On the physical host, create a bridge
eth2 in it, and another bridge
eth3 in it.
Now any guest NIC you place in
br2 will have access to whatever's on the other end of
eth2. Likewise any guest NIC bridged into
br3 will have access to whatever's on the other end of
Just like you can have the physical system with
eth3, you can have a guest with one interface in
br2 and one interface in
br3. You can also just have a guest with one interface in one bridge only.
You can put a host IP address on the new bridges if you want to. If the host has no need to see what's happening on these other networks, then the host doesn't need to have an IP on these bridges. The Linux bridge is a software implementation of a network switch, so everything works at Layer 2, there's no routing involved which would require the host to have an IP address on the bridge.
You can call the bridges whatever you want (
i_like_bacon, etc) but I'd suggest something that makes it easy to identify either what the bridge is connected to, or what it's for. I try to follow the numbering of the underlying NIC, or an intrinsic name like
I'd suggest not to have the names swapped around like
br3 and vice versa. One day you're almost sure to put an interface in the wrong bridge then spend a while wondering why it's not working.