xref: /netbsd-src/sys/netipsec/ipsecif.c (revision 404ee5b9334f618040b6cdef96a0ff35a6fc4636)
1 /*	$NetBSD: ipsecif.c,v 1.18 2019/11/01 04:28:14 knakahara Exp $  */
2 
3 /*
4  * Copyright (c) 2017 Internet Initiative Japan Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __KERNEL_RCSID(0, "$NetBSD: ipsecif.c,v 1.18 2019/11/01 04:28:14 knakahara Exp $");
31 
32 #ifdef _KERNEL_OPT
33 #include "opt_inet.h"
34 #include "opt_ipsec.h"
35 #endif
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/socket.h>
40 #include <sys/sockio.h>
41 #include <sys/mbuf.h>
42 #include <sys/errno.h>
43 #include <sys/ioctl.h>
44 #include <sys/syslog.h>
45 #include <sys/kernel.h>
46 
47 #include <net/if.h>
48 #include <net/route.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/ip.h>
53 #include <netinet/ip_var.h>
54 #include <netinet/in_var.h>
55 #include <netinet/ip_encap.h>
56 #include <netinet/ip_ecn.h>
57 #include <netinet/ip_private.h>
58 #include <netinet/udp.h>
59 
60 #ifdef INET6
61 #include <netinet/ip6.h>
62 #include <netinet6/ip6_var.h>
63 #include <netinet6/ip6_private.h>
64 #include <netinet6/in6_var.h>
65 #include <netinet6/ip6protosw.h> /* for struct ip6ctlparam */
66 #include <netinet/ip_ecn.h>
67 #endif
68 
69 #include <netipsec/key.h>
70 #include <netipsec/ipsecif.h>
71 
72 #include <net/if_ipsec.h>
73 
74 static int ipsecif_set_natt_ports(struct ipsec_variant *, struct mbuf *);
75 static void ipsecif4_input(struct mbuf *, int, int, void *);
76 static int ipsecif4_output(struct ipsec_variant *, int, struct mbuf *);
77 static int ipsecif4_filter4(const struct ip *, struct ipsec_variant *,
78 	struct ifnet *);
79 
80 #ifdef INET6
81 static int ipsecif6_input(struct mbuf **, int *, int, void *);
82 static int ipsecif6_output(struct ipsec_variant *, int, struct mbuf *);
83 static int ipsecif6_filter6(const struct ip6_hdr *, struct ipsec_variant *,
84 	struct ifnet *);
85 #endif
86 
87 static int ip_ipsec_ttl = IPSEC_TTL;
88 static int ip_ipsec_copy_tos = 0;
89 #ifdef INET6
90 int ip6_ipsec_hlim = IPSEC_HLIM;
91 int ip6_ipsec_pmtu = 0;
92 static int ip6_ipsec_copy_tos = 0;
93 #endif
94 
95 static const struct encapsw ipsecif4_encapsw = {
96 	.encapsw4 = {
97 		.pr_input = ipsecif4_input,
98 		.pr_ctlinput = NULL,
99 	}
100 };
101 
102 #ifdef INET6
103 static const struct encapsw ipsecif6_encapsw;
104 #endif
105 
106 static int
107 ipsecif_set_natt_ports(struct ipsec_variant *var, struct mbuf *m)
108 {
109 
110 	KASSERT(if_ipsec_heldref_variant(var));
111 
112 	if (var->iv_sport || var->iv_dport) {
113 		struct m_tag *mtag;
114 
115 		mtag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
116 		    sizeof(uint16_t) + sizeof(uint16_t), M_DONTWAIT);
117 		if (mtag) {
118 			uint16_t *natt_port;
119 
120 			natt_port = (uint16_t *)(mtag + 1);
121 			natt_port[0] = var->iv_dport;
122 			natt_port[1] = var->iv_sport;
123 			m_tag_prepend(m, mtag);
124 		} else {
125 			return ENOBUFS;
126 		}
127 	}
128 
129 	return 0;
130 }
131 
132 static struct mbuf *
133 ipsecif4_prepend_hdr(struct ipsec_variant *var, struct mbuf *m,
134     uint8_t proto, uint8_t tos)
135 {
136 	struct ip *ip;
137 	struct sockaddr_in *src, *dst;
138 
139 	src = satosin(var->iv_psrc);
140 	dst = satosin(var->iv_pdst);
141 
142 	if (in_nullhost(src->sin_addr) || in_nullhost(src->sin_addr) ||
143 	    src->sin_addr.s_addr == INADDR_BROADCAST ||
144 	    dst->sin_addr.s_addr == INADDR_BROADCAST) {
145 		m_freem(m);
146 		return NULL;
147 	}
148 	m->m_flags &= ~M_BCAST;
149 
150 	if (IN_MULTICAST(src->sin_addr.s_addr) ||
151 	    IN_MULTICAST(dst->sin_addr.s_addr)) {
152 		m_freem(m);
153 		return NULL;
154 	}
155 
156 	M_PREPEND(m, sizeof(struct ip), M_DONTWAIT);
157 	if (m && M_UNWRITABLE(m, sizeof(struct ip)))
158 		m = m_pullup(m, sizeof(struct ip));
159 	if (m == NULL)
160 		return NULL;
161 
162 	ip = mtod(m, struct ip *);
163 	ip->ip_v = IPVERSION;
164 	ip->ip_off = htons(0);
165 	if (m->m_pkthdr.len < IP_MINFRAGSIZE)
166 		ip->ip_id = 0;
167 	else
168 		ip->ip_id = ip_newid(NULL);
169 	ip->ip_hl = sizeof(*ip) >> 2;
170 	if (ip_ipsec_copy_tos)
171 		ip->ip_tos = tos;
172 	else
173 		ip->ip_tos = 0;
174 	ip->ip_sum = 0;
175 	ip->ip_src = src->sin_addr;
176 	ip->ip_dst = dst->sin_addr;
177 	ip->ip_p = proto;
178 	ip->ip_ttl = ip_ipsec_ttl;
179 	ip->ip_len = htons(m->m_pkthdr.len);
180 #ifndef IPSEC_TX_TOS_CLEAR
181 	struct ifnet *ifp = &var->iv_softc->ipsec_if;
182 	if (ifp->if_flags & IFF_ECN)
183 		ip_ecn_ingress(ECN_ALLOWED, &ip->ip_tos, &tos);
184 	else
185 		ip_ecn_ingress(ECN_NOCARE, &ip->ip_tos, &tos);
186 #endif
187 
188 	return m;
189 }
190 
191 static int
192 ipsecif4_needfrag(struct mbuf *m, struct ipsecrequest *isr)
193 {
194 	struct ip ip0;
195 	struct ip *ip;
196 	int mtu;
197 	struct secasvar *sav;
198 
199 	sav = key_lookup_sa_bysaidx(&isr->saidx);
200 	if (sav == NULL)
201 		return 0;
202 
203 	if (!(sav->natt_type & UDP_ENCAP_ESPINUDP)) {
204 		mtu = 0;
205 		goto out;
206 	}
207 
208 	if (m->m_len < sizeof(struct ip)) {
209 		m_copydata(m, 0, sizeof(ip0), &ip0);
210 		ip = &ip0;
211 	} else {
212 		ip = mtod(m, struct ip *);
213 	}
214 	mtu = sav->esp_frag;
215 	if (ntohs(ip->ip_len) <= mtu)
216 		mtu = 0;
217 
218 out:
219 	KEY_SA_UNREF(&sav);
220 	return mtu;
221 }
222 
223 static struct mbuf *
224 ipsecif4_flowinfo(struct mbuf *m, int family, int *proto0, u_int8_t *tos0)
225 {
226 	const struct ip *ip;
227 	int proto;
228 	int tos;
229 
230 	KASSERT(proto0 != NULL);
231 	KASSERT(tos0 != NULL);
232 
233 	switch (family) {
234 	case AF_INET:
235 		proto = IPPROTO_IPV4;
236 		if (m->m_len < sizeof(*ip)) {
237 			m = m_pullup(m, sizeof(*ip));
238 			if (m == NULL) {
239 				*tos0 = 0;
240 				*proto0 = 0;
241 				return NULL;
242 			}
243 		}
244 		ip = mtod(m, const struct ip *);
245 		tos = ip->ip_tos;
246 		/* TODO: support ALTQ for innner packet */
247 		break;
248 #ifdef INET6
249 	case AF_INET6: {
250 		const struct ip6_hdr *ip6;
251 		proto = IPPROTO_IPV6;
252 		if (m->m_len < sizeof(*ip6)) {
253 			m = m_pullup(m, sizeof(*ip6));
254 			if (m == NULL) {
255 				*tos0 = 0;
256 				*proto0 = 0;
257 				return NULL;
258 			}
259 		}
260 		ip6 = mtod(m, const struct ip6_hdr *);
261 		tos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
262 		/* TODO: support ALTQ for innner packet */
263 		break;
264 	}
265 #endif /* INET6 */
266 	default:
267 		*tos0 = 0;
268 		*proto0 = 0;
269 		return NULL;
270 	}
271 
272 	*proto0 = proto;
273 	*tos0 = tos;
274 	return m;
275 }
276 
277 static int
278 ipsecif4_fragout(struct ipsec_variant *var, int family, struct mbuf *m, int mtu)
279 {
280 	struct ifnet *ifp = &var->iv_softc->ipsec_if;
281 	struct mbuf *next;
282 	struct m_tag *mtag;
283 	int error;
284 
285 	KASSERT(if_ipsec_heldref_variant(var));
286 
287 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_NAT_T_PORTS);
288 	if (mtag)
289 		m_tag_delete(m, mtag);
290 
291 	/* consider new IP header prepended in ipsecif4_output() */
292 	if (mtu <= sizeof(struct ip)) {
293 		m_freem(m);
294 		return ENETUNREACH;
295 	}
296 	m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
297 	error = ip_fragment(m, ifp, mtu - sizeof(struct ip));
298 	if (error)
299 		return error;
300 
301 	for (error = 0; m; m = next) {
302 		next = m->m_nextpkt;
303 		m->m_nextpkt = NULL;
304 		if (error) {
305 			m_freem(m);
306 			continue;
307 		}
308 
309 		error = ipsecif4_output(var, family, m);
310 	}
311 	if (error == 0)
312 		IP_STATINC(IP_STAT_FRAGMENTED);
313 
314 	return error;
315 }
316 
317 int
318 ipsecif4_encap_func(struct mbuf *m, struct ip *ip, struct ipsec_variant *var)
319 {
320 	struct m_tag *mtag;
321 	struct sockaddr_in *src, *dst;
322 	u_int16_t src_port = 0;
323 	u_int16_t dst_port = 0;
324 
325 	KASSERT(var != NULL);
326 
327 	src = satosin(var->iv_psrc);
328 	dst = satosin(var->iv_pdst);
329 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_NAT_T_PORTS);
330 	if (mtag) {
331 		u_int16_t *ports;
332 
333 		ports = (u_int16_t *)(mtag + 1);
334 		src_port = ports[0];
335 		dst_port = ports[1];
336 	}
337 
338 	/* address match */
339 	if (src->sin_addr.s_addr != ip->ip_dst.s_addr ||
340 	    dst->sin_addr.s_addr != ip->ip_src.s_addr)
341 		return 0;
342 
343 	/* UDP encap? */
344 	if (mtag == NULL && var->iv_sport == 0 && var->iv_dport == 0)
345 		goto match;
346 
347 	/* port match */
348 	if (src_port != var->iv_dport ||
349 	    dst_port != var->iv_sport) {
350 #ifdef DEBUG
351 		printf("%s: port mismatch: pkt(%u, %u), if(%u, %u)\n",
352 		    __func__, ntohs(src_port), ntohs(dst_port),
353 		    ntohs(var->iv_sport), ntohs(var->iv_dport));
354 #endif
355 		return 0;
356 	}
357 
358 match:
359 	/*
360 	 * hide NAT-T information from encapsulated traffics.
361 	 * they don't know about IPsec.
362 	 */
363 	if (mtag)
364 		m_tag_delete(m, mtag);
365 	return sizeof(src->sin_addr) + sizeof(dst->sin_addr);
366 }
367 
368 static int
369 ipsecif4_output(struct ipsec_variant *var, int family, struct mbuf *m)
370 {
371 	struct secpolicy *sp = NULL;
372 	u_int8_t tos;
373 	int proto;
374 	int error;
375 	int mtu;
376 	u_long sa_mtu = 0;
377 
378 	KASSERT(if_ipsec_heldref_variant(var));
379 	KASSERT(if_ipsec_variant_is_configured(var));
380 	KASSERT(var->iv_psrc->sa_family == AF_INET);
381 	KASSERT(var->iv_pdst->sa_family == AF_INET);
382 
383 	sp = IV_SP_OUT(var);
384 	KASSERT(sp != NULL);
385 	/*
386 	 * The SPs in ipsec_variant are prevented from freed by
387 	 * ipsec_variant->iv_psref. So, KEY_SP_REF() is unnecessary here.
388 	 */
389 
390 	KASSERT(sp->policy != IPSEC_POLICY_NONE);
391 	KASSERT(sp->policy != IPSEC_POLICY_ENTRUST);
392 	KASSERT(sp->policy != IPSEC_POLICY_BYPASS);
393 	if (sp->policy != IPSEC_POLICY_IPSEC) {
394 		m_freem(m);
395 		error = ENETUNREACH;
396 		goto done;
397 	}
398 
399 	/* get flowinfo */
400 	m = ipsecif4_flowinfo(m, family, &proto, &tos);
401 	if (m == NULL) {
402 		error = ENETUNREACH;
403 		goto done;
404 	}
405 
406 	/* prepend new IP header */
407 	m = ipsecif4_prepend_hdr(var, m, proto, tos);
408 	if (m == NULL) {
409 		error = ENETUNREACH;
410 		goto done;
411 	}
412 
413 	/*
414 	 * Normal netipsec's NAT-T fragmentation is done in ip_output().
415 	 * See "natt_frag" processing.
416 	 * However, ipsec(4) interface's one is not done in the same way,
417 	 * so we must do NAT-T fragmentation by own code.
418 	 */
419 	/* NAT-T ESP fragmentation */
420 	mtu = ipsecif4_needfrag(m, sp->req);
421 	if (mtu > 0)
422 		return ipsecif4_fragout(var, family, m, mtu);
423 
424 	/* set NAT-T ports */
425 	error = ipsecif_set_natt_ports(var, m);
426 	if (error) {
427 		m_freem(m);
428 		goto done;
429 	}
430 
431 	/* IPsec output */
432 	IP_STATINC(IP_STAT_LOCALOUT);
433 	error = ipsec4_process_packet(m, sp->req, &sa_mtu);
434 	if (error == ENOENT)
435 		error = 0;
436 	/*
437 	 * frangmentation is already done in ipsecif4_fragout(),
438 	 * so ipsec4_process_packet() must not do fragmentation here.
439 	 */
440 	KASSERT(sa_mtu == 0);
441 
442 done:
443 	return error;
444 }
445 
446 #ifdef INET6
447 int
448 ipsecif6_encap_func(struct mbuf *m, struct ip6_hdr *ip6, struct ipsec_variant *var)
449 {
450 	struct m_tag *mtag;
451 	struct sockaddr_in6 *src, *dst;
452 	u_int16_t src_port = 0;
453 	u_int16_t dst_port = 0;
454 
455 	KASSERT(var != NULL);
456 
457 	src = satosin6(var->iv_psrc);
458 	dst = satosin6(var->iv_pdst);
459 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_NAT_T_PORTS);
460 	if (mtag) {
461 		u_int16_t *ports;
462 
463 		ports = (u_int16_t *)(mtag + 1);
464 		src_port = ports[0];
465 		dst_port = ports[1];
466 	}
467 
468 	/* address match */
469 	if (!IN6_ARE_ADDR_EQUAL(&src->sin6_addr, &ip6->ip6_dst) ||
470 	    !IN6_ARE_ADDR_EQUAL(&dst->sin6_addr, &ip6->ip6_src))
471 		return 0;
472 
473 	/* UDP encap? */
474 	if (mtag == NULL && var->iv_sport == 0 && var->iv_dport == 0)
475 		goto match;
476 
477 	/* port match */
478 	if (src_port != var->iv_dport ||
479 	    dst_port != var->iv_sport) {
480 #ifdef DEBUG
481 		printf("%s: port mismatch: pkt(%u, %u), if(%u, %u)\n",
482 		    __func__, ntohs(src_port), ntohs(dst_port),
483 		    ntohs(var->iv_sport), ntohs(var->iv_dport));
484 #endif
485 		return 0;
486 	}
487 
488 match:
489 	/*
490 	 * hide NAT-T information from encapsulated traffics.
491 	 * they don't know about IPsec.
492 	 */
493 	if (mtag)
494 		m_tag_delete(m, mtag);
495 	return sizeof(src->sin6_addr) + sizeof(dst->sin6_addr);
496 }
497 
498 static int
499 ipsecif6_output(struct ipsec_variant *var, int family, struct mbuf *m)
500 {
501 	struct ifnet *ifp = &var->iv_softc->ipsec_if;
502 	struct ipsec_softc *sc = ifp->if_softc;
503 	struct route *ro_pc;
504 	kmutex_t *lock_pc;
505 	struct rtentry *rt;
506 	struct sockaddr_in6 *sin6_src;
507 	struct sockaddr_in6 *sin6_dst;
508 	struct ip6_hdr *ip6;
509 	int proto, error, flags;
510 	u_int8_t itos, otos;
511 	union {
512 		struct sockaddr		dst;
513 		struct sockaddr_in6	dst6;
514 	} u;
515 
516 	KASSERT(if_ipsec_heldref_variant(var));
517 	KASSERT(if_ipsec_variant_is_configured(var));
518 
519 	sin6_src = satosin6(var->iv_psrc);
520 	sin6_dst = satosin6(var->iv_pdst);
521 
522 	KASSERT(sin6_src->sin6_family == AF_INET6);
523 	KASSERT(sin6_dst->sin6_family == AF_INET6);
524 
525 	switch (family) {
526 #ifdef INET
527 	case AF_INET:
528 	    {
529 		struct ip *ip;
530 
531 		proto = IPPROTO_IPV4;
532 		if (m->m_len < sizeof(*ip)) {
533 			m = m_pullup(m, sizeof(*ip));
534 			if (m == NULL)
535 				return ENOBUFS;
536 		}
537 		ip = mtod(m, struct ip *);
538 		itos = ip->ip_tos;
539 		/* TODO: support ALTQ for innner packet */
540 		break;
541 	    }
542 #endif /* INET */
543 	case AF_INET6:
544 	    {
545 		struct ip6_hdr *xip6;
546 		proto = IPPROTO_IPV6;
547 		if (m->m_len < sizeof(*xip6)) {
548 			m = m_pullup(m, sizeof(*xip6));
549 			if (m == NULL)
550 				return ENOBUFS;
551 		}
552 		xip6 = mtod(m, struct ip6_hdr *);
553 		itos = (ntohl(xip6->ip6_flow) >> 20) & 0xff;
554 		/* TODO: support ALTQ for innner packet */
555 		break;
556 	    }
557 	default:
558 		m_freem(m);
559 		return EAFNOSUPPORT;
560 	}
561 
562 	/* prepend new IP header */
563 	M_PREPEND(m, sizeof(struct ip6_hdr), M_DONTWAIT);
564 	if (m && M_UNWRITABLE(m, sizeof(struct ip6_hdr)))
565 		m = m_pullup(m, sizeof(struct ip6_hdr));
566 	if (m == NULL)
567 		return ENOBUFS;
568 
569 	ip6 = mtod(m, struct ip6_hdr *);
570 	ip6->ip6_flow	= 0;
571 	ip6->ip6_vfc	&= ~IPV6_VERSION_MASK;
572 	ip6->ip6_vfc	|= IPV6_VERSION;
573 #if 0	/* ip6->ip6_plen will be filled by ip6_output */
574 	ip6->ip6_plen	= htons((u_short)m->m_pkthdr.len - sizeof(*ip6));
575 #endif
576 	ip6->ip6_nxt	= proto;
577 	ip6->ip6_hlim	= ip6_ipsec_hlim;
578 	ip6->ip6_src	= sin6_src->sin6_addr;
579 	/* bidirectional configured tunnel mode */
580 	if (!IN6_IS_ADDR_UNSPECIFIED(&sin6_dst->sin6_addr)) {
581 		ip6->ip6_dst = sin6_dst->sin6_addr;
582 	} else  {
583 		m_freem(m);
584 		return ENETUNREACH;
585 	}
586 #ifndef IPSEC_TX_TOS_CLEAR
587 	if (!ip6_ipsec_copy_tos)
588 		otos = 0;
589 
590 	if (ifp->if_flags & IFF_ECN)
591 		ip_ecn_ingress(ECN_ALLOWED, &otos, &itos);
592 	else
593 		ip_ecn_ingress(ECN_NOCARE, &otos, &itos);
594 #else
595 	if (ip6_ipsec_copy_tos)
596 		otos = itos;
597 	else
598 		otos = 0;
599 #endif
600 	ip6->ip6_flow &= ~ntohl(0xff00000);
601 	ip6->ip6_flow |= htonl((u_int32_t)otos << 20);
602 
603 	sockaddr_in6_init(&u.dst6, &sin6_dst->sin6_addr, 0, 0, 0);
604 
605 	if_tunnel_get_ro(sc->ipsec_ro_percpu, &ro_pc, &lock_pc);
606 	if ((rt = rtcache_lookup(ro_pc, &u.dst)) == NULL) {
607 		if_tunnel_put_ro(sc->ipsec_ro_percpu, lock_pc);
608 		m_freem(m);
609 		return ENETUNREACH;
610 	}
611 
612 	if (rt->rt_ifp == ifp) {
613 		rtcache_unref(rt, ro_pc);
614 		rtcache_free(ro_pc);
615 		if_tunnel_put_ro(sc->ipsec_ro_percpu, lock_pc);
616 		m_freem(m);
617 		return ENETUNREACH;
618 	}
619 	rtcache_unref(rt, ro_pc);
620 
621 	/* set NAT-T ports */
622 	error = ipsecif_set_natt_ports(var, m);
623 	if (error) {
624 		m_freem(m);
625 		goto out;
626 	}
627 
628 	/*
629 	 * - IPSEC_PMTU_MINMTU
630 	 *   Force fragmentation to minimum MTU to avoid path MTU discovery
631 	 * - IPSEC_PMTU_OUTERMTU
632 	 *   Trust outer MTU is large enough to send all packets
633 	 *
634 	 * It is too painful to ask for resend of inner packet, to achieve
635 	 * path MTU discovery for encapsulated packets.
636 	 *
637 	 * See RFC4459.
638 	 */
639 	if (sc->ipsec_pmtu == IPSEC_PMTU_SYSDEFAULT) {
640 		switch (ip6_ipsec_pmtu) {
641 		case IPSEC_PMTU_MINMTU:
642 			flags = IPV6_MINMTU;
643 			break;
644 		case IPSEC_PMTU_OUTERMTU:
645 			flags = 0;
646 			break;
647 		default:
648 #ifdef DEBUG
649 			log(LOG_DEBUG, "%s: ignore unexpected ip6_ipsec_pmtu %d\n",
650 			    __func__, ip6_ipsec_pmtu);
651 #endif
652 			flags = IPV6_MINMTU;
653 			break;
654 		}
655 	} else {
656 		switch (sc->ipsec_pmtu) {
657 		case IPSEC_PMTU_MINMTU:
658 			flags = IPV6_MINMTU;
659 			break;
660 		case IPSEC_PMTU_OUTERMTU:
661 			flags = 0;
662 			break;
663 		default:
664 #ifdef DEBUG
665 			log(LOG_DEBUG, "%s: ignore unexpected ipsec_pmtu of %s %d\n",
666 			    __func__, ifp->if_xname, sc->ipsec_pmtu);
667 #endif
668 			flags = IPV6_MINMTU;
669 			break;
670 		}
671 	}
672 	error = ip6_output(m, 0, ro_pc, flags, 0, NULL, NULL);
673 
674 out:
675 	if (error)
676 		rtcache_free(ro_pc);
677 	if_tunnel_put_ro(sc->ipsec_ro_percpu, lock_pc);
678 
679 	return error;
680 }
681 #endif /* INET6 */
682 
683 static void
684 ipsecif4_input(struct mbuf *m, int off, int proto, void *eparg)
685 {
686 	struct ifnet *ipsecp;
687 	struct ipsec_softc *sc = eparg;
688 	struct ipsec_variant *var;
689 	const struct ip *ip;
690 	int af;
691 #ifndef IPSEC_TX_TOS_CLEAR
692 	u_int8_t otos;
693 #endif
694 	struct psref psref_rcvif;
695 	struct psref psref_var;
696 	struct ifnet *rcvif;
697 
698 	KASSERT(sc != NULL);
699 
700 	ipsecp = &sc->ipsec_if;
701 	if ((ipsecp->if_flags & IFF_UP) == 0) {
702 		m_freem(m);
703 		ip_statinc(IP_STAT_NOIPSEC);
704 		return;
705 	}
706 
707 	var = if_ipsec_getref_variant(sc, &psref_var);
708 	if (if_ipsec_variant_is_unconfigured(var)) {
709 		if_ipsec_putref_variant(var, &psref_var);
710 		m_freem(m);
711 		ip_statinc(IP_STAT_NOIPSEC);
712 		return;
713 	}
714 
715 	ip = mtod(m, const struct ip *);
716 
717 	rcvif = m_get_rcvif_psref(m, &psref_rcvif);
718 	if (rcvif == NULL || !ipsecif4_filter4(ip, var, rcvif)) {
719 		m_put_rcvif_psref(rcvif, &psref_rcvif);
720 		if_ipsec_putref_variant(var, &psref_var);
721 		m_freem(m);
722 		ip_statinc(IP_STAT_NOIPSEC);
723 		return;
724 	}
725 	m_put_rcvif_psref(rcvif, &psref_rcvif);
726 	if_ipsec_putref_variant(var, &psref_var);
727 #ifndef IPSEC_TX_TOS_CLEAR
728 	otos = ip->ip_tos;
729 #endif
730 	m_adj(m, off);
731 
732 	switch (proto) {
733 	case IPPROTO_IPV4:
734 	    {
735 		struct ip *xip;
736 		af = AF_INET;
737 		if (M_UNWRITABLE(m, sizeof(*xip))) {
738 			m = m_pullup(m, sizeof(*xip));
739 			if (m == NULL)
740 				return;
741 		}
742 		xip = mtod(m, struct ip *);
743 #ifndef IPSEC_TX_TOS_CLEAR
744 		if (ipsecp->if_flags & IFF_ECN)
745 			ip_ecn_egress(ECN_ALLOWED, &otos, &xip->ip_tos);
746 		else
747 			ip_ecn_egress(ECN_NOCARE, &otos, &xip->ip_tos);
748 #endif
749 		break;
750 	    }
751 #ifdef INET6
752 	case IPPROTO_IPV6:
753 	    {
754 		struct ip6_hdr *ip6;
755 		u_int8_t itos;
756 		af = AF_INET6;
757 		if (M_UNWRITABLE(m, sizeof(*ip6))) {
758 			m = m_pullup(m, sizeof(*ip6));
759 			if (m == NULL)
760 				return;
761 		}
762 		ip6 = mtod(m, struct ip6_hdr *);
763 		itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
764 #ifndef IPSEC_TX_TOS_CLEAR
765 		if (ipsecp->if_flags & IFF_ECN)
766 			ip_ecn_egress(ECN_ALLOWED, &otos, &itos);
767 		else
768 			ip_ecn_egress(ECN_NOCARE, &otos, &itos);
769 #endif
770 		ip6->ip6_flow &= ~htonl(0xff << 20);
771 		ip6->ip6_flow |= htonl((u_int32_t)itos << 20);
772 		break;
773 	    }
774 #endif /* INET6 */
775 	default:
776 		ip_statinc(IP_STAT_NOIPSEC);
777 		m_freem(m);
778 		return;
779 	}
780 	if_ipsec_input(m, af, ipsecp);
781 
782 	return;
783 }
784 
785 /*
786  * validate and filter the packet
787  */
788 static int
789 ipsecif4_filter4(const struct ip *ip, struct ipsec_variant *var,
790     struct ifnet *ifp)
791 {
792 	struct sockaddr_in *src, *dst;
793 
794 	src = satosin(var->iv_psrc);
795 	dst = satosin(var->iv_pdst);
796 
797 	return in_tunnel_validate(ip, src->sin_addr, dst->sin_addr);
798 }
799 
800 #ifdef INET6
801 static int
802 ipsecif6_input(struct mbuf **mp, int *offp, int proto, void *eparg)
803 {
804 	struct mbuf *m = *mp;
805 	struct ifnet *ipsecp;
806 	struct ipsec_softc *sc = eparg;
807 	struct ipsec_variant *var;
808 	struct ip6_hdr *ip6;
809 	int af = 0;
810 #ifndef IPSEC_TX_TOS_CLEAR
811 	u_int32_t otos;
812 #endif
813 	struct psref psref_rcvif;
814 	struct psref psref_var;
815 	struct ifnet *rcvif;
816 
817 	KASSERT(eparg != NULL);
818 
819 	ipsecp = &sc->ipsec_if;
820 	if ((ipsecp->if_flags & IFF_UP) == 0) {
821 		m_freem(m);
822 		IP6_STATINC(IP6_STAT_NOIPSEC);
823 		return IPPROTO_DONE;
824 	}
825 
826 	var = if_ipsec_getref_variant(sc, &psref_var);
827 	if (if_ipsec_variant_is_unconfigured(var)) {
828 		if_ipsec_putref_variant(var, &psref_var);
829 		m_freem(m);
830 		IP6_STATINC(IP6_STAT_NOIPSEC);
831 		return IPPROTO_DONE;
832 	}
833 
834 	ip6 = mtod(m, struct ip6_hdr *);
835 
836 	rcvif = m_get_rcvif_psref(m, &psref_rcvif);
837 	if (rcvif == NULL || !ipsecif6_filter6(ip6, var, rcvif)) {
838 		m_put_rcvif_psref(rcvif, &psref_rcvif);
839 		if_ipsec_putref_variant(var, &psref_var);
840 		m_freem(m);
841 		IP6_STATINC(IP6_STAT_NOIPSEC);
842 		return IPPROTO_DONE;
843 	}
844 	m_put_rcvif_psref(rcvif, &psref_rcvif);
845 	if_ipsec_putref_variant(var, &psref_var);
846 
847 #ifndef IPSEC_TX_TOS_CLEAR
848 	otos = ip6->ip6_flow;
849 #endif
850 	m_adj(m, *offp);
851 
852 	switch (proto) {
853 #ifdef INET
854 	case IPPROTO_IPV4:
855 	    {
856 		af = AF_INET;
857 #ifndef IPSEC_TX_TOS_CLEAR
858 		struct ip *ip;
859 		u_int8_t otos8;
860 		otos8 = (ntohl(otos) >> 20) & 0xff;
861 
862 		if (M_UNWRITABLE(m, sizeof(*ip))) {
863 			m = m_pullup(m, sizeof(*ip));
864 			if (m == NULL)
865 				return IPPROTO_DONE;
866 		}
867 		ip = mtod(m, struct ip *);
868 		if (ipsecp->if_flags & IFF_ECN)
869 			ip_ecn_egress(ECN_ALLOWED, &otos8, &ip->ip_tos);
870 		else
871 			ip_ecn_egress(ECN_NOCARE, &otos8, &ip->ip_tos);
872 #endif
873 		break;
874 	    }
875 #endif /* INET */
876 	case IPPROTO_IPV6:
877 	    {
878 		af = AF_INET6;
879 #ifndef IPSEC_TX_TOS_CLEAR
880 		struct ip6_hdr *xip6;
881 
882 		if (M_UNWRITABLE(m, sizeof(*xip6))) {
883 			m = m_pullup(m, sizeof(*xip6));
884 			if (m == NULL)
885 				return IPPROTO_DONE;
886 		}
887 		xip6 = mtod(m, struct ip6_hdr *);
888 		if (ipsecp->if_flags & IFF_ECN)
889 			ip6_ecn_egress(ECN_ALLOWED, &otos, &xip6->ip6_flow);
890 		else
891 			ip6_ecn_egress(ECN_NOCARE, &otos, &xip6->ip6_flow);
892 		break;
893 #endif
894 	    }
895 	default:
896 		IP6_STATINC(IP6_STAT_NOIPSEC);
897 		m_freem(m);
898 		return IPPROTO_DONE;
899 	}
900 
901 	if_ipsec_input(m, af, ipsecp);
902 	return IPPROTO_DONE;
903 }
904 
905 /*
906  * validate and filter the packet.
907  */
908 static int
909 ipsecif6_filter6(const struct ip6_hdr *ip6, struct ipsec_variant *var,
910     struct ifnet *ifp)
911 {
912 	struct sockaddr_in6 *src, *dst;
913 
914 	src = satosin6(var->iv_psrc);
915 	dst = satosin6(var->iv_pdst);
916 
917 	return in6_tunnel_validate(ip6, &src->sin6_addr, &dst->sin6_addr);
918 }
919 #endif /* INET6 */
920 
921 int
922 ipsecif4_attach(struct ipsec_variant *var)
923 {
924 	struct ipsec_softc *sc = var->iv_softc;
925 
926 	KASSERT(if_ipsec_variant_is_configured(var));
927 
928 	if (var->iv_encap_cookie4 != NULL)
929 		return EALREADY;
930 	var->iv_encap_cookie4 = encap_attach_func(AF_INET, -1, if_ipsec_encap_func,
931 	    &ipsecif4_encapsw, sc);
932 	if (var->iv_encap_cookie4 == NULL)
933 		return EEXIST;
934 
935 	var->iv_output = ipsecif4_output;
936 	return 0;
937 }
938 
939 int
940 ipsecif4_detach(struct ipsec_variant *var)
941 {
942 	int error;
943 
944 	if (var->iv_encap_cookie4 == NULL)
945 		return 0;
946 
947 	var->iv_output = NULL;
948 	error = encap_detach(var->iv_encap_cookie4);
949 	if (error == 0)
950 		var->iv_encap_cookie4 = NULL;
951 
952 	return error;
953 }
954 
955 #ifdef INET6
956 int
957 ipsecif6_attach(struct ipsec_variant *var)
958 {
959 	struct ipsec_softc *sc = var->iv_softc;
960 
961 	KASSERT(if_ipsec_variant_is_configured(var));
962 	KASSERT(var->iv_encap_cookie6 == NULL);
963 
964 	var->iv_encap_cookie6 = encap_attach_func(AF_INET6, -1, if_ipsec_encap_func,
965 	    &ipsecif6_encapsw, sc);
966 	if (var->iv_encap_cookie6 == NULL)
967 		return EEXIST;
968 
969 	var->iv_output = ipsecif6_output;
970 	return 0;
971 }
972 
973 int
974 ipsecif6_detach(struct ipsec_variant *var)
975 {
976 	struct ipsec_softc *sc = var->iv_softc;
977 	int error;
978 
979 	KASSERT(var->iv_encap_cookie6 != NULL);
980 
981 	if_tunnel_ro_percpu_rtcache_free(sc->ipsec_ro_percpu);
982 
983 	var->iv_output = NULL;
984 	error = encap_detach(var->iv_encap_cookie6);
985 	if (error == 0)
986 		var->iv_encap_cookie6 = NULL;
987 	return error;
988 }
989 
990 void *
991 ipsecif6_ctlinput(int cmd, const struct sockaddr *sa, void *d, void *eparg)
992 {
993 	struct ipsec_softc *sc = eparg;
994 	struct ip6ctlparam *ip6cp = NULL;
995 	struct ip6_hdr *ip6;
996 	const struct sockaddr_in6 *dst6;
997 	struct route *ro_pc;
998 	kmutex_t *lock_pc;
999 
1000 	if (sa->sa_family != AF_INET6 ||
1001 	    sa->sa_len != sizeof(struct sockaddr_in6))
1002 		return NULL;
1003 
1004 	if ((unsigned)cmd >= PRC_NCMDS)
1005 		return NULL;
1006 	if (cmd == PRC_HOSTDEAD)
1007 		d = NULL;
1008 	else if (inet6ctlerrmap[cmd] == 0)
1009 		return NULL;
1010 
1011 	/* if the parameter is from icmp6, decode it. */
1012 	if (d != NULL) {
1013 		ip6cp = (struct ip6ctlparam *)d;
1014 		ip6 = ip6cp->ip6c_ip6;
1015 	} else {
1016 		ip6 = NULL;
1017 	}
1018 
1019 	if (!ip6)
1020 		return NULL;
1021 
1022 	if_tunnel_get_ro(sc->ipsec_ro_percpu, &ro_pc, &lock_pc);
1023 	dst6 = satocsin6(rtcache_getdst(ro_pc));
1024 	/* XXX scope */
1025 	if (dst6 == NULL)
1026 		;
1027 	else if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &dst6->sin6_addr))
1028 		/* flush route cache */
1029 		rtcache_free(ro_pc);
1030 
1031 	if_tunnel_put_ro(sc->ipsec_ro_percpu, lock_pc);
1032 
1033 	return NULL;
1034 }
1035 
1036 ENCAP_PR_WRAP_CTLINPUT(ipsecif6_ctlinput)
1037 #define	ipsecif6_ctlinput	ipsecif6_ctlinput_wrapper
1038 
1039 static const struct encapsw ipsecif6_encapsw = {
1040 	.encapsw6 = {
1041 		.pr_input = ipsecif6_input,
1042 		.pr_ctlinput = ipsecif6_ctlinput,
1043 	}
1044 };
1045 #endif /* INET6 */
1046