svn commit: samba r21903 - in branches/SAMBA_3_0/source/libsmb: .

[Jeremy Allison]

On Tue, Mar 27, 2007 at 06:03:38AM +1000, Andrew Bartlett wrote:
> 
> Urgh.  I *really* don't like this.  Why can't we negotiate it in the
> session setup itself?

I did consider this, but it's much easier to enforce a separate
encryption setup, as the RPC transport does. That way you're
not coping with layer violations all over the place, plus you
need a "rekey" mechanism, without having to do another sessionsetup
call, so as you've got to have a separate step anyway, so just 
make it the same for initial as replacement key setups.

That way to re-key, you just do the trans2 again (under the
encryption of the first key) and once it's complete you reset
the sealing engine with the new key and continue. Completely
transparent to the CIFS auth.

> Part of why I don't like it is the race conditions that may occurs in
> trying to authenticate a user twice.  What do you do if the first NTLM
> authentication succeeds, but the second fails?

The encryption setup failed.

Jeremy.