The write_lock is released after the first time already.

This leaves the hole in the log_write_and_flush logic, some
flush_lock/write_lock waiters could be hanging, until the
background redo log flush comes.

The possibility of that is relatively tiny though.