well, if netproto doesnt need old prototypes inet6 doesnt either
[dragonfly.git] / sys / netinet6 / ip6_forward.c
1 /*      $FreeBSD: src/sys/netinet6/ip6_forward.c,v 1.4.2.7 2003/01/24 05:11:35 sam Exp $        */
2 /*      $DragonFly: src/sys/netinet6/ip6_forward.c,v 1.5 2003/08/23 11:02:45 rob Exp $  */
3 /*      $KAME: ip6_forward.c,v 1.69 2001/05/17 03:48:30 itojun Exp $    */
4
5 /*
6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the project nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 #include "opt_ip6fw.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_ipsec.h"
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/domain.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/errno.h>
47 #include <sys/time.h>
48 #include <sys/kernel.h>
49 #include <sys/syslog.h>
50
51 #include <net/if.h>
52 #include <net/route.h>
53
54 #include <netinet/in.h>
55 #include <netinet/in_var.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/ip.h>
58 #include <netinet/ip_var.h>
59 #include <netinet6/in6_var.h>
60 #include <netinet/ip6.h>
61 #include <netinet6/ip6_var.h>
62 #include <netinet/icmp6.h>
63 #include <netinet6/nd6.h>
64
65 #include <netinet/in_pcb.h>
66
67 #ifdef IPSEC
68 #include <netinet6/ipsec.h>
69 #ifdef INET6
70 #include <netinet6/ipsec6.h>
71 #endif
72 #include <netproto/key/key.h>
73 #endif /* IPSEC */
74
75 #ifdef FAST_IPSEC
76 #include "ipsec.h"
77 #include "ipsec6.h"
78 #include <netipsec/key.h>
79 #define IPSEC
80 #endif /* FAST_IPSEC */
81
82 #include <net/ip6fw/ip6_fw.h>
83
84 #include <net/net_osdep.h>
85
86 extern int (*fr_checkp) (struct ip *, int, struct ifnet *, int, struct mbuf **);
87
88 struct  route_in6 ip6_forward_rt;
89
90 /*
91  * Forward a packet.  If some error occurs return the sender
92  * an icmp packet.  Note we can't always generate a meaningful
93  * icmp message because icmp doesn't have a large enough repertoire
94  * of codes and types.
95  *
96  * If not forwarding, just drop the packet.  This could be confusing
97  * if ipforwarding was zero but some routing protocol was advancing
98  * us as a gateway to somewhere.  However, we must let the routing
99  * protocol deal with that.
100  *
101  */
102
103 void
104 ip6_forward(m, srcrt)
105         struct mbuf *m;
106         int srcrt;
107 {
108         struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
109         struct sockaddr_in6 *dst;
110         struct rtentry *rt;
111         int error, type = 0, code = 0;
112         struct mbuf *mcopy = NULL;
113         struct ifnet *origifp;  /* maybe unnecessary */
114 #ifdef IPSEC
115         struct secpolicy *sp = NULL;
116 #endif
117
118 #ifdef IPSEC
119         /*
120          * Check AH/ESP integrity.
121          */
122         /*
123          * Don't increment ip6s_cantforward because this is the check
124          * before forwarding packet actually.
125          */
126         if (ipsec6_in_reject(m, NULL)) {
127 #if !defined(FAST_IPSEC)
128                 ipsec6stat.in_polvio++;
129 #endif
130                 m_freem(m);
131                 return;
132         }
133 #endif /* IPSEC */
134
135         /*
136          * Do not forward packets to multicast destination (should be handled
137          * by ip6_mforward().
138          * Do not forward packets with unspecified source.  It was discussed
139          * in July 2000, on ipngwg mailing list.
140          */
141         if ((m->m_flags & (M_BCAST|M_MCAST)) != 0 ||
142             IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
143             IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
144                 ip6stat.ip6s_cantforward++;
145                 /* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard) */
146                 if (ip6_log_time + ip6_log_interval < time_second) {
147                         ip6_log_time = time_second;
148                         log(LOG_DEBUG,
149                             "cannot forward "
150                             "from %s to %s nxt %d received on %s\n",
151                             ip6_sprintf(&ip6->ip6_src),
152                             ip6_sprintf(&ip6->ip6_dst),
153                             ip6->ip6_nxt,
154                             if_name(m->m_pkthdr.rcvif));
155                 }
156                 m_freem(m);
157                 return;
158         }
159
160         if (ip6->ip6_hlim <= IPV6_HLIMDEC) {
161                 /* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard) */
162                 icmp6_error(m, ICMP6_TIME_EXCEEDED,
163                                 ICMP6_TIME_EXCEED_TRANSIT, 0);
164                 return;
165         }
166         ip6->ip6_hlim -= IPV6_HLIMDEC;
167
168         /*
169          * Save at most ICMPV6_PLD_MAXLEN (= the min IPv6 MTU -
170          * size of IPv6 + ICMPv6 headers) bytes of the packet in case
171          * we need to generate an ICMP6 message to the src.
172          * Thanks to M_EXT, in most cases copy will not occur.
173          *
174          * It is important to save it before IPsec processing as IPsec
175          * processing may modify the mbuf.
176          */
177         mcopy = m_copy(m, 0, imin(m->m_pkthdr.len, ICMPV6_PLD_MAXLEN));
178
179 #ifdef IPSEC
180         /* get a security policy for this packet */
181         sp = ipsec6_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, IP_FORWARDING,
182             &error);
183         if (sp == NULL) {
184                 ipsec6stat.out_inval++;
185                 ip6stat.ip6s_cantforward++;
186                 if (mcopy) {
187 #if 0
188                         /* XXX: what icmp ? */
189 #else
190                         m_freem(mcopy);
191 #endif
192                 }
193                 m_freem(m);
194                 return;
195         }
196
197         error = 0;
198
199         /* check policy */
200         switch (sp->policy) {
201         case IPSEC_POLICY_DISCARD:
202                 /*
203                  * This packet is just discarded.
204                  */
205                 ipsec6stat.out_polvio++;
206                 ip6stat.ip6s_cantforward++;
207                 key_freesp(sp);
208                 if (mcopy) {
209 #if 0
210                         /* XXX: what icmp ? */
211 #else
212                         m_freem(mcopy);
213 #endif
214                 }
215                 m_freem(m);
216                 return;
217
218         case IPSEC_POLICY_BYPASS:
219         case IPSEC_POLICY_NONE:
220                 /* no need to do IPsec. */
221                 key_freesp(sp);
222                 goto skip_ipsec;
223
224         case IPSEC_POLICY_IPSEC:
225                 if (sp->req == NULL) {
226                         /* XXX should be panic ? */
227                         printf("ip6_forward: No IPsec request specified.\n");
228                         ip6stat.ip6s_cantforward++;
229                         key_freesp(sp);
230                         if (mcopy) {
231 #if 0
232                                 /* XXX: what icmp ? */
233 #else
234                                 m_freem(mcopy);
235 #endif
236                         }
237                         m_freem(m);
238                         return;
239                 }
240                 /* do IPsec */
241                 break;
242
243         case IPSEC_POLICY_ENTRUST:
244         default:
245                 /* should be panic ?? */
246                 printf("ip6_forward: Invalid policy found. %d\n", sp->policy);
247                 key_freesp(sp);
248                 goto skip_ipsec;
249         }
250
251     {
252         struct ipsec_output_state state;
253
254         /*
255          * All the extension headers will become inaccessible
256          * (since they can be encrypted).
257          * Don't panic, we need no more updates to extension headers
258          * on inner IPv6 packet (since they are now encapsulated).
259          *
260          * IPv6 [ESP|AH] IPv6 [extension headers] payload
261          */
262         bzero(&state, sizeof(state));
263         state.m = m;
264         state.ro = NULL;        /* update at ipsec6_output_tunnel() */
265         state.dst = NULL;       /* update at ipsec6_output_tunnel() */
266
267         error = ipsec6_output_tunnel(&state, sp, 0);
268
269         m = state.m;
270         key_freesp(sp);
271
272         if (error) {
273                 /* mbuf is already reclaimed in ipsec6_output_tunnel. */
274                 switch (error) {
275                 case EHOSTUNREACH:
276                 case ENETUNREACH:
277                 case EMSGSIZE:
278                 case ENOBUFS:
279                 case ENOMEM:
280                         break;
281                 default:
282                         printf("ip6_output (ipsec): error code %d\n", error);
283                         /* fall through */
284                 case ENOENT:
285                         /* don't show these error codes to the user */
286                         break;
287                 }
288                 ip6stat.ip6s_cantforward++;
289                 if (mcopy) {
290 #if 0
291                         /* XXX: what icmp ? */
292 #else
293                         m_freem(mcopy);
294 #endif
295                 }
296                 m_freem(m);
297                 return;
298         }
299     }
300     skip_ipsec:
301 #endif /* IPSEC */
302
303         dst = (struct sockaddr_in6 *)&ip6_forward_rt.ro_dst;
304         if (!srcrt) {
305                 /*
306                  * ip6_forward_rt.ro_dst.sin6_addr is equal to ip6->ip6_dst
307                  */
308                 if (ip6_forward_rt.ro_rt == 0 ||
309                     (ip6_forward_rt.ro_rt->rt_flags & RTF_UP) == 0) {
310                         if (ip6_forward_rt.ro_rt) {
311                                 RTFREE(ip6_forward_rt.ro_rt);
312                                 ip6_forward_rt.ro_rt = 0;
313                         }
314                         /* this probably fails but give it a try again */
315                         rtalloc_ign((struct route *)&ip6_forward_rt,
316                                     RTF_PRCLONING);
317                 }
318
319                 if (ip6_forward_rt.ro_rt == 0) {
320                         ip6stat.ip6s_noroute++;
321                         in6_ifstat_inc(m->m_pkthdr.rcvif, ifs6_in_noroute);
322                         if (mcopy) {
323                                 icmp6_error(mcopy, ICMP6_DST_UNREACH,
324                                             ICMP6_DST_UNREACH_NOROUTE, 0);
325                         }
326                         m_freem(m);
327                         return;
328                 }
329         } else if ((rt = ip6_forward_rt.ro_rt) == 0 ||
330                  !IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &dst->sin6_addr)) {
331                 if (ip6_forward_rt.ro_rt) {
332                         RTFREE(ip6_forward_rt.ro_rt);
333                         ip6_forward_rt.ro_rt = 0;
334                 }
335                 bzero(dst, sizeof(*dst));
336                 dst->sin6_len = sizeof(struct sockaddr_in6);
337                 dst->sin6_family = AF_INET6;
338                 dst->sin6_addr = ip6->ip6_dst;
339
340                 rtalloc_ign((struct route *)&ip6_forward_rt, RTF_PRCLONING);
341                 if (ip6_forward_rt.ro_rt == 0) {
342                         ip6stat.ip6s_noroute++;
343                         in6_ifstat_inc(m->m_pkthdr.rcvif, ifs6_in_noroute);
344                         if (mcopy) {
345                                 icmp6_error(mcopy, ICMP6_DST_UNREACH,
346                                             ICMP6_DST_UNREACH_NOROUTE, 0);
347                         }
348                         m_freem(m);
349                         return;
350                 }
351         }
352         rt = ip6_forward_rt.ro_rt;
353
354         /*
355          * Scope check: if a packet can't be delivered to its destination
356          * for the reason that the destination is beyond the scope of the
357          * source address, discard the packet and return an icmp6 destination
358          * unreachable error with Code 2 (beyond scope of source address).
359          * [draft-ietf-ipngwg-icmp-v3-02.txt, Section 3.1]
360          */
361         if (in6_addr2scopeid(m->m_pkthdr.rcvif, &ip6->ip6_src) !=
362             in6_addr2scopeid(rt->rt_ifp, &ip6->ip6_src)) {
363                 ip6stat.ip6s_cantforward++;
364                 ip6stat.ip6s_badscope++;
365                 in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard);
366
367                 if (ip6_log_time + ip6_log_interval < time_second) {
368                         ip6_log_time = time_second;
369                         log(LOG_DEBUG,
370                             "cannot forward "
371                             "src %s, dst %s, nxt %d, rcvif %s, outif %s\n",
372                             ip6_sprintf(&ip6->ip6_src),
373                             ip6_sprintf(&ip6->ip6_dst),
374                             ip6->ip6_nxt,
375                             if_name(m->m_pkthdr.rcvif), if_name(rt->rt_ifp));
376                 }
377                 if (mcopy)
378                         icmp6_error(mcopy, ICMP6_DST_UNREACH,
379                                     ICMP6_DST_UNREACH_BEYONDSCOPE, 0);
380                 m_freem(m);
381                 return;
382         }
383
384         if (m->m_pkthdr.len > rt->rt_ifp->if_mtu) {
385                 in6_ifstat_inc(rt->rt_ifp, ifs6_in_toobig);
386                 if (mcopy) {
387                         u_long mtu;
388 #ifdef IPSEC
389                         struct secpolicy *sp;
390                         int ipsecerror;
391                         size_t ipsechdrsiz;
392 #endif
393
394                         mtu = rt->rt_ifp->if_mtu;
395 #ifdef IPSEC
396                         /*
397                          * When we do IPsec tunnel ingress, we need to play
398                          * with if_mtu value (decrement IPsec header size
399                          * from mtu value).  The code is much simpler than v4
400                          * case, as we have the outgoing interface for
401                          * encapsulated packet as "rt->rt_ifp".
402                          */
403                         sp = ipsec6_getpolicybyaddr(mcopy, IPSEC_DIR_OUTBOUND,
404                                 IP_FORWARDING, &ipsecerror);
405                         if (sp) {
406                                 ipsechdrsiz = ipsec6_hdrsiz(mcopy,
407                                         IPSEC_DIR_OUTBOUND, NULL);
408                                 if (ipsechdrsiz < mtu)
409                                         mtu -= ipsechdrsiz;
410                         }
411
412                         /*
413                          * if mtu becomes less than minimum MTU,
414                          * tell minimum MTU (and I'll need to fragment it).
415                          */
416                         if (mtu < IPV6_MMTU)
417                                 mtu = IPV6_MMTU;
418 #endif
419                         icmp6_error(mcopy, ICMP6_PACKET_TOO_BIG, 0, mtu);
420                 }
421                 m_freem(m);
422                 return;
423         }
424
425         if (rt->rt_flags & RTF_GATEWAY)
426                 dst = (struct sockaddr_in6 *)rt->rt_gateway;
427
428         /*
429          * If we are to forward the packet using the same interface
430          * as one we got the packet from, perhaps we should send a redirect
431          * to sender to shortcut a hop.
432          * Only send redirect if source is sending directly to us,
433          * and if packet was not source routed (or has any options).
434          * Also, don't send redirect if forwarding using a route
435          * modified by a redirect.
436          */
437         if (rt->rt_ifp == m->m_pkthdr.rcvif && !srcrt &&
438             (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0) {
439                 if ((rt->rt_ifp->if_flags & IFF_POINTOPOINT) != 0) {
440                         /*
441                          * If the incoming interface is equal to the outgoing
442                          * one, and the link attached to the interface is
443                          * point-to-point, then it will be highly probable
444                          * that a routing loop occurs. Thus, we immediately
445                          * drop the packet and send an ICMPv6 error message.
446                          *
447                          * type/code is based on suggestion by Rich Draves.
448                          * not sure if it is the best pick.
449                          */
450                         icmp6_error(mcopy, ICMP6_DST_UNREACH,
451                                     ICMP6_DST_UNREACH_ADDR, 0);
452                         m_freem(m);
453                         return;
454                 }
455                 type = ND_REDIRECT;
456         }
457
458         /*
459          * Check with the firewall...
460          */
461         if (ip6_fw_enable && ip6_fw_chk_ptr) {
462                 u_short port = 0;
463                 /* If ipfw says divert, we have to just drop packet */
464                 if ((*ip6_fw_chk_ptr)(&ip6, rt->rt_ifp, &port, &m)) {
465                         m_freem(m);
466                         goto freecopy;
467                 }
468                 if (!m)
469                         goto freecopy;
470         }
471
472         /*
473          * Fake scoped addresses. Note that even link-local source or
474          * destinaion can appear, if the originating node just sends the
475          * packet to us (without address resolution for the destination).
476          * Since both icmp6_error and icmp6_redirect_output fill the embedded
477          * link identifiers, we can do this stuff after making a copy for
478          * returning an error.
479          */
480         if ((rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
481                 /*
482                  * See corresponding comments in ip6_output.
483                  * XXX: but is it possible that ip6_forward() sends a packet
484                  *      to a loopback interface? I don't think so, and thus
485                  *      I bark here. (jinmei@kame.net)
486                  * XXX: it is common to route invalid packets to loopback.
487                  *      also, the codepath will be visited on use of ::1 in
488                  *      rthdr. (itojun)
489                  */
490 #if 1
491                 if (0)
492 #else
493                 if ((rt->rt_flags & (RTF_BLACKHOLE|RTF_REJECT)) == 0)
494 #endif
495                 {
496                         printf("ip6_forward: outgoing interface is loopback. "
497                                 "src %s, dst %s, nxt %d, rcvif %s, outif %s\n",
498                                 ip6_sprintf(&ip6->ip6_src),
499                                 ip6_sprintf(&ip6->ip6_dst),
500                                 ip6->ip6_nxt, if_name(m->m_pkthdr.rcvif),
501                                 if_name(rt->rt_ifp));
502                 }
503
504                 /* we can just use rcvif in forwarding. */
505                 origifp = m->m_pkthdr.rcvif;
506         }
507         else
508                 origifp = rt->rt_ifp;
509 #ifndef SCOPEDROUTING
510         /*
511          * clear embedded scope identifiers if necessary.
512          * in6_clearscope will touch the addresses only when necessary.
513          */
514         in6_clearscope(&ip6->ip6_src);
515         in6_clearscope(&ip6->ip6_dst);
516 #endif
517
518         /*
519          * Check if we want to allow this packet to be processed.
520          * Consider it to be bad if not.
521          */
522         if (fr_checkp) {
523                 struct  mbuf    *m1 = m;
524
525                 if ((*fr_checkp)((struct ip *)ip6, sizeof(*ip6),
526                                  rt->rt_ifp, 1, &m1) != 0)
527                         goto freecopy;
528                 m = m1;
529                 if (m == NULL)
530                         goto freecopy;
531                 ip6 = mtod(m, struct ip6_hdr *);
532         }
533
534         error = nd6_output(rt->rt_ifp, origifp, m, dst, rt);
535         if (error) {
536                 in6_ifstat_inc(rt->rt_ifp, ifs6_out_discard);
537                 ip6stat.ip6s_cantforward++;
538         } else {
539                 ip6stat.ip6s_forward++;
540                 in6_ifstat_inc(rt->rt_ifp, ifs6_out_forward);
541                 if (type)
542                         ip6stat.ip6s_redirectsent++;
543                 else {
544                         if (mcopy)
545                                 goto freecopy;
546                 }
547         }
548         if (mcopy == NULL)
549                 return;
550         switch (error) {
551         case 0:
552 #if 1
553                 if (type == ND_REDIRECT) {
554                         icmp6_redirect_output(mcopy, rt);
555                         return;
556                 }
557 #endif
558                 goto freecopy;
559
560         case EMSGSIZE:
561                 /* xxx MTU is constant in PPP? */
562                 goto freecopy;
563
564         case ENOBUFS:
565                 /* Tell source to slow down like source quench in IP? */
566                 goto freecopy;
567
568         case ENETUNREACH:       /* shouldn't happen, checked above */
569         case EHOSTUNREACH:
570         case ENETDOWN:
571         case EHOSTDOWN:
572         default:
573                 type = ICMP6_DST_UNREACH;
574                 code = ICMP6_DST_UNREACH_ADDR;
575                 break;
576         }
577         icmp6_error(mcopy, type, code, 0);
578         return;
579
580  freecopy:
581         m_freem(mcopy);
582         return;
583 }