2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
34 #include "xfs_dir2_priv.h"
35 #include "xfs_ioctl.h"
36 #include "xfs_trace.h"
38 #include "xfs_icache.h"
40 #include "xfs_iomap.h"
41 #include "xfs_reflink.h"
43 #include <linux/dcache.h>
44 #include <linux/falloc.h>
45 #include <linux/pagevec.h>
46 #include <linux/backing-dev.h>
48 static const struct vm_operations_struct xfs_file_vm_ops
;
51 * Locking primitives for read and write IO paths to ensure we consistently use
52 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
59 if (type
& XFS_IOLOCK_EXCL
)
60 inode_lock(VFS_I(ip
));
69 xfs_iunlock(ip
, type
);
70 if (type
& XFS_IOLOCK_EXCL
)
71 inode_unlock(VFS_I(ip
));
79 xfs_ilock_demote(ip
, type
);
80 if (type
& XFS_IOLOCK_EXCL
)
81 inode_unlock(VFS_I(ip
));
85 * Clear the specified ranges to zero through either the pagecache or DAX.
86 * Holes and unwritten extents will be left as-is as they already are zeroed.
95 return iomap_zero_range(VFS_I(ip
), pos
, count
, NULL
, &xfs_iomap_ops
);
99 xfs_update_prealloc_flags(
100 struct xfs_inode
*ip
,
101 enum xfs_prealloc_flags flags
)
103 struct xfs_trans
*tp
;
106 error
= xfs_trans_alloc(ip
->i_mount
, &M_RES(ip
->i_mount
)->tr_writeid
,
111 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
112 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
114 if (!(flags
& XFS_PREALLOC_INVISIBLE
)) {
115 VFS_I(ip
)->i_mode
&= ~S_ISUID
;
116 if (VFS_I(ip
)->i_mode
& S_IXGRP
)
117 VFS_I(ip
)->i_mode
&= ~S_ISGID
;
118 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
121 if (flags
& XFS_PREALLOC_SET
)
122 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
123 if (flags
& XFS_PREALLOC_CLEAR
)
124 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
126 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
127 if (flags
& XFS_PREALLOC_SYNC
)
128 xfs_trans_set_sync(tp
);
129 return xfs_trans_commit(tp
);
133 * Fsync operations on directories are much simpler than on regular files,
134 * as there is no file data to flush, and thus also no need for explicit
135 * cache flush operations, and there are no non-transaction metadata updates
136 * on directories either.
145 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
146 struct xfs_mount
*mp
= ip
->i_mount
;
149 trace_xfs_dir_fsync(ip
);
151 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
152 if (xfs_ipincount(ip
))
153 lsn
= ip
->i_itemp
->ili_last_lsn
;
154 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
158 return _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, NULL
);
168 struct inode
*inode
= file
->f_mapping
->host
;
169 struct xfs_inode
*ip
= XFS_I(inode
);
170 struct xfs_mount
*mp
= ip
->i_mount
;
175 trace_xfs_file_fsync(ip
);
177 error
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
181 if (XFS_FORCED_SHUTDOWN(mp
))
184 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
186 if (mp
->m_flags
& XFS_MOUNT_BARRIER
) {
188 * If we have an RT and/or log subvolume we need to make sure
189 * to flush the write cache the device used for file data
190 * first. This is to ensure newly written file data make
191 * it to disk before logging the new inode size in case of
192 * an extending write.
194 if (XFS_IS_REALTIME_INODE(ip
))
195 xfs_blkdev_issue_flush(mp
->m_rtdev_targp
);
196 else if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
197 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
201 * All metadata updates are logged, which means that we just have to
202 * flush the log up to the latest LSN that touched the inode. If we have
203 * concurrent fsync/fdatasync() calls, we need them to all block on the
204 * log force before we clear the ili_fsync_fields field. This ensures
205 * that we don't get a racing sync operation that does not wait for the
206 * metadata to hit the journal before returning. If we race with
207 * clearing the ili_fsync_fields, then all that will happen is the log
208 * force will do nothing as the lsn will already be on disk. We can't
209 * race with setting ili_fsync_fields because that is done under
210 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
211 * until after the ili_fsync_fields is cleared.
213 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
214 if (xfs_ipincount(ip
)) {
216 (ip
->i_itemp
->ili_fsync_fields
& ~XFS_ILOG_TIMESTAMP
))
217 lsn
= ip
->i_itemp
->ili_last_lsn
;
221 error
= _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, &log_flushed
);
222 ip
->i_itemp
->ili_fsync_fields
= 0;
224 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
227 * If we only have a single device, and the log force about was
228 * a no-op we might have to flush the data device cache here.
229 * This can only happen for fdatasync/O_DSYNC if we were overwriting
230 * an already allocated file and thus do not have any metadata to
233 if ((mp
->m_flags
& XFS_MOUNT_BARRIER
) &&
234 mp
->m_logdev_targp
== mp
->m_ddev_targp
&&
235 !XFS_IS_REALTIME_INODE(ip
) &&
237 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
243 xfs_file_dio_aio_read(
247 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
248 struct inode
*inode
= mapping
->host
;
249 struct xfs_inode
*ip
= XFS_I(inode
);
250 loff_t isize
= i_size_read(inode
);
251 size_t count
= iov_iter_count(to
);
252 struct iov_iter data
;
253 struct xfs_buftarg
*target
;
256 trace_xfs_file_direct_read(ip
, count
, iocb
->ki_pos
);
259 return 0; /* skip atime */
261 if (XFS_IS_REALTIME_INODE(ip
))
262 target
= ip
->i_mount
->m_rtdev_targp
;
264 target
= ip
->i_mount
->m_ddev_targp
;
266 /* DIO must be aligned to device logical sector size */
267 if ((iocb
->ki_pos
| count
) & target
->bt_logical_sectormask
) {
268 if (iocb
->ki_pos
== isize
)
273 file_accessed(iocb
->ki_filp
);
276 * Locking is a bit tricky here. If we take an exclusive lock for direct
277 * IO, we effectively serialise all new concurrent read IO to this file
278 * and block it behind IO that is currently in progress because IO in
279 * progress holds the IO lock shared. We only need to hold the lock
280 * exclusive to blow away the page cache, so only take lock exclusively
281 * if the page cache needs invalidation. This allows the normal direct
282 * IO case of no page cache pages to proceeed concurrently without
285 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
286 if (mapping
->nrpages
) {
287 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
288 xfs_rw_ilock(ip
, XFS_IOLOCK_EXCL
);
291 * The generic dio code only flushes the range of the particular
292 * I/O. Because we take an exclusive lock here, this whole
293 * sequence is considerably more expensive for us. This has a
294 * noticeable performance impact for any file with cached pages,
295 * even when outside of the range of the particular I/O.
297 * Hence, amortize the cost of the lock against a full file
298 * flush and reduce the chances of repeated iolock cycles going
301 if (mapping
->nrpages
) {
302 ret
= filemap_write_and_wait(mapping
);
304 xfs_rw_iunlock(ip
, XFS_IOLOCK_EXCL
);
309 * Invalidate whole pages. This can return an error if
310 * we fail to invalidate a page, but this should never
311 * happen on XFS. Warn if it does fail.
313 ret
= invalidate_inode_pages2(mapping
);
317 xfs_rw_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
321 ret
= __blockdev_direct_IO(iocb
, inode
, target
->bt_bdev
, &data
,
322 xfs_get_blocks_direct
, NULL
, NULL
, 0);
325 iov_iter_advance(to
, ret
);
327 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
332 static noinline ssize_t
337 struct xfs_inode
*ip
= XFS_I(iocb
->ki_filp
->f_mapping
->host
);
338 size_t count
= iov_iter_count(to
);
341 trace_xfs_file_dax_read(ip
, count
, iocb
->ki_pos
);
344 return 0; /* skip atime */
346 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
347 ret
= iomap_dax_rw(iocb
, to
, &xfs_iomap_ops
);
348 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
350 file_accessed(iocb
->ki_filp
);
355 xfs_file_buffered_aio_read(
359 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
362 trace_xfs_file_buffered_read(ip
, iov_iter_count(to
), iocb
->ki_pos
);
364 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
365 ret
= generic_file_read_iter(iocb
, to
);
366 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
376 struct inode
*inode
= file_inode(iocb
->ki_filp
);
377 struct xfs_mount
*mp
= XFS_I(inode
)->i_mount
;
380 XFS_STATS_INC(mp
, xs_read_calls
);
382 if (XFS_FORCED_SHUTDOWN(mp
))
386 ret
= xfs_file_dax_read(iocb
, to
);
387 else if (iocb
->ki_flags
& IOCB_DIRECT
)
388 ret
= xfs_file_dio_aio_read(iocb
, to
);
390 ret
= xfs_file_buffered_aio_read(iocb
, to
);
393 XFS_STATS_ADD(mp
, xs_read_bytes
, ret
);
398 xfs_file_splice_read(
401 struct pipe_inode_info
*pipe
,
405 struct xfs_inode
*ip
= XFS_I(infilp
->f_mapping
->host
);
408 XFS_STATS_INC(ip
->i_mount
, xs_read_calls
);
410 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
413 trace_xfs_file_splice_read(ip
, count
, *ppos
);
416 * DAX inodes cannot ues the page cache for splice, so we have to push
417 * them through the VFS IO path. This means it goes through
418 * ->read_iter, which for us takes the XFS_IOLOCK_SHARED. Hence we
419 * cannot lock the splice operation at this level for DAX inodes.
421 if (IS_DAX(VFS_I(ip
))) {
422 ret
= default_file_splice_read(infilp
, ppos
, pipe
, count
,
427 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
428 ret
= generic_file_splice_read(infilp
, ppos
, pipe
, count
, flags
);
429 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
432 XFS_STATS_ADD(ip
->i_mount
, xs_read_bytes
, ret
);
437 * Zero any on disk space between the current EOF and the new, larger EOF.
439 * This handles the normal case of zeroing the remainder of the last block in
440 * the file and the unusual case of zeroing blocks out beyond the size of the
441 * file. This second case only happens with fixed size extents and when the
442 * system crashes before the inode size was updated but after blocks were
445 * Expects the iolock to be held exclusive, and will take the ilock internally.
447 int /* error (positive) */
449 struct xfs_inode
*ip
,
450 xfs_off_t offset
, /* starting I/O offset */
451 xfs_fsize_t isize
, /* current inode size */
454 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
455 ASSERT(offset
> isize
);
457 trace_xfs_zero_eof(ip
, isize
, offset
- isize
);
458 return xfs_zero_range(ip
, isize
, offset
- isize
, did_zeroing
);
462 * Common pre-write limit and setup checks.
464 * Called with the iolocked held either shared and exclusive according to
465 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
466 * if called for a direct write beyond i_size.
469 xfs_file_aio_write_checks(
471 struct iov_iter
*from
,
474 struct file
*file
= iocb
->ki_filp
;
475 struct inode
*inode
= file
->f_mapping
->host
;
476 struct xfs_inode
*ip
= XFS_I(inode
);
478 size_t count
= iov_iter_count(from
);
479 bool drained_dio
= false;
482 error
= generic_write_checks(iocb
, from
);
486 error
= xfs_break_layouts(inode
, iolock
, true);
490 /* For changing security info in file_remove_privs() we need i_mutex */
491 if (*iolock
== XFS_IOLOCK_SHARED
&& !IS_NOSEC(inode
)) {
492 xfs_rw_iunlock(ip
, *iolock
);
493 *iolock
= XFS_IOLOCK_EXCL
;
494 xfs_rw_ilock(ip
, *iolock
);
498 * If the offset is beyond the size of the file, we need to zero any
499 * blocks that fall between the existing EOF and the start of this
500 * write. If zeroing is needed and we are currently holding the
501 * iolock shared, we need to update it to exclusive which implies
502 * having to redo all checks before.
504 * We need to serialise against EOF updates that occur in IO
505 * completions here. We want to make sure that nobody is changing the
506 * size while we do this check until we have placed an IO barrier (i.e.
507 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
508 * The spinlock effectively forms a memory barrier once we have the
509 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
510 * and hence be able to correctly determine if we need to run zeroing.
512 spin_lock(&ip
->i_flags_lock
);
513 if (iocb
->ki_pos
> i_size_read(inode
)) {
516 spin_unlock(&ip
->i_flags_lock
);
518 if (*iolock
== XFS_IOLOCK_SHARED
) {
519 xfs_rw_iunlock(ip
, *iolock
);
520 *iolock
= XFS_IOLOCK_EXCL
;
521 xfs_rw_ilock(ip
, *iolock
);
522 iov_iter_reexpand(from
, count
);
525 * We now have an IO submission barrier in place, but
526 * AIO can do EOF updates during IO completion and hence
527 * we now need to wait for all of them to drain. Non-AIO
528 * DIO will have drained before we are given the
529 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
532 inode_dio_wait(inode
);
536 error
= xfs_zero_eof(ip
, iocb
->ki_pos
, i_size_read(inode
), &zero
);
540 spin_unlock(&ip
->i_flags_lock
);
543 * Updating the timestamps will grab the ilock again from
544 * xfs_fs_dirty_inode, so we have to call it after dropping the
545 * lock above. Eventually we should look into a way to avoid
546 * the pointless lock roundtrip.
548 if (likely(!(file
->f_mode
& FMODE_NOCMTIME
))) {
549 error
= file_update_time(file
);
555 * If we're writing the file then make sure to clear the setuid and
556 * setgid bits if the process is not being run by root. This keeps
557 * people from modifying setuid and setgid binaries.
559 if (!IS_NOSEC(inode
))
560 return file_remove_privs(file
);
565 * xfs_file_dio_aio_write - handle direct IO writes
567 * Lock the inode appropriately to prepare for and issue a direct IO write.
568 * By separating it from the buffered write path we remove all the tricky to
569 * follow locking changes and looping.
571 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
572 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
573 * pages are flushed out.
575 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
576 * allowing them to be done in parallel with reads and other direct IO writes.
577 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
578 * needs to do sub-block zeroing and that requires serialisation against other
579 * direct IOs to the same block. In this case we need to serialise the
580 * submission of the unaligned IOs so that we don't get racing block zeroing in
581 * the dio layer. To avoid the problem with aio, we also need to wait for
582 * outstanding IOs to complete so that unwritten extent conversion is completed
583 * before we try to map the overlapping block. This is currently implemented by
584 * hitting it with a big hammer (i.e. inode_dio_wait()).
586 * Returns with locks held indicated by @iolock and errors indicated by
587 * negative return values.
590 xfs_file_dio_aio_write(
592 struct iov_iter
*from
)
594 struct file
*file
= iocb
->ki_filp
;
595 struct address_space
*mapping
= file
->f_mapping
;
596 struct inode
*inode
= mapping
->host
;
597 struct xfs_inode
*ip
= XFS_I(inode
);
598 struct xfs_mount
*mp
= ip
->i_mount
;
600 int unaligned_io
= 0;
602 size_t count
= iov_iter_count(from
);
604 struct iov_iter data
;
605 struct xfs_buftarg
*target
= XFS_IS_REALTIME_INODE(ip
) ?
606 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
608 /* DIO must be aligned to device logical sector size */
609 if ((iocb
->ki_pos
| count
) & target
->bt_logical_sectormask
)
612 /* "unaligned" here means not aligned to a filesystem block */
613 if ((iocb
->ki_pos
& mp
->m_blockmask
) ||
614 ((iocb
->ki_pos
+ count
) & mp
->m_blockmask
))
618 * We don't need to take an exclusive lock unless there page cache needs
619 * to be invalidated or unaligned IO is being executed. We don't need to
620 * consider the EOF extension case here because
621 * xfs_file_aio_write_checks() will relock the inode as necessary for
622 * EOF zeroing cases and fill out the new inode size as appropriate.
624 if (unaligned_io
|| mapping
->nrpages
)
625 iolock
= XFS_IOLOCK_EXCL
;
627 iolock
= XFS_IOLOCK_SHARED
;
628 xfs_rw_ilock(ip
, iolock
);
631 * Recheck if there are cached pages that need invalidate after we got
632 * the iolock to protect against other threads adding new pages while
633 * we were waiting for the iolock.
635 if (mapping
->nrpages
&& iolock
== XFS_IOLOCK_SHARED
) {
636 xfs_rw_iunlock(ip
, iolock
);
637 iolock
= XFS_IOLOCK_EXCL
;
638 xfs_rw_ilock(ip
, iolock
);
641 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
644 count
= iov_iter_count(from
);
645 end
= iocb
->ki_pos
+ count
- 1;
648 * See xfs_file_dio_aio_read() for why we do a full-file flush here.
650 if (mapping
->nrpages
) {
651 ret
= filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
655 * Invalidate whole pages. This can return an error if we fail
656 * to invalidate a page, but this should never happen on XFS.
657 * Warn if it does fail.
659 ret
= invalidate_inode_pages2(VFS_I(ip
)->i_mapping
);
665 * If we are doing unaligned IO, wait for all other IO to drain,
666 * otherwise demote the lock if we had to flush cached pages
669 inode_dio_wait(inode
);
670 else if (iolock
== XFS_IOLOCK_EXCL
) {
671 xfs_rw_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
672 iolock
= XFS_IOLOCK_SHARED
;
675 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
);
677 /* If this is a block-aligned directio CoW, remap immediately. */
678 if (xfs_is_reflink_inode(ip
) && !unaligned_io
) {
679 ret
= xfs_reflink_allocate_cow_range(ip
, iocb
->ki_pos
, count
);
685 ret
= __blockdev_direct_IO(iocb
, inode
, target
->bt_bdev
, &data
,
686 xfs_get_blocks_direct
, xfs_end_io_direct_write
,
687 NULL
, DIO_ASYNC_EXTEND
);
689 /* see generic_file_direct_write() for why this is necessary */
690 if (mapping
->nrpages
) {
691 invalidate_inode_pages2_range(mapping
,
692 iocb
->ki_pos
>> PAGE_SHIFT
,
698 iov_iter_advance(from
, ret
);
701 xfs_rw_iunlock(ip
, iolock
);
704 * No fallback to buffered IO on errors for XFS, direct IO will either
705 * complete fully or fail.
707 ASSERT(ret
< 0 || ret
== count
);
711 static noinline ssize_t
714 struct iov_iter
*from
)
716 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
717 struct xfs_inode
*ip
= XFS_I(inode
);
718 int iolock
= XFS_IOLOCK_EXCL
;
719 ssize_t ret
, error
= 0;
723 xfs_rw_ilock(ip
, iolock
);
724 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
729 count
= iov_iter_count(from
);
731 trace_xfs_file_dax_write(ip
, count
, pos
);
733 ret
= iomap_dax_rw(iocb
, from
, &xfs_iomap_ops
);
734 if (ret
> 0 && iocb
->ki_pos
> i_size_read(inode
)) {
735 i_size_write(inode
, iocb
->ki_pos
);
736 error
= xfs_setfilesize(ip
, pos
, ret
);
740 xfs_rw_iunlock(ip
, iolock
);
741 return error
? error
: ret
;
745 xfs_file_buffered_aio_write(
747 struct iov_iter
*from
)
749 struct file
*file
= iocb
->ki_filp
;
750 struct address_space
*mapping
= file
->f_mapping
;
751 struct inode
*inode
= mapping
->host
;
752 struct xfs_inode
*ip
= XFS_I(inode
);
755 int iolock
= XFS_IOLOCK_EXCL
;
757 xfs_rw_ilock(ip
, iolock
);
759 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
763 /* We can write back this queue in page reclaim */
764 current
->backing_dev_info
= inode_to_bdi(inode
);
767 trace_xfs_file_buffered_write(ip
, iov_iter_count(from
), iocb
->ki_pos
);
768 ret
= iomap_file_buffered_write(iocb
, from
, &xfs_iomap_ops
);
769 if (likely(ret
>= 0))
773 * If we hit a space limit, try to free up some lingering preallocated
774 * space before returning an error. In the case of ENOSPC, first try to
775 * write back all dirty inodes to free up some of the excess reserved
776 * metadata space. This reduces the chances that the eofblocks scan
777 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
778 * also behaves as a filter to prevent too many eofblocks scans from
779 * running at the same time.
781 if (ret
== -EDQUOT
&& !enospc
) {
782 enospc
= xfs_inode_free_quota_eofblocks(ip
);
785 } else if (ret
== -ENOSPC
&& !enospc
) {
786 struct xfs_eofblocks eofb
= {0};
789 xfs_flush_inodes(ip
->i_mount
);
790 eofb
.eof_scan_owner
= ip
->i_ino
; /* for locking */
791 eofb
.eof_flags
= XFS_EOF_FLAGS_SYNC
;
792 xfs_icache_free_eofblocks(ip
->i_mount
, &eofb
);
796 current
->backing_dev_info
= NULL
;
798 xfs_rw_iunlock(ip
, iolock
);
805 struct iov_iter
*from
)
807 struct file
*file
= iocb
->ki_filp
;
808 struct address_space
*mapping
= file
->f_mapping
;
809 struct inode
*inode
= mapping
->host
;
810 struct xfs_inode
*ip
= XFS_I(inode
);
812 size_t ocount
= iov_iter_count(from
);
814 XFS_STATS_INC(ip
->i_mount
, xs_write_calls
);
819 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
823 ret
= xfs_file_dax_write(iocb
, from
);
824 else if (iocb
->ki_flags
& IOCB_DIRECT
) {
826 * Allow a directio write to fall back to a buffered
827 * write *only* in the case that we're doing a reflink
828 * CoW. In all other directio scenarios we do not
829 * allow an operation to fall back to buffered mode.
831 ret
= xfs_file_dio_aio_write(iocb
, from
);
836 ret
= xfs_file_buffered_aio_write(iocb
, from
);
840 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
842 /* Handle various SYNC-type writes */
843 ret
= generic_write_sync(iocb
, ret
);
848 #define XFS_FALLOC_FL_SUPPORTED \
849 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
850 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
851 FALLOC_FL_INSERT_RANGE)
860 struct inode
*inode
= file_inode(file
);
861 struct xfs_inode
*ip
= XFS_I(inode
);
863 enum xfs_prealloc_flags flags
= 0;
864 uint iolock
= XFS_IOLOCK_EXCL
;
866 bool do_file_insert
= 0;
868 if (!S_ISREG(inode
->i_mode
))
870 if (mode
& ~XFS_FALLOC_FL_SUPPORTED
)
873 xfs_ilock(ip
, iolock
);
874 error
= xfs_break_layouts(inode
, &iolock
, false);
878 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
879 iolock
|= XFS_MMAPLOCK_EXCL
;
881 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
882 error
= xfs_free_file_space(ip
, offset
, len
);
885 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
886 unsigned blksize_mask
= (1 << inode
->i_blkbits
) - 1;
888 if (offset
& blksize_mask
|| len
& blksize_mask
) {
894 * There is no need to overlap collapse range with EOF,
895 * in which case it is effectively a truncate operation
897 if (offset
+ len
>= i_size_read(inode
)) {
902 new_size
= i_size_read(inode
) - len
;
904 error
= xfs_collapse_file_space(ip
, offset
, len
);
907 } else if (mode
& FALLOC_FL_INSERT_RANGE
) {
908 unsigned blksize_mask
= (1 << inode
->i_blkbits
) - 1;
910 new_size
= i_size_read(inode
) + len
;
911 if (offset
& blksize_mask
|| len
& blksize_mask
) {
916 /* check the new inode size does not wrap through zero */
917 if (new_size
> inode
->i_sb
->s_maxbytes
) {
922 /* Offset should be less than i_size */
923 if (offset
>= i_size_read(inode
)) {
929 flags
|= XFS_PREALLOC_SET
;
931 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
932 offset
+ len
> i_size_read(inode
)) {
933 new_size
= offset
+ len
;
934 error
= inode_newsize_ok(inode
, new_size
);
939 if (mode
& FALLOC_FL_ZERO_RANGE
)
940 error
= xfs_zero_file_space(ip
, offset
, len
);
942 error
= xfs_alloc_file_space(ip
, offset
, len
,
948 if (file
->f_flags
& O_DSYNC
)
949 flags
|= XFS_PREALLOC_SYNC
;
951 error
= xfs_update_prealloc_flags(ip
, flags
);
955 /* Change file size if needed */
959 iattr
.ia_valid
= ATTR_SIZE
;
960 iattr
.ia_size
= new_size
;
961 error
= xfs_setattr_size(ip
, &iattr
);
967 * Perform hole insertion now that the file size has been
968 * updated so that if we crash during the operation we don't
969 * leave shifted extents past EOF and hence losing access to
970 * the data that is contained within them.
973 error
= xfs_insert_file_space(ip
, offset
, len
);
976 xfs_iunlock(ip
, iolock
);
981 * Flush all file writes out to disk.
984 xfs_file_wait_for_io(
995 bs
= inode
->i_sb
->s_blocksize
;
996 inode_dio_wait(inode
);
998 rounding
= max_t(xfs_off_t
, bs
, PAGE_SIZE
);
999 ioffset
= round_down(offset
, rounding
);
1000 iendoffset
= round_up(offset
+ len
, rounding
) - 1;
1001 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
,
1006 /* Hook up to the VFS reflink function */
1008 xfs_file_share_range(
1009 struct file
*file_in
,
1011 struct file
*file_out
,
1015 struct inode
*inode_in
;
1016 struct inode
*inode_out
;
1023 inode_in
= file_inode(file_in
);
1024 inode_out
= file_inode(file_out
);
1025 bs
= inode_out
->i_sb
->s_blocksize
;
1027 /* Don't touch certain kinds of inodes */
1028 if (IS_IMMUTABLE(inode_out
))
1030 if (IS_SWAPFILE(inode_in
) ||
1031 IS_SWAPFILE(inode_out
))
1034 /* Reflink only works within this filesystem. */
1035 if (inode_in
->i_sb
!= inode_out
->i_sb
)
1037 same_inode
= (inode_in
->i_ino
== inode_out
->i_ino
);
1039 /* Don't reflink dirs, pipes, sockets... */
1040 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
1042 if (S_ISFIFO(inode_in
->i_mode
) || S_ISFIFO(inode_out
->i_mode
))
1044 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
1047 /* Are we going all the way to the end? */
1048 isize
= i_size_read(inode_in
);
1052 len
= isize
- pos_in
;
1054 /* Ensure offsets don't wrap and the input is inside i_size */
1055 if (pos_in
+ len
< pos_in
|| pos_out
+ len
< pos_out
||
1056 pos_in
+ len
> isize
)
1059 /* If we're linking to EOF, continue to the block boundary. */
1060 if (pos_in
+ len
== isize
)
1061 blen
= ALIGN(isize
, bs
) - pos_in
;
1065 /* Only reflink if we're aligned to block boundaries */
1066 if (!IS_ALIGNED(pos_in
, bs
) || !IS_ALIGNED(pos_in
+ blen
, bs
) ||
1067 !IS_ALIGNED(pos_out
, bs
) || !IS_ALIGNED(pos_out
+ blen
, bs
))
1070 /* Don't allow overlapped reflink within the same file */
1071 if (same_inode
&& pos_out
+ blen
> pos_in
&& pos_out
< pos_in
+ blen
)
1074 /* Wait for the completion of any pending IOs on srcfile */
1075 ret
= xfs_file_wait_for_io(inode_in
, pos_in
, len
);
1078 ret
= xfs_file_wait_for_io(inode_out
, pos_out
, len
);
1082 ret
= xfs_reflink_remap_range(XFS_I(inode_in
), pos_in
, XFS_I(inode_out
),
1092 xfs_file_copy_range(
1093 struct file
*file_in
,
1095 struct file
*file_out
,
1102 error
= xfs_file_share_range(file_in
, pos_in
, file_out
, pos_out
,
1110 xfs_file_clone_range(
1111 struct file
*file_in
,
1113 struct file
*file_out
,
1117 return xfs_file_share_range(file_in
, pos_in
, file_out
, pos_out
,
1123 struct inode
*inode
,
1126 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
1128 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
1135 struct inode
*inode
,
1138 struct xfs_inode
*ip
= XFS_I(inode
);
1142 error
= xfs_file_open(inode
, file
);
1147 * If there are any blocks, read-ahead block 0 as we're almost
1148 * certain to have the next operation be a read there.
1150 mode
= xfs_ilock_data_map_shared(ip
);
1151 if (ip
->i_d
.di_nextents
> 0)
1152 xfs_dir3_data_readahead(ip
, 0, -1);
1153 xfs_iunlock(ip
, mode
);
1159 struct inode
*inode
,
1162 return xfs_release(XFS_I(inode
));
1168 struct dir_context
*ctx
)
1170 struct inode
*inode
= file_inode(file
);
1171 xfs_inode_t
*ip
= XFS_I(inode
);
1175 * The Linux API doesn't pass down the total size of the buffer
1176 * we read into down to the filesystem. With the filldir concept
1177 * it's not needed for correct information, but the XFS dir2 leaf
1178 * code wants an estimate of the buffer size to calculate it's
1179 * readahead window and size the buffers used for mapping to
1182 * Try to give it an estimate that's good enough, maybe at some
1183 * point we can change the ->readdir prototype to include the
1184 * buffer size. For now we use the current glibc buffer size.
1186 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
1188 return xfs_readdir(ip
, ctx
, bufsize
);
1192 * This type is designed to indicate the type of offset we would like
1193 * to search from page cache for xfs_seek_hole_data().
1201 * Lookup the desired type of offset from the given page.
1203 * On success, return true and the offset argument will point to the
1204 * start of the region that was found. Otherwise this function will
1205 * return false and keep the offset argument unchanged.
1208 xfs_lookup_buffer_offset(
1213 loff_t lastoff
= page_offset(page
);
1215 struct buffer_head
*bh
, *head
;
1217 bh
= head
= page_buffers(page
);
1220 * Unwritten extents that have data in the page
1221 * cache covering them can be identified by the
1222 * BH_Unwritten state flag. Pages with multiple
1223 * buffers might have a mix of holes, data and
1224 * unwritten extents - any buffer with valid
1225 * data in it should have BH_Uptodate flag set
1228 if (buffer_unwritten(bh
) ||
1229 buffer_uptodate(bh
)) {
1230 if (type
== DATA_OFF
)
1233 if (type
== HOLE_OFF
)
1241 lastoff
+= bh
->b_size
;
1242 } while ((bh
= bh
->b_this_page
) != head
);
1248 * This routine is called to find out and return a data or hole offset
1249 * from the page cache for unwritten extents according to the desired
1250 * type for xfs_seek_hole_data().
1252 * The argument offset is used to tell where we start to search from the
1253 * page cache. Map is used to figure out the end points of the range to
1256 * Return true if the desired type of offset was found, and the argument
1257 * offset is filled with that address. Otherwise, return false and keep
1261 xfs_find_get_desired_pgoff(
1262 struct inode
*inode
,
1263 struct xfs_bmbt_irec
*map
,
1267 struct xfs_inode
*ip
= XFS_I(inode
);
1268 struct xfs_mount
*mp
= ip
->i_mount
;
1269 struct pagevec pvec
;
1273 loff_t startoff
= *offset
;
1274 loff_t lastoff
= startoff
;
1277 pagevec_init(&pvec
, 0);
1279 index
= startoff
>> PAGE_SHIFT
;
1280 endoff
= XFS_FSB_TO_B(mp
, map
->br_startoff
+ map
->br_blockcount
);
1281 end
= endoff
>> PAGE_SHIFT
;
1287 want
= min_t(pgoff_t
, end
- index
, PAGEVEC_SIZE
);
1288 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, index
,
1291 * No page mapped into given range. If we are searching holes
1292 * and if this is the first time we got into the loop, it means
1293 * that the given offset is landed in a hole, return it.
1295 * If we have already stepped through some block buffers to find
1296 * holes but they all contains data. In this case, the last
1297 * offset is already updated and pointed to the end of the last
1298 * mapped page, if it does not reach the endpoint to search,
1299 * that means there should be a hole between them.
1301 if (nr_pages
== 0) {
1302 /* Data search found nothing */
1303 if (type
== DATA_OFF
)
1306 ASSERT(type
== HOLE_OFF
);
1307 if (lastoff
== startoff
|| lastoff
< endoff
) {
1315 * At lease we found one page. If this is the first time we
1316 * step into the loop, and if the first page index offset is
1317 * greater than the given search offset, a hole was found.
1319 if (type
== HOLE_OFF
&& lastoff
== startoff
&&
1320 lastoff
< page_offset(pvec
.pages
[0])) {
1325 for (i
= 0; i
< nr_pages
; i
++) {
1326 struct page
*page
= pvec
.pages
[i
];
1330 * At this point, the page may be truncated or
1331 * invalidated (changing page->mapping to NULL),
1332 * or even swizzled back from swapper_space to tmpfs
1333 * file mapping. However, page->index will not change
1334 * because we have a reference on the page.
1336 * Searching done if the page index is out of range.
1337 * If the current offset is not reaches the end of
1338 * the specified search range, there should be a hole
1341 if (page
->index
> end
) {
1342 if (type
== HOLE_OFF
&& lastoff
< endoff
) {
1351 * Page truncated or invalidated(page->mapping == NULL).
1352 * We can freely skip it and proceed to check the next
1355 if (unlikely(page
->mapping
!= inode
->i_mapping
)) {
1360 if (!page_has_buffers(page
)) {
1365 found
= xfs_lookup_buffer_offset(page
, &b_offset
, type
);
1368 * The found offset may be less than the start
1369 * point to search if this is the first time to
1372 *offset
= max_t(loff_t
, startoff
, b_offset
);
1378 * We either searching data but nothing was found, or
1379 * searching hole but found a data buffer. In either
1380 * case, probably the next page contains the desired
1381 * things, update the last offset to it so.
1383 lastoff
= page_offset(page
) + PAGE_SIZE
;
1388 * The number of returned pages less than our desired, search
1389 * done. In this case, nothing was found for searching data,
1390 * but we found a hole behind the last offset.
1392 if (nr_pages
< want
) {
1393 if (type
== HOLE_OFF
) {
1400 index
= pvec
.pages
[i
- 1]->index
+ 1;
1401 pagevec_release(&pvec
);
1402 } while (index
<= end
);
1405 pagevec_release(&pvec
);
1410 * caller must lock inode with xfs_ilock_data_map_shared,
1411 * can we craft an appropriate ASSERT?
1413 * end is because the VFS-level lseek interface is defined such that any
1414 * offset past i_size shall return -ENXIO, but we use this for quota code
1415 * which does not maintain i_size, and we want to SEEK_DATA past i_size.
1418 __xfs_seek_hole_data(
1419 struct inode
*inode
,
1424 struct xfs_inode
*ip
= XFS_I(inode
);
1425 struct xfs_mount
*mp
= ip
->i_mount
;
1426 loff_t
uninitialized_var(offset
);
1427 xfs_fileoff_t fsbno
;
1428 xfs_filblks_t lastbno
;
1437 * Try to read extents from the first block indicated
1438 * by fsbno to the end block of the file.
1440 fsbno
= XFS_B_TO_FSBT(mp
, start
);
1441 lastbno
= XFS_B_TO_FSB(mp
, end
);
1444 struct xfs_bmbt_irec map
[2];
1448 error
= xfs_bmapi_read(ip
, fsbno
, lastbno
- fsbno
, map
, &nmap
,
1453 /* No extents at given offset, must be beyond EOF */
1459 for (i
= 0; i
< nmap
; i
++) {
1460 offset
= max_t(loff_t
, start
,
1461 XFS_FSB_TO_B(mp
, map
[i
].br_startoff
));
1463 /* Landed in the hole we wanted? */
1464 if (whence
== SEEK_HOLE
&&
1465 map
[i
].br_startblock
== HOLESTARTBLOCK
)
1468 /* Landed in the data extent we wanted? */
1469 if (whence
== SEEK_DATA
&&
1470 (map
[i
].br_startblock
== DELAYSTARTBLOCK
||
1471 (map
[i
].br_state
== XFS_EXT_NORM
&&
1472 !isnullstartblock(map
[i
].br_startblock
))))
1476 * Landed in an unwritten extent, try to search
1477 * for hole or data from page cache.
1479 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
) {
1480 if (xfs_find_get_desired_pgoff(inode
, &map
[i
],
1481 whence
== SEEK_HOLE
? HOLE_OFF
: DATA_OFF
,
1488 * We only received one extent out of the two requested. This
1489 * means we've hit EOF and didn't find what we are looking for.
1493 * If we were looking for a hole, set offset to
1494 * the end of the file (i.e., there is an implicit
1495 * hole at the end of any file).
1497 if (whence
== SEEK_HOLE
) {
1502 * If we were looking for data, it's nowhere to be found
1504 ASSERT(whence
== SEEK_DATA
);
1512 * Nothing was found, proceed to the next round of search
1513 * if the next reading offset is not at or beyond EOF.
1515 fsbno
= map
[i
- 1].br_startoff
+ map
[i
- 1].br_blockcount
;
1516 start
= XFS_FSB_TO_B(mp
, fsbno
);
1518 if (whence
== SEEK_HOLE
) {
1522 ASSERT(whence
== SEEK_DATA
);
1530 * If at this point we have found the hole we wanted, the returned
1531 * offset may be bigger than the file size as it may be aligned to
1532 * page boundary for unwritten extents. We need to deal with this
1533 * situation in particular.
1535 if (whence
== SEEK_HOLE
)
1536 offset
= min_t(loff_t
, offset
, end
);
1550 struct inode
*inode
= file
->f_mapping
->host
;
1551 struct xfs_inode
*ip
= XFS_I(inode
);
1552 struct xfs_mount
*mp
= ip
->i_mount
;
1557 if (XFS_FORCED_SHUTDOWN(mp
))
1560 lock
= xfs_ilock_data_map_shared(ip
);
1562 end
= i_size_read(inode
);
1563 offset
= __xfs_seek_hole_data(inode
, start
, end
, whence
);
1569 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1572 xfs_iunlock(ip
, lock
);
1589 return generic_file_llseek(file
, offset
, whence
);
1592 return xfs_seek_hole_data(file
, offset
, whence
);
1599 * Locking for serialisation of IO during page faults. This results in a lock
1603 * sb_start_pagefault(vfs, freeze)
1604 * i_mmaplock (XFS - truncate serialisation)
1606 * i_lock (XFS - extent map serialisation)
1610 * mmap()d file has taken write protection fault and is being made writable. We
1611 * can set the page state up correctly for a writable page, which means we can
1612 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1616 xfs_filemap_page_mkwrite(
1617 struct vm_area_struct
*vma
,
1618 struct vm_fault
*vmf
)
1620 struct inode
*inode
= file_inode(vma
->vm_file
);
1623 trace_xfs_filemap_page_mkwrite(XFS_I(inode
));
1625 sb_start_pagefault(inode
->i_sb
);
1626 file_update_time(vma
->vm_file
);
1627 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1629 if (IS_DAX(inode
)) {
1630 ret
= iomap_dax_fault(vma
, vmf
, &xfs_iomap_ops
);
1632 ret
= iomap_page_mkwrite(vma
, vmf
, &xfs_iomap_ops
);
1633 ret
= block_page_mkwrite_return(ret
);
1636 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1637 sb_end_pagefault(inode
->i_sb
);
1644 struct vm_area_struct
*vma
,
1645 struct vm_fault
*vmf
)
1647 struct inode
*inode
= file_inode(vma
->vm_file
);
1650 trace_xfs_filemap_fault(XFS_I(inode
));
1652 /* DAX can shortcut the normal fault path on write faults! */
1653 if ((vmf
->flags
& FAULT_FLAG_WRITE
) && IS_DAX(inode
))
1654 return xfs_filemap_page_mkwrite(vma
, vmf
);
1656 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1657 if (IS_DAX(inode
)) {
1659 * we do not want to trigger unwritten extent conversion on read
1660 * faults - that is unnecessary overhead and would also require
1661 * changes to xfs_get_blocks_direct() to map unwritten extent
1662 * ioend for conversion on read-only mappings.
1664 ret
= iomap_dax_fault(vma
, vmf
, &xfs_iomap_ops
);
1666 ret
= filemap_fault(vma
, vmf
);
1667 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1673 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1674 * both read and write faults. Hence we need to handle both cases. There is no
1675 * ->pmd_mkwrite callout for huge pages, so we have a single function here to
1676 * handle both cases here. @flags carries the information on the type of fault
1680 xfs_filemap_pmd_fault(
1681 struct vm_area_struct
*vma
,
1686 struct inode
*inode
= file_inode(vma
->vm_file
);
1687 struct xfs_inode
*ip
= XFS_I(inode
);
1691 return VM_FAULT_FALLBACK
;
1693 trace_xfs_filemap_pmd_fault(ip
);
1695 if (flags
& FAULT_FLAG_WRITE
) {
1696 sb_start_pagefault(inode
->i_sb
);
1697 file_update_time(vma
->vm_file
);
1700 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1701 ret
= dax_pmd_fault(vma
, addr
, pmd
, flags
, xfs_get_blocks_dax_fault
);
1702 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1704 if (flags
& FAULT_FLAG_WRITE
)
1705 sb_end_pagefault(inode
->i_sb
);
1711 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1712 * updates on write faults. In reality, it's need to serialise against
1713 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1714 * to ensure we serialise the fault barrier in place.
1717 xfs_filemap_pfn_mkwrite(
1718 struct vm_area_struct
*vma
,
1719 struct vm_fault
*vmf
)
1722 struct inode
*inode
= file_inode(vma
->vm_file
);
1723 struct xfs_inode
*ip
= XFS_I(inode
);
1724 int ret
= VM_FAULT_NOPAGE
;
1727 trace_xfs_filemap_pfn_mkwrite(ip
);
1729 sb_start_pagefault(inode
->i_sb
);
1730 file_update_time(vma
->vm_file
);
1732 /* check if the faulting page hasn't raced with truncate */
1733 xfs_ilock(ip
, XFS_MMAPLOCK_SHARED
);
1734 size
= (i_size_read(inode
) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1735 if (vmf
->pgoff
>= size
)
1736 ret
= VM_FAULT_SIGBUS
;
1737 else if (IS_DAX(inode
))
1738 ret
= dax_pfn_mkwrite(vma
, vmf
);
1739 xfs_iunlock(ip
, XFS_MMAPLOCK_SHARED
);
1740 sb_end_pagefault(inode
->i_sb
);
1745 static const struct vm_operations_struct xfs_file_vm_ops
= {
1746 .fault
= xfs_filemap_fault
,
1747 .pmd_fault
= xfs_filemap_pmd_fault
,
1748 .map_pages
= filemap_map_pages
,
1749 .page_mkwrite
= xfs_filemap_page_mkwrite
,
1750 .pfn_mkwrite
= xfs_filemap_pfn_mkwrite
,
1756 struct vm_area_struct
*vma
)
1758 file_accessed(filp
);
1759 vma
->vm_ops
= &xfs_file_vm_ops
;
1760 if (IS_DAX(file_inode(filp
)))
1761 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
1765 const struct file_operations xfs_file_operations
= {
1766 .llseek
= xfs_file_llseek
,
1767 .read_iter
= xfs_file_read_iter
,
1768 .write_iter
= xfs_file_write_iter
,
1769 .splice_read
= xfs_file_splice_read
,
1770 .splice_write
= iter_file_splice_write
,
1771 .unlocked_ioctl
= xfs_file_ioctl
,
1772 #ifdef CONFIG_COMPAT
1773 .compat_ioctl
= xfs_file_compat_ioctl
,
1775 .mmap
= xfs_file_mmap
,
1776 .open
= xfs_file_open
,
1777 .release
= xfs_file_release
,
1778 .fsync
= xfs_file_fsync
,
1779 .fallocate
= xfs_file_fallocate
,
1780 .copy_file_range
= xfs_file_copy_range
,
1781 .clone_file_range
= xfs_file_clone_range
,
1784 const struct file_operations xfs_dir_file_operations
= {
1785 .open
= xfs_dir_open
,
1786 .read
= generic_read_dir
,
1787 .iterate_shared
= xfs_file_readdir
,
1788 .llseek
= generic_file_llseek
,
1789 .unlocked_ioctl
= xfs_file_ioctl
,
1790 #ifdef CONFIG_COMPAT
1791 .compat_ioctl
= xfs_file_compat_ioctl
,
1793 .fsync
= xfs_dir_fsync
,