Index: net/if.h
===================================================================
RCS file: /home/ncvs/src/sys/net/if.h,v
retrieving revision 1.103
diff -u -p -r1.103 if.h
--- net/if.h	19 Jun 2006 22:20:44 -0000	1.103
+++ net/if.h	3 Sep 2006 19:01:01 -0000
@@ -180,7 +180,12 @@ struct if_data {
 #define	IF_Mbps(x)	(IF_Kbps((x) * 1000))	/* megabits/sec. */
 #define	IF_Gbps(x)	(IF_Mbps((x) * 1000))	/* gigabits/sec. */
 
-/* Capabilities that interfaces can advertise. */
+/*
+ * Capabilities that interfaces can advertise.
+ * if_capabilities
+ * if_capenabled
+ * ifi_hwassist in mbuf CSUM_ flag form, controlled by above
+ */
 #define IFCAP_RXCSUM		0x0001  /* can offload checksum on RX */
 #define IFCAP_TXCSUM		0x0002  /* can offload checksum on TX */
 #define IFCAP_NETCONS		0x0004  /* can be a network console */
@@ -189,6 +194,7 @@ struct if_data {
 #define	IFCAP_JUMBO_MTU		0x0020	/* 9000 byte MTU supported */
 #define	IFCAP_POLLING		0x0040	/* driver supports polling */
 #define	IFCAP_VLAN_HWCSUM	0x0080	/* can do IFCAP_HWCSUM on VLANs */
+#define	IFCAP_TSO		0x0100	/* can do TCP Segmentation Offload */
 
 #define IFCAP_HWCSUM		(IFCAP_RXCSUM | IFCAP_TXCSUM)
 
Index: netinet/ip_output.c
===================================================================
RCS file: /home/ncvs/src/sys/netinet/ip_output.c,v
retrieving revision 1.259
diff -u -p -r1.259 ip_output.c
--- netinet/ip_output.c	17 Aug 2006 00:37:03 -0000	1.259
+++ netinet/ip_output.c	3 Sep 2006 19:01:01 -0000
@@ -486,17 +495,24 @@ passout:
 
 	/*
 	 * If small enough for interface, or the interface will take
-	 * care of the fragmentation for us, can just send directly.
+	 * care of the fragmentation for us, we can just send directly.
 	 */
-	if (ip->ip_len <= ifp->if_mtu || (ifp->if_hwassist & CSUM_FRAGMENT &&
-	    ((ip->ip_off & IP_DF) == 0))) {
+	if (ip->ip_len <= ifp->if_mtu ||
+	    (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
+	    ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
 		ip->ip_len = htons(ip->ip_len);
 		ip->ip_off = htons(ip->ip_off);
 		ip->ip_sum = 0;
 		if (sw_csum & CSUM_DELAY_IP)
 			ip->ip_sum = in_cksum(m, hlen);
 
-		/* Record statistics for this interface address. */
+		/*
+		 * Record statistics for this interface address.
+		 * XXX: With CSUM_TSO the packet count will be
+		 * incorrect because the packetization of a large
+		 * number of bytes is done in the network card
+		 * instead of the stack.
+		 */
 		if (!(flags & IP_FORWARDING) && ia) {
 			ia->ia_ifa.if_opackets++;
 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
Index: netinet/tcp_input.c
===================================================================
RCS file: /home/ncvs/src/sys/netinet/tcp_input.c,v
retrieving revision 1.304
diff -u -p -r1.304 tcp_input.c
--- netinet/tcp_input.c	11 Aug 2006 21:15:23 -0000	1.304
+++ netinet/tcp_input.c	3 Sep 2006 19:01:02 -0000
@@ -2857,6 +2867,7 @@ tcp_mss(tp, offer)
 	struct socket *so;
 	struct hc_metrics_lite metrics;
 	int origoffer = offer;
+	int mtuflags = 0;
 #ifdef INET6
 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
 	size_t min_protoh = isipv6 ?
@@ -2869,12 +2880,12 @@ tcp_mss(tp, offer)
 	/* initialize */
 #ifdef INET6
 	if (isipv6) {
-		maxmtu = tcp_maxmtu6(&inp->inp_inc);
+		maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags);
 		tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
 	} else
 #endif
 	{
-		maxmtu = tcp_maxmtu(&inp->inp_inc);
+		maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags);
 		tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
 	}
 	so = inp->inp_socket;
@@ -3081,6 +3092,10 @@ tcp_mss(tp, offer)
 		tp->snd_cwnd = mss * ss_fltsz_local;
 	else
 		tp->snd_cwnd = mss * ss_fltsz;
+
+	/* Check the interface for TSO capabilities. */
+	if (mtuflags & CSUM_TSO)
+		tp->t_flags |= TF_TSO;
 }
 
 /*
@@ -3103,14 +3118,14 @@ tcp_mssopt(inc)
 #ifdef INET6
 	if (isipv6) {
 		mss = tcp_v6mssdflt;
-		maxmtu = tcp_maxmtu6(inc);
+		maxmtu = tcp_maxmtu6(inc, NULL);
 		thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
 	} else
 #endif
 	{
 		mss = tcp_mssdflt;
-		maxmtu = tcp_maxmtu(inc);
+		maxmtu = tcp_maxmtu(inc, NULL);
 		thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
 		min_protoh = sizeof(struct tcpiphdr);
 	}
Index: netinet/tcp_output.c
===================================================================
RCS file: /home/ncvs/src/sys/netinet/tcp_output.c,v
retrieving revision 1.115
diff -u -p -r1.115 tcp_output.c
--- netinet/tcp_output.c	23 Feb 2006 21:14:34 -0000	1.115
+++ netinet/tcp_output.c	3 Sep 2006 19:01:02 -0000
@@ -102,8 +102,12 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, loca
 	&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");
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW,
+	&tcp_do_newreno, 0, "Enable NewReno Algorithms");
+
+int	tcp_do_tso = 1;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_RW,
+	&tcp_do_tso, 0, "Enable TCP Segmentation Offload");
 
 /*
  * Tcp output routine: figure out what should be sent and send it.
@@ -127,6 +131,7 @@ tcp_output(struct tcpcb *tp)
 	int i, sack_rxmit;
 	int sack_bytes_rxmt;
 	struct sackhole *p;
+	int tso = 0;
 #if 0
 	int maxburst = TCP_MAXBURST;
 #endif
@@ -376,12 +381,33 @@ after_sack_rexmit:
 
 	/*
 	 * len will be >= 0 after this point.  Truncate to the maximum
-	 * segment length and ensure that FIN is removed if the length
-	 * no longer contains the last data byte.
+	 * segment length or enable TCP Segmentation Offloading (if supported
+	 * by hardware) and ensure that FIN is removed if the length no longer
+	 * contains the last data byte.
+	 *
+	 * TSO may only be used if we are in a pure bulk sending state.  The
+	 * presence of TCP-MD5, SACK retransmits, SACK advertizements and
+	 * IP options prevent using TSO.  With TSO the TCP header is the same
+	 * (except for the sequence number) for all generated packets.  This
+	 * makes it impossible to transmit any options which vary per generated
+	 * segment or packet.
+	 *
+	 * The length of TSO bursts is limited by TCP_MAXWIN.  That limit and
+	 * removal of FIN (if not already catched here) are handled later after
+	 * the exact length of the TCP options are known.
 	 */
 	if (len > tp->t_maxseg) {
-		len = tp->t_maxseg;
-		sendalot = 1;
+		if ((tp->t_flags & TF_TSO) && tcp_do_tso &&
+		    ((tp->t_flags & TF_SIGNATURE) == 0) &&
+		    tp->rcv_numsacks == 0 && sack_rxmit == 0 &&
+		    tp->t_inpcb->inp_options == NULL &&
+		    tp->t_inpcb->in6p_options == NULL) {
+			tso = 1;
+		} else {
+			len = tp->t_maxseg;
+			sendalot = 1;
+			tso = 0;
+		}
 	}
 	if (sack_rxmit) {
 		if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc))
@@ -397,7 +423,7 @@ after_sack_rexmit:
 	 * Sender silly window avoidance.   We transmit under the following
 	 * conditions when len is non-zero:
 	 *
-	 *	- We have a full segment
+	 *	- We have a full segment (or more with TSO)
 	 *	- This is the last buffer in a write()/send() and we are
 	 *	  either idle or running NODELAY
 	 *	- we've timed out (e.g. persist timer)
@@ -406,7 +432,7 @@ after_sack_rexmit:
 	 *	- we need to retransmit
 	 */
 	if (len) {
-		if (len == tp->t_maxseg)
+		if (len >= tp->t_maxseg)
 			goto send;
 		/*
 		 * NOTE! on localhost connections an 'ack' from the remote
@@ -702,14 +728,27 @@ send:
 	 * bump the packet length beyond the t_maxopd length.
 	 * Clear the FIN bit because we cut off the tail of
 	 * the segment.
+	 *
+	 * When doing TSO limit a burst to TCP_MAXWIN and set the
+	 * flag to continue sending and prevent the last segment
+	 * from being fractional thus making them all equal sized.
 	 */
 	if (len + optlen + ipoptlen > tp->t_maxopd) {
-		/*
-		 * If there is still more to send, don't close the connection.
-		 */
 		flags &= ~TH_FIN;
-		len = tp->t_maxopd - optlen - ipoptlen;
-		sendalot = 1;
+		if (tso) {
+			if (len > TCP_MAXWIN) {
+				len = TCP_MAXWIN - TCP_MAXWIN %
+					(tp->t_maxopd - optlen);
+				sendalot = 1;
+			} else {
+				len = min(len, TCP_MAXWIN);
+				if (tp->t_flags & TF_NEEDFIN)
+					sendalot = 1;
+			}
+		} else {
+			len = tp->t_maxopd - optlen - ipoptlen;
+			sendalot = 1;
+		}
 	}
 
 /*#ifdef DIAGNOSTIC*/
@@ -926,14 +965,20 @@ send:
 	 */
 	m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
 #ifdef INET6
-	if (isipv6)
+	if (isipv6) {
 		/*
 		 * ip6_plen is not need to be filled now, and will be filled
 		 * in ip6_output.
 		 */
 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
 				       sizeof(struct tcphdr) + optlen + len);
-	else
+
+		/* Enable TSO and specify the size of the segments. */
+		if (tso) {
+			m->m_pkthdr.csum_flags = CSUM_TSO;
+			m->m_pkthdr.tso_segsz = tp->t_maxopd - optlen;
+		}
+	} else
 #endif /* INET6 */
 	{
 		m->m_pkthdr.csum_flags = CSUM_TCP;
@@ -941,6 +986,12 @@ send:
 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen));
 
+		/* Enable TSO and specify the size of the segments. */
+		if (tso) {
+			m->m_pkthdr.csum_flags = CSUM_TSO;
+			m->m_pkthdr.tso_segsz = tp->t_maxopd - optlen;
+		}
+
 		/* IP version must be set here for ipv4/ipv6 checking later */
 		KASSERT(ip->ip_v == IPVERSION,
 		    ("%s: IP version incorrect: %d", __func__, ip->ip_v));
@@ -1119,11 +1170,22 @@ out:
 		}
 		if (error == EMSGSIZE) {
 			/*
-			 * ip_output() will have already fixed the route
-			 * for us.  tcp_mtudisc() will, as its last action,
-			 * initiate retransmission, so it is important to
-			 * not do so here.
+			 * For some reason the interface we used initially
+			 * to send segments changed to another or lowered
+			 * its MTU.
+			 *
+			 * tcp_mtudisc() will find out the new MTU and as
+			 * its last action, initiate retransmission, so it
+			 * is important to not do so here.
+			 *
+			 * If TSO was active we either got an interface
+			 * without TSO capabilits or TSO was turned off.
+			 * Disable it for this connection as too and
+			 * immediatly retry with MSS sized segments generated
+			 * by this function.
 			 */
+			if (tso)
+				tp->t_flags &= ~TF_TSO;
 			tcp_mtudisc(tp->t_inpcb, 0);
 			return 0;
 		}
@@ -1132,6 +1194,7 @@ out:
 			tp->t_softerror = error;
 			return (0);
 		}
+		/* if (error == EACCES) */
 		return (error);
 	}
 	tcpstat.tcps_sndtotal++;
Index: netinet/tcp_subr.c
===================================================================
RCS file: /home/ncvs/src/sys/netinet/tcp_subr.c,v
retrieving revision 1.255
diff -u -p -r1.255 tcp_subr.c
--- netinet/tcp_subr.c	26 Aug 2006 17:53:19 -0000	1.255
+++ netinet/tcp_subr.c	3 Sep 2006 19:01:02 -0000
@@ -1230,7 +1230,7 @@ tcp_ctlinput(int cmd, struct sockaddr *s
 					     * or route MTU.  tcp_mtudisc()
 					     * will do right thing by itself.
 					     */
-					    if (mtu <= tcp_maxmtu(&inc))
+					    if (mtu <= tcp_maxmtu(&inc, NULL))
 						tcp_hc_updatemtu(&inc, mtu);
 					}
 
@@ -1509,9 +1509,9 @@ tcp_mtudisc(struct inpcb *inp, int errno
 	maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
 	romtu =
 #ifdef INET6
-	    isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
+	    isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) :
 #endif /* INET6 */
-	    tcp_maxmtu(&inp->inp_inc);
+	    tcp_maxmtu(&inp->inp_inc, NULL);
 	if (!maxmtu)
 		maxmtu = romtu;
 	else
@@ -1588,7 +1588,7 @@ tcp_mtudisc(struct inpcb *inp, int errno
  * to get the interface MTU.
  */
 u_long
-tcp_maxmtu(struct in_conninfo *inc)
+tcp_maxmtu(struct in_conninfo *inc, int *flags)
 {
 	struct route sro;
 	struct sockaddr_in *dst;
@@ -1611,6 +1611,12 @@ tcp_maxmtu(struct in_conninfo *inc)
 			maxmtu = ifp->if_mtu;
 		else
 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
+
+		/* Report additional interface capabilities or information. */
+		if (flags != NULL) {
+			if (ifp->if_hwassist & CSUM_TSO)
+				*flags |= CSUM_TSO;
+		}
 		RTFREE(sro.ro_rt);
 	}
 	return (maxmtu);
@@ -1618,7 +1624,7 @@ tcp_maxmtu(struct in_conninfo *inc)
 
 #ifdef INET6
 u_long
-tcp_maxmtu6(struct in_conninfo *inc)
+tcp_maxmtu6(struct in_conninfo *inc, int *flags)
 {
 	struct route_in6 sro6;
 	struct ifnet *ifp;
@@ -1640,6 +1646,12 @@ tcp_maxmtu6(struct in_conninfo *inc)
 		else
 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
+
+		/* Report additional interface capabilities or information. */
+		if (flags != NULL) {
+			if (ifp->if_hwassist & CSUM_TSO)
+				*flags |= CSUM_TSO;
+		}
 		RTFREE(sro6.ro_rt);
 	}
 
Index: netinet/tcp_var.h
===================================================================
RCS file: /home/ncvs/src/sys/netinet/tcp_var.h,v
retrieving revision 1.133
diff -u -p -r1.133 tcp_var.h
--- netinet/tcp_var.h	26 Jun 2006 15:35:25 -0000	1.133
+++ netinet/tcp_var.h	3 Sep 2006 19:01:02 -0000
@@ -114,6 +114,7 @@ struct tcpcb {
 #define	TF_WASFRECOVERY	0x200000	/* was in NewReno Fast Recovery */
 #define	TF_SIGNATURE	0x400000	/* require MD5 digests (RFC2385) */
 #define	TF_FORCEDATA	0x800000	/* force out a byte */
+#define	TF_TSO		0x1000000	/* TSO enabled on this connection */
 
 	tcp_seq	snd_una;		/* send unacknowledged */
 	tcp_seq	snd_max;		/* highest sequence number sent;
@@ -506,8 +508,8 @@ void	 tcp_init(void);
 void	 tcp_fini(void *);
 void	 tcp_reass_init(void);
 void	 tcp_input(struct mbuf *, int);
-u_long	 tcp_maxmtu(struct in_conninfo *);
-u_long	 tcp_maxmtu6(struct in_conninfo *);
+u_long	 tcp_maxmtu(struct in_conninfo *, int *);
+u_long	 tcp_maxmtu6(struct in_conninfo *, int *);
 void	 tcp_mss(struct tcpcb *, int);
 int	 tcp_mssopt(struct in_conninfo *);
 struct inpcb *
Index: netinet6/icmp6.c
===================================================================
RCS file: /home/ncvs/src/sys/netinet6/icmp6.c,v
retrieving revision 1.71
diff -u -p -r1.71 icmp6.c
--- netinet6/icmp6.c	4 Aug 2006 21:27:38 -0000	1.71
+++ netinet6/icmp6.c	3 Sep 2006 19:01:02 -0000
@@ -1132,7 +1132,7 @@ icmp6_mtudisc_update(ip6cp, validated)
 	if (in6_setscope(&inc.inc6_faddr, m->m_pkthdr.rcvif, NULL))
 		return;
 
-	if (mtu < tcp_maxmtu6(&inc)) {
+	if (mtu < tcp_maxmtu6(&inc, NULL)) {
 		tcp_hc_updatemtu(&inc, mtu);
 		icmp6stat.icp6s_pmtuchg++;
 	}
Index: sys/mbuf.h
===================================================================
RCS file: /home/ncvs/src/sys/sys/mbuf.h,v
retrieving revision 1.192
diff -u -p -r1.192 mbuf.h
--- sys/mbuf.h	24 Jul 2006 01:49:57 -0000	1.192
+++ sys/mbuf.h	3 Sep 2006 19:01:03 -0000
@@ -110,6 +110,7 @@ struct pkthdr {
 	/* variables for hardware checksum */
 	int	csum_flags;		/* flags regarding checksum */
 	int	csum_data;		/* data field used by csum routines */
+	int	tso_segsz;		/* TSO segment size */
 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
 };
 
@@ -215,6 +216,7 @@ struct mbuf {
 #define	CSUM_UDP		0x0004		/* will csum UDP */
 #define	CSUM_IP_FRAGS		0x0008		/* will csum IP fragments */
 #define	CSUM_FRAGMENT		0x0010		/* will do IP fragmentation */
+#define	CSUM_TSO		0x0020		/* will do TSO */
 
 #define	CSUM_IP_CHECKED		0x0100		/* did csum IP */
 #define	CSUM_IP_VALID		0x0200		/*   ... the csum is valid */