IndexedDB: Prevent terminated worker from hanging connections

Source/modules/indexeddb/IDBDatabase.cpp

 /*
  * Copyright (C) 2010 Google Inc. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  *
  * 1.  Redistributions of source code must retain the above copyright
  *     notice, this list of conditions and the following disclaimer.
  * 2.  Redistributions in binary form must reproduce the above copyright
@@ skipping 313 matching lines @@
 
     if (m_transactions.isEmpty())
         closeConnection();
 }
 
 void IDBDatabase::closeConnection()
 {
     ASSERT(m_closePending);
     ASSERT(m_transactions.isEmpty());
 
-    m_backend->close();
-    m_backend.clear();
+    if (m_backend) {
+        m_backend->close();
+        m_backend.clear();
+    }
 
     if (m_contextStopped || !executionContext())
         return;
 
     EventQueue* eventQueue = executionContext()->eventQueue();
     // Remove any pending versionchange events scheduled to fire on this
     // connection. They would have been scheduled by the backend when another
     // connection called setVersion, but the frontend connection is being
     // closed before they could fire.
     for (size_t i = 0; i < m_enqueuedEvents.size(); ++i) {
@@ skipping 49 matching lines @@
 
 bool IDBDatabase::hasPendingActivity() const
 {
     // The script wrapper must not be collected before the object is closed or
     // we can't fire a "versionchange" event to let script manually close the connection.
     return !m_closePending && hasEventListeners() && !m_contextStopped;
 }
 
 void IDBDatabase::stop()
 {
-    // Stop fires at a deterministic time, so we need to call close in it.
-    close();
+    m_contextStopped = true;
 
-    m_contextStopped = true;
+    // Normally, connections are not closed until all transactions are
+    // finished, which requires a round-trip to the back end. In a worker
+    // context, no more messages will be received once the context is torn
+    // down. So in the case of context stop, just close immediately.

[alecflett, 2014/01/15 23:40:23]

The comment could be generalized here - I think something along the lines of "We
know the context is stopping, so we can proactively close the backend without
waiting for an onabort/onclose event from the backend."

meaning, there's no other "Normally" behavior now, right? since
IDBDatabase::stop is actually closing the backend unconditionally...

[jsbell, 2014/01/15 23:56:56]

By "Normally" I meant "on explicit close() or GC", i.e. not context shutdown.
But I see what you mean, the comment is implying knowledge of the history of the
code. I'll fix.

+    if (m_backend) {
+        m_backend->close();
+        m_backend.clear();
+    }
 }
 
 const AtomicString& IDBDatabase::interfaceName() const
 {
     return EventTargetNames::IDBDatabase;
 }
 
 ExecutionContext* IDBDatabase::executionContext() const
 {
     return ActiveDOMObject::executionContext();
 }
 
 } // namespace WebCore