Re: [PATCH] netlink: Drop point-to-point peer information when we copy addresses

[Stefano Brivio <sbrivio@redhat.com>]

On Tue, 23 Apr 2024 11:02:43 +1000
David Gibson <david@gibson.dropbear.id.au> wrote:

> On Fri, Apr 12, 2024 at 12:18:00AM +0200, Stefano Brivio wrote:
> > If the template host interface is of type tun, and it's configured
> > with a point-to-point peer address (that's what happens for example
> > with openvpn and '--topology net30'), pasta will copy the peer
> > information onto the namespace interface.
> > 
> > But the namespace interface is not actually a point-to-point tunnel,
> > and we won't resolve the peer address via ARP either, so we have
> > to drop this information to get the expected behaviour (traffic
> > regularly sent over our tap interface).
> > 
> > Link: https://github.com/containers/podman/issues/22320
> > Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
> > ---
> >  netlink.c | 23 ++++++++++++++++++++++-
> >  1 file changed, 22 insertions(+), 1 deletion(-)
> > 
> > diff --git a/netlink.c b/netlink.c
> > index 89c0641..73aaa4b 100644
> > --- a/netlink.c
> > +++ b/netlink.c
> > @@ -792,8 +792,8 @@ int nl_addr_dup(int s_src, unsigned int ifi_src,
> >  
> >  	seq = nl_send(s_src, &req, RTM_GETADDR, NLM_F_DUMP, sizeof(req));
> >  	nl_foreach_oftype(nh, status, s_src, buf, seq, RTM_NEWADDR) {
> > +		struct rtattr *rta, *rta_local = NULL;
> >  		struct ifaddrmsg *ifa;
> > -		struct rtattr *rta;
> >  		size_t na;
> >  
> >  		ifa = (struct ifaddrmsg *)NLMSG_DATA(nh);
> > @@ -804,12 +804,33 @@ int nl_addr_dup(int s_src, unsigned int ifi_src,
> >  
> >  		ifa->ifa_index = ifi_dst;
> >  
> > +		for (rta = IFA_RTA(ifa), na = IFA_PAYLOAD(nh); RTA_OK(rta, na);
> > +		     rta = RTA_NEXT(rta, na)) {
> > +			if (rta->rta_type == IFA_LOCAL) {
> > +				rta_local = rta;
> > +				break;
> > +			}
> > +		}
> > +
> >  		for (rta = IFA_RTA(ifa), na = IFA_PAYLOAD(nh); RTA_OK(rta, na);
> >  		     rta = RTA_NEXT(rta, na)) {
> >  			/* Strip label and expiry (cacheinfo) information */
> >  			if (rta->rta_type == IFA_LABEL ||
> >  			    rta->rta_type == IFA_CACHEINFO)
> >  				rta->rta_type = IFA_UNSPEC;
> > +
> > +			/* Different values for IFA_ADDRESS and IFA_LOCAL mean
> > +			 * that IFA_LOCAL is the locally configured address, and
> > +			 * IFA_ADDRESS is the peer address for a point-to-point
> > +			 * interface. But our namespace interface isn't really a
> > +			 * point-to-point tunnel, and we can't resolve that peer
> > +			 * address via ARP: simply drop it, and keep the local
> > +			 * address.  
> 
> Could we just unconditionally remove IFA_ADDRESS properties (by
> setting them to IFA_UNSPEC)?  That we we could avoid having two passes
> through the attributes.

Ah, thanks, that sounds better, but I haven't tried it yet. By the way,
this patch doesn't fix the issue:
  https://github.com/containers/podman/issues/22320#issuecomment-2051279807

so I think we need something on top of this, but I'm not sure yet what.
Other than tweaking routes, another idea might be to adjust the netmask
here.

-- 
Stefano