turn off path mtu when icmp needfrag messages get blocked, okay itojun@

3 files changed, 74 insertions, 25 deletions

diff --git a/sys/netinet/ip_icmp.c b/sys/netinet/ip_icmp.c
index bd317ee070d..6946c82df2c 100644
--- a/sys/netinet/ip_icmp.c
+++ b/sys/netinet/ip_icmp.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_icmp.c,v 1.30 2000/10/17 22:33:51 provos Exp $	*/
+/*	$OpenBSD: ip_icmp.c,v 1.31 2000/12/11 19:12:22 provos Exp $	*/
 /*	$NetBSD: ip_icmp.c,v 1.19 1996/02/13 23:42:22 christos Exp $	*/
 
 /*
@@ -747,23 +747,15 @@ icmp_sysctl(name, namelen, oldp, oldlenp, newp, newlen)
 	/* NOTREACHED */
 }
 
-void
-icmp_mtudisc(icp)
-	struct icmp *icp;
+struct rtentry *
+icmp_mtudisc_clone(struct sockaddr *dst)
 {
 	struct rtentry *rt;
-	struct sockaddr *dst = sintosa(&icmpsrc);
-	u_long mtu = ntohs(icp->icmp_nextmtu);  /* Why a long?  IPv6 */
-	int    error;
-	
-	/* Table of common MTUs: */
+	int error;
 
-	static u_short mtu_table[] = {65535, 65280, 32000, 17914, 9180, 8166, 
-				      4352, 2002, 1492, 1006, 508, 296, 68, 0};
-    
 	rt = rtalloc1(dst, 1);
 	if (rt == 0)
-		return;
+		return (NULL);
     
 	/* If we didn't get a host route, allocate one */
     
@@ -777,7 +769,7 @@ icmp_mtudisc(icp)
 		if (error) {
 			rtfree(rt);
 			rtfree(nrt);
-			return;
+			return (NULL);
 		}
 		nrt->rt_rmx = rt->rt_rmx;
 		rtfree(rt);
@@ -786,9 +778,29 @@ icmp_mtudisc(icp)
 	error = rt_timer_add(rt, icmp_mtudisc_timeout, ip_mtudisc_timeout_q);
 	if (error) {
 		rtfree(rt);
-		return;
+		return (NULL);
 	}
 
+	return (rt);
+}
+
+void
+icmp_mtudisc(icp)
+	struct icmp *icp;
+{
+	struct rtentry *rt;
+	struct sockaddr *dst = sintosa(&icmpsrc);
+	u_long mtu = ntohs(icp->icmp_nextmtu);  /* Why a long?  IPv6 */
+	
+	/* Table of common MTUs: */
+
+	static u_short mtu_table[] = {65535, 65280, 32000, 17914, 9180, 8166, 
+				      4352, 2002, 1492, 1006, 508, 296, 68, 0};
+
+	rt = icmp_mtudisc_clone(dst);
+	if (rt == 0)
+		return;
+
 	if (mtu == 0) {
 		int i = 0;
 
@@ -831,8 +843,7 @@ icmp_mtudisc(icp)
 			rt->rt_rmx.rmx_mtu = mtu;
 	}
 
-	if (rt)
-		rtfree(rt);
+	rtfree(rt);
 }
 
 void
diff --git a/sys/netinet/ip_icmp.h b/sys/netinet/ip_icmp.h
index 387e3b1b2c0..11436f34dca 100644
--- a/sys/netinet/ip_icmp.h
+++ b/sys/netinet/ip_icmp.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_icmp.h,v 1.10 2000/10/10 15:16:02 provos Exp $	*/
+/*	$OpenBSD: ip_icmp.h,v 1.11 2000/12/11 19:12:22 provos Exp $	*/
 /*	$NetBSD: ip_icmp.h,v 1.10 1996/02/13 23:42:28 christos Exp $	*/
 
 /*
@@ -186,5 +186,7 @@ void	icmp_input __P((struct mbuf *, ...));
 void	icmp_reflect __P((struct mbuf *));
 void	icmp_send __P((struct mbuf *, struct mbuf *));
 int	icmp_sysctl __P((int *, u_int, void *, size_t *, void *, size_t));
+struct rtentry *
+	icmp_mtudisc_clone(struct sockaddr *);
 void	icmp_mtudisc __P((struct icmp *));
 #endif
diff --git a/sys/netinet/tcp_timer.c b/sys/netinet/tcp_timer.c
index 4f7b1726439..c1e08b91f27 100644
--- a/sys/netinet/tcp_timer.c
+++ b/sys/netinet/tcp_timer.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_timer.c,v 1.17 2000/09/18 22:06:38 provos Exp $	*/
+/*	$OpenBSD: tcp_timer.c,v 1.18 2000/12/11 19:12:22 provos Exp $	*/
 /*	$NetBSD: tcp_timer.c,v 1.14 1996/02/13 23:44:09 christos Exp $	*/
 
 /*
@@ -60,6 +60,7 @@
 #include <netinet/tcp_timer.h>
 #include <netinet/tcp_var.h>
 #include <netinet/tcpip.h>
+#include <netinet/ip_icmp.h>
 #include <dev/rndvar.h>
 
 int	tcp_keepidle = TCPTV_KEEP_IDLE;
@@ -236,7 +237,7 @@ tcp_timers(tp, timer)
 		    rto * tcp_backoff[tp->t_rxtshift],
 		    tp->t_rttmin, TCPTV_REXMTMAX);
 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
-#if 0
+
 		/* 
 		 * If we are losing and we are trying path MTU discovery,
 		 * try turning it off.  This will avoid black holes in
@@ -245,15 +246,50 @@ tcp_timers(tp, timer)
 		 * lots more sophisticated searching to find the right
 		 * value here...
 		 */
-		if (ip_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
+		if (ip_mtudisc && tp->t_inpcb &&
+		    tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
+			struct inpcb *inp = tp->t_inpcb;
 			struct rtentry *rt = NULL;
+			struct sockaddr_in sin;
 
-			if (tp->t_inpcb)
-				rt = in_pcbrtentry(tp->t_inpcb);
 
-			/* XXX:  Black hole recovery code goes here */
-		}
+			rt = in_pcbrtentry(inp);
+			/* Check if path MTU discovery is disabled already */
+			if (rt && (rt->rt_flags & RTF_HOST) &&
+			    (rt->rt_rmx.rmx_locks & RTV_MTU))
+				goto out;
+
+			rt = NULL;
+			switch(tp->pf) {
+#ifdef INET6
+			case PF_INET6:
+				/*
+				 * We can not turn off path MTU for IPv6.
+				 * Do nothing for now, maybe lower to
+				 * minimum MTU.
+				 */
+				break;
 #endif
+			case PF_INET:
+				bzero(&sin, sizeof(struct sockaddr_in));
+				sin.sin_family = AF_INET;
+				sin.sin_len = sizeof(struct sockaddr_in);
+				sin.sin_addr = inp->inp_faddr;
+				rt = icmp_mtudisc_clone(sintosa(&sin));
+				break;
+			}
+			if (rt != NULL) {
+				/* Disable path MTU discovery */
+				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0) {
+					rt->rt_rmx.rmx_locks |= RTV_MTU;
+					in_rtchange(inp, 0);
+				}
+
+				rtfree(rt);
+			}
+			out:
+		}
+
 		/*
 		 * If losing, let the lower level know and try for
 		 * a better route.  Also, if we backed off this far,