Re: draft-gundavelli-v6ops-l2-unicast WGLC

[Brian Haberman <[e-mail]>]

Hemant,

On 8/17/10 11:37 AM, Hemant Singh (shemant) wrote:
> -----Original Message-----
> From: Brian Haberman [mailto:[e-mail]] 
> Sent: Tuesday, August 17, 2010 8:59 AM
> To: Ole Troan
> Cc: Hemant Singh (shemant); Sri Gundavelli (sgundave); Suresh Krishnan;
> IPv6 WG Mailing List
> Subject: Re: Consensus call on
> adopting:draft-krishnan-6man-rs-mark-06.txt
> 
> 
>> Correct.  The MLD snooping functionality only looks at L3 information
> if
>> the L2 destination address is a multicast address.  In this case, L2
> has
>> a unicast address and the MLD snooping function will never see the
>> packet (it will be forwarded using standard L2 logic).

I do not grok this scenario.  Let me see if I can understand what you
are concerned about.

> 
> Brian, I changed the subject back to the Gundavelli document.  I am
> saying a host sent an MLDv2 Report to a router.  There is no network
> switch between this host and the router.  Now with the rule in the
> Gundavelli document, the host sent the MLDv2 Report with the L3
> multicast destination but L2 unicast destination.  The L2 sniffer on the
> router fails to capture this packet and fails to forward the packet to
> its ULP (Upper Layer Protocol) for multicast.  So now the packet is
> shipped to the unicast ULP. Why can't the unicast ULP barf that it
> received a packet with a L3 destination when it's a unicast ULP?  Why
> shouldn't we test such a case with a router and a host sending such a
> doctored MLDv2 Report?  One should use more than one router to test such
> a case.  Or am I missing something - if yes, my humble apologies.

A host generates an MLDv2 Report message that expresses interest in
receiving multicast traffic on group X.  The destination address in the
IPv6 header will be the All-MLDv2-Routers address (FF02::16).  However,
the Gundavelli draft allows the node to set the destination L2 address
to a unicast address of the only router on the link.  The host transmits
this message.

Now, what I see you arguing above is that the router will receive this
message, but may try and pass it to a *different* ULP.  I think we agree
that the router will receive the packet since it is addressed to one of
its L2 addresses.  Now, when it is parsing this message I don't see how
the L2 destination address will have any impact on which protocol this
message is handed to.  All the routing platforms I am familiar with
follow a coarse (and high-level) logic like:

1. Verify L2 validity (e.g., checksum)
2. Pass packet to protocol handler based on e.g., ethertype
3. For IPv6, parse Next Header field
4. Process MLDv2 Report

So, I can't see a scenario where the L2 destination address interferes
with the proper handling of the L3 information.

Regards,
Brian