Home | History | Annotate | Line # | Download | only in netipsec
ipsec_output.c revision 1.77
      1 /*	$NetBSD: ipsec_output.c,v 1.77 2018/05/07 09:25:04 maxv Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  *
     16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     26  * SUCH DAMAGE.
     27  *
     28  * $FreeBSD: sys/netipsec/ipsec_output.c,v 1.3.2.2 2003/03/28 20:32:53 sam Exp $
     29  */
     30 
     31 #include <sys/cdefs.h>
     32 __KERNEL_RCSID(0, "$NetBSD: ipsec_output.c,v 1.77 2018/05/07 09:25:04 maxv Exp $");
     33 
     34 #if defined(_KERNEL_OPT)
     35 #include "opt_inet.h"
     36 #include "opt_net_mpsafe.h"
     37 #endif
     38 
     39 #include <sys/param.h>
     40 #include <sys/systm.h>
     41 #include <sys/mbuf.h>
     42 #include <sys/domain.h>
     43 #include <sys/protosw.h>
     44 #include <sys/socket.h>
     45 #include <sys/errno.h>
     46 #include <sys/syslog.h>
     47 
     48 #include <net/if.h>
     49 #include <net/route.h>
     50 
     51 #include <netinet/in.h>
     52 #include <netinet/in_systm.h>
     53 #include <netinet/ip.h>
     54 #include <netinet/ip_var.h>
     55 #include <netinet/in_var.h>
     56 #include <netinet/ip_ecn.h>
     57 
     58 #include <netinet/ip6.h>
     59 #ifdef INET6
     60 #include <netinet6/ip6_var.h>
     61 #endif
     62 #include <netinet/in_pcb.h>
     63 #ifdef INET6
     64 #include <netinet/icmp6.h>
     65 #endif
     66 #include <netinet/udp.h>
     67 
     68 #include <netipsec/ipsec.h>
     69 #include <netipsec/ipsec_var.h>
     70 #include <netipsec/ipsec_private.h>
     71 #ifdef INET6
     72 #include <netipsec/ipsec6.h>
     73 #endif
     74 #include <netipsec/ah_var.h>
     75 #include <netipsec/esp_var.h>
     76 #include <netipsec/ipcomp_var.h>
     77 
     78 #include <netipsec/xform.h>
     79 
     80 #include <netipsec/key.h>
     81 #include <netipsec/keydb.h>
     82 #include <netipsec/key_debug.h>
     83 
     84 static percpu_t *ipsec_rtcache_percpu __cacheline_aligned;
     85 
     86 /*
     87  * Add a IPSEC_OUT_DONE tag to mark that we have finished the ipsec processing
     88  * It will be used by ip{,6}_output to check if we have already or not
     89  * processed this packet.
     90  */
     91 static int
     92 ipsec_register_done(struct mbuf *m, int *error)
     93 {
     94 	struct m_tag *mtag;
     95 
     96 	mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE, 0, M_NOWAIT);
     97 	if (mtag == NULL) {
     98 		IPSECLOG(LOG_DEBUG, "could not get packet tag\n");
     99 		*error = ENOMEM;
    100 		return -1;
    101 	}
    102 
    103 	m_tag_prepend(m, mtag);
    104 	return 0;
    105 }
    106 
    107 static int
    108 ipsec_reinject_ipstack(struct mbuf *m, int af)
    109 {
    110 	int rv = -1;
    111 	struct route *ro;
    112 
    113 	KASSERT(af == AF_INET || af == AF_INET6);
    114 
    115 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
    116 	ro = percpu_getref(ipsec_rtcache_percpu);
    117 	switch (af) {
    118 #ifdef INET
    119 	case AF_INET:
    120 		rv = ip_output(m, NULL, ro, IP_RAWOUTPUT|IP_NOIPNEWID,
    121 		    NULL, NULL);
    122 		break;
    123 #endif
    124 #ifdef INET6
    125 	case AF_INET6:
    126 		/*
    127 		 * We don't need massage, IPv6 header fields are always in
    128 		 * net endian.
    129 		 */
    130 		rv = ip6_output(m, NULL, ro, 0, NULL, NULL, NULL);
    131 		break;
    132 #endif
    133 	}
    134 	percpu_putref(ipsec_rtcache_percpu);
    135 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    136 
    137 	return rv;
    138 }
    139 
    140 int
    141 ipsec_process_done(struct mbuf *m, const struct ipsecrequest *isr,
    142     struct secasvar *sav)
    143 {
    144 	struct secasindex *saidx;
    145 	int error;
    146 #ifdef INET
    147 	struct ip *ip;
    148 #endif
    149 #ifdef INET6
    150 	struct ip6_hdr *ip6;
    151 #endif
    152 	struct mbuf *mo;
    153 	struct udphdr *udp = NULL;
    154 	uint64_t *data = NULL;
    155 	int hlen, roff;
    156 
    157 	KASSERT(m != NULL);
    158 	KASSERT(isr != NULL);
    159 	KASSERT(sav != NULL);
    160 
    161 	saidx = &sav->sah->saidx;
    162 
    163 	if (sav->natt_type != 0) {
    164 		ip = mtod(m, struct ip *);
    165 
    166 		hlen = sizeof(struct udphdr);
    167 		if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
    168 			hlen += sizeof(uint64_t);
    169 
    170 		mo = m_makespace(m, sizeof(struct ip), hlen, &roff);
    171 		if (mo == NULL) {
    172 			char buf[IPSEC_ADDRSTRLEN];
    173 			IPSECLOG(LOG_DEBUG,
    174 			    "failed to inject %u byte UDP for SA %s/%08lx\n",
    175 			    hlen, ipsec_address(&saidx->dst, buf, sizeof(buf)),
    176 			    (u_long)ntohl(sav->spi));
    177 			error = ENOBUFS;
    178 			goto bad;
    179 		}
    180 
    181 		udp = (struct udphdr *)(mtod(mo, char *) + roff);
    182 		data = (uint64_t *)(udp + 1);
    183 
    184 		if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
    185 			*data = 0; /* NON-IKE Marker */
    186 
    187 		if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
    188 			udp->uh_sport = htons(UDP_ENCAP_ESPINUDP_PORT);
    189 		else
    190 			udp->uh_sport = key_portfromsaddr(&saidx->src);
    191 
    192 		udp->uh_dport = key_portfromsaddr(&saidx->dst);
    193 		udp->uh_sum = 0;
    194 		udp->uh_ulen = htons(m->m_pkthdr.len - (ip->ip_hl << 2));
    195 	}
    196 
    197 	/*
    198 	 * Fix the header length, for AH processing.
    199 	 */
    200 	switch (saidx->dst.sa.sa_family) {
    201 #ifdef INET
    202 	case AF_INET:
    203 		ip = mtod(m, struct ip *);
    204 		ip->ip_len = htons(m->m_pkthdr.len);
    205 		if (sav->natt_type != 0)
    206 			ip->ip_p = IPPROTO_UDP;
    207 		break;
    208 #endif
    209 #ifdef INET6
    210 	case AF_INET6:
    211 		if (m->m_pkthdr.len < sizeof(struct ip6_hdr)) {
    212 			error = ENXIO;
    213 			goto bad;
    214 		}
    215 		if (m->m_pkthdr.len - sizeof(struct ip6_hdr) > IPV6_MAXPACKET) {
    216 			/* No jumbogram support. */
    217 			error = ENXIO;	/*?*/
    218 			goto bad;
    219 		}
    220 		ip6 = mtod(m, struct ip6_hdr *);
    221 		ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
    222 		if (sav->natt_type != 0)
    223 			ip6->ip6_nxt = IPPROTO_UDP;
    224 		break;
    225 #endif
    226 	default:
    227 		IPSECLOG(LOG_DEBUG, "unknown protocol family %u\n",
    228 		    saidx->dst.sa.sa_family);
    229 		error = ENXIO;
    230 		goto bad;
    231 	}
    232 
    233 	key_sa_recordxfer(sav, m);
    234 
    235 	/*
    236 	 * If there's another (bundled) SA to apply, do so.
    237 	 * Note that this puts a burden on the kernel stack size.
    238 	 * If this is a problem we'll need to introduce a queue
    239 	 * to set the packet on so we can unwind the stack before
    240 	 * doing further processing.
    241 	 */
    242 	if (isr->next) {
    243 		IPSEC_STATINC(IPSEC_STAT_OUT_BUNDLESA);
    244 		switch (saidx->dst.sa.sa_family) {
    245 #ifdef INET
    246 		case AF_INET:
    247 			return ipsec4_process_packet(m, isr->next, NULL);
    248 #endif
    249 #ifdef INET6
    250 		case AF_INET6:
    251 			return ipsec6_process_packet(m, isr->next);
    252 #endif
    253 		default:
    254 			IPSECLOG(LOG_DEBUG, "unknown protocol family %u\n",
    255 			    saidx->dst.sa.sa_family);
    256 			error = ENXIO;
    257 			goto bad;
    258 		}
    259 	}
    260 
    261 	/*
    262 	 * We're done with IPsec processing, mark the packet as processed,
    263 	 * and transmit it using the appropriate network protocol
    264 	 * (IPv4/IPv6).
    265 	 */
    266 
    267 	if (ipsec_register_done(m, &error) < 0)
    268 		goto bad;
    269 
    270 	return ipsec_reinject_ipstack(m, saidx->dst.sa.sa_family);
    271 
    272 bad:
    273 	m_freem(m);
    274 	return error;
    275 }
    276 
    277 static void
    278 ipsec_fill_saidx_bymbuf(struct secasindex *saidx, const struct mbuf *m,
    279     const int af)
    280 {
    281 
    282 	if (af == AF_INET) {
    283 		struct sockaddr_in *sin;
    284 		struct ip *ip = mtod(m, struct ip *);
    285 
    286 		if (saidx->src.sa.sa_len == 0) {
    287 			sin = &saidx->src.sin;
    288 			sin->sin_len = sizeof(*sin);
    289 			sin->sin_family = AF_INET;
    290 			sin->sin_port = IPSEC_PORT_ANY;
    291 			sin->sin_addr = ip->ip_src;
    292 		}
    293 		if (saidx->dst.sa.sa_len == 0) {
    294 			sin = &saidx->dst.sin;
    295 			sin->sin_len = sizeof(*sin);
    296 			sin->sin_family = AF_INET;
    297 			sin->sin_port = IPSEC_PORT_ANY;
    298 			sin->sin_addr = ip->ip_dst;
    299 		}
    300 	} else {
    301 		struct sockaddr_in6 *sin6;
    302 		struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
    303 
    304 		if (saidx->src.sin6.sin6_len == 0) {
    305 			sin6 = (struct sockaddr_in6 *)&saidx->src;
    306 			sin6->sin6_len = sizeof(*sin6);
    307 			sin6->sin6_family = AF_INET6;
    308 			sin6->sin6_port = IPSEC_PORT_ANY;
    309 			sin6->sin6_addr = ip6->ip6_src;
    310 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
    311 				/* fix scope id for comparing SPD */
    312 				sin6->sin6_addr.s6_addr16[1] = 0;
    313 				sin6->sin6_scope_id =
    314 				    ntohs(ip6->ip6_src.s6_addr16[1]);
    315 			}
    316 		}
    317 		if (saidx->dst.sin6.sin6_len == 0) {
    318 			sin6 = (struct sockaddr_in6 *)&saidx->dst;
    319 			sin6->sin6_len = sizeof(*sin6);
    320 			sin6->sin6_family = AF_INET6;
    321 			sin6->sin6_port = IPSEC_PORT_ANY;
    322 			sin6->sin6_addr = ip6->ip6_dst;
    323 			if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
    324 				/* fix scope id for comparing SPD */
    325 				sin6->sin6_addr.s6_addr16[1] = 0;
    326 				sin6->sin6_scope_id =
    327 				    ntohs(ip6->ip6_dst.s6_addr16[1]);
    328 			}
    329 		}
    330 	}
    331 }
    332 
    333 struct secasvar *
    334 ipsec_lookup_sa(const struct ipsecrequest *isr, const struct mbuf *m)
    335 {
    336 	struct secasindex saidx;
    337 
    338 	saidx = isr->saidx;
    339 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
    340 		/* Fillin unspecified SA peers only for transport mode */
    341 		ipsec_fill_saidx_bymbuf(&saidx, m, isr->saidx.dst.sa.sa_family);
    342 	}
    343 
    344 	return key_lookup_sa_bysaidx(&saidx);
    345 }
    346 
    347 /*
    348  * ipsec_nextisr can return :
    349  * - isr == NULL and error != 0 => something is bad : the packet must be
    350  *   discarded
    351  * - isr == NULL and error == 0 => no more rules to apply, ipsec processing
    352  *   is done, reinject it in ip stack
    353  * - isr != NULL (error == 0) => we need to apply one rule to the packet
    354  */
    355 static const struct ipsecrequest *
    356 ipsec_nextisr(struct mbuf *m, const struct ipsecrequest *isr, int af,
    357     int *error, struct secasvar **ret)
    358 {
    359 #define	IPSEC_OSTAT(type)						\
    360 do {									\
    361 	switch (isr->saidx.proto) {					\
    362 	case IPPROTO_ESP:						\
    363 		ESP_STATINC(ESP_STAT_ ## type);				\
    364 		break;							\
    365 	case IPPROTO_AH:						\
    366 		AH_STATINC(AH_STAT_ ## type);				\
    367 		break;							\
    368 	default:							\
    369 		IPCOMP_STATINC(IPCOMP_STAT_ ## type);			\
    370 		break;							\
    371 	}								\
    372 } while (/*CONSTCOND*/0)
    373 
    374 	struct secasvar *sav = NULL;
    375 	struct secasindex saidx;
    376 
    377 	KASSERTMSG(af == AF_INET || af == AF_INET6,
    378 	    "invalid address family %u", af);
    379 again:
    380 	/*
    381 	 * Craft SA index to search for proper SA.  Note that
    382 	 * we only fillin unspecified SA peers for transport
    383 	 * mode; for tunnel mode they must already be filled in.
    384 	 */
    385 	saidx = isr->saidx;
    386 	if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
    387 		/* Fillin unspecified SA peers only for transport mode */
    388 		ipsec_fill_saidx_bymbuf(&saidx, m, af);
    389 	}
    390 
    391 	/*
    392 	 * Lookup SA and validate it.
    393 	 */
    394 	*error = key_checkrequest(isr, &saidx, &sav);
    395 	if (*error != 0) {
    396 		/*
    397 		 * IPsec processing is required, but no SA found.
    398 		 * I assume that key_acquire() had been called
    399 		 * to get/establish the SA. Here I discard
    400 		 * this packet because it is responsibility for
    401 		 * upper layer to retransmit the packet.
    402 		 */
    403 		IPSEC_STATINC(IPSEC_STAT_OUT_NOSA);
    404 		goto bad;
    405 	}
    406 	/* sav may be NULL here if we have an USE rule */
    407 	if (sav == NULL) {
    408 		KASSERTMSG(ipsec_get_reqlevel(isr) == IPSEC_LEVEL_USE,
    409 		    "no SA found, but required; level %u",
    410 		    ipsec_get_reqlevel(isr));
    411 		isr = isr->next;
    412 		/*
    413 		 * No more rules to apply, return NULL isr and no error.
    414 		 * It can happen when the last rules are USE rules.
    415 		 */
    416 		if (isr == NULL) {
    417 			*ret = NULL;
    418 			*error = 0;
    419 			return isr;
    420 		}
    421 		goto again;
    422 	}
    423 
    424 	/*
    425 	 * Check system global policy controls.
    426 	 */
    427 	if ((isr->saidx.proto == IPPROTO_ESP && !esp_enable) ||
    428 	    (isr->saidx.proto == IPPROTO_AH && !ah_enable) ||
    429 	    (isr->saidx.proto == IPPROTO_IPCOMP && !ipcomp_enable)) {
    430 		IPSECLOG(LOG_DEBUG, "IPsec outbound packet dropped due"
    431 		    " to policy (check your sysctls)\n");
    432 		IPSEC_OSTAT(PDROPS);
    433 		*error = EHOSTUNREACH;
    434 		KEY_SA_UNREF(&sav);
    435 		goto bad;
    436 	}
    437 
    438 	/*
    439 	 * Sanity check the SA contents for the caller
    440 	 * before they invoke the xform output method.
    441 	 */
    442 	KASSERT(sav->tdb_xform != NULL);
    443 	*ret = sav;
    444 	return isr;
    445 
    446 bad:
    447 	KASSERTMSG(*error != 0, "error return w/ no error code");
    448 	return NULL;
    449 #undef IPSEC_OSTAT
    450 }
    451 
    452 #ifdef INET
    453 /*
    454  * IPsec output logic for IPv4.
    455  */
    456 int
    457 ipsec4_process_packet(struct mbuf *m, const struct ipsecrequest *isr,
    458     u_long *mtu)
    459 {
    460 	struct secasvar *sav = NULL;
    461 	struct ip *ip;
    462 	int s, error, i, off;
    463 	union sockaddr_union *dst;
    464 	int setdf;
    465 
    466 	KASSERT(m != NULL);
    467 	KASSERT(m->m_nextpkt == NULL);
    468 	KASSERT(isr != NULL);
    469 
    470 	s = splsoftnet();	/* insure SA contents don't change */
    471 
    472 	isr = ipsec_nextisr(m, isr, AF_INET, &error, &sav);
    473 	if (isr == NULL) {
    474 		if (error != 0) {
    475 			goto bad;
    476 		} else {
    477 			if (ipsec_register_done(m, &error) < 0)
    478 				goto bad;
    479 
    480 			splx(s);
    481 			return ipsec_reinject_ipstack(m, AF_INET);
    482 		}
    483 	}
    484 	KASSERT(sav != NULL);
    485 
    486 	if (m->m_len < sizeof(struct ip) &&
    487 	    (m = m_pullup(m, sizeof(struct ip))) == NULL) {
    488 		error = ENOBUFS;
    489 		goto unrefsav;
    490 	}
    491 
    492 	/*
    493 	 * Check if we need to handle NAT-T fragmentation.
    494 	 */
    495 	if (isr == isr->sp->req) { /* Check only if called from ipsec4_output */
    496 		KASSERT(mtu != NULL);
    497 		ip = mtod(m, struct ip *);
    498 		if (!(sav->natt_type &
    499 		    (UDP_ENCAP_ESPINUDP|UDP_ENCAP_ESPINUDP_NON_IKE))) {
    500 			goto noneed;
    501 		}
    502 		if (ntohs(ip->ip_len) <= sav->esp_frag)
    503 			goto noneed;
    504 		*mtu = sav->esp_frag;
    505 		KEY_SA_UNREF(&sav);
    506 		splx(s);
    507 		return 0;
    508 	}
    509 noneed:
    510 	dst = &sav->sah->saidx.dst;
    511 
    512 	/*
    513 	 * Collect IP_DF state from the outer header.
    514 	 */
    515 	if (dst->sa.sa_family == AF_INET) {
    516 		ip = mtod(m, struct ip *);
    517 		/* Honor system-wide control of how to handle IP_DF */
    518 		switch (ip4_ipsec_dfbit) {
    519 		case 0:			/* clear in outer header */
    520 		case 1:			/* set in outer header */
    521 			setdf = ip4_ipsec_dfbit;
    522 			break;
    523 		default:		/* propagate to outer header */
    524 			setdf = ip->ip_off;
    525 			setdf = ntohs(setdf);
    526 			setdf = htons(setdf & IP_DF);
    527 			break;
    528 		}
    529 	} else {
    530 		ip = NULL;		/* keep compiler happy */
    531 		setdf = 0;
    532 	}
    533 
    534 	/* Do the appropriate encapsulation, if necessary */
    535 	if (isr->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
    536 	    dst->sa.sa_family != AF_INET ||	    /* PF mismatch */
    537 #if 0
    538 	    (sav->flags & SADB_X_SAFLAGS_TUNNEL) || /* Tunnel requ'd */
    539 	    sav->tdb_xform->xf_type == XF_IP4 ||    /* ditto */
    540 #endif
    541 	    (dst->sa.sa_family == AF_INET &&	    /* Proxy */
    542 	     dst->sin.sin_addr.s_addr != INADDR_ANY &&
    543 	     dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
    544 		struct mbuf *mp;
    545 
    546 		/* Fix IPv4 header checksum and length */
    547 		ip = mtod(m, struct ip *);
    548 		ip->ip_len = htons(m->m_pkthdr.len);
    549 		ip->ip_sum = 0;
    550 		ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
    551 
    552 		/* Encapsulate the packet */
    553 		error = ipip_output(m, sav, &mp);
    554 		if (mp == NULL && !error) {
    555 			/* Should never happen. */
    556 			IPSECLOG(LOG_DEBUG,
    557 			    "ipip_output returns no mbuf and no error!");
    558 			error = EFAULT;
    559 		}
    560 		if (error) {
    561 			if (mp) {
    562 				/* XXX: Should never happen! */
    563 				m_freem(mp);
    564 			}
    565 			m = NULL; /* ipip_output() already freed it */
    566 			goto unrefsav;
    567 		}
    568 		m = mp, mp = NULL;
    569 
    570 		/*
    571 		 * ipip_output clears IP_DF in the new header.  If
    572 		 * we need to propagate IP_DF from the outer header,
    573 		 * then we have to do it here.
    574 		 *
    575 		 * XXX shouldn't assume what ipip_output does.
    576 		 */
    577 		if (dst->sa.sa_family == AF_INET && setdf) {
    578 			if (m->m_len < sizeof(struct ip) &&
    579 			    (m = m_pullup(m, sizeof(struct ip))) == NULL) {
    580 				error = ENOBUFS;
    581 				goto unrefsav;
    582 			}
    583 			ip = mtod(m, struct ip *);
    584 			ip->ip_off |= htons(IP_DF);
    585 		}
    586 	}
    587 
    588 	/*
    589 	 * Dispatch to the appropriate IPsec transform logic.  The
    590 	 * packet will be returned for transmission after crypto
    591 	 * processing, etc. are completed.  For encapsulation we
    592 	 * bypass this call because of the explicit call done above
    593 	 * (necessary to deal with IP_DF handling for IPv4).
    594 	 *
    595 	 * NB: m & sav are ``passed to caller'' who's reponsible for
    596 	 *     for reclaiming their resources.
    597 	 */
    598 	if (sav->tdb_xform->xf_type != XF_IP4) {
    599 		if (dst->sa.sa_family == AF_INET) {
    600 			ip = mtod(m, struct ip *);
    601 			i = ip->ip_hl << 2;
    602 			off = offsetof(struct ip, ip_p);
    603 		} else {
    604 			i = sizeof(struct ip6_hdr);
    605 			off = offsetof(struct ip6_hdr, ip6_nxt);
    606 		}
    607 		error = (*sav->tdb_xform->xf_output)(m, isr, sav, i, off);
    608 	} else {
    609 		error = ipsec_process_done(m, isr, sav);
    610 	}
    611 	KEY_SA_UNREF(&sav);
    612 	splx(s);
    613 	return error;
    614 
    615 unrefsav:
    616 	KEY_SA_UNREF(&sav);
    617 bad:
    618 	splx(s);
    619 	if (m)
    620 		m_freem(m);
    621 	return error;
    622 }
    623 #endif
    624 
    625 #ifdef INET6
    626 static int
    627 compute_ipsec_pos(struct mbuf *m, int *i, int *off)
    628 {
    629 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
    630 	struct ip6_ext ip6e;
    631 	int dstopt = 0;
    632 	int nxt;
    633 
    634 	*i = sizeof(struct ip6_hdr);
    635 	*off = offsetof(struct ip6_hdr, ip6_nxt);
    636 	nxt = ip6->ip6_nxt;
    637 
    638 	/*
    639 	 * chase mbuf chain to find the appropriate place to
    640 	 * put AH/ESP/IPcomp header.
    641 	 *     IPv6 hbh dest1 rthdr ah* [esp* dest2 payload]
    642 	 */
    643 	while (1) {
    644 		if (*i + sizeof(ip6e) > m->m_pkthdr.len) {
    645 			return EINVAL;
    646 		}
    647 
    648 		switch (nxt) {
    649 		case IPPROTO_AH:
    650 		case IPPROTO_ESP:
    651 		case IPPROTO_IPCOMP:
    652 		/*
    653 		 * we should not skip security header added
    654 		 * beforehand.
    655 		 */
    656 			return 0;
    657 
    658 		case IPPROTO_HOPOPTS:
    659 		case IPPROTO_DSTOPTS:
    660 		case IPPROTO_ROUTING:
    661 		/*
    662 		 * if we see 2nd destination option header,
    663 		 * we should stop there.
    664 		 */
    665 			if (nxt == IPPROTO_DSTOPTS && dstopt)
    666 				return 0;
    667 
    668 			if (nxt == IPPROTO_DSTOPTS) {
    669 				/*
    670 				 * seen 1st or 2nd destination option.
    671 				 * next time we see one, it must be 2nd.
    672 				 */
    673 				dstopt = 1;
    674 			} else if (nxt == IPPROTO_ROUTING) {
    675 				/*
    676 				 * if we see destination option next
    677 				 * time, it must be dest2.
    678 				 */
    679 				dstopt = 2;
    680 			}
    681 
    682 			/* skip this header */
    683 			m_copydata(m, *i, sizeof(ip6e), &ip6e);
    684 			nxt = ip6e.ip6e_nxt;
    685 			*off = *i + offsetof(struct ip6_ext, ip6e_nxt);
    686 			*i += (ip6e.ip6e_len + 1) << 3;
    687 			break;
    688 		default:
    689 			return 0;
    690 		}
    691 	}
    692 
    693 	return 0;
    694 }
    695 
    696 static int
    697 in6_sa_equal_addrwithscope(const struct sockaddr_in6 *sa,
    698     const struct in6_addr *ia)
    699 {
    700 	struct in6_addr ia2;
    701 
    702 	memcpy(&ia2, &sa->sin6_addr, sizeof(ia2));
    703 	if (IN6_IS_SCOPE_LINKLOCAL(&sa->sin6_addr))
    704 		ia2.s6_addr16[1] = htons(sa->sin6_scope_id);
    705 
    706 	return IN6_ARE_ADDR_EQUAL(ia, &ia2);
    707 }
    708 
    709 int
    710 ipsec6_process_packet(struct mbuf *m, const struct ipsecrequest *isr)
    711 {
    712 	struct secasvar *sav = NULL;
    713 	struct ip6_hdr *ip6;
    714 	int s, error, i, off;
    715 	union sockaddr_union *dst;
    716 
    717 	KASSERT(m != NULL);
    718 	KASSERT(m->m_nextpkt == NULL);
    719 	KASSERT(isr != NULL);
    720 
    721 	s = splsoftnet();   /* insure SA contents don't change */
    722 
    723 	isr = ipsec_nextisr(m, isr, AF_INET6, &error, &sav);
    724 	if (isr == NULL) {
    725 		if (error != 0) {
    726 			/* XXX Should we send a notification ? */
    727 			goto bad;
    728 		} else {
    729 			if (ipsec_register_done(m, &error) < 0)
    730 				goto bad;
    731 
    732 			splx(s);
    733 			return ipsec_reinject_ipstack(m, AF_INET6);
    734 		}
    735 	}
    736 
    737 	KASSERT(sav != NULL);
    738 	dst = &sav->sah->saidx.dst;
    739 
    740 	if (m->m_len < sizeof(struct ip6_hdr)) {
    741 		if ((m = m_pullup(m,sizeof(struct ip6_hdr))) == NULL) {
    742 			error = ENOBUFS;
    743 			goto unrefsav;
    744 		}
    745 	}
    746 	ip6 = mtod(m, struct ip6_hdr *);
    747 
    748 	/* Do the appropriate encapsulation, if necessary */
    749 	if (isr->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
    750 	    dst->sa.sa_family != AF_INET6 ||        /* AF mismatch */
    751 	    ((dst->sa.sa_family == AF_INET6) &&
    752 	     (!IN6_IS_ADDR_UNSPECIFIED(&dst->sin6.sin6_addr)) &&
    753 	     (!in6_sa_equal_addrwithscope(&dst->sin6, &ip6->ip6_dst)))) {
    754 		struct mbuf *mp;
    755 
    756 		if (m->m_pkthdr.len - sizeof(*ip6) > IPV6_MAXPACKET) {
    757 			/* No jumbogram support. */
    758 			error = ENXIO;   /*XXX*/
    759 			goto unrefsav;
    760 		}
    761 
    762 		/* Fix IPv6 header payload length. */
    763 		ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
    764 
    765 		/* Encapsulate the packet */
    766 		error = ipip_output(m, sav, &mp);
    767 		if (mp == NULL && !error) {
    768 			/* Should never happen. */
    769 			IPSECLOG(LOG_DEBUG,
    770 			    "ipip_output returns no mbuf and no error!");
    771 			error = EFAULT;
    772 		}
    773 
    774 		if (error) {
    775 			if (mp) {
    776 				/* XXX: Should never happen! */
    777 				m_freem(mp);
    778 			}
    779 			m = NULL; /* ipip_output() already freed it */
    780 			goto unrefsav;
    781 		}
    782 
    783 		m = mp;
    784 		mp = NULL;
    785 	}
    786 
    787 	if (dst->sa.sa_family == AF_INET) {
    788 		struct ip *ip;
    789 		ip = mtod(m, struct ip *);
    790 		i = ip->ip_hl << 2;
    791 		off = offsetof(struct ip, ip_p);
    792 	} else {
    793 		error = compute_ipsec_pos(m, &i, &off);
    794 		if (error)
    795 			goto unrefsav;
    796 	}
    797 	error = (*sav->tdb_xform->xf_output)(m, isr, sav, i, off);
    798 	KEY_SA_UNREF(&sav);
    799 	splx(s);
    800 	return error;
    801 
    802 unrefsav:
    803 	KEY_SA_UNREF(&sav);
    804 bad:
    805 	splx(s);
    806 	if (m)
    807 		m_freem(m);
    808 	return error;
    809 }
    810 #endif /* INET6 */
    811 
    812 void
    813 ipsec_output_init(void)
    814 {
    815 
    816 	ipsec_rtcache_percpu = percpu_alloc(sizeof(struct route));
    817 }
    818