xref: /freebsd/sys/netipsec/ipsec_output.c (revision 8e9313ca)
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 			m = mb_unmapped_to_ext(m);
327 			if (m == NULL) {
328 				IPSECSTAT_INC(ips_out_nomem);
329 				key_freesp(&sp);
330 				return (ENOBUFS);
331 			}
332 			in_delayed_cksum(m);
333 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
334 		}
335 #if defined(SCTP) || defined(SCTP_SUPPORT)
336 		if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
337 			struct ip *ip;
338 
339 			m = mb_unmapped_to_ext(m);
340 			if (m == NULL) {
341 				IPSECSTAT_INC(ips_out_nomem);
342 				key_freesp(&sp);
343 				return (ENOBUFS);
344 			}
345 			ip = mtod(m, struct ip *);
346 			sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
347 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
348 		}
349 #endif
350 	}
351 	/* NB: callee frees mbuf and releases reference to SP */
352 	error = ipsec4_process_packet(m, sp, inp);
353 	if (error == EJUSTRETURN) {
354 		/*
355 		 * We had a SP with a level of 'use' and no SA. We
356 		 * will just continue to process the packet without
357 		 * IPsec processing and return without error.
358 		 */
359 		return (0);
360 	}
361 	if (error == 0)
362 		return (EINPROGRESS); /* consumed by IPsec */
363 	return (error);
364 }
365 
366 /*
367  * IPSEC_OUTPUT() method implementation for IPv4.
368  * 0 - no IPsec handling needed
369  * other values - mbuf consumed by IPsec.
370  */
371 int
372 ipsec4_output(struct mbuf *m, struct inpcb *inp)
373 {
374 
375 	/*
376 	 * If the packet is resubmitted to ip_output (e.g. after
377 	 * AH, ESP, etc. processing), there will be a tag to bypass
378 	 * the lookup and related policy checking.
379 	 */
380 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
381 		return (0);
382 
383 	return (ipsec4_common_output(m, inp, 0));
384 }
385 
386 /*
387  * IPSEC_FORWARD() method implementation for IPv4.
388  * 0 - no IPsec handling needed
389  * other values - mbuf consumed by IPsec.
390  */
391 int
392 ipsec4_forward(struct mbuf *m)
393 {
394 
395 	/*
396 	 * Check if this packet has an active inbound SP and needs to be
397 	 * dropped instead of forwarded.
398 	 */
399 	if (ipsec4_in_reject(m, NULL) != 0) {
400 		m_freem(m);
401 		return (EACCES);
402 	}
403 	return (ipsec4_common_output(m, NULL, 1));
404 }
405 #endif
406 
407 #ifdef INET6
408 static int
409 in6_sa_equal_addrwithscope(const struct sockaddr_in6 *sa,
410     const struct in6_addr *ia)
411 {
412 	struct in6_addr ia2;
413 
414 	if (IN6_IS_SCOPE_LINKLOCAL(&sa->sin6_addr)) {
415 		memcpy(&ia2, &sa->sin6_addr, sizeof(ia2));
416 		ia2.s6_addr16[1] = htons(sa->sin6_scope_id);
417 		return (IN6_ARE_ADDR_EQUAL(ia, &ia2));
418 	}
419 	return (IN6_ARE_ADDR_EQUAL(&sa->sin6_addr, ia));
420 }
421 
422 static struct secasvar *
423 ipsec6_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
424 {
425 	struct secasindex *saidx, tmpsaidx;
426 	struct ipsecrequest *isr;
427 	struct sockaddr_in6 *sin6;
428 	struct secasvar *sav;
429 	struct ip6_hdr *ip6;
430 
431 	/*
432 	 * Check system global policy controls.
433 	 */
434 next:
435 	isr = sp->req[*pidx];
436 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
437 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
438 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
439 		DPRINTF(("%s: IPsec outbound packet dropped due"
440 			" to policy (check your sysctls)\n", __func__));
441 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
442 		*error = EHOSTUNREACH;
443 		return (NULL);
444 	}
445 	/*
446 	 * Craft SA index to search for proper SA.  Note that
447 	 * we only fillin unspecified SA peers for transport
448 	 * mode; for tunnel mode they must already be filled in.
449 	 */
450 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
451 		saidx = &tmpsaidx;
452 		*saidx = isr->saidx;
453 		ip6 = mtod(m, struct ip6_hdr *);
454 		if (saidx->src.sin6.sin6_len == 0) {
455 			sin6 = (struct sockaddr_in6 *)&saidx->src;
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_src;
460 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
461 				/* fix scope id for comparing SPD */
462 				sin6->sin6_addr.s6_addr16[1] = 0;
463 				sin6->sin6_scope_id =
464 				    ntohs(ip6->ip6_src.s6_addr16[1]);
465 			}
466 		}
467 		if (saidx->dst.sin6.sin6_len == 0) {
468 			sin6 = (struct sockaddr_in6 *)&saidx->dst;
469 			sin6->sin6_len = sizeof(*sin6);
470 			sin6->sin6_family = AF_INET6;
471 			sin6->sin6_port = IPSEC_PORT_ANY;
472 			sin6->sin6_addr = ip6->ip6_dst;
473 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
474 				/* fix scope id for comparing SPD */
475 				sin6->sin6_addr.s6_addr16[1] = 0;
476 				sin6->sin6_scope_id =
477 				    ntohs(ip6->ip6_dst.s6_addr16[1]);
478 			}
479 		}
480 	} else
481 		saidx = &sp->req[*pidx]->saidx;
482 	/*
483 	 * Lookup SA and validate it.
484 	 */
485 	sav = key_allocsa_policy(sp, saidx, error);
486 	if (sav == NULL) {
487 		IPSEC6STAT_INC(ips_out_nosa);
488 		if (*error != 0)
489 			return (NULL);
490 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
491 			/*
492 			 * We have no SA and policy that doesn't require
493 			 * this IPsec transform, thus we can continue w/o
494 			 * IPsec processing, i.e. return EJUSTRETURN.
495 			 * But first check if there is some bundled transform.
496 			 */
497 			if (sp->tcount > ++(*pidx))
498 				goto next;
499 			*error = EJUSTRETURN;
500 		}
501 		return (NULL);
502 	}
503 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
504 	return (sav);
505 }
506 
507 /*
508  * IPsec output logic for IPv6.
509  */
510 static int
511 ipsec6_perform_request(struct mbuf *m, struct secpolicy *sp,
512     struct inpcb *inp, u_int idx)
513 {
514 	struct ipsec_ctx_data ctx;
515 	union sockaddr_union *dst;
516 	struct secasvar *sav;
517 	struct ip6_hdr *ip6;
518 	int error, i, off;
519 
520 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
521 
522 	sav = ipsec6_allocsa(m, sp, &idx, &error);
523 	if (sav == NULL) {
524 		if (error == EJUSTRETURN) { /* No IPsec required */
525 			key_freesp(&sp);
526 			return (error);
527 		}
528 		goto bad;
529 	}
530 
531 	/* Fix IP length in case if it is not set yet. */
532 	ip6 = mtod(m, struct ip6_hdr *);
533 	ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
534 
535 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET6, IPSEC_ENC_BEFORE);
536 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
537 		goto bad;
538 
539 	ip6 = mtod(m, struct ip6_hdr *); /* pfil can change mbuf */
540 	dst = &sav->sah->saidx.dst;
541 
542 	/* Do the appropriate encapsulation, if necessary */
543 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
544 	    dst->sa.sa_family != AF_INET6 ||        /* PF mismatch */
545 	    ((dst->sa.sa_family == AF_INET6) &&
546 	     (!IN6_IS_ADDR_UNSPECIFIED(&dst->sin6.sin6_addr)) &&
547 	     (!in6_sa_equal_addrwithscope(&dst->sin6, &ip6->ip6_dst)))) {
548 		if (m->m_pkthdr.len - sizeof(*ip6) > IPV6_MAXPACKET) {
549 			/* No jumbogram support. */
550 			error = ENXIO;   /*XXX*/
551 			goto bad;
552 		}
553 		error = ipsec_encap(&m, &sav->sah->saidx);
554 		if (error != 0) {
555 			DPRINTF(("%s: encapsulation for SPI 0x%08x failed "
556 			    "with error %d\n", __func__, ntohl(sav->spi),
557 			    error));
558 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
559 			goto bad;
560 		}
561 		inp = NULL;
562 	}
563 
564 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, dst->sa.sa_family, IPSEC_ENC_AFTER);
565 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
566 		goto bad;
567 
568 	switch(dst->sa.sa_family) {
569 #ifdef INET
570 	case AF_INET:
571 		{
572 		struct ip *ip;
573 		ip = mtod(m, struct ip *);
574 		i = ip->ip_hl << 2;
575 		off = offsetof(struct ip, ip_p);
576 		}
577 		break;
578 #endif /* AF_INET */
579 	case AF_INET6:
580 		i = sizeof(struct ip6_hdr);
581 		off = offsetof(struct ip6_hdr, ip6_nxt);
582 		break;
583 	default:
584 		DPRINTF(("%s: unsupported protocol family %u\n",
585 				 __func__, dst->sa.sa_family));
586 		error = EPFNOSUPPORT;
587 		IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf);
588 		goto bad;
589 	}
590 	error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off);
591 	return (error);
592 bad:
593 	IPSEC6STAT_INC(ips_out_inval);
594 	if (m != NULL)
595 		m_freem(m);
596 	if (sav != NULL)
597 		key_freesav(&sav);
598 	key_freesp(&sp);
599 	return (error);
600 }
601 
602 int
603 ipsec6_process_packet(struct mbuf *m, struct secpolicy *sp,
604     struct inpcb *inp)
605 {
606 
607 	return (ipsec6_perform_request(m, sp, inp, 0));
608 }
609 
610 static int
611 ipsec6_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
612 {
613 	struct secpolicy *sp;
614 	int error;
615 
616 	/* Lookup for the corresponding outbound security policy */
617 	sp = ipsec6_checkpolicy(m, inp, &error, !forwarding);
618 	if (sp == NULL) {
619 		if (error == -EINVAL) {
620 			/* Discarded by policy. */
621 			m_freem(m);
622 			return (EACCES);
623 		}
624 		return (0); /* No IPsec required. */
625 	}
626 
627 	if (!forwarding) {
628 		/*
629 		 * Do delayed checksums now because we send before
630 		 * this is done in the normal processing path.
631 		 */
632 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
633 			m = mb_unmapped_to_ext(m);
634 			if (m == NULL) {
635 				IPSEC6STAT_INC(ips_out_nomem);
636 				key_freesp(&sp);
637 				return (ENOBUFS);
638 			}
639 			in6_delayed_cksum(m, m->m_pkthdr.len -
640 			    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
641 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
642 		}
643 #if defined(SCTP) || defined(SCTP_SUPPORT)
644 		if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
645 			m = mb_unmapped_to_ext(m);
646 			if (m == NULL) {
647 				IPSEC6STAT_INC(ips_out_nomem);
648 				key_freesp(&sp);
649 				return (ENOBUFS);
650 			}
651 			sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
652 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
653 		}
654 #endif
655 	}
656 	/* NB: callee frees mbuf and releases reference to SP */
657 	error = ipsec6_process_packet(m, sp, inp);
658 	if (error == EJUSTRETURN) {
659 		/*
660 		 * We had a SP with a level of 'use' and no SA. We
661 		 * will just continue to process the packet without
662 		 * IPsec processing and return without error.
663 		 */
664 		return (0);
665 	}
666 	if (error == 0)
667 		return (EINPROGRESS); /* consumed by IPsec */
668 	return (error);
669 }
670 
671 /*
672  * IPSEC_OUTPUT() method implementation for IPv6.
673  * 0 - no IPsec handling needed
674  * other values - mbuf consumed by IPsec.
675  */
676 int
677 ipsec6_output(struct mbuf *m, struct inpcb *inp)
678 {
679 
680 	/*
681 	 * If the packet is resubmitted to ip_output (e.g. after
682 	 * AH, ESP, etc. processing), there will be a tag to bypass
683 	 * the lookup and related policy checking.
684 	 */
685 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
686 		return (0);
687 
688 	return (ipsec6_common_output(m, inp, 0));
689 }
690 
691 /*
692  * IPSEC_FORWARD() method implementation for IPv6.
693  * 0 - no IPsec handling needed
694  * other values - mbuf consumed by IPsec.
695  */
696 int
697 ipsec6_forward(struct mbuf *m)
698 {
699 
700 	/*
701 	 * Check if this packet has an active inbound SP and needs to be
702 	 * dropped instead of forwarded.
703 	 */
704 	if (ipsec6_in_reject(m, NULL) != 0) {
705 		m_freem(m);
706 		return (EACCES);
707 	}
708 	return (ipsec6_common_output(m, NULL, 1));
709 }
710 #endif /* INET6 */
711 
712 int
713 ipsec_process_done(struct mbuf *m, struct secpolicy *sp, struct secasvar *sav,
714     u_int idx)
715 {
716 	struct epoch_tracker et;
717 	struct xform_history *xh;
718 	struct secasindex *saidx;
719 	struct m_tag *mtag;
720 	int error;
721 
722 	saidx = &sav->sah->saidx;
723 	switch (saidx->dst.sa.sa_family) {
724 #ifdef INET
725 	case AF_INET:
726 		/* Fix the header length, for AH processing. */
727 		mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
728 		break;
729 #endif /* INET */
730 #ifdef INET6
731 	case AF_INET6:
732 		/* Fix the header length, for AH processing. */
733 		if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
734 			error = ENXIO;
735 			goto bad;
736 		}
737 		if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
738 			/* No jumbogram support. */
739 			error = ENXIO;	/*?*/
740 			goto bad;
741 		}
742 		mtod(m, struct ip6_hdr *)->ip6_plen =
743 			htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
744 		break;
745 #endif /* INET6 */
746 	default:
747 		DPRINTF(("%s: unknown protocol family %u\n", __func__,
748 		    saidx->dst.sa.sa_family));
749 		error = ENXIO;
750 		goto bad;
751 	}
752 
753 	/*
754 	 * Add a record of what we've done to the packet.
755 	 */
756 	mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE, sizeof(*xh), M_NOWAIT);
757 	if (mtag == NULL) {
758 		DPRINTF(("%s: could not get packet tag\n", __func__));
759 		error = ENOMEM;
760 		goto bad;
761 	}
762 
763 	xh = (struct xform_history *)(mtag + 1);
764 	xh->dst = saidx->dst;
765 	xh->proto = saidx->proto;
766 	xh->mode = saidx->mode;
767 	xh->spi = sav->spi;
768 	m_tag_prepend(m, mtag);
769 
770 	key_sa_recordxfer(sav, m);		/* record data transfer */
771 
772 	/*
773 	 * If there's another (bundled) SA to apply, do so.
774 	 * Note that this puts a burden on the kernel stack size.
775 	 * If this is a problem we'll need to introduce a queue
776 	 * to set the packet on so we can unwind the stack before
777 	 * doing further processing.
778 	 */
779 	if (++idx < sp->tcount) {
780 		switch (saidx->dst.sa.sa_family) {
781 #ifdef INET
782 		case AF_INET:
783 			key_freesav(&sav);
784 			IPSECSTAT_INC(ips_out_bundlesa);
785 			return (ipsec4_perform_request(m, sp, NULL, idx));
786 			/* NOTREACHED */
787 #endif
788 #ifdef INET6
789 		case AF_INET6:
790 			key_freesav(&sav);
791 			IPSEC6STAT_INC(ips_out_bundlesa);
792 			return (ipsec6_perform_request(m, sp, NULL, idx));
793 			/* NOTREACHED */
794 #endif /* INET6 */
795 		default:
796 			DPRINTF(("%s: unknown protocol family %u\n", __func__,
797 			    saidx->dst.sa.sa_family));
798 			error = EPFNOSUPPORT;
799 			goto bad;
800 		}
801 	}
802 
803 	key_freesp(&sp), sp = NULL;	/* Release reference to SP */
804 #ifdef INET
805 	/*
806 	 * Do UDP encapsulation if SA requires it.
807 	 */
808 	if (sav->natt != NULL) {
809 		error = udp_ipsec_output(m, sav);
810 		if (error != 0)
811 			goto bad;
812 	}
813 #endif /* INET */
814 	/*
815 	 * We're done with IPsec processing, transmit the packet using the
816 	 * appropriate network protocol (IP or IPv6).
817 	 */
818 	NET_EPOCH_ENTER(et);
819 	switch (saidx->dst.sa.sa_family) {
820 #ifdef INET
821 	case AF_INET:
822 		key_freesav(&sav);
823 		error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
824 		break;
825 #endif /* INET */
826 #ifdef INET6
827 	case AF_INET6:
828 		key_freesav(&sav);
829 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
830 		break;
831 #endif /* INET6 */
832 	default:
833 		panic("ipsec_process_done");
834 	}
835 	NET_EPOCH_EXIT(et);
836 	return (error);
837 bad:
838 	m_freem(m);
839 	key_freesav(&sav);
840 	if (sp != NULL)
841 		key_freesp(&sp);
842 	return (error);
843 }
844 
845 /*
846  * ipsec_prepend() is optimized version of M_PREPEND().
847  * ipsec_encap() is called by IPsec output routine for tunnel mode SA.
848  * It is expected that after IP encapsulation some IPsec transform will
849  * be performed. Each IPsec transform inserts its variable length header
850  * just after outer IP header using m_makespace(). If given mbuf has not
851  * enough free space at the beginning, we allocate new mbuf and reserve
852  * some space at the beginning and at the end.
853  * This helps avoid allocating of new mbuf and data copying in m_makespace(),
854  * we place outer header in the middle of mbuf's data with reserved leading
855  * and trailing space:
856  *	[ LEADINGSPACE ][ Outer IP header ][ TRAILINGSPACE ]
857  * LEADINGSPACE will be used to add ethernet header, TRAILINGSPACE will
858  * be used to inject AH/ESP/IPCOMP header.
859  */
860 #define	IPSEC_TRAILINGSPACE	(sizeof(struct udphdr) +/* NAT-T */	\
861     max(sizeof(struct newesp) + EALG_MAX_BLOCK_LEN,	/* ESP + IV */	\
862 	sizeof(struct newah) + HASH_MAX_LEN		/* AH + ICV */))
863 static struct mbuf *
864 ipsec_prepend(struct mbuf *m, int len, int how)
865 {
866 	struct mbuf *n;
867 
868 	M_ASSERTPKTHDR(m);
869 	IPSEC_ASSERT(len < MHLEN, ("wrong length"));
870 	if (M_LEADINGSPACE(m) >= len) {
871 		/* No need to allocate new mbuf. */
872 		m->m_data -= len;
873 		m->m_len += len;
874 		m->m_pkthdr.len += len;
875 		return (m);
876 	}
877 	n = m_gethdr(how, m->m_type);
878 	if (n == NULL) {
879 		m_freem(m);
880 		return (NULL);
881 	}
882 	m_move_pkthdr(n, m);
883 	n->m_next = m;
884 	if (len + IPSEC_TRAILINGSPACE < M_SIZE(n))
885 		m_align(n, len + IPSEC_TRAILINGSPACE);
886 	n->m_len = len;
887 	n->m_pkthdr.len += len;
888 	return (n);
889 }
890 
891 static int
892 ipsec_encap(struct mbuf **mp, struct secasindex *saidx)
893 {
894 #ifdef INET6
895 	struct ip6_hdr *ip6;
896 #endif
897 	struct ip *ip;
898 	int setdf;
899 	uint8_t itos, proto;
900 
901 	ip = mtod(*mp, struct ip *);
902 	switch (ip->ip_v) {
903 #ifdef INET
904 	case IPVERSION:
905 		proto = IPPROTO_IPIP;
906 		/*
907 		 * Collect IP_DF state from the inner header
908 		 * and honor system-wide control of how to handle it.
909 		 */
910 		switch (V_ip4_ipsec_dfbit) {
911 		case 0:	/* clear in outer header */
912 		case 1:	/* set in outer header */
913 			setdf = V_ip4_ipsec_dfbit;
914 			break;
915 		default:/* propagate to outer header */
916 			setdf = (ip->ip_off & htons(IP_DF)) != 0;
917 		}
918 		itos = ip->ip_tos;
919 		break;
920 #endif
921 #ifdef INET6
922 	case (IPV6_VERSION >> 4):
923 		proto = IPPROTO_IPV6;
924 		ip6 = mtod(*mp, struct ip6_hdr *);
925 		itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
926 		setdf = V_ip4_ipsec_dfbit ? 1: 0;
927 		/* scoped address handling */
928 		in6_clearscope(&ip6->ip6_src);
929 		in6_clearscope(&ip6->ip6_dst);
930 		break;
931 #endif
932 	default:
933 		return (EAFNOSUPPORT);
934 	}
935 	switch (saidx->dst.sa.sa_family) {
936 #ifdef INET
937 	case AF_INET:
938 		if (saidx->src.sa.sa_family != AF_INET ||
939 		    saidx->src.sin.sin_addr.s_addr == INADDR_ANY ||
940 		    saidx->dst.sin.sin_addr.s_addr == INADDR_ANY)
941 			return (EINVAL);
942 		*mp = ipsec_prepend(*mp, sizeof(struct ip), M_NOWAIT);
943 		if (*mp == NULL)
944 			return (ENOBUFS);
945 		ip = mtod(*mp, struct ip *);
946 		ip->ip_v = IPVERSION;
947 		ip->ip_hl = sizeof(struct ip) >> 2;
948 		ip->ip_p = proto;
949 		ip->ip_len = htons((*mp)->m_pkthdr.len);
950 		ip->ip_ttl = V_ip_defttl;
951 		ip->ip_sum = 0;
952 		ip->ip_off = setdf ? htons(IP_DF): 0;
953 		ip->ip_src = saidx->src.sin.sin_addr;
954 		ip->ip_dst = saidx->dst.sin.sin_addr;
955 		ip_ecn_ingress(V_ip4_ipsec_ecn, &ip->ip_tos, &itos);
956 		ip_fillid(ip);
957 		break;
958 #endif /* INET */
959 #ifdef INET6
960 	case AF_INET6:
961 		if (saidx->src.sa.sa_family != AF_INET6 ||
962 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->src.sin6.sin6_addr) ||
963 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->dst.sin6.sin6_addr))
964 			return (EINVAL);
965 		*mp = ipsec_prepend(*mp, sizeof(struct ip6_hdr), M_NOWAIT);
966 		if (*mp == NULL)
967 			return (ENOBUFS);
968 		ip6 = mtod(*mp, struct ip6_hdr *);
969 		ip6->ip6_flow = 0;
970 		ip6->ip6_vfc = IPV6_VERSION;
971 		ip6->ip6_hlim = V_ip6_defhlim;
972 		ip6->ip6_nxt = proto;
973 		ip6->ip6_dst = saidx->dst.sin6.sin6_addr;
974 		/* For link-local address embed scope zone id */
975 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst))
976 			ip6->ip6_dst.s6_addr16[1] =
977 			    htons(saidx->dst.sin6.sin6_scope_id & 0xffff);
978 		ip6->ip6_src = saidx->src.sin6.sin6_addr;
979 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src))
980 			ip6->ip6_src.s6_addr16[1] =
981 			    htons(saidx->src.sin6.sin6_scope_id & 0xffff);
982 		ip6->ip6_plen = htons((*mp)->m_pkthdr.len - sizeof(*ip6));
983 		ip_ecn_ingress(V_ip6_ipsec_ecn, &proto, &itos);
984 		ip6->ip6_flow |= htonl((uint32_t)proto << 20);
985 		break;
986 #endif /* INET6 */
987 	default:
988 		return (EAFNOSUPPORT);
989 	}
990 	(*mp)->m_flags &= ~(M_BCAST | M_MCAST);
991 	return (0);
992 }
993