tcp: add support for usec resolution in TCP TS values

Back in 2015, Van Jacobson suggested to use usec resolution in TCP TS values.
This has been implemented in our private kernels.

Goals were :

1) better observability of delays in networking stacks.
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 negotiate the feature.

For upstream submission, we chose to use a route attribute,
because this feature is probably going to be used in private
networks [1] [2].

ip route add 10/8 ... features tcp_usec_ts

Note that RFC 7323 recommends a
  "timestamp clock frequency in the range 1 ms to 1 sec per tick.",
but also mentions
  "the maximum acceptable clock frequency is one tick every 59 ns."

[1] Unfortunately RFC 7323 5.5 (Outdated Timestamps) suggests
to invalidate TS.Recent values after a flow was idle for more
than 24 days. This is the part making usec_ts a problem
for peers following this recommendation for long living
idle flows.

[2] Attempts to standardize usec ts went nowhere:

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/

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

7 files changed, 72 insertions, 29 deletions

diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index 62395fdb0ca5..c64334363230 100644
--- a/net/ipv4/syncookies.c
+++ b/net/ipv4/syncookies.c
@@ -84,7 +84,9 @@ u64 cookie_init_timestamp(struct request_sock *req, u64 now)
 	if (ts > ts_now)
 		ts -= (1UL << TSBITS);
 
-	return ts * (NSEC_PER_SEC / TCP_TS_HZ);
+	if (tcp_rsk(req)->req_usec_ts)
+		return ts * NSEC_PER_USEC;
+	return ts * NSEC_PER_MSEC;
 }
 
 
@@ -304,6 +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) {
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 805f8341064f..b961364b4961 100644
--- a/net/ipv4/tcp.c
+++ b/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_clock_ts(false));
+			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);
@@ -4143,7 +4149,11 @@ int do_tcp_getsockopt(struct sock *sk, int level,
 		break;
 
 	case TCP_TIMESTAMP:
-		val = tcp_clock_ts(false) + 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);
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 5666f6137167..18b858597af4 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -698,6 +698,8 @@ 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)
@@ -2452,7 +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_ts(false, skb));
+	       tcp_tsopt_ecr_before(tp, tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb));
 }
 
 /* Nothing was retransmitted or returned timestamp is less
@@ -7045,6 +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
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index cdd65cc594bc..7583d4e34c8c 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/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);
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index a9fdba897a28..ace806c5bd0c 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/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;
@@ -554,21 +555,29 @@ 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->undo_marker = treq->snt_isn;
-		newtp->retrans_stamp = div_u64(treq->snt_synack,
-					       USEC_PER_SEC / TCP_TS_HZ);
 		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_time = tcp_clock_ms() -
+						newtp->retrans_stamp;
+		}
 		newtp->total_rto_recoveries = 1;
-		newtp->total_rto_time = tcp_clock_ms() -
-					newtp->retrans_stamp;
 	}
 	newtp->tsoffset = treq->ts_off;
 #ifdef CONFIG_TCP_MD5SIG
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index a1fec8be9ac3..2866ccbccde0 100644
--- a/net/ipv4/tcp_output.c
+++ b/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_ts(false, 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_ts(false, 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_ts(false, 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_ts(false, 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))
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index bfcf3fe44c72..1f9f6c1c196b 100644
--- a/net/ipv4/tcp_timer.c
+++ b/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_ts(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_ts(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. */
@@ -468,20 +478,18 @@ static void tcp_fastopen_synack_timer(struct sock *sk, struct request_sock *req)
 }
 
 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_ts(tp) -
-			(tp->retrans_stamp ?: tcp_skb_timestamp_ts(false, 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_ts(tp) - (tp->retrans_stamp ?: tcp_skb_timestamp_ts(false, 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;
 		}