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_INCOMPLETE:
 479                         if (ln->ln_asked < nd6_mmaxtries) {
 480                                 ln->ln_asked++;
 481                                 ln->ln_expire = time_uptime +
 482                                         ND_IFINFO(ifp)->retrans / 1000;
 483                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 484                                         ln, 0);
 485                         } else {
 486                                 struct mbuf *m = ln->ln_hold;
 487                                 if (m) {
 488                                         if (rt->rt_ifp) {
 489                                                 /*
 490                                                  * Fake rcvif to make ICMP error
 491                                                  * more helpful in diagnosing
 492                                                  * for the receiver.
 493                                                  * XXX: should we consider
 494                                                  * older rcvif?
 495                                                  */
 496                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 497                                         }
 498                                         icmp6_error(m, ICMP6_DST_UNREACH,
 499                                                     ICMP6_DST_UNREACH_ADDR, 0);
 500                                         ln->ln_hold = NULL;
 501                                 }
 502                                 next = nd6_free(rt);
 503                         }
 504                         break;
 505                 case ND6_LLINFO_REACHABLE:
 506                         if (ln->ln_expire) {
 507                                 ln->ln_state = ND6_LLINFO_STALE;
 508                                 ln->ln_expire = time_uptime + nd6_gctimer;
 509                         }
 510                         break;
 511
 512                 case ND6_LLINFO_STALE:
 513                         /* Garbage Collection(RFC 2461 5.3) */
 514                         if (ln->ln_expire)
 515                                 next = nd6_free(rt);
 516                         break;
 517
 518                 case ND6_LLINFO_DELAY:
 519                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 520                                 /* We need NUD */
 521                                 ln->ln_asked = 1;
 522                                 ln->ln_state = ND6_LLINFO_PROBE;
 523                                 ln->ln_expire = time_uptime +
 524                                         ndi->retrans / 1000;
 525                                 nd6_ns_output(ifp, &dst->sin6_addr,
 526                                               &dst->sin6_addr,
 527                                               ln, 0);
 528                         } else {
 529                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 530                                 ln->ln_expire = time_uptime + nd6_gctimer;
 531                         }
 532                         break;
 533                 case ND6_LLINFO_PROBE:
 534                         if (ln->ln_asked < nd6_umaxtries) {
 535                                 ln->ln_asked++;
 536                                 ln->ln_expire = time_uptime +
 537                                         ND_IFINFO(ifp)->retrans / 1000;
 538                                 nd6_ns_output(ifp, &dst->sin6_addr,
 539                                                &dst->sin6_addr, ln, 0);
 540                         } else {
 541                                 next = nd6_free(rt);
 542                         }
 543                         break;
 544                 }
 545                 ln = next;
 546         }
 547
 548         /* expire default router list */
 549         dr = TAILQ_FIRST(&nd_defrouter);
 550         while (dr) {
 551                 if (dr->expire && dr->expire < time_uptime) {
 552                         struct nd_defrouter *t;
 553                         t = TAILQ_NEXT(dr, dr_entry);
 554                         defrtrlist_del(dr);
 555                         dr = t;
 556                 } else {
 557                         dr = TAILQ_NEXT(dr, dr_entry);
 558                 }
 559         }
 560
 561         /*
 562          * expire interface addresses.
 563          * in the past the loop was inside prefix expiry processing.
 564          * However, from a stricter speci-confrmance standpoint, we should
 565          * rather separate address lifetimes and prefix lifetimes.
 566          */
 567 addrloop:
 568         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 569                 nia6 = ia6->ia_next;
 570                 /* check address lifetime */
 571                 if (IFA6_IS_INVALID(ia6)) {
 572                         int regen = 0;
 573
 574                         /*
 575                          * If the expiring address is temporary, try
 576                          * regenerating a new one.  This would be useful when
 577                          * we suspended a laptop PC, then turned it on after a
 578                          * period that could invalidate all temporary
 579                          * addresses.  Although we may have to restart the
 580                          * loop (see below), it must be after purging the
 581                          * address.  Otherwise, we'd see an infinite loop of
 582                          * regeneration.
 583                          */
 584                         if (ip6_use_tempaddr &&
 585                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 586                                 if (regen_tmpaddr(ia6) == 0)
 587                                         regen = 1;
 588                         }
 589
 590                         in6_purgeaddr(&ia6->ia_ifa);
 591
 592                         if (regen)
 593                                 goto addrloop; /* XXX: see below */
 594                 }
 595                 if (IFA6_IS_DEPRECATED(ia6)) {
 596                         int oldflags = ia6->ia6_flags;
 597
 598                         if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
 599                                 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 600                                 in6_newaddrmsg((struct ifaddr *)ia6);
 601                         }
 602
 603                         /*
 604                          * If a temporary address has just become deprecated,
 605                          * regenerate a new one if possible.
 606                          */
 607                         if (ip6_use_tempaddr &&
 608                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 609                             !(oldflags & IN6_IFF_DEPRECATED)) {
 610
 611                                 if (regen_tmpaddr(ia6) == 0) {
 612                                         /*
 613                                          * A new temporary address is
 614                                          * generated.
 615                                          * XXX: this means the address chain
 616                                          * has changed while we are still in
 617                                          * the loop.  Although the change
 618                                          * would not cause disaster (because
 619                                          * it's not a deletion, but an
 620                                          * addition,) we'd rather restart the
 621                                          * loop just for safety.  Or does this
 622                                          * significantly reduce performance??
 623                                          */
 624                                         goto addrloop;
 625                                 }
 626                         }
 627                 } else {
 628                         /*
 629                          * A new RA might have made a deprecated address
 630                          * preferred.
 631                          */
 632                         if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
 633                                 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 634                                 in6_newaddrmsg((struct ifaddr *)ia6);
 635                         }
 636                 }
 637         }
 638
 639         /* expire prefix list */
 640         pr = nd_prefix.lh_first;
 641         while (pr) {
 642                 /*
 643                  * check prefix lifetime.
 644                  * since pltime is just for autoconf, pltime processing for
 645                  * prefix is not necessary.
 646                  */
 647                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
 648                         struct nd_prefix *t;
 649                         t = pr->ndpr_next;
 650
 651                         /*
 652                          * address expiration and prefix expiration are
 653                          * separate.  NEVER perform in6_purgeaddr here.
 654                          */
 655
 656                         prelist_remove(pr);
 657                         pr = t;
 658                 } else
 659                         pr = pr->ndpr_next;
 660         }
 661
 662         mtx_unlock(&nd6_mtx);
 663
 664         callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
 665 }
 666
 667 static void
 668 nd6_timer(void *arg __unused)
 669 {
 670         struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
 671
 672         KASSERT(mycpuid == 0, ("not on cpu0"));
 673         crit_enter();
 674         if (lmsg->ms_flags & MSGF_DONE)
 675                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
 676         crit_exit();
 677 }
 678
 679 void
 680 nd6_timer_init(void)
 681 {
 682         callout_init_mp(&nd6_timer_ch);
 683         netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
 684             MSGF_PRIORITY, nd6_timer_dispatch);
 685         callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
 686 }
 687
 688 static int
 689 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 690                                          address */
 691 {
 692         struct ifaddr_container *ifac;
 693         struct ifnet *ifp;
 694         struct in6_ifaddr *public_ifa6 = NULL;
 695
 696         ifp = ia6->ia_ifa.ifa_ifp;
 697         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 698                 struct ifaddr *ifa = ifac->ifa;
 699                 struct in6_ifaddr *it6;
 700
 701                 if (ifa->ifa_addr->sa_family != AF_INET6)
 702                         continue;
 703
 704                 it6 = (struct in6_ifaddr *)ifa;
 705
 706                 /* ignore no autoconf addresses. */
 707                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 708                         continue;
 709
 710                 /* ignore autoconf addresses with different prefixes. */
 711                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 712                         continue;
 713
 714                 /*
 715                  * Now we are looking at an autoconf address with the same
 716                  * prefix as ours.  If the address is temporary and is still
 717                  * preferred, do not create another one.  It would be rare, but
 718                  * could happen, for example, when we resume a laptop PC after
 719                  * a long period.
 720                  */
 721                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 722                     !IFA6_IS_DEPRECATED(it6)) {
 723                         public_ifa6 = NULL;
 724                         break;
 725                 }
 726
 727                 /*
 728                  * This is a public autoconf address that has the same prefix
 729                  * as ours.  If it is preferred, keep it.  We can't break the
 730                  * loop here, because there may be a still-preferred temporary
 731                  * address with the prefix.
 732                  */
 733                 if (!IFA6_IS_DEPRECATED(it6))
 734                     public_ifa6 = it6;
 735         }
 736
 737         if (public_ifa6 != NULL) {
 738                 int e;
 739
 740                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 741                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 742                             " tmp addr,errno=%d\n", e);
 743                         return (-1);
 744                 }
 745                 return (0);
 746         }
 747
 748         return (-1);
 749 }
 750
 751 /*
 752  * Nuke neighbor cache/prefix/default router management table, right before
 753  * ifp goes away.
 754  */
 755 void
 756 nd6_purge(struct ifnet *ifp)
 757 {
 758         struct llinfo_nd6 *ln, *nln;
 759         struct nd_defrouter *dr, *ndr, drany;
 760         struct nd_prefix *pr, *npr;
 761
 762         /* Nuke default router list entries toward ifp */
 763         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 764                 /*
 765                  * The first entry of the list may be stored in
 766                  * the routing table, so we'll delete it later.
 767                  */
 768                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 769                         ndr = TAILQ_NEXT(dr, dr_entry);
 770                         if (dr->ifp == ifp)
 771                                 defrtrlist_del(dr);
 772                 }
 773                 dr = TAILQ_FIRST(&nd_defrouter);
 774                 if (dr->ifp == ifp)
 775                         defrtrlist_del(dr);
 776         }
 777
 778         /* Nuke prefix list entries toward ifp */
 779         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 780                 npr = pr->ndpr_next;
 781                 if (pr->ndpr_ifp == ifp) {
 782                         /*
 783                          * Previously, pr->ndpr_addr is removed as well,
 784                          * but I strongly believe we don't have to do it.
 785                          * nd6_purge() is only called from in6_ifdetach(),
 786                          * which removes all the associated interface addresses
 787                          * by itself.
 788                          * (jinmei@kame.net 20010129)
 789                          */
 790                         prelist_remove(pr);
 791                 }
 792         }
 793
 794         /* cancel default outgoing interface setting */
 795         if (nd6_defifindex == ifp->if_index)
 796                 nd6_setdefaultiface(0);
 797
 798         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 799                 /* refresh default router list */
 800                 bzero(&drany, sizeof(drany));
 801                 defrouter_delreq(&drany, 0);
 802                 defrouter_select();
 803         }
 804
 805         /*
 806          * Nuke neighbor cache entries for the ifp.
 807          * Note that rt->rt_ifp may not be the same as ifp,
 808          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 809          * nd6_rtrequest(), and ip6_input().
 810          */
 811         ln = llinfo_nd6.ln_next;
 812         while (ln && ln != &llinfo_nd6) {
 813                 struct rtentry *rt;
 814                 struct sockaddr_dl *sdl;
 815
 816                 nln = ln->ln_next;
 817                 rt = ln->ln_rt;
 818                 if (rt && rt->rt_gateway &&
 819                     rt->rt_gateway->sa_family == AF_LINK) {
 820                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 821                         if (sdl->sdl_index == ifp->if_index)
 822                                 nln = nd6_free(rt);
 823                 }
 824                 ln = nln;
 825         }
 826 }
 827
 828 struct rtentry *
 829 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 830 {
 831         struct rtentry *rt;
 832         struct sockaddr_in6 sin6;
 833
 834         bzero(&sin6, sizeof(sin6));
 835         sin6.sin6_len = sizeof(struct sockaddr_in6);
 836         sin6.sin6_family = AF_INET6;
 837         sin6.sin6_addr = *addr6;
 838
 839         if (create)
 840                 rt = rtlookup((struct sockaddr *)&sin6);
 841         else
 842                 rt = rtpurelookup((struct sockaddr *)&sin6);
 843         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 844                 /*
 845                  * This is the case for the default route.
 846                  * If we want to create a neighbor cache for the address, we
 847                  * should free the route for the destination and allocate an
 848                  * interface route.
 849                  */
 850                 if (create) {
 851                         --rt->rt_refcnt;
 852                         rt = NULL;
 853                 }
 854         }
 855         if (!rt) {
 856                 if (create && ifp) {
 857                         int e;
 858
 859                         /*
 860                          * If no route is available and create is set,
 861                          * we allocate a host route for the destination
 862                          * and treat it like an interface route.
 863                          * This hack is necessary for a neighbor which can't
 864                          * be covered by our own prefix.
 865                          */
 866                         struct ifaddr *ifa;
 867
 868                         ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 869                         if (ifa == NULL)
 870                                 return (NULL);
 871
 872                         /*
 873                          * Create a new route.  RTF_LLINFO is necessary
 874                          * to create a Neighbor Cache entry for the
 875                          * destination in nd6_rtrequest which will be
 876                          * called in rtrequest via ifa->ifa_rtrequest.
 877                          */
 878                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 879                              ifa->ifa_addr, (struct sockaddr *)&all1_sa,
 880                              (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
 881                              ~RTF_CLONING, &rt)) != 0) {
 882                                 log(LOG_ERR,
 883                                     "nd6_lookup: failed to add route for a "
 884                                     "neighbor(%s), errno=%d\n",
 885                                     ip6_sprintf(addr6), e);
 886                         }
 887                         if (rt == NULL)
 888                                 return (NULL);
 889                         if (rt->rt_llinfo) {
 890                                 struct llinfo_nd6 *ln =
 891                                     (struct llinfo_nd6 *)rt->rt_llinfo;
 892
 893                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 894                         }
 895                 } else
 896                         return (NULL);
 897         }
 898         rt->rt_refcnt--;
 899
 900         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
 901                 if (create) {
 902                         log(LOG_DEBUG,
 903                             "nd6_lookup: failed to lookup %s (if = %s)\n",
 904                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 905                         /* xxx more logs... kazu */
 906                 }
 907                 return (NULL);
 908         }
 909         return (rt);
 910 }
 911
 912 static struct rtentry *
 913 nd6_neighbor_lookup(struct in6_addr *addr6, struct ifnet *ifp)
 914 {
 915         struct rtentry *rt;
 916         struct sockaddr_in6 sin6;
 917
 918         bzero(&sin6, sizeof(sin6));
 919         sin6.sin6_len = sizeof(struct sockaddr_in6);
 920         sin6.sin6_family = AF_INET6;
 921         sin6.sin6_addr = *addr6;
 922
 923         rt = rtpurelookup((struct sockaddr *)&sin6);
 924         if (rt == NULL)
 925                 return (NULL);
 926         rt->rt_refcnt--;
 927
 928         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
 929                 if (nd6_onlink_ns_rfc4861 &&
 930                     (ND6_RTENTRY_IS_LLCLONING(rt) ||    /* not cloned yet */
 931                      (rt->rt_parent != NULL &&          /* cloning */
 932                       ND6_RTENTRY_IS_LLCLONING(rt->rt_parent)))) {
 933                         /*
 934                          * If cloning ever happened or is happening,
 935                          * rtentry for addr6 would or will become a
 936                          * neighbor cache.
 937                          */
 938                 } else {
 939                         rt = NULL;
 940                 }
 941         }
 942         return (rt);
 943 }
 944
 945 /*
 946  * Detect if a given IPv6 address identifies a neighbor on a given link.
 947  * XXX: should take care of the destination of a p2p link?
 948  */
 949 int
 950 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 951 {
 952         struct ifaddr_container *ifac;
 953         int i;
 954
 955 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 956 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 957
 958         /*
 959          * A link-local address is always a neighbor.
 960          * XXX: we should use the sin6_scope_id field rather than the embedded
 961          * interface index.
 962          */
 963         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 964             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 965                 return (1);
 966
 967         /*
 968          * If the address matches one of our addresses,
 969          * it should be a neighbor.
 970          */
 971         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 972                 struct ifaddr *ifa = ifac->ifa;
 973
 974                 if (ifa->ifa_addr->sa_family != AF_INET6)
 975                         next: continue;
 976
 977                 for (i = 0; i < 4; i++) {
 978                         if ((IFADDR6(ifa).s6_addr32[i] ^
 979                              addr->sin6_addr.s6_addr32[i]) &
 980                             IFMASK6(ifa).s6_addr32[i])
 981                                 goto next;
 982                 }
 983                 return (1);
 984         }
 985
 986         /*
 987          * Even if the address matches none of our addresses, it might be
 988          * in the neighbor cache.
 989          */
 990         if (nd6_neighbor_lookup(&addr->sin6_addr, ifp) != NULL)
 991                 return (1);
 992
 993         return (0);
 994 #undef IFADDR6
 995 #undef IFMASK6
 996 }
 997
 998 /*
 999  * Free an nd6 llinfo entry.
1000  */
1001 struct llinfo_nd6 *
1002 nd6_free(struct rtentry *rt)
1003 {
1004         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1005         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
1006         struct nd_defrouter *dr;
1007
1008         /*
1009          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
1010          * even though it is not harmful, it was not really necessary.
1011          */
1012
1013         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
1014                 mtx_lock(&nd6_mtx);
1015                 dr = defrouter_lookup(
1016                     &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1017                     rt->rt_ifp);
1018
1019                 if (ln->ln_router || dr) {
1020                         /*
1021                          * rt6_flush must be called whether or not the neighbor
1022                          * is in the Default Router List.
1023                          * See a corresponding comment in nd6_na_input().
1024                          */
1025                         rt6_flush(&in6, rt->rt_ifp);
1026                 }
1027
1028                 if (dr) {
1029                         /*
1030                          * Unreachablity of a router might affect the default
1031                          * router selection and on-link detection of advertised
1032                          * prefixes.
1033                          */
1034
1035                         /*
1036                          * Temporarily fake the state to choose a new default
1037                          * router and to perform on-link determination of
1038                          * prefixes correctly.
1039                          * Below the state will be set correctly,
1040                          * or the entry itself will be deleted.
1041                          */
1042                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
1043
1044                         /*
1045                          * Since defrouter_select() does not affect the
1046                          * on-link determination and MIP6 needs the check
1047                          * before the default router selection, we perform
1048                          * the check now.
1049                          */
1050                         pfxlist_onlink_check();
1051
1052                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1053                                 /*
1054                                  * It is used as the current default router,
1055                                  * so we have to move it to the end of the
1056                                  * list and choose a new one.
1057                                  * XXX: it is not very efficient if this is
1058                                  *      the only router.
1059                                  */
1060                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1061                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1062
1063                                 defrouter_select();
1064                         }
1065                 }
1066                 mtx_unlock(&nd6_mtx);
1067         }
1068
1069         /*
1070          * Before deleting the entry, remember the next entry as the
1071          * return value.  We need this because pfxlist_onlink_check() above
1072          * might have freed other entries (particularly the old next entry) as
1073          * a side effect (XXX).
1074          */
1075         next = ln->ln_next;
1076
1077         /*
1078          * Detach the route from the routing tree and the list of neighbor
1079          * caches, and disable the route entry not to be used in already
1080          * cached routes.
1081          */
1082         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1083
1084         return (next);
1085 }
1086
1087 /*
1088  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1089  *
1090  * XXX cost-effective metods?
1091  */
1092 void
1093 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1094 {
1095         struct llinfo_nd6 *ln;
1096
1097         /*
1098          * If the caller specified "rt", use that.  Otherwise, resolve the
1099          * routing table by supplied "dst6".
1100          */
1101         if (!rt) {
1102                 if (!dst6)
1103                         return;
1104                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1105                         return;
1106         }
1107
1108         if ((rt->rt_flags & RTF_GATEWAY) ||
1109             !(rt->rt_flags & RTF_LLINFO) ||
1110             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1111             rt->rt_gateway->sa_family != AF_LINK) {
1112                 /* This is not a host route. */
1113                 return;
1114         }
1115
1116         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1117         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1118                 return;
1119
1120         /*
1121          * if we get upper-layer reachability confirmation many times,
1122          * it is possible we have false information.
1123          */
1124         if (!force) {
1125                 ln->ln_byhint++;
1126                 if (ln->ln_byhint > nd6_maxnudhint)
1127                         return;
1128         }
1129
1130         ln->ln_state = ND6_LLINFO_REACHABLE;
1131         if (ln->ln_expire)
1132                 ln->ln_expire = time_uptime +
1133                         ND_IFINFO(rt->rt_ifp)->reachable;
1134 }
1135
1136 void
1137 nd6_rtrequest(int req, struct rtentry *rt)
1138 {
1139         struct sockaddr *gate = rt->rt_gateway;
1140         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1141         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1142         struct ifnet *ifp = rt->rt_ifp;
1143         struct ifaddr *ifa;
1144
1145         if ((rt->rt_flags & RTF_GATEWAY))
1146                 return;
1147
1148         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1149                 /*
1150                  * This is probably an interface direct route for a link
1151                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1152                  * We do not need special treatment below for such a route.
1153                  * Moreover, the RTF_LLINFO flag which would be set below
1154                  * would annoy the ndp(8) command.
1155                  */
1156                 return;
1157         }
1158
1159         if (req == RTM_RESOLVE &&
1160             (nd6_need_cache(ifp) == 0 || /* stf case */
1161              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1162                 /*
1163                  * FreeBSD and BSD/OS often make a cloned host route based
1164                  * on a less-specific route (e.g. the default route).
1165                  * If the less specific route does not have a "gateway"
1166                  * (this is the case when the route just goes to a p2p or an
1167                  * stf interface), we'll mistakenly make a neighbor cache for
1168                  * the host route, and will see strange neighbor solicitation
1169                  * for the corresponding destination.  In order to avoid the
1170                  * confusion, we check if the destination of the route is
1171                  * a neighbor in terms of neighbor discovery, and stop the
1172                  * process if not.  Additionally, we remove the LLINFO flag
1173                  * so that ndp(8) will not try to get the neighbor information
1174                  * of the destination.
1175                  */
1176                 rt->rt_flags &= ~RTF_LLINFO;
1177                 return;
1178         }
1179
1180         switch (req) {
1181         case RTM_ADD:
1182                 /*
1183                  * There is no backward compatibility :)
1184                  *
1185                  * if (!(rt->rt_flags & RTF_HOST) &&
1186                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1187                  *         rt->rt_flags |= RTF_CLONING;
1188                  */
1189                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1190                         /*
1191                          * Case 1: This route should come from
1192                          * a route to interface.  RTF_LLINFO flag is set
1193                          * for a host route whose destination should be
1194                          * treated as on-link.
1195                          */
1196                         rt_setgate(rt, rt_key(rt),
1197                                    (struct sockaddr *)&null_sdl,
1198                                    RTL_DONTREPORT);
1199                         gate = rt->rt_gateway;
1200                         SDL(gate)->sdl_type = ifp->if_type;
1201                         SDL(gate)->sdl_index = ifp->if_index;
1202                         if (ln)
1203                                 ln->ln_expire = time_uptime;
1204                         if (ln && ln->ln_expire == 0) {
1205                                 /* kludge for desktops */
1206                                 ln->ln_expire = 1;
1207                         }
1208                         if ((rt->rt_flags & RTF_CLONING))
1209                                 break;
1210                 }
1211                 /*
1212                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1213                  * We don't do that here since llinfo is not ready yet.
1214                  *
1215                  * There are also couple of other things to be discussed:
1216                  * - unsolicited NA code needs improvement beforehand
1217                  * - RFC2461 says we MAY send multicast unsolicited NA
1218                  *   (7.2.6 paragraph 4), however, it also says that we
1219                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1220                  *   we don't have anything like it right now.
1221                  *   note that the mechanism needs a mutual agreement
1222                  *   between proxies, which means that we need to implement
1223                  *   a new protocol, or a new kludge.
1224                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1225                  *   we need to check ip6forwarding before sending it.
1226                  *   (or should we allow proxy ND configuration only for
1227                  *   routers?  there's no mention about proxy ND from hosts)
1228                  */
1229 #if 0
1230                 /* XXX it does not work */
1231                 if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
1232                         nd6_na_output(ifp,
1233                               &SIN6(rt_key(rt))->sin6_addr,
1234                               &SIN6(rt_key(rt))->sin6_addr,
1235                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1236                               1, NULL);
1237                 }
1238 #endif
1239                 /* FALLTHROUGH */
1240         case RTM_RESOLVE:
1241                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1242                         /*
1243                          * Address resolution isn't necessary for a point to
1244                          * point link, so we can skip this test for a p2p link.
1245                          */
1246                         if (gate->sa_family != AF_LINK ||
1247                             gate->sa_len < sizeof(null_sdl)) {
1248                                 log(LOG_DEBUG,
1249                                     "nd6_rtrequest: bad gateway value: %s\n",
1250                                     if_name(ifp));
1251                                 break;
1252                         }
1253                         SDL(gate)->sdl_type = ifp->if_type;
1254                         SDL(gate)->sdl_index = ifp->if_index;
1255                 }
1256                 if (ln != NULL)
1257                         break;  /* This happens on a route change */
1258                 /*
1259                  * Case 2: This route may come from cloning, or a manual route
1260                  * add with a LL address.
1261                  */
1262                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1263                 rt->rt_llinfo = (caddr_t)ln;
1264                 if (!ln) {
1265                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1266                         break;
1267                 }
1268                 nd6_inuse++;
1269                 nd6_allocated++;
1270                 bzero(ln, sizeof(*ln));
1271                 ln->ln_rt = rt;
1272                 /* this is required for "ndp" command. - shin */
1273                 if (req == RTM_ADD) {
1274                         /*
1275                          * gate should have some valid AF_LINK entry,
1276                          * and ln->ln_expire should have some lifetime
1277                          * which is specified by ndp command.
1278                          */
1279                         ln->ln_state = ND6_LLINFO_REACHABLE;
1280                         ln->ln_byhint = 0;
1281                 } else {
1282                         /*
1283                          * When req == RTM_RESOLVE, rt is created and
1284                          * initialized in rtrequest(), so rt_expire is 0.
1285                          */
1286                         ln->ln_state = ND6_LLINFO_NOSTATE;
1287                         ln->ln_expire = time_uptime;
1288                 }
1289                 rt->rt_flags |= RTF_LLINFO;
1290                 ln->ln_next = llinfo_nd6.ln_next;
1291                 llinfo_nd6.ln_next = ln;
1292                 ln->ln_prev = &llinfo_nd6;
1293                 ln->ln_next->ln_prev = ln;
1294
1295                 /*
1296                  * check if rt_key(rt) is one of my address assigned
1297                  * to the interface.
1298                  */
1299                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1300                     &SIN6(rt_key(rt))->sin6_addr);
1301                 if (ifa) {
1302                         caddr_t macp = nd6_ifptomac(ifp);
1303                         ln->ln_expire = 0;
1304                         ln->ln_state = ND6_LLINFO_REACHABLE;
1305                         ln->ln_byhint = 0;
1306                         if (macp) {
1307                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1308                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1309                         }
1310                         if (nd6_useloopback) {
1311                                 rt->rt_ifp = loif;      /* XXX */
1312                                 /*
1313                                  * Make sure rt_ifa be equal to the ifaddr
1314                                  * corresponding to the address.
1315                                  * We need this because when we refer
1316                                  * rt_ifa->ia6_flags in ip6_input, we assume
1317                                  * that the rt_ifa points to the address instead
1318                                  * of the loopback address.
1319                                  */
1320                                 if (ifa != rt->rt_ifa) {
1321                                         IFAFREE(rt->rt_ifa);
1322                                         IFAREF(ifa);
1323                                         rt->rt_ifa = ifa;
1324                                 }
1325                         }
1326                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1327                         ln->ln_expire = 0;
1328                         ln->ln_state = ND6_LLINFO_REACHABLE;
1329                         ln->ln_byhint = 0;
1330
1331                         /*
1332                          * Join solicited node multicast for proxy ND, and only
1333                          * join it once on cpu0.
1334                          */
1335                         if ((ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1336                                 struct in6_addr llsol;
1337                                 int error;
1338
1339                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1340                                 llsol.s6_addr16[0] = htons(0xff02);
1341                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1342                                 llsol.s6_addr32[1] = 0;
1343                                 llsol.s6_addr32[2] = htonl(1);
1344                                 llsol.s6_addr8[12] = 0xff;
1345
1346                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1347                                         nd6log((LOG_ERR, "%s: failed to join "
1348                                             "%s (errno=%d)\n", if_name(ifp),
1349                                             ip6_sprintf(&llsol), error));
1350                                 }
1351                         }
1352                 }
1353                 break;
1354
1355         case RTM_DELETE:
1356                 if (!ln)
1357                         break;
1358                 /*
1359                  * Leave from solicited node multicast for proxy ND, and only
1360                  * leave it once on cpu0 (since we joined it once on cpu0).
1361                  */
1362                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1363                     (ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1364                         struct in6_addr llsol;
1365                         struct in6_multi *in6m;
1366
1367                         llsol = SIN6(rt_key(rt))->sin6_addr;
1368                         llsol.s6_addr16[0] = htons(0xff02);
1369                         llsol.s6_addr16[1] = htons(ifp->if_index);
1370                         llsol.s6_addr32[1] = 0;
1371                         llsol.s6_addr32[2] = htonl(1);
1372                         llsol.s6_addr8[12] = 0xff;
1373
1374                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1375                         if (in6m)
1376                                 in6_delmulti(in6m);
1377                 }
1378                 nd6_inuse--;
1379                 ln->ln_next->ln_prev = ln->ln_prev;
1380                 ln->ln_prev->ln_next = ln->ln_next;
1381                 ln->ln_prev = NULL;
1382                 rt->rt_llinfo = 0;
1383                 rt->rt_flags &= ~RTF_LLINFO;
1384                 if (ln->ln_hold)
1385                         m_freem(ln->ln_hold);
1386                 Free((caddr_t)ln);
1387         }
1388 }
1389
1390 int
1391 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1392 {
1393         struct in6_drlist *drl = (struct in6_drlist *)data;
1394         struct in6_prlist *prl = (struct in6_prlist *)data;
1395         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1396         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1397         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1398         struct nd_defrouter *dr, any;
1399         struct nd_prefix *pr;
1400         struct rtentry *rt;
1401         int i = 0, error = 0;
1402
1403         switch (cmd) {
1404         case SIOCGDRLST_IN6:
1405                 /*
1406                  * obsolete API, use sysctl under net.inet6.icmp6
1407                  */
1408                 bzero(drl, sizeof(*drl));
1409                 mtx_lock(&nd6_mtx);
1410                 dr = TAILQ_FIRST(&nd_defrouter);
1411                 while (dr && i < DRLSTSIZ) {
1412                         drl->defrouter[i].rtaddr = dr->rtaddr;
1413                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1414                                 /* XXX: need to this hack for KAME stack */
1415                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1416                         } else
1417                                 log(LOG_ERR,
1418                                     "default router list contains a "
1419                                     "non-linklocal address(%s)\n",
1420                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1421
1422                         drl->defrouter[i].flags = dr->flags;
1423                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1424                         drl->defrouter[i].expire = dr->expire;
1425                         drl->defrouter[i].if_index = dr->ifp->if_index;
1426                         i++;
1427                         dr = TAILQ_NEXT(dr, dr_entry);
1428                 }
1429                 mtx_unlock(&nd6_mtx);
1430                 break;
1431         case SIOCGPRLST_IN6:
1432                 /*
1433                  * obsolete API, use sysctl under net.inet6.icmp6
1434                  */
1435                 /*
1436                  * XXX meaning of fields, especialy "raflags", is very
1437                  * differnet between RA prefix list and RR/static prefix list.
1438                  * how about separating ioctls into two?
1439                  */
1440                 bzero(prl, sizeof(*prl));
1441                 mtx_lock(&nd6_mtx);
1442                 pr = nd_prefix.lh_first;
1443                 while (pr && i < PRLSTSIZ) {
1444                         struct nd_pfxrouter *pfr;
1445                         int j;
1446
1447                         in6_embedscope(&prl->prefix[i].prefix,
1448                             &pr->ndpr_prefix, NULL, NULL);
1449                         prl->prefix[i].raflags = pr->ndpr_raf;
1450                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1451                         prl->prefix[i].vltime = pr->ndpr_vltime;
1452                         prl->prefix[i].pltime = pr->ndpr_pltime;
1453                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1454                         prl->prefix[i].expire = pr->ndpr_expire;
1455
1456                         pfr = pr->ndpr_advrtrs.lh_first;
1457                         j = 0;
1458                         while (pfr) {
1459                                 if (j < DRLSTSIZ) {
1460 #define RTRADDR prl->prefix[i].advrtr[j]
1461                                         RTRADDR = pfr->router->rtaddr;
1462                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1463                                                 /* XXX: hack for KAME */
1464                                                 RTRADDR.s6_addr16[1] = 0;
1465                                         } else
1466                                                 log(LOG_ERR,
1467                                                     "a router(%s) advertises "
1468                                                     "a prefix with "
1469                                                     "non-link local address\n",
1470                                                     ip6_sprintf(&RTRADDR));
1471 #undef RTRADDR
1472                                 }
1473                                 j++;
1474                                 pfr = pfr->pfr_next;
1475                         }
1476                         prl->prefix[i].advrtrs = j;
1477                         prl->prefix[i].origin = PR_ORIG_RA;
1478
1479                         i++;
1480                         pr = pr->ndpr_next;
1481                 }
1482                 mtx_unlock(&nd6_mtx);
1483
1484                 break;
1485         case OSIOCGIFINFO_IN6:
1486                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1487                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1488                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1489                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1490                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1491                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1492                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1493                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1494                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1495                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1496                 break;
1497         case SIOCGIFINFO_IN6:
1498                 ndi->ndi = *ND_IFINFO(ifp);
1499                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1500                 break;
1501         case SIOCSIFINFO_FLAGS:
1502                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1503                 break;
1504         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1505                 /* flush default router list */
1506                 /*
1507                  * xxx sumikawa: should not delete route if default
1508                  * route equals to the top of default router list
1509                  */
1510                 bzero(&any, sizeof(any));
1511                 defrouter_delreq(&any, 0);
1512                 defrouter_select();
1513                 /* xxx sumikawa: flush prefix list */
1514                 break;
1515         case SIOCSPFXFLUSH_IN6:
1516         {
1517                 /* flush all the prefix advertised by routers */
1518                 struct nd_prefix *pr, *next;
1519
1520                 mtx_lock(&nd6_mtx);
1521                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1522                         struct in6_ifaddr *ia, *ia_next;
1523
1524                         next = pr->ndpr_next;
1525
1526                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1527                                 continue; /* XXX */
1528
1529                         /* do we really have to remove addresses as well? */
1530                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1531                                 /* ia might be removed.  keep the next ptr. */
1532                                 ia_next = ia->ia_next;
1533
1534                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1535                                         continue;
1536
1537                                 if (ia->ia6_ndpr == pr)
1538                                         in6_purgeaddr(&ia->ia_ifa);
1539                         }
1540                         prelist_remove(pr);
1541                 }
1542                 mtx_unlock(&nd6_mtx);
1543                 break;
1544         }
1545         case SIOCSRTRFLUSH_IN6:
1546         {
1547                 /* flush all the default routers */
1548                 struct nd_defrouter *dr, *next;
1549
1550                 mtx_lock(&nd6_mtx);
1551                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1552                         /*
1553                          * The first entry of the list may be stored in
1554                          * the routing table, so we'll delete it later.
1555                          */
1556                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1557                                 next = TAILQ_NEXT(dr, dr_entry);
1558                                 defrtrlist_del(dr);
1559                         }
1560                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1561                 }
1562                 mtx_unlock(&nd6_mtx);
1563                 break;
1564         }
1565         case SIOCGNBRINFO_IN6:
1566         {
1567                 struct llinfo_nd6 *ln;
1568                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1569
1570                 /*
1571                  * XXX: KAME specific hack for scoped addresses
1572                  *      XXXX: for other scopes than link-local?
1573                  */
1574                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1575                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1576                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1577
1578                         if (*idp == 0)
1579                                 *idp = htons(ifp->if_index);
1580                 }
1581
1582                 mtx_lock(&nd6_mtx);
1583                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1584                         error = EINVAL;
1585                         mtx_unlock(&nd6_mtx);
1586                         break;
1587                 }
1588                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1589                 nbi->state = ln->ln_state;
1590                 nbi->asked = ln->ln_asked;
1591                 nbi->isrouter = ln->ln_router;
1592                 nbi->expire = ln->ln_expire;
1593                 mtx_unlock(&nd6_mtx);
1594
1595                 break;
1596         }
1597         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1598                 ndif->ifindex = nd6_defifindex;
1599                 break;
1600         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1601                 return (nd6_setdefaultiface(ndif->ifindex));
1602         }
1603         return (error);
1604 }
1605
1606 /*
1607  * Create neighbor cache entry and cache link-layer address,
1608  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1609  */
1610 struct rtentry *
1611 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1612                  int lladdrlen,
1613                  int type,      /* ICMP6 type */
1614                  int code       /* type dependent information */)
1615 {
1616         struct rtentry *rt = NULL;
1617         struct llinfo_nd6 *ln = NULL;
1618         int is_newentry;
1619         struct sockaddr_dl *sdl = NULL;
1620         int do_update;
1621         int olladdr;
1622         int llchange;
1623         int newstate = 0;
1624
1625         if (!ifp)
1626                 panic("ifp == NULL in nd6_cache_lladdr");
1627         if (!from)
1628                 panic("from == NULL in nd6_cache_lladdr");
1629
1630         /* nothing must be updated for unspecified address */
1631         if (IN6_IS_ADDR_UNSPECIFIED(from))
1632                 return NULL;
1633
1634         /*
1635          * Validation about ifp->if_addrlen and lladdrlen must be done in
1636          * the caller.
1637          *
1638          * XXX If the link does not have link-layer adderss, what should
1639          * we do? (ifp->if_addrlen == 0)
1640          * Spec says nothing in sections for RA, RS and NA.  There's small
1641          * description on it in NS section (RFC 2461 7.2.3).
1642          */
1643
1644         rt = nd6_lookup(from, 0, ifp);
1645         if (!rt) {
1646 #if 0
1647                 /* nothing must be done if there's no lladdr */
1648                 if (!lladdr || !lladdrlen)
1649                         return NULL;
1650 #endif
1651
1652                 rt = nd6_lookup(from, 1, ifp);
1653                 is_newentry = 1;
1654         } else {
1655                 /* do nothing if static ndp is set */
1656                 if (rt->rt_flags & RTF_STATIC)
1657                         return NULL;
1658                 is_newentry = 0;
1659         }
1660
1661         if (!rt)
1662                 return NULL;
1663         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1664 fail:
1665                 nd6_free(rt);
1666                 return NULL;
1667         }
1668         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1669         if (!ln)
1670                 goto fail;
1671         if (!rt->rt_gateway)
1672                 goto fail;
1673         if (rt->rt_gateway->sa_family != AF_LINK)
1674                 goto fail;
1675         sdl = SDL(rt->rt_gateway);
1676
1677         olladdr = (sdl->sdl_alen) ? 1 : 0;
1678         if (olladdr && lladdr) {
1679                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1680                         llchange = 1;
1681                 else
1682                         llchange = 0;
1683         } else
1684                 llchange = 0;
1685
1686         /*
1687          * newentry olladdr  lladdr  llchange   (*=record)
1688          *      0       n       n       --      (1)
1689          *      0       y       n       --      (2)
1690          *      0       n       y       --      (3) * STALE
1691          *      0       y       y       n       (4) *
1692          *      0       y       y       y       (5) * STALE
1693          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1694          *      1       --      y       --      (7) * STALE
1695          */
1696
1697         if (lladdr) {           /* (3-5) and (7) */
1698                 /*
1699                  * Record source link-layer address
1700                  * XXX is it dependent to ifp->if_type?
1701                  */
1702                 sdl->sdl_alen = ifp->if_addrlen;
1703                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1704         }
1705
1706         if (!is_newentry) {
1707                 if ((!olladdr && lladdr) ||             /* (3) */
1708                     (olladdr && lladdr && llchange)) {  /* (5) */
1709                         do_update = 1;
1710                         newstate = ND6_LLINFO_STALE;
1711                 } else {                                /* (1-2,4) */
1712                         do_update = 0;
1713                 }
1714         } else {
1715                 do_update = 1;
1716                 if (!lladdr)                            /* (6) */
1717                         newstate = ND6_LLINFO_NOSTATE;
1718                 else                                    /* (7) */
1719                         newstate = ND6_LLINFO_STALE;
1720         }
1721
1722         if (do_update) {
1723                 /*
1724                  * Update the state of the neighbor cache.
1725                  */
1726                 ln->ln_state = newstate;
1727
1728                 if (ln->ln_state == ND6_LLINFO_STALE) {
1729                         /*
1730                          * XXX: since nd6_output() below will cause
1731                          * state tansition to DELAY and reset the timer,
1732                          * we must set the timer now, although it is actually
1733                          * meaningless.
1734                          */
1735                         ln->ln_expire = time_uptime + nd6_gctimer;
1736
1737                         if (ln->ln_hold) {
1738                                 /*
1739                                  * we assume ifp is not a p2p here, so just
1740                                  * set the 2nd argument as the 1st one.
1741                                  */
1742                                 nd6_output(ifp, ifp, ln->ln_hold,
1743                                     (struct sockaddr_in6 *)rt_key(rt), rt);
1744                                 ln->ln_hold = NULL;
1745                         }
1746                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1747                         /* probe right away */
1748                         ln->ln_expire = time_uptime;
1749                 }
1750         }
1751
1752         /*
1753          * ICMP6 type dependent behavior.
1754          *
1755          * NS: clear IsRouter if new entry
1756          * RS: clear IsRouter
1757          * RA: set IsRouter if there's lladdr
1758          * redir: clear IsRouter if new entry
1759          *
1760          * RA case, (1):
1761          * The spec says that we must set IsRouter in the following cases:
1762          * - If lladdr exist, set IsRouter.  This means (1-5).
1763          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1764          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1765          * A quetion arises for (1) case.  (1) case has no lladdr in the
1766          * neighbor cache, this is similar to (6).
1767          * This case is rare but we figured that we MUST NOT set IsRouter.
1768          *
1769          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1770          *                                                      D R
1771          *      0       n       n       --      (1)     c   ?     s
1772          *      0       y       n       --      (2)     c   s     s
1773          *      0       n       y       --      (3)     c   s     s
1774          *      0       y       y       n       (4)     c   s     s
1775          *      0       y       y       y       (5)     c   s     s
1776          *      1       --      n       --      (6) c   c       c s
1777          *      1       --      y       --      (7) c   c   s   c s
1778          *
1779          *                                      (c=clear s=set)
1780          */
1781         switch (type & 0xff) {
1782         case ND_NEIGHBOR_SOLICIT:
1783                 /*
1784                  * New entry must have is_router flag cleared.
1785                  */
1786                 if (is_newentry)        /* (6-7) */
1787                         ln->ln_router = 0;
1788                 break;
1789         case ND_REDIRECT:
1790                 /*
1791                  * If the icmp is a redirect to a better router, always set the
1792                  * is_router flag.  Otherwise, if the entry is newly created,
1793                  * clear the flag.  [RFC 2461, sec 8.3]
1794                  */
1795                 if (code == ND_REDIRECT_ROUTER)
1796                         ln->ln_router = 1;
1797                 else if (is_newentry) /* (6-7) */
1798                         ln->ln_router = 0;
1799                 break;
1800         case ND_ROUTER_SOLICIT:
1801                 /*
1802                  * is_router flag must always be cleared.
1803                  */
1804                 ln->ln_router = 0;
1805                 break;
1806         case ND_ROUTER_ADVERT:
1807                 /*
1808                  * Mark an entry with lladdr as a router.
1809                  */
1810                 if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
1811                     (is_newentry && lladdr)) {                  /* (7) */
1812                         ln->ln_router = 1;
1813                 }
1814                 break;
1815         }
1816
1817         /*
1818          * When the link-layer address of a router changes, select the
1819          * best router again.  In particular, when the neighbor entry is newly
1820          * created, it might affect the selection policy.
1821          * Question: can we restrict the first condition to the "is_newentry"
1822          * case?
1823          * XXX: when we hear an RA from a new router with the link-layer
1824          * address option, defrouter_select() is called twice, since
1825          * defrtrlist_update called the function as well.  However, I believe
1826          * we can compromise the overhead, since it only happens the first
1827          * time.
1828          * XXX: although defrouter_select() should not have a bad effect
1829          * for those are not autoconfigured hosts, we explicitly avoid such
1830          * cases for safety.
1831          */
1832         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1833                 defrouter_select();
1834
1835         return rt;
1836 }
1837
1838 static void
1839 nd6_slowtimo(void *arg __unused)
1840 {
1841         struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1842
1843         KASSERT(mycpuid == 0, ("not on cpu0"));
1844         crit_enter();
1845         if (lmsg->ms_flags & MSGF_DONE)
1846                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1847         crit_exit();
1848 }
1849
1850 static void
1851 nd6_slowtimo_dispatch(netmsg_t nmsg)
1852 {
1853         const struct ifnet_array *arr;
1854         struct nd_ifinfo *nd6if;
1855         int i;
1856
1857         ASSERT_NETISR0;
1858
1859         crit_enter();
1860         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1861         crit_exit();
1862
1863         arr = ifnet_array_get();
1864
1865         mtx_lock(&nd6_mtx);
1866         for (i = 0; i < arr->ifnet_count; ++i) {
1867                 struct ifnet *ifp = arr->ifnet_arr[i];
1868
1869                 if (ifp->if_afdata[AF_INET6] == NULL)
1870                         continue;
1871                 nd6if = ND_IFINFO(ifp);
1872                 if (nd6if->basereachable && /* already initialized */
1873                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1874                         /*
1875                          * Since reachable time rarely changes by router
1876                          * advertisements, we SHOULD insure that a new random
1877                          * value gets recomputed at least once every few hours.
1878                          * (RFC 2461, 6.3.4)
1879                          */
1880                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1881                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1882                 }
1883         }
1884         mtx_unlock(&nd6_mtx);
1885
1886         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1887             nd6_slowtimo, NULL);
1888 }
1889
1890 #define gotoerr(e) { error = (e); goto bad;}
1891
1892 int
1893 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1894            struct sockaddr_in6 *dst, struct rtentry *rt)
1895 {
1896         struct llinfo_nd6 *ln = NULL;
1897         int error = 0;
1898
1899         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1900                 goto sendpkt;
1901
1902         if (nd6_need_cache(ifp) == 0)
1903                 goto sendpkt;
1904
1905         /*
1906          * Next hop determination.  This routine is derived from rt_llroute.
1907          */
1908         if (rt != NULL) {
1909                 if (!(rt->rt_flags & RTF_UP)) {
1910                         rt = rtlookup((struct sockaddr *)dst);
1911                         if (rt == NULL)
1912                                 gotoerr(EHOSTUNREACH);
1913                         rt->rt_refcnt--;
1914                         if (rt->rt_ifp != ifp) {
1915                                 /* XXX: loop care? */
1916                                 return nd6_output(ifp, origifp, m, dst, rt);
1917                         }
1918                 }
1919                 if (rt->rt_flags & RTF_GATEWAY) {
1920                         struct sockaddr_in6 *gw6;
1921
1922                         /*
1923                          * We skip link-layer address resolution and NUD
1924                          * if the gateway is not a neighbor from ND point
1925                          * of view, regardless of the value of nd_ifinfo.flags.
1926                          * The second condition is a bit tricky; we skip
1927                          * if the gateway is our own address, which is
1928                          * sometimes used to install a route to a p2p link.
1929                          */
1930                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1931                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1932                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1933                                 /*
1934                                  * We allow this kind of tricky route only
1935                                  * when the outgoing interface is p2p.
1936                                  * XXX: we may need a more generic rule here.
1937                                  */
1938                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1939                                         gotoerr(EHOSTUNREACH);
1940
1941                                 goto sendpkt;
1942                         }
1943
1944                         if (rt->rt_gwroute == NULL) {
1945                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1946                                 if (rt->rt_gwroute == NULL)
1947                                         gotoerr(EHOSTUNREACH);
1948                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1949                                 rtfree(rt->rt_gwroute);
1950                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1951                                 if (rt->rt_gwroute == NULL)
1952                                         gotoerr(EHOSTUNREACH);
1953                         }
1954                         rt = rt->rt_gwroute;
1955                 }
1956         }
1957
1958         /*
1959          * Address resolution or Neighbor Unreachability Detection
1960          * for the next hop.
1961          * At this point, the destination of the packet must be a unicast
1962          * or an anycast address(i.e. not a multicast).
1963          */
1964
1965         /* Look up the neighbor cache for the nexthop */
1966         if (rt && (rt->rt_flags & RTF_LLINFO))
1967                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1968         else {
1969                 /*
1970                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1971                  * the condition below is not very efficient.  But we believe
1972                  * it is tolerable, because this should be a rare case.
1973                  */
1974                 if (nd6_is_addr_neighbor(dst, ifp) &&
1975                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1976                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1977         }
1978         if (!ln || !rt) {
1979                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1980                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1981                         log(LOG_DEBUG,
1982                             "nd6_output: can't allocate llinfo for %s "
1983                             "(ln=%p, rt=%p)\n",
1984                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1985                         gotoerr(EIO);   /* XXX: good error? */
1986                 }
1987
1988                 goto sendpkt;   /* send anyway */
1989         }
1990
1991         /* We don't have to do link-layer address resolution on a p2p link. */
1992         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1993             ln->ln_state < ND6_LLINFO_REACHABLE) {
1994                 ln->ln_state = ND6_LLINFO_STALE;
1995                 ln->ln_expire = time_uptime + nd6_gctimer;
1996         }
1997
1998         /*
1999          * The first time we send a packet to a neighbor whose entry is
2000          * STALE, we have to change the state to DELAY and a sets a timer to
2001          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2002          * neighbor unreachability detection on expiration.
2003          * (RFC 2461 7.3.3)
2004          */
2005         if (ln->ln_state == ND6_LLINFO_STALE) {
2006                 ln->ln_asked = 0;
2007                 ln->ln_state = ND6_LLINFO_DELAY;
2008                 ln->ln_expire = time_uptime + nd6_delay;
2009         }
2010
2011         /*
2012          * If the neighbor cache entry has a state other than INCOMPLETE
2013          * (i.e. its link-layer address is already resolved), just
2014          * send the packet.
2015          */
2016         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2017                 goto sendpkt;
2018
2019         /*
2020          * There is a neighbor cache entry, but no ethernet address
2021          * response yet.  Replace the held mbuf (if any) with this
2022          * latest one.
2023          *
2024          * This code conforms to the rate-limiting rule described in Section
2025          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2026          * an NS below.
2027          */
2028         if (ln->ln_state == ND6_LLINFO_NOSTATE) {
2029                 /*
2030                  * This neighbor cache entry was just created; change its
2031                  * state to INCOMPLETE and start its life cycle.
2032                  *
2033                  * We force an NS output below by setting ln_expire to 1
2034                  * (nd6_rtrequest() could set it to the current time_uptime)
2035                  * and zeroing out ln_asked (XXX this may not be necessary).
2036                  */
2037                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
2038                 ln->ln_expire = 1;
2039                 ln->ln_asked = 0;
2040         }
2041         if (ln->ln_hold)
2042                 m_freem(ln->ln_hold);
2043         ln->ln_hold = m;
2044         if (ln->ln_expire) {
2045                 if (ln->ln_asked < nd6_mmaxtries &&
2046                     ln->ln_expire < time_uptime) {
2047                         ln->ln_asked++;
2048                         ln->ln_expire = time_uptime +
2049                                 ND_IFINFO(ifp)->retrans / 1000;
2050                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2051                 }
2052         }
2053         return (0);
2054
2055 sendpkt:
2056         if (ifp->if_flags & IFF_LOOPBACK)
2057                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
2058         else
2059                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
2060         return (error);
2061
2062 bad:
2063         m_freem(m);
2064         return (error);
2065 }
2066 #undef gotoerr
2067
2068 int
2069 nd6_need_cache(struct ifnet *ifp)
2070 {
2071         /*
2072          * XXX: we currently do not make neighbor cache on any interface
2073          * other than Ethernet and GIF.
2074          *
2075          * RFC2893 says:
2076          * - unidirectional tunnels needs no ND
2077          */
2078         switch (ifp->if_type) {
2079         case IFT_ETHER:
2080         case IFT_IEEE1394:
2081 #ifdef IFT_L2VLAN
2082         case IFT_L2VLAN:
2083 #endif
2084 #ifdef IFT_IEEE80211
2085         case IFT_IEEE80211:
2086 #endif
2087 #ifdef IFT_CARP
2088         case IFT_CARP:
2089 #endif
2090         case IFT_GIF:           /* XXX need more cases? */
2091                 return (1);
2092         default:
2093                 return (0);
2094         }
2095 }
2096
2097 int
2098 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2099                 struct sockaddr *dst, u_char *desten)
2100 {
2101         struct sockaddr_dl *sdl;
2102         struct rtentry *rt;
2103
2104
2105         if (m->m_flags & M_MCAST) {
2106                 switch (ifp->if_type) {
2107                 case IFT_ETHER:
2108 #ifdef IFT_L2VLAN
2109         case IFT_L2VLAN:
2110 #endif
2111 #ifdef IFT_IEEE80211
2112                 case IFT_IEEE80211:
2113 #endif
2114                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2115                                                  desten);
2116                         return (1);
2117                 case IFT_IEEE1394:
2118                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2119                         return (1);
2120                 default:
2121                         m_freem(m);
2122                         return (0);
2123                 }
2124         }
2125         if (rt0 == NULL) {
2126                 /* this could happen, if we could not allocate memory */
2127                 m_freem(m);
2128                 return (0);
2129         }
2130         if (rt_llroute(dst, rt0, &rt) != 0) {
2131                 m_freem(m);
2132                 return (0);
2133         }
2134         if (rt->rt_gateway->sa_family != AF_LINK) {
2135                 kprintf("nd6_storelladdr: something odd happens\n");
2136                 m_freem(m);
2137                 return (0);
2138         }
2139         sdl = SDL(rt->rt_gateway);
2140         if (sdl->sdl_alen == 0) {
2141                 /* this should be impossible, but we bark here for debugging */
2142                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2143                 m_freem(m);
2144                 return (0);
2145         }
2146
2147         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2148         return (1);
2149 }
2150
2151 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2152 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2153 #ifdef SYSCTL_DECL
2154 SYSCTL_DECL(_net_inet6_icmp6);
2155 #endif
2156 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2157         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2158 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2159         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2160
2161 static int
2162 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2163 {
2164         int error;
2165         char buf[1024];
2166         struct in6_defrouter *d, *de;
2167         struct nd_defrouter *dr;
2168
2169         if (req->newptr)
2170                 return EPERM;
2171         error = 0;
2172
2173         for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2174              dr = TAILQ_NEXT(dr, dr_entry)) {
2175                 d = (struct in6_defrouter *)buf;
2176                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2177
2178                 if (d + 1 <= de) {
2179                         bzero(d, sizeof(*d));
2180                         d->rtaddr.sin6_family = AF_INET6;
2181                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2182                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2183                             dr->ifp) != 0)
2184                                 log(LOG_ERR,
2185                                     "scope error in "
2186                                     "default router list (%s)\n",
2187                                     ip6_sprintf(&dr->rtaddr));
2188                         d->flags = dr->flags;
2189                         d->rtlifetime = dr->rtlifetime;
2190                         d->expire = dr->expire;
2191                         d->if_index = dr->ifp->if_index;
2192                 } else
2193                         panic("buffer too short");
2194
2195                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2196                 if (error)
2197                         break;
2198         }
2199         return error;
2200 }
2201
2202 static int
2203 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2204 {
2205         int error;
2206         char buf[1024];
2207         struct in6_prefix *p, *pe;
2208         struct nd_prefix *pr;
2209
2210         if (req->newptr)
2211                 return EPERM;
2212         error = 0;
2213
2214         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2215                 u_short advrtrs;
2216                 size_t advance;
2217                 struct sockaddr_in6 *sin6, *s6;
2218                 struct nd_pfxrouter *pfr;
2219
2220                 p = (struct in6_prefix *)buf;
2221                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2222
2223                 if (p + 1 <= pe) {
2224                         bzero(p, sizeof(*p));
2225                         sin6 = (struct sockaddr_in6 *)(p + 1);
2226
2227                         p->prefix = pr->ndpr_prefix;
2228                         if (in6_recoverscope(&p->prefix,
2229                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2230                                 log(LOG_ERR,
2231                                     "scope error in prefix list (%s)\n",
2232                                     ip6_sprintf(&p->prefix.sin6_addr));
2233                         p->raflags = pr->ndpr_raf;
2234                         p->prefixlen = pr->ndpr_plen;
2235                         p->vltime = pr->ndpr_vltime;
2236                         p->pltime = pr->ndpr_pltime;
2237                         p->if_index = pr->ndpr_ifp->if_index;
2238                         p->expire = pr->ndpr_expire;
2239                         p->refcnt = pr->ndpr_refcnt;
2240                         p->flags = pr->ndpr_stateflags;
2241                         p->origin = PR_ORIG_RA;
2242                         advrtrs = 0;
2243                         for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2244                              pfr = pfr->pfr_next) {
2245                                 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2246                                         advrtrs++;
2247                                         continue;
2248                                 }
2249                                 s6 = &sin6[advrtrs];
2250                                 bzero(s6, sizeof(*s6));
2251                                 s6->sin6_family = AF_INET6;
2252                                 s6->sin6_len = sizeof(*sin6);
2253                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2254                                     pfr->router->ifp) != 0)
2255                                         log(LOG_ERR,
2256                                             "scope error in "
2257                                             "prefix list (%s)\n",
2258                                             ip6_sprintf(&pfr->router->rtaddr));
2259                                 advrtrs++;
2260                         }
2261                         p->advrtrs = advrtrs;
2262                 } else {
2263                         panic("buffer too short");
2264                 }
2265
2266                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2267                 error = SYSCTL_OUT(req, buf, advance);
2268                 if (error)
2269                         break;
2270         }
2271         return error;
2272 }