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 request_send(fc
, req
);
43 err
= req
->out
.h
.error
;
44 fuse_put_request(fc
, req
);
49 struct fuse_file
*fuse_file_alloc(void)
52 ff
= kmalloc(sizeof(struct fuse_file
), GFP_KERNEL
);
54 ff
->reserved_req
= fuse_request_alloc();
55 if (!ff
->reserved_req
) {
59 INIT_LIST_HEAD(&ff
->write_entry
);
60 atomic_set(&ff
->count
, 0);
66 void fuse_file_free(struct fuse_file
*ff
)
68 fuse_request_free(ff
->reserved_req
);
72 static struct fuse_file
*fuse_file_get(struct fuse_file
*ff
)
74 atomic_inc(&ff
->count
);
78 static void fuse_release_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
80 dput(req
->misc
.release
.dentry
);
81 mntput(req
->misc
.release
.vfsmount
);
84 static void fuse_file_put(struct fuse_file
*ff
)
86 if (atomic_dec_and_test(&ff
->count
)) {
87 struct fuse_req
*req
= ff
->reserved_req
;
88 struct inode
*inode
= req
->misc
.release
.dentry
->d_inode
;
89 struct fuse_conn
*fc
= get_fuse_conn(inode
);
90 req
->end
= fuse_release_end
;
91 request_send_background(fc
, req
);
96 void fuse_finish_open(struct inode
*inode
, struct file
*file
,
97 struct fuse_file
*ff
, struct fuse_open_out
*outarg
)
99 if (outarg
->open_flags
& FOPEN_DIRECT_IO
)
100 file
->f_op
= &fuse_direct_io_file_operations
;
101 if (!(outarg
->open_flags
& FOPEN_KEEP_CACHE
))
102 invalidate_inode_pages2(inode
->i_mapping
);
103 if (outarg
->open_flags
& FOPEN_NONSEEKABLE
)
104 nonseekable_open(inode
, file
);
106 file
->private_data
= fuse_file_get(ff
);
109 int fuse_open_common(struct inode
*inode
, struct file
*file
, int isdir
)
111 struct fuse_open_out outarg
;
112 struct fuse_file
*ff
;
115 /* VFS checks this, but only _after_ ->open() */
116 if (file
->f_flags
& O_DIRECT
)
119 err
= generic_file_open(inode
, file
);
123 ff
= fuse_file_alloc();
127 err
= fuse_send_open(inode
, file
, isdir
, &outarg
);
132 outarg
.open_flags
&= ~FOPEN_DIRECT_IO
;
133 fuse_finish_open(inode
, file
, ff
, &outarg
);
139 void fuse_release_fill(struct fuse_file
*ff
, u64 nodeid
, int flags
, int opcode
)
141 struct fuse_req
*req
= ff
->reserved_req
;
142 struct fuse_release_in
*inarg
= &req
->misc
.release
.in
;
145 inarg
->flags
= flags
;
146 req
->in
.h
.opcode
= opcode
;
147 req
->in
.h
.nodeid
= nodeid
;
149 req
->in
.args
[0].size
= sizeof(struct fuse_release_in
);
150 req
->in
.args
[0].value
= inarg
;
153 int fuse_release_common(struct inode
*inode
, struct file
*file
, int isdir
)
155 struct fuse_file
*ff
= file
->private_data
;
157 struct fuse_conn
*fc
= get_fuse_conn(inode
);
158 struct fuse_req
*req
= ff
->reserved_req
;
160 fuse_release_fill(ff
, get_node_id(inode
), file
->f_flags
,
161 isdir
? FUSE_RELEASEDIR
: FUSE_RELEASE
);
163 /* Hold vfsmount and dentry until release is finished */
164 req
->misc
.release
.vfsmount
= mntget(file
->f_path
.mnt
);
165 req
->misc
.release
.dentry
= dget(file
->f_path
.dentry
);
167 spin_lock(&fc
->lock
);
168 list_del(&ff
->write_entry
);
169 spin_unlock(&fc
->lock
);
171 * Normally this will send the RELEASE request,
172 * however if some asynchronous READ or WRITE requests
173 * are outstanding, the sending will be delayed
178 /* Return value is ignored by VFS */
182 static int fuse_open(struct inode
*inode
, struct file
*file
)
184 return fuse_open_common(inode
, file
, 0);
187 static int fuse_release(struct inode
*inode
, struct file
*file
)
189 return fuse_release_common(inode
, file
, 0);
193 * Scramble the ID space with XTEA, so that the value of the files_struct
194 * pointer is not exposed to userspace.
196 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
)
198 u32
*k
= fc
->scramble_key
;
199 u64 v
= (unsigned long) id
;
205 for (i
= 0; i
< 32; i
++) {
206 v0
+= ((v1
<< 4 ^ v1
>> 5) + v1
) ^ (sum
+ k
[sum
& 3]);
208 v1
+= ((v0
<< 4 ^ v0
>> 5) + v0
) ^ (sum
+ k
[sum
>>11 & 3]);
211 return (u64
) v0
+ ((u64
) v1
<< 32);
215 * Check if page is under writeback
217 * This is currently done by walking the list of writepage requests
218 * for the inode, which can be pretty inefficient.
220 static bool fuse_page_is_writeback(struct inode
*inode
, pgoff_t index
)
222 struct fuse_conn
*fc
= get_fuse_conn(inode
);
223 struct fuse_inode
*fi
= get_fuse_inode(inode
);
224 struct fuse_req
*req
;
227 spin_lock(&fc
->lock
);
228 list_for_each_entry(req
, &fi
->writepages
, writepages_entry
) {
231 BUG_ON(req
->inode
!= inode
);
232 curr_index
= req
->misc
.write
.in
.offset
>> PAGE_CACHE_SHIFT
;
233 if (curr_index
== index
) {
238 spin_unlock(&fc
->lock
);
244 * Wait for page writeback to be completed.
246 * Since fuse doesn't rely on the VM writeback tracking, this has to
247 * use some other means.
249 static int fuse_wait_on_page_writeback(struct inode
*inode
, pgoff_t index
)
251 struct fuse_inode
*fi
= get_fuse_inode(inode
);
253 wait_event(fi
->page_waitq
, !fuse_page_is_writeback(inode
, index
));
257 static int fuse_flush(struct file
*file
, fl_owner_t id
)
259 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
260 struct fuse_conn
*fc
= get_fuse_conn(inode
);
261 struct fuse_file
*ff
= file
->private_data
;
262 struct fuse_req
*req
;
263 struct fuse_flush_in inarg
;
266 if (is_bad_inode(inode
))
272 req
= fuse_get_req_nofail(fc
, file
);
273 memset(&inarg
, 0, sizeof(inarg
));
275 inarg
.lock_owner
= fuse_lock_owner_id(fc
, id
);
276 req
->in
.h
.opcode
= FUSE_FLUSH
;
277 req
->in
.h
.nodeid
= get_node_id(inode
);
279 req
->in
.args
[0].size
= sizeof(inarg
);
280 req
->in
.args
[0].value
= &inarg
;
282 request_send(fc
, req
);
283 err
= req
->out
.h
.error
;
284 fuse_put_request(fc
, req
);
285 if (err
== -ENOSYS
) {
293 * Wait for all pending writepages on the inode to finish.
295 * This is currently done by blocking further writes with FUSE_NOWRITE
296 * and waiting for all sent writes to complete.
298 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
299 * could conflict with truncation.
301 static void fuse_sync_writes(struct inode
*inode
)
303 fuse_set_nowrite(inode
);
304 fuse_release_nowrite(inode
);
307 int fuse_fsync_common(struct file
*file
, struct dentry
*de
, int datasync
,
310 struct inode
*inode
= de
->d_inode
;
311 struct fuse_conn
*fc
= get_fuse_conn(inode
);
312 struct fuse_file
*ff
= file
->private_data
;
313 struct fuse_req
*req
;
314 struct fuse_fsync_in inarg
;
317 if (is_bad_inode(inode
))
320 if ((!isdir
&& fc
->no_fsync
) || (isdir
&& fc
->no_fsyncdir
))
324 * Start writeback against all dirty pages of the inode, then
325 * wait for all outstanding writes, before sending the FSYNC
328 err
= write_inode_now(inode
, 0);
332 fuse_sync_writes(inode
);
334 req
= fuse_get_req(fc
);
338 memset(&inarg
, 0, sizeof(inarg
));
340 inarg
.fsync_flags
= datasync
? 1 : 0;
341 req
->in
.h
.opcode
= isdir
? FUSE_FSYNCDIR
: FUSE_FSYNC
;
342 req
->in
.h
.nodeid
= get_node_id(inode
);
344 req
->in
.args
[0].size
= sizeof(inarg
);
345 req
->in
.args
[0].value
= &inarg
;
346 request_send(fc
, req
);
347 err
= req
->out
.h
.error
;
348 fuse_put_request(fc
, req
);
349 if (err
== -ENOSYS
) {
359 static int fuse_fsync(struct file
*file
, struct dentry
*de
, int datasync
)
361 return fuse_fsync_common(file
, de
, datasync
, 0);
364 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
365 struct inode
*inode
, loff_t pos
, size_t count
, int opcode
)
367 struct fuse_read_in
*inarg
= &req
->misc
.read
.in
;
368 struct fuse_file
*ff
= file
->private_data
;
373 inarg
->flags
= file
->f_flags
;
374 req
->in
.h
.opcode
= opcode
;
375 req
->in
.h
.nodeid
= get_node_id(inode
);
377 req
->in
.args
[0].size
= sizeof(struct fuse_read_in
);
378 req
->in
.args
[0].value
= inarg
;
379 req
->out
.argpages
= 1;
381 req
->out
.numargs
= 1;
382 req
->out
.args
[0].size
= count
;
385 static size_t fuse_send_read(struct fuse_req
*req
, struct file
*file
,
386 struct inode
*inode
, loff_t pos
, size_t count
,
389 struct fuse_conn
*fc
= get_fuse_conn(inode
);
391 fuse_read_fill(req
, file
, inode
, pos
, count
, FUSE_READ
);
393 struct fuse_read_in
*inarg
= &req
->misc
.read
.in
;
395 inarg
->read_flags
|= FUSE_READ_LOCKOWNER
;
396 inarg
->lock_owner
= fuse_lock_owner_id(fc
, owner
);
398 request_send(fc
, req
);
399 return req
->out
.args
[0].size
;
402 static void fuse_read_update_size(struct inode
*inode
, loff_t size
,
405 struct fuse_conn
*fc
= get_fuse_conn(inode
);
406 struct fuse_inode
*fi
= get_fuse_inode(inode
);
408 spin_lock(&fc
->lock
);
409 if (attr_ver
== fi
->attr_version
&& size
< inode
->i_size
) {
410 fi
->attr_version
= ++fc
->attr_version
;
411 i_size_write(inode
, size
);
413 spin_unlock(&fc
->lock
);
416 static int fuse_readpage(struct file
*file
, struct page
*page
)
418 struct inode
*inode
= page
->mapping
->host
;
419 struct fuse_conn
*fc
= get_fuse_conn(inode
);
420 struct fuse_req
*req
;
422 loff_t pos
= page_offset(page
);
423 size_t count
= PAGE_CACHE_SIZE
;
428 if (is_bad_inode(inode
))
432 * Page writeback can extend beyond the liftime of the
433 * page-cache page, so make sure we read a properly synced
436 fuse_wait_on_page_writeback(inode
, page
->index
);
438 req
= fuse_get_req(fc
);
443 attr_ver
= fuse_get_attr_version(fc
);
445 req
->out
.page_zeroing
= 1;
447 req
->pages
[0] = page
;
448 num_read
= fuse_send_read(req
, file
, inode
, pos
, count
, NULL
);
449 err
= req
->out
.h
.error
;
450 fuse_put_request(fc
, req
);
454 * Short read means EOF. If file size is larger, truncate it
456 if (num_read
< count
)
457 fuse_read_update_size(inode
, pos
+ num_read
, attr_ver
);
459 SetPageUptodate(page
);
462 fuse_invalidate_attr(inode
); /* atime changed */
468 static void fuse_readpages_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
471 size_t count
= req
->misc
.read
.in
.size
;
472 size_t num_read
= req
->out
.args
[0].size
;
473 struct inode
*inode
= req
->pages
[0]->mapping
->host
;
476 * Short read means EOF. If file size is larger, truncate it
478 if (!req
->out
.h
.error
&& num_read
< count
) {
479 loff_t pos
= page_offset(req
->pages
[0]) + num_read
;
480 fuse_read_update_size(inode
, pos
, req
->misc
.read
.attr_ver
);
483 fuse_invalidate_attr(inode
); /* atime changed */
485 for (i
= 0; i
< req
->num_pages
; i
++) {
486 struct page
*page
= req
->pages
[i
];
487 if (!req
->out
.h
.error
)
488 SetPageUptodate(page
);
494 fuse_file_put(req
->ff
);
497 static void fuse_send_readpages(struct fuse_req
*req
, struct file
*file
,
500 struct fuse_conn
*fc
= get_fuse_conn(inode
);
501 loff_t pos
= page_offset(req
->pages
[0]);
502 size_t count
= req
->num_pages
<< PAGE_CACHE_SHIFT
;
503 req
->out
.page_zeroing
= 1;
504 fuse_read_fill(req
, file
, inode
, pos
, count
, FUSE_READ
);
505 req
->misc
.read
.attr_ver
= fuse_get_attr_version(fc
);
506 if (fc
->async_read
) {
507 struct fuse_file
*ff
= file
->private_data
;
508 req
->ff
= fuse_file_get(ff
);
509 req
->end
= fuse_readpages_end
;
510 request_send_background(fc
, req
);
512 request_send(fc
, req
);
513 fuse_readpages_end(fc
, req
);
514 fuse_put_request(fc
, req
);
518 struct fuse_fill_data
{
519 struct fuse_req
*req
;
524 static int fuse_readpages_fill(void *_data
, struct page
*page
)
526 struct fuse_fill_data
*data
= _data
;
527 struct fuse_req
*req
= data
->req
;
528 struct inode
*inode
= data
->inode
;
529 struct fuse_conn
*fc
= get_fuse_conn(inode
);
531 fuse_wait_on_page_writeback(inode
, page
->index
);
533 if (req
->num_pages
&&
534 (req
->num_pages
== FUSE_MAX_PAGES_PER_REQ
||
535 (req
->num_pages
+ 1) * PAGE_CACHE_SIZE
> fc
->max_read
||
536 req
->pages
[req
->num_pages
- 1]->index
+ 1 != page
->index
)) {
537 fuse_send_readpages(req
, data
->file
, inode
);
538 data
->req
= req
= fuse_get_req(fc
);
544 req
->pages
[req
->num_pages
] = page
;
549 static int fuse_readpages(struct file
*file
, struct address_space
*mapping
,
550 struct list_head
*pages
, unsigned nr_pages
)
552 struct inode
*inode
= mapping
->host
;
553 struct fuse_conn
*fc
= get_fuse_conn(inode
);
554 struct fuse_fill_data data
;
558 if (is_bad_inode(inode
))
563 data
.req
= fuse_get_req(fc
);
564 err
= PTR_ERR(data
.req
);
565 if (IS_ERR(data
.req
))
568 err
= read_cache_pages(mapping
, pages
, fuse_readpages_fill
, &data
);
570 if (data
.req
->num_pages
)
571 fuse_send_readpages(data
.req
, file
, inode
);
573 fuse_put_request(fc
, data
.req
);
579 static ssize_t
fuse_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
580 unsigned long nr_segs
, loff_t pos
)
582 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
584 if (pos
+ iov_length(iov
, nr_segs
) > i_size_read(inode
)) {
587 * If trying to read past EOF, make sure the i_size
588 * attribute is up-to-date.
590 err
= fuse_update_attributes(inode
, NULL
, iocb
->ki_filp
, NULL
);
595 return generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
598 static void fuse_write_fill(struct fuse_req
*req
, struct file
*file
,
599 struct fuse_file
*ff
, struct inode
*inode
,
600 loff_t pos
, size_t count
, int writepage
)
602 struct fuse_conn
*fc
= get_fuse_conn(inode
);
603 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
604 struct fuse_write_out
*outarg
= &req
->misc
.write
.out
;
606 memset(inarg
, 0, sizeof(struct fuse_write_in
));
610 inarg
->write_flags
= writepage
? FUSE_WRITE_CACHE
: 0;
611 inarg
->flags
= file
? file
->f_flags
: 0;
612 req
->in
.h
.opcode
= FUSE_WRITE
;
613 req
->in
.h
.nodeid
= get_node_id(inode
);
614 req
->in
.argpages
= 1;
617 req
->in
.args
[0].size
= FUSE_COMPAT_WRITE_IN_SIZE
;
619 req
->in
.args
[0].size
= sizeof(struct fuse_write_in
);
620 req
->in
.args
[0].value
= inarg
;
621 req
->in
.args
[1].size
= count
;
622 req
->out
.numargs
= 1;
623 req
->out
.args
[0].size
= sizeof(struct fuse_write_out
);
624 req
->out
.args
[0].value
= outarg
;
627 static size_t fuse_send_write(struct fuse_req
*req
, struct file
*file
,
628 struct inode
*inode
, loff_t pos
, size_t count
,
631 struct fuse_conn
*fc
= get_fuse_conn(inode
);
632 fuse_write_fill(req
, file
, file
->private_data
, inode
, pos
, count
, 0);
634 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
635 inarg
->write_flags
|= FUSE_WRITE_LOCKOWNER
;
636 inarg
->lock_owner
= fuse_lock_owner_id(fc
, owner
);
638 request_send(fc
, req
);
639 return req
->misc
.write
.out
.size
;
642 static int fuse_write_begin(struct file
*file
, struct address_space
*mapping
,
643 loff_t pos
, unsigned len
, unsigned flags
,
644 struct page
**pagep
, void **fsdata
)
646 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
648 *pagep
= __grab_cache_page(mapping
, index
);
654 static void fuse_write_update_size(struct inode
*inode
, loff_t pos
)
656 struct fuse_conn
*fc
= get_fuse_conn(inode
);
657 struct fuse_inode
*fi
= get_fuse_inode(inode
);
659 spin_lock(&fc
->lock
);
660 fi
->attr_version
= ++fc
->attr_version
;
661 if (pos
> inode
->i_size
)
662 i_size_write(inode
, pos
);
663 spin_unlock(&fc
->lock
);
666 static int fuse_buffered_write(struct file
*file
, struct inode
*inode
,
667 loff_t pos
, unsigned count
, struct page
*page
)
671 struct fuse_conn
*fc
= get_fuse_conn(inode
);
672 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
673 struct fuse_req
*req
;
675 if (is_bad_inode(inode
))
679 * Make sure writepages on the same page are not mixed up with
682 fuse_wait_on_page_writeback(inode
, page
->index
);
684 req
= fuse_get_req(fc
);
689 req
->pages
[0] = page
;
690 req
->page_offset
= offset
;
691 nres
= fuse_send_write(req
, file
, inode
, pos
, count
, NULL
);
692 err
= req
->out
.h
.error
;
693 fuse_put_request(fc
, req
);
698 fuse_write_update_size(inode
, pos
);
699 if (count
== PAGE_CACHE_SIZE
)
700 SetPageUptodate(page
);
702 fuse_invalidate_attr(inode
);
703 return err
? err
: nres
;
706 static int fuse_write_end(struct file
*file
, struct address_space
*mapping
,
707 loff_t pos
, unsigned len
, unsigned copied
,
708 struct page
*page
, void *fsdata
)
710 struct inode
*inode
= mapping
->host
;
714 res
= fuse_buffered_write(file
, inode
, pos
, copied
, page
);
717 page_cache_release(page
);
721 static size_t fuse_send_write_pages(struct fuse_req
*req
, struct file
*file
,
722 struct inode
*inode
, loff_t pos
,
729 for (i
= 0; i
< req
->num_pages
; i
++)
730 fuse_wait_on_page_writeback(inode
, req
->pages
[i
]->index
);
732 res
= fuse_send_write(req
, file
, inode
, pos
, count
, NULL
);
734 offset
= req
->page_offset
;
736 for (i
= 0; i
< req
->num_pages
; i
++) {
737 struct page
*page
= req
->pages
[i
];
739 if (!req
->out
.h
.error
&& !offset
&& count
>= PAGE_CACHE_SIZE
)
740 SetPageUptodate(page
);
742 if (count
> PAGE_CACHE_SIZE
- offset
)
743 count
-= PAGE_CACHE_SIZE
- offset
;
749 page_cache_release(page
);
755 static ssize_t
fuse_fill_write_pages(struct fuse_req
*req
,
756 struct address_space
*mapping
,
757 struct iov_iter
*ii
, loff_t pos
)
759 struct fuse_conn
*fc
= get_fuse_conn(mapping
->host
);
760 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
764 req
->page_offset
= offset
;
769 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
770 size_t bytes
= min_t(size_t, PAGE_CACHE_SIZE
- offset
,
773 bytes
= min_t(size_t, bytes
, fc
->max_write
- count
);
777 if (iov_iter_fault_in_readable(ii
, bytes
))
781 page
= __grab_cache_page(mapping
, index
);
786 tmp
= iov_iter_copy_from_user_atomic(page
, ii
, offset
, bytes
);
788 flush_dcache_page(page
);
792 page_cache_release(page
);
793 bytes
= min(bytes
, iov_iter_single_seg_count(ii
));
798 req
->pages
[req
->num_pages
] = page
;
801 iov_iter_advance(ii
, tmp
);
805 if (offset
== PAGE_CACHE_SIZE
)
810 } while (iov_iter_count(ii
) && count
< fc
->max_write
&&
811 req
->num_pages
< FUSE_MAX_PAGES_PER_REQ
&& offset
== 0);
813 return count
> 0 ? count
: err
;
816 static ssize_t
fuse_perform_write(struct file
*file
,
817 struct address_space
*mapping
,
818 struct iov_iter
*ii
, loff_t pos
)
820 struct inode
*inode
= mapping
->host
;
821 struct fuse_conn
*fc
= get_fuse_conn(inode
);
825 if (is_bad_inode(inode
))
829 struct fuse_req
*req
;
832 req
= fuse_get_req(fc
);
838 count
= fuse_fill_write_pages(req
, mapping
, ii
, pos
);
844 num_written
= fuse_send_write_pages(req
, file
, inode
,
846 err
= req
->out
.h
.error
;
851 /* break out of the loop on short write */
852 if (num_written
!= count
)
856 fuse_put_request(fc
, req
);
857 } while (!err
&& iov_iter_count(ii
));
860 fuse_write_update_size(inode
, pos
);
862 fuse_invalidate_attr(inode
);
864 return res
> 0 ? res
: err
;
867 static ssize_t
fuse_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
868 unsigned long nr_segs
, loff_t pos
)
870 struct file
*file
= iocb
->ki_filp
;
871 struct address_space
*mapping
= file
->f_mapping
;
874 struct inode
*inode
= mapping
->host
;
878 WARN_ON(iocb
->ki_pos
!= pos
);
880 err
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_READ
);
884 mutex_lock(&inode
->i_mutex
);
885 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
887 /* We can write back this queue in page reclaim */
888 current
->backing_dev_info
= mapping
->backing_dev_info
;
890 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
897 err
= file_remove_suid(file
);
901 file_update_time(file
);
903 iov_iter_init(&i
, iov
, nr_segs
, count
, 0);
904 written
= fuse_perform_write(file
, mapping
, &i
, pos
);
906 iocb
->ki_pos
= pos
+ written
;
909 current
->backing_dev_info
= NULL
;
910 mutex_unlock(&inode
->i_mutex
);
912 return written
? written
: err
;
915 static void fuse_release_user_pages(struct fuse_req
*req
, int write
)
919 for (i
= 0; i
< req
->num_pages
; i
++) {
920 struct page
*page
= req
->pages
[i
];
922 set_page_dirty_lock(page
);
927 static int fuse_get_user_pages(struct fuse_req
*req
, const char __user
*buf
,
928 unsigned nbytes
, int write
)
930 unsigned long user_addr
= (unsigned long) buf
;
931 unsigned offset
= user_addr
& ~PAGE_MASK
;
934 /* This doesn't work with nfsd */
938 nbytes
= min(nbytes
, (unsigned) FUSE_MAX_PAGES_PER_REQ
<< PAGE_SHIFT
);
939 npages
= (nbytes
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
940 npages
= clamp(npages
, 1, FUSE_MAX_PAGES_PER_REQ
);
941 down_read(¤t
->mm
->mmap_sem
);
942 npages
= get_user_pages(current
, current
->mm
, user_addr
, npages
, write
,
943 0, req
->pages
, NULL
);
944 up_read(¤t
->mm
->mmap_sem
);
948 req
->num_pages
= npages
;
949 req
->page_offset
= offset
;
953 static ssize_t
fuse_direct_io(struct file
*file
, const char __user
*buf
,
954 size_t count
, loff_t
*ppos
, int write
)
956 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
957 struct fuse_conn
*fc
= get_fuse_conn(inode
);
958 size_t nmax
= write
? fc
->max_write
: fc
->max_read
;
961 struct fuse_req
*req
;
963 if (is_bad_inode(inode
))
966 req
= fuse_get_req(fc
);
972 size_t nbytes_limit
= min(count
, nmax
);
974 int err
= fuse_get_user_pages(req
, buf
, nbytes_limit
, !write
);
979 nbytes
= (req
->num_pages
<< PAGE_SHIFT
) - req
->page_offset
;
980 nbytes
= min(nbytes_limit
, nbytes
);
982 nres
= fuse_send_write(req
, file
, inode
, pos
, nbytes
,
985 nres
= fuse_send_read(req
, file
, inode
, pos
, nbytes
,
987 fuse_release_user_pages(req
, !write
);
988 if (req
->out
.h
.error
) {
990 res
= req
->out
.h
.error
;
992 } else if (nres
> nbytes
) {
1003 fuse_put_request(fc
, req
);
1004 req
= fuse_get_req(fc
);
1009 fuse_put_request(fc
, req
);
1012 fuse_write_update_size(inode
, pos
);
1015 fuse_invalidate_attr(inode
);
1020 static ssize_t
fuse_direct_read(struct file
*file
, char __user
*buf
,
1021 size_t count
, loff_t
*ppos
)
1023 return fuse_direct_io(file
, buf
, count
, ppos
, 0);
1026 static ssize_t
fuse_direct_write(struct file
*file
, const char __user
*buf
,
1027 size_t count
, loff_t
*ppos
)
1029 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1031 /* Don't allow parallel writes to the same file */
1032 mutex_lock(&inode
->i_mutex
);
1033 res
= generic_write_checks(file
, ppos
, &count
, 0);
1035 res
= fuse_direct_io(file
, buf
, count
, ppos
, 1);
1036 mutex_unlock(&inode
->i_mutex
);
1040 static void fuse_writepage_free(struct fuse_conn
*fc
, struct fuse_req
*req
)
1042 __free_page(req
->pages
[0]);
1043 fuse_file_put(req
->ff
);
1046 static void fuse_writepage_finish(struct fuse_conn
*fc
, struct fuse_req
*req
)
1048 struct inode
*inode
= req
->inode
;
1049 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1050 struct backing_dev_info
*bdi
= inode
->i_mapping
->backing_dev_info
;
1052 list_del(&req
->writepages_entry
);
1053 dec_bdi_stat(bdi
, BDI_WRITEBACK
);
1054 dec_zone_page_state(req
->pages
[0], NR_WRITEBACK_TEMP
);
1055 bdi_writeout_inc(bdi
);
1056 wake_up(&fi
->page_waitq
);
1059 /* Called under fc->lock, may release and reacquire it */
1060 static void fuse_send_writepage(struct fuse_conn
*fc
, struct fuse_req
*req
)
1062 struct fuse_inode
*fi
= get_fuse_inode(req
->inode
);
1063 loff_t size
= i_size_read(req
->inode
);
1064 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
1069 if (inarg
->offset
+ PAGE_CACHE_SIZE
<= size
) {
1070 inarg
->size
= PAGE_CACHE_SIZE
;
1071 } else if (inarg
->offset
< size
) {
1072 inarg
->size
= size
& (PAGE_CACHE_SIZE
- 1);
1074 /* Got truncated off completely */
1078 req
->in
.args
[1].size
= inarg
->size
;
1080 request_send_background_locked(fc
, req
);
1084 fuse_writepage_finish(fc
, req
);
1085 spin_unlock(&fc
->lock
);
1086 fuse_writepage_free(fc
, req
);
1087 fuse_put_request(fc
, req
);
1088 spin_lock(&fc
->lock
);
1092 * If fi->writectr is positive (no truncate or fsync going on) send
1093 * all queued writepage requests.
1095 * Called with fc->lock
1097 void fuse_flush_writepages(struct inode
*inode
)
1099 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1100 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1101 struct fuse_req
*req
;
1103 while (fi
->writectr
>= 0 && !list_empty(&fi
->queued_writes
)) {
1104 req
= list_entry(fi
->queued_writes
.next
, struct fuse_req
, list
);
1105 list_del_init(&req
->list
);
1106 fuse_send_writepage(fc
, req
);
1110 static void fuse_writepage_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1112 struct inode
*inode
= req
->inode
;
1113 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1115 mapping_set_error(inode
->i_mapping
, req
->out
.h
.error
);
1116 spin_lock(&fc
->lock
);
1118 fuse_writepage_finish(fc
, req
);
1119 spin_unlock(&fc
->lock
);
1120 fuse_writepage_free(fc
, req
);
1123 static int fuse_writepage_locked(struct page
*page
)
1125 struct address_space
*mapping
= page
->mapping
;
1126 struct inode
*inode
= mapping
->host
;
1127 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1128 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1129 struct fuse_req
*req
;
1130 struct fuse_file
*ff
;
1131 struct page
*tmp_page
;
1133 set_page_writeback(page
);
1135 req
= fuse_request_alloc_nofs();
1139 tmp_page
= alloc_page(GFP_NOFS
| __GFP_HIGHMEM
);
1143 spin_lock(&fc
->lock
);
1144 BUG_ON(list_empty(&fi
->write_files
));
1145 ff
= list_entry(fi
->write_files
.next
, struct fuse_file
, write_entry
);
1146 req
->ff
= fuse_file_get(ff
);
1147 spin_unlock(&fc
->lock
);
1149 fuse_write_fill(req
, NULL
, ff
, inode
, page_offset(page
), 0, 1);
1151 copy_highpage(tmp_page
, page
);
1153 req
->pages
[0] = tmp_page
;
1154 req
->page_offset
= 0;
1155 req
->end
= fuse_writepage_end
;
1158 inc_bdi_stat(mapping
->backing_dev_info
, BDI_WRITEBACK
);
1159 inc_zone_page_state(tmp_page
, NR_WRITEBACK_TEMP
);
1160 end_page_writeback(page
);
1162 spin_lock(&fc
->lock
);
1163 list_add(&req
->writepages_entry
, &fi
->writepages
);
1164 list_add_tail(&req
->list
, &fi
->queued_writes
);
1165 fuse_flush_writepages(inode
);
1166 spin_unlock(&fc
->lock
);
1171 fuse_request_free(req
);
1173 end_page_writeback(page
);
1177 static int fuse_writepage(struct page
*page
, struct writeback_control
*wbc
)
1181 err
= fuse_writepage_locked(page
);
1187 static int fuse_launder_page(struct page
*page
)
1190 if (clear_page_dirty_for_io(page
)) {
1191 struct inode
*inode
= page
->mapping
->host
;
1192 err
= fuse_writepage_locked(page
);
1194 fuse_wait_on_page_writeback(inode
, page
->index
);
1200 * Write back dirty pages now, because there may not be any suitable
1203 static void fuse_vma_close(struct vm_area_struct
*vma
)
1205 filemap_write_and_wait(vma
->vm_file
->f_mapping
);
1209 * Wait for writeback against this page to complete before allowing it
1210 * to be marked dirty again, and hence written back again, possibly
1211 * before the previous writepage completed.
1213 * Block here, instead of in ->writepage(), so that the userspace fs
1214 * can only block processes actually operating on the filesystem.
1216 * Otherwise unprivileged userspace fs would be able to block
1221 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1223 static int fuse_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
1226 * Don't use page->mapping as it may become NULL from a
1227 * concurrent truncate.
1229 struct inode
*inode
= vma
->vm_file
->f_mapping
->host
;
1231 fuse_wait_on_page_writeback(inode
, page
->index
);
1235 static struct vm_operations_struct fuse_file_vm_ops
= {
1236 .close
= fuse_vma_close
,
1237 .fault
= filemap_fault
,
1238 .page_mkwrite
= fuse_page_mkwrite
,
1241 static int fuse_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1243 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
)) {
1244 struct inode
*inode
= file
->f_dentry
->d_inode
;
1245 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1246 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1247 struct fuse_file
*ff
= file
->private_data
;
1249 * file may be written through mmap, so chain it onto the
1250 * inodes's write_file list
1252 spin_lock(&fc
->lock
);
1253 if (list_empty(&ff
->write_entry
))
1254 list_add(&ff
->write_entry
, &fi
->write_files
);
1255 spin_unlock(&fc
->lock
);
1257 file_accessed(file
);
1258 vma
->vm_ops
= &fuse_file_vm_ops
;
1262 static int convert_fuse_file_lock(const struct fuse_file_lock
*ffl
,
1263 struct file_lock
*fl
)
1265 switch (ffl
->type
) {
1271 if (ffl
->start
> OFFSET_MAX
|| ffl
->end
> OFFSET_MAX
||
1272 ffl
->end
< ffl
->start
)
1275 fl
->fl_start
= ffl
->start
;
1276 fl
->fl_end
= ffl
->end
;
1277 fl
->fl_pid
= ffl
->pid
;
1283 fl
->fl_type
= ffl
->type
;
1287 static void fuse_lk_fill(struct fuse_req
*req
, struct file
*file
,
1288 const struct file_lock
*fl
, int opcode
, pid_t pid
,
1291 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1292 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1293 struct fuse_file
*ff
= file
->private_data
;
1294 struct fuse_lk_in
*arg
= &req
->misc
.lk_in
;
1297 arg
->owner
= fuse_lock_owner_id(fc
, fl
->fl_owner
);
1298 arg
->lk
.start
= fl
->fl_start
;
1299 arg
->lk
.end
= fl
->fl_end
;
1300 arg
->lk
.type
= fl
->fl_type
;
1303 arg
->lk_flags
|= FUSE_LK_FLOCK
;
1304 req
->in
.h
.opcode
= opcode
;
1305 req
->in
.h
.nodeid
= get_node_id(inode
);
1306 req
->in
.numargs
= 1;
1307 req
->in
.args
[0].size
= sizeof(*arg
);
1308 req
->in
.args
[0].value
= arg
;
1311 static int fuse_getlk(struct file
*file
, struct file_lock
*fl
)
1313 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1314 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1315 struct fuse_req
*req
;
1316 struct fuse_lk_out outarg
;
1319 req
= fuse_get_req(fc
);
1321 return PTR_ERR(req
);
1323 fuse_lk_fill(req
, file
, fl
, FUSE_GETLK
, 0, 0);
1324 req
->out
.numargs
= 1;
1325 req
->out
.args
[0].size
= sizeof(outarg
);
1326 req
->out
.args
[0].value
= &outarg
;
1327 request_send(fc
, req
);
1328 err
= req
->out
.h
.error
;
1329 fuse_put_request(fc
, req
);
1331 err
= convert_fuse_file_lock(&outarg
.lk
, fl
);
1336 static int fuse_setlk(struct file
*file
, struct file_lock
*fl
, int flock
)
1338 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1339 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1340 struct fuse_req
*req
;
1341 int opcode
= (fl
->fl_flags
& FL_SLEEP
) ? FUSE_SETLKW
: FUSE_SETLK
;
1342 pid_t pid
= fl
->fl_type
!= F_UNLCK
? current
->tgid
: 0;
1345 if (fl
->fl_lmops
&& fl
->fl_lmops
->fl_grant
) {
1346 /* NLM needs asynchronous locks, which we don't support yet */
1350 /* Unlock on close is handled by the flush method */
1351 if (fl
->fl_flags
& FL_CLOSE
)
1354 req
= fuse_get_req(fc
);
1356 return PTR_ERR(req
);
1358 fuse_lk_fill(req
, file
, fl
, opcode
, pid
, flock
);
1359 request_send(fc
, req
);
1360 err
= req
->out
.h
.error
;
1361 /* locking is restartable */
1364 fuse_put_request(fc
, req
);
1368 static int fuse_file_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1370 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1371 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1374 if (cmd
== F_CANCELLK
) {
1376 } else if (cmd
== F_GETLK
) {
1378 posix_test_lock(file
, fl
);
1381 err
= fuse_getlk(file
, fl
);
1384 err
= posix_lock_file(file
, fl
, NULL
);
1386 err
= fuse_setlk(file
, fl
, 0);
1391 static int fuse_file_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1393 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1394 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1398 err
= flock_lock_file_wait(file
, fl
);
1400 /* emulate flock with POSIX locks */
1401 fl
->fl_owner
= (fl_owner_t
) file
;
1402 err
= fuse_setlk(file
, fl
, 1);
1408 static sector_t
fuse_bmap(struct address_space
*mapping
, sector_t block
)
1410 struct inode
*inode
= mapping
->host
;
1411 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1412 struct fuse_req
*req
;
1413 struct fuse_bmap_in inarg
;
1414 struct fuse_bmap_out outarg
;
1417 if (!inode
->i_sb
->s_bdev
|| fc
->no_bmap
)
1420 req
= fuse_get_req(fc
);
1424 memset(&inarg
, 0, sizeof(inarg
));
1425 inarg
.block
= block
;
1426 inarg
.blocksize
= inode
->i_sb
->s_blocksize
;
1427 req
->in
.h
.opcode
= FUSE_BMAP
;
1428 req
->in
.h
.nodeid
= get_node_id(inode
);
1429 req
->in
.numargs
= 1;
1430 req
->in
.args
[0].size
= sizeof(inarg
);
1431 req
->in
.args
[0].value
= &inarg
;
1432 req
->out
.numargs
= 1;
1433 req
->out
.args
[0].size
= sizeof(outarg
);
1434 req
->out
.args
[0].value
= &outarg
;
1435 request_send(fc
, req
);
1436 err
= req
->out
.h
.error
;
1437 fuse_put_request(fc
, req
);
1441 return err
? 0 : outarg
.block
;
1444 static loff_t
fuse_file_llseek(struct file
*file
, loff_t offset
, int origin
)
1447 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1449 mutex_lock(&inode
->i_mutex
);
1452 retval
= fuse_update_attributes(inode
, NULL
, file
, NULL
);
1455 offset
+= i_size_read(inode
);
1458 offset
+= file
->f_pos
;
1461 if (offset
>= 0 && offset
<= inode
->i_sb
->s_maxbytes
) {
1462 if (offset
!= file
->f_pos
) {
1463 file
->f_pos
= offset
;
1464 file
->f_version
= 0;
1468 mutex_unlock(&inode
->i_mutex
);
1472 static const struct file_operations fuse_file_operations
= {
1473 .llseek
= fuse_file_llseek
,
1474 .read
= do_sync_read
,
1475 .aio_read
= fuse_file_aio_read
,
1476 .write
= do_sync_write
,
1477 .aio_write
= fuse_file_aio_write
,
1478 .mmap
= fuse_file_mmap
,
1480 .flush
= fuse_flush
,
1481 .release
= fuse_release
,
1482 .fsync
= fuse_fsync
,
1483 .lock
= fuse_file_lock
,
1484 .flock
= fuse_file_flock
,
1485 .splice_read
= generic_file_splice_read
,
1488 static const struct file_operations fuse_direct_io_file_operations
= {
1489 .llseek
= fuse_file_llseek
,
1490 .read
= fuse_direct_read
,
1491 .write
= fuse_direct_write
,
1493 .flush
= fuse_flush
,
1494 .release
= fuse_release
,
1495 .fsync
= fuse_fsync
,
1496 .lock
= fuse_file_lock
,
1497 .flock
= fuse_file_flock
,
1498 /* no mmap and splice_read */
1501 static const struct address_space_operations fuse_file_aops
= {
1502 .readpage
= fuse_readpage
,
1503 .writepage
= fuse_writepage
,
1504 .launder_page
= fuse_launder_page
,
1505 .write_begin
= fuse_write_begin
,
1506 .write_end
= fuse_write_end
,
1507 .readpages
= fuse_readpages
,
1508 .set_page_dirty
= __set_page_dirty_nobuffers
,
1512 void fuse_init_file_inode(struct inode
*inode
)
1514 inode
->i_fop
= &fuse_file_operations
;
1515 inode
->i_data
.a_ops
= &fuse_file_aops
;