Add the TCP socket option TCP_NOPUSH to delay sending the stream.

This is useful to aggregate data in the kernel from multiple sources
like writes and socket splicing.  It avoids sending small packets.
From FreeBSD via David Hill; OK mikeb@ henning@

4 files changed, 24 insertions, 6 deletions

diff --git a/sys/netinet/tcp.h b/sys/netinet/tcp.h
index c4ae9533f90..ab62ec0bc34 100644
--- a/sys/netinet/tcp.h
+++ b/sys/netinet/tcp.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp.h,v 1.19 2013/06/02 23:12:23 henning Exp $	*/
+/*	$OpenBSD: tcp.h,v 1.20 2013/08/12 21:57:16 bluhm Exp $	*/
 /*	$NetBSD: tcp.h,v 1.8 1995/04/17 05:32:58 cgd Exp $	*/
 
 /*
@@ -127,5 +127,6 @@ struct tcphdr {
 #define	TCP_MAXSEG		0x02   /* set maximum segment size */
 #define	TCP_MD5SIG		0x04   /* enable TCP MD5 signature option */
 #define	TCP_SACK_ENABLE		0x08   /* enable SACKs (if disabled by def.) */
+#define	TCP_NOPUSH		0x10   /* don't push last block of write */
 
 #endif /* _NETINET_TCP_H_ */
diff --git a/sys/netinet/tcp_output.c b/sys/netinet/tcp_output.c
index c5cd8a8eaec..9d5b469a51f 100644
--- a/sys/netinet/tcp_output.c
+++ b/sys/netinet/tcp_output.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_output.c,v 1.98 2013/06/03 16:57:06 bluhm Exp $	*/
+/*	$OpenBSD: tcp_output.c,v 1.99 2013/08/12 21:57:16 bluhm Exp $	*/
 /*	$NetBSD: tcp_output.c,v 1.16 1997/06/03 16:17:09 kml Exp $	*/
 
 /*
@@ -415,8 +415,9 @@ again:
 	if (len) {
 		if (len == txmaxseg)
 			goto send;
-		if ((idle || tp->t_flags & TF_NODELAY) &&
-		    len + off >= so->so_snd.sb_cc && !soissending(so))
+		if ((idle || (tp->t_flags & TF_NODELAY)) &&
+		    len + off >= so->so_snd.sb_cc && !soissending(so) &&
+		    (tp->t_flags & TF_NOPUSH) == 0)
 			goto send;
 		if (tp->t_force)
 			goto send;
diff --git a/sys/netinet/tcp_usrreq.c b/sys/netinet/tcp_usrreq.c
index 4e108d144fe..9fd519615ab 100644
--- a/sys/netinet/tcp_usrreq.c
+++ b/sys/netinet/tcp_usrreq.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_usrreq.c,v 1.113 2013/08/06 07:31:48 bluhm Exp $	*/
+/*	$OpenBSD: tcp_usrreq.c,v 1.114 2013/08/12 21:57:16 bluhm Exp $	*/
 /*	$NetBSD: tcp_usrreq.c,v 1.20 1996/02/13 23:44:16 christos Exp $	*/
 
 /*
@@ -532,6 +532,18 @@ tcp_ctloutput(op, so, level, optname, mp)
 				tp->t_flags &= ~TF_NODELAY;
 			break;
 
+		case TCP_NOPUSH:
+			if (m == NULL || m->m_len < sizeof (int))
+				error = EINVAL;
+			else if (*mtod(m, int *))
+				tp->t_flags |= TF_NOPUSH;
+			else if (tp->t_flags & TF_NOPUSH) {
+				tp->t_flags &= ~TF_NOPUSH;
+				if (TCPS_HAVEESTABLISHED(tp->t_state))
+					error = tcp_output(tp);
+			}
+			break;
+
 		case TCP_MAXSEG:
 			if (m == NULL || m->m_len < sizeof (int)) {
 				error = EINVAL;
@@ -605,6 +617,9 @@ tcp_ctloutput(op, so, level, optname, mp)
 		case TCP_NODELAY:
 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
 			break;
+		case TCP_NOPUSH:
+			*mtod(m, int *) = tp->t_flags & TF_NOPUSH;
+			break;
 		case TCP_MAXSEG:
 			*mtod(m, int *) = tp->t_maxseg;
 			break;
diff --git a/sys/netinet/tcp_var.h b/sys/netinet/tcp_var.h
index f3053598a58..acded92556b 100644
--- a/sys/netinet/tcp_var.h
+++ b/sys/netinet/tcp_var.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_var.h,v 1.101 2013/06/01 16:22:05 bluhm Exp $	*/
+/*	$OpenBSD: tcp_var.h,v 1.102 2013/08/12 21:57:16 bluhm Exp $	*/
 /*	$NetBSD: tcp_var.h,v 1.17 1996/02/13 23:44:24 christos Exp $	*/
 
 /*
@@ -97,6 +97,7 @@ struct tcpcb {
 #define TF_PMTUD_PEND	0x00400000	/* Path MTU Discovery pending */
 #define TF_NEEDOUTPUT	0x00800000	/* call tcp_output after tcp_input */
 #define TF_BLOCKOUTPUT	0x01000000	/* avert tcp_output during tcp_input */
+#define TF_NOPUSH	0x02000000	/* don't push */
 
 	struct	mbuf *t_template;	/* skeletal packet for transmit */
 	struct	inpcb *t_inpcb;		/* back pointer to internet pcb */