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/falloc.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mman.h>
31 #include <linux/fadvise.h>
33 static const struct vm_operations_struct xfs_file_vm_ops
;
36 xfs_update_prealloc_flags(
38 enum xfs_prealloc_flags flags
)
43 error
= xfs_trans_alloc(ip
->i_mount
, &M_RES(ip
->i_mount
)->tr_writeid
,
48 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
49 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
51 if (!(flags
& XFS_PREALLOC_INVISIBLE
)) {
52 VFS_I(ip
)->i_mode
&= ~S_ISUID
;
53 if (VFS_I(ip
)->i_mode
& S_IXGRP
)
54 VFS_I(ip
)->i_mode
&= ~S_ISGID
;
55 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
58 if (flags
& XFS_PREALLOC_SET
)
59 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
60 if (flags
& XFS_PREALLOC_CLEAR
)
61 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
63 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
64 if (flags
& XFS_PREALLOC_SYNC
)
65 xfs_trans_set_sync(tp
);
66 return xfs_trans_commit(tp
);
70 * Fsync operations on directories are much simpler than on regular files,
71 * as there is no file data to flush, and thus also no need for explicit
72 * cache flush operations, and there are no non-transaction metadata updates
73 * on directories either.
82 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
83 struct xfs_mount
*mp
= ip
->i_mount
;
86 trace_xfs_dir_fsync(ip
);
88 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
89 if (xfs_ipincount(ip
))
90 lsn
= ip
->i_itemp
->ili_last_lsn
;
91 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
95 return xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, NULL
);
105 struct inode
*inode
= file
->f_mapping
->host
;
106 struct xfs_inode
*ip
= XFS_I(inode
);
107 struct xfs_mount
*mp
= ip
->i_mount
;
112 trace_xfs_file_fsync(ip
);
114 error
= file_write_and_wait_range(file
, start
, end
);
118 if (XFS_FORCED_SHUTDOWN(mp
))
121 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
124 * If we have an RT and/or log subvolume we need to make sure to flush
125 * the write cache the device used for file data first. This is to
126 * ensure newly written file data make it to disk before logging the new
127 * inode size in case of an extending write.
129 if (XFS_IS_REALTIME_INODE(ip
))
130 xfs_blkdev_issue_flush(mp
->m_rtdev_targp
);
131 else if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
132 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
135 * All metadata updates are logged, which means that we just have to
136 * flush the log up to the latest LSN that touched the inode. If we have
137 * concurrent fsync/fdatasync() calls, we need them to all block on the
138 * log force before we clear the ili_fsync_fields field. This ensures
139 * that we don't get a racing sync operation that does not wait for the
140 * metadata to hit the journal before returning. If we race with
141 * clearing the ili_fsync_fields, then all that will happen is the log
142 * force will do nothing as the lsn will already be on disk. We can't
143 * race with setting ili_fsync_fields because that is done under
144 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
145 * until after the ili_fsync_fields is cleared.
147 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
148 if (xfs_ipincount(ip
)) {
150 (ip
->i_itemp
->ili_fsync_fields
& ~XFS_ILOG_TIMESTAMP
))
151 lsn
= ip
->i_itemp
->ili_last_lsn
;
155 error
= xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, &log_flushed
);
156 ip
->i_itemp
->ili_fsync_fields
= 0;
158 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
161 * If we only have a single device, and the log force about was
162 * a no-op we might have to flush the data device cache here.
163 * This can only happen for fdatasync/O_DSYNC if we were overwriting
164 * an already allocated file and thus do not have any metadata to
167 if (!log_flushed
&& !XFS_IS_REALTIME_INODE(ip
) &&
168 mp
->m_logdev_targp
== mp
->m_ddev_targp
)
169 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
175 xfs_file_dio_aio_read(
179 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
180 size_t count
= iov_iter_count(to
);
183 trace_xfs_file_direct_read(ip
, count
, iocb
->ki_pos
);
186 return 0; /* skip atime */
188 file_accessed(iocb
->ki_filp
);
190 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
191 ret
= iomap_dio_rw(iocb
, to
, &xfs_read_iomap_ops
, NULL
,
192 is_sync_kiocb(iocb
));
193 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
198 static noinline ssize_t
203 struct xfs_inode
*ip
= XFS_I(iocb
->ki_filp
->f_mapping
->host
);
204 size_t count
= iov_iter_count(to
);
207 trace_xfs_file_dax_read(ip
, count
, iocb
->ki_pos
);
210 return 0; /* skip atime */
212 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
213 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
))
216 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
219 ret
= dax_iomap_rw(iocb
, to
, &xfs_read_iomap_ops
);
220 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
222 file_accessed(iocb
->ki_filp
);
227 xfs_file_buffered_aio_read(
231 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
234 trace_xfs_file_buffered_read(ip
, iov_iter_count(to
), iocb
->ki_pos
);
236 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
237 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
))
240 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
242 ret
= generic_file_read_iter(iocb
, to
);
243 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
253 struct inode
*inode
= file_inode(iocb
->ki_filp
);
254 struct xfs_mount
*mp
= XFS_I(inode
)->i_mount
;
257 XFS_STATS_INC(mp
, xs_read_calls
);
259 if (XFS_FORCED_SHUTDOWN(mp
))
263 ret
= xfs_file_dax_read(iocb
, to
);
264 else if (iocb
->ki_flags
& IOCB_DIRECT
)
265 ret
= xfs_file_dio_aio_read(iocb
, to
);
267 ret
= xfs_file_buffered_aio_read(iocb
, to
);
270 XFS_STATS_ADD(mp
, xs_read_bytes
, ret
);
275 * Common pre-write limit and setup checks.
277 * Called with the iolocked held either shared and exclusive according to
278 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
279 * if called for a direct write beyond i_size.
282 xfs_file_aio_write_checks(
284 struct iov_iter
*from
,
287 struct file
*file
= iocb
->ki_filp
;
288 struct inode
*inode
= file
->f_mapping
->host
;
289 struct xfs_inode
*ip
= XFS_I(inode
);
291 size_t count
= iov_iter_count(from
);
292 bool drained_dio
= false;
296 error
= generic_write_checks(iocb
, from
);
300 error
= xfs_break_layouts(inode
, iolock
, BREAK_WRITE
);
305 * For changing security info in file_remove_privs() we need i_rwsem
308 if (*iolock
== XFS_IOLOCK_SHARED
&& !IS_NOSEC(inode
)) {
309 xfs_iunlock(ip
, *iolock
);
310 *iolock
= XFS_IOLOCK_EXCL
;
311 xfs_ilock(ip
, *iolock
);
315 * If the offset is beyond the size of the file, we need to zero any
316 * blocks that fall between the existing EOF and the start of this
317 * write. If zeroing is needed and we are currently holding the
318 * iolock shared, we need to update it to exclusive which implies
319 * having to redo all checks before.
321 * We need to serialise against EOF updates that occur in IO
322 * completions here. We want to make sure that nobody is changing the
323 * size while we do this check until we have placed an IO barrier (i.e.
324 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
325 * The spinlock effectively forms a memory barrier once we have the
326 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
327 * and hence be able to correctly determine if we need to run zeroing.
329 spin_lock(&ip
->i_flags_lock
);
330 isize
= i_size_read(inode
);
331 if (iocb
->ki_pos
> isize
) {
332 spin_unlock(&ip
->i_flags_lock
);
334 if (*iolock
== XFS_IOLOCK_SHARED
) {
335 xfs_iunlock(ip
, *iolock
);
336 *iolock
= XFS_IOLOCK_EXCL
;
337 xfs_ilock(ip
, *iolock
);
338 iov_iter_reexpand(from
, count
);
341 * We now have an IO submission barrier in place, but
342 * AIO can do EOF updates during IO completion and hence
343 * we now need to wait for all of them to drain. Non-AIO
344 * DIO will have drained before we are given the
345 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
348 inode_dio_wait(inode
);
353 trace_xfs_zero_eof(ip
, isize
, iocb
->ki_pos
- isize
);
354 error
= iomap_zero_range(inode
, isize
, iocb
->ki_pos
- isize
,
355 NULL
, &xfs_buffered_write_iomap_ops
);
359 spin_unlock(&ip
->i_flags_lock
);
362 * Updating the timestamps will grab the ilock again from
363 * xfs_fs_dirty_inode, so we have to call it after dropping the
364 * lock above. Eventually we should look into a way to avoid
365 * the pointless lock roundtrip.
367 return file_modified(file
);
371 xfs_dio_write_end_io(
377 struct inode
*inode
= file_inode(iocb
->ki_filp
);
378 struct xfs_inode
*ip
= XFS_I(inode
);
379 loff_t offset
= iocb
->ki_pos
;
380 unsigned int nofs_flag
;
382 trace_xfs_end_io_direct_write(ip
, offset
, size
);
384 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
393 * Capture amount written on completion as we can't reliably account
394 * for it on submission.
396 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, size
);
399 * We can allocate memory here while doing writeback on behalf of
400 * memory reclaim. To avoid memory allocation deadlocks set the
401 * task-wide nofs context for the following operations.
403 nofs_flag
= memalloc_nofs_save();
405 if (flags
& IOMAP_DIO_COW
) {
406 error
= xfs_reflink_end_cow(ip
, offset
, size
);
412 * Unwritten conversion updates the in-core isize after extent
413 * conversion but before updating the on-disk size. Updating isize any
414 * earlier allows a racing dio read to find unwritten extents before
415 * they are converted.
417 if (flags
& IOMAP_DIO_UNWRITTEN
) {
418 error
= xfs_iomap_write_unwritten(ip
, offset
, size
, true);
423 * We need to update the in-core inode size here so that we don't end up
424 * with the on-disk inode size being outside the in-core inode size. We
425 * have no other method of updating EOF for AIO, so always do it here
428 * We need to lock the test/set EOF update as we can be racing with
429 * other IO completions here to update the EOF. Failing to serialise
430 * here can result in EOF moving backwards and Bad Things Happen when
433 spin_lock(&ip
->i_flags_lock
);
434 if (offset
+ size
> i_size_read(inode
)) {
435 i_size_write(inode
, offset
+ size
);
436 spin_unlock(&ip
->i_flags_lock
);
437 error
= xfs_setfilesize(ip
, offset
, size
);
439 spin_unlock(&ip
->i_flags_lock
);
443 memalloc_nofs_restore(nofs_flag
);
447 static const struct iomap_dio_ops xfs_dio_write_ops
= {
448 .end_io
= xfs_dio_write_end_io
,
452 * xfs_file_dio_aio_write - handle direct IO writes
454 * Lock the inode appropriately to prepare for and issue a direct IO write.
455 * By separating it from the buffered write path we remove all the tricky to
456 * follow locking changes and looping.
458 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
459 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
460 * pages are flushed out.
462 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
463 * allowing them to be done in parallel with reads and other direct IO writes.
464 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
465 * needs to do sub-block zeroing and that requires serialisation against other
466 * direct IOs to the same block. In this case we need to serialise the
467 * submission of the unaligned IOs so that we don't get racing block zeroing in
468 * the dio layer. To avoid the problem with aio, we also need to wait for
469 * outstanding IOs to complete so that unwritten extent conversion is completed
470 * before we try to map the overlapping block. This is currently implemented by
471 * hitting it with a big hammer (i.e. inode_dio_wait()).
473 * Returns with locks held indicated by @iolock and errors indicated by
474 * negative return values.
477 xfs_file_dio_aio_write(
479 struct iov_iter
*from
)
481 struct file
*file
= iocb
->ki_filp
;
482 struct address_space
*mapping
= file
->f_mapping
;
483 struct inode
*inode
= mapping
->host
;
484 struct xfs_inode
*ip
= XFS_I(inode
);
485 struct xfs_mount
*mp
= ip
->i_mount
;
487 int unaligned_io
= 0;
489 size_t count
= iov_iter_count(from
);
490 struct xfs_buftarg
*target
= xfs_inode_buftarg(ip
);
492 /* DIO must be aligned to device logical sector size */
493 if ((iocb
->ki_pos
| count
) & target
->bt_logical_sectormask
)
497 * Don't take the exclusive iolock here unless the I/O is unaligned to
498 * the file system block size. We don't need to consider the EOF
499 * extension case here because xfs_file_aio_write_checks() will relock
500 * the inode as necessary for EOF zeroing cases and fill out the new
501 * inode size as appropriate.
503 if ((iocb
->ki_pos
& mp
->m_blockmask
) ||
504 ((iocb
->ki_pos
+ count
) & mp
->m_blockmask
)) {
508 * We can't properly handle unaligned direct I/O to reflink
509 * files yet, as we can't unshare a partial block.
511 if (xfs_is_cow_inode(ip
)) {
512 trace_xfs_reflink_bounce_dio_write(ip
, iocb
->ki_pos
, count
);
515 iolock
= XFS_IOLOCK_EXCL
;
517 iolock
= XFS_IOLOCK_SHARED
;
520 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
521 /* unaligned dio always waits, bail */
524 if (!xfs_ilock_nowait(ip
, iolock
))
527 xfs_ilock(ip
, iolock
);
530 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
533 count
= iov_iter_count(from
);
536 * If we are doing unaligned IO, we can't allow any other overlapping IO
537 * in-flight at the same time or we risk data corruption. Wait for all
538 * other IO to drain before we submit. If the IO is aligned, demote the
539 * iolock if we had to take the exclusive lock in
540 * xfs_file_aio_write_checks() for other reasons.
543 inode_dio_wait(inode
);
544 } else if (iolock
== XFS_IOLOCK_EXCL
) {
545 xfs_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
546 iolock
= XFS_IOLOCK_SHARED
;
549 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
);
551 * If unaligned, this is the only IO in-flight. Wait on it before we
552 * release the iolock to prevent subsequent overlapping IO.
554 ret
= iomap_dio_rw(iocb
, from
, &xfs_direct_write_iomap_ops
,
556 is_sync_kiocb(iocb
) || unaligned_io
);
558 xfs_iunlock(ip
, iolock
);
561 * No fallback to buffered IO on errors for XFS, direct IO will either
562 * complete fully or fail.
564 ASSERT(ret
< 0 || ret
== count
);
568 static noinline ssize_t
571 struct iov_iter
*from
)
573 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
574 struct xfs_inode
*ip
= XFS_I(inode
);
575 int iolock
= XFS_IOLOCK_EXCL
;
576 ssize_t ret
, error
= 0;
580 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
581 if (!xfs_ilock_nowait(ip
, iolock
))
584 xfs_ilock(ip
, iolock
);
587 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
592 count
= iov_iter_count(from
);
594 trace_xfs_file_dax_write(ip
, count
, pos
);
595 ret
= dax_iomap_rw(iocb
, from
, &xfs_direct_write_iomap_ops
);
596 if (ret
> 0 && iocb
->ki_pos
> i_size_read(inode
)) {
597 i_size_write(inode
, iocb
->ki_pos
);
598 error
= xfs_setfilesize(ip
, pos
, ret
);
601 xfs_iunlock(ip
, iolock
);
606 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
608 /* Handle various SYNC-type writes */
609 ret
= generic_write_sync(iocb
, ret
);
615 xfs_file_buffered_aio_write(
617 struct iov_iter
*from
)
619 struct file
*file
= iocb
->ki_filp
;
620 struct address_space
*mapping
= file
->f_mapping
;
621 struct inode
*inode
= mapping
->host
;
622 struct xfs_inode
*ip
= XFS_I(inode
);
627 if (iocb
->ki_flags
& IOCB_NOWAIT
)
631 iolock
= XFS_IOLOCK_EXCL
;
632 xfs_ilock(ip
, iolock
);
634 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
638 /* We can write back this queue in page reclaim */
639 current
->backing_dev_info
= inode_to_bdi(inode
);
641 trace_xfs_file_buffered_write(ip
, iov_iter_count(from
), iocb
->ki_pos
);
642 ret
= iomap_file_buffered_write(iocb
, from
,
643 &xfs_buffered_write_iomap_ops
);
644 if (likely(ret
>= 0))
648 * If we hit a space limit, try to free up some lingering preallocated
649 * space before returning an error. In the case of ENOSPC, first try to
650 * write back all dirty inodes to free up some of the excess reserved
651 * metadata space. This reduces the chances that the eofblocks scan
652 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
653 * also behaves as a filter to prevent too many eofblocks scans from
654 * running at the same time.
656 if (ret
== -EDQUOT
&& !enospc
) {
657 xfs_iunlock(ip
, iolock
);
658 enospc
= xfs_inode_free_quota_eofblocks(ip
);
661 enospc
= xfs_inode_free_quota_cowblocks(ip
);
665 } else if (ret
== -ENOSPC
&& !enospc
) {
666 struct xfs_eofblocks eofb
= {0};
669 xfs_flush_inodes(ip
->i_mount
);
671 xfs_iunlock(ip
, iolock
);
672 eofb
.eof_flags
= XFS_EOF_FLAGS_SYNC
;
673 xfs_icache_free_eofblocks(ip
->i_mount
, &eofb
);
674 xfs_icache_free_cowblocks(ip
->i_mount
, &eofb
);
678 current
->backing_dev_info
= NULL
;
681 xfs_iunlock(ip
, iolock
);
684 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
685 /* Handle various SYNC-type writes */
686 ret
= generic_write_sync(iocb
, ret
);
694 struct iov_iter
*from
)
696 struct file
*file
= iocb
->ki_filp
;
697 struct address_space
*mapping
= file
->f_mapping
;
698 struct inode
*inode
= mapping
->host
;
699 struct xfs_inode
*ip
= XFS_I(inode
);
701 size_t ocount
= iov_iter_count(from
);
703 XFS_STATS_INC(ip
->i_mount
, xs_write_calls
);
708 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
712 return xfs_file_dax_write(iocb
, from
);
714 if (iocb
->ki_flags
& IOCB_DIRECT
) {
716 * Allow a directio write to fall back to a buffered
717 * write *only* in the case that we're doing a reflink
718 * CoW. In all other directio scenarios we do not
719 * allow an operation to fall back to buffered mode.
721 ret
= xfs_file_dio_aio_write(iocb
, from
);
726 return xfs_file_buffered_aio_write(iocb
, from
);
733 struct xfs_inode
*ip
= XFS_I(inode
);
735 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
737 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
741 xfs_break_dax_layouts(
747 ASSERT(xfs_isilocked(XFS_I(inode
), XFS_MMAPLOCK_EXCL
));
749 page
= dax_layout_busy_page(inode
->i_mapping
);
754 return ___wait_var_event(&page
->_refcount
,
755 atomic_read(&page
->_refcount
) == 1, TASK_INTERRUPTIBLE
,
756 0, 0, xfs_wait_dax_page(inode
));
763 enum layout_break_reason reason
)
768 ASSERT(xfs_isilocked(XFS_I(inode
), XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
));
774 error
= xfs_break_dax_layouts(inode
, &retry
);
779 error
= xfs_break_leased_layouts(inode
, iolock
, &retry
);
785 } while (error
== 0 && retry
);
790 #define XFS_FALLOC_FL_SUPPORTED \
791 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
792 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
793 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
802 struct inode
*inode
= file_inode(file
);
803 struct xfs_inode
*ip
= XFS_I(inode
);
805 enum xfs_prealloc_flags flags
= 0;
806 uint iolock
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
808 bool do_file_insert
= false;
810 if (!S_ISREG(inode
->i_mode
))
812 if (mode
& ~XFS_FALLOC_FL_SUPPORTED
)
815 xfs_ilock(ip
, iolock
);
816 error
= xfs_break_layouts(inode
, &iolock
, BREAK_UNMAP
);
821 * Must wait for all AIO to complete before we continue as AIO can
822 * change the file size on completion without holding any locks we
823 * currently hold. We must do this first because AIO can update both
824 * the on disk and in memory inode sizes, and the operations that follow
825 * require the in-memory size to be fully up-to-date.
827 inode_dio_wait(inode
);
830 * Now AIO and DIO has drained we flush and (if necessary) invalidate
831 * the cached range over the first operation we are about to run.
833 * We care about zero and collapse here because they both run a hole
834 * punch over the range first. Because that can zero data, and the range
835 * of invalidation for the shift operations is much larger, we still do
836 * the required flush for collapse in xfs_prepare_shift().
838 * Insert has the same range requirements as collapse, and we extend the
839 * file first which can zero data. Hence insert has the same
840 * flush/invalidate requirements as collapse and so they are both
841 * handled at the right time by xfs_prepare_shift().
843 if (mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
|
844 FALLOC_FL_COLLAPSE_RANGE
)) {
845 error
= xfs_flush_unmap_range(ip
, offset
, len
);
850 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
851 error
= xfs_free_file_space(ip
, offset
, len
);
854 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
855 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
857 if (offset
& blksize_mask
|| len
& blksize_mask
) {
863 * There is no need to overlap collapse range with EOF,
864 * in which case it is effectively a truncate operation
866 if (offset
+ len
>= i_size_read(inode
)) {
871 new_size
= i_size_read(inode
) - len
;
873 error
= xfs_collapse_file_space(ip
, offset
, len
);
876 } else if (mode
& FALLOC_FL_INSERT_RANGE
) {
877 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
878 loff_t isize
= i_size_read(inode
);
880 if (offset
& blksize_mask
|| len
& blksize_mask
) {
886 * New inode size must not exceed ->s_maxbytes, accounting for
887 * possible signed overflow.
889 if (inode
->i_sb
->s_maxbytes
- isize
< len
) {
893 new_size
= isize
+ len
;
895 /* Offset should be less than i_size */
896 if (offset
>= isize
) {
900 do_file_insert
= true;
902 flags
|= XFS_PREALLOC_SET
;
904 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
905 offset
+ len
> i_size_read(inode
)) {
906 new_size
= offset
+ len
;
907 error
= inode_newsize_ok(inode
, new_size
);
912 if (mode
& FALLOC_FL_ZERO_RANGE
) {
914 * Punch a hole and prealloc the range. We use a hole
915 * punch rather than unwritten extent conversion for two
918 * 1.) Hole punch handles partial block zeroing for us.
919 * 2.) If prealloc returns ENOSPC, the file range is
920 * still zero-valued by virtue of the hole punch.
922 unsigned int blksize
= i_blocksize(inode
);
924 trace_xfs_zero_file_space(ip
);
926 error
= xfs_free_file_space(ip
, offset
, len
);
930 len
= round_up(offset
+ len
, blksize
) -
931 round_down(offset
, blksize
);
932 offset
= round_down(offset
, blksize
);
933 } else if (mode
& FALLOC_FL_UNSHARE_RANGE
) {
934 error
= xfs_reflink_unshare(ip
, offset
, len
);
939 * If always_cow mode we can't use preallocations and
940 * thus should not create them.
942 if (xfs_is_always_cow_inode(ip
)) {
948 if (!xfs_is_always_cow_inode(ip
)) {
949 error
= xfs_alloc_file_space(ip
, offset
, len
,
956 if (file
->f_flags
& O_DSYNC
)
957 flags
|= XFS_PREALLOC_SYNC
;
959 error
= xfs_update_prealloc_flags(ip
, flags
);
963 /* Change file size if needed */
967 iattr
.ia_valid
= ATTR_SIZE
;
968 iattr
.ia_size
= new_size
;
969 error
= xfs_vn_setattr_size(file_dentry(file
), &iattr
);
975 * Perform hole insertion now that the file size has been
976 * updated so that if we crash during the operation we don't
977 * leave shifted extents past EOF and hence losing access to
978 * the data that is contained within them.
981 error
= xfs_insert_file_space(ip
, offset
, len
);
984 xfs_iunlock(ip
, iolock
);
995 struct xfs_inode
*ip
= XFS_I(file_inode(file
));
1000 * Operations creating pages in page cache need protection from hole
1001 * punching and similar ops
1003 if (advice
== POSIX_FADV_WILLNEED
) {
1004 lockflags
= XFS_IOLOCK_SHARED
;
1005 xfs_ilock(ip
, lockflags
);
1007 ret
= generic_fadvise(file
, start
, end
, advice
);
1009 xfs_iunlock(ip
, lockflags
);
1014 xfs_file_remap_range(
1015 struct file
*file_in
,
1017 struct file
*file_out
,
1020 unsigned int remap_flags
)
1022 struct inode
*inode_in
= file_inode(file_in
);
1023 struct xfs_inode
*src
= XFS_I(inode_in
);
1024 struct inode
*inode_out
= file_inode(file_out
);
1025 struct xfs_inode
*dest
= XFS_I(inode_out
);
1026 struct xfs_mount
*mp
= src
->i_mount
;
1027 loff_t remapped
= 0;
1028 xfs_extlen_t cowextsize
;
1031 if (remap_flags
& ~(REMAP_FILE_DEDUP
| REMAP_FILE_ADVISORY
))
1034 if (!xfs_sb_version_hasreflink(&mp
->m_sb
))
1037 if (XFS_FORCED_SHUTDOWN(mp
))
1040 /* Prepare and then clone file data. */
1041 ret
= xfs_reflink_remap_prep(file_in
, pos_in
, file_out
, pos_out
,
1043 if (ret
< 0 || len
== 0)
1046 trace_xfs_reflink_remap_range(src
, pos_in
, len
, dest
, pos_out
);
1048 ret
= xfs_reflink_remap_blocks(src
, pos_in
, dest
, pos_out
, len
,
1054 * Carry the cowextsize hint from src to dest if we're sharing the
1055 * entire source file to the entire destination file, the source file
1056 * has a cowextsize hint, and the destination file does not.
1059 if (pos_in
== 0 && len
== i_size_read(inode_in
) &&
1060 (src
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
) &&
1061 pos_out
== 0 && len
>= i_size_read(inode_out
) &&
1062 !(dest
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
))
1063 cowextsize
= src
->i_d
.di_cowextsize
;
1065 ret
= xfs_reflink_update_dest(dest
, pos_out
+ len
, cowextsize
,
1069 xfs_reflink_remap_unlock(file_in
, file_out
);
1071 trace_xfs_reflink_remap_range_error(dest
, ret
, _RET_IP_
);
1072 return remapped
> 0 ? remapped
: ret
;
1077 struct inode
*inode
,
1080 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
1082 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
1084 file
->f_mode
|= FMODE_NOWAIT
;
1090 struct inode
*inode
,
1093 struct xfs_inode
*ip
= XFS_I(inode
);
1097 error
= xfs_file_open(inode
, file
);
1102 * If there are any blocks, read-ahead block 0 as we're almost
1103 * certain to have the next operation be a read there.
1105 mode
= xfs_ilock_data_map_shared(ip
);
1106 if (ip
->i_d
.di_nextents
> 0)
1107 error
= xfs_dir3_data_readahead(ip
, 0, 0);
1108 xfs_iunlock(ip
, mode
);
1114 struct inode
*inode
,
1117 return xfs_release(XFS_I(inode
));
1123 struct dir_context
*ctx
)
1125 struct inode
*inode
= file_inode(file
);
1126 xfs_inode_t
*ip
= XFS_I(inode
);
1130 * The Linux API doesn't pass down the total size of the buffer
1131 * we read into down to the filesystem. With the filldir concept
1132 * it's not needed for correct information, but the XFS dir2 leaf
1133 * code wants an estimate of the buffer size to calculate it's
1134 * readahead window and size the buffers used for mapping to
1137 * Try to give it an estimate that's good enough, maybe at some
1138 * point we can change the ->readdir prototype to include the
1139 * buffer size. For now we use the current glibc buffer size.
1141 bufsize
= (size_t)min_t(loff_t
, XFS_READDIR_BUFSIZE
, ip
->i_d
.di_size
);
1143 return xfs_readdir(NULL
, ip
, ctx
, bufsize
);
1152 struct inode
*inode
= file
->f_mapping
->host
;
1154 if (XFS_FORCED_SHUTDOWN(XFS_I(inode
)->i_mount
))
1159 return generic_file_llseek(file
, offset
, whence
);
1161 offset
= iomap_seek_hole(inode
, offset
, &xfs_seek_iomap_ops
);
1164 offset
= iomap_seek_data(inode
, offset
, &xfs_seek_iomap_ops
);
1170 return vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1174 * Locking for serialisation of IO during page faults. This results in a lock
1178 * sb_start_pagefault(vfs, freeze)
1179 * i_mmaplock (XFS - truncate serialisation)
1181 * i_lock (XFS - extent map serialisation)
1184 __xfs_filemap_fault(
1185 struct vm_fault
*vmf
,
1186 enum page_entry_size pe_size
,
1189 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1190 struct xfs_inode
*ip
= XFS_I(inode
);
1193 trace_xfs_filemap_fault(ip
, pe_size
, write_fault
);
1196 sb_start_pagefault(inode
->i_sb
);
1197 file_update_time(vmf
->vma
->vm_file
);
1200 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1201 if (IS_DAX(inode
)) {
1204 ret
= dax_iomap_fault(vmf
, pe_size
, &pfn
, NULL
,
1205 (write_fault
&& !vmf
->cow_page
) ?
1206 &xfs_direct_write_iomap_ops
:
1207 &xfs_read_iomap_ops
);
1208 if (ret
& VM_FAULT_NEEDDSYNC
)
1209 ret
= dax_finish_sync_fault(vmf
, pe_size
, pfn
);
1212 ret
= iomap_page_mkwrite(vmf
,
1213 &xfs_buffered_write_iomap_ops
);
1215 ret
= filemap_fault(vmf
);
1217 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1220 sb_end_pagefault(inode
->i_sb
);
1226 struct vm_fault
*vmf
)
1228 /* DAX can shortcut the normal fault path on write faults! */
1229 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
,
1230 IS_DAX(file_inode(vmf
->vma
->vm_file
)) &&
1231 (vmf
->flags
& FAULT_FLAG_WRITE
));
1235 xfs_filemap_huge_fault(
1236 struct vm_fault
*vmf
,
1237 enum page_entry_size pe_size
)
1239 if (!IS_DAX(file_inode(vmf
->vma
->vm_file
)))
1240 return VM_FAULT_FALLBACK
;
1242 /* DAX can shortcut the normal fault path on write faults! */
1243 return __xfs_filemap_fault(vmf
, pe_size
,
1244 (vmf
->flags
& FAULT_FLAG_WRITE
));
1248 xfs_filemap_page_mkwrite(
1249 struct vm_fault
*vmf
)
1251 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
, true);
1255 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1256 * on write faults. In reality, it needs to serialise against truncate and
1257 * prepare memory for writing so handle is as standard write fault.
1260 xfs_filemap_pfn_mkwrite(
1261 struct vm_fault
*vmf
)
1264 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
, true);
1267 static const struct vm_operations_struct xfs_file_vm_ops
= {
1268 .fault
= xfs_filemap_fault
,
1269 .huge_fault
= xfs_filemap_huge_fault
,
1270 .map_pages
= filemap_map_pages
,
1271 .page_mkwrite
= xfs_filemap_page_mkwrite
,
1272 .pfn_mkwrite
= xfs_filemap_pfn_mkwrite
,
1278 struct vm_area_struct
*vma
)
1280 struct inode
*inode
= file_inode(file
);
1281 struct xfs_buftarg
*target
= xfs_inode_buftarg(XFS_I(inode
));
1284 * We don't support synchronous mappings for non-DAX files and
1285 * for DAX files if underneath dax_device is not synchronous.
1287 if (!daxdev_mapping_supported(vma
, target
->bt_daxdev
))
1290 file_accessed(file
);
1291 vma
->vm_ops
= &xfs_file_vm_ops
;
1293 vma
->vm_flags
|= VM_HUGEPAGE
;
1297 const struct file_operations xfs_file_operations
= {
1298 .llseek
= xfs_file_llseek
,
1299 .read_iter
= xfs_file_read_iter
,
1300 .write_iter
= xfs_file_write_iter
,
1301 .splice_read
= generic_file_splice_read
,
1302 .splice_write
= iter_file_splice_write
,
1303 .iopoll
= iomap_dio_iopoll
,
1304 .unlocked_ioctl
= xfs_file_ioctl
,
1305 #ifdef CONFIG_COMPAT
1306 .compat_ioctl
= xfs_file_compat_ioctl
,
1308 .mmap
= xfs_file_mmap
,
1309 .mmap_supported_flags
= MAP_SYNC
,
1310 .open
= xfs_file_open
,
1311 .release
= xfs_file_release
,
1312 .fsync
= xfs_file_fsync
,
1313 .get_unmapped_area
= thp_get_unmapped_area
,
1314 .fallocate
= xfs_file_fallocate
,
1315 .fadvise
= xfs_file_fadvise
,
1316 .remap_file_range
= xfs_file_remap_range
,
1319 const struct file_operations xfs_dir_file_operations
= {
1320 .open
= xfs_dir_open
,
1321 .read
= generic_read_dir
,
1322 .iterate_shared
= xfs_file_readdir
,
1323 .llseek
= generic_file_llseek
,
1324 .unlocked_ioctl
= xfs_file_ioctl
,
1325 #ifdef CONFIG_COMPAT
1326 .compat_ioctl
= xfs_file_compat_ioctl
,
1328 .fsync
= xfs_dir_fsync
,