xref: /freebsd/sys/netipsec/ipsec_output.c (revision b28624fde638caadd4a89f50c9b7e7da0f98c4d2)
1 /*-
2  * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 /*
30  * IPsec output processing.
31  */
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 #include "opt_ipsec.h"
35 #include "opt_enc.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/mbuf.h>
40 #include <sys/domain.h>
41 #include <sys/protosw.h>
42 #include <sys/socket.h>
43 #include <sys/errno.h>
44 #include <sys/syslog.h>
45 
46 #include <net/if.h>
47 #include <net/pfil.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_ecn.h>
56 #ifdef INET6
57 #include <netinet6/ip6_ecn.h>
58 #endif
59 
60 #include <netinet/ip6.h>
61 #ifdef INET6
62 #include <netinet6/ip6_var.h>
63 #endif
64 #include <netinet/in_pcb.h>
65 #ifdef INET6
66 #include <netinet/icmp6.h>
67 #endif
68 
69 #include <netipsec/ipsec.h>
70 #ifdef INET6
71 #include <netipsec/ipsec6.h>
72 #endif
73 #include <netipsec/ah_var.h>
74 #include <netipsec/esp_var.h>
75 #include <netipsec/ipcomp_var.h>
76 
77 #include <netipsec/xform.h>
78 
79 #include <netipsec/key.h>
80 #include <netipsec/keydb.h>
81 #include <netipsec/key_debug.h>
82 
83 #include <machine/in_cksum.h>
84 
85 int
86 ipsec_process_done(struct mbuf *m, struct ipsecrequest *isr)
87 {
88 	struct tdb_ident *tdbi;
89 	struct m_tag *mtag;
90 	struct secasvar *sav;
91 	struct secasindex *saidx;
92 	int error;
93 
94 	IPSEC_SPLASSERT_SOFTNET(__func__);
95 
96 	IPSEC_ASSERT(m != NULL, ("null mbuf"));
97 	IPSEC_ASSERT(isr != NULL, ("null ISR"));
98 	sav = isr->sav;
99 	IPSEC_ASSERT(sav != NULL, ("null SA"));
100 	IPSEC_ASSERT(sav->sah != NULL, ("null SAH"));
101 
102 	saidx = &sav->sah->saidx;
103 	switch (saidx->dst.sa.sa_family) {
104 #ifdef INET
105 	case AF_INET:
106 		/* Fix the header length, for AH processing. */
107 		mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
108 		break;
109 #endif /* INET */
110 #ifdef INET6
111 	case AF_INET6:
112 		/* Fix the header length, for AH processing. */
113 		if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
114 			error = ENXIO;
115 			goto bad;
116 		}
117 		if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
118 			/* No jumbogram support. */
119 			error = ENXIO;	/*?*/
120 			goto bad;
121 		}
122 		mtod(m, struct ip6_hdr *)->ip6_plen =
123 			htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
124 		break;
125 #endif /* INET6 */
126 	default:
127 		DPRINTF(("%s: unknown protocol family %u\n", __func__,
128 		    saidx->dst.sa.sa_family));
129 		error = ENXIO;
130 		goto bad;
131 	}
132 
133 	/*
134 	 * Add a record of what we've done or what needs to be done to the
135 	 * packet.
136 	 */
137 	mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE,
138 			sizeof(struct tdb_ident), M_NOWAIT);
139 	if (mtag == NULL) {
140 		DPRINTF(("%s: could not get packet tag\n", __func__));
141 		error = ENOMEM;
142 		goto bad;
143 	}
144 
145 	tdbi = (struct tdb_ident *)(mtag + 1);
146 	tdbi->dst = saidx->dst;
147 	tdbi->proto = saidx->proto;
148 	tdbi->spi = sav->spi;
149 	m_tag_prepend(m, mtag);
150 
151 	/*
152 	 * If there's another (bundled) SA to apply, do so.
153 	 * Note that this puts a burden on the kernel stack size.
154 	 * If this is a problem we'll need to introduce a queue
155 	 * to set the packet on so we can unwind the stack before
156 	 * doing further processing.
157 	 */
158 	if (isr->next) {
159 		ipsec4stat.ips_out_bundlesa++;
160 		return ipsec4_process_packet(m, isr->next, 0, 0);
161 	}
162 	key_sa_recordxfer(sav, m);		/* record data transfer */
163 
164 	/*
165 	 * We're done with IPsec processing, transmit the packet using the
166 	 * appropriate network protocol (IP or IPv6). SPD lookup will be
167 	 * performed again there.
168 	 */
169 	switch (saidx->dst.sa.sa_family) {
170 #ifdef INET
171 	struct ip *ip;
172 	case AF_INET:
173 		ip = mtod(m, struct ip *);
174 		ip->ip_len = ntohs(ip->ip_len);
175 		ip->ip_off = ntohs(ip->ip_off);
176 
177 		return ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
178 #endif /* INET */
179 #ifdef INET6
180 	case AF_INET6:
181 		/*
182 		 * We don't need massage, IPv6 header fields are always in
183 		 * net endian.
184 		 */
185 		return ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
186 #endif /* INET6 */
187 	}
188 	panic("ipsec_process_done");
189 bad:
190 	m_freem(m);
191 	KEY_FREESAV(&sav);
192 	return (error);
193 }
194 
195 static struct ipsecrequest *
196 ipsec_nextisr(
197 	struct mbuf *m,
198 	struct ipsecrequest *isr,
199 	int af,
200 	struct secasindex *saidx,
201 	int *error
202 )
203 {
204 #define IPSEC_OSTAT(x,y,z) (isr->saidx.proto == IPPROTO_ESP ? (x)++ : \
205 			    isr->saidx.proto == IPPROTO_AH ? (y)++ : (z)++)
206 	struct secasvar *sav;
207 
208 	IPSEC_SPLASSERT_SOFTNET(__func__);
209 	IPSECREQUEST_LOCK_ASSERT(isr);
210 
211 	IPSEC_ASSERT(af == AF_INET || af == AF_INET6,
212 		("invalid address family %u", af));
213 again:
214 	/*
215 	 * Craft SA index to search for proper SA.  Note that
216 	 * we only fillin unspecified SA peers for transport
217 	 * mode; for tunnel mode they must already be filled in.
218 	 */
219 	*saidx = isr->saidx;
220 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
221 		/* Fillin unspecified SA peers only for transport mode */
222 		if (af == AF_INET) {
223 			struct sockaddr_in *sin;
224 			struct ip *ip = mtod(m, struct ip *);
225 
226 			if (saidx->src.sa.sa_len == 0) {
227 				sin = &saidx->src.sin;
228 				sin->sin_len = sizeof(*sin);
229 				sin->sin_family = AF_INET;
230 				sin->sin_port = IPSEC_PORT_ANY;
231 				sin->sin_addr = ip->ip_src;
232 			}
233 			if (saidx->dst.sa.sa_len == 0) {
234 				sin = &saidx->dst.sin;
235 				sin->sin_len = sizeof(*sin);
236 				sin->sin_family = AF_INET;
237 				sin->sin_port = IPSEC_PORT_ANY;
238 				sin->sin_addr = ip->ip_dst;
239 			}
240 		} else {
241 			struct sockaddr_in6 *sin6;
242 			struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
243 
244 			if (saidx->src.sin6.sin6_len == 0) {
245 				sin6 = (struct sockaddr_in6 *)&saidx->src;
246 				sin6->sin6_len = sizeof(*sin6);
247 				sin6->sin6_family = AF_INET6;
248 				sin6->sin6_port = IPSEC_PORT_ANY;
249 				sin6->sin6_addr = ip6->ip6_src;
250 				if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
251 					/* fix scope id for comparing SPD */
252 					sin6->sin6_addr.s6_addr16[1] = 0;
253 					sin6->sin6_scope_id =
254 					    ntohs(ip6->ip6_src.s6_addr16[1]);
255 				}
256 			}
257 			if (saidx->dst.sin6.sin6_len == 0) {
258 				sin6 = (struct sockaddr_in6 *)&saidx->dst;
259 				sin6->sin6_len = sizeof(*sin6);
260 				sin6->sin6_family = AF_INET6;
261 				sin6->sin6_port = IPSEC_PORT_ANY;
262 				sin6->sin6_addr = ip6->ip6_dst;
263 				if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
264 					/* fix scope id for comparing SPD */
265 					sin6->sin6_addr.s6_addr16[1] = 0;
266 					sin6->sin6_scope_id =
267 					    ntohs(ip6->ip6_dst.s6_addr16[1]);
268 				}
269 			}
270 		}
271 	}
272 
273 	/*
274 	 * Lookup SA and validate it.
275 	 */
276 	*error = key_checkrequest(isr, saidx);
277 	if (*error != 0) {
278 		/*
279 		 * IPsec processing is required, but no SA found.
280 		 * I assume that key_acquire() had been called
281 		 * to get/establish the SA. Here I discard
282 		 * this packet because it is responsibility for
283 		 * upper layer to retransmit the packet.
284 		 */
285 		ipsec4stat.ips_out_nosa++;
286 		goto bad;
287 	}
288 	sav = isr->sav;
289 	if (sav == NULL) {		/* XXX valid return */
290 		IPSEC_ASSERT(ipsec_get_reqlevel(isr) == IPSEC_LEVEL_USE,
291 			("no SA found, but required; level %u",
292 			ipsec_get_reqlevel(isr)));
293 		IPSECREQUEST_UNLOCK(isr);
294 		isr = isr->next;
295 		if (isr == NULL) {
296 			/*XXXstatistic??*/
297 			*error = EINVAL;		/*XXX*/
298 			return isr;
299 		}
300 		IPSECREQUEST_LOCK(isr);
301 		goto again;
302 	}
303 
304 	/*
305 	 * Check system global policy controls.
306 	 */
307 	if ((isr->saidx.proto == IPPROTO_ESP && !esp_enable) ||
308 	    (isr->saidx.proto == IPPROTO_AH && !ah_enable) ||
309 	    (isr->saidx.proto == IPPROTO_IPCOMP && !ipcomp_enable)) {
310 		DPRINTF(("%s: IPsec outbound packet dropped due"
311 			" to policy (check your sysctls)\n", __func__));
312 		IPSEC_OSTAT(espstat.esps_pdrops, ahstat.ahs_pdrops,
313 		    ipcompstat.ipcomps_pdrops);
314 		*error = EHOSTUNREACH;
315 		goto bad;
316 	}
317 
318 	/*
319 	 * Sanity check the SA contents for the caller
320 	 * before they invoke the xform output method.
321 	 */
322 	if (sav->tdb_xform == NULL) {
323 		DPRINTF(("%s: no transform for SA\n", __func__));
324 		IPSEC_OSTAT(espstat.esps_noxform, ahstat.ahs_noxform,
325 		    ipcompstat.ipcomps_noxform);
326 		*error = EHOSTUNREACH;
327 		goto bad;
328 	}
329 	return isr;
330 bad:
331 	IPSEC_ASSERT(*error != 0, ("error return w/ no error code"));
332 	IPSECREQUEST_UNLOCK(isr);
333 	return NULL;
334 #undef IPSEC_OSTAT
335 }
336 
337 #ifdef INET
338 /*
339  * IPsec output logic for IPv4.
340  */
341 int
342 ipsec4_process_packet(
343 	struct mbuf *m,
344 	struct ipsecrequest *isr,
345 	int flags,
346 	int tunalready)
347 {
348 	struct secasindex saidx;
349 	struct secasvar *sav;
350 	struct ip *ip;
351 	int error, i, off;
352 
353 	IPSEC_ASSERT(m != NULL, ("null mbuf"));
354 	IPSEC_ASSERT(isr != NULL, ("null isr"));
355 
356 	IPSECREQUEST_LOCK(isr);		/* insure SA contents don't change */
357 
358 	isr = ipsec_nextisr(m, isr, AF_INET, &saidx, &error);
359 	if (isr == NULL)
360 		goto bad;
361 
362 	sav = isr->sav;
363 
364 #ifdef DEV_ENC
365 	/* pass the mbuf to enc0 for packet filtering */
366 	if ((error = ipsec_filter(&m, PFIL_OUT)) != 0)
367 		goto bad;
368 #endif
369 
370 	if (!tunalready) {
371 		union sockaddr_union *dst = &sav->sah->saidx.dst;
372 		int setdf;
373 
374 		/*
375 		 * Collect IP_DF state from the outer header.
376 		 */
377 		if (dst->sa.sa_family == AF_INET) {
378 			if (m->m_len < sizeof (struct ip) &&
379 			    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
380 				error = ENOBUFS;
381 				goto bad;
382 			}
383 			ip = mtod(m, struct ip *);
384 			/* Honor system-wide control of how to handle IP_DF */
385 			switch (ip4_ipsec_dfbit) {
386 			case 0:			/* clear in outer header */
387 			case 1:			/* set in outer header */
388 				setdf = ip4_ipsec_dfbit;
389 				break;
390 			default:		/* propagate to outer header */
391 				setdf = ntohs(ip->ip_off & IP_DF);
392 				break;
393 			}
394 		} else {
395 			ip = NULL;		/* keep compiler happy */
396 			setdf = 0;
397 		}
398 		/* Do the appropriate encapsulation, if necessary */
399 		if (isr->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
400 		    dst->sa.sa_family != AF_INET ||	    /* PF mismatch */
401 #if 0
402 		    (sav->flags & SADB_X_SAFLAGS_TUNNEL) || /* Tunnel requ'd */
403 		    sav->tdb_xform->xf_type == XF_IP4 ||    /* ditto */
404 #endif
405 		    (dst->sa.sa_family == AF_INET &&	    /* Proxy */
406 		     dst->sin.sin_addr.s_addr != INADDR_ANY &&
407 		     dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
408 			struct mbuf *mp;
409 
410 			/* Fix IPv4 header checksum and length */
411 			if (m->m_len < sizeof (struct ip) &&
412 			    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
413 				error = ENOBUFS;
414 				goto bad;
415 			}
416 			ip = mtod(m, struct ip *);
417 			ip->ip_len = htons(m->m_pkthdr.len);
418 			ip->ip_sum = 0;
419 #ifdef _IP_VHL
420 			if (ip->ip_vhl == IP_VHL_BORING)
421 				ip->ip_sum = in_cksum_hdr(ip);
422 			else
423 				ip->ip_sum = in_cksum(m,
424 					_IP_VHL_HL(ip->ip_vhl) << 2);
425 #else
426 			ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
427 #endif
428 
429 			/* Encapsulate the packet */
430 			error = ipip_output(m, isr, &mp, 0, 0);
431 			if (mp == NULL && !error) {
432 				/* Should never happen. */
433 				DPRINTF(("%s: ipip_output returns no mbuf and "
434 					"no error!", __func__));
435 				error = EFAULT;
436 			}
437 			if (error) {
438 				if (mp) {
439 					/* XXX: Should never happen! */
440 					m_freem(mp);
441 				}
442 				m = NULL; /* ipip_output() already freed it */
443 				goto bad;
444 			}
445 			m = mp, mp = NULL;
446 			/*
447 			 * ipip_output clears IP_DF in the new header.  If
448 			 * we need to propagate IP_DF from the outer header,
449 			 * then we have to do it here.
450 			 *
451 			 * XXX shouldn't assume what ipip_output does.
452 			 */
453 			if (dst->sa.sa_family == AF_INET && setdf) {
454 				if (m->m_len < sizeof (struct ip) &&
455 				    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
456 					error = ENOBUFS;
457 					goto bad;
458 				}
459 				ip = mtod(m, struct ip *);
460 				ip->ip_off = ntohs(ip->ip_off);
461 				ip->ip_off |= IP_DF;
462 				ip->ip_off = htons(ip->ip_off);
463 			}
464 		}
465 	}
466 
467 #ifdef DEV_ENC
468 	/* pass the mbuf to enc0 for bpf processing */
469 	ipsec_bpf(m, sav, AF_INET);
470 #endif
471 
472 	/*
473 	 * Dispatch to the appropriate IPsec transform logic.  The
474 	 * packet will be returned for transmission after crypto
475 	 * processing, etc. are completed.  For encapsulation we
476 	 * bypass this call because of the explicit call done above
477 	 * (necessary to deal with IP_DF handling for IPv4).
478 	 *
479 	 * NB: m & sav are ``passed to caller'' who's reponsible for
480 	 *     for reclaiming their resources.
481 	 */
482 	if (sav->tdb_xform->xf_type != XF_IP4) {
483 		ip = mtod(m, struct ip *);
484 		i = ip->ip_hl << 2;
485 		off = offsetof(struct ip, ip_p);
486 		error = (*sav->tdb_xform->xf_output)(m, isr, NULL, i, off);
487 	} else {
488 		error = ipsec_process_done(m, isr);
489 	}
490 	IPSECREQUEST_UNLOCK(isr);
491 	return error;
492 bad:
493 	if (isr)
494 		IPSECREQUEST_UNLOCK(isr);
495 	if (m)
496 		m_freem(m);
497 	return error;
498 }
499 #endif
500 
501 #ifdef INET6
502 /*
503  * Chop IP6 header from the payload.
504  */
505 static struct mbuf *
506 ipsec6_splithdr(struct mbuf *m)
507 {
508 	struct mbuf *mh;
509 	struct ip6_hdr *ip6;
510 	int hlen;
511 
512 	IPSEC_ASSERT(m->m_len >= sizeof (struct ip6_hdr),
513 		("first mbuf too short, len %u", m->m_len));
514 	ip6 = mtod(m, struct ip6_hdr *);
515 	hlen = sizeof(struct ip6_hdr);
516 	if (m->m_len > hlen) {
517 		MGETHDR(mh, M_DONTWAIT, MT_DATA);
518 		if (!mh) {
519 			m_freem(m);
520 			return NULL;
521 		}
522 		M_MOVE_PKTHDR(mh, m);
523 		MH_ALIGN(mh, hlen);
524 		m->m_len -= hlen;
525 		m->m_data += hlen;
526 		mh->m_next = m;
527 		m = mh;
528 		m->m_len = hlen;
529 		bcopy((caddr_t)ip6, mtod(m, caddr_t), hlen);
530 	} else if (m->m_len < hlen) {
531 		m = m_pullup(m, hlen);
532 		if (!m)
533 			return NULL;
534 	}
535 	return m;
536 }
537 
538 /*
539  * IPsec output logic for IPv6, transport mode.
540  */
541 int
542 ipsec6_output_trans(
543 	struct ipsec_output_state *state,
544 	u_char *nexthdrp,
545 	struct mbuf *mprev,
546 	struct secpolicy *sp,
547 	int flags,
548 	int *tun)
549 {
550 	struct ipsecrequest *isr;
551 	struct secasindex saidx;
552 	int error = 0;
553 	struct mbuf *m;
554 
555 	IPSEC_ASSERT(state != NULL, ("null state"));
556 	IPSEC_ASSERT(state->m != NULL, ("null m"));
557 	IPSEC_ASSERT(nexthdrp != NULL, ("null nexthdrp"));
558 	IPSEC_ASSERT(mprev != NULL, ("null mprev"));
559 	IPSEC_ASSERT(sp != NULL, ("null sp"));
560 	IPSEC_ASSERT(tun != NULL, ("null tun"));
561 
562 	KEYDEBUG(KEYDEBUG_IPSEC_DATA,
563 		printf("%s: applied SP\n", __func__);
564 		kdebug_secpolicy(sp));
565 
566 	isr = sp->req;
567 	if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
568 		/* the rest will be handled by ipsec6_output_tunnel() */
569 		*tun = 1;		/* need tunnel-mode processing */
570 		return 0;
571 	}
572 
573 	*tun = 0;
574 	m = state->m;
575 
576 	IPSECREQUEST_LOCK(isr);		/* insure SA contents don't change */
577 	isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
578 	if (isr == NULL) {
579 #ifdef notdef
580 		/* XXX should notification be done for all errors ? */
581 		/*
582 		 * Notify the fact that the packet is discarded
583 		 * to ourselves. I believe this is better than
584 		 * just silently discarding. (jinmei@kame.net)
585 		 * XXX: should we restrict the error to TCP packets?
586 		 * XXX: should we directly notify sockets via
587 		 *      pfctlinputs?
588 		 */
589 		icmp6_error(m, ICMP6_DST_UNREACH,
590 			    ICMP6_DST_UNREACH_ADMIN, 0);
591 		m = NULL;	/* NB: icmp6_error frees mbuf */
592 #endif
593 		goto bad;
594 	}
595 
596 	error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
597 						  sizeof (struct ip6_hdr),
598 						  offsetof(struct ip6_hdr,
599 							   ip6_nxt));
600 	IPSECREQUEST_UNLOCK(isr);
601 	return error;
602 bad:
603 	if (isr)
604 		IPSECREQUEST_UNLOCK(isr);
605 	if (m)
606 		m_freem(m);
607 	state->m = NULL;
608 	return error;
609 }
610 
611 static int
612 ipsec6_encapsulate(struct mbuf *m, struct secasvar *sav)
613 {
614 	struct ip6_hdr *oip6;
615 	struct ip6_hdr *ip6;
616 	size_t plen;
617 
618 	/* can't tunnel between different AFs */
619 	if (sav->sah->saidx.src.sa.sa_family != AF_INET6 ||
620 	    sav->sah->saidx.dst.sa.sa_family != AF_INET6) {
621 		m_freem(m);
622 		return EINVAL;
623 	}
624 	IPSEC_ASSERT(m->m_len == sizeof (struct ip6_hdr),
625 		("mbuf wrong size; len %u", m->m_len));
626 
627 
628 	/*
629 	 * grow the mbuf to accomodate the new IPv6 header.
630 	 */
631 	plen = m->m_pkthdr.len;
632 	if (M_LEADINGSPACE(m->m_next) < sizeof(struct ip6_hdr)) {
633 		struct mbuf *n;
634 		MGET(n, M_DONTWAIT, MT_DATA);
635 		if (!n) {
636 			m_freem(m);
637 			return ENOBUFS;
638 		}
639 		n->m_len = sizeof(struct ip6_hdr);
640 		n->m_next = m->m_next;
641 		m->m_next = n;
642 		m->m_pkthdr.len += sizeof(struct ip6_hdr);
643 		oip6 = mtod(n, struct ip6_hdr *);
644 	} else {
645 		m->m_next->m_len += sizeof(struct ip6_hdr);
646 		m->m_next->m_data -= sizeof(struct ip6_hdr);
647 		m->m_pkthdr.len += sizeof(struct ip6_hdr);
648 		oip6 = mtod(m->m_next, struct ip6_hdr *);
649 	}
650 	ip6 = mtod(m, struct ip6_hdr *);
651 	bcopy((caddr_t)ip6, (caddr_t)oip6, sizeof(struct ip6_hdr));
652 
653 	/* Fake link-local scope-class addresses */
654 	if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_src))
655 		oip6->ip6_src.s6_addr16[1] = 0;
656 	if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_dst))
657 		oip6->ip6_dst.s6_addr16[1] = 0;
658 
659 	/* construct new IPv6 header. see RFC 2401 5.1.2.2 */
660 	/* ECN consideration. */
661 	ip6_ecn_ingress(ip6_ipsec_ecn, &ip6->ip6_flow, &oip6->ip6_flow);
662 	if (plen < IPV6_MAXPACKET - sizeof(struct ip6_hdr))
663 		ip6->ip6_plen = htons(plen);
664 	else {
665 		/* ip6->ip6_plen will be updated in ip6_output() */
666 	}
667 	ip6->ip6_nxt = IPPROTO_IPV6;
668 	ip6->ip6_src = sav->sah->saidx.src.sin6.sin6_addr;
669 	ip6->ip6_dst = sav->sah->saidx.dst.sin6.sin6_addr;
670 	ip6->ip6_hlim = IPV6_DEFHLIM;
671 
672 	/* XXX Should ip6_src be updated later ? */
673 
674 	return 0;
675 }
676 
677 /*
678  * IPsec output logic for IPv6, tunnel mode.
679  */
680 int
681 ipsec6_output_tunnel(struct ipsec_output_state *state, struct secpolicy *sp, int flags)
682 {
683 	struct ip6_hdr *ip6;
684 	struct ipsecrequest *isr;
685 	struct secasindex saidx;
686 	int error;
687 	struct sockaddr_in6* dst6;
688 	struct mbuf *m;
689 
690 	IPSEC_ASSERT(state != NULL, ("null state"));
691 	IPSEC_ASSERT(state->m != NULL, ("null m"));
692 	IPSEC_ASSERT(sp != NULL, ("null sp"));
693 
694 	KEYDEBUG(KEYDEBUG_IPSEC_DATA,
695 		printf("%s: applied SP\n", __func__);
696 		kdebug_secpolicy(sp));
697 
698 	m = state->m;
699 	/*
700 	 * transport mode ipsec (before the 1st tunnel mode) is already
701 	 * processed by ipsec6_output_trans().
702 	 */
703 	for (isr = sp->req; isr; isr = isr->next) {
704 		if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
705 			break;
706 	}
707 
708 	IPSECREQUEST_LOCK(isr);		/* insure SA contents don't change */
709 	isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
710 	if (isr == NULL)
711 		goto bad;
712 
713 	/*
714 	 * There may be the case that SA status will be changed when
715 	 * we are refering to one. So calling splsoftnet().
716 	 */
717 	if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
718 		/*
719 		 * build IPsec tunnel.
720 		 */
721 		/* XXX should be processed with other familiy */
722 		if (isr->sav->sah->saidx.src.sa.sa_family != AF_INET6) {
723 			ipseclog((LOG_ERR, "%s: family mismatched between "
724 			    "inner and outer, spi=%u\n", __func__,
725 			    ntohl(isr->sav->spi)));
726 			ipsec6stat.ips_out_inval++;
727 			error = EAFNOSUPPORT;
728 			goto bad;
729 		}
730 
731 		m = ipsec6_splithdr(m);
732 		if (!m) {
733 			ipsec6stat.ips_out_nomem++;
734 			error = ENOMEM;
735 			goto bad;
736 		}
737 		error = ipsec6_encapsulate(m, isr->sav);
738 		if (error) {
739 			m = NULL;
740 			goto bad;
741 		}
742 		ip6 = mtod(m, struct ip6_hdr *);
743 
744 		state->ro = &isr->sav->sah->sa_route;
745 		state->dst = (struct sockaddr *)&state->ro->ro_dst;
746 		dst6 = (struct sockaddr_in6 *)state->dst;
747 		if (state->ro->ro_rt
748 		 && ((state->ro->ro_rt->rt_flags & RTF_UP) == 0
749 		  || !IN6_ARE_ADDR_EQUAL(&dst6->sin6_addr, &ip6->ip6_dst))) {
750 			RTFREE(state->ro->ro_rt);
751 			state->ro->ro_rt = NULL;
752 		}
753 		if (state->ro->ro_rt == 0) {
754 			bzero(dst6, sizeof(*dst6));
755 			dst6->sin6_family = AF_INET6;
756 			dst6->sin6_len = sizeof(*dst6);
757 			dst6->sin6_addr = ip6->ip6_dst;
758 			rtalloc(state->ro);
759 		}
760 		if (state->ro->ro_rt == 0) {
761 			ip6stat.ip6s_noroute++;
762 			ipsec6stat.ips_out_noroute++;
763 			error = EHOSTUNREACH;
764 			goto bad;
765 		}
766 
767 		/* adjust state->dst if tunnel endpoint is offlink */
768 		if (state->ro->ro_rt->rt_flags & RTF_GATEWAY) {
769 			state->dst = (struct sockaddr *)state->ro->ro_rt->rt_gateway;
770 			dst6 = (struct sockaddr_in6 *)state->dst;
771 		}
772 	}
773 
774 	m = ipsec6_splithdr(m);
775 	if (!m) {
776 		ipsec6stat.ips_out_nomem++;
777 		error = ENOMEM;
778 		goto bad;
779 	}
780 	ip6 = mtod(m, struct ip6_hdr *);
781 	error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
782 		sizeof (struct ip6_hdr),
783 		offsetof(struct ip6_hdr, ip6_nxt));
784 	IPSECREQUEST_UNLOCK(isr);
785 	return error;
786 bad:
787 	if (isr)
788 		IPSECREQUEST_UNLOCK(isr);
789 	if (m)
790 		m_freem(m);
791 	state->m = NULL;
792 	return error;
793 }
794 #endif /*INET6*/
795