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 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
)) {
263 if (iocb
->ki_flags
& IOCB_NOWAIT
)
265 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
267 ret
= generic_file_read_iter(iocb
, to
);
268 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
278 struct inode
*inode
= file_inode(iocb
->ki_filp
);
279 struct xfs_mount
*mp
= XFS_I(inode
)->i_mount
;
282 XFS_STATS_INC(mp
, xs_read_calls
);
284 if (XFS_FORCED_SHUTDOWN(mp
))
288 ret
= xfs_file_dax_read(iocb
, to
);
289 else if (iocb
->ki_flags
& IOCB_DIRECT
)
290 ret
= xfs_file_dio_aio_read(iocb
, to
);
292 ret
= xfs_file_buffered_aio_read(iocb
, to
);
295 XFS_STATS_ADD(mp
, xs_read_bytes
, ret
);
300 * Zero any on disk space between the current EOF and the new, larger EOF.
302 * This handles the normal case of zeroing the remainder of the last block in
303 * the file and the unusual case of zeroing blocks out beyond the size of the
304 * file. This second case only happens with fixed size extents and when the
305 * system crashes before the inode size was updated but after blocks were
308 * Expects the iolock to be held exclusive, and will take the ilock internally.
310 int /* error (positive) */
312 struct xfs_inode
*ip
,
313 xfs_off_t offset
, /* starting I/O offset */
314 xfs_fsize_t isize
, /* current inode size */
317 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
318 ASSERT(offset
> isize
);
320 trace_xfs_zero_eof(ip
, isize
, offset
- isize
);
321 return xfs_zero_range(ip
, isize
, offset
- isize
, did_zeroing
);
325 * Common pre-write limit and setup checks.
327 * Called with the iolocked held either shared and exclusive according to
328 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
329 * if called for a direct write beyond i_size.
332 xfs_file_aio_write_checks(
334 struct iov_iter
*from
,
337 struct file
*file
= iocb
->ki_filp
;
338 struct inode
*inode
= file
->f_mapping
->host
;
339 struct xfs_inode
*ip
= XFS_I(inode
);
341 size_t count
= iov_iter_count(from
);
342 bool drained_dio
= false;
345 error
= generic_write_checks(iocb
, from
);
349 error
= xfs_break_layouts(inode
, iolock
);
354 * For changing security info in file_remove_privs() we need i_rwsem
357 if (*iolock
== XFS_IOLOCK_SHARED
&& !IS_NOSEC(inode
)) {
358 xfs_iunlock(ip
, *iolock
);
359 *iolock
= XFS_IOLOCK_EXCL
;
360 xfs_ilock(ip
, *iolock
);
364 * If the offset is beyond the size of the file, we need to zero any
365 * blocks that fall between the existing EOF and the start of this
366 * write. If zeroing is needed and we are currently holding the
367 * iolock shared, we need to update it to exclusive which implies
368 * having to redo all checks before.
370 * We need to serialise against EOF updates that occur in IO
371 * completions here. We want to make sure that nobody is changing the
372 * size while we do this check until we have placed an IO barrier (i.e.
373 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
374 * The spinlock effectively forms a memory barrier once we have the
375 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
376 * and hence be able to correctly determine if we need to run zeroing.
378 spin_lock(&ip
->i_flags_lock
);
379 if (iocb
->ki_pos
> i_size_read(inode
)) {
380 spin_unlock(&ip
->i_flags_lock
);
382 if (*iolock
== XFS_IOLOCK_SHARED
) {
383 xfs_iunlock(ip
, *iolock
);
384 *iolock
= XFS_IOLOCK_EXCL
;
385 xfs_ilock(ip
, *iolock
);
386 iov_iter_reexpand(from
, count
);
389 * We now have an IO submission barrier in place, but
390 * AIO can do EOF updates during IO completion and hence
391 * we now need to wait for all of them to drain. Non-AIO
392 * DIO will have drained before we are given the
393 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
396 inode_dio_wait(inode
);
400 error
= xfs_zero_eof(ip
, iocb
->ki_pos
, i_size_read(inode
), NULL
);
404 spin_unlock(&ip
->i_flags_lock
);
407 * Updating the timestamps will grab the ilock again from
408 * xfs_fs_dirty_inode, so we have to call it after dropping the
409 * lock above. Eventually we should look into a way to avoid
410 * the pointless lock roundtrip.
412 if (likely(!(file
->f_mode
& FMODE_NOCMTIME
))) {
413 error
= file_update_time(file
);
419 * If we're writing the file then make sure to clear the setuid and
420 * setgid bits if the process is not being run by root. This keeps
421 * people from modifying setuid and setgid binaries.
423 if (!IS_NOSEC(inode
))
424 return file_remove_privs(file
);
429 xfs_dio_write_end_io(
434 struct inode
*inode
= file_inode(iocb
->ki_filp
);
435 struct xfs_inode
*ip
= XFS_I(inode
);
436 loff_t offset
= iocb
->ki_pos
;
439 trace_xfs_end_io_direct_write(ip
, offset
, size
);
441 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
447 if (flags
& IOMAP_DIO_COW
) {
448 error
= xfs_reflink_end_cow(ip
, offset
, size
);
454 * Unwritten conversion updates the in-core isize after extent
455 * conversion but before updating the on-disk size. Updating isize any
456 * earlier allows a racing dio read to find unwritten extents before
457 * they are converted.
459 if (flags
& IOMAP_DIO_UNWRITTEN
)
460 return xfs_iomap_write_unwritten(ip
, offset
, size
, true);
463 * We need to update the in-core inode size here so that we don't end up
464 * with the on-disk inode size being outside the in-core inode size. We
465 * have no other method of updating EOF for AIO, so always do it here
468 * We need to lock the test/set EOF update as we can be racing with
469 * other IO completions here to update the EOF. Failing to serialise
470 * here can result in EOF moving backwards and Bad Things Happen when
473 spin_lock(&ip
->i_flags_lock
);
474 if (offset
+ size
> i_size_read(inode
)) {
475 i_size_write(inode
, offset
+ size
);
476 spin_unlock(&ip
->i_flags_lock
);
477 error
= xfs_setfilesize(ip
, offset
, size
);
479 spin_unlock(&ip
->i_flags_lock
);
486 * xfs_file_dio_aio_write - handle direct IO writes
488 * Lock the inode appropriately to prepare for and issue a direct IO write.
489 * By separating it from the buffered write path we remove all the tricky to
490 * follow locking changes and looping.
492 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
493 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
494 * pages are flushed out.
496 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
497 * allowing them to be done in parallel with reads and other direct IO writes.
498 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
499 * needs to do sub-block zeroing and that requires serialisation against other
500 * direct IOs to the same block. In this case we need to serialise the
501 * submission of the unaligned IOs so that we don't get racing block zeroing in
502 * the dio layer. To avoid the problem with aio, we also need to wait for
503 * outstanding IOs to complete so that unwritten extent conversion is completed
504 * before we try to map the overlapping block. This is currently implemented by
505 * hitting it with a big hammer (i.e. inode_dio_wait()).
507 * Returns with locks held indicated by @iolock and errors indicated by
508 * negative return values.
511 xfs_file_dio_aio_write(
513 struct iov_iter
*from
)
515 struct file
*file
= iocb
->ki_filp
;
516 struct address_space
*mapping
= file
->f_mapping
;
517 struct inode
*inode
= mapping
->host
;
518 struct xfs_inode
*ip
= XFS_I(inode
);
519 struct xfs_mount
*mp
= ip
->i_mount
;
521 int unaligned_io
= 0;
523 size_t count
= iov_iter_count(from
);
524 struct xfs_buftarg
*target
= XFS_IS_REALTIME_INODE(ip
) ?
525 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
527 /* DIO must be aligned to device logical sector size */
528 if ((iocb
->ki_pos
| count
) & target
->bt_logical_sectormask
)
532 * Don't take the exclusive iolock here unless the I/O is unaligned to
533 * the file system block size. We don't need to consider the EOF
534 * extension case here because xfs_file_aio_write_checks() will relock
535 * the inode as necessary for EOF zeroing cases and fill out the new
536 * inode size as appropriate.
538 if ((iocb
->ki_pos
& mp
->m_blockmask
) ||
539 ((iocb
->ki_pos
+ count
) & mp
->m_blockmask
)) {
543 * We can't properly handle unaligned direct I/O to reflink
544 * files yet, as we can't unshare a partial block.
546 if (xfs_is_reflink_inode(ip
)) {
547 trace_xfs_reflink_bounce_dio_write(ip
, iocb
->ki_pos
, count
);
550 iolock
= XFS_IOLOCK_EXCL
;
552 iolock
= XFS_IOLOCK_SHARED
;
555 if (!xfs_ilock_nowait(ip
, iolock
)) {
556 if (iocb
->ki_flags
& IOCB_NOWAIT
)
558 xfs_ilock(ip
, iolock
);
561 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
564 count
= iov_iter_count(from
);
567 * If we are doing unaligned IO, wait for all other IO to drain,
568 * otherwise demote the lock if we had to take the exclusive lock
569 * for other reasons in xfs_file_aio_write_checks.
572 /* If we are going to wait for other DIO to finish, bail */
573 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
574 if (atomic_read(&inode
->i_dio_count
))
577 inode_dio_wait(inode
);
579 } else if (iolock
== XFS_IOLOCK_EXCL
) {
580 xfs_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
581 iolock
= XFS_IOLOCK_SHARED
;
584 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
);
585 ret
= iomap_dio_rw(iocb
, from
, &xfs_iomap_ops
, xfs_dio_write_end_io
);
587 xfs_iunlock(ip
, iolock
);
590 * No fallback to buffered IO on errors for XFS, direct IO will either
591 * complete fully or fail.
593 ASSERT(ret
< 0 || ret
== count
);
597 static noinline ssize_t
600 struct iov_iter
*from
)
602 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
603 struct xfs_inode
*ip
= XFS_I(inode
);
604 int iolock
= XFS_IOLOCK_EXCL
;
605 ssize_t ret
, error
= 0;
609 if (!xfs_ilock_nowait(ip
, iolock
)) {
610 if (iocb
->ki_flags
& IOCB_NOWAIT
)
612 xfs_ilock(ip
, iolock
);
615 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
620 count
= iov_iter_count(from
);
622 trace_xfs_file_dax_write(ip
, count
, pos
);
623 ret
= dax_iomap_rw(iocb
, from
, &xfs_iomap_ops
);
624 if (ret
> 0 && iocb
->ki_pos
> i_size_read(inode
)) {
625 i_size_write(inode
, iocb
->ki_pos
);
626 error
= xfs_setfilesize(ip
, pos
, ret
);
629 xfs_iunlock(ip
, iolock
);
630 return error
? error
: ret
;
634 xfs_file_buffered_aio_write(
636 struct iov_iter
*from
)
638 struct file
*file
= iocb
->ki_filp
;
639 struct address_space
*mapping
= file
->f_mapping
;
640 struct inode
*inode
= mapping
->host
;
641 struct xfs_inode
*ip
= XFS_I(inode
);
646 if (iocb
->ki_flags
& IOCB_NOWAIT
)
650 iolock
= XFS_IOLOCK_EXCL
;
651 xfs_ilock(ip
, iolock
);
653 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
657 /* We can write back this queue in page reclaim */
658 current
->backing_dev_info
= inode_to_bdi(inode
);
660 trace_xfs_file_buffered_write(ip
, iov_iter_count(from
), iocb
->ki_pos
);
661 ret
= iomap_file_buffered_write(iocb
, from
, &xfs_iomap_ops
);
662 if (likely(ret
>= 0))
666 * If we hit a space limit, try to free up some lingering preallocated
667 * space before returning an error. In the case of ENOSPC, first try to
668 * write back all dirty inodes to free up some of the excess reserved
669 * metadata space. This reduces the chances that the eofblocks scan
670 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
671 * also behaves as a filter to prevent too many eofblocks scans from
672 * running at the same time.
674 if (ret
== -EDQUOT
&& !enospc
) {
675 xfs_iunlock(ip
, iolock
);
676 enospc
= xfs_inode_free_quota_eofblocks(ip
);
679 enospc
= xfs_inode_free_quota_cowblocks(ip
);
683 } else if (ret
== -ENOSPC
&& !enospc
) {
684 struct xfs_eofblocks eofb
= {0};
687 xfs_flush_inodes(ip
->i_mount
);
689 xfs_iunlock(ip
, iolock
);
690 eofb
.eof_flags
= XFS_EOF_FLAGS_SYNC
;
691 xfs_icache_free_eofblocks(ip
->i_mount
, &eofb
);
692 xfs_icache_free_cowblocks(ip
->i_mount
, &eofb
);
696 current
->backing_dev_info
= NULL
;
699 xfs_iunlock(ip
, iolock
);
706 struct iov_iter
*from
)
708 struct file
*file
= iocb
->ki_filp
;
709 struct address_space
*mapping
= file
->f_mapping
;
710 struct inode
*inode
= mapping
->host
;
711 struct xfs_inode
*ip
= XFS_I(inode
);
713 size_t ocount
= iov_iter_count(from
);
715 XFS_STATS_INC(ip
->i_mount
, xs_write_calls
);
720 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
724 ret
= xfs_file_dax_write(iocb
, from
);
725 else if (iocb
->ki_flags
& IOCB_DIRECT
) {
727 * Allow a directio write to fall back to a buffered
728 * write *only* in the case that we're doing a reflink
729 * CoW. In all other directio scenarios we do not
730 * allow an operation to fall back to buffered mode.
732 ret
= xfs_file_dio_aio_write(iocb
, from
);
737 ret
= xfs_file_buffered_aio_write(iocb
, from
);
741 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
743 /* Handle various SYNC-type writes */
744 ret
= generic_write_sync(iocb
, ret
);
749 #define XFS_FALLOC_FL_SUPPORTED \
750 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
751 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
752 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
761 struct inode
*inode
= file_inode(file
);
762 struct xfs_inode
*ip
= XFS_I(inode
);
764 enum xfs_prealloc_flags flags
= 0;
765 uint iolock
= XFS_IOLOCK_EXCL
;
767 bool do_file_insert
= false;
769 if (!S_ISREG(inode
->i_mode
))
771 if (mode
& ~XFS_FALLOC_FL_SUPPORTED
)
774 xfs_ilock(ip
, iolock
);
775 error
= xfs_break_layouts(inode
, &iolock
);
779 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
780 iolock
|= XFS_MMAPLOCK_EXCL
;
782 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
783 error
= xfs_free_file_space(ip
, offset
, len
);
786 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
787 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
789 if (offset
& blksize_mask
|| len
& blksize_mask
) {
795 * There is no need to overlap collapse range with EOF,
796 * in which case it is effectively a truncate operation
798 if (offset
+ len
>= i_size_read(inode
)) {
803 new_size
= i_size_read(inode
) - len
;
805 error
= xfs_collapse_file_space(ip
, offset
, len
);
808 } else if (mode
& FALLOC_FL_INSERT_RANGE
) {
809 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
811 new_size
= i_size_read(inode
) + len
;
812 if (offset
& blksize_mask
|| len
& blksize_mask
) {
817 /* check the new inode size does not wrap through zero */
818 if (new_size
> inode
->i_sb
->s_maxbytes
) {
823 /* Offset should be less than i_size */
824 if (offset
>= i_size_read(inode
)) {
828 do_file_insert
= true;
830 flags
|= XFS_PREALLOC_SET
;
832 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
833 offset
+ len
> i_size_read(inode
)) {
834 new_size
= offset
+ len
;
835 error
= inode_newsize_ok(inode
, new_size
);
840 if (mode
& FALLOC_FL_ZERO_RANGE
)
841 error
= xfs_zero_file_space(ip
, offset
, len
);
843 if (mode
& FALLOC_FL_UNSHARE_RANGE
) {
844 error
= xfs_reflink_unshare(ip
, offset
, len
);
848 error
= xfs_alloc_file_space(ip
, offset
, len
,
855 if (file
->f_flags
& O_DSYNC
)
856 flags
|= XFS_PREALLOC_SYNC
;
858 error
= xfs_update_prealloc_flags(ip
, flags
);
862 /* Change file size if needed */
866 iattr
.ia_valid
= ATTR_SIZE
;
867 iattr
.ia_size
= new_size
;
868 error
= xfs_vn_setattr_size(file_dentry(file
), &iattr
);
874 * Perform hole insertion now that the file size has been
875 * updated so that if we crash during the operation we don't
876 * leave shifted extents past EOF and hence losing access to
877 * the data that is contained within them.
880 error
= xfs_insert_file_space(ip
, offset
, len
);
883 xfs_iunlock(ip
, iolock
);
888 xfs_file_clone_range(
889 struct file
*file_in
,
891 struct file
*file_out
,
895 return xfs_reflink_remap_range(file_in
, pos_in
, file_out
, pos_out
,
900 xfs_file_dedupe_range(
901 struct file
*src_file
,
904 struct file
*dst_file
,
909 error
= xfs_reflink_remap_range(src_file
, loff
, dst_file
, dst_loff
,
921 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
923 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
925 file
->f_mode
|= FMODE_NOWAIT
;
934 struct xfs_inode
*ip
= XFS_I(inode
);
938 error
= xfs_file_open(inode
, file
);
943 * If there are any blocks, read-ahead block 0 as we're almost
944 * certain to have the next operation be a read there.
946 mode
= xfs_ilock_data_map_shared(ip
);
947 if (ip
->i_d
.di_nextents
> 0)
948 error
= xfs_dir3_data_readahead(ip
, 0, -1);
949 xfs_iunlock(ip
, mode
);
958 return xfs_release(XFS_I(inode
));
964 struct dir_context
*ctx
)
966 struct inode
*inode
= file_inode(file
);
967 xfs_inode_t
*ip
= XFS_I(inode
);
971 * The Linux API doesn't pass down the total size of the buffer
972 * we read into down to the filesystem. With the filldir concept
973 * it's not needed for correct information, but the XFS dir2 leaf
974 * code wants an estimate of the buffer size to calculate it's
975 * readahead window and size the buffers used for mapping to
978 * Try to give it an estimate that's good enough, maybe at some
979 * point we can change the ->readdir prototype to include the
980 * buffer size. For now we use the current glibc buffer size.
982 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
984 return xfs_readdir(NULL
, ip
, ctx
, bufsize
);
993 struct inode
*inode
= file
->f_mapping
->host
;
995 if (XFS_FORCED_SHUTDOWN(XFS_I(inode
)->i_mount
))
1000 return generic_file_llseek(file
, offset
, whence
);
1002 offset
= iomap_seek_hole(inode
, offset
, &xfs_iomap_ops
);
1005 offset
= iomap_seek_data(inode
, offset
, &xfs_iomap_ops
);
1011 return vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1015 * Locking for serialisation of IO during page faults. This results in a lock
1019 * sb_start_pagefault(vfs, freeze)
1020 * i_mmaplock (XFS - truncate serialisation)
1022 * i_lock (XFS - extent map serialisation)
1025 __xfs_filemap_fault(
1026 struct vm_fault
*vmf
,
1027 enum page_entry_size pe_size
,
1030 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1031 struct xfs_inode
*ip
= XFS_I(inode
);
1034 trace_xfs_filemap_fault(ip
, pe_size
, write_fault
);
1037 sb_start_pagefault(inode
->i_sb
);
1038 file_update_time(vmf
->vma
->vm_file
);
1041 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1042 if (IS_DAX(inode
)) {
1043 ret
= dax_iomap_fault(vmf
, pe_size
, &xfs_iomap_ops
);
1046 ret
= iomap_page_mkwrite(vmf
, &xfs_iomap_ops
);
1048 ret
= filemap_fault(vmf
);
1050 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1053 sb_end_pagefault(inode
->i_sb
);
1059 struct vm_fault
*vmf
)
1061 /* DAX can shortcut the normal fault path on write faults! */
1062 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
,
1063 IS_DAX(file_inode(vmf
->vma
->vm_file
)) &&
1064 (vmf
->flags
& FAULT_FLAG_WRITE
));
1068 xfs_filemap_huge_fault(
1069 struct vm_fault
*vmf
,
1070 enum page_entry_size pe_size
)
1072 if (!IS_DAX(file_inode(vmf
->vma
->vm_file
)))
1073 return VM_FAULT_FALLBACK
;
1075 /* DAX can shortcut the normal fault path on write faults! */
1076 return __xfs_filemap_fault(vmf
, pe_size
,
1077 (vmf
->flags
& FAULT_FLAG_WRITE
));
1081 xfs_filemap_page_mkwrite(
1082 struct vm_fault
*vmf
)
1084 return __xfs_filemap_fault(vmf
, PE_SIZE_PTE
, true);
1088 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1089 * updates on write faults. In reality, it's need to serialise against
1090 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1091 * to ensure we serialise the fault barrier in place.
1094 xfs_filemap_pfn_mkwrite(
1095 struct vm_fault
*vmf
)
1098 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1099 struct xfs_inode
*ip
= XFS_I(inode
);
1100 int ret
= VM_FAULT_NOPAGE
;
1103 trace_xfs_filemap_pfn_mkwrite(ip
);
1105 sb_start_pagefault(inode
->i_sb
);
1106 file_update_time(vmf
->vma
->vm_file
);
1108 /* check if the faulting page hasn't raced with truncate */
1109 xfs_ilock(ip
, XFS_MMAPLOCK_SHARED
);
1110 size
= (i_size_read(inode
) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1111 if (vmf
->pgoff
>= size
)
1112 ret
= VM_FAULT_SIGBUS
;
1113 else if (IS_DAX(inode
))
1114 ret
= dax_iomap_fault(vmf
, PE_SIZE_PTE
, &xfs_iomap_ops
);
1115 xfs_iunlock(ip
, XFS_MMAPLOCK_SHARED
);
1116 sb_end_pagefault(inode
->i_sb
);
1121 static const struct vm_operations_struct xfs_file_vm_ops
= {
1122 .fault
= xfs_filemap_fault
,
1123 .huge_fault
= xfs_filemap_huge_fault
,
1124 .map_pages
= filemap_map_pages
,
1125 .page_mkwrite
= xfs_filemap_page_mkwrite
,
1126 .pfn_mkwrite
= xfs_filemap_pfn_mkwrite
,
1132 struct vm_area_struct
*vma
)
1134 file_accessed(filp
);
1135 vma
->vm_ops
= &xfs_file_vm_ops
;
1136 if (IS_DAX(file_inode(filp
)))
1137 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
1141 const struct file_operations xfs_file_operations
= {
1142 .llseek
= xfs_file_llseek
,
1143 .read_iter
= xfs_file_read_iter
,
1144 .write_iter
= xfs_file_write_iter
,
1145 .splice_read
= generic_file_splice_read
,
1146 .splice_write
= iter_file_splice_write
,
1147 .unlocked_ioctl
= xfs_file_ioctl
,
1148 #ifdef CONFIG_COMPAT
1149 .compat_ioctl
= xfs_file_compat_ioctl
,
1151 .mmap
= xfs_file_mmap
,
1152 .open
= xfs_file_open
,
1153 .release
= xfs_file_release
,
1154 .fsync
= xfs_file_fsync
,
1155 .get_unmapped_area
= thp_get_unmapped_area
,
1156 .fallocate
= xfs_file_fallocate
,
1157 .clone_file_range
= xfs_file_clone_range
,
1158 .dedupe_file_range
= xfs_file_dedupe_range
,
1161 const struct file_operations xfs_dir_file_operations
= {
1162 .open
= xfs_dir_open
,
1163 .read
= generic_read_dir
,
1164 .iterate_shared
= xfs_file_readdir
,
1165 .llseek
= generic_file_llseek
,
1166 .unlocked_ioctl
= xfs_file_ioctl
,
1167 #ifdef CONFIG_COMPAT
1168 .compat_ioctl
= xfs_file_compat_ioctl
,
1170 .fsync
= xfs_dir_fsync
,