possible write to freed memory in EventTargetNode::dispatchSimulatedMouseEvent()

[Eric Roman]

Not certain about this, but it looks like if the event handler deletes the node within dispatchSimulatedMouseEvent(), there can be a subsequent assignment to freed memory:

EventTargetNode::dispatchSimulatedMouseEvent(...) {
    ...

    m_dispatchingSimulatedEvent = true;

   
    // <--- The event handler may end up deleting "this" --->
    dispatchMouseEvent(...);
   
    // <--- Could this assignment happen to freed memory? --->
    m_dispatchingSimulatedEvent = false;
}

I ran into a problem when using an access key that deleted the anchor (LayoutTests/fast/events/access-key-self-destruct.html) -- however I am not certain if there were other modifications in my code that could have caused this problem.

Can someone familiar with the method comment if this is a problem?

(I see that EventTargetNode::dispatchMouseEvent() protects "this" from deletion, why does this block not require it too?)

[Mark Rowe (bdash)]

It should be possible to construct a test case which crashes due to this issue.  It may require GuardMalloc to be enabled to reproducibly crash though.

[Mark Rowe (bdash)]

<rdar://problem/5808517>

[Adele Peterson]

I've been trying to create a test case where the node gets destroyed that early, and I've been unsuccessful.  I think its because the Event keeps a RefPtr to the EventTarget...and dispatchSimulatedClick has a RefPtr for the event, so the event and the target node can't get destroyed until after dispatchSimulatedClick exits.  Someone should check my logic on that though...

[Eric Seidel (no email)]

The m_dispatchingSimulatedMouseEvent bool is gone from TOT.  This bug can be closed.

[David Kilzer (:ddkilzer)]

(In reply to comment #4)
> The m_dispatchingSimulatedMouseEvent bool is gone from TOT.  This bug can be
> closed.

The m_dispatchingSimulatedEvent bool was removed in r31767.

<http://trac.webkit.org/changeset/31767>