From 4730cb344fefcadc57fc089c0642bce97df4b76a Mon Sep 17 00:00:00 2001
From: tuexen <tuexen@FreeBSD.org>
Date: Thu, 9 May 2019 07:34:15 +0000
Subject: [PATCH] Receiver side DSACK implemenation.

This adds initial support for RFC 2883.

Submitted by:		Richard Scheffenegger
Reviewed by:		rrs@
Differential Revision:	https://reviews.freebsd.org/D19334
---
 sys/netinet/tcp_input.c | 36 ++++++++++++++++++++++++++---
 sys/netinet/tcp_sack.c  | 50 +++++++++++++++++++++++++++++++++++++----
 2 files changed, 79 insertions(+), 7 deletions(-)

diff --git a/sys/netinet/tcp_input.c b/sys/netinet/tcp_input.c
index bedfa856d5d3..066c0ccdcf04 100644
--- a/sys/netinet/tcp_input.c
+++ b/sys/netinet/tcp_input.c
@@ -2258,6 +2258,17 @@ tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
 			TCPSTAT_INC(tcps_rcvpartduppack);
 			TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
 		}
+		/*
+		 * DSACK - add SACK block for dropped range
+		 */
+		if (tp->t_flags & TF_SACK_PERMIT) {
+			tcp_update_sack_list(tp, th->th_seq, th->th_seq+tlen);
+			/*
+			 * ACK now, as the next in-sequence segment
+			 * will clear the DSACK block again
+			 */
+			tp->t_flags |= TF_ACKNOW;
+		}
 		drop_hdrlen += todrop;	/* drop from the top afterwards */
 		th->th_seq += todrop;
 		tlen -= todrop;
@@ -2986,6 +2997,8 @@ tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
 	if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
 		tcp_seq save_start = th->th_seq;
+		tcp_seq save_rnxt  = tp->rcv_nxt;
+		int     save_tlen  = tlen;
 		m_adj(m, drop_hdrlen);	/* delayed header drop */
 		/*
 		 * Insert segment which includes th into TCP reassembly queue
@@ -3025,11 +3038,28 @@ tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
 			 * m_adj() doesn't actually frees any mbufs
 			 * when trimming from the head.
 			 */
-			thflags = tcp_reass(tp, th, &save_start, &tlen, m);
+			tcp_seq temp = save_start;
+			thflags = tcp_reass(tp, th, &temp, &tlen, m);
 			tp->t_flags |= TF_ACKNOW;
 		}
-		if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
-			tcp_update_sack_list(tp, save_start, save_start + tlen);
+		if (tp->t_flags & TF_SACK_PERMIT) {
+			if (((tlen == 0) && (save_tlen > 0) &&
+			    (SEQ_LT(save_start, save_rnxt)))) {
+				// DSACK actually handled in the fastpath above
+				tcp_update_sack_list(tp, save_start, save_start + save_tlen);
+			} else
+			if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
+				// cleaning sackblks by using zero length update
+				tcp_update_sack_list(tp, save_start, save_start);
+			} else
+			if ((tlen > 0) && (tlen >= save_tlen)) {
+				// update of sackblks
+				tcp_update_sack_list(tp, save_start, save_start + save_tlen);
+			} else
+			if (tlen > 0) {
+				tcp_update_sack_list(tp, save_start, save_start+tlen);
+			}
+		}
 #if 0
 		/*
 		 * Note the amount of data that peer has sent into
diff --git a/sys/netinet/tcp_sack.c b/sys/netinet/tcp_sack.c
index 985b35122d5c..eb4064d29cf7 100644
--- a/sys/netinet/tcp_sack.c
+++ b/sys/netinet/tcp_sack.c
@@ -168,7 +168,7 @@ tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
 	INP_WLOCK_ASSERT(tp->t_inpcb);
 
 	/* Check arguments. */
-	KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end"));
+	KASSERT(SEQ_LEQ(rcv_start, rcv_end), ("rcv_start <= rcv_end"));
 
 	/* SACK block for the received segment. */
 	head_blk.start = rcv_start;
@@ -193,11 +193,53 @@ tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
 			 * Merge this SACK block into head_blk.  This SACK
 			 * block itself will be discarded.
 			 */
-			if (SEQ_GT(head_blk.start, start))
+			/*
+			 * |-|
+			 *   |---|  merge
+			 *
+			 *     |-|
+			 * |---|    merge
+			 *
+			 * |-----|
+			 *   |-|    DSACK smaller
+			 *
+			 *   |-|
+			 * |-----|  DSACK smaller
+			 */
+			if (head_blk.start == end)
 				head_blk.start = start;
-			if (SEQ_LT(head_blk.end, end))
+			else if (head_blk.end == start)
 				head_blk.end = end;
+			else {
+				if (SEQ_LT(head_blk.start, start)) {
+					tcp_seq temp = start;
+					start = head_blk.start;
+					head_blk.start = temp;
+				}
+				if (SEQ_GT(head_blk.end, end)) {
+					tcp_seq temp = end;
+					end = head_blk.end;
+					head_blk.end = temp;
+				}
+				if ((head_blk.start != start) ||
+				    (head_blk.end != end)) {
+					if ((num_saved >= 1) &&
+					   SEQ_GEQ(saved_blks[num_saved-1].start, start) &&
+					   SEQ_LEQ(saved_blks[num_saved-1].end, end))
+						num_saved--;
+					saved_blks[num_saved].start = start;
+					saved_blks[num_saved].end = end;
+					num_saved++;
+				}
+			}
 		} else {
+			/*
+			 * This block supercedes the prior block
+			 */
+			if ((num_saved >= 1) &&
+			   SEQ_GEQ(saved_blks[num_saved-1].start, start) &&
+			   SEQ_LEQ(saved_blks[num_saved-1].end, end))
+				num_saved--;
 			/*
 			 * Save this SACK block.
 			 */
@@ -211,7 +253,7 @@ tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
 	 * Update SACK list in tp->sackblks[].
 	 */
 	num_head = 0;
-	if (SEQ_GT(head_blk.start, tp->rcv_nxt)) {
+	if (SEQ_LT(rcv_start, rcv_end)) {
 		/*
 		 * The received data segment is an out-of-order segment.  Put
 		 * head_blk at the top of SACK list.