Merge branch 'tcp-add-tos-reflection-feature'

Wei Wang says:

====================
tcp: add tos reflection feature

This patch series adds a new tcp feature to reflect TOS value received in
SYN, and send it out in SYN-ACK, and eventually set the TOS value of the
established socket with this reflected TOS value. This provides a way to
set the traffic class/QoS level for all traffic in the same connection
to be the same as the incoming SYN. It could be useful for datacenters
to provide equivalent QoS according to the incoming request.
This feature is guarded by /proc/sys/net/ipv4/tcp_reflect_tos, and is by
default turned off.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/tcp.h

@@ -134,6 +134,7 @@ struct tcp_request_sock {
 						  * FastOpen it's the seq#
 						  * after data-in-SYN.
 						  */
+	u8				syn_tos;
 };
 
 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)

include/net/ip.h

@@ -151,7 +151,7 @@ int igmp_mc_init(void);
 
 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
 			  __be32 saddr, __be32 daddr,
-			  struct ip_options_rcu *opt);
+			  struct ip_options_rcu *opt, u8 tos);
 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
 	   struct net_device *orig_dev);
 void ip_list_rcv(struct list_head *head, struct packet_type *pt,

include/net/netns/ipv4.h

@@ -183,6 +183,7 @@ struct netns_ipv4 {
 	unsigned int sysctl_tcp_fastopen_blackhole_timeout;
 	atomic_t tfo_active_disable_times;
 	unsigned long tfo_active_disable_stamp;
+	int sysctl_tcp_reflect_tos;
 
 	int sysctl_udp_wmem_min;
 	int sysctl_udp_rmem_min;

net/dccp/ipv4.c

@@ -495,7 +495,8 @@ static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req
 		rcu_read_lock();
 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 					    ireq->ir_rmt_addr,
-					    rcu_dereference(ireq->ireq_opt));
+					    rcu_dereference(ireq->ireq_opt),
+					    inet_sk(sk)->tos);
 		rcu_read_unlock();
 		err = net_xmit_eval(err);
 	}
@@ -537,7 +538,8 @@ static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 	local_bh_disable();
 	bh_lock_sock(ctl_sk);
 	err = ip_build_and_send_pkt(skb, ctl_sk,
-				    rxiph->daddr, rxiph->saddr, NULL);
+				    rxiph->daddr, rxiph->saddr, NULL,
+				    inet_sk(ctl_sk)->tos);
 	bh_unlock_sock(ctl_sk);
 
 	if (net_xmit_eval(err) == 0) {

net/ipv4/ip_output.c

@@ -142,7 +142,8 @@ static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
  *
  */
 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
-			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
+			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
+			  u8 tos)
 {
 	struct inet_sock *inet = inet_sk(sk);
 	struct rtable *rt = skb_rtable(skb);
@@ -155,7 +156,7 @@ int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
 	iph = ip_hdr(skb);
 	iph->version  = 4;
 	iph->ihl      = 5;
-	iph->tos      = inet->tos;
+	iph->tos      = tos;
 	iph->ttl      = ip_select_ttl(inet, &rt->dst);
 	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
 	iph->saddr    = saddr;

net/ipv4/syncookies.c

@@ -286,11 +286,10 @@ struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
 					    struct sock *sk,
 					    struct sk_buff *skb)
 {
+	struct tcp_request_sock *treq;
 	struct request_sock *req;
 
 #ifdef CONFIG_MPTCP
-	struct tcp_request_sock *treq;
-
 	if (sk_is_mptcp(sk))
 		ops = &mptcp_subflow_request_sock_ops;
 #endif
@@ -299,8 +298,9 @@ struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
 	if (!req)
 		return NULL;
 
-#if IS_ENABLED(CONFIG_MPTCP)
 	treq = tcp_rsk(req);
+	treq->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
+#if IS_ENABLED(CONFIG_MPTCP)
 	treq->is_mptcp = sk_is_mptcp(sk);
 	if (treq->is_mptcp) {
 		int err = mptcp_subflow_init_cookie_req(req, sk, skb);

net/ipv4/sysctl_net_ipv4.c

@@ -1329,6 +1329,15 @@ static struct ctl_table ipv4_net_table[] = {
 		.extra1		= SYSCTL_ZERO,
 		.extra2		= &comp_sack_nr_max,
 	},
+	{
+		.procname       = "tcp_reflect_tos",
+		.data           = &init_net.ipv4.sysctl_tcp_reflect_tos,
+		.maxlen         = sizeof(int),
+		.mode           = 0644,
+		.proc_handler   = proc_dointvec_minmax,
+		.extra1         = SYSCTL_ZERO,
+		.extra2         = SYSCTL_ONE,
+	},
 	{
 		.procname	= "udp_rmem_min",
 		.data		= &init_net.ipv4.sysctl_udp_rmem_min,

net/ipv4/tcp_input.c

@@ -6834,6 +6834,7 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops,
 
 	tcp_rsk(req)->snt_isn = isn;
 	tcp_rsk(req)->txhash = net_tx_rndhash();
+	tcp_rsk(req)->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
 	tcp_openreq_init_rwin(req, sk, dst);
 	sk_rx_queue_set(req_to_sk(req), skb);
 	if (!want_cookie) {

net/ipv4/tcp_ipv4.c

@@ -972,6 +972,7 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 	struct flowi4 fl4;
 	int err = -1;
 	struct sk_buff *skb;
+	u8 tos;
 
 	/* First, grab a route. */
 	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
@@ -979,13 +980,17 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 
 	skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
 
+	tos = sock_net(sk)->ipv4.sysctl_tcp_reflect_tos ?
+			tcp_rsk(req)->syn_tos : inet_sk(sk)->tos;
+
 	if (skb) {
 		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
 
 		rcu_read_lock();
 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 					    ireq->ir_rmt_addr,
-					    rcu_dereference(ireq->ireq_opt));
+					    rcu_dereference(ireq->ireq_opt),
+					    tos & ~INET_ECN_MASK);
 		rcu_read_unlock();
 		err = net_xmit_eval(err);
 	}
@@ -1530,6 +1535,10 @@ struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
 		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 	newinet->inet_id = prandom_u32();
 
+	/* Set ToS of the new socket based upon the value of incoming SYN. */
+	if (sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)
+		newinet->tos = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
+
 	if (!dst) {
 		dst = inet_csk_route_child_sock(sk, newsk, req);
 		if (!dst)

net/ipv6/tcp_ipv6.c

@@ -510,6 +510,7 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 	struct flowi6 *fl6 = &fl->u.ip6;
 	struct sk_buff *skb;
 	int err = -ENOMEM;
+	u8 tclass;
 
 	/* First, grab a route. */
 	if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
@@ -528,9 +529,12 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 
 		rcu_read_lock();
 		opt = ireq->ipv6_opt;
+		tclass = sock_net(sk)->ipv4.sysctl_tcp_reflect_tos ?
+				tcp_rsk(req)->syn_tos : np->tclass;
 		if (!opt)
 			opt = rcu_dereference(np->opt);
-		err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass,
+		err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt,
+			       tclass & ~INET_ECN_MASK,
 			       sk->sk_priority);
 		rcu_read_unlock();
 		err = net_xmit_eval(err);
@@ -1310,6 +1314,10 @@ static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *
 	if (np->repflow)
 		newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
 
+	/* Set ToS of the new socket based upon the value of incoming SYN. */
+	if (sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)
+		newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
+
 	/* Clone native IPv6 options from listening socket (if any)
 
 	   Yes, keeping reference count would be much more clever,