net: add a refcount tracker for kernel sockets

Commit ffa84b5ffb37 ("net: add netns refcount tracker to struct sock")
added a tracker to sockets, but did not track kernel sockets.

We still have syzbot reports hinting about netns being destroyed
while some kernel TCP sockets had not been dismantled.

This patch tracks kernel sockets, and adds a ref_tracker_dir_print()
call to net_free() right before the netns is freed.

Normally, each layer is responsible for properly releasing its
kernel sockets before last call to net_free().

This debugging facility is enabled with CONFIG_NET_NS_REFCNT_TRACKER=y

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Tested-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

5 files changed, 55 insertions, 8 deletions

diff --git a/include/net/net_namespace.h b/include/net/net_namespace.h
index 8c3587d5c308..78beaa765c73 100644
--- a/include/net/net_namespace.h
+++ b/include/net/net_namespace.h
@@ -92,7 +92,9 @@ struct net {
 
 	struct ns_common	ns;
 	struct ref_tracker_dir  refcnt_tracker;
-
+	struct ref_tracker_dir  notrefcnt_tracker; /* tracker for objects not
+						    * refcounted against netns
+						    */
 	struct list_head 	dev_base_head;
 	struct proc_dir_entry 	*proc_net;
 	struct proc_dir_entry 	*proc_net_stat;
@@ -320,19 +322,31 @@ static inline int check_net(const struct net *net)
 #endif
 
 
-static inline void netns_tracker_alloc(struct net *net,
-				       netns_tracker *tracker, gfp_t gfp)
+static inline void __netns_tracker_alloc(struct net *net,
+					 netns_tracker *tracker,
+					 bool refcounted,
+					 gfp_t gfp)
 {
 #ifdef CONFIG_NET_NS_REFCNT_TRACKER
-	ref_tracker_alloc(&net->refcnt_tracker, tracker, gfp);
+	ref_tracker_alloc(refcounted ? &net->refcnt_tracker :
+				       &net->notrefcnt_tracker,
+			  tracker, gfp);
 #endif
 }
 
-static inline void netns_tracker_free(struct net *net,
-				      netns_tracker *tracker)
+static inline void netns_tracker_alloc(struct net *net, netns_tracker *tracker,
+				       gfp_t gfp)
+{
+	__netns_tracker_alloc(net, tracker, true, gfp);
+}
+
+static inline void __netns_tracker_free(struct net *net,
+					netns_tracker *tracker,
+					bool refcounted)
 {
 #ifdef CONFIG_NET_NS_REFCNT_TRACKER
-       ref_tracker_free(&net->refcnt_tracker, tracker);
+       ref_tracker_free(refcounted ? &net->refcnt_tracker :
+				     &net->notrefcnt_tracker, tracker);
 #endif
 }
 
@@ -346,7 +360,7 @@ static inline struct net *get_net_track(struct net *net,
 
 static inline void put_net_track(struct net *net, netns_tracker *tracker)
 {
-	netns_tracker_free(net, tracker);
+	__netns_tracker_free(net, tracker, true);
 	put_net(net);
 }
 
diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index 0ec2f5906a27..12c68edf7682 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -309,6 +309,7 @@ static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 
 	refcount_set(&net->ns.count, 1);
 	ref_tracker_dir_init(&net->refcnt_tracker, 128);
+	ref_tracker_dir_init(&net->notrefcnt_tracker, 128);
 
 	refcount_set(&net->passive, 1);
 	get_random_bytes(&net->hash_mix, sizeof(u32));
@@ -429,6 +430,10 @@ static void net_free(struct net *net)
 {
 	if (refcount_dec_and_test(&net->passive)) {
 		kfree(rcu_access_pointer(net->gen));
+
+		/* There should not be any trackers left there. */
+		ref_tracker_dir_exit(&net->notrefcnt_tracker);
+
 		kmem_cache_free(net_cachep, net);
 	}
 }
diff --git a/net/core/sock.c b/net/core/sock.c
index a3ba0358c77c..aa608dc0930b 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -2094,6 +2094,9 @@ struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
 		if (likely(sk->sk_net_refcnt)) {
 			get_net_track(net, &sk->ns_tracker, priority);
 			sock_inuse_add(net, 1);
+		} else {
+			__netns_tracker_alloc(net, &sk->ns_tracker,
+					      false, priority);
 		}
 
 		sock_net_set(sk, net);
@@ -2149,6 +2152,9 @@ static void __sk_destruct(struct rcu_head *head)
 
 	if (likely(sk->sk_net_refcnt))
 		put_net_track(sock_net(sk), &sk->ns_tracker);
+	else
+		__netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
+
 	sk_prot_free(sk->sk_prot_creator, sk);
 }
 
@@ -2237,6 +2243,14 @@ struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
 	if (likely(newsk->sk_net_refcnt)) {
 		get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
 		sock_inuse_add(sock_net(newsk), 1);
+	} else {
+		/* Kernel sockets are not elevating the struct net refcount.
+		 * Instead, use a tracker to more easily detect if a layer
+		 * is not properly dismantling its kernel sockets at netns
+		 * destroy time.
+		 */
+		__netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
+				      false, priority);
 	}
 	sk_node_init(&newsk->sk_node);
 	sock_lock_init(newsk);
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index a662e8a5ff84..f0c94d394ab1 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -812,6 +812,17 @@ static int netlink_release(struct socket *sock)
 	}
 
 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
+
+	/* Because struct net might disappear soon, do not keep a pointer. */
+	if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
+		__netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
+		/* Because of deferred_put_nlk_sk and use of work queue,
+		 * it is possible  netns will be freed before this socket.
+		 */
+		sock_net_set(sk, &init_net);
+		__netns_tracker_alloc(&init_net, &sk->ns_tracker,
+				      false, GFP_KERNEL);
+	}
 	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
 	return 0;
 }
diff --git a/net/rds/tcp.c b/net/rds/tcp.c
index 4444fd82b66d..c5b86066ff66 100644
--- a/net/rds/tcp.c
+++ b/net/rds/tcp.c
@@ -503,6 +503,9 @@ bool rds_tcp_tune(struct socket *sock)
 			release_sock(sk);
 			return false;
 		}
+		/* Update ns_tracker to current stack trace and refcounted tracker */
+		__netns_tracker_free(net, &sk->ns_tracker, false);
+
 		sk->sk_net_refcnt = 1;
 		netns_tracker_alloc(net, &sk->ns_tracker, GFP_KERNEL);
 		sock_inuse_add(net, 1);