[PATCH] clean the event context after fork in fork_domain_child()

[Jeremy Allison]

On Tue, Jan 06, 2009 at 01:19:41PM +0800, boyang wrote:
> Jeremy Allison wrote:
> > I know it causes a memory leak, but it can't lead to
> > double free. Remember, Metze's patch removes the
> > event pointers from the linked list, but doesn't
> > free them. So even if some codepath has a saved
> > pointer and calls TALLOC_FREE() on it there won't
> > be a double free, the saved pointer hasn't been
> > freed. That's why it causes a memory leak.
> >
> > Doing it this way is safer than trying to free
> > the event pointers inside the event code, that
> > way can lead to crashes if the saved pointer
> > is referenced.
> >
> > When I've added the code to convert the
> > reinit_after_fork() calls to the individual
> > winbindd_reinit_after_fork() etc., this
> > won't be a problem as part of the 
> > winbindd_reinit_after_fork() call will be
> > code to free all pending events and null out
> > their saved pointers, so the event_context_reinit()
> > should be a no-op (as all the events on the
> > list will already have been freed).
> >   
> v2-event-context-master.diff frees check_online_event. But I think you
> don't need it now. Because all events freeing work will be done in
> winbindd_reinit_after_fork() :-). As you said.
> >   
> >> Looks like ccache_regain_all_now relies on entry->refresh_time to judge
> >> whether the handler is krb5_ticket_refresh_handler or
> >> krb5_ticket_gain_handler. Therefore, we must set entry->refresh_time
> >> before event_add_timed to make ccache_regain_all_now work correctly.
> >> I'll post a separate patch later.
> >>     
> >
> > Please send asap so I can evaluate it as I
> > back-port Metze's code to 3.3-test.
> >   
> v2-krb5-refresh-master.diff addresses this. :-) Pls review it, thanks!

Ok, I've finished doing a lot of cleanup work on the
event code in winbindd. It's all checked into 3.3 and
master. Please review.

If you feel it's working, feel free to back-port to 3.2
and/or 3.0.x and I'll commit patches for you. I'm not
doing that work yet as I'm not sure if we're doing any
more 3.2.x releases or just moving on to 3.3.0.

Jeremy.