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/uio.h>
15 #include <linux/miscdevice.h>
16 #include <linux/pagemap.h>
17 #include <linux/file.h>
18 #include <linux/slab.h>
19 #include <linux/pipe_fs_i.h>
20 #include <linux/swap.h>
21 #include <linux/splice.h>
22 #include <linux/sched.h>
24 MODULE_ALIAS_MISCDEV(FUSE_MINOR
);
25 MODULE_ALIAS("devname:fuse");
27 static struct kmem_cache
*fuse_req_cachep
;
29 static struct fuse_dev
*fuse_get_dev(struct file
*file
)
32 * Lockless access is OK, because file->private data is set
33 * once during mount and is valid until the file is released.
35 return ACCESS_ONCE(file
->private_data
);
38 static void fuse_request_init(struct fuse_req
*req
, struct page
**pages
,
39 struct fuse_page_desc
*page_descs
,
42 memset(req
, 0, sizeof(*req
));
43 memset(pages
, 0, sizeof(*pages
) * npages
);
44 memset(page_descs
, 0, sizeof(*page_descs
) * npages
);
45 INIT_LIST_HEAD(&req
->list
);
46 INIT_LIST_HEAD(&req
->intr_entry
);
47 init_waitqueue_head(&req
->waitq
);
48 atomic_set(&req
->count
, 1);
50 req
->page_descs
= page_descs
;
51 req
->max_pages
= npages
;
52 __set_bit(FR_PENDING
, &req
->flags
);
55 static struct fuse_req
*__fuse_request_alloc(unsigned npages
, gfp_t flags
)
57 struct fuse_req
*req
= kmem_cache_alloc(fuse_req_cachep
, flags
);
60 struct fuse_page_desc
*page_descs
;
62 if (npages
<= FUSE_REQ_INLINE_PAGES
) {
63 pages
= req
->inline_pages
;
64 page_descs
= req
->inline_page_descs
;
66 pages
= kmalloc(sizeof(struct page
*) * npages
, flags
);
67 page_descs
= kmalloc(sizeof(struct fuse_page_desc
) *
71 if (!pages
|| !page_descs
) {
74 kmem_cache_free(fuse_req_cachep
, req
);
78 fuse_request_init(req
, pages
, page_descs
, npages
);
83 struct fuse_req
*fuse_request_alloc(unsigned npages
)
85 return __fuse_request_alloc(npages
, GFP_KERNEL
);
87 EXPORT_SYMBOL_GPL(fuse_request_alloc
);
89 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
)
91 return __fuse_request_alloc(npages
, GFP_NOFS
);
94 void fuse_request_free(struct fuse_req
*req
)
96 if (req
->pages
!= req
->inline_pages
) {
98 kfree(req
->page_descs
);
100 kmem_cache_free(fuse_req_cachep
, req
);
103 void __fuse_get_request(struct fuse_req
*req
)
105 atomic_inc(&req
->count
);
108 /* Must be called with > 1 refcount */
109 static void __fuse_put_request(struct fuse_req
*req
)
111 BUG_ON(atomic_read(&req
->count
) < 2);
112 atomic_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 struct fuse_req
*__fuse_get_req(struct fuse_conn
*fc
, unsigned npages
,
137 struct fuse_req
*req
;
139 atomic_inc(&fc
->num_waiting
);
141 if (fuse_block_alloc(fc
, for_background
)) {
143 if (wait_event_killable_exclusive(fc
->blocked_waitq
,
144 !fuse_block_alloc(fc
, for_background
)))
147 /* Matches smp_wmb() in fuse_set_initialized() */
158 req
= fuse_request_alloc(npages
);
162 wake_up(&fc
->blocked_waitq
);
166 fuse_req_init_context(fc
, req
);
167 __set_bit(FR_WAITING
, &req
->flags
);
169 __set_bit(FR_BACKGROUND
, &req
->flags
);
170 if (req
->in
.h
.uid
== (uid_t
)-1 || req
->in
.h
.gid
== (gid_t
)-1) {
171 fuse_put_request(fc
, req
);
172 return ERR_PTR(-EOVERFLOW
);
178 atomic_dec(&fc
->num_waiting
);
182 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
)
184 return __fuse_get_req(fc
, npages
, false);
186 EXPORT_SYMBOL_GPL(fuse_get_req
);
188 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
191 return __fuse_get_req(fc
, npages
, true);
193 EXPORT_SYMBOL_GPL(fuse_get_req_for_background
);
196 * Return request in fuse_file->reserved_req. However that may
197 * currently be in use. If that is the case, wait for it to become
200 static struct fuse_req
*get_reserved_req(struct fuse_conn
*fc
,
203 struct fuse_req
*req
= NULL
;
204 struct fuse_file
*ff
= file
->private_data
;
207 wait_event(fc
->reserved_req_waitq
, ff
->reserved_req
);
208 spin_lock(&fc
->lock
);
209 if (ff
->reserved_req
) {
210 req
= ff
->reserved_req
;
211 ff
->reserved_req
= NULL
;
212 req
->stolen_file
= get_file(file
);
214 spin_unlock(&fc
->lock
);
221 * Put stolen request back into fuse_file->reserved_req
223 static void put_reserved_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
225 struct file
*file
= req
->stolen_file
;
226 struct fuse_file
*ff
= file
->private_data
;
228 spin_lock(&fc
->lock
);
229 fuse_request_init(req
, req
->pages
, req
->page_descs
, req
->max_pages
);
230 BUG_ON(ff
->reserved_req
);
231 ff
->reserved_req
= req
;
232 wake_up_all(&fc
->reserved_req_waitq
);
233 spin_unlock(&fc
->lock
);
238 * Gets a requests for a file operation, always succeeds
240 * This is used for sending the FLUSH request, which must get to
241 * userspace, due to POSIX locks which may need to be unlocked.
243 * If allocation fails due to OOM, use the reserved request in
246 * This is very unlikely to deadlock accidentally, since the
247 * filesystem should not have it's own file open. If deadlock is
248 * intentional, it can still be broken by "aborting" the filesystem.
250 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
253 struct fuse_req
*req
;
255 atomic_inc(&fc
->num_waiting
);
256 wait_event(fc
->blocked_waitq
, fc
->initialized
);
257 /* Matches smp_wmb() in fuse_set_initialized() */
259 req
= fuse_request_alloc(0);
261 req
= get_reserved_req(fc
, file
);
263 fuse_req_init_context(fc
, req
);
264 __set_bit(FR_WAITING
, &req
->flags
);
265 __clear_bit(FR_BACKGROUND
, &req
->flags
);
269 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
271 if (atomic_dec_and_test(&req
->count
)) {
272 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
274 * We get here in the unlikely case that a background
275 * request was allocated but not sent
277 spin_lock(&fc
->lock
);
279 wake_up(&fc
->blocked_waitq
);
280 spin_unlock(&fc
->lock
);
283 if (test_bit(FR_WAITING
, &req
->flags
)) {
284 __clear_bit(FR_WAITING
, &req
->flags
);
285 atomic_dec(&fc
->num_waiting
);
288 if (req
->stolen_file
)
289 put_reserved_req(fc
, req
);
291 fuse_request_free(req
);
294 EXPORT_SYMBOL_GPL(fuse_put_request
);
296 static unsigned len_args(unsigned numargs
, struct fuse_arg
*args
)
301 for (i
= 0; i
< numargs
; i
++)
302 nbytes
+= args
[i
].size
;
307 static u64
fuse_get_unique(struct fuse_iqueue
*fiq
)
309 return ++fiq
->reqctr
;
312 static void queue_request(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
314 req
->in
.h
.len
= sizeof(struct fuse_in_header
) +
315 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
316 list_add_tail(&req
->list
, &fiq
->pending
);
317 wake_up_locked(&fiq
->waitq
);
318 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
321 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
322 u64 nodeid
, u64 nlookup
)
324 struct fuse_iqueue
*fiq
= &fc
->iq
;
326 forget
->forget_one
.nodeid
= nodeid
;
327 forget
->forget_one
.nlookup
= nlookup
;
329 spin_lock(&fiq
->waitq
.lock
);
330 if (fiq
->connected
) {
331 fiq
->forget_list_tail
->next
= forget
;
332 fiq
->forget_list_tail
= forget
;
333 wake_up_locked(&fiq
->waitq
);
334 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
338 spin_unlock(&fiq
->waitq
.lock
);
341 static void flush_bg_queue(struct fuse_conn
*fc
)
343 while (fc
->active_background
< fc
->max_background
&&
344 !list_empty(&fc
->bg_queue
)) {
345 struct fuse_req
*req
;
346 struct fuse_iqueue
*fiq
= &fc
->iq
;
348 req
= list_entry(fc
->bg_queue
.next
, struct fuse_req
, list
);
349 list_del(&req
->list
);
350 fc
->active_background
++;
351 spin_lock(&fiq
->waitq
.lock
);
352 req
->in
.h
.unique
= fuse_get_unique(fiq
);
353 queue_request(fiq
, req
);
354 spin_unlock(&fiq
->waitq
.lock
);
359 * This function is called when a request is finished. Either a reply
360 * has arrived or it was aborted (and not yet sent) or some error
361 * occurred during communication with userspace, or the device file
362 * was closed. The requester thread is woken up (if still waiting),
363 * the 'end' callback is called if given, else the reference to the
364 * request is released
366 static void request_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
368 struct fuse_iqueue
*fiq
= &fc
->iq
;
370 if (test_and_set_bit(FR_FINISHED
, &req
->flags
))
373 spin_lock(&fiq
->waitq
.lock
);
374 list_del_init(&req
->intr_entry
);
375 spin_unlock(&fiq
->waitq
.lock
);
376 WARN_ON(test_bit(FR_PENDING
, &req
->flags
));
377 WARN_ON(test_bit(FR_SENT
, &req
->flags
));
378 if (test_bit(FR_BACKGROUND
, &req
->flags
)) {
379 spin_lock(&fc
->lock
);
380 clear_bit(FR_BACKGROUND
, &req
->flags
);
381 if (fc
->num_background
== fc
->max_background
)
384 /* Wake up next waiter, if any */
385 if (!fc
->blocked
&& waitqueue_active(&fc
->blocked_waitq
))
386 wake_up(&fc
->blocked_waitq
);
388 if (fc
->num_background
== fc
->congestion_threshold
&&
389 fc
->connected
&& fc
->bdi_initialized
) {
390 clear_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
391 clear_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
393 fc
->num_background
--;
394 fc
->active_background
--;
396 spin_unlock(&fc
->lock
);
398 wake_up(&req
->waitq
);
401 fuse_put_request(fc
, req
);
404 static void queue_interrupt(struct fuse_iqueue
*fiq
, struct fuse_req
*req
)
406 spin_lock(&fiq
->waitq
.lock
);
407 if (test_bit(FR_FINISHED
, &req
->flags
)) {
408 spin_unlock(&fiq
->waitq
.lock
);
411 if (list_empty(&req
->intr_entry
)) {
412 list_add_tail(&req
->intr_entry
, &fiq
->interrupts
);
413 wake_up_locked(&fiq
->waitq
);
415 spin_unlock(&fiq
->waitq
.lock
);
416 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
419 static void request_wait_answer(struct fuse_conn
*fc
, struct fuse_req
*req
)
421 struct fuse_iqueue
*fiq
= &fc
->iq
;
424 if (!fc
->no_interrupt
) {
425 /* Any signal may interrupt this */
426 err
= wait_event_interruptible(req
->waitq
,
427 test_bit(FR_FINISHED
, &req
->flags
));
431 set_bit(FR_INTERRUPTED
, &req
->flags
);
432 /* matches barrier in fuse_dev_do_read() */
433 smp_mb__after_atomic();
434 if (test_bit(FR_SENT
, &req
->flags
))
435 queue_interrupt(fiq
, req
);
438 if (!test_bit(FR_FORCE
, &req
->flags
)) {
439 /* Only fatal signals may interrupt this */
440 err
= wait_event_killable(req
->waitq
,
441 test_bit(FR_FINISHED
, &req
->flags
));
445 spin_lock(&fiq
->waitq
.lock
);
446 /* Request is not yet in userspace, bail out */
447 if (test_bit(FR_PENDING
, &req
->flags
)) {
448 list_del(&req
->list
);
449 spin_unlock(&fiq
->waitq
.lock
);
450 __fuse_put_request(req
);
451 req
->out
.h
.error
= -EINTR
;
454 spin_unlock(&fiq
->waitq
.lock
);
458 * Either request is already in userspace, or it was forced.
461 wait_event(req
->waitq
, test_bit(FR_FINISHED
, &req
->flags
));
464 static void __fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
466 struct fuse_iqueue
*fiq
= &fc
->iq
;
468 BUG_ON(test_bit(FR_BACKGROUND
, &req
->flags
));
469 spin_lock(&fiq
->waitq
.lock
);
470 if (!fiq
->connected
) {
471 spin_unlock(&fiq
->waitq
.lock
);
472 req
->out
.h
.error
= -ENOTCONN
;
474 req
->in
.h
.unique
= fuse_get_unique(fiq
);
475 queue_request(fiq
, req
);
476 /* acquire extra reference, since request is still needed
477 after request_end() */
478 __fuse_get_request(req
);
479 spin_unlock(&fiq
->waitq
.lock
);
481 request_wait_answer(fc
, req
);
482 /* Pairs with smp_wmb() in request_end() */
487 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
489 __set_bit(FR_ISREPLY
, &req
->flags
);
490 if (!test_bit(FR_WAITING
, &req
->flags
)) {
491 __set_bit(FR_WAITING
, &req
->flags
);
492 atomic_inc(&fc
->num_waiting
);
494 __fuse_request_send(fc
, req
);
496 EXPORT_SYMBOL_GPL(fuse_request_send
);
498 static void fuse_adjust_compat(struct fuse_conn
*fc
, struct fuse_args
*args
)
500 if (fc
->minor
< 4 && args
->in
.h
.opcode
== FUSE_STATFS
)
501 args
->out
.args
[0].size
= FUSE_COMPAT_STATFS_SIZE
;
504 switch (args
->in
.h
.opcode
) {
511 args
->out
.args
[0].size
= FUSE_COMPAT_ENTRY_OUT_SIZE
;
515 args
->out
.args
[0].size
= FUSE_COMPAT_ATTR_OUT_SIZE
;
519 if (fc
->minor
< 12) {
520 switch (args
->in
.h
.opcode
) {
522 args
->in
.args
[0].size
= sizeof(struct fuse_open_in
);
525 args
->in
.args
[0].size
= FUSE_COMPAT_MKNOD_IN_SIZE
;
531 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
)
533 struct fuse_req
*req
;
536 req
= fuse_get_req(fc
, 0);
540 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
541 fuse_adjust_compat(fc
, args
);
543 req
->in
.h
.opcode
= args
->in
.h
.opcode
;
544 req
->in
.h
.nodeid
= args
->in
.h
.nodeid
;
545 req
->in
.numargs
= args
->in
.numargs
;
546 memcpy(req
->in
.args
, args
->in
.args
,
547 args
->in
.numargs
* sizeof(struct fuse_in_arg
));
548 req
->out
.argvar
= args
->out
.argvar
;
549 req
->out
.numargs
= args
->out
.numargs
;
550 memcpy(req
->out
.args
, args
->out
.args
,
551 args
->out
.numargs
* sizeof(struct fuse_arg
));
552 fuse_request_send(fc
, req
);
553 ret
= req
->out
.h
.error
;
554 if (!ret
&& args
->out
.argvar
) {
555 BUG_ON(args
->out
.numargs
!= 1);
556 ret
= req
->out
.args
[0].size
;
558 fuse_put_request(fc
, req
);
564 * Called under fc->lock
566 * fc->connected must have been checked previously
568 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
569 struct fuse_req
*req
)
571 BUG_ON(!test_bit(FR_BACKGROUND
, &req
->flags
));
572 if (!test_bit(FR_WAITING
, &req
->flags
)) {
573 __set_bit(FR_WAITING
, &req
->flags
);
574 atomic_inc(&fc
->num_waiting
);
576 __set_bit(FR_ISREPLY
, &req
->flags
);
577 fc
->num_background
++;
578 if (fc
->num_background
== fc
->max_background
)
580 if (fc
->num_background
== fc
->congestion_threshold
&&
581 fc
->bdi_initialized
) {
582 set_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
583 set_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
585 list_add_tail(&req
->list
, &fc
->bg_queue
);
589 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
592 spin_lock(&fc
->lock
);
594 fuse_request_send_background_locked(fc
, req
);
595 spin_unlock(&fc
->lock
);
597 spin_unlock(&fc
->lock
);
598 req
->out
.h
.error
= -ENOTCONN
;
600 fuse_put_request(fc
, req
);
603 EXPORT_SYMBOL_GPL(fuse_request_send_background
);
605 static int fuse_request_send_notify_reply(struct fuse_conn
*fc
,
606 struct fuse_req
*req
, u64 unique
)
609 struct fuse_iqueue
*fiq
= &fc
->iq
;
611 __clear_bit(FR_ISREPLY
, &req
->flags
);
612 req
->in
.h
.unique
= unique
;
613 spin_lock(&fiq
->waitq
.lock
);
614 if (fiq
->connected
) {
615 queue_request(fiq
, req
);
618 spin_unlock(&fiq
->waitq
.lock
);
623 void fuse_force_forget(struct file
*file
, u64 nodeid
)
625 struct inode
*inode
= file_inode(file
);
626 struct fuse_conn
*fc
= get_fuse_conn(inode
);
627 struct fuse_req
*req
;
628 struct fuse_forget_in inarg
;
630 memset(&inarg
, 0, sizeof(inarg
));
632 req
= fuse_get_req_nofail_nopages(fc
, file
);
633 req
->in
.h
.opcode
= FUSE_FORGET
;
634 req
->in
.h
.nodeid
= nodeid
;
636 req
->in
.args
[0].size
= sizeof(inarg
);
637 req
->in
.args
[0].value
= &inarg
;
638 __clear_bit(FR_ISREPLY
, &req
->flags
);
639 __fuse_request_send(fc
, req
);
641 fuse_put_request(fc
, req
);
645 * Lock the request. Up to the next unlock_request() there mustn't be
646 * anything that could cause a page-fault. If the request was already
649 static int lock_request(struct fuse_req
*req
)
653 spin_lock(&req
->waitq
.lock
);
654 if (test_bit(FR_ABORTED
, &req
->flags
))
657 set_bit(FR_LOCKED
, &req
->flags
);
658 spin_unlock(&req
->waitq
.lock
);
664 * Unlock request. If it was aborted while locked, caller is responsible
665 * for unlocking and ending the request.
667 static int unlock_request(struct fuse_req
*req
)
671 spin_lock(&req
->waitq
.lock
);
672 if (test_bit(FR_ABORTED
, &req
->flags
))
675 clear_bit(FR_LOCKED
, &req
->flags
);
676 spin_unlock(&req
->waitq
.lock
);
681 struct fuse_copy_state
{
683 struct fuse_req
*req
;
684 struct iov_iter
*iter
;
685 struct pipe_buffer
*pipebufs
;
686 struct pipe_buffer
*currbuf
;
687 struct pipe_inode_info
*pipe
;
688 unsigned long nr_segs
;
692 unsigned move_pages
:1;
695 static void fuse_copy_init(struct fuse_copy_state
*cs
, int write
,
696 struct iov_iter
*iter
)
698 memset(cs
, 0, sizeof(*cs
));
703 /* Unmap and put previous page of userspace buffer */
704 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
707 struct pipe_buffer
*buf
= cs
->currbuf
;
710 buf
->len
= PAGE_SIZE
- cs
->len
;
714 flush_dcache_page(cs
->pg
);
715 set_page_dirty_lock(cs
->pg
);
723 * Get another pagefull of userspace buffer, and map it to kernel
724 * address space, and lock request
726 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
731 err
= unlock_request(cs
->req
);
735 fuse_copy_finish(cs
);
737 struct pipe_buffer
*buf
= cs
->pipebufs
;
740 err
= pipe_buf_confirm(cs
->pipe
, buf
);
744 BUG_ON(!cs
->nr_segs
);
747 cs
->offset
= buf
->offset
;
752 if (cs
->nr_segs
== cs
->pipe
->buffers
)
755 page
= alloc_page(GFP_HIGHUSER
);
772 err
= iov_iter_get_pages(cs
->iter
, &page
, PAGE_SIZE
, 1, &off
);
779 iov_iter_advance(cs
->iter
, err
);
782 return lock_request(cs
->req
);
785 /* Do as much copy to/from userspace buffer as we can */
786 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
788 unsigned ncpy
= min(*size
, cs
->len
);
790 void *pgaddr
= kmap_atomic(cs
->pg
);
791 void *buf
= pgaddr
+ cs
->offset
;
794 memcpy(buf
, *val
, ncpy
);
796 memcpy(*val
, buf
, ncpy
);
798 kunmap_atomic(pgaddr
);
807 static int fuse_check_page(struct page
*page
)
809 if (page_mapcount(page
) ||
810 page
->mapping
!= NULL
||
811 page_count(page
) != 1 ||
812 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
819 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
820 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
);
826 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
829 struct page
*oldpage
= *pagep
;
830 struct page
*newpage
;
831 struct pipe_buffer
*buf
= cs
->pipebufs
;
833 err
= unlock_request(cs
->req
);
837 fuse_copy_finish(cs
);
839 err
= pipe_buf_confirm(cs
->pipe
, buf
);
843 BUG_ON(!cs
->nr_segs
);
849 if (cs
->len
!= PAGE_SIZE
)
852 if (pipe_buf_steal(cs
->pipe
, buf
) != 0)
857 if (!PageUptodate(newpage
))
858 SetPageUptodate(newpage
);
860 ClearPageMappedToDisk(newpage
);
862 if (fuse_check_page(newpage
) != 0)
863 goto out_fallback_unlock
;
866 * This is a new and locked page, it shouldn't be mapped or
867 * have any special flags on it
869 if (WARN_ON(page_mapped(oldpage
)))
870 goto out_fallback_unlock
;
871 if (WARN_ON(page_has_private(oldpage
)))
872 goto out_fallback_unlock
;
873 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
874 goto out_fallback_unlock
;
875 if (WARN_ON(PageMlocked(oldpage
)))
876 goto out_fallback_unlock
;
878 err
= replace_page_cache_page(oldpage
, newpage
, GFP_KERNEL
);
880 unlock_page(newpage
);
886 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
887 lru_cache_add_file(newpage
);
890 spin_lock(&cs
->req
->waitq
.lock
);
891 if (test_bit(FR_ABORTED
, &cs
->req
->flags
))
895 spin_unlock(&cs
->req
->waitq
.lock
);
898 unlock_page(newpage
);
903 unlock_page(oldpage
);
910 unlock_page(newpage
);
913 cs
->offset
= buf
->offset
;
915 err
= lock_request(cs
->req
);
922 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
923 unsigned offset
, unsigned count
)
925 struct pipe_buffer
*buf
;
928 if (cs
->nr_segs
== cs
->pipe
->buffers
)
931 err
= unlock_request(cs
->req
);
935 fuse_copy_finish(cs
);
940 buf
->offset
= offset
;
951 * Copy a page in the request to/from the userspace buffer. Must be
954 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
955 unsigned offset
, unsigned count
, int zeroing
)
958 struct page
*page
= *pagep
;
960 if (page
&& zeroing
&& count
< PAGE_SIZE
)
961 clear_highpage(page
);
964 if (cs
->write
&& cs
->pipebufs
&& page
) {
965 return fuse_ref_page(cs
, page
, offset
, count
);
966 } else if (!cs
->len
) {
967 if (cs
->move_pages
&& page
&&
968 offset
== 0 && count
== PAGE_SIZE
) {
969 err
= fuse_try_move_page(cs
, pagep
);
973 err
= fuse_copy_fill(cs
);
979 void *mapaddr
= kmap_atomic(page
);
980 void *buf
= mapaddr
+ offset
;
981 offset
+= fuse_copy_do(cs
, &buf
, &count
);
982 kunmap_atomic(mapaddr
);
984 offset
+= fuse_copy_do(cs
, NULL
, &count
);
986 if (page
&& !cs
->write
)
987 flush_dcache_page(page
);
991 /* Copy pages in the request to/from userspace buffer */
992 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
996 struct fuse_req
*req
= cs
->req
;
998 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
1000 unsigned offset
= req
->page_descs
[i
].offset
;
1001 unsigned count
= min(nbytes
, req
->page_descs
[i
].length
);
1003 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
1013 /* Copy a single argument in the request to/from userspace buffer */
1014 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
1018 int err
= fuse_copy_fill(cs
);
1022 fuse_copy_do(cs
, &val
, &size
);
1027 /* Copy request arguments to/from userspace buffer */
1028 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
1029 unsigned argpages
, struct fuse_arg
*args
,
1035 for (i
= 0; !err
&& i
< numargs
; i
++) {
1036 struct fuse_arg
*arg
= &args
[i
];
1037 if (i
== numargs
- 1 && argpages
)
1038 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
1040 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
1045 static int forget_pending(struct fuse_iqueue
*fiq
)
1047 return fiq
->forget_list_head
.next
!= NULL
;
1050 static int request_pending(struct fuse_iqueue
*fiq
)
1052 return !list_empty(&fiq
->pending
) || !list_empty(&fiq
->interrupts
) ||
1053 forget_pending(fiq
);
1057 * Transfer an interrupt request to userspace
1059 * Unlike other requests this is assembled on demand, without a need
1060 * to allocate a separate fuse_req structure.
1062 * Called with fiq->waitq.lock held, releases it
1064 static int fuse_read_interrupt(struct fuse_iqueue
*fiq
,
1065 struct fuse_copy_state
*cs
,
1066 size_t nbytes
, struct fuse_req
*req
)
1067 __releases(fiq
->waitq
.lock
)
1069 struct fuse_in_header ih
;
1070 struct fuse_interrupt_in arg
;
1071 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
1074 list_del_init(&req
->intr_entry
);
1075 req
->intr_unique
= fuse_get_unique(fiq
);
1076 memset(&ih
, 0, sizeof(ih
));
1077 memset(&arg
, 0, sizeof(arg
));
1079 ih
.opcode
= FUSE_INTERRUPT
;
1080 ih
.unique
= req
->intr_unique
;
1081 arg
.unique
= req
->in
.h
.unique
;
1083 spin_unlock(&fiq
->waitq
.lock
);
1084 if (nbytes
< reqsize
)
1087 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1089 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1090 fuse_copy_finish(cs
);
1092 return err
? err
: reqsize
;
1095 static struct fuse_forget_link
*dequeue_forget(struct fuse_iqueue
*fiq
,
1099 struct fuse_forget_link
*head
= fiq
->forget_list_head
.next
;
1100 struct fuse_forget_link
**newhead
= &head
;
1103 for (count
= 0; *newhead
!= NULL
&& count
< max
; count
++)
1104 newhead
= &(*newhead
)->next
;
1106 fiq
->forget_list_head
.next
= *newhead
;
1108 if (fiq
->forget_list_head
.next
== NULL
)
1109 fiq
->forget_list_tail
= &fiq
->forget_list_head
;
1117 static int fuse_read_single_forget(struct fuse_iqueue
*fiq
,
1118 struct fuse_copy_state
*cs
,
1120 __releases(fiq
->waitq
.lock
)
1123 struct fuse_forget_link
*forget
= dequeue_forget(fiq
, 1, NULL
);
1124 struct fuse_forget_in arg
= {
1125 .nlookup
= forget
->forget_one
.nlookup
,
1127 struct fuse_in_header ih
= {
1128 .opcode
= FUSE_FORGET
,
1129 .nodeid
= forget
->forget_one
.nodeid
,
1130 .unique
= fuse_get_unique(fiq
),
1131 .len
= sizeof(ih
) + sizeof(arg
),
1134 spin_unlock(&fiq
->waitq
.lock
);
1136 if (nbytes
< ih
.len
)
1139 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1141 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1142 fuse_copy_finish(cs
);
1150 static int fuse_read_batch_forget(struct fuse_iqueue
*fiq
,
1151 struct fuse_copy_state
*cs
, size_t nbytes
)
1152 __releases(fiq
->waitq
.lock
)
1155 unsigned max_forgets
;
1157 struct fuse_forget_link
*head
;
1158 struct fuse_batch_forget_in arg
= { .count
= 0 };
1159 struct fuse_in_header ih
= {
1160 .opcode
= FUSE_BATCH_FORGET
,
1161 .unique
= fuse_get_unique(fiq
),
1162 .len
= sizeof(ih
) + sizeof(arg
),
1165 if (nbytes
< ih
.len
) {
1166 spin_unlock(&fiq
->waitq
.lock
);
1170 max_forgets
= (nbytes
- ih
.len
) / sizeof(struct fuse_forget_one
);
1171 head
= dequeue_forget(fiq
, max_forgets
, &count
);
1172 spin_unlock(&fiq
->waitq
.lock
);
1175 ih
.len
+= count
* sizeof(struct fuse_forget_one
);
1176 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1178 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1181 struct fuse_forget_link
*forget
= head
;
1184 err
= fuse_copy_one(cs
, &forget
->forget_one
,
1185 sizeof(forget
->forget_one
));
1187 head
= forget
->next
;
1191 fuse_copy_finish(cs
);
1199 static int fuse_read_forget(struct fuse_conn
*fc
, struct fuse_iqueue
*fiq
,
1200 struct fuse_copy_state
*cs
,
1202 __releases(fiq
->waitq
.lock
)
1204 if (fc
->minor
< 16 || fiq
->forget_list_head
.next
->next
== NULL
)
1205 return fuse_read_single_forget(fiq
, cs
, nbytes
);
1207 return fuse_read_batch_forget(fiq
, cs
, nbytes
);
1211 * Read a single request into the userspace filesystem's buffer. This
1212 * function waits until a request is available, then removes it from
1213 * the pending list and copies request data to userspace buffer. If
1214 * no reply is needed (FORGET) or request has been aborted or there
1215 * was an error during the copying then it's finished by calling
1216 * request_end(). Otherwise add it to the processing list, and set
1219 static ssize_t
fuse_dev_do_read(struct fuse_dev
*fud
, struct file
*file
,
1220 struct fuse_copy_state
*cs
, size_t nbytes
)
1223 struct fuse_conn
*fc
= fud
->fc
;
1224 struct fuse_iqueue
*fiq
= &fc
->iq
;
1225 struct fuse_pqueue
*fpq
= &fud
->pq
;
1226 struct fuse_req
*req
;
1230 if (task_active_pid_ns(current
) != fc
->pid_ns
||
1231 current_user_ns() != fc
->user_ns
)
1235 spin_lock(&fiq
->waitq
.lock
);
1237 if ((file
->f_flags
& O_NONBLOCK
) && fiq
->connected
&&
1238 !request_pending(fiq
))
1241 err
= wait_event_interruptible_exclusive_locked(fiq
->waitq
,
1242 !fiq
->connected
|| request_pending(fiq
));
1247 if (!fiq
->connected
)
1250 if (!list_empty(&fiq
->interrupts
)) {
1251 req
= list_entry(fiq
->interrupts
.next
, struct fuse_req
,
1253 return fuse_read_interrupt(fiq
, cs
, nbytes
, req
);
1256 if (forget_pending(fiq
)) {
1257 if (list_empty(&fiq
->pending
) || fiq
->forget_batch
-- > 0)
1258 return fuse_read_forget(fc
, fiq
, cs
, nbytes
);
1260 if (fiq
->forget_batch
<= -8)
1261 fiq
->forget_batch
= 16;
1264 req
= list_entry(fiq
->pending
.next
, struct fuse_req
, list
);
1265 clear_bit(FR_PENDING
, &req
->flags
);
1266 list_del_init(&req
->list
);
1267 spin_unlock(&fiq
->waitq
.lock
);
1270 reqsize
= in
->h
.len
;
1271 /* If request is too large, reply with an error and restart the read */
1272 if (nbytes
< reqsize
) {
1273 req
->out
.h
.error
= -EIO
;
1274 /* SETXATTR is special, since it may contain too large data */
1275 if (in
->h
.opcode
== FUSE_SETXATTR
)
1276 req
->out
.h
.error
= -E2BIG
;
1277 request_end(fc
, req
);
1280 spin_lock(&fpq
->lock
);
1281 list_add(&req
->list
, &fpq
->io
);
1282 spin_unlock(&fpq
->lock
);
1284 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1286 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1287 (struct fuse_arg
*) in
->args
, 0);
1288 fuse_copy_finish(cs
);
1289 spin_lock(&fpq
->lock
);
1290 clear_bit(FR_LOCKED
, &req
->flags
);
1291 if (!fpq
->connected
) {
1296 req
->out
.h
.error
= -EIO
;
1299 if (!test_bit(FR_ISREPLY
, &req
->flags
)) {
1303 list_move_tail(&req
->list
, &fpq
->processing
);
1304 spin_unlock(&fpq
->lock
);
1305 set_bit(FR_SENT
, &req
->flags
);
1306 /* matches barrier in request_wait_answer() */
1307 smp_mb__after_atomic();
1308 if (test_bit(FR_INTERRUPTED
, &req
->flags
))
1309 queue_interrupt(fiq
, req
);
1314 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1315 list_del_init(&req
->list
);
1316 spin_unlock(&fpq
->lock
);
1317 request_end(fc
, req
);
1321 spin_unlock(&fiq
->waitq
.lock
);
1325 static int fuse_dev_open(struct inode
*inode
, struct file
*file
)
1328 * The fuse device's file's private_data is used to hold
1329 * the fuse_conn(ection) when it is mounted, and is used to
1330 * keep track of whether the file has been mounted already.
1332 file
->private_data
= NULL
;
1336 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, struct iov_iter
*to
)
1338 struct fuse_copy_state cs
;
1339 struct file
*file
= iocb
->ki_filp
;
1340 struct fuse_dev
*fud
= fuse_get_dev(file
);
1345 if (!iter_is_iovec(to
))
1348 fuse_copy_init(&cs
, 1, to
);
1350 return fuse_dev_do_read(fud
, file
, &cs
, iov_iter_count(to
));
1353 static ssize_t
fuse_dev_splice_read(struct file
*in
, loff_t
*ppos
,
1354 struct pipe_inode_info
*pipe
,
1355 size_t len
, unsigned int flags
)
1359 struct pipe_buffer
*bufs
;
1360 struct fuse_copy_state cs
;
1361 struct fuse_dev
*fud
= fuse_get_dev(in
);
1366 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1370 fuse_copy_init(&cs
, 1, NULL
);
1373 ret
= fuse_dev_do_read(fud
, in
, &cs
, len
);
1377 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1382 for (ret
= total
= 0; page_nr
< cs
.nr_segs
; total
+= ret
) {
1384 * Need to be careful about this. Having buf->ops in module
1385 * code can Oops if the buffer persists after module unload.
1387 bufs
[page_nr
].ops
= &nosteal_pipe_buf_ops
;
1388 bufs
[page_nr
].flags
= 0;
1389 ret
= add_to_pipe(pipe
, &bufs
[page_nr
++]);
1390 if (unlikely(ret
< 0))
1396 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1397 put_page(bufs
[page_nr
].page
);
1403 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1404 struct fuse_copy_state
*cs
)
1406 struct fuse_notify_poll_wakeup_out outarg
;
1409 if (size
!= sizeof(outarg
))
1412 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1416 fuse_copy_finish(cs
);
1417 return fuse_notify_poll_wakeup(fc
, &outarg
);
1420 fuse_copy_finish(cs
);
1424 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1425 struct fuse_copy_state
*cs
)
1427 struct fuse_notify_inval_inode_out outarg
;
1430 if (size
!= sizeof(outarg
))
1433 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1436 fuse_copy_finish(cs
);
1438 down_read(&fc
->killsb
);
1441 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1442 outarg
.off
, outarg
.len
);
1444 up_read(&fc
->killsb
);
1448 fuse_copy_finish(cs
);
1452 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1453 struct fuse_copy_state
*cs
)
1455 struct fuse_notify_inval_entry_out outarg
;
1460 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1465 if (size
< sizeof(outarg
))
1468 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1472 err
= -ENAMETOOLONG
;
1473 if (outarg
.namelen
> FUSE_NAME_MAX
)
1477 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1481 name
.len
= outarg
.namelen
;
1482 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1485 fuse_copy_finish(cs
);
1486 buf
[outarg
.namelen
] = 0;
1488 down_read(&fc
->killsb
);
1491 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, 0, &name
);
1492 up_read(&fc
->killsb
);
1498 fuse_copy_finish(cs
);
1502 static int fuse_notify_delete(struct fuse_conn
*fc
, unsigned int size
,
1503 struct fuse_copy_state
*cs
)
1505 struct fuse_notify_delete_out outarg
;
1510 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1515 if (size
< sizeof(outarg
))
1518 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1522 err
= -ENAMETOOLONG
;
1523 if (outarg
.namelen
> FUSE_NAME_MAX
)
1527 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1531 name
.len
= outarg
.namelen
;
1532 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1535 fuse_copy_finish(cs
);
1536 buf
[outarg
.namelen
] = 0;
1538 down_read(&fc
->killsb
);
1541 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
,
1542 outarg
.child
, &name
);
1543 up_read(&fc
->killsb
);
1549 fuse_copy_finish(cs
);
1553 static int fuse_notify_store(struct fuse_conn
*fc
, unsigned int size
,
1554 struct fuse_copy_state
*cs
)
1556 struct fuse_notify_store_out outarg
;
1557 struct inode
*inode
;
1558 struct address_space
*mapping
;
1562 unsigned int offset
;
1568 if (size
< sizeof(outarg
))
1571 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1576 if (size
- sizeof(outarg
) != outarg
.size
)
1579 nodeid
= outarg
.nodeid
;
1581 down_read(&fc
->killsb
);
1587 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1591 mapping
= inode
->i_mapping
;
1592 index
= outarg
.offset
>> PAGE_SHIFT
;
1593 offset
= outarg
.offset
& ~PAGE_MASK
;
1594 file_size
= i_size_read(inode
);
1595 end
= outarg
.offset
+ outarg
.size
;
1596 if (end
> file_size
) {
1598 fuse_write_update_size(inode
, file_size
);
1604 unsigned int this_num
;
1607 page
= find_or_create_page(mapping
, index
,
1608 mapping_gfp_mask(mapping
));
1612 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1613 err
= fuse_copy_page(cs
, &page
, offset
, this_num
, 0);
1614 if (!err
&& offset
== 0 &&
1615 (this_num
== PAGE_SIZE
|| file_size
== end
))
1616 SetPageUptodate(page
);
1633 up_read(&fc
->killsb
);
1635 fuse_copy_finish(cs
);
1639 static void fuse_retrieve_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1641 release_pages(req
->pages
, req
->num_pages
, false);
1644 static int fuse_retrieve(struct fuse_conn
*fc
, struct inode
*inode
,
1645 struct fuse_notify_retrieve_out
*outarg
)
1648 struct address_space
*mapping
= inode
->i_mapping
;
1649 struct fuse_req
*req
;
1653 unsigned int offset
;
1654 size_t total_len
= 0;
1657 offset
= outarg
->offset
& ~PAGE_MASK
;
1658 file_size
= i_size_read(inode
);
1661 if (outarg
->offset
> file_size
)
1663 else if (outarg
->offset
+ num
> file_size
)
1664 num
= file_size
- outarg
->offset
;
1666 num_pages
= (num
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1667 num_pages
= min(num_pages
, FUSE_MAX_PAGES_PER_REQ
);
1669 req
= fuse_get_req(fc
, num_pages
);
1671 return PTR_ERR(req
);
1673 req
->in
.h
.opcode
= FUSE_NOTIFY_REPLY
;
1674 req
->in
.h
.nodeid
= outarg
->nodeid
;
1675 req
->in
.numargs
= 2;
1676 req
->in
.argpages
= 1;
1677 req
->page_descs
[0].offset
= offset
;
1678 req
->end
= fuse_retrieve_end
;
1680 index
= outarg
->offset
>> PAGE_SHIFT
;
1682 while (num
&& req
->num_pages
< num_pages
) {
1684 unsigned int this_num
;
1686 page
= find_get_page(mapping
, index
);
1690 this_num
= min_t(unsigned, num
, PAGE_SIZE
- offset
);
1691 req
->pages
[req
->num_pages
] = page
;
1692 req
->page_descs
[req
->num_pages
].length
= this_num
;
1697 total_len
+= this_num
;
1700 req
->misc
.retrieve_in
.offset
= outarg
->offset
;
1701 req
->misc
.retrieve_in
.size
= total_len
;
1702 req
->in
.args
[0].size
= sizeof(req
->misc
.retrieve_in
);
1703 req
->in
.args
[0].value
= &req
->misc
.retrieve_in
;
1704 req
->in
.args
[1].size
= total_len
;
1706 err
= fuse_request_send_notify_reply(fc
, req
, outarg
->notify_unique
);
1708 fuse_retrieve_end(fc
, req
);
1713 static int fuse_notify_retrieve(struct fuse_conn
*fc
, unsigned int size
,
1714 struct fuse_copy_state
*cs
)
1716 struct fuse_notify_retrieve_out outarg
;
1717 struct inode
*inode
;
1721 if (size
!= sizeof(outarg
))
1724 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1728 fuse_copy_finish(cs
);
1730 down_read(&fc
->killsb
);
1733 u64 nodeid
= outarg
.nodeid
;
1735 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1737 err
= fuse_retrieve(fc
, inode
, &outarg
);
1741 up_read(&fc
->killsb
);
1746 fuse_copy_finish(cs
);
1750 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1751 unsigned int size
, struct fuse_copy_state
*cs
)
1753 /* Don't try to move pages (yet) */
1757 case FUSE_NOTIFY_POLL
:
1758 return fuse_notify_poll(fc
, size
, cs
);
1760 case FUSE_NOTIFY_INVAL_INODE
:
1761 return fuse_notify_inval_inode(fc
, size
, cs
);
1763 case FUSE_NOTIFY_INVAL_ENTRY
:
1764 return fuse_notify_inval_entry(fc
, size
, cs
);
1766 case FUSE_NOTIFY_STORE
:
1767 return fuse_notify_store(fc
, size
, cs
);
1769 case FUSE_NOTIFY_RETRIEVE
:
1770 return fuse_notify_retrieve(fc
, size
, cs
);
1772 case FUSE_NOTIFY_DELETE
:
1773 return fuse_notify_delete(fc
, size
, cs
);
1776 fuse_copy_finish(cs
);
1781 /* Look up request on processing list by unique ID */
1782 static struct fuse_req
*request_find(struct fuse_pqueue
*fpq
, u64 unique
)
1784 struct fuse_req
*req
;
1786 list_for_each_entry(req
, &fpq
->processing
, list
) {
1787 if (req
->in
.h
.unique
== unique
|| req
->intr_unique
== unique
)
1793 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1796 unsigned reqsize
= sizeof(struct fuse_out_header
);
1799 return nbytes
!= reqsize
? -EINVAL
: 0;
1801 reqsize
+= len_args(out
->numargs
, out
->args
);
1803 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1805 else if (reqsize
> nbytes
) {
1806 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1807 unsigned diffsize
= reqsize
- nbytes
;
1808 if (diffsize
> lastarg
->size
)
1810 lastarg
->size
-= diffsize
;
1812 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1817 * Write a single reply to a request. First the header is copied from
1818 * the write buffer. The request is then searched on the processing
1819 * list by the unique ID found in the header. If found, then remove
1820 * it from the list and copy the rest of the buffer to the request.
1821 * The request is finished by calling request_end()
1823 static ssize_t
fuse_dev_do_write(struct fuse_dev
*fud
,
1824 struct fuse_copy_state
*cs
, size_t nbytes
)
1827 struct fuse_conn
*fc
= fud
->fc
;
1828 struct fuse_pqueue
*fpq
= &fud
->pq
;
1829 struct fuse_req
*req
;
1830 struct fuse_out_header oh
;
1832 if (task_active_pid_ns(current
) != fc
->pid_ns
||
1833 current_user_ns() != fc
->user_ns
)
1836 if (nbytes
< sizeof(struct fuse_out_header
))
1839 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1844 if (oh
.len
!= nbytes
)
1848 * Zero oh.unique indicates unsolicited notification message
1849 * and error contains notification code.
1852 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1853 return err
? err
: nbytes
;
1857 if (oh
.error
<= -1000 || oh
.error
> 0)
1860 spin_lock(&fpq
->lock
);
1862 if (!fpq
->connected
)
1865 req
= request_find(fpq
, oh
.unique
);
1869 /* Is it an interrupt reply? */
1870 if (req
->intr_unique
== oh
.unique
) {
1871 spin_unlock(&fpq
->lock
);
1874 if (nbytes
!= sizeof(struct fuse_out_header
))
1877 if (oh
.error
== -ENOSYS
)
1878 fc
->no_interrupt
= 1;
1879 else if (oh
.error
== -EAGAIN
)
1880 queue_interrupt(&fc
->iq
, req
);
1882 fuse_copy_finish(cs
);
1886 clear_bit(FR_SENT
, &req
->flags
);
1887 list_move(&req
->list
, &fpq
->io
);
1889 set_bit(FR_LOCKED
, &req
->flags
);
1890 spin_unlock(&fpq
->lock
);
1892 if (!req
->out
.page_replace
)
1895 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1896 fuse_copy_finish(cs
);
1898 spin_lock(&fpq
->lock
);
1899 clear_bit(FR_LOCKED
, &req
->flags
);
1900 if (!fpq
->connected
)
1903 req
->out
.h
.error
= -EIO
;
1904 if (!test_bit(FR_PRIVATE
, &req
->flags
))
1905 list_del_init(&req
->list
);
1906 spin_unlock(&fpq
->lock
);
1908 request_end(fc
, req
);
1910 return err
? err
: nbytes
;
1913 spin_unlock(&fpq
->lock
);
1915 fuse_copy_finish(cs
);
1919 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, struct iov_iter
*from
)
1921 struct fuse_copy_state cs
;
1922 struct fuse_dev
*fud
= fuse_get_dev(iocb
->ki_filp
);
1927 if (!iter_is_iovec(from
))
1930 fuse_copy_init(&cs
, 0, from
);
1932 return fuse_dev_do_write(fud
, &cs
, iov_iter_count(from
));
1935 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1936 struct file
*out
, loff_t
*ppos
,
1937 size_t len
, unsigned int flags
)
1941 struct pipe_buffer
*bufs
;
1942 struct fuse_copy_state cs
;
1943 struct fuse_dev
*fud
;
1947 fud
= fuse_get_dev(out
);
1951 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1958 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
1959 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
1969 struct pipe_buffer
*ibuf
;
1970 struct pipe_buffer
*obuf
;
1972 BUG_ON(nbuf
>= pipe
->buffers
);
1973 BUG_ON(!pipe
->nrbufs
);
1974 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
1977 if (rem
>= ibuf
->len
) {
1980 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
1983 pipe_buf_get(pipe
, ibuf
);
1985 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1987 ibuf
->offset
+= obuf
->len
;
1988 ibuf
->len
-= obuf
->len
;
1995 fuse_copy_init(&cs
, 0, NULL
);
2000 if (flags
& SPLICE_F_MOVE
)
2003 ret
= fuse_dev_do_write(fud
, &cs
, len
);
2005 for (idx
= 0; idx
< nbuf
; idx
++)
2006 pipe_buf_release(pipe
, &bufs
[idx
]);
2013 static unsigned fuse_dev_poll(struct file
*file
, poll_table
*wait
)
2015 unsigned mask
= POLLOUT
| POLLWRNORM
;
2016 struct fuse_iqueue
*fiq
;
2017 struct fuse_dev
*fud
= fuse_get_dev(file
);
2023 poll_wait(file
, &fiq
->waitq
, wait
);
2025 spin_lock(&fiq
->waitq
.lock
);
2026 if (!fiq
->connected
)
2028 else if (request_pending(fiq
))
2029 mask
|= POLLIN
| POLLRDNORM
;
2030 spin_unlock(&fiq
->waitq
.lock
);
2036 * Abort all requests on the given list (pending or processing)
2038 * This function releases and reacquires fc->lock
2040 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
2042 while (!list_empty(head
)) {
2043 struct fuse_req
*req
;
2044 req
= list_entry(head
->next
, struct fuse_req
, list
);
2045 req
->out
.h
.error
= -ECONNABORTED
;
2046 clear_bit(FR_SENT
, &req
->flags
);
2047 list_del_init(&req
->list
);
2048 request_end(fc
, req
);
2052 static void end_polls(struct fuse_conn
*fc
)
2056 p
= rb_first(&fc
->polled_files
);
2059 struct fuse_file
*ff
;
2060 ff
= rb_entry(p
, struct fuse_file
, polled_node
);
2061 wake_up_interruptible_all(&ff
->poll_wait
);
2068 * Abort all requests.
2070 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2073 * The same effect is usually achievable through killing the filesystem daemon
2074 * and all users of the filesystem. The exception is the combination of an
2075 * asynchronous request and the tricky deadlock (see
2076 * Documentation/filesystems/fuse.txt).
2078 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2079 * requests, they should be finished off immediately. Locked requests will be
2080 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2081 * requests. It is possible that some request will finish before we can. This
2082 * is OK, the request will in that case be removed from the list before we touch
2085 void fuse_abort_conn(struct fuse_conn
*fc
)
2087 struct fuse_iqueue
*fiq
= &fc
->iq
;
2089 spin_lock(&fc
->lock
);
2090 if (fc
->connected
) {
2091 struct fuse_dev
*fud
;
2092 struct fuse_req
*req
, *next
;
2098 fuse_set_initialized(fc
);
2099 list_for_each_entry(fud
, &fc
->devices
, entry
) {
2100 struct fuse_pqueue
*fpq
= &fud
->pq
;
2102 spin_lock(&fpq
->lock
);
2104 list_for_each_entry_safe(req
, next
, &fpq
->io
, list
) {
2105 req
->out
.h
.error
= -ECONNABORTED
;
2106 spin_lock(&req
->waitq
.lock
);
2107 set_bit(FR_ABORTED
, &req
->flags
);
2108 if (!test_bit(FR_LOCKED
, &req
->flags
)) {
2109 set_bit(FR_PRIVATE
, &req
->flags
);
2110 list_move(&req
->list
, &to_end1
);
2112 spin_unlock(&req
->waitq
.lock
);
2114 list_splice_init(&fpq
->processing
, &to_end2
);
2115 spin_unlock(&fpq
->lock
);
2117 fc
->max_background
= UINT_MAX
;
2120 spin_lock(&fiq
->waitq
.lock
);
2122 list_splice_init(&fiq
->pending
, &to_end2
);
2123 list_for_each_entry(req
, &to_end2
, list
)
2124 clear_bit(FR_PENDING
, &req
->flags
);
2125 while (forget_pending(fiq
))
2126 kfree(dequeue_forget(fiq
, 1, NULL
));
2127 wake_up_all_locked(&fiq
->waitq
);
2128 spin_unlock(&fiq
->waitq
.lock
);
2129 kill_fasync(&fiq
->fasync
, SIGIO
, POLL_IN
);
2131 wake_up_all(&fc
->blocked_waitq
);
2132 spin_unlock(&fc
->lock
);
2134 while (!list_empty(&to_end1
)) {
2135 req
= list_first_entry(&to_end1
, struct fuse_req
, list
);
2136 __fuse_get_request(req
);
2137 list_del_init(&req
->list
);
2138 request_end(fc
, req
);
2140 end_requests(fc
, &to_end2
);
2142 spin_unlock(&fc
->lock
);
2145 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
2147 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
2149 struct fuse_dev
*fud
= fuse_get_dev(file
);
2152 struct fuse_conn
*fc
= fud
->fc
;
2153 struct fuse_pqueue
*fpq
= &fud
->pq
;
2155 WARN_ON(!list_empty(&fpq
->io
));
2156 end_requests(fc
, &fpq
->processing
);
2157 /* Are we the last open device? */
2158 if (atomic_dec_and_test(&fc
->dev_count
)) {
2159 WARN_ON(fc
->iq
.fasync
!= NULL
);
2160 fuse_abort_conn(fc
);
2166 EXPORT_SYMBOL_GPL(fuse_dev_release
);
2168 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
2170 struct fuse_dev
*fud
= fuse_get_dev(file
);
2175 /* No locking - fasync_helper does its own locking */
2176 return fasync_helper(fd
, file
, on
, &fud
->fc
->iq
.fasync
);
2179 static int fuse_device_clone(struct fuse_conn
*fc
, struct file
*new)
2181 struct fuse_dev
*fud
;
2183 if (new->private_data
)
2186 fud
= fuse_dev_alloc(fc
);
2190 new->private_data
= fud
;
2191 atomic_inc(&fc
->dev_count
);
2196 static long fuse_dev_ioctl(struct file
*file
, unsigned int cmd
,
2201 if (cmd
== FUSE_DEV_IOC_CLONE
) {
2205 if (!get_user(oldfd
, (__u32 __user
*) arg
)) {
2206 struct file
*old
= fget(oldfd
);
2210 struct fuse_dev
*fud
= NULL
;
2213 * Check against file->f_op because CUSE
2214 * uses the same ioctl handler.
2216 if (old
->f_op
== file
->f_op
&&
2217 old
->f_cred
->user_ns
== file
->f_cred
->user_ns
)
2218 fud
= fuse_get_dev(old
);
2221 mutex_lock(&fuse_mutex
);
2222 err
= fuse_device_clone(fud
->fc
, file
);
2223 mutex_unlock(&fuse_mutex
);
2232 const struct file_operations fuse_dev_operations
= {
2233 .owner
= THIS_MODULE
,
2234 .open
= fuse_dev_open
,
2235 .llseek
= no_llseek
,
2236 .read_iter
= fuse_dev_read
,
2237 .splice_read
= fuse_dev_splice_read
,
2238 .write_iter
= fuse_dev_write
,
2239 .splice_write
= fuse_dev_splice_write
,
2240 .poll
= fuse_dev_poll
,
2241 .release
= fuse_dev_release
,
2242 .fasync
= fuse_dev_fasync
,
2243 .unlocked_ioctl
= fuse_dev_ioctl
,
2244 .compat_ioctl
= fuse_dev_ioctl
,
2246 EXPORT_SYMBOL_GPL(fuse_dev_operations
);
2248 static struct miscdevice fuse_miscdevice
= {
2249 .minor
= FUSE_MINOR
,
2251 .fops
= &fuse_dev_operations
,
2254 int __init
fuse_dev_init(void)
2257 fuse_req_cachep
= kmem_cache_create("fuse_request",
2258 sizeof(struct fuse_req
),
2260 if (!fuse_req_cachep
)
2263 err
= misc_register(&fuse_miscdevice
);
2265 goto out_cache_clean
;
2270 kmem_cache_destroy(fuse_req_cachep
);
2275 void fuse_dev_cleanup(void)
2277 misc_deregister(&fuse_miscdevice
);
2278 kmem_cache_destroy(fuse_req_cachep
);