ext3_sync_fs will start a commit and will wait on that commit.  This
means that on its return, all journalled file data has been dirtied and
exposed to sync_inodes_sb().  Which is sufficient to fix the umount
data loss problem.