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 1.9 2004/08/13 14:27:29 drhodus 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_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 147             nd6_slowtimo, NULL);
 148 }
 149
 150 void
 151 nd6_ifattach(struct ifnet *ifp)
 152 {
 153
 154         /*
 155          * We have some arrays that should be indexed by if_index.
 156          * since if_index will grow dynamically, they should grow too.
 157          */
 158         if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
 159                 size_t n;
 160                 caddr_t q;
 161
 162                 while (if_index >= nd_ifinfo_indexlim)
 163                         nd_ifinfo_indexlim <<= 1;
 164
 165                 /* grow nd_ifinfo */
 166                 n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
 167                 q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
 168                 bzero(q, n);
 169                 if (nd_ifinfo) {
 170                         bcopy((caddr_t)nd_ifinfo, q, n/2);
 171                         free((caddr_t)nd_ifinfo, M_IP6NDP);
 172                 }
 173                 nd_ifinfo = (struct nd_ifinfo *)q;
 174         }
 175
 176 #define ND nd_ifinfo[ifp->if_index]
 177
 178         /*
 179          * Don't initialize if called twice.
 180          * XXX: to detect this, we should choose a member that is never set
 181          * before initialization of the ND structure itself.  We formaly used
 182          * the linkmtu member, which was not suitable because it could be
 183          * initialized via "ifconfig mtu".
 184          */
 185         if (ND.basereachable)
 186                 return;
 187
 188         ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
 189         ND.chlim = IPV6_DEFHLIM;
 190         ND.basereachable = REACHABLE_TIME;
 191         ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
 192         ND.retrans = RETRANS_TIMER;
 193         ND.receivedra = 0;
 194         ND.flags = ND6_IFF_PERFORMNUD;
 195         nd6_setmtu(ifp);
 196 #undef ND
 197 }
 198
 199 /*
 200  * Reset ND level link MTU. This function is called when the physical MTU
 201  * changes, which means we might have to adjust the ND level MTU.
 202  */
 203 void
 204 nd6_setmtu(struct ifnet *ifp)
 205 {
 206         struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
 207         u_long oldmaxmtu = ndi->maxmtu;
 208         u_long oldlinkmtu = ndi->linkmtu;
 209
 210         switch (ifp->if_type) {
 211         case IFT_ARCNET:        /* XXX MTU handling needs more work */
 212                 ndi->maxmtu = MIN(60480, ifp->if_mtu);
 213                 break;
 214         case IFT_ETHER:
 215                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 216                 break;
 217         case IFT_FDDI:
 218                 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
 219                 break;
 220         case IFT_ATM:
 221                 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
 222                 break;
 223         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 224                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 225                 break;
 226 #ifdef IFT_IEEE80211
 227         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 228                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 229                 break;
 230 #endif
 231         default:
 232                 ndi->maxmtu = ifp->if_mtu;
 233                 break;
 234         }
 235
 236         if (oldmaxmtu != ndi->maxmtu) {
 237                 /*
 238                  * If the ND level MTU is not set yet, or if the maxmtu
 239                  * is reset to a smaller value than the ND level MTU,
 240                  * also reset the ND level MTU.
 241                  */
 242                 if (ndi->linkmtu == 0 ||
 243                     ndi->maxmtu < ndi->linkmtu) {
 244                         ndi->linkmtu = ndi->maxmtu;
 245                         /* also adjust in6_maxmtu if necessary. */
 246                         if (oldlinkmtu == 0) {
 247                                 /*
 248                                  * XXX: the case analysis is grotty, but
 249                                  * it is not efficient to call in6_setmaxmtu()
 250                                  * here when we are during the initialization
 251                                  * procedure.
 252                                  */
 253                                 if (in6_maxmtu < ndi->linkmtu)
 254                                         in6_maxmtu = ndi->linkmtu;
 255                         } else
 256                                 in6_setmaxmtu();
 257                 }
 258         }
 259 #undef MIN
 260 }
 261
 262 void
 263 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 264 {
 265         bzero(ndopts, sizeof(*ndopts));
 266         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 267         ndopts->nd_opts_last
 268                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 269
 270         if (icmp6len == 0) {
 271                 ndopts->nd_opts_done = 1;
 272                 ndopts->nd_opts_search = NULL;
 273         }
 274 }
 275
 276 /*
 277  * Take one ND option.
 278  */
 279 struct nd_opt_hdr *
 280 nd6_option(union nd_opts *ndopts)
 281 {
 282         struct nd_opt_hdr *nd_opt;
 283         int olen;
 284
 285         if (!ndopts)
 286                 panic("ndopts == NULL in nd6_option");
 287         if (!ndopts->nd_opts_last)
 288                 panic("uninitialized ndopts in nd6_option");
 289         if (!ndopts->nd_opts_search)
 290                 return NULL;
 291         if (ndopts->nd_opts_done)
 292                 return NULL;
 293
 294         nd_opt = ndopts->nd_opts_search;
 295
 296         /* make sure nd_opt_len is inside the buffer */
 297         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 298                 bzero(ndopts, sizeof(*ndopts));
 299                 return NULL;
 300         }
 301
 302         olen = nd_opt->nd_opt_len << 3;
 303         if (olen == 0) {
 304                 /*
 305                  * Message validation requires that all included
 306                  * options have a length that is greater than zero.
 307                  */
 308                 bzero(ndopts, sizeof(*ndopts));
 309                 return NULL;
 310         }
 311
 312         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 313         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 314                 /* option overruns the end of buffer, invalid */
 315                 bzero(ndopts, sizeof(*ndopts));
 316                 return NULL;
 317         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 318                 /* reached the end of options chain */
 319                 ndopts->nd_opts_done = 1;
 320                 ndopts->nd_opts_search = NULL;
 321         }
 322         return nd_opt;
 323 }
 324
 325 /*
 326  * Parse multiple ND options.
 327  * This function is much easier to use, for ND routines that do not need
 328  * multiple options of the same type.
 329  */
 330 int
 331 nd6_options(union nd_opts *ndopts)
 332 {
 333         struct nd_opt_hdr *nd_opt;
 334         int i = 0;
 335
 336         if (!ndopts)
 337                 panic("ndopts == NULL in nd6_options");
 338         if (!ndopts->nd_opts_last)
 339                 panic("uninitialized ndopts in nd6_options");
 340         if (!ndopts->nd_opts_search)
 341                 return 0;
 342
 343         while (1) {
 344                 nd_opt = nd6_option(ndopts);
 345                 if (!nd_opt && !ndopts->nd_opts_last) {
 346                         /*
 347                          * Message validation requires that all included
 348                          * options have a length that is greater than zero.
 349                          */
 350                         icmp6stat.icp6s_nd_badopt++;
 351                         bzero(ndopts, sizeof(*ndopts));
 352                         return -1;
 353                 }
 354
 355                 if (!nd_opt)
 356                         goto skip1;
 357
 358                 switch (nd_opt->nd_opt_type) {
 359                 case ND_OPT_SOURCE_LINKADDR:
 360                 case ND_OPT_TARGET_LINKADDR:
 361                 case ND_OPT_MTU:
 362                 case ND_OPT_REDIRECTED_HEADER:
 363                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 364                                 nd6log((LOG_INFO,
 365                                     "duplicated ND6 option found (type=%d)\n",
 366                                     nd_opt->nd_opt_type));
 367                                 /* XXX bark? */
 368                         } else {
 369                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 370                                         = nd_opt;
 371                         }
 372                         break;
 373                 case ND_OPT_PREFIX_INFORMATION:
 374                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 375                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 376                                         = nd_opt;
 377                         }
 378                         ndopts->nd_opts_pi_end =
 379                                 (struct nd_opt_prefix_info *)nd_opt;
 380                         break;
 381                 default:
 382                         /*
 383                          * Unknown options must be silently ignored,
 384                          * to accomodate future extension to the protocol.
 385                          */
 386                         nd6log((LOG_DEBUG,
 387                             "nd6_options: unsupported option %d - "
 388                             "option ignored\n", nd_opt->nd_opt_type));
 389                 }
 390
 391 skip1:
 392                 i++;
 393                 if (i > nd6_maxndopt) {
 394                         icmp6stat.icp6s_nd_toomanyopt++;
 395                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 396                         break;
 397                 }
 398
 399                 if (ndopts->nd_opts_done)
 400                         break;
 401         }
 402
 403         return 0;
 404 }
 405
 406 /*
 407  * ND6 timer routine to expire default route list and prefix list
 408  */
 409 void
 410 nd6_timer(void *ignored_arg)
 411 {
 412         int s;
 413         struct llinfo_nd6 *ln;
 414         struct nd_defrouter *dr;
 415         struct nd_prefix *pr;
 416         struct ifnet *ifp;
 417         struct in6_ifaddr *ia6, *nia6;
 418         struct in6_addrlifetime *lt6;
 419
 420         s = splnet();
 421         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 422                       nd6_timer, NULL);
 423
 424         ln = llinfo_nd6.ln_next;
 425         while (ln && ln != &llinfo_nd6) {
 426                 struct rtentry *rt;
 427                 struct sockaddr_in6 *dst;
 428                 struct llinfo_nd6 *next = ln->ln_next;
 429                 /* XXX: used for the DELAY case only: */
 430                 struct nd_ifinfo *ndi = NULL;
 431
 432                 if ((rt = ln->ln_rt) == NULL) {
 433                         ln = next;
 434                         continue;
 435                 }
 436                 if ((ifp = rt->rt_ifp) == NULL) {
 437                         ln = next;
 438                         continue;
 439                 }
 440                 ndi = &nd_ifinfo[ifp->if_index];
 441                 dst = (struct sockaddr_in6 *)rt_key(rt);
 442
 443                 if (ln->ln_expire > time_second) {
 444                         ln = next;
 445                         continue;
 446                 }
 447
 448                 /* sanity check */
 449                 if (!rt)
 450                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 451                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 452                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 453                               rt->rt_llinfo, ln);
 454                 if (!dst)
 455                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 456
 457                 switch (ln->ln_state) {
 458                 case ND6_LLINFO_INCOMPLETE:
 459                         if (ln->ln_asked < nd6_mmaxtries) {
 460                                 ln->ln_asked++;
 461                                 ln->ln_expire = time_second +
 462                                         nd_ifinfo[ifp->if_index].retrans / 1000;
 463                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 464                                         ln, 0);
 465                         } else {
 466                                 struct mbuf *m = ln->ln_hold;
 467                                 if (m) {
 468                                         if (rt->rt_ifp) {
 469                                                 /*
 470                                                  * Fake rcvif to make ICMP error
 471                                                  * more helpful in diagnosing
 472                                                  * for the receiver.
 473                                                  * XXX: should we consider
 474                                                  * older rcvif?
 475                                                  */
 476                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 477                                         }
 478                                         icmp6_error(m, ICMP6_DST_UNREACH,
 479                                                     ICMP6_DST_UNREACH_ADDR, 0);
 480                                         ln->ln_hold = NULL;
 481                                 }
 482                                 next = nd6_free(rt);
 483                         }
 484                         break;
 485                 case ND6_LLINFO_REACHABLE:
 486                         if (ln->ln_expire) {
 487                                 ln->ln_state = ND6_LLINFO_STALE;
 488                                 ln->ln_expire = time_second + nd6_gctimer;
 489                         }
 490                         break;
 491
 492                 case ND6_LLINFO_STALE:
 493                         /* Garbage Collection(RFC 2461 5.3) */
 494                         if (ln->ln_expire)
 495                                 next = nd6_free(rt);
 496                         break;
 497
 498                 case ND6_LLINFO_DELAY:
 499                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
 500                                 /* We need NUD */
 501                                 ln->ln_asked = 1;
 502                                 ln->ln_state = ND6_LLINFO_PROBE;
 503                                 ln->ln_expire = time_second +
 504                                         ndi->retrans / 1000;
 505                                 nd6_ns_output(ifp, &dst->sin6_addr,
 506                                               &dst->sin6_addr,
 507                                               ln, 0);
 508                         } else {
 509                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 510                                 ln->ln_expire = time_second + nd6_gctimer;
 511                         }
 512                         break;
 513                 case ND6_LLINFO_PROBE:
 514                         if (ln->ln_asked < nd6_umaxtries) {
 515                                 ln->ln_asked++;
 516                                 ln->ln_expire = time_second +
 517                                         nd_ifinfo[ifp->if_index].retrans / 1000;
 518                                 nd6_ns_output(ifp, &dst->sin6_addr,
 519                                                &dst->sin6_addr, ln, 0);
 520                         } else {
 521                                 next = nd6_free(rt);
 522                         }
 523                         break;
 524                 }
 525                 ln = next;
 526         }
 527
 528         /* expire default router list */
 529         dr = TAILQ_FIRST(&nd_defrouter);
 530         while (dr) {
 531                 if (dr->expire && dr->expire < time_second) {
 532                         struct nd_defrouter *t;
 533                         t = TAILQ_NEXT(dr, dr_entry);
 534                         defrtrlist_del(dr);
 535                         dr = t;
 536                 } else {
 537                         dr = TAILQ_NEXT(dr, dr_entry);
 538                 }
 539         }
 540
 541         /*
 542          * expire interface addresses.
 543          * in the past the loop was inside prefix expiry processing.
 544          * However, from a stricter speci-confrmance standpoint, we should
 545          * rather separate address lifetimes and prefix lifetimes.
 546          */
 547   addrloop:
 548         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 549                 nia6 = ia6->ia_next;
 550                 /* check address lifetime */
 551                 lt6 = &ia6->ia6_lifetime;
 552                 if (IFA6_IS_INVALID(ia6)) {
 553                         int regen = 0;
 554
 555                         /*
 556                          * If the expiring address is temporary, try
 557                          * regenerating a new one.  This would be useful when
 558                          * we suspended a laptop PC, then turned it on after a
 559                          * period that could invalidate all temporary
 560                          * addresses.  Although we may have to restart the
 561                          * loop (see below), it must be after purging the
 562                          * address.  Otherwise, we'd see an infinite loop of
 563                          * regeneration.
 564                          */
 565                         if (ip6_use_tempaddr &&
 566                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
 567                                 if (regen_tmpaddr(ia6) == 0)
 568                                         regen = 1;
 569                         }
 570
 571                         in6_purgeaddr(&ia6->ia_ifa);
 572
 573                         if (regen)
 574                                 goto addrloop; /* XXX: see below */
 575                 }
 576                 if (IFA6_IS_DEPRECATED(ia6)) {
 577                         int oldflags = ia6->ia6_flags;
 578
 579                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 580
 581                         /*
 582                          * If a temporary address has just become deprecated,
 583                          * regenerate a new one if possible.
 584                          */
 585                         if (ip6_use_tempaddr &&
 586                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 587                             (oldflags & IN6_IFF_DEPRECATED) == 0) {
 588
 589                                 if (regen_tmpaddr(ia6) == 0) {
 590                                         /*
 591                                          * A new temporary address is
 592                                          * generated.
 593                                          * XXX: this means the address chain
 594                                          * has changed while we are still in
 595                                          * the loop.  Although the change
 596                                          * would not cause disaster (because
 597                                          * it's not a deletion, but an
 598                                          * addition,) we'd rather restart the
 599                                          * loop just for safety.  Or does this
 600                                          * significantly reduce performance??
 601                                          */
 602                                         goto addrloop;
 603                                 }
 604                         }
 605                 } else {
 606                         /*
 607                          * A new RA might have made a deprecated address
 608                          * preferred.
 609                          */
 610                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 611                 }
 612         }
 613
 614         /* expire prefix list */
 615         pr = nd_prefix.lh_first;
 616         while (pr) {
 617                 /*
 618                  * check prefix lifetime.
 619                  * since pltime is just for autoconf, pltime processing for
 620                  * prefix is not necessary.
 621                  */
 622                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
 623                         struct nd_prefix *t;
 624                         t = pr->ndpr_next;
 625
 626                         /*
 627                          * address expiration and prefix expiration are
 628                          * separate.  NEVER perform in6_purgeaddr here.
 629                          */
 630
 631                         prelist_remove(pr);
 632                         pr = t;
 633                 } else
 634                         pr = pr->ndpr_next;
 635         }
 636         splx(s);
 637 }
 638
 639 static int
 640 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 641                                          address */
 642 {
 643         struct ifaddr *ifa;
 644         struct ifnet *ifp;
 645         struct in6_ifaddr *public_ifa6 = NULL;
 646
 647         ifp = ia6->ia_ifa.ifa_ifp;
 648         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_list) {
 649                 struct in6_ifaddr *it6;
 650
 651                 if (ifa->ifa_addr->sa_family != AF_INET6)
 652                         continue;
 653
 654                 it6 = (struct in6_ifaddr *)ifa;
 655
 656                 /* ignore no autoconf addresses. */
 657                 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
 658                         continue;
 659
 660                 /* ignore autoconf addresses with different prefixes. */
 661                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 662                         continue;
 663
 664                 /*
 665                  * Now we are looking at an autoconf address with the same
 666                  * prefix as ours.  If the address is temporary and is still
 667                  * preferred, do not create another one.  It would be rare, but
 668                  * could happen, for example, when we resume a laptop PC after
 669                  * a long period.
 670                  */
 671                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 672                     !IFA6_IS_DEPRECATED(it6)) {
 673                         public_ifa6 = NULL;
 674                         break;
 675                 }
 676
 677                 /*
 678                  * This is a public autoconf address that has the same prefix
 679                  * as ours.  If it is preferred, keep it.  We can't break the
 680                  * loop here, because there may be a still-preferred temporary
 681                  * address with the prefix.
 682                  */
 683                 if (!IFA6_IS_DEPRECATED(it6))
 684                     public_ifa6 = it6;
 685         }
 686
 687         if (public_ifa6 != NULL) {
 688                 int e;
 689
 690                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 691                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 692                             " tmp addr,errno=%d\n", e);
 693                         return(-1);
 694                 }
 695                 return(0);
 696         }
 697
 698         return(-1);
 699 }
 700
 701 /*
 702  * Nuke neighbor cache/prefix/default router management table, right before
 703  * ifp goes away.
 704  */
 705 void
 706 nd6_purge(struct ifnet *ifp)
 707 {
 708         struct llinfo_nd6 *ln, *nln;
 709         struct nd_defrouter *dr, *ndr, drany;
 710         struct nd_prefix *pr, *npr;
 711
 712         /* Nuke default router list entries toward ifp */
 713         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 714                 /*
 715                  * The first entry of the list may be stored in
 716                  * the routing table, so we'll delete it later.
 717                  */
 718                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 719                         ndr = TAILQ_NEXT(dr, dr_entry);
 720                         if (dr->ifp == ifp)
 721                                 defrtrlist_del(dr);
 722                 }
 723                 dr = TAILQ_FIRST(&nd_defrouter);
 724                 if (dr->ifp == ifp)
 725                         defrtrlist_del(dr);
 726         }
 727
 728         /* Nuke prefix list entries toward ifp */
 729         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 730                 npr = pr->ndpr_next;
 731                 if (pr->ndpr_ifp == ifp) {
 732                         /*
 733                          * Previously, pr->ndpr_addr is removed as well,
 734                          * but I strongly believe we don't have to do it.
 735                          * nd6_purge() is only called from in6_ifdetach(),
 736                          * which removes all the associated interface addresses
 737                          * by itself.
 738                          * (jinmei@kame.net 20010129)
 739                          */
 740                         prelist_remove(pr);
 741                 }
 742         }
 743
 744         /* cancel default outgoing interface setting */
 745         if (nd6_defifindex == ifp->if_index)
 746                 nd6_setdefaultiface(0);
 747
 748         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 749                 /* refresh default router list */
 750                 bzero(&drany, sizeof(drany));
 751                 defrouter_delreq(&drany, 0);
 752                 defrouter_select();
 753         }
 754
 755         /*
 756          * Nuke neighbor cache entries for the ifp.
 757          * Note that rt->rt_ifp may not be the same as ifp,
 758          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 759          * nd6_rtrequest(), and ip6_input().
 760          */
 761         ln = llinfo_nd6.ln_next;
 762         while (ln && ln != &llinfo_nd6) {
 763                 struct rtentry *rt;
 764                 struct sockaddr_dl *sdl;
 765
 766                 nln = ln->ln_next;
 767                 rt = ln->ln_rt;
 768                 if (rt && rt->rt_gateway &&
 769                     rt->rt_gateway->sa_family == AF_LINK) {
 770                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 771                         if (sdl->sdl_index == ifp->if_index)
 772                                 nln = nd6_free(rt);
 773                 }
 774                 ln = nln;
 775         }
 776 }
 777
 778 struct rtentry *
 779 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 780 {
 781         struct rtentry *rt;
 782         struct sockaddr_in6 sin6;
 783
 784         bzero(&sin6, sizeof(sin6));
 785         sin6.sin6_len = sizeof(struct sockaddr_in6);
 786         sin6.sin6_family = AF_INET6;
 787         sin6.sin6_addr = *addr6;
 788 #ifdef SCOPEDROUTING
 789         sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
 790 #endif
 791         rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
 792         if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
 793                 /*
 794                  * This is the case for the default route.
 795                  * If we want to create a neighbor cache for the address, we
 796                  * should free the route for the destination and allocate an
 797                  * interface route.
 798                  */
 799                 if (create) {
 800                         RTFREE(rt);
 801                         rt = 0;
 802                 }
 803         }
 804         if (!rt) {
 805                 if (create && ifp) {
 806                         int e;
 807
 808                         /*
 809                          * If no route is available and create is set,
 810                          * we allocate a host route for the destination
 811                          * and treat it like an interface route.
 812                          * This hack is necessary for a neighbor which can't
 813                          * be covered by our own prefix.
 814                          */
 815                         struct ifaddr *ifa =
 816                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 817                         if (ifa == NULL)
 818                                 return(NULL);
 819
 820                         /*
 821                          * Create a new route.  RTF_LLINFO is necessary
 822                          * to create a Neighbor Cache entry for the
 823                          * destination in nd6_rtrequest which will be
 824                          * called in rtrequest via ifa->ifa_rtrequest.
 825                          */
 826                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 827                                            ifa->ifa_addr,
 828                                            (struct sockaddr *)&all1_sa,
 829                                            (ifa->ifa_flags |
 830                                             RTF_HOST | RTF_LLINFO) &
 831                                            ~RTF_CLONING,
 832                                            &rt)) != 0)
 833                                 log(LOG_ERR,
 834                                     "nd6_lookup: failed to add route for a "
 835                                     "neighbor(%s), errno=%d\n",
 836                                     ip6_sprintf(addr6), e);
 837                         if (rt == NULL)
 838                                 return(NULL);
 839                         if (rt->rt_llinfo) {
 840                                 struct llinfo_nd6 *ln =
 841                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 842                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 843                         }
 844                 } else
 845                         return(NULL);
 846         }
 847         rt->rt_refcnt--;
 848         /*
 849          * Validation for the entry.
 850          * Note that the check for rt_llinfo is necessary because a cloned
 851          * route from a parent route that has the L flag (e.g. the default
 852          * route to a p2p interface) may have the flag, too, while the
 853          * destination is not actually a neighbor.
 854          * XXX: we can't use rt->rt_ifp to check for the interface, since
 855          *      it might be the loopback interface if the entry is for our
 856          *      own address on a non-loopback interface. Instead, we should
 857          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 858          *      interface.
 859          */
 860         if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
 861             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 862             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 863                 if (create) {
 864                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 865                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 866                         /* xxx more logs... kazu */
 867                 }
 868                 return(NULL);
 869         }
 870         return(rt);
 871 }
 872
 873 /*
 874  * Detect if a given IPv6 address identifies a neighbor on a given link.
 875  * XXX: should take care of the destination of a p2p link?
 876  */
 877 int
 878 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 879 {
 880         struct ifaddr *ifa;
 881         int i;
 882
 883 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 884 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 885
 886         /*
 887          * A link-local address is always a neighbor.
 888          * XXX: we should use the sin6_scope_id field rather than the embedded
 889          * interface index.
 890          */
 891         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 892             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 893                 return(1);
 894
 895         /*
 896          * If the address matches one of our addresses,
 897          * it should be a neighbor.
 898          */
 899         TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
 900                 if (ifa->ifa_addr->sa_family != AF_INET6)
 901                         next: continue;
 902
 903                 for (i = 0; i < 4; i++) {
 904                         if ((IFADDR6(ifa).s6_addr32[i] ^
 905                              addr->sin6_addr.s6_addr32[i]) &
 906                             IFMASK6(ifa).s6_addr32[i])
 907                                 goto next;
 908                 }
 909                 return(1);
 910         }
 911
 912         /*
 913          * Even if the address matches none of our addresses, it might be
 914          * in the neighbor cache.
 915          */
 916         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 917                 return(1);
 918
 919         return(0);
 920 #undef IFADDR6
 921 #undef IFMASK6
 922 }
 923
 924 /*
 925  * Free an nd6 llinfo entry.
 926  */
 927 struct llinfo_nd6 *
 928 nd6_free(struct rtentry *rt)
 929 {
 930         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 931         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 932         struct nd_defrouter *dr;
 933
 934         /*
 935          * we used to have pfctlinput(PRC_HOSTDEAD) here.
 936          * even though it is not harmful, it was not really necessary.
 937          */
 938
 939         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 940                 int s;
 941                 s = splnet();
 942                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 943                                       rt->rt_ifp);
 944
 945                 if (ln->ln_router || dr) {
 946                         /*
 947                          * rt6_flush must be called whether or not the neighbor
 948                          * is in the Default Router List.
 949                          * See a corresponding comment in nd6_na_input().
 950                          */
 951                         rt6_flush(&in6, rt->rt_ifp);
 952                 }
 953
 954                 if (dr) {
 955                         /*
 956                          * Unreachablity of a router might affect the default
 957                          * router selection and on-link detection of advertised
 958                          * prefixes.
 959                          */
 960
 961                         /*
 962                          * Temporarily fake the state to choose a new default
 963                          * router and to perform on-link determination of
 964                          * prefixes correctly.
 965                          * Below the state will be set correctly,
 966                          * or the entry itself will be deleted.
 967                          */
 968                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 969
 970                         /*
 971                          * Since defrouter_select() does not affect the
 972                          * on-link determination and MIP6 needs the check
 973                          * before the default router selection, we perform
 974                          * the check now.
 975                          */
 976                         pfxlist_onlink_check();
 977
 978                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 979                                 /*
 980                                  * It is used as the current default router,
 981                                  * so we have to move it to the end of the
 982                                  * list and choose a new one.
 983                                  * XXX: it is not very efficient if this is
 984                                  *      the only router.
 985                                  */
 986                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
 987                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
 988
 989                                 defrouter_select();
 990                         }
 991                 }
 992                 splx(s);
 993         }
 994
 995         /*
 996          * Before deleting the entry, remember the next entry as the
 997          * return value.  We need this because pfxlist_onlink_check() above
 998          * might have freed other entries (particularly the old next entry) as
 999          * a side effect (XXX).
1000          */
1001         next = ln->ln_next;
1002
1003         /*
1004          * Detach the route from the routing tree and the list of neighbor
1005          * caches, and disable the route entry not to be used in already
1006          * cached routes.
1007          */
1008         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1009                   rt_mask(rt), 0, (struct rtentry **)0);
1010
1011         return(next);
1012 }
1013
1014 /*
1015  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1016  *
1017  * XXX cost-effective metods?
1018  */
1019 void
1020 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1021 {
1022         struct llinfo_nd6 *ln;
1023
1024         /*
1025          * If the caller specified "rt", use that.  Otherwise, resolve the
1026          * routing table by supplied "dst6".
1027          */
1028         if (!rt) {
1029                 if (!dst6)
1030                         return;
1031                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1032                         return;
1033         }
1034
1035         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1036             (rt->rt_flags & RTF_LLINFO) == 0 ||
1037             !rt->rt_llinfo || !rt->rt_gateway ||
1038             rt->rt_gateway->sa_family != AF_LINK) {
1039                 /* This is not a host route. */
1040                 return;
1041         }
1042
1043         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1044         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1045                 return;
1046
1047         /*
1048          * if we get upper-layer reachability confirmation many times,
1049          * it is possible we have false information.
1050          */
1051         if (!force) {
1052                 ln->ln_byhint++;
1053                 if (ln->ln_byhint > nd6_maxnudhint)
1054                         return;
1055         }
1056
1057         ln->ln_state = ND6_LLINFO_REACHABLE;
1058         if (ln->ln_expire)
1059                 ln->ln_expire = time_second +
1060                         nd_ifinfo[rt->rt_ifp->if_index].reachable;
1061 }
1062
1063 void
1064 nd6_rtrequest(int req, struct rtentry *rt,
1065               struct rt_addrinfo *info) /* xxx unused */
1066 {
1067         struct sockaddr *gate = rt->rt_gateway;
1068         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1069         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1070         struct ifnet *ifp = rt->rt_ifp;
1071         struct ifaddr *ifa;
1072
1073         if ((rt->rt_flags & RTF_GATEWAY))
1074                 return;
1075
1076         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1077                 /*
1078                  * This is probably an interface direct route for a link
1079                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1080                  * We do not need special treatment below for such a route.
1081                  * Moreover, the RTF_LLINFO flag which would be set below
1082                  * would annoy the ndp(8) command.
1083                  */
1084                 return;
1085         }
1086
1087         if (req == RTM_RESOLVE &&
1088             (nd6_need_cache(ifp) == 0 || /* stf case */
1089              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1090                 /*
1091                  * FreeBSD and BSD/OS often make a cloned host route based
1092                  * on a less-specific route (e.g. the default route).
1093                  * If the less specific route does not have a "gateway"
1094                  * (this is the case when the route just goes to a p2p or an
1095                  * stf interface), we'll mistakenly make a neighbor cache for
1096                  * the host route, and will see strange neighbor solicitation
1097                  * for the corresponding destination.  In order to avoid the
1098                  * confusion, we check if the destination of the route is
1099                  * a neighbor in terms of neighbor discovery, and stop the
1100                  * process if not.  Additionally, we remove the LLINFO flag
1101                  * so that ndp(8) will not try to get the neighbor information
1102                  * of the destination.
1103                  */
1104                 rt->rt_flags &= ~RTF_LLINFO;
1105                 return;
1106         }
1107
1108         switch (req) {
1109         case RTM_ADD:
1110                 /*
1111                  * There is no backward compatibility :)
1112                  *
1113                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1114                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1115                  *         rt->rt_flags |= RTF_CLONING;
1116                  */
1117                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1118                         /*
1119                          * Case 1: This route should come from
1120                          * a route to interface.  RTF_LLINFO flag is set
1121                          * for a host route whose destination should be
1122                          * treated as on-link.
1123                          */
1124                         rt_setgate(rt, rt_key(rt),
1125                                    (struct sockaddr *)&null_sdl);
1126                         gate = rt->rt_gateway;
1127                         SDL(gate)->sdl_type = ifp->if_type;
1128                         SDL(gate)->sdl_index = ifp->if_index;
1129                         if (ln)
1130                                 ln->ln_expire = time_second;
1131 #if 1
1132                         if (ln && ln->ln_expire == 0) {
1133                                 /* kludge for desktops */
1134 #if 0
1135                                 printf("nd6_rtequest: time.tv_sec is zero; "
1136                                        "treat it as 1\n");
1137 #endif
1138                                 ln->ln_expire = 1;
1139                         }
1140 #endif
1141                         if ((rt->rt_flags & RTF_CLONING))
1142                                 break;
1143                 }
1144                 /*
1145                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1146                  * We don't do that here since llinfo is not ready yet.
1147                  *
1148                  * There are also couple of other things to be discussed:
1149                  * - unsolicited NA code needs improvement beforehand
1150                  * - RFC2461 says we MAY send multicast unsolicited NA
1151                  *   (7.2.6 paragraph 4), however, it also says that we
1152                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1153                  *   we don't have anything like it right now.
1154                  *   note that the mechanism needs a mutual agreement
1155                  *   between proxies, which means that we need to implement
1156                  *   a new protocol, or a new kludge.
1157                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1158                  *   we need to check ip6forwarding before sending it.
1159                  *   (or should we allow proxy ND configuration only for
1160                  *   routers?  there's no mention about proxy ND from hosts)
1161                  */
1162 #if 0
1163                 /* XXX it does not work */
1164                 if (rt->rt_flags & RTF_ANNOUNCE)
1165                         nd6_na_output(ifp,
1166                               &SIN6(rt_key(rt))->sin6_addr,
1167                               &SIN6(rt_key(rt))->sin6_addr,
1168                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1169                               1, NULL);
1170 #endif
1171                 /* FALLTHROUGH */
1172         case RTM_RESOLVE:
1173                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1174                         /*
1175                          * Address resolution isn't necessary for a point to
1176                          * point link, so we can skip this test for a p2p link.
1177                          */
1178                         if (gate->sa_family != AF_LINK ||
1179                             gate->sa_len < sizeof(null_sdl)) {
1180                                 log(LOG_DEBUG,
1181                                     "nd6_rtrequest: bad gateway value: %s\n",
1182                                     if_name(ifp));
1183                                 break;
1184                         }
1185                         SDL(gate)->sdl_type = ifp->if_type;
1186                         SDL(gate)->sdl_index = ifp->if_index;
1187                 }
1188                 if (ln != NULL)
1189                         break;  /* This happens on a route change */
1190                 /*
1191                  * Case 2: This route may come from cloning, or a manual route
1192                  * add with a LL address.
1193                  */
1194                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1195                 rt->rt_llinfo = (caddr_t)ln;
1196                 if (!ln) {
1197                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1198                         break;
1199                 }
1200                 nd6_inuse++;
1201                 nd6_allocated++;
1202                 Bzero(ln, sizeof(*ln));
1203                 ln->ln_rt = rt;
1204                 /* this is required for "ndp" command. - shin */
1205                 if (req == RTM_ADD) {
1206                         /*
1207                          * gate should have some valid AF_LINK entry,
1208                          * and ln->ln_expire should have some lifetime
1209                          * which is specified by ndp command.
1210                          */
1211                         ln->ln_state = ND6_LLINFO_REACHABLE;
1212                         ln->ln_byhint = 0;
1213                 } else {
1214                         /*
1215                          * When req == RTM_RESOLVE, rt is created and
1216                          * initialized in rtrequest(), so rt_expire is 0.
1217                          */
1218                         ln->ln_state = ND6_LLINFO_NOSTATE;
1219                         ln->ln_expire = time_second;
1220                 }
1221                 rt->rt_flags |= RTF_LLINFO;
1222                 ln->ln_next = llinfo_nd6.ln_next;
1223                 llinfo_nd6.ln_next = ln;
1224                 ln->ln_prev = &llinfo_nd6;
1225                 ln->ln_next->ln_prev = ln;
1226
1227                 /*
1228                  * check if rt_key(rt) is one of my address assigned
1229                  * to the interface.
1230                  */
1231                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1232                                           &SIN6(rt_key(rt))->sin6_addr);
1233                 if (ifa) {
1234                         caddr_t macp = nd6_ifptomac(ifp);
1235                         ln->ln_expire = 0;
1236                         ln->ln_state = ND6_LLINFO_REACHABLE;
1237                         ln->ln_byhint = 0;
1238                         if (macp) {
1239                                 Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1240                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1241                         }
1242                         if (nd6_useloopback) {
1243                                 rt->rt_ifp = &loif[0];  /* XXX */
1244                                 /*
1245                                  * Make sure rt_ifa be equal to the ifaddr
1246                                  * corresponding to the address.
1247                                  * We need this because when we refer
1248                                  * rt_ifa->ia6_flags in ip6_input, we assume
1249                                  * that the rt_ifa points to the address instead
1250                                  * of the loopback address.
1251                                  */
1252                                 if (ifa != rt->rt_ifa) {
1253                                         IFAFREE(rt->rt_ifa);
1254                                         IFAREF(ifa);
1255                                         rt->rt_ifa = ifa;
1256                                 }
1257                         }
1258                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1259                         ln->ln_expire = 0;
1260                         ln->ln_state = ND6_LLINFO_REACHABLE;
1261                         ln->ln_byhint = 0;
1262
1263                         /* join solicited node multicast for proxy ND */
1264                         if (ifp->if_flags & IFF_MULTICAST) {
1265                                 struct in6_addr llsol;
1266                                 int error;
1267
1268                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1269                                 llsol.s6_addr16[0] = htons(0xff02);
1270                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1271                                 llsol.s6_addr32[1] = 0;
1272                                 llsol.s6_addr32[2] = htonl(1);
1273                                 llsol.s6_addr8[12] = 0xff;
1274
1275                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1276                                         nd6log((LOG_ERR, "%s: failed to join "
1277                                             "%s (errno=%d)\n", if_name(ifp),
1278                                             ip6_sprintf(&llsol), error));
1279                                 }
1280                         }
1281                 }
1282                 break;
1283
1284         case RTM_DELETE:
1285                 if (!ln)
1286                         break;
1287                 /* leave from solicited node multicast for proxy ND */
1288                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1289                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1290                         struct in6_addr llsol;
1291                         struct in6_multi *in6m;
1292
1293                         llsol = SIN6(rt_key(rt))->sin6_addr;
1294                         llsol.s6_addr16[0] = htons(0xff02);
1295                         llsol.s6_addr16[1] = htons(ifp->if_index);
1296                         llsol.s6_addr32[1] = 0;
1297                         llsol.s6_addr32[2] = htonl(1);
1298                         llsol.s6_addr8[12] = 0xff;
1299
1300                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1301                         if (in6m)
1302                                 in6_delmulti(in6m);
1303                 }
1304                 nd6_inuse--;
1305                 ln->ln_next->ln_prev = ln->ln_prev;
1306                 ln->ln_prev->ln_next = ln->ln_next;
1307                 ln->ln_prev = NULL;
1308                 rt->rt_llinfo = 0;
1309                 rt->rt_flags &= ~RTF_LLINFO;
1310                 if (ln->ln_hold)
1311                         m_freem(ln->ln_hold);
1312                 Free((caddr_t)ln);
1313         }
1314 }
1315
1316 int
1317 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1318 {
1319         struct in6_drlist *drl = (struct in6_drlist *)data;
1320         struct in6_prlist *prl = (struct in6_prlist *)data;
1321         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1322         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1323         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1324         struct nd_defrouter *dr, any;
1325         struct nd_prefix *pr;
1326         struct rtentry *rt;
1327         int i = 0, error = 0;
1328         int s;
1329
1330         switch (cmd) {
1331         case SIOCGDRLST_IN6:
1332                 /*
1333                  * obsolete API, use sysctl under net.inet6.icmp6
1334                  */
1335                 bzero(drl, sizeof(*drl));
1336                 s = splnet();
1337                 dr = TAILQ_FIRST(&nd_defrouter);
1338                 while (dr && i < DRLSTSIZ) {
1339                         drl->defrouter[i].rtaddr = dr->rtaddr;
1340                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1341                                 /* XXX: need to this hack for KAME stack */
1342                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1343                         } else
1344                                 log(LOG_ERR,
1345                                     "default router list contains a "
1346                                     "non-linklocal address(%s)\n",
1347                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1348
1349                         drl->defrouter[i].flags = dr->flags;
1350                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1351                         drl->defrouter[i].expire = dr->expire;
1352                         drl->defrouter[i].if_index = dr->ifp->if_index;
1353                         i++;
1354                         dr = TAILQ_NEXT(dr, dr_entry);
1355                 }
1356                 splx(s);
1357                 break;
1358         case SIOCGPRLST_IN6:
1359                 /*
1360                  * obsolete API, use sysctl under net.inet6.icmp6
1361                  */
1362                 /*
1363                  * XXX meaning of fields, especialy "raflags", is very
1364                  * differnet between RA prefix list and RR/static prefix list.
1365                  * how about separating ioctls into two?
1366                  */
1367                 bzero(prl, sizeof(*prl));
1368                 s = splnet();
1369                 pr = nd_prefix.lh_first;
1370                 while (pr && i < PRLSTSIZ) {
1371                         struct nd_pfxrouter *pfr;
1372                         int j;
1373
1374                         (void)in6_embedscope(&prl->prefix[i].prefix,
1375                             &pr->ndpr_prefix, NULL, NULL);
1376                         prl->prefix[i].raflags = pr->ndpr_raf;
1377                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1378                         prl->prefix[i].vltime = pr->ndpr_vltime;
1379                         prl->prefix[i].pltime = pr->ndpr_pltime;
1380                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1381                         prl->prefix[i].expire = pr->ndpr_expire;
1382
1383                         pfr = pr->ndpr_advrtrs.lh_first;
1384                         j = 0;
1385                         while (pfr) {
1386                                 if (j < DRLSTSIZ) {
1387 #define RTRADDR prl->prefix[i].advrtr[j]
1388                                         RTRADDR = pfr->router->rtaddr;
1389                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1390                                                 /* XXX: hack for KAME */
1391                                                 RTRADDR.s6_addr16[1] = 0;
1392                                         } else
1393                                                 log(LOG_ERR,
1394                                                     "a router(%s) advertises "
1395                                                     "a prefix with "
1396                                                     "non-link local address\n",
1397                                                     ip6_sprintf(&RTRADDR));
1398 #undef RTRADDR
1399                                 }
1400                                 j++;
1401                                 pfr = pfr->pfr_next;
1402                         }
1403                         prl->prefix[i].advrtrs = j;
1404                         prl->prefix[i].origin = PR_ORIG_RA;
1405
1406                         i++;
1407                         pr = pr->ndpr_next;
1408                 }
1409               {
1410                 struct rr_prefix *rpp;
1411
1412                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1413                      rpp = LIST_NEXT(rpp, rp_entry)) {
1414                         if (i >= PRLSTSIZ)
1415                                 break;
1416                         (void)in6_embedscope(&prl->prefix[i].prefix,
1417                             &pr->ndpr_prefix, NULL, NULL);
1418                         prl->prefix[i].raflags = rpp->rp_raf;
1419                         prl->prefix[i].prefixlen = rpp->rp_plen;
1420                         prl->prefix[i].vltime = rpp->rp_vltime;
1421                         prl->prefix[i].pltime = rpp->rp_pltime;
1422                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1423                         prl->prefix[i].expire = rpp->rp_expire;
1424                         prl->prefix[i].advrtrs = 0;
1425                         prl->prefix[i].origin = rpp->rp_origin;
1426                         i++;
1427                 }
1428               }
1429                 splx(s);
1430
1431                 break;
1432         case OSIOCGIFINFO_IN6:
1433                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1434                         error = EINVAL;
1435                         break;
1436                 }
1437                 ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1438                 ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1439                 ndi->ndi.basereachable =
1440                     nd_ifinfo[ifp->if_index].basereachable;
1441                 ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1442                 ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1443                 ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1444                 ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1445                 ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1446                 ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1447                 break;
1448         case SIOCGIFINFO_IN6:
1449                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1450                         error = EINVAL;
1451                         break;
1452                 }
1453                 ndi->ndi = nd_ifinfo[ifp->if_index];
1454                 break;
1455         case SIOCSIFINFO_FLAGS:
1456                 /* XXX: almost all other fields of ndi->ndi is unused */
1457                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1458                         error = EINVAL;
1459                         break;
1460                 }
1461                 nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1462                 break;
1463         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1464                 /* flush default router list */
1465                 /*
1466                  * xxx sumikawa: should not delete route if default
1467                  * route equals to the top of default router list
1468                  */
1469                 bzero(&any, sizeof(any));
1470                 defrouter_delreq(&any, 0);
1471                 defrouter_select();
1472                 /* xxx sumikawa: flush prefix list */
1473                 break;
1474         case SIOCSPFXFLUSH_IN6:
1475             {
1476                 /* flush all the prefix advertised by routers */
1477                 struct nd_prefix *pr, *next;
1478
1479                 s = splnet();
1480                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1481                         struct in6_ifaddr *ia, *ia_next;
1482
1483                         next = pr->ndpr_next;
1484
1485                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1486                                 continue; /* XXX */
1487
1488                         /* do we really have to remove addresses as well? */
1489                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1490                                 /* ia might be removed.  keep the next ptr. */
1491                                 ia_next = ia->ia_next;
1492
1493                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1494                                         continue;
1495
1496                                 if (ia->ia6_ndpr == pr)
1497                                         in6_purgeaddr(&ia->ia_ifa);
1498                         }
1499                         prelist_remove(pr);
1500                 }
1501                 splx(s);
1502                 break;
1503             }
1504         case SIOCSRTRFLUSH_IN6:
1505             {
1506                 /* flush all the default routers */
1507                 struct nd_defrouter *dr, *next;
1508
1509                 s = splnet();
1510                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1511                         /*
1512                          * The first entry of the list may be stored in
1513                          * the routing table, so we'll delete it later.
1514                          */
1515                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1516                                 next = TAILQ_NEXT(dr, dr_entry);
1517                                 defrtrlist_del(dr);
1518                         }
1519                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1520                 }
1521                 splx(s);
1522                 break;
1523             }
1524         case SIOCGNBRINFO_IN6:
1525             {
1526                 struct llinfo_nd6 *ln;
1527                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1528
1529                 /*
1530                  * XXX: KAME specific hack for scoped addresses
1531                  *      XXXX: for other scopes than link-local?
1532                  */
1533                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1534                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1535                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1536
1537                         if (*idp == 0)
1538                                 *idp = htons(ifp->if_index);
1539                 }
1540
1541                 s = splnet();
1542                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1543                         error = EINVAL;
1544                         splx(s);
1545                         break;
1546                 }
1547                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1548                 nbi->state = ln->ln_state;
1549                 nbi->asked = ln->ln_asked;
1550                 nbi->isrouter = ln->ln_router;
1551                 nbi->expire = ln->ln_expire;
1552                 splx(s);
1553
1554                 break;
1555             }
1556         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1557                 ndif->ifindex = nd6_defifindex;
1558                 break;
1559         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1560                 return(nd6_setdefaultiface(ndif->ifindex));
1561                 break;
1562         }
1563         return(error);
1564 }
1565
1566 /*
1567  * Create neighbor cache entry and cache link-layer address,
1568  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1569  */
1570 struct rtentry *
1571 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1572                  int lladdrlen,
1573                  int type,      /* ICMP6 type */
1574                  int code       /* type dependent information */)
1575 {
1576         struct rtentry *rt = NULL;
1577         struct llinfo_nd6 *ln = NULL;
1578         int is_newentry;
1579         struct sockaddr_dl *sdl = NULL;
1580         int do_update;
1581         int olladdr;
1582         int llchange;
1583         int newstate = 0;
1584
1585         if (!ifp)
1586                 panic("ifp == NULL in nd6_cache_lladdr");
1587         if (!from)
1588                 panic("from == NULL in nd6_cache_lladdr");
1589
1590         /* nothing must be updated for unspecified address */
1591         if (IN6_IS_ADDR_UNSPECIFIED(from))
1592                 return NULL;
1593
1594         /*
1595          * Validation about ifp->if_addrlen and lladdrlen must be done in
1596          * the caller.
1597          *
1598          * XXX If the link does not have link-layer adderss, what should
1599          * we do? (ifp->if_addrlen == 0)
1600          * Spec says nothing in sections for RA, RS and NA.  There's small
1601          * description on it in NS section (RFC 2461 7.2.3).
1602          */
1603
1604         rt = nd6_lookup(from, 0, ifp);
1605         if (!rt) {
1606 #if 0
1607                 /* nothing must be done if there's no lladdr */
1608                 if (!lladdr || !lladdrlen)
1609                         return NULL;
1610 #endif
1611
1612                 rt = nd6_lookup(from, 1, ifp);
1613                 is_newentry = 1;
1614         } else {
1615                 /* do nothing if static ndp is set */
1616                 if (rt->rt_flags & RTF_STATIC)
1617                         return NULL;
1618                 is_newentry = 0;
1619         }
1620
1621         if (!rt)
1622                 return NULL;
1623         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1624 fail:
1625                 (void)nd6_free(rt);
1626                 return NULL;
1627         }
1628         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1629         if (!ln)
1630                 goto fail;
1631         if (!rt->rt_gateway)
1632                 goto fail;
1633         if (rt->rt_gateway->sa_family != AF_LINK)
1634                 goto fail;
1635         sdl = SDL(rt->rt_gateway);
1636
1637         olladdr = (sdl->sdl_alen) ? 1 : 0;
1638         if (olladdr && lladdr) {
1639                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1640                         llchange = 1;
1641                 else
1642                         llchange = 0;
1643         } else
1644                 llchange = 0;
1645
1646         /*
1647          * newentry olladdr  lladdr  llchange   (*=record)
1648          *      0       n       n       --      (1)
1649          *      0       y       n       --      (2)
1650          *      0       n       y       --      (3) * STALE
1651          *      0       y       y       n       (4) *
1652          *      0       y       y       y       (5) * STALE
1653          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1654          *      1       --      y       --      (7) * STALE
1655          */
1656
1657         if (lladdr) {           /* (3-5) and (7) */
1658                 /*
1659                  * Record source link-layer address
1660                  * XXX is it dependent to ifp->if_type?
1661                  */
1662                 sdl->sdl_alen = ifp->if_addrlen;
1663                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1664         }
1665
1666         if (!is_newentry) {
1667                 if ((!olladdr && lladdr)                /* (3) */
1668                  || (olladdr && lladdr && llchange)) {  /* (5) */
1669                         do_update = 1;
1670                         newstate = ND6_LLINFO_STALE;
1671                 } else                                  /* (1-2,4) */
1672                         do_update = 0;
1673         } else {
1674                 do_update = 1;
1675                 if (!lladdr)                            /* (6) */
1676                         newstate = ND6_LLINFO_NOSTATE;
1677                 else                                    /* (7) */
1678                         newstate = ND6_LLINFO_STALE;
1679         }
1680
1681         if (do_update) {
1682                 /*
1683                  * Update the state of the neighbor cache.
1684                  */
1685                 ln->ln_state = newstate;
1686
1687                 if (ln->ln_state == ND6_LLINFO_STALE) {
1688                         /*
1689                          * XXX: since nd6_output() below will cause
1690                          * state tansition to DELAY and reset the timer,
1691                          * we must set the timer now, although it is actually
1692                          * meaningless.
1693                          */
1694                         ln->ln_expire = time_second + nd6_gctimer;
1695
1696                         if (ln->ln_hold) {
1697                                 /*
1698                                  * we assume ifp is not a p2p here, so just
1699                                  * set the 2nd argument as the 1st one.
1700                                  */
1701                                 nd6_output(ifp, ifp, ln->ln_hold,
1702                                            (struct sockaddr_in6 *)rt_key(rt),
1703                                            rt);
1704                                 ln->ln_hold = NULL;
1705                         }
1706                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1707                         /* probe right away */
1708                         ln->ln_expire = time_second;
1709                 }
1710         }
1711
1712         /*
1713          * ICMP6 type dependent behavior.
1714          *
1715          * NS: clear IsRouter if new entry
1716          * RS: clear IsRouter
1717          * RA: set IsRouter if there's lladdr
1718          * redir: clear IsRouter if new entry
1719          *
1720          * RA case, (1):
1721          * The spec says that we must set IsRouter in the following cases:
1722          * - If lladdr exist, set IsRouter.  This means (1-5).
1723          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1724          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1725          * A quetion arises for (1) case.  (1) case has no lladdr in the
1726          * neighbor cache, this is similar to (6).
1727          * This case is rare but we figured that we MUST NOT set IsRouter.
1728          *
1729          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1730          *                                                      D R
1731          *      0       n       n       --      (1)     c   ?     s
1732          *      0       y       n       --      (2)     c   s     s
1733          *      0       n       y       --      (3)     c   s     s
1734          *      0       y       y       n       (4)     c   s     s
1735          *      0       y       y       y       (5)     c   s     s
1736          *      1       --      n       --      (6) c   c       c s
1737          *      1       --      y       --      (7) c   c   s   c s
1738          *
1739          *                                      (c=clear s=set)
1740          */
1741         switch (type & 0xff) {
1742         case ND_NEIGHBOR_SOLICIT:
1743                 /*
1744                  * New entry must have is_router flag cleared.
1745                  */
1746                 if (is_newentry)        /* (6-7) */
1747                         ln->ln_router = 0;
1748                 break;
1749         case ND_REDIRECT:
1750                 /*
1751                  * If the icmp is a redirect to a better router, always set the
1752                  * is_router flag. Otherwise, if the entry is newly created,
1753                  * clear the flag. [RFC 2461, sec 8.3]
1754                  */
1755                 if (code == ND_REDIRECT_ROUTER)
1756                         ln->ln_router = 1;
1757                 else if (is_newentry) /* (6-7) */
1758                         ln->ln_router = 0;
1759                 break;
1760         case ND_ROUTER_SOLICIT:
1761                 /*
1762                  * is_router flag must always be cleared.
1763                  */
1764                 ln->ln_router = 0;
1765                 break;
1766         case ND_ROUTER_ADVERT:
1767                 /*
1768                  * Mark an entry with lladdr as a router.
1769                  */
1770                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1771                  || (is_newentry && lladdr)) {                  /* (7) */
1772                         ln->ln_router = 1;
1773                 }
1774                 break;
1775         }
1776
1777         /*
1778          * When the link-layer address of a router changes, select the
1779          * best router again.  In particular, when the neighbor entry is newly
1780          * created, it might affect the selection policy.
1781          * Question: can we restrict the first condition to the "is_newentry"
1782          * case?
1783          * XXX: when we hear an RA from a new router with the link-layer
1784          * address option, defrouter_select() is called twice, since
1785          * defrtrlist_update called the function as well.  However, I believe
1786          * we can compromise the overhead, since it only happens the first
1787          * time.
1788          * XXX: although defrouter_select() should not have a bad effect
1789          * for those are not autoconfigured hosts, we explicitly avoid such
1790          * cases for safety.
1791          */
1792         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1793                 defrouter_select();
1794
1795         return rt;
1796 }
1797
1798 static void
1799 nd6_slowtimo(void *ignored_arg)
1800 {
1801         int s = splnet();
1802         int i;
1803         struct nd_ifinfo *nd6if;
1804
1805         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1806             nd6_slowtimo, NULL);
1807         for (i = 1; i < if_index + 1; i++) {
1808                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1809                         continue;
1810                 nd6if = &nd_ifinfo[i];
1811                 if (nd6if->basereachable && /* already initialized */
1812                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1813                         /*
1814                          * Since reachable time rarely changes by router
1815                          * advertisements, we SHOULD insure that a new random
1816                          * value gets recomputed at least once every few hours.
1817                          * (RFC 2461, 6.3.4)
1818                          */
1819                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1820                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1821                 }
1822         }
1823         splx(s);
1824 }
1825
1826 #define senderr(e) { error = (e); goto bad;}
1827 int
1828 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1829            struct sockaddr_in6 *dst, struct rtentry *rt0)
1830 {
1831         struct mbuf *m = m0;
1832         struct rtentry *rt = rt0;
1833         struct sockaddr_in6 *gw6 = NULL;
1834         struct llinfo_nd6 *ln = NULL;
1835         int error = 0;
1836
1837         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1838                 goto sendpkt;
1839
1840         if (nd6_need_cache(ifp) == 0)
1841                 goto sendpkt;
1842
1843         /*
1844          * next hop determination.  This routine is derived from ether_outpout.
1845          */
1846         if (rt) {
1847                 if ((rt->rt_flags & RTF_UP) == 0) {
1848                         if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL)) !=
1849                                 NULL)
1850                         {
1851                                 rt->rt_refcnt--;
1852                                 if (rt->rt_ifp != ifp) {
1853                                         /* XXX: loop care? */
1854                                         return nd6_output(ifp, origifp, m0,
1855                                                           dst, rt);
1856                                 }
1857                         } else
1858                                 senderr(EHOSTUNREACH);
1859                 }
1860
1861                 if (rt->rt_flags & RTF_GATEWAY) {
1862                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1863
1864                         /*
1865                          * We skip link-layer address resolution and NUD
1866                          * if the gateway is not a neighbor from ND point
1867                          * of view, regardless of the value of nd_ifinfo.flags.
1868                          * The second condition is a bit tricky; we skip
1869                          * if the gateway is our own address, which is
1870                          * sometimes used to install a route to a p2p link.
1871                          */
1872                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1873                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1874                                 /*
1875                                  * We allow this kind of tricky route only
1876                                  * when the outgoing interface is p2p.
1877                                  * XXX: we may need a more generic rule here.
1878                                  */
1879                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1880                                         senderr(EHOSTUNREACH);
1881
1882                                 goto sendpkt;
1883                         }
1884
1885                         if (rt->rt_gwroute == 0)
1886                                 goto lookup;
1887                         if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1888                                 rtfree(rt); rt = rt0;
1889                         lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1890                                 if ((rt = rt->rt_gwroute) == 0)
1891                                         senderr(EHOSTUNREACH);
1892                         }
1893                 }
1894         }
1895
1896         /*
1897          * Address resolution or Neighbor Unreachability Detection
1898          * for the next hop.
1899          * At this point, the destination of the packet must be a unicast
1900          * or an anycast address(i.e. not a multicast).
1901          */
1902
1903         /* Look up the neighbor cache for the nexthop */
1904         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1905                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1906         else {
1907                 /*
1908                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1909                  * the condition below is not very efficient.  But we believe
1910                  * it is tolerable, because this should be a rare case.
1911                  */
1912                 if (nd6_is_addr_neighbor(dst, ifp) &&
1913                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1914                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1915         }
1916         if (!ln || !rt) {
1917                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1918                     !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1919                         log(LOG_DEBUG,
1920                             "nd6_output: can't allocate llinfo for %s "
1921                             "(ln=%p, rt=%p)\n",
1922                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1923                         senderr(EIO);   /* XXX: good error? */
1924                 }
1925
1926                 goto sendpkt;   /* send anyway */
1927         }
1928
1929         /* We don't have to do link-layer address resolution on a p2p link. */
1930         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1931             ln->ln_state < ND6_LLINFO_REACHABLE) {
1932                 ln->ln_state = ND6_LLINFO_STALE;
1933                 ln->ln_expire = time_second + nd6_gctimer;
1934         }
1935
1936         /*
1937          * The first time we send a packet to a neighbor whose entry is
1938          * STALE, we have to change the state to DELAY and a sets a timer to
1939          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1940          * neighbor unreachability detection on expiration.
1941          * (RFC 2461 7.3.3)
1942          */
1943         if (ln->ln_state == ND6_LLINFO_STALE) {
1944                 ln->ln_asked = 0;
1945                 ln->ln_state = ND6_LLINFO_DELAY;
1946                 ln->ln_expire = time_second + nd6_delay;
1947         }
1948
1949         /*
1950          * If the neighbor cache entry has a state other than INCOMPLETE
1951          * (i.e. its link-layer address is already resolved), just
1952          * send the packet.
1953          */
1954         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1955                 goto sendpkt;
1956
1957         /*
1958          * There is a neighbor cache entry, but no ethernet address
1959          * response yet.  Replace the held mbuf (if any) with this
1960          * latest one.
1961          *
1962          * This code conforms to the rate-limiting rule described in Section
1963          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1964          * an NS below.
1965          */
1966         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1967                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1968         if (ln->ln_hold)
1969                 m_freem(ln->ln_hold);
1970         ln->ln_hold = m;
1971         if (ln->ln_expire) {
1972                 if (ln->ln_asked < nd6_mmaxtries &&
1973                     ln->ln_expire < time_second) {
1974                         ln->ln_asked++;
1975                         ln->ln_expire = time_second +
1976                                 nd_ifinfo[ifp->if_index].retrans / 1000;
1977                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1978                 }
1979         }
1980         return(0);
1981
1982   sendpkt:
1983
1984         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1985                 return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1986                                          rt));
1987         }
1988         return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1989
1990   bad:
1991         if (m)
1992                 m_freem(m);
1993         return (error);
1994 }
1995 #undef senderr
1996
1997 int
1998 nd6_need_cache(struct ifnet *ifp)
1999 {
2000         /*
2001          * XXX: we currently do not make neighbor cache on any interface
2002          * other than ARCnet, Ethernet, FDDI and GIF.
2003          *
2004          * RFC2893 says:
2005          * - unidirectional tunnels needs no ND
2006          */
2007         switch (ifp->if_type) {
2008         case IFT_ARCNET:
2009         case IFT_ETHER:
2010         case IFT_FDDI:
2011         case IFT_IEEE1394:
2012 #ifdef IFT_L2VLAN
2013         case IFT_L2VLAN:
2014 #endif
2015 #ifdef IFT_IEEE80211
2016         case IFT_IEEE80211:
2017 #endif
2018         case IFT_GIF:           /* XXX need more cases? */
2019                 return(1);
2020         default:
2021                 return(0);
2022         }
2023 }
2024
2025 int
2026 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt, struct mbuf *m,
2027                 struct sockaddr *dst, u_char *desten)
2028 {
2029         struct sockaddr_dl *sdl;
2030
2031         if (m->m_flags & M_MCAST) {
2032                 switch (ifp->if_type) {
2033                 case IFT_ETHER:
2034                 case IFT_FDDI:
2035 #ifdef IFT_L2VLAN
2036         case IFT_L2VLAN:
2037 #endif
2038 #ifdef IFT_IEEE80211
2039                 case IFT_IEEE80211:
2040 #endif
2041                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2042                                                  desten);
2043                         return(1);
2044                 case IFT_IEEE1394:
2045                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2046                         return(1);
2047                 case IFT_ARCNET:
2048                         *desten = 0;
2049                         return(1);
2050                 default:
2051                         m_freem(m);
2052                         return(0);
2053                 }
2054         }
2055
2056         if (rt == NULL) {
2057                 /* this could happen, if we could not allocate memory */
2058                 m_freem(m);
2059                 return(0);
2060         }
2061         if (rt->rt_gateway->sa_family != AF_LINK) {
2062                 printf("nd6_storelladdr: something odd happens\n");
2063                 m_freem(m);
2064                 return(0);
2065         }
2066         sdl = SDL(rt->rt_gateway);
2067         if (sdl->sdl_alen == 0) {
2068                 /* this should be impossible, but we bark here for debugging */
2069                 printf("nd6_storelladdr: sdl_alen == 0\n");
2070                 m_freem(m);
2071                 return(0);
2072         }
2073
2074         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2075         return(1);
2076 }
2077
2078 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2079 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2080 #ifdef SYSCTL_DECL
2081 SYSCTL_DECL(_net_inet6_icmp6);
2082 #endif
2083 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2084         CTLFLAG_RD, nd6_sysctl_drlist, "");
2085 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2086         CTLFLAG_RD, nd6_sysctl_prlist, "");
2087
2088 static int
2089 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2090 {
2091         int error;
2092         char buf[1024];
2093         struct in6_defrouter *d, *de;
2094         struct nd_defrouter *dr;
2095
2096         if (req->newptr)
2097                 return EPERM;
2098         error = 0;
2099
2100         for (dr = TAILQ_FIRST(&nd_defrouter);
2101              dr;
2102              dr = TAILQ_NEXT(dr, dr_entry)) {
2103                 d = (struct in6_defrouter *)buf;
2104                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2105
2106                 if (d + 1 <= de) {
2107                         bzero(d, sizeof(*d));
2108                         d->rtaddr.sin6_family = AF_INET6;
2109                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2110                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2111                             dr->ifp) != 0)
2112                                 log(LOG_ERR,
2113                                     "scope error in "
2114                                     "default router list (%s)\n",
2115                                     ip6_sprintf(&dr->rtaddr));
2116                         d->flags = dr->flags;
2117                         d->rtlifetime = dr->rtlifetime;
2118                         d->expire = dr->expire;
2119                         d->if_index = dr->ifp->if_index;
2120                 } else
2121                         panic("buffer too short");
2122
2123                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2124                 if (error)
2125                         break;
2126         }
2127         return error;
2128 }
2129
2130 static int
2131 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2132 {
2133         int error;
2134         char buf[1024];
2135         struct in6_prefix *p, *pe;
2136         struct nd_prefix *pr;
2137
2138         if (req->newptr)
2139                 return EPERM;
2140         error = 0;
2141
2142         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2143                 u_short advrtrs;
2144                 size_t advance;
2145                 struct sockaddr_in6 *sin6, *s6;
2146                 struct nd_pfxrouter *pfr;
2147
2148                 p = (struct in6_prefix *)buf;
2149                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2150
2151                 if (p + 1 <= pe) {
2152                         bzero(p, sizeof(*p));
2153                         sin6 = (struct sockaddr_in6 *)(p + 1);
2154
2155                         p->prefix = pr->ndpr_prefix;
2156                         if (in6_recoverscope(&p->prefix,
2157                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2158                                 log(LOG_ERR,
2159                                     "scope error in prefix list (%s)\n",
2160                                     ip6_sprintf(&p->prefix.sin6_addr));
2161                         p->raflags = pr->ndpr_raf;
2162                         p->prefixlen = pr->ndpr_plen;
2163                         p->vltime = pr->ndpr_vltime;
2164                         p->pltime = pr->ndpr_pltime;
2165                         p->if_index = pr->ndpr_ifp->if_index;
2166                         p->expire = pr->ndpr_expire;
2167                         p->refcnt = pr->ndpr_refcnt;
2168                         p->flags = pr->ndpr_stateflags;
2169                         p->origin = PR_ORIG_RA;
2170                         advrtrs = 0;
2171                         for (pfr = pr->ndpr_advrtrs.lh_first;
2172                              pfr;
2173                              pfr = pfr->pfr_next) {
2174                                 if ((void *)&sin6[advrtrs + 1] >
2175                                     (void *)pe) {
2176                                         advrtrs++;
2177                                         continue;
2178                                 }
2179                                 s6 = &sin6[advrtrs];
2180                                 bzero(s6, sizeof(*s6));
2181                                 s6->sin6_family = AF_INET6;
2182                                 s6->sin6_len = sizeof(*sin6);
2183                                 if (in6_recoverscope(s6,
2184                                     &pfr->router->rtaddr,
2185                                     pfr->router->ifp) != 0)
2186                                         log(LOG_ERR,
2187                                             "scope error in "
2188                                             "prefix list (%s)\n",
2189                                             ip6_sprintf(&pfr->router->rtaddr));
2190                                 advrtrs++;
2191                         }
2192                         p->advrtrs = advrtrs;
2193                 } else
2194                         panic("buffer too short");
2195
2196                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2197                 error = SYSCTL_OUT(req, buf, advance);
2198                 if (error)
2199                         break;
2200         }
2201         return error;
2202 }