Re: [RFC PATCH 0/4] tracing: fix open/delete fixes

[Oleg Nesterov <oleg@redhat.com>]

On 07/17, Masami Hiramatsu wrote:
>
> At a glance, you're trying to change which operation will be
> failed. Currently, user can not remove an event while someone
> opens files which related to the event. And this approach
> changes that the someone can remove the event even if the
> files are opened (and read/write operation will be failed).
> Am I understand correctly?

Yes.

Once again, I am still not sure and I am asking for your review.

But to me this looks much better. To simplify the discussion, lets
consider ftrace_enable_fops in particular.

	- Why should .open() block rmdir or unregister_uprobe_event?

	- Why do we need .open() at all? Whatever it can do to
	  validate file/call/etc, .read/write can do the same.

	- If we kill .open/release, we do not need the nontrivial
	  refcounting. Everything becomes simple, no need to keep
	  the state "in between".

	  We need event_mutex anyway (and note that other f_op's can
	  also rely on other locks taken by trace_remove_event_call),
	  the validation degrades to the trivial != NULL check.

	- This also simplifies trace_remove_event_call() paths, we
	  know that once it takes event_mutex nobody can play with
	  ftrace_event_file/ftrace_event_call we are going to free.

Oleg.