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                 V_ipsec4stat.ips_out_bundlesa++;
 168                 return ipsec4_process_packet(m, isr->next, 0, 0);
 169         }
 170         key_sa_recordxfer(sav, m);              /* record data transfer */
 171
 172         /*
 173          * We're done with IPsec processing, transmit the packet using the
 174          * appropriate network protocol (IP or IPv6). SPD lookup will be
 175          * performed again there.
 176          */
 177         switch (saidx->dst.sa.sa_family) {
 178 #ifdef INET
 179         struct ip *ip;
 180         case AF_INET:
 181                 ip = mtod(m, struct ip *);
 182                 ip->ip_len = ntohs(ip->ip_len);
 183                 ip->ip_off = ntohs(ip->ip_off);
 184
 185 #ifdef IPSEC_NAT_T
 186                 /*
 187                  * If NAT-T is enabled, now that all IPsec processing is done
 188                  * insert UDP encapsulation header after IP header.
 189                  */
 190                 if (sav->natt_type) {
 191 #ifdef _IP_VHL
 192                         const int hlen = IP_VHL_HL(ip->ip_vhl);
 193 #else
 194                         const int hlen = (ip->ip_hl << 2);
 195 #endif
 196                         int size, off;
 197                         struct mbuf *mi;
 198                         struct udphdr *udp;
 199
 200                         size = sizeof(struct udphdr);
 201                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
 202                                 /*
 203                                  * draft-ietf-ipsec-nat-t-ike-0[01].txt and
 204                                  * draft-ietf-ipsec-udp-encaps-(00/)01.txt,
 205                                  * ignoring possible AH mode
 206                                  * non-IKE marker + non-ESP marker
 207                                  * from draft-ietf-ipsec-udp-encaps-00.txt.
 208                                  */
 209                                 size += sizeof(u_int64_t);
 210                         }
 211                         mi = m_makespace(m, hlen, size, &off);
 212                         if (mi == NULL) {
 213                                 DPRINTF(("%s: m_makespace for udphdr failed\n",
 214                                     __func__));
 215                                 error = ENOBUFS;
 216                                 goto bad;
 217                         }
 218
 219                         udp = (struct udphdr *)(mtod(mi, caddr_t) + off);
 220                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
 221                                 udp->uh_sport = htons(UDP_ENCAP_ESPINUDP_PORT);
 222                         else
 223                                 udp->uh_sport =
 224                                         KEY_PORTFROMSADDR(&sav->sah->saidx.src);
 225                         udp->uh_dport = KEY_PORTFROMSADDR(&sav->sah->saidx.dst);
 226                         udp->uh_sum = 0;
 227                         udp->uh_ulen = htons(m->m_pkthdr.len - hlen);
 228                         ip->ip_len = m->m_pkthdr.len;
 229                         ip->ip_p = IPPROTO_UDP;
 230
 231                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
 232                                 *(u_int64_t *)(udp + 1) = 0;
 233                 }
 234 #endif /* IPSEC_NAT_T */
 235
 236                 return ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
 237 #endif /* INET */
 238 #ifdef INET6
 239         case AF_INET6:
 240                 /*
 241                  * We don't need massage, IPv6 header fields are always in
 242                  * net endian.
 243                  */
 244                 return ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
 245 #endif /* INET6 */
 246         }
 247         panic("ipsec_process_done");
 248 bad:
 249         m_freem(m);
 250         KEY_FREESAV(&sav);
 251         return (error);
 252 }
 253
 254 static struct ipsecrequest *
 255 ipsec_nextisr(
 256         struct mbuf *m,
 257         struct ipsecrequest *isr,
 258         int af,
 259         struct secasindex *saidx,
 260         int *error
 261 )
 262 {
 263 #define IPSEC_OSTAT(x,y,z) (isr->saidx.proto == IPPROTO_ESP ? (x)++ : \
 264                             isr->saidx.proto == IPPROTO_AH ? (y)++ : (z)++)
 265         struct secasvar *sav;
 266
 267         IPSECREQUEST_LOCK_ASSERT(isr);
 268
 269         IPSEC_ASSERT(af == AF_INET || af == AF_INET6,
 270                 ("invalid address family %u", af));
 271 again:
 272         /*
 273          * Craft SA index to search for proper SA.  Note that
 274          * we only fillin unspecified SA peers for transport
 275          * mode; for tunnel mode they must already be filled in.
 276          */
 277         *saidx = isr->saidx;
 278         if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
 279                 /* Fillin unspecified SA peers only for transport mode */
 280                 if (af == AF_INET) {
 281                         struct sockaddr_in *sin;
 282                         struct ip *ip = mtod(m, struct ip *);
 283
 284                         if (saidx->src.sa.sa_len == 0) {
 285                                 sin = &saidx->src.sin;
 286                                 sin->sin_len = sizeof(*sin);
 287                                 sin->sin_family = AF_INET;
 288                                 sin->sin_port = IPSEC_PORT_ANY;
 289                                 sin->sin_addr = ip->ip_src;
 290                         }
 291                         if (saidx->dst.sa.sa_len == 0) {
 292                                 sin = &saidx->dst.sin;
 293                                 sin->sin_len = sizeof(*sin);
 294                                 sin->sin_family = AF_INET;
 295                                 sin->sin_port = IPSEC_PORT_ANY;
 296                                 sin->sin_addr = ip->ip_dst;
 297                         }
 298                 } else {
 299                         struct sockaddr_in6 *sin6;
 300                         struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
 301
 302                         if (saidx->src.sin6.sin6_len == 0) {
 303                                 sin6 = (struct sockaddr_in6 *)&saidx->src;
 304                                 sin6->sin6_len = sizeof(*sin6);
 305                                 sin6->sin6_family = AF_INET6;
 306                                 sin6->sin6_port = IPSEC_PORT_ANY;
 307                                 sin6->sin6_addr = ip6->ip6_src;
 308                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
 309                                         /* fix scope id for comparing SPD */
 310                                         sin6->sin6_addr.s6_addr16[1] = 0;
 311                                         sin6->sin6_scope_id =
 312                                             ntohs(ip6->ip6_src.s6_addr16[1]);
 313                                 }
 314                         }
 315                         if (saidx->dst.sin6.sin6_len == 0) {
 316                                 sin6 = (struct sockaddr_in6 *)&saidx->dst;
 317                                 sin6->sin6_len = sizeof(*sin6);
 318                                 sin6->sin6_family = AF_INET6;
 319                                 sin6->sin6_port = IPSEC_PORT_ANY;
 320                                 sin6->sin6_addr = ip6->ip6_dst;
 321                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
 322                                         /* fix scope id for comparing SPD */
 323                                         sin6->sin6_addr.s6_addr16[1] = 0;
 324                                         sin6->sin6_scope_id =
 325                                             ntohs(ip6->ip6_dst.s6_addr16[1]);
 326                                 }
 327                         }
 328                 }
 329         }
 330
 331         /*
 332          * Lookup SA and validate it.
 333          */
 334         *error = key_checkrequest(isr, saidx);
 335         if (*error != 0) {
 336                 /*
 337                  * IPsec processing is required, but no SA found.
 338                  * I assume that key_acquire() had been called
 339                  * to get/establish the SA. Here I discard
 340                  * this packet because it is responsibility for
 341                  * upper layer to retransmit the packet.
 342                  */
 343                 V_ipsec4stat.ips_out_nosa++;
 344                 goto bad;
 345         }
 346         sav = isr->sav;
 347         if (sav == NULL) {
 348                 IPSEC_ASSERT(ipsec_get_reqlevel(isr) == IPSEC_LEVEL_USE,
 349                         ("no SA found, but required; level %u",
 350                         ipsec_get_reqlevel(isr)));
 351                 IPSECREQUEST_UNLOCK(isr);
 352                 isr = isr->next;
 353                 /*
 354                  * If isr is NULL, we found a 'use' policy w/o SA.
 355                  * Return w/o error and w/o isr so we can drop out
 356                  * and continue w/o IPsec processing.
 357                  */
 358                 if (isr == NULL)
 359                         return isr;
 360                 IPSECREQUEST_LOCK(isr);
 361                 goto again;
 362         }
 363
 364         /*
 365          * Check system global policy controls.
 366          */
 367         if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
 368             (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
 369             (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
 370                 DPRINTF(("%s: IPsec outbound packet dropped due"
 371                         " to policy (check your sysctls)\n", __func__));
 372                 IPSEC_OSTAT(V_espstat.esps_pdrops, V_ahstat.ahs_pdrops,
 373                     V_ipcompstat.ipcomps_pdrops);
 374                 *error = EHOSTUNREACH;
 375                 goto bad;
 376         }
 377
 378         /*
 379          * Sanity check the SA contents for the caller
 380          * before they invoke the xform output method.
 381          */
 382         if (sav->tdb_xform == NULL) {
 383                 DPRINTF(("%s: no transform for SA\n", __func__));
 384                 IPSEC_OSTAT(V_espstat.esps_noxform, V_ahstat.ahs_noxform,
 385                     V_ipcompstat.ipcomps_noxform);
 386                 *error = EHOSTUNREACH;
 387                 goto bad;
 388         }
 389         return isr;
 390 bad:
 391         IPSEC_ASSERT(*error != 0, ("error return w/ no error code"));
 392         IPSECREQUEST_UNLOCK(isr);
 393         return NULL;
 394 #undef IPSEC_OSTAT
 395 }
 396
 397 #ifdef INET
 398 /*
 399  * IPsec output logic for IPv4.
 400  */
 401 int
 402 ipsec4_process_packet(
 403         struct mbuf *m,
 404         struct ipsecrequest *isr,
 405         int flags,
 406         int tunalready)
 407 {
 408         struct secasindex saidx;
 409         struct secasvar *sav;
 410         struct ip *ip;
 411         int error, i, off;
 412
 413         IPSEC_ASSERT(m != NULL, ("null mbuf"));
 414         IPSEC_ASSERT(isr != NULL, ("null isr"));
 415
 416         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 417
 418         isr = ipsec_nextisr(m, isr, AF_INET, &saidx, &error);
 419         if (isr == NULL) {
 420                 if (error != 0)
 421                         goto bad;
 422                 return EJUSTRETURN;
 423         }
 424
 425         sav = isr->sav;
 426
 427 #ifdef DEV_ENC
 428         encif->if_opackets++;
 429         encif->if_obytes += m->m_pkthdr.len;
 430
 431         /* pass the mbuf to enc0 for bpf processing */
 432         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_BEFORE);
 433         /* pass the mbuf to enc0 for packet filtering */
 434         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 435                 goto bad;
 436 #endif
 437
 438         if (!tunalready) {
 439                 union sockaddr_union *dst = &sav->sah->saidx.dst;
 440                 int setdf;
 441
 442                 /*
 443                  * Collect IP_DF state from the outer header.
 444                  */
 445                 if (dst->sa.sa_family == AF_INET) {
 446                         if (m->m_len < sizeof (struct ip) &&
 447                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 448                                 error = ENOBUFS;
 449                                 goto bad;
 450                         }
 451                         ip = mtod(m, struct ip *);
 452                         /* Honor system-wide control of how to handle IP_DF */
 453                         switch (V_ip4_ipsec_dfbit) {
 454                         case 0:                 /* clear in outer header */
 455                         case 1:                 /* set in outer header */
 456                                 setdf = V_ip4_ipsec_dfbit;
 457                                 break;
 458                         default:                /* propagate to outer header */
 459                                 setdf = ntohs(ip->ip_off & IP_DF);
 460                                 break;
 461                         }
 462                 } else {
 463                         ip = NULL;              /* keep compiler happy */
 464                         setdf = 0;
 465                 }
 466                 /* Do the appropriate encapsulation, if necessary */
 467                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
 468                     dst->sa.sa_family != AF_INET ||         /* PF mismatch */
 469 #if 0
 470                     (sav->flags & SADB_X_SAFLAGS_TUNNEL) || /* Tunnel requ'd */
 471                     sav->tdb_xform->xf_type == XF_IP4 ||    /* ditto */
 472 #endif
 473                     (dst->sa.sa_family == AF_INET &&        /* Proxy */
 474                      dst->sin.sin_addr.s_addr != INADDR_ANY &&
 475                      dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
 476                         struct mbuf *mp;
 477
 478                         /* Fix IPv4 header checksum and length */
 479                         if (m->m_len < sizeof (struct ip) &&
 480                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 481                                 error = ENOBUFS;
 482                                 goto bad;
 483                         }
 484                         ip = mtod(m, struct ip *);
 485                         ip->ip_len = htons(m->m_pkthdr.len);
 486                         ip->ip_sum = 0;
 487 #ifdef _IP_VHL
 488                         if (ip->ip_vhl == IP_VHL_BORING)
 489                                 ip->ip_sum = in_cksum_hdr(ip);
 490                         else
 491                                 ip->ip_sum = in_cksum(m,
 492                                         _IP_VHL_HL(ip->ip_vhl) << 2);
 493 #else
 494                         ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
 495 #endif
 496
 497                         /* Encapsulate the packet */
 498                         error = ipip_output(m, isr, &mp, 0, 0);
 499                         if (mp == NULL && !error) {
 500                                 /* Should never happen. */
 501                                 DPRINTF(("%s: ipip_output returns no mbuf and "
 502                                         "no error!", __func__));
 503                                 error = EFAULT;
 504                         }
 505                         if (error) {
 506                                 if (mp) {
 507                                         /* XXX: Should never happen! */
 508                                         m_freem(mp);
 509                                 }
 510                                 m = NULL; /* ipip_output() already freed it */
 511                                 goto bad;
 512                         }
 513                         m = mp, mp = NULL;
 514                         /*
 515                          * ipip_output clears IP_DF in the new header.  If
 516                          * we need to propagate IP_DF from the outer header,
 517                          * then we have to do it here.
 518                          *
 519                          * XXX shouldn't assume what ipip_output does.
 520                          */
 521                         if (dst->sa.sa_family == AF_INET && setdf) {
 522                                 if (m->m_len < sizeof (struct ip) &&
 523                                     (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 524                                         error = ENOBUFS;
 525                                         goto bad;
 526                                 }
 527                                 ip = mtod(m, struct ip *);
 528                                 ip->ip_off = ntohs(ip->ip_off);
 529                                 ip->ip_off |= IP_DF;
 530                                 ip->ip_off = htons(ip->ip_off);
 531                         }
 532                 }
 533         }
 534
 535 #ifdef DEV_ENC
 536         /* pass the mbuf to enc0 for bpf processing */
 537         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_AFTER);
 538         /* pass the mbuf to enc0 for packet filtering */
 539         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 540                 goto bad;
 541 #endif
 542
 543         /*
 544          * Dispatch to the appropriate IPsec transform logic.  The
 545          * packet will be returned for transmission after crypto
 546          * processing, etc. are completed.  For encapsulation we
 547          * bypass this call because of the explicit call done above
 548          * (necessary to deal with IP_DF handling for IPv4).
 549          *
 550          * NB: m & sav are ``passed to caller'' who's reponsible for
 551          *     for reclaiming their resources.
 552          */
 553         if (sav->tdb_xform->xf_type != XF_IP4) {
 554                 ip = mtod(m, struct ip *);
 555                 i = ip->ip_hl << 2;
 556                 off = offsetof(struct ip, ip_p);
 557                 error = (*sav->tdb_xform->xf_output)(m, isr, NULL, i, off);
 558         } else {
 559                 error = ipsec_process_done(m, isr);
 560         }
 561         IPSECREQUEST_UNLOCK(isr);
 562         return error;
 563 bad:
 564         if (isr)
 565                 IPSECREQUEST_UNLOCK(isr);
 566         if (m)
 567                 m_freem(m);
 568         return error;
 569 }
 570 #endif
 571
 572 #ifdef INET6
 573 /*
 574  * Chop IP6 header from the payload.
 575  */
 576 static struct mbuf *
 577 ipsec6_splithdr(struct mbuf *m)
 578 {
 579         struct mbuf *mh;
 580         struct ip6_hdr *ip6;
 581         int hlen;
 582
 583         IPSEC_ASSERT(m->m_len >= sizeof (struct ip6_hdr),
 584                 ("first mbuf too short, len %u", m->m_len));
 585         ip6 = mtod(m, struct ip6_hdr *);
 586         hlen = sizeof(struct ip6_hdr);
 587         if (m->m_len > hlen) {
 588                 MGETHDR(mh, M_DONTWAIT, MT_DATA);
 589                 if (!mh) {
 590                         m_freem(m);
 591                         return NULL;
 592                 }
 593                 M_MOVE_PKTHDR(mh, m);
 594                 MH_ALIGN(mh, hlen);
 595                 m->m_len -= hlen;
 596                 m->m_data += hlen;
 597                 mh->m_next = m;
 598                 m = mh;
 599                 m->m_len = hlen;
 600                 bcopy((caddr_t)ip6, mtod(m, caddr_t), hlen);
 601         } else if (m->m_len < hlen) {
 602                 m = m_pullup(m, hlen);
 603                 if (!m)
 604                         return NULL;
 605         }
 606         return m;
 607 }
 608
 609 /*
 610  * IPsec output logic for IPv6, transport mode.
 611  */
 612 int
 613 ipsec6_output_trans(
 614         struct ipsec_output_state *state,
 615         u_char *nexthdrp,
 616         struct mbuf *mprev,
 617         struct secpolicy *sp,
 618         int flags,
 619         int *tun)
 620 {
 621         struct ipsecrequest *isr;
 622         struct secasindex saidx;
 623         int error = 0;
 624         struct mbuf *m;
 625
 626         IPSEC_ASSERT(state != NULL, ("null state"));
 627         IPSEC_ASSERT(state->m != NULL, ("null m"));
 628         IPSEC_ASSERT(nexthdrp != NULL, ("null nexthdrp"));
 629         IPSEC_ASSERT(mprev != NULL, ("null mprev"));
 630         IPSEC_ASSERT(sp != NULL, ("null sp"));
 631         IPSEC_ASSERT(tun != NULL, ("null tun"));
 632
 633         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 634                 printf("%s: applied SP\n", __func__);
 635                 kdebug_secpolicy(sp));
 636
 637         isr = sp->req;
 638         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 639                 /* the rest will be handled by ipsec6_output_tunnel() */
 640                 *tun = 1;               /* need tunnel-mode processing */
 641                 return 0;
 642         }
 643
 644         *tun = 0;
 645         m = state->m;
 646
 647         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 648         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 649         if (isr == NULL) {
 650                 if (error != 0) {
 651 #ifdef notdef
 652                         /* XXX should notification be done for all errors ? */
 653                         /*
 654                          * Notify the fact that the packet is discarded
 655                          * to ourselves. I believe this is better than
 656                          * just silently discarding. (jinmei@kame.net)
 657                          * XXX: should we restrict the error to TCP packets?
 658                          * XXX: should we directly notify sockets via
 659                          *      pfctlinputs?
 660                          */
 661                         icmp6_error(m, ICMP6_DST_UNREACH,
 662                                     ICMP6_DST_UNREACH_ADMIN, 0);
 663                         m = NULL;       /* NB: icmp6_error frees mbuf */
 664 #endif
 665                         goto bad;
 666                 }
 667                 return EJUSTRETURN;
 668         }
 669
 670         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 671                                                   sizeof (struct ip6_hdr),
 672                                                   offsetof(struct ip6_hdr,
 673                                                            ip6_nxt));
 674         IPSECREQUEST_UNLOCK(isr);
 675         return error;
 676 bad:
 677         if (isr)
 678                 IPSECREQUEST_UNLOCK(isr);
 679         if (m)
 680                 m_freem(m);
 681         state->m = NULL;
 682         return error;
 683 }
 684
 685 static int
 686 ipsec6_encapsulate(struct mbuf *m, struct secasvar *sav)
 687 {
 688         struct ip6_hdr *oip6;
 689         struct ip6_hdr *ip6;
 690         size_t plen;
 691
 692         /* can't tunnel between different AFs */
 693         if (sav->sah->saidx.src.sa.sa_family != AF_INET6 ||
 694             sav->sah->saidx.dst.sa.sa_family != AF_INET6) {
 695                 m_freem(m);
 696                 return EINVAL;
 697         }
 698         IPSEC_ASSERT(m->m_len == sizeof (struct ip6_hdr),
 699                 ("mbuf wrong size; len %u", m->m_len));
 700
 701
 702         /*
 703          * grow the mbuf to accomodate the new IPv6 header.
 704          */
 705         plen = m->m_pkthdr.len;
 706         if (M_LEADINGSPACE(m->m_next) < sizeof(struct ip6_hdr)) {
 707                 struct mbuf *n;
 708                 MGET(n, M_DONTWAIT, MT_DATA);
 709                 if (!n) {
 710                         m_freem(m);
 711                         return ENOBUFS;
 712                 }
 713                 n->m_len = sizeof(struct ip6_hdr);
 714                 n->m_next = m->m_next;
 715                 m->m_next = n;
 716                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 717                 oip6 = mtod(n, struct ip6_hdr *);
 718         } else {
 719                 m->m_next->m_len += sizeof(struct ip6_hdr);
 720                 m->m_next->m_data -= sizeof(struct ip6_hdr);
 721                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 722                 oip6 = mtod(m->m_next, struct ip6_hdr *);
 723         }
 724         ip6 = mtod(m, struct ip6_hdr *);
 725         bcopy((caddr_t)ip6, (caddr_t)oip6, sizeof(struct ip6_hdr));
 726
 727         /* Fake link-local scope-class addresses */
 728         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_src))
 729                 oip6->ip6_src.s6_addr16[1] = 0;
 730         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_dst))
 731                 oip6->ip6_dst.s6_addr16[1] = 0;
 732
 733         /* construct new IPv6 header. see RFC 2401 5.1.2.2 */
 734         /* ECN consideration. */
 735         ip6_ecn_ingress(V_ip6_ipsec_ecn, &ip6->ip6_flow, &oip6->ip6_flow);
 736         if (plen < IPV6_MAXPACKET - sizeof(struct ip6_hdr))
 737                 ip6->ip6_plen = htons(plen);
 738         else {
 739                 /* ip6->ip6_plen will be updated in ip6_output() */
 740         }
 741         ip6->ip6_nxt = IPPROTO_IPV6;
 742         ip6->ip6_src = sav->sah->saidx.src.sin6.sin6_addr;
 743         ip6->ip6_dst = sav->sah->saidx.dst.sin6.sin6_addr;
 744         ip6->ip6_hlim = IPV6_DEFHLIM;
 745
 746         /* XXX Should ip6_src be updated later ? */
 747
 748         return 0;
 749 }
 750
 751 /*
 752  * IPsec output logic for IPv6, tunnel mode.
 753  */
 754 int
 755 ipsec6_output_tunnel(struct ipsec_output_state *state, struct secpolicy *sp, int flags)
 756 {
 757         struct ip6_hdr *ip6;
 758         struct ipsecrequest *isr;
 759         struct secasindex saidx;
 760         int error;
 761         struct sockaddr_in6* dst6;
 762         struct mbuf *m;
 763
 764         IPSEC_ASSERT(state != NULL, ("null state"));
 765         IPSEC_ASSERT(state->m != NULL, ("null m"));
 766         IPSEC_ASSERT(sp != NULL, ("null sp"));
 767
 768         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 769                 printf("%s: applied SP\n", __func__);
 770                 kdebug_secpolicy(sp));
 771
 772         m = state->m;
 773         /*
 774          * transport mode ipsec (before the 1st tunnel mode) is already
 775          * processed by ipsec6_output_trans().
 776          */
 777         for (isr = sp->req; isr; isr = isr->next) {
 778                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
 779                         break;
 780         }
 781
 782         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 783         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 784         if (isr == NULL) {
 785                 if (error != 0)
 786                         goto bad;
 787                 return EJUSTRETURN;
 788         }
 789
 790 #ifdef DEV_ENC
 791         encif->if_opackets++;
 792         encif->if_obytes += m->m_pkthdr.len;
 793
 794         /* pass the mbuf to enc0 for bpf processing */
 795         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_BEFORE);
 796         /* pass the mbuf to enc0 for packet filtering */
 797         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 798                 goto bad;
 799 #endif
 800
 801         /*
 802          * There may be the case that SA status will be changed when
 803          * we are refering to one. So calling splsoftnet().
 804          */
 805         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 806                 /*
 807                  * build IPsec tunnel.
 808                  */
 809                 /* XXX should be processed with other familiy */
 810                 if (isr->sav->sah->saidx.src.sa.sa_family != AF_INET6) {
 811                         ipseclog((LOG_ERR, "%s: family mismatched between "
 812                             "inner and outer, spi=%u\n", __func__,
 813                             ntohl(isr->sav->spi)));
 814                         V_ipsec6stat.ips_out_inval++;
 815                         error = EAFNOSUPPORT;
 816                         goto bad;
 817                 }
 818
 819                 m = ipsec6_splithdr(m);
 820                 if (!m) {
 821                         V_ipsec6stat.ips_out_nomem++;
 822                         error = ENOMEM;
 823                         goto bad;
 824                 }
 825                 error = ipsec6_encapsulate(m, isr->sav);
 826                 if (error) {
 827                         m = NULL;
 828                         goto bad;
 829                 }
 830                 ip6 = mtod(m, struct ip6_hdr *);
 831
 832                 state->ro = &isr->sav->sah->sa_route;
 833                 state->dst = (struct sockaddr *)&state->ro->ro_dst;
 834                 dst6 = (struct sockaddr_in6 *)state->dst;
 835                 if (state->ro->ro_rt
 836                  && ((state->ro->ro_rt->rt_flags & RTF_UP) == 0
 837                   || !IN6_ARE_ADDR_EQUAL(&dst6->sin6_addr, &ip6->ip6_dst))) {
 838                         RTFREE(state->ro->ro_rt);
 839                         state->ro->ro_rt = NULL;
 840                 }
 841                 if (state->ro->ro_rt == NULL) {
 842                         bzero(dst6, sizeof(*dst6));
 843                         dst6->sin6_family = AF_INET6;
 844                         dst6->sin6_len = sizeof(*dst6);
 845                         dst6->sin6_addr = ip6->ip6_dst;
 846                         rtalloc(state->ro);
 847                 }
 848                 if (state->ro->ro_rt == NULL) {
 849                         V_ip6stat.ip6s_noroute++;
 850                         V_ipsec6stat.ips_out_noroute++;
 851                         error = EHOSTUNREACH;
 852                         goto bad;
 853                 }
 854
 855                 /* adjust state->dst if tunnel endpoint is offlink */
 856                 if (state->ro->ro_rt->rt_flags & RTF_GATEWAY) {
 857                         state->dst = (struct sockaddr *)state->ro->ro_rt->rt_gateway;
 858                         dst6 = (struct sockaddr_in6 *)state->dst;
 859                 }
 860         }
 861
 862         m = ipsec6_splithdr(m);
 863         if (!m) {
 864                 V_ipsec6stat.ips_out_nomem++;
 865                 error = ENOMEM;
 866                 goto bad;
 867         }
 868         ip6 = mtod(m, struct ip6_hdr *);
 869
 870 #ifdef DEV_ENC
 871         /* pass the mbuf to enc0 for bpf processing */
 872         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_AFTER);
 873         /* pass the mbuf to enc0 for packet filtering */
 874         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 875                 goto bad;
 876 #endif
 877
 878         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 879                 sizeof (struct ip6_hdr),
 880                 offsetof(struct ip6_hdr, ip6_nxt));
 881         IPSECREQUEST_UNLOCK(isr);
 882         return error;
 883 bad:
 884         if (isr)
 885                 IPSECREQUEST_UNLOCK(isr);
 886         if (m)
 887                 m_freem(m);
 888         state->m = NULL;
 889         return error;
 890 }
 891 #endif /*INET6*/