Re: [PATCH 13/16] ptrace: reorganize __ptrace_unlink() and ptrace_untrace()

[Oleg Nesterov <oleg@redhat.com>]

A bit off-topic note,

On 12/06, Tejun Heo wrote:
>
> -static void ptrace_untrace(struct task_struct *child)
> +void __ptrace_unlink(struct task_struct *child)
>  {
> +	struct signal_struct *sig = child->signal;
> +
> +	BUG_ON(!child->ptrace);
> +
>  	spin_lock(&child->sighand->siglock);
> +
>  	if (task_is_traced(child)) {
>  		/*
>  		 * If group stop is completed or in progress, it should
>  		 * participate in the group stop.  Set GROUP_STOP_PENDING
>  		 * before kicking it.
>  		 */
> -		if (child->signal->flags & SIGNAL_STOP_STOPPED ||
> -		    child->signal->group_stop_count)
> +		if (sig->flags & SIGNAL_STOP_STOPPED || sig->group_stop_count)
>  			child->group_stop |= GROUP_STOP_PENDING;
>  		signal_wake_up(child, 1);

OK. Of course, I do not blame this patch, this mimics the current
behaviour.

But, afaics, this is not exactly right in the long term. Suppose
that SIGNAL_STOP_STOPPED is set but the tracee is running (this can
happen if, say, debugger resumes the tracee and exits). In this case,
I think this thread should be stopped too.

IIRC, I already tried to do this, but the patch (or idea) was nacked
because it means another user-visible change. However, if we want to
really fix things, we should fix this case too. If SIGNAL_STOP_STOPPED
is set, there should be no running threads after detach.

Or. We can change the rules for ptrace_resume(), more on this later.

Oleg.