Re: [PATCH 9/9] ptrace: Don't change __state

[Oleg Nesterov]

On 04/27, Oleg Nesterov wrote:
>
> On 04/27, Eric W. Biederman wrote:
> >
> > Oleg Nesterov <oleg@xxxxxxxxxx> writes:
> >
> > > On 04/26, Eric W. Biederman wrote:
> > >>
> > >> @@ -253,7 +252,7 @@ static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
> > >>  	 */
> > >>  	if (lock_task_sighand(child, &flags)) {
> > >>  		if (child->ptrace && child->parent == current) {
> > >> -			WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
> > >> +			WARN_ON(child->jobctl & JOBCTL_DELAY_WAKEKILL);
> > >
> > > This WARN_ON() doesn't look right.
> > >
> > > It is possible that this child was traced by another task and PTRACE_DETACH'ed,
> > > but it didn't clear DELAY_WAKEKILL.
> >
> > That would be a bug.  That would mean that PTRACE_DETACHED process can
> > not be SIGKILL'd.
>
> Why? The tracee will take siglock, clear JOBCTL_DELAY_WAKEKILL and notice
> SIGKILL after that.

Not to mention that the tracee is TASK_RUNNING after PTRACE_DETACH wakes it
up, so the pending JOBCTL_DELAY_WAKEKILL simply has no effect.

Oleg.

> > > If the new debugger attaches and calls ptrace() before the child takes siglock
> > > ptrace_freeze_traced() will fail, but we can hit this WARN_ON().
> > 
> > Eric
> >