Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 25 May 2020 21:21:43 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Jens Axboe <axboe@kernel.dk>
Cc:     Stefano Garzarella <sgarzare@redhat.com>, io-uring@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        Ingo Molnar <mingo@redhat.com>
Subject: Re: io_uring: BUG: kernel NULL pointer dereference
Message-ID: <20200525192143.GG317569@hirez.programming.kicks-ass.net>
References: <20200525103051.lztpbl33hsgv6grz@steredhat>
 <20200525134552.5dyldwmeks3t6vj6@steredhat>
 <b1689238-b236-cc93-9909-c09120e7975c@kernel.dk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <b1689238-b236-cc93-9909-c09120e7975c@kernel.dk>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, May 25, 2020 at 08:10:27AM -0600, Jens Axboe wrote:
> I think the odd part here is that task_tick_numa() checks for a
> valid mm, and queues work if the task has it. But for the sqpoll
> kthread, the mm can come and go. By the time the task work is run,
> the mm is gone and we oops on current->mm == NULL.
> 
> I think the below should fix it:
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 538ba5d94e99..24a8557f001f 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -2908,7 +2908,8 @@ static void task_tick_numa(struct rq *rq, struct task_struct *curr)
>  	/*
>  	 * We don't care about NUMA placement if we don't have memory.
>  	 */
> -	if (!curr->mm || (curr->flags & PF_EXITING) || work->next != work)
> +	if (!curr->mm || (curr->flags & (PF_EXITING | PF_KTHREAD)) ||
> +	    work->next != work)
>  		return;

Ah, I think that's one more instance of '!p->mm' != is_kthread(). A
while ago someone went and cleaned a bunch of them up. Clearly this one
was missed.

I'm thinking just:

	if ((curr->flags & (PF_EXITING | PF_KTHREAD)) || work->next != work)

should be enough.