The down-script previously flushed the kill-switch source set on tunnel-down, which
disarmed the 'ip saddr @vpn_egress_gateway_src oifname != tun0 drop' rule exactly when
the tunnel was down — leaving fail-closed to ride only on the stale SNAT source being
non-routable (could leak if the edge NATed it). Retain the set instead: the drop rule
stays armed so any vpn-class packet leaving the WAN while the tunnel is down is EXPLICITLY
dropped (edge-independent). Safe because the up-script flush+re-adds the source on every
reconnect. Validated behavior: WS-NET fault-injection on k3s-dev-3 (2026-08-24).
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Author the vpn_egress_gateway role + playbook that stands up the NODE-level
PIA OpenVPN tunnel on egress-gateway node(s) — the host half of ADR-0015. The
CiliumEgressGatewayPolicy steering vpn-class pods is separate GitOps against the
same contract (group vpn_egress_gateway, node label
homelab.keithjackson.dev/egress-gateway=true, host iface tun0).
Honors all four ADR-0015 invariants:
- Split-tunnel (inv 1&2): OpenVPN route-noexec + pull-filter ignore
redirect-gateway => installs no routes; policy routing in a dedicated table
(51820) with `ip rule from <tun0-addr>` steers ONLY Cilium-SNAT'd vpn-class
traffic. Node default route stays on the WAN, tunnel up or down.
- Scoped nftables kill-switch (inv 3&4): own inet table, policy accept, drops
only tunnel-sourced (@vpn_egress_gateway_src set) packets not leaving tun0 =>
vpn-class fails closed (blackhole), node/internet-class fail open (WAN),
never matched.
PIA creds + the proven sweden-aes-128-cbc-udp-ip .ovpn come from Ansible vault
(host-side per ADR-0015/docs-07); vault keys documented, no secret values
committed. ansible-lint (production) + yamllint clean. Not deployed: green
conformance is gated on the live tunnel (operator host creds) — see role README.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
jakman85andClaude Opus 4.8