xref: /netbsd-src/sys/netinet6/ip6_forward.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: ip6_forward.c,v 1.35 2003/07/03 05:03:55 itojun Exp $	*/
2 /*	$KAME: ip6_forward.c,v 1.109 2002/09/11 08:10:17 sakane Exp $	*/
3 
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: ip6_forward.c,v 1.35 2003/07/03 05:03:55 itojun Exp $");
35 
36 #include "opt_ipsec.h"
37 #include "opt_pfil_hooks.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/ip_var.h>
57 #include <netinet/ip6.h>
58 #include <netinet6/ip6_var.h>
59 #include <netinet/icmp6.h>
60 #include <netinet6/nd6.h>
61 
62 #ifdef IPSEC
63 #include <netinet6/ipsec.h>
64 #include <netkey/key.h>
65 #endif /* IPSEC */
66 
67 #ifdef PFIL_HOOKS
68 #include <net/pfil.h>
69 #endif
70 
71 #include <net/net_osdep.h>
72 
73 struct	route_in6 ip6_forward_rt;
74 
75 #ifdef PFIL_HOOKS
76 extern struct pfil_head inet6_pfil_hook;	/* XXX */
77 #endif
78 
79 /*
80  * Forward a packet.  If some error occurs return the sender
81  * an icmp packet.  Note we can't always generate a meaningful
82  * icmp message because icmp doesn't have a large enough repertoire
83  * of codes and types.
84  *
85  * If not forwarding, just drop the packet.  This could be confusing
86  * if ipforwarding was zero but some routing protocol was advancing
87  * us as a gateway to somewhere.  However, we must let the routing
88  * protocol deal with that.
89  *
90  */
91 
92 void
93 ip6_forward(m, srcrt)
94 	struct mbuf *m;
95 	int srcrt;
96 {
97 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
98 	struct sockaddr_in6 *dst;
99 	struct rtentry *rt;
100 	int error, type = 0, code = 0;
101 	struct mbuf *mcopy = NULL;
102 	struct ifnet *origifp;	/* maybe unnecessary */
103 #ifdef IPSEC
104 	struct secpolicy *sp = NULL;
105 #endif
106 
107 #ifdef IPSEC
108 	/*
109 	 * Check AH/ESP integrity.
110 	 */
111 	/*
112 	 * Don't increment ip6s_cantforward because this is the check
113 	 * before forwarding packet actually.
114 	 */
115 	if (ipsec6_in_reject(m, NULL)) {
116 		ipsec6stat.in_polvio++;
117 		m_freem(m);
118 		return;
119 	}
120 #endif /* IPSEC */
121 
122 	/*
123 	 * Do not forward packets to multicast destination (should be handled
124 	 * by ip6_mforward().
125 	 * Do not forward packets with unspecified source.  It was discussed
126 	 * in July 2000, on ipngwg mailing list.
127 	 */
128 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0 ||
129 	    IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
130 	    IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
131 		ip6stat.ip6s_cantforward++;
132 		/* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard) */
133 		if (ip6_log_time + ip6_log_interval < time.tv_sec) {
134 			ip6_log_time = time.tv_sec;
135 			log(LOG_DEBUG,
136 			    "cannot forward "
137 			    "from %s to %s nxt %d received on %s\n",
138 			    ip6_sprintf(&ip6->ip6_src),
139 			    ip6_sprintf(&ip6->ip6_dst),
140 			    ip6->ip6_nxt,
141 			    if_name(m->m_pkthdr.rcvif));
142 		}
143 		m_freem(m);
144 		return;
145 	}
146 
147 	if (ip6->ip6_hlim <= IPV6_HLIMDEC) {
148 		/* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard) */
149 		icmp6_error(m, ICMP6_TIME_EXCEEDED,
150 				ICMP6_TIME_EXCEED_TRANSIT, 0);
151 		return;
152 	}
153 	ip6->ip6_hlim -= IPV6_HLIMDEC;
154 
155 	/*
156 	 * Save at most ICMPV6_PLD_MAXLEN (= the min IPv6 MTU -
157 	 * size of IPv6 + ICMPv6 headers) bytes of the packet in case
158 	 * we need to generate an ICMP6 message to the src.
159 	 * Thanks to M_EXT, in most cases copy will not occur.
160 	 *
161 	 * It is important to save it before IPsec processing as IPsec
162 	 * processing may modify the mbuf.
163 	 */
164 	mcopy = m_copy(m, 0, imin(m->m_pkthdr.len, ICMPV6_PLD_MAXLEN));
165 
166 #ifdef IPSEC
167 	/* get a security policy for this packet */
168 	sp = ipsec6_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND,
169 	    IP_FORWARDING, &error);
170 	if (sp == NULL) {
171 		ipsec6stat.out_inval++;
172 		ip6stat.ip6s_cantforward++;
173 		if (mcopy) {
174 #if 0
175 			/* XXX: what icmp ? */
176 #else
177 			m_freem(mcopy);
178 #endif
179 		}
180 		m_freem(m);
181 		return;
182 	}
183 
184 	error = 0;
185 
186 	/* check policy */
187 	switch (sp->policy) {
188 	case IPSEC_POLICY_DISCARD:
189 		/*
190 		 * This packet is just discarded.
191 		 */
192 		ipsec6stat.out_polvio++;
193 		ip6stat.ip6s_cantforward++;
194 		key_freesp(sp);
195 		if (mcopy) {
196 #if 0
197 			/* XXX: what icmp ? */
198 #else
199 			m_freem(mcopy);
200 #endif
201 		}
202 		m_freem(m);
203 		return;
204 
205 	case IPSEC_POLICY_BYPASS:
206 	case IPSEC_POLICY_NONE:
207 		/* no need to do IPsec. */
208 		key_freesp(sp);
209 		goto skip_ipsec;
210 
211 	case IPSEC_POLICY_IPSEC:
212 		if (sp->req == NULL) {
213 			/* XXX should be panic ? */
214 			printf("ip6_forward: No IPsec request specified.\n");
215 			ip6stat.ip6s_cantforward++;
216 			key_freesp(sp);
217 			if (mcopy) {
218 #if 0
219 				/* XXX: what icmp ? */
220 #else
221 				m_freem(mcopy);
222 #endif
223 			}
224 			m_freem(m);
225 			return;
226 		}
227 		/* do IPsec */
228 		break;
229 
230 	case IPSEC_POLICY_ENTRUST:
231 	default:
232 		/* should be panic ?? */
233 		printf("ip6_forward: Invalid policy found. %d\n", sp->policy);
234 		key_freesp(sp);
235 		goto skip_ipsec;
236 	}
237 
238     {
239 	struct ipsecrequest *isr = NULL;
240 	struct ipsec_output_state state;
241 
242 	/*
243 	 * when the kernel forwards a packet, it is not proper to apply
244 	 * IPsec transport mode to the packet is not proper.  this check
245 	 * avoid from this.
246 	 * at present, if there is even a transport mode SA request in the
247 	 * security policy, the kernel does not apply IPsec to the packet.
248 	 * this check is not enough because the following case is valid.
249 	 *      ipsec esp/tunnel/xxx-xxx/require esp/transport//require;
250 	 */
251 	for (isr = sp->req; isr; isr = isr->next) {
252 		if (isr->saidx.mode == IPSEC_MODE_TRANSPORT)
253 			goto skip_ipsec;
254 	}
255 
256 	/*
257 	 * All the extension headers will become inaccessible
258 	 * (since they can be encrypted).
259 	 * Don't panic, we need no more updates to extension headers
260 	 * on inner IPv6 packet (since they are now encapsulated).
261 	 *
262 	 * IPv6 [ESP|AH] IPv6 [extension headers] payload
263 	 */
264 	bzero(&state, sizeof(state));
265 	state.m = m;
266 	state.ro = NULL;	/* update at ipsec6_output_tunnel() */
267 	state.dst = NULL;	/* update at ipsec6_output_tunnel() */
268 
269 	error = ipsec6_output_tunnel(&state, sp, 0);
270 
271 	m = state.m;
272 #if 0	/* XXX allocate a route (ro, dst) again later */
273 	ro = (struct route_in6 *)state.ro;
274 	dst = (struct sockaddr_in6 *)state.dst;
275 #endif
276 	key_freesp(sp);
277 
278 	if (error) {
279 		/* mbuf is already reclaimed in ipsec6_output_tunnel. */
280 		switch (error) {
281 		case EHOSTUNREACH:
282 		case ENETUNREACH:
283 		case EMSGSIZE:
284 		case ENOBUFS:
285 		case ENOMEM:
286 			break;
287 		default:
288 			printf("ip6_output (ipsec): error code %d\n", error);
289 			/* FALLTHROUGH */
290 		case ENOENT:
291 			/* don't show these error codes to the user */
292 			break;
293 		}
294 		ip6stat.ip6s_cantforward++;
295 		if (mcopy) {
296 #if 0
297 			/* XXX: what icmp ? */
298 #else
299 			m_freem(mcopy);
300 #endif
301 		}
302 		m_freem(m);
303 		return;
304 	}
305     }
306     skip_ipsec:
307 #endif /* IPSEC */
308 
309 	dst = &ip6_forward_rt.ro_dst;
310 	if (!srcrt) {
311 		/*
312 		 * ip6_forward_rt.ro_dst.sin6_addr is equal to ip6->ip6_dst
313 		 */
314 		if (ip6_forward_rt.ro_rt == 0 ||
315 		    (ip6_forward_rt.ro_rt->rt_flags & RTF_UP) == 0) {
316 			if (ip6_forward_rt.ro_rt) {
317 				RTFREE(ip6_forward_rt.ro_rt);
318 				ip6_forward_rt.ro_rt = 0;
319 			}
320 			/* this probably fails but give it a try again */
321 			rtalloc((struct route *)&ip6_forward_rt);
322 		}
323 
324 		if (ip6_forward_rt.ro_rt == 0) {
325 			ip6stat.ip6s_noroute++;
326 			/* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_noroute) */
327 			if (mcopy) {
328 				icmp6_error(mcopy, ICMP6_DST_UNREACH,
329 					    ICMP6_DST_UNREACH_NOROUTE, 0);
330 			}
331 			m_freem(m);
332 			return;
333 		}
334 	} else if ((rt = ip6_forward_rt.ro_rt) == 0 ||
335 		 !IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &dst->sin6_addr)) {
336 		if (ip6_forward_rt.ro_rt) {
337 			RTFREE(ip6_forward_rt.ro_rt);
338 			ip6_forward_rt.ro_rt = 0;
339 		}
340 		bzero(dst, sizeof(*dst));
341 		dst->sin6_len = sizeof(struct sockaddr_in6);
342 		dst->sin6_family = AF_INET6;
343 		dst->sin6_addr = ip6->ip6_dst;
344 
345 		rtalloc((struct route *)&ip6_forward_rt);
346 		if (ip6_forward_rt.ro_rt == 0) {
347 			ip6stat.ip6s_noroute++;
348 			/* XXX in6_ifstat_inc(rt->rt_ifp, ifs6_in_noroute) */
349 			if (mcopy) {
350 				icmp6_error(mcopy, ICMP6_DST_UNREACH,
351 					    ICMP6_DST_UNREACH_NOROUTE, 0);
352 			}
353 			m_freem(m);
354 			return;
355 		}
356 	}
357 	rt = ip6_forward_rt.ro_rt;
358 
359 	/*
360 	 * Scope check: if a packet can't be delivered to its destination
361 	 * for the reason that the destination is beyond the scope of the
362 	 * source address, discard the packet and return an icmp6 destination
363 	 * unreachable error with Code 2 (beyond scope of source address).
364 	 * [draft-ietf-ipngwg-icmp-v3-00.txt, Section 3.1]
365 	 */
366 	if (in6_addr2scopeid(m->m_pkthdr.rcvif, &ip6->ip6_src) !=
367 	    in6_addr2scopeid(rt->rt_ifp, &ip6->ip6_src)) {
368 		ip6stat.ip6s_cantforward++;
369 		ip6stat.ip6s_badscope++;
370 		in6_ifstat_inc(rt->rt_ifp, ifs6_in_discard);
371 
372 		if (ip6_log_time + ip6_log_interval < time.tv_sec) {
373 			ip6_log_time = time.tv_sec;
374 			log(LOG_DEBUG,
375 			    "cannot forward "
376 			    "src %s, dst %s, nxt %d, rcvif %s, outif %s\n",
377 			    ip6_sprintf(&ip6->ip6_src),
378 			    ip6_sprintf(&ip6->ip6_dst),
379 			    ip6->ip6_nxt,
380 			    if_name(m->m_pkthdr.rcvif), if_name(rt->rt_ifp));
381 		}
382 		if (mcopy)
383 			icmp6_error(mcopy, ICMP6_DST_UNREACH,
384 				    ICMP6_DST_UNREACH_BEYONDSCOPE, 0);
385 		m_freem(m);
386 		return;
387 	}
388 
389 	if (m->m_pkthdr.len > IN6_LINKMTU(rt->rt_ifp)) {
390 		in6_ifstat_inc(rt->rt_ifp, ifs6_in_toobig);
391 		if (mcopy) {
392 			u_long mtu;
393 #ifdef IPSEC
394 			struct secpolicy *sp;
395 			int ipsecerror;
396 			size_t ipsechdrsiz;
397 #endif
398 
399 			mtu = IN6_LINKMTU(rt->rt_ifp);
400 #ifdef IPSEC
401 			/*
402 			 * When we do IPsec tunnel ingress, we need to play
403 			 * with the link value (decrement IPsec header size
404 			 * from mtu value).  The code is much simpler than v4
405 			 * case, as we have the outgoing interface for
406 			 * encapsulated packet as "rt->rt_ifp".
407 			 */
408 			sp = ipsec6_getpolicybyaddr(mcopy, IPSEC_DIR_OUTBOUND,
409 				IP_FORWARDING, &ipsecerror);
410 			if (sp) {
411 				ipsechdrsiz = ipsec6_hdrsiz(mcopy,
412 					IPSEC_DIR_OUTBOUND, NULL);
413 				if (ipsechdrsiz < mtu)
414 					mtu -= ipsechdrsiz;
415 			}
416 
417 			/*
418 			 * if mtu becomes less than minimum MTU,
419 			 * tell minimum MTU (and I'll need to fragment it).
420 			 */
421 			if (mtu < IPV6_MMTU)
422 				mtu = IPV6_MMTU;
423 #endif
424 			icmp6_error(mcopy, ICMP6_PACKET_TOO_BIG, 0, mtu);
425 		}
426 		m_freem(m);
427 		return;
428 	}
429 
430 	if (rt->rt_flags & RTF_GATEWAY)
431 		dst = (struct sockaddr_in6 *)rt->rt_gateway;
432 
433 	/*
434 	 * If we are to forward the packet using the same interface
435 	 * as one we got the packet from, perhaps we should send a redirect
436 	 * to sender to shortcut a hop.
437 	 * Only send redirect if source is sending directly to us,
438 	 * and if packet was not source routed (or has any options).
439 	 * Also, don't send redirect if forwarding using a route
440 	 * modified by a redirect.
441 	 */
442 	if (rt->rt_ifp == m->m_pkthdr.rcvif && !srcrt &&
443 	    (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0) {
444 		if ((rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
445 		    nd6_is_addr_neighbor((struct sockaddr_in6 *)&ip6_forward_rt.ro_dst, rt->rt_ifp)) {
446 			/*
447 			 * If the incoming interface is equal to the outgoing
448 			 * one, the link attached to the interface is
449 			 * point-to-point, and the IPv6 destination is
450 			 * regarded as on-link on the link, then it will be
451 			 * highly probable that the destination address does
452 			 * not exist on the link and that the packet is going
453 			 * to loop.  Thus, we immediately drop the packet and
454 			 * send an ICMPv6 error message.
455 			 * For other routing loops, we dare to let the packet
456 			 * go to the loop, so that a remote diagnosing host
457 			 * can detect the loop by traceroute.
458 			 * type/code is based on suggestion by Rich Draves.
459 			 * not sure if it is the best pick.
460 			 */
461 			icmp6_error(mcopy, ICMP6_DST_UNREACH,
462 				    ICMP6_DST_UNREACH_ADDR, 0);
463 			m_freem(m);
464 			return;
465 		}
466 		type = ND_REDIRECT;
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 	if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src))
507 		ip6->ip6_src.s6_addr16[1] = 0;
508 	if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst))
509 		ip6->ip6_dst.s6_addr16[1] = 0;
510 
511 #ifdef PFIL_HOOKS
512 	/*
513 	 * Run through list of hooks for output packets.
514 	 */
515 	if ((error = pfil_run_hooks(&inet6_pfil_hook, &m, rt->rt_ifp,
516 	    PFIL_OUT)) != 0)
517 		goto senderr;
518 	if (m == NULL)
519 		goto freecopy;
520 	ip6 = mtod(m, struct ip6_hdr *);
521 #endif /* PFIL_HOOKS */
522 
523 	error = nd6_output(rt->rt_ifp, origifp, m, dst, rt);
524 	if (error) {
525 		in6_ifstat_inc(rt->rt_ifp, ifs6_out_discard);
526 		ip6stat.ip6s_cantforward++;
527 	} else {
528 		ip6stat.ip6s_forward++;
529 		in6_ifstat_inc(rt->rt_ifp, ifs6_out_forward);
530 		if (type)
531 			ip6stat.ip6s_redirectsent++;
532 		else {
533 			if (mcopy)
534 				goto freecopy;
535 		}
536 	}
537 
538 #ifdef PFIL_HOOKS
539  senderr:
540 #endif
541 	if (mcopy == NULL)
542 		return;
543 	switch (error) {
544 	case 0:
545 		if (type == ND_REDIRECT) {
546 			icmp6_redirect_output(mcopy, rt);
547 			return;
548 		}
549 		goto freecopy;
550 
551 	case EMSGSIZE:
552 		/* xxx MTU is constant in PPP? */
553 		goto freecopy;
554 
555 	case ENOBUFS:
556 		/* Tell source to slow down like source quench in IP? */
557 		goto freecopy;
558 
559 	case ENETUNREACH:	/* shouldn't happen, checked above */
560 	case EHOSTUNREACH:
561 	case ENETDOWN:
562 	case EHOSTDOWN:
563 	default:
564 		type = ICMP6_DST_UNREACH;
565 		code = ICMP6_DST_UNREACH_ADDR;
566 		break;
567 	}
568 	icmp6_error(mcopy, type, code, 0);
569 	return;
570 
571  freecopy:
572 	m_freem(mcopy);
573 	return;
574 }
575