android: a retry the device slept through is resumed by the watchdog alarm: the countdown runs in uptime and stood still in deep sleep, so a tunnel that dropped overnight came back only when the phone was picked up; while a retry is pending the wakeup alarm stays at 10 min instead of backing off to 1 h

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

src/frontends/android/app/src/main/java/org/strongswan/android/logic/VpnStateService.java

@@ -90,6 +90,9 @@ public class VpnStateService extends Service
 	private long mRetryIn;
 	/* true while the countdown is parked because the ISP whitelist blocks the server */
 	private boolean mRetryFrozen;
+	/* when the pending retry is due in elapsedRealtime: the countdown itself
+	 * runs in uptime and stands still while the device sleeps */
+	private long mRetryDue;
 	/* timestamp of the last network-change-triggered immediate retry */
 	private long mLastNetworkRetry;
 	/* double-start guard: profile and issue time of the last connect() that
@@ -610,6 +613,37 @@ public class VpnStateService extends Service
 		});
 	}
 
+	/** Whether a retry countdown is pending (an error with an automatic retry). */
+	public boolean isRetryPending()
+	{
+		return mRetryTimeout > 0;
+	}
+
+	/**
+	 * Runs the pending retry now if its time has passed on the wall clock.
+	 * The countdown ticks in uptime, so it stands still while the device
+	 * sleeps; the watchdog's wakeup alarm calls this so a tunnel that dropped
+	 * overnight comes back without the user picking up the phone.  Never
+	 * shortens a backoff.  Main thread only.
+	 *
+	 * @return true if a reconnect was started
+	 */
+	public boolean retryIfOverdue()
+	{
+		if (mRetryTimeout <= 0 || mProfile == null || SystemClock.elapsedRealtime() < mRetryDue ||
+			WhiteListManager.getInstance(this).shouldBlockReconnect())
+		{
+			return false;
+		}
+		DiagLog.i(TAG, "retry overdue (device slept through the countdown), reconnecting");
+		mRetryTimeout = 0;
+		mRetryIn = 0;
+		mRetryFrozen = false;
+		mHandler.removeMessages(RETRY_MSG);
+		connect(null, false);
+		return true;
+	}
+
 	/**
 	 * Update state and notify all listeners about the change. By using a Handler
 	 * this is done from the main UI thread and not the initial reporter thread.
@@ -826,6 +860,7 @@ public class VpnStateService extends Service
 	{
 		mRetryTimeout = mRetryIn = mTimeoutProvider.getTimeout(error);
 		mRetryFrozen = false;
+		mRetryDue = SystemClock.elapsedRealtime() + mRetryTimeout;
 		DiagLog.i(TAG, "error " + error + ", retry in " + (mRetryTimeout / 1000) + "s");
 		/* never run two countdown chains in parallel (double-speed countdown,
 		 * double notification updates and two reconnects at zero) */

src/frontends/android/app/src/main/java/org/strongswan/android/logic/WatchdogReceiver.java

@@ -154,15 +154,16 @@ public class WatchdogReceiver extends BroadcastReceiver
 
 	/**
 	 * Whether a check counts towards the steady-state backoff of a live
-	 * process: connected, retrying/connecting, or held back by the ISP
+	 * process: connected, connecting, or held back by the ISP whitelist.  In
-	 * whitelist.  In all of these the in-process machinery owns recovery and
+	 * all of these the in-process machinery owns recovery and the alarm can
-	 * the alarm can only guard against the process dying - firing every
+	 * only guard against the process dying - firing every minute (and waking
-	 * minute (and waking Doze every 10 minutes) for hours of failed retries
+	 * Doze every 10 minutes) for hours of whitelist blocking only burns
-	 * or whitelist blocking only burns battery.
+	 * battery.
 	 */
-	static boolean isAliveSteady(Action action, boolean processAlive, boolean parked)
+	static boolean isAliveSteady(Action action, boolean processAlive, boolean parked, boolean retryPending)
-	{
+	{	/* a pending retry needs the fast wakeup alarm: its countdown stands
-		return action == Action.NONE && processAlive && !parked;
+		 * still while the device sleeps and only this alarm resumes it */
+		return action == Action.NONE && processAlive && !parked && !retryPending;
 	}
 
 	@Override
@@ -190,7 +191,13 @@ public class WatchdogReceiver extends BroadcastReceiver
 		Action action = decideAction(enabled, uuid != null, autoExpected, whitelistBlocks, serviceTracking,
 									 processAlive, wifiPresent, autoProfileConfigured);
 		boolean parked = uuid == null && autoExpected && processAlive && !serviceTracking && action == Action.NONE;
-		boolean alive = isAliveSteady(action, processAlive, parked);
+		/* the whitelist parks retries on purpose, that state may back off */
+		boolean retryPending = serviceTracking && !whitelistBlocks && service.isRetryPending();
+		if (retryPending && service.retryIfOverdue())
+		{
+			DiagLog.i(TAG, "resumed a retry the device slept through");
+		}
+		boolean alive = isAliveSteady(action, processAlive, parked, retryPending);
 		if (parked)
 		{
 			sParkedChecks++;

src/frontends/android/app/src/test/java/org/strongswan/android/logic/WatchdogReceiverTest.kt

@@ -132,21 +132,23 @@ class WatchdogReceiverTest
 		/* retrying/connecting: the service tracks the connection */
 		val retrying = decide(expectedProfileSet = true, serviceTracking = true, processAlive = true)
 		assertThat(retrying).isEqualTo(Action.NONE)
-		assertThat(WatchdogReceiver.isAliveSteady(retrying, true, false)).isTrue()
+		assertThat(WatchdogReceiver.isAliveSteady(retrying, true, false, false)).isTrue()
+		/* a pending retry stands still while the device sleeps: keep the fast wakeup alarm */
+		assertThat(WatchdogReceiver.isAliveSteady(retrying, true, false, true)).isFalse()
 		/* whitelist blocks the server while the process is alive */
 		val blocked = decide(expectedProfileSet = true, whitelistBlocks = true, processAlive = true)
-		assertThat(WatchdogReceiver.isAliveSteady(blocked, true, false)).isTrue()
+		assertThat(WatchdogReceiver.isAliveSteady(blocked, true, false, false)).isTrue()
 	}
 
 	@Test
 	fun deadProcessOrRestartNeverBacksOff()
 	{
 		val restart = decide(expectedProfileSet = true, processAlive = false)
-		assertThat(WatchdogReceiver.isAliveSteady(restart, false, false)).isFalse()
+		assertThat(WatchdogReceiver.isAliveSteady(restart, false, false, false)).isFalse()
 		/* whitelist blocks, but no live process: keep the fast cadence */
 		val blocked = decide(expectedProfileSet = true, whitelistBlocks = true, processAlive = false)
-		assertThat(WatchdogReceiver.isAliveSteady(blocked, false, false)).isFalse()
+		assertThat(WatchdogReceiver.isAliveSteady(blocked, false, false, false)).isFalse()
 		/* parked sessions use their own backoff */
-		assertThat(WatchdogReceiver.isAliveSteady(Action.NONE, true, true)).isFalse()
+		assertThat(WatchdogReceiver.isAliveSteady(Action.NONE, true, true, false)).isFalse()
 	}
 }