Add support in TCP for Path MTU discovery. This is highly experimental

and gated on `options MTUDISC' in the source.  It is also practically
untested becausse (sniff!) I don't have easy access to a network with
an MTU of less than an Ethernet.  If you have a small MTU network,
please try it and tell me if it works!

3 files changed, 108 insertions, 12 deletions

diff --git a/sys/netinet/tcp_output.c b/sys/netinet/tcp_output.c
index 6c7ccd1..f5500ee 100644
--- a/sys/netinet/tcp_output.c
+++ b/sys/netinet/tcp_output.c
@@ -31,7 +31,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)tcp_output.c	8.3 (Berkeley) 12/30/93
- * $Id: tcp_output.c,v 1.11 1995/05/30 08:09:56 rgrimes Exp $
+ * $Id: tcp_output.c,v 1.12 1995/09/13 17:36:31 wollman Exp $
  */
 
 #include <sys/param.h>
@@ -661,16 +661,28 @@ send:
 	else
 #endif
     {
+#ifdef MTUDISC
+	struct rtentry *rt;
+#endif
 	((struct ip *)ti)->ip_len = m->m_pkthdr.len;
 	((struct ip *)ti)->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl;	/* XXX */
 	((struct ip *)ti)->ip_tos = tp->t_inpcb->inp_ip.ip_tos;	/* XXX */
-#if BSD >= 43
+#ifdef MTUDISC
+	/*
+	 * See if we should do MTU discovery.  We do it only if the following
+	 * are true:
+	 *	1) we have a valid route to the destination
+	 *	2) the MTU is not locked (if it is, then discovery has been
+	 *	   disabled)
+	 */
+	if ((rt = tp->t_inpcb->inp_route.ro_rt)
+	    && rt->rt_flags & RTF_UP
+	    && !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
+		((struct ip *)ti)->ip_off |= IP_DF;
+	}
+#endif /* MTUDISC */
 	error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
 	    so->so_options & SO_DONTROUTE, 0);
-#else
-	error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route,
-	    so->so_options & SO_DONTROUTE);
-#endif
     }
 	if (error) {
 out:
diff --git a/sys/netinet/tcp_subr.c b/sys/netinet/tcp_subr.c
index be84884..feec90b 100644
--- a/sys/netinet/tcp_subr.c
+++ b/sys/netinet/tcp_subr.c
@@ -479,6 +479,7 @@ tcp_quench(inp, errno)
  * When `need fragmentation' ICMP is received, update our idea of the MSS
  * based on the new value in the route.  Also nudge TCP to send something,
  * since we know the packet we just sent was dropped.
+ * This duplicates some code in the tcp_mss() function in tcp_input.c.
  */
 static void
 tcp_mtudisc(inp, errno)
@@ -486,11 +487,52 @@ tcp_mtudisc(inp, errno)
 	int errno;
 {
 	struct tcpcb *tp = intotcpcb(inp);
+	struct rtentry *rt;
+	struct rmxp_tao *taop;
+	struct socket *so = inp->inp_socket;
+	int offered;
+	int mss;
 
-#if 0
-	if (tp)
-		;		/* XXX implement */
+	if (tp) {
+		rt = tcp_rtlookup(inp);
+		if (!rt || !rt->rt_rmx.rmx_mtu) {
+			tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
+			return;
+		}
+		taop = rmx_taop(rt->rt_rmx);
+		offered = taop->tao_mssopt;
+		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
+		mss = min(mss, offer);
+		tp->t_maxopd = mss;
+
+		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
+		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
+			mss -= TCPOLEN_TSTAMP_APPA;
+		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
+		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
+			mss -= TCPOLEN_CC_APPA;
+#if	(MCLBYTES & (MCLBYTES - 1)) == 0
+		if (mss > MCLBYTES)
+			mss &= ~(MCLBYTES-1);
+#else
+		if (mss > MCLBYTES)
+			mss = mss / MCLBYTES * MCLBYTES;
 #endif
+		if (so->so_snd.sb_hiwat < mss)
+			mss = so->so_snd.sb_hiwat;
+
+		tp->t_maxseg = mss;
+
+		/*
+		 * Nudge TCP output.  Unfortunately, we have no way to know
+		 * which packet that we sent is the failing one, but in the
+		 * vast majority of cases we expect that it will be at the
+		 * beginning of the window, so this should do the right
+		 * thing (I hope).
+		 */
+		tp->snd_nxt = tp->snd_una;
+		tcp_output(tp);
+	}
 }
 #endif /* MTUDISC */
 
diff --git a/sys/netinet/tcp_timewait.c b/sys/netinet/tcp_timewait.c
index be84884..feec90b 100644
--- a/sys/netinet/tcp_timewait.c
+++ b/sys/netinet/tcp_timewait.c
@@ -479,6 +479,7 @@ tcp_quench(inp, errno)
  * When `need fragmentation' ICMP is received, update our idea of the MSS
  * based on the new value in the route.  Also nudge TCP to send something,
  * since we know the packet we just sent was dropped.
+ * This duplicates some code in the tcp_mss() function in tcp_input.c.
  */
 static void
 tcp_mtudisc(inp, errno)
@@ -486,11 +487,52 @@ tcp_mtudisc(inp, errno)
 	int errno;
 {
 	struct tcpcb *tp = intotcpcb(inp);
+	struct rtentry *rt;
+	struct rmxp_tao *taop;
+	struct socket *so = inp->inp_socket;
+	int offered;
+	int mss;
 
-#if 0
-	if (tp)
-		;		/* XXX implement */
+	if (tp) {
+		rt = tcp_rtlookup(inp);
+		if (!rt || !rt->rt_rmx.rmx_mtu) {
+			tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
+			return;
+		}
+		taop = rmx_taop(rt->rt_rmx);
+		offered = taop->tao_mssopt;
+		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
+		mss = min(mss, offer);
+		tp->t_maxopd = mss;
+
+		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
+		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
+			mss -= TCPOLEN_TSTAMP_APPA;
+		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
+		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
+			mss -= TCPOLEN_CC_APPA;
+#if	(MCLBYTES & (MCLBYTES - 1)) == 0
+		if (mss > MCLBYTES)
+			mss &= ~(MCLBYTES-1);
+#else
+		if (mss > MCLBYTES)
+			mss = mss / MCLBYTES * MCLBYTES;
 #endif
+		if (so->so_snd.sb_hiwat < mss)
+			mss = so->so_snd.sb_hiwat;
+
+		tp->t_maxseg = mss;
+
+		/*
+		 * Nudge TCP output.  Unfortunately, we have no way to know
+		 * which packet that we sent is the failing one, but in the
+		 * vast majority of cases we expect that it will be at the
+		 * beginning of the window, so this should do the right
+		 * thing (I hope).
+		 */
+		tp->snd_nxt = tp->snd_una;
+		tcp_output(tp);
+	}
 }
 #endif /* MTUDISC */