WireGuard kill switch crashes with AttributeError when a bridge-slave (port) Ethernet device is present

[thongor77]

WireGuard kill switch crashes with AttributeError when a bridge-slave (port) Ethernet device is present

Summary

When connecting via WireGuard, the kill switch route injection code (add_vpn_server_route) crashes with AttributeError: 'NoneType' object has no attribute 'get_num_routes' if a physical Ethernet interface is enslaved to a NetworkManager bridge (e.g. br0 with eth0 as a bridge port). The connection attempt aborts every time, regardless of the kill switch setting (reproduced with kill switch Off).

Root cause

KillSwitchConnectionHandler.add_vpn_server_route() (killswitch_connection_handler.py) iterates over NMClient.get_physical_devices():

def get_physical_devices(self) -> List[NM.Device]:
    """Returns all the active ethernet/wifi devices."""
    return [
        device for device in self._nm_client.get_devices() if (
            device.get_device_type() in (NM.DeviceType.ETHERNET, NM.DeviceType.WIFI) and
            device.get_state() is NM.DeviceState.ACTIVATED and
            device.get_active_connection()
        )
    ]

This does not exclude Ethernet devices that are bridge ports/slaves. When an interface is a bridge slave, its own NM.RemoteConnection (the "Bridge slave" profile, e.g. bridge-slave-eth0) has no ipv4 setting group — the IPv4 config lives on the bridge master connection (br0), not on the port. Calling nmcli connection show bridge-slave-eth0 confirms it: there is no ipv4.* property block at all for a slave profile.

add_route_to_device() then calls _remove_ipv4_routes():

connection = active_connection.get_connection()
config = connection.get_setting_ip4_config()   # returns None for a bridge-slave profile

for i in range(config.get_num_routes()):        # AttributeError: 'NoneType' object has no attribute 'get_num_routes'

config is None, so the loop crashes immediately.

Reproduction

1. Set up a NetworkManager bridge with a physical Ethernet device as a port, e.g.:

   nmcli connection show
   NAME               TYPE      DEVICE
   bridge-br0         bridge    br0
   bridge-slave-eth0  ethernet  eth0
   

   (br0 carries the IP/default route; eth0 is enslaved with no IP config of its own.)
2. Optionally also have a second active interface (Wi-Fi in my case) — not required to reproduce, but present in my setup.
3. Connect via WireGuard protocol (kill switch setting: Off).
4. Connection fails immediately every time.

Traceback (proton-vpn-gtk-app 4.16.5, Flatpak com.protonvpn.www)

proton.vpn.app.gtk.utils.exception_handler:300 | CRITICAL | APP:CRASH | Unexpected error.
Traceback (most recent call last):
  File ".../proton/vpn/core/vpnconnector.py", line 394, in connect
    await self._on_connection_event(events.Up(...))
  File ".../proton/vpn/core/vpnconnector.py", line 463, in _on_connection_event
    event = await self._handle_on_event(event)
  File ".../proton/vpn/core/vpnconnector.py", line 448, in _handle_on_event
    return await self._update_state(new_state)
  File ".../proton/vpn/core/vpnconnector.py", line 499, in _update_state
    new_event = await self._current_state.run_tasks()
  File ".../proton/vpn/connection/states.py", line 257, in run_tasks
    await self.context.kill_switch.enable(...)
  File ".../proton/vpn/backend/networkmanager/killswitch/wireguard/wgkillswitch.py", line 70, in enable
    await self._ks_handler.add_vpn_server_route(server_ip=vpn_server.server_ip)
  File ".../proton/vpn/backend/networkmanager/killswitch/wireguard/killswitch_connection_handler.py", line 154, in add_vpn_server_route
    await _wrap_future(...)
  File ".../proton/vpn/backend/networkmanager/killswitch/wireguard/killswitch_connection_handler.py", line 56, in _wrap_future
    return await asyncio.wait_for(...)
  File ".../proton/vpn/backend/networkmanager/killswitch/wireguard/nmclient.py", line 375, in _add_ipv4_route
    cls._remove_ipv4_routes(active_connection, new_server_ip)
  File ".../proton/vpn/backend/networkmanager/killswitch/wireguard/nmclient.py", line 314, in _remove_ipv4_routes
    for i in range(config.get_num_routes()):
AttributeError: 'NoneType' object has no attribute 'get_num_routes'

Same crash reproduced with kill switch Off, on both the WireGuard and default-gateway server ("Fastest" pick), on two different WireGuard servers.

This is a regression

The exact same bridge topology (br0 + eth0 as bridge slave) worked fine with WireGuard on 2026-03-07 using the then-current native package proton-vpn-gtk-app 4.14.1-2 (Arch Linux). Log excerpt from that successful connection:

proton.vpn.core.connection:400 | INFO | CONN.CONNECT:START | ... Protocol: wireguard
proton.vpn.backend.networkmanager.core.networkmanager:91 | INFO | VPN server REACHABLE.
proton.vpn.core.connection:508 | INFO | CONN:STATE_CHANGED | Connected

So this isn't an inherent WireGuard+bridge incompatibility — something in the kill switch route-injection code regressed between then and the current release (reproduced on 4.16.5).

Environment

• OS: Arch Linux
• proton-vpn-gtk-app: 4.16.5 (Flatpak, com.protonvpn.www, flathub)
• Network: NetworkManager-managed bridge br0 with a USB-C-attached Ethernet adapter (eth0) as bridge port, plus Wi-Fi (wlan0) active concurrently
• Protocol: WireGuard (OpenVPN works fine as a workaround — its kill switch implementation uses exclusion routes on a dedicated connection profile instead of device_reapply on physical devices, so it never hits this code path)

Suggested fix

In NMClient.get_physical_devices(), exclude devices that are bridge/bond/team ports (e.g. check device.get_active_connection().get_connection().get_setting_connection().get_master() is None, or check device.get_master() is None on the NM.Device), since a port device never carries its own routable IPv4 config — only the master (bridge) device does. Alternatively, guard _add_ipv4_route/_remove_ipv4_routes against config is None and skip the device gracefully instead of crashing.