Hi Jim,
It sounds like you are assuming that the Client would have to remember only one
security context, but it is possible that the client has several (old) security
contexts associated with one access token, for example if the RS has rebooted
several times. It would be unreasonable to require the client to store and
check all of those when deriving a new one (even if it is just the nonces).
Also, it is not required that the AS provides a new access token to the Client,
as it may be pre-generated and not expired. So in that case, if the client does
not keep track of access tokens and security contexts it has used, it might end
up with reused keys and nonces.
For both these reasons, we think that the client generated nonce is needed. If
you agree with that, we can discuss where to transport these nonces, for
example we can avoid echoing N1 if we use it as salt. Also, as you suggested,
it seems like a good idea to make this mechanism optional. We prepare for
discussion about this in Bangkok.
Thanks,
Francesca
On 25/10/2018, 23:25, "Jim Schaad" <i...@augustcellars.com> wrote:
> -----Original Message-----
> From: Francesca Palombini <francesca.palomb...@ericsson.com>
> Sent: Thursday, October 25, 2018 4:47 AM
> To: Jim Schaad <i...@augustcellars.com>; draft-ietf-ace-oscore-
> prof...@ietf.org
> Cc: ace@ietf.org
> Subject: Re: [Ace] WGLC for draft-ietf-ace-oscore-profile
>
> Hi Jim,
>
> Thank you for your review comments. We agree with all your points, and
> have opened issues: https://github.com/ace-wg/ace-oscore-profile/issues to
> get this fixed.
>
> Inline some detailed answer.
>
> Thanks,
> Francesca
>
> On 22/10/2018, 21:09, "Jim Schaad" <i...@augustcellars.com> wrote:
>
> * Section 1 - I understand the reasoning behind having the server
send back
> a nonce, although it would be good to have a description someplace
about
> why
> this is being done. (I would also make it optional as not all RS
need to do
> this.) I do not understand the reasoning behind having the client
send a
> nonce to the server.
>
> FP: The motivation for the nonce construction was in the security
> considerations, but I agree that having it in the Protocol Overview makes
> sense, so I opened an issue to fix that. The reason behind having the
client
> create a nonce is that we are protecting against an attacker replaying an
old
> RS message (containing an old nonce), which would provoke the creation of
> an old security context on the Client, and reuse of keys and nonces for a
> different (new) message.
So what you are looking at is a client which is going to "forget" the
current security context that it possesses for an RS but not the token,
re-posting the token to the RS because it apparently thinks that the RS has
forgotten the token and context without and then accepting the replay as being
valid.
I think that a client which has forgotten the security context is also
going to forget the token and thus need a new one.
I think that a client which is going to remember the security context is
also going to remember the nonce from the server and thus not rebuild the
context if the same nonce comes back.
If this needs to be explicit - then it should be made so.
>
> * Section 3.1 - This is more general than the section, but you should
not
> use the URI path in the text, instead you should be using the name
that is
> in the authz document.
>
> FP: Agreed, issue opened to fix this.
>
> * Section 3.2 - Does it really make sense to use 'COSE_Key' to
transport the
> key data? Would a different field name be better?
>
> FP: This was brought up several times, so we will make this change now.
>
> * Section 3.2 - Please provide a justification for the requirement
that the
> ids must be unique over the set of all clients and RS. I can see
that the
> client ids need to be unique on a single RS and RS ids need to be
unique for
> any given client but not the broader statement.
>
> FP: You are right, this requirement is too wide. We will replace with your
> suggestion.
>
> * Please add an explicit section on when a RS and a client should
discard
> the security context.
>
> FP: Ok we will add this. As mentioned in the issue, I have now only this 3
> cases in mind: Partial IV space ends (either C or RS); the kid context on
the RS
> side does not match with N1 (RS); C receives a number of Unauthorized (C),
> although that is a consideration/recommendations, details would be
> application specific. Do you see any other case relevant for this section?
Well - there are thing like expiration of the token which would make sense
as well. If you re-register the token because of an unauthorized and get
back a different nonce value.
>
> * Section 6 - Ok I'll bite - how does not echoing the nonce allow
for a
> man-in-the-middle attack given that the salt and shared secret are
still
> going to be known only to the C and RS and not to the MITM. I can
see a
> DOS
> attack being made, but that can be done even without this just by
causing
> the response to never be delivered.
>
> FP: Ok so our mistake here is to use the term MitM, so to solve this we
will
> replace with "on-path attacker". The following sentence should be correct
> with that fix "Moreover, the client echoes the nonce created by the RS,
> which verifies it before deriving the Security Context, and this protects
> against an adversary acting as an on path attacker and substituting the
nonce
> in transit from client to RS to provoke the creation of different Security
> Contexts in the client and RS." Yes this is a DOS, but could also lead to
reuse
> of keys and nonces, as mentioned before.
See my above comment. But there are no problems here with the server doing
anything interesting as it would presumably not have generated the same nonce
value as the last time. This means that the client would generate new keys
only if the previous comment was wrong.
I just don't see this as being a problem. Merits discussion in Bangkok
Jim
>
> * Appendix - I am not sure that I think that the EDHOC profile should
be in
> this document as oppose to being in it's own document. The fact that
we
> have not even tried to get this to work in any of the interop tests
means
> that I am less sure that it is well baked.
>
> FP: Agreed, we will remove this from this document and move it to its own
> document
>
> Jim
>
>
> > -----Original Message-----
> > From: Ace <ace-boun...@ietf.org> On Behalf Of Jim Schaad
> > Sent: Monday, October 8, 2018 2:35 PM
> > To: ace@ietf.org
> > Subject: [Ace] WGLC for draft-ietf-ace-oscore-profile
> >
> > The chairs believe that the set of documents dealing with the OAuth
> > framework for constrained environments is nearing the point that we
> should
> > be able to advance it to the IESG for publication. We therefore
want to
> > have a full list of issues that need to be dealt with at the Bangkok
> > meeting.
> >
> > This starts a 2 week WGLC for draft-ietf-ace-oscore-profile
> >
> > We know that the following issues are outstanding:
> >
> > draft-ietf-ace-oscore-profile:
> > * No current known issues
> >
> >
> > Jim & Roman
> >
> >
> > _______________________________________________
> > Ace mailing list
> > Ace@ietf.org
> > https://www.ietf.org/mailman/listinfo/ace
>
>
_______________________________________________
Ace mailing list
Ace@ietf.org
https://www.ietf.org/mailman/listinfo/ace
