4

In SRP-6 $B$ is calculated as $B=kv+ g^b, k=3$. What is the purpose of $k$, and why was it fixed as $3$?

(In SRP-6a, this value $3$ is replaced by $k = H(N,g)$, but this question is about SRP-6.)

Community
  • 1
user236501
  • 213
  • 1
  • 5

1 Answers1

3

The purpose is to prevent a two-for-one guessing attack, where an active adversary, impersonating the server, can test two password guesses per attempt. The attack and why the multiplier prevents it is described in Section 2 of the SRP-6 paper (ps). (According to MacKenzie, it was discovered by Bleichenbacher.)

In brief, the attack goes like this:

  • Instead of $B = v + g^b = g^x + g^b$ with a random $b$, the attacker calculates $B = g^{x_1} + g^{x_2}$ with two password guesses.
  • The client uses the value $B - g^x$, meaning if $x=x_1$ they get $g^{x_2}$ and if $x=x_2$ they get $g^{x_1}$.
  • The attacker calculates two session keys, based on $x=x_1, b=x_2$ and $x=x_2, b=x_1$. If either of these matches with the $M_1$ sent by the client, they have found the password.

If the attacker does not know the discrete logarithm of $k$, i.e. the number $l$ for which $g^l = k$, they cannot try two guesses at once with the version 6 protocol where $v$ is multiplied by $k$.

The paper shows why $k=3$ is a safe choice for generic $g$ and $N$.

(The hashed $k$ fixes it for maliciously chosen $g$ and $N$ as well.)

Community
  • 1
otus
  • 32,132
  • 5
  • 70
  • 165
  • I notice that there necessarily still exists a fast classical attack for any given $g$ and $N\hspace{-0.02 in}$. $:$ It seems to me that using k = 3 + H(N,g,A) would make it plausible that there is no fast classical attack. $:$ Is there a better reason than "it takes slightly more computation" for not doing that? $;;;;$ –  Aug 19 '14 at 20:46
  • @RickyDemer, isn't that just equivalent to switching to another $H$? I don't think it can help. – otus Aug 20 '14 at 05:32
  • That's certainly not obviously equivalent, since it makes $k$ depend on the client's message. $\hspace{.91 in}$ –  Aug 20 '14 at 05:41
  • 1
    @RickyDemer, oh I missed that you added the $A$. I don't think the $3$ in your formula matters, but how is adding $A$ better than what SRP-6a does? – otus Aug 20 '14 at 05:46
  • It makes the resulting protocol be such that, as far as I can see, there does not necessarily exist a fast classical two-for-one guessing attack on a large $N$ that can handle arbitrary pairs of candidate passwords. $;$ –  Aug 20 '14 at 06:03
  • 1
    Whereas for SRP-6a one exists, but cannot be found quickly? I guess that would be an advantage, but not necessarily worth the cost of an extra hash on each run and serializing the protocol ($B$ cannot be calculated in advance). – otus Aug 20 '14 at 06:36