Re: heimdal and replay cache

[Frank Cusack <fcusack@fcusack.com>]

On Fri, Mar 28, 2003 at 12:39:30AM +0100, Love wrote:
> Its not very useful since its not turned on by default and it doesn't
> truncate the log so will eventially fill up your disk.
> 
> Fixing either of the problems above shouldn't be too hard.

Yup, sounds easy enough.

> However, protocols that failes to verify that the content of the data
> stream is correct are flawed (integrity check) and reply cache wont really
> help you. A replay cache will only protect you against replay attacks only
> for protocols that are secure by design.

That sounds like a cop-out.  What if the returned (to the client) data
isn't especially important?

ssh is secure by design yet is vulnerable to replay if there is no cache,
assuming encryption type "none".

What if you're doing nfs and don't want/need the overhead, you just want
the strong authentication?

It sounds like you are extending the point I made in another thread; that
use of the replay cache is meaningless for applications that do integrity
checks or encryption of the data stream using the krb5 session key.  Here,
I am making the additional argument that not doing integrity/encryption
can be useful and legitimate, and requires a functional replay cache.

What do you think?
/fc