- Port rtw(4) from NetBSD, which supports various RealTek 8180 chip based
[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 1.11 2005/02/01 16:09:37 hrs 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 #include <net/pfil.h>
54
55 #include <netinet/in.h>
56 #include <netinet/in_var.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/ip.h>
59 #include <netinet/ip_var.h>
60 #include <netinet6/in6_var.h>
61 #include <netinet/ip6.h>
62 #include <netinet6/ip6_var.h>
63 #include <netinet/icmp6.h>
64 #include <netinet6/nd6.h>
65
66 #include <netinet/in_pcb.h>
67
68 #ifdef IPSEC
69 #include <netinet6/ipsec.h>
70 #ifdef INET6
71 #include <netinet6/ipsec6.h>
72 #endif
73 #include <netproto/key/key.h>
74 #endif /* IPSEC */
75
76 #ifdef FAST_IPSEC
77 #include <netproto/ipsec/ipsec.h>
78 #include <netproto/ipsec/ipsec6.h>
79 #include <netproto/ipsec/key.h>
80 #define IPSEC
81 #endif /* FAST_IPSEC */
82
83 #include <net/ip6fw/ip6_fw.h>
84
85 #include <net/net_osdep.h>
86
87 struct  route_in6 ip6_forward_rt;
88
89 /*
90  * Forward a packet.  If some error occurs return the sender
91  * an icmp packet.  Note we can't always generate a meaningful
92  * icmp message because icmp doesn't have a large enough repertoire
93  * of codes and types.
94  *
95  * If not forwarding, just drop the packet.  This could be confusing
96  * if ipforwarding was zero but some routing protocol was advancing
97  * us as a gateway to somewhere.  However, we must let the routing
98  * protocol deal with that.
99  *
100  */
101
102 void
103 ip6_forward(struct mbuf *m, int srcrt)
104 {
105         struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
106         struct sockaddr_in6 *dst;
107         struct rtentry *rt;
108         int error, type = 0, code = 0;
109         struct mbuf *mcopy = NULL;
110         struct ifnet *origifp;  /* maybe unnecessary */
111 #ifdef IPSEC
112         struct secpolicy *sp = NULL;
113 #endif
114
115 #ifdef IPSEC
116         /*
117          * Check AH/ESP integrity.
118          */
119         /*
120          * Don't increment ip6s_cantforward because this is the check
121          * before forwarding packet actually.
122          */
123         if (ipsec6_in_reject(m, NULL)) {
124 #if !defined(FAST_IPSEC)
125                 ipsec6stat.in_polvio++;
126 #endif
127                 m_freem(m);
128                 return;
129         }
130 #endif /* IPSEC */
131
132         /*
133          * Do not forward packets to multicast destination (should be handled
134          * by ip6_mforward().
135          * Do not forward packets with unspecified source.  It was discussed
136          * in July 2000, on ipngwg mailing list.
137          */
138         if ((m->m_flags & (M_BCAST|M_MCAST)) != 0 ||
139             IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
140             IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
141                 ip6stat.ip6s_cantforward++;
142                 /* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard) */
143                 if (ip6_log_time + ip6_log_interval < time_second) {
144                         ip6_log_time = time_second;
145                         log(LOG_DEBUG,
146                             "cannot forward "
147                             "from %s to %s nxt %d received on %s\n",
148                             ip6_sprintf(&ip6->ip6_src),
149                             ip6_sprintf(&ip6->ip6_dst),
150                             ip6->ip6_nxt,
151                             if_name(m->m_pkthdr.rcvif));
152                 }
153                 m_freem(m);
154                 return;
155         }
156
157         if (ip6->ip6_hlim <= IPV6_HLIMDEC) {
158                 /* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard) */
159                 icmp6_error(m, ICMP6_TIME_EXCEEDED,
160                                 ICMP6_TIME_EXCEED_TRANSIT, 0);
161                 return;
162         }
163         ip6->ip6_hlim -= IPV6_HLIMDEC;
164
165         /*
166          * Save at most ICMPV6_PLD_MAXLEN (= the min IPv6 MTU -
167          * size of IPv6 + ICMPv6 headers) bytes of the packet in case
168          * we need to generate an ICMP6 message to the src.
169          * Thanks to M_EXT, in most cases copy will not occur.
170          *
171          * It is important to save it before IPsec processing as IPsec
172          * processing may modify the mbuf.
173          */
174         mcopy = m_copy(m, 0, imin(m->m_pkthdr.len, ICMPV6_PLD_MAXLEN));
175
176 #ifdef IPSEC
177         /* get a security policy for this packet */
178         sp = ipsec6_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, IP_FORWARDING,
179             &error);
180         if (sp == NULL) {
181                 ipsec6stat.out_inval++;
182                 ip6stat.ip6s_cantforward++;
183                 if (mcopy) {
184 #if 0
185                         /* XXX: what icmp ? */
186 #else
187                         m_freem(mcopy);
188 #endif
189                 }
190                 m_freem(m);
191                 return;
192         }
193
194         error = 0;
195
196         /* check policy */
197         switch (sp->policy) {
198         case IPSEC_POLICY_DISCARD:
199                 /*
200                  * This packet is just discarded.
201                  */
202                 ipsec6stat.out_polvio++;
203                 ip6stat.ip6s_cantforward++;
204                 key_freesp(sp);
205                 if (mcopy) {
206 #if 0
207                         /* XXX: what icmp ? */
208 #else
209                         m_freem(mcopy);
210 #endif
211                 }
212                 m_freem(m);
213                 return;
214
215         case IPSEC_POLICY_BYPASS:
216         case IPSEC_POLICY_NONE:
217                 /* no need to do IPsec. */
218                 key_freesp(sp);
219                 goto skip_ipsec;
220
221         case IPSEC_POLICY_IPSEC:
222                 if (sp->req == NULL) {
223                         /* XXX should be panic ? */
224                         printf("ip6_forward: No IPsec request specified.\n");
225                         ip6stat.ip6s_cantforward++;
226                         key_freesp(sp);
227                         if (mcopy) {
228 #if 0
229                                 /* XXX: what icmp ? */
230 #else
231                                 m_freem(mcopy);
232 #endif
233                         }
234                         m_freem(m);
235                         return;
236                 }
237                 /* do IPsec */
238                 break;
239
240         case IPSEC_POLICY_ENTRUST:
241         default:
242                 /* should be panic ?? */
243                 printf("ip6_forward: Invalid policy found. %d\n", sp->policy);
244                 key_freesp(sp);
245                 goto skip_ipsec;
246         }
247
248     {
249         struct ipsec_output_state state;
250
251         /*
252          * All the extension headers will become inaccessible
253          * (since they can be encrypted).
254          * Don't panic, we need no more updates to extension headers
255          * on inner IPv6 packet (since they are now encapsulated).
256          *
257          * IPv6 [ESP|AH] IPv6 [extension headers] payload
258          */
259         bzero(&state, sizeof(state));
260         state.m = m;
261         state.ro = NULL;        /* update at ipsec6_output_tunnel() */
262         state.dst = NULL;       /* update at ipsec6_output_tunnel() */
263
264         error = ipsec6_output_tunnel(&state, sp, 0);
265
266         m = state.m;
267         key_freesp(sp);
268
269         if (error) {
270                 /* mbuf is already reclaimed in ipsec6_output_tunnel. */
271                 switch (error) {
272                 case EHOSTUNREACH:
273                 case ENETUNREACH:
274                 case EMSGSIZE:
275                 case ENOBUFS:
276                 case ENOMEM:
277                         break;
278                 default:
279                         printf("ip6_output (ipsec): error code %d\n", error);
280                         /* fall through */
281                 case ENOENT:
282                         /* don't show these error codes to the user */
283                         break;
284                 }
285                 ip6stat.ip6s_cantforward++;
286                 if (mcopy) {
287 #if 0
288                         /* XXX: what icmp ? */
289 #else
290                         m_freem(mcopy);
291 #endif
292                 }
293                 m_freem(m);
294                 return;
295         }
296     }
297     skip_ipsec:
298 #endif /* IPSEC */
299
300         dst = (struct sockaddr_in6 *)&ip6_forward_rt.ro_dst;
301         if (!srcrt) {
302                 /*
303                  * ip6_forward_rt.ro_dst.sin6_addr is equal to ip6->ip6_dst
304                  */
305                 if (ip6_forward_rt.ro_rt == 0 ||
306                     (ip6_forward_rt.ro_rt->rt_flags & RTF_UP) == 0) {
307                         if (ip6_forward_rt.ro_rt) {
308                                 RTFREE(ip6_forward_rt.ro_rt);
309                                 ip6_forward_rt.ro_rt = 0;
310                         }
311                         /* this probably fails but give it a try again */
312                         rtalloc_ign((struct route *)&ip6_forward_rt,
313                                     RTF_PRCLONING);
314                 }
315
316                 if (ip6_forward_rt.ro_rt == 0) {
317                         ip6stat.ip6s_noroute++;
318                         in6_ifstat_inc(m->m_pkthdr.rcvif, ifs6_in_noroute);
319                         if (mcopy) {
320                                 icmp6_error(mcopy, ICMP6_DST_UNREACH,
321                                             ICMP6_DST_UNREACH_NOROUTE, 0);
322                         }
323                         m_freem(m);
324                         return;
325                 }
326         } else if ((rt = ip6_forward_rt.ro_rt) == 0 ||
327                  !IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &dst->sin6_addr)) {
328                 if (ip6_forward_rt.ro_rt) {
329                         RTFREE(ip6_forward_rt.ro_rt);
330                         ip6_forward_rt.ro_rt = 0;
331                 }
332                 bzero(dst, sizeof(*dst));
333                 dst->sin6_len = sizeof(struct sockaddr_in6);
334                 dst->sin6_family = AF_INET6;
335                 dst->sin6_addr = ip6->ip6_dst;
336
337                 rtalloc_ign((struct route *)&ip6_forward_rt, RTF_PRCLONING);
338                 if (ip6_forward_rt.ro_rt == 0) {
339                         ip6stat.ip6s_noroute++;
340                         in6_ifstat_inc(m->m_pkthdr.rcvif, ifs6_in_noroute);
341                         if (mcopy) {
342                                 icmp6_error(mcopy, ICMP6_DST_UNREACH,
343                                             ICMP6_DST_UNREACH_NOROUTE, 0);
344                         }
345                         m_freem(m);
346                         return;
347                 }
348         }
349         rt = ip6_forward_rt.ro_rt;
350
351         /*
352          * Scope check: if a packet can't be delivered to its destination
353          * for the reason that the destination is beyond the scope of the
354          * source address, discard the packet and return an icmp6 destination
355          * unreachable error with Code 2 (beyond scope of source address).
356          * [draft-ietf-ipngwg-icmp-v3-02.txt, Section 3.1]
357          */
358         if (in6_addr2scopeid(m->m_pkthdr.rcvif, &ip6->ip6_src) !=
359             in6_addr2scopeid(rt->rt_ifp, &ip6->ip6_src)) {
360                 ip6stat.ip6s_cantforward++;
361                 ip6stat.ip6s_badscope++;
362                 in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard);
363
364                 if (ip6_log_time + ip6_log_interval < time_second) {
365                         ip6_log_time = time_second;
366                         log(LOG_DEBUG,
367                             "cannot forward "
368                             "src %s, dst %s, nxt %d, rcvif %s, outif %s\n",
369                             ip6_sprintf(&ip6->ip6_src),
370                             ip6_sprintf(&ip6->ip6_dst),
371                             ip6->ip6_nxt,
372                             if_name(m->m_pkthdr.rcvif), if_name(rt->rt_ifp));
373                 }
374                 if (mcopy)
375                         icmp6_error(mcopy, ICMP6_DST_UNREACH,
376                                     ICMP6_DST_UNREACH_BEYONDSCOPE, 0);
377                 m_freem(m);
378                 return;
379         }
380
381         if (m->m_pkthdr.len > rt->rt_ifp->if_mtu) {
382                 in6_ifstat_inc(rt->rt_ifp, ifs6_in_toobig);
383                 if (mcopy) {
384                         u_long mtu;
385 #ifdef IPSEC
386                         struct secpolicy *sp;
387                         int ipsecerror;
388                         size_t ipsechdrsiz;
389 #endif
390
391                         mtu = rt->rt_ifp->if_mtu;
392 #ifdef IPSEC
393                         /*
394                          * When we do IPsec tunnel ingress, we need to play
395                          * with if_mtu value (decrement IPsec header size
396                          * from mtu value).  The code is much simpler than v4
397                          * case, as we have the outgoing interface for
398                          * encapsulated packet as "rt->rt_ifp".
399                          */
400                         sp = ipsec6_getpolicybyaddr(mcopy, IPSEC_DIR_OUTBOUND,
401                                 IP_FORWARDING, &ipsecerror);
402                         if (sp) {
403                                 ipsechdrsiz = ipsec6_hdrsiz(mcopy,
404                                         IPSEC_DIR_OUTBOUND, NULL);
405                                 if (ipsechdrsiz < mtu)
406                                         mtu -= ipsechdrsiz;
407                         }
408
409                         /*
410                          * if mtu becomes less than minimum MTU,
411                          * tell minimum MTU (and I'll need to fragment it).
412                          */
413                         if (mtu < IPV6_MMTU)
414                                 mtu = IPV6_MMTU;
415 #endif
416                         icmp6_error(mcopy, ICMP6_PACKET_TOO_BIG, 0, mtu);
417                 }
418                 m_freem(m);
419                 return;
420         }
421
422         if (rt->rt_flags & RTF_GATEWAY)
423                 dst = (struct sockaddr_in6 *)rt->rt_gateway;
424
425         /*
426          * If we are to forward the packet using the same interface
427          * as one we got the packet from, perhaps we should send a redirect
428          * to sender to shortcut a hop.
429          * Only send redirect if source is sending directly to us,
430          * and if packet was not source routed (or has any options).
431          * Also, don't send redirect if forwarding using a route
432          * modified by a redirect.
433          */
434         if (rt->rt_ifp == m->m_pkthdr.rcvif && !srcrt &&
435             (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0) {
436                 if ((rt->rt_ifp->if_flags & IFF_POINTOPOINT) != 0) {
437                         /*
438                          * If the incoming interface is equal to the outgoing
439                          * one, and the link attached to the interface is
440                          * point-to-point, then it will be highly probable
441                          * that a routing loop occurs. Thus, we immediately
442                          * drop the packet and send an ICMPv6 error message.
443                          *
444                          * type/code is based on suggestion by Rich Draves.
445                          * not sure if it is the best pick.
446                          */
447                         icmp6_error(mcopy, ICMP6_DST_UNREACH,
448                                     ICMP6_DST_UNREACH_ADDR, 0);
449                         m_freem(m);
450                         return;
451                 }
452                 type = ND_REDIRECT;
453         }
454
455         /*
456          * Check with the firewall...
457          */
458         if (ip6_fw_enable && ip6_fw_chk_ptr) {
459                 u_short port = 0;
460                 /* If ipfw says divert, we have to just drop packet */
461                 if ((*ip6_fw_chk_ptr)(&ip6, rt->rt_ifp, &port, &m)) {
462                         m_freem(m);
463                         goto freecopy;
464                 }
465                 if (!m)
466                         goto freecopy;
467         }
468
469         /*
470          * Fake scoped addresses. Note that even link-local source or
471          * destinaion can appear, if the originating node just sends the
472          * packet to us (without address resolution for the destination).
473          * Since both icmp6_error and icmp6_redirect_output fill the embedded
474          * link identifiers, we can do this stuff after making a copy for
475          * returning an error.
476          */
477         if ((rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
478                 /*
479                  * See corresponding comments in ip6_output.
480                  * XXX: but is it possible that ip6_forward() sends a packet
481                  *      to a loopback interface? I don't think so, and thus
482                  *      I bark here. (jinmei@kame.net)
483                  * XXX: it is common to route invalid packets to loopback.
484                  *      also, the codepath will be visited on use of ::1 in
485                  *      rthdr. (itojun)
486                  */
487 #if 1
488                 if (0)
489 #else
490                 if ((rt->rt_flags & (RTF_BLACKHOLE|RTF_REJECT)) == 0)
491 #endif
492                 {
493                         printf("ip6_forward: outgoing interface is loopback. "
494                                 "src %s, dst %s, nxt %d, rcvif %s, outif %s\n",
495                                 ip6_sprintf(&ip6->ip6_src),
496                                 ip6_sprintf(&ip6->ip6_dst),
497                                 ip6->ip6_nxt, if_name(m->m_pkthdr.rcvif),
498                                 if_name(rt->rt_ifp));
499                 }
500
501                 /* we can just use rcvif in forwarding. */
502                 origifp = m->m_pkthdr.rcvif;
503         }
504         else
505                 origifp = rt->rt_ifp;
506         /*
507          * clear embedded scope identifiers if necessary.
508          * in6_clearscope will touch the addresses only when necessary.
509          */
510         in6_clearscope(&ip6->ip6_src);
511         in6_clearscope(&ip6->ip6_dst);
512
513         /*
514          * Run through list of hooks for output packets.
515          */
516         if (pfil_has_hooks(&inet6_pfil_hook)) {
517                 error = pfil_run_hooks(&inet6_pfil_hook, &m, 
518                                         rt->rt_ifp, PFIL_OUT);
519                 if (error != 0)
520                         goto senderr;
521                 if (m == NULL)
522                         goto freecopy;
523                 ip6 = mtod(m, struct ip6_hdr *);
524         }
525
526         error = nd6_output(rt->rt_ifp, origifp, m, dst, rt);
527         if (error) {
528                 in6_ifstat_inc(rt->rt_ifp, ifs6_out_discard);
529                 ip6stat.ip6s_cantforward++;
530         } else {
531                 ip6stat.ip6s_forward++;
532                 in6_ifstat_inc(rt->rt_ifp, ifs6_out_forward);
533                 if (type)
534                         ip6stat.ip6s_redirectsent++;
535                 else {
536                         if (mcopy)
537                                 goto freecopy;
538                 }
539         }
540 senderr:
541         if (mcopy == NULL)
542                 return;
543         switch (error) {
544         case 0:
545 #if 1
546                 if (type == ND_REDIRECT) {
547                         icmp6_redirect_output(mcopy, rt);
548                         return;
549                 }
550 #endif
551                 goto freecopy;
552
553         case EMSGSIZE:
554                 /* xxx MTU is constant in PPP? */
555                 goto freecopy;
556
557         case ENOBUFS:
558                 /* Tell source to slow down like source quench in IP? */
559                 goto freecopy;
560
561         case ENETUNREACH:       /* shouldn't happen, checked above */
562         case EHOSTUNREACH:
563         case ENETDOWN:
564         case EHOSTDOWN:
565         default:
566                 type = ICMP6_DST_UNREACH;
567                 code = ICMP6_DST_UNREACH_ADDR;
568                 break;
569         }
570         icmp6_error(mcopy, type, code, 0);
571         return;
572
573  freecopy:
574         m_freem(mcopy);
575         return;
576 }