mpls: Protect net->mpls.platform_label with a per-netns mutex.

MPLS (re)uses RTNL to protect net->mpls.platform_label,
but the lock does not need to be RTNL at all.

Let's protect net->mpls.platform_label with a dedicated
per-netns mutex.

Signed-off-by: Kuniyuki Iwashima <[e-mail]>
Reviewed-by: Guillaume Nault <[e-mail]>
Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Jakub Kicinski <[e-mail]>

include/net/netns/mpls.h

···
 	int default_ttl;
 	size_t platform_labels;
 	struct mpls_route __rcu * __rcu *platform_label;
+	struct mutex platform_mutex;
 
 	struct ctl_table_header *ctl;
 };

net/mpls/af_mpls.c

···
 {
 	struct mpls_route __rcu **platform_label;
 
-	platform_label = rtnl_dereference(net->mpls.platform_label);
-	return rtnl_dereference(platform_label[index]);
+	platform_label = mpls_dereference(net, net->mpls.platform_label);
+	return mpls_dereference(net, platform_label[index]);
 }
 
 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned int index)
···
 	struct mpls_route __rcu **platform_label;
 	struct mpls_route *rt;
 
-	ASSERT_RTNL();
-
-	platform_label = rtnl_dereference(net->mpls.platform_label);
-	rt = rtnl_dereference(platform_label[index]);
+	platform_label = mpls_dereference(net, net->mpls.platform_label);
+	rt = mpls_dereference(net, platform_label[index]);
 	rcu_assign_pointer(platform_label[index], new);
 
 	mpls_notify_route(net, index, rt, new, info);
···
 	int err = -ENOMEM;
 	int i;
 
-	ASSERT_RTNL();
-
 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
 	if (!mdev)
 		return ERR_PTR(err);
···
 	unsigned int flags;
 	int err;
 
+	mutex_lock(&net->mpls.platform_mutex);
+
 	if (event == NETDEV_REGISTER) {
 		mdev = mpls_add_dev(dev);
 		if (IS_ERR(mdev)) {
···
 	}
 
 out:
+	mutex_unlock(&net->mpls.platform_mutex);
 	return NOTIFY_OK;
 
 err:
+	mutex_unlock(&net->mpls.platform_mutex);
 	return notifier_from_errno(err);
 }
 
···
 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 			     struct netlink_ext_ack *extack)
 {
+	struct net *net = sock_net(skb->sk);
 	struct mpls_route_config *cfg;
 	int err;
 
···
 	if (err < 0)
 		goto out;
 
+	mutex_lock(&net->mpls.platform_mutex);
 	err = mpls_route_del(cfg, extack);
+	mutex_unlock(&net->mpls.platform_mutex);
 out:
 	kfree(cfg);
 
···
 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 			     struct netlink_ext_ack *extack)
 {
+	struct net *net = sock_net(skb->sk);
 	struct mpls_route_config *cfg;
 	int err;
 
···
 	if (err < 0)
 		goto out;
 
+	mutex_lock(&net->mpls.platform_mutex);
 	err = mpls_route_add(cfg, extack);
+	mutex_unlock(&net->mpls.platform_mutex);
 out:
 	kfree(cfg);
 
···
 	u8 n_labels;
 	int err;
 
+	mutex_lock(&net->mpls.platform_mutex);
+
 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
 	if (err < 0)
 		goto errout;
···
 			goto errout_free;
 		}
 
-		return rtnl_unicast(skb, net, portid);
+		err = rtnl_unicast(skb, net, portid);
+		goto errout;
 	}
 
 	if (tb[RTA_NEWDST]) {
···
 
 	err = rtnl_unicast(skb, net, portid);
 errout:
+	mutex_unlock(&net->mpls.platform_mutex);
 	return err;
 
 nla_put_failure:
 	nlmsg_cancel(skb, nlh);
 	err = -EMSGSIZE;
 errout_free:
+	mutex_unlock(&net->mpls.platform_mutex);
 	kfree_skb(skb);
 	return err;
 }
···
 		       lo->addr_len);
 	}
 
-	rtnl_lock();
+	mutex_lock(&net->mpls.platform_mutex);
+
 	/* Remember the original table */
-	old = rtnl_dereference(net->mpls.platform_label);
+	old = mpls_dereference(net, net->mpls.platform_label);
 	old_limit = net->mpls.platform_labels;
 
 	/* Free any labels beyond the new table */
···
 	net->mpls.platform_labels = limit;
 	rcu_assign_pointer(net->mpls.platform_label, labels);
 
-	rtnl_unlock();
+	mutex_unlock(&net->mpls.platform_mutex);
 
 	mpls_rt_free(rt2);
 	mpls_rt_free(rt0);
···
 	},
 };
 
-static int mpls_net_init(struct net *net)
+static __net_init int mpls_net_init(struct net *net)
 {
 	size_t table_size = ARRAY_SIZE(mpls_table);
 	struct ctl_table *table;
 	int i;
 
+	mutex_init(&net->mpls.platform_mutex);
 	net->mpls.platform_labels = 0;
 	net->mpls.platform_label = NULL;
 	net->mpls.ip_ttl_propagate = 1;
···
 	return 0;
 }
 
-static void mpls_net_exit(struct net *net)
+static __net_exit void mpls_net_exit(struct net *net)
 {
 	struct mpls_route __rcu **platform_label;
 	size_t platform_labels;
···
 	 * As such no additional rcu synchronization is necessary when
 	 * freeing the platform_label table.
 	 */
-	rtnl_lock();
-	platform_label = rtnl_dereference(net->mpls.platform_label);
+	mutex_lock(&net->mpls.platform_mutex);
+
+	platform_label = mpls_dereference(net, net->mpls.platform_label);
 	platform_labels = net->mpls.platform_labels;
+
 	for (index = 0; index < platform_labels; index++) {
-		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
-		RCU_INIT_POINTER(platform_label[index], NULL);
+		struct mpls_route *rt;
+
+		rt = mpls_dereference(net, platform_label[index]);
 		mpls_notify_route(net, index, rt, NULL, NULL);
 		mpls_rt_free(rt);
 	}
-	rtnl_unlock();
+
+	mutex_unlock(&net->mpls.platform_mutex);
 
 	kvfree(platform_label);
 }

net/mpls/internal.h

···
 	return result;
 }
 
+#define mpls_dereference(net, p)					\
+	rcu_dereference_protected(					\
+		(p),							\
+		lockdep_is_held(&(net)->mpls.platform_mutex))
+
 static inline struct mpls_dev *mpls_dev_rcu(const struct net_device *dev)
 {
 	return rcu_dereference(dev->mpls_ptr);
···
 static inline struct mpls_dev *mpls_dev_get(const struct net *net,
 					    const struct net_device *dev)
 {
-	return rcu_dereference_rtnl(dev->mpls_ptr);
+	return mpls_dereference(net, dev->mpls_ptr);
 }
 
 int nla_put_labels(struct sk_buff *skb, int attrtype,  u8 labels,