pbr: invalid default route if gateway is not configured for interface

[sumpfralle]

I am using pbr for routing outgoing SMTP traffic via a wireguard connection instead of the WAN interface.

My Setup

• interface wan is the local uplink
• interface smtp_vpn is the wireguard interface
  • my local IP on this interface is 10.90.12.1
  • a single peer is configured (and connected) with a few allowed_ips is configured (including 0.0.0.0/0)
  • resulting default route for the wireguard interface:

    root@router:~# ip r | grep -w default | grep smtp_vpn
    default dev smtp_vpn proto static scope link metric 20
    

Problem

The default route configured in the dynamically generated routing table pbr_smtp_vpn is wrong:

root@router:~# ip r show table pbr_smtp_vpn | grep -w default
default via 10.90.12.1 dev smtp_vpn

Here the local IP address (10.90.12.1) is used as a default gateway.
This obviously does not work.
Instead the original link-level route should have been applied (see "My Setup" above).

Workaround

In order to let the configured packets flow via the smtp_vpn interface, I configured a static route via OpenWrt for the table dynamically maintained by pbr:

network.cfg21c8b4=route
network.cfg21c8b4.interface='smtp_vpn'
network.cfg21c8b4.target='0.0.0.0/0'
network.cfg21c8b4.metric='15'
network.cfg21c8b4.table='pbr_smtp_vpn'

Probably the above is not robust enough, but it works as for now.

And now there are two default routes in this dynamic pbr table:

root@router:~# ip r show table pbr_smtp_vpn | grep -w default
default via 10.90.12.1 dev smtp_vpn
default dev smtp_vpn proto static scope link metric 15

For unknown reasons, the (unwanted) host-based route (via 10.90.12.1) is ignored by the kernel. This lets the static route handle the traffic.

Possible cause

/etc/init.d/pbr currently contains the following function:

pbr_get_gateway4() {
        local iface="$2" dev="$3" gw
        network_get_gateway gw "$iface" true
        if [ -z "$gw" ] || [ "$gw" = '0.0.0.0' ]; then
#               gw="$(ubus call "network.interface.${iface}" status | jsonfilter -e "@.route[0].nexthop")"
                gw="$(ip -4 a list dev "$dev" 2>/dev/null | grep inet | awk '{print $2}' | awk -F "/" '{print $1}')"
        fi
        eval "$1"='$gw'
}

The currently active gw= assignment picks the local IP as the gateway IP.
I cannot imagine a situation, where the local IP could be the proper choice for a default route.
Thus, the above fallback is probably of no use at all?

I would recommend one of the following changes:

• A) do not set a route, if no default gateway is found
  • no surprises, no problems
  • a static route should be a reasonable workaround for most affected users
• B) set a link-scope route, if no default gateway is found
  • this works for wireguard interfaces (just by accident)

Thanks for maintaining the great pbr package!