Re: [PATCH] net: seeq: Fix use after free vulnerability in ether3 Driver Due to Race Condition

[Przemek Kitszel <przemyslaw.kitszel@intel.com>]

On 9/9/24 19:58, Kaixin Wang wrote:
> In the ether3_probe function, a timer is initialized with a callback
> function ether3_ledoff, bound to &prev(dev)->timer. Once the timer is
> started, there is a risk of a race condition if the module or device
> is removed, triggering the ether3_remove function to perform cleanup.
> The sequence of operations that may lead to a UAF bug is as follows:
> 
> CPU0                                    CPU1
> 
>                        |  ether3_ledoff
> ether3_remove         |
>    free_netdev(dev);   |
>    put_devic           |
>    kfree(dev);         |
>   |  ether3_outw(priv(dev)->regs.config2 |= CFG2_CTRLO, REG_CONFIG2);
>                        | // use dev
> 
> Fix it by ensuring that the timer is canceled before proceeding with
> the cleanup in ether3_remove.

this code change indeed prevents UAF bug
but as is, the CFG2_CTRLO flag of REG_CONFIG2 will be left in state "ON"

it would be better to first turn the LED off unconditionally

> 
> Signed-off-by: Kaixin Wang <kxwang23@m.fudan.edu.cn>
> ---
>   drivers/net/ethernet/seeq/ether3.c | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/net/ethernet/seeq/ether3.c b/drivers/net/ethernet/seeq/ether3.c
> index c672f92d65e9..f9d27c9d6808 100644
> --- a/drivers/net/ethernet/seeq/ether3.c
> +++ b/drivers/net/ethernet/seeq/ether3.c
> @@ -850,6 +850,7 @@ static void ether3_remove(struct expansion_card *ec)
>   	ecard_set_drvdata(ec, NULL);
>   
>   	unregister_netdev(dev);
> +	del_timer_sync(&priv(dev)->timer);
>   	free_netdev(dev);
>   	ecard_release_resources(ec);
>   }