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 1.7 2004/07/17 09:43:06 joerg 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         for (ifa = ifp->if_addrlist.tqh_first; ifa;
 649              ifa = ifa->ifa_list.tqe_next)
 650         {
 651                 struct in6_ifaddr *it6;
 652
 653                 if (ifa->ifa_addr->sa_family != AF_INET6)
 654                         continue;
 655
 656                 it6 = (struct in6_ifaddr *)ifa;
 657
 658                 /* ignore no autoconf addresses. */
 659                 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
 660                         continue;
 661
 662                 /* ignore autoconf addresses with different prefixes. */
 663                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 664                         continue;
 665
 666                 /*
 667                  * Now we are looking at an autoconf address with the same
 668                  * prefix as ours.  If the address is temporary and is still
 669                  * preferred, do not create another one.  It would be rare, but
 670                  * could happen, for example, when we resume a laptop PC after
 671                  * a long period.
 672                  */
 673                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 674                     !IFA6_IS_DEPRECATED(it6)) {
 675                         public_ifa6 = NULL;
 676                         break;
 677                 }
 678
 679                 /*
 680                  * This is a public autoconf address that has the same prefix
 681                  * as ours.  If it is preferred, keep it.  We can't break the
 682                  * loop here, because there may be a still-preferred temporary
 683                  * address with the prefix.
 684                  */
 685                 if (!IFA6_IS_DEPRECATED(it6))
 686                     public_ifa6 = it6;
 687         }
 688
 689         if (public_ifa6 != NULL) {
 690                 int e;
 691
 692                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 693                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 694                             " tmp addr,errno=%d\n", e);
 695                         return(-1);
 696                 }
 697                 return(0);
 698         }
 699
 700         return(-1);
 701 }
 702
 703 /*
 704  * Nuke neighbor cache/prefix/default router management table, right before
 705  * ifp goes away.
 706  */
 707 void
 708 nd6_purge(struct ifnet *ifp)
 709 {
 710         struct llinfo_nd6 *ln, *nln;
 711         struct nd_defrouter *dr, *ndr, drany;
 712         struct nd_prefix *pr, *npr;
 713
 714         /* Nuke default router list entries toward ifp */
 715         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 716                 /*
 717                  * The first entry of the list may be stored in
 718                  * the routing table, so we'll delete it later.
 719                  */
 720                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 721                         ndr = TAILQ_NEXT(dr, dr_entry);
 722                         if (dr->ifp == ifp)
 723                                 defrtrlist_del(dr);
 724                 }
 725                 dr = TAILQ_FIRST(&nd_defrouter);
 726                 if (dr->ifp == ifp)
 727                         defrtrlist_del(dr);
 728         }
 729
 730         /* Nuke prefix list entries toward ifp */
 731         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 732                 npr = pr->ndpr_next;
 733                 if (pr->ndpr_ifp == ifp) {
 734                         /*
 735                          * Previously, pr->ndpr_addr is removed as well,
 736                          * but I strongly believe we don't have to do it.
 737                          * nd6_purge() is only called from in6_ifdetach(),
 738                          * which removes all the associated interface addresses
 739                          * by itself.
 740                          * (jinmei@kame.net 20010129)
 741                          */
 742                         prelist_remove(pr);
 743                 }
 744         }
 745
 746         /* cancel default outgoing interface setting */
 747         if (nd6_defifindex == ifp->if_index)
 748                 nd6_setdefaultiface(0);
 749
 750         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 751                 /* refresh default router list */
 752                 bzero(&drany, sizeof(drany));
 753                 defrouter_delreq(&drany, 0);
 754                 defrouter_select();
 755         }
 756
 757         /*
 758          * Nuke neighbor cache entries for the ifp.
 759          * Note that rt->rt_ifp may not be the same as ifp,
 760          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 761          * nd6_rtrequest(), and ip6_input().
 762          */
 763         ln = llinfo_nd6.ln_next;
 764         while (ln && ln != &llinfo_nd6) {
 765                 struct rtentry *rt;
 766                 struct sockaddr_dl *sdl;
 767
 768                 nln = ln->ln_next;
 769                 rt = ln->ln_rt;
 770                 if (rt && rt->rt_gateway &&
 771                     rt->rt_gateway->sa_family == AF_LINK) {
 772                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 773                         if (sdl->sdl_index == ifp->if_index)
 774                                 nln = nd6_free(rt);
 775                 }
 776                 ln = nln;
 777         }
 778 }
 779
 780 struct rtentry *
 781 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 782 {
 783         struct rtentry *rt;
 784         struct sockaddr_in6 sin6;
 785
 786         bzero(&sin6, sizeof(sin6));
 787         sin6.sin6_len = sizeof(struct sockaddr_in6);
 788         sin6.sin6_family = AF_INET6;
 789         sin6.sin6_addr = *addr6;
 790 #ifdef SCOPEDROUTING
 791         sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
 792 #endif
 793         rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
 794         if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
 795                 /*
 796                  * This is the case for the default route.
 797                  * If we want to create a neighbor cache for the address, we
 798                  * should free the route for the destination and allocate an
 799                  * interface route.
 800                  */
 801                 if (create) {
 802                         RTFREE(rt);
 803                         rt = 0;
 804                 }
 805         }
 806         if (!rt) {
 807                 if (create && ifp) {
 808                         int e;
 809
 810                         /*
 811                          * If no route is available and create is set,
 812                          * we allocate a host route for the destination
 813                          * and treat it like an interface route.
 814                          * This hack is necessary for a neighbor which can't
 815                          * be covered by our own prefix.
 816                          */
 817                         struct ifaddr *ifa =
 818                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 819                         if (ifa == NULL)
 820                                 return(NULL);
 821
 822                         /*
 823                          * Create a new route.  RTF_LLINFO is necessary
 824                          * to create a Neighbor Cache entry for the
 825                          * destination in nd6_rtrequest which will be
 826                          * called in rtrequest via ifa->ifa_rtrequest.
 827                          */
 828                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 829                                            ifa->ifa_addr,
 830                                            (struct sockaddr *)&all1_sa,
 831                                            (ifa->ifa_flags |
 832                                             RTF_HOST | RTF_LLINFO) &
 833                                            ~RTF_CLONING,
 834                                            &rt)) != 0)
 835                                 log(LOG_ERR,
 836                                     "nd6_lookup: failed to add route for a "
 837                                     "neighbor(%s), errno=%d\n",
 838                                     ip6_sprintf(addr6), e);
 839                         if (rt == NULL)
 840                                 return(NULL);
 841                         if (rt->rt_llinfo) {
 842                                 struct llinfo_nd6 *ln =
 843                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 844                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 845                         }
 846                 } else
 847                         return(NULL);
 848         }
 849         rt->rt_refcnt--;
 850         /*
 851          * Validation for the entry.
 852          * Note that the check for rt_llinfo is necessary because a cloned
 853          * route from a parent route that has the L flag (e.g. the default
 854          * route to a p2p interface) may have the flag, too, while the
 855          * destination is not actually a neighbor.
 856          * XXX: we can't use rt->rt_ifp to check for the interface, since
 857          *      it might be the loopback interface if the entry is for our
 858          *      own address on a non-loopback interface. Instead, we should
 859          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 860          *      interface.
 861          */
 862         if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
 863             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 864             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 865                 if (create) {
 866                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 867                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 868                         /* xxx more logs... kazu */
 869                 }
 870                 return(NULL);
 871         }
 872         return(rt);
 873 }
 874
 875 /*
 876  * Detect if a given IPv6 address identifies a neighbor on a given link.
 877  * XXX: should take care of the destination of a p2p link?
 878  */
 879 int
 880 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 881 {
 882         struct ifaddr *ifa;
 883         int i;
 884
 885 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 886 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 887
 888         /*
 889          * A link-local address is always a neighbor.
 890          * XXX: we should use the sin6_scope_id field rather than the embedded
 891          * interface index.
 892          */
 893         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 894             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 895                 return(1);
 896
 897         /*
 898          * If the address matches one of our addresses,
 899          * it should be a neighbor.
 900          */
 901         for (ifa = ifp->if_addrlist.tqh_first;
 902              ifa;
 903              ifa = ifa->ifa_list.tqe_next)
 904         {
 905                 if (ifa->ifa_addr->sa_family != AF_INET6)
 906                         next: continue;
 907
 908                 for (i = 0; i < 4; i++) {
 909                         if ((IFADDR6(ifa).s6_addr32[i] ^
 910                              addr->sin6_addr.s6_addr32[i]) &
 911                             IFMASK6(ifa).s6_addr32[i])
 912                                 goto next;
 913                 }
 914                 return(1);
 915         }
 916
 917         /*
 918          * Even if the address matches none of our addresses, it might be
 919          * in the neighbor cache.
 920          */
 921         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 922                 return(1);
 923
 924         return(0);
 925 #undef IFADDR6
 926 #undef IFMASK6
 927 }
 928
 929 /*
 930  * Free an nd6 llinfo entry.
 931  */
 932 struct llinfo_nd6 *
 933 nd6_free(struct rtentry *rt)
 934 {
 935         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 936         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 937         struct nd_defrouter *dr;
 938
 939         /*
 940          * we used to have pfctlinput(PRC_HOSTDEAD) here.
 941          * even though it is not harmful, it was not really necessary.
 942          */
 943
 944         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 945                 int s;
 946                 s = splnet();
 947                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 948                                       rt->rt_ifp);
 949
 950                 if (ln->ln_router || dr) {
 951                         /*
 952                          * rt6_flush must be called whether or not the neighbor
 953                          * is in the Default Router List.
 954                          * See a corresponding comment in nd6_na_input().
 955                          */
 956                         rt6_flush(&in6, rt->rt_ifp);
 957                 }
 958
 959                 if (dr) {
 960                         /*
 961                          * Unreachablity of a router might affect the default
 962                          * router selection and on-link detection of advertised
 963                          * prefixes.
 964                          */
 965
 966                         /*
 967                          * Temporarily fake the state to choose a new default
 968                          * router and to perform on-link determination of
 969                          * prefixes correctly.
 970                          * Below the state will be set correctly,
 971                          * or the entry itself will be deleted.
 972                          */
 973                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 974
 975                         /*
 976                          * Since defrouter_select() does not affect the
 977                          * on-link determination and MIP6 needs the check
 978                          * before the default router selection, we perform
 979                          * the check now.
 980                          */
 981                         pfxlist_onlink_check();
 982
 983                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 984                                 /*
 985                                  * It is used as the current default router,
 986                                  * so we have to move it to the end of the
 987                                  * list and choose a new one.
 988                                  * XXX: it is not very efficient if this is
 989                                  *      the only router.
 990                                  */
 991                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
 992                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
 993
 994                                 defrouter_select();
 995                         }
 996                 }
 997                 splx(s);
 998         }
 999
1000         /*
1001          * Before deleting the entry, remember the next entry as the
1002          * return value.  We need this because pfxlist_onlink_check() above
1003          * might have freed other entries (particularly the old next entry) as
1004          * a side effect (XXX).
1005          */
1006         next = ln->ln_next;
1007
1008         /*
1009          * Detach the route from the routing tree and the list of neighbor
1010          * caches, and disable the route entry not to be used in already
1011          * cached routes.
1012          */
1013         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1014                   rt_mask(rt), 0, (struct rtentry **)0);
1015
1016         return(next);
1017 }
1018
1019 /*
1020  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1021  *
1022  * XXX cost-effective metods?
1023  */
1024 void
1025 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1026 {
1027         struct llinfo_nd6 *ln;
1028
1029         /*
1030          * If the caller specified "rt", use that.  Otherwise, resolve the
1031          * routing table by supplied "dst6".
1032          */
1033         if (!rt) {
1034                 if (!dst6)
1035                         return;
1036                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1037                         return;
1038         }
1039
1040         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1041             (rt->rt_flags & RTF_LLINFO) == 0 ||
1042             !rt->rt_llinfo || !rt->rt_gateway ||
1043             rt->rt_gateway->sa_family != AF_LINK) {
1044                 /* This is not a host route. */
1045                 return;
1046         }
1047
1048         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1049         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1050                 return;
1051
1052         /*
1053          * if we get upper-layer reachability confirmation many times,
1054          * it is possible we have false information.
1055          */
1056         if (!force) {
1057                 ln->ln_byhint++;
1058                 if (ln->ln_byhint > nd6_maxnudhint)
1059                         return;
1060         }
1061
1062         ln->ln_state = ND6_LLINFO_REACHABLE;
1063         if (ln->ln_expire)
1064                 ln->ln_expire = time_second +
1065                         nd_ifinfo[rt->rt_ifp->if_index].reachable;
1066 }
1067
1068 void
1069 nd6_rtrequest(int req, struct rtentry *rt,
1070               struct rt_addrinfo *info) /* xxx unused */
1071 {
1072         struct sockaddr *gate = rt->rt_gateway;
1073         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1074         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1075         struct ifnet *ifp = rt->rt_ifp;
1076         struct ifaddr *ifa;
1077
1078         if ((rt->rt_flags & RTF_GATEWAY))
1079                 return;
1080
1081         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1082                 /*
1083                  * This is probably an interface direct route for a link
1084                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1085                  * We do not need special treatment below for such a route.
1086                  * Moreover, the RTF_LLINFO flag which would be set below
1087                  * would annoy the ndp(8) command.
1088                  */
1089                 return;
1090         }
1091
1092         if (req == RTM_RESOLVE &&
1093             (nd6_need_cache(ifp) == 0 || /* stf case */
1094              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1095                 /*
1096                  * FreeBSD and BSD/OS often make a cloned host route based
1097                  * on a less-specific route (e.g. the default route).
1098                  * If the less specific route does not have a "gateway"
1099                  * (this is the case when the route just goes to a p2p or an
1100                  * stf interface), we'll mistakenly make a neighbor cache for
1101                  * the host route, and will see strange neighbor solicitation
1102                  * for the corresponding destination.  In order to avoid the
1103                  * confusion, we check if the destination of the route is
1104                  * a neighbor in terms of neighbor discovery, and stop the
1105                  * process if not.  Additionally, we remove the LLINFO flag
1106                  * so that ndp(8) will not try to get the neighbor information
1107                  * of the destination.
1108                  */
1109                 rt->rt_flags &= ~RTF_LLINFO;
1110                 return;
1111         }
1112
1113         switch (req) {
1114         case RTM_ADD:
1115                 /*
1116                  * There is no backward compatibility :)
1117                  *
1118                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1119                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1120                  *         rt->rt_flags |= RTF_CLONING;
1121                  */
1122                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1123                         /*
1124                          * Case 1: This route should come from
1125                          * a route to interface.  RTF_LLINFO flag is set
1126                          * for a host route whose destination should be
1127                          * treated as on-link.
1128                          */
1129                         rt_setgate(rt, rt_key(rt),
1130                                    (struct sockaddr *)&null_sdl);
1131                         gate = rt->rt_gateway;
1132                         SDL(gate)->sdl_type = ifp->if_type;
1133                         SDL(gate)->sdl_index = ifp->if_index;
1134                         if (ln)
1135                                 ln->ln_expire = time_second;
1136 #if 1
1137                         if (ln && ln->ln_expire == 0) {
1138                                 /* kludge for desktops */
1139 #if 0
1140                                 printf("nd6_rtequest: time.tv_sec is zero; "
1141                                        "treat it as 1\n");
1142 #endif
1143                                 ln->ln_expire = 1;
1144                         }
1145 #endif
1146                         if ((rt->rt_flags & RTF_CLONING))
1147                                 break;
1148                 }
1149                 /*
1150                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1151                  * We don't do that here since llinfo is not ready yet.
1152                  *
1153                  * There are also couple of other things to be discussed:
1154                  * - unsolicited NA code needs improvement beforehand
1155                  * - RFC2461 says we MAY send multicast unsolicited NA
1156                  *   (7.2.6 paragraph 4), however, it also says that we
1157                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1158                  *   we don't have anything like it right now.
1159                  *   note that the mechanism needs a mutual agreement
1160                  *   between proxies, which means that we need to implement
1161                  *   a new protocol, or a new kludge.
1162                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1163                  *   we need to check ip6forwarding before sending it.
1164                  *   (or should we allow proxy ND configuration only for
1165                  *   routers?  there's no mention about proxy ND from hosts)
1166                  */
1167 #if 0
1168                 /* XXX it does not work */
1169                 if (rt->rt_flags & RTF_ANNOUNCE)
1170                         nd6_na_output(ifp,
1171                               &SIN6(rt_key(rt))->sin6_addr,
1172                               &SIN6(rt_key(rt))->sin6_addr,
1173                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1174                               1, NULL);
1175 #endif
1176                 /* FALLTHROUGH */
1177         case RTM_RESOLVE:
1178                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1179                         /*
1180                          * Address resolution isn't necessary for a point to
1181                          * point link, so we can skip this test for a p2p link.
1182                          */
1183                         if (gate->sa_family != AF_LINK ||
1184                             gate->sa_len < sizeof(null_sdl)) {
1185                                 log(LOG_DEBUG,
1186                                     "nd6_rtrequest: bad gateway value: %s\n",
1187                                     if_name(ifp));
1188                                 break;
1189                         }
1190                         SDL(gate)->sdl_type = ifp->if_type;
1191                         SDL(gate)->sdl_index = ifp->if_index;
1192                 }
1193                 if (ln != NULL)
1194                         break;  /* This happens on a route change */
1195                 /*
1196                  * Case 2: This route may come from cloning, or a manual route
1197                  * add with a LL address.
1198                  */
1199                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1200                 rt->rt_llinfo = (caddr_t)ln;
1201                 if (!ln) {
1202                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1203                         break;
1204                 }
1205                 nd6_inuse++;
1206                 nd6_allocated++;
1207                 Bzero(ln, sizeof(*ln));
1208                 ln->ln_rt = rt;
1209                 /* this is required for "ndp" command. - shin */
1210                 if (req == RTM_ADD) {
1211                         /*
1212                          * gate should have some valid AF_LINK entry,
1213                          * and ln->ln_expire should have some lifetime
1214                          * which is specified by ndp command.
1215                          */
1216                         ln->ln_state = ND6_LLINFO_REACHABLE;
1217                         ln->ln_byhint = 0;
1218                 } else {
1219                         /*
1220                          * When req == RTM_RESOLVE, rt is created and
1221                          * initialized in rtrequest(), so rt_expire is 0.
1222                          */
1223                         ln->ln_state = ND6_LLINFO_NOSTATE;
1224                         ln->ln_expire = time_second;
1225                 }
1226                 rt->rt_flags |= RTF_LLINFO;
1227                 ln->ln_next = llinfo_nd6.ln_next;
1228                 llinfo_nd6.ln_next = ln;
1229                 ln->ln_prev = &llinfo_nd6;
1230                 ln->ln_next->ln_prev = ln;
1231
1232                 /*
1233                  * check if rt_key(rt) is one of my address assigned
1234                  * to the interface.
1235                  */
1236                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1237                                           &SIN6(rt_key(rt))->sin6_addr);
1238                 if (ifa) {
1239                         caddr_t macp = nd6_ifptomac(ifp);
1240                         ln->ln_expire = 0;
1241                         ln->ln_state = ND6_LLINFO_REACHABLE;
1242                         ln->ln_byhint = 0;
1243                         if (macp) {
1244                                 Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1245                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1246                         }
1247                         if (nd6_useloopback) {
1248                                 rt->rt_ifp = &loif[0];  /* XXX */
1249                                 /*
1250                                  * Make sure rt_ifa be equal to the ifaddr
1251                                  * corresponding to the address.
1252                                  * We need this because when we refer
1253                                  * rt_ifa->ia6_flags in ip6_input, we assume
1254                                  * that the rt_ifa points to the address instead
1255                                  * of the loopback address.
1256                                  */
1257                                 if (ifa != rt->rt_ifa) {
1258                                         IFAFREE(rt->rt_ifa);
1259                                         IFAREF(ifa);
1260                                         rt->rt_ifa = ifa;
1261                                 }
1262                         }
1263                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1264                         ln->ln_expire = 0;
1265                         ln->ln_state = ND6_LLINFO_REACHABLE;
1266                         ln->ln_byhint = 0;
1267
1268                         /* join solicited node multicast for proxy ND */
1269                         if (ifp->if_flags & IFF_MULTICAST) {
1270                                 struct in6_addr llsol;
1271                                 int error;
1272
1273                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1274                                 llsol.s6_addr16[0] = htons(0xff02);
1275                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1276                                 llsol.s6_addr32[1] = 0;
1277                                 llsol.s6_addr32[2] = htonl(1);
1278                                 llsol.s6_addr8[12] = 0xff;
1279
1280                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1281                                         nd6log((LOG_ERR, "%s: failed to join "
1282                                             "%s (errno=%d)\n", if_name(ifp),
1283                                             ip6_sprintf(&llsol), error));
1284                                 }
1285                         }
1286                 }
1287                 break;
1288
1289         case RTM_DELETE:
1290                 if (!ln)
1291                         break;
1292                 /* leave from solicited node multicast for proxy ND */
1293                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1294                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1295                         struct in6_addr llsol;
1296                         struct in6_multi *in6m;
1297
1298                         llsol = SIN6(rt_key(rt))->sin6_addr;
1299                         llsol.s6_addr16[0] = htons(0xff02);
1300                         llsol.s6_addr16[1] = htons(ifp->if_index);
1301                         llsol.s6_addr32[1] = 0;
1302                         llsol.s6_addr32[2] = htonl(1);
1303                         llsol.s6_addr8[12] = 0xff;
1304
1305                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1306                         if (in6m)
1307                                 in6_delmulti(in6m);
1308                 }
1309                 nd6_inuse--;
1310                 ln->ln_next->ln_prev = ln->ln_prev;
1311                 ln->ln_prev->ln_next = ln->ln_next;
1312                 ln->ln_prev = NULL;
1313                 rt->rt_llinfo = 0;
1314                 rt->rt_flags &= ~RTF_LLINFO;
1315                 if (ln->ln_hold)
1316                         m_freem(ln->ln_hold);
1317                 Free((caddr_t)ln);
1318         }
1319 }
1320
1321 int
1322 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1323 {
1324         struct in6_drlist *drl = (struct in6_drlist *)data;
1325         struct in6_prlist *prl = (struct in6_prlist *)data;
1326         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1327         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1328         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1329         struct nd_defrouter *dr, any;
1330         struct nd_prefix *pr;
1331         struct rtentry *rt;
1332         int i = 0, error = 0;
1333         int s;
1334
1335         switch (cmd) {
1336         case SIOCGDRLST_IN6:
1337                 /*
1338                  * obsolete API, use sysctl under net.inet6.icmp6
1339                  */
1340                 bzero(drl, sizeof(*drl));
1341                 s = splnet();
1342                 dr = TAILQ_FIRST(&nd_defrouter);
1343                 while (dr && i < DRLSTSIZ) {
1344                         drl->defrouter[i].rtaddr = dr->rtaddr;
1345                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1346                                 /* XXX: need to this hack for KAME stack */
1347                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1348                         } else
1349                                 log(LOG_ERR,
1350                                     "default router list contains a "
1351                                     "non-linklocal address(%s)\n",
1352                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1353
1354                         drl->defrouter[i].flags = dr->flags;
1355                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1356                         drl->defrouter[i].expire = dr->expire;
1357                         drl->defrouter[i].if_index = dr->ifp->if_index;
1358                         i++;
1359                         dr = TAILQ_NEXT(dr, dr_entry);
1360                 }
1361                 splx(s);
1362                 break;
1363         case SIOCGPRLST_IN6:
1364                 /*
1365                  * obsolete API, use sysctl under net.inet6.icmp6
1366                  */
1367                 /*
1368                  * XXX meaning of fields, especialy "raflags", is very
1369                  * differnet between RA prefix list and RR/static prefix list.
1370                  * how about separating ioctls into two?
1371                  */
1372                 bzero(prl, sizeof(*prl));
1373                 s = splnet();
1374                 pr = nd_prefix.lh_first;
1375                 while (pr && i < PRLSTSIZ) {
1376                         struct nd_pfxrouter *pfr;
1377                         int j;
1378
1379                         (void)in6_embedscope(&prl->prefix[i].prefix,
1380                             &pr->ndpr_prefix, NULL, NULL);
1381                         prl->prefix[i].raflags = pr->ndpr_raf;
1382                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1383                         prl->prefix[i].vltime = pr->ndpr_vltime;
1384                         prl->prefix[i].pltime = pr->ndpr_pltime;
1385                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1386                         prl->prefix[i].expire = pr->ndpr_expire;
1387
1388                         pfr = pr->ndpr_advrtrs.lh_first;
1389                         j = 0;
1390                         while (pfr) {
1391                                 if (j < DRLSTSIZ) {
1392 #define RTRADDR prl->prefix[i].advrtr[j]
1393                                         RTRADDR = pfr->router->rtaddr;
1394                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1395                                                 /* XXX: hack for KAME */
1396                                                 RTRADDR.s6_addr16[1] = 0;
1397                                         } else
1398                                                 log(LOG_ERR,
1399                                                     "a router(%s) advertises "
1400                                                     "a prefix with "
1401                                                     "non-link local address\n",
1402                                                     ip6_sprintf(&RTRADDR));
1403 #undef RTRADDR
1404                                 }
1405                                 j++;
1406                                 pfr = pfr->pfr_next;
1407                         }
1408                         prl->prefix[i].advrtrs = j;
1409                         prl->prefix[i].origin = PR_ORIG_RA;
1410
1411                         i++;
1412                         pr = pr->ndpr_next;
1413                 }
1414               {
1415                 struct rr_prefix *rpp;
1416
1417                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1418                      rpp = LIST_NEXT(rpp, rp_entry)) {
1419                         if (i >= PRLSTSIZ)
1420                                 break;
1421                         (void)in6_embedscope(&prl->prefix[i].prefix,
1422                             &pr->ndpr_prefix, NULL, NULL);
1423                         prl->prefix[i].raflags = rpp->rp_raf;
1424                         prl->prefix[i].prefixlen = rpp->rp_plen;
1425                         prl->prefix[i].vltime = rpp->rp_vltime;
1426                         prl->prefix[i].pltime = rpp->rp_pltime;
1427                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1428                         prl->prefix[i].expire = rpp->rp_expire;
1429                         prl->prefix[i].advrtrs = 0;
1430                         prl->prefix[i].origin = rpp->rp_origin;
1431                         i++;
1432                 }
1433               }
1434                 splx(s);
1435
1436                 break;
1437         case OSIOCGIFINFO_IN6:
1438                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1439                         error = EINVAL;
1440                         break;
1441                 }
1442                 ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1443                 ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1444                 ndi->ndi.basereachable =
1445                     nd_ifinfo[ifp->if_index].basereachable;
1446                 ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1447                 ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1448                 ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1449                 ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1450                 ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1451                 ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1452                 break;
1453         case SIOCGIFINFO_IN6:
1454                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1455                         error = EINVAL;
1456                         break;
1457                 }
1458                 ndi->ndi = nd_ifinfo[ifp->if_index];
1459                 break;
1460         case SIOCSIFINFO_FLAGS:
1461                 /* XXX: almost all other fields of ndi->ndi is unused */
1462                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1463                         error = EINVAL;
1464                         break;
1465                 }
1466                 nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1467                 break;
1468         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1469                 /* flush default router list */
1470                 /*
1471                  * xxx sumikawa: should not delete route if default
1472                  * route equals to the top of default router list
1473                  */
1474                 bzero(&any, sizeof(any));
1475                 defrouter_delreq(&any, 0);
1476                 defrouter_select();
1477                 /* xxx sumikawa: flush prefix list */
1478                 break;
1479         case SIOCSPFXFLUSH_IN6:
1480             {
1481                 /* flush all the prefix advertised by routers */
1482                 struct nd_prefix *pr, *next;
1483
1484                 s = splnet();
1485                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1486                         struct in6_ifaddr *ia, *ia_next;
1487
1488                         next = pr->ndpr_next;
1489
1490                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1491                                 continue; /* XXX */
1492
1493                         /* do we really have to remove addresses as well? */
1494                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1495                                 /* ia might be removed.  keep the next ptr. */
1496                                 ia_next = ia->ia_next;
1497
1498                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1499                                         continue;
1500
1501                                 if (ia->ia6_ndpr == pr)
1502                                         in6_purgeaddr(&ia->ia_ifa);
1503                         }
1504                         prelist_remove(pr);
1505                 }
1506                 splx(s);
1507                 break;
1508             }
1509         case SIOCSRTRFLUSH_IN6:
1510             {
1511                 /* flush all the default routers */
1512                 struct nd_defrouter *dr, *next;
1513
1514                 s = splnet();
1515                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1516                         /*
1517                          * The first entry of the list may be stored in
1518                          * the routing table, so we'll delete it later.
1519                          */
1520                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1521                                 next = TAILQ_NEXT(dr, dr_entry);
1522                                 defrtrlist_del(dr);
1523                         }
1524                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1525                 }
1526                 splx(s);
1527                 break;
1528             }
1529         case SIOCGNBRINFO_IN6:
1530             {
1531                 struct llinfo_nd6 *ln;
1532                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1533
1534                 /*
1535                  * XXX: KAME specific hack for scoped addresses
1536                  *      XXXX: for other scopes than link-local?
1537                  */
1538                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1539                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1540                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1541
1542                         if (*idp == 0)
1543                                 *idp = htons(ifp->if_index);
1544                 }
1545
1546                 s = splnet();
1547                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1548                         error = EINVAL;
1549                         splx(s);
1550                         break;
1551                 }
1552                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1553                 nbi->state = ln->ln_state;
1554                 nbi->asked = ln->ln_asked;
1555                 nbi->isrouter = ln->ln_router;
1556                 nbi->expire = ln->ln_expire;
1557                 splx(s);
1558
1559                 break;
1560             }
1561         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1562                 ndif->ifindex = nd6_defifindex;
1563                 break;
1564         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1565                 return(nd6_setdefaultiface(ndif->ifindex));
1566                 break;
1567         }
1568         return(error);
1569 }
1570
1571 /*
1572  * Create neighbor cache entry and cache link-layer address,
1573  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1574  */
1575 struct rtentry *
1576 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1577                  int lladdrlen,
1578                  int type,      /* ICMP6 type */
1579                  int code       /* type dependent information */)
1580 {
1581         struct rtentry *rt = NULL;
1582         struct llinfo_nd6 *ln = NULL;
1583         int is_newentry;
1584         struct sockaddr_dl *sdl = NULL;
1585         int do_update;
1586         int olladdr;
1587         int llchange;
1588         int newstate = 0;
1589
1590         if (!ifp)
1591                 panic("ifp == NULL in nd6_cache_lladdr");
1592         if (!from)
1593                 panic("from == NULL in nd6_cache_lladdr");
1594
1595         /* nothing must be updated for unspecified address */
1596         if (IN6_IS_ADDR_UNSPECIFIED(from))
1597                 return NULL;
1598
1599         /*
1600          * Validation about ifp->if_addrlen and lladdrlen must be done in
1601          * the caller.
1602          *
1603          * XXX If the link does not have link-layer adderss, what should
1604          * we do? (ifp->if_addrlen == 0)
1605          * Spec says nothing in sections for RA, RS and NA.  There's small
1606          * description on it in NS section (RFC 2461 7.2.3).
1607          */
1608
1609         rt = nd6_lookup(from, 0, ifp);
1610         if (!rt) {
1611 #if 0
1612                 /* nothing must be done if there's no lladdr */
1613                 if (!lladdr || !lladdrlen)
1614                         return NULL;
1615 #endif
1616
1617                 rt = nd6_lookup(from, 1, ifp);
1618                 is_newentry = 1;
1619         } else {
1620                 /* do nothing if static ndp is set */
1621                 if (rt->rt_flags & RTF_STATIC)
1622                         return NULL;
1623                 is_newentry = 0;
1624         }
1625
1626         if (!rt)
1627                 return NULL;
1628         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1629 fail:
1630                 (void)nd6_free(rt);
1631                 return NULL;
1632         }
1633         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1634         if (!ln)
1635                 goto fail;
1636         if (!rt->rt_gateway)
1637                 goto fail;
1638         if (rt->rt_gateway->sa_family != AF_LINK)
1639                 goto fail;
1640         sdl = SDL(rt->rt_gateway);
1641
1642         olladdr = (sdl->sdl_alen) ? 1 : 0;
1643         if (olladdr && lladdr) {
1644                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1645                         llchange = 1;
1646                 else
1647                         llchange = 0;
1648         } else
1649                 llchange = 0;
1650
1651         /*
1652          * newentry olladdr  lladdr  llchange   (*=record)
1653          *      0       n       n       --      (1)
1654          *      0       y       n       --      (2)
1655          *      0       n       y       --      (3) * STALE
1656          *      0       y       y       n       (4) *
1657          *      0       y       y       y       (5) * STALE
1658          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1659          *      1       --      y       --      (7) * STALE
1660          */
1661
1662         if (lladdr) {           /* (3-5) and (7) */
1663                 /*
1664                  * Record source link-layer address
1665                  * XXX is it dependent to ifp->if_type?
1666                  */
1667                 sdl->sdl_alen = ifp->if_addrlen;
1668                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1669         }
1670
1671         if (!is_newentry) {
1672                 if ((!olladdr && lladdr)                /* (3) */
1673                  || (olladdr && lladdr && llchange)) {  /* (5) */
1674                         do_update = 1;
1675                         newstate = ND6_LLINFO_STALE;
1676                 } else                                  /* (1-2,4) */
1677                         do_update = 0;
1678         } else {
1679                 do_update = 1;
1680                 if (!lladdr)                            /* (6) */
1681                         newstate = ND6_LLINFO_NOSTATE;
1682                 else                                    /* (7) */
1683                         newstate = ND6_LLINFO_STALE;
1684         }
1685
1686         if (do_update) {
1687                 /*
1688                  * Update the state of the neighbor cache.
1689                  */
1690                 ln->ln_state = newstate;
1691
1692                 if (ln->ln_state == ND6_LLINFO_STALE) {
1693                         /*
1694                          * XXX: since nd6_output() below will cause
1695                          * state tansition to DELAY and reset the timer,
1696                          * we must set the timer now, although it is actually
1697                          * meaningless.
1698                          */
1699                         ln->ln_expire = time_second + nd6_gctimer;
1700
1701                         if (ln->ln_hold) {
1702                                 /*
1703                                  * we assume ifp is not a p2p here, so just
1704                                  * set the 2nd argument as the 1st one.
1705                                  */
1706                                 nd6_output(ifp, ifp, ln->ln_hold,
1707                                            (struct sockaddr_in6 *)rt_key(rt),
1708                                            rt);
1709                                 ln->ln_hold = NULL;
1710                         }
1711                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1712                         /* probe right away */
1713                         ln->ln_expire = time_second;
1714                 }
1715         }
1716
1717         /*
1718          * ICMP6 type dependent behavior.
1719          *
1720          * NS: clear IsRouter if new entry
1721          * RS: clear IsRouter
1722          * RA: set IsRouter if there's lladdr
1723          * redir: clear IsRouter if new entry
1724          *
1725          * RA case, (1):
1726          * The spec says that we must set IsRouter in the following cases:
1727          * - If lladdr exist, set IsRouter.  This means (1-5).
1728          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1729          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1730          * A quetion arises for (1) case.  (1) case has no lladdr in the
1731          * neighbor cache, this is similar to (6).
1732          * This case is rare but we figured that we MUST NOT set IsRouter.
1733          *
1734          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1735          *                                                      D R
1736          *      0       n       n       --      (1)     c   ?     s
1737          *      0       y       n       --      (2)     c   s     s
1738          *      0       n       y       --      (3)     c   s     s
1739          *      0       y       y       n       (4)     c   s     s
1740          *      0       y       y       y       (5)     c   s     s
1741          *      1       --      n       --      (6) c   c       c s
1742          *      1       --      y       --      (7) c   c   s   c s
1743          *
1744          *                                      (c=clear s=set)
1745          */
1746         switch (type & 0xff) {
1747         case ND_NEIGHBOR_SOLICIT:
1748                 /*
1749                  * New entry must have is_router flag cleared.
1750                  */
1751                 if (is_newentry)        /* (6-7) */
1752                         ln->ln_router = 0;
1753                 break;
1754         case ND_REDIRECT:
1755                 /*
1756                  * If the icmp is a redirect to a better router, always set the
1757                  * is_router flag. Otherwise, if the entry is newly created,
1758                  * clear the flag. [RFC 2461, sec 8.3]
1759                  */
1760                 if (code == ND_REDIRECT_ROUTER)
1761                         ln->ln_router = 1;
1762                 else if (is_newentry) /* (6-7) */
1763                         ln->ln_router = 0;
1764                 break;
1765         case ND_ROUTER_SOLICIT:
1766                 /*
1767                  * is_router flag must always be cleared.
1768                  */
1769                 ln->ln_router = 0;
1770                 break;
1771         case ND_ROUTER_ADVERT:
1772                 /*
1773                  * Mark an entry with lladdr as a router.
1774                  */
1775                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1776                  || (is_newentry && lladdr)) {                  /* (7) */
1777                         ln->ln_router = 1;
1778                 }
1779                 break;
1780         }
1781
1782         /*
1783          * When the link-layer address of a router changes, select the
1784          * best router again.  In particular, when the neighbor entry is newly
1785          * created, it might affect the selection policy.
1786          * Question: can we restrict the first condition to the "is_newentry"
1787          * case?
1788          * XXX: when we hear an RA from a new router with the link-layer
1789          * address option, defrouter_select() is called twice, since
1790          * defrtrlist_update called the function as well.  However, I believe
1791          * we can compromise the overhead, since it only happens the first
1792          * time.
1793          * XXX: although defrouter_select() should not have a bad effect
1794          * for those are not autoconfigured hosts, we explicitly avoid such
1795          * cases for safety.
1796          */
1797         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1798                 defrouter_select();
1799
1800         return rt;
1801 }
1802
1803 static void
1804 nd6_slowtimo(void *ignored_arg)
1805 {
1806         int s = splnet();
1807         int i;
1808         struct nd_ifinfo *nd6if;
1809
1810         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1811             nd6_slowtimo, NULL);
1812         for (i = 1; i < if_index + 1; i++) {
1813                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1814                         continue;
1815                 nd6if = &nd_ifinfo[i];
1816                 if (nd6if->basereachable && /* already initialized */
1817                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1818                         /*
1819                          * Since reachable time rarely changes by router
1820                          * advertisements, we SHOULD insure that a new random
1821                          * value gets recomputed at least once every few hours.
1822                          * (RFC 2461, 6.3.4)
1823                          */
1824                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1825                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1826                 }
1827         }
1828         splx(s);
1829 }
1830
1831 #define senderr(e) { error = (e); goto bad;}
1832 int
1833 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1834            struct sockaddr_in6 *dst, struct rtentry *rt0)
1835 {
1836         struct mbuf *m = m0;
1837         struct rtentry *rt = rt0;
1838         struct sockaddr_in6 *gw6 = NULL;
1839         struct llinfo_nd6 *ln = NULL;
1840         int error = 0;
1841
1842         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1843                 goto sendpkt;
1844
1845         if (nd6_need_cache(ifp) == 0)
1846                 goto sendpkt;
1847
1848         /*
1849          * next hop determination.  This routine is derived from ether_outpout.
1850          */
1851         if (rt) {
1852                 if ((rt->rt_flags & RTF_UP) == 0) {
1853                         if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL)) !=
1854                                 NULL)
1855                         {
1856                                 rt->rt_refcnt--;
1857                                 if (rt->rt_ifp != ifp) {
1858                                         /* XXX: loop care? */
1859                                         return nd6_output(ifp, origifp, m0,
1860                                                           dst, rt);
1861                                 }
1862                         } else
1863                                 senderr(EHOSTUNREACH);
1864                 }
1865
1866                 if (rt->rt_flags & RTF_GATEWAY) {
1867                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1868
1869                         /*
1870                          * We skip link-layer address resolution and NUD
1871                          * if the gateway is not a neighbor from ND point
1872                          * of view, regardless of the value of nd_ifinfo.flags.
1873                          * The second condition is a bit tricky; we skip
1874                          * if the gateway is our own address, which is
1875                          * sometimes used to install a route to a p2p link.
1876                          */
1877                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1878                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1879                                 /*
1880                                  * We allow this kind of tricky route only
1881                                  * when the outgoing interface is p2p.
1882                                  * XXX: we may need a more generic rule here.
1883                                  */
1884                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1885                                         senderr(EHOSTUNREACH);
1886
1887                                 goto sendpkt;
1888                         }
1889
1890                         if (rt->rt_gwroute == 0)
1891                                 goto lookup;
1892                         if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1893                                 rtfree(rt); rt = rt0;
1894                         lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1895                                 if ((rt = rt->rt_gwroute) == 0)
1896                                         senderr(EHOSTUNREACH);
1897                         }
1898                 }
1899         }
1900
1901         /*
1902          * Address resolution or Neighbor Unreachability Detection
1903          * for the next hop.
1904          * At this point, the destination of the packet must be a unicast
1905          * or an anycast address(i.e. not a multicast).
1906          */
1907
1908         /* Look up the neighbor cache for the nexthop */
1909         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1910                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1911         else {
1912                 /*
1913                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1914                  * the condition below is not very efficient.  But we believe
1915                  * it is tolerable, because this should be a rare case.
1916                  */
1917                 if (nd6_is_addr_neighbor(dst, ifp) &&
1918                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1919                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1920         }
1921         if (!ln || !rt) {
1922                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1923                     !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1924                         log(LOG_DEBUG,
1925                             "nd6_output: can't allocate llinfo for %s "
1926                             "(ln=%p, rt=%p)\n",
1927                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1928                         senderr(EIO);   /* XXX: good error? */
1929                 }
1930
1931                 goto sendpkt;   /* send anyway */
1932         }
1933
1934         /* We don't have to do link-layer address resolution on a p2p link. */
1935         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1936             ln->ln_state < ND6_LLINFO_REACHABLE) {
1937                 ln->ln_state = ND6_LLINFO_STALE;
1938                 ln->ln_expire = time_second + nd6_gctimer;
1939         }
1940
1941         /*
1942          * The first time we send a packet to a neighbor whose entry is
1943          * STALE, we have to change the state to DELAY and a sets a timer to
1944          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1945          * neighbor unreachability detection on expiration.
1946          * (RFC 2461 7.3.3)
1947          */
1948         if (ln->ln_state == ND6_LLINFO_STALE) {
1949                 ln->ln_asked = 0;
1950                 ln->ln_state = ND6_LLINFO_DELAY;
1951                 ln->ln_expire = time_second + nd6_delay;
1952         }
1953
1954         /*
1955          * If the neighbor cache entry has a state other than INCOMPLETE
1956          * (i.e. its link-layer address is already resolved), just
1957          * send the packet.
1958          */
1959         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1960                 goto sendpkt;
1961
1962         /*
1963          * There is a neighbor cache entry, but no ethernet address
1964          * response yet.  Replace the held mbuf (if any) with this
1965          * latest one.
1966          *
1967          * This code conforms to the rate-limiting rule described in Section
1968          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1969          * an NS below.
1970          */
1971         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1972                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1973         if (ln->ln_hold)
1974                 m_freem(ln->ln_hold);
1975         ln->ln_hold = m;
1976         if (ln->ln_expire) {
1977                 if (ln->ln_asked < nd6_mmaxtries &&
1978                     ln->ln_expire < time_second) {
1979                         ln->ln_asked++;
1980                         ln->ln_expire = time_second +
1981                                 nd_ifinfo[ifp->if_index].retrans / 1000;
1982                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1983                 }
1984         }
1985         return(0);
1986
1987   sendpkt:
1988
1989         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1990                 return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1991                                          rt));
1992         }
1993         return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1994
1995   bad:
1996         if (m)
1997                 m_freem(m);
1998         return (error);
1999 }
2000 #undef senderr
2001
2002 int
2003 nd6_need_cache(struct ifnet *ifp)
2004 {
2005         /*
2006          * XXX: we currently do not make neighbor cache on any interface
2007          * other than ARCnet, Ethernet, FDDI and GIF.
2008          *
2009          * RFC2893 says:
2010          * - unidirectional tunnels needs no ND
2011          */
2012         switch (ifp->if_type) {
2013         case IFT_ARCNET:
2014         case IFT_ETHER:
2015         case IFT_FDDI:
2016         case IFT_IEEE1394:
2017 #ifdef IFT_L2VLAN
2018         case IFT_L2VLAN:
2019 #endif
2020 #ifdef IFT_IEEE80211
2021         case IFT_IEEE80211:
2022 #endif
2023         case IFT_GIF:           /* XXX need more cases? */
2024                 return(1);
2025         default:
2026                 return(0);
2027         }
2028 }
2029
2030 int
2031 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt, struct mbuf *m,
2032                 struct sockaddr *dst, u_char *desten)
2033 {
2034         int i;
2035         struct sockaddr_dl *sdl;
2036
2037         if (m->m_flags & M_MCAST) {
2038                 switch (ifp->if_type) {
2039                 case IFT_ETHER:
2040                 case IFT_FDDI:
2041 #ifdef IFT_L2VLAN
2042         case IFT_L2VLAN:
2043 #endif
2044 #ifdef IFT_IEEE80211
2045                 case IFT_IEEE80211:
2046 #endif
2047                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2048                                                  desten);
2049                         return(1);
2050                 case IFT_IEEE1394:
2051                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2052                         return(1);
2053                 case IFT_ARCNET:
2054                         *desten = 0;
2055                         return(1);
2056                 default:
2057                         m_freem(m);
2058                         return(0);
2059                 }
2060         }
2061
2062         if (rt == NULL) {
2063                 /* this could happen, if we could not allocate memory */
2064                 m_freem(m);
2065                 return(0);
2066         }
2067         if (rt->rt_gateway->sa_family != AF_LINK) {
2068                 printf("nd6_storelladdr: something odd happens\n");
2069                 m_freem(m);
2070                 return(0);
2071         }
2072         sdl = SDL(rt->rt_gateway);
2073         if (sdl->sdl_alen == 0) {
2074                 /* this should be impossible, but we bark here for debugging */
2075                 printf("nd6_storelladdr: sdl_alen == 0\n");
2076                 m_freem(m);
2077                 return(0);
2078         }
2079
2080         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2081         return(1);
2082 }
2083
2084 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2085 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2086 #ifdef SYSCTL_DECL
2087 SYSCTL_DECL(_net_inet6_icmp6);
2088 #endif
2089 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2090         CTLFLAG_RD, nd6_sysctl_drlist, "");
2091 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2092         CTLFLAG_RD, nd6_sysctl_prlist, "");
2093
2094 static int
2095 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2096 {
2097         int error;
2098         char buf[1024];
2099         struct in6_defrouter *d, *de;
2100         struct nd_defrouter *dr;
2101
2102         if (req->newptr)
2103                 return EPERM;
2104         error = 0;
2105
2106         for (dr = TAILQ_FIRST(&nd_defrouter);
2107              dr;
2108              dr = TAILQ_NEXT(dr, dr_entry)) {
2109                 d = (struct in6_defrouter *)buf;
2110                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2111
2112                 if (d + 1 <= de) {
2113                         bzero(d, sizeof(*d));
2114                         d->rtaddr.sin6_family = AF_INET6;
2115                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2116                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2117                             dr->ifp) != 0)
2118                                 log(LOG_ERR,
2119                                     "scope error in "
2120                                     "default router list (%s)\n",
2121                                     ip6_sprintf(&dr->rtaddr));
2122                         d->flags = dr->flags;
2123                         d->rtlifetime = dr->rtlifetime;
2124                         d->expire = dr->expire;
2125                         d->if_index = dr->ifp->if_index;
2126                 } else
2127                         panic("buffer too short");
2128
2129                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2130                 if (error)
2131                         break;
2132         }
2133         return error;
2134 }
2135
2136 static int
2137 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2138 {
2139         int error;
2140         char buf[1024];
2141         struct in6_prefix *p, *pe;
2142         struct nd_prefix *pr;
2143
2144         if (req->newptr)
2145                 return EPERM;
2146         error = 0;
2147
2148         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2149                 u_short advrtrs;
2150                 size_t advance;
2151                 struct sockaddr_in6 *sin6, *s6;
2152                 struct nd_pfxrouter *pfr;
2153
2154                 p = (struct in6_prefix *)buf;
2155                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2156
2157                 if (p + 1 <= pe) {
2158                         bzero(p, sizeof(*p));
2159                         sin6 = (struct sockaddr_in6 *)(p + 1);
2160
2161                         p->prefix = pr->ndpr_prefix;
2162                         if (in6_recoverscope(&p->prefix,
2163                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2164                                 log(LOG_ERR,
2165                                     "scope error in prefix list (%s)\n",
2166                                     ip6_sprintf(&p->prefix.sin6_addr));
2167                         p->raflags = pr->ndpr_raf;
2168                         p->prefixlen = pr->ndpr_plen;
2169                         p->vltime = pr->ndpr_vltime;
2170                         p->pltime = pr->ndpr_pltime;
2171                         p->if_index = pr->ndpr_ifp->if_index;
2172                         p->expire = pr->ndpr_expire;
2173                         p->refcnt = pr->ndpr_refcnt;
2174                         p->flags = pr->ndpr_stateflags;
2175                         p->origin = PR_ORIG_RA;
2176                         advrtrs = 0;
2177                         for (pfr = pr->ndpr_advrtrs.lh_first;
2178                              pfr;
2179                              pfr = pfr->pfr_next) {
2180                                 if ((void *)&sin6[advrtrs + 1] >
2181                                     (void *)pe) {
2182                                         advrtrs++;
2183                                         continue;
2184                                 }
2185                                 s6 = &sin6[advrtrs];
2186                                 bzero(s6, sizeof(*s6));
2187                                 s6->sin6_family = AF_INET6;
2188                                 s6->sin6_len = sizeof(*sin6);
2189                                 if (in6_recoverscope(s6,
2190                                     &pfr->router->rtaddr,
2191                                     pfr->router->ifp) != 0)
2192                                         log(LOG_ERR,
2193                                             "scope error in "
2194                                             "prefix list (%s)\n",
2195                                             ip6_sprintf(&pfr->router->rtaddr));
2196                                 advrtrs++;
2197                         }
2198                         p->advrtrs = advrtrs;
2199                 } else
2200                         panic("buffer too short");
2201
2202                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2203                 error = SYSCTL_OUT(req, buf, advance);
2204                 if (error)
2205                         break;
2206         }
2207         return error;
2208 }