Fix race between stopping and starting the receive thread (#2142)

* Fix race between stopping and starting the receive thread

When the last receive handler was removed, receiveThread was cleared
before the thread had stopped. A handler added in that window, e.g. by
a client reconnecting to the daemon in on-connection receive mode,
started a second receive thread. While the stop request for the old
thread was still set, each new thread ended immediately and started the
next one. When the old thread ended, it cleared the reference to the
newer thread and started yet another one, leaving two receive loops on
the same websocket. The unreferenced loop was never stopped and kept
receiving and acknowledging messages after all clients had gone.

Keep receiveThread set until the receive thread has finished and only
let the thread itself clear it. Its end block already starts a new
receive thread if handlers were added in the meantime. Clear a stop
request right before a new receive loop is started instead of when a
loop ends, so it can neither be cleared by another loop nor stop a
later one.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Add test for adding a receive handler while the receive thread stops

The receive loop is replaced by a fake one that can be held while it
stops, so the test doesn't connect to the server.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: iddo <917594+iddo@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
iddo 2026-10-09 17:28:42 +03:00 committed by GitHub
parent 2a9a4b1f3c
commit 6f2ea80060
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 281 additions and 8 deletions

View File

@ -42,8 +42,8 @@ public class ReceiveHelper {
private ReceiveConfig receiveConfig = new ReceiveConfig(false, false, false, false, false);
private boolean hasCaughtUpWithOldMessages = false;
private boolean isWaitingForMessage = false;
private boolean shouldStop = false;
private volatile boolean isWaitingForMessage = false;
private volatile boolean shouldStop = false;
private Callable authenticationFailureListener;
private Callable caughtUpWithOldMessagesListener;
@ -71,6 +71,15 @@ public class ReceiveHelper {
return isWaitingForMessage;
}
/**
* Must only be called right before a new receive loop is started, while no other receive loop is running.
* Ending a receive loop doesn't clear the stop request, so a loop that was asked to stop can't continue
* receiving, e.g. by retrying in receiveMessagesContinuously after an IOException.
*/
public void clearStopRequest() {
this.shouldStop = false;
}
public void receiveMessagesContinuously(Manager.ReceiveMessageHandler handler) {
while (!shouldStop) {
try {
@ -111,7 +120,6 @@ public class ReceiveHelper {
signalWebSocket.removeKeepAliveToken("receive");
signalWebSocket.disconnect();
webSocketStateDisposable.dispose();
shouldStop = false;
}
}

View File

@ -1510,15 +1510,18 @@ public class ManagerImpl implements Manager {
private static final AtomicInteger threadNumber = new AtomicInteger(0);
private void startReceiveThreadIfRequired() {
if (receiveThread != null || isReceivingSynchronous) {
if (isReceiveThreadAlive() || isReceivingSynchronous) {
return;
}
context.getReceiveHelper().clearStopRequest();
receiveThread = Thread.ofPlatform().name("receive-" + threadNumber.getAndIncrement()).start(() -> {
logger.debug("Starting receiving messages");
context.getReceiveHelper().receiveMessagesContinuously(this::passReceivedMessageToHandlers);
logger.debug("Finished receiving messages");
synchronized (messageHandlers) {
receiveThread = null;
if (receiveThread == Thread.currentThread()) {
receiveThread = null;
}
// Check if in the meantime another handler has been registered
if (!messageHandlers.isEmpty()) {
@ -1529,6 +1532,13 @@ public class ManagerImpl implements Manager {
});
}
private boolean isReceiveThreadAlive() {
// The receive thread clears receiveThread itself when it has finished, so it stays set while the thread is
// still stopping and no second receive thread can be started in the meantime.
// A receive thread that died from an uncaught exception never clears it.
return receiveThread != null && receiveThread.isAlive();
}
private void passReceivedMessageToHandlers(MessageEnvelope envelope, Throwable e) {
synchronized (messageHandlers) {
Stream.concat(messageHandlers.stream(), weakHandlers.stream()).forEach(h -> {
@ -1551,7 +1561,6 @@ public class ManagerImpl implements Manager {
return;
}
thread = receiveThread;
receiveThread = null;
}
stopReceiveThread(thread);
@ -1607,11 +1616,12 @@ public class ManagerImpl implements Manager {
ReceiveMessageHandler handler
) throws IOException, AlreadyReceivingException {
synchronized (messageHandlers) {
if (isReceiving()) {
if (isReceiving() || isReceiveThreadAlive()) {
throw new AlreadyReceivingException("Already receiving message.");
}
isReceivingSynchronous = true;
receiveThread = Thread.currentThread();
context.getReceiveHelper().clearStopRequest();
}
try {
context.getReceiveHelper().receiveMessages(timeout, maxMessages, (envelope, e) -> {
@ -2036,7 +2046,6 @@ public class ManagerImpl implements Manager {
weakHandlers.clear();
messageHandlers.clear();
thread = receiveThread;
receiveThread = null;
}
if (thread != null) {
stopReceiveThread(thread);

View File

@ -0,0 +1,256 @@
package org.asamk.signal.manager.internal;
import org.asamk.signal.manager.Manager;
import org.asamk.signal.manager.Settings;
import org.asamk.signal.manager.api.MessageEnvelope;
import org.asamk.signal.manager.api.ServiceEnvironment;
import org.asamk.signal.manager.config.ServiceConfig;
import org.asamk.signal.manager.helper.AccountFileUpdater;
import org.asamk.signal.manager.helper.Context;
import org.asamk.signal.manager.helper.ReceiveHelper;
import org.asamk.signal.manager.storage.SignalAccount;
import org.asamk.signal.manager.util.KeyUtils;
import org.bouncycastle.jce.provider.BouncyCastleProvider;
import org.junit.jupiter.api.AfterEach;
import org.junit.jupiter.api.BeforeEach;
import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.Timeout;
import org.junit.jupiter.api.io.TempDir;
import org.signal.core.models.ServiceId.ACI;
import org.whispersystems.signalservice.api.push.ServiceIdType;
import java.nio.file.Path;
import java.security.Security;
import java.time.Duration;
import java.util.Optional;
import java.util.function.BooleanSupplier;
import static org.junit.jupiter.api.Assertions.assertTrue;
@Timeout(60)
class ReceiveThreadTest {
private static final Duration TIMEOUT = Duration.ofSeconds(10);
@TempDir
Path directory;
private ManagerImpl manager;
private FakeReceiveHelper receiveHelper;
@BeforeEach
void setUp() throws Exception {
// ManagerImpl.close() creates the push service socket, whose trust store needs BouncyCastle
if (Security.getProvider(BouncyCastleProvider.PROVIDER_NAME) == null) {
Security.addProvider(new BouncyCastleProvider());
}
final var account = SignalAccount.createLinkedAccount(directory.toFile(),
"account",
ServiceEnvironment.STAGING,
Settings.DEFAULT);
account.setProvisioningData(null,
ACI.parseOrThrow("11111111-1111-4111-8111-111111111111"),
null,
"test-password",
new byte[]{1},
KeyUtils.generateIdentityKeyPair(),
null,
KeyUtils.createProfileKey(),
null,
new byte[32],
null);
account.finishLinking(2, KeyUtils.generatePreKeysForType(account.getAccountData(ServiceIdType.ACI)), null);
manager = new ManagerImpl(account,
PathConfig.createDefault(directory.toFile()),
new AccountFileUpdater() {
@Override
public void updateAccountIdentifiers(final String number, final ACI aci) {
}
@Override
public void removeAccount() {
}
},
ServiceConfig.getServiceEnvironmentConfig(ServiceEnvironment.STAGING, "signal-cli-test"),
"signal-cli-test");
// Replace the receive helper, so the receive thread doesn't connect to the server
final var contextField = ManagerImpl.class.getDeclaredField("context");
contextField.setAccessible(true);
final var context = (Context) contextField.get(manager);
receiveHelper = new FakeReceiveHelper(context);
final var receiveHelperField = Context.class.getDeclaredField("receiveHelper");
receiveHelperField.setAccessible(true);
receiveHelperField.set(context, receiveHelper);
}
@AfterEach
void tearDown() {
receiveHelper.setHoldStoppingLoops(false);
manager.close();
}
@Test
void handlerAddedWhileReceiveThreadIsStoppingDoesNotStartSecondReceiveThread() throws Exception {
final var first = newHandler();
final var second = newHandler();
manager.addReceiveHandler(first);
assertTrue(receiveHelper.await(() -> receiveHelper.runningLoops == 1));
receiveHelper.setHoldStoppingLoops(true);
final var remover = Thread.ofPlatform().start(() -> manager.removeReceiveHandler(first));
assertTrue(receiveHelper.await(() -> receiveHelper.stoppingLoops == 1));
// Like a client that reconnects while the receive thread for its previous connection is still stopping
manager.addReceiveHandler(second);
assertTrue(receiveHelper.await(() -> receiveHelper.startedThreads == 1));
receiveHelper.setHoldStoppingLoops(false);
assertTrue(remover.join(TIMEOUT));
assertTrue(receiveHelper.await(() -> receiveHelper.startedLoops == 2 && receiveHelper.runningLoops == 1));
assertTrue(receiveHelper.await(() -> receiveHelper.startedThreads == 2 && receiveHelper.maxRunningLoops == 1));
manager.removeReceiveHandler(second);
assertTrue(receiveHelper.await(() -> receiveHelper.runningLoops == 0));
}
@Test
void stopRequestedForStoppingReceiveThreadDoesNotStopNextReceiveThread() throws Exception {
final var first = newHandler();
final var second = newHandler();
final var third = newHandler();
manager.addReceiveHandler(first);
assertTrue(receiveHelper.await(() -> receiveHelper.runningLoops == 1));
receiveHelper.setHoldStoppingLoops(true);
final var firstRemover = Thread.ofPlatform().start(() -> manager.removeReceiveHandler(first));
assertTrue(receiveHelper.await(() -> receiveHelper.stoppingLoops == 1));
manager.addReceiveHandler(second);
final var secondRemover = Thread.ofPlatform().start(() -> manager.removeReceiveHandler(second));
assertTrue(receiveHelper.await(() -> receiveHelper.stopRequests == 2));
receiveHelper.setHoldStoppingLoops(false);
assertTrue(firstRemover.join(TIMEOUT));
assertTrue(secondRemover.join(TIMEOUT));
assertTrue(receiveHelper.await(() -> receiveHelper.runningLoops == 0 && receiveHelper.startedThreads == 1));
manager.addReceiveHandler(third);
assertTrue(receiveHelper.await(() -> receiveHelper.startedLoops == 2 && receiveHelper.runningLoops == 1));
assertTrue(receiveHelper.await(() -> receiveHelper.startedThreads == 2 && receiveHelper.maxRunningLoops == 1));
}
private static Manager.ReceiveMessageHandler newHandler() {
// An anonymous class instead of a lambda, so every handler is a separate instance
return new Manager.ReceiveMessageHandler() {
@Override
public void handleMessage(final MessageEnvelope envelope, final Throwable e) {
}
};
}
/**
* Instead of receiving messages from the server, a receive loop waits until it has been asked to stop.
* While holdStoppingLoops is set, a loop that has been asked to stop doesn't return yet, like a loop that
* is still busy when the last handler is removed.
*/
private static class FakeReceiveHelper extends ReceiveHelper {
private final Object lock = new Object();
private boolean stopRequested;
private boolean holdStoppingLoops;
private int stopRequests;
private int startedThreads;
private int startedLoops;
private int runningLoops;
private int maxRunningLoops;
private int stoppingLoops;
FakeReceiveHelper(final Context context) {
super(context);
}
@Override
public boolean requestStopReceiveMessages() {
super.requestStopReceiveMessages();
synchronized (lock) {
stopRequested = true;
stopRequests++;
lock.notifyAll();
}
// The fake loop isn't blocked in a read that needs to be interrupted
return false;
}
@Override
public void clearStopRequest() {
super.clearStopRequest();
synchronized (lock) {
stopRequested = false;
}
}
@Override
public void receiveMessagesContinuously(final Manager.ReceiveMessageHandler handler) {
synchronized (lock) {
startedThreads++;
lock.notifyAll();
}
super.receiveMessagesContinuously(handler);
}
@Override
public void receiveMessages(
final Optional<Duration> timeout,
final Integer maxMessages,
final Manager.ReceiveMessageHandler handler
) {
synchronized (lock) {
startedLoops++;
runningLoops++;
maxRunningLoops = Math.max(maxRunningLoops, runningLoops);
lock.notifyAll();
if (await(() -> stopRequested)) {
stoppingLoops++;
lock.notifyAll();
await(() -> !holdStoppingLoops);
stoppingLoops--;
}
runningLoops--;
lock.notifyAll();
}
}
void setHoldStoppingLoops(final boolean hold) {
synchronized (lock) {
holdStoppingLoops = hold;
lock.notifyAll();
}
}
/**
* Waits until the condition, which is evaluated while holding the lock, is true.
* Returns false if it didn't become true within the timeout.
*/
boolean await(final BooleanSupplier condition) {
synchronized (lock) {
final var deadline = System.nanoTime() + TIMEOUT.toNanos();
while (!condition.getAsBoolean()) {
final var remainingMillis = (deadline - System.nanoTime()) / 1_000_000;
if (remainingMillis <= 0) {
return false;
}
try {
lock.wait(remainingMillis);
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
return false;
}
}
return true;
}
}
}
}