Whenever there is a mount/unmount failure - there is a chance of calling the
callbacks functions once - transaction ail mount pointer is destroyed. So, it results
in NULL pointer exception followed by hang. So, before unmount of the log - flush all
the pending buffers.

 void
 xfs_log_unmount(xfs_mount_t *mp)
 {
+	int error = 0;
+	/*
+	 * Make sure all buffers have been flushed and completed before
+	 * unmounting the log.
+	 */
+	error = xfs_flush_buftarg(mp->m_ddev_targp, 1);
+	if (error)
+		cmn_err(CE_WARN, "%d busy buffers during log unmount.", error);
+	xfs_wait_buftarg(mp->m_ddev_targp);
+	

We do exactly that sequence before the xfs_log_unmount_write call on
umount.  Care to explain what code in xfs_log_unmount_write would
require this to be called again?

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs