diff mbox series

[13/16] xfs: add zero-around controls to iomap

Message ID 20181107063127.3902-14-david@fromorbit.com (mailing list archive)
State New, archived
Headers show
Series xfs: Block size > PAGE_SIZE support | expand

Commit Message

Dave Chinner Nov. 7, 2018, 6:31 a.m. UTC 
From: Dave Chinner <dchinner@redhat.com>

For buffered writes into block size > page size filesystems, XFS
will need to trigger zero-around for delayed allocation mappings
and writes over unwritten extents.

Unwritten extents will only remain unwritten until the first data
gets written back, so we have to ensure that any write mapping
triggers zero-around for them.

Delayed allocation occurs over holes, which then require the data in
the page cache to completely cover them at write-out time. Hence we
have to trigger write-around for these mappings, too.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
---
 fs/xfs/xfs_iomap.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
diff mbox series

Patch

diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index 27c93b5f029d..35626855e207 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -575,6 +575,15 @@  xfs_file_iomap_begin_delay(
 				goto out_unlock;
 		}
 
+		/*
+		 * If we are about to write to an unwritten extent and the
+		 * the block size is larger than page size, then we need to make sure
+		 * that the caller zeroes blocks partially covered by data.
+		 */
+		if (got.br_state == XFS_EXT_UNWRITTEN &&
+		    mp->m_sb.sb_blocksize > PAGE_SIZE)
+			iomap->flags |= IOMAP_F_ZERO_AROUND;
+
 		trace_xfs_iomap_found(ip, offset, count, 0, &got);
 		goto done;
 	}
@@ -647,6 +656,8 @@  xfs_file_iomap_begin_delay(
 	 * them out if the write happens to fail.
 	 */
 	iomap->flags |= IOMAP_F_NEW;
+	if (mp->m_sb.sb_blocksize > PAGE_SIZE)
+		iomap->flags |= IOMAP_F_ZERO_AROUND;
 	trace_xfs_iomap_alloc(ip, offset, count, 0, &got);
 done:
 	if (isnullstartblock(got.br_startblock))
@@ -1107,6 +1118,15 @@  xfs_file_iomap_begin(
 	if (flags & IOMAP_ZERO)
 		goto out_found;
 
+	/*
+	 * If we are about to write to an unwritten extent and the
+	 * the block size is larger than page size, then we need to make sure
+	 * that the caller zeroes blocks partially covered by data.
+	 */
+	if (imap.br_state == XFS_EXT_UNWRITTEN &&
+	    mp->m_sb.sb_blocksize > PAGE_SIZE)
+		iomap->flags |= IOMAP_F_ZERO_AROUND;
+
 	if (!imap_needs_alloc(inode, &imap, nimaps))
 		goto out_found;