Re: [PATCH net-next V6] ptp: fix corrupted list in ptp_open

[Richard Cochran <richardcochran@gmail.com>]

Edward!

On Sun, Nov 05, 2023 at 10:12:08AM +0800, Edward Adam Davis wrote:
> There is no lock protection when writing ptp->tsevqs in ptp_open() and
> ptp_release(), which can cause data corruption, use spin 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.

Change to:  "it should be deleted altogether"
 
> Reported-and-tested-by: syzbot+df3f3ef31f60781fa911@syzkaller.appspotmail.com
> Fixes: 8f5de6fb2453 ("ptp: support multiple timestamp event readers")
> Signed-off-by: Edward Adam Davis <eadavis@qq.com>
> ---
>  drivers/ptp/ptp_chardev.c | 11 +++++++----
>  drivers/ptp/ptp_clock.c   |  1 +
>  drivers/ptp/ptp_private.h |  1 +
>  3 files changed, 9 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/ptp/ptp_chardev.c b/drivers/ptp/ptp_chardev.c
> index 282cd7d24077..31594f40a21e 100644
> --- a/drivers/ptp/ptp_chardev.c
> +++ b/drivers/ptp/ptp_chardev.c
> @@ -108,6 +108,7 @@ int ptp_open(struct posix_clock_context *pccontext, fmode_t fmode)
>  		container_of(pccontext->clk, struct ptp_clock, clock);
>  	struct timestamp_event_queue *queue;
>  	char debugfsname[32];
> +	unsigned long flags;
>  
>  	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
>  	if (!queue)
> @@ -119,8 +120,10 @@ int ptp_open(struct posix_clock_context *pccontext, fmode_t fmode)
>  	}
>  	bitmap_set(queue->mask, 0, PTP_MAX_CHANNELS);
>  	spin_lock_init(&queue->lock);
> +	spin_lock_irqsave(&ptp->tsevqs_lock, flags);
>  	list_add_tail(&queue->qlist, &ptp->tsevqs);
>  	pccontext->private_clkdata = queue;

Move this assignment outside of locked region, i.e. after spin_unlock_irqrestore().

> +	spin_unlock_irqrestore(&ptp->tsevqs_lock, flags);
>  
>  	/* Debugfs contents */
>  	sprintf(debugfsname, "0x%p", queue);
> @@ -139,13 +142,15 @@ int ptp_release(struct posix_clock_context *pccontext)
>  {
>  	struct timestamp_event_queue *queue = pccontext->private_clkdata;
>  	unsigned long flags;
> +	struct ptp_clock *ptp =
> +		container_of(pccontext->clk, struct ptp_clock, clock);
>  
>  	if (queue) {

Please remove this test.  Since you removed ptp_release() from
ptp_read(), the queue cannot be NULL.

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

Move this assignment outside of locked region.

> -		spin_lock_irqsave(&queue->lock, flags);
>  		list_del(&queue->qlist);
> -		spin_unlock_irqrestore(&queue->lock, flags);
> +		spin_unlock_irqrestore(&ptp->tsevqs_lock, flags);
>  		bitmap_free(queue->mask);
>  		kfree(queue);
>  	}
> @@ -585,7 +590,5 @@ ssize_t ptp_read(struct posix_clock_context *pccontext, uint rdflags,
>  free_event:
>  	kfree(event);
>  exit:
> -	if (result < 0)
> -		ptp_release(pccontext);

This is good, but please put it into a separate patch, along with the
removal of the bogus "if (queue)" test in ptp_release().

>  	return result;
>  }
> diff --git a/drivers/ptp/ptp_clock.c b/drivers/ptp/ptp_clock.c
> index 3d1b0a97301c..ea82648ad557 100644
> --- a/drivers/ptp/ptp_clock.c
> +++ b/drivers/ptp/ptp_clock.c
> @@ -247,6 +247,7 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,
>  	if (!queue)
>  		goto no_memory_queue;
>  	list_add_tail(&queue->qlist, &ptp->tsevqs);
> +	spin_lock_init(&ptp->tsevqs_lock);
>  	queue->mask = bitmap_alloc(PTP_MAX_CHANNELS, GFP_KERNEL);
>  	if (!queue->mask)
>  		goto no_memory_bitmap;
> diff --git a/drivers/ptp/ptp_private.h b/drivers/ptp/ptp_private.h
> index 52f87e394aa6..63af246f17eb 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 */
> +	spinlock_t tsevqs_lock; /* one process at a time writing the timestamp fifo list*/

Please change this comment to "protects tsevqs from concurrent access"

>  	struct mutex pincfg_mux; /* protect concurrent info->pin_config access */
>  	wait_queue_head_t tsev_wq;
>  	int defunct; /* tells readers to go away when clock is being removed */
> -- 
> 2.25.1
> 

Since v6.6 now contains the original commit, please change subject to [net].

Thanks,
Richard