Also - SSH offers more than just encryption, but also data integrity - you can modify / manipulate a telnet session in ways you just can’t via SSH
(I do run Wireguard, it just feels like sometimes a VPN is a sledgehammer to solve a port forwarding problem)
You know what wireguard is?
> If telnet can use keypairs
Kerberos exists, so, yes, it can.
> You know what wireguard is?
I suppose if you prefer, I can write "over the network". The point is that the password leaves my machine. As a practical example: With password auth, if an attacker gets root on a server then they can read your password and log in to other machines. With SSH keypairs, this isn't possible (unless you go out of your way to forward an SSH agent, and even then there are mitigations).
>> If telnet can use keypairs
> Kerberos exists, so, yes, it can.
This sounds promising, and in fact at least one page I found about it claims that kerberos+telnet encrypts the session, at which point I don't immediately see what we need wireguard or ssh for. On the other hand, it looks like eg. GNU inetutils telnet doesn't support it? In fact, https://web.mit.edu/kerberos/krb5-1.5/krb5-1.5.4/doc/krb5-us... says
> The Kerberos V5 telnet command works exactly like the standard UNIX telnet program, with the following Kerberos options added:
which makes it sound like they've just made a special telnet variant with these features, at which point it rather feels like we've just re-invented ssh under a different name.
Which is just an abstraction of a password. Instead of storing it in your mind and sending it over the network you store it on your harddrive and use to calculate things on the network. Haven't you just moved the security boundary slightly? You're now also mixing up authentication and encryption into one package which isn't strictly an upside. Granted you can secure your key with a password or a yubikey but now you're messing with agents or you're typing that password an incredible amount.
I see your point. The dead simplicity of telnet session over encrypted tunnel still appeals to me in the face of the structures above. It also means you can elide the whole "how do I do this complicated port forwarding with ssh" question entirely.
> we've just re-invented ssh under a different name.
ssh is the best implementation of self signed security you can get. With kerberos we actually get a central authority and it separates authentication and encryption rather nicely without requiring user agents running as daemons.
I don't think that's functionally true. The important thing is that with telnet or password-auth ssh, you send the actual text of your password to the server. With ssh keys, the server and client do some magic crypto math to let you prove you control the private key without ever sending it. Therefore, a compromised server can steal a password, but not a ssh key.
(Yes, in theory perhaps you could do some fancier way of proving that you know a password without sending it, but 1. I'm no cryptographer, but the fact that openssh hasn't done this feels suggestive, and 2. that is once again a pretty big change for the nominal goal of keeping telnet)
> ssh is the best implementation of self signed security you can get. With kerberos we actually get a central authority and it separates authentication and encryption rather nicely without requiring user agents running as daemons.
Sure. Like, I've never thoroughly evaluated kerberos in depth (again, I'm no cryptographer), but I hear generally good things. My point is that by the time you have kerberos, you aren't really using what I would call telnet anymore, you're using something that acts like telnet but backs into a completely different authentication and communication system, and at that point you might as well use a completely different authentication and communication system without pretending to be telnet. This goes double because (open)ssh does support kerberos.
This use case is perfectly secure, because IBM mainframe/midrange telnet servers support telnet-over-TLS, and that’s what people run in production
For connecting to mainframes, SSH has no real advantage over TLS, and its major disadvantage is that there is no standardised way to transmit 3270/5250 data streams over it
But people looking for telnet traffic over the public Internet probably won’t even notice this, because they aren’t looking for telnet over TLS - which is difficult to distinguish from whatever else over TLS - and because almost all of it goes over VPNs not the public Internet
The two entities which have reported on this event are looking for tcp traffic on port 23, not TELNET protocol traffic. So indeed, as you say, if they are tunneled in VPN, or encapsulated or using an alternate port, tn3270 traffic will not be detect on port 23/tcp. Telnet over TLS is assigned to port 992, so any RFC-compliant implementation would be found there, and irrelevant, again, to the telnetd CVE reported this year.
There are two facets to January's incident: the vulnerability in the GNU implementation of telnetd, and the purported, widespread blocking of port 23. The original report went out because of the coincidence they perceived there, and especially because the latter preceded the disclosure of the vulnerability!
Mainframe tn3270 servers would not be subject to this vulnerability. If there had been a port filter in place, it only would've tripped-up the mainframes that still used port 23, which is evidently optional, and it says here that many admins want to keep AIX's telnetd bound to port 23 anyway.
So it is good to know that TELNET protocol, and its extensions, are alive and well. We may not actually know how many clients and servers implement the protocol itself, since MUDs made this a routine thing, but certainly the deployment of IBM systems is formidable, considering the sheer mass of the iron in their rack mounts.
And IMO, X.509 (used in TLS) is virtually superior over SSH’s bespoke certificate format in every way. You get both regular certificate pinning (like what SSH uses now) AND full certificate authority chains (if you want).
The main downside is that X.509 is more complex.
It doesn't do full chains, but SSH does have certificate authorities. I agree that the lack of intermediate CAs is a limitation (a CA can only sign a leaf node public key directly), but it's still super useful.
I do remember reading a long time ago telnet does/can support encryption. But when I looked at the systems I have access to, the manuals have no mention of that.