IPv6 / ndp taking a long time to resolve in this weeks STABLE

[Pete French]

> I haven't modified the ndp related code for quite a long time, but recently have seen
> some postings regarding incomplete ndp entries (still catching up on emails).
>
> Changes committed in the past year or so were related to locking and memory leak, not
> functional updates.
>
> I have seen this issue with ARP, where incoming ARP requests are not completing
> the entries until an explicit outgoing packet is generated right after the
> system boots.
>
> Email me when you recreate the problem.

Hi, thanks for the email - I canb recreate this very easily here
now. I have also shown that it the problem doesnt happen
on a kernel from Feb 25th, but does happen on a kernel from
this week.

cery easy to demonstarte - set up 2 machines. Ping one
from the other so that it is pinging nicely. Then delete the ndp
entry. The ping freezes - sometimes for 20 seconds! It
eventually starts again with all the inrtervening packets lost.
This only happens if the machone being oinged is running the
newer kernel. If you do it the other way round then all
is fine.

I havent tried to reproduce what I saw with CARP - but that
feels like the same problem to me. i.e. ndp cannot find the
CARP interface's ether address, and tahst why connectivity does
not work.

-pete.