"James A. Donald" <[EMAIL PROTECTED]> writes:

>However, seems to me that logging into the website using SRP is a non trivial
>refactoring, and not just a matter of dropping in TLS-SRP as a simple
>replacement of TLS-DSA-X509

I've discussed this with (so far) a small sample of assorted corporate TLS
users to get at least a general idea of what'd be involved.  At a very
abstract level all they see is "username + password + TLS" ->
"permitted/denied", the only change is that by moving the verification into
TLS this process happens a bit earlier than when it's done in HTML (and
obviously the failsafe nature means the other side never gets the password if
the auth fails).

At an implementation level it's also fairly simple, it's maybe 2-3 pages of
code added to my SSL implementation, and I spoke to another SSL developer who
gave similar figures.  All you're doing is mixing a little extra keying
material into the premaster secret, it's not a major piece of programming.

The real issues occur in two locations:

1. In the browser UI.
2. In the server processing, which no longer gets the password via an HTTP
   POST but as a side-effect of the TLS connect.

(1) is a one-off cost for the browser developers, (2) is a bit more complex to
estimate because it's on a per-site basis, but in general since the raw data
(username+pw) is already present it's mostly a case of redoing the data flow a
bit, and not necessarily rebuilding the whole system from scratch.  To give
one example, a healthcare provider, they currently trigger an SQL query from
an HTTP POST that looks up the password with the username as key, and the
change would be to do the same thing at the TLS stage rather than the post-TLS
HTTP stage.

With the folks I've discussed this with their concern has been far more "We
want this yesterday, why isn't it here yet" rather than "We can't integrate
this with our existing back-ends".


The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to [EMAIL PROTECTED]

Reply via email to