Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 5 Mar 2018 06:52:26 +0000 (UTC)
From:      Eitan Adler <eadler@FreeBSD.org>
To:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-11@freebsd.org
Subject:   svn commit: r330445 - in stable/11/sys/netinet: . cc tcp_stacks
Message-ID:  <201803050652.w256qQvu071205@repo.freebsd.org>

next in thread | raw e-mail | index | archive | help
Author: eadler
Date: Mon Mar  5 06:52:26 2018
New Revision: 330445
URL: https://svnweb.freebsd.org/changeset/base/330445

Log:
  MFC r307901,r308180:
  
  FreeBSD tcp stack used to inform respective congestion control module about the
  loss event but not use or obay the recommendations i.e. values set by it in some
  cases.
  
  Here is an attempt to solve that confusion by following relevant RFCs/drafts.
  Stack only sets congestion window/slow start threshold values when there is no
  CC module availalbe to take that action. All CC modules are inspected and
  updated when needed to take appropriate action on loss.
  
  tcp_stacks/fastpath module has been updated to adapt these changes.
  
  Note: Probably, the most significant change would be to not bring congestion
  window down to 1MSS on a loss signaled by 3-duplicate acks and letting
  respective CC decide that value.

Modified:
  stable/11/sys/netinet/cc/cc_cdg.c
  stable/11/sys/netinet/cc/cc_chd.c
  stable/11/sys/netinet/cc/cc_cubic.c
  stable/11/sys/netinet/cc/cc_dctcp.c
  stable/11/sys/netinet/cc/cc_htcp.c
  stable/11/sys/netinet/cc/cc_newreno.c
  stable/11/sys/netinet/tcp_input.c
  stable/11/sys/netinet/tcp_stacks/fastpath.c
Directory Properties:
  stable/11/   (props changed)

Modified: stable/11/sys/netinet/cc/cc_cdg.c
==============================================================================
--- stable/11/sys/netinet/cc/cc_cdg.c	Mon Mar  5 06:47:28 2018	(r330444)
+++ stable/11/sys/netinet/cc/cc_cdg.c	Mon Mar  5 06:52:26 2018	(r330445)
@@ -431,7 +431,12 @@ static void
 cdg_cong_signal(struct cc_var *ccv, uint32_t signal_type)
 {
 	struct cdg *cdg_data = ccv->cc_data;
+	uint32_t cwin;
+	u_int mss;
 
+	cwin = CCV(ccv, snd_cwnd);
+	mss = CCV(ccv, t_maxseg);
+
 	switch(signal_type) {
 	case CC_CDG_DELAY:
 		CCV(ccv, snd_ssthresh) = cdg_window_decrease(ccv,
@@ -448,7 +453,7 @@ cdg_cong_signal(struct cc_var *ccv, uint32_t signal_ty
 		 */
 		if (IN_CONGRECOVERY(CCV(ccv, t_flags)) ||
 		    cdg_data->queue_state < CDG_Q_FULL) {
-			CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd);
+			CCV(ccv, snd_ssthresh) = cwin;
 			CCV(ccv, snd_recover) = CCV(ccv, snd_max);
 		} else {
 			/*
@@ -461,12 +466,18 @@ cdg_cong_signal(struct cc_var *ccv, uint32_t signal_ty
 				    cdg_data->shadow_w, RENO_BETA);
 
 			CCV(ccv, snd_ssthresh) = ulmax(cdg_data->shadow_w,
-			    cdg_window_decrease(ccv, CCV(ccv, snd_cwnd),
-			    V_cdg_beta_loss));
+			    cdg_window_decrease(ccv, cwin, V_cdg_beta_loss));
+			CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
 
 			cdg_data->window_incr = cdg_data->rtt_count = 0;
 		}
 		ENTER_RECOVERY(CCV(ccv, t_flags));
+		break;
+	case CC_RTO:
+		CCV(ccv, snd_ssthresh) =
+		    max((CCV(ccv, snd_max) - CCV(ccv, snd_una)) / 2 / mss, 2)
+			* mss;
+		CCV(ccv, snd_cwnd) = mss;
 		break;
 	default:
 		newreno_cc_algo.cong_signal(ccv, signal_type);

Modified: stable/11/sys/netinet/cc/cc_chd.c
==============================================================================
--- stable/11/sys/netinet/cc/cc_chd.c	Mon Mar  5 06:47:28 2018	(r330444)
+++ stable/11/sys/netinet/cc/cc_chd.c	Mon Mar  5 06:52:26 2018	(r330445)
@@ -330,10 +330,12 @@ chd_cong_signal(struct cc_var *ccv, uint32_t signal_ty
 	struct ertt *e_t;
 	struct chd *chd_data;
 	int qdly;
+	u_int mss;
 
 	e_t = khelp_get_osd(CCV(ccv, osd), ertt_id);
 	chd_data = ccv->cc_data;
 	qdly = imax(e_t->rtt, chd_data->maxrtt_in_rtt) - e_t->minrtt;
+	mss = CCV(ccv, t_maxseg);
 
 	switch(signal_type) {
 	case CC_CHD_DELAY:
@@ -372,6 +374,12 @@ chd_cong_signal(struct cc_var *ccv, uint32_t signal_ty
 			    CCV(ccv, t_maxseg) / 2, 2) * CCV(ccv, t_maxseg);
 		}
 		ENTER_FASTRECOVERY(CCV(ccv, t_flags));
+		break;
+	case CC_RTO:
+		CCV(ccv, snd_ssthresh) =
+		    max((CCV(ccv, snd_max) - CCV(ccv, snd_una)) / 2 / mss, 2)
+			* mss;
+		CCV(ccv, snd_cwnd) = mss;
 		break;
 
 	default:

Modified: stable/11/sys/netinet/cc/cc_cubic.c
==============================================================================
--- stable/11/sys/netinet/cc/cc_cubic.c	Mon Mar  5 06:47:28 2018	(r330444)
+++ stable/11/sys/netinet/cc/cc_cubic.c	Mon Mar  5 06:52:26 2018	(r330445)
@@ -225,8 +225,12 @@ static void
 cubic_cong_signal(struct cc_var *ccv, uint32_t type)
 {
 	struct cubic *cubic_data;
+	uint32_t cwin;
+	u_int mss;
 
 	cubic_data = ccv->cc_data;
+	cwin = CCV(ccv, snd_cwnd);
+	mss = CCV(ccv, t_maxseg);
 
 	switch (type) {
 	case CC_NDUPACK:
@@ -235,7 +239,8 @@ cubic_cong_signal(struct cc_var *ccv, uint32_t type)
 				cubic_ssthresh_update(ccv);
 				cubic_data->num_cong_events++;
 				cubic_data->prev_max_cwnd = cubic_data->max_cwnd;
-				cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
+				cubic_data->max_cwnd = cwin;
+				CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
 			}
 			ENTER_RECOVERY(CCV(ccv, t_flags));
 		}
@@ -246,7 +251,7 @@ cubic_cong_signal(struct cc_var *ccv, uint32_t type)
 			cubic_ssthresh_update(ccv);
 			cubic_data->num_cong_events++;
 			cubic_data->prev_max_cwnd = cubic_data->max_cwnd;
-			cubic_data->max_cwnd = CCV(ccv, snd_cwnd);
+			cubic_data->max_cwnd = cwin;
 			cubic_data->t_last_cong = ticks;
 			CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
 			ENTER_CONGRECOVERY(CCV(ccv, t_flags));
@@ -261,9 +266,13 @@ cubic_cong_signal(struct cc_var *ccv, uint32_t type)
 		 * chance the first one is a false alarm and may not indicate
 		 * congestion.
 		 */
-		if (CCV(ccv, t_rxtshift) >= 2)
+		if (CCV(ccv, t_rxtshift) >= 2) {
 			cubic_data->num_cong_events++;
 			cubic_data->t_last_cong = ticks;
+			cubic_ssthresh_update(ccv);
+			cubic_data->max_cwnd = cwin;
+			CCV(ccv, snd_cwnd) = mss;
+		}
 		break;
 	}
 }

Modified: stable/11/sys/netinet/cc/cc_dctcp.c
==============================================================================
--- stable/11/sys/netinet/cc/cc_dctcp.c	Mon Mar  5 06:47:28 2018	(r330444)
+++ stable/11/sys/netinet/cc/cc_dctcp.c	Mon Mar  5 06:52:26 2018	(r330445)
@@ -230,27 +230,29 @@ static void
 dctcp_cong_signal(struct cc_var *ccv, uint32_t type)
 {
 	struct dctcp *dctcp_data;
-	u_int win, mss;
+	uint32_t cwin, ssthresh_on_loss;
+	u_int mss;
 
 	dctcp_data = ccv->cc_data;
-	win = CCV(ccv, snd_cwnd);
+	cwin = CCV(ccv, snd_cwnd);
 	mss = CCV(ccv, t_maxseg);
+	ssthresh_on_loss =
+	    max((CCV(ccv, snd_max) - CCV(ccv, snd_una)) / 2 / mss, 2)
+		* mss;
 
 	switch (type) {
 	case CC_NDUPACK:
 		if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) {
 			if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
-				CCV(ccv, snd_ssthresh) = mss *
-				    max(win / 2 / mss, 2);
+				CCV(ccv, snd_ssthresh) = ssthresh_on_loss;
 				dctcp_data->num_cong_events++;
 			} else {
 				/* cwnd has already updated as congestion
 				 * recovery. Reverse cwnd value using
 				 * snd_cwnd_prev and recalculate snd_ssthresh
 				 */
-				win = CCV(ccv, snd_cwnd_prev);
-				CCV(ccv, snd_ssthresh) =
-				    max(win / 2 / mss, 2) * mss;
+				cwin = CCV(ccv, snd_cwnd_prev);
+				CCV(ccv, snd_ssthresh) = ssthresh_on_loss;
 			}
 			ENTER_RECOVERY(CCV(ccv, t_flags));
 		}
@@ -260,18 +262,17 @@ dctcp_cong_signal(struct cc_var *ccv, uint32_t type)
 		 * Save current snd_cwnd when the host encounters both
 		 * congestion recovery and fast recovery.
 		 */
-		CCV(ccv, snd_cwnd_prev) = win;
+		CCV(ccv, snd_cwnd_prev) = cwin;
 		if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
 			if (V_dctcp_slowstart &&
 			    dctcp_data->num_cong_events++ == 0) {
-				CCV(ccv, snd_ssthresh) =
-				    mss * max(win / 2 / mss, 2);
+				CCV(ccv, snd_ssthresh) = ssthresh_on_loss;
 				dctcp_data->alpha = MAX_ALPHA_VALUE;
 				dctcp_data->bytes_ecn = 0;
 				dctcp_data->bytes_total = 0;
 				dctcp_data->save_sndnxt = CCV(ccv, snd_nxt);
 			} else
-				CCV(ccv, snd_ssthresh) = max((win - ((win *
+				CCV(ccv, snd_ssthresh) = max((cwin - ((cwin *
 				    dctcp_data->alpha) >> 11)) / mss, 2) * mss;
 			CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
 			ENTER_CONGRECOVERY(CCV(ccv, t_flags));
@@ -284,6 +285,8 @@ dctcp_cong_signal(struct cc_var *ccv, uint32_t type)
 			dctcp_update_alpha(ccv);
 			dctcp_data->save_sndnxt += CCV(ccv, t_maxseg);
 			dctcp_data->num_cong_events++;
+			CCV(ccv, snd_ssthresh) = ssthresh_on_loss;
+			CCV(ccv, snd_cwnd) = mss;
 		}
 		break;
 	}

Modified: stable/11/sys/netinet/cc/cc_htcp.c
==============================================================================
--- stable/11/sys/netinet/cc/cc_htcp.c	Mon Mar  5 06:47:28 2018	(r330444)
+++ stable/11/sys/netinet/cc/cc_htcp.c	Mon Mar  5 06:52:26 2018	(r330445)
@@ -271,8 +271,12 @@ static void
 htcp_cong_signal(struct cc_var *ccv, uint32_t type)
 {
 	struct htcp *htcp_data;
+	uint32_t cwin;
+	u_int mss;
 
 	htcp_data = ccv->cc_data;
+	cwin = CCV(ccv, snd_cwnd);
+	mss = CCV(ccv, t_maxseg);
 
 	switch (type) {
 	case CC_NDUPACK:
@@ -287,8 +291,9 @@ htcp_cong_signal(struct cc_var *ccv, uint32_t type)
 				    (htcp_data->maxrtt - htcp_data->minrtt) *
 				    95) / 100;
 				htcp_ssthresh_update(ccv);
+				CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
 				htcp_data->t_last_cong = ticks;
-				htcp_data->prev_cwnd = CCV(ccv, snd_cwnd);
+				htcp_data->prev_cwnd = cwin;
 			}
 			ENTER_RECOVERY(CCV(ccv, t_flags));
 		}
@@ -305,7 +310,7 @@ htcp_cong_signal(struct cc_var *ccv, uint32_t type)
 			htcp_ssthresh_update(ccv);
 			CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
 			htcp_data->t_last_cong = ticks;
-			htcp_data->prev_cwnd = CCV(ccv, snd_cwnd);
+			htcp_data->prev_cwnd = cwin;
 			ENTER_CONGRECOVERY(CCV(ccv, t_flags));
 		}
 		break;
@@ -320,6 +325,10 @@ htcp_cong_signal(struct cc_var *ccv, uint32_t type)
 		 */
 		if (CCV(ccv, t_rxtshift) >= 2)
 			htcp_data->t_last_cong = ticks;
+		CCV(ccv, snd_ssthresh) =
+		    max((CCV(ccv, snd_max) - CCV(ccv, snd_una)) / 2 / mss, 2)
+			* mss;
+		CCV(ccv, snd_cwnd) = mss;
 		break;
 	}
 }
@@ -511,6 +520,10 @@ htcp_ssthresh_update(struct cc_var *ccv)
 		CCV(ccv, snd_ssthresh) = (CCV(ccv, snd_cwnd) * htcp_data->beta)
 		    >> HTCP_SHIFT;
 	}
+
+	/* Align ssthresh to MSS boundary */
+	CCV(ccv, snd_ssthresh) = (CCV(ccv, snd_ssthresh) / CCV(ccv, t_maxseg))
+	    * CCV(ccv, t_maxseg);
 }
 
 

Modified: stable/11/sys/netinet/cc/cc_newreno.c
==============================================================================
--- stable/11/sys/netinet/cc/cc_newreno.c	Mon Mar  5 06:47:28 2018	(r330444)
+++ stable/11/sys/netinet/cc/cc_newreno.c	Mon Mar  5 06:52:26 2018	(r330445)
@@ -181,29 +181,41 @@ newreno_after_idle(struct cc_var *ccv)
 static void
 newreno_cong_signal(struct cc_var *ccv, uint32_t type)
 {
-	u_int win;
+	uint32_t cwin, ssthresh_on_loss;
+	u_int mss;
 
+	cwin = CCV(ccv, snd_cwnd);
+	mss = CCV(ccv, t_maxseg);
+	ssthresh_on_loss =
+	    max((CCV(ccv, snd_max) - CCV(ccv, snd_una)) / 2 / mss, 2)
+		* mss;
+
 	/* Catch algos which mistakenly leak private signal types. */
 	KASSERT((type & CC_SIGPRIVMASK) == 0,
 	    ("%s: congestion signal type 0x%08x is private\n", __func__, type));
 
-	win = max(CCV(ccv, snd_cwnd) / 2 / CCV(ccv, t_maxseg), 2) *
-	    CCV(ccv, t_maxseg);
+	cwin = max(cwin / 2 / mss, 2) * mss;
 
 	switch (type) {
 	case CC_NDUPACK:
 		if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) {
-			if (!IN_CONGRECOVERY(CCV(ccv, t_flags)))
-				CCV(ccv, snd_ssthresh) = win;
+			if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
+				CCV(ccv, snd_ssthresh) = ssthresh_on_loss;
+				CCV(ccv, snd_cwnd) = cwin;
+			}
 			ENTER_RECOVERY(CCV(ccv, t_flags));
 		}
 		break;
 	case CC_ECN:
 		if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
-			CCV(ccv, snd_ssthresh) = win;
-			CCV(ccv, snd_cwnd) = win;
+			CCV(ccv, snd_ssthresh) = ssthresh_on_loss;
+			CCV(ccv, snd_cwnd) = cwin;
 			ENTER_CONGRECOVERY(CCV(ccv, t_flags));
 		}
+		break;
+	case CC_RTO:
+		CCV(ccv, snd_ssthresh) = ssthresh_on_loss;
+		CCV(ccv, snd_cwnd) = mss;
 		break;
 	}
 }

Modified: stable/11/sys/netinet/tcp_input.c
==============================================================================
--- stable/11/sys/netinet/tcp_input.c	Mon Mar  5 06:47:28 2018	(r330444)
+++ stable/11/sys/netinet/tcp_input.c	Mon Mar  5 06:52:26 2018	(r330445)
@@ -429,9 +429,16 @@ cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, ui
 		tp->t_dupacks = 0;
 		tp->t_bytes_acked = 0;
 		EXIT_RECOVERY(tp->t_flags);
-		tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
-		    maxseg) * maxseg;
-		tp->snd_cwnd = maxseg;
+		if (CC_ALGO(tp)->cong_signal == NULL) {
+			/*
+			 * RFC5681 Section 3.1 
+			 * ssthresh = max (FlightSize / 2, 2*SMSS) eq (4)
+			 */
+			tp->snd_ssthresh =
+			    max((tp->snd_max - tp->snd_una) / 2 / maxseg, 2)
+				* maxseg;
+			tp->snd_cwnd = maxseg;
+		}
 		break;
 	case CC_RTO_ERR:
 		TCPSTAT_INC(tcps_sndrexmitbad);
@@ -2592,6 +2599,15 @@ tcp_do_segment(struct mbuf *m, struct tcphdr *th, stru
 
 						if (awnd < tp->snd_ssthresh) {
 							tp->snd_cwnd += maxseg;
+							/*
+							 * RFC5681 Section 3.2 talks about cwnd
+							 * inflation on additional dupacks and
+							 * deflation on recovering from loss.
+							 *
+							 * We keep cwnd into check so that
+							 * we don't have to 'deflate' it when we
+							 * get out of recovery.
+							 */
 							if (tp->snd_cwnd > tp->snd_ssthresh)
 								tp->snd_cwnd = tp->snd_ssthresh;
 						}
@@ -2630,19 +2646,22 @@ tcp_do_segment(struct mbuf *m, struct tcphdr *th, stru
 						TCPSTAT_INC(
 						    tcps_sack_recovery_episode);
 						tp->sack_newdata = tp->snd_nxt;
-						tp->snd_cwnd = maxseg;
+						if (CC_ALGO(tp)->cong_signal == NULL)
+							tp->snd_cwnd = maxseg;
 						(void) tp->t_fb->tfb_tcp_output(tp);
 						goto drop;
 					}
 					tp->snd_nxt = th->th_ack;
-					tp->snd_cwnd = maxseg;
+					if (CC_ALGO(tp)->cong_signal == NULL)
+						tp->snd_cwnd = maxseg;
 					(void) tp->t_fb->tfb_tcp_output(tp);
 					KASSERT(tp->snd_limited <= 2,
 					    ("%s: tp->snd_limited too big",
 					    __func__));
-					tp->snd_cwnd = tp->snd_ssthresh +
-					     maxseg *
-					     (tp->t_dupacks - tp->snd_limited);
+					if (CC_ALGO(tp)->cong_signal == NULL)
+						tp->snd_cwnd = tp->snd_ssthresh +
+						    maxseg *
+						    (tp->t_dupacks - tp->snd_limited);
 					if (SEQ_GT(onxt, tp->snd_nxt))
 						tp->snd_nxt = onxt;
 					goto drop;

Modified: stable/11/sys/netinet/tcp_stacks/fastpath.c
==============================================================================
--- stable/11/sys/netinet/tcp_stacks/fastpath.c	Mon Mar  5 06:47:28 2018	(r330444)
+++ stable/11/sys/netinet/tcp_stacks/fastpath.c	Mon Mar  5 06:52:26 2018	(r330445)
@@ -1035,6 +1035,15 @@ tcp_do_slowpath(struct mbuf *m, struct tcphdr *th, str
 
 						if (awnd < tp->snd_ssthresh) {
 							tp->snd_cwnd += tp->t_maxseg;
+							/*
+							 * RFC5681 Section 3.2 talks about cwnd
+							 * inflation on additional dupacks and
+							 * deflation on recovering from loss.
+							 *
+							 * We keep cwnd into check so that
+							 * we don't have to 'deflate' it when we
+							 * get out of recovery.
+							 */
 							if (tp->snd_cwnd > tp->snd_ssthresh)
 								tp->snd_cwnd = tp->snd_ssthresh;
 						}
@@ -1073,19 +1082,22 @@ tcp_do_slowpath(struct mbuf *m, struct tcphdr *th, str
 						TCPSTAT_INC(
 						    tcps_sack_recovery_episode);
 						tp->sack_newdata = tp->snd_nxt;
-						tp->snd_cwnd = tp->t_maxseg;
+						if (CC_ALGO(tp)->cong_signal == NULL)
+							tp->snd_cwnd = tp->t_maxseg;
 						(void) tp->t_fb->tfb_tcp_output(tp);
 						goto drop;
 					}
 					tp->snd_nxt = th->th_ack;
-					tp->snd_cwnd = tp->t_maxseg;
+					if (CC_ALGO(tp)->cong_signal == NULL)
+						tp->snd_cwnd = tp->t_maxseg;
 					(void) tp->t_fb->tfb_tcp_output(tp);
 					KASSERT(tp->snd_limited <= 2,
 					    ("%s: tp->snd_limited too big",
 					    __func__));
-					tp->snd_cwnd = tp->snd_ssthresh +
-					     tp->t_maxseg *
-					     (tp->t_dupacks - tp->snd_limited);
+					if (CC_ALGO(tp)->cong_signal == NULL)
+						tp->snd_cwnd = tp->snd_ssthresh +
+						    tp->t_maxseg *
+						    (tp->t_dupacks - tp->snd_limited);
 					if (SEQ_GT(onxt, tp->snd_nxt))
 						tp->snd_nxt = onxt;
 					goto drop;



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201803050652.w256qQvu071205>