[quagga-dev 8512] Re: [PATCH 3/3] ospfd: use setsockopt_so_bindtodevice to specify out interface.
joakim.tjernlund at transmode.se
Wed Feb 9 12:22:23 GMT 2011
Joakim Tjernlund/Transmode wrote on 2011/02/09 13:12:36:
> paul at jakma.org wrote on 2011/02/09 12:01:06:
> > On Wed, 9 Feb 2011, Joakim Tjernlund wrote:
> > > AFAIK virtual links are not connected to a real interface but
> > > lets the OS select interface. This is not what we want when
> > > we have multiple links with same src and dst address but for
> > > now there is no choice.
> > Not commenting specifically on the details of the patch, but
> > virtual-links should follow routing to the destination. The nexthop
> > (and hence the local address, to whatever degree the interface
> > constrains selection, which is meant to reflect the interface) is set
> > via the SPF process.
> > I.e. the VL *is* associated with a local interface, fwiw. So
> > potentially this patch isn't needed (at least, not at this level - if
> > something's going wrong, then perhaps it's in or around vl_up_check
> > not working in the presence of unnumbered interfaces).
> hmm, but oi->ifp->name isn't updated to match the local I/F I think.
> It is still VLINK%d, isn't it?
> So to get at the real IF something else is needed. We don't use VLINKS
> so I haven't looked into this area.
Possibly one could use oi->vl_data->nexthop.oi->ifp->name ?
More information about the Quagga-dev