1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_inode_item.h"
17 #include "xfs_bmap_util.h"
19 #include "xfs_dir2_priv.h"
20 #include "xfs_ioctl.h"
21 #include "xfs_trace.h"
23 #include "xfs_icache.h"
25 #include "xfs_iomap.h"
26 #include "xfs_reflink.h"
28 #include <linux/dax.h>
29 #include <linux/falloc.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mman.h>
32 #include <linux/fadvise.h>
33 #include <linux/mount.h>
35 static const struct vm_operations_struct xfs_file_vm_ops
;
38 * Decide if the given file range is aligned to the size of the fundamental
39 * allocation unit for the file.
42 xfs_is_falloc_aligned(
47 struct xfs_mount
*mp
= ip
->i_mount
;
50 if (XFS_IS_REALTIME_INODE(ip
)) {
51 if (!is_power_of_2(mp
->m_sb
.sb_rextsize
)) {
55 rextbytes
= XFS_FSB_TO_B(mp
, mp
->m_sb
.sb_rextsize
);
56 div_u64_rem(pos
, rextbytes
, &mod
);
59 div_u64_rem(len
, rextbytes
, &mod
);
62 mask
= XFS_FSB_TO_B(mp
, mp
->m_sb
.sb_rextsize
) - 1;
64 mask
= mp
->m_sb
.sb_blocksize
- 1;
67 return !((pos
| len
) & mask
);
71 * Fsync operations on directories are much simpler than on regular files,
72 * as there is no file data to flush, and thus also no need for explicit
73 * cache flush operations, and there are no non-transaction metadata updates
74 * on directories either.
83 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
85 trace_xfs_dir_fsync(ip
);
86 return xfs_log_force_inode(ip
);
94 if (!xfs_ipincount(ip
))
96 if (datasync
&& !(ip
->i_itemp
->ili_fsync_fields
& ~XFS_ILOG_TIMESTAMP
))
98 return ip
->i_itemp
->ili_commit_seq
;
102 * All metadata updates are logged, which means that we just have to flush the
103 * log up to the latest LSN that touched the inode.
105 * If we have concurrent fsync/fdatasync() calls, we need them to all block on
106 * the log force before we clear the ili_fsync_fields field. This ensures that
107 * we don't get a racing sync operation that does not wait for the metadata to
108 * hit the journal before returning. If we race with clearing ili_fsync_fields,
109 * then all that will happen is the log force will do nothing as the lsn will
110 * already be on disk. We can't race with setting ili_fsync_fields because that
111 * is done under XFS_ILOCK_EXCL, and that can't happen because we hold the lock
112 * shared until after the ili_fsync_fields is cleared.
116 struct xfs_inode
*ip
,
123 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
124 seq
= xfs_fsync_seq(ip
, datasync
);
126 error
= xfs_log_force_seq(ip
->i_mount
, seq
, XFS_LOG_SYNC
,
129 spin_lock(&ip
->i_itemp
->ili_lock
);
130 ip
->i_itemp
->ili_fsync_fields
= 0;
131 spin_unlock(&ip
->i_itemp
->ili_lock
);
133 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
144 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
145 struct xfs_mount
*mp
= ip
->i_mount
;
149 trace_xfs_file_fsync(ip
);
151 error
= file_write_and_wait_range(file
, start
, end
);
155 if (xfs_is_shutdown(mp
))
158 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
161 * If we have an RT and/or log subvolume we need to make sure to flush
162 * the write cache the device used for file data first. This is to
163 * ensure newly written file data make it to disk before logging the new
164 * inode size in case of an extending write.
166 if (XFS_IS_REALTIME_INODE(ip
))
167 error
= blkdev_issue_flush(mp
->m_rtdev_targp
->bt_bdev
);
168 else if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
169 error
= blkdev_issue_flush(mp
->m_ddev_targp
->bt_bdev
);
172 * Any inode that has dirty modifications in the log is pinned. The
173 * racy check here for a pinned inode will not catch modifications
174 * that happen concurrently to the fsync call, but fsync semantics
175 * only require to sync previously completed I/O.
177 if (xfs_ipincount(ip
)) {
178 err2
= xfs_fsync_flush_log(ip
, datasync
, &log_flushed
);
184 * If we only have a single device, and the log force about was
185 * a no-op we might have to flush the data device cache here.
186 * This can only happen for fdatasync/O_DSYNC if we were overwriting
187 * an already allocated file and thus do not have any metadata to
190 if (!log_flushed
&& !XFS_IS_REALTIME_INODE(ip
) &&
191 mp
->m_logdev_targp
== mp
->m_ddev_targp
) {
192 err2
= blkdev_issue_flush(mp
->m_ddev_targp
->bt_bdev
);
203 unsigned int lock_mode
)
205 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
207 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
208 if (!xfs_ilock_nowait(ip
, lock_mode
))
211 xfs_ilock(ip
, lock_mode
);
222 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
225 trace_xfs_file_direct_read(iocb
, to
);
227 if (!iov_iter_count(to
))
228 return 0; /* skip atime */
230 file_accessed(iocb
->ki_filp
);
232 ret
= xfs_ilock_iocb(iocb
, XFS_IOLOCK_SHARED
);
235 ret
= iomap_dio_rw(iocb
, to
, &xfs_read_iomap_ops
, NULL
, 0, NULL
, 0);
236 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
241 static noinline ssize_t
246 struct xfs_inode
*ip
= XFS_I(iocb
->ki_filp
->f_mapping
->host
);
249 trace_xfs_file_dax_read(iocb
, to
);
251 if (!iov_iter_count(to
))
252 return 0; /* skip atime */
254 ret
= xfs_ilock_iocb(iocb
, XFS_IOLOCK_SHARED
);
257 ret
= dax_iomap_rw(iocb
, to
, &xfs_read_iomap_ops
);
258 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
260 file_accessed(iocb
->ki_filp
);
265 xfs_file_buffered_read(
269 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
272 trace_xfs_file_buffered_read(iocb
, to
);
274 ret
= xfs_ilock_iocb(iocb
, XFS_IOLOCK_SHARED
);
277 ret
= generic_file_read_iter(iocb
, to
);
278 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
288 struct inode
*inode
= file_inode(iocb
->ki_filp
);
289 struct xfs_mount
*mp
= XFS_I(inode
)->i_mount
;
292 XFS_STATS_INC(mp
, xs_read_calls
);
294 if (xfs_is_shutdown(mp
))
298 ret
= xfs_file_dax_read(iocb
, to
);
299 else if (iocb
->ki_flags
& IOCB_DIRECT
)
300 ret
= xfs_file_dio_read(iocb
, to
);
302 ret
= xfs_file_buffered_read(iocb
, to
);
305 XFS_STATS_ADD(mp
, xs_read_bytes
, ret
);
310 * Common pre-write limit and setup checks.
312 * Called with the iolocked held either shared and exclusive according to
313 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
314 * if called for a direct write beyond i_size.
317 xfs_file_write_checks(
319 struct iov_iter
*from
,
320 unsigned int *iolock
)
322 struct file
*file
= iocb
->ki_filp
;
323 struct inode
*inode
= file
->f_mapping
->host
;
324 struct xfs_inode
*ip
= XFS_I(inode
);
326 size_t count
= iov_iter_count(from
);
327 bool drained_dio
= false;
331 error
= generic_write_checks(iocb
, from
);
335 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
336 error
= break_layout(inode
, false);
337 if (error
== -EWOULDBLOCK
)
340 error
= xfs_break_layouts(inode
, iolock
, BREAK_WRITE
);
347 * For changing security info in file_remove_privs() we need i_rwsem
350 if (*iolock
== XFS_IOLOCK_SHARED
&& !IS_NOSEC(inode
)) {
351 xfs_iunlock(ip
, *iolock
);
352 *iolock
= XFS_IOLOCK_EXCL
;
353 error
= xfs_ilock_iocb(iocb
, *iolock
);
362 * If the offset is beyond the size of the file, we need to zero any
363 * blocks that fall between the existing EOF and the start of this
364 * write. If zeroing is needed and we are currently holding the iolock
365 * shared, we need to update it to exclusive which implies having to
366 * redo all checks before.
368 * We need to serialise against EOF updates that occur in IO completions
369 * here. We want to make sure that nobody is changing the size while we
370 * do this check until we have placed an IO barrier (i.e. hold the
371 * XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. The
372 * spinlock effectively forms a memory barrier once we have the
373 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value and
374 * hence be able to correctly determine if we need to run zeroing.
376 * We can do an unlocked check here safely as IO completion can only
377 * extend EOF. Truncate is locked out at this point, so the EOF can
378 * not move backwards, only forwards. Hence we only need to take the
379 * slow path and spin locks when we are at or beyond the current EOF.
381 if (iocb
->ki_pos
<= i_size_read(inode
))
384 spin_lock(&ip
->i_flags_lock
);
385 isize
= i_size_read(inode
);
386 if (iocb
->ki_pos
> isize
) {
387 spin_unlock(&ip
->i_flags_lock
);
389 if (iocb
->ki_flags
& IOCB_NOWAIT
)
393 if (*iolock
== XFS_IOLOCK_SHARED
) {
394 xfs_iunlock(ip
, *iolock
);
395 *iolock
= XFS_IOLOCK_EXCL
;
396 xfs_ilock(ip
, *iolock
);
397 iov_iter_reexpand(from
, count
);
400 * We now have an IO submission barrier in place, but
401 * AIO can do EOF updates during IO completion and hence
402 * we now need to wait for all of them to drain. Non-AIO
403 * DIO will have drained before we are given the
404 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
407 inode_dio_wait(inode
);
412 trace_xfs_zero_eof(ip
, isize
, iocb
->ki_pos
- isize
);
413 error
= xfs_zero_range(ip
, isize
, iocb
->ki_pos
- isize
, NULL
);
417 spin_unlock(&ip
->i_flags_lock
);
420 return kiocb_modified(iocb
);
424 xfs_dio_write_end_io(
430 struct inode
*inode
= file_inode(iocb
->ki_filp
);
431 struct xfs_inode
*ip
= XFS_I(inode
);
432 loff_t offset
= iocb
->ki_pos
;
433 unsigned int nofs_flag
;
435 trace_xfs_end_io_direct_write(ip
, offset
, size
);
437 if (xfs_is_shutdown(ip
->i_mount
))
446 * Capture amount written on completion as we can't reliably account
447 * for it on submission.
449 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, size
);
452 * We can allocate memory here while doing writeback on behalf of
453 * memory reclaim. To avoid memory allocation deadlocks set the
454 * task-wide nofs context for the following operations.
456 nofs_flag
= memalloc_nofs_save();
458 if (flags
& IOMAP_DIO_COW
) {
459 error
= xfs_reflink_end_cow(ip
, offset
, size
);
465 * Unwritten conversion updates the in-core isize after extent
466 * conversion but before updating the on-disk size. Updating isize any
467 * earlier allows a racing dio read to find unwritten extents before
468 * they are converted.
470 if (flags
& IOMAP_DIO_UNWRITTEN
) {
471 error
= xfs_iomap_write_unwritten(ip
, offset
, size
, true);
476 * We need to update the in-core inode size here so that we don't end up
477 * with the on-disk inode size being outside the in-core inode size. We
478 * have no other method of updating EOF for AIO, so always do it here
481 * We need to lock the test/set EOF update as we can be racing with
482 * other IO completions here to update the EOF. Failing to serialise
483 * here can result in EOF moving backwards and Bad Things Happen when
486 * As IO completion only ever extends EOF, we can do an unlocked check
487 * here to avoid taking the spinlock. If we land within the current EOF,
488 * then we do not need to do an extending update at all, and we don't
489 * need to take the lock to check this. If we race with an update moving
490 * EOF, then we'll either still be beyond EOF and need to take the lock,
491 * or we'll be within EOF and we don't need to take it at all.
493 if (offset
+ size
<= i_size_read(inode
))
496 spin_lock(&ip
->i_flags_lock
);
497 if (offset
+ size
> i_size_read(inode
)) {
498 i_size_write(inode
, offset
+ size
);
499 spin_unlock(&ip
->i_flags_lock
);
500 error
= xfs_setfilesize(ip
, offset
, size
);
502 spin_unlock(&ip
->i_flags_lock
);
506 memalloc_nofs_restore(nofs_flag
);
510 static const struct iomap_dio_ops xfs_dio_write_ops
= {
511 .end_io
= xfs_dio_write_end_io
,
515 * Handle block aligned direct I/O writes
517 static noinline ssize_t
518 xfs_file_dio_write_aligned(
519 struct xfs_inode
*ip
,
521 struct iov_iter
*from
)
523 unsigned int iolock
= XFS_IOLOCK_SHARED
;
526 ret
= xfs_ilock_iocb(iocb
, iolock
);
529 ret
= xfs_file_write_checks(iocb
, from
, &iolock
);
534 * We don't need to hold the IOLOCK exclusively across the IO, so demote
535 * the iolock back to shared if we had to take the exclusive lock in
536 * xfs_file_write_checks() for other reasons.
538 if (iolock
== XFS_IOLOCK_EXCL
) {
539 xfs_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
540 iolock
= XFS_IOLOCK_SHARED
;
542 trace_xfs_file_direct_write(iocb
, from
);
543 ret
= iomap_dio_rw(iocb
, from
, &xfs_direct_write_iomap_ops
,
544 &xfs_dio_write_ops
, 0, NULL
, 0);
547 xfs_iunlock(ip
, iolock
);
552 * Handle block unaligned direct I/O writes
554 * In most cases direct I/O writes will be done holding IOLOCK_SHARED, allowing
555 * them to be done in parallel with reads and other direct I/O writes. However,
556 * if the I/O is not aligned to filesystem blocks, the direct I/O layer may need
557 * to do sub-block zeroing and that requires serialisation against other direct
558 * I/O to the same block. In this case we need to serialise the submission of
559 * the unaligned I/O so that we don't get racing block zeroing in the dio layer.
560 * In the case where sub-block zeroing is not required, we can do concurrent
561 * sub-block dios to the same block successfully.
563 * Optimistically submit the I/O using the shared lock first, but use the
564 * IOMAP_DIO_OVERWRITE_ONLY flag to tell the lower layers to return -EAGAIN
565 * if block allocation or partial block zeroing would be required. In that case
566 * we try again with the exclusive lock.
568 static noinline ssize_t
569 xfs_file_dio_write_unaligned(
570 struct xfs_inode
*ip
,
572 struct iov_iter
*from
)
574 size_t isize
= i_size_read(VFS_I(ip
));
575 size_t count
= iov_iter_count(from
);
576 unsigned int iolock
= XFS_IOLOCK_SHARED
;
577 unsigned int flags
= IOMAP_DIO_OVERWRITE_ONLY
;
581 * Extending writes need exclusivity because of the sub-block zeroing
582 * that the DIO code always does for partial tail blocks beyond EOF, so
583 * don't even bother trying the fast path in this case.
585 if (iocb
->ki_pos
> isize
|| iocb
->ki_pos
+ count
>= isize
) {
586 if (iocb
->ki_flags
& IOCB_NOWAIT
)
589 iolock
= XFS_IOLOCK_EXCL
;
590 flags
= IOMAP_DIO_FORCE_WAIT
;
593 ret
= xfs_ilock_iocb(iocb
, iolock
);
598 * We can't properly handle unaligned direct I/O to reflink files yet,
599 * as we can't unshare a partial block.
601 if (xfs_is_cow_inode(ip
)) {
602 trace_xfs_reflink_bounce_dio_write(iocb
, from
);
607 ret
= xfs_file_write_checks(iocb
, from
, &iolock
);
612 * If we are doing exclusive unaligned I/O, this must be the only I/O
613 * in-flight. Otherwise we risk data corruption due to unwritten extent
614 * conversions from the AIO end_io handler. Wait for all other I/O to
617 if (flags
& IOMAP_DIO_FORCE_WAIT
)
618 inode_dio_wait(VFS_I(ip
));
620 trace_xfs_file_direct_write(iocb
, from
);
621 ret
= iomap_dio_rw(iocb
, from
, &xfs_direct_write_iomap_ops
,
622 &xfs_dio_write_ops
, flags
, NULL
, 0);
625 * Retry unaligned I/O with exclusive blocking semantics if the DIO
626 * layer rejected it for mapping or locking reasons. If we are doing
627 * nonblocking user I/O, propagate the error.
629 if (ret
== -EAGAIN
&& !(iocb
->ki_flags
& IOCB_NOWAIT
)) {
630 ASSERT(flags
& IOMAP_DIO_OVERWRITE_ONLY
);
631 xfs_iunlock(ip
, iolock
);
632 goto retry_exclusive
;
637 xfs_iunlock(ip
, iolock
);
644 struct iov_iter
*from
)
646 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
647 struct xfs_buftarg
*target
= xfs_inode_buftarg(ip
);
648 size_t count
= iov_iter_count(from
);
650 /* direct I/O must be aligned to device logical sector size */
651 if ((iocb
->ki_pos
| count
) & target
->bt_logical_sectormask
)
653 if ((iocb
->ki_pos
| count
) & ip
->i_mount
->m_blockmask
)
654 return xfs_file_dio_write_unaligned(ip
, iocb
, from
);
655 return xfs_file_dio_write_aligned(ip
, iocb
, from
);
658 static noinline ssize_t
661 struct iov_iter
*from
)
663 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
664 struct xfs_inode
*ip
= XFS_I(inode
);
665 unsigned int iolock
= XFS_IOLOCK_EXCL
;
666 ssize_t ret
, error
= 0;
669 ret
= xfs_ilock_iocb(iocb
, iolock
);
672 ret
= xfs_file_write_checks(iocb
, from
, &iolock
);
678 trace_xfs_file_dax_write(iocb
, from
);
679 ret
= dax_iomap_rw(iocb
, from
, &xfs_dax_write_iomap_ops
);
680 if (ret
> 0 && iocb
->ki_pos
> i_size_read(inode
)) {
681 i_size_write(inode
, iocb
->ki_pos
);
682 error
= xfs_setfilesize(ip
, pos
, ret
);
686 xfs_iunlock(ip
, iolock
);
691 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
693 /* Handle various SYNC-type writes */
694 ret
= generic_write_sync(iocb
, ret
);
700 xfs_file_buffered_write(
702 struct iov_iter
*from
)
704 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
705 struct xfs_inode
*ip
= XFS_I(inode
);
707 bool cleared_space
= false;
711 iolock
= XFS_IOLOCK_EXCL
;
712 ret
= xfs_ilock_iocb(iocb
, iolock
);
716 ret
= xfs_file_write_checks(iocb
, from
, &iolock
);
720 /* We can write back this queue in page reclaim */
721 current
->backing_dev_info
= inode_to_bdi(inode
);
723 trace_xfs_file_buffered_write(iocb
, from
);
724 ret
= iomap_file_buffered_write(iocb
, from
,
725 &xfs_buffered_write_iomap_ops
);
726 if (likely(ret
>= 0))
730 * If we hit a space limit, try to free up some lingering preallocated
731 * space before returning an error. In the case of ENOSPC, first try to
732 * write back all dirty inodes to free up some of the excess reserved
733 * metadata space. This reduces the chances that the eofblocks scan
734 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
735 * also behaves as a filter to prevent too many eofblocks scans from
736 * running at the same time. Use a synchronous scan to increase the
737 * effectiveness of the scan.
739 if (ret
== -EDQUOT
&& !cleared_space
) {
740 xfs_iunlock(ip
, iolock
);
741 xfs_blockgc_free_quota(ip
, XFS_ICWALK_FLAG_SYNC
);
742 cleared_space
= true;
744 } else if (ret
== -ENOSPC
&& !cleared_space
) {
745 struct xfs_icwalk icw
= {0};
747 cleared_space
= true;
748 xfs_flush_inodes(ip
->i_mount
);
750 xfs_iunlock(ip
, iolock
);
751 icw
.icw_flags
= XFS_ICWALK_FLAG_SYNC
;
752 xfs_blockgc_free_space(ip
->i_mount
, &icw
);
756 current
->backing_dev_info
= NULL
;
759 xfs_iunlock(ip
, iolock
);
762 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
763 /* Handle various SYNC-type writes */
764 ret
= generic_write_sync(iocb
, ret
);
772 struct iov_iter
*from
)
774 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
775 struct xfs_inode
*ip
= XFS_I(inode
);
777 size_t ocount
= iov_iter_count(from
);
779 XFS_STATS_INC(ip
->i_mount
, xs_write_calls
);
784 if (xfs_is_shutdown(ip
->i_mount
))
788 return xfs_file_dax_write(iocb
, from
);
790 if (iocb
->ki_flags
& IOCB_DIRECT
) {
792 * Allow a directio write to fall back to a buffered
793 * write *only* in the case that we're doing a reflink
794 * CoW. In all other directio scenarios we do not
795 * allow an operation to fall back to buffered mode.
797 ret
= xfs_file_dio_write(iocb
, from
);
802 return xfs_file_buffered_write(iocb
, from
);
809 struct xfs_inode
*ip
= XFS_I(inode
);
811 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
813 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
817 xfs_break_dax_layouts(
823 ASSERT(xfs_isilocked(XFS_I(inode
), XFS_MMAPLOCK_EXCL
));
825 page
= dax_layout_busy_page(inode
->i_mapping
);
830 return ___wait_var_event(&page
->_refcount
,
831 atomic_read(&page
->_refcount
) == 1, TASK_INTERRUPTIBLE
,
832 0, 0, xfs_wait_dax_page(inode
));
839 enum layout_break_reason reason
)
844 ASSERT(xfs_isilocked(XFS_I(inode
), XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
));
850 error
= xfs_break_dax_layouts(inode
, &retry
);
855 error
= xfs_break_leased_layouts(inode
, iolock
, &retry
);
861 } while (error
== 0 && retry
);
866 /* Does this file, inode, or mount want synchronous writes? */
867 static inline bool xfs_file_sync_writes(struct file
*filp
)
869 struct xfs_inode
*ip
= XFS_I(file_inode(filp
));
871 if (xfs_has_wsync(ip
->i_mount
))
873 if (filp
->f_flags
& (__O_SYNC
| O_DSYNC
))
875 if (IS_SYNC(file_inode(filp
)))
881 #define XFS_FALLOC_FL_SUPPORTED \
882 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
883 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
884 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
893 struct inode
*inode
= file_inode(file
);
894 struct xfs_inode
*ip
= XFS_I(inode
);
896 uint iolock
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
898 bool do_file_insert
= false;
900 if (!S_ISREG(inode
->i_mode
))
902 if (mode
& ~XFS_FALLOC_FL_SUPPORTED
)
905 xfs_ilock(ip
, iolock
);
906 error
= xfs_break_layouts(inode
, &iolock
, BREAK_UNMAP
);
911 * Must wait for all AIO to complete before we continue as AIO can
912 * change the file size on completion without holding any locks we
913 * currently hold. We must do this first because AIO can update both
914 * the on disk and in memory inode sizes, and the operations that follow
915 * require the in-memory size to be fully up-to-date.
917 inode_dio_wait(inode
);
920 * Now AIO and DIO has drained we flush and (if necessary) invalidate
921 * the cached range over the first operation we are about to run.
923 * We care about zero and collapse here because they both run a hole
924 * punch over the range first. Because that can zero data, and the range
925 * of invalidation for the shift operations is much larger, we still do
926 * the required flush for collapse in xfs_prepare_shift().
928 * Insert has the same range requirements as collapse, and we extend the
929 * file first which can zero data. Hence insert has the same
930 * flush/invalidate requirements as collapse and so they are both
931 * handled at the right time by xfs_prepare_shift().
933 if (mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
|
934 FALLOC_FL_COLLAPSE_RANGE
)) {
935 error
= xfs_flush_unmap_range(ip
, offset
, len
);
940 error
= file_modified(file
);
944 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
945 error
= xfs_free_file_space(ip
, offset
, len
);
948 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
949 if (!xfs_is_falloc_aligned(ip
, offset
, len
)) {
955 * There is no need to overlap collapse range with EOF,
956 * in which case it is effectively a truncate operation
958 if (offset
+ len
>= i_size_read(inode
)) {
963 new_size
= i_size_read(inode
) - len
;
965 error
= xfs_collapse_file_space(ip
, offset
, len
);
968 } else if (mode
& FALLOC_FL_INSERT_RANGE
) {
969 loff_t isize
= i_size_read(inode
);
971 if (!xfs_is_falloc_aligned(ip
, offset
, len
)) {
977 * New inode size must not exceed ->s_maxbytes, accounting for
978 * possible signed overflow.
980 if (inode
->i_sb
->s_maxbytes
- isize
< len
) {
984 new_size
= isize
+ len
;
986 /* Offset should be less than i_size */
987 if (offset
>= isize
) {
991 do_file_insert
= true;
993 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
994 offset
+ len
> i_size_read(inode
)) {
995 new_size
= offset
+ len
;
996 error
= inode_newsize_ok(inode
, new_size
);
1001 if (mode
& FALLOC_FL_ZERO_RANGE
) {
1003 * Punch a hole and prealloc the range. We use a hole
1004 * punch rather than unwritten extent conversion for two
1007 * 1.) Hole punch handles partial block zeroing for us.
1008 * 2.) If prealloc returns ENOSPC, the file range is
1009 * still zero-valued by virtue of the hole punch.
1011 unsigned int blksize
= i_blocksize(inode
);
1013 trace_xfs_zero_file_space(ip
);
1015 error
= xfs_free_file_space(ip
, offset
, len
);
1019 len
= round_up(offset
+ len
, blksize
) -
1020 round_down(offset
, blksize
);
1021 offset
= round_down(offset
, blksize
);
1022 } else if (mode
& FALLOC_FL_UNSHARE_RANGE
) {
1023 error
= xfs_reflink_unshare(ip
, offset
, len
);
1028 * If always_cow mode we can't use preallocations and
1029 * thus should not create them.
1031 if (xfs_is_always_cow_inode(ip
)) {
1032 error
= -EOPNOTSUPP
;
1037 if (!xfs_is_always_cow_inode(ip
)) {
1038 error
= xfs_alloc_file_space(ip
, offset
, len
);
1044 /* Change file size if needed */
1048 iattr
.ia_valid
= ATTR_SIZE
;
1049 iattr
.ia_size
= new_size
;
1050 error
= xfs_vn_setattr_size(file_mnt_idmap(file
),
1051 file_dentry(file
), &iattr
);
1057 * Perform hole insertion now that the file size has been
1058 * updated so that if we crash during the operation we don't
1059 * leave shifted extents past EOF and hence losing access to
1060 * the data that is contained within them.
1062 if (do_file_insert
) {
1063 error
= xfs_insert_file_space(ip
, offset
, len
);
1068 if (xfs_file_sync_writes(file
))
1069 error
= xfs_log_force_inode(ip
);
1072 xfs_iunlock(ip
, iolock
);
1083 struct xfs_inode
*ip
= XFS_I(file_inode(file
));
1088 * Operations creating pages in page cache need protection from hole
1089 * punching and similar ops
1091 if (advice
== POSIX_FADV_WILLNEED
) {
1092 lockflags
= XFS_IOLOCK_SHARED
;
1093 xfs_ilock(ip
, lockflags
);
1095 ret
= generic_fadvise(file
, start
, end
, advice
);
1097 xfs_iunlock(ip
, lockflags
);
1102 xfs_file_remap_range(
1103 struct file
*file_in
,
1105 struct file
*file_out
,
1108 unsigned int remap_flags
)
1110 struct inode
*inode_in
= file_inode(file_in
);
1111 struct xfs_inode
*src
= XFS_I(inode_in
);
1112 struct inode
*inode_out
= file_inode(file_out
);
1113 struct xfs_inode
*dest
= XFS_I(inode_out
);
1114 struct xfs_mount
*mp
= src
->i_mount
;
1115 loff_t remapped
= 0;
1116 xfs_extlen_t cowextsize
;
1119 if (remap_flags
& ~(REMAP_FILE_DEDUP
| REMAP_FILE_ADVISORY
))
1122 if (!xfs_has_reflink(mp
))
1125 if (xfs_is_shutdown(mp
))
1128 /* Prepare and then clone file data. */
1129 ret
= xfs_reflink_remap_prep(file_in
, pos_in
, file_out
, pos_out
,
1131 if (ret
|| len
== 0)
1134 trace_xfs_reflink_remap_range(src
, pos_in
, len
, dest
, pos_out
);
1136 ret
= xfs_reflink_remap_blocks(src
, pos_in
, dest
, pos_out
, len
,
1142 * Carry the cowextsize hint from src to dest if we're sharing the
1143 * entire source file to the entire destination file, the source file
1144 * has a cowextsize hint, and the destination file does not.
1147 if (pos_in
== 0 && len
== i_size_read(inode_in
) &&
1148 (src
->i_diflags2
& XFS_DIFLAG2_COWEXTSIZE
) &&
1149 pos_out
== 0 && len
>= i_size_read(inode_out
) &&
1150 !(dest
->i_diflags2
& XFS_DIFLAG2_COWEXTSIZE
))
1151 cowextsize
= src
->i_cowextsize
;
1153 ret
= xfs_reflink_update_dest(dest
, pos_out
+ len
, cowextsize
,
1158 if (xfs_file_sync_writes(file_in
) || xfs_file_sync_writes(file_out
))
1159 xfs_log_force_inode(dest
);
1161 xfs_iunlock2_io_mmap(src
, dest
);
1163 trace_xfs_reflink_remap_range_error(dest
, ret
, _RET_IP_
);
1164 return remapped
> 0 ? remapped
: ret
;
1169 struct inode
*inode
,
1172 if (xfs_is_shutdown(XFS_M(inode
->i_sb
)))
1174 file
->f_mode
|= FMODE_NOWAIT
| FMODE_BUF_RASYNC
| FMODE_BUF_WASYNC
|
1175 FMODE_DIO_PARALLEL_WRITE
;
1176 return generic_file_open(inode
, file
);
1181 struct inode
*inode
,
1184 struct xfs_inode
*ip
= XFS_I(inode
);
1188 error
= xfs_file_open(inode
, file
);
1193 * If there are any blocks, read-ahead block 0 as we're almost
1194 * certain to have the next operation be a read there.
1196 mode
= xfs_ilock_data_map_shared(ip
);
1197 if (ip
->i_df
.if_nextents
> 0)
1198 error
= xfs_dir3_data_readahead(ip
, 0, 0);
1199 xfs_iunlock(ip
, mode
);
1205 struct inode
*inode
,
1208 return xfs_release(XFS_I(inode
));
1214 struct dir_context
*ctx
)
1216 struct inode
*inode
= file_inode(file
);
1217 xfs_inode_t
*ip
= XFS_I(inode
);
1221 * The Linux API doesn't pass down the total size of the buffer
1222 * we read into down to the filesystem. With the filldir concept
1223 * it's not needed for correct information, but the XFS dir2 leaf
1224 * code wants an estimate of the buffer size to calculate it's
1225 * readahead window and size the buffers used for mapping to
1228 * Try to give it an estimate that's good enough, maybe at some
1229 * point we can change the ->readdir prototype to include the
1230 * buffer size. For now we use the current glibc buffer size.
1232 bufsize
= (size_t)min_t(loff_t
, XFS_READDIR_BUFSIZE
, ip
->i_disk_size
);
1234 return xfs_readdir(NULL
, ip
, ctx
, bufsize
);
1243 struct inode
*inode
= file
->f_mapping
->host
;
1245 if (xfs_is_shutdown(XFS_I(inode
)->i_mount
))
1250 return generic_file_llseek(file
, offset
, whence
);
1252 offset
= iomap_seek_hole(inode
, offset
, &xfs_seek_iomap_ops
);
1255 offset
= iomap_seek_data(inode
, offset
, &xfs_seek_iomap_ops
);
1261 return vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1264 #ifdef CONFIG_FS_DAX
1265 static inline vm_fault_t
1267 struct vm_fault
*vmf
,
1268 enum page_entry_size pe_size
,
1272 return dax_iomap_fault(vmf
, pe_size
, pfn
, NULL
,
1273 (write_fault
&& !vmf
->cow_page
) ?
1274 &xfs_dax_write_iomap_ops
:
1275 &xfs_read_iomap_ops
);
1278 static inline vm_fault_t
1280 struct vm_fault
*vmf
,
1281 enum page_entry_size pe_size
,
1286 return VM_FAULT_SIGBUS
;
1291 * Locking for serialisation of IO during page faults. This results in a lock
1295 * sb_start_pagefault(vfs, freeze)
1296 * invalidate_lock (vfs/XFS_MMAPLOCK - truncate serialisation)
1298 * i_lock (XFS - extent map serialisation)
1301 __xfs_filemap_fault(
1302 struct vm_fault
*vmf
,
1303 enum page_entry_size pe_size
,
1306 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1307 struct xfs_inode
*ip
= XFS_I(inode
);
1310 trace_xfs_filemap_fault(ip
, pe_size
, write_fault
);
1313 sb_start_pagefault(inode
->i_sb
);
1314 file_update_time(vmf
->vma
->vm_file
);
1317 if (IS_DAX(inode
)) {
1320 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1321 ret
= xfs_dax_fault(vmf
, pe_size
, write_fault
, &pfn
);
1322 if (ret
& VM_FAULT_NEEDDSYNC
)
1323 ret
= dax_finish_sync_fault(vmf
, pe_size
, pfn
);
1324 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1327 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1328 ret
= iomap_page_mkwrite(vmf
,
1329 &xfs_page_mkwrite_iomap_ops
);
1330 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1332 ret
= filemap_fault(vmf
);
1337 sb_end_pagefault(inode
->i_sb
);
1343 struct vm_fault
*vmf
)
1345 return (vmf
->flags
& FAULT_FLAG_WRITE
) &&
1346 (vmf
->vma
->vm_flags
& VM_SHARED
);
1351 struct vm_fault
*vmf
)
1353 /* DAX can shortcut the normal fault path on write faults! */
1354 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
,
1355 IS_DAX(file_inode(vmf
->vma
->vm_file
)) &&
1356 xfs_is_write_fault(vmf
));
1360 xfs_filemap_huge_fault(
1361 struct vm_fault
*vmf
,
1362 enum page_entry_size pe_size
)
1364 if (!IS_DAX(file_inode(vmf
->vma
->vm_file
)))
1365 return VM_FAULT_FALLBACK
;
1367 /* DAX can shortcut the normal fault path on write faults! */
1368 return __xfs_filemap_fault(vmf
, pe_size
,
1369 xfs_is_write_fault(vmf
));
1373 xfs_filemap_page_mkwrite(
1374 struct vm_fault
*vmf
)
1376 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
, true);
1380 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1381 * on write faults. In reality, it needs to serialise against truncate and
1382 * prepare memory for writing so handle is as standard write fault.
1385 xfs_filemap_pfn_mkwrite(
1386 struct vm_fault
*vmf
)
1389 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
, true);
1392 static const struct vm_operations_struct xfs_file_vm_ops
= {
1393 .fault
= xfs_filemap_fault
,
1394 .huge_fault
= xfs_filemap_huge_fault
,
1395 .map_pages
= filemap_map_pages
,
1396 .page_mkwrite
= xfs_filemap_page_mkwrite
,
1397 .pfn_mkwrite
= xfs_filemap_pfn_mkwrite
,
1403 struct vm_area_struct
*vma
)
1405 struct inode
*inode
= file_inode(file
);
1406 struct xfs_buftarg
*target
= xfs_inode_buftarg(XFS_I(inode
));
1409 * We don't support synchronous mappings for non-DAX files and
1410 * for DAX files if underneath dax_device is not synchronous.
1412 if (!daxdev_mapping_supported(vma
, target
->bt_daxdev
))
1415 file_accessed(file
);
1416 vma
->vm_ops
= &xfs_file_vm_ops
;
1418 vm_flags_set(vma
, VM_HUGEPAGE
);
1422 const struct file_operations xfs_file_operations
= {
1423 .llseek
= xfs_file_llseek
,
1424 .read_iter
= xfs_file_read_iter
,
1425 .write_iter
= xfs_file_write_iter
,
1426 .splice_read
= generic_file_splice_read
,
1427 .splice_write
= iter_file_splice_write
,
1428 .iopoll
= iocb_bio_iopoll
,
1429 .unlocked_ioctl
= xfs_file_ioctl
,
1430 #ifdef CONFIG_COMPAT
1431 .compat_ioctl
= xfs_file_compat_ioctl
,
1433 .mmap
= xfs_file_mmap
,
1434 .mmap_supported_flags
= MAP_SYNC
,
1435 .open
= xfs_file_open
,
1436 .release
= xfs_file_release
,
1437 .fsync
= xfs_file_fsync
,
1438 .get_unmapped_area
= thp_get_unmapped_area
,
1439 .fallocate
= xfs_file_fallocate
,
1440 .fadvise
= xfs_file_fadvise
,
1441 .remap_file_range
= xfs_file_remap_range
,
1444 const struct file_operations xfs_dir_file_operations
= {
1445 .open
= xfs_dir_open
,
1446 .read
= generic_read_dir
,
1447 .iterate_shared
= xfs_file_readdir
,
1448 .llseek
= generic_file_llseek
,
1449 .unlocked_ioctl
= xfs_file_ioctl
,
1450 #ifdef CONFIG_COMPAT
1451 .compat_ioctl
= xfs_file_compat_ioctl
,
1453 .fsync
= xfs_dir_fsync
,