sys/netinet6/nd6.c

   1 /*      $FreeBSD$       */
   2 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   3
   4 /*-
   5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of the project nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 #include "opt_inet.h"
  34 #include "opt_inet6.h"
  35 #include "opt_mac.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/callout.h>
  40 #include <sys/mac.h>
  41 #include <sys/malloc.h>
  42 #include <sys/mbuf.h>
  43 #include <sys/socket.h>
  44 #include <sys/sockio.h>
  45 #include <sys/time.h>
  46 #include <sys/kernel.h>
  47 #include <sys/protosw.h>
  48 #include <sys/errno.h>
  49 #include <sys/syslog.h>
  50 #include <sys/queue.h>
  51 #include <sys/sysctl.h>
  52
  53 #include <net/if.h>
  54 #include <net/if_arc.h>
  55 #include <net/if_dl.h>
  56 #include <net/if_types.h>
  57 #include <net/iso88025.h>
  58 #include <net/fddi.h>
  59 #include <net/route.h>
  60
  61 #include <netinet/in.h>
  62 #include <netinet/if_ether.h>
  63 #include <netinet6/in6_var.h>
  64 #include <netinet/ip6.h>
  65 #include <netinet6/ip6_var.h>
  66 #include <netinet6/scope6_var.h>
  67 #include <netinet6/nd6.h>
  68 #include <netinet/icmp6.h>
  69
  70 #include <net/net_osdep.h>
  71
  72 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  73 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  74
  75 #define SIN6(s) ((struct sockaddr_in6 *)s)
  76 #define SDL(s) ((struct sockaddr_dl *)s)
  77
  78 /* timer values */
  79 int     nd6_prune       = 1;    /* walk list every 1 seconds */
  80 int     nd6_delay       = 5;    /* delay first probe time 5 second */
  81 int     nd6_umaxtries   = 3;    /* maximum unicast query */
  82 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
  83 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
  84 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
  85
  86 /* preventing too many loops in ND option parsing */
  87 int nd6_maxndopt = 10;  /* max # of ND options allowed */
  88
  89 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
  90
  91 #ifdef ND6_DEBUG
  92 int nd6_debug = 1;
  93 #else
  94 int nd6_debug = 0;
  95 #endif
  96
  97 /* for debugging? */
  98 static int nd6_inuse, nd6_allocated;
  99
 100 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 101 struct nd_drhead nd_defrouter;
 102 struct nd_prhead nd_prefix = { 0 };
 103
 104 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 105 static struct sockaddr_in6 all1_sa;
 106
 107 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
 108         struct ifnet *));
 109 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *));
 110 static void nd6_slowtimo __P((void *));
 111 static int regen_tmpaddr __P((struct in6_ifaddr *));
 112
 113 struct callout nd6_slowtimo_ch;
 114 struct callout nd6_timer_ch;
 115 extern struct callout in6_tmpaddrtimer_ch;
 116
 117 void
 118 nd6_init()
 119 {
 120         static int nd6_init_done = 0;
 121         int i;
 122
 123         if (nd6_init_done) {
 124                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 125                 return;
 126         }
 127
 128         all1_sa.sin6_family = AF_INET6;
 129         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 130         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 131                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 132
 133         /* initialization of the default router list */
 134         TAILQ_INIT(&nd_defrouter);
 135
 136         nd6_init_done = 1;
 137
 138         /* start timer */
 139         callout_init(&nd6_slowtimo_ch, 0);
 140         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 141             nd6_slowtimo, NULL);
 142 }
 143
 144 struct nd_ifinfo *
 145 nd6_ifattach(ifp)
 146         struct ifnet *ifp;
 147 {
 148         struct nd_ifinfo *nd;
 149
 150         nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
 151         bzero(nd, sizeof(*nd));
 152
 153         nd->initialized = 1;
 154
 155         nd->chlim = IPV6_DEFHLIM;
 156         nd->basereachable = REACHABLE_TIME;
 157         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 158         nd->retrans = RETRANS_TIMER;
 159         /*
 160          * Note that the default value of ip6_accept_rtadv is 0, which means
 161          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 162          * here.
 163          */
 164         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 165
 166         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 167         nd6_setmtu0(ifp, nd);
 168
 169         return nd;
 170 }
 171
 172 void
 173 nd6_ifdetach(nd)
 174         struct nd_ifinfo *nd;
 175 {
 176
 177         free(nd, M_IP6NDP);
 178 }
 179
 180 /*
 181  * Reset ND level link MTU. This function is called when the physical MTU
 182  * changes, which means we might have to adjust the ND level MTU.
 183  */
 184 void
 185 nd6_setmtu(ifp)
 186         struct ifnet *ifp;
 187 {
 188
 189         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 190 }
 191
 192 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 193 void
 194 nd6_setmtu0(ifp, ndi)
 195         struct ifnet *ifp;
 196         struct nd_ifinfo *ndi;
 197 {
 198         u_int32_t omaxmtu;
 199
 200         omaxmtu = ndi->maxmtu;
 201
 202         switch (ifp->if_type) {
 203         case IFT_ARCNET:
 204                 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
 205                 break;
 206         case IFT_FDDI:
 207                 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
 208                 break;
 209         case IFT_ISO88025:
 210                  ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
 211                  break;
 212         default:
 213                 ndi->maxmtu = ifp->if_mtu;
 214                 break;
 215         }
 216
 217         /*
 218          * Decreasing the interface MTU under IPV6 minimum MTU may cause
 219          * undesirable situation.  We thus notify the operator of the change
 220          * explicitly.  The check for omaxmtu is necessary to restrict the
 221          * log to the case of changing the MTU, not initializing it.
 222          */
 223         if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
 224                 log(LOG_NOTICE, "nd6_setmtu0: "
 225                     "new link MTU on %s (%lu) is too small for IPv6\n",
 226                     if_name(ifp), (unsigned long)ndi->maxmtu);
 227         }
 228
 229         if (ndi->maxmtu > in6_maxmtu)
 230                 in6_setmaxmtu(); /* check all interfaces just in case */
 231
 232 #undef MIN
 233 }
 234
 235 void
 236 nd6_option_init(opt, icmp6len, ndopts)
 237         void *opt;
 238         int icmp6len;
 239         union nd_opts *ndopts;
 240 {
 241
 242         bzero(ndopts, sizeof(*ndopts));
 243         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 244         ndopts->nd_opts_last
 245                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 246
 247         if (icmp6len == 0) {
 248                 ndopts->nd_opts_done = 1;
 249                 ndopts->nd_opts_search = NULL;
 250         }
 251 }
 252
 253 /*
 254  * Take one ND option.
 255  */
 256 struct nd_opt_hdr *
 257 nd6_option(ndopts)
 258         union nd_opts *ndopts;
 259 {
 260         struct nd_opt_hdr *nd_opt;
 261         int olen;
 262
 263         if (ndopts == NULL)
 264                 panic("ndopts == NULL in nd6_option");
 265         if (ndopts->nd_opts_last == NULL)
 266                 panic("uninitialized ndopts in nd6_option");
 267         if (ndopts->nd_opts_search == NULL)
 268                 return NULL;
 269         if (ndopts->nd_opts_done)
 270                 return NULL;
 271
 272         nd_opt = ndopts->nd_opts_search;
 273
 274         /* make sure nd_opt_len is inside the buffer */
 275         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 276                 bzero(ndopts, sizeof(*ndopts));
 277                 return NULL;
 278         }
 279
 280         olen = nd_opt->nd_opt_len << 3;
 281         if (olen == 0) {
 282                 /*
 283                  * Message validation requires that all included
 284                  * options have a length that is greater than zero.
 285                  */
 286                 bzero(ndopts, sizeof(*ndopts));
 287                 return NULL;
 288         }
 289
 290         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 291         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 292                 /* option overruns the end of buffer, invalid */
 293                 bzero(ndopts, sizeof(*ndopts));
 294                 return NULL;
 295         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 296                 /* reached the end of options chain */
 297                 ndopts->nd_opts_done = 1;
 298                 ndopts->nd_opts_search = NULL;
 299         }
 300         return nd_opt;
 301 }
 302
 303 /*
 304  * Parse multiple ND options.
 305  * This function is much easier to use, for ND routines that do not need
 306  * multiple options of the same type.
 307  */
 308 int
 309 nd6_options(ndopts)
 310         union nd_opts *ndopts;
 311 {
 312         struct nd_opt_hdr *nd_opt;
 313         int i = 0;
 314
 315         if (ndopts == NULL)
 316                 panic("ndopts == NULL in nd6_options");
 317         if (ndopts->nd_opts_last == NULL)
 318                 panic("uninitialized ndopts in nd6_options");
 319         if (ndopts->nd_opts_search == NULL)
 320                 return 0;
 321
 322         while (1) {
 323                 nd_opt = nd6_option(ndopts);
 324                 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
 325                         /*
 326                          * Message validation requires that all included
 327                          * options have a length that is greater than zero.
 328                          */
 329                         icmp6stat.icp6s_nd_badopt++;
 330                         bzero(ndopts, sizeof(*ndopts));
 331                         return -1;
 332                 }
 333
 334                 if (nd_opt == NULL)
 335                         goto skip1;
 336
 337                 switch (nd_opt->nd_opt_type) {
 338                 case ND_OPT_SOURCE_LINKADDR:
 339                 case ND_OPT_TARGET_LINKADDR:
 340                 case ND_OPT_MTU:
 341                 case ND_OPT_REDIRECTED_HEADER:
 342                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 343                                 nd6log((LOG_INFO,
 344                                     "duplicated ND6 option found (type=%d)\n",
 345                                     nd_opt->nd_opt_type));
 346                                 /* XXX bark? */
 347                         } else {
 348                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 349                                         = nd_opt;
 350                         }
 351                         break;
 352                 case ND_OPT_PREFIX_INFORMATION:
 353                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 354                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 355                                         = nd_opt;
 356                         }
 357                         ndopts->nd_opts_pi_end =
 358                                 (struct nd_opt_prefix_info *)nd_opt;
 359                         break;
 360                 default:
 361                         /*
 362                          * Unknown options must be silently ignored,
 363                          * to accomodate future extension to the protocol.
 364                          */
 365                         nd6log((LOG_DEBUG,
 366                             "nd6_options: unsupported option %d - "
 367                             "option ignored\n", nd_opt->nd_opt_type));
 368                 }
 369
 370 skip1:
 371                 i++;
 372                 if (i > nd6_maxndopt) {
 373                         icmp6stat.icp6s_nd_toomanyopt++;
 374                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 375                         break;
 376                 }
 377
 378                 if (ndopts->nd_opts_done)
 379                         break;
 380         }
 381
 382         return 0;
 383 }
 384
 385 /*
 386  * ND6 timer routine to expire default route list and prefix list
 387  */
 388 void
 389 nd6_timer(ignored_arg)
 390         void    *ignored_arg;
 391 {
 392         int s;
 393         struct llinfo_nd6 *ln;
 394         struct nd_defrouter *dr;
 395         struct nd_prefix *pr;
 396         struct ifnet *ifp;
 397         struct in6_ifaddr *ia6, *nia6;
 398         struct in6_addrlifetime *lt6;
 399
 400         s = splnet();
 401         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 402             nd6_timer, NULL);
 403
 404         ln = llinfo_nd6.ln_next;
 405         while (ln && ln != &llinfo_nd6) {
 406                 struct rtentry *rt;
 407                 struct sockaddr_in6 *dst;
 408                 struct llinfo_nd6 *next = ln->ln_next;
 409                 /* XXX: used for the DELAY case only: */
 410                 struct nd_ifinfo *ndi = NULL;
 411
 412                 if ((rt = ln->ln_rt) == NULL) {
 413                         ln = next;
 414                         continue;
 415                 }
 416                 if ((ifp = rt->rt_ifp) == NULL) {
 417                         ln = next;
 418                         continue;
 419                 }
 420                 ndi = ND_IFINFO(ifp);
 421                 dst = (struct sockaddr_in6 *)rt_key(rt);
 422
 423                 if (ln->ln_expire > time_second) {
 424                         ln = next;
 425                         continue;
 426                 }
 427
 428                 /* sanity check */
 429                 if (!rt)
 430                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 431                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 432                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 433                               rt->rt_llinfo, ln);
 434                 if (!dst)
 435                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 436
 437                 switch (ln->ln_state) {
 438                 case ND6_LLINFO_INCOMPLETE:
 439                         if (ln->ln_asked < nd6_mmaxtries) {
 440                                 ln->ln_asked++;
 441                                 ln->ln_expire = time_second +
 442                                         ND_IFINFO(ifp)->retrans / 1000;
 443                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 444                                         ln, 0);
 445                         } else {
 446                                 struct mbuf *m = ln->ln_hold;
 447                                 if (m) {
 448                                         /*
 449                                          * assuming every packet in ln_hold has
 450                                          * the same IP header
 451                                          */
 452                                         ln->ln_hold = NULL;
 453                                         icmp6_error2(m, ICMP6_DST_UNREACH,
 454                                                      ICMP6_DST_UNREACH_ADDR, 0,
 455                                                      rt->rt_ifp);
 456                                 }
 457                                 next = nd6_free(rt);
 458                         }
 459                         break;
 460                 case ND6_LLINFO_REACHABLE:
 461                         if (ln->ln_expire) {
 462                                 ln->ln_state = ND6_LLINFO_STALE;
 463                                 ln->ln_expire = time_second + nd6_gctimer;
 464                         }
 465                         break;
 466
 467                 case ND6_LLINFO_STALE:
 468                         /* Garbage Collection(RFC 2461 5.3) */
 469                         if (ln->ln_expire)
 470                                 next = nd6_free(rt);
 471                         break;
 472
 473                 case ND6_LLINFO_DELAY:
 474                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
 475                                 /* We need NUD */
 476                                 ln->ln_asked = 1;
 477                                 ln->ln_state = ND6_LLINFO_PROBE;
 478                                 ln->ln_expire = time_second +
 479                                         ndi->retrans / 1000;
 480                                 nd6_ns_output(ifp, &dst->sin6_addr,
 481                                               &dst->sin6_addr,
 482                                               ln, 0);
 483                         } else {
 484                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 485                                 ln->ln_expire = time_second + nd6_gctimer;
 486                         }
 487                         break;
 488                 case ND6_LLINFO_PROBE:
 489                         if (ln->ln_asked < nd6_umaxtries) {
 490                                 ln->ln_asked++;
 491                                 ln->ln_expire = time_second +
 492                                         ND_IFINFO(ifp)->retrans / 1000;
 493                                 nd6_ns_output(ifp, &dst->sin6_addr,
 494                                                &dst->sin6_addr, ln, 0);
 495                         } else {
 496                                 next = nd6_free(rt);
 497                         }
 498                         break;
 499                 }
 500                 ln = next;
 501         }
 502
 503         /* expire default router list */
 504         dr = TAILQ_FIRST(&nd_defrouter);
 505         while (dr) {
 506                 if (dr->expire && dr->expire < time_second) {
 507                         struct nd_defrouter *t;
 508                         t = TAILQ_NEXT(dr, dr_entry);
 509                         defrtrlist_del(dr);
 510                         dr = t;
 511                 } else {
 512                         dr = TAILQ_NEXT(dr, dr_entry);
 513                 }
 514         }
 515
 516         /*
 517          * expire interface addresses.
 518          * in the past the loop was inside prefix expiry processing.
 519          * However, from a stricter speci-confrmance standpoint, we should
 520          * rather separate address lifetimes and prefix lifetimes.
 521          */
 522   addrloop:
 523         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 524                 nia6 = ia6->ia_next;
 525                 /* check address lifetime */
 526                 lt6 = &ia6->ia6_lifetime;
 527                 if (IFA6_IS_INVALID(ia6)) {
 528                         int regen = 0;
 529
 530                         /*
 531                          * If the expiring address is temporary, try
 532                          * regenerating a new one.  This would be useful when
 533                          * we suspended a laptop PC, then turned it on after a
 534                          * period that could invalidate all temporary
 535                          * addresses.  Although we may have to restart the
 536                          * loop (see below), it must be after purging the
 537                          * address.  Otherwise, we'd see an infinite loop of
 538                          * regeneration.
 539                          */
 540                         if (ip6_use_tempaddr &&
 541                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
 542                                 if (regen_tmpaddr(ia6) == 0)
 543                                         regen = 1;
 544                         }
 545
 546                         in6_purgeaddr(&ia6->ia_ifa);
 547
 548                         if (regen)
 549                                 goto addrloop; /* XXX: see below */
 550                 }
 551                 if (IFA6_IS_DEPRECATED(ia6)) {
 552                         int oldflags = ia6->ia6_flags;
 553
 554                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 555
 556                         /*
 557                          * If a temporary address has just become deprecated,
 558                          * regenerate a new one if possible.
 559                          */
 560                         if (ip6_use_tempaddr &&
 561                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 562                             (oldflags & IN6_IFF_DEPRECATED) == 0) {
 563
 564                                 if (regen_tmpaddr(ia6) == 0) {
 565                                         /*
 566                                          * A new temporary address is
 567                                          * generated.
 568                                          * XXX: this means the address chain
 569                                          * has changed while we are still in
 570                                          * the loop.  Although the change
 571                                          * would not cause disaster (because
 572                                          * it's not a deletion, but an
 573                                          * addition,) we'd rather restart the
 574                                          * loop just for safety.  Or does this
 575                                          * significantly reduce performance??
 576                                          */
 577                                         goto addrloop;
 578                                 }
 579                         }
 580                 } else {
 581                         /*
 582                          * A new RA might have made a deprecated address
 583                          * preferred.
 584                          */
 585                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 586                 }
 587         }
 588
 589         /* expire prefix list */
 590         pr = nd_prefix.lh_first;
 591         while (pr) {
 592                 /*
 593                  * check prefix lifetime.
 594                  * since pltime is just for autoconf, pltime processing for
 595                  * prefix is not necessary.
 596                  */
 597                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
 598                         struct nd_prefix *t;
 599                         t = pr->ndpr_next;
 600
 601                         /*
 602                          * address expiration and prefix expiration are
 603                          * separate.  NEVER perform in6_purgeaddr here.
 604                          */
 605
 606                         prelist_remove(pr);
 607                         pr = t;
 608                 } else
 609                         pr = pr->ndpr_next;
 610         }
 611         splx(s);
 612 }
 613
 614 static int
 615 regen_tmpaddr(ia6)
 616         struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
 617 {
 618         struct ifaddr *ifa;
 619         struct ifnet *ifp;
 620         struct in6_ifaddr *public_ifa6 = NULL;
 621
 622         ifp = ia6->ia_ifa.ifa_ifp;
 623         for (ifa = ifp->if_addrlist.tqh_first; ifa;
 624              ifa = ifa->ifa_list.tqe_next) {
 625                 struct in6_ifaddr *it6;
 626
 627                 if (ifa->ifa_addr->sa_family != AF_INET6)
 628                         continue;
 629
 630                 it6 = (struct in6_ifaddr *)ifa;
 631
 632                 /* ignore no autoconf addresses. */
 633                 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
 634                         continue;
 635
 636                 /* ignore autoconf addresses with different prefixes. */
 637                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 638                         continue;
 639
 640                 /*
 641                  * Now we are looking at an autoconf address with the same
 642                  * prefix as ours.  If the address is temporary and is still
 643                  * preferred, do not create another one.  It would be rare, but
 644                  * could happen, for example, when we resume a laptop PC after
 645                  * a long period.
 646                  */
 647                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 648                     !IFA6_IS_DEPRECATED(it6)) {
 649                         public_ifa6 = NULL;
 650                         break;
 651                 }
 652
 653                 /*
 654                  * This is a public autoconf address that has the same prefix
 655                  * as ours.  If it is preferred, keep it.  We can't break the
 656                  * loop here, because there may be a still-preferred temporary
 657                  * address with the prefix.
 658                  */
 659                 if (!IFA6_IS_DEPRECATED(it6))
 660                     public_ifa6 = it6;
 661         }
 662
 663         if (public_ifa6 != NULL) {
 664                 int e;
 665
 666                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 667                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 668                             " tmp addr,errno=%d\n", e);
 669                         return (-1);
 670                 }
 671                 return (0);
 672         }
 673
 674         return (-1);
 675 }
 676
 677 /*
 678  * Nuke neighbor cache/prefix/default router management table, right before
 679  * ifp goes away.
 680  */
 681 void
 682 nd6_purge(ifp)
 683         struct ifnet *ifp;
 684 {
 685         struct llinfo_nd6 *ln, *nln;
 686         struct nd_defrouter *dr, *ndr, drany;
 687         struct nd_prefix *pr, *npr;
 688
 689         /* Nuke default router list entries toward ifp */
 690         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 691                 /*
 692                  * The first entry of the list may be stored in
 693                  * the routing table, so we'll delete it later.
 694                  */
 695                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 696                         ndr = TAILQ_NEXT(dr, dr_entry);
 697                         if (dr->ifp == ifp)
 698                                 defrtrlist_del(dr);
 699                 }
 700                 dr = TAILQ_FIRST(&nd_defrouter);
 701                 if (dr->ifp == ifp)
 702                         defrtrlist_del(dr);
 703         }
 704
 705         /* Nuke prefix list entries toward ifp */
 706         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 707                 npr = pr->ndpr_next;
 708                 if (pr->ndpr_ifp == ifp) {
 709                         /*
 710                          * Previously, pr->ndpr_addr is removed as well,
 711                          * but I strongly believe we don't have to do it.
 712                          * nd6_purge() is only called from in6_ifdetach(),
 713                          * which removes all the associated interface addresses
 714                          * by itself.
 715                          * (jinmei@kame.net 20010129)
 716                          */
 717                         prelist_remove(pr);
 718                 }
 719         }
 720
 721         /* cancel default outgoing interface setting */
 722         if (nd6_defifindex == ifp->if_index)
 723                 nd6_setdefaultiface(0);
 724
 725         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 726                 /* refresh default router list */
 727                 bzero(&drany, sizeof(drany));
 728                 defrouter_delreq(&drany, 0);
 729                 defrouter_select();
 730         }
 731
 732         /*
 733          * Nuke neighbor cache entries for the ifp.
 734          * Note that rt->rt_ifp may not be the same as ifp,
 735          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 736          * nd6_rtrequest(), and ip6_input().
 737          */
 738         ln = llinfo_nd6.ln_next;
 739         while (ln && ln != &llinfo_nd6) {
 740                 struct rtentry *rt;
 741                 struct sockaddr_dl *sdl;
 742
 743                 nln = ln->ln_next;
 744                 rt = ln->ln_rt;
 745                 if (rt && rt->rt_gateway &&
 746                     rt->rt_gateway->sa_family == AF_LINK) {
 747                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 748                         if (sdl->sdl_index == ifp->if_index)
 749                                 nln = nd6_free(rt);
 750                 }
 751                 ln = nln;
 752         }
 753 }
 754
 755 struct rtentry *
 756 nd6_lookup(addr6, create, ifp)
 757         struct in6_addr *addr6;
 758         int create;
 759         struct ifnet *ifp;
 760 {
 761         struct rtentry *rt;
 762         struct sockaddr_in6 sin6;
 763
 764         bzero(&sin6, sizeof(sin6));
 765         sin6.sin6_len = sizeof(struct sockaddr_in6);
 766         sin6.sin6_family = AF_INET6;
 767         sin6.sin6_addr = *addr6;
 768         rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
 769         if (rt) {
 770                 if ((rt->rt_flags & RTF_LLINFO) == 0 && create) {
 771                         /*
 772                          * This is the case for the default route.
 773                          * If we want to create a neighbor cache for the
 774                          * address, we should free the route for the
 775                          * destination and allocate an interface route.
 776                          */
 777                         RTFREE_LOCKED(rt);
 778                         rt = NULL;
 779                 }
 780         }
 781         if (rt == NULL) {
 782                 if (create && ifp) {
 783                         int e;
 784
 785                         /*
 786                          * If no route is available and create is set,
 787                          * we allocate a host route for the destination
 788                          * and treat it like an interface route.
 789                          * This hack is necessary for a neighbor which can't
 790                          * be covered by our own prefix.
 791                          */
 792                         struct ifaddr *ifa =
 793                             ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 794                         if (ifa == NULL)
 795                                 return (NULL);
 796
 797                         /*
 798                          * Create a new route.  RTF_LLINFO is necessary
 799                          * to create a Neighbor Cache entry for the
 800                          * destination in nd6_rtrequest which will be
 801                          * called in rtrequest via ifa->ifa_rtrequest.
 802                          */
 803                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 804                             ifa->ifa_addr, (struct sockaddr *)&all1_sa,
 805                             (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
 806                             ~RTF_CLONING, &rt)) != 0) {
 807                                 log(LOG_ERR,
 808                                     "nd6_lookup: failed to add route for a "
 809                                     "neighbor(%s), errno=%d\n",
 810                                     ip6_sprintf(addr6), e);
 811                         }
 812                         if (rt == NULL)
 813                                 return (NULL);
 814                         RT_LOCK(rt);
 815                         if (rt->rt_llinfo) {
 816                                 struct llinfo_nd6 *ln =
 817                                     (struct llinfo_nd6 *)rt->rt_llinfo;
 818                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 819                         }
 820                 } else
 821                         return (NULL);
 822         }
 823         RT_LOCK_ASSERT(rt);
 824         RT_REMREF(rt);
 825         /*
 826          * Validation for the entry.
 827          * Note that the check for rt_llinfo is necessary because a cloned
 828          * route from a parent route that has the L flag (e.g. the default
 829          * route to a p2p interface) may have the flag, too, while the
 830          * destination is not actually a neighbor.
 831          * XXX: we can't use rt->rt_ifp to check for the interface, since
 832          *      it might be the loopback interface if the entry is for our
 833          *      own address on a non-loopback interface. Instead, we should
 834          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 835          *      interface.
 836          */
 837         if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
 838             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 839             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 840                 if (create) {
 841                         nd6log((LOG_DEBUG,
 842                             "nd6_lookup: failed to lookup %s (if = %s)\n",
 843                             ip6_sprintf(addr6),
 844                             ifp ? if_name(ifp) : "unspec"));
 845                 }
 846                 RT_UNLOCK(rt);
 847                 return (NULL);
 848         }
 849         RT_UNLOCK(rt);          /* XXX not ready to return rt locked */
 850         return (rt);
 851 }
 852
 853 /*
 854  * Test whether a given IPv6 address is a neighbor or not, ignoring
 855  * the actual neighbor cache.  The neighbor cache is ignored in order
 856  * to not reenter the routing code from within itself.
 857  */
 858 static int
 859 nd6_is_new_addr_neighbor(addr, ifp)
 860         struct sockaddr_in6 *addr;
 861         struct ifnet *ifp;
 862 {
 863         struct nd_prefix *pr;
 864
 865         /*
 866          * A link-local address is always a neighbor.
 867          * XXX: a link does not necessarily specify a single interface.
 868          */
 869         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
 870                 struct sockaddr_in6 sin6_copy;
 871                 u_int32_t zone;
 872
 873                 /*
 874                  * We need sin6_copy since sa6_recoverscope() may modify the
 875                  * content (XXX).
 876                  */
 877                 sin6_copy = *addr;
 878                 if (sa6_recoverscope(&sin6_copy))
 879                         return (0); /* XXX: should be impossible */
 880                 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
 881                         return (0);
 882                 if (sin6_copy.sin6_scope_id == zone)
 883                         return (1);
 884                 else
 885                         return (0);
 886         }
 887
 888         /*
 889          * If the address matches one of our addresses,
 890          * it should be a neighbor.
 891          * If the address matches one of our on-link prefixes, it should be a
 892          * neighbor.
 893          */
 894         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
 895                 if (pr->ndpr_ifp != ifp)
 896                         continue;
 897
 898                 if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
 899                         continue;
 900
 901                 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
 902                     &addr->sin6_addr, &pr->ndpr_mask))
 903                         return (1);
 904         }
 905
 906         /*
 907          * If the default router list is empty, all addresses are regarded
 908          * as on-link, and thus, as a neighbor.
 909          * XXX: we restrict the condition to hosts, because routers usually do
 910          * not have the "default router list".
 911          */
 912         if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
 913             nd6_defifindex == ifp->if_index) {
 914                 return (1);
 915         }
 916
 917         return (0);
 918 }
 919
 920
 921 /*
 922  * Detect if a given IPv6 address identifies a neighbor on a given link.
 923  * XXX: should take care of the destination of a p2p link?
 924  */
 925 int
 926 nd6_is_addr_neighbor(addr, ifp)
 927         struct sockaddr_in6 *addr;
 928         struct ifnet *ifp;
 929 {
 930
 931         if (nd6_is_new_addr_neighbor(addr, ifp))
 932                 return (1);
 933
 934         /*
 935          * Even if the address matches none of our addresses, it might be
 936          * in the neighbor cache.
 937          */
 938         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 939                 return (1);
 940
 941         return (0);
 942 }
 943
 944 /*
 945  * Free an nd6 llinfo entry.
 946  */
 947 struct llinfo_nd6 *
 948 nd6_free(rt)
 949         struct rtentry *rt;
 950 {
 951         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 952         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 953         struct nd_defrouter *dr;
 954
 955         /*
 956          * we used to have pfctlinput(PRC_HOSTDEAD) here.
 957          * even though it is not harmful, it was not really necessary.
 958          */
 959
 960         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 961                 int s;
 962                 s = splnet();
 963                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 964                     rt->rt_ifp);
 965
 966                 if (ln->ln_router || dr) {
 967                         /*
 968                          * rt6_flush must be called whether or not the neighbor
 969                          * is in the Default Router List.
 970                          * See a corresponding comment in nd6_na_input().
 971                          */
 972                         rt6_flush(&in6, rt->rt_ifp);
 973                 }
 974
 975                 if (dr) {
 976                         /*
 977                          * Unreachablity of a router might affect the default
 978                          * router selection and on-link detection of advertised
 979                          * prefixes.
 980                          */
 981
 982                         /*
 983                          * Temporarily fake the state to choose a new default
 984                          * router and to perform on-link determination of
 985                          * prefixes correctly.
 986                          * Below the state will be set correctly,
 987                          * or the entry itself will be deleted.
 988                          */
 989                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 990
 991                         /*
 992                          * Since defrouter_select() does not affect the
 993                          * on-link determination and MIP6 needs the check
 994                          * before the default router selection, we perform
 995                          * the check now.
 996                          */
 997                         pfxlist_onlink_check();
 998
 999                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1000                                 /*
1001                                  * It is used as the current default router,
1002                                  * so we have to move it to the end of the
1003                                  * list and choose a new one.
1004                                  * XXX: it is not very efficient if this is
1005                                  *      the only router.
1006                                  */
1007                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1008                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1009
1010                                 defrouter_select();
1011                         }
1012                 }
1013                 splx(s);
1014         }
1015
1016         /*
1017          * Before deleting the entry, remember the next entry as the
1018          * return value.  We need this because pfxlist_onlink_check() above
1019          * might have freed other entries (particularly the old next entry) as
1020          * a side effect (XXX).
1021          */
1022         next = ln->ln_next;
1023
1024         /*
1025          * Detach the route from the routing tree and the list of neighbor
1026          * caches, and disable the route entry not to be used in already
1027          * cached routes.
1028          */
1029         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1030             rt_mask(rt), 0, (struct rtentry **)0);
1031
1032         return (next);
1033 }
1034
1035 /*
1036  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1037  *
1038  * XXX cost-effective methods?
1039  */
1040 void
1041 nd6_nud_hint(rt, dst6, force)
1042         struct rtentry *rt;
1043         struct in6_addr *dst6;
1044         int force;
1045 {
1046         struct llinfo_nd6 *ln;
1047
1048         /*
1049          * If the caller specified "rt", use that.  Otherwise, resolve the
1050          * routing table by supplied "dst6".
1051          */
1052         if (rt == NULL) {
1053                 if (dst6 == NULL)
1054                         return;
1055                 if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL)
1056                         return;
1057         }
1058
1059         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1060             (rt->rt_flags & RTF_LLINFO) == 0 ||
1061             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1062             rt->rt_gateway->sa_family != AF_LINK) {
1063                 /* This is not a host route. */
1064                 return;
1065         }
1066
1067         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1068         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1069                 return;
1070
1071         /*
1072          * if we get upper-layer reachability confirmation many times,
1073          * it is possible we have false information.
1074          */
1075         if (!force) {
1076                 ln->ln_byhint++;
1077                 if (ln->ln_byhint > nd6_maxnudhint)
1078                         return;
1079         }
1080
1081         ln->ln_state = ND6_LLINFO_REACHABLE;
1082         if (ln->ln_expire)
1083                 ln->ln_expire = time_second +
1084                         ND_IFINFO(rt->rt_ifp)->reachable;
1085 }
1086
1087 void
1088 nd6_rtrequest(req, rt, info)
1089         int     req;
1090         struct rtentry *rt;
1091         struct rt_addrinfo *info; /* xxx unused */
1092 {
1093         struct sockaddr *gate = rt->rt_gateway;
1094         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1095         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1096         struct ifnet *ifp = rt->rt_ifp;
1097         struct ifaddr *ifa;
1098
1099         RT_LOCK_ASSERT(rt);
1100
1101         if ((rt->rt_flags & RTF_GATEWAY) != 0)
1102                 return;
1103
1104         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1105                 /*
1106                  * This is probably an interface direct route for a link
1107                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1108                  * We do not need special treatment below for such a route.
1109                  * Moreover, the RTF_LLINFO flag which would be set below
1110                  * would annoy the ndp(8) command.
1111                  */
1112                 return;
1113         }
1114
1115         if (req == RTM_RESOLVE &&
1116             (nd6_need_cache(ifp) == 0 || /* stf case */
1117              !nd6_is_new_addr_neighbor((struct sockaddr_in6 *)rt_key(rt),
1118              ifp))) {
1119                 /*
1120                  * FreeBSD and BSD/OS often make a cloned host route based
1121                  * on a less-specific route (e.g. the default route).
1122                  * If the less specific route does not have a "gateway"
1123                  * (this is the case when the route just goes to a p2p or an
1124                  * stf interface), we'll mistakenly make a neighbor cache for
1125                  * the host route, and will see strange neighbor solicitation
1126                  * for the corresponding destination.  In order to avoid the
1127                  * confusion, we check if the destination of the route is
1128                  * a neighbor in terms of neighbor discovery, and stop the
1129                  * process if not.  Additionally, we remove the LLINFO flag
1130                  * so that ndp(8) will not try to get the neighbor information
1131                  * of the destination.
1132                  */
1133                 rt->rt_flags &= ~RTF_LLINFO;
1134                 return;
1135         }
1136
1137         switch (req) {
1138         case RTM_ADD:
1139                 /*
1140                  * There is no backward compatibility :)
1141                  *
1142                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1143                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1144                  *         rt->rt_flags |= RTF_CLONING;
1145                  */
1146                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1147                         /*
1148                          * Case 1: This route should come from
1149                          * a route to interface.  RTF_LLINFO flag is set
1150                          * for a host route whose destination should be
1151                          * treated as on-link.
1152                          */
1153                         rt_setgate(rt, rt_key(rt),
1154                                    (struct sockaddr *)&null_sdl);
1155                         gate = rt->rt_gateway;
1156                         SDL(gate)->sdl_type = ifp->if_type;
1157                         SDL(gate)->sdl_index = ifp->if_index;
1158                         if (ln)
1159                                 ln->ln_expire = time_second;
1160                         if (ln && ln->ln_expire == 0) {
1161                                 /* kludge for desktops */
1162                                 ln->ln_expire = 1;
1163                         }
1164                         if ((rt->rt_flags & RTF_CLONING) != 0)
1165                                 break;
1166                 }
1167                 /*
1168                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1169                  * We don't do that here since llinfo is not ready yet.
1170                  *
1171                  * There are also couple of other things to be discussed:
1172                  * - unsolicited NA code needs improvement beforehand
1173                  * - RFC2461 says we MAY send multicast unsolicited NA
1174                  *   (7.2.6 paragraph 4), however, it also says that we
1175                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1176                  *   we don't have anything like it right now.
1177                  *   note that the mechanism needs a mutual agreement
1178                  *   between proxies, which means that we need to implement
1179                  *   a new protocol, or a new kludge.
1180                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1181                  *   we need to check ip6forwarding before sending it.
1182                  *   (or should we allow proxy ND configuration only for
1183                  *   routers?  there's no mention about proxy ND from hosts)
1184                  */
1185                 /* FALLTHROUGH */
1186         case RTM_RESOLVE:
1187                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1188                         /*
1189                          * Address resolution isn't necessary for a point to
1190                          * point link, so we can skip this test for a p2p link.
1191                          */
1192                         if (gate->sa_family != AF_LINK ||
1193                             gate->sa_len < sizeof(null_sdl)) {
1194                                 log(LOG_DEBUG,
1195                                     "nd6_rtrequest: bad gateway value: %s\n",
1196                                     if_name(ifp));
1197                                 break;
1198                         }
1199                         SDL(gate)->sdl_type = ifp->if_type;
1200                         SDL(gate)->sdl_index = ifp->if_index;
1201                 }
1202                 if (ln != NULL)
1203                         break;  /* This happens on a route change */
1204                 /*
1205                  * Case 2: This route may come from cloning, or a manual route
1206                  * add with a LL address.
1207                  */
1208                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1209                 rt->rt_llinfo = (caddr_t)ln;
1210                 if (ln == NULL) {
1211                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1212                         break;
1213                 }
1214                 nd6_inuse++;
1215                 nd6_allocated++;
1216                 bzero(ln, sizeof(*ln));
1217                 ln->ln_rt = rt;
1218                 /* this is required for "ndp" command. - shin */
1219                 if (req == RTM_ADD) {
1220                         /*
1221                          * gate should have some valid AF_LINK entry,
1222                          * and ln->ln_expire should have some lifetime
1223                          * which is specified by ndp command.
1224                          */
1225                         ln->ln_state = ND6_LLINFO_REACHABLE;
1226                         ln->ln_byhint = 0;
1227                 } else {
1228                         /*
1229                          * When req == RTM_RESOLVE, rt is created and
1230                          * initialized in rtrequest(), so rt_expire is 0.
1231                          */
1232                         ln->ln_state = ND6_LLINFO_NOSTATE;
1233                         ln->ln_expire = time_second;
1234                 }
1235                 rt->rt_flags |= RTF_LLINFO;
1236                 ln->ln_next = llinfo_nd6.ln_next;
1237                 llinfo_nd6.ln_next = ln;
1238                 ln->ln_prev = &llinfo_nd6;
1239                 ln->ln_next->ln_prev = ln;
1240
1241                 /*
1242                  * check if rt_key(rt) is one of my address assigned
1243                  * to the interface.
1244                  */
1245                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1246                     &SIN6(rt_key(rt))->sin6_addr);
1247                 if (ifa) {
1248                         caddr_t macp = nd6_ifptomac(ifp);
1249                         ln->ln_expire = 0;
1250                         ln->ln_state = ND6_LLINFO_REACHABLE;
1251                         ln->ln_byhint = 0;
1252                         if (macp) {
1253                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1254                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1255                         }
1256                         if (nd6_useloopback) {
1257                                 rt->rt_ifp = &loif[0];  /* XXX */
1258                                 /*
1259                                  * Make sure rt_ifa be equal to the ifaddr
1260                                  * corresponding to the address.
1261                                  * We need this because when we refer
1262                                  * rt_ifa->ia6_flags in ip6_input, we assume
1263                                  * that the rt_ifa points to the address instead
1264                                  * of the loopback address.
1265                                  */
1266                                 if (ifa != rt->rt_ifa) {
1267                                         IFAFREE(rt->rt_ifa);
1268                                         IFAREF(ifa);
1269                                         rt->rt_ifa = ifa;
1270                                 }
1271                         }
1272                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1273                         ln->ln_expire = 0;
1274                         ln->ln_state = ND6_LLINFO_REACHABLE;
1275                         ln->ln_byhint = 0;
1276
1277                         /* join solicited node multicast for proxy ND */
1278                         if (ifp->if_flags & IFF_MULTICAST) {
1279                                 struct in6_addr llsol;
1280                                 int error;
1281
1282                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1283                                 llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL;
1284                                 llsol.s6_addr32[1] = 0;
1285                                 llsol.s6_addr32[2] = htonl(1);
1286                                 llsol.s6_addr8[12] = 0xff;
1287                                 if (in6_setscope(&llsol, ifp, NULL))
1288                                         break;
1289                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1290                                         nd6log((LOG_ERR, "%s: failed to join "
1291                                             "%s (errno=%d)\n", if_name(ifp),
1292                                             ip6_sprintf(&llsol), error));
1293                                 }
1294                         }
1295                 }
1296                 break;
1297
1298         case RTM_DELETE:
1299                 if (ln == NULL)
1300                         break;
1301                 /* leave from solicited node multicast for proxy ND */
1302                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1303                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1304                         struct in6_addr llsol;
1305                         struct in6_multi *in6m;
1306
1307                         llsol = SIN6(rt_key(rt))->sin6_addr;
1308                         llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL;
1309                         llsol.s6_addr32[1] = 0;
1310                         llsol.s6_addr32[2] = htonl(1);
1311                         llsol.s6_addr8[12] = 0xff;
1312                         if (in6_setscope(&llsol, ifp, NULL) == 0) {
1313                                 IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1314                                 if (in6m)
1315                                         in6_delmulti(in6m);
1316                         } else
1317                                 ; /* XXX: should not happen. bark here? */
1318                 }
1319                 nd6_inuse--;
1320                 ln->ln_next->ln_prev = ln->ln_prev;
1321                 ln->ln_prev->ln_next = ln->ln_next;
1322                 ln->ln_prev = NULL;
1323                 rt->rt_llinfo = 0;
1324                 rt->rt_flags &= ~RTF_LLINFO;
1325                 if (ln->ln_hold)
1326                         m_freem(ln->ln_hold);
1327                 Free((caddr_t)ln);
1328         }
1329 }
1330
1331 int
1332 nd6_ioctl(cmd, data, ifp)
1333         u_long cmd;
1334         caddr_t data;
1335         struct ifnet *ifp;
1336 {
1337         struct in6_drlist *drl = (struct in6_drlist *)data;
1338         struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1339         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1340         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1341         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1342         struct nd_defrouter *dr, any;
1343         struct nd_prefix *pr;
1344         struct rtentry *rt;
1345         int i = 0, error = 0;
1346         int s;
1347
1348         switch (cmd) {
1349         case SIOCGDRLST_IN6:
1350                 /*
1351                  * obsolete API, use sysctl under net.inet6.icmp6
1352                  */
1353                 bzero(drl, sizeof(*drl));
1354                 s = splnet();
1355                 dr = TAILQ_FIRST(&nd_defrouter);
1356                 while (dr && i < DRLSTSIZ) {
1357                         drl->defrouter[i].rtaddr = dr->rtaddr;
1358                         in6_clearscope(&drl->defrouter[i].rtaddr);
1359
1360                         drl->defrouter[i].flags = dr->flags;
1361                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1362                         drl->defrouter[i].expire = dr->expire;
1363                         drl->defrouter[i].if_index = dr->ifp->if_index;
1364                         i++;
1365                         dr = TAILQ_NEXT(dr, dr_entry);
1366                 }
1367                 splx(s);
1368                 break;
1369         case SIOCGPRLST_IN6:
1370                 /*
1371                  * obsolete API, use sysctl under net.inet6.icmp6
1372                  *
1373                  * XXX the structure in6_prlist was changed in backward-
1374                  * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1375                  * in6_prlist is used for nd6_sysctl() - fill_prlist().
1376                  */
1377                 /*
1378                  * XXX meaning of fields, especialy "raflags", is very
1379                  * differnet between RA prefix list and RR/static prefix list.
1380                  * how about separating ioctls into two?
1381                  */
1382                 bzero(oprl, sizeof(*oprl));
1383                 s = splnet();
1384                 pr = nd_prefix.lh_first;
1385                 while (pr && i < PRLSTSIZ) {
1386                         struct nd_pfxrouter *pfr;
1387                         int j;
1388
1389                         oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1390                         oprl->prefix[i].raflags = pr->ndpr_raf;
1391                         oprl->prefix[i].prefixlen = pr->ndpr_plen;
1392                         oprl->prefix[i].vltime = pr->ndpr_vltime;
1393                         oprl->prefix[i].pltime = pr->ndpr_pltime;
1394                         oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1395                         oprl->prefix[i].expire = pr->ndpr_expire;
1396
1397                         pfr = pr->ndpr_advrtrs.lh_first;
1398                         j = 0;
1399                         while (pfr) {
1400                                 if (j < DRLSTSIZ) {
1401 #define RTRADDR oprl->prefix[i].advrtr[j]
1402                                         RTRADDR = pfr->router->rtaddr;
1403                                         in6_clearscope(&RTRADDR);
1404 #undef RTRADDR
1405                                 }
1406                                 j++;
1407                                 pfr = pfr->pfr_next;
1408                         }
1409                         oprl->prefix[i].advrtrs = j;
1410                         oprl->prefix[i].origin = PR_ORIG_RA;
1411
1412                         i++;
1413                         pr = pr->ndpr_next;
1414                 }
1415                 splx(s);
1416
1417                 break;
1418         case OSIOCGIFINFO_IN6:
1419 #define ND      ndi->ndi
1420                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1421                 bzero(&ND, sizeof(ND));
1422                 ND.linkmtu = IN6_LINKMTU(ifp);
1423                 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1424                 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1425                 ND.reachable = ND_IFINFO(ifp)->reachable;
1426                 ND.retrans = ND_IFINFO(ifp)->retrans;
1427                 ND.flags = ND_IFINFO(ifp)->flags;
1428                 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1429                 ND.chlim = ND_IFINFO(ifp)->chlim;
1430                 break;
1431         case SIOCGIFINFO_IN6:
1432                 ND = *ND_IFINFO(ifp);
1433                 ND.linkmtu = IN6_LINKMTU(ifp);
1434                 break;
1435         case SIOCSIFINFO_IN6:
1436                 /*
1437                  * used to change host variables from userland.
1438                  * intented for a use on router to reflect RA configurations.
1439                  */
1440                 /* 0 means 'unspecified' */
1441                 if (ND.linkmtu != 0) {
1442                         if (ND.linkmtu < IPV6_MMTU ||
1443                             ND.linkmtu > IN6_LINKMTU(ifp)) {
1444                                 error = EINVAL;
1445                                 break;
1446                         }
1447                         ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1448                 }
1449
1450                 if (ND.basereachable != 0) {
1451                         int obasereachable = ND_IFINFO(ifp)->basereachable;
1452
1453                         ND_IFINFO(ifp)->basereachable = ND.basereachable;
1454                         if (ND.basereachable != obasereachable)
1455                                 ND_IFINFO(ifp)->reachable =
1456                                     ND_COMPUTE_RTIME(ND.basereachable);
1457                 }
1458                 if (ND.retrans != 0)
1459                         ND_IFINFO(ifp)->retrans = ND.retrans;
1460                 if (ND.chlim != 0)
1461                         ND_IFINFO(ifp)->chlim = ND.chlim;
1462                 /* FALLTHROUGH */
1463         case SIOCSIFINFO_FLAGS:
1464                 ND_IFINFO(ifp)->flags = ND.flags;
1465                 break;
1466 #undef ND
1467         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1468                 /* flush default router list */
1469                 /*
1470                  * xxx sumikawa: should not delete route if default
1471                  * route equals to the top of default router list
1472                  */
1473                 bzero(&any, sizeof(any));
1474                 defrouter_delreq(&any, 0);
1475                 defrouter_select();
1476                 /* xxx sumikawa: flush prefix list */
1477                 break;
1478         case SIOCSPFXFLUSH_IN6:
1479         {
1480                 /* flush all the prefix advertised by routers */
1481                 struct nd_prefix *pr, *next;
1482
1483                 s = splnet();
1484                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1485                         struct in6_ifaddr *ia, *ia_next;
1486
1487                         next = pr->ndpr_next;
1488
1489                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1490                                 continue; /* XXX */
1491
1492                         /* do we really have to remove addresses as well? */
1493                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1494                                 /* ia might be removed.  keep the next ptr. */
1495                                 ia_next = ia->ia_next;
1496
1497                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1498                                         continue;
1499
1500                                 if (ia->ia6_ndpr == pr)
1501                                         in6_purgeaddr(&ia->ia_ifa);
1502                         }
1503                         prelist_remove(pr);
1504                 }
1505                 splx(s);
1506                 break;
1507         }
1508         case SIOCSRTRFLUSH_IN6:
1509         {
1510                 /* flush all the default routers */
1511                 struct nd_defrouter *dr, *next;
1512
1513                 s = splnet();
1514                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1515                         /*
1516                          * The first entry of the list may be stored in
1517                          * the routing table, so we'll delete it later.
1518                          */
1519                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1520                                 next = TAILQ_NEXT(dr, dr_entry);
1521                                 defrtrlist_del(dr);
1522                         }
1523                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1524                 }
1525                 splx(s);
1526                 break;
1527         }
1528         case SIOCGNBRINFO_IN6:
1529         {
1530                 struct llinfo_nd6 *ln;
1531                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1532
1533                 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1534                         return (error);
1535
1536                 s = splnet();
1537                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1538                         error = EINVAL;
1539                         splx(s);
1540                         break;
1541                 }
1542                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1543                 nbi->state = ln->ln_state;
1544                 nbi->asked = ln->ln_asked;
1545                 nbi->isrouter = ln->ln_router;
1546                 nbi->expire = ln->ln_expire;
1547                 splx(s);
1548
1549                 break;
1550         }
1551         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1552                 ndif->ifindex = nd6_defifindex;
1553                 break;
1554         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1555                 return (nd6_setdefaultiface(ndif->ifindex));
1556         }
1557         return (error);
1558 }
1559
1560 /*
1561  * Create neighbor cache entry and cache link-layer address,
1562  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1563  */
1564 struct rtentry *
1565 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1566         struct ifnet *ifp;
1567         struct in6_addr *from;
1568         char *lladdr;
1569         int lladdrlen;
1570         int type;       /* ICMP6 type */
1571         int code;       /* type dependent information */
1572 {
1573         struct rtentry *rt = NULL;
1574         struct llinfo_nd6 *ln = NULL;
1575         int is_newentry;
1576         struct sockaddr_dl *sdl = NULL;
1577         int do_update;
1578         int olladdr;
1579         int llchange;
1580         int newstate = 0;
1581
1582         if (ifp == NULL)
1583                 panic("ifp == NULL in nd6_cache_lladdr");
1584         if (from == NULL)
1585                 panic("from == NULL in nd6_cache_lladdr");
1586
1587         /* nothing must be updated for unspecified address */
1588         if (IN6_IS_ADDR_UNSPECIFIED(from))
1589                 return NULL;
1590
1591         /*
1592          * Validation about ifp->if_addrlen and lladdrlen must be done in
1593          * the caller.
1594          *
1595          * XXX If the link does not have link-layer adderss, what should
1596          * we do? (ifp->if_addrlen == 0)
1597          * Spec says nothing in sections for RA, RS and NA.  There's small
1598          * description on it in NS section (RFC 2461 7.2.3).
1599          */
1600
1601         rt = nd6_lookup(from, 0, ifp);
1602         if (rt == NULL) {
1603                 rt = nd6_lookup(from, 1, ifp);
1604                 is_newentry = 1;
1605         } else {
1606                 /* do nothing if static ndp is set */
1607                 if (rt->rt_flags & RTF_STATIC)
1608                         return NULL;
1609                 is_newentry = 0;
1610         }
1611
1612         if (rt == NULL)
1613                 return NULL;
1614         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1615 fail:
1616                 (void)nd6_free(rt);
1617                 return NULL;
1618         }
1619         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1620         if (ln == NULL)
1621                 goto fail;
1622         if (rt->rt_gateway == NULL)
1623                 goto fail;
1624         if (rt->rt_gateway->sa_family != AF_LINK)
1625                 goto fail;
1626         sdl = SDL(rt->rt_gateway);
1627
1628         olladdr = (sdl->sdl_alen) ? 1 : 0;
1629         if (olladdr && lladdr) {
1630                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1631                         llchange = 1;
1632                 else
1633                         llchange = 0;
1634         } else
1635                 llchange = 0;
1636
1637         /*
1638          * newentry olladdr  lladdr  llchange   (*=record)
1639          *      0       n       n       --      (1)
1640          *      0       y       n       --      (2)
1641          *      0       n       y       --      (3) * STALE
1642          *      0       y       y       n       (4) *
1643          *      0       y       y       y       (5) * STALE
1644          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1645          *      1       --      y       --      (7) * STALE
1646          */
1647
1648         if (lladdr) {           /* (3-5) and (7) */
1649                 /*
1650                  * Record source link-layer address
1651                  * XXX is it dependent to ifp->if_type?
1652                  */
1653                 sdl->sdl_alen = ifp->if_addrlen;
1654                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1655         }
1656
1657         if (!is_newentry) {
1658                 if ((!olladdr && lladdr != NULL) ||     /* (3) */
1659                     (olladdr && lladdr != NULL && llchange)) {  /* (5) */
1660                         do_update = 1;
1661                         newstate = ND6_LLINFO_STALE;
1662                 } else                                  /* (1-2,4) */
1663                         do_update = 0;
1664         } else {
1665                 do_update = 1;
1666                 if (lladdr == NULL)                     /* (6) */
1667                         newstate = ND6_LLINFO_NOSTATE;
1668                 else                                    /* (7) */
1669                         newstate = ND6_LLINFO_STALE;
1670         }
1671
1672         if (do_update) {
1673                 /*
1674                  * Update the state of the neighbor cache.
1675                  */
1676                 ln->ln_state = newstate;
1677
1678                 if (ln->ln_state == ND6_LLINFO_STALE) {
1679                         /*
1680                          * XXX: since nd6_output() below will cause
1681                          * state tansition to DELAY and reset the timer,
1682                          * we must set the timer now, although it is actually
1683                          * meaningless.
1684                          */
1685                         ln->ln_expire = time_second + nd6_gctimer;
1686
1687                         if (ln->ln_hold) {
1688                                 /*
1689                                  * we assume ifp is not a p2p here, so just
1690                                  * set the 2nd argument as the 1st one.
1691                                  */
1692                                 nd6_output(ifp, ifp, ln->ln_hold,
1693                                     (struct sockaddr_in6 *)rt_key(rt), rt);
1694                                 ln->ln_hold = NULL;
1695                         }
1696                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1697                         /* probe right away */
1698                         ln->ln_expire = time_second;
1699                 }
1700         }
1701
1702         /*
1703          * ICMP6 type dependent behavior.
1704          *
1705          * NS: clear IsRouter if new entry
1706          * RS: clear IsRouter
1707          * RA: set IsRouter if there's lladdr
1708          * redir: clear IsRouter if new entry
1709          *
1710          * RA case, (1):
1711          * The spec says that we must set IsRouter in the following cases:
1712          * - If lladdr exist, set IsRouter.  This means (1-5).
1713          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1714          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1715          * A quetion arises for (1) case.  (1) case has no lladdr in the
1716          * neighbor cache, this is similar to (6).
1717          * This case is rare but we figured that we MUST NOT set IsRouter.
1718          *
1719          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1720          *                                                      D R
1721          *      0       n       n       --      (1)     c   ?     s
1722          *      0       y       n       --      (2)     c   s     s
1723          *      0       n       y       --      (3)     c   s     s
1724          *      0       y       y       n       (4)     c   s     s
1725          *      0       y       y       y       (5)     c   s     s
1726          *      1       --      n       --      (6) c   c       c s
1727          *      1       --      y       --      (7) c   c   s   c s
1728          *
1729          *                                      (c=clear s=set)
1730          */
1731         switch (type & 0xff) {
1732         case ND_NEIGHBOR_SOLICIT:
1733                 /*
1734                  * New entry must have is_router flag cleared.
1735                  */
1736                 if (is_newentry)        /* (6-7) */
1737                         ln->ln_router = 0;
1738                 break;
1739         case ND_REDIRECT:
1740                 /*
1741                  * If the icmp is a redirect to a better router, always set the
1742                  * is_router flag.  Otherwise, if the entry is newly created,
1743                  * clear the flag.  [RFC 2461, sec 8.3]
1744                  */
1745                 if (code == ND_REDIRECT_ROUTER)
1746                         ln->ln_router = 1;
1747                 else if (is_newentry) /* (6-7) */
1748                         ln->ln_router = 0;
1749                 break;
1750         case ND_ROUTER_SOLICIT:
1751                 /*
1752                  * is_router flag must always be cleared.
1753                  */
1754                 ln->ln_router = 0;
1755                 break;
1756         case ND_ROUTER_ADVERT:
1757                 /*
1758                  * Mark an entry with lladdr as a router.
1759                  */
1760                 if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
1761                     (is_newentry && lladdr)) {                  /* (7) */
1762                         ln->ln_router = 1;
1763                 }
1764                 break;
1765         }
1766
1767         /*
1768          * When the link-layer address of a router changes, select the
1769          * best router again.  In particular, when the neighbor entry is newly
1770          * created, it might affect the selection policy.
1771          * Question: can we restrict the first condition to the "is_newentry"
1772          * case?
1773          * XXX: when we hear an RA from a new router with the link-layer
1774          * address option, defrouter_select() is called twice, since
1775          * defrtrlist_update called the function as well.  However, I believe
1776          * we can compromise the overhead, since it only happens the first
1777          * time.
1778          * XXX: although defrouter_select() should not have a bad effect
1779          * for those are not autoconfigured hosts, we explicitly avoid such
1780          * cases for safety.
1781          */
1782         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1783                 defrouter_select();
1784
1785         return rt;
1786 }
1787
1788 static void
1789 nd6_slowtimo(ignored_arg)
1790     void *ignored_arg;
1791 {
1792         int s = splnet();
1793         struct nd_ifinfo *nd6if;
1794         struct ifnet *ifp;
1795
1796         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1797             nd6_slowtimo, NULL);
1798         IFNET_RLOCK();
1799         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1800                 nd6if = ND_IFINFO(ifp);
1801                 if (nd6if->basereachable && /* already initialized */
1802                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1803                         /*
1804                          * Since reachable time rarely changes by router
1805                          * advertisements, we SHOULD insure that a new random
1806                          * value gets recomputed at least once every few hours.
1807                          * (RFC 2461, 6.3.4)
1808                          */
1809                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1810                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1811                 }
1812         }
1813         IFNET_RUNLOCK();
1814         splx(s);
1815 }
1816
1817 #define senderr(e) { error = (e); goto bad;}
1818 int
1819 nd6_output(ifp, origifp, m0, dst, rt0)
1820         struct ifnet *ifp;
1821         struct ifnet *origifp;
1822         struct mbuf *m0;
1823         struct sockaddr_in6 *dst;
1824         struct rtentry *rt0;
1825 {
1826         struct mbuf *m = m0;
1827         struct rtentry *rt = rt0;
1828         struct sockaddr_in6 *gw6 = NULL;
1829         struct llinfo_nd6 *ln = NULL;
1830         int error = 0;
1831
1832         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1833                 goto sendpkt;
1834
1835         if (nd6_need_cache(ifp) == 0)
1836                 goto sendpkt;
1837
1838         /*
1839          * next hop determination.  This routine is derived from ether_output.
1840          */
1841 again:
1842         if (rt) {
1843                 if ((rt->rt_flags & RTF_UP) == 0) {
1844                         rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL);
1845                         if (rt != NULL) {
1846                                 RT_REMREF(rt);
1847                                 RT_UNLOCK(rt);
1848                                 if (rt->rt_ifp != ifp)
1849                                         /*
1850                                          * XXX maybe we should update ifp too,
1851                                          * but the original code didn't and I
1852                                          * don't know what is correct here.
1853                                          */
1854                                         goto again;
1855                         } else
1856                                 senderr(EHOSTUNREACH);
1857                 }
1858
1859                 if (rt->rt_flags & RTF_GATEWAY) {
1860                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1861
1862                         /*
1863                          * We skip link-layer address resolution and NUD
1864                          * if the gateway is not a neighbor from ND point
1865                          * of view, regardless of the value of nd_ifinfo.flags.
1866                          * The second condition is a bit tricky; we skip
1867                          * if the gateway is our own address, which is
1868                          * sometimes used to install a route to a p2p link.
1869                          */
1870                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1871                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1872                                 /*
1873                                  * We allow this kind of tricky route only
1874                                  * when the outgoing interface is p2p.
1875                                  * XXX: we may need a more generic rule here.
1876                                  */
1877                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1878                                         senderr(EHOSTUNREACH);
1879
1880                                 goto sendpkt;
1881                         }
1882
1883                         if (rt->rt_gwroute == 0)
1884                                 goto lookup;
1885                         if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1886                                 RT_LOCK(rt);
1887                                 rtfree(rt); rt = rt0;
1888                         lookup:
1889                                 rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1890                                 if ((rt = rt->rt_gwroute) == 0)
1891                                         senderr(EHOSTUNREACH);
1892                                 RT_UNLOCK(rt);
1893                         }
1894                 }
1895         }
1896
1897         /*
1898          * Address resolution or Neighbor Unreachability Detection
1899          * for the next hop.
1900          * At this point, the destination of the packet must be a unicast
1901          * or an anycast address(i.e. not a multicast).
1902          */
1903
1904         /* Look up the neighbor cache for the nexthop */
1905         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1906                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1907         else {
1908                 /*
1909                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1910                  * the condition below is not very efficient.  But we believe
1911                  * it is tolerable, because this should be a rare case.
1912                  */
1913                 if (nd6_is_addr_neighbor(dst, ifp) &&
1914                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1915                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1916         }
1917         if (ln == NULL || rt == NULL) {
1918                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1919                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1920                         log(LOG_DEBUG,
1921                             "nd6_output: can't allocate llinfo for %s "
1922                             "(ln=%p, rt=%p)\n",
1923                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1924                         senderr(EIO);   /* XXX: good error? */
1925                 }
1926
1927                 goto sendpkt;   /* send anyway */
1928         }
1929
1930         /* We don't have to do link-layer address resolution on a p2p link. */
1931         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1932             ln->ln_state < ND6_LLINFO_REACHABLE) {
1933                 ln->ln_state = ND6_LLINFO_STALE;
1934                 ln->ln_expire = time_second + nd6_gctimer;
1935         }
1936
1937         /*
1938          * The first time we send a packet to a neighbor whose entry is
1939          * STALE, we have to change the state to DELAY and a sets a timer to
1940          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1941          * neighbor unreachability detection on expiration.
1942          * (RFC 2461 7.3.3)
1943          */
1944         if (ln->ln_state == ND6_LLINFO_STALE) {
1945                 ln->ln_asked = 0;
1946                 ln->ln_state = ND6_LLINFO_DELAY;
1947                 ln->ln_expire = time_second + nd6_delay;
1948         }
1949
1950         /*
1951          * If the neighbor cache entry has a state other than INCOMPLETE
1952          * (i.e. its link-layer address is already resolved), just
1953          * send the packet.
1954          */
1955         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1956                 goto sendpkt;
1957
1958         /*
1959          * There is a neighbor cache entry, but no ethernet address
1960          * response yet.  Replace the held mbuf (if any) with this
1961          * latest one.
1962          *
1963          * This code conforms to the rate-limiting rule described in Section
1964          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1965          * an NS below.
1966          */
1967         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1968                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1969         if (ln->ln_hold)
1970                 m_freem(ln->ln_hold);
1971         ln->ln_hold = m;
1972         if (ln->ln_expire) {
1973                 if (ln->ln_asked < nd6_mmaxtries &&
1974                     ln->ln_expire < time_second) {
1975                         ln->ln_asked++;
1976                         ln->ln_expire = time_second +
1977                                 ND_IFINFO(ifp)->retrans / 1000;
1978                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1979                 }
1980         }
1981         return (0);
1982
1983   sendpkt:
1984         /* discard the packet if IPv6 operation is disabled on the interface */
1985         if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1986                 error = ENETDOWN; /* better error? */
1987                 goto bad;
1988         }
1989
1990 #ifdef IPSEC
1991         /* clean ipsec history once it goes out of the node */
1992         ipsec_delaux(m);
1993 #endif
1994
1995 #ifdef MAC
1996         mac_create_mbuf_linklayer(ifp, m);
1997 #endif
1998         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1999                 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2000                     rt));
2001         }
2002         return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2003
2004   bad:
2005         if (m)
2006                 m_freem(m);
2007         return (error);
2008 }
2009 #undef senderr
2010
2011 int
2012 nd6_need_cache(ifp)
2013         struct ifnet *ifp;
2014 {
2015         /*
2016          * XXX: we currently do not make neighbor cache on any interface
2017          * other than ARCnet, Ethernet, FDDI and GIF.
2018          *
2019          * RFC2893 says:
2020          * - unidirectional tunnels needs no ND
2021          */
2022         switch (ifp->if_type) {
2023         case IFT_ARCNET:
2024         case IFT_ETHER:
2025         case IFT_FDDI:
2026         case IFT_IEEE1394:
2027 #ifdef IFT_L2VLAN
2028         case IFT_L2VLAN:
2029 #endif
2030 #ifdef IFT_IEEE80211
2031         case IFT_IEEE80211:
2032 #endif
2033 #ifdef IFT_CARP
2034         case IFT_CARP:
2035 #endif
2036         case IFT_GIF:           /* XXX need more cases? */
2037         case IFT_PPP:
2038         case IFT_TUNNEL:
2039         case IFT_BRIDGE:
2040                 return (1);
2041         default:
2042                 return (0);
2043         }
2044 }
2045
2046 int
2047 nd6_storelladdr(ifp, rt0, m, dst, desten)
2048         struct ifnet *ifp;
2049         struct rtentry *rt0;
2050         struct mbuf *m;
2051         struct sockaddr *dst;
2052         u_char *desten;
2053 {
2054         struct sockaddr_dl *sdl;
2055         struct rtentry *rt;
2056         int error;
2057
2058         if (m->m_flags & M_MCAST) {
2059                 int i;
2060
2061                 switch (ifp->if_type) {
2062                 case IFT_ETHER:
2063                 case IFT_FDDI:
2064 #ifdef IFT_L2VLAN
2065                 case IFT_L2VLAN:
2066 #endif
2067 #ifdef IFT_IEEE80211
2068                 case IFT_IEEE80211:
2069 #endif
2070                 case IFT_BRIDGE:
2071                 case IFT_ISO88025:
2072                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2073                                                  desten);
2074                         return (0);
2075                 case IFT_IEEE1394:
2076                         /*
2077                          * netbsd can use if_broadcastaddr, but we don't do so
2078                          * to reduce # of ifdef.
2079                          */
2080                         for (i = 0; i < ifp->if_addrlen; i++)
2081                                 desten[i] = ~0;
2082                         return (0);
2083                 case IFT_ARCNET:
2084                         *desten = 0;
2085                         return (0);
2086                 default:
2087                         m_freem(m);
2088                         return (EAFNOSUPPORT);
2089                 }
2090         }
2091
2092         if (rt0 == NULL) {
2093                 /* this could happen, if we could not allocate memory */
2094                 m_freem(m);
2095                 return (ENOMEM);
2096         }
2097
2098         error = rt_check(&rt, &rt0, dst);
2099         if (error) {
2100                 m_freem(m);
2101                 return (error);
2102         }
2103         RT_UNLOCK(rt);
2104
2105         if (rt->rt_gateway->sa_family != AF_LINK) {
2106                 printf("nd6_storelladdr: something odd happens\n");
2107                 m_freem(m);
2108                 return (EINVAL);
2109         }
2110         sdl = SDL(rt->rt_gateway);
2111         if (sdl->sdl_alen == 0) {
2112                 /* this should be impossible, but we bark here for debugging */
2113                 printf("nd6_storelladdr: sdl_alen == 0\n");
2114                 m_freem(m);
2115                 return (EINVAL);
2116         }
2117
2118         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2119         return (0);
2120 }
2121
2122 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2123 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2124 #ifdef SYSCTL_DECL
2125 SYSCTL_DECL(_net_inet6_icmp6);
2126 #endif
2127 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2128         CTLFLAG_RD, nd6_sysctl_drlist, "");
2129 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2130         CTLFLAG_RD, nd6_sysctl_prlist, "");
2131
2132 static int
2133 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2134 {
2135         int error;
2136         char buf[1024];
2137         struct in6_defrouter *d, *de;
2138         struct nd_defrouter *dr;
2139
2140         if (req->newptr)
2141                 return EPERM;
2142         error = 0;
2143
2144         for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2145              dr = TAILQ_NEXT(dr, dr_entry)) {
2146                 d = (struct in6_defrouter *)buf;
2147                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2148
2149                 if (d + 1 <= de) {
2150                         bzero(d, sizeof(*d));
2151                         d->rtaddr.sin6_family = AF_INET6;
2152                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2153                         d->rtaddr.sin6_addr = dr->rtaddr;
2154                         if (sa6_recoverscope(&d->rtaddr)) {
2155                                 log(LOG_ERR,
2156                                     "scope error in router list (%s)\n",
2157                                     ip6_sprintf(&d->rtaddr.sin6_addr));
2158                                 /* XXX: press on... */
2159                         }
2160                         d->flags = dr->flags;
2161                         d->rtlifetime = dr->rtlifetime;
2162                         d->expire = dr->expire;
2163                         d->if_index = dr->ifp->if_index;
2164                 } else
2165                         panic("buffer too short");
2166
2167                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2168                 if (error)
2169                         break;
2170         }
2171
2172         return (error);
2173 }
2174
2175 static int
2176 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2177 {
2178         int error;
2179         char buf[1024];
2180         struct in6_prefix *p, *pe;
2181         struct nd_prefix *pr;
2182
2183         if (req->newptr)
2184                 return EPERM;
2185         error = 0;
2186
2187         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2188                 u_short advrtrs;
2189                 size_t advance;
2190                 struct sockaddr_in6 *sin6, *s6;
2191                 struct nd_pfxrouter *pfr;
2192
2193                 p = (struct in6_prefix *)buf;
2194                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2195
2196                 if (p + 1 <= pe) {
2197                         bzero(p, sizeof(*p));
2198                         sin6 = (struct sockaddr_in6 *)(p + 1);
2199
2200                         p->prefix = pr->ndpr_prefix;
2201                         if (sa6_recoverscope(&p->prefix)) {
2202                                 log(LOG_ERR,
2203                                     "scope error in prefix list (%s)\n",
2204                                     ip6_sprintf(&p->prefix.sin6_addr));
2205                                 /* XXX: press on... */
2206                         }
2207                         p->raflags = pr->ndpr_raf;
2208                         p->prefixlen = pr->ndpr_plen;
2209                         p->vltime = pr->ndpr_vltime;
2210                         p->pltime = pr->ndpr_pltime;
2211                         p->if_index = pr->ndpr_ifp->if_index;
2212                         p->expire = pr->ndpr_expire;
2213                         p->refcnt = pr->ndpr_refcnt;
2214                         p->flags = pr->ndpr_stateflags;
2215                         p->origin = PR_ORIG_RA;
2216                         advrtrs = 0;
2217                         for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2218                              pfr = pfr->pfr_next) {
2219                                 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2220                                         advrtrs++;
2221                                         continue;
2222                                 }
2223                                 s6 = &sin6[advrtrs];
2224                                 bzero(s6, sizeof(*s6));
2225                                 s6->sin6_family = AF_INET6;
2226                                 s6->sin6_len = sizeof(*sin6);
2227                                 s6->sin6_addr = pfr->router->rtaddr;
2228                                 if (sa6_recoverscope(s6)) {
2229                                         log(LOG_ERR,
2230                                             "scope error in "
2231                                             "prefix list (%s)\n",
2232                                             ip6_sprintf(&pfr->router->rtaddr));
2233                                 }
2234                                 advrtrs++;
2235                         }
2236                         p->advrtrs = advrtrs;
2237                 } else
2238                         panic("buffer too short");
2239
2240                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2241                 error = SYSCTL_OUT(req, buf, advance);
2242                 if (error)
2243                         break;
2244         }
2245
2246         return (error);
2247 }