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>
15 #include <linux/module.h>
16 #include <linux/compat.h>
17 #include <linux/swap.h>
18 #include <linux/aio.h>
19 #include <linux/falloc.h>
21 static const struct file_operations fuse_direct_io_file_operations
;
23 static int fuse_send_open(struct fuse_conn
*fc
, u64 nodeid
, struct file
*file
,
24 int opcode
, struct fuse_open_out
*outargp
)
26 struct fuse_open_in inarg
;
30 req
= fuse_get_req_nopages(fc
);
34 memset(&inarg
, 0, sizeof(inarg
));
35 inarg
.flags
= file
->f_flags
& ~(O_CREAT
| O_EXCL
| O_NOCTTY
);
36 if (!fc
->atomic_o_trunc
)
37 inarg
.flags
&= ~O_TRUNC
;
38 req
->in
.h
.opcode
= opcode
;
39 req
->in
.h
.nodeid
= nodeid
;
41 req
->in
.args
[0].size
= sizeof(inarg
);
42 req
->in
.args
[0].value
= &inarg
;
44 req
->out
.args
[0].size
= sizeof(*outargp
);
45 req
->out
.args
[0].value
= outargp
;
46 fuse_request_send(fc
, req
);
47 err
= req
->out
.h
.error
;
48 fuse_put_request(fc
, req
);
53 struct fuse_file
*fuse_file_alloc(struct fuse_conn
*fc
)
57 ff
= kmalloc(sizeof(struct fuse_file
), GFP_KERNEL
);
62 ff
->reserved_req
= fuse_request_alloc(0);
63 if (unlikely(!ff
->reserved_req
)) {
68 INIT_LIST_HEAD(&ff
->write_entry
);
69 atomic_set(&ff
->count
, 0);
70 RB_CLEAR_NODE(&ff
->polled_node
);
71 init_waitqueue_head(&ff
->poll_wait
);
75 spin_unlock(&fc
->lock
);
80 void fuse_file_free(struct fuse_file
*ff
)
82 fuse_request_free(ff
->reserved_req
);
86 struct fuse_file
*fuse_file_get(struct fuse_file
*ff
)
88 atomic_inc(&ff
->count
);
92 static void fuse_release_async(struct work_struct
*work
)
98 req
= container_of(work
, struct fuse_req
, misc
.release
.work
);
99 path
= req
->misc
.release
.path
;
100 fc
= get_fuse_conn(path
.dentry
->d_inode
);
102 fuse_put_request(fc
, req
);
106 static void fuse_release_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
108 if (fc
->destroy_req
) {
110 * If this is a fuseblk mount, then it's possible that
111 * releasing the path will result in releasing the
112 * super block and sending the DESTROY request. If
113 * the server is single threaded, this would hang.
114 * For this reason do the path_put() in a separate
117 atomic_inc(&req
->count
);
118 INIT_WORK(&req
->misc
.release
.work
, fuse_release_async
);
119 schedule_work(&req
->misc
.release
.work
);
121 path_put(&req
->misc
.release
.path
);
125 static void fuse_file_put(struct fuse_file
*ff
, bool sync
)
127 if (atomic_dec_and_test(&ff
->count
)) {
128 struct fuse_req
*req
= ff
->reserved_req
;
132 fuse_request_send(ff
->fc
, req
);
133 path_put(&req
->misc
.release
.path
);
134 fuse_put_request(ff
->fc
, req
);
136 req
->end
= fuse_release_end
;
138 fuse_request_send_background(ff
->fc
, req
);
144 int fuse_do_open(struct fuse_conn
*fc
, u64 nodeid
, struct file
*file
,
147 struct fuse_open_out outarg
;
148 struct fuse_file
*ff
;
150 int opcode
= isdir
? FUSE_OPENDIR
: FUSE_OPEN
;
152 ff
= fuse_file_alloc(fc
);
156 err
= fuse_send_open(fc
, nodeid
, file
, opcode
, &outarg
);
163 outarg
.open_flags
&= ~FOPEN_DIRECT_IO
;
167 ff
->open_flags
= outarg
.open_flags
;
168 file
->private_data
= fuse_file_get(ff
);
172 EXPORT_SYMBOL_GPL(fuse_do_open
);
174 void fuse_finish_open(struct inode
*inode
, struct file
*file
)
176 struct fuse_file
*ff
= file
->private_data
;
177 struct fuse_conn
*fc
= get_fuse_conn(inode
);
179 if (ff
->open_flags
& FOPEN_DIRECT_IO
)
180 file
->f_op
= &fuse_direct_io_file_operations
;
181 if (!(ff
->open_flags
& FOPEN_KEEP_CACHE
))
182 invalidate_inode_pages2(inode
->i_mapping
);
183 if (ff
->open_flags
& FOPEN_NONSEEKABLE
)
184 nonseekable_open(inode
, file
);
185 if (fc
->atomic_o_trunc
&& (file
->f_flags
& O_TRUNC
)) {
186 struct fuse_inode
*fi
= get_fuse_inode(inode
);
188 spin_lock(&fc
->lock
);
189 fi
->attr_version
= ++fc
->attr_version
;
190 i_size_write(inode
, 0);
191 spin_unlock(&fc
->lock
);
192 fuse_invalidate_attr(inode
);
196 int fuse_open_common(struct inode
*inode
, struct file
*file
, bool isdir
)
198 struct fuse_conn
*fc
= get_fuse_conn(inode
);
201 err
= generic_file_open(inode
, file
);
205 err
= fuse_do_open(fc
, get_node_id(inode
), file
, isdir
);
209 fuse_finish_open(inode
, file
);
214 static void fuse_prepare_release(struct fuse_file
*ff
, int flags
, int opcode
)
216 struct fuse_conn
*fc
= ff
->fc
;
217 struct fuse_req
*req
= ff
->reserved_req
;
218 struct fuse_release_in
*inarg
= &req
->misc
.release
.in
;
220 spin_lock(&fc
->lock
);
221 list_del(&ff
->write_entry
);
222 if (!RB_EMPTY_NODE(&ff
->polled_node
))
223 rb_erase(&ff
->polled_node
, &fc
->polled_files
);
224 spin_unlock(&fc
->lock
);
226 wake_up_interruptible_all(&ff
->poll_wait
);
229 inarg
->flags
= flags
;
230 req
->in
.h
.opcode
= opcode
;
231 req
->in
.h
.nodeid
= ff
->nodeid
;
233 req
->in
.args
[0].size
= sizeof(struct fuse_release_in
);
234 req
->in
.args
[0].value
= inarg
;
237 void fuse_release_common(struct file
*file
, int opcode
)
239 struct fuse_file
*ff
;
240 struct fuse_req
*req
;
242 ff
= file
->private_data
;
246 req
= ff
->reserved_req
;
247 fuse_prepare_release(ff
, file
->f_flags
, opcode
);
250 struct fuse_release_in
*inarg
= &req
->misc
.release
.in
;
251 inarg
->release_flags
|= FUSE_RELEASE_FLOCK_UNLOCK
;
252 inarg
->lock_owner
= fuse_lock_owner_id(ff
->fc
,
255 /* Hold vfsmount and dentry until release is finished */
256 path_get(&file
->f_path
);
257 req
->misc
.release
.path
= file
->f_path
;
260 * Normally this will send the RELEASE request, however if
261 * some asynchronous READ or WRITE requests are outstanding,
262 * the sending will be delayed.
264 * Make the release synchronous if this is a fuseblk mount,
265 * synchronous RELEASE is allowed (and desirable) in this case
266 * because the server can be trusted not to screw up.
268 fuse_file_put(ff
, ff
->fc
->destroy_req
!= NULL
);
271 static int fuse_open(struct inode
*inode
, struct file
*file
)
273 return fuse_open_common(inode
, file
, false);
276 static int fuse_release(struct inode
*inode
, struct file
*file
)
278 fuse_release_common(file
, FUSE_RELEASE
);
280 /* return value is ignored by VFS */
284 void fuse_sync_release(struct fuse_file
*ff
, int flags
)
286 WARN_ON(atomic_read(&ff
->count
) > 1);
287 fuse_prepare_release(ff
, flags
, FUSE_RELEASE
);
288 ff
->reserved_req
->force
= 1;
289 ff
->reserved_req
->background
= 0;
290 fuse_request_send(ff
->fc
, ff
->reserved_req
);
291 fuse_put_request(ff
->fc
, ff
->reserved_req
);
294 EXPORT_SYMBOL_GPL(fuse_sync_release
);
297 * Scramble the ID space with XTEA, so that the value of the files_struct
298 * pointer is not exposed to userspace.
300 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
)
302 u32
*k
= fc
->scramble_key
;
303 u64 v
= (unsigned long) id
;
309 for (i
= 0; i
< 32; i
++) {
310 v0
+= ((v1
<< 4 ^ v1
>> 5) + v1
) ^ (sum
+ k
[sum
& 3]);
312 v1
+= ((v0
<< 4 ^ v0
>> 5) + v0
) ^ (sum
+ k
[sum
>>11 & 3]);
315 return (u64
) v0
+ ((u64
) v1
<< 32);
319 * Check if page is under writeback
321 * This is currently done by walking the list of writepage requests
322 * for the inode, which can be pretty inefficient.
324 static bool fuse_page_is_writeback(struct inode
*inode
, pgoff_t index
)
326 struct fuse_conn
*fc
= get_fuse_conn(inode
);
327 struct fuse_inode
*fi
= get_fuse_inode(inode
);
328 struct fuse_req
*req
;
331 spin_lock(&fc
->lock
);
332 list_for_each_entry(req
, &fi
->writepages
, writepages_entry
) {
335 BUG_ON(req
->inode
!= inode
);
336 curr_index
= req
->misc
.write
.in
.offset
>> PAGE_CACHE_SHIFT
;
337 if (curr_index
== index
) {
342 spin_unlock(&fc
->lock
);
348 * Wait for page writeback to be completed.
350 * Since fuse doesn't rely on the VM writeback tracking, this has to
351 * use some other means.
353 static int fuse_wait_on_page_writeback(struct inode
*inode
, pgoff_t index
)
355 struct fuse_inode
*fi
= get_fuse_inode(inode
);
357 wait_event(fi
->page_waitq
, !fuse_page_is_writeback(inode
, index
));
361 static int fuse_flush(struct file
*file
, fl_owner_t id
)
363 struct inode
*inode
= file_inode(file
);
364 struct fuse_conn
*fc
= get_fuse_conn(inode
);
365 struct fuse_file
*ff
= file
->private_data
;
366 struct fuse_req
*req
;
367 struct fuse_flush_in inarg
;
370 if (is_bad_inode(inode
))
376 req
= fuse_get_req_nofail_nopages(fc
, file
);
377 memset(&inarg
, 0, sizeof(inarg
));
379 inarg
.lock_owner
= fuse_lock_owner_id(fc
, id
);
380 req
->in
.h
.opcode
= FUSE_FLUSH
;
381 req
->in
.h
.nodeid
= get_node_id(inode
);
383 req
->in
.args
[0].size
= sizeof(inarg
);
384 req
->in
.args
[0].value
= &inarg
;
386 fuse_request_send(fc
, req
);
387 err
= req
->out
.h
.error
;
388 fuse_put_request(fc
, req
);
389 if (err
== -ENOSYS
) {
397 * Wait for all pending writepages on the inode to finish.
399 * This is currently done by blocking further writes with FUSE_NOWRITE
400 * and waiting for all sent writes to complete.
402 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
403 * could conflict with truncation.
405 static void fuse_sync_writes(struct inode
*inode
)
407 fuse_set_nowrite(inode
);
408 fuse_release_nowrite(inode
);
411 int fuse_fsync_common(struct file
*file
, loff_t start
, loff_t end
,
412 int datasync
, int isdir
)
414 struct inode
*inode
= file
->f_mapping
->host
;
415 struct fuse_conn
*fc
= get_fuse_conn(inode
);
416 struct fuse_file
*ff
= file
->private_data
;
417 struct fuse_req
*req
;
418 struct fuse_fsync_in inarg
;
421 if (is_bad_inode(inode
))
424 err
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
428 if ((!isdir
&& fc
->no_fsync
) || (isdir
&& fc
->no_fsyncdir
))
431 mutex_lock(&inode
->i_mutex
);
434 * Start writeback against all dirty pages of the inode, then
435 * wait for all outstanding writes, before sending the FSYNC
438 err
= write_inode_now(inode
, 0);
442 fuse_sync_writes(inode
);
444 req
= fuse_get_req_nopages(fc
);
450 memset(&inarg
, 0, sizeof(inarg
));
452 inarg
.fsync_flags
= datasync
? 1 : 0;
453 req
->in
.h
.opcode
= isdir
? FUSE_FSYNCDIR
: FUSE_FSYNC
;
454 req
->in
.h
.nodeid
= get_node_id(inode
);
456 req
->in
.args
[0].size
= sizeof(inarg
);
457 req
->in
.args
[0].value
= &inarg
;
458 fuse_request_send(fc
, req
);
459 err
= req
->out
.h
.error
;
460 fuse_put_request(fc
, req
);
461 if (err
== -ENOSYS
) {
469 mutex_unlock(&inode
->i_mutex
);
473 static int fuse_fsync(struct file
*file
, loff_t start
, loff_t end
,
476 return fuse_fsync_common(file
, start
, end
, datasync
, 0);
479 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
, loff_t pos
,
480 size_t count
, int opcode
)
482 struct fuse_read_in
*inarg
= &req
->misc
.read
.in
;
483 struct fuse_file
*ff
= file
->private_data
;
488 inarg
->flags
= file
->f_flags
;
489 req
->in
.h
.opcode
= opcode
;
490 req
->in
.h
.nodeid
= ff
->nodeid
;
492 req
->in
.args
[0].size
= sizeof(struct fuse_read_in
);
493 req
->in
.args
[0].value
= inarg
;
495 req
->out
.numargs
= 1;
496 req
->out
.args
[0].size
= count
;
499 static void fuse_release_user_pages(struct fuse_req
*req
, int write
)
503 for (i
= 0; i
< req
->num_pages
; i
++) {
504 struct page
*page
= req
->pages
[i
];
506 set_page_dirty_lock(page
);
512 * In case of short read, the caller sets 'pos' to the position of
513 * actual end of fuse request in IO request. Otherwise, if bytes_requested
514 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
517 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
518 * both submitted asynchronously. The first of them was ACKed by userspace as
519 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
520 * second request was ACKed as short, e.g. only 1K was read, resulting in
523 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
524 * will be equal to the length of the longest contiguous fragment of
525 * transferred data starting from the beginning of IO request.
527 static void fuse_aio_complete(struct fuse_io_priv
*io
, int err
, ssize_t pos
)
531 spin_lock(&io
->lock
);
533 io
->err
= io
->err
? : err
;
534 else if (pos
>= 0 && (io
->bytes
< 0 || pos
< io
->bytes
))
538 spin_unlock(&io
->lock
);
545 else if (io
->bytes
>= 0 && io
->write
)
548 res
= io
->bytes
< 0 ? io
->size
: io
->bytes
;
550 if (!is_sync_kiocb(io
->iocb
)) {
551 struct path
*path
= &io
->iocb
->ki_filp
->f_path
;
552 struct inode
*inode
= path
->dentry
->d_inode
;
553 struct fuse_conn
*fc
= get_fuse_conn(inode
);
554 struct fuse_inode
*fi
= get_fuse_inode(inode
);
556 spin_lock(&fc
->lock
);
557 fi
->attr_version
= ++fc
->attr_version
;
558 spin_unlock(&fc
->lock
);
562 aio_complete(io
->iocb
, res
, 0);
567 static void fuse_aio_complete_req(struct fuse_conn
*fc
, struct fuse_req
*req
)
569 struct fuse_io_priv
*io
= req
->io
;
572 fuse_release_user_pages(req
, !io
->write
);
575 if (req
->misc
.write
.in
.size
!= req
->misc
.write
.out
.size
)
576 pos
= req
->misc
.write
.in
.offset
- io
->offset
+
577 req
->misc
.write
.out
.size
;
579 if (req
->misc
.read
.in
.size
!= req
->out
.args
[0].size
)
580 pos
= req
->misc
.read
.in
.offset
- io
->offset
+
581 req
->out
.args
[0].size
;
584 fuse_aio_complete(io
, req
->out
.h
.error
, pos
);
587 static size_t fuse_async_req_send(struct fuse_conn
*fc
, struct fuse_req
*req
,
588 size_t num_bytes
, struct fuse_io_priv
*io
)
590 spin_lock(&io
->lock
);
591 io
->size
+= num_bytes
;
593 spin_unlock(&io
->lock
);
596 req
->end
= fuse_aio_complete_req
;
598 __fuse_get_request(req
);
599 fuse_request_send_background(fc
, req
);
604 static size_t fuse_send_read(struct fuse_req
*req
, struct fuse_io_priv
*io
,
605 loff_t pos
, size_t count
, fl_owner_t owner
)
607 struct file
*file
= io
->file
;
608 struct fuse_file
*ff
= file
->private_data
;
609 struct fuse_conn
*fc
= ff
->fc
;
611 fuse_read_fill(req
, file
, pos
, count
, FUSE_READ
);
613 struct fuse_read_in
*inarg
= &req
->misc
.read
.in
;
615 inarg
->read_flags
|= FUSE_READ_LOCKOWNER
;
616 inarg
->lock_owner
= fuse_lock_owner_id(fc
, owner
);
620 return fuse_async_req_send(fc
, req
, count
, io
);
622 fuse_request_send(fc
, req
);
623 return req
->out
.args
[0].size
;
626 static void fuse_read_update_size(struct inode
*inode
, loff_t size
,
629 struct fuse_conn
*fc
= get_fuse_conn(inode
);
630 struct fuse_inode
*fi
= get_fuse_inode(inode
);
632 spin_lock(&fc
->lock
);
633 if (attr_ver
== fi
->attr_version
&& size
< inode
->i_size
&&
634 !test_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
)) {
635 fi
->attr_version
= ++fc
->attr_version
;
636 i_size_write(inode
, size
);
638 spin_unlock(&fc
->lock
);
641 static int fuse_readpage(struct file
*file
, struct page
*page
)
643 struct fuse_io_priv io
= { .async
= 0, .file
= file
};
644 struct inode
*inode
= page
->mapping
->host
;
645 struct fuse_conn
*fc
= get_fuse_conn(inode
);
646 struct fuse_req
*req
;
648 loff_t pos
= page_offset(page
);
649 size_t count
= PAGE_CACHE_SIZE
;
654 if (is_bad_inode(inode
))
658 * Page writeback can extend beyond the lifetime of the
659 * page-cache page, so make sure we read a properly synced
662 fuse_wait_on_page_writeback(inode
, page
->index
);
664 req
= fuse_get_req(fc
, 1);
669 attr_ver
= fuse_get_attr_version(fc
);
671 req
->out
.page_zeroing
= 1;
672 req
->out
.argpages
= 1;
674 req
->pages
[0] = page
;
675 req
->page_descs
[0].length
= count
;
676 num_read
= fuse_send_read(req
, &io
, pos
, count
, NULL
);
677 err
= req
->out
.h
.error
;
678 fuse_put_request(fc
, req
);
682 * Short read means EOF. If file size is larger, truncate it
684 if (num_read
< count
)
685 fuse_read_update_size(inode
, pos
+ num_read
, attr_ver
);
687 SetPageUptodate(page
);
690 fuse_invalidate_attr(inode
); /* atime changed */
696 static void fuse_readpages_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
699 size_t count
= req
->misc
.read
.in
.size
;
700 size_t num_read
= req
->out
.args
[0].size
;
701 struct address_space
*mapping
= NULL
;
703 for (i
= 0; mapping
== NULL
&& i
< req
->num_pages
; i
++)
704 mapping
= req
->pages
[i
]->mapping
;
707 struct inode
*inode
= mapping
->host
;
710 * Short read means EOF. If file size is larger, truncate it
712 if (!req
->out
.h
.error
&& num_read
< count
) {
715 pos
= page_offset(req
->pages
[0]) + num_read
;
716 fuse_read_update_size(inode
, pos
,
717 req
->misc
.read
.attr_ver
);
719 fuse_invalidate_attr(inode
); /* atime changed */
722 for (i
= 0; i
< req
->num_pages
; i
++) {
723 struct page
*page
= req
->pages
[i
];
724 if (!req
->out
.h
.error
)
725 SetPageUptodate(page
);
729 page_cache_release(page
);
732 fuse_file_put(req
->ff
, false);
735 static void fuse_send_readpages(struct fuse_req
*req
, struct file
*file
)
737 struct fuse_file
*ff
= file
->private_data
;
738 struct fuse_conn
*fc
= ff
->fc
;
739 loff_t pos
= page_offset(req
->pages
[0]);
740 size_t count
= req
->num_pages
<< PAGE_CACHE_SHIFT
;
742 req
->out
.argpages
= 1;
743 req
->out
.page_zeroing
= 1;
744 req
->out
.page_replace
= 1;
745 fuse_read_fill(req
, file
, pos
, count
, FUSE_READ
);
746 req
->misc
.read
.attr_ver
= fuse_get_attr_version(fc
);
747 if (fc
->async_read
) {
748 req
->ff
= fuse_file_get(ff
);
749 req
->end
= fuse_readpages_end
;
750 fuse_request_send_background(fc
, req
);
752 fuse_request_send(fc
, req
);
753 fuse_readpages_end(fc
, req
);
754 fuse_put_request(fc
, req
);
758 struct fuse_fill_data
{
759 struct fuse_req
*req
;
765 static int fuse_readpages_fill(void *_data
, struct page
*page
)
767 struct fuse_fill_data
*data
= _data
;
768 struct fuse_req
*req
= data
->req
;
769 struct inode
*inode
= data
->inode
;
770 struct fuse_conn
*fc
= get_fuse_conn(inode
);
772 fuse_wait_on_page_writeback(inode
, page
->index
);
774 if (req
->num_pages
&&
775 (req
->num_pages
== FUSE_MAX_PAGES_PER_REQ
||
776 (req
->num_pages
+ 1) * PAGE_CACHE_SIZE
> fc
->max_read
||
777 req
->pages
[req
->num_pages
- 1]->index
+ 1 != page
->index
)) {
778 int nr_alloc
= min_t(unsigned, data
->nr_pages
,
779 FUSE_MAX_PAGES_PER_REQ
);
780 fuse_send_readpages(req
, data
->file
);
782 req
= fuse_get_req_for_background(fc
, nr_alloc
);
784 req
= fuse_get_req(fc
, nr_alloc
);
793 if (WARN_ON(req
->num_pages
>= req
->max_pages
)) {
794 fuse_put_request(fc
, req
);
798 page_cache_get(page
);
799 req
->pages
[req
->num_pages
] = page
;
800 req
->page_descs
[req
->num_pages
].length
= PAGE_SIZE
;
806 static int fuse_readpages(struct file
*file
, struct address_space
*mapping
,
807 struct list_head
*pages
, unsigned nr_pages
)
809 struct inode
*inode
= mapping
->host
;
810 struct fuse_conn
*fc
= get_fuse_conn(inode
);
811 struct fuse_fill_data data
;
813 int nr_alloc
= min_t(unsigned, nr_pages
, FUSE_MAX_PAGES_PER_REQ
);
816 if (is_bad_inode(inode
))
822 data
.req
= fuse_get_req_for_background(fc
, nr_alloc
);
824 data
.req
= fuse_get_req(fc
, nr_alloc
);
825 data
.nr_pages
= nr_pages
;
826 err
= PTR_ERR(data
.req
);
827 if (IS_ERR(data
.req
))
830 err
= read_cache_pages(mapping
, pages
, fuse_readpages_fill
, &data
);
832 if (data
.req
->num_pages
)
833 fuse_send_readpages(data
.req
, file
);
835 fuse_put_request(fc
, data
.req
);
841 static ssize_t
fuse_file_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
842 unsigned long nr_segs
, loff_t pos
)
844 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
845 struct fuse_conn
*fc
= get_fuse_conn(inode
);
848 * In auto invalidate mode, always update attributes on read.
849 * Otherwise, only update if we attempt to read past EOF (to ensure
850 * i_size is up to date).
852 if (fc
->auto_inval_data
||
853 (pos
+ iov_length(iov
, nr_segs
) > i_size_read(inode
))) {
855 err
= fuse_update_attributes(inode
, NULL
, iocb
->ki_filp
, NULL
);
860 return generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
863 static void fuse_write_fill(struct fuse_req
*req
, struct fuse_file
*ff
,
864 loff_t pos
, size_t count
)
866 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
867 struct fuse_write_out
*outarg
= &req
->misc
.write
.out
;
872 req
->in
.h
.opcode
= FUSE_WRITE
;
873 req
->in
.h
.nodeid
= ff
->nodeid
;
875 if (ff
->fc
->minor
< 9)
876 req
->in
.args
[0].size
= FUSE_COMPAT_WRITE_IN_SIZE
;
878 req
->in
.args
[0].size
= sizeof(struct fuse_write_in
);
879 req
->in
.args
[0].value
= inarg
;
880 req
->in
.args
[1].size
= count
;
881 req
->out
.numargs
= 1;
882 req
->out
.args
[0].size
= sizeof(struct fuse_write_out
);
883 req
->out
.args
[0].value
= outarg
;
886 static size_t fuse_send_write(struct fuse_req
*req
, struct fuse_io_priv
*io
,
887 loff_t pos
, size_t count
, fl_owner_t owner
)
889 struct file
*file
= io
->file
;
890 struct fuse_file
*ff
= file
->private_data
;
891 struct fuse_conn
*fc
= ff
->fc
;
892 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
894 fuse_write_fill(req
, ff
, pos
, count
);
895 inarg
->flags
= file
->f_flags
;
897 inarg
->write_flags
|= FUSE_WRITE_LOCKOWNER
;
898 inarg
->lock_owner
= fuse_lock_owner_id(fc
, owner
);
902 return fuse_async_req_send(fc
, req
, count
, io
);
904 fuse_request_send(fc
, req
);
905 return req
->misc
.write
.out
.size
;
908 void fuse_write_update_size(struct inode
*inode
, loff_t pos
)
910 struct fuse_conn
*fc
= get_fuse_conn(inode
);
911 struct fuse_inode
*fi
= get_fuse_inode(inode
);
913 spin_lock(&fc
->lock
);
914 fi
->attr_version
= ++fc
->attr_version
;
915 if (pos
> inode
->i_size
)
916 i_size_write(inode
, pos
);
917 spin_unlock(&fc
->lock
);
920 static size_t fuse_send_write_pages(struct fuse_req
*req
, struct file
*file
,
921 struct inode
*inode
, loff_t pos
,
927 struct fuse_io_priv io
= { .async
= 0, .file
= file
};
929 for (i
= 0; i
< req
->num_pages
; i
++)
930 fuse_wait_on_page_writeback(inode
, req
->pages
[i
]->index
);
932 res
= fuse_send_write(req
, &io
, pos
, count
, NULL
);
934 offset
= req
->page_descs
[0].offset
;
936 for (i
= 0; i
< req
->num_pages
; i
++) {
937 struct page
*page
= req
->pages
[i
];
939 if (!req
->out
.h
.error
&& !offset
&& count
>= PAGE_CACHE_SIZE
)
940 SetPageUptodate(page
);
942 if (count
> PAGE_CACHE_SIZE
- offset
)
943 count
-= PAGE_CACHE_SIZE
- offset
;
949 page_cache_release(page
);
955 static ssize_t
fuse_fill_write_pages(struct fuse_req
*req
,
956 struct address_space
*mapping
,
957 struct iov_iter
*ii
, loff_t pos
)
959 struct fuse_conn
*fc
= get_fuse_conn(mapping
->host
);
960 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
964 req
->in
.argpages
= 1;
965 req
->page_descs
[0].offset
= offset
;
970 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
971 size_t bytes
= min_t(size_t, PAGE_CACHE_SIZE
- offset
,
974 bytes
= min_t(size_t, bytes
, fc
->max_write
- count
);
978 if (iov_iter_fault_in_readable(ii
, bytes
))
982 page
= grab_cache_page_write_begin(mapping
, index
, 0);
986 if (mapping_writably_mapped(mapping
))
987 flush_dcache_page(page
);
990 tmp
= iov_iter_copy_from_user_atomic(page
, ii
, offset
, bytes
);
992 flush_dcache_page(page
);
994 mark_page_accessed(page
);
996 iov_iter_advance(ii
, tmp
);
999 page_cache_release(page
);
1000 bytes
= min(bytes
, iov_iter_single_seg_count(ii
));
1005 req
->pages
[req
->num_pages
] = page
;
1006 req
->page_descs
[req
->num_pages
].length
= tmp
;
1012 if (offset
== PAGE_CACHE_SIZE
)
1015 if (!fc
->big_writes
)
1017 } while (iov_iter_count(ii
) && count
< fc
->max_write
&&
1018 req
->num_pages
< req
->max_pages
&& offset
== 0);
1020 return count
> 0 ? count
: err
;
1023 static inline unsigned fuse_wr_pages(loff_t pos
, size_t len
)
1025 return min_t(unsigned,
1026 ((pos
+ len
- 1) >> PAGE_CACHE_SHIFT
) -
1027 (pos
>> PAGE_CACHE_SHIFT
) + 1,
1028 FUSE_MAX_PAGES_PER_REQ
);
1031 static ssize_t
fuse_perform_write(struct file
*file
,
1032 struct address_space
*mapping
,
1033 struct iov_iter
*ii
, loff_t pos
)
1035 struct inode
*inode
= mapping
->host
;
1036 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1037 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1041 if (is_bad_inode(inode
))
1044 if (inode
->i_size
< pos
+ iov_iter_count(ii
))
1045 set_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
1048 struct fuse_req
*req
;
1050 unsigned nr_pages
= fuse_wr_pages(pos
, iov_iter_count(ii
));
1052 req
= fuse_get_req(fc
, nr_pages
);
1058 count
= fuse_fill_write_pages(req
, mapping
, ii
, pos
);
1064 num_written
= fuse_send_write_pages(req
, file
, inode
,
1066 err
= req
->out
.h
.error
;
1071 /* break out of the loop on short write */
1072 if (num_written
!= count
)
1076 fuse_put_request(fc
, req
);
1077 } while (!err
&& iov_iter_count(ii
));
1080 fuse_write_update_size(inode
, pos
);
1082 clear_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
1083 fuse_invalidate_attr(inode
);
1085 return res
> 0 ? res
: err
;
1088 static ssize_t
fuse_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1089 unsigned long nr_segs
, loff_t pos
)
1091 struct file
*file
= iocb
->ki_filp
;
1092 struct address_space
*mapping
= file
->f_mapping
;
1095 ssize_t written
= 0;
1096 ssize_t written_buffered
= 0;
1097 struct inode
*inode
= mapping
->host
;
1102 WARN_ON(iocb
->ki_pos
!= pos
);
1105 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
1110 mutex_lock(&inode
->i_mutex
);
1112 /* We can write back this queue in page reclaim */
1113 current
->backing_dev_info
= mapping
->backing_dev_info
;
1115 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
1122 err
= file_remove_suid(file
);
1126 err
= file_update_time(file
);
1130 if (file
->f_flags
& O_DIRECT
) {
1131 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
,
1134 if (written
< 0 || written
== count
)
1140 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
1141 written_buffered
= fuse_perform_write(file
, mapping
, &i
, pos
);
1142 if (written_buffered
< 0) {
1143 err
= written_buffered
;
1146 endbyte
= pos
+ written_buffered
- 1;
1148 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
,
1153 invalidate_mapping_pages(file
->f_mapping
,
1154 pos
>> PAGE_CACHE_SHIFT
,
1155 endbyte
>> PAGE_CACHE_SHIFT
);
1157 written
+= written_buffered
;
1158 iocb
->ki_pos
= pos
+ written_buffered
;
1160 iov_iter_init(&i
, iov
, nr_segs
, count
, 0);
1161 written
= fuse_perform_write(file
, mapping
, &i
, pos
);
1163 iocb
->ki_pos
= pos
+ written
;
1166 current
->backing_dev_info
= NULL
;
1167 mutex_unlock(&inode
->i_mutex
);
1169 return written
? written
: err
;
1172 static inline void fuse_page_descs_length_init(struct fuse_req
*req
,
1173 unsigned index
, unsigned nr_pages
)
1177 for (i
= index
; i
< index
+ nr_pages
; i
++)
1178 req
->page_descs
[i
].length
= PAGE_SIZE
-
1179 req
->page_descs
[i
].offset
;
1182 static inline unsigned long fuse_get_user_addr(const struct iov_iter
*ii
)
1184 return (unsigned long)ii
->iov
->iov_base
+ ii
->iov_offset
;
1187 static inline size_t fuse_get_frag_size(const struct iov_iter
*ii
,
1190 return min(iov_iter_single_seg_count(ii
), max_size
);
1193 static int fuse_get_user_pages(struct fuse_req
*req
, struct iov_iter
*ii
,
1194 size_t *nbytesp
, int write
)
1196 size_t nbytes
= 0; /* # bytes already packed in req */
1198 /* Special case for kernel I/O: can copy directly into the buffer */
1199 if (segment_eq(get_fs(), KERNEL_DS
)) {
1200 unsigned long user_addr
= fuse_get_user_addr(ii
);
1201 size_t frag_size
= fuse_get_frag_size(ii
, *nbytesp
);
1204 req
->in
.args
[1].value
= (void *) user_addr
;
1206 req
->out
.args
[0].value
= (void *) user_addr
;
1208 iov_iter_advance(ii
, frag_size
);
1209 *nbytesp
= frag_size
;
1213 while (nbytes
< *nbytesp
&& req
->num_pages
< req
->max_pages
) {
1215 unsigned long user_addr
= fuse_get_user_addr(ii
);
1216 unsigned offset
= user_addr
& ~PAGE_MASK
;
1217 size_t frag_size
= fuse_get_frag_size(ii
, *nbytesp
- nbytes
);
1220 unsigned n
= req
->max_pages
- req
->num_pages
;
1221 frag_size
= min_t(size_t, frag_size
, n
<< PAGE_SHIFT
);
1223 npages
= (frag_size
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1224 npages
= clamp(npages
, 1U, n
);
1226 ret
= get_user_pages_fast(user_addr
, npages
, !write
,
1227 &req
->pages
[req
->num_pages
]);
1232 frag_size
= min_t(size_t, frag_size
,
1233 (npages
<< PAGE_SHIFT
) - offset
);
1234 iov_iter_advance(ii
, frag_size
);
1236 req
->page_descs
[req
->num_pages
].offset
= offset
;
1237 fuse_page_descs_length_init(req
, req
->num_pages
, npages
);
1239 req
->num_pages
+= npages
;
1240 req
->page_descs
[req
->num_pages
- 1].length
-=
1241 (npages
<< PAGE_SHIFT
) - offset
- frag_size
;
1243 nbytes
+= frag_size
;
1247 req
->in
.argpages
= 1;
1249 req
->out
.argpages
= 1;
1256 static inline int fuse_iter_npages(const struct iov_iter
*ii_p
)
1258 struct iov_iter ii
= *ii_p
;
1261 while (iov_iter_count(&ii
) && npages
< FUSE_MAX_PAGES_PER_REQ
) {
1262 unsigned long user_addr
= fuse_get_user_addr(&ii
);
1263 unsigned offset
= user_addr
& ~PAGE_MASK
;
1264 size_t frag_size
= iov_iter_single_seg_count(&ii
);
1266 npages
+= (frag_size
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1267 iov_iter_advance(&ii
, frag_size
);
1270 return min(npages
, FUSE_MAX_PAGES_PER_REQ
);
1273 ssize_t
fuse_direct_io(struct fuse_io_priv
*io
, const struct iovec
*iov
,
1274 unsigned long nr_segs
, size_t count
, loff_t
*ppos
,
1277 struct file
*file
= io
->file
;
1278 struct fuse_file
*ff
= file
->private_data
;
1279 struct fuse_conn
*fc
= ff
->fc
;
1280 size_t nmax
= write
? fc
->max_write
: fc
->max_read
;
1283 struct fuse_req
*req
;
1286 iov_iter_init(&ii
, iov
, nr_segs
, count
, 0);
1289 req
= fuse_get_req_for_background(fc
, fuse_iter_npages(&ii
));
1291 req
= fuse_get_req(fc
, fuse_iter_npages(&ii
));
1293 return PTR_ERR(req
);
1297 fl_owner_t owner
= current
->files
;
1298 size_t nbytes
= min(count
, nmax
);
1299 int err
= fuse_get_user_pages(req
, &ii
, &nbytes
, write
);
1306 nres
= fuse_send_write(req
, io
, pos
, nbytes
, owner
);
1308 nres
= fuse_send_read(req
, io
, pos
, nbytes
, owner
);
1311 fuse_release_user_pages(req
, !write
);
1312 if (req
->out
.h
.error
) {
1314 res
= req
->out
.h
.error
;
1316 } else if (nres
> nbytes
) {
1326 fuse_put_request(fc
, req
);
1328 req
= fuse_get_req_for_background(fc
,
1329 fuse_iter_npages(&ii
));
1331 req
= fuse_get_req(fc
, fuse_iter_npages(&ii
));
1337 fuse_put_request(fc
, req
);
1343 EXPORT_SYMBOL_GPL(fuse_direct_io
);
1345 static ssize_t
__fuse_direct_read(struct fuse_io_priv
*io
,
1346 const struct iovec
*iov
,
1347 unsigned long nr_segs
, loff_t
*ppos
,
1351 struct file
*file
= io
->file
;
1352 struct inode
*inode
= file_inode(file
);
1354 if (is_bad_inode(inode
))
1357 res
= fuse_direct_io(io
, iov
, nr_segs
, count
, ppos
, 0);
1359 fuse_invalidate_attr(inode
);
1364 static ssize_t
fuse_direct_read(struct file
*file
, char __user
*buf
,
1365 size_t count
, loff_t
*ppos
)
1367 struct fuse_io_priv io
= { .async
= 0, .file
= file
};
1368 struct iovec iov
= { .iov_base
= buf
, .iov_len
= count
};
1369 return __fuse_direct_read(&io
, &iov
, 1, ppos
, count
);
1372 static ssize_t
__fuse_direct_write(struct fuse_io_priv
*io
,
1373 const struct iovec
*iov
,
1374 unsigned long nr_segs
, loff_t
*ppos
)
1376 struct file
*file
= io
->file
;
1377 struct inode
*inode
= file_inode(file
);
1378 size_t count
= iov_length(iov
, nr_segs
);
1381 res
= generic_write_checks(file
, ppos
, &count
, 0);
1383 res
= fuse_direct_io(io
, iov
, nr_segs
, count
, ppos
, 1);
1385 fuse_invalidate_attr(inode
);
1390 static ssize_t
fuse_direct_write(struct file
*file
, const char __user
*buf
,
1391 size_t count
, loff_t
*ppos
)
1393 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= count
};
1394 struct inode
*inode
= file_inode(file
);
1396 struct fuse_io_priv io
= { .async
= 0, .file
= file
};
1398 if (is_bad_inode(inode
))
1401 /* Don't allow parallel writes to the same file */
1402 mutex_lock(&inode
->i_mutex
);
1403 res
= __fuse_direct_write(&io
, &iov
, 1, ppos
);
1405 fuse_write_update_size(inode
, *ppos
);
1406 mutex_unlock(&inode
->i_mutex
);
1411 static void fuse_writepage_free(struct fuse_conn
*fc
, struct fuse_req
*req
)
1413 __free_page(req
->pages
[0]);
1414 fuse_file_put(req
->ff
, false);
1417 static void fuse_writepage_finish(struct fuse_conn
*fc
, struct fuse_req
*req
)
1419 struct inode
*inode
= req
->inode
;
1420 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1421 struct backing_dev_info
*bdi
= inode
->i_mapping
->backing_dev_info
;
1423 list_del(&req
->writepages_entry
);
1424 dec_bdi_stat(bdi
, BDI_WRITEBACK
);
1425 dec_zone_page_state(req
->pages
[0], NR_WRITEBACK_TEMP
);
1426 bdi_writeout_inc(bdi
);
1427 wake_up(&fi
->page_waitq
);
1430 /* Called under fc->lock, may release and reacquire it */
1431 static void fuse_send_writepage(struct fuse_conn
*fc
, struct fuse_req
*req
)
1432 __releases(fc
->lock
)
1433 __acquires(fc
->lock
)
1435 struct fuse_inode
*fi
= get_fuse_inode(req
->inode
);
1436 loff_t size
= i_size_read(req
->inode
);
1437 struct fuse_write_in
*inarg
= &req
->misc
.write
.in
;
1442 if (inarg
->offset
+ PAGE_CACHE_SIZE
<= size
) {
1443 inarg
->size
= PAGE_CACHE_SIZE
;
1444 } else if (inarg
->offset
< size
) {
1445 inarg
->size
= size
& (PAGE_CACHE_SIZE
- 1);
1447 /* Got truncated off completely */
1451 req
->in
.args
[1].size
= inarg
->size
;
1453 fuse_request_send_background_locked(fc
, req
);
1457 fuse_writepage_finish(fc
, req
);
1458 spin_unlock(&fc
->lock
);
1459 fuse_writepage_free(fc
, req
);
1460 fuse_put_request(fc
, req
);
1461 spin_lock(&fc
->lock
);
1465 * If fi->writectr is positive (no truncate or fsync going on) send
1466 * all queued writepage requests.
1468 * Called with fc->lock
1470 void fuse_flush_writepages(struct inode
*inode
)
1471 __releases(fc
->lock
)
1472 __acquires(fc
->lock
)
1474 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1475 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1476 struct fuse_req
*req
;
1478 while (fi
->writectr
>= 0 && !list_empty(&fi
->queued_writes
)) {
1479 req
= list_entry(fi
->queued_writes
.next
, struct fuse_req
, list
);
1480 list_del_init(&req
->list
);
1481 fuse_send_writepage(fc
, req
);
1485 static void fuse_writepage_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
1487 struct inode
*inode
= req
->inode
;
1488 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1490 mapping_set_error(inode
->i_mapping
, req
->out
.h
.error
);
1491 spin_lock(&fc
->lock
);
1493 fuse_writepage_finish(fc
, req
);
1494 spin_unlock(&fc
->lock
);
1495 fuse_writepage_free(fc
, req
);
1498 static int fuse_writepage_locked(struct page
*page
)
1500 struct address_space
*mapping
= page
->mapping
;
1501 struct inode
*inode
= mapping
->host
;
1502 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1503 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1504 struct fuse_req
*req
;
1505 struct fuse_file
*ff
;
1506 struct page
*tmp_page
;
1508 set_page_writeback(page
);
1510 req
= fuse_request_alloc_nofs(1);
1514 req
->background
= 1; /* writeback always goes to bg_queue */
1515 tmp_page
= alloc_page(GFP_NOFS
| __GFP_HIGHMEM
);
1519 spin_lock(&fc
->lock
);
1520 BUG_ON(list_empty(&fi
->write_files
));
1521 ff
= list_entry(fi
->write_files
.next
, struct fuse_file
, write_entry
);
1522 req
->ff
= fuse_file_get(ff
);
1523 spin_unlock(&fc
->lock
);
1525 fuse_write_fill(req
, ff
, page_offset(page
), 0);
1527 copy_highpage(tmp_page
, page
);
1528 req
->misc
.write
.in
.write_flags
|= FUSE_WRITE_CACHE
;
1529 req
->in
.argpages
= 1;
1531 req
->pages
[0] = tmp_page
;
1532 req
->page_descs
[0].offset
= 0;
1533 req
->page_descs
[0].length
= PAGE_SIZE
;
1534 req
->end
= fuse_writepage_end
;
1537 inc_bdi_stat(mapping
->backing_dev_info
, BDI_WRITEBACK
);
1538 inc_zone_page_state(tmp_page
, NR_WRITEBACK_TEMP
);
1540 spin_lock(&fc
->lock
);
1541 list_add(&req
->writepages_entry
, &fi
->writepages
);
1542 list_add_tail(&req
->list
, &fi
->queued_writes
);
1543 fuse_flush_writepages(inode
);
1544 spin_unlock(&fc
->lock
);
1546 end_page_writeback(page
);
1551 fuse_request_free(req
);
1553 end_page_writeback(page
);
1557 static int fuse_writepage(struct page
*page
, struct writeback_control
*wbc
)
1561 err
= fuse_writepage_locked(page
);
1567 static int fuse_launder_page(struct page
*page
)
1570 if (clear_page_dirty_for_io(page
)) {
1571 struct inode
*inode
= page
->mapping
->host
;
1572 err
= fuse_writepage_locked(page
);
1574 fuse_wait_on_page_writeback(inode
, page
->index
);
1580 * Write back dirty pages now, because there may not be any suitable
1583 static void fuse_vma_close(struct vm_area_struct
*vma
)
1585 filemap_write_and_wait(vma
->vm_file
->f_mapping
);
1589 * Wait for writeback against this page to complete before allowing it
1590 * to be marked dirty again, and hence written back again, possibly
1591 * before the previous writepage completed.
1593 * Block here, instead of in ->writepage(), so that the userspace fs
1594 * can only block processes actually operating on the filesystem.
1596 * Otherwise unprivileged userspace fs would be able to block
1601 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1603 static int fuse_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1605 struct page
*page
= vmf
->page
;
1607 * Don't use page->mapping as it may become NULL from a
1608 * concurrent truncate.
1610 struct inode
*inode
= vma
->vm_file
->f_mapping
->host
;
1612 fuse_wait_on_page_writeback(inode
, page
->index
);
1616 static const struct vm_operations_struct fuse_file_vm_ops
= {
1617 .close
= fuse_vma_close
,
1618 .fault
= filemap_fault
,
1619 .page_mkwrite
= fuse_page_mkwrite
,
1620 .remap_pages
= generic_file_remap_pages
,
1623 static int fuse_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1625 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
)) {
1626 struct inode
*inode
= file_inode(file
);
1627 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1628 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1629 struct fuse_file
*ff
= file
->private_data
;
1631 * file may be written through mmap, so chain it onto the
1632 * inodes's write_file list
1634 spin_lock(&fc
->lock
);
1635 if (list_empty(&ff
->write_entry
))
1636 list_add(&ff
->write_entry
, &fi
->write_files
);
1637 spin_unlock(&fc
->lock
);
1639 file_accessed(file
);
1640 vma
->vm_ops
= &fuse_file_vm_ops
;
1644 static int fuse_direct_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1646 /* Can't provide the coherency needed for MAP_SHARED */
1647 if (vma
->vm_flags
& VM_MAYSHARE
)
1650 invalidate_inode_pages2(file
->f_mapping
);
1652 return generic_file_mmap(file
, vma
);
1655 static int convert_fuse_file_lock(const struct fuse_file_lock
*ffl
,
1656 struct file_lock
*fl
)
1658 switch (ffl
->type
) {
1664 if (ffl
->start
> OFFSET_MAX
|| ffl
->end
> OFFSET_MAX
||
1665 ffl
->end
< ffl
->start
)
1668 fl
->fl_start
= ffl
->start
;
1669 fl
->fl_end
= ffl
->end
;
1670 fl
->fl_pid
= ffl
->pid
;
1676 fl
->fl_type
= ffl
->type
;
1680 static void fuse_lk_fill(struct fuse_req
*req
, struct file
*file
,
1681 const struct file_lock
*fl
, int opcode
, pid_t pid
,
1684 struct inode
*inode
= file_inode(file
);
1685 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1686 struct fuse_file
*ff
= file
->private_data
;
1687 struct fuse_lk_in
*arg
= &req
->misc
.lk_in
;
1690 arg
->owner
= fuse_lock_owner_id(fc
, fl
->fl_owner
);
1691 arg
->lk
.start
= fl
->fl_start
;
1692 arg
->lk
.end
= fl
->fl_end
;
1693 arg
->lk
.type
= fl
->fl_type
;
1696 arg
->lk_flags
|= FUSE_LK_FLOCK
;
1697 req
->in
.h
.opcode
= opcode
;
1698 req
->in
.h
.nodeid
= get_node_id(inode
);
1699 req
->in
.numargs
= 1;
1700 req
->in
.args
[0].size
= sizeof(*arg
);
1701 req
->in
.args
[0].value
= arg
;
1704 static int fuse_getlk(struct file
*file
, struct file_lock
*fl
)
1706 struct inode
*inode
= file_inode(file
);
1707 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1708 struct fuse_req
*req
;
1709 struct fuse_lk_out outarg
;
1712 req
= fuse_get_req_nopages(fc
);
1714 return PTR_ERR(req
);
1716 fuse_lk_fill(req
, file
, fl
, FUSE_GETLK
, 0, 0);
1717 req
->out
.numargs
= 1;
1718 req
->out
.args
[0].size
= sizeof(outarg
);
1719 req
->out
.args
[0].value
= &outarg
;
1720 fuse_request_send(fc
, req
);
1721 err
= req
->out
.h
.error
;
1722 fuse_put_request(fc
, req
);
1724 err
= convert_fuse_file_lock(&outarg
.lk
, fl
);
1729 static int fuse_setlk(struct file
*file
, struct file_lock
*fl
, int flock
)
1731 struct inode
*inode
= file_inode(file
);
1732 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1733 struct fuse_req
*req
;
1734 int opcode
= (fl
->fl_flags
& FL_SLEEP
) ? FUSE_SETLKW
: FUSE_SETLK
;
1735 pid_t pid
= fl
->fl_type
!= F_UNLCK
? current
->tgid
: 0;
1738 if (fl
->fl_lmops
&& fl
->fl_lmops
->lm_grant
) {
1739 /* NLM needs asynchronous locks, which we don't support yet */
1743 /* Unlock on close is handled by the flush method */
1744 if (fl
->fl_flags
& FL_CLOSE
)
1747 req
= fuse_get_req_nopages(fc
);
1749 return PTR_ERR(req
);
1751 fuse_lk_fill(req
, file
, fl
, opcode
, pid
, flock
);
1752 fuse_request_send(fc
, req
);
1753 err
= req
->out
.h
.error
;
1754 /* locking is restartable */
1757 fuse_put_request(fc
, req
);
1761 static int fuse_file_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1763 struct inode
*inode
= file_inode(file
);
1764 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1767 if (cmd
== F_CANCELLK
) {
1769 } else if (cmd
== F_GETLK
) {
1771 posix_test_lock(file
, fl
);
1774 err
= fuse_getlk(file
, fl
);
1777 err
= posix_lock_file(file
, fl
, NULL
);
1779 err
= fuse_setlk(file
, fl
, 0);
1784 static int fuse_file_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1786 struct inode
*inode
= file_inode(file
);
1787 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1791 err
= flock_lock_file_wait(file
, fl
);
1793 struct fuse_file
*ff
= file
->private_data
;
1795 /* emulate flock with POSIX locks */
1796 fl
->fl_owner
= (fl_owner_t
) file
;
1798 err
= fuse_setlk(file
, fl
, 1);
1804 static sector_t
fuse_bmap(struct address_space
*mapping
, sector_t block
)
1806 struct inode
*inode
= mapping
->host
;
1807 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1808 struct fuse_req
*req
;
1809 struct fuse_bmap_in inarg
;
1810 struct fuse_bmap_out outarg
;
1813 if (!inode
->i_sb
->s_bdev
|| fc
->no_bmap
)
1816 req
= fuse_get_req_nopages(fc
);
1820 memset(&inarg
, 0, sizeof(inarg
));
1821 inarg
.block
= block
;
1822 inarg
.blocksize
= inode
->i_sb
->s_blocksize
;
1823 req
->in
.h
.opcode
= FUSE_BMAP
;
1824 req
->in
.h
.nodeid
= get_node_id(inode
);
1825 req
->in
.numargs
= 1;
1826 req
->in
.args
[0].size
= sizeof(inarg
);
1827 req
->in
.args
[0].value
= &inarg
;
1828 req
->out
.numargs
= 1;
1829 req
->out
.args
[0].size
= sizeof(outarg
);
1830 req
->out
.args
[0].value
= &outarg
;
1831 fuse_request_send(fc
, req
);
1832 err
= req
->out
.h
.error
;
1833 fuse_put_request(fc
, req
);
1837 return err
? 0 : outarg
.block
;
1840 static loff_t
fuse_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1843 struct inode
*inode
= file_inode(file
);
1845 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
1846 if (whence
== SEEK_CUR
|| whence
== SEEK_SET
)
1847 return generic_file_llseek(file
, offset
, whence
);
1849 mutex_lock(&inode
->i_mutex
);
1850 retval
= fuse_update_attributes(inode
, NULL
, file
, NULL
);
1852 retval
= generic_file_llseek(file
, offset
, whence
);
1853 mutex_unlock(&inode
->i_mutex
);
1858 static int fuse_ioctl_copy_user(struct page
**pages
, struct iovec
*iov
,
1859 unsigned int nr_segs
, size_t bytes
, bool to_user
)
1867 iov_iter_init(&ii
, iov
, nr_segs
, bytes
, 0);
1869 while (iov_iter_count(&ii
)) {
1870 struct page
*page
= pages
[page_idx
++];
1871 size_t todo
= min_t(size_t, PAGE_SIZE
, iov_iter_count(&ii
));
1877 char __user
*uaddr
= ii
.iov
->iov_base
+ ii
.iov_offset
;
1878 size_t iov_len
= ii
.iov
->iov_len
- ii
.iov_offset
;
1879 size_t copy
= min(todo
, iov_len
);
1883 left
= copy_from_user(kaddr
, uaddr
, copy
);
1885 left
= copy_to_user(uaddr
, kaddr
, copy
);
1890 iov_iter_advance(&ii
, copy
);
1902 * CUSE servers compiled on 32bit broke on 64bit kernels because the
1903 * ABI was defined to be 'struct iovec' which is different on 32bit
1904 * and 64bit. Fortunately we can determine which structure the server
1905 * used from the size of the reply.
1907 static int fuse_copy_ioctl_iovec_old(struct iovec
*dst
, void *src
,
1908 size_t transferred
, unsigned count
,
1911 #ifdef CONFIG_COMPAT
1912 if (count
* sizeof(struct compat_iovec
) == transferred
) {
1913 struct compat_iovec
*ciov
= src
;
1917 * With this interface a 32bit server cannot support
1918 * non-compat (i.e. ones coming from 64bit apps) ioctl
1924 for (i
= 0; i
< count
; i
++) {
1925 dst
[i
].iov_base
= compat_ptr(ciov
[i
].iov_base
);
1926 dst
[i
].iov_len
= ciov
[i
].iov_len
;
1932 if (count
* sizeof(struct iovec
) != transferred
)
1935 memcpy(dst
, src
, transferred
);
1939 /* Make sure iov_length() won't overflow */
1940 static int fuse_verify_ioctl_iov(struct iovec
*iov
, size_t count
)
1943 u32 max
= FUSE_MAX_PAGES_PER_REQ
<< PAGE_SHIFT
;
1945 for (n
= 0; n
< count
; n
++, iov
++) {
1946 if (iov
->iov_len
> (size_t) max
)
1948 max
-= iov
->iov_len
;
1953 static int fuse_copy_ioctl_iovec(struct fuse_conn
*fc
, struct iovec
*dst
,
1954 void *src
, size_t transferred
, unsigned count
,
1958 struct fuse_ioctl_iovec
*fiov
= src
;
1960 if (fc
->minor
< 16) {
1961 return fuse_copy_ioctl_iovec_old(dst
, src
, transferred
,
1965 if (count
* sizeof(struct fuse_ioctl_iovec
) != transferred
)
1968 for (i
= 0; i
< count
; i
++) {
1969 /* Did the server supply an inappropriate value? */
1970 if (fiov
[i
].base
!= (unsigned long) fiov
[i
].base
||
1971 fiov
[i
].len
!= (unsigned long) fiov
[i
].len
)
1974 dst
[i
].iov_base
= (void __user
*) (unsigned long) fiov
[i
].base
;
1975 dst
[i
].iov_len
= (size_t) fiov
[i
].len
;
1977 #ifdef CONFIG_COMPAT
1979 (ptr_to_compat(dst
[i
].iov_base
) != fiov
[i
].base
||
1980 (compat_size_t
) dst
[i
].iov_len
!= fiov
[i
].len
))
1990 * For ioctls, there is no generic way to determine how much memory
1991 * needs to be read and/or written. Furthermore, ioctls are allowed
1992 * to dereference the passed pointer, so the parameter requires deep
1993 * copying but FUSE has no idea whatsoever about what to copy in or
1996 * This is solved by allowing FUSE server to retry ioctl with
1997 * necessary in/out iovecs. Let's assume the ioctl implementation
1998 * needs to read in the following structure.
2005 * On the first callout to FUSE server, inarg->in_size and
2006 * inarg->out_size will be NULL; then, the server completes the ioctl
2007 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
2008 * the actual iov array to
2010 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
2012 * which tells FUSE to copy in the requested area and retry the ioctl.
2013 * On the second round, the server has access to the structure and
2014 * from that it can tell what to look for next, so on the invocation,
2015 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
2017 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
2018 * { .iov_base = a.buf, .iov_len = a.buflen } }
2020 * FUSE will copy both struct a and the pointed buffer from the
2021 * process doing the ioctl and retry ioctl with both struct a and the
2024 * This time, FUSE server has everything it needs and completes ioctl
2025 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
2027 * Copying data out works the same way.
2029 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
2030 * automatically initializes in and out iovs by decoding @cmd with
2031 * _IOC_* macros and the server is not allowed to request RETRY. This
2032 * limits ioctl data transfers to well-formed ioctls and is the forced
2033 * behavior for all FUSE servers.
2035 long fuse_do_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
,
2038 struct fuse_file
*ff
= file
->private_data
;
2039 struct fuse_conn
*fc
= ff
->fc
;
2040 struct fuse_ioctl_in inarg
= {
2046 struct fuse_ioctl_out outarg
;
2047 struct fuse_req
*req
= NULL
;
2048 struct page
**pages
= NULL
;
2049 struct iovec
*iov_page
= NULL
;
2050 struct iovec
*in_iov
= NULL
, *out_iov
= NULL
;
2051 unsigned int in_iovs
= 0, out_iovs
= 0, num_pages
= 0, max_pages
;
2052 size_t in_size
, out_size
, transferred
;
2055 #if BITS_PER_LONG == 32
2056 inarg
.flags
|= FUSE_IOCTL_32BIT
;
2058 if (flags
& FUSE_IOCTL_COMPAT
)
2059 inarg
.flags
|= FUSE_IOCTL_32BIT
;
2062 /* assume all the iovs returned by client always fits in a page */
2063 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec
) * FUSE_IOCTL_MAX_IOV
> PAGE_SIZE
);
2066 pages
= kcalloc(FUSE_MAX_PAGES_PER_REQ
, sizeof(pages
[0]), GFP_KERNEL
);
2067 iov_page
= (struct iovec
*) __get_free_page(GFP_KERNEL
);
2068 if (!pages
|| !iov_page
)
2072 * If restricted, initialize IO parameters as encoded in @cmd.
2073 * RETRY from server is not allowed.
2075 if (!(flags
& FUSE_IOCTL_UNRESTRICTED
)) {
2076 struct iovec
*iov
= iov_page
;
2078 iov
->iov_base
= (void __user
*)arg
;
2079 iov
->iov_len
= _IOC_SIZE(cmd
);
2081 if (_IOC_DIR(cmd
) & _IOC_WRITE
) {
2086 if (_IOC_DIR(cmd
) & _IOC_READ
) {
2093 inarg
.in_size
= in_size
= iov_length(in_iov
, in_iovs
);
2094 inarg
.out_size
= out_size
= iov_length(out_iov
, out_iovs
);
2097 * Out data can be used either for actual out data or iovs,
2098 * make sure there always is at least one page.
2100 out_size
= max_t(size_t, out_size
, PAGE_SIZE
);
2101 max_pages
= DIV_ROUND_UP(max(in_size
, out_size
), PAGE_SIZE
);
2103 /* make sure there are enough buffer pages and init request with them */
2105 if (max_pages
> FUSE_MAX_PAGES_PER_REQ
)
2107 while (num_pages
< max_pages
) {
2108 pages
[num_pages
] = alloc_page(GFP_KERNEL
| __GFP_HIGHMEM
);
2109 if (!pages
[num_pages
])
2114 req
= fuse_get_req(fc
, num_pages
);
2120 memcpy(req
->pages
, pages
, sizeof(req
->pages
[0]) * num_pages
);
2121 req
->num_pages
= num_pages
;
2122 fuse_page_descs_length_init(req
, 0, req
->num_pages
);
2124 /* okay, let's send it to the client */
2125 req
->in
.h
.opcode
= FUSE_IOCTL
;
2126 req
->in
.h
.nodeid
= ff
->nodeid
;
2127 req
->in
.numargs
= 1;
2128 req
->in
.args
[0].size
= sizeof(inarg
);
2129 req
->in
.args
[0].value
= &inarg
;
2132 req
->in
.args
[1].size
= in_size
;
2133 req
->in
.argpages
= 1;
2135 err
= fuse_ioctl_copy_user(pages
, in_iov
, in_iovs
, in_size
,
2141 req
->out
.numargs
= 2;
2142 req
->out
.args
[0].size
= sizeof(outarg
);
2143 req
->out
.args
[0].value
= &outarg
;
2144 req
->out
.args
[1].size
= out_size
;
2145 req
->out
.argpages
= 1;
2146 req
->out
.argvar
= 1;
2148 fuse_request_send(fc
, req
);
2149 err
= req
->out
.h
.error
;
2150 transferred
= req
->out
.args
[1].size
;
2151 fuse_put_request(fc
, req
);
2156 /* did it ask for retry? */
2157 if (outarg
.flags
& FUSE_IOCTL_RETRY
) {
2160 /* no retry if in restricted mode */
2162 if (!(flags
& FUSE_IOCTL_UNRESTRICTED
))
2165 in_iovs
= outarg
.in_iovs
;
2166 out_iovs
= outarg
.out_iovs
;
2169 * Make sure things are in boundary, separate checks
2170 * are to protect against overflow.
2173 if (in_iovs
> FUSE_IOCTL_MAX_IOV
||
2174 out_iovs
> FUSE_IOCTL_MAX_IOV
||
2175 in_iovs
+ out_iovs
> FUSE_IOCTL_MAX_IOV
)
2178 vaddr
= kmap_atomic(pages
[0]);
2179 err
= fuse_copy_ioctl_iovec(fc
, iov_page
, vaddr
,
2180 transferred
, in_iovs
+ out_iovs
,
2181 (flags
& FUSE_IOCTL_COMPAT
) != 0);
2182 kunmap_atomic(vaddr
);
2187 out_iov
= in_iov
+ in_iovs
;
2189 err
= fuse_verify_ioctl_iov(in_iov
, in_iovs
);
2193 err
= fuse_verify_ioctl_iov(out_iov
, out_iovs
);
2201 if (transferred
> inarg
.out_size
)
2204 err
= fuse_ioctl_copy_user(pages
, out_iov
, out_iovs
, transferred
, true);
2207 fuse_put_request(fc
, req
);
2208 free_page((unsigned long) iov_page
);
2210 __free_page(pages
[--num_pages
]);
2213 return err
? err
: outarg
.result
;
2215 EXPORT_SYMBOL_GPL(fuse_do_ioctl
);
2217 long fuse_ioctl_common(struct file
*file
, unsigned int cmd
,
2218 unsigned long arg
, unsigned int flags
)
2220 struct inode
*inode
= file_inode(file
);
2221 struct fuse_conn
*fc
= get_fuse_conn(inode
);
2223 if (!fuse_allow_current_process(fc
))
2226 if (is_bad_inode(inode
))
2229 return fuse_do_ioctl(file
, cmd
, arg
, flags
);
2232 static long fuse_file_ioctl(struct file
*file
, unsigned int cmd
,
2235 return fuse_ioctl_common(file
, cmd
, arg
, 0);
2238 static long fuse_file_compat_ioctl(struct file
*file
, unsigned int cmd
,
2241 return fuse_ioctl_common(file
, cmd
, arg
, FUSE_IOCTL_COMPAT
);
2245 * All files which have been polled are linked to RB tree
2246 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2247 * find the matching one.
2249 static struct rb_node
**fuse_find_polled_node(struct fuse_conn
*fc
, u64 kh
,
2250 struct rb_node
**parent_out
)
2252 struct rb_node
**link
= &fc
->polled_files
.rb_node
;
2253 struct rb_node
*last
= NULL
;
2256 struct fuse_file
*ff
;
2259 ff
= rb_entry(last
, struct fuse_file
, polled_node
);
2262 link
= &last
->rb_left
;
2263 else if (kh
> ff
->kh
)
2264 link
= &last
->rb_right
;
2275 * The file is about to be polled. Make sure it's on the polled_files
2276 * RB tree. Note that files once added to the polled_files tree are
2277 * not removed before the file is released. This is because a file
2278 * polled once is likely to be polled again.
2280 static void fuse_register_polled_file(struct fuse_conn
*fc
,
2281 struct fuse_file
*ff
)
2283 spin_lock(&fc
->lock
);
2284 if (RB_EMPTY_NODE(&ff
->polled_node
)) {
2285 struct rb_node
**link
, *parent
;
2287 link
= fuse_find_polled_node(fc
, ff
->kh
, &parent
);
2289 rb_link_node(&ff
->polled_node
, parent
, link
);
2290 rb_insert_color(&ff
->polled_node
, &fc
->polled_files
);
2292 spin_unlock(&fc
->lock
);
2295 unsigned fuse_file_poll(struct file
*file
, poll_table
*wait
)
2297 struct fuse_file
*ff
= file
->private_data
;
2298 struct fuse_conn
*fc
= ff
->fc
;
2299 struct fuse_poll_in inarg
= { .fh
= ff
->fh
, .kh
= ff
->kh
};
2300 struct fuse_poll_out outarg
;
2301 struct fuse_req
*req
;
2305 return DEFAULT_POLLMASK
;
2307 poll_wait(file
, &ff
->poll_wait
, wait
);
2308 inarg
.events
= (__u32
)poll_requested_events(wait
);
2311 * Ask for notification iff there's someone waiting for it.
2312 * The client may ignore the flag and always notify.
2314 if (waitqueue_active(&ff
->poll_wait
)) {
2315 inarg
.flags
|= FUSE_POLL_SCHEDULE_NOTIFY
;
2316 fuse_register_polled_file(fc
, ff
);
2319 req
= fuse_get_req_nopages(fc
);
2323 req
->in
.h
.opcode
= FUSE_POLL
;
2324 req
->in
.h
.nodeid
= ff
->nodeid
;
2325 req
->in
.numargs
= 1;
2326 req
->in
.args
[0].size
= sizeof(inarg
);
2327 req
->in
.args
[0].value
= &inarg
;
2328 req
->out
.numargs
= 1;
2329 req
->out
.args
[0].size
= sizeof(outarg
);
2330 req
->out
.args
[0].value
= &outarg
;
2331 fuse_request_send(fc
, req
);
2332 err
= req
->out
.h
.error
;
2333 fuse_put_request(fc
, req
);
2336 return outarg
.revents
;
2337 if (err
== -ENOSYS
) {
2339 return DEFAULT_POLLMASK
;
2343 EXPORT_SYMBOL_GPL(fuse_file_poll
);
2346 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2347 * wakes up the poll waiters.
2349 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
2350 struct fuse_notify_poll_wakeup_out
*outarg
)
2352 u64 kh
= outarg
->kh
;
2353 struct rb_node
**link
;
2355 spin_lock(&fc
->lock
);
2357 link
= fuse_find_polled_node(fc
, kh
, NULL
);
2359 struct fuse_file
*ff
;
2361 ff
= rb_entry(*link
, struct fuse_file
, polled_node
);
2362 wake_up_interruptible_sync(&ff
->poll_wait
);
2365 spin_unlock(&fc
->lock
);
2369 static void fuse_do_truncate(struct file
*file
)
2371 struct inode
*inode
= file
->f_mapping
->host
;
2374 attr
.ia_valid
= ATTR_SIZE
;
2375 attr
.ia_size
= i_size_read(inode
);
2377 attr
.ia_file
= file
;
2378 attr
.ia_valid
|= ATTR_FILE
;
2380 fuse_do_setattr(inode
, &attr
, file
);
2383 static inline loff_t
fuse_round_up(loff_t off
)
2385 return round_up(off
, FUSE_MAX_PAGES_PER_REQ
<< PAGE_SHIFT
);
2389 fuse_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
2390 loff_t offset
, unsigned long nr_segs
)
2393 struct file
*file
= iocb
->ki_filp
;
2394 struct fuse_file
*ff
= file
->private_data
;
2395 bool async_dio
= ff
->fc
->async_dio
;
2397 struct inode
*inode
;
2399 size_t count
= iov_length(iov
, nr_segs
);
2400 struct fuse_io_priv
*io
;
2403 inode
= file
->f_mapping
->host
;
2404 i_size
= i_size_read(inode
);
2406 /* optimization for short read */
2407 if (async_dio
&& rw
!= WRITE
&& offset
+ count
> i_size
) {
2408 if (offset
>= i_size
)
2410 count
= min_t(loff_t
, count
, fuse_round_up(i_size
- offset
));
2413 io
= kmalloc(sizeof(struct fuse_io_priv
), GFP_KERNEL
);
2416 spin_lock_init(&io
->lock
);
2420 io
->offset
= offset
;
2421 io
->write
= (rw
== WRITE
);
2425 * By default, we want to optimize all I/Os with async request
2426 * submission to the client filesystem if supported.
2428 io
->async
= async_dio
;
2432 * We cannot asynchronously extend the size of a file. We have no method
2433 * to wait on real async I/O requests, so we must submit this request
2436 if (!is_sync_kiocb(iocb
) && (offset
+ count
> i_size
) && rw
== WRITE
)
2440 ret
= __fuse_direct_write(io
, iov
, nr_segs
, &pos
);
2442 ret
= __fuse_direct_read(io
, iov
, nr_segs
, &pos
, count
);
2445 fuse_aio_complete(io
, ret
< 0 ? ret
: 0, -1);
2447 /* we have a non-extending, async request, so return */
2448 if (!is_sync_kiocb(iocb
))
2449 return -EIOCBQUEUED
;
2451 ret
= wait_on_sync_kiocb(iocb
);
2458 fuse_write_update_size(inode
, pos
);
2459 else if (ret
< 0 && offset
+ count
> i_size
)
2460 fuse_do_truncate(file
);
2466 static long fuse_file_fallocate(struct file
*file
, int mode
, loff_t offset
,
2469 struct fuse_file
*ff
= file
->private_data
;
2470 struct inode
*inode
= file
->f_inode
;
2471 struct fuse_inode
*fi
= get_fuse_inode(inode
);
2472 struct fuse_conn
*fc
= ff
->fc
;
2473 struct fuse_req
*req
;
2474 struct fuse_fallocate_in inarg
= {
2481 bool lock_inode
= !(mode
& FALLOC_FL_KEEP_SIZE
) ||
2482 (mode
& FALLOC_FL_PUNCH_HOLE
);
2484 if (fc
->no_fallocate
)
2488 mutex_lock(&inode
->i_mutex
);
2489 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2490 loff_t endbyte
= offset
+ length
- 1;
2491 err
= filemap_write_and_wait_range(inode
->i_mapping
,
2496 fuse_sync_writes(inode
);
2500 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
2501 set_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
2503 req
= fuse_get_req_nopages(fc
);
2509 req
->in
.h
.opcode
= FUSE_FALLOCATE
;
2510 req
->in
.h
.nodeid
= ff
->nodeid
;
2511 req
->in
.numargs
= 1;
2512 req
->in
.args
[0].size
= sizeof(inarg
);
2513 req
->in
.args
[0].value
= &inarg
;
2514 fuse_request_send(fc
, req
);
2515 err
= req
->out
.h
.error
;
2516 if (err
== -ENOSYS
) {
2517 fc
->no_fallocate
= 1;
2520 fuse_put_request(fc
, req
);
2525 /* we could have extended the file */
2526 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
2527 fuse_write_update_size(inode
, offset
+ length
);
2529 if (mode
& FALLOC_FL_PUNCH_HOLE
)
2530 truncate_pagecache_range(inode
, offset
, offset
+ length
- 1);
2532 fuse_invalidate_attr(inode
);
2535 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
2536 clear_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
2539 mutex_unlock(&inode
->i_mutex
);
2544 static const struct file_operations fuse_file_operations
= {
2545 .llseek
= fuse_file_llseek
,
2546 .read
= do_sync_read
,
2547 .aio_read
= fuse_file_aio_read
,
2548 .write
= do_sync_write
,
2549 .aio_write
= fuse_file_aio_write
,
2550 .mmap
= fuse_file_mmap
,
2552 .flush
= fuse_flush
,
2553 .release
= fuse_release
,
2554 .fsync
= fuse_fsync
,
2555 .lock
= fuse_file_lock
,
2556 .flock
= fuse_file_flock
,
2557 .splice_read
= generic_file_splice_read
,
2558 .unlocked_ioctl
= fuse_file_ioctl
,
2559 .compat_ioctl
= fuse_file_compat_ioctl
,
2560 .poll
= fuse_file_poll
,
2561 .fallocate
= fuse_file_fallocate
,
2564 static const struct file_operations fuse_direct_io_file_operations
= {
2565 .llseek
= fuse_file_llseek
,
2566 .read
= fuse_direct_read
,
2567 .write
= fuse_direct_write
,
2568 .mmap
= fuse_direct_mmap
,
2570 .flush
= fuse_flush
,
2571 .release
= fuse_release
,
2572 .fsync
= fuse_fsync
,
2573 .lock
= fuse_file_lock
,
2574 .flock
= fuse_file_flock
,
2575 .unlocked_ioctl
= fuse_file_ioctl
,
2576 .compat_ioctl
= fuse_file_compat_ioctl
,
2577 .poll
= fuse_file_poll
,
2578 .fallocate
= fuse_file_fallocate
,
2579 /* no splice_read */
2582 static const struct address_space_operations fuse_file_aops
= {
2583 .readpage
= fuse_readpage
,
2584 .writepage
= fuse_writepage
,
2585 .launder_page
= fuse_launder_page
,
2586 .readpages
= fuse_readpages
,
2587 .set_page_dirty
= __set_page_dirty_nobuffers
,
2589 .direct_IO
= fuse_direct_IO
,
2592 void fuse_init_file_inode(struct inode
*inode
)
2594 inode
->i_fop
= &fuse_file_operations
;
2595 inode
->i_data
.a_ops
= &fuse_file_aops
;