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/sched/signal.h>
16 #include <linux/module.h>
17 #include <linux/swap.h>
18 #include <linux/falloc.h>
19 #include <linux/uio.h>
22 static int fuse_send_open(struct fuse_mount
*fm
, u64 nodeid
,
23 unsigned int open_flags
, int opcode
,
24 struct fuse_open_out
*outargp
)
26 struct fuse_open_in inarg
;
29 memset(&inarg
, 0, sizeof(inarg
));
30 inarg
.flags
= open_flags
& ~(O_CREAT
| O_EXCL
| O_NOCTTY
);
31 if (!fm
->fc
->atomic_o_trunc
)
32 inarg
.flags
&= ~O_TRUNC
;
34 if (fm
->fc
->handle_killpriv_v2
&&
35 (inarg
.flags
& O_TRUNC
) && !capable(CAP_FSETID
)) {
36 inarg
.open_flags
|= FUSE_OPEN_KILL_SUIDGID
;
42 args
.in_args
[0].size
= sizeof(inarg
);
43 args
.in_args
[0].value
= &inarg
;
45 args
.out_args
[0].size
= sizeof(*outargp
);
46 args
.out_args
[0].value
= outargp
;
48 return fuse_simple_request(fm
, &args
);
51 struct fuse_release_args
{
52 struct fuse_args args
;
53 struct fuse_release_in inarg
;
57 struct fuse_file
*fuse_file_alloc(struct fuse_mount
*fm
)
61 ff
= kzalloc(sizeof(struct fuse_file
), GFP_KERNEL_ACCOUNT
);
66 ff
->release_args
= kzalloc(sizeof(*ff
->release_args
),
68 if (!ff
->release_args
) {
73 INIT_LIST_HEAD(&ff
->write_entry
);
74 mutex_init(&ff
->readdir
.lock
);
75 refcount_set(&ff
->count
, 1);
76 RB_CLEAR_NODE(&ff
->polled_node
);
77 init_waitqueue_head(&ff
->poll_wait
);
79 ff
->kh
= atomic64_inc_return(&fm
->fc
->khctr
);
84 void fuse_file_free(struct fuse_file
*ff
)
86 kfree(ff
->release_args
);
87 mutex_destroy(&ff
->readdir
.lock
);
91 static struct fuse_file
*fuse_file_get(struct fuse_file
*ff
)
93 refcount_inc(&ff
->count
);
97 static void fuse_release_end(struct fuse_mount
*fm
, struct fuse_args
*args
,
100 struct fuse_release_args
*ra
= container_of(args
, typeof(*ra
), args
);
106 static void fuse_file_put(struct fuse_file
*ff
, bool sync
, bool isdir
)
108 if (refcount_dec_and_test(&ff
->count
)) {
109 struct fuse_args
*args
= &ff
->release_args
->args
;
111 if (isdir
? ff
->fm
->fc
->no_opendir
: ff
->fm
->fc
->no_open
) {
112 /* Do nothing when client does not implement 'open' */
113 fuse_release_end(ff
->fm
, args
, 0);
115 fuse_simple_request(ff
->fm
, args
);
116 fuse_release_end(ff
->fm
, args
, 0);
118 args
->end
= fuse_release_end
;
119 if (fuse_simple_background(ff
->fm
, args
,
120 GFP_KERNEL
| __GFP_NOFAIL
))
121 fuse_release_end(ff
->fm
, args
, -ENOTCONN
);
127 struct fuse_file
*fuse_file_open(struct fuse_mount
*fm
, u64 nodeid
,
128 unsigned int open_flags
, bool isdir
)
130 struct fuse_conn
*fc
= fm
->fc
;
131 struct fuse_file
*ff
;
132 int opcode
= isdir
? FUSE_OPENDIR
: FUSE_OPEN
;
134 ff
= fuse_file_alloc(fm
);
136 return ERR_PTR(-ENOMEM
);
139 /* Default for no-open */
140 ff
->open_flags
= FOPEN_KEEP_CACHE
| (isdir
? FOPEN_CACHE_DIR
: 0);
141 if (isdir
? !fc
->no_opendir
: !fc
->no_open
) {
142 struct fuse_open_out outarg
;
145 err
= fuse_send_open(fm
, nodeid
, open_flags
, opcode
, &outarg
);
148 ff
->open_flags
= outarg
.open_flags
;
150 } else if (err
!= -ENOSYS
) {
162 ff
->open_flags
&= ~FOPEN_DIRECT_IO
;
169 int fuse_do_open(struct fuse_mount
*fm
, u64 nodeid
, struct file
*file
,
172 struct fuse_file
*ff
= fuse_file_open(fm
, nodeid
, file
->f_flags
, isdir
);
175 file
->private_data
= ff
;
177 return PTR_ERR_OR_ZERO(ff
);
179 EXPORT_SYMBOL_GPL(fuse_do_open
);
181 static void fuse_link_write_file(struct file
*file
)
183 struct inode
*inode
= file_inode(file
);
184 struct fuse_inode
*fi
= get_fuse_inode(inode
);
185 struct fuse_file
*ff
= file
->private_data
;
187 * file may be written through mmap, so chain it onto the
188 * inodes's write_file list
190 spin_lock(&fi
->lock
);
191 if (list_empty(&ff
->write_entry
))
192 list_add(&ff
->write_entry
, &fi
->write_files
);
193 spin_unlock(&fi
->lock
);
196 void fuse_finish_open(struct inode
*inode
, struct file
*file
)
198 struct fuse_file
*ff
= file
->private_data
;
199 struct fuse_conn
*fc
= get_fuse_conn(inode
);
201 if (!(ff
->open_flags
& FOPEN_KEEP_CACHE
))
202 invalidate_inode_pages2(inode
->i_mapping
);
203 if (ff
->open_flags
& FOPEN_STREAM
)
204 stream_open(inode
, file
);
205 else if (ff
->open_flags
& FOPEN_NONSEEKABLE
)
206 nonseekable_open(inode
, file
);
207 if (fc
->atomic_o_trunc
&& (file
->f_flags
& O_TRUNC
)) {
208 struct fuse_inode
*fi
= get_fuse_inode(inode
);
210 spin_lock(&fi
->lock
);
211 fi
->attr_version
= atomic64_inc_return(&fc
->attr_version
);
212 i_size_write(inode
, 0);
213 spin_unlock(&fi
->lock
);
214 fuse_invalidate_attr(inode
);
215 if (fc
->writeback_cache
)
216 file_update_time(file
);
218 if ((file
->f_mode
& FMODE_WRITE
) && fc
->writeback_cache
)
219 fuse_link_write_file(file
);
222 int fuse_open_common(struct inode
*inode
, struct file
*file
, bool isdir
)
224 struct fuse_mount
*fm
= get_fuse_mount(inode
);
225 struct fuse_conn
*fc
= fm
->fc
;
227 bool is_wb_truncate
= (file
->f_flags
& O_TRUNC
) &&
228 fc
->atomic_o_trunc
&&
230 bool dax_truncate
= (file
->f_flags
& O_TRUNC
) &&
231 fc
->atomic_o_trunc
&& FUSE_IS_DAX(inode
);
233 if (fuse_is_bad(inode
))
236 err
= generic_file_open(inode
, file
);
240 if (is_wb_truncate
|| dax_truncate
) {
242 fuse_set_nowrite(inode
);
246 filemap_invalidate_lock(inode
->i_mapping
);
247 err
= fuse_dax_break_layouts(inode
, 0, 0);
252 err
= fuse_do_open(fm
, get_node_id(inode
), file
, isdir
);
254 fuse_finish_open(inode
, file
);
258 filemap_invalidate_unlock(inode
->i_mapping
);
260 if (is_wb_truncate
| dax_truncate
) {
261 fuse_release_nowrite(inode
);
268 static void fuse_prepare_release(struct fuse_inode
*fi
, struct fuse_file
*ff
,
269 unsigned int flags
, int opcode
)
271 struct fuse_conn
*fc
= ff
->fm
->fc
;
272 struct fuse_release_args
*ra
= ff
->release_args
;
274 /* Inode is NULL on error path of fuse_create_open() */
276 spin_lock(&fi
->lock
);
277 list_del(&ff
->write_entry
);
278 spin_unlock(&fi
->lock
);
280 spin_lock(&fc
->lock
);
281 if (!RB_EMPTY_NODE(&ff
->polled_node
))
282 rb_erase(&ff
->polled_node
, &fc
->polled_files
);
283 spin_unlock(&fc
->lock
);
285 wake_up_interruptible_all(&ff
->poll_wait
);
287 ra
->inarg
.fh
= ff
->fh
;
288 ra
->inarg
.flags
= flags
;
289 ra
->args
.in_numargs
= 1;
290 ra
->args
.in_args
[0].size
= sizeof(struct fuse_release_in
);
291 ra
->args
.in_args
[0].value
= &ra
->inarg
;
292 ra
->args
.opcode
= opcode
;
293 ra
->args
.nodeid
= ff
->nodeid
;
294 ra
->args
.force
= true;
295 ra
->args
.nocreds
= true;
298 void fuse_file_release(struct inode
*inode
, struct fuse_file
*ff
,
299 unsigned int open_flags
, fl_owner_t id
, bool isdir
)
301 struct fuse_inode
*fi
= get_fuse_inode(inode
);
302 struct fuse_release_args
*ra
= ff
->release_args
;
303 int opcode
= isdir
? FUSE_RELEASEDIR
: FUSE_RELEASE
;
305 fuse_prepare_release(fi
, ff
, open_flags
, opcode
);
308 ra
->inarg
.release_flags
|= FUSE_RELEASE_FLOCK_UNLOCK
;
309 ra
->inarg
.lock_owner
= fuse_lock_owner_id(ff
->fm
->fc
, id
);
311 /* Hold inode until release is finished */
312 ra
->inode
= igrab(inode
);
315 * Normally this will send the RELEASE request, however if
316 * some asynchronous READ or WRITE requests are outstanding,
317 * the sending will be delayed.
319 * Make the release synchronous if this is a fuseblk mount,
320 * synchronous RELEASE is allowed (and desirable) in this case
321 * because the server can be trusted not to screw up.
323 fuse_file_put(ff
, ff
->fm
->fc
->destroy
, isdir
);
326 void fuse_release_common(struct file
*file
, bool isdir
)
328 fuse_file_release(file_inode(file
), file
->private_data
, file
->f_flags
,
329 (fl_owner_t
) file
, isdir
);
332 static int fuse_open(struct inode
*inode
, struct file
*file
)
334 return fuse_open_common(inode
, file
, false);
337 static int fuse_release(struct inode
*inode
, struct file
*file
)
339 struct fuse_conn
*fc
= get_fuse_conn(inode
);
341 /* see fuse_vma_close() for !writeback_cache case */
342 if (fc
->writeback_cache
)
343 write_inode_now(inode
, 1);
345 fuse_release_common(file
, false);
347 /* return value is ignored by VFS */
351 void fuse_sync_release(struct fuse_inode
*fi
, struct fuse_file
*ff
,
354 WARN_ON(refcount_read(&ff
->count
) > 1);
355 fuse_prepare_release(fi
, ff
, flags
, FUSE_RELEASE
);
357 * iput(NULL) is a no-op and since the refcount is 1 and everything's
358 * synchronous, we are fine with not doing igrab() here"
360 fuse_file_put(ff
, true, false);
362 EXPORT_SYMBOL_GPL(fuse_sync_release
);
365 * Scramble the ID space with XTEA, so that the value of the files_struct
366 * pointer is not exposed to userspace.
368 u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
)
370 u32
*k
= fc
->scramble_key
;
371 u64 v
= (unsigned long) id
;
377 for (i
= 0; i
< 32; i
++) {
378 v0
+= ((v1
<< 4 ^ v1
>> 5) + v1
) ^ (sum
+ k
[sum
& 3]);
380 v1
+= ((v0
<< 4 ^ v0
>> 5) + v0
) ^ (sum
+ k
[sum
>>11 & 3]);
383 return (u64
) v0
+ ((u64
) v1
<< 32);
386 struct fuse_writepage_args
{
387 struct fuse_io_args ia
;
388 struct rb_node writepages_entry
;
389 struct list_head queue_entry
;
390 struct fuse_writepage_args
*next
;
394 static struct fuse_writepage_args
*fuse_find_writeback(struct fuse_inode
*fi
,
395 pgoff_t idx_from
, pgoff_t idx_to
)
399 n
= fi
->writepages
.rb_node
;
402 struct fuse_writepage_args
*wpa
;
405 wpa
= rb_entry(n
, struct fuse_writepage_args
, writepages_entry
);
406 WARN_ON(get_fuse_inode(wpa
->inode
) != fi
);
407 curr_index
= wpa
->ia
.write
.in
.offset
>> PAGE_SHIFT
;
408 if (idx_from
>= curr_index
+ wpa
->ia
.ap
.num_pages
)
410 else if (idx_to
< curr_index
)
419 * Check if any page in a range is under writeback
421 * This is currently done by walking the list of writepage requests
422 * for the inode, which can be pretty inefficient.
424 static bool fuse_range_is_writeback(struct inode
*inode
, pgoff_t idx_from
,
427 struct fuse_inode
*fi
= get_fuse_inode(inode
);
430 spin_lock(&fi
->lock
);
431 found
= fuse_find_writeback(fi
, idx_from
, idx_to
);
432 spin_unlock(&fi
->lock
);
437 static inline bool fuse_page_is_writeback(struct inode
*inode
, pgoff_t index
)
439 return fuse_range_is_writeback(inode
, index
, index
);
443 * Wait for page writeback to be completed.
445 * Since fuse doesn't rely on the VM writeback tracking, this has to
446 * use some other means.
448 static void fuse_wait_on_page_writeback(struct inode
*inode
, pgoff_t index
)
450 struct fuse_inode
*fi
= get_fuse_inode(inode
);
452 wait_event(fi
->page_waitq
, !fuse_page_is_writeback(inode
, index
));
456 * Wait for all pending writepages on the inode to finish.
458 * This is currently done by blocking further writes with FUSE_NOWRITE
459 * and waiting for all sent writes to complete.
461 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
462 * could conflict with truncation.
464 static void fuse_sync_writes(struct inode
*inode
)
466 fuse_set_nowrite(inode
);
467 fuse_release_nowrite(inode
);
470 static int fuse_flush(struct file
*file
, fl_owner_t id
)
472 struct inode
*inode
= file_inode(file
);
473 struct fuse_mount
*fm
= get_fuse_mount(inode
);
474 struct fuse_file
*ff
= file
->private_data
;
475 struct fuse_flush_in inarg
;
479 if (fuse_is_bad(inode
))
482 err
= write_inode_now(inode
, 1);
487 fuse_sync_writes(inode
);
490 err
= filemap_check_errors(file
->f_mapping
);
495 if (fm
->fc
->no_flush
)
498 memset(&inarg
, 0, sizeof(inarg
));
500 inarg
.lock_owner
= fuse_lock_owner_id(fm
->fc
, id
);
501 args
.opcode
= FUSE_FLUSH
;
502 args
.nodeid
= get_node_id(inode
);
504 args
.in_args
[0].size
= sizeof(inarg
);
505 args
.in_args
[0].value
= &inarg
;
508 err
= fuse_simple_request(fm
, &args
);
509 if (err
== -ENOSYS
) {
510 fm
->fc
->no_flush
= 1;
516 * In memory i_blocks is not maintained by fuse, if writeback cache is
517 * enabled, i_blocks from cached attr may not be accurate.
519 if (!err
&& fm
->fc
->writeback_cache
)
520 fuse_invalidate_attr(inode
);
524 int fuse_fsync_common(struct file
*file
, loff_t start
, loff_t end
,
525 int datasync
, int opcode
)
527 struct inode
*inode
= file
->f_mapping
->host
;
528 struct fuse_mount
*fm
= get_fuse_mount(inode
);
529 struct fuse_file
*ff
= file
->private_data
;
531 struct fuse_fsync_in inarg
;
533 memset(&inarg
, 0, sizeof(inarg
));
535 inarg
.fsync_flags
= datasync
? FUSE_FSYNC_FDATASYNC
: 0;
536 args
.opcode
= opcode
;
537 args
.nodeid
= get_node_id(inode
);
539 args
.in_args
[0].size
= sizeof(inarg
);
540 args
.in_args
[0].value
= &inarg
;
541 return fuse_simple_request(fm
, &args
);
544 static int fuse_fsync(struct file
*file
, loff_t start
, loff_t end
,
547 struct inode
*inode
= file
->f_mapping
->host
;
548 struct fuse_conn
*fc
= get_fuse_conn(inode
);
551 if (fuse_is_bad(inode
))
557 * Start writeback against all dirty pages of the inode, then
558 * wait for all outstanding writes, before sending the FSYNC
561 err
= file_write_and_wait_range(file
, start
, end
);
565 fuse_sync_writes(inode
);
568 * Due to implementation of fuse writeback
569 * file_write_and_wait_range() does not catch errors.
570 * We have to do this directly after fuse_sync_writes()
572 err
= file_check_and_advance_wb_err(file
);
576 err
= sync_inode_metadata(inode
, 1);
583 err
= fuse_fsync_common(file
, start
, end
, datasync
, FUSE_FSYNC
);
584 if (err
== -ENOSYS
) {
594 void fuse_read_args_fill(struct fuse_io_args
*ia
, struct file
*file
, loff_t pos
,
595 size_t count
, int opcode
)
597 struct fuse_file
*ff
= file
->private_data
;
598 struct fuse_args
*args
= &ia
->ap
.args
;
600 ia
->read
.in
.fh
= ff
->fh
;
601 ia
->read
.in
.offset
= pos
;
602 ia
->read
.in
.size
= count
;
603 ia
->read
.in
.flags
= file
->f_flags
;
604 args
->opcode
= opcode
;
605 args
->nodeid
= ff
->nodeid
;
606 args
->in_numargs
= 1;
607 args
->in_args
[0].size
= sizeof(ia
->read
.in
);
608 args
->in_args
[0].value
= &ia
->read
.in
;
609 args
->out_argvar
= true;
610 args
->out_numargs
= 1;
611 args
->out_args
[0].size
= count
;
614 static void fuse_release_user_pages(struct fuse_args_pages
*ap
,
619 for (i
= 0; i
< ap
->num_pages
; i
++) {
621 set_page_dirty_lock(ap
->pages
[i
]);
622 put_page(ap
->pages
[i
]);
626 static void fuse_io_release(struct kref
*kref
)
628 kfree(container_of(kref
, struct fuse_io_priv
, refcnt
));
631 static ssize_t
fuse_get_res_by_io(struct fuse_io_priv
*io
)
636 if (io
->bytes
>= 0 && io
->write
)
639 return io
->bytes
< 0 ? io
->size
: io
->bytes
;
643 * In case of short read, the caller sets 'pos' to the position of
644 * actual end of fuse request in IO request. Otherwise, if bytes_requested
645 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
648 * User requested DIO read of 64K. It was split into two 32K fuse requests,
649 * both submitted asynchronously. The first of them was ACKed by userspace as
650 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
651 * second request was ACKed as short, e.g. only 1K was read, resulting in
654 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
655 * will be equal to the length of the longest contiguous fragment of
656 * transferred data starting from the beginning of IO request.
658 static void fuse_aio_complete(struct fuse_io_priv
*io
, int err
, ssize_t pos
)
662 spin_lock(&io
->lock
);
664 io
->err
= io
->err
? : err
;
665 else if (pos
>= 0 && (io
->bytes
< 0 || pos
< io
->bytes
))
669 if (!left
&& io
->blocking
)
671 spin_unlock(&io
->lock
);
673 if (!left
&& !io
->blocking
) {
674 ssize_t res
= fuse_get_res_by_io(io
);
677 struct inode
*inode
= file_inode(io
->iocb
->ki_filp
);
678 struct fuse_conn
*fc
= get_fuse_conn(inode
);
679 struct fuse_inode
*fi
= get_fuse_inode(inode
);
681 spin_lock(&fi
->lock
);
682 fi
->attr_version
= atomic64_inc_return(&fc
->attr_version
);
683 spin_unlock(&fi
->lock
);
686 io
->iocb
->ki_complete(io
->iocb
, res
, 0);
689 kref_put(&io
->refcnt
, fuse_io_release
);
692 static struct fuse_io_args
*fuse_io_alloc(struct fuse_io_priv
*io
,
695 struct fuse_io_args
*ia
;
697 ia
= kzalloc(sizeof(*ia
), GFP_KERNEL
);
700 ia
->ap
.pages
= fuse_pages_alloc(npages
, GFP_KERNEL
,
710 static void fuse_io_free(struct fuse_io_args
*ia
)
716 static void fuse_aio_complete_req(struct fuse_mount
*fm
, struct fuse_args
*args
,
719 struct fuse_io_args
*ia
= container_of(args
, typeof(*ia
), ap
.args
);
720 struct fuse_io_priv
*io
= ia
->io
;
723 fuse_release_user_pages(&ia
->ap
, io
->should_dirty
);
727 } else if (io
->write
) {
728 if (ia
->write
.out
.size
> ia
->write
.in
.size
) {
730 } else if (ia
->write
.in
.size
!= ia
->write
.out
.size
) {
731 pos
= ia
->write
.in
.offset
- io
->offset
+
735 u32 outsize
= args
->out_args
[0].size
;
737 if (ia
->read
.in
.size
!= outsize
)
738 pos
= ia
->read
.in
.offset
- io
->offset
+ outsize
;
741 fuse_aio_complete(io
, err
, pos
);
745 static ssize_t
fuse_async_req_send(struct fuse_mount
*fm
,
746 struct fuse_io_args
*ia
, size_t num_bytes
)
749 struct fuse_io_priv
*io
= ia
->io
;
751 spin_lock(&io
->lock
);
752 kref_get(&io
->refcnt
);
753 io
->size
+= num_bytes
;
755 spin_unlock(&io
->lock
);
757 ia
->ap
.args
.end
= fuse_aio_complete_req
;
758 ia
->ap
.args
.may_block
= io
->should_dirty
;
759 err
= fuse_simple_background(fm
, &ia
->ap
.args
, GFP_KERNEL
);
761 fuse_aio_complete_req(fm
, &ia
->ap
.args
, err
);
766 static ssize_t
fuse_send_read(struct fuse_io_args
*ia
, loff_t pos
, size_t count
,
769 struct file
*file
= ia
->io
->iocb
->ki_filp
;
770 struct fuse_file
*ff
= file
->private_data
;
771 struct fuse_mount
*fm
= ff
->fm
;
773 fuse_read_args_fill(ia
, file
, pos
, count
, FUSE_READ
);
775 ia
->read
.in
.read_flags
|= FUSE_READ_LOCKOWNER
;
776 ia
->read
.in
.lock_owner
= fuse_lock_owner_id(fm
->fc
, owner
);
780 return fuse_async_req_send(fm
, ia
, count
);
782 return fuse_simple_request(fm
, &ia
->ap
.args
);
785 static void fuse_read_update_size(struct inode
*inode
, loff_t size
,
788 struct fuse_conn
*fc
= get_fuse_conn(inode
);
789 struct fuse_inode
*fi
= get_fuse_inode(inode
);
791 spin_lock(&fi
->lock
);
792 if (attr_ver
== fi
->attr_version
&& size
< inode
->i_size
&&
793 !test_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
)) {
794 fi
->attr_version
= atomic64_inc_return(&fc
->attr_version
);
795 i_size_write(inode
, size
);
797 spin_unlock(&fi
->lock
);
800 static void fuse_short_read(struct inode
*inode
, u64 attr_ver
, size_t num_read
,
801 struct fuse_args_pages
*ap
)
803 struct fuse_conn
*fc
= get_fuse_conn(inode
);
806 * If writeback_cache is enabled, a short read means there's a hole in
807 * the file. Some data after the hole is in page cache, but has not
808 * reached the client fs yet. So the hole is not present there.
810 if (!fc
->writeback_cache
) {
811 loff_t pos
= page_offset(ap
->pages
[0]) + num_read
;
812 fuse_read_update_size(inode
, pos
, attr_ver
);
816 static int fuse_do_readpage(struct file
*file
, struct page
*page
)
818 struct inode
*inode
= page
->mapping
->host
;
819 struct fuse_mount
*fm
= get_fuse_mount(inode
);
820 loff_t pos
= page_offset(page
);
821 struct fuse_page_desc desc
= { .length
= PAGE_SIZE
};
822 struct fuse_io_args ia
= {
823 .ap
.args
.page_zeroing
= true,
824 .ap
.args
.out_pages
= true,
833 * Page writeback can extend beyond the lifetime of the
834 * page-cache page, so make sure we read a properly synced
837 fuse_wait_on_page_writeback(inode
, page
->index
);
839 attr_ver
= fuse_get_attr_version(fm
->fc
);
841 /* Don't overflow end offset */
842 if (pos
+ (desc
.length
- 1) == LLONG_MAX
)
845 fuse_read_args_fill(&ia
, file
, pos
, desc
.length
, FUSE_READ
);
846 res
= fuse_simple_request(fm
, &ia
.ap
.args
);
850 * Short read means EOF. If file size is larger, truncate it
852 if (res
< desc
.length
)
853 fuse_short_read(inode
, attr_ver
, res
, &ia
.ap
);
855 SetPageUptodate(page
);
860 static int fuse_readpage(struct file
*file
, struct page
*page
)
862 struct inode
*inode
= page
->mapping
->host
;
866 if (fuse_is_bad(inode
))
869 err
= fuse_do_readpage(file
, page
);
870 fuse_invalidate_atime(inode
);
876 static void fuse_readpages_end(struct fuse_mount
*fm
, struct fuse_args
*args
,
880 struct fuse_io_args
*ia
= container_of(args
, typeof(*ia
), ap
.args
);
881 struct fuse_args_pages
*ap
= &ia
->ap
;
882 size_t count
= ia
->read
.in
.size
;
883 size_t num_read
= args
->out_args
[0].size
;
884 struct address_space
*mapping
= NULL
;
886 for (i
= 0; mapping
== NULL
&& i
< ap
->num_pages
; i
++)
887 mapping
= ap
->pages
[i
]->mapping
;
890 struct inode
*inode
= mapping
->host
;
893 * Short read means EOF. If file size is larger, truncate it
895 if (!err
&& num_read
< count
)
896 fuse_short_read(inode
, ia
->read
.attr_ver
, num_read
, ap
);
898 fuse_invalidate_atime(inode
);
901 for (i
= 0; i
< ap
->num_pages
; i
++) {
902 struct page
*page
= ap
->pages
[i
];
905 SetPageUptodate(page
);
912 fuse_file_put(ia
->ff
, false, false);
917 static void fuse_send_readpages(struct fuse_io_args
*ia
, struct file
*file
)
919 struct fuse_file
*ff
= file
->private_data
;
920 struct fuse_mount
*fm
= ff
->fm
;
921 struct fuse_args_pages
*ap
= &ia
->ap
;
922 loff_t pos
= page_offset(ap
->pages
[0]);
923 size_t count
= ap
->num_pages
<< PAGE_SHIFT
;
927 ap
->args
.out_pages
= true;
928 ap
->args
.page_zeroing
= true;
929 ap
->args
.page_replace
= true;
931 /* Don't overflow end offset */
932 if (pos
+ (count
- 1) == LLONG_MAX
) {
934 ap
->descs
[ap
->num_pages
- 1].length
--;
936 WARN_ON((loff_t
) (pos
+ count
) < 0);
938 fuse_read_args_fill(ia
, file
, pos
, count
, FUSE_READ
);
939 ia
->read
.attr_ver
= fuse_get_attr_version(fm
->fc
);
940 if (fm
->fc
->async_read
) {
941 ia
->ff
= fuse_file_get(ff
);
942 ap
->args
.end
= fuse_readpages_end
;
943 err
= fuse_simple_background(fm
, &ap
->args
, GFP_KERNEL
);
947 res
= fuse_simple_request(fm
, &ap
->args
);
948 err
= res
< 0 ? res
: 0;
950 fuse_readpages_end(fm
, &ap
->args
, err
);
953 static void fuse_readahead(struct readahead_control
*rac
)
955 struct inode
*inode
= rac
->mapping
->host
;
956 struct fuse_conn
*fc
= get_fuse_conn(inode
);
957 unsigned int i
, max_pages
, nr_pages
= 0;
959 if (fuse_is_bad(inode
))
962 max_pages
= min_t(unsigned int, fc
->max_pages
,
963 fc
->max_read
/ PAGE_SIZE
);
966 struct fuse_io_args
*ia
;
967 struct fuse_args_pages
*ap
;
969 nr_pages
= readahead_count(rac
) - nr_pages
;
970 if (nr_pages
> max_pages
)
971 nr_pages
= max_pages
;
974 ia
= fuse_io_alloc(NULL
, nr_pages
);
978 nr_pages
= __readahead_batch(rac
, ap
->pages
, nr_pages
);
979 for (i
= 0; i
< nr_pages
; i
++) {
980 fuse_wait_on_page_writeback(inode
,
981 readahead_index(rac
) + i
);
982 ap
->descs
[i
].length
= PAGE_SIZE
;
984 ap
->num_pages
= nr_pages
;
985 fuse_send_readpages(ia
, rac
->file
);
989 static ssize_t
fuse_cache_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
991 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
992 struct fuse_conn
*fc
= get_fuse_conn(inode
);
995 * In auto invalidate mode, always update attributes on read.
996 * Otherwise, only update if we attempt to read past EOF (to ensure
997 * i_size is up to date).
999 if (fc
->auto_inval_data
||
1000 (iocb
->ki_pos
+ iov_iter_count(to
) > i_size_read(inode
))) {
1002 err
= fuse_update_attributes(inode
, iocb
->ki_filp
);
1007 return generic_file_read_iter(iocb
, to
);
1010 static void fuse_write_args_fill(struct fuse_io_args
*ia
, struct fuse_file
*ff
,
1011 loff_t pos
, size_t count
)
1013 struct fuse_args
*args
= &ia
->ap
.args
;
1015 ia
->write
.in
.fh
= ff
->fh
;
1016 ia
->write
.in
.offset
= pos
;
1017 ia
->write
.in
.size
= count
;
1018 args
->opcode
= FUSE_WRITE
;
1019 args
->nodeid
= ff
->nodeid
;
1020 args
->in_numargs
= 2;
1021 if (ff
->fm
->fc
->minor
< 9)
1022 args
->in_args
[0].size
= FUSE_COMPAT_WRITE_IN_SIZE
;
1024 args
->in_args
[0].size
= sizeof(ia
->write
.in
);
1025 args
->in_args
[0].value
= &ia
->write
.in
;
1026 args
->in_args
[1].size
= count
;
1027 args
->out_numargs
= 1;
1028 args
->out_args
[0].size
= sizeof(ia
->write
.out
);
1029 args
->out_args
[0].value
= &ia
->write
.out
;
1032 static unsigned int fuse_write_flags(struct kiocb
*iocb
)
1034 unsigned int flags
= iocb
->ki_filp
->f_flags
;
1036 if (iocb
->ki_flags
& IOCB_DSYNC
)
1038 if (iocb
->ki_flags
& IOCB_SYNC
)
1044 static ssize_t
fuse_send_write(struct fuse_io_args
*ia
, loff_t pos
,
1045 size_t count
, fl_owner_t owner
)
1047 struct kiocb
*iocb
= ia
->io
->iocb
;
1048 struct file
*file
= iocb
->ki_filp
;
1049 struct fuse_file
*ff
= file
->private_data
;
1050 struct fuse_mount
*fm
= ff
->fm
;
1051 struct fuse_write_in
*inarg
= &ia
->write
.in
;
1054 fuse_write_args_fill(ia
, ff
, pos
, count
);
1055 inarg
->flags
= fuse_write_flags(iocb
);
1056 if (owner
!= NULL
) {
1057 inarg
->write_flags
|= FUSE_WRITE_LOCKOWNER
;
1058 inarg
->lock_owner
= fuse_lock_owner_id(fm
->fc
, owner
);
1062 return fuse_async_req_send(fm
, ia
, count
);
1064 err
= fuse_simple_request(fm
, &ia
->ap
.args
);
1065 if (!err
&& ia
->write
.out
.size
> count
)
1068 return err
?: ia
->write
.out
.size
;
1071 bool fuse_write_update_size(struct inode
*inode
, loff_t pos
)
1073 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1074 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1077 spin_lock(&fi
->lock
);
1078 fi
->attr_version
= atomic64_inc_return(&fc
->attr_version
);
1079 if (pos
> inode
->i_size
) {
1080 i_size_write(inode
, pos
);
1083 spin_unlock(&fi
->lock
);
1088 static ssize_t
fuse_send_write_pages(struct fuse_io_args
*ia
,
1089 struct kiocb
*iocb
, struct inode
*inode
,
1090 loff_t pos
, size_t count
)
1092 struct fuse_args_pages
*ap
= &ia
->ap
;
1093 struct file
*file
= iocb
->ki_filp
;
1094 struct fuse_file
*ff
= file
->private_data
;
1095 struct fuse_mount
*fm
= ff
->fm
;
1096 unsigned int offset
, i
;
1100 for (i
= 0; i
< ap
->num_pages
; i
++)
1101 fuse_wait_on_page_writeback(inode
, ap
->pages
[i
]->index
);
1103 fuse_write_args_fill(ia
, ff
, pos
, count
);
1104 ia
->write
.in
.flags
= fuse_write_flags(iocb
);
1105 if (fm
->fc
->handle_killpriv_v2
&& !capable(CAP_FSETID
))
1106 ia
->write
.in
.write_flags
|= FUSE_WRITE_KILL_SUIDGID
;
1108 err
= fuse_simple_request(fm
, &ap
->args
);
1109 if (!err
&& ia
->write
.out
.size
> count
)
1112 short_write
= ia
->write
.out
.size
< count
;
1113 offset
= ap
->descs
[0].offset
;
1114 count
= ia
->write
.out
.size
;
1115 for (i
= 0; i
< ap
->num_pages
; i
++) {
1116 struct page
*page
= ap
->pages
[i
];
1119 ClearPageUptodate(page
);
1121 if (count
>= PAGE_SIZE
- offset
)
1122 count
-= PAGE_SIZE
- offset
;
1125 ClearPageUptodate(page
);
1130 if (ia
->write
.page_locked
&& (i
== ap
->num_pages
- 1))
1138 static ssize_t
fuse_fill_write_pages(struct fuse_io_args
*ia
,
1139 struct address_space
*mapping
,
1140 struct iov_iter
*ii
, loff_t pos
,
1141 unsigned int max_pages
)
1143 struct fuse_args_pages
*ap
= &ia
->ap
;
1144 struct fuse_conn
*fc
= get_fuse_conn(mapping
->host
);
1145 unsigned offset
= pos
& (PAGE_SIZE
- 1);
1149 ap
->args
.in_pages
= true;
1150 ap
->descs
[0].offset
= offset
;
1155 pgoff_t index
= pos
>> PAGE_SHIFT
;
1156 size_t bytes
= min_t(size_t, PAGE_SIZE
- offset
,
1157 iov_iter_count(ii
));
1159 bytes
= min_t(size_t, bytes
, fc
->max_write
- count
);
1163 if (iov_iter_fault_in_readable(ii
, bytes
))
1167 page
= grab_cache_page_write_begin(mapping
, index
, 0);
1171 if (mapping_writably_mapped(mapping
))
1172 flush_dcache_page(page
);
1174 tmp
= copy_page_from_iter_atomic(page
, offset
, bytes
, ii
);
1175 flush_dcache_page(page
);
1184 ap
->pages
[ap
->num_pages
] = page
;
1185 ap
->descs
[ap
->num_pages
].length
= tmp
;
1191 if (offset
== PAGE_SIZE
)
1194 /* If we copied full page, mark it uptodate */
1195 if (tmp
== PAGE_SIZE
)
1196 SetPageUptodate(page
);
1198 if (PageUptodate(page
)) {
1201 ia
->write
.page_locked
= true;
1204 if (!fc
->big_writes
)
1206 } while (iov_iter_count(ii
) && count
< fc
->max_write
&&
1207 ap
->num_pages
< max_pages
&& offset
== 0);
1209 return count
> 0 ? count
: err
;
1212 static inline unsigned int fuse_wr_pages(loff_t pos
, size_t len
,
1213 unsigned int max_pages
)
1215 return min_t(unsigned int,
1216 ((pos
+ len
- 1) >> PAGE_SHIFT
) -
1217 (pos
>> PAGE_SHIFT
) + 1,
1221 static ssize_t
fuse_perform_write(struct kiocb
*iocb
,
1222 struct address_space
*mapping
,
1223 struct iov_iter
*ii
, loff_t pos
)
1225 struct inode
*inode
= mapping
->host
;
1226 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1227 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1231 if (inode
->i_size
< pos
+ iov_iter_count(ii
))
1232 set_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
1236 struct fuse_io_args ia
= {};
1237 struct fuse_args_pages
*ap
= &ia
.ap
;
1238 unsigned int nr_pages
= fuse_wr_pages(pos
, iov_iter_count(ii
),
1241 ap
->pages
= fuse_pages_alloc(nr_pages
, GFP_KERNEL
, &ap
->descs
);
1247 count
= fuse_fill_write_pages(&ia
, mapping
, ii
, pos
, nr_pages
);
1251 err
= fuse_send_write_pages(&ia
, iocb
, inode
,
1254 size_t num_written
= ia
.write
.out
.size
;
1259 /* break out of the loop on short write */
1260 if (num_written
!= count
)
1265 } while (!err
&& iov_iter_count(ii
));
1268 fuse_write_update_size(inode
, pos
);
1270 clear_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
1271 fuse_invalidate_attr(inode
);
1273 return res
> 0 ? res
: err
;
1276 static ssize_t
fuse_cache_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1278 struct file
*file
= iocb
->ki_filp
;
1279 struct address_space
*mapping
= file
->f_mapping
;
1280 ssize_t written
= 0;
1281 ssize_t written_buffered
= 0;
1282 struct inode
*inode
= mapping
->host
;
1284 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1287 if (fc
->writeback_cache
) {
1288 /* Update size (EOF optimization) and mode (SUID clearing) */
1289 err
= fuse_update_attributes(mapping
->host
, file
);
1293 if (fc
->handle_killpriv_v2
&&
1294 should_remove_suid(file_dentry(file
))) {
1298 return generic_file_write_iter(iocb
, from
);
1304 /* We can write back this queue in page reclaim */
1305 current
->backing_dev_info
= inode_to_bdi(inode
);
1307 err
= generic_write_checks(iocb
, from
);
1311 err
= file_remove_privs(file
);
1315 err
= file_update_time(file
);
1319 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1320 loff_t pos
= iocb
->ki_pos
;
1321 written
= generic_file_direct_write(iocb
, from
);
1322 if (written
< 0 || !iov_iter_count(from
))
1327 written_buffered
= fuse_perform_write(iocb
, mapping
, from
, pos
);
1328 if (written_buffered
< 0) {
1329 err
= written_buffered
;
1332 endbyte
= pos
+ written_buffered
- 1;
1334 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
,
1339 invalidate_mapping_pages(file
->f_mapping
,
1341 endbyte
>> PAGE_SHIFT
);
1343 written
+= written_buffered
;
1344 iocb
->ki_pos
= pos
+ written_buffered
;
1346 written
= fuse_perform_write(iocb
, mapping
, from
, iocb
->ki_pos
);
1348 iocb
->ki_pos
+= written
;
1351 current
->backing_dev_info
= NULL
;
1352 inode_unlock(inode
);
1354 written
= generic_write_sync(iocb
, written
);
1356 return written
? written
: err
;
1359 static inline unsigned long fuse_get_user_addr(const struct iov_iter
*ii
)
1361 return (unsigned long)ii
->iov
->iov_base
+ ii
->iov_offset
;
1364 static inline size_t fuse_get_frag_size(const struct iov_iter
*ii
,
1367 return min(iov_iter_single_seg_count(ii
), max_size
);
1370 static int fuse_get_user_pages(struct fuse_args_pages
*ap
, struct iov_iter
*ii
,
1371 size_t *nbytesp
, int write
,
1372 unsigned int max_pages
)
1374 size_t nbytes
= 0; /* # bytes already packed in req */
1377 /* Special case for kernel I/O: can copy directly into the buffer */
1378 if (iov_iter_is_kvec(ii
)) {
1379 unsigned long user_addr
= fuse_get_user_addr(ii
);
1380 size_t frag_size
= fuse_get_frag_size(ii
, *nbytesp
);
1383 ap
->args
.in_args
[1].value
= (void *) user_addr
;
1385 ap
->args
.out_args
[0].value
= (void *) user_addr
;
1387 iov_iter_advance(ii
, frag_size
);
1388 *nbytesp
= frag_size
;
1392 while (nbytes
< *nbytesp
&& ap
->num_pages
< max_pages
) {
1395 ret
= iov_iter_get_pages(ii
, &ap
->pages
[ap
->num_pages
],
1397 max_pages
- ap
->num_pages
,
1402 iov_iter_advance(ii
, ret
);
1406 npages
= DIV_ROUND_UP(ret
, PAGE_SIZE
);
1408 ap
->descs
[ap
->num_pages
].offset
= start
;
1409 fuse_page_descs_length_init(ap
->descs
, ap
->num_pages
, npages
);
1411 ap
->num_pages
+= npages
;
1412 ap
->descs
[ap
->num_pages
- 1].length
-=
1413 (PAGE_SIZE
- ret
) & (PAGE_SIZE
- 1);
1417 ap
->args
.in_pages
= true;
1419 ap
->args
.out_pages
= true;
1423 return ret
< 0 ? ret
: 0;
1426 ssize_t
fuse_direct_io(struct fuse_io_priv
*io
, struct iov_iter
*iter
,
1427 loff_t
*ppos
, int flags
)
1429 int write
= flags
& FUSE_DIO_WRITE
;
1430 int cuse
= flags
& FUSE_DIO_CUSE
;
1431 struct file
*file
= io
->iocb
->ki_filp
;
1432 struct inode
*inode
= file
->f_mapping
->host
;
1433 struct fuse_file
*ff
= file
->private_data
;
1434 struct fuse_conn
*fc
= ff
->fm
->fc
;
1435 size_t nmax
= write
? fc
->max_write
: fc
->max_read
;
1437 size_t count
= iov_iter_count(iter
);
1438 pgoff_t idx_from
= pos
>> PAGE_SHIFT
;
1439 pgoff_t idx_to
= (pos
+ count
- 1) >> PAGE_SHIFT
;
1442 struct fuse_io_args
*ia
;
1443 unsigned int max_pages
;
1445 max_pages
= iov_iter_npages(iter
, fc
->max_pages
);
1446 ia
= fuse_io_alloc(io
, max_pages
);
1451 if (!cuse
&& fuse_range_is_writeback(inode
, idx_from
, idx_to
)) {
1454 fuse_sync_writes(inode
);
1456 inode_unlock(inode
);
1459 io
->should_dirty
= !write
&& iter_is_iovec(iter
);
1462 fl_owner_t owner
= current
->files
;
1463 size_t nbytes
= min(count
, nmax
);
1465 err
= fuse_get_user_pages(&ia
->ap
, iter
, &nbytes
, write
,
1471 if (!capable(CAP_FSETID
))
1472 ia
->write
.in
.write_flags
|= FUSE_WRITE_KILL_SUIDGID
;
1474 nres
= fuse_send_write(ia
, pos
, nbytes
, owner
);
1476 nres
= fuse_send_read(ia
, pos
, nbytes
, owner
);
1479 if (!io
->async
|| nres
< 0) {
1480 fuse_release_user_pages(&ia
->ap
, io
->should_dirty
);
1485 iov_iter_revert(iter
, nbytes
);
1489 WARN_ON(nres
> nbytes
);
1494 if (nres
!= nbytes
) {
1495 iov_iter_revert(iter
, nbytes
- nres
);
1499 max_pages
= iov_iter_npages(iter
, fc
->max_pages
);
1500 ia
= fuse_io_alloc(io
, max_pages
);
1510 return res
> 0 ? res
: err
;
1512 EXPORT_SYMBOL_GPL(fuse_direct_io
);
1514 static ssize_t
__fuse_direct_read(struct fuse_io_priv
*io
,
1515 struct iov_iter
*iter
,
1519 struct inode
*inode
= file_inode(io
->iocb
->ki_filp
);
1521 res
= fuse_direct_io(io
, iter
, ppos
, 0);
1523 fuse_invalidate_atime(inode
);
1528 static ssize_t
fuse_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
);
1530 static ssize_t
fuse_direct_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1534 if (!is_sync_kiocb(iocb
) && iocb
->ki_flags
& IOCB_DIRECT
) {
1535 res
= fuse_direct_IO(iocb
, to
);
1537 struct fuse_io_priv io
= FUSE_IO_PRIV_SYNC(iocb
);
1539 res
= __fuse_direct_read(&io
, to
, &iocb
->ki_pos
);
1545 static ssize_t
fuse_direct_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1547 struct inode
*inode
= file_inode(iocb
->ki_filp
);
1548 struct fuse_io_priv io
= FUSE_IO_PRIV_SYNC(iocb
);
1551 /* Don't allow parallel writes to the same file */
1553 res
= generic_write_checks(iocb
, from
);
1555 if (!is_sync_kiocb(iocb
) && iocb
->ki_flags
& IOCB_DIRECT
) {
1556 res
= fuse_direct_IO(iocb
, from
);
1558 res
= fuse_direct_io(&io
, from
, &iocb
->ki_pos
,
1562 fuse_invalidate_attr(inode
);
1564 fuse_write_update_size(inode
, iocb
->ki_pos
);
1565 inode_unlock(inode
);
1570 static ssize_t
fuse_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1572 struct file
*file
= iocb
->ki_filp
;
1573 struct fuse_file
*ff
= file
->private_data
;
1574 struct inode
*inode
= file_inode(file
);
1576 if (fuse_is_bad(inode
))
1579 if (FUSE_IS_DAX(inode
))
1580 return fuse_dax_read_iter(iocb
, to
);
1582 if (!(ff
->open_flags
& FOPEN_DIRECT_IO
))
1583 return fuse_cache_read_iter(iocb
, to
);
1585 return fuse_direct_read_iter(iocb
, to
);
1588 static ssize_t
fuse_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1590 struct file
*file
= iocb
->ki_filp
;
1591 struct fuse_file
*ff
= file
->private_data
;
1592 struct inode
*inode
= file_inode(file
);
1594 if (fuse_is_bad(inode
))
1597 if (FUSE_IS_DAX(inode
))
1598 return fuse_dax_write_iter(iocb
, from
);
1600 if (!(ff
->open_flags
& FOPEN_DIRECT_IO
))
1601 return fuse_cache_write_iter(iocb
, from
);
1603 return fuse_direct_write_iter(iocb
, from
);
1606 static void fuse_writepage_free(struct fuse_writepage_args
*wpa
)
1608 struct fuse_args_pages
*ap
= &wpa
->ia
.ap
;
1611 for (i
= 0; i
< ap
->num_pages
; i
++)
1612 __free_page(ap
->pages
[i
]);
1615 fuse_file_put(wpa
->ia
.ff
, false, false);
1621 static void fuse_writepage_finish(struct fuse_mount
*fm
,
1622 struct fuse_writepage_args
*wpa
)
1624 struct fuse_args_pages
*ap
= &wpa
->ia
.ap
;
1625 struct inode
*inode
= wpa
->inode
;
1626 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1627 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
1630 for (i
= 0; i
< ap
->num_pages
; i
++) {
1631 dec_wb_stat(&bdi
->wb
, WB_WRITEBACK
);
1632 dec_node_page_state(ap
->pages
[i
], NR_WRITEBACK_TEMP
);
1633 wb_writeout_inc(&bdi
->wb
);
1635 wake_up(&fi
->page_waitq
);
1638 /* Called under fi->lock, may release and reacquire it */
1639 static void fuse_send_writepage(struct fuse_mount
*fm
,
1640 struct fuse_writepage_args
*wpa
, loff_t size
)
1641 __releases(fi
->lock
)
1642 __acquires(fi
->lock
)
1644 struct fuse_writepage_args
*aux
, *next
;
1645 struct fuse_inode
*fi
= get_fuse_inode(wpa
->inode
);
1646 struct fuse_write_in
*inarg
= &wpa
->ia
.write
.in
;
1647 struct fuse_args
*args
= &wpa
->ia
.ap
.args
;
1648 __u64 data_size
= wpa
->ia
.ap
.num_pages
* PAGE_SIZE
;
1652 if (inarg
->offset
+ data_size
<= size
) {
1653 inarg
->size
= data_size
;
1654 } else if (inarg
->offset
< size
) {
1655 inarg
->size
= size
- inarg
->offset
;
1657 /* Got truncated off completely */
1661 args
->in_args
[1].size
= inarg
->size
;
1663 args
->nocreds
= true;
1665 err
= fuse_simple_background(fm
, args
, GFP_ATOMIC
);
1666 if (err
== -ENOMEM
) {
1667 spin_unlock(&fi
->lock
);
1668 err
= fuse_simple_background(fm
, args
, GFP_NOFS
| __GFP_NOFAIL
);
1669 spin_lock(&fi
->lock
);
1672 /* Fails on broken connection only */
1680 rb_erase(&wpa
->writepages_entry
, &fi
->writepages
);
1681 fuse_writepage_finish(fm
, wpa
);
1682 spin_unlock(&fi
->lock
);
1684 /* After fuse_writepage_finish() aux request list is private */
1685 for (aux
= wpa
->next
; aux
; aux
= next
) {
1688 fuse_writepage_free(aux
);
1691 fuse_writepage_free(wpa
);
1692 spin_lock(&fi
->lock
);
1696 * If fi->writectr is positive (no truncate or fsync going on) send
1697 * all queued writepage requests.
1699 * Called with fi->lock
1701 void fuse_flush_writepages(struct inode
*inode
)
1702 __releases(fi
->lock
)
1703 __acquires(fi
->lock
)
1705 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1706 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1707 loff_t crop
= i_size_read(inode
);
1708 struct fuse_writepage_args
*wpa
;
1710 while (fi
->writectr
>= 0 && !list_empty(&fi
->queued_writes
)) {
1711 wpa
= list_entry(fi
->queued_writes
.next
,
1712 struct fuse_writepage_args
, queue_entry
);
1713 list_del_init(&wpa
->queue_entry
);
1714 fuse_send_writepage(fm
, wpa
, crop
);
1718 static struct fuse_writepage_args
*fuse_insert_writeback(struct rb_root
*root
,
1719 struct fuse_writepage_args
*wpa
)
1721 pgoff_t idx_from
= wpa
->ia
.write
.in
.offset
>> PAGE_SHIFT
;
1722 pgoff_t idx_to
= idx_from
+ wpa
->ia
.ap
.num_pages
- 1;
1723 struct rb_node
**p
= &root
->rb_node
;
1724 struct rb_node
*parent
= NULL
;
1726 WARN_ON(!wpa
->ia
.ap
.num_pages
);
1728 struct fuse_writepage_args
*curr
;
1732 curr
= rb_entry(parent
, struct fuse_writepage_args
,
1734 WARN_ON(curr
->inode
!= wpa
->inode
);
1735 curr_index
= curr
->ia
.write
.in
.offset
>> PAGE_SHIFT
;
1737 if (idx_from
>= curr_index
+ curr
->ia
.ap
.num_pages
)
1738 p
= &(*p
)->rb_right
;
1739 else if (idx_to
< curr_index
)
1745 rb_link_node(&wpa
->writepages_entry
, parent
, p
);
1746 rb_insert_color(&wpa
->writepages_entry
, root
);
1750 static void tree_insert(struct rb_root
*root
, struct fuse_writepage_args
*wpa
)
1752 WARN_ON(fuse_insert_writeback(root
, wpa
));
1755 static void fuse_writepage_end(struct fuse_mount
*fm
, struct fuse_args
*args
,
1758 struct fuse_writepage_args
*wpa
=
1759 container_of(args
, typeof(*wpa
), ia
.ap
.args
);
1760 struct inode
*inode
= wpa
->inode
;
1761 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1762 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1764 mapping_set_error(inode
->i_mapping
, error
);
1766 * A writeback finished and this might have updated mtime/ctime on
1767 * server making local mtime/ctime stale. Hence invalidate attrs.
1768 * Do this only if writeback_cache is not enabled. If writeback_cache
1769 * is enabled, we trust local ctime/mtime.
1771 if (!fc
->writeback_cache
)
1772 fuse_invalidate_attr(inode
);
1773 spin_lock(&fi
->lock
);
1774 rb_erase(&wpa
->writepages_entry
, &fi
->writepages
);
1776 struct fuse_mount
*fm
= get_fuse_mount(inode
);
1777 struct fuse_write_in
*inarg
= &wpa
->ia
.write
.in
;
1778 struct fuse_writepage_args
*next
= wpa
->next
;
1780 wpa
->next
= next
->next
;
1782 next
->ia
.ff
= fuse_file_get(wpa
->ia
.ff
);
1783 tree_insert(&fi
->writepages
, next
);
1786 * Skip fuse_flush_writepages() to make it easy to crop requests
1787 * based on primary request size.
1789 * 1st case (trivial): there are no concurrent activities using
1790 * fuse_set/release_nowrite. Then we're on safe side because
1791 * fuse_flush_writepages() would call fuse_send_writepage()
1794 * 2nd case: someone called fuse_set_nowrite and it is waiting
1795 * now for completion of all in-flight requests. This happens
1796 * rarely and no more than once per page, so this should be
1799 * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle
1800 * of fuse_set_nowrite..fuse_release_nowrite section. The fact
1801 * that fuse_set_nowrite returned implies that all in-flight
1802 * requests were completed along with all of their secondary
1803 * requests. Further primary requests are blocked by negative
1804 * writectr. Hence there cannot be any in-flight requests and
1805 * no invocations of fuse_writepage_end() while we're in
1806 * fuse_set_nowrite..fuse_release_nowrite section.
1808 fuse_send_writepage(fm
, next
, inarg
->offset
+ inarg
->size
);
1811 fuse_writepage_finish(fm
, wpa
);
1812 spin_unlock(&fi
->lock
);
1813 fuse_writepage_free(wpa
);
1816 static struct fuse_file
*__fuse_write_file_get(struct fuse_conn
*fc
,
1817 struct fuse_inode
*fi
)
1819 struct fuse_file
*ff
= NULL
;
1821 spin_lock(&fi
->lock
);
1822 if (!list_empty(&fi
->write_files
)) {
1823 ff
= list_entry(fi
->write_files
.next
, struct fuse_file
,
1827 spin_unlock(&fi
->lock
);
1832 static struct fuse_file
*fuse_write_file_get(struct fuse_conn
*fc
,
1833 struct fuse_inode
*fi
)
1835 struct fuse_file
*ff
= __fuse_write_file_get(fc
, fi
);
1840 int fuse_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1842 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1843 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1844 struct fuse_file
*ff
;
1847 ff
= __fuse_write_file_get(fc
, fi
);
1848 err
= fuse_flush_times(inode
, ff
);
1850 fuse_file_put(ff
, false, false);
1855 static struct fuse_writepage_args
*fuse_writepage_args_alloc(void)
1857 struct fuse_writepage_args
*wpa
;
1858 struct fuse_args_pages
*ap
;
1860 wpa
= kzalloc(sizeof(*wpa
), GFP_NOFS
);
1864 ap
->pages
= fuse_pages_alloc(1, GFP_NOFS
, &ap
->descs
);
1874 static int fuse_writepage_locked(struct page
*page
)
1876 struct address_space
*mapping
= page
->mapping
;
1877 struct inode
*inode
= mapping
->host
;
1878 struct fuse_conn
*fc
= get_fuse_conn(inode
);
1879 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1880 struct fuse_writepage_args
*wpa
;
1881 struct fuse_args_pages
*ap
;
1882 struct page
*tmp_page
;
1883 int error
= -ENOMEM
;
1885 set_page_writeback(page
);
1887 wpa
= fuse_writepage_args_alloc();
1892 tmp_page
= alloc_page(GFP_NOFS
| __GFP_HIGHMEM
);
1897 wpa
->ia
.ff
= fuse_write_file_get(fc
, fi
);
1901 fuse_write_args_fill(&wpa
->ia
, wpa
->ia
.ff
, page_offset(page
), 0);
1903 copy_highpage(tmp_page
, page
);
1904 wpa
->ia
.write
.in
.write_flags
|= FUSE_WRITE_CACHE
;
1906 ap
->args
.in_pages
= true;
1908 ap
->pages
[0] = tmp_page
;
1909 ap
->descs
[0].offset
= 0;
1910 ap
->descs
[0].length
= PAGE_SIZE
;
1911 ap
->args
.end
= fuse_writepage_end
;
1914 inc_wb_stat(&inode_to_bdi(inode
)->wb
, WB_WRITEBACK
);
1915 inc_node_page_state(tmp_page
, NR_WRITEBACK_TEMP
);
1917 spin_lock(&fi
->lock
);
1918 tree_insert(&fi
->writepages
, wpa
);
1919 list_add_tail(&wpa
->queue_entry
, &fi
->queued_writes
);
1920 fuse_flush_writepages(inode
);
1921 spin_unlock(&fi
->lock
);
1923 end_page_writeback(page
);
1928 __free_page(tmp_page
);
1932 mapping_set_error(page
->mapping
, error
);
1933 end_page_writeback(page
);
1937 static int fuse_writepage(struct page
*page
, struct writeback_control
*wbc
)
1941 if (fuse_page_is_writeback(page
->mapping
->host
, page
->index
)) {
1943 * ->writepages() should be called for sync() and friends. We
1944 * should only get here on direct reclaim and then we are
1945 * allowed to skip a page which is already in flight
1947 WARN_ON(wbc
->sync_mode
== WB_SYNC_ALL
);
1949 redirty_page_for_writepage(wbc
, page
);
1954 err
= fuse_writepage_locked(page
);
1960 struct fuse_fill_wb_data
{
1961 struct fuse_writepage_args
*wpa
;
1962 struct fuse_file
*ff
;
1963 struct inode
*inode
;
1964 struct page
**orig_pages
;
1965 unsigned int max_pages
;
1968 static bool fuse_pages_realloc(struct fuse_fill_wb_data
*data
)
1970 struct fuse_args_pages
*ap
= &data
->wpa
->ia
.ap
;
1971 struct fuse_conn
*fc
= get_fuse_conn(data
->inode
);
1972 struct page
**pages
;
1973 struct fuse_page_desc
*descs
;
1974 unsigned int npages
= min_t(unsigned int,
1975 max_t(unsigned int, data
->max_pages
* 2,
1976 FUSE_DEFAULT_MAX_PAGES_PER_REQ
),
1978 WARN_ON(npages
<= data
->max_pages
);
1980 pages
= fuse_pages_alloc(npages
, GFP_NOFS
, &descs
);
1984 memcpy(pages
, ap
->pages
, sizeof(struct page
*) * ap
->num_pages
);
1985 memcpy(descs
, ap
->descs
, sizeof(struct fuse_page_desc
) * ap
->num_pages
);
1989 data
->max_pages
= npages
;
1994 static void fuse_writepages_send(struct fuse_fill_wb_data
*data
)
1996 struct fuse_writepage_args
*wpa
= data
->wpa
;
1997 struct inode
*inode
= data
->inode
;
1998 struct fuse_inode
*fi
= get_fuse_inode(inode
);
1999 int num_pages
= wpa
->ia
.ap
.num_pages
;
2002 wpa
->ia
.ff
= fuse_file_get(data
->ff
);
2003 spin_lock(&fi
->lock
);
2004 list_add_tail(&wpa
->queue_entry
, &fi
->queued_writes
);
2005 fuse_flush_writepages(inode
);
2006 spin_unlock(&fi
->lock
);
2008 for (i
= 0; i
< num_pages
; i
++)
2009 end_page_writeback(data
->orig_pages
[i
]);
2013 * Check under fi->lock if the page is under writeback, and insert it onto the
2014 * rb_tree if not. Otherwise iterate auxiliary write requests, to see if there's
2015 * one already added for a page at this offset. If there's none, then insert
2016 * this new request onto the auxiliary list, otherwise reuse the existing one by
2017 * swapping the new temp page with the old one.
2019 static bool fuse_writepage_add(struct fuse_writepage_args
*new_wpa
,
2022 struct fuse_inode
*fi
= get_fuse_inode(new_wpa
->inode
);
2023 struct fuse_writepage_args
*tmp
;
2024 struct fuse_writepage_args
*old_wpa
;
2025 struct fuse_args_pages
*new_ap
= &new_wpa
->ia
.ap
;
2027 WARN_ON(new_ap
->num_pages
!= 0);
2028 new_ap
->num_pages
= 1;
2030 spin_lock(&fi
->lock
);
2031 old_wpa
= fuse_insert_writeback(&fi
->writepages
, new_wpa
);
2033 spin_unlock(&fi
->lock
);
2037 for (tmp
= old_wpa
->next
; tmp
; tmp
= tmp
->next
) {
2040 WARN_ON(tmp
->inode
!= new_wpa
->inode
);
2041 curr_index
= tmp
->ia
.write
.in
.offset
>> PAGE_SHIFT
;
2042 if (curr_index
== page
->index
) {
2043 WARN_ON(tmp
->ia
.ap
.num_pages
!= 1);
2044 swap(tmp
->ia
.ap
.pages
[0], new_ap
->pages
[0]);
2050 new_wpa
->next
= old_wpa
->next
;
2051 old_wpa
->next
= new_wpa
;
2054 spin_unlock(&fi
->lock
);
2057 struct backing_dev_info
*bdi
= inode_to_bdi(new_wpa
->inode
);
2059 dec_wb_stat(&bdi
->wb
, WB_WRITEBACK
);
2060 dec_node_page_state(new_ap
->pages
[0], NR_WRITEBACK_TEMP
);
2061 wb_writeout_inc(&bdi
->wb
);
2062 fuse_writepage_free(new_wpa
);
2068 static bool fuse_writepage_need_send(struct fuse_conn
*fc
, struct page
*page
,
2069 struct fuse_args_pages
*ap
,
2070 struct fuse_fill_wb_data
*data
)
2072 WARN_ON(!ap
->num_pages
);
2075 * Being under writeback is unlikely but possible. For example direct
2076 * read to an mmaped fuse file will set the page dirty twice; once when
2077 * the pages are faulted with get_user_pages(), and then after the read
2080 if (fuse_page_is_writeback(data
->inode
, page
->index
))
2083 /* Reached max pages */
2084 if (ap
->num_pages
== fc
->max_pages
)
2087 /* Reached max write bytes */
2088 if ((ap
->num_pages
+ 1) * PAGE_SIZE
> fc
->max_write
)
2092 if (data
->orig_pages
[ap
->num_pages
- 1]->index
+ 1 != page
->index
)
2095 /* Need to grow the pages array? If so, did the expansion fail? */
2096 if (ap
->num_pages
== data
->max_pages
&& !fuse_pages_realloc(data
))
2102 static int fuse_writepages_fill(struct page
*page
,
2103 struct writeback_control
*wbc
, void *_data
)
2105 struct fuse_fill_wb_data
*data
= _data
;
2106 struct fuse_writepage_args
*wpa
= data
->wpa
;
2107 struct fuse_args_pages
*ap
= &wpa
->ia
.ap
;
2108 struct inode
*inode
= data
->inode
;
2109 struct fuse_inode
*fi
= get_fuse_inode(inode
);
2110 struct fuse_conn
*fc
= get_fuse_conn(inode
);
2111 struct page
*tmp_page
;
2116 data
->ff
= fuse_write_file_get(fc
, fi
);
2121 if (wpa
&& fuse_writepage_need_send(fc
, page
, ap
, data
)) {
2122 fuse_writepages_send(data
);
2127 tmp_page
= alloc_page(GFP_NOFS
| __GFP_HIGHMEM
);
2132 * The page must not be redirtied until the writeout is completed
2133 * (i.e. userspace has sent a reply to the write request). Otherwise
2134 * there could be more than one temporary page instance for each real
2137 * This is ensured by holding the page lock in page_mkwrite() while
2138 * checking fuse_page_is_writeback(). We already hold the page lock
2139 * since clear_page_dirty_for_io() and keep it held until we add the
2140 * request to the fi->writepages list and increment ap->num_pages.
2141 * After this fuse_page_is_writeback() will indicate that the page is
2142 * under writeback, so we can release the page lock.
2144 if (data
->wpa
== NULL
) {
2146 wpa
= fuse_writepage_args_alloc();
2148 __free_page(tmp_page
);
2151 data
->max_pages
= 1;
2154 fuse_write_args_fill(&wpa
->ia
, data
->ff
, page_offset(page
), 0);
2155 wpa
->ia
.write
.in
.write_flags
|= FUSE_WRITE_CACHE
;
2157 ap
->args
.in_pages
= true;
2158 ap
->args
.end
= fuse_writepage_end
;
2162 set_page_writeback(page
);
2164 copy_highpage(tmp_page
, page
);
2165 ap
->pages
[ap
->num_pages
] = tmp_page
;
2166 ap
->descs
[ap
->num_pages
].offset
= 0;
2167 ap
->descs
[ap
->num_pages
].length
= PAGE_SIZE
;
2168 data
->orig_pages
[ap
->num_pages
] = page
;
2170 inc_wb_stat(&inode_to_bdi(inode
)->wb
, WB_WRITEBACK
);
2171 inc_node_page_state(tmp_page
, NR_WRITEBACK_TEMP
);
2176 * Protected by fi->lock against concurrent access by
2177 * fuse_page_is_writeback().
2179 spin_lock(&fi
->lock
);
2181 spin_unlock(&fi
->lock
);
2182 } else if (fuse_writepage_add(wpa
, page
)) {
2185 end_page_writeback(page
);
2193 static int fuse_writepages(struct address_space
*mapping
,
2194 struct writeback_control
*wbc
)
2196 struct inode
*inode
= mapping
->host
;
2197 struct fuse_conn
*fc
= get_fuse_conn(inode
);
2198 struct fuse_fill_wb_data data
;
2202 if (fuse_is_bad(inode
))
2210 data
.orig_pages
= kcalloc(fc
->max_pages
,
2211 sizeof(struct page
*),
2213 if (!data
.orig_pages
)
2216 err
= write_cache_pages(mapping
, wbc
, fuse_writepages_fill
, &data
);
2218 WARN_ON(!data
.wpa
->ia
.ap
.num_pages
);
2219 fuse_writepages_send(&data
);
2222 fuse_file_put(data
.ff
, false, false);
2224 kfree(data
.orig_pages
);
2230 * It's worthy to make sure that space is reserved on disk for the write,
2231 * but how to implement it without killing performance need more thinking.
2233 static int fuse_write_begin(struct file
*file
, struct address_space
*mapping
,
2234 loff_t pos
, unsigned len
, unsigned flags
,
2235 struct page
**pagep
, void **fsdata
)
2237 pgoff_t index
= pos
>> PAGE_SHIFT
;
2238 struct fuse_conn
*fc
= get_fuse_conn(file_inode(file
));
2243 WARN_ON(!fc
->writeback_cache
);
2245 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2249 fuse_wait_on_page_writeback(mapping
->host
, page
->index
);
2251 if (PageUptodate(page
) || len
== PAGE_SIZE
)
2254 * Check if the start this page comes after the end of file, in which
2255 * case the readpage can be optimized away.
2257 fsize
= i_size_read(mapping
->host
);
2258 if (fsize
<= (pos
& PAGE_MASK
)) {
2259 size_t off
= pos
& ~PAGE_MASK
;
2261 zero_user_segment(page
, 0, off
);
2264 err
= fuse_do_readpage(file
, page
);
2278 static int fuse_write_end(struct file
*file
, struct address_space
*mapping
,
2279 loff_t pos
, unsigned len
, unsigned copied
,
2280 struct page
*page
, void *fsdata
)
2282 struct inode
*inode
= page
->mapping
->host
;
2284 /* Haven't copied anything? Skip zeroing, size extending, dirtying. */
2288 if (!PageUptodate(page
)) {
2289 /* Zero any unwritten bytes at the end of the page */
2290 size_t endoff
= (pos
+ copied
) & ~PAGE_MASK
;
2292 zero_user_segment(page
, endoff
, PAGE_SIZE
);
2293 SetPageUptodate(page
);
2296 fuse_write_update_size(inode
, pos
+ copied
);
2297 set_page_dirty(page
);
2306 static int fuse_launder_page(struct page
*page
)
2309 if (clear_page_dirty_for_io(page
)) {
2310 struct inode
*inode
= page
->mapping
->host
;
2312 /* Serialize with pending writeback for the same page */
2313 fuse_wait_on_page_writeback(inode
, page
->index
);
2314 err
= fuse_writepage_locked(page
);
2316 fuse_wait_on_page_writeback(inode
, page
->index
);
2322 * Write back dirty pages now, because there may not be any suitable
2325 static void fuse_vma_close(struct vm_area_struct
*vma
)
2327 filemap_write_and_wait(vma
->vm_file
->f_mapping
);
2331 * Wait for writeback against this page to complete before allowing it
2332 * to be marked dirty again, and hence written back again, possibly
2333 * before the previous writepage completed.
2335 * Block here, instead of in ->writepage(), so that the userspace fs
2336 * can only block processes actually operating on the filesystem.
2338 * Otherwise unprivileged userspace fs would be able to block
2343 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
2345 static vm_fault_t
fuse_page_mkwrite(struct vm_fault
*vmf
)
2347 struct page
*page
= vmf
->page
;
2348 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2350 file_update_time(vmf
->vma
->vm_file
);
2352 if (page
->mapping
!= inode
->i_mapping
) {
2354 return VM_FAULT_NOPAGE
;
2357 fuse_wait_on_page_writeback(inode
, page
->index
);
2358 return VM_FAULT_LOCKED
;
2361 static const struct vm_operations_struct fuse_file_vm_ops
= {
2362 .close
= fuse_vma_close
,
2363 .fault
= filemap_fault
,
2364 .map_pages
= filemap_map_pages
,
2365 .page_mkwrite
= fuse_page_mkwrite
,
2368 static int fuse_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2370 struct fuse_file
*ff
= file
->private_data
;
2372 /* DAX mmap is superior to direct_io mmap */
2373 if (FUSE_IS_DAX(file_inode(file
)))
2374 return fuse_dax_mmap(file
, vma
);
2376 if (ff
->open_flags
& FOPEN_DIRECT_IO
) {
2377 /* Can't provide the coherency needed for MAP_SHARED */
2378 if (vma
->vm_flags
& VM_MAYSHARE
)
2381 invalidate_inode_pages2(file
->f_mapping
);
2383 return generic_file_mmap(file
, vma
);
2386 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2387 fuse_link_write_file(file
);
2389 file_accessed(file
);
2390 vma
->vm_ops
= &fuse_file_vm_ops
;
2394 static int convert_fuse_file_lock(struct fuse_conn
*fc
,
2395 const struct fuse_file_lock
*ffl
,
2396 struct file_lock
*fl
)
2398 switch (ffl
->type
) {
2404 if (ffl
->start
> OFFSET_MAX
|| ffl
->end
> OFFSET_MAX
||
2405 ffl
->end
< ffl
->start
)
2408 fl
->fl_start
= ffl
->start
;
2409 fl
->fl_end
= ffl
->end
;
2412 * Convert pid into init's pid namespace. The locks API will
2413 * translate it into the caller's pid namespace.
2416 fl
->fl_pid
= pid_nr_ns(find_pid_ns(ffl
->pid
, fc
->pid_ns
), &init_pid_ns
);
2423 fl
->fl_type
= ffl
->type
;
2427 static void fuse_lk_fill(struct fuse_args
*args
, struct file
*file
,
2428 const struct file_lock
*fl
, int opcode
, pid_t pid
,
2429 int flock
, struct fuse_lk_in
*inarg
)
2431 struct inode
*inode
= file_inode(file
);
2432 struct fuse_conn
*fc
= get_fuse_conn(inode
);
2433 struct fuse_file
*ff
= file
->private_data
;
2435 memset(inarg
, 0, sizeof(*inarg
));
2437 inarg
->owner
= fuse_lock_owner_id(fc
, fl
->fl_owner
);
2438 inarg
->lk
.start
= fl
->fl_start
;
2439 inarg
->lk
.end
= fl
->fl_end
;
2440 inarg
->lk
.type
= fl
->fl_type
;
2441 inarg
->lk
.pid
= pid
;
2443 inarg
->lk_flags
|= FUSE_LK_FLOCK
;
2444 args
->opcode
= opcode
;
2445 args
->nodeid
= get_node_id(inode
);
2446 args
->in_numargs
= 1;
2447 args
->in_args
[0].size
= sizeof(*inarg
);
2448 args
->in_args
[0].value
= inarg
;
2451 static int fuse_getlk(struct file
*file
, struct file_lock
*fl
)
2453 struct inode
*inode
= file_inode(file
);
2454 struct fuse_mount
*fm
= get_fuse_mount(inode
);
2456 struct fuse_lk_in inarg
;
2457 struct fuse_lk_out outarg
;
2460 fuse_lk_fill(&args
, file
, fl
, FUSE_GETLK
, 0, 0, &inarg
);
2461 args
.out_numargs
= 1;
2462 args
.out_args
[0].size
= sizeof(outarg
);
2463 args
.out_args
[0].value
= &outarg
;
2464 err
= fuse_simple_request(fm
, &args
);
2466 err
= convert_fuse_file_lock(fm
->fc
, &outarg
.lk
, fl
);
2471 static int fuse_setlk(struct file
*file
, struct file_lock
*fl
, int flock
)
2473 struct inode
*inode
= file_inode(file
);
2474 struct fuse_mount
*fm
= get_fuse_mount(inode
);
2476 struct fuse_lk_in inarg
;
2477 int opcode
= (fl
->fl_flags
& FL_SLEEP
) ? FUSE_SETLKW
: FUSE_SETLK
;
2478 struct pid
*pid
= fl
->fl_type
!= F_UNLCK
? task_tgid(current
) : NULL
;
2479 pid_t pid_nr
= pid_nr_ns(pid
, fm
->fc
->pid_ns
);
2482 if (fl
->fl_lmops
&& fl
->fl_lmops
->lm_grant
) {
2483 /* NLM needs asynchronous locks, which we don't support yet */
2487 /* Unlock on close is handled by the flush method */
2488 if ((fl
->fl_flags
& FL_CLOSE_POSIX
) == FL_CLOSE_POSIX
)
2491 fuse_lk_fill(&args
, file
, fl
, opcode
, pid_nr
, flock
, &inarg
);
2492 err
= fuse_simple_request(fm
, &args
);
2494 /* locking is restartable */
2501 static int fuse_file_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
2503 struct inode
*inode
= file_inode(file
);
2504 struct fuse_conn
*fc
= get_fuse_conn(inode
);
2507 if (cmd
== F_CANCELLK
) {
2509 } else if (cmd
== F_GETLK
) {
2511 posix_test_lock(file
, fl
);
2514 err
= fuse_getlk(file
, fl
);
2517 err
= posix_lock_file(file
, fl
, NULL
);
2519 err
= fuse_setlk(file
, fl
, 0);
2524 static int fuse_file_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
2526 struct inode
*inode
= file_inode(file
);
2527 struct fuse_conn
*fc
= get_fuse_conn(inode
);
2531 err
= locks_lock_file_wait(file
, fl
);
2533 struct fuse_file
*ff
= file
->private_data
;
2535 /* emulate flock with POSIX locks */
2537 err
= fuse_setlk(file
, fl
, 1);
2543 static sector_t
fuse_bmap(struct address_space
*mapping
, sector_t block
)
2545 struct inode
*inode
= mapping
->host
;
2546 struct fuse_mount
*fm
= get_fuse_mount(inode
);
2548 struct fuse_bmap_in inarg
;
2549 struct fuse_bmap_out outarg
;
2552 if (!inode
->i_sb
->s_bdev
|| fm
->fc
->no_bmap
)
2555 memset(&inarg
, 0, sizeof(inarg
));
2556 inarg
.block
= block
;
2557 inarg
.blocksize
= inode
->i_sb
->s_blocksize
;
2558 args
.opcode
= FUSE_BMAP
;
2559 args
.nodeid
= get_node_id(inode
);
2560 args
.in_numargs
= 1;
2561 args
.in_args
[0].size
= sizeof(inarg
);
2562 args
.in_args
[0].value
= &inarg
;
2563 args
.out_numargs
= 1;
2564 args
.out_args
[0].size
= sizeof(outarg
);
2565 args
.out_args
[0].value
= &outarg
;
2566 err
= fuse_simple_request(fm
, &args
);
2568 fm
->fc
->no_bmap
= 1;
2570 return err
? 0 : outarg
.block
;
2573 static loff_t
fuse_lseek(struct file
*file
, loff_t offset
, int whence
)
2575 struct inode
*inode
= file
->f_mapping
->host
;
2576 struct fuse_mount
*fm
= get_fuse_mount(inode
);
2577 struct fuse_file
*ff
= file
->private_data
;
2579 struct fuse_lseek_in inarg
= {
2584 struct fuse_lseek_out outarg
;
2587 if (fm
->fc
->no_lseek
)
2590 args
.opcode
= FUSE_LSEEK
;
2591 args
.nodeid
= ff
->nodeid
;
2592 args
.in_numargs
= 1;
2593 args
.in_args
[0].size
= sizeof(inarg
);
2594 args
.in_args
[0].value
= &inarg
;
2595 args
.out_numargs
= 1;
2596 args
.out_args
[0].size
= sizeof(outarg
);
2597 args
.out_args
[0].value
= &outarg
;
2598 err
= fuse_simple_request(fm
, &args
);
2600 if (err
== -ENOSYS
) {
2601 fm
->fc
->no_lseek
= 1;
2607 return vfs_setpos(file
, outarg
.offset
, inode
->i_sb
->s_maxbytes
);
2610 err
= fuse_update_attributes(inode
, file
);
2612 return generic_file_llseek(file
, offset
, whence
);
2617 static loff_t
fuse_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2620 struct inode
*inode
= file_inode(file
);
2625 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
2626 retval
= generic_file_llseek(file
, offset
, whence
);
2630 retval
= fuse_update_attributes(inode
, file
);
2632 retval
= generic_file_llseek(file
, offset
, whence
);
2633 inode_unlock(inode
);
2638 retval
= fuse_lseek(file
, offset
, whence
);
2639 inode_unlock(inode
);
2649 * All files which have been polled are linked to RB tree
2650 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2651 * find the matching one.
2653 static struct rb_node
**fuse_find_polled_node(struct fuse_conn
*fc
, u64 kh
,
2654 struct rb_node
**parent_out
)
2656 struct rb_node
**link
= &fc
->polled_files
.rb_node
;
2657 struct rb_node
*last
= NULL
;
2660 struct fuse_file
*ff
;
2663 ff
= rb_entry(last
, struct fuse_file
, polled_node
);
2666 link
= &last
->rb_left
;
2667 else if (kh
> ff
->kh
)
2668 link
= &last
->rb_right
;
2679 * The file is about to be polled. Make sure it's on the polled_files
2680 * RB tree. Note that files once added to the polled_files tree are
2681 * not removed before the file is released. This is because a file
2682 * polled once is likely to be polled again.
2684 static void fuse_register_polled_file(struct fuse_conn
*fc
,
2685 struct fuse_file
*ff
)
2687 spin_lock(&fc
->lock
);
2688 if (RB_EMPTY_NODE(&ff
->polled_node
)) {
2689 struct rb_node
**link
, *parent
;
2691 link
= fuse_find_polled_node(fc
, ff
->kh
, &parent
);
2693 rb_link_node(&ff
->polled_node
, parent
, link
);
2694 rb_insert_color(&ff
->polled_node
, &fc
->polled_files
);
2696 spin_unlock(&fc
->lock
);
2699 __poll_t
fuse_file_poll(struct file
*file
, poll_table
*wait
)
2701 struct fuse_file
*ff
= file
->private_data
;
2702 struct fuse_mount
*fm
= ff
->fm
;
2703 struct fuse_poll_in inarg
= { .fh
= ff
->fh
, .kh
= ff
->kh
};
2704 struct fuse_poll_out outarg
;
2708 if (fm
->fc
->no_poll
)
2709 return DEFAULT_POLLMASK
;
2711 poll_wait(file
, &ff
->poll_wait
, wait
);
2712 inarg
.events
= mangle_poll(poll_requested_events(wait
));
2715 * Ask for notification iff there's someone waiting for it.
2716 * The client may ignore the flag and always notify.
2718 if (waitqueue_active(&ff
->poll_wait
)) {
2719 inarg
.flags
|= FUSE_POLL_SCHEDULE_NOTIFY
;
2720 fuse_register_polled_file(fm
->fc
, ff
);
2723 args
.opcode
= FUSE_POLL
;
2724 args
.nodeid
= ff
->nodeid
;
2725 args
.in_numargs
= 1;
2726 args
.in_args
[0].size
= sizeof(inarg
);
2727 args
.in_args
[0].value
= &inarg
;
2728 args
.out_numargs
= 1;
2729 args
.out_args
[0].size
= sizeof(outarg
);
2730 args
.out_args
[0].value
= &outarg
;
2731 err
= fuse_simple_request(fm
, &args
);
2734 return demangle_poll(outarg
.revents
);
2735 if (err
== -ENOSYS
) {
2736 fm
->fc
->no_poll
= 1;
2737 return DEFAULT_POLLMASK
;
2741 EXPORT_SYMBOL_GPL(fuse_file_poll
);
2744 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2745 * wakes up the poll waiters.
2747 int fuse_notify_poll_wakeup(struct fuse_conn
*fc
,
2748 struct fuse_notify_poll_wakeup_out
*outarg
)
2750 u64 kh
= outarg
->kh
;
2751 struct rb_node
**link
;
2753 spin_lock(&fc
->lock
);
2755 link
= fuse_find_polled_node(fc
, kh
, NULL
);
2757 struct fuse_file
*ff
;
2759 ff
= rb_entry(*link
, struct fuse_file
, polled_node
);
2760 wake_up_interruptible_sync(&ff
->poll_wait
);
2763 spin_unlock(&fc
->lock
);
2767 static void fuse_do_truncate(struct file
*file
)
2769 struct inode
*inode
= file
->f_mapping
->host
;
2772 attr
.ia_valid
= ATTR_SIZE
;
2773 attr
.ia_size
= i_size_read(inode
);
2775 attr
.ia_file
= file
;
2776 attr
.ia_valid
|= ATTR_FILE
;
2778 fuse_do_setattr(file_dentry(file
), &attr
, file
);
2781 static inline loff_t
fuse_round_up(struct fuse_conn
*fc
, loff_t off
)
2783 return round_up(off
, fc
->max_pages
<< PAGE_SHIFT
);
2787 fuse_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
2789 DECLARE_COMPLETION_ONSTACK(wait
);
2791 struct file
*file
= iocb
->ki_filp
;
2792 struct fuse_file
*ff
= file
->private_data
;
2794 struct inode
*inode
;
2796 size_t count
= iov_iter_count(iter
), shortened
= 0;
2797 loff_t offset
= iocb
->ki_pos
;
2798 struct fuse_io_priv
*io
;
2801 inode
= file
->f_mapping
->host
;
2802 i_size
= i_size_read(inode
);
2804 if ((iov_iter_rw(iter
) == READ
) && (offset
>= i_size
))
2807 io
= kmalloc(sizeof(struct fuse_io_priv
), GFP_KERNEL
);
2810 spin_lock_init(&io
->lock
);
2811 kref_init(&io
->refcnt
);
2815 io
->offset
= offset
;
2816 io
->write
= (iov_iter_rw(iter
) == WRITE
);
2819 * By default, we want to optimize all I/Os with async request
2820 * submission to the client filesystem if supported.
2822 io
->async
= ff
->fm
->fc
->async_dio
;
2824 io
->blocking
= is_sync_kiocb(iocb
);
2826 /* optimization for short read */
2827 if (io
->async
&& !io
->write
&& offset
+ count
> i_size
) {
2828 iov_iter_truncate(iter
, fuse_round_up(ff
->fm
->fc
, i_size
- offset
));
2829 shortened
= count
- iov_iter_count(iter
);
2834 * We cannot asynchronously extend the size of a file.
2835 * In such case the aio will behave exactly like sync io.
2837 if ((offset
+ count
> i_size
) && io
->write
)
2838 io
->blocking
= true;
2840 if (io
->async
&& io
->blocking
) {
2842 * Additional reference to keep io around after
2843 * calling fuse_aio_complete()
2845 kref_get(&io
->refcnt
);
2849 if (iov_iter_rw(iter
) == WRITE
) {
2850 ret
= fuse_direct_io(io
, iter
, &pos
, FUSE_DIO_WRITE
);
2851 fuse_invalidate_attr(inode
);
2853 ret
= __fuse_direct_read(io
, iter
, &pos
);
2855 iov_iter_reexpand(iter
, iov_iter_count(iter
) + shortened
);
2858 bool blocking
= io
->blocking
;
2860 fuse_aio_complete(io
, ret
< 0 ? ret
: 0, -1);
2862 /* we have a non-extending, async request, so return */
2864 return -EIOCBQUEUED
;
2866 wait_for_completion(&wait
);
2867 ret
= fuse_get_res_by_io(io
);
2870 kref_put(&io
->refcnt
, fuse_io_release
);
2872 if (iov_iter_rw(iter
) == WRITE
) {
2874 fuse_write_update_size(inode
, pos
);
2875 else if (ret
< 0 && offset
+ count
> i_size
)
2876 fuse_do_truncate(file
);
2882 static int fuse_writeback_range(struct inode
*inode
, loff_t start
, loff_t end
)
2884 int err
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
2887 fuse_sync_writes(inode
);
2892 static long fuse_file_fallocate(struct file
*file
, int mode
, loff_t offset
,
2895 struct fuse_file
*ff
= file
->private_data
;
2896 struct inode
*inode
= file_inode(file
);
2897 struct fuse_inode
*fi
= get_fuse_inode(inode
);
2898 struct fuse_mount
*fm
= ff
->fm
;
2900 struct fuse_fallocate_in inarg
= {
2907 bool lock_inode
= !(mode
& FALLOC_FL_KEEP_SIZE
) ||
2908 (mode
& (FALLOC_FL_PUNCH_HOLE
|
2909 FALLOC_FL_ZERO_RANGE
));
2911 bool block_faults
= FUSE_IS_DAX(inode
) && lock_inode
;
2913 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
|
2914 FALLOC_FL_ZERO_RANGE
))
2917 if (fm
->fc
->no_fallocate
)
2923 filemap_invalidate_lock(inode
->i_mapping
);
2924 err
= fuse_dax_break_layouts(inode
, 0, 0);
2929 if (mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
)) {
2930 loff_t endbyte
= offset
+ length
- 1;
2932 err
= fuse_writeback_range(inode
, offset
, endbyte
);
2938 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
2939 offset
+ length
> i_size_read(inode
)) {
2940 err
= inode_newsize_ok(inode
, offset
+ length
);
2945 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
2946 set_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
2948 args
.opcode
= FUSE_FALLOCATE
;
2949 args
.nodeid
= ff
->nodeid
;
2950 args
.in_numargs
= 1;
2951 args
.in_args
[0].size
= sizeof(inarg
);
2952 args
.in_args
[0].value
= &inarg
;
2953 err
= fuse_simple_request(fm
, &args
);
2954 if (err
== -ENOSYS
) {
2955 fm
->fc
->no_fallocate
= 1;
2961 /* we could have extended the file */
2962 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
2963 bool changed
= fuse_write_update_size(inode
, offset
+ length
);
2965 if (changed
&& fm
->fc
->writeback_cache
)
2966 file_update_time(file
);
2969 if (mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
))
2970 truncate_pagecache_range(inode
, offset
, offset
+ length
- 1);
2972 fuse_invalidate_attr(inode
);
2975 if (!(mode
& FALLOC_FL_KEEP_SIZE
))
2976 clear_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
);
2979 filemap_invalidate_unlock(inode
->i_mapping
);
2982 inode_unlock(inode
);
2987 static ssize_t
__fuse_copy_file_range(struct file
*file_in
, loff_t pos_in
,
2988 struct file
*file_out
, loff_t pos_out
,
2989 size_t len
, unsigned int flags
)
2991 struct fuse_file
*ff_in
= file_in
->private_data
;
2992 struct fuse_file
*ff_out
= file_out
->private_data
;
2993 struct inode
*inode_in
= file_inode(file_in
);
2994 struct inode
*inode_out
= file_inode(file_out
);
2995 struct fuse_inode
*fi_out
= get_fuse_inode(inode_out
);
2996 struct fuse_mount
*fm
= ff_in
->fm
;
2997 struct fuse_conn
*fc
= fm
->fc
;
2999 struct fuse_copy_file_range_in inarg
= {
3002 .nodeid_out
= ff_out
->nodeid
,
3003 .fh_out
= ff_out
->fh
,
3008 struct fuse_write_out outarg
;
3010 /* mark unstable when write-back is not used, and file_out gets
3012 bool is_unstable
= (!fc
->writeback_cache
) &&
3013 ((pos_out
+ len
) > inode_out
->i_size
);
3015 if (fc
->no_copy_file_range
)
3018 if (file_inode(file_in
)->i_sb
!= file_inode(file_out
)->i_sb
)
3021 inode_lock(inode_in
);
3022 err
= fuse_writeback_range(inode_in
, pos_in
, pos_in
+ len
- 1);
3023 inode_unlock(inode_in
);
3027 inode_lock(inode_out
);
3029 err
= file_modified(file_out
);
3034 * Write out dirty pages in the destination file before sending the COPY
3035 * request to userspace. After the request is completed, truncate off
3036 * pages (including partial ones) from the cache that have been copied,
3037 * since these contain stale data at that point.
3039 * This should be mostly correct, but if the COPY writes to partial
3040 * pages (at the start or end) and the parts not covered by the COPY are
3041 * written through a memory map after calling fuse_writeback_range(),
3042 * then these partial page modifications will be lost on truncation.
3044 * It is unlikely that someone would rely on such mixed style
3045 * modifications. Yet this does give less guarantees than if the
3046 * copying was performed with write(2).
3048 * To fix this a mapping->invalidate_lock could be used to prevent new
3049 * faults while the copy is ongoing.
3051 err
= fuse_writeback_range(inode_out
, pos_out
, pos_out
+ len
- 1);
3056 set_bit(FUSE_I_SIZE_UNSTABLE
, &fi_out
->state
);
3058 args
.opcode
= FUSE_COPY_FILE_RANGE
;
3059 args
.nodeid
= ff_in
->nodeid
;
3060 args
.in_numargs
= 1;
3061 args
.in_args
[0].size
= sizeof(inarg
);
3062 args
.in_args
[0].value
= &inarg
;
3063 args
.out_numargs
= 1;
3064 args
.out_args
[0].size
= sizeof(outarg
);
3065 args
.out_args
[0].value
= &outarg
;
3066 err
= fuse_simple_request(fm
, &args
);
3067 if (err
== -ENOSYS
) {
3068 fc
->no_copy_file_range
= 1;
3074 truncate_inode_pages_range(inode_out
->i_mapping
,
3075 ALIGN_DOWN(pos_out
, PAGE_SIZE
),
3076 ALIGN(pos_out
+ outarg
.size
, PAGE_SIZE
) - 1);
3078 if (fc
->writeback_cache
) {
3079 fuse_write_update_size(inode_out
, pos_out
+ outarg
.size
);
3080 file_update_time(file_out
);
3083 fuse_invalidate_attr(inode_out
);
3088 clear_bit(FUSE_I_SIZE_UNSTABLE
, &fi_out
->state
);
3090 inode_unlock(inode_out
);
3091 file_accessed(file_in
);
3096 static ssize_t
fuse_copy_file_range(struct file
*src_file
, loff_t src_off
,
3097 struct file
*dst_file
, loff_t dst_off
,
3098 size_t len
, unsigned int flags
)
3102 ret
= __fuse_copy_file_range(src_file
, src_off
, dst_file
, dst_off
,
3105 if (ret
== -EOPNOTSUPP
|| ret
== -EXDEV
)
3106 ret
= generic_copy_file_range(src_file
, src_off
, dst_file
,
3107 dst_off
, len
, flags
);
3111 static const struct file_operations fuse_file_operations
= {
3112 .llseek
= fuse_file_llseek
,
3113 .read_iter
= fuse_file_read_iter
,
3114 .write_iter
= fuse_file_write_iter
,
3115 .mmap
= fuse_file_mmap
,
3117 .flush
= fuse_flush
,
3118 .release
= fuse_release
,
3119 .fsync
= fuse_fsync
,
3120 .lock
= fuse_file_lock
,
3121 .get_unmapped_area
= thp_get_unmapped_area
,
3122 .flock
= fuse_file_flock
,
3123 .splice_read
= generic_file_splice_read
,
3124 .splice_write
= iter_file_splice_write
,
3125 .unlocked_ioctl
= fuse_file_ioctl
,
3126 .compat_ioctl
= fuse_file_compat_ioctl
,
3127 .poll
= fuse_file_poll
,
3128 .fallocate
= fuse_file_fallocate
,
3129 .copy_file_range
= fuse_copy_file_range
,
3132 static const struct address_space_operations fuse_file_aops
= {
3133 .readpage
= fuse_readpage
,
3134 .readahead
= fuse_readahead
,
3135 .writepage
= fuse_writepage
,
3136 .writepages
= fuse_writepages
,
3137 .launder_page
= fuse_launder_page
,
3138 .set_page_dirty
= __set_page_dirty_nobuffers
,
3140 .direct_IO
= fuse_direct_IO
,
3141 .write_begin
= fuse_write_begin
,
3142 .write_end
= fuse_write_end
,
3145 void fuse_init_file_inode(struct inode
*inode
)
3147 struct fuse_inode
*fi
= get_fuse_inode(inode
);
3149 inode
->i_fop
= &fuse_file_operations
;
3150 inode
->i_data
.a_ops
= &fuse_file_aops
;
3152 INIT_LIST_HEAD(&fi
->write_files
);
3153 INIT_LIST_HEAD(&fi
->queued_writes
);
3155 init_waitqueue_head(&fi
->page_waitq
);
3156 fi
->writepages
= RB_ROOT
;
3158 if (IS_ENABLED(CONFIG_FUSE_DAX
))
3159 fuse_dax_inode_init(inode
);