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 <[e-mail]>
Signed-off-by: David S. Miller <[e-mail]>

include/net/inet6_hashtables.h

···
 					  const __be16 sport,
 					  const struct in6_addr *daddr,
 					  const u16 hnum,
-					  const int dif)
+					  const int dif,
+					  bool *refcounted)
 {
 	struct sock *sk = __inet6_lookup_established(net, hashinfo, saddr,
 						sport, daddr, hnum, dif);
+	*refcounted = true;
 	if (sk)
 		return sk;
-
+	*refcounted = false;
 	return inet6_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
 				     daddr, hnum, dif);
 }
···
 					      struct sk_buff *skb, int doff,
 					      const __be16 sport,
 					      const __be16 dport,
-					      int iif)
+					      int iif,
+					      bool *refcounted)
 {
 	struct sock *sk = skb_steal_sock(skb);
 
+	*refcounted = true;
 	if (sk)
 		return sk;
 
 	return __inet6_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
 			      doff, &ipv6_hdr(skb)->saddr, sport,
 			      &ipv6_hdr(skb)->daddr, ntohs(dport),
-			      iif);
+			      iif, refcounted);
 }
 
 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,

include/net/inet_hashtables.h

···
 
 /*
  * 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. */
···
 					 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,
···
 				       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;
 }
 
···
 					     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);

net/dccp/ipv4.c

···
 {
 	const struct dccp_hdr *dh;
 	const struct iphdr *iph;
+	bool refcounted;
 	struct sock *sk;
 	int min_cov;
 
···
 
 lookup:
 	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
-			       dh->dccph_sport, dh->dccph_dport);
+			       dh->dccph_sport, dh->dccph_dport, &refcounted);
 	if (!sk) {
 		dccp_pr_debug("failed to look up flow ID in table and "
 			      "get corresponding socket\n");
···
 			goto lookup;
 		}
 		sock_hold(sk);
+		refcounted = true;
 		nsk = dccp_check_req(sk, skb, req);
 		if (!nsk) {
 			reqsk_put(req);
···
 	return 0;
 
 discard_and_relse:
-	sock_put(sk);
+	if (refcounted)
+		sock_put(sk);
 	goto discard_it;
 }
 

net/dccp/ipv6.c

···
 static int dccp_v6_rcv(struct sk_buff *skb)
 {
 	const struct dccp_hdr *dh;
+	bool refcounted;
 	struct sock *sk;
 	int min_cov;
 
···
 lookup:
 	sk = __inet6_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
 			        dh->dccph_sport, dh->dccph_dport,
-				inet6_iif(skb));
+				inet6_iif(skb), &refcounted);
 	if (!sk) {
 		dccp_pr_debug("failed to look up flow ID in table and "
 			      "get corresponding socket\n");
···
 			goto lookup;
 		}
 		sock_hold(sk);
+		refcounted = true;
 		nsk = dccp_check_req(sk, skb, req);
 		if (!nsk) {
 			reqsk_put(req);
···
 	return 0;
 
 discard_and_relse:
-	sock_put(sk);
+	if (refcounted)
+		sock_put(sk);
 	goto discard_it;
 }
 

net/ipv4/inet_diag.c

···
 
 		for (i = s_i; i < INET_LHTABLE_SIZE; i++) {
 			struct inet_listen_hashbucket *ilb;
-			struct hlist_nulls_node *node;
 			struct sock *sk;
 
 			num = 0;
 			ilb = &hashinfo->listening_hash[i];
 			spin_lock_bh(&ilb->lock);
-			sk_nulls_for_each(sk, node, &ilb->head) {
+			sk_for_each(sk, &ilb->head) {
 				struct inet_sock *inet = inet_sk(sk);
 
 				if (!net_eq(sock_net(sk), net))

net/ipv4/inet_hashtables.c

···
 }
 
 /*
- * 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,
···
 				    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)
 				result = sk;
 			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;
···
 		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);
···
 {
 	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);
···
 
 	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);
 

net/ipv4/tcp_ipv4.c

···
 
 	net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 #ifdef CONFIG_TCP_MD5SIG
+	rcu_read_lock();
 	hash_location = tcp_parse_md5sig_option(th);
 	if (sk && sk_fullsock(sk)) {
 		key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
···
 					     ntohs(th->source), inet_iif(skb));
 		/* don't send rst if it can't find key */
 		if (!sk1)
-			return;
-		rcu_read_lock();
+			goto out;
+
 		key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
 					&ip_hdr(skb)->saddr, AF_INET);
 		if (!key)
-			goto release_sk1;
+			goto out;
+
 
 		genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
 		if (genhash || memcmp(hash_location, newhash, 16) != 0)
-			goto release_sk1;
+			goto out;
+
 	}
 
 	if (key) {
···
 	TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 
 #ifdef CONFIG_TCP_MD5SIG
-release_sk1:
-	if (sk1) {
-		rcu_read_unlock();
-		sock_put(sk1);
-	}
+out:
+	rcu_read_unlock();
 #endif
 }
 
···
 
 int tcp_v4_rcv(struct sk_buff *skb)
 {
+	struct net *net = dev_net(skb->dev);
 	const struct iphdr *iph;
 	const struct tcphdr *th;
+	bool refcounted;
 	struct sock *sk;
 	int ret;
-	struct net *net = dev_net(skb->dev);
 
 	if (skb->pkt_type != PACKET_HOST)
 		goto discard_it;
···
 
 lookup:
 	sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
-			       th->dest);
+			       th->dest, &refcounted);
 	if (!sk)
 		goto no_tcp_socket;
 
···
 			inet_csk_reqsk_queue_drop_and_put(sk, req);
 			goto lookup;
 		}
+		/* We own a reference on the listener, increase it again
+		 * as we might lose it too soon.
+		 */
 		sock_hold(sk);
+		refcounted = true;
 		nsk = tcp_check_req(sk, skb, req, false);
 		if (!nsk) {
 			reqsk_put(req);
···
 	bh_unlock_sock(sk);
 
 put_and_return:
-	sock_put(sk);
+	if (refcounted)
+		sock_put(sk);
 
 	return ret;
 
···
 	return 0;
 
 discard_and_relse:
-	sock_put(sk);
+	if (refcounted)
+		sock_put(sk);
 	goto discard_it;
 
 do_time_wait:
···
 		if (sk2) {
 			inet_twsk_deschedule_put(inet_twsk(sk));
 			sk = sk2;
+			refcounted = false;
 			goto process;
 		}
 		/* Fall through to ACK */
···
  */
 static void *listening_get_next(struct seq_file *seq, void *cur)
 {
-	struct inet_connection_sock *icsk;
-	struct hlist_nulls_node *node;
-	struct sock *sk = cur;
-	struct inet_listen_hashbucket *ilb;
 	struct tcp_iter_state *st = seq->private;
 	struct net *net = seq_file_net(seq);
+	struct inet_listen_hashbucket *ilb;
+	struct inet_connection_sock *icsk;
+	struct sock *sk = cur;
 
 	if (!sk) {
+get_head:
 		ilb = &tcp_hashinfo.listening_hash[st->bucket];
 		spin_lock_bh(&ilb->lock);
-		sk = sk_nulls_head(&ilb->head);
+		sk = sk_head(&ilb->head);
 		st->offset = 0;
 		goto get_sk;
 	}
···
 	++st->num;
 	++st->offset;
 
-	sk = sk_nulls_next(sk);
+	sk = sk_next(sk);
 get_sk:
-	sk_nulls_for_each_from(sk, node) {
+	sk_for_each_from(sk) {
 		if (!net_eq(sock_net(sk), net))
 			continue;
-		if (sk->sk_family == st->family) {
-			cur = sk;
-			goto out;
-		}
+		if (sk->sk_family == st->family)
+			return sk;
 		icsk = inet_csk(sk);
 	}
 	spin_unlock_bh(&ilb->lock);
 	st->offset = 0;
-	if (++st->bucket < INET_LHTABLE_SIZE) {
-		ilb = &tcp_hashinfo.listening_hash[st->bucket];
-		spin_lock_bh(&ilb->lock);
-		sk = sk_nulls_head(&ilb->head);
-		goto get_sk;
-	}
-	cur = NULL;
-out:
-	return cur;
+	if (++st->bucket < INET_LHTABLE_SIZE)
+		goto get_head;
+	return NULL;
 }
 
 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)

net/ipv6/inet6_hashtables.c

···
 	return score;
 }
 
+/* called with rcu_read_lock() */
 struct sock *inet6_lookup_listener(struct net *net,
 		struct inet_hashinfo *hashinfo,
 		struct sk_buff *skb, int doff,
···
 		const __be16 sport, const struct in6_addr *daddr,
 		const unsigned short hnum, const int dif)
 {
-	struct sock *sk;
-	const struct hlist_nulls_node *node;
-	struct sock *result;
-	int score, hiscore, matches = 0, reuseport = 0;
-	bool select_ok = true;
-	u32 phash = 0;
 	unsigned int hash = inet_lhashfn(net, hnum);
 	struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
+	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(sk, node, &ilb->head) {
+	sk_for_each(sk, &ilb->head) {
 		score = compute_score(sk, net, hnum, daddr, dif);
 		if (score > hiscore) {
 			hiscore = score;
-			result = sk;
-			reuseport = sk->sk_reuseport;
 			if (reuseport) {
 				phash = inet6_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;
+			reuseport = sk->sk_reuseport;
 		} else if (score == hiscore && reuseport) {
 			matches++;
 			if (reciprocal_scale(phash, matches) == 0)
 				result = sk;
 			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;
···
 			  const int dif)
 {
 	struct sock *sk;
+	bool refcounted;
 
 	sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
-			    ntohs(dport), dif);
-
+			    ntohs(dport), dif, &refcounted);
+	if (sk && !refcounted && !atomic_inc_not_zero(&sk->sk_refcnt))
+		sk = NULL;
 	return sk;
 }
 EXPORT_SYMBOL_GPL(inet6_lookup);

net/ipv6/tcp_ipv6.c

···
 		return;
 
 #ifdef CONFIG_TCP_MD5SIG
+	rcu_read_lock();
 	hash_location = tcp_parse_md5sig_option(th);
 	if (sk && sk_fullsock(sk)) {
 		key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
···
 					   th->source, &ipv6h->daddr,
 					   ntohs(th->source), tcp_v6_iif(skb));
 		if (!sk1)
-			return;
+			goto out;
 
-		rcu_read_lock();
 		key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 		if (!key)
-			goto release_sk1;
+			goto out;
 
 		genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
 		if (genhash || memcmp(hash_location, newhash, 16) != 0)
-			goto release_sk1;
+			goto out;
 	}
 #endif
 
···
 	tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
 
 #ifdef CONFIG_TCP_MD5SIG
-release_sk1:
-	if (sk1) {
-		rcu_read_unlock();
-		sock_put(sk1);
-	}
+out:
+	rcu_read_unlock();
 #endif
 }
 
···
 {
 	const struct tcphdr *th;
 	const struct ipv6hdr *hdr;
+	bool refcounted;
 	struct sock *sk;
 	int ret;
 	struct net *net = dev_net(skb->dev);
···
 
 lookup:
 	sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
-				th->source, th->dest, inet6_iif(skb));
+				th->source, th->dest, inet6_iif(skb),
+				&refcounted);
 	if (!sk)
 		goto no_tcp_socket;
 
···
 			goto lookup;
 		}
 		sock_hold(sk);
+		refcounted = true;
 		nsk = tcp_check_req(sk, skb, req, false);
 		if (!nsk) {
 			reqsk_put(req);
···
 	bh_unlock_sock(sk);
 
 put_and_return:
-	sock_put(sk);
+	if (refcounted)
+		sock_put(sk);
 	return ret ? -1 : 0;
 
 no_tcp_socket:
···
 	return 0;
 
 discard_and_relse:
-	sock_put(sk);
+	if (refcounted)
+		sock_put(sk);
 	goto discard_it;
 
 do_time_wait:
···
 			inet_twsk_deschedule_put(tw);
 			sk = sk2;
 			tcp_v6_restore_cb(skb);
+			refcounted = false;
 			goto process;
 		}
 		/* Fall through to ACK */

net/netfilter/xt_socket.c

···
 {
 	switch (protocol) {
 	case IPPROTO_TCP:
-		return __inet_lookup(net, &tcp_hashinfo, skb, doff,
-				     saddr, sport, daddr, dport,
-				     in->ifindex);
+		return inet_lookup(net, &tcp_hashinfo, skb, doff,
+				   saddr, sport, daddr, dport,
+				   in->ifindex);
 	case IPPROTO_UDP:
 		return udp4_lib_lookup(net, saddr, sport, daddr, dport,
 				       in->ifindex);