tcp: switch rtt estimations to usec resolution

Upcoming congestion controls for TCP require usec resolution for RTT
estimations. Millisecond resolution is simply not enough these days.

FQ/pacing in DC environments also require this change for finer control
and removal of bimodal behavior due to the current hack in
tcp_update_pacing_rate() for 'small rtt'

TCP_CONG_RTT_STAMP is no longer needed.

As Julian Anastasov pointed out, we need to keep user compatibility :
tcp_metrics used to export RTT and RTTVAR in msec resolution,
so we added RTT_US and RTTVAR_US. An iproute2 patch is needed
to use the new attributes if provided by the kernel.

In this example ss command displays a srtt of 32 usecs (10Gbit link)

lpk51:~# ./ss -i dst lpk52
Netid  State      Recv-Q Send-Q   Local Address:Port       Peer
Address:Port
tcp    ESTAB      0      1         10.246.11.51:42959
10.246.11.52:64614
         cubic wscale:6,6 rto:201 rtt:0.032/0.001 ato:40 mss:1448
cwnd:10 send
3620.0Mbps pacing_rate 7240.0Mbps unacked:1 rcv_rtt:993 rcv_space:29559

Updated iproute2 ip command displays :

lpk51:~# ./ip tcp_metrics | grep 10.246.11.52
10.246.11.52 age 561.914sec cwnd 10 rtt 274us rttvar 213us source
10.246.11.51

Old binary displays :

lpk51:~# ip tcp_metrics | grep 10.246.11.52
10.246.11.52 age 561.914sec cwnd 10 rtt 250us rttvar 125us source
10.246.11.51

With help from Julian Anastasov, Stephen Hemminger and Yuchung Cheng

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Larry Brakmo <brakmo@google.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/tcp.h

@@ -201,10 +201,10 @@ struct tcp_sock {
 	u32	tlp_high_seq;	/* snd_nxt at the time of TLP retransmit. */
 
 /* RTT measurement */
-	u32	srtt;		/* smoothed round trip time << 3	*/
-	u32	mdev;		/* medium deviation			*/
-	u32	mdev_max;	/* maximal mdev for the last rtt period	*/
-	u32	rttvar;		/* smoothed mdev_max			*/
+	u32	srtt_us;	/* smoothed round trip time << 3 in usecs */
+	u32	mdev_us;	/* medium deviation			*/
+	u32	mdev_max_us;	/* maximal mdev for the last rtt period	*/
+	u32	rttvar_us;	/* smoothed mdev_max			*/
 	u32	rtt_seq;	/* sequence number to update rttvar	*/
 
 	u32	packets_out;	/* Packets which are "in flight"	*/

include/net/tcp.h

@@ -31,6 +31,7 @@
 #include <linux/crypto.h>
 #include <linux/cryptohash.h>
 #include <linux/kref.h>
+#include <linux/ktime.h>
 
 #include <net/inet_connection_sock.h>
 #include <net/inet_timewait_sock.h>
@@ -478,7 +479,6 @@ int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
 			     struct ip_options *opt);
 #ifdef CONFIG_SYN_COOKIES
-#include <linux/ktime.h>
 
 /* Syncookies use a monotonic timer which increments every 64 seconds.
  * This counter is used both as a hash input and partially encoded into
@@ -619,7 +619,7 @@ static inline void tcp_bound_rto(const struct sock *sk)
 
 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
 {
-	return (tp->srtt >> 3) + tp->rttvar;
+	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
 }
 
 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
@@ -656,6 +656,11 @@ static inline u32 tcp_rto_min(struct sock *sk)
 	return rto_min;
 }
 
+static inline u32 tcp_rto_min_us(struct sock *sk)
+{
+	return jiffies_to_usecs(tcp_rto_min(sk));
+}
+
 /* Compute the actual receive window we are currently advertising.
  * Rcv_nxt can be after the window if our peer push more data
  * than the offered window.
@@ -778,7 +783,6 @@ enum tcp_ca_event {
 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
 
 #define TCP_CONG_NON_RESTRICTED 0x1
-#define TCP_CONG_RTT_STAMP	0x2
 
 struct tcp_congestion_ops {
 	struct list_head	list;

include/uapi/linux/tcp_metrics.h

@@ -11,12 +11,15 @@
 #define TCP_METRICS_GENL_VERSION	0x1
 
 enum tcp_metric_index {
-	TCP_METRIC_RTT,
-	TCP_METRIC_RTTVAR,
+	TCP_METRIC_RTT,		/* in ms units */
+	TCP_METRIC_RTTVAR,	/* in ms units */
 	TCP_METRIC_SSTHRESH,
 	TCP_METRIC_CWND,
 	TCP_METRIC_REORDERING,
 
+	TCP_METRIC_RTT_US,	/* in usec units */
+	TCP_METRIC_RTTVAR_US,	/* in usec units */
+
 	/* Always last.  */
 	__TCP_METRIC_MAX,
 };

net/ipv4/tcp.c

@@ -387,7 +387,7 @@ void tcp_init_sock(struct sock *sk)
 	INIT_LIST_HEAD(&tp->tsq_node);
 
 	icsk->icsk_rto = TCP_TIMEOUT_INIT;
-	tp->mdev = TCP_TIMEOUT_INIT;
+	tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
 
 	/* So many TCP implementations out there (incorrectly) count the
 	 * initial SYN frame in their delayed-ACK and congestion control
@@ -2339,7 +2339,7 @@ int tcp_disconnect(struct sock *sk, int flags)
 
 	sk->sk_shutdown = 0;
 	sock_reset_flag(sk, SOCK_DONE);
-	tp->srtt = 0;
+	tp->srtt_us = 0;
 	if ((tp->write_seq += tp->max_window + 2) == 0)
 		tp->write_seq = 1;
 	icsk->icsk_backoff = 0;
@@ -2783,8 +2783,8 @@ void tcp_get_info(const struct sock *sk, struct tcp_info *info)
 
 	info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
 	info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
-	info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
-	info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
+	info->tcpi_rtt = tp->srtt_us >> 3;
+	info->tcpi_rttvar = tp->mdev_us >> 2;
 	info->tcpi_snd_ssthresh = tp->snd_ssthresh;
 	info->tcpi_snd_cwnd = tp->snd_cwnd;
 	info->tcpi_advmss = tp->advmss;

net/ipv4/tcp_cubic.c

@@ -476,10 +476,6 @@ static int __init cubictcp_register(void)
 	/* divide by bic_scale and by constant Srtt (100ms) */
 	do_div(cube_factor, bic_scale * 10);
 
-	/* hystart needs ms clock resolution */
-	if (hystart && HZ < 1000)
-		cubictcp.flags |= TCP_CONG_RTT_STAMP;
-
 	return tcp_register_congestion_control(&cubictcp);
 }
 

net/ipv4/tcp_hybla.c

@@ -21,7 +21,7 @@ struct hybla {
 	u32   rho2;	      /* Rho * Rho, integer part */
 	u32   rho_3ls;	      /* Rho parameter, <<3 */
 	u32   rho2_7ls;	      /* Rho^2, <<7	*/
-	u32   minrtt;	      /* Minimum smoothed round trip time value seen */
+	u32   minrtt_us;      /* Minimum smoothed round trip time value seen */
 };
 
 /* Hybla reference round trip time (default= 1/40 sec = 25 ms), in ms */
@@ -35,7 +35,9 @@ static inline void hybla_recalc_param (struct sock *sk)
 {
 	struct hybla *ca = inet_csk_ca(sk);
 
-	ca->rho_3ls = max_t(u32, tcp_sk(sk)->srtt / msecs_to_jiffies(rtt0), 8);
+	ca->rho_3ls = max_t(u32,
+			    tcp_sk(sk)->srtt_us / (rtt0 * USEC_PER_MSEC),
+			    8U);
 	ca->rho = ca->rho_3ls >> 3;
 	ca->rho2_7ls = (ca->rho_3ls * ca->rho_3ls) << 1;
 	ca->rho2 = ca->rho2_7ls >> 7;
@@ -59,7 +61,7 @@ static void hybla_init(struct sock *sk)
 	hybla_recalc_param(sk);
 
 	/* set minimum rtt as this is the 1st ever seen */
-	ca->minrtt = tp->srtt;
+	ca->minrtt_us = tp->srtt_us;
 	tp->snd_cwnd = ca->rho;
 }
 
@@ -94,9 +96,9 @@ static void hybla_cong_avoid(struct sock *sk, u32 ack, u32 acked,
 	int is_slowstart = 0;
 
 	/*  Recalculate rho only if this srtt is the lowest */
-	if (tp->srtt < ca->minrtt){
+	if (tp->srtt_us < ca->minrtt_us) {
 		hybla_recalc_param(sk);
-		ca->minrtt = tp->srtt;
+		ca->minrtt_us = tp->srtt_us;
 	}
 
 	if (!tcp_is_cwnd_limited(sk, in_flight))

net/ipv4/tcp_illinois.c

@@ -325,7 +325,6 @@ static void tcp_illinois_info(struct sock *sk, u32 ext,
 }
 
 static struct tcp_congestion_ops tcp_illinois __read_mostly = {
-	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_illinois_init,
 	.ssthresh	= tcp_illinois_ssthresh,
 	.cong_avoid	= tcp_illinois_cong_avoid,

net/ipv4/tcp_input.c

@@ -667,11 +667,11 @@ static void tcp_event_data_recv(struct sock *sk, struct sk_buff *skb)
  * To save cycles in the RFC 1323 implementation it was better to break
  * it up into three procedures. -- erics
  */
-static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt)
+static void tcp_rtt_estimator(struct sock *sk, long mrtt_us)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
-	long m = mrtt; /* RTT */
-	u32 srtt = tp->srtt;
+	long m = mrtt_us; /* RTT */
+	u32 srtt = tp->srtt_us;
 
 	/*	The following amusing code comes from Jacobson's
 	 *	article in SIGCOMM '88.  Note that rtt and mdev
@@ -694,7 +694,7 @@ static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt)
 		srtt += m;		/* rtt = 7/8 rtt + 1/8 new */
 		if (m < 0) {
 			m = -m;		/* m is now abs(error) */
-			m -= (tp->mdev >> 2);   /* similar update on mdev */
+			m -= (tp->mdev_us >> 2);   /* similar update on mdev */
 			/* This is similar to one of Eifel findings.
 			 * Eifel blocks mdev updates when rtt decreases.
 			 * This solution is a bit different: we use finer gain
@@ -706,28 +706,29 @@ static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt)
 			if (m > 0)
 				m >>= 3;
 		} else {
-			m -= (tp->mdev >> 2);   /* similar update on mdev */
+			m -= (tp->mdev_us >> 2);   /* similar update on mdev */
 		}
-		tp->mdev += m;	    	/* mdev = 3/4 mdev + 1/4 new */
-		if (tp->mdev > tp->mdev_max) {
-			tp->mdev_max = tp->mdev;
-			if (tp->mdev_max > tp->rttvar)
-				tp->rttvar = tp->mdev_max;
+		tp->mdev_us += m;		/* mdev = 3/4 mdev + 1/4 new */
+		if (tp->mdev_us > tp->mdev_max_us) {
+			tp->mdev_max_us = tp->mdev_us;
+			if (tp->mdev_max_us > tp->rttvar_us)
+				tp->rttvar_us = tp->mdev_max_us;
 		}
 		if (after(tp->snd_una, tp->rtt_seq)) {
-			if (tp->mdev_max < tp->rttvar)
-				tp->rttvar -= (tp->rttvar - tp->mdev_max) >> 2;
+			if (tp->mdev_max_us < tp->rttvar_us)
+				tp->rttvar_us -= (tp->rttvar_us - tp->mdev_max_us) >> 2;
 			tp->rtt_seq = tp->snd_nxt;
-			tp->mdev_max = tcp_rto_min(sk);
+			tp->mdev_max_us = tcp_rto_min_us(sk);
 		}
 	} else {
 		/* no previous measure. */
 		srtt = m << 3;		/* take the measured time to be rtt */
-		tp->mdev = m << 1;	/* make sure rto = 3*rtt */
-		tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk));
+		tp->mdev_us = m << 1;	/* make sure rto = 3*rtt */
+		tp->rttvar_us = max(tp->mdev_us, tcp_rto_min_us(sk));
+		tp->mdev_max_us = tp->rttvar_us;
 		tp->rtt_seq = tp->snd_nxt;
 	}
-	tp->srtt = max(1U, srtt);
+	tp->srtt_us = max(1U, srtt);
 }
 
 /* Set the sk_pacing_rate to allow proper sizing of TSO packets.
@@ -742,20 +743,12 @@ static void tcp_update_pacing_rate(struct sock *sk)
 	u64 rate;
 
 	/* set sk_pacing_rate to 200 % of current rate (mss * cwnd / srtt) */
-	rate = (u64)tp->mss_cache * 2 * (HZ << 3);
+	rate = (u64)tp->mss_cache * 2 * (USEC_PER_SEC << 3);
 
 	rate *= max(tp->snd_cwnd, tp->packets_out);
 
-	/* Correction for small srtt and scheduling constraints.
-	 * For small rtt, consider noise is too high, and use
-	 * the minimal value (srtt = 1 -> 125 us for HZ=1000)
-	 *
-	 * We probably need usec resolution in the future.
-	 * Note: This also takes care of possible srtt=0 case,
-	 * when tcp_rtt_estimator() was not yet called.
-	 */
-	if (tp->srtt > 8 + 2)
-		do_div(rate, tp->srtt);
+	if (likely(tp->srtt_us))
+		do_div(rate, tp->srtt_us);
 
 	/* ACCESS_ONCE() is needed because sch_fq fetches sk_pacing_rate
 	 * without any lock. We want to make sure compiler wont store
@@ -1122,10 +1115,10 @@ static bool tcp_check_dsack(struct sock *sk, const struct sk_buff *ack_skb,
 }
 
 struct tcp_sacktag_state {
-	int reord;
-	int fack_count;
-	int flag;
-	s32 rtt; /* RTT measured by SACKing never-retransmitted data */
+	int	reord;
+	int	fack_count;
+	long	rtt_us; /* RTT measured by SACKing never-retransmitted data */
+	int	flag;
 };
 
 /* Check if skb is fully within the SACK block. In presence of GSO skbs,
@@ -1186,7 +1179,8 @@ static int tcp_match_skb_to_sack(struct sock *sk, struct sk_buff *skb,
 static u8 tcp_sacktag_one(struct sock *sk,
 			  struct tcp_sacktag_state *state, u8 sacked,
 			  u32 start_seq, u32 end_seq,
-			  int dup_sack, int pcount, u32 xmit_time)
+			  int dup_sack, int pcount,
+			  const struct skb_mstamp *xmit_time)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	int fack_count = state->fack_count;
@@ -1227,8 +1221,13 @@ static u8 tcp_sacktag_one(struct sock *sk,
 				if (!after(end_seq, tp->high_seq))
 					state->flag |= FLAG_ORIG_SACK_ACKED;
 				/* Pick the earliest sequence sacked for RTT */
-				if (state->rtt < 0)
-					state->rtt = tcp_time_stamp - xmit_time;
+				if (state->rtt_us < 0) {
+					struct skb_mstamp now;
+
+					skb_mstamp_get(&now);
+					state->rtt_us = skb_mstamp_us_delta(&now,
+								xmit_time);
+				}
 			}
 
 			if (sacked & TCPCB_LOST) {
@@ -1287,7 +1286,7 @@ static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *skb,
 	 */
 	tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked,
 			start_seq, end_seq, dup_sack, pcount,
-			TCP_SKB_CB(skb)->when);
+			&skb->skb_mstamp);
 
 	if (skb == tp->lost_skb_hint)
 		tp->lost_cnt_hint += pcount;
@@ -1565,7 +1564,7 @@ static struct sk_buff *tcp_sacktag_walk(struct sk_buff *skb, struct sock *sk,
 						TCP_SKB_CB(skb)->end_seq,
 						dup_sack,
 						tcp_skb_pcount(skb),
-						TCP_SKB_CB(skb)->when);
+						&skb->skb_mstamp);
 
 			if (!before(TCP_SKB_CB(skb)->seq,
 				    tcp_highest_sack_seq(tp)))
@@ -1622,7 +1621,7 @@ static int tcp_sack_cache_ok(const struct tcp_sock *tp, const struct tcp_sack_bl
 
 static int
 tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
-			u32 prior_snd_una, s32 *sack_rtt)
+			u32 prior_snd_una, long *sack_rtt_us)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	const unsigned char *ptr = (skb_transport_header(ack_skb) +
@@ -1640,7 +1639,7 @@ tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
 
 	state.flag = 0;
 	state.reord = tp->packets_out;
-	state.rtt = -1;
+	state.rtt_us = -1L;
 
 	if (!tp->sacked_out) {
 		if (WARN_ON(tp->fackets_out))
@@ -1824,7 +1823,7 @@ out:
 	WARN_ON((int)tp->retrans_out < 0);
 	WARN_ON((int)tcp_packets_in_flight(tp) < 0);
 #endif
-	*sack_rtt = state.rtt;
+	*sack_rtt_us = state.rtt_us;
 	return state.flag;
 }
 
@@ -2034,10 +2033,12 @@ static bool tcp_pause_early_retransmit(struct sock *sk, int flag)
 	 * available, or RTO is scheduled to fire first.
 	 */
 	if (sysctl_tcp_early_retrans < 2 || sysctl_tcp_early_retrans > 3 ||
-	    (flag & FLAG_ECE) || !tp->srtt)
+	    (flag & FLAG_ECE) || !tp->srtt_us)
 		return false;
 
-	delay = max_t(unsigned long, (tp->srtt >> 5), msecs_to_jiffies(2));
+	delay = max(usecs_to_jiffies(tp->srtt_us >> 5),
+		    msecs_to_jiffies(2));
+
 	if (!time_after(inet_csk(sk)->icsk_timeout, (jiffies + delay)))
 		return false;
 
@@ -2884,7 +2885,7 @@ static void tcp_fastretrans_alert(struct sock *sk, const int acked,
 }
 
 static inline bool tcp_ack_update_rtt(struct sock *sk, const int flag,
-				      s32 seq_rtt, s32 sack_rtt)
+				      long seq_rtt_us, long sack_rtt_us)
 {
 	const struct tcp_sock *tp = tcp_sk(sk);
 
@@ -2894,10 +2895,10 @@ static inline bool tcp_ack_update_rtt(struct sock *sk, const int flag,
 	 * is acked (RFC6298).
 	 */
 	if (flag & FLAG_RETRANS_DATA_ACKED)
-		seq_rtt = -1;
+		seq_rtt_us = -1L;
 
-	if (seq_rtt < 0)
-		seq_rtt = sack_rtt;
+	if (seq_rtt_us < 0)
+		seq_rtt_us = sack_rtt_us;
 
 	/* RTTM Rule: A TSecr value received in a segment is used to
 	 * update the averaged RTT measurement only if the segment
@@ -2905,14 +2906,14 @@ static inline bool tcp_ack_update_rtt(struct sock *sk, const int flag,
 	 * left edge of the send window.
 	 * See draft-ietf-tcplw-high-performance-00, section 3.3.
 	 */
-	if (seq_rtt < 0 && tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&
+	if (seq_rtt_us < 0 && tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&
 	    flag & FLAG_ACKED)
-		seq_rtt = tcp_time_stamp - tp->rx_opt.rcv_tsecr;
+		seq_rtt_us = jiffies_to_usecs(tcp_time_stamp - tp->rx_opt.rcv_tsecr);
 
-	if (seq_rtt < 0)
+	if (seq_rtt_us < 0)
 		return false;
 
-	tcp_rtt_estimator(sk, seq_rtt);
+	tcp_rtt_estimator(sk, seq_rtt_us);
 	tcp_set_rto(sk);
 
 	/* RFC6298: only reset backoff on valid RTT measurement. */
@@ -2924,16 +2925,16 @@ static inline bool tcp_ack_update_rtt(struct sock *sk, const int flag,
 static void tcp_synack_rtt_meas(struct sock *sk, const u32 synack_stamp)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
-	s32 seq_rtt = -1;
+	long seq_rtt_us = -1L;
 
 	if (synack_stamp && !tp->total_retrans)
-		seq_rtt = tcp_time_stamp - synack_stamp;
+		seq_rtt_us = jiffies_to_usecs(tcp_time_stamp - synack_stamp);
 
 	/* If the ACK acks both the SYNACK and the (Fast Open'd) data packets
 	 * sent in SYN_RECV, SYNACK RTT is the smooth RTT computed in tcp_ack()
 	 */
-	if (!tp->srtt)
-		tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, seq_rtt, -1);
+	if (!tp->srtt_us)
+		tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, seq_rtt_us, -1L);
 }
 
 static void tcp_cong_avoid(struct sock *sk, u32 ack, u32 acked, u32 in_flight)
@@ -3022,26 +3023,27 @@ static u32 tcp_tso_acked(struct sock *sk, struct sk_buff *skb)
  * arrived at the other end.
  */
 static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
-			       u32 prior_snd_una, s32 sack_rtt)
+			       u32 prior_snd_una, long sack_rtt_us)
 {
-	struct tcp_sock *tp = tcp_sk(sk);
 	const struct inet_connection_sock *icsk = inet_csk(sk);
-	struct sk_buff *skb;
-	u32 now = tcp_time_stamp;
+	struct skb_mstamp first_ackt, last_ackt, now;
+	struct tcp_sock *tp = tcp_sk(sk);
+	u32 prior_sacked = tp->sacked_out;
+	u32 reord = tp->packets_out;
 	bool fully_acked = true;
-	int flag = 0;
+	long ca_seq_rtt_us = -1L;
+	long seq_rtt_us = -1L;
+	struct sk_buff *skb;
 	u32 pkts_acked = 0;
-	u32 reord = tp->packets_out;
-	u32 prior_sacked = tp->sacked_out;
-	s32 seq_rtt = -1;
-	s32 ca_seq_rtt = -1;
-	ktime_t last_ackt = net_invalid_timestamp();
 	bool rtt_update;
+	int flag = 0;
+
+	first_ackt.v64 = 0;
 
 	while ((skb = tcp_write_queue_head(sk)) && skb != tcp_send_head(sk)) {
 		struct tcp_skb_cb *scb = TCP_SKB_CB(skb);
-		u32 acked_pcount;
 		u8 sacked = scb->sacked;
+		u32 acked_pcount;
 
 		/* Determine how many packets and what bytes were acked, tso and else */
 		if (after(scb->end_seq, tp->snd_una)) {
@@ -3063,11 +3065,10 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 				tp->retrans_out -= acked_pcount;
 			flag |= FLAG_RETRANS_DATA_ACKED;
 		} else {
-			ca_seq_rtt = now - scb->when;
-			last_ackt = skb->tstamp;
-			if (seq_rtt < 0) {
-				seq_rtt = ca_seq_rtt;
-			}
+			last_ackt = skb->skb_mstamp;
+			if (!first_ackt.v64)
+				first_ackt = last_ackt;
+
 			if (!(sacked & TCPCB_SACKED_ACKED))
 				reord = min(pkts_acked, reord);
 			if (!after(scb->end_seq, tp->high_seq))
@@ -3113,7 +3114,13 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 	if (skb && (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
 		flag |= FLAG_SACK_RENEGING;
 
-	rtt_update = tcp_ack_update_rtt(sk, flag, seq_rtt, sack_rtt);
+	skb_mstamp_get(&now);
+	if (first_ackt.v64) {
+		seq_rtt_us = skb_mstamp_us_delta(&now, &first_ackt);
+		ca_seq_rtt_us = skb_mstamp_us_delta(&now, &last_ackt);
+	}
+
+	rtt_update = tcp_ack_update_rtt(sk, flag, seq_rtt_us, sack_rtt_us);
 
 	if (flag & FLAG_ACKED) {
 		const struct tcp_congestion_ops *ca_ops
@@ -3141,25 +3148,11 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 
 		tp->fackets_out -= min(pkts_acked, tp->fackets_out);
 
-		if (ca_ops->pkts_acked) {
-			s32 rtt_us = -1;
-
-			/* Is the ACK triggering packet unambiguous? */
-			if (!(flag & FLAG_RETRANS_DATA_ACKED)) {
-				/* High resolution needed and available? */
-				if (ca_ops->flags & TCP_CONG_RTT_STAMP &&
-				    !ktime_equal(last_ackt,
-						 net_invalid_timestamp()))
-					rtt_us = ktime_us_delta(ktime_get_real(),
-								last_ackt);
-				else if (ca_seq_rtt >= 0)
-					rtt_us = jiffies_to_usecs(ca_seq_rtt);
-			}
+		if (ca_ops->pkts_acked)
+			ca_ops->pkts_acked(sk, pkts_acked, ca_seq_rtt_us);
 
-			ca_ops->pkts_acked(sk, pkts_acked, rtt_us);
-		}
-	} else if (skb && rtt_update && sack_rtt >= 0 &&
-		   sack_rtt > (s32)(now - TCP_SKB_CB(skb)->when)) {
+	} else if (skb && rtt_update && sack_rtt_us >= 0 &&
+		   sack_rtt_us > skb_mstamp_us_delta(&now, &skb->skb_mstamp)) {
 		/* Do not re-arm RTO if the sack RTT is measured from data sent
 		 * after when the head was last (re)transmitted. Otherwise the
 		 * timeout may continue to extend in loss recovery.
@@ -3369,12 +3362,12 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 	u32 ack_seq = TCP_SKB_CB(skb)->seq;
 	u32 ack = TCP_SKB_CB(skb)->ack_seq;
 	bool is_dupack = false;
-	u32 prior_in_flight, prior_cwnd = tp->snd_cwnd, prior_rtt = tp->srtt;
+	u32 prior_in_flight;
 	u32 prior_fackets;
 	int prior_packets = tp->packets_out;
 	const int prior_unsacked = tp->packets_out - tp->sacked_out;
 	int acked = 0; /* Number of packets newly acked */
-	s32 sack_rtt = -1;
+	long sack_rtt_us = -1L;
 
 	/* If the ack is older than previous acks
 	 * then we can probably ignore it.
@@ -3432,7 +3425,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 
 		if (TCP_SKB_CB(skb)->sacked)
 			flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una,
-							&sack_rtt);
+							&sack_rtt_us);
 
 		if (TCP_ECN_rcv_ecn_echo(tp, tcp_hdr(skb)))
 			flag |= FLAG_ECE;
@@ -3451,7 +3444,8 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 
 	/* See if we can take anything off of the retransmit queue. */
 	acked = tp->packets_out;
-	flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una, sack_rtt);
+	flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una,
+				    sack_rtt_us);
 	acked -= tp->packets_out;
 
 	/* Advance cwnd if state allows */
@@ -3474,8 +3468,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 
 	if (icsk->icsk_pending == ICSK_TIME_RETRANS)
 		tcp_schedule_loss_probe(sk);
-	if (tp->srtt != prior_rtt || tp->snd_cwnd != prior_cwnd)
-		tcp_update_pacing_rate(sk);
+	tcp_update_pacing_rate(sk);
 	return 1;
 
 no_queue:
@@ -3504,7 +3497,7 @@ old_ack:
 	 */
 	if (TCP_SKB_CB(skb)->sacked) {
 		flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una,
-						&sack_rtt);
+						&sack_rtt_us);
 		tcp_fastretrans_alert(sk, acked, prior_unsacked,
 				      is_dupack, flag);
 	}

net/ipv4/tcp_ipv4.c

@@ -435,7 +435,7 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
 			break;
 
 		icsk->icsk_backoff--;
-		inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
+		inet_csk(sk)->icsk_rto = (tp->srtt_us ? __tcp_set_rto(tp) :
 			TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
 		tcp_bound_rto(sk);
 

net/ipv4/tcp_lp.c

@@ -315,7 +315,6 @@ static void tcp_lp_pkts_acked(struct sock *sk, u32 num_acked, s32 rtt_us)
 }
 
 static struct tcp_congestion_ops tcp_lp __read_mostly = {
-	.flags = TCP_CONG_RTT_STAMP,
 	.init = tcp_lp_init,
 	.ssthresh = tcp_reno_ssthresh,
 	.cong_avoid = tcp_lp_cong_avoid,

net/ipv4/tcp_metrics.c

@@ -33,6 +33,11 @@ struct tcp_fastopen_metrics {
 	struct	tcp_fastopen_cookie	cookie;
 };
 
+/* TCP_METRIC_MAX includes 2 extra fields for userspace compatibility
+ * Kernel only stores RTT and RTTVAR in usec resolution
+ */
+#define TCP_METRIC_MAX_KERNEL (TCP_METRIC_MAX - 2)
+
 struct tcp_metrics_block {
 	struct tcp_metrics_block __rcu	*tcpm_next;
 	struct inetpeer_addr		tcpm_saddr;
@@ -41,7 +46,7 @@ struct tcp_metrics_block {
 	u32				tcpm_ts;
 	u32				tcpm_ts_stamp;
 	u32				tcpm_lock;
-	u32				tcpm_vals[TCP_METRIC_MAX + 1];
+	u32				tcpm_vals[TCP_METRIC_MAX_KERNEL + 1];
 	struct tcp_fastopen_metrics	tcpm_fastopen;
 
 	struct rcu_head			rcu_head;
@@ -59,12 +64,6 @@ static u32 tcp_metric_get(struct tcp_metrics_block *tm,
 	return tm->tcpm_vals[idx];
 }
 
-static u32 tcp_metric_get_jiffies(struct tcp_metrics_block *tm,
-				  enum tcp_metric_index idx)
-{
-	return msecs_to_jiffies(tm->tcpm_vals[idx]);
-}
-
 static void tcp_metric_set(struct tcp_metrics_block *tm,
 			   enum tcp_metric_index idx,
 			   u32 val)
@@ -72,13 +71,6 @@ static void tcp_metric_set(struct tcp_metrics_block *tm,
 	tm->tcpm_vals[idx] = val;
 }
 
-static void tcp_metric_set_msecs(struct tcp_metrics_block *tm,
-				 enum tcp_metric_index idx,
-				 u32 val)
-{
-	tm->tcpm_vals[idx] = jiffies_to_msecs(val);
-}
-
 static bool addr_same(const struct inetpeer_addr *a,
 		      const struct inetpeer_addr *b)
 {
@@ -101,9 +93,11 @@ struct tcpm_hash_bucket {
 
 static DEFINE_SPINLOCK(tcp_metrics_lock);
 
-static void tcpm_suck_dst(struct tcp_metrics_block *tm, struct dst_entry *dst,
+static void tcpm_suck_dst(struct tcp_metrics_block *tm,
+			  const struct dst_entry *dst,
 			  bool fastopen_clear)
 {
+	u32 msval;
 	u32 val;
 
 	tm->tcpm_stamp = jiffies;
@@ -121,8 +115,11 @@ static void tcpm_suck_dst(struct tcp_metrics_block *tm, struct dst_entry *dst,
 		val |= 1 << TCP_METRIC_REORDERING;
 	tm->tcpm_lock = val;
 
-	tm->tcpm_vals[TCP_METRIC_RTT] = dst_metric_raw(dst, RTAX_RTT);
-	tm->tcpm_vals[TCP_METRIC_RTTVAR] = dst_metric_raw(dst, RTAX_RTTVAR);
+	msval = dst_metric_raw(dst, RTAX_RTT);
+	tm->tcpm_vals[TCP_METRIC_RTT] = msval * USEC_PER_MSEC;
+
+	msval = dst_metric_raw(dst, RTAX_RTTVAR);
+	tm->tcpm_vals[TCP_METRIC_RTTVAR] = msval * USEC_PER_MSEC;
 	tm->tcpm_vals[TCP_METRIC_SSTHRESH] = dst_metric_raw(dst, RTAX_SSTHRESH);
 	tm->tcpm_vals[TCP_METRIC_CWND] = dst_metric_raw(dst, RTAX_CWND);
 	tm->tcpm_vals[TCP_METRIC_REORDERING] = dst_metric_raw(dst, RTAX_REORDERING);
@@ -384,7 +381,7 @@ void tcp_update_metrics(struct sock *sk)
 		dst_confirm(dst);
 
 	rcu_read_lock();
-	if (icsk->icsk_backoff || !tp->srtt) {
+	if (icsk->icsk_backoff || !tp->srtt_us) {
 		/* This session failed to estimate rtt. Why?
 		 * Probably, no packets returned in time.  Reset our
 		 * results.
@@ -399,8 +396,8 @@ void tcp_update_metrics(struct sock *sk)
 	if (!tm)
 		goto out_unlock;
 
-	rtt = tcp_metric_get_jiffies(tm, TCP_METRIC_RTT);
-	m = rtt - tp->srtt;
+	rtt = tcp_metric_get(tm, TCP_METRIC_RTT);
+	m = rtt - tp->srtt_us;
 
 	/* If newly calculated rtt larger than stored one, store new
 	 * one. Otherwise, use EWMA. Remember, rtt overestimation is
@@ -408,10 +405,10 @@ void tcp_update_metrics(struct sock *sk)
 	 */
 	if (!tcp_metric_locked(tm, TCP_METRIC_RTT)) {
 		if (m <= 0)
-			rtt = tp->srtt;
+			rtt = tp->srtt_us;
 		else
 			rtt -= (m >> 3);
-		tcp_metric_set_msecs(tm, TCP_METRIC_RTT, rtt);
+		tcp_metric_set(tm, TCP_METRIC_RTT, rtt);
 	}
 
 	if (!tcp_metric_locked(tm, TCP_METRIC_RTTVAR)) {
@@ -422,16 +419,16 @@ void tcp_update_metrics(struct sock *sk)
 
 		/* Scale deviation to rttvar fixed point */
 		m >>= 1;
-		if (m < tp->mdev)
-			m = tp->mdev;
+		if (m < tp->mdev_us)
+			m = tp->mdev_us;
 
-		var = tcp_metric_get_jiffies(tm, TCP_METRIC_RTTVAR);
+		var = tcp_metric_get(tm, TCP_METRIC_RTTVAR);
 		if (m >= var)
 			var = m;
 		else
 			var -= (var - m) >> 2;
 
-		tcp_metric_set_msecs(tm, TCP_METRIC_RTTVAR, var);
+		tcp_metric_set(tm, TCP_METRIC_RTTVAR, var);
 	}
 
 	if (tcp_in_initial_slowstart(tp)) {
@@ -528,7 +525,7 @@ void tcp_init_metrics(struct sock *sk)
 		tp->reordering = val;
 	}
 
-	crtt = tcp_metric_get_jiffies(tm, TCP_METRIC_RTT);
+	crtt = tcp_metric_get(tm, TCP_METRIC_RTT);
 	rcu_read_unlock();
 reset:
 	/* The initial RTT measurement from the SYN/SYN-ACK is not ideal
@@ -551,18 +548,20 @@ reset:
 	 * to low value, and then abruptly stops to do it and starts to delay
 	 * ACKs, wait for troubles.
 	 */
-	if (crtt > tp->srtt) {
+	if (crtt > tp->srtt_us) {
 		/* Set RTO like tcp_rtt_estimator(), but from cached RTT. */
-		crtt >>= 3;
+		crtt /= 8 * USEC_PER_MSEC;
 		inet_csk(sk)->icsk_rto = crtt + max(2 * crtt, tcp_rto_min(sk));
-	} else if (tp->srtt == 0) {
+	} else if (tp->srtt_us == 0) {
 		/* RFC6298: 5.7 We've failed to get a valid RTT sample from
 		 * 3WHS. This is most likely due to retransmission,
 		 * including spurious one. Reset the RTO back to 3secs
 		 * from the more aggressive 1sec to avoid more spurious
 		 * retransmission.
 		 */
-		tp->mdev = tp->mdev_max = tp->rttvar = TCP_TIMEOUT_FALLBACK;
+		tp->rttvar_us = jiffies_to_usecs(TCP_TIMEOUT_FALLBACK);
+		tp->mdev_us = tp->mdev_max_us = tp->rttvar_us;
+
 		inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
 	}
 	/* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
@@ -809,10 +808,26 @@ static int tcp_metrics_fill_info(struct sk_buff *msg,
 		nest = nla_nest_start(msg, TCP_METRICS_ATTR_VALS);
 		if (!nest)
 			goto nla_put_failure;
-		for (i = 0; i < TCP_METRIC_MAX + 1; i++) {
-			if (!tm->tcpm_vals[i])
+		for (i = 0; i < TCP_METRIC_MAX_KERNEL + 1; i++) {
+			u32 val = tm->tcpm_vals[i];
+
+			if (!val)
 				continue;
-			if (nla_put_u32(msg, i + 1, tm->tcpm_vals[i]) < 0)
+			if (i == TCP_METRIC_RTT) {
+				if (nla_put_u32(msg, TCP_METRIC_RTT_US + 1,
+						val) < 0)
+					goto nla_put_failure;
+				n++;
+				val = max(val / 1000, 1U);
+			}
+			if (i == TCP_METRIC_RTTVAR) {
+				if (nla_put_u32(msg, TCP_METRIC_RTTVAR_US + 1,
+						val) < 0)
+					goto nla_put_failure;
+				n++;
+				val = max(val / 1000, 1U);
+			}
+			if (nla_put_u32(msg, i + 1, val) < 0)
 				goto nla_put_failure;
 			n++;
 		}

net/ipv4/tcp_minisocks.c

@@ -398,8 +398,8 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
 
 		tcp_init_wl(newtp, treq->rcv_isn);
 
-		newtp->srtt = 0;
-		newtp->mdev = TCP_TIMEOUT_INIT;
+		newtp->srtt_us = 0;
+		newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
 		newicsk->icsk_rto = TCP_TIMEOUT_INIT;
 
 		newtp->packets_out = 0;

net/ipv4/tcp_output.c

@@ -866,11 +866,7 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
 	if (clone_it) {
 		const struct sk_buff *fclone = skb + 1;
 
-		/* If congestion control is doing timestamping, we must
-		 * take such a timestamp before we potentially clone/copy.
-		 */
-		if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP)
-			__net_timestamp(skb);
+		skb_mstamp_get(&skb->skb_mstamp);
 
 		if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
 			     fclone->fclone == SKB_FCLONE_CLONE))
@@ -1974,7 +1970,7 @@ bool tcp_schedule_loss_probe(struct sock *sk)
 	struct inet_connection_sock *icsk = inet_csk(sk);
 	struct tcp_sock *tp = tcp_sk(sk);
 	u32 timeout, tlp_time_stamp, rto_time_stamp;
-	u32 rtt = tp->srtt >> 3;
+	u32 rtt = usecs_to_jiffies(tp->srtt_us >> 3);
 
 	if (WARN_ON(icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS))
 		return false;
@@ -1996,7 +1992,7 @@ bool tcp_schedule_loss_probe(struct sock *sk)
 	/* Schedule a loss probe in 2*RTT for SACK capable connections
 	 * in Open state, that are either limited by cwnd or application.
 	 */
-	if (sysctl_tcp_early_retrans < 3 || !tp->srtt || !tp->packets_out ||
+	if (sysctl_tcp_early_retrans < 3 || !tp->srtt_us || !tp->packets_out ||
 	    !tcp_is_sack(tp) || inet_csk(sk)->icsk_ca_state != TCP_CA_Open)
 		return false;
 
@@ -3050,8 +3046,9 @@ void tcp_send_delayed_ack(struct sock *sk)
 		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
 		 * directly.
 		 */
-		if (tp->srtt) {
-			int rtt = max(tp->srtt >> 3, TCP_DELACK_MIN);
+		if (tp->srtt_us) {
+			int rtt = max_t(int, usecs_to_jiffies(tp->srtt_us >> 3),
+					TCP_DELACK_MIN);
 
 			if (rtt < max_ato)
 				max_ato = rtt;

net/ipv4/tcp_probe.c

@@ -154,7 +154,7 @@ static void jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 			p->snd_wnd = tp->snd_wnd;
 			p->rcv_wnd = tp->rcv_wnd;
 			p->ssthresh = tcp_current_ssthresh(sk);
-			p->srtt = tp->srtt >> 3;
+			p->srtt = tp->srtt_us >> 3;
 
 			tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
 		}

net/ipv4/tcp_vegas.c

@@ -306,7 +306,6 @@ void tcp_vegas_get_info(struct sock *sk, u32 ext, struct sk_buff *skb)
 EXPORT_SYMBOL_GPL(tcp_vegas_get_info);
 
 static struct tcp_congestion_ops tcp_vegas __read_mostly = {
-	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_vegas_init,
 	.ssthresh	= tcp_reno_ssthresh,
 	.cong_avoid	= tcp_vegas_cong_avoid,

net/ipv4/tcp_veno.c

@@ -203,7 +203,6 @@ static u32 tcp_veno_ssthresh(struct sock *sk)
 }
 
 static struct tcp_congestion_ops tcp_veno __read_mostly = {
-	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_veno_init,
 	.ssthresh	= tcp_veno_ssthresh,
 	.cong_avoid	= tcp_veno_cong_avoid,

net/ipv4/tcp_yeah.c

@@ -227,7 +227,6 @@ static u32 tcp_yeah_ssthresh(struct sock *sk) {
 }
 
 static struct tcp_congestion_ops tcp_yeah __read_mostly = {
-	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_yeah_init,
 	.ssthresh	= tcp_yeah_ssthresh,
 	.cong_avoid	= tcp_yeah_cong_avoid,