Re: [TLS] about the PWD Proposal

[Yaron Sheffer <[e-mail]>]

In a somewhat related protocol (IKEv2 with PACE for authentication), we 
added an appendix on password salting: 
http://tools.ietf.org/html/draft-kuegler-ipsecme-pace-ikev2-08#appendix-B. 
The setting there is a little different from tls-pwd, because IKE is 
normally symmetric (peer to peer vs. client to server). But some of our 
analysis does apply. In particular, dynamically replacing the short 
credentials by a long shared secret ( 
http://tools.ietf.org/html/draft-kuegler-ipsecme-pace-ikev2-08#appendix-B.3) 
would mitigate much of the risk.

Thanks,
     Yaron

-----
  Hi,
  I am interested in the PWD proposal at the meeting, and run through it(http://tools.ietf.org/html/draft-harkins-tls-pwd-01#section-3.4),
  and have the following concerns:

1. In the document, it  says both Client and Server should be able to compute PE, which is derived from username, password, salt,etc.
    That means Client or the user should also remember the salt value, in that case, salt is meaningless and can be replaced by a longer password.
    This is quite different from the widely known usage of salt, e.g. in UNIX operating system, salt is not required to remember by the user, but by server to fetch locally by reference to the username.
   And as far as I know, I am sure the authors also know that, other password based key exchanges protocols don't involve salt, but store a hash value of password among other things locally.

> I realize that this is very different from the usage in UNIX operating systems, but it is necessary, as the UNIX (or other) operating system running the web server has to store the password. Storing the password in the clear on as an unsalted hash is worse than storing a salted hash (or "base" in the draft's terminology), so that is what the TLS server does. The use of base instead of password is not to protect the password during handshake, but to protect it during storage on the server.
>
> Having said that, I think it is now very practical to take a list of the 100,000 most popular passwords on the web, a database of 10,000 salted hashes (the file described in section 3.4), and find all the matches. That's 1,000,000,000 operations, which would take 15 minutes on a MacBook Air (1.4 GHz). I think the days of saying that a database of salted hashes makes theft of the password file irrelevant should be over. Yes, I know that with unsalted hash it would be practical to test the 100,000,000 most popular passwords.
>
> session cookies somehow (e.g., via the BEAST attack): but if you don't
> have my OBCs' private keys then you can't actually make use of my
> cookies.
>
> With OBC cookie theft is not enough.  With OBC the attacker must get
> at the private keys for the OBCs.
>
> Also, with OBC you get closer to true logout: destroy ("forget") your
> OBC private keys and your sessions are as good as dead.  Moreover, the
> user-agent, not the server, is in control of maximum session lifetime,
> since there's no point in the server setting cookies that will outlive
> the OBCs.
> I we can find no other use case, we might want to rename it to "Origin Bound Cookie".
>
> But that would require some interesting interaction between layers.
>
> Are all cookies received while using OBC considered (by the client!) bound to the OBC? If not, the client might use an old cookie after a logout event, leading to the server falsely detecting an attempt to forge a cookie. Maybe we need a new keyword on the cookie header.
>
> Where is the "logout" button?  If it's part of the website (in the "canvas") how does the HTML/HTTP tell the browser to log out? Probably best to have it in the Chrome for sites where OBC are supported.