Bug 83542

When a device configured with DHCPServer=yes loses carrier, networkd stops its DHCP server. But when the carrier re-appears later on, the server is not restarted.

My use case is a bridge I use to connect virtual machines to; the bridge is NAT-ed to the outside and the host interface is *not* connected to it; veth and tap devices are added and removed as needed.

First virtual machine to start gets its address successfully. After the last machine is shut down, the bridge has no ports and networkd stops the DHCP server. When a new machine is started after that, the bridge gets a carrier again, but networkd doesn't restart DHCP server, so that and subsequent machines can't acquire a lease.


$ cat /etc/systemd/network/br0.netdev
[NetDev]
Name=br0
Kind=bridge

$ cat /etc/systemd/network/br0.network
[Match]
Name=br0
[Network]
Address=10.10.10.1/24
DHCPServer=yes


This behaviour follows from the state machine in networkd-link.c. It looks like it could be fixed by moving set-up calls to sd_dhcp_*() from link_enter_configure() to link_acquire_conf() to regain symmetry between setup that happens on carrier_gained and teardown on carrier_lost, but I'm not confident enough in my understanding of the state machine invariants to patch it myself.