1 files changed, 32 insertions, 6 deletions

diff --git a/net/core/sock_map.c b/net/core/sock_map.c
index 00a26cf2cfe9..4059f94e9bb5 100644
--- a/net/core/sock_map.c
+++ b/net/core/sock_map.c
@@ -424,10 +424,7 @@ static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 	return 0;
 }
 
-static bool sock_map_redirect_allowed(const struct sock *sk)
-{
-	return sk->sk_state != TCP_LISTEN;
-}
+static bool sock_map_redirect_allowed(const struct sock *sk);
 
 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 				  struct sock *sk, u64 flags)
@@ -508,6 +505,11 @@ static bool sk_is_udp(const struct sock *sk)
 	       sk->sk_protocol == IPPROTO_UDP;
 }
 
+static bool sock_map_redirect_allowed(const struct sock *sk)
+{
+	return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
+}
+
 static bool sock_map_sk_is_suitable(const struct sock *sk)
 {
 	return sk_is_tcp(sk) || sk_is_udp(sk);
@@ -989,11 +991,15 @@ static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
 		err = -EINVAL;
 		goto free_htab;
 	}
+	err = bpf_map_charge_init(&htab->map.memory, cost);
+	if (err)
+		goto free_htab;
 
 	htab->buckets = bpf_map_area_alloc(htab->buckets_num *
 					   sizeof(struct bpf_htab_bucket),
 					   htab->map.numa_node);
 	if (!htab->buckets) {
+		bpf_map_charge_finish(&htab->map.memory);
 		err = -ENOMEM;
 		goto free_htab;
 	}
@@ -1013,6 +1019,7 @@ static void sock_hash_free(struct bpf_map *map)
 {
 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 	struct bpf_htab_bucket *bucket;
+	struct hlist_head unlink_list;
 	struct bpf_htab_elem *elem;
 	struct hlist_node *node;
 	int i;
@@ -1024,13 +1031,32 @@ static void sock_hash_free(struct bpf_map *map)
 	synchronize_rcu();
 	for (i = 0; i < htab->buckets_num; i++) {
 		bucket = sock_hash_select_bucket(htab, i);
-		hlist_for_each_entry_safe(elem, node, &bucket->head, node) {
-			hlist_del_rcu(&elem->node);
+
+		/* We are racing with sock_hash_delete_from_link to
+		 * enter the spin-lock critical section. Every socket on
+		 * the list is still linked to sockhash. Since link
+		 * exists, psock exists and holds a ref to socket. That
+		 * lets us to grab a socket ref too.
+		 */
+		raw_spin_lock_bh(&bucket->lock);
+		hlist_for_each_entry(elem, &bucket->head, node)
+			sock_hold(elem->sk);
+		hlist_move_list(&bucket->head, &unlink_list);
+		raw_spin_unlock_bh(&bucket->lock);
+
+		/* Process removed entries out of atomic context to
+		 * block for socket lock before deleting the psock's
+		 * link to sockhash.
+		 */
+		hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
+			hlist_del(&elem->node);
 			lock_sock(elem->sk);
 			rcu_read_lock();
 			sock_map_unref(elem->sk, elem);
 			rcu_read_unlock();
 			release_sock(elem->sk);
+			sock_put(elem->sk);
+			sock_hash_free_elem(htab, elem);
 		}
 	}