Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:2 as permitted sender) client-ip=2620:137:e000::3:2;
Date:   Fri, 3 Nov 2023 19:07:03 -0700
From:   Richard Cochran <richardcochran@gmail.com>
To:     Edward Adam Davis <eadavis@qq.com>
Cc:     jeremy@jcline.org, davem@davemloft.net, habetsm.xilinx@gmail.com,
        linux-kernel@vger.kernel.org, netdev@vger.kernel.org,
        reibax@gmail.com,
        syzbot+df3f3ef31f60781fa911@syzkaller.appspotmail.com
Subject: Re: [PATCH net-next V3] ptp: fix corrupted list in ptp_open
Message-ID: <ZUWnR6T6RZfc6LeF@hoboy.vegasvil.org>
References: <tencent_97D1BA12BBF933129EC01B1D4BB71BE92508@qq.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <tencent_97D1BA12BBF933129EC01B1D4BB71BE92508@qq.com>
Precedence: bulk

On Fri, Nov 03, 2023 at 09:15:03PM +0800, Edward Adam Davis wrote:
> There is no lock protection when writing ptp->tsevqs in ptp_open(),
> ptp_release(), which can cause data corruption, use mutex lock to avoid this
> issue.
> 
> Moreover, ptp_release() should not be used to release the queue in ptp_read(),
> and it should be deleted together.

Oh, now I see what you are fixing...

> @@ -138,14 +143,19 @@ int ptp_open(struct posix_clock_context *pccontext, fmode_t fmode)
>  int ptp_release(struct posix_clock_context *pccontext)
>  {
>  	struct timestamp_event_queue *queue = pccontext->private_clkdata;
> +	struct ptp_clock *ptp =
> +		container_of(pccontext->clk, struct ptp_clock, clock);
>  	unsigned long flags;
>  
>  	if (queue) {
> +		if (mutex_lock_interruptible(&ptp->tsevq_mux)) 
> +			return -ERESTARTSYS;

I don't think it is a good idea to return ERESTARTSYS on signal here.
The release method needs to succeed.

>  		debugfs_remove(queue->debugfs_instance);
>  		pccontext->private_clkdata = NULL;
>  		spin_lock_irqsave(&queue->lock, flags);

This spin lock is wrong.  The spin lock protects the queue, not the
list of queues.

The spin lock/unlock needs to be replaced with mutex lock/unlock.

>  		list_del(&queue->qlist);
>  		spin_unlock_irqrestore(&queue->lock, flags);
> +		mutex_unlock(&ptp->tsevq_mux);
>  		bitmap_free(queue->mask);
>  		kfree(queue);
>  	}

> diff --git a/drivers/ptp/ptp_private.h b/drivers/ptp/ptp_private.h
> index 52f87e394aa6..1525bd2059ba 100644
> --- a/drivers/ptp/ptp_private.h
> +++ b/drivers/ptp/ptp_private.h
> @@ -44,6 +44,7 @@ struct ptp_clock {
>  	struct pps_device *pps_source;
>  	long dialed_frequency; /* remembers the frequency adjustment */
>  	struct list_head tsevqs; /* timestamp fifo list */
> +	struct mutex tsevq_mux; /* one process at a time reading the fifo */

This comment is very misleading.  The mutex does not protect the
fifo.  It protects 'tsevqs' from concurrent access.

Thanks,
Richard