[PATCH 2/3] mailbox: Avoid NULL dereference in mbox_chan_received_data

[Andre Przywara]

Hi,

On 28/02/18 02:27, Samuel Holland wrote:
> If a reception IRQ is pending when a mailbox channel is shut down (for
> example, if the controller uses threaded interrupts), it is possible for
> mbox_chan_received_data to be called while chan->cl is NULL.
> 
> This was found while developing a mailbox controller driver for use with
> SCPI. The SCPI protocol driver frees its mailbox channel during probing
> if the SCP firmware does not respond within a specified timeout. In this
> case, if the SCP firmware takes slightly too long to respond,
> mbox_chan_received_data races with mbox_free_channel clearing chan->cl.
> 
> Signed-off-by: Samuel Holland <samuel at sholland.org>
> ---
>  drivers/mailbox/mailbox.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/mailbox/mailbox.c b/drivers/mailbox/mailbox.c
> index 674b35f402f5..a0258d8672d5 100644
> --- a/drivers/mailbox/mailbox.c
> +++ b/drivers/mailbox/mailbox.c
> @@ -152,9 +152,11 @@ static enum hrtimer_restart txdone_hrtimer(struct hrtimer *hrtimer)
>   */
>  void mbox_chan_received_data(struct mbox_chan *chan, void *mssg)
>  {
> +	struct mbox_client *cl = READ_ONCE(chan->cl);
> +
>  	/* No buffering the received data */
> -	if (chan->cl->rx_callback)
> -		chan->cl->rx_callback(chan->cl, mssg);
> +	if (cl && cl->rx_callback)
> +		cl->rx_callback(cl, mssg);

I don't think this is the proper fix. This sounds like we should have a
lock here. If mbox_free_channel now frees or clears chan->cl between the
READ_ONCE and the second part of the "&&", we will have a
use-after-free. If it's being cleared after the comparison, we will end
up with a NULL pointer dereference again, IIUC.

Or am I missing something?

Cheers,
Andre.

>  }
>  EXPORT_SYMBOL_GPL(mbox_chan_received_data);
>  
>