Probably works

.git_ignore

@@ -0,0 +1,4 @@
+site-a.yml
+site-b.yml
+.env
+.DS_Store

README.md

@@ -5,18 +5,17 @@ is one TCP port on one host per side.
 
 ```
 ansible/
-├── inventory.ini
 ├── wg-relay.yml           # the whole thing
-├── group_vars/all.yml     # defaults
+├── defaults.yml           # defaults
-├── host_vars/site-a.yml   # Site A (Site B writes the mirror image)
+├── vars/
+│   ├── site-a.yml         # Site A (Site B writes the mirror image)
+│   └── site-b.yml
 └── templates/
     ├── wg0.conf.j2
     ├── socat-relay.service.j2
     └── nftables-wgrelay.conf.j2
 ```
 
-Needs `ansible-core` and nothing else — no collections to install.
-
 ## How it works
 
 The relay VM never routes. `ip_forward` is pinned to 0 and the nftables
@@ -24,7 +23,7 @@ forward chain policy is `drop`, so there is no path from the tunnel to the
 LAN. The only things listening are socat processes you declared:
 
 - an **inbound** relay bound to the tunnel address, forwarding to one
-  `host:port` on the local LAN
+`host:port` on the local LAN
 - an **outbound** relay bound to the LAN address, forwarding down the tunnel
 
 Local clients point at the relay VM's LAN address and never learn a tunnel
@@ -43,20 +42,48 @@ Each side's restrictions live entirely on its own machine. There is no
 mechanism for one site to prove to the other how its half is configured — and
 none is needed, since the half that protects you is the half you control.
 
-## First run
+## Prerequisites
 
-1. Run it on each side with the peer key left as the placeholder. It
+You can use the ansible repositories or another installation method if desired, all that matters is you have ansible
-   generates a keypair and prints the public key.
-2. Exchange public keys, fill in `wg_peers[].public_key`, run again.
-3. Point the local client at the relay VM's LAN address and port.
 
-```bash
+Needs `ansible-core`, no other packages or collections are needed. The playbook will install additional required packages
-ansible-playbook -i inventory.ini wg-relay.yml --check --diff
+
-ansible-playbook -i inventory.ini wg-relay.yml
+```
+sudo apt update && sudo apt install ansible
 ```
 
-The playbook ends by asserting forwarding is off, waiting for each relay port
+## Setup
-to accept connections, and printing `wg show`.
+
+The playbook and `defaults.yml` are identical at both sites. Everything that
+differs lives in that site's vars file — copy `vars/site-a.yml` and edit:
+
+- `relay_host` / `relay_user` — which box to configure. Set it to the relay
+  VM to run over SSH, or leave it as `localhost` to configure the machine
+  you're sitting on.
+- `wg_address` — this side's tunnel address, as a `/32`.
+- `wg_peers` — the other site's public key, endpoint and `/32`.
+- `relays` — the inbound and outbound listeners.
+
+Values in the vars file override `defaults.yml`, because `-e` outranks
+everything else. Adding a site later means one more file in `vars/`; the
+playbook doesn't change.
+
+## Installation
+
+Run it at each site against that site's vars file. No inventory file needed.
+
+```
+ansible-playbook wg-relay.yml -e @vars/site-a.yml --check --diff
+ansible-playbook wg-relay.yml -e @vars/site-a.yml
+```
+
+Leave the peer's public key as the placeholder on the first run — it generates
+a keypair and prints this side's public key. Exchange public keys, fill in
+`wg_peers[].public_key` in each vars file, then run again. Finally, point the
+local client at the relay VM's LAN address and port.
+
+Each run ends by asserting forwarding is off, waiting for each relay port to
+accept connections, and printing `wg show`.
 
 ## Notes
 
@@ -65,8 +92,8 @@ to accept connections, and printing `wg show`.
 - Keys live in `/etc/wireguard/private.key` (0600, root-only). To manage them
   centrally instead, set `wg_private_key` from Ansible Vault and the playbook
   skips generation.
-- Removing a relay from `host_vars` stops and deletes its systemd unit on the
+- Removing a relay from the vars file stops and deletes its systemd unit on
-  next run — no orphaned listeners.
+  the next run — no orphaned listeners.
 - If either uplink is PPPoE or tunnelled, drop `wg_mtu` to 1380. Sustained
   bulk transfers are what find PMTU blackholes, and the symptom is a stall
   mid-transfer rather than a clean failure.

defaults.yml

@@ -0,0 +1,63 @@
+---
+# ---------------------------------------------------------------------------
+# Defaults shared by every site. Do not edit this file per-site — put the
+# differences in your site vars file (see vars/site-a.yml) and pass it with
+# -e @vars/<site>.yml, which overrides everything here.
+#
+# Anything marked REQUIRED has no sensible default and must be set there.
+# ---------------------------------------------------------------------------
+
+wg_interface: wg0
+wg_listen_port: 51820
+
+# 1420 leaves room for the WireGuard header inside a 1500-byte path.
+# Drop to 1380 if either side is on PPPoE or a tunnelled uplink.
+wg_mtu: 1420
+
+# REQUIRED: this box's tunnel address, as a /32. Never a /24 — the mask is
+# what stops wg-quick installing a route to the whole tunnel subnet.
+wg_address: ""
+
+# Leave undefined to have the playbook generate a keypair on first run and print
+# the public key. Set it (from Ansible Vault) if you manage keys centrally.
+# wg_private_key: "..."
+
+# REQUIRED: the remote site.
+#   allowed_ips  - their /32 and nothing wider. This is the kernel's source
+#                  filter, not just a route.
+#   endpoint     - omit entirely if they're behind CGNAT and dial out to you.
+wg_peers: []
+#   - name: site-b
+#     public_key: "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx="
+#     allowed_ips: 10.99.0.2/32
+#     endpoint: site-b.example.com:51820
+#     keepalive: 25
+
+# REQUIRED: the socat listeners.
+#   direction: inbound  - listens on the tunnel, forwards to your LAN
+#   direction: outbound - listens on your LAN, forwards down the tunnel
+#   source_allow        - nftables source restriction for this listener
+relays: []
+#   - name: in-site-b
+#     direction: inbound
+#     listen_address: 10.99.0.1
+#     listen_port: 9000
+#     target: 192.168.1.50:9000
+#     source_allow: 10.99.0.2/32
+#   - name: out-site-b
+#     direction: outbound
+#     listen_address: 192.168.1.10
+#     listen_port: 9001
+#     target: 10.99.0.2:9000
+#     source_allow: 192.168.1.50/32
+
+# Unprivileged account the socat units run as. 'nogroup' is Debian/Ubuntu;
+# on RHEL-family set both to 'nobody'.
+socat_user: nobody
+socat_group: nogroup
+
+# Safety rails. Only flip these if you know why.
+#   forwarding_enabled: true turns this box into a router between the tunnel
+#   and your LAN, which defeats the point of the relay design.
+wg_relay_forwarding_enabled: false
+wg_relay_manage_nftables: true

site-a-example.yml

@@ -9,10 +9,8 @@
 # ===========================================================================
 
 # --- which box to configure ------------------------------------------------
-# Set to the relay VM's address to configure it over SSH, or leave as
+relay_host: localhost
-# localhost to configure the machine you're running this on.
+#relay_user: admin
-relay_host: 192.168.1.10
-relay_user: admin
 
 # --- this side of the tunnel -----------------------------------------------
 wg_address: 10.99.0.1/32
@@ -22,8 +20,6 @@ wg_listen_port: 51820
 wg_peers:
   - name: site-b
     public_key: "PASTE_SITE_B_PUBLIC_KEY_HERE"
-    # Omit endpoint on whichever side isn't reachable from outside; the
-    # other side dials in and keepalives hold the path open.
     endpoint: site-b.example.com:51820
     allowed_ips: 10.99.0.2/32
     keepalive: 25

site-b-example.yml

@@ -10,8 +10,8 @@
 # ===========================================================================
 
 # --- which box to configure ------------------------------------------------
-relay_host: 10.0.0.10
+relay_host: localhost
-relay_user: admin
+#
 
 # --- this side of the tunnel -----------------------------------------------
 wg_address: 10.99.0.2/32

templates/nftables-wgrelay.conf.j2

@@ -0,0 +1,40 @@
+{{ ansible_managed | comment }}
+
+table inet wgrelay
+delete table inet wgrelay
+
+table inet wgrelay {
+    chain input {
+        type filter hook input priority 0; policy accept;
+
+        ct state established,related accept
+        ct state invalid drop
+
+        # --- traffic arriving over the tunnel -------------------------------
+        # Ping across the tunnel is useful for debugging and harmless.
+        iifname "{{ wg_interface }}" icmp type echo-request accept
+        iifname "{{ wg_interface }}" icmpv6 type echo-request accept
+
+{% for r in relays if r.direction == 'inbound' %}
+        # {{ r.name }}: only {{ r.source_allow }} may reach this listener
+        iifname "{{ wg_interface }}" ip saddr {{ r.source_allow }} tcp dport {{ r.listen_port }} accept
+{% endfor %}
+
+        # Anything else off the tunnel is not ours. This is the rule that
+        # makes "one port, one host" true rather than aspirational.
+        iifname "{{ wg_interface }}" counter drop
+
+        # --- LAN-facing relay listeners -------------------------------------
+{% for r in relays if r.direction == 'outbound' %}
+        iifname != "{{ wg_interface }}" ip saddr {{ r.source_allow }} tcp dport {{ r.listen_port }} accept
+        tcp dport {{ r.listen_port }} counter drop
+{% endfor %}
+    }
+
+    chain forward {
+        type filter hook forward priority 0; policy drop;
+        # Nothing is routed through this box, by design. Combined with
+        # net.ipv4.ip_forward=0 this is belt and braces.
+        counter
+    }
+}

templates/socat-relay.service.j2

@@ -0,0 +1,39 @@
+{{ ansible_managed | comment }}
+[Unit]
+Description=Relay {{ item.name }} ({{ item.listen_address }}:{{ item.listen_port }} -> {{ item.target }})
+After=network-online.target wg-quick@{{ wg_interface }}.service
+Wants=network-online.target
+{% if item.direction == 'inbound' %}
+# Inbound listeners bind to the tunnel address, which doesn't exist until
+# wg-quick has brought the interface up.
+BindsTo=wg-quick@{{ wg_interface }}.service
+{% endif %}
+
+[Service]
+Type=simple
+ExecStart=/usr/bin/socat -d \
+  TCP-LISTEN:{{ item.listen_port }},bind={{ item.listen_address }},fork,reuseaddr,keepalive,keepidle=60,keepintvl=10,keepcnt=5 \
+  TCP:{{ item.target }},keepalive,keepidle=60,keepintvl=10,keepcnt=5
+
+# Bind may fail for a few seconds after boot while wg0 settles; just retry.
+Restart=always
+RestartSec=5
+
+# socat needs no privileges beyond binding a high port.
+User={{ socat_user }}
+Group={{ socat_group }}
+NoNewPrivileges=yes
+PrivateTmp=yes
+PrivateDevices=yes
+ProtectSystem=strict
+ProtectHome=yes
+ProtectKernelTunables=yes
+ProtectKernelModules=yes
+ProtectControlGroups=yes
+RestrictAddressFamilies=AF_INET AF_INET6
+RestrictNamespaces=yes
+MemoryDenyWriteExecute=yes
+CapabilityBoundingSet=
+
+[Install]
+WantedBy=multi-user.target

templates/wg0.conf.j2

@@ -0,0 +1,24 @@
+{{ ansible_managed | comment }}
+# Relay tunnel. This box forwards nothing; see the sysctl and nftables
+# tasks in wg-relay.yml.
+
+[Interface]
+Address = {{ wg_address }}
+ListenPort = {{ wg_listen_port }}
+MTU = {{ wg_mtu }}
+PrivateKey = {{ wg_relay_effective_private_key }}
+
+{% for peer in wg_peers %}
+# {{ peer.name }}
+[Peer]
+PublicKey = {{ peer.public_key }}
+AllowedIPs = {{ peer.allowed_ips }}
+{% if peer.endpoint is defined and peer.endpoint %}
+Endpoint = {{ peer.endpoint }}
+{% endif %}
+{% if peer.preshared_key is defined and peer.preshared_key %}
+PresharedKey = {{ peer.preshared_key }}
+{% endif %}
+PersistentKeepalive = {{ peer.keepalive | default(25) }}
+
+{% endfor %}

wg-relay.yml

@@ -8,12 +8,10 @@
 #   ansible-playbook wg-relay.yml -e @vars/site-a.yml --check --diff
 #   ansible-playbook wg-relay.yml -e @vars/site-a.yml
 #
-# No inventory file needed. Set relay_host to localhost (or leave it unset)
+# No inventory file needed. Set relay_host to localhost or leave it unset
 # to configure the machine you're running on. Defaults come from
-# defaults.yml; anything in the vars file wins, since -e outranks it.
+# defaults.yml.
-#
+# The restrictions are enforced locally at each site.
-# The restrictions are enforced locally at each site, so neither side can
-# widen the other's exposure.
 
 - name: Resolve which box to configure
   hosts: localhost