linux-kernel - Re: [PATCH] media: rp1-cfe: Fix double-free on video device re-registration

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20260211081530.GB2553356@killaraus.ideasonboard.com>
Date: Wed, 11 Feb 2026 10:15:30 +0200
From: Laurent Pinchart <laurent.pinchart@...asonboard.com>
To: Xiaolei Wang <xiaolei.wang@...driver.com>
Cc: tomi.valkeinen@...asonboard.com, kernel-list@...pberrypi.com,
	mchehab@...nel.org, florian.fainelli@...adcom.com,
	bcm-kernel-feedback-list@...adcom.com, hverkuil@...all.nl,
	sakari.ailus@...ux.intel.com, linux-media@...r.kernel.org,
	linux-rpi-kernel@...ts.infradead.org,
	linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] media: rp1-cfe: Fix double-free on video device
 re-registration

Hi Xiaolei,

On Wed, Feb 11, 2026 at 11:45:01AM +0800, Xiaolei Wang wrote:
> When a sensor driver is unloaded and reloaded (e.g., rmmod/insmod ov5647),
> the cfe_async_complete callback is invoked again, attempting to re-register
> video nodes that are still registered. This causes multiple issues:
> 
> 1. KASAN double-free in kfree_const when dev_set_name tries to free the
>    kobject name that was already freed during video_unregister_device
> 2. "tried to init an initialized object" warnings because the video_device
>    kobject is re-initialized before being fully released
> 
> Fix this by:
> - Adding a check in cfe_probe_complete() to skip nodes already in
>   NODE_REGISTERED state, preventing duplicate registration attempts
> - Implementing cfe_async_unbind() callback to properly clear the
>   source_sd pointer when the subdevice is unbound

I think a better fix would be to register video nodes at probe time, not
when sensors are bound.

> Signed-off-by: Xiaolei Wang <xiaolei.wang@...driver.com>
> ---
>  drivers/media/platform/raspberrypi/rp1-cfe/cfe.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
> 
> diff --git a/drivers/media/platform/raspberrypi/rp1-cfe/cfe.c b/drivers/media/platform/raspberrypi/rp1-cfe/cfe.c
> index 62dca76b468d..d3813c79316d 100644
> --- a/drivers/media/platform/raspberrypi/rp1-cfe/cfe.c
> +++ b/drivers/media/platform/raspberrypi/rp1-cfe/cfe.c
> @@ -2152,6 +2152,9 @@ static int cfe_probe_complete(struct cfe_device *cfe)
>  	cfe->v4l2_dev.notify = cfe_notify;
>  
>  	for (unsigned int i = 0; i < NUM_NODES; i++) {
> +		if (check_state(cfe, NODE_REGISTERED, i))
> +			continue;
> +
>  		ret = cfe_register_node(cfe, i);
>  		if (ret) {
>  			cfe_err(cfe, "Unable to register video node %u.\n", i);
> @@ -2204,8 +2207,19 @@ static int cfe_async_complete(struct v4l2_async_notifier *notifier)
>  	return cfe_probe_complete(cfe);
>  }
>  
> +static void cfe_async_unbind(struct v4l2_async_notifier *notifier,
> +			     struct v4l2_subdev *subdev,
> +			     struct v4l2_async_connection *asd)
> +{
> +	struct cfe_device *cfe = to_cfe_device(notifier->v4l2_dev);
> +
> +	cfe->source_sd = NULL;
> +	cfe_info(cfe, "Unbinding subdev %s\n", subdev->name);
> +}
> +
>  static const struct v4l2_async_notifier_operations cfe_async_ops = {
>  	.bound = cfe_async_bound,
> +	.unbind = cfe_async_unbind,
>  	.complete = cfe_async_complete,
>  };
>  

-- 
Regards,

Laurent Pinchart