[openib-general] IPoIB Multicast

[Hal Rosenstock]

On Sat, 2004-11-06 at 14:27, Roland Dreier wrote:
>     Hal> 1. If you down the interface and bring it back up, the second
>     Hal> time up, there are 2 identical join requests for the
>     Hal> broadcast group rather than just 1. These 2 come out very
>     Hal> close to one another (217 usec apart). Is there some counting
>     Hal> issue that is causing this ?
> 
>     Hal> 2. When leaving an IP multicast group, there appears to be an
>     Hal> extra join to 0x16 (something like 224.0.0.22 which would be
>     Hal> for IGMP). Any ideas on this ?
> 
> If you or someone else doesn't debug these issues first, I'll take a
> look at the code.

I'll take a first crack and look at the code to see what I can
determine. 

On the second issue, I partially understand what is going on: IPmc group
changes need to be reported via IGMP so the IPmc router knows to prune
the multicast tree, but... I don't understand why it joins here (and not
earlier when a IPmc group is first joined by this node and second after
the join is successful, I do not see any IGMP packet come out of the
node (onto IB; maybe it is going out the ethernet instead).

-- Hal