[ofa-general] [PATCH/RFC] IPoIB: Don't drop multicast packets sent before group is joined

[Christoph Lameter]

On Thu, 11 Jun 2009, Roland Dreier wrote:

>  > Could you reapply the patch? The semantics are exaclty as demanded by
>  > David Miller. It blocks the process that is doing the join and not the
>  > others.
>
> No, the semantics are not the same.  It doesn't block until socket space
> is exhausted, and it will no longer work if the net stack does
> skb_orphan before giving the packet to the network driver.

So it does block (the socket space for one process is exhausted!) and it
will work until the skb_orphan goes in. At that point different methods
will be available for the regular NICs as well.