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 * lockess check of fc->connected is okay, because atomic_dec_and_test()
138 * provides a memory barrier mached with the one in fuse_wait_aborted()
139 * to ensure no wake-up is missed.
141 if (atomic_dec_and_test(&fc
->num_waiting
) &&
142 !READ_ONCE(fc
->connected
)) {
143 /* wake up aborters */
144 wake_up_all(&fc
->blocked_waitq
);
148 static struct fuse_req
*__fuse_get_req(struct fuse_conn
*fc
, unsigned npages
,
151 struct fuse_req
*req
;
153 atomic_inc(&fc
->num_waiting
);
155 if (fuse_block_alloc(fc
, for_background
)) {
157 if (wait_event_killable_exclusive(fc
->blocked_waitq
,
158 !fuse_block_alloc(fc
, for_background
)))
161 /* Matches smp_wmb() in fuse_set_initialized() */
172 req
= fuse_request_alloc(npages
);
176 wake_up(&fc
->blocked_waitq
);
180 fuse_req_init_context(fc
, req
);
181 __set_bit(FR_WAITING
, &req
->flags
);
183 __set_bit(FR_BACKGROUND
, &req
->flags
);
184 if (req
->in
.h
.uid
== (uid_t
)-1 || req
->in
.h
.gid
== (gid_t
)-1) {
185 fuse_put_request(fc
, req
);
186 return ERR_PTR(-EOVERFLOW
);
192 fuse_drop_waiting(fc
);
196 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
)
198 return __fuse_get_req(fc
, npages
, false);
200 EXPORT_SYMBOL_GPL(fuse_get_req
);
202 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
205 return __fuse_get_req(fc
, npages
, true);
207 EXPORT_SYMBOL_GPL(fuse_get_req_for_background
);
210 * Return request in fuse_file->reserved_req. However that may
211 * currently be in use. If that is the case, wait for it to become
214 static struct fuse_req
*get_reserved_req(struct fuse_conn
*fc
,
217 struct fuse_req
*req
= NULL
;
218 struct fuse_file
*ff
= file
->private_data
;
221 wait_event(fc
->reserved_req_waitq
, ff
->reserved_req
);
222 spin_lock(&fc
->lock
);
223 if (ff
->reserved_req
) {
224 req
= ff
->reserved_req
;
225 ff
->reserved_req
= NULL
;
226 req
->stolen_file
= get_file(file
);
228 spin_unlock(&fc
->lock
);
235 * Put stolen request back into fuse_file->reserved_req
237 static void put_reserved_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
239 struct file
*file
= req
->stolen_file
;
240 struct fuse_file
*ff
= file
->private_data
;
242 spin_lock(&fc
->lock
);
243 fuse_request_init(req
, req
->pages
, req
->page_descs
, req
->max_pages
);
244 BUG_ON(ff
->reserved_req
);
245 ff
->reserved_req
= req
;
246 wake_up_all(&fc
->reserved_req_waitq
);
247 spin_unlock(&fc
->lock
);
252 * Gets a requests for a file operation, always succeeds
254 * This is used for sending the FLUSH request, which must get to
255 * userspace, due to POSIX locks which may need to be unlocked.
257 * If allocation fails due to OOM, use the reserved request in
260 * This is very unlikely to deadlock accidentally, since the
261 * filesystem should not have it's own file open. If deadlock is
262 * intentional, it can still be broken by "aborting" the filesystem.
264 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
267 struct fuse_req
*req
;
269 atomic_inc(&fc
->num_waiting
);
270 wait_event(fc
->blocked_waitq
, fc
->initialized
);
271 /* Matches smp_wmb() in fuse_set_initialized() */
273 req
= fuse_request_alloc(0);
275 req
= get_reserved_req(fc
, file
);
277 fuse_req_init_context(fc
, req
);
278 __set_bit(FR_WAITING
, &req
->flags
);
279 __clear_bit(FR_BACKGROUND
, &req
->flags
);
283 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
285 if (refcount_dec_and_test(&req
->count
)) {
286 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
288 * We get here in the unlikely case that a background
289 * request was allocated but not sent
291 spin_lock(&fc
->lock
);
293 wake_up(&fc
->blocked_waitq
);
294 spin_unlock(&fc
->lock
);
297 if (test_bit(FR_WAITING
, &req
->flags
)) {
298 __clear_bit(FR_WAITING
, &req
->flags
);
299 fuse_drop_waiting(fc
);
302 if (req
->stolen_file
)
303 put_reserved_req(fc
, req
);
305 fuse_request_free(req
);
308 EXPORT_SYMBOL_GPL(fuse_put_request
);
310 static unsigned len_args(unsigned numargs
, struct fuse_arg
*args
)
315 for (i
= 0; i
< numargs
; i
++)
316 nbytes
+= args
[i
].size
;
321 static u64
fuse_get_unique(struct fuse_iqueue
*fiq
)
323 return ++fiq
->reqctr
;
326 static void queue_request(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
328 req
->in
.h
.len
= sizeof(struct fuse_in_header
) +
329 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
330 list_add_tail(&req
->list
, &fiq
->pending
);
331 wake_up_locked(&fiq
->waitq
);
332 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
335 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
336 u64 nodeid
, u64 nlookup
)
338 struct fuse_iqueue
*fiq
= &fc
->iq
;
340 forget
->forget_one
.nodeid
= nodeid
;
341 forget
->forget_one
.nlookup
= nlookup
;
343 spin_lock(&fiq
->waitq
.lock
);
344 if (fiq
->connected
) {
345 fiq
->forget_list_tail
->next
= forget
;
346 fiq
->forget_list_tail
= forget
;
347 wake_up_locked(&fiq
->waitq
);
348 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
352 spin_unlock(&fiq
->waitq
.lock
);
355 static void flush_bg_queue(struct fuse_conn
*fc
)
357 while (fc
->active_background
< fc
->max_background
&&
358 !list_empty(&fc
->bg_queue
)) {
359 struct fuse_req
*req
;
360 struct fuse_iqueue
*fiq
= &fc
->iq
;
362 req
= list_entry(fc
->bg_queue
.next
, struct fuse_req
, list
);
363 list_del(&req
->list
);
364 fc
->active_background
++;
365 spin_lock(&fiq
->waitq
.lock
);
366 req
->in
.h
.unique
= fuse_get_unique(fiq
);
367 queue_request(fiq
, req
);
368 spin_unlock(&fiq
->waitq
.lock
);
373 * This function is called when a request is finished. Either a reply
374 * has arrived or it was aborted (and not yet sent) or some error
375 * occurred during communication with userspace, or the device file
376 * was closed. The requester thread is woken up (if still waiting),
377 * the 'end' callback is called if given, else the reference to the
378 * request is released
380 static void request_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
382 struct fuse_iqueue
*fiq
= &fc
->iq
;
384 if (test_and_set_bit(FR_FINISHED
, &req
->flags
))
387 spin_lock(&fiq
->waitq
.lock
);
388 list_del_init(&req
->intr_entry
);
389 spin_unlock(&fiq
->waitq
.lock
);
390 WARN_ON(test_bit(FR_PENDING
, &req
->flags
));
391 WARN_ON(test_bit(FR_SENT
, &req
->flags
));
392 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
393 spin_lock(&fc
->lock
);
394 clear_bit(FR_BACKGROUND
, &req
->flags
);
395 if (fc
->num_background
== fc
->max_background
) {
397 wake_up(&fc
->blocked_waitq
);
398 } else if (!fc
->blocked
) {
400 * Wake up next waiter, if any. It's okay to use
401 * waitqueue_active(), as we've already synced up
402 * fc->blocked with waiters with the wake_up() call
405 if (waitqueue_active(&fc
->blocked_waitq
))
406 wake_up(&fc
->blocked_waitq
);
409 if (fc
->num_background
== fc
->congestion_threshold
&& fc
->sb
) {
410 clear_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_SYNC
);
411 clear_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_ASYNC
);
413 fc
->num_background
--;
414 fc
->active_background
--;
416 spin_unlock(&fc
->lock
);
418 wake_up(&req
->waitq
);
422 fuse_put_request(fc
, req
);
425 static void queue_interrupt(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
427 spin_lock(&fiq
->waitq
.lock
);
428 if (test_bit(FR_FINISHED
, &req
->flags
)) {
429 spin_unlock(&fiq
->waitq
.lock
);
432 if (list_empty(&req
->intr_entry
)) {
433 list_add_tail(&req
->intr_entry
, &fiq
->interrupts
);
434 wake_up_locked(&fiq
->waitq
);
436 spin_unlock(&fiq
->waitq
.lock
);
437 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
440 static void request_wait_answer(struct fuse_conn
*fc
, struct fuse_req
*req
)
442 struct fuse_iqueue
*fiq
= &fc
->iq
;
445 if (!fc
->no_interrupt
) {
446 /* Any signal may interrupt this */
447 err
= wait_event_interruptible(req
->waitq
,
448 test_bit(FR_FINISHED
, &req
->flags
));
452 set_bit(FR_INTERRUPTED
, &req
->flags
);
453 /* matches barrier in fuse_dev_do_read() */
454 smp_mb__after_atomic();
455 if (test_bit(FR_SENT
, &req
->flags
))
456 queue_interrupt(fiq
, req
);
459 if (!test_bit(FR_FORCE
, &req
->flags
)) {
460 /* Only fatal signals may interrupt this */
461 err
= wait_event_killable(req
->waitq
,
462 test_bit(FR_FINISHED
, &req
->flags
));
466 spin_lock(&fiq
->waitq
.lock
);
467 /* Request is not yet in userspace, bail out */
468 if (test_bit(FR_PENDING
, &req
->flags
)) {
469 list_del(&req
->list
);
470 spin_unlock(&fiq
->waitq
.lock
);
471 __fuse_put_request(req
);
472 req
->out
.h
.error
= -EINTR
;
475 spin_unlock(&fiq
->waitq
.lock
);
479 * Either request is already in userspace, or it was forced.
482 wait_event(req
->waitq
, test_bit(FR_FINISHED
, &req
->flags
));
485 static void __fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
487 struct fuse_iqueue
*fiq
= &fc
->iq
;
489 BUG_ON(test_bit(FR_BACKGROUND
, &req
->flags
));
490 spin_lock(&fiq
->waitq
.lock
);
491 if (!fiq
->connected
) {
492 spin_unlock(&fiq
->waitq
.lock
);
493 req
->out
.h
.error
= -ENOTCONN
;
495 req
->in
.h
.unique
= fuse_get_unique(fiq
);
496 queue_request(fiq
, req
);
497 /* acquire extra reference, since request is still needed
498 after request_end() */
499 __fuse_get_request(req
);
500 spin_unlock(&fiq
->waitq
.lock
);
502 request_wait_answer(fc
, req
);
503 /* Pairs with smp_wmb() in request_end() */
508 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
510 __set_bit(FR_ISREPLY
, &req
->flags
);
511 if (!test_bit(FR_WAITING
, &req
->flags
)) {
512 __set_bit(FR_WAITING
, &req
->flags
);
513 atomic_inc(&fc
->num_waiting
);
515 __fuse_request_send(fc
, req
);
517 EXPORT_SYMBOL_GPL(fuse_request_send
);
519 static void fuse_adjust_compat(struct fuse_conn
*fc
, struct fuse_args
*args
)
521 if (fc
->minor
< 4 && args
->in
.h
.opcode
== FUSE_STATFS
)
522 args
->out
.args
[0].size
= FUSE_COMPAT_STATFS_SIZE
;
525 switch (args
->in
.h
.opcode
) {
532 args
->out
.args
[0].size
= FUSE_COMPAT_ENTRY_OUT_SIZE
;
536 args
->out
.args
[0].size
= FUSE_COMPAT_ATTR_OUT_SIZE
;
540 if (fc
->minor
< 12) {
541 switch (args
->in
.h
.opcode
) {
543 args
->in
.args
[0].size
= sizeof(struct fuse_open_in
);
546 args
->in
.args
[0].size
= FUSE_COMPAT_MKNOD_IN_SIZE
;
552 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
)
554 struct fuse_req
*req
;
557 req
= fuse_get_req(fc
, 0);
561 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
562 fuse_adjust_compat(fc
, args
);
564 req
->in
.h
.opcode
= args
->in
.h
.opcode
;
565 req
->in
.h
.nodeid
= args
->in
.h
.nodeid
;
566 req
->in
.numargs
= args
->in
.numargs
;
567 memcpy(req
->in
.args
, args
->in
.args
,
568 args
->in
.numargs
* sizeof(struct fuse_in_arg
));
569 req
->out
.argvar
= args
->out
.argvar
;
570 req
->out
.numargs
= args
->out
.numargs
;
571 memcpy(req
->out
.args
, args
->out
.args
,
572 args
->out
.numargs
* sizeof(struct fuse_arg
));
573 fuse_request_send(fc
, req
);
574 ret
= req
->out
.h
.error
;
575 if (!ret
&& args
->out
.argvar
) {
576 BUG_ON(args
->out
.numargs
!= 1);
577 ret
= req
->out
.args
[0].size
;
579 fuse_put_request(fc
, req
);
585 * Called under fc->lock
587 * fc->connected must have been checked previously
589 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
590 struct fuse_req
*req
)
592 BUG_ON(!test_bit(FR_BACKGROUND
, &req
->flags
));
593 if (!test_bit(FR_WAITING
, &req
->flags
)) {
594 __set_bit(FR_WAITING
, &req
->flags
);
595 atomic_inc(&fc
->num_waiting
);
597 __set_bit(FR_ISREPLY
, &req
->flags
);
598 fc
->num_background
++;
599 if (fc
->num_background
== fc
->max_background
)
601 if (fc
->num_background
== fc
->congestion_threshold
&& fc
->sb
) {
602 set_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_SYNC
);
603 set_bdi_congested(fc
->sb
->s_bdi
, BLK_RW_ASYNC
);
605 list_add_tail(&req
->list
, &fc
->bg_queue
);
609 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
612 spin_lock(&fc
->lock
);
614 fuse_request_send_background_locked(fc
, req
);
615 spin_unlock(&fc
->lock
);
617 spin_unlock(&fc
->lock
);
618 req
->out
.h
.error
= -ENOTCONN
;
620 fuse_put_request(fc
, req
);
623 EXPORT_SYMBOL_GPL(fuse_request_send_background
);
625 static int fuse_request_send_notify_reply(struct fuse_conn
*fc
,
626 struct fuse_req
*req
, u64 unique
)
629 struct fuse_iqueue
*fiq
= &fc
->iq
;
631 __clear_bit(FR_ISREPLY
, &req
->flags
);
632 req
->in
.h
.unique
= unique
;
633 spin_lock(&fiq
->waitq
.lock
);
634 if (fiq
->connected
) {
635 queue_request(fiq
, req
);
638 spin_unlock(&fiq
->waitq
.lock
);
643 void fuse_force_forget(struct file
*file
, u64 nodeid
)
645 struct inode
*inode
= file_inode(file
);
646 struct fuse_conn
*fc
= get_fuse_conn(inode
);
647 struct fuse_req
*req
;
648 struct fuse_forget_in inarg
;
650 memset(&inarg
, 0, sizeof(inarg
));
652 req
= fuse_get_req_nofail_nopages(fc
, file
);
653 req
->in
.h
.opcode
= FUSE_FORGET
;
654 req
->in
.h
.nodeid
= nodeid
;
656 req
->in
.args
[0].size
= sizeof(inarg
);
657 req
->in
.args
[0].value
= &inarg
;
658 __clear_bit(FR_ISREPLY
, &req
->flags
);
659 __fuse_request_send(fc
, req
);
661 fuse_put_request(fc
, req
);
665 * Lock the request. Up to the next unlock_request() there mustn't be
666 * anything that could cause a page-fault. If the request was already
669 static int lock_request(struct fuse_req
*req
)
673 spin_lock(&req
->waitq
.lock
);
674 if (test_bit(FR_ABORTED
, &req
->flags
))
677 set_bit(FR_LOCKED
, &req
->flags
);
678 spin_unlock(&req
->waitq
.lock
);
684 * Unlock request. If it was aborted while locked, caller is responsible
685 * for unlocking and ending the request.
687 static int unlock_request(struct fuse_req
*req
)
691 spin_lock(&req
->waitq
.lock
);
692 if (test_bit(FR_ABORTED
, &req
->flags
))
695 clear_bit(FR_LOCKED
, &req
->flags
);
696 spin_unlock(&req
->waitq
.lock
);
701 struct fuse_copy_state
{
703 struct fuse_req
*req
;
704 struct iov_iter
*iter
;
705 struct pipe_buffer
*pipebufs
;
706 struct pipe_buffer
*currbuf
;
707 struct pipe_inode_info
*pipe
;
708 unsigned long nr_segs
;
712 unsigned move_pages
:1;
715 static void fuse_copy_init(struct fuse_copy_state
*cs
, int write
,
716 struct iov_iter
*iter
)
718 memset(cs
, 0, sizeof(*cs
));
723 /* Unmap and put previous page of userspace buffer */
724 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
727 struct pipe_buffer
*buf
= cs
->currbuf
;
730 buf
->len
= PAGE_SIZE
- cs
->len
;
734 flush_dcache_page(cs
->pg
);
735 set_page_dirty_lock(cs
->pg
);
743 * Get another pagefull of userspace buffer, and map it to kernel
744 * address space, and lock request
746 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
751 err
= unlock_request(cs
->req
);
755 fuse_copy_finish(cs
);
757 struct pipe_buffer
*buf
= cs
->pipebufs
;
760 err
= pipe_buf_confirm(cs
->pipe
, buf
);
764 BUG_ON(!cs
->nr_segs
);
767 cs
->offset
= buf
->offset
;
772 if (cs
->nr_segs
== cs
->pipe
->buffers
)
775 page
= alloc_page(GFP_HIGHUSER
);
792 err
= iov_iter_get_pages(cs
->iter
, &page
, PAGE_SIZE
, 1, &off
);
799 iov_iter_advance(cs
->iter
, err
);
802 return lock_request(cs
->req
);
805 /* Do as much copy to/from userspace buffer as we can */
806 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
808 unsigned ncpy
= min(*size
, cs
->len
);
810 void *pgaddr
= kmap_atomic(cs
->pg
);
811 void *buf
= pgaddr
+ cs
->offset
;
814 memcpy(buf
, *val
, ncpy
);
816 memcpy(*val
, buf
, ncpy
);
818 kunmap_atomic(pgaddr
);
827 static int fuse_check_page(struct page
*page
)
829 if (page_mapcount(page
) ||
830 page
->mapping
!= NULL
||
831 page_count(page
) != 1 ||
832 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
839 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
840 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
);
846 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
849 struct page
*oldpage
= *pagep
;
850 struct page
*newpage
;
851 struct pipe_buffer
*buf
= cs
->pipebufs
;
853 err
= unlock_request(cs
->req
);
857 fuse_copy_finish(cs
);
859 err
= pipe_buf_confirm(cs
->pipe
, buf
);
863 BUG_ON(!cs
->nr_segs
);
869 if (cs
->len
!= PAGE_SIZE
)
872 if (pipe_buf_steal(cs
->pipe
, buf
) != 0)
877 if (!PageUptodate(newpage
))
878 SetPageUptodate(newpage
);
880 ClearPageMappedToDisk(newpage
);
882 if (fuse_check_page(newpage
) != 0)
883 goto out_fallback_unlock
;
886 * This is a new and locked page, it shouldn't be mapped or
887 * have any special flags on it
889 if (WARN_ON(page_mapped(oldpage
)))
890 goto out_fallback_unlock
;
891 if (WARN_ON(page_has_private(oldpage
)))
892 goto out_fallback_unlock
;
893 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
894 goto out_fallback_unlock
;
895 if (WARN_ON(PageMlocked(oldpage
)))
896 goto out_fallback_unlock
;
898 err
= replace_page_cache_page(oldpage
, newpage
, GFP_KERNEL
);
900 unlock_page(newpage
);
906 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
907 lru_cache_add_file(newpage
);
910 spin_lock(&cs
->req
->waitq
.lock
);
911 if (test_bit(FR_ABORTED
, &cs
->req
->flags
))
915 spin_unlock(&cs
->req
->waitq
.lock
);
918 unlock_page(newpage
);
923 unlock_page(oldpage
);
930 unlock_page(newpage
);
933 cs
->offset
= buf
->offset
;
935 err
= lock_request(cs
->req
);
942 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
943 unsigned offset
, unsigned count
)
945 struct pipe_buffer
*buf
;
948 if (cs
->nr_segs
== cs
->pipe
->buffers
)
951 err
= unlock_request(cs
->req
);
955 fuse_copy_finish(cs
);
960 buf
->offset
= offset
;
971 * Copy a page in the request to/from the userspace buffer. Must be
974 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
975 unsigned offset
, unsigned count
, int zeroing
)
978 struct page
*page
= *pagep
;
980 if (page
&& zeroing
&& count
< PAGE_SIZE
)
981 clear_highpage(page
);
984 if (cs
->write
&& cs
->pipebufs
&& page
) {
985 return fuse_ref_page(cs
, page
, offset
, count
);
986 } else if (!cs
->len
) {
987 if (cs
->move_pages
&& page
&&
988 offset
== 0 && count
== PAGE_SIZE
) {
989 err
= fuse_try_move_page(cs
, pagep
);
993 err
= fuse_copy_fill(cs
);
999 void *mapaddr
= kmap_atomic(page
);
1000 void *buf
= mapaddr
+ offset
;
1001 offset
+= fuse_copy_do(cs
, &buf
, &count
);
1002 kunmap_atomic(mapaddr
);
1004 offset
+= fuse_copy_do(cs
, NULL
, &count
);
1006 if (page
&& !cs
->write
)
1007 flush_dcache_page(page
);
1011 /* Copy pages in the request to/from userspace buffer */
1012 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
1016 struct fuse_req
*req
= cs
->req
;
1018 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
1020 unsigned offset
= req
->page_descs
[i
].offset
;
1021 unsigned count
= min(nbytes
, req
->page_descs
[i
].length
);
1023 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
1033 /* Copy a single argument in the request to/from userspace buffer */
1034 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
1038 int err
= fuse_copy_fill(cs
);
1042 fuse_copy_do(cs
, &val
, &size
);
1047 /* Copy request arguments to/from userspace buffer */
1048 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
1049 unsigned argpages
, struct fuse_arg
*args
,
1055 for (i
= 0; !err
&& i
< numargs
; i
++) {
1056 struct fuse_arg
*arg
= &args
[i
];
1057 if (i
== numargs
- 1 && argpages
)
1058 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
1060 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
1065 static int forget_pending(struct fuse_iqueue
*fiq
)
1067 return fiq
->forget_list_head
.next
!= NULL
;
1070 static int request_pending(struct fuse_iqueue
*fiq
)
1072 return !list_empty(&fiq
->pending
) || !list_empty(&fiq
->interrupts
) ||
1073 forget_pending(fiq
);
1077 * Transfer an interrupt request to userspace
1079 * Unlike other requests this is assembled on demand, without a need
1080 * to allocate a separate fuse_req structure.
1082 * Called with fiq->waitq.lock held, releases it
1084 static int fuse_read_interrupt(struct fuse_iqueue
*fiq
,
1085 struct fuse_copy_state
*cs
,
1086 size_t nbytes
, struct fuse_req
*req
)
1087 __releases(fiq
->waitq
.lock
)
1089 struct fuse_in_header ih
;
1090 struct fuse_interrupt_in arg
;
1091 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
1094 list_del_init(&req
->intr_entry
);
1095 req
->intr_unique
= fuse_get_unique(fiq
);
1096 memset(&ih
, 0, sizeof(ih
));
1097 memset(&arg
, 0, sizeof(arg
));
1099 ih
.opcode
= FUSE_INTERRUPT
;
1100 ih
.unique
= req
->intr_unique
;
1101 arg
.unique
= req
->in
.h
.unique
;
1103 spin_unlock(&fiq
->waitq
.lock
);
1104 if (nbytes
< reqsize
)
1107 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1109 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1110 fuse_copy_finish(cs
);
1112 return err
? err
: reqsize
;
1115 static struct fuse_forget_link
*dequeue_forget(struct fuse_iqueue
*fiq
,
1119 struct fuse_forget_link
*head
= fiq
->forget_list_head
.next
;
1120 struct fuse_forget_link
**newhead
= &head
;
1123 for (count
= 0; *newhead
!= NULL
&& count
< max
; count
++)
1124 newhead
= &(*newhead
)->next
;
1126 fiq
->forget_list_head
.next
= *newhead
;
1128 if (fiq
->forget_list_head
.next
== NULL
)
1129 fiq
->forget_list_tail
= &fiq
->forget_list_head
;
1137 static int fuse_read_single_forget(struct fuse_iqueue
*fiq
,
1138 struct fuse_copy_state
*cs
,
1140 __releases(fiq
->waitq
.lock
)
1143 struct fuse_forget_link
*forget
= dequeue_forget(fiq
, 1, NULL
);
1144 struct fuse_forget_in arg
= {
1145 .nlookup
= forget
->forget_one
.nlookup
,
1147 struct fuse_in_header ih
= {
1148 .opcode
= FUSE_FORGET
,
1149 .nodeid
= forget
->forget_one
.nodeid
,
1150 .unique
= fuse_get_unique(fiq
),
1151 .len
= sizeof(ih
) + sizeof(arg
),
1154 spin_unlock(&fiq
->waitq
.lock
);
1156 if (nbytes
< ih
.len
)
1159 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1161 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1162 fuse_copy_finish(cs
);
1170 static int fuse_read_batch_forget(struct fuse_iqueue
*fiq
,
1171 struct fuse_copy_state
*cs
, size_t nbytes
)
1172 __releases(fiq
->waitq
.lock
)
1175 unsigned max_forgets
;
1177 struct fuse_forget_link
*head
;
1178 struct fuse_batch_forget_in arg
= { .count
= 0 };
1179 struct fuse_in_header ih
= {
1180 .opcode
= FUSE_BATCH_FORGET
,
1181 .unique
= fuse_get_unique(fiq
),
1182 .len
= sizeof(ih
) + sizeof(arg
),
1185 if (nbytes
< ih
.len
) {
1186 spin_unlock(&fiq
->waitq
.lock
);
1190 max_forgets
= (nbytes
- ih
.len
) / sizeof(struct fuse_forget_one
);
1191 head
= dequeue_forget(fiq
, max_forgets
, &count
);
1192 spin_unlock(&fiq
->waitq
.lock
);
1195 ih
.len
+= count
* sizeof(struct fuse_forget_one
);
1196 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1198 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1201 struct fuse_forget_link
*forget
= head
;
1204 err
= fuse_copy_one(cs
, &forget
->forget_one
,
1205 sizeof(forget
->forget_one
));
1207 head
= forget
->next
;
1211 fuse_copy_finish(cs
);
1219 static int fuse_read_forget(struct fuse_conn
*fc
, struct fuse_iqueue
*fiq
,
1220 struct fuse_copy_state
*cs
,
1222 __releases(fiq
->waitq
.lock
)
1224 if (fc
->minor
< 16 || fiq
->forget_list_head
.next
->next
== NULL
)
1225 return fuse_read_single_forget(fiq
, cs
, nbytes
);
1227 return fuse_read_batch_forget(fiq
, cs
, nbytes
);
1231 * Read a single request into the userspace filesystem's buffer. This
1232 * function waits until a request is available, then removes it from
1233 * the pending list and copies request data to userspace buffer. If
1234 * no reply is needed (FORGET) or request has been aborted or there
1235 * was an error during the copying then it's finished by calling
1236 * request_end(). Otherwise add it to the processing list, and set
1239 static ssize_t
fuse_dev_do_read(struct fuse_dev
*fud
, struct file
*file
,
1240 struct fuse_copy_state
*cs
, size_t nbytes
)
1243 struct fuse_conn
*fc
= fud
->fc
;
1244 struct fuse_iqueue
*fiq
= &fc
->iq
;
1245 struct fuse_pqueue
*fpq
= &fud
->pq
;
1246 struct fuse_req
*req
;
1250 if (current_user_ns() != fc
->user_ns
)
1254 spin_lock(&fiq
->waitq
.lock
);
1256 if ((file
->f_flags
& O_NONBLOCK
) && fiq
->connected
&&
1257 !request_pending(fiq
))
1260 err
= wait_event_interruptible_exclusive_locked(fiq
->waitq
,
1261 !fiq
->connected
|| request_pending(fiq
));
1266 if (!fiq
->connected
)
1269 if (!list_empty(&fiq
->interrupts
)) {
1270 req
= list_entry(fiq
->interrupts
.next
, struct fuse_req
,
1272 return fuse_read_interrupt(fiq
, cs
, nbytes
, req
);
1275 if (forget_pending(fiq
)) {
1276 if (list_empty(&fiq
->pending
) || fiq
->forget_batch
-- > 0)
1277 return fuse_read_forget(fc
, fiq
, cs
, nbytes
);
1279 if (fiq
->forget_batch
<= -8)
1280 fiq
->forget_batch
= 16;
1283 req
= list_entry(fiq
->pending
.next
, struct fuse_req
, list
);
1284 clear_bit(FR_PENDING
, &req
->flags
);
1285 list_del_init(&req
->list
);
1286 spin_unlock(&fiq
->waitq
.lock
);
1289 reqsize
= in
->h
.len
;
1291 if (task_active_pid_ns(current
) != fc
->pid_ns
) {
1293 in
->h
.pid
= pid_vnr(find_pid_ns(in
->h
.pid
, fc
->pid_ns
));
1297 /* If request is too large, reply with an error and restart the read */
1298 if (nbytes
< reqsize
) {
1299 req
->out
.h
.error
= -EIO
;
1300 /* SETXATTR is special, since it may contain too large data */
1301 if (in
->h
.opcode
== FUSE_SETXATTR
)
1302 req
->out
.h
.error
= -E2BIG
;
1303 request_end(fc
, req
);
1306 spin_lock(&fpq
->lock
);
1307 list_add(&req
->list
, &fpq
->io
);
1308 spin_unlock(&fpq
->lock
);
1310 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1312 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1313 (struct fuse_arg
*) in
->args
, 0);
1314 fuse_copy_finish(cs
);
1315 spin_lock(&fpq
->lock
);
1316 clear_bit(FR_LOCKED
, &req
->flags
);
1317 if (!fpq
->connected
) {
1322 req
->out
.h
.error
= -EIO
;
1325 if (!test_bit(FR_ISREPLY
, &req
->flags
)) {
1329 list_move_tail(&req
->list
, &fpq
->processing
);
1330 __fuse_get_request(req
);
1331 set_bit(FR_SENT
, &req
->flags
);
1332 spin_unlock(&fpq
->lock
);
1333 /* matches barrier in request_wait_answer() */
1334 smp_mb__after_atomic();
1335 if (test_bit(FR_INTERRUPTED
, &req
->flags
))
1336 queue_interrupt(fiq
, req
);
1337 fuse_put_request(fc
, req
);
1342 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1343 list_del_init(&req
->list
);
1344 spin_unlock(&fpq
->lock
);
1345 request_end(fc
, req
);
1349 spin_unlock(&fiq
->waitq
.lock
);
1353 static int fuse_dev_open(struct inode
*inode
, struct file
*file
)
1356 * The fuse device's file's private_data is used to hold
1357 * the fuse_conn(ection) when it is mounted, and is used to
1358 * keep track of whether the file has been mounted already.
1360 file
->private_data
= NULL
;
1364 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, struct iov_iter
*to
)
1366 struct fuse_copy_state cs
;
1367 struct file
*file
= iocb
->ki_filp
;
1368 struct fuse_dev
*fud
= fuse_get_dev(file
);
1373 if (!iter_is_iovec(to
))
1376 fuse_copy_init(&cs
, 1, to
);
1378 return fuse_dev_do_read(fud
, file
, &cs
, iov_iter_count(to
));
1381 static ssize_t
fuse_dev_splice_read(struct file
*in
, loff_t
*ppos
,
1382 struct pipe_inode_info
*pipe
,
1383 size_t len
, unsigned int flags
)
1387 struct pipe_buffer
*bufs
;
1388 struct fuse_copy_state cs
;
1389 struct fuse_dev
*fud
= fuse_get_dev(in
);
1394 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1398 fuse_copy_init(&cs
, 1, NULL
);
1401 ret
= fuse_dev_do_read(fud
, in
, &cs
, len
);
1405 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1410 for (ret
= total
= 0; page_nr
< cs
.nr_segs
; total
+= ret
) {
1412 * Need to be careful about this. Having buf->ops in module
1413 * code can Oops if the buffer persists after module unload.
1415 bufs
[page_nr
].ops
= &nosteal_pipe_buf_ops
;
1416 bufs
[page_nr
].flags
= 0;
1417 ret
= add_to_pipe(pipe
, &bufs
[page_nr
++]);
1418 if (unlikely(ret
< 0))
1424 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1425 put_page(bufs
[page_nr
].page
);
1431 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1432 struct fuse_copy_state
*cs
)
1434 struct fuse_notify_poll_wakeup_out outarg
;
1437 if (size
!= sizeof(outarg
))
1440 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1444 fuse_copy_finish(cs
);
1445 return fuse_notify_poll_wakeup(fc
, &outarg
);
1448 fuse_copy_finish(cs
);
1452 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1453 struct fuse_copy_state
*cs
)
1455 struct fuse_notify_inval_inode_out outarg
;
1458 if (size
!= sizeof(outarg
))
1461 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1464 fuse_copy_finish(cs
);
1466 down_read(&fc
->killsb
);
1469 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1470 outarg
.off
, outarg
.len
);
1472 up_read(&fc
->killsb
);
1476 fuse_copy_finish(cs
);
1480 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1481 struct fuse_copy_state
*cs
)
1483 struct fuse_notify_inval_entry_out outarg
;
1488 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1493 if (size
< sizeof(outarg
))
1496 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1500 err
= -ENAMETOOLONG
;
1501 if (outarg
.namelen
> FUSE_NAME_MAX
)
1505 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1509 name
.len
= outarg
.namelen
;
1510 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1513 fuse_copy_finish(cs
);
1514 buf
[outarg
.namelen
] = 0;
1516 down_read(&fc
->killsb
);
1519 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, 0, &name
);
1520 up_read(&fc
->killsb
);
1526 fuse_copy_finish(cs
);
1530 static int fuse_notify_delete(struct fuse_conn
*fc
, unsigned int size
,
1531 struct fuse_copy_state
*cs
)
1533 struct fuse_notify_delete_out outarg
;
1538 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1543 if (size
< sizeof(outarg
))
1546 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1550 err
= -ENAMETOOLONG
;
1551 if (outarg
.namelen
> FUSE_NAME_MAX
)
1555 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1559 name
.len
= outarg
.namelen
;
1560 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1563 fuse_copy_finish(cs
);
1564 buf
[outarg
.namelen
] = 0;
1566 down_read(&fc
->killsb
);
1569 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
,
1570 outarg
.child
, &name
);
1571 up_read(&fc
->killsb
);
1577 fuse_copy_finish(cs
);
1581 static int fuse_notify_store(struct fuse_conn
*fc
, unsigned int size
,
1582 struct fuse_copy_state
*cs
)
1584 struct fuse_notify_store_out outarg
;
1585 struct inode
*inode
;
1586 struct address_space
*mapping
;
1590 unsigned int offset
;
1596 if (size
< sizeof(outarg
))
1599 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1604 if (size
- sizeof(outarg
) != outarg
.size
)
1607 nodeid
= outarg
.nodeid
;
1609 down_read(&fc
->killsb
);
1615 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1619 mapping
= inode
->i_mapping
;
1620 index
= outarg
.offset
>> PAGE_SHIFT
;
1621 offset
= outarg
.offset
& ~PAGE_MASK
;
1622 file_size
= i_size_read(inode
);
1623 end
= outarg
.offset
+ outarg
.size
;
1624 if (end
> file_size
) {
1626 fuse_write_update_size(inode
, file_size
);
1632 unsigned int this_num
;
1635 page
= find_or_create_page(mapping
, index
,
1636 mapping_gfp_mask(mapping
));
1640 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1641 err
= fuse_copy_page(cs
, &page
, offset
, this_num
, 0);
1642 if (!err
&& offset
== 0 &&
1643 (this_num
== PAGE_SIZE
|| file_size
== end
))
1644 SetPageUptodate(page
);
1661 up_read(&fc
->killsb
);
1663 fuse_copy_finish(cs
);
1667 static void fuse_retrieve_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1669 release_pages(req
->pages
, req
->num_pages
);
1672 static int fuse_retrieve(struct fuse_conn
*fc
, struct inode
*inode
,
1673 struct fuse_notify_retrieve_out
*outarg
)
1676 struct address_space
*mapping
= inode
->i_mapping
;
1677 struct fuse_req
*req
;
1681 unsigned int offset
;
1682 size_t total_len
= 0;
1685 offset
= outarg
->offset
& ~PAGE_MASK
;
1686 file_size
= i_size_read(inode
);
1689 if (outarg
->offset
> file_size
)
1691 else if (outarg
->offset
+ num
> file_size
)
1692 num
= file_size
- outarg
->offset
;
1694 num_pages
= (num
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1695 num_pages
= min(num_pages
, FUSE_MAX_PAGES_PER_REQ
);
1697 req
= fuse_get_req(fc
, num_pages
);
1699 return PTR_ERR(req
);
1701 req
->in
.h
.opcode
= FUSE_NOTIFY_REPLY
;
1702 req
->in
.h
.nodeid
= outarg
->nodeid
;
1703 req
->in
.numargs
= 2;
1704 req
->in
.argpages
= 1;
1705 req
->page_descs
[0].offset
= offset
;
1706 req
->end
= fuse_retrieve_end
;
1708 index
= outarg
->offset
>> PAGE_SHIFT
;
1710 while (num
&& req
->num_pages
< num_pages
) {
1712 unsigned int this_num
;
1714 page
= find_get_page(mapping
, index
);
1718 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1719 req
->pages
[req
->num_pages
] = page
;
1720 req
->page_descs
[req
->num_pages
].length
= this_num
;
1725 total_len
+= this_num
;
1728 req
->misc
.retrieve_in
.offset
= outarg
->offset
;
1729 req
->misc
.retrieve_in
.size
= total_len
;
1730 req
->in
.args
[0].size
= sizeof(req
->misc
.retrieve_in
);
1731 req
->in
.args
[0].value
= &req
->misc
.retrieve_in
;
1732 req
->in
.args
[1].size
= total_len
;
1734 err
= fuse_request_send_notify_reply(fc
, req
, outarg
->notify_unique
);
1736 fuse_retrieve_end(fc
, req
);
1737 fuse_put_request(fc
, req
);
1743 static int fuse_notify_retrieve(struct fuse_conn
*fc
, unsigned int size
,
1744 struct fuse_copy_state
*cs
)
1746 struct fuse_notify_retrieve_out outarg
;
1747 struct inode
*inode
;
1751 if (size
!= sizeof(outarg
))
1754 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1758 fuse_copy_finish(cs
);
1760 down_read(&fc
->killsb
);
1763 u64 nodeid
= outarg
.nodeid
;
1765 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1767 err
= fuse_retrieve(fc
, inode
, &outarg
);
1771 up_read(&fc
->killsb
);
1776 fuse_copy_finish(cs
);
1780 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1781 unsigned int size
, struct fuse_copy_state
*cs
)
1783 /* Don't try to move pages (yet) */
1787 case FUSE_NOTIFY_POLL
:
1788 return fuse_notify_poll(fc
, size
, cs
);
1790 case FUSE_NOTIFY_INVAL_INODE
:
1791 return fuse_notify_inval_inode(fc
, size
, cs
);
1793 case FUSE_NOTIFY_INVAL_ENTRY
:
1794 return fuse_notify_inval_entry(fc
, size
, cs
);
1796 case FUSE_NOTIFY_STORE
:
1797 return fuse_notify_store(fc
, size
, cs
);
1799 case FUSE_NOTIFY_RETRIEVE
:
1800 return fuse_notify_retrieve(fc
, size
, cs
);
1802 case FUSE_NOTIFY_DELETE
:
1803 return fuse_notify_delete(fc
, size
, cs
);
1806 fuse_copy_finish(cs
);
1811 /* Look up request on processing list by unique ID */
1812 static struct fuse_req
*request_find(struct fuse_pqueue
*fpq
, u64 unique
)
1814 struct fuse_req
*req
;
1816 list_for_each_entry(req
, &fpq
->processing
, list
) {
1817 if (req
->in
.h
.unique
== unique
|| req
->intr_unique
== unique
)
1823 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1826 unsigned reqsize
= sizeof(struct fuse_out_header
);
1829 return nbytes
!= reqsize
? -EINVAL
: 0;
1831 reqsize
+= len_args(out
->numargs
, out
->args
);
1833 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1835 else if (reqsize
> nbytes
) {
1836 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1837 unsigned diffsize
= reqsize
- nbytes
;
1838 if (diffsize
> lastarg
->size
)
1840 lastarg
->size
-= diffsize
;
1842 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1847 * Write a single reply to a request. First the header is copied from
1848 * the write buffer. The request is then searched on the processing
1849 * list by the unique ID found in the header. If found, then remove
1850 * it from the list and copy the rest of the buffer to the request.
1851 * The request is finished by calling request_end()
1853 static ssize_t
fuse_dev_do_write(struct fuse_dev
*fud
,
1854 struct fuse_copy_state
*cs
, size_t nbytes
)
1857 struct fuse_conn
*fc
= fud
->fc
;
1858 struct fuse_pqueue
*fpq
= &fud
->pq
;
1859 struct fuse_req
*req
;
1860 struct fuse_out_header oh
;
1862 if (current_user_ns() != fc
->user_ns
)
1865 if (nbytes
< sizeof(struct fuse_out_header
))
1868 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1873 if (oh
.len
!= nbytes
)
1877 * Zero oh.unique indicates unsolicited notification message
1878 * and error contains notification code.
1881 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1882 return err
? err
: nbytes
;
1886 if (oh
.error
<= -1000 || oh
.error
> 0)
1889 spin_lock(&fpq
->lock
);
1891 if (!fpq
->connected
)
1894 req
= request_find(fpq
, oh
.unique
);
1898 /* Is it an interrupt reply? */
1899 if (req
->intr_unique
== oh
.unique
) {
1900 __fuse_get_request(req
);
1901 spin_unlock(&fpq
->lock
);
1904 if (nbytes
!= sizeof(struct fuse_out_header
)) {
1905 fuse_put_request(fc
, req
);
1909 if (oh
.error
== -ENOSYS
)
1910 fc
->no_interrupt
= 1;
1911 else if (oh
.error
== -EAGAIN
)
1912 queue_interrupt(&fc
->iq
, req
);
1913 fuse_put_request(fc
, req
);
1915 fuse_copy_finish(cs
);
1919 clear_bit(FR_SENT
, &req
->flags
);
1920 list_move(&req
->list
, &fpq
->io
);
1922 set_bit(FR_LOCKED
, &req
->flags
);
1923 spin_unlock(&fpq
->lock
);
1925 if (!req
->out
.page_replace
)
1928 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1929 fuse_copy_finish(cs
);
1931 spin_lock(&fpq
->lock
);
1932 clear_bit(FR_LOCKED
, &req
->flags
);
1933 if (!fpq
->connected
)
1936 req
->out
.h
.error
= -EIO
;
1937 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1938 list_del_init(&req
->list
);
1939 spin_unlock(&fpq
->lock
);
1941 request_end(fc
, req
);
1943 return err
? err
: nbytes
;
1946 spin_unlock(&fpq
->lock
);
1948 fuse_copy_finish(cs
);
1952 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, struct iov_iter
*from
)
1954 struct fuse_copy_state cs
;
1955 struct fuse_dev
*fud
= fuse_get_dev(iocb
->ki_filp
);
1960 if (!iter_is_iovec(from
))
1963 fuse_copy_init(&cs
, 0, from
);
1965 return fuse_dev_do_write(fud
, &cs
, iov_iter_count(from
));
1968 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1969 struct file
*out
, loff_t
*ppos
,
1970 size_t len
, unsigned int flags
)
1974 struct pipe_buffer
*bufs
;
1975 struct fuse_copy_state cs
;
1976 struct fuse_dev
*fud
;
1980 fud
= fuse_get_dev(out
);
1986 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1994 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
1995 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
2005 struct pipe_buffer
*ibuf
;
2006 struct pipe_buffer
*obuf
;
2008 BUG_ON(nbuf
>= pipe
->buffers
);
2009 BUG_ON(!pipe
->nrbufs
);
2010 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
2013 if (rem
>= ibuf
->len
) {
2016 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
2019 pipe_buf_get(pipe
, ibuf
);
2021 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
2023 ibuf
->offset
+= obuf
->len
;
2024 ibuf
->len
-= obuf
->len
;
2031 fuse_copy_init(&cs
, 0, NULL
);
2036 if (flags
& SPLICE_F_MOVE
)
2039 ret
= fuse_dev_do_write(fud
, &cs
, len
);
2041 for (idx
= 0; idx
< nbuf
; idx
++)
2042 pipe_buf_release(pipe
, &bufs
[idx
]);
2049 static unsigned fuse_dev_poll(struct file
*file
, poll_table
*wait
)
2051 unsigned mask
= POLLOUT
| POLLWRNORM
;
2052 struct fuse_iqueue
*fiq
;
2053 struct fuse_dev
*fud
= fuse_get_dev(file
);
2059 poll_wait(file
, &fiq
->waitq
, wait
);
2061 spin_lock(&fiq
->waitq
.lock
);
2062 if (!fiq
->connected
)
2064 else if (request_pending(fiq
))
2065 mask
|= POLLIN
| POLLRDNORM
;
2066 spin_unlock(&fiq
->waitq
.lock
);
2072 * Abort all requests on the given list (pending or processing)
2074 * This function releases and reacquires fc->lock
2076 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
2078 while (!list_empty(head
)) {
2079 struct fuse_req
*req
;
2080 req
= list_entry(head
->next
, struct fuse_req
, list
);
2081 req
->out
.h
.error
= -ECONNABORTED
;
2082 clear_bit(FR_SENT
, &req
->flags
);
2083 list_del_init(&req
->list
);
2084 request_end(fc
, req
);
2088 static void end_polls(struct fuse_conn
*fc
)
2092 p
= rb_first(&fc
->polled_files
);
2095 struct fuse_file
*ff
;
2096 ff
= rb_entry(p
, struct fuse_file
, polled_node
);
2097 wake_up_interruptible_all(&ff
->poll_wait
);
2104 * Abort all requests.
2106 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2109 * The same effect is usually achievable through killing the filesystem daemon
2110 * and all users of the filesystem. The exception is the combination of an
2111 * asynchronous request and the tricky deadlock (see
2112 * Documentation/filesystems/fuse.txt).
2114 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2115 * requests, they should be finished off immediately. Locked requests will be
2116 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2117 * requests. It is possible that some request will finish before we can. This
2118 * is OK, the request will in that case be removed from the list before we touch
2121 void fuse_abort_conn(struct fuse_conn
*fc
)
2123 struct fuse_iqueue
*fiq
= &fc
->iq
;
2125 spin_lock(&fc
->lock
);
2126 if (fc
->connected
) {
2127 struct fuse_dev
*fud
;
2128 struct fuse_req
*req
, *next
;
2134 fuse_set_initialized(fc
);
2135 list_for_each_entry(fud
, &fc
->devices
, entry
) {
2136 struct fuse_pqueue
*fpq
= &fud
->pq
;
2138 spin_lock(&fpq
->lock
);
2140 list_for_each_entry_safe(req
, next
, &fpq
->io
, list
) {
2141 req
->out
.h
.error
= -ECONNABORTED
;
2142 spin_lock(&req
->waitq
.lock
);
2143 set_bit(FR_ABORTED
, &req
->flags
);
2144 if (!test_bit(FR_LOCKED
, &req
->flags
)) {
2145 set_bit(FR_PRIVATE
, &req
->flags
);
2146 __fuse_get_request(req
);
2147 list_move(&req
->list
, &to_end1
);
2149 spin_unlock(&req
->waitq
.lock
);
2151 list_splice_init(&fpq
->processing
, &to_end2
);
2152 spin_unlock(&fpq
->lock
);
2154 fc
->max_background
= UINT_MAX
;
2157 spin_lock(&fiq
->waitq
.lock
);
2159 list_splice_init(&fiq
->pending
, &to_end2
);
2160 list_for_each_entry(req
, &to_end2
, list
)
2161 clear_bit(FR_PENDING
, &req
->flags
);
2162 while (forget_pending(fiq
))
2163 kfree(dequeue_forget(fiq
, 1, NULL
));
2164 wake_up_all_locked(&fiq
->waitq
);
2165 spin_unlock(&fiq
->waitq
.lock
);
2166 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
2168 wake_up_all(&fc
->blocked_waitq
);
2169 spin_unlock(&fc
->lock
);
2171 while (!list_empty(&to_end1
)) {
2172 req
= list_first_entry(&to_end1
, struct fuse_req
, list
);
2173 list_del_init(&req
->list
);
2174 request_end(fc
, req
);
2176 end_requests(fc
, &to_end2
);
2178 spin_unlock(&fc
->lock
);
2181 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
2183 void fuse_wait_aborted(struct fuse_conn
*fc
)
2185 /* matches implicit memory barrier in fuse_drop_waiting() */
2187 wait_event(fc
->blocked_waitq
, atomic_read(&fc
->num_waiting
) == 0);
2190 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
2192 struct fuse_dev
*fud
= fuse_get_dev(file
);
2195 struct fuse_conn
*fc
= fud
->fc
;
2196 struct fuse_pqueue
*fpq
= &fud
->pq
;
2199 spin_lock(&fpq
->lock
);
2200 WARN_ON(!list_empty(&fpq
->io
));
2201 list_splice_init(&fpq
->processing
, &to_end
);
2202 spin_unlock(&fpq
->lock
);
2204 end_requests(fc
, &to_end
);
2206 /* Are we the last open device? */
2207 if (atomic_dec_and_test(&fc
->dev_count
)) {
2208 WARN_ON(fc
->iq
.fasync
!= NULL
);
2209 fuse_abort_conn(fc
);
2215 EXPORT_SYMBOL_GPL(fuse_dev_release
);
2217 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
2219 struct fuse_dev
*fud
= fuse_get_dev(file
);
2224 /* No locking - fasync_helper does its own locking */
2225 return fasync_helper(fd
, file
, on
, &fud
->fc
->iq
.fasync
);
2228 static int fuse_device_clone(struct fuse_conn
*fc
, struct file
*new)
2230 struct fuse_dev
*fud
;
2232 if (new->private_data
)
2235 fud
= fuse_dev_alloc(fc
);
2239 new->private_data
= fud
;
2240 atomic_inc(&fc
->dev_count
);
2245 static long fuse_dev_ioctl(struct file
*file
, unsigned int cmd
,
2250 if (cmd
== FUSE_DEV_IOC_CLONE
) {
2254 if (!get_user(oldfd
, (__u32 __user
*) arg
)) {
2255 struct file
*old
= fget(oldfd
);
2259 struct fuse_dev
*fud
= NULL
;
2262 * Check against file->f_op because CUSE
2263 * uses the same ioctl handler.
2265 if (old
->f_op
== file
->f_op
&&
2266 old
->f_cred
->user_ns
== file
->f_cred
->user_ns
)
2267 fud
= fuse_get_dev(old
);
2270 mutex_lock(&fuse_mutex
);
2271 err
= fuse_device_clone(fud
->fc
, file
);
2272 mutex_unlock(&fuse_mutex
);
2281 const struct file_operations fuse_dev_operations
= {
2282 .owner
= THIS_MODULE
,
2283 .open
= fuse_dev_open
,
2284 .llseek
= no_llseek
,
2285 .read_iter
= fuse_dev_read
,
2286 .splice_read
= fuse_dev_splice_read
,
2287 .write_iter
= fuse_dev_write
,
2288 .splice_write
= fuse_dev_splice_write
,
2289 .poll
= fuse_dev_poll
,
2290 .release
= fuse_dev_release
,
2291 .fasync
= fuse_dev_fasync
,
2292 .unlocked_ioctl
= fuse_dev_ioctl
,
2293 .compat_ioctl
= fuse_dev_ioctl
,
2295 EXPORT_SYMBOL_GPL(fuse_dev_operations
);
2297 static struct miscdevice fuse_miscdevice
= {
2298 .minor
= FUSE_MINOR
,
2300 .fops
= &fuse_dev_operations
,
2303 int __init
fuse_dev_init(void)
2306 fuse_req_cachep
= kmem_cache_create("fuse_request",
2307 sizeof(struct fuse_req
),
2309 if (!fuse_req_cachep
)
2312 err
= misc_register(&fuse_miscdevice
);
2314 goto out_cache_clean
;
2319 kmem_cache_destroy(fuse_req_cachep
);
2324 void fuse_dev_cleanup(void)
2326 misc_deregister(&fuse_miscdevice
);
2327 kmem_cache_destroy(fuse_req_cachep
);