Re: [pve-devel] [PATCH container 1/2] net: Add `link_down` config to allow setting interfaces as disconnected

[Christoph Heiss <c.heiss@proxmox.com>]

Thanks for the review!

On Mon, Feb 13, 2023 at 03:31:32PM +0100, Wolfgang Bumiller wrote:
> On Mon, Feb 13, 2023 at 02:56:59PM +0100, Christoph Heiss wrote:
> > [..]
> >
> > diff --git a/src/PVE/LXC.pm b/src/PVE/LXC.pm
> > index ce6d5a5..039a476 100644
> > --- a/src/PVE/LXC.pm
> > +++ b/src/PVE/LXC.pm
> > @@ -668,7 +668,7 @@ sub update_lxc_config {
> >
> >      # some init scripts expect a linux terminal (turnkey).
> >      $raw .= "lxc.environment = TERM=linux\n";
> > -
> > +
> >      my $utsname = $conf->{hostname} || "CT$vmid";
> >      $raw .= "lxc.uts.name = $utsname\n";
> >
> > @@ -932,8 +932,9 @@ sub update_net {
> >  	my $oldnet = PVE::LXC::Config->parse_lxc_network($oldnetcfg);
> >
> >  	if (safe_string_ne($oldnet->{hwaddr}, $newnet->{hwaddr}) ||
> > -	    safe_string_ne($oldnet->{name}, $newnet->{name})) {
> > -
> > +	    safe_string_ne($oldnet->{name}, $newnet->{name}) ||
> > +	    defined($oldnet->{link_down}) != defined($newnet->{link_down})
>
> Doing this here would cause the interface to be deleted and recreated in
> a "down" state, which is much more disruptive than it needs to be.
> Instead, this should be treated more like we do changing the 'bridge'
> property (the 'else' case just below this), and stop after the
> `tap_unplug` for the 'down' case.
Ack. I see, that's a good point.

>
> > +	) {
> >  	    PVE::Network::veth_delete($veth);
> >  	    delete $conf->{$opt};
> >  	    PVE::LXC::Config->write_config($vmid, $conf);
> > @@ -1010,6 +1011,11 @@ sub hotplug_net {
> >      $cmd = ['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', '/sbin/ip', 'link', 'set', $eth ,'up'  ];
> >      PVE::Tools::run_command($cmd);
> >
> > +    # In case the network device should be disconnected, force the host-link down
> > +    if (defined($newnet->{link_down})) {
> > +	PVE::Tools::run_command(['/sbin/ip', 'link', 'set', 'dev', $veth, 'down']);
>
> These interfaces are usually part of a bridge, and therefore the next
> `ifreload` (and probably some other things) would re-enable them
> automatically.
I did not know this, thanks for explaining!

>
> We do need to actually "unplug" them from the bridge (tap_unplug) to
> avoid this.
Will rework that for v2 and do it properly.

>
> > +    }
> > +
> >      my $done = { type => 'veth' };
> >      foreach (qw(bridge tag firewall hwaddr name)) {
> >  	$done->{$_} = $newnet->{$_} if $newnet->{$_};
> >
> > [..]
> > diff --git a/src/lxcnetaddbr b/src/lxcnetaddbr
> > index 83052e1..d8c6767 100755
> > --- a/src/lxcnetaddbr
> > +++ b/src/lxcnetaddbr
> > @@ -58,7 +58,8 @@ if (-d "/sys/class/net/$iface") {
> >      #avoid insecure dependency;
> >      ($bridgemtu) = $bridgemtu =~ /(\d+)/;
> >
> > -    PVE::Tools::run_command("/sbin/ip link set dev $iface up mtu $bridgemtu");
> > +    my $linkstate = defined($net->{link_down}) ? 'down' : 'up';
>
> We need to skip the rest altogether if the link is supposed to stay down
> reliably. As mentioned above, a 'down' interface that is still plugged
> into a bridge will get activated sooner or later...
Funnily enough, in my first draft I actually had it like that, so that
it wasn't connected to the brigde at all. Now I know what cases to also
test for v2.

>
> > +    PVE::Tools::run_command("/sbin/ip link set dev $iface $linkstate mtu $bridgemtu");
> >      PVE::Tools::run_command("/sbin/ip addr add 0.0.0.0/0 dev $iface");
> >
> >      if ($have_sdn) {
> > --
> > 2.39.1