(In reply to Thomas Haller from comment #0)
> Devices that are "ignore-carrier" still don't allow to activate a DHCP
> connection without having carrier -- which is correct.
> 
> But when loosing carrier, the connection stays up -- which is probably
> correct too, even for DHCP.

Agreed.

> 
> When carrier returns, NM does not try to renew the addresses -- wrong.
> 

Fixed, published in  branch jk/renew-dhcp-bgo743368.

> 
> Also TODO: check correct behavior if the lease expires while being
> unplugged. Correct behavior might be to tear down the connection.

Tested. The connection is correctly deactivated when the lease expired.

(In reply to Jiri Klimes from comment #1)
> > Also TODO: check correct behavior if the lease expires while being
> > unplugged. Correct behavior might be to tear down the connection.
> 
> Tested. The connection is correctly deactivated when the lease expired.

yeah sounds right for now.

But thinking about having the root-fs mounted from the network, in the future we might need a way to flag connections to keep them up at all cost (and ignore DHCP expires). 




>> device: renew DHCP lease for active "ignore-carrier" devices on carrier-on (bgo #743368)
    
+    if (priv->rdisc) {
+         /* FIXME: todo */
+    }
+    if (priv->dnsmasq_manager) {
+         /* FIXME: todo */
+    }

these comments don't belong there after your change. Maybe leave update_for_ip_ifname_change() where it was, calling into update_dynamic_ip_setup()?


Other then that, LGTM

I think the comments are still relevant.

master: 348452f device: renew DHCP lease for active "ignore-carrier" devices on carrier-on (bgo #743368)
nm-1-0: 94931d4 device: renew DHCP lease for active "ignore-carrier" devices on carrier-on (bgo #743368)