Merge branch 'tcp-ts-usec-resolution'

Eric Dumazet says:

====================
tcp: add optional usec resolution to TCP TS

As discussed in various public places in 2016, Google adopted
usec resolution in RFC 7323 TS values, at Van Jacobson suggestion.

Goals were :

1) better observability of delays in networking stacks/fabrics.

2) better disambiguation of events based on TSval/ecr values.

3) building block for congestion control modules needing usec resolution.

Back then we implemented a schem based on private SYN options
to safely negotiate the feature.

For upstream submission, we chose to use a much simpler route
attribute because this feature is probably going to be used
in private networks.

ip route add 10/8 ... features tcp_usec_ts

References:

https://www.ietf.org/proceedings/97/slides/slides-97-tcpm-tcp-options-for-low-latency-00.pdf
https://datatracker.ietf.org/doc/draft-wang-tcpm-low-latency-opt/

First two patches are fixing old minor bugs and might be taken
by stable teams (thanks to appropriate Fixes: tags)
====================
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/chelsio/inline_crypto/chtls/chtls_cm.c

@@ -2259,7 +2259,7 @@ static void chtls_rx_ack(struct sock *sk, struct sk_buff *skb)
 
 		if (tp->snd_una != snd_una) {
 			tp->snd_una = snd_una;
-			tp->rcv_tstamp = tcp_time_stamp(tp);
+			tp->rcv_tstamp = tcp_jiffies32;
 			if (tp->snd_una == tp->snd_nxt &&
 			    !csk_flag_nochk(csk, CSK_TX_FAILOVER))
 				csk_reset_flag(csk, CSK_TX_WAIT_IDLE);

include/linux/tcp.h

@@ -152,6 +152,7 @@ struct tcp_request_sock {
 	u64				snt_synack; /* first SYNACK sent time */
 	bool				tfo_listener;
 	bool				is_mptcp;
+	s8				req_usec_ts;
 #if IS_ENABLED(CONFIG_MPTCP)
 	bool				drop_req;
 #endif
@@ -257,7 +258,8 @@ struct tcp_sock {
 	u8	compressed_ack;
 	u8	dup_ack_counter:2,
 		tlp_retrans:1,	/* TLP is a retransmission */
-		unused:5;
+		tcp_usec_ts:1, /* TSval values in usec */
+		unused:4;
 	u32	chrono_start;	/* Start time in jiffies of a TCP chrono */
 	u32	chrono_stat[3];	/* Time in jiffies for chrono_stat stats */
 	u8	chrono_type:2,	/* current chronograph type */
@@ -576,4 +578,9 @@ void tcp_sock_set_quickack(struct sock *sk, int val);
 int tcp_sock_set_syncnt(struct sock *sk, int val);
 int tcp_sock_set_user_timeout(struct sock *sk, int val);
 
+static inline bool dst_tcp_usec_ts(const struct dst_entry *dst)
+{
+	return dst_feature(dst, RTAX_FEATURE_TCP_USEC_TS);
+}
+
 #endif	/* _LINUX_TCP_H */

include/net/inet_timewait_sock.h

@@ -67,7 +67,8 @@ struct inet_timewait_sock {
 	/* And these are ours. */
 	unsigned int		tw_transparent  : 1,
 				tw_flowlabel	: 20,
-				tw_pad		: 3,	/* 3 bits hole */
+				tw_usec_ts	: 1,
+				tw_pad		: 2,	/* 2 bits hole */
 				tw_tos		: 8;
 	u32			tw_txhash;
 	u32			tw_priority;

include/net/tcp.h

@@ -166,7 +166,12 @@ static_assert((1 << ATO_BITS) > TCP_DELACK_MAX);
 #define MAX_TCP_KEEPCNT		127
 #define MAX_TCP_SYNCNT		127
 
-#define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
+/* Ensure that TCP PAWS checks are relaxed after ~2147 seconds
+ * to avoid overflows. This assumes a clock smaller than 1 Mhz.
+ * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz.
+ */
+#define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC)
+
 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
 					 * after this time. It should be equal
 					 * (or greater than) TCP_TIMEWAIT_LEN
@@ -798,22 +803,31 @@ static inline u64 tcp_clock_us(void)
 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
 }
 
-/* This should only be used in contexts where tp->tcp_mstamp is up to date */
-static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
+static inline u64 tcp_clock_ms(void)
 {
-	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
+	return div_u64(tcp_clock_ns(), NSEC_PER_MSEC);
 }
 
-/* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
-static inline u32 tcp_ns_to_ts(u64 ns)
+/* TCP Timestamp included in TS option (RFC 1323) can either use ms
+ * or usec resolution. Each socket carries a flag to select one or other
+ * resolution, as the route attribute could change anytime.
+ * Each flow must stick to initial resolution.
+ */
+static inline u32 tcp_clock_ts(bool usec_ts)
 {
-	return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
+	return usec_ts ? tcp_clock_us() : tcp_clock_ms();
 }
 
-/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
-static inline u32 tcp_time_stamp_raw(void)
+static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
 {
-	return tcp_ns_to_ts(tcp_clock_ns());
+	return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
+}
+
+static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp)
+{
+	if (tp->tcp_usec_ts)
+		return tp->tcp_mstamp;
+	return tcp_time_stamp_ms(tp);
 }
 
 void tcp_mstamp_refresh(struct tcp_sock *tp);
@@ -823,17 +837,30 @@ static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
 	return max_t(s64, t1 - t0, 0);
 }
 
-static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
-{
-	return tcp_ns_to_ts(skb->skb_mstamp_ns);
-}
-
 /* provide the departure time in us unit */
 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
 {
 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
 }
 
+/* Provide skb TSval in usec or ms unit */
+static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb)
+{
+	if (usec_ts)
+		return tcp_skb_timestamp_us(skb);
+
+	return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC);
+}
+
+static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw)
+{
+	return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset;
+}
+
+static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq)
+{
+	return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off;
+}
 
 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
 
@@ -1599,7 +1626,7 @@ static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
 		return true;
 	if (unlikely(!time_before32(ktime_get_seconds(),
-				    rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
+				    rx_opt->ts_recent_stamp + TCP_PAWS_WRAP)))
 		return true;
 	/*
 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,

include/uapi/linux/rtnetlink.h

@@ -503,12 +503,16 @@ enum {
 #define RTAX_MAX (__RTAX_MAX - 1)
 
 #define RTAX_FEATURE_ECN		(1 << 0)
-#define RTAX_FEATURE_SACK	(1 << 1)
-#define RTAX_FEATURE_TIMESTAMP	(1 << 2)
+#define RTAX_FEATURE_SACK		(1 << 1) /* unused */
+#define RTAX_FEATURE_TIMESTAMP		(1 << 2) /* unused */
 #define RTAX_FEATURE_ALLFRAG		(1 << 3)
+#define RTAX_FEATURE_TCP_USEC_TS	(1 << 4)
 
-#define RTAX_FEATURE_MASK	(RTAX_FEATURE_ECN | RTAX_FEATURE_SACK | \
-				 RTAX_FEATURE_TIMESTAMP | RTAX_FEATURE_ALLFRAG)
+#define RTAX_FEATURE_MASK	(RTAX_FEATURE_ECN |		\
+				 RTAX_FEATURE_SACK |		\
+				 RTAX_FEATURE_TIMESTAMP |	\
+				 RTAX_FEATURE_ALLFRAG |		\
+				 RTAX_FEATURE_TCP_USEC_TS)
 
 struct rta_session {
 	__u8	proto;

include/uapi/linux/tcp.h

@@ -170,6 +170,7 @@ enum tcp_fastopen_client_fail {
 #define TCPI_OPT_ECN		8 /* ECN was negociated at TCP session init */
 #define TCPI_OPT_ECN_SEEN	16 /* we received at least one packet with ECT */
 #define TCPI_OPT_SYN_DATA	32 /* SYN-ACK acked data in SYN sent or rcvd */
+#define TCPI_OPT_USEC_TS	64 /* usec timestamps */
 
 /*
  * Sender's congestion state indicating normal or abnormal situations

net/ipv4/syncookies.c

@@ -41,7 +41,6 @@ static siphash_aligned_key_t syncookie_secret[2];
  * requested/supported by the syn/synack exchange.
  */
 #define TSBITS	6
-#define TSMASK	(((__u32)1 << TSBITS) - 1)
 
 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
 		       u32 count, int c)
@@ -52,6 +51,14 @@ static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
 			    count, &syncookie_secret[c]);
 }
 
+/* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */
+static u64 tcp_ns_to_ts(bool usec_ts, u64 val)
+{
+	if (usec_ts)
+		return div_u64(val, NSEC_PER_USEC);
+
+	return div_u64(val, NSEC_PER_MSEC);
+}
 
 /*
  * when syncookies are in effect and tcp timestamps are enabled we encode
@@ -62,27 +69,24 @@ static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
  */
 u64 cookie_init_timestamp(struct request_sock *req, u64 now)
 {
-	struct inet_request_sock *ireq;
-	u32 ts, ts_now = tcp_ns_to_ts(now);
+	const struct inet_request_sock *ireq = inet_rsk(req);
+	u64 ts, ts_now = tcp_ns_to_ts(false, now);
 	u32 options = 0;
 
-	ireq = inet_rsk(req);
-
 	options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
 	if (ireq->sack_ok)
 		options |= TS_OPT_SACK;
 	if (ireq->ecn_ok)
 		options |= TS_OPT_ECN;
 
-	ts = ts_now & ~TSMASK;
-	ts |= options;
-	if (ts > ts_now) {
-		ts >>= TSBITS;
-		ts--;
-		ts <<= TSBITS;
+	ts = (ts_now >> TSBITS) << TSBITS;
 	ts |= options;
-	}
-	return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
+	if (ts > ts_now)
+		ts -= (1UL << TSBITS);
+
+	if (tcp_rsk(req)->req_usec_ts)
+		return ts * NSEC_PER_USEC;
+	return ts * NSEC_PER_MSEC;
 }
 
 
@@ -302,6 +306,8 @@ struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
 	treq->af_specific = af_ops;
 
 	treq->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
+	treq->req_usec_ts = -1;
+
 #if IS_ENABLED(CONFIG_MPTCP)
 	treq->is_mptcp = sk_is_mptcp(sk);
 	if (treq->is_mptcp) {

net/ipv4/tcp.c

@@ -3629,10 +3629,16 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,
 			tp->fastopen_no_cookie = val;
 		break;
 	case TCP_TIMESTAMP:
-		if (!tp->repair)
+		if (!tp->repair) {
 			err = -EPERM;
-		else
-			WRITE_ONCE(tp->tsoffset, val - tcp_time_stamp_raw());
+			break;
+		}
+		/* val is an opaque field,
+		 * and low order bit contains usec_ts enable bit.
+		 * Its a best effort, and we do not care if user makes an error.
+		 */
+		tp->tcp_usec_ts = val & 1;
+		WRITE_ONCE(tp->tsoffset, val - tcp_clock_ts(tp->tcp_usec_ts));
 		break;
 	case TCP_REPAIR_WINDOW:
 		err = tcp_repair_set_window(tp, optval, optlen);
@@ -3754,6 +3760,8 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
 		info->tcpi_options |= TCPI_OPT_ECN_SEEN;
 	if (tp->syn_data_acked)
 		info->tcpi_options |= TCPI_OPT_SYN_DATA;
+	if (tp->tcp_usec_ts)
+		info->tcpi_options |= TCPI_OPT_USEC_TS;
 
 	info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
 	info->tcpi_ato = jiffies_to_usecs(min_t(u32, icsk->icsk_ack.ato,
@@ -3817,10 +3825,8 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
 	info->tcpi_total_rto = tp->total_rto;
 	info->tcpi_total_rto_recoveries = tp->total_rto_recoveries;
 	info->tcpi_total_rto_time = tp->total_rto_time;
-	if (tp->rto_stamp) {
-		info->tcpi_total_rto_time += tcp_time_stamp_raw() -
-						tp->rto_stamp;
-	}
+	if (tp->rto_stamp)
+		info->tcpi_total_rto_time += tcp_clock_ms() - tp->rto_stamp;
 
 	unlock_sock_fast(sk, slow);
 }
@@ -4145,7 +4151,11 @@ int do_tcp_getsockopt(struct sock *sk, int level,
 		break;
 
 	case TCP_TIMESTAMP:
-		val = tcp_time_stamp_raw() + READ_ONCE(tp->tsoffset);
+		val = tcp_clock_ts(tp->tcp_usec_ts) + READ_ONCE(tp->tsoffset);
+		if (tp->tcp_usec_ts)
+			val |= 1;
+		else
+			val &= ~1;
 		break;
 	case TCP_NOTSENT_LOWAT:
 		val = READ_ONCE(tp->notsent_lowat);

net/ipv4/tcp_input.c

@@ -693,6 +693,23 @@ static inline void tcp_rcv_rtt_measure(struct tcp_sock *tp)
 	tp->rcv_rtt_est.time = tp->tcp_mstamp;
 }
 
+static s32 tcp_rtt_tsopt_us(const struct tcp_sock *tp)
+{
+	u32 delta, delta_us;
+
+	delta = tcp_time_stamp_ts(tp) - tp->rx_opt.rcv_tsecr;
+	if (tp->tcp_usec_ts)
+		return delta;
+
+	if (likely(delta < INT_MAX / (USEC_PER_SEC / TCP_TS_HZ))) {
+		if (!delta)
+			delta = 1;
+		delta_us = delta * (USEC_PER_SEC / TCP_TS_HZ);
+		return delta_us;
+	}
+	return -1;
+}
+
 static inline void tcp_rcv_rtt_measure_ts(struct sock *sk,
 					  const struct sk_buff *skb)
 {
@@ -704,15 +721,10 @@ static inline void tcp_rcv_rtt_measure_ts(struct sock *sk,
 
 	if (TCP_SKB_CB(skb)->end_seq -
 	    TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss) {
-		u32 delta = tcp_time_stamp(tp) - tp->rx_opt.rcv_tsecr;
-		u32 delta_us;
+		s32 delta = tcp_rtt_tsopt_us(tp);
 
-		if (likely(delta < INT_MAX / (USEC_PER_SEC / TCP_TS_HZ))) {
-			if (!delta)
-				delta = 1;
-			delta_us = delta * (USEC_PER_SEC / TCP_TS_HZ);
-			tcp_rcv_rtt_update(tp, delta_us, 0);
-		}
+		if (delta >= 0)
+			tcp_rcv_rtt_update(tp, delta, 0);
 	}
 }
 
@@ -2442,7 +2454,7 @@ static bool tcp_skb_spurious_retrans(const struct tcp_sock *tp,
 				     const struct sk_buff *skb)
 {
 	return (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS) &&
-	       tcp_tsopt_ecr_before(tp, tcp_skb_timestamp(skb));
+	       tcp_tsopt_ecr_before(tp, tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb));
 }
 
 /* Nothing was retransmitted or returned timestamp is less
@@ -2856,7 +2868,7 @@ void tcp_enter_recovery(struct sock *sk, bool ece_ack)
 static void tcp_update_rto_time(struct tcp_sock *tp)
 {
 	if (tp->rto_stamp) {
-		tp->total_rto_time += tcp_time_stamp(tp) - tp->rto_stamp;
+		tp->total_rto_time += tcp_time_stamp_ms(tp) - tp->rto_stamp;
 		tp->rto_stamp = 0;
 	}
 }
@@ -3146,17 +3158,10 @@ static 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_us < 0 && tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&
-	    flag & FLAG_ACKED) {
-		u32 delta = tcp_time_stamp(tp) - 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_us = ca_rtt_us = tcp_rtt_tsopt_us(tp);
 
-		if (likely(delta < INT_MAX / (USEC_PER_SEC / TCP_TS_HZ))) {
-			if (!delta)
-				delta = 1;
-			seq_rtt_us = delta * (USEC_PER_SEC / TCP_TS_HZ);
-			ca_rtt_us = seq_rtt_us;
-		}
-	}
 	rs->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet (or -1) */
 	if (seq_rtt_us < 0)
 		return false;
@@ -6293,7 +6298,7 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
 
 		if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&
 		    !between(tp->rx_opt.rcv_tsecr, tp->retrans_stamp,
-			     tcp_time_stamp(tp))) {
+			     tcp_time_stamp_ts(tp))) {
 			NET_INC_STATS(sock_net(sk),
 					LINUX_MIB_PAWSACTIVEREJECTED);
 			goto reset_and_undo;
@@ -7042,6 +7047,7 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops,
 	req->syncookie = want_cookie;
 	tcp_rsk(req)->af_specific = af_ops;
 	tcp_rsk(req)->ts_off = 0;
+	tcp_rsk(req)->req_usec_ts = -1;
 #if IS_ENABLED(CONFIG_MPTCP)
 	tcp_rsk(req)->is_mptcp = 0;
 #endif

net/ipv4/tcp_ipv4.c

@@ -296,6 +296,7 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 		rt = NULL;
 		goto failure;
 	}
+	tp->tcp_usec_ts = dst_tcp_usec_ts(&rt->dst);
 	/* OK, now commit destination to socket.  */
 	sk->sk_gso_type = SKB_GSO_TCPV4;
 	sk_setup_caps(sk, &rt->dst);
@@ -954,7 +955,7 @@ static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 	tcp_v4_send_ack(sk, skb,
 			tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
-			tcp_time_stamp_raw() + tcptw->tw_ts_offset,
+			tcp_tw_tsval(tcptw),
 			tcptw->tw_ts_recent,
 			tw->tw_bound_dev_if,
 			tcp_twsk_md5_key(tcptw),
@@ -988,7 +989,7 @@ static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 	tcp_v4_send_ack(sk, skb, seq,
 			tcp_rsk(req)->rcv_nxt,
 			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
-			tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
+			tcp_rsk_tsval(tcp_rsk(req)),
 			READ_ONCE(req->ts_recent),
 			0,
 			tcp_md5_do_lookup(sk, l3index, addr, AF_INET),

net/ipv4/tcp_lp.c

@@ -272,7 +272,7 @@ static void tcp_lp_pkts_acked(struct sock *sk, const struct ack_sample *sample)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct lp *lp = inet_csk_ca(sk);
-	u32 now = tcp_time_stamp(tp);
+	u32 now = tcp_time_stamp_ts(tp);
 	u32 delta;
 
 	if (sample->rtt_us > 0)

net/ipv4/tcp_minisocks.c

@@ -300,6 +300,7 @@ void tcp_time_wait(struct sock *sk, int state, int timeo)
 		tcptw->tw_ts_recent	= tp->rx_opt.ts_recent;
 		tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
 		tcptw->tw_ts_offset	= tp->tsoffset;
+		tw->tw_usec_ts		= tp->tcp_usec_ts;
 		tcptw->tw_last_oow_ack_time = 0;
 		tcptw->tw_tx_delay	= tp->tcp_tx_delay;
 		tw->tw_txhash		= sk->sk_txhash;
@@ -554,22 +555,30 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
 	newtp->max_window = newtp->snd_wnd;
 
 	if (newtp->rx_opt.tstamp_ok) {
+		newtp->tcp_usec_ts = treq->req_usec_ts;
 		newtp->rx_opt.ts_recent = READ_ONCE(req->ts_recent);
 		newtp->rx_opt.ts_recent_stamp = ktime_get_seconds();
 		newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
 	} else {
+		newtp->tcp_usec_ts = 0;
 		newtp->rx_opt.ts_recent_stamp = 0;
 		newtp->tcp_header_len = sizeof(struct tcphdr);
 	}
 	if (req->num_timeout) {
+		newtp->total_rto = req->num_timeout;
 		newtp->undo_marker = treq->snt_isn;
+		if (newtp->tcp_usec_ts) {
+			newtp->retrans_stamp = treq->snt_synack;
+			newtp->total_rto_time = (u32)(tcp_clock_us() -
+						      newtp->retrans_stamp) / USEC_PER_MSEC;
+		} else {
 			newtp->retrans_stamp = div_u64(treq->snt_synack,
 						       USEC_PER_SEC / TCP_TS_HZ);
-		newtp->total_rto = req->num_timeout;
-		newtp->total_rto_recoveries = 1;
-		newtp->total_rto_time = tcp_time_stamp_raw() -
+			newtp->total_rto_time = tcp_clock_ms() -
 						newtp->retrans_stamp;
 		}
+		newtp->total_rto_recoveries = 1;
+	}
 	newtp->tsoffset = treq->ts_off;
 #ifdef CONFIG_TCP_MD5SIG
 	newtp->md5sig_info = NULL;	/*XXX*/

net/ipv4/tcp_output.c

@@ -799,7 +799,7 @@ static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
 
 	if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps) && !*md5)) {
 		opts->options |= OPTION_TS;
-		opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset;
+		opts->tsval = tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb) + tp->tsoffset;
 		opts->tsecr = tp->rx_opt.ts_recent;
 		remaining -= TCPOLEN_TSTAMP_ALIGNED;
 	}
@@ -884,7 +884,8 @@ static unsigned int tcp_synack_options(const struct sock *sk,
 	}
 	if (likely(ireq->tstamp_ok)) {
 		opts->options |= OPTION_TS;
-		opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off;
+		opts->tsval = tcp_skb_timestamp_ts(tcp_rsk(req)->req_usec_ts, skb) +
+			      tcp_rsk(req)->ts_off;
 		opts->tsecr = READ_ONCE(req->ts_recent);
 		remaining -= TCPOLEN_TSTAMP_ALIGNED;
 	}
@@ -943,7 +944,8 @@ static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb
 
 	if (likely(tp->rx_opt.tstamp_ok)) {
 		opts->options |= OPTION_TS;
-		opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0;
+		opts->tsval = skb ? tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb) +
+				tp->tsoffset : 0;
 		opts->tsecr = tp->rx_opt.ts_recent;
 		size += TCPOLEN_TSTAMP_ALIGNED;
 	}
@@ -3379,7 +3381,7 @@ int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
 
 	/* Save stamp of the first (attempted) retransmit. */
 	if (!tp->retrans_stamp)
-		tp->retrans_stamp = tcp_skb_timestamp(skb);
+		tp->retrans_stamp = tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb);
 
 	if (tp->undo_retrans < 0)
 		tp->undo_retrans = 0;
@@ -3665,6 +3667,8 @@ struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
 	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
 
 	memset(&opts, 0, sizeof(opts));
+	if (tcp_rsk(req)->req_usec_ts < 0)
+		tcp_rsk(req)->req_usec_ts = dst_tcp_usec_ts(dst);
 	now = tcp_clock_ns();
 #ifdef CONFIG_SYN_COOKIES
 	if (unlikely(synack_type == TCP_SYNACK_COOKIE && ireq->tstamp_ok))
@@ -3961,7 +3965,7 @@ int tcp_connect(struct sock *sk)
 
 	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
 	tcp_mstamp_refresh(tp);
-	tp->retrans_stamp = tcp_time_stamp(tp);
+	tp->retrans_stamp = tcp_time_stamp_ts(tp);
 	tcp_connect_queue_skb(sk, buff);
 	tcp_ecn_send_syn(sk, buff);
 	tcp_rbtree_insert(&sk->tcp_rtx_queue, buff);

net/ipv4/tcp_timer.c

@@ -26,14 +26,18 @@
 static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
-	u32 elapsed, start_ts, user_timeout;
+	const struct tcp_sock *tp = tcp_sk(sk);
+	u32 elapsed, user_timeout;
 	s32 remaining;
 
-	start_ts = tcp_sk(sk)->retrans_stamp;
 	user_timeout = READ_ONCE(icsk->icsk_user_timeout);
 	if (!user_timeout)
 		return icsk->icsk_rto;
-	elapsed = tcp_time_stamp(tcp_sk(sk)) - start_ts;
+
+	elapsed = tcp_time_stamp_ts(tp) - tp->retrans_stamp;
+	if (tp->tcp_usec_ts)
+		elapsed /= USEC_PER_MSEC;
+
 	remaining = user_timeout - elapsed;
 	if (remaining <= 0)
 		return 1; /* user timeout has passed; fire ASAP */
@@ -212,12 +216,13 @@ static bool retransmits_timed_out(struct sock *sk,
 				  unsigned int boundary,
 				  unsigned int timeout)
 {
-	unsigned int start_ts;
+	struct tcp_sock *tp = tcp_sk(sk);
+	unsigned int start_ts, delta;
 
 	if (!inet_csk(sk)->icsk_retransmits)
 		return false;
 
-	start_ts = tcp_sk(sk)->retrans_stamp;
+	start_ts = tp->retrans_stamp;
 	if (likely(timeout == 0)) {
 		unsigned int rto_base = TCP_RTO_MIN;
 
@@ -226,7 +231,12 @@ static bool retransmits_timed_out(struct sock *sk,
 		timeout = tcp_model_timeout(sk, boundary, rto_base);
 	}
 
-	return (s32)(tcp_time_stamp(tcp_sk(sk)) - start_ts - timeout) >= 0;
+	if (tp->tcp_usec_ts) {
+		/* delta maybe off up to a jiffy due to timer granularity. */
+		delta = tp->tcp_mstamp - start_ts + jiffies_to_usecs(1);
+		return (s32)(delta - timeout * USEC_PER_MSEC) >= 0;
+	}
+	return (s32)(tcp_time_stamp_ts(tp) - start_ts - timeout) >= 0;
 }
 
 /* A write timeout has occurred. Process the after effects. */
@@ -422,7 +432,7 @@ static void tcp_update_rto_stats(struct sock *sk)
 
 	if (!icsk->icsk_retransmits) {
 		tp->total_rto_recoveries++;
-		tp->rto_stamp = tcp_time_stamp(tp);
+		tp->rto_stamp = tcp_time_stamp_ms(tp);
 	}
 	icsk->icsk_retransmits++;
 	tp->total_rto++;
@@ -462,26 +472,24 @@ static void tcp_fastopen_synack_timer(struct sock *sk, struct request_sock *req)
 	req->num_timeout++;
 	tcp_update_rto_stats(sk);
 	if (!tp->retrans_stamp)
-		tp->retrans_stamp = tcp_time_stamp(tp);
+		tp->retrans_stamp = tcp_time_stamp_ts(tp);
 	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 			  req->timeout << req->num_timeout, TCP_RTO_MAX);
 }
 
 static bool tcp_rtx_probe0_timed_out(const struct sock *sk,
-				     const struct sk_buff *skb)
+				     const struct sk_buff *skb,
+				     u32 rtx_delta)
 {
 	const struct tcp_sock *tp = tcp_sk(sk);
 	const int timeout = TCP_RTO_MAX * 2;
-	u32 rcv_delta, rtx_delta;
+	u32 rcv_delta;
 
 	rcv_delta = inet_csk(sk)->icsk_timeout - tp->rcv_tstamp;
 	if (rcv_delta <= timeout)
 		return false;
 
-	rtx_delta = (u32)msecs_to_jiffies(tcp_time_stamp(tp) -
-			(tp->retrans_stamp ?: tcp_skb_timestamp(skb)));
-
-	return rtx_delta > timeout;
+	return msecs_to_jiffies(rtx_delta) > timeout;
 }
 
 /**
@@ -534,7 +542,11 @@ void tcp_retransmit_timer(struct sock *sk)
 		struct inet_sock *inet = inet_sk(sk);
 		u32 rtx_delta;
 
-		rtx_delta = tcp_time_stamp(tp) - (tp->retrans_stamp ?: tcp_skb_timestamp(skb));
+		rtx_delta = tcp_time_stamp_ts(tp) - (tp->retrans_stamp ?: 
+				tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb));
+		if (tp->tcp_usec_ts)
+			rtx_delta /= USEC_PER_MSEC;
+
 		if (sk->sk_family == AF_INET) {
 			net_dbg_ratelimited("Probing zero-window on %pI4:%u/%u, seq=%u:%u, recv %ums ago, lasting %ums\n",
 				&inet->inet_daddr, ntohs(inet->inet_dport),
@@ -551,7 +563,7 @@ void tcp_retransmit_timer(struct sock *sk)
 				rtx_delta);
 		}
 #endif
-		if (tcp_rtx_probe0_timed_out(sk, skb)) {
+		if (tcp_rtx_probe0_timed_out(sk, skb, rtx_delta)) {
 			tcp_write_err(sk);
 			goto out;
 		}

net/ipv6/tcp_ipv6.c

@@ -286,6 +286,7 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 		goto failure;
 	}
 
+	tp->tcp_usec_ts = dst_tcp_usec_ts(dst);
 	tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 
 	if (!saddr) {
@@ -1096,7 +1097,7 @@ static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 
 	tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
-			tcp_time_stamp_raw() + tcptw->tw_ts_offset,
+			tcp_tw_tsval(tcptw),
 			tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
 			tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority,
 			tw->tw_txhash);
@@ -1123,7 +1124,7 @@ static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 			tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 			tcp_rsk(req)->rcv_nxt,
 			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
-			tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
+			tcp_rsk_tsval(tcp_rsk(req)),
 			READ_ONCE(req->ts_recent), sk->sk_bound_dev_if,
 			tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
 			ipv6_get_dsfield(ipv6_hdr(skb)), 0,

net/netfilter/nf_synproxy_core.c

@@ -153,7 +153,7 @@ void synproxy_init_timestamp_cookie(const struct nf_synproxy_info *info,
 				    struct synproxy_options *opts)
 {
 	opts->tsecr = opts->tsval;
-	opts->tsval = tcp_time_stamp_raw() & ~0x3f;
+	opts->tsval = tcp_clock_ms() & ~0x3f;
 
 	if (opts->options & NF_SYNPROXY_OPT_WSCALE) {
 		opts->tsval |= opts->wscale;

tools/testing/selftests/bpf/progs/xdp_synproxy_kern.c

@@ -177,7 +177,7 @@ static __always_inline __u32 tcp_ns_to_ts(__u64 ns)
 	return ns / (NSEC_PER_SEC / TCP_TS_HZ);
 }
 
-static __always_inline __u32 tcp_time_stamp_raw(void)
+static __always_inline __u32 tcp_clock_ms(void)
 {
 	return tcp_ns_to_ts(tcp_clock_ns());
 }
@@ -274,7 +274,7 @@ static __always_inline bool tscookie_init(struct tcphdr *tcp_header,
 	if (!loop_ctx.option_timestamp)
 		return false;
 
-	cookie = tcp_time_stamp_raw() & ~TSMASK;
+	cookie = tcp_clock_ms() & ~TSMASK;
 	cookie |= loop_ctx.wscale & TS_OPT_WSCALE_MASK;
 	if (loop_ctx.option_sack)
 		cookie |= TS_OPT_SACK;