keep USETRAILERS from SIOCSARP
[unix-history] / usr / src / sys / netinet / ip_input.c
index 5499ddc..5983463 100644 (file)
@@ -1,29 +1,50 @@
-/*     ip_input.c      1.67    83/05/12        */
-
-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/mbuf.h"
-#include "../h/domain.h"
-#include "../h/protosw.h"
-#include "../h/socket.h"
-#include "../h/errno.h"
-#include "../h/time.h"
-#include "../h/kernel.h"
+/*
+ * Copyright (c) 1982, 1986 Regents of the University of California.
+ * All rights reserved.  The Berkeley software License Agreement
+ * specifies the terms and conditions for redistribution.
+ *
+ *     @(#)ip_input.c  7.6.1.2 (Berkeley) %G%
+ */
+
+#include "param.h"
+#include "systm.h"
+#include "mbuf.h"
+#include "domain.h"
+#include "protosw.h"
+#include "socket.h"
+#include "errno.h"
+#include "time.h"
+#include "kernel.h"
 
 #include "../net/if.h"
 #include "../net/route.h"
 
 
 #include "../net/if.h"
 #include "../net/route.h"
 
-#include "../netinet/in.h"
-#include "../netinet/in_pcb.h"
-#include "../netinet/in_systm.h"
-#include "../netinet/ip.h"
-#include "../netinet/ip_var.h"
-#include "../netinet/ip_icmp.h"
-#include "../netinet/tcp.h"
+#include "in.h"
+#include "in_pcb.h"
+#include "in_systm.h"
+#include "in_var.h"
+#include "ip.h"
+#include "ip_var.h"
+#include "ip_icmp.h"
+#include "tcp.h"
 
 u_char ip_protox[IPPROTO_MAX];
 int    ipqmaxlen = IFQ_MAXLEN;
 
 u_char ip_protox[IPPROTO_MAX];
 int    ipqmaxlen = IFQ_MAXLEN;
-struct ifnet *ifinet;                  /* first inet interface */
+struct in_ifaddr *in_ifaddr;                   /* first inet address */
+
+/*
+ * We need to save the IP options in case a protocol wants to respond
+ * to an incoming packet over the same route if the packet got here
+ * using IP source routing.  This allows connection establishment and
+ * maintenance when the remote end is on a network that is not known
+ * to us.
+ */
+int    ip_nhops = 0;
+static struct ip_srcrt {
+       char    nop;                            /* one NOP to align */
+       char    srcopt[IPOPT_OFFSET + 1];       /* OPTVAL, OLEN and OFFSET */
+       struct  in_addr route[MAX_IPOPTLEN];
+} ip_srcrt;
 
 /*
  * IP initialization: fill in IP protocol switch table.
 
 /*
  * IP initialization: fill in IP protocol switch table.
@@ -34,25 +55,25 @@ ip_init()
        register struct protosw *pr;
        register int i;
 
        register struct protosw *pr;
        register int i;
 
-       pr = pffindproto(PF_INET, IPPROTO_RAW);
+       pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
        if (pr == 0)
                panic("ip_init");
        for (i = 0; i < IPPROTO_MAX; i++)
                ip_protox[i] = pr - inetsw;
        for (pr = inetdomain.dom_protosw;
        if (pr == 0)
                panic("ip_init");
        for (i = 0; i < IPPROTO_MAX; i++)
                ip_protox[i] = pr - inetsw;
        for (pr = inetdomain.dom_protosw;
-           pr <= inetdomain.dom_protoswNPROTOSW; pr++)
-               if (pr->pr_family == PF_INET &&
+           pr < inetdomain.dom_protoswNPROTOSW; pr++)
+               if (pr->pr_domain->dom_family == PF_INET &&
                    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
                        ip_protox[pr->pr_protocol] = pr - inetsw;
        ipq.next = ipq.prev = &ipq;
        ip_id = time.tv_sec & 0xffff;
        ipintrq.ifq_maxlen = ipqmaxlen;
                    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
                        ip_protox[pr->pr_protocol] = pr - inetsw;
        ipq.next = ipq.prev = &ipq;
        ip_id = time.tv_sec & 0xffff;
        ipintrq.ifq_maxlen = ipqmaxlen;
-       ifinet = if_ifwithaf(AF_INET);
 }
 
 u_char ipcksum = 1;
 struct ip *ip_reass();
 struct sockaddr_in ipaddr = { AF_INET };
 }
 
 u_char ipcksum = 1;
 struct ip *ip_reass();
 struct sockaddr_in ipaddr = { AF_INET };
+struct route ipforward_rt;
 
 /*
  * Ip input routine.  Checksum and byte swap header.  If fragmented
 
 /*
  * Ip input routine.  Checksum and byte swap header.  If fragmented
@@ -66,6 +87,8 @@ ipintr()
        struct mbuf *m0;
        register int i;
        register struct ipq *fp;
        struct mbuf *m0;
        register int i;
        register struct ipq *fp;
+       register struct in_ifaddr *ia;
+       struct ifnet *ifp;
        int hlen, s;
 
 next:
        int hlen, s;
 
 next:
@@ -74,17 +97,29 @@ next:
         * in first mbuf.
         */
        s = splimp();
         * in first mbuf.
         */
        s = splimp();
-       IF_DEQUEUE(&ipintrq, m);
+       IF_DEQUEUEIF(&ipintrq, m, ifp);
        splx(s);
        if (m == 0)
                return;
        splx(s);
        if (m == 0)
                return;
+       /*
+        * If no IP addresses have been set yet but the interfaces
+        * are receiving, can't do anything with incoming packets yet.
+        */
+       if (in_ifaddr == NULL)
+               goto bad;
+       ipstat.ips_total++;
        if ((m->m_off > MMAXOFF || m->m_len < sizeof (struct ip)) &&
            (m = m_pullup(m, sizeof (struct ip))) == 0) {
                ipstat.ips_toosmall++;
                goto next;
        }
        ip = mtod(m, struct ip *);
        if ((m->m_off > MMAXOFF || m->m_len < sizeof (struct ip)) &&
            (m = m_pullup(m, sizeof (struct ip))) == 0) {
                ipstat.ips_toosmall++;
                goto next;
        }
        ip = mtod(m, struct ip *);
-       if ((hlen = ip->ip_hl << 2) > m->m_len) {
+       hlen = ip->ip_hl << 2;
+       if (hlen < sizeof(struct ip)) { /* minimum header length */
+               ipstat.ips_badhlen++;
+               goto bad;
+       }
+       if (hlen > m->m_len) {
                if ((m = m_pullup(m, hlen)) == 0) {
                        ipstat.ips_badhlen++;
                        goto next;
                if ((m = m_pullup(m, hlen)) == 0) {
                        ipstat.ips_badhlen++;
                        goto next;
@@ -114,7 +149,7 @@ next:
         * Trim mbufs if longer than we expect.
         * Drop packet if shorter than we expect.
         */
         * Trim mbufs if longer than we expect.
         * Drop packet if shorter than we expect.
         */
-       i = -ip->ip_len;
+       i = -(u_short)ip->ip_len;
        m0 = m;
        for (;;) {
                i += m->m_len;
        m0 = m;
        for (;;) {
                i += m->m_len;
@@ -125,6 +160,7 @@ next:
        if (i != 0) {
                if (i < 0) {
                        ipstat.ips_tooshort++;
        if (i != 0) {
                if (i < 0) {
                        ipstat.ips_tooshort++;
+                       m = m0;
                        goto bad;
                }
                if (i <= m->m_len)
                        goto bad;
                }
                if (i <= m->m_len)
@@ -138,45 +174,53 @@ next:
         * Process options and, if not destined for us,
         * ship it on.  ip_dooptions returns 1 when an
         * error was detected (causing an icmp message
         * Process options and, if not destined for us,
         * ship it on.  ip_dooptions returns 1 when an
         * error was detected (causing an icmp message
-        * to be sent).
+        * to be sent and the original packet to be freed).
         */
         */
-       if (hlen > sizeof (struct ip) && ip_dooptions(ip))
+       ip_nhops = 0;           /* for source routed packets */
+       if (hlen > sizeof (struct ip) && ip_dooptions(ip, ifp))
                goto next;
 
        /*
                goto next;
 
        /*
-        * Fast check on the first internet
-        * interface in the list.
+        * Check our list of addresses, to see if the packet is for us.
         */
         */
-       if (ifinet) {
-               struct sockaddr_in *sin;
+       for (ia = in_ifaddr; ia; ia = ia->ia_next) {
+#define        satosin(sa)     ((struct sockaddr_in *)(sa))
 
 
-               sin = (struct sockaddr_in *)&ifinet->if_addr;
-               if (sin->sin_addr.s_addr == ip->ip_dst.s_addr)
-                       goto ours;
-               sin = (struct sockaddr_in *)&ifinet->if_broadaddr;
-               if ((ifinet->if_flags & IFF_BROADCAST) &&
-                   sin->sin_addr.s_addr == ip->ip_dst.s_addr)
+               if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
                        goto ours;
                        goto ours;
+               if (
+#ifdef DIRECTED_BROADCAST
+                   ia->ia_ifp == ifp &&
+#endif
+                   (ia->ia_ifp->if_flags & IFF_BROADCAST)) {
+                       u_long t;
+
+                       if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
+                           ip->ip_dst.s_addr)
+                               goto ours;
+                       if (ip->ip_dst.s_addr == ia->ia_netbroadcast.s_addr)
+                               goto ours;
+                       /*
+                        * Look for all-0's host part (old broadcast addr),
+                        * either for subnet or net.
+                        */
+                       t = ntohl(ip->ip_dst.s_addr);
+                       if (t == ia->ia_subnet)
+                               goto ours;
+                       if (t == ia->ia_net)
+                               goto ours;
+               }
        }
        }
-/* BEGIN GROT */
-#include "nd.h"
-#if NND > 0
+       if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
+               goto ours;
+       if (ip->ip_dst.s_addr == INADDR_ANY)
+               goto ours;
+
        /*
        /*
-        * Diskless machines don't initially know
-        * their address, so take packets from them
-        * if we're acting as a network disk server.
+        * Not for us; forward if possible and desirable.
         */
         */
-       if (ip->ip_dst.s_addr == INADDR_ANY &&
-           (in_netof(ip->ip_src) == INADDR_ANY &&
-            in_lnaof(ip->ip_src) != INADDR_ANY))
-               goto ours;
-#endif
-/* END GROT */
-       ipaddr.sin_addr = ip->ip_dst;
-       if (if_ifwithaddr((struct sockaddr *)&ipaddr) == 0) {
-               ip_forward(ip);
-               goto next;
-       }
+       ip_forward(ip, ifp);
+       goto next;
 
 ours:
        /*
 
 ours:
        /*
@@ -209,10 +253,10 @@ found:
         * attempt reassembly; if it succeeds, proceed.
         */
        if (((struct ipasfrag *)ip)->ipf_mff || ip->ip_off) {
         * attempt reassembly; if it succeeds, proceed.
         */
        if (((struct ipasfrag *)ip)->ipf_mff || ip->ip_off) {
+               ipstat.ips_fragments++;
                ip = ip_reass((struct ipasfrag *)ip, fp);
                if (ip == 0)
                        goto next;
                ip = ip_reass((struct ipasfrag *)ip, fp);
                if (ip == 0)
                        goto next;
-               hlen = ip->ip_hl << 2;
                m = dtom(ip);
        } else
                if (fp)
                m = dtom(ip);
        } else
                if (fp)
@@ -221,7 +265,7 @@ found:
        /*
         * Switch out to protocol's input routine.
         */
        /*
         * Switch out to protocol's input routine.
         */
-       (*inetsw[ip_protox[ip->ip_p]].pr_input)(m);
+       (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, ifp);
        goto next;
 bad:
        m_freem(m);
        goto next;
 bad:
        m_freem(m);
@@ -256,7 +300,7 @@ ip_reass(ip, fp)
         * If first fragment to arrive, create a reassembly queue.
         */
        if (fp == 0) {
         * If first fragment to arrive, create a reassembly queue.
         */
        if (fp == 0) {
-               if ((t = m_get(M_WAIT, MT_FTABLE)) == NULL)
+               if ((t = m_get(M_DONTWAIT, MT_FTABLE)) == NULL)
                        goto dropfrag;
                fp = mtod(t, struct ipq *);
                insque(fp, &ipq);
                        goto dropfrag;
                fp = mtod(t, struct ipq *);
                insque(fp, &ipq);
@@ -353,11 +397,12 @@ insert:
        remque(fp);
        (void) m_free(dtom(fp));
        m = dtom(ip);
        remque(fp);
        (void) m_free(dtom(fp));
        m = dtom(ip);
-       m->m_len += sizeof (struct ipasfrag);
-       m->m_off -= sizeof (struct ipasfrag);
+       m->m_len += (ip->ip_hl << 2);
+       m->m_off -= (ip->ip_hl << 2);
        return ((struct ip *)ip);
 
 dropfrag:
        return ((struct ip *)ip);
 
 dropfrag:
+       ipstat.ips_fragdropped++;
        m_freem(m);
        return (0);
 }
        m_freem(m);
        return (0);
 }
@@ -423,8 +468,10 @@ ip_slowtimo()
        while (fp != &ipq) {
                --fp->ipq_ttl;
                fp = fp->next;
        while (fp != &ipq) {
                --fp->ipq_ttl;
                fp = fp->next;
-               if (fp->prev->ipq_ttl == 0)
+               if (fp->prev->ipq_ttl == 0) {
+                       ipstat.ips_fragtimeout++;
                        ip_freef(fp->prev);
                        ip_freef(fp->prev);
+               }
        }
        splx(s);
 }
        }
        splx(s);
 }
@@ -435,35 +482,46 @@ ip_slowtimo()
 ip_drain()
 {
 
 ip_drain()
 {
 
-       while (ipq.next != &ipq)
+       while (ipq.next != &ipq) {
+               ipstat.ips_fragdropped++;
                ip_freef(ipq.next);
                ip_freef(ipq.next);
+       }
 }
 
 }
 
+extern struct in_ifaddr *ifptoia();
+struct in_ifaddr *ip_rtaddr();
+
 /*
  * Do option processing on a datagram,
  * possibly discarding it if bad options
  * are encountered.
  */
 /*
  * Do option processing on a datagram,
  * possibly discarding it if bad options
  * are encountered.
  */
-ip_dooptions(ip)
-       struct ip *ip;
+ip_dooptions(ip, ifp)
+       register struct ip *ip;
+       struct ifnet *ifp;
 {
        register u_char *cp;
 {
        register u_char *cp;
-       int opt, optlen, cnt, code, type;
-       struct in_addr *sin;
+       int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
        register struct ip_timestamp *ipt;
        register struct ip_timestamp *ipt;
-       register struct ifnet *ifp;
-       struct in_addr t;
+       register struct in_ifaddr *ia;
+       struct in_addr *sin;
+       n_time ntime;
 
        cp = (u_char *)(ip + 1);
        cnt = (ip->ip_hl << 2) - sizeof (struct ip);
        for (; cnt > 0; cnt -= optlen, cp += optlen) {
 
        cp = (u_char *)(ip + 1);
        cnt = (ip->ip_hl << 2) - sizeof (struct ip);
        for (; cnt > 0; cnt -= optlen, cp += optlen) {
-               opt = cp[0];
+               opt = cp[IPOPT_OPTVAL];
                if (opt == IPOPT_EOL)
                        break;
                if (opt == IPOPT_NOP)
                        optlen = 1;
                if (opt == IPOPT_EOL)
                        break;
                if (opt == IPOPT_NOP)
                        optlen = 1;
-               else
-                       optlen = cp[1];
+               else {
+                       optlen = cp[IPOPT_OLEN];
+                       if (optlen <= 0 || optlen > cnt) {
+                               code = &cp[IPOPT_OLEN] - (u_char *)ip;
+                               goto bad;
+                       }
+               }
                switch (opt) {
 
                default:
                switch (opt) {
 
                default:
@@ -480,30 +538,79 @@ ip_dooptions(ip)
                 */
                case IPOPT_LSRR:
                case IPOPT_SSRR:
                 */
                case IPOPT_LSRR:
                case IPOPT_SSRR:
-                       if (cp[2] < 4 || cp[2] > optlen - (sizeof (long) - 1))
-                               break;
-                       sin = (struct in_addr *)(cp + cp[2]);
-                       ipaddr.sin_addr = *sin;
-                       ifp = if_ifwithaddr((struct sockaddr *)&ipaddr);
-                       type = ICMP_UNREACH, code = ICMP_UNREACH_SRCFAIL;
-                       if (ifp == 0) {
-                               if (opt == IPOPT_SSRR)
+                       if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
+                               code = &cp[IPOPT_OFFSET] - (u_char *)ip;
+                               goto bad;
+                       }
+                       ipaddr.sin_addr = ip->ip_dst;
+                       ia = (struct in_ifaddr *)
+                               ifa_ifwithaddr((struct sockaddr *)&ipaddr);
+                       if (ia == 0) {
+                               if (opt == IPOPT_SSRR) {
+                                       type = ICMP_UNREACH;
+                                       code = ICMP_UNREACH_SRCFAIL;
                                        goto bad;
                                        goto bad;
+                               }
+                               /*
+                                * Loose routing, and not at next destination
+                                * yet; nothing to do except forward.
+                                */
                                break;
                        }
                                break;
                        }
-                       t = ip->ip_dst; ip->ip_dst = *sin; *sin = t;
-                       cp[2] += 4;
-                       if (cp[2] > optlen - (sizeof (long) - 1))
+                       off--;                  /* 0 origin */
+                       if (off > optlen - sizeof(struct in_addr)) {
+                               /*
+                                * End of source route.  Should be for us.
+                                */
+                               save_rte(cp, ip->ip_src);
                                break;
                                break;
-                       ip->ip_dst = sin[1];
-                       if (opt == IPOPT_SSRR &&
-                           if_ifonnetof(in_netof(ip->ip_dst)) == 0)
+                       }
+                       /*
+                        * locate outgoing interface
+                        */
+                       bcopy((caddr_t)(cp + off), (caddr_t)&ipaddr.sin_addr,
+                           sizeof(ipaddr.sin_addr));
+                       if ((opt == IPOPT_SSRR &&
+                           in_iaonnetof(in_netof(ipaddr.sin_addr)) == 0) ||
+                           (ia = ip_rtaddr(ipaddr.sin_addr)) == 0) {
+                               type = ICMP_UNREACH;
+                               code = ICMP_UNREACH_SRCFAIL;
                                goto bad;
                                goto bad;
+                       }
+                       ip->ip_dst = ipaddr.sin_addr;
+                       bcopy((caddr_t)&(IA_SIN(ia)->sin_addr),
+                           (caddr_t)(cp + off), sizeof(struct in_addr));
+                       cp[IPOPT_OFFSET] += sizeof(struct in_addr);
+                       break;
+
+               case IPOPT_RR:
+                       if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
+                               code = &cp[IPOPT_OFFSET] - (u_char *)ip;
+                               goto bad;
+                       }
+                       /*
+                        * If no space remains, ignore.
+                        */
+                       off--;                  /* 0 origin */
+                       if (off > optlen - sizeof(struct in_addr))
+                               break;
+                       bcopy((caddr_t)(&ip->ip_dst), (caddr_t)&ipaddr.sin_addr,
+                           sizeof(ipaddr.sin_addr));
+                       /*
+                        * locate outgoing interface
+                        */
+                       if ((ia = ip_rtaddr(ipaddr.sin_addr)) == 0) {
+                               type = ICMP_UNREACH;
+                               code = ICMP_UNREACH_HOST;
+                               goto bad;
+                       }
+                       bcopy((caddr_t)&(IA_SIN(ia)->sin_addr),
+                           (caddr_t)(cp + off), sizeof(struct in_addr));
+                       cp[IPOPT_OFFSET] += sizeof(struct in_addr);
                        break;
 
                case IPOPT_TS:
                        code = cp - (u_char *)ip;
                        break;
 
                case IPOPT_TS:
                        code = cp - (u_char *)ip;
-                       type = ICMP_PARAMPROB;
                        ipt = (struct ip_timestamp *)cp;
                        if (ipt->ipt_len < 5)
                                goto bad;
                        ipt = (struct ip_timestamp *)cp;
                        if (ipt->ipt_len < 5)
                                goto bad;
@@ -512,42 +619,145 @@ ip_dooptions(ip)
                                        goto bad;
                                break;
                        }
                                        goto bad;
                                break;
                        }
-                       sin = (struct in_addr *)(cp+cp[2]);
+                       sin = (struct in_addr *)(cp + ipt->ipt_ptr - 1);
                        switch (ipt->ipt_flg) {
 
                        case IPOPT_TS_TSONLY:
                                break;
 
                        case IPOPT_TS_TSANDADDR:
                        switch (ipt->ipt_flg) {
 
                        case IPOPT_TS_TSONLY:
                                break;
 
                        case IPOPT_TS_TSANDADDR:
-                               if (ipt->ipt_ptr + 8 > ipt->ipt_len)
+                               if (ipt->ipt_ptr + sizeof(n_time) +
+                                   sizeof(struct in_addr) > ipt->ipt_len)
                                        goto bad;
                                        goto bad;
-                               if (ifinet == 0)
-                                       goto bad;       /* ??? */
-                               *sin++ = ((struct sockaddr_in *)&ifinet->if_addr)->sin_addr;
+                               ia = ifptoia(ifp);
+                               bcopy((caddr_t)&IA_SIN(ia)->sin_addr,
+                                   (caddr_t)sin, sizeof(struct in_addr));
+                               ipt->ipt_ptr += sizeof(struct in_addr);
                                break;
 
                        case IPOPT_TS_PRESPEC:
                                break;
 
                        case IPOPT_TS_PRESPEC:
-                               ipaddr.sin_addr = *sin;
-                               if (if_ifwithaddr((struct sockaddr *)&ipaddr) == 0)
-                                       continue;
-                               if (ipt->ipt_ptr + 8 > ipt->ipt_len)
+                               if (ipt->ipt_ptr + sizeof(n_time) +
+                                   sizeof(struct in_addr) > ipt->ipt_len)
                                        goto bad;
                                        goto bad;
-                               ipt->ipt_ptr += 4;
+                               bcopy((caddr_t)sin, (caddr_t)&ipaddr.sin_addr,
+                                   sizeof(struct in_addr));
+                               if (ifa_ifwithaddr((struct sockaddr *)&ipaddr) == 0)
+                                       continue;
+                               ipt->ipt_ptr += sizeof(struct in_addr);
                                break;
 
                        default:
                                goto bad;
                        }
                                break;
 
                        default:
                                goto bad;
                        }
-                       *(n_time *)sin = iptime();
-                       ipt->ipt_ptr += 4;
+                       ntime = iptime();
+                       bcopy((caddr_t)&ntime, (caddr_t)cp + ipt->ipt_ptr - 1,
+                           sizeof(n_time));
+                       ipt->ipt_ptr += sizeof(n_time);
                }
        }
        return (0);
 bad:
                }
        }
        return (0);
 bad:
-       icmp_error(ip, type, code);
+       icmp_error(ip, type, code, ifp);
        return (1);
 }
 
        return (1);
 }
 
+/*
+ * Given address of next destination (final or next hop),
+ * return internet address info of interface to be used to get there.
+ */
+struct in_ifaddr *
+ip_rtaddr(dst)
+        struct in_addr dst;
+{
+       register struct sockaddr_in *sin;
+       register struct in_ifaddr *ia;
+
+       sin = (struct sockaddr_in *) &ipforward_rt.ro_dst;
+
+       if (ipforward_rt.ro_rt == 0 || dst.s_addr != sin->sin_addr.s_addr) {
+               if (ipforward_rt.ro_rt) {
+                       RTFREE(ipforward_rt.ro_rt);
+                       ipforward_rt.ro_rt = 0;
+               }
+               sin->sin_family = AF_INET;
+               sin->sin_addr = dst;
+
+               rtalloc(&ipforward_rt);
+       }
+       if (ipforward_rt.ro_rt == 0)
+               return ((struct in_ifaddr *)0);
+       /*
+        * Find address associated with outgoing interface.
+        */
+       for (ia = in_ifaddr; ia; ia = ia->ia_next)
+               if (ia->ia_ifp == ipforward_rt.ro_rt->rt_ifp)
+                       break;
+       return (ia);
+}
+
+/*
+ * Save incoming source route for use in replies,
+ * to be picked up later by ip_srcroute if the receiver is interested.
+ */
+save_rte(option, dst)
+       u_char *option;
+       struct in_addr dst;
+{
+       unsigned olen;
+       extern ipprintfs;
+
+       olen = option[IPOPT_OLEN];
+       if (olen > sizeof(ip_srcrt) - 1) {
+               if (ipprintfs)
+                       printf("save_rte: olen %d\n", olen);
+               return;
+       }
+       bcopy((caddr_t)option, (caddr_t)ip_srcrt.srcopt, olen);
+       ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
+       ip_srcrt.route[ip_nhops++] = dst;
+}
+
+/*
+ * Retrieve incoming source route for use in replies,
+ * in the same form used by setsockopt.
+ * The first hop is placed before the options, will be removed later.
+ */
+struct mbuf *
+ip_srcroute()
+{
+       register struct in_addr *p, *q;
+       register struct mbuf *m;
+
+       if (ip_nhops == 0)
+               return ((struct mbuf *)0);
+       m = m_get(M_DONTWAIT, MT_SOOPTS);
+       if (m == 0)
+               return ((struct mbuf *)0);
+       m->m_len = ip_nhops * sizeof(struct in_addr) + IPOPT_OFFSET + 1 + 1;
+
+       /*
+        * First save first hop for return route
+        */
+       p = &ip_srcrt.route[ip_nhops - 1];
+       *(mtod(m, struct in_addr *)) = *p--;
+
+       /*
+        * Copy option fields and padding (nop) to mbuf.
+        */
+       ip_srcrt.nop = IPOPT_NOP;
+       bcopy((caddr_t)&ip_srcrt, mtod(m, caddr_t) + sizeof(struct in_addr),
+           IPOPT_OFFSET + 1 + 1);
+       q = (struct in_addr *)(mtod(m, caddr_t) +
+           sizeof(struct in_addr) + IPOPT_OFFSET + 1 + 1);
+       /*
+        * Record return path as an IP source route,
+        * reversing the path (pointers are now aligned).
+        */
+       while (p >= ip_srcrt.route)
+               *q++ = *p--;
+       return (m);
+}
+
 /*
  * Strip out IP options, at higher
  * level protocol in the kernel.
 /*
  * Strip out IP options, at higher
  * level protocol in the kernel.
@@ -560,104 +770,170 @@ ip_stripoptions(ip, mopt)
 {
        register int i;
        register struct mbuf *m;
 {
        register int i;
        register struct mbuf *m;
+       register caddr_t opts;
        int olen;
 
        olen = (ip->ip_hl<<2) - sizeof (struct ip);
        m = dtom(ip);
        int olen;
 
        olen = (ip->ip_hl<<2) - sizeof (struct ip);
        m = dtom(ip);
-       ip++;
+       opts = (caddr_t)(ip + 1);
        if (mopt) {
                mopt->m_len = olen;
                mopt->m_off = MMINOFF;
        if (mopt) {
                mopt->m_len = olen;
                mopt->m_off = MMINOFF;
-               bcopy((caddr_t)ip, mtod(m, caddr_t), (unsigned)olen);
+               bcopy(opts, mtod(mopt, caddr_t), (unsigned)olen);
        }
        i = m->m_len - (sizeof (struct ip) + olen);
        }
        i = m->m_len - (sizeof (struct ip) + olen);
-       bcopy((caddr_t)ip+olen, (caddr_t)ip, (unsigned)i);
+       bcopy(opts  + olen, opts, (unsigned)i);
        m->m_len -= olen;
        m->m_len -= olen;
+       ip->ip_hl = sizeof(struct ip) >> 2;
 }
 
 }
 
-u_char inetctlerrmap[] = {
-       ECONNABORTED,   ECONNABORTED,   0,              0,
-       0,              0,
-       EHOSTDOWN,      EHOSTUNREACH,   ENETUNREACH,    EHOSTUNREACH,
-       ECONNREFUSED,   ECONNREFUSED,   EMSGSIZE,       0,
-       0,              0,              0,              0
+u_char inetctlerrmap[PRC_NCMDS] = {
+       0,              0,              0,              0,
+       0,              0,              EHOSTDOWN,      EHOSTUNREACH,
+       ENETUNREACH,    EHOSTUNREACH,   ECONNREFUSED,   ECONNREFUSED,
+       EMSGSIZE,       EHOSTUNREACH,   0,              0,
+       0,              0,              0,              0,
+       ENOPROTOOPT
 };
 
 };
 
-ip_ctlinput(cmd, arg)
-       int cmd;
-       caddr_t arg;
-{
-       struct in_addr *in;
-       int tcp_abort(), udp_abort();
-       extern struct inpcb tcb, udb;
-
-       if (cmd < 0 || cmd > PRC_NCMDS)
-               return;
-       if (inetctlerrmap[cmd] == 0)
-               return;         /* XXX */
-       if (cmd == PRC_IFDOWN)
-               in = &((struct sockaddr_in *)arg)->sin_addr;
-       else if (cmd == PRC_HOSTDEAD || cmd == PRC_HOSTUNREACH)
-               in = (struct in_addr *)arg;
-       else
-               in = &((struct icmp *)arg)->icmp_ip.ip_dst;
-/* THIS IS VERY QUESTIONABLE, SHOULD HIT ALL PROTOCOLS */
-       in_pcbnotify(&tcb, in, (int)inetctlerrmap[cmd], tcp_abort);
-       in_pcbnotify(&udb, in, (int)inetctlerrmap[cmd], udp_abort);
-}
-
+#ifndef        IPFORWARDING
+#define        IPFORWARDING    1
+#endif
+#ifndef        IPSENDREDIRECTS
+#define        IPSENDREDIRECTS 1
+#endif
 int    ipprintfs = 0;
 int    ipprintfs = 0;
-int    ipforwarding = 1;
+int    ipforwarding = IPFORWARDING;
+extern int in_interfaces;
+int    ipsendredirects = IPSENDREDIRECTS;
+
 /*
  * Forward a packet.  If some error occurs return the sender
 /*
  * Forward a packet.  If some error occurs return the sender
- * and icmp packet.  Note we can't always generate a meaningful
- * icmp message because icmp doesn't have a large enough repetoire
+ * an icmp packet.  Note we can't always generate a meaningful
+ * icmp message because icmp doesn't have a large enough repertoire
  * of codes and types.
  * of codes and types.
+ *
+ * If not forwarding (possibly because we have only a single external
+ * network), just drop the packet.  This could be confusing if ipforwarding
+ * was zero but some routing protocol was advancing us as a gateway
+ * to somewhere.  However, we must let the routing protocol deal with that.
  */
  */
-ip_forward(ip)
+ip_forward(ip, ifp)
        register struct ip *ip;
        register struct ip *ip;
+       struct ifnet *ifp;
 {
 {
-       register int error, type, code;
-       struct mbuf *mopt, *mcopy;
+       register int error, type = 0, code;
+       register struct sockaddr_in *sin;
+       struct mbuf *mcopy;
+       struct in_addr dest;
 
 
+       dest.s_addr = 0;
        if (ipprintfs)
                printf("forward: src %x dst %x ttl %x\n", ip->ip_src,
                        ip->ip_dst, ip->ip_ttl);
        if (ipprintfs)
                printf("forward: src %x dst %x ttl %x\n", ip->ip_src,
                        ip->ip_dst, ip->ip_ttl);
-       if (ipforwarding == 0) {
-               /* can't tell difference between net and host */
+       ip->ip_id = htons(ip->ip_id);
+       if (ipforwarding == 0 || in_interfaces <= 1) {
+               ipstat.ips_cantforward++;
+#ifdef GATEWAY
                type = ICMP_UNREACH, code = ICMP_UNREACH_NET;
                goto sendicmp;
                type = ICMP_UNREACH, code = ICMP_UNREACH_NET;
                goto sendicmp;
+#else
+               m_freem(dtom(ip));
+               return;
+#endif
        }
        }
-       if (ip->ip_ttl < IPTTLDEC) {
+       if (ip->ip_ttl <= IPTTLDEC) {
                type = ICMP_TIMXCEED, code = ICMP_TIMXCEED_INTRANS;
                goto sendicmp;
        }
        ip->ip_ttl -= IPTTLDEC;
                type = ICMP_TIMXCEED, code = ICMP_TIMXCEED_INTRANS;
                goto sendicmp;
        }
        ip->ip_ttl -= IPTTLDEC;
-       mopt = m_get(M_DONTWAIT, MT_DATA);
-       if (mopt == NULL) {
-               m_freem(dtom(ip));
-               return;
-       }
 
        /*
         * Save at most 64 bytes of the packet in case
         * we need to generate an ICMP message to the src.
         */
 
        /*
         * Save at most 64 bytes of the packet in case
         * we need to generate an ICMP message to the src.
         */
-       mcopy = m_copy(dtom(ip), 0, imin(ip->ip_len, 64));
-       ip_stripoptions(ip, mopt);
+       mcopy = m_copy(dtom(ip), 0, imin((int)ip->ip_len, 64));
+
+       sin = (struct sockaddr_in *)&ipforward_rt.ro_dst;
+       if (ipforward_rt.ro_rt == 0 ||
+           ip->ip_dst.s_addr != sin->sin_addr.s_addr) {
+               if (ipforward_rt.ro_rt) {
+                       RTFREE(ipforward_rt.ro_rt);
+                       ipforward_rt.ro_rt = 0;
+               }
+               sin->sin_family = AF_INET;
+               sin->sin_addr = ip->ip_dst;
 
 
-       error = ip_output(dtom(ip), mopt, (struct route *)0, IP_FORWARDING);
-       if (error == 0) {
+               rtalloc(&ipforward_rt);
+       }
+       /*
+        * If forwarding packet using same interface that it came in on,
+        * perhaps should send a redirect to sender to shortcut a hop.
+        * Only send redirect if source is sending directly to us,
+        * and if packet was not source routed (or has any options).
+        * Also, don't send redirect if forwarding using a default route
+        * or a route modfied by a redirect.
+        */
+#define        satosin(sa)     ((struct sockaddr_in *)(sa))
+       if (ipforward_rt.ro_rt && ipforward_rt.ro_rt->rt_ifp == ifp &&
+           (ipforward_rt.ro_rt->rt_flags & RTF_DYNAMIC) == 0 &&
+           satosin(&ipforward_rt.ro_rt->rt_dst)->sin_addr.s_addr != 0 &&
+           ipsendredirects && ip->ip_hl == (sizeof(struct ip) >> 2)) {
+               struct in_ifaddr *ia;
+               u_long src = ntohl(ip->ip_src.s_addr);
+               u_long dst = ntohl(ip->ip_dst.s_addr);
+
+               if ((ia = ifptoia(ifp)) &&
+                  (src & ia->ia_subnetmask) == ia->ia_subnet) {
+                   if (ipforward_rt.ro_rt->rt_flags & RTF_GATEWAY)
+                       dest = satosin(&ipforward_rt.ro_rt->rt_gateway)->sin_addr;
+                   else
+                       dest = ip->ip_dst;
+                   /*
+                    * If the destination is reached by a route to host,
+                    * is on a subnet of a local net, or is directly
+                    * on the attached net (!), use host redirect.
+                    * (We may be the correct first hop for other subnets.)
+                    */
+                   type = ICMP_REDIRECT;
+                   code = ICMP_REDIRECT_NET;
+                   if ((ipforward_rt.ro_rt->rt_flags & RTF_HOST) ||
+                      (ipforward_rt.ro_rt->rt_flags & RTF_GATEWAY) == 0)
+                       code = ICMP_REDIRECT_HOST;
+                   else for (ia = in_ifaddr; ia = ia->ia_next; )
+                       if ((dst & ia->ia_netmask) == ia->ia_net) {
+                           if (ia->ia_subnetmask != ia->ia_netmask)
+                                   code = ICMP_REDIRECT_HOST;
+                           break;
+                       }
+                   if (ipprintfs)
+                       printf("redirect (%d) to %x\n", code, dest);
+               }
+       }
+
+       error = ip_output(dtom(ip), (struct mbuf *)0, &ipforward_rt,
+               IP_FORWARDING);
+       if (error)
+               ipstat.ips_cantforward++;
+       else if (type)
+               ipstat.ips_redirectsent++;
+       else {
                if (mcopy)
                        m_freem(mcopy);
                if (mcopy)
                        m_freem(mcopy);
+               ipstat.ips_forward++;
                return;
        }
        if (mcopy == NULL)
                return;
        ip = mtod(mcopy, struct ip *);
                return;
        }
        if (mcopy == NULL)
                return;
        ip = mtod(mcopy, struct ip *);
-       type = ICMP_UNREACH, code = 0;          /* need ``undefined'' */
+       type = ICMP_UNREACH;
        switch (error) {
 
        switch (error) {
 
+       case 0:                         /* forwarded, but need redirect */
+               type = ICMP_REDIRECT;
+               /* code set above */
+               break;
+
        case ENETUNREACH:
        case ENETDOWN:
                code = ICMP_UNREACH_NET;
        case ENETUNREACH:
        case ENETDOWN:
                code = ICMP_UNREACH_NET;
@@ -681,5 +957,5 @@ ip_forward(ip)
                break;
        }
 sendicmp:
                break;
        }
 sendicmp:
-       icmp_error(ip, type, code);
+       icmp_error(ip, type, code, ifp, dest);
 }
 }