misconfigured networkd may break after networkd restart

[Dan Streetman]

[impact]

bug 1929560 fixed an error in setting/clearing an interface 'master' during systemd-networkd startup if the interface was already up, however for any users who have been relying on that bug to manually set an interface 'master' for an interface that's otherwise controlled by systemd-networkd, this will break their system by removing the interface from the 'master' on a systemd-networkd restart.

For example, if a user has interface 'eth0' configured to be controlled by systemd-networkd, but then manually does 'ip l set dev eth0 master br0' to add it to the 'br0' bridge, previous behavior would leave 'eth0' in the bridge after a systemd-networkd restart, even though it should not be in the bridge based on the systemd-networkd configuration. The new behavior matches systemd-networkd behavior for all other aspects (i.e. networkd removes all addresses, routes, and other settings it doesn't know about or that are different from its configuration), however it breaks users expecting the previous incorrect behavior.

[test case]

configure systemd-networkd to control an interface that isn't set with any 'master', and start networkd. For example, an instance configured for DHCP. After networkd has started, manually set the interface 'master' to an existing bridge. Restart networkd, and check if the interface still has the bridge 'master' or not.

e.g., in a newly created cloud-image container, with normal netplan, check the networkd control of eth0:

root@lp1937117-f:~# networkctl list eth0
IDX LINK TYPE OPERATIONAL SETUP
159 eth0 ether routable configured

create a new bridge and manually place eth0 into it:

root@lp1937117-f:~# ip l add dev br0 type bridge
root@lp1937117-f:~# ip l set up dev br0
root@lp1937117-f:~# ip l set dev eth0 master br0
root@lp1937117-f:~# ip l show dev eth0
159: eth0@if160: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 00:16:3e:56:f8:07 brd ff:ff:ff:ff:ff:ff link-netnsid 0

now restart systemd-networkd, and check if eth0 remains under br0:

root@lp1937117-f:~# systemctl restart systemd-networkd
root@lp1937117-f:~# ip l show dev eth0
159: eth0@if160: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT group default qlen 1000
    link/ether 00:16:3e:56:f8:07 brd ff:ff:ff:ff:ff:ff link-netnsid 0

[regression potential]

the main potential is for problems for those with misconfigured systems, relying on specific previous incorrect behavior, specifically interfaces controlled by systemd-networkd in part, but also with manual configuration performed outside of systemd-networkd configuration.

[scope]

this is needed for bionic and focal to restore the previous incorrect behavior; the correct behavior should remain in impish, and (arguably) should remain in hirsute as well

[other info]

see bug 1929560 for details of the original bug

[Dan Streetman (ddstreet)]

Note that the *proper* way to place eth0 into a bridge while also being under systemd-networkd management is to create a .netdev for the bridge and assign eth0 to it, e.g.:

$ cat 10-br0.netdev
[NetDev]
Name=br0
Kind=bridge

cat 10-netplan-eth0.network
[Match]
Name=eth0

[Network]
DHCP=ipv4
LinkLocalAddressing=ipv6
Bridge=br0

[DHCP]
RouteMetric=100
UseMTU=true

Or alternately, just remove the .network file for 'eth0' so it's not under systemd-networkd control, and manually and/or externally control it (i.e. manually bring up eth0 and manually place it into br0).

[Dan Streetman (ddstreet)]

Attached debdiffs for b/f.

Note that Bionic only needs a revert of the patch for the original bug, but Focal requires the patch revert plus an additional small patch to restore the previous behavior, due to other patches added for bug 1664844.

[Launchpad Janitor (janitor)]

This bug was fixed in the package systemd - 245.4-4ubuntu3.11

---------------
systemd (245.4-4ubuntu3.11) focal-security; urgency=medium

  * d/p/lp1937117/0001-revert-lp1929560-network-move-set-MAC-and-set-nomaster-operations-out.patch,
    d/p/lp1937117/0002-avoid-changing-interface-master-if-interface-already-up.patch:
    - Don't change interface master if interface is already up,
      due to users expecting previous buggy behavior (LP: #1937117)

 -- Dan Streetman <email address hidden> Wed, 21 Jul 2021 15:00:21 -0400

[Launchpad Janitor (janitor)]

This bug was fixed in the package systemd - 237-3ubuntu10.50

---------------
systemd (237-3ubuntu10.50) bionic-security; urgency=medium

  * d/p/lp1937117-revert-lp1929560-network-move-set-MAC-and-set-nomaster-operations-out.patch:
    Revert patch due to users expecting previous buggy behavior
    (LP: #1937117)

 -- Dan Streetman <email address hidden> Wed, 21 Jul 2021 14:51:38 -0400

[Jan Graichen (jgraichen)]

> Note that the *proper* way to place eth0 into a bridge while also being under systemd-networkd management is to create a .netdev for the bridge and assign eth0 to it.

What is the recommended way for Ubuntu's OpenStack offering that dynamically creates bridges (neutron-linuxbridge-agent)? Not using systemd-network at all, never?