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 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   3
   4 /*
   5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of the project nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 #include "opt_inet.h"
  34 #include "opt_inet6.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/callout.h>
  39 #include <sys/malloc.h>
  40 #include <sys/mbuf.h>
  41 #include <sys/socket.h>
  42 #include <sys/sockio.h>
  43 #include <sys/time.h>
  44 #include <sys/kernel.h>
  45 #include <sys/protosw.h>
  46 #include <sys/errno.h>
  47 #include <sys/syslog.h>
  48 #include <sys/queue.h>
  49 #include <sys/sysctl.h>
  50 #include <sys/mutex.h>
  51
  52 #include <sys/thread2.h>
  53 #include <sys/mutex2.h>
  54
  55 #include <net/if.h>
  56 #include <net/if_dl.h>
  57 #include <net/if_types.h>
  58 #include <net/route.h>
  59 #include <net/netisr2.h>
  60 #include <net/netmsg2.h>
  61
  62 #include <netinet/in.h>
  63 #include <netinet/if_ether.h>
  64 #include <netinet6/in6_var.h>
  65 #include <netinet/ip6.h>
  66 #include <netinet6/ip6_var.h>
  67 #include <netinet6/nd6.h>
  68 #include <netinet/icmp6.h>
  69
  70 #include <net/net_osdep.h>
  71
  72 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  73 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  74
  75 #define SIN6(s) ((struct sockaddr_in6 *)s)
  76 #define SDL(s) ((struct sockaddr_dl *)s)
  77
  78 /*
  79  * Note that the check for rt_llinfo is necessary because a cloned
  80  * route from a parent route that has the L flag (e.g. the default
  81  * route to a p2p interface) may have the flag, too, while the
  82  * destination is not actually a neighbor.
  83  * XXX: we can't use rt->rt_ifp to check for the interface, since
  84  *      it might be the loopback interface if the entry is for our
  85  *      own address on a non-loopback interface. Instead, we should
  86  *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
  87  *      interface.
  88  */
  89 #define ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)                        \
  90         !(((rt)->rt_flags & RTF_GATEWAY) ||                     \
  91           ((rt)->rt_flags & RTF_LLINFO) == 0 ||                 \
  92           (rt)->rt_gateway->sa_family != AF_LINK ||             \
  93           (rt)->rt_llinfo == NULL ||                            \
  94           ((ifp) != NULL && (rt)->rt_ifa->ifa_ifp != (ifp)))
  95
  96 #define ND6_RTENTRY_IS_LLCLONING(rt)                            \
  97         (((rt)->rt_flags & (RTF_PRCLONING | RTF_LLINFO)) ==     \
  98          (RTF_PRCLONING | RTF_LLINFO) ||                        \
  99          ((rt)->rt_flags & RTF_CLONING))
 100
 101 /* timer values */
 102 int     nd6_prune       = 1;    /* walk list every 1 seconds */
 103 int     nd6_delay       = 5;    /* delay first probe time 5 second */
 104 int     nd6_umaxtries   = 3;    /* maximum unicast query */
 105 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
 106 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
 107 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
 108
 109 /* preventing too many loops in ND option parsing */
 110 int nd6_maxndopt = 10;  /* max # of ND options allowed */
 111
 112 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
 113
 114 #ifdef ND6_DEBUG
 115 int nd6_debug = 1;
 116 #else
 117 int nd6_debug = 0;
 118 #endif
 119
 120 /* for debugging? */
 121 static int nd6_inuse, nd6_allocated;
 122
 123 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 124 struct nd_drhead nd_defrouter;
 125 struct nd_prhead nd_prefix = { 0 };
 126 struct mtx nd6_mtx = MTX_INITIALIZER("nd6");
 127
 128 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 129 static struct sockaddr_in6 all1_sa;
 130
 131 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
 132 static int regen_tmpaddr (struct in6_ifaddr *);
 133 static void nd6_slowtimo(void *);
 134 static void nd6_slowtimo_dispatch(netmsg_t);
 135 static void nd6_timer(void *);
 136 static void nd6_timer_dispatch(netmsg_t);
 137
 138 static struct callout nd6_slowtimo_ch;
 139 static struct netmsg_base nd6_slowtimo_netmsg;
 140
 141 static struct callout nd6_timer_ch;
 142 static struct netmsg_base nd6_timer_netmsg;
 143
 144 void
 145 nd6_init(void)
 146 {
 147         static int nd6_init_done = 0;
 148         int i;
 149
 150         if (nd6_init_done) {
 151                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 152                 return;
 153         }
 154
 155         all1_sa.sin6_family = AF_INET6;
 156         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 157         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 158                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 159
 160         /* initialization of the default router list */
 161         TAILQ_INIT(&nd_defrouter);
 162
 163         nd6_init_done = 1;
 164
 165         /* start timer */
 166         callout_init_mp(&nd6_slowtimo_ch);
 167         netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
 168             MSGF_PRIORITY, nd6_slowtimo_dispatch);
 169         callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 170             nd6_slowtimo, NULL, 0);
 171 }
 172
 173 struct nd_ifinfo *
 174 nd6_ifattach(struct ifnet *ifp)
 175 {
 176         struct nd_ifinfo *nd;
 177
 178         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
 179             M_WAITOK | M_ZERO);
 180
 181         nd->initialized = 1;
 182
 183         nd->chlim = IPV6_DEFHLIM;
 184         nd->basereachable = REACHABLE_TIME;
 185         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 186         nd->retrans = RETRANS_TIMER;
 187
 188         /*
 189          * Note that the default value of ip6_accept_rtadv is 0, which means
 190          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 191          * here.
 192          */
 193         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 194
 195         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 196         nd6_setmtu0(ifp, nd);
 197         return nd;
 198 }
 199
 200 void
 201 nd6_ifdetach(struct nd_ifinfo *nd)
 202 {
 203         kfree(nd, M_IP6NDP);
 204 }
 205
 206 /*
 207  * Reset ND level link MTU. This function is called when the physical MTU
 208  * changes, which means we might have to adjust the ND level MTU.
 209  */
 210 void
 211 nd6_setmtu(struct ifnet *ifp)
 212 {
 213         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 214 }
 215
 216 struct netmsg_nd6setmtu {
 217         struct netmsg_base      nmsg;
 218         struct ifnet            *ifp;
 219         struct nd_ifinfo        *ndi;
 220 };
 221
 222 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 223 static void
 224 nd6_setmtu0_dispatch(netmsg_t msg)
 225 {
 226         struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
 227         struct ifnet *ifp = nmsg->ifp;
 228         struct nd_ifinfo *ndi = nmsg->ndi;
 229         uint32_t omaxmtu;
 230
 231         omaxmtu = ndi->maxmtu;
 232
 233         switch (ifp->if_type) {
 234         case IFT_ETHER:
 235                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 236                 break;
 237         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 238                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 239                 break;
 240 #ifdef IFT_IEEE80211
 241         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 242                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 243                 break;
 244 #endif
 245         default:
 246                 ndi->maxmtu = ifp->if_mtu;
 247                 break;
 248         }
 249
 250         /*
 251          * Decreasing the interface MTU under IPV6 minimum MTU may cause
 252          * undesirable situation.  We thus notify the operator of the change
 253          * explicitly.  The check for omaxmtu is necessary to restrict the
 254          * log to the case of changing the MTU, not initializing it.
 255          */
 256         if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
 257                 log(LOG_NOTICE, "nd6_setmtu0: "
 258                     "new link MTU on %s (%lu) is too small for IPv6\n",
 259                     if_name(ifp), (unsigned long)ndi->maxmtu);
 260         }
 261
 262         if (ndi->maxmtu > in6_maxmtu)
 263                 in6_setmaxmtu(); /* check all interfaces just in case */
 264
 265         lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
 266 }
 267
 268 void
 269 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 270 {
 271         struct netmsg_nd6setmtu nmsg;
 272
 273         netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
 274             nd6_setmtu0_dispatch);
 275         nmsg.ifp = ifp;
 276         nmsg.ndi = ndi;
 277         lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
 278 }
 279
 280 void
 281 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 282 {
 283         bzero(ndopts, sizeof(*ndopts));
 284         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 285         ndopts->nd_opts_last
 286                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 287
 288         if (icmp6len == 0) {
 289                 ndopts->nd_opts_done = 1;
 290                 ndopts->nd_opts_search = NULL;
 291         }
 292 }
 293
 294 /*
 295  * Take one ND option.
 296  */
 297 struct nd_opt_hdr *
 298 nd6_option(union nd_opts *ndopts)
 299 {
 300         struct nd_opt_hdr *nd_opt;
 301         int olen;
 302
 303         if (!ndopts)
 304                 panic("ndopts == NULL in nd6_option");
 305         if (!ndopts->nd_opts_last)
 306                 panic("uninitialized ndopts in nd6_option");
 307         if (!ndopts->nd_opts_search)
 308                 return NULL;
 309         if (ndopts->nd_opts_done)
 310                 return NULL;
 311
 312         nd_opt = ndopts->nd_opts_search;
 313
 314         /* make sure nd_opt_len is inside the buffer */
 315         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 316                 bzero(ndopts, sizeof(*ndopts));
 317                 return NULL;
 318         }
 319
 320         olen = nd_opt->nd_opt_len << 3;
 321         if (olen == 0) {
 322                 /*
 323                  * Message validation requires that all included
 324                  * options have a length that is greater than zero.
 325                  */
 326                 bzero(ndopts, sizeof(*ndopts));
 327                 return NULL;
 328         }
 329
 330         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 331         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 332                 /* option overruns the end of buffer, invalid */
 333                 bzero(ndopts, sizeof(*ndopts));
 334                 return NULL;
 335         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 336                 /* reached the end of options chain */
 337                 ndopts->nd_opts_done = 1;
 338                 ndopts->nd_opts_search = NULL;
 339         }
 340         return nd_opt;
 341 }
 342
 343 /*
 344  * Parse multiple ND options.
 345  * This function is much easier to use, for ND routines that do not need
 346  * multiple options of the same type.
 347  */
 348 int
 349 nd6_options(union nd_opts *ndopts)
 350 {
 351         struct nd_opt_hdr *nd_opt;
 352         int i = 0;
 353
 354         if (!ndopts)
 355                 panic("ndopts == NULL in nd6_options");
 356         if (!ndopts->nd_opts_last)
 357                 panic("uninitialized ndopts in nd6_options");
 358         if (!ndopts->nd_opts_search)
 359                 return 0;
 360
 361         while (1) {
 362                 nd_opt = nd6_option(ndopts);
 363                 if (!nd_opt && !ndopts->nd_opts_last) {
 364                         /*
 365                          * Message validation requires that all included
 366                          * options have a length that is greater than zero.
 367                          */
 368                         icmp6stat.icp6s_nd_badopt++;
 369                         bzero(ndopts, sizeof(*ndopts));
 370                         return -1;
 371                 }
 372
 373                 if (!nd_opt)
 374                         goto skip1;
 375
 376                 switch (nd_opt->nd_opt_type) {
 377                 case ND_OPT_SOURCE_LINKADDR:
 378                 case ND_OPT_TARGET_LINKADDR:
 379                 case ND_OPT_MTU:
 380                 case ND_OPT_REDIRECTED_HEADER:
 381                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 382                                 nd6log((LOG_INFO,
 383                                     "duplicated ND6 option found (type=%d)\n",
 384                                     nd_opt->nd_opt_type));
 385                                 /* XXX bark? */
 386                         } else {
 387                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 388                                         = nd_opt;
 389                         }
 390                         break;
 391                 case ND_OPT_PREFIX_INFORMATION:
 392                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 393                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 394                                         = nd_opt;
 395                         }
 396                         ndopts->nd_opts_pi_end =
 397                                 (struct nd_opt_prefix_info *)nd_opt;
 398                         break;
 399                 default:
 400                         /*
 401                          * Unknown options must be silently ignored,
 402                          * to accomodate future extension to the protocol.
 403                          */
 404                         nd6log((LOG_DEBUG,
 405                             "nd6_options: unsupported option %d - "
 406                             "option ignored\n", nd_opt->nd_opt_type));
 407                 }
 408
 409 skip1:
 410                 i++;
 411                 if (i > nd6_maxndopt) {
 412                         icmp6stat.icp6s_nd_toomanyopt++;
 413                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 414                         break;
 415                 }
 416
 417                 if (ndopts->nd_opts_done)
 418                         break;
 419         }
 420
 421         return 0;
 422 }
 423
 424 /*
 425  * ND6 timer routine to expire default route list and prefix list
 426  */
 427 static void
 428 nd6_timer_dispatch(netmsg_t nmsg)
 429 {
 430         struct llinfo_nd6 *ln;
 431         struct nd_defrouter *dr;
 432         struct nd_prefix *pr;
 433         struct ifnet *ifp;
 434         struct in6_ifaddr *ia6, *nia6;
 435
 436         ASSERT_NETISR0;
 437
 438         crit_enter();
 439         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
 440         crit_exit();
 441
 442         mtx_lock(&nd6_mtx);
 443
 444         ln = llinfo_nd6.ln_next;
 445         while (ln && ln != &llinfo_nd6) {
 446                 struct rtentry *rt;
 447                 struct sockaddr_in6 *dst;
 448                 struct llinfo_nd6 *next = ln->ln_next;
 449                 /* XXX: used for the DELAY case only: */
 450                 struct nd_ifinfo *ndi = NULL;
 451
 452                 if ((rt = ln->ln_rt) == NULL) {
 453                         ln = next;
 454                         continue;
 455                 }
 456                 if ((ifp = rt->rt_ifp) == NULL) {
 457                         ln = next;
 458                         continue;
 459                 }
 460                 ndi = ND_IFINFO(ifp);
 461                 dst = (struct sockaddr_in6 *)rt_key(rt);
 462
 463                 if (ln->ln_expire > time_uptime) {
 464                         ln = next;
 465                         continue;
 466                 }
 467
 468                 /* sanity check */
 469                 if (!rt)
 470                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 471                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 472                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 473                               rt->rt_llinfo, ln);
 474                 if (!dst)
 475                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 476
 477                 switch (ln->ln_state) {
 478                 case ND6_LLINFO_WAITDELETE:
 479                         next = nd6_free(rt);
 480                         break;
 481                 case ND6_LLINFO_INCOMPLETE:
 482                         if (ln->ln_asked++ >= nd6_mmaxtries) {
 483                                 struct mbuf *m = ln->ln_hold;
 484                                 if (m) {
 485                                         if (rt->rt_ifp) {
 486                                                 /*
 487                                                  * Fake rcvif to make ICMP error
 488                                                  * more helpful in diagnosing
 489                                                  * for the receiver.
 490                                                  * XXX: should we consider
 491                                                  * older rcvif?
 492                                                  */
 493                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 494                                         }
 495                                         /* XXX This will log a diagnostic
 496                                          * that it's not an IPv6 header. */
 497                                         icmp6_error(m, ICMP6_DST_UNREACH,
 498                                                     ICMP6_DST_UNREACH_ADDR, 0);
 499                                         ln->ln_hold = NULL;
 500                                 }
 501                                 ln->ln_state = ND6_LLINFO_WAITDELETE;
 502                                 rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
 503                         }
 504                         ln->ln_expire = time_uptime +
 505                                 ND_IFINFO(ifp)->retrans / 1000;
 506                         nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 507                                 ln, 0);
 508                         break;
 509                 case ND6_LLINFO_REACHABLE:
 510                         if (ln->ln_expire) {
 511                                 ln->ln_state = ND6_LLINFO_STALE;
 512                                 ln->ln_expire = time_uptime + nd6_gctimer;
 513                         }
 514                         break;
 515
 516                 case ND6_LLINFO_STALE:
 517                         /* Garbage Collection(RFC 2461 5.3) */
 518                         if (ln->ln_expire)
 519                                 next = nd6_free(rt);
 520                         break;
 521
 522                 case ND6_LLINFO_DELAY:
 523                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 524                                 /* We need NUD */
 525                                 ln->ln_asked = 1;
 526                                 ln->ln_state = ND6_LLINFO_PROBE;
 527                                 ln->ln_expire = time_uptime +
 528                                         ndi->retrans / 1000;
 529                                 nd6_ns_output(ifp, &dst->sin6_addr,
 530                                               &dst->sin6_addr,
 531                                               ln, 0);
 532                         } else {
 533                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 534                                 ln->ln_expire = time_uptime + nd6_gctimer;
 535                         }
 536                         break;
 537                 case ND6_LLINFO_PROBE:
 538                         if (ln->ln_asked < nd6_umaxtries) {
 539                                 ln->ln_asked++;
 540                                 ln->ln_expire = time_uptime +
 541                                         ND_IFINFO(ifp)->retrans / 1000;
 542                                 nd6_ns_output(ifp, &dst->sin6_addr,
 543                                                &dst->sin6_addr, ln, 0);
 544                         } else {
 545                                 rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
 546                                 next = nd6_free(rt);
 547                         }
 548                         break;
 549                 }
 550                 ln = next;
 551         }
 552
 553         /* expire default router list */
 554         dr = TAILQ_FIRST(&nd_defrouter);
 555         while (dr) {
 556                 if (dr->expire && dr->expire < time_uptime) {
 557                         struct nd_defrouter *t;
 558                         t = TAILQ_NEXT(dr, dr_entry);
 559                         defrtrlist_del(dr);
 560                         dr = t;
 561                 } else {
 562                         dr = TAILQ_NEXT(dr, dr_entry);
 563                 }
 564         }
 565
 566         /*
 567          * expire interface addresses.
 568          * in the past the loop was inside prefix expiry processing.
 569          * However, from a stricter speci-confrmance standpoint, we should
 570          * rather separate address lifetimes and prefix lifetimes.
 571          */
 572 addrloop:
 573         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 574                 nia6 = ia6->ia_next;
 575                 /* check address lifetime */
 576                 if (IFA6_IS_INVALID(ia6)) {
 577                         int regen = 0;
 578
 579                         /*
 580                          * If the expiring address is temporary, try
 581                          * regenerating a new one.  This would be useful when
 582                          * we suspended a laptop PC, then turned it on after a
 583                          * period that could invalidate all temporary
 584                          * addresses.  Although we may have to restart the
 585                          * loop (see below), it must be after purging the
 586                          * address.  Otherwise, we'd see an infinite loop of
 587                          * regeneration.
 588                          */
 589                         if (ip6_use_tempaddr &&
 590                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 591                                 if (regen_tmpaddr(ia6) == 0)
 592                                         regen = 1;
 593                         }
 594
 595                         in6_purgeaddr(&ia6->ia_ifa);
 596
 597                         if (regen)
 598                                 goto addrloop; /* XXX: see below */
 599                 }
 600                 if (IFA6_IS_DEPRECATED(ia6)) {
 601                         int oldflags = ia6->ia6_flags;
 602
 603                         if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
 604                                 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 605                                 in6_newaddrmsg((struct ifaddr *)ia6);
 606                         }
 607
 608                         /*
 609                          * If a temporary address has just become deprecated,
 610                          * regenerate a new one if possible.
 611                          */
 612                         if (ip6_use_tempaddr &&
 613                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 614                             !(oldflags & IN6_IFF_DEPRECATED)) {
 615
 616                                 if (regen_tmpaddr(ia6) == 0) {
 617                                         /*
 618                                          * A new temporary address is
 619                                          * generated.
 620                                          * XXX: this means the address chain
 621                                          * has changed while we are still in
 622                                          * the loop.  Although the change
 623                                          * would not cause disaster (because
 624                                          * it's not a deletion, but an
 625                                          * addition,) we'd rather restart the
 626                                          * loop just for safety.  Or does this
 627                                          * significantly reduce performance??
 628                                          */
 629                                         goto addrloop;
 630                                 }
 631                         }
 632                 } else {
 633                         /*
 634                          * A new RA might have made a deprecated address
 635                          * preferred.
 636                          */
 637                         if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
 638                                 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 639                                 in6_newaddrmsg((struct ifaddr *)ia6);
 640                         }
 641                 }
 642         }
 643
 644         /* expire prefix list */
 645         pr = nd_prefix.lh_first;
 646         while (pr) {
 647                 /*
 648                  * check prefix lifetime.
 649                  * since pltime is just for autoconf, pltime processing for
 650                  * prefix is not necessary.
 651                  */
 652                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
 653                         struct nd_prefix *t;
 654                         t = pr->ndpr_next;
 655
 656                         /*
 657                          * address expiration and prefix expiration are
 658                          * separate.  NEVER perform in6_purgeaddr here.
 659                          */
 660
 661                         prelist_remove(pr);
 662                         pr = t;
 663                 } else
 664                         pr = pr->ndpr_next;
 665         }
 666
 667         mtx_unlock(&nd6_mtx);
 668
 669         callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
 670 }
 671
 672 static void
 673 nd6_timer(void *arg __unused)
 674 {
 675         struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
 676
 677         KASSERT(mycpuid == 0, ("not on cpu0"));
 678         crit_enter();
 679         if (lmsg->ms_flags & MSGF_DONE)
 680                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
 681         crit_exit();
 682 }
 683
 684 void
 685 nd6_timer_init(void)
 686 {
 687         callout_init_mp(&nd6_timer_ch);
 688         netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
 689             MSGF_PRIORITY, nd6_timer_dispatch);
 690         callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
 691 }
 692
 693 static int
 694 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 695                                          address */
 696 {
 697         struct ifaddr_container *ifac;
 698         struct ifnet *ifp;
 699         struct in6_ifaddr *public_ifa6 = NULL;
 700
 701         ifp = ia6->ia_ifa.ifa_ifp;
 702         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 703                 struct ifaddr *ifa = ifac->ifa;
 704                 struct in6_ifaddr *it6;
 705
 706                 if (ifa->ifa_addr->sa_family != AF_INET6)
 707                         continue;
 708
 709                 it6 = (struct in6_ifaddr *)ifa;
 710
 711                 /* ignore no autoconf addresses. */
 712                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 713                         continue;
 714
 715                 /* ignore autoconf addresses with different prefixes. */
 716                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 717                         continue;
 718
 719                 /*
 720                  * Now we are looking at an autoconf address with the same
 721                  * prefix as ours.  If the address is temporary and is still
 722                  * preferred, do not create another one.  It would be rare, but
 723                  * could happen, for example, when we resume a laptop PC after
 724                  * a long period.
 725                  */
 726                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 727                     !IFA6_IS_DEPRECATED(it6)) {
 728                         public_ifa6 = NULL;
 729                         break;
 730                 }
 731
 732                 /*
 733                  * This is a public autoconf address that has the same prefix
 734                  * as ours.  If it is preferred, keep it.  We can't break the
 735                  * loop here, because there may be a still-preferred temporary
 736                  * address with the prefix.
 737                  */
 738                 if (!IFA6_IS_DEPRECATED(it6))
 739                     public_ifa6 = it6;
 740         }
 741
 742         if (public_ifa6 != NULL) {
 743                 int e;
 744
 745                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 746                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 747                             " tmp addr,errno=%d\n", e);
 748                         return (-1);
 749                 }
 750                 return (0);
 751         }
 752
 753         return (-1);
 754 }
 755
 756 /*
 757  * Nuke neighbor cache/prefix/default router management table, right before
 758  * ifp goes away.
 759  */
 760 void
 761 nd6_purge(struct ifnet *ifp)
 762 {
 763         struct llinfo_nd6 *ln, *nln;
 764         struct nd_defrouter *dr, *ndr, drany;
 765         struct nd_prefix *pr, *npr;
 766
 767         /* Nuke default router list entries toward ifp */
 768         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 769                 /*
 770                  * The first entry of the list may be stored in
 771                  * the routing table, so we'll delete it later.
 772                  */
 773                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 774                         ndr = TAILQ_NEXT(dr, dr_entry);
 775                         if (dr->ifp == ifp)
 776                                 defrtrlist_del(dr);
 777                 }
 778                 dr = TAILQ_FIRST(&nd_defrouter);
 779                 if (dr->ifp == ifp)
 780                         defrtrlist_del(dr);
 781         }
 782
 783         /* Nuke prefix list entries toward ifp */
 784         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 785                 npr = pr->ndpr_next;
 786                 if (pr->ndpr_ifp == ifp) {
 787                         /*
 788                          * Previously, pr->ndpr_addr is removed as well,
 789                          * but I strongly believe we don't have to do it.
 790                          * nd6_purge() is only called from in6_ifdetach(),
 791                          * which removes all the associated interface addresses
 792                          * by itself.
 793                          * (jinmei@kame.net 20010129)
 794                          */
 795                         prelist_remove(pr);
 796                 }
 797         }
 798
 799         /* cancel default outgoing interface setting */
 800         if (nd6_defifindex == ifp->if_index)
 801                 nd6_setdefaultiface(0);
 802
 803         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 804                 /* refresh default router list */
 805                 bzero(&drany, sizeof(drany));
 806                 defrouter_delreq(&drany, 0);
 807                 defrouter_select();
 808         }
 809
 810         /*
 811          * Nuke neighbor cache entries for the ifp.
 812          * Note that rt->rt_ifp may not be the same as ifp,
 813          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 814          * nd6_rtrequest(), and ip6_input().
 815          */
 816         ln = llinfo_nd6.ln_next;
 817         while (ln && ln != &llinfo_nd6) {
 818                 struct rtentry *rt;
 819                 struct sockaddr_dl *sdl;
 820
 821                 nln = ln->ln_next;
 822                 rt = ln->ln_rt;
 823                 if (rt && rt->rt_gateway &&
 824                     rt->rt_gateway->sa_family == AF_LINK) {
 825                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 826                         if (sdl->sdl_index == ifp->if_index)
 827                                 nln = nd6_free(rt);
 828                 }
 829                 ln = nln;
 830         }
 831 }
 832
 833 struct rtentry *
 834 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 835 {
 836         struct rtentry *rt;
 837         struct sockaddr_in6 sin6;
 838
 839         bzero(&sin6, sizeof(sin6));
 840         sin6.sin6_len = sizeof(struct sockaddr_in6);
 841         sin6.sin6_family = AF_INET6;
 842         sin6.sin6_addr = *addr6;
 843
 844         if (create)
 845                 rt = rtlookup((struct sockaddr *)&sin6);
 846         else
 847                 rt = rtpurelookup((struct sockaddr *)&sin6);
 848         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 849                 /*
 850                  * This is the case for the default route.
 851                  * If we want to create a neighbor cache for the address, we
 852                  * should free the route for the destination and allocate an
 853                  * interface route.
 854                  */
 855                 if (create) {
 856                         --rt->rt_refcnt;
 857                         rt = NULL;
 858                 }
 859         }
 860         if (!rt) {
 861                 if (create && ifp) {
 862                         int e;
 863
 864                         /*
 865                          * If no route is available and create is set,
 866                          * we allocate a host route for the destination
 867                          * and treat it like an interface route.
 868                          * This hack is necessary for a neighbor which can't
 869                          * be covered by our own prefix.
 870                          */
 871                         struct ifaddr *ifa;
 872
 873                         ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 874                         if (ifa == NULL)
 875                                 return (NULL);
 876
 877                         /*
 878                          * Create a new route.  RTF_LLINFO is necessary
 879                          * to create a Neighbor Cache entry for the
 880                          * destination in nd6_rtrequest which will be
 881                          * called in rtrequest via ifa->ifa_rtrequest.
 882                          */
 883                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 884                              ifa->ifa_addr, (struct sockaddr *)&all1_sa,
 885                              (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
 886                              ~RTF_CLONING, &rt)) != 0) {
 887                                 log(LOG_ERR,
 888                                     "nd6_lookup: failed to add route for a "
 889                                     "neighbor(%s), errno=%d\n",
 890                                     ip6_sprintf(addr6), e);
 891                         }
 892                         if (rt == NULL)
 893                                 return (NULL);
 894                         if (rt->rt_llinfo) {
 895                                 struct llinfo_nd6 *ln =
 896                                     (struct llinfo_nd6 *)rt->rt_llinfo;
 897
 898                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 899                         }
 900                 } else
 901                         return (NULL);
 902         }
 903         rt->rt_refcnt--;
 904
 905         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
 906                 if (create) {
 907                         log(LOG_DEBUG,
 908                             "nd6_lookup: failed to lookup %s (if = %s)\n",
 909                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 910                         /* xxx more logs... kazu */
 911                 }
 912                 return (NULL);
 913         }
 914         return (rt);
 915 }
 916
 917 static struct rtentry *
 918 nd6_neighbor_lookup(struct in6_addr *addr6, struct ifnet *ifp)
 919 {
 920         struct rtentry *rt;
 921         struct sockaddr_in6 sin6;
 922
 923         bzero(&sin6, sizeof(sin6));
 924         sin6.sin6_len = sizeof(struct sockaddr_in6);
 925         sin6.sin6_family = AF_INET6;
 926         sin6.sin6_addr = *addr6;
 927
 928         rt = rtpurelookup((struct sockaddr *)&sin6);
 929         if (rt == NULL)
 930                 return (NULL);
 931         rt->rt_refcnt--;
 932
 933         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
 934                 if (nd6_onlink_ns_rfc4861 &&
 935                     (ND6_RTENTRY_IS_LLCLONING(rt) ||    /* not cloned yet */
 936                      (rt->rt_parent != NULL &&          /* cloning */
 937                       ND6_RTENTRY_IS_LLCLONING(rt->rt_parent)))) {
 938                         /*
 939                          * If cloning ever happened or is happening,
 940                          * rtentry for addr6 would or will become a
 941                          * neighbor cache.
 942                          */
 943                 } else {
 944                         rt = NULL;
 945                 }
 946         }
 947         return (rt);
 948 }
 949
 950 /*
 951  * Detect if a given IPv6 address identifies a neighbor on a given link.
 952  * XXX: should take care of the destination of a p2p link?
 953  */
 954 int
 955 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 956 {
 957         struct ifaddr_container *ifac;
 958         int i;
 959
 960 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 961 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 962
 963         /*
 964          * A link-local address is always a neighbor.
 965          * XXX: we should use the sin6_scope_id field rather than the embedded
 966          * interface index.
 967          */
 968         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 969             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 970                 return (1);
 971
 972         /*
 973          * If the address matches one of our addresses,
 974          * it should be a neighbor.
 975          */
 976         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 977                 struct ifaddr *ifa = ifac->ifa;
 978
 979                 if (ifa->ifa_addr->sa_family != AF_INET6)
 980                         next: continue;
 981
 982                 for (i = 0; i < 4; i++) {
 983                         if ((IFADDR6(ifa).s6_addr32[i] ^
 984                              addr->sin6_addr.s6_addr32[i]) &
 985                             IFMASK6(ifa).s6_addr32[i])
 986                                 goto next;
 987                 }
 988                 return (1);
 989         }
 990
 991         /*
 992          * Even if the address matches none of our addresses, it might be
 993          * in the neighbor cache.
 994          */
 995         if (nd6_neighbor_lookup(&addr->sin6_addr, ifp) != NULL)
 996                 return (1);
 997
 998         return (0);
 999 #undef IFADDR6
1000 #undef IFMASK6
1001 }
1002
1003 /*
1004  * Free an nd6 llinfo entry.
1005  */
1006 struct llinfo_nd6 *
1007 nd6_free(struct rtentry *rt)
1008 {
1009         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1010         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
1011         struct nd_defrouter *dr;
1012
1013         /*
1014          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
1015          * even though it is not harmful, it was not really necessary.
1016          */
1017
1018         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
1019                 mtx_lock(&nd6_mtx);
1020                 dr = defrouter_lookup(
1021                     &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1022                     rt->rt_ifp);
1023
1024                 if (ln->ln_router || dr) {
1025                         /*
1026                          * rt6_flush must be called whether or not the neighbor
1027                          * is in the Default Router List.
1028                          * See a corresponding comment in nd6_na_input().
1029                          */
1030                         rt6_flush(&in6, rt->rt_ifp);
1031                 }
1032
1033                 if (dr) {
1034                         /*
1035                          * Unreachablity of a router might affect the default
1036                          * router selection and on-link detection of advertised
1037                          * prefixes.
1038                          */
1039
1040                         /*
1041                          * Temporarily fake the state to choose a new default
1042                          * router and to perform on-link determination of
1043                          * prefixes correctly.
1044                          * Below the state will be set correctly,
1045                          * or the entry itself will be deleted.
1046                          */
1047                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
1048
1049                         /*
1050                          * Since defrouter_select() does not affect the
1051                          * on-link determination and MIP6 needs the check
1052                          * before the default router selection, we perform
1053                          * the check now.
1054                          */
1055                         pfxlist_onlink_check();
1056
1057                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1058                                 /*
1059                                  * It is used as the current default router,
1060                                  * so we have to move it to the end of the
1061                                  * list and choose a new one.
1062                                  * XXX: it is not very efficient if this is
1063                                  *      the only router.
1064                                  */
1065                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1066                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1067
1068                                 defrouter_select();
1069                         }
1070                 }
1071                 mtx_unlock(&nd6_mtx);
1072         }
1073
1074         /*
1075          * Before deleting the entry, remember the next entry as the
1076          * return value.  We need this because pfxlist_onlink_check() above
1077          * might have freed other entries (particularly the old next entry) as
1078          * a side effect (XXX).
1079          */
1080         next = ln->ln_next;
1081
1082         /*
1083          * Detach the route from the routing tree and the list of neighbor
1084          * caches, and disable the route entry not to be used in already
1085          * cached routes.
1086          *
1087          * ND expiry happens under one big timer.
1088          * To avoid overflowing the route socket, don't report this.
1089          * Now that RTM_MISS is reported when an address is unresolvable
1090          * the benefit of reporting this deletion is questionable.
1091          */
1092         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1093
1094         return (next);
1095 }
1096
1097 /*
1098  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1099  *
1100  * XXX cost-effective metods?
1101  */
1102 void
1103 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1104 {
1105         struct llinfo_nd6 *ln;
1106
1107         /*
1108          * If the caller specified "rt", use that.  Otherwise, resolve the
1109          * routing table by supplied "dst6".
1110          */
1111         if (!rt) {
1112                 if (!dst6)
1113                         return;
1114                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1115                         return;
1116         }
1117
1118         if ((rt->rt_flags & RTF_GATEWAY) ||
1119             !(rt->rt_flags & RTF_LLINFO) ||
1120             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1121             rt->rt_gateway->sa_family != AF_LINK) {
1122                 /* This is not a host route. */
1123                 return;
1124         }
1125
1126         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1127         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1128                 return;
1129
1130         /*
1131          * if we get upper-layer reachability confirmation many times,
1132          * it is possible we have false information.
1133          */
1134         if (!force) {
1135                 ln->ln_byhint++;
1136                 if (ln->ln_byhint > nd6_maxnudhint)
1137                         return;
1138         }
1139
1140         ln->ln_state = ND6_LLINFO_REACHABLE;
1141         if (ln->ln_expire)
1142                 ln->ln_expire = time_uptime +
1143                         ND_IFINFO(rt->rt_ifp)->reachable;
1144 }
1145
1146 void
1147 nd6_rtrequest(int req, struct rtentry *rt)
1148 {
1149         struct sockaddr *gate = rt->rt_gateway;
1150         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1151         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1152         struct ifnet *ifp = rt->rt_ifp;
1153         struct ifaddr *ifa;
1154
1155         if ((rt->rt_flags & RTF_GATEWAY))
1156                 return;
1157
1158         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1159                 /*
1160                  * This is probably an interface direct route for a link
1161                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1162                  * We do not need special treatment below for such a route.
1163                  * Moreover, the RTF_LLINFO flag which would be set below
1164                  * would annoy the ndp(8) command.
1165                  */
1166                 return;
1167         }
1168
1169         if (req == RTM_RESOLVE &&
1170             (nd6_need_cache(ifp) == 0 || /* stf case */
1171              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1172                 /*
1173                  * FreeBSD and BSD/OS often make a cloned host route based
1174                  * on a less-specific route (e.g. the default route).
1175                  * If the less specific route does not have a "gateway"
1176                  * (this is the case when the route just goes to a p2p or an
1177                  * stf interface), we'll mistakenly make a neighbor cache for
1178                  * the host route, and will see strange neighbor solicitation
1179                  * for the corresponding destination.  In order to avoid the
1180                  * confusion, we check if the destination of the route is
1181                  * a neighbor in terms of neighbor discovery, and stop the
1182                  * process if not.  Additionally, we remove the LLINFO flag
1183                  * so that ndp(8) will not try to get the neighbor information
1184                  * of the destination.
1185                  */
1186                 rt->rt_flags &= ~RTF_LLINFO;
1187                 return;
1188         }
1189
1190         switch (req) {
1191         case RTM_ADD:
1192                 /*
1193                  * There is no backward compatibility :)
1194                  *
1195                  * if (!(rt->rt_flags & RTF_HOST) &&
1196                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1197                  *         rt->rt_flags |= RTF_CLONING;
1198                  */
1199                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1200                         /*
1201                          * Case 1: This route should come from
1202                          * a route to interface.  RTF_LLINFO flag is set
1203                          * for a host route whose destination should be
1204                          * treated as on-link.
1205                          */
1206                         rt_setgate(rt, rt_key(rt),
1207                                    (struct sockaddr *)&null_sdl);
1208                         gate = rt->rt_gateway;
1209                         SDL(gate)->sdl_type = ifp->if_type;
1210                         SDL(gate)->sdl_index = ifp->if_index;
1211                         if (ln)
1212                                 ln->ln_expire = time_uptime;
1213                         if (ln && ln->ln_expire == 0) {
1214                                 /* kludge for desktops */
1215                                 ln->ln_expire = 1;
1216                         }
1217                         if ((rt->rt_flags & RTF_CLONING))
1218                                 break;
1219                 }
1220                 /*
1221                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1222                  * We don't do that here since llinfo is not ready yet.
1223                  *
1224                  * There are also couple of other things to be discussed:
1225                  * - unsolicited NA code needs improvement beforehand
1226                  * - RFC2461 says we MAY send multicast unsolicited NA
1227                  *   (7.2.6 paragraph 4), however, it also says that we
1228                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1229                  *   we don't have anything like it right now.
1230                  *   note that the mechanism needs a mutual agreement
1231                  *   between proxies, which means that we need to implement
1232                  *   a new protocol, or a new kludge.
1233                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1234                  *   we need to check ip6forwarding before sending it.
1235                  *   (or should we allow proxy ND configuration only for
1236                  *   routers?  there's no mention about proxy ND from hosts)
1237                  */
1238 #if 0
1239                 /* XXX it does not work */
1240                 if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
1241                         nd6_na_output(ifp,
1242                               &SIN6(rt_key(rt))->sin6_addr,
1243                               &SIN6(rt_key(rt))->sin6_addr,
1244                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1245                               1, NULL);
1246                 }
1247 #endif
1248                 /* FALLTHROUGH */
1249         case RTM_RESOLVE:
1250                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1251                         /*
1252                          * Address resolution isn't necessary for a point to
1253                          * point link, so we can skip this test for a p2p link.
1254                          */
1255                         if (gate->sa_family != AF_LINK ||
1256                             gate->sa_len < sizeof(null_sdl)) {
1257                                 log(LOG_DEBUG,
1258                                     "nd6_rtrequest: bad gateway value: %s\n",
1259                                     if_name(ifp));
1260                                 break;
1261                         }
1262                         SDL(gate)->sdl_type = ifp->if_type;
1263                         SDL(gate)->sdl_index = ifp->if_index;
1264                 }
1265                 if (ln != NULL)
1266                         break;  /* This happens on a route change */
1267                 /*
1268                  * Case 2: This route may come from cloning, or a manual route
1269                  * add with a LL address.
1270                  */
1271                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1272                 rt->rt_llinfo = (caddr_t)ln;
1273                 if (!ln) {
1274                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1275                         break;
1276                 }
1277                 nd6_inuse++;
1278                 nd6_allocated++;
1279                 bzero(ln, sizeof(*ln));
1280                 ln->ln_rt = rt;
1281                 /* this is required for "ndp" command. - shin */
1282                 if (req == RTM_ADD) {
1283                         /*
1284                          * gate should have some valid AF_LINK entry,
1285                          * and ln->ln_expire should have some lifetime
1286                          * which is specified by ndp command.
1287                          */
1288                         ln->ln_state = ND6_LLINFO_REACHABLE;
1289                         ln->ln_byhint = 0;
1290                 } else {
1291                         /*
1292                          * When req == RTM_RESOLVE, rt is created and
1293                          * initialized in rtrequest(), so rt_expire is 0.
1294                          */
1295                         ln->ln_state = ND6_LLINFO_NOSTATE;
1296                         ln->ln_expire = time_uptime;
1297                 }
1298                 rt->rt_flags |= RTF_LLINFO;
1299                 ln->ln_next = llinfo_nd6.ln_next;
1300                 llinfo_nd6.ln_next = ln;
1301                 ln->ln_prev = &llinfo_nd6;
1302                 ln->ln_next->ln_prev = ln;
1303
1304                 /*
1305                  * check if rt_key(rt) is one of my address assigned
1306                  * to the interface.
1307                  */
1308                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1309                     &SIN6(rt_key(rt))->sin6_addr);
1310                 if (ifa) {
1311                         caddr_t macp = nd6_ifptomac(ifp);
1312                         ln->ln_expire = 0;
1313                         ln->ln_state = ND6_LLINFO_REACHABLE;
1314                         ln->ln_byhint = 0;
1315                         if (macp) {
1316                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1317                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1318                         }
1319                         if (nd6_useloopback) {
1320                                 rt->rt_ifp = loif;      /* XXX */
1321                                 /*
1322                                  * Make sure rt_ifa be equal to the ifaddr
1323                                  * corresponding to the address.
1324                                  * We need this because when we refer
1325                                  * rt_ifa->ia6_flags in ip6_input, we assume
1326                                  * that the rt_ifa points to the address instead
1327                                  * of the loopback address.
1328                                  */
1329                                 if (ifa != rt->rt_ifa) {
1330                                         IFAFREE(rt->rt_ifa);
1331                                         IFAREF(ifa);
1332                                         rt->rt_ifa = ifa;
1333                                 }
1334                         }
1335                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1336                         ln->ln_expire = 0;
1337                         ln->ln_state = ND6_LLINFO_REACHABLE;
1338                         ln->ln_byhint = 0;
1339
1340                         /*
1341                          * Join solicited node multicast for proxy ND, and only
1342                          * join it once on cpu0.
1343                          */
1344                         if ((ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1345                                 struct in6_addr llsol;
1346                                 int error;
1347
1348                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1349                                 llsol.s6_addr16[0] = htons(0xff02);
1350                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1351                                 llsol.s6_addr32[1] = 0;
1352                                 llsol.s6_addr32[2] = htonl(1);
1353                                 llsol.s6_addr8[12] = 0xff;
1354
1355                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1356                                         nd6log((LOG_ERR, "%s: failed to join "
1357                                             "%s (errno=%d)\n", if_name(ifp),
1358                                             ip6_sprintf(&llsol), error));
1359                                 }
1360                         }
1361                 }
1362                 break;
1363
1364         case RTM_DELETE:
1365                 if (!ln)
1366                         break;
1367                 /*
1368                  * Leave from solicited node multicast for proxy ND, and only
1369                  * leave it once on cpu0 (since we joined it once on cpu0).
1370                  */
1371                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1372                     (ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1373                         struct in6_addr llsol;
1374                         struct in6_multi *in6m;
1375
1376                         llsol = SIN6(rt_key(rt))->sin6_addr;
1377                         llsol.s6_addr16[0] = htons(0xff02);
1378                         llsol.s6_addr16[1] = htons(ifp->if_index);
1379                         llsol.s6_addr32[1] = 0;
1380                         llsol.s6_addr32[2] = htonl(1);
1381                         llsol.s6_addr8[12] = 0xff;
1382
1383                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1384                         if (in6m)
1385                                 in6_delmulti(in6m);
1386                 }
1387                 nd6_inuse--;
1388                 ln->ln_next->ln_prev = ln->ln_prev;
1389                 ln->ln_prev->ln_next = ln->ln_next;
1390                 ln->ln_prev = NULL;
1391                 rt->rt_llinfo = 0;
1392                 rt->rt_flags &= ~RTF_LLINFO;
1393                 if (ln->ln_hold)
1394                         m_freem(ln->ln_hold);
1395                 Free((caddr_t)ln);
1396         }
1397 }
1398
1399 int
1400 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1401 {
1402         struct in6_drlist *drl = (struct in6_drlist *)data;
1403         struct in6_prlist *prl = (struct in6_prlist *)data;
1404         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1405         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1406         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1407         struct nd_defrouter *dr, any;
1408         struct nd_prefix *pr;
1409         struct rtentry *rt;
1410         int i = 0, error = 0;
1411
1412         switch (cmd) {
1413         case SIOCGDRLST_IN6:
1414                 /*
1415                  * obsolete API, use sysctl under net.inet6.icmp6
1416                  */
1417                 bzero(drl, sizeof(*drl));
1418                 mtx_lock(&nd6_mtx);
1419                 dr = TAILQ_FIRST(&nd_defrouter);
1420                 while (dr && i < DRLSTSIZ) {
1421                         drl->defrouter[i].rtaddr = dr->rtaddr;
1422                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1423                                 /* XXX: need to this hack for KAME stack */
1424                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1425                         } else
1426                                 log(LOG_ERR,
1427                                     "default router list contains a "
1428                                     "non-linklocal address(%s)\n",
1429                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1430
1431                         drl->defrouter[i].flags = dr->flags;
1432                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1433                         drl->defrouter[i].expire = dr->expire;
1434                         drl->defrouter[i].if_index = dr->ifp->if_index;
1435                         i++;
1436                         dr = TAILQ_NEXT(dr, dr_entry);
1437                 }
1438                 mtx_unlock(&nd6_mtx);
1439                 break;
1440         case SIOCGPRLST_IN6:
1441                 /*
1442                  * obsolete API, use sysctl under net.inet6.icmp6
1443                  */
1444                 /*
1445                  * XXX meaning of fields, especialy "raflags", is very
1446                  * differnet between RA prefix list and RR/static prefix list.
1447                  * how about separating ioctls into two?
1448                  */
1449                 bzero(prl, sizeof(*prl));
1450                 mtx_lock(&nd6_mtx);
1451                 pr = nd_prefix.lh_first;
1452                 while (pr && i < PRLSTSIZ) {
1453                         struct nd_pfxrouter *pfr;
1454                         int j;
1455
1456                         in6_embedscope(&prl->prefix[i].prefix,
1457                             &pr->ndpr_prefix, NULL, NULL);
1458                         prl->prefix[i].raflags = pr->ndpr_raf;
1459                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1460                         prl->prefix[i].vltime = pr->ndpr_vltime;
1461                         prl->prefix[i].pltime = pr->ndpr_pltime;
1462                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1463                         prl->prefix[i].expire = pr->ndpr_expire;
1464
1465                         pfr = pr->ndpr_advrtrs.lh_first;
1466                         j = 0;
1467                         while (pfr) {
1468                                 if (j < DRLSTSIZ) {
1469 #define RTRADDR prl->prefix[i].advrtr[j]
1470                                         RTRADDR = pfr->router->rtaddr;
1471                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1472                                                 /* XXX: hack for KAME */
1473                                                 RTRADDR.s6_addr16[1] = 0;
1474                                         } else
1475                                                 log(LOG_ERR,
1476                                                     "a router(%s) advertises "
1477                                                     "a prefix with "
1478                                                     "non-link local address\n",
1479                                                     ip6_sprintf(&RTRADDR));
1480 #undef RTRADDR
1481                                 }
1482                                 j++;
1483                                 pfr = pfr->pfr_next;
1484                         }
1485                         prl->prefix[i].advrtrs = j;
1486                         prl->prefix[i].origin = PR_ORIG_RA;
1487
1488                         i++;
1489                         pr = pr->ndpr_next;
1490                 }
1491                 mtx_unlock(&nd6_mtx);
1492
1493                 break;
1494         case OSIOCGIFINFO_IN6:
1495                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1496                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1497                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1498                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1499                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1500                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1501                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1502                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1503                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1504                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1505                 break;
1506         case SIOCGIFINFO_IN6:
1507                 ndi->ndi = *ND_IFINFO(ifp);
1508                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1509                 break;
1510         case SIOCSIFINFO_IN6:
1511                 /*
1512                  * used to change host variables from userland.
1513                  * intented for a use on router to reflect RA configurations.
1514                  */
1515                 /* 0 means 'unspecified' */
1516                 if (ndi->ndi.linkmtu != 0) {
1517                         if (ndi->ndi.linkmtu < IPV6_MMTU ||
1518                             ndi->ndi.linkmtu > IN6_LINKMTU(ifp)) {
1519                                 error = EINVAL;
1520                                 break;
1521                         }
1522                         ND_IFINFO(ifp)->linkmtu = ndi->ndi.linkmtu;
1523                 }
1524
1525                 if (ndi->ndi.basereachable != 0) {
1526                         int obasereachable = ND_IFINFO(ifp)->basereachable;
1527
1528                         ND_IFINFO(ifp)->basereachable = ndi->ndi.basereachable;
1529                         if (ndi->ndi.basereachable != obasereachable)
1530                                 ND_IFINFO(ifp)->reachable =
1531                                     ND_COMPUTE_RTIME(ndi->ndi.basereachable);
1532                 }
1533                 if (ndi->ndi.retrans != 0)
1534                         ND_IFINFO(ifp)->retrans = ndi->ndi.retrans;
1535                 if (ndi->ndi.chlim != 0)
1536                         ND_IFINFO(ifp)->chlim = ndi->ndi.chlim;
1537                 /* FALLTHROUGH */
1538         case SIOCSIFINFO_FLAGS:
1539                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1540                 break;
1541         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1542                 /* flush default router list */
1543                 /*
1544                  * xxx sumikawa: should not delete route if default
1545                  * route equals to the top of default router list
1546                  */
1547                 bzero(&any, sizeof(any));
1548                 defrouter_delreq(&any, 0);
1549                 defrouter_select();
1550                 /* xxx sumikawa: flush prefix list */
1551                 break;
1552         case SIOCSPFXFLUSH_IN6:
1553         {
1554                 /* flush all the prefix advertised by routers */
1555                 struct nd_prefix *pr, *next;
1556
1557                 mtx_lock(&nd6_mtx);
1558                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1559                         struct in6_ifaddr *ia, *ia_next;
1560
1561                         next = pr->ndpr_next;
1562
1563                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1564                                 continue; /* XXX */
1565
1566                         /* do we really have to remove addresses as well? */
1567                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1568                                 /* ia might be removed.  keep the next ptr. */
1569                                 ia_next = ia->ia_next;
1570
1571                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1572                                         continue;
1573
1574                                 if (ia->ia6_ndpr == pr)
1575                                         in6_purgeaddr(&ia->ia_ifa);
1576                         }
1577                         prelist_remove(pr);
1578                 }
1579                 mtx_unlock(&nd6_mtx);
1580                 break;
1581         }
1582         case SIOCSRTRFLUSH_IN6:
1583         {
1584                 /* flush all the default routers */
1585                 struct nd_defrouter *dr, *next;
1586
1587                 mtx_lock(&nd6_mtx);
1588                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1589                         /*
1590                          * The first entry of the list may be stored in
1591                          * the routing table, so we'll delete it later.
1592                          */
1593                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1594                                 next = TAILQ_NEXT(dr, dr_entry);
1595                                 defrtrlist_del(dr);
1596                         }
1597                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1598                 }
1599                 mtx_unlock(&nd6_mtx);
1600                 break;
1601         }
1602         case SIOCGNBRINFO_IN6:
1603         {
1604                 struct llinfo_nd6 *ln;
1605                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1606
1607                 /*
1608                  * XXX: KAME specific hack for scoped addresses
1609                  *      XXXX: for other scopes than link-local?
1610                  */
1611                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1612                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1613                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1614
1615                         if (*idp == 0)
1616                                 *idp = htons(ifp->if_index);
1617                 }
1618
1619                 mtx_lock(&nd6_mtx);
1620                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1621                         error = EINVAL;
1622                         mtx_unlock(&nd6_mtx);
1623                         break;
1624                 }
1625                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1626                 nbi->state = ln->ln_state;
1627                 nbi->asked = ln->ln_asked;
1628                 nbi->isrouter = ln->ln_router;
1629                 nbi->expire = ln->ln_expire;
1630                 mtx_unlock(&nd6_mtx);
1631
1632                 break;
1633         }
1634         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1635                 ndif->ifindex = nd6_defifindex;
1636                 break;
1637         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1638                 return (nd6_setdefaultiface(ndif->ifindex));
1639         }
1640         return (error);
1641 }
1642
1643 /*
1644  * Create neighbor cache entry and cache link-layer address,
1645  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1646  */
1647 struct rtentry *
1648 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1649                  int lladdrlen,
1650                  int type,      /* ICMP6 type */
1651                  int code       /* type dependent information */)
1652 {
1653         struct rtentry *rt = NULL;
1654         struct llinfo_nd6 *ln = NULL;
1655         int is_newentry;
1656         struct sockaddr_dl *sdl = NULL;
1657         int do_update;
1658         int olladdr;
1659         int llchange;
1660         int newstate = 0;
1661
1662         if (!ifp)
1663                 panic("ifp == NULL in nd6_cache_lladdr");
1664         if (!from)
1665                 panic("from == NULL in nd6_cache_lladdr");
1666
1667         /* nothing must be updated for unspecified address */
1668         if (IN6_IS_ADDR_UNSPECIFIED(from))
1669                 return NULL;
1670
1671         /*
1672          * Validation about ifp->if_addrlen and lladdrlen must be done in
1673          * the caller.
1674          *
1675          * XXX If the link does not have link-layer adderss, what should
1676          * we do? (ifp->if_addrlen == 0)
1677          * Spec says nothing in sections for RA, RS and NA.  There's small
1678          * description on it in NS section (RFC 2461 7.2.3).
1679          */
1680
1681         rt = nd6_lookup(from, 0, ifp);
1682         if (!rt) {
1683 #if 0
1684                 /* nothing must be done if there's no lladdr */
1685                 if (!lladdr || !lladdrlen)
1686                         return NULL;
1687 #endif
1688
1689                 rt = nd6_lookup(from, 1, ifp);
1690                 is_newentry = 1;
1691         } else {
1692                 /* do nothing if static ndp is set */
1693                 if (rt->rt_flags & RTF_STATIC)
1694                         return NULL;
1695                 is_newentry = 0;
1696         }
1697
1698         if (!rt)
1699                 return NULL;
1700         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1701 fail:
1702                 nd6_free(rt);
1703                 return NULL;
1704         }
1705         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1706         if (!ln)
1707                 goto fail;
1708         if (!rt->rt_gateway)
1709                 goto fail;
1710         if (rt->rt_gateway->sa_family != AF_LINK)
1711                 goto fail;
1712         sdl = SDL(rt->rt_gateway);
1713
1714         olladdr = (sdl->sdl_alen) ? 1 : 0;
1715         if (olladdr && lladdr) {
1716                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1717                         llchange = 1;
1718                 else
1719                         llchange = 0;
1720         } else
1721                 llchange = 0;
1722
1723         /*
1724          * newentry olladdr  lladdr  llchange   (*=record)
1725          *      0       n       n       --      (1)
1726          *      0       y       n       --      (2)
1727          *      0       n       y       --      (3) * STALE
1728          *      0       y       y       n       (4) *
1729          *      0       y       y       y       (5) * STALE
1730          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1731          *      1       --      y       --      (7) * STALE
1732          */
1733
1734         if (lladdr) {           /* (3-5) and (7) */
1735                 /*
1736                  * Record source link-layer address
1737                  * XXX is it dependent to ifp->if_type?
1738                  */
1739                 sdl->sdl_alen = ifp->if_addrlen;
1740                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1741         }
1742
1743         if (!is_newentry) {
1744                 if ((!olladdr && lladdr) ||             /* (3) */
1745                     (olladdr && lladdr && llchange)) {  /* (5) */
1746                         do_update = 1;
1747                         newstate = ND6_LLINFO_STALE;
1748                 } else {                                /* (1-2,4) */
1749                         do_update = 0;
1750                 }
1751         } else {
1752                 do_update = 1;
1753                 if (!lladdr)                            /* (6) */
1754                         newstate = ND6_LLINFO_NOSTATE;
1755                 else                                    /* (7) */
1756                         newstate = ND6_LLINFO_STALE;
1757         }
1758
1759         if (do_update) {
1760                 /*
1761                  * Update the state of the neighbor cache.
1762                  */
1763                 ln->ln_state = newstate;
1764
1765                 if (ln->ln_state == ND6_LLINFO_STALE) {
1766                         /*
1767                          * XXX: since nd6_output() below will cause
1768                          * state tansition to DELAY and reset the timer,
1769                          * we must set the timer now, although it is actually
1770                          * meaningless.
1771                          */
1772                         ln->ln_expire = time_uptime + nd6_gctimer;
1773
1774                         if (ln->ln_hold) {
1775                                 /*
1776                                  * we assume ifp is not a p2p here, so just
1777                                  * set the 2nd argument as the 1st one.
1778                                  */
1779                                 nd6_output(ifp, ifp, ln->ln_hold,
1780                                     (struct sockaddr_in6 *)rt_key(rt), rt);
1781                                 ln->ln_hold = NULL;
1782                         }
1783                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1784                         /* probe right away */
1785                         ln->ln_expire = time_uptime;
1786                 }
1787         }
1788
1789         /*
1790          * ICMP6 type dependent behavior.
1791          *
1792          * NS: clear IsRouter if new entry
1793          * RS: clear IsRouter
1794          * RA: set IsRouter if there's lladdr
1795          * redir: clear IsRouter if new entry
1796          *
1797          * RA case, (1):
1798          * The spec says that we must set IsRouter in the following cases:
1799          * - If lladdr exist, set IsRouter.  This means (1-5).
1800          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1801          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1802          * A quetion arises for (1) case.  (1) case has no lladdr in the
1803          * neighbor cache, this is similar to (6).
1804          * This case is rare but we figured that we MUST NOT set IsRouter.
1805          *
1806          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1807          *                                                      D R
1808          *      0       n       n       --      (1)     c   ?     s
1809          *      0       y       n       --      (2)     c   s     s
1810          *      0       n       y       --      (3)     c   s     s
1811          *      0       y       y       n       (4)     c   s     s
1812          *      0       y       y       y       (5)     c   s     s
1813          *      1       --      n       --      (6) c   c       c s
1814          *      1       --      y       --      (7) c   c   s   c s
1815          *
1816          *                                      (c=clear s=set)
1817          */
1818         switch (type & 0xff) {
1819         case ND_NEIGHBOR_SOLICIT:
1820                 /*
1821                  * New entry must have is_router flag cleared.
1822                  */
1823                 if (is_newentry)        /* (6-7) */
1824                         ln->ln_router = 0;
1825                 break;
1826         case ND_REDIRECT:
1827                 /*
1828                  * If the icmp is a redirect to a better router, always set the
1829                  * is_router flag.  Otherwise, if the entry is newly created,
1830                  * clear the flag.  [RFC 2461, sec 8.3]
1831                  */
1832                 if (code == ND_REDIRECT_ROUTER)
1833                         ln->ln_router = 1;
1834                 else if (is_newentry) /* (6-7) */
1835                         ln->ln_router = 0;
1836                 break;
1837         case ND_ROUTER_SOLICIT:
1838                 /*
1839                  * is_router flag must always be cleared.
1840                  */
1841                 ln->ln_router = 0;
1842                 break;
1843         case ND_ROUTER_ADVERT:
1844                 /*
1845                  * Mark an entry with lladdr as a router.
1846                  */
1847                 if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
1848                     (is_newentry && lladdr)) {                  /* (7) */
1849                         ln->ln_router = 1;
1850                 }
1851                 break;
1852         }
1853
1854         if (llchange || lladdr)
1855                 rt_rtmsg(llchange ? RTM_CHANGE : RTM_ADD, rt, rt->rt_ifp, 0);
1856
1857         /*
1858          * When the link-layer address of a router changes, select the
1859          * best router again.  In particular, when the neighbor entry is newly
1860          * created, it might affect the selection policy.
1861          * Question: can we restrict the first condition to the "is_newentry"
1862          * case?
1863          * XXX: when we hear an RA from a new router with the link-layer
1864          * address option, defrouter_select() is called twice, since
1865          * defrtrlist_update called the function as well.  However, I believe
1866          * we can compromise the overhead, since it only happens the first
1867          * time.
1868          * XXX: although defrouter_select() should not have a bad effect
1869          * for those are not autoconfigured hosts, we explicitly avoid such
1870          * cases for safety.
1871          */
1872         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1873                 defrouter_select();
1874
1875         return rt;
1876 }
1877
1878 static void
1879 nd6_slowtimo(void *arg __unused)
1880 {
1881         struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1882
1883         KASSERT(mycpuid == 0, ("not on cpu0"));
1884         crit_enter();
1885         if (lmsg->ms_flags & MSGF_DONE)
1886                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1887         crit_exit();
1888 }
1889
1890 static void
1891 nd6_slowtimo_dispatch(netmsg_t nmsg)
1892 {
1893         const struct ifnet_array *arr;
1894         struct nd_ifinfo *nd6if;
1895         int i;
1896
1897         ASSERT_NETISR0;
1898
1899         crit_enter();
1900         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1901         crit_exit();
1902
1903         arr = ifnet_array_get();
1904
1905         mtx_lock(&nd6_mtx);
1906         for (i = 0; i < arr->ifnet_count; ++i) {
1907                 struct ifnet *ifp = arr->ifnet_arr[i];
1908
1909                 if (ifp->if_afdata[AF_INET6] == NULL)
1910                         continue;
1911                 nd6if = ND_IFINFO(ifp);
1912                 if (nd6if->basereachable && /* already initialized */
1913                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1914                         /*
1915                          * Since reachable time rarely changes by router
1916                          * advertisements, we SHOULD insure that a new random
1917                          * value gets recomputed at least once every few hours.
1918                          * (RFC 2461, 6.3.4)
1919                          */
1920                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1921                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1922                 }
1923         }
1924         mtx_unlock(&nd6_mtx);
1925
1926         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1927             nd6_slowtimo, NULL);
1928 }
1929
1930 int
1931 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1932            struct sockaddr_in6 *dst, struct rtentry *rt)
1933 {
1934         int error;
1935
1936         if (ifp->if_flags & IFF_LOOPBACK)
1937                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1938         else
1939                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1940         return error;
1941 }
1942
1943 int
1944 nd6_resolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1945     struct sockaddr *dst0, u_char *desten)
1946 {
1947         struct sockaddr_in6 *dst = SIN6(dst0);
1948         struct rtentry *rt = NULL;
1949         struct llinfo_nd6 *ln = NULL;
1950         int error;
1951
1952         if (m->m_flags & M_MCAST) {
1953                 switch (ifp->if_type) {
1954                 case IFT_ETHER:
1955 #ifdef IFT_L2VLAN
1956                 case IFT_L2VLAN:
1957 #endif
1958 #ifdef IFT_IEEE80211
1959                 case IFT_IEEE80211:
1960 #endif
1961                         ETHER_MAP_IPV6_MULTICAST(&dst->sin6_addr,
1962                                                  desten);
1963                         return 0;
1964                 case IFT_IEEE1394:
1965                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1966                         return 0;
1967                 default:
1968                         error = EAFNOSUPPORT;
1969                         goto bad;
1970                 }
1971         }
1972
1973         if (rt0 != NULL) {
1974                 error = rt_llroute(dst0, rt0, &rt);
1975                 if (error != 0)
1976                         goto bad;
1977                 ln = rt->rt_llinfo;
1978         }
1979
1980         /*
1981          * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1982          * the condition below is not very efficient.  But we believe
1983          * it is tolerable, because this should be a rare case.
1984          */
1985         if (ln == NULL && nd6_is_addr_neighbor(dst, ifp)) {
1986                 rt = nd6_lookup(&dst->sin6_addr, 1, ifp);
1987                 if (rt != NULL)
1988                         ln = rt->rt_llinfo;
1989         }
1990
1991         if (ln == NULL || rt == NULL) {
1992                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1993                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1994                         log(LOG_DEBUG,
1995                             "nd6_output: can't allocate llinfo for %s "
1996                             "(ln=%p, rt=%p)\n",
1997                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1998                         error = ENOBUFS;
1999                         goto bad;
2000                 }
2001                 return 0;
2002         }
2003
2004         /* We don't have to do link-layer address resolution on a p2p link. */
2005         if ((ifp->if_flags & IFF_POINTOPOINT) &&
2006             ln->ln_state < ND6_LLINFO_REACHABLE) {
2007                 ln->ln_state = ND6_LLINFO_STALE;
2008                 ln->ln_expire = time_uptime + nd6_gctimer;
2009         }
2010
2011         /*
2012          * The first time we send a packet to a neighbor whose entry is
2013          * STALE, we have to change the state to DELAY and a sets a timer to
2014          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2015          * neighbor unreachability detection on expiration.
2016          * (RFC 2461 7.3.3)
2017          */
2018         if (ln->ln_state == ND6_LLINFO_STALE) {
2019                 ln->ln_asked = 0;
2020                 ln->ln_state = ND6_LLINFO_DELAY;
2021                 ln->ln_expire = time_uptime + nd6_delay;
2022         }
2023
2024         /*
2025          * If the neighbor cache entry has a state other than INCOMPLETE
2026          * (i.e. its link-layer address is already resolved), return it.
2027          */
2028         if (ln->ln_state > ND6_LLINFO_INCOMPLETE) {
2029                 struct sockaddr_dl *sdl = SDL(rt->rt_gateway);
2030
2031                 KKASSERT(sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0);
2032                 bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2033                 return 0;
2034         }
2035
2036         /*
2037          * There is a neighbor cache entry, but no ethernet address
2038          * response yet.  Replace the held mbuf (if any) with this
2039          * latest one.
2040          */
2041         if (ln->ln_hold)
2042                 m_freem(ln->ln_hold);
2043         ln->ln_hold = m;
2044
2045         /*
2046          * This code conforms to the rate-limiting rule described in Section
2047          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2048          * an NS below.
2049          */
2050         if (ln->ln_state == ND6_LLINFO_NOSTATE ||
2051             ln->ln_state == ND6_LLINFO_WAITDELETE) {
2052                 /*
2053                  * This neighbor cache entry was just created; change its
2054                  * state to INCOMPLETE and start its life cycle.
2055                  *
2056                  * We force an NS output below by setting ln_expire to 1
2057                  * (nd6_rtrequest() could set it to the current time_uptime)
2058                  * and zeroing out ln_asked (XXX this may not be necessary).
2059                  */
2060                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
2061                 ln->ln_expire = 1;
2062                 ln->ln_asked = 0;
2063         }
2064         if (ln->ln_expire && ln->ln_expire < time_uptime && ln->ln_asked == 0) {
2065                 ln->ln_asked++;
2066                 ln->ln_expire = time_uptime + ND_IFINFO(ifp)->retrans / 1000;
2067                 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2068         }
2069
2070         if (ln->ln_asked >= nd6_mmaxtries)
2071                 return (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
2072                     EHOSTUNREACH : EHOSTDOWN;
2073         return EWOULDBLOCK;
2074
2075 bad:
2076         m_freem(m);
2077         return error;
2078 }
2079
2080 int
2081 nd6_need_cache(struct ifnet *ifp)
2082 {
2083         /*
2084          * XXX: we currently do not make neighbor cache on any interface
2085          * other than Ethernet and GIF.
2086          *
2087          * RFC2893 says:
2088          * - unidirectional tunnels needs no ND
2089          */
2090         switch (ifp->if_type) {
2091         case IFT_ETHER:
2092         case IFT_IEEE1394:
2093 #ifdef IFT_L2VLAN
2094         case IFT_L2VLAN:
2095 #endif
2096 #ifdef IFT_IEEE80211
2097         case IFT_IEEE80211:
2098 #endif
2099 #ifdef IFT_CARP
2100         case IFT_CARP:
2101 #endif
2102         case IFT_GIF:           /* XXX need more cases? */
2103                 return (1);
2104         default:
2105                 return (0);
2106         }
2107 }
2108
2109 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2110 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2111 #ifdef SYSCTL_DECL
2112 SYSCTL_DECL(_net_inet6_icmp6);
2113 #endif
2114 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2115         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2116 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2117         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2118
2119 static int
2120 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2121 {
2122         int error;
2123         char buf[1024];
2124         struct in6_defrouter *d, *de;
2125         struct nd_defrouter *dr;
2126
2127         if (req->newptr)
2128                 return EPERM;
2129         error = 0;
2130
2131         for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2132              dr = TAILQ_NEXT(dr, dr_entry)) {
2133                 d = (struct in6_defrouter *)buf;
2134                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2135
2136                 if (d + 1 <= de) {
2137                         bzero(d, sizeof(*d));
2138                         d->rtaddr.sin6_family = AF_INET6;
2139                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2140                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2141                             dr->ifp) != 0)
2142                                 log(LOG_ERR,
2143                                     "scope error in "
2144                                     "default router list (%s)\n",
2145                                     ip6_sprintf(&dr->rtaddr));
2146                         d->flags = dr->flags;
2147                         d->rtlifetime = dr->rtlifetime;
2148                         d->expire = dr->expire;
2149                         d->if_index = dr->ifp->if_index;
2150                 } else
2151                         panic("buffer too short");
2152
2153                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2154                 if (error)
2155                         break;
2156         }
2157         return error;
2158 }
2159
2160 static int
2161 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2162 {
2163         int error;
2164         char buf[1024];
2165         struct in6_prefix *p, *pe;
2166         struct nd_prefix *pr;
2167
2168         if (req->newptr)
2169                 return EPERM;
2170         error = 0;
2171
2172         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2173                 u_short advrtrs;
2174                 size_t advance;
2175                 struct sockaddr_in6 *sin6, *s6;
2176                 struct nd_pfxrouter *pfr;
2177
2178                 p = (struct in6_prefix *)buf;
2179                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2180
2181                 if (p + 1 <= pe) {
2182                         bzero(p, sizeof(*p));
2183                         sin6 = (struct sockaddr_in6 *)(p + 1);
2184
2185                         p->prefix = pr->ndpr_prefix;
2186                         if (in6_recoverscope(&p->prefix,
2187                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2188                                 log(LOG_ERR,
2189                                     "scope error in prefix list (%s)\n",
2190                                     ip6_sprintf(&p->prefix.sin6_addr));
2191                         p->raflags = pr->ndpr_raf;
2192                         p->prefixlen = pr->ndpr_plen;
2193                         p->vltime = pr->ndpr_vltime;
2194                         p->pltime = pr->ndpr_pltime;
2195                         p->if_index = pr->ndpr_ifp->if_index;
2196                         p->expire = pr->ndpr_expire;
2197                         p->refcnt = pr->ndpr_refcnt;
2198                         p->flags = pr->ndpr_stateflags;
2199                         p->origin = PR_ORIG_RA;
2200                         advrtrs = 0;
2201                         for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2202                              pfr = pfr->pfr_next) {
2203                                 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2204                                         advrtrs++;
2205                                         continue;
2206                                 }
2207                                 s6 = &sin6[advrtrs];
2208                                 bzero(s6, sizeof(*s6));
2209                                 s6->sin6_family = AF_INET6;
2210                                 s6->sin6_len = sizeof(*sin6);
2211                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2212                                     pfr->router->ifp) != 0)
2213                                         log(LOG_ERR,
2214                                             "scope error in "
2215                                             "prefix list (%s)\n",
2216                                             ip6_sprintf(&pfr->router->rtaddr));
2217                                 advrtrs++;
2218                         }
2219                         p->advrtrs = advrtrs;
2220                 } else {
2221                         panic("buffer too short");
2222                 }
2223
2224                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2225                 error = SYSCTL_OUT(req, buf, advance);
2226                 if (error)
2227                         break;
2228         }
2229         return error;
2230 }