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 * Clear the specified ranges to zero through either the pagecache or DAX.
52 * Holes and unwritten extents will be left as-is as they already are zeroed.
61 return iomap_zero_range(VFS_I(ip
), pos
, count
, did_zero
, &xfs_iomap_ops
);
65 xfs_update_prealloc_flags(
67 enum xfs_prealloc_flags flags
)
72 error
= xfs_trans_alloc(ip
->i_mount
, &M_RES(ip
->i_mount
)->tr_writeid
,
77 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
78 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
80 if (!(flags
& XFS_PREALLOC_INVISIBLE
)) {
81 VFS_I(ip
)->i_mode
&= ~S_ISUID
;
82 if (VFS_I(ip
)->i_mode
& S_IXGRP
)
83 VFS_I(ip
)->i_mode
&= ~S_ISGID
;
84 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
87 if (flags
& XFS_PREALLOC_SET
)
88 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
89 if (flags
& XFS_PREALLOC_CLEAR
)
90 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
92 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
93 if (flags
& XFS_PREALLOC_SYNC
)
94 xfs_trans_set_sync(tp
);
95 return xfs_trans_commit(tp
);
99 * Fsync operations on directories are much simpler than on regular files,
100 * as there is no file data to flush, and thus also no need for explicit
101 * cache flush operations, and there are no non-transaction metadata updates
102 * on directories either.
111 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
112 struct xfs_mount
*mp
= ip
->i_mount
;
115 trace_xfs_dir_fsync(ip
);
117 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
118 if (xfs_ipincount(ip
))
119 lsn
= ip
->i_itemp
->ili_last_lsn
;
120 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
124 return _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, NULL
);
134 struct inode
*inode
= file
->f_mapping
->host
;
135 struct xfs_inode
*ip
= XFS_I(inode
);
136 struct xfs_mount
*mp
= ip
->i_mount
;
141 trace_xfs_file_fsync(ip
);
143 error
= file_write_and_wait_range(file
, start
, end
);
147 if (XFS_FORCED_SHUTDOWN(mp
))
150 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
153 * If we have an RT and/or log subvolume we need to make sure to flush
154 * the write cache the device used for file data first. This is to
155 * ensure newly written file data make it to disk before logging the new
156 * inode size in case of an extending write.
158 if (XFS_IS_REALTIME_INODE(ip
))
159 xfs_blkdev_issue_flush(mp
->m_rtdev_targp
);
160 else if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
161 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
164 * All metadata updates are logged, which means that we just have to
165 * flush the log up to the latest LSN that touched the inode. If we have
166 * concurrent fsync/fdatasync() calls, we need them to all block on the
167 * log force before we clear the ili_fsync_fields field. This ensures
168 * that we don't get a racing sync operation that does not wait for the
169 * metadata to hit the journal before returning. If we race with
170 * clearing the ili_fsync_fields, then all that will happen is the log
171 * force will do nothing as the lsn will already be on disk. We can't
172 * race with setting ili_fsync_fields because that is done under
173 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
174 * until after the ili_fsync_fields is cleared.
176 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
177 if (xfs_ipincount(ip
)) {
179 (ip
->i_itemp
->ili_fsync_fields
& ~XFS_ILOG_TIMESTAMP
))
180 lsn
= ip
->i_itemp
->ili_last_lsn
;
184 error
= _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, &log_flushed
);
185 ip
->i_itemp
->ili_fsync_fields
= 0;
187 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
190 * If we only have a single device, and the log force about was
191 * a no-op we might have to flush the data device cache here.
192 * This can only happen for fdatasync/O_DSYNC if we were overwriting
193 * an already allocated file and thus do not have any metadata to
196 if (!log_flushed
&& !XFS_IS_REALTIME_INODE(ip
) &&
197 mp
->m_logdev_targp
== mp
->m_ddev_targp
)
198 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
204 xfs_file_dio_aio_read(
208 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
209 size_t count
= iov_iter_count(to
);
212 trace_xfs_file_direct_read(ip
, count
, iocb
->ki_pos
);
215 return 0; /* skip atime */
217 file_accessed(iocb
->ki_filp
);
219 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
220 ret
= iomap_dio_rw(iocb
, to
, &xfs_iomap_ops
, NULL
);
221 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
226 static noinline ssize_t
231 struct xfs_inode
*ip
= XFS_I(iocb
->ki_filp
->f_mapping
->host
);
232 size_t count
= iov_iter_count(to
);
235 trace_xfs_file_dax_read(ip
, count
, iocb
->ki_pos
);
238 return 0; /* skip atime */
240 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
)) {
241 if (iocb
->ki_flags
& IOCB_NOWAIT
)
243 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
245 ret
= dax_iomap_rw(iocb
, to
, &xfs_iomap_ops
);
246 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
248 file_accessed(iocb
->ki_filp
);
253 xfs_file_buffered_aio_read(
257 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
260 trace_xfs_file_buffered_read(ip
, iov_iter_count(to
), iocb
->ki_pos
);
262 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
263 ret
= generic_file_read_iter(iocb
, to
);
264 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
274 struct inode
*inode
= file_inode(iocb
->ki_filp
);
275 struct xfs_mount
*mp
= XFS_I(inode
)->i_mount
;
278 XFS_STATS_INC(mp
, xs_read_calls
);
280 if (XFS_FORCED_SHUTDOWN(mp
))
284 ret
= xfs_file_dax_read(iocb
, to
);
285 else if (iocb
->ki_flags
& IOCB_DIRECT
)
286 ret
= xfs_file_dio_aio_read(iocb
, to
);
288 ret
= xfs_file_buffered_aio_read(iocb
, to
);
291 XFS_STATS_ADD(mp
, xs_read_bytes
, ret
);
296 * Zero any on disk space between the current EOF and the new, larger EOF.
298 * This handles the normal case of zeroing the remainder of the last block in
299 * the file and the unusual case of zeroing blocks out beyond the size of the
300 * file. This second case only happens with fixed size extents and when the
301 * system crashes before the inode size was updated but after blocks were
304 * Expects the iolock to be held exclusive, and will take the ilock internally.
306 int /* error (positive) */
308 struct xfs_inode
*ip
,
309 xfs_off_t offset
, /* starting I/O offset */
310 xfs_fsize_t isize
, /* current inode size */
313 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
314 ASSERT(offset
> isize
);
316 trace_xfs_zero_eof(ip
, isize
, offset
- isize
);
317 return xfs_zero_range(ip
, isize
, offset
- isize
, did_zeroing
);
321 * Common pre-write limit and setup checks.
323 * Called with the iolocked held either shared and exclusive according to
324 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
325 * if called for a direct write beyond i_size.
328 xfs_file_aio_write_checks(
330 struct iov_iter
*from
,
333 struct file
*file
= iocb
->ki_filp
;
334 struct inode
*inode
= file
->f_mapping
->host
;
335 struct xfs_inode
*ip
= XFS_I(inode
);
337 size_t count
= iov_iter_count(from
);
338 bool drained_dio
= false;
341 error
= generic_write_checks(iocb
, from
);
345 error
= xfs_break_layouts(inode
, iolock
);
350 * For changing security info in file_remove_privs() we need i_rwsem
353 if (*iolock
== XFS_IOLOCK_SHARED
&& !IS_NOSEC(inode
)) {
354 xfs_iunlock(ip
, *iolock
);
355 *iolock
= XFS_IOLOCK_EXCL
;
356 xfs_ilock(ip
, *iolock
);
360 * If the offset is beyond the size of the file, we need to zero any
361 * blocks that fall between the existing EOF and the start of this
362 * write. If zeroing is needed and we are currently holding the
363 * iolock shared, we need to update it to exclusive which implies
364 * having to redo all checks before.
366 * We need to serialise against EOF updates that occur in IO
367 * completions here. We want to make sure that nobody is changing the
368 * size while we do this check until we have placed an IO barrier (i.e.
369 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
370 * The spinlock effectively forms a memory barrier once we have the
371 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
372 * and hence be able to correctly determine if we need to run zeroing.
374 spin_lock(&ip
->i_flags_lock
);
375 if (iocb
->ki_pos
> i_size_read(inode
)) {
378 spin_unlock(&ip
->i_flags_lock
);
380 if (*iolock
== XFS_IOLOCK_SHARED
) {
381 xfs_iunlock(ip
, *iolock
);
382 *iolock
= XFS_IOLOCK_EXCL
;
383 xfs_ilock(ip
, *iolock
);
384 iov_iter_reexpand(from
, count
);
387 * We now have an IO submission barrier in place, but
388 * AIO can do EOF updates during IO completion and hence
389 * we now need to wait for all of them to drain. Non-AIO
390 * DIO will have drained before we are given the
391 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
394 inode_dio_wait(inode
);
398 error
= xfs_zero_eof(ip
, iocb
->ki_pos
, i_size_read(inode
), &zero
);
402 spin_unlock(&ip
->i_flags_lock
);
405 * Updating the timestamps will grab the ilock again from
406 * xfs_fs_dirty_inode, so we have to call it after dropping the
407 * lock above. Eventually we should look into a way to avoid
408 * the pointless lock roundtrip.
410 if (likely(!(file
->f_mode
& FMODE_NOCMTIME
))) {
411 error
= file_update_time(file
);
417 * If we're writing the file then make sure to clear the setuid and
418 * setgid bits if the process is not being run by root. This keeps
419 * people from modifying setuid and setgid binaries.
421 if (!IS_NOSEC(inode
))
422 return file_remove_privs(file
);
427 xfs_dio_write_end_io(
432 struct inode
*inode
= file_inode(iocb
->ki_filp
);
433 struct xfs_inode
*ip
= XFS_I(inode
);
434 loff_t offset
= iocb
->ki_pos
;
437 trace_xfs_end_io_direct_write(ip
, offset
, size
);
439 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
445 if (flags
& IOMAP_DIO_COW
) {
446 error
= xfs_reflink_end_cow(ip
, offset
, size
);
452 * Unwritten conversion updates the in-core isize after extent
453 * conversion but before updating the on-disk size. Updating isize any
454 * earlier allows a racing dio read to find unwritten extents before
455 * they are converted.
457 if (flags
& IOMAP_DIO_UNWRITTEN
)
458 return xfs_iomap_write_unwritten(ip
, offset
, size
, true);
461 * We need to update the in-core inode size here so that we don't end up
462 * with the on-disk inode size being outside the in-core inode size. We
463 * have no other method of updating EOF for AIO, so always do it here
466 * We need to lock the test/set EOF update as we can be racing with
467 * other IO completions here to update the EOF. Failing to serialise
468 * here can result in EOF moving backwards and Bad Things Happen when
471 spin_lock(&ip
->i_flags_lock
);
472 if (offset
+ size
> i_size_read(inode
)) {
473 i_size_write(inode
, offset
+ size
);
474 spin_unlock(&ip
->i_flags_lock
);
475 error
= xfs_setfilesize(ip
, offset
, size
);
477 spin_unlock(&ip
->i_flags_lock
);
484 * xfs_file_dio_aio_write - handle direct IO writes
486 * Lock the inode appropriately to prepare for and issue a direct IO write.
487 * By separating it from the buffered write path we remove all the tricky to
488 * follow locking changes and looping.
490 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
491 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
492 * pages are flushed out.
494 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
495 * allowing them to be done in parallel with reads and other direct IO writes.
496 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
497 * needs to do sub-block zeroing and that requires serialisation against other
498 * direct IOs to the same block. In this case we need to serialise the
499 * submission of the unaligned IOs so that we don't get racing block zeroing in
500 * the dio layer. To avoid the problem with aio, we also need to wait for
501 * outstanding IOs to complete so that unwritten extent conversion is completed
502 * before we try to map the overlapping block. This is currently implemented by
503 * hitting it with a big hammer (i.e. inode_dio_wait()).
505 * Returns with locks held indicated by @iolock and errors indicated by
506 * negative return values.
509 xfs_file_dio_aio_write(
511 struct iov_iter
*from
)
513 struct file
*file
= iocb
->ki_filp
;
514 struct address_space
*mapping
= file
->f_mapping
;
515 struct inode
*inode
= mapping
->host
;
516 struct xfs_inode
*ip
= XFS_I(inode
);
517 struct xfs_mount
*mp
= ip
->i_mount
;
519 int unaligned_io
= 0;
521 size_t count
= iov_iter_count(from
);
522 struct xfs_buftarg
*target
= XFS_IS_REALTIME_INODE(ip
) ?
523 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
525 /* DIO must be aligned to device logical sector size */
526 if ((iocb
->ki_pos
| count
) & target
->bt_logical_sectormask
)
530 * Don't take the exclusive iolock here unless the I/O is unaligned to
531 * the file system block size. We don't need to consider the EOF
532 * extension case here because xfs_file_aio_write_checks() will relock
533 * the inode as necessary for EOF zeroing cases and fill out the new
534 * inode size as appropriate.
536 if ((iocb
->ki_pos
& mp
->m_blockmask
) ||
537 ((iocb
->ki_pos
+ count
) & mp
->m_blockmask
)) {
541 * We can't properly handle unaligned direct I/O to reflink
542 * files yet, as we can't unshare a partial block.
544 if (xfs_is_reflink_inode(ip
)) {
545 trace_xfs_reflink_bounce_dio_write(ip
, iocb
->ki_pos
, count
);
548 iolock
= XFS_IOLOCK_EXCL
;
550 iolock
= XFS_IOLOCK_SHARED
;
553 if (!xfs_ilock_nowait(ip
, iolock
)) {
554 if (iocb
->ki_flags
& IOCB_NOWAIT
)
556 xfs_ilock(ip
, iolock
);
559 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
562 count
= iov_iter_count(from
);
565 * If we are doing unaligned IO, wait for all other IO to drain,
566 * otherwise demote the lock if we had to take the exclusive lock
567 * for other reasons in xfs_file_aio_write_checks.
570 /* If we are going to wait for other DIO to finish, bail */
571 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
572 if (atomic_read(&inode
->i_dio_count
))
575 inode_dio_wait(inode
);
577 } else if (iolock
== XFS_IOLOCK_EXCL
) {
578 xfs_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
579 iolock
= XFS_IOLOCK_SHARED
;
582 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
);
583 ret
= iomap_dio_rw(iocb
, from
, &xfs_iomap_ops
, xfs_dio_write_end_io
);
585 xfs_iunlock(ip
, iolock
);
588 * No fallback to buffered IO on errors for XFS, direct IO will either
589 * complete fully or fail.
591 ASSERT(ret
< 0 || ret
== count
);
595 static noinline ssize_t
598 struct iov_iter
*from
)
600 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
601 struct xfs_inode
*ip
= XFS_I(inode
);
602 int iolock
= XFS_IOLOCK_EXCL
;
603 ssize_t ret
, error
= 0;
607 if (!xfs_ilock_nowait(ip
, iolock
)) {
608 if (iocb
->ki_flags
& IOCB_NOWAIT
)
610 xfs_ilock(ip
, iolock
);
613 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
618 count
= iov_iter_count(from
);
620 trace_xfs_file_dax_write(ip
, count
, pos
);
621 ret
= dax_iomap_rw(iocb
, from
, &xfs_iomap_ops
);
622 if (ret
> 0 && iocb
->ki_pos
> i_size_read(inode
)) {
623 i_size_write(inode
, iocb
->ki_pos
);
624 error
= xfs_setfilesize(ip
, pos
, ret
);
627 xfs_iunlock(ip
, iolock
);
628 return error
? error
: ret
;
632 xfs_file_buffered_aio_write(
634 struct iov_iter
*from
)
636 struct file
*file
= iocb
->ki_filp
;
637 struct address_space
*mapping
= file
->f_mapping
;
638 struct inode
*inode
= mapping
->host
;
639 struct xfs_inode
*ip
= XFS_I(inode
);
645 iolock
= XFS_IOLOCK_EXCL
;
646 xfs_ilock(ip
, iolock
);
648 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
652 /* We can write back this queue in page reclaim */
653 current
->backing_dev_info
= inode_to_bdi(inode
);
655 trace_xfs_file_buffered_write(ip
, iov_iter_count(from
), iocb
->ki_pos
);
656 ret
= iomap_file_buffered_write(iocb
, from
, &xfs_iomap_ops
);
657 if (likely(ret
>= 0))
661 * If we hit a space limit, try to free up some lingering preallocated
662 * space before returning an error. In the case of ENOSPC, first try to
663 * write back all dirty inodes to free up some of the excess reserved
664 * metadata space. This reduces the chances that the eofblocks scan
665 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
666 * also behaves as a filter to prevent too many eofblocks scans from
667 * running at the same time.
669 if (ret
== -EDQUOT
&& !enospc
) {
670 xfs_iunlock(ip
, iolock
);
671 enospc
= xfs_inode_free_quota_eofblocks(ip
);
674 enospc
= xfs_inode_free_quota_cowblocks(ip
);
678 } else if (ret
== -ENOSPC
&& !enospc
) {
679 struct xfs_eofblocks eofb
= {0};
682 xfs_flush_inodes(ip
->i_mount
);
684 xfs_iunlock(ip
, iolock
);
685 eofb
.eof_flags
= XFS_EOF_FLAGS_SYNC
;
686 xfs_icache_free_eofblocks(ip
->i_mount
, &eofb
);
687 xfs_icache_free_cowblocks(ip
->i_mount
, &eofb
);
691 current
->backing_dev_info
= NULL
;
694 xfs_iunlock(ip
, iolock
);
701 struct iov_iter
*from
)
703 struct file
*file
= iocb
->ki_filp
;
704 struct address_space
*mapping
= file
->f_mapping
;
705 struct inode
*inode
= mapping
->host
;
706 struct xfs_inode
*ip
= XFS_I(inode
);
708 size_t ocount
= iov_iter_count(from
);
710 XFS_STATS_INC(ip
->i_mount
, xs_write_calls
);
715 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
719 ret
= xfs_file_dax_write(iocb
, from
);
720 else if (iocb
->ki_flags
& IOCB_DIRECT
) {
722 * Allow a directio write to fall back to a buffered
723 * write *only* in the case that we're doing a reflink
724 * CoW. In all other directio scenarios we do not
725 * allow an operation to fall back to buffered mode.
727 ret
= xfs_file_dio_aio_write(iocb
, from
);
732 ret
= xfs_file_buffered_aio_write(iocb
, from
);
736 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
738 /* Handle various SYNC-type writes */
739 ret
= generic_write_sync(iocb
, ret
);
744 #define XFS_FALLOC_FL_SUPPORTED \
745 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
746 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
747 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
756 struct inode
*inode
= file_inode(file
);
757 struct xfs_inode
*ip
= XFS_I(inode
);
759 enum xfs_prealloc_flags flags
= 0;
760 uint iolock
= XFS_IOLOCK_EXCL
;
762 bool do_file_insert
= 0;
764 if (!S_ISREG(inode
->i_mode
))
766 if (mode
& ~XFS_FALLOC_FL_SUPPORTED
)
769 xfs_ilock(ip
, iolock
);
770 error
= xfs_break_layouts(inode
, &iolock
);
774 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
775 iolock
|= XFS_MMAPLOCK_EXCL
;
777 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
778 error
= xfs_free_file_space(ip
, offset
, len
);
781 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
782 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
784 if (offset
& blksize_mask
|| len
& blksize_mask
) {
790 * There is no need to overlap collapse range with EOF,
791 * in which case it is effectively a truncate operation
793 if (offset
+ len
>= i_size_read(inode
)) {
798 new_size
= i_size_read(inode
) - len
;
800 error
= xfs_collapse_file_space(ip
, offset
, len
);
803 } else if (mode
& FALLOC_FL_INSERT_RANGE
) {
804 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
806 new_size
= i_size_read(inode
) + len
;
807 if (offset
& blksize_mask
|| len
& blksize_mask
) {
812 /* check the new inode size does not wrap through zero */
813 if (new_size
> inode
->i_sb
->s_maxbytes
) {
818 /* Offset should be less than i_size */
819 if (offset
>= i_size_read(inode
)) {
825 flags
|= XFS_PREALLOC_SET
;
827 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
828 offset
+ len
> i_size_read(inode
)) {
829 new_size
= offset
+ len
;
830 error
= inode_newsize_ok(inode
, new_size
);
835 if (mode
& FALLOC_FL_ZERO_RANGE
)
836 error
= xfs_zero_file_space(ip
, offset
, len
);
838 if (mode
& FALLOC_FL_UNSHARE_RANGE
) {
839 error
= xfs_reflink_unshare(ip
, offset
, len
);
843 error
= xfs_alloc_file_space(ip
, offset
, len
,
850 if (file
->f_flags
& O_DSYNC
)
851 flags
|= XFS_PREALLOC_SYNC
;
853 error
= xfs_update_prealloc_flags(ip
, flags
);
857 /* Change file size if needed */
861 iattr
.ia_valid
= ATTR_SIZE
;
862 iattr
.ia_size
= new_size
;
863 error
= xfs_vn_setattr_size(file_dentry(file
), &iattr
);
869 * Perform hole insertion now that the file size has been
870 * updated so that if we crash during the operation we don't
871 * leave shifted extents past EOF and hence losing access to
872 * the data that is contained within them.
875 error
= xfs_insert_file_space(ip
, offset
, len
);
878 xfs_iunlock(ip
, iolock
);
883 xfs_file_clone_range(
884 struct file
*file_in
,
886 struct file
*file_out
,
890 return xfs_reflink_remap_range(file_in
, pos_in
, file_out
, pos_out
,
895 xfs_file_dedupe_range(
896 struct file
*src_file
,
899 struct file
*dst_file
,
904 error
= xfs_reflink_remap_range(src_file
, loff
, dst_file
, dst_loff
,
916 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
918 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
920 file
->f_mode
|= FMODE_AIO_NOWAIT
;
929 struct xfs_inode
*ip
= XFS_I(inode
);
933 error
= xfs_file_open(inode
, file
);
938 * If there are any blocks, read-ahead block 0 as we're almost
939 * certain to have the next operation be a read there.
941 mode
= xfs_ilock_data_map_shared(ip
);
942 if (ip
->i_d
.di_nextents
> 0)
943 error
= xfs_dir3_data_readahead(ip
, 0, -1);
944 xfs_iunlock(ip
, mode
);
953 return xfs_release(XFS_I(inode
));
959 struct dir_context
*ctx
)
961 struct inode
*inode
= file_inode(file
);
962 xfs_inode_t
*ip
= XFS_I(inode
);
966 * The Linux API doesn't pass down the total size of the buffer
967 * we read into down to the filesystem. With the filldir concept
968 * it's not needed for correct information, but the XFS dir2 leaf
969 * code wants an estimate of the buffer size to calculate it's
970 * readahead window and size the buffers used for mapping to
973 * Try to give it an estimate that's good enough, maybe at some
974 * point we can change the ->readdir prototype to include the
975 * buffer size. For now we use the current glibc buffer size.
977 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
979 return xfs_readdir(NULL
, ip
, ctx
, bufsize
);
988 struct inode
*inode
= file
->f_mapping
->host
;
990 if (XFS_FORCED_SHUTDOWN(XFS_I(inode
)->i_mount
))
995 return generic_file_llseek(file
, offset
, whence
);
997 offset
= iomap_seek_hole(inode
, offset
, &xfs_iomap_ops
);
1000 offset
= iomap_seek_data(inode
, offset
, &xfs_iomap_ops
);
1006 return vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1010 * Locking for serialisation of IO during page faults. This results in a lock
1014 * sb_start_pagefault(vfs, freeze)
1015 * i_mmaplock (XFS - truncate serialisation)
1017 * i_lock (XFS - extent map serialisation)
1021 * mmap()d file has taken write protection fault and is being made writable. We
1022 * can set the page state up correctly for a writable page, which means we can
1023 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1027 xfs_filemap_page_mkwrite(
1028 struct vm_fault
*vmf
)
1030 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1033 trace_xfs_filemap_page_mkwrite(XFS_I(inode
));
1035 sb_start_pagefault(inode
->i_sb
);
1036 file_update_time(vmf
->vma
->vm_file
);
1037 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1039 if (IS_DAX(inode
)) {
1040 ret
= dax_iomap_fault(vmf
, PE_SIZE_PTE
, &xfs_iomap_ops
);
1042 ret
= iomap_page_mkwrite(vmf
, &xfs_iomap_ops
);
1043 ret
= block_page_mkwrite_return(ret
);
1046 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1047 sb_end_pagefault(inode
->i_sb
);
1054 struct vm_fault
*vmf
)
1056 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1059 trace_xfs_filemap_fault(XFS_I(inode
));
1061 /* DAX can shortcut the normal fault path on write faults! */
1062 if ((vmf
->flags
& FAULT_FLAG_WRITE
) && IS_DAX(inode
))
1063 return xfs_filemap_page_mkwrite(vmf
);
1065 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1067 ret
= dax_iomap_fault(vmf
, PE_SIZE_PTE
, &xfs_iomap_ops
);
1069 ret
= filemap_fault(vmf
);
1070 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1076 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1077 * both read and write faults. Hence we need to handle both cases. There is no
1078 * ->huge_mkwrite callout for huge pages, so we have a single function here to
1079 * handle both cases here. @flags carries the information on the type of fault
1083 xfs_filemap_huge_fault(
1084 struct vm_fault
*vmf
,
1085 enum page_entry_size pe_size
)
1087 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1088 struct xfs_inode
*ip
= XFS_I(inode
);
1092 return VM_FAULT_FALLBACK
;
1094 trace_xfs_filemap_huge_fault(ip
);
1096 if (vmf
->flags
& FAULT_FLAG_WRITE
) {
1097 sb_start_pagefault(inode
->i_sb
);
1098 file_update_time(vmf
->vma
->vm_file
);
1101 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1102 ret
= dax_iomap_fault(vmf
, pe_size
, &xfs_iomap_ops
);
1103 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1105 if (vmf
->flags
& FAULT_FLAG_WRITE
)
1106 sb_end_pagefault(inode
->i_sb
);
1112 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1113 * updates on write faults. In reality, it's need to serialise against
1114 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1115 * to ensure we serialise the fault barrier in place.
1118 xfs_filemap_pfn_mkwrite(
1119 struct vm_fault
*vmf
)
1122 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1123 struct xfs_inode
*ip
= XFS_I(inode
);
1124 int ret
= VM_FAULT_NOPAGE
;
1127 trace_xfs_filemap_pfn_mkwrite(ip
);
1129 sb_start_pagefault(inode
->i_sb
);
1130 file_update_time(vmf
->vma
->vm_file
);
1132 /* check if the faulting page hasn't raced with truncate */
1133 xfs_ilock(ip
, XFS_MMAPLOCK_SHARED
);
1134 size
= (i_size_read(inode
) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1135 if (vmf
->pgoff
>= size
)
1136 ret
= VM_FAULT_SIGBUS
;
1137 else if (IS_DAX(inode
))
1138 ret
= dax_iomap_fault(vmf
, PE_SIZE_PTE
, &xfs_iomap_ops
);
1139 xfs_iunlock(ip
, XFS_MMAPLOCK_SHARED
);
1140 sb_end_pagefault(inode
->i_sb
);
1145 static const struct vm_operations_struct xfs_file_vm_ops
= {
1146 .fault
= xfs_filemap_fault
,
1147 .huge_fault
= xfs_filemap_huge_fault
,
1148 .map_pages
= filemap_map_pages
,
1149 .page_mkwrite
= xfs_filemap_page_mkwrite
,
1150 .pfn_mkwrite
= xfs_filemap_pfn_mkwrite
,
1156 struct vm_area_struct
*vma
)
1158 file_accessed(filp
);
1159 vma
->vm_ops
= &xfs_file_vm_ops
;
1160 if (IS_DAX(file_inode(filp
)))
1161 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
1165 const struct file_operations xfs_file_operations
= {
1166 .llseek
= xfs_file_llseek
,
1167 .read_iter
= xfs_file_read_iter
,
1168 .write_iter
= xfs_file_write_iter
,
1169 .splice_read
= generic_file_splice_read
,
1170 .splice_write
= iter_file_splice_write
,
1171 .unlocked_ioctl
= xfs_file_ioctl
,
1172 #ifdef CONFIG_COMPAT
1173 .compat_ioctl
= xfs_file_compat_ioctl
,
1175 .mmap
= xfs_file_mmap
,
1176 .open
= xfs_file_open
,
1177 .release
= xfs_file_release
,
1178 .fsync
= xfs_file_fsync
,
1179 .get_unmapped_area
= thp_get_unmapped_area
,
1180 .fallocate
= xfs_file_fallocate
,
1181 .clone_file_range
= xfs_file_clone_range
,
1182 .dedupe_file_range
= xfs_file_dedupe_range
,
1185 const struct file_operations xfs_dir_file_operations
= {
1186 .open
= xfs_dir_open
,
1187 .read
= generic_read_dir
,
1188 .iterate_shared
= xfs_file_readdir
,
1189 .llseek
= generic_file_llseek
,
1190 .unlocked_ioctl
= xfs_file_ioctl
,
1191 #ifdef CONFIG_COMPAT
1192 .compat_ioctl
= xfs_file_compat_ioctl
,
1194 .fsync
= xfs_dir_fsync
,