Comment # 6 on bug 1231050 from Neil Brown

I think this is simply an omission from the switch statement in 
xs_tcp_setup_socket().  -ENOTCONN should be handled the same way as
-ECONNREFUSED set.

If it was, the error would be passed to xprt_wake_pending_tasks which end up
calling  call_connect_status() which does handle -ENOTCONN.

I'll look thorough the code again when I get a bit more time and ensure that
makes sense - then will suggest a patch upstream.

Thanks for the report.