Hello,
On Tue, 27 Oct 2015, Andy Gospodarek wrote:
> > Of course, we have a semantic problem when setting
> > RTNH_F_LINKDOWN on last address removal, i.e. this event
> > has nothing to do with the link state. But it works because
> > RTNH_F_LINKDOWN is valid for lookups only when DEAD flag
> > is not set, so that is why my patch looks this way.
> The problem you describe here was a concern of mine as well. I would
> really like the output of 'ip route show' to properly reflect the link
> state and fix the problem you describe, but it seems like it will not in
> this case with your current patch. I'll do a bit more testing and let
> you know.
Another option is to remove the RTNH_F_LINKDOWN
usage from any event, from FIB structs. In case only
netif_carrier_ok check is needed, it looks cheap to just
call it after the IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN check.
When showing it to user space we can again use netif_carrier_ok.
The new fib_rebalance() is special, it still
needs to be called, just like now on NETDEV_CHANGE
event but only when ignore_routes_with_linkdown=1, i.e.
we can avoid calling fib_sync_up/fib_sync_down_dev in
fib_netdev_event if the flag is not set, guarded by
CONFIG_IP_ROUTE_MULTIPATH check.
To make this work we can also call somehow
fib_rebalance() from devinet_conf_proc() context when
the flag is changed, all this is for the
CONFIG_IP_ROUTE_MULTIPATH case.
Regards
--
Julian Anastasov <j...@ssi.bg>
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
