wireguard: device: avoid circular netns references

Before, we took a reference to the creating netns if the new netns was
different. This caused issues with circular references, with two
wireguard interfaces swapping namespaces. The solution is to rather not
take any extra references at all, but instead simply invalidate the
creating netns pointer when that netns is deleted.

In order to prevent this from happening again, this commit improves the
rough object leak tracking by allowing it to account for created and
destroyed interfaces, aside from just peers and keys. That then makes it
possible to check for the object leak when having two interfaces take a
reference to each others' namespaces.

Fixes: e7096c131e51 ("net: WireGuard secure network tunnel")
Signed-off-by: Jason A. Donenfeld <[e-mail]>
Signed-off-by: David S. Miller <[e-mail]>

drivers/net/wireguard/device.c

···
 	if (dev_v6)
 		dev_v6->cnf.addr_gen_mode = IN6_ADDR_GEN_MODE_NONE;
 
+	mutex_lock(&wg->device_update_lock);
 	ret = wg_socket_init(wg, wg->incoming_port);
 	if (ret < 0)
-		return ret;
-	mutex_lock(&wg->device_update_lock);
+		goto out;
 	list_for_each_entry(peer, &wg->peer_list, peer_list) {
 		wg_packet_send_staged_packets(peer);
 		if (peer->persistent_keepalive_interval)
 			wg_packet_send_keepalive(peer);
 	}
+out:
 	mutex_unlock(&wg->device_update_lock);
-	return 0;
+	return ret;
 }
 
 #ifdef CONFIG_PM_SLEEP
···
 	list_del(&wg->device_list);
 	rtnl_unlock();
 	mutex_lock(&wg->device_update_lock);
+	rcu_assign_pointer(wg->creating_net, NULL);
 	wg->incoming_port = 0;
 	wg_socket_reinit(wg, NULL, NULL);
 	/* The final references are cleared in the below calls to destroy_workqueue. */
···
 	skb_queue_purge(&wg->incoming_handshakes);
 	free_percpu(dev->tstats);
 	free_percpu(wg->incoming_handshakes_worker);
-	if (wg->have_creating_net_ref)
-		put_net(wg->creating_net);
 	kvfree(wg->index_hashtable);
 	kvfree(wg->peer_hashtable);
 	mutex_unlock(&wg->device_update_lock);
 
-	pr_debug("%s: Interface deleted\n", dev->name);
+	pr_debug("%s: Interface destroyed\n", dev->name);
 	free_netdev(dev);
 }
 
···
 	struct wg_device *wg = netdev_priv(dev);
 	int ret = -ENOMEM;
 
-	wg->creating_net = src_net;
+	rcu_assign_pointer(wg->creating_net, src_net);
 	init_rwsem(&wg->static_identity.lock);
 	mutex_init(&wg->socket_update_lock);
 	mutex_init(&wg->device_update_lock);
···
 	.newlink		= wg_newlink,
 };
 
-static int wg_netdevice_notification(struct notifier_block *nb,
-				     unsigned long action, void *data)
+static void wg_netns_pre_exit(struct net *net)
 {
-	struct net_device *dev = ((struct netdev_notifier_info *)data)->dev;
-	struct wg_device *wg = netdev_priv(dev);
+	struct wg_device *wg;
 
-	ASSERT_RTNL();
-
-	if (action != NETDEV_REGISTER || dev->netdev_ops != &netdev_ops)
-		return 0;
-
-	if (dev_net(dev) == wg->creating_net && wg->have_creating_net_ref) {
-		put_net(wg->creating_net);
-		wg->have_creating_net_ref = false;
-	} else if (dev_net(dev) != wg->creating_net &&
-		   !wg->have_creating_net_ref) {
-		wg->have_creating_net_ref = true;
-		get_net(wg->creating_net);
+	rtnl_lock();
+	list_for_each_entry(wg, &device_list, device_list) {
+		if (rcu_access_pointer(wg->creating_net) == net) {
+			pr_debug("%s: Creating namespace exiting\n", wg->dev->name);
+			netif_carrier_off(wg->dev);
+			mutex_lock(&wg->device_update_lock);
+			rcu_assign_pointer(wg->creating_net, NULL);
+			wg_socket_reinit(wg, NULL, NULL);
+			mutex_unlock(&wg->device_update_lock);
+		}
 	}
-	return 0;
+	rtnl_unlock();
 }
 
-static struct notifier_block netdevice_notifier = {
-	.notifier_call = wg_netdevice_notification
+static struct pernet_operations pernet_ops = {
+	.pre_exit = wg_netns_pre_exit
 };
 
 int __init wg_device_init(void)
···
 		return ret;
 #endif
 
-	ret = register_netdevice_notifier(&netdevice_notifier);
+	ret = register_pernet_device(&pernet_ops);
 	if (ret)
 		goto error_pm;
 
 	ret = rtnl_link_register(&link_ops);
 	if (ret)
-		goto error_netdevice;
+		goto error_pernet;
 
 	return 0;
 
-error_netdevice:
-	unregister_netdevice_notifier(&netdevice_notifier);
+error_pernet:
+	unregister_pernet_device(&pernet_ops);
 error_pm:
 #ifdef CONFIG_PM_SLEEP
 	unregister_pm_notifier(&pm_notifier);
···
 void wg_device_uninit(void)
 {
 	rtnl_link_unregister(&link_ops);
-	unregister_netdevice_notifier(&netdevice_notifier);
+	unregister_pernet_device(&pernet_ops);
 #ifdef CONFIG_PM_SLEEP
 	unregister_pm_notifier(&pm_notifier);
 #endif

drivers/net/wireguard/device.h

···
 	struct net_device *dev;
 	struct crypt_queue encrypt_queue, decrypt_queue;
 	struct sock __rcu *sock4, *sock6;
-	struct net *creating_net;
+	struct net __rcu *creating_net;
 	struct noise_static_identity static_identity;
 	struct workqueue_struct *handshake_receive_wq, *handshake_send_wq;
 	struct workqueue_struct *packet_crypt_wq;
···
 	unsigned int num_peers, device_update_gen;
 	u32 fwmark;
 	u16 incoming_port;
-	bool have_creating_net_ref;
 };
 
 int wg_device_init(void);

drivers/net/wireguard/netlink.c

···
 	if (flags & ~__WGDEVICE_F_ALL)
 		goto out;
 
-	ret = -EPERM;
-	if ((info->attrs[WGDEVICE_A_LISTEN_PORT] ||
-	     info->attrs[WGDEVICE_A_FWMARK]) &&
-	    !ns_capable(wg->creating_net->user_ns, CAP_NET_ADMIN))
-		goto out;
+	if (info->attrs[WGDEVICE_A_LISTEN_PORT] || info->attrs[WGDEVICE_A_FWMARK]) {
+		struct net *net;
+		rcu_read_lock();
+		net = rcu_dereference(wg->creating_net);
+		ret = !net || !ns_capable(net->user_ns, CAP_NET_ADMIN) ? -EPERM : 0;
+		rcu_read_unlock();
+		if (ret)
+			goto out;
+	}
 
 	++wg->device_update_gen;
 

drivers/net/wireguard/socket.c

···
 
 int wg_socket_init(struct wg_device *wg, u16 port)
 {
+	struct net *net;
 	int ret;
 	struct udp_tunnel_sock_cfg cfg = {
 		.sk_user_data = wg,
···
 	};
 #endif
 
+	rcu_read_lock();
+	net = rcu_dereference(wg->creating_net);
+	net = net ? maybe_get_net(net) : NULL;
+	rcu_read_unlock();
+	if (unlikely(!net))
+		return -ENONET;
+
 #if IS_ENABLED(CONFIG_IPV6)
 retry:
 #endif
 
-	ret = udp_sock_create(wg->creating_net, &port4, &new4);
+	ret = udp_sock_create(net, &port4, &new4);
 	if (ret < 0) {
 		pr_err("%s: Could not create IPv4 socket\n", wg->dev->name);
-		return ret;
+		goto out;
 	}
 	set_sock_opts(new4);
-	setup_udp_tunnel_sock(wg->creating_net, new4, &cfg);
+	setup_udp_tunnel_sock(net, new4, &cfg);
 
 #if IS_ENABLED(CONFIG_IPV6)
 	if (ipv6_mod_enabled()) {
 		port6.local_udp_port = inet_sk(new4->sk)->inet_sport;
-		ret = udp_sock_create(wg->creating_net, &port6, &new6);
+		ret = udp_sock_create(net, &port6, &new6);
 		if (ret < 0) {
 			udp_tunnel_sock_release(new4);
 			if (ret == -EADDRINUSE && !port && retries++ < 100)
 				goto retry;
 			pr_err("%s: Could not create IPv6 socket\n",
 			       wg->dev->name);
-			return ret;
+			goto out;
 		}
 		set_sock_opts(new6);
-		setup_udp_tunnel_sock(wg->creating_net, new6, &cfg);
+		setup_udp_tunnel_sock(net, new6, &cfg);
 	}
 #endif
 
 	wg_socket_reinit(wg, new4->sk, new6 ? new6->sk : NULL);
-	return 0;
+	ret = 0;
+out:
+	put_net(net);
+	return ret;
 }
 
 void wg_socket_reinit(struct wg_device *wg, struct sock *new4,

tools/testing/selftests/wireguard/netns.sh

···
 kill $ncat_pid
 ip0 link del wg0
 
+# Ensure there aren't circular reference loops
+ip1 link add wg1 type wireguard
+ip2 link add wg2 type wireguard
+ip1 link set wg1 netns $netns2
+ip2 link set wg2 netns $netns1
+pp ip netns delete $netns1
+pp ip netns delete $netns2
+pp ip netns add $netns1
+pp ip netns add $netns2
+
+sleep 2 # Wait for cleanup and grace periods
 declare -A objects
 while read -t 0.1 -r line 2>/dev/null || [[ $? -ne 142 ]]; do
-	[[ $line =~ .*(wg[0-9]+:\ [A-Z][a-z]+\ [0-9]+)\ .*(created|destroyed).* ]] || continue
+	[[ $line =~ .*(wg[0-9]+:\ [A-Z][a-z]+\ ?[0-9]*)\ .*(created|destroyed).* ]] || continue
 	objects["${BASH_REMATCH[1]}"]+="${BASH_REMATCH[2]}"
 done < /dev/kmsg
 alldeleted=1