[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <26614b92-4d24-4aff-8fc3-25aa8ed83cb6@intel.com>
Date: Wed, 11 Sep 2024 11:29:44 +0200
From: Przemek Kitszel <przemyslaw.kitszel@...el.com>
To: Kaixin Wang <kxwang23@...udan.edu.cn>
CC: <wtdeng24@...udan.edu.cn>, <21210240012@...udan.edu.cn>,
<netdev@...r.kernel.org>, <linux-kernel@...r.kernel.org>,
<edumazet@...gle.com>, <kuba@...nel.org>, <davem@...emloft.net>
Subject: Re: [PATCH] net: seeq: Fix use after free vulnerability in ether3
Driver Due to Race Condition
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@...udan.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);
> }
Powered by blists - more mailing lists