tcp: Add hystart++ to our cubic implementation.

As promised to the transport call on 11/4/22 here is an implementation
of hystart++ for cubic. It also cleans up the tcp_congestion function
to have a better name. Common variables are moved into the general
cc.h structure so that both cubic and newreno can use them for
hystart++

Reviewed by: Michael Tuexen, Richard Scheffenegger
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D33035

10 files changed, 492 insertions, 204 deletions

diff --git a/share/man/man4/cc_newreno.4 b/share/man/man4/cc_newreno.4
index 1c4f4a39cb36..71d3a239ecd1 100644
--- a/share/man/man4/cc_newreno.4
+++ b/share/man/man4/cc_newreno.4
@@ -75,21 +75,10 @@ the congestion window in response to an ECN congestion signal when
 .Va net.inet.tcp.cc.abe=1
 per: cwnd = (cwnd * CC_NEWRENO_BETA_ECN) / 100.
 Default is 80.
-.It Va CC_NEWRENO_ENABLE_HYSTART
-will enable or disable the application of Hystart++.
-The current implementation allows the values 0, 1, 2 and 3.
-A value of 0 (the default) disables the use of Hystart++.
-Setting the value to 1 enables Hystart++.
-Setting the value to 2 enables Hystart++ but also will cause, on exit from Hystart++'s CSS, to
-set the cwnd to the value of where the increase in RTT first began as
-well as setting ssthresh to the flight at send when we exit CSS.
-Setting a value of 3 will keep the setting of the cwnd the same as 2, but will cause ssthresh
-to be set to the average value between the lowest fas rtt (the value cwnd is
-set to) and the fas value at exit of CSS.
 .PP
 Note that currently the only way to enable
 hystart++ is to enable it via socket option.
-When enabling it a value of 1 will enable precise internet-draft behavior
+When enabling it a value of 1 will enable precise internet-draft (version 4) behavior
 (subject to any MIB variable settings), other setting (2 and 3) are experimental.
 .El
 .PP
@@ -120,32 +109,6 @@ the congestion window in response to an ECN congestion signal when
 .Va net.inet.tcp.cc.abe=1
 per: cwnd = (cwnd * beta_ecn) / 100.
 Default is 80.
-.It Va hystartplusplus.bblogs
-This boolean controls if black box logging will be done for hystart++ events. If set
-to zero (the default) no logging is performed.
-If set to one then black box logs will be generated on all hystart++ events.
-.It Va hystartplusplus.css_rounds
-This value controls the number of rounds that CSS runs for.
-The default value matches the current internet-draft of 5.
-.It Va hystartplusplus.css_growth_div
-This value controls the divisor applied to slowstart during CSS.
-The default value matches the current internet-draft of 4.
-.It Va hystartplusplus.n_rttsamples
-This value controls how many rtt samples must be collected in each round for
-hystart++ to be active.
-The default value matches the current internet-draft of 8.
-.It Va hystartplusplus.maxrtt_thresh
-This value controls the maximum rtt variance clamp when considering if CSS is needed.
-The default value matches the current internet-draft of 16000 (in microseconds).
-For further explanation please see the internet-draft.
-.It Va hystartplusplus.minrtt_thresh
-This value controls the minimum rtt variance clamp when considering if CSS is needed.
-The default value matches the current internet-draft of 4000 (in microseconds).
-For further explanation please see the internet-draft.
-.It Va hystartplusplus.lowcwnd
-This value controls what is the lowest congestion window that the tcp
-stack must be at before hystart++ engages.
-The default value matches the current internet-draft of 16.
 .El
 .Sh SEE ALSO
 .Xr cc_cdg 4 ,
diff --git a/share/man/man4/mod_cc.4 b/share/man/man4/mod_cc.4
index 84edf0b2a8ed..ddc7440a4ca7 100644
--- a/share/man/man4/mod_cc.4
+++ b/share/man/man4/mod_cc.4
@@ -83,7 +83,7 @@ The framework exposes the following variables in the
 branch of the
 .Xr sysctl 3
 MIB:
-.Bl -tag -width ".Va abe_frlossreduce"
+.Bl -tag -width ".Va hystartplusplus.css_growth_div"
 .It Va available
 Read-only list of currently available congestion control algorithms by name.
 .It Va algorithm
@@ -102,10 +102,33 @@ support for ABE and for configuration details.
 .It Va abe_frlossreduce
 If non-zero, apply standard beta instead of ABE-beta during ECN-signalled
 congestion recovery episodes if loss also needs to be repaired.
+.It Va hystartplusplus.bblogs
+This boolean controls if black box logging will be done for hystart++ events.
+If set to zero (the default) no logging is performed.
+If set to one then black box logs will be generated on all hystart++ events.
+.It Va hystartplusplus.css_rounds
+This value controls the number of rounds that CSS runs for.
+The default value matches the current internet-draft of 5.
+.It Va hystartplusplus.css_growth_div
+This value controls the divisor applied to slowstart during CSS.
+The default value matches the current internet-draft of 4.
+.It Va hystartplusplus.n_rttsamples
+This value controls how many rtt samples must be collected in each round for
+hystart++ to be active.
+The default value matches the current internet-draft of 8.
+.It Va hystartplusplus.maxrtt_thresh
+This value controls the maximum rtt variance clamp when considering if CSS is needed.
+The default value matches the current internet-draft of 16000 (in microseconds).
+For further explanation please see the internet-draft.
+.It Va hystartplusplus.minrtt_thresh
+This value controls the minimum rtt variance clamp when considering if CSS is needed.
+The default value matches the current internet-draft of 4000 (in microseconds).
+For further explanation please see the internet-draft.
 .El
 .Pp
 Each congestion control module may also expose other MIB variables
 to control their behaviour.
+Note that both newreno and cubic now support hystart++ based on the version 3 of the internet-draft.
 .Sh Kernel Configuration
 .Pp
 All of the available congestion control modules may also be loaded
diff --git a/sys/netinet/cc/cc.c b/sys/netinet/cc/cc.c
index bcda9276d4d9..647ecb47f18d 100644
--- a/sys/netinet/cc/cc.c
+++ b/sys/netinet/cc/cc.c
@@ -84,6 +84,13 @@ __FBSDID("$FreeBSD$");
 #define CC_DEFAULT "newreno"
 #endif
 
+uint32_t hystart_minrtt_thresh = 4000;
+uint32_t hystart_maxrtt_thresh = 16000;
+uint32_t hystart_n_rttsamples = 8;
+uint32_t hystart_css_growth_div = 4;
+uint32_t hystart_css_rounds = 5;
+uint32_t hystart_bblogs = 0;
+
 MALLOC_DEFINE(M_CC_MEM, "CC Mem", "Congestion Control State memory");
 
 /*
@@ -582,6 +589,40 @@ SYSCTL_PROC(_net_inet_tcp_cc, OID_AUTO, available,
     NULL, 0, cc_list_available, "A",
     "List available congestion control algorithms");
 
+SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, hystartplusplus,
+    CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
+    "New Reno related HyStart++ settings");
+
+SYSCTL_UINT(_net_inet_tcp_cc_hystartplusplus, OID_AUTO, minrtt_thresh,
+    CTLFLAG_RW,
+    &hystart_minrtt_thresh, 4000,
+   "HyStarts++ minimum RTT thresh used in clamp (in microseconds)");
+
+SYSCTL_UINT(_net_inet_tcp_cc_hystartplusplus, OID_AUTO, maxrtt_thresh,
+    CTLFLAG_RW,
+    &hystart_maxrtt_thresh, 16000,
+   "HyStarts++ maximum RTT thresh used in clamp (in microseconds)");
+
+SYSCTL_UINT(_net_inet_tcp_cc_hystartplusplus, OID_AUTO, n_rttsamples,
+    CTLFLAG_RW,
+    &hystart_n_rttsamples, 8,
+   "The number of RTT samples that must be seen to consider HyStart++");
+
+SYSCTL_UINT(_net_inet_tcp_cc_hystartplusplus, OID_AUTO, css_growth_div,
+    CTLFLAG_RW,
+    &hystart_css_growth_div, 4,
+   "The divisor to the growth when in Hystart++ CSS");
+
+SYSCTL_UINT(_net_inet_tcp_cc_hystartplusplus, OID_AUTO, css_rounds,
+    CTLFLAG_RW,
+    &hystart_css_rounds, 5,
+   "The number of rounds HyStart++ lasts in CSS before falling to CA");
+
+SYSCTL_UINT(_net_inet_tcp_cc_hystartplusplus, OID_AUTO, bblogs,
+    CTLFLAG_RW,
+    &hystart_bblogs, 0,
+   "Do we enable HyStart++ Black Box logs to be generated if BB logging is on");
+
 VNET_DEFINE(int, cc_do_abe) = 0;
 SYSCTL_INT(_net_inet_tcp_cc, OID_AUTO, abe, CTLFLAG_VNET | CTLFLAG_RW,
     &VNET_NAME(cc_do_abe), 0,
diff --git a/sys/netinet/cc/cc.h b/sys/netinet/cc/cc.h
index 4282f0585bbd..6f942da7aa83 100644
--- a/sys/netinet/cc/cc.h
+++ b/sys/netinet/cc/cc.h
@@ -72,6 +72,15 @@ VNET_DECLARE(int, cc_abe_frlossreduce);
 /* Define the new net.inet.tcp.cc sysctl tree. */
 SYSCTL_DECL(_net_inet_tcp_cc);
 
+/* For CC modules that use hystart++ */
+extern uint32_t hystart_lowcwnd;
+extern uint32_t hystart_minrtt_thresh;
+extern uint32_t hystart_maxrtt_thresh;
+extern uint32_t hystart_n_rttsamples;
+extern uint32_t hystart_css_growth_div;
+extern uint32_t hystart_css_rounds;
+extern uint32_t hystart_bblogs;
+
 /* CC housekeeping functions. */
 int	cc_register_algo(struct cc_algo *add_cc);
 int	cc_deregister_algo(struct cc_algo *remove_cc);
@@ -106,6 +115,9 @@ struct cc_var {
 #define	CCF_CHG_MAX_CWND	0x0080	/* Cubic max_cwnd changed, for K */
 #define	CCF_USR_IWND		0x0100	/* User specified initial window */
 #define	CCF_USR_IWND_INIT_NSEG	0x0200	/* Convert segs to bytes on conn init */
+#define CCF_HYSTART_ALLOWED	0x0400	/* If the CC supports it Hystart is allowed */
+#define CCF_HYSTART_CAN_SH_CWND	0x0800  /* Can hystart when going CSS -> CA slam the cwnd */
+#define CCF_HYSTART_CONS_SSTH	0x1000	/* Should hystart use the more conservative ssthresh */
 
 /* ACK types passed to the ack_received() hook. */
 #define	CC_ACK		0x0001	/* Regular in sequence ACK. */
diff --git a/sys/netinet/cc/cc_cubic.c b/sys/netinet/cc/cc_cubic.c
index a4146423b0ae..e7d86a25156a 100644
--- a/sys/netinet/cc/cc_cubic.c
+++ b/sys/netinet/cc/cc_cubic.c
@@ -70,6 +70,8 @@ __FBSDID("$FreeBSD$");
 #include <netinet/tcp_seq.h>
 #include <netinet/tcp_timer.h>
 #include <netinet/tcp_var.h>
+#include <netinet/tcp_log_buf.h>
+#include <netinet/tcp_hpts.h>
 #include <netinet/cc/cc.h>
 #include <netinet/cc/cc_cubic.h>
 #include <netinet/cc/cc_module.h>
@@ -85,39 +87,9 @@ static void	cubic_record_rtt(struct cc_var *ccv);
 static void	cubic_ssthresh_update(struct cc_var *ccv, uint32_t maxseg);
 static void	cubic_after_idle(struct cc_var *ccv);
 static size_t	cubic_data_sz(void);
-
-struct cubic {
-	/* Cubic K in fixed point form with CUBIC_SHIFT worth of precision. */
-	int64_t		K;
-	/* Sum of RTT samples across an epoch in ticks. */
-	int64_t		sum_rtt_ticks;
-	/* cwnd at the most recent congestion event. */
-	unsigned long	max_cwnd;
-	/* cwnd at the previous congestion event. */
-	unsigned long	prev_max_cwnd;
-	/* A copy of prev_max_cwnd. Used for CC_RTO_ERR */
-	unsigned long	prev_max_cwnd_cp;
-	/* various flags */
-	uint32_t	flags;
-#define CUBICFLAG_CONG_EVENT	0x00000001	/* congestion experienced */
-#define CUBICFLAG_IN_SLOWSTART	0x00000002	/* in slow start */
-#define CUBICFLAG_IN_APPLIMIT	0x00000004	/* application limited */
-#define CUBICFLAG_RTO_EVENT	0x00000008	/* RTO experienced */
-	/* Minimum observed rtt in ticks. */
-	int		min_rtt_ticks;
-	/* Mean observed rtt between congestion epochs. */
-	int		mean_rtt_ticks;
-	/* ACKs since last congestion event. */
-	int		epoch_ack_count;
-	/* Timestamp (in ticks) of arriving in congestion avoidance from last
-	 * congestion event.
-	 */
-	int		t_last_cong;
-	/* Timestamp (in ticks) of a previous congestion event. Used for
-	 * CC_RTO_ERR.
-	 */
-	int		t_last_cong_prev;
-};
+static void	cubic_newround(struct cc_var *ccv, uint32_t round_cnt);
+static void	cubic_rttsample(struct cc_var *ccv, uint32_t usec_rtt,
+       uint32_t rxtcnt, uint32_t fas);
 
 struct cc_algo cubic_cc_algo = {
 	.name = "cubic",
@@ -129,10 +101,141 @@ struct cc_algo cubic_cc_algo = {
 	.mod_init = cubic_mod_init,
 	.post_recovery = cubic_post_recovery,
 	.after_idle = cubic_after_idle,
-	.cc_data_sz = cubic_data_sz
+	.cc_data_sz = cubic_data_sz,
+	.rttsample = cubic_rttsample,
+	.newround = cubic_newround
 };
 
 static void
+cubic_log_hystart_event(struct cc_var *ccv, struct cubic *cubicd, uint8_t mod, uint32_t flex1)
+{
+	/*
+	 * Types of logs (mod value)
+	 * 1 - rtt_thresh in flex1, checking to see if RTT is to great.
+	 * 2 - rtt is too great, rtt_thresh in flex1.
+	 * 3 - CSS is active incr in flex1
+	 * 4 - A new round is beginning flex1 is round count
+	 * 5 - A new RTT measurement flex1 is the new measurement.
+	 * 6 - We enter CA ssthresh is also in flex1.
+	 * 7 - Socket option to change hystart executed opt.val in flex1.
+	 * 8 - Back out of CSS into SS, flex1 is the css_baseline_minrtt
+	 * 9 - We enter CA, via an ECN mark.
+	 * 10 - We enter CA, via a loss.
+	 * 11 - We have slipped out of SS into CA via cwnd growth.
+	 * 12 - After idle has re-enabled hystart++
+	 */
+	struct tcpcb *tp;
+
+	if (hystart_bblogs == 0)
+		return;
+	tp = ccv->ccvc.tcp;
+	if (tp->t_logstate != TCP_LOG_STATE_OFF) {
+		union tcp_log_stackspecific log;
+		struct timeval tv;
+
+		memset(&log, 0, sizeof(log));
+		log.u_bbr.flex1 = flex1;
+		log.u_bbr.flex2 = cubicd->css_current_round_minrtt;
+		log.u_bbr.flex3 = cubicd->css_lastround_minrtt;
+		log.u_bbr.flex4 = cubicd->css_rttsample_count;
+		log.u_bbr.flex5 = cubicd->css_entered_at_round;
+		log.u_bbr.flex6 = cubicd->css_baseline_minrtt;
+		/* We only need bottom 16 bits of flags */
+		log.u_bbr.flex7 = cubicd->flags & 0x0000ffff;
+		log.u_bbr.flex8 = mod;
+		log.u_bbr.epoch = cubicd->css_current_round;
+		log.u_bbr.timeStamp = tcp_get_usecs(&tv);
+		log.u_bbr.lt_epoch = cubicd->css_fas_at_css_entry;
+		log.u_bbr.pkts_out = cubicd->css_last_fas;
+		log.u_bbr.delivered = cubicd->css_lowrtt_fas;
+		log.u_bbr.pkt_epoch = ccv->flags;
+		TCP_LOG_EVENTP(tp, NULL,
+		    &tp->t_inpcb->inp_socket->so_rcv,
+		    &tp->t_inpcb->inp_socket->so_snd,
+		    TCP_HYSTART, 0,
+		    0, &log, false, &tv);
+	}
+}
+
+static void
+cubic_does_slow_start(struct cc_var *ccv, struct cubic *cubicd)
+{
+	/*
+	 * In slow-start with ABC enabled and no RTO in sight?
+	 * (Must not use abc_l_var > 1 if slow starting after
+	 * an RTO. On RTO, snd_nxt = snd_una, so the
+	 * snd_nxt == snd_max check is sufficient to
+	 * handle this).
+	 *
+	 * XXXLAS: Find a way to signal SS after RTO that
+	 * doesn't rely on tcpcb vars.
+	 */
+	u_int cw = CCV(ccv, snd_cwnd);
+	u_int incr = CCV(ccv, t_maxseg);
+	uint16_t abc_val;
+
+	cubicd->flags |= CUBICFLAG_IN_SLOWSTART;
+	if (ccv->flags & CCF_USE_LOCAL_ABC)
+		abc_val = ccv->labc;
+	else
+		abc_val = V_tcp_abc_l_var;
+	if ((ccv->flags & CCF_HYSTART_ALLOWED) &&
+	    (cubicd->flags & CUBICFLAG_HYSTART_ENABLED) &&
+	    ((cubicd->flags & CUBICFLAG_HYSTART_IN_CSS) == 0)) {
+		/*
+		 * Hystart is allowed and still enabled and we are not yet
+		 * in CSS. Lets check to see if we can make a decision on
+		 * if we need to go into CSS.
+		 */
+		if ((cubicd->css_rttsample_count >= hystart_n_rttsamples) &&
+		    (cubicd->css_current_round_minrtt != 0xffffffff) &&
+		    (cubicd->css_lastround_minrtt != 0xffffffff)) {
+			uint32_t rtt_thresh;
+
+			/* Clamp (minrtt_thresh, lastround/8, maxrtt_thresh) */
+			rtt_thresh = (cubicd->css_lastround_minrtt >> 3);
+			if (rtt_thresh < hystart_minrtt_thresh)
+				rtt_thresh = hystart_minrtt_thresh;
+			if (rtt_thresh > hystart_maxrtt_thresh)
+				rtt_thresh = hystart_maxrtt_thresh;
+			cubic_log_hystart_event(ccv, cubicd, 1, rtt_thresh);
+
+			if (cubicd->css_current_round_minrtt >= (cubicd->css_lastround_minrtt + rtt_thresh)) {
+				/* Enter CSS */
+				cubicd->flags |= CUBICFLAG_HYSTART_IN_CSS;
+				cubicd->css_fas_at_css_entry = cubicd->css_lowrtt_fas;
+				/* 
+				 * The draft (v4) calls for us to set baseline to css_current_round_min
+				 * but that can cause an oscillation. We probably shoudl be using
+				 * css_lastround_minrtt, but the authors insist that will cause
+				 * issues on exiting early. We will leave the draft version for now
+				 * but I suspect this is incorrect.
+				 */
+				cubicd->css_baseline_minrtt = cubicd->css_current_round_minrtt;
+				cubicd->css_entered_at_round = cubicd->css_current_round;
+				cubic_log_hystart_event(ccv, cubicd, 2, rtt_thresh);
+			}
+		}
+	}
+	if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max))
+		incr = min(ccv->bytes_this_ack,
+			   ccv->nsegs * abc_val *
+			   CCV(ccv, t_maxseg));
+	else
+		incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg));
+
+	/* Only if Hystart is enabled will the flag get set */
+	if (cubicd->flags & CUBICFLAG_HYSTART_IN_CSS) {
+		incr /= hystart_css_growth_div;
+		cubic_log_hystart_event(ccv, cubicd, 3, incr);
+	}
+	/* ABC is on by default, so incr equals 0 frequently. */
+	if (incr > 0)
+		CCV(ccv, snd_cwnd) = min((cw + incr),
+					 TCP_MAXWIN << CCV(ccv, snd_scale));
+}
+
+static void
 cubic_ack_received(struct cc_var *ccv, uint16_t type)
 {
 	struct cubic *cubic_data;
@@ -151,9 +254,19 @@ cubic_ack_received(struct cc_var *ccv, uint16_t type)
 		 /* Use the logic in NewReno ack_received() for slow start. */
 		if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh) ||
 		    cubic_data->min_rtt_ticks == TCPTV_SRTTBASE) {
-			cubic_data->flags |= CUBICFLAG_IN_SLOWSTART;
-			newreno_cc_ack_received(ccv, type);
+			cubic_does_slow_start(ccv, cubic_data);
 		} else {
+			if (cubic_data->flags & CUBICFLAG_HYSTART_IN_CSS) {
+				/*
+				 * We have slipped into CA with
+				 * CSS active. Deactivate all.
+				 */
+				/* Turn off the CSS flag */
+				cubic_data->flags &= ~CUBICFLAG_HYSTART_IN_CSS;
+				/* Disable use of CSS in the future except long idle  */
+				cubic_data->flags &= ~CUBICFLAG_HYSTART_ENABLED;
+				cubic_log_hystart_event(ccv, cubic_data, 11, CCV(ccv, snd_ssthresh));
+			}
 			if ((cubic_data->flags & CUBICFLAG_RTO_EVENT) &&
 			    (cubic_data->flags & CUBICFLAG_IN_SLOWSTART)) {
 				/* RFC8312 Section 4.7 */
@@ -245,7 +358,14 @@ cubic_after_idle(struct cc_var *ccv)
 
 	cubic_data->max_cwnd = ulmax(cubic_data->max_cwnd, CCV(ccv, snd_cwnd));
 	cubic_data->K = cubic_k(cubic_data->max_cwnd / CCV(ccv, t_maxseg));
-
+	if ((cubic_data->flags & CUBICFLAG_HYSTART_ENABLED) == 0) {
+		/*
+		 * Re-enable hystart if we have been idle.
+		 */
+		cubic_data->flags &= ~CUBICFLAG_HYSTART_IN_CSS;
+		cubic_data->flags |= CUBICFLAG_HYSTART_ENABLED;
+		cubic_log_hystart_event(ccv, cubic_data, 12, CCV(ccv, snd_ssthresh));
+	}
 	newreno_cc_after_idle(ccv);
 	cubic_data->t_last_cong = ticks;
 }
@@ -281,6 +401,17 @@ cubic_cb_init(struct cc_var *ccv, void *ptr)
 	cubic_data->mean_rtt_ticks = 1;
 
 	ccv->cc_data = cubic_data;
+	cubic_data->flags = CUBICFLAG_HYSTART_ENABLED;
+	/* At init set both to infinity */
+	cubic_data->css_lastround_minrtt = 0xffffffff;
+	cubic_data->css_current_round_minrtt = 0xffffffff;
+	cubic_data->css_current_round = 0;
+	cubic_data->css_baseline_minrtt = 0xffffffff;
+	cubic_data->css_rttsample_count = 0;
+	cubic_data->css_entered_at_round = 0;
+	cubic_data->css_fas_at_css_entry = 0;
+	cubic_data->css_lowrtt_fas = 0;
+	cubic_data->css_last_fas = 0;
 
 	return (0);
 }
@@ -299,6 +430,12 @@ cubic_cong_signal(struct cc_var *ccv, uint32_t type)
 
 	switch (type) {
 	case CC_NDUPACK:
+		if (cubic_data->flags & CUBICFLAG_HYSTART_ENABLED) {
+			/* Make sure the flags are all off we had a loss */
+			cubic_data->flags &= ~CUBICFLAG_HYSTART_ENABLED;
+			cubic_data->flags &= ~CUBICFLAG_HYSTART_IN_CSS;
+			cubic_log_hystart_event(ccv, cubic_data, 10, CCV(ccv, snd_ssthresh));
+		}
 		if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) {
 			if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
 				cubic_ssthresh_update(ccv, mss);
@@ -311,6 +448,12 @@ cubic_cong_signal(struct cc_var *ccv, uint32_t type)
 		break;
 
 	case CC_ECN:
+		if (cubic_data->flags & CUBICFLAG_HYSTART_ENABLED) {
+			/* Make sure the flags are all off we had a loss */
+			cubic_data->flags &= ~CUBICFLAG_HYSTART_ENABLED;
+			cubic_data->flags &= ~CUBICFLAG_HYSTART_IN_CSS;
+			cubic_log_hystart_event(ccv, cubic_data, 9, CCV(ccv, snd_ssthresh));
+		}
 		if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
 			cubic_ssthresh_update(ccv, mss);
 			cubic_data->flags |= CUBICFLAG_CONG_EVENT;
@@ -495,5 +638,79 @@ cubic_ssthresh_update(struct cc_var *ccv, uint32_t maxseg)
 	CCV(ccv, snd_ssthresh) = max(ssthresh, 2 * maxseg);
 }
 
+static void
+cubic_rttsample(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32_t fas)
+{
+	struct cubic *cubicd;
+
+	cubicd = ccv->cc_data;
+	if (rxtcnt > 1) {
+		/*
+		 * Only look at RTT's that are non-ambiguous.
+		 */
+		return;
+	}
+	cubicd->css_rttsample_count++;
+	cubicd->css_last_fas = fas;
+	if (cubicd->css_current_round_minrtt > usec_rtt) {
+		cubicd->css_current_round_minrtt = usec_rtt;
+		cubicd->css_lowrtt_fas = cubicd->css_last_fas;
+	}
+	if ((cubicd->css_rttsample_count >= hystart_n_rttsamples) &&
+	    (cubicd->css_current_round_minrtt != 0xffffffff) &&
+	    (cubicd->css_lastround_minrtt != 0xffffffff)) {
+		/*
+		 * We were in CSS and the RTT is now less, we
+		 * entered CSS erroneously.
+		 */
+		cubicd->flags &= ~CUBICFLAG_HYSTART_IN_CSS;
+		cubic_log_hystart_event(ccv, cubicd, 8, cubicd->css_baseline_minrtt);
+		cubicd->css_baseline_minrtt = 0xffffffff;
+	}
+	if (cubicd->flags & CUBICFLAG_HYSTART_ENABLED)
+		cubic_log_hystart_event(ccv, cubicd, 5, usec_rtt);
+}
+
+static void
+cubic_newround(struct cc_var *ccv, uint32_t round_cnt)
+{
+	struct cubic *cubicd;
+
+	cubicd = ccv->cc_data;
+	/* We have entered a new round */
+	cubicd->css_lastround_minrtt = cubicd->css_current_round_minrtt;
+	cubicd->css_current_round_minrtt = 0xffffffff;
+	cubicd->css_rttsample_count = 0;
+	cubicd->css_current_round = round_cnt;
+	if ((cubicd->flags & CUBICFLAG_HYSTART_IN_CSS) &&
+	    ((round_cnt - cubicd->css_entered_at_round) >= hystart_css_rounds)) {
+		/* Enter CA */
+		if (ccv->flags & CCF_HYSTART_CAN_SH_CWND) {
+			/*
+			 * We engage more than snd_ssthresh, engage
+			 * the brakes!! Though we will stay in SS to
+			 * creep back up again, so lets leave CSS active
+			 * and give us hystart_css_rounds more rounds.
+			 */
+			if (ccv->flags & CCF_HYSTART_CONS_SSTH) {
+				CCV(ccv, snd_ssthresh) = ((cubicd->css_lowrtt_fas + cubicd->css_fas_at_css_entry) / 2);
+			} else {
+				CCV(ccv, snd_ssthresh) = cubicd->css_lowrtt_fas;
+			}
+			CCV(ccv, snd_cwnd) = cubicd->css_fas_at_css_entry;
+			cubicd->css_entered_at_round = round_cnt;
+		} else {
+			CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd);
+			/* Turn off the CSS flag */
+			cubicd->flags &= ~CUBICFLAG_HYSTART_IN_CSS;
+			/* Disable use of CSS in the future except long idle  */
+			cubicd->flags &= ~CUBICFLAG_HYSTART_ENABLED;
+		}
+		cubic_log_hystart_event(ccv, cubicd, 6, CCV(ccv, snd_ssthresh));
+	}
+	if (cubicd->flags & CUBICFLAG_HYSTART_ENABLED)
+		cubic_log_hystart_event(ccv, cubicd, 4, round_cnt);
+}
+
 DECLARE_CC_MODULE(cubic, &cubic_cc_algo);
 MODULE_VERSION(cubic, 2);
diff --git a/sys/netinet/cc/cc_cubic.h b/sys/netinet/cc/cc_cubic.h
index 13681aae793c..2042590182fc 100644
--- a/sys/netinet/cc/cc_cubic.h
+++ b/sys/netinet/cc/cc_cubic.h
@@ -78,6 +78,55 @@
  */
 #define	CUBED_ROOT_MAX_ULONG	448845
 
+/* Flags used in the cubic structure */
+#define CUBICFLAG_CONG_EVENT		0x00000001	/* congestion experienced */
+#define CUBICFLAG_IN_SLOWSTART		0x00000002	/* in slow start */
+#define CUBICFLAG_IN_APPLIMIT		0x00000004	/* application limited */
+#define CUBICFLAG_RTO_EVENT		0x00000008	/* RTO experienced */
+#define CUBICFLAG_HYSTART_ENABLED	0x00000010	/* Hystart++ is enabled */
+#define CUBICFLAG_HYSTART_IN_CSS	0x00000020	/* We are in Hystart++ CSS */
+
+/* Kernel only bits */
+#ifdef _KERNEL
+struct cubic {
+	/* Cubic K in fixed point form with CUBIC_SHIFT worth of precision. */
+	int64_t		K;
+	/* Sum of RTT samples across an epoch in ticks. */
+	int64_t		sum_rtt_ticks;
+	/* cwnd at the most recent congestion event. */
+	unsigned long	max_cwnd;
+	/* cwnd at the previous congestion event. */
+	unsigned long	prev_max_cwnd;
+	/* A copy of prev_max_cwnd. Used for CC_RTO_ERR */
+	unsigned long	prev_max_cwnd_cp;
+	/* various flags */
+	uint32_t	flags;
+	/* Minimum observed rtt in ticks. */
+	int		min_rtt_ticks;
+	/* Mean observed rtt between congestion epochs. */
+	int		mean_rtt_ticks;
+	/* ACKs since last congestion event. */
+	int		epoch_ack_count;
+	/* Timestamp (in ticks) of arriving in congestion avoidance from last
+	 * congestion event.
+	 */
+	int		t_last_cong;
+	/* Timestamp (in ticks) of a previous congestion event. Used for
+	 * CC_RTO_ERR.
+	 */
+	int		t_last_cong_prev;
+	uint32_t css_baseline_minrtt;
+	uint32_t css_current_round_minrtt;
+	uint32_t css_lastround_minrtt;
+	uint32_t css_rttsample_count;
+	uint32_t css_entered_at_round;
+	uint32_t css_current_round;
+	uint32_t css_fas_at_css_entry;
+	uint32_t css_lowrtt_fas;
+	uint32_t css_last_fas;
+};
+#endif
+
 /* Userland only bits. */
 #ifndef _KERNEL
 
diff --git a/sys/netinet/cc/cc_newreno.c b/sys/netinet/cc/cc_newreno.c
index 3280e270ede3..06b53d76d8e9 100644
--- a/sys/netinet/cc/cc_newreno.c
+++ b/sys/netinet/cc/cc_newreno.c
@@ -116,14 +116,6 @@ struct cc_algo newreno_cc_algo = {
 	.cc_data_sz = newreno_data_sz,
 };
 
-static uint32_t hystart_lowcwnd = 16;
-static uint32_t hystart_minrtt_thresh = 4000;
-static uint32_t hystart_maxrtt_thresh = 16000;
-static uint32_t hystart_n_rttsamples = 8;
-static uint32_t hystart_css_growth_div = 4;
-static uint32_t hystart_css_rounds = 5;
-static uint32_t hystart_bblogs = 0;
-
 static void
 newreno_log_hystart_event(struct cc_var *ccv, struct newreno *nreno, uint8_t mod, uint32_t flex1)
 {
@@ -137,6 +129,10 @@ newreno_log_hystart_event(struct cc_var *ccv, struct newreno *nreno, uint8_t mod
 	 * 6 - We enter CA ssthresh is also in flex1.
 	 * 7 - Socket option to change hystart executed opt.val in flex1.
 	 * 8 - Back out of CSS into SS, flex1 is the css_baseline_minrtt
+	 * 9 - We enter CA, via an ECN mark.
+	 * 10 - We enter CA, via a loss.
+	 * 11 - We have slipped out of SS into CA via cwnd growth.
+	 * 12 - After idle has re-enabled hystart++
 	 */
 	struct tcpcb *tp;
 
@@ -162,6 +158,7 @@ newreno_log_hystart_event(struct cc_var *ccv, struct newreno *nreno, uint8_t mod
 		log.u_bbr.lt_epoch = nreno->css_fas_at_css_entry;
 		log.u_bbr.pkts_out = nreno->css_last_fas;
 		log.u_bbr.delivered = nreno->css_lowrtt_fas;
+		log.u_bbr.pkt_epoch = ccv->flags;
 		TCP_LOG_EVENTP(tp, NULL,
 		    &tp->t_inpcb->inp_socket->so_rcv,
 		    &tp->t_inpcb->inp_socket->so_snd,
@@ -265,6 +262,7 @@ newreno_ack_received(struct cc_var *ccv, uint16_t type)
 				nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS;
 				/* Disable use of CSS in the future except long idle  */
 				nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED;
+				newreno_log_hystart_event(ccv, nreno, 11, CCV(ccv, snd_ssthresh));
 			}
 			if (V_tcp_do_rfc3465) {
 				if (ccv->flags & CCF_ABC_SENTAWND)
@@ -290,7 +288,7 @@ newreno_ack_received(struct cc_var *ccv, uint16_t type)
 				abc_val = ccv->labc;
 			else
 				abc_val = V_tcp_abc_l_var;
-			if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_ALLOWED) &&
+			if ((ccv->flags & CCF_HYSTART_ALLOWED) &&
 			    (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED) &&
 			    ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) == 0)) {
 				/*
@@ -299,8 +297,8 @@ newreno_ack_received(struct cc_var *ccv, uint16_t type)
 				 * if we need to go into CSS.
 				 */
 				if ((nreno->css_rttsample_count >= hystart_n_rttsamples) &&
-				    (CCV(ccv, snd_cwnd) >
-				     (hystart_lowcwnd * tcp_fixed_maxseg(ccv->ccvc.tcp)))) {
+				    (nreno->css_current_round_minrtt != 0xffffffff) &&
+				    (nreno->css_lastround_minrtt != 0xffffffff)) {
 					uint32_t rtt_thresh;
 
 					/* Clamp (minrtt_thresh, lastround/8, maxrtt_thresh) */
@@ -314,6 +312,13 @@ newreno_ack_received(struct cc_var *ccv, uint16_t type)
 						/* Enter CSS */
 						nreno->newreno_flags |= CC_NEWRENO_HYSTART_IN_CSS;
 						nreno->css_fas_at_css_entry = nreno->css_lowrtt_fas;
+						/*
+						 * The draft (v4) calls for us to set baseline to css_current_round_min
+						 * but that can cause an oscillation. We probably shoudl be using
+						 * css_lastround_minrtt, but the authors insist that will cause
+						 * issues on exiting early. We will leave the draft version for now
+						 * but I suspect this is incorrect.
+						 */
 						nreno->css_baseline_minrtt = nreno->css_current_round_minrtt;
 						nreno->css_entered_at_round = nreno->css_current_round;
 						newreno_log_hystart_event(ccv, nreno, 2, rtt_thresh);
@@ -348,14 +353,12 @@ newreno_after_idle(struct cc_var *ccv)
 	nreno = ccv->cc_data;
 	newreno_cc_after_idle(ccv);
 	if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED) == 0) {
-		if (CCV(ccv, snd_cwnd) <= (hystart_lowcwnd * tcp_fixed_maxseg(ccv->ccvc.tcp))) {
-			/*
-			 * Re-enable hystart if our cwnd has fallen below
-			 * the hystart lowcwnd point.
-			 */
-			nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS;
-			nreno->newreno_flags |= CC_NEWRENO_HYSTART_ENABLED;
-		}
+		/*
+		 * Re-enable hystart if we have been idle.
+		 */
+		nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS;
+		nreno->newreno_flags |= CC_NEWRENO_HYSTART_ENABLED;
+		newreno_log_hystart_event(ccv, nreno, 12, CCV(ccv, snd_ssthresh));
 	}
 }
 
@@ -400,6 +403,7 @@ newreno_cong_signal(struct cc_var *ccv, uint32_t type)
 			/* Make sure the flags are all off we had a loss */
 			nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED;
 			nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS;
+			newreno_log_hystart_event(ccv, nreno, 10, CCV(ccv, snd_ssthresh));
 		}
 		if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) {
 			if (IN_CONGRECOVERY(CCV(ccv, t_flags) &&
@@ -418,6 +422,7 @@ newreno_cong_signal(struct cc_var *ccv, uint32_t type)
 			/* Make sure the flags are all off we had a loss */
 			nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_ENABLED;
 			nreno->newreno_flags &= ~CC_NEWRENO_HYSTART_IN_CSS;
+			newreno_log_hystart_event(ccv, nreno, 9, CCV(ccv, snd_ssthresh));
 		}
 		if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
 			CCV(ccv, snd_ssthresh) = cwin;
@@ -460,18 +465,6 @@ newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf)
 			nreno->beta_ecn = opt->val;
 			nreno->newreno_flags |= CC_NEWRENO_BETA_ECN_ENABLED;
 			break;
-		case CC_NEWRENO_ENABLE_HYSTART:
-			/* Allow hystart on this connection */
-			if (opt->val != 0) {
-				nreno->newreno_flags |= CC_NEWRENO_HYSTART_ALLOWED;
-				if (opt->val > 1)
-					nreno->newreno_flags |= CC_NEWRENO_HYSTART_CAN_SH_CWND;
-				if (opt->val > 2)
-					nreno->newreno_flags |= CC_NEWRENO_HYSTART_CONS_SSTH;
-			} else
-				nreno->newreno_flags &= ~(CC_NEWRENO_HYSTART_ALLOWED|CC_NEWRENO_HYSTART_CAN_SH_CWND|CC_NEWRENO_HYSTART_CONS_SSTH);
-			newreno_log_hystart_event(ccv, nreno, 7, opt->val);
-			break;
 		default:
 			return (ENOPROTOOPT);
 		}
@@ -486,17 +479,6 @@ newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf)
 			opt->val = (nreno == NULL) ?
 			    V_newreno_beta_ecn : nreno->beta_ecn;
 			break;
-		case CC_NEWRENO_ENABLE_HYSTART:
-			if (nreno->newreno_flags & CC_NEWRENO_HYSTART_ALLOWED) {
-				if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CONS_SSTH)
-					opt->val = 3;
-				else if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CAN_SH_CWND)
-					opt->val = 2;
-				else
-					opt->val = 1;
-			} else
-				opt->val = 0;
-			break;
 		default:
 			return (ENOPROTOOPT);
 		}
@@ -542,14 +524,14 @@ newreno_newround(struct cc_var *ccv, uint32_t round_cnt)
 	if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) &&
 	    ((round_cnt - nreno->css_entered_at_round) >= hystart_css_rounds)) {
 		/* Enter CA */
-		if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CAN_SH_CWND) {
+		if (ccv->flags & CCF_HYSTART_CAN_SH_CWND) {
 			/*
 			 * We engage more than snd_ssthresh, engage
 			 * the brakes!! Though we will stay in SS to
 			 * creep back up again, so lets leave CSS active
 			 * and give us hystart_css_rounds more rounds.
 			 */
-			if (nreno->newreno_flags & CC_NEWRENO_HYSTART_CONS_SSTH) {
+			if (ccv->flags & CCF_HYSTART_CONS_SSTH) {
 				CCV(ccv, snd_ssthresh) = ((nreno->css_lowrtt_fas + nreno->css_fas_at_css_entry) / 2);
 			} else {
 				CCV(ccv, snd_ssthresh) = nreno->css_lowrtt_fas;
@@ -565,7 +547,8 @@ newreno_newround(struct cc_var *ccv, uint32_t round_cnt)
 		}
 		newreno_log_hystart_event(ccv, nreno, 6, CCV(ccv, snd_ssthresh));
 	}
-	newreno_log_hystart_event(ccv, nreno, 4, round_cnt);
+	if (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED)
+		newreno_log_hystart_event(ccv, nreno, 4, round_cnt);
 }
 
 static void
@@ -586,9 +569,9 @@ newreno_rttsample(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32
 		nreno->css_current_round_minrtt = usec_rtt;
 		nreno->css_lowrtt_fas = nreno->css_last_fas;
 	}
-	if ((nreno->newreno_flags & CC_NEWRENO_HYSTART_IN_CSS) &&
-	    (nreno->css_rttsample_count >= hystart_n_rttsamples) &&
-	    (nreno->css_baseline_minrtt > nreno->css_current_round_minrtt)) {
+	if ((nreno->css_rttsample_count >= hystart_n_rttsamples) &&
+	    (nreno->css_current_round_minrtt != 0xffffffff) &&
+	    (nreno->css_lastround_minrtt != 0xffffffff)) {
 		/*
 		 * We were in CSS and the RTT is now less, we
 		 * entered CSS erroneously.
@@ -597,7 +580,8 @@ newreno_rttsample(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32
 		newreno_log_hystart_event(ccv, nreno, 8, nreno->css_baseline_minrtt);
 		nreno->css_baseline_minrtt = 0xffffffff;
 	}
-	newreno_log_hystart_event(ccv, nreno, 5, usec_rtt);
+	if (nreno->newreno_flags & CC_NEWRENO_HYSTART_ENABLED)
+		newreno_log_hystart_event(ccv, nreno, 5, usec_rtt);
 }
 
 SYSCTL_DECL(_net_inet_tcp_cc_newreno);
@@ -615,45 +599,5 @@ SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta_ecn,
     &VNET_NAME(newreno_beta_ecn), 3, &newreno_beta_handler, "IU",
     "New Reno beta ecn, specified as number between 1 and 100");
 
-SYSCTL_NODE(_net_inet_tcp_cc_newreno, OID_AUTO, hystartplusplus,
-    CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
-    "New Reno related HyStart++ settings");
-
-SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, lowcwnd,
-    CTLFLAG_RW,
-    &hystart_lowcwnd, 16,
-   "The number of MSS in the CWND before HyStart++ is active");
-
-SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, minrtt_thresh,
-    CTLFLAG_RW,
-    &hystart_minrtt_thresh, 4000,
-   "HyStarts++ minimum RTT thresh used in clamp (in microseconds)");
-
-SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, maxrtt_thresh,
-    CTLFLAG_RW,
-    &hystart_maxrtt_thresh, 16000,
-   "HyStarts++ maximum RTT thresh used in clamp (in microseconds)");
-
-SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, n_rttsamples,
-    CTLFLAG_RW,
-    &hystart_n_rttsamples, 8,
-   "The number of RTT samples that must be seen to consider HyStart++");
-
-SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, css_growth_div,
-    CTLFLAG_RW,
-    &hystart_css_growth_div, 4,
-   "The divisor to the growth when in Hystart++ CSS");
-
-SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, css_rounds,
-    CTLFLAG_RW,
-    &hystart_css_rounds, 5,
-   "The number of rounds HyStart++ lasts in CSS before falling to CA");
-
-SYSCTL_UINT(_net_inet_tcp_cc_newreno_hystartplusplus, OID_AUTO, bblogs,
-    CTLFLAG_RW,
-    &hystart_bblogs, 0,
-   "Do we enable HyStart++ Black Box logs to be generated if BB logging is on");
-
-
 DECLARE_CC_MODULE(newreno, &newreno_cc_algo);
 MODULE_VERSION(newreno, 2);
diff --git a/sys/netinet/cc/cc_newreno.h b/sys/netinet/cc/cc_newreno.h
index 42c41c7db287..791b64dfdc00 100644
--- a/sys/netinet/cc/cc_newreno.h
+++ b/sys/netinet/cc/cc_newreno.h
@@ -53,13 +53,9 @@ struct cc_newreno_opts {
 
 #define CC_NEWRENO_BETA			1	/* Beta for normal DUP-ACK/Sack recovery */
 #define CC_NEWRENO_BETA_ECN		2	/* ECN Beta for Abe */
-#define CC_NEWRENO_ENABLE_HYSTART	3	/* Enable hystart */
 
 /* Flags values */
-#define CC_NEWRENO_HYSTART_ALLOWED	0x0001	/* Does the tcp connection allow hystart? */
 #define CC_NEWRENO_HYSTART_ENABLED	0x0002	/* We can do hystart, a loss removes this flag */
 #define CC_NEWRENO_HYSTART_IN_CSS	0x0004	/* If we enter hystart CSS this flag is set */
-#define CC_NEWRENO_HYSTART_CAN_SH_CWND	0x0008	/* Can hystart when going CSS -> CA slam the cwnd */
-#define CC_NEWRENO_HYSTART_CONS_SSTH	0x0010	/* Should hystart use the more conservative sstrhesh */
 #define CC_NEWRENO_BETA_ECN_ENABLED	0x0020
 #endif /* _CC_NEWRENO_H */
diff --git a/sys/netinet/tcp_stacks/rack.c b/sys/netinet/tcp_stacks/rack.c
index 4ad6a12ee43b..459520a4a0bb 100644
--- a/sys/netinet/tcp_stacks/rack.c
+++ b/sys/netinet/tcp_stacks/rack.c
@@ -12858,15 +12858,11 @@ rack_init(struct tcpcb *tp)
 	rack_convert_rtts(tp);
 	tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow);
 	if (rack_do_hystart) {
-		struct sockopt sopt;
-		struct cc_newreno_opts opt;
-
-		sopt.sopt_valsize = sizeof(struct cc_newreno_opts);
-		sopt.sopt_dir = SOPT_SET;
-		opt.name = CC_NEWRENO_ENABLE_HYSTART;
-		opt.val = rack_do_hystart;
-		if (CC_ALGO(tp)->ctl_output != NULL)
-			(void)CC_ALGO(tp)->ctl_output(tp->ccv, &sopt, &opt);
+		tp->ccv->flags |= CCF_HYSTART_ALLOWED;
+		if (rack_do_hystart > 1) 
+			tp->ccv->flags |= CCF_HYSTART_CAN_SH_CWND;
+		if (rack_do_hystart > 2) 
+			tp->ccv->flags |= CCF_HYSTART_CONS_SSTH;
 	}
 	if (rack_def_profile)
 		rack_set_profile(rack, rack_def_profile);
@@ -13515,7 +13511,6 @@ rack_handle_probe_response(struct tcp_rack *rack, uint32_t tiwin, uint32_t us_ct
 	}
 }
 
-
 static int
 rack_do_compressed_ack_processing(struct tcpcb *tp, struct socket *so, struct mbuf *m, int nxt_pkt, struct timeval *tv)
 {
@@ -13766,7 +13761,29 @@ rack_do_compressed_ack_processing(struct tcpcb *tp, struct socket *so, struct mb
 						(((ae->ack - high_seq) + segsiz - 1) / segsiz));
 #endif
 				high_seq = ae->ack;
-				if (SEQ_GEQ(high_seq, rack->r_ctl.roundends)) {
+				if (rack_verbose_logging && (rack->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) {
+					union tcp_log_stackspecific log;
+					struct timeval tv;
+
+					memset(&log.u_bbr, 0, sizeof(log.u_bbr));
+					log.u_bbr.timeStamp = tcp_get_usecs(&tv);
+					log.u_bbr.flex1 = high_seq;
+					log.u_bbr.flex2 = rack->r_ctl.roundends;
+					log.u_bbr.flex3 = rack->r_ctl.current_round;
+					log.u_bbr.rttProp = (uint64_t)CC_ALGO(tp)->newround;
+					log.u_bbr.flex8 = 8;
+					tcp_log_event_(tp, NULL, NULL, NULL, BBR_LOG_CWND, 0,
+						       0, &log, false, NULL, NULL, 0, &tv);
+				}
+				/* 
+				 * The draft (v3) calls for us to use SEQ_GEQ, but that
+				 * causes issues when we are just going app limited. Lets
+				 * instead use SEQ_GT <or> where its equal but more data
+				 * is outstanding.
+				 */
+				if ((SEQ_GT(high_seq, rack->r_ctl.roundends)) ||
+				    ((high_seq == rack->r_ctl.roundends) &&
+				     SEQ_GT(tp->snd_max, tp->snd_una))) {
 					rack->r_ctl.current_round++;
 					rack->r_ctl.roundends = tp->snd_max;
 					if (CC_ALGO(tp)->newround != NULL) {
@@ -14194,7 +14211,7 @@ rack_do_segment_nounlock(struct mbuf *m, struct tcphdr *th, struct socket *so,
 	 * us_cts - is the time that LRO or hardware actually got the packet in microseconds.
 	 */
 	uint32_t cts, us_cts, ms_cts;
-	uint32_t tiwin;
+	uint32_t tiwin, high_seq;
 	struct timespec ts;
 	struct tcpopt to;
 	struct tcp_rack *rack;
@@ -14301,6 +14318,7 @@ rack_do_segment_nounlock(struct mbuf *m, struct tcphdr *th, struct socket *so,
 			tp->t_flags &= ~TF_GPUTINPROG;
 		}
 	}
+	high_seq = th->th_ack;
 	if (tp->t_logstate != TCP_LOG_STATE_OFF) {
 		union tcp_log_stackspecific log;
 		struct timeval ltv;
@@ -14627,7 +14645,28 @@ do_output_now:
 			rack_free_trim(rack);
 		}
 		/* Update any rounds needed */
-		if (SEQ_GEQ(tp->snd_una, rack->r_ctl.roundends)) {
+		if (rack_verbose_logging &&  (rack->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) {
+			union tcp_log_stackspecific log;
+			struct timeval tv;
+
+			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
+			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
+			log.u_bbr.flex1 = high_seq;
+			log.u_bbr.flex2 = rack->r_ctl.roundends;
+			log.u_bbr.flex3 = rack->r_ctl.current_round;
+			log.u_bbr.rttProp = (uint64_t)CC_ALGO(tp)->newround;
+			log.u_bbr.flex8 = 9;
+			tcp_log_event_(tp, NULL, NULL, NULL, BBR_LOG_CWND, 0,
+				       0, &log, false, NULL, NULL, 0, &tv);
+		}
+		/* 
+		 * The draft (v3) calls for us to use SEQ_GEQ, but that
+		 * causes issues when we are just going app limited. Lets
+		 * instead use SEQ_GT <or> where its equal but more data
+		 * is outstanding.
+		 */
+		if ((SEQ_GT(tp->snd_una, rack->r_ctl.roundends)) ||
+		    ((tp->snd_una == rack->r_ctl.roundends) && SEQ_GT(tp->snd_max, tp->snd_una))) {
 			rack->r_ctl.current_round++;
 			rack->r_ctl.roundends = tp->snd_max;
 			if (CC_ALGO(tp)->newround != NULL) {
@@ -20204,17 +20243,15 @@ rack_process_option(struct tcpcb *tp, struct tcp_rack *rack, int sopt_name,
 		break;
 	case TCP_RACK_ENABLE_HYSTART:
 	{
-		struct sockopt sopt;
-		struct cc_newreno_opts opt;
-
-		sopt.sopt_valsize = sizeof(struct cc_newreno_opts);
-		sopt.sopt_dir = SOPT_SET;
-		opt.name = CC_NEWRENO_ENABLE_HYSTART;
-		opt.val = optval;
-		if (CC_ALGO(tp)->ctl_output != NULL)
-			error = CC_ALGO(tp)->ctl_output(tp->ccv, &sopt, &opt);
-		else
-			error = EINVAL;
+		if (optval) {
+			tp->ccv->flags |= CCF_HYSTART_ALLOWED;
+			if (rack_do_hystart > RACK_HYSTART_ON) 
+				tp->ccv->flags |= CCF_HYSTART_CAN_SH_CWND;
+			if (rack_do_hystart > RACK_HYSTART_ON_W_SC) 
+				tp->ccv->flags |= CCF_HYSTART_CONS_SSTH;
+		} else {
+			tp->ccv->flags &= ~(CCF_HYSTART_ALLOWED|CCF_HYSTART_CAN_SH_CWND|CCF_HYSTART_CONS_SSTH);
+		}
 	}
 	break;
 	case TCP_RACK_REORD_THRESH:
@@ -20684,17 +20721,15 @@ rack_get_sockopt(struct inpcb *inp, struct sockopt *sopt)
 		break;
  	case TCP_RACK_ENABLE_HYSTART:
 	{
-		struct sockopt sopt;
-		struct cc_newreno_opts opt;
-
-		sopt.sopt_valsize = sizeof(struct cc_newreno_opts);
-		sopt.sopt_dir = SOPT_GET;
-		opt.name = CC_NEWRENO_ENABLE_HYSTART;
-		if (CC_ALGO(tp)->ctl_output != NULL)
-			error = CC_ALGO(tp)->ctl_output(tp->ccv, &sopt, &opt);
-		else
-			error = EINVAL;
-		optval = opt.val;
+		if (tp->ccv->flags & CCF_HYSTART_ALLOWED) {
+			optval = RACK_HYSTART_ON;
+			if (tp->ccv->flags & CCF_HYSTART_CAN_SH_CWND)
+				optval = RACK_HYSTART_ON_W_SC;
+			if (tp->ccv->flags & CCF_HYSTART_CONS_SSTH)
+				optval = RACK_HYSTART_ON_W_SC_C;
+		} else {
+			optval = RACK_HYSTART_OFF;
+		}
 	}
 	break;
 	case TCP_FAST_RSM_HACK:
diff --git a/sys/netinet/tcp_stacks/tcp_rack.h b/sys/netinet/tcp_stacks/tcp_rack.h
index 4b1f8513055d..d9a7f98a4075 100644
--- a/sys/netinet/tcp_stacks/tcp_rack.h
+++ b/sys/netinet/tcp_stacks/tcp_rack.h
@@ -525,6 +525,14 @@ struct rack_control {
 #define RACK_TIMELY_CNT_BOOST 5	/* At 5th increase boost */
 #define RACK_MINRTT_FILTER_TIM 10 /* Seconds */
 
+#define RACK_HYSTART_OFF	0
+#define RACK_HYSTART_ON		1	/* hystart++ on */
+#define RACK_HYSTART_ON_W_SC	2	/* hystart++ on +Slam Cwnd */
+#define RACK_HYSTART_ON_W_SC_C	3	/* hystart++ on,
+					 * Conservative ssthresh and 
+					 * +Slam cwnd 
+					 */
+
 #ifdef _KERNEL
 
 struct tcp_rack {