Re: [PATCH v2] drm/rockchip: rk3066_hdmi: convert to devm_drm_bridge_alloc() API

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

Pre-existing issues:
- [High] This problem wasn't introduced by this patch, but there is a Use-After-Free vulnerability on the error path of `rk3066_hdmi_bind()`. If `rk3066_hdmi_register()` fails, the code jumps to `err_disable_hclk`, skipping the encoder cleanup label `err_cleanup_hdmi`. Since the DRM encoder was added to the DRM device's list during `drm_encoder_init()` but is not removed, and the `hdmi` struct containing the encoder is freed by devres upon bind failure, the DRM core retains a dangling pointer. This will cause a Use-After-Free when the DRM core later iterates over the encoder list or during teardown.
--

--- Patch [1]: [PATCH v2] drm/rockchip: rk3066_hdmi: convert to devm_drm_bridge_alloc() API ---
commit bdad67e77f8f693946bd674b1a40ce7af3ce54c3
Author: Hüseyin BIYIK <[email protected]>

drm/rockchip: rk3066_hdmi: convert to devm_drm_bridge_alloc() API

This commit converts the rk3066_hdmi driver to use devm_drm_bridge_alloc()
for allocating DRM bridges. This conversion was missed during the initial
conversion of all bridges to the new API.

> diff --git a/drivers/gpu/drm/rockchip/rk3066_hdmi.c b/drivers/gpu/drm/rockchip/rk3066_hdmi.c
> index f28d6a883e099..39c65ee07caac 100644
> --- a/drivers/gpu/drm/rockchip/rk3066_hdmi.c
> +++ b/drivers/gpu/drm/rockchip/rk3066_hdmi.c
[ ... ]
> @@ -747,9 +746,10 @@ static int rk3066_hdmi_bind(struct device *dev, struct device *master,
>  	int irq;
>  	int ret;
>  
> -	hdmi = devm_kzalloc(dev, sizeof(*hdmi), GFP_KERNEL);
> -	if (!hdmi)
> -		return -ENOMEM;
> +	hdmi = devm_drm_bridge_alloc(dev, struct rk3066_hdmi, bridge,
> +				     &rk3066_hdmi_bridge_funcs);
> +	if (IS_ERR(hdmi))
> +		return PTR_ERR(hdmi);

[Severity: High]
This isn't a bug introduced by this patch, but while reviewing the
allocation of the hdmi struct in rk3066_hdmi_bind() inside
drivers/gpu/drm/rockchip/rk3066_hdmi.c, I noticed a potential
use-after-free on the error path.

If rk3066_hdmi_register() fails after drm_encoder_init() has been called,
the error path in rk3066_hdmi_bind() jumps to err_disable_hclk:

	ret = rk3066_hdmi_register(drm, hdmi);
	if (ret)
		goto err_disable_hclk;

This bypasses the err_cleanup_hdmi label which destroys the encoder:

err_cleanup_hdmi:
	hdmi->encoder.encoder.funcs->destroy(&hdmi->encoder.encoder);
err_disable_hclk:
	clk_disable_unprepare(hdmi->hclk);

Since the encoder was added to the DRM device's internal lists during
drm_encoder_init(), and the devm-allocated hdmi struct will be freed when
rk3066_hdmi_bind() returns an error, does this leave a dangling pointer in
the DRM core that will cause a use-after-free when the list is iterated or
torn down?

>  
>  	hdmi->dev = dev;
>  	hdmi->drm_dev = drm;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1