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 #ifdef DEV_ENC
  86 #include <net/if_enc.h>
  87 #endif
  88
  89
  90 int
  91 ipsec_process_done(struct mbuf *m, struct ipsecrequest *isr)
  92 {
  93         struct tdb_ident *tdbi;
  94         struct m_tag *mtag;
  95         struct secasvar *sav;
  96         struct secasindex *saidx;
  97         int error;
  98
  99         IPSEC_SPLASSERT_SOFTNET(__func__);
 100
 101         IPSEC_ASSERT(m != NULL, ("null mbuf"));
 102         IPSEC_ASSERT(isr != NULL, ("null ISR"));
 103         sav = isr->sav;
 104         IPSEC_ASSERT(sav != NULL, ("null SA"));
 105         IPSEC_ASSERT(sav->sah != NULL, ("null SAH"));
 106
 107         saidx = &sav->sah->saidx;
 108         switch (saidx->dst.sa.sa_family) {
 109 #ifdef INET
 110         case AF_INET:
 111                 /* Fix the header length, for AH processing. */
 112                 mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
 113                 break;
 114 #endif /* INET */
 115 #ifdef INET6
 116         case AF_INET6:
 117                 /* Fix the header length, for AH processing. */
 118                 if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
 119                         error = ENXIO;
 120                         goto bad;
 121                 }
 122                 if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
 123                         /* No jumbogram support. */
 124                         error = ENXIO;  /*?*/
 125                         goto bad;
 126                 }
 127                 mtod(m, struct ip6_hdr *)->ip6_plen =
 128                         htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
 129                 break;
 130 #endif /* INET6 */
 131         default:
 132                 DPRINTF(("%s: unknown protocol family %u\n", __func__,
 133                     saidx->dst.sa.sa_family));
 134                 error = ENXIO;
 135                 goto bad;
 136         }
 137
 138         /*
 139          * Add a record of what we've done or what needs to be done to the
 140          * packet.
 141          */
 142         mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE,
 143                         sizeof(struct tdb_ident), M_NOWAIT);
 144         if (mtag == NULL) {
 145                 DPRINTF(("%s: could not get packet tag\n", __func__));
 146                 error = ENOMEM;
 147                 goto bad;
 148         }
 149
 150         tdbi = (struct tdb_ident *)(mtag + 1);
 151         tdbi->dst = saidx->dst;
 152         tdbi->proto = saidx->proto;
 153         tdbi->spi = sav->spi;
 154         m_tag_prepend(m, mtag);
 155
 156         /*
 157          * If there's another (bundled) SA to apply, do so.
 158          * Note that this puts a burden on the kernel stack size.
 159          * If this is a problem we'll need to introduce a queue
 160          * to set the packet on so we can unwind the stack before
 161          * doing further processing.
 162          */
 163         if (isr->next) {
 164                 ipsec4stat.ips_out_bundlesa++;
 165                 return ipsec4_process_packet(m, isr->next, 0, 0);
 166         }
 167         key_sa_recordxfer(sav, m);              /* record data transfer */
 168
 169         /*
 170          * We're done with IPsec processing, transmit the packet using the
 171          * appropriate network protocol (IP or IPv6). SPD lookup will be
 172          * performed again there.
 173          */
 174         switch (saidx->dst.sa.sa_family) {
 175 #ifdef INET
 176         struct ip *ip;
 177         case AF_INET:
 178                 ip = mtod(m, struct ip *);
 179                 ip->ip_len = ntohs(ip->ip_len);
 180                 ip->ip_off = ntohs(ip->ip_off);
 181
 182                 return ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
 183 #endif /* INET */
 184 #ifdef INET6
 185         case AF_INET6:
 186                 /*
 187                  * We don't need massage, IPv6 header fields are always in
 188                  * net endian.
 189                  */
 190                 return ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
 191 #endif /* INET6 */
 192         }
 193         panic("ipsec_process_done");
 194 bad:
 195         m_freem(m);
 196         KEY_FREESAV(&sav);
 197         return (error);
 198 }
 199
 200 static struct ipsecrequest *
 201 ipsec_nextisr(
 202         struct mbuf *m,
 203         struct ipsecrequest *isr,
 204         int af,
 205         struct secasindex *saidx,
 206         int *error
 207 )
 208 {
 209 #define IPSEC_OSTAT(x,y,z) (isr->saidx.proto == IPPROTO_ESP ? (x)++ : \
 210                             isr->saidx.proto == IPPROTO_AH ? (y)++ : (z)++)
 211         struct secasvar *sav;
 212
 213         IPSEC_SPLASSERT_SOFTNET(__func__);
 214         IPSECREQUEST_LOCK_ASSERT(isr);
 215
 216         IPSEC_ASSERT(af == AF_INET || af == AF_INET6,
 217                 ("invalid address family %u", af));
 218 again:
 219         /*
 220          * Craft SA index to search for proper SA.  Note that
 221          * we only fillin unspecified SA peers for transport
 222          * mode; for tunnel mode they must already be filled in.
 223          */
 224         *saidx = isr->saidx;
 225         if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
 226                 /* Fillin unspecified SA peers only for transport mode */
 227                 if (af == AF_INET) {
 228                         struct sockaddr_in *sin;
 229                         struct ip *ip = mtod(m, struct ip *);
 230
 231                         if (saidx->src.sa.sa_len == 0) {
 232                                 sin = &saidx->src.sin;
 233                                 sin->sin_len = sizeof(*sin);
 234                                 sin->sin_family = AF_INET;
 235                                 sin->sin_port = IPSEC_PORT_ANY;
 236                                 sin->sin_addr = ip->ip_src;
 237                         }
 238                         if (saidx->dst.sa.sa_len == 0) {
 239                                 sin = &saidx->dst.sin;
 240                                 sin->sin_len = sizeof(*sin);
 241                                 sin->sin_family = AF_INET;
 242                                 sin->sin_port = IPSEC_PORT_ANY;
 243                                 sin->sin_addr = ip->ip_dst;
 244                         }
 245                 } else {
 246                         struct sockaddr_in6 *sin6;
 247                         struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
 248
 249                         if (saidx->src.sin6.sin6_len == 0) {
 250                                 sin6 = (struct sockaddr_in6 *)&saidx->src;
 251                                 sin6->sin6_len = sizeof(*sin6);
 252                                 sin6->sin6_family = AF_INET6;
 253                                 sin6->sin6_port = IPSEC_PORT_ANY;
 254                                 sin6->sin6_addr = ip6->ip6_src;
 255                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
 256                                         /* fix scope id for comparing SPD */
 257                                         sin6->sin6_addr.s6_addr16[1] = 0;
 258                                         sin6->sin6_scope_id =
 259                                             ntohs(ip6->ip6_src.s6_addr16[1]);
 260                                 }
 261                         }
 262                         if (saidx->dst.sin6.sin6_len == 0) {
 263                                 sin6 = (struct sockaddr_in6 *)&saidx->dst;
 264                                 sin6->sin6_len = sizeof(*sin6);
 265                                 sin6->sin6_family = AF_INET6;
 266                                 sin6->sin6_port = IPSEC_PORT_ANY;
 267                                 sin6->sin6_addr = ip6->ip6_dst;
 268                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
 269                                         /* fix scope id for comparing SPD */
 270                                         sin6->sin6_addr.s6_addr16[1] = 0;
 271                                         sin6->sin6_scope_id =
 272                                             ntohs(ip6->ip6_dst.s6_addr16[1]);
 273                                 }
 274                         }
 275                 }
 276         }
 277
 278         /*
 279          * Lookup SA and validate it.
 280          */
 281         *error = key_checkrequest(isr, saidx);
 282         if (*error != 0) {
 283                 /*
 284                  * IPsec processing is required, but no SA found.
 285                  * I assume that key_acquire() had been called
 286                  * to get/establish the SA. Here I discard
 287                  * this packet because it is responsibility for
 288                  * upper layer to retransmit the packet.
 289                  */
 290                 ipsec4stat.ips_out_nosa++;
 291                 goto bad;
 292         }
 293         sav = isr->sav;
 294         if (sav == NULL) {
 295                 IPSEC_ASSERT(ipsec_get_reqlevel(isr) == IPSEC_LEVEL_USE,
 296                         ("no SA found, but required; level %u",
 297                         ipsec_get_reqlevel(isr)));
 298                 IPSECREQUEST_UNLOCK(isr);
 299                 isr = isr->next;
 300                 /*
 301                  * If isr is NULL, we found a 'use' policy w/o SA.
 302                  * Return w/o error and w/o isr so we can drop out
 303                  * and continue w/o IPsec processing.
 304                  */
 305                 if (isr == NULL)
 306                         return isr;
 307                 IPSECREQUEST_LOCK(isr);
 308                 goto again;
 309         }
 310
 311         /*
 312          * Check system global policy controls.
 313          */
 314         if ((isr->saidx.proto == IPPROTO_ESP && !esp_enable) ||
 315             (isr->saidx.proto == IPPROTO_AH && !ah_enable) ||
 316             (isr->saidx.proto == IPPROTO_IPCOMP && !ipcomp_enable)) {
 317                 DPRINTF(("%s: IPsec outbound packet dropped due"
 318                         " to policy (check your sysctls)\n", __func__));
 319                 IPSEC_OSTAT(espstat.esps_pdrops, ahstat.ahs_pdrops,
 320                     ipcompstat.ipcomps_pdrops);
 321                 *error = EHOSTUNREACH;
 322                 goto bad;
 323         }
 324
 325         /*
 326          * Sanity check the SA contents for the caller
 327          * before they invoke the xform output method.
 328          */
 329         if (sav->tdb_xform == NULL) {
 330                 DPRINTF(("%s: no transform for SA\n", __func__));
 331                 IPSEC_OSTAT(espstat.esps_noxform, ahstat.ahs_noxform,
 332                     ipcompstat.ipcomps_noxform);
 333                 *error = EHOSTUNREACH;
 334                 goto bad;
 335         }
 336         return isr;
 337 bad:
 338         IPSEC_ASSERT(*error != 0, ("error return w/ no error code"));
 339         IPSECREQUEST_UNLOCK(isr);
 340         return NULL;
 341 #undef IPSEC_OSTAT
 342 }
 343
 344 #ifdef INET
 345 /*
 346  * IPsec output logic for IPv4.
 347  */
 348 int
 349 ipsec4_process_packet(
 350         struct mbuf *m,
 351         struct ipsecrequest *isr,
 352         int flags,
 353         int tunalready)
 354 {
 355         struct secasindex saidx;
 356         struct secasvar *sav;
 357         struct ip *ip;
 358         int error, i, off;
 359
 360         IPSEC_ASSERT(m != NULL, ("null mbuf"));
 361         IPSEC_ASSERT(isr != NULL, ("null isr"));
 362
 363         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 364
 365         isr = ipsec_nextisr(m, isr, AF_INET, &saidx, &error);
 366         if (isr == NULL) {
 367                 if (error != 0)
 368                         goto bad;
 369                 return EJUSTRETURN;
 370         }
 371
 372         sav = isr->sav;
 373
 374 #ifdef DEV_ENC
 375         encif->if_opackets++;
 376         encif->if_obytes += m->m_pkthdr.len;
 377
 378         /* pass the mbuf to enc0 for bpf processing */
 379         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_BEFORE);
 380         /* pass the mbuf to enc0 for packet filtering */
 381         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 382                 goto bad;
 383 #endif
 384
 385         if (!tunalready) {
 386                 union sockaddr_union *dst = &sav->sah->saidx.dst;
 387                 int setdf;
 388
 389                 /*
 390                  * Collect IP_DF state from the outer header.
 391                  */
 392                 if (dst->sa.sa_family == AF_INET) {
 393                         if (m->m_len < sizeof (struct ip) &&
 394                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 395                                 error = ENOBUFS;
 396                                 goto bad;
 397                         }
 398                         ip = mtod(m, struct ip *);
 399                         /* Honor system-wide control of how to handle IP_DF */
 400                         switch (ip4_ipsec_dfbit) {
 401                         case 0:                 /* clear in outer header */
 402                         case 1:                 /* set in outer header */
 403                                 setdf = ip4_ipsec_dfbit;
 404                                 break;
 405                         default:                /* propagate to outer header */
 406                                 setdf = ntohs(ip->ip_off & IP_DF);
 407                                 break;
 408                         }
 409                 } else {
 410                         ip = NULL;              /* keep compiler happy */
 411                         setdf = 0;
 412                 }
 413                 /* Do the appropriate encapsulation, if necessary */
 414                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
 415                     dst->sa.sa_family != AF_INET ||         /* PF mismatch */
 416 #if 0
 417                     (sav->flags & SADB_X_SAFLAGS_TUNNEL) || /* Tunnel requ'd */
 418                     sav->tdb_xform->xf_type == XF_IP4 ||    /* ditto */
 419 #endif
 420                     (dst->sa.sa_family == AF_INET &&        /* Proxy */
 421                      dst->sin.sin_addr.s_addr != INADDR_ANY &&
 422                      dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
 423                         struct mbuf *mp;
 424
 425                         /* Fix IPv4 header checksum and length */
 426                         if (m->m_len < sizeof (struct ip) &&
 427                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 428                                 error = ENOBUFS;
 429                                 goto bad;
 430                         }
 431                         ip = mtod(m, struct ip *);
 432                         ip->ip_len = htons(m->m_pkthdr.len);
 433                         ip->ip_sum = 0;
 434 #ifdef _IP_VHL
 435                         if (ip->ip_vhl == IP_VHL_BORING)
 436                                 ip->ip_sum = in_cksum_hdr(ip);
 437                         else
 438                                 ip->ip_sum = in_cksum(m,
 439                                         _IP_VHL_HL(ip->ip_vhl) << 2);
 440 #else
 441                         ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
 442 #endif
 443
 444                         /* Encapsulate the packet */
 445                         error = ipip_output(m, isr, &mp, 0, 0);
 446                         if (mp == NULL && !error) {
 447                                 /* Should never happen. */
 448                                 DPRINTF(("%s: ipip_output returns no mbuf and "
 449                                         "no error!", __func__));
 450                                 error = EFAULT;
 451                         }
 452                         if (error) {
 453                                 if (mp) {
 454                                         /* XXX: Should never happen! */
 455                                         m_freem(mp);
 456                                 }
 457                                 m = NULL; /* ipip_output() already freed it */
 458                                 goto bad;
 459                         }
 460                         m = mp, mp = NULL;
 461                         /*
 462                          * ipip_output clears IP_DF in the new header.  If
 463                          * we need to propagate IP_DF from the outer header,
 464                          * then we have to do it here.
 465                          *
 466                          * XXX shouldn't assume what ipip_output does.
 467                          */
 468                         if (dst->sa.sa_family == AF_INET && setdf) {
 469                                 if (m->m_len < sizeof (struct ip) &&
 470                                     (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 471                                         error = ENOBUFS;
 472                                         goto bad;
 473                                 }
 474                                 ip = mtod(m, struct ip *);
 475                                 ip->ip_off = ntohs(ip->ip_off);
 476                                 ip->ip_off |= IP_DF;
 477                                 ip->ip_off = htons(ip->ip_off);
 478                         }
 479                 }
 480         }
 481
 482 #ifdef DEV_ENC
 483         /* pass the mbuf to enc0 for bpf processing */
 484         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_AFTER);
 485         /* pass the mbuf to enc0 for packet filtering */
 486         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 487                 goto bad;
 488 #endif
 489
 490         /*
 491          * Dispatch to the appropriate IPsec transform logic.  The
 492          * packet will be returned for transmission after crypto
 493          * processing, etc. are completed.  For encapsulation we
 494          * bypass this call because of the explicit call done above
 495          * (necessary to deal with IP_DF handling for IPv4).
 496          *
 497          * NB: m & sav are ``passed to caller'' who's reponsible for
 498          *     for reclaiming their resources.
 499          */
 500         if (sav->tdb_xform->xf_type != XF_IP4) {
 501                 ip = mtod(m, struct ip *);
 502                 i = ip->ip_hl << 2;
 503                 off = offsetof(struct ip, ip_p);
 504                 error = (*sav->tdb_xform->xf_output)(m, isr, NULL, i, off);
 505         } else {
 506                 error = ipsec_process_done(m, isr);
 507         }
 508         IPSECREQUEST_UNLOCK(isr);
 509         return error;
 510 bad:
 511         if (isr)
 512                 IPSECREQUEST_UNLOCK(isr);
 513         if (m)
 514                 m_freem(m);
 515         return error;
 516 }
 517 #endif
 518
 519 #ifdef INET6
 520 /*
 521  * Chop IP6 header from the payload.
 522  */
 523 static struct mbuf *
 524 ipsec6_splithdr(struct mbuf *m)
 525 {
 526         struct mbuf *mh;
 527         struct ip6_hdr *ip6;
 528         int hlen;
 529
 530         IPSEC_ASSERT(m->m_len >= sizeof (struct ip6_hdr),
 531                 ("first mbuf too short, len %u", m->m_len));
 532         ip6 = mtod(m, struct ip6_hdr *);
 533         hlen = sizeof(struct ip6_hdr);
 534         if (m->m_len > hlen) {
 535                 MGETHDR(mh, M_DONTWAIT, MT_DATA);
 536                 if (!mh) {
 537                         m_freem(m);
 538                         return NULL;
 539                 }
 540                 M_MOVE_PKTHDR(mh, m);
 541                 MH_ALIGN(mh, hlen);
 542                 m->m_len -= hlen;
 543                 m->m_data += hlen;
 544                 mh->m_next = m;
 545                 m = mh;
 546                 m->m_len = hlen;
 547                 bcopy((caddr_t)ip6, mtod(m, caddr_t), hlen);
 548         } else if (m->m_len < hlen) {
 549                 m = m_pullup(m, hlen);
 550                 if (!m)
 551                         return NULL;
 552         }
 553         return m;
 554 }
 555
 556 /*
 557  * IPsec output logic for IPv6, transport mode.
 558  */
 559 int
 560 ipsec6_output_trans(
 561         struct ipsec_output_state *state,
 562         u_char *nexthdrp,
 563         struct mbuf *mprev,
 564         struct secpolicy *sp,
 565         int flags,
 566         int *tun)
 567 {
 568         struct ipsecrequest *isr;
 569         struct secasindex saidx;
 570         int error = 0;
 571         struct mbuf *m;
 572
 573         IPSEC_ASSERT(state != NULL, ("null state"));
 574         IPSEC_ASSERT(state->m != NULL, ("null m"));
 575         IPSEC_ASSERT(nexthdrp != NULL, ("null nexthdrp"));
 576         IPSEC_ASSERT(mprev != NULL, ("null mprev"));
 577         IPSEC_ASSERT(sp != NULL, ("null sp"));
 578         IPSEC_ASSERT(tun != NULL, ("null tun"));
 579
 580         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 581                 printf("%s: applied SP\n", __func__);
 582                 kdebug_secpolicy(sp));
 583
 584         isr = sp->req;
 585         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 586                 /* the rest will be handled by ipsec6_output_tunnel() */
 587                 *tun = 1;               /* need tunnel-mode processing */
 588                 return 0;
 589         }
 590
 591         *tun = 0;
 592         m = state->m;
 593
 594         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 595         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 596         if (isr == NULL) {
 597                 if (error != 0) {
 598 #ifdef notdef
 599                         /* XXX should notification be done for all errors ? */
 600                         /*
 601                          * Notify the fact that the packet is discarded
 602                          * to ourselves. I believe this is better than
 603                          * just silently discarding. (jinmei@kame.net)
 604                          * XXX: should we restrict the error to TCP packets?
 605                          * XXX: should we directly notify sockets via
 606                          *      pfctlinputs?
 607                          */
 608                         icmp6_error(m, ICMP6_DST_UNREACH,
 609                                     ICMP6_DST_UNREACH_ADMIN, 0);
 610                         m = NULL;       /* NB: icmp6_error frees mbuf */
 611 #endif
 612                         goto bad;
 613                 }
 614                 return EJUSTRETURN;
 615         }
 616
 617         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 618                                                   sizeof (struct ip6_hdr),
 619                                                   offsetof(struct ip6_hdr,
 620                                                            ip6_nxt));
 621         IPSECREQUEST_UNLOCK(isr);
 622         return error;
 623 bad:
 624         if (isr)
 625                 IPSECREQUEST_UNLOCK(isr);
 626         if (m)
 627                 m_freem(m);
 628         state->m = NULL;
 629         return error;
 630 }
 631
 632 static int
 633 ipsec6_encapsulate(struct mbuf *m, struct secasvar *sav)
 634 {
 635         struct ip6_hdr *oip6;
 636         struct ip6_hdr *ip6;
 637         size_t plen;
 638
 639         /* can't tunnel between different AFs */
 640         if (sav->sah->saidx.src.sa.sa_family != AF_INET6 ||
 641             sav->sah->saidx.dst.sa.sa_family != AF_INET6) {
 642                 m_freem(m);
 643                 return EINVAL;
 644         }
 645         IPSEC_ASSERT(m->m_len == sizeof (struct ip6_hdr),
 646                 ("mbuf wrong size; len %u", m->m_len));
 647
 648
 649         /*
 650          * grow the mbuf to accomodate the new IPv6 header.
 651          */
 652         plen = m->m_pkthdr.len;
 653         if (M_LEADINGSPACE(m->m_next) < sizeof(struct ip6_hdr)) {
 654                 struct mbuf *n;
 655                 MGET(n, M_DONTWAIT, MT_DATA);
 656                 if (!n) {
 657                         m_freem(m);
 658                         return ENOBUFS;
 659                 }
 660                 n->m_len = sizeof(struct ip6_hdr);
 661                 n->m_next = m->m_next;
 662                 m->m_next = n;
 663                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 664                 oip6 = mtod(n, struct ip6_hdr *);
 665         } else {
 666                 m->m_next->m_len += sizeof(struct ip6_hdr);
 667                 m->m_next->m_data -= sizeof(struct ip6_hdr);
 668                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 669                 oip6 = mtod(m->m_next, struct ip6_hdr *);
 670         }
 671         ip6 = mtod(m, struct ip6_hdr *);
 672         bcopy((caddr_t)ip6, (caddr_t)oip6, sizeof(struct ip6_hdr));
 673
 674         /* Fake link-local scope-class addresses */
 675         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_src))
 676                 oip6->ip6_src.s6_addr16[1] = 0;
 677         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_dst))
 678                 oip6->ip6_dst.s6_addr16[1] = 0;
 679
 680         /* construct new IPv6 header. see RFC 2401 5.1.2.2 */
 681         /* ECN consideration. */
 682         ip6_ecn_ingress(ip6_ipsec_ecn, &ip6->ip6_flow, &oip6->ip6_flow);
 683         if (plen < IPV6_MAXPACKET - sizeof(struct ip6_hdr))
 684                 ip6->ip6_plen = htons(plen);
 685         else {
 686                 /* ip6->ip6_plen will be updated in ip6_output() */
 687         }
 688         ip6->ip6_nxt = IPPROTO_IPV6;
 689         ip6->ip6_src = sav->sah->saidx.src.sin6.sin6_addr;
 690         ip6->ip6_dst = sav->sah->saidx.dst.sin6.sin6_addr;
 691         ip6->ip6_hlim = IPV6_DEFHLIM;
 692
 693         /* XXX Should ip6_src be updated later ? */
 694
 695         return 0;
 696 }
 697
 698 /*
 699  * IPsec output logic for IPv6, tunnel mode.
 700  */
 701 int
 702 ipsec6_output_tunnel(struct ipsec_output_state *state, struct secpolicy *sp, int flags)
 703 {
 704         struct ip6_hdr *ip6;
 705         struct ipsecrequest *isr;
 706         struct secasindex saidx;
 707         int error;
 708         struct sockaddr_in6* dst6;
 709         struct mbuf *m;
 710
 711         IPSEC_ASSERT(state != NULL, ("null state"));
 712         IPSEC_ASSERT(state->m != NULL, ("null m"));
 713         IPSEC_ASSERT(sp != NULL, ("null sp"));
 714
 715         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 716                 printf("%s: applied SP\n", __func__);
 717                 kdebug_secpolicy(sp));
 718
 719         m = state->m;
 720         /*
 721          * transport mode ipsec (before the 1st tunnel mode) is already
 722          * processed by ipsec6_output_trans().
 723          */
 724         for (isr = sp->req; isr; isr = isr->next) {
 725                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
 726                         break;
 727         }
 728
 729         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 730         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 731         if (isr == NULL) {
 732                 if (error != 0)
 733                         goto bad;
 734                 return EJUSTRETURN;
 735         }
 736
 737 #ifdef DEV_ENC
 738         /* pass the mbuf to enc0 for bpf processing */
 739         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_BEFORE);
 740         /* pass the mbuf to enc0 for packet filtering */
 741         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 742                 goto bad;
 743 #endif
 744
 745         /*
 746          * There may be the case that SA status will be changed when
 747          * we are refering to one. So calling splsoftnet().
 748          */
 749         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 750                 /*
 751                  * build IPsec tunnel.
 752                  */
 753                 /* XXX should be processed with other familiy */
 754                 if (isr->sav->sah->saidx.src.sa.sa_family != AF_INET6) {
 755                         ipseclog((LOG_ERR, "%s: family mismatched between "
 756                             "inner and outer, spi=%u\n", __func__,
 757                             ntohl(isr->sav->spi)));
 758                         ipsec6stat.ips_out_inval++;
 759                         error = EAFNOSUPPORT;
 760                         goto bad;
 761                 }
 762
 763                 m = ipsec6_splithdr(m);
 764                 if (!m) {
 765                         ipsec6stat.ips_out_nomem++;
 766                         error = ENOMEM;
 767                         goto bad;
 768                 }
 769                 error = ipsec6_encapsulate(m, isr->sav);
 770                 if (error) {
 771                         m = NULL;
 772                         goto bad;
 773                 }
 774                 ip6 = mtod(m, struct ip6_hdr *);
 775
 776                 state->ro = &isr->sav->sah->sa_route;
 777                 state->dst = (struct sockaddr *)&state->ro->ro_dst;
 778                 dst6 = (struct sockaddr_in6 *)state->dst;
 779                 if (state->ro->ro_rt
 780                  && ((state->ro->ro_rt->rt_flags & RTF_UP) == 0
 781                   || !IN6_ARE_ADDR_EQUAL(&dst6->sin6_addr, &ip6->ip6_dst))) {
 782                         RTFREE(state->ro->ro_rt);
 783                         state->ro->ro_rt = NULL;
 784                 }
 785                 if (state->ro->ro_rt == 0) {
 786                         bzero(dst6, sizeof(*dst6));
 787                         dst6->sin6_family = AF_INET6;
 788                         dst6->sin6_len = sizeof(*dst6);
 789                         dst6->sin6_addr = ip6->ip6_dst;
 790                         rtalloc(state->ro);
 791                 }
 792                 if (state->ro->ro_rt == 0) {
 793                         ip6stat.ip6s_noroute++;
 794                         ipsec6stat.ips_out_noroute++;
 795                         error = EHOSTUNREACH;
 796                         goto bad;
 797                 }
 798
 799                 /* adjust state->dst if tunnel endpoint is offlink */
 800                 if (state->ro->ro_rt->rt_flags & RTF_GATEWAY) {
 801                         state->dst = (struct sockaddr *)state->ro->ro_rt->rt_gateway;
 802                         dst6 = (struct sockaddr_in6 *)state->dst;
 803                 }
 804         }
 805
 806         m = ipsec6_splithdr(m);
 807         if (!m) {
 808                 ipsec6stat.ips_out_nomem++;
 809                 error = ENOMEM;
 810                 goto bad;
 811         }
 812         ip6 = mtod(m, struct ip6_hdr *);
 813
 814 #ifdef DEV_ENC
 815         /* pass the mbuf to enc0 for bpf processing */
 816         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_AFTER);
 817         /* pass the mbuf to enc0 for packet filtering */
 818         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 819                 goto bad;
 820 #endif
 821
 822         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 823                 sizeof (struct ip6_hdr),
 824                 offsetof(struct ip6_hdr, ip6_nxt));
 825         IPSECREQUEST_UNLOCK(isr);
 826         return error;
 827 bad:
 828         if (isr)
 829                 IPSECREQUEST_UNLOCK(isr);
 830         if (m)
 831                 m_freem(m);
 832         state->m = NULL;
 833         return error;
 834 }
 835 #endif /*INET6*/