Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 17 Jul 2019 19:55:57 +0200
From:   Christian Brauner <christian@brauner.io>
To:     "Joel Fernandes (Google)" <joel@joelfernandes.org>
Cc:     linux-kernel@vger.kernel.org,
        Suren Baghdasaryan <surenb@google.com>,
        kernel-team@android.com, Andrea Arcangeli <aarcange@redhat.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        "Eric W. Biederman" <ebiederm@xmission.com>,
        Oleg Nesterov <oleg@redhat.com>, Tejun Heo <tj@kernel.org>,
        jannh@google.com
Subject: Re: [PATCH RFC v1] pidfd: fix a race in setting exit_state for pidfd
 polling
Message-ID: <20190717175556.axe2pne7lcrkmtzr@brauner.io>
References: <20190717172100.261204-1-joel@joelfernandes.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20190717172100.261204-1-joel@joelfernandes.org>
User-Agent: NeoMutt/20180716
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Jul 17, 2019 at 01:21:00PM -0400, Joel Fernandes wrote:
> From: Suren Baghdasaryan <surenb@google.com>
> 
> There is a race between reading task->exit_state in pidfd_poll and writing
> it after do_notify_parent calls do_notify_pidfd. Expected sequence of
> events is:
> 
> CPU 0                            CPU 1
> ------------------------------------------------
> exit_notify
>   do_notify_parent
>     do_notify_pidfd
>   tsk->exit_state = EXIT_DEAD
>                                   pidfd_poll
>                                      if (tsk->exit_state)
> 
> However nothing prevents the following sequence:
> 
> CPU 0                            CPU 1
> ------------------------------------------------
> exit_notify
>   do_notify_parent
>     do_notify_pidfd
>                                    pidfd_poll
>                                       if (tsk->exit_state)
>   tsk->exit_state = EXIT_DEAD
> 
> This causes a polling task to wait forever, since poll blocks because
> exit_state is 0 and the waiting task is not notified again. A stress
> test continuously doing pidfd poll and process exits uncovered this bug,
> and the below patch fixes it.
> 
> To fix this, we set tsk->exit_state before calling do_notify_pidfd.
> 
> Cc: kernel-team@android.com
> Signed-off-by: Suren Baghdasaryan <surenb@google.com>
> Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>

That means in such a situation other users will see EXIT_ZOMBIE where
they didn't see that before until after the parent failed to get
notified.

That's a rather subtle internal change. I was worried about
__ptrace_detach() since it explicitly checks for EXIT_ZOMBIE but it
seems to me that this is fine since we hold write_lock_irq(&tasklist_lock);
at the point when we do set p->exit_signal.

Acked-by: Christian Brauner <christian@brauner.io>

Once Oleg confirms that I'm right not to worty I'll pick this up.

Thanks!
Christian

> 
> ---
>  kernel/exit.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/exit.c b/kernel/exit.c
> index a75b6a7f458a..740ceacb4b76 100644
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -720,6 +720,7 @@ static void exit_notify(struct task_struct *tsk, int group_dead)
>  	if (group_dead)
>  		kill_orphaned_pgrp(tsk->group_leader, NULL);
>  
> +	tsk->exit_state = EXIT_ZOMBIE;
>  	if (unlikely(tsk->ptrace)) {
>  		int sig = thread_group_leader(tsk) &&
>  				thread_group_empty(tsk) &&
> @@ -1156,10 +1157,11 @@ static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
>  		ptrace_unlink(p);
>  
>  		/* If parent wants a zombie, don't release it now */
> -		state = EXIT_ZOMBIE;
> +		p->exit_state = EXIT_ZOMBIE;
>  		if (do_notify_parent(p, p->exit_signal))
> -			state = EXIT_DEAD;
> -		p->exit_state = state;
> +			p->exit_state = EXIT_DEAD;
> +
> +		state = p->exit_state;
>  		write_unlock_irq(&tasklist_lock);
>  	}
>  	if (state == EXIT_DEAD)
> -- 
> 2.22.0.657.g960e92d24f-goog
>