Currently, I have two VPN clients on most of my devices:
- One for connecting to a LAN
- One commercial VPN for privacy reasons
I usually stay connected to the commercial VPN on all my devices, unless I need to access something on that LAN.
This setup has a few drawbacks:
- Most commercial VPN providers have a limit on the number of simulations connected clients
- I either obfuscate my IP or am able to access resources on that LAN, including my Pi-Hole fur custom DNS-based blocking
One possible solution for this would be to route all internet traffic through a VPN client on the router in the LAN and figuring out how to still be able to at least have a port open for the VPN docker container allowing access to the LAN. But then the ability to split tunnel around that would be pretty hard to achieve.
I want to be able to connect to a VPN host container on the LAN, which in turn routes all internet traffic through another VPN client container while allowing LAN traffic, but still be able to split tunnel specific applications on my Android/Linux/iOS devices.
Basically this:
+---------------------+ internet traffic +--------------------+
| | remote LAN traffic | |
| Client |------------------->|VPN Host Container |
| (Android/iOS/Linux) | |in remote LAN |
| | | |
+---------------------+ +--------------------+
| | |
| remote LAN traffic| | internet traffic
split tunneled traffic| |-------- |
| | v
v | +---------------------------+
+---------------------+ v | |
| regular LAN or | +-----------+ | VPN Client Container |
| internet connection | |remote LAN | | connects to commercial VPN|
+---------------------+ +-----------+ | |
| |
+---------------------------+
Any recommendations on how to achieve this, especially considering client apps for Android and iOS with the ability to split tunnel per application?
Update:
~~Got it by following this guide.~~
Ended up modifying this setup to have better control over potential IP leakage
I've been tempted by Tailscale a few times before, but I don't want to depend on their proprietary clients and control server. The latter could be solved by selfhosting Headscale, but at this point I figure that going for a basic Wireguard setup is probably easier to maintain.
I'd like to have a look at your rules setup, I'm especially curious if/how you approached the event of the commercial VPN Wireguard tunnel(s) on your exit node(s) going down, which depending on the setup may send requests meant to go through the commercial VPN through your VPS exit node.
Personally, I ended up with two Wireguard containers in the target LAN, a wireguard-server and a **wireguard-client **container.
They both share a docker network with a specific subnet {DOCKER_SUBNET} and wireguard-client has a static IP {WG_CLIENT_IP} in that subnet.
The wireguard-client has a slightly altered standard config to establish a tunnel to an external endpoint, a commercial VPN in this case:
where
are responsible for properly routing traffic coming in from outside the container and
is your standard kill-switch meant to block traffic going out of any network interface except the tunnel interface in the event of the tunnel going down.
The wireguard-server container has these PostUPs and -Downs:
PostUp = iptables -A FORWARD -i %i -j ACCEPT; iptables -A FORWARD -o %i -j ACCEPT; iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADEdefault rules that come with the template and allow for routing packets through the server tunnel
PostUp = wg set wg0 fwmark 51820the traffic out of the tunnel interface get marked
PostUp = ip -4 route add 0.0.0.0/0 via {WG_CLIENT_IP} table 51820add a rule to routing table 51820 for routing all packets through the wireguard-client container
PostUp = ip -4 rule add not fwmark 51820 table 51820packets not marked should use routing table 51820
PostUp = ip -4 rule add table main suppress_prefixlength 0respect manual rules added to main routing table
PostUp = ip route add {LAN_SUBNET} via {DOCKER_SUBNET_GATEWAY_IP} dev eth0route packages with a destination in {LAN_SUBNET} to the actual {LAN_SUBNET} of the host
PostDown = iptables -D FORWARD -i %i -j ACCEPT; iptables -D FORWARD -o %i -j ACCEPT; iptables -t nat -D POSTROUTING -o eth0 -j MASQUERADE; ip route del {LAN_SUBNET} via {DOCKER_SUBNET_GATEWAY_IP} dev eth0delete those rules after the tunnel goes down
Basically the same kill-switch as in wireguard-client, but with the mark manually substituted since the command it relied on didn't work in my server container for some reason and AFAIK the mark actually doesn't change.
Now do I actually need the kill-switch in wireguard-server? Is the kill-switch in wireguard-client sufficient? I'm not even sure anymore.
Your setup looks more advanced than mine, and I'd really like to do something similar. I'm just going to copy/paste what I have with some addresses replaced by:
VPN_IPV4_CLIENT_ADDRESS: The WireGuard IPv4 address of the VPN provider's interface (e.g. 172.0.0.1)VPN_IPV6_CLIENT_ADDRESS: The WireGuard IPv6 address of the VPN provider's interfaceVPN_IPV6_CLIENT_ADDRESS_PLUS_ONE: The next IPv6 address that comes afterVPN_IPV6_CLIENT_ADDRESS. I can't remember the logic behinds this but I'd found an article online explaining it.WG_INTERFACE: The WireGuard network interface name (e.g. wg0) for the commercial VPNI left
100.64.0.0/10,fd7a:115c:a1e0::/96in my example because those are the networks Tailscale traffic will come from. I also lefttailscale0because that is the typical interface. Obviously these can be changed to support any network.I'm using Alpine Linux so I don't have the
PostUp,PostDown, etc. in my WireGuard configuration. I'm not usingwg-quickat all.Before I hit paste, one thing I'll say is I haven't addressed the "kill switch" yet. But so far (~4 months) when the VPN provider's tunnel goes down nothing leaks. 🤞
Forgot to mention that I run a DNS server for blocking too. When using Tailscale I’ve found it’s important to use their resolver as upstream otherwise App Connectors won’t work (the VPN provider tunnels on each VPS routes to different countries so DNS wasn’t in sync). This kind of sucks but I make do with it after a month or two of App Connectors being very iffy.