[PATCH net-next 00/13] tcp: add optional usec resolution to TCP TS

[PATCH net-next 00/13] tcp: add optional usec resolution to TCP TS

[Eric Dumazet]

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)

Eric Dumazet (13):
  chtls: fix tp->rcv_tstamp initialization
  tcp: fix cookie_init_timestamp() overflows
  tcp: add tcp_time_stamp_ms() helper
  tcp: introduce tcp_clock_ms()
  tcp: replace tcp_time_stamp_raw()
  tcp: rename tcp_skb_timestamp()
  tcp: move tcp_ns_to_ts() to net/ipv4/syncookies.c
  tcp: rename tcp_time_stamp() to tcp_time_stamp_ts()
  tcp: add tcp_rtt_tsopt_us()
  tcp: add RTAX_FEATURE_TCP_USEC_TS
  tcp: introduce TCP_PAWS_WRAP
  tcp: add support for usec resolution in TCP TS values
  tcp: add TCPI_OPT_USEC_TS

 .../chelsio/inline_crypto/chtls/chtls_cm.c    |  2 +-
 include/linux/tcp.h                           |  9 ++-
 include/net/inet_timewait_sock.h              |  3 +-
 include/net/tcp.h                             | 59 ++++++++++++++-----
 include/uapi/linux/rtnetlink.h                | 18 +++---
 include/uapi/linux/tcp.h                      |  1 +
 net/ipv4/syncookies.c                         | 32 ++++++----
 net/ipv4/tcp.c                                | 26 +++++---
 net/ipv4/tcp_input.c                          | 52 ++++++++--------
 net/ipv4/tcp_ipv4.c                           |  5 +-
 net/ipv4/tcp_lp.c                             |  2 +-
 net/ipv4/tcp_minisocks.c                      | 19 ++++--
 net/ipv4/tcp_output.c                         | 14 +++--
 net/ipv4/tcp_timer.c                          | 44 +++++++++-----
 net/ipv6/tcp_ipv6.c                           |  5 +-
 net/netfilter/nf_synproxy_core.c              |  2 +-
 .../selftests/bpf/progs/xdp_synproxy_kern.c   |  4 +-
 17 files changed, 193 insertions(+), 104 deletions(-)

-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 01/13] chtls: fix tp->rcv_tstamp initialization

[Eric Dumazet]

tp->rcv_tstamp should be set to tcp_jiffies, not tcp_time_stamp().

Fixes: cc35c88ae4db ("crypto : chtls - CPL handler definition")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ayush Sawal <ayush.sawal@chelsio.com>
---
 drivers/net/ethernet/chelsio/inline_crypto/chtls/chtls_cm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/chelsio/inline_crypto/chtls/chtls_cm.c b/drivers/net/ethernet/chelsio/inline_crypto/chtls/chtls_cm.c
index 7750702900fa60b3405992e6c6c465b43abff116..6f6525983130e7701e9c6530afbecb9ae374d7ee 100644
--- a/drivers/net/ethernet/chelsio/inline_crypto/chtls/chtls_cm.c
+++ b/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);
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 02/13] tcp: fix cookie_init_timestamp() overflows

[Eric Dumazet]

cookie_init_timestamp() is supposed to return a 64bit timestamp
suitable for both TSval determination and setting of skb->tstamp.

Unfortunately it uses 32bit fields and overflows after
2^32 * 10^6 nsec (~49 days) of uptime.

Generated TSval are still correct, but skb->tstamp might be set
far away in the past, potentially confusing other layers.

tcp_ns_to_ts() is changed to return a full 64bit value,
ts and ts_now variables are changed to u64 type,
and TSMASK is removed in favor of shifts operations.

While we are at it, change this sequence:
		ts >>= TSBITS;
		ts--;
		ts <<= TSBITS;
		ts |= options;
to:
		ts -= (1UL << TSBITS);

Fixes: 9a568de4818d ("tcp: switch TCP TS option (RFC 7323) to 1ms clock")
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/tcp.h     |  2 +-
 net/ipv4/syncookies.c | 20 +++++++-------------
 2 files changed, 8 insertions(+), 14 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index bad304d173a56c768e1f87fa740e8e568d220f00..d47a57a47b50b4bbc7ff45c76371d39cf6207c54 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -805,7 +805,7 @@ static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
 }
 
 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
-static inline u32 tcp_ns_to_ts(u64 ns)
+static inline u64 tcp_ns_to_ts(u64 ns)
 {
 	return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
 }
diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index dc478a0574cbe71d392522ed4a690f12654f3e08..3b4dafefb4b036c661bf52f5e7e304b943a4fd5e 100644
--- a/net/ipv4/syncookies.c
+++ b/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)
@@ -62,27 +61,22 @@ 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(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 = (ts_now >> TSBITS) << TSBITS;
 	ts |= options;
-	if (ts > ts_now) {
-		ts >>= TSBITS;
-		ts--;
-		ts <<= TSBITS;
-		ts |= options;
-	}
-	return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
+	if (ts > ts_now)
+		ts -= (1UL << TSBITS);
+
+	return ts * (NSEC_PER_SEC / TCP_TS_HZ);
 }
 
 
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 03/13] tcp: add tcp_time_stamp_ms() helper

[Eric Dumazet]

In preparation of adding usec TCP TS values, add tcp_time_stamp_ms()
for contexts needing ms based values.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/tcp.h    | 5 +++++
 net/ipv4/tcp_input.c | 2 +-
 net/ipv4/tcp_timer.c | 2 +-
 3 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index d47a57a47b50b4bbc7ff45c76371d39cf6207c54..9fc6dc4ba9e2e2be44318d4495ceb19523395b18 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -804,6 +804,11 @@ static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
 	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
 }
 
+static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
+{
+	return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
+}
+
 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
 static inline u64 tcp_ns_to_ts(u64 ns)
 {
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index ab87f0285b728f3829f1e409833ed4feebd7920e..ffce17545b62c78595c5dd569665a6ebe6a29bbc 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -2856,7 +2856,7 @@ 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;
 	}
 }
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index 0862b73dd3b5299d2b201e9e93dbef8a0617f75b..63247c78dc13d445c1e1c5cf24e7ffd7a1faa403 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -422,7 +422,7 @@ 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++;
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 04/13] tcp: introduce tcp_clock_ms()

[Eric Dumazet]

It delivers current TCP time stamp in ms unit, and is used
in place of confusing tcp_time_stamp_raw()

It is the same family than tcp_clock_ns() and tcp_clock_ms().

tcp_time_stamp_raw() will be replaced later for TSval
contexts with a more descriptive name.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/tcp.h                                     | 5 +++++
 net/ipv4/tcp.c                                        | 6 ++----
 net/ipv4/tcp_minisocks.c                              | 4 ++--
 net/netfilter/nf_synproxy_core.c                      | 2 +-
 tools/testing/selftests/bpf/progs/xdp_synproxy_kern.c | 4 ++--
 5 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 9fc6dc4ba9e2e2be44318d4495ceb19523395b18..3bdf1141f5a2c11e30ad85c68aafd062e7bf548c 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -798,6 +798,11 @@ static inline u64 tcp_clock_us(void)
 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
 }
 
+static inline u64 tcp_clock_ms(void)
+{
+	return div_u64(tcp_clock_ns(), NSEC_PER_MSEC);
+}
+
 /* 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)
 {
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 56a8d936000f610566e46694fec5ddd1dccd7102..5b034b0356ecbd2b7d2dcafd9caac2b8de5886f1 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3817,10 +3817,8 @@ 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);
 }
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 3f87611077ef21edb61f3d6c751c88c515bb4b5b..a9fdba897a28f3a5835a1bd1426f82de0e81a633 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -567,8 +567,8 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
 					       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->retrans_stamp;
+		newtp->total_rto_time = tcp_clock_ms() -
+					newtp->retrans_stamp;
 	}
 	newtp->tsoffset = treq->ts_off;
 #ifdef CONFIG_TCP_MD5SIG
diff --git a/net/netfilter/nf_synproxy_core.c b/net/netfilter/nf_synproxy_core.c
index 16915f8eef2b16eec7da7ecaaf8b4a4af5dd94e2..467671f2d42f742554d95b017738aaa4ea4c9121 100644
--- a/net/netfilter/nf_synproxy_core.c
+++ b/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;
diff --git a/tools/testing/selftests/bpf/progs/xdp_synproxy_kern.c b/tools/testing/selftests/bpf/progs/xdp_synproxy_kern.c
index 07d786329105da371dd280868981e8e19ac5f6a8..e959336c7a7304be409ffac7d3a34f64538d5f74 100644
--- a/tools/testing/selftests/bpf/progs/xdp_synproxy_kern.c
+++ b/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;
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 05/13] tcp: replace tcp_time_stamp_raw()

[Eric Dumazet]

In preparation of usec TCP TS support, remove tcp_time_stamp_raw()
in favor of tcp_clock_ts() helper. This helper will return a suitable
32bit result to feed TS values, depending on a socket field.

Also add tcp_tw_tsval() and tcp_rsk_tsval() helpers to factorize
the details.

We do not yet support usec timestamps.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/tcp.h   | 25 +++++++++++++++++++------
 net/ipv4/tcp.c      |  4 ++--
 net/ipv4/tcp_ipv4.c |  4 ++--
 net/ipv6/tcp_ipv6.c |  4 ++--
 4 files changed, 25 insertions(+), 12 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 3bdf1141f5a2c11e30ad85c68aafd062e7bf548c..0534526a535da7cee7d8d49fd556fe4d7a4eefb6 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -803,6 +803,16 @@ static inline u64 tcp_clock_ms(void)
 	return div_u64(tcp_clock_ns(), NSEC_PER_MSEC);
 }
 
+/* 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 usec_ts ? tcp_clock_us() : tcp_clock_ms();
+}
+
 /* 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)
 {
@@ -820,12 +830,6 @@ static inline u64 tcp_ns_to_ts(u64 ns)
 	return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
 }
 
-/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
-static inline u32 tcp_time_stamp_raw(void)
-{
-	return tcp_ns_to_ts(tcp_clock_ns());
-}
-
 void tcp_mstamp_refresh(struct tcp_sock *tp);
 
 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
@@ -844,6 +848,15 @@ static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
 }
 
+static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw)
+{
+	return tcp_clock_ts(false) + tcptw->tw_ts_offset;
+}
+
+static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq)
+{
+	return tcp_clock_ts(false) + treq->ts_off;
+}
 
 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
 
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 5b034b0356ecbd2b7d2dcafd9caac2b8de5886f1..805f8341064fec4fe0504e14c579185cfe11d896 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3632,7 +3632,7 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,
 		if (!tp->repair)
 			err = -EPERM;
 		else
-			WRITE_ONCE(tp->tsoffset, val - tcp_time_stamp_raw());
+			WRITE_ONCE(tp->tsoffset, val - tcp_clock_ts(false));
 		break;
 	case TCP_REPAIR_WINDOW:
 		err = tcp_repair_set_window(tp, optval, optlen);
@@ -4143,7 +4143,7 @@ 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(false) + READ_ONCE(tp->tsoffset);
 		break;
 	case TCP_NOTSENT_LOWAT:
 		val = READ_ONCE(tp->notsent_lowat);
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index a67a5de86253ba40e38154e81ffef11f68a55a3a..cdd65cc594bc4571fa5793bc14d6e9ab892dfd2a 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -954,7 +954,7 @@ 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 +988,7 @@ 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),
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index d410703bb5a1e483848104e7e4cb379e45d7b216..1ee6517e9b2f9d549268dea240a725976a9e2720 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1096,7 +1096,7 @@ 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 +1123,7 @@ 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,
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 06/13] tcp: rename tcp_skb_timestamp()

[Eric Dumazet]

This helper returns a 32bit TCP TSval from skb->tstamp.

As we are going to support usec or ms units soon, rename it
to tcp_skb_timestamp_ts() and add a boolean to select the unit.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/tcp.h     | 14 +++++++++-----
 net/ipv4/tcp_input.c  |  2 +-
 net/ipv4/tcp_output.c |  8 ++++----
 net/ipv4/tcp_timer.c  |  4 ++--
 4 files changed, 16 insertions(+), 12 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 0534526a535da7cee7d8d49fd556fe4d7a4eefb6..493f8550055bca09b69a9d3129d6ba781a1233f8 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -837,17 +837,21 @@ 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(false) + tcptw->tw_ts_offset;
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index ffce17545b62c78595c5dd569665a6ebe6a29bbc..de68cad82d19e37171deadc45c5acc0cfd90c315 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -2442,7 +2442,7 @@ 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(false, skb));
 }
 
 /* Nothing was retransmitted or returned timestamp is less
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 909f85aefd7401aeffcd098356c5e3823bffd89e..03a2a9fc0dc191d7066d679913d41bd2ef2d685a 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(skb) + tp->tsoffset;
+		opts->tsval = tcp_skb_timestamp_ts(false, skb) + tp->tsoffset;
 		opts->tsecr = tp->rx_opt.ts_recent;
 		remaining -= TCPOLEN_TSTAMP_ALIGNED;
 	}
@@ -884,7 +884,7 @@ 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(false, skb) + tcp_rsk(req)->ts_off;
 		opts->tsecr = READ_ONCE(req->ts_recent);
 		remaining -= TCPOLEN_TSTAMP_ALIGNED;
 	}
@@ -943,7 +943,7 @@ 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(false, skb) + tp->tsoffset : 0;
 		opts->tsecr = tp->rx_opt.ts_recent;
 		size += TCPOLEN_TSTAMP_ALIGNED;
 	}
@@ -3379,7 +3379,7 @@ 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(false, skb);
 
 	if (tp->undo_retrans < 0)
 		tp->undo_retrans = 0;
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index 63247c78dc13d445c1e1c5cf24e7ffd7a1faa403..8764a9a2dc213f648ffc64f79950037b1f44ee99 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -479,7 +479,7 @@ static bool tcp_rtx_probe0_timed_out(const struct sock *sk,
 		return false;
 
 	rtx_delta = (u32)msecs_to_jiffies(tcp_time_stamp(tp) -
-			(tp->retrans_stamp ?: tcp_skb_timestamp(skb)));
+			(tp->retrans_stamp ?: tcp_skb_timestamp_ts(false, skb)));
 
 	return rtx_delta > timeout;
 }
@@ -534,7 +534,7 @@ 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(tp) - (tp->retrans_stamp ?: tcp_skb_timestamp_ts(false, skb));
 		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),
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 07/13] tcp: move tcp_ns_to_ts() to net/ipv4/syncookies.c

[Eric Dumazet]

tcp_ns_to_ts() is only used once from cookie_init_timestamp().

Also add the 'bool usec_ts' parameter to enable usec TS later.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/tcp.h     |  6 ------
 net/ipv4/syncookies.c | 10 +++++++++-
 2 files changed, 9 insertions(+), 7 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 493f8550055bca09b69a9d3129d6ba781a1233f8..b86abf1fbe46061a00dbd202323792f01a307969 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -824,12 +824,6 @@ static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
 	return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
 }
 
-/* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
-static inline u64 tcp_ns_to_ts(u64 ns)
-{
-	return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
-}
-
 void tcp_mstamp_refresh(struct tcp_sock *tp);
 
 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index 3b4dafefb4b036c661bf52f5e7e304b943a4fd5e..62395fdb0ca557caee78d84ae70273eb42a837b9 100644
--- a/net/ipv4/syncookies.c
+++ b/net/ipv4/syncookies.c
@@ -51,6 +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
@@ -62,7 +70,7 @@ static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
 u64 cookie_init_timestamp(struct request_sock *req, u64 now)
 {
 	const struct inet_request_sock *ireq = inet_rsk(req);
-	u64 ts, ts_now = tcp_ns_to_ts(now);
+	u64 ts, ts_now = tcp_ns_to_ts(false, now);
 	u32 options = 0;
 
 	options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 08/13] tcp: rename tcp_time_stamp() to tcp_time_stamp_ts()

[Eric Dumazet]

This helper returns a TSval from a TCP socket.

It currently calls tcp_time_stamp_ms() but will soon
be able to return a usec based TSval, depending
on an upcoming tp->tcp_usec_ts field.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/tcp.h     |  9 ++++-----
 net/ipv4/tcp_input.c  |  6 +++---
 net/ipv4/tcp_lp.c     |  2 +-
 net/ipv4/tcp_output.c |  2 +-
 net/ipv4/tcp_timer.c  | 10 +++++-----
 5 files changed, 14 insertions(+), 15 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index b86abf1fbe46061a00dbd202323792f01a307969..af72c1dc37f3dd4cd6858e9c8f6aa7ef31541652 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -813,15 +813,14 @@ static inline u32 tcp_clock_ts(bool usec_ts)
 	return usec_ts ? tcp_clock_us() : tcp_clock_ms();
 }
 
-/* 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 u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
 {
-	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
+	return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
 }
 
-static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
+static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp)
 {
-	return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
+	return tcp_time_stamp_ms(tp);
 }
 
 void tcp_mstamp_refresh(struct tcp_sock *tp);
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index de68cad82d19e37171deadc45c5acc0cfd90c315..e7e38fc1d62ff16d7afd7f2ba58a1990f01e17b6 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -704,7 +704,7 @@ 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 = tcp_time_stamp_ts(tp) - tp->rx_opt.rcv_tsecr;
 		u32 delta_us;
 
 		if (likely(delta < INT_MAX / (USEC_PER_SEC / TCP_TS_HZ))) {
@@ -3148,7 +3148,7 @@ static bool tcp_ack_update_rtt(struct sock *sk, const int flag,
 	 */
 	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;
+		u32 delta = tcp_time_stamp_ts(tp) - tp->rx_opt.rcv_tsecr;
 
 		if (likely(delta < INT_MAX / (USEC_PER_SEC / TCP_TS_HZ))) {
 			if (!delta)
@@ -6293,7 +6293,7 @@ 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;
diff --git a/net/ipv4/tcp_lp.c b/net/ipv4/tcp_lp.c
index ae36780977d2762066cdd59e40116d1240492b90..52fe17167460fc433ec84434795f7cbef8144767 100644
--- a/net/ipv4/tcp_lp.c
+++ b/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)
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 03a2a9fc0dc191d7066d679913d41bd2ef2d685a..a1fec8be9ac36c67022c90b08b0a5faa935725f0 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -3961,7 +3961,7 @@ 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);
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index 8764a9a2dc213f648ffc64f79950037b1f44ee99..bfcf3fe44c72427eccb37376bec15fb71b594c56 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -33,7 +33,7 @@ static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)
 	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(tcp_sk(sk)) - start_ts;
 	remaining = user_timeout - elapsed;
 	if (remaining <= 0)
 		return 1; /* user timeout has passed; fire ASAP */
@@ -226,7 +226,7 @@ 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;
+	return (s32)(tcp_time_stamp_ts(tcp_sk(sk)) - start_ts - timeout) >= 0;
 }
 
 /* A write timeout has occurred. Process the after effects. */
@@ -462,7 +462,7 @@ 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);
 }
@@ -478,7 +478,7 @@ static bool tcp_rtx_probe0_timed_out(const struct sock *sk,
 	if (rcv_delta <= timeout)
 		return false;
 
-	rtx_delta = (u32)msecs_to_jiffies(tcp_time_stamp(tp) -
+	rtx_delta = (u32)msecs_to_jiffies(tcp_time_stamp_ts(tp) -
 			(tp->retrans_stamp ?: tcp_skb_timestamp_ts(false, skb)));
 
 	return rtx_delta > timeout;
@@ -534,7 +534,7 @@ 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_ts(false, skb));
+		rtx_delta = tcp_time_stamp_ts(tp) - (tp->retrans_stamp ?: tcp_skb_timestamp_ts(false, skb));
 		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),
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 09/13] tcp: add tcp_rtt_tsopt_us()

[Eric Dumazet]

Before adding usec TS support, add tcp_rtt_tsopt_us() helper
to factorize code.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/tcp_input.c | 43 +++++++++++++++++++++++--------------------
 1 file changed, 23 insertions(+), 20 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index e7e38fc1d62ff16d7afd7f2ba58a1990f01e17b6..5666f61371678ce1c7bf466e0d54c0b7a396841b 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -693,6 +693,21 @@ 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 (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 +719,10 @@ 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_ts(tp) - tp->rx_opt.rcv_tsecr;
-		u32 delta_us;
-
-		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);
-		}
+		s32 delta = tcp_rtt_tsopt_us(tp);
+
+		if (delta >= 0)
+			tcp_rcv_rtt_update(tp, delta, 0);
 	}
 }
 
@@ -3146,17 +3156,10 @@ 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_ts(tp) - tp->rx_opt.rcv_tsecr;
-
-		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;
-		}
-	}
+	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);
+
 	rs->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet (or -1) */
 	if (seq_rtt_us < 0)
 		return false;
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 10/13] tcp: add RTAX_FEATURE_TCP_USEC_TS

[Eric Dumazet]

This new dst feature flag will be used to allow TCP to use usec
based timestamps instead of msec ones.

ip route .... feature tcp_usec_ts

Also document that RTAX_FEATURE_SACK and RTAX_FEATURE_TIMESTAMP
are unused.

RTAX_FEATURE_ALLFRAG is also going away soon.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/tcp.h            |  5 +++++
 include/uapi/linux/rtnetlink.h | 18 +++++++++++-------
 2 files changed, 16 insertions(+), 7 deletions(-)

diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index e15452df9804f6d5badddaed3445c5586b1fcea3..04a0e647ef747e5c83520cd1b1c2156d3a315ba5 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -576,4 +576,9 @@ 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 */
diff --git a/include/uapi/linux/rtnetlink.h b/include/uapi/linux/rtnetlink.h
index 51c13cf9c5aee4a2d1ab33c1a89043383d67b9cf..aa2482a0614aa685590fcc73819cbe1baac63d66 100644
--- a/include/uapi/linux/rtnetlink.h
+++ b/include/uapi/linux/rtnetlink.h
@@ -502,13 +502,17 @@ 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_ALLFRAG	(1 << 3)
-
-#define RTAX_FEATURE_MASK	(RTAX_FEATURE_ECN | RTAX_FEATURE_SACK | \
-				 RTAX_FEATURE_TIMESTAMP | RTAX_FEATURE_ALLFRAG)
+#define RTAX_FEATURE_ECN		(1 << 0)
+#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 |		\
+				 RTAX_FEATURE_TCP_USEC_TS)
 
 struct rta_session {
 	__u8	proto;
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 11/13] tcp: introduce TCP_PAWS_WRAP

[Eric Dumazet]

tcp_paws_check() uses TCP_PAWS_24DAYS constant to detect if TCP TS
values might have wrapped after a long idle period.

This mechanism is described in RFC 7323 5.5 (Outdated Timestamps)

TCP_PAWS_24DAYS value was based on the assumption of a clock
of 1 Khz.

As we want to adopt a 1 Mhz clock in the future, we reduce
this constant.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/tcp.h | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index af72c1dc37f3dd4cd6858e9c8f6aa7ef31541652..0ab577869d7ac61c5c3b687e98068a16b69bb3af 100644
--- a/include/net/tcp.h
+++ b/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
@@ -1619,7 +1624,7 @@ 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,
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 12/13] tcp: add support for usec resolution in TCP TS values

[Eric Dumazet]

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>
---
 include/linux/tcp.h              |  4 +++-
 include/net/inet_timewait_sock.h |  3 ++-
 include/net/tcp.h                |  6 +++--
 net/ipv4/syncookies.c            |  6 ++++-
 net/ipv4/tcp.c                   | 18 ++++++++++----
 net/ipv4/tcp_input.c             |  5 +++-
 net/ipv4/tcp_ipv4.c              |  1 +
 net/ipv4/tcp_minisocks.c         | 19 +++++++++++----
 net/ipv4/tcp_output.c            | 12 ++++++----
 net/ipv4/tcp_timer.c             | 40 +++++++++++++++++++++-----------
 net/ipv6/tcp_ipv6.c              |  1 +
 11 files changed, 82 insertions(+), 33 deletions(-)

diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 04a0e647ef747e5c83520cd1b1c2156d3a315ba5..6df715b6e51d4420d8835146a75d0655e71dea5b 100644
--- a/include/linux/tcp.h
+++ b/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 */
diff --git a/include/net/inet_timewait_sock.h b/include/net/inet_timewait_sock.h
index 4a8e578405cb37856f07dac6ad0f1809f786682a..b14999ff55db1be505f371be48c26ff946062e83 100644
--- a/include/net/inet_timewait_sock.h
+++ b/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;
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 0ab577869d7ac61c5c3b687e98068a16b69bb3af..39b731c900dd53fa2683f316fcd779e8e16ec7e2 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -825,6 +825,8 @@ static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
 
 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);
 }
 
@@ -852,12 +854,12 @@ static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb)
 
 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw)
 {
-	return tcp_clock_ts(false) + tcptw->tw_ts_offset;
+	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(false) + treq->ts_off;
+	return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off;
 }
 
 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index 62395fdb0ca557caee78d84ae70273eb42a837b9..c6433436323096fe0e63f89e889687f74b57e387 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 805f8341064fec4fe0504e14c579185cfe11d896..b961364b4961c5f6fada9b2f6828413dfc8307ed 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 5666f61371678ce1c7bf466e0d54c0b7a396841b..18b858597af4ef62248f7c759cddf7224a681fc8 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 cdd65cc594bc4571fa5793bc14d6e9ab892dfd2a..7583d4e34c8c5fedaf29177f458f7aa39ca0519a 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 a9fdba897a28f3a5835a1bd1426f82de0e81a633..ace806c5bd0cdc2ec79eca5065e125418dd0c5b1 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 a1fec8be9ac36c67022c90b08b0a5faa935725f0..2866ccbccde073bff26182a0ad52446e84f789de 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 bfcf3fe44c72427eccb37376bec15fb71b594c56..1f9f6c1c196b2de35b0bc2f734484f09ba90541a 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;
 		}
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 1ee6517e9b2f9d549268dea240a725976a9e2720..0c8a14ba104f29877a17beae0c378313e2b206b3 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/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) {
-- 
2.42.0.655.g421f12c284-goog

[PATCH net-next 13/13] tcp: add TCPI_OPT_USEC_TS

[Eric Dumazet]

Add the ability to report in tcp_info.tcpi_options if
a flow is using usec resolution in TCP TS val.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/uapi/linux/tcp.h | 1 +
 net/ipv4/tcp.c           | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h
index d1d08da6331ab5e1f07baacd7057c7a2207229e6..8aa3916e14f6d09322a2a7e90402129d2aaa40b8 100644
--- a/include/uapi/linux/tcp.h
+++ b/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
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index b961364b4961c5f6fada9b2f6828413dfc8307ed..a86d8200a1e861b7c1292f480013c97eea141126 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3760,6 +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,
-- 
2.42.0.655.g421f12c284-goog

Re: [PATCH net-next 00/13] tcp: add optional usec resolution to TCP TS

[Yuchung Cheng]

On Fri, Oct 20, 2023 at 5:57 AM Eric Dumazet <edumazet@google.com> wrote:
>
> 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)

Huge thanks to Eric for making this happen for real :-) an immediate
benefit is enabling TCP timestamp based undos (Eifel) for short RTT
transactions. This allows datacenter TCP to use more aggressive
timeout w/o worrying too much of spurious timeout cwnd effect

Acked-by: Yuchung Cheng <ycheng@google.com>


>
> Eric Dumazet (13):
>   chtls: fix tp->rcv_tstamp initialization
>   tcp: fix cookie_init_timestamp() overflows
>   tcp: add tcp_time_stamp_ms() helper
>   tcp: introduce tcp_clock_ms()
>   tcp: replace tcp_time_stamp_raw()
>   tcp: rename tcp_skb_timestamp()
>   tcp: move tcp_ns_to_ts() to net/ipv4/syncookies.c
>   tcp: rename tcp_time_stamp() to tcp_time_stamp_ts()
>   tcp: add tcp_rtt_tsopt_us()
>   tcp: add RTAX_FEATURE_TCP_USEC_TS
>   tcp: introduce TCP_PAWS_WRAP
>   tcp: add support for usec resolution in TCP TS values
>   tcp: add TCPI_OPT_USEC_TS
>
>  .../chelsio/inline_crypto/chtls/chtls_cm.c    |  2 +-
>  include/linux/tcp.h                           |  9 ++-
>  include/net/inet_timewait_sock.h              |  3 +-
>  include/net/tcp.h                             | 59 ++++++++++++++-----
>  include/uapi/linux/rtnetlink.h                | 18 +++---
>  include/uapi/linux/tcp.h                      |  1 +
>  net/ipv4/syncookies.c                         | 32 ++++++----
>  net/ipv4/tcp.c                                | 26 +++++---
>  net/ipv4/tcp_input.c                          | 52 ++++++++--------
>  net/ipv4/tcp_ipv4.c                           |  5 +-
>  net/ipv4/tcp_lp.c                             |  2 +-
>  net/ipv4/tcp_minisocks.c                      | 19 ++++--
>  net/ipv4/tcp_output.c                         | 14 +++--
>  net/ipv4/tcp_timer.c                          | 44 +++++++++-----
>  net/ipv6/tcp_ipv6.c                           |  5 +-
>  net/netfilter/nf_synproxy_core.c              |  2 +-
>  .../selftests/bpf/progs/xdp_synproxy_kern.c   |  4 +-
>  17 files changed, 193 insertions(+), 104 deletions(-)
>
> --
> 2.42.0.655.g421f12c284-goog
>

Re: [PATCH net-next 00/13] tcp: add optional usec resolution to TCP TS

[Soheil Hassas Yeganeh]

On Fri, Oct 20, 2023 at 12:11 PM Yuchung Cheng <ycheng@google.com> wrote:
>
> On Fri, Oct 20, 2023 at 5:57 AM Eric Dumazet <edumazet@google.com> wrote:
> >
> > 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)
>
> Huge thanks to Eric for making this happen for real :-) an immediate
> benefit is enabling TCP timestamp based undos (Eifel) for short RTT
> transactions. This allows datacenter TCP to use more aggressive
> timeout w/o worrying too much of spurious timeout cwnd effect
>
> Acked-by: Yuchung Cheng <ycheng@google.com>

Thank you so much, Eric, for upstreaming the feature!  This is a major
milestone.

Acked-by: Soheil Hassas Yeganeh <soheil@google.com>

>
> >
> > Eric Dumazet (13):
> >   chtls: fix tp->rcv_tstamp initialization
> >   tcp: fix cookie_init_timestamp() overflows
> >   tcp: add tcp_time_stamp_ms() helper
> >   tcp: introduce tcp_clock_ms()
> >   tcp: replace tcp_time_stamp_raw()
> >   tcp: rename tcp_skb_timestamp()
> >   tcp: move tcp_ns_to_ts() to net/ipv4/syncookies.c
> >   tcp: rename tcp_time_stamp() to tcp_time_stamp_ts()
> >   tcp: add tcp_rtt_tsopt_us()
> >   tcp: add RTAX_FEATURE_TCP_USEC_TS
> >   tcp: introduce TCP_PAWS_WRAP
> >   tcp: add support for usec resolution in TCP TS values
> >   tcp: add TCPI_OPT_USEC_TS
> >
> >  .../chelsio/inline_crypto/chtls/chtls_cm.c    |  2 +-
> >  include/linux/tcp.h                           |  9 ++-
> >  include/net/inet_timewait_sock.h              |  3 +-
> >  include/net/tcp.h                             | 59 ++++++++++++++-----
> >  include/uapi/linux/rtnetlink.h                | 18 +++---
> >  include/uapi/linux/tcp.h                      |  1 +
> >  net/ipv4/syncookies.c                         | 32 ++++++----
> >  net/ipv4/tcp.c                                | 26 +++++---
> >  net/ipv4/tcp_input.c                          | 52 ++++++++--------
> >  net/ipv4/tcp_ipv4.c                           |  5 +-
> >  net/ipv4/tcp_lp.c                             |  2 +-
> >  net/ipv4/tcp_minisocks.c                      | 19 ++++--
> >  net/ipv4/tcp_output.c                         | 14 +++--
> >  net/ipv4/tcp_timer.c                          | 44 +++++++++-----
> >  net/ipv6/tcp_ipv6.c                           |  5 +-
> >  net/netfilter/nf_synproxy_core.c              |  2 +-
> >  .../selftests/bpf/progs/xdp_synproxy_kern.c   |  4 +-
> >  17 files changed, 193 insertions(+), 104 deletions(-)
> >
> > --
> > 2.42.0.655.g421f12c284-goog
> >

Re: [PATCH net-next 00/13] tcp: add optional usec resolution to TCP TS

[Kevin Yang]

Thanks Eric, this nice feature has been beneficial to Google for so
many years and is finally happening upstream :)

Acked-by: Kevin Yang <yyd@google.com>


On Fri, Oct 20, 2023 at 12:20 PM Soheil Hassas Yeganeh
<soheil@google.com> wrote:
>
> On Fri, Oct 20, 2023 at 12:11 PM Yuchung Cheng <ycheng@google.com> wrote:
> >
> > On Fri, Oct 20, 2023 at 5:57 AM Eric Dumazet <edumazet@google.com> wrote:
> > >
> > > 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)
> >
> > Huge thanks to Eric for making this happen for real :-) an immediate
> > benefit is enabling TCP timestamp based undos (Eifel) for short RTT
> > transactions. This allows datacenter TCP to use more aggressive
> > timeout w/o worrying too much of spurious timeout cwnd effect
> >
> > Acked-by: Yuchung Cheng <ycheng@google.com>
>
> Thank you so much, Eric, for upstreaming the feature!  This is a major
> milestone.
>
> Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
>
> >
> > >
> > > Eric Dumazet (13):
> > >   chtls: fix tp->rcv_tstamp initialization
> > >   tcp: fix cookie_init_timestamp() overflows
> > >   tcp: add tcp_time_stamp_ms() helper
> > >   tcp: introduce tcp_clock_ms()
> > >   tcp: replace tcp_time_stamp_raw()
> > >   tcp: rename tcp_skb_timestamp()
> > >   tcp: move tcp_ns_to_ts() to net/ipv4/syncookies.c
> > >   tcp: rename tcp_time_stamp() to tcp_time_stamp_ts()
> > >   tcp: add tcp_rtt_tsopt_us()
> > >   tcp: add RTAX_FEATURE_TCP_USEC_TS
> > >   tcp: introduce TCP_PAWS_WRAP
> > >   tcp: add support for usec resolution in TCP TS values
> > >   tcp: add TCPI_OPT_USEC_TS
> > >
> > >  .../chelsio/inline_crypto/chtls/chtls_cm.c    |  2 +-
> > >  include/linux/tcp.h                           |  9 ++-
> > >  include/net/inet_timewait_sock.h              |  3 +-
> > >  include/net/tcp.h                             | 59 ++++++++++++++-----
> > >  include/uapi/linux/rtnetlink.h                | 18 +++---
> > >  include/uapi/linux/tcp.h                      |  1 +
> > >  net/ipv4/syncookies.c                         | 32 ++++++----
> > >  net/ipv4/tcp.c                                | 26 +++++---
> > >  net/ipv4/tcp_input.c                          | 52 ++++++++--------
> > >  net/ipv4/tcp_ipv4.c                           |  5 +-
> > >  net/ipv4/tcp_lp.c                             |  2 +-
> > >  net/ipv4/tcp_minisocks.c                      | 19 ++++--
> > >  net/ipv4/tcp_output.c                         | 14 +++--
> > >  net/ipv4/tcp_timer.c                          | 44 +++++++++-----
> > >  net/ipv6/tcp_ipv6.c                           |  5 +-
> > >  net/netfilter/nf_synproxy_core.c              |  2 +-
> > >  .../selftests/bpf/progs/xdp_synproxy_kern.c   |  4 +-
> > >  17 files changed, 193 insertions(+), 104 deletions(-)
> > >
> > > --
> > > 2.42.0.655.g421f12c284-goog
> > >

Re: [PATCH net-next 00/13] tcp: add optional usec resolution to TCP TS

[Neal Cardwell]

On Fri, Oct 20, 2023 at 8:57 AM Eric Dumazet <edumazet@google.com> wrote:
>
> 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)
>
> Eric Dumazet (13):
>   chtls: fix tp->rcv_tstamp initialization
>   tcp: fix cookie_init_timestamp() overflows
>   tcp: add tcp_time_stamp_ms() helper
>   tcp: introduce tcp_clock_ms()
>   tcp: replace tcp_time_stamp_raw()
>   tcp: rename tcp_skb_timestamp()
>   tcp: move tcp_ns_to_ts() to net/ipv4/syncookies.c
>   tcp: rename tcp_time_stamp() to tcp_time_stamp_ts()
>   tcp: add tcp_rtt_tsopt_us()
>   tcp: add RTAX_FEATURE_TCP_USEC_TS
>   tcp: introduce TCP_PAWS_WRAP
>   tcp: add support for usec resolution in TCP TS values
>   tcp: add TCPI_OPT_USEC_TS
>
>  .../chelsio/inline_crypto/chtls/chtls_cm.c    |  2 +-
>  include/linux/tcp.h                           |  9 ++-
>  include/net/inet_timewait_sock.h              |  3 +-
>  include/net/tcp.h                             | 59 ++++++++++++++-----
>  include/uapi/linux/rtnetlink.h                | 18 +++---
>  include/uapi/linux/tcp.h                      |  1 +
>  net/ipv4/syncookies.c                         | 32 ++++++----
>  net/ipv4/tcp.c                                | 26 +++++---
>  net/ipv4/tcp_input.c                          | 52 ++++++++--------
>  net/ipv4/tcp_ipv4.c                           |  5 +-
>  net/ipv4/tcp_lp.c                             |  2 +-
>  net/ipv4/tcp_minisocks.c                      | 19 ++++--
>  net/ipv4/tcp_output.c                         | 14 +++--
>  net/ipv4/tcp_timer.c                          | 44 +++++++++-----
>  net/ipv6/tcp_ipv6.c                           |  5 +-
>  net/netfilter/nf_synproxy_core.c              |  2 +-
>  .../selftests/bpf/progs/xdp_synproxy_kern.c   |  4 +-
>  17 files changed, 193 insertions(+), 104 deletions(-)
>
> --

Thanks for sending this upstream, Eric! Great to have this upstream.
+1 to the nice benefits mentioned by Yuchung and Van.

The whole patch series looks great to me.

Acked-by: Neal Cardwell <ncardwell@google.com>

thanks,
neal

Re: [PATCH net-next 00/13] tcp: add optional usec resolution to TCP TS

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (main)
by David S. Miller <davem@davemloft.net>:

On Fri, 20 Oct 2023 12:57:35 +0000 you wrote:
> 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.
> 
> [...]

Here is the summary with links:
  - [net-next,01/13] chtls: fix tp->rcv_tstamp initialization
    https://git.kernel.org/netdev/net-next/c/225d9ddbacb1
  - [net-next,02/13] tcp: fix cookie_init_timestamp() overflows
    https://git.kernel.org/netdev/net-next/c/73ed8e03388d
  - [net-next,03/13] tcp: add tcp_time_stamp_ms() helper
    https://git.kernel.org/netdev/net-next/c/99d679556d73
  - [net-next,04/13] tcp: introduce tcp_clock_ms()
    https://git.kernel.org/netdev/net-next/c/2a7c8d291ffe
  - [net-next,05/13] tcp: replace tcp_time_stamp_raw()
    https://git.kernel.org/netdev/net-next/c/16cf6477741b
  - [net-next,06/13] tcp: rename tcp_skb_timestamp()
    https://git.kernel.org/netdev/net-next/c/d1a02ed66fe6
  - [net-next,07/13] tcp: move tcp_ns_to_ts() to net/ipv4/syncookies.c
    https://git.kernel.org/netdev/net-next/c/003e07a1e48e
  - [net-next,08/13] tcp: rename tcp_time_stamp() to tcp_time_stamp_ts()
    https://git.kernel.org/netdev/net-next/c/9d0c00f5ca05
  - [net-next,09/13] tcp: add tcp_rtt_tsopt_us()
    https://git.kernel.org/netdev/net-next/c/b04c3320885a
  - [net-next,10/13] tcp: add RTAX_FEATURE_TCP_USEC_TS
    https://git.kernel.org/netdev/net-next/c/3d44de9a10ea
  - [net-next,11/13] tcp: introduce TCP_PAWS_WRAP
    https://git.kernel.org/netdev/net-next/c/af7721448a60
  - [net-next,12/13] tcp: add support for usec resolution in TCP TS values
    https://git.kernel.org/netdev/net-next/c/614e8316aa4c
  - [net-next,13/13] tcp: add TCPI_OPT_USEC_TS
    https://git.kernel.org/netdev/net-next/c/a77a0f5c7f23

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html