Slashdot Mirror
New OpenSSL Man-in-the-Middle Flaw Affects All Clients
Trailrunner7 (1100399) writes 'There is a new, remotely exploitable vulnerability in OpenSSL that could enable an attacker to intercept and decrypt traffic between vulnerable clients and servers. The flaw affects all versions of the OpenSSL client and versions 1.0.1 and 1.0.2-beta1 of the server software. The new vulnerability could only be exploited to decrypt traffic between a vulnerable client and a vulnerable server, and the attacker would need to have a man-in-the-middle position on a network in order to do so. That's not an insignificant set of conditions that must be present for a successful attack, but in the current environment, where open wireless networks are everywhere and many users connect to them without a second thought, gaining a MITM position is not an insurmountable hurdle. Researchers who have looked at the vulnerable piece of code say that it appears to have existed, nearly unchanged, in the OpenSSL source since 1998.'
"but in the current environment, where open wireless networks are everywhere and many users connect to them without a second thought"
But if you have a man in the middle position, most of those same users would have just clicked "ignore" or typed yes to the "connect anyway" prompt.
"Ultimately you still need to get the encryption information across securely"
This is provably impossible with an active MITM attacker (Though a solution to passive listeners does exist and is in common use). If it could be done, we wouldn't be messing around with complicated certificate signing systems.
Just to be clear, versions 1.01 and 1.02(beta) is the same as saying "Any OpenSSL version released since early 2012", right? It sounds like the summary is trying to downplay the threat a little bit.
The more of these we find, the more secure OpenSSL will be. I hope we continue to find these kinds of problems and see them fixed. If open source has one strength, it's that when many skilled eyes DO converge on the code it can be tested and fixed far more quickly than a corporation with limited resources and only paid developers can do the same sort of debugging work. The trick is getting the eyes there in the first place.
This is a flaw, but it requires both ends use vulnerable OpenSSL versions. Which means your day-to-day life may or may not be affected that much.
I mean, if you use iOS, OS X, or Windows, you're more than likely NOT using OpenSSL on the client side (except say, if you use Firefox on Windows) - since Apple and Microsoft have their own SSL implementations. If you have an Android phone or tablet, then yes, it's quite likely an issue, and while both are popular, people generally don't use them that much for data (iOS traffic, after 7 years, has finally dropped to below 50% of all mobile traffic out there, despite Android outselling iOS by a huge margin). And nevermind the oddball Linux user.
So the real question is, how many people really ARE affected?
Heartbleed affects everyone because it exposes server secrets irrespective of the client side. But this vulnerability is only really present if both ends use OpenSSL.
Given all the open-source SSL/TLS security flaws (OpenSSL, gnutls, Apple SSL) that have come out these past few months - mostly thanks to renewed interest in hunting flaws, thanks to the Snowden revelations, I suspect - I hope that companies like Microsoft are also seeing this as a wake-up call driving them to do code reviews on their closed-source SSL/TLS code.
#DeleteChrome
I mean, if you use iOS, OS X, or Windows, you're more than likely NOT using OpenSSL on the client side (except say, if you use Firefox on Windows)
Ummm, firefox uses NSS, not OpenSSL: http://en.wikipedia.org/wiki/Network_Security_Services
Firefox uses NSS, not OpenSSL.
That's quite the parabolic sentence there. I hope you didn't give yourself whiplash.
Corporation, n. An ingenious device for obtaining individual profit without individual responsibility. - Ambrose Bierce
After reading the advisory from OpenSSL, I'm still confused by what is vulnerable and what isn't. The flaw requires both the client and server to be vulnerable. If the client is using OpenSSL, they're vulnerable for 0.9.8/1.0.0/1.0.1. But if the server is using OpenSSL, they're only vulnerable if using 1.0.1/1.0.2(beta). Yet the bullet list of recommendations points out that servers should upgrade even if they're using 0.9.8: * OpenSSL 0.9.8 SSL/TLS users (client and/or server) should upgrade to 0.9.8za. Let's say I have a server using 0.9.8 and client using 1.0.0. If I understand their explanation correctly, then this scenario is *not* vulnerable. Is that the same conclusion others would draw from their explanation?
Yes. If you're using libressl and the snapshot of the source you compiled is older than Thu Jun 5 15:46:24 2014 UTC, you're affected by this, too. See: http://www.openbsd.org/cgi-bin/cvsweb/src/lib/libssl/src/ssl/s3_clnt.c
"Ultimately you still need to get the encryption information across securely"
This is provably impossible with an active MITM attacker (Though a solution to passive listeners does exist and is in common use). If it could be done, we wouldn't be messing around with complicated certificate signing systems.
Correct. A MITM beats everything done over the wire. You need to secure your shit before you use the wire. You need a pre-shared key to encrypt the initial communication. A certificate sort of does this, but not really because we still trust them blindly and we initially accept them over the same wire. The proper way to do shit would be for you to go to your bank in person, for example, and generate 2 keys - one for you to use to talk to your bank and one for the bank to use to talk to you. You then use that key when establishing your first communication with the bank, and they use theirs. You can use whatever encryption you want, you can deploy a key-changing scheme, or a certificate scheme like we have now, whatever.
Your initial key exchange must be done securely. Doing it in person is the most secure way possible, but it's also the least convenient. Doing it over the wire is NEVER secure against a MITM.
Diffie-Hellman Key Exchange allows you to securely share a secret key over an insecure medium. Combining this with asymmetric cryptography to identify parties is how modern handshake protocols work.
The problem here how to trust Bob's asymmetric key really came from Bob and not Eve.
You are correct in that the ideal solution would be to talk to Bob over a different medium (like phone) and ask him if that is his key but there are ways to do this over the wire. As an example, several Linux distros sign their LiveCD images with cryptographic keys and post the keys' fingerprints on their web page. Can these be spoofed? Sure, hack the server hosting the files. That requires additional effort (and risk) though which would dissuade most cyber-criminals from attempting it.
Is this perfect security? No, but there is no such thing short of chucking whatever you want secured into a black hole.