tcp/dccp: do not touch listener sk_refcnt under synflood

When a SYNFLOOD targets a non SO_REUSEPORT listener, multiple
cpus contend on sk->sk_refcnt and sk->sk_wmem_alloc changes.

By letting listeners use SOCK_RCU_FREE infrastructure,
we can relax TCP_LISTEN lookup rules and avoid touching sk_refcnt

Note that we still use SLAB_DESTROY_BY_RCU rules for other sockets,
only listeners are impacted by this change.

Peak performance under SYNFLOOD is increased by ~33% :

On my test machine, I could process 3.2 Mpps instead of 2.4 Mpps

Most consuming functions are now skb_set_owner_w() and sock_wfree()
contending on sk->sk_wmem_alloc when cooking SYNACK and freeing them.

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

1 files changed, 25 insertions, 48 deletions

diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
index 387338d71dcd..98ba03b6f87d 100644
--- a/net/ipv4/inet_hashtables.c
+++ b/net/ipv4/inet_hashtables.c
@@ -198,13 +198,13 @@ static inline int compute_score(struct sock *sk, struct net *net,
 }
 
 /*
- * Don't inline this cruft. Here are some nice properties to exploit here. The
- * BSD API does not allow a listening sock to specify the remote port nor the
+ * Here are some nice properties to exploit here. The BSD API
+ * does not allow a listening sock to specify the remote port nor the
  * remote address for the connection. So always assume those are both
  * wildcarded during the search since they can never be otherwise.
  */
 
-
+/* called with rcu_read_lock() : No refcount taken on the socket */
 struct sock *__inet_lookup_listener(struct net *net,
 				    struct inet_hashinfo *hashinfo,
 				    struct sk_buff *skb, int doff,
@@ -212,37 +212,27 @@ struct sock *__inet_lookup_listener(struct net *net,
 				    const __be32 daddr, const unsigned short hnum,
 				    const int dif)
 {
-	struct sock *sk, *result;
-	struct hlist_nulls_node *node;
 	unsigned int hash = inet_lhashfn(net, hnum);
 	struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
-	int score, hiscore, matches = 0, reuseport = 0;
-	bool select_ok = true;
+	int score, hiscore = 0, matches = 0, reuseport = 0;
+	struct sock *sk, *result = NULL;
 	u32 phash = 0;
 
-begin:
-	result = NULL;
-	hiscore = 0;
-	sk_nulls_for_each_rcu(sk, node, &ilb->head) {
+	sk_for_each_rcu(sk, &ilb->head) {
 		score = compute_score(sk, net, hnum, daddr, dif);
 		if (score > hiscore) {
-			result = sk;
-			hiscore = score;
 			reuseport = sk->sk_reuseport;
 			if (reuseport) {
 				phash = inet_ehashfn(net, daddr, hnum,
 						     saddr, sport);
-				if (select_ok) {
-					struct sock *sk2;
-					sk2 = reuseport_select_sock(sk, phash,
-								    skb, doff);
-					if (sk2) {
-						result = sk2;
-						goto found;
-					}
-				}
+				result = reuseport_select_sock(sk, phash,
+							       skb, doff);
+				if (result)
+					return result;
 				matches = 1;
 			}
+			result = sk;
+			hiscore = score;
 		} else if (score == hiscore && reuseport) {
 			matches++;
 			if (reciprocal_scale(phash, matches) == 0)
@@ -250,24 +240,6 @@ begin:
 			phash = next_pseudo_random32(phash);
 		}
 	}
-	/*
-	 * if the nulls value we got at the end of this lookup is
-	 * not the expected one, we must restart lookup.
-	 * We probably met an item that was moved to another chain.
-	 */
-	if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
-		goto begin;
-	if (result) {
-found:
-		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
-			result = NULL;
-		else if (unlikely(compute_score(result, net, hnum, daddr,
-				  dif) < hiscore)) {
-			sock_put(result);
-			select_ok = false;
-			goto begin;
-		}
-	}
 	return result;
 }
 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
@@ -508,7 +480,8 @@ int __inet_hash(struct sock *sk, struct sock *osk,
 		if (err)
 			goto unlock;
 	}
-	__sk_nulls_add_node_rcu(sk, &ilb->head);
+	hlist_add_head_rcu(&sk->sk_node, &ilb->head);
+	sock_set_flag(sk, SOCK_RCU_FREE);
 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 unlock:
 	spin_unlock(&ilb->lock);
@@ -535,20 +508,25 @@ void inet_unhash(struct sock *sk)
 {
 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 	spinlock_t *lock;
+	bool listener = false;
 	int done;
 
 	if (sk_unhashed(sk))
 		return;
 
-	if (sk->sk_state == TCP_LISTEN)
+	if (sk->sk_state == TCP_LISTEN) {
 		lock = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)].lock;
-	else
+		listener = true;
+	} else {
 		lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
-
+	}
 	spin_lock_bh(lock);
 	if (rcu_access_pointer(sk->sk_reuseport_cb))
 		reuseport_detach_sock(sk);
-	done = __sk_nulls_del_node_init_rcu(sk);
+	if (listener)
+		done = __sk_del_node_init(sk);
+	else
+		done = __sk_nulls_del_node_init_rcu(sk);
 	if (done)
 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 	spin_unlock_bh(lock);
@@ -684,9 +662,8 @@ void inet_hashinfo_init(struct inet_hashinfo *h)
 
 	for (i = 0; i < INET_LHTABLE_SIZE; i++) {
 		spin_lock_init(&h->listening_hash[i].lock);
-		INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].head,
-				      i + LISTENING_NULLS_BASE);
-		}
+		INIT_HLIST_HEAD(&h->listening_hash[i].head);
+	}
 }
 EXPORT_SYMBOL_GPL(inet_hashinfo_init);