netns: add pre_exit method to struct pernet_operations

Current struct pernet_operations exit() handlers are highly
discouraged to call synchronize_rcu().

There are cases where we need them, and exit_batch() does
not help the common case where a single netns is dismantled.

This patch leverages the existing synchronize_rcu() call
in cleanup_net()

Calling optional ->pre_exit() method before ->exit() or
->exit_batch() allows to benefit from a single synchronize_rcu()
call.

Note that the synchronize_rcu() calls added in this patch
are only in error paths or slow paths.

Tested:

$ time for i in {1..1000}; do unshare -n /bin/false;done

real	0m2.612s
user	0m0.171s
sys	0m2.216s

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 33 insertions, 0 deletions

diff --git a/include/net/net_namespace.h b/include/net/net_namespace.h
index abb4f92456e1..ad9243afac67 100644
--- a/include/net/net_namespace.h
+++ b/include/net/net_namespace.h
@@ -355,8 +355,13 @@ struct pernet_operations {
 	 * synchronize_rcu() related to these pernet_operations,
 	 * instead of separate synchronize_rcu() for every net.
 	 * Please, avoid synchronize_rcu() at all, where it's possible.
+	 *
+	 * Note that a combination of pre_exit() and exit() can
+	 * be used, since a synchronize_rcu() is guaranteed between
+	 * the calls.
 	 */
 	int (*init)(struct net *net);
+	void (*pre_exit)(struct net *net);
 	void (*exit)(struct net *net);
 	void (*exit_batch)(struct list_head *net_exit_list);
 	unsigned int *id;
diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index 15f68842ac6b..89dc99a28978 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -145,6 +145,17 @@ static void ops_free(const struct pernet_operations *ops, struct net *net)
 	}
 }
 
+static void ops_pre_exit_list(const struct pernet_operations *ops,
+			      struct list_head *net_exit_list)
+{
+	struct net *net;
+
+	if (ops->pre_exit) {
+		list_for_each_entry(net, net_exit_list, exit_list)
+			ops->pre_exit(net);
+	}
+}
+
 static void ops_exit_list(const struct pernet_operations *ops,
 			  struct list_head *net_exit_list)
 {
@@ -330,6 +341,12 @@ out_undo:
 	list_add(&net->exit_list, &net_exit_list);
 	saved_ops = ops;
 	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
+		ops_pre_exit_list(ops, &net_exit_list);
+
+	synchronize_rcu();
+
+	saved_ops = ops;
+	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 		ops_exit_list(ops, &net_exit_list);
 
 	ops = saved_ops;
@@ -541,10 +558,15 @@ static void cleanup_net(struct work_struct *work)
 		list_add_tail(&net->exit_list, &net_exit_list);
 	}
 
+	/* Run all of the network namespace pre_exit methods */
+	list_for_each_entry_reverse(ops, &pernet_list, list)
+		ops_pre_exit_list(ops, &net_exit_list);
+
 	/*
 	 * Another CPU might be rcu-iterating the list, wait for it.
 	 * This needs to be before calling the exit() notifiers, so
 	 * the rcu_barrier() below isn't sufficient alone.
+	 * Also the pre_exit() and exit() methods need this barrier.
 	 */
 	synchronize_rcu();
 
@@ -1101,6 +1123,8 @@ static int __register_pernet_operations(struct list_head *list,
 out_undo:
 	/* If I have an error cleanup all namespaces I initialized */
 	list_del(&ops->list);
+	ops_pre_exit_list(ops, &net_exit_list);
+	synchronize_rcu();
 	ops_exit_list(ops, &net_exit_list);
 	ops_free_list(ops, &net_exit_list);
 	return error;
@@ -1115,6 +1139,8 @@ static void __unregister_pernet_operations(struct pernet_operations *ops)
 	/* See comment in __register_pernet_operations() */
 	for_each_net(net)
 		list_add_tail(&net->exit_list, &net_exit_list);
+	ops_pre_exit_list(ops, &net_exit_list);
+	synchronize_rcu();
 	ops_exit_list(ops, &net_exit_list);
 	ops_free_list(ops, &net_exit_list);
 }
@@ -1139,6 +1165,8 @@ static void __unregister_pernet_operations(struct pernet_operations *ops)
 	} else {
 		LIST_HEAD(net_exit_list);
 		list_add(&init_net.exit_list, &net_exit_list);
+		ops_pre_exit_list(ops, &net_exit_list);
+		synchronize_rcu();
 		ops_exit_list(ops, &net_exit_list);
 		ops_free_list(ops, &net_exit_list);
 	}