6 files changed, 163 insertions, 15 deletions

diff --git a/sys/netinet/tcp.h b/sys/netinet/tcp.h
index ba88dd3c89ea..6c214e80b557 100644
--- a/sys/netinet/tcp.h
+++ b/sys/netinet/tcp.h
@@ -118,6 +118,8 @@ struct tcphdr {
 
 #define TCP_MAX_WINSHIFT	14	/* maximum window shift */
 
+#define TCP_MAXBURST		4 	/* maximum segments in a burst */
+
 #define TCP_MAXHLEN	(0xf<<2)	/* max length of header in bytes */
 #define TCP_MAXOLEN	(TCP_MAXHLEN - sizeof(struct tcphdr))
 					/* max space left for options */
diff --git a/sys/netinet/tcp_input.c b/sys/netinet/tcp_input.c
index 792225399ef4..82ab08fbb957 100644
--- a/sys/netinet/tcp_input.c
+++ b/sys/netinet/tcp_input.c
@@ -148,6 +148,7 @@ static void	 tcp_pulloutofband __P((struct socket *,
 static int	 tcp_reass __P((struct tcpcb *, struct tcphdr *, int *,
 				struct mbuf *));
 static void	 tcp_xmit_timer __P((struct tcpcb *, int));
+static int	 tcp_newreno __P((struct tcpcb *, struct tcphdr *));
 
 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
 #ifdef INET6
@@ -1104,6 +1105,7 @@ findpcb:
 		tp->irs = th->th_seq;
 		tcp_sendseqinit(tp);
 		tcp_rcvseqinit(tp);
+		tp->snd_recover = tp->snd_una;
 		/*
 		 * Initialization of the tcpcb for transaction;
 		 *   set SND.WND = SEG.WND,
@@ -1780,10 +1782,20 @@ trimthenstep6:
 					u_int win =
 					    min(tp->snd_wnd, tp->snd_cwnd) / 2 /
 						tp->t_maxseg;
-
+					if (tcp_do_newreno && SEQ_LT(th->th_ack,
+					    tp->snd_recover)) {
+						/* False retransmit, should not
+						 * cut window
+						 */
+						tp->snd_cwnd += tp->t_maxseg;
+						tp->t_dupacks = 0;
+						(void) tcp_output(tp);
+						goto drop;
+					}
 					if (win < 2)
 						win = 2;
 					tp->snd_ssthresh = win * tp->t_maxseg;
+					tp->snd_recover = tp->snd_max;
 					callout_stop(tp->tt_rexmt);
 					tp->t_rtttime = 0;
 					tp->snd_nxt = th->th_ack;
@@ -1807,10 +1819,26 @@ trimthenstep6:
 		 * If the congestion window was inflated to account
 		 * for the other side's cached packets, retract it.
 		 */
-		if (tp->t_dupacks >= tcprexmtthresh &&
-		    tp->snd_cwnd > tp->snd_ssthresh)
-			tp->snd_cwnd = tp->snd_ssthresh;
-		tp->t_dupacks = 0;
+		if (tcp_do_newreno == 0) {
+                        if (tp->t_dupacks >= tcprexmtthresh &&
+                                tp->snd_cwnd > tp->snd_ssthresh)
+                                tp->snd_cwnd = tp->snd_ssthresh;
+                        tp->t_dupacks = 0;
+                } else if (tp->t_dupacks >= tcprexmtthresh &&
+		    !tcp_newreno(tp, th)) {
+                        /*
+                         * Window inflation should have left us with approx.
+                         * snd_ssthresh outstanding data.  But in case we
+                         * would be inclined to send a burst, better to do
+                         * it via the slow start mechanism.
+                         */
+			if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max))
+                                tp->snd_cwnd =
+				    tp->snd_max - th->th_ack + tp->t_maxseg;
+			else
+                        	tp->snd_cwnd = tp->snd_ssthresh;
+                        tp->t_dupacks = 0;
+                }
 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
 			tcpstat.tcps_rcvacktoomuch++;
 			goto dropafterack;
@@ -1903,7 +1931,8 @@ process_ACK:
 
 		if (cw > tp->snd_ssthresh)
 			incr = incr * incr / cw;
-		tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale);
+		if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover))
+			tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale);
 		}
 		if (acked > so->so_snd.sb_cc) {
 			tp->snd_wnd -= so->so_snd.sb_cc;
@@ -2791,3 +2820,41 @@ tcp_mssopt(tp)
 
 	return rt->rt_ifp->if_mtu - min_protoh;
 }
+
+
+/*
+ * Checks for partial ack.  If partial ack arrives, force the retransmission
+ * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
+ * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
+ * be started again.  If the ack advances at least to tp->snd_recover, return 0.
+ */
+static int
+tcp_newreno(tp, th)
+	struct tcpcb *tp;
+	struct tcphdr *th;
+{
+	if (SEQ_LT(th->th_ack, tp->snd_recover)) {
+		tcp_seq onxt = tp->snd_nxt;
+		tcp_seq ouna = tp->snd_una;  /* Haven't updated snd_una yet*/
+		u_long  ocwnd = tp->snd_cwnd;
+
+		callout_stop(tp->tt_rexmt);
+		tp->t_rtttime = 0;
+		tp->snd_nxt = th->th_ack;
+		tp->snd_cwnd = tp->t_maxseg;
+		tp->snd_una = th->th_ack;
+		(void) tcp_output(tp);
+
+		tp->snd_cwnd = ocwnd;
+		tp->snd_una = ouna;
+		if (SEQ_GT(onxt, tp->snd_nxt))
+			tp->snd_nxt = onxt;
+		/*
+		 * Partial window deflation.  Relies on fact that tp->snd_una
+		 * not updated yet.
+		 */
+		tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
+		return (1);
+	}
+	return (0);
+}
diff --git a/sys/netinet/tcp_output.c b/sys/netinet/tcp_output.c
index 238d25a6b15d..736e128ecafe 100644
--- a/sys/netinet/tcp_output.c
+++ b/sys/netinet/tcp_output.c
@@ -98,6 +98,9 @@ int ss_fltsz_local = TCP_MAXWIN;               /* something large */
 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW,
 	&ss_fltsz_local, 1, "Slow start flight size for local networks");
 
+int     tcp_do_newreno = 1;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, &tcp_do_newreno,
+        0, "Enable NewReno Algorithms");
 /*
  * Tcp output routine: figure out what should be sent and send it.
  */
@@ -118,6 +121,7 @@ tcp_output(tp)
 	u_char opt[TCP_MAXOLEN];
 	unsigned ipoptlen, optlen, hdrlen;
 	int idle, sendalot;
+	int maxburst = TCP_MAXBURST;
 	struct rmxp_tao *taop;
 	struct rmxp_tao tao_noncached;
 #ifdef INET6
@@ -778,12 +782,12 @@ send:
 		 */
 		if (!callout_active(tp->tt_rexmt) &&
 		    tp->snd_nxt != tp->snd_una) {
-			callout_reset(tp->tt_rexmt, tp->t_rxtcur,
-				      tcp_timer_rexmt, tp);
 			if (callout_active(tp->tt_persist)) {
 				callout_stop(tp->tt_persist);
 				tp->t_rxtshift = 0;
 			}
+			callout_reset(tp->tt_rexmt, tp->t_rxtcur,
+				      tcp_timer_rexmt, tp);
 		}
 	} else
 		if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
@@ -889,7 +893,7 @@ out:
 	tp->t_flags &= ~TF_ACKNOW;
 	if (tcp_delack_enabled)
 		callout_stop(tp->tt_delack);
-	if (sendalot)
+	if (sendalot && (!tcp_do_newreno || --maxburst))
 		goto again;
 	return (0);
 }
diff --git a/sys/netinet/tcp_reass.c b/sys/netinet/tcp_reass.c
index 792225399ef4..82ab08fbb957 100644
--- a/sys/netinet/tcp_reass.c
+++ b/sys/netinet/tcp_reass.c
@@ -148,6 +148,7 @@ static void	 tcp_pulloutofband __P((struct socket *,
 static int	 tcp_reass __P((struct tcpcb *, struct tcphdr *, int *,
 				struct mbuf *));
 static void	 tcp_xmit_timer __P((struct tcpcb *, int));
+static int	 tcp_newreno __P((struct tcpcb *, struct tcphdr *));
 
 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
 #ifdef INET6
@@ -1104,6 +1105,7 @@ findpcb:
 		tp->irs = th->th_seq;
 		tcp_sendseqinit(tp);
 		tcp_rcvseqinit(tp);
+		tp->snd_recover = tp->snd_una;
 		/*
 		 * Initialization of the tcpcb for transaction;
 		 *   set SND.WND = SEG.WND,
@@ -1780,10 +1782,20 @@ trimthenstep6:
 					u_int win =
 					    min(tp->snd_wnd, tp->snd_cwnd) / 2 /
 						tp->t_maxseg;
-
+					if (tcp_do_newreno && SEQ_LT(th->th_ack,
+					    tp->snd_recover)) {
+						/* False retransmit, should not
+						 * cut window
+						 */
+						tp->snd_cwnd += tp->t_maxseg;
+						tp->t_dupacks = 0;
+						(void) tcp_output(tp);
+						goto drop;
+					}
 					if (win < 2)
 						win = 2;
 					tp->snd_ssthresh = win * tp->t_maxseg;
+					tp->snd_recover = tp->snd_max;
 					callout_stop(tp->tt_rexmt);
 					tp->t_rtttime = 0;
 					tp->snd_nxt = th->th_ack;
@@ -1807,10 +1819,26 @@ trimthenstep6:
 		 * If the congestion window was inflated to account
 		 * for the other side's cached packets, retract it.
 		 */
-		if (tp->t_dupacks >= tcprexmtthresh &&
-		    tp->snd_cwnd > tp->snd_ssthresh)
-			tp->snd_cwnd = tp->snd_ssthresh;
-		tp->t_dupacks = 0;
+		if (tcp_do_newreno == 0) {
+                        if (tp->t_dupacks >= tcprexmtthresh &&
+                                tp->snd_cwnd > tp->snd_ssthresh)
+                                tp->snd_cwnd = tp->snd_ssthresh;
+                        tp->t_dupacks = 0;
+                } else if (tp->t_dupacks >= tcprexmtthresh &&
+		    !tcp_newreno(tp, th)) {
+                        /*
+                         * Window inflation should have left us with approx.
+                         * snd_ssthresh outstanding data.  But in case we
+                         * would be inclined to send a burst, better to do
+                         * it via the slow start mechanism.
+                         */
+			if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max))
+                                tp->snd_cwnd =
+				    tp->snd_max - th->th_ack + tp->t_maxseg;
+			else
+                        	tp->snd_cwnd = tp->snd_ssthresh;
+                        tp->t_dupacks = 0;
+                }
 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
 			tcpstat.tcps_rcvacktoomuch++;
 			goto dropafterack;
@@ -1903,7 +1931,8 @@ process_ACK:
 
 		if (cw > tp->snd_ssthresh)
 			incr = incr * incr / cw;
-		tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale);
+		if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover))
+			tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale);
 		}
 		if (acked > so->so_snd.sb_cc) {
 			tp->snd_wnd -= so->so_snd.sb_cc;
@@ -2791,3 +2820,41 @@ tcp_mssopt(tp)
 
 	return rt->rt_ifp->if_mtu - min_protoh;
 }
+
+
+/*
+ * Checks for partial ack.  If partial ack arrives, force the retransmission
+ * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
+ * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
+ * be started again.  If the ack advances at least to tp->snd_recover, return 0.
+ */
+static int
+tcp_newreno(tp, th)
+	struct tcpcb *tp;
+	struct tcphdr *th;
+{
+	if (SEQ_LT(th->th_ack, tp->snd_recover)) {
+		tcp_seq onxt = tp->snd_nxt;
+		tcp_seq ouna = tp->snd_una;  /* Haven't updated snd_una yet*/
+		u_long  ocwnd = tp->snd_cwnd;
+
+		callout_stop(tp->tt_rexmt);
+		tp->t_rtttime = 0;
+		tp->snd_nxt = th->th_ack;
+		tp->snd_cwnd = tp->t_maxseg;
+		tp->snd_una = th->th_ack;
+		(void) tcp_output(tp);
+
+		tp->snd_cwnd = ocwnd;
+		tp->snd_una = ouna;
+		if (SEQ_GT(onxt, tp->snd_nxt))
+			tp->snd_nxt = onxt;
+		/*
+		 * Partial window deflation.  Relies on fact that tp->snd_una
+		 * not updated yet.
+		 */
+		tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
+		return (1);
+	}
+	return (0);
+}
diff --git a/sys/netinet/tcp_timer.c b/sys/netinet/tcp_timer.c
index 4555eac60079..1cf64fe2cb96 100644
--- a/sys/netinet/tcp_timer.c
+++ b/sys/netinet/tcp_timer.c
@@ -416,6 +416,11 @@ tcp_timer_rexmt(xtp)
 	}
 	tp->snd_nxt = tp->snd_una;
 	/*
+	 * Note:  We overload snd_recover to function also as the
+	 * snd_last variable described in RFC 2582
+	 */
+	tp->snd_recover = tp->snd_max;
+	/*
 	 * Force a segment to be sent.
 	 */
 	tp->t_flags |= TF_ACKNOW;
diff --git a/sys/netinet/tcp_var.h b/sys/netinet/tcp_var.h
index 5d64cf28ec53..5798c84cba52 100644
--- a/sys/netinet/tcp_var.h
+++ b/sys/netinet/tcp_var.h
@@ -119,6 +119,8 @@ struct tcpcb {
 					 * for slow start exponential to
 					 * linear switch
 					 */
+	tcp_seq	snd_recover;		/* for use in fast recovery */
+
 	u_int	t_maxopd;		/* mss plus options */
 
 	u_long	t_rcvtime;		/* inactivity time */
@@ -365,6 +367,7 @@ extern	struct inpcbinfo tcbinfo;
 extern	struct tcpstat tcpstat;	/* tcp statistics */
 extern	int tcp_mssdflt;	/* XXX */
 extern	int tcp_delack_enabled;
+extern	int tcp_do_newreno;
 extern	int ss_fltsz;
 extern	int ss_fltsz_local;