[PATCH] s390: cmf: fix use-after-free in enable_cmf

[PATCH] s390: cmf: fix use-after-free in enable_cmf

[Qinglang Miao]

kfree(cdev) is called in put_device in the error branch. So that
device_unlock(&cdev->dev) would raise a use-after-free bug. In fact,
there's no need to call device_unlock after put_device.

Fix it by adding simply return after put_device.

Fixes: a6ef15652d26 ("s390/cio: fix use after free in cmb processing")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Qinglang Miao <miaoqinglang@huawei.com>
---
 drivers/s390/cio/cmf.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/s390/cio/cmf.c b/drivers/s390/cio/cmf.c
index 72dd2471e..e95ca476f 100644
--- a/drivers/s390/cio/cmf.c
+++ b/drivers/s390/cio/cmf.c
@@ -1149,9 +1149,12 @@ int enable_cmf(struct ccw_device *cdev)
 		sysfs_remove_group(&cdev->dev.kobj, cmbops->attr_group);
 		cmbops->free(cdev);
 	}
+
 out:
-	if (ret)
+	if (ret) {
 		put_device(&cdev->dev);
+		return ret;
+	}
 out_unlock:
 	device_unlock(&cdev->dev);
 	return ret;
-- 
2.23.0

Re: [PATCH] s390: cmf: fix use-after-free in enable_cmf

[Cornelia Huck]

On Fri, 20 Nov 2020 15:48:50 +0800
Qinglang Miao <miaoqinglang@huawei.com> wrote:

> kfree(cdev) is called in put_device in the error branch. So that
> device_unlock(&cdev->dev) would raise a use-after-free bug. In fact,
> there's no need to call device_unlock after put_device.
> 
> Fix it by adding simply return after put_device.
> 
> Fixes: a6ef15652d26 ("s390/cio: fix use after free in cmb processing")
> Reported-by: Hulk Robot <hulkci@huawei.com>
> Signed-off-by: Qinglang Miao <miaoqinglang@huawei.com>
> ---
>  drivers/s390/cio/cmf.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/s390/cio/cmf.c b/drivers/s390/cio/cmf.c
> index 72dd2471e..e95ca476f 100644
> --- a/drivers/s390/cio/cmf.c
> +++ b/drivers/s390/cio/cmf.c
> @@ -1149,9 +1149,12 @@ int enable_cmf(struct ccw_device *cdev)
>  		sysfs_remove_group(&cdev->dev.kobj, cmbops->attr_group);
>  		cmbops->free(cdev);
>  	}
> +
>  out:
> -	if (ret)
> +	if (ret) {
>  		put_device(&cdev->dev);

The put_device() here undoes a get_device() further up in the function.
There is at least one more reference remaining, held by the caller of
enable_cmf(). Returning here would actually introduce a bug (missing
unlock).

> +		return ret;
> +	}
>  out_unlock:
>  	device_unlock(&cdev->dev);
>  	return ret;