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
.len
= sizeof(struct fuse_in_header
) +
243 len_args(req
->in
.numargs
, (struct fuse_arg
*) req
->in
.args
);
244 list_add_tail(&req
->list
, &fc
->pending
);
245 req
->state
= FUSE_REQ_PENDING
;
248 atomic_inc(&fc
->num_waiting
);
251 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
254 static void flush_bg_queue(struct fuse_conn
*fc
)
256 while (fc
->active_background
< fc
->max_background
&&
257 !list_empty(&fc
->bg_queue
)) {
258 struct fuse_req
*req
;
260 req
= list_entry(fc
->bg_queue
.next
, struct fuse_req
, list
);
261 list_del(&req
->list
);
262 fc
->active_background
++;
263 req
->in
.h
.unique
= fuse_get_unique(fc
);
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
)
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
)
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
)
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 req
->in
.h
.unique
= fuse_get_unique(fc
);
402 queue_request(fc
, req
);
403 /* acquire extra reference, since request is still needed
404 after request_end() */
405 __fuse_get_request(req
);
407 request_wait_answer(fc
, req
);
409 spin_unlock(&fc
->lock
);
411 EXPORT_SYMBOL_GPL(fuse_request_send
);
413 static void fuse_request_send_nowait_locked(struct fuse_conn
*fc
,
414 struct fuse_req
*req
)
417 fc
->num_background
++;
418 if (fc
->num_background
== fc
->max_background
)
420 if (fc
->num_background
== fc
->congestion_threshold
&&
421 fc
->bdi_initialized
) {
422 set_bdi_congested(&fc
->bdi
, BLK_RW_SYNC
);
423 set_bdi_congested(&fc
->bdi
, BLK_RW_ASYNC
);
425 list_add_tail(&req
->list
, &fc
->bg_queue
);
429 static void fuse_request_send_nowait(struct fuse_conn
*fc
, struct fuse_req
*req
)
431 spin_lock(&fc
->lock
);
433 fuse_request_send_nowait_locked(fc
, req
);
434 spin_unlock(&fc
->lock
);
436 req
->out
.h
.error
= -ENOTCONN
;
437 request_end(fc
, req
);
441 void fuse_request_send_noreply(struct fuse_conn
*fc
, struct fuse_req
*req
)
444 fuse_request_send_nowait(fc
, req
);
447 void fuse_request_send_background(struct fuse_conn
*fc
, struct fuse_req
*req
)
450 fuse_request_send_nowait(fc
, req
);
452 EXPORT_SYMBOL_GPL(fuse_request_send_background
);
454 static int fuse_request_send_notify_reply(struct fuse_conn
*fc
,
455 struct fuse_req
*req
, u64 unique
)
460 req
->in
.h
.unique
= unique
;
461 spin_lock(&fc
->lock
);
463 queue_request(fc
, req
);
466 spin_unlock(&fc
->lock
);
472 * Called under fc->lock
474 * fc->connected must have been checked previously
476 void fuse_request_send_background_locked(struct fuse_conn
*fc
,
477 struct fuse_req
*req
)
480 fuse_request_send_nowait_locked(fc
, req
);
484 * Lock the request. Up to the next unlock_request() there mustn't be
485 * anything that could cause a page-fault. If the request was already
488 static int lock_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
492 spin_lock(&fc
->lock
);
497 spin_unlock(&fc
->lock
);
503 * Unlock request. If it was aborted during being locked, the
504 * requester thread is currently waiting for it to be unlocked, so
507 static void unlock_request(struct fuse_conn
*fc
, struct fuse_req
*req
)
510 spin_lock(&fc
->lock
);
513 wake_up(&req
->waitq
);
514 spin_unlock(&fc
->lock
);
518 struct fuse_copy_state
{
519 struct fuse_conn
*fc
;
521 struct fuse_req
*req
;
522 const struct iovec
*iov
;
523 struct pipe_buffer
*pipebufs
;
524 struct pipe_buffer
*currbuf
;
525 struct pipe_inode_info
*pipe
;
526 unsigned long nr_segs
;
527 unsigned long seglen
;
533 unsigned move_pages
:1;
536 static void fuse_copy_init(struct fuse_copy_state
*cs
, struct fuse_conn
*fc
,
538 const struct iovec
*iov
, unsigned long nr_segs
)
540 memset(cs
, 0, sizeof(*cs
));
544 cs
->nr_segs
= nr_segs
;
547 /* Unmap and put previous page of userspace buffer */
548 static void fuse_copy_finish(struct fuse_copy_state
*cs
)
551 struct pipe_buffer
*buf
= cs
->currbuf
;
554 buf
->ops
->unmap(cs
->pipe
, buf
, cs
->mapaddr
);
557 buf
->len
= PAGE_SIZE
- cs
->len
;
561 } else if (cs
->mapaddr
) {
564 flush_dcache_page(cs
->pg
);
565 set_page_dirty_lock(cs
->pg
);
573 * Get another pagefull of userspace buffer, and map it to kernel
574 * address space, and lock request
576 static int fuse_copy_fill(struct fuse_copy_state
*cs
)
578 unsigned long offset
;
581 unlock_request(cs
->fc
, cs
->req
);
582 fuse_copy_finish(cs
);
584 struct pipe_buffer
*buf
= cs
->pipebufs
;
587 err
= buf
->ops
->confirm(cs
->pipe
, buf
);
591 BUG_ON(!cs
->nr_segs
);
593 cs
->mapaddr
= buf
->ops
->map(cs
->pipe
, buf
, 0);
595 cs
->buf
= cs
->mapaddr
+ buf
->offset
;
601 if (cs
->nr_segs
== cs
->pipe
->buffers
)
604 page
= alloc_page(GFP_HIGHUSER
);
613 cs
->mapaddr
= kmap(page
);
614 cs
->buf
= cs
->mapaddr
;
621 BUG_ON(!cs
->nr_segs
);
622 cs
->seglen
= cs
->iov
[0].iov_len
;
623 cs
->addr
= (unsigned long) cs
->iov
[0].iov_base
;
627 err
= get_user_pages_fast(cs
->addr
, 1, cs
->write
, &cs
->pg
);
631 offset
= cs
->addr
% PAGE_SIZE
;
632 cs
->mapaddr
= kmap(cs
->pg
);
633 cs
->buf
= cs
->mapaddr
+ offset
;
634 cs
->len
= min(PAGE_SIZE
- offset
, cs
->seglen
);
635 cs
->seglen
-= cs
->len
;
639 return lock_request(cs
->fc
, cs
->req
);
642 /* Do as much copy to/from userspace buffer as we can */
643 static int fuse_copy_do(struct fuse_copy_state
*cs
, void **val
, unsigned *size
)
645 unsigned ncpy
= min(*size
, cs
->len
);
648 memcpy(cs
->buf
, *val
, ncpy
);
650 memcpy(*val
, cs
->buf
, ncpy
);
659 static int fuse_check_page(struct page
*page
)
661 if (page_mapcount(page
) ||
662 page
->mapping
!= NULL
||
663 page_count(page
) != 1 ||
664 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
&
671 printk(KERN_WARNING
"fuse: trying to steal weird page\n");
672 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
);
678 static int fuse_try_move_page(struct fuse_copy_state
*cs
, struct page
**pagep
)
681 struct page
*oldpage
= *pagep
;
682 struct page
*newpage
;
683 struct pipe_buffer
*buf
= cs
->pipebufs
;
684 struct address_space
*mapping
;
687 unlock_request(cs
->fc
, cs
->req
);
688 fuse_copy_finish(cs
);
690 err
= buf
->ops
->confirm(cs
->pipe
, buf
);
694 BUG_ON(!cs
->nr_segs
);
700 if (cs
->len
!= PAGE_SIZE
)
703 if (buf
->ops
->steal(cs
->pipe
, buf
) != 0)
708 if (WARN_ON(!PageUptodate(newpage
)))
711 ClearPageMappedToDisk(newpage
);
713 if (fuse_check_page(newpage
) != 0)
714 goto out_fallback_unlock
;
716 mapping
= oldpage
->mapping
;
717 index
= oldpage
->index
;
720 * This is a new and locked page, it shouldn't be mapped or
721 * have any special flags on it
723 if (WARN_ON(page_mapped(oldpage
)))
724 goto out_fallback_unlock
;
725 if (WARN_ON(page_has_private(oldpage
)))
726 goto out_fallback_unlock
;
727 if (WARN_ON(PageDirty(oldpage
) || PageWriteback(oldpage
)))
728 goto out_fallback_unlock
;
729 if (WARN_ON(PageMlocked(oldpage
)))
730 goto out_fallback_unlock
;
732 remove_from_page_cache(oldpage
);
733 page_cache_release(oldpage
);
735 err
= add_to_page_cache_locked(newpage
, mapping
, index
, GFP_KERNEL
);
737 printk(KERN_WARNING
"fuse_try_move_page: failed to add page");
738 goto out_fallback_unlock
;
740 page_cache_get(newpage
);
742 if (!(buf
->flags
& PIPE_BUF_FLAG_LRU
))
743 lru_cache_add_file(newpage
);
746 spin_lock(&cs
->fc
->lock
);
747 if (cs
->req
->aborted
)
751 spin_unlock(&cs
->fc
->lock
);
754 unlock_page(newpage
);
755 page_cache_release(newpage
);
759 unlock_page(oldpage
);
760 page_cache_release(oldpage
);
766 unlock_page(newpage
);
768 cs
->mapaddr
= buf
->ops
->map(cs
->pipe
, buf
, 1);
769 cs
->buf
= cs
->mapaddr
+ buf
->offset
;
771 err
= lock_request(cs
->fc
, cs
->req
);
778 static int fuse_ref_page(struct fuse_copy_state
*cs
, struct page
*page
,
779 unsigned offset
, unsigned count
)
781 struct pipe_buffer
*buf
;
783 if (cs
->nr_segs
== cs
->pipe
->buffers
)
786 unlock_request(cs
->fc
, cs
->req
);
787 fuse_copy_finish(cs
);
790 page_cache_get(page
);
792 buf
->offset
= offset
;
803 * Copy a page in the request to/from the userspace buffer. Must be
806 static int fuse_copy_page(struct fuse_copy_state
*cs
, struct page
**pagep
,
807 unsigned offset
, unsigned count
, int zeroing
)
810 struct page
*page
= *pagep
;
812 if (page
&& zeroing
&& count
< PAGE_SIZE
)
813 clear_highpage(page
);
816 if (cs
->write
&& cs
->pipebufs
&& page
) {
817 return fuse_ref_page(cs
, page
, offset
, count
);
818 } else if (!cs
->len
) {
819 if (cs
->move_pages
&& page
&&
820 offset
== 0 && count
== PAGE_SIZE
) {
821 err
= fuse_try_move_page(cs
, pagep
);
825 err
= fuse_copy_fill(cs
);
831 void *mapaddr
= kmap_atomic(page
, KM_USER0
);
832 void *buf
= mapaddr
+ offset
;
833 offset
+= fuse_copy_do(cs
, &buf
, &count
);
834 kunmap_atomic(mapaddr
, KM_USER0
);
836 offset
+= fuse_copy_do(cs
, NULL
, &count
);
838 if (page
&& !cs
->write
)
839 flush_dcache_page(page
);
843 /* Copy pages in the request to/from userspace buffer */
844 static int fuse_copy_pages(struct fuse_copy_state
*cs
, unsigned nbytes
,
848 struct fuse_req
*req
= cs
->req
;
849 unsigned offset
= req
->page_offset
;
850 unsigned count
= min(nbytes
, (unsigned) PAGE_SIZE
- offset
);
852 for (i
= 0; i
< req
->num_pages
&& (nbytes
|| zeroing
); i
++) {
855 err
= fuse_copy_page(cs
, &req
->pages
[i
], offset
, count
,
861 count
= min(nbytes
, (unsigned) PAGE_SIZE
);
867 /* Copy a single argument in the request to/from userspace buffer */
868 static int fuse_copy_one(struct fuse_copy_state
*cs
, void *val
, unsigned size
)
872 int err
= fuse_copy_fill(cs
);
876 fuse_copy_do(cs
, &val
, &size
);
881 /* Copy request arguments to/from userspace buffer */
882 static int fuse_copy_args(struct fuse_copy_state
*cs
, unsigned numargs
,
883 unsigned argpages
, struct fuse_arg
*args
,
889 for (i
= 0; !err
&& i
< numargs
; i
++) {
890 struct fuse_arg
*arg
= &args
[i
];
891 if (i
== numargs
- 1 && argpages
)
892 err
= fuse_copy_pages(cs
, arg
->size
, zeroing
);
894 err
= fuse_copy_one(cs
, arg
->value
, arg
->size
);
899 static int request_pending(struct fuse_conn
*fc
)
901 return !list_empty(&fc
->pending
) || !list_empty(&fc
->interrupts
);
904 /* Wait until a request is available on the pending list */
905 static void request_wait(struct fuse_conn
*fc
)
909 DECLARE_WAITQUEUE(wait
, current
);
911 add_wait_queue_exclusive(&fc
->waitq
, &wait
);
912 while (fc
->connected
&& !request_pending(fc
)) {
913 set_current_state(TASK_INTERRUPTIBLE
);
914 if (signal_pending(current
))
917 spin_unlock(&fc
->lock
);
919 spin_lock(&fc
->lock
);
921 set_current_state(TASK_RUNNING
);
922 remove_wait_queue(&fc
->waitq
, &wait
);
926 * Transfer an interrupt request to userspace
928 * Unlike other requests this is assembled on demand, without a need
929 * to allocate a separate fuse_req structure.
931 * Called with fc->lock held, releases it
933 static int fuse_read_interrupt(struct fuse_conn
*fc
, struct fuse_copy_state
*cs
,
934 size_t nbytes
, struct fuse_req
*req
)
937 struct fuse_in_header ih
;
938 struct fuse_interrupt_in arg
;
939 unsigned reqsize
= sizeof(ih
) + sizeof(arg
);
942 list_del_init(&req
->intr_entry
);
943 req
->intr_unique
= fuse_get_unique(fc
);
944 memset(&ih
, 0, sizeof(ih
));
945 memset(&arg
, 0, sizeof(arg
));
947 ih
.opcode
= FUSE_INTERRUPT
;
948 ih
.unique
= req
->intr_unique
;
949 arg
.unique
= req
->in
.h
.unique
;
951 spin_unlock(&fc
->lock
);
952 if (nbytes
< reqsize
)
955 err
= fuse_copy_one(cs
, &ih
, sizeof(ih
));
957 err
= fuse_copy_one(cs
, &arg
, sizeof(arg
));
958 fuse_copy_finish(cs
);
960 return err
? err
: reqsize
;
964 * Read a single request into the userspace filesystem's buffer. This
965 * function waits until a request is available, then removes it from
966 * the pending list and copies request data to userspace buffer. If
967 * no reply is needed (FORGET) or request has been aborted or there
968 * was an error during the copying then it's finished by calling
969 * request_end(). Otherwise add it to the processing list, and set
972 static ssize_t
fuse_dev_do_read(struct fuse_conn
*fc
, struct file
*file
,
973 struct fuse_copy_state
*cs
, size_t nbytes
)
976 struct fuse_req
*req
;
981 spin_lock(&fc
->lock
);
983 if ((file
->f_flags
& O_NONBLOCK
) && fc
->connected
&&
984 !request_pending(fc
))
992 if (!request_pending(fc
))
995 if (!list_empty(&fc
->interrupts
)) {
996 req
= list_entry(fc
->interrupts
.next
, struct fuse_req
,
998 return fuse_read_interrupt(fc
, cs
, nbytes
, req
);
1001 req
= list_entry(fc
->pending
.next
, struct fuse_req
, list
);
1002 req
->state
= FUSE_REQ_READING
;
1003 list_move(&req
->list
, &fc
->io
);
1006 reqsize
= in
->h
.len
;
1007 /* If request is too large, reply with an error and restart the read */
1008 if (nbytes
< reqsize
) {
1009 req
->out
.h
.error
= -EIO
;
1010 /* SETXATTR is special, since it may contain too large data */
1011 if (in
->h
.opcode
== FUSE_SETXATTR
)
1012 req
->out
.h
.error
= -E2BIG
;
1013 request_end(fc
, req
);
1016 spin_unlock(&fc
->lock
);
1018 err
= fuse_copy_one(cs
, &in
->h
, sizeof(in
->h
));
1020 err
= fuse_copy_args(cs
, in
->numargs
, in
->argpages
,
1021 (struct fuse_arg
*) in
->args
, 0);
1022 fuse_copy_finish(cs
);
1023 spin_lock(&fc
->lock
);
1026 request_end(fc
, req
);
1030 req
->out
.h
.error
= -EIO
;
1031 request_end(fc
, req
);
1035 request_end(fc
, req
);
1037 req
->state
= FUSE_REQ_SENT
;
1038 list_move_tail(&req
->list
, &fc
->processing
);
1039 if (req
->interrupted
)
1040 queue_interrupt(fc
, req
);
1041 spin_unlock(&fc
->lock
);
1046 spin_unlock(&fc
->lock
);
1050 static ssize_t
fuse_dev_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1051 unsigned long nr_segs
, loff_t pos
)
1053 struct fuse_copy_state cs
;
1054 struct file
*file
= iocb
->ki_filp
;
1055 struct fuse_conn
*fc
= fuse_get_conn(file
);
1059 fuse_copy_init(&cs
, fc
, 1, iov
, nr_segs
);
1061 return fuse_dev_do_read(fc
, file
, &cs
, iov_length(iov
, nr_segs
));
1064 static int fuse_dev_pipe_buf_steal(struct pipe_inode_info
*pipe
,
1065 struct pipe_buffer
*buf
)
1070 static const struct pipe_buf_operations fuse_dev_pipe_buf_ops
= {
1072 .map
= generic_pipe_buf_map
,
1073 .unmap
= generic_pipe_buf_unmap
,
1074 .confirm
= generic_pipe_buf_confirm
,
1075 .release
= generic_pipe_buf_release
,
1076 .steal
= fuse_dev_pipe_buf_steal
,
1077 .get
= generic_pipe_buf_get
,
1080 static ssize_t
fuse_dev_splice_read(struct file
*in
, loff_t
*ppos
,
1081 struct pipe_inode_info
*pipe
,
1082 size_t len
, unsigned int flags
)
1087 struct pipe_buffer
*bufs
;
1088 struct fuse_copy_state cs
;
1089 struct fuse_conn
*fc
= fuse_get_conn(in
);
1093 bufs
= kmalloc(pipe
->buffers
* sizeof (struct pipe_buffer
), GFP_KERNEL
);
1097 fuse_copy_init(&cs
, fc
, 1, NULL
, 0);
1100 ret
= fuse_dev_do_read(fc
, in
, &cs
, len
);
1107 if (!pipe
->readers
) {
1108 send_sig(SIGPIPE
, current
, 0);
1114 if (pipe
->nrbufs
+ cs
.nr_segs
> pipe
->buffers
) {
1119 while (page_nr
< cs
.nr_segs
) {
1120 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
1121 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
1123 buf
->page
= bufs
[page_nr
].page
;
1124 buf
->offset
= bufs
[page_nr
].offset
;
1125 buf
->len
= bufs
[page_nr
].len
;
1126 buf
->ops
= &fuse_dev_pipe_buf_ops
;
1141 if (waitqueue_active(&pipe
->wait
))
1142 wake_up_interruptible(&pipe
->wait
);
1143 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
1147 for (; page_nr
< cs
.nr_segs
; page_nr
++)
1148 page_cache_release(bufs
[page_nr
].page
);
1154 static int fuse_notify_poll(struct fuse_conn
*fc
, unsigned int size
,
1155 struct fuse_copy_state
*cs
)
1157 struct fuse_notify_poll_wakeup_out outarg
;
1160 if (size
!= sizeof(outarg
))
1163 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1167 fuse_copy_finish(cs
);
1168 return fuse_notify_poll_wakeup(fc
, &outarg
);
1171 fuse_copy_finish(cs
);
1175 static int fuse_notify_inval_inode(struct fuse_conn
*fc
, unsigned int size
,
1176 struct fuse_copy_state
*cs
)
1178 struct fuse_notify_inval_inode_out outarg
;
1181 if (size
!= sizeof(outarg
))
1184 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1187 fuse_copy_finish(cs
);
1189 down_read(&fc
->killsb
);
1192 err
= fuse_reverse_inval_inode(fc
->sb
, outarg
.ino
,
1193 outarg
.off
, outarg
.len
);
1195 up_read(&fc
->killsb
);
1199 fuse_copy_finish(cs
);
1203 static int fuse_notify_inval_entry(struct fuse_conn
*fc
, unsigned int size
,
1204 struct fuse_copy_state
*cs
)
1206 struct fuse_notify_inval_entry_out outarg
;
1211 buf
= kzalloc(FUSE_NAME_MAX
+ 1, GFP_KERNEL
);
1216 if (size
< sizeof(outarg
))
1219 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1223 err
= -ENAMETOOLONG
;
1224 if (outarg
.namelen
> FUSE_NAME_MAX
)
1228 name
.len
= outarg
.namelen
;
1229 err
= fuse_copy_one(cs
, buf
, outarg
.namelen
+ 1);
1232 fuse_copy_finish(cs
);
1233 buf
[outarg
.namelen
] = 0;
1234 name
.hash
= full_name_hash(name
.name
, name
.len
);
1236 down_read(&fc
->killsb
);
1239 err
= fuse_reverse_inval_entry(fc
->sb
, outarg
.parent
, &name
);
1240 up_read(&fc
->killsb
);
1246 fuse_copy_finish(cs
);
1250 static int fuse_notify_store(struct fuse_conn
*fc
, unsigned int size
,
1251 struct fuse_copy_state
*cs
)
1253 struct fuse_notify_store_out outarg
;
1254 struct inode
*inode
;
1255 struct address_space
*mapping
;
1259 unsigned int offset
;
1265 if (size
< sizeof(outarg
))
1268 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1273 if (size
- sizeof(outarg
) != outarg
.size
)
1276 nodeid
= outarg
.nodeid
;
1278 down_read(&fc
->killsb
);
1284 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1288 mapping
= inode
->i_mapping
;
1289 index
= outarg
.offset
>> PAGE_CACHE_SHIFT
;
1290 offset
= outarg
.offset
& ~PAGE_CACHE_MASK
;
1291 file_size
= i_size_read(inode
);
1292 end
= outarg
.offset
+ outarg
.size
;
1293 if (end
> file_size
) {
1295 fuse_write_update_size(inode
, file_size
);
1301 unsigned int this_num
;
1304 page
= find_or_create_page(mapping
, index
,
1305 mapping_gfp_mask(mapping
));
1309 this_num
= min_t(unsigned, num
, PAGE_CACHE_SIZE
- offset
);
1310 err
= fuse_copy_page(cs
, &page
, offset
, this_num
, 0);
1311 if (!err
&& offset
== 0 && (num
!= 0 || file_size
== end
))
1312 SetPageUptodate(page
);
1314 page_cache_release(page
);
1329 up_read(&fc
->killsb
);
1331 fuse_copy_finish(cs
);
1335 static void fuse_retrieve_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1339 for (i
= 0; i
< req
->num_pages
; i
++) {
1340 struct page
*page
= req
->pages
[i
];
1341 page_cache_release(page
);
1345 static int fuse_retrieve(struct fuse_conn
*fc
, struct inode
*inode
,
1346 struct fuse_notify_retrieve_out
*outarg
)
1349 struct address_space
*mapping
= inode
->i_mapping
;
1350 struct fuse_req
*req
;
1354 unsigned int offset
;
1355 size_t total_len
= 0;
1357 req
= fuse_get_req(fc
);
1359 return PTR_ERR(req
);
1361 offset
= outarg
->offset
& ~PAGE_CACHE_MASK
;
1363 req
->in
.h
.opcode
= FUSE_NOTIFY_REPLY
;
1364 req
->in
.h
.nodeid
= outarg
->nodeid
;
1365 req
->in
.numargs
= 2;
1366 req
->in
.argpages
= 1;
1367 req
->page_offset
= offset
;
1368 req
->end
= fuse_retrieve_end
;
1370 index
= outarg
->offset
>> PAGE_CACHE_SHIFT
;
1371 file_size
= i_size_read(inode
);
1373 if (outarg
->offset
> file_size
)
1375 else if (outarg
->offset
+ num
> file_size
)
1376 num
= file_size
- outarg
->offset
;
1380 unsigned int this_num
;
1382 page
= find_get_page(mapping
, index
);
1386 this_num
= min_t(unsigned, num
, PAGE_CACHE_SIZE
- offset
);
1387 req
->pages
[req
->num_pages
] = page
;
1391 total_len
+= this_num
;
1393 req
->misc
.retrieve_in
.offset
= outarg
->offset
;
1394 req
->misc
.retrieve_in
.size
= total_len
;
1395 req
->in
.args
[0].size
= sizeof(req
->misc
.retrieve_in
);
1396 req
->in
.args
[0].value
= &req
->misc
.retrieve_in
;
1397 req
->in
.args
[1].size
= total_len
;
1399 err
= fuse_request_send_notify_reply(fc
, req
, outarg
->notify_unique
);
1401 fuse_retrieve_end(fc
, req
);
1406 static int fuse_notify_retrieve(struct fuse_conn
*fc
, unsigned int size
,
1407 struct fuse_copy_state
*cs
)
1409 struct fuse_notify_retrieve_out outarg
;
1410 struct inode
*inode
;
1414 if (size
!= sizeof(outarg
))
1417 err
= fuse_copy_one(cs
, &outarg
, sizeof(outarg
));
1421 fuse_copy_finish(cs
);
1423 down_read(&fc
->killsb
);
1426 u64 nodeid
= outarg
.nodeid
;
1428 inode
= ilookup5(fc
->sb
, nodeid
, fuse_inode_eq
, &nodeid
);
1430 err
= fuse_retrieve(fc
, inode
, &outarg
);
1434 up_read(&fc
->killsb
);
1439 fuse_copy_finish(cs
);
1443 static int fuse_notify(struct fuse_conn
*fc
, enum fuse_notify_code code
,
1444 unsigned int size
, struct fuse_copy_state
*cs
)
1447 case FUSE_NOTIFY_POLL
:
1448 return fuse_notify_poll(fc
, size
, cs
);
1450 case FUSE_NOTIFY_INVAL_INODE
:
1451 return fuse_notify_inval_inode(fc
, size
, cs
);
1453 case FUSE_NOTIFY_INVAL_ENTRY
:
1454 return fuse_notify_inval_entry(fc
, size
, cs
);
1456 case FUSE_NOTIFY_STORE
:
1457 return fuse_notify_store(fc
, size
, cs
);
1459 case FUSE_NOTIFY_RETRIEVE
:
1460 return fuse_notify_retrieve(fc
, size
, cs
);
1463 fuse_copy_finish(cs
);
1468 /* Look up request on processing list by unique ID */
1469 static struct fuse_req
*request_find(struct fuse_conn
*fc
, u64 unique
)
1471 struct list_head
*entry
;
1473 list_for_each(entry
, &fc
->processing
) {
1474 struct fuse_req
*req
;
1475 req
= list_entry(entry
, struct fuse_req
, list
);
1476 if (req
->in
.h
.unique
== unique
|| req
->intr_unique
== unique
)
1482 static int copy_out_args(struct fuse_copy_state
*cs
, struct fuse_out
*out
,
1485 unsigned reqsize
= sizeof(struct fuse_out_header
);
1488 return nbytes
!= reqsize
? -EINVAL
: 0;
1490 reqsize
+= len_args(out
->numargs
, out
->args
);
1492 if (reqsize
< nbytes
|| (reqsize
> nbytes
&& !out
->argvar
))
1494 else if (reqsize
> nbytes
) {
1495 struct fuse_arg
*lastarg
= &out
->args
[out
->numargs
-1];
1496 unsigned diffsize
= reqsize
- nbytes
;
1497 if (diffsize
> lastarg
->size
)
1499 lastarg
->size
-= diffsize
;
1501 return fuse_copy_args(cs
, out
->numargs
, out
->argpages
, out
->args
,
1506 * Write a single reply to a request. First the header is copied from
1507 * the write buffer. The request is then searched on the processing
1508 * list by the unique ID found in the header. If found, then remove
1509 * it from the list and copy the rest of the buffer to the request.
1510 * The request is finished by calling request_end()
1512 static ssize_t
fuse_dev_do_write(struct fuse_conn
*fc
,
1513 struct fuse_copy_state
*cs
, size_t nbytes
)
1516 struct fuse_req
*req
;
1517 struct fuse_out_header oh
;
1519 if (nbytes
< sizeof(struct fuse_out_header
))
1522 err
= fuse_copy_one(cs
, &oh
, sizeof(oh
));
1527 if (oh
.len
!= nbytes
)
1531 * Zero oh.unique indicates unsolicited notification message
1532 * and error contains notification code.
1535 err
= fuse_notify(fc
, oh
.error
, nbytes
- sizeof(oh
), cs
);
1536 return err
? err
: nbytes
;
1540 if (oh
.error
<= -1000 || oh
.error
> 0)
1543 spin_lock(&fc
->lock
);
1548 req
= request_find(fc
, oh
.unique
);
1553 spin_unlock(&fc
->lock
);
1554 fuse_copy_finish(cs
);
1555 spin_lock(&fc
->lock
);
1556 request_end(fc
, req
);
1559 /* Is it an interrupt reply? */
1560 if (req
->intr_unique
== oh
.unique
) {
1562 if (nbytes
!= sizeof(struct fuse_out_header
))
1565 if (oh
.error
== -ENOSYS
)
1566 fc
->no_interrupt
= 1;
1567 else if (oh
.error
== -EAGAIN
)
1568 queue_interrupt(fc
, req
);
1570 spin_unlock(&fc
->lock
);
1571 fuse_copy_finish(cs
);
1575 req
->state
= FUSE_REQ_WRITING
;
1576 list_move(&req
->list
, &fc
->io
);
1580 if (!req
->out
.page_replace
)
1582 spin_unlock(&fc
->lock
);
1584 err
= copy_out_args(cs
, &req
->out
, nbytes
);
1585 fuse_copy_finish(cs
);
1587 spin_lock(&fc
->lock
);
1592 } else if (!req
->aborted
)
1593 req
->out
.h
.error
= -EIO
;
1594 request_end(fc
, req
);
1596 return err
? err
: nbytes
;
1599 spin_unlock(&fc
->lock
);
1601 fuse_copy_finish(cs
);
1605 static ssize_t
fuse_dev_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1606 unsigned long nr_segs
, loff_t pos
)
1608 struct fuse_copy_state cs
;
1609 struct fuse_conn
*fc
= fuse_get_conn(iocb
->ki_filp
);
1613 fuse_copy_init(&cs
, fc
, 0, iov
, nr_segs
);
1615 return fuse_dev_do_write(fc
, &cs
, iov_length(iov
, nr_segs
));
1618 static ssize_t
fuse_dev_splice_write(struct pipe_inode_info
*pipe
,
1619 struct file
*out
, loff_t
*ppos
,
1620 size_t len
, unsigned int flags
)
1624 struct pipe_buffer
*bufs
;
1625 struct fuse_copy_state cs
;
1626 struct fuse_conn
*fc
;
1630 fc
= fuse_get_conn(out
);
1634 bufs
= kmalloc(pipe
->buffers
* sizeof (struct pipe_buffer
), GFP_KERNEL
);
1641 for (idx
= 0; idx
< pipe
->nrbufs
&& rem
< len
; idx
++)
1642 rem
+= pipe
->bufs
[(pipe
->curbuf
+ idx
) & (pipe
->buffers
- 1)].len
;
1652 struct pipe_buffer
*ibuf
;
1653 struct pipe_buffer
*obuf
;
1655 BUG_ON(nbuf
>= pipe
->buffers
);
1656 BUG_ON(!pipe
->nrbufs
);
1657 ibuf
= &pipe
->bufs
[pipe
->curbuf
];
1660 if (rem
>= ibuf
->len
) {
1663 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
1666 ibuf
->ops
->get(pipe
, ibuf
);
1668 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1670 ibuf
->offset
+= obuf
->len
;
1671 ibuf
->len
-= obuf
->len
;
1678 fuse_copy_init(&cs
, fc
, 0, NULL
, nbuf
);
1682 if (flags
& SPLICE_F_MOVE
)
1685 ret
= fuse_dev_do_write(fc
, &cs
, len
);
1687 for (idx
= 0; idx
< nbuf
; idx
++) {
1688 struct pipe_buffer
*buf
= &bufs
[idx
];
1689 buf
->ops
->release(pipe
, buf
);
1696 static unsigned fuse_dev_poll(struct file
*file
, poll_table
*wait
)
1698 unsigned mask
= POLLOUT
| POLLWRNORM
;
1699 struct fuse_conn
*fc
= fuse_get_conn(file
);
1703 poll_wait(file
, &fc
->waitq
, wait
);
1705 spin_lock(&fc
->lock
);
1708 else if (request_pending(fc
))
1709 mask
|= POLLIN
| POLLRDNORM
;
1710 spin_unlock(&fc
->lock
);
1716 * Abort all requests on the given list (pending or processing)
1718 * This function releases and reacquires fc->lock
1720 static void end_requests(struct fuse_conn
*fc
, struct list_head
*head
)
1721 __releases(fc
->lock
)
1722 __acquires(fc
->lock
)
1724 while (!list_empty(head
)) {
1725 struct fuse_req
*req
;
1726 req
= list_entry(head
->next
, struct fuse_req
, list
);
1727 req
->out
.h
.error
= -ECONNABORTED
;
1728 request_end(fc
, req
);
1729 spin_lock(&fc
->lock
);
1734 * Abort requests under I/O
1736 * The requests are set to aborted and finished, and the request
1737 * waiter is woken up. This will make request_wait_answer() wait
1738 * until the request is unlocked and then return.
1740 * If the request is asynchronous, then the end function needs to be
1741 * called after waiting for the request to be unlocked (if it was
1744 static void end_io_requests(struct fuse_conn
*fc
)
1745 __releases(fc
->lock
)
1746 __acquires(fc
->lock
)
1748 while (!list_empty(&fc
->io
)) {
1749 struct fuse_req
*req
=
1750 list_entry(fc
->io
.next
, struct fuse_req
, list
);
1751 void (*end
) (struct fuse_conn
*, struct fuse_req
*) = req
->end
;
1754 req
->out
.h
.error
= -ECONNABORTED
;
1755 req
->state
= FUSE_REQ_FINISHED
;
1756 list_del_init(&req
->list
);
1757 wake_up(&req
->waitq
);
1760 __fuse_get_request(req
);
1761 spin_unlock(&fc
->lock
);
1762 wait_event(req
->waitq
, !req
->locked
);
1764 fuse_put_request(fc
, req
);
1765 spin_lock(&fc
->lock
);
1770 static void end_queued_requests(struct fuse_conn
*fc
)
1771 __releases(fc
->lock
)
1772 __acquires(fc
->lock
)
1774 fc
->max_background
= UINT_MAX
;
1776 end_requests(fc
, &fc
->pending
);
1777 end_requests(fc
, &fc
->processing
);
1781 * Abort all requests.
1783 * Emergency exit in case of a malicious or accidental deadlock, or
1784 * just a hung filesystem.
1786 * The same effect is usually achievable through killing the
1787 * filesystem daemon and all users of the filesystem. The exception
1788 * is the combination of an asynchronous request and the tricky
1789 * deadlock (see Documentation/filesystems/fuse.txt).
1791 * During the aborting, progression of requests from the pending and
1792 * processing lists onto the io list, and progression of new requests
1793 * onto the pending list is prevented by req->connected being false.
1795 * Progression of requests under I/O to the processing list is
1796 * prevented by the req->aborted flag being true for these requests.
1797 * For this reason requests on the io list must be aborted first.
1799 void fuse_abort_conn(struct fuse_conn
*fc
)
1801 spin_lock(&fc
->lock
);
1802 if (fc
->connected
) {
1805 end_io_requests(fc
);
1806 end_queued_requests(fc
);
1807 wake_up_all(&fc
->waitq
);
1808 wake_up_all(&fc
->blocked_waitq
);
1809 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
1811 spin_unlock(&fc
->lock
);
1813 EXPORT_SYMBOL_GPL(fuse_abort_conn
);
1815 int fuse_dev_release(struct inode
*inode
, struct file
*file
)
1817 struct fuse_conn
*fc
= fuse_get_conn(file
);
1819 spin_lock(&fc
->lock
);
1822 end_queued_requests(fc
);
1823 wake_up_all(&fc
->blocked_waitq
);
1824 spin_unlock(&fc
->lock
);
1830 EXPORT_SYMBOL_GPL(fuse_dev_release
);
1832 static int fuse_dev_fasync(int fd
, struct file
*file
, int on
)
1834 struct fuse_conn
*fc
= fuse_get_conn(file
);
1838 /* No locking - fasync_helper does its own locking */
1839 return fasync_helper(fd
, file
, on
, &fc
->fasync
);
1842 const struct file_operations fuse_dev_operations
= {
1843 .owner
= THIS_MODULE
,
1844 .llseek
= no_llseek
,
1845 .read
= do_sync_read
,
1846 .aio_read
= fuse_dev_read
,
1847 .splice_read
= fuse_dev_splice_read
,
1848 .write
= do_sync_write
,
1849 .aio_write
= fuse_dev_write
,
1850 .splice_write
= fuse_dev_splice_write
,
1851 .poll
= fuse_dev_poll
,
1852 .release
= fuse_dev_release
,
1853 .fasync
= fuse_dev_fasync
,
1855 EXPORT_SYMBOL_GPL(fuse_dev_operations
);
1857 static struct miscdevice fuse_miscdevice
= {
1858 .minor
= FUSE_MINOR
,
1860 .fops
= &fuse_dev_operations
,
1863 int __init
fuse_dev_init(void)
1866 fuse_req_cachep
= kmem_cache_create("fuse_request",
1867 sizeof(struct fuse_req
),
1869 if (!fuse_req_cachep
)
1872 err
= misc_register(&fuse_miscdevice
);
1874 goto out_cache_clean
;
1879 kmem_cache_destroy(fuse_req_cachep
);
1884 void fuse_dev_cleanup(void)
1886 misc_deregister(&fuse_miscdevice
);
1887 kmem_cache_destroy(fuse_req_cachep
);