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_da_format.h"
14 #include "xfs_da_btree.h"
15 #include "xfs_inode.h"
16 #include "xfs_trans.h"
17 #include "xfs_inode_item.h"
19 #include "xfs_bmap_util.h"
20 #include "xfs_error.h"
22 #include "xfs_dir2_priv.h"
23 #include "xfs_ioctl.h"
24 #include "xfs_trace.h"
26 #include "xfs_icache.h"
28 #include "xfs_iomap.h"
29 #include "xfs_reflink.h"
31 #include <linux/dcache.h>
32 #include <linux/falloc.h>
33 #include <linux/pagevec.h>
34 #include <linux/backing-dev.h>
35 #include <linux/mman.h>
37 static const struct vm_operations_struct xfs_file_vm_ops
;
40 xfs_update_prealloc_flags(
42 enum xfs_prealloc_flags flags
)
47 error
= xfs_trans_alloc(ip
->i_mount
, &M_RES(ip
->i_mount
)->tr_writeid
,
52 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
53 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
55 if (!(flags
& XFS_PREALLOC_INVISIBLE
)) {
56 VFS_I(ip
)->i_mode
&= ~S_ISUID
;
57 if (VFS_I(ip
)->i_mode
& S_IXGRP
)
58 VFS_I(ip
)->i_mode
&= ~S_ISGID
;
59 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
62 if (flags
& XFS_PREALLOC_SET
)
63 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
64 if (flags
& XFS_PREALLOC_CLEAR
)
65 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
67 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
68 if (flags
& XFS_PREALLOC_SYNC
)
69 xfs_trans_set_sync(tp
);
70 return xfs_trans_commit(tp
);
74 * Fsync operations on directories are much simpler than on regular files,
75 * as there is no file data to flush, and thus also no need for explicit
76 * cache flush operations, and there are no non-transaction metadata updates
77 * on directories either.
86 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
87 struct xfs_mount
*mp
= ip
->i_mount
;
90 trace_xfs_dir_fsync(ip
);
92 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
93 if (xfs_ipincount(ip
))
94 lsn
= ip
->i_itemp
->ili_last_lsn
;
95 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
99 return xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, NULL
);
109 struct inode
*inode
= file
->f_mapping
->host
;
110 struct xfs_inode
*ip
= XFS_I(inode
);
111 struct xfs_mount
*mp
= ip
->i_mount
;
116 trace_xfs_file_fsync(ip
);
118 error
= file_write_and_wait_range(file
, start
, end
);
122 if (XFS_FORCED_SHUTDOWN(mp
))
125 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
128 * If we have an RT and/or log subvolume we need to make sure to flush
129 * the write cache the device used for file data first. This is to
130 * ensure newly written file data make it to disk before logging the new
131 * inode size in case of an extending write.
133 if (XFS_IS_REALTIME_INODE(ip
))
134 xfs_blkdev_issue_flush(mp
->m_rtdev_targp
);
135 else if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
136 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
139 * All metadata updates are logged, which means that we just have to
140 * flush the log up to the latest LSN that touched the inode. If we have
141 * concurrent fsync/fdatasync() calls, we need them to all block on the
142 * log force before we clear the ili_fsync_fields field. This ensures
143 * that we don't get a racing sync operation that does not wait for the
144 * metadata to hit the journal before returning. If we race with
145 * clearing the ili_fsync_fields, then all that will happen is the log
146 * force will do nothing as the lsn will already be on disk. We can't
147 * race with setting ili_fsync_fields because that is done under
148 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
149 * until after the ili_fsync_fields is cleared.
151 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
152 if (xfs_ipincount(ip
)) {
154 (ip
->i_itemp
->ili_fsync_fields
& ~XFS_ILOG_TIMESTAMP
))
155 lsn
= ip
->i_itemp
->ili_last_lsn
;
159 error
= xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, &log_flushed
);
160 ip
->i_itemp
->ili_fsync_fields
= 0;
162 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
165 * If we only have a single device, and the log force about was
166 * a no-op we might have to flush the data device cache here.
167 * This can only happen for fdatasync/O_DSYNC if we were overwriting
168 * an already allocated file and thus do not have any metadata to
171 if (!log_flushed
&& !XFS_IS_REALTIME_INODE(ip
) &&
172 mp
->m_logdev_targp
== mp
->m_ddev_targp
)
173 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
179 xfs_file_dio_aio_read(
183 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
184 size_t count
= iov_iter_count(to
);
187 trace_xfs_file_direct_read(ip
, count
, iocb
->ki_pos
);
190 return 0; /* skip atime */
192 file_accessed(iocb
->ki_filp
);
194 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
195 ret
= iomap_dio_rw(iocb
, to
, &xfs_iomap_ops
, NULL
);
196 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
201 static noinline ssize_t
206 struct xfs_inode
*ip
= XFS_I(iocb
->ki_filp
->f_mapping
->host
);
207 size_t count
= iov_iter_count(to
);
210 trace_xfs_file_dax_read(ip
, count
, iocb
->ki_pos
);
213 return 0; /* skip atime */
215 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
216 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
))
219 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
222 ret
= dax_iomap_rw(iocb
, to
, &xfs_iomap_ops
);
223 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
225 file_accessed(iocb
->ki_filp
);
230 xfs_file_buffered_aio_read(
234 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
237 trace_xfs_file_buffered_read(ip
, iov_iter_count(to
), iocb
->ki_pos
);
239 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
240 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
))
243 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
245 ret
= generic_file_read_iter(iocb
, to
);
246 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
256 struct inode
*inode
= file_inode(iocb
->ki_filp
);
257 struct xfs_mount
*mp
= XFS_I(inode
)->i_mount
;
260 XFS_STATS_INC(mp
, xs_read_calls
);
262 if (XFS_FORCED_SHUTDOWN(mp
))
266 ret
= xfs_file_dax_read(iocb
, to
);
267 else if (iocb
->ki_flags
& IOCB_DIRECT
)
268 ret
= xfs_file_dio_aio_read(iocb
, to
);
270 ret
= xfs_file_buffered_aio_read(iocb
, to
);
273 XFS_STATS_ADD(mp
, xs_read_bytes
, ret
);
278 * Common pre-write limit and setup checks.
280 * Called with the iolocked held either shared and exclusive according to
281 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
282 * if called for a direct write beyond i_size.
285 xfs_file_aio_write_checks(
287 struct iov_iter
*from
,
290 struct file
*file
= iocb
->ki_filp
;
291 struct inode
*inode
= file
->f_mapping
->host
;
292 struct xfs_inode
*ip
= XFS_I(inode
);
294 size_t count
= iov_iter_count(from
);
295 bool drained_dio
= false;
299 error
= generic_write_checks(iocb
, from
);
303 error
= xfs_break_layouts(inode
, iolock
, BREAK_WRITE
);
308 * For changing security info in file_remove_privs() we need i_rwsem
311 if (*iolock
== XFS_IOLOCK_SHARED
&& !IS_NOSEC(inode
)) {
312 xfs_iunlock(ip
, *iolock
);
313 *iolock
= XFS_IOLOCK_EXCL
;
314 xfs_ilock(ip
, *iolock
);
318 * If the offset is beyond the size of the file, we need to zero any
319 * blocks that fall between the existing EOF and the start of this
320 * write. If zeroing is needed and we are currently holding the
321 * iolock shared, we need to update it to exclusive which implies
322 * having to redo all checks before.
324 * We need to serialise against EOF updates that occur in IO
325 * completions here. We want to make sure that nobody is changing the
326 * size while we do this check until we have placed an IO barrier (i.e.
327 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
328 * The spinlock effectively forms a memory barrier once we have the
329 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
330 * and hence be able to correctly determine if we need to run zeroing.
332 spin_lock(&ip
->i_flags_lock
);
333 isize
= i_size_read(inode
);
334 if (iocb
->ki_pos
> isize
) {
335 spin_unlock(&ip
->i_flags_lock
);
337 if (*iolock
== XFS_IOLOCK_SHARED
) {
338 xfs_iunlock(ip
, *iolock
);
339 *iolock
= XFS_IOLOCK_EXCL
;
340 xfs_ilock(ip
, *iolock
);
341 iov_iter_reexpand(from
, count
);
344 * We now have an IO submission barrier in place, but
345 * AIO can do EOF updates during IO completion and hence
346 * we now need to wait for all of them to drain. Non-AIO
347 * DIO will have drained before we are given the
348 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
351 inode_dio_wait(inode
);
356 trace_xfs_zero_eof(ip
, isize
, iocb
->ki_pos
- isize
);
357 error
= iomap_zero_range(inode
, isize
, iocb
->ki_pos
- isize
,
358 NULL
, &xfs_iomap_ops
);
362 spin_unlock(&ip
->i_flags_lock
);
365 * Updating the timestamps will grab the ilock again from
366 * xfs_fs_dirty_inode, so we have to call it after dropping the
367 * lock above. Eventually we should look into a way to avoid
368 * the pointless lock roundtrip.
370 if (likely(!(file
->f_mode
& FMODE_NOCMTIME
))) {
371 error
= file_update_time(file
);
377 * If we're writing the file then make sure to clear the setuid and
378 * setgid bits if the process is not being run by root. This keeps
379 * people from modifying setuid and setgid binaries.
381 if (!IS_NOSEC(inode
))
382 return file_remove_privs(file
);
387 xfs_dio_write_end_io(
392 struct inode
*inode
= file_inode(iocb
->ki_filp
);
393 struct xfs_inode
*ip
= XFS_I(inode
);
394 loff_t offset
= iocb
->ki_pos
;
397 trace_xfs_end_io_direct_write(ip
, offset
, size
);
399 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
406 * Capture amount written on completion as we can't reliably account
407 * for it on submission.
409 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, size
);
411 if (flags
& IOMAP_DIO_COW
) {
412 error
= xfs_reflink_end_cow(ip
, offset
, size
);
418 * Unwritten conversion updates the in-core isize after extent
419 * conversion but before updating the on-disk size. Updating isize any
420 * earlier allows a racing dio read to find unwritten extents before
421 * they are converted.
423 if (flags
& IOMAP_DIO_UNWRITTEN
)
424 return xfs_iomap_write_unwritten(ip
, offset
, size
, true);
427 * We need to update the in-core inode size here so that we don't end up
428 * with the on-disk inode size being outside the in-core inode size. We
429 * have no other method of updating EOF for AIO, so always do it here
432 * We need to lock the test/set EOF update as we can be racing with
433 * other IO completions here to update the EOF. Failing to serialise
434 * here can result in EOF moving backwards and Bad Things Happen when
437 spin_lock(&ip
->i_flags_lock
);
438 if (offset
+ size
> i_size_read(inode
)) {
439 i_size_write(inode
, offset
+ size
);
440 spin_unlock(&ip
->i_flags_lock
);
441 error
= xfs_setfilesize(ip
, offset
, size
);
443 spin_unlock(&ip
->i_flags_lock
);
450 * xfs_file_dio_aio_write - handle direct IO writes
452 * Lock the inode appropriately to prepare for and issue a direct IO write.
453 * By separating it from the buffered write path we remove all the tricky to
454 * follow locking changes and looping.
456 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
457 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
458 * pages are flushed out.
460 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
461 * allowing them to be done in parallel with reads and other direct IO writes.
462 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
463 * needs to do sub-block zeroing and that requires serialisation against other
464 * direct IOs to the same block. In this case we need to serialise the
465 * submission of the unaligned IOs so that we don't get racing block zeroing in
466 * the dio layer. To avoid the problem with aio, we also need to wait for
467 * outstanding IOs to complete so that unwritten extent conversion is completed
468 * before we try to map the overlapping block. This is currently implemented by
469 * hitting it with a big hammer (i.e. inode_dio_wait()).
471 * Returns with locks held indicated by @iolock and errors indicated by
472 * negative return values.
475 xfs_file_dio_aio_write(
477 struct iov_iter
*from
)
479 struct file
*file
= iocb
->ki_filp
;
480 struct address_space
*mapping
= file
->f_mapping
;
481 struct inode
*inode
= mapping
->host
;
482 struct xfs_inode
*ip
= XFS_I(inode
);
483 struct xfs_mount
*mp
= ip
->i_mount
;
485 int unaligned_io
= 0;
487 size_t count
= iov_iter_count(from
);
488 struct xfs_buftarg
*target
= XFS_IS_REALTIME_INODE(ip
) ?
489 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
491 /* DIO must be aligned to device logical sector size */
492 if ((iocb
->ki_pos
| count
) & target
->bt_logical_sectormask
)
496 * Don't take the exclusive iolock here unless the I/O is unaligned to
497 * the file system block size. We don't need to consider the EOF
498 * extension case here because xfs_file_aio_write_checks() will relock
499 * the inode as necessary for EOF zeroing cases and fill out the new
500 * inode size as appropriate.
502 if ((iocb
->ki_pos
& mp
->m_blockmask
) ||
503 ((iocb
->ki_pos
+ count
) & mp
->m_blockmask
)) {
507 * We can't properly handle unaligned direct I/O to reflink
508 * files yet, as we can't unshare a partial block.
510 if (xfs_is_reflink_inode(ip
)) {
511 trace_xfs_reflink_bounce_dio_write(ip
, iocb
->ki_pos
, count
);
514 iolock
= XFS_IOLOCK_EXCL
;
516 iolock
= XFS_IOLOCK_SHARED
;
519 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
520 if (!xfs_ilock_nowait(ip
, iolock
))
523 xfs_ilock(ip
, iolock
);
526 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
529 count
= iov_iter_count(from
);
532 * If we are doing unaligned IO, wait for all other IO to drain,
533 * otherwise demote the lock if we had to take the exclusive lock
534 * for other reasons in xfs_file_aio_write_checks.
537 /* If we are going to wait for other DIO to finish, bail */
538 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
539 if (atomic_read(&inode
->i_dio_count
))
542 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
);
550 ret
= iomap_dio_rw(iocb
, from
, &xfs_iomap_ops
, xfs_dio_write_end_io
);
552 xfs_iunlock(ip
, iolock
);
555 * No fallback to buffered IO on errors for XFS, direct IO will either
556 * complete fully or fail.
558 ASSERT(ret
< 0 || ret
== count
);
562 static noinline ssize_t
565 struct iov_iter
*from
)
567 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
568 struct xfs_inode
*ip
= XFS_I(inode
);
569 int iolock
= XFS_IOLOCK_EXCL
;
570 ssize_t ret
, error
= 0;
574 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
575 if (!xfs_ilock_nowait(ip
, iolock
))
578 xfs_ilock(ip
, iolock
);
581 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
586 count
= iov_iter_count(from
);
588 trace_xfs_file_dax_write(ip
, count
, pos
);
589 ret
= dax_iomap_rw(iocb
, from
, &xfs_iomap_ops
);
590 if (ret
> 0 && iocb
->ki_pos
> i_size_read(inode
)) {
591 i_size_write(inode
, iocb
->ki_pos
);
592 error
= xfs_setfilesize(ip
, pos
, ret
);
595 xfs_iunlock(ip
, iolock
);
600 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
602 /* Handle various SYNC-type writes */
603 ret
= generic_write_sync(iocb
, ret
);
609 xfs_file_buffered_aio_write(
611 struct iov_iter
*from
)
613 struct file
*file
= iocb
->ki_filp
;
614 struct address_space
*mapping
= file
->f_mapping
;
615 struct inode
*inode
= mapping
->host
;
616 struct xfs_inode
*ip
= XFS_I(inode
);
621 if (iocb
->ki_flags
& IOCB_NOWAIT
)
625 iolock
= XFS_IOLOCK_EXCL
;
626 xfs_ilock(ip
, iolock
);
628 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
632 /* We can write back this queue in page reclaim */
633 current
->backing_dev_info
= inode_to_bdi(inode
);
635 trace_xfs_file_buffered_write(ip
, iov_iter_count(from
), iocb
->ki_pos
);
636 ret
= iomap_file_buffered_write(iocb
, from
, &xfs_iomap_ops
);
637 if (likely(ret
>= 0))
641 * If we hit a space limit, try to free up some lingering preallocated
642 * space before returning an error. In the case of ENOSPC, first try to
643 * write back all dirty inodes to free up some of the excess reserved
644 * metadata space. This reduces the chances that the eofblocks scan
645 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
646 * also behaves as a filter to prevent too many eofblocks scans from
647 * running at the same time.
649 if (ret
== -EDQUOT
&& !enospc
) {
650 xfs_iunlock(ip
, iolock
);
651 enospc
= xfs_inode_free_quota_eofblocks(ip
);
654 enospc
= xfs_inode_free_quota_cowblocks(ip
);
658 } else if (ret
== -ENOSPC
&& !enospc
) {
659 struct xfs_eofblocks eofb
= {0};
662 xfs_flush_inodes(ip
->i_mount
);
664 xfs_iunlock(ip
, iolock
);
665 eofb
.eof_flags
= XFS_EOF_FLAGS_SYNC
;
666 xfs_icache_free_eofblocks(ip
->i_mount
, &eofb
);
667 xfs_icache_free_cowblocks(ip
->i_mount
, &eofb
);
671 current
->backing_dev_info
= NULL
;
674 xfs_iunlock(ip
, iolock
);
677 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
678 /* Handle various SYNC-type writes */
679 ret
= generic_write_sync(iocb
, ret
);
687 struct iov_iter
*from
)
689 struct file
*file
= iocb
->ki_filp
;
690 struct address_space
*mapping
= file
->f_mapping
;
691 struct inode
*inode
= mapping
->host
;
692 struct xfs_inode
*ip
= XFS_I(inode
);
694 size_t ocount
= iov_iter_count(from
);
696 XFS_STATS_INC(ip
->i_mount
, xs_write_calls
);
701 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
705 return xfs_file_dax_write(iocb
, from
);
707 if (iocb
->ki_flags
& IOCB_DIRECT
) {
709 * Allow a directio write to fall back to a buffered
710 * write *only* in the case that we're doing a reflink
711 * CoW. In all other directio scenarios we do not
712 * allow an operation to fall back to buffered mode.
714 ret
= xfs_file_dio_aio_write(iocb
, from
);
719 return xfs_file_buffered_aio_write(iocb
, from
);
726 struct xfs_inode
*ip
= XFS_I(inode
);
728 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
730 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
734 xfs_break_dax_layouts(
740 ASSERT(xfs_isilocked(XFS_I(inode
), XFS_MMAPLOCK_EXCL
));
742 page
= dax_layout_busy_page(inode
->i_mapping
);
747 return ___wait_var_event(&page
->_refcount
,
748 atomic_read(&page
->_refcount
) == 1, TASK_INTERRUPTIBLE
,
749 0, 0, xfs_wait_dax_page(inode
));
756 enum layout_break_reason reason
)
761 ASSERT(xfs_isilocked(XFS_I(inode
), XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
));
767 error
= xfs_break_dax_layouts(inode
, &retry
);
772 error
= xfs_break_leased_layouts(inode
, iolock
, &retry
);
778 } while (error
== 0 && retry
);
783 #define XFS_FALLOC_FL_SUPPORTED \
784 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
785 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
786 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
795 struct inode
*inode
= file_inode(file
);
796 struct xfs_inode
*ip
= XFS_I(inode
);
798 enum xfs_prealloc_flags flags
= 0;
799 uint iolock
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
801 bool do_file_insert
= false;
803 if (!S_ISREG(inode
->i_mode
))
805 if (mode
& ~XFS_FALLOC_FL_SUPPORTED
)
808 xfs_ilock(ip
, iolock
);
809 error
= xfs_break_layouts(inode
, &iolock
, BREAK_UNMAP
);
813 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
814 error
= xfs_free_file_space(ip
, offset
, len
);
817 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
818 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
820 if (offset
& blksize_mask
|| len
& blksize_mask
) {
826 * There is no need to overlap collapse range with EOF,
827 * in which case it is effectively a truncate operation
829 if (offset
+ len
>= i_size_read(inode
)) {
834 new_size
= i_size_read(inode
) - len
;
836 error
= xfs_collapse_file_space(ip
, offset
, len
);
839 } else if (mode
& FALLOC_FL_INSERT_RANGE
) {
840 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
841 loff_t isize
= i_size_read(inode
);
843 if (offset
& blksize_mask
|| len
& blksize_mask
) {
849 * New inode size must not exceed ->s_maxbytes, accounting for
850 * possible signed overflow.
852 if (inode
->i_sb
->s_maxbytes
- isize
< len
) {
856 new_size
= isize
+ len
;
858 /* Offset should be less than i_size */
859 if (offset
>= isize
) {
863 do_file_insert
= true;
865 flags
|= XFS_PREALLOC_SET
;
867 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
868 offset
+ len
> i_size_read(inode
)) {
869 new_size
= offset
+ len
;
870 error
= inode_newsize_ok(inode
, new_size
);
875 if (mode
& FALLOC_FL_ZERO_RANGE
)
876 error
= xfs_zero_file_space(ip
, offset
, len
);
878 if (mode
& FALLOC_FL_UNSHARE_RANGE
) {
879 error
= xfs_reflink_unshare(ip
, offset
, len
);
883 error
= xfs_alloc_file_space(ip
, offset
, len
,
890 if (file
->f_flags
& O_DSYNC
)
891 flags
|= XFS_PREALLOC_SYNC
;
893 error
= xfs_update_prealloc_flags(ip
, flags
);
897 /* Change file size if needed */
901 iattr
.ia_valid
= ATTR_SIZE
;
902 iattr
.ia_size
= new_size
;
903 error
= xfs_vn_setattr_size(file_dentry(file
), &iattr
);
909 * Perform hole insertion now that the file size has been
910 * updated so that if we crash during the operation we don't
911 * leave shifted extents past EOF and hence losing access to
912 * the data that is contained within them.
915 error
= xfs_insert_file_space(ip
, offset
, len
);
918 xfs_iunlock(ip
, iolock
);
924 xfs_file_remap_range(
925 struct file
*file_in
,
927 struct file
*file_out
,
930 unsigned int remap_flags
)
932 struct inode
*inode_in
= file_inode(file_in
);
933 struct xfs_inode
*src
= XFS_I(inode_in
);
934 struct inode
*inode_out
= file_inode(file_out
);
935 struct xfs_inode
*dest
= XFS_I(inode_out
);
936 struct xfs_mount
*mp
= src
->i_mount
;
938 xfs_extlen_t cowextsize
;
941 if (remap_flags
& ~(REMAP_FILE_DEDUP
| REMAP_FILE_ADVISORY
))
944 if (!xfs_sb_version_hasreflink(&mp
->m_sb
))
947 if (XFS_FORCED_SHUTDOWN(mp
))
950 /* Prepare and then clone file data. */
951 ret
= xfs_reflink_remap_prep(file_in
, pos_in
, file_out
, pos_out
,
953 if (ret
< 0 || len
== 0)
956 trace_xfs_reflink_remap_range(src
, pos_in
, len
, dest
, pos_out
);
958 ret
= xfs_reflink_remap_blocks(src
, pos_in
, dest
, pos_out
, len
,
964 * Carry the cowextsize hint from src to dest if we're sharing the
965 * entire source file to the entire destination file, the source file
966 * has a cowextsize hint, and the destination file does not.
969 if (pos_in
== 0 && len
== i_size_read(inode_in
) &&
970 (src
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
) &&
971 pos_out
== 0 && len
>= i_size_read(inode_out
) &&
972 !(dest
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
))
973 cowextsize
= src
->i_d
.di_cowextsize
;
975 ret
= xfs_reflink_update_dest(dest
, pos_out
+ len
, cowextsize
,
979 xfs_reflink_remap_unlock(file_in
, file_out
);
981 trace_xfs_reflink_remap_range_error(dest
, ret
, _RET_IP_
);
982 return remapped
> 0 ? remapped
: ret
;
990 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
992 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
994 file
->f_mode
|= FMODE_NOWAIT
;
1000 struct inode
*inode
,
1003 struct xfs_inode
*ip
= XFS_I(inode
);
1007 error
= xfs_file_open(inode
, file
);
1012 * If there are any blocks, read-ahead block 0 as we're almost
1013 * certain to have the next operation be a read there.
1015 mode
= xfs_ilock_data_map_shared(ip
);
1016 if (ip
->i_d
.di_nextents
> 0)
1017 error
= xfs_dir3_data_readahead(ip
, 0, -1);
1018 xfs_iunlock(ip
, mode
);
1024 struct inode
*inode
,
1027 return xfs_release(XFS_I(inode
));
1033 struct dir_context
*ctx
)
1035 struct inode
*inode
= file_inode(file
);
1036 xfs_inode_t
*ip
= XFS_I(inode
);
1040 * The Linux API doesn't pass down the total size of the buffer
1041 * we read into down to the filesystem. With the filldir concept
1042 * it's not needed for correct information, but the XFS dir2 leaf
1043 * code wants an estimate of the buffer size to calculate it's
1044 * readahead window and size the buffers used for mapping to
1047 * Try to give it an estimate that's good enough, maybe at some
1048 * point we can change the ->readdir prototype to include the
1049 * buffer size. For now we use the current glibc buffer size.
1051 bufsize
= (size_t)min_t(loff_t
, XFS_READDIR_BUFSIZE
, ip
->i_d
.di_size
);
1053 return xfs_readdir(NULL
, ip
, ctx
, bufsize
);
1062 struct inode
*inode
= file
->f_mapping
->host
;
1064 if (XFS_FORCED_SHUTDOWN(XFS_I(inode
)->i_mount
))
1069 return generic_file_llseek(file
, offset
, whence
);
1071 offset
= iomap_seek_hole(inode
, offset
, &xfs_iomap_ops
);
1074 offset
= iomap_seek_data(inode
, offset
, &xfs_iomap_ops
);
1080 return vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1084 * Locking for serialisation of IO during page faults. This results in a lock
1088 * sb_start_pagefault(vfs, freeze)
1089 * i_mmaplock (XFS - truncate serialisation)
1091 * i_lock (XFS - extent map serialisation)
1094 __xfs_filemap_fault(
1095 struct vm_fault
*vmf
,
1096 enum page_entry_size pe_size
,
1099 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1100 struct xfs_inode
*ip
= XFS_I(inode
);
1103 trace_xfs_filemap_fault(ip
, pe_size
, write_fault
);
1106 sb_start_pagefault(inode
->i_sb
);
1107 file_update_time(vmf
->vma
->vm_file
);
1110 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1111 if (IS_DAX(inode
)) {
1114 ret
= dax_iomap_fault(vmf
, pe_size
, &pfn
, NULL
, &xfs_iomap_ops
);
1115 if (ret
& VM_FAULT_NEEDDSYNC
)
1116 ret
= dax_finish_sync_fault(vmf
, pe_size
, pfn
);
1119 ret
= iomap_page_mkwrite(vmf
, &xfs_iomap_ops
);
1121 ret
= filemap_fault(vmf
);
1123 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1126 sb_end_pagefault(inode
->i_sb
);
1132 struct vm_fault
*vmf
)
1134 /* DAX can shortcut the normal fault path on write faults! */
1135 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
,
1136 IS_DAX(file_inode(vmf
->vma
->vm_file
)) &&
1137 (vmf
->flags
& FAULT_FLAG_WRITE
));
1141 xfs_filemap_huge_fault(
1142 struct vm_fault
*vmf
,
1143 enum page_entry_size pe_size
)
1145 if (!IS_DAX(file_inode(vmf
->vma
->vm_file
)))
1146 return VM_FAULT_FALLBACK
;
1148 /* DAX can shortcut the normal fault path on write faults! */
1149 return __xfs_filemap_fault(vmf
, pe_size
,
1150 (vmf
->flags
& FAULT_FLAG_WRITE
));
1154 xfs_filemap_page_mkwrite(
1155 struct vm_fault
*vmf
)
1157 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
, true);
1161 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1162 * on write faults. In reality, it needs to serialise against truncate and
1163 * prepare memory for writing so handle is as standard write fault.
1166 xfs_filemap_pfn_mkwrite(
1167 struct vm_fault
*vmf
)
1170 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
, true);
1173 static const struct vm_operations_struct xfs_file_vm_ops
= {
1174 .fault
= xfs_filemap_fault
,
1175 .huge_fault
= xfs_filemap_huge_fault
,
1176 .map_pages
= filemap_map_pages
,
1177 .page_mkwrite
= xfs_filemap_page_mkwrite
,
1178 .pfn_mkwrite
= xfs_filemap_pfn_mkwrite
,
1184 struct vm_area_struct
*vma
)
1187 * We don't support synchronous mappings for non-DAX files. At least
1188 * until someone comes with a sensible use case.
1190 if (!IS_DAX(file_inode(filp
)) && (vma
->vm_flags
& VM_SYNC
))
1193 file_accessed(filp
);
1194 vma
->vm_ops
= &xfs_file_vm_ops
;
1195 if (IS_DAX(file_inode(filp
)))
1196 vma
->vm_flags
|= VM_HUGEPAGE
;
1200 const struct file_operations xfs_file_operations
= {
1201 .llseek
= xfs_file_llseek
,
1202 .read_iter
= xfs_file_read_iter
,
1203 .write_iter
= xfs_file_write_iter
,
1204 .splice_read
= generic_file_splice_read
,
1205 .splice_write
= iter_file_splice_write
,
1206 .unlocked_ioctl
= xfs_file_ioctl
,
1207 #ifdef CONFIG_COMPAT
1208 .compat_ioctl
= xfs_file_compat_ioctl
,
1210 .mmap
= xfs_file_mmap
,
1211 .mmap_supported_flags
= MAP_SYNC
,
1212 .open
= xfs_file_open
,
1213 .release
= xfs_file_release
,
1214 .fsync
= xfs_file_fsync
,
1215 .get_unmapped_area
= thp_get_unmapped_area
,
1216 .fallocate
= xfs_file_fallocate
,
1217 .remap_file_range
= xfs_file_remap_range
,
1220 const struct file_operations xfs_dir_file_operations
= {
1221 .open
= xfs_dir_open
,
1222 .read
= generic_read_dir
,
1223 .iterate_shared
= xfs_file_readdir
,
1224 .llseek
= generic_file_llseek
,
1225 .unlocked_ioctl
= xfs_file_ioctl
,
1226 #ifdef CONFIG_COMPAT
1227 .compat_ioctl
= xfs_file_compat_ioctl
,
1229 .fsync
= xfs_dir_fsync
,