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/aio.h>
24 MODULE_ALIAS_MISCDEV(FUSE_MINOR
);
25 MODULE_ALIAS("devname:fuse");
27 static struct kmem_cache
*fuse_req_cachep
;
29 static struct fuse_conn
*fuse_get_conn(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 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
;
54 static struct fuse_req
*__fuse_request_alloc(unsigned npages
, gfp_t flags
)
56 struct fuse_req
*req
= kmem_cache_alloc(fuse_req_cachep
, flags
);
59 struct fuse_page_desc
*page_descs
;
61 if (npages
<= FUSE_REQ_INLINE_PAGES
) {
62 pages
= req
->inline_pages
;
63 page_descs
= req
->inline_page_descs
;
65 pages
= kmalloc(sizeof(struct page
*) * npages
, flags
);
66 page_descs
= kmalloc(sizeof(struct fuse_page_desc
) *
70 if (!pages
|| !page_descs
) {
73 kmem_cache_free(fuse_req_cachep
, req
);
77 fuse_request_init(req
, pages
, page_descs
, npages
);
82 struct fuse_req
*fuse_request_alloc(unsigned npages
)
84 return __fuse_request_alloc(npages
, GFP_KERNEL
);
86 EXPORT_SYMBOL_GPL(fuse_request_alloc
);
88 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
)
90 return __fuse_request_alloc(npages
, GFP_NOFS
);
93 void fuse_request_free(struct fuse_req
*req
)
95 if (req
->pages
!= req
->inline_pages
) {
97 kfree(req
->page_descs
);
99 kmem_cache_free(fuse_req_cachep
, req
);
102 static void block_sigs(sigset_t
*oldset
)
106 siginitsetinv(&mask
, sigmask(SIGKILL
));
107 sigprocmask(SIG_BLOCK
, &mask
, oldset
);
110 static void restore_sigs(sigset_t
*oldset
)
112 sigprocmask(SIG_SETMASK
, oldset
, NULL
);
115 void __fuse_get_request(struct fuse_req
*req
)
117 atomic_inc(&req
->count
);
120 /* Must be called with > 1 refcount */
121 static void __fuse_put_request(struct fuse_req
*req
)
123 BUG_ON(atomic_read(&req
->count
) < 2);
124 atomic_dec(&req
->count
);
127 static void fuse_req_init_context(struct fuse_req
*req
)
129 req
->in
.h
.uid
= from_kuid_munged(&init_user_ns
, current_fsuid());
130 req
->in
.h
.gid
= from_kgid_munged(&init_user_ns
, current_fsgid());
131 req
->in
.h
.pid
= current
->pid
;
134 void fuse_set_initialized(struct fuse_conn
*fc
)
136 /* Make sure stores before this are seen on another CPU */
141 static bool fuse_block_alloc(struct fuse_conn
*fc
, bool for_background
)
143 return !fc
->initialized
|| (for_background
&& fc
->blocked
);
146 static struct fuse_req
*__fuse_get_req(struct fuse_conn
*fc
, unsigned npages
,
149 struct fuse_req
*req
;
151 atomic_inc(&fc
->num_waiting
);
153 if (fuse_block_alloc(fc
, for_background
)) {
158 intr
= wait_event_interruptible_exclusive(fc
->blocked_waitq
,
159 !fuse_block_alloc(fc
, for_background
));
160 restore_sigs(&oldset
);
165 /* Matches smp_wmb() in fuse_set_initialized() */
172 req
= fuse_request_alloc(npages
);
176 wake_up(&fc
->blocked_waitq
);
180 fuse_req_init_context(req
);
182 req
->background
= for_background
;
186 atomic_dec(&fc
->num_waiting
);
190 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
)
192 return __fuse_get_req(fc
, npages
, false);
194 EXPORT_SYMBOL_GPL(fuse_get_req
);
196 struct fuse_req
*fuse_get_req_for_background(struct fuse_conn
*fc
,
199 return __fuse_get_req(fc
, npages
, true);
201 EXPORT_SYMBOL_GPL(fuse_get_req_for_background
);
204 * Return request in fuse_file->reserved_req. However that may
205 * currently be in use. If that is the case, wait for it to become
208 static struct fuse_req
*get_reserved_req(struct fuse_conn
*fc
,
211 struct fuse_req
*req
= NULL
;
212 struct fuse_file
*ff
= file
->private_data
;
215 wait_event(fc
->reserved_req_waitq
, ff
->reserved_req
);
216 spin_lock(&fc
->lock
);
217 if (ff
->reserved_req
) {
218 req
= ff
->reserved_req
;
219 ff
->reserved_req
= NULL
;
220 req
->stolen_file
= get_file(file
);
222 spin_unlock(&fc
->lock
);
229 * Put stolen request back into fuse_file->reserved_req
231 static void put_reserved_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
233 struct file
*file
= req
->stolen_file
;
234 struct fuse_file
*ff
= file
->private_data
;
236 spin_lock(&fc
->lock
);
237 fuse_request_init(req
, req
->pages
, req
->page_descs
, req
->max_pages
);
238 BUG_ON(ff
->reserved_req
);
239 ff
->reserved_req
= req
;
240 wake_up_all(&fc
->reserved_req_waitq
);
241 spin_unlock(&fc
->lock
);
246 * Gets a requests for a file operation, always succeeds
248 * This is used for sending the FLUSH request, which must get to
249 * userspace, due to POSIX locks which may need to be unlocked.
251 * If allocation fails due to OOM, use the reserved request in
254 * This is very unlikely to deadlock accidentally, since the
255 * filesystem should not have it's own file open. If deadlock is
256 * intentional, it can still be broken by "aborting" the filesystem.
258 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
261 struct fuse_req
*req
;
263 atomic_inc(&fc
->num_waiting
);
264 wait_event(fc
->blocked_waitq
, fc
->initialized
);
265 /* Matches smp_wmb() in fuse_set_initialized() */
267 req
= fuse_request_alloc(0);
269 req
= get_reserved_req(fc
, file
);
271 fuse_req_init_context(req
);
277 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
279 if (atomic_dec_and_test(&req
->count
)) {
280 if (unlikely(req
->background
)) {
282 * We get here in the unlikely case that a background
283 * request was allocated but not sent
285 spin_lock(&fc
->lock
);
287 wake_up(&fc
->blocked_waitq
);
288 spin_unlock(&fc
->lock
);
292 atomic_dec(&fc
->num_waiting
);
294 if (req
->stolen_file
)
295 put_reserved_req(fc
, req
);
297 fuse_request_free(req
);
300 EXPORT_SYMBOL_GPL(fuse_put_request
);
302 static unsigned len_args(unsigned numargs
, struct fuse_arg
*args
)
307 for (i
= 0; i
< numargs
; i
++)
308 nbytes
+= args
[i
].size
;
313 static u64
fuse_get_unique(struct fuse_conn
*fc
)
316 /* zero is special */
323 static void queue_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
325 req
->in
.h
.len
= sizeof(struct fuse_in_header
) +
326 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
327 list_add_tail(&req
->list
, &fc
->pending
);
328 req
->state
= FUSE_REQ_PENDING
;
331 atomic_inc(&fc
->num_waiting
);
334 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
337 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
338 u64 nodeid
, u64 nlookup
)
340 forget
->forget_one
.nodeid
= nodeid
;
341 forget
->forget_one
.nlookup
= nlookup
;
343 spin_lock(&fc
->lock
);
345 fc
->forget_list_tail
->next
= forget
;
346 fc
->forget_list_tail
= forget
;
348 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
352 spin_unlock(&fc
->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
;
361 req
= list_entry(fc
->bg_queue
.next
, struct fuse_req
, list
);
362 list_del(&req
->list
);
363 fc
->active_background
++;
364 req
->in
.h
.unique
= fuse_get_unique(fc
);
365 queue_request(fc
, req
);
370 * This function is called when a request is finished. Either a reply
371 * has arrived or it was aborted (and not yet sent) or some error
372 * occurred during communication with userspace, or the device file
373 * was closed. The requester thread is woken up (if still waiting),
374 * the 'end' callback is called if given, else the reference to the
375 * request is released
377 * Called with fc->lock, unlocks it
379 static void request_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
382 void (*end
) (struct fuse_conn
*, struct fuse_req
*) = req
->end
;
384 list_del(&req
->list
);
385 list_del(&req
->intr_entry
);
386 req
->state
= FUSE_REQ_FINISHED
;
387 if (req
->background
) {
390 if (fc
->num_background
== fc
->max_background
)
393 /* Wake up next waiter, if any */
394 if (!fc
->blocked
&& waitqueue_active(&fc
->blocked_waitq
))
395 wake_up(&fc
->blocked_waitq
);
397 if (fc
->num_background
== fc
->congestion_threshold
&&
398 fc
->connected
&& fc
->bdi_initialized
) {
399 clear_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
400 clear_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
402 fc
->num_background
--;
403 fc
->active_background
--;
406 spin_unlock(&fc
->lock
);
407 wake_up(&req
->waitq
);
410 fuse_put_request(fc
, req
);
413 static void wait_answer_interruptible(struct fuse_conn
*fc
,
414 struct fuse_req
*req
)
418 if (signal_pending(current
))
421 spin_unlock(&fc
->lock
);
422 wait_event_interruptible(req
->waitq
, req
->state
== FUSE_REQ_FINISHED
);
423 spin_lock(&fc
->lock
);
426 static void queue_interrupt(struct fuse_conn
*fc
, struct fuse_req
*req
)
428 list_add_tail(&req
->intr_entry
, &fc
->interrupts
);
430 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
433 static void request_wait_answer(struct fuse_conn
*fc
, struct fuse_req
*req
)
437 if (!fc
->no_interrupt
) {
438 /* Any signal may interrupt this */
439 wait_answer_interruptible(fc
, req
);
443 if (req
->state
== FUSE_REQ_FINISHED
)
446 req
->interrupted
= 1;
447 if (req
->state
== FUSE_REQ_SENT
)
448 queue_interrupt(fc
, req
);
454 /* Only fatal signals may interrupt this */
456 wait_answer_interruptible(fc
, req
);
457 restore_sigs(&oldset
);
461 if (req
->state
== FUSE_REQ_FINISHED
)
464 /* Request is not yet in userspace, bail out */
465 if (req
->state
== FUSE_REQ_PENDING
) {
466 list_del(&req
->list
);
467 __fuse_put_request(req
);
468 req
->out
.h
.error
= -EINTR
;
474 * Either request is already in userspace, or it was forced.
477 spin_unlock(&fc
->lock
);
478 wait_event(req
->waitq
, req
->state
== FUSE_REQ_FINISHED
);
479 spin_lock(&fc
->lock
);
485 BUG_ON(req
->state
!= FUSE_REQ_FINISHED
);
487 /* This is uninterruptible sleep, because data is
488 being copied to/from the buffers of req. During
489 locked state, there mustn't be any filesystem
490 operation (e.g. page fault), since that could lead
492 spin_unlock(&fc
->lock
);
493 wait_event(req
->waitq
, !req
->locked
);
494 spin_lock(&fc
->lock
);
498 static void __fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
500 BUG_ON(req
->background
);
501 spin_lock(&fc
->lock
);
503 req
->out
.h
.error
= -ENOTCONN
;
504 else if (fc
->conn_error
)
505 req
->out
.h
.error
= -ECONNREFUSED
;
507 req
->in
.h
.unique
= fuse_get_unique(fc
);
508 queue_request(fc
, req
);
509 /* acquire extra reference, since request is still needed
510 after request_end() */
511 __fuse_get_request(req
);
513 request_wait_answer(fc
, req
);
515 spin_unlock(&fc
->lock
);
518 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
521 __fuse_request_send(fc
, req
);
523 EXPORT_SYMBOL_GPL(fuse_request_send
);
525 static void fuse_adjust_compat(struct fuse_conn
*fc
, struct fuse_args
*args
)
527 if (fc
->minor
< 4 && args
->in
.h
.opcode
== FUSE_STATFS
)
528 args
->out
.args
[0].size
= FUSE_COMPAT_STATFS_SIZE
;
531 switch (args
->in
.h
.opcode
) {
538 args
->out
.args
[0].size
= FUSE_COMPAT_ENTRY_OUT_SIZE
;
542 args
->out
.args
[0].size
= FUSE_COMPAT_ATTR_OUT_SIZE
;
546 if (fc
->minor
< 12) {
547 switch (args
->in
.h
.opcode
) {
549 args
->in
.args
[0].size
= sizeof(struct fuse_open_in
);
552 args
->in
.args
[0].size
= FUSE_COMPAT_MKNOD_IN_SIZE
;
558 ssize_t
fuse_simple_request(struct fuse_conn
*fc
, struct fuse_args
*args
)
560 struct fuse_req
*req
;
563 req
= fuse_get_req(fc
, 0);
567 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
568 fuse_adjust_compat(fc
, args
);
570 req
->in
.h
.opcode
= args
->in
.h
.opcode
;
571 req
->in
.h
.nodeid
= args
->in
.h
.nodeid
;
572 req
->in
.numargs
= args
->in
.numargs
;
573 memcpy(req
->in
.args
, args
->in
.args
,
574 args
->in
.numargs
* sizeof(struct fuse_in_arg
));
575 req
->out
.argvar
= args
->out
.argvar
;
576 req
->out
.numargs
= args
->out
.numargs
;
577 memcpy(req
->out
.args
, args
->out
.args
,
578 args
->out
.numargs
* sizeof(struct fuse_arg
));
579 fuse_request_send(fc
, req
);
580 ret
= req
->out
.h
.error
;
581 if (!ret
&& args
->out
.argvar
) {
582 BUG_ON(args
->out
.numargs
!= 1);
583 ret
= req
->out
.args
[0].size
;
585 fuse_put_request(fc
, req
);
590 static void fuse_request_send_nowait_locked(struct fuse_conn
*fc
,
591 struct fuse_req
*req
)
593 BUG_ON(!req
->background
);
594 fc
->num_background
++;
595 if (fc
->num_background
== fc
->max_background
)
597 if (fc
->num_background
== fc
->congestion_threshold
&&
598 fc
->bdi_initialized
) {
599 set_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
600 set_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
602 list_add_tail(&req
->list
, &fc
->bg_queue
);
606 static void fuse_request_send_nowait(struct fuse_conn
*fc
, struct fuse_req
*req
)
608 spin_lock(&fc
->lock
);
610 fuse_request_send_nowait_locked(fc
, req
);
611 spin_unlock(&fc
->lock
);
613 req
->out
.h
.error
= -ENOTCONN
;
614 request_end(fc
, req
);
618 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
621 fuse_request_send_nowait(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
)
631 req
->in
.h
.unique
= unique
;
632 spin_lock(&fc
->lock
);
634 queue_request(fc
, req
);
637 spin_unlock(&fc
->lock
);
643 * Called under fc->lock
645 * fc->connected must have been checked previously
647 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
648 struct fuse_req
*req
)
651 fuse_request_send_nowait_locked(fc
, req
);
654 void fuse_force_forget(struct file
*file
, u64 nodeid
)
656 struct inode
*inode
= file_inode(file
);
657 struct fuse_conn
*fc
= get_fuse_conn(inode
);
658 struct fuse_req
*req
;
659 struct fuse_forget_in inarg
;
661 memset(&inarg
, 0, sizeof(inarg
));
663 req
= fuse_get_req_nofail_nopages(fc
, file
);
664 req
->in
.h
.opcode
= FUSE_FORGET
;
665 req
->in
.h
.nodeid
= nodeid
;
667 req
->in
.args
[0].size
= sizeof(inarg
);
668 req
->in
.args
[0].value
= &inarg
;
670 __fuse_request_send(fc
, req
);
672 fuse_put_request(fc
, req
);
676 * Lock the request. Up to the next unlock_request() there mustn't be
677 * anything that could cause a page-fault. If the request was already
680 static int lock_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
684 spin_lock(&fc
->lock
);
689 spin_unlock(&fc
->lock
);
695 * Unlock request. If it was aborted during being locked, the
696 * requester thread is currently waiting for it to be unlocked, so
699 static void unlock_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
702 spin_lock(&fc
->lock
);
705 wake_up(&req
->waitq
);
706 spin_unlock(&fc
->lock
);
710 struct fuse_copy_state
{
711 struct fuse_conn
*fc
;
713 struct fuse_req
*req
;
714 const struct iovec
*iov
;
715 struct pipe_buffer
*pipebufs
;
716 struct pipe_buffer
*currbuf
;
717 struct pipe_inode_info
*pipe
;
718 unsigned long nr_segs
;
719 unsigned long seglen
;
724 unsigned move_pages
:1;
727 static void fuse_copy_init(struct fuse_copy_state
*cs
, struct fuse_conn
*fc
,
729 const struct iovec
*iov
, unsigned long nr_segs
)
731 memset(cs
, 0, sizeof(*cs
));
735 cs
->nr_segs
= nr_segs
;
738 /* Unmap and put previous page of userspace buffer */
739 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
742 struct pipe_buffer
*buf
= cs
->currbuf
;
745 buf
->len
= PAGE_SIZE
- cs
->len
;
749 flush_dcache_page(cs
->pg
);
750 set_page_dirty_lock(cs
->pg
);
758 * Get another pagefull of userspace buffer, and map it to kernel
759 * address space, and lock request
761 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
766 unlock_request(cs
->fc
, cs
->req
);
767 fuse_copy_finish(cs
);
769 struct pipe_buffer
*buf
= cs
->pipebufs
;
772 err
= buf
->ops
->confirm(cs
->pipe
, buf
);
776 BUG_ON(!cs
->nr_segs
);
779 cs
->offset
= buf
->offset
;
784 if (cs
->nr_segs
== cs
->pipe
->buffers
)
787 page
= alloc_page(GFP_HIGHUSER
);
804 BUG_ON(!cs
->nr_segs
);
805 cs
->seglen
= cs
->iov
[0].iov_len
;
806 cs
->addr
= (unsigned long) cs
->iov
[0].iov_base
;
810 err
= get_user_pages_fast(cs
->addr
, 1, cs
->write
, &page
);
815 cs
->offset
= cs
->addr
% PAGE_SIZE
;
816 cs
->len
= min(PAGE_SIZE
- cs
->offset
, cs
->seglen
);
817 cs
->seglen
-= cs
->len
;
821 return lock_request(cs
->fc
, cs
->req
);
824 /* Do as much copy to/from userspace buffer as we can */
825 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
827 unsigned ncpy
= min(*size
, cs
->len
);
829 void *pgaddr
= kmap_atomic(cs
->pg
);
830 void *buf
= pgaddr
+ cs
->offset
;
833 memcpy(buf
, *val
, ncpy
);
835 memcpy(*val
, buf
, ncpy
);
837 kunmap_atomic(pgaddr
);
846 static int fuse_check_page(struct page
*page
)
848 if (page_mapcount(page
) ||
849 page
->mapping
!= NULL
||
850 page_count(page
) != 1 ||
851 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
858 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
859 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
);
865 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
868 struct page
*oldpage
= *pagep
;
869 struct page
*newpage
;
870 struct pipe_buffer
*buf
= cs
->pipebufs
;
872 unlock_request(cs
->fc
, cs
->req
);
873 fuse_copy_finish(cs
);
875 err
= buf
->ops
->confirm(cs
->pipe
, buf
);
879 BUG_ON(!cs
->nr_segs
);
885 if (cs
->len
!= PAGE_SIZE
)
888 if (buf
->ops
->steal(cs
->pipe
, buf
) != 0)
893 if (!PageUptodate(newpage
))
894 SetPageUptodate(newpage
);
896 ClearPageMappedToDisk(newpage
);
898 if (fuse_check_page(newpage
) != 0)
899 goto out_fallback_unlock
;
902 * This is a new and locked page, it shouldn't be mapped or
903 * have any special flags on it
905 if (WARN_ON(page_mapped(oldpage
)))
906 goto out_fallback_unlock
;
907 if (WARN_ON(page_has_private(oldpage
)))
908 goto out_fallback_unlock
;
909 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
910 goto out_fallback_unlock
;
911 if (WARN_ON(PageMlocked(oldpage
)))
912 goto out_fallback_unlock
;
914 err
= replace_page_cache_page(oldpage
, newpage
, GFP_KERNEL
);
916 unlock_page(newpage
);
920 page_cache_get(newpage
);
922 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
923 lru_cache_add_file(newpage
);
926 spin_lock(&cs
->fc
->lock
);
927 if (cs
->req
->aborted
)
931 spin_unlock(&cs
->fc
->lock
);
934 unlock_page(newpage
);
935 page_cache_release(newpage
);
939 unlock_page(oldpage
);
940 page_cache_release(oldpage
);
946 unlock_page(newpage
);
949 cs
->offset
= buf
->offset
;
951 err
= lock_request(cs
->fc
, cs
->req
);
958 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
959 unsigned offset
, unsigned count
)
961 struct pipe_buffer
*buf
;
963 if (cs
->nr_segs
== cs
->pipe
->buffers
)
966 unlock_request(cs
->fc
, cs
->req
);
967 fuse_copy_finish(cs
);
970 page_cache_get(page
);
972 buf
->offset
= offset
;
983 * Copy a page in the request to/from the userspace buffer. Must be
986 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
987 unsigned offset
, unsigned count
, int zeroing
)
990 struct page
*page
= *pagep
;
992 if (page
&& zeroing
&& count
< PAGE_SIZE
)
993 clear_highpage(page
);
996 if (cs
->write
&& cs
->pipebufs
&& page
) {
997 return fuse_ref_page(cs
, page
, offset
, count
);
998 } else if (!cs
->len
) {
999 if (cs
->move_pages
&& page
&&
1000 offset
== 0 && count
== PAGE_SIZE
) {
1001 err
= fuse_try_move_page(cs
, pagep
);
1005 err
= fuse_copy_fill(cs
);
1011 void *mapaddr
= kmap_atomic(page
);
1012 void *buf
= mapaddr
+ offset
;
1013 offset
+= fuse_copy_do(cs
, &buf
, &count
);
1014 kunmap_atomic(mapaddr
);
1016 offset
+= fuse_copy_do(cs
, NULL
, &count
);
1018 if (page
&& !cs
->write
)
1019 flush_dcache_page(page
);
1023 /* Copy pages in the request to/from userspace buffer */
1024 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
1028 struct fuse_req
*req
= cs
->req
;
1030 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
1032 unsigned offset
= req
->page_descs
[i
].offset
;
1033 unsigned count
= min(nbytes
, req
->page_descs
[i
].length
);
1035 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
1045 /* Copy a single argument in the request to/from userspace buffer */
1046 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
1050 int err
= fuse_copy_fill(cs
);
1054 fuse_copy_do(cs
, &val
, &size
);
1059 /* Copy request arguments to/from userspace buffer */
1060 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
1061 unsigned argpages
, struct fuse_arg
*args
,
1067 for (i
= 0; !err
&& i
< numargs
; i
++) {
1068 struct fuse_arg
*arg
= &args
[i
];
1069 if (i
== numargs
- 1 && argpages
)
1070 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
1072 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
1077 static int forget_pending(struct fuse_conn
*fc
)
1079 return fc
->forget_list_head
.next
!= NULL
;
1082 static int request_pending(struct fuse_conn
*fc
)
1084 return !list_empty(&fc
->pending
) || !list_empty(&fc
->interrupts
) ||
1088 /* Wait until a request is available on the pending list */
1089 static void request_wait(struct fuse_conn
*fc
)
1090 __releases(fc
->lock
)
1091 __acquires(fc
->lock
)
1093 DECLARE_WAITQUEUE(wait
, current
);
1095 add_wait_queue_exclusive(&fc
->waitq
, &wait
);
1096 while (fc
->connected
&& !request_pending(fc
)) {
1097 set_current_state(TASK_INTERRUPTIBLE
);
1098 if (signal_pending(current
))
1101 spin_unlock(&fc
->lock
);
1103 spin_lock(&fc
->lock
);
1105 set_current_state(TASK_RUNNING
);
1106 remove_wait_queue(&fc
->waitq
, &wait
);
1110 * Transfer an interrupt request to userspace
1112 * Unlike other requests this is assembled on demand, without a need
1113 * to allocate a separate fuse_req structure.
1115 * Called with fc->lock held, releases it
1117 static int fuse_read_interrupt(struct fuse_conn
*fc
, struct fuse_copy_state
*cs
,
1118 size_t nbytes
, struct fuse_req
*req
)
1119 __releases(fc
->lock
)
1121 struct fuse_in_header ih
;
1122 struct fuse_interrupt_in arg
;
1123 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
1126 list_del_init(&req
->intr_entry
);
1127 req
->intr_unique
= fuse_get_unique(fc
);
1128 memset(&ih
, 0, sizeof(ih
));
1129 memset(&arg
, 0, sizeof(arg
));
1131 ih
.opcode
= FUSE_INTERRUPT
;
1132 ih
.unique
= req
->intr_unique
;
1133 arg
.unique
= req
->in
.h
.unique
;
1135 spin_unlock(&fc
->lock
);
1136 if (nbytes
< reqsize
)
1139 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1141 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1142 fuse_copy_finish(cs
);
1144 return err
? err
: reqsize
;
1147 static struct fuse_forget_link
*dequeue_forget(struct fuse_conn
*fc
,
1151 struct fuse_forget_link
*head
= fc
->forget_list_head
.next
;
1152 struct fuse_forget_link
**newhead
= &head
;
1155 for (count
= 0; *newhead
!= NULL
&& count
< max
; count
++)
1156 newhead
= &(*newhead
)->next
;
1158 fc
->forget_list_head
.next
= *newhead
;
1160 if (fc
->forget_list_head
.next
== NULL
)
1161 fc
->forget_list_tail
= &fc
->forget_list_head
;
1169 static int fuse_read_single_forget(struct fuse_conn
*fc
,
1170 struct fuse_copy_state
*cs
,
1172 __releases(fc
->lock
)
1175 struct fuse_forget_link
*forget
= dequeue_forget(fc
, 1, NULL
);
1176 struct fuse_forget_in arg
= {
1177 .nlookup
= forget
->forget_one
.nlookup
,
1179 struct fuse_in_header ih
= {
1180 .opcode
= FUSE_FORGET
,
1181 .nodeid
= forget
->forget_one
.nodeid
,
1182 .unique
= fuse_get_unique(fc
),
1183 .len
= sizeof(ih
) + sizeof(arg
),
1186 spin_unlock(&fc
->lock
);
1188 if (nbytes
< ih
.len
)
1191 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1193 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1194 fuse_copy_finish(cs
);
1202 static int fuse_read_batch_forget(struct fuse_conn
*fc
,
1203 struct fuse_copy_state
*cs
, size_t nbytes
)
1204 __releases(fc
->lock
)
1207 unsigned max_forgets
;
1209 struct fuse_forget_link
*head
;
1210 struct fuse_batch_forget_in arg
= { .count
= 0 };
1211 struct fuse_in_header ih
= {
1212 .opcode
= FUSE_BATCH_FORGET
,
1213 .unique
= fuse_get_unique(fc
),
1214 .len
= sizeof(ih
) + sizeof(arg
),
1217 if (nbytes
< ih
.len
) {
1218 spin_unlock(&fc
->lock
);
1222 max_forgets
= (nbytes
- ih
.len
) / sizeof(struct fuse_forget_one
);
1223 head
= dequeue_forget(fc
, max_forgets
, &count
);
1224 spin_unlock(&fc
->lock
);
1227 ih
.len
+= count
* sizeof(struct fuse_forget_one
);
1228 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1230 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1233 struct fuse_forget_link
*forget
= head
;
1236 err
= fuse_copy_one(cs
, &forget
->forget_one
,
1237 sizeof(forget
->forget_one
));
1239 head
= forget
->next
;
1243 fuse_copy_finish(cs
);
1251 static int fuse_read_forget(struct fuse_conn
*fc
, struct fuse_copy_state
*cs
,
1253 __releases(fc
->lock
)
1255 if (fc
->minor
< 16 || fc
->forget_list_head
.next
->next
== NULL
)
1256 return fuse_read_single_forget(fc
, cs
, nbytes
);
1258 return fuse_read_batch_forget(fc
, cs
, nbytes
);
1262 * Read a single request into the userspace filesystem's buffer. This
1263 * function waits until a request is available, then removes it from
1264 * the pending list and copies request data to userspace buffer. If
1265 * no reply is needed (FORGET) or request has been aborted or there
1266 * was an error during the copying then it's finished by calling
1267 * request_end(). Otherwise add it to the processing list, and set
1270 static ssize_t
fuse_dev_do_read(struct fuse_conn
*fc
, struct file
*file
,
1271 struct fuse_copy_state
*cs
, size_t nbytes
)
1274 struct fuse_req
*req
;
1279 spin_lock(&fc
->lock
);
1281 if ((file
->f_flags
& O_NONBLOCK
) && fc
->connected
&&
1282 !request_pending(fc
))
1290 if (!request_pending(fc
))
1293 if (!list_empty(&fc
->interrupts
)) {
1294 req
= list_entry(fc
->interrupts
.next
, struct fuse_req
,
1296 return fuse_read_interrupt(fc
, cs
, nbytes
, req
);
1299 if (forget_pending(fc
)) {
1300 if (list_empty(&fc
->pending
) || fc
->forget_batch
-- > 0)
1301 return fuse_read_forget(fc
, cs
, nbytes
);
1303 if (fc
->forget_batch
<= -8)
1304 fc
->forget_batch
= 16;
1307 req
= list_entry(fc
->pending
.next
, struct fuse_req
, list
);
1308 req
->state
= FUSE_REQ_READING
;
1309 list_move(&req
->list
, &fc
->io
);
1312 reqsize
= in
->h
.len
;
1313 /* If request is too large, reply with an error and restart the read */
1314 if (nbytes
< reqsize
) {
1315 req
->out
.h
.error
= -EIO
;
1316 /* SETXATTR is special, since it may contain too large data */
1317 if (in
->h
.opcode
== FUSE_SETXATTR
)
1318 req
->out
.h
.error
= -E2BIG
;
1319 request_end(fc
, req
);
1322 spin_unlock(&fc
->lock
);
1324 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1326 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1327 (struct fuse_arg
*) in
->args
, 0);
1328 fuse_copy_finish(cs
);
1329 spin_lock(&fc
->lock
);
1332 request_end(fc
, req
);
1336 req
->out
.h
.error
= -EIO
;
1337 request_end(fc
, req
);
1341 request_end(fc
, req
);
1343 req
->state
= FUSE_REQ_SENT
;
1344 list_move_tail(&req
->list
, &fc
->processing
);
1345 if (req
->interrupted
)
1346 queue_interrupt(fc
, req
);
1347 spin_unlock(&fc
->lock
);
1352 spin_unlock(&fc
->lock
);
1356 static int fuse_dev_open(struct inode
*inode
, struct file
*file
)
1359 * The fuse device's file's private_data is used to hold
1360 * the fuse_conn(ection) when it is mounted, and is used to
1361 * keep track of whether the file has been mounted already.
1363 file
->private_data
= NULL
;
1367 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1368 unsigned long nr_segs
, loff_t pos
)
1370 struct fuse_copy_state cs
;
1371 struct file
*file
= iocb
->ki_filp
;
1372 struct fuse_conn
*fc
= fuse_get_conn(file
);
1376 fuse_copy_init(&cs
, fc
, 1, iov
, nr_segs
);
1378 return fuse_dev_do_read(fc
, file
, &cs
, iov_length(iov
, nr_segs
));
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
)
1388 struct pipe_buffer
*bufs
;
1389 struct fuse_copy_state cs
;
1390 struct fuse_conn
*fc
= fuse_get_conn(in
);
1394 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1398 fuse_copy_init(&cs
, fc
, 1, NULL
, 0);
1401 ret
= fuse_dev_do_read(fc
, in
, &cs
, len
);
1408 if (!pipe
->readers
) {
1409 send_sig(SIGPIPE
, current
, 0);
1415 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1420 while (page_nr
< cs
.nr_segs
) {
1421 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
1422 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
1424 buf
->page
= bufs
[page_nr
].page
;
1425 buf
->offset
= bufs
[page_nr
].offset
;
1426 buf
->len
= bufs
[page_nr
].len
;
1428 * Need to be careful about this. Having buf->ops in module
1429 * code can Oops if the buffer persists after module unload.
1431 buf
->ops
= &nosteal_pipe_buf_ops
;
1446 if (waitqueue_active(&pipe
->wait
))
1447 wake_up_interruptible(&pipe
->wait
);
1448 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
1452 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1453 page_cache_release(bufs
[page_nr
].page
);
1459 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1460 struct fuse_copy_state
*cs
)
1462 struct fuse_notify_poll_wakeup_out outarg
;
1465 if (size
!= sizeof(outarg
))
1468 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1472 fuse_copy_finish(cs
);
1473 return fuse_notify_poll_wakeup(fc
, &outarg
);
1476 fuse_copy_finish(cs
);
1480 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1481 struct fuse_copy_state
*cs
)
1483 struct fuse_notify_inval_inode_out outarg
;
1486 if (size
!= sizeof(outarg
))
1489 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1492 fuse_copy_finish(cs
);
1494 down_read(&fc
->killsb
);
1497 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1498 outarg
.off
, outarg
.len
);
1500 up_read(&fc
->killsb
);
1504 fuse_copy_finish(cs
);
1508 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1509 struct fuse_copy_state
*cs
)
1511 struct fuse_notify_inval_entry_out outarg
;
1516 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1521 if (size
< sizeof(outarg
))
1524 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1528 err
= -ENAMETOOLONG
;
1529 if (outarg
.namelen
> FUSE_NAME_MAX
)
1533 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1537 name
.len
= outarg
.namelen
;
1538 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1541 fuse_copy_finish(cs
);
1542 buf
[outarg
.namelen
] = 0;
1543 name
.hash
= full_name_hash(name
.name
, name
.len
);
1545 down_read(&fc
->killsb
);
1548 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, 0, &name
);
1549 up_read(&fc
->killsb
);
1555 fuse_copy_finish(cs
);
1559 static int fuse_notify_delete(struct fuse_conn
*fc
, unsigned int size
,
1560 struct fuse_copy_state
*cs
)
1562 struct fuse_notify_delete_out outarg
;
1567 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1572 if (size
< sizeof(outarg
))
1575 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1579 err
= -ENAMETOOLONG
;
1580 if (outarg
.namelen
> FUSE_NAME_MAX
)
1584 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1588 name
.len
= outarg
.namelen
;
1589 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1592 fuse_copy_finish(cs
);
1593 buf
[outarg
.namelen
] = 0;
1594 name
.hash
= full_name_hash(name
.name
, name
.len
);
1596 down_read(&fc
->killsb
);
1599 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
,
1600 outarg
.child
, &name
);
1601 up_read(&fc
->killsb
);
1607 fuse_copy_finish(cs
);
1611 static int fuse_notify_store(struct fuse_conn
*fc
, unsigned int size
,
1612 struct fuse_copy_state
*cs
)
1614 struct fuse_notify_store_out outarg
;
1615 struct inode
*inode
;
1616 struct address_space
*mapping
;
1620 unsigned int offset
;
1626 if (size
< sizeof(outarg
))
1629 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1634 if (size
- sizeof(outarg
) != outarg
.size
)
1637 nodeid
= outarg
.nodeid
;
1639 down_read(&fc
->killsb
);
1645 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1649 mapping
= inode
->i_mapping
;
1650 index
= outarg
.offset
>> PAGE_CACHE_SHIFT
;
1651 offset
= outarg
.offset
& ~PAGE_CACHE_MASK
;
1652 file_size
= i_size_read(inode
);
1653 end
= outarg
.offset
+ outarg
.size
;
1654 if (end
> file_size
) {
1656 fuse_write_update_size(inode
, file_size
);
1662 unsigned int this_num
;
1665 page
= find_or_create_page(mapping
, index
,
1666 mapping_gfp_mask(mapping
));
1670 this_num
= min_t(unsigned, num
, PAGE_CACHE_SIZE
- offset
);
1671 err
= fuse_copy_page(cs
, &page
, offset
, this_num
, 0);
1672 if (!err
&& offset
== 0 &&
1673 (this_num
== PAGE_CACHE_SIZE
|| file_size
== end
))
1674 SetPageUptodate(page
);
1676 page_cache_release(page
);
1691 up_read(&fc
->killsb
);
1693 fuse_copy_finish(cs
);
1697 static void fuse_retrieve_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1699 release_pages(req
->pages
, req
->num_pages
, false);
1702 static int fuse_retrieve(struct fuse_conn
*fc
, struct inode
*inode
,
1703 struct fuse_notify_retrieve_out
*outarg
)
1706 struct address_space
*mapping
= inode
->i_mapping
;
1707 struct fuse_req
*req
;
1711 unsigned int offset
;
1712 size_t total_len
= 0;
1715 offset
= outarg
->offset
& ~PAGE_CACHE_MASK
;
1716 file_size
= i_size_read(inode
);
1719 if (outarg
->offset
> file_size
)
1721 else if (outarg
->offset
+ num
> file_size
)
1722 num
= file_size
- outarg
->offset
;
1724 num_pages
= (num
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1725 num_pages
= min(num_pages
, FUSE_MAX_PAGES_PER_REQ
);
1727 req
= fuse_get_req(fc
, num_pages
);
1729 return PTR_ERR(req
);
1731 req
->in
.h
.opcode
= FUSE_NOTIFY_REPLY
;
1732 req
->in
.h
.nodeid
= outarg
->nodeid
;
1733 req
->in
.numargs
= 2;
1734 req
->in
.argpages
= 1;
1735 req
->page_descs
[0].offset
= offset
;
1736 req
->end
= fuse_retrieve_end
;
1738 index
= outarg
->offset
>> PAGE_CACHE_SHIFT
;
1740 while (num
&& req
->num_pages
< num_pages
) {
1742 unsigned int this_num
;
1744 page
= find_get_page(mapping
, index
);
1748 this_num
= min_t(unsigned, num
, PAGE_CACHE_SIZE
- offset
);
1749 req
->pages
[req
->num_pages
] = page
;
1750 req
->page_descs
[req
->num_pages
].length
= this_num
;
1755 total_len
+= this_num
;
1758 req
->misc
.retrieve_in
.offset
= outarg
->offset
;
1759 req
->misc
.retrieve_in
.size
= total_len
;
1760 req
->in
.args
[0].size
= sizeof(req
->misc
.retrieve_in
);
1761 req
->in
.args
[0].value
= &req
->misc
.retrieve_in
;
1762 req
->in
.args
[1].size
= total_len
;
1764 err
= fuse_request_send_notify_reply(fc
, req
, outarg
->notify_unique
);
1766 fuse_retrieve_end(fc
, req
);
1771 static int fuse_notify_retrieve(struct fuse_conn
*fc
, unsigned int size
,
1772 struct fuse_copy_state
*cs
)
1774 struct fuse_notify_retrieve_out outarg
;
1775 struct inode
*inode
;
1779 if (size
!= sizeof(outarg
))
1782 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1786 fuse_copy_finish(cs
);
1788 down_read(&fc
->killsb
);
1791 u64 nodeid
= outarg
.nodeid
;
1793 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1795 err
= fuse_retrieve(fc
, inode
, &outarg
);
1799 up_read(&fc
->killsb
);
1804 fuse_copy_finish(cs
);
1808 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1809 unsigned int size
, struct fuse_copy_state
*cs
)
1811 /* Don't try to move pages (yet) */
1815 case FUSE_NOTIFY_POLL
:
1816 return fuse_notify_poll(fc
, size
, cs
);
1818 case FUSE_NOTIFY_INVAL_INODE
:
1819 return fuse_notify_inval_inode(fc
, size
, cs
);
1821 case FUSE_NOTIFY_INVAL_ENTRY
:
1822 return fuse_notify_inval_entry(fc
, size
, cs
);
1824 case FUSE_NOTIFY_STORE
:
1825 return fuse_notify_store(fc
, size
, cs
);
1827 case FUSE_NOTIFY_RETRIEVE
:
1828 return fuse_notify_retrieve(fc
, size
, cs
);
1830 case FUSE_NOTIFY_DELETE
:
1831 return fuse_notify_delete(fc
, size
, cs
);
1834 fuse_copy_finish(cs
);
1839 /* Look up request on processing list by unique ID */
1840 static struct fuse_req
*request_find(struct fuse_conn
*fc
, u64 unique
)
1842 struct fuse_req
*req
;
1844 list_for_each_entry(req
, &fc
->processing
, list
) {
1845 if (req
->in
.h
.unique
== unique
|| req
->intr_unique
== unique
)
1851 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1854 unsigned reqsize
= sizeof(struct fuse_out_header
);
1857 return nbytes
!= reqsize
? -EINVAL
: 0;
1859 reqsize
+= len_args(out
->numargs
, out
->args
);
1861 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1863 else if (reqsize
> nbytes
) {
1864 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1865 unsigned diffsize
= reqsize
- nbytes
;
1866 if (diffsize
> lastarg
->size
)
1868 lastarg
->size
-= diffsize
;
1870 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1875 * Write a single reply to a request. First the header is copied from
1876 * the write buffer. The request is then searched on the processing
1877 * list by the unique ID found in the header. If found, then remove
1878 * it from the list and copy the rest of the buffer to the request.
1879 * The request is finished by calling request_end()
1881 static ssize_t
fuse_dev_do_write(struct fuse_conn
*fc
,
1882 struct fuse_copy_state
*cs
, size_t nbytes
)
1885 struct fuse_req
*req
;
1886 struct fuse_out_header oh
;
1888 if (nbytes
< sizeof(struct fuse_out_header
))
1891 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1896 if (oh
.len
!= nbytes
)
1900 * Zero oh.unique indicates unsolicited notification message
1901 * and error contains notification code.
1904 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1905 return err
? err
: nbytes
;
1909 if (oh
.error
<= -1000 || oh
.error
> 0)
1912 spin_lock(&fc
->lock
);
1917 req
= request_find(fc
, oh
.unique
);
1922 spin_unlock(&fc
->lock
);
1923 fuse_copy_finish(cs
);
1924 spin_lock(&fc
->lock
);
1925 request_end(fc
, req
);
1928 /* Is it an interrupt reply? */
1929 if (req
->intr_unique
== oh
.unique
) {
1931 if (nbytes
!= sizeof(struct fuse_out_header
))
1934 if (oh
.error
== -ENOSYS
)
1935 fc
->no_interrupt
= 1;
1936 else if (oh
.error
== -EAGAIN
)
1937 queue_interrupt(fc
, req
);
1939 spin_unlock(&fc
->lock
);
1940 fuse_copy_finish(cs
);
1944 req
->state
= FUSE_REQ_WRITING
;
1945 list_move(&req
->list
, &fc
->io
);
1949 if (!req
->out
.page_replace
)
1951 spin_unlock(&fc
->lock
);
1953 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1954 fuse_copy_finish(cs
);
1956 spin_lock(&fc
->lock
);
1961 } else if (!req
->aborted
)
1962 req
->out
.h
.error
= -EIO
;
1963 request_end(fc
, req
);
1965 return err
? err
: nbytes
;
1968 spin_unlock(&fc
->lock
);
1970 fuse_copy_finish(cs
);
1974 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1975 unsigned long nr_segs
, loff_t pos
)
1977 struct fuse_copy_state cs
;
1978 struct fuse_conn
*fc
= fuse_get_conn(iocb
->ki_filp
);
1982 fuse_copy_init(&cs
, fc
, 0, iov
, nr_segs
);
1984 return fuse_dev_do_write(fc
, &cs
, iov_length(iov
, nr_segs
));
1987 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1988 struct file
*out
, loff_t
*ppos
,
1989 size_t len
, unsigned int flags
)
1993 struct pipe_buffer
*bufs
;
1994 struct fuse_copy_state cs
;
1995 struct fuse_conn
*fc
;
1999 fc
= fuse_get_conn(out
);
2003 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
2010 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
2011 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
2021 struct pipe_buffer
*ibuf
;
2022 struct pipe_buffer
*obuf
;
2024 BUG_ON(nbuf
>= pipe
->buffers
);
2025 BUG_ON(!pipe
->nrbufs
);
2026 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
2029 if (rem
>= ibuf
->len
) {
2032 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
2035 ibuf
->ops
->get(pipe
, ibuf
);
2037 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
2039 ibuf
->offset
+= obuf
->len
;
2040 ibuf
->len
-= obuf
->len
;
2047 fuse_copy_init(&cs
, fc
, 0, NULL
, nbuf
);
2051 if (flags
& SPLICE_F_MOVE
)
2054 ret
= fuse_dev_do_write(fc
, &cs
, len
);
2056 for (idx
= 0; idx
< nbuf
; idx
++) {
2057 struct pipe_buffer
*buf
= &bufs
[idx
];
2058 buf
->ops
->release(pipe
, buf
);
2065 static unsigned fuse_dev_poll(struct file
*file
, poll_table
*wait
)
2067 unsigned mask
= POLLOUT
| POLLWRNORM
;
2068 struct fuse_conn
*fc
= fuse_get_conn(file
);
2072 poll_wait(file
, &fc
->waitq
, wait
);
2074 spin_lock(&fc
->lock
);
2077 else if (request_pending(fc
))
2078 mask
|= POLLIN
| POLLRDNORM
;
2079 spin_unlock(&fc
->lock
);
2085 * Abort all requests on the given list (pending or processing)
2087 * This function releases and reacquires fc->lock
2089 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
2090 __releases(fc
->lock
)
2091 __acquires(fc
->lock
)
2093 while (!list_empty(head
)) {
2094 struct fuse_req
*req
;
2095 req
= list_entry(head
->next
, struct fuse_req
, list
);
2096 req
->out
.h
.error
= -ECONNABORTED
;
2097 request_end(fc
, req
);
2098 spin_lock(&fc
->lock
);
2103 * Abort requests under I/O
2105 * The requests are set to aborted and finished, and the request
2106 * waiter is woken up. This will make request_wait_answer() wait
2107 * until the request is unlocked and then return.
2109 * If the request is asynchronous, then the end function needs to be
2110 * called after waiting for the request to be unlocked (if it was
2113 static void end_io_requests(struct fuse_conn
*fc
)
2114 __releases(fc
->lock
)
2115 __acquires(fc
->lock
)
2117 while (!list_empty(&fc
->io
)) {
2118 struct fuse_req
*req
=
2119 list_entry(fc
->io
.next
, struct fuse_req
, list
);
2120 void (*end
) (struct fuse_conn
*, struct fuse_req
*) = req
->end
;
2123 req
->out
.h
.error
= -ECONNABORTED
;
2124 req
->state
= FUSE_REQ_FINISHED
;
2125 list_del_init(&req
->list
);
2126 wake_up(&req
->waitq
);
2129 __fuse_get_request(req
);
2130 spin_unlock(&fc
->lock
);
2131 wait_event(req
->waitq
, !req
->locked
);
2133 fuse_put_request(fc
, req
);
2134 spin_lock(&fc
->lock
);
2139 static void end_queued_requests(struct fuse_conn
*fc
)
2140 __releases(fc
->lock
)
2141 __acquires(fc
->lock
)
2143 fc
->max_background
= UINT_MAX
;
2145 end_requests(fc
, &fc
->pending
);
2146 end_requests(fc
, &fc
->processing
);
2147 while (forget_pending(fc
))
2148 kfree(dequeue_forget(fc
, 1, NULL
));
2151 static void end_polls(struct fuse_conn
*fc
)
2155 p
= rb_first(&fc
->polled_files
);
2158 struct fuse_file
*ff
;
2159 ff
= rb_entry(p
, struct fuse_file
, polled_node
);
2160 wake_up_interruptible_all(&ff
->poll_wait
);
2167 * Abort all requests.
2169 * Emergency exit in case of a malicious or accidental deadlock, or
2170 * just a hung filesystem.
2172 * The same effect is usually achievable through killing the
2173 * filesystem daemon and all users of the filesystem. The exception
2174 * is the combination of an asynchronous request and the tricky
2175 * deadlock (see Documentation/filesystems/fuse.txt).
2177 * During the aborting, progression of requests from the pending and
2178 * processing lists onto the io list, and progression of new requests
2179 * onto the pending list is prevented by req->connected being false.
2181 * Progression of requests under I/O to the processing list is
2182 * prevented by the req->aborted flag being true for these requests.
2183 * For this reason requests on the io list must be aborted first.
2185 void fuse_abort_conn(struct fuse_conn
*fc
)
2187 spin_lock(&fc
->lock
);
2188 if (fc
->connected
) {
2191 fuse_set_initialized(fc
);
2192 end_io_requests(fc
);
2193 end_queued_requests(fc
);
2195 wake_up_all(&fc
->waitq
);
2196 wake_up_all(&fc
->blocked_waitq
);
2197 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
2199 spin_unlock(&fc
->lock
);
2201 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
2203 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
2205 struct fuse_conn
*fc
= fuse_get_conn(file
);
2207 spin_lock(&fc
->lock
);
2210 fuse_set_initialized(fc
);
2211 end_queued_requests(fc
);
2213 wake_up_all(&fc
->blocked_waitq
);
2214 spin_unlock(&fc
->lock
);
2220 EXPORT_SYMBOL_GPL(fuse_dev_release
);
2222 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
2224 struct fuse_conn
*fc
= fuse_get_conn(file
);
2228 /* No locking - fasync_helper does its own locking */
2229 return fasync_helper(fd
, file
, on
, &fc
->fasync
);
2232 const struct file_operations fuse_dev_operations
= {
2233 .owner
= THIS_MODULE
,
2234 .open
= fuse_dev_open
,
2235 .llseek
= no_llseek
,
2236 .read
= do_sync_read
,
2237 .aio_read
= fuse_dev_read
,
2238 .splice_read
= fuse_dev_splice_read
,
2239 .write
= do_sync_write
,
2240 .aio_write
= fuse_dev_write
,
2241 .splice_write
= fuse_dev_splice_write
,
2242 .poll
= fuse_dev_poll
,
2243 .release
= fuse_dev_release
,
2244 .fasync
= fuse_dev_fasync
,
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
);