In LXC containers, chroots, or similar virtualization setups, one typically creates veth devices to communicate with/route through the host. These are create pairwise, one end for the container (usually called "ethN"), the other end for the host (usually called "vethN"):

  # ip link add name eth1 type veth peer name veth0

This creates a veth0/eth1 veth pair. As these have no driver, NM classifies these as "generic" devices:

DEVICE     TYPE              STATE        
eth1       unknown           unmanaged    
veth0      unknown           unmanaged    

In order to use NM in containers with these devices, we need to teach it (in particular, nm-udev-manager.c's dev_get_attrs() and nm-system.c's nm_system_get_iface_type()) to recognize veth devices, and then filter out ones which are called "veth.." (as these are usually for the "host" side).

This also makes it possible to create automatic tests for ethernet (my current ones use veth devices as well, to be independent of any physical devices).