Slashdot Mirror
Keep SSH Sessions Active, Or Reconnect?
borjonx writes "Is it safer to log out of an SSH session, and re-establish it later, or just keep the connection open? Like many of you, I use OpenSSH to connect to my Slackware Linux boxes remotely from Linux and WinXP (putty.exe) clients. At home and at work, I wonder if it would be safer to just leave the connection open (my clients are physically secured, the servers limit connections with hosts.allow). Is it more secure to re-establish the connection over an insecure link (big bad internet) where people can sniff that handshaking, or is it more secure to just remain connected? I connect 1 to 4 times per day, most days."
It is good that you are concerned about security. It is bad that you are asking Slashdot for security advice.
If I told you that it is far more secure to leave your connection open all day, would you take my word for it?
Do some research on the subject. Learn what terms like IND-CPA, IND-CCA, and IND-CCA2 mean and how to evaluate this situation for yourself. In terms of security, blindly following someone's advice is the less secure choice.
Ask Slashdot: Where bad ideas meet poor googling skills.
This is the wrong place to ask. I doubt we'll get a single response from a person on the cutting edge of cryptanalysis who can give you a meaningful answer on the relative strength of Diffe-Hellman vs AES, which is what your question comes down to.
Realistically, it makes no difference. Both mechanisms are highly secure, cutting edge cryptographic systems. I doubt that either have been broken by anyone. If there is someone powerful enough to break those systems *and* keep the discovery secret, they're waaay above the league where they'd be interested in your SSH connections. That is, unless you work for the military of a major world power and are known to be transmitting valuable intel.
The ability to secretly break DH or AES would be such a huge weapon that they wouldn't use it unless the stakes were high enough to risk losing the advantage if their capability were detected. Somehow, I think your connections to your servers aren't that important.
I hate printers.
This is the wrong place to ask. I doubt we'll get a single response from a person on the cutting edge of cryptanalysis who can give you a meaningful answer on the relative strength of Diffe-Hellman vs AES, which is what your question comes down to.
No, it doesn't.
Currently, the relative strength of both of those is "much stronger than the chance of some kind of user screwup". Something like typing a password and "enter" into the wrong window, connecting to the wrong server, being tired and cranky about having to get work done and so ignoring a KEY CHANGE warning, etc is far more likely than an attacker breaking AES or Diffie-Hellman to get to your data.
So, do what you can to minimize the chance of user error. To me, that probably means stay connected (I'm willing to be persuaded otherwise, though, whether in general or for particular work patterns).
rage, rage against the dying of the light
Breaking the crypto is almost assuredly not the weakest point in your connection. I'd stay connected,
You're right about the crypto not being a concern, but I think the bigger danger is that he gets up to go to the bathroom or printer or something and he forgets to lock the client machine. Cert change alerts are hard to ignore, at least with OpenSSH. Logout.
Penny - plain text accounting
Reconnect. Leaving the sessions constantly open means if your workstation is compromised, you may have compromised the servers as well.... at least you've increased the risk profile of the servers.
Connect as needed - use proper key management and passwords, etc.
That has no bearing on comparing logout/login vs. staying logged in. Yes, the very very first handshake can be bad (there are methods to mitigate, but that's beyond the scope of this discussion), but once you establish that trust, logging out does not break it.
XML is like violence. If it doesn't solve the problem, use more.
Great, now you have something that will work for 5% of the cases in which people need to remotely connect.
I never suggested that this is a general crypto solution for the masses. I am pointing out that if you think you do need to security offered by an OTP system, it's not really that hard to communicate the pads securely. If I can't afford a $1000 plane ticket to deliver the pad in person, chances are my data isn't important enough to need that level of security in the first place.
Cutting edge cryptanalyst here (PhD in IBE, works for major global security company)
A disclaimer: Conventional crypto is not my game anymore (post-quantum crypto is the way of the future). As any expert will tell you, I am not an expert, but I'll try to shed some light on some aspects of the discussion here.
To begin, we first have to make some reasonable assumptions about the choice of keys in SSH2. There exist known weak primes and weak generators in the DH (Diffie-Hellman) protocol that can be exploited. Assuming the SSH key generator algorithm is smart enough not to choose any known weak primes or generators, we can say the following.
The default OpenSSH implementation uses a 2048-bit prime order field. The security of the DH key exchange protocol is based on the discrete logarithm problem, of which the best known conventional attacks are generally O(sqrt(n)). ie. in laymans terms, roughly equivalent to a keysearch of 2^1024. Quantum computers are another story, but unless you're transferring data that will need to be secure in the order of decades (like you're that important), I doubt you have much to worry about in that regard for a while to come.
AES (the symmetric cipher used in SSH) uses by default 128 bit keys. There are no known attacks on AES better than brute force (ie. on average a keysearch of 2^127, since on average only half the keys will need to be checked before finding your session key). I would say however that there is a far greater chance of someone in the future strongly breaking AES than someone strongly breaking DH. New techniques for attacking symmetric cryptosystems appear all the time (see: Linear cryptanalysis, Differential Cryptanalysis, Impossible Differential Cryptanalysis, Integral Cryptanalysis, Boomerang attacks etc.) whereas DH is based on a very well known and studied number theory problem. Crypto-God Bruce Schneier seems to think AES will be broken in the future, but not enough to allow practical cryptanalysis of traffic.
It's hard to make any definite statements about a comparative analysis of the two schemes, due to the constants (or indeed polynomial terms) of the above complexity statements being unknown. From a purely theoretical standpoint, DH is the weakest link due to it having a better attack than brute force. However, when given this specific set of values to be used, the real-world security comparison is generally seen to be in the favour of DH with 2048 bit prime rather than AES-128. One author suggests Regardless, cycling the session key seems to be free (I can't find any known attacks that use past key exchanges). The SecSH RFC suggests session key cycling after a gigabyte of data, however more often can't hurt.
In short, you don't need to be worried about either DH or AES for a long time to come, but in terms of security, cycling the session key more often than necessary (ie. logging out and back in again) is probably technically more secure. As others have said in this thread however, crypto is very very rarely the weakest link. I'd be looking far more closely at the security of the computers involved than worrying about the crypto being broken.
ssh-agent is its own profound issue: by keeping the key unlocked in a format usable by other shells or software, it makes all your unlocked keys available to anyone who can gain access to the same server as you. This means that I, as an admin, can probably borrow the ssh keys of anyone I've educated in how to use ssh-agent on any of my systems.
Isn't that _convenient_ for me?
Come on people what is this? Tagging such a story where someone asks about some security where some obscure attack may be possible and then tagging it "you aren't that important"?!
This is the same messageboard that wants https for everything, even for this board.
This is the same board that seems to hold privacy above all.
And on top of it, it is full of nerds that tend to love to go into this kind of obscure detail.
And then tag it "you aren't that important" implying "what are you worried about", or with a little further stretch "you have nothing to hide, so don't bother". This is quite ridiculous.
To me I am the most important person in the world, and I would like to live safe and secure. The poster is likely the most important person to himself, and he also wishes to live safe and secure. I wouldn't go as far as poster does, but that's besides the point. He does want to go this far, and has a genuine question that many may consider over the top for personal security but which may have consequences for entities that are under constant attack, where any minute attack vector may mean the difference between safe and 0wned.
"youarentthatimportant" is the worst tag I have ever seen. It's denigrating at best. It's stupid, and shows lack of respect for other people. I may hope this was intended as a joke and a joke alone.