[PATCH] Fix deadlock in an error path

[Havoc Pennington]

Havoc Pennington wrote:
> I think this is just a mistake in naming the function. _unlocked usually 
> refers to the connection lock, but this is the global lock in dbus-bus.c 
> which is different. I don't know a reason that the dbus-bus.c lock would 
> be held already when calling this.
> 

Oh, I see in the redhat.com bug there's a backtrace showing an attempt 
to lock recursively here.

It's harder to fix than your initial patch though. With your patch, you 
would have to *guarantee* that dbus-bus.c global lock is held when doing 
the notify. I don't think that's guaranteed right now.

It might be easier to drop the lock in dbus-bus.c before calling 
dbus-connection.c to close a connection, or something like that.

Havoc