tcp: calculate ssthresh on RTO according to RFC5681

per RFC5681, only adjust ssthresh on the initital
retransmission timeout. Since RTO often happens
during loss recovery, while cwnd no longer tracks
all data in flight, calculcate pipe properly.

Reviewed By:           tuexen, #transport
Sponsored by:          NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D43768

sys/netinet/cc/cc.c

@@ -454,8 +454,7 @@ newreno_cc_after_idle(struct cc_var *ccv)
 void
 newreno_cc_cong_signal(struct cc_var *ccv, uint32_t type)
 {
-	uint32_t cwin, factor;
-	u_int mss;
+	uint32_t cwin, factor, mss, pipe;
 
 	cwin = CCV(ccv, snd_cwnd);
 	mss = tcp_fixed_maxseg(ccv->ccvc.tcp);
@@ -489,9 +488,17 @@ newreno_cc_cong_signal(struct cc_var *ccv, uint32_t type)
 		}
 		break;
 	case CC_RTO:
-		CCV(ccv, snd_ssthresh) = max(min(CCV(ccv, snd_wnd),
-						 CCV(ccv, snd_cwnd)) / 2 / mss,
-					     2) * mss;
+		if (CCV(ccv, t_rxtshift) == 1) {
+			if (V_tcp_do_newsack) {
+				pipe = tcp_compute_pipe(ccv->ccvc.tcp);
+			} else {
+				pipe = CCV(ccv, snd_nxt) -
+					CCV(ccv, snd_fack) +
+					CCV(ccv, sackhint.sack_bytes_rexmit);
+			}
+			CCV(ccv, snd_ssthresh) = max(2,
+				min(CCV(ccv, snd_wnd), pipe) / 2 / mss) * mss;
+		}
 		CCV(ccv, snd_cwnd) = mss;
 		break;
 	}

sys/netinet/cc/cc_cubic.c

@@ -420,7 +420,7 @@ static void
 cubic_cong_signal(struct cc_var *ccv, uint32_t type)
 {
 	struct cubic *cubic_data;
-	u_int mss;
+	uint32_t mss, pipe;
 
 	cubic_data = ccv->cc_data;
 	mss = tcp_fixed_maxseg(ccv->ccvc.tcp);
@@ -476,12 +476,20 @@ cubic_cong_signal(struct cc_var *ccv, uint32_t type)
 			cubic_data->undo_cwnd_prior = cubic_data->cwnd_prior;
 			cubic_data->undo_W_max = cubic_data->W_max;
 			cubic_data->undo_K = cubic_data->K;
+			if (V_tcp_do_newsack) {
+				pipe = tcp_compute_pipe(ccv->ccvc.tcp);
+			} else {
+				pipe = CCV(ccv, snd_nxt) -
+					CCV(ccv, snd_fack) +
+					CCV(ccv, sackhint.sack_bytes_rexmit);
+			}
+			CCV(ccv, snd_ssthresh) = max(2,
+				(((uint64_t)min(CCV(ccv, snd_wnd), pipe) *
+				CUBIC_BETA) >> CUBIC_SHIFT) / mss) * mss;
 		}
 		cubic_data->flags |= CUBICFLAG_CONG_EVENT | CUBICFLAG_RTO_EVENT;
 		cubic_data->undo_W_max = cubic_data->W_max;
 		cubic_data->num_cong_events++;
-			CCV(ccv, snd_ssthresh) = ((uint64_t)CCV(ccv, snd_cwnd) *
-					  CUBIC_BETA) >> CUBIC_SHIFT;
 		CCV(ccv, snd_cwnd) = mss;
 		break;
 

sys/netinet/cc/cc_dctcp.c

@@ -240,7 +240,7 @@ static void
 dctcp_cong_signal(struct cc_var *ccv, uint32_t type)
 {
 	struct dctcp *dctcp_data;
-	u_int cwin, mss;
+	uint32_t cwin, mss, pipe;
 
 	if (CCV(ccv, t_flags2) & TF2_ECN_PERMIT) {
 		dctcp_data = ccv->cc_data;
@@ -292,9 +292,17 @@ dctcp_cong_signal(struct cc_var *ccv, uint32_t type)
 			dctcp_data->ece_curr = 1;
 			break;
 		case CC_RTO:
-			CCV(ccv, snd_ssthresh) = max(min(CCV(ccv, snd_wnd),
-							 CCV(ccv, snd_cwnd)) / 2 / mss,
-						     2) * mss;
+			if (CCV(ccv, t_rxtshift) == 1) {
+				if (V_tcp_do_newsack) {
+					pipe = tcp_compute_pipe(ccv->ccvc.tcp);
+				} else {
+					pipe = CCV(ccv, snd_nxt) -
+						CCV(ccv, snd_fack) +
+						CCV(ccv, sackhint.sack_bytes_rexmit);
+				}
+				CCV(ccv, snd_ssthresh) = max(2,
+					min(CCV(ccv, snd_wnd), pipe) / 2 / mss) * mss;
+			}
 			CCV(ccv, snd_cwnd) = mss;
 			dctcp_update_alpha(ccv);
 			dctcp_data->save_sndnxt += CCV(ccv, t_maxseg);

sys/netinet/cc/cc_htcp.c

@@ -281,7 +281,7 @@ static void
 htcp_cong_signal(struct cc_var *ccv, uint32_t type)
 {
 	struct htcp *htcp_data;
-	u_int mss;
+	uint32_t mss, pipe;
 
 	htcp_data = ccv->cc_data;
 	mss = tcp_fixed_maxseg(ccv->ccvc.tcp);
@@ -323,9 +323,17 @@ htcp_cong_signal(struct cc_var *ccv, uint32_t type)
 		break;
 
 	case CC_RTO:
-		CCV(ccv, snd_ssthresh) = max(min(CCV(ccv, snd_wnd),
-						 CCV(ccv, snd_cwnd)) / 2 / mss,
-					     2) * mss;
+		if (CCV(ccv, t_rxtshift) == 1) {
+			if (V_tcp_do_newsack) {
+				pipe = tcp_compute_pipe(ccv->ccvc.tcp);
+			} else {
+				pipe = CCV(ccv, snd_nxt) -
+					CCV(ccv, snd_fack) +
+					CCV(ccv, sackhint.sack_bytes_rexmit);
+			}
+			CCV(ccv, snd_ssthresh) = max(2,
+				min(CCV(ccv, snd_wnd), pipe) / 2 / mss) * mss;
+		}
 		CCV(ccv, snd_cwnd) = mss;
 		/*
 		 * Grab the current time and record it so we know when the

sys/netinet/cc/cc_newreno.c

@@ -366,8 +366,7 @@ static void
 newreno_cong_signal(struct cc_var *ccv, uint32_t type)
 {
 	struct newreno *nreno;
-	uint32_t beta, beta_ecn, cwin, factor;
-	u_int mss;
+	uint32_t beta, beta_ecn, cwin, factor, mss, pipe;
 
 	cwin = CCV(ccv, snd_cwnd);
 	mss = tcp_fixed_maxseg(ccv->ccvc.tcp);
@@ -428,9 +427,19 @@ newreno_cong_signal(struct cc_var *ccv, uint32_t type)
 		}
 		break;
 	case CC_RTO:
-		CCV(ccv, snd_ssthresh) = max(min(CCV(ccv, snd_wnd),
-						 CCV(ccv, snd_cwnd)) / 2 / mss,
-					     2) * mss;
+		if (CCV(ccv, t_rxtshift) == 1) {
+			if (V_tcp_do_newsack) {
+				pipe = tcp_compute_pipe(ccv->ccvc.tcp);
+			} else {
+				pipe = CCV(ccv, snd_nxt) -
+					CCV(ccv, snd_fack) +
+					CCV(ccv, sackhint.sack_bytes_rexmit);
+			}
+			CCV(ccv, snd_ssthresh) = max(2,
+				((uint64_t)min(CCV(ccv, snd_wnd), pipe) *
+				    (uint64_t)factor) /
+				    (100ULL * (uint64_t)mss)) * mss;
+		}
 		CCV(ccv, snd_cwnd) = mss;
 		break;
 	}

sys/netinet/tcp_stacks/rack.c

@@ -5948,8 +5948,11 @@ rack_cong_signal(struct tcpcb *tp, uint32_t type, uint32_t ack, int line)
 		tp->t_bytes_acked = 0;
 		rack->r_fast_output = 0;
 		EXIT_RECOVERY(tp->t_flags);
-		tp->snd_ssthresh = max(2, min(tp->snd_wnd, rack->r_ctl.cwnd_to_use) / 2 /
-		    ctf_fixed_maxseg(tp)) * ctf_fixed_maxseg(tp);
+		if (tp->t_rxtshift == 1) {
+			tp->snd_ssthresh = max(2,
+			    min(tp->snd_wnd, rack->r_ctl.cwnd_to_use) / 2 /
+			    ctf_fixed_maxseg(tp)) * ctf_fixed_maxseg(tp);
+		}
 		orig_cwnd = tp->snd_cwnd;
 		tp->snd_cwnd = ctf_fixed_maxseg(tp);
 		rack_log_to_prr(rack, 16, orig_cwnd, line);