[PATCH] vfio: check vfio_register_iommu_driver() return value

[PATCH] vfio: check vfio_register_iommu_driver() return value

[Bo Liu]

As vfio_register_iommu_driver() can fail, we should check the return value.

Signed-off-by: Bo Liu <liubo03@inspur.com>
---
 drivers/vfio/vfio.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/vfio/vfio.c b/drivers/vfio/vfio.c
index 61e71c1154be..7d4b6dfafd27 100644
--- a/drivers/vfio/vfio.c
+++ b/drivers/vfio/vfio.c
@@ -2159,10 +2159,16 @@ static int __init vfio_init(void)
 	pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n");
 
 #ifdef CONFIG_VFIO_NOIOMMU
-	vfio_register_iommu_driver(&vfio_noiommu_ops);
+	ret = vfio_register_iommu_driver(&vfio_noiommu_ops);
+	if (ret)
+		goto err_driver_register;
 #endif
 	return 0;
 
+#ifdef CONFIG_VFIO_NOIOMMU
+err_driver_register:
+	unregister_chrdev_region(vfio.group_devt, MINORMASK + 1);
+#endif
 err_alloc_chrdev:
 	class_destroy(vfio.class);
 	vfio.class = NULL;
-- 
2.27.0

Re: [PATCH] vfio: check vfio_register_iommu_driver() return value

[Alex Williamson]

On Tue, 21 Jun 2022 02:21:12 -0400
Bo Liu <liubo03@inspur.com> wrote:

> As vfio_register_iommu_driver() can fail, we should check the return value.
> 
> Signed-off-by: Bo Liu <liubo03@inspur.com>
> ---
>  drivers/vfio/vfio.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/vfio/vfio.c b/drivers/vfio/vfio.c
> index 61e71c1154be..7d4b6dfafd27 100644
> --- a/drivers/vfio/vfio.c
> +++ b/drivers/vfio/vfio.c
> @@ -2159,10 +2159,16 @@ static int __init vfio_init(void)
>  	pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n");

If we're going to acknowledge that the below can fail, then we should
move the pr_info so that we only advertise that vfio has loaded after
all potential failure points.

>  
>  #ifdef CONFIG_VFIO_NOIOMMU
> -	vfio_register_iommu_driver(&vfio_noiommu_ops);
> +	ret = vfio_register_iommu_driver(&vfio_noiommu_ops);
> +	if (ret)
> +		goto err_driver_register;
>  #endif

We could move this #endif above the ret test so that the goto exists
outside of the #ifdef, then we don't need to wrap the goto target in
#ifdefs below.  Thanks,

Alex

>  	return 0;
>  
> +#ifdef CONFIG_VFIO_NOIOMMU
> +err_driver_register:
> +	unregister_chrdev_region(vfio.group_devt, MINORMASK + 1);
> +#endif
>  err_alloc_chrdev:
>  	class_destroy(vfio.class);
>  	vfio.class = NULL;