xfs: remove incorrect log write optimization
authorChristoph Hellwig <hch@infradead.org>
Mon, 28 Jun 2010 14:34:34 +0000 (10:34 -0400)
committerAlex Elder <aelder@sgi.com>
Mon, 26 Jul 2010 18:16:45 +0000 (13:16 -0500)
We do need a barrier for the first buffer of a split log write.
Otherwise we might incorrectly stamp the tail LSN into transactions
in the first part of the split write, or not flush data I/O before
updating the inode size.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
fs/xfs/xfs_log.c

index 5a5a54ad36850e623b4b5f36a8c0ef4682140383..f309e1404fd6314d27446c2d9756adbc80c67976 100644 (file)
@@ -1423,11 +1423,8 @@ xlog_sync(xlog_t         *log,
        XFS_BUF_BUSY(bp);
        XFS_BUF_ASYNC(bp);
        bp->b_flags |= XBF_LOG_BUFFER;
-       /*
-        * Do an ordered write for the log block.
-        * Its unnecessary to flush the first split block in the log wrap case.
-        */
-       if (!split && (log->l_mp->m_flags & XFS_MOUNT_BARRIER))
+
+       if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
                XFS_BUF_ORDERED(bp);
 
        ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);