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)
 264                 panic("ndopts == NULL in nd6_option");
 265         if (!ndopts->nd_opts_last)
 266                 panic("uninitialized ndopts in nd6_option");
 267         if (!ndopts->nd_opts_search)
 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)
 316                 panic("ndopts == NULL in nd6_options");
 317         if (!ndopts->nd_opts_last)
 318                 panic("uninitialized ndopts in nd6_options");
 319         if (!ndopts->nd_opts_search)
 320                 return 0;
 321
 322         while (1) {
 323                 nd_opt = nd6_option(ndopts);
 324                 if (!nd_opt && !ndopts->nd_opts_last) {
 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)
 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 = 0;
 779                 }
 780         }
 781         if (!rt) {
 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                         log(LOG_DEBUG,
 842                             "nd6_lookup: failed to lookup %s (if = %s)\n",
 843                             ip6_sprintf(addr6),
 844                             ifp ? if_name(ifp) : "unspec");
 845                         /* xxx more logs... kazu */
 846                 }
 847                 RT_UNLOCK(rt);
 848                 return (NULL);
 849         }
 850         RT_UNLOCK(rt);          /* XXX not ready to return rt locked */
 851         return (rt);
 852 }
 853
 854 /*
 855  * Test whether a given IPv6 address is a neighbor or not, ignoring
 856  * the actual neighbor cache.  The neighbor cache is ignored in order
 857  * to not reenter the routing code from within itself.
 858  */
 859 static int
 860 nd6_is_new_addr_neighbor(addr, ifp)
 861         struct sockaddr_in6 *addr;
 862         struct ifnet *ifp;
 863 {
 864         struct nd_prefix *pr;
 865
 866         /*
 867          * A link-local address is always a neighbor.
 868          * XXX: a link does not necessarily specify a single interface.
 869          */
 870         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
 871                 struct sockaddr_in6 sin6_copy;
 872                 u_int32_t zone;
 873
 874                 /*
 875                  * We need sin6_copy since sa6_recoverscope() may modify the
 876                  * content (XXX).
 877                  */
 878                 sin6_copy = *addr;
 879                 if (sa6_recoverscope(&sin6_copy))
 880                         return (0); /* XXX: should be impossible */
 881                 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
 882                         return (0);
 883                 if (sin6_copy.sin6_scope_id == zone)
 884                         return (1);
 885                 else
 886                         return (0);
 887         }
 888
 889         /*
 890          * If the address matches one of our addresses,
 891          * it should be a neighbor.
 892          * If the address matches one of our on-link prefixes, it should be a
 893          * neighbor.
 894          */
 895         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
 896                 if (pr->ndpr_ifp != ifp)
 897                         continue;
 898
 899                 if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
 900                         continue;
 901
 902                 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
 903                     &addr->sin6_addr, &pr->ndpr_mask))
 904                         return (1);
 905         }
 906
 907         /*
 908          * If the default router list is empty, all addresses are regarded
 909          * as on-link, and thus, as a neighbor.
 910          * XXX: we restrict the condition to hosts, because routers usually do
 911          * not have the "default router list".
 912          */
 913         if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
 914             nd6_defifindex == ifp->if_index) {
 915                 return (1);
 916         }
 917
 918         return (0);
 919 }
 920
 921
 922 /*
 923  * Detect if a given IPv6 address identifies a neighbor on a given link.
 924  * XXX: should take care of the destination of a p2p link?
 925  */
 926 int
 927 nd6_is_addr_neighbor(addr, ifp)
 928         struct sockaddr_in6 *addr;
 929         struct ifnet *ifp;
 930 {
 931
 932         if (nd6_is_new_addr_neighbor(addr, ifp))
 933                 return (1);
 934
 935         /*
 936          * Even if the address matches none of our addresses, it might be
 937          * in the neighbor cache.
 938          */
 939         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 940                 return (1);
 941
 942         return (0);
 943 }
 944
 945 /*
 946  * Free an nd6 llinfo entry.
 947  */
 948 struct llinfo_nd6 *
 949 nd6_free(rt)
 950         struct rtentry *rt;
 951 {
 952         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 953         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 954         struct nd_defrouter *dr;
 955
 956         /*
 957          * we used to have pfctlinput(PRC_HOSTDEAD) here.
 958          * even though it is not harmful, it was not really necessary.
 959          */
 960
 961         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 962                 int s;
 963                 s = splnet();
 964                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 965                     rt->rt_ifp);
 966
 967                 if (ln->ln_router || dr) {
 968                         /*
 969                          * rt6_flush must be called whether or not the neighbor
 970                          * is in the Default Router List.
 971                          * See a corresponding comment in nd6_na_input().
 972                          */
 973                         rt6_flush(&in6, rt->rt_ifp);
 974                 }
 975
 976                 if (dr) {
 977                         /*
 978                          * Unreachablity of a router might affect the default
 979                          * router selection and on-link detection of advertised
 980                          * prefixes.
 981                          */
 982
 983                         /*
 984                          * Temporarily fake the state to choose a new default
 985                          * router and to perform on-link determination of
 986                          * prefixes correctly.
 987                          * Below the state will be set correctly,
 988                          * or the entry itself will be deleted.
 989                          */
 990                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 991
 992                         /*
 993                          * Since defrouter_select() does not affect the
 994                          * on-link determination and MIP6 needs the check
 995                          * before the default router selection, we perform
 996                          * the check now.
 997                          */
 998                         pfxlist_onlink_check();
 999
1000                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1001                                 /*
1002                                  * It is used as the current default router,
1003                                  * so we have to move it to the end of the
1004                                  * list and choose a new one.
1005                                  * XXX: it is not very efficient if this is
1006                                  *      the only router.
1007                                  */
1008                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1009                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1010
1011                                 defrouter_select();
1012                         }
1013                 }
1014                 splx(s);
1015         }
1016
1017         /*
1018          * Before deleting the entry, remember the next entry as the
1019          * return value.  We need this because pfxlist_onlink_check() above
1020          * might have freed other entries (particularly the old next entry) as
1021          * a side effect (XXX).
1022          */
1023         next = ln->ln_next;
1024
1025         /*
1026          * Detach the route from the routing tree and the list of neighbor
1027          * caches, and disable the route entry not to be used in already
1028          * cached routes.
1029          */
1030         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1031             rt_mask(rt), 0, (struct rtentry **)0);
1032
1033         return (next);
1034 }
1035
1036 /*
1037  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1038  *
1039  * XXX cost-effective methods?
1040  */
1041 void
1042 nd6_nud_hint(rt, dst6, force)
1043         struct rtentry *rt;
1044         struct in6_addr *dst6;
1045         int force;
1046 {
1047         struct llinfo_nd6 *ln;
1048
1049         /*
1050          * If the caller specified "rt", use that.  Otherwise, resolve the
1051          * routing table by supplied "dst6".
1052          */
1053         if (!rt) {
1054                 if (!dst6)
1055                         return;
1056                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1057                         return;
1058         }
1059
1060         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1061             (rt->rt_flags & RTF_LLINFO) == 0 ||
1062             !rt->rt_llinfo || !rt->rt_gateway ||
1063             rt->rt_gateway->sa_family != AF_LINK) {
1064                 /* This is not a host route. */
1065                 return;
1066         }
1067
1068         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1069         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1070                 return;
1071
1072         /*
1073          * if we get upper-layer reachability confirmation many times,
1074          * it is possible we have false information.
1075          */
1076         if (!force) {
1077                 ln->ln_byhint++;
1078                 if (ln->ln_byhint > nd6_maxnudhint)
1079                         return;
1080         }
1081
1082         ln->ln_state = ND6_LLINFO_REACHABLE;
1083         if (ln->ln_expire)
1084                 ln->ln_expire = time_second +
1085                         ND_IFINFO(rt->rt_ifp)->reachable;
1086 }
1087
1088 void
1089 nd6_rtrequest(req, rt, info)
1090         int     req;
1091         struct rtentry *rt;
1092         struct rt_addrinfo *info; /* xxx unused */
1093 {
1094         struct sockaddr *gate = rt->rt_gateway;
1095         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1096         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1097         struct ifnet *ifp = rt->rt_ifp;
1098         struct ifaddr *ifa;
1099
1100         RT_LOCK_ASSERT(rt);
1101
1102         if ((rt->rt_flags & RTF_GATEWAY) != 0)
1103                 return;
1104
1105         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1106                 /*
1107                  * This is probably an interface direct route for a link
1108                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1109                  * We do not need special treatment below for such a route.
1110                  * Moreover, the RTF_LLINFO flag which would be set below
1111                  * would annoy the ndp(8) command.
1112                  */
1113                 return;
1114         }
1115
1116         if (req == RTM_RESOLVE &&
1117             (nd6_need_cache(ifp) == 0 || /* stf case */
1118              !nd6_is_new_addr_neighbor((struct sockaddr_in6 *)rt_key(rt),
1119              ifp))) {
1120                 /*
1121                  * FreeBSD and BSD/OS often make a cloned host route based
1122                  * on a less-specific route (e.g. the default route).
1123                  * If the less specific route does not have a "gateway"
1124                  * (this is the case when the route just goes to a p2p or an
1125                  * stf interface), we'll mistakenly make a neighbor cache for
1126                  * the host route, and will see strange neighbor solicitation
1127                  * for the corresponding destination.  In order to avoid the
1128                  * confusion, we check if the destination of the route is
1129                  * a neighbor in terms of neighbor discovery, and stop the
1130                  * process if not.  Additionally, we remove the LLINFO flag
1131                  * so that ndp(8) will not try to get the neighbor information
1132                  * of the destination.
1133                  */
1134                 rt->rt_flags &= ~RTF_LLINFO;
1135                 return;
1136         }
1137
1138         switch (req) {
1139         case RTM_ADD:
1140                 /*
1141                  * There is no backward compatibility :)
1142                  *
1143                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1144                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1145                  *         rt->rt_flags |= RTF_CLONING;
1146                  */
1147                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1148                         /*
1149                          * Case 1: This route should come from
1150                          * a route to interface.  RTF_LLINFO flag is set
1151                          * for a host route whose destination should be
1152                          * treated as on-link.
1153                          */
1154                         rt_setgate(rt, rt_key(rt),
1155                                    (struct sockaddr *)&null_sdl);
1156                         gate = rt->rt_gateway;
1157                         SDL(gate)->sdl_type = ifp->if_type;
1158                         SDL(gate)->sdl_index = ifp->if_index;
1159                         if (ln)
1160                                 ln->ln_expire = time_second;
1161                         if (ln && ln->ln_expire == 0) {
1162                                 /* kludge for desktops */
1163                                 ln->ln_expire = 1;
1164                         }
1165                         if ((rt->rt_flags & RTF_CLONING) != 0)
1166                                 break;
1167                 }
1168                 /*
1169                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1170                  * We don't do that here since llinfo is not ready yet.
1171                  *
1172                  * There are also couple of other things to be discussed:
1173                  * - unsolicited NA code needs improvement beforehand
1174                  * - RFC2461 says we MAY send multicast unsolicited NA
1175                  *   (7.2.6 paragraph 4), however, it also says that we
1176                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1177                  *   we don't have anything like it right now.
1178                  *   note that the mechanism needs a mutual agreement
1179                  *   between proxies, which means that we need to implement
1180                  *   a new protocol, or a new kludge.
1181                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1182                  *   we need to check ip6forwarding before sending it.
1183                  *   (or should we allow proxy ND configuration only for
1184                  *   routers?  there's no mention about proxy ND from hosts)
1185                  */
1186 #if 0
1187                 /* XXX it does not work */
1188                 if (rt->rt_flags & RTF_ANNOUNCE)
1189                         nd6_na_output(ifp,
1190                               &SIN6(rt_key(rt))->sin6_addr,
1191                               &SIN6(rt_key(rt))->sin6_addr,
1192                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1193                               1, NULL);
1194 #endif
1195                 /* FALLTHROUGH */
1196         case RTM_RESOLVE:
1197                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1198                         /*
1199                          * Address resolution isn't necessary for a point to
1200                          * point link, so we can skip this test for a p2p link.
1201                          */
1202                         if (gate->sa_family != AF_LINK ||
1203                             gate->sa_len < sizeof(null_sdl)) {
1204                                 log(LOG_DEBUG,
1205                                     "nd6_rtrequest: bad gateway value: %s\n",
1206                                     if_name(ifp));
1207                                 break;
1208                         }
1209                         SDL(gate)->sdl_type = ifp->if_type;
1210                         SDL(gate)->sdl_index = ifp->if_index;
1211                 }
1212                 if (ln != NULL)
1213                         break;  /* This happens on a route change */
1214                 /*
1215                  * Case 2: This route may come from cloning, or a manual route
1216                  * add with a LL address.
1217                  */
1218                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1219                 rt->rt_llinfo = (caddr_t)ln;
1220                 if (!ln) {
1221                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1222                         break;
1223                 }
1224                 nd6_inuse++;
1225                 nd6_allocated++;
1226                 bzero(ln, sizeof(*ln));
1227                 ln->ln_rt = rt;
1228                 /* this is required for "ndp" command. - shin */
1229                 if (req == RTM_ADD) {
1230                         /*
1231                          * gate should have some valid AF_LINK entry,
1232                          * and ln->ln_expire should have some lifetime
1233                          * which is specified by ndp command.
1234                          */
1235                         ln->ln_state = ND6_LLINFO_REACHABLE;
1236                         ln->ln_byhint = 0;
1237                 } else {
1238                         /*
1239                          * When req == RTM_RESOLVE, rt is created and
1240                          * initialized in rtrequest(), so rt_expire is 0.
1241                          */
1242                         ln->ln_state = ND6_LLINFO_NOSTATE;
1243                         ln->ln_expire = time_second;
1244                 }
1245                 rt->rt_flags |= RTF_LLINFO;
1246                 ln->ln_next = llinfo_nd6.ln_next;
1247                 llinfo_nd6.ln_next = ln;
1248                 ln->ln_prev = &llinfo_nd6;
1249                 ln->ln_next->ln_prev = ln;
1250
1251                 /*
1252                  * check if rt_key(rt) is one of my address assigned
1253                  * to the interface.
1254                  */
1255                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1256                     &SIN6(rt_key(rt))->sin6_addr);
1257                 if (ifa) {
1258                         caddr_t macp = nd6_ifptomac(ifp);
1259                         ln->ln_expire = 0;
1260                         ln->ln_state = ND6_LLINFO_REACHABLE;
1261                         ln->ln_byhint = 0;
1262                         if (macp) {
1263                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1264                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1265                         }
1266                         if (nd6_useloopback) {
1267                                 rt->rt_ifp = &loif[0];  /* XXX */
1268                                 /*
1269                                  * Make sure rt_ifa be equal to the ifaddr
1270                                  * corresponding to the address.
1271                                  * We need this because when we refer
1272                                  * rt_ifa->ia6_flags in ip6_input, we assume
1273                                  * that the rt_ifa points to the address instead
1274                                  * of the loopback address.
1275                                  */
1276                                 if (ifa != rt->rt_ifa) {
1277                                         IFAFREE(rt->rt_ifa);
1278                                         IFAREF(ifa);
1279                                         rt->rt_ifa = ifa;
1280                                 }
1281                         }
1282                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1283                         ln->ln_expire = 0;
1284                         ln->ln_state = ND6_LLINFO_REACHABLE;
1285                         ln->ln_byhint = 0;
1286
1287                         /* join solicited node multicast for proxy ND */
1288                         if (ifp->if_flags & IFF_MULTICAST) {
1289                                 struct in6_addr llsol;
1290                                 int error;
1291
1292                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1293                                 llsol.s6_addr16[0] = htons(0xff02);
1294                                 llsol.s6_addr32[1] = 0;
1295                                 llsol.s6_addr32[2] = htonl(1);
1296                                 llsol.s6_addr8[12] = 0xff;
1297                                 if (in6_setscope(&llsol, ifp, NULL))
1298                                         break;
1299                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1300                                         nd6log((LOG_ERR, "%s: failed to join "
1301                                             "%s (errno=%d)\n", if_name(ifp),
1302                                             ip6_sprintf(&llsol), error));
1303                                 }
1304                         }
1305                 }
1306                 break;
1307
1308         case RTM_DELETE:
1309                 if (!ln)
1310                         break;
1311                 /* leave from solicited node multicast for proxy ND */
1312                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1313                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1314                         struct in6_addr llsol;
1315                         struct in6_multi *in6m;
1316
1317                         llsol = SIN6(rt_key(rt))->sin6_addr;
1318                         llsol.s6_addr16[0] = htons(0xff02);
1319                         llsol.s6_addr32[1] = 0;
1320                         llsol.s6_addr32[2] = htonl(1);
1321                         llsol.s6_addr8[12] = 0xff;
1322                         if (in6_setscope(&llsol, ifp, NULL) == 0) {
1323                                 IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1324                                 if (in6m)
1325                                         in6_delmulti(in6m);
1326                         } else
1327                                 ; /* XXX: should not happen. bark here? */
1328                 }
1329                 nd6_inuse--;
1330                 ln->ln_next->ln_prev = ln->ln_prev;
1331                 ln->ln_prev->ln_next = ln->ln_next;
1332                 ln->ln_prev = NULL;
1333                 rt->rt_llinfo = 0;
1334                 rt->rt_flags &= ~RTF_LLINFO;
1335                 if (ln->ln_hold)
1336                         m_freem(ln->ln_hold);
1337                 Free((caddr_t)ln);
1338         }
1339 }
1340
1341 int
1342 nd6_ioctl(cmd, data, ifp)
1343         u_long cmd;
1344         caddr_t data;
1345         struct ifnet *ifp;
1346 {
1347         struct in6_drlist *drl = (struct in6_drlist *)data;
1348         struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1349         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1350         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1351         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1352         struct nd_defrouter *dr, any;
1353         struct nd_prefix *pr;
1354         struct rtentry *rt;
1355         int i = 0, error = 0;
1356         int s;
1357
1358         switch (cmd) {
1359         case SIOCGDRLST_IN6:
1360                 /*
1361                  * obsolete API, use sysctl under net.inet6.icmp6
1362                  */
1363                 bzero(drl, sizeof(*drl));
1364                 s = splnet();
1365                 dr = TAILQ_FIRST(&nd_defrouter);
1366                 while (dr && i < DRLSTSIZ) {
1367                         drl->defrouter[i].rtaddr = dr->rtaddr;
1368                         in6_clearscope(&drl->defrouter[i].rtaddr);
1369
1370                         drl->defrouter[i].flags = dr->flags;
1371                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1372                         drl->defrouter[i].expire = dr->expire;
1373                         drl->defrouter[i].if_index = dr->ifp->if_index;
1374                         i++;
1375                         dr = TAILQ_NEXT(dr, dr_entry);
1376                 }
1377                 splx(s);
1378                 break;
1379         case SIOCGPRLST_IN6:
1380                 /*
1381                  * obsolete API, use sysctl under net.inet6.icmp6
1382                  *
1383                  * XXX the structure in6_prlist was changed in backward-
1384                  * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1385                  * in6_prlist is used for nd6_sysctl() - fill_prlist().
1386                  */
1387                 /*
1388                  * XXX meaning of fields, especialy "raflags", is very
1389                  * differnet between RA prefix list and RR/static prefix list.
1390                  * how about separating ioctls into two?
1391                  */
1392                 bzero(oprl, sizeof(*oprl));
1393                 s = splnet();
1394                 pr = nd_prefix.lh_first;
1395                 while (pr && i < PRLSTSIZ) {
1396                         struct nd_pfxrouter *pfr;
1397                         int j;
1398
1399                         oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1400                         oprl->prefix[i].raflags = pr->ndpr_raf;
1401                         oprl->prefix[i].prefixlen = pr->ndpr_plen;
1402                         oprl->prefix[i].vltime = pr->ndpr_vltime;
1403                         oprl->prefix[i].pltime = pr->ndpr_pltime;
1404                         oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1405                         oprl->prefix[i].expire = pr->ndpr_expire;
1406
1407                         pfr = pr->ndpr_advrtrs.lh_first;
1408                         j = 0;
1409                         while (pfr) {
1410                                 if (j < DRLSTSIZ) {
1411 #define RTRADDR oprl->prefix[i].advrtr[j]
1412                                         RTRADDR = pfr->router->rtaddr;
1413                                         in6_clearscope(&RTRADDR);
1414 #undef RTRADDR
1415                                 }
1416                                 j++;
1417                                 pfr = pfr->pfr_next;
1418                         }
1419                         oprl->prefix[i].advrtrs = j;
1420                         oprl->prefix[i].origin = PR_ORIG_RA;
1421
1422                         i++;
1423                         pr = pr->ndpr_next;
1424                 }
1425                 splx(s);
1426
1427                 break;
1428         case OSIOCGIFINFO_IN6:
1429                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1430                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1431                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1432                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1433                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1434                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1435                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1436                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1437                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1438                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1439                 break;
1440         case SIOCGIFINFO_IN6:
1441                 ndi->ndi = *ND_IFINFO(ifp);
1442                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1443                 break;
1444         case SIOCSIFINFO_FLAGS:
1445                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1446                 break;
1447         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1448                 /* flush default router list */
1449                 /*
1450                  * xxx sumikawa: should not delete route if default
1451                  * route equals to the top of default router list
1452                  */
1453                 bzero(&any, sizeof(any));
1454                 defrouter_delreq(&any, 0);
1455                 defrouter_select();
1456                 /* xxx sumikawa: flush prefix list */
1457                 break;
1458         case SIOCSPFXFLUSH_IN6:
1459         {
1460                 /* flush all the prefix advertised by routers */
1461                 struct nd_prefix *pr, *next;
1462
1463                 s = splnet();
1464                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1465                         struct in6_ifaddr *ia, *ia_next;
1466
1467                         next = pr->ndpr_next;
1468
1469                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1470                                 continue; /* XXX */
1471
1472                         /* do we really have to remove addresses as well? */
1473                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1474                                 /* ia might be removed.  keep the next ptr. */
1475                                 ia_next = ia->ia_next;
1476
1477                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1478                                         continue;
1479
1480                                 if (ia->ia6_ndpr == pr)
1481                                         in6_purgeaddr(&ia->ia_ifa);
1482                         }
1483                         prelist_remove(pr);
1484                 }
1485                 splx(s);
1486                 break;
1487         }
1488         case SIOCSRTRFLUSH_IN6:
1489         {
1490                 /* flush all the default routers */
1491                 struct nd_defrouter *dr, *next;
1492
1493                 s = splnet();
1494                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1495                         /*
1496                          * The first entry of the list may be stored in
1497                          * the routing table, so we'll delete it later.
1498                          */
1499                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1500                                 next = TAILQ_NEXT(dr, dr_entry);
1501                                 defrtrlist_del(dr);
1502                         }
1503                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1504                 }
1505                 splx(s);
1506                 break;
1507         }
1508         case SIOCGNBRINFO_IN6:
1509         {
1510                 struct llinfo_nd6 *ln;
1511                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1512
1513                 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1514                         return (error);
1515
1516                 s = splnet();
1517                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1518                         error = EINVAL;
1519                         splx(s);
1520                         break;
1521                 }
1522                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1523                 nbi->state = ln->ln_state;
1524                 nbi->asked = ln->ln_asked;
1525                 nbi->isrouter = ln->ln_router;
1526                 nbi->expire = ln->ln_expire;
1527                 splx(s);
1528
1529                 break;
1530         }
1531         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1532                 ndif->ifindex = nd6_defifindex;
1533                 break;
1534         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1535                 return (nd6_setdefaultiface(ndif->ifindex));
1536         }
1537         return (error);
1538 }
1539
1540 /*
1541  * Create neighbor cache entry and cache link-layer address,
1542  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1543  */
1544 struct rtentry *
1545 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1546         struct ifnet *ifp;
1547         struct in6_addr *from;
1548         char *lladdr;
1549         int lladdrlen;
1550         int type;       /* ICMP6 type */
1551         int code;       /* type dependent information */
1552 {
1553         struct rtentry *rt = NULL;
1554         struct llinfo_nd6 *ln = NULL;
1555         int is_newentry;
1556         struct sockaddr_dl *sdl = NULL;
1557         int do_update;
1558         int olladdr;
1559         int llchange;
1560         int newstate = 0;
1561
1562         if (!ifp)
1563                 panic("ifp == NULL in nd6_cache_lladdr");
1564         if (!from)
1565                 panic("from == NULL in nd6_cache_lladdr");
1566
1567         /* nothing must be updated for unspecified address */
1568         if (IN6_IS_ADDR_UNSPECIFIED(from))
1569                 return NULL;
1570
1571         /*
1572          * Validation about ifp->if_addrlen and lladdrlen must be done in
1573          * the caller.
1574          *
1575          * XXX If the link does not have link-layer adderss, what should
1576          * we do? (ifp->if_addrlen == 0)
1577          * Spec says nothing in sections for RA, RS and NA.  There's small
1578          * description on it in NS section (RFC 2461 7.2.3).
1579          */
1580
1581         rt = nd6_lookup(from, 0, ifp);
1582         if (!rt) {
1583 #if 0
1584                 /* nothing must be done if there's no lladdr */
1585                 if (!lladdr || !lladdrlen)
1586                         return NULL;
1587 #endif
1588
1589                 rt = nd6_lookup(from, 1, ifp);
1590                 is_newentry = 1;
1591         } else {
1592                 /* do nothing if static ndp is set */
1593                 if (rt->rt_flags & RTF_STATIC)
1594                         return NULL;
1595                 is_newentry = 0;
1596         }
1597
1598         if (!rt)
1599                 return NULL;
1600         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1601 fail:
1602                 (void)nd6_free(rt);
1603                 return NULL;
1604         }
1605         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1606         if (!ln)
1607                 goto fail;
1608         if (!rt->rt_gateway)
1609                 goto fail;
1610         if (rt->rt_gateway->sa_family != AF_LINK)
1611                 goto fail;
1612         sdl = SDL(rt->rt_gateway);
1613
1614         olladdr = (sdl->sdl_alen) ? 1 : 0;
1615         if (olladdr && lladdr) {
1616                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1617                         llchange = 1;
1618                 else
1619                         llchange = 0;
1620         } else
1621                 llchange = 0;
1622
1623         /*
1624          * newentry olladdr  lladdr  llchange   (*=record)
1625          *      0       n       n       --      (1)
1626          *      0       y       n       --      (2)
1627          *      0       n       y       --      (3) * STALE
1628          *      0       y       y       n       (4) *
1629          *      0       y       y       y       (5) * STALE
1630          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1631          *      1       --      y       --      (7) * STALE
1632          */
1633
1634         if (lladdr) {           /* (3-5) and (7) */
1635                 /*
1636                  * Record source link-layer address
1637                  * XXX is it dependent to ifp->if_type?
1638                  */
1639                 sdl->sdl_alen = ifp->if_addrlen;
1640                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1641         }
1642
1643         if (!is_newentry) {
1644                 if ((!olladdr && lladdr) ||             /* (3) */
1645                     (olladdr && lladdr && llchange)) {  /* (5) */
1646                         do_update = 1;
1647                         newstate = ND6_LLINFO_STALE;
1648                 } else                                  /* (1-2,4) */
1649                         do_update = 0;
1650         } else {
1651                 do_update = 1;
1652                 if (!lladdr)                            /* (6) */
1653                         newstate = ND6_LLINFO_NOSTATE;
1654                 else                                    /* (7) */
1655                         newstate = ND6_LLINFO_STALE;
1656         }
1657
1658         if (do_update) {
1659                 /*
1660                  * Update the state of the neighbor cache.
1661                  */
1662                 ln->ln_state = newstate;
1663
1664                 if (ln->ln_state == ND6_LLINFO_STALE) {
1665                         /*
1666                          * XXX: since nd6_output() below will cause
1667                          * state tansition to DELAY and reset the timer,
1668                          * we must set the timer now, although it is actually
1669                          * meaningless.
1670                          */
1671                         ln->ln_expire = time_second + nd6_gctimer;
1672
1673                         if (ln->ln_hold) {
1674                                 /*
1675                                  * we assume ifp is not a p2p here, so just
1676                                  * set the 2nd argument as the 1st one.
1677                                  */
1678                                 nd6_output(ifp, ifp, ln->ln_hold,
1679                                     (struct sockaddr_in6 *)rt_key(rt), rt);
1680                                 ln->ln_hold = NULL;
1681                         }
1682                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1683                         /* probe right away */
1684                         ln->ln_expire = time_second;
1685                 }
1686         }
1687
1688         /*
1689          * ICMP6 type dependent behavior.
1690          *
1691          * NS: clear IsRouter if new entry
1692          * RS: clear IsRouter
1693          * RA: set IsRouter if there's lladdr
1694          * redir: clear IsRouter if new entry
1695          *
1696          * RA case, (1):
1697          * The spec says that we must set IsRouter in the following cases:
1698          * - If lladdr exist, set IsRouter.  This means (1-5).
1699          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1700          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1701          * A quetion arises for (1) case.  (1) case has no lladdr in the
1702          * neighbor cache, this is similar to (6).
1703          * This case is rare but we figured that we MUST NOT set IsRouter.
1704          *
1705          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1706          *                                                      D R
1707          *      0       n       n       --      (1)     c   ?     s
1708          *      0       y       n       --      (2)     c   s     s
1709          *      0       n       y       --      (3)     c   s     s
1710          *      0       y       y       n       (4)     c   s     s
1711          *      0       y       y       y       (5)     c   s     s
1712          *      1       --      n       --      (6) c   c       c s
1713          *      1       --      y       --      (7) c   c   s   c s
1714          *
1715          *                                      (c=clear s=set)
1716          */
1717         switch (type & 0xff) {
1718         case ND_NEIGHBOR_SOLICIT:
1719                 /*
1720                  * New entry must have is_router flag cleared.
1721                  */
1722                 if (is_newentry)        /* (6-7) */
1723                         ln->ln_router = 0;
1724                 break;
1725         case ND_REDIRECT:
1726                 /*
1727                  * If the icmp is a redirect to a better router, always set the
1728                  * is_router flag.  Otherwise, if the entry is newly created,
1729                  * clear the flag.  [RFC 2461, sec 8.3]
1730                  */
1731                 if (code == ND_REDIRECT_ROUTER)
1732                         ln->ln_router = 1;
1733                 else if (is_newentry) /* (6-7) */
1734                         ln->ln_router = 0;
1735                 break;
1736         case ND_ROUTER_SOLICIT:
1737                 /*
1738                  * is_router flag must always be cleared.
1739                  */
1740                 ln->ln_router = 0;
1741                 break;
1742         case ND_ROUTER_ADVERT:
1743                 /*
1744                  * Mark an entry with lladdr as a router.
1745                  */
1746                 if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
1747                     (is_newentry && lladdr)) {                  /* (7) */
1748                         ln->ln_router = 1;
1749                 }
1750                 break;
1751         }
1752
1753         /*
1754          * When the link-layer address of a router changes, select the
1755          * best router again.  In particular, when the neighbor entry is newly
1756          * created, it might affect the selection policy.
1757          * Question: can we restrict the first condition to the "is_newentry"
1758          * case?
1759          * XXX: when we hear an RA from a new router with the link-layer
1760          * address option, defrouter_select() is called twice, since
1761          * defrtrlist_update called the function as well.  However, I believe
1762          * we can compromise the overhead, since it only happens the first
1763          * time.
1764          * XXX: although defrouter_select() should not have a bad effect
1765          * for those are not autoconfigured hosts, we explicitly avoid such
1766          * cases for safety.
1767          */
1768         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1769                 defrouter_select();
1770
1771         return rt;
1772 }
1773
1774 static void
1775 nd6_slowtimo(ignored_arg)
1776     void *ignored_arg;
1777 {
1778         int s = splnet();
1779         struct nd_ifinfo *nd6if;
1780         struct ifnet *ifp;
1781
1782         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1783             nd6_slowtimo, NULL);
1784         IFNET_RLOCK();
1785         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1786                 nd6if = ND_IFINFO(ifp);
1787                 if (nd6if->basereachable && /* already initialized */
1788                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1789                         /*
1790                          * Since reachable time rarely changes by router
1791                          * advertisements, we SHOULD insure that a new random
1792                          * value gets recomputed at least once every few hours.
1793                          * (RFC 2461, 6.3.4)
1794                          */
1795                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1796                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1797                 }
1798         }
1799         IFNET_RUNLOCK();
1800         splx(s);
1801 }
1802
1803 #define senderr(e) { error = (e); goto bad;}
1804 int
1805 nd6_output(ifp, origifp, m0, dst, rt0)
1806         struct ifnet *ifp;
1807         struct ifnet *origifp;
1808         struct mbuf *m0;
1809         struct sockaddr_in6 *dst;
1810         struct rtentry *rt0;
1811 {
1812         struct mbuf *m = m0;
1813         struct rtentry *rt = rt0;
1814         struct sockaddr_in6 *gw6 = NULL;
1815         struct llinfo_nd6 *ln = NULL;
1816         int error = 0;
1817
1818         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1819                 goto sendpkt;
1820
1821         if (nd6_need_cache(ifp) == 0)
1822                 goto sendpkt;
1823
1824         /*
1825          * next hop determination.  This routine is derived from ether_output.
1826          */
1827 again:
1828         if (rt) {
1829                 if ((rt->rt_flags & RTF_UP) == 0) {
1830                         rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL);
1831                         if (rt != NULL) {
1832                                 RT_REMREF(rt);
1833                                 RT_UNLOCK(rt);
1834                                 if (rt->rt_ifp != ifp)
1835                                         /*
1836                                          * XXX maybe we should update ifp too,
1837                                          * but the original code didn't and I
1838                                          * don't know what is correct here.
1839                                          */
1840                                         goto again;
1841                         } else
1842                                 senderr(EHOSTUNREACH);
1843                 }
1844
1845                 if (rt->rt_flags & RTF_GATEWAY) {
1846                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1847
1848                         /*
1849                          * We skip link-layer address resolution and NUD
1850                          * if the gateway is not a neighbor from ND point
1851                          * of view, regardless of the value of nd_ifinfo.flags.
1852                          * The second condition is a bit tricky; we skip
1853                          * if the gateway is our own address, which is
1854                          * sometimes used to install a route to a p2p link.
1855                          */
1856                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1857                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1858                                 /*
1859                                  * We allow this kind of tricky route only
1860                                  * when the outgoing interface is p2p.
1861                                  * XXX: we may need a more generic rule here.
1862                                  */
1863                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1864                                         senderr(EHOSTUNREACH);
1865
1866                                 goto sendpkt;
1867                         }
1868
1869                         if (rt->rt_gwroute == 0)
1870                                 goto lookup;
1871                         if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1872                                 RT_LOCK(rt);
1873                                 rtfree(rt); rt = rt0;
1874                         lookup:
1875                                 rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1876                                 if ((rt = rt->rt_gwroute) == 0)
1877                                         senderr(EHOSTUNREACH);
1878                                 RT_UNLOCK(rt);
1879                         }
1880                 }
1881         }
1882
1883         /*
1884          * Address resolution or Neighbor Unreachability Detection
1885          * for the next hop.
1886          * At this point, the destination of the packet must be a unicast
1887          * or an anycast address(i.e. not a multicast).
1888          */
1889
1890         /* Look up the neighbor cache for the nexthop */
1891         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1892                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1893         else {
1894                 /*
1895                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1896                  * the condition below is not very efficient.  But we believe
1897                  * it is tolerable, because this should be a rare case.
1898                  */
1899                 if (nd6_is_addr_neighbor(dst, ifp) &&
1900                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1901                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1902         }
1903         if (!ln || !rt) {
1904                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1905                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1906                         log(LOG_DEBUG,
1907                             "nd6_output: can't allocate llinfo for %s "
1908                             "(ln=%p, rt=%p)\n",
1909                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1910                         senderr(EIO);   /* XXX: good error? */
1911                 }
1912
1913                 goto sendpkt;   /* send anyway */
1914         }
1915
1916         /* We don't have to do link-layer address resolution on a p2p link. */
1917         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1918             ln->ln_state < ND6_LLINFO_REACHABLE) {
1919                 ln->ln_state = ND6_LLINFO_STALE;
1920                 ln->ln_expire = time_second + nd6_gctimer;
1921         }
1922
1923         /*
1924          * The first time we send a packet to a neighbor whose entry is
1925          * STALE, we have to change the state to DELAY and a sets a timer to
1926          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1927          * neighbor unreachability detection on expiration.
1928          * (RFC 2461 7.3.3)
1929          */
1930         if (ln->ln_state == ND6_LLINFO_STALE) {
1931                 ln->ln_asked = 0;
1932                 ln->ln_state = ND6_LLINFO_DELAY;
1933                 ln->ln_expire = time_second + nd6_delay;
1934         }
1935
1936         /*
1937          * If the neighbor cache entry has a state other than INCOMPLETE
1938          * (i.e. its link-layer address is already resolved), just
1939          * send the packet.
1940          */
1941         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1942                 goto sendpkt;
1943
1944         /*
1945          * There is a neighbor cache entry, but no ethernet address
1946          * response yet.  Replace the held mbuf (if any) with this
1947          * latest one.
1948          *
1949          * This code conforms to the rate-limiting rule described in Section
1950          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1951          * an NS below.
1952          */
1953         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1954                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1955         if (ln->ln_hold)
1956                 m_freem(ln->ln_hold);
1957         ln->ln_hold = m;
1958         if (ln->ln_expire) {
1959                 if (ln->ln_asked < nd6_mmaxtries &&
1960                     ln->ln_expire < time_second) {
1961                         ln->ln_asked++;
1962                         ln->ln_expire = time_second +
1963                                 ND_IFINFO(ifp)->retrans / 1000;
1964                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1965                 }
1966         }
1967         return (0);
1968
1969   sendpkt:
1970 #ifdef IPSEC
1971         /* clean ipsec history once it goes out of the node */
1972         ipsec_delaux(m);
1973 #endif
1974
1975 #ifdef MAC
1976         mac_create_mbuf_linklayer(ifp, m);
1977 #endif
1978         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1979                 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1980                     rt));
1981         }
1982         return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1983
1984   bad:
1985         if (m)
1986                 m_freem(m);
1987         return (error);
1988 }
1989 #undef senderr
1990
1991 int
1992 nd6_need_cache(ifp)
1993         struct ifnet *ifp;
1994 {
1995         /*
1996          * XXX: we currently do not make neighbor cache on any interface
1997          * other than ARCnet, Ethernet, FDDI and GIF.
1998          *
1999          * RFC2893 says:
2000          * - unidirectional tunnels needs no ND
2001          */
2002         switch (ifp->if_type) {
2003         case IFT_ARCNET:
2004         case IFT_ETHER:
2005         case IFT_FDDI:
2006         case IFT_IEEE1394:
2007 #ifdef IFT_L2VLAN
2008         case IFT_L2VLAN:
2009 #endif
2010 #ifdef IFT_IEEE80211
2011         case IFT_IEEE80211:
2012 #endif
2013 #ifdef IFT_CARP
2014         case IFT_CARP:
2015 #endif
2016         case IFT_GIF:           /* XXX need more cases? */
2017         case IFT_BRIDGE:
2018                 return (1);
2019         default:
2020                 return (0);
2021         }
2022 }
2023
2024 int
2025 nd6_storelladdr(ifp, rt0, m, dst, desten)
2026         struct ifnet *ifp;
2027         struct rtentry *rt0;
2028         struct mbuf *m;
2029         struct sockaddr *dst;
2030         u_char *desten;
2031 {
2032         struct sockaddr_dl *sdl;
2033         struct rtentry *rt;
2034         int error;
2035
2036         if (m->m_flags & M_MCAST) {
2037                 int i;
2038
2039                 switch (ifp->if_type) {
2040                 case IFT_ETHER:
2041                 case IFT_FDDI:
2042 #ifdef IFT_L2VLAN
2043                 case IFT_L2VLAN:
2044 #endif
2045 #ifdef IFT_IEEE80211
2046                 case IFT_IEEE80211:
2047 #endif
2048                 case IFT_BRIDGE:
2049                 case IFT_ISO88025:
2050                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2051                                                  desten);
2052                         return (0);
2053                 case IFT_IEEE1394:
2054                         /*
2055                          * netbsd can use if_broadcastaddr, but we don't do so
2056                          * to reduce # of ifdef.
2057                          */
2058                         for (i = 0; i < ifp->if_addrlen; i++)
2059                                 desten[i] = ~0;
2060                         return (0);
2061                 case IFT_ARCNET:
2062                         *desten = 0;
2063                         return (0);
2064                 default:
2065                         m_freem(m);
2066                         return (EAFNOSUPPORT);
2067                 }
2068         }
2069
2070         if (rt0 == NULL) {
2071                 /* this could happen, if we could not allocate memory */
2072                 m_freem(m);
2073                 return (ENOMEM);
2074         }
2075
2076         error = rt_check(&rt, &rt0, dst);
2077         if (error) {
2078                 m_freem(m);
2079                 return (error);
2080         }
2081         RT_UNLOCK(rt);
2082
2083         if (rt->rt_gateway->sa_family != AF_LINK) {
2084                 printf("nd6_storelladdr: something odd happens\n");
2085                 m_freem(m);
2086                 return (EINVAL);
2087         }
2088         sdl = SDL(rt->rt_gateway);
2089         if (sdl->sdl_alen == 0) {
2090                 /* this should be impossible, but we bark here for debugging */
2091                 printf("nd6_storelladdr: sdl_alen == 0\n");
2092                 m_freem(m);
2093                 return (EINVAL);
2094         }
2095
2096         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2097         return (0);
2098 }
2099
2100 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2101 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2102 #ifdef SYSCTL_DECL
2103 SYSCTL_DECL(_net_inet6_icmp6);
2104 #endif
2105 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2106         CTLFLAG_RD, nd6_sysctl_drlist, "");
2107 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2108         CTLFLAG_RD, nd6_sysctl_prlist, "");
2109
2110 static int
2111 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2112 {
2113         int error;
2114         char buf[1024];
2115         struct in6_defrouter *d, *de;
2116         struct nd_defrouter *dr;
2117
2118         if (req->newptr)
2119                 return EPERM;
2120         error = 0;
2121
2122         for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2123              dr = TAILQ_NEXT(dr, dr_entry)) {
2124                 d = (struct in6_defrouter *)buf;
2125                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2126
2127                 if (d + 1 <= de) {
2128                         bzero(d, sizeof(*d));
2129                         d->rtaddr.sin6_family = AF_INET6;
2130                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2131                         d->rtaddr.sin6_addr = dr->rtaddr;
2132                         if (sa6_recoverscope(&d->rtaddr)) {
2133                                 log(LOG_ERR,
2134                                     "scope error in router list (%s)\n",
2135                                     ip6_sprintf(&d->rtaddr.sin6_addr));
2136                                 /* XXX: press on... */
2137                         }
2138                         d->flags = dr->flags;
2139                         d->rtlifetime = dr->rtlifetime;
2140                         d->expire = dr->expire;
2141                         d->if_index = dr->ifp->if_index;
2142                 } else
2143                         panic("buffer too short");
2144
2145                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2146                 if (error)
2147                         break;
2148         }
2149
2150         return (error);
2151 }
2152
2153 static int
2154 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2155 {
2156         int error;
2157         char buf[1024];
2158         struct in6_prefix *p, *pe;
2159         struct nd_prefix *pr;
2160
2161         if (req->newptr)
2162                 return EPERM;
2163         error = 0;
2164
2165         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2166                 u_short advrtrs;
2167                 size_t advance;
2168                 struct sockaddr_in6 *sin6, *s6;
2169                 struct nd_pfxrouter *pfr;
2170
2171                 p = (struct in6_prefix *)buf;
2172                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2173
2174                 if (p + 1 <= pe) {
2175                         bzero(p, sizeof(*p));
2176                         sin6 = (struct sockaddr_in6 *)(p + 1);
2177
2178                         p->prefix = pr->ndpr_prefix;
2179                         if (sa6_recoverscope(&p->prefix)) {
2180                                 log(LOG_ERR,
2181                                     "scope error in prefix list (%s)\n",
2182                                     ip6_sprintf(&p->prefix.sin6_addr));
2183                                 /* XXX: press on... */
2184                         }
2185                         p->raflags = pr->ndpr_raf;
2186                         p->prefixlen = pr->ndpr_plen;
2187                         p->vltime = pr->ndpr_vltime;
2188                         p->pltime = pr->ndpr_pltime;
2189                         p->if_index = pr->ndpr_ifp->if_index;
2190                         p->expire = pr->ndpr_expire;
2191                         p->refcnt = pr->ndpr_refcnt;
2192                         p->flags = pr->ndpr_stateflags;
2193                         p->origin = PR_ORIG_RA;
2194                         advrtrs = 0;
2195                         for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2196                              pfr = pfr->pfr_next) {
2197                                 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2198                                         advrtrs++;
2199                                         continue;
2200                                 }
2201                                 s6 = &sin6[advrtrs];
2202                                 bzero(s6, sizeof(*s6));
2203                                 s6->sin6_family = AF_INET6;
2204                                 s6->sin6_len = sizeof(*sin6);
2205                                 s6->sin6_addr = pfr->router->rtaddr;
2206                                 if (sa6_recoverscope(s6)) {
2207                                         log(LOG_ERR,
2208                                             "scope error in "
2209                                             "prefix list (%s)\n",
2210                                             ip6_sprintf(&pfr->router->rtaddr));
2211                                 }
2212                                 advrtrs++;
2213                         }
2214                         p->advrtrs = advrtrs;
2215                 } else
2216                         panic("buffer too short");
2217
2218                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2219                 error = SYSCTL_OUT(req, buf, advance);
2220                 if (error)
2221                         break;
2222         }
2223
2224         return (error);
2225 }