xref: /freebsd/sys/netipsec/ipsec_output.c (revision ef36b3f75658d201edb495068db5e1be49593de5)
1 /*-
2  * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
3  * Copyright (c) 2016 Andrey V. Elsukov <ae@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD$
28  */
29 
30 /*
31  * IPsec output processing.
32  */
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_ipsec.h"
36 #include "opt_sctp.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/mbuf.h>
41 #include <sys/domain.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/errno.h>
45 #include <sys/hhook.h>
46 #include <sys/syslog.h>
47 
48 #include <net/if.h>
49 #include <net/if_enc.h>
50 #include <net/if_var.h>
51 #include <net/vnet.h>
52 
53 #include <netinet/in.h>
54 #include <netinet/in_systm.h>
55 #include <netinet/ip.h>
56 #include <netinet/ip_var.h>
57 #include <netinet/in_var.h>
58 #include <netinet/ip_ecn.h>
59 #ifdef INET6
60 #include <netinet6/ip6_ecn.h>
61 #endif
62 
63 #include <netinet/ip6.h>
64 #ifdef INET6
65 #include <netinet6/ip6_var.h>
66 #include <netinet6/scope6_var.h>
67 #endif
68 #include <netinet/in_pcb.h>
69 #ifdef INET6
70 #include <netinet/icmp6.h>
71 #endif
72 #ifdef SCTP
73 #include <netinet/sctp_crc32.h>
74 #endif
75 
76 #include <netinet/udp.h>
77 #include <netipsec/ah.h>
78 #include <netipsec/esp.h>
79 #include <netipsec/ipsec.h>
80 #ifdef INET6
81 #include <netipsec/ipsec6.h>
82 #endif
83 #include <netipsec/ah_var.h>
84 #include <netipsec/esp_var.h>
85 #include <netipsec/ipcomp_var.h>
86 
87 #include <netipsec/xform.h>
88 
89 #include <netipsec/key.h>
90 #include <netipsec/keydb.h>
91 #include <netipsec/key_debug.h>
92 
93 #include <machine/in_cksum.h>
94 
95 #define	IPSEC_OSTAT_INC(proto, name)	do {		\
96 	if ((proto) == IPPROTO_ESP)	\
97 		ESPSTAT_INC(esps_##name);	\
98 	else if ((proto) == IPPROTO_AH)\
99 		AHSTAT_INC(ahs_##name);		\
100 	else					\
101 		IPCOMPSTAT_INC(ipcomps_##name);	\
102 } while (0)
103 
104 static int ipsec_encap(struct mbuf **mp, struct secasindex *saidx);
105 
106 #ifdef INET
107 static struct secasvar *
108 ipsec4_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
109 {
110 	struct secasindex *saidx, tmpsaidx;
111 	struct ipsecrequest *isr;
112 	struct sockaddr_in *sin;
113 	struct secasvar *sav;
114 	struct ip *ip;
115 
116 	/*
117 	 * Check system global policy controls.
118 	 */
119 next:
120 	isr = sp->req[*pidx];
121 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
122 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
123 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
124 		DPRINTF(("%s: IPsec outbound packet dropped due"
125 			" to policy (check your sysctls)\n", __func__));
126 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
127 		*error = EHOSTUNREACH;
128 		return (NULL);
129 	}
130 	/*
131 	 * Craft SA index to search for proper SA.  Note that
132 	 * we only initialize unspecified SA peers for transport
133 	 * mode; for tunnel mode they must already be filled in.
134 	 */
135 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
136 		saidx = &tmpsaidx;
137 		*saidx = isr->saidx;
138 		ip = mtod(m, struct ip *);
139 		if (saidx->src.sa.sa_len == 0) {
140 			sin = &saidx->src.sin;
141 			sin->sin_len = sizeof(*sin);
142 			sin->sin_family = AF_INET;
143 			sin->sin_port = IPSEC_PORT_ANY;
144 			sin->sin_addr = ip->ip_src;
145 		}
146 		if (saidx->dst.sa.sa_len == 0) {
147 			sin = &saidx->dst.sin;
148 			sin->sin_len = sizeof(*sin);
149 			sin->sin_family = AF_INET;
150 			sin->sin_port = IPSEC_PORT_ANY;
151 			sin->sin_addr = ip->ip_dst;
152 		}
153 	} else
154 		saidx = &sp->req[*pidx]->saidx;
155 	/*
156 	 * Lookup SA and validate it.
157 	 */
158 	sav = key_allocsa_policy(sp, saidx, error);
159 	if (sav == NULL) {
160 		IPSECSTAT_INC(ips_out_nosa);
161 		if (*error != 0)
162 			return (NULL);
163 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
164 			/*
165 			 * We have no SA and policy that doesn't require
166 			 * this IPsec transform, thus we can continue w/o
167 			 * IPsec processing, i.e. return EJUSTRETURN.
168 			 * But first check if there is some bundled transform.
169 			 */
170 			if (sp->tcount > ++(*pidx))
171 				goto next;
172 			*error = EJUSTRETURN;
173 		}
174 		return (NULL);
175 	}
176 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
177 	return (sav);
178 }
179 
180 /*
181  * IPsec output logic for IPv4.
182  */
183 static int
184 ipsec4_perform_request(struct mbuf *m, struct secpolicy *sp,
185     struct inpcb *inp, u_int idx)
186 {
187 	struct ipsec_ctx_data ctx;
188 	union sockaddr_union *dst;
189 	struct secasvar *sav;
190 	struct ip *ip;
191 	int error, i, off;
192 
193 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
194 
195 	/*
196 	 * We hold the reference to SP. Content of SP couldn't be changed.
197 	 * Craft secasindex and do lookup for suitable SA.
198 	 * Then do encapsulation if needed and call xform's output.
199 	 * We need to store SP in the xform callback parameters.
200 	 * In xform callback we will extract SP and it can be used to
201 	 * determine next transform. At the end of transform we can
202 	 * release reference to SP.
203 	 */
204 	sav = ipsec4_allocsa(m, sp, &idx, &error);
205 	if (sav == NULL) {
206 		if (error == EJUSTRETURN) { /* No IPsec required */
207 			key_freesp(&sp);
208 			return (error);
209 		}
210 		goto bad;
211 	}
212 	/*
213 	 * XXXAE: most likely ip_sum at this point is wrong.
214 	 */
215 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET, IPSEC_ENC_BEFORE);
216 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
217 		goto bad;
218 
219 	ip = mtod(m, struct ip *);
220 	dst = &sav->sah->saidx.dst;
221 	/* Do the appropriate encapsulation, if necessary */
222 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
223 	    dst->sa.sa_family != AF_INET ||	    /* PF mismatch */
224 	    (dst->sa.sa_family == AF_INET &&	    /* Proxy */
225 	     dst->sin.sin_addr.s_addr != INADDR_ANY &&
226 	     dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
227 		/* Fix IPv4 header checksum and length */
228 		ip->ip_len = htons(m->m_pkthdr.len);
229 		ip->ip_sum = 0;
230 		ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
231 		error = ipsec_encap(&m, &sav->sah->saidx);
232 		if (error != 0) {
233 			DPRINTF(("%s: encapsulation for SPI 0x%08x failed "
234 			    "with error %d\n", __func__, ntohl(sav->spi),
235 			    error));
236 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
237 			goto bad;
238 		}
239 		inp = NULL;
240 	}
241 
242 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, dst->sa.sa_family, IPSEC_ENC_AFTER);
243 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
244 		goto bad;
245 
246 	/*
247 	 * Dispatch to the appropriate IPsec transform logic.  The
248 	 * packet will be returned for transmission after crypto
249 	 * processing, etc. are completed.
250 	 *
251 	 * NB: m & sav are ``passed to caller'' who's responsible for
252 	 *     reclaiming their resources.
253 	 */
254 	switch(dst->sa.sa_family) {
255 	case AF_INET:
256 		ip = mtod(m, struct ip *);
257 		i = ip->ip_hl << 2;
258 		off = offsetof(struct ip, ip_p);
259 		break;
260 #ifdef INET6
261 	case AF_INET6:
262 		i = sizeof(struct ip6_hdr);
263 		off = offsetof(struct ip6_hdr, ip6_nxt);
264 		break;
265 #endif /* INET6 */
266 	default:
267 		DPRINTF(("%s: unsupported protocol family %u\n",
268 		    __func__, dst->sa.sa_family));
269 		error = EPFNOSUPPORT;
270 		IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf);
271 		goto bad;
272 	}
273 	error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off);
274 	return (error);
275 bad:
276 	IPSECSTAT_INC(ips_out_inval);
277 	if (m != NULL)
278 		m_freem(m);
279 	if (sav != NULL)
280 		key_freesav(&sav);
281 	key_freesp(&sp);
282 	return (error);
283 }
284 
285 int
286 ipsec4_process_packet(struct mbuf *m, struct secpolicy *sp,
287     struct inpcb *inp)
288 {
289 
290 	return (ipsec4_perform_request(m, sp, inp, 0));
291 }
292 
293 static int
294 ipsec4_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
295 {
296 	struct secpolicy *sp;
297 	int error;
298 
299 	/* Lookup for the corresponding outbound security policy */
300 	sp = ipsec4_checkpolicy(m, inp, &error, !forwarding);
301 	if (sp == NULL) {
302 		if (error == -EINVAL) {
303 			/* Discarded by policy. */
304 			m_freem(m);
305 			return (EACCES);
306 		}
307 		return (0); /* No IPsec required. */
308 	}
309 
310 	/*
311 	 * Usually we have to have tunnel mode IPsec security policy
312 	 * when we are forwarding a packet. Otherwise we could not handle
313 	 * encrypted replies, because they are not destined for us. But
314 	 * some users are doing source address translation for forwarded
315 	 * packets, and thus, even if they are forwarded, the replies will
316 	 * return back to us.
317 	 */
318 	if (!forwarding) {
319 		/*
320 		 * Do delayed checksums now because we send before
321 		 * this is done in the normal processing path.
322 		 */
323 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
324 			in_delayed_cksum(m);
325 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
326 		}
327 #ifdef SCTP
328 		if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
329 			struct ip *ip = mtod(m, struct ip *);
330 
331 			sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
332 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
333 		}
334 #endif
335 	}
336 	/* NB: callee frees mbuf and releases reference to SP */
337 	error = ipsec4_process_packet(m, sp, inp);
338 	if (error == EJUSTRETURN) {
339 		/*
340 		 * We had a SP with a level of 'use' and no SA. We
341 		 * will just continue to process the packet without
342 		 * IPsec processing and return without error.
343 		 */
344 		return (0);
345 	}
346 	if (error == 0)
347 		return (EINPROGRESS); /* consumed by IPsec */
348 	return (error);
349 }
350 
351 /*
352  * IPSEC_OUTPUT() method implementation for IPv4.
353  * 0 - no IPsec handling needed
354  * other values - mbuf consumed by IPsec.
355  */
356 int
357 ipsec4_output(struct mbuf *m, struct inpcb *inp)
358 {
359 
360 	/*
361 	 * If the packet is resubmitted to ip_output (e.g. after
362 	 * AH, ESP, etc. processing), there will be a tag to bypass
363 	 * the lookup and related policy checking.
364 	 */
365 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
366 		return (0);
367 
368 	return (ipsec4_common_output(m, inp, 0));
369 }
370 
371 /*
372  * IPSEC_FORWARD() method implementation for IPv4.
373  * 0 - no IPsec handling needed
374  * other values - mbuf consumed by IPsec.
375  */
376 int
377 ipsec4_forward(struct mbuf *m)
378 {
379 
380 	/*
381 	 * Check if this packet has an active inbound SP and needs to be
382 	 * dropped instead of forwarded.
383 	 */
384 	if (ipsec4_in_reject(m, NULL) != 0) {
385 		m_freem(m);
386 		return (EACCES);
387 	}
388 	return (ipsec4_common_output(m, NULL, 1));
389 }
390 #endif
391 
392 #ifdef INET6
393 static int
394 in6_sa_equal_addrwithscope(const struct sockaddr_in6 *sa,
395     const struct in6_addr *ia)
396 {
397 	struct in6_addr ia2;
398 
399 	if (IN6_IS_SCOPE_LINKLOCAL(&sa->sin6_addr)) {
400 		memcpy(&ia2, &sa->sin6_addr, sizeof(ia2));
401 		ia2.s6_addr16[1] = htons(sa->sin6_scope_id);
402 		return (IN6_ARE_ADDR_EQUAL(ia, &ia2));
403 	}
404 	return (IN6_ARE_ADDR_EQUAL(&sa->sin6_addr, ia));
405 }
406 
407 static struct secasvar *
408 ipsec6_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error)
409 {
410 	struct secasindex *saidx, tmpsaidx;
411 	struct ipsecrequest *isr;
412 	struct sockaddr_in6 *sin6;
413 	struct secasvar *sav;
414 	struct ip6_hdr *ip6;
415 
416 	/*
417 	 * Check system global policy controls.
418 	 */
419 next:
420 	isr = sp->req[*pidx];
421 	if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
422 	    (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
423 	    (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
424 		DPRINTF(("%s: IPsec outbound packet dropped due"
425 			" to policy (check your sysctls)\n", __func__));
426 		IPSEC_OSTAT_INC(isr->saidx.proto, pdrops);
427 		*error = EHOSTUNREACH;
428 		return (NULL);
429 	}
430 	/*
431 	 * Craft SA index to search for proper SA.  Note that
432 	 * we only fillin unspecified SA peers for transport
433 	 * mode; for tunnel mode they must already be filled in.
434 	 */
435 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
436 		saidx = &tmpsaidx;
437 		*saidx = isr->saidx;
438 		ip6 = mtod(m, struct ip6_hdr *);
439 		if (saidx->src.sin6.sin6_len == 0) {
440 			sin6 = (struct sockaddr_in6 *)&saidx->src;
441 			sin6->sin6_len = sizeof(*sin6);
442 			sin6->sin6_family = AF_INET6;
443 			sin6->sin6_port = IPSEC_PORT_ANY;
444 			sin6->sin6_addr = ip6->ip6_src;
445 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
446 				/* fix scope id for comparing SPD */
447 				sin6->sin6_addr.s6_addr16[1] = 0;
448 				sin6->sin6_scope_id =
449 				    ntohs(ip6->ip6_src.s6_addr16[1]);
450 			}
451 		}
452 		if (saidx->dst.sin6.sin6_len == 0) {
453 			sin6 = (struct sockaddr_in6 *)&saidx->dst;
454 			sin6->sin6_len = sizeof(*sin6);
455 			sin6->sin6_family = AF_INET6;
456 			sin6->sin6_port = IPSEC_PORT_ANY;
457 			sin6->sin6_addr = ip6->ip6_dst;
458 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
459 				/* fix scope id for comparing SPD */
460 				sin6->sin6_addr.s6_addr16[1] = 0;
461 				sin6->sin6_scope_id =
462 				    ntohs(ip6->ip6_dst.s6_addr16[1]);
463 			}
464 		}
465 	} else
466 		saidx = &sp->req[*pidx]->saidx;
467 	/*
468 	 * Lookup SA and validate it.
469 	 */
470 	sav = key_allocsa_policy(sp, saidx, error);
471 	if (sav == NULL) {
472 		IPSEC6STAT_INC(ips_out_nosa);
473 		if (*error != 0)
474 			return (NULL);
475 		if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) {
476 			/*
477 			 * We have no SA and policy that doesn't require
478 			 * this IPsec transform, thus we can continue w/o
479 			 * IPsec processing, i.e. return EJUSTRETURN.
480 			 * But first check if there is some bundled transform.
481 			 */
482 			if (sp->tcount > ++(*pidx))
483 				goto next;
484 			*error = EJUSTRETURN;
485 		}
486 		return (NULL);
487 	}
488 	IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform"));
489 	return (sav);
490 }
491 
492 /*
493  * IPsec output logic for IPv6.
494  */
495 static int
496 ipsec6_perform_request(struct mbuf *m, struct secpolicy *sp,
497     struct inpcb *inp, u_int idx)
498 {
499 	struct ipsec_ctx_data ctx;
500 	union sockaddr_union *dst;
501 	struct secasvar *sav;
502 	struct ip6_hdr *ip6;
503 	int error, i, off;
504 
505 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
506 
507 	sav = ipsec6_allocsa(m, sp, &idx, &error);
508 	if (sav == NULL) {
509 		if (error == EJUSTRETURN) { /* No IPsec required */
510 			key_freesp(&sp);
511 			return (error);
512 		}
513 		goto bad;
514 	}
515 
516 	/* Fix IP length in case if it is not set yet. */
517 	ip6 = mtod(m, struct ip6_hdr *);
518 	ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
519 
520 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET6, IPSEC_ENC_BEFORE);
521 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
522 		goto bad;
523 
524 	ip6 = mtod(m, struct ip6_hdr *); /* pfil can change mbuf */
525 	dst = &sav->sah->saidx.dst;
526 
527 	/* Do the appropriate encapsulation, if necessary */
528 	if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
529 	    dst->sa.sa_family != AF_INET6 ||        /* PF mismatch */
530 	    ((dst->sa.sa_family == AF_INET6) &&
531 	     (!IN6_IS_ADDR_UNSPECIFIED(&dst->sin6.sin6_addr)) &&
532 	     (!in6_sa_equal_addrwithscope(&dst->sin6, &ip6->ip6_dst)))) {
533 		if (m->m_pkthdr.len - sizeof(*ip6) > IPV6_MAXPACKET) {
534 			/* No jumbogram support. */
535 			error = ENXIO;   /*XXX*/
536 			goto bad;
537 		}
538 		error = ipsec_encap(&m, &sav->sah->saidx);
539 		if (error != 0) {
540 			DPRINTF(("%s: encapsulation for SPI 0x%08x failed "
541 			    "with error %d\n", __func__, ntohl(sav->spi),
542 			    error));
543 			/* XXXAE: IPSEC_OSTAT_INC(tunnel); */
544 			goto bad;
545 		}
546 		inp = NULL;
547 	}
548 
549 	IPSEC_INIT_CTX(&ctx, &m, inp, sav, dst->sa.sa_family, IPSEC_ENC_AFTER);
550 	if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0)
551 		goto bad;
552 
553 	switch(dst->sa.sa_family) {
554 #ifdef INET
555 	case AF_INET:
556 		{
557 		struct ip *ip;
558 		ip = mtod(m, struct ip *);
559 		i = ip->ip_hl << 2;
560 		off = offsetof(struct ip, ip_p);
561 		}
562 		break;
563 #endif /* AF_INET */
564 	case AF_INET6:
565 		i = sizeof(struct ip6_hdr);
566 		off = offsetof(struct ip6_hdr, ip6_nxt);
567 		break;
568 	default:
569 		DPRINTF(("%s: unsupported protocol family %u\n",
570 				 __func__, dst->sa.sa_family));
571 		error = EPFNOSUPPORT;
572 		IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf);
573 		goto bad;
574 	}
575 	error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off);
576 	return (error);
577 bad:
578 	IPSEC6STAT_INC(ips_out_inval);
579 	if (m != NULL)
580 		m_freem(m);
581 	if (sav != NULL)
582 		key_freesav(&sav);
583 	key_freesp(&sp);
584 	return (error);
585 }
586 
587 int
588 ipsec6_process_packet(struct mbuf *m, struct secpolicy *sp,
589     struct inpcb *inp)
590 {
591 
592 	return (ipsec6_perform_request(m, sp, inp, 0));
593 }
594 
595 static int
596 ipsec6_common_output(struct mbuf *m, struct inpcb *inp, int forwarding)
597 {
598 	struct secpolicy *sp;
599 	int error;
600 
601 	/* Lookup for the corresponding outbound security policy */
602 	sp = ipsec6_checkpolicy(m, inp, &error, !forwarding);
603 	if (sp == NULL) {
604 		if (error == -EINVAL) {
605 			/* Discarded by policy. */
606 			m_freem(m);
607 			return (EACCES);
608 		}
609 		return (0); /* No IPsec required. */
610 	}
611 
612 	if (!forwarding) {
613 		/*
614 		 * Do delayed checksums now because we send before
615 		 * this is done in the normal processing path.
616 		 */
617 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
618 			in6_delayed_cksum(m, m->m_pkthdr.len -
619 			    sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
620 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
621 		}
622 #ifdef SCTP
623 		if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
624 			sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
625 			m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
626 		}
627 #endif
628 	}
629 	/* NB: callee frees mbuf and releases reference to SP */
630 	error = ipsec6_process_packet(m, sp, inp);
631 	if (error == EJUSTRETURN) {
632 		/*
633 		 * We had a SP with a level of 'use' and no SA. We
634 		 * will just continue to process the packet without
635 		 * IPsec processing and return without error.
636 		 */
637 		return (0);
638 	}
639 	if (error == 0)
640 		return (EINPROGRESS); /* consumed by IPsec */
641 	return (error);
642 }
643 
644 /*
645  * IPSEC_OUTPUT() method implementation for IPv6.
646  * 0 - no IPsec handling needed
647  * other values - mbuf consumed by IPsec.
648  */
649 int
650 ipsec6_output(struct mbuf *m, struct inpcb *inp)
651 {
652 
653 	/*
654 	 * If the packet is resubmitted to ip_output (e.g. after
655 	 * AH, ESP, etc. processing), there will be a tag to bypass
656 	 * the lookup and related policy checking.
657 	 */
658 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL)
659 		return (0);
660 
661 	return (ipsec6_common_output(m, inp, 0));
662 }
663 
664 /*
665  * IPSEC_FORWARD() method implementation for IPv6.
666  * 0 - no IPsec handling needed
667  * other values - mbuf consumed by IPsec.
668  */
669 int
670 ipsec6_forward(struct mbuf *m)
671 {
672 
673 	/*
674 	 * Check if this packet has an active inbound SP and needs to be
675 	 * dropped instead of forwarded.
676 	 */
677 	if (ipsec6_in_reject(m, NULL) != 0) {
678 		m_freem(m);
679 		return (EACCES);
680 	}
681 	return (ipsec6_common_output(m, NULL, 1));
682 }
683 #endif /* INET6 */
684 
685 int
686 ipsec_process_done(struct mbuf *m, struct secpolicy *sp, struct secasvar *sav,
687     u_int idx)
688 {
689 	struct xform_history *xh;
690 	struct secasindex *saidx;
691 	struct m_tag *mtag;
692 	int error;
693 
694 	saidx = &sav->sah->saidx;
695 	switch (saidx->dst.sa.sa_family) {
696 #ifdef INET
697 	case AF_INET:
698 		/* Fix the header length, for AH processing. */
699 		mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
700 		break;
701 #endif /* INET */
702 #ifdef INET6
703 	case AF_INET6:
704 		/* Fix the header length, for AH processing. */
705 		if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
706 			error = ENXIO;
707 			goto bad;
708 		}
709 		if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
710 			/* No jumbogram support. */
711 			error = ENXIO;	/*?*/
712 			goto bad;
713 		}
714 		mtod(m, struct ip6_hdr *)->ip6_plen =
715 			htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
716 		break;
717 #endif /* INET6 */
718 	default:
719 		DPRINTF(("%s: unknown protocol family %u\n", __func__,
720 		    saidx->dst.sa.sa_family));
721 		error = ENXIO;
722 		goto bad;
723 	}
724 
725 	/*
726 	 * Add a record of what we've done to the packet.
727 	 */
728 	mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE, sizeof(*xh), M_NOWAIT);
729 	if (mtag == NULL) {
730 		DPRINTF(("%s: could not get packet tag\n", __func__));
731 		error = ENOMEM;
732 		goto bad;
733 	}
734 
735 	xh = (struct xform_history *)(mtag + 1);
736 	xh->dst = saidx->dst;
737 	xh->proto = saidx->proto;
738 	xh->mode = saidx->mode;
739 	xh->spi = sav->spi;
740 	m_tag_prepend(m, mtag);
741 
742 	key_sa_recordxfer(sav, m);		/* record data transfer */
743 
744 	/*
745 	 * If there's another (bundled) SA to apply, do so.
746 	 * Note that this puts a burden on the kernel stack size.
747 	 * If this is a problem we'll need to introduce a queue
748 	 * to set the packet on so we can unwind the stack before
749 	 * doing further processing.
750 	 */
751 	if (++idx < sp->tcount) {
752 		switch (saidx->dst.sa.sa_family) {
753 #ifdef INET
754 		case AF_INET:
755 			key_freesav(&sav);
756 			IPSECSTAT_INC(ips_out_bundlesa);
757 			return (ipsec4_perform_request(m, sp, NULL, idx));
758 			/* NOTREACHED */
759 #endif
760 #ifdef INET6
761 		case AF_INET6:
762 			key_freesav(&sav);
763 			IPSEC6STAT_INC(ips_out_bundlesa);
764 			return (ipsec6_perform_request(m, sp, NULL, idx));
765 			/* NOTREACHED */
766 #endif /* INET6 */
767 		default:
768 			DPRINTF(("%s: unknown protocol family %u\n", __func__,
769 			    saidx->dst.sa.sa_family));
770 			error = EPFNOSUPPORT;
771 			goto bad;
772 		}
773 	}
774 
775 	key_freesp(&sp), sp = NULL;	/* Release reference to SP */
776 #ifdef INET
777 	/*
778 	 * Do UDP encapsulation if SA requires it.
779 	 */
780 	if (sav->natt != NULL) {
781 		error = udp_ipsec_output(m, sav);
782 		if (error != 0)
783 			goto bad;
784 	}
785 #endif /* INET */
786 	/*
787 	 * We're done with IPsec processing, transmit the packet using the
788 	 * appropriate network protocol (IP or IPv6).
789 	 */
790 	switch (saidx->dst.sa.sa_family) {
791 #ifdef INET
792 	case AF_INET:
793 		key_freesav(&sav);
794 		return ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
795 #endif /* INET */
796 #ifdef INET6
797 	case AF_INET6:
798 		key_freesav(&sav);
799 		return ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
800 #endif /* INET6 */
801 	}
802 	panic("ipsec_process_done");
803 bad:
804 	m_freem(m);
805 	key_freesav(&sav);
806 	if (sp != NULL)
807 		key_freesp(&sp);
808 	return (error);
809 }
810 
811 /*
812  * ipsec_prepend() is optimized version of M_PREPEND().
813  * ipsec_encap() is called by IPsec output routine for tunnel mode SA.
814  * It is expected that after IP encapsulation some IPsec transform will
815  * be performed. Each IPsec transform inserts its variable length header
816  * just after outer IP header using m_makespace(). If given mbuf has not
817  * enough free space at the beginning, we allocate new mbuf and reserve
818  * some space at the beginning and at the end.
819  * This helps avoid allocating of new mbuf and data copying in m_makespace(),
820  * we place outer header in the middle of mbuf's data with reserved leading
821  * and trailing space:
822  *	[ LEADINGSPACE ][ Outer IP header ][ TRAILINGSPACE ]
823  * LEADINGSPACE will be used to add ethernet header, TRAILINGSPACE will
824  * be used to inject AH/ESP/IPCOMP header.
825  */
826 #define	IPSEC_TRAILINGSPACE	(sizeof(struct udphdr) +/* NAT-T */	\
827     max(sizeof(struct newesp) + EALG_MAX_BLOCK_LEN,	/* ESP + IV */	\
828 	sizeof(struct newah) + HASH_MAX_LEN		/* AH + ICV */))
829 static struct mbuf *
830 ipsec_prepend(struct mbuf *m, int len, int how)
831 {
832 	struct mbuf *n;
833 
834 	M_ASSERTPKTHDR(m);
835 	IPSEC_ASSERT(len < MHLEN, ("wrong length"));
836 	if (M_LEADINGSPACE(m) >= len) {
837 		/* No need to allocate new mbuf. */
838 		m->m_data -= len;
839 		m->m_len += len;
840 		m->m_pkthdr.len += len;
841 		return (m);
842 	}
843 	n = m_gethdr(how, m->m_type);
844 	if (n == NULL) {
845 		m_freem(m);
846 		return (NULL);
847 	}
848 	m_move_pkthdr(n, m);
849 	n->m_next = m;
850 	if (len + IPSEC_TRAILINGSPACE < M_SIZE(n))
851 		m_align(n, len + IPSEC_TRAILINGSPACE);
852 	n->m_len = len;
853 	n->m_pkthdr.len += len;
854 	return (n);
855 }
856 
857 static int
858 ipsec_encap(struct mbuf **mp, struct secasindex *saidx)
859 {
860 #ifdef INET6
861 	struct ip6_hdr *ip6;
862 #endif
863 	struct ip *ip;
864 	int setdf;
865 	uint8_t itos, proto;
866 
867 	ip = mtod(*mp, struct ip *);
868 	switch (ip->ip_v) {
869 #ifdef INET
870 	case IPVERSION:
871 		proto = IPPROTO_IPIP;
872 		/*
873 		 * Collect IP_DF state from the inner header
874 		 * and honor system-wide control of how to handle it.
875 		 */
876 		switch (V_ip4_ipsec_dfbit) {
877 		case 0:	/* clear in outer header */
878 		case 1:	/* set in outer header */
879 			setdf = V_ip4_ipsec_dfbit;
880 			break;
881 		default:/* propagate to outer header */
882 			setdf = (ip->ip_off & htons(IP_DF)) != 0;
883 		}
884 		itos = ip->ip_tos;
885 		break;
886 #endif
887 #ifdef INET6
888 	case (IPV6_VERSION >> 4):
889 		proto = IPPROTO_IPV6;
890 		ip6 = mtod(*mp, struct ip6_hdr *);
891 		itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
892 		setdf = V_ip4_ipsec_dfbit ? 1: 0;
893 		/* scoped address handling */
894 		in6_clearscope(&ip6->ip6_src);
895 		in6_clearscope(&ip6->ip6_dst);
896 		break;
897 #endif
898 	default:
899 		return (EAFNOSUPPORT);
900 	}
901 	switch (saidx->dst.sa.sa_family) {
902 #ifdef INET
903 	case AF_INET:
904 		if (saidx->src.sa.sa_family != AF_INET ||
905 		    saidx->src.sin.sin_addr.s_addr == INADDR_ANY ||
906 		    saidx->dst.sin.sin_addr.s_addr == INADDR_ANY)
907 			return (EINVAL);
908 		*mp = ipsec_prepend(*mp, sizeof(struct ip), M_NOWAIT);
909 		if (*mp == NULL)
910 			return (ENOBUFS);
911 		ip = mtod(*mp, struct ip *);
912 		ip->ip_v = IPVERSION;
913 		ip->ip_hl = sizeof(struct ip) >> 2;
914 		ip->ip_p = proto;
915 		ip->ip_len = htons((*mp)->m_pkthdr.len);
916 		ip->ip_ttl = V_ip_defttl;
917 		ip->ip_sum = 0;
918 		ip->ip_off = setdf ? htons(IP_DF): 0;
919 		ip->ip_src = saidx->src.sin.sin_addr;
920 		ip->ip_dst = saidx->dst.sin.sin_addr;
921 		ip_ecn_ingress(V_ip4_ipsec_ecn, &ip->ip_tos, &itos);
922 		ip_fillid(ip);
923 		break;
924 #endif /* INET */
925 #ifdef INET6
926 	case AF_INET6:
927 		if (saidx->src.sa.sa_family != AF_INET6 ||
928 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->src.sin6.sin6_addr) ||
929 		    IN6_IS_ADDR_UNSPECIFIED(&saidx->dst.sin6.sin6_addr))
930 			return (EINVAL);
931 		*mp = ipsec_prepend(*mp, sizeof(struct ip6_hdr), M_NOWAIT);
932 		if (*mp == NULL)
933 			return (ENOBUFS);
934 		ip6 = mtod(*mp, struct ip6_hdr *);
935 		ip6->ip6_flow = 0;
936 		ip6->ip6_vfc = IPV6_VERSION;
937 		ip6->ip6_hlim = V_ip6_defhlim;
938 		ip6->ip6_nxt = proto;
939 		ip6->ip6_dst = saidx->dst.sin6.sin6_addr;
940 		/* For link-local address embed scope zone id */
941 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst))
942 			ip6->ip6_dst.s6_addr16[1] =
943 			    htons(saidx->dst.sin6.sin6_scope_id & 0xffff);
944 		ip6->ip6_src = saidx->src.sin6.sin6_addr;
945 		if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src))
946 			ip6->ip6_src.s6_addr16[1] =
947 			    htons(saidx->src.sin6.sin6_scope_id & 0xffff);
948 		ip6->ip6_plen = htons((*mp)->m_pkthdr.len - sizeof(*ip6));
949 		ip_ecn_ingress(V_ip6_ipsec_ecn, &proto, &itos);
950 		ip6->ip6_flow |= htonl((uint32_t)proto << 20);
951 		break;
952 #endif /* INET6 */
953 	default:
954 		return (EAFNOSUPPORT);
955 	}
956 	(*mp)->m_flags &= ~(M_BCAST | M_MCAST);
957 	return (0);
958 }
959 
960