sys/netipsec/ipsec_output.c

   1 /*-
   2  * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  *
  26  * $FreeBSD$
  27  */
  28
  29 /*
  30  * IPsec output processing.
  31  */
  32 #include "opt_inet.h"
  33 #include "opt_inet6.h"
  34 #include "opt_ipsec.h"
  35 #include "opt_enc.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/mbuf.h>
  40 #include <sys/domain.h>
  41 #include <sys/protosw.h>
  42 #include <sys/socket.h>
  43 #include <sys/errno.h>
  44 #include <sys/syslog.h>
  45
  46 #include <net/if.h>
  47 #include <net/pfil.h>
  48 #include <net/route.h>
  49 #include <net/vnet.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 #ifdef INET6
  58 #include <netinet6/ip6_ecn.h>
  59 #endif
  60
  61 #include <netinet/ip6.h>
  62 #ifdef INET6
  63 #include <netinet6/ip6_var.h>
  64 #endif
  65 #include <netinet/in_pcb.h>
  66 #ifdef INET6
  67 #include <netinet/icmp6.h>
  68 #endif
  69
  70 #include <netipsec/ipsec.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 #include <machine/in_cksum.h>
  85
  86 #ifdef IPSEC_NAT_T
  87 #include <netinet/udp.h>
  88 #endif
  89
  90 #ifdef DEV_ENC
  91 #include <net/if_enc.h>
  92 #endif
  93
  94
  95 int
  96 ipsec_process_done(struct mbuf *m, struct ipsecrequest *isr)
  97 {
  98         struct tdb_ident *tdbi;
  99         struct m_tag *mtag;
 100         struct secasvar *sav;
 101         struct secasindex *saidx;
 102         int error;
 103
 104         IPSEC_ASSERT(m != NULL, ("null mbuf"));
 105         IPSEC_ASSERT(isr != NULL, ("null ISR"));
 106         sav = isr->sav;
 107         IPSEC_ASSERT(sav != NULL, ("null SA"));
 108         IPSEC_ASSERT(sav->sah != NULL, ("null SAH"));
 109
 110         saidx = &sav->sah->saidx;
 111         switch (saidx->dst.sa.sa_family) {
 112 #ifdef INET
 113         case AF_INET:
 114                 /* Fix the header length, for AH processing. */
 115                 mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
 116                 break;
 117 #endif /* INET */
 118 #ifdef INET6
 119         case AF_INET6:
 120                 /* Fix the header length, for AH processing. */
 121                 if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
 122                         error = ENXIO;
 123                         goto bad;
 124                 }
 125                 if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
 126                         /* No jumbogram support. */
 127                         error = ENXIO;  /*?*/
 128                         goto bad;
 129                 }
 130                 mtod(m, struct ip6_hdr *)->ip6_plen =
 131                         htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
 132                 break;
 133 #endif /* INET6 */
 134         default:
 135                 DPRINTF(("%s: unknown protocol family %u\n", __func__,
 136                     saidx->dst.sa.sa_family));
 137                 error = ENXIO;
 138                 goto bad;
 139         }
 140
 141         /*
 142          * Add a record of what we've done or what needs to be done to the
 143          * packet.
 144          */
 145         mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE,
 146                         sizeof(struct tdb_ident), M_NOWAIT);
 147         if (mtag == NULL) {
 148                 DPRINTF(("%s: could not get packet tag\n", __func__));
 149                 error = ENOMEM;
 150                 goto bad;
 151         }
 152
 153         tdbi = (struct tdb_ident *)(mtag + 1);
 154         tdbi->dst = saidx->dst;
 155         tdbi->proto = saidx->proto;
 156         tdbi->spi = sav->spi;
 157         m_tag_prepend(m, mtag);
 158
 159         /*
 160          * If there's another (bundled) SA to apply, do so.
 161          * Note that this puts a burden on the kernel stack size.
 162          * If this is a problem we'll need to introduce a queue
 163          * to set the packet on so we can unwind the stack before
 164          * doing further processing.
 165          */
 166         if (isr->next) {
 167                 IPSECSTAT_INC(ips_out_bundlesa);
 168                 /* XXX-BZ currently only support same AF bundles. */
 169                 switch (saidx->dst.sa.sa_family) {
 170 #ifdef INET
 171                 case AF_INET:
 172                         return ipsec4_process_packet(m, isr->next, 0, 0);
 173                         /* NOTREACHED */
 174 #endif
 175 #ifdef notyet
 176 #ifdef INET6
 177                 case AF_INET6:
 178                         /* XXX */
 179                         ipsec6_output_trans()
 180                         ipsec6_output_tunnel()
 181                         /* NOTREACHED */
 182 #endif /* INET6 */
 183 #endif
 184                 default:
 185                         DPRINTF(("%s: unknown protocol family %u\n", __func__,
 186                             saidx->dst.sa.sa_family));
 187                         error = ENXIO;
 188                         goto bad;
 189                 }
 190         }
 191         key_sa_recordxfer(sav, m);              /* record data transfer */
 192
 193         /*
 194          * We're done with IPsec processing, transmit the packet using the
 195          * appropriate network protocol (IP or IPv6). SPD lookup will be
 196          * performed again there.
 197          */
 198         switch (saidx->dst.sa.sa_family) {
 199 #ifdef INET
 200         case AF_INET:
 201 #ifdef IPSEC_NAT_T
 202                 /*
 203                  * If NAT-T is enabled, now that all IPsec processing is done
 204                  * insert UDP encapsulation header after IP header.
 205                  */
 206                 if (sav->natt_type) {
 207                         struct ip *ip = mtod(m, struct ip *);
 208                         const int hlen = (ip->ip_hl << 2);
 209                         int size, off;
 210                         struct mbuf *mi;
 211                         struct udphdr *udp;
 212
 213                         size = sizeof(struct udphdr);
 214                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
 215                                 /*
 216                                  * draft-ietf-ipsec-nat-t-ike-0[01].txt and
 217                                  * draft-ietf-ipsec-udp-encaps-(00/)01.txt,
 218                                  * ignoring possible AH mode
 219                                  * non-IKE marker + non-ESP marker
 220                                  * from draft-ietf-ipsec-udp-encaps-00.txt.
 221                                  */
 222                                 size += sizeof(u_int64_t);
 223                         }
 224                         mi = m_makespace(m, hlen, size, &off);
 225                         if (mi == NULL) {
 226                                 DPRINTF(("%s: m_makespace for udphdr failed\n",
 227                                     __func__));
 228                                 error = ENOBUFS;
 229                                 goto bad;
 230                         }
 231
 232                         udp = (struct udphdr *)(mtod(mi, caddr_t) + off);
 233                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
 234                                 udp->uh_sport = htons(UDP_ENCAP_ESPINUDP_PORT);
 235                         else
 236                                 udp->uh_sport =
 237                                         KEY_PORTFROMSADDR(&sav->sah->saidx.src);
 238                         udp->uh_dport = KEY_PORTFROMSADDR(&sav->sah->saidx.dst);
 239                         udp->uh_sum = 0;
 240                         udp->uh_ulen = htons(m->m_pkthdr.len - hlen);
 241                         ip->ip_len = htons(m->m_pkthdr.len);
 242                         ip->ip_p = IPPROTO_UDP;
 243
 244                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
 245                                 *(u_int64_t *)(udp + 1) = 0;
 246                 }
 247 #endif /* IPSEC_NAT_T */
 248
 249                 return ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
 250 #endif /* INET */
 251 #ifdef INET6
 252         case AF_INET6:
 253                 /*
 254                  * We don't need massage, IPv6 header fields are always in
 255                  * net endian.
 256                  */
 257                 return ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
 258 #endif /* INET6 */
 259         }
 260         panic("ipsec_process_done");
 261 bad:
 262         m_freem(m);
 263         return (error);
 264 }
 265
 266 static struct ipsecrequest *
 267 ipsec_nextisr(
 268         struct mbuf *m,
 269         struct ipsecrequest *isr,
 270         int af,
 271         struct secasindex *saidx,
 272         int *error
 273 )
 274 {
 275 #define IPSEC_OSTAT(name)       do {            \
 276         if (isr->saidx.proto == IPPROTO_ESP)    \
 277                 ESPSTAT_INC(esps_##name);       \
 278         else if (isr->saidx.proto == IPPROTO_AH)\
 279                 AHSTAT_INC(ahs_##name);         \
 280         else                                    \
 281                 IPCOMPSTAT_INC(ipcomps_##name); \
 282 } while (0)
 283         struct secasvar *sav;
 284
 285         IPSECREQUEST_LOCK_ASSERT(isr);
 286
 287         IPSEC_ASSERT(af == AF_INET || af == AF_INET6,
 288                 ("invalid address family %u", af));
 289 again:
 290         /*
 291          * Craft SA index to search for proper SA.  Note that
 292          * we only fillin unspecified SA peers for transport
 293          * mode; for tunnel mode they must already be filled in.
 294          */
 295         *saidx = isr->saidx;
 296         if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
 297                 /* Fillin unspecified SA peers only for transport mode */
 298                 if (af == AF_INET) {
 299                         struct sockaddr_in *sin;
 300                         struct ip *ip = mtod(m, struct ip *);
 301
 302                         if (saidx->src.sa.sa_len == 0) {
 303                                 sin = &saidx->src.sin;
 304                                 sin->sin_len = sizeof(*sin);
 305                                 sin->sin_family = AF_INET;
 306                                 sin->sin_port = IPSEC_PORT_ANY;
 307                                 sin->sin_addr = ip->ip_src;
 308                         }
 309                         if (saidx->dst.sa.sa_len == 0) {
 310                                 sin = &saidx->dst.sin;
 311                                 sin->sin_len = sizeof(*sin);
 312                                 sin->sin_family = AF_INET;
 313                                 sin->sin_port = IPSEC_PORT_ANY;
 314                                 sin->sin_addr = ip->ip_dst;
 315                         }
 316                 } else {
 317                         struct sockaddr_in6 *sin6;
 318                         struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
 319
 320                         if (saidx->src.sin6.sin6_len == 0) {
 321                                 sin6 = (struct sockaddr_in6 *)&saidx->src;
 322                                 sin6->sin6_len = sizeof(*sin6);
 323                                 sin6->sin6_family = AF_INET6;
 324                                 sin6->sin6_port = IPSEC_PORT_ANY;
 325                                 sin6->sin6_addr = ip6->ip6_src;
 326                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
 327                                         /* fix scope id for comparing SPD */
 328                                         sin6->sin6_addr.s6_addr16[1] = 0;
 329                                         sin6->sin6_scope_id =
 330                                             ntohs(ip6->ip6_src.s6_addr16[1]);
 331                                 }
 332                         }
 333                         if (saidx->dst.sin6.sin6_len == 0) {
 334                                 sin6 = (struct sockaddr_in6 *)&saidx->dst;
 335                                 sin6->sin6_len = sizeof(*sin6);
 336                                 sin6->sin6_family = AF_INET6;
 337                                 sin6->sin6_port = IPSEC_PORT_ANY;
 338                                 sin6->sin6_addr = ip6->ip6_dst;
 339                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
 340                                         /* fix scope id for comparing SPD */
 341                                         sin6->sin6_addr.s6_addr16[1] = 0;
 342                                         sin6->sin6_scope_id =
 343                                             ntohs(ip6->ip6_dst.s6_addr16[1]);
 344                                 }
 345                         }
 346                 }
 347         }
 348
 349         /*
 350          * Lookup SA and validate it.
 351          */
 352         *error = key_checkrequest(isr, saidx);
 353         if (*error != 0) {
 354                 /*
 355                  * IPsec processing is required, but no SA found.
 356                  * I assume that key_acquire() had been called
 357                  * to get/establish the SA. Here I discard
 358                  * this packet because it is responsibility for
 359                  * upper layer to retransmit the packet.
 360                  */
 361                 IPSECSTAT_INC(ips_out_nosa);
 362                 goto bad;
 363         }
 364         sav = isr->sav;
 365         if (sav == NULL) {
 366                 IPSEC_ASSERT(ipsec_get_reqlevel(isr) == IPSEC_LEVEL_USE,
 367                         ("no SA found, but required; level %u",
 368                         ipsec_get_reqlevel(isr)));
 369                 IPSECREQUEST_UNLOCK(isr);
 370                 isr = isr->next;
 371                 /*
 372                  * If isr is NULL, we found a 'use' policy w/o SA.
 373                  * Return w/o error and w/o isr so we can drop out
 374                  * and continue w/o IPsec processing.
 375                  */
 376                 if (isr == NULL)
 377                         return isr;
 378                 IPSECREQUEST_LOCK(isr);
 379                 goto again;
 380         }
 381
 382         /*
 383          * Check system global policy controls.
 384          */
 385         if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
 386             (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
 387             (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
 388                 DPRINTF(("%s: IPsec outbound packet dropped due"
 389                         " to policy (check your sysctls)\n", __func__));
 390                 IPSEC_OSTAT(pdrops);
 391                 *error = EHOSTUNREACH;
 392                 goto bad;
 393         }
 394
 395         /*
 396          * Sanity check the SA contents for the caller
 397          * before they invoke the xform output method.
 398          */
 399         if (sav->tdb_xform == NULL) {
 400                 DPRINTF(("%s: no transform for SA\n", __func__));
 401                 IPSEC_OSTAT(noxform);
 402                 *error = EHOSTUNREACH;
 403                 goto bad;
 404         }
 405         return isr;
 406 bad:
 407         IPSEC_ASSERT(*error != 0, ("error return w/ no error code"));
 408         IPSECREQUEST_UNLOCK(isr);
 409         return NULL;
 410 #undef IPSEC_OSTAT
 411 }
 412
 413 #ifdef INET
 414 /*
 415  * IPsec output logic for IPv4.
 416  */
 417 int
 418 ipsec4_process_packet(
 419         struct mbuf *m,
 420         struct ipsecrequest *isr,
 421         int flags,
 422         int tunalready)
 423 {
 424         struct secasindex saidx;
 425         struct secasvar *sav;
 426         struct ip *ip;
 427         int error, i, off;
 428
 429         IPSEC_ASSERT(m != NULL, ("null mbuf"));
 430         IPSEC_ASSERT(isr != NULL, ("null isr"));
 431
 432         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 433
 434         isr = ipsec_nextisr(m, isr, AF_INET, &saidx, &error);
 435         if (isr == NULL) {
 436                 if (error != 0)
 437                         goto bad;
 438                 return EJUSTRETURN;
 439         }
 440
 441         sav = isr->sav;
 442
 443 #ifdef DEV_ENC
 444         encif->if_opackets++;
 445         encif->if_obytes += m->m_pkthdr.len;
 446
 447         /* pass the mbuf to enc0 for bpf processing */
 448         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_BEFORE);
 449         /* pass the mbuf to enc0 for packet filtering */
 450         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 451                 goto bad;
 452 #endif
 453
 454         if (!tunalready) {
 455                 union sockaddr_union *dst = &sav->sah->saidx.dst;
 456                 int setdf;
 457
 458                 /*
 459                  * Collect IP_DF state from the outer header.
 460                  */
 461                 if (dst->sa.sa_family == AF_INET) {
 462                         if (m->m_len < sizeof (struct ip) &&
 463                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 464                                 error = ENOBUFS;
 465                                 goto bad;
 466                         }
 467                         ip = mtod(m, struct ip *);
 468                         /* Honor system-wide control of how to handle IP_DF */
 469                         switch (V_ip4_ipsec_dfbit) {
 470                         case 0:                 /* clear in outer header */
 471                         case 1:                 /* set in outer header */
 472                                 setdf = V_ip4_ipsec_dfbit;
 473                                 break;
 474                         default:                /* propagate to outer header */
 475                                 setdf = ntohs(ip->ip_off & IP_DF);
 476                                 break;
 477                         }
 478                 } else {
 479                         ip = NULL;              /* keep compiler happy */
 480                         setdf = 0;
 481                 }
 482                 /* Do the appropriate encapsulation, if necessary */
 483                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
 484                     dst->sa.sa_family != AF_INET ||         /* PF mismatch */
 485 #if 0
 486                     (sav->flags & SADB_X_SAFLAGS_TUNNEL) || /* Tunnel requ'd */
 487                     sav->tdb_xform->xf_type == XF_IP4 ||    /* ditto */
 488 #endif
 489                     (dst->sa.sa_family == AF_INET &&        /* Proxy */
 490                      dst->sin.sin_addr.s_addr != INADDR_ANY &&
 491                      dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
 492                         struct mbuf *mp;
 493
 494                         /* Fix IPv4 header checksum and length */
 495                         if (m->m_len < sizeof (struct ip) &&
 496                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 497                                 error = ENOBUFS;
 498                                 goto bad;
 499                         }
 500                         ip = mtod(m, struct ip *);
 501                         ip->ip_len = htons(m->m_pkthdr.len);
 502                         ip->ip_sum = 0;
 503                         ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
 504
 505                         /* Encapsulate the packet */
 506                         error = ipip_output(m, isr, &mp, 0, 0);
 507                         if (mp == NULL && !error) {
 508                                 /* Should never happen. */
 509                                 DPRINTF(("%s: ipip_output returns no mbuf and "
 510                                         "no error!", __func__));
 511                                 error = EFAULT;
 512                         }
 513                         if (error) {
 514                                 if (mp) {
 515                                         /* XXX: Should never happen! */
 516                                         m_freem(mp);
 517                                 }
 518                                 m = NULL; /* ipip_output() already freed it */
 519                                 goto bad;
 520                         }
 521                         m = mp, mp = NULL;
 522                         /*
 523                          * ipip_output clears IP_DF in the new header.  If
 524                          * we need to propagate IP_DF from the outer header,
 525                          * then we have to do it here.
 526                          *
 527                          * XXX shouldn't assume what ipip_output does.
 528                          */
 529                         if (dst->sa.sa_family == AF_INET && setdf) {
 530                                 if (m->m_len < sizeof (struct ip) &&
 531                                     (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 532                                         error = ENOBUFS;
 533                                         goto bad;
 534                                 }
 535                                 ip = mtod(m, struct ip *);
 536                                 ip->ip_off = ntohs(ip->ip_off);
 537                                 ip->ip_off |= IP_DF;
 538                                 ip->ip_off = htons(ip->ip_off);
 539                         }
 540                 }
 541         }
 542
 543 #ifdef DEV_ENC
 544         /* pass the mbuf to enc0 for bpf processing */
 545         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_AFTER);
 546         /* pass the mbuf to enc0 for packet filtering */
 547         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 548                 goto bad;
 549 #endif
 550
 551         /*
 552          * Dispatch to the appropriate IPsec transform logic.  The
 553          * packet will be returned for transmission after crypto
 554          * processing, etc. are completed.  For encapsulation we
 555          * bypass this call because of the explicit call done above
 556          * (necessary to deal with IP_DF handling for IPv4).
 557          *
 558          * NB: m & sav are ``passed to caller'' who's reponsible for
 559          *     for reclaiming their resources.
 560          */
 561         if (sav->tdb_xform->xf_type != XF_IP4) {
 562                 ip = mtod(m, struct ip *);
 563                 i = ip->ip_hl << 2;
 564                 off = offsetof(struct ip, ip_p);
 565                 error = (*sav->tdb_xform->xf_output)(m, isr, NULL, i, off);
 566         } else {
 567                 error = ipsec_process_done(m, isr);
 568         }
 569         IPSECREQUEST_UNLOCK(isr);
 570         return error;
 571 bad:
 572         if (isr)
 573                 IPSECREQUEST_UNLOCK(isr);
 574         if (m)
 575                 m_freem(m);
 576         return error;
 577 }
 578 #endif
 579
 580 #ifdef INET6
 581 /*
 582  * Chop IP6 header from the payload.
 583  */
 584 static struct mbuf *
 585 ipsec6_splithdr(struct mbuf *m)
 586 {
 587         struct mbuf *mh;
 588         struct ip6_hdr *ip6;
 589         int hlen;
 590
 591         IPSEC_ASSERT(m->m_len >= sizeof (struct ip6_hdr),
 592                 ("first mbuf too short, len %u", m->m_len));
 593         ip6 = mtod(m, struct ip6_hdr *);
 594         hlen = sizeof(struct ip6_hdr);
 595         if (m->m_len > hlen) {
 596                 MGETHDR(mh, M_NOWAIT, MT_DATA);
 597                 if (!mh) {
 598                         m_freem(m);
 599                         return NULL;
 600                 }
 601                 M_MOVE_PKTHDR(mh, m);
 602                 MH_ALIGN(mh, hlen);
 603                 m->m_len -= hlen;
 604                 m->m_data += hlen;
 605                 mh->m_next = m;
 606                 m = mh;
 607                 m->m_len = hlen;
 608                 bcopy((caddr_t)ip6, mtod(m, caddr_t), hlen);
 609         } else if (m->m_len < hlen) {
 610                 m = m_pullup(m, hlen);
 611                 if (!m)
 612                         return NULL;
 613         }
 614         return m;
 615 }
 616
 617 /*
 618  * IPsec output logic for IPv6, transport mode.
 619  */
 620 int
 621 ipsec6_output_trans(
 622         struct ipsec_output_state *state,
 623         u_char *nexthdrp,
 624         struct mbuf *mprev,
 625         struct secpolicy *sp,
 626         int flags,
 627         int *tun)
 628 {
 629         struct ipsecrequest *isr;
 630         struct secasindex saidx;
 631         int error = 0;
 632         struct mbuf *m;
 633
 634         IPSEC_ASSERT(state != NULL, ("null state"));
 635         IPSEC_ASSERT(state->m != NULL, ("null m"));
 636         IPSEC_ASSERT(nexthdrp != NULL, ("null nexthdrp"));
 637         IPSEC_ASSERT(mprev != NULL, ("null mprev"));
 638         IPSEC_ASSERT(sp != NULL, ("null sp"));
 639         IPSEC_ASSERT(tun != NULL, ("null tun"));
 640
 641         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 642                 printf("%s: applied SP\n", __func__);
 643                 kdebug_secpolicy(sp));
 644
 645         isr = sp->req;
 646         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 647                 /* the rest will be handled by ipsec6_output_tunnel() */
 648                 *tun = 1;               /* need tunnel-mode processing */
 649                 return 0;
 650         }
 651
 652         *tun = 0;
 653         m = state->m;
 654
 655         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 656         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 657         if (isr == NULL) {
 658                 if (error != 0) {
 659 #ifdef notdef
 660                         /* XXX should notification be done for all errors ? */
 661                         /*
 662                          * Notify the fact that the packet is discarded
 663                          * to ourselves. I believe this is better than
 664                          * just silently discarding. (jinmei@kame.net)
 665                          * XXX: should we restrict the error to TCP packets?
 666                          * XXX: should we directly notify sockets via
 667                          *      pfctlinputs?
 668                          */
 669                         icmp6_error(m, ICMP6_DST_UNREACH,
 670                                     ICMP6_DST_UNREACH_ADMIN, 0);
 671                         m = NULL;       /* NB: icmp6_error frees mbuf */
 672 #endif
 673                         goto bad;
 674                 }
 675                 return EJUSTRETURN;
 676         }
 677
 678         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 679                                                   sizeof (struct ip6_hdr),
 680                                                   offsetof(struct ip6_hdr,
 681                                                            ip6_nxt));
 682         IPSECREQUEST_UNLOCK(isr);
 683         return error;
 684 bad:
 685         if (isr)
 686                 IPSECREQUEST_UNLOCK(isr);
 687         if (m)
 688                 m_freem(m);
 689         state->m = NULL;
 690         return error;
 691 }
 692
 693 static int
 694 ipsec6_encapsulate(struct mbuf *m, struct secasvar *sav)
 695 {
 696         struct ip6_hdr *oip6;
 697         struct ip6_hdr *ip6;
 698         size_t plen;
 699
 700         /* can't tunnel between different AFs */
 701         if (sav->sah->saidx.src.sa.sa_family != AF_INET6 ||
 702             sav->sah->saidx.dst.sa.sa_family != AF_INET6) {
 703                 m_freem(m);
 704                 return EINVAL;
 705         }
 706         IPSEC_ASSERT(m->m_len == sizeof (struct ip6_hdr),
 707                 ("mbuf wrong size; len %u", m->m_len));
 708
 709
 710         /*
 711          * grow the mbuf to accomodate the new IPv6 header.
 712          */
 713         plen = m->m_pkthdr.len;
 714         if (M_LEADINGSPACE(m->m_next) < sizeof(struct ip6_hdr)) {
 715                 struct mbuf *n;
 716                 MGET(n, M_NOWAIT, MT_DATA);
 717                 if (!n) {
 718                         m_freem(m);
 719                         return ENOBUFS;
 720                 }
 721                 n->m_len = sizeof(struct ip6_hdr);
 722                 n->m_next = m->m_next;
 723                 m->m_next = n;
 724                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 725                 oip6 = mtod(n, struct ip6_hdr *);
 726         } else {
 727                 m->m_next->m_len += sizeof(struct ip6_hdr);
 728                 m->m_next->m_data -= sizeof(struct ip6_hdr);
 729                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 730                 oip6 = mtod(m->m_next, struct ip6_hdr *);
 731         }
 732         ip6 = mtod(m, struct ip6_hdr *);
 733         bcopy((caddr_t)ip6, (caddr_t)oip6, sizeof(struct ip6_hdr));
 734
 735         /* Fake link-local scope-class addresses */
 736         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_src))
 737                 oip6->ip6_src.s6_addr16[1] = 0;
 738         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_dst))
 739                 oip6->ip6_dst.s6_addr16[1] = 0;
 740
 741         /* construct new IPv6 header. see RFC 2401 5.1.2.2 */
 742         /* ECN consideration. */
 743         ip6_ecn_ingress(V_ip6_ipsec_ecn, &ip6->ip6_flow, &oip6->ip6_flow);
 744         if (plen < IPV6_MAXPACKET - sizeof(struct ip6_hdr))
 745                 ip6->ip6_plen = htons(plen);
 746         else {
 747                 /* ip6->ip6_plen will be updated in ip6_output() */
 748         }
 749         ip6->ip6_nxt = IPPROTO_IPV6;
 750         ip6->ip6_src = sav->sah->saidx.src.sin6.sin6_addr;
 751         ip6->ip6_dst = sav->sah->saidx.dst.sin6.sin6_addr;
 752         ip6->ip6_hlim = IPV6_DEFHLIM;
 753
 754         /* XXX Should ip6_src be updated later ? */
 755
 756         return 0;
 757 }
 758
 759 /*
 760  * IPsec output logic for IPv6, tunnel mode.
 761  */
 762 int
 763 ipsec6_output_tunnel(struct ipsec_output_state *state, struct secpolicy *sp, int flags)
 764 {
 765         struct ip6_hdr *ip6;
 766         struct ipsecrequest *isr;
 767         struct secasindex saidx;
 768         int error;
 769         struct sockaddr_in6 *dst6;
 770         struct mbuf *m;
 771
 772         IPSEC_ASSERT(state != NULL, ("null state"));
 773         IPSEC_ASSERT(state->m != NULL, ("null m"));
 774         IPSEC_ASSERT(sp != NULL, ("null sp"));
 775
 776         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 777                 printf("%s: applied SP\n", __func__);
 778                 kdebug_secpolicy(sp));
 779
 780         m = state->m;
 781         /*
 782          * transport mode ipsec (before the 1st tunnel mode) is already
 783          * processed by ipsec6_output_trans().
 784          */
 785         for (isr = sp->req; isr; isr = isr->next) {
 786                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
 787                         break;
 788         }
 789
 790         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 791         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 792         if (isr == NULL) {
 793                 if (error != 0)
 794                         goto bad;
 795                 return EJUSTRETURN;
 796         }
 797
 798 #ifdef DEV_ENC
 799         encif->if_opackets++;
 800         encif->if_obytes += m->m_pkthdr.len;
 801
 802         /* pass the mbuf to enc0 for bpf processing */
 803         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_BEFORE);
 804         /* pass the mbuf to enc0 for packet filtering */
 805         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 806                 goto bad;
 807 #endif
 808
 809         /*
 810          * There may be the case that SA status will be changed when
 811          * we are refering to one. So calling splsoftnet().
 812          */
 813         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 814                 /*
 815                  * build IPsec tunnel.
 816                  */
 817                 /* XXX should be processed with other familiy */
 818                 if (isr->sav->sah->saidx.src.sa.sa_family != AF_INET6) {
 819                         ipseclog((LOG_ERR, "%s: family mismatched between "
 820                             "inner and outer, spi=%u\n", __func__,
 821                             ntohl(isr->sav->spi)));
 822                         IPSEC6STAT_INC(ips_out_inval);
 823                         error = EAFNOSUPPORT;
 824                         goto bad;
 825                 }
 826
 827                 m = ipsec6_splithdr(m);
 828                 if (!m) {
 829                         IPSEC6STAT_INC(ips_out_nomem);
 830                         error = ENOMEM;
 831                         goto bad;
 832                 }
 833                 error = ipsec6_encapsulate(m, isr->sav);
 834                 if (error) {
 835                         m = NULL;
 836                         goto bad;
 837                 }
 838                 ip6 = mtod(m, struct ip6_hdr *);
 839
 840                 state->ro =
 841                     (struct route *)&isr->sav->sah->route_cache.sin6_route;
 842                 state->dst = (struct sockaddr *)&state->ro->ro_dst;
 843                 dst6 = (struct sockaddr_in6 *)state->dst;
 844                 if (state->ro->ro_rt
 845                  && ((state->ro->ro_rt->rt_flags & RTF_UP) == 0
 846                   || !IN6_ARE_ADDR_EQUAL(&dst6->sin6_addr, &ip6->ip6_dst))) {
 847                         RTFREE(state->ro->ro_rt);
 848                         state->ro->ro_rt = NULL;
 849                 }
 850                 if (state->ro->ro_rt == NULL) {
 851                         bzero(dst6, sizeof(*dst6));
 852                         dst6->sin6_family = AF_INET6;
 853                         dst6->sin6_len = sizeof(*dst6);
 854                         dst6->sin6_addr = ip6->ip6_dst;
 855                         rtalloc_ign_fib(state->ro, 0UL, M_GETFIB(m));
 856                 }
 857                 if (state->ro->ro_rt == NULL) {
 858                         IP6STAT_INC(ip6s_noroute);
 859                         IPSEC6STAT_INC(ips_out_noroute);
 860                         error = EHOSTUNREACH;
 861                         goto bad;
 862                 }
 863
 864                 /* adjust state->dst if tunnel endpoint is offlink */
 865                 if (state->ro->ro_rt->rt_flags & RTF_GATEWAY)
 866                         state->dst = (struct sockaddr *)state->ro->ro_rt->rt_gateway;
 867         }
 868
 869         m = ipsec6_splithdr(m);
 870         if (!m) {
 871                 IPSEC6STAT_INC(ips_out_nomem);
 872                 error = ENOMEM;
 873                 goto bad;
 874         }
 875         ip6 = mtod(m, struct ip6_hdr *);
 876
 877 #ifdef DEV_ENC
 878         /* pass the mbuf to enc0 for bpf processing */
 879         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_AFTER);
 880         /* pass the mbuf to enc0 for packet filtering */
 881         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 882                 goto bad;
 883 #endif
 884
 885         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 886                 sizeof (struct ip6_hdr),
 887                 offsetof(struct ip6_hdr, ip6_nxt));
 888         IPSECREQUEST_UNLOCK(isr);
 889         return error;
 890 bad:
 891         if (isr)
 892                 IPSECREQUEST_UNLOCK(isr);
 893         if (m)
 894                 m_freem(m);
 895         state->m = NULL;
 896         return error;
 897 }
 898 #endif /*INET6*/