next up previous
Next: About this document ... Up: No Title Previous: PAK-R

PAK-RY

Finally we present the protocol used in the PAK distribution that is resilient to server compromise. That is, when a server's password file is compromised, the passwords are still hidden, and one cannot use the compromised password file to successfully impersonate a user (at least not without having to run a dictionary attack on the password file).

We do this by having the server store a salt value s and a password verifier for a user A of the form V=gv, where $v=v[A]=H_0'(A,s,\pi)$. Then the PAK-RY protocol is run to determine that both the user and server know V, and additionally, we have the user prove that he knows the discrete log of V (i.e., v). This is done using a non-interactive Schnorr proof. We call this the PAK-RY protocol.


 
Figure 6: The PAK-RY protocol
\begin{figure}
{
 \begin{center}

 $\begin{array}
{lcr}
 Alice \mbox{(user)} ...
 ... K = H_3(A, s, X, Y, Z, V) \\ 
 \end{array}$
 \end{center}
 }
 
\end{figure}

As stated above, this is the protocol used in the PAK distribution (PAKtelnet and PAKftp). It is slightly different than the PAK-RY protocol from the paper (coming soon) in that the salt value replaces the identity of the server. This could only affect a user in the following situation: The user uses the same password on two servers B and B', but somehow does not reveal the password to B'. However B' is allowed to set the salt, and sets it to be the same as the salt value for B. Then the user has the same V value for B and B', thus allowing B' to impersonate B.

However, it seems most likely that a user will either

1.
trust a server completely, i.e., type in the plaintext password on the server itself, and trust that server to erase the plaintext password and only store the password verifier, or
2.
not trust a server, i.e., generate random salt and then the password verifier itself, and send these to the server.
In the first case, it would be easier for a cheating server to simply remember the plaintext password of the user. In the second case, the cheating server can't run the attack because it can't set the salt value.

If the bad situation we described above is possible, then the design of the protocol must be changed back to include the server identity in the computation of the password verifier.


next up previous
Next: About this document ... Up: No Title Previous: PAK-R
Phil MacKenzie
9/29/2000