Bug 2811

The situation:

1. Server is configured with ClientAliveInterval=3 and ClientAliveCountMax=5
2. Client machine runs "ssh -24xaTnN -i/some/keyfile ... -oBatchMode=yes -oExitOnForwardFailure=yes -oServerAliveInterval=3 -oServerAliveCountMax=5 -Rxx:localhost:zz user@server". Note the purpose of this SSH connection is only to setup the reverse tunnel (there is no pty or interactive session). This ssh process is run as a background daemon by a separate monitoring script.
3. A TCP connection is established on the reverse-forwarded port xx from server back to client through the SSH connection, and then is idle (no traffic).
4. Network connectivity is lost between server and client - suddenly, no traffic can get through (so no clean shutdown of TCP connection)
5. The ssh client eventually detects the failure and exits, then is restarted automatically (by a separate monitoring script) and retries repeatedly.
6. However, the sshd process running on server never exits (THIS IS THE BUG)
6. Once connectivity is restored, the client's retrying eventually creates a successfull ssh connection, but it then fails immediately because the old sshd process is still hanging around listening on localhost:xx and the new connection has -oExitOnForwardFailure=yes.

So it LOOKS as if the ClientAlive* keep-alive setting on server is not working at all...

In trying to understand this behavior, I'm guessing what's really happening is that the ClientAlive* keep-alive setting is actually working and does detect the lost connection, but the existence of the established reverse-forwarded TCP connection is preventing sshd from exiting - just like when you exit an interactive SSH session with a still-connected forwarded TCP connection and it hangs until that forwarded TCP session closes.

First of all - is this behavior expected?

If not - consider this a bug report.

If so - consider this a feature request, namely, it would be nice if there were a way to configure around this.

E.g., add new options ClientKillForwardedConnectionsOnClose and ServerKillForwardedConnectionsOnClose that mean "if the main SSH connection dies (or completes normally), forcibly kill any forwarded connections immediately instead of waiting for them to close.