Comment by jerf
2 days ago
"Though, if someone gets that far, couldn't they also install a key logger on the users system?"
There are a ton of situations where that is not the case. They may be using directory traversal from something else to read a key without even necessarily being on the system. They may be on the system at 1am local time, and want to get in and get their job done before the user is even there. They may be on a server somewhere where someone left a key they shouldn't have. The attacker may have gotten enough of a secret to compromise some other secret store where the key is being held. They're probably on a system with user-level access only and that may not be enough to "just" install a keylogger, depending on how the system is set up and how the user accesses it. These are examples and not even remotely a full enumeration of all the possibilities. I won't tell you which ones, but some of these are things I've personally seen attackers take advantage of, so they're not just theory, either.
When you're under personal attack, not just getting popped by some vuln scanner scanning over the entire Internet, the situation becomes very different than a lot of people here on HN are used to. Ever been locked out of a system accidentally, then thought for a moment and strung together three other things to reach back into the system you were locked out of, like "oh yeah, I can push a software update to this automatic deployment system, which will run a bash script that checks the IP address and if it is this system restarts the ssh server, and so after 10 minutes we should be in"? Imagine someone who does that every day, all the time, as their full time job, and then imagine they're on a team of other people who also do it every day as their full time job, then imagine they've gotten a foothold into your system. Which, by the way, they immediately used to put a command-and-control client on your system, loaded with all kinds of exploits, and the ability to push arbitrary code to any number of systems at a time and all the tooling to use that as if they've been developing it for 20 years, which they have. What's the transitive closure of what they could work out how to access? The answer would probably surprise you.
I appreciate the additional insights, but the premise I'm pushing back on is whenever a SSH key is read, then the user account is by necessity compromised in order to do so. Given that level of a breach, there are myriad ways for an attacker to escalate privilege and exploit their access without worrying about a password on the SSH key. Namely, at that point, cracking the password on the SSH key is a tractable problem.
> They may be using directory traversal from something else to read a key without even necessarily being on the system.
At least on linux - to read the directory containing a SSH key requires the ability to also write to that directory, as the user. Therefore you can also write to '.bashrc' and all sorts of other places. I suspect Windows might have a larger attack surface, but nonetheless, a directory traversal that is able to read and write is also able to install a keylogger.
> They may be on a server somewhere where someone left a key they shouldn't have
Private key should never be transmitted over a network boundary. SSH key passwords can be bruteforced as well. Having a password on the SSH key, when the SSH key is somewhere it really should never have been, is closing the barn door after the horses have left.
> The attacker may have gotten enough of a secret to compromise some other secret store where the key is being held.
Again, getting access to the secret is enough to also have write access and be able to install a key logger. A password on the SSH key still does not help.
> They're probably on a system with user-level access only and that may not be enough to "just" install a keylogger
If a person has enough access to read a SSH key, they can also install a key logger for at least that user account. They are equivalent levels of compromise, a user account having its SSH key read is already compromised.
edit: addendum: There are certainly attacks that can only read the contents of a system, with root that can read the full system. It's just odd to think about, since at that rate the SSH keys being on a prod system is already such a big no-no. SSH keys really need to live exactly just on the personal devices of the people who own those keys - EG: it should never be the case that say a SQL injection attack that gains root level read permission over everything on a filesystem can then ever read SSH keys - cause those keys should never be on the remote system to begin with. Putting a password on private keys that are then copied to servers _is_ security theater; the keys ought to never be copied to a remote server to begin with.
I can tell you've not been involved in defending against an active attack. You, as the defender, do not get to play the game of "well, if I squint and read it that way, that attack wouldn't work". The attackers get to play "well, hey, if it turns out I do this and that and push it through the other thing, I get access". They are the ones who get to flow through any crack they can find. They are the ones who get to do logic chopping like you're trying to do. You don't get to argue "Well gosh, that team shouldn't have left that one permission open on that one system, that's not a best practice, if they'd followed best practices 100% of the time the attackers couldn't have gotten in...". Your job is to pick up the pieces.