Re: [PATCH v9 5/5] HID: wacom: Redesign shared sibling data lifecycle

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Use-After-Free in wacom_mode_change_work due to unsafe RCU access and lack of sibling synchronization.
--

commit 144702a97dac139acf4377ec2bc37fa17f37b11a
Author: Lee Jones <lee@kernel.org>

HID: wacom: Redesign shared sibling data lifecycle

The Wacom driver coordinates state between sibling interfaces of
the same physical device using a shared structure 'wacom_shared'.
This redesigns the sibling data lifecycle by eliminating a volatile
representative pointer and storing stable static copies of the required
attributes for sibling matching. It also switches to kref_put_mutex()
to secure the lifecycle against concurrent probe/disconnect races.

> diff --git a/drivers/hid/wacom_sys.c b/drivers/hid/wacom_sys.c
> index 12f6a32cbc5fc..9b991078a36d8 100644
> --- a/drivers/hid/wacom_sys.c
> +++ b/drivers/hid/wacom_sys.c
[ ... ]
> @@ -883,17 +900,19 @@ static void wacom_remove_shared_data(void *res)
>  				rcu_dereference_protected(wacom_wac->shared->pen,
>  							  lockdep_is_held(&wacom_udev_list_lock));
>  
> -			if (touch == wacom->hdev) {
> +			if (touch == res_wacom->hdev) {
>  				rcu_assign_pointer(wacom_wac->shared->touch, NULL);
>  				rcu_assign_pointer(wacom_wac->shared->touch_input, NULL);
> -			} else if (pen == wacom->hdev) {
> +			} else if (pen == res_wacom->hdev) {
>  				rcu_assign_pointer(wacom_wac->shared->pen, NULL);
>  			}
>  		}
>  
>  		synchronize_rcu();

[Severity: High]
This is a pre-existing issue, but does synchronize_rcu() properly protect
against concurrent access in wacom_mode_change_work()?

When a sibling device is removed, this devres cleanup clears the sibling
pointers and calls synchronize_rcu(). However, wacom_mode_change_work()
retrieves the sibling pointer outside of an RCU critical section:

	struct hid_device *pen = rcu_access_pointer(shared->pen);
	struct hid_device *touch = rcu_access_pointer(shared->touch);

	if (pen) {
		wacom1 = hid_get_drvdata(pen);
		wacom_release_resources(wacom1);
		hid_hw_stop(wacom1->hdev);

Since wacom_mode_change_work() executes in a sleepable workqueue context and
uses rcu_access_pointer() instead of rcu_dereference() inside an
rcu_read_lock(), does synchronize_rcu() actually wait for the workqueue to
finish?

If wacom_mode_change_work() is executing concurrently with sibling removal,
could it dereference a dangling sibling pointer leading to a use-after-free
when the sibling's devres cleanup frees the memory?

>  
> -		kref_put(&data->kref, wacom_release_shared_data);
> +		if (kref_put_mutex(&data->kref, wacom_release_shared_data, &wacom_udev_list_lock))
> +			mutex_unlock(&wacom_udev_list_lock);
> +
>  		wacom_wac->shared = NULL;
>  	}
>  }

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260902134417.1270552-1-lee@kernel.org?part=5