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
, NULL
, &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
= filemap_write_and_wait_range(inode
->i_mapping
, 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 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
241 ret
= dax_iomap_rw(iocb
, to
, &xfs_iomap_ops
);
242 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
244 file_accessed(iocb
->ki_filp
);
249 xfs_file_buffered_aio_read(
253 struct xfs_inode
*ip
= XFS_I(file_inode(iocb
->ki_filp
));
256 trace_xfs_file_buffered_read(ip
, iov_iter_count(to
), iocb
->ki_pos
);
258 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
259 ret
= generic_file_read_iter(iocb
, to
);
260 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
270 struct inode
*inode
= file_inode(iocb
->ki_filp
);
271 struct xfs_mount
*mp
= XFS_I(inode
)->i_mount
;
274 XFS_STATS_INC(mp
, xs_read_calls
);
276 if (XFS_FORCED_SHUTDOWN(mp
))
280 ret
= xfs_file_dax_read(iocb
, to
);
281 else if (iocb
->ki_flags
& IOCB_DIRECT
)
282 ret
= xfs_file_dio_aio_read(iocb
, to
);
284 ret
= xfs_file_buffered_aio_read(iocb
, to
);
287 XFS_STATS_ADD(mp
, xs_read_bytes
, ret
);
292 * Zero any on disk space between the current EOF and the new, larger EOF.
294 * This handles the normal case of zeroing the remainder of the last block in
295 * the file and the unusual case of zeroing blocks out beyond the size of the
296 * file. This second case only happens with fixed size extents and when the
297 * system crashes before the inode size was updated but after blocks were
300 * Expects the iolock to be held exclusive, and will take the ilock internally.
302 int /* error (positive) */
304 struct xfs_inode
*ip
,
305 xfs_off_t offset
, /* starting I/O offset */
306 xfs_fsize_t isize
, /* current inode size */
309 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
310 ASSERT(offset
> isize
);
312 trace_xfs_zero_eof(ip
, isize
, offset
- isize
);
313 return xfs_zero_range(ip
, isize
, offset
- isize
, did_zeroing
);
317 * Common pre-write limit and setup checks.
319 * Called with the iolocked held either shared and exclusive according to
320 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
321 * if called for a direct write beyond i_size.
324 xfs_file_aio_write_checks(
326 struct iov_iter
*from
,
329 struct file
*file
= iocb
->ki_filp
;
330 struct inode
*inode
= file
->f_mapping
->host
;
331 struct xfs_inode
*ip
= XFS_I(inode
);
333 size_t count
= iov_iter_count(from
);
334 bool drained_dio
= false;
337 error
= generic_write_checks(iocb
, from
);
341 error
= xfs_break_layouts(inode
, iolock
);
346 * For changing security info in file_remove_privs() we need i_rwsem
349 if (*iolock
== XFS_IOLOCK_SHARED
&& !IS_NOSEC(inode
)) {
350 xfs_iunlock(ip
, *iolock
);
351 *iolock
= XFS_IOLOCK_EXCL
;
352 xfs_ilock(ip
, *iolock
);
356 * If the offset is beyond the size of the file, we need to zero any
357 * blocks that fall between the existing EOF and the start of this
358 * write. If zeroing is needed and we are currently holding the
359 * iolock shared, we need to update it to exclusive which implies
360 * having to redo all checks before.
362 * We need to serialise against EOF updates that occur in IO
363 * completions here. We want to make sure that nobody is changing the
364 * size while we do this check until we have placed an IO barrier (i.e.
365 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
366 * The spinlock effectively forms a memory barrier once we have the
367 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
368 * and hence be able to correctly determine if we need to run zeroing.
370 spin_lock(&ip
->i_flags_lock
);
371 if (iocb
->ki_pos
> i_size_read(inode
)) {
374 spin_unlock(&ip
->i_flags_lock
);
376 if (*iolock
== XFS_IOLOCK_SHARED
) {
377 xfs_iunlock(ip
, *iolock
);
378 *iolock
= XFS_IOLOCK_EXCL
;
379 xfs_ilock(ip
, *iolock
);
380 iov_iter_reexpand(from
, count
);
383 * We now have an IO submission barrier in place, but
384 * AIO can do EOF updates during IO completion and hence
385 * we now need to wait for all of them to drain. Non-AIO
386 * DIO will have drained before we are given the
387 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
390 inode_dio_wait(inode
);
394 error
= xfs_zero_eof(ip
, iocb
->ki_pos
, i_size_read(inode
), &zero
);
398 spin_unlock(&ip
->i_flags_lock
);
401 * Updating the timestamps will grab the ilock again from
402 * xfs_fs_dirty_inode, so we have to call it after dropping the
403 * lock above. Eventually we should look into a way to avoid
404 * the pointless lock roundtrip.
406 if (likely(!(file
->f_mode
& FMODE_NOCMTIME
))) {
407 error
= file_update_time(file
);
413 * If we're writing the file then make sure to clear the setuid and
414 * setgid bits if the process is not being run by root. This keeps
415 * people from modifying setuid and setgid binaries.
417 if (!IS_NOSEC(inode
))
418 return file_remove_privs(file
);
423 xfs_dio_write_end_io(
428 struct inode
*inode
= file_inode(iocb
->ki_filp
);
429 struct xfs_inode
*ip
= XFS_I(inode
);
430 loff_t offset
= iocb
->ki_pos
;
431 bool update_size
= false;
434 trace_xfs_end_io_direct_write(ip
, offset
, size
);
436 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
443 * We need to update the in-core inode size here so that we don't end up
444 * with the on-disk inode size being outside the in-core inode size. We
445 * have no other method of updating EOF for AIO, so always do it here
448 * We need to lock the test/set EOF update as we can be racing with
449 * other IO completions here to update the EOF. Failing to serialise
450 * here can result in EOF moving backwards and Bad Things Happen when
453 spin_lock(&ip
->i_flags_lock
);
454 if (offset
+ size
> i_size_read(inode
)) {
455 i_size_write(inode
, offset
+ size
);
458 spin_unlock(&ip
->i_flags_lock
);
460 if (flags
& IOMAP_DIO_COW
) {
461 error
= xfs_reflink_end_cow(ip
, offset
, size
);
466 if (flags
& IOMAP_DIO_UNWRITTEN
)
467 error
= xfs_iomap_write_unwritten(ip
, offset
, size
);
468 else if (update_size
)
469 error
= xfs_setfilesize(ip
, offset
, size
);
475 * xfs_file_dio_aio_write - handle direct IO writes
477 * Lock the inode appropriately to prepare for and issue a direct IO write.
478 * By separating it from the buffered write path we remove all the tricky to
479 * follow locking changes and looping.
481 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
482 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
483 * pages are flushed out.
485 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
486 * allowing them to be done in parallel with reads and other direct IO writes.
487 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
488 * needs to do sub-block zeroing and that requires serialisation against other
489 * direct IOs to the same block. In this case we need to serialise the
490 * submission of the unaligned IOs so that we don't get racing block zeroing in
491 * the dio layer. To avoid the problem with aio, we also need to wait for
492 * outstanding IOs to complete so that unwritten extent conversion is completed
493 * before we try to map the overlapping block. This is currently implemented by
494 * hitting it with a big hammer (i.e. inode_dio_wait()).
496 * Returns with locks held indicated by @iolock and errors indicated by
497 * negative return values.
500 xfs_file_dio_aio_write(
502 struct iov_iter
*from
)
504 struct file
*file
= iocb
->ki_filp
;
505 struct address_space
*mapping
= file
->f_mapping
;
506 struct inode
*inode
= mapping
->host
;
507 struct xfs_inode
*ip
= XFS_I(inode
);
508 struct xfs_mount
*mp
= ip
->i_mount
;
510 int unaligned_io
= 0;
512 size_t count
= iov_iter_count(from
);
513 struct xfs_buftarg
*target
= XFS_IS_REALTIME_INODE(ip
) ?
514 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
516 /* DIO must be aligned to device logical sector size */
517 if ((iocb
->ki_pos
| count
) & target
->bt_logical_sectormask
)
521 * Don't take the exclusive iolock here unless the I/O is unaligned to
522 * the file system block size. We don't need to consider the EOF
523 * extension case here because xfs_file_aio_write_checks() will relock
524 * the inode as necessary for EOF zeroing cases and fill out the new
525 * inode size as appropriate.
527 if ((iocb
->ki_pos
& mp
->m_blockmask
) ||
528 ((iocb
->ki_pos
+ count
) & mp
->m_blockmask
)) {
532 * We can't properly handle unaligned direct I/O to reflink
533 * files yet, as we can't unshare a partial block.
535 if (xfs_is_reflink_inode(ip
)) {
536 trace_xfs_reflink_bounce_dio_write(ip
, iocb
->ki_pos
, count
);
539 iolock
= XFS_IOLOCK_EXCL
;
541 iolock
= XFS_IOLOCK_SHARED
;
544 xfs_ilock(ip
, iolock
);
546 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
549 count
= iov_iter_count(from
);
552 * If we are doing unaligned IO, wait for all other IO to drain,
553 * otherwise demote the lock if we had to take the exclusive lock
554 * for other reasons in xfs_file_aio_write_checks.
557 inode_dio_wait(inode
);
558 else if (iolock
== XFS_IOLOCK_EXCL
) {
559 xfs_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
560 iolock
= XFS_IOLOCK_SHARED
;
563 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
);
564 ret
= iomap_dio_rw(iocb
, from
, &xfs_iomap_ops
, xfs_dio_write_end_io
);
566 xfs_iunlock(ip
, iolock
);
569 * No fallback to buffered IO on errors for XFS, direct IO will either
570 * complete fully or fail.
572 ASSERT(ret
< 0 || ret
== count
);
576 static noinline ssize_t
579 struct iov_iter
*from
)
581 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
582 struct xfs_inode
*ip
= XFS_I(inode
);
583 int iolock
= XFS_IOLOCK_EXCL
;
584 ssize_t ret
, error
= 0;
588 xfs_ilock(ip
, iolock
);
589 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
594 count
= iov_iter_count(from
);
596 trace_xfs_file_dax_write(ip
, count
, pos
);
597 ret
= dax_iomap_rw(iocb
, from
, &xfs_iomap_ops
);
598 if (ret
> 0 && iocb
->ki_pos
> i_size_read(inode
)) {
599 i_size_write(inode
, iocb
->ki_pos
);
600 error
= xfs_setfilesize(ip
, pos
, ret
);
603 xfs_iunlock(ip
, iolock
);
604 return error
? error
: ret
;
608 xfs_file_buffered_aio_write(
610 struct iov_iter
*from
)
612 struct file
*file
= iocb
->ki_filp
;
613 struct address_space
*mapping
= file
->f_mapping
;
614 struct inode
*inode
= mapping
->host
;
615 struct xfs_inode
*ip
= XFS_I(inode
);
621 iolock
= XFS_IOLOCK_EXCL
;
622 xfs_ilock(ip
, iolock
);
624 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
628 /* We can write back this queue in page reclaim */
629 current
->backing_dev_info
= inode_to_bdi(inode
);
631 trace_xfs_file_buffered_write(ip
, iov_iter_count(from
), iocb
->ki_pos
);
632 ret
= iomap_file_buffered_write(iocb
, from
, &xfs_iomap_ops
);
633 if (likely(ret
>= 0))
637 * If we hit a space limit, try to free up some lingering preallocated
638 * space before returning an error. In the case of ENOSPC, first try to
639 * write back all dirty inodes to free up some of the excess reserved
640 * metadata space. This reduces the chances that the eofblocks scan
641 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
642 * also behaves as a filter to prevent too many eofblocks scans from
643 * running at the same time.
645 if (ret
== -EDQUOT
&& !enospc
) {
646 xfs_iunlock(ip
, iolock
);
647 enospc
= xfs_inode_free_quota_eofblocks(ip
);
650 enospc
= xfs_inode_free_quota_cowblocks(ip
);
654 } else if (ret
== -ENOSPC
&& !enospc
) {
655 struct xfs_eofblocks eofb
= {0};
658 xfs_flush_inodes(ip
->i_mount
);
660 xfs_iunlock(ip
, iolock
);
661 eofb
.eof_flags
= XFS_EOF_FLAGS_SYNC
;
662 xfs_icache_free_eofblocks(ip
->i_mount
, &eofb
);
666 current
->backing_dev_info
= NULL
;
669 xfs_iunlock(ip
, iolock
);
676 struct iov_iter
*from
)
678 struct file
*file
= iocb
->ki_filp
;
679 struct address_space
*mapping
= file
->f_mapping
;
680 struct inode
*inode
= mapping
->host
;
681 struct xfs_inode
*ip
= XFS_I(inode
);
683 size_t ocount
= iov_iter_count(from
);
685 XFS_STATS_INC(ip
->i_mount
, xs_write_calls
);
690 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
694 ret
= xfs_file_dax_write(iocb
, from
);
695 else if (iocb
->ki_flags
& IOCB_DIRECT
) {
697 * Allow a directio write to fall back to a buffered
698 * write *only* in the case that we're doing a reflink
699 * CoW. In all other directio scenarios we do not
700 * allow an operation to fall back to buffered mode.
702 ret
= xfs_file_dio_aio_write(iocb
, from
);
707 ret
= xfs_file_buffered_aio_write(iocb
, from
);
711 XFS_STATS_ADD(ip
->i_mount
, xs_write_bytes
, ret
);
713 /* Handle various SYNC-type writes */
714 ret
= generic_write_sync(iocb
, ret
);
719 #define XFS_FALLOC_FL_SUPPORTED \
720 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
721 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
722 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
731 struct inode
*inode
= file_inode(file
);
732 struct xfs_inode
*ip
= XFS_I(inode
);
734 enum xfs_prealloc_flags flags
= 0;
735 uint iolock
= XFS_IOLOCK_EXCL
;
737 bool do_file_insert
= 0;
739 if (!S_ISREG(inode
->i_mode
))
741 if (mode
& ~XFS_FALLOC_FL_SUPPORTED
)
744 xfs_ilock(ip
, iolock
);
745 error
= xfs_break_layouts(inode
, &iolock
);
749 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
750 iolock
|= XFS_MMAPLOCK_EXCL
;
752 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
753 error
= xfs_free_file_space(ip
, offset
, len
);
756 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
757 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
759 if (offset
& blksize_mask
|| len
& blksize_mask
) {
765 * There is no need to overlap collapse range with EOF,
766 * in which case it is effectively a truncate operation
768 if (offset
+ len
>= i_size_read(inode
)) {
773 new_size
= i_size_read(inode
) - len
;
775 error
= xfs_collapse_file_space(ip
, offset
, len
);
778 } else if (mode
& FALLOC_FL_INSERT_RANGE
) {
779 unsigned int blksize_mask
= i_blocksize(inode
) - 1;
781 new_size
= i_size_read(inode
) + len
;
782 if (offset
& blksize_mask
|| len
& blksize_mask
) {
787 /* check the new inode size does not wrap through zero */
788 if (new_size
> inode
->i_sb
->s_maxbytes
) {
793 /* Offset should be less than i_size */
794 if (offset
>= i_size_read(inode
)) {
800 flags
|= XFS_PREALLOC_SET
;
802 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
803 offset
+ len
> i_size_read(inode
)) {
804 new_size
= offset
+ len
;
805 error
= inode_newsize_ok(inode
, new_size
);
810 if (mode
& FALLOC_FL_ZERO_RANGE
)
811 error
= xfs_zero_file_space(ip
, offset
, len
);
813 if (mode
& FALLOC_FL_UNSHARE_RANGE
) {
814 error
= xfs_reflink_unshare(ip
, offset
, len
);
818 error
= xfs_alloc_file_space(ip
, offset
, len
,
825 if (file
->f_flags
& O_DSYNC
)
826 flags
|= XFS_PREALLOC_SYNC
;
828 error
= xfs_update_prealloc_flags(ip
, flags
);
832 /* Change file size if needed */
836 iattr
.ia_valid
= ATTR_SIZE
;
837 iattr
.ia_size
= new_size
;
838 error
= xfs_vn_setattr_size(file_dentry(file
), &iattr
);
844 * Perform hole insertion now that the file size has been
845 * updated so that if we crash during the operation we don't
846 * leave shifted extents past EOF and hence losing access to
847 * the data that is contained within them.
850 error
= xfs_insert_file_space(ip
, offset
, len
);
853 xfs_iunlock(ip
, iolock
);
858 xfs_file_clone_range(
859 struct file
*file_in
,
861 struct file
*file_out
,
865 return xfs_reflink_remap_range(file_in
, pos_in
, file_out
, pos_out
,
870 xfs_file_dedupe_range(
871 struct file
*src_file
,
874 struct file
*dst_file
,
879 error
= xfs_reflink_remap_range(src_file
, loff
, dst_file
, dst_loff
,
891 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
893 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
903 struct xfs_inode
*ip
= XFS_I(inode
);
907 error
= xfs_file_open(inode
, file
);
912 * If there are any blocks, read-ahead block 0 as we're almost
913 * certain to have the next operation be a read there.
915 mode
= xfs_ilock_data_map_shared(ip
);
916 if (ip
->i_d
.di_nextents
> 0)
917 error
= xfs_dir3_data_readahead(ip
, 0, -1);
918 xfs_iunlock(ip
, mode
);
927 return xfs_release(XFS_I(inode
));
933 struct dir_context
*ctx
)
935 struct inode
*inode
= file_inode(file
);
936 xfs_inode_t
*ip
= XFS_I(inode
);
940 * The Linux API doesn't pass down the total size of the buffer
941 * we read into down to the filesystem. With the filldir concept
942 * it's not needed for correct information, but the XFS dir2 leaf
943 * code wants an estimate of the buffer size to calculate it's
944 * readahead window and size the buffers used for mapping to
947 * Try to give it an estimate that's good enough, maybe at some
948 * point we can change the ->readdir prototype to include the
949 * buffer size. For now we use the current glibc buffer size.
951 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
953 return xfs_readdir(ip
, ctx
, bufsize
);
957 * This type is designed to indicate the type of offset we would like
958 * to search from page cache for xfs_seek_hole_data().
966 * Lookup the desired type of offset from the given page.
968 * On success, return true and the offset argument will point to the
969 * start of the region that was found. Otherwise this function will
970 * return false and keep the offset argument unchanged.
973 xfs_lookup_buffer_offset(
978 loff_t lastoff
= page_offset(page
);
980 struct buffer_head
*bh
, *head
;
982 bh
= head
= page_buffers(page
);
985 * Unwritten extents that have data in the page
986 * cache covering them can be identified by the
987 * BH_Unwritten state flag. Pages with multiple
988 * buffers might have a mix of holes, data and
989 * unwritten extents - any buffer with valid
990 * data in it should have BH_Uptodate flag set
993 if (buffer_unwritten(bh
) ||
994 buffer_uptodate(bh
)) {
995 if (type
== DATA_OFF
)
998 if (type
== HOLE_OFF
)
1006 lastoff
+= bh
->b_size
;
1007 } while ((bh
= bh
->b_this_page
) != head
);
1013 * This routine is called to find out and return a data or hole offset
1014 * from the page cache for unwritten extents according to the desired
1015 * type for xfs_seek_hole_data().
1017 * The argument offset is used to tell where we start to search from the
1018 * page cache. Map is used to figure out the end points of the range to
1021 * Return true if the desired type of offset was found, and the argument
1022 * offset is filled with that address. Otherwise, return false and keep
1026 xfs_find_get_desired_pgoff(
1027 struct inode
*inode
,
1028 struct xfs_bmbt_irec
*map
,
1032 struct xfs_inode
*ip
= XFS_I(inode
);
1033 struct xfs_mount
*mp
= ip
->i_mount
;
1034 struct pagevec pvec
;
1038 loff_t startoff
= *offset
;
1039 loff_t lastoff
= startoff
;
1042 pagevec_init(&pvec
, 0);
1044 index
= startoff
>> PAGE_SHIFT
;
1045 endoff
= XFS_FSB_TO_B(mp
, map
->br_startoff
+ map
->br_blockcount
);
1046 end
= (endoff
- 1) >> PAGE_SHIFT
;
1052 want
= min_t(pgoff_t
, end
- index
, PAGEVEC_SIZE
- 1) + 1;
1053 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, index
,
1058 for (i
= 0; i
< nr_pages
; i
++) {
1059 struct page
*page
= pvec
.pages
[i
];
1063 * At this point, the page may be truncated or
1064 * invalidated (changing page->mapping to NULL),
1065 * or even swizzled back from swapper_space to tmpfs
1066 * file mapping. However, page->index will not change
1067 * because we have a reference on the page.
1069 * If current page offset is beyond where we've ended,
1070 * we've found a hole.
1072 if (type
== HOLE_OFF
&& lastoff
< endoff
&&
1073 lastoff
< page_offset(pvec
.pages
[i
])) {
1078 /* Searching done if the page index is out of range. */
1079 if (page
->index
> end
)
1084 * Page truncated or invalidated(page->mapping == NULL).
1085 * We can freely skip it and proceed to check the next
1088 if (unlikely(page
->mapping
!= inode
->i_mapping
)) {
1093 if (!page_has_buffers(page
)) {
1098 found
= xfs_lookup_buffer_offset(page
, &b_offset
, type
);
1101 * The found offset may be less than the start
1102 * point to search if this is the first time to
1105 *offset
= max_t(loff_t
, startoff
, b_offset
);
1111 * We either searching data but nothing was found, or
1112 * searching hole but found a data buffer. In either
1113 * case, probably the next page contains the desired
1114 * things, update the last offset to it so.
1116 lastoff
= page_offset(page
) + PAGE_SIZE
;
1121 * The number of returned pages less than our desired, search
1124 if (nr_pages
< want
)
1127 index
= pvec
.pages
[i
- 1]->index
+ 1;
1128 pagevec_release(&pvec
);
1129 } while (index
<= end
);
1131 /* No page at lastoff and we are not done - we found a hole. */
1132 if (type
== HOLE_OFF
&& lastoff
< endoff
) {
1137 pagevec_release(&pvec
);
1142 * caller must lock inode with xfs_ilock_data_map_shared,
1143 * can we craft an appropriate ASSERT?
1145 * end is because the VFS-level lseek interface is defined such that any
1146 * offset past i_size shall return -ENXIO, but we use this for quota code
1147 * which does not maintain i_size, and we want to SEEK_DATA past i_size.
1150 __xfs_seek_hole_data(
1151 struct inode
*inode
,
1156 struct xfs_inode
*ip
= XFS_I(inode
);
1157 struct xfs_mount
*mp
= ip
->i_mount
;
1158 loff_t
uninitialized_var(offset
);
1159 xfs_fileoff_t fsbno
;
1160 xfs_filblks_t lastbno
;
1169 * Try to read extents from the first block indicated
1170 * by fsbno to the end block of the file.
1172 fsbno
= XFS_B_TO_FSBT(mp
, start
);
1173 lastbno
= XFS_B_TO_FSB(mp
, end
);
1176 struct xfs_bmbt_irec map
[2];
1180 error
= xfs_bmapi_read(ip
, fsbno
, lastbno
- fsbno
, map
, &nmap
,
1185 /* No extents at given offset, must be beyond EOF */
1191 for (i
= 0; i
< nmap
; i
++) {
1192 offset
= max_t(loff_t
, start
,
1193 XFS_FSB_TO_B(mp
, map
[i
].br_startoff
));
1195 /* Landed in the hole we wanted? */
1196 if (whence
== SEEK_HOLE
&&
1197 map
[i
].br_startblock
== HOLESTARTBLOCK
)
1200 /* Landed in the data extent we wanted? */
1201 if (whence
== SEEK_DATA
&&
1202 (map
[i
].br_startblock
== DELAYSTARTBLOCK
||
1203 (map
[i
].br_state
== XFS_EXT_NORM
&&
1204 !isnullstartblock(map
[i
].br_startblock
))))
1208 * Landed in an unwritten extent, try to search
1209 * for hole or data from page cache.
1211 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
) {
1212 if (xfs_find_get_desired_pgoff(inode
, &map
[i
],
1213 whence
== SEEK_HOLE
? HOLE_OFF
: DATA_OFF
,
1220 * We only received one extent out of the two requested. This
1221 * means we've hit EOF and didn't find what we are looking for.
1225 * If we were looking for a hole, set offset to
1226 * the end of the file (i.e., there is an implicit
1227 * hole at the end of any file).
1229 if (whence
== SEEK_HOLE
) {
1234 * If we were looking for data, it's nowhere to be found
1236 ASSERT(whence
== SEEK_DATA
);
1244 * Nothing was found, proceed to the next round of search
1245 * if the next reading offset is not at or beyond EOF.
1247 fsbno
= map
[i
- 1].br_startoff
+ map
[i
- 1].br_blockcount
;
1248 start
= XFS_FSB_TO_B(mp
, fsbno
);
1250 if (whence
== SEEK_HOLE
) {
1254 ASSERT(whence
== SEEK_DATA
);
1262 * If at this point we have found the hole we wanted, the returned
1263 * offset may be bigger than the file size as it may be aligned to
1264 * page boundary for unwritten extents. We need to deal with this
1265 * situation in particular.
1267 if (whence
== SEEK_HOLE
)
1268 offset
= min_t(loff_t
, offset
, end
);
1282 struct inode
*inode
= file
->f_mapping
->host
;
1283 struct xfs_inode
*ip
= XFS_I(inode
);
1284 struct xfs_mount
*mp
= ip
->i_mount
;
1289 if (XFS_FORCED_SHUTDOWN(mp
))
1292 lock
= xfs_ilock_data_map_shared(ip
);
1294 end
= i_size_read(inode
);
1295 offset
= __xfs_seek_hole_data(inode
, start
, end
, whence
);
1301 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1304 xfs_iunlock(ip
, lock
);
1321 return generic_file_llseek(file
, offset
, whence
);
1324 return xfs_seek_hole_data(file
, offset
, whence
);
1331 * Locking for serialisation of IO during page faults. This results in a lock
1335 * sb_start_pagefault(vfs, freeze)
1336 * i_mmaplock (XFS - truncate serialisation)
1338 * i_lock (XFS - extent map serialisation)
1342 * mmap()d file has taken write protection fault and is being made writable. We
1343 * can set the page state up correctly for a writable page, which means we can
1344 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1348 xfs_filemap_page_mkwrite(
1349 struct vm_fault
*vmf
)
1351 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1354 trace_xfs_filemap_page_mkwrite(XFS_I(inode
));
1356 sb_start_pagefault(inode
->i_sb
);
1357 file_update_time(vmf
->vma
->vm_file
);
1358 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1360 if (IS_DAX(inode
)) {
1361 ret
= dax_iomap_fault(vmf
, PE_SIZE_PTE
, &xfs_iomap_ops
);
1363 ret
= iomap_page_mkwrite(vmf
, &xfs_iomap_ops
);
1364 ret
= block_page_mkwrite_return(ret
);
1367 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1368 sb_end_pagefault(inode
->i_sb
);
1375 struct vm_fault
*vmf
)
1377 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1380 trace_xfs_filemap_fault(XFS_I(inode
));
1382 /* DAX can shortcut the normal fault path on write faults! */
1383 if ((vmf
->flags
& FAULT_FLAG_WRITE
) && IS_DAX(inode
))
1384 return xfs_filemap_page_mkwrite(vmf
);
1386 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1388 ret
= dax_iomap_fault(vmf
, PE_SIZE_PTE
, &xfs_iomap_ops
);
1390 ret
= filemap_fault(vmf
);
1391 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1397 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1398 * both read and write faults. Hence we need to handle both cases. There is no
1399 * ->huge_mkwrite callout for huge pages, so we have a single function here to
1400 * handle both cases here. @flags carries the information on the type of fault
1404 xfs_filemap_huge_fault(
1405 struct vm_fault
*vmf
,
1406 enum page_entry_size pe_size
)
1408 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1409 struct xfs_inode
*ip
= XFS_I(inode
);
1413 return VM_FAULT_FALLBACK
;
1415 trace_xfs_filemap_huge_fault(ip
);
1417 if (vmf
->flags
& FAULT_FLAG_WRITE
) {
1418 sb_start_pagefault(inode
->i_sb
);
1419 file_update_time(vmf
->vma
->vm_file
);
1422 xfs_ilock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1423 ret
= dax_iomap_fault(vmf
, pe_size
, &xfs_iomap_ops
);
1424 xfs_iunlock(XFS_I(inode
), XFS_MMAPLOCK_SHARED
);
1426 if (vmf
->flags
& FAULT_FLAG_WRITE
)
1427 sb_end_pagefault(inode
->i_sb
);
1433 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1434 * updates on write faults. In reality, it's need to serialise against
1435 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1436 * to ensure we serialise the fault barrier in place.
1439 xfs_filemap_pfn_mkwrite(
1440 struct vm_fault
*vmf
)
1443 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1444 struct xfs_inode
*ip
= XFS_I(inode
);
1445 int ret
= VM_FAULT_NOPAGE
;
1448 trace_xfs_filemap_pfn_mkwrite(ip
);
1450 sb_start_pagefault(inode
->i_sb
);
1451 file_update_time(vmf
->vma
->vm_file
);
1453 /* check if the faulting page hasn't raced with truncate */
1454 xfs_ilock(ip
, XFS_MMAPLOCK_SHARED
);
1455 size
= (i_size_read(inode
) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1456 if (vmf
->pgoff
>= size
)
1457 ret
= VM_FAULT_SIGBUS
;
1458 else if (IS_DAX(inode
))
1459 ret
= dax_pfn_mkwrite(vmf
);
1460 xfs_iunlock(ip
, XFS_MMAPLOCK_SHARED
);
1461 sb_end_pagefault(inode
->i_sb
);
1466 static const struct vm_operations_struct xfs_file_vm_ops
= {
1467 .fault
= xfs_filemap_fault
,
1468 .huge_fault
= xfs_filemap_huge_fault
,
1469 .map_pages
= filemap_map_pages
,
1470 .page_mkwrite
= xfs_filemap_page_mkwrite
,
1471 .pfn_mkwrite
= xfs_filemap_pfn_mkwrite
,
1477 struct vm_area_struct
*vma
)
1479 file_accessed(filp
);
1480 vma
->vm_ops
= &xfs_file_vm_ops
;
1481 if (IS_DAX(file_inode(filp
)))
1482 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
1486 const struct file_operations xfs_file_operations
= {
1487 .llseek
= xfs_file_llseek
,
1488 .read_iter
= xfs_file_read_iter
,
1489 .write_iter
= xfs_file_write_iter
,
1490 .splice_read
= generic_file_splice_read
,
1491 .splice_write
= iter_file_splice_write
,
1492 .unlocked_ioctl
= xfs_file_ioctl
,
1493 #ifdef CONFIG_COMPAT
1494 .compat_ioctl
= xfs_file_compat_ioctl
,
1496 .mmap
= xfs_file_mmap
,
1497 .open
= xfs_file_open
,
1498 .release
= xfs_file_release
,
1499 .fsync
= xfs_file_fsync
,
1500 .get_unmapped_area
= thp_get_unmapped_area
,
1501 .fallocate
= xfs_file_fallocate
,
1502 .clone_file_range
= xfs_file_clone_range
,
1503 .dedupe_file_range
= xfs_file_dedupe_range
,
1506 const struct file_operations xfs_dir_file_operations
= {
1507 .open
= xfs_dir_open
,
1508 .read
= generic_read_dir
,
1509 .iterate_shared
= xfs_file_readdir
,
1510 .llseek
= generic_file_llseek
,
1511 .unlocked_ioctl
= xfs_file_ioctl
,
1512 #ifdef CONFIG_COMPAT
1513 .compat_ioctl
= xfs_file_compat_ioctl
,
1515 .fsync
= xfs_dir_fsync
,