Of course, this only applies to linux systems where you don't trust local unprivileged users. Or software that you are running as an unprivileged user. So every system.
Good point, but
> they could start up a counterfeit one and collect your password.
you missed the part where I disable password logins on all of my boxes :-) The important point was that the system was already secure enough due to the key requirement, and moving the port was indeed just to stop the "doorknob rattling". If I suddenly find that a box I control is asking me for a password, I'm not going to just type my social security number in and hope for the best.
One could argue that using a port < 1024 makes it easier for the scanners to find, but frankly anything other than 22 (or a frequently scanned port) would work well enough.
It's really pretty easy to set up with iptables in Linux.
The only thing I do externally from sshd is have a cronscript run every 15 seconds to grep my authlog file to find any IP addresses that fail authentication and put them in a global blacklist for my PF firewall. It cuts down the ssh authlog noise considerably, and rarely accidentally blacklists myself.
Big ups for this -- since moving SSH to port 8022 I get zero bruteforce attacks. Even blacklisting tools like fail2ban can't get that kind of result. Of course it can't be the only defense, but I've always configured my SSH daemons to be key login only (no passwords). Moving SSH just cuts down on the CPU cycles burned at rejecting drone scans.
Every time I come across posts like this I consider maybe setting up a honeypot but I'm not sure what I'd do with the results other than look over them every few months for my own amusement. Are there any honeypots that can automatically forward information about the attacks to a central location?