Re: [PATCH] watchdog: core: assign parent with devm registration

[Guenter Roeck <linux@roeck-us.net>]

On 5/23/26 19:23, Rosen Penev wrote:
> If the user did not pass a parent in the struct watchdog_device
> then use the device used for devres as parent.
> 
> This is quite intuitive and can help avoiding having to
> assign parent explicitly in every driver using devres
> to add the watchdog_device.
> 
> Signed-off-by: Rosen Penev <rosenp@gmail.com>
> ---
>   Linus: hope you don't mind me flat out copying the description from
>   https://lore.kernel.org/all/20260427-gpio-mmio-more-v3-1-fe1882351424@kernel.org/
>   drivers/watchdog/watchdog_core.c | 7 +++++++
>   1 file changed, 7 insertions(+)
> 
> diff --git a/drivers/watchdog/watchdog_core.c b/drivers/watchdog/watchdog_core.c
> index 8300520688d0..726c85debabc 100644
> --- a/drivers/watchdog/watchdog_core.c
> +++ b/drivers/watchdog/watchdog_core.c
> @@ -440,6 +440,13 @@ int devm_watchdog_register_device(struct device *dev,
>   	if (!rcwdd)
>   		return -ENOMEM;
>   
> +	/*
> +	 * We are passing the devres device here so if the user did not pass
> +	 * another parent, it's this one.
> +	 */
> +	if (!wdd->parent)
> +		wdd->parent = dev;
> +

I understand that you consider it to be garbage, but in my opinion Sashiko has a
point. If the watchdog device data structure is static, wdd->parent will be retained
over multiple remove/reinsert instantiations of the driver, which in turn would result
in subsequent UAF. That means I can not accept your patch.

Guenter