Re: Searching for a special challenge&response algorithm

Carlos Moreno schrieb:

>> Sorry, DoS is the wrong description. I had more something similar to an SSH
>> brute force attack in mind.
>
> Ok. A tiny nitpick: SSH login attempts do not constitute a brute
> force attack -- that falls more in the category of a dictionary
> attack (they're not trying all possible names with all possible
> passwords -- they're only trying a list of the commonly used
> combinations).

> Ok, but then -- a C-R implies that you're simply trusting a remote
> user as soon as they prove to you that they know the right password;
> so, there's nothing that the protocol or the implementation can do
> about it; if the passwords *can be guessed*, then the remote
> attacker will be able to log in.

All passwords can be guessed. It is just a matter of time (theoretically).

> The solution is: make the passwords non-guessable. Restrict the
> frequency with which logins may be attempted (with configurable
> exceptions, so that "privileged" IP addresses can always attempt
> the login).

That is exactly what I want do to but not on a per IP-basis (may be
additionally to it) but with the outlined c-r algorithm.
 
> Youcan also do what people do with SSH: do not use password
> authentication at all -- use public-key authentication;

Thanks but I already know that. The problem is that I don't want to secure
SSH - I am just looking for a algorithm which could slow down password
guessing/brute force attacks on a more general level.

I got the idea while studying PKCS#5 (algorithm for deriving crypto-keys
from a password). It slows down password guessing/brute force attacks by
adding 1000+ encryption rounds. Applying PKCS#5 to the usual c-r
implementations the random number sent from the server to the client would
be the password and the PKCS#5-password-derived-key would be the correct
response. The only problem is that testing the response if it is the right
one needs the same steps as creating it (on the client side). This makes it
unusable for an client-server scenario for c-r.

So I am searching for a replacement algorithm. It should be act like
backdoor-function (mathematically). The server has the secret for using
this backdoor - the client doesn't have it. The server only grants access
to the authentication function to clients which successfully applied the
backdoor-function to the challenge. Because of the missing secret the
client can not use the backdoor and has to use therefore much CPU power to
compute the function. The server knows the secret and can compute the
result quite easily. When the client returns the response to the challenge
the server compares it with it's own computation and grant access to the
authentication method if it is equal to the result computed by the server.

That is what I want to achieve...

Jan
Received on Mon May 1 02:06:09 2006