Re: Detecting Brute-Force and Dictionary attacks

Quoting Greg Metcalfe <metcalfegreg@xxxxxxxxx>:

On Thursday 09 November 2006 10:45, fabio wrote:
The idea is simple and good, but there's a problem in its
implementation: usually modern systems doesn't compare the password you
write with the saved password; instead, they compare an hash of your
password attempt with the saved hash of your current password. By
design, two similar string have strongly different hashes. So you can't
compare two hashes and say if they correspond to two similar words.
Greets,
Fabio

Sebastiaan Veenstra wrote:
> Hi,
>
> I didn't read the whole discussion about this issue but I came up with
> an idea which might be usefull to detect brute force attempt. By
> storing the passwords a certain user has used in the past along with
> the current password you could be able to compare to password (by
> pattern matching) used at the login attempts with the passwords list.
> If the password used differs significantly (this excludes typos) from
> the entries in the password list, there could be a possible brute
> force attempt. The reason for storing the previous passwords is that
> people tend to use every password they've used in the past when they
> forgot their password. Maybe this idea can be used along with the
> other methods of detecting brute force attempts. Anyway, it's just a
> random thought.
>
> Greets,
>
> Sebastiaan
Most diplomatic of Fabio. Here's an example, using md5 hashing. Results will
be similar, if any sort of valid crypto hash function is used.

# echo 12345678 | md5sum
23cdc18507b52418db7740cbb5543e54
# echo 12345679 | md5sum
0f4fd7804fbbcf67df5dc8ef8dc946fb

The difficulty still lies in whether you choose to use modified binaries to
record the submitted password (and there are huge downsides to doing this in
anything other than a lab environment) or take the decision that x number of
failed logins constitutes an attack. That's generally a wise move, depending
upon your weighting scheme (time, IP number, etc.) and threat model.

Even if you take the risky step of recording submitted passwords, you still
have to write analysis software (and that's not nearly as simple as it
sounds), and decide what to do with the results. There Are Issues.

Personally, I'm against the whole idea of authenticating via passwords, at
least as corporate password policies are currently and commonly implemented.
But that's all about dealing with a threat model that may have nothing to do
with your situation.

Nothing said here constitutes good advice. Everything depends upon context..
How you protect the Big Red Button will be far different than how you protect
generic httpd logs.

Regards,
--
Greg Metcalfe


It is my personal opinion that evaluating the passwords so closely, such as mentioned in the previous email to Greg's, would open yourself up to a far more complicated world than you might be thinking. By evaluating the passwords, you are utilizing more resources than normal during the authentication process. You would also open yourself to the fact that the more system resources dedicated to this evaluation process, the easier it would be for someone to perform a denial of service attack to your system.
In the past, when I've had people attempting to attack my systems, the easiest way to tell is the number of login attempts against the frequency of the attempts. I typically end-up with a log full of attempts to login, which makes it quite obvious.
To sum it up, make sure that if you do take this approach, you're extremely defensive. You would want any evaluation you perform to be done in the background to determine if there's an attack, not actively while the login session is in progress.

John Hall