Re: [PATCH V4 4/5] virtio: introduce a vDPA based transport

[Jason Gunthorpe <jgg@mellanox.com>]

On Thu, Feb 20, 2020 at 02:11:40PM +0800, Jason Wang wrote:
> +static int virtio_vdpa_probe(struct vdpa_device *vdpa)
> +{
> +	const struct vdpa_config_ops *ops = vdpa->config;
> +	struct virtio_vdpa_device *vd_dev;
> +	int ret = -EINVAL;
> +
> +	vd_dev = kzalloc(sizeof(*vd_dev), GFP_KERNEL);
> +	if (!vd_dev)
> +		return -ENOMEM;
> +
> +	vd_dev->vdev.dev.parent = vdpa_get_dma_dev(vdpa);
> +	vd_dev->vdev.dev.release = virtio_vdpa_release_dev;
> +	vd_dev->vdev.config = &virtio_vdpa_config_ops;
> +	vd_dev->vdpa = vdpa;
> +	INIT_LIST_HEAD(&vd_dev->virtqueues);
> +	spin_lock_init(&vd_dev->lock);
> +
> +	vd_dev->vdev.id.device = ops->get_device_id(vdpa);
> +	if (vd_dev->vdev.id.device == 0)
> +		goto err;
> +
> +	vd_dev->vdev.id.vendor = ops->get_vendor_id(vdpa);
> +	ret = register_virtio_device(&vd_dev->vdev);
> +	if (ret)
> +		goto err;

This error unwind is wrong. register_virtio_device() does
device_initialize() as it's first action. After that point error
unwind must be done with put_device() - particularly calling
kfree(vd_dev) after doing dev_set_name() leaks memory.

Looks like about half of the register_virtio_device() users did this
right, the others not. Perhaps you should fix them too...

Jason