bond interfaces stop working after restart of systemd-networkd

[impact]

restarting systemd-networkd drops carrier on all bond slaves, temporarily interrupting networking over the bond.

[test case]

on a bionic system with 2 interfaces that can be put into a bond, create config files such as:

root@lp1833671:~# cat /etc/systemd/network/10-bond0.netdev
[NetDev]
Name=bond0
Kind=bond

root@lp1833671:~# cat /etc/systemd/network/20-ens8.network
[Match]
Name=ens8

[Network]
Bond=bond0

root@lp1833671:~# cat /etc/systemd/network/20-ens9.network
[Match]
Name=ens9

[Network]
Bond=bond0

root@lp1833671:~# cat /etc/systemd/network/30-bond0.network
[Match]
Name=bond0

[Network]
Address=1.2.3.4/32

restart networkd, or reboot, and verify the bond is up:

root@lp1833671:~# ip a
3: ens8: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc fq_codel master bond0 state UP group default qlen 1000
    link/ether 42:30:62:cc:36:2b brd ff:ff:ff:ff:ff:ff
4: ens9: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc fq_codel master bond0 state UP group default qlen 1000
    link/ether 42:30:62:cc:36:2b brd ff:ff:ff:ff:ff:ff
5: bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
    link/ether 42:30:62:cc:36:2b brd ff:ff:ff:ff:ff:ff
    inet 1.2.3.4/32 scope global bond0
       valid_lft forever preferred_lft forever
    inet6 fe80::4030:62ff:fecc:362b/64 scope link
       valid_lft forever preferred_lft forever

restart networkd and check /var/log/syslog:

root@lp1833671:~# systemctl restart systemd-networkd
root@lp1833671:~# cat /var/log/syslog
...
Jul 23 21:08:07 lp1833671 systemd-networkd[1805]: ens9: Lost carrier
Jul 23 21:08:07 lp1833671 systemd-networkd[1805]: ens8: Lost carrier
Jul 23 21:08:07 lp1833671 systemd-networkd[1805]: ens9: Gained carrier
Jul 23 21:08:07 lp1833671 systemd-networkd[1805]: ens8: Gained carrier

[regression potential]

this changes how bond slaves are managed, so regressions could affect any configurations using bonding.

[other info]

the patch is already included in d, and ifupdown manages networking in x, so this is needed only for b.

[original description]

Running systemd-networkd from systemd 237-3ubuntu10.23 on Ubuntu 18.04.2 I have one machine where, every time systemd-networkd restarts (ie every time there is an update to systemd) the bond0 interface stops working.

I see both physical interfaces go soft down and then come back again:

Jun 21 07:28:24 odin.openstreetmap.org systemd[1]: systemd 237 running in system mode. (+PAM +AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP +GCRYPT +GNUTLS +ACL +XZ +LZ4 +SEC
Jun 21 07:28:24 odin.openstreetmap.org systemd[1]: Detected architecture x86-64.
Jun 21 07:28:24 odin.openstreetmap.org kernel: bond0: link status down for backup interface eno2, disabling it in 200 ms
Jun 21 07:28:24 odin.openstreetmap.org kernel: bond0: link status down for active interface eno1, disabling it in 200 ms
Jun 21 07:28:24 odin.openstreetmap.org kernel: 8021q: adding VLAN 0 to HW filter on device eno2
Jun 21 07:28:25 odin.openstreetmap.org kernel: 8021q: adding VLAN 0 to HW filter on device eno1
Jun 21 07:28:25 odin.openstreetmap.org kernel: bond0: link status up again after 200 ms for interface eno2
Jun 21 07:28:25 odin.openstreetmap.org kernel: bond0: link status up again after 100 ms for interface eno1

and after that nothing until I stop systemd-networkd, delete the bond interface, and then start systemd-networkd again.

On most machines the cycle seems to take a bit longer and the interfaces reach a hard down start before coming back and in that case there seems to be no problem.

I think this is likely an instance of this upstream bug:

https://github.com/systemd/systemd/issues/10118

which has a fix here:

https://github.com/systemd/systemd/pull/10465