DescriptionAdam Williamson
2012-03-29 02:24:44 UTC
I'm no networking expert, so there may be errors in the description here, but I hope the gist comes through.
If you just use Fedora's default virt stack to set up a VM, in the dumb pointy-clicky virt-manager way, you'll get a 'virbr0' interface on the host machine with an IP of 192.168.122.1. VMs will have 'eth0' interfaces with IP address in the 192.168.122.[2-255] range (actually probably a narrower higher range, but you get the idea). Effectively libvirt is handling NAT for the VMs behind it. The VMs get their DNS server set to 192.168.122.1, and everything works, and I don't have to think too hard about why.
If I boot a VM using a live image which has Boxes included (and probably if you just 'yum install boxes' into a VM and reboot), the *VM itself* has a virbr0 interface with an IP of 192.168.122.1. The VM can ping out to numeric IP addresses, it seems - ping 8.8.8.8 works - but it can't resolve any hostnames, presumably because the guest virbr0 is conflicting with the host virbr0 and messing things up.
I have no idea how this should be fixed, I just know it should, because the user-visible impact is that 'networking is horked in KVMs with Boxes installed'.
This may be a libvirt issue rather than a Boxes one, I guess. I'm not sure. CCing the Dans Berrange and Veillard.
If we wanted to have Boxes in the F17 release, this would be a release blocker.