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
  50 #include <netinet/in.h>
  51 #include <netinet/in_systm.h>
  52 #include <netinet/ip.h>
  53 #include <netinet/ip_var.h>
  54 #include <netinet/in_var.h>
  55 #include <netinet/ip_ecn.h>
  56 #ifdef INET6
  57 #include <netinet6/ip6_ecn.h>
  58 #endif
  59
  60 #include <netinet/ip6.h>
  61 #ifdef INET6
  62 #include <netinet6/ip6_var.h>
  63 #endif
  64 #include <netinet/in_pcb.h>
  65 #ifdef INET6
  66 #include <netinet/icmp6.h>
  67 #endif
  68
  69 #include <netipsec/ipsec.h>
  70 #ifdef INET6
  71 #include <netipsec/ipsec6.h>
  72 #endif
  73 #include <netipsec/ah_var.h>
  74 #include <netipsec/esp_var.h>
  75 #include <netipsec/ipcomp_var.h>
  76
  77 #include <netipsec/xform.h>
  78
  79 #include <netipsec/key.h>
  80 #include <netipsec/keydb.h>
  81 #include <netipsec/key_debug.h>
  82
  83 #include <machine/in_cksum.h>
  84
  85 int
  86 ipsec_process_done(struct mbuf *m, struct ipsecrequest *isr)
  87 {
  88         struct tdb_ident *tdbi;
  89         struct m_tag *mtag;
  90         struct secasvar *sav;
  91         struct secasindex *saidx;
  92         int error;
  93
  94         IPSEC_SPLASSERT_SOFTNET(__func__);
  95
  96         IPSEC_ASSERT(m != NULL, ("null mbuf"));
  97         IPSEC_ASSERT(isr != NULL, ("null ISR"));
  98         sav = isr->sav;
  99         IPSEC_ASSERT(sav != NULL, ("null SA"));
 100         IPSEC_ASSERT(sav->sah != NULL, ("null SAH"));
 101
 102         saidx = &sav->sah->saidx;
 103         switch (saidx->dst.sa.sa_family) {
 104 #ifdef INET
 105         case AF_INET:
 106                 /* Fix the header length, for AH processing. */
 107                 mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
 108                 break;
 109 #endif /* INET */
 110 #ifdef INET6
 111         case AF_INET6:
 112                 /* Fix the header length, for AH processing. */
 113                 if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
 114                         error = ENXIO;
 115                         goto bad;
 116                 }
 117                 if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
 118                         /* No jumbogram support. */
 119                         error = ENXIO;  /*?*/
 120                         goto bad;
 121                 }
 122                 mtod(m, struct ip6_hdr *)->ip6_plen =
 123                         htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
 124                 break;
 125 #endif /* INET6 */
 126         default:
 127                 DPRINTF(("%s: unknown protocol family %u\n", __func__,
 128                     saidx->dst.sa.sa_family));
 129                 error = ENXIO;
 130                 goto bad;
 131         }
 132
 133         /*
 134          * Add a record of what we've done or what needs to be done to the
 135          * packet.
 136          */
 137         mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE,
 138                         sizeof(struct tdb_ident), M_NOWAIT);
 139         if (mtag == NULL) {
 140                 DPRINTF(("%s: could not get packet tag\n", __func__));
 141                 error = ENOMEM;
 142                 goto bad;
 143         }
 144
 145         tdbi = (struct tdb_ident *)(mtag + 1);
 146         tdbi->dst = saidx->dst;
 147         tdbi->proto = saidx->proto;
 148         tdbi->spi = sav->spi;
 149         m_tag_prepend(m, mtag);
 150
 151         /*
 152          * If there's another (bundled) SA to apply, do so.
 153          * Note that this puts a burden on the kernel stack size.
 154          * If this is a problem we'll need to introduce a queue
 155          * to set the packet on so we can unwind the stack before
 156          * doing further processing.
 157          */
 158         if (isr->next) {
 159                 ipsec4stat.ips_out_bundlesa++;
 160                 return ipsec4_process_packet(m, isr->next, 0, 0);
 161         }
 162         key_sa_recordxfer(sav, m);              /* record data transfer */
 163
 164         /*
 165          * We're done with IPsec processing, transmit the packet using the
 166          * appropriate network protocol (IP or IPv6). SPD lookup will be
 167          * performed again there.
 168          */
 169         switch (saidx->dst.sa.sa_family) {
 170 #ifdef INET
 171         struct ip *ip;
 172         case AF_INET:
 173                 ip = mtod(m, struct ip *);
 174                 ip->ip_len = ntohs(ip->ip_len);
 175                 ip->ip_off = ntohs(ip->ip_off);
 176
 177                 return ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
 178 #endif /* INET */
 179 #ifdef INET6
 180         case AF_INET6:
 181                 /*
 182                  * We don't need massage, IPv6 header fields are always in
 183                  * net endian.
 184                  */
 185                 return ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
 186 #endif /* INET6 */
 187         }
 188         panic("ipsec_process_done");
 189 bad:
 190         m_freem(m);
 191         KEY_FREESAV(&sav);
 192         return (error);
 193 }
 194
 195 static struct ipsecrequest *
 196 ipsec_nextisr(
 197         struct mbuf *m,
 198         struct ipsecrequest *isr,
 199         int af,
 200         struct secasindex *saidx,
 201         int *error
 202 )
 203 {
 204 #define IPSEC_OSTAT(x,y,z) (isr->saidx.proto == IPPROTO_ESP ? (x)++ : \
 205                             isr->saidx.proto == IPPROTO_AH ? (y)++ : (z)++)
 206         struct secasvar *sav;
 207
 208         IPSEC_SPLASSERT_SOFTNET(__func__);
 209         IPSECREQUEST_LOCK_ASSERT(isr);
 210
 211         IPSEC_ASSERT(af == AF_INET || af == AF_INET6,
 212                 ("invalid address family %u", af));
 213 again:
 214         /*
 215          * Craft SA index to search for proper SA.  Note that
 216          * we only fillin unspecified SA peers for transport
 217          * mode; for tunnel mode they must already be filled in.
 218          */
 219         *saidx = isr->saidx;
 220         if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
 221                 /* Fillin unspecified SA peers only for transport mode */
 222                 if (af == AF_INET) {
 223                         struct sockaddr_in *sin;
 224                         struct ip *ip = mtod(m, struct ip *);
 225
 226                         if (saidx->src.sa.sa_len == 0) {
 227                                 sin = &saidx->src.sin;
 228                                 sin->sin_len = sizeof(*sin);
 229                                 sin->sin_family = AF_INET;
 230                                 sin->sin_port = IPSEC_PORT_ANY;
 231                                 sin->sin_addr = ip->ip_src;
 232                         }
 233                         if (saidx->dst.sa.sa_len == 0) {
 234                                 sin = &saidx->dst.sin;
 235                                 sin->sin_len = sizeof(*sin);
 236                                 sin->sin_family = AF_INET;
 237                                 sin->sin_port = IPSEC_PORT_ANY;
 238                                 sin->sin_addr = ip->ip_dst;
 239                         }
 240                 } else {
 241                         struct sockaddr_in6 *sin6;
 242                         struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
 243
 244                         if (saidx->src.sin6.sin6_len == 0) {
 245                                 sin6 = (struct sockaddr_in6 *)&saidx->src;
 246                                 sin6->sin6_len = sizeof(*sin6);
 247                                 sin6->sin6_family = AF_INET6;
 248                                 sin6->sin6_port = IPSEC_PORT_ANY;
 249                                 sin6->sin6_addr = ip6->ip6_src;
 250                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
 251                                         /* fix scope id for comparing SPD */
 252                                         sin6->sin6_addr.s6_addr16[1] = 0;
 253                                         sin6->sin6_scope_id =
 254                                             ntohs(ip6->ip6_src.s6_addr16[1]);
 255                                 }
 256                         }
 257                         if (saidx->dst.sin6.sin6_len == 0) {
 258                                 sin6 = (struct sockaddr_in6 *)&saidx->dst;
 259                                 sin6->sin6_len = sizeof(*sin6);
 260                                 sin6->sin6_family = AF_INET6;
 261                                 sin6->sin6_port = IPSEC_PORT_ANY;
 262                                 sin6->sin6_addr = ip6->ip6_dst;
 263                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
 264                                         /* fix scope id for comparing SPD */
 265                                         sin6->sin6_addr.s6_addr16[1] = 0;
 266                                         sin6->sin6_scope_id =
 267                                             ntohs(ip6->ip6_dst.s6_addr16[1]);
 268                                 }
 269                         }
 270                 }
 271         }
 272
 273         /*
 274          * Lookup SA and validate it.
 275          */
 276         *error = key_checkrequest(isr, saidx);
 277         if (*error != 0) {
 278                 /*
 279                  * IPsec processing is required, but no SA found.
 280                  * I assume that key_acquire() had been called
 281                  * to get/establish the SA. Here I discard
 282                  * this packet because it is responsibility for
 283                  * upper layer to retransmit the packet.
 284                  */
 285                 ipsec4stat.ips_out_nosa++;
 286                 goto bad;
 287         }
 288         sav = isr->sav;
 289         if (sav == NULL) {              /* XXX valid return */
 290                 IPSEC_ASSERT(ipsec_get_reqlevel(isr) == IPSEC_LEVEL_USE,
 291                         ("no SA found, but required; level %u",
 292                         ipsec_get_reqlevel(isr)));
 293                 IPSECREQUEST_UNLOCK(isr);
 294                 isr = isr->next;
 295                 if (isr == NULL) {
 296                         /*XXXstatistic??*/
 297                         *error = EINVAL;                /*XXX*/
 298                         return isr;
 299                 }
 300                 IPSECREQUEST_LOCK(isr);
 301                 goto again;
 302         }
 303
 304         /*
 305          * Check system global policy controls.
 306          */
 307         if ((isr->saidx.proto == IPPROTO_ESP && !esp_enable) ||
 308             (isr->saidx.proto == IPPROTO_AH && !ah_enable) ||
 309             (isr->saidx.proto == IPPROTO_IPCOMP && !ipcomp_enable)) {
 310                 DPRINTF(("%s: IPsec outbound packet dropped due"
 311                         " to policy (check your sysctls)\n", __func__));
 312                 IPSEC_OSTAT(espstat.esps_pdrops, ahstat.ahs_pdrops,
 313                     ipcompstat.ipcomps_pdrops);
 314                 *error = EHOSTUNREACH;
 315                 goto bad;
 316         }
 317
 318         /*
 319          * Sanity check the SA contents for the caller
 320          * before they invoke the xform output method.
 321          */
 322         if (sav->tdb_xform == NULL) {
 323                 DPRINTF(("%s: no transform for SA\n", __func__));
 324                 IPSEC_OSTAT(espstat.esps_noxform, ahstat.ahs_noxform,
 325                     ipcompstat.ipcomps_noxform);
 326                 *error = EHOSTUNREACH;
 327                 goto bad;
 328         }
 329         return isr;
 330 bad:
 331         IPSEC_ASSERT(*error != 0, ("error return w/ no error code"));
 332         IPSECREQUEST_UNLOCK(isr);
 333         return NULL;
 334 #undef IPSEC_OSTAT
 335 }
 336
 337 #ifdef INET
 338 /*
 339  * IPsec output logic for IPv4.
 340  */
 341 int
 342 ipsec4_process_packet(
 343         struct mbuf *m,
 344         struct ipsecrequest *isr,
 345         int flags,
 346         int tunalready)
 347 {
 348         struct secasindex saidx;
 349         struct secasvar *sav;
 350         struct ip *ip;
 351         int error, i, off;
 352
 353         IPSEC_ASSERT(m != NULL, ("null mbuf"));
 354         IPSEC_ASSERT(isr != NULL, ("null isr"));
 355
 356         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 357
 358         isr = ipsec_nextisr(m, isr, AF_INET, &saidx, &error);
 359         if (isr == NULL)
 360                 goto bad;
 361
 362         sav = isr->sav;
 363
 364 #ifdef DEV_ENC
 365         /* pass the mbuf to enc0 for bpf processing */
 366         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_BEFORE);
 367         /* pass the mbuf to enc0 for packet filtering */
 368         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 369                 goto bad;
 370 #endif
 371
 372         if (!tunalready) {
 373                 union sockaddr_union *dst = &sav->sah->saidx.dst;
 374                 int setdf;
 375
 376                 /*
 377                  * Collect IP_DF state from the outer header.
 378                  */
 379                 if (dst->sa.sa_family == AF_INET) {
 380                         if (m->m_len < sizeof (struct ip) &&
 381                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 382                                 error = ENOBUFS;
 383                                 goto bad;
 384                         }
 385                         ip = mtod(m, struct ip *);
 386                         /* Honor system-wide control of how to handle IP_DF */
 387                         switch (ip4_ipsec_dfbit) {
 388                         case 0:                 /* clear in outer header */
 389                         case 1:                 /* set in outer header */
 390                                 setdf = ip4_ipsec_dfbit;
 391                                 break;
 392                         default:                /* propagate to outer header */
 393                                 setdf = ntohs(ip->ip_off & IP_DF);
 394                                 break;
 395                         }
 396                 } else {
 397                         ip = NULL;              /* keep compiler happy */
 398                         setdf = 0;
 399                 }
 400                 /* Do the appropriate encapsulation, if necessary */
 401                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
 402                     dst->sa.sa_family != AF_INET ||         /* PF mismatch */
 403 #if 0
 404                     (sav->flags & SADB_X_SAFLAGS_TUNNEL) || /* Tunnel requ'd */
 405                     sav->tdb_xform->xf_type == XF_IP4 ||    /* ditto */
 406 #endif
 407                     (dst->sa.sa_family == AF_INET &&        /* Proxy */
 408                      dst->sin.sin_addr.s_addr != INADDR_ANY &&
 409                      dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
 410                         struct mbuf *mp;
 411
 412                         /* Fix IPv4 header checksum and length */
 413                         if (m->m_len < sizeof (struct ip) &&
 414                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 415                                 error = ENOBUFS;
 416                                 goto bad;
 417                         }
 418                         ip = mtod(m, struct ip *);
 419                         ip->ip_len = htons(m->m_pkthdr.len);
 420                         ip->ip_sum = 0;
 421 #ifdef _IP_VHL
 422                         if (ip->ip_vhl == IP_VHL_BORING)
 423                                 ip->ip_sum = in_cksum_hdr(ip);
 424                         else
 425                                 ip->ip_sum = in_cksum(m,
 426                                         _IP_VHL_HL(ip->ip_vhl) << 2);
 427 #else
 428                         ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
 429 #endif
 430
 431                         /* Encapsulate the packet */
 432                         error = ipip_output(m, isr, &mp, 0, 0);
 433                         if (mp == NULL && !error) {
 434                                 /* Should never happen. */
 435                                 DPRINTF(("%s: ipip_output returns no mbuf and "
 436                                         "no error!", __func__));
 437                                 error = EFAULT;
 438                         }
 439                         if (error) {
 440                                 if (mp) {
 441                                         /* XXX: Should never happen! */
 442                                         m_freem(mp);
 443                                 }
 444                                 m = NULL; /* ipip_output() already freed it */
 445                                 goto bad;
 446                         }
 447                         m = mp, mp = NULL;
 448                         /*
 449                          * ipip_output clears IP_DF in the new header.  If
 450                          * we need to propagate IP_DF from the outer header,
 451                          * then we have to do it here.
 452                          *
 453                          * XXX shouldn't assume what ipip_output does.
 454                          */
 455                         if (dst->sa.sa_family == AF_INET && setdf) {
 456                                 if (m->m_len < sizeof (struct ip) &&
 457                                     (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 458                                         error = ENOBUFS;
 459                                         goto bad;
 460                                 }
 461                                 ip = mtod(m, struct ip *);
 462                                 ip->ip_off = ntohs(ip->ip_off);
 463                                 ip->ip_off |= IP_DF;
 464                                 ip->ip_off = htons(ip->ip_off);
 465                         }
 466                 }
 467         }
 468
 469 #ifdef DEV_ENC
 470         /* pass the mbuf to enc0 for bpf processing */
 471         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_AFTER);
 472         /* pass the mbuf to enc0 for packet filtering */
 473         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 474                 goto bad;
 475 #endif
 476
 477         /*
 478          * Dispatch to the appropriate IPsec transform logic.  The
 479          * packet will be returned for transmission after crypto
 480          * processing, etc. are completed.  For encapsulation we
 481          * bypass this call because of the explicit call done above
 482          * (necessary to deal with IP_DF handling for IPv4).
 483          *
 484          * NB: m & sav are ``passed to caller'' who's reponsible for
 485          *     for reclaiming their resources.
 486          */
 487         if (sav->tdb_xform->xf_type != XF_IP4) {
 488                 ip = mtod(m, struct ip *);
 489                 i = ip->ip_hl << 2;
 490                 off = offsetof(struct ip, ip_p);
 491                 error = (*sav->tdb_xform->xf_output)(m, isr, NULL, i, off);
 492         } else {
 493                 error = ipsec_process_done(m, isr);
 494         }
 495         IPSECREQUEST_UNLOCK(isr);
 496         return error;
 497 bad:
 498         if (isr)
 499                 IPSECREQUEST_UNLOCK(isr);
 500         if (m)
 501                 m_freem(m);
 502         return error;
 503 }
 504 #endif
 505
 506 #ifdef INET6
 507 /*
 508  * Chop IP6 header from the payload.
 509  */
 510 static struct mbuf *
 511 ipsec6_splithdr(struct mbuf *m)
 512 {
 513         struct mbuf *mh;
 514         struct ip6_hdr *ip6;
 515         int hlen;
 516
 517         IPSEC_ASSERT(m->m_len >= sizeof (struct ip6_hdr),
 518                 ("first mbuf too short, len %u", m->m_len));
 519         ip6 = mtod(m, struct ip6_hdr *);
 520         hlen = sizeof(struct ip6_hdr);
 521         if (m->m_len > hlen) {
 522                 MGETHDR(mh, M_DONTWAIT, MT_DATA);
 523                 if (!mh) {
 524                         m_freem(m);
 525                         return NULL;
 526                 }
 527                 M_MOVE_PKTHDR(mh, m);
 528                 MH_ALIGN(mh, hlen);
 529                 m->m_len -= hlen;
 530                 m->m_data += hlen;
 531                 mh->m_next = m;
 532                 m = mh;
 533                 m->m_len = hlen;
 534                 bcopy((caddr_t)ip6, mtod(m, caddr_t), hlen);
 535         } else if (m->m_len < hlen) {
 536                 m = m_pullup(m, hlen);
 537                 if (!m)
 538                         return NULL;
 539         }
 540         return m;
 541 }
 542
 543 /*
 544  * IPsec output logic for IPv6, transport mode.
 545  */
 546 int
 547 ipsec6_output_trans(
 548         struct ipsec_output_state *state,
 549         u_char *nexthdrp,
 550         struct mbuf *mprev,
 551         struct secpolicy *sp,
 552         int flags,
 553         int *tun)
 554 {
 555         struct ipsecrequest *isr;
 556         struct secasindex saidx;
 557         int error = 0;
 558         struct mbuf *m;
 559
 560         IPSEC_ASSERT(state != NULL, ("null state"));
 561         IPSEC_ASSERT(state->m != NULL, ("null m"));
 562         IPSEC_ASSERT(nexthdrp != NULL, ("null nexthdrp"));
 563         IPSEC_ASSERT(mprev != NULL, ("null mprev"));
 564         IPSEC_ASSERT(sp != NULL, ("null sp"));
 565         IPSEC_ASSERT(tun != NULL, ("null tun"));
 566
 567         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 568                 printf("%s: applied SP\n", __func__);
 569                 kdebug_secpolicy(sp));
 570
 571         isr = sp->req;
 572         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 573                 /* the rest will be handled by ipsec6_output_tunnel() */
 574                 *tun = 1;               /* need tunnel-mode processing */
 575                 return 0;
 576         }
 577
 578         *tun = 0;
 579         m = state->m;
 580
 581         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 582         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 583         if (isr == NULL) {
 584 #ifdef notdef
 585                 /* XXX should notification be done for all errors ? */
 586                 /*
 587                  * Notify the fact that the packet is discarded
 588                  * to ourselves. I believe this is better than
 589                  * just silently discarding. (jinmei@kame.net)
 590                  * XXX: should we restrict the error to TCP packets?
 591                  * XXX: should we directly notify sockets via
 592                  *      pfctlinputs?
 593                  */
 594                 icmp6_error(m, ICMP6_DST_UNREACH,
 595                             ICMP6_DST_UNREACH_ADMIN, 0);
 596                 m = NULL;       /* NB: icmp6_error frees mbuf */
 597 #endif
 598                 goto bad;
 599         }
 600
 601         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 602                                                   sizeof (struct ip6_hdr),
 603                                                   offsetof(struct ip6_hdr,
 604                                                            ip6_nxt));
 605         IPSECREQUEST_UNLOCK(isr);
 606         return error;
 607 bad:
 608         if (isr)
 609                 IPSECREQUEST_UNLOCK(isr);
 610         if (m)
 611                 m_freem(m);
 612         state->m = NULL;
 613         return error;
 614 }
 615
 616 static int
 617 ipsec6_encapsulate(struct mbuf *m, struct secasvar *sav)
 618 {
 619         struct ip6_hdr *oip6;
 620         struct ip6_hdr *ip6;
 621         size_t plen;
 622
 623         /* can't tunnel between different AFs */
 624         if (sav->sah->saidx.src.sa.sa_family != AF_INET6 ||
 625             sav->sah->saidx.dst.sa.sa_family != AF_INET6) {
 626                 m_freem(m);
 627                 return EINVAL;
 628         }
 629         IPSEC_ASSERT(m->m_len == sizeof (struct ip6_hdr),
 630                 ("mbuf wrong size; len %u", m->m_len));
 631
 632
 633         /*
 634          * grow the mbuf to accomodate the new IPv6 header.
 635          */
 636         plen = m->m_pkthdr.len;
 637         if (M_LEADINGSPACE(m->m_next) < sizeof(struct ip6_hdr)) {
 638                 struct mbuf *n;
 639                 MGET(n, M_DONTWAIT, MT_DATA);
 640                 if (!n) {
 641                         m_freem(m);
 642                         return ENOBUFS;
 643                 }
 644                 n->m_len = sizeof(struct ip6_hdr);
 645                 n->m_next = m->m_next;
 646                 m->m_next = n;
 647                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 648                 oip6 = mtod(n, struct ip6_hdr *);
 649         } else {
 650                 m->m_next->m_len += sizeof(struct ip6_hdr);
 651                 m->m_next->m_data -= sizeof(struct ip6_hdr);
 652                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 653                 oip6 = mtod(m->m_next, struct ip6_hdr *);
 654         }
 655         ip6 = mtod(m, struct ip6_hdr *);
 656         bcopy((caddr_t)ip6, (caddr_t)oip6, sizeof(struct ip6_hdr));
 657
 658         /* Fake link-local scope-class addresses */
 659         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_src))
 660                 oip6->ip6_src.s6_addr16[1] = 0;
 661         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_dst))
 662                 oip6->ip6_dst.s6_addr16[1] = 0;
 663
 664         /* construct new IPv6 header. see RFC 2401 5.1.2.2 */
 665         /* ECN consideration. */
 666         ip6_ecn_ingress(ip6_ipsec_ecn, &ip6->ip6_flow, &oip6->ip6_flow);
 667         if (plen < IPV6_MAXPACKET - sizeof(struct ip6_hdr))
 668                 ip6->ip6_plen = htons(plen);
 669         else {
 670                 /* ip6->ip6_plen will be updated in ip6_output() */
 671         }
 672         ip6->ip6_nxt = IPPROTO_IPV6;
 673         ip6->ip6_src = sav->sah->saidx.src.sin6.sin6_addr;
 674         ip6->ip6_dst = sav->sah->saidx.dst.sin6.sin6_addr;
 675         ip6->ip6_hlim = IPV6_DEFHLIM;
 676
 677         /* XXX Should ip6_src be updated later ? */
 678
 679         return 0;
 680 }
 681
 682 /*
 683  * IPsec output logic for IPv6, tunnel mode.
 684  */
 685 int
 686 ipsec6_output_tunnel(struct ipsec_output_state *state, struct secpolicy *sp, int flags)
 687 {
 688         struct ip6_hdr *ip6;
 689         struct ipsecrequest *isr;
 690         struct secasindex saidx;
 691         int error;
 692         struct sockaddr_in6* dst6;
 693         struct mbuf *m;
 694
 695         IPSEC_ASSERT(state != NULL, ("null state"));
 696         IPSEC_ASSERT(state->m != NULL, ("null m"));
 697         IPSEC_ASSERT(sp != NULL, ("null sp"));
 698
 699         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 700                 printf("%s: applied SP\n", __func__);
 701                 kdebug_secpolicy(sp));
 702
 703         m = state->m;
 704         /*
 705          * transport mode ipsec (before the 1st tunnel mode) is already
 706          * processed by ipsec6_output_trans().
 707          */
 708         for (isr = sp->req; isr; isr = isr->next) {
 709                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
 710                         break;
 711         }
 712
 713         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 714         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 715         if (isr == NULL)
 716                 goto bad;
 717
 718 #ifdef DEV_ENC
 719         /* pass the mbuf to enc0 for bpf processing */
 720         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_BEFORE);
 721         /* pass the mbuf to enc0 for packet filtering */
 722         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 723                 goto bad;
 724 #endif
 725
 726         /*
 727          * There may be the case that SA status will be changed when
 728          * we are refering to one. So calling splsoftnet().
 729          */
 730         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 731                 /*
 732                  * build IPsec tunnel.
 733                  */
 734                 /* XXX should be processed with other familiy */
 735                 if (isr->sav->sah->saidx.src.sa.sa_family != AF_INET6) {
 736                         ipseclog((LOG_ERR, "%s: family mismatched between "
 737                             "inner and outer, spi=%u\n", __func__,
 738                             ntohl(isr->sav->spi)));
 739                         ipsec6stat.ips_out_inval++;
 740                         error = EAFNOSUPPORT;
 741                         goto bad;
 742                 }
 743
 744                 m = ipsec6_splithdr(m);
 745                 if (!m) {
 746                         ipsec6stat.ips_out_nomem++;
 747                         error = ENOMEM;
 748                         goto bad;
 749                 }
 750                 error = ipsec6_encapsulate(m, isr->sav);
 751                 if (error) {
 752                         m = NULL;
 753                         goto bad;
 754                 }
 755                 ip6 = mtod(m, struct ip6_hdr *);
 756
 757                 state->ro = &isr->sav->sah->sa_route;
 758                 state->dst = (struct sockaddr *)&state->ro->ro_dst;
 759                 dst6 = (struct sockaddr_in6 *)state->dst;
 760                 if (state->ro->ro_rt
 761                  && ((state->ro->ro_rt->rt_flags & RTF_UP) == 0
 762                   || !IN6_ARE_ADDR_EQUAL(&dst6->sin6_addr, &ip6->ip6_dst))) {
 763                         RTFREE(state->ro->ro_rt);
 764                         state->ro->ro_rt = NULL;
 765                 }
 766                 if (state->ro->ro_rt == 0) {
 767                         bzero(dst6, sizeof(*dst6));
 768                         dst6->sin6_family = AF_INET6;
 769                         dst6->sin6_len = sizeof(*dst6);
 770                         dst6->sin6_addr = ip6->ip6_dst;
 771                         rtalloc(state->ro);
 772                 }
 773                 if (state->ro->ro_rt == 0) {
 774                         ip6stat.ip6s_noroute++;
 775                         ipsec6stat.ips_out_noroute++;
 776                         error = EHOSTUNREACH;
 777                         goto bad;
 778                 }
 779
 780                 /* adjust state->dst if tunnel endpoint is offlink */
 781                 if (state->ro->ro_rt->rt_flags & RTF_GATEWAY) {
 782                         state->dst = (struct sockaddr *)state->ro->ro_rt->rt_gateway;
 783                         dst6 = (struct sockaddr_in6 *)state->dst;
 784                 }
 785         }
 786
 787         m = ipsec6_splithdr(m);
 788         if (!m) {
 789                 ipsec6stat.ips_out_nomem++;
 790                 error = ENOMEM;
 791                 goto bad;
 792         }
 793         ip6 = mtod(m, struct ip6_hdr *);
 794
 795 #ifdef DEV_ENC
 796         /* pass the mbuf to enc0 for bpf processing */
 797         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_AFTER);
 798         /* pass the mbuf to enc0 for packet filtering */
 799         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 800                 goto bad;
 801 #endif
 802
 803         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 804                 sizeof (struct ip6_hdr),
 805                 offsetof(struct ip6_hdr, ip6_nxt));
 806         IPSECREQUEST_UNLOCK(isr);
 807         return error;
 808 bad:
 809         if (isr)
 810                 IPSECREQUEST_UNLOCK(isr);
 811         if (m)
 812                 m_freem(m);
 813         state->m = NULL;
 814         return error;
 815 }
 816 #endif /*INET6*/