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

[Avi Lior (avi]

Hi Joe,

I have something to say about both your replies to I included them
together.


> 
> Hi Madjid
> 
> Seems that there a lot of good reasons for keeping EMSK around after 
> it is used to generate AMSKs.  Hopefully everyone gets that.
> 
[Joe] No. After the AMSKs are generated what are you going to do with
the EMSK?  The purpos of the EMSK is to generate AMSKs.   

I think it is sufficient to say that exactly what you said: "The purpose
of the EMSK is to generate AMSKs."

Why should we generate all the AMSKs at once?  A session may have
several applications that may require AMSK.  Some applications will be
used 100% of the time, some application will be used rarely.  Why should
an implmeentation genrate an AMSK and cache it just in case it will need
the key?  That doesn't make sense.


> -----Original Message-----
> From: Salowey, Joe [mailto:jsalowey [at] cisco.com] 
> Sent: Monday, March 06, 2006 12:46 PM
> To: Avi Lior; Nakhjiri Madjid-MNAKHJI1; Rafa Marin Lopez; 
> Bernard Aboba
> Cc: eap [at] frascone.com
> Subject: RE: [eap] Strawman -10/EMSK deletion requirement?
> 
>  
> 
> > -----Original Message-----
> > From: Avi Lior [mailto:avi [at] bridgewatersystems.com]
> > Sent: Thursday, March 02, 2006 12:16 PM
> > To: Salowey, Joe; Nakhjiri Madjid-MNAKHJI1; Rafa Marin 
> Lopez; Bernard 
> > Aboba
> > Cc: eap [at] frascone.com
> > Subject: RE: [eap] Strawman -10/EMSK deletion requirement?
> > 
> > Hi Joe and Madjid,
> > 
> > The only reason for caching the EMSK is if you have to generate an 
> > AMSK for another application associated with the current session.
> > 
> > So the question is do you have to generate all possible 
> AMSKs and then 
> > delete the EMSK or can you generate the AMSKs as needed?
> > 
> [Joe] I understand that there may be need for caching of 
> certain quantities.  I think Jari explained it well in a 
> separate message.  In my opinion you should try to identify 
> the applications in use and the keys needed as early as 
> possible.  
I disagree.  I think from a resource point of view we should not genrate
keys for something that we may not be using.  At any rate this needs to
be an implmeentation choice and we should not prohibit this choice in a
standard.

> If this is not possible then caching of the MSK 
> may be necessary.  

I think you mean EMSK not MSK. Right?

If yes, we agree then caching of EMSK may be necessary.  Great!!!

I don't think we should write a specification that MUST require an
implmeentation to dervie all the keys that it is necessary on the chance
that the keys may be needed.  

I do agree that EMSK MUST ONLY BE USED for key derivation (AMSKs) and
MUST NOT be transported out of the EAP Authentication Server layer.

> Do we have examples where the caching of 
> the EMSK is necessary.  

What if I don't have an example today?  Does that mean that in the
future I won't? or that somewhere in the world someone does have such a
need?

> In the case where key caching is used 
> the cache needs to be managed.  If we are caching EMSKs then 
> the document that describes EMSK usage must describe how key 
> cache can be managed. 

I agree.
 
> 
> > Finally I have an slight issue with the statement that a 
> KEY should be 
> > deleted so that it wont be used for other purposes.  I find that 
> > statements like that are an over specification.  What makes anyone 
> > think that making a statement that "X should/must be 
> deleted" is going 
> > to be taken more seriously then "X should/must only be used for 
> > purpose X and should/must not be transported out of Y"?
> > 
> > I think that whether it gets deleted or not is an implementation 
> > issue.
> > 
> 
> [Joe] The goal of the architecture is to prevent the re-use 
> of keys for multiple different purposes.  I agree that this 
> is not necessarily the same a deletion.  It is very important 
> that keys derived from the EMSK don't collide purposes across 
> applications.  

I agree and therefore lets stick to the important issue and drop the
deletion argument.  The important issue being that EMSK shall be used
for AMSK derivation and for no other purpose.

> What happens in a particular AMSK branch of 
> the key hierarchy is not necessarily as important since this 
> branch is cryptographically separated from the rest of the 
> system through the AMSK derivation.  I think the current text 
> could probably do better in motivating why it is good 
> practice to delete keys.  

I don't think we need to talk about key deletion at all, except when it
comes to the key's end-of-life or end of session.  I think we need to
stress exactly what you said.

 
> > Finally if we insist on saying that X must be deleted we 
> should really 
> > say that X must be securely deleted.  There is a difference.
> > When a key
> > is securely deleted the software takes extra care that the memory 
> > location containing the key is first erased before the 
> memory location 
> > is freed to the heap.  Perhaps we can state that somewhere.
> > 
> > > -----Original Message-----
> > > From: Salowey, Joe [mailto:jsalowey [at] cisco.com]
> > > Sent: Thursday, March 02, 2006 2:49 PM
> > > To: Nakhjiri Madjid-MNAKHJI1; Rafa Marin Lopez; Bernard Aboba
> > > Cc: eap [at] frascone.com
> > > Subject: RE: [eap] Strawman -10/EMSK deletion requirement?
> > > 
> > >  
> > > 
> > > > -----Original Message-----
> > > > From: Nakhjiri Madjid-MNAKHJI1
> > [mailto:Madjid.Nakhjiri [at] motorola.com]
> > > > Sent: Thursday, March 02, 2006 8:44 AM
> > > > To: Salowey, Joe; Rafa Marin Lopez; Bernard Aboba
> > > > Cc: eap [at] frascone.com
> > > > Subject: RE: [eap] Strawman -10/EMSK deletion requirement?
> > > > 
> > > > Hi Joe,
> > > > 
> > > > Thanks for the email. I think you responded to the old
> > piece of the
> > > > email from Rafa and I am to blame for causing that
> > confusion, as I
> > > > kept that part to provide context.
> > > > Again, my question was why an entity needs to delete EMSK after 
> > > > generating the first AMSK (or first set of AMSKs?)? This
> > > seems to be
> > > > the requirement regardless of two options:
> > > > 
> > > > 1) keep EMSK at EAP layer, create AMSK at EAP layer based
> > > request from
> > > > AAA layer, delete EMSK Immediately (this means EAP layer
> > must have
> > > > KDFs for AMSK=KDF(EMSK, etc)
> > > > 2) push EMSK down to AAA layer at backend server, create
> > > AMSK at AAA
> > > > layer and delete EMSK immediately (this means AAA layer 
> must have
> > > > KDFs)
> > > >
> > > [Joe] If the AAA layer contains the AAA client and AAA 
> server then 
> > > the EMSK should not be available to this layer, if the AAA layer 
> > > means something else then I don't know about (1).
> > > The AMSK should be generated in the EAP and exported, option (2).
> > >  
> > > > 
> > > > In both cases we require deletion of EMSK after
> > generation of AMSK,
> > > > why?
> > > > 
> > > [Joe] To minimize the chance of exposure of the EMSK.  Why do you 
> > > need to cache it? Could you generate and cache an AMSK
> > instead?  
> > > 
> > > 
> > > > Thanks,
> > > > 
> > > > Madjid
> > > > 
> > > > -----Original Message-----
> > > > From: Salowey, Joe [mailto:jsalowey [at] cisco.com]
> > > > Sent: Wednesday, March 01, 2006 5:17 PM
> > > > To: Nakhjiri Madjid-MNAKHJI1; Rafa Marin Lopez; Bernard Aboba
> > > > Cc: eap [at] frascone.com
> > > > Subject: RE: [eap] Strawman -10
> > > > 
> > > >  
> > > > 
> > > > > -----Original Message-----
> > > > > From: Nakhjiri Madjid-MNAKHJI1
> > > [mailto:Madjid.Nakhjiri [at] motorola.com]
> > > > > Sent: Wednesday, March 01, 2006 2:38 PM
> > > > > To: Rafa Marin Lopez; Bernard Aboba
> > > > > Cc: eap [at] frascone.com
> > > > > Subject: RE: [eap] Strawman -10
> > > > > 
> > > > > Madjid>>Again, why is deletion of EMSK after generation of
> > > > > one AMSK is a
> > > > > requirements. What if we need to create multiple AMSKs
> > > and that at
> > > > > multiple occassions?
> > > > > 
> > > > > 
> > > > > 
> > > > > Well, actually, lower layer authenticator implementation
> > > > should expect
> > > > > (MSK,EMSK) in the case EAP method is executed by the 
> standalone 
> > > > > authenticator or (MSK,AMSK) in the case EAP method is
> > executed by
> > > > > backend authentication server. If it receives (MSK,EMSK)
> > > > should create
> > > > 
> > > > > AMSK and delete EMSK. If it receives (MSK,AMSK) , that's
> > > > all, correct?
> > > > 
> > > > [Joe] Not really, strictly speaking the lower layer
> > > shouldn't expect
> > > > to receive the EMSK as that would break mode independence.  An 
> > > > architectural description should not have the lower layer
> > receiving
> > > > the keys. From a supplicant perspective it must appear the same 
> > > > whether an external EAP-Server or a collocated EAP server
> > is used.  
> > > > Now I don't know what is going on inside your box, it
> > could all be
> > > > monolithic when a internal EAP server is used but that
> > shouldn't be
> > > > visible to the external world.  If I was interested in
> > > cryptographic
> > > > module separation I might not be too happy if you shared
> > > the EMSK with
> > > > the lower layer.
> > > > 
> > > > > 
> > _________________________________________________________________
> > > > > To unsubscribe or modify your subscription options,
> > please visit:
> > > > > http://lists.frascone.com/mailman/listinfo/eap
> > > > > 
> > > > > Arhives: http://lists.frascone.com/pipermail/eap
> > > > > 
> > > > 
> > > _________________________________________________________________
> > > To unsubscribe or modify your subscription options, please visit:
> > > http://lists.frascone.com/mailman/listinfo/eap
> > > 
> > > Arhives: http://lists.frascone.com/pipermail/eap
> > > 
> > 
>