[2933] in Kerberos

home help back first fref pref prev next nref lref last post

Re: About principals' secret keys & attacks

daemon@ATHENA.MIT.EDU (Scott Dawson)
Mon Dec 20 18:41:21 1993

To: Carlos Horowicz <carlos@athea.ar>
Cc: sdawson@engin.umich.edu, kerberos@MIT.EDU
In-Reply-To: Your message of "Mon, 20 Dec 1993 17:20:19 -0300."
Date: Mon, 20 Dec 1993 18:28:38 -0500
From: Scott Dawson <sdawson@engin.umich.edu>


>  1. I think there should be nothing "recognizable" if timestamps and/or random
>  numbers go in the authenticator.  

Timestamps are recognizable because I pretty much know what time it is
and if I decrypt and get anywhere near that time I know I've hit the
right key.

The thing is, in krb4 at least, what is in the authenticator is
something of the form { username, address, timestamp }Ku,s

The chaining works like this:

you get a tgt which looks like:

{ Ku,tgs , {Ttgs}Ktgs }Kuser

none of the packet contents here is recognizable if I just guess at Kuser.

However, you now send a message to the tgs asking to talk to service Y.  When 
you do this, you also send the authenticator for the tgs.  What you send
looks like:

{ username, address, timestamp }Ku,tgs along with {Ttgs}Ktgs


So, here's the attack if I have sniffed your packets.  I decrypt the
initial tgt using dictionary attack.  I don't know that what I got was
any good.  I use the Ku,tgs which I get to decrypt the authenticator
which you sent to the tgs.  If I see { username, address, timestamp }, 
all of which are recognizable, then I have guessed your user key correctly.  
If not, I move on to the next guess and repeat the procedure.

I can't do this though if I'm sitting here in Michigan and have no way
of sniffing your packets.  The best I can do in that case is hit your
kerberos server and ask for a tgt.  Then I keep decrypting and getting
some TGT which I can't recognize.  The only way for me to see whether
it's good is to try it out (by trying to get a ticket for a specific
service).  If I do this 10 billion times though, you might notice it and
start filtering my traffic.

>  2. Under krb5, does the attacker have less possibilities than here, to run
>  a dictionary of possible passwords and maybe hit the password ?

not sure.  I'm not familiar with krb5.

-Scott

home help back first fref pref prev next nref lref last post