Re: [PATCH] fprobe: Ensure running fprobe_exit_handler() finished before calling rethook_free()

[Google]

On Thu, 6 Jul 2023 09:56:24 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Thu, 6 Jul 2023 14:10:12 +0900
> Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx> wrote:
> 
> > With only Jiri's patch, following flow can happen;
> > 
> > ------
> >  CPU1                              CPU2
> >  call unregister_fprobe()
> >  ...
> >                                    __fprobe_handler()
> >                                    rethook_hook() on probed function
> >  unregister_ftrace_function()
> >                                    return from probed function
> >                                    rethook hooks
> >                                    find rh->handler == fprobe_exit_handler
> >                                    call fprobe_exit_handler()
> >  rethook_free():
> >    set rh->handler = NULL;
> >  return from unreigster_fprobe;
> >                                    call fp->exit_handler() <- (*)
> > 
> > (*) In this point, the exit handler is called after returning from 
> > unregister_fprobe().
> > ------
> > 
> > So, this patch changes it as following;
> > ------
> >  CPU1                              CPU2
> >  call unregister_fprobe()
> >  ...
> >  rethook_stop():
> >    set rh->handler = NULL;
> >                                    __fprobe_handler()
> >                                    rethook_hook() on probed function
> >  unregister_ftrace_function()
> >                                    return from probed function
> >                                    rethook hooks
> >                                    find rh->handler == NULL
> >                                    return from rethook
> >  rethook_free()
> >  return from unreigster_fprobe;
> > ------
> > 
> > I can also just put a synchronize_sched_rcu() right after rethook_free()
> > to wait for all running fprobe_exit_handler() too.
> > 
> 
> This makes more sense. Can you please add the above to the change log.

OK, let me update it.

Thanks!

> 
> Thanks,
> 
> -- Steve


-- 
Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>