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>
23 MODULE_ALIAS_MISCDEV(FUSE_MINOR
);
24 MODULE_ALIAS("devname:fuse");
26 static struct kmem_cache
*fuse_req_cachep
;
28 static struct fuse_conn
*fuse_get_conn(struct file
*file
)
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
34 return file
->private_data
;
37 static void fuse_request_init(struct fuse_req
*req
, struct page
**pages
,
38 struct fuse_page_desc
*page_descs
,
41 memset(req
, 0, sizeof(*req
));
42 memset(pages
, 0, sizeof(*pages
) * npages
);
43 memset(page_descs
, 0, sizeof(*page_descs
) * npages
);
44 INIT_LIST_HEAD(&req
->list
);
45 INIT_LIST_HEAD(&req
->intr_entry
);
46 init_waitqueue_head(&req
->waitq
);
47 atomic_set(&req
->count
, 1);
49 req
->page_descs
= page_descs
;
50 req
->max_pages
= npages
;
53 static struct fuse_req
*__fuse_request_alloc(unsigned npages
, gfp_t flags
)
55 struct fuse_req
*req
= kmem_cache_alloc(fuse_req_cachep
, flags
);
58 struct fuse_page_desc
*page_descs
;
60 if (npages
<= FUSE_REQ_INLINE_PAGES
) {
61 pages
= req
->inline_pages
;
62 page_descs
= req
->inline_page_descs
;
64 pages
= kmalloc(sizeof(struct page
*) * npages
, flags
);
65 page_descs
= kmalloc(sizeof(struct fuse_page_desc
) *
69 if (!pages
|| !page_descs
) {
72 kmem_cache_free(fuse_req_cachep
, req
);
76 fuse_request_init(req
, pages
, page_descs
, npages
);
81 struct fuse_req
*fuse_request_alloc(unsigned npages
)
83 return __fuse_request_alloc(npages
, GFP_KERNEL
);
85 EXPORT_SYMBOL_GPL(fuse_request_alloc
);
87 struct fuse_req
*fuse_request_alloc_nofs(unsigned npages
)
89 return __fuse_request_alloc(npages
, GFP_NOFS
);
92 void fuse_request_free(struct fuse_req
*req
)
94 if (req
->pages
!= req
->inline_pages
) {
96 kfree(req
->page_descs
);
98 kmem_cache_free(fuse_req_cachep
, req
);
101 static void block_sigs(sigset_t
*oldset
)
105 siginitsetinv(&mask
, sigmask(SIGKILL
));
106 sigprocmask(SIG_BLOCK
, &mask
, oldset
);
109 static void restore_sigs(sigset_t
*oldset
)
111 sigprocmask(SIG_SETMASK
, oldset
, NULL
);
114 static void __fuse_get_request(struct fuse_req
*req
)
116 atomic_inc(&req
->count
);
119 /* Must be called with > 1 refcount */
120 static void __fuse_put_request(struct fuse_req
*req
)
122 BUG_ON(atomic_read(&req
->count
) < 2);
123 atomic_dec(&req
->count
);
126 static void fuse_req_init_context(struct fuse_req
*req
)
128 req
->in
.h
.uid
= from_kuid_munged(&init_user_ns
, current_fsuid());
129 req
->in
.h
.gid
= from_kgid_munged(&init_user_ns
, current_fsgid());
130 req
->in
.h
.pid
= current
->pid
;
133 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
, unsigned npages
)
135 struct fuse_req
*req
;
140 atomic_inc(&fc
->num_waiting
);
142 intr
= wait_event_interruptible(fc
->blocked_waitq
, !fc
->blocked
);
143 restore_sigs(&oldset
);
152 req
= fuse_request_alloc(npages
);
157 fuse_req_init_context(req
);
162 atomic_dec(&fc
->num_waiting
);
165 EXPORT_SYMBOL_GPL(fuse_get_req
);
168 * Return request in fuse_file->reserved_req. However that may
169 * currently be in use. If that is the case, wait for it to become
172 static struct fuse_req
*get_reserved_req(struct fuse_conn
*fc
,
175 struct fuse_req
*req
= NULL
;
176 struct fuse_file
*ff
= file
->private_data
;
179 wait_event(fc
->reserved_req_waitq
, ff
->reserved_req
);
180 spin_lock(&fc
->lock
);
181 if (ff
->reserved_req
) {
182 req
= ff
->reserved_req
;
183 ff
->reserved_req
= NULL
;
184 req
->stolen_file
= get_file(file
);
186 spin_unlock(&fc
->lock
);
193 * Put stolen request back into fuse_file->reserved_req
195 static void put_reserved_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
197 struct file
*file
= req
->stolen_file
;
198 struct fuse_file
*ff
= file
->private_data
;
200 spin_lock(&fc
->lock
);
201 fuse_request_init(req
, req
->pages
, req
->page_descs
, req
->max_pages
);
202 BUG_ON(ff
->reserved_req
);
203 ff
->reserved_req
= req
;
204 wake_up_all(&fc
->reserved_req_waitq
);
205 spin_unlock(&fc
->lock
);
210 * Gets a requests for a file operation, always succeeds
212 * This is used for sending the FLUSH request, which must get to
213 * userspace, due to POSIX locks which may need to be unlocked.
215 * If allocation fails due to OOM, use the reserved request in
218 * This is very unlikely to deadlock accidentally, since the
219 * filesystem should not have it's own file open. If deadlock is
220 * intentional, it can still be broken by "aborting" the filesystem.
222 struct fuse_req
*fuse_get_req_nofail_nopages(struct fuse_conn
*fc
,
225 struct fuse_req
*req
;
227 atomic_inc(&fc
->num_waiting
);
228 wait_event(fc
->blocked_waitq
, !fc
->blocked
);
229 req
= fuse_request_alloc(0);
231 req
= get_reserved_req(fc
, file
);
233 fuse_req_init_context(req
);
238 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
240 if (atomic_dec_and_test(&req
->count
)) {
242 atomic_dec(&fc
->num_waiting
);
244 if (req
->stolen_file
)
245 put_reserved_req(fc
, req
);
247 fuse_request_free(req
);
250 EXPORT_SYMBOL_GPL(fuse_put_request
);
252 static unsigned len_args(unsigned numargs
, struct fuse_arg
*args
)
257 for (i
= 0; i
< numargs
; i
++)
258 nbytes
+= args
[i
].size
;
263 static u64
fuse_get_unique(struct fuse_conn
*fc
)
266 /* zero is special */
273 static void queue_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
275 req
->in
.h
.len
= sizeof(struct fuse_in_header
) +
276 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
277 list_add_tail(&req
->list
, &fc
->pending
);
278 req
->state
= FUSE_REQ_PENDING
;
281 atomic_inc(&fc
->num_waiting
);
284 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
287 void fuse_queue_forget(struct fuse_conn
*fc
, struct fuse_forget_link
*forget
,
288 u64 nodeid
, u64 nlookup
)
290 forget
->forget_one
.nodeid
= nodeid
;
291 forget
->forget_one
.nlookup
= nlookup
;
293 spin_lock(&fc
->lock
);
295 fc
->forget_list_tail
->next
= forget
;
296 fc
->forget_list_tail
= forget
;
298 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
302 spin_unlock(&fc
->lock
);
305 static void flush_bg_queue(struct fuse_conn
*fc
)
307 while (fc
->active_background
< fc
->max_background
&&
308 !list_empty(&fc
->bg_queue
)) {
309 struct fuse_req
*req
;
311 req
= list_entry(fc
->bg_queue
.next
, struct fuse_req
, list
);
312 list_del(&req
->list
);
313 fc
->active_background
++;
314 req
->in
.h
.unique
= fuse_get_unique(fc
);
315 queue_request(fc
, req
);
320 * This function is called when a request is finished. Either a reply
321 * has arrived or it was aborted (and not yet sent) or some error
322 * occurred during communication with userspace, or the device file
323 * was closed. The requester thread is woken up (if still waiting),
324 * the 'end' callback is called if given, else the reference to the
325 * request is released
327 * Called with fc->lock, unlocks it
329 static void request_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
332 void (*end
) (struct fuse_conn
*, struct fuse_req
*) = req
->end
;
334 list_del(&req
->list
);
335 list_del(&req
->intr_entry
);
336 req
->state
= FUSE_REQ_FINISHED
;
337 if (req
->background
) {
338 if (fc
->num_background
== fc
->max_background
) {
340 wake_up_all(&fc
->blocked_waitq
);
342 if (fc
->num_background
== fc
->congestion_threshold
&&
343 fc
->connected
&& fc
->bdi_initialized
) {
344 clear_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
345 clear_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
347 fc
->num_background
--;
348 fc
->active_background
--;
351 spin_unlock(&fc
->lock
);
352 wake_up(&req
->waitq
);
355 fuse_put_request(fc
, req
);
358 static void wait_answer_interruptible(struct fuse_conn
*fc
,
359 struct fuse_req
*req
)
363 if (signal_pending(current
))
366 spin_unlock(&fc
->lock
);
367 wait_event_interruptible(req
->waitq
, req
->state
== FUSE_REQ_FINISHED
);
368 spin_lock(&fc
->lock
);
371 static void queue_interrupt(struct fuse_conn
*fc
, struct fuse_req
*req
)
373 list_add_tail(&req
->intr_entry
, &fc
->interrupts
);
375 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
378 static void request_wait_answer(struct fuse_conn
*fc
, struct fuse_req
*req
)
382 if (!fc
->no_interrupt
) {
383 /* Any signal may interrupt this */
384 wait_answer_interruptible(fc
, req
);
388 if (req
->state
== FUSE_REQ_FINISHED
)
391 req
->interrupted
= 1;
392 if (req
->state
== FUSE_REQ_SENT
)
393 queue_interrupt(fc
, req
);
399 /* Only fatal signals may interrupt this */
401 wait_answer_interruptible(fc
, req
);
402 restore_sigs(&oldset
);
406 if (req
->state
== FUSE_REQ_FINISHED
)
409 /* Request is not yet in userspace, bail out */
410 if (req
->state
== FUSE_REQ_PENDING
) {
411 list_del(&req
->list
);
412 __fuse_put_request(req
);
413 req
->out
.h
.error
= -EINTR
;
419 * Either request is already in userspace, or it was forced.
422 spin_unlock(&fc
->lock
);
423 wait_event(req
->waitq
, req
->state
== FUSE_REQ_FINISHED
);
424 spin_lock(&fc
->lock
);
430 BUG_ON(req
->state
!= FUSE_REQ_FINISHED
);
432 /* This is uninterruptible sleep, because data is
433 being copied to/from the buffers of req. During
434 locked state, there mustn't be any filesystem
435 operation (e.g. page fault), since that could lead
437 spin_unlock(&fc
->lock
);
438 wait_event(req
->waitq
, !req
->locked
);
439 spin_lock(&fc
->lock
);
443 static void __fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
445 spin_lock(&fc
->lock
);
447 req
->out
.h
.error
= -ENOTCONN
;
448 else if (fc
->conn_error
)
449 req
->out
.h
.error
= -ECONNREFUSED
;
451 req
->in
.h
.unique
= fuse_get_unique(fc
);
452 queue_request(fc
, req
);
453 /* acquire extra reference, since request is still needed
454 after request_end() */
455 __fuse_get_request(req
);
457 request_wait_answer(fc
, req
);
459 spin_unlock(&fc
->lock
);
462 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
465 __fuse_request_send(fc
, req
);
467 EXPORT_SYMBOL_GPL(fuse_request_send
);
469 static void fuse_request_send_nowait_locked(struct fuse_conn
*fc
,
470 struct fuse_req
*req
)
473 fc
->num_background
++;
474 if (fc
->num_background
== fc
->max_background
)
476 if (fc
->num_background
== fc
->congestion_threshold
&&
477 fc
->bdi_initialized
) {
478 set_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
479 set_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
481 list_add_tail(&req
->list
, &fc
->bg_queue
);
485 static void fuse_request_send_nowait(struct fuse_conn
*fc
, struct fuse_req
*req
)
487 spin_lock(&fc
->lock
);
489 fuse_request_send_nowait_locked(fc
, req
);
490 spin_unlock(&fc
->lock
);
492 req
->out
.h
.error
= -ENOTCONN
;
493 request_end(fc
, req
);
497 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
500 fuse_request_send_nowait(fc
, req
);
502 EXPORT_SYMBOL_GPL(fuse_request_send_background
);
504 static int fuse_request_send_notify_reply(struct fuse_conn
*fc
,
505 struct fuse_req
*req
, u64 unique
)
510 req
->in
.h
.unique
= unique
;
511 spin_lock(&fc
->lock
);
513 queue_request(fc
, req
);
516 spin_unlock(&fc
->lock
);
522 * Called under fc->lock
524 * fc->connected must have been checked previously
526 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
527 struct fuse_req
*req
)
530 fuse_request_send_nowait_locked(fc
, req
);
533 void fuse_force_forget(struct file
*file
, u64 nodeid
)
535 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
536 struct fuse_conn
*fc
= get_fuse_conn(inode
);
537 struct fuse_req
*req
;
538 struct fuse_forget_in inarg
;
540 memset(&inarg
, 0, sizeof(inarg
));
542 req
= fuse_get_req_nofail_nopages(fc
, file
);
543 req
->in
.h
.opcode
= FUSE_FORGET
;
544 req
->in
.h
.nodeid
= nodeid
;
546 req
->in
.args
[0].size
= sizeof(inarg
);
547 req
->in
.args
[0].value
= &inarg
;
549 __fuse_request_send(fc
, req
);
551 fuse_put_request(fc
, req
);
555 * Lock the request. Up to the next unlock_request() there mustn't be
556 * anything that could cause a page-fault. If the request was already
559 static int lock_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
563 spin_lock(&fc
->lock
);
568 spin_unlock(&fc
->lock
);
574 * Unlock request. If it was aborted during being locked, the
575 * requester thread is currently waiting for it to be unlocked, so
578 static void unlock_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
581 spin_lock(&fc
->lock
);
584 wake_up(&req
->waitq
);
585 spin_unlock(&fc
->lock
);
589 struct fuse_copy_state
{
590 struct fuse_conn
*fc
;
592 struct fuse_req
*req
;
593 const struct iovec
*iov
;
594 struct pipe_buffer
*pipebufs
;
595 struct pipe_buffer
*currbuf
;
596 struct pipe_inode_info
*pipe
;
597 unsigned long nr_segs
;
598 unsigned long seglen
;
604 unsigned move_pages
:1;
607 static void fuse_copy_init(struct fuse_copy_state
*cs
, struct fuse_conn
*fc
,
609 const struct iovec
*iov
, unsigned long nr_segs
)
611 memset(cs
, 0, sizeof(*cs
));
615 cs
->nr_segs
= nr_segs
;
618 /* Unmap and put previous page of userspace buffer */
619 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
622 struct pipe_buffer
*buf
= cs
->currbuf
;
625 buf
->ops
->unmap(cs
->pipe
, buf
, cs
->mapaddr
);
628 buf
->len
= PAGE_SIZE
- cs
->len
;
632 } else if (cs
->mapaddr
) {
635 flush_dcache_page(cs
->pg
);
636 set_page_dirty_lock(cs
->pg
);
644 * Get another pagefull of userspace buffer, and map it to kernel
645 * address space, and lock request
647 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
649 unsigned long offset
;
652 unlock_request(cs
->fc
, cs
->req
);
653 fuse_copy_finish(cs
);
655 struct pipe_buffer
*buf
= cs
->pipebufs
;
658 err
= buf
->ops
->confirm(cs
->pipe
, buf
);
662 BUG_ON(!cs
->nr_segs
);
664 cs
->mapaddr
= buf
->ops
->map(cs
->pipe
, buf
, 0);
666 cs
->buf
= cs
->mapaddr
+ buf
->offset
;
672 if (cs
->nr_segs
== cs
->pipe
->buffers
)
675 page
= alloc_page(GFP_HIGHUSER
);
684 cs
->mapaddr
= kmap(page
);
685 cs
->buf
= cs
->mapaddr
;
692 BUG_ON(!cs
->nr_segs
);
693 cs
->seglen
= cs
->iov
[0].iov_len
;
694 cs
->addr
= (unsigned long) cs
->iov
[0].iov_base
;
698 err
= get_user_pages_fast(cs
->addr
, 1, cs
->write
, &cs
->pg
);
702 offset
= cs
->addr
% PAGE_SIZE
;
703 cs
->mapaddr
= kmap(cs
->pg
);
704 cs
->buf
= cs
->mapaddr
+ offset
;
705 cs
->len
= min(PAGE_SIZE
- offset
, cs
->seglen
);
706 cs
->seglen
-= cs
->len
;
710 return lock_request(cs
->fc
, cs
->req
);
713 /* Do as much copy to/from userspace buffer as we can */
714 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
716 unsigned ncpy
= min(*size
, cs
->len
);
719 memcpy(cs
->buf
, *val
, ncpy
);
721 memcpy(*val
, cs
->buf
, ncpy
);
730 static int fuse_check_page(struct page
*page
)
732 if (page_mapcount(page
) ||
733 page
->mapping
!= NULL
||
734 page_count(page
) != 1 ||
735 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
742 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
743 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
);
749 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
752 struct page
*oldpage
= *pagep
;
753 struct page
*newpage
;
754 struct pipe_buffer
*buf
= cs
->pipebufs
;
756 unlock_request(cs
->fc
, cs
->req
);
757 fuse_copy_finish(cs
);
759 err
= buf
->ops
->confirm(cs
->pipe
, buf
);
763 BUG_ON(!cs
->nr_segs
);
769 if (cs
->len
!= PAGE_SIZE
)
772 if (buf
->ops
->steal(cs
->pipe
, buf
) != 0)
777 if (WARN_ON(!PageUptodate(newpage
)))
780 ClearPageMappedToDisk(newpage
);
782 if (fuse_check_page(newpage
) != 0)
783 goto out_fallback_unlock
;
786 * This is a new and locked page, it shouldn't be mapped or
787 * have any special flags on it
789 if (WARN_ON(page_mapped(oldpage
)))
790 goto out_fallback_unlock
;
791 if (WARN_ON(page_has_private(oldpage
)))
792 goto out_fallback_unlock
;
793 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
794 goto out_fallback_unlock
;
795 if (WARN_ON(PageMlocked(oldpage
)))
796 goto out_fallback_unlock
;
798 err
= replace_page_cache_page(oldpage
, newpage
, GFP_KERNEL
);
800 unlock_page(newpage
);
804 page_cache_get(newpage
);
806 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
807 lru_cache_add_file(newpage
);
810 spin_lock(&cs
->fc
->lock
);
811 if (cs
->req
->aborted
)
815 spin_unlock(&cs
->fc
->lock
);
818 unlock_page(newpage
);
819 page_cache_release(newpage
);
823 unlock_page(oldpage
);
824 page_cache_release(oldpage
);
830 unlock_page(newpage
);
832 cs
->mapaddr
= buf
->ops
->map(cs
->pipe
, buf
, 1);
833 cs
->buf
= cs
->mapaddr
+ buf
->offset
;
835 err
= lock_request(cs
->fc
, cs
->req
);
842 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
843 unsigned offset
, unsigned count
)
845 struct pipe_buffer
*buf
;
847 if (cs
->nr_segs
== cs
->pipe
->buffers
)
850 unlock_request(cs
->fc
, cs
->req
);
851 fuse_copy_finish(cs
);
854 page_cache_get(page
);
856 buf
->offset
= offset
;
867 * Copy a page in the request to/from the userspace buffer. Must be
870 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
871 unsigned offset
, unsigned count
, int zeroing
)
874 struct page
*page
= *pagep
;
876 if (page
&& zeroing
&& count
< PAGE_SIZE
)
877 clear_highpage(page
);
880 if (cs
->write
&& cs
->pipebufs
&& page
) {
881 return fuse_ref_page(cs
, page
, offset
, count
);
882 } else if (!cs
->len
) {
883 if (cs
->move_pages
&& page
&&
884 offset
== 0 && count
== PAGE_SIZE
) {
885 err
= fuse_try_move_page(cs
, pagep
);
889 err
= fuse_copy_fill(cs
);
895 void *mapaddr
= kmap_atomic(page
);
896 void *buf
= mapaddr
+ offset
;
897 offset
+= fuse_copy_do(cs
, &buf
, &count
);
898 kunmap_atomic(mapaddr
);
900 offset
+= fuse_copy_do(cs
, NULL
, &count
);
902 if (page
&& !cs
->write
)
903 flush_dcache_page(page
);
907 /* Copy pages in the request to/from userspace buffer */
908 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
912 struct fuse_req
*req
= cs
->req
;
914 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
916 unsigned offset
= req
->page_descs
[i
].offset
;
917 unsigned count
= min(nbytes
, req
->page_descs
[i
].length
);
919 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
929 /* Copy a single argument in the request to/from userspace buffer */
930 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
934 int err
= fuse_copy_fill(cs
);
938 fuse_copy_do(cs
, &val
, &size
);
943 /* Copy request arguments to/from userspace buffer */
944 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
945 unsigned argpages
, struct fuse_arg
*args
,
951 for (i
= 0; !err
&& i
< numargs
; i
++) {
952 struct fuse_arg
*arg
= &args
[i
];
953 if (i
== numargs
- 1 && argpages
)
954 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
956 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
961 static int forget_pending(struct fuse_conn
*fc
)
963 return fc
->forget_list_head
.next
!= NULL
;
966 static int request_pending(struct fuse_conn
*fc
)
968 return !list_empty(&fc
->pending
) || !list_empty(&fc
->interrupts
) ||
972 /* Wait until a request is available on the pending list */
973 static void request_wait(struct fuse_conn
*fc
)
977 DECLARE_WAITQUEUE(wait
, current
);
979 add_wait_queue_exclusive(&fc
->waitq
, &wait
);
980 while (fc
->connected
&& !request_pending(fc
)) {
981 set_current_state(TASK_INTERRUPTIBLE
);
982 if (signal_pending(current
))
985 spin_unlock(&fc
->lock
);
987 spin_lock(&fc
->lock
);
989 set_current_state(TASK_RUNNING
);
990 remove_wait_queue(&fc
->waitq
, &wait
);
994 * Transfer an interrupt request to userspace
996 * Unlike other requests this is assembled on demand, without a need
997 * to allocate a separate fuse_req structure.
999 * Called with fc->lock held, releases it
1001 static int fuse_read_interrupt(struct fuse_conn
*fc
, struct fuse_copy_state
*cs
,
1002 size_t nbytes
, struct fuse_req
*req
)
1003 __releases(fc
->lock
)
1005 struct fuse_in_header ih
;
1006 struct fuse_interrupt_in arg
;
1007 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
1010 list_del_init(&req
->intr_entry
);
1011 req
->intr_unique
= fuse_get_unique(fc
);
1012 memset(&ih
, 0, sizeof(ih
));
1013 memset(&arg
, 0, sizeof(arg
));
1015 ih
.opcode
= FUSE_INTERRUPT
;
1016 ih
.unique
= req
->intr_unique
;
1017 arg
.unique
= req
->in
.h
.unique
;
1019 spin_unlock(&fc
->lock
);
1020 if (nbytes
< reqsize
)
1023 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1025 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1026 fuse_copy_finish(cs
);
1028 return err
? err
: reqsize
;
1031 static struct fuse_forget_link
*dequeue_forget(struct fuse_conn
*fc
,
1035 struct fuse_forget_link
*head
= fc
->forget_list_head
.next
;
1036 struct fuse_forget_link
**newhead
= &head
;
1039 for (count
= 0; *newhead
!= NULL
&& count
< max
; count
++)
1040 newhead
= &(*newhead
)->next
;
1042 fc
->forget_list_head
.next
= *newhead
;
1044 if (fc
->forget_list_head
.next
== NULL
)
1045 fc
->forget_list_tail
= &fc
->forget_list_head
;
1053 static int fuse_read_single_forget(struct fuse_conn
*fc
,
1054 struct fuse_copy_state
*cs
,
1056 __releases(fc
->lock
)
1059 struct fuse_forget_link
*forget
= dequeue_forget(fc
, 1, NULL
);
1060 struct fuse_forget_in arg
= {
1061 .nlookup
= forget
->forget_one
.nlookup
,
1063 struct fuse_in_header ih
= {
1064 .opcode
= FUSE_FORGET
,
1065 .nodeid
= forget
->forget_one
.nodeid
,
1066 .unique
= fuse_get_unique(fc
),
1067 .len
= sizeof(ih
) + sizeof(arg
),
1070 spin_unlock(&fc
->lock
);
1072 if (nbytes
< ih
.len
)
1075 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1077 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1078 fuse_copy_finish(cs
);
1086 static int fuse_read_batch_forget(struct fuse_conn
*fc
,
1087 struct fuse_copy_state
*cs
, size_t nbytes
)
1088 __releases(fc
->lock
)
1091 unsigned max_forgets
;
1093 struct fuse_forget_link
*head
;
1094 struct fuse_batch_forget_in arg
= { .count
= 0 };
1095 struct fuse_in_header ih
= {
1096 .opcode
= FUSE_BATCH_FORGET
,
1097 .unique
= fuse_get_unique(fc
),
1098 .len
= sizeof(ih
) + sizeof(arg
),
1101 if (nbytes
< ih
.len
) {
1102 spin_unlock(&fc
->lock
);
1106 max_forgets
= (nbytes
- ih
.len
) / sizeof(struct fuse_forget_one
);
1107 head
= dequeue_forget(fc
, max_forgets
, &count
);
1108 spin_unlock(&fc
->lock
);
1111 ih
.len
+= count
* sizeof(struct fuse_forget_one
);
1112 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
1114 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
1117 struct fuse_forget_link
*forget
= head
;
1120 err
= fuse_copy_one(cs
, &forget
->forget_one
,
1121 sizeof(forget
->forget_one
));
1123 head
= forget
->next
;
1127 fuse_copy_finish(cs
);
1135 static int fuse_read_forget(struct fuse_conn
*fc
, struct fuse_copy_state
*cs
,
1137 __releases(fc
->lock
)
1139 if (fc
->minor
< 16 || fc
->forget_list_head
.next
->next
== NULL
)
1140 return fuse_read_single_forget(fc
, cs
, nbytes
);
1142 return fuse_read_batch_forget(fc
, cs
, nbytes
);
1146 * Read a single request into the userspace filesystem's buffer. This
1147 * function waits until a request is available, then removes it from
1148 * the pending list and copies request data to userspace buffer. If
1149 * no reply is needed (FORGET) or request has been aborted or there
1150 * was an error during the copying then it's finished by calling
1151 * request_end(). Otherwise add it to the processing list, and set
1154 static ssize_t
fuse_dev_do_read(struct fuse_conn
*fc
, struct file
*file
,
1155 struct fuse_copy_state
*cs
, size_t nbytes
)
1158 struct fuse_req
*req
;
1163 spin_lock(&fc
->lock
);
1165 if ((file
->f_flags
& O_NONBLOCK
) && fc
->connected
&&
1166 !request_pending(fc
))
1174 if (!request_pending(fc
))
1177 if (!list_empty(&fc
->interrupts
)) {
1178 req
= list_entry(fc
->interrupts
.next
, struct fuse_req
,
1180 return fuse_read_interrupt(fc
, cs
, nbytes
, req
);
1183 if (forget_pending(fc
)) {
1184 if (list_empty(&fc
->pending
) || fc
->forget_batch
-- > 0)
1185 return fuse_read_forget(fc
, cs
, nbytes
);
1187 if (fc
->forget_batch
<= -8)
1188 fc
->forget_batch
= 16;
1191 req
= list_entry(fc
->pending
.next
, struct fuse_req
, list
);
1192 req
->state
= FUSE_REQ_READING
;
1193 list_move(&req
->list
, &fc
->io
);
1196 reqsize
= in
->h
.len
;
1197 /* If request is too large, reply with an error and restart the read */
1198 if (nbytes
< reqsize
) {
1199 req
->out
.h
.error
= -EIO
;
1200 /* SETXATTR is special, since it may contain too large data */
1201 if (in
->h
.opcode
== FUSE_SETXATTR
)
1202 req
->out
.h
.error
= -E2BIG
;
1203 request_end(fc
, req
);
1206 spin_unlock(&fc
->lock
);
1208 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1210 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1211 (struct fuse_arg
*) in
->args
, 0);
1212 fuse_copy_finish(cs
);
1213 spin_lock(&fc
->lock
);
1216 request_end(fc
, req
);
1220 req
->out
.h
.error
= -EIO
;
1221 request_end(fc
, req
);
1225 request_end(fc
, req
);
1227 req
->state
= FUSE_REQ_SENT
;
1228 list_move_tail(&req
->list
, &fc
->processing
);
1229 if (req
->interrupted
)
1230 queue_interrupt(fc
, req
);
1231 spin_unlock(&fc
->lock
);
1236 spin_unlock(&fc
->lock
);
1240 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1241 unsigned long nr_segs
, loff_t pos
)
1243 struct fuse_copy_state cs
;
1244 struct file
*file
= iocb
->ki_filp
;
1245 struct fuse_conn
*fc
= fuse_get_conn(file
);
1249 fuse_copy_init(&cs
, fc
, 1, iov
, nr_segs
);
1251 return fuse_dev_do_read(fc
, file
, &cs
, iov_length(iov
, nr_segs
));
1254 static int fuse_dev_pipe_buf_steal(struct pipe_inode_info
*pipe
,
1255 struct pipe_buffer
*buf
)
1260 static const struct pipe_buf_operations fuse_dev_pipe_buf_ops
= {
1262 .map
= generic_pipe_buf_map
,
1263 .unmap
= generic_pipe_buf_unmap
,
1264 .confirm
= generic_pipe_buf_confirm
,
1265 .release
= generic_pipe_buf_release
,
1266 .steal
= fuse_dev_pipe_buf_steal
,
1267 .get
= generic_pipe_buf_get
,
1270 static ssize_t
fuse_dev_splice_read(struct file
*in
, loff_t
*ppos
,
1271 struct pipe_inode_info
*pipe
,
1272 size_t len
, unsigned int flags
)
1277 struct pipe_buffer
*bufs
;
1278 struct fuse_copy_state cs
;
1279 struct fuse_conn
*fc
= fuse_get_conn(in
);
1283 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1287 fuse_copy_init(&cs
, fc
, 1, NULL
, 0);
1290 ret
= fuse_dev_do_read(fc
, in
, &cs
, len
);
1297 if (!pipe
->readers
) {
1298 send_sig(SIGPIPE
, current
, 0);
1304 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1309 while (page_nr
< cs
.nr_segs
) {
1310 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
1311 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
1313 buf
->page
= bufs
[page_nr
].page
;
1314 buf
->offset
= bufs
[page_nr
].offset
;
1315 buf
->len
= bufs
[page_nr
].len
;
1316 buf
->ops
= &fuse_dev_pipe_buf_ops
;
1331 if (waitqueue_active(&pipe
->wait
))
1332 wake_up_interruptible(&pipe
->wait
);
1333 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
1337 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1338 page_cache_release(bufs
[page_nr
].page
);
1344 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1345 struct fuse_copy_state
*cs
)
1347 struct fuse_notify_poll_wakeup_out outarg
;
1350 if (size
!= sizeof(outarg
))
1353 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1357 fuse_copy_finish(cs
);
1358 return fuse_notify_poll_wakeup(fc
, &outarg
);
1361 fuse_copy_finish(cs
);
1365 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1366 struct fuse_copy_state
*cs
)
1368 struct fuse_notify_inval_inode_out outarg
;
1371 if (size
!= sizeof(outarg
))
1374 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1377 fuse_copy_finish(cs
);
1379 down_read(&fc
->killsb
);
1382 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1383 outarg
.off
, outarg
.len
);
1385 up_read(&fc
->killsb
);
1389 fuse_copy_finish(cs
);
1393 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1394 struct fuse_copy_state
*cs
)
1396 struct fuse_notify_inval_entry_out outarg
;
1401 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1406 if (size
< sizeof(outarg
))
1409 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1413 err
= -ENAMETOOLONG
;
1414 if (outarg
.namelen
> FUSE_NAME_MAX
)
1418 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1422 name
.len
= outarg
.namelen
;
1423 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1426 fuse_copy_finish(cs
);
1427 buf
[outarg
.namelen
] = 0;
1428 name
.hash
= full_name_hash(name
.name
, name
.len
);
1430 down_read(&fc
->killsb
);
1433 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, 0, &name
);
1434 up_read(&fc
->killsb
);
1440 fuse_copy_finish(cs
);
1444 static int fuse_notify_delete(struct fuse_conn
*fc
, unsigned int size
,
1445 struct fuse_copy_state
*cs
)
1447 struct fuse_notify_delete_out outarg
;
1452 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1457 if (size
< sizeof(outarg
))
1460 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1464 err
= -ENAMETOOLONG
;
1465 if (outarg
.namelen
> FUSE_NAME_MAX
)
1469 if (size
!= sizeof(outarg
) + outarg
.namelen
+ 1)
1473 name
.len
= outarg
.namelen
;
1474 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1477 fuse_copy_finish(cs
);
1478 buf
[outarg
.namelen
] = 0;
1479 name
.hash
= full_name_hash(name
.name
, name
.len
);
1481 down_read(&fc
->killsb
);
1484 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
,
1485 outarg
.child
, &name
);
1486 up_read(&fc
->killsb
);
1492 fuse_copy_finish(cs
);
1496 static int fuse_notify_store(struct fuse_conn
*fc
, unsigned int size
,
1497 struct fuse_copy_state
*cs
)
1499 struct fuse_notify_store_out outarg
;
1500 struct inode
*inode
;
1501 struct address_space
*mapping
;
1505 unsigned int offset
;
1511 if (size
< sizeof(outarg
))
1514 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1519 if (size
- sizeof(outarg
) != outarg
.size
)
1522 nodeid
= outarg
.nodeid
;
1524 down_read(&fc
->killsb
);
1530 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1534 mapping
= inode
->i_mapping
;
1535 index
= outarg
.offset
>> PAGE_CACHE_SHIFT
;
1536 offset
= outarg
.offset
& ~PAGE_CACHE_MASK
;
1537 file_size
= i_size_read(inode
);
1538 end
= outarg
.offset
+ outarg
.size
;
1539 if (end
> file_size
) {
1541 fuse_write_update_size(inode
, file_size
);
1547 unsigned int this_num
;
1550 page
= find_or_create_page(mapping
, index
,
1551 mapping_gfp_mask(mapping
));
1555 this_num
= min_t(unsigned, num
, PAGE_CACHE_SIZE
- offset
);
1556 err
= fuse_copy_page(cs
, &page
, offset
, this_num
, 0);
1557 if (!err
&& offset
== 0 && (num
!= 0 || file_size
== end
))
1558 SetPageUptodate(page
);
1560 page_cache_release(page
);
1575 up_read(&fc
->killsb
);
1577 fuse_copy_finish(cs
);
1581 static void fuse_retrieve_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1583 release_pages(req
->pages
, req
->num_pages
, 0);
1586 static int fuse_retrieve(struct fuse_conn
*fc
, struct inode
*inode
,
1587 struct fuse_notify_retrieve_out
*outarg
)
1590 struct address_space
*mapping
= inode
->i_mapping
;
1591 struct fuse_req
*req
;
1595 unsigned int offset
;
1596 size_t total_len
= 0;
1599 offset
= outarg
->offset
& ~PAGE_CACHE_MASK
;
1600 file_size
= i_size_read(inode
);
1603 if (outarg
->offset
> file_size
)
1605 else if (outarg
->offset
+ num
> file_size
)
1606 num
= file_size
- outarg
->offset
;
1608 num_pages
= (num
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1609 num_pages
= min(num_pages
, FUSE_MAX_PAGES_PER_REQ
);
1611 req
= fuse_get_req(fc
, num_pages
);
1613 return PTR_ERR(req
);
1615 req
->in
.h
.opcode
= FUSE_NOTIFY_REPLY
;
1616 req
->in
.h
.nodeid
= outarg
->nodeid
;
1617 req
->in
.numargs
= 2;
1618 req
->in
.argpages
= 1;
1619 req
->page_descs
[0].offset
= offset
;
1620 req
->end
= fuse_retrieve_end
;
1622 index
= outarg
->offset
>> PAGE_CACHE_SHIFT
;
1624 while (num
&& req
->num_pages
< num_pages
) {
1626 unsigned int this_num
;
1628 page
= find_get_page(mapping
, index
);
1632 this_num
= min_t(unsigned, num
, PAGE_CACHE_SIZE
- offset
);
1633 req
->pages
[req
->num_pages
] = page
;
1634 req
->page_descs
[req
->num_pages
].length
= this_num
;
1639 total_len
+= this_num
;
1642 req
->misc
.retrieve_in
.offset
= outarg
->offset
;
1643 req
->misc
.retrieve_in
.size
= total_len
;
1644 req
->in
.args
[0].size
= sizeof(req
->misc
.retrieve_in
);
1645 req
->in
.args
[0].value
= &req
->misc
.retrieve_in
;
1646 req
->in
.args
[1].size
= total_len
;
1648 err
= fuse_request_send_notify_reply(fc
, req
, outarg
->notify_unique
);
1650 fuse_retrieve_end(fc
, req
);
1655 static int fuse_notify_retrieve(struct fuse_conn
*fc
, unsigned int size
,
1656 struct fuse_copy_state
*cs
)
1658 struct fuse_notify_retrieve_out outarg
;
1659 struct inode
*inode
;
1663 if (size
!= sizeof(outarg
))
1666 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1670 fuse_copy_finish(cs
);
1672 down_read(&fc
->killsb
);
1675 u64 nodeid
= outarg
.nodeid
;
1677 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1679 err
= fuse_retrieve(fc
, inode
, &outarg
);
1683 up_read(&fc
->killsb
);
1688 fuse_copy_finish(cs
);
1692 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1693 unsigned int size
, struct fuse_copy_state
*cs
)
1696 case FUSE_NOTIFY_POLL
:
1697 return fuse_notify_poll(fc
, size
, cs
);
1699 case FUSE_NOTIFY_INVAL_INODE
:
1700 return fuse_notify_inval_inode(fc
, size
, cs
);
1702 case FUSE_NOTIFY_INVAL_ENTRY
:
1703 return fuse_notify_inval_entry(fc
, size
, cs
);
1705 case FUSE_NOTIFY_STORE
:
1706 return fuse_notify_store(fc
, size
, cs
);
1708 case FUSE_NOTIFY_RETRIEVE
:
1709 return fuse_notify_retrieve(fc
, size
, cs
);
1711 case FUSE_NOTIFY_DELETE
:
1712 return fuse_notify_delete(fc
, size
, cs
);
1715 fuse_copy_finish(cs
);
1720 /* Look up request on processing list by unique ID */
1721 static struct fuse_req
*request_find(struct fuse_conn
*fc
, u64 unique
)
1723 struct list_head
*entry
;
1725 list_for_each(entry
, &fc
->processing
) {
1726 struct fuse_req
*req
;
1727 req
= list_entry(entry
, struct fuse_req
, list
);
1728 if (req
->in
.h
.unique
== unique
|| req
->intr_unique
== unique
)
1734 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1737 unsigned reqsize
= sizeof(struct fuse_out_header
);
1740 return nbytes
!= reqsize
? -EINVAL
: 0;
1742 reqsize
+= len_args(out
->numargs
, out
->args
);
1744 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1746 else if (reqsize
> nbytes
) {
1747 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1748 unsigned diffsize
= reqsize
- nbytes
;
1749 if (diffsize
> lastarg
->size
)
1751 lastarg
->size
-= diffsize
;
1753 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1758 * Write a single reply to a request. First the header is copied from
1759 * the write buffer. The request is then searched on the processing
1760 * list by the unique ID found in the header. If found, then remove
1761 * it from the list and copy the rest of the buffer to the request.
1762 * The request is finished by calling request_end()
1764 static ssize_t
fuse_dev_do_write(struct fuse_conn
*fc
,
1765 struct fuse_copy_state
*cs
, size_t nbytes
)
1768 struct fuse_req
*req
;
1769 struct fuse_out_header oh
;
1771 if (nbytes
< sizeof(struct fuse_out_header
))
1774 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1779 if (oh
.len
!= nbytes
)
1783 * Zero oh.unique indicates unsolicited notification message
1784 * and error contains notification code.
1787 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1788 return err
? err
: nbytes
;
1792 if (oh
.error
<= -1000 || oh
.error
> 0)
1795 spin_lock(&fc
->lock
);
1800 req
= request_find(fc
, oh
.unique
);
1805 spin_unlock(&fc
->lock
);
1806 fuse_copy_finish(cs
);
1807 spin_lock(&fc
->lock
);
1808 request_end(fc
, req
);
1811 /* Is it an interrupt reply? */
1812 if (req
->intr_unique
== oh
.unique
) {
1814 if (nbytes
!= sizeof(struct fuse_out_header
))
1817 if (oh
.error
== -ENOSYS
)
1818 fc
->no_interrupt
= 1;
1819 else if (oh
.error
== -EAGAIN
)
1820 queue_interrupt(fc
, req
);
1822 spin_unlock(&fc
->lock
);
1823 fuse_copy_finish(cs
);
1827 req
->state
= FUSE_REQ_WRITING
;
1828 list_move(&req
->list
, &fc
->io
);
1832 if (!req
->out
.page_replace
)
1834 spin_unlock(&fc
->lock
);
1836 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1837 fuse_copy_finish(cs
);
1839 spin_lock(&fc
->lock
);
1844 } else if (!req
->aborted
)
1845 req
->out
.h
.error
= -EIO
;
1846 request_end(fc
, req
);
1848 return err
? err
: nbytes
;
1851 spin_unlock(&fc
->lock
);
1853 fuse_copy_finish(cs
);
1857 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1858 unsigned long nr_segs
, loff_t pos
)
1860 struct fuse_copy_state cs
;
1861 struct fuse_conn
*fc
= fuse_get_conn(iocb
->ki_filp
);
1865 fuse_copy_init(&cs
, fc
, 0, iov
, nr_segs
);
1867 return fuse_dev_do_write(fc
, &cs
, iov_length(iov
, nr_segs
));
1870 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1871 struct file
*out
, loff_t
*ppos
,
1872 size_t len
, unsigned int flags
)
1876 struct pipe_buffer
*bufs
;
1877 struct fuse_copy_state cs
;
1878 struct fuse_conn
*fc
;
1882 fc
= fuse_get_conn(out
);
1886 bufs
= kmalloc(pipe
->buffers
* sizeof(struct pipe_buffer
), GFP_KERNEL
);
1893 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
1894 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
1904 struct pipe_buffer
*ibuf
;
1905 struct pipe_buffer
*obuf
;
1907 BUG_ON(nbuf
>= pipe
->buffers
);
1908 BUG_ON(!pipe
->nrbufs
);
1909 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
1912 if (rem
>= ibuf
->len
) {
1915 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
1918 ibuf
->ops
->get(pipe
, ibuf
);
1920 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1922 ibuf
->offset
+= obuf
->len
;
1923 ibuf
->len
-= obuf
->len
;
1930 fuse_copy_init(&cs
, fc
, 0, NULL
, nbuf
);
1934 if (flags
& SPLICE_F_MOVE
)
1937 ret
= fuse_dev_do_write(fc
, &cs
, len
);
1939 for (idx
= 0; idx
< nbuf
; idx
++) {
1940 struct pipe_buffer
*buf
= &bufs
[idx
];
1941 buf
->ops
->release(pipe
, buf
);
1948 static unsigned fuse_dev_poll(struct file
*file
, poll_table
*wait
)
1950 unsigned mask
= POLLOUT
| POLLWRNORM
;
1951 struct fuse_conn
*fc
= fuse_get_conn(file
);
1955 poll_wait(file
, &fc
->waitq
, wait
);
1957 spin_lock(&fc
->lock
);
1960 else if (request_pending(fc
))
1961 mask
|= POLLIN
| POLLRDNORM
;
1962 spin_unlock(&fc
->lock
);
1968 * Abort all requests on the given list (pending or processing)
1970 * This function releases and reacquires fc->lock
1972 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
1973 __releases(fc
->lock
)
1974 __acquires(fc
->lock
)
1976 while (!list_empty(head
)) {
1977 struct fuse_req
*req
;
1978 req
= list_entry(head
->next
, struct fuse_req
, list
);
1979 req
->out
.h
.error
= -ECONNABORTED
;
1980 request_end(fc
, req
);
1981 spin_lock(&fc
->lock
);
1986 * Abort requests under I/O
1988 * The requests are set to aborted and finished, and the request
1989 * waiter is woken up. This will make request_wait_answer() wait
1990 * until the request is unlocked and then return.
1992 * If the request is asynchronous, then the end function needs to be
1993 * called after waiting for the request to be unlocked (if it was
1996 static void end_io_requests(struct fuse_conn
*fc
)
1997 __releases(fc
->lock
)
1998 __acquires(fc
->lock
)
2000 while (!list_empty(&fc
->io
)) {
2001 struct fuse_req
*req
=
2002 list_entry(fc
->io
.next
, struct fuse_req
, list
);
2003 void (*end
) (struct fuse_conn
*, struct fuse_req
*) = req
->end
;
2006 req
->out
.h
.error
= -ECONNABORTED
;
2007 req
->state
= FUSE_REQ_FINISHED
;
2008 list_del_init(&req
->list
);
2009 wake_up(&req
->waitq
);
2012 __fuse_get_request(req
);
2013 spin_unlock(&fc
->lock
);
2014 wait_event(req
->waitq
, !req
->locked
);
2016 fuse_put_request(fc
, req
);
2017 spin_lock(&fc
->lock
);
2022 static void end_queued_requests(struct fuse_conn
*fc
)
2023 __releases(fc
->lock
)
2024 __acquires(fc
->lock
)
2026 fc
->max_background
= UINT_MAX
;
2028 end_requests(fc
, &fc
->pending
);
2029 end_requests(fc
, &fc
->processing
);
2030 while (forget_pending(fc
))
2031 kfree(dequeue_forget(fc
, 1, NULL
));
2034 static void end_polls(struct fuse_conn
*fc
)
2038 p
= rb_first(&fc
->polled_files
);
2041 struct fuse_file
*ff
;
2042 ff
= rb_entry(p
, struct fuse_file
, polled_node
);
2043 wake_up_interruptible_all(&ff
->poll_wait
);
2050 * Abort all requests.
2052 * Emergency exit in case of a malicious or accidental deadlock, or
2053 * just a hung filesystem.
2055 * The same effect is usually achievable through killing the
2056 * filesystem daemon and all users of the filesystem. The exception
2057 * is the combination of an asynchronous request and the tricky
2058 * deadlock (see Documentation/filesystems/fuse.txt).
2060 * During the aborting, progression of requests from the pending and
2061 * processing lists onto the io list, and progression of new requests
2062 * onto the pending list is prevented by req->connected being false.
2064 * Progression of requests under I/O to the processing list is
2065 * prevented by the req->aborted flag being true for these requests.
2066 * For this reason requests on the io list must be aborted first.
2068 void fuse_abort_conn(struct fuse_conn
*fc
)
2070 spin_lock(&fc
->lock
);
2071 if (fc
->connected
) {
2074 end_io_requests(fc
);
2075 end_queued_requests(fc
);
2077 wake_up_all(&fc
->waitq
);
2078 wake_up_all(&fc
->blocked_waitq
);
2079 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
2081 spin_unlock(&fc
->lock
);
2083 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
2085 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
2087 struct fuse_conn
*fc
= fuse_get_conn(file
);
2089 spin_lock(&fc
->lock
);
2092 end_queued_requests(fc
);
2094 wake_up_all(&fc
->blocked_waitq
);
2095 spin_unlock(&fc
->lock
);
2101 EXPORT_SYMBOL_GPL(fuse_dev_release
);
2103 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
2105 struct fuse_conn
*fc
= fuse_get_conn(file
);
2109 /* No locking - fasync_helper does its own locking */
2110 return fasync_helper(fd
, file
, on
, &fc
->fasync
);
2113 const struct file_operations fuse_dev_operations
= {
2114 .owner
= THIS_MODULE
,
2115 .llseek
= no_llseek
,
2116 .read
= do_sync_read
,
2117 .aio_read
= fuse_dev_read
,
2118 .splice_read
= fuse_dev_splice_read
,
2119 .write
= do_sync_write
,
2120 .aio_write
= fuse_dev_write
,
2121 .splice_write
= fuse_dev_splice_write
,
2122 .poll
= fuse_dev_poll
,
2123 .release
= fuse_dev_release
,
2124 .fasync
= fuse_dev_fasync
,
2126 EXPORT_SYMBOL_GPL(fuse_dev_operations
);
2128 static struct miscdevice fuse_miscdevice
= {
2129 .minor
= FUSE_MINOR
,
2131 .fops
= &fuse_dev_operations
,
2134 int __init
fuse_dev_init(void)
2137 fuse_req_cachep
= kmem_cache_create("fuse_request",
2138 sizeof(struct fuse_req
),
2140 if (!fuse_req_cachep
)
2143 err
= misc_register(&fuse_miscdevice
);
2145 goto out_cache_clean
;
2150 kmem_cache_destroy(fuse_req_cachep
);
2155 void fuse_dev_cleanup(void)
2157 misc_deregister(&fuse_miscdevice
);
2158 kmem_cache_destroy(fuse_req_cachep
);