Setting up a Proxmox VM to act as a VPN gateway for other containers

By on
Warning
Edit from 2024-06-01: I wrote this guide for Proxmox 7.x in early 2023. I have since then upgraded my Proxmox server to v8, and with some minor tweaks here and there (because network cards suddenly got a new naming scheme again) it still works great for me. But I haven't re-tested this guide from scratch on Proxmox v8, so keep that in mind.

On my brand new Proxmox server I want to transparently have some of the containers connect to the internet via a VPN.

Instead of setting up a VPN connection in each container separately, I want to create one single Virtual Machine and have it act as a router/gateway for other containers.

This isn't super trivial apparently. I've found quite a few posts on the internet of people trying to do a similar thing, but never with a detailed guide.

So I decided to make sense of all the information that is out there and give it a go.

Proxmox host changes

A new network bridge interface is needed, and this has to be made on the Proxmox host.

In the "Server View" sidebar go to "Datacenter » your-host-name" and then select "System » Network" on the right. There you can create new network interfaces. Click on the Create dropdown button and select Linux Bridge.

A new dialog will open, here's how I configured it:

Linux bridge config

192.168.15.0/24 is the range this "gateway" will serve to other virtual machines & containers. By already setting this here, you'll still be able to actually connect to these machines from the Proxmox host, and thus configure some port forwarding.

The Virtual Machine itself

Create the VM

I'm going to use a Wireguard VPN connection, from PIA, Private Internet Access

Even though you can technically get a Wireguard VPN connection to work in an unprivileged LXC container, doing all the actual forwarding seems to fail, no matter if you're in a priviliged or unpriviliged container. And I've tried for quite a while. So I'm opting for a VM.

I decided to go with Debian 11 for this using the netinstall iso. Obviously I kept the install quite small: no desktop environment, just the essentials.

This VM does not need a lot of resources. Just 512 megabytes of memory, 1 CPU core and 8 gigabyte of storage will be plenty.

Warning
At some point the installation process will ask you what APT mirror you want to use. Do NOT select the default deb.debian.org mirror, because it blocks VPN connections.

Don't forget to add both network bridges to it. Once you're done, it's hardware setup should look like this:

VM hardware settings

Configure the network

After installing Debian, it's time to get started. You should already have these 2 interfaces:

  • ens18: This is the interface connected to vmbr0, so the "unsecure" connection
  • ens19: This is the interface connected to vmbr1, so the interface other containers/vms will use to get "secured" internet

Later, a pia interface will also be added.

Let's configure the ens19 interface first. This is the interface on which you'll be receiving incoming connections & serving IP addresses. But first it needs an IP address itself. In this guide I'm using 192.168.15.1, to match the subnet we defined earlier.

Let's add it to the interfaces file:

nano /etc/network/interfaces

And paste this:

auto ens19
iface ens19 inet static
        address 192.168.15.1/24
Warning
If you want, you can also give ens18 a static IP address. This might be a good idea, since our next action (restarting the network stack) seems to always make it forget its IP address.

Restart networking so the changes are applied

systemctl restart networking
Info
If ens18 forgot its IP address, simply doing dhclient ens18 should make it get a new one.
Info
If the restart command fails, and you get something like an RTNETLINK: File exists error, you just need to flush the device: ip addr flush dev ens18

Enable forwarding

IP forwarding is not enabled by default, so let's do that now.

nano /etc/sysctl.d/10-wireguard.conf

Paste in the following:

net.ipv4.ip_forward=1
net.ipv6.conf.all.forwarding=1

And load in the changes:

sysctl -p /etc/sysctl.d/10-ipforward.conf

From now on this will happen at boot

Setup your PIA VPN connection

PIA's wireguard VPN has a downside: it uses tokens that expire every 24 hours. So trying to reconnect after 24 hours have passed, will fail.

Luckily, plenty of scripts to automate this process exist, even some made by PIA themselves. I found a very interesting and short one, made by jbagwell27 called pia-autoconnect-wireguard

I've made a few changes to the script, so it'll work for our usecase. First of all: install certain dependencies:

apt install git curl jq iptables wireguard-tools netselect-apt

Download PIA's certificate from their Github repository:

mkdir /var/lib/pia
curl https://raw.githubusercontent.com/pia-foss/manual-connections/master/ca.rsa.4096.crt --output /var/lib/pia/ca.rsa.4096.crt

To make it easier to manage changing any values in the future, we'll create a single variable file:

mkdir /etc/wireguard/helper/
nano /etc/wireguard/helper/vars

Then paste in the following:

PREFERRED_REGION="us_atlanta" # The preferred region to connect to
PIA_USER="p0123456"           # Your PIA username
PIA_PASS="xxx"                # Your PIA password

INSECURE_FACE="ens18"         # NIC connected to the internet
SECURE_FACE="ens19"           # NIC other containers will connect to
WG_FACE="pia"                 # Wireguard/PIA NIC 

I had to make some modifications to the script, so that it includes some PostUp and PostDown actions. Let's create it manually:

nano /usr/local/bin/pia-autoconnect.sh

Paste in the following, but don't forget to edit the PREFERRED_REGION, PIA_USER and PIA_PASS variables.

#!/usr/bin/bash

source /etc/wireguard/helper/vars

# This only works with Wireguard with port forwarding disabled.
VPN_PROTOCOL=wireguard
PIA_PF="false"

selectedRegion=$PREFERRED_REGION

serverlist_url='https://serverlist.piaservers.net/vpninfo/servers/v6'

# Make sure curl always uses the "insecure" network to download something
curl() {
    command curl --interface $INSECURE_FACE "$@"
}

all_region_data=$(curl -s "$serverlist_url" | head -1)

get_selected_region_data() {
    regionData="$(echo "$all_region_data" |
        jq --arg REGION_ID "$selectedRegion" -r \
            '.regions[] | select(.id==$REGION_ID)')"
}

get_selected_region_data

WG_SERVER_IP=$(echo "$regionData" | jq -r '.servers.wg[0].ip')
WG_HOSTNAME=$(echo "$regionData" | jq -r '.servers.wg[0].cn')

timeout_timestamp() {
    date +"%c" --date='1 day' # Timestamp 24 hours
}

generateTokenResponse=$(curl -s --location --request POST \
    'https://www.privateinternetaccess.com/api/client/v2/token' \
    --form "username=$PIA_USER" \
    --form "password=$PIA_PASS")
token=$(echo "$generateTokenResponse" | jq -r '.token')
tokenExpiration=$(timeout_timestamp)
PIA_TOKEN=$token${nc}

export PIA_TOKEN

if [[ -f /proc/net/if_inet6 ]] &&
    [[ $(sysctl -n net.ipv6.conf.all.disable_ipv6) -ne 1 ||
    $(sysctl -n net.ipv6.conf.default.disable_ipv6) -ne 1 ]]; then

    sysctl -n net.ipv6.conf.all.disable_ipv6
    sysctl -n net.ipv6.conf.default.disable_ipv6
fi

privKey=$(wg genkey)
export privKey
pubKey=$(echo "$privKey" | wg pubkey)
export pubKey
wireguard_json="$(curl -s -G \
    --connect-to "$WG_HOSTNAME::$WG_SERVER_IP:" \
    --cacert "/var/lib/pia/ca.rsa.4096.crt" \
    --data-urlencode "pt=${PIA_TOKEN}" \
    --data-urlencode "pubkey=$pubKey" \
    "https://${WG_HOSTNAME}:1337/addKey")"

export wireguard_json

if [[ $(echo "$wireguard_json" | jq -r '.status') != "OK" ]]; then
    echo >&2 -e "Server did not return OK. Stopping now.${nc}"
    exit 1
fi

wg-quick down pia

dnsServer=$(echo "$wireguard_json" | jq -r '.dns_servers[0]')
dnsSettingForVPN="DNS = $dnsServer"

mkdir -p /etc/wireguard

echo "
[Interface]
Address = $(echo "$wireguard_json" | jq -r '.peer_ip')
PrivateKey = $privKey
PostUp = /etc/wireguard/helper/add-nat-routing.sh
PostDown = /etc/wireguard/helper/remove-nat-routing.sh
$dnsSettingsForVPN
[Peer]
PersistentKeepalive = 25
PublicKey = $(echo "$wireguard_json" | jq -r '.server_key')
AllowedIPs = 0.0.0.0/0
Endpoint = ${WG_SERVER_IP}:$(echo "$wireguard_json" | jq -r '.server_port')
" >/etc/wireguard/pia.conf || exit 1

wg-quick up pia

Don't forget to make it executable:

chmod +x /usr/local/bin/pia-autoconnect.sh

Create the up & down hooks

Now create the file that sets up the NAT after PIA is connected:

touch /etc/wireguard/helper/add-nat-routing.sh
chmod +x /etc/wireguard/helper/add-nat-routing.sh
nano /etc/wireguard/helper/add-nat-routing.sh

Paste in the following lines that will actually route the data:

#!/bin/bash
source /etc/wireguard/helper/vars

iptables -t nat -A POSTROUTING -o $WG_FACE -j MASQUERADE
iptables -A FORWARD -i $SECURE_FACE -o $WG_FACE -j ACCEPT
#iptables -A FORWARD -i $SECURE_FACE -o $INSECURE_FACE$ -d 192.168.0.0/16 -j ACCEPT

systemctl restart isc-dhcp-server

Now create the file that brings down the NAT when PIA disconnects:

touch /etc/wireguard/helper/remove-nat-routing.sh
chmod +x /etc/wireguard/helper/remove-nat-routing.sh
nano /etc/wireguard/helper/remove-nat-routing.sh

And make it flush all the iptable rules:

#!/bin/bash
source /etc/wireguard/helper/vars

iptables -F

systemctl restart isc-dhcp-server

Create a SystemD service

Now let's create a systemd service, which will start the connection at boot.

nano /etc/systemd/system/pia-connect.service

And pase this in:

[Unit]
Description=PIA-Wireguard Connection

[Service]
Type=oneshot
ExecStart=/bin/bash /usr/local/bin/pia-autoconnect.sh

[Install]
WantedBy=multi-user.target

And enable the service (but don't start it yet):

systemctl enable pia-connect.service

Setup cronjob

We want to restart this script automatically every day, so that it'll always have a valid connection key. We'll use a cronjob for that:

crontab -e

And paste in this line:

0 4 * * * /usr/bin/systemctl restart pia-connect.service

This will run the script every day at 04:00 at night.

Configure DHCP server

Let's automatically serve IP addresses to containers that connect to us without setting a static IP address

Warning
The DHCP server will try to automatically start already, but it will fail. This is not an issue.
apt install dnsutils isc-dhcp-server

We have to make sure the DHCP server does not start by itself, the network config will take care of that

systemctl disable isc-dhcp-server.service

Now actually add the DHCP configuration:

nano /etc/dhcp/dhcpd.conf

And add the following:

authoritative;

subnet 192.168.15.0 netmask 255.255.255.0 {
  range 192.168.15.128 192.168.15.254;
  
  option routers 192.168.15.1;
  option subnet-mask 255.255.255.0;
  option broadcast-address 192.168.15.255;
  option domain-name-servers 8.8.8.8;
  
  default-lease-time 86400;
  max-lease-time 259200;
}

Then change the default interface for the DHCP server:

nano /etc/default/isc-dhcp-server

And make sure it looks like this:

INTERFACESv4="ens19"
INTERFACESv6=""

Test it all!

Bring up the connection:

systemctl start pia-connect

Now test the connection by getting your current IP address:

curl http://wtfismyip.com/text

Potential problems

DNS resolver doesn't work

When creating a new container, it's DNS is set to the same value as the Proxmox host. This is probbaly not reachable over the new bridge we've made. So don't forget to configure this to something else, like 8.8.8.8

Unable to install containers from "Proxmox Helper Scripts"

It's best to install the containers using the default bridge, and then switch it to the new one afterwards.

Comments

How did you end setting up the reverse proxy in the container?
Hi, thanks for this - it worked great. I didn't use your main config, just used `wg-quick@wg0` as I have a static main peer. I honestly didn't realise it would be so easy to configure Wireguard over the network so that the other containers/VMs on the same interface would then be able to use it. I observed during peak network load (eg maxing out my gigabit connection) on a container/VM, the WG VM would spike CPU like crazy. I'm unsure if something needs to be configured, but I tried changing the VM type to `host`, turning on AES etc, but it didn't solve it. I changed to using a container, and it worked fine. I didn't see any leaks - though privacy is not my main concern, I'm sure you can implement some sort of killswitch. The container didn't seem to have the same spikes the VM under load (as Wireguard is really CPU friendly), and the install steps are pretty much identical. It also uses <30mb of memory :-). Couple of notes from my findings: 1. I recommend installing qemu-guest-agent on the VM so you get all the niceties that come with that. 2. I also noticed you had e1000 in your screenshots, you'd probably want virtio instead. 3. I had weird issues with ipv6 during this, where it would be assigned to my interfaces, despite me telling it not to (Proxmox 8.x). You might need to force disable IPv6 on the VMs/containers so you don't accidentally send traffic over ipv6. 4. Setting the WG peers DNS Server in the containers DNS Servers & Domain in the Proxmox GUI would prevent accidentally assigning the wrong DNS server.
Thanks for this guide. Managed to set this up with Windscribe and get a LXC container connected to the gateway. Only thing I haven't figured out is "How to force the LXC to use the windscribe DNS server rather than 8.8.8.8. Essentially to prevent DNS leak. " The VM with the wireguard config correctly uses the Windscribe DNS server and there are no leaks. Just not sure how to get the LXC to use this same DNS server.
@Jaeger: Could you try running the Bash script directly, and see what the error is?
Went through this step by step and got the following: ar 24 02:54:34 piavpn systemd[1]: Starting PIA-Wireguard Connection... Mar 24 02:54:35 piavpn bash[487]: 0 Mar 24 02:54:35 piavpn bash[488]: 0 Mar 24 02:54:35 piavpn bash[465]: Server did not return OK. Stopping now. Mar 24 02:54:35 piavpn systemd[1]: pia-connect.service: Main process exited, code=exited, status=1/FAILURE Mar 24 02:54:35 piavpn systemd[1]: pia-connect.service: Failed with result 'exit-code'. Mar 24 02:54:35 piavpn systemd[1]: Failed to start PIA-Wireguard Connection. Any ideas?
That's a good question. Currently I have 1 LXC container that uses this system, and it's received an IP address from the VM's DHCP server of 192.168.15.2 On the Proxmox host itself, I'm able to reach this IP. But indeed: outside of the Proxmox host (or even the other containers) I can't reach it. There's 2 ways to fix this, both involve a reverse proxy. First: install a reverse proxy on the Proxmox host itself. Which is the least clean solution. Second (and the one I went with): install a reverse proxy as another container, and give it access to both the `vmbr0` and `vmbr1` bridge.
Thanks for the guide. Is it possible to configure this so that we can access a GUI on one of the clients through the local network? With this configuration the clients are not reachable from outside proxmox as the router is not aware of the IP being assigned by the VPN VM's dhcp server.
Name
Email
Website
Body
submit error done Busy
Jelle De Loecker