eap Mailing List: RE: Strawman -10/EMSK deletion requirement?

[Avi Lior (avi]

Hi Rafa,

You raise a good point in the following: 

> I have seen during discussions and also under my 
> understanding of draft-aboba-eap-keying-extns-00.txt that 
> either 1) AMSK would be tranported from AAA server to some 
> entity or 2) AMSK could be used as a root and cached by the 
> AAA server to derive new keys which would be eventually 
> transported to different entities.
> 
> will that decision between both cases be specified for 
> application? or would it be better to select one approach (it 
> seems people like second one)?

I have recently seen the need for both models.  IMO keeping the AMSK at
the AAA and EAP Peer and only transporting derived keys to external
entities makes a lot of sense but consider the following:

Supposing I have an two entities E1 and E2 that received DE-KEY(s)
derived from E-AMSK.  DE-KEY(s) are used to secure the communication
between E1 and E2.

Now E2 wants to delegate responsibility of communicate with E1 to E3
which it trusts.  E2 can pass the DE-KEY(s) to E3 but it would be better
to derive another set of keys for that communication.

As I understand it, E2 and E1 cannot derive DE-KEY'(s) from DE-KEY(s) --
since DE-KEY(s) are used for one purpose already and using them for
another purpose (key derivation) will make them weak.  So in this case
the only way to achieve this tranasaction is to involve AAA again.  This
could be very expensive.

Another approach is to send E1 and E2 E-AMSK so that they can generate
subsequent keys to deal with this scenario.  Of course you could also
send E1 and E2 DE-KEY(S) to secure their communication and also another
key derived from E-AMSK for the purpose of generating new keys.  But is
that necessary?  Why couldn't we just distribute E-AMSK.


> -----Original Message-----
> From: Rafa Marin Lopez [mailto:rafa [at] dif.um.es] 
> Sent: Friday, March 10, 2006 11:15 AM
> To: Jari Arkko
> Cc: Avi Lior; eap [at] frascone.com
> Subject: Re: [eap] Strawman -10/EMSK deletion requirement?
> 
> Hi Jari
> 
> >First of all, I think this part of the discussion is largely just a 
> >matter of vague definitions making it hard to make exact statements. 
> >E.g. where is the line between AAA as a "transport" and as an 
> >"application" that has intelligence to make decisions and grant 
> >resources such as keys?
> >  
> >
> I agree on you this. In my mind the AAA server (where EAP 
> server might be co-located) is (or should be) intelligent 
> enough to manage keys from users.
> 
> >Just to repeat what we have already established, I think we have 
> >consensus that (a) EMSK is generated by EAP methods and that (b) its 
> >not going to be shipped away to the access devices. In addition, it 
> >seems obvious that the document we are discussing will not 
> define how 
> >AMSKs are used or transported.
> >  
> >
> >So the conclusion is that the EMSK is kept somewhere between the EAP 
> >method and AAA transport layers.
> >  
> >
> So far, between EAP methods and AAA transpor layers (AAA 
> server?? ) we have EAP authenticator layer and EAP layer. An 
> as specified draft-ietf-eap-keying-10.txt both layers cannot 
> cache anything. I think it does not preclude the use of EMSK 
> but it establishes limits to the creation of AMSK. That is, 
> the AMSK should be created just in the moment the EMSK is 
> exported. If EMSK wants to be cached, that part of text 
> should be relaxed, no?
> 
> >I would note that this means that it is, in an IETF sense, 
> within a box 
> >and not transported via protocols. I don't think it makes a lot of 
> >sense to debate too long about what the internal structure 
> of the box 
> >is. I'd be happy saying that it resides in the EAP server. I 
> would in 
> >addition only say the following about the internal module 
> structures: 
> >The EMSK is not communicated either to the lower layer or the AAA 
> >transport layer.
> >But AMSKs derived from it can be,
> >and that any such derivation would have to be defined in future 
> >documents.
> >  
> >
> I have seen during discussions and also under my 
> understanding of draft-aboba-eap-keying-extns-00.txt that 
> either 1) AMSK would be tranported from AAA server to some 
> entity or 2) AMSK could be used as a root and cached by the 
> AAA server to derive new keys which would be eventually 
> transported to different entities.
> 
> will that decision between both cases be specified for 
> application? or would it be better to select one approach (it 
> seems people like second one)?
> 
> Regards.
> 
> --
> ------------------------------------------------------
> Rafael Marin Lopez
> Faculty of Computer Science-University of Murcia
> 30071 Murcia - Spain
> Telf: +34968367645    e-mail: rafa [at] dif.um.es
> ------------------------------------------------------
> 
>