Originally Posted by beaker_
Why bridge when it will never be trustworthy?
Why wouldn't bridging be stable and trustworthy ?! I've used bridged (as in brctl) connections for commercial systems since the early 2.4 kernel days. I've used it in custom network hardware configs. My mail server is a VM connected by bridge. All of these work 100%.
Here is why brctl doesn't work with wifi. It's a 80211 stack usage issue.
Is there a reason why you don't route your virbr interface instead? Personally, I'd buy a nic and pull a wire.
One reason to not want route/NAT virbr connection is that the performance is a lot slower (but still decent). But it certainly avoids the 'wifi' can't bridge' problem. So *IF* you have a wired connection - yes you can improve net performance by just adding the wired connection to the bridge and using this bridge directly as your net interface (no routing, no NAT, just a software hub).
Originally Posted by geekaume
I can't understand what you said. What do you mean by "why you don't route your virbr interface instead" ? Is there a way for me to have VMs accessing the network (and Internet) with a Wifi interface ?
I can't have a wire : my router is too far from the PC....
The default Fedora KVM configuration creates a virbr0 bridge just for the VMs and then it creates a tap/tun device for each VM that connects the VM to this bridge. Then it creates a virbr0 tap/tun that is NAT/routed to the (primary?) Host network (like your wifi).
You can change it but the default bridge is virbr0 with LAN range 192.168.122.*/24
The VMs can NAT to your LAN and even out to the Inet.
Your Host can access the VMs and the virbr0 LAN according to the iptables entries created (disables ping/icmp btw).
To access the VMs from across your LAN you need to add route entries to the other systems. You neet to route the address range for the VMs (192.168.122.*/24 by default) to the VM host.
Look at the virt-manager' virtual network setup screens.