[systemd-devel] systemd-networkd not discovering all devices at bootup, and thus no network is configured

[Zbigniew Jędrzejewski-Szmek]

On Wed, Feb 11, 2015 at 09:32:00PM +0100, Lennart Poettering wrote:
> On Wed, 11.02.15 20:22, Keller, Jacob E (jacob.e.keller at intel.com) wrote:
> 
> > > > I actually see the same behavior now again, so the removal of
> > > > biosdevname does not solve this problem! :(
> > > > 
> > > > Is there any more information I can provide?
> > > 
> > > Hmm, it appears as if networkd completely misses the netlink messages
> > > describing your em0 link.
> > > 
> > 
> > Yes. I believe that possibly the netlink messages are occurring too
> > early before networkd has started. Thus, when I restart the service
> > after startup it works fine.
> 
> Well, the first thing after subscribing to links coming/going that
> networkd does is query the kernel for the list of devices it has. This
> means that it should always get all links, regardless when it is started.
> 
> > > To debug this it might be worse adding debug log messages to
> > > manager_rtnl_process_link() to see if any rtnl messages announcing the
> > > interface are received by networkd. If no such message arrives there,
> > > then this indicates a kernel issue, otherwise a bug in networkd.
> > > 
> > > Lennart
> > > 
> > 
> > How would I go about doing that? I am guessing that means modifying the
> > source of networkd?
> 
> Correct.
> 
> > I'm also somewhat unfamiliar with the best practice for installing a
> > local copy of systemd from source rather than from the Fedora RPMs..
> 
> Most of us tend to run git versions of systemd, and simply build them
> with "./autogen.sh c && sudo make install". But of course, you should
> know what you do then, and there's no easy path back to the FEdora version...

For the case of trying to install a specific utility from git, full
installation is overkill. It's enough to do
  git clone
  ./autogen.sh c && make
  install ./systemd-networkd /usr/lib/systemd/

Zbyszek