#!/usr/bin/bash

# VPN monitor — connects OpenVPN when internet is up, disconnects when down.
# Drops and reconnects when WiFi SSID changes (stale tunnel prevention).
# On non-home networks, resolves VPN hostname via 8.8.8.8 and passes IP directly.
# Keepalive: pings gateway through tunnel, two failures 10s apart = reconnect.
# SIGTERM: gracefully stops tunnel and exits.

SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
CONF="$SCRIPT_DIR/vpn.ovpn"
VPN_HOST="vpn.hanson.xyz"
VPN_PORT="1194"
HOME_SSID="risa"
VPN_GW="192.168.69.1"
CHECK_HOST="1.1.1.1"
INTERVAL=30
CONNECT_TIMEOUT=30
MAX_FAILURES=3
PREV_SSID=""
FAIL_COUNT=0
ACTIVE_VPN_IP=""

kill_vpn() {
  killall openvpn 2>/dev/null
  # Clean up host route to VPN server
  if [ -n "$ACTIVE_VPN_IP" ]; then
    ip route del "$ACTIVE_VPN_IP/32" 2>/dev/null
    ACTIVE_VPN_IP=""
  fi
}

get_default_gw() {
  ip route show default | awk '/via/ {print $3; exit}'
}

resolve_vpn() {
  if [ "$CURR_SSID" != "$HOME_SSID" ]; then
    dig +short @8.8.8.8 "$VPN_HOST" 2>/dev/null | tail -1
  else
    dig +short "$VPN_HOST" 2>/dev/null | tail -1
  fi
}

# Graceful shutdown on SIGTERM
shutdown() {
  echo "$(date): SIGTERM received, stopping vpn and exiting"
  kill_vpn
  exit 0
}
trap shutdown SIGTERM SIGINT

# Kill other instances of this script and wait for graceful shutdown
for pid in $(pgrep -f 'vpn-monitor.sh' | grep -v $$); do
  kill "$pid" 2>/dev/null
done
sleep 5
# Force kill any that didn't exit
for pid in $(pgrep -f 'vpn-monitor.sh' | grep -v $$); do
  kill -9 "$pid" 2>/dev/null
done

# Kill any existing VPN and clean up
kill_vpn
sleep 1

while true; do
  CURR_SSID="$(iwgetid -r 2>/dev/null)"

  # Detect SSID change (only when switching between two known networks)
  if [ -n "$PREV_SSID" ] && [ -n "$CURR_SSID" ] && [ "$PREV_SSID" != "$CURR_SSID" ]; then
    echo "$(date): wifi changed from '$PREV_SSID' to '$CURR_SSID', dropping vpn"
    kill_vpn
    FAIL_COUNT=0
    sleep 5
  fi
  PREV_SSID="$CURR_SSID"

  if ping -c 1 -W 3 "$CHECK_HOST" > /dev/null 2>&1; then
    # Internet is up — check tunnel health if connected
    if ip link show tun0 > /dev/null 2>&1; then
      # Keepalive: ping gateway through tunnel, two failures 10s apart = dead
      if ! ping -c 1 -W 3 -I tun0 "$VPN_GW" > /dev/null 2>&1; then
        sleep 10
        if ! ping -c 1 -W 3 -I tun0 "$VPN_GW" > /dev/null 2>&1; then
          echo "$(date): keepalive failed twice, dropping vpn"
          kill_vpn
          sleep 5
          continue
        fi
      fi
    fi

    # Start VPN if not running
    if ! ip link show tun0 > /dev/null 2>&1; then
      if [ "$FAIL_COUNT" -ge "$MAX_FAILURES" ]; then
        # Back off after repeated failures — just wait for next interval
        sleep "$INTERVAL"
        continue
      fi

      # Resolve VPN server IP (via 8.8.8.8 on non-home networks)
      RESOLVED_IP="$(resolve_vpn)"
      if [ -z "$RESOLVED_IP" ]; then
        echo "$(date): failed to resolve $VPN_HOST (ssid=$CURR_SSID)"
        FAIL_COUNT=$((FAIL_COUNT + 1))
        sleep "$INTERVAL"
        continue
      fi

      # Add host route to VPN server via current default gateway
      # so VPN traffic survives tun0 coming up
      GW="$(get_default_gw)"
      if [ -n "$GW" ]; then
        ip route replace "$RESOLVED_IP/32" via "$GW"
        echo "$(date): host route $RESOLVED_IP via $GW"
      fi
      ACTIVE_VPN_IP="$RESOLVED_IP"

      echo "$(date): starting openvpn -> $RESOLVED_IP (attempt $((FAIL_COUNT + 1))/$MAX_FAILURES, ssid=$CURR_SSID)"
      nice -n 19 openvpn --config "$CONF" --remote "$RESOLVED_IP" "$VPN_PORT" --daemon --log-append /tmp/openvpn.log

      # Wait for tunnel to come up
      CONNECTED=0
      for i in $(seq 1 "$CONNECT_TIMEOUT"); do
        if ip link show tun0 > /dev/null 2>&1; then
          CONNECTED=1
          break
        fi
        sleep 1
      done

      if [ "$CONNECTED" -eq 1 ]; then
        echo "$(date): vpn connected (took ${i}s)"
        FAIL_COUNT=0
      else
        echo "$(date): vpn failed to connect within ${CONNECT_TIMEOUT}s, killing"
        kill_vpn
        FAIL_COUNT=$((FAIL_COUNT + 1))
        if [ "$FAIL_COUNT" -ge "$MAX_FAILURES" ]; then
          echo "$(date): $MAX_FAILURES consecutive failures, backing off"
        fi
      fi
    fi
  else
    # Internet is down — kill VPN if running
    if ip link show tun0 > /dev/null 2>&1; then
      echo "$(date): internet down, stopping openvpn"
      kill_vpn
    fi
    FAIL_COUNT=0
  fi
  sleep "$INTERVAL"
done