Re: [PATCH] mfd: dln2: Fix use-after-free risk in list iterator

[Lee Jones <lee@kernel.org>]

On Mon, 12 Oct 2026, Riccardo Oliva wrote:

> In dln2_unregister_event_cb(), the iterator variable 'i' is used after
> the list_for_each_entry() loop terminates to free memory.

How is this a use-after-free risk?  'i' is only ever used after the loop
if 'found' is true, which means we broke out of the loop early while 'i'
still points to a valid entry.

Is this an actual bug or are we just silencing a static checker false
positive?

>
> Fix this by introducing a dedicated 'target' pointer to hold the entry,
> isolating the iterator to the loop scope and removing the 'found' bool.
>
> Assisted-by: coccinelle sparse
> Signed-off-by: Riccardo Oliva <oliva.scorza@gmail.com>
> ---
>  drivers/mfd/dln2.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/mfd/dln2.c b/drivers/mfd/dln2.c
> index 0b1e7fd7f..7cd64aad1 100644
> --- a/drivers/mfd/dln2.c
> +++ b/drivers/mfd/dln2.c
> @@ -157,25 +157,24 @@ EXPORT_SYMBOL(dln2_register_event_cb);
>  void dln2_unregister_event_cb(struct platform_device *pdev, u16 id)
>  {
>  	struct dln2_dev *dln2 = dev_get_drvdata(pdev->dev.parent);
> -	struct dln2_event_cb_entry *i;
> +	struct dln2_event_cb_entry *i, *target = NULL;

If we do keep this change as a cleanup, does it make sense to use
'entry' instead of 'target' to match 'dln2_register_event_cb()' above?

>  	unsigned long flags;
> -	bool found = false;
>
>  	spin_lock_irqsave(&dln2->event_cb_lock, flags);
>
>  	list_for_each_entry(i, &dln2->event_cb_list, list) {
>  		if (i->id == id) {
>  			list_del_rcu(&i->list);
> -			found = true;
> +			target = i;
>  			break;
>  		}
>  	}
>
>  	spin_unlock_irqrestore(&dln2->event_cb_lock, flags);
>
> -	if (found) {
> +	if (target) {
>  		synchronize_rcu();
> -		kfree(i);
> +		kfree(target);
>  	}
>  }
>  EXPORT_SYMBOL(dln2_unregister_event_cb);
> -- 
> 2.43.0
>

-- 
Lee Jones