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
)
39 memset(req
, 0, sizeof(*req
));
40 INIT_LIST_HEAD(&req
->list
);
41 INIT_LIST_HEAD(&req
->intr_entry
);
42 init_waitqueue_head(&req
->waitq
);
43 atomic_set(&req
->count
, 1);
46 struct fuse_req
*fuse_request_alloc(void)
48 struct fuse_req
*req
= kmem_cache_alloc(fuse_req_cachep
, GFP_KERNEL
);
50 fuse_request_init(req
);
53 EXPORT_SYMBOL_GPL(fuse_request_alloc
);
55 struct fuse_req
*fuse_request_alloc_nofs(void)
57 struct fuse_req
*req
= kmem_cache_alloc(fuse_req_cachep
, GFP_NOFS
);
59 fuse_request_init(req
);
63 void fuse_request_free(struct fuse_req
*req
)
65 kmem_cache_free(fuse_req_cachep
, req
);
68 static void block_sigs(sigset_t
*oldset
)
72 siginitsetinv(&mask
, sigmask(SIGKILL
));
73 sigprocmask(SIG_BLOCK
, &mask
, oldset
);
76 static void restore_sigs(sigset_t
*oldset
)
78 sigprocmask(SIG_SETMASK
, oldset
, NULL
);
81 static void __fuse_get_request(struct fuse_req
*req
)
83 atomic_inc(&req
->count
);
86 /* Must be called with > 1 refcount */
87 static void __fuse_put_request(struct fuse_req
*req
)
89 BUG_ON(atomic_read(&req
->count
) < 2);
90 atomic_dec(&req
->count
);
93 static void fuse_req_init_context(struct fuse_req
*req
)
95 req
->in
.h
.uid
= current_fsuid();
96 req
->in
.h
.gid
= current_fsgid();
97 req
->in
.h
.pid
= current
->pid
;
100 struct fuse_req
*fuse_get_req(struct fuse_conn
*fc
)
102 struct fuse_req
*req
;
107 atomic_inc(&fc
->num_waiting
);
109 intr
= wait_event_interruptible(fc
->blocked_waitq
, !fc
->blocked
);
110 restore_sigs(&oldset
);
119 req
= fuse_request_alloc();
124 fuse_req_init_context(req
);
129 atomic_dec(&fc
->num_waiting
);
132 EXPORT_SYMBOL_GPL(fuse_get_req
);
135 * Return request in fuse_file->reserved_req. However that may
136 * currently be in use. If that is the case, wait for it to become
139 static struct fuse_req
*get_reserved_req(struct fuse_conn
*fc
,
142 struct fuse_req
*req
= NULL
;
143 struct fuse_file
*ff
= file
->private_data
;
146 wait_event(fc
->reserved_req_waitq
, ff
->reserved_req
);
147 spin_lock(&fc
->lock
);
148 if (ff
->reserved_req
) {
149 req
= ff
->reserved_req
;
150 ff
->reserved_req
= NULL
;
152 req
->stolen_file
= file
;
154 spin_unlock(&fc
->lock
);
161 * Put stolen request back into fuse_file->reserved_req
163 static void put_reserved_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
165 struct file
*file
= req
->stolen_file
;
166 struct fuse_file
*ff
= file
->private_data
;
168 spin_lock(&fc
->lock
);
169 fuse_request_init(req
);
170 BUG_ON(ff
->reserved_req
);
171 ff
->reserved_req
= req
;
172 wake_up_all(&fc
->reserved_req_waitq
);
173 spin_unlock(&fc
->lock
);
178 * Gets a requests for a file operation, always succeeds
180 * This is used for sending the FLUSH request, which must get to
181 * userspace, due to POSIX locks which may need to be unlocked.
183 * If allocation fails due to OOM, use the reserved request in
186 * This is very unlikely to deadlock accidentally, since the
187 * filesystem should not have it's own file open. If deadlock is
188 * intentional, it can still be broken by "aborting" the filesystem.
190 struct fuse_req
*fuse_get_req_nofail(struct fuse_conn
*fc
, struct file
*file
)
192 struct fuse_req
*req
;
194 atomic_inc(&fc
->num_waiting
);
195 wait_event(fc
->blocked_waitq
, !fc
->blocked
);
196 req
= fuse_request_alloc();
198 req
= get_reserved_req(fc
, file
);
200 fuse_req_init_context(req
);
205 void fuse_put_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
207 if (atomic_dec_and_test(&req
->count
)) {
209 atomic_dec(&fc
->num_waiting
);
211 if (req
->stolen_file
)
212 put_reserved_req(fc
, req
);
214 fuse_request_free(req
);
217 EXPORT_SYMBOL_GPL(fuse_put_request
);
219 static unsigned len_args(unsigned numargs
, struct fuse_arg
*args
)
224 for (i
= 0; i
< numargs
; i
++)
225 nbytes
+= args
[i
].size
;
230 static u64
fuse_get_unique(struct fuse_conn
*fc
)
233 /* zero is special */
240 static void queue_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
242 req
->in
.h
.unique
= fuse_get_unique(fc
);
243 req
->in
.h
.len
= sizeof(struct fuse_in_header
) +
244 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
245 list_add_tail(&req
->list
, &fc
->pending
);
246 req
->state
= FUSE_REQ_PENDING
;
249 atomic_inc(&fc
->num_waiting
);
252 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
255 static void flush_bg_queue(struct fuse_conn
*fc
)
257 while (fc
->active_background
< fc
->max_background
&&
258 !list_empty(&fc
->bg_queue
)) {
259 struct fuse_req
*req
;
261 req
= list_entry(fc
->bg_queue
.next
, struct fuse_req
, list
);
262 list_del(&req
->list
);
263 fc
->active_background
++;
264 queue_request(fc
, req
);
269 * This function is called when a request is finished. Either a reply
270 * has arrived or it was aborted (and not yet sent) or some error
271 * occurred during communication with userspace, or the device file
272 * was closed. The requester thread is woken up (if still waiting),
273 * the 'end' callback is called if given, else the reference to the
274 * request is released
276 * Called with fc->lock, unlocks it
278 static void request_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
279 __releases(&fc
->lock
)
281 void (*end
) (struct fuse_conn
*, struct fuse_req
*) = req
->end
;
283 list_del(&req
->list
);
284 list_del(&req
->intr_entry
);
285 req
->state
= FUSE_REQ_FINISHED
;
286 if (req
->background
) {
287 if (fc
->num_background
== fc
->max_background
) {
289 wake_up_all(&fc
->blocked_waitq
);
291 if (fc
->num_background
== fc
->congestion_threshold
&&
292 fc
->connected
&& fc
->bdi_initialized
) {
293 clear_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
294 clear_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
296 fc
->num_background
--;
297 fc
->active_background
--;
300 spin_unlock(&fc
->lock
);
301 wake_up(&req
->waitq
);
304 fuse_put_request(fc
, req
);
307 static void wait_answer_interruptible(struct fuse_conn
*fc
,
308 struct fuse_req
*req
)
309 __releases(&fc
->lock
)
310 __acquires(&fc
->lock
)
312 if (signal_pending(current
))
315 spin_unlock(&fc
->lock
);
316 wait_event_interruptible(req
->waitq
, req
->state
== FUSE_REQ_FINISHED
);
317 spin_lock(&fc
->lock
);
320 static void queue_interrupt(struct fuse_conn
*fc
, struct fuse_req
*req
)
322 list_add_tail(&req
->intr_entry
, &fc
->interrupts
);
324 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
327 static void request_wait_answer(struct fuse_conn
*fc
, struct fuse_req
*req
)
328 __releases(&fc
->lock
)
329 __acquires(&fc
->lock
)
331 if (!fc
->no_interrupt
) {
332 /* Any signal may interrupt this */
333 wait_answer_interruptible(fc
, req
);
337 if (req
->state
== FUSE_REQ_FINISHED
)
340 req
->interrupted
= 1;
341 if (req
->state
== FUSE_REQ_SENT
)
342 queue_interrupt(fc
, req
);
348 /* Only fatal signals may interrupt this */
350 wait_answer_interruptible(fc
, req
);
351 restore_sigs(&oldset
);
355 if (req
->state
== FUSE_REQ_FINISHED
)
358 /* Request is not yet in userspace, bail out */
359 if (req
->state
== FUSE_REQ_PENDING
) {
360 list_del(&req
->list
);
361 __fuse_put_request(req
);
362 req
->out
.h
.error
= -EINTR
;
368 * Either request is already in userspace, or it was forced.
371 spin_unlock(&fc
->lock
);
372 wait_event(req
->waitq
, req
->state
== FUSE_REQ_FINISHED
);
373 spin_lock(&fc
->lock
);
379 BUG_ON(req
->state
!= FUSE_REQ_FINISHED
);
381 /* This is uninterruptible sleep, because data is
382 being copied to/from the buffers of req. During
383 locked state, there mustn't be any filesystem
384 operation (e.g. page fault), since that could lead
386 spin_unlock(&fc
->lock
);
387 wait_event(req
->waitq
, !req
->locked
);
388 spin_lock(&fc
->lock
);
392 void fuse_request_send(struct fuse_conn
*fc
, struct fuse_req
*req
)
395 spin_lock(&fc
->lock
);
397 req
->out
.h
.error
= -ENOTCONN
;
398 else if (fc
->conn_error
)
399 req
->out
.h
.error
= -ECONNREFUSED
;
401 queue_request(fc
, req
);
402 /* acquire extra reference, since request is still needed
403 after request_end() */
404 __fuse_get_request(req
);
406 request_wait_answer(fc
, req
);
408 spin_unlock(&fc
->lock
);
410 EXPORT_SYMBOL_GPL(fuse_request_send
);
412 static void fuse_request_send_nowait_locked(struct fuse_conn
*fc
,
413 struct fuse_req
*req
)
416 fc
->num_background
++;
417 if (fc
->num_background
== fc
->max_background
)
419 if (fc
->num_background
== fc
->congestion_threshold
&&
420 fc
->bdi_initialized
) {
421 set_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
422 set_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
424 list_add_tail(&req
->list
, &fc
->bg_queue
);
428 static void fuse_request_send_nowait(struct fuse_conn
*fc
, struct fuse_req
*req
)
430 spin_lock(&fc
->lock
);
432 fuse_request_send_nowait_locked(fc
, req
);
433 spin_unlock(&fc
->lock
);
435 req
->out
.h
.error
= -ENOTCONN
;
436 request_end(fc
, req
);
440 void fuse_request_send_noreply(struct fuse_conn
*fc
, struct fuse_req
*req
)
443 fuse_request_send_nowait(fc
, req
);
446 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
449 fuse_request_send_nowait(fc
, req
);
451 EXPORT_SYMBOL_GPL(fuse_request_send_background
);
454 * Called under fc->lock
456 * fc->connected must have been checked previously
458 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
459 struct fuse_req
*req
)
462 fuse_request_send_nowait_locked(fc
, req
);
466 * Lock the request. Up to the next unlock_request() there mustn't be
467 * anything that could cause a page-fault. If the request was already
470 static int lock_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
474 spin_lock(&fc
->lock
);
479 spin_unlock(&fc
->lock
);
485 * Unlock request. If it was aborted during being locked, the
486 * requester thread is currently waiting for it to be unlocked, so
489 static void unlock_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
492 spin_lock(&fc
->lock
);
495 wake_up(&req
->waitq
);
496 spin_unlock(&fc
->lock
);
500 struct fuse_copy_state
{
501 struct fuse_conn
*fc
;
503 struct fuse_req
*req
;
504 const struct iovec
*iov
;
505 struct pipe_buffer
*pipebufs
;
506 struct pipe_buffer
*currbuf
;
507 struct pipe_inode_info
*pipe
;
508 unsigned long nr_segs
;
509 unsigned long seglen
;
515 unsigned move_pages
:1;
518 static void fuse_copy_init(struct fuse_copy_state
*cs
, struct fuse_conn
*fc
,
520 const struct iovec
*iov
, unsigned long nr_segs
)
522 memset(cs
, 0, sizeof(*cs
));
526 cs
->nr_segs
= nr_segs
;
529 /* Unmap and put previous page of userspace buffer */
530 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
533 struct pipe_buffer
*buf
= cs
->currbuf
;
536 buf
->ops
->unmap(cs
->pipe
, buf
, cs
->mapaddr
);
538 kunmap_atomic(cs
->mapaddr
, KM_USER0
);
539 buf
->len
= PAGE_SIZE
- cs
->len
;
543 } else if (cs
->mapaddr
) {
544 kunmap_atomic(cs
->mapaddr
, KM_USER0
);
546 flush_dcache_page(cs
->pg
);
547 set_page_dirty_lock(cs
->pg
);
555 * Get another pagefull of userspace buffer, and map it to kernel
556 * address space, and lock request
558 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
560 unsigned long offset
;
563 unlock_request(cs
->fc
, cs
->req
);
564 fuse_copy_finish(cs
);
566 struct pipe_buffer
*buf
= cs
->pipebufs
;
569 err
= buf
->ops
->confirm(cs
->pipe
, buf
);
573 BUG_ON(!cs
->nr_segs
);
575 cs
->mapaddr
= buf
->ops
->map(cs
->pipe
, buf
, 1);
577 cs
->buf
= cs
->mapaddr
+ buf
->offset
;
583 if (cs
->nr_segs
== cs
->pipe
->buffers
)
586 page
= alloc_page(GFP_HIGHUSER
);
595 cs
->mapaddr
= kmap_atomic(page
, KM_USER0
);
596 cs
->buf
= cs
->mapaddr
;
603 BUG_ON(!cs
->nr_segs
);
604 cs
->seglen
= cs
->iov
[0].iov_len
;
605 cs
->addr
= (unsigned long) cs
->iov
[0].iov_base
;
609 err
= get_user_pages_fast(cs
->addr
, 1, cs
->write
, &cs
->pg
);
613 offset
= cs
->addr
% PAGE_SIZE
;
614 cs
->mapaddr
= kmap_atomic(cs
->pg
, KM_USER0
);
615 cs
->buf
= cs
->mapaddr
+ offset
;
616 cs
->len
= min(PAGE_SIZE
- offset
, cs
->seglen
);
617 cs
->seglen
-= cs
->len
;
621 return lock_request(cs
->fc
, cs
->req
);
624 /* Do as much copy to/from userspace buffer as we can */
625 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
627 unsigned ncpy
= min(*size
, cs
->len
);
630 memcpy(cs
->buf
, *val
, ncpy
);
632 memcpy(*val
, cs
->buf
, ncpy
);
641 static int fuse_check_page(struct page
*page
)
643 if (page_mapcount(page
) ||
644 page
->mapping
!= NULL
||
645 page_count(page
) != 1 ||
646 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
653 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
654 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
);
660 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
663 struct page
*oldpage
= *pagep
;
664 struct page
*newpage
;
665 struct pipe_buffer
*buf
= cs
->pipebufs
;
666 struct address_space
*mapping
;
669 unlock_request(cs
->fc
, cs
->req
);
670 fuse_copy_finish(cs
);
672 err
= buf
->ops
->confirm(cs
->pipe
, buf
);
676 BUG_ON(!cs
->nr_segs
);
682 if (cs
->len
!= PAGE_SIZE
)
685 if (buf
->ops
->steal(cs
->pipe
, buf
) != 0)
690 if (WARN_ON(!PageUptodate(newpage
)))
693 ClearPageMappedToDisk(newpage
);
695 if (fuse_check_page(newpage
) != 0)
696 goto out_fallback_unlock
;
698 mapping
= oldpage
->mapping
;
699 index
= oldpage
->index
;
702 * This is a new and locked page, it shouldn't be mapped or
703 * have any special flags on it
705 if (WARN_ON(page_mapped(oldpage
)))
706 goto out_fallback_unlock
;
707 if (WARN_ON(page_has_private(oldpage
)))
708 goto out_fallback_unlock
;
709 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
710 goto out_fallback_unlock
;
711 if (WARN_ON(PageMlocked(oldpage
)))
712 goto out_fallback_unlock
;
714 remove_from_page_cache(oldpage
);
715 page_cache_release(oldpage
);
717 err
= add_to_page_cache_locked(newpage
, mapping
, index
, GFP_KERNEL
);
719 printk(KERN_WARNING
"fuse_try_move_page: failed to add page");
720 goto out_fallback_unlock
;
722 page_cache_get(newpage
);
724 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
725 lru_cache_add_file(newpage
);
728 spin_lock(&cs
->fc
->lock
);
729 if (cs
->req
->aborted
)
733 spin_unlock(&cs
->fc
->lock
);
736 unlock_page(newpage
);
737 page_cache_release(newpage
);
741 unlock_page(oldpage
);
742 page_cache_release(oldpage
);
748 unlock_page(newpage
);
750 cs
->mapaddr
= buf
->ops
->map(cs
->pipe
, buf
, 1);
751 cs
->buf
= cs
->mapaddr
+ buf
->offset
;
753 err
= lock_request(cs
->fc
, cs
->req
);
760 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
761 unsigned offset
, unsigned count
)
763 struct pipe_buffer
*buf
;
765 if (cs
->nr_segs
== cs
->pipe
->buffers
)
768 unlock_request(cs
->fc
, cs
->req
);
769 fuse_copy_finish(cs
);
772 page_cache_get(page
);
774 buf
->offset
= offset
;
785 * Copy a page in the request to/from the userspace buffer. Must be
788 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
789 unsigned offset
, unsigned count
, int zeroing
)
792 struct page
*page
= *pagep
;
794 if (page
&& zeroing
&& count
< PAGE_SIZE
) {
795 void *mapaddr
= kmap_atomic(page
, KM_USER1
);
796 memset(mapaddr
, 0, PAGE_SIZE
);
797 kunmap_atomic(mapaddr
, KM_USER1
);
800 if (cs
->write
&& cs
->pipebufs
&& page
) {
801 return fuse_ref_page(cs
, page
, offset
, count
);
802 } else if (!cs
->len
) {
803 if (cs
->move_pages
&& page
&&
804 offset
== 0 && count
== PAGE_SIZE
) {
805 err
= fuse_try_move_page(cs
, pagep
);
809 err
= fuse_copy_fill(cs
);
815 void *mapaddr
= kmap_atomic(page
, KM_USER1
);
816 void *buf
= mapaddr
+ offset
;
817 offset
+= fuse_copy_do(cs
, &buf
, &count
);
818 kunmap_atomic(mapaddr
, KM_USER1
);
820 offset
+= fuse_copy_do(cs
, NULL
, &count
);
822 if (page
&& !cs
->write
)
823 flush_dcache_page(page
);
827 /* Copy pages in the request to/from userspace buffer */
828 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
832 struct fuse_req
*req
= cs
->req
;
833 unsigned offset
= req
->page_offset
;
834 unsigned count
= min(nbytes
, (unsigned) PAGE_SIZE
- offset
);
836 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
839 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
845 count
= min(nbytes
, (unsigned) PAGE_SIZE
);
851 /* Copy a single argument in the request to/from userspace buffer */
852 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
856 int err
= fuse_copy_fill(cs
);
860 fuse_copy_do(cs
, &val
, &size
);
865 /* Copy request arguments to/from userspace buffer */
866 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
867 unsigned argpages
, struct fuse_arg
*args
,
873 for (i
= 0; !err
&& i
< numargs
; i
++) {
874 struct fuse_arg
*arg
= &args
[i
];
875 if (i
== numargs
- 1 && argpages
)
876 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
878 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
883 static int request_pending(struct fuse_conn
*fc
)
885 return !list_empty(&fc
->pending
) || !list_empty(&fc
->interrupts
);
888 /* Wait until a request is available on the pending list */
889 static void request_wait(struct fuse_conn
*fc
)
890 __releases(&fc
->lock
)
891 __acquires(&fc
->lock
)
893 DECLARE_WAITQUEUE(wait
, current
);
895 add_wait_queue_exclusive(&fc
->waitq
, &wait
);
896 while (fc
->connected
&& !request_pending(fc
)) {
897 set_current_state(TASK_INTERRUPTIBLE
);
898 if (signal_pending(current
))
901 spin_unlock(&fc
->lock
);
903 spin_lock(&fc
->lock
);
905 set_current_state(TASK_RUNNING
);
906 remove_wait_queue(&fc
->waitq
, &wait
);
910 * Transfer an interrupt request to userspace
912 * Unlike other requests this is assembled on demand, without a need
913 * to allocate a separate fuse_req structure.
915 * Called with fc->lock held, releases it
917 static int fuse_read_interrupt(struct fuse_conn
*fc
, struct fuse_copy_state
*cs
,
918 size_t nbytes
, struct fuse_req
*req
)
919 __releases(&fc
->lock
)
921 struct fuse_in_header ih
;
922 struct fuse_interrupt_in arg
;
923 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
926 list_del_init(&req
->intr_entry
);
927 req
->intr_unique
= fuse_get_unique(fc
);
928 memset(&ih
, 0, sizeof(ih
));
929 memset(&arg
, 0, sizeof(arg
));
931 ih
.opcode
= FUSE_INTERRUPT
;
932 ih
.unique
= req
->intr_unique
;
933 arg
.unique
= req
->in
.h
.unique
;
935 spin_unlock(&fc
->lock
);
936 if (nbytes
< reqsize
)
939 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
941 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
942 fuse_copy_finish(cs
);
944 return err
? err
: reqsize
;
948 * Read a single request into the userspace filesystem's buffer. This
949 * function waits until a request is available, then removes it from
950 * the pending list and copies request data to userspace buffer. If
951 * no reply is needed (FORGET) or request has been aborted or there
952 * was an error during the copying then it's finished by calling
953 * request_end(). Otherwise add it to the processing list, and set
956 static ssize_t
fuse_dev_do_read(struct fuse_conn
*fc
, struct file
*file
,
957 struct fuse_copy_state
*cs
, size_t nbytes
)
960 struct fuse_req
*req
;
965 spin_lock(&fc
->lock
);
967 if ((file
->f_flags
& O_NONBLOCK
) && fc
->connected
&&
968 !request_pending(fc
))
976 if (!request_pending(fc
))
979 if (!list_empty(&fc
->interrupts
)) {
980 req
= list_entry(fc
->interrupts
.next
, struct fuse_req
,
982 return fuse_read_interrupt(fc
, cs
, nbytes
, req
);
985 req
= list_entry(fc
->pending
.next
, struct fuse_req
, list
);
986 req
->state
= FUSE_REQ_READING
;
987 list_move(&req
->list
, &fc
->io
);
991 /* If request is too large, reply with an error and restart the read */
992 if (nbytes
< reqsize
) {
993 req
->out
.h
.error
= -EIO
;
994 /* SETXATTR is special, since it may contain too large data */
995 if (in
->h
.opcode
== FUSE_SETXATTR
)
996 req
->out
.h
.error
= -E2BIG
;
997 request_end(fc
, req
);
1000 spin_unlock(&fc
->lock
);
1002 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1004 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1005 (struct fuse_arg
*) in
->args
, 0);
1006 fuse_copy_finish(cs
);
1007 spin_lock(&fc
->lock
);
1010 request_end(fc
, req
);
1014 req
->out
.h
.error
= -EIO
;
1015 request_end(fc
, req
);
1019 request_end(fc
, req
);
1021 req
->state
= FUSE_REQ_SENT
;
1022 list_move_tail(&req
->list
, &fc
->processing
);
1023 if (req
->interrupted
)
1024 queue_interrupt(fc
, req
);
1025 spin_unlock(&fc
->lock
);
1030 spin_unlock(&fc
->lock
);
1034 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1035 unsigned long nr_segs
, loff_t pos
)
1037 struct fuse_copy_state cs
;
1038 struct file
*file
= iocb
->ki_filp
;
1039 struct fuse_conn
*fc
= fuse_get_conn(file
);
1043 fuse_copy_init(&cs
, fc
, 1, iov
, nr_segs
);
1045 return fuse_dev_do_read(fc
, file
, &cs
, iov_length(iov
, nr_segs
));
1048 static int fuse_dev_pipe_buf_steal(struct pipe_inode_info
*pipe
,
1049 struct pipe_buffer
*buf
)
1054 static const struct pipe_buf_operations fuse_dev_pipe_buf_ops
= {
1056 .map
= generic_pipe_buf_map
,
1057 .unmap
= generic_pipe_buf_unmap
,
1058 .confirm
= generic_pipe_buf_confirm
,
1059 .release
= generic_pipe_buf_release
,
1060 .steal
= fuse_dev_pipe_buf_steal
,
1061 .get
= generic_pipe_buf_get
,
1064 static ssize_t
fuse_dev_splice_read(struct file
*in
, loff_t
*ppos
,
1065 struct pipe_inode_info
*pipe
,
1066 size_t len
, unsigned int flags
)
1071 struct pipe_buffer
*bufs
;
1072 struct fuse_copy_state cs
;
1073 struct fuse_conn
*fc
= fuse_get_conn(in
);
1077 bufs
= kmalloc(pipe
->buffers
* sizeof (struct pipe_buffer
), GFP_KERNEL
);
1081 fuse_copy_init(&cs
, fc
, 1, NULL
, 0);
1084 ret
= fuse_dev_do_read(fc
, in
, &cs
, len
);
1091 if (!pipe
->readers
) {
1092 send_sig(SIGPIPE
, current
, 0);
1098 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1103 while (page_nr
< cs
.nr_segs
) {
1104 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
1105 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
1107 buf
->page
= bufs
[page_nr
].page
;
1108 buf
->offset
= bufs
[page_nr
].offset
;
1109 buf
->len
= bufs
[page_nr
].len
;
1110 buf
->ops
= &fuse_dev_pipe_buf_ops
;
1125 if (waitqueue_active(&pipe
->wait
))
1126 wake_up_interruptible(&pipe
->wait
);
1127 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
1131 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1132 page_cache_release(bufs
[page_nr
].page
);
1138 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1139 struct fuse_copy_state
*cs
)
1141 struct fuse_notify_poll_wakeup_out outarg
;
1144 if (size
!= sizeof(outarg
))
1147 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1151 fuse_copy_finish(cs
);
1152 return fuse_notify_poll_wakeup(fc
, &outarg
);
1155 fuse_copy_finish(cs
);
1159 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1160 struct fuse_copy_state
*cs
)
1162 struct fuse_notify_inval_inode_out outarg
;
1165 if (size
!= sizeof(outarg
))
1168 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1171 fuse_copy_finish(cs
);
1173 down_read(&fc
->killsb
);
1176 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1177 outarg
.off
, outarg
.len
);
1179 up_read(&fc
->killsb
);
1183 fuse_copy_finish(cs
);
1187 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1188 struct fuse_copy_state
*cs
)
1190 struct fuse_notify_inval_entry_out outarg
;
1195 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1200 if (size
< sizeof(outarg
))
1203 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1207 err
= -ENAMETOOLONG
;
1208 if (outarg
.namelen
> FUSE_NAME_MAX
)
1212 name
.len
= outarg
.namelen
;
1213 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1216 fuse_copy_finish(cs
);
1217 buf
[outarg
.namelen
] = 0;
1218 name
.hash
= full_name_hash(name
.name
, name
.len
);
1220 down_read(&fc
->killsb
);
1223 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, &name
);
1224 up_read(&fc
->killsb
);
1230 fuse_copy_finish(cs
);
1234 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1235 unsigned int size
, struct fuse_copy_state
*cs
)
1238 case FUSE_NOTIFY_POLL
:
1239 return fuse_notify_poll(fc
, size
, cs
);
1241 case FUSE_NOTIFY_INVAL_INODE
:
1242 return fuse_notify_inval_inode(fc
, size
, cs
);
1244 case FUSE_NOTIFY_INVAL_ENTRY
:
1245 return fuse_notify_inval_entry(fc
, size
, cs
);
1248 fuse_copy_finish(cs
);
1253 /* Look up request on processing list by unique ID */
1254 static struct fuse_req
*request_find(struct fuse_conn
*fc
, u64 unique
)
1256 struct list_head
*entry
;
1258 list_for_each(entry
, &fc
->processing
) {
1259 struct fuse_req
*req
;
1260 req
= list_entry(entry
, struct fuse_req
, list
);
1261 if (req
->in
.h
.unique
== unique
|| req
->intr_unique
== unique
)
1267 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1270 unsigned reqsize
= sizeof(struct fuse_out_header
);
1273 return nbytes
!= reqsize
? -EINVAL
: 0;
1275 reqsize
+= len_args(out
->numargs
, out
->args
);
1277 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1279 else if (reqsize
> nbytes
) {
1280 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1281 unsigned diffsize
= reqsize
- nbytes
;
1282 if (diffsize
> lastarg
->size
)
1284 lastarg
->size
-= diffsize
;
1286 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1291 * Write a single reply to a request. First the header is copied from
1292 * the write buffer. The request is then searched on the processing
1293 * list by the unique ID found in the header. If found, then remove
1294 * it from the list and copy the rest of the buffer to the request.
1295 * The request is finished by calling request_end()
1297 static ssize_t
fuse_dev_do_write(struct fuse_conn
*fc
,
1298 struct fuse_copy_state
*cs
, size_t nbytes
)
1301 struct fuse_req
*req
;
1302 struct fuse_out_header oh
;
1304 if (nbytes
< sizeof(struct fuse_out_header
))
1307 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1312 if (oh
.len
!= nbytes
)
1316 * Zero oh.unique indicates unsolicited notification message
1317 * and error contains notification code.
1320 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1321 return err
? err
: nbytes
;
1325 if (oh
.error
<= -1000 || oh
.error
> 0)
1328 spin_lock(&fc
->lock
);
1333 req
= request_find(fc
, oh
.unique
);
1338 spin_unlock(&fc
->lock
);
1339 fuse_copy_finish(cs
);
1340 spin_lock(&fc
->lock
);
1341 request_end(fc
, req
);
1344 /* Is it an interrupt reply? */
1345 if (req
->intr_unique
== oh
.unique
) {
1347 if (nbytes
!= sizeof(struct fuse_out_header
))
1350 if (oh
.error
== -ENOSYS
)
1351 fc
->no_interrupt
= 1;
1352 else if (oh
.error
== -EAGAIN
)
1353 queue_interrupt(fc
, req
);
1355 spin_unlock(&fc
->lock
);
1356 fuse_copy_finish(cs
);
1360 req
->state
= FUSE_REQ_WRITING
;
1361 list_move(&req
->list
, &fc
->io
);
1365 if (!req
->out
.page_replace
)
1367 spin_unlock(&fc
->lock
);
1369 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1370 fuse_copy_finish(cs
);
1372 spin_lock(&fc
->lock
);
1377 } else if (!req
->aborted
)
1378 req
->out
.h
.error
= -EIO
;
1379 request_end(fc
, req
);
1381 return err
? err
: nbytes
;
1384 spin_unlock(&fc
->lock
);
1386 fuse_copy_finish(cs
);
1390 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1391 unsigned long nr_segs
, loff_t pos
)
1393 struct fuse_copy_state cs
;
1394 struct fuse_conn
*fc
= fuse_get_conn(iocb
->ki_filp
);
1398 fuse_copy_init(&cs
, fc
, 0, iov
, nr_segs
);
1400 return fuse_dev_do_write(fc
, &cs
, iov_length(iov
, nr_segs
));
1403 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1404 struct file
*out
, loff_t
*ppos
,
1405 size_t len
, unsigned int flags
)
1409 struct pipe_buffer
*bufs
;
1410 struct fuse_copy_state cs
;
1411 struct fuse_conn
*fc
;
1415 fc
= fuse_get_conn(out
);
1419 bufs
= kmalloc(pipe
->buffers
* sizeof (struct pipe_buffer
), GFP_KERNEL
);
1426 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
1427 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
1437 struct pipe_buffer
*ibuf
;
1438 struct pipe_buffer
*obuf
;
1440 BUG_ON(nbuf
>= pipe
->buffers
);
1441 BUG_ON(!pipe
->nrbufs
);
1442 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
1445 if (rem
>= ibuf
->len
) {
1448 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
1451 ibuf
->ops
->get(pipe
, ibuf
);
1453 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1455 ibuf
->offset
+= obuf
->len
;
1456 ibuf
->len
-= obuf
->len
;
1463 fuse_copy_init(&cs
, fc
, 0, NULL
, nbuf
);
1467 if (flags
& SPLICE_F_MOVE
)
1470 ret
= fuse_dev_do_write(fc
, &cs
, len
);
1472 for (idx
= 0; idx
< nbuf
; idx
++) {
1473 struct pipe_buffer
*buf
= &bufs
[idx
];
1474 buf
->ops
->release(pipe
, buf
);
1481 static unsigned fuse_dev_poll(struct file
*file
, poll_table
*wait
)
1483 unsigned mask
= POLLOUT
| POLLWRNORM
;
1484 struct fuse_conn
*fc
= fuse_get_conn(file
);
1488 poll_wait(file
, &fc
->waitq
, wait
);
1490 spin_lock(&fc
->lock
);
1493 else if (request_pending(fc
))
1494 mask
|= POLLIN
| POLLRDNORM
;
1495 spin_unlock(&fc
->lock
);
1501 * Abort all requests on the given list (pending or processing)
1503 * This function releases and reacquires fc->lock
1505 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
1506 __releases(&fc
->lock
)
1507 __acquires(&fc
->lock
)
1509 while (!list_empty(head
)) {
1510 struct fuse_req
*req
;
1511 req
= list_entry(head
->next
, struct fuse_req
, list
);
1512 req
->out
.h
.error
= -ECONNABORTED
;
1513 request_end(fc
, req
);
1514 spin_lock(&fc
->lock
);
1519 * Abort requests under I/O
1521 * The requests are set to aborted and finished, and the request
1522 * waiter is woken up. This will make request_wait_answer() wait
1523 * until the request is unlocked and then return.
1525 * If the request is asynchronous, then the end function needs to be
1526 * called after waiting for the request to be unlocked (if it was
1529 static void end_io_requests(struct fuse_conn
*fc
)
1530 __releases(&fc
->lock
)
1531 __acquires(&fc
->lock
)
1533 while (!list_empty(&fc
->io
)) {
1534 struct fuse_req
*req
=
1535 list_entry(fc
->io
.next
, struct fuse_req
, list
);
1536 void (*end
) (struct fuse_conn
*, struct fuse_req
*) = req
->end
;
1539 req
->out
.h
.error
= -ECONNABORTED
;
1540 req
->state
= FUSE_REQ_FINISHED
;
1541 list_del_init(&req
->list
);
1542 wake_up(&req
->waitq
);
1545 __fuse_get_request(req
);
1546 spin_unlock(&fc
->lock
);
1547 wait_event(req
->waitq
, !req
->locked
);
1549 fuse_put_request(fc
, req
);
1550 spin_lock(&fc
->lock
);
1556 * Abort all requests.
1558 * Emergency exit in case of a malicious or accidental deadlock, or
1559 * just a hung filesystem.
1561 * The same effect is usually achievable through killing the
1562 * filesystem daemon and all users of the filesystem. The exception
1563 * is the combination of an asynchronous request and the tricky
1564 * deadlock (see Documentation/filesystems/fuse.txt).
1566 * During the aborting, progression of requests from the pending and
1567 * processing lists onto the io list, and progression of new requests
1568 * onto the pending list is prevented by req->connected being false.
1570 * Progression of requests under I/O to the processing list is
1571 * prevented by the req->aborted flag being true for these requests.
1572 * For this reason requests on the io list must be aborted first.
1574 void fuse_abort_conn(struct fuse_conn
*fc
)
1576 spin_lock(&fc
->lock
);
1577 if (fc
->connected
) {
1580 end_io_requests(fc
);
1581 end_requests(fc
, &fc
->pending
);
1582 end_requests(fc
, &fc
->processing
);
1583 wake_up_all(&fc
->waitq
);
1584 wake_up_all(&fc
->blocked_waitq
);
1585 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
1587 spin_unlock(&fc
->lock
);
1589 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
1591 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
1593 struct fuse_conn
*fc
= fuse_get_conn(file
);
1595 spin_lock(&fc
->lock
);
1597 end_requests(fc
, &fc
->pending
);
1598 end_requests(fc
, &fc
->processing
);
1599 spin_unlock(&fc
->lock
);
1605 EXPORT_SYMBOL_GPL(fuse_dev_release
);
1607 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
1609 struct fuse_conn
*fc
= fuse_get_conn(file
);
1613 /* No locking - fasync_helper does its own locking */
1614 return fasync_helper(fd
, file
, on
, &fc
->fasync
);
1617 const struct file_operations fuse_dev_operations
= {
1618 .owner
= THIS_MODULE
,
1619 .llseek
= no_llseek
,
1620 .read
= do_sync_read
,
1621 .aio_read
= fuse_dev_read
,
1622 .splice_read
= fuse_dev_splice_read
,
1623 .write
= do_sync_write
,
1624 .aio_write
= fuse_dev_write
,
1625 .splice_write
= fuse_dev_splice_write
,
1626 .poll
= fuse_dev_poll
,
1627 .release
= fuse_dev_release
,
1628 .fasync
= fuse_dev_fasync
,
1630 EXPORT_SYMBOL_GPL(fuse_dev_operations
);
1632 static struct miscdevice fuse_miscdevice
= {
1633 .minor
= FUSE_MINOR
,
1635 .fops
= &fuse_dev_operations
,
1638 int __init
fuse_dev_init(void)
1641 fuse_req_cachep
= kmem_cache_create("fuse_request",
1642 sizeof(struct fuse_req
),
1644 if (!fuse_req_cachep
)
1647 err
= misc_register(&fuse_miscdevice
);
1649 goto out_cache_clean
;
1654 kmem_cache_destroy(fuse_req_cachep
);
1659 void fuse_dev_cleanup(void)
1661 misc_deregister(&fuse_miscdevice
);
1662 kmem_cache_destroy(fuse_req_cachep
);