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, 24 insertions, 16 deletions

diff --git a/include/net/inet_hashtables.h b/include/net/inet_hashtables.h
index a77acee93aaf..0574493e3899 100644
--- a/include/net/inet_hashtables.h
+++ b/include/net/inet_hashtables.h
@@ -100,14 +100,10 @@ struct inet_bind_hashbucket {
 
 /*
  * Sockets can be hashed in established or listening table
- * We must use different 'nulls' end-of-chain value for listening
- * hash table, or we might find a socket that was closed and
- * reallocated/inserted into established hash table
  */
-#define LISTENING_NULLS_BASE (1U << 29)
 struct inet_listen_hashbucket {
 	spinlock_t		lock;
-	struct hlist_nulls_head	head;
+	struct hlist_head	head;
 };
 
 /* This is for listening sockets, thus all sockets which possess wildcards. */
@@ -304,14 +300,20 @@ static inline struct sock *__inet_lookup(struct net *net,
 					 struct sk_buff *skb, int doff,
 					 const __be32 saddr, const __be16 sport,
 					 const __be32 daddr, const __be16 dport,
-					 const int dif)
+					 const int dif,
+					 bool *refcounted)
 {
 	u16 hnum = ntohs(dport);
-	struct sock *sk = __inet_lookup_established(net, hashinfo,
-				saddr, sport, daddr, hnum, dif);
+	struct sock *sk;
 
-	return sk ? : __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
-					     sport, daddr, hnum, dif);
+	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
+				       daddr, hnum, dif);
+	*refcounted = true;
+	if (sk)
+		return sk;
+	*refcounted = false;
+	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
+				      sport, daddr, hnum, dif);
 }
 
 static inline struct sock *inet_lookup(struct net *net,
@@ -322,10 +324,13 @@ static inline struct sock *inet_lookup(struct net *net,
 				       const int dif)
 {
 	struct sock *sk;
+	bool refcounted;
 
 	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
-			   dport, dif);
+			   dport, dif, &refcounted);
 
+	if (sk && !refcounted && !atomic_inc_not_zero(&sk->sk_refcnt))
+		sk = NULL;
 	return sk;
 }
 
@@ -333,17 +338,20 @@ static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
 					     struct sk_buff *skb,
 					     int doff,
 					     const __be16 sport,
-					     const __be16 dport)
+					     const __be16 dport,
+					     bool *refcounted)
 {
 	struct sock *sk = skb_steal_sock(skb);
 	const struct iphdr *iph = ip_hdr(skb);
 
+	*refcounted = true;
 	if (sk)
 		return sk;
-	else
-		return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
-				     doff, iph->saddr, sport,
-				     iph->daddr, dport, inet_iif(skb));
+
+	return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
+			     doff, iph->saddr, sport,
+			     iph->daddr, dport, inet_iif(skb),
+			     refcounted);
 }
 
 u32 sk_ehashfn(const struct sock *sk);