ipv4: Early TCP socket demux.

Input packet processing for local sockets involves two major demuxes.
One for the route and one for the socket.

But we can optimize this down to one demux for certain kinds of local
sockets.

Currently we only do this for established TCP sockets, but it could
at least in theory be expanded to other kinds of connections.

If a TCP socket is established then it's identity is fully specified.

This means that whatever input route was used during the three-way
handshake must work equally well for the rest of the connection since
the keys will not change.

Once we move to established state, we cache the receive packet's input
route to use later.

Like the existing cached route in sk->sk_dst_cache used for output
packets, we have to check for route invalidations using dst->obsolete
and dst->ops->check().

Early demux occurs outside of a socket locked section, so when a route
invalidation occurs we defer the fixup of sk->sk_rx_dst until we are
actually inside of established state packet processing and thus have
the socket locked.
Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/inet_hashtables.h

@@ -379,10 +379,10 @@ static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
 					     const __be16 sport,
 					     const __be16 dport)
 {
-	struct sock *sk;
+	struct sock *sk = skb_steal_sock(skb);
 	const struct iphdr *iph = ip_hdr(skb);
 
-	if (unlikely(sk = skb_steal_sock(skb)))
+	if (sk)
 		return sk;
 	else
 		return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,

include/net/protocol.h

@@ -37,6 +37,7 @@
 
 /* This is used to register protocols. */
 struct net_protocol {
+	int			(*early_demux)(struct sk_buff *skb);
 	int			(*handler)(struct sk_buff *skb);
 	void			(*err_handler)(struct sk_buff *skb, u32 info);
 	int			(*gso_send_check)(struct sk_buff *skb);

include/net/sock.h

@@ -319,6 +319,7 @@ struct sock {
 	unsigned long 		sk_flags;
 	struct dst_entry	*sk_dst_cache;
 	spinlock_t		sk_dst_lock;
+	struct dst_entry	*sk_rx_dst;
 	atomic_t		sk_wmem_alloc;
 	atomic_t		sk_omem_alloc;
 	int			sk_sndbuf;
@@ -1426,6 +1427,7 @@ extern struct sk_buff		*sock_rmalloc(struct sock *sk,
 					      gfp_t priority);
 extern void			sock_wfree(struct sk_buff *skb);
 extern void			sock_rfree(struct sk_buff *skb);
+extern void			sock_edemux(struct sk_buff *skb);
 
 extern int			sock_setsockopt(struct socket *sock, int level,
 						int op, char __user *optval,

include/net/tcp.h

@@ -325,6 +325,7 @@ extern void tcp_v4_err(struct sk_buff *skb, u32);
 
 extern void tcp_shutdown (struct sock *sk, int how);
 
+extern int tcp_v4_early_demux(struct sk_buff *skb);
 extern int tcp_v4_rcv(struct sk_buff *skb);
 
 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk);

net/core/sock.c

@@ -1465,6 +1465,11 @@ void sock_rfree(struct sk_buff *skb)
 }
 EXPORT_SYMBOL(sock_rfree);
 
+void sock_edemux(struct sk_buff *skb)
+{
+	sock_put(skb->sk);
+}
+EXPORT_SYMBOL(sock_edemux);
 
 int sock_i_uid(struct sock *sk)
 {

net/ipv4/af_inet.c

@@ -157,6 +157,7 @@ void inet_sock_destruct(struct sock *sk)
 
 	kfree(rcu_dereference_protected(inet->inet_opt, 1));
 	dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
+	dst_release(sk->sk_rx_dst);
 	sk_refcnt_debug_dec(sk);
 }
 EXPORT_SYMBOL(inet_sock_destruct);
@@ -1518,6 +1519,7 @@ static const struct net_protocol igmp_protocol = {
 #endif
 
 static const struct net_protocol tcp_protocol = {
+	.early_demux	=	tcp_v4_early_demux,
 	.handler	=	tcp_v4_rcv,
 	.err_handler	=	tcp_v4_err,
 	.gso_send_check	=	tcp_v4_gso_send_check,

net/ipv4/ip_input.c

@@ -323,7 +323,19 @@ static int ip_rcv_finish(struct sk_buff *skb)
 	 *	how the packet travels inside Linux networking.
 	 */
 	if (skb_dst(skb) == NULL) {
-		int err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
+		const struct net_protocol *ipprot;
+		int protocol = iph->protocol;
+		int err;
+
+		rcu_read_lock();
+		ipprot = rcu_dereference(inet_protos[protocol]);
+		err = -ENOENT;
+		if (ipprot && ipprot->early_demux)
+			err = ipprot->early_demux(skb);
+		rcu_read_unlock();
+
+		if (err) {
+			err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
 						   iph->tos, skb->dev);
 			if (unlikely(err)) {
 				if (err == -EHOSTUNREACH)
@@ -338,6 +350,7 @@ static int ip_rcv_finish(struct sk_buff *skb)
 				goto drop;
 			}
 		}
+	}
 
 #ifdef CONFIG_IP_ROUTE_CLASSID
 	if (unlikely(skb_dst(skb)->tclassid)) {

net/ipv4/tcp_input.c

@@ -5518,6 +5518,18 @@ int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 	struct tcp_sock *tp = tcp_sk(sk);
 	int res;
 
+	if (sk->sk_rx_dst) {
+		struct dst_entry *dst = sk->sk_rx_dst;
+		if (unlikely(dst->obsolete)) {
+			if (dst->ops->check(dst, 0) == NULL) {
+				dst_release(dst);
+				sk->sk_rx_dst = NULL;
+			}
+		}
+	}
+	if (unlikely(sk->sk_rx_dst == NULL))
+		sk->sk_rx_dst = dst_clone(skb_dst(skb));
+
 	/*
 	 *	Header prediction.
 	 *	The code loosely follows the one in the famous
@@ -5729,8 +5741,10 @@ void tcp_finish_connect(struct sock *sk, struct sk_buff *skb)
 
 	tcp_set_state(sk, TCP_ESTABLISHED);
 
-	if (skb != NULL)
+	if (skb != NULL) {
+		sk->sk_rx_dst = dst_clone(skb_dst(skb));
 		security_inet_conn_established(sk, skb);
+	}
 
 	/* Make sure socket is routed, for correct metrics.  */
 	icsk->icsk_af_ops->rebuild_header(sk);

net/ipv4/tcp_ipv4.c

@@ -1671,6 +1671,52 @@ int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 }
 EXPORT_SYMBOL(tcp_v4_do_rcv);
 
+int tcp_v4_early_demux(struct sk_buff *skb)
+{
+	struct net *net = dev_net(skb->dev);
+	const struct iphdr *iph;
+	const struct tcphdr *th;
+	struct sock *sk;
+	int err;
+
+	err = -ENOENT;
+	if (skb->pkt_type != PACKET_HOST)
+		goto out_err;
+
+	if (!pskb_may_pull(skb, ip_hdrlen(skb) + sizeof(struct tcphdr)))
+		goto out_err;
+
+	iph = ip_hdr(skb);
+	th = (struct tcphdr *) ((char *)iph + ip_hdrlen(skb));
+
+	if (th->doff < sizeof(struct tcphdr) / 4)
+		goto out_err;
+
+	if (!pskb_may_pull(skb, ip_hdrlen(skb) + th->doff * 4))
+		goto out_err;
+
+	sk = __inet_lookup_established(net, &tcp_hashinfo,
+				       iph->saddr, th->source,
+				       iph->daddr, th->dest,
+				       skb->dev->ifindex);
+	if (sk) {
+		skb->sk = sk;
+		skb->destructor = sock_edemux;
+		if (sk->sk_state != TCP_TIME_WAIT) {
+			struct dst_entry *dst = sk->sk_rx_dst;
+			if (dst)
+				dst = dst_check(dst, 0);
+			if (dst) {
+				skb_dst_set_noref(skb, dst);
+				err = 0;
+			}
+		}
+	}
+
+out_err:
+	return err;
+}
+
 /*
  *	From tcp_input.c
  */

net/ipv4/tcp_minisocks.c

@@ -445,6 +445,8 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
 		struct tcp_sock *oldtp = tcp_sk(sk);
 		struct tcp_cookie_values *oldcvp = oldtp->cookie_values;
 
+		newsk->sk_rx_dst = dst_clone(skb_dst(skb));
+
 		/* TCP Cookie Transactions require space for the cookie pair,
 		 * as it differs for each connection.  There is no need to
 		 * copy any s_data_payload stored at the original socket.