xref: /freebsd/sys/netipsec/ipsec_output.c (revision 662c13053f4bf2d6245ba7e2b66c10d1cd5c1fb9)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
5  * Copyright (c) 2016 Andrey V. Elsukov <ae@FreeBSD.org>
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  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $FreeBSD$
30  */
31 
32 /*
33  * IPsec output processing.
34  */
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_ipsec.h"
38 #include "opt_sctp.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.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/hhook.h>
48 #include <sys/syslog.h>
49 
50 #include <net/if.h>
51 #include <net/if_enc.h>
52 #include <net/if_var.h>
53 #include <net/vnet.h>
54 
55 #include <netinet/in.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/ip.h>
58 #include <netinet/ip_var.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip_ecn.h>
61 #ifdef INET6
62 #include <netinet6/ip6_ecn.h>
63 #endif
64 
65 #include <netinet/ip6.h>
66 #ifdef INET6
67 #include <netinet6/ip6_var.h>
68 #include <netinet6/scope6_var.h>
69 #endif
70 #include <netinet/in_pcb.h>
71 #ifdef INET6
72 #include <netinet/icmp6.h>
73 #endif
74 #if defined(SCTP) || defined(SCTP_SUPPORT)
75 #include <netinet/sctp_crc32.h>
76 #endif
77 
78 #include <netinet/udp.h>
79 #include <netipsec/ah.h>
80 #include <netipsec/esp.h>
81 #include <netipsec/ipsec.h>
82 #ifdef INET6
83 #include <netipsec/ipsec6.h>
84 #endif
85 #include <netipsec/ah_var.h>
86 #include <netipsec/esp_var.h>
87 #include <netipsec/ipcomp_var.h>
88 
89 #include <netipsec/xform.h>
90 
91 #include <netipsec/key.h>
92 #include <netipsec/keydb.h>
93 #include <netipsec/key_debug.h>
94 
95 #include <machine/in_cksum.h>
96 
97 #define	IPSEC_OSTAT_INC(proto, name)	do {		\
98 	if ((proto) == IPPROTO_ESP)	\
99 		ESPSTAT_INC(esps_##name);	\
100 	else if ((proto) == IPPROTO_AH)\
101 		AHSTAT_INC(ahs_##name);		\
102 	else					\
103 		IPCOMPSTAT_INC(ipcomps_##name);	\
104 } while (0)
105 
106 static int ipsec_encap(struct mbuf **mp, struct secasindex *saidx);
107 
108 #ifdef INET
109 static struct secasvar *
110 ipsec4_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
111 {
112 	struct secasindex *saidx, tmpsaidx;
113 	struct ipsecrequest *isr;
114 	struct sockaddr_in *sin;
115 	struct secasvar *sav;
116 	struct ip *ip;
117 
118 	/*
119 	 * Check system global policy controls.
120 	 */
121 next:
122 	isr = sp->req[*pidx];
123 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
124 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
125 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
126 		DPRINTF(("%s: IPsec outbound packet dropped due"
127 			" to policy (check your sysctls)\n", __func__));
128 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
129 		*error = EHOSTUNREACH;
130 		return (NULL);
131 	}
132 	/*
133 	 * Craft SA index to search for proper SA.  Note that
134 	 * we only initialize unspecified SA peers for transport
135 	 * mode; for tunnel mode they must already be filled in.
136 	 */
137 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
138 		saidx = &tmpsaidx;
139 		*saidx = isr->saidx;
140 		ip = mtod(m, struct ip *);
141 		if (saidx->src.sa.sa_len == 0) {
142 			sin = &saidx->src.sin;
143 			sin->sin_len = sizeof(*sin);
144 			sin->sin_family = AF_INET;
145 			sin->sin_port = IPSEC_PORT_ANY;
146 			sin->sin_addr = ip->ip_src;
147 		}
148 		if (saidx->dst.sa.sa_len == 0) {
149 			sin = &saidx->dst.sin;
150 			sin->sin_len = sizeof(*sin);
151 			sin->sin_family = AF_INET;
152 			sin->sin_port = IPSEC_PORT_ANY;
153 			sin->sin_addr = ip->ip_dst;
154 		}
155 	} else
156 		saidx = &sp->req[*pidx]->saidx;
157 	/*
158 	 * Lookup SA and validate it.
159 	 */
160 	sav = key_allocsa_policy(sp, saidx, error);
161 	if (sav == NULL) {
162 		IPSECSTAT_INC(ips_out_nosa);
163 		if (*error != 0)
164 			return (NULL);
165 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
166 			/*
167 			 * We have no SA and policy that doesn't require
168 			 * this IPsec transform, thus we can continue w/o
169 			 * IPsec processing, i.e. return EJUSTRETURN.
170 			 * But first check if there is some bundled transform.
171 			 */
172 			if (sp->tcount > ++(*pidx))
173 				goto next;
174 			*error = EJUSTRETURN;
175 		}
176 		return (NULL);
177 	}
178 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
179 	return (sav);
180 }
181 
182 /*
183  * IPsec output logic for IPv4.
184  */
185 static int
186 ipsec4_perform_request(struct mbuf *m, struct secpolicy *sp,
187     struct inpcb *inp, u_int idx)
188 {
189 	struct ipsec_ctx_data ctx;
190 	union sockaddr_union *dst;
191 	struct secasvar *sav;
192 	struct ip *ip;
193 	int error, i, off;
194 
195 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
196 
197 	/*
198 	 * We hold the reference to SP. Content of SP couldn't be changed.
199 	 * Craft secasindex and do lookup for suitable SA.
200 	 * Then do encapsulation if needed and call xform's output.
201 	 * We need to store SP in the xform callback parameters.
202 	 * In xform callback we will extract SP and it can be used to
203 	 * determine next transform. At the end of transform we can
204 	 * release reference to SP.
205 	 */
206 	sav = ipsec4_allocsa(m, sp, &idx, &error);
207 	if (sav == NULL) {
208 		if (error == EJUSTRETURN) { /* No IPsec required */
209 			key_freesp(&sp);
210 			return (error);
211 		}
212 		goto bad;
213 	}
214 	/*
215 	 * XXXAE: most likely ip_sum at this point is wrong.
216 	 */
217 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET, IPSEC_ENC_BEFORE);
218 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
219 		goto bad;
220 
221 	ip = mtod(m, struct ip *);
222 	dst = &sav->sah->saidx.dst;
223 	/* Do the appropriate encapsulation, if necessary */
224 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
225 	    dst->sa.sa_family != AF_INET ||	    /* PF mismatch */
226 	    (dst->sa.sa_family == AF_INET &&	    /* Proxy */
227 	     dst->sin.sin_addr.s_addr != INADDR_ANY &&
228 	     dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
229 		/* Fix IPv4 header checksum and length */
230 		ip->ip_len = htons(m->m_pkthdr.len);
231 		ip->ip_sum = 0;
232 		ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
233 		error = ipsec_encap(&m, &sav->sah->saidx);
234 		if (error != 0) {
235 			DPRINTF(("%s: encapsulation for SPI 0x%08x failed "
236 			    "with error %d\n", __func__, ntohl(sav->spi),
237 			    error));
238 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
239 			goto bad;
240 		}
241 		inp = NULL;
242 	}
243 
244 	IPSEC_INIT_CTX(&ctx, &m, inp, 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 	return (error);
277 bad:
278 	IPSECSTAT_INC(ips_out_inval);
279 	if (m != NULL)
280 		m_freem(m);
281 	if (sav != NULL)
282 		key_freesav(&sav);
283 	key_freesp(&sp);
284 	return (error);
285 }
286 
287 int
288 ipsec4_process_packet(struct mbuf *m, struct secpolicy *sp,
289     struct inpcb *inp)
290 {
291 
292 	return (ipsec4_perform_request(m, sp, inp, 0));
293 }
294 
295 static int
296 ipsec4_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
297 {
298 	struct secpolicy *sp;
299 	int error;
300 
301 	/* Lookup for the corresponding outbound security policy */
302 	sp = ipsec4_checkpolicy(m, inp, &error, !forwarding);
303 	if (sp == NULL) {
304 		if (error == -EINVAL) {
305 			/* Discarded by policy. */
306 			m_freem(m);
307 			return (EACCES);
308 		}
309 		return (0); /* No IPsec required. */
310 	}
311 
312 	/*
313 	 * Usually we have to have tunnel mode IPsec security policy
314 	 * when we are forwarding a packet. Otherwise we could not handle
315 	 * encrypted replies, because they are not destined for us. But
316 	 * some users are doing source address translation for forwarded
317 	 * packets, and thus, even if they are forwarded, the replies will
318 	 * return back to us.
319 	 */
320 	if (!forwarding) {
321 		/*
322 		 * Do delayed checksums now because we send before
323 		 * this is done in the normal processing path.
324 		 */
325 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
326 			in_delayed_cksum(m);
327 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
328 		}
329 #if defined(SCTP) || defined(SCTP_SUPPORT)
330 		if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
331 			struct ip *ip = mtod(m, struct ip *);
332 
333 			sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
334 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
335 		}
336 #endif
337 	}
338 	/* NB: callee frees mbuf and releases reference to SP */
339 	error = ipsec4_process_packet(m, sp, inp);
340 	if (error == EJUSTRETURN) {
341 		/*
342 		 * We had a SP with a level of 'use' and no SA. We
343 		 * will just continue to process the packet without
344 		 * IPsec processing and return without error.
345 		 */
346 		return (0);
347 	}
348 	if (error == 0)
349 		return (EINPROGRESS); /* consumed by IPsec */
350 	return (error);
351 }
352 
353 /*
354  * IPSEC_OUTPUT() method implementation for IPv4.
355  * 0 - no IPsec handling needed
356  * other values - mbuf consumed by IPsec.
357  */
358 int
359 ipsec4_output(struct mbuf *m, struct inpcb *inp)
360 {
361 
362 	/*
363 	 * If the packet is resubmitted to ip_output (e.g. after
364 	 * AH, ESP, etc. processing), there will be a tag to bypass
365 	 * the lookup and related policy checking.
366 	 */
367 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
368 		return (0);
369 
370 	return (ipsec4_common_output(m, inp, 0));
371 }
372 
373 /*
374  * IPSEC_FORWARD() method implementation for IPv4.
375  * 0 - no IPsec handling needed
376  * other values - mbuf consumed by IPsec.
377  */
378 int
379 ipsec4_forward(struct mbuf *m)
380 {
381 
382 	/*
383 	 * Check if this packet has an active inbound SP and needs to be
384 	 * dropped instead of forwarded.
385 	 */
386 	if (ipsec4_in_reject(m, NULL) != 0) {
387 		m_freem(m);
388 		return (EACCES);
389 	}
390 	return (ipsec4_common_output(m, NULL, 1));
391 }
392 #endif
393 
394 #ifdef INET6
395 static int
396 in6_sa_equal_addrwithscope(const struct sockaddr_in6 *sa,
397     const struct in6_addr *ia)
398 {
399 	struct in6_addr ia2;
400 
401 	if (IN6_IS_SCOPE_LINKLOCAL(&sa->sin6_addr)) {
402 		memcpy(&ia2, &sa->sin6_addr, sizeof(ia2));
403 		ia2.s6_addr16[1] = htons(sa->sin6_scope_id);
404 		return (IN6_ARE_ADDR_EQUAL(ia, &ia2));
405 	}
406 	return (IN6_ARE_ADDR_EQUAL(&sa->sin6_addr, ia));
407 }
408 
409 static struct secasvar *
410 ipsec6_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
411 {
412 	struct secasindex *saidx, tmpsaidx;
413 	struct ipsecrequest *isr;
414 	struct sockaddr_in6 *sin6;
415 	struct secasvar *sav;
416 	struct ip6_hdr *ip6;
417 
418 	/*
419 	 * Check system global policy controls.
420 	 */
421 next:
422 	isr = sp->req[*pidx];
423 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
424 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
425 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
426 		DPRINTF(("%s: IPsec outbound packet dropped due"
427 			" to policy (check your sysctls)\n", __func__));
428 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
429 		*error = EHOSTUNREACH;
430 		return (NULL);
431 	}
432 	/*
433 	 * Craft SA index to search for proper SA.  Note that
434 	 * we only fillin unspecified SA peers for transport
435 	 * mode; for tunnel mode they must already be filled in.
436 	 */
437 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
438 		saidx = &tmpsaidx;
439 		*saidx = isr->saidx;
440 		ip6 = mtod(m, struct ip6_hdr *);
441 		if (saidx->src.sin6.sin6_len == 0) {
442 			sin6 = (struct sockaddr_in6 *)&saidx->src;
443 			sin6->sin6_len = sizeof(*sin6);
444 			sin6->sin6_family = AF_INET6;
445 			sin6->sin6_port = IPSEC_PORT_ANY;
446 			sin6->sin6_addr = ip6->ip6_src;
447 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
448 				/* fix scope id for comparing SPD */
449 				sin6->sin6_addr.s6_addr16[1] = 0;
450 				sin6->sin6_scope_id =
451 				    ntohs(ip6->ip6_src.s6_addr16[1]);
452 			}
453 		}
454 		if (saidx->dst.sin6.sin6_len == 0) {
455 			sin6 = (struct sockaddr_in6 *)&saidx->dst;
456 			sin6->sin6_len = sizeof(*sin6);
457 			sin6->sin6_family = AF_INET6;
458 			sin6->sin6_port = IPSEC_PORT_ANY;
459 			sin6->sin6_addr = ip6->ip6_dst;
460 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
461 				/* fix scope id for comparing SPD */
462 				sin6->sin6_addr.s6_addr16[1] = 0;
463 				sin6->sin6_scope_id =
464 				    ntohs(ip6->ip6_dst.s6_addr16[1]);
465 			}
466 		}
467 	} else
468 		saidx = &sp->req[*pidx]->saidx;
469 	/*
470 	 * Lookup SA and validate it.
471 	 */
472 	sav = key_allocsa_policy(sp, saidx, error);
473 	if (sav == NULL) {
474 		IPSEC6STAT_INC(ips_out_nosa);
475 		if (*error != 0)
476 			return (NULL);
477 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
478 			/*
479 			 * We have no SA and policy that doesn't require
480 			 * this IPsec transform, thus we can continue w/o
481 			 * IPsec processing, i.e. return EJUSTRETURN.
482 			 * But first check if there is some bundled transform.
483 			 */
484 			if (sp->tcount > ++(*pidx))
485 				goto next;
486 			*error = EJUSTRETURN;
487 		}
488 		return (NULL);
489 	}
490 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
491 	return (sav);
492 }
493 
494 /*
495  * IPsec output logic for IPv6.
496  */
497 static int
498 ipsec6_perform_request(struct mbuf *m, struct secpolicy *sp,
499     struct inpcb *inp, u_int idx)
500 {
501 	struct ipsec_ctx_data ctx;
502 	union sockaddr_union *dst;
503 	struct secasvar *sav;
504 	struct ip6_hdr *ip6;
505 	int error, i, off;
506 
507 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
508 
509 	sav = ipsec6_allocsa(m, sp, &idx, &error);
510 	if (sav == NULL) {
511 		if (error == EJUSTRETURN) { /* No IPsec required */
512 			key_freesp(&sp);
513 			return (error);
514 		}
515 		goto bad;
516 	}
517 
518 	/* Fix IP length in case if it is not set yet. */
519 	ip6 = mtod(m, struct ip6_hdr *);
520 	ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
521 
522 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET6, IPSEC_ENC_BEFORE);
523 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
524 		goto bad;
525 
526 	ip6 = mtod(m, struct ip6_hdr *); /* pfil can change mbuf */
527 	dst = &sav->sah->saidx.dst;
528 
529 	/* Do the appropriate encapsulation, if necessary */
530 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
531 	    dst->sa.sa_family != AF_INET6 ||        /* PF mismatch */
532 	    ((dst->sa.sa_family == AF_INET6) &&
533 	     (!IN6_IS_ADDR_UNSPECIFIED(&dst->sin6.sin6_addr)) &&
534 	     (!in6_sa_equal_addrwithscope(&dst->sin6, &ip6->ip6_dst)))) {
535 		if (m->m_pkthdr.len - sizeof(*ip6) > IPV6_MAXPACKET) {
536 			/* No jumbogram support. */
537 			error = ENXIO;   /*XXX*/
538 			goto bad;
539 		}
540 		error = ipsec_encap(&m, &sav->sah->saidx);
541 		if (error != 0) {
542 			DPRINTF(("%s: encapsulation for SPI 0x%08x failed "
543 			    "with error %d\n", __func__, ntohl(sav->spi),
544 			    error));
545 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
546 			goto bad;
547 		}
548 		inp = NULL;
549 	}
550 
551 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, dst->sa.sa_family, IPSEC_ENC_AFTER);
552 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
553 		goto bad;
554 
555 	switch(dst->sa.sa_family) {
556 #ifdef INET
557 	case AF_INET:
558 		{
559 		struct ip *ip;
560 		ip = mtod(m, struct ip *);
561 		i = ip->ip_hl << 2;
562 		off = offsetof(struct ip, ip_p);
563 		}
564 		break;
565 #endif /* AF_INET */
566 	case AF_INET6:
567 		i = sizeof(struct ip6_hdr);
568 		off = offsetof(struct ip6_hdr, ip6_nxt);
569 		break;
570 	default:
571 		DPRINTF(("%s: unsupported protocol family %u\n",
572 				 __func__, dst->sa.sa_family));
573 		error = EPFNOSUPPORT;
574 		IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf);
575 		goto bad;
576 	}
577 	error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off);
578 	return (error);
579 bad:
580 	IPSEC6STAT_INC(ips_out_inval);
581 	if (m != NULL)
582 		m_freem(m);
583 	if (sav != NULL)
584 		key_freesav(&sav);
585 	key_freesp(&sp);
586 	return (error);
587 }
588 
589 int
590 ipsec6_process_packet(struct mbuf *m, struct secpolicy *sp,
591     struct inpcb *inp)
592 {
593 
594 	return (ipsec6_perform_request(m, sp, inp, 0));
595 }
596 
597 static int
598 ipsec6_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
599 {
600 	struct secpolicy *sp;
601 	int error;
602 
603 	/* Lookup for the corresponding outbound security policy */
604 	sp = ipsec6_checkpolicy(m, inp, &error, !forwarding);
605 	if (sp == NULL) {
606 		if (error == -EINVAL) {
607 			/* Discarded by policy. */
608 			m_freem(m);
609 			return (EACCES);
610 		}
611 		return (0); /* No IPsec required. */
612 	}
613 
614 	if (!forwarding) {
615 		/*
616 		 * Do delayed checksums now because we send before
617 		 * this is done in the normal processing path.
618 		 */
619 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
620 			in6_delayed_cksum(m, m->m_pkthdr.len -
621 			    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
622 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
623 		}
624 #if defined(SCTP) || defined(SCTP_SUPPORT)
625 		if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
626 			sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
627 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
628 		}
629 #endif
630 	}
631 	/* NB: callee frees mbuf and releases reference to SP */
632 	error = ipsec6_process_packet(m, sp, inp);
633 	if (error == EJUSTRETURN) {
634 		/*
635 		 * We had a SP with a level of 'use' and no SA. We
636 		 * will just continue to process the packet without
637 		 * IPsec processing and return without error.
638 		 */
639 		return (0);
640 	}
641 	if (error == 0)
642 		return (EINPROGRESS); /* consumed by IPsec */
643 	return (error);
644 }
645 
646 /*
647  * IPSEC_OUTPUT() method implementation for IPv6.
648  * 0 - no IPsec handling needed
649  * other values - mbuf consumed by IPsec.
650  */
651 int
652 ipsec6_output(struct mbuf *m, struct inpcb *inp)
653 {
654 
655 	/*
656 	 * If the packet is resubmitted to ip_output (e.g. after
657 	 * AH, ESP, etc. processing), there will be a tag to bypass
658 	 * the lookup and related policy checking.
659 	 */
660 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
661 		return (0);
662 
663 	return (ipsec6_common_output(m, inp, 0));
664 }
665 
666 /*
667  * IPSEC_FORWARD() method implementation for IPv6.
668  * 0 - no IPsec handling needed
669  * other values - mbuf consumed by IPsec.
670  */
671 int
672 ipsec6_forward(struct mbuf *m)
673 {
674 
675 	/*
676 	 * Check if this packet has an active inbound SP and needs to be
677 	 * dropped instead of forwarded.
678 	 */
679 	if (ipsec6_in_reject(m, NULL) != 0) {
680 		m_freem(m);
681 		return (EACCES);
682 	}
683 	return (ipsec6_common_output(m, NULL, 1));
684 }
685 #endif /* INET6 */
686 
687 int
688 ipsec_process_done(struct mbuf *m, struct secpolicy *sp, struct secasvar *sav,
689     u_int idx)
690 {
691 	struct epoch_tracker et;
692 	struct xform_history *xh;
693 	struct secasindex *saidx;
694 	struct m_tag *mtag;
695 	int error;
696 
697 	saidx = &sav->sah->saidx;
698 	switch (saidx->dst.sa.sa_family) {
699 #ifdef INET
700 	case AF_INET:
701 		/* Fix the header length, for AH processing. */
702 		mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
703 		break;
704 #endif /* INET */
705 #ifdef INET6
706 	case AF_INET6:
707 		/* Fix the header length, for AH processing. */
708 		if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
709 			error = ENXIO;
710 			goto bad;
711 		}
712 		if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
713 			/* No jumbogram support. */
714 			error = ENXIO;	/*?*/
715 			goto bad;
716 		}
717 		mtod(m, struct ip6_hdr *)->ip6_plen =
718 			htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
719 		break;
720 #endif /* INET6 */
721 	default:
722 		DPRINTF(("%s: unknown protocol family %u\n", __func__,
723 		    saidx->dst.sa.sa_family));
724 		error = ENXIO;
725 		goto bad;
726 	}
727 
728 	/*
729 	 * Add a record of what we've done to the packet.
730 	 */
731 	mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE, sizeof(*xh), M_NOWAIT);
732 	if (mtag == NULL) {
733 		DPRINTF(("%s: could not get packet tag\n", __func__));
734 		error = ENOMEM;
735 		goto bad;
736 	}
737 
738 	xh = (struct xform_history *)(mtag + 1);
739 	xh->dst = saidx->dst;
740 	xh->proto = saidx->proto;
741 	xh->mode = saidx->mode;
742 	xh->spi = sav->spi;
743 	m_tag_prepend(m, mtag);
744 
745 	key_sa_recordxfer(sav, m);		/* record data transfer */
746 
747 	/*
748 	 * If there's another (bundled) SA to apply, do so.
749 	 * Note that this puts a burden on the kernel stack size.
750 	 * If this is a problem we'll need to introduce a queue
751 	 * to set the packet on so we can unwind the stack before
752 	 * doing further processing.
753 	 */
754 	if (++idx < sp->tcount) {
755 		switch (saidx->dst.sa.sa_family) {
756 #ifdef INET
757 		case AF_INET:
758 			key_freesav(&sav);
759 			IPSECSTAT_INC(ips_out_bundlesa);
760 			return (ipsec4_perform_request(m, sp, NULL, idx));
761 			/* NOTREACHED */
762 #endif
763 #ifdef INET6
764 		case AF_INET6:
765 			key_freesav(&sav);
766 			IPSEC6STAT_INC(ips_out_bundlesa);
767 			return (ipsec6_perform_request(m, sp, NULL, idx));
768 			/* NOTREACHED */
769 #endif /* INET6 */
770 		default:
771 			DPRINTF(("%s: unknown protocol family %u\n", __func__,
772 			    saidx->dst.sa.sa_family));
773 			error = EPFNOSUPPORT;
774 			goto bad;
775 		}
776 	}
777 
778 	key_freesp(&sp), sp = NULL;	/* Release reference to SP */
779 #ifdef INET
780 	/*
781 	 * Do UDP encapsulation if SA requires it.
782 	 */
783 	if (sav->natt != NULL) {
784 		error = udp_ipsec_output(m, sav);
785 		if (error != 0)
786 			goto bad;
787 	}
788 #endif /* INET */
789 	/*
790 	 * We're done with IPsec processing, transmit the packet using the
791 	 * appropriate network protocol (IP or IPv6).
792 	 */
793 	NET_EPOCH_ENTER(et);
794 	switch (saidx->dst.sa.sa_family) {
795 #ifdef INET
796 	case AF_INET:
797 		key_freesav(&sav);
798 		error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
799 		break;
800 #endif /* INET */
801 #ifdef INET6
802 	case AF_INET6:
803 		key_freesav(&sav);
804 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
805 		break;
806 #endif /* INET6 */
807 	default:
808 		panic("ipsec_process_done");
809 	}
810 	NET_EPOCH_EXIT(et);
811 	return (error);
812 bad:
813 	m_freem(m);
814 	key_freesav(&sav);
815 	if (sp != NULL)
816 		key_freesp(&sp);
817 	return (error);
818 }
819 
820 /*
821  * ipsec_prepend() is optimized version of M_PREPEND().
822  * ipsec_encap() is called by IPsec output routine for tunnel mode SA.
823  * It is expected that after IP encapsulation some IPsec transform will
824  * be performed. Each IPsec transform inserts its variable length header
825  * just after outer IP header using m_makespace(). If given mbuf has not
826  * enough free space at the beginning, we allocate new mbuf and reserve
827  * some space at the beginning and at the end.
828  * This helps avoid allocating of new mbuf and data copying in m_makespace(),
829  * we place outer header in the middle of mbuf's data with reserved leading
830  * and trailing space:
831  *	[ LEADINGSPACE ][ Outer IP header ][ TRAILINGSPACE ]
832  * LEADINGSPACE will be used to add ethernet header, TRAILINGSPACE will
833  * be used to inject AH/ESP/IPCOMP header.
834  */
835 #define	IPSEC_TRAILINGSPACE	(sizeof(struct udphdr) +/* NAT-T */	\
836     max(sizeof(struct newesp) + EALG_MAX_BLOCK_LEN,	/* ESP + IV */	\
837 	sizeof(struct newah) + HASH_MAX_LEN		/* AH + ICV */))
838 static struct mbuf *
839 ipsec_prepend(struct mbuf *m, int len, int how)
840 {
841 	struct mbuf *n;
842 
843 	M_ASSERTPKTHDR(m);
844 	IPSEC_ASSERT(len < MHLEN, ("wrong length"));
845 	if (M_LEADINGSPACE(m) >= len) {
846 		/* No need to allocate new mbuf. */
847 		m->m_data -= len;
848 		m->m_len += len;
849 		m->m_pkthdr.len += len;
850 		return (m);
851 	}
852 	n = m_gethdr(how, m->m_type);
853 	if (n == NULL) {
854 		m_freem(m);
855 		return (NULL);
856 	}
857 	m_move_pkthdr(n, m);
858 	n->m_next = m;
859 	if (len + IPSEC_TRAILINGSPACE < M_SIZE(n))
860 		m_align(n, len + IPSEC_TRAILINGSPACE);
861 	n->m_len = len;
862 	n->m_pkthdr.len += len;
863 	return (n);
864 }
865 
866 static int
867 ipsec_encap(struct mbuf **mp, struct secasindex *saidx)
868 {
869 #ifdef INET6
870 	struct ip6_hdr *ip6;
871 #endif
872 	struct ip *ip;
873 	int setdf;
874 	uint8_t itos, proto;
875 
876 	ip = mtod(*mp, struct ip *);
877 	switch (ip->ip_v) {
878 #ifdef INET
879 	case IPVERSION:
880 		proto = IPPROTO_IPIP;
881 		/*
882 		 * Collect IP_DF state from the inner header
883 		 * and honor system-wide control of how to handle it.
884 		 */
885 		switch (V_ip4_ipsec_dfbit) {
886 		case 0:	/* clear in outer header */
887 		case 1:	/* set in outer header */
888 			setdf = V_ip4_ipsec_dfbit;
889 			break;
890 		default:/* propagate to outer header */
891 			setdf = (ip->ip_off & htons(IP_DF)) != 0;
892 		}
893 		itos = ip->ip_tos;
894 		break;
895 #endif
896 #ifdef INET6
897 	case (IPV6_VERSION >> 4):
898 		proto = IPPROTO_IPV6;
899 		ip6 = mtod(*mp, struct ip6_hdr *);
900 		itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
901 		setdf = V_ip4_ipsec_dfbit ? 1: 0;
902 		/* scoped address handling */
903 		in6_clearscope(&ip6->ip6_src);
904 		in6_clearscope(&ip6->ip6_dst);
905 		break;
906 #endif
907 	default:
908 		return (EAFNOSUPPORT);
909 	}
910 	switch (saidx->dst.sa.sa_family) {
911 #ifdef INET
912 	case AF_INET:
913 		if (saidx->src.sa.sa_family != AF_INET ||
914 		    saidx->src.sin.sin_addr.s_addr == INADDR_ANY ||
915 		    saidx->dst.sin.sin_addr.s_addr == INADDR_ANY)
916 			return (EINVAL);
917 		*mp = ipsec_prepend(*mp, sizeof(struct ip), M_NOWAIT);
918 		if (*mp == NULL)
919 			return (ENOBUFS);
920 		ip = mtod(*mp, struct ip *);
921 		ip->ip_v = IPVERSION;
922 		ip->ip_hl = sizeof(struct ip) >> 2;
923 		ip->ip_p = proto;
924 		ip->ip_len = htons((*mp)->m_pkthdr.len);
925 		ip->ip_ttl = V_ip_defttl;
926 		ip->ip_sum = 0;
927 		ip->ip_off = setdf ? htons(IP_DF): 0;
928 		ip->ip_src = saidx->src.sin.sin_addr;
929 		ip->ip_dst = saidx->dst.sin.sin_addr;
930 		ip_ecn_ingress(V_ip4_ipsec_ecn, &ip->ip_tos, &itos);
931 		ip_fillid(ip);
932 		break;
933 #endif /* INET */
934 #ifdef INET6
935 	case AF_INET6:
936 		if (saidx->src.sa.sa_family != AF_INET6 ||
937 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->src.sin6.sin6_addr) ||
938 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->dst.sin6.sin6_addr))
939 			return (EINVAL);
940 		*mp = ipsec_prepend(*mp, sizeof(struct ip6_hdr), M_NOWAIT);
941 		if (*mp == NULL)
942 			return (ENOBUFS);
943 		ip6 = mtod(*mp, struct ip6_hdr *);
944 		ip6->ip6_flow = 0;
945 		ip6->ip6_vfc = IPV6_VERSION;
946 		ip6->ip6_hlim = V_ip6_defhlim;
947 		ip6->ip6_nxt = proto;
948 		ip6->ip6_dst = saidx->dst.sin6.sin6_addr;
949 		/* For link-local address embed scope zone id */
950 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst))
951 			ip6->ip6_dst.s6_addr16[1] =
952 			    htons(saidx->dst.sin6.sin6_scope_id & 0xffff);
953 		ip6->ip6_src = saidx->src.sin6.sin6_addr;
954 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src))
955 			ip6->ip6_src.s6_addr16[1] =
956 			    htons(saidx->src.sin6.sin6_scope_id & 0xffff);
957 		ip6->ip6_plen = htons((*mp)->m_pkthdr.len - sizeof(*ip6));
958 		ip_ecn_ingress(V_ip6_ipsec_ecn, &proto, &itos);
959 		ip6->ip6_flow |= htonl((uint32_t)proto << 20);
960 		break;
961 #endif /* INET6 */
962 	default:
963 		return (EAFNOSUPPORT);
964 	}
965 	(*mp)->m_flags &= ~(M_BCAST | M_MCAST);
966 	return (0);
967 }
968