[Olsr-users] olsrd 0.6.6.1 (and earlier) ipv6 problems

[Russell Senior]

We have a network of around 60 nodes, running olsrd 0.6.6.1 (and
earlier) in a hub and spoke topology, primarily over openvpn tunnels.
The nodes are running recent trunk openwrt, the hub is a gentoo box.
The nodes are a mixture of ar71xx, x86 and brcm47xx devices.  We
provision ipv6 to the nodes through hurricane electric 6in4 tunnel to
the well-connected hub machine, and then statically provising /64's to
each of the nodes and let olsrd manage the routing tables.  We also
run an ipv4 olsrd in parallel.

Recently, we encountered our first node that had its own native ipv6
upstream, which is preferable (as it takes the traffic load off our hub
box).  To avoid routing confusion, we ultimately turned olsrd ipv6 off
on that node.  Around the same time, we noticed that some of our nodes
seemed to lose their ipv6 routing.  When I checked the routing tables
on our hub machine (e.g. "ip -6 r"), all I saw were the local routes.

I have a set of devices in a test-bed in my basement where I checkout
newly built images where I have serial console.  My basement network
has its own ipv6 network through a separate hurricane electric tunnel.

I noticed that if I unplugged one of the devices in my test-bed, the
hub machine gradually (over about 30-60 seconds) got all its ipv6
routes back.  When I plugged in the test-device, boom, in 10-15
seconds, the hub machine's ipv6 routing table collapsed again.  Unplug
it, the routes come back.  

Meanwhile, the ipv4 routing table is steady as a rock.

Eventually, I discover that an openwrt uci option ip6assign seems to
be causing trouble, and I disable it.  Meanwhile the same image at
another site does not appear to be causing any trouble, perhaps
because it does not have an upstream ipv6 network?

Then a few days ago, I look and notice that all (or virtually all) the
ipv6 routes are gone from our hub machine again.  This time, I go to
another of my test-bed devices and turn off the ipv6 olsrd there, and
after 30-60 seconds the routes all come back again.  This time it is
an x86 device (an alix2).

Does this sound at all familiar?  Any suggestion where to start looking?


-- 
Russell Senior, President
(spam-protected)