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

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 6 Jul 2023 14:10:12 +0900
Masami Hiramatsu (Google) <mhiramat@kernel.org> 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.

Thanks,

-- Steve