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 1.10 2004/09/17 00:07:27 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 (void *);
 118 static int regen_tmpaddr (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(void)
 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_init(&nd6_slowtimo_ch);
 147         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 148             nd6_slowtimo, NULL);
 149 }
 150
 151 void
 152 nd6_ifattach(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(struct ifnet *ifp)
 206 {
 207         struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
 208         u_long oldmaxmtu = ndi->maxmtu;
 209         u_long oldlinkmtu = ndi->linkmtu;
 210
 211         switch (ifp->if_type) {
 212         case IFT_ARCNET:        /* XXX MTU handling needs more work */
 213                 ndi->maxmtu = MIN(60480, ifp->if_mtu);
 214                 break;
 215         case IFT_ETHER:
 216                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 217                 break;
 218         case IFT_FDDI:
 219                 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
 220                 break;
 221         case IFT_ATM:
 222                 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
 223                 break;
 224         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 225                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 226                 break;
 227 #ifdef IFT_IEEE80211
 228         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 229                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 230                 break;
 231 #endif
 232         default:
 233                 ndi->maxmtu = ifp->if_mtu;
 234                 break;
 235         }
 236
 237         if (oldmaxmtu != ndi->maxmtu) {
 238                 /*
 239                  * If the ND level MTU is not set yet, or if the maxmtu
 240                  * is reset to a smaller value than the ND level MTU,
 241                  * also reset the ND level MTU.
 242                  */
 243                 if (ndi->linkmtu == 0 ||
 244                     ndi->maxmtu < ndi->linkmtu) {
 245                         ndi->linkmtu = ndi->maxmtu;
 246                         /* also adjust in6_maxmtu if necessary. */
 247                         if (oldlinkmtu == 0) {
 248                                 /*
 249                                  * XXX: the case analysis is grotty, but
 250                                  * it is not efficient to call in6_setmaxmtu()
 251                                  * here when we are during the initialization
 252                                  * procedure.
 253                                  */
 254                                 if (in6_maxmtu < ndi->linkmtu)
 255                                         in6_maxmtu = ndi->linkmtu;
 256                         } else
 257                                 in6_setmaxmtu();
 258                 }
 259         }
 260 #undef MIN
 261 }
 262
 263 void
 264 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 265 {
 266         bzero(ndopts, sizeof(*ndopts));
 267         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 268         ndopts->nd_opts_last
 269                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 270
 271         if (icmp6len == 0) {
 272                 ndopts->nd_opts_done = 1;
 273                 ndopts->nd_opts_search = NULL;
 274         }
 275 }
 276
 277 /*
 278  * Take one ND option.
 279  */
 280 struct nd_opt_hdr *
 281 nd6_option(union nd_opts *ndopts)
 282 {
 283         struct nd_opt_hdr *nd_opt;
 284         int olen;
 285
 286         if (!ndopts)
 287                 panic("ndopts == NULL in nd6_option");
 288         if (!ndopts->nd_opts_last)
 289                 panic("uninitialized ndopts in nd6_option");
 290         if (!ndopts->nd_opts_search)
 291                 return NULL;
 292         if (ndopts->nd_opts_done)
 293                 return NULL;
 294
 295         nd_opt = ndopts->nd_opts_search;
 296
 297         /* make sure nd_opt_len is inside the buffer */
 298         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 299                 bzero(ndopts, sizeof(*ndopts));
 300                 return NULL;
 301         }
 302
 303         olen = nd_opt->nd_opt_len << 3;
 304         if (olen == 0) {
 305                 /*
 306                  * Message validation requires that all included
 307                  * options have a length that is greater than zero.
 308                  */
 309                 bzero(ndopts, sizeof(*ndopts));
 310                 return NULL;
 311         }
 312
 313         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 314         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 315                 /* option overruns the end of buffer, invalid */
 316                 bzero(ndopts, sizeof(*ndopts));
 317                 return NULL;
 318         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 319                 /* reached the end of options chain */
 320                 ndopts->nd_opts_done = 1;
 321                 ndopts->nd_opts_search = NULL;
 322         }
 323         return nd_opt;
 324 }
 325
 326 /*
 327  * Parse multiple ND options.
 328  * This function is much easier to use, for ND routines that do not need
 329  * multiple options of the same type.
 330  */
 331 int
 332 nd6_options(union nd_opts *ndopts)
 333 {
 334         struct nd_opt_hdr *nd_opt;
 335         int i = 0;
 336
 337         if (!ndopts)
 338                 panic("ndopts == NULL in nd6_options");
 339         if (!ndopts->nd_opts_last)
 340                 panic("uninitialized ndopts in nd6_options");
 341         if (!ndopts->nd_opts_search)
 342                 return 0;
 343
 344         while (1) {
 345                 nd_opt = nd6_option(ndopts);
 346                 if (!nd_opt && !ndopts->nd_opts_last) {
 347                         /*
 348                          * Message validation requires that all included
 349                          * options have a length that is greater than zero.
 350                          */
 351                         icmp6stat.icp6s_nd_badopt++;
 352                         bzero(ndopts, sizeof(*ndopts));
 353                         return -1;
 354                 }
 355
 356                 if (!nd_opt)
 357                         goto skip1;
 358
 359                 switch (nd_opt->nd_opt_type) {
 360                 case ND_OPT_SOURCE_LINKADDR:
 361                 case ND_OPT_TARGET_LINKADDR:
 362                 case ND_OPT_MTU:
 363                 case ND_OPT_REDIRECTED_HEADER:
 364                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 365                                 nd6log((LOG_INFO,
 366                                     "duplicated ND6 option found (type=%d)\n",
 367                                     nd_opt->nd_opt_type));
 368                                 /* XXX bark? */
 369                         } else {
 370                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 371                                         = nd_opt;
 372                         }
 373                         break;
 374                 case ND_OPT_PREFIX_INFORMATION:
 375                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 376                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 377                                         = nd_opt;
 378                         }
 379                         ndopts->nd_opts_pi_end =
 380                                 (struct nd_opt_prefix_info *)nd_opt;
 381                         break;
 382                 default:
 383                         /*
 384                          * Unknown options must be silently ignored,
 385                          * to accomodate future extension to the protocol.
 386                          */
 387                         nd6log((LOG_DEBUG,
 388                             "nd6_options: unsupported option %d - "
 389                             "option ignored\n", nd_opt->nd_opt_type));
 390                 }
 391
 392 skip1:
 393                 i++;
 394                 if (i > nd6_maxndopt) {
 395                         icmp6stat.icp6s_nd_toomanyopt++;
 396                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 397                         break;
 398                 }
 399
 400                 if (ndopts->nd_opts_done)
 401                         break;
 402         }
 403
 404         return 0;
 405 }
 406
 407 /*
 408  * ND6 timer routine to expire default route list and prefix list
 409  */
 410 void
 411 nd6_timer(void *ignored_arg)
 412 {
 413         int s;
 414         struct llinfo_nd6 *ln;
 415         struct nd_defrouter *dr;
 416         struct nd_prefix *pr;
 417         struct ifnet *ifp;
 418         struct in6_ifaddr *ia6, *nia6;
 419         struct in6_addrlifetime *lt6;
 420
 421         s = splnet();
 422         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 423                       nd6_timer, NULL);
 424
 425         ln = llinfo_nd6.ln_next;
 426         while (ln && ln != &llinfo_nd6) {
 427                 struct rtentry *rt;
 428                 struct sockaddr_in6 *dst;
 429                 struct llinfo_nd6 *next = ln->ln_next;
 430                 /* XXX: used for the DELAY case only: */
 431                 struct nd_ifinfo *ndi = NULL;
 432
 433                 if ((rt = ln->ln_rt) == NULL) {
 434                         ln = next;
 435                         continue;
 436                 }
 437                 if ((ifp = rt->rt_ifp) == NULL) {
 438                         ln = next;
 439                         continue;
 440                 }
 441                 ndi = &nd_ifinfo[ifp->if_index];
 442                 dst = (struct sockaddr_in6 *)rt_key(rt);
 443
 444                 if (ln->ln_expire > time_second) {
 445                         ln = next;
 446                         continue;
 447                 }
 448
 449                 /* sanity check */
 450                 if (!rt)
 451                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 452                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 453                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 454                               rt->rt_llinfo, ln);
 455                 if (!dst)
 456                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 457
 458                 switch (ln->ln_state) {
 459                 case ND6_LLINFO_INCOMPLETE:
 460                         if (ln->ln_asked < nd6_mmaxtries) {
 461                                 ln->ln_asked++;
 462                                 ln->ln_expire = time_second +
 463                                         nd_ifinfo[ifp->if_index].retrans / 1000;
 464                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 465                                         ln, 0);
 466                         } else {
 467                                 struct mbuf *m = ln->ln_hold;
 468                                 if (m) {
 469                                         if (rt->rt_ifp) {
 470                                                 /*
 471                                                  * Fake rcvif to make ICMP error
 472                                                  * more helpful in diagnosing
 473                                                  * for the receiver.
 474                                                  * XXX: should we consider
 475                                                  * older rcvif?
 476                                                  */
 477                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 478                                         }
 479                                         icmp6_error(m, ICMP6_DST_UNREACH,
 480                                                     ICMP6_DST_UNREACH_ADDR, 0);
 481                                         ln->ln_hold = NULL;
 482                                 }
 483                                 next = nd6_free(rt);
 484                         }
 485                         break;
 486                 case ND6_LLINFO_REACHABLE:
 487                         if (ln->ln_expire) {
 488                                 ln->ln_state = ND6_LLINFO_STALE;
 489                                 ln->ln_expire = time_second + nd6_gctimer;
 490                         }
 491                         break;
 492
 493                 case ND6_LLINFO_STALE:
 494                         /* Garbage Collection(RFC 2461 5.3) */
 495                         if (ln->ln_expire)
 496                                 next = nd6_free(rt);
 497                         break;
 498
 499                 case ND6_LLINFO_DELAY:
 500                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
 501                                 /* We need NUD */
 502                                 ln->ln_asked = 1;
 503                                 ln->ln_state = ND6_LLINFO_PROBE;
 504                                 ln->ln_expire = time_second +
 505                                         ndi->retrans / 1000;
 506                                 nd6_ns_output(ifp, &dst->sin6_addr,
 507                                               &dst->sin6_addr,
 508                                               ln, 0);
 509                         } else {
 510                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 511                                 ln->ln_expire = time_second + nd6_gctimer;
 512                         }
 513                         break;
 514                 case ND6_LLINFO_PROBE:
 515                         if (ln->ln_asked < nd6_umaxtries) {
 516                                 ln->ln_asked++;
 517                                 ln->ln_expire = time_second +
 518                                         nd_ifinfo[ifp->if_index].retrans / 1000;
 519                                 nd6_ns_output(ifp, &dst->sin6_addr,
 520                                                &dst->sin6_addr, ln, 0);
 521                         } else {
 522                                 next = nd6_free(rt);
 523                         }
 524                         break;
 525                 }
 526                 ln = next;
 527         }
 528
 529         /* expire default router list */
 530         dr = TAILQ_FIRST(&nd_defrouter);
 531         while (dr) {
 532                 if (dr->expire && dr->expire < time_second) {
 533                         struct nd_defrouter *t;
 534                         t = TAILQ_NEXT(dr, dr_entry);
 535                         defrtrlist_del(dr);
 536                         dr = t;
 537                 } else {
 538                         dr = TAILQ_NEXT(dr, dr_entry);
 539                 }
 540         }
 541
 542         /*
 543          * expire interface addresses.
 544          * in the past the loop was inside prefix expiry processing.
 545          * However, from a stricter speci-confrmance standpoint, we should
 546          * rather separate address lifetimes and prefix lifetimes.
 547          */
 548   addrloop:
 549         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 550                 nia6 = ia6->ia_next;
 551                 /* check address lifetime */
 552                 lt6 = &ia6->ia6_lifetime;
 553                 if (IFA6_IS_INVALID(ia6)) {
 554                         int regen = 0;
 555
 556                         /*
 557                          * If the expiring address is temporary, try
 558                          * regenerating a new one.  This would be useful when
 559                          * we suspended a laptop PC, then turned it on after a
 560                          * period that could invalidate all temporary
 561                          * addresses.  Although we may have to restart the
 562                          * loop (see below), it must be after purging the
 563                          * address.  Otherwise, we'd see an infinite loop of
 564                          * regeneration.
 565                          */
 566                         if (ip6_use_tempaddr &&
 567                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
 568                                 if (regen_tmpaddr(ia6) == 0)
 569                                         regen = 1;
 570                         }
 571
 572                         in6_purgeaddr(&ia6->ia_ifa);
 573
 574                         if (regen)
 575                                 goto addrloop; /* XXX: see below */
 576                 }
 577                 if (IFA6_IS_DEPRECATED(ia6)) {
 578                         int oldflags = ia6->ia6_flags;
 579
 580                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 581
 582                         /*
 583                          * If a temporary address has just become deprecated,
 584                          * regenerate a new one if possible.
 585                          */
 586                         if (ip6_use_tempaddr &&
 587                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 588                             (oldflags & IN6_IFF_DEPRECATED) == 0) {
 589
 590                                 if (regen_tmpaddr(ia6) == 0) {
 591                                         /*
 592                                          * A new temporary address is
 593                                          * generated.
 594                                          * XXX: this means the address chain
 595                                          * has changed while we are still in
 596                                          * the loop.  Although the change
 597                                          * would not cause disaster (because
 598                                          * it's not a deletion, but an
 599                                          * addition,) we'd rather restart the
 600                                          * loop just for safety.  Or does this
 601                                          * significantly reduce performance??
 602                                          */
 603                                         goto addrloop;
 604                                 }
 605                         }
 606                 } else {
 607                         /*
 608                          * A new RA might have made a deprecated address
 609                          * preferred.
 610                          */
 611                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 612                 }
 613         }
 614
 615         /* expire prefix list */
 616         pr = nd_prefix.lh_first;
 617         while (pr) {
 618                 /*
 619                  * check prefix lifetime.
 620                  * since pltime is just for autoconf, pltime processing for
 621                  * prefix is not necessary.
 622                  */
 623                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
 624                         struct nd_prefix *t;
 625                         t = pr->ndpr_next;
 626
 627                         /*
 628                          * address expiration and prefix expiration are
 629                          * separate.  NEVER perform in6_purgeaddr here.
 630                          */
 631
 632                         prelist_remove(pr);
 633                         pr = t;
 634                 } else
 635                         pr = pr->ndpr_next;
 636         }
 637         splx(s);
 638 }
 639
 640 static int
 641 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 642                                          address */
 643 {
 644         struct ifaddr *ifa;
 645         struct ifnet *ifp;
 646         struct in6_ifaddr *public_ifa6 = NULL;
 647
 648         ifp = ia6->ia_ifa.ifa_ifp;
 649         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_list) {
 650                 struct in6_ifaddr *it6;
 651
 652                 if (ifa->ifa_addr->sa_family != AF_INET6)
 653                         continue;
 654
 655                 it6 = (struct in6_ifaddr *)ifa;
 656
 657                 /* ignore no autoconf addresses. */
 658                 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
 659                         continue;
 660
 661                 /* ignore autoconf addresses with different prefixes. */
 662                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 663                         continue;
 664
 665                 /*
 666                  * Now we are looking at an autoconf address with the same
 667                  * prefix as ours.  If the address is temporary and is still
 668                  * preferred, do not create another one.  It would be rare, but
 669                  * could happen, for example, when we resume a laptop PC after
 670                  * a long period.
 671                  */
 672                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 673                     !IFA6_IS_DEPRECATED(it6)) {
 674                         public_ifa6 = NULL;
 675                         break;
 676                 }
 677
 678                 /*
 679                  * This is a public autoconf address that has the same prefix
 680                  * as ours.  If it is preferred, keep it.  We can't break the
 681                  * loop here, because there may be a still-preferred temporary
 682                  * address with the prefix.
 683                  */
 684                 if (!IFA6_IS_DEPRECATED(it6))
 685                     public_ifa6 = it6;
 686         }
 687
 688         if (public_ifa6 != NULL) {
 689                 int e;
 690
 691                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 692                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 693                             " tmp addr,errno=%d\n", e);
 694                         return(-1);
 695                 }
 696                 return(0);
 697         }
 698
 699         return(-1);
 700 }
 701
 702 /*
 703  * Nuke neighbor cache/prefix/default router management table, right before
 704  * ifp goes away.
 705  */
 706 void
 707 nd6_purge(struct ifnet *ifp)
 708 {
 709         struct llinfo_nd6 *ln, *nln;
 710         struct nd_defrouter *dr, *ndr, drany;
 711         struct nd_prefix *pr, *npr;
 712
 713         /* Nuke default router list entries toward ifp */
 714         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 715                 /*
 716                  * The first entry of the list may be stored in
 717                  * the routing table, so we'll delete it later.
 718                  */
 719                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 720                         ndr = TAILQ_NEXT(dr, dr_entry);
 721                         if (dr->ifp == ifp)
 722                                 defrtrlist_del(dr);
 723                 }
 724                 dr = TAILQ_FIRST(&nd_defrouter);
 725                 if (dr->ifp == ifp)
 726                         defrtrlist_del(dr);
 727         }
 728
 729         /* Nuke prefix list entries toward ifp */
 730         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 731                 npr = pr->ndpr_next;
 732                 if (pr->ndpr_ifp == ifp) {
 733                         /*
 734                          * Previously, pr->ndpr_addr is removed as well,
 735                          * but I strongly believe we don't have to do it.
 736                          * nd6_purge() is only called from in6_ifdetach(),
 737                          * which removes all the associated interface addresses
 738                          * by itself.
 739                          * (jinmei@kame.net 20010129)
 740                          */
 741                         prelist_remove(pr);
 742                 }
 743         }
 744
 745         /* cancel default outgoing interface setting */
 746         if (nd6_defifindex == ifp->if_index)
 747                 nd6_setdefaultiface(0);
 748
 749         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 750                 /* refresh default router list */
 751                 bzero(&drany, sizeof(drany));
 752                 defrouter_delreq(&drany, 0);
 753                 defrouter_select();
 754         }
 755
 756         /*
 757          * Nuke neighbor cache entries for the ifp.
 758          * Note that rt->rt_ifp may not be the same as ifp,
 759          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 760          * nd6_rtrequest(), and ip6_input().
 761          */
 762         ln = llinfo_nd6.ln_next;
 763         while (ln && ln != &llinfo_nd6) {
 764                 struct rtentry *rt;
 765                 struct sockaddr_dl *sdl;
 766
 767                 nln = ln->ln_next;
 768                 rt = ln->ln_rt;
 769                 if (rt && rt->rt_gateway &&
 770                     rt->rt_gateway->sa_family == AF_LINK) {
 771                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 772                         if (sdl->sdl_index == ifp->if_index)
 773                                 nln = nd6_free(rt);
 774                 }
 775                 ln = nln;
 776         }
 777 }
 778
 779 struct rtentry *
 780 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 781 {
 782         struct rtentry *rt;
 783         struct sockaddr_in6 sin6;
 784
 785         bzero(&sin6, sizeof(sin6));
 786         sin6.sin6_len = sizeof(struct sockaddr_in6);
 787         sin6.sin6_family = AF_INET6;
 788         sin6.sin6_addr = *addr6;
 789 #ifdef SCOPEDROUTING
 790         sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
 791 #endif
 792         rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
 793         if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
 794                 /*
 795                  * This is the case for the default route.
 796                  * If we want to create a neighbor cache for the address, we
 797                  * should free the route for the destination and allocate an
 798                  * interface route.
 799                  */
 800                 if (create) {
 801                         RTFREE(rt);
 802                         rt = 0;
 803                 }
 804         }
 805         if (!rt) {
 806                 if (create && ifp) {
 807                         int e;
 808
 809                         /*
 810                          * If no route is available and create is set,
 811                          * we allocate a host route for the destination
 812                          * and treat it like an interface route.
 813                          * This hack is necessary for a neighbor which can't
 814                          * be covered by our own prefix.
 815                          */
 816                         struct ifaddr *ifa =
 817                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 818                         if (ifa == NULL)
 819                                 return(NULL);
 820
 821                         /*
 822                          * Create a new route.  RTF_LLINFO is necessary
 823                          * to create a Neighbor Cache entry for the
 824                          * destination in nd6_rtrequest which will be
 825                          * called in rtrequest via ifa->ifa_rtrequest.
 826                          */
 827                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 828                                            ifa->ifa_addr,
 829                                            (struct sockaddr *)&all1_sa,
 830                                            (ifa->ifa_flags |
 831                                             RTF_HOST | RTF_LLINFO) &
 832                                            ~RTF_CLONING,
 833                                            &rt)) != 0)
 834                                 log(LOG_ERR,
 835                                     "nd6_lookup: failed to add route for a "
 836                                     "neighbor(%s), errno=%d\n",
 837                                     ip6_sprintf(addr6), e);
 838                         if (rt == NULL)
 839                                 return(NULL);
 840                         if (rt->rt_llinfo) {
 841                                 struct llinfo_nd6 *ln =
 842                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 843                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 844                         }
 845                 } else
 846                         return(NULL);
 847         }
 848         rt->rt_refcnt--;
 849         /*
 850          * Validation for the entry.
 851          * Note that the check for rt_llinfo is necessary because a cloned
 852          * route from a parent route that has the L flag (e.g. the default
 853          * route to a p2p interface) may have the flag, too, while the
 854          * destination is not actually a neighbor.
 855          * XXX: we can't use rt->rt_ifp to check for the interface, since
 856          *      it might be the loopback interface if the entry is for our
 857          *      own address on a non-loopback interface. Instead, we should
 858          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 859          *      interface.
 860          */
 861         if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
 862             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 863             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 864                 if (create) {
 865                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 866                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 867                         /* xxx more logs... kazu */
 868                 }
 869                 return(NULL);
 870         }
 871         return(rt);
 872 }
 873
 874 /*
 875  * Detect if a given IPv6 address identifies a neighbor on a given link.
 876  * XXX: should take care of the destination of a p2p link?
 877  */
 878 int
 879 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 880 {
 881         struct ifaddr *ifa;
 882         int i;
 883
 884 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 885 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 886
 887         /*
 888          * A link-local address is always a neighbor.
 889          * XXX: we should use the sin6_scope_id field rather than the embedded
 890          * interface index.
 891          */
 892         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 893             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 894                 return(1);
 895
 896         /*
 897          * If the address matches one of our addresses,
 898          * it should be a neighbor.
 899          */
 900         TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
 901                 if (ifa->ifa_addr->sa_family != AF_INET6)
 902                         next: continue;
 903
 904                 for (i = 0; i < 4; i++) {
 905                         if ((IFADDR6(ifa).s6_addr32[i] ^
 906                              addr->sin6_addr.s6_addr32[i]) &
 907                             IFMASK6(ifa).s6_addr32[i])
 908                                 goto next;
 909                 }
 910                 return(1);
 911         }
 912
 913         /*
 914          * Even if the address matches none of our addresses, it might be
 915          * in the neighbor cache.
 916          */
 917         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 918                 return(1);
 919
 920         return(0);
 921 #undef IFADDR6
 922 #undef IFMASK6
 923 }
 924
 925 /*
 926  * Free an nd6 llinfo entry.
 927  */
 928 struct llinfo_nd6 *
 929 nd6_free(struct rtentry *rt)
 930 {
 931         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 932         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 933         struct nd_defrouter *dr;
 934
 935         /*
 936          * we used to have pfctlinput(PRC_HOSTDEAD) here.
 937          * even though it is not harmful, it was not really necessary.
 938          */
 939
 940         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 941                 int s;
 942                 s = splnet();
 943                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 944                                       rt->rt_ifp);
 945
 946                 if (ln->ln_router || dr) {
 947                         /*
 948                          * rt6_flush must be called whether or not the neighbor
 949                          * is in the Default Router List.
 950                          * See a corresponding comment in nd6_na_input().
 951                          */
 952                         rt6_flush(&in6, rt->rt_ifp);
 953                 }
 954
 955                 if (dr) {
 956                         /*
 957                          * Unreachablity of a router might affect the default
 958                          * router selection and on-link detection of advertised
 959                          * prefixes.
 960                          */
 961
 962                         /*
 963                          * Temporarily fake the state to choose a new default
 964                          * router and to perform on-link determination of
 965                          * prefixes correctly.
 966                          * Below the state will be set correctly,
 967                          * or the entry itself will be deleted.
 968                          */
 969                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 970
 971                         /*
 972                          * Since defrouter_select() does not affect the
 973                          * on-link determination and MIP6 needs the check
 974                          * before the default router selection, we perform
 975                          * the check now.
 976                          */
 977                         pfxlist_onlink_check();
 978
 979                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 980                                 /*
 981                                  * It is used as the current default router,
 982                                  * so we have to move it to the end of the
 983                                  * list and choose a new one.
 984                                  * XXX: it is not very efficient if this is
 985                                  *      the only router.
 986                                  */
 987                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
 988                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
 989
 990                                 defrouter_select();
 991                         }
 992                 }
 993                 splx(s);
 994         }
 995
 996         /*
 997          * Before deleting the entry, remember the next entry as the
 998          * return value.  We need this because pfxlist_onlink_check() above
 999          * might have freed other entries (particularly the old next entry) as
1000          * a side effect (XXX).
1001          */
1002         next = ln->ln_next;
1003
1004         /*
1005          * Detach the route from the routing tree and the list of neighbor
1006          * caches, and disable the route entry not to be used in already
1007          * cached routes.
1008          */
1009         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1010                   rt_mask(rt), 0, (struct rtentry **)0);
1011
1012         return(next);
1013 }
1014
1015 /*
1016  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1017  *
1018  * XXX cost-effective metods?
1019  */
1020 void
1021 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1022 {
1023         struct llinfo_nd6 *ln;
1024
1025         /*
1026          * If the caller specified "rt", use that.  Otherwise, resolve the
1027          * routing table by supplied "dst6".
1028          */
1029         if (!rt) {
1030                 if (!dst6)
1031                         return;
1032                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1033                         return;
1034         }
1035
1036         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1037             (rt->rt_flags & RTF_LLINFO) == 0 ||
1038             !rt->rt_llinfo || !rt->rt_gateway ||
1039             rt->rt_gateway->sa_family != AF_LINK) {
1040                 /* This is not a host route. */
1041                 return;
1042         }
1043
1044         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1045         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1046                 return;
1047
1048         /*
1049          * if we get upper-layer reachability confirmation many times,
1050          * it is possible we have false information.
1051          */
1052         if (!force) {
1053                 ln->ln_byhint++;
1054                 if (ln->ln_byhint > nd6_maxnudhint)
1055                         return;
1056         }
1057
1058         ln->ln_state = ND6_LLINFO_REACHABLE;
1059         if (ln->ln_expire)
1060                 ln->ln_expire = time_second +
1061                         nd_ifinfo[rt->rt_ifp->if_index].reachable;
1062 }
1063
1064 void
1065 nd6_rtrequest(int req, struct rtentry *rt,
1066               struct rt_addrinfo *info) /* xxx unused */
1067 {
1068         struct sockaddr *gate = rt->rt_gateway;
1069         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1070         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1071         struct ifnet *ifp = rt->rt_ifp;
1072         struct ifaddr *ifa;
1073
1074         if ((rt->rt_flags & RTF_GATEWAY))
1075                 return;
1076
1077         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1078                 /*
1079                  * This is probably an interface direct route for a link
1080                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1081                  * We do not need special treatment below for such a route.
1082                  * Moreover, the RTF_LLINFO flag which would be set below
1083                  * would annoy the ndp(8) command.
1084                  */
1085                 return;
1086         }
1087
1088         if (req == RTM_RESOLVE &&
1089             (nd6_need_cache(ifp) == 0 || /* stf case */
1090              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1091                 /*
1092                  * FreeBSD and BSD/OS often make a cloned host route based
1093                  * on a less-specific route (e.g. the default route).
1094                  * If the less specific route does not have a "gateway"
1095                  * (this is the case when the route just goes to a p2p or an
1096                  * stf interface), we'll mistakenly make a neighbor cache for
1097                  * the host route, and will see strange neighbor solicitation
1098                  * for the corresponding destination.  In order to avoid the
1099                  * confusion, we check if the destination of the route is
1100                  * a neighbor in terms of neighbor discovery, and stop the
1101                  * process if not.  Additionally, we remove the LLINFO flag
1102                  * so that ndp(8) will not try to get the neighbor information
1103                  * of the destination.
1104                  */
1105                 rt->rt_flags &= ~RTF_LLINFO;
1106                 return;
1107         }
1108
1109         switch (req) {
1110         case RTM_ADD:
1111                 /*
1112                  * There is no backward compatibility :)
1113                  *
1114                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1115                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1116                  *         rt->rt_flags |= RTF_CLONING;
1117                  */
1118                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1119                         /*
1120                          * Case 1: This route should come from
1121                          * a route to interface.  RTF_LLINFO flag is set
1122                          * for a host route whose destination should be
1123                          * treated as on-link.
1124                          */
1125                         rt_setgate(rt, rt_key(rt),
1126                                    (struct sockaddr *)&null_sdl);
1127                         gate = rt->rt_gateway;
1128                         SDL(gate)->sdl_type = ifp->if_type;
1129                         SDL(gate)->sdl_index = ifp->if_index;
1130                         if (ln)
1131                                 ln->ln_expire = time_second;
1132 #if 1
1133                         if (ln && ln->ln_expire == 0) {
1134                                 /* kludge for desktops */
1135 #if 0
1136                                 printf("nd6_rtequest: time.tv_sec is zero; "
1137                                        "treat it as 1\n");
1138 #endif
1139                                 ln->ln_expire = 1;
1140                         }
1141 #endif
1142                         if ((rt->rt_flags & RTF_CLONING))
1143                                 break;
1144                 }
1145                 /*
1146                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1147                  * We don't do that here since llinfo is not ready yet.
1148                  *
1149                  * There are also couple of other things to be discussed:
1150                  * - unsolicited NA code needs improvement beforehand
1151                  * - RFC2461 says we MAY send multicast unsolicited NA
1152                  *   (7.2.6 paragraph 4), however, it also says that we
1153                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1154                  *   we don't have anything like it right now.
1155                  *   note that the mechanism needs a mutual agreement
1156                  *   between proxies, which means that we need to implement
1157                  *   a new protocol, or a new kludge.
1158                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1159                  *   we need to check ip6forwarding before sending it.
1160                  *   (or should we allow proxy ND configuration only for
1161                  *   routers?  there's no mention about proxy ND from hosts)
1162                  */
1163 #if 0
1164                 /* XXX it does not work */
1165                 if (rt->rt_flags & RTF_ANNOUNCE)
1166                         nd6_na_output(ifp,
1167                               &SIN6(rt_key(rt))->sin6_addr,
1168                               &SIN6(rt_key(rt))->sin6_addr,
1169                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1170                               1, NULL);
1171 #endif
1172                 /* FALLTHROUGH */
1173         case RTM_RESOLVE:
1174                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1175                         /*
1176                          * Address resolution isn't necessary for a point to
1177                          * point link, so we can skip this test for a p2p link.
1178                          */
1179                         if (gate->sa_family != AF_LINK ||
1180                             gate->sa_len < sizeof(null_sdl)) {
1181                                 log(LOG_DEBUG,
1182                                     "nd6_rtrequest: bad gateway value: %s\n",
1183                                     if_name(ifp));
1184                                 break;
1185                         }
1186                         SDL(gate)->sdl_type = ifp->if_type;
1187                         SDL(gate)->sdl_index = ifp->if_index;
1188                 }
1189                 if (ln != NULL)
1190                         break;  /* This happens on a route change */
1191                 /*
1192                  * Case 2: This route may come from cloning, or a manual route
1193                  * add with a LL address.
1194                  */
1195                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1196                 rt->rt_llinfo = (caddr_t)ln;
1197                 if (!ln) {
1198                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1199                         break;
1200                 }
1201                 nd6_inuse++;
1202                 nd6_allocated++;
1203                 Bzero(ln, sizeof(*ln));
1204                 ln->ln_rt = rt;
1205                 /* this is required for "ndp" command. - shin */
1206                 if (req == RTM_ADD) {
1207                         /*
1208                          * gate should have some valid AF_LINK entry,
1209                          * and ln->ln_expire should have some lifetime
1210                          * which is specified by ndp command.
1211                          */
1212                         ln->ln_state = ND6_LLINFO_REACHABLE;
1213                         ln->ln_byhint = 0;
1214                 } else {
1215                         /*
1216                          * When req == RTM_RESOLVE, rt is created and
1217                          * initialized in rtrequest(), so rt_expire is 0.
1218                          */
1219                         ln->ln_state = ND6_LLINFO_NOSTATE;
1220                         ln->ln_expire = time_second;
1221                 }
1222                 rt->rt_flags |= RTF_LLINFO;
1223                 ln->ln_next = llinfo_nd6.ln_next;
1224                 llinfo_nd6.ln_next = ln;
1225                 ln->ln_prev = &llinfo_nd6;
1226                 ln->ln_next->ln_prev = ln;
1227
1228                 /*
1229                  * check if rt_key(rt) is one of my address assigned
1230                  * to the interface.
1231                  */
1232                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1233                                           &SIN6(rt_key(rt))->sin6_addr);
1234                 if (ifa) {
1235                         caddr_t macp = nd6_ifptomac(ifp);
1236                         ln->ln_expire = 0;
1237                         ln->ln_state = ND6_LLINFO_REACHABLE;
1238                         ln->ln_byhint = 0;
1239                         if (macp) {
1240                                 Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1241                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1242                         }
1243                         if (nd6_useloopback) {
1244                                 rt->rt_ifp = &loif[0];  /* XXX */
1245                                 /*
1246                                  * Make sure rt_ifa be equal to the ifaddr
1247                                  * corresponding to the address.
1248                                  * We need this because when we refer
1249                                  * rt_ifa->ia6_flags in ip6_input, we assume
1250                                  * that the rt_ifa points to the address instead
1251                                  * of the loopback address.
1252                                  */
1253                                 if (ifa != rt->rt_ifa) {
1254                                         IFAFREE(rt->rt_ifa);
1255                                         IFAREF(ifa);
1256                                         rt->rt_ifa = ifa;
1257                                 }
1258                         }
1259                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1260                         ln->ln_expire = 0;
1261                         ln->ln_state = ND6_LLINFO_REACHABLE;
1262                         ln->ln_byhint = 0;
1263
1264                         /* join solicited node multicast for proxy ND */
1265                         if (ifp->if_flags & IFF_MULTICAST) {
1266                                 struct in6_addr llsol;
1267                                 int error;
1268
1269                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1270                                 llsol.s6_addr16[0] = htons(0xff02);
1271                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1272                                 llsol.s6_addr32[1] = 0;
1273                                 llsol.s6_addr32[2] = htonl(1);
1274                                 llsol.s6_addr8[12] = 0xff;
1275
1276                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1277                                         nd6log((LOG_ERR, "%s: failed to join "
1278                                             "%s (errno=%d)\n", if_name(ifp),
1279                                             ip6_sprintf(&llsol), error));
1280                                 }
1281                         }
1282                 }
1283                 break;
1284
1285         case RTM_DELETE:
1286                 if (!ln)
1287                         break;
1288                 /* leave from solicited node multicast for proxy ND */
1289                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1290                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1291                         struct in6_addr llsol;
1292                         struct in6_multi *in6m;
1293
1294                         llsol = SIN6(rt_key(rt))->sin6_addr;
1295                         llsol.s6_addr16[0] = htons(0xff02);
1296                         llsol.s6_addr16[1] = htons(ifp->if_index);
1297                         llsol.s6_addr32[1] = 0;
1298                         llsol.s6_addr32[2] = htonl(1);
1299                         llsol.s6_addr8[12] = 0xff;
1300
1301                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1302                         if (in6m)
1303                                 in6_delmulti(in6m);
1304                 }
1305                 nd6_inuse--;
1306                 ln->ln_next->ln_prev = ln->ln_prev;
1307                 ln->ln_prev->ln_next = ln->ln_next;
1308                 ln->ln_prev = NULL;
1309                 rt->rt_llinfo = 0;
1310                 rt->rt_flags &= ~RTF_LLINFO;
1311                 if (ln->ln_hold)
1312                         m_freem(ln->ln_hold);
1313                 Free((caddr_t)ln);
1314         }
1315 }
1316
1317 int
1318 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1319 {
1320         struct in6_drlist *drl = (struct in6_drlist *)data;
1321         struct in6_prlist *prl = (struct in6_prlist *)data;
1322         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1323         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1324         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1325         struct nd_defrouter *dr, any;
1326         struct nd_prefix *pr;
1327         struct rtentry *rt;
1328         int i = 0, error = 0;
1329         int s;
1330
1331         switch (cmd) {
1332         case SIOCGDRLST_IN6:
1333                 /*
1334                  * obsolete API, use sysctl under net.inet6.icmp6
1335                  */
1336                 bzero(drl, sizeof(*drl));
1337                 s = splnet();
1338                 dr = TAILQ_FIRST(&nd_defrouter);
1339                 while (dr && i < DRLSTSIZ) {
1340                         drl->defrouter[i].rtaddr = dr->rtaddr;
1341                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1342                                 /* XXX: need to this hack for KAME stack */
1343                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1344                         } else
1345                                 log(LOG_ERR,
1346                                     "default router list contains a "
1347                                     "non-linklocal address(%s)\n",
1348                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1349
1350                         drl->defrouter[i].flags = dr->flags;
1351                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1352                         drl->defrouter[i].expire = dr->expire;
1353                         drl->defrouter[i].if_index = dr->ifp->if_index;
1354                         i++;
1355                         dr = TAILQ_NEXT(dr, dr_entry);
1356                 }
1357                 splx(s);
1358                 break;
1359         case SIOCGPRLST_IN6:
1360                 /*
1361                  * obsolete API, use sysctl under net.inet6.icmp6
1362                  */
1363                 /*
1364                  * XXX meaning of fields, especialy "raflags", is very
1365                  * differnet between RA prefix list and RR/static prefix list.
1366                  * how about separating ioctls into two?
1367                  */
1368                 bzero(prl, sizeof(*prl));
1369                 s = splnet();
1370                 pr = nd_prefix.lh_first;
1371                 while (pr && i < PRLSTSIZ) {
1372                         struct nd_pfxrouter *pfr;
1373                         int j;
1374
1375                         (void)in6_embedscope(&prl->prefix[i].prefix,
1376                             &pr->ndpr_prefix, NULL, NULL);
1377                         prl->prefix[i].raflags = pr->ndpr_raf;
1378                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1379                         prl->prefix[i].vltime = pr->ndpr_vltime;
1380                         prl->prefix[i].pltime = pr->ndpr_pltime;
1381                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1382                         prl->prefix[i].expire = pr->ndpr_expire;
1383
1384                         pfr = pr->ndpr_advrtrs.lh_first;
1385                         j = 0;
1386                         while (pfr) {
1387                                 if (j < DRLSTSIZ) {
1388 #define RTRADDR prl->prefix[i].advrtr[j]
1389                                         RTRADDR = pfr->router->rtaddr;
1390                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1391                                                 /* XXX: hack for KAME */
1392                                                 RTRADDR.s6_addr16[1] = 0;
1393                                         } else
1394                                                 log(LOG_ERR,
1395                                                     "a router(%s) advertises "
1396                                                     "a prefix with "
1397                                                     "non-link local address\n",
1398                                                     ip6_sprintf(&RTRADDR));
1399 #undef RTRADDR
1400                                 }
1401                                 j++;
1402                                 pfr = pfr->pfr_next;
1403                         }
1404                         prl->prefix[i].advrtrs = j;
1405                         prl->prefix[i].origin = PR_ORIG_RA;
1406
1407                         i++;
1408                         pr = pr->ndpr_next;
1409                 }
1410               {
1411                 struct rr_prefix *rpp;
1412
1413                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1414                      rpp = LIST_NEXT(rpp, rp_entry)) {
1415                         if (i >= PRLSTSIZ)
1416                                 break;
1417                         (void)in6_embedscope(&prl->prefix[i].prefix,
1418                             &pr->ndpr_prefix, NULL, NULL);
1419                         prl->prefix[i].raflags = rpp->rp_raf;
1420                         prl->prefix[i].prefixlen = rpp->rp_plen;
1421                         prl->prefix[i].vltime = rpp->rp_vltime;
1422                         prl->prefix[i].pltime = rpp->rp_pltime;
1423                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1424                         prl->prefix[i].expire = rpp->rp_expire;
1425                         prl->prefix[i].advrtrs = 0;
1426                         prl->prefix[i].origin = rpp->rp_origin;
1427                         i++;
1428                 }
1429               }
1430                 splx(s);
1431
1432                 break;
1433         case OSIOCGIFINFO_IN6:
1434                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1435                         error = EINVAL;
1436                         break;
1437                 }
1438                 ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1439                 ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1440                 ndi->ndi.basereachable =
1441                     nd_ifinfo[ifp->if_index].basereachable;
1442                 ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1443                 ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1444                 ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1445                 ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1446                 ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1447                 ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1448                 break;
1449         case SIOCGIFINFO_IN6:
1450                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1451                         error = EINVAL;
1452                         break;
1453                 }
1454                 ndi->ndi = nd_ifinfo[ifp->if_index];
1455                 break;
1456         case SIOCSIFINFO_FLAGS:
1457                 /* XXX: almost all other fields of ndi->ndi is unused */
1458                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1459                         error = EINVAL;
1460                         break;
1461                 }
1462                 nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1463                 break;
1464         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1465                 /* flush default router list */
1466                 /*
1467                  * xxx sumikawa: should not delete route if default
1468                  * route equals to the top of default router list
1469                  */
1470                 bzero(&any, sizeof(any));
1471                 defrouter_delreq(&any, 0);
1472                 defrouter_select();
1473                 /* xxx sumikawa: flush prefix list */
1474                 break;
1475         case SIOCSPFXFLUSH_IN6:
1476             {
1477                 /* flush all the prefix advertised by routers */
1478                 struct nd_prefix *pr, *next;
1479
1480                 s = splnet();
1481                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1482                         struct in6_ifaddr *ia, *ia_next;
1483
1484                         next = pr->ndpr_next;
1485
1486                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1487                                 continue; /* XXX */
1488
1489                         /* do we really have to remove addresses as well? */
1490                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1491                                 /* ia might be removed.  keep the next ptr. */
1492                                 ia_next = ia->ia_next;
1493
1494                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1495                                         continue;
1496
1497                                 if (ia->ia6_ndpr == pr)
1498                                         in6_purgeaddr(&ia->ia_ifa);
1499                         }
1500                         prelist_remove(pr);
1501                 }
1502                 splx(s);
1503                 break;
1504             }
1505         case SIOCSRTRFLUSH_IN6:
1506             {
1507                 /* flush all the default routers */
1508                 struct nd_defrouter *dr, *next;
1509
1510                 s = splnet();
1511                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1512                         /*
1513                          * The first entry of the list may be stored in
1514                          * the routing table, so we'll delete it later.
1515                          */
1516                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1517                                 next = TAILQ_NEXT(dr, dr_entry);
1518                                 defrtrlist_del(dr);
1519                         }
1520                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1521                 }
1522                 splx(s);
1523                 break;
1524             }
1525         case SIOCGNBRINFO_IN6:
1526             {
1527                 struct llinfo_nd6 *ln;
1528                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1529
1530                 /*
1531                  * XXX: KAME specific hack for scoped addresses
1532                  *      XXXX: for other scopes than link-local?
1533                  */
1534                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1535                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1536                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1537
1538                         if (*idp == 0)
1539                                 *idp = htons(ifp->if_index);
1540                 }
1541
1542                 s = splnet();
1543                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1544                         error = EINVAL;
1545                         splx(s);
1546                         break;
1547                 }
1548                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1549                 nbi->state = ln->ln_state;
1550                 nbi->asked = ln->ln_asked;
1551                 nbi->isrouter = ln->ln_router;
1552                 nbi->expire = ln->ln_expire;
1553                 splx(s);
1554
1555                 break;
1556             }
1557         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1558                 ndif->ifindex = nd6_defifindex;
1559                 break;
1560         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1561                 return(nd6_setdefaultiface(ndif->ifindex));
1562                 break;
1563         }
1564         return(error);
1565 }
1566
1567 /*
1568  * Create neighbor cache entry and cache link-layer address,
1569  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1570  */
1571 struct rtentry *
1572 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1573                  int lladdrlen,
1574                  int type,      /* ICMP6 type */
1575                  int code       /* type dependent information */)
1576 {
1577         struct rtentry *rt = NULL;
1578         struct llinfo_nd6 *ln = NULL;
1579         int is_newentry;
1580         struct sockaddr_dl *sdl = NULL;
1581         int do_update;
1582         int olladdr;
1583         int llchange;
1584         int newstate = 0;
1585
1586         if (!ifp)
1587                 panic("ifp == NULL in nd6_cache_lladdr");
1588         if (!from)
1589                 panic("from == NULL in nd6_cache_lladdr");
1590
1591         /* nothing must be updated for unspecified address */
1592         if (IN6_IS_ADDR_UNSPECIFIED(from))
1593                 return NULL;
1594
1595         /*
1596          * Validation about ifp->if_addrlen and lladdrlen must be done in
1597          * the caller.
1598          *
1599          * XXX If the link does not have link-layer adderss, what should
1600          * we do? (ifp->if_addrlen == 0)
1601          * Spec says nothing in sections for RA, RS and NA.  There's small
1602          * description on it in NS section (RFC 2461 7.2.3).
1603          */
1604
1605         rt = nd6_lookup(from, 0, ifp);
1606         if (!rt) {
1607 #if 0
1608                 /* nothing must be done if there's no lladdr */
1609                 if (!lladdr || !lladdrlen)
1610                         return NULL;
1611 #endif
1612
1613                 rt = nd6_lookup(from, 1, ifp);
1614                 is_newentry = 1;
1615         } else {
1616                 /* do nothing if static ndp is set */
1617                 if (rt->rt_flags & RTF_STATIC)
1618                         return NULL;
1619                 is_newentry = 0;
1620         }
1621
1622         if (!rt)
1623                 return NULL;
1624         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1625 fail:
1626                 (void)nd6_free(rt);
1627                 return NULL;
1628         }
1629         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1630         if (!ln)
1631                 goto fail;
1632         if (!rt->rt_gateway)
1633                 goto fail;
1634         if (rt->rt_gateway->sa_family != AF_LINK)
1635                 goto fail;
1636         sdl = SDL(rt->rt_gateway);
1637
1638         olladdr = (sdl->sdl_alen) ? 1 : 0;
1639         if (olladdr && lladdr) {
1640                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1641                         llchange = 1;
1642                 else
1643                         llchange = 0;
1644         } else
1645                 llchange = 0;
1646
1647         /*
1648          * newentry olladdr  lladdr  llchange   (*=record)
1649          *      0       n       n       --      (1)
1650          *      0       y       n       --      (2)
1651          *      0       n       y       --      (3) * STALE
1652          *      0       y       y       n       (4) *
1653          *      0       y       y       y       (5) * STALE
1654          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1655          *      1       --      y       --      (7) * STALE
1656          */
1657
1658         if (lladdr) {           /* (3-5) and (7) */
1659                 /*
1660                  * Record source link-layer address
1661                  * XXX is it dependent to ifp->if_type?
1662                  */
1663                 sdl->sdl_alen = ifp->if_addrlen;
1664                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1665         }
1666
1667         if (!is_newentry) {
1668                 if ((!olladdr && lladdr)                /* (3) */
1669                  || (olladdr && lladdr && llchange)) {  /* (5) */
1670                         do_update = 1;
1671                         newstate = ND6_LLINFO_STALE;
1672                 } else                                  /* (1-2,4) */
1673                         do_update = 0;
1674         } else {
1675                 do_update = 1;
1676                 if (!lladdr)                            /* (6) */
1677                         newstate = ND6_LLINFO_NOSTATE;
1678                 else                                    /* (7) */
1679                         newstate = ND6_LLINFO_STALE;
1680         }
1681
1682         if (do_update) {
1683                 /*
1684                  * Update the state of the neighbor cache.
1685                  */
1686                 ln->ln_state = newstate;
1687
1688                 if (ln->ln_state == ND6_LLINFO_STALE) {
1689                         /*
1690                          * XXX: since nd6_output() below will cause
1691                          * state tansition to DELAY and reset the timer,
1692                          * we must set the timer now, although it is actually
1693                          * meaningless.
1694                          */
1695                         ln->ln_expire = time_second + nd6_gctimer;
1696
1697                         if (ln->ln_hold) {
1698                                 /*
1699                                  * we assume ifp is not a p2p here, so just
1700                                  * set the 2nd argument as the 1st one.
1701                                  */
1702                                 nd6_output(ifp, ifp, ln->ln_hold,
1703                                            (struct sockaddr_in6 *)rt_key(rt),
1704                                            rt);
1705                                 ln->ln_hold = NULL;
1706                         }
1707                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1708                         /* probe right away */
1709                         ln->ln_expire = time_second;
1710                 }
1711         }
1712
1713         /*
1714          * ICMP6 type dependent behavior.
1715          *
1716          * NS: clear IsRouter if new entry
1717          * RS: clear IsRouter
1718          * RA: set IsRouter if there's lladdr
1719          * redir: clear IsRouter if new entry
1720          *
1721          * RA case, (1):
1722          * The spec says that we must set IsRouter in the following cases:
1723          * - If lladdr exist, set IsRouter.  This means (1-5).
1724          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1725          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1726          * A quetion arises for (1) case.  (1) case has no lladdr in the
1727          * neighbor cache, this is similar to (6).
1728          * This case is rare but we figured that we MUST NOT set IsRouter.
1729          *
1730          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1731          *                                                      D R
1732          *      0       n       n       --      (1)     c   ?     s
1733          *      0       y       n       --      (2)     c   s     s
1734          *      0       n       y       --      (3)     c   s     s
1735          *      0       y       y       n       (4)     c   s     s
1736          *      0       y       y       y       (5)     c   s     s
1737          *      1       --      n       --      (6) c   c       c s
1738          *      1       --      y       --      (7) c   c   s   c s
1739          *
1740          *                                      (c=clear s=set)
1741          */
1742         switch (type & 0xff) {
1743         case ND_NEIGHBOR_SOLICIT:
1744                 /*
1745                  * New entry must have is_router flag cleared.
1746                  */
1747                 if (is_newentry)        /* (6-7) */
1748                         ln->ln_router = 0;
1749                 break;
1750         case ND_REDIRECT:
1751                 /*
1752                  * If the icmp is a redirect to a better router, always set the
1753                  * is_router flag. Otherwise, if the entry is newly created,
1754                  * clear the flag. [RFC 2461, sec 8.3]
1755                  */
1756                 if (code == ND_REDIRECT_ROUTER)
1757                         ln->ln_router = 1;
1758                 else if (is_newentry) /* (6-7) */
1759                         ln->ln_router = 0;
1760                 break;
1761         case ND_ROUTER_SOLICIT:
1762                 /*
1763                  * is_router flag must always be cleared.
1764                  */
1765                 ln->ln_router = 0;
1766                 break;
1767         case ND_ROUTER_ADVERT:
1768                 /*
1769                  * Mark an entry with lladdr as a router.
1770                  */
1771                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1772                  || (is_newentry && lladdr)) {                  /* (7) */
1773                         ln->ln_router = 1;
1774                 }
1775                 break;
1776         }
1777
1778         /*
1779          * When the link-layer address of a router changes, select the
1780          * best router again.  In particular, when the neighbor entry is newly
1781          * created, it might affect the selection policy.
1782          * Question: can we restrict the first condition to the "is_newentry"
1783          * case?
1784          * XXX: when we hear an RA from a new router with the link-layer
1785          * address option, defrouter_select() is called twice, since
1786          * defrtrlist_update called the function as well.  However, I believe
1787          * we can compromise the overhead, since it only happens the first
1788          * time.
1789          * XXX: although defrouter_select() should not have a bad effect
1790          * for those are not autoconfigured hosts, we explicitly avoid such
1791          * cases for safety.
1792          */
1793         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1794                 defrouter_select();
1795
1796         return rt;
1797 }
1798
1799 static void
1800 nd6_slowtimo(void *ignored_arg)
1801 {
1802         int s = splnet();
1803         int i;
1804         struct nd_ifinfo *nd6if;
1805
1806         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1807             nd6_slowtimo, NULL);
1808         for (i = 1; i < if_index + 1; i++) {
1809                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1810                         continue;
1811                 nd6if = &nd_ifinfo[i];
1812                 if (nd6if->basereachable && /* already initialized */
1813                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1814                         /*
1815                          * Since reachable time rarely changes by router
1816                          * advertisements, we SHOULD insure that a new random
1817                          * value gets recomputed at least once every few hours.
1818                          * (RFC 2461, 6.3.4)
1819                          */
1820                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1821                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1822                 }
1823         }
1824         splx(s);
1825 }
1826
1827 #define senderr(e) { error = (e); goto bad;}
1828 int
1829 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1830            struct sockaddr_in6 *dst, struct rtentry *rt0)
1831 {
1832         struct mbuf *m = m0;
1833         struct rtentry *rt = rt0;
1834         struct sockaddr_in6 *gw6 = NULL;
1835         struct llinfo_nd6 *ln = NULL;
1836         int error = 0;
1837
1838         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1839                 goto sendpkt;
1840
1841         if (nd6_need_cache(ifp) == 0)
1842                 goto sendpkt;
1843
1844         /*
1845          * next hop determination.  This routine is derived from ether_outpout.
1846          */
1847         if (rt) {
1848                 if ((rt->rt_flags & RTF_UP) == 0) {
1849                         if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL)) !=
1850                                 NULL)
1851                         {
1852                                 rt->rt_refcnt--;
1853                                 if (rt->rt_ifp != ifp) {
1854                                         /* XXX: loop care? */
1855                                         return nd6_output(ifp, origifp, m0,
1856                                                           dst, rt);
1857                                 }
1858                         } else
1859                                 senderr(EHOSTUNREACH);
1860                 }
1861
1862                 if (rt->rt_flags & RTF_GATEWAY) {
1863                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1864
1865                         /*
1866                          * We skip link-layer address resolution and NUD
1867                          * if the gateway is not a neighbor from ND point
1868                          * of view, regardless of the value of nd_ifinfo.flags.
1869                          * The second condition is a bit tricky; we skip
1870                          * if the gateway is our own address, which is
1871                          * sometimes used to install a route to a p2p link.
1872                          */
1873                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1874                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1875                                 /*
1876                                  * We allow this kind of tricky route only
1877                                  * when the outgoing interface is p2p.
1878                                  * XXX: we may need a more generic rule here.
1879                                  */
1880                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1881                                         senderr(EHOSTUNREACH);
1882
1883                                 goto sendpkt;
1884                         }
1885
1886                         if (rt->rt_gwroute == 0)
1887                                 goto lookup;
1888                         if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1889                                 rtfree(rt); rt = rt0;
1890                         lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1891                                 if ((rt = rt->rt_gwroute) == 0)
1892                                         senderr(EHOSTUNREACH);
1893                         }
1894                 }
1895         }
1896
1897         /*
1898          * Address resolution or Neighbor Unreachability Detection
1899          * for the next hop.
1900          * At this point, the destination of the packet must be a unicast
1901          * or an anycast address(i.e. not a multicast).
1902          */
1903
1904         /* Look up the neighbor cache for the nexthop */
1905         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1906                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1907         else {
1908                 /*
1909                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1910                  * the condition below is not very efficient.  But we believe
1911                  * it is tolerable, because this should be a rare case.
1912                  */
1913                 if (nd6_is_addr_neighbor(dst, ifp) &&
1914                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1915                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1916         }
1917         if (!ln || !rt) {
1918                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1919                     !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1920                         log(LOG_DEBUG,
1921                             "nd6_output: can't allocate llinfo for %s "
1922                             "(ln=%p, rt=%p)\n",
1923                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1924                         senderr(EIO);   /* XXX: good error? */
1925                 }
1926
1927                 goto sendpkt;   /* send anyway */
1928         }
1929
1930         /* We don't have to do link-layer address resolution on a p2p link. */
1931         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1932             ln->ln_state < ND6_LLINFO_REACHABLE) {
1933                 ln->ln_state = ND6_LLINFO_STALE;
1934                 ln->ln_expire = time_second + nd6_gctimer;
1935         }
1936
1937         /*
1938          * The first time we send a packet to a neighbor whose entry is
1939          * STALE, we have to change the state to DELAY and a sets a timer to
1940          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1941          * neighbor unreachability detection on expiration.
1942          * (RFC 2461 7.3.3)
1943          */
1944         if (ln->ln_state == ND6_LLINFO_STALE) {
1945                 ln->ln_asked = 0;
1946                 ln->ln_state = ND6_LLINFO_DELAY;
1947                 ln->ln_expire = time_second + nd6_delay;
1948         }
1949
1950         /*
1951          * If the neighbor cache entry has a state other than INCOMPLETE
1952          * (i.e. its link-layer address is already resolved), just
1953          * send the packet.
1954          */
1955         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1956                 goto sendpkt;
1957
1958         /*
1959          * There is a neighbor cache entry, but no ethernet address
1960          * response yet.  Replace the held mbuf (if any) with this
1961          * latest one.
1962          *
1963          * This code conforms to the rate-limiting rule described in Section
1964          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1965          * an NS below.
1966          */
1967         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1968                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1969         if (ln->ln_hold)
1970                 m_freem(ln->ln_hold);
1971         ln->ln_hold = m;
1972         if (ln->ln_expire) {
1973                 if (ln->ln_asked < nd6_mmaxtries &&
1974                     ln->ln_expire < time_second) {
1975                         ln->ln_asked++;
1976                         ln->ln_expire = time_second +
1977                                 nd_ifinfo[ifp->if_index].retrans / 1000;
1978                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1979                 }
1980         }
1981         return(0);
1982
1983   sendpkt:
1984
1985         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1986                 return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1987                                          rt));
1988         }
1989         return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1990
1991   bad:
1992         if (m)
1993                 m_freem(m);
1994         return (error);
1995 }
1996 #undef senderr
1997
1998 int
1999 nd6_need_cache(struct ifnet *ifp)
2000 {
2001         /*
2002          * XXX: we currently do not make neighbor cache on any interface
2003          * other than ARCnet, Ethernet, FDDI and GIF.
2004          *
2005          * RFC2893 says:
2006          * - unidirectional tunnels needs no ND
2007          */
2008         switch (ifp->if_type) {
2009         case IFT_ARCNET:
2010         case IFT_ETHER:
2011         case IFT_FDDI:
2012         case IFT_IEEE1394:
2013 #ifdef IFT_L2VLAN
2014         case IFT_L2VLAN:
2015 #endif
2016 #ifdef IFT_IEEE80211
2017         case IFT_IEEE80211:
2018 #endif
2019         case IFT_GIF:           /* XXX need more cases? */
2020                 return(1);
2021         default:
2022                 return(0);
2023         }
2024 }
2025
2026 int
2027 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt, struct mbuf *m,
2028                 struct sockaddr *dst, u_char *desten)
2029 {
2030         struct sockaddr_dl *sdl;
2031
2032         if (m->m_flags & M_MCAST) {
2033                 switch (ifp->if_type) {
2034                 case IFT_ETHER:
2035                 case IFT_FDDI:
2036 #ifdef IFT_L2VLAN
2037         case IFT_L2VLAN:
2038 #endif
2039 #ifdef IFT_IEEE80211
2040                 case IFT_IEEE80211:
2041 #endif
2042                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2043                                                  desten);
2044                         return(1);
2045                 case IFT_IEEE1394:
2046                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2047                         return(1);
2048                 case IFT_ARCNET:
2049                         *desten = 0;
2050                         return(1);
2051                 default:
2052                         m_freem(m);
2053                         return(0);
2054                 }
2055         }
2056
2057         if (rt == NULL) {
2058                 /* this could happen, if we could not allocate memory */
2059                 m_freem(m);
2060                 return(0);
2061         }
2062         if (rt->rt_gateway->sa_family != AF_LINK) {
2063                 printf("nd6_storelladdr: something odd happens\n");
2064                 m_freem(m);
2065                 return(0);
2066         }
2067         sdl = SDL(rt->rt_gateway);
2068         if (sdl->sdl_alen == 0) {
2069                 /* this should be impossible, but we bark here for debugging */
2070                 printf("nd6_storelladdr: sdl_alen == 0\n");
2071                 m_freem(m);
2072                 return(0);
2073         }
2074
2075         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2076         return(1);
2077 }
2078
2079 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2080 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2081 #ifdef SYSCTL_DECL
2082 SYSCTL_DECL(_net_inet6_icmp6);
2083 #endif
2084 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2085         CTLFLAG_RD, nd6_sysctl_drlist, "");
2086 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2087         CTLFLAG_RD, nd6_sysctl_prlist, "");
2088
2089 static int
2090 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2091 {
2092         int error;
2093         char buf[1024];
2094         struct in6_defrouter *d, *de;
2095         struct nd_defrouter *dr;
2096
2097         if (req->newptr)
2098                 return EPERM;
2099         error = 0;
2100
2101         for (dr = TAILQ_FIRST(&nd_defrouter);
2102              dr;
2103              dr = TAILQ_NEXT(dr, dr_entry)) {
2104                 d = (struct in6_defrouter *)buf;
2105                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2106
2107                 if (d + 1 <= de) {
2108                         bzero(d, sizeof(*d));
2109                         d->rtaddr.sin6_family = AF_INET6;
2110                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2111                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2112                             dr->ifp) != 0)
2113                                 log(LOG_ERR,
2114                                     "scope error in "
2115                                     "default router list (%s)\n",
2116                                     ip6_sprintf(&dr->rtaddr));
2117                         d->flags = dr->flags;
2118                         d->rtlifetime = dr->rtlifetime;
2119                         d->expire = dr->expire;
2120                         d->if_index = dr->ifp->if_index;
2121                 } else
2122                         panic("buffer too short");
2123
2124                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2125                 if (error)
2126                         break;
2127         }
2128         return error;
2129 }
2130
2131 static int
2132 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2133 {
2134         int error;
2135         char buf[1024];
2136         struct in6_prefix *p, *pe;
2137         struct nd_prefix *pr;
2138
2139         if (req->newptr)
2140                 return EPERM;
2141         error = 0;
2142
2143         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2144                 u_short advrtrs;
2145                 size_t advance;
2146                 struct sockaddr_in6 *sin6, *s6;
2147                 struct nd_pfxrouter *pfr;
2148
2149                 p = (struct in6_prefix *)buf;
2150                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2151
2152                 if (p + 1 <= pe) {
2153                         bzero(p, sizeof(*p));
2154                         sin6 = (struct sockaddr_in6 *)(p + 1);
2155
2156                         p->prefix = pr->ndpr_prefix;
2157                         if (in6_recoverscope(&p->prefix,
2158                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2159                                 log(LOG_ERR,
2160                                     "scope error in prefix list (%s)\n",
2161                                     ip6_sprintf(&p->prefix.sin6_addr));
2162                         p->raflags = pr->ndpr_raf;
2163                         p->prefixlen = pr->ndpr_plen;
2164                         p->vltime = pr->ndpr_vltime;
2165                         p->pltime = pr->ndpr_pltime;
2166                         p->if_index = pr->ndpr_ifp->if_index;
2167                         p->expire = pr->ndpr_expire;
2168                         p->refcnt = pr->ndpr_refcnt;
2169                         p->flags = pr->ndpr_stateflags;
2170                         p->origin = PR_ORIG_RA;
2171                         advrtrs = 0;
2172                         for (pfr = pr->ndpr_advrtrs.lh_first;
2173                              pfr;
2174                              pfr = pfr->pfr_next) {
2175                                 if ((void *)&sin6[advrtrs + 1] >
2176                                     (void *)pe) {
2177                                         advrtrs++;
2178                                         continue;
2179                                 }
2180                                 s6 = &sin6[advrtrs];
2181                                 bzero(s6, sizeof(*s6));
2182                                 s6->sin6_family = AF_INET6;
2183                                 s6->sin6_len = sizeof(*sin6);
2184                                 if (in6_recoverscope(s6,
2185                                     &pfr->router->rtaddr,
2186                                     pfr->router->ifp) != 0)
2187                                         log(LOG_ERR,
2188                                             "scope error in "
2189                                             "prefix list (%s)\n",
2190                                             ip6_sprintf(&pfr->router->rtaddr));
2191                                 advrtrs++;
2192                         }
2193                         p->advrtrs = advrtrs;
2194                 } else
2195                         panic("buffer too short");
2196
2197                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2198                 error = SYSCTL_OUT(req, buf, advance);
2199                 if (error)
2200                         break;
2201         }
2202         return error;
2203 }