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/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/kernel.h>
14 #include <linux/sched.h>
16 static const struct file_operations fuse_direct_io_file_operations
;
18 static int fuse_send_open(struct inode
*inode
, struct file
*file
, int isdir
,
19 struct fuse_open_out
*outargp
)
21 struct fuse_conn
*fc
= get_fuse_conn(inode
);
22 struct fuse_open_in inarg
;
26 req
= fuse_get_req(fc
);
30 memset(&inarg
, 0, sizeof(inarg
));
31 inarg
.flags
= file
->f_flags
& ~(O_CREAT
| O_EXCL
| O_NOCTTY
);
32 if (!fc
->atomic_o_trunc
)
33 inarg
.flags
&= ~O_TRUNC
;
34 req
->in
.h
.opcode
= isdir
? FUSE_OPENDIR
: FUSE_OPEN
;
35 req
->in
.h
.nodeid
= get_node_id(inode
);
37 req
->in
.args
[0].size
= sizeof(inarg
);
38 req
->in
.args
[0].value
= &inarg
;
40 req
->out
.args
[0].size
= sizeof(*outargp
);
41 req
->out
.args
[0].value
= outargp
;
42 fuse_request_send(fc
, req
);
43 err
= req
->out
.h
.error
;
44 fuse_put_request(fc
, req
);
49 struct fuse_file
*fuse_file_alloc(struct fuse_conn
*fc
)
53 ff
= kmalloc(sizeof(struct fuse_file
), GFP_KERNEL
);
57 ff
->reserved_req
= fuse_request_alloc();
58 if (unlikely(!ff
->reserved_req
)) {
63 INIT_LIST_HEAD(&ff
->write_entry
);
64 atomic_set(&ff
->count
, 0);
65 RB_CLEAR_NODE(&ff
->polled_node
);
66 init_waitqueue_head(&ff
->poll_wait
);
70 spin_unlock(&fc
->lock
);
75 void fuse_file_free(struct fuse_file
*ff
)
77 fuse_request_free(ff
->reserved_req
);
81 static struct fuse_file
*fuse_file_get(struct fuse_file
*ff
)
83 atomic_inc(&ff
->count
);
87 static void fuse_release_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
89 path_put(&req
->misc
.release
.path
);
92 static void fuse_file_put(struct fuse_file
*ff
)
94 if (atomic_dec_and_test(&ff
->count
)) {
95 struct fuse_req
*req
= ff
->reserved_req
;
96 struct inode
*inode
= req
->misc
.release
.path
.dentry
->d_inode
;
97 struct fuse_conn
*fc
= get_fuse_conn(inode
);
98 req
->end
= fuse_release_end
;
99 fuse_request_send_background(fc
, req
);
104 void fuse_finish_open(struct inode
*inode
, struct file
*file
,
105 struct fuse_file
*ff
, struct fuse_open_out
*outarg
)
107 if (outarg
->open_flags
& FOPEN_DIRECT_IO
)
108 file
->f_op
= &fuse_direct_io_file_operations
;
109 if (!(outarg
->open_flags
& FOPEN_KEEP_CACHE
))
110 invalidate_inode_pages2(inode
->i_mapping
);
111 if (outarg
->open_flags
& FOPEN_NONSEEKABLE
)
112 nonseekable_open(inode
, file
);
114 file
->private_data
= fuse_file_get(ff
);
117 int fuse_open_common(struct inode
*inode
, struct file
*file
, int isdir
)
119 struct fuse_conn
*fc
= get_fuse_conn(inode
);
120 struct fuse_open_out outarg
;
121 struct fuse_file
*ff
;
124 /* VFS checks this, but only _after_ ->open() */
125 if (file
->f_flags
& O_DIRECT
)
128 err
= generic_file_open(inode
, file
);
132 ff
= fuse_file_alloc(fc
);
136 err
= fuse_send_open(inode
, file
, isdir
, &outarg
);
141 outarg
.open_flags
&= ~FOPEN_DIRECT_IO
;
142 fuse_finish_open(inode
, file
, ff
, &outarg
);
148 void fuse_release_fill(struct fuse_file
*ff
, u64 nodeid
, int flags
, int opcode
)
150 struct fuse_req
*req
= ff
->reserved_req
;
151 struct fuse_release_in
*inarg
= &req
->misc
.release
.in
;
154 inarg
->flags
= flags
;
155 req
->in
.h
.opcode
= opcode
;
156 req
->in
.h
.nodeid
= nodeid
;
158 req
->in
.args
[0].size
= sizeof(struct fuse_release_in
);
159 req
->in
.args
[0].value
= inarg
;
162 int fuse_release_common(struct inode
*inode
, struct file
*file
, int isdir
)
164 struct fuse_conn
*fc
;
165 struct fuse_file
*ff
;
166 struct fuse_req
*req
;
168 ff
= file
->private_data
;
170 return 0; /* return value is ignored by VFS */
172 fc
= get_fuse_conn(inode
);
173 req
= ff
->reserved_req
;
175 fuse_release_fill(ff
, get_node_id(inode
), file
->f_flags
,
176 isdir
? FUSE_RELEASEDIR
: FUSE_RELEASE
);
178 /* Hold vfsmount and dentry until release is finished */
179 path_get(&file
->f_path
);
180 req
->misc
.release
.path
= file
->f_path
;
182 spin_lock(&fc
->lock
);
183 list_del(&ff
->write_entry
);
184 if (!RB_EMPTY_NODE(&ff
->polled_node
))
185 rb_erase(&ff
->polled_node
, &fc
->polled_files
);
186 spin_unlock(&fc
->lock
);
188 wake_up_interruptible_sync(&ff
->poll_wait
);
190 * Normally this will send the RELEASE request, however if
191 * some asynchronous READ or WRITE requests are outstanding,
192 * the sending will be delayed.
198 static int fuse_open(struct inode
*inode
, struct file
*file
)
200 return fuse_open_common(inode
, file
, 0);
203 static int fuse_release(struct inode
*inode
, struct file
*file
)
205 return fuse_release_common(inode
, file
, 0);
209 * Scramble the ID space with XTEA, so that the value of the files_struct
210 * pointer is not exposed to userspace.
212 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
)
214 u32
*k
= fc
->scramble_key
;
215 u64 v
= (unsigned long) id
;
221 for (i
= 0; i
< 32; i
++) {
222 v0
+= ((v1
<< 4 ^ v1
>> 5) + v1
) ^ (sum
+ k
[sum
& 3]);
224 v1
+= ((v0
<< 4 ^ v0
>> 5) + v0
) ^ (sum
+ k
[sum
>>11 & 3]);
227 return (u64
) v0
+ ((u64
) v1
<< 32);
231 * Check if page is under writeback
233 * This is currently done by walking the list of writepage requests
234 * for the inode, which can be pretty inefficient.
236 static bool fuse_page_is_writeback(struct inode
*inode
, pgoff_t index
)
238 struct fuse_conn
*fc
= get_fuse_conn(inode
);
239 struct fuse_inode
*fi
= get_fuse_inode(inode
);
240 struct fuse_req
*req
;
243 spin_lock(&fc
->lock
);
244 list_for_each_entry(req
, &fi
->writepages
, writepages_entry
) {
247 BUG_ON(req
->inode
!= inode
);
248 curr_index
= req
->misc
.write
.in
.offset
>> PAGE_CACHE_SHIFT
;
249 if (curr_index
== index
) {
254 spin_unlock(&fc
->lock
);
260 * Wait for page writeback to be completed.
262 * Since fuse doesn't rely on the VM writeback tracking, this has to
263 * use some other means.
265 static int fuse_wait_on_page_writeback(struct inode
*inode
, pgoff_t index
)
267 struct fuse_inode
*fi
= get_fuse_inode(inode
);
269 wait_event(fi
->page_waitq
, !fuse_page_is_writeback(inode
, index
));
273 static int fuse_flush(struct file
*file
, fl_owner_t id
)
275 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
276 struct fuse_conn
*fc
= get_fuse_conn(inode
);
277 struct fuse_file
*ff
= file
->private_data
;
278 struct fuse_req
*req
;
279 struct fuse_flush_in inarg
;
282 if (is_bad_inode(inode
))
288 req
= fuse_get_req_nofail(fc
, file
);
289 memset(&inarg
, 0, sizeof(inarg
));
291 inarg
.lock_owner
= fuse_lock_owner_id(fc
, id
);
292 req
->in
.h
.opcode
= FUSE_FLUSH
;
293 req
->in
.h
.nodeid
= get_node_id(inode
);
295 req
->in
.args
[0].size
= sizeof(inarg
);
296 req
->in
.args
[0].value
= &inarg
;
298 fuse_request_send(fc
, req
);
299 err
= req
->out
.h
.error
;
300 fuse_put_request(fc
, req
);
301 if (err
== -ENOSYS
) {
309 * Wait for all pending writepages on the inode to finish.
311 * This is currently done by blocking further writes with FUSE_NOWRITE
312 * and waiting for all sent writes to complete.
314 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
315 * could conflict with truncation.
317 static void fuse_sync_writes(struct inode
*inode
)
319 fuse_set_nowrite(inode
);
320 fuse_release_nowrite(inode
);
323 int fuse_fsync_common(struct file
*file
, struct dentry
*de
, int datasync
,
326 struct inode
*inode
= de
->d_inode
;
327 struct fuse_conn
*fc
= get_fuse_conn(inode
);
328 struct fuse_file
*ff
= file
->private_data
;
329 struct fuse_req
*req
;
330 struct fuse_fsync_in inarg
;
333 if (is_bad_inode(inode
))
336 if ((!isdir
&& fc
->no_fsync
) || (isdir
&& fc
->no_fsyncdir
))
340 * Start writeback against all dirty pages of the inode, then
341 * wait for all outstanding writes, before sending the FSYNC
344 err
= write_inode_now(inode
, 0);
348 fuse_sync_writes(inode
);
350 req
= fuse_get_req(fc
);
354 memset(&inarg
, 0, sizeof(inarg
));
356 inarg
.fsync_flags
= datasync
? 1 : 0;
357 req
->in
.h
.opcode
= isdir
? FUSE_FSYNCDIR
: FUSE_FSYNC
;
358 req
->in
.h
.nodeid
= get_node_id(inode
);
360 req
->in
.args
[0].size
= sizeof(inarg
);
361 req
->in
.args
[0].value
= &inarg
;
362 fuse_request_send(fc
, req
);
363 err
= req
->out
.h
.error
;
364 fuse_put_request(fc
, req
);
365 if (err
== -ENOSYS
) {
375 static int fuse_fsync(struct file
*file
, struct dentry
*de
, int datasync
)
377 return fuse_fsync_common(file
, de
, datasync
, 0);
380 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
381 struct inode
*inode
, loff_t pos
, size_t count
, int opcode
)
383 struct fuse_read_in
*inarg
= &req
->misc
.read
.in
;
384 struct fuse_file
*ff
= file
->private_data
;
389 inarg
->flags
= file
->f_flags
;
390 req
->in
.h
.opcode
= opcode
;
391 req
->in
.h
.nodeid
= get_node_id(inode
);
393 req
->in
.args
[0].size
= sizeof(struct fuse_read_in
);
394 req
->in
.args
[0].value
= inarg
;
396 req
->out
.numargs
= 1;
397 req
->out
.args
[0].size
= count
;
400 static size_t fuse_send_read(struct fuse_req
*req
, struct file
*file
,
401 struct inode
*inode
, loff_t pos
, size_t count
,
404 struct fuse_conn
*fc
= get_fuse_conn(inode
);
406 fuse_read_fill(req
, file
, inode
, pos
, count
, FUSE_READ
);
408 struct fuse_read_in
*inarg
= &req
->misc
.read
.in
;
410 inarg
->read_flags
|= FUSE_READ_LOCKOWNER
;
411 inarg
->lock_owner
= fuse_lock_owner_id(fc
, owner
);
413 fuse_request_send(fc
, req
);
414 return req
->out
.args
[0].size
;
417 static void fuse_read_update_size(struct inode
*inode
, loff_t size
,
420 struct fuse_conn
*fc
= get_fuse_conn(inode
);
421 struct fuse_inode
*fi
= get_fuse_inode(inode
);
423 spin_lock(&fc
->lock
);
424 if (attr_ver
== fi
->attr_version
&& size
< inode
->i_size
) {
425 fi
->attr_version
= ++fc
->attr_version
;
426 i_size_write(inode
, size
);
428 spin_unlock(&fc
->lock
);
431 static int fuse_readpage(struct file
*file
, struct page
*page
)
433 struct inode
*inode
= page
->mapping
->host
;
434 struct fuse_conn
*fc
= get_fuse_conn(inode
);
435 struct fuse_req
*req
;
437 loff_t pos
= page_offset(page
);
438 size_t count
= PAGE_CACHE_SIZE
;
443 if (is_bad_inode(inode
))
447 * Page writeback can extend beyond the liftime of the
448 * page-cache page, so make sure we read a properly synced
451 fuse_wait_on_page_writeback(inode
, page
->index
);
453 req
= fuse_get_req(fc
);
458 attr_ver
= fuse_get_attr_version(fc
);
460 req
->out
.page_zeroing
= 1;
461 req
->out
.argpages
= 1;
463 req
->pages
[0] = page
;
464 num_read
= fuse_send_read(req
, file
, inode
, pos
, count
, NULL
);
465 err
= req
->out
.h
.error
;
466 fuse_put_request(fc
, req
);
470 * Short read means EOF. If file size is larger, truncate it
472 if (num_read
< count
)
473 fuse_read_update_size(inode
, pos
+ num_read
, attr_ver
);
475 SetPageUptodate(page
);
478 fuse_invalidate_attr(inode
); /* atime changed */
484 static void fuse_readpages_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
487 size_t count
= req
->misc
.read
.in
.size
;
488 size_t num_read
= req
->out
.args
[0].size
;
489 struct inode
*inode
= req
->pages
[0]->mapping
->host
;
492 * Short read means EOF. If file size is larger, truncate it
494 if (!req
->out
.h
.error
&& num_read
< count
) {
495 loff_t pos
= page_offset(req
->pages
[0]) + num_read
;
496 fuse_read_update_size(inode
, pos
, req
->misc
.read
.attr_ver
);
499 fuse_invalidate_attr(inode
); /* atime changed */
501 for (i
= 0; i
< req
->num_pages
; i
++) {
502 struct page
*page
= req
->pages
[i
];
503 if (!req
->out
.h
.error
)
504 SetPageUptodate(page
);
510 fuse_file_put(req
->ff
);
513 static void fuse_send_readpages(struct fuse_req
*req
, struct file
*file
,
516 struct fuse_conn
*fc
= get_fuse_conn(inode
);
517 loff_t pos
= page_offset(req
->pages
[0]);
518 size_t count
= req
->num_pages
<< PAGE_CACHE_SHIFT
;
520 req
->out
.argpages
= 1;
521 req
->out
.page_zeroing
= 1;
522 fuse_read_fill(req
, file
, inode
, pos
, count
, FUSE_READ
);
523 req
->misc
.read
.attr_ver
= fuse_get_attr_version(fc
);
524 if (fc
->async_read
) {
525 struct fuse_file
*ff
= file
->private_data
;
526 req
->ff
= fuse_file_get(ff
);
527 req
->end
= fuse_readpages_end
;
528 fuse_request_send_background(fc
, req
);
530 fuse_request_send(fc
, req
);
531 fuse_readpages_end(fc
, req
);
532 fuse_put_request(fc
, req
);
536 struct fuse_fill_data
{
537 struct fuse_req
*req
;
542 static int fuse_readpages_fill(void *_data
, struct page
*page
)
544 struct fuse_fill_data
*data
= _data
;
545 struct fuse_req
*req
= data
->req
;
546 struct inode
*inode
= data
->inode
;
547 struct fuse_conn
*fc
= get_fuse_conn(inode
);
549 fuse_wait_on_page_writeback(inode
, page
->index
);
551 if (req
->num_pages
&&
552 (req
->num_pages
== FUSE_MAX_PAGES_PER_REQ
||
553 (req
->num_pages
+ 1) * PAGE_CACHE_SIZE
> fc
->max_read
||
554 req
->pages
[req
->num_pages
- 1]->index
+ 1 != page
->index
)) {
555 fuse_send_readpages(req
, data
->file
, inode
);
556 data
->req
= req
= fuse_get_req(fc
);
562 req
->pages
[req
->num_pages
] = page
;
567 static int fuse_readpages(struct file
*file
, struct address_space
*mapping
,
568 struct list_head
*pages
, unsigned nr_pages
)
570 struct inode
*inode
= mapping
->host
;
571 struct fuse_conn
*fc
= get_fuse_conn(inode
);
572 struct fuse_fill_data data
;
576 if (is_bad_inode(inode
))
581 data
.req
= fuse_get_req(fc
);
582 err
= PTR_ERR(data
.req
);
583 if (IS_ERR(data
.req
))
586 err
= read_cache_pages(mapping
, pages
, fuse_readpages_fill
, &data
);
588 if (data
.req
->num_pages
)
589 fuse_send_readpages(data
.req
, file
, inode
);
591 fuse_put_request(fc
, data
.req
);
597 static ssize_t
fuse_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
598 unsigned long nr_segs
, loff_t pos
)
600 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
602 if (pos
+ iov_length(iov
, nr_segs
) > i_size_read(inode
)) {
605 * If trying to read past EOF, make sure the i_size
606 * attribute is up-to-date.
608 err
= fuse_update_attributes(inode
, NULL
, iocb
->ki_filp
, NULL
);
613 return generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
616 static void fuse_write_fill(struct fuse_req
*req
, struct fuse_file
*ff
,
617 struct inode
*inode
, loff_t pos
, size_t count
)
619 struct fuse_conn
*fc
= get_fuse_conn(inode
);
620 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
621 struct fuse_write_out
*outarg
= &req
->misc
.write
.out
;
626 req
->in
.h
.opcode
= FUSE_WRITE
;
627 req
->in
.h
.nodeid
= get_node_id(inode
);
630 req
->in
.args
[0].size
= FUSE_COMPAT_WRITE_IN_SIZE
;
632 req
->in
.args
[0].size
= sizeof(struct fuse_write_in
);
633 req
->in
.args
[0].value
= inarg
;
634 req
->in
.args
[1].size
= count
;
635 req
->out
.numargs
= 1;
636 req
->out
.args
[0].size
= sizeof(struct fuse_write_out
);
637 req
->out
.args
[0].value
= outarg
;
640 static size_t fuse_send_write(struct fuse_req
*req
, struct file
*file
,
641 struct inode
*inode
, loff_t pos
, size_t count
,
644 struct fuse_conn
*fc
= get_fuse_conn(inode
);
645 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
647 fuse_write_fill(req
, file
->private_data
, inode
, pos
, count
);
648 inarg
->flags
= file
->f_flags
;
650 inarg
->write_flags
|= FUSE_WRITE_LOCKOWNER
;
651 inarg
->lock_owner
= fuse_lock_owner_id(fc
, owner
);
653 fuse_request_send(fc
, req
);
654 return req
->misc
.write
.out
.size
;
657 static int fuse_write_begin(struct file
*file
, struct address_space
*mapping
,
658 loff_t pos
, unsigned len
, unsigned flags
,
659 struct page
**pagep
, void **fsdata
)
661 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
663 *pagep
= grab_cache_page_write_begin(mapping
, index
, flags
);
669 static void fuse_write_update_size(struct inode
*inode
, loff_t pos
)
671 struct fuse_conn
*fc
= get_fuse_conn(inode
);
672 struct fuse_inode
*fi
= get_fuse_inode(inode
);
674 spin_lock(&fc
->lock
);
675 fi
->attr_version
= ++fc
->attr_version
;
676 if (pos
> inode
->i_size
)
677 i_size_write(inode
, pos
);
678 spin_unlock(&fc
->lock
);
681 static int fuse_buffered_write(struct file
*file
, struct inode
*inode
,
682 loff_t pos
, unsigned count
, struct page
*page
)
686 struct fuse_conn
*fc
= get_fuse_conn(inode
);
687 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
688 struct fuse_req
*req
;
690 if (is_bad_inode(inode
))
694 * Make sure writepages on the same page are not mixed up with
697 fuse_wait_on_page_writeback(inode
, page
->index
);
699 req
= fuse_get_req(fc
);
703 req
->in
.argpages
= 1;
705 req
->pages
[0] = page
;
706 req
->page_offset
= offset
;
707 nres
= fuse_send_write(req
, file
, inode
, pos
, count
, NULL
);
708 err
= req
->out
.h
.error
;
709 fuse_put_request(fc
, req
);
714 fuse_write_update_size(inode
, pos
);
715 if (count
== PAGE_CACHE_SIZE
)
716 SetPageUptodate(page
);
718 fuse_invalidate_attr(inode
);
719 return err
? err
: nres
;
722 static int fuse_write_end(struct file
*file
, struct address_space
*mapping
,
723 loff_t pos
, unsigned len
, unsigned copied
,
724 struct page
*page
, void *fsdata
)
726 struct inode
*inode
= mapping
->host
;
730 res
= fuse_buffered_write(file
, inode
, pos
, copied
, page
);
733 page_cache_release(page
);
737 static size_t fuse_send_write_pages(struct fuse_req
*req
, struct file
*file
,
738 struct inode
*inode
, loff_t pos
,
745 for (i
= 0; i
< req
->num_pages
; i
++)
746 fuse_wait_on_page_writeback(inode
, req
->pages
[i
]->index
);
748 res
= fuse_send_write(req
, file
, inode
, pos
, count
, NULL
);
750 offset
= req
->page_offset
;
752 for (i
= 0; i
< req
->num_pages
; i
++) {
753 struct page
*page
= req
->pages
[i
];
755 if (!req
->out
.h
.error
&& !offset
&& count
>= PAGE_CACHE_SIZE
)
756 SetPageUptodate(page
);
758 if (count
> PAGE_CACHE_SIZE
- offset
)
759 count
-= PAGE_CACHE_SIZE
- offset
;
765 page_cache_release(page
);
771 static ssize_t
fuse_fill_write_pages(struct fuse_req
*req
,
772 struct address_space
*mapping
,
773 struct iov_iter
*ii
, loff_t pos
)
775 struct fuse_conn
*fc
= get_fuse_conn(mapping
->host
);
776 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
780 req
->in
.argpages
= 1;
781 req
->page_offset
= offset
;
786 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
787 size_t bytes
= min_t(size_t, PAGE_CACHE_SIZE
- offset
,
790 bytes
= min_t(size_t, bytes
, fc
->max_write
- count
);
794 if (iov_iter_fault_in_readable(ii
, bytes
))
798 page
= grab_cache_page_write_begin(mapping
, index
, 0);
803 tmp
= iov_iter_copy_from_user_atomic(page
, ii
, offset
, bytes
);
805 flush_dcache_page(page
);
809 page_cache_release(page
);
810 bytes
= min(bytes
, iov_iter_single_seg_count(ii
));
815 req
->pages
[req
->num_pages
] = page
;
818 iov_iter_advance(ii
, tmp
);
822 if (offset
== PAGE_CACHE_SIZE
)
827 } while (iov_iter_count(ii
) && count
< fc
->max_write
&&
828 req
->num_pages
< FUSE_MAX_PAGES_PER_REQ
&& offset
== 0);
830 return count
> 0 ? count
: err
;
833 static ssize_t
fuse_perform_write(struct file
*file
,
834 struct address_space
*mapping
,
835 struct iov_iter
*ii
, loff_t pos
)
837 struct inode
*inode
= mapping
->host
;
838 struct fuse_conn
*fc
= get_fuse_conn(inode
);
842 if (is_bad_inode(inode
))
846 struct fuse_req
*req
;
849 req
= fuse_get_req(fc
);
855 count
= fuse_fill_write_pages(req
, mapping
, ii
, pos
);
861 num_written
= fuse_send_write_pages(req
, file
, inode
,
863 err
= req
->out
.h
.error
;
868 /* break out of the loop on short write */
869 if (num_written
!= count
)
873 fuse_put_request(fc
, req
);
874 } while (!err
&& iov_iter_count(ii
));
877 fuse_write_update_size(inode
, pos
);
879 fuse_invalidate_attr(inode
);
881 return res
> 0 ? res
: err
;
884 static ssize_t
fuse_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
885 unsigned long nr_segs
, loff_t pos
)
887 struct file
*file
= iocb
->ki_filp
;
888 struct address_space
*mapping
= file
->f_mapping
;
891 struct inode
*inode
= mapping
->host
;
895 WARN_ON(iocb
->ki_pos
!= pos
);
897 err
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_READ
);
901 mutex_lock(&inode
->i_mutex
);
902 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
904 /* We can write back this queue in page reclaim */
905 current
->backing_dev_info
= mapping
->backing_dev_info
;
907 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
914 err
= file_remove_suid(file
);
918 file_update_time(file
);
920 iov_iter_init(&i
, iov
, nr_segs
, count
, 0);
921 written
= fuse_perform_write(file
, mapping
, &i
, pos
);
923 iocb
->ki_pos
= pos
+ written
;
926 current
->backing_dev_info
= NULL
;
927 mutex_unlock(&inode
->i_mutex
);
929 return written
? written
: err
;
932 static void fuse_release_user_pages(struct fuse_req
*req
, int write
)
936 for (i
= 0; i
< req
->num_pages
; i
++) {
937 struct page
*page
= req
->pages
[i
];
939 set_page_dirty_lock(page
);
944 static int fuse_get_user_pages(struct fuse_req
*req
, const char __user
*buf
,
945 size_t *nbytesp
, int write
)
947 size_t nbytes
= *nbytesp
;
948 unsigned long user_addr
= (unsigned long) buf
;
949 unsigned offset
= user_addr
& ~PAGE_MASK
;
952 /* Special case for kernel I/O: can copy directly into the buffer */
953 if (segment_eq(get_fs(), KERNEL_DS
)) {
955 req
->in
.args
[1].value
= (void *) user_addr
;
957 req
->out
.args
[0].value
= (void *) user_addr
;
962 nbytes
= min_t(size_t, nbytes
, FUSE_MAX_PAGES_PER_REQ
<< PAGE_SHIFT
);
963 npages
= (nbytes
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
964 npages
= clamp(npages
, 1, FUSE_MAX_PAGES_PER_REQ
);
965 down_read(¤t
->mm
->mmap_sem
);
966 npages
= get_user_pages(current
, current
->mm
, user_addr
, npages
, !write
,
967 0, req
->pages
, NULL
);
968 up_read(¤t
->mm
->mmap_sem
);
972 req
->num_pages
= npages
;
973 req
->page_offset
= offset
;
976 req
->in
.argpages
= 1;
978 req
->out
.argpages
= 1;
980 nbytes
= (req
->num_pages
<< PAGE_SHIFT
) - req
->page_offset
;
981 *nbytesp
= min(*nbytesp
, nbytes
);
986 static ssize_t
fuse_direct_io(struct file
*file
, const char __user
*buf
,
987 size_t count
, loff_t
*ppos
, int write
)
989 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
990 struct fuse_conn
*fc
= get_fuse_conn(inode
);
991 size_t nmax
= write
? fc
->max_write
: fc
->max_read
;
994 struct fuse_req
*req
;
996 req
= fuse_get_req(fc
);
1002 size_t nbytes
= min(count
, nmax
);
1003 int err
= fuse_get_user_pages(req
, buf
, &nbytes
, write
);
1010 nres
= fuse_send_write(req
, file
, inode
, pos
, nbytes
,
1013 nres
= fuse_send_read(req
, file
, inode
, pos
, nbytes
,
1015 fuse_release_user_pages(req
, !write
);
1016 if (req
->out
.h
.error
) {
1018 res
= req
->out
.h
.error
;
1020 } else if (nres
> nbytes
) {
1031 fuse_put_request(fc
, req
);
1032 req
= fuse_get_req(fc
);
1037 fuse_put_request(fc
, req
);
1044 static ssize_t
fuse_direct_read(struct file
*file
, char __user
*buf
,
1045 size_t count
, loff_t
*ppos
)
1048 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1050 if (is_bad_inode(inode
))
1053 res
= fuse_direct_io(file
, buf
, count
, ppos
, 0);
1055 fuse_invalidate_attr(inode
);
1060 static ssize_t
fuse_direct_write(struct file
*file
, const char __user
*buf
,
1061 size_t count
, loff_t
*ppos
)
1063 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1066 if (is_bad_inode(inode
))
1069 /* Don't allow parallel writes to the same file */
1070 mutex_lock(&inode
->i_mutex
);
1071 res
= generic_write_checks(file
, ppos
, &count
, 0);
1073 res
= fuse_direct_io(file
, buf
, count
, ppos
, 1);
1075 fuse_write_update_size(inode
, *ppos
);
1077 mutex_unlock(&inode
->i_mutex
);
1079 fuse_invalidate_attr(inode
);
1084 static void fuse_writepage_free(struct fuse_conn
*fc
, struct fuse_req
*req
)
1086 __free_page(req
->pages
[0]);
1087 fuse_file_put(req
->ff
);
1090 static void fuse_writepage_finish(struct fuse_conn
*fc
, struct fuse_req
*req
)
1092 struct inode
*inode
= req
->inode
;
1093 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1094 struct backing_dev_info
*bdi
= inode
->i_mapping
->backing_dev_info
;
1096 list_del(&req
->writepages_entry
);
1097 dec_bdi_stat(bdi
, BDI_WRITEBACK
);
1098 dec_zone_page_state(req
->pages
[0], NR_WRITEBACK_TEMP
);
1099 bdi_writeout_inc(bdi
);
1100 wake_up(&fi
->page_waitq
);
1103 /* Called under fc->lock, may release and reacquire it */
1104 static void fuse_send_writepage(struct fuse_conn
*fc
, struct fuse_req
*req
)
1105 __releases(&fc
->lock
)
1106 __acquires(&fc
->lock
)
1108 struct fuse_inode
*fi
= get_fuse_inode(req
->inode
);
1109 loff_t size
= i_size_read(req
->inode
);
1110 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
1115 if (inarg
->offset
+ PAGE_CACHE_SIZE
<= size
) {
1116 inarg
->size
= PAGE_CACHE_SIZE
;
1117 } else if (inarg
->offset
< size
) {
1118 inarg
->size
= size
& (PAGE_CACHE_SIZE
- 1);
1120 /* Got truncated off completely */
1124 req
->in
.args
[1].size
= inarg
->size
;
1126 fuse_request_send_background_locked(fc
, req
);
1130 fuse_writepage_finish(fc
, req
);
1131 spin_unlock(&fc
->lock
);
1132 fuse_writepage_free(fc
, req
);
1133 fuse_put_request(fc
, req
);
1134 spin_lock(&fc
->lock
);
1138 * If fi->writectr is positive (no truncate or fsync going on) send
1139 * all queued writepage requests.
1141 * Called with fc->lock
1143 void fuse_flush_writepages(struct inode
*inode
)
1144 __releases(&fc
->lock
)
1145 __acquires(&fc
->lock
)
1147 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1148 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1149 struct fuse_req
*req
;
1151 while (fi
->writectr
>= 0 && !list_empty(&fi
->queued_writes
)) {
1152 req
= list_entry(fi
->queued_writes
.next
, struct fuse_req
, list
);
1153 list_del_init(&req
->list
);
1154 fuse_send_writepage(fc
, req
);
1158 static void fuse_writepage_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1160 struct inode
*inode
= req
->inode
;
1161 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1163 mapping_set_error(inode
->i_mapping
, req
->out
.h
.error
);
1164 spin_lock(&fc
->lock
);
1166 fuse_writepage_finish(fc
, req
);
1167 spin_unlock(&fc
->lock
);
1168 fuse_writepage_free(fc
, req
);
1171 static int fuse_writepage_locked(struct page
*page
)
1173 struct address_space
*mapping
= page
->mapping
;
1174 struct inode
*inode
= mapping
->host
;
1175 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1176 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1177 struct fuse_req
*req
;
1178 struct fuse_file
*ff
;
1179 struct page
*tmp_page
;
1181 set_page_writeback(page
);
1183 req
= fuse_request_alloc_nofs();
1187 tmp_page
= alloc_page(GFP_NOFS
| __GFP_HIGHMEM
);
1191 spin_lock(&fc
->lock
);
1192 BUG_ON(list_empty(&fi
->write_files
));
1193 ff
= list_entry(fi
->write_files
.next
, struct fuse_file
, write_entry
);
1194 req
->ff
= fuse_file_get(ff
);
1195 spin_unlock(&fc
->lock
);
1197 fuse_write_fill(req
, ff
, inode
, page_offset(page
), 0);
1199 copy_highpage(tmp_page
, page
);
1200 req
->misc
.write
.in
.write_flags
|= FUSE_WRITE_CACHE
;
1201 req
->in
.argpages
= 1;
1203 req
->pages
[0] = tmp_page
;
1204 req
->page_offset
= 0;
1205 req
->end
= fuse_writepage_end
;
1208 inc_bdi_stat(mapping
->backing_dev_info
, BDI_WRITEBACK
);
1209 inc_zone_page_state(tmp_page
, NR_WRITEBACK_TEMP
);
1210 end_page_writeback(page
);
1212 spin_lock(&fc
->lock
);
1213 list_add(&req
->writepages_entry
, &fi
->writepages
);
1214 list_add_tail(&req
->list
, &fi
->queued_writes
);
1215 fuse_flush_writepages(inode
);
1216 spin_unlock(&fc
->lock
);
1221 fuse_request_free(req
);
1223 end_page_writeback(page
);
1227 static int fuse_writepage(struct page
*page
, struct writeback_control
*wbc
)
1231 err
= fuse_writepage_locked(page
);
1237 static int fuse_launder_page(struct page
*page
)
1240 if (clear_page_dirty_for_io(page
)) {
1241 struct inode
*inode
= page
->mapping
->host
;
1242 err
= fuse_writepage_locked(page
);
1244 fuse_wait_on_page_writeback(inode
, page
->index
);
1250 * Write back dirty pages now, because there may not be any suitable
1253 static void fuse_vma_close(struct vm_area_struct
*vma
)
1255 filemap_write_and_wait(vma
->vm_file
->f_mapping
);
1259 * Wait for writeback against this page to complete before allowing it
1260 * to be marked dirty again, and hence written back again, possibly
1261 * before the previous writepage completed.
1263 * Block here, instead of in ->writepage(), so that the userspace fs
1264 * can only block processes actually operating on the filesystem.
1266 * Otherwise unprivileged userspace fs would be able to block
1271 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1273 static int fuse_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1275 struct page
*page
= vmf
->page
;
1277 * Don't use page->mapping as it may become NULL from a
1278 * concurrent truncate.
1280 struct inode
*inode
= vma
->vm_file
->f_mapping
->host
;
1282 fuse_wait_on_page_writeback(inode
, page
->index
);
1286 static struct vm_operations_struct fuse_file_vm_ops
= {
1287 .close
= fuse_vma_close
,
1288 .fault
= filemap_fault
,
1289 .page_mkwrite
= fuse_page_mkwrite
,
1292 static int fuse_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1294 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
)) {
1295 struct inode
*inode
= file
->f_dentry
->d_inode
;
1296 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1297 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1298 struct fuse_file
*ff
= file
->private_data
;
1300 * file may be written through mmap, so chain it onto the
1301 * inodes's write_file list
1303 spin_lock(&fc
->lock
);
1304 if (list_empty(&ff
->write_entry
))
1305 list_add(&ff
->write_entry
, &fi
->write_files
);
1306 spin_unlock(&fc
->lock
);
1308 file_accessed(file
);
1309 vma
->vm_ops
= &fuse_file_vm_ops
;
1313 static int fuse_direct_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1315 /* Can't provide the coherency needed for MAP_SHARED */
1316 if (vma
->vm_flags
& VM_MAYSHARE
)
1319 invalidate_inode_pages2(file
->f_mapping
);
1321 return generic_file_mmap(file
, vma
);
1324 static int convert_fuse_file_lock(const struct fuse_file_lock
*ffl
,
1325 struct file_lock
*fl
)
1327 switch (ffl
->type
) {
1333 if (ffl
->start
> OFFSET_MAX
|| ffl
->end
> OFFSET_MAX
||
1334 ffl
->end
< ffl
->start
)
1337 fl
->fl_start
= ffl
->start
;
1338 fl
->fl_end
= ffl
->end
;
1339 fl
->fl_pid
= ffl
->pid
;
1345 fl
->fl_type
= ffl
->type
;
1349 static void fuse_lk_fill(struct fuse_req
*req
, struct file
*file
,
1350 const struct file_lock
*fl
, int opcode
, pid_t pid
,
1353 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1354 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1355 struct fuse_file
*ff
= file
->private_data
;
1356 struct fuse_lk_in
*arg
= &req
->misc
.lk_in
;
1359 arg
->owner
= fuse_lock_owner_id(fc
, fl
->fl_owner
);
1360 arg
->lk
.start
= fl
->fl_start
;
1361 arg
->lk
.end
= fl
->fl_end
;
1362 arg
->lk
.type
= fl
->fl_type
;
1365 arg
->lk_flags
|= FUSE_LK_FLOCK
;
1366 req
->in
.h
.opcode
= opcode
;
1367 req
->in
.h
.nodeid
= get_node_id(inode
);
1368 req
->in
.numargs
= 1;
1369 req
->in
.args
[0].size
= sizeof(*arg
);
1370 req
->in
.args
[0].value
= arg
;
1373 static int fuse_getlk(struct file
*file
, struct file_lock
*fl
)
1375 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1376 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1377 struct fuse_req
*req
;
1378 struct fuse_lk_out outarg
;
1381 req
= fuse_get_req(fc
);
1383 return PTR_ERR(req
);
1385 fuse_lk_fill(req
, file
, fl
, FUSE_GETLK
, 0, 0);
1386 req
->out
.numargs
= 1;
1387 req
->out
.args
[0].size
= sizeof(outarg
);
1388 req
->out
.args
[0].value
= &outarg
;
1389 fuse_request_send(fc
, req
);
1390 err
= req
->out
.h
.error
;
1391 fuse_put_request(fc
, req
);
1393 err
= convert_fuse_file_lock(&outarg
.lk
, fl
);
1398 static int fuse_setlk(struct file
*file
, struct file_lock
*fl
, int flock
)
1400 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1401 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1402 struct fuse_req
*req
;
1403 int opcode
= (fl
->fl_flags
& FL_SLEEP
) ? FUSE_SETLKW
: FUSE_SETLK
;
1404 pid_t pid
= fl
->fl_type
!= F_UNLCK
? current
->tgid
: 0;
1407 if (fl
->fl_lmops
&& fl
->fl_lmops
->fl_grant
) {
1408 /* NLM needs asynchronous locks, which we don't support yet */
1412 /* Unlock on close is handled by the flush method */
1413 if (fl
->fl_flags
& FL_CLOSE
)
1416 req
= fuse_get_req(fc
);
1418 return PTR_ERR(req
);
1420 fuse_lk_fill(req
, file
, fl
, opcode
, pid
, flock
);
1421 fuse_request_send(fc
, req
);
1422 err
= req
->out
.h
.error
;
1423 /* locking is restartable */
1426 fuse_put_request(fc
, req
);
1430 static int fuse_file_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1432 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1433 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1436 if (cmd
== F_CANCELLK
) {
1438 } else if (cmd
== F_GETLK
) {
1440 posix_test_lock(file
, fl
);
1443 err
= fuse_getlk(file
, fl
);
1446 err
= posix_lock_file(file
, fl
, NULL
);
1448 err
= fuse_setlk(file
, fl
, 0);
1453 static int fuse_file_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1455 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1456 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1460 err
= flock_lock_file_wait(file
, fl
);
1462 /* emulate flock with POSIX locks */
1463 fl
->fl_owner
= (fl_owner_t
) file
;
1464 err
= fuse_setlk(file
, fl
, 1);
1470 static sector_t
fuse_bmap(struct address_space
*mapping
, sector_t block
)
1472 struct inode
*inode
= mapping
->host
;
1473 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1474 struct fuse_req
*req
;
1475 struct fuse_bmap_in inarg
;
1476 struct fuse_bmap_out outarg
;
1479 if (!inode
->i_sb
->s_bdev
|| fc
->no_bmap
)
1482 req
= fuse_get_req(fc
);
1486 memset(&inarg
, 0, sizeof(inarg
));
1487 inarg
.block
= block
;
1488 inarg
.blocksize
= inode
->i_sb
->s_blocksize
;
1489 req
->in
.h
.opcode
= FUSE_BMAP
;
1490 req
->in
.h
.nodeid
= get_node_id(inode
);
1491 req
->in
.numargs
= 1;
1492 req
->in
.args
[0].size
= sizeof(inarg
);
1493 req
->in
.args
[0].value
= &inarg
;
1494 req
->out
.numargs
= 1;
1495 req
->out
.args
[0].size
= sizeof(outarg
);
1496 req
->out
.args
[0].value
= &outarg
;
1497 fuse_request_send(fc
, req
);
1498 err
= req
->out
.h
.error
;
1499 fuse_put_request(fc
, req
);
1503 return err
? 0 : outarg
.block
;
1506 static loff_t
fuse_file_llseek(struct file
*file
, loff_t offset
, int origin
)
1509 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1511 mutex_lock(&inode
->i_mutex
);
1514 retval
= fuse_update_attributes(inode
, NULL
, file
, NULL
);
1517 offset
+= i_size_read(inode
);
1520 offset
+= file
->f_pos
;
1523 if (offset
>= 0 && offset
<= inode
->i_sb
->s_maxbytes
) {
1524 if (offset
!= file
->f_pos
) {
1525 file
->f_pos
= offset
;
1526 file
->f_version
= 0;
1531 mutex_unlock(&inode
->i_mutex
);
1535 static int fuse_ioctl_copy_user(struct page
**pages
, struct iovec
*iov
,
1536 unsigned int nr_segs
, size_t bytes
, bool to_user
)
1544 iov_iter_init(&ii
, iov
, nr_segs
, bytes
, 0);
1546 while (iov_iter_count(&ii
)) {
1547 struct page
*page
= pages
[page_idx
++];
1548 size_t todo
= min_t(size_t, PAGE_SIZE
, iov_iter_count(&ii
));
1551 kaddr
= map
= kmap(page
);
1554 char __user
*uaddr
= ii
.iov
->iov_base
+ ii
.iov_offset
;
1555 size_t iov_len
= ii
.iov
->iov_len
- ii
.iov_offset
;
1556 size_t copy
= min(todo
, iov_len
);
1560 left
= copy_from_user(kaddr
, uaddr
, copy
);
1562 left
= copy_to_user(uaddr
, kaddr
, copy
);
1567 iov_iter_advance(&ii
, copy
);
1579 * For ioctls, there is no generic way to determine how much memory
1580 * needs to be read and/or written. Furthermore, ioctls are allowed
1581 * to dereference the passed pointer, so the parameter requires deep
1582 * copying but FUSE has no idea whatsoever about what to copy in or
1585 * This is solved by allowing FUSE server to retry ioctl with
1586 * necessary in/out iovecs. Let's assume the ioctl implementation
1587 * needs to read in the following structure.
1594 * On the first callout to FUSE server, inarg->in_size and
1595 * inarg->out_size will be NULL; then, the server completes the ioctl
1596 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
1597 * the actual iov array to
1599 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
1601 * which tells FUSE to copy in the requested area and retry the ioctl.
1602 * On the second round, the server has access to the structure and
1603 * from that it can tell what to look for next, so on the invocation,
1604 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
1606 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
1607 * { .iov_base = a.buf, .iov_len = a.buflen } }
1609 * FUSE will copy both struct a and the pointed buffer from the
1610 * process doing the ioctl and retry ioctl with both struct a and the
1613 * This time, FUSE server has everything it needs and completes ioctl
1614 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
1616 * Copying data out works the same way.
1618 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
1619 * automatically initializes in and out iovs by decoding @cmd with
1620 * _IOC_* macros and the server is not allowed to request RETRY. This
1621 * limits ioctl data transfers to well-formed ioctls and is the forced
1622 * behavior for all FUSE servers.
1624 static long fuse_file_do_ioctl(struct file
*file
, unsigned int cmd
,
1625 unsigned long arg
, unsigned int flags
)
1627 struct inode
*inode
= file
->f_dentry
->d_inode
;
1628 struct fuse_file
*ff
= file
->private_data
;
1629 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1630 struct fuse_ioctl_in inarg
= {
1636 struct fuse_ioctl_out outarg
;
1637 struct fuse_req
*req
= NULL
;
1638 struct page
**pages
= NULL
;
1639 struct page
*iov_page
= NULL
;
1640 struct iovec
*in_iov
= NULL
, *out_iov
= NULL
;
1641 unsigned int in_iovs
= 0, out_iovs
= 0, num_pages
= 0, max_pages
;
1642 size_t in_size
, out_size
, transferred
;
1645 /* assume all the iovs returned by client always fits in a page */
1646 BUILD_BUG_ON(sizeof(struct iovec
) * FUSE_IOCTL_MAX_IOV
> PAGE_SIZE
);
1648 if (!fuse_allow_task(fc
, current
))
1652 if (is_bad_inode(inode
))
1656 pages
= kzalloc(sizeof(pages
[0]) * FUSE_MAX_PAGES_PER_REQ
, GFP_KERNEL
);
1657 iov_page
= alloc_page(GFP_KERNEL
);
1658 if (!pages
|| !iov_page
)
1662 * If restricted, initialize IO parameters as encoded in @cmd.
1663 * RETRY from server is not allowed.
1665 if (!(flags
& FUSE_IOCTL_UNRESTRICTED
)) {
1666 struct iovec
*iov
= page_address(iov_page
);
1668 iov
->iov_base
= (void __user
*)arg
;
1669 iov
->iov_len
= _IOC_SIZE(cmd
);
1671 if (_IOC_DIR(cmd
) & _IOC_WRITE
) {
1676 if (_IOC_DIR(cmd
) & _IOC_READ
) {
1683 inarg
.in_size
= in_size
= iov_length(in_iov
, in_iovs
);
1684 inarg
.out_size
= out_size
= iov_length(out_iov
, out_iovs
);
1687 * Out data can be used either for actual out data or iovs,
1688 * make sure there always is at least one page.
1690 out_size
= max_t(size_t, out_size
, PAGE_SIZE
);
1691 max_pages
= DIV_ROUND_UP(max(in_size
, out_size
), PAGE_SIZE
);
1693 /* make sure there are enough buffer pages and init request with them */
1695 if (max_pages
> FUSE_MAX_PAGES_PER_REQ
)
1697 while (num_pages
< max_pages
) {
1698 pages
[num_pages
] = alloc_page(GFP_KERNEL
| __GFP_HIGHMEM
);
1699 if (!pages
[num_pages
])
1704 req
= fuse_get_req(fc
);
1710 memcpy(req
->pages
, pages
, sizeof(req
->pages
[0]) * num_pages
);
1711 req
->num_pages
= num_pages
;
1713 /* okay, let's send it to the client */
1714 req
->in
.h
.opcode
= FUSE_IOCTL
;
1715 req
->in
.h
.nodeid
= get_node_id(inode
);
1716 req
->in
.numargs
= 1;
1717 req
->in
.args
[0].size
= sizeof(inarg
);
1718 req
->in
.args
[0].value
= &inarg
;
1721 req
->in
.args
[1].size
= in_size
;
1722 req
->in
.argpages
= 1;
1724 err
= fuse_ioctl_copy_user(pages
, in_iov
, in_iovs
, in_size
,
1730 req
->out
.numargs
= 2;
1731 req
->out
.args
[0].size
= sizeof(outarg
);
1732 req
->out
.args
[0].value
= &outarg
;
1733 req
->out
.args
[1].size
= out_size
;
1734 req
->out
.argpages
= 1;
1735 req
->out
.argvar
= 1;
1737 fuse_request_send(fc
, req
);
1738 err
= req
->out
.h
.error
;
1739 transferred
= req
->out
.args
[1].size
;
1740 fuse_put_request(fc
, req
);
1745 /* did it ask for retry? */
1746 if (outarg
.flags
& FUSE_IOCTL_RETRY
) {
1749 /* no retry if in restricted mode */
1751 if (!(flags
& FUSE_IOCTL_UNRESTRICTED
))
1754 in_iovs
= outarg
.in_iovs
;
1755 out_iovs
= outarg
.out_iovs
;
1758 * Make sure things are in boundary, separate checks
1759 * are to protect against overflow.
1762 if (in_iovs
> FUSE_IOCTL_MAX_IOV
||
1763 out_iovs
> FUSE_IOCTL_MAX_IOV
||
1764 in_iovs
+ out_iovs
> FUSE_IOCTL_MAX_IOV
)
1768 if ((in_iovs
+ out_iovs
) * sizeof(struct iovec
) != transferred
)
1771 /* okay, copy in iovs and retry */
1772 vaddr
= kmap_atomic(pages
[0], KM_USER0
);
1773 memcpy(page_address(iov_page
), vaddr
, transferred
);
1774 kunmap_atomic(vaddr
, KM_USER0
);
1776 in_iov
= page_address(iov_page
);
1777 out_iov
= in_iov
+ in_iovs
;
1783 if (transferred
> inarg
.out_size
)
1786 err
= fuse_ioctl_copy_user(pages
, out_iov
, out_iovs
, transferred
, true);
1789 fuse_put_request(fc
, req
);
1791 __free_page(iov_page
);
1793 __free_page(pages
[--num_pages
]);
1796 return err
? err
: outarg
.result
;
1799 static long fuse_file_ioctl(struct file
*file
, unsigned int cmd
,
1802 return fuse_file_do_ioctl(file
, cmd
, arg
, 0);
1805 static long fuse_file_compat_ioctl(struct file
*file
, unsigned int cmd
,
1808 return fuse_file_do_ioctl(file
, cmd
, arg
, FUSE_IOCTL_COMPAT
);
1812 * All files which have been polled are linked to RB tree
1813 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
1814 * find the matching one.
1816 static struct rb_node
**fuse_find_polled_node(struct fuse_conn
*fc
, u64 kh
,
1817 struct rb_node
**parent_out
)
1819 struct rb_node
**link
= &fc
->polled_files
.rb_node
;
1820 struct rb_node
*last
= NULL
;
1823 struct fuse_file
*ff
;
1826 ff
= rb_entry(last
, struct fuse_file
, polled_node
);
1829 link
= &last
->rb_left
;
1830 else if (kh
> ff
->kh
)
1831 link
= &last
->rb_right
;
1842 * The file is about to be polled. Make sure it's on the polled_files
1843 * RB tree. Note that files once added to the polled_files tree are
1844 * not removed before the file is released. This is because a file
1845 * polled once is likely to be polled again.
1847 static void fuse_register_polled_file(struct fuse_conn
*fc
,
1848 struct fuse_file
*ff
)
1850 spin_lock(&fc
->lock
);
1851 if (RB_EMPTY_NODE(&ff
->polled_node
)) {
1852 struct rb_node
**link
, *parent
;
1854 link
= fuse_find_polled_node(fc
, ff
->kh
, &parent
);
1856 rb_link_node(&ff
->polled_node
, parent
, link
);
1857 rb_insert_color(&ff
->polled_node
, &fc
->polled_files
);
1859 spin_unlock(&fc
->lock
);
1862 static unsigned fuse_file_poll(struct file
*file
, poll_table
*wait
)
1864 struct inode
*inode
= file
->f_dentry
->d_inode
;
1865 struct fuse_file
*ff
= file
->private_data
;
1866 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1867 struct fuse_poll_in inarg
= { .fh
= ff
->fh
, .kh
= ff
->kh
};
1868 struct fuse_poll_out outarg
;
1869 struct fuse_req
*req
;
1873 return DEFAULT_POLLMASK
;
1875 poll_wait(file
, &ff
->poll_wait
, wait
);
1878 * Ask for notification iff there's someone waiting for it.
1879 * The client may ignore the flag and always notify.
1881 if (waitqueue_active(&ff
->poll_wait
)) {
1882 inarg
.flags
|= FUSE_POLL_SCHEDULE_NOTIFY
;
1883 fuse_register_polled_file(fc
, ff
);
1886 req
= fuse_get_req(fc
);
1888 return PTR_ERR(req
);
1890 req
->in
.h
.opcode
= FUSE_POLL
;
1891 req
->in
.h
.nodeid
= get_node_id(inode
);
1892 req
->in
.numargs
= 1;
1893 req
->in
.args
[0].size
= sizeof(inarg
);
1894 req
->in
.args
[0].value
= &inarg
;
1895 req
->out
.numargs
= 1;
1896 req
->out
.args
[0].size
= sizeof(outarg
);
1897 req
->out
.args
[0].value
= &outarg
;
1898 fuse_request_send(fc
, req
);
1899 err
= req
->out
.h
.error
;
1900 fuse_put_request(fc
, req
);
1903 return outarg
.revents
;
1904 if (err
== -ENOSYS
) {
1906 return DEFAULT_POLLMASK
;
1912 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
1913 * wakes up the poll waiters.
1915 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
1916 struct fuse_notify_poll_wakeup_out
*outarg
)
1918 u64 kh
= outarg
->kh
;
1919 struct rb_node
**link
;
1921 spin_lock(&fc
->lock
);
1923 link
= fuse_find_polled_node(fc
, kh
, NULL
);
1925 struct fuse_file
*ff
;
1927 ff
= rb_entry(*link
, struct fuse_file
, polled_node
);
1928 wake_up_interruptible_sync(&ff
->poll_wait
);
1931 spin_unlock(&fc
->lock
);
1935 static const struct file_operations fuse_file_operations
= {
1936 .llseek
= fuse_file_llseek
,
1937 .read
= do_sync_read
,
1938 .aio_read
= fuse_file_aio_read
,
1939 .write
= do_sync_write
,
1940 .aio_write
= fuse_file_aio_write
,
1941 .mmap
= fuse_file_mmap
,
1943 .flush
= fuse_flush
,
1944 .release
= fuse_release
,
1945 .fsync
= fuse_fsync
,
1946 .lock
= fuse_file_lock
,
1947 .flock
= fuse_file_flock
,
1948 .splice_read
= generic_file_splice_read
,
1949 .unlocked_ioctl
= fuse_file_ioctl
,
1950 .compat_ioctl
= fuse_file_compat_ioctl
,
1951 .poll
= fuse_file_poll
,
1954 static const struct file_operations fuse_direct_io_file_operations
= {
1955 .llseek
= fuse_file_llseek
,
1956 .read
= fuse_direct_read
,
1957 .write
= fuse_direct_write
,
1958 .mmap
= fuse_direct_mmap
,
1960 .flush
= fuse_flush
,
1961 .release
= fuse_release
,
1962 .fsync
= fuse_fsync
,
1963 .lock
= fuse_file_lock
,
1964 .flock
= fuse_file_flock
,
1965 .unlocked_ioctl
= fuse_file_ioctl
,
1966 .compat_ioctl
= fuse_file_compat_ioctl
,
1967 .poll
= fuse_file_poll
,
1968 /* no splice_read */
1971 static const struct address_space_operations fuse_file_aops
= {
1972 .readpage
= fuse_readpage
,
1973 .writepage
= fuse_writepage
,
1974 .launder_page
= fuse_launder_page
,
1975 .write_begin
= fuse_write_begin
,
1976 .write_end
= fuse_write_end
,
1977 .readpages
= fuse_readpages
,
1978 .set_page_dirty
= __set_page_dirty_nobuffers
,
1982 void fuse_init_file_inode(struct inode
*inode
)
1984 inode
->i_fop
= &fuse_file_operations
;
1985 inode
->i_data
.a_ops
= &fuse_file_aops
;