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 static struct kmem_cache
*fuse_req_cachep
;
30 static struct fuse_dev
*fuse_get_dev(struct file
*file
)
33 * Lockless access is OK, because file->private data is set
34 * once during mount and is valid until the file is released.
36 return READ_ONCE(file
->private_data
);
39 static void fuse_request_init(struct fuse_req
*req
, struct page
**pages
,
40 struct fuse_page_desc
*page_descs
,
43 memset(req
, 0, sizeof(*req
));
44 memset(pages
, 0, sizeof(*pages
) * npages
);
45 memset(page_descs
, 0, sizeof(*page_descs
) * npages
);
46 INIT_LIST_HEAD(&req
->list
);
47 INIT_LIST_HEAD(&req
->intr_entry
);
48 init_waitqueue_head(&req
->waitq
);
49 refcount_set(&req
->count
, 1);
51 req
->page_descs
= page_descs
;
52 req
->max_pages
= npages
;
53 __set_bit(FR_PENDING
, &req
->flags
);
56 static struct fuse_req
*__fuse_request_alloc(unsigned npages
, gfp_t flags
)
58 struct fuse_req
*req
= kmem_cache_alloc(fuse_req_cachep
, flags
);
61 struct fuse_page_desc
*page_descs
;
63 if (npages
<= FUSE_REQ_INLINE_PAGES
) {
64 pages
= req
->inline_pages
;
65 page_descs
= req
->inline_page_descs
;
67 pages
= kmalloc(sizeof(struct page
*) * npages
, flags
);
68 page_descs
= kmalloc(sizeof(struct fuse_page_desc
) *
72 if (!pages
|| !page_descs
) {
75 kmem_cache_free(fuse_req_cachep
, req
);
79 fuse_request_init(req
, pages
, page_descs
, npages
);
84 struct fuse_req
*fuse_request_alloc(unsigned npages
)
86 return __fuse_request_alloc(npages
, GFP_KERNEL
);
88 EXPORT_SYMBOL_GPL(fuse_request_alloc
);
90 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
)
92 return __fuse_request_alloc(npages
, GFP_NOFS
);
95 void fuse_request_free(struct fuse_req
*req
)
97 if (req
->pages
!= req
->inline_pages
) {
99 kfree(req
->page_descs
);
101 kmem_cache_free(fuse_req_cachep
, req
);
104 void __fuse_get_request(struct fuse_req
*req
)
106 refcount_inc(&req
->count
);
109 /* Must be called with > 1 refcount */
110 static void __fuse_put_request(struct fuse_req
*req
)
112 refcount_dec(&req
->count
);
115 static void fuse_req_init_context(struct fuse_conn
*fc
, struct fuse_req
*req
)
117 req
->in
.h
.uid
= from_kuid(fc
->user_ns
, current_fsuid());
118 req
->in
.h
.gid
= from_kgid(fc
->user_ns
, current_fsgid());
119 req
->in
.h
.pid
= pid_nr_ns(task_pid(current
), fc
->pid_ns
);
122 void fuse_set_initialized(struct fuse_conn
*fc
)
124 /* Make sure stores before this are seen on another CPU */
129 static bool fuse_block_alloc(struct fuse_conn
*fc
, bool for_background
)
131 return !fc
->initialized
|| (for_background
&& fc
->blocked
);
134 static void fuse_drop_waiting(struct fuse_conn
*fc
)
137 atomic_dec(&fc
->num_waiting
);
138 } else if (atomic_dec_and_test(&fc
->num_waiting
)) {
139 /* wake up aborters */
140 wake_up_all(&fc
->blocked_waitq
);
144 static struct fuse_req
*__fuse_get_req(struct fuse_conn
*fc
, unsigned npages
,
147 struct fuse_req
*req
;
149 atomic_inc(&fc
->num_waiting
);
151 if (fuse_block_alloc(fc
, for_background
)) {
153 if (wait_event_killable_exclusive(fc
->blocked_waitq
,
154 !fuse_block_alloc(fc
, for_background
)))
157 /* Matches smp_wmb() in fuse_set_initialized() */
168 req
= fuse_request_alloc(npages
);
172 wake_up(&fc
->blocked_waitq
);
176 fuse_req_init_context(fc
, req
);
177 __set_bit(FR_WAITING
, &req
->flags
);
179 __set_bit(FR_BACKGROUND
, &req
->flags
);
180 if (req
->in
.h
.uid
== (uid_t
)-1 || req
->in
.h
.gid
== (gid_t
)-1) {
181 fuse_put_request(fc
, req
);
182 return ERR_PTR(-EOVERFLOW
);
188 fuse_drop_waiting(fc
);
192 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
)
194 return __fuse_get_req(fc
, npages
, false);
196 EXPORT_SYMBOL_GPL(fuse_get_req
);
198 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
201 return __fuse_get_req(fc
, npages
, true);
203 EXPORT_SYMBOL_GPL(fuse_get_req_for_background
);
206 * Return request in fuse_file->reserved_req. However that may
207 * currently be in use. If that is the case, wait for it to become
210 static struct fuse_req
*get_reserved_req(struct fuse_conn
*fc
,
213 struct fuse_req
*req
= NULL
;
214 struct fuse_file
*ff
= file
->private_data
;
217 wait_event(fc
->reserved_req_waitq
, ff
->reserved_req
);
218 spin_lock(&fc
->lock
);
219 if (ff
->reserved_req
) {
220 req
= ff
->reserved_req
;
221 ff
->reserved_req
= NULL
;
222 req
->stolen_file
= get_file(file
);
224 spin_unlock(&fc
->lock
);
231 * Put stolen request back into fuse_file->reserved_req
233 static void put_reserved_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
235 struct file
*file
= req
->stolen_file
;
236 struct fuse_file
*ff
= file
->private_data
;
238 spin_lock(&fc
->lock
);
239 fuse_request_init(req
, req
->pages
, req
->page_descs
, req
->max_pages
);
240 BUG_ON(ff
->reserved_req
);
241 ff
->reserved_req
= req
;
242 wake_up_all(&fc
->reserved_req_waitq
);
243 spin_unlock(&fc
->lock
);
248 * Gets a requests for a file operation, always succeeds
250 * This is used for sending the FLUSH request, which must get to
251 * userspace, due to POSIX locks which may need to be unlocked.
253 * If allocation fails due to OOM, use the reserved request in
256 * This is very unlikely to deadlock accidentally, since the
257 * filesystem should not have it's own file open. If deadlock is
258 * intentional, it can still be broken by "aborting" the filesystem.
260 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
263 struct fuse_req
*req
;
265 atomic_inc(&fc
->num_waiting
);
266 wait_event(fc
->blocked_waitq
, fc
->initialized
);
267 /* Matches smp_wmb() in fuse_set_initialized() */
269 req
= fuse_request_alloc(0);
271 req
= get_reserved_req(fc
, file
);
273 fuse_req_init_context(fc
, req
);
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
->lock
);
289 wake_up(&fc
->blocked_waitq
);
290 spin_unlock(&fc
->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 return ++fiq
->reqctr
;
322 static void queue_request(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
324 req
->in
.h
.len
= sizeof(struct fuse_in_header
) +
325 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
326 list_add_tail(&req
->list
, &fiq
->pending
);
327 wake_up_locked(&fiq
->waitq
);
328 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
331 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
332 u64 nodeid
, u64 nlookup
)
334 struct fuse_iqueue
*fiq
= &fc
->iq
;
336 forget
->forget_one
.nodeid
= nodeid
;
337 forget
->forget_one
.nlookup
= nlookup
;
339 spin_lock(&fiq
->waitq
.lock
);
340 if (fiq
->connected
) {
341 fiq
->forget_list_tail
->next
= forget
;
342 fiq
->forget_list_tail
= forget
;
343 wake_up_locked(&fiq
->waitq
);
344 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
348 spin_unlock(&fiq
->waitq
.lock
);
351 static void flush_bg_queue(struct fuse_conn
*fc
)
353 while (fc
->active_background
< fc
->max_background
&&
354 !list_empty(&fc
->bg_queue
)) {
355 struct fuse_req
*req
;
356 struct fuse_iqueue
*fiq
= &fc
->iq
;
358 req
= list_entry(fc
->bg_queue
.next
, struct fuse_req
, list
);
359 list_del(&req
->list
);
360 fc
->active_background
++;
361 spin_lock(&fiq
->waitq
.lock
);
362 req
->in
.h
.unique
= fuse_get_unique(fiq
);
363 queue_request(fiq
, req
);
364 spin_unlock(&fiq
->waitq
.lock
);
369 * This function is called when a request is finished. Either a reply
370 * has arrived or it was aborted (and not yet sent) or some error
371 * occurred during communication with userspace, or the device file
372 * was closed. The requester thread is woken up (if still waiting),
373 * the 'end' callback is called if given, else the reference to the
374 * request is released
376 static void request_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
378 struct fuse_iqueue
*fiq
= &fc
->iq
;
380 if (test_and_set_bit(FR_FINISHED
, &req
->flags
))
383 spin_lock(&fiq
->waitq
.lock
);
384 list_del_init(&req
->intr_entry
);
385 spin_unlock(&fiq
->waitq
.lock
);
386 WARN_ON(test_bit(FR_PENDING
, &req
->flags
));
387 WARN_ON(test_bit(FR_SENT
, &req
->flags
));
388 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
389 spin_lock(&fc
->lock
);
390 clear_bit(FR_BACKGROUND
, &req
->flags
);
391 if (fc
->num_background
== fc
->max_background
)
394 /* Wake up next waiter, if any */
395 if (!fc
->blocked
&& waitqueue_active(&fc
->blocked_waitq
))
396 wake_up(&fc
->blocked_waitq
);
398 if (fc
->num_background
== fc
->congestion_threshold
&& fc
->sb
) {
399 clear_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_SYNC
);
400 clear_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_ASYNC
);
402 fc
->num_background
--;
403 fc
->active_background
--;
405 spin_unlock(&fc
->lock
);
407 wake_up(&req
->waitq
);
411 fuse_put_request(fc
, req
);
414 static void queue_interrupt(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
416 spin_lock(&fiq
->waitq
.lock
);
417 if (test_bit(FR_FINISHED
, &req
->flags
)) {
418 spin_unlock(&fiq
->waitq
.lock
);
421 if (list_empty(&req
->intr_entry
)) {
422 list_add_tail(&req
->intr_entry
, &fiq
->interrupts
);
423 wake_up_locked(&fiq
->waitq
);
425 spin_unlock(&fiq
->waitq
.lock
);
426 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
429 static void request_wait_answer(struct fuse_conn
*fc
, struct fuse_req
*req
)
431 struct fuse_iqueue
*fiq
= &fc
->iq
;
434 if (!fc
->no_interrupt
) {
435 /* Any signal may interrupt this */
436 err
= wait_event_interruptible(req
->waitq
,
437 test_bit(FR_FINISHED
, &req
->flags
));
441 set_bit(FR_INTERRUPTED
, &req
->flags
);
442 /* matches barrier in fuse_dev_do_read() */
443 smp_mb__after_atomic();
444 if (test_bit(FR_SENT
, &req
->flags
))
445 queue_interrupt(fiq
, req
);
448 if (!test_bit(FR_FORCE
, &req
->flags
)) {
449 /* Only fatal signals may interrupt this */
450 err
= wait_event_killable(req
->waitq
,
451 test_bit(FR_FINISHED
, &req
->flags
));
455 spin_lock(&fiq
->waitq
.lock
);
456 /* Request is not yet in userspace, bail out */
457 if (test_bit(FR_PENDING
, &req
->flags
)) {
458 list_del(&req
->list
);
459 spin_unlock(&fiq
->waitq
.lock
);
460 __fuse_put_request(req
);
461 req
->out
.h
.error
= -EINTR
;
464 spin_unlock(&fiq
->waitq
.lock
);
468 * Either request is already in userspace, or it was forced.
471 wait_event(req
->waitq
, test_bit(FR_FINISHED
, &req
->flags
));
474 static void __fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
476 struct fuse_iqueue
*fiq
= &fc
->iq
;
478 BUG_ON(test_bit(FR_BACKGROUND
, &req
->flags
));
479 spin_lock(&fiq
->waitq
.lock
);
480 if (!fiq
->connected
) {
481 spin_unlock(&fiq
->waitq
.lock
);
482 req
->out
.h
.error
= -ENOTCONN
;
484 req
->in
.h
.unique
= fuse_get_unique(fiq
);
485 queue_request(fiq
, req
);
486 /* acquire extra reference, since request is still needed
487 after request_end() */
488 __fuse_get_request(req
);
489 spin_unlock(&fiq
->waitq
.lock
);
491 request_wait_answer(fc
, req
);
492 /* Pairs with smp_wmb() in request_end() */
497 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
499 __set_bit(FR_ISREPLY
, &req
->flags
);
500 if (!test_bit(FR_WAITING
, &req
->flags
)) {
501 __set_bit(FR_WAITING
, &req
->flags
);
502 atomic_inc(&fc
->num_waiting
);
504 __fuse_request_send(fc
, req
);
506 EXPORT_SYMBOL_GPL(fuse_request_send
);
508 static void fuse_adjust_compat(struct fuse_conn
*fc
, struct fuse_args
*args
)
510 if (fc
->minor
< 4 && args
->in
.h
.opcode
== FUSE_STATFS
)
511 args
->out
.args
[0].size
= FUSE_COMPAT_STATFS_SIZE
;
514 switch (args
->in
.h
.opcode
) {
521 args
->out
.args
[0].size
= FUSE_COMPAT_ENTRY_OUT_SIZE
;
525 args
->out
.args
[0].size
= FUSE_COMPAT_ATTR_OUT_SIZE
;
529 if (fc
->minor
< 12) {
530 switch (args
->in
.h
.opcode
) {
532 args
->in
.args
[0].size
= sizeof(struct fuse_open_in
);
535 args
->in
.args
[0].size
= FUSE_COMPAT_MKNOD_IN_SIZE
;
541 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
)
543 struct fuse_req
*req
;
546 req
= fuse_get_req(fc
, 0);
550 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
551 fuse_adjust_compat(fc
, args
);
553 req
->in
.h
.opcode
= args
->in
.h
.opcode
;
554 req
->in
.h
.nodeid
= args
->in
.h
.nodeid
;
555 req
->in
.numargs
= args
->in
.numargs
;
556 memcpy(req
->in
.args
, args
->in
.args
,
557 args
->in
.numargs
* sizeof(struct fuse_in_arg
));
558 req
->out
.argvar
= args
->out
.argvar
;
559 req
->out
.numargs
= args
->out
.numargs
;
560 memcpy(req
->out
.args
, args
->out
.args
,
561 args
->out
.numargs
* sizeof(struct fuse_arg
));
562 fuse_request_send(fc
, req
);
563 ret
= req
->out
.h
.error
;
564 if (!ret
&& args
->out
.argvar
) {
565 BUG_ON(args
->out
.numargs
!= 1);
566 ret
= req
->out
.args
[0].size
;
568 fuse_put_request(fc
, req
);
574 * Called under fc->lock
576 * fc->connected must have been checked previously
578 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
579 struct fuse_req
*req
)
581 BUG_ON(!test_bit(FR_BACKGROUND
, &req
->flags
));
582 if (!test_bit(FR_WAITING
, &req
->flags
)) {
583 __set_bit(FR_WAITING
, &req
->flags
);
584 atomic_inc(&fc
->num_waiting
);
586 __set_bit(FR_ISREPLY
, &req
->flags
);
587 fc
->num_background
++;
588 if (fc
->num_background
== fc
->max_background
)
590 if (fc
->num_background
== fc
->congestion_threshold
&& fc
->sb
) {
591 set_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_SYNC
);
592 set_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_ASYNC
);
594 list_add_tail(&req
->list
, &fc
->bg_queue
);
598 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
601 spin_lock(&fc
->lock
);
603 fuse_request_send_background_locked(fc
, req
);
604 spin_unlock(&fc
->lock
);
606 spin_unlock(&fc
->lock
);
607 req
->out
.h
.error
= -ENOTCONN
;
609 fuse_put_request(fc
, req
);
612 EXPORT_SYMBOL_GPL(fuse_request_send_background
);
614 static int fuse_request_send_notify_reply(struct fuse_conn
*fc
,
615 struct fuse_req
*req
, u64 unique
)
618 struct fuse_iqueue
*fiq
= &fc
->iq
;
620 __clear_bit(FR_ISREPLY
, &req
->flags
);
621 req
->in
.h
.unique
= unique
;
622 spin_lock(&fiq
->waitq
.lock
);
623 if (fiq
->connected
) {
624 queue_request(fiq
, req
);
627 spin_unlock(&fiq
->waitq
.lock
);
632 void fuse_force_forget(struct file
*file
, u64 nodeid
)
634 struct inode
*inode
= file_inode(file
);
635 struct fuse_conn
*fc
= get_fuse_conn(inode
);
636 struct fuse_req
*req
;
637 struct fuse_forget_in inarg
;
639 memset(&inarg
, 0, sizeof(inarg
));
641 req
= fuse_get_req_nofail_nopages(fc
, file
);
642 req
->in
.h
.opcode
= FUSE_FORGET
;
643 req
->in
.h
.nodeid
= nodeid
;
645 req
->in
.args
[0].size
= sizeof(inarg
);
646 req
->in
.args
[0].value
= &inarg
;
647 __clear_bit(FR_ISREPLY
, &req
->flags
);
648 __fuse_request_send(fc
, req
);
650 fuse_put_request(fc
, req
);
654 * Lock the request. Up to the next unlock_request() there mustn't be
655 * anything that could cause a page-fault. If the request was already
658 static int lock_request(struct fuse_req
*req
)
662 spin_lock(&req
->waitq
.lock
);
663 if (test_bit(FR_ABORTED
, &req
->flags
))
666 set_bit(FR_LOCKED
, &req
->flags
);
667 spin_unlock(&req
->waitq
.lock
);
673 * Unlock request. If it was aborted while locked, caller is responsible
674 * for unlocking and ending the request.
676 static int unlock_request(struct fuse_req
*req
)
680 spin_lock(&req
->waitq
.lock
);
681 if (test_bit(FR_ABORTED
, &req
->flags
))
684 clear_bit(FR_LOCKED
, &req
->flags
);
685 spin_unlock(&req
->waitq
.lock
);
690 struct fuse_copy_state
{
692 struct fuse_req
*req
;
693 struct iov_iter
*iter
;
694 struct pipe_buffer
*pipebufs
;
695 struct pipe_buffer
*currbuf
;
696 struct pipe_inode_info
*pipe
;
697 unsigned long nr_segs
;
701 unsigned move_pages
:1;
704 static void fuse_copy_init(struct fuse_copy_state
*cs
, int write
,
705 struct iov_iter
*iter
)
707 memset(cs
, 0, sizeof(*cs
));
712 /* Unmap and put previous page of userspace buffer */
713 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
716 struct pipe_buffer
*buf
= cs
->currbuf
;
719 buf
->len
= PAGE_SIZE
- cs
->len
;
723 flush_dcache_page(cs
->pg
);
724 set_page_dirty_lock(cs
->pg
);
732 * Get another pagefull of userspace buffer, and map it to kernel
733 * address space, and lock request
735 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
740 err
= unlock_request(cs
->req
);
744 fuse_copy_finish(cs
);
746 struct pipe_buffer
*buf
= cs
->pipebufs
;
749 err
= pipe_buf_confirm(cs
->pipe
, buf
);
753 BUG_ON(!cs
->nr_segs
);
756 cs
->offset
= buf
->offset
;
761 if (cs
->nr_segs
== cs
->pipe
->buffers
)
764 page
= alloc_page(GFP_HIGHUSER
);
781 err
= iov_iter_get_pages(cs
->iter
, &page
, PAGE_SIZE
, 1, &off
);
788 iov_iter_advance(cs
->iter
, err
);
791 return lock_request(cs
->req
);
794 /* Do as much copy to/from userspace buffer as we can */
795 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
797 unsigned ncpy
= min(*size
, cs
->len
);
799 void *pgaddr
= kmap_atomic(cs
->pg
);
800 void *buf
= pgaddr
+ cs
->offset
;
803 memcpy(buf
, *val
, ncpy
);
805 memcpy(*val
, buf
, ncpy
);
807 kunmap_atomic(pgaddr
);
816 static int fuse_check_page(struct page
*page
)
818 if (page_mapcount(page
) ||
819 page
->mapping
!= NULL
||
820 page_count(page
) != 1 ||
821 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
828 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
829 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
);
835 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
838 struct page
*oldpage
= *pagep
;
839 struct page
*newpage
;
840 struct pipe_buffer
*buf
= cs
->pipebufs
;
842 err
= unlock_request(cs
->req
);
846 fuse_copy_finish(cs
);
848 err
= pipe_buf_confirm(cs
->pipe
, buf
);
852 BUG_ON(!cs
->nr_segs
);
858 if (cs
->len
!= PAGE_SIZE
)
861 if (pipe_buf_steal(cs
->pipe
, buf
) != 0)
866 if (!PageUptodate(newpage
))
867 SetPageUptodate(newpage
);
869 ClearPageMappedToDisk(newpage
);
871 if (fuse_check_page(newpage
) != 0)
872 goto out_fallback_unlock
;
875 * This is a new and locked page, it shouldn't be mapped or
876 * have any special flags on it
878 if (WARN_ON(page_mapped(oldpage
)))
879 goto out_fallback_unlock
;
880 if (WARN_ON(page_has_private(oldpage
)))
881 goto out_fallback_unlock
;
882 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
883 goto out_fallback_unlock
;
884 if (WARN_ON(PageMlocked(oldpage
)))
885 goto out_fallback_unlock
;
887 err
= replace_page_cache_page(oldpage
, newpage
, GFP_KERNEL
);
889 unlock_page(newpage
);
895 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
896 lru_cache_add_file(newpage
);
899 spin_lock(&cs
->req
->waitq
.lock
);
900 if (test_bit(FR_ABORTED
, &cs
->req
->flags
))
904 spin_unlock(&cs
->req
->waitq
.lock
);
907 unlock_page(newpage
);
912 unlock_page(oldpage
);
919 unlock_page(newpage
);
922 cs
->offset
= buf
->offset
;
924 err
= lock_request(cs
->req
);
931 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
932 unsigned offset
, unsigned count
)
934 struct pipe_buffer
*buf
;
937 if (cs
->nr_segs
== cs
->pipe
->buffers
)
940 err
= unlock_request(cs
->req
);
944 fuse_copy_finish(cs
);
949 buf
->offset
= offset
;
960 * Copy a page in the request to/from the userspace buffer. Must be
963 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
964 unsigned offset
, unsigned count
, int zeroing
)
967 struct page
*page
= *pagep
;
969 if (page
&& zeroing
&& count
< PAGE_SIZE
)
970 clear_highpage(page
);
973 if (cs
->write
&& cs
->pipebufs
&& page
) {
974 return fuse_ref_page(cs
, page
, offset
, count
);
975 } else if (!cs
->len
) {
976 if (cs
->move_pages
&& page
&&
977 offset
== 0 && count
== PAGE_SIZE
) {
978 err
= fuse_try_move_page(cs
, pagep
);
982 err
= fuse_copy_fill(cs
);
988 void *mapaddr
= kmap_atomic(page
);
989 void *buf
= mapaddr
+ offset
;
990 offset
+= fuse_copy_do(cs
, &buf
, &count
);
991 kunmap_atomic(mapaddr
);
993 offset
+= fuse_copy_do(cs
, NULL
, &count
);
995 if (page
&& !cs
->write
)
996 flush_dcache_page(page
);
1000 /* Copy pages in the request to/from userspace buffer */
1001 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
1005 struct fuse_req
*req
= cs
->req
;
1007 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
1009 unsigned offset
= req
->page_descs
[i
].offset
;
1010 unsigned count
= min(nbytes
, req
->page_descs
[i
].length
);
1012 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
1022 /* Copy a single argument in the request to/from userspace buffer */
1023 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
1027 int err
= fuse_copy_fill(cs
);
1031 fuse_copy_do(cs
, &val
, &size
);
1036 /* Copy request arguments to/from userspace buffer */
1037 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
1038 unsigned argpages
, struct fuse_arg
*args
,
1044 for (i
= 0; !err
&& i
< numargs
; i
++) {
1045 struct fuse_arg
*arg
= &args
[i
];
1046 if (i
== numargs
- 1 && argpages
)
1047 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
1049 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
1054 static int forget_pending(struct fuse_iqueue
*fiq
)
1056 return fiq
->forget_list_head
.next
!= NULL
;
1059 static int request_pending(struct fuse_iqueue
*fiq
)
1061 return !list_empty(&fiq
->pending
) || !list_empty(&fiq
->interrupts
) ||
1062 forget_pending(fiq
);
1066 * Transfer an interrupt request to userspace
1068 * Unlike other requests this is assembled on demand, without a need
1069 * to allocate a separate fuse_req structure.
1071 * Called with fiq->waitq.lock held, releases it
1073 static int fuse_read_interrupt(struct fuse_iqueue
*fiq
,
1074 struct fuse_copy_state
*cs
,
1075 size_t nbytes
, struct fuse_req
*req
)
1076 __releases(fiq
->waitq
.lock
)
1078 struct fuse_in_header ih
;
1079 struct fuse_interrupt_in arg
;
1080 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
1083 list_del_init(&req
->intr_entry
);
1084 req
->intr_unique
= fuse_get_unique(fiq
);
1085 memset(&ih
, 0, sizeof(ih
));
1086 memset(&arg
, 0, sizeof(arg
));
1088 ih
.opcode
= FUSE_INTERRUPT
;
1089 ih
.unique
= req
->intr_unique
;
1090 arg
.unique
= req
->in
.h
.unique
;
1092 spin_unlock(&fiq
->waitq
.lock
);
1093 if (nbytes
< reqsize
)
1096 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1098 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1099 fuse_copy_finish(cs
);
1101 return err
? err
: reqsize
;
1104 static struct fuse_forget_link
*dequeue_forget(struct fuse_iqueue
*fiq
,
1108 struct fuse_forget_link
*head
= fiq
->forget_list_head
.next
;
1109 struct fuse_forget_link
**newhead
= &head
;
1112 for (count
= 0; *newhead
!= NULL
&& count
< max
; count
++)
1113 newhead
= &(*newhead
)->next
;
1115 fiq
->forget_list_head
.next
= *newhead
;
1117 if (fiq
->forget_list_head
.next
== NULL
)
1118 fiq
->forget_list_tail
= &fiq
->forget_list_head
;
1126 static int fuse_read_single_forget(struct fuse_iqueue
*fiq
,
1127 struct fuse_copy_state
*cs
,
1129 __releases(fiq
->waitq
.lock
)
1132 struct fuse_forget_link
*forget
= dequeue_forget(fiq
, 1, NULL
);
1133 struct fuse_forget_in arg
= {
1134 .nlookup
= forget
->forget_one
.nlookup
,
1136 struct fuse_in_header ih
= {
1137 .opcode
= FUSE_FORGET
,
1138 .nodeid
= forget
->forget_one
.nodeid
,
1139 .unique
= fuse_get_unique(fiq
),
1140 .len
= sizeof(ih
) + sizeof(arg
),
1143 spin_unlock(&fiq
->waitq
.lock
);
1145 if (nbytes
< ih
.len
)
1148 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1150 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1151 fuse_copy_finish(cs
);
1159 static int fuse_read_batch_forget(struct fuse_iqueue
*fiq
,
1160 struct fuse_copy_state
*cs
, size_t nbytes
)
1161 __releases(fiq
->waitq
.lock
)
1164 unsigned max_forgets
;
1166 struct fuse_forget_link
*head
;
1167 struct fuse_batch_forget_in arg
= { .count
= 0 };
1168 struct fuse_in_header ih
= {
1169 .opcode
= FUSE_BATCH_FORGET
,
1170 .unique
= fuse_get_unique(fiq
),
1171 .len
= sizeof(ih
) + sizeof(arg
),
1174 if (nbytes
< ih
.len
) {
1175 spin_unlock(&fiq
->waitq
.lock
);
1179 max_forgets
= (nbytes
- ih
.len
) / sizeof(struct fuse_forget_one
);
1180 head
= dequeue_forget(fiq
, max_forgets
, &count
);
1181 spin_unlock(&fiq
->waitq
.lock
);
1184 ih
.len
+= count
* sizeof(struct fuse_forget_one
);
1185 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1187 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1190 struct fuse_forget_link
*forget
= head
;
1193 err
= fuse_copy_one(cs
, &forget
->forget_one
,
1194 sizeof(forget
->forget_one
));
1196 head
= forget
->next
;
1200 fuse_copy_finish(cs
);
1208 static int fuse_read_forget(struct fuse_conn
*fc
, struct fuse_iqueue
*fiq
,
1209 struct fuse_copy_state
*cs
,
1211 __releases(fiq
->waitq
.lock
)
1213 if (fc
->minor
< 16 || fiq
->forget_list_head
.next
->next
== NULL
)
1214 return fuse_read_single_forget(fiq
, cs
, nbytes
);
1216 return fuse_read_batch_forget(fiq
, cs
, nbytes
);
1220 * Read a single request into the userspace filesystem's buffer. This
1221 * function waits until a request is available, then removes it from
1222 * the pending list and copies request data to userspace buffer. If
1223 * no reply is needed (FORGET) or request has been aborted or there
1224 * was an error during the copying then it's finished by calling
1225 * request_end(). Otherwise add it to the processing list, and set
1228 static ssize_t
fuse_dev_do_read(struct fuse_dev
*fud
, struct file
*file
,
1229 struct fuse_copy_state
*cs
, size_t nbytes
)
1232 struct fuse_conn
*fc
= fud
->fc
;
1233 struct fuse_iqueue
*fiq
= &fc
->iq
;
1234 struct fuse_pqueue
*fpq
= &fud
->pq
;
1235 struct fuse_req
*req
;
1239 if (current_user_ns() != fc
->user_ns
)
1243 spin_lock(&fiq
->waitq
.lock
);
1245 if ((file
->f_flags
& O_NONBLOCK
) && fiq
->connected
&&
1246 !request_pending(fiq
))
1249 err
= wait_event_interruptible_exclusive_locked(fiq
->waitq
,
1250 !fiq
->connected
|| request_pending(fiq
));
1255 if (!fiq
->connected
)
1258 if (!list_empty(&fiq
->interrupts
)) {
1259 req
= list_entry(fiq
->interrupts
.next
, struct fuse_req
,
1261 return fuse_read_interrupt(fiq
, cs
, nbytes
, req
);
1264 if (forget_pending(fiq
)) {
1265 if (list_empty(&fiq
->pending
) || fiq
->forget_batch
-- > 0)
1266 return fuse_read_forget(fc
, fiq
, cs
, nbytes
);
1268 if (fiq
->forget_batch
<= -8)
1269 fiq
->forget_batch
= 16;
1272 req
= list_entry(fiq
->pending
.next
, struct fuse_req
, list
);
1273 clear_bit(FR_PENDING
, &req
->flags
);
1274 list_del_init(&req
->list
);
1275 spin_unlock(&fiq
->waitq
.lock
);
1278 reqsize
= in
->h
.len
;
1280 if (task_active_pid_ns(current
) != fc
->pid_ns
) {
1282 in
->h
.pid
= pid_vnr(find_pid_ns(in
->h
.pid
, fc
->pid_ns
));
1286 /* If request is too large, reply with an error and restart the read */
1287 if (nbytes
< reqsize
) {
1288 req
->out
.h
.error
= -EIO
;
1289 /* SETXATTR is special, since it may contain too large data */
1290 if (in
->h
.opcode
== FUSE_SETXATTR
)
1291 req
->out
.h
.error
= -E2BIG
;
1292 request_end(fc
, req
);
1295 spin_lock(&fpq
->lock
);
1296 list_add(&req
->list
, &fpq
->io
);
1297 spin_unlock(&fpq
->lock
);
1299 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1301 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1302 (struct fuse_arg
*) in
->args
, 0);
1303 fuse_copy_finish(cs
);
1304 spin_lock(&fpq
->lock
);
1305 clear_bit(FR_LOCKED
, &req
->flags
);
1306 if (!fpq
->connected
) {
1311 req
->out
.h
.error
= -EIO
;
1314 if (!test_bit(FR_ISREPLY
, &req
->flags
)) {
1318 list_move_tail(&req
->list
, &fpq
->processing
);
1319 __fuse_get_request(req
);
1320 spin_unlock(&fpq
->lock
);
1321 set_bit(FR_SENT
, &req
->flags
);
1322 /* matches barrier in request_wait_answer() */
1323 smp_mb__after_atomic();
1324 if (test_bit(FR_INTERRUPTED
, &req
->flags
))
1325 queue_interrupt(fiq
, req
);
1326 fuse_put_request(fc
, req
);
1331 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1332 list_del_init(&req
->list
);
1333 spin_unlock(&fpq
->lock
);
1334 request_end(fc
, req
);
1338 spin_unlock(&fiq
->waitq
.lock
);
1342 static int fuse_dev_open(struct inode
*inode
, struct file
*file
)
1345 * The fuse device's file's private_data is used to hold
1346 * the fuse_conn(ection) when it is mounted, and is used to
1347 * keep track of whether the file has been mounted already.
1349 file
->private_data
= NULL
;
1353 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, struct iov_iter
*to
)
1355 struct fuse_copy_state cs
;
1356 struct file
*file
= iocb
->ki_filp
;
1357 struct fuse_dev
*fud
= fuse_get_dev(file
);
1362 if (!iter_is_iovec(to
))
1365 fuse_copy_init(&cs
, 1, to
);
1367 return fuse_dev_do_read(fud
, file
, &cs
, iov_iter_count(to
));
1370 static ssize_t
fuse_dev_splice_read(struct file
*in
, loff_t
*ppos
,
1371 struct pipe_inode_info
*pipe
,
1372 size_t len
, unsigned int flags
)
1376 struct pipe_buffer
*bufs
;
1377 struct fuse_copy_state cs
;
1378 struct fuse_dev
*fud
= fuse_get_dev(in
);
1383 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1387 fuse_copy_init(&cs
, 1, NULL
);
1390 ret
= fuse_dev_do_read(fud
, in
, &cs
, len
);
1394 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1399 for (ret
= total
= 0; page_nr
< cs
.nr_segs
; total
+= ret
) {
1401 * Need to be careful about this. Having buf->ops in module
1402 * code can Oops if the buffer persists after module unload.
1404 bufs
[page_nr
].ops
= &nosteal_pipe_buf_ops
;
1405 bufs
[page_nr
].flags
= 0;
1406 ret
= add_to_pipe(pipe
, &bufs
[page_nr
++]);
1407 if (unlikely(ret
< 0))
1413 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1414 put_page(bufs
[page_nr
].page
);
1420 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1421 struct fuse_copy_state
*cs
)
1423 struct fuse_notify_poll_wakeup_out outarg
;
1426 if (size
!= sizeof(outarg
))
1429 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1433 fuse_copy_finish(cs
);
1434 return fuse_notify_poll_wakeup(fc
, &outarg
);
1437 fuse_copy_finish(cs
);
1441 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1442 struct fuse_copy_state
*cs
)
1444 struct fuse_notify_inval_inode_out outarg
;
1447 if (size
!= sizeof(outarg
))
1450 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1453 fuse_copy_finish(cs
);
1455 down_read(&fc
->killsb
);
1458 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1459 outarg
.off
, outarg
.len
);
1461 up_read(&fc
->killsb
);
1465 fuse_copy_finish(cs
);
1469 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1470 struct fuse_copy_state
*cs
)
1472 struct fuse_notify_inval_entry_out outarg
;
1477 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1482 if (size
< sizeof(outarg
))
1485 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1489 err
= -ENAMETOOLONG
;
1490 if (outarg
.namelen
> FUSE_NAME_MAX
)
1494 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1498 name
.len
= outarg
.namelen
;
1499 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1502 fuse_copy_finish(cs
);
1503 buf
[outarg
.namelen
] = 0;
1505 down_read(&fc
->killsb
);
1508 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, 0, &name
);
1509 up_read(&fc
->killsb
);
1515 fuse_copy_finish(cs
);
1519 static int fuse_notify_delete(struct fuse_conn
*fc
, unsigned int size
,
1520 struct fuse_copy_state
*cs
)
1522 struct fuse_notify_delete_out outarg
;
1527 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1532 if (size
< sizeof(outarg
))
1535 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1539 err
= -ENAMETOOLONG
;
1540 if (outarg
.namelen
> FUSE_NAME_MAX
)
1544 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1548 name
.len
= outarg
.namelen
;
1549 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1552 fuse_copy_finish(cs
);
1553 buf
[outarg
.namelen
] = 0;
1555 down_read(&fc
->killsb
);
1558 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
,
1559 outarg
.child
, &name
);
1560 up_read(&fc
->killsb
);
1566 fuse_copy_finish(cs
);
1570 static int fuse_notify_store(struct fuse_conn
*fc
, unsigned int size
,
1571 struct fuse_copy_state
*cs
)
1573 struct fuse_notify_store_out outarg
;
1574 struct inode
*inode
;
1575 struct address_space
*mapping
;
1579 unsigned int offset
;
1585 if (size
< sizeof(outarg
))
1588 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1593 if (size
- sizeof(outarg
) != outarg
.size
)
1596 nodeid
= outarg
.nodeid
;
1598 down_read(&fc
->killsb
);
1604 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1608 mapping
= inode
->i_mapping
;
1609 index
= outarg
.offset
>> PAGE_SHIFT
;
1610 offset
= outarg
.offset
& ~PAGE_MASK
;
1611 file_size
= i_size_read(inode
);
1612 end
= outarg
.offset
+ outarg
.size
;
1613 if (end
> file_size
) {
1615 fuse_write_update_size(inode
, file_size
);
1621 unsigned int this_num
;
1624 page
= find_or_create_page(mapping
, index
,
1625 mapping_gfp_mask(mapping
));
1629 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1630 err
= fuse_copy_page(cs
, &page
, offset
, this_num
, 0);
1631 if (!err
&& offset
== 0 &&
1632 (this_num
== PAGE_SIZE
|| file_size
== end
))
1633 SetPageUptodate(page
);
1650 up_read(&fc
->killsb
);
1652 fuse_copy_finish(cs
);
1656 static void fuse_retrieve_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1658 release_pages(req
->pages
, req
->num_pages
);
1661 static int fuse_retrieve(struct fuse_conn
*fc
, struct inode
*inode
,
1662 struct fuse_notify_retrieve_out
*outarg
)
1665 struct address_space
*mapping
= inode
->i_mapping
;
1666 struct fuse_req
*req
;
1670 unsigned int offset
;
1671 size_t total_len
= 0;
1674 offset
= outarg
->offset
& ~PAGE_MASK
;
1675 file_size
= i_size_read(inode
);
1678 if (outarg
->offset
> file_size
)
1680 else if (outarg
->offset
+ num
> file_size
)
1681 num
= file_size
- outarg
->offset
;
1683 num_pages
= (num
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1684 num_pages
= min(num_pages
, FUSE_MAX_PAGES_PER_REQ
);
1686 req
= fuse_get_req(fc
, num_pages
);
1688 return PTR_ERR(req
);
1690 req
->in
.h
.opcode
= FUSE_NOTIFY_REPLY
;
1691 req
->in
.h
.nodeid
= outarg
->nodeid
;
1692 req
->in
.numargs
= 2;
1693 req
->in
.argpages
= 1;
1694 req
->page_descs
[0].offset
= offset
;
1695 req
->end
= fuse_retrieve_end
;
1697 index
= outarg
->offset
>> PAGE_SHIFT
;
1699 while (num
&& req
->num_pages
< num_pages
) {
1701 unsigned int this_num
;
1703 page
= find_get_page(mapping
, index
);
1707 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1708 req
->pages
[req
->num_pages
] = page
;
1709 req
->page_descs
[req
->num_pages
].length
= this_num
;
1714 total_len
+= this_num
;
1717 req
->misc
.retrieve_in
.offset
= outarg
->offset
;
1718 req
->misc
.retrieve_in
.size
= total_len
;
1719 req
->in
.args
[0].size
= sizeof(req
->misc
.retrieve_in
);
1720 req
->in
.args
[0].value
= &req
->misc
.retrieve_in
;
1721 req
->in
.args
[1].size
= total_len
;
1723 err
= fuse_request_send_notify_reply(fc
, req
, outarg
->notify_unique
);
1725 fuse_retrieve_end(fc
, req
);
1730 static int fuse_notify_retrieve(struct fuse_conn
*fc
, unsigned int size
,
1731 struct fuse_copy_state
*cs
)
1733 struct fuse_notify_retrieve_out outarg
;
1734 struct inode
*inode
;
1738 if (size
!= sizeof(outarg
))
1741 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1745 fuse_copy_finish(cs
);
1747 down_read(&fc
->killsb
);
1750 u64 nodeid
= outarg
.nodeid
;
1752 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1754 err
= fuse_retrieve(fc
, inode
, &outarg
);
1758 up_read(&fc
->killsb
);
1763 fuse_copy_finish(cs
);
1767 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1768 unsigned int size
, struct fuse_copy_state
*cs
)
1770 /* Don't try to move pages (yet) */
1774 case FUSE_NOTIFY_POLL
:
1775 return fuse_notify_poll(fc
, size
, cs
);
1777 case FUSE_NOTIFY_INVAL_INODE
:
1778 return fuse_notify_inval_inode(fc
, size
, cs
);
1780 case FUSE_NOTIFY_INVAL_ENTRY
:
1781 return fuse_notify_inval_entry(fc
, size
, cs
);
1783 case FUSE_NOTIFY_STORE
:
1784 return fuse_notify_store(fc
, size
, cs
);
1786 case FUSE_NOTIFY_RETRIEVE
:
1787 return fuse_notify_retrieve(fc
, size
, cs
);
1789 case FUSE_NOTIFY_DELETE
:
1790 return fuse_notify_delete(fc
, size
, cs
);
1793 fuse_copy_finish(cs
);
1798 /* Look up request on processing list by unique ID */
1799 static struct fuse_req
*request_find(struct fuse_pqueue
*fpq
, u64 unique
)
1801 struct fuse_req
*req
;
1803 list_for_each_entry(req
, &fpq
->processing
, list
) {
1804 if (req
->in
.h
.unique
== unique
|| req
->intr_unique
== unique
)
1810 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1813 unsigned reqsize
= sizeof(struct fuse_out_header
);
1816 return nbytes
!= reqsize
? -EINVAL
: 0;
1818 reqsize
+= len_args(out
->numargs
, out
->args
);
1820 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1822 else if (reqsize
> nbytes
) {
1823 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1824 unsigned diffsize
= reqsize
- nbytes
;
1825 if (diffsize
> lastarg
->size
)
1827 lastarg
->size
-= diffsize
;
1829 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1834 * Write a single reply to a request. First the header is copied from
1835 * the write buffer. The request is then searched on the processing
1836 * list by the unique ID found in the header. If found, then remove
1837 * it from the list and copy the rest of the buffer to the request.
1838 * The request is finished by calling request_end()
1840 static ssize_t
fuse_dev_do_write(struct fuse_dev
*fud
,
1841 struct fuse_copy_state
*cs
, size_t nbytes
)
1844 struct fuse_conn
*fc
= fud
->fc
;
1845 struct fuse_pqueue
*fpq
= &fud
->pq
;
1846 struct fuse_req
*req
;
1847 struct fuse_out_header oh
;
1849 if (current_user_ns() != fc
->user_ns
)
1852 if (nbytes
< sizeof(struct fuse_out_header
))
1855 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1860 if (oh
.len
!= nbytes
)
1864 * Zero oh.unique indicates unsolicited notification message
1865 * and error contains notification code.
1868 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1869 return err
? err
: nbytes
;
1873 if (oh
.error
<= -1000 || oh
.error
> 0)
1876 spin_lock(&fpq
->lock
);
1878 if (!fpq
->connected
)
1881 req
= request_find(fpq
, oh
.unique
);
1885 /* Is it an interrupt reply? */
1886 if (req
->intr_unique
== oh
.unique
) {
1887 __fuse_get_request(req
);
1888 spin_unlock(&fpq
->lock
);
1891 if (nbytes
!= sizeof(struct fuse_out_header
)) {
1892 fuse_put_request(fc
, req
);
1896 if (oh
.error
== -ENOSYS
)
1897 fc
->no_interrupt
= 1;
1898 else if (oh
.error
== -EAGAIN
)
1899 queue_interrupt(&fc
->iq
, req
);
1900 fuse_put_request(fc
, req
);
1902 fuse_copy_finish(cs
);
1906 clear_bit(FR_SENT
, &req
->flags
);
1907 list_move(&req
->list
, &fpq
->io
);
1909 set_bit(FR_LOCKED
, &req
->flags
);
1910 spin_unlock(&fpq
->lock
);
1912 if (!req
->out
.page_replace
)
1915 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1916 fuse_copy_finish(cs
);
1918 spin_lock(&fpq
->lock
);
1919 clear_bit(FR_LOCKED
, &req
->flags
);
1920 if (!fpq
->connected
)
1923 req
->out
.h
.error
= -EIO
;
1924 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1925 list_del_init(&req
->list
);
1926 spin_unlock(&fpq
->lock
);
1928 request_end(fc
, req
);
1930 return err
? err
: nbytes
;
1933 spin_unlock(&fpq
->lock
);
1935 fuse_copy_finish(cs
);
1939 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, struct iov_iter
*from
)
1941 struct fuse_copy_state cs
;
1942 struct fuse_dev
*fud
= fuse_get_dev(iocb
->ki_filp
);
1947 if (!iter_is_iovec(from
))
1950 fuse_copy_init(&cs
, 0, from
);
1952 return fuse_dev_do_write(fud
, &cs
, iov_iter_count(from
));
1955 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1956 struct file
*out
, loff_t
*ppos
,
1957 size_t len
, unsigned int flags
)
1961 struct pipe_buffer
*bufs
;
1962 struct fuse_copy_state cs
;
1963 struct fuse_dev
*fud
;
1967 fud
= fuse_get_dev(out
);
1973 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1981 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
1982 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
1992 struct pipe_buffer
*ibuf
;
1993 struct pipe_buffer
*obuf
;
1995 BUG_ON(nbuf
>= pipe
->buffers
);
1996 BUG_ON(!pipe
->nrbufs
);
1997 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
2000 if (rem
>= ibuf
->len
) {
2003 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
2006 pipe_buf_get(pipe
, ibuf
);
2008 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
2010 ibuf
->offset
+= obuf
->len
;
2011 ibuf
->len
-= obuf
->len
;
2018 fuse_copy_init(&cs
, 0, NULL
);
2023 if (flags
& SPLICE_F_MOVE
)
2026 ret
= fuse_dev_do_write(fud
, &cs
, len
);
2028 for (idx
= 0; idx
< nbuf
; idx
++)
2029 pipe_buf_release(pipe
, &bufs
[idx
]);
2036 static unsigned fuse_dev_poll(struct file
*file
, poll_table
*wait
)
2038 unsigned mask
= POLLOUT
| POLLWRNORM
;
2039 struct fuse_iqueue
*fiq
;
2040 struct fuse_dev
*fud
= fuse_get_dev(file
);
2046 poll_wait(file
, &fiq
->waitq
, wait
);
2048 spin_lock(&fiq
->waitq
.lock
);
2049 if (!fiq
->connected
)
2051 else if (request_pending(fiq
))
2052 mask
|= POLLIN
| POLLRDNORM
;
2053 spin_unlock(&fiq
->waitq
.lock
);
2059 * Abort all requests on the given list (pending or processing)
2061 * This function releases and reacquires fc->lock
2063 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
2065 while (!list_empty(head
)) {
2066 struct fuse_req
*req
;
2067 req
= list_entry(head
->next
, struct fuse_req
, list
);
2068 req
->out
.h
.error
= -ECONNABORTED
;
2069 clear_bit(FR_SENT
, &req
->flags
);
2070 list_del_init(&req
->list
);
2071 request_end(fc
, req
);
2075 static void end_polls(struct fuse_conn
*fc
)
2079 p
= rb_first(&fc
->polled_files
);
2082 struct fuse_file
*ff
;
2083 ff
= rb_entry(p
, struct fuse_file
, polled_node
);
2084 wake_up_interruptible_all(&ff
->poll_wait
);
2091 * Abort all requests.
2093 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2096 * The same effect is usually achievable through killing the filesystem daemon
2097 * and all users of the filesystem. The exception is the combination of an
2098 * asynchronous request and the tricky deadlock (see
2099 * Documentation/filesystems/fuse.txt).
2101 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2102 * requests, they should be finished off immediately. Locked requests will be
2103 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2104 * requests. It is possible that some request will finish before we can. This
2105 * is OK, the request will in that case be removed from the list before we touch
2108 void fuse_abort_conn(struct fuse_conn
*fc
)
2110 struct fuse_iqueue
*fiq
= &fc
->iq
;
2112 spin_lock(&fc
->lock
);
2113 if (fc
->connected
) {
2114 struct fuse_dev
*fud
;
2115 struct fuse_req
*req
, *next
;
2121 fuse_set_initialized(fc
);
2122 list_for_each_entry(fud
, &fc
->devices
, entry
) {
2123 struct fuse_pqueue
*fpq
= &fud
->pq
;
2125 spin_lock(&fpq
->lock
);
2127 list_for_each_entry_safe(req
, next
, &fpq
->io
, list
) {
2128 req
->out
.h
.error
= -ECONNABORTED
;
2129 spin_lock(&req
->waitq
.lock
);
2130 set_bit(FR_ABORTED
, &req
->flags
);
2131 if (!test_bit(FR_LOCKED
, &req
->flags
)) {
2132 set_bit(FR_PRIVATE
, &req
->flags
);
2133 __fuse_get_request(req
);
2134 list_move(&req
->list
, &to_end1
);
2136 spin_unlock(&req
->waitq
.lock
);
2138 list_splice_init(&fpq
->processing
, &to_end2
);
2139 spin_unlock(&fpq
->lock
);
2141 fc
->max_background
= UINT_MAX
;
2144 spin_lock(&fiq
->waitq
.lock
);
2146 list_splice_init(&fiq
->pending
, &to_end2
);
2147 list_for_each_entry(req
, &to_end2
, list
)
2148 clear_bit(FR_PENDING
, &req
->flags
);
2149 while (forget_pending(fiq
))
2150 kfree(dequeue_forget(fiq
, 1, NULL
));
2151 wake_up_all_locked(&fiq
->waitq
);
2152 spin_unlock(&fiq
->waitq
.lock
);
2153 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
2155 wake_up_all(&fc
->blocked_waitq
);
2156 spin_unlock(&fc
->lock
);
2158 while (!list_empty(&to_end1
)) {
2159 req
= list_first_entry(&to_end1
, struct fuse_req
, list
);
2160 list_del_init(&req
->list
);
2161 request_end(fc
, req
);
2163 end_requests(fc
, &to_end2
);
2165 spin_unlock(&fc
->lock
);
2168 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
2170 void fuse_wait_aborted(struct fuse_conn
*fc
)
2172 wait_event(fc
->blocked_waitq
, atomic_read(&fc
->num_waiting
) == 0);
2175 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
2177 struct fuse_dev
*fud
= fuse_get_dev(file
);
2180 struct fuse_conn
*fc
= fud
->fc
;
2181 struct fuse_pqueue
*fpq
= &fud
->pq
;
2184 spin_lock(&fpq
->lock
);
2185 WARN_ON(!list_empty(&fpq
->io
));
2186 list_splice_init(&fpq
->processing
, &to_end
);
2187 spin_unlock(&fpq
->lock
);
2189 end_requests(fc
, &to_end
);
2191 /* Are we the last open device? */
2192 if (atomic_dec_and_test(&fc
->dev_count
)) {
2193 WARN_ON(fc
->iq
.fasync
!= NULL
);
2194 fuse_abort_conn(fc
);
2200 EXPORT_SYMBOL_GPL(fuse_dev_release
);
2202 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
2204 struct fuse_dev
*fud
= fuse_get_dev(file
);
2209 /* No locking - fasync_helper does its own locking */
2210 return fasync_helper(fd
, file
, on
, &fud
->fc
->iq
.fasync
);
2213 static int fuse_device_clone(struct fuse_conn
*fc
, struct file
*new)
2215 struct fuse_dev
*fud
;
2217 if (new->private_data
)
2220 fud
= fuse_dev_alloc(fc
);
2224 new->private_data
= fud
;
2225 atomic_inc(&fc
->dev_count
);
2230 static long fuse_dev_ioctl(struct file
*file
, unsigned int cmd
,
2235 if (cmd
== FUSE_DEV_IOC_CLONE
) {
2239 if (!get_user(oldfd
, (__u32 __user
*) arg
)) {
2240 struct file
*old
= fget(oldfd
);
2244 struct fuse_dev
*fud
= NULL
;
2247 * Check against file->f_op because CUSE
2248 * uses the same ioctl handler.
2250 if (old
->f_op
== file
->f_op
&&
2251 old
->f_cred
->user_ns
== file
->f_cred
->user_ns
)
2252 fud
= fuse_get_dev(old
);
2255 mutex_lock(&fuse_mutex
);
2256 err
= fuse_device_clone(fud
->fc
, file
);
2257 mutex_unlock(&fuse_mutex
);
2266 const struct file_operations fuse_dev_operations
= {
2267 .owner
= THIS_MODULE
,
2268 .open
= fuse_dev_open
,
2269 .llseek
= no_llseek
,
2270 .read_iter
= fuse_dev_read
,
2271 .splice_read
= fuse_dev_splice_read
,
2272 .write_iter
= fuse_dev_write
,
2273 .splice_write
= fuse_dev_splice_write
,
2274 .poll
= fuse_dev_poll
,
2275 .release
= fuse_dev_release
,
2276 .fasync
= fuse_dev_fasync
,
2277 .unlocked_ioctl
= fuse_dev_ioctl
,
2278 .compat_ioctl
= fuse_dev_ioctl
,
2280 EXPORT_SYMBOL_GPL(fuse_dev_operations
);
2282 static struct miscdevice fuse_miscdevice
= {
2283 .minor
= FUSE_MINOR
,
2285 .fops
= &fuse_dev_operations
,
2288 int __init
fuse_dev_init(void)
2291 fuse_req_cachep
= kmem_cache_create("fuse_request",
2292 sizeof(struct fuse_req
),
2294 if (!fuse_req_cachep
)
2297 err
= misc_register(&fuse_miscdevice
);
2299 goto out_cache_clean
;
2304 kmem_cache_destroy(fuse_req_cachep
);
2309 void fuse_dev_cleanup(void)
2311 misc_deregister(&fuse_miscdevice
);
2312 kmem_cache_destroy(fuse_req_cachep
);