add ECN support in layer-3.

 - implement the tunnel egress rule in ip_ecn_egress() in ip_ecn.c.
   make ip{,6}_ecn_egress() return integer to tell the caller that
   this packet should be dropped.
 - handle ECN at fragment reassembly in ip_input.c and frag6.c.

Obtained from:	KAME

10 files changed, 186 insertions, 54 deletions

diff --git a/sys/netinet/in_gif.c b/sys/netinet/in_gif.c
index 3d2ff66..fd6397e 100644
--- a/sys/netinet/in_gif.c
+++ b/sys/netinet/in_gif.c
@@ -161,10 +161,8 @@ in_gif_output(ifp, family, m)
 	/* version will be set in ip_output() */
 	iphdr.ip_ttl = ip_gif_ttl;
 	iphdr.ip_len = m->m_pkthdr.len + sizeof(struct ip);
-	if (ifp->if_flags & IFF_LINK1)
-		ip_ecn_ingress(ECN_ALLOWED, &iphdr.ip_tos, &tos);
-	else
-		ip_ecn_ingress(ECN_NOCARE, &iphdr.ip_tos, &tos);
+	ip_ecn_ingress((ifp->if_flags & IFF_LINK1) ? ECN_ALLOWED : ECN_NOCARE,
+		       &iphdr.ip_tos, &tos);
 
 	/* prepend new IP header */
 	M_PREPEND(m, sizeof(struct ip), M_DONTWAIT);
@@ -250,10 +248,12 @@ in_gif_input(m, off)
 				return;
 		}
 		ip = mtod(m, struct ip *);
-		if (gifp->if_flags & IFF_LINK1)
-			ip_ecn_egress(ECN_ALLOWED, &otos, &ip->ip_tos);
-		else
-			ip_ecn_egress(ECN_NOCARE, &otos, &ip->ip_tos);
+		if (ip_ecn_egress((gifp->if_flags & IFF_LINK1) ?
+				  ECN_ALLOWED : ECN_NOCARE,
+				  &otos, &ip->ip_tos) == 0) {
+			m_freem(m);
+			return;
+		}
 		break;
 	    }
 #endif
@@ -261,7 +261,8 @@ in_gif_input(m, off)
 	case IPPROTO_IPV6:
 	    {
 		struct ip6_hdr *ip6;
-		u_int8_t itos;
+		u_int8_t itos, oitos;
+
 		af = AF_INET6;
 		if (m->m_len < sizeof(*ip6)) {
 			m = m_pullup(m, sizeof(*ip6));
@@ -269,13 +270,17 @@ in_gif_input(m, off)
 				return;
 		}
 		ip6 = mtod(m, struct ip6_hdr *);
-		itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
-		if (gifp->if_flags & IFF_LINK1)
-			ip_ecn_egress(ECN_ALLOWED, &otos, &itos);
-		else
-			ip_ecn_egress(ECN_NOCARE, &otos, &itos);
-		ip6->ip6_flow &= ~htonl(0xff << 20);
-		ip6->ip6_flow |= htonl((u_int32_t)itos << 20);
+		itos = oitos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
+		if (ip_ecn_egress((gifp->if_flags & IFF_LINK1) ?
+				  ECN_ALLOWED : ECN_NOCARE,
+				  &otos, &itos) == 0) {
+			m_freem(m);
+			return;
+		}
+		if (itos != oitos) {
+			ip6->ip6_flow &= ~htonl(0xff << 20);
+			ip6->ip6_flow |= htonl((u_int32_t)itos << 20);
+		}
 		break;
 	    }
 #endif /* INET6 */
diff --git a/sys/netinet/ip.h b/sys/netinet/ip.h
index 24f66ce..025ad08 100644
--- a/sys/netinet/ip.h
+++ b/sys/netinet/ip.h
@@ -84,10 +84,11 @@ CTASSERT(sizeof (struct ip) == 20);
 #define	IPTOS_THROUGHPUT	0x08
 #define	IPTOS_RELIABILITY	0x04
 #define	IPTOS_MINCOST		0x02
-/* ECN bits proposed by Sally Floyd */
-#define	IPTOS_CE		0x01	/* congestion experienced */
-#define	IPTOS_ECT		0x02	/* ECN-capable transport */
-
+#if 1
+/* ECN RFC3168 obsoletes RFC2481, and these will be deprecated soon. */
+#define	IPTOS_CE		0x01
+#define	IPTOS_ECT		0x02
+#endif
 
 /*
  * Definitions for IP precedence (also in ip_tos) (hopefully unused)
@@ -102,6 +103,16 @@ CTASSERT(sizeof (struct ip) == 20);
 #define	IPTOS_PREC_ROUTINE		0x00
 
 /*
+ * ECN (Explicit Congestion Notification) codepoints in RFC3168
+ * mapped to the lower 2 bits of the TOS field.
+ */
+#define	IPTOS_ECN_NOTECT	0x00	/* not-ECT */
+#define	IPTOS_ECN_ECT1		0x01	/* ECN-capable transport (1) */
+#define	IPTOS_ECN_ECT0		0x02	/* ECN-capable transport (0) */
+#define	IPTOS_ECN_CE		0x03	/* congestion experienced */
+#define	IPTOS_ECN_MASK		0x03	/* ECN field mask */
+
+/*
  * Definitions for options.
  */
 #define	IPOPT_COPIED(o)		((o)&0x80)
diff --git a/sys/netinet/ip_ecn.c b/sys/netinet/ip_ecn.c
index de3d38e..9ea2f6b 100644
--- a/sys/netinet/ip_ecn.c
+++ b/sys/netinet/ip_ecn.c
@@ -1,5 +1,5 @@
 /*	$FreeBSD$	*/
-/*	$KAME: ip_ecn.c,v 1.11 2001/05/03 16:09:29 itojun Exp $	*/
+/*	$KAME: ip_ecn.c,v 1.12 2002/01/07 11:34:47 kjc Exp $	*/
 
 /*
  * Copyright (C) 1999 WIDE Project.
@@ -56,6 +56,37 @@
 #endif
 
 /*
+ * ECN and TOS (or TCLASS) processing rules at tunnel encapsulation and
+ * decapsulation from RFC3168:
+ *
+ *                      Outer Hdr at                 Inner Hdr at
+ *                      Encapsulator                 Decapsulator
+ *   Header fields:     --------------------         ------------
+ *     DS Field         copied from inner hdr        no change
+ *     ECN Field        constructed by (I)           constructed by (E)
+ *
+ * ECN_ALLOWED (full functionality):
+ *    (I) if the ECN field in the inner header is set to CE, then set the
+ *    ECN field in the outer header to ECT(0).
+ *    otherwise, copy the ECN field to the outer header.
+ *
+ *    (E) if the ECN field in the outer header is set to CE and the ECN
+ *    field of the inner header is not-ECT, drop the packet.
+ *    if the ECN field in the inner header is set to ECT(0) or ECT(1)
+ *    and the ECN field in the outer header is set to CE, then copy CE to
+ *    the inner header.  otherwise, make no change to the inner header.
+ *
+ * ECN_FORBIDDEN (limited functionality):
+ *    (I) set the ECN field to not-ECT in the outer header.
+ *
+ *    (E) if the ECN field in the outer header is set to CE, drop the packet.
+ *    otherwise, make no change to the ECN field in the inner header.
+ *
+ * the drop rule is for backward compatibility and protection against
+ * erasure of CE.
+ */
+
+/*
  * modify outer ECN (TOS) field on ingress operation (tunnel encapsulation).
  */
 void
@@ -70,10 +101,18 @@ ip_ecn_ingress(mode, outer, inner)
 	*outer = *inner;
 	switch (mode) {
 	case ECN_ALLOWED:		/* ECN allowed */
-		*outer &= ~IPTOS_CE;
+		/*
+		 * full-functionality: if the inner is CE, set ECT(0)
+		 * to the outer.  otherwise, copy the ECN field.
+		 */
+		if ((*inner & IPTOS_ECN_MASK) == IPTOS_ECN_CE)
+			*outer &= ~IPTOS_ECN_ECT1;
 		break;
 	case ECN_FORBIDDEN:		/* ECN forbidden */
-		*outer &= ~(IPTOS_ECT | IPTOS_CE);
+		/*
+		 * limited-functionality: set not-ECT to the outer
+		 */
+		*outer &= ~IPTOS_ECN_MASK;
 		break;
 	case ECN_NOCARE:	/* no consideration to ECN */
 		break;
@@ -82,8 +121,9 @@ ip_ecn_ingress(mode, outer, inner)
 
 /*
  * modify inner ECN (TOS) field on egress operation (tunnel decapsulation).
+ * the caller should drop the packet if the return value is 0.
  */
-void
+int
 ip_ecn_egress(mode, outer, inner)
 	int mode;
 	const u_int8_t *outer;
@@ -94,13 +134,28 @@ ip_ecn_egress(mode, outer, inner)
 
 	switch (mode) {
 	case ECN_ALLOWED:
-		if (*outer & IPTOS_CE)
-			*inner |= IPTOS_CE;
+		/*
+		 * full-functionality: if the outer is CE and the inner is
+		 * not-ECT, should drop it.  otherwise, copy CE.
+		 */
+		if ((*outer & IPTOS_ECN_MASK) == IPTOS_ECN_CE) {
+			if ((*inner & IPTOS_ECN_MASK) == IPTOS_ECN_NOTECT)
+				return (0);
+			*inner |= IPTOS_ECN_CE;
+		}
 		break;
 	case ECN_FORBIDDEN:		/* ECN forbidden */
+		/*
+		 * limited-functionality: if the outer is CE, should drop it.
+		 * otherwise, leave the inner.
+		 */
+		if ((*outer & IPTOS_ECN_MASK) == IPTOS_ECN_CE)
+			return (0);
+		break;
 	case ECN_NOCARE:	/* no consideration to ECN */
 		break;
 	}
+	return (1);
 }
 
 #ifdef INET6
@@ -115,28 +170,31 @@ ip6_ecn_ingress(mode, outer, inner)
 	if (!outer || !inner)
 		panic("NULL pointer passed to ip6_ecn_ingress");
 
-	outer8 = (ntohl(*outer) >> 20) & 0xff;
 	inner8 = (ntohl(*inner) >> 20) & 0xff;
 	ip_ecn_ingress(mode, &outer8, &inner8);
 	*outer &= ~htonl(0xff << 20);
 	*outer |= htonl((u_int32_t)outer8 << 20);
 }
 
-void
+int
 ip6_ecn_egress(mode, outer, inner)
 	int mode;
 	const u_int32_t *outer;
 	u_int32_t *inner;
 {
-	u_int8_t outer8, inner8;
+	u_int8_t outer8, inner8, oinner8;
 
 	if (!outer || !inner)
 		panic("NULL pointer passed to ip6_ecn_egress");
 
 	outer8 = (ntohl(*outer) >> 20) & 0xff;
-	inner8 = (ntohl(*inner) >> 20) & 0xff;
-	ip_ecn_egress(mode, &outer8, &inner8);
-	*inner &= ~htonl(0xff << 20);
-	*inner |= htonl((u_int32_t)inner8 << 20);
+	inner8 = oinner8 = (ntohl(*inner) >> 20) & 0xff;
+	if (ip_ecn_egress(mode, &outer8, &inner8) == 0)
+		return (0);
+	if (inner8 != oinner8) {
+		*inner &= ~htonl(0xff << 20);
+		*inner |= htonl((u_int32_t)inner8 << 20);
+	}
+	return (1);
 }
 #endif
diff --git a/sys/netinet/ip_ecn.h b/sys/netinet/ip_ecn.h
index 1a38a48..01163ed 100644
--- a/sys/netinet/ip_ecn.h
+++ b/sys/netinet/ip_ecn.h
@@ -1,5 +1,5 @@
 /*	$FreeBSD$	*/
-/*	$KAME: ip_ecn.h,v 1.6 2001/05/03 14:51:48 itojun Exp $	*/
+/*	$KAME: ip_ecn.h,v 1.8 2002/01/07 11:34:47 kjc Exp $	*/
 
 /*
  * Copyright (C) 1999 WIDE Project.
@@ -35,6 +35,9 @@
  * http://www.aciri.org/floyd/papers/draft-ipsec-ecn-00.txt
  */
 
+#ifndef _NETINET_IP_ECN_H_
+#define _NETINET_IP_ECN_H_
+
 #if defined(_KERNEL) && !defined(_LKM)
 #include "opt_inet.h"
 #endif
@@ -45,5 +48,6 @@
 
 #ifdef _KERNEL
 extern void ip_ecn_ingress(int, u_int8_t *, const u_int8_t *);
-extern void ip_ecn_egress(int, const u_int8_t *, u_int8_t *);
+extern int ip_ecn_egress(int, const u_int8_t *, u_int8_t *);
+#endif
 #endif
diff --git a/sys/netinet/ip_input.c b/sys/netinet/ip_input.c
index 02bd132..30a5b75 100644
--- a/sys/netinet/ip_input.c
+++ b/sys/netinet/ip_input.c
@@ -1037,6 +1037,7 @@ ip_reass(struct mbuf *m, struct ipqhead *head, struct ipq *fp,
 	struct mbuf *t;
 	int hlen = ip->ip_hl << 2;
 	int i, next;
+	u_int8_t ecn, ecn0;
 
 	IPQ_LOCK_ASSERT();
 
@@ -1086,6 +1087,22 @@ ip_reass(struct mbuf *m, struct ipqhead *head, struct ipq *fp,
 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
 
 	/*
+	 * Handle ECN by comparing this segment with the first one;
+	 * if CE is set, do not lose CE.
+	 * drop if CE and not-ECT are mixed for the same packet.
+	 */
+	ecn = ip->ip_tos & IPTOS_ECN_MASK;
+	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
+	if (ecn == IPTOS_ECN_CE) {
+		if (ecn0 == IPTOS_ECN_NOTECT)
+			goto dropfrag;
+		if (ecn0 != IPTOS_ECN_CE)
+			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
+	}
+	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
+		goto dropfrag;
+
+	/*
 	 * Find a segment which begins after this one does.
 	 */
 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
diff --git a/sys/netinet6/ah_input.c b/sys/netinet6/ah_input.c
index 6fb8071..78076b6 100644
--- a/sys/netinet6/ah_input.c
+++ b/sys/netinet6/ah_input.c
@@ -405,7 +405,10 @@ ah4_input(m, off)
 		}
 		ip = mtod(m, struct ip *);
 		/* ECN consideration. */
-		ip_ecn_egress(ip4_ipsec_ecn, &tos, &ip->ip_tos);
+		if (!ip_ecn_egress(ip4_ipsec_ecn, &tos, &ip->ip_tos)) {
+			ipsecstat.in_inval++;
+			goto fail;
+		}
 		if (!key_checktunnelsanity(sav, AF_INET,
 			    (caddr_t)&ip->ip_src, (caddr_t)&ip->ip_dst)) {
 			ipseclog((LOG_NOTICE, "ipsec tunnel address mismatch "
@@ -812,7 +815,10 @@ ah6_input(mp, offp, proto)
 		}
 		ip6 = mtod(m, struct ip6_hdr *);
 		/* ECN consideration. */
-		ip6_ecn_egress(ip6_ipsec_ecn, &flowinfo, &ip6->ip6_flow);
+		if (!ip6_ecn_egress(ip6_ipsec_ecn, &flowinfo, &ip6->ip6_flow)) {
+			ipsec6stat.in_inval++;
+			goto fail;
+		}
 		if (!key_checktunnelsanity(sav, AF_INET6,
 			    (caddr_t)&ip6->ip6_src, (caddr_t)&ip6->ip6_dst)) {
 			ipseclog((LOG_NOTICE, "ipsec tunnel address mismatch "
diff --git a/sys/netinet6/esp_input.c b/sys/netinet6/esp_input.c
index f2d802c..f25d0f3 100644
--- a/sys/netinet6/esp_input.c
+++ b/sys/netinet6/esp_input.c
@@ -371,7 +371,10 @@ noreplaycheck:
 		}
 		ip = mtod(m, struct ip *);
 		/* ECN consideration. */
-		ip_ecn_egress(ip4_ipsec_ecn, &tos, &ip->ip_tos);
+		if (!ip_ecn_egress(ip4_ipsec_ecn, &tos, &ip->ip_tos)) {
+			ipsecstat.in_inval++;
+			goto bad;
+		}
 		if (!key_checktunnelsanity(sav, AF_INET,
 			    (caddr_t)&ip->ip_src, (caddr_t)&ip->ip_dst)) {
 			ipseclog((LOG_ERR, "ipsec tunnel address mismatch "
@@ -723,7 +726,10 @@ noreplaycheck:
 		}
 		ip6 = mtod(m, struct ip6_hdr *);
 		/* ECN consideration. */
-		ip6_ecn_egress(ip6_ipsec_ecn, &flowinfo, &ip6->ip6_flow);
+		if (!ip6_ecn_egress(ip6_ipsec_ecn, &flowinfo, &ip6->ip6_flow)) {
+			ipsec6stat.in_inval++;
+			goto bad;
+		}
 		if (!key_checktunnelsanity(sav, AF_INET6,
 			    (caddr_t)&ip6->ip6_src, (caddr_t)&ip6->ip6_dst)) {
 			ipseclog((LOG_ERR, "ipsec tunnel address mismatch "
diff --git a/sys/netinet6/frag6.c b/sys/netinet6/frag6.c
index 3bb425f..f8a86a1 100644
--- a/sys/netinet6/frag6.c
+++ b/sys/netinet6/frag6.c
@@ -52,6 +52,8 @@
 #include <netinet/ip6.h>
 #include <netinet6/ip6_var.h>
 #include <netinet/icmp6.h>
+#include <netinet/in_systm.h>	/* for ECN definitions */
+#include <netinet/ip.h>		/* for ECN definitions */
 
 #include <net/net_osdep.h>
 
@@ -151,6 +153,7 @@ frag6_input(mp, offp, proto)
 	int first_frag = 0;
 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
 	struct ifnet *dstifp;
+	u_int8_t ecn, ecn0;
 
 	ip6 = mtod(m, struct ip6_hdr *);
 #ifndef PULLDOWN_TEST
@@ -350,6 +353,26 @@ frag6_input(mp, offp, proto)
 	}
 
 	/*
+	 * Handle ECN by comparing this segment with the first one;
+	 * if CE is set, do not lose CE.
+	 * drop if CE and not-ECT are mixed for the same packet.
+	 */
+	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
+	ecn0 = (ntohl(q6->ip6q_down->ip6af_head) >> 20) & IPTOS_ECN_MASK;
+	if (ecn == IPTOS_ECN_CE) {
+		if (ecn0 == IPTOS_ECN_NOTECT) {
+			free(ip6af, M_FTABLE);
+			goto dropfrag;
+		}
+		if (ecn0 != IPTOS_ECN_CE)
+			q6->ip6q_down->ip6af_head |= htonl(IPTOS_ECN_CE << 20);
+	}
+	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
+		free(ip6af, M_FTABLE);
+		goto dropfrag;
+	}
+
+	/*
 	 * Find a segment which begins after this one does.
 	 */
 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
diff --git a/sys/netinet6/in6_gif.c b/sys/netinet6/in6_gif.c
index fae8a98..e3cd124 100644
--- a/sys/netinet6/in6_gif.c
+++ b/sys/netinet6/in6_gif.c
@@ -168,11 +168,9 @@ in6_gif_output(ifp, family, m)
 		m_freem(m);
 		return ENETUNREACH;
 	}
-	if (ifp->if_flags & IFF_LINK1)
-		ip_ecn_ingress(ECN_ALLOWED, &otos, &itos);
-	else
-		ip_ecn_ingress(ECN_NOCARE, &otos, &itos);
-	ip6->ip6_flow &= ~ntohl(0xff00000);
+	ip_ecn_ingress((ifp->if_flags & IFF_LINK1) ? ECN_ALLOWED : ECN_NOCARE,
+		       &otos, &itos);
+	ip6->ip6_flow &= ~htonl(0xff << 20);
 	ip6->ip6_flow |= htonl((u_int32_t)otos << 20);
 
 	if (dst->sin6_family != sin6_dst->sin6_family ||
@@ -259,10 +257,12 @@ in6_gif_input(mp, offp, proto)
 				return IPPROTO_DONE;
 		}
 		ip = mtod(m, struct ip *);
-		if (gifp->if_flags & IFF_LINK1)
-			ip_ecn_egress(ECN_ALLOWED, &otos8, &ip->ip_tos);
-		else
-			ip_ecn_egress(ECN_NOCARE, &otos8, &ip->ip_tos);
+		if (ip_ecn_egress((gifp->if_flags & IFF_LINK1) ?
+				  ECN_ALLOWED : ECN_NOCARE,
+				  &otos8, &ip->ip_tos) == 0) {
+			m_freem(m);
+			return IPPROTO_DONE;
+		}
 		break;
 	    }
 #endif /* INET */
@@ -277,10 +277,12 @@ in6_gif_input(mp, offp, proto)
 				return IPPROTO_DONE;
 		}
 		ip6 = mtod(m, struct ip6_hdr *);
-		if (gifp->if_flags & IFF_LINK1)
-			ip6_ecn_egress(ECN_ALLOWED, &otos, &ip6->ip6_flow);
-		else
-			ip6_ecn_egress(ECN_NOCARE, &otos, &ip6->ip6_flow);
+		if (ip6_ecn_egress((gifp->if_flags & IFF_LINK1) ?
+				   ECN_ALLOWED : ECN_NOCARE,
+				   &otos, &ip6->ip6_flow) == 0) {
+			m_freem(m);
+			return IPPROTO_DONE;
+		}
 		break;
 	    }
 #endif
diff --git a/sys/netinet6/ip6_ecn.h b/sys/netinet6/ip6_ecn.h
index 4107cf0..6644b56 100644
--- a/sys/netinet6/ip6_ecn.h
+++ b/sys/netinet6/ip6_ecn.h
@@ -36,6 +36,6 @@
  */
 
 #ifdef _KERNEL
-extern void ip6_ecn_ingress __P((int, u_int32_t *, const u_int32_t *));
-extern void ip6_ecn_egress __P((int, const u_int32_t *, u_int32_t *));
+extern void ip6_ecn_ingress(int, u_int32_t *, const u_int32_t *);
+extern int ip6_ecn_egress(int, const u_int32_t *, u_int32_t *);
 #endif