10.3.2 breaks bridges in ifupdown

[perkelix]

10.3.2 breaks bridges in ifupdown:

$ LC_ALL=C journalctl -b -u networking.service
Apr 30 17:37:49 p8b75 systemd[1]: Starting networking.service - Raise network interfaces...
Apr 30 17:37:54 p8b75 ifup[1123]: device br0 already exists; can't create bridge with the same name
Apr 30 17:37:54 p8b75 ifup[1121]: run-parts: /etc/network/if-pre-up.d/bridge exited with return code 1
Apr 30 17:37:54 p8b75 ifup[902]: ifup: failed to bring up br0
Apr 30 17:37:54 p8b75 systemd[1]: networking.service: Main process exited, code=exited, status=1/FAILURE
Apr 30 17:37:54 p8b75 systemd[1]: networking.service: Failed with result 'exit-code'.
Apr 30 17:37:54 p8b75 systemd[1]: Failed to start networking.service - Raise network interfaces.

The above "br0 already exists" results in failure to assign a static IPv4 address to the bridge, which in turn makes sshd and dnsmasq fail to launch on this router. As far as I can tell, dhcpcd 10.3.2 tries to bring br0 up before ifupdown's bridge-utils plug-in has had a chance to create the bridge and add interfaces to it.

$ cat /etc/dhcpcd.conf 
require dhcp_server_identifier, domain_name_servers, routers
option static_routes, classless_static_routes, interface_mtu, ntp_servers, sixrd
duid
slaac private temporary
noipv4ll
release
timeout 0
vendorclassid
ia_pd 1 br0

$ cat /etc/network/interfaces
# This file describes the network interfaces available on your system
# and how to activate them. For more information, see interfaces(5).
source /etc/network/interfaces.d/*
allow-hotplug /en*
iface enp2s0 inet manual
iface enp3s0 inet dhcp
	post-up /usr/libexec/funkyware/IPv4-MASQ_IPv6-Bridge $IFACE
iface enp4s0 inet manual
iface enp5s0 inet manual
iface enp6s0 inet manual
iface enp6s1 inet manual
iface enp6s2 inet manual
auto br0
iface br0 inet static
	address 172.16.1.1
	bridge_ports regex (wl|enp6s|enp5s|enp4s|enp2s).*

[rsmarples]

I fail to see how this is a dhcpcd problem as dhcpcd doesn't create interfaces.

[perkelix]

The problem was introduced with dhcpcd 10.3.2.

[rsmarples]

dhcpcd cannot bring up interfaces that don't exist and I'm struggling to see how the two very small patches in this release (1 line change each!!!) can possibly be the cause here.

[perkelix]

Beats me. All I'm saying is:

• My configs haven't changed in a couple of years.
• This setup worked rock-solid until 10.3.1.
• It no longer does since 10.3.2.

If the only thing that changed before the reboot is the dhcpcd binary, it's fairly safe to say that the problem is there.

[rsmarples]

• does the same error happen after rebooting?
• does the error go away if you downgrade to 10.3.1 and reboot?

[perkelix]

• The error consistently occurs if I reboot with 10.3.2.
• The error goes away when I downgrade to 10.3.1 and reboot.

[rsmarples]

Does it work if you revert 0edb765?

[perkelix]

Well... This is odd. After shutting the host down long enough to let the prefix expire, then powering up, everything looks normal again. I'll be monitoring this for the next few days. If we're lucky, this was an oddball fluke. If not... I'll enable debug and add evidence to this report.

[perkelix]

Actually, there is another breakage, noted on Debian CI: Whatever those 2 changes related to carrier handling do, they result in the exit script failing to restore NTP server settings upon releasing the lease. Our test script hasn't changed in a couple of years and was working rock-solid until 10.3.1, but it no longer does. I tried restarting the job a few times, but it systematically fails.

[rsmarples]

Can you put different versions through your CI please - either patch works without the other and I want to know which one is breaking it for you.

[perkelix]

It's Debian's CI, not mine. I just noticed the breakage this morning and tried restarting autopkgtest, but it always fails at the point where it's supposed to restore the initial NTP config. I really have to think of how I could test 2 variants there.

[rsmarples]

Upload new packages? dhcpcd-test1 and dhcpcd-test2. I think that is less effort than me trying to replicate your CI tests right now. Also I'm out of town until Monday so no immediate fixes from me.