id	summary	reporter	owner	description	type	status	priority	milestone	component	version	severity	resolution	keywords	cc
4506	Fix systemd-networkd to work with 5.2+ kernels	Douglas R. Reno	Douglas R. Reno	"In Linux Kernel 5.2+, The IPv6 implementation was changed in the kernel, in particular the way that it responds to calls towards it's API for bringing interfaces up.

As a result, systemd-networkd will refuse to bring up any network interfaces that are on the system, either at startup or manually. systemd-networkd will return the following output:

{{{
enp3s0: Could not bring up interface: Invalid argument
}}}

You must instead bring up the interface manually through the kernel, and then systemd will recognize the flag change and begin bringing up DHCP services or assigning a static if necessary:

{{{
sudo ip link set enp3s0 up
}}}

However, as Paul Menzel from Debian noted, this isn't always guaranteed to work either.

As far as I can tell (I'm running dhcpcd on my system with networkd brought down), dhcpcd isn't affected. I have NOT tested dhclient though, I'll probably do that when I'm done with this seeing as it could be related.

Upstream bug report: [https://github.com/systemd/systemd/issues/12784]

We need to apply the following patches for systemd to allow this to work:

[https://github.com/systemd/systemd/commit/4eb086a38712ea98faf41e075b84555b11b54362]
[https://github.com/systemd/systemd/commit/9f6e82e6eb3b6e73d66d00d1d6eee60691fb702f]

On that same note, we should apply the following fix as well, for users of AMD Ryzen systems:

[https://github.com/systemd/systemd/commit/1c53d4a070edbec8ad2d384ba0014d0eb6bae077]
[https://github.com/systemd/systemd/commit/7f2cdceaed4d37c4e601e531c7d863fca1bd1460]
[https://github.com/systemd/systemd/commit/328f850e36e86d14ab06d11fa8f2397e9575a7f9]

(This fixes issues with the Random Number Generator on Ryzen systems that will cause the system to fail to start due to firmware changes and the way that they interact with 5.2+ kernels. AMD is still preparing a firmware update for their CPUs to fix this. This seems to be due to the RDRAND instruction in AMD Excavator-era CPUs in particular).

Upstream bug report for that: [https://github.com/systemd/systemd/issues/11810]

Thank you, Wayne, for initially discovering the networkd issue:
[http://lists.linuxfromscratch.org/pipermail/lfs-support/2019-July/052952.html]"	defect	closed	normal	9.0	Book	SVN	normal	fixed