BUG: scheduling while atomic: sbd_reaper in sheepdog block device drive

[Bambo-Huang]

submit_sheep_request should not be in atomic context but submit_blocking_sheep_request did so.
So I did a fix as the following:

static void submit_blocking_sheep_request(struct sbd_device *dev, uint64_t oid)
{
	struct sheep_request *req, *t;

	write_lock(&dev->blocking_lock);
	list_for_each_entry_safe(req, t, &dev->blocking_head, list) {
		if (req->oid != oid)
			continue;
		list_del_init(&req->list);
// Fix here, we unlock, do request, and lock it again.
		write_unlock(&dev->blocking_lock);
		submit_sheep_request(req);
		write_lock(&dev->blocking_lock);
	}
	write_unlock(&dev->blocking_lock);
}