sys/netipsec/ipsec_output.c

   1 /*-
   2  * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  *
  26  * $FreeBSD$
  27  */
  28
  29 /*
  30  * IPsec output processing.
  31  */
  32 #include "opt_inet.h"
  33 #include "opt_inet6.h"
  34 #include "opt_ipsec.h"
  35 #include "opt_enc.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/mbuf.h>
  40 #include <sys/domain.h>
  41 #include <sys/protosw.h>
  42 #include <sys/socket.h>
  43 #include <sys/errno.h>
  44 #include <sys/syslog.h>
  45
  46 #include <net/if.h>
  47 #include <net/pfil.h>
  48 #include <net/route.h>
  49 #include <net/vnet.h>
  50
  51 #include <netinet/in.h>
  52 #include <netinet/in_systm.h>
  53 #include <netinet/ip.h>
  54 #include <netinet/ip_var.h>
  55 #include <netinet/in_var.h>
  56 #include <netinet/ip_ecn.h>
  57 #ifdef INET6
  58 #include <netinet6/ip6_ecn.h>
  59 #endif
  60
  61 #include <netinet/ip6.h>
  62 #ifdef INET6
  63 #include <netinet6/ip6_var.h>
  64 #endif
  65 #include <netinet/in_pcb.h>
  66 #ifdef INET6
  67 #include <netinet/icmp6.h>
  68 #endif
  69
  70 #include <netipsec/ipsec.h>
  71 #ifdef INET6
  72 #include <netipsec/ipsec6.h>
  73 #endif
  74 #include <netipsec/ah_var.h>
  75 #include <netipsec/esp_var.h>
  76 #include <netipsec/ipcomp_var.h>
  77
  78 #include <netipsec/xform.h>
  79
  80 #include <netipsec/key.h>
  81 #include <netipsec/keydb.h>
  82 #include <netipsec/key_debug.h>
  83
  84 #include <machine/in_cksum.h>
  85
  86 #ifdef IPSEC_NAT_T
  87 #include <netinet/udp.h>
  88 #endif
  89
  90 #ifdef DEV_ENC
  91 #include <net/if_enc.h>
  92 #endif
  93
  94
  95 int
  96 ipsec_process_done(struct mbuf *m, struct ipsecrequest *isr)
  97 {
  98         struct tdb_ident *tdbi;
  99         struct m_tag *mtag;
 100         struct secasvar *sav;
 101         struct secasindex *saidx;
 102         int error;
 103
 104         IPSEC_ASSERT(m != NULL, ("null mbuf"));
 105         IPSEC_ASSERT(isr != NULL, ("null ISR"));
 106         sav = isr->sav;
 107         IPSEC_ASSERT(sav != NULL, ("null SA"));
 108         IPSEC_ASSERT(sav->sah != NULL, ("null SAH"));
 109
 110         saidx = &sav->sah->saidx;
 111         switch (saidx->dst.sa.sa_family) {
 112 #ifdef INET
 113         case AF_INET:
 114                 /* Fix the header length, for AH processing. */
 115                 mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len);
 116                 break;
 117 #endif /* INET */
 118 #ifdef INET6
 119         case AF_INET6:
 120                 /* Fix the header length, for AH processing. */
 121                 if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) {
 122                         error = ENXIO;
 123                         goto bad;
 124                 }
 125                 if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) {
 126                         /* No jumbogram support. */
 127                         error = ENXIO;  /*?*/
 128                         goto bad;
 129                 }
 130                 mtod(m, struct ip6_hdr *)->ip6_plen =
 131                         htons(m->m_pkthdr.len - sizeof(struct ip6_hdr));
 132                 break;
 133 #endif /* INET6 */
 134         default:
 135                 DPRINTF(("%s: unknown protocol family %u\n", __func__,
 136                     saidx->dst.sa.sa_family));
 137                 error = ENXIO;
 138                 goto bad;
 139         }
 140
 141         /*
 142          * Add a record of what we've done or what needs to be done to the
 143          * packet.
 144          */
 145         mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE,
 146                         sizeof(struct tdb_ident), M_NOWAIT);
 147         if (mtag == NULL) {
 148                 DPRINTF(("%s: could not get packet tag\n", __func__));
 149                 error = ENOMEM;
 150                 goto bad;
 151         }
 152
 153         tdbi = (struct tdb_ident *)(mtag + 1);
 154         tdbi->dst = saidx->dst;
 155         tdbi->proto = saidx->proto;
 156         tdbi->spi = sav->spi;
 157         m_tag_prepend(m, mtag);
 158
 159         /*
 160          * If there's another (bundled) SA to apply, do so.
 161          * Note that this puts a burden on the kernel stack size.
 162          * If this is a problem we'll need to introduce a queue
 163          * to set the packet on so we can unwind the stack before
 164          * doing further processing.
 165          */
 166         if (isr->next) {
 167                 IPSECSTAT_INC(ips_out_bundlesa);
 168                 /* XXX-BZ currently only support same AF bundles. */
 169                 switch (saidx->dst.sa.sa_family) {
 170 #ifdef INET
 171                 case AF_INET:
 172                         return ipsec4_process_packet(m, isr->next, 0, 0);
 173                         /* NOTREACHED */
 174 #endif
 175 #ifdef notyet
 176 #ifdef INET6
 177                 case AF_INET6:
 178                         /* XXX */
 179                         ipsec6_output_trans()
 180                         ipsec6_output_tunnel()
 181                         /* NOTREACHED */
 182 #endif /* INET6 */
 183 #endif
 184                 default:
 185                         DPRINTF(("%s: unknown protocol family %u\n", __func__,
 186                             saidx->dst.sa.sa_family));
 187                         error = ENXIO;
 188                         goto bad;
 189                 }
 190         }
 191         key_sa_recordxfer(sav, m);              /* record data transfer */
 192
 193         m_addr_changed(m);
 194
 195         /*
 196          * We're done with IPsec processing, transmit the packet using the
 197          * appropriate network protocol (IP or IPv6). SPD lookup will be
 198          * performed again there.
 199          */
 200         switch (saidx->dst.sa.sa_family) {
 201 #ifdef INET
 202         struct ip *ip;
 203         case AF_INET:
 204                 ip = mtod(m, struct ip *);
 205                 ip->ip_len = ntohs(ip->ip_len);
 206                 ip->ip_off = ntohs(ip->ip_off);
 207
 208 #ifdef IPSEC_NAT_T
 209                 /*
 210                  * If NAT-T is enabled, now that all IPsec processing is done
 211                  * insert UDP encapsulation header after IP header.
 212                  */
 213                 if (sav->natt_type) {
 214 #ifdef _IP_VHL
 215                         const int hlen = IP_VHL_HL(ip->ip_vhl);
 216 #else
 217                         const int hlen = (ip->ip_hl << 2);
 218 #endif
 219                         int size, off;
 220                         struct mbuf *mi;
 221                         struct udphdr *udp;
 222
 223                         size = sizeof(struct udphdr);
 224                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
 225                                 /*
 226                                  * draft-ietf-ipsec-nat-t-ike-0[01].txt and
 227                                  * draft-ietf-ipsec-udp-encaps-(00/)01.txt,
 228                                  * ignoring possible AH mode
 229                                  * non-IKE marker + non-ESP marker
 230                                  * from draft-ietf-ipsec-udp-encaps-00.txt.
 231                                  */
 232                                 size += sizeof(u_int64_t);
 233                         }
 234                         mi = m_makespace(m, hlen, size, &off);
 235                         if (mi == NULL) {
 236                                 DPRINTF(("%s: m_makespace for udphdr failed\n",
 237                                     __func__));
 238                                 error = ENOBUFS;
 239                                 goto bad;
 240                         }
 241
 242                         udp = (struct udphdr *)(mtod(mi, caddr_t) + off);
 243                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
 244                                 udp->uh_sport = htons(UDP_ENCAP_ESPINUDP_PORT);
 245                         else
 246                                 udp->uh_sport =
 247                                         KEY_PORTFROMSADDR(&sav->sah->saidx.src);
 248                         udp->uh_dport = KEY_PORTFROMSADDR(&sav->sah->saidx.dst);
 249                         udp->uh_sum = 0;
 250                         udp->uh_ulen = htons(m->m_pkthdr.len - hlen);
 251                         ip->ip_len = m->m_pkthdr.len;
 252                         ip->ip_p = IPPROTO_UDP;
 253
 254                         if (sav->natt_type == UDP_ENCAP_ESPINUDP_NON_IKE)
 255                                 *(u_int64_t *)(udp + 1) = 0;
 256                 }
 257 #endif /* IPSEC_NAT_T */
 258
 259                 return ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL);
 260 #endif /* INET */
 261 #ifdef INET6
 262         case AF_INET6:
 263                 /*
 264                  * We don't need massage, IPv6 header fields are always in
 265                  * net endian.
 266                  */
 267                 return ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
 268 #endif /* INET6 */
 269         }
 270         panic("ipsec_process_done");
 271 bad:
 272         m_freem(m);
 273         return (error);
 274 }
 275
 276 static struct ipsecrequest *
 277 ipsec_nextisr(
 278         struct mbuf *m,
 279         struct ipsecrequest *isr,
 280         int af,
 281         struct secasindex *saidx,
 282         int *error
 283 )
 284 {
 285 #define IPSEC_OSTAT(name)       do {            \
 286         if (isr->saidx.proto == IPPROTO_ESP)    \
 287                 ESPSTAT_INC(esps_##name);       \
 288         else if (isr->saidx.proto == IPPROTO_AH)\
 289                 AHSTAT_INC(ahs_##name);         \
 290         else                                    \
 291                 IPCOMPSTAT_INC(ipcomps_##name); \
 292 } while (0)
 293         struct secasvar *sav;
 294
 295         IPSECREQUEST_LOCK_ASSERT(isr);
 296
 297         IPSEC_ASSERT(af == AF_INET || af == AF_INET6,
 298                 ("invalid address family %u", af));
 299 again:
 300         /*
 301          * Craft SA index to search for proper SA.  Note that
 302          * we only fillin unspecified SA peers for transport
 303          * mode; for tunnel mode they must already be filled in.
 304          */
 305         *saidx = isr->saidx;
 306         if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) {
 307                 /* Fillin unspecified SA peers only for transport mode */
 308                 if (af == AF_INET) {
 309                         struct sockaddr_in *sin;
 310                         struct ip *ip = mtod(m, struct ip *);
 311
 312                         if (saidx->src.sa.sa_len == 0) {
 313                                 sin = &saidx->src.sin;
 314                                 sin->sin_len = sizeof(*sin);
 315                                 sin->sin_family = AF_INET;
 316                                 sin->sin_port = IPSEC_PORT_ANY;
 317                                 sin->sin_addr = ip->ip_src;
 318                         }
 319                         if (saidx->dst.sa.sa_len == 0) {
 320                                 sin = &saidx->dst.sin;
 321                                 sin->sin_len = sizeof(*sin);
 322                                 sin->sin_family = AF_INET;
 323                                 sin->sin_port = IPSEC_PORT_ANY;
 324                                 sin->sin_addr = ip->ip_dst;
 325                         }
 326                 } else {
 327                         struct sockaddr_in6 *sin6;
 328                         struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
 329
 330                         if (saidx->src.sin6.sin6_len == 0) {
 331                                 sin6 = (struct sockaddr_in6 *)&saidx->src;
 332                                 sin6->sin6_len = sizeof(*sin6);
 333                                 sin6->sin6_family = AF_INET6;
 334                                 sin6->sin6_port = IPSEC_PORT_ANY;
 335                                 sin6->sin6_addr = ip6->ip6_src;
 336                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
 337                                         /* fix scope id for comparing SPD */
 338                                         sin6->sin6_addr.s6_addr16[1] = 0;
 339                                         sin6->sin6_scope_id =
 340                                             ntohs(ip6->ip6_src.s6_addr16[1]);
 341                                 }
 342                         }
 343                         if (saidx->dst.sin6.sin6_len == 0) {
 344                                 sin6 = (struct sockaddr_in6 *)&saidx->dst;
 345                                 sin6->sin6_len = sizeof(*sin6);
 346                                 sin6->sin6_family = AF_INET6;
 347                                 sin6->sin6_port = IPSEC_PORT_ANY;
 348                                 sin6->sin6_addr = ip6->ip6_dst;
 349                                 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) {
 350                                         /* fix scope id for comparing SPD */
 351                                         sin6->sin6_addr.s6_addr16[1] = 0;
 352                                         sin6->sin6_scope_id =
 353                                             ntohs(ip6->ip6_dst.s6_addr16[1]);
 354                                 }
 355                         }
 356                 }
 357         }
 358
 359         /*
 360          * Lookup SA and validate it.
 361          */
 362         *error = key_checkrequest(isr, saidx);
 363         if (*error != 0) {
 364                 /*
 365                  * IPsec processing is required, but no SA found.
 366                  * I assume that key_acquire() had been called
 367                  * to get/establish the SA. Here I discard
 368                  * this packet because it is responsibility for
 369                  * upper layer to retransmit the packet.
 370                  */
 371                 IPSECSTAT_INC(ips_out_nosa);
 372                 goto bad;
 373         }
 374         sav = isr->sav;
 375         if (sav == NULL) {
 376                 IPSEC_ASSERT(ipsec_get_reqlevel(isr) == IPSEC_LEVEL_USE,
 377                         ("no SA found, but required; level %u",
 378                         ipsec_get_reqlevel(isr)));
 379                 IPSECREQUEST_UNLOCK(isr);
 380                 isr = isr->next;
 381                 /*
 382                  * If isr is NULL, we found a 'use' policy w/o SA.
 383                  * Return w/o error and w/o isr so we can drop out
 384                  * and continue w/o IPsec processing.
 385                  */
 386                 if (isr == NULL)
 387                         return isr;
 388                 IPSECREQUEST_LOCK(isr);
 389                 goto again;
 390         }
 391
 392         /*
 393          * Check system global policy controls.
 394          */
 395         if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) ||
 396             (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) ||
 397             (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) {
 398                 DPRINTF(("%s: IPsec outbound packet dropped due"
 399                         " to policy (check your sysctls)\n", __func__));
 400                 IPSEC_OSTAT(pdrops);
 401                 *error = EHOSTUNREACH;
 402                 goto bad;
 403         }
 404
 405         /*
 406          * Sanity check the SA contents for the caller
 407          * before they invoke the xform output method.
 408          */
 409         if (sav->tdb_xform == NULL) {
 410                 DPRINTF(("%s: no transform for SA\n", __func__));
 411                 IPSEC_OSTAT(noxform);
 412                 *error = EHOSTUNREACH;
 413                 goto bad;
 414         }
 415         return isr;
 416 bad:
 417         IPSEC_ASSERT(*error != 0, ("error return w/ no error code"));
 418         IPSECREQUEST_UNLOCK(isr);
 419         return NULL;
 420 #undef IPSEC_OSTAT
 421 }
 422
 423 #ifdef INET
 424 /*
 425  * IPsec output logic for IPv4.
 426  */
 427 int
 428 ipsec4_process_packet(
 429         struct mbuf *m,
 430         struct ipsecrequest *isr,
 431         int flags,
 432         int tunalready)
 433 {
 434         struct secasindex saidx;
 435         struct secasvar *sav;
 436         struct ip *ip;
 437         int error, i, off;
 438
 439         IPSEC_ASSERT(m != NULL, ("null mbuf"));
 440         IPSEC_ASSERT(isr != NULL, ("null isr"));
 441
 442         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 443
 444         isr = ipsec_nextisr(m, isr, AF_INET, &saidx, &error);
 445         if (isr == NULL) {
 446                 if (error != 0)
 447                         goto bad;
 448                 return EJUSTRETURN;
 449         }
 450
 451         sav = isr->sav;
 452
 453 #ifdef DEV_ENC
 454         encif->if_opackets++;
 455         encif->if_obytes += m->m_pkthdr.len;
 456
 457         /* pass the mbuf to enc0 for bpf processing */
 458         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_BEFORE);
 459         /* pass the mbuf to enc0 for packet filtering */
 460         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 461                 goto bad;
 462 #endif
 463
 464         if (!tunalready) {
 465                 union sockaddr_union *dst = &sav->sah->saidx.dst;
 466                 int setdf;
 467
 468                 /*
 469                  * Collect IP_DF state from the outer header.
 470                  */
 471                 if (dst->sa.sa_family == AF_INET) {
 472                         if (m->m_len < sizeof (struct ip) &&
 473                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 474                                 error = ENOBUFS;
 475                                 goto bad;
 476                         }
 477                         ip = mtod(m, struct ip *);
 478                         /* Honor system-wide control of how to handle IP_DF */
 479                         switch (V_ip4_ipsec_dfbit) {
 480                         case 0:                 /* clear in outer header */
 481                         case 1:                 /* set in outer header */
 482                                 setdf = V_ip4_ipsec_dfbit;
 483                                 break;
 484                         default:                /* propagate to outer header */
 485                                 setdf = ntohs(ip->ip_off & IP_DF);
 486                                 break;
 487                         }
 488                 } else {
 489                         ip = NULL;              /* keep compiler happy */
 490                         setdf = 0;
 491                 }
 492                 /* Do the appropriate encapsulation, if necessary */
 493                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */
 494                     dst->sa.sa_family != AF_INET ||         /* PF mismatch */
 495 #if 0
 496                     (sav->flags & SADB_X_SAFLAGS_TUNNEL) || /* Tunnel requ'd */
 497                     sav->tdb_xform->xf_type == XF_IP4 ||    /* ditto */
 498 #endif
 499                     (dst->sa.sa_family == AF_INET &&        /* Proxy */
 500                      dst->sin.sin_addr.s_addr != INADDR_ANY &&
 501                      dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) {
 502                         struct mbuf *mp;
 503
 504                         /* Fix IPv4 header checksum and length */
 505                         if (m->m_len < sizeof (struct ip) &&
 506                             (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 507                                 error = ENOBUFS;
 508                                 goto bad;
 509                         }
 510                         ip = mtod(m, struct ip *);
 511                         ip->ip_len = htons(m->m_pkthdr.len);
 512                         ip->ip_sum = 0;
 513 #ifdef _IP_VHL
 514                         if (ip->ip_vhl == IP_VHL_BORING)
 515                                 ip->ip_sum = in_cksum_hdr(ip);
 516                         else
 517                                 ip->ip_sum = in_cksum(m,
 518                                         _IP_VHL_HL(ip->ip_vhl) << 2);
 519 #else
 520                         ip->ip_sum = in_cksum(m, ip->ip_hl << 2);
 521 #endif
 522
 523                         /* Encapsulate the packet */
 524                         error = ipip_output(m, isr, &mp, 0, 0);
 525                         if (mp == NULL && !error) {
 526                                 /* Should never happen. */
 527                                 DPRINTF(("%s: ipip_output returns no mbuf and "
 528                                         "no error!", __func__));
 529                                 error = EFAULT;
 530                         }
 531                         if (error) {
 532                                 if (mp) {
 533                                         /* XXX: Should never happen! */
 534                                         m_freem(mp);
 535                                 }
 536                                 m = NULL; /* ipip_output() already freed it */
 537                                 goto bad;
 538                         }
 539                         m = mp, mp = NULL;
 540                         /*
 541                          * ipip_output clears IP_DF in the new header.  If
 542                          * we need to propagate IP_DF from the outer header,
 543                          * then we have to do it here.
 544                          *
 545                          * XXX shouldn't assume what ipip_output does.
 546                          */
 547                         if (dst->sa.sa_family == AF_INET && setdf) {
 548                                 if (m->m_len < sizeof (struct ip) &&
 549                                     (m = m_pullup(m, sizeof (struct ip))) == NULL) {
 550                                         error = ENOBUFS;
 551                                         goto bad;
 552                                 }
 553                                 ip = mtod(m, struct ip *);
 554                                 ip->ip_off = ntohs(ip->ip_off);
 555                                 ip->ip_off |= IP_DF;
 556                                 ip->ip_off = htons(ip->ip_off);
 557                         }
 558                 }
 559         }
 560
 561 #ifdef DEV_ENC
 562         /* pass the mbuf to enc0 for bpf processing */
 563         ipsec_bpf(m, sav, AF_INET, ENC_OUT|ENC_AFTER);
 564         /* pass the mbuf to enc0 for packet filtering */
 565         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 566                 goto bad;
 567 #endif
 568
 569         /*
 570          * Dispatch to the appropriate IPsec transform logic.  The
 571          * packet will be returned for transmission after crypto
 572          * processing, etc. are completed.  For encapsulation we
 573          * bypass this call because of the explicit call done above
 574          * (necessary to deal with IP_DF handling for IPv4).
 575          *
 576          * NB: m & sav are ``passed to caller'' who's reponsible for
 577          *     for reclaiming their resources.
 578          */
 579         if (sav->tdb_xform->xf_type != XF_IP4) {
 580                 ip = mtod(m, struct ip *);
 581                 i = ip->ip_hl << 2;
 582                 off = offsetof(struct ip, ip_p);
 583                 error = (*sav->tdb_xform->xf_output)(m, isr, NULL, i, off);
 584         } else {
 585                 error = ipsec_process_done(m, isr);
 586         }
 587         IPSECREQUEST_UNLOCK(isr);
 588         return error;
 589 bad:
 590         if (isr)
 591                 IPSECREQUEST_UNLOCK(isr);
 592         if (m)
 593                 m_freem(m);
 594         return error;
 595 }
 596 #endif
 597
 598 #ifdef INET6
 599 /*
 600  * Chop IP6 header from the payload.
 601  */
 602 static struct mbuf *
 603 ipsec6_splithdr(struct mbuf *m)
 604 {
 605         struct mbuf *mh;
 606         struct ip6_hdr *ip6;
 607         int hlen;
 608
 609         IPSEC_ASSERT(m->m_len >= sizeof (struct ip6_hdr),
 610                 ("first mbuf too short, len %u", m->m_len));
 611         ip6 = mtod(m, struct ip6_hdr *);
 612         hlen = sizeof(struct ip6_hdr);
 613         if (m->m_len > hlen) {
 614                 MGETHDR(mh, M_DONTWAIT, MT_DATA);
 615                 if (!mh) {
 616                         m_freem(m);
 617                         return NULL;
 618                 }
 619                 M_MOVE_PKTHDR(mh, m);
 620                 MH_ALIGN(mh, hlen);
 621                 m->m_len -= hlen;
 622                 m->m_data += hlen;
 623                 mh->m_next = m;
 624                 m = mh;
 625                 m->m_len = hlen;
 626                 bcopy((caddr_t)ip6, mtod(m, caddr_t), hlen);
 627         } else if (m->m_len < hlen) {
 628                 m = m_pullup(m, hlen);
 629                 if (!m)
 630                         return NULL;
 631         }
 632         return m;
 633 }
 634
 635 /*
 636  * IPsec output logic for IPv6, transport mode.
 637  */
 638 int
 639 ipsec6_output_trans(
 640         struct ipsec_output_state *state,
 641         u_char *nexthdrp,
 642         struct mbuf *mprev,
 643         struct secpolicy *sp,
 644         int flags,
 645         int *tun)
 646 {
 647         struct ipsecrequest *isr;
 648         struct secasindex saidx;
 649         int error = 0;
 650         struct mbuf *m;
 651
 652         IPSEC_ASSERT(state != NULL, ("null state"));
 653         IPSEC_ASSERT(state->m != NULL, ("null m"));
 654         IPSEC_ASSERT(nexthdrp != NULL, ("null nexthdrp"));
 655         IPSEC_ASSERT(mprev != NULL, ("null mprev"));
 656         IPSEC_ASSERT(sp != NULL, ("null sp"));
 657         IPSEC_ASSERT(tun != NULL, ("null tun"));
 658
 659         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 660                 printf("%s: applied SP\n", __func__);
 661                 kdebug_secpolicy(sp));
 662
 663         isr = sp->req;
 664         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 665                 /* the rest will be handled by ipsec6_output_tunnel() */
 666                 *tun = 1;               /* need tunnel-mode processing */
 667                 return 0;
 668         }
 669
 670         *tun = 0;
 671         m = state->m;
 672
 673         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 674         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 675         if (isr == NULL) {
 676                 if (error != 0) {
 677 #ifdef notdef
 678                         /* XXX should notification be done for all errors ? */
 679                         /*
 680                          * Notify the fact that the packet is discarded
 681                          * to ourselves. I believe this is better than
 682                          * just silently discarding. (jinmei@kame.net)
 683                          * XXX: should we restrict the error to TCP packets?
 684                          * XXX: should we directly notify sockets via
 685                          *      pfctlinputs?
 686                          */
 687                         icmp6_error(m, ICMP6_DST_UNREACH,
 688                                     ICMP6_DST_UNREACH_ADMIN, 0);
 689                         m = NULL;       /* NB: icmp6_error frees mbuf */
 690 #endif
 691                         goto bad;
 692                 }
 693                 return EJUSTRETURN;
 694         }
 695
 696         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 697                                                   sizeof (struct ip6_hdr),
 698                                                   offsetof(struct ip6_hdr,
 699                                                            ip6_nxt));
 700         IPSECREQUEST_UNLOCK(isr);
 701         return error;
 702 bad:
 703         if (isr)
 704                 IPSECREQUEST_UNLOCK(isr);
 705         if (m)
 706                 m_freem(m);
 707         state->m = NULL;
 708         return error;
 709 }
 710
 711 static int
 712 ipsec6_encapsulate(struct mbuf *m, struct secasvar *sav)
 713 {
 714         struct ip6_hdr *oip6;
 715         struct ip6_hdr *ip6;
 716         size_t plen;
 717
 718         /* can't tunnel between different AFs */
 719         if (sav->sah->saidx.src.sa.sa_family != AF_INET6 ||
 720             sav->sah->saidx.dst.sa.sa_family != AF_INET6) {
 721                 m_freem(m);
 722                 return EINVAL;
 723         }
 724         IPSEC_ASSERT(m->m_len == sizeof (struct ip6_hdr),
 725                 ("mbuf wrong size; len %u", m->m_len));
 726
 727
 728         /*
 729          * grow the mbuf to accomodate the new IPv6 header.
 730          */
 731         plen = m->m_pkthdr.len;
 732         if (M_LEADINGSPACE(m->m_next) < sizeof(struct ip6_hdr)) {
 733                 struct mbuf *n;
 734                 MGET(n, M_DONTWAIT, MT_DATA);
 735                 if (!n) {
 736                         m_freem(m);
 737                         return ENOBUFS;
 738                 }
 739                 n->m_len = sizeof(struct ip6_hdr);
 740                 n->m_next = m->m_next;
 741                 m->m_next = n;
 742                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 743                 oip6 = mtod(n, struct ip6_hdr *);
 744         } else {
 745                 m->m_next->m_len += sizeof(struct ip6_hdr);
 746                 m->m_next->m_data -= sizeof(struct ip6_hdr);
 747                 m->m_pkthdr.len += sizeof(struct ip6_hdr);
 748                 oip6 = mtod(m->m_next, struct ip6_hdr *);
 749         }
 750         ip6 = mtod(m, struct ip6_hdr *);
 751         bcopy((caddr_t)ip6, (caddr_t)oip6, sizeof(struct ip6_hdr));
 752
 753         /* Fake link-local scope-class addresses */
 754         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_src))
 755                 oip6->ip6_src.s6_addr16[1] = 0;
 756         if (IN6_IS_SCOPE_LINKLOCAL(&oip6->ip6_dst))
 757                 oip6->ip6_dst.s6_addr16[1] = 0;
 758
 759         /* construct new IPv6 header. see RFC 2401 5.1.2.2 */
 760         /* ECN consideration. */
 761         ip6_ecn_ingress(V_ip6_ipsec_ecn, &ip6->ip6_flow, &oip6->ip6_flow);
 762         if (plen < IPV6_MAXPACKET - sizeof(struct ip6_hdr))
 763                 ip6->ip6_plen = htons(plen);
 764         else {
 765                 /* ip6->ip6_plen will be updated in ip6_output() */
 766         }
 767         ip6->ip6_nxt = IPPROTO_IPV6;
 768         ip6->ip6_src = sav->sah->saidx.src.sin6.sin6_addr;
 769         ip6->ip6_dst = sav->sah->saidx.dst.sin6.sin6_addr;
 770         ip6->ip6_hlim = IPV6_DEFHLIM;
 771
 772         /* XXX Should ip6_src be updated later ? */
 773
 774         return 0;
 775 }
 776
 777 /*
 778  * IPsec output logic for IPv6, tunnel mode.
 779  */
 780 int
 781 ipsec6_output_tunnel(struct ipsec_output_state *state, struct secpolicy *sp, int flags)
 782 {
 783         struct ip6_hdr *ip6;
 784         struct ipsecrequest *isr;
 785         struct secasindex saidx;
 786         int error;
 787         struct sockaddr_in6 *dst6;
 788         struct mbuf *m;
 789
 790         IPSEC_ASSERT(state != NULL, ("null state"));
 791         IPSEC_ASSERT(state->m != NULL, ("null m"));
 792         IPSEC_ASSERT(sp != NULL, ("null sp"));
 793
 794         KEYDEBUG(KEYDEBUG_IPSEC_DATA,
 795                 printf("%s: applied SP\n", __func__);
 796                 kdebug_secpolicy(sp));
 797
 798         m = state->m;
 799         /*
 800          * transport mode ipsec (before the 1st tunnel mode) is already
 801          * processed by ipsec6_output_trans().
 802          */
 803         for (isr = sp->req; isr; isr = isr->next) {
 804                 if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
 805                         break;
 806         }
 807
 808         IPSECREQUEST_LOCK(isr);         /* insure SA contents don't change */
 809         isr = ipsec_nextisr(m, isr, AF_INET6, &saidx, &error);
 810         if (isr == NULL) {
 811                 if (error != 0)
 812                         goto bad;
 813                 return EJUSTRETURN;
 814         }
 815
 816 #ifdef DEV_ENC
 817         encif->if_opackets++;
 818         encif->if_obytes += m->m_pkthdr.len;
 819
 820         /* pass the mbuf to enc0 for bpf processing */
 821         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_BEFORE);
 822         /* pass the mbuf to enc0 for packet filtering */
 823         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_BEFORE)) != 0)
 824                 goto bad;
 825 #endif
 826
 827         /*
 828          * There may be the case that SA status will be changed when
 829          * we are refering to one. So calling splsoftnet().
 830          */
 831         if (isr->saidx.mode == IPSEC_MODE_TUNNEL) {
 832                 /*
 833                  * build IPsec tunnel.
 834                  */
 835                 /* XXX should be processed with other familiy */
 836                 if (isr->sav->sah->saidx.src.sa.sa_family != AF_INET6) {
 837                         ipseclog((LOG_ERR, "%s: family mismatched between "
 838                             "inner and outer, spi=%u\n", __func__,
 839                             ntohl(isr->sav->spi)));
 840                         IPSEC6STAT_INC(ips_out_inval);
 841                         error = EAFNOSUPPORT;
 842                         goto bad;
 843                 }
 844
 845                 m = ipsec6_splithdr(m);
 846                 if (!m) {
 847                         IPSEC6STAT_INC(ips_out_nomem);
 848                         error = ENOMEM;
 849                         goto bad;
 850                 }
 851                 error = ipsec6_encapsulate(m, isr->sav);
 852                 if (error) {
 853                         m = NULL;
 854                         goto bad;
 855                 }
 856                 ip6 = mtod(m, struct ip6_hdr *);
 857
 858                 state->ro =
 859                     (struct route *)&isr->sav->sah->route_cache.sin6_route;
 860                 state->dst = (struct sockaddr *)&state->ro->ro_dst;
 861                 dst6 = (struct sockaddr_in6 *)state->dst;
 862                 if (state->ro->ro_rt
 863                  && ((state->ro->ro_rt->rt_flags & RTF_UP) == 0
 864                   || !IN6_ARE_ADDR_EQUAL(&dst6->sin6_addr, &ip6->ip6_dst))) {
 865                         RTFREE(state->ro->ro_rt);
 866                         state->ro->ro_rt = NULL;
 867                 }
 868                 if (state->ro->ro_rt == NULL) {
 869                         bzero(dst6, sizeof(*dst6));
 870                         dst6->sin6_family = AF_INET6;
 871                         dst6->sin6_len = sizeof(*dst6);
 872                         dst6->sin6_addr = ip6->ip6_dst;
 873                         rtalloc_ign_fib(state->ro, 0UL, M_GETFIB(m));
 874                 }
 875                 if (state->ro->ro_rt == NULL) {
 876                         IP6STAT_INC(ip6s_noroute);
 877                         IPSEC6STAT_INC(ips_out_noroute);
 878                         error = EHOSTUNREACH;
 879                         goto bad;
 880                 }
 881
 882                 /* adjust state->dst if tunnel endpoint is offlink */
 883                 if (state->ro->ro_rt->rt_flags & RTF_GATEWAY)
 884                         state->dst = (struct sockaddr *)state->ro->ro_rt->rt_gateway;
 885         }
 886
 887         m = ipsec6_splithdr(m);
 888         if (!m) {
 889                 IPSEC6STAT_INC(ips_out_nomem);
 890                 error = ENOMEM;
 891                 goto bad;
 892         }
 893         ip6 = mtod(m, struct ip6_hdr *);
 894
 895 #ifdef DEV_ENC
 896         /* pass the mbuf to enc0 for bpf processing */
 897         ipsec_bpf(m, isr->sav, AF_INET6, ENC_OUT|ENC_AFTER);
 898         /* pass the mbuf to enc0 for packet filtering */
 899         if ((error = ipsec_filter(&m, PFIL_OUT, ENC_OUT|ENC_AFTER)) != 0)
 900                 goto bad;
 901 #endif
 902
 903         error = (*isr->sav->tdb_xform->xf_output)(m, isr, NULL,
 904                 sizeof (struct ip6_hdr),
 905                 offsetof(struct ip6_hdr, ip6_nxt));
 906         IPSECREQUEST_UNLOCK(isr);
 907         return error;
 908 bad:
 909         if (isr)
 910                 IPSECREQUEST_UNLOCK(isr);
 911         if (m)
 912                 m_freem(m);
 913         state->m = NULL;
 914         return error;
 915 }
 916 #endif /*INET6*/