * function waits until a request is available, then removes it from
1212 * the pending list and copies request data to userspace buffer. If
1213 * no reply is needed (FORGET) or request has been aborted or there
1214 * was an error during the copying then it's finished by calling
1215 * request_end(). Otherwise add it to the processing list, and set
1218 static ssize_t
fuse_dev_do_read(struct fuse_dev
*fud
, struct file
*file
,
1219 struct fuse_copy_state
*cs
, size_t nbytes
)
1222 struct fuse_conn
*fc
= fud
->fc
;
1223 struct fuse_iqueue
*fiq
= &fc
->iq
;
1224 struct fuse_pqueue
*fpq
= &fud
->pq
;
1225 struct fuse_req
*req
;
1230 spin_lock(&fiq
->waitq
.lock
);
1232 if ((file
->f_flags
& O_NONBLOCK
) && fiq
->connected
&&
1233 !request_pending(fiq
))
1236 err
= wait_event_interruptible_exclusive_locked(fiq
->waitq
,
1237 !fiq
->connected
|| request_pending(fiq
));
1242 if (!fiq
->connected
)
1245 if (!list_empty(&fiq
->interrupts
)) {
1246 req
= list_entry(fiq
->interrupts
.next
, struct fuse_req
,
1248 return fuse_read_interrupt(fiq
, cs
, nbytes
, req
);
1251 if (forget_pending(fiq
)) {
1252 if (list_empty(&fiq
->pending
) || fiq
->forget_batch
-- > 0)
1253 return fuse_read_forget(fc
, fiq
, cs
, nbytes
);
1255 if (fiq
->forget_batch
<= -8)
1256 fiq
->forget_batch
= 16;
1259 req
= list_entry(fiq
->pending
.next
, struct fuse_req
, list
);
1260 clear_bit(FR_PENDING
, &req
->flags
);
1261 list_del_init(&req
->list
);
1262 spin_unlock(&fiq
->waitq
.lock
);
1265 reqsize
= in
->h
.len
;
1267 if (task_active_pid_ns(current
) != fc
->pid_ns
) {
1269 in
->h
.pid
= pid_vnr(find_pid_ns(in
->h
.pid
, fc
->pid_ns
));
1273 /* If request is too large, reply with an error and restart the read */
1274 if (nbytes
< reqsize
) {
1275 req
->out
.h
.error
= -EIO
;
1276 /* SETXATTR is special, since it may contain too large data */
1277 if (in
->h
.opcode
== FUSE_SETXATTR
)
1278 req
->out
.h
.error
= -E2BIG
;
1279 request_end(fc
, req
);
1282 spin_lock(&fpq
->lock
);
1283 list_add(&req
->list
, &fpq
->io
);
1284 spin_unlock(&fpq
->lock
);
1286 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1288 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1289 (struct fuse_arg
*) in
->args
, 0);
1290 fuse_copy_finish(cs
);
1291 spin_lock(&fpq
->lock
);
1292 clear_bit(FR_LOCKED
, &req
->flags
);
1293 if (!fpq
->connected
) {
1298 req
->out
.h
.error
= -EIO
;
1301 if (!test_bit(FR_ISREPLY
, &req
->flags
)) {
1305 list_move_tail(&req
->list
, &fpq
->processing
);
1306 spin_unlock(&fpq
->lock
);
1307 set_bit(FR_SENT
, &req
->flags
);
1308 /* matches barrier in request_wait_answer() */
1309 smp_mb__after_atomic();
1310 if (test_bit(FR_INTERRUPTED
, &req
->flags
))
1311 queue_interrupt(fiq
, req
);
1316 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1317 list_del_init(&req
->list
);
1318 spin_unlock(&fpq
->lock
);
1319 request_end(fc
, req
);
1323 spin_unlock(&fiq
->waitq
.lock
);
1327 static int fuse_dev_open(struct inode
*inode
, struct file
*file
)
1330 * The fuse device's file's private_data is used to hold
1331 * the fuse_conn(ection) when it is mounted, and is used to
1332 * keep track of whether the file has been mounted already.
1334 file
->private_data
= NULL
;
1338 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, struct iov_iter
*to
)
1340 struct fuse_copy_state cs
;
1341 struct file
*file
= iocb
->ki_filp
;
1342 struct fuse_dev
*fud
= fuse_get_dev(file
);
1347 if (!iter_is_iovec(to
))
1350 fuse_copy_init(&cs
, 1, to
);
1352 return fuse_dev_do_read(fud
, file
, &cs
, iov_iter_count(to
));
1355 static ssize_t
fuse_dev_splice_read(struct file
*in
, loff_t
*ppos
,
1356 struct pipe_inode_info
*pipe
,
1357 size_t len
, unsigned int flags
)
1361 struct pipe_buffer
*bufs
;
1362 struct fuse_copy_state cs
;
1363 struct fuse_dev
*fud
= fuse_get_dev(in
);
1368 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1372 fuse_copy_init(&cs
, 1, NULL
);
1375 ret
= fuse_dev_do_read(fud
, in
, &cs
, len
);
1379 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1384 for (ret
= total
= 0; page_nr
< cs
.nr_segs
; total
+= ret
) {
1386 * Need to be careful about this. Having buf->ops in module
1387 * code can Oops if the buffer persists after module unload.
1389 bufs
[page_nr
].ops
= &nosteal_pipe_buf_ops
;
1390 bufs
[page_nr
].flags
= 0;
1391 ret
= add_to_pipe(pipe
, &bufs
[page_nr
++]);
1392 if (unlikely(ret
< 0))
1398 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1399 put_page(bufs
[page_nr
].page
);
1405 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1406 struct fuse_copy_state
*cs
)
1408 struct fuse_notify_poll_wakeup_out outarg
;
1411 if (size
!= sizeof(outarg
))
1414 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1418 fuse_copy_finish(cs
);
1419 return fuse_notify_poll_wakeup(fc
, &outarg
);
1422 fuse_copy_finish(cs
);
1426 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1427 struct fuse_copy_state
*cs
)
1429 struct fuse_notify_inval_inode_out outarg
;
1432 if (size
!= sizeof(outarg
))
1435 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1438 fuse_copy_finish(cs
);
1440 down_read(&fc
->killsb
);
1443 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1444 outarg
.off
, outarg
.len
);
1446 up_read(&fc
->killsb
);
1450 fuse_copy_finish(cs
);
1454 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1455 struct fuse_copy_state
*cs
)
1457 struct fuse_notify_inval_entry_out outarg
;
1462 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1467 if (size
< sizeof(outarg
))
1470 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1474 err
= -ENAMETOOLONG
;
1475 if (outarg
.namelen
> FUSE_NAME_MAX
)
1479 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1483 name
.len
= outarg
.namelen
;
1484 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1487 fuse_copy_finish(cs
);
1488 buf
[outarg
.namelen
] = 0;
1490 down_read(&fc
->killsb
);
1493 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, 0, &name
);
1494 up_read(&fc
->killsb
);
1500 fuse_copy_finish(cs
);
1504 static int fuse_notify_delete(struct fuse_conn
*fc
, unsigned int size
,
1505 struct fuse_copy_state
*cs
)
1507 struct fuse_notify_delete_out outarg
;
1512 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1517 if (size
< sizeof(outarg
))
1520 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1524 err
= -ENAMETOOLONG
;
1525 if (outarg
.namelen
> FUSE_NAME_MAX
)
1529 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1533 name
.len
= outarg
.namelen
;
1534 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1537 fuse_copy_finish(cs
);
1538 buf
[outarg
.namelen
] = 0;
1540 down_read(&fc
->killsb
);
1543 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
,
1544 outarg
.child
, &name
);
1545 up_read(&fc
->killsb
);
1551 fuse_copy_finish(cs
);
1555 static int fuse_notify_store(struct fuse_conn
*fc
, unsigned int size
,
1556 struct fuse_copy_state
*cs
)
1558 struct fuse_notify_store_out outarg
;
1559 struct inode
*inode
;
1560 struct address_space
*mapping
;
1564 unsigned int offset
;
1570 if (size
< sizeof(outarg
))
1573 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1578 if (size
- sizeof(outarg
) != outarg
.size
)
1581 nodeid
= outarg
.nodeid
;
1583 down_read(&fc
->killsb
);
1589 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1593 mapping
= inode
->i_mapping
;
1594 index
= outarg
.offset
>> PAGE_SHIFT
;
1595 offset
= outarg
.offset
& ~PAGE_MASK
;
1596 file_size
= i_size_read(inode
);
1597 end
= outarg
.offset
+ outarg
.size
;
1598 if (end
> file_size
) {
1600 fuse_write_update_size(inode
, file_size
);
1606 unsigned int this_num
;
1609 page
= find_or_create_page(mapping
, index
,
1610 mapping_gfp_mask(mapping
));
1614 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1615 err
= fuse_copy_page(cs
, &page
, offset
, this_num
, 0);
1616 if (!err
&& offset
== 0 &&
1617 (this_num
== PAGE_SIZE
|| file_size
== end
))
1618 SetPageUptodate(page
);
1635 up_read(&fc
->killsb
);
1637 fuse_copy_finish(cs
);
1641 static void fuse_retrieve_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1643 release_pages(req
->pages
, req
->num_pages
, false);
1646 static int fuse_retrieve(struct fuse_conn
*fc
, struct inode
*inode
,
1647 struct fuse_notify_retrieve_out
*outarg
)
1650 struct address_space
*mapping
= inode
->i_mapping
;
1651 struct fuse_req
*req
;
1655 unsigned int offset
;
1656 size_t total_len
= 0;
1659 offset
= outarg
->offset
& ~PAGE_MASK
;
1660 file_size
= i_size_read(inode
);
1663 if (outarg
->offset
> file_size
)
1665 else if (outarg
->offset
+ num
> file_size
)
1666 num
= file_size
- outarg
->offset
;
1668 num_pages
= (num
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1669 num_pages
= min(num_pages
, FUSE_MAX_PAGES_PER_REQ
);
1671 req
= fuse_get_req(fc
, num_pages
);
1673 return PTR_ERR(req
);
1675 req
->in
.h
.opcode
= FUSE_NOTIFY_REPLY
;
1676 req
->in
.h
.nodeid
= outarg
->nodeid
;
1677 req
->in
.numargs
= 2;
1678 req
->in
.argpages
= 1;
1679 req
->page_descs
[0].offset
= offset
;
1680 req
->end
= fuse_retrieve_end
;
1682 index
= outarg
->offset
>> PAGE_SHIFT
;
1684 while (num
&& req
->num_pages
< num_pages
) {
1686 unsigned int this_num
;
1688 page
= find_get_page(mapping
, index
);
1692 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1693 req
->pages
[req
->num_pages
] = page
;
1694 req
->page_descs
[req
->num_pages
].length
= this_num
;
1699 total_len
+= this_num
;
1702 req
->misc
.retrieve_in
.offset
= outarg
->offset
;
1703 req
->misc
.retrieve_in
.size
= total_len
;
1704 req
->in
.args
[0].size
= sizeof(req
->misc
.retrieve_in
);
1705 req
->in
.args
[0].value
= &req
->misc
.retrieve_in
;
1706 req
->in
.args
[1].size
= total_len
;
1708 err
= fuse_request_send_notify_reply(fc
, req
, outarg
->notify_unique
);
1710 fuse_retrieve_end(fc
, req
);
1715 static int fuse_notify_retrieve(struct fuse_conn
*fc
, unsigned int size
,
1716 struct fuse_copy_state
*cs
)
1718 struct fuse_notify_retrieve_out outarg
;
1719 struct inode
*inode
;
1723 if (size
!= sizeof(outarg
))
1726 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1730 fuse_copy_finish(cs
);
1732 down_read(&fc
->killsb
);
1735 u64 nodeid
= outarg
.nodeid
;
1737 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1739 err
= fuse_retrieve(fc
, inode
, &outarg
);
1743 up_read(&fc
->killsb
);
1748 fuse_copy_finish(cs
);
1752 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1753 unsigned int size
, struct fuse_copy_state
*cs
)
1755 /* Don't try to move pages (yet) */
1759 case FUSE_NOTIFY_POLL
:
1760 return fuse_notify_poll(fc
, size
, cs
);
1762 case FUSE_NOTIFY_INVAL_INODE
:
1763 return fuse_notify_inval_inode(fc
, size
, cs
);
1765 case FUSE_NOTIFY_INVAL_ENTRY
:
1766 return fuse_notify_inval_entry(fc
, size
, cs
);
1768 case FUSE_NOTIFY_STORE
:
1769 return fuse_notify_store(fc
, size
, cs
);
1771 case FUSE_NOTIFY_RETRIEVE
:
1772 return fuse_notify_retrieve(fc
, size
, cs
);
1774 case FUSE_NOTIFY_DELETE
:
1775 return fuse_notify_delete(fc
, size
, cs
);
1778 fuse_copy_finish(cs
);
1783 /* Look up request on processing list by unique ID */
1784 static struct fuse_req
*request_find(struct fuse_pqueue
*fpq
, u64 unique
)
1786 struct fuse_req
*req
;
1788 list_for_each_entry(req
, &fpq
->processing
, list
) {
1789 if (req
->in
.h
.unique
== unique
|| req
->intr_unique
== unique
)
1795 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1798 unsigned reqsize
= sizeof(struct fuse_out_header
);
1801 return nbytes
!= reqsize
? -EINVAL
: 0;
1803 reqsize
+= len_args(out
->numargs
, out
->args
);
1805 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1807 else if (reqsize
> nbytes
) {
1808 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1809 unsigned diffsize
= reqsize
- nbytes
;
1810 if (diffsize
> lastarg
->size
)
1812 lastarg
->size
-= diffsize
;
1814 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1819 * Write a single reply to a request. First the header is copied from
1820 * the write buffer. The request is then searched on the processing
1821 * list by the unique ID found in the header. If found, then remove
1822 * it from the list and copy the rest of the buffer to the request.
1823 * The request is finished by calling request_end()
1825 static ssize_t
fuse_dev_do_write(struct fuse_dev
*fud
,
1826 struct fuse_copy_state
*cs
, size_t nbytes
)
1829 struct fuse_conn
*fc
= fud
->fc
;
1830 struct fuse_pqueue
*fpq
= &fud
->pq
;
1831 struct fuse_req
*req
;
1832 struct fuse_out_header oh
;
1834 if (nbytes
< sizeof(struct fuse_out_header
))
1837 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1842 if (oh
.len
!= nbytes
)
1846 * Zero oh.unique indicates unsolicited notification message
1847 * and error contains notification code.
1850 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1851 return err
? err
: nbytes
;
1855 if (oh
.error
<= -1000 || oh
.error
> 0)
1858 spin_lock(&fpq
->lock
);
1860 if (!fpq
->connected
)
1863 req
= request_find(fpq
, oh
.unique
);
1867 /* Is it an interrupt reply? */
1868 if (req
->intr_unique
== oh
.unique
) {
1869 spin_unlock(&fpq
->lock
);
1872 if (nbytes
!= sizeof(struct fuse_out_header
))
1875 if (oh
.error
== -ENOSYS
)
1876 fc
->no_interrupt
= 1;
1877 else if (oh
.error
== -EAGAIN
)
1878 queue_interrupt(&fc
->iq
, req
);
1880 fuse_copy_finish(cs
);
1884 clear_bit(FR_SENT
, &req
->flags
);
1885 list_move(&req
->list
, &fpq
->io
);
1887 set_bit(FR_LOCKED
, &req
->flags
);
1888 spin_unlock(&fpq
->lock
);
1890 if (!req
->out
.page_replace
)
1893 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1894 if (req
->in
.h
.opcode
== FUSE_CANONICAL_PATH
) {
1895 req
->out
.h
.error
= kern_path((char *)req
->out
.args
[0].value
, 0,
1896 req
->canonical_path
);
1898 fuse_copy_finish(cs
);
1900 spin_lock(&fpq
->lock
);
1901 clear_bit(FR_LOCKED
, &req
->flags
);
1902 if (!fpq
->connected
)
1905 req
->out
.h
.error
= -EIO
;
1906 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1907 list_del_init(&req
->list
);
1908 spin_unlock(&fpq
->lock
);
1910 request_end(fc
, req
);
1912 return err
? err
: nbytes
;
1915 spin_unlock(&fpq
->lock
);
1917 fuse_copy_finish(cs
);
1921 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, struct iov_iter
*from
)
1923 struct fuse_copy_state cs
;
1924 struct fuse_dev
*fud
= fuse_get_dev(iocb
->ki_filp
);
1929 if (!iter_is_iovec(from
))
1932 fuse_copy_init(&cs
, 0, from
);
1934 return fuse_dev_do_write(fud
, &cs
, iov_iter_count(from
));
1937 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1938 struct file
*out
, loff_t
*ppos
,
1939 size_t len
, unsigned int flags
)
1943 struct pipe_buffer
*bufs
;
1944 struct fuse_copy_state cs
;
1945 struct fuse_dev
*fud
;
1949 fud
= fuse_get_dev(out
);
1953 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1960 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
1961 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
1971 struct pipe_buffer
*ibuf
;
1972 struct pipe_buffer
*obuf
;
1974 BUG_ON(nbuf
>= pipe
->buffers
);
1975 BUG_ON(!pipe
->nrbufs
);
1976 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
1979 if (rem
>= ibuf
->len
) {
1982 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
1985 pipe_buf_get(pipe
, ibuf
);
1987 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1989 ibuf
->offset
+= obuf
->len
;
1990 ibuf
->len
-= obuf
->len
;
1997 fuse_copy_init(&cs
, 0, NULL
);
2002 if (flags
& SPLICE_F_MOVE
)
2005 ret
= fuse_dev_do_write(fud
, &cs
, len
);
2007 for (idx
= 0; idx
< nbuf
; idx
++)
2008 pipe_buf_release(pipe
, &bufs
[idx
]);
2015 static unsigned fuse_dev_poll(struct file
*file
, poll_table
*wait
)
2017 unsigned mask
= POLLOUT
| POLLWRNORM
;
2018 struct fuse_iqueue
*fiq
;
2019 struct fuse_dev
*fud
= fuse_get_dev(file
);
2025 poll_wait(file
, &fiq
->waitq
, wait
);
2027 spin_lock(&fiq
->waitq
.lock
);
2028 if (!fiq
->connected
)
2030 else if (request_pending(fiq
))
2031 mask
|= POLLIN
| POLLRDNORM
;
2032 spin_unlock(&fiq
->waitq
.lock
);
2038 * Abort all requests on the given list (pending or processing)
2040 * This function releases and reacquires fc->lock
2042 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
2044 while (!list_empty(head
)) {
2045 struct fuse_req
*req
;
2046 req
= list_entry(head
->next
, struct fuse_req
, list
);
2047 req
->out
.h
.error
= -ECONNABORTED
;
2048 clear_bit(FR_SENT
, &req
->flags
);
2049 list_del_init(&req
->list
);
2050 request_end(fc
, req
);
2054 static void end_polls(struct fuse_conn
*fc
)
2058 p
= rb_first(&fc
->polled_files
);
2061 struct fuse_file
*ff
;
2062 ff
= rb_entry(p
, struct fuse_file
, polled_node
);
2063 wake_up_interruptible_all(&ff
->poll_wait
);
2070 * Abort all requests.
2072 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2075 * The same effect is usually achievable through killing the filesystem daemon
2076 * and all users of the filesystem. The exception is the combination of an
2077 * asynchronous request and the tricky deadlock (see
2078 * Documentation/filesystems/fuse.txt).
2080 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2081 * requests, they should be finished off immediately. Locked requests will be
2082 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2083 * requests. It is possible that some request will finish before we can. This
2084 * is OK, the request will in that case be removed from the list before we touch
2087 void fuse_abort_conn(struct fuse_conn
*fc
)
2089 struct fuse_iqueue
*fiq
= &fc
->iq
;
2091 spin_lock(&fc
->lock
);
2092 if (fc
->connected
) {
2093 struct fuse_dev
*fud
;
2094 struct fuse_req
*req
, *next
;
2100 fuse_set_initialized(fc
);
2101 list_for_each_entry(fud
, &fc
->devices
, entry
) {
2102 struct fuse_pqueue
*fpq
= &fud
->pq
;
2104 spin_lock(&fpq
->lock
);
2106 list_for_each_entry_safe(req
, next
, &fpq
->io
, list
) {
2107 req
->out
.h
.error
= -ECONNABORTED
;
2108 spin_lock(&req
->waitq
.lock
);
2109 set_bit(FR_ABORTED
, &req
->flags
);
2110 if (!test_bit(FR_LOCKED
, &req
->flags
)) {
2111 set_bit(FR_PRIVATE
, &req
->flags
);
2112 list_move(&req
->list
, &to_end1
);
2114 spin_unlock(&req
->waitq
.lock
);
2116 list_splice_init(&fpq
->processing
, &to_end2
);
2117 spin_unlock(&fpq
->lock
);
2119 fc
->max_background
= UINT_MAX
;
2122 spin_lock(&fiq
->waitq
.lock
);
2124 list_splice_init(&fiq
->pending
, &to_end2
);
2125 list_for_each_entry(req
, &to_end2
, list
)
2126 clear_bit(FR_PENDING
, &req
->flags
);
2127 while (forget_pending(fiq
))
2128 kfree(dequeue_forget(fiq
, 1, NULL
));
2129 wake_up_all_locked(&fiq
->waitq
);
2130 spin_unlock(&fiq
->waitq
.lock
);
2131 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
2133 wake_up_all(&fc
->blocked_waitq
);
2134 spin_unlock(&fc
->lock
);
2136 while (!list_empty(&to_end1
)) {
2137 req
= list_first_entry(&to_end1
, struct fuse_req
, list
);
2138 __fuse_get_request(req
);
2139 list_del_init(&req
->list
);
2140 request_end(fc
, req
);
2142 end_requests(fc
, &to_end2
);
2144 spin_unlock(&fc
->lock
);
2147 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
2149 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
2151 struct fuse_dev
*fud
= fuse_get_dev(file
);
2154 struct fuse_conn
*fc
= fud
->fc
;
2155 struct fuse_pqueue
*fpq
= &fud
->pq
;
2157 WARN_ON(!list_empty(&fpq
->io
));
2158 end_requests(fc
, &fpq
->processing
);
2159 /* Are we the last open device? */
2160 if (atomic_dec_and_test(&fc
->dev_count
)) {
2161 WARN_ON(fc
->iq
.fasync
!= NULL
);
2162 fuse_abort_conn(fc
);
2168 EXPORT_SYMBOL_GPL(fuse_dev_release
);
2170 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
2172 struct fuse_dev
*fud
= fuse_get_dev(file
);
2177 /* No locking - fasync_helper does its own locking */
2178 return fasync_helper(fd
, file
, on
, &fud
->fc
->iq
.fasync
);
2181 static int fuse_device_clone(struct fuse_conn
*fc
, struct file
*new)
2183 struct fuse_dev
*fud
;
2185 if (new->private_data
)
2188 fud
= fuse_dev_alloc(fc
);
2192 new->private_data
= fud
;
2193 atomic_inc(&fc
->dev_count
);
2198 static long fuse_dev_ioctl(struct file
*file
, unsigned int cmd
,
2203 if (cmd
== FUSE_DEV_IOC_CLONE
) {
2207 if (!get_user(oldfd
, (__u32 __user
*) arg
)) {
2208 struct file
*old
= fget(oldfd
);
2212 struct fuse_dev
*fud
= NULL
;
2215 * Check against file->f_op because CUSE
2216 * uses the same ioctl handler.
2218 if (old
->f_op
== file
->f_op
&&
2219 old
->f_cred
->user_ns
== file
->f_cred
->user_ns
)
2220 fud
= fuse_get_dev(old
);
2223 mutex_lock(&fuse_mutex
);
2224 err
= fuse_device_clone(fud
->fc
, file
);
2225 mutex_unlock(&fuse_mutex
);
2234 const struct file_operations fuse_dev_operations
= {
2235 .owner
= THIS_MODULE
,
2236 .open
= fuse_dev_open
,
2237 .llseek
= no_llseek
,
2238 .read_iter
= fuse_dev_read
,
2239 .splice_read
= fuse_dev_splice_read
,
2240 .write_iter
= fuse_dev_write
,
2241 .splice_write
= fuse_dev_splice_write
,
2242 .poll
= fuse_dev_poll
,
2243 .release
= fuse_dev_release
,
2244 .fasync
= fuse_dev_fasync
,
2245 .unlocked_ioctl
= fuse_dev_ioctl
,
2246 .compat_ioctl
= fuse_dev_ioctl
,
2248 EXPORT_SYMBOL_GPL(fuse_dev_operations
);
2250 static struct miscdevice fuse_miscdevice
= {
2251 .minor
= FUSE_MINOR
,
2253 .fops
= &fuse_dev_operations
,
2256 int __init
fuse_dev_init(void)
2259 fuse_req_cachep
= kmem_cache_create("fuse_request",
2260 sizeof(struct fuse_req
),
2262 if (!fuse_req_cachep
)
2265 err
= misc_register(&fuse_miscdevice
);
2267 goto out_cache_clean
;
2272 kmem_cache_destroy(fuse_req_cachep
);
2277 void fuse_dev_cleanup(void)
2279 misc_deregister(&fuse_miscdevice
);
2280 kmem_cache_destroy(fuse_req_cachep
);