Re: [RFC] fs, proc: Introduce /proc/<pid>/task/<tid>/children entry v6

[Oleg Nesterov <oleg@redhat.com>]

On 01/18, Cyrill Gorcunov wrote:
>
> So Oleg, I think you meant something like below? Comment is moved down an
> list_empty over siblings remans, right?

Yes, except the comment still looks misleading to me.

Otherwise looks correct, but I'll try to re-check once again with
the fresh head. Although I think you should remove me from CC: after
I found the nonexistent bug ;)

> +get_children_pid(struct proc_pid_children_iter *iter, struct pid *pid_prev, loff_t pos)
> +{
> ...
> +	/*
> +	 * Slow search case
> +	 *
> +	 * We might miss some freshly created children
> +	 * here, but it was never promised to be
> +	 * accurate.
> +	 */
> +	list_for_each_entry(task, &start->children, sibling) {
> +		if (pos-- == 0) {
> +			pid = get_pid(task_pid(task));
> +			break;
> +		}
> +	}

This is minor, but "freshly created" looks very confusing to me.
What does it mean? We hold tasklist, we can't race with fork().

Yes we can miss a child, but this has nothing to do with "freshly".
Just suppose that the parent sleeps, but N children exit after we
printed their tids. Now the slow paths skips N extra children, we
miss N tasks.

Oleg.