(usagi-users 02467) Re: IPv6 problem in 2.5 kernel

["Jeroen Massar" <jeroen@xxxxxxxxx>]

Hasso Tepper [mailto:hasso@xxxxxxxxx] wrote:

> Jeroen Massar wrote:
> > Hasso Tepper [mailto:hasso@xxxxxxxxx] wrote:
> > > http://bugme.osdl.org/show_bug.cgi?id=746
> > >
> > > I can confirm that this bug still exist in stock 2.5.74 and in
> > > Usagi cvs.
> >
> > Is multicast correctly supported on your network card?
> 
> I think so. I'm using Intel eepro100 card with e100 module. Not 
> completely sure of course, there might be bugs as well. But I doupt 
> that it is related to multicast.

You should not be having any troubles with intel eepro100 cards.
I've been using those for a couple of years along with IPv6
and had no problems whatsoever. I think they are one of the
better NIC's around next to DEC Tulips...

The reason for checking multicast is that some buggy drivers
don't properly pass along multicast packets and thus one won't
send/receive RA's... and as they are buggy one moment they
could work, and after your expire time they will stop being
on your interface...

> > ip link set eth0 multicast on
> > Or throw it into promisc might solve the problem.
> 
> If IPv6 stops working throwing it into promisc mode doesn't make it 
> working again.

Did you throw the interface up/down to let it generate a route request?
RA's are only sent out periodically... and there is no rtsol command
generically available on linux boxes unfortunatly. Which would be
better than throwing up/down interfaces (or is there another trick)

> "echo 8000 > /proc/sys/net/ipv6/route/max_size" does.

Are you hitting the max route table? 4096 entries, fun :)
Maybe a dump of your routing tables could help.
But I would suspect that an overflow would at least log something
in the dmesg or something similar.

Biggest ip -6 ro sho | wc -l  I found was ~1500 entries...
And that's a lot already imho :)

Greets,
 Jeroen