My server machine (which hosts an array of VMs that do the actual server-ing) did its nightly update run last night. It got firewalld-0.3.9-1.fc20 . Unfortunately, the update seemed to have a bad interaction with libvirt, and there's a whole bunch of errors in the logs right at the time the update is installed and firewalld restarted. When I woke up, none of my server VMs were online. After rebooting the host (probably not necessary, but eh, there was a new kernel anyway) they're back up. I'll attach the logs, in case there's enough here to figure out what went wrong.
Created attachment 850601 [details] journalctl output from around the relevant time
danpb suggests this may be the same as https://bugzilla.redhat.com/show_bug.cgi?id=1031102 .
0.3.9 contains the fix for the problem: - Use rmmod instead of 'modprobe -r' (RHBZ#1031102) of course when you upgrade to the 0.3.9 version, as part of restarting the service, the 0.3.8 version will be stopped - which triggers the flaw in that version. Only future upgrades thereafter will be safe.
OK, if thomas confirms this looks like the same problem, I guess he can close this. and i'll try and keep an eye out for the next firewalld update and see if it causes the same thing.
Yes, this very much looks like the "modprobe -r" problem we had. Closing as suggested in comment 4 with CURRENTRELEASE.