xref: /freebsd/sys/netipsec/ipsec_output.c (revision 8b959dd6a3921c35395bef4a6d7ad2426a3bd88e)
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 #include <netinet/ip_icmp.h>
65 #include <netinet/tcp_var.h>
66 
67 #include <netinet/ip6.h>
68 #ifdef INET6
69 #include <netinet6/ip6_var.h>
70 #include <netinet6/scope6_var.h>
71 #endif
72 #include <netinet/in_pcb.h>
73 #ifdef INET6
74 #include <netinet/icmp6.h>
75 #endif
76 #if defined(SCTP) || defined(SCTP_SUPPORT)
77 #include <netinet/sctp_crc32.h>
78 #endif
79 
80 #include <netinet/udp.h>
81 #include <netipsec/ah.h>
82 #include <netipsec/esp.h>
83 #include <netipsec/ipsec.h>
84 #ifdef INET6
85 #include <netipsec/ipsec6.h>
86 #endif
87 #include <netipsec/ah_var.h>
88 #include <netipsec/esp_var.h>
89 #include <netipsec/ipcomp_var.h>
90 
91 #include <netipsec/xform.h>
92 
93 #include <netipsec/key.h>
94 #include <netipsec/keydb.h>
95 #include <netipsec/key_debug.h>
96 
97 #include <machine/in_cksum.h>
98 
99 #define	IPSEC_OSTAT_INC(proto, name)	do {		\
100 	if ((proto) == IPPROTO_ESP)	\
101 		ESPSTAT_INC(esps_##name);	\
102 	else if ((proto) == IPPROTO_AH)\
103 		AHSTAT_INC(ahs_##name);		\
104 	else					\
105 		IPCOMPSTAT_INC(ipcomps_##name);	\
106 } while (0)
107 
108 static int ipsec_encap(struct mbuf **mp, struct secasindex *saidx);
109 static size_t ipsec_get_pmtu(struct secasvar *sav);
110 
111 #ifdef INET
112 static struct secasvar *
113 ipsec4_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
114 {
115 	struct secasindex *saidx, tmpsaidx;
116 	struct ipsecrequest *isr;
117 	struct sockaddr_in *sin;
118 	struct secasvar *sav;
119 	struct ip *ip;
120 
121 	/*
122 	 * Check system global policy controls.
123 	 */
124 next:
125 	isr = sp->req[*pidx];
126 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
127 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
128 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
129 		DPRINTF(("%s: IPsec outbound packet dropped due"
130 			" to policy (check your sysctls)\n", __func__));
131 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
132 		*error = EHOSTUNREACH;
133 		return (NULL);
134 	}
135 	/*
136 	 * Craft SA index to search for proper SA.  Note that
137 	 * we only initialize unspecified SA peers for transport
138 	 * mode; for tunnel mode they must already be filled in.
139 	 */
140 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
141 		saidx = &tmpsaidx;
142 		*saidx = isr->saidx;
143 		ip = mtod(m, struct ip *);
144 		if (saidx->src.sa.sa_len == 0) {
145 			sin = &saidx->src.sin;
146 			sin->sin_len = sizeof(*sin);
147 			sin->sin_family = AF_INET;
148 			sin->sin_port = IPSEC_PORT_ANY;
149 			sin->sin_addr = ip->ip_src;
150 		}
151 		if (saidx->dst.sa.sa_len == 0) {
152 			sin = &saidx->dst.sin;
153 			sin->sin_len = sizeof(*sin);
154 			sin->sin_family = AF_INET;
155 			sin->sin_port = IPSEC_PORT_ANY;
156 			sin->sin_addr = ip->ip_dst;
157 		}
158 	} else
159 		saidx = &sp->req[*pidx]->saidx;
160 	/*
161 	 * Lookup SA and validate it.
162 	 */
163 	sav = key_allocsa_policy(sp, saidx, error);
164 	if (sav == NULL) {
165 		IPSECSTAT_INC(ips_out_nosa);
166 		if (*error != 0)
167 			return (NULL);
168 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
169 			/*
170 			 * We have no SA and policy that doesn't require
171 			 * this IPsec transform, thus we can continue w/o
172 			 * IPsec processing, i.e. return EJUSTRETURN.
173 			 * But first check if there is some bundled transform.
174 			 */
175 			if (sp->tcount > ++(*pidx))
176 				goto next;
177 			*error = EJUSTRETURN;
178 		}
179 		return (NULL);
180 	}
181 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
182 	return (sav);
183 }
184 
185 /*
186  * IPsec output logic for IPv4.
187  */
188 static int
189 ipsec4_perform_request(struct mbuf *m, struct secpolicy *sp,
190     struct inpcb *inp, u_int idx)
191 {
192 	struct ipsec_ctx_data ctx;
193 	union sockaddr_union *dst;
194 	struct secasvar *sav;
195 	struct ip *ip;
196 	int error, i, off;
197 
198 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
199 
200 	/*
201 	 * We hold the reference to SP. Content of SP couldn't be changed.
202 	 * Craft secasindex and do lookup for suitable SA.
203 	 * Then do encapsulation if needed and call xform's output.
204 	 * We need to store SP in the xform callback parameters.
205 	 * In xform callback we will extract SP and it can be used to
206 	 * determine next transform. At the end of transform we can
207 	 * release reference to SP.
208 	 */
209 	sav = ipsec4_allocsa(m, sp, &idx, &error);
210 	if (sav == NULL) {
211 		if (error == EJUSTRETURN) { /* No IPsec required */
212 			key_freesp(&sp);
213 			return (error);
214 		}
215 		goto bad;
216 	}
217 	/*
218 	 * XXXAE: most likely ip_sum at this point is wrong.
219 	 */
220 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET, IPSEC_ENC_BEFORE);
221 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
222 		goto bad;
223 
224 	ip = mtod(m, struct ip *);
225 	dst = &sav->sah->saidx.dst;
226 	/* Do the appropriate encapsulation, if necessary */
227 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
228 	    dst->sa.sa_family != AF_INET ||	    /* PF mismatch */
229 	    (dst->sa.sa_family == AF_INET &&	    /* Proxy */
230 	     dst->sin.sin_addr.s_addr != INADDR_ANY &&
231 	     dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
232 		/* Fix IPv4 header checksum and length */
233 		ip->ip_len = htons(m->m_pkthdr.len);
234 		ip->ip_sum = 0;
235 		ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
236 		error = ipsec_encap(&m, &sav->sah->saidx);
237 		if (error != 0) {
238 			DPRINTF(("%s: encapsulation for SPI 0x%08x failed "
239 			    "with error %d\n", __func__, ntohl(sav->spi),
240 			    error));
241 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
242 			goto bad;
243 		}
244 		inp = NULL;
245 	}
246 
247 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, dst->sa.sa_family, IPSEC_ENC_AFTER);
248 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
249 		goto bad;
250 
251 	/*
252 	 * Dispatch to the appropriate IPsec transform logic.  The
253 	 * packet will be returned for transmission after crypto
254 	 * processing, etc. are completed.
255 	 *
256 	 * NB: m & sav are ``passed to caller'' who's responsible for
257 	 *     reclaiming their resources.
258 	 */
259 	switch(dst->sa.sa_family) {
260 	case AF_INET:
261 		ip = mtod(m, struct ip *);
262 		i = ip->ip_hl << 2;
263 		off = offsetof(struct ip, ip_p);
264 		break;
265 #ifdef INET6
266 	case AF_INET6:
267 		i = sizeof(struct ip6_hdr);
268 		off = offsetof(struct ip6_hdr, ip6_nxt);
269 		break;
270 #endif /* INET6 */
271 	default:
272 		DPRINTF(("%s: unsupported protocol family %u\n",
273 		    __func__, dst->sa.sa_family));
274 		error = EPFNOSUPPORT;
275 		IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf);
276 		goto bad;
277 	}
278 	error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off);
279 	return (error);
280 bad:
281 	IPSECSTAT_INC(ips_out_inval);
282 	if (m != NULL)
283 		m_freem(m);
284 	if (sav != NULL)
285 		key_freesav(&sav);
286 	key_freesp(&sp);
287 	return (error);
288 }
289 
290 int
291 ipsec4_process_packet(struct mbuf *m, struct secpolicy *sp,
292     struct inpcb *inp)
293 {
294 
295 	return (ipsec4_perform_request(m, sp, inp, 0));
296 }
297 
298 int
299 ipsec4_check_pmtu(struct mbuf *m, struct secpolicy *sp, int forwarding)
300 {
301 	struct secasvar *sav;
302 	struct ip *ip;
303 	size_t hlen, pmtu;
304 	uint32_t idx;
305 	int error;
306 
307 	/* Don't check PMTU if the frame won't have DF bit set. */
308 	if (!V_ip4_ipsec_dfbit)
309 		return (0);
310 	if (V_ip4_ipsec_dfbit == 1)
311 		goto setdf;
312 
313 	/* V_ip4_ipsec_dfbit > 1 - we will copy it from inner header. */
314 	ip = mtod(m, struct ip *);
315 	if (!(ip->ip_off & htons(IP_DF)))
316 		return (0);
317 
318 setdf:
319 	idx = sp->tcount - 1;
320 	sav = ipsec4_allocsa(m, sp, &idx, &error);
321 	if (sav == NULL) {
322 		key_freesp(&sp);
323 		/*
324 		 * No matching SA was found and SADB_ACQUIRE message was generated.
325 		 * Since we have matched a SP to this packet drop it silently.
326 		 */
327 		if (error == 0)
328 			error = EINPROGRESS;
329 		if (error != EJUSTRETURN)
330 			m_freem(m);
331 
332 		return (error);
333 	}
334 
335 	pmtu = ipsec_get_pmtu(sav);
336 	if (pmtu == 0) {
337 		key_freesav(&sav);
338 		return (0);
339 	}
340 
341 	hlen = ipsec_hdrsiz_internal(sp);
342 	key_freesav(&sav);
343 
344 	if (m_length(m, NULL) + hlen > pmtu) {
345 		/*
346 		 * If we're forwarding generate ICMP message here,
347 		 * so that it contains pmtu subtracted by header size.
348 		 * Set error to EINPROGRESS, in order for the frame
349 		 * to be dropped silently.
350 		 */
351 		if (forwarding) {
352 			if (pmtu > hlen)
353 				icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
354 				    0, pmtu - hlen);
355 			else
356 				m_freem(m);
357 
358 			key_freesp(&sp);
359 			return (EINPROGRESS); /* Pretend that we consumed it. */
360 		} else {
361 			m_freem(m);
362 			key_freesp(&sp);
363 			return (EMSGSIZE);
364 		}
365 	}
366 
367 	return (0);
368 }
369 
370 static int
371 ipsec4_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
372 {
373 	struct secpolicy *sp;
374 	int error;
375 
376 	/* Lookup for the corresponding outbound security policy */
377 	sp = ipsec4_checkpolicy(m, inp, &error, !forwarding);
378 	if (sp == NULL) {
379 		if (error == -EINVAL) {
380 			/* Discarded by policy. */
381 			m_freem(m);
382 			return (EACCES);
383 		}
384 		return (0); /* No IPsec required. */
385 	}
386 
387 	/*
388 	 * Usually we have to have tunnel mode IPsec security policy
389 	 * when we are forwarding a packet. Otherwise we could not handle
390 	 * encrypted replies, because they are not destined for us. But
391 	 * some users are doing source address translation for forwarded
392 	 * packets, and thus, even if they are forwarded, the replies will
393 	 * return back to us.
394 	 */
395 	if (!forwarding) {
396 		/*
397 		 * Do delayed checksums now because we send before
398 		 * this is done in the normal processing path.
399 		 */
400 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
401 			m = mb_unmapped_to_ext(m);
402 			if (m == NULL) {
403 				IPSECSTAT_INC(ips_out_nomem);
404 				key_freesp(&sp);
405 				return (ENOBUFS);
406 			}
407 			in_delayed_cksum(m);
408 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
409 		}
410 #if defined(SCTP) || defined(SCTP_SUPPORT)
411 		if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
412 			struct ip *ip;
413 
414 			m = mb_unmapped_to_ext(m);
415 			if (m == NULL) {
416 				IPSECSTAT_INC(ips_out_nomem);
417 				key_freesp(&sp);
418 				return (ENOBUFS);
419 			}
420 			ip = mtod(m, struct ip *);
421 			sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
422 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
423 		}
424 #endif
425 	}
426 	/* NB: callee frees mbuf and releases reference to SP */
427 	error = ipsec4_check_pmtu(m, sp, forwarding);
428 	if (error != 0) {
429 		if (error == EJUSTRETURN)
430 			return (0);
431 
432 		return (error);
433 	}
434 
435 	error = ipsec4_process_packet(m, sp, inp);
436 	if (error == EJUSTRETURN) {
437 		/*
438 		 * We had a SP with a level of 'use' and no SA. We
439 		 * will just continue to process the packet without
440 		 * IPsec processing and return without error.
441 		 */
442 		return (0);
443 	}
444 	if (error == 0)
445 		return (EINPROGRESS); /* consumed by IPsec */
446 	return (error);
447 }
448 
449 /*
450  * IPSEC_OUTPUT() method implementation for IPv4.
451  * 0 - no IPsec handling needed
452  * other values - mbuf consumed by IPsec.
453  */
454 int
455 ipsec4_output(struct mbuf *m, struct inpcb *inp)
456 {
457 
458 	/*
459 	 * If the packet is resubmitted to ip_output (e.g. after
460 	 * AH, ESP, etc. processing), there will be a tag to bypass
461 	 * the lookup and related policy checking.
462 	 */
463 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
464 		return (0);
465 
466 	return (ipsec4_common_output(m, inp, 0));
467 }
468 
469 /*
470  * IPSEC_FORWARD() method implementation for IPv4.
471  * 0 - no IPsec handling needed
472  * other values - mbuf consumed by IPsec.
473  */
474 int
475 ipsec4_forward(struct mbuf *m)
476 {
477 
478 	/*
479 	 * Check if this packet has an active inbound SP and needs to be
480 	 * dropped instead of forwarded.
481 	 */
482 	if (ipsec4_in_reject(m, NULL) != 0) {
483 		m_freem(m);
484 		return (EACCES);
485 	}
486 	return (ipsec4_common_output(m, NULL, 1));
487 }
488 #endif
489 
490 #ifdef INET6
491 static int
492 in6_sa_equal_addrwithscope(const struct sockaddr_in6 *sa,
493     const struct in6_addr *ia)
494 {
495 	struct in6_addr ia2;
496 
497 	if (IN6_IS_SCOPE_LINKLOCAL(&sa->sin6_addr)) {
498 		memcpy(&ia2, &sa->sin6_addr, sizeof(ia2));
499 		ia2.s6_addr16[1] = htons(sa->sin6_scope_id);
500 		return (IN6_ARE_ADDR_EQUAL(ia, &ia2));
501 	}
502 	return (IN6_ARE_ADDR_EQUAL(&sa->sin6_addr, ia));
503 }
504 
505 static struct secasvar *
506 ipsec6_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
507 {
508 	struct secasindex *saidx, tmpsaidx;
509 	struct ipsecrequest *isr;
510 	struct sockaddr_in6 *sin6;
511 	struct secasvar *sav;
512 	struct ip6_hdr *ip6;
513 
514 	/*
515 	 * Check system global policy controls.
516 	 */
517 next:
518 	isr = sp->req[*pidx];
519 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
520 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
521 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
522 		DPRINTF(("%s: IPsec outbound packet dropped due"
523 			" to policy (check your sysctls)\n", __func__));
524 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
525 		*error = EHOSTUNREACH;
526 		return (NULL);
527 	}
528 	/*
529 	 * Craft SA index to search for proper SA.  Note that
530 	 * we only fillin unspecified SA peers for transport
531 	 * mode; for tunnel mode they must already be filled in.
532 	 */
533 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
534 		saidx = &tmpsaidx;
535 		*saidx = isr->saidx;
536 		ip6 = mtod(m, struct ip6_hdr *);
537 		if (saidx->src.sin6.sin6_len == 0) {
538 			sin6 = (struct sockaddr_in6 *)&saidx->src;
539 			sin6->sin6_len = sizeof(*sin6);
540 			sin6->sin6_family = AF_INET6;
541 			sin6->sin6_port = IPSEC_PORT_ANY;
542 			sin6->sin6_addr = ip6->ip6_src;
543 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
544 				/* fix scope id for comparing SPD */
545 				sin6->sin6_addr.s6_addr16[1] = 0;
546 				sin6->sin6_scope_id =
547 				    ntohs(ip6->ip6_src.s6_addr16[1]);
548 			}
549 		}
550 		if (saidx->dst.sin6.sin6_len == 0) {
551 			sin6 = (struct sockaddr_in6 *)&saidx->dst;
552 			sin6->sin6_len = sizeof(*sin6);
553 			sin6->sin6_family = AF_INET6;
554 			sin6->sin6_port = IPSEC_PORT_ANY;
555 			sin6->sin6_addr = ip6->ip6_dst;
556 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
557 				/* fix scope id for comparing SPD */
558 				sin6->sin6_addr.s6_addr16[1] = 0;
559 				sin6->sin6_scope_id =
560 				    ntohs(ip6->ip6_dst.s6_addr16[1]);
561 			}
562 		}
563 	} else
564 		saidx = &sp->req[*pidx]->saidx;
565 	/*
566 	 * Lookup SA and validate it.
567 	 */
568 	sav = key_allocsa_policy(sp, saidx, error);
569 	if (sav == NULL) {
570 		IPSEC6STAT_INC(ips_out_nosa);
571 		if (*error != 0)
572 			return (NULL);
573 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
574 			/*
575 			 * We have no SA and policy that doesn't require
576 			 * this IPsec transform, thus we can continue w/o
577 			 * IPsec processing, i.e. return EJUSTRETURN.
578 			 * But first check if there is some bundled transform.
579 			 */
580 			if (sp->tcount > ++(*pidx))
581 				goto next;
582 			*error = EJUSTRETURN;
583 		}
584 		return (NULL);
585 	}
586 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
587 	return (sav);
588 }
589 
590 /*
591  * IPsec output logic for IPv6.
592  */
593 static int
594 ipsec6_perform_request(struct mbuf *m, struct secpolicy *sp,
595     struct inpcb *inp, u_int idx)
596 {
597 	struct ipsec_ctx_data ctx;
598 	union sockaddr_union *dst;
599 	struct secasvar *sav;
600 	struct ip6_hdr *ip6;
601 	int error, i, off;
602 
603 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
604 
605 	sav = ipsec6_allocsa(m, sp, &idx, &error);
606 	if (sav == NULL) {
607 		if (error == EJUSTRETURN) { /* No IPsec required */
608 			key_freesp(&sp);
609 			return (error);
610 		}
611 		goto bad;
612 	}
613 
614 	/* Fix IP length in case if it is not set yet. */
615 	ip6 = mtod(m, struct ip6_hdr *);
616 	ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
617 
618 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET6, IPSEC_ENC_BEFORE);
619 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
620 		goto bad;
621 
622 	ip6 = mtod(m, struct ip6_hdr *); /* pfil can change mbuf */
623 	dst = &sav->sah->saidx.dst;
624 
625 	/* Do the appropriate encapsulation, if necessary */
626 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
627 	    dst->sa.sa_family != AF_INET6 ||        /* PF mismatch */
628 	    ((dst->sa.sa_family == AF_INET6) &&
629 	     (!IN6_IS_ADDR_UNSPECIFIED(&dst->sin6.sin6_addr)) &&
630 	     (!in6_sa_equal_addrwithscope(&dst->sin6, &ip6->ip6_dst)))) {
631 		if (m->m_pkthdr.len - sizeof(*ip6) > IPV6_MAXPACKET) {
632 			/* No jumbogram support. */
633 			error = ENXIO;   /*XXX*/
634 			goto bad;
635 		}
636 		error = ipsec_encap(&m, &sav->sah->saidx);
637 		if (error != 0) {
638 			DPRINTF(("%s: encapsulation for SPI 0x%08x failed "
639 			    "with error %d\n", __func__, ntohl(sav->spi),
640 			    error));
641 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
642 			goto bad;
643 		}
644 		inp = NULL;
645 	}
646 
647 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, dst->sa.sa_family, IPSEC_ENC_AFTER);
648 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
649 		goto bad;
650 
651 	switch(dst->sa.sa_family) {
652 #ifdef INET
653 	case AF_INET:
654 		{
655 		struct ip *ip;
656 		ip = mtod(m, struct ip *);
657 		i = ip->ip_hl << 2;
658 		off = offsetof(struct ip, ip_p);
659 		}
660 		break;
661 #endif /* AF_INET */
662 	case AF_INET6:
663 		i = sizeof(struct ip6_hdr);
664 		off = offsetof(struct ip6_hdr, ip6_nxt);
665 		break;
666 	default:
667 		DPRINTF(("%s: unsupported protocol family %u\n",
668 				 __func__, dst->sa.sa_family));
669 		error = EPFNOSUPPORT;
670 		IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf);
671 		goto bad;
672 	}
673 	error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off);
674 	return (error);
675 bad:
676 	IPSEC6STAT_INC(ips_out_inval);
677 	if (m != NULL)
678 		m_freem(m);
679 	if (sav != NULL)
680 		key_freesav(&sav);
681 	key_freesp(&sp);
682 	return (error);
683 }
684 
685 int
686 ipsec6_process_packet(struct mbuf *m, struct secpolicy *sp,
687     struct inpcb *inp)
688 {
689 
690 	return (ipsec6_perform_request(m, sp, inp, 0));
691 }
692 
693 /*
694  * IPv6 implementation is based on IPv4 implementation.
695  */
696 int
697 ipsec6_check_pmtu(struct mbuf *m, struct secpolicy *sp, int forwarding)
698 {
699 	struct secasvar *sav;
700 	size_t hlen, pmtu;
701 	uint32_t idx;
702 	int error;
703 
704 	/*
705 	 * According to RFC8200 L3 fragmentation is supposed to be done only on
706 	 * locally generated packets. During L3 forwarding packets that are too
707 	 * big are always supposed to be dropped, with an ICMPv6 packet being
708 	 * sent back.
709 	 */
710 	if (!forwarding)
711 		return (0);
712 
713 	idx = sp->tcount - 1;
714 	sav = ipsec6_allocsa(m, sp, &idx, &error);
715 	if (sav == NULL) {
716 		key_freesp(&sp);
717 		/*
718 		 * No matching SA was found and SADB_ACQUIRE message was generated.
719 		 * Since we have matched a SP to this packet drop it silently.
720 		 */
721 		if (error == 0)
722 			error = EINPROGRESS;
723 		if (error != EJUSTRETURN)
724 			m_freem(m);
725 
726 		return (error);
727 	}
728 
729 	pmtu = ipsec_get_pmtu(sav);
730 	if (pmtu == 0) {
731 		key_freesav(&sav);
732 		return (0);
733 	}
734 
735 	hlen = ipsec_hdrsiz_internal(sp);
736 	key_freesav(&sav);
737 
738 	if (m_length(m, NULL) + hlen > pmtu) {
739 		/*
740 		 * If we're forwarding generate ICMPv6 message here,
741 		 * so that it contains pmtu subtracted by header size.
742 		 * Set error to EINPROGRESS, in order for the frame
743 		 * to be dropped silently.
744 		 */
745 		if (forwarding) {
746 			if (pmtu > hlen)
747 				icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, pmtu - hlen);
748 			else
749 				m_freem(m);
750 
751 			key_freesp(&sp);
752 			return (EINPROGRESS); /* Pretend that we consumed it. */
753 		}
754 	}
755 
756 	return (0);
757 }
758 
759 static int
760 ipsec6_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
761 {
762 	struct secpolicy *sp;
763 	int error;
764 
765 	/* Lookup for the corresponding outbound security policy */
766 	sp = ipsec6_checkpolicy(m, inp, &error, !forwarding);
767 	if (sp == NULL) {
768 		if (error == -EINVAL) {
769 			/* Discarded by policy. */
770 			m_freem(m);
771 			return (EACCES);
772 		}
773 		return (0); /* No IPsec required. */
774 	}
775 
776 	if (!forwarding) {
777 		/*
778 		 * Do delayed checksums now because we send before
779 		 * this is done in the normal processing path.
780 		 */
781 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
782 			m = mb_unmapped_to_ext(m);
783 			if (m == NULL) {
784 				IPSEC6STAT_INC(ips_out_nomem);
785 				key_freesp(&sp);
786 				return (ENOBUFS);
787 			}
788 			in6_delayed_cksum(m, m->m_pkthdr.len -
789 			    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
790 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
791 		}
792 #if defined(SCTP) || defined(SCTP_SUPPORT)
793 		if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
794 			m = mb_unmapped_to_ext(m);
795 			if (m == NULL) {
796 				IPSEC6STAT_INC(ips_out_nomem);
797 				key_freesp(&sp);
798 				return (ENOBUFS);
799 			}
800 			sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
801 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
802 		}
803 #endif
804 	}
805 
806 	error = ipsec6_check_pmtu(m, sp, forwarding);
807 	if (error != 0) {
808 		if (error == EJUSTRETURN)
809 			return (0);
810 
811 		return (error);
812 	}
813 
814 	/* NB: callee frees mbuf and releases reference to SP */
815 	error = ipsec6_process_packet(m, sp, inp);
816 	if (error == EJUSTRETURN) {
817 		/*
818 		 * We had a SP with a level of 'use' and no SA. We
819 		 * will just continue to process the packet without
820 		 * IPsec processing and return without error.
821 		 */
822 		return (0);
823 	}
824 	if (error == 0)
825 		return (EINPROGRESS); /* consumed by IPsec */
826 	return (error);
827 }
828 
829 /*
830  * IPSEC_OUTPUT() method implementation for IPv6.
831  * 0 - no IPsec handling needed
832  * other values - mbuf consumed by IPsec.
833  */
834 int
835 ipsec6_output(struct mbuf *m, struct inpcb *inp)
836 {
837 
838 	/*
839 	 * If the packet is resubmitted to ip_output (e.g. after
840 	 * AH, ESP, etc. processing), there will be a tag to bypass
841 	 * the lookup and related policy checking.
842 	 */
843 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
844 		return (0);
845 
846 	return (ipsec6_common_output(m, inp, 0));
847 }
848 
849 /*
850  * IPSEC_FORWARD() method implementation for IPv6.
851  * 0 - no IPsec handling needed
852  * other values - mbuf consumed by IPsec.
853  */
854 int
855 ipsec6_forward(struct mbuf *m)
856 {
857 
858 	/*
859 	 * Check if this packet has an active inbound SP and needs to be
860 	 * dropped instead of forwarded.
861 	 */
862 	if (ipsec6_in_reject(m, NULL) != 0) {
863 		m_freem(m);
864 		return (EACCES);
865 	}
866 	return (ipsec6_common_output(m, NULL, 1));
867 }
868 #endif /* INET6 */
869 
870 int
871 ipsec_process_done(struct mbuf *m, struct secpolicy *sp, struct secasvar *sav,
872     u_int idx)
873 {
874 	struct epoch_tracker et;
875 	struct xform_history *xh;
876 	struct secasindex *saidx;
877 	struct m_tag *mtag;
878 	int error;
879 
880 	saidx = &sav->sah->saidx;
881 	switch (saidx->dst.sa.sa_family) {
882 #ifdef INET
883 	case AF_INET:
884 		/* Fix the header length, for AH processing. */
885 		mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
886 		break;
887 #endif /* INET */
888 #ifdef INET6
889 	case AF_INET6:
890 		/* Fix the header length, for AH processing. */
891 		if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
892 			error = ENXIO;
893 			goto bad;
894 		}
895 		if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
896 			/* No jumbogram support. */
897 			error = ENXIO;	/*?*/
898 			goto bad;
899 		}
900 		mtod(m, struct ip6_hdr *)->ip6_plen =
901 			htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
902 		break;
903 #endif /* INET6 */
904 	default:
905 		DPRINTF(("%s: unknown protocol family %u\n", __func__,
906 		    saidx->dst.sa.sa_family));
907 		error = ENXIO;
908 		goto bad;
909 	}
910 
911 	/*
912 	 * Add a record of what we've done to the packet.
913 	 */
914 	mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE, sizeof(*xh), M_NOWAIT);
915 	if (mtag == NULL) {
916 		DPRINTF(("%s: could not get packet tag\n", __func__));
917 		error = ENOMEM;
918 		goto bad;
919 	}
920 
921 	xh = (struct xform_history *)(mtag + 1);
922 	xh->dst = saidx->dst;
923 	xh->proto = saidx->proto;
924 	xh->mode = saidx->mode;
925 	xh->spi = sav->spi;
926 	m_tag_prepend(m, mtag);
927 
928 	key_sa_recordxfer(sav, m);		/* record data transfer */
929 
930 	/*
931 	 * If there's another (bundled) SA to apply, do so.
932 	 * Note that this puts a burden on the kernel stack size.
933 	 * If this is a problem we'll need to introduce a queue
934 	 * to set the packet on so we can unwind the stack before
935 	 * doing further processing.
936 	 */
937 	if (++idx < sp->tcount) {
938 		switch (saidx->dst.sa.sa_family) {
939 #ifdef INET
940 		case AF_INET:
941 			key_freesav(&sav);
942 			IPSECSTAT_INC(ips_out_bundlesa);
943 			return (ipsec4_perform_request(m, sp, NULL, idx));
944 			/* NOTREACHED */
945 #endif
946 #ifdef INET6
947 		case AF_INET6:
948 			key_freesav(&sav);
949 			IPSEC6STAT_INC(ips_out_bundlesa);
950 			return (ipsec6_perform_request(m, sp, NULL, idx));
951 			/* NOTREACHED */
952 #endif /* INET6 */
953 		default:
954 			DPRINTF(("%s: unknown protocol family %u\n", __func__,
955 			    saidx->dst.sa.sa_family));
956 			error = EPFNOSUPPORT;
957 			goto bad;
958 		}
959 	}
960 
961 	key_freesp(&sp), sp = NULL;	/* Release reference to SP */
962 #ifdef INET
963 	/*
964 	 * Do UDP encapsulation if SA requires it.
965 	 */
966 	if (sav->natt != NULL) {
967 		error = udp_ipsec_output(m, sav);
968 		if (error != 0)
969 			goto bad;
970 	}
971 #endif /* INET */
972 	/*
973 	 * We're done with IPsec processing, transmit the packet using the
974 	 * appropriate network protocol (IP or IPv6).
975 	 */
976 	NET_EPOCH_ENTER(et);
977 	switch (saidx->dst.sa.sa_family) {
978 #ifdef INET
979 	case AF_INET:
980 		key_freesav(&sav);
981 		error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
982 		break;
983 #endif /* INET */
984 #ifdef INET6
985 	case AF_INET6:
986 		key_freesav(&sav);
987 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
988 		break;
989 #endif /* INET6 */
990 	default:
991 		panic("ipsec_process_done");
992 	}
993 	NET_EPOCH_EXIT(et);
994 	return (error);
995 bad:
996 	m_freem(m);
997 	key_freesav(&sav);
998 	if (sp != NULL)
999 		key_freesp(&sp);
1000 	return (error);
1001 }
1002 
1003 /*
1004  * ipsec_prepend() is optimized version of M_PREPEND().
1005  * ipsec_encap() is called by IPsec output routine for tunnel mode SA.
1006  * It is expected that after IP encapsulation some IPsec transform will
1007  * be performed. Each IPsec transform inserts its variable length header
1008  * just after outer IP header using m_makespace(). If given mbuf has not
1009  * enough free space at the beginning, we allocate new mbuf and reserve
1010  * some space at the beginning and at the end.
1011  * This helps avoid allocating of new mbuf and data copying in m_makespace(),
1012  * we place outer header in the middle of mbuf's data with reserved leading
1013  * and trailing space:
1014  *	[ LEADINGSPACE ][ Outer IP header ][ TRAILINGSPACE ]
1015  * LEADINGSPACE will be used to add ethernet header, TRAILINGSPACE will
1016  * be used to inject AH/ESP/IPCOMP header.
1017  */
1018 #define	IPSEC_TRAILINGSPACE	(sizeof(struct udphdr) +/* NAT-T */	\
1019     max(sizeof(struct newesp) + EALG_MAX_BLOCK_LEN,	/* ESP + IV */	\
1020 	sizeof(struct newah) + HASH_MAX_LEN		/* AH + ICV */))
1021 static struct mbuf *
1022 ipsec_prepend(struct mbuf *m, int len, int how)
1023 {
1024 	struct mbuf *n;
1025 
1026 	M_ASSERTPKTHDR(m);
1027 	IPSEC_ASSERT(len < MHLEN, ("wrong length"));
1028 	if (M_LEADINGSPACE(m) >= len) {
1029 		/* No need to allocate new mbuf. */
1030 		m->m_data -= len;
1031 		m->m_len += len;
1032 		m->m_pkthdr.len += len;
1033 		return (m);
1034 	}
1035 	n = m_gethdr(how, m->m_type);
1036 	if (n == NULL) {
1037 		m_freem(m);
1038 		return (NULL);
1039 	}
1040 	m_move_pkthdr(n, m);
1041 	n->m_next = m;
1042 	if (len + IPSEC_TRAILINGSPACE < M_SIZE(n))
1043 		m_align(n, len + IPSEC_TRAILINGSPACE);
1044 	n->m_len = len;
1045 	n->m_pkthdr.len += len;
1046 	return (n);
1047 }
1048 
1049 static size_t
1050 ipsec_get_pmtu(struct secasvar *sav)
1051 {
1052 	union sockaddr_union *dst;
1053 	struct in_conninfo inc;
1054 	size_t pmtu;
1055 
1056 	dst = &sav->sah->saidx.dst;
1057 	memset(&inc, 0, sizeof(inc));
1058 
1059 	switch (dst->sa.sa_family) {
1060 #ifdef INET
1061 	case AF_INET:
1062 		inc.inc_faddr = satosin(&dst->sa)->sin_addr;
1063 		break;
1064 #endif
1065 #ifdef INET6
1066 	case AF_INET6:
1067 		inc.inc6_faddr = satosin6(&dst->sa)->sin6_addr;
1068 		inc.inc_flags |= INC_ISIPV6;
1069 		break;
1070 #endif
1071 	default:
1072 		return (0);
1073 	}
1074 
1075 	pmtu = tcp_hc_getmtu(&inc);
1076 	if (pmtu != 0)
1077 		return (pmtu);
1078 
1079 	/* No entry in hostcache. Assume that PMTU is equal to link's MTU */
1080 	switch (dst->sa.sa_family) {
1081 #ifdef INET
1082 	case AF_INET:
1083 		pmtu = tcp_maxmtu(&inc, NULL);
1084 		break;
1085 #endif
1086 #ifdef INET6
1087 	case AF_INET6:
1088 		pmtu = tcp_maxmtu6(&inc, NULL);
1089 		break;
1090 #endif
1091 	default:
1092 		return (0);
1093 	}
1094 	if (pmtu == 0)
1095 		return (0);
1096 
1097 	tcp_hc_updatemtu(&inc, pmtu);
1098 
1099 	return (pmtu);
1100 }
1101 
1102 static int
1103 ipsec_encap(struct mbuf **mp, struct secasindex *saidx)
1104 {
1105 #ifdef INET6
1106 	struct ip6_hdr *ip6;
1107 #endif
1108 	struct ip *ip;
1109 	int setdf;
1110 	uint8_t itos, proto;
1111 
1112 	ip = mtod(*mp, struct ip *);
1113 	switch (ip->ip_v) {
1114 #ifdef INET
1115 	case IPVERSION:
1116 		proto = IPPROTO_IPIP;
1117 		/*
1118 		 * Collect IP_DF state from the inner header
1119 		 * and honor system-wide control of how to handle it.
1120 		 */
1121 		switch (V_ip4_ipsec_dfbit) {
1122 		case 0:	/* clear in outer header */
1123 		case 1:	/* set in outer header */
1124 			setdf = V_ip4_ipsec_dfbit;
1125 			break;
1126 		default:/* propagate to outer header */
1127 			setdf = (ip->ip_off & htons(IP_DF)) != 0;
1128 		}
1129 		itos = ip->ip_tos;
1130 		break;
1131 #endif
1132 #ifdef INET6
1133 	case (IPV6_VERSION >> 4):
1134 		proto = IPPROTO_IPV6;
1135 		ip6 = mtod(*mp, struct ip6_hdr *);
1136 		itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1137 		setdf = V_ip4_ipsec_dfbit ? 1: 0;
1138 		/* scoped address handling */
1139 		in6_clearscope(&ip6->ip6_src);
1140 		in6_clearscope(&ip6->ip6_dst);
1141 		break;
1142 #endif
1143 	default:
1144 		return (EAFNOSUPPORT);
1145 	}
1146 	switch (saidx->dst.sa.sa_family) {
1147 #ifdef INET
1148 	case AF_INET:
1149 		if (saidx->src.sa.sa_family != AF_INET ||
1150 		    saidx->src.sin.sin_addr.s_addr == INADDR_ANY ||
1151 		    saidx->dst.sin.sin_addr.s_addr == INADDR_ANY)
1152 			return (EINVAL);
1153 		*mp = ipsec_prepend(*mp, sizeof(struct ip), M_NOWAIT);
1154 		if (*mp == NULL)
1155 			return (ENOBUFS);
1156 		ip = mtod(*mp, struct ip *);
1157 		ip->ip_v = IPVERSION;
1158 		ip->ip_hl = sizeof(struct ip) >> 2;
1159 		ip->ip_p = proto;
1160 		ip->ip_len = htons((*mp)->m_pkthdr.len);
1161 		ip->ip_ttl = V_ip_defttl;
1162 		ip->ip_sum = 0;
1163 		ip->ip_off = setdf ? htons(IP_DF): 0;
1164 		ip->ip_src = saidx->src.sin.sin_addr;
1165 		ip->ip_dst = saidx->dst.sin.sin_addr;
1166 		ip_ecn_ingress(V_ip4_ipsec_ecn, &ip->ip_tos, &itos);
1167 		ip_fillid(ip);
1168 		break;
1169 #endif /* INET */
1170 #ifdef INET6
1171 	case AF_INET6:
1172 		if (saidx->src.sa.sa_family != AF_INET6 ||
1173 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->src.sin6.sin6_addr) ||
1174 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->dst.sin6.sin6_addr))
1175 			return (EINVAL);
1176 		*mp = ipsec_prepend(*mp, sizeof(struct ip6_hdr), M_NOWAIT);
1177 		if (*mp == NULL)
1178 			return (ENOBUFS);
1179 		ip6 = mtod(*mp, struct ip6_hdr *);
1180 		ip6->ip6_flow = 0;
1181 		ip6->ip6_vfc = IPV6_VERSION;
1182 		ip6->ip6_hlim = V_ip6_defhlim;
1183 		ip6->ip6_nxt = proto;
1184 		ip6->ip6_dst = saidx->dst.sin6.sin6_addr;
1185 		/* For link-local address embed scope zone id */
1186 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst))
1187 			ip6->ip6_dst.s6_addr16[1] =
1188 			    htons(saidx->dst.sin6.sin6_scope_id & 0xffff);
1189 		ip6->ip6_src = saidx->src.sin6.sin6_addr;
1190 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src))
1191 			ip6->ip6_src.s6_addr16[1] =
1192 			    htons(saidx->src.sin6.sin6_scope_id & 0xffff);
1193 		ip6->ip6_plen = htons((*mp)->m_pkthdr.len - sizeof(*ip6));
1194 		ip_ecn_ingress(V_ip6_ipsec_ecn, &proto, &itos);
1195 		ip6->ip6_flow |= htonl((uint32_t)proto << 20);
1196 		break;
1197 #endif /* INET6 */
1198 	default:
1199 		return (EAFNOSUPPORT);
1200 	}
1201 	(*mp)->m_flags &= ~(M_BCAST | M_MCAST);
1202 	return (0);
1203 }
1204