Merge branch 'tcp_skb_cb'

Eric Dumazet says:

====================
tcp: better TCP_SKB_CB layout

TCP had the assumption that IPCB and IP6CB are first members of skb->cb[]

This is fine, except that IPCB/IP6CB are used in TCP for a very short time
in input path.

What really matters for TCP stack is to get skb->next,
TCP_SKB_CB(skb)->seq, and TCP_SKB_CB(skb)->end_seq in the same cache line.

skb that are immediately consumed do not care because whole skb->cb[] is
hot in cpu cache, while skb that sit in wocket write queue or receive queues
do not need TCP_SKB_CB(skb)->header at all.

This patch set implements the prereq for IPv4, IPv6, and TCP to make this
possible. This makes TCP more efficient.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/ip.h

@@ -180,8 +180,10 @@ static inline __u8 ip_reply_arg_flowi_flags(const struct ip_reply_arg *arg)
 	return (arg->flags & IP_REPLY_ARG_NOSRCCHECK) ? FLOWI_FLAG_ANYSRC : 0;
 }
 
-void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
-			   __be32 saddr, const struct ip_reply_arg *arg,
+void ip_send_unicast_reply(struct net *net, struct sk_buff *skb,
+			   const struct ip_options *sopt,
+			   __be32 daddr, __be32 saddr,
+			   const struct ip_reply_arg *arg,
 			   unsigned int len);
 
 #define IP_INC_STATS(net, field)	SNMP_INC_STATS64((net)->mib.ip_statistics, field)
@@ -511,7 +513,14 @@ int ip_forward(struct sk_buff *skb);
  
 void ip_options_build(struct sk_buff *skb, struct ip_options *opt,
 		      __be32 daddr, struct rtable *rt, int is_frag);
-int ip_options_echo(struct ip_options *dopt, struct sk_buff *skb);
+
+int __ip_options_echo(struct ip_options *dopt, struct sk_buff *skb,
+		      const struct ip_options *sopt);
+static inline int ip_options_echo(struct ip_options *dopt, struct sk_buff *skb)
+{
+	return __ip_options_echo(dopt, skb, &IPCB(skb)->opt);
+}
+
 void ip_options_fragment(struct sk_buff *skb);
 int ip_options_compile(struct net *net, struct ip_options *opt,
 		       struct sk_buff *skb);

include/net/ipv6.h

@@ -288,7 +288,8 @@ struct ipv6_txoptions *ipv6_renew_options(struct sock *sk,
 struct ipv6_txoptions *ipv6_fixup_options(struct ipv6_txoptions *opt_space,
 					  struct ipv6_txoptions *opt);
 
-bool ipv6_opt_accepted(const struct sock *sk, const struct sk_buff *skb);
+bool ipv6_opt_accepted(const struct sock *sk, const struct sk_buff *skb,
+		       const struct inet6_skb_parm *opt);
 
 static inline bool ipv6_accept_ra(struct inet6_dev *idev)
 {

include/net/tcp.h

@@ -696,12 +696,6 @@ static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
  */
 struct tcp_skb_cb {
-	union {
-		struct inet_skb_parm	h4;
-#if IS_ENABLED(CONFIG_IPV6)
-		struct inet6_skb_parm	h6;
-#endif
-	} header;	/* For incoming frames		*/
 	__u32		seq;		/* Starting sequence number	*/
 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
 	__u32		tcp_tw_isn;	/* isn chosen by tcp_timewait_state_process() */
@@ -720,6 +714,12 @@ struct tcp_skb_cb {
 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
 	/* 1 byte hole */
 	__u32		ack_seq;	/* Sequence number ACK'd	*/
+	union {
+		struct inet_skb_parm	h4;
+#if IS_ENABLED(CONFIG_IPV6)
+		struct inet6_skb_parm	h6;
+#endif
+	} header;	/* For incoming frames		*/
 };
 
 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))

net/dccp/ipv6.c

@@ -404,7 +404,7 @@ static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 	ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 	ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 
-	if (ipv6_opt_accepted(sk, skb) ||
+	if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
 	    np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
 	    np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
 		atomic_inc(&skb->users);

net/ipv4/ip_options.c

@@ -87,17 +87,15 @@ void ip_options_build(struct sk_buff *skb, struct ip_options *opt,
  * NOTE: dopt cannot point to skb.
  */
 
-int ip_options_echo(struct ip_options *dopt, struct sk_buff *skb)
+int __ip_options_echo(struct ip_options *dopt, struct sk_buff *skb,
+		      const struct ip_options *sopt)
 {
-	const struct ip_options *sopt;
 	unsigned char *sptr, *dptr;
 	int soffset, doffset;
 	int	optlen;
 
 	memset(dopt, 0, sizeof(struct ip_options));
 
-	sopt = &(IPCB(skb)->opt);
-
 	if (sopt->optlen == 0)
 		return 0;
 

net/ipv4/ip_output.c

@@ -1522,8 +1522,10 @@ static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
 	.uc_ttl		= -1,
 };
 
-void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
-			   __be32 saddr, const struct ip_reply_arg *arg,
+void ip_send_unicast_reply(struct net *net, struct sk_buff *skb,
+			   const struct ip_options *sopt,
+			   __be32 daddr, __be32 saddr,
+			   const struct ip_reply_arg *arg,
 			   unsigned int len)
 {
 	struct ip_options_data replyopts;
@@ -1534,7 +1536,7 @@ void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
 	struct sock *sk;
 	struct inet_sock *inet;
 
-	if (ip_options_echo(&replyopts.opt.opt, skb))
+	if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
 		return;
 
 	ipc.addr = daddr;

net/ipv4/tcp_ipv4.c

@@ -681,8 +681,9 @@ static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
 
 	net = dev_net(skb_dst(skb)->dev);
 	arg.tos = ip_hdr(skb)->tos;
-	ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
-			      ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
+	ip_send_unicast_reply(net, skb, &TCP_SKB_CB(skb)->header.h4.opt,
+			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
+			      &arg, arg.iov[0].iov_len);
 
 	TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 	TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
@@ -764,8 +765,9 @@ static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 	if (oif)
 		arg.bound_dev_if = oif;
 	arg.tos = tos;
-	ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
-			      ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
+	ip_send_unicast_reply(net, skb, &TCP_SKB_CB(skb)->header.h4.opt,
+			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
+			      &arg, arg.iov[0].iov_len);
 
 	TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 }
@@ -884,18 +886,16 @@ EXPORT_SYMBOL(tcp_syn_flood_action);
  */
 static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
 {
-	const struct ip_options *opt = &(IPCB(skb)->opt);
+	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
 	struct ip_options_rcu *dopt = NULL;
 
 	if (opt && opt->optlen) {
 		int opt_size = sizeof(*dopt) + opt->optlen;
 
 		dopt = kmalloc(opt_size, GFP_ATOMIC);
-		if (dopt) {
-			if (ip_options_echo(&dopt->opt, skb)) {
-				kfree(dopt);
-				dopt = NULL;
-			}
+		if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
+			kfree(dopt);
+			dopt = NULL;
 		}
 	}
 	return dopt;
@@ -1429,7 +1429,7 @@ static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 
 #ifdef CONFIG_SYN_COOKIES
 	if (!th->syn)
-		sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
+		sk = cookie_v4_check(sk, skb, &TCP_SKB_CB(skb)->header.h4.opt);
 #endif
 	return sk;
 }
@@ -1634,6 +1634,13 @@ int tcp_v4_rcv(struct sk_buff *skb)
 
 	th = tcp_hdr(skb);
 	iph = ip_hdr(skb);
+	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
+	 * barrier() makes sure compiler wont play fool^Waliasing games.
+	 */
+	memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
+		sizeof(struct inet_skb_parm));
+	barrier();
+
 	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
 	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
 				    skb->len - th->doff * 4);

net/ipv4/tcp_output.c

@@ -974,6 +974,11 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
 
 	/* Our usage of tstamp should remain private */
 	skb->tstamp.tv64 = 0;
+
+	/* Cleanup our debris for IP stacks */
+	memset(skb->cb, 0, max(sizeof(struct inet_skb_parm),
+			       sizeof(struct inet6_skb_parm)));
+
 	err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
 
 	if (likely(err <= 0))

net/ipv6/af_inet6.c

@@ -672,10 +672,10 @@ int inet6_sk_rebuild_header(struct sock *sk)
 }
 EXPORT_SYMBOL_GPL(inet6_sk_rebuild_header);
 
-bool ipv6_opt_accepted(const struct sock *sk, const struct sk_buff *skb)
+bool ipv6_opt_accepted(const struct sock *sk, const struct sk_buff *skb,
+		       const struct inet6_skb_parm *opt)
 {
 	const struct ipv6_pinfo *np = inet6_sk(sk);
-	const struct inet6_skb_parm *opt = IP6CB(skb);
 
 	if (np->rxopt.all) {
 		if ((opt->hop && (np->rxopt.bits.hopopts ||

net/ipv6/syncookies.c

@@ -203,7 +203,7 @@ struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)
 	ireq->ir_num = ntohs(th->dest);
 	ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 	ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
-	if (ipv6_opt_accepted(sk, skb) ||
+	if (ipv6_opt_accepted(sk, skb, &TCP_SKB_CB(skb)->header.h6) ||
 	    np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
 	    np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
 		atomic_inc(&skb->users);

net/ipv6/tcp_ipv6.c

@@ -742,7 +742,8 @@ static void tcp_v6_init_req(struct request_sock *req, struct sock *sk,
 		ireq->ir_iif = inet6_iif(skb);
 
 	if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
-	    (ipv6_opt_accepted(sk, skb) || np->rxopt.bits.rxinfo ||
+	    (ipv6_opt_accepted(sk, skb, &TCP_SKB_CB(skb)->header.h6) ||
+	     np->rxopt.bits.rxinfo ||
 	     np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
 	     np->rxopt.bits.rxohlim || np->repflow)) {
 		atomic_inc(&skb->users);
@@ -1367,7 +1368,7 @@ ipv6_pktoptions:
 			np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
 		if (np->repflow)
 			np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
-		if (ipv6_opt_accepted(sk, opt_skb)) {
+		if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
 			skb_set_owner_r(opt_skb, sk);
 			opt_skb = xchg(&np->pktoptions, opt_skb);
 		} else {
@@ -1411,6 +1412,13 @@ static int tcp_v6_rcv(struct sk_buff *skb)
 
 	th = tcp_hdr(skb);
 	hdr = ipv6_hdr(skb);
+	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
+	 * barrier() makes sure compiler wont play fool^Waliasing games.
+	 */
+	memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
+		sizeof(struct inet6_skb_parm));
+	barrier();
+
 	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
 	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
 				    skb->len - th->doff*4);