Enhance WireGuard startup and keepalive monitoring logic in entrypoint and persistentkeepalive scripts

Dockerfile

@@ -25,4 +25,4 @@ VOLUME ["/etc/wireguard"]
 ENV WG_QUICK_USERSPACE_IMPLEMENTATION=boringtun LOGNAME=nobody INTERFACE=wg0
 COPY set_gateway.sh /etc/wireguard/
 COPY persistentkeepalive.sh /etc/wireguard/
-ENTRYPOINT ["/sbin/tini", "--", "/bin/sh", "-c", "test -f \"/etc/wireguard/$INTERFACE.conf\" || ( umask 077 && printf '%s\\n' '[Interface]' 'Address = 10.8.0.1/24' 'PostUp = iptables --table nat --append POSTROUTING --jump MASQUERADE' 'PostDown = iptables --table nat --delete POSTROUTING --jump MASQUERADE' 'ListenPort = 51820' \"PrivateKey = $(wg genkey)\" | tee \"/etc/wireguard/$INTERFACE.conf\" ) && test -c /dev/net/tun || { mkdir -p /dev/net && mknod -m 666 /dev/net/tun c 10 200; } && exec wg-quick up \"$INTERFACE\"", "--"]
+ENTRYPOINT ["/sbin/tini", "--", "/bin/sh", "-c", "test -f \"/etc/wireguard/$INTERFACE.conf\" || ( umask 077 && printf '%s\\n' '[Interface]' 'Address = 10.8.0.1/24' 'PostUp = iptables --table nat --append POSTROUTING --jump MASQUERADE' 'PostDown = iptables --table nat --delete POSTROUTING --jump MASQUERADE' 'ListenPort = 51820' \"PrivateKey = $(wg genkey)\" | tee \"/etc/wireguard/$INTERFACE.conf\" ) && test -c /dev/net/tun || { mkdir -p /dev/net && mknod -m 666 /dev/net/tun c 10 200; } && /etc/wireguard/persistentkeepalive.sh & exec wg-quick up \"$INTERFACE\"", "--"]

entrypoint.sh

@@ -4,29 +4,43 @@ set -e
 
 declare -r WG_INTERFACE=${1:-wg0}
 
-function infinite_loop() {
-  # Handle shutdown behavior
-  trap 'shutdown_wg "$1"' SIGTERM SIGINT SIGQUIT
-
-  sleep infinity &
-  wait $!
-}
+KEEPALIVE_PID=""
+WG_PID=""
 
 function shutdown_wg() {
-  echo "Shutting down Wireguard (boringtun)"
+  echo "Shutting down WireGuard (boringtun)"
+  [ -n "$KEEPALIVE_PID" ] && kill "$KEEPALIVE_PID" 2>/dev/null || true
   wg-quick down "$1"
   exit 0
 }
 
 function start_wg() {
-  echo "Starting up Wireguard (boringtun)"
-  wg-quick up "$1"
-  infinite_loop "$1"
+  echo "Starting WireGuard (boringtun)"
+  # Run wg-quick in the background so we retain control of this shell
+  wg-quick up "$1" &
+  WG_PID=$!
+
+  echo "Starting keepalive connection monitor"
+  /etc/wireguard/persistentkeepalive.sh &
+  KEEPALIVE_PID=$!
+
+  # Handle clean shutdown signals
+  trap 'shutdown_wg "$1"' SIGTERM SIGINT SIGQUIT
+
+  # Block until the keepalive monitor exits.
+  # It exits with code 1 only when the connection is stuck — trigger a restart.
+  if ! wait "$KEEPALIVE_PID"; then
+    echo "WireGuard keepalive monitor detected a stuck connection — forcing container restart (exit 1)"
+    [ -n "$WG_PID" ] && kill -9 "$WG_PID" 2>/dev/null || true
+    exit 1
+  fi
+
+  # Keepalive exited cleanly (shouldn't happen); wait for wg process too
+  wait "$WG_PID" || exit $?
 }
 
-#if [[ "$1" =~ ^wg.*$ ]]; then
 if [ -f "/etc/wireguard/${WG_INTERFACE}.conf" ]; then
-  start_wg ${WG_INTERFACE}
+  start_wg "${WG_INTERFACE}"
 else
   exec "$@"
 fi

persistentkeepalive.sh

@@ -1,6 +1,37 @@
 #!/bin/sh
 
-sleep 15;
-PERSISTENT_KEEP_ALIVE=$(cat /etc/wireguard/wg0.conf | grep PersistentKeepalive | awk '{print $3}');
-WG_SERVER_IP="$(cat /etc/wireguard/wg0.conf | grep Address | awk '{print $3}' | cut -d . -f1-3).1";
-ping -s 0 -I wg0 -i $PERSISTENT_KEEP_ALIVE $WG_SERVER_IP
+CONF="/etc/wireguard/${INTERFACE:-wg0}.conf"
+IFACE="${INTERFACE:-wg0}"
+
+# Wait for the WireGuard interface to be fully up
+echo "WireGuard keepalive monitor: waiting 15s for interface $IFACE to come up..."
+sleep 15
+
+PERSISTENT_KEEP_ALIVE=$(grep PersistentKeepalive "$CONF" | awk '{print $3}')
+# Prefer the Endpoint IP; fall back to the Address-derived gateway
+WG_SERVER_IP=$(grep Endpoint "$CONF" | awk '{print $3}' | cut -d: -f1)
+if [ -z "$WG_SERVER_IP" ]; then
+  WG_SERVER_IP="$(grep Address "$CONF" | awk '{print $3}' | cut -d. -f1-3).1"
+fi
+
+PING_INTERVAL=${PERSISTENT_KEEP_ALIVE:-25}
+MAX_FAILURES=3
+fail_count=0
+
+echo "WireGuard keepalive monitor started (target: $WG_SERVER_IP, interval: ${PING_INTERVAL}s, threshold: $MAX_FAILURES)"
+
+while true; do
+  if ping -c 1 -W 5 -I "$IFACE" "$WG_SERVER_IP" > /dev/null 2>&1; then
+    fail_count=0
+  else
+    fail_count=$((fail_count + 1))
+    echo "WireGuard keepalive ping failed ($fail_count/$MAX_FAILURES) to $WG_SERVER_IP via $IFACE"
+    if [ "$fail_count" -ge "$MAX_FAILURES" ]; then
+      echo "WireGuard connection is stuck after $MAX_FAILURES consecutive failures — forcing container restart..."
+      # SIGKILL tini (PID 1) so the container exits with code 137 and Docker restarts it
+      kill -9 1
+      exit 1
+    fi
+  fi
+  sleep "$PING_INTERVAL"
+done