diff --git a/[refs] b/[refs]
index eb19d104f9ef..45c0c50b21d4 100644
--- a/[refs]
+++ b/[refs]
@@ -1,2 +1,2 @@
 ---
-refs/heads/master: e7dfb09a3624a32df2dd00683875d0e56406a603
+refs/heads/master: 91c46e2e60da0b43a1e8740729737318862f64eb
diff --git a/trunk/net/ipv4/netfilter/ip_conntrack_core.c b/trunk/net/ipv4/netfilter/ip_conntrack_core.c
index e23e8ca476c0..babce304c619 100644
--- a/trunk/net/ipv4/netfilter/ip_conntrack_core.c
+++ b/trunk/net/ipv4/netfilter/ip_conntrack_core.c
@@ -938,6 +938,9 @@ void ip_conntrack_unexpect_related(struct ip_conntrack_expect *exp)
 	write_unlock_bh(&ip_conntrack_lock);
 }
 
+/* We don't increase the master conntrack refcount for non-fulfilled
+ * conntracks. During the conntrack destruction, the expectations are 
+ * always killed before the conntrack itself */
 struct ip_conntrack_expect *ip_conntrack_expect_alloc(struct ip_conntrack *me)
 {
 	struct ip_conntrack_expect *new;
@@ -948,17 +951,14 @@ struct ip_conntrack_expect *ip_conntrack_expect_alloc(struct ip_conntrack *me)
 		return NULL;
 	}
 	new->master = me;
-	atomic_inc(&new->master->ct_general.use);
 	atomic_set(&new->use, 1);
 	return new;
 }
 
 void ip_conntrack_expect_put(struct ip_conntrack_expect *exp)
 {
-	if (atomic_dec_and_test(&exp->use)) {
-		ip_conntrack_put(exp->master);
+	if (atomic_dec_and_test(&exp->use))
 		kmem_cache_free(ip_conntrack_expect_cachep, exp);
-	}
 }
 
 static void ip_conntrack_expect_insert(struct ip_conntrack_expect *exp)