]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/netinet6/ip6_forward.c
xnu-3248.40.184.tar.gz
[apple/xnu.git] / bsd / netinet6 / ip6_forward.c
index 95838ed3c46e6adf46b3ebc20e8fddf77032662e..ee5a7045369e7f99e4ad65b3dcc4b5ea926ac305 100644 (file)
@@ -1,4 +1,33 @@
-/*     $KAME: ip6_forward.c,v 1.29 2000/02/26 18:08:38 itojun Exp $    */
+/*
+ * Copyright (c) 2009-2013 Apple Inc. All rights reserved.
+ *
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
+ * 
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ * 
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
+ * 
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
+ */
+
+/*     $FreeBSD: src/sys/netinet6/ip6_forward.c,v 1.16 2002/10/16 02:25:05 sam Exp $   */
+/*     $KAME: ip6_forward.c,v 1.69 2001/05/17 03:48:30 itojun Exp $    */
 
 /*
  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
  * SUCH DAMAGE.
  */
 
-#if (defined(__FreeBSD__) && __FreeBSD__ >= 3)
-#include "opt_ip6fw.h"
-#include "opt_inet.h"
-#endif
 
 #include <sys/param.h>
 #include <sys/systm.h>
 
 #include <netinet/in.h>
 #include <netinet/in_var.h>
+#include <netinet/in_systm.h>
+#include <netinet/ip.h>
 #include <netinet/ip_var.h>
+#include <netinet6/in6_var.h>
 #include <netinet/ip6.h>
 #include <netinet6/ip6_var.h>
 #include <netinet/icmp6.h>
 #include <netinet6/nd6.h>
+#include <netinet6/scope6_var.h>
+
+#include <netinet/in_pcb.h>
 
-#if IPSEC_IPV6FWD
+#if IPSEC
 #include <netinet6/ipsec.h>
+#if INET6
+#include <netinet6/ipsec6.h>
+#endif
 #include <netkey/key.h>
-#include <netkey/key_debug.h>
-#endif /* IPSEC_IPV6FWD */
+extern int ipsec_bypass;
+#endif /* IPSEC */
 
-#if IPV6FIREWALL
 #include <netinet6/ip6_fw.h>
-#endif
-
-#if MIP6
-#include <netinet6/mip6.h>
-#endif
 
 #include <net/net_osdep.h>
 
-struct route_in6 ip6_forward_rt;
+#if PF
+#include <net/pfvar.h>
+#endif /* PF */
 
 /*
  * Forward a packet.  If some error occurs return the sender
@@ -88,24 +118,69 @@ struct     route_in6 ip6_forward_rt;
  *
  */
 
-void
-ip6_forward(m, srcrt)
-       struct mbuf *m;
-       int srcrt;
+struct mbuf *
+ip6_forward(struct mbuf *m, struct route_in6 *ip6forward_rt,
+    int srcrt)
 {
        struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
-       register struct sockaddr_in6 *dst;
-       register struct rtentry *rt;
+       struct sockaddr_in6 *dst;
+       struct rtentry *rt;
        int error, type = 0, code = 0;
+       boolean_t proxy = FALSE;
        struct mbuf *mcopy = NULL;
-#if IPSEC_IPV6FWD
+       struct ifnet *ifp, *rcvifp, *origifp;   /* maybe unnecessary */
+       u_int32_t inzone, outzone, len;
+       struct in6_addr src_in6, dst_in6;
+       uint64_t curtime = net_uptime();
+#if IPSEC
        struct secpolicy *sp = NULL;
 #endif
-#if !(defined(__FreeBSD__) && __FreeBSD__ >= 3) && !defined(__APPLE__)
-       long time_second = time.tv_sec;
-#endif
+       unsigned int ifscope = IFSCOPE_NONE;
+#if PF
+       struct pf_mtag *pf_mtag;
+#endif /* PF */
+
+       /*
+        * In the prefix proxying case, the route to the proxied node normally
+        * gets created by nd6_prproxy_ns_output(), as part of forwarding a
+        * NS (NUD/AR) packet to the proxied node.  In the event that such
+        * packet did not arrive in time before the correct route gets created,
+        * ip6_input() would have performed a rtalloc() which most likely will
+        * create the wrong cloned route; this route points back to the same
+        * interface as the inbound interface, since the parent non-scoped
+        * prefix route points there.  Therefore we check if that is the case
+        * and perform the necessary fixup to get the correct route installed.
+        */
+       if (!srcrt && nd6_prproxy &&
+           (rt = ip6forward_rt->ro_rt) != NULL && (rt->rt_flags & RTF_PROXY)) {
+               nd6_proxy_find_fwdroute(m->m_pkthdr.rcvif, ip6forward_rt);
+               if ((rt = ip6forward_rt->ro_rt) != NULL)
+                       ifscope = rt->rt_ifp->if_index;
+       }
 
-#if IPSEC_IPV6FWD
+#if PF
+       pf_mtag = pf_find_mtag(m);
+       if (pf_mtag != NULL && pf_mtag->pftag_rtableid != IFSCOPE_NONE)
+               ifscope = pf_mtag->pftag_rtableid;
+
+       /*
+        * If the caller provides a route which is on a different interface
+        * than the one specified for scoped forwarding, discard the route
+        * and do a lookup below.
+        */
+       if (ifscope != IFSCOPE_NONE && (rt = ip6forward_rt->ro_rt) != NULL) {
+               RT_LOCK(rt);
+               if (rt->rt_ifp->if_index != ifscope) {
+                       RT_UNLOCK(rt);
+                       ROUTE_RELEASE(ip6forward_rt);
+                       rt = NULL;
+               } else {
+                       RT_UNLOCK(rt);
+               }
+       }
+#endif /* PF */
+
+#if IPSEC
        /*
         * Check AH/ESP integrity.
         */
@@ -113,19 +188,27 @@ ip6_forward(m, srcrt)
         * Don't increment ip6s_cantforward because this is the check
         * before forwarding packet actually.
         */
-       if (ipsec6_in_reject(m, NULL)) {
-               ipsec6stat.in_polvio++;
-               m_freem(m);
-               return;
+       if (ipsec_bypass == 0) {
+               if (ipsec6_in_reject(m, NULL)) {
+                       IPSEC_STAT_INCREMENT(ipsec6stat.in_polvio);
+                       m_freem(m);
+                       return (NULL);
+               }
        }
-#endif /*IPSEC_IPV6FWD*/
+#endif /*IPSEC*/
 
+       /*
+        * Do not forward packets to multicast destination.
+        * Do not forward packets with unspecified source.  It was discussed
+        * in July 2000, on ipngwg mailing list.
+        */
        if ((m->m_flags & (M_BCAST|M_MCAST)) != 0 ||
-           IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
+           IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
+           IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
                ip6stat.ip6s_cantforward++;
                /* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard) */
-               if (ip6_log_time + ip6_log_interval < time_second) {
-                       ip6_log_time = time_second;
+               if (ip6_log_time + ip6_log_interval < curtime) {
+                       ip6_log_time = curtime;
                        log(LOG_DEBUG,
                            "cannot forward "
                            "from %s to %s nxt %d received on %s\n",
@@ -135,15 +218,34 @@ ip6_forward(m, srcrt)
                            if_name(m->m_pkthdr.rcvif));
                }
                m_freem(m);
-               return;
+               return (NULL);
        }
 
        if (ip6->ip6_hlim <= IPV6_HLIMDEC) {
                /* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard) */
-               icmp6_error(m, ICMP6_TIME_EXCEEDED,
-                               ICMP6_TIME_EXCEED_TRANSIT, 0);
-               return;
+               icmp6_error_flag(m, ICMP6_TIME_EXCEEDED,
+                               ICMP6_TIME_EXCEED_TRANSIT, 0, 0);
+               return (NULL);
        }
+
+       /*
+        * See if the destination is a proxied address, and if so pretend
+        * that it's for us.  This is mostly to handle NUD probes against
+        * the proxied addresses.  We filter for ICMPv6 here and will let
+        * icmp6_input handle the rest.
+        */
+       if (!srcrt && nd6_prproxy) {
+               VERIFY(!IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst));
+               proxy = nd6_prproxy_isours(m, ip6, ip6forward_rt, ifscope);
+               /*
+                * Don't update hop limit while proxying; RFC 4389 4.1.
+                * Also skip IPsec forwarding path processing as this
+                * packet is not to be forwarded.
+                */
+               if (proxy)
+                       goto skip_ipsec;
+       }
+
        ip6->ip6_hlim -= IPV6_HLIMDEC;
 
        /*
@@ -157,12 +259,14 @@ ip6_forward(m, srcrt)
         */
        mcopy = m_copy(m, 0, imin(m->m_pkthdr.len, ICMPV6_PLD_MAXLEN));
 
-
-#if IPSEC_IPV6FWD
+#if IPSEC
+       if (ipsec_bypass != 0)
+               goto skip_ipsec;
        /* get a security policy for this packet */
-       sp = ipsec6_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, 0, &error);
+       sp = ipsec6_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, IP_FORWARDING,
+           &error);
        if (sp == NULL) {
-               ipsec6stat.out_inval++;
+               IPSEC_STAT_INCREMENT(ipsec6stat.out_inval);
                ip6stat.ip6s_cantforward++;
                if (mcopy) {
 #if 0
@@ -172,7 +276,7 @@ ip6_forward(m, srcrt)
 #endif
                }
                m_freem(m);
-               return;
+               return (NULL);
        }
 
        error = 0;
@@ -180,12 +284,13 @@ ip6_forward(m, srcrt)
        /* check policy */
        switch (sp->policy) {
        case IPSEC_POLICY_DISCARD:
+        case IPSEC_POLICY_GENERATE:
                /*
                 * This packet is just discarded.
                 */
-               ipsec6stat.out_polvio++;
+               IPSEC_STAT_INCREMENT(ipsec6stat.out_polvio);
                ip6stat.ip6s_cantforward++;
-               key_freesp(sp);
+               key_freesp(sp, KEY_SADB_UNLOCKED);
                if (mcopy) {
 #if 0
                        /* XXX: what icmp ? */
@@ -194,20 +299,20 @@ ip6_forward(m, srcrt)
 #endif
                }
                m_freem(m);
-               return;
+               return (NULL);
 
        case IPSEC_POLICY_BYPASS:
        case IPSEC_POLICY_NONE:
                /* no need to do IPsec. */
-               key_freesp(sp);
+               key_freesp(sp, KEY_SADB_UNLOCKED);
                goto skip_ipsec;
-       
+
        case IPSEC_POLICY_IPSEC:
                if (sp->req == NULL) {
                        /* XXX should be panic ? */
                        printf("ip6_forward: No IPsec request specified.\n");
                        ip6stat.ip6s_cantforward++;
-                       key_freesp(sp);
+                       key_freesp(sp, KEY_SADB_UNLOCKED);
                        if (mcopy) {
 #if 0
                                /* XXX: what icmp ? */
@@ -216,7 +321,7 @@ ip6_forward(m, srcrt)
 #endif
                        }
                        m_freem(m);
-                       return;
+                       return (NULL);
                }
                /* do IPsec */
                break;
@@ -225,7 +330,7 @@ ip6_forward(m, srcrt)
        default:
                /* should be panic ?? */
                printf("ip6_forward: Invalid policy found. %d\n", sp->policy);
-               key_freesp(sp);
+               key_freesp(sp, KEY_SADB_UNLOCKED);
                goto skip_ipsec;
        }
 
@@ -242,17 +347,17 @@ ip6_forward(m, srcrt)
         */
        bzero(&state, sizeof(state));
        state.m = m;
-       state.ro = NULL;        /* update at ipsec6_output_tunnel() */
        state.dst = NULL;       /* update at ipsec6_output_tunnel() */
 
        error = ipsec6_output_tunnel(&state, sp, 0);
+       key_freesp(sp, KEY_SADB_UNLOCKED);
+       if (state.tunneled == 4) {
+               ROUTE_RELEASE(&state.ro);
+               return (NULL);  /* packet is gone - sent over IPv4 */
+       }
 
        m = state.m;
-#if 0  /* XXX allocate a route (ro, dst) again later */
-       ro = (struct route_in6 *)state.ro;
-       dst = (struct sockaddr_in6 *)state.dst;
-#endif
-       key_freesp(sp);
+       ROUTE_RELEASE(&state.ro);
 
        if (error) {
                /* mbuf is already reclaimed in ipsec6_output_tunnel. */
@@ -265,7 +370,7 @@ ip6_forward(m, srcrt)
                        break;
                default:
                        printf("ip6_output (ipsec): error code %d\n", error);
-                       /*fall through*/
+                       /* fall through */
                case ENOENT:
                        /* don't show these error codes to the user */
                        break;
@@ -279,104 +384,118 @@ ip6_forward(m, srcrt)
 #endif
                }
                m_freem(m);
-               return;
+               return (NULL);
        }
     }
+#endif /* IPSEC */
     skip_ipsec:
-#endif /* IPSEC_IPV6FWD */
-       
-#if MIP6
-       {
-               struct   mip6_bc   *bc;
-
-               bc = mip6_bc_find(&ip6->ip6_dst);
-               if ((bc != NULL) && (bc->hr_flag)) {
-                       if (mip6_tunnel_output(&m, bc) != 0) {
-                               ip6stat.ip6s_cantforward++;
-                               if (mcopy)
-                                       m_freem(mcopy);
-                               m_freem(m);
-                               return;
-                       }
-               }
-               ip6 = mtod(m, struct ip6_hdr *);        /* m has changed */
+
+       dst = (struct sockaddr_in6 *)&ip6forward_rt->ro_dst;
+       if ((rt = ip6forward_rt->ro_rt) != NULL) {
+               RT_LOCK(rt);
+               /* Take an extra ref for ourselves */
+               RT_ADDREF_LOCKED(rt);
        }
-#endif
-       
-       dst = &ip6_forward_rt.ro_dst;
+
+       VERIFY(rt == NULL || rt == ip6forward_rt->ro_rt);
        if (!srcrt) {
                /*
-                * ip6_forward_rt.ro_dst.sin6_addr is equal to ip6->ip6_dst
+                * ip6forward_rt->ro_dst.sin6_addr is equal to ip6->ip6_dst
                 */
-               if (ip6_forward_rt.ro_rt == 0 ||
-                   (ip6_forward_rt.ro_rt->rt_flags & RTF_UP) == 0) {
-                       if (ip6_forward_rt.ro_rt) {
-                               RTFREE(ip6_forward_rt.ro_rt);
-                               ip6_forward_rt.ro_rt = 0;
+               if (ROUTE_UNUSABLE(ip6forward_rt)) {
+                       if (rt != NULL) {
+                               /* Release extra ref */
+                               RT_REMREF_LOCKED(rt);
+                               RT_UNLOCK(rt);
                        }
+                       ROUTE_RELEASE(ip6forward_rt);
+
                        /* this probably fails but give it a try again */
-#if __FreeBSD__ || defined(__APPLE__)
-                       rtalloc_ign((struct route *)&ip6_forward_rt,
-                                   RTF_PRCLONING);
-#else
-                       rtalloc((struct route *)&ip6_forward_rt);
-#endif
+                       rtalloc_scoped_ign((struct route *)ip6forward_rt,
+                           RTF_PRCLONING, ifscope);
+                       if ((rt = ip6forward_rt->ro_rt) != NULL) {
+                               RT_LOCK(rt);
+                               /* Take an extra ref for ourselves */
+                               RT_ADDREF_LOCKED(rt);
+                       }
                }
-               
-               if (ip6_forward_rt.ro_rt == 0) {
+
+               if (rt == NULL) {
                        ip6stat.ip6s_noroute++;
-                       /* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_noroute) */
-                       if (mcopy) {
+                       in6_ifstat_inc(m->m_pkthdr.rcvif, ifs6_in_noroute);
+                       if (mcopy)
                                icmp6_error(mcopy, ICMP6_DST_UNREACH,
                                            ICMP6_DST_UNREACH_NOROUTE, 0);
-                       }
                        m_freem(m);
-                       return;
+                       return (NULL);
                }
-       } else if ((rt = ip6_forward_rt.ro_rt) == 0 ||
-                !IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &dst->sin6_addr)) {
-               if (ip6_forward_rt.ro_rt) {
-                       RTFREE(ip6_forward_rt.ro_rt);
-                       ip6_forward_rt.ro_rt = 0;
+               RT_LOCK_ASSERT_HELD(rt);
+       } else if (ROUTE_UNUSABLE(ip6forward_rt) ||
+           !IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &dst->sin6_addr)) {
+               if (rt != NULL) {
+                       /* Release extra ref */
+                       RT_REMREF_LOCKED(rt);
+                       RT_UNLOCK(rt);
                }
+               ROUTE_RELEASE(ip6forward_rt);
+
                bzero(dst, sizeof(*dst));
                dst->sin6_len = sizeof(struct sockaddr_in6);
                dst->sin6_family = AF_INET6;
                dst->sin6_addr = ip6->ip6_dst;
 
-#if __FreeBSD__ || defined(__APPLE__)
-               rtalloc_ign((struct route *)&ip6_forward_rt, RTF_PRCLONING);
-#else
-               rtalloc((struct route *)&ip6_forward_rt);
-#endif
-               if (ip6_forward_rt.ro_rt == 0) {
+               rtalloc_scoped_ign((struct route *)ip6forward_rt,
+                   RTF_PRCLONING, ifscope);
+               if ((rt = ip6forward_rt->ro_rt) == NULL) {
                        ip6stat.ip6s_noroute++;
-                       /* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_noroute) */
-                       if (mcopy) {
+                       in6_ifstat_inc(m->m_pkthdr.rcvif, ifs6_in_noroute);
+                       if (mcopy)
                                icmp6_error(mcopy, ICMP6_DST_UNREACH,
-                                           ICMP6_DST_UNREACH_NOROUTE, 0);
-                       }
+                                   ICMP6_DST_UNREACH_NOROUTE, 0);
                        m_freem(m);
-                       return;
+                       return (NULL);
                }
+               RT_LOCK(rt);
+               /* Take an extra ref for ourselves */
+               RT_ADDREF_LOCKED(rt);
        }
-       rt = ip6_forward_rt.ro_rt;
 
        /*
-        * Scope check: if a packet can't be delivered to its destination
-        * for the reason that the destination is beyond the scope of the
-        * source address, discard the packet and return an icmp6 destination
-        * unreachable error with Code 2 (beyond scope of source address).
-        * [draft-ietf-ipngwg-icmp-v3-00.txt, Section 3.1]
+        * Source scope check: if a packet can't be delivered to its
+        * destination for the reason that the destination is beyond the scope
+        * of the source address, discard the packet and return an icmp6
+        * destination unreachable error with Code 2 (beyond scope of source
+        * address) unless we are proxying (source address is link local
+        * for NUDs.)  We use a local copy of ip6_src, since in6_setscope()
+        * will possibly modify its first argument.
+        * [draft-ietf-ipngwg-icmp-v3-04.txt, Section 3.1]
         */
-       if (in6_addr2scopeid(m->m_pkthdr.rcvif, &ip6->ip6_src) !=
-           in6_addr2scopeid(rt->rt_ifp, &ip6->ip6_src)) {
+       src_in6 = ip6->ip6_src;
+       if (in6_setscope(&src_in6, rt->rt_ifp, &outzone)) {
+               RT_REMREF_LOCKED(rt);
+               RT_UNLOCK(rt);
+               /* XXX: this should not happen */
+               ip6stat.ip6s_cantforward++;
+               ip6stat.ip6s_badscope++;
+               m_freem(m);
+               return (NULL);
+       }
+       if (in6_setscope(&src_in6, m->m_pkthdr.rcvif, &inzone)) {
+               RT_REMREF_LOCKED(rt);
+               RT_UNLOCK(rt);
+               ip6stat.ip6s_cantforward++;
+               ip6stat.ip6s_badscope++;
+               m_freem(m);
+               return (NULL);
+       }
+
+       if (inzone != outzone && !proxy) {
                ip6stat.ip6s_cantforward++;
                ip6stat.ip6s_badscope++;
                in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard);
 
-               if (ip6_log_time + ip6_log_interval < time_second) {
-                       ip6_log_time = time_second;
+               if (ip6_log_time + ip6_log_interval < curtime) {
+                       ip6_log_time = curtime;
                        log(LOG_DEBUG,
                            "cannot forward "
                            "src %s, dst %s, nxt %d, rcvif %s, outif %s\n",
@@ -385,41 +504,64 @@ ip6_forward(m, srcrt)
                            ip6->ip6_nxt,
                            if_name(m->m_pkthdr.rcvif), if_name(rt->rt_ifp));
                }
-               if (mcopy)
+               /* Release extra ref */
+               RT_REMREF_LOCKED(rt);
+               RT_UNLOCK(rt);
+               if (mcopy) {
                        icmp6_error(mcopy, ICMP6_DST_UNREACH,
                                    ICMP6_DST_UNREACH_BEYONDSCOPE, 0);
+               }
+               m_freem(m);
+               return (NULL);
+       }
+
+       /*
+        * Destination scope check: if a packet is going to break the scope
+        * zone of packet's destination address, discard it.  This case should
+        * usually be prevented by appropriately-configured routing table, but
+        * we need an explicit check because we may mistakenly forward the
+        * packet to a different zone by (e.g.) a default route.
+        */
+       dst_in6 = ip6->ip6_dst;
+       if (in6_setscope(&dst_in6, m->m_pkthdr.rcvif, &inzone) != 0 ||
+           in6_setscope(&dst_in6, rt->rt_ifp, &outzone) != 0 ||
+           inzone != outzone) {
+               RT_REMREF_LOCKED(rt);
+               RT_UNLOCK(rt);
+               ip6stat.ip6s_cantforward++;
+               ip6stat.ip6s_badscope++;
                m_freem(m);
-               return;
+               return (NULL);
        }
 
        if (m->m_pkthdr.len > rt->rt_ifp->if_mtu) {
                in6_ifstat_inc(rt->rt_ifp, ifs6_in_toobig);
                if (mcopy) {
-                       u_long mtu;
-#if IPSEC_IPV6FWD
-                       struct secpolicy *sp;
+                       uint32_t mtu;
+#if IPSEC
+                       struct secpolicy *sp2;
                        int ipsecerror;
                        size_t ipsechdrsiz;
 #endif
 
                        mtu = rt->rt_ifp->if_mtu;
-#if IPSEC_IPV6FWD
+#if IPSEC
                        /*
                         * When we do IPsec tunnel ingress, we need to play
-                        * with if_mtu value (decrement IPsec header size
+                        * with the link value (decrement IPsec header size
                         * from mtu value).  The code is much simpler than v4
                         * case, as we have the outgoing interface for
                         * encapsulated packet as "rt->rt_ifp".
                         */
-                       sp = ipsec6_getpolicybyaddr(mcopy, IPSEC_DIR_OUTBOUND,
+                       sp2 = ipsec6_getpolicybyaddr(mcopy, IPSEC_DIR_OUTBOUND,
                                IP_FORWARDING, &ipsecerror);
-                       if (sp) {
+                       if (sp2) {
                                ipsechdrsiz = ipsec6_hdrsiz(mcopy,
                                        IPSEC_DIR_OUTBOUND, NULL);
                                if (ipsechdrsiz < mtu)
                                        mtu -= ipsechdrsiz;
+                               key_freesp(sp2, KEY_SADB_UNLOCKED);
                        }
-
                        /*
                         * if mtu becomes less than minimum MTU,
                         * tell minimum MTU (and I'll need to fragment it).
@@ -427,14 +569,21 @@ ip6_forward(m, srcrt)
                        if (mtu < IPV6_MMTU)
                                mtu = IPV6_MMTU;
 #endif
+                       /* Release extra ref */
+                       RT_REMREF_LOCKED(rt);
+                       RT_UNLOCK(rt);
                        icmp6_error(mcopy, ICMP6_PACKET_TOO_BIG, 0, mtu);
+               } else {
+                       /* Release extra ref */
+                       RT_REMREF_LOCKED(rt);
+                       RT_UNLOCK(rt);
                }
                m_freem(m);
-               return;
+               return (NULL);
        }
 
        if (rt->rt_flags & RTF_GATEWAY)
-               dst = (struct sockaddr_in6 *)rt->rt_gateway;
+               dst = (struct sockaddr_in6 *)(void *)rt->rt_gateway;
 
        /*
         * If we are to forward the packet using the same interface
@@ -445,55 +594,179 @@ ip6_forward(m, srcrt)
         * Also, don't send redirect if forwarding using a route
         * modified by a redirect.
         */
-       if (rt->rt_ifp == m->m_pkthdr.rcvif && !srcrt &&
-           (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0)
+       if (!proxy &&
+           ip6_sendredirects && rt->rt_ifp == m->m_pkthdr.rcvif && !srcrt &&
+           (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0) {
+               if ((rt->rt_ifp->if_flags & IFF_POINTOPOINT) != 0) {
+                       /*
+                        * If the incoming interface is equal to the outgoing
+                        * one, and the link attached to the interface is
+                        * point-to-point, then it will be highly probable
+                        * that a routing loop occurs. Thus, we immediately
+                        * drop the packet and send an ICMPv6 error message.
+                        *
+                        * type/code is based on suggestion by Rich Draves.
+                        * not sure if it is the best pick.
+                        */
+                       RT_REMREF_LOCKED(rt);   /* Release extra ref */
+                       RT_UNLOCK(rt);
+                       icmp6_error(mcopy, ICMP6_DST_UNREACH,
+                                   ICMP6_DST_UNREACH_ADDR, 0);
+                       m_freem(m);
+                       return (NULL);
+               }
                type = ND_REDIRECT;
+       }
 
-#if IPV6FIREWALL
+#if IPFW2
        /*
         * Check with the firewall...
         */
-       if (ip6_fw_chk_ptr) {
+       if (ip6_fw_enable && ip6_fw_chk_ptr) {
                u_short port = 0;
+               ifp = rt->rt_ifp;
+               /* Drop the lock but retain the extra ref */
+               RT_UNLOCK(rt);
                /* If ipfw says divert, we have to just drop packet */
-               if ((*ip6_fw_chk_ptr)(&ip6, rt->rt_ifp, &port, &m)) {
+               if (ip6_fw_chk_ptr(&ip6, ifp, &port, &m)) {
                        m_freem(m);
                        goto freecopy;
                }
-               if (!m)
+               if (!m) {
                        goto freecopy;
+               }
+               /* We still have the extra ref on rt */
+               RT_LOCK(rt);
        }
 #endif
 
-#if OLDIP6OUTPUT
-       error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m,
-                                        (struct sockaddr *)dst,
-                                        ip6_forward_rt.ro_rt);
+       /*
+        * Fake scoped addresses. Note that even link-local source or
+        * destinaion can appear, if the originating node just sends the
+        * packet to us (without address resolution for the destination).
+        * Since both icmp6_error and icmp6_redirect_output fill the embedded
+        * link identifiers, we can do this stuff after making a copy for
+        * returning an error.
+        */
+       if ((rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
+               /*
+                * See corresponding comments in ip6_output.
+                * XXX: but is it possible that ip6_forward() sends a packet
+                *      to a loopback interface? I don't think so, and thus
+                *      I bark here. (jinmei@kame.net)
+                * XXX: it is common to route invalid packets to loopback.
+                *      also, the codepath will be visited on use of ::1 in
+                *      rthdr. (itojun)
+                */
+#if 1
+               if ((0))
 #else
-       error = nd6_output(rt->rt_ifp, m, dst, rt);
-#endif 
+               if ((rt->rt_flags & (RTF_BLACKHOLE|RTF_REJECT)) == 0)
+#endif
+               {
+                       printf("ip6_forward: outgoing interface is loopback. "
+                               "src %s, dst %s, nxt %d, rcvif %s, outif %s\n",
+                               ip6_sprintf(&ip6->ip6_src),
+                               ip6_sprintf(&ip6->ip6_dst),
+                               ip6->ip6_nxt, if_name(m->m_pkthdr.rcvif),
+                               if_name(rt->rt_ifp));
+               }
+
+               /* we can just use rcvif in forwarding. */
+               origifp = rcvifp = m->m_pkthdr.rcvif;
+       } else if (nd6_prproxy) {
+               /*
+                * In the prefix proxying case, we need to inform nd6_output()
+                * about the inbound interface, so that any subsequent NS
+                * packets generated by nd6_prproxy_ns_output() will not be
+                * sent back to that same interface.
+                */
+               origifp = rcvifp = m->m_pkthdr.rcvif;
+       } else {
+               rcvifp = m->m_pkthdr.rcvif;
+               origifp = rt->rt_ifp;
+       }
+       /*
+        * clear embedded scope identifiers if necessary.
+        * in6_clearscope will touch the addresses only when necessary.
+        */
+       in6_clearscope(&ip6->ip6_src);
+       in6_clearscope(&ip6->ip6_dst);
+
+       ifp = rt->rt_ifp;
+       /* Drop the lock but retain the extra ref */
+       RT_UNLOCK(rt);
+
+       /*
+        * If this is to be processed locally, let ip6_input have it.
+        */
+       if (proxy) {
+               VERIFY(m->m_pkthdr.pkt_flags & PKTF_PROXY_DST);
+               /* Release extra ref */
+               RT_REMREF(rt);
+               if (mcopy != NULL)
+                       m_freem(mcopy);
+               return (m);
+       }
+
+#if PF
+       /* Invoke outbound packet filter */
+       error = pf_af_hook(ifp, NULL, &m, AF_INET6, FALSE, NULL);
+
+       if (error != 0 || m == NULL) {
+               if (m != NULL) {
+                       panic("%s: unexpected packet %p\n", __func__, m);
+                       /* NOTREACHED */
+               }
+               /* Already freed by callee */
+               goto senderr;
+       }
+       ip6 = mtod(m, struct ip6_hdr *);
+#endif /* PF */
+
+       /* Mark this packet as being forwarded from another interface */
+       m->m_pkthdr.pkt_flags |= PKTF_FORWARDED;
+       len = m_pktlen(m);
+
+       error = nd6_output(ifp, origifp, m, dst, rt, NULL);
        if (error) {
-               in6_ifstat_inc(rt->rt_ifp, ifs6_out_discard);
+               in6_ifstat_inc(ifp, ifs6_out_discard);
                ip6stat.ip6s_cantforward++;
        } else {
+               /*
+                * Increment stats on the source interface; the ones
+                * for destination interface has been taken care of
+                * during output above by virtue of PKTF_FORWARDED.
+                */
+               rcvifp->if_fpackets++;
+               rcvifp->if_fbytes += len;
+
                ip6stat.ip6s_forward++;
-               in6_ifstat_inc(rt->rt_ifp, ifs6_out_forward);
+               in6_ifstat_inc(ifp, ifs6_out_forward);
                if (type)
                        ip6stat.ip6s_redirectsent++;
                else {
-                       if (mcopy)
+                       if (mcopy) {
                                goto freecopy;
+                       }
                }
        }
-       if (mcopy == NULL)
-               return;
-
+#if PF
+senderr:
+#endif /* PF */
+       if (mcopy == NULL) {
+               /* Release extra ref */
+               RT_REMREF(rt);
+               return (NULL);
+       }
        switch (error) {
        case 0:
 #if 1
                if (type == ND_REDIRECT) {
                        icmp6_redirect_output(mcopy, rt);
-                       return;
+                       /* Release extra ref */
+                       RT_REMREF(rt);
+                       return (NULL);
                }
 #endif
                goto freecopy;
@@ -516,9 +789,13 @@ ip6_forward(m, srcrt)
                break;
        }
        icmp6_error(mcopy, type, code, 0);
-       return;
+       /* Release extra ref */
+       RT_REMREF(rt);
+       return (NULL);
 
  freecopy:
        m_freem(mcopy);
-       return;
+       /* Release extra ref */
+       RT_REMREF(rt);
+       return (NULL);
 }