Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 5 Jun 2023 15:48:44 +0200
From:   Oleg Nesterov <oleg@redhat.com>
To:     Mike Christie <michael.christie@oracle.com>
Cc:     linux@leemhuis.info, nicolas.dichtel@6wind.com, axboe@kernel.dk,
        ebiederm@xmission.com, torvalds@linux-foundation.org,
        linux-kernel@vger.kernel.org,
        virtualization@lists.linux-foundation.org, mst@redhat.com,
        sgarzare@redhat.com, jasowang@redhat.com, stefanha@redhat.com,
        brauner@kernel.org
Subject: Re: [PATCH 1/1] fork, vhost: Use CLONE_THREAD to fix freezer/ps
 regression
Message-ID: <20230605134844.GC32275@redhat.com>
References: <20230601183232.8384-1-michael.christie@oracle.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230601183232.8384-1-michael.christie@oracle.com>
User-Agent: Mutt/1.5.24 (2015-08-30)
Precedence: bulk

On 06/01, Mike Christie wrote:
>
> --- a/kernel/signal.c
> +++ b/kernel/signal.c
> @@ -1368,7 +1368,9 @@ int zap_other_threads(struct task_struct *p)
>  
>  	while_each_thread(p, t) {
>  		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
> -		count++;
> +		/* Don't require de_thread to wait for the vhost_worker */
> +		if ((t->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)
> +			count++;

Well if you do this, then you should also change __exit_signal() to
not decrement sig->notify_count. Otherwise de_thread() can succeed
before the "normal" sub-threads exit.

But this can't be right anyway... If nothing else, suppose we have
a process with 3 threads:

	M	- the main thread, group leader
	T	- sub-thread
	V	- vhost worker

T does exec and calls de_thread().

M exits. T takes the leadership and does release_task()

V is still running but V->group_leader points to already freed M.

Or unshare_sighand() after that... If nothing else this means that
lock_task_sighand(T) and lock_task_sighand(V) will take different
locks.

Oleg.