Slashdot Mirror
Flaw Made Public In OpenSSH Encryption
alimo20 writes "Researchers at the Royal Holloway, University of London have discovered a flaw in Version 4.7 of OpenSSH on Debian/GNU Linux. According to ISG lead professor Kenny Patterson, an attacker has a 2^{-18} (that is, one in 262,144) chance of success. Patterson tells that this is more significant than past discoveries because 'This is a design flaw in OpenSSH. The other vulnerabilities have been more about coding errors.' The vulnerability is possible by a man-in-the-middle intercepting blocks of encrypted material as it passes. The attacker then re-transmits the data back to the server and counts the number of bytes before the server to throws error messages and disconnects the attacker. Using this information, the attacker can work backwards to figure out the first 4 bytes of data before encryption. 'The attack relies on flaws in the RFC (Request for Comments) internet standards that define SSH, said Patterson. ... Patterson said that he did not believe this flaw had been exploited in the wild, and that to deduce a message of appreciable length could take days.'"
OpenSSH 5.2 was released in February already which has builtin countermeasures against this form of "attack." Next.
Whew. Glad I use Telnet.
MABASPLOOM!
From the article it seems that it is more of a design flaw of SSH and not specifically OpenSSH
And in other news it also appears that the word "chink" is banned in the comments section.
Anyone else remember when Unix was the usual target, and MS/DOS the "safe" OS?
-Space for rent
Did you read the article?
It indicates that it effects SSH in general, not only one particular implementation.
Go green: turn off your refrigerator.
If the flaw is in the design of SSH, wouldn't all OS's be effected? Why does this only effect Debian?
The 2^-18 is _really_scary_
The 'first 4 bytes', not so much.
So, meh. Of course true hardcore cryptanalysts are sure to be already ditching OpenSSH or maybe piping it through GPG first.
how long until
". . .discovered a flaw in Version 4.7 of OpenSSH on Debian/GNU Linux."
"The attack relies on flaws in the RFC (Request for Comments) internet standards that define SSH"
So which is it, is it an implementation specific bug, which is specific to OpenSSH on Linux specifically, OpenSSH on all O/Ses, or is it a flaw in the RFC, which should make it exploitable on *all* implementations of SSH, shouldn't it? How can a flaw in the standard only be exploitable in one version of one implementation of the standard on one specific target OS?
According to TFA, "OpenSSH version 5.2 contain[s] countermeasures". For Ubuntu users, note that Ubuntu 8.04 LTS (Hardy) is using the vulnerable version 4.7. Versions 8.10 (Intrepid) and above appear to use version 5.1.
Does anyone know whether 5.1 contains the flaw and/or the "countermeasures"?
Also, can any security gurus comment on the danger level here? It sounds like there is a low-probability to access a small amount of information... but the very existence of this vulnerability makes me uncomfortable. Also, why does it mention Debian specifically? Don't most distros use an OpenSSH package based on the exact same design? Shouldn't they all be vulnerable?
http://www.cpni.gov.uk/Docs/Vulnerability_Advisory_SSH.txt
So is it a flaw in the design of SSH or in the Debian patched OpenSSH implementation? If it's the former (as the quote seems to imply), why does it matter what SSH implementation, OS they found it on? Shouldn't it affect everyone?
sic transit gloria mundi
FTA, this vulnerability is addressed in newer versions of OpenSSH, not by fixing the specification, but by employing some kind of workaround to make it impractical. I didn't know that from the summary, since I don't really keep current on where OpenSSH is with their releases.
It seems like this attack has an awfully small chance of success. I am wondering if there is that small chance of success to decode a message after many days, or if because of the small chance of success, it would take multiple days before you had anything.
If this is really something that has almost no chance of working at all--period, I'm not too worried. If it is something that makes the encryption breakable in a few days, that's a pretty big deal and am surprised that it didn't get outed sooner as a flaw.
Can/should the RFC be revised to close this hole? Are there other (perhaps more obvious) examples of weakness in the RFC that have implementation-specific fixes applied?
This flaw was published in Nov 2008 with simple configuration fix, and OpenSSH released a default fixed version in March 2009.
Also, this attack gives only 4 bytes of unencrypted output after crashing your session many thousands of times, which is sure to be noticed. If you were repeating the exact same network traffic in millions of SSH sessions, an attacker might get something interesting after weeks of crashing your sessions. It's just one of the lamest exploits I've seen, worth mitigating eventually, but not worth all the press it's getting, especially 6 months after release...
The fix is simple, just use CTR mode encryption instead of CBC, or upgrade to OpenSSH 5.2 or later.
For more details go to the OpenSSH security page.
Blessed are the pessimists, for they have made backups.
It is because that happened to be the system that they found the vulnerability on.
Nothing more than that, really.
This is not your lawn. The property line clearly indicates...
wait a minute...
you are on MY LAWN!
Anyone else remember when stone tablets were the usual target, and cave drawings considered "safe"?
-Space for rent
This is a technicality that no one cares about anymore I think even Stallman gave up on it.
ftfa:
"The flaw, which lies in version 4.7 of OpenSSH on Debian/GNU Linux"
"Patterson said his group had worked with OpenSSH developers to mitigate the flaw, and that OpenSSH version 5.2 contained countermeasures."
They are unclear on whether or not it's only debian's repos that are affected, so I'd suggest upgrading to 5.2 or later.
Command to determine openSSH version: ssh -V Output: OpenSSH_5.1p1 Debian-5ubuntu1, OpenSSL 0.9.8g 19 Oct 2007 I assume if you get the same results as I did you are not affected because you are running version 5.1.
Yes. That's why we now have replaced telnet/rsh/rcp and authenticated FTP with ssh and scp, NIS with LDAP+Kerberos, /etc/shadow, authentication in NFS, support for other filesystems like CIFS, etc.
Microsoft, for their part, haven't changed all that much.
My blog
Implies man-in-the-middle already, and the odds of figuring 32 bits of a particular packet are somewhat low. The worst case i could think about is when i pass my password or a very specific short data (cc number?). That particular data must be the decripted one, and in the 1/256k odds of happening, IF i have someone in the middle actively trying to get it. With that chances, the attacker well could expend his time playing lotto that have more chances to win.
Of course, those numbers are regarding a specific distribution/ssh version, could be different for other versions, but still, looks somewhat hard to happen ever.
> Patterson said that he did not believe this flaw > had been exploited in the wild, and that to > deduce a message of appreciable length could take > days.
Is my social security number a "message of appreciable length?"
Probably not on its own. Full packed it would take 33 bits, 11 bytes (88 bits, though if the attacker knew for sure that an SSN was being sent in those bytes the search space would not significantly greater than the 33 bits) if represented in pure ASCII text with separators.
As each attempt to read each 32 bits has a 11/2^18 chance of success, and assuming failure of one attempt does no extra clue as to which other patterns to try next, each 4 byte block is going to take on average 131,072 connections to infer from the server response so for the 11 byte ASCII string that is an average attack length of 393,216 connections.
While that isn't going to take long (at 4.5 connections per second you are looking at a day), any message being sent containing your SSN is going to be significatly longer than the SSN on its own so I wouldn't worry just yet.
We are still in "it would be a lot easier for the attacker to raid your bins, burgle your house, or steal records from your bank" territory here. Though there is the chance that someone improve the attack (or already has) so be vigilant and apply updated SSH packages as soon as practical once your distribution offers them.
This was never a real threat, just another piece of Academic FUD. To be vulnerable as an interactive ssh user you would have to ignore 100,000 aborted sessions to expose 14 bits of plaintext, I think I would notice, and block the attacker.
There are a whole suite of cyphers, including AES aka Rijhndael are configurable, have you done yours?, and not vulnerable.
Finally the protocol is trivially fixed.
Now I for one, whilst I have the highest respect for the work done by people like Ross Anderson and Schnieer am fed up to the back teeth with alarmism from governments, NGOs and academics -- all of which add up to give us more money.
If you dont know these researchers were working for the UK equivalent of Homeland Security and failed to inform SSH of the details of the attack, doubtless quoting National Security.
These people who parade nonsense should be tarred an feathered and sent on the next rail.
Yes. That's why we now have replaced telnet/rsh/rcp and authenticated FTP with ssh and scp, NIS with LDAP+Kerberos, /etc/shadow, authentication in NFS, support for other filesystems like CIFS, etc.
Microsoft, for their part, haven't changed all that much.
You're talking about changes since the tools developed in the 1980s. Microsoft completely replaced their consumer OS, moving from the Win95-based platform to the infinitely more secure NT-based platform, moved to AD, replaced an unsecure method of storing passwords with a secure one, invented CIFS, moved to a filesystem where all actions can be audited, etc.
The non-Linux Unixes, for their part, are mostly the same as they were in 1985 (that Oracle OS, Sol-something or other, being the one exception).
Socialism: a lie told by totalitarians and believed by fools.
Any of them that have successfully set up IPSEC? I've used it to secure an application which screen-scrapes a telnet session and uses ftp to transfer files.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
I assume the parent failed every numerate course he ever took, and never took any kind of statistics for idiots course, and has not bothered to read and understand the attack.
... then I have to tolerate 100000 failures/connection.
An continues to spread FUD.
The attack requires a Man in the Middle scenario, if I ssh into a random machine the MiM is __VERY__ unlikely to work unless the NSA and my ISP are in cahoots and I dont notice that connection takes a nonsensically longer time (15min v 3s),
Now comes the idiot who talks about rsync, and scripted connections, which, for brevity I did not address and tries to use multipicative probability leveling to try to convince us we have a problem; sorry now you need 1000+ MiM attacks, and fail to notice that they are failing, so you also dont notice a net cable kicked out, or a cable cut by a backhoe.
Look guys, get real, if you want to use statistics or probability learn some and understand contingent probability which is essential to risk evaluation.
My understanding is that one packager screwed up one package (openssl), which (while a _big_ screw-up) would hardly amount to a "history of fucking up packages".
Can you cite any other examples that would indicate such a trend?
To mitigate on clients and servers: in /etc/ssh/sshd_config and /etc/ssh/ssh_config and/or any ssh clients you use, add:
Ciphers aes256-ctr,aes192-ctr,aes128-ctr
MACs hmac-sha1
To verify:
ssh -v host
look for the output:
debug1: kex: server->client aes128-ctr hmac-sha1 zlib@openssh.com
debug1: kex: client->server aes128-ctr hmac-sha1 zlib@openssh.com
You are particularly interested in the aesXXX-ctr segment. If that specifies a CBC mode, then you probably need to change that server's config. For the blowfish-using type, I'm uncertain of the attack's applicability to blowfish-cbc. YMMV. For server testing, you probably want to make sure your ssh client isn't forcing the CTR mode. To test that, do
ssh -v -o Ciphers=aes256-cbc,aes192-cbc,aes128-cbc,aes256-ctr,aes192-ctr,aes128-ctr
and look for similar debugging output as above.