Comment # 6 on bug 986395 from Neil Brown

> This is not true and it's the main cause of this bug.

Ahh, thanks.

So it seems there are two ways we could handle this.

One is to just use PUBLIC addresses for all NFS, or even all RPC connections. 
That would be easy, but feels ugly. It circumvents the purpose of the privacy
extensions.

The other is to re-establish a new connection periodically, making sure to do
so before the current address becomes invalid.

NFS will already close an idle connection after 5 minutes.  We could reduce
this timeout if the address is no longer preferred, and then force a shutdown
in the address is invalid.

How easy is it to determine the preferred/valid state of an address given a
bound socket?