sys/netinet6/nd6.c

   1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
   2 /*      $DragonFly: src/sys/netinet6/nd6.c,v 1.4 2003/08/07 21:17:33 dillon Exp $       */
   3 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   4
   5 /*
   6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. Neither the name of the project nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * XXX
  36  * KAME 970409 note:
  37  * BSD/OS version heavily modifies this code, related to llinfo.
  38  * Since we don't have BSD/OS version of net/route.c in our hand,
  39  * I left the code mostly as it was in 970310.  -- itojun
  40  */
  41
  42 #include "opt_inet.h"
  43 #include "opt_inet6.h"
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/callout.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mbuf.h>
  50 #include <sys/socket.h>
  51 #include <sys/sockio.h>
  52 #include <sys/time.h>
  53 #include <sys/kernel.h>
  54 #include <sys/protosw.h>
  55 #include <sys/errno.h>
  56 #include <sys/syslog.h>
  57 #include <sys/queue.h>
  58 #include <sys/sysctl.h>
  59
  60 #include <net/if.h>
  61 #include <net/if_dl.h>
  62 #include <net/if_types.h>
  63 #include <net/if_atm.h>
  64 #include <net/route.h>
  65
  66 #include <netinet/in.h>
  67 #include <netinet/if_ether.h>
  68 #include <netinet/if_fddi.h>
  69 #include <netinet6/in6_var.h>
  70 #include <netinet/ip6.h>
  71 #include <netinet6/ip6_var.h>
  72 #include <netinet6/nd6.h>
  73 #include <netinet6/in6_prefix.h>
  74 #include <netinet/icmp6.h>
  75
  76 #include "use_loop.h"
  77
  78 #include <net/net_osdep.h>
  79
  80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  82
  83 #define SIN6(s) ((struct sockaddr_in6 *)s)
  84 #define SDL(s) ((struct sockaddr_dl *)s)
  85
  86 /* timer values */
  87 int     nd6_prune       = 1;    /* walk list every 1 seconds */
  88 int     nd6_delay       = 5;    /* delay first probe time 5 second */
  89 int     nd6_umaxtries   = 3;    /* maximum unicast query */
  90 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
  91 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
  92 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
  93
  94 /* preventing too many loops in ND option parsing */
  95 int nd6_maxndopt = 10;  /* max # of ND options allowed */
  96
  97 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
  98
  99 #ifdef ND6_DEBUG
 100 int nd6_debug = 1;
 101 #else
 102 int nd6_debug = 0;
 103 #endif
 104
 105 /* for debugging? */
 106 static int nd6_inuse, nd6_allocated;
 107
 108 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 109 static size_t nd_ifinfo_indexlim = 8;
 110 struct nd_ifinfo *nd_ifinfo = NULL;
 111 struct nd_drhead nd_defrouter;
 112 struct nd_prhead nd_prefix = { 0 };
 113
 114 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 115 static struct sockaddr_in6 all1_sa;
 116
 117 static void nd6_slowtimo __P((void *));
 118 static int regen_tmpaddr __P((struct in6_ifaddr *));
 119
 120 struct callout nd6_slowtimo_ch;
 121 struct callout nd6_timer_ch;
 122 extern struct callout in6_tmpaddrtimer_ch;
 123
 124 void
 125 nd6_init()
 126 {
 127         static int nd6_init_done = 0;
 128         int i;
 129
 130         if (nd6_init_done) {
 131                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 132                 return;
 133         }
 134
 135         all1_sa.sin6_family = AF_INET6;
 136         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 137         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 138                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 139
 140         /* initialization of the default router list */
 141         TAILQ_INIT(&nd_defrouter);
 142
 143         nd6_init_done = 1;
 144
 145         /* start timer */
 146         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 147             nd6_slowtimo, NULL);
 148 }
 149
 150 void
 151 nd6_ifattach(ifp)
 152         struct ifnet *ifp;
 153 {
 154
 155         /*
 156          * We have some arrays that should be indexed by if_index.
 157          * since if_index will grow dynamically, they should grow too.
 158          */
 159         if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
 160                 size_t n;
 161                 caddr_t q;
 162
 163                 while (if_index >= nd_ifinfo_indexlim)
 164                         nd_ifinfo_indexlim <<= 1;
 165
 166                 /* grow nd_ifinfo */
 167                 n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
 168                 q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
 169                 bzero(q, n);
 170                 if (nd_ifinfo) {
 171                         bcopy((caddr_t)nd_ifinfo, q, n/2);
 172                         free((caddr_t)nd_ifinfo, M_IP6NDP);
 173                 }
 174                 nd_ifinfo = (struct nd_ifinfo *)q;
 175         }
 176
 177 #define ND nd_ifinfo[ifp->if_index]
 178
 179         /*
 180          * Don't initialize if called twice.
 181          * XXX: to detect this, we should choose a member that is never set
 182          * before initialization of the ND structure itself.  We formaly used
 183          * the linkmtu member, which was not suitable because it could be
 184          * initialized via "ifconfig mtu".
 185          */
 186         if (ND.basereachable)
 187                 return;
 188
 189         ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
 190         ND.chlim = IPV6_DEFHLIM;
 191         ND.basereachable = REACHABLE_TIME;
 192         ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
 193         ND.retrans = RETRANS_TIMER;
 194         ND.receivedra = 0;
 195         ND.flags = ND6_IFF_PERFORMNUD;
 196         nd6_setmtu(ifp);
 197 #undef ND
 198 }
 199
 200 /*
 201  * Reset ND level link MTU. This function is called when the physical MTU
 202  * changes, which means we might have to adjust the ND level MTU.
 203  */
 204 void
 205 nd6_setmtu(ifp)
 206         struct ifnet *ifp;
 207 {
 208         struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
 209         u_long oldmaxmtu = ndi->maxmtu;
 210         u_long oldlinkmtu = ndi->linkmtu;
 211
 212         switch (ifp->if_type) {
 213         case IFT_ARCNET:        /* XXX MTU handling needs more work */
 214                 ndi->maxmtu = MIN(60480, ifp->if_mtu);
 215                 break;
 216         case IFT_ETHER:
 217                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 218                 break;
 219         case IFT_FDDI:
 220                 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
 221                 break;
 222         case IFT_ATM:
 223                 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
 224                 break;
 225         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 226                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 227                 break;
 228 #ifdef IFT_IEEE80211
 229         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 230                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 231                 break;
 232 #endif
 233         default:
 234                 ndi->maxmtu = ifp->if_mtu;
 235                 break;
 236         }
 237
 238         if (oldmaxmtu != ndi->maxmtu) {
 239                 /*
 240                  * If the ND level MTU is not set yet, or if the maxmtu
 241                  * is reset to a smaller value than the ND level MTU,
 242                  * also reset the ND level MTU.
 243                  */
 244                 if (ndi->linkmtu == 0 ||
 245                     ndi->maxmtu < ndi->linkmtu) {
 246                         ndi->linkmtu = ndi->maxmtu;
 247                         /* also adjust in6_maxmtu if necessary. */
 248                         if (oldlinkmtu == 0) {
 249                                 /*
 250                                  * XXX: the case analysis is grotty, but
 251                                  * it is not efficient to call in6_setmaxmtu()
 252                                  * here when we are during the initialization
 253                                  * procedure.
 254                                  */
 255                                 if (in6_maxmtu < ndi->linkmtu)
 256                                         in6_maxmtu = ndi->linkmtu;
 257                         } else
 258                                 in6_setmaxmtu();
 259                 }
 260         }
 261 #undef MIN
 262 }
 263
 264 void
 265 nd6_option_init(opt, icmp6len, ndopts)
 266         void *opt;
 267         int icmp6len;
 268         union nd_opts *ndopts;
 269 {
 270         bzero(ndopts, sizeof(*ndopts));
 271         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 272         ndopts->nd_opts_last
 273                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 274
 275         if (icmp6len == 0) {
 276                 ndopts->nd_opts_done = 1;
 277                 ndopts->nd_opts_search = NULL;
 278         }
 279 }
 280
 281 /*
 282  * Take one ND option.
 283  */
 284 struct nd_opt_hdr *
 285 nd6_option(ndopts)
 286         union nd_opts *ndopts;
 287 {
 288         struct nd_opt_hdr *nd_opt;
 289         int olen;
 290
 291         if (!ndopts)
 292                 panic("ndopts == NULL in nd6_option");
 293         if (!ndopts->nd_opts_last)
 294                 panic("uninitialized ndopts in nd6_option");
 295         if (!ndopts->nd_opts_search)
 296                 return NULL;
 297         if (ndopts->nd_opts_done)
 298                 return NULL;
 299
 300         nd_opt = ndopts->nd_opts_search;
 301
 302         /* make sure nd_opt_len is inside the buffer */
 303         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 304                 bzero(ndopts, sizeof(*ndopts));
 305                 return NULL;
 306         }
 307
 308         olen = nd_opt->nd_opt_len << 3;
 309         if (olen == 0) {
 310                 /*
 311                  * Message validation requires that all included
 312                  * options have a length that is greater than zero.
 313                  */
 314                 bzero(ndopts, sizeof(*ndopts));
 315                 return NULL;
 316         }
 317
 318         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 319         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 320                 /* option overruns the end of buffer, invalid */
 321                 bzero(ndopts, sizeof(*ndopts));
 322                 return NULL;
 323         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 324                 /* reached the end of options chain */
 325                 ndopts->nd_opts_done = 1;
 326                 ndopts->nd_opts_search = NULL;
 327         }
 328         return nd_opt;
 329 }
 330
 331 /*
 332  * Parse multiple ND options.
 333  * This function is much easier to use, for ND routines that do not need
 334  * multiple options of the same type.
 335  */
 336 int
 337 nd6_options(ndopts)
 338         union nd_opts *ndopts;
 339 {
 340         struct nd_opt_hdr *nd_opt;
 341         int i = 0;
 342
 343         if (!ndopts)
 344                 panic("ndopts == NULL in nd6_options");
 345         if (!ndopts->nd_opts_last)
 346                 panic("uninitialized ndopts in nd6_options");
 347         if (!ndopts->nd_opts_search)
 348                 return 0;
 349
 350         while (1) {
 351                 nd_opt = nd6_option(ndopts);
 352                 if (!nd_opt && !ndopts->nd_opts_last) {
 353                         /*
 354                          * Message validation requires that all included
 355                          * options have a length that is greater than zero.
 356                          */
 357                         icmp6stat.icp6s_nd_badopt++;
 358                         bzero(ndopts, sizeof(*ndopts));
 359                         return -1;
 360                 }
 361
 362                 if (!nd_opt)
 363                         goto skip1;
 364
 365                 switch (nd_opt->nd_opt_type) {
 366                 case ND_OPT_SOURCE_LINKADDR:
 367                 case ND_OPT_TARGET_LINKADDR:
 368                 case ND_OPT_MTU:
 369                 case ND_OPT_REDIRECTED_HEADER:
 370                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 371                                 nd6log((LOG_INFO,
 372                                     "duplicated ND6 option found (type=%d)\n",
 373                                     nd_opt->nd_opt_type));
 374                                 /* XXX bark? */
 375                         } else {
 376                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 377                                         = nd_opt;
 378                         }
 379                         break;
 380                 case ND_OPT_PREFIX_INFORMATION:
 381                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 382                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 383                                         = nd_opt;
 384                         }
 385                         ndopts->nd_opts_pi_end =
 386                                 (struct nd_opt_prefix_info *)nd_opt;
 387                         break;
 388                 default:
 389                         /*
 390                          * Unknown options must be silently ignored,
 391                          * to accomodate future extension to the protocol.
 392                          */
 393                         nd6log((LOG_DEBUG,
 394                             "nd6_options: unsupported option %d - "
 395                             "option ignored\n", nd_opt->nd_opt_type));
 396                 }
 397
 398 skip1:
 399                 i++;
 400                 if (i > nd6_maxndopt) {
 401                         icmp6stat.icp6s_nd_toomanyopt++;
 402                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 403                         break;
 404                 }
 405
 406                 if (ndopts->nd_opts_done)
 407                         break;
 408         }
 409
 410         return 0;
 411 }
 412
 413 /*
 414  * ND6 timer routine to expire default route list and prefix list
 415  */
 416 void
 417 nd6_timer(ignored_arg)
 418         void    *ignored_arg;
 419 {
 420         int s;
 421         struct llinfo_nd6 *ln;
 422         struct nd_defrouter *dr;
 423         struct nd_prefix *pr;
 424         struct ifnet *ifp;
 425         struct in6_ifaddr *ia6, *nia6;
 426         struct in6_addrlifetime *lt6;
 427
 428         s = splnet();
 429         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 430                       nd6_timer, NULL);
 431
 432         ln = llinfo_nd6.ln_next;
 433         while (ln && ln != &llinfo_nd6) {
 434                 struct rtentry *rt;
 435                 struct sockaddr_in6 *dst;
 436                 struct llinfo_nd6 *next = ln->ln_next;
 437                 /* XXX: used for the DELAY case only: */
 438                 struct nd_ifinfo *ndi = NULL;
 439
 440                 if ((rt = ln->ln_rt) == NULL) {
 441                         ln = next;
 442                         continue;
 443                 }
 444                 if ((ifp = rt->rt_ifp) == NULL) {
 445                         ln = next;
 446                         continue;
 447                 }
 448                 ndi = &nd_ifinfo[ifp->if_index];
 449                 dst = (struct sockaddr_in6 *)rt_key(rt);
 450
 451                 if (ln->ln_expire > time_second) {
 452                         ln = next;
 453                         continue;
 454                 }
 455
 456                 /* sanity check */
 457                 if (!rt)
 458                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 459                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 460                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 461                               rt->rt_llinfo, ln);
 462                 if (!dst)
 463                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 464
 465                 switch (ln->ln_state) {
 466                 case ND6_LLINFO_INCOMPLETE:
 467                         if (ln->ln_asked < nd6_mmaxtries) {
 468                                 ln->ln_asked++;
 469                                 ln->ln_expire = time_second +
 470                                         nd_ifinfo[ifp->if_index].retrans / 1000;
 471                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 472                                         ln, 0);
 473                         } else {
 474                                 struct mbuf *m = ln->ln_hold;
 475                                 if (m) {
 476                                         if (rt->rt_ifp) {
 477                                                 /*
 478                                                  * Fake rcvif to make ICMP error
 479                                                  * more helpful in diagnosing
 480                                                  * for the receiver.
 481                                                  * XXX: should we consider
 482                                                  * older rcvif?
 483                                                  */
 484                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 485                                         }
 486                                         icmp6_error(m, ICMP6_DST_UNREACH,
 487                                                     ICMP6_DST_UNREACH_ADDR, 0);
 488                                         ln->ln_hold = NULL;
 489                                 }
 490                                 next = nd6_free(rt);
 491                         }
 492                         break;
 493                 case ND6_LLINFO_REACHABLE:
 494                         if (ln->ln_expire) {
 495                                 ln->ln_state = ND6_LLINFO_STALE;
 496                                 ln->ln_expire = time_second + nd6_gctimer;
 497                         }
 498                         break;
 499
 500                 case ND6_LLINFO_STALE:
 501                         /* Garbage Collection(RFC 2461 5.3) */
 502                         if (ln->ln_expire)
 503                                 next = nd6_free(rt);
 504                         break;
 505
 506                 case ND6_LLINFO_DELAY:
 507                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
 508                                 /* We need NUD */
 509                                 ln->ln_asked = 1;
 510                                 ln->ln_state = ND6_LLINFO_PROBE;
 511                                 ln->ln_expire = time_second +
 512                                         ndi->retrans / 1000;
 513                                 nd6_ns_output(ifp, &dst->sin6_addr,
 514                                               &dst->sin6_addr,
 515                                               ln, 0);
 516                         } else {
 517                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 518                                 ln->ln_expire = time_second + nd6_gctimer;
 519                         }
 520                         break;
 521                 case ND6_LLINFO_PROBE:
 522                         if (ln->ln_asked < nd6_umaxtries) {
 523                                 ln->ln_asked++;
 524                                 ln->ln_expire = time_second +
 525                                         nd_ifinfo[ifp->if_index].retrans / 1000;
 526                                 nd6_ns_output(ifp, &dst->sin6_addr,
 527                                                &dst->sin6_addr, ln, 0);
 528                         } else {
 529                                 next = nd6_free(rt);
 530                         }
 531                         break;
 532                 }
 533                 ln = next;
 534         }
 535
 536         /* expire default router list */
 537         dr = TAILQ_FIRST(&nd_defrouter);
 538         while (dr) {
 539                 if (dr->expire && dr->expire < time_second) {
 540                         struct nd_defrouter *t;
 541                         t = TAILQ_NEXT(dr, dr_entry);
 542                         defrtrlist_del(dr);
 543                         dr = t;
 544                 } else {
 545                         dr = TAILQ_NEXT(dr, dr_entry);
 546                 }
 547         }
 548
 549         /*
 550          * expire interface addresses.
 551          * in the past the loop was inside prefix expiry processing.
 552          * However, from a stricter speci-confrmance standpoint, we should
 553          * rather separate address lifetimes and prefix lifetimes.
 554          */
 555   addrloop:
 556         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 557                 nia6 = ia6->ia_next;
 558                 /* check address lifetime */
 559                 lt6 = &ia6->ia6_lifetime;
 560                 if (IFA6_IS_INVALID(ia6)) {
 561                         int regen = 0;
 562
 563                         /*
 564                          * If the expiring address is temporary, try
 565                          * regenerating a new one.  This would be useful when
 566                          * we suspended a laptop PC, then turned it on after a
 567                          * period that could invalidate all temporary
 568                          * addresses.  Although we may have to restart the
 569                          * loop (see below), it must be after purging the
 570                          * address.  Otherwise, we'd see an infinite loop of
 571                          * regeneration.
 572                          */
 573                         if (ip6_use_tempaddr &&
 574                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
 575                                 if (regen_tmpaddr(ia6) == 0)
 576                                         regen = 1;
 577                         }
 578
 579                         in6_purgeaddr(&ia6->ia_ifa);
 580
 581                         if (regen)
 582                                 goto addrloop; /* XXX: see below */
 583                 }
 584                 if (IFA6_IS_DEPRECATED(ia6)) {
 585                         int oldflags = ia6->ia6_flags;
 586
 587                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 588
 589                         /*
 590                          * If a temporary address has just become deprecated,
 591                          * regenerate a new one if possible.
 592                          */
 593                         if (ip6_use_tempaddr &&
 594                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 595                             (oldflags & IN6_IFF_DEPRECATED) == 0) {
 596
 597                                 if (regen_tmpaddr(ia6) == 0) {
 598                                         /*
 599                                          * A new temporary address is
 600                                          * generated.
 601                                          * XXX: this means the address chain
 602                                          * has changed while we are still in
 603                                          * the loop.  Although the change
 604                                          * would not cause disaster (because
 605                                          * it's not a deletion, but an
 606                                          * addition,) we'd rather restart the
 607                                          * loop just for safety.  Or does this
 608                                          * significantly reduce performance??
 609                                          */
 610                                         goto addrloop;
 611                                 }
 612                         }
 613                 } else {
 614                         /*
 615                          * A new RA might have made a deprecated address
 616                          * preferred.
 617                          */
 618                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 619                 }
 620         }
 621
 622         /* expire prefix list */
 623         pr = nd_prefix.lh_first;
 624         while (pr) {
 625                 /*
 626                  * check prefix lifetime.
 627                  * since pltime is just for autoconf, pltime processing for
 628                  * prefix is not necessary.
 629                  */
 630                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
 631                         struct nd_prefix *t;
 632                         t = pr->ndpr_next;
 633
 634                         /*
 635                          * address expiration and prefix expiration are
 636                          * separate.  NEVER perform in6_purgeaddr here.
 637                          */
 638
 639                         prelist_remove(pr);
 640                         pr = t;
 641                 } else
 642                         pr = pr->ndpr_next;
 643         }
 644         splx(s);
 645 }
 646
 647 static int
 648 regen_tmpaddr(ia6)
 649         struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
 650 {
 651         struct ifaddr *ifa;
 652         struct ifnet *ifp;
 653         struct in6_ifaddr *public_ifa6 = NULL;
 654
 655         ifp = ia6->ia_ifa.ifa_ifp;
 656         for (ifa = ifp->if_addrlist.tqh_first; ifa;
 657              ifa = ifa->ifa_list.tqe_next)
 658         {
 659                 struct in6_ifaddr *it6;
 660
 661                 if (ifa->ifa_addr->sa_family != AF_INET6)
 662                         continue;
 663
 664                 it6 = (struct in6_ifaddr *)ifa;
 665
 666                 /* ignore no autoconf addresses. */
 667                 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
 668                         continue;
 669
 670                 /* ignore autoconf addresses with different prefixes. */
 671                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 672                         continue;
 673
 674                 /*
 675                  * Now we are looking at an autoconf address with the same
 676                  * prefix as ours.  If the address is temporary and is still
 677                  * preferred, do not create another one.  It would be rare, but
 678                  * could happen, for example, when we resume a laptop PC after
 679                  * a long period.
 680                  */
 681                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 682                     !IFA6_IS_DEPRECATED(it6)) {
 683                         public_ifa6 = NULL;
 684                         break;
 685                 }
 686
 687                 /*
 688                  * This is a public autoconf address that has the same prefix
 689                  * as ours.  If it is preferred, keep it.  We can't break the
 690                  * loop here, because there may be a still-preferred temporary
 691                  * address with the prefix.
 692                  */
 693                 if (!IFA6_IS_DEPRECATED(it6))
 694                     public_ifa6 = it6;
 695         }
 696
 697         if (public_ifa6 != NULL) {
 698                 int e;
 699
 700                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 701                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 702                             " tmp addr,errno=%d\n", e);
 703                         return(-1);
 704                 }
 705                 return(0);
 706         }
 707
 708         return(-1);
 709 }
 710
 711 /*
 712  * Nuke neighbor cache/prefix/default router management table, right before
 713  * ifp goes away.
 714  */
 715 void
 716 nd6_purge(ifp)
 717         struct ifnet *ifp;
 718 {
 719         struct llinfo_nd6 *ln, *nln;
 720         struct nd_defrouter *dr, *ndr, drany;
 721         struct nd_prefix *pr, *npr;
 722
 723         /* Nuke default router list entries toward ifp */
 724         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 725                 /*
 726                  * The first entry of the list may be stored in
 727                  * the routing table, so we'll delete it later.
 728                  */
 729                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 730                         ndr = TAILQ_NEXT(dr, dr_entry);
 731                         if (dr->ifp == ifp)
 732                                 defrtrlist_del(dr);
 733                 }
 734                 dr = TAILQ_FIRST(&nd_defrouter);
 735                 if (dr->ifp == ifp)
 736                         defrtrlist_del(dr);
 737         }
 738
 739         /* Nuke prefix list entries toward ifp */
 740         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 741                 npr = pr->ndpr_next;
 742                 if (pr->ndpr_ifp == ifp) {
 743                         /*
 744                          * Previously, pr->ndpr_addr is removed as well,
 745                          * but I strongly believe we don't have to do it.
 746                          * nd6_purge() is only called from in6_ifdetach(),
 747                          * which removes all the associated interface addresses
 748                          * by itself.
 749                          * (jinmei@kame.net 20010129)
 750                          */
 751                         prelist_remove(pr);
 752                 }
 753         }
 754
 755         /* cancel default outgoing interface setting */
 756         if (nd6_defifindex == ifp->if_index)
 757                 nd6_setdefaultiface(0);
 758
 759         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 760                 /* refresh default router list */
 761                 bzero(&drany, sizeof(drany));
 762                 defrouter_delreq(&drany, 0);
 763                 defrouter_select();
 764         }
 765
 766         /*
 767          * Nuke neighbor cache entries for the ifp.
 768          * Note that rt->rt_ifp may not be the same as ifp,
 769          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 770          * nd6_rtrequest(), and ip6_input().
 771          */
 772         ln = llinfo_nd6.ln_next;
 773         while (ln && ln != &llinfo_nd6) {
 774                 struct rtentry *rt;
 775                 struct sockaddr_dl *sdl;
 776
 777                 nln = ln->ln_next;
 778                 rt = ln->ln_rt;
 779                 if (rt && rt->rt_gateway &&
 780                     rt->rt_gateway->sa_family == AF_LINK) {
 781                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 782                         if (sdl->sdl_index == ifp->if_index)
 783                                 nln = nd6_free(rt);
 784                 }
 785                 ln = nln;
 786         }
 787 }
 788
 789 struct rtentry *
 790 nd6_lookup(addr6, create, ifp)
 791         struct in6_addr *addr6;
 792         int create;
 793         struct ifnet *ifp;
 794 {
 795         struct rtentry *rt;
 796         struct sockaddr_in6 sin6;
 797
 798         bzero(&sin6, sizeof(sin6));
 799         sin6.sin6_len = sizeof(struct sockaddr_in6);
 800         sin6.sin6_family = AF_INET6;
 801         sin6.sin6_addr = *addr6;
 802 #ifdef SCOPEDROUTING
 803         sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
 804 #endif
 805         rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
 806         if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
 807                 /*
 808                  * This is the case for the default route.
 809                  * If we want to create a neighbor cache for the address, we
 810                  * should free the route for the destination and allocate an
 811                  * interface route.
 812                  */
 813                 if (create) {
 814                         RTFREE(rt);
 815                         rt = 0;
 816                 }
 817         }
 818         if (!rt) {
 819                 if (create && ifp) {
 820                         int e;
 821
 822                         /*
 823                          * If no route is available and create is set,
 824                          * we allocate a host route for the destination
 825                          * and treat it like an interface route.
 826                          * This hack is necessary for a neighbor which can't
 827                          * be covered by our own prefix.
 828                          */
 829                         struct ifaddr *ifa =
 830                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 831                         if (ifa == NULL)
 832                                 return(NULL);
 833
 834                         /*
 835                          * Create a new route.  RTF_LLINFO is necessary
 836                          * to create a Neighbor Cache entry for the
 837                          * destination in nd6_rtrequest which will be
 838                          * called in rtrequest via ifa->ifa_rtrequest.
 839                          */
 840                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 841                                            ifa->ifa_addr,
 842                                            (struct sockaddr *)&all1_sa,
 843                                            (ifa->ifa_flags |
 844                                             RTF_HOST | RTF_LLINFO) &
 845                                            ~RTF_CLONING,
 846                                            &rt)) != 0)
 847                                 log(LOG_ERR,
 848                                     "nd6_lookup: failed to add route for a "
 849                                     "neighbor(%s), errno=%d\n",
 850                                     ip6_sprintf(addr6), e);
 851                         if (rt == NULL)
 852                                 return(NULL);
 853                         if (rt->rt_llinfo) {
 854                                 struct llinfo_nd6 *ln =
 855                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 856                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 857                         }
 858                 } else
 859                         return(NULL);
 860         }
 861         rt->rt_refcnt--;
 862         /*
 863          * Validation for the entry.
 864          * Note that the check for rt_llinfo is necessary because a cloned
 865          * route from a parent route that has the L flag (e.g. the default
 866          * route to a p2p interface) may have the flag, too, while the
 867          * destination is not actually a neighbor.
 868          * XXX: we can't use rt->rt_ifp to check for the interface, since
 869          *      it might be the loopback interface if the entry is for our
 870          *      own address on a non-loopback interface. Instead, we should
 871          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 872          *      interface.
 873          */
 874         if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
 875             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 876             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 877                 if (create) {
 878                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 879                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 880                         /* xxx more logs... kazu */
 881                 }
 882                 return(NULL);
 883         }
 884         return(rt);
 885 }
 886
 887 /*
 888  * Detect if a given IPv6 address identifies a neighbor on a given link.
 889  * XXX: should take care of the destination of a p2p link?
 890  */
 891 int
 892 nd6_is_addr_neighbor(addr, ifp)
 893         struct sockaddr_in6 *addr;
 894         struct ifnet *ifp;
 895 {
 896         struct ifaddr *ifa;
 897         int i;
 898
 899 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 900 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 901
 902         /*
 903          * A link-local address is always a neighbor.
 904          * XXX: we should use the sin6_scope_id field rather than the embedded
 905          * interface index.
 906          */
 907         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 908             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 909                 return(1);
 910
 911         /*
 912          * If the address matches one of our addresses,
 913          * it should be a neighbor.
 914          */
 915         for (ifa = ifp->if_addrlist.tqh_first;
 916              ifa;
 917              ifa = ifa->ifa_list.tqe_next)
 918         {
 919                 if (ifa->ifa_addr->sa_family != AF_INET6)
 920                         next: continue;
 921
 922                 for (i = 0; i < 4; i++) {
 923                         if ((IFADDR6(ifa).s6_addr32[i] ^
 924                              addr->sin6_addr.s6_addr32[i]) &
 925                             IFMASK6(ifa).s6_addr32[i])
 926                                 goto next;
 927                 }
 928                 return(1);
 929         }
 930
 931         /*
 932          * Even if the address matches none of our addresses, it might be
 933          * in the neighbor cache.
 934          */
 935         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 936                 return(1);
 937
 938         return(0);
 939 #undef IFADDR6
 940 #undef IFMASK6
 941 }
 942
 943 /*
 944  * Free an nd6 llinfo entry.
 945  */
 946 struct llinfo_nd6 *
 947 nd6_free(rt)
 948         struct rtentry *rt;
 949 {
 950         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 951         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 952         struct nd_defrouter *dr;
 953
 954         /*
 955          * we used to have pfctlinput(PRC_HOSTDEAD) here.
 956          * even though it is not harmful, it was not really necessary.
 957          */
 958
 959         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 960                 int s;
 961                 s = splnet();
 962                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 963                                       rt->rt_ifp);
 964
 965                 if (ln->ln_router || dr) {
 966                         /*
 967                          * rt6_flush must be called whether or not the neighbor
 968                          * is in the Default Router List.
 969                          * See a corresponding comment in nd6_na_input().
 970                          */
 971                         rt6_flush(&in6, rt->rt_ifp);
 972                 }
 973
 974                 if (dr) {
 975                         /*
 976                          * Unreachablity of a router might affect the default
 977                          * router selection and on-link detection of advertised
 978                          * prefixes.
 979                          */
 980
 981                         /*
 982                          * Temporarily fake the state to choose a new default
 983                          * router and to perform on-link determination of
 984                          * prefixes correctly.
 985                          * Below the state will be set correctly,
 986                          * or the entry itself will be deleted.
 987                          */
 988                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 989
 990                         /*
 991                          * Since defrouter_select() does not affect the
 992                          * on-link determination and MIP6 needs the check
 993                          * before the default router selection, we perform
 994                          * the check now.
 995                          */
 996                         pfxlist_onlink_check();
 997
 998                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 999                                 /*
1000                                  * It is used as the current default router,
1001                                  * so we have to move it to the end of the
1002                                  * list and choose a new one.
1003                                  * XXX: it is not very efficient if this is
1004                                  *      the only router.
1005                                  */
1006                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1007                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1008
1009                                 defrouter_select();
1010                         }
1011                 }
1012                 splx(s);
1013         }
1014
1015         /*
1016          * Before deleting the entry, remember the next entry as the
1017          * return value.  We need this because pfxlist_onlink_check() above
1018          * might have freed other entries (particularly the old next entry) as
1019          * a side effect (XXX).
1020          */
1021         next = ln->ln_next;
1022
1023         /*
1024          * Detach the route from the routing tree and the list of neighbor
1025          * caches, and disable the route entry not to be used in already
1026          * cached routes.
1027          */
1028         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1029                   rt_mask(rt), 0, (struct rtentry **)0);
1030
1031         return(next);
1032 }
1033
1034 /*
1035  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1036  *
1037  * XXX cost-effective metods?
1038  */
1039 void
1040 nd6_nud_hint(rt, dst6, force)
1041         struct rtentry *rt;
1042         struct in6_addr *dst6;
1043         int force;
1044 {
1045         struct llinfo_nd6 *ln;
1046
1047         /*
1048          * If the caller specified "rt", use that.  Otherwise, resolve the
1049          * routing table by supplied "dst6".
1050          */
1051         if (!rt) {
1052                 if (!dst6)
1053                         return;
1054                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1055                         return;
1056         }
1057
1058         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1059             (rt->rt_flags & RTF_LLINFO) == 0 ||
1060             !rt->rt_llinfo || !rt->rt_gateway ||
1061             rt->rt_gateway->sa_family != AF_LINK) {
1062                 /* This is not a host route. */
1063                 return;
1064         }
1065
1066         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1067         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1068                 return;
1069
1070         /*
1071          * if we get upper-layer reachability confirmation many times,
1072          * it is possible we have false information.
1073          */
1074         if (!force) {
1075                 ln->ln_byhint++;
1076                 if (ln->ln_byhint > nd6_maxnudhint)
1077                         return;
1078         }
1079
1080         ln->ln_state = ND6_LLINFO_REACHABLE;
1081         if (ln->ln_expire)
1082                 ln->ln_expire = time_second +
1083                         nd_ifinfo[rt->rt_ifp->if_index].reachable;
1084 }
1085
1086 void
1087 nd6_rtrequest(req, rt, info)
1088         int     req;
1089         struct rtentry *rt;
1090         struct rt_addrinfo *info; /* xxx unused */
1091 {
1092         struct sockaddr *gate = rt->rt_gateway;
1093         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1094         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1095         struct ifnet *ifp = rt->rt_ifp;
1096         struct ifaddr *ifa;
1097
1098         if ((rt->rt_flags & RTF_GATEWAY))
1099                 return;
1100
1101         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1102                 /*
1103                  * This is probably an interface direct route for a link
1104                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1105                  * We do not need special treatment below for such a route.
1106                  * Moreover, the RTF_LLINFO flag which would be set below
1107                  * would annoy the ndp(8) command.
1108                  */
1109                 return;
1110         }
1111
1112         if (req == RTM_RESOLVE &&
1113             (nd6_need_cache(ifp) == 0 || /* stf case */
1114              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1115                 /*
1116                  * FreeBSD and BSD/OS often make a cloned host route based
1117                  * on a less-specific route (e.g. the default route).
1118                  * If the less specific route does not have a "gateway"
1119                  * (this is the case when the route just goes to a p2p or an
1120                  * stf interface), we'll mistakenly make a neighbor cache for
1121                  * the host route, and will see strange neighbor solicitation
1122                  * for the corresponding destination.  In order to avoid the
1123                  * confusion, we check if the destination of the route is
1124                  * a neighbor in terms of neighbor discovery, and stop the
1125                  * process if not.  Additionally, we remove the LLINFO flag
1126                  * so that ndp(8) will not try to get the neighbor information
1127                  * of the destination.
1128                  */
1129                 rt->rt_flags &= ~RTF_LLINFO;
1130                 return;
1131         }
1132
1133         switch (req) {
1134         case RTM_ADD:
1135                 /*
1136                  * There is no backward compatibility :)
1137                  *
1138                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1139                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1140                  *         rt->rt_flags |= RTF_CLONING;
1141                  */
1142                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1143                         /*
1144                          * Case 1: This route should come from
1145                          * a route to interface.  RTF_LLINFO flag is set
1146                          * for a host route whose destination should be
1147                          * treated as on-link.
1148                          */
1149                         rt_setgate(rt, rt_key(rt),
1150                                    (struct sockaddr *)&null_sdl);
1151                         gate = rt->rt_gateway;
1152                         SDL(gate)->sdl_type = ifp->if_type;
1153                         SDL(gate)->sdl_index = ifp->if_index;
1154                         if (ln)
1155                                 ln->ln_expire = time_second;
1156 #if 1
1157                         if (ln && ln->ln_expire == 0) {
1158                                 /* kludge for desktops */
1159 #if 0
1160                                 printf("nd6_rtequest: time.tv_sec is zero; "
1161                                        "treat it as 1\n");
1162 #endif
1163                                 ln->ln_expire = 1;
1164                         }
1165 #endif
1166                         if ((rt->rt_flags & RTF_CLONING))
1167                                 break;
1168                 }
1169                 /*
1170                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1171                  * We don't do that here since llinfo is not ready yet.
1172                  *
1173                  * There are also couple of other things to be discussed:
1174                  * - unsolicited NA code needs improvement beforehand
1175                  * - RFC2461 says we MAY send multicast unsolicited NA
1176                  *   (7.2.6 paragraph 4), however, it also says that we
1177                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1178                  *   we don't have anything like it right now.
1179                  *   note that the mechanism needs a mutual agreement
1180                  *   between proxies, which means that we need to implement
1181                  *   a new protocol, or a new kludge.
1182                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1183                  *   we need to check ip6forwarding before sending it.
1184                  *   (or should we allow proxy ND configuration only for
1185                  *   routers?  there's no mention about proxy ND from hosts)
1186                  */
1187 #if 0
1188                 /* XXX it does not work */
1189                 if (rt->rt_flags & RTF_ANNOUNCE)
1190                         nd6_na_output(ifp,
1191                               &SIN6(rt_key(rt))->sin6_addr,
1192                               &SIN6(rt_key(rt))->sin6_addr,
1193                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1194                               1, NULL);
1195 #endif
1196                 /* FALLTHROUGH */
1197         case RTM_RESOLVE:
1198                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1199                         /*
1200                          * Address resolution isn't necessary for a point to
1201                          * point link, so we can skip this test for a p2p link.
1202                          */
1203                         if (gate->sa_family != AF_LINK ||
1204                             gate->sa_len < sizeof(null_sdl)) {
1205                                 log(LOG_DEBUG,
1206                                     "nd6_rtrequest: bad gateway value: %s\n",
1207                                     if_name(ifp));
1208                                 break;
1209                         }
1210                         SDL(gate)->sdl_type = ifp->if_type;
1211                         SDL(gate)->sdl_index = ifp->if_index;
1212                 }
1213                 if (ln != NULL)
1214                         break;  /* This happens on a route change */
1215                 /*
1216                  * Case 2: This route may come from cloning, or a manual route
1217                  * add with a LL address.
1218                  */
1219                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1220                 rt->rt_llinfo = (caddr_t)ln;
1221                 if (!ln) {
1222                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1223                         break;
1224                 }
1225                 nd6_inuse++;
1226                 nd6_allocated++;
1227                 Bzero(ln, sizeof(*ln));
1228                 ln->ln_rt = rt;
1229                 /* this is required for "ndp" command. - shin */
1230                 if (req == RTM_ADD) {
1231                         /*
1232                          * gate should have some valid AF_LINK entry,
1233                          * and ln->ln_expire should have some lifetime
1234                          * which is specified by ndp command.
1235                          */
1236                         ln->ln_state = ND6_LLINFO_REACHABLE;
1237                         ln->ln_byhint = 0;
1238                 } else {
1239                         /*
1240                          * When req == RTM_RESOLVE, rt is created and
1241                          * initialized in rtrequest(), so rt_expire is 0.
1242                          */
1243                         ln->ln_state = ND6_LLINFO_NOSTATE;
1244                         ln->ln_expire = time_second;
1245                 }
1246                 rt->rt_flags |= RTF_LLINFO;
1247                 ln->ln_next = llinfo_nd6.ln_next;
1248                 llinfo_nd6.ln_next = ln;
1249                 ln->ln_prev = &llinfo_nd6;
1250                 ln->ln_next->ln_prev = ln;
1251
1252                 /*
1253                  * check if rt_key(rt) is one of my address assigned
1254                  * to the interface.
1255                  */
1256                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1257                                           &SIN6(rt_key(rt))->sin6_addr);
1258                 if (ifa) {
1259                         caddr_t macp = nd6_ifptomac(ifp);
1260                         ln->ln_expire = 0;
1261                         ln->ln_state = ND6_LLINFO_REACHABLE;
1262                         ln->ln_byhint = 0;
1263                         if (macp) {
1264                                 Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1265                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1266                         }
1267                         if (nd6_useloopback) {
1268                                 rt->rt_ifp = &loif[0];  /* XXX */
1269                                 /*
1270                                  * Make sure rt_ifa be equal to the ifaddr
1271                                  * corresponding to the address.
1272                                  * We need this because when we refer
1273                                  * rt_ifa->ia6_flags in ip6_input, we assume
1274                                  * that the rt_ifa points to the address instead
1275                                  * of the loopback address.
1276                                  */
1277                                 if (ifa != rt->rt_ifa) {
1278                                         IFAFREE(rt->rt_ifa);
1279                                         IFAREF(ifa);
1280                                         rt->rt_ifa = ifa;
1281                                 }
1282                         }
1283                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1284                         ln->ln_expire = 0;
1285                         ln->ln_state = ND6_LLINFO_REACHABLE;
1286                         ln->ln_byhint = 0;
1287
1288                         /* join solicited node multicast for proxy ND */
1289                         if (ifp->if_flags & IFF_MULTICAST) {
1290                                 struct in6_addr llsol;
1291                                 int error;
1292
1293                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1294                                 llsol.s6_addr16[0] = htons(0xff02);
1295                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1296                                 llsol.s6_addr32[1] = 0;
1297                                 llsol.s6_addr32[2] = htonl(1);
1298                                 llsol.s6_addr8[12] = 0xff;
1299
1300                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1301                                         nd6log((LOG_ERR, "%s: failed to join "
1302                                             "%s (errno=%d)\n", if_name(ifp),
1303                                             ip6_sprintf(&llsol), error));
1304                                 }
1305                         }
1306                 }
1307                 break;
1308
1309         case RTM_DELETE:
1310                 if (!ln)
1311                         break;
1312                 /* leave from solicited node multicast for proxy ND */
1313                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1314                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1315                         struct in6_addr llsol;
1316                         struct in6_multi *in6m;
1317
1318                         llsol = SIN6(rt_key(rt))->sin6_addr;
1319                         llsol.s6_addr16[0] = htons(0xff02);
1320                         llsol.s6_addr16[1] = htons(ifp->if_index);
1321                         llsol.s6_addr32[1] = 0;
1322                         llsol.s6_addr32[2] = htonl(1);
1323                         llsol.s6_addr8[12] = 0xff;
1324
1325                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1326                         if (in6m)
1327                                 in6_delmulti(in6m);
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_prlist *prl = (struct in6_prlist *)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                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1369                                 /* XXX: need to this hack for KAME stack */
1370                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1371                         } else
1372                                 log(LOG_ERR,
1373                                     "default router list contains a "
1374                                     "non-linklocal address(%s)\n",
1375                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1376
1377                         drl->defrouter[i].flags = dr->flags;
1378                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1379                         drl->defrouter[i].expire = dr->expire;
1380                         drl->defrouter[i].if_index = dr->ifp->if_index;
1381                         i++;
1382                         dr = TAILQ_NEXT(dr, dr_entry);
1383                 }
1384                 splx(s);
1385                 break;
1386         case SIOCGPRLST_IN6:
1387                 /*
1388                  * obsolete API, use sysctl under net.inet6.icmp6
1389                  */
1390                 /*
1391                  * XXX meaning of fields, especialy "raflags", is very
1392                  * differnet between RA prefix list and RR/static prefix list.
1393                  * how about separating ioctls into two?
1394                  */
1395                 bzero(prl, sizeof(*prl));
1396                 s = splnet();
1397                 pr = nd_prefix.lh_first;
1398                 while (pr && i < PRLSTSIZ) {
1399                         struct nd_pfxrouter *pfr;
1400                         int j;
1401
1402                         (void)in6_embedscope(&prl->prefix[i].prefix,
1403                             &pr->ndpr_prefix, NULL, NULL);
1404                         prl->prefix[i].raflags = pr->ndpr_raf;
1405                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1406                         prl->prefix[i].vltime = pr->ndpr_vltime;
1407                         prl->prefix[i].pltime = pr->ndpr_pltime;
1408                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1409                         prl->prefix[i].expire = pr->ndpr_expire;
1410
1411                         pfr = pr->ndpr_advrtrs.lh_first;
1412                         j = 0;
1413                         while (pfr) {
1414                                 if (j < DRLSTSIZ) {
1415 #define RTRADDR prl->prefix[i].advrtr[j]
1416                                         RTRADDR = pfr->router->rtaddr;
1417                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1418                                                 /* XXX: hack for KAME */
1419                                                 RTRADDR.s6_addr16[1] = 0;
1420                                         } else
1421                                                 log(LOG_ERR,
1422                                                     "a router(%s) advertises "
1423                                                     "a prefix with "
1424                                                     "non-link local address\n",
1425                                                     ip6_sprintf(&RTRADDR));
1426 #undef RTRADDR
1427                                 }
1428                                 j++;
1429                                 pfr = pfr->pfr_next;
1430                         }
1431                         prl->prefix[i].advrtrs = j;
1432                         prl->prefix[i].origin = PR_ORIG_RA;
1433
1434                         i++;
1435                         pr = pr->ndpr_next;
1436                 }
1437               {
1438                 struct rr_prefix *rpp;
1439
1440                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1441                      rpp = LIST_NEXT(rpp, rp_entry)) {
1442                         if (i >= PRLSTSIZ)
1443                                 break;
1444                         (void)in6_embedscope(&prl->prefix[i].prefix,
1445                             &pr->ndpr_prefix, NULL, NULL);
1446                         prl->prefix[i].raflags = rpp->rp_raf;
1447                         prl->prefix[i].prefixlen = rpp->rp_plen;
1448                         prl->prefix[i].vltime = rpp->rp_vltime;
1449                         prl->prefix[i].pltime = rpp->rp_pltime;
1450                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1451                         prl->prefix[i].expire = rpp->rp_expire;
1452                         prl->prefix[i].advrtrs = 0;
1453                         prl->prefix[i].origin = rpp->rp_origin;
1454                         i++;
1455                 }
1456               }
1457                 splx(s);
1458
1459                 break;
1460         case OSIOCGIFINFO_IN6:
1461                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1462                         error = EINVAL;
1463                         break;
1464                 }
1465                 ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1466                 ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1467                 ndi->ndi.basereachable =
1468                     nd_ifinfo[ifp->if_index].basereachable;
1469                 ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1470                 ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1471                 ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1472                 ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1473                 ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1474                 ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1475                 break;
1476         case SIOCGIFINFO_IN6:
1477                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1478                         error = EINVAL;
1479                         break;
1480                 }
1481                 ndi->ndi = nd_ifinfo[ifp->if_index];
1482                 break;
1483         case SIOCSIFINFO_FLAGS:
1484                 /* XXX: almost all other fields of ndi->ndi is unused */
1485                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1486                         error = EINVAL;
1487                         break;
1488                 }
1489                 nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1490                 break;
1491         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1492                 /* flush default router list */
1493                 /*
1494                  * xxx sumikawa: should not delete route if default
1495                  * route equals to the top of default router list
1496                  */
1497                 bzero(&any, sizeof(any));
1498                 defrouter_delreq(&any, 0);
1499                 defrouter_select();
1500                 /* xxx sumikawa: flush prefix list */
1501                 break;
1502         case SIOCSPFXFLUSH_IN6:
1503             {
1504                 /* flush all the prefix advertised by routers */
1505                 struct nd_prefix *pr, *next;
1506
1507                 s = splnet();
1508                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1509                         struct in6_ifaddr *ia, *ia_next;
1510
1511                         next = pr->ndpr_next;
1512
1513                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1514                                 continue; /* XXX */
1515
1516                         /* do we really have to remove addresses as well? */
1517                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1518                                 /* ia might be removed.  keep the next ptr. */
1519                                 ia_next = ia->ia_next;
1520
1521                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1522                                         continue;
1523
1524                                 if (ia->ia6_ndpr == pr)
1525                                         in6_purgeaddr(&ia->ia_ifa);
1526                         }
1527                         prelist_remove(pr);
1528                 }
1529                 splx(s);
1530                 break;
1531             }
1532         case SIOCSRTRFLUSH_IN6:
1533             {
1534                 /* flush all the default routers */
1535                 struct nd_defrouter *dr, *next;
1536
1537                 s = splnet();
1538                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1539                         /*
1540                          * The first entry of the list may be stored in
1541                          * the routing table, so we'll delete it later.
1542                          */
1543                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1544                                 next = TAILQ_NEXT(dr, dr_entry);
1545                                 defrtrlist_del(dr);
1546                         }
1547                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1548                 }
1549                 splx(s);
1550                 break;
1551             }
1552         case SIOCGNBRINFO_IN6:
1553             {
1554                 struct llinfo_nd6 *ln;
1555                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1556
1557                 /*
1558                  * XXX: KAME specific hack for scoped addresses
1559                  *      XXXX: for other scopes than link-local?
1560                  */
1561                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1562                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1563                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1564
1565                         if (*idp == 0)
1566                                 *idp = htons(ifp->if_index);
1567                 }
1568
1569                 s = splnet();
1570                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1571                         error = EINVAL;
1572                         splx(s);
1573                         break;
1574                 }
1575                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1576                 nbi->state = ln->ln_state;
1577                 nbi->asked = ln->ln_asked;
1578                 nbi->isrouter = ln->ln_router;
1579                 nbi->expire = ln->ln_expire;
1580                 splx(s);
1581
1582                 break;
1583             }
1584         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1585                 ndif->ifindex = nd6_defifindex;
1586                 break;
1587         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1588                 return(nd6_setdefaultiface(ndif->ifindex));
1589                 break;
1590         }
1591         return(error);
1592 }
1593
1594 /*
1595  * Create neighbor cache entry and cache link-layer address,
1596  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1597  */
1598 struct rtentry *
1599 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1600         struct ifnet *ifp;
1601         struct in6_addr *from;
1602         char *lladdr;
1603         int lladdrlen;
1604         int type;       /* ICMP6 type */
1605         int code;       /* type dependent information */
1606 {
1607         struct rtentry *rt = NULL;
1608         struct llinfo_nd6 *ln = NULL;
1609         int is_newentry;
1610         struct sockaddr_dl *sdl = NULL;
1611         int do_update;
1612         int olladdr;
1613         int llchange;
1614         int newstate = 0;
1615
1616         if (!ifp)
1617                 panic("ifp == NULL in nd6_cache_lladdr");
1618         if (!from)
1619                 panic("from == NULL in nd6_cache_lladdr");
1620
1621         /* nothing must be updated for unspecified address */
1622         if (IN6_IS_ADDR_UNSPECIFIED(from))
1623                 return NULL;
1624
1625         /*
1626          * Validation about ifp->if_addrlen and lladdrlen must be done in
1627          * the caller.
1628          *
1629          * XXX If the link does not have link-layer adderss, what should
1630          * we do? (ifp->if_addrlen == 0)
1631          * Spec says nothing in sections for RA, RS and NA.  There's small
1632          * description on it in NS section (RFC 2461 7.2.3).
1633          */
1634
1635         rt = nd6_lookup(from, 0, ifp);
1636         if (!rt) {
1637 #if 0
1638                 /* nothing must be done if there's no lladdr */
1639                 if (!lladdr || !lladdrlen)
1640                         return NULL;
1641 #endif
1642
1643                 rt = nd6_lookup(from, 1, ifp);
1644                 is_newentry = 1;
1645         } else {
1646                 /* do nothing if static ndp is set */
1647                 if (rt->rt_flags & RTF_STATIC)
1648                         return NULL;
1649                 is_newentry = 0;
1650         }
1651
1652         if (!rt)
1653                 return NULL;
1654         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1655 fail:
1656                 (void)nd6_free(rt);
1657                 return NULL;
1658         }
1659         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1660         if (!ln)
1661                 goto fail;
1662         if (!rt->rt_gateway)
1663                 goto fail;
1664         if (rt->rt_gateway->sa_family != AF_LINK)
1665                 goto fail;
1666         sdl = SDL(rt->rt_gateway);
1667
1668         olladdr = (sdl->sdl_alen) ? 1 : 0;
1669         if (olladdr && lladdr) {
1670                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1671                         llchange = 1;
1672                 else
1673                         llchange = 0;
1674         } else
1675                 llchange = 0;
1676
1677         /*
1678          * newentry olladdr  lladdr  llchange   (*=record)
1679          *      0       n       n       --      (1)
1680          *      0       y       n       --      (2)
1681          *      0       n       y       --      (3) * STALE
1682          *      0       y       y       n       (4) *
1683          *      0       y       y       y       (5) * STALE
1684          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1685          *      1       --      y       --      (7) * STALE
1686          */
1687
1688         if (lladdr) {           /* (3-5) and (7) */
1689                 /*
1690                  * Record source link-layer address
1691                  * XXX is it dependent to ifp->if_type?
1692                  */
1693                 sdl->sdl_alen = ifp->if_addrlen;
1694                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1695         }
1696
1697         if (!is_newentry) {
1698                 if ((!olladdr && lladdr)                /* (3) */
1699                  || (olladdr && lladdr && llchange)) {  /* (5) */
1700                         do_update = 1;
1701                         newstate = ND6_LLINFO_STALE;
1702                 } else                                  /* (1-2,4) */
1703                         do_update = 0;
1704         } else {
1705                 do_update = 1;
1706                 if (!lladdr)                            /* (6) */
1707                         newstate = ND6_LLINFO_NOSTATE;
1708                 else                                    /* (7) */
1709                         newstate = ND6_LLINFO_STALE;
1710         }
1711
1712         if (do_update) {
1713                 /*
1714                  * Update the state of the neighbor cache.
1715                  */
1716                 ln->ln_state = newstate;
1717
1718                 if (ln->ln_state == ND6_LLINFO_STALE) {
1719                         /*
1720                          * XXX: since nd6_output() below will cause
1721                          * state tansition to DELAY and reset the timer,
1722                          * we must set the timer now, although it is actually
1723                          * meaningless.
1724                          */
1725                         ln->ln_expire = time_second + nd6_gctimer;
1726
1727                         if (ln->ln_hold) {
1728                                 /*
1729                                  * we assume ifp is not a p2p here, so just
1730                                  * set the 2nd argument as the 1st one.
1731                                  */
1732                                 nd6_output(ifp, ifp, ln->ln_hold,
1733                                            (struct sockaddr_in6 *)rt_key(rt),
1734                                            rt);
1735                                 ln->ln_hold = NULL;
1736                         }
1737                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1738                         /* probe right away */
1739                         ln->ln_expire = time_second;
1740                 }
1741         }
1742
1743         /*
1744          * ICMP6 type dependent behavior.
1745          *
1746          * NS: clear IsRouter if new entry
1747          * RS: clear IsRouter
1748          * RA: set IsRouter if there's lladdr
1749          * redir: clear IsRouter if new entry
1750          *
1751          * RA case, (1):
1752          * The spec says that we must set IsRouter in the following cases:
1753          * - If lladdr exist, set IsRouter.  This means (1-5).
1754          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1755          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1756          * A quetion arises for (1) case.  (1) case has no lladdr in the
1757          * neighbor cache, this is similar to (6).
1758          * This case is rare but we figured that we MUST NOT set IsRouter.
1759          *
1760          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1761          *                                                      D R
1762          *      0       n       n       --      (1)     c   ?     s
1763          *      0       y       n       --      (2)     c   s     s
1764          *      0       n       y       --      (3)     c   s     s
1765          *      0       y       y       n       (4)     c   s     s
1766          *      0       y       y       y       (5)     c   s     s
1767          *      1       --      n       --      (6) c   c       c s
1768          *      1       --      y       --      (7) c   c   s   c s
1769          *
1770          *                                      (c=clear s=set)
1771          */
1772         switch (type & 0xff) {
1773         case ND_NEIGHBOR_SOLICIT:
1774                 /*
1775                  * New entry must have is_router flag cleared.
1776                  */
1777                 if (is_newentry)        /* (6-7) */
1778                         ln->ln_router = 0;
1779                 break;
1780         case ND_REDIRECT:
1781                 /*
1782                  * If the icmp is a redirect to a better router, always set the
1783                  * is_router flag. Otherwise, if the entry is newly created,
1784                  * clear the flag. [RFC 2461, sec 8.3]
1785                  */
1786                 if (code == ND_REDIRECT_ROUTER)
1787                         ln->ln_router = 1;
1788                 else if (is_newentry) /* (6-7) */
1789                         ln->ln_router = 0;
1790                 break;
1791         case ND_ROUTER_SOLICIT:
1792                 /*
1793                  * is_router flag must always be cleared.
1794                  */
1795                 ln->ln_router = 0;
1796                 break;
1797         case ND_ROUTER_ADVERT:
1798                 /*
1799                  * Mark an entry with lladdr as a router.
1800                  */
1801                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1802                  || (is_newentry && lladdr)) {                  /* (7) */
1803                         ln->ln_router = 1;
1804                 }
1805                 break;
1806         }
1807
1808         /*
1809          * When the link-layer address of a router changes, select the
1810          * best router again.  In particular, when the neighbor entry is newly
1811          * created, it might affect the selection policy.
1812          * Question: can we restrict the first condition to the "is_newentry"
1813          * case?
1814          * XXX: when we hear an RA from a new router with the link-layer
1815          * address option, defrouter_select() is called twice, since
1816          * defrtrlist_update called the function as well.  However, I believe
1817          * we can compromise the overhead, since it only happens the first
1818          * time.
1819          * XXX: although defrouter_select() should not have a bad effect
1820          * for those are not autoconfigured hosts, we explicitly avoid such
1821          * cases for safety.
1822          */
1823         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1824                 defrouter_select();
1825
1826         return rt;
1827 }
1828
1829 static void
1830 nd6_slowtimo(ignored_arg)
1831     void *ignored_arg;
1832 {
1833         int s = splnet();
1834         int i;
1835         struct nd_ifinfo *nd6if;
1836
1837         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1838             nd6_slowtimo, NULL);
1839         for (i = 1; i < if_index + 1; i++) {
1840                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1841                         continue;
1842                 nd6if = &nd_ifinfo[i];
1843                 if (nd6if->basereachable && /* already initialized */
1844                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1845                         /*
1846                          * Since reachable time rarely changes by router
1847                          * advertisements, we SHOULD insure that a new random
1848                          * value gets recomputed at least once every few hours.
1849                          * (RFC 2461, 6.3.4)
1850                          */
1851                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1852                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1853                 }
1854         }
1855         splx(s);
1856 }
1857
1858 #define senderr(e) { error = (e); goto bad;}
1859 int
1860 nd6_output(ifp, origifp, m0, dst, rt0)
1861         struct ifnet *ifp;
1862         struct ifnet *origifp;
1863         struct mbuf *m0;
1864         struct sockaddr_in6 *dst;
1865         struct rtentry *rt0;
1866 {
1867         struct mbuf *m = m0;
1868         struct rtentry *rt = rt0;
1869         struct sockaddr_in6 *gw6 = NULL;
1870         struct llinfo_nd6 *ln = NULL;
1871         int error = 0;
1872
1873         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1874                 goto sendpkt;
1875
1876         if (nd6_need_cache(ifp) == 0)
1877                 goto sendpkt;
1878
1879         /*
1880          * next hop determination.  This routine is derived from ether_outpout.
1881          */
1882         if (rt) {
1883                 if ((rt->rt_flags & RTF_UP) == 0) {
1884                         if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL)) !=
1885                                 NULL)
1886                         {
1887                                 rt->rt_refcnt--;
1888                                 if (rt->rt_ifp != ifp) {
1889                                         /* XXX: loop care? */
1890                                         return nd6_output(ifp, origifp, m0,
1891                                                           dst, rt);
1892                                 }
1893                         } else
1894                                 senderr(EHOSTUNREACH);
1895                 }
1896
1897                 if (rt->rt_flags & RTF_GATEWAY) {
1898                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1899
1900                         /*
1901                          * We skip link-layer address resolution and NUD
1902                          * if the gateway is not a neighbor from ND point
1903                          * of view, regardless of the value of nd_ifinfo.flags.
1904                          * The second condition is a bit tricky; we skip
1905                          * if the gateway is our own address, which is
1906                          * sometimes used to install a route to a p2p link.
1907                          */
1908                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1909                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1910                                 /*
1911                                  * We allow this kind of tricky route only
1912                                  * when the outgoing interface is p2p.
1913                                  * XXX: we may need a more generic rule here.
1914                                  */
1915                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1916                                         senderr(EHOSTUNREACH);
1917
1918                                 goto sendpkt;
1919                         }
1920
1921                         if (rt->rt_gwroute == 0)
1922                                 goto lookup;
1923                         if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1924                                 rtfree(rt); rt = rt0;
1925                         lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1926                                 if ((rt = rt->rt_gwroute) == 0)
1927                                         senderr(EHOSTUNREACH);
1928                         }
1929                 }
1930         }
1931
1932         /*
1933          * Address resolution or Neighbor Unreachability Detection
1934          * for the next hop.
1935          * At this point, the destination of the packet must be a unicast
1936          * or an anycast address(i.e. not a multicast).
1937          */
1938
1939         /* Look up the neighbor cache for the nexthop */
1940         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1941                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1942         else {
1943                 /*
1944                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1945                  * the condition below is not very efficient.  But we believe
1946                  * it is tolerable, because this should be a rare case.
1947                  */
1948                 if (nd6_is_addr_neighbor(dst, ifp) &&
1949                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1950                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1951         }
1952         if (!ln || !rt) {
1953                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1954                     !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1955                         log(LOG_DEBUG,
1956                             "nd6_output: can't allocate llinfo for %s "
1957                             "(ln=%p, rt=%p)\n",
1958                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1959                         senderr(EIO);   /* XXX: good error? */
1960                 }
1961
1962                 goto sendpkt;   /* send anyway */
1963         }
1964
1965         /* We don't have to do link-layer address resolution on a p2p link. */
1966         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1967             ln->ln_state < ND6_LLINFO_REACHABLE) {
1968                 ln->ln_state = ND6_LLINFO_STALE;
1969                 ln->ln_expire = time_second + nd6_gctimer;
1970         }
1971
1972         /*
1973          * The first time we send a packet to a neighbor whose entry is
1974          * STALE, we have to change the state to DELAY and a sets a timer to
1975          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1976          * neighbor unreachability detection on expiration.
1977          * (RFC 2461 7.3.3)
1978          */
1979         if (ln->ln_state == ND6_LLINFO_STALE) {
1980                 ln->ln_asked = 0;
1981                 ln->ln_state = ND6_LLINFO_DELAY;
1982                 ln->ln_expire = time_second + nd6_delay;
1983         }
1984
1985         /*
1986          * If the neighbor cache entry has a state other than INCOMPLETE
1987          * (i.e. its link-layer address is already resolved), just
1988          * send the packet.
1989          */
1990         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1991                 goto sendpkt;
1992
1993         /*
1994          * There is a neighbor cache entry, but no ethernet address
1995          * response yet.  Replace the held mbuf (if any) with this
1996          * latest one.
1997          *
1998          * This code conforms to the rate-limiting rule described in Section
1999          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2000          * an NS below.
2001          */
2002         if (ln->ln_state == ND6_LLINFO_NOSTATE)
2003                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
2004         if (ln->ln_hold)
2005                 m_freem(ln->ln_hold);
2006         ln->ln_hold = m;
2007         if (ln->ln_expire) {
2008                 if (ln->ln_asked < nd6_mmaxtries &&
2009                     ln->ln_expire < time_second) {
2010                         ln->ln_asked++;
2011                         ln->ln_expire = time_second +
2012                                 nd_ifinfo[ifp->if_index].retrans / 1000;
2013                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2014                 }
2015         }
2016         return(0);
2017
2018   sendpkt:
2019
2020         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2021                 return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2022                                          rt));
2023         }
2024         return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2025
2026   bad:
2027         if (m)
2028                 m_freem(m);
2029         return (error);
2030 }
2031 #undef senderr
2032
2033 int
2034 nd6_need_cache(ifp)
2035         struct ifnet *ifp;
2036 {
2037         /*
2038          * XXX: we currently do not make neighbor cache on any interface
2039          * other than ARCnet, Ethernet, FDDI and GIF.
2040          *
2041          * RFC2893 says:
2042          * - unidirectional tunnels needs no ND
2043          */
2044         switch (ifp->if_type) {
2045         case IFT_ARCNET:
2046         case IFT_ETHER:
2047         case IFT_FDDI:
2048         case IFT_IEEE1394:
2049 #ifdef IFT_L2VLAN
2050         case IFT_L2VLAN:
2051 #endif
2052 #ifdef IFT_IEEE80211
2053         case IFT_IEEE80211:
2054 #endif
2055         case IFT_GIF:           /* XXX need more cases? */
2056                 return(1);
2057         default:
2058                 return(0);
2059         }
2060 }
2061
2062 int
2063 nd6_storelladdr(ifp, rt, m, dst, desten)
2064         struct ifnet *ifp;
2065         struct rtentry *rt;
2066         struct mbuf *m;
2067         struct sockaddr *dst;
2068         u_char *desten;
2069 {
2070         int i;
2071         struct sockaddr_dl *sdl;
2072
2073         if (m->m_flags & M_MCAST) {
2074                 switch (ifp->if_type) {
2075                 case IFT_ETHER:
2076                 case IFT_FDDI:
2077 #ifdef IFT_L2VLAN
2078         case IFT_L2VLAN:
2079 #endif
2080 #ifdef IFT_IEEE80211
2081                 case IFT_IEEE80211:
2082 #endif
2083                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2084                                                  desten);
2085                         return(1);
2086                 case IFT_IEEE1394:
2087                         /*
2088                          * netbsd can use if_broadcastaddr, but we don't do so
2089                          * to reduce # of ifdef.
2090                          */
2091                         for (i = 0; i < ifp->if_addrlen; i++)
2092                                 desten[i] = ~0;
2093                         return(1);
2094                 case IFT_ARCNET:
2095                         *desten = 0;
2096                         return(1);
2097                 default:
2098                         m_freem(m);
2099                         return(0);
2100                 }
2101         }
2102
2103         if (rt == NULL) {
2104                 /* this could happen, if we could not allocate memory */
2105                 m_freem(m);
2106                 return(0);
2107         }
2108         if (rt->rt_gateway->sa_family != AF_LINK) {
2109                 printf("nd6_storelladdr: something odd happens\n");
2110                 m_freem(m);
2111                 return(0);
2112         }
2113         sdl = SDL(rt->rt_gateway);
2114         if (sdl->sdl_alen == 0) {
2115                 /* this should be impossible, but we bark here for debugging */
2116                 printf("nd6_storelladdr: sdl_alen == 0\n");
2117                 m_freem(m);
2118                 return(0);
2119         }
2120
2121         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2122         return(1);
2123 }
2124
2125 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2126 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2127 #ifdef SYSCTL_DECL
2128 SYSCTL_DECL(_net_inet6_icmp6);
2129 #endif
2130 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2131         CTLFLAG_RD, nd6_sysctl_drlist, "");
2132 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2133         CTLFLAG_RD, nd6_sysctl_prlist, "");
2134
2135 static int
2136 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2137 {
2138         int error;
2139         char buf[1024];
2140         struct in6_defrouter *d, *de;
2141         struct nd_defrouter *dr;
2142
2143         if (req->newptr)
2144                 return EPERM;
2145         error = 0;
2146
2147         for (dr = TAILQ_FIRST(&nd_defrouter);
2148              dr;
2149              dr = TAILQ_NEXT(dr, dr_entry)) {
2150                 d = (struct in6_defrouter *)buf;
2151                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2152
2153                 if (d + 1 <= de) {
2154                         bzero(d, sizeof(*d));
2155                         d->rtaddr.sin6_family = AF_INET6;
2156                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2157                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2158                             dr->ifp) != 0)
2159                                 log(LOG_ERR,
2160                                     "scope error in "
2161                                     "default router list (%s)\n",
2162                                     ip6_sprintf(&dr->rtaddr));
2163                         d->flags = dr->flags;
2164                         d->rtlifetime = dr->rtlifetime;
2165                         d->expire = dr->expire;
2166                         d->if_index = dr->ifp->if_index;
2167                 } else
2168                         panic("buffer too short");
2169
2170                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2171                 if (error)
2172                         break;
2173         }
2174         return error;
2175 }
2176
2177 static int
2178 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2179 {
2180         int error;
2181         char buf[1024];
2182         struct in6_prefix *p, *pe;
2183         struct nd_prefix *pr;
2184
2185         if (req->newptr)
2186                 return EPERM;
2187         error = 0;
2188
2189         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2190                 u_short advrtrs;
2191                 size_t advance;
2192                 struct sockaddr_in6 *sin6, *s6;
2193                 struct nd_pfxrouter *pfr;
2194
2195                 p = (struct in6_prefix *)buf;
2196                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2197
2198                 if (p + 1 <= pe) {
2199                         bzero(p, sizeof(*p));
2200                         sin6 = (struct sockaddr_in6 *)(p + 1);
2201
2202                         p->prefix = pr->ndpr_prefix;
2203                         if (in6_recoverscope(&p->prefix,
2204                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2205                                 log(LOG_ERR,
2206                                     "scope error in prefix list (%s)\n",
2207                                     ip6_sprintf(&p->prefix.sin6_addr));
2208                         p->raflags = pr->ndpr_raf;
2209                         p->prefixlen = pr->ndpr_plen;
2210                         p->vltime = pr->ndpr_vltime;
2211                         p->pltime = pr->ndpr_pltime;
2212                         p->if_index = pr->ndpr_ifp->if_index;
2213                         p->expire = pr->ndpr_expire;
2214                         p->refcnt = pr->ndpr_refcnt;
2215                         p->flags = pr->ndpr_stateflags;
2216                         p->origin = PR_ORIG_RA;
2217                         advrtrs = 0;
2218                         for (pfr = pr->ndpr_advrtrs.lh_first;
2219                              pfr;
2220                              pfr = pfr->pfr_next) {
2221                                 if ((void *)&sin6[advrtrs + 1] >
2222                                     (void *)pe) {
2223                                         advrtrs++;
2224                                         continue;
2225                                 }
2226                                 s6 = &sin6[advrtrs];
2227                                 bzero(s6, sizeof(*s6));
2228                                 s6->sin6_family = AF_INET6;
2229                                 s6->sin6_len = sizeof(*sin6);
2230                                 if (in6_recoverscope(s6,
2231                                     &pfr->router->rtaddr,
2232                                     pfr->router->ifp) != 0)
2233                                         log(LOG_ERR,
2234                                             "scope error in "
2235                                             "prefix list (%s)\n",
2236                                             ip6_sprintf(&pfr->router->rtaddr));
2237                                 advrtrs++;
2238                         }
2239                         p->advrtrs = advrtrs;
2240                 } else
2241                         panic("buffer too short");
2242
2243                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2244                 error = SYSCTL_OUT(req, buf, advance);
2245                 if (error)
2246                         break;
2247         }
2248         return error;
2249 }