Network and access
- Public: nginx 80/443 with Certbot TLS. Catch-all
default_server (000-default-catchall) hardened 2026-09-28.
- WireGuard (10.88.0.0/24): internal tools only answer to this range. Managed by wireguard-ui (127.0.0.1:5000, vhost on :8082).
- Adding a VPN-only site is a two-part change: the nginx vhost (
listen 10.88.0.1:<port>; allow 10.88.0.0/24; deny all;) and a matching ufw rule sudo ufw allow in on wg0 to any port <port> proto tcp. INPUT policy is DROP and nothing blanket-allows wg0, so a vhost alone is silently unreachable — the client just hangs with no log line. 8094/SATNA was dead for this reason until 2026-11-15. Verify with sudo iptables -S ufw-user-input | grep <port> and a loopback-WG curl.
- DNS: NextDNS CLI + DoH is the host resolver (profile 6e78c6) since 2026-10-26;
nextdns deactivate to roll back.
- fail2ban active.
- Related: Docker perimeter hardening (the DOCKER-USER safeguard — read this before touching anything Docker-facing)
★ 2026-11-17 — DOCKER-USER safeguard was SILENTLY GONE (found and fixed)
The chain had regressed to a bare -A DOCKER-USER -j RETURN — no -j DROP, no conntrack
accept, no wg0 accept. Every container port published as 0.0.0.0:<port> in a compose file was
therefore WAN-reachable and only invisible because those ports happened to matter little.
Read Docker perimeter hardening for the full mechanism, the reproduction and the fix.
Verified live now: none of 3333 (agnt) · 5678 (n8n) · 9443 (portainer) · 5432 · 6379 · 8888 ·
9120/9121 · 8096 · 8098 · 13378 · 11434 is reachable on 45.61.176.215. Exactly four ports are
public: 22/tcp (key-only), 80, 443, 51820/udp. That is the intended posture — keep it.
Second service on 10.88.0.1:80 — the Command Center
http://10.88.0.1/ serves the private launcher page. Adding Command Center as a card does
not open a port; but the reports vhost binds the literal address 10.88.0.1 in addition to
:80, which is the same shape as the vhost row above — so it needs its own ufw rule
(sudo ufw allow in on wg0 to any port 80 proto tcp) or the WG-side copy is silently dead.
Adding a new WG-only listener — checklist
1. listen 10.88.0.1:<port>; allow 10.88.0.0/24; deny all; in the vhost.
2. sudo ufw allow in on wg0 to any port <port> proto tcp.
3. sudo iptables -S ufw-user-input | grep <port> to confirm the rule landed.
4. curl -m3 -o/dev/null -w '%{http_code}' http://10.88.0.1:<port>/ from the host.
5. Loopback-BG curl from a phone/peer — a vhost with no ufw rule hangs with no log line at all.