Re: [Emu] Working Group Last Call for draft-ietf-emu-tls-eap-types

[Mohit Sethi <[e-mail]>]

Hi:


On 7/4/22 22:28, Alan DeKok wrote:
> On Jul 4, 2022, at 2:56 PM, Mohit Sethi <[e-mail]> wrote:
>> Some late last call comments:
>>
>> 1. For PEAP and TTLS, it is no longer possible to use client certificates without phase 2 authentication. Does the same restriction apply to TEAP. I think it would make sense to add an explanation on why this was done?
>    I think it should be done for TEAP, too.  The text could mirror TEAP:
>
> The practice of using client certificates with no "inner tunnel" method is forbidden when TEAP
> is used with TLS 1.3.  If there is a requirement to use client
> certificates with no inner tunnel methods, then EAP-TLS should be used
> instead of TEAP.
>
>> How about using server certificate in phase 1 and client certificate in phase 2 (with no further inner methods)? TEAP supports such behavior for TLS 1.2 to hide client identity?
>    My inclination is to forbid that for TLS 1.3.  However, TEAP can do multiple inner method authentication (host + user), unlike TTLS and PEAP.  Which means that it is possible to do TEAP with inner EAP-TLS, and then also inner "other method".
>
>    Perhaps the document could just say it's NOT RECOMMENDED to do TEAP with only inner EAP-TLS.
>
>> Would it not be better to simply mandate at least one NewSessionTicket message?
>    I can see situations where resumption isn't used.
You don't have to use resumption. You just need to send NewSessionTicket 
as a protected success indication. I thought the reason for for 
forbidding use of client certificates without phase 2 authentication was 
lack of protected success indication. Perhaps there is some other 
reason? This is why I wrote in my original comment: "I think it would 
make sense to add an explanation on why this was done? ".
>
>> I can think of a TTLS deployment where some peers only authenticate with client certificates while other peers authenticate with client certificates and one-time passwords in phases 2. Depending on the type of authentication, peers are put in different VLANs.
>    I'm not sure how that affects NewSessionTicket?
>
>    One-time passwords in phase 2 are a significant issue with EAP methods.   Where people have tried to deploy them, there are many issues.  They just don't work in practice for a host of reasons.
One-time password was only an example. Perhaps a wrong one. I just 
wanted to highlight that we were forbidding TTLS/PEAP and now TEAP 
deployments from using client certificates without phase 2 
authentication. But there can be deployments where some peers use client 
certificates without phase 2 authentication while other peers use client 
certificates with additional phase 2 authentication (password, etc.).
>
>    The simplest one is that the supplicant caches the inner identity, and re-uses it when the WiFi signal is lost / re-gained.  While a NewSessionTicket will help here, it won't always help.  There are situations where the end user will be prompted again (sometimes repeatedly!) in a short period of time.
>
>    I would suggest that one-time passwords in phase 2 are NOT RECOMMENDED, unless the passwords can be automatically entered without human intervention.
>
>> 2. I don't think this makes sense:
>>
>>
>>> Implementations SHOULD NOT use inner identities which contain an NAI
>>>     realm.
>>>
>>
>>>   if the inner identity does contain an NAI realm, the inner
>>>     realm SHOULD be either an exact copy of the outer realm, or be a
>>>     subdomain of the outer realm.
>>>
>> Eliot would agree that there are all kinds of IoT uses cases where the outer NAI has a realm of the device manufacturer while the inner NAI has a realm of the enterprise where the device is installed (or vice-versa).
>    I think that's best described as "surprising" from an EAP standpoint.  The outer NAI is overwhelmingly used to determine routing, as with Eduroam.  Having different outer / inner NAIs means that the routing is broken:
>
> * packets get sent to "example.com"
> * the example.com server sees authentication requests for "example.net", and says "that's not me, why did you send these packets to me?"
>
>    For the IoT case, we would never route the authentication packets to the device manufacturer.  So it's not clear to me why the outer NAI has to contain that value.
>
>
>    If this is a necessary use-case, I would suggest requiring that these authentications MUST be handled locally, with no routing.
>
>    This issue is why I was proposing to define the outer NAI realm of "@eap.arpa".  Those packets are clearly labelled as "site local", and are not forwarded.
>
>> I also don't understand why this is bad:
>>
>>
>>> For example, if a user
>>>     has an inner identity of
>>> "[e-mail]"
>>> , then it generally makes
>>>     no sense to have an outer identity of "@example.org".
>>>
>> Most university guidelines for eduroam recommend exactly what you are trying to prevent. For example: https://www.aalto.fi/en/services/installation-instructions-for-eduroam says:
>    I haven't seen this recommendation in most universities.  My discussions with the eduroam people make me believe that such configurations are in fact discouraged by the wider Eduroam recommendations.
>
>>> Review your settings in the Wi-Fi Settings window:
>>> Wireless security: WPA & WPA2 Enterprise
>>> Authentication: Protected EAP (PEAP)
>>> Anonymous identity: [e-mail]
>>> CA certificate: ca-certificates.crt
>>> PEAP version: Automatic
>>> Inner authentication: MSCHAPv2
>>> Username: [e-mail]
>>> Password: <your Aalto password>
>    Which is using the same realm for both outer / inner identities.  So I don't understand why there's an issue.

The draft currently says that inner identity should not have a realm. 
However, the example above, and many other guidance documents I have 
seen do use a realm in the inner identity.

--Mohit

>
>    Incidentally the recommendation violates the RFCV 7542 recommendation to just use @aalto.fi as the outer NAI.   i.e. an outer *user* portion of the NAI is not needed.  But both *realms* are identical for inner and outer NAIs.
>
>> @Joe: I am not confident that this section has had sufficient review. I am generally not comfortable with the text. This draft was anyways about TLS 1.3 for TEAP/PEAP/TTLS etc. I think this is going way beyond what the draft originally was trying to solve.
>    This section describes wide-spread practice in the community for about a decade.
>
>    While the updated text is not specific to TLS 1.3 and TTLS / PEAP / FAST / TEAP, it does give guidelines to users of TTLS / etc. for TLS 1.3.  WhichI think is critical for success in real-world use-cases.
>
>> 3. Section 2.4 says:
>>
>>>   the response from the EAP peer MUST be either
>>>     EAP-Success or EAP-Failure.
>>>
>> I though the Success and Failure messages are sent by the EAP server?
>    That was a typo, already addressed in a previous comment.
>
>> 4. Section 4 says:
>>
>>
>>> If either peer or server instead
>>>     initiates an inner tunnel method
>>>
>> I thought you have mandated the use of an inner tunnel method? So why the 'if'?
>    Resumption doesn't use an inner tunnel method.  So the "if" is necessary.
>
>    Alan DeKok.
>