Re: [PATCH 2/4] ipgre: follow state of lower device

[Stephen Hemminger <shemminger@vyatta.com>]

On Thu, 12 Apr 2012 18:32:07 +0100
Ben Hutchings <bhutchings@solarflare.com> wrote:

> On Thu, 2012-04-12 at 09:31 -0700, Stephen Hemminger wrote:
> 
> > GRE tunnels like other layered devices should propogate
> > carrier and RFC2863 state from lower device to tunnel.
> > 
> > Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
> > 
> > --- a/net/ipv4/ip_gre.c	2012-04-12 08:07:39.508107847 -0700
> > +++ b/net/ipv4/ip_gre.c	2012-04-12 08:10:14.177499183 -0700
> [...]
> > @@ -1732,6 +1734,36 @@ static struct rtnl_link_ops ipgre_tap_op
> >  	.fill_info	= ipgre_fill_info,
> >  };
> >  
> > +/* If lower device changes state, reflect that to the tunnel. */
> > +static int ipgre_notify(struct notifier_block *unused,
> > +			unsigned long event, void *ptr)
> > +{
> > +	struct net_device *dev = ptr;
> > +	struct net *net = dev_net(dev);
> > +	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
> > +	unsigned int prio, h;
> > +	struct ip_tunnel *t;
> > +
> > +	if (event == NETDEV_CHANGE)
> > +		return NOTIFY_DONE;
> 
> Surely we should handle NETDEV_UP, NETDEV_CHANGE, NETDEV_DOWN here?  Not
> everything other than NETDEV_CHANGE.

yes, up and down needed as well.

> > +	for (prio = 0; prio < 4; prio++)
> > +		for (h = 0; h < HASH_SIZE; h++) {
> > +			for (t = rtnl_dereference(ign->tunnels[prio][h]);
> > +			     t; t = rtnl_dereference(t->next)) {
> > +				if (dev->ifindex != t->dev->iflink)
> > +					continue;
> > +				netif_stacked_transfer_operstate(dev, t->dev);
> > +			}
> > +		}
> [...]
> 
> This seems potentially very inefficient.

Yes, but there is no list of tunnels per device.