On Nov 29, 2012, at 3:15 AM, Sam Hartman wrote:

>>>>>> "Alper" == Alper Yegin <[email protected]> writes:
> 
>    Alper> Sam, If you search for the keyword "discard" in RFC 3748, you
>    Alper> see lots of instances. That's for "EAP layer discarding."
> 
> BTW, thanks for this citation; it was useful. I had read most of the
> text in question before and was in fact even aware of the behavior, but
> for some reason didn't think of it as discarding.. I know, that makes no
> sense at all: the spec calls it discarding.
> 
> I think Jim's proposed fix for this is good. I just wanted to thank you
> for helping me solve my confusion.
> 

No problem.

But I don't understand how you think the text is good. Do you care to explain 
your views on the issues I raised below:

Begin forwarded message:

> From: Alper Yegin <[email protected]>
> Subject: Re: [abfab] Retransmission Text for EAP applicability
> Date: November 22, 2012 12:39:27 PM GMT+02:00
> To: Jim Schaad <[email protected]>
> Cc: [email protected]
> 
>>>> input, these methods discard the input rather than generating an error
>>>> response. If the erroneous input was generated by an attacker,
>>>> legitimate input can sometimes be received after the erroneous input.
>>>> Applications MUST handle an EAP method discarding a message, although
>>>> the specific way in which discarded messages will be handled depend on
>>>> the characteristics of the application.
>>> 
>>> Now, saying "MUST handle" and then leaving the way it's handled out-of
>>> scope does not make sense.
>> 
>> I think this makes perfect sense.  We have just said that how it is going to
>> happen - in fact what the correct response to happen - is going to be based
>> on the specific application we are discussing.  This means that we can't say
>> what it should do.
>> 
> 
> Why would the application care what happens when an EAP-layer or EAP 
> method-layer message is discarded?
> If an EAP-layer message is discarded, it's for the EAP-layer to deal with 
> that.
> If an EAP method-layer message is discarded, it's for the EAP method layer to 
> deal with that.
> 
> If you are designing a protocol that needs to deal with what happens at 1 or 
> 2 layers above, then I'd say there's an issue with this design.
> This is like saying, if HTTP discards a packet, then IP MUST handle this. 
> 
> 
>> Some applications will assume transmission error and send the fact that the
>> EAP packet was not processed.  (I kind of expect TEAP to do this behavior).
>> 
>> Some applications will treat it as an Access-Fail and kill the whole
>> process.
>> 
>> We cannot dictate what the correct behavior should be.
>> 
> 
> Because, that's not the business of this layer. Having a MUST w/o a clear 
> normative text and this explanation is the indication of the problem. Don't 
> you think?
> 
>>> 
>>>> Options include
>>>> failing the authentication at the application level
>>> 
>>> This is problematic. If the EAP method has discarded the message, now you
>>> need this be conveyed down the stack to the EAP lower-layer. This does not
>>> happen today. And enforcing that requires changing existing EAP methods,
>>> creating additional requirement on future methods.
>> 
>> If I make a call to the EAP handler, it does not say that I got an
>> Access-Accept or an Access-Reject, and it does not return anything for me to
>> send back continuing the conversation.   This sounds like a pretty good clue
>> the packet was dropped on the floor given the lock-step behavior of EAP
>> itself.
>> 
> 
> Nope. Maybe the EAP method is awaiting an input from the user. The "EAP lower 
> layer" cannot know what's going on at the EAP method layer.
> 
>> I don't see any need to change either current EAP behavior or future
>> behavior.  
>> 
>>> 
>>>> and waiting for
>>>> additional EAP input, possibly after an EAP retransmit.
>>>> 
>>> 
>>> Who is retransmitting here? The EAP peer? If so, we need to clarify that.
>> And
>>> also described how the EAP peer decides to retransmit, and what it really
>>> retransmits (the previous EAP response?).
>>> 
>> 
>> That is going to be application specific behavior.  In part it is going to
>> depend on who dropped the message on the floor and who actually has the
>> ability to even try and do the resend.    Without knowing your application
>> data flow, I cannot answer this question.
>> 
> 
> The options provided by the text is ambiguous. 
> 
> 
>>> 
>>> Is this text meant to cover how the EAP conversation can be made reliable
>>> with client-driven retransmissions? I see the implications of it, but not
>> clear
>>> and complete description of it.
>> 
>> No.  That may be one of the things that an application may define, but it is
>> not required in any way.
> 
> 
> I'd say if this text is meant to cover that possibility, then we need to make 
> it very clear. Otherwise, the text is rather vague…
> 
> Alper
> 

_______________________________________________
abfab mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/abfab

Reply via email to