xref: /freebsd/sys/netipsec/ipsec_output.c (revision fcf596178b5f2be36424ecbc1b6a3224b29c91d2)
1 /*-
2  * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
3  * Copyright (c) 2016 Andrey V. Elsukov <ae@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD$
28  */
29 
30 /*
31  * IPsec output processing.
32  */
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_ipsec.h"
36 #include "opt_sctp.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/mbuf.h>
41 #include <sys/domain.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/errno.h>
45 #include <sys/hhook.h>
46 #include <sys/syslog.h>
47 
48 #include <net/if.h>
49 #include <net/if_enc.h>
50 #include <net/if_var.h>
51 #include <net/vnet.h>
52 
53 #include <netinet/in.h>
54 #include <netinet/in_systm.h>
55 #include <netinet/ip.h>
56 #include <netinet/ip_var.h>
57 #include <netinet/in_var.h>
58 #include <netinet/ip_ecn.h>
59 #ifdef INET6
60 #include <netinet6/ip6_ecn.h>
61 #endif
62 
63 #include <netinet/ip6.h>
64 #ifdef INET6
65 #include <netinet6/ip6_var.h>
66 #include <netinet6/scope6_var.h>
67 #endif
68 #include <netinet/in_pcb.h>
69 #ifdef INET6
70 #include <netinet/icmp6.h>
71 #endif
72 #ifdef SCTP
73 #include <netinet/sctp_crc32.h>
74 #endif
75 
76 #include <netinet/udp.h>
77 #include <netipsec/ah.h>
78 #include <netipsec/esp.h>
79 #include <netipsec/ipsec.h>
80 #ifdef INET6
81 #include <netipsec/ipsec6.h>
82 #endif
83 #include <netipsec/ah_var.h>
84 #include <netipsec/esp_var.h>
85 #include <netipsec/ipcomp_var.h>
86 
87 #include <netipsec/xform.h>
88 
89 #include <netipsec/key.h>
90 #include <netipsec/keydb.h>
91 #include <netipsec/key_debug.h>
92 
93 #include <machine/in_cksum.h>
94 
95 #define	IPSEC_OSTAT_INC(proto, name)	do {		\
96 	if ((proto) == IPPROTO_ESP)	\
97 		ESPSTAT_INC(esps_##name);	\
98 	else if ((proto) == IPPROTO_AH)\
99 		AHSTAT_INC(ahs_##name);		\
100 	else					\
101 		IPCOMPSTAT_INC(ipcomps_##name);	\
102 } while (0)
103 
104 static int ipsec_encap(struct mbuf **mp, struct secasindex *saidx);
105 
106 #ifdef INET
107 static struct secasvar *
108 ipsec4_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
109 {
110 	struct secasindex *saidx, tmpsaidx;
111 	struct ipsecrequest *isr;
112 	struct sockaddr_in *sin;
113 	struct secasvar *sav;
114 	struct ip *ip;
115 
116 	/*
117 	 * Check system global policy controls.
118 	 */
119 next:
120 	isr = sp->req[*pidx];
121 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
122 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
123 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
124 		DPRINTF(("%s: IPsec outbound packet dropped due"
125 			" to policy (check your sysctls)\n", __func__));
126 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
127 		*error = EHOSTUNREACH;
128 		return (NULL);
129 	}
130 	/*
131 	 * Craft SA index to search for proper SA.  Note that
132 	 * we only initialize unspecified SA peers for transport
133 	 * mode; for tunnel mode they must already be filled in.
134 	 */
135 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
136 		saidx = &tmpsaidx;
137 		*saidx = isr->saidx;
138 		ip = mtod(m, struct ip *);
139 		if (saidx->src.sa.sa_len == 0) {
140 			sin = &saidx->src.sin;
141 			sin->sin_len = sizeof(*sin);
142 			sin->sin_family = AF_INET;
143 			sin->sin_port = IPSEC_PORT_ANY;
144 			sin->sin_addr = ip->ip_src;
145 		}
146 		if (saidx->dst.sa.sa_len == 0) {
147 			sin = &saidx->dst.sin;
148 			sin->sin_len = sizeof(*sin);
149 			sin->sin_family = AF_INET;
150 			sin->sin_port = IPSEC_PORT_ANY;
151 			sin->sin_addr = ip->ip_dst;
152 		}
153 	} else
154 		saidx = &sp->req[*pidx]->saidx;
155 	/*
156 	 * Lookup SA and validate it.
157 	 */
158 	sav = key_allocsa_policy(sp, saidx, error);
159 	if (sav == NULL) {
160 		IPSECSTAT_INC(ips_out_nosa);
161 		if (*error != 0)
162 			return (NULL);
163 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
164 			/*
165 			 * We have no SA and policy that doesn't require
166 			 * this IPsec transform, thus we can continue w/o
167 			 * IPsec processing, i.e. return EJUSTRETURN.
168 			 * But first check if there is some bundled transform.
169 			 */
170 			if (sp->tcount > ++(*pidx))
171 				goto next;
172 			*error = EJUSTRETURN;
173 		}
174 		return (NULL);
175 	}
176 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
177 	return (sav);
178 }
179 
180 /*
181  * IPsec output logic for IPv4.
182  */
183 static int
184 ipsec4_perform_request(struct mbuf *m, struct secpolicy *sp, u_int idx)
185 {
186 	char sbuf[IPSEC_ADDRSTRLEN], dbuf[IPSEC_ADDRSTRLEN];
187 	struct ipsec_ctx_data ctx;
188 	union sockaddr_union *dst;
189 	struct secasvar *sav;
190 	struct ip *ip;
191 	int error, i, off;
192 
193 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
194 
195 	/*
196 	 * We hold the reference to SP. Content of SP couldn't be changed.
197 	 * Craft secasindex and do lookup for suitable SA.
198 	 * Then do encapsulation if needed and call xform's output.
199 	 * We need to store SP in the xform callback parameters.
200 	 * In xform callback we will extract SP and it can be used to
201 	 * determine next transform. At the end of transform we can
202 	 * release reference to SP.
203 	 */
204 	sav = ipsec4_allocsa(m, sp, &idx, &error);
205 	if (sav == NULL) {
206 		if (error == EJUSTRETURN) { /* No IPsec required */
207 			key_freesp(&sp);
208 			return (error);
209 		}
210 		goto bad;
211 	}
212 	/*
213 	 * XXXAE: most likely ip_sum at this point is wrong.
214 	 */
215 	IPSEC_INIT_CTX(&ctx, &m, sav, AF_INET, IPSEC_ENC_BEFORE);
216 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
217 		goto bad;
218 
219 	ip = mtod(m, struct ip *);
220 	dst = &sav->sah->saidx.dst;
221 	/* Do the appropriate encapsulation, if necessary */
222 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
223 	    dst->sa.sa_family != AF_INET ||	    /* PF mismatch */
224 	    (dst->sa.sa_family == AF_INET &&	    /* Proxy */
225 	     dst->sin.sin_addr.s_addr != INADDR_ANY &&
226 	     dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
227 		/* Fix IPv4 header checksum and length */
228 		ip->ip_len = htons(m->m_pkthdr.len);
229 		ip->ip_sum = 0;
230 		ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
231 		error = ipsec_encap(&m, &sav->sah->saidx);
232 		if (error != 0) {
233 			DPRINTF(("%s: encapsulation for SA %s->%s "
234 			    "SPI 0x%08x failed with error %d\n", __func__,
235 			    ipsec_address(&sav->sah->saidx.src, sbuf,
236 				sizeof(sbuf)),
237 			    ipsec_address(&sav->sah->saidx.dst, dbuf,
238 				sizeof(dbuf)), ntohl(sav->spi), error));
239 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
240 			goto bad;
241 		}
242 	}
243 
244 	IPSEC_INIT_CTX(&ctx, &m, sav, dst->sa.sa_family, IPSEC_ENC_AFTER);
245 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
246 		goto bad;
247 
248 	/*
249 	 * Dispatch to the appropriate IPsec transform logic.  The
250 	 * packet will be returned for transmission after crypto
251 	 * processing, etc. are completed.
252 	 *
253 	 * NB: m & sav are ``passed to caller'' who's responsible for
254 	 *     reclaiming their resources.
255 	 */
256 	switch(dst->sa.sa_family) {
257 	case AF_INET:
258 		ip = mtod(m, struct ip *);
259 		i = ip->ip_hl << 2;
260 		off = offsetof(struct ip, ip_p);
261 		break;
262 #ifdef INET6
263 	case AF_INET6:
264 		i = sizeof(struct ip6_hdr);
265 		off = offsetof(struct ip6_hdr, ip6_nxt);
266 		break;
267 #endif /* INET6 */
268 	default:
269 		DPRINTF(("%s: unsupported protocol family %u\n",
270 		    __func__, dst->sa.sa_family));
271 		error = EPFNOSUPPORT;
272 		IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf);
273 		goto bad;
274 	}
275 	error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off);
276 	if (error != 0) {
277 		key_freesav(&sav);
278 		key_freesp(&sp);
279 	}
280 	return (error);
281 bad:
282 	IPSECSTAT_INC(ips_out_inval);
283 	if (m != NULL)
284 		m_freem(m);
285 	if (sav != NULL)
286 		key_freesav(&sav);
287 	key_freesp(&sp);
288 	return (error);
289 }
290 
291 int
292 ipsec4_process_packet(struct mbuf *m, struct secpolicy *sp,
293     struct inpcb *inp)
294 {
295 
296 	return (ipsec4_perform_request(m, sp, 0));
297 }
298 
299 static int
300 ipsec4_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
301 {
302 	struct secpolicy *sp;
303 	int error;
304 
305 	/* Lookup for the corresponding outbound security policy */
306 	sp = ipsec4_checkpolicy(m, inp, &error);
307 	if (sp == NULL) {
308 		if (error == -EINVAL) {
309 			/* Discarded by policy. */
310 			m_freem(m);
311 			return (EACCES);
312 		}
313 		return (0); /* No IPsec required. */
314 	}
315 
316 	/*
317 	 * Usually we have to have tunnel mode IPsec security policy
318 	 * when we are forwarding a packet. Otherwise we could not handle
319 	 * encrypted replies, because they are not destined for us. But
320 	 * some users are doing source address translation for forwarded
321 	 * packets, and thus, even if they are forwarded, the replies will
322 	 * return back to us.
323 	 */
324 	if (!forwarding) {
325 		/*
326 		 * Do delayed checksums now because we send before
327 		 * this is done in the normal processing path.
328 		 */
329 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
330 			in_delayed_cksum(m);
331 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
332 		}
333 #ifdef SCTP
334 		if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
335 			struct ip *ip = mtod(m, struct ip *);
336 
337 			sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
338 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
339 		}
340 #endif
341 	}
342 	/* NB: callee frees mbuf and releases reference to SP */
343 	error = ipsec4_process_packet(m, sp, inp);
344 	if (error == EJUSTRETURN) {
345 		/*
346 		 * We had a SP with a level of 'use' and no SA. We
347 		 * will just continue to process the packet without
348 		 * IPsec processing and return without error.
349 		 */
350 		return (0);
351 	}
352 	if (error == 0)
353 		return (EINPROGRESS); /* consumed by IPsec */
354 	return (error);
355 }
356 
357 /*
358  * IPSEC_OUTPUT() method implementation for IPv4.
359  * 0 - no IPsec handling needed
360  * other values - mbuf consumed by IPsec.
361  */
362 int
363 ipsec4_output(struct mbuf *m, struct inpcb *inp)
364 {
365 
366 	/*
367 	 * If the packet is resubmitted to ip_output (e.g. after
368 	 * AH, ESP, etc. processing), there will be a tag to bypass
369 	 * the lookup and related policy checking.
370 	 */
371 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
372 		return (0);
373 
374 	return (ipsec4_common_output(m, inp, 0));
375 }
376 
377 /*
378  * IPSEC_FORWARD() method implementation for IPv4.
379  * 0 - no IPsec handling needed
380  * other values - mbuf consumed by IPsec.
381  */
382 int
383 ipsec4_forward(struct mbuf *m)
384 {
385 
386 	/*
387 	 * Check if this packet has an active inbound SP and needs to be
388 	 * dropped instead of forwarded.
389 	 */
390 	if (ipsec4_in_reject(m, NULL) != 0) {
391 		m_freem(m);
392 		return (EACCES);
393 	}
394 	return (ipsec4_common_output(m, NULL, 1));
395 }
396 #endif
397 
398 #ifdef INET6
399 static int
400 in6_sa_equal_addrwithscope(const struct sockaddr_in6 *sa,
401     const struct in6_addr *ia)
402 {
403 	struct in6_addr ia2;
404 
405 	if (IN6_IS_SCOPE_LINKLOCAL(&sa->sin6_addr)) {
406 		memcpy(&ia2, &sa->sin6_addr, sizeof(ia2));
407 		ia2.s6_addr16[1] = htons(sa->sin6_scope_id);
408 		return (IN6_ARE_ADDR_EQUAL(ia, &ia2));
409 	}
410 	return (IN6_ARE_ADDR_EQUAL(&sa->sin6_addr, ia));
411 }
412 
413 static struct secasvar *
414 ipsec6_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
415 {
416 	struct secasindex *saidx, tmpsaidx;
417 	struct ipsecrequest *isr;
418 	struct sockaddr_in6 *sin6;
419 	struct secasvar *sav;
420 	struct ip6_hdr *ip6;
421 
422 	/*
423 	 * Check system global policy controls.
424 	 */
425 next:
426 	isr = sp->req[*pidx];
427 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
428 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
429 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
430 		DPRINTF(("%s: IPsec outbound packet dropped due"
431 			" to policy (check your sysctls)\n", __func__));
432 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
433 		*error = EHOSTUNREACH;
434 		return (NULL);
435 	}
436 	/*
437 	 * Craft SA index to search for proper SA.  Note that
438 	 * we only fillin unspecified SA peers for transport
439 	 * mode; for tunnel mode they must already be filled in.
440 	 */
441 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
442 		saidx = &tmpsaidx;
443 		*saidx = isr->saidx;
444 		ip6 = mtod(m, struct ip6_hdr *);
445 		if (saidx->src.sin6.sin6_len == 0) {
446 			sin6 = (struct sockaddr_in6 *)&saidx->src;
447 			sin6->sin6_len = sizeof(*sin6);
448 			sin6->sin6_family = AF_INET6;
449 			sin6->sin6_port = IPSEC_PORT_ANY;
450 			sin6->sin6_addr = ip6->ip6_src;
451 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
452 				/* fix scope id for comparing SPD */
453 				sin6->sin6_addr.s6_addr16[1] = 0;
454 				sin6->sin6_scope_id =
455 				    ntohs(ip6->ip6_src.s6_addr16[1]);
456 			}
457 		}
458 		if (saidx->dst.sin6.sin6_len == 0) {
459 			sin6 = (struct sockaddr_in6 *)&saidx->dst;
460 			sin6->sin6_len = sizeof(*sin6);
461 			sin6->sin6_family = AF_INET6;
462 			sin6->sin6_port = IPSEC_PORT_ANY;
463 			sin6->sin6_addr = ip6->ip6_dst;
464 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
465 				/* fix scope id for comparing SPD */
466 				sin6->sin6_addr.s6_addr16[1] = 0;
467 				sin6->sin6_scope_id =
468 				    ntohs(ip6->ip6_dst.s6_addr16[1]);
469 			}
470 		}
471 	} else
472 		saidx = &sp->req[*pidx]->saidx;
473 	/*
474 	 * Lookup SA and validate it.
475 	 */
476 	sav = key_allocsa_policy(sp, saidx, error);
477 	if (sav == NULL) {
478 		IPSEC6STAT_INC(ips_out_nosa);
479 		if (*error != 0)
480 			return (NULL);
481 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
482 			/*
483 			 * We have no SA and policy that doesn't require
484 			 * this IPsec transform, thus we can continue w/o
485 			 * IPsec processing, i.e. return EJUSTRETURN.
486 			 * But first check if there is some bundled transform.
487 			 */
488 			if (sp->tcount > ++(*pidx))
489 				goto next;
490 			*error = EJUSTRETURN;
491 		}
492 		return (NULL);
493 	}
494 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
495 	return (sav);
496 }
497 
498 /*
499  * IPsec output logic for IPv6.
500  */
501 static int
502 ipsec6_perform_request(struct mbuf *m, struct secpolicy *sp, u_int idx)
503 {
504 	char sbuf[IPSEC_ADDRSTRLEN], dbuf[IPSEC_ADDRSTRLEN];
505 	struct ipsec_ctx_data ctx;
506 	union sockaddr_union *dst;
507 	struct secasvar *sav;
508 	struct ip6_hdr *ip6;
509 	int error, i, off;
510 
511 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
512 
513 	sav = ipsec6_allocsa(m, sp, &idx, &error);
514 	if (sav == NULL) {
515 		if (error == EJUSTRETURN) { /* No IPsec required */
516 			key_freesp(&sp);
517 			return (error);
518 		}
519 		goto bad;
520 	}
521 
522 	/* Fix IP length in case if it is not set yet. */
523 	ip6 = mtod(m, struct ip6_hdr *);
524 	ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
525 
526 	IPSEC_INIT_CTX(&ctx, &m, sav, AF_INET6, IPSEC_ENC_BEFORE);
527 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
528 		goto bad;
529 
530 	ip6 = mtod(m, struct ip6_hdr *); /* pfil can change mbuf */
531 	dst = &sav->sah->saidx.dst;
532 
533 	/* Do the appropriate encapsulation, if necessary */
534 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
535 	    dst->sa.sa_family != AF_INET6 ||        /* PF mismatch */
536 	    ((dst->sa.sa_family == AF_INET6) &&
537 	     (!IN6_IS_ADDR_UNSPECIFIED(&dst->sin6.sin6_addr)) &&
538 	     (!in6_sa_equal_addrwithscope(&dst->sin6, &ip6->ip6_dst)))) {
539 		if (m->m_pkthdr.len - sizeof(*ip6) > IPV6_MAXPACKET) {
540 			/* No jumbogram support. */
541 			error = ENXIO;   /*XXX*/
542 			goto bad;
543 		}
544 		error = ipsec_encap(&m, &sav->sah->saidx);
545 		if (error != 0) {
546 			DPRINTF(("%s: encapsulation for SA %s->%s "
547 			    "SPI 0x%08x failed with error %d\n", __func__,
548 			    ipsec_address(&sav->sah->saidx.src, sbuf,
549 				sizeof(sbuf)),
550 			    ipsec_address(&sav->sah->saidx.dst, dbuf,
551 				sizeof(dbuf)), ntohl(sav->spi), error));
552 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
553 			goto bad;
554 		}
555 	}
556 
557 	IPSEC_INIT_CTX(&ctx, &m, sav, dst->sa.sa_family, IPSEC_ENC_AFTER);
558 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
559 		goto bad;
560 
561 	switch(dst->sa.sa_family) {
562 #ifdef INET
563 	case AF_INET:
564 		{
565 		struct ip *ip;
566 		ip = mtod(m, struct ip *);
567 		i = ip->ip_hl << 2;
568 		off = offsetof(struct ip, ip_p);
569 		}
570 		break;
571 #endif /* AF_INET */
572 	case AF_INET6:
573 		i = sizeof(struct ip6_hdr);
574 		off = offsetof(struct ip6_hdr, ip6_nxt);
575 		break;
576 	default:
577 		DPRINTF(("%s: unsupported protocol family %u\n",
578 				 __func__, dst->sa.sa_family));
579 		error = EPFNOSUPPORT;
580 		IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf);
581 		goto bad;
582 	}
583 	error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off);
584 	if (error != 0) {
585 		key_freesav(&sav);
586 		key_freesp(&sp);
587 	}
588 	return (error);
589 bad:
590 	IPSEC6STAT_INC(ips_out_inval);
591 	if (m != NULL)
592 		m_freem(m);
593 	if (sav != NULL)
594 		key_freesav(&sav);
595 	key_freesp(&sp);
596 	return (error);
597 }
598 
599 int
600 ipsec6_process_packet(struct mbuf *m, struct secpolicy *sp,
601     struct inpcb *inp)
602 {
603 
604 	return (ipsec6_perform_request(m, sp, 0));
605 }
606 
607 static int
608 ipsec6_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
609 {
610 	struct secpolicy *sp;
611 	int error;
612 
613 	/* Lookup for the corresponding outbound security policy */
614 	sp = ipsec6_checkpolicy(m, inp, &error);
615 	if (sp == NULL) {
616 		if (error == -EINVAL) {
617 			/* Discarded by policy. */
618 			m_freem(m);
619 			return (EACCES);
620 		}
621 		return (0); /* No IPsec required. */
622 	}
623 
624 	if (!forwarding) {
625 		/*
626 		 * Do delayed checksums now because we send before
627 		 * this is done in the normal processing path.
628 		 */
629 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
630 			in6_delayed_cksum(m, m->m_pkthdr.len -
631 			    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
632 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
633 		}
634 #ifdef SCTP
635 		if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
636 			sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
637 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
638 		}
639 #endif
640 	}
641 	/* NB: callee frees mbuf and releases reference to SP */
642 	error = ipsec6_process_packet(m, sp, inp);
643 	if (error == EJUSTRETURN) {
644 		/*
645 		 * We had a SP with a level of 'use' and no SA. We
646 		 * will just continue to process the packet without
647 		 * IPsec processing and return without error.
648 		 */
649 		return (0);
650 	}
651 	if (error == 0)
652 		return (EINPROGRESS); /* consumed by IPsec */
653 	return (error);
654 }
655 
656 /*
657  * IPSEC_OUTPUT() method implementation for IPv6.
658  * 0 - no IPsec handling needed
659  * other values - mbuf consumed by IPsec.
660  */
661 int
662 ipsec6_output(struct mbuf *m, struct inpcb *inp)
663 {
664 
665 	/*
666 	 * If the packet is resubmitted to ip_output (e.g. after
667 	 * AH, ESP, etc. processing), there will be a tag to bypass
668 	 * the lookup and related policy checking.
669 	 */
670 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
671 		return (0);
672 
673 	return (ipsec6_common_output(m, inp, 0));
674 }
675 
676 /*
677  * IPSEC_FORWARD() method implementation for IPv6.
678  * 0 - no IPsec handling needed
679  * other values - mbuf consumed by IPsec.
680  */
681 int
682 ipsec6_forward(struct mbuf *m)
683 {
684 
685 	/*
686 	 * Check if this packet has an active inbound SP and needs to be
687 	 * dropped instead of forwarded.
688 	 */
689 	if (ipsec6_in_reject(m, NULL) != 0) {
690 		m_freem(m);
691 		return (EACCES);
692 	}
693 	return (ipsec6_common_output(m, NULL, 1));
694 }
695 #endif /* INET6 */
696 
697 int
698 ipsec_process_done(struct mbuf *m, struct secpolicy *sp, struct secasvar *sav,
699     u_int idx)
700 {
701 	struct xform_history *xh;
702 	struct secasindex *saidx;
703 	struct m_tag *mtag;
704 	int error;
705 
706 	saidx = &sav->sah->saidx;
707 	switch (saidx->dst.sa.sa_family) {
708 #ifdef INET
709 	case AF_INET:
710 		/* Fix the header length, for AH processing. */
711 		mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
712 		break;
713 #endif /* INET */
714 #ifdef INET6
715 	case AF_INET6:
716 		/* Fix the header length, for AH processing. */
717 		if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
718 			error = ENXIO;
719 			goto bad;
720 		}
721 		if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
722 			/* No jumbogram support. */
723 			error = ENXIO;	/*?*/
724 			goto bad;
725 		}
726 		mtod(m, struct ip6_hdr *)->ip6_plen =
727 			htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
728 		break;
729 #endif /* INET6 */
730 	default:
731 		DPRINTF(("%s: unknown protocol family %u\n", __func__,
732 		    saidx->dst.sa.sa_family));
733 		error = ENXIO;
734 		goto bad;
735 	}
736 
737 	/*
738 	 * Add a record of what we've done to the packet.
739 	 */
740 	mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE, sizeof(*xh), M_NOWAIT);
741 	if (mtag == NULL) {
742 		DPRINTF(("%s: could not get packet tag\n", __func__));
743 		error = ENOMEM;
744 		goto bad;
745 	}
746 
747 	xh = (struct xform_history *)(mtag + 1);
748 	xh->dst = saidx->dst;
749 	xh->proto = saidx->proto;
750 	xh->mode = saidx->mode;
751 	xh->spi = sav->spi;
752 	m_tag_prepend(m, mtag);
753 
754 	key_sa_recordxfer(sav, m);		/* record data transfer */
755 
756 	/*
757 	 * If there's another (bundled) SA to apply, do so.
758 	 * Note that this puts a burden on the kernel stack size.
759 	 * If this is a problem we'll need to introduce a queue
760 	 * to set the packet on so we can unwind the stack before
761 	 * doing further processing.
762 	 */
763 	if (++idx < sp->tcount) {
764 		switch (saidx->dst.sa.sa_family) {
765 #ifdef INET
766 		case AF_INET:
767 			key_freesav(&sav);
768 			IPSECSTAT_INC(ips_out_bundlesa);
769 			return (ipsec4_perform_request(m, sp, idx));
770 			/* NOTREACHED */
771 #endif
772 #ifdef INET6
773 		case AF_INET6:
774 			key_freesav(&sav);
775 			IPSEC6STAT_INC(ips_out_bundlesa);
776 			return (ipsec6_perform_request(m, sp, idx));
777 			/* NOTREACHED */
778 #endif /* INET6 */
779 		default:
780 			DPRINTF(("%s: unknown protocol family %u\n", __func__,
781 			    saidx->dst.sa.sa_family));
782 			error = EPFNOSUPPORT;
783 			goto bad;
784 		}
785 	}
786 
787 	key_freesp(&sp), sp = NULL;	/* Release reference to SP */
788 #ifdef INET
789 	/*
790 	 * Do UDP encapsulation if SA requires it.
791 	 */
792 	if (sav->natt != NULL) {
793 		error = udp_ipsec_output(m, sav);
794 		if (error != 0)
795 			goto bad;
796 	}
797 #endif /* INET */
798 	/*
799 	 * We're done with IPsec processing, transmit the packet using the
800 	 * appropriate network protocol (IP or IPv6).
801 	 */
802 	switch (saidx->dst.sa.sa_family) {
803 #ifdef INET
804 	case AF_INET:
805 		key_freesav(&sav);
806 		return ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
807 #endif /* INET */
808 #ifdef INET6
809 	case AF_INET6:
810 		key_freesav(&sav);
811 		return ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
812 #endif /* INET6 */
813 	}
814 	panic("ipsec_process_done");
815 bad:
816 	m_freem(m);
817 	key_freesav(&sav);
818 	if (sp != NULL)
819 		key_freesp(&sp);
820 	return (error);
821 }
822 
823 /*
824  * ipsec_prepend() is optimized version of M_PREPEND().
825  * ipsec_encap() is called by IPsec output routine for tunnel mode SA.
826  * It is expected that after IP encapsulation some IPsec transform will
827  * be performed. Each IPsec transform inserts its variable length header
828  * just after outer IP header using m_makespace(). If given mbuf has not
829  * enough free space at the beginning, we allocate new mbuf and reserve
830  * some space at the beginning and at the end.
831  * This helps avoid allocating of new mbuf and data copying in m_makespace(),
832  * we place outer header in the middle of mbuf's data with reserved leading
833  * and trailing space:
834  *	[ LEADINGSPACE ][ Outer IP header ][ TRAILINGSPACE ]
835  * LEADINGSPACE will be used to add ethernet header, TRAILINGSPACE will
836  * be used to inject AH/ESP/IPCOMP header.
837  */
838 #define	IPSEC_TRAILINGSPACE	(sizeof(struct udphdr) +/* NAT-T */	\
839     max(sizeof(struct newesp) + EALG_MAX_BLOCK_LEN,	/* ESP + IV */	\
840 	sizeof(struct newah) + HASH_MAX_LEN		/* AH + ICV */))
841 static struct mbuf *
842 ipsec_prepend(struct mbuf *m, int len, int how)
843 {
844 	struct mbuf *n;
845 
846 	M_ASSERTPKTHDR(m);
847 	IPSEC_ASSERT(len < MHLEN, ("wrong length"));
848 	if (M_LEADINGSPACE(m) >= len) {
849 		/* No need to allocate new mbuf. */
850 		m->m_data -= len;
851 		m->m_len += len;
852 		m->m_pkthdr.len += len;
853 		return (m);
854 	}
855 	n = m_gethdr(how, m->m_type);
856 	if (n == NULL) {
857 		m_freem(m);
858 		return (NULL);
859 	}
860 	m_move_pkthdr(n, m);
861 	n->m_next = m;
862 	if (len + IPSEC_TRAILINGSPACE < M_SIZE(n))
863 		m_align(n, len + IPSEC_TRAILINGSPACE);
864 	n->m_len = len;
865 	n->m_pkthdr.len += len;
866 	return (n);
867 }
868 
869 static int
870 ipsec_encap(struct mbuf **mp, struct secasindex *saidx)
871 {
872 #ifdef INET6
873 	struct ip6_hdr *ip6;
874 #endif
875 	struct ip *ip;
876 	int setdf;
877 	uint8_t itos, proto;
878 
879 	ip = mtod(*mp, struct ip *);
880 	switch (ip->ip_v) {
881 #ifdef INET
882 	case IPVERSION:
883 		proto = IPPROTO_IPIP;
884 		/*
885 		 * Collect IP_DF state from the inner header
886 		 * and honor system-wide control of how to handle it.
887 		 */
888 		switch (V_ip4_ipsec_dfbit) {
889 		case 0:	/* clear in outer header */
890 		case 1:	/* set in outer header */
891 			setdf = V_ip4_ipsec_dfbit;
892 			break;
893 		default:/* propagate to outer header */
894 			setdf = (ip->ip_off & htons(IP_DF)) != 0;
895 		}
896 		itos = ip->ip_tos;
897 		break;
898 #endif
899 #ifdef INET6
900 	case (IPV6_VERSION >> 4):
901 		proto = IPPROTO_IPV6;
902 		ip6 = mtod(*mp, struct ip6_hdr *);
903 		itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
904 		setdf = V_ip4_ipsec_dfbit ? 1: 0;
905 		/* scoped address handling */
906 		in6_clearscope(&ip6->ip6_src);
907 		in6_clearscope(&ip6->ip6_dst);
908 		break;
909 #endif
910 	default:
911 		return (EAFNOSUPPORT);
912 	}
913 	switch (saidx->dst.sa.sa_family) {
914 #ifdef INET
915 	case AF_INET:
916 		if (saidx->src.sa.sa_family != AF_INET ||
917 		    saidx->src.sin.sin_addr.s_addr == INADDR_ANY ||
918 		    saidx->dst.sin.sin_addr.s_addr == INADDR_ANY)
919 			return (EINVAL);
920 		*mp = ipsec_prepend(*mp, sizeof(struct ip), M_NOWAIT);
921 		if (*mp == NULL)
922 			return (ENOBUFS);
923 		ip = mtod(*mp, struct ip *);
924 		ip->ip_v = IPVERSION;
925 		ip->ip_hl = sizeof(struct ip) >> 2;
926 		ip->ip_p = proto;
927 		ip->ip_len = htons((*mp)->m_pkthdr.len);
928 		ip->ip_ttl = V_ip_defttl;
929 		ip->ip_sum = 0;
930 		ip->ip_off = setdf ? htons(IP_DF): 0;
931 		ip->ip_src = saidx->src.sin.sin_addr;
932 		ip->ip_dst = saidx->dst.sin.sin_addr;
933 		ip_ecn_ingress(V_ip4_ipsec_ecn, &ip->ip_tos, &itos);
934 		ip_fillid(ip);
935 		break;
936 #endif /* INET */
937 #ifdef INET6
938 	case AF_INET6:
939 		if (saidx->src.sa.sa_family != AF_INET6 ||
940 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->src.sin6.sin6_addr) ||
941 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->dst.sin6.sin6_addr))
942 			return (EINVAL);
943 		*mp = ipsec_prepend(*mp, sizeof(struct ip6_hdr), M_NOWAIT);
944 		if (*mp == NULL)
945 			return (ENOBUFS);
946 		ip6 = mtod(*mp, struct ip6_hdr *);
947 		ip6->ip6_flow = 0;
948 		ip6->ip6_vfc = IPV6_VERSION;
949 		ip6->ip6_hlim = V_ip6_defhlim;
950 		ip6->ip6_nxt = proto;
951 		ip6->ip6_dst = saidx->dst.sin6.sin6_addr;
952 		/* For link-local address embed scope zone id */
953 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst))
954 			ip6->ip6_dst.s6_addr16[1] =
955 			    htons(saidx->dst.sin6.sin6_scope_id & 0xffff);
956 		ip6->ip6_src = saidx->src.sin6.sin6_addr;
957 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src))
958 			ip6->ip6_src.s6_addr16[1] =
959 			    htons(saidx->src.sin6.sin6_scope_id & 0xffff);
960 		ip6->ip6_plen = htons((*mp)->m_pkthdr.len - sizeof(*ip6));
961 		ip_ecn_ingress(V_ip6_ipsec_ecn, &proto, &itos);
962 		ip6->ip6_flow |= htonl((uint32_t)proto << 20);
963 		break;
964 #endif /* INET6 */
965 	default:
966 		return (EAFNOSUPPORT);
967 	}
968 	(*mp)->m_flags &= ~(M_BCAST | M_MCAST);
969 	return (0);
970 }
971 
972