[PATCH] Fix intermittent crash in NTVFS server.

[Jeremy Allison]

On Sun, Apr 02, 2017 at 06:23:46AM +1200, Andrew Bartlett via samba-technical wrote:
> On Sat, 2017-04-01 at 14:15 -0400, Jim Brown via samba-technical wrote:
> > You should add "/* NOT REACHED */" after model_ops->terminate() to
> > show 
> > control never returns from this call.
> > 
> > Do any other callers of model_ops->terminate() assume control
> > returns?
> 
> Thanks for taking the time to look over our patches! 
> 
> In this case it is a little more tricky than that, it can return if the
> server is in single process mode (which tries to be a giant state
> machine). 
> 
> /* called when a task goes down */
> static void single_terminate(struct tevent_context *ev, struct
> loadparm_context *lp_ctx, const char *reason)
> {
>         DEBUG(3,("single_terminate: reason[%s]\n",reason));
> }
> 
> Thanks,

Yep, but that doesn't affect this patch (which fixes
the "standard" process mode). For the single process
mode it does:

model_ops->terminate()
talloc_free(srv_conn);

as single_terminate() actually does return.

So changing the order of these calls doesn't
change anything for the "single" process mode.

It's only the standard_terminate() call that does:

talloc_free(event_ctx);
exit(0)

So this patch is good for both the "single" and "standard"
process mode (and it certainly eliminates the reproducible
crash !).

Cheers,

	Jeremy.