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 /*
  34  * XXX
  35  * KAME 970409 note:
  36  * BSD/OS version heavily modifies this code, related to llinfo.
  37  * Since we don't have BSD/OS version of net/route.c in our hand,
  38  * I left the code mostly as it was in 970310.  -- itojun
  39  */
  40
  41 #include "opt_inet.h"
  42 #include "opt_inet6.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/callout.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/socket.h>
  50 #include <sys/sockio.h>
  51 #include <sys/time.h>
  52 #include <sys/kernel.h>
  53 #include <sys/protosw.h>
  54 #include <sys/errno.h>
  55 #include <sys/syslog.h>
  56 #include <sys/queue.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/mutex.h>
  59
  60 #include <sys/thread2.h>
  61 #include <sys/mutex2.h>
  62
  63 #include <net/if.h>
  64 #include <net/if_dl.h>
  65 #include <net/if_types.h>
  66 #include <net/route.h>
  67
  68 #include <netinet/in.h>
  69 #include <netinet/if_ether.h>
  70 #include <netinet6/in6_var.h>
  71 #include <netinet/ip6.h>
  72 #include <netinet6/ip6_var.h>
  73 #include <netinet6/nd6.h>
  74 #include <netinet6/in6_prefix.h>
  75 #include <netinet/icmp6.h>
  76
  77 #include <net/net_osdep.h>
  78
  79 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  80 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  81
  82 #define SIN6(s) ((struct sockaddr_in6 *)s)
  83 #define SDL(s) ((struct sockaddr_dl *)s)
  84
  85 /* timer values */
  86 int     nd6_prune       = 1;    /* walk list every 1 seconds */
  87 int     nd6_delay       = 5;    /* delay first probe time 5 second */
  88 int     nd6_umaxtries   = 3;    /* maximum unicast query */
  89 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
  90 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
  91 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
  92
  93 /* preventing too many loops in ND option parsing */
  94 int nd6_maxndopt = 10;  /* max # of ND options allowed */
  95
  96 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
  97
  98 #ifdef ND6_DEBUG
  99 int nd6_debug = 1;
 100 #else
 101 int nd6_debug = 0;
 102 #endif
 103
 104 /* for debugging? */
 105 static int nd6_inuse, nd6_allocated;
 106
 107 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 108 struct nd_drhead nd_defrouter;
 109 struct nd_prhead nd_prefix = { 0 };
 110 struct mtx nd6_mtx = MTX_INITIALIZER;
 111
 112 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 113 static struct sockaddr_in6 all1_sa;
 114
 115 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
 116 static void nd6_slowtimo (void *);
 117 static int regen_tmpaddr (struct in6_ifaddr *);
 118
 119 struct callout nd6_slowtimo_ch;
 120 struct callout nd6_timer_ch;
 121 extern struct callout in6_tmpaddrtimer_ch;
 122
 123 void
 124 nd6_init(void)
 125 {
 126         static int nd6_init_done = 0;
 127         int i;
 128
 129         if (nd6_init_done) {
 130                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 131                 return;
 132         }
 133
 134         all1_sa.sin6_family = AF_INET6;
 135         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 136         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 137                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 138
 139         /* initialization of the default router list */
 140         TAILQ_INIT(&nd_defrouter);
 141
 142         nd6_init_done = 1;
 143
 144         /* start timer */
 145         callout_init(&nd6_slowtimo_ch);
 146         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 147             nd6_slowtimo, NULL);
 148 }
 149
 150 struct nd_ifinfo *
 151 nd6_ifattach(struct ifnet *ifp)
 152 {
 153         struct nd_ifinfo *nd;
 154
 155         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
 156             M_WAITOK | M_ZERO);
 157
 158         nd->initialized = 1;
 159
 160         nd->linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
 161         nd->chlim = IPV6_DEFHLIM;
 162         nd->basereachable = REACHABLE_TIME;
 163         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 164         nd->retrans = RETRANS_TIMER;
 165         nd->receivedra = 0;
 166
 167         /*
 168          * Note that the default value of ip6_accept_rtadv is 0, which means
 169          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 170          * here.
 171          */
 172         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 173
 174         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 175         nd6_setmtu0(ifp, nd);
 176         return nd;
 177 }
 178
 179 void
 180 nd6_ifdetach(struct nd_ifinfo *nd)
 181 {
 182         kfree(nd, M_IP6NDP);
 183 }
 184
 185 /*
 186  * Reset ND level link MTU. This function is called when the physical MTU
 187  * changes, which means we might have to adjust the ND level MTU.
 188  */
 189 void
 190 nd6_setmtu(struct ifnet *ifp)
 191 {
 192         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 193 }
 194
 195 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 196 void
 197 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 198 {
 199         u_long oldmaxmtu;
 200         u_long oldlinkmtu;
 201
 202         oldmaxmtu = ndi->maxmtu;
 203         oldlinkmtu = ndi->linkmtu;
 204
 205         switch (ifp->if_type) {
 206         case IFT_ETHER:
 207                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 208                 break;
 209         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 210                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 211                 break;
 212 #ifdef IFT_IEEE80211
 213         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 214                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 215                 break;
 216 #endif
 217         default:
 218                 ndi->maxmtu = ifp->if_mtu;
 219                 break;
 220         }
 221
 222         if (oldmaxmtu != ndi->maxmtu) {
 223                 /*
 224                  * If the ND level MTU is not set yet, or if the maxmtu
 225                  * is reset to a smaller value than the ND level MTU,
 226                  * also reset the ND level MTU.
 227                  */
 228                 if (ndi->linkmtu == 0 ||
 229                     ndi->maxmtu < ndi->linkmtu) {
 230                         ndi->linkmtu = ndi->maxmtu;
 231                         /* also adjust in6_maxmtu if necessary. */
 232                         if (oldlinkmtu == 0) {
 233                                 /*
 234                                  * XXX: the case analysis is grotty, but
 235                                  * it is not efficient to call in6_setmaxmtu()
 236                                  * here when we are during the initialization
 237                                  * procedure.
 238                                  */
 239                                 if (in6_maxmtu < ndi->linkmtu)
 240                                         in6_maxmtu = ndi->linkmtu;
 241                         } else
 242                                 in6_setmaxmtu();
 243                 }
 244         }
 245 #undef MIN
 246 }
 247
 248 void
 249 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 250 {
 251         bzero(ndopts, sizeof(*ndopts));
 252         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 253         ndopts->nd_opts_last
 254                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 255
 256         if (icmp6len == 0) {
 257                 ndopts->nd_opts_done = 1;
 258                 ndopts->nd_opts_search = NULL;
 259         }
 260 }
 261
 262 /*
 263  * Take one ND option.
 264  */
 265 struct nd_opt_hdr *
 266 nd6_option(union nd_opts *ndopts)
 267 {
 268         struct nd_opt_hdr *nd_opt;
 269         int olen;
 270
 271         if (!ndopts)
 272                 panic("ndopts == NULL in nd6_option");
 273         if (!ndopts->nd_opts_last)
 274                 panic("uninitialized ndopts in nd6_option");
 275         if (!ndopts->nd_opts_search)
 276                 return NULL;
 277         if (ndopts->nd_opts_done)
 278                 return NULL;
 279
 280         nd_opt = ndopts->nd_opts_search;
 281
 282         /* make sure nd_opt_len is inside the buffer */
 283         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 284                 bzero(ndopts, sizeof(*ndopts));
 285                 return NULL;
 286         }
 287
 288         olen = nd_opt->nd_opt_len << 3;
 289         if (olen == 0) {
 290                 /*
 291                  * Message validation requires that all included
 292                  * options have a length that is greater than zero.
 293                  */
 294                 bzero(ndopts, sizeof(*ndopts));
 295                 return NULL;
 296         }
 297
 298         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 299         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 300                 /* option overruns the end of buffer, invalid */
 301                 bzero(ndopts, sizeof(*ndopts));
 302                 return NULL;
 303         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 304                 /* reached the end of options chain */
 305                 ndopts->nd_opts_done = 1;
 306                 ndopts->nd_opts_search = NULL;
 307         }
 308         return nd_opt;
 309 }
 310
 311 /*
 312  * Parse multiple ND options.
 313  * This function is much easier to use, for ND routines that do not need
 314  * multiple options of the same type.
 315  */
 316 int
 317 nd6_options(union nd_opts *ndopts)
 318 {
 319         struct nd_opt_hdr *nd_opt;
 320         int i = 0;
 321
 322         if (!ndopts)
 323                 panic("ndopts == NULL in nd6_options");
 324         if (!ndopts->nd_opts_last)
 325                 panic("uninitialized ndopts in nd6_options");
 326         if (!ndopts->nd_opts_search)
 327                 return 0;
 328
 329         while (1) {
 330                 nd_opt = nd6_option(ndopts);
 331                 if (!nd_opt && !ndopts->nd_opts_last) {
 332                         /*
 333                          * Message validation requires that all included
 334                          * options have a length that is greater than zero.
 335                          */
 336                         icmp6stat.icp6s_nd_badopt++;
 337                         bzero(ndopts, sizeof(*ndopts));
 338                         return -1;
 339                 }
 340
 341                 if (!nd_opt)
 342                         goto skip1;
 343
 344                 switch (nd_opt->nd_opt_type) {
 345                 case ND_OPT_SOURCE_LINKADDR:
 346                 case ND_OPT_TARGET_LINKADDR:
 347                 case ND_OPT_MTU:
 348                 case ND_OPT_REDIRECTED_HEADER:
 349                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 350                                 nd6log((LOG_INFO,
 351                                     "duplicated ND6 option found (type=%d)\n",
 352                                     nd_opt->nd_opt_type));
 353                                 /* XXX bark? */
 354                         } else {
 355                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 356                                         = nd_opt;
 357                         }
 358                         break;
 359                 case ND_OPT_PREFIX_INFORMATION:
 360                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 361                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 362                                         = nd_opt;
 363                         }
 364                         ndopts->nd_opts_pi_end =
 365                                 (struct nd_opt_prefix_info *)nd_opt;
 366                         break;
 367                 default:
 368                         /*
 369                          * Unknown options must be silently ignored,
 370                          * to accomodate future extension to the protocol.
 371                          */
 372                         nd6log((LOG_DEBUG,
 373                             "nd6_options: unsupported option %d - "
 374                             "option ignored\n", nd_opt->nd_opt_type));
 375                 }
 376
 377 skip1:
 378                 i++;
 379                 if (i > nd6_maxndopt) {
 380                         icmp6stat.icp6s_nd_toomanyopt++;
 381                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 382                         break;
 383                 }
 384
 385                 if (ndopts->nd_opts_done)
 386                         break;
 387         }
 388
 389         return 0;
 390 }
 391
 392 /*
 393  * ND6 timer routine to expire default route list and prefix list
 394  */
 395 void
 396 nd6_timer(void *ignored_arg)
 397 {
 398         struct llinfo_nd6 *ln;
 399         struct nd_defrouter *dr;
 400         struct nd_prefix *pr;
 401         struct ifnet *ifp;
 402         struct in6_ifaddr *ia6, *nia6;
 403
 404         mtx_lock(&nd6_mtx);
 405         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 406                       nd6_timer, NULL);
 407
 408         ln = llinfo_nd6.ln_next;
 409         while (ln && ln != &llinfo_nd6) {
 410                 struct rtentry *rt;
 411                 struct sockaddr_in6 *dst;
 412                 struct llinfo_nd6 *next = ln->ln_next;
 413                 /* XXX: used for the DELAY case only: */
 414                 struct nd_ifinfo *ndi = NULL;
 415
 416                 if ((rt = ln->ln_rt) == NULL) {
 417                         ln = next;
 418                         continue;
 419                 }
 420                 if ((ifp = rt->rt_ifp) == NULL) {
 421                         ln = next;
 422                         continue;
 423                 }
 424                 ndi = ND_IFINFO(ifp);
 425                 dst = (struct sockaddr_in6 *)rt_key(rt);
 426
 427                 if (ln->ln_expire > time_uptime) {
 428                         ln = next;
 429                         continue;
 430                 }
 431
 432                 /* sanity check */
 433                 if (!rt)
 434                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 435                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 436                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 437                               rt->rt_llinfo, ln);
 438                 if (!dst)
 439                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 440
 441                 switch (ln->ln_state) {
 442                 case ND6_LLINFO_INCOMPLETE:
 443                         if (ln->ln_asked < nd6_mmaxtries) {
 444                                 ln->ln_asked++;
 445                                 ln->ln_expire = time_uptime +
 446                                         ND_IFINFO(ifp)->retrans / 1000;
 447                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 448                                         ln, 0);
 449                         } else {
 450                                 struct mbuf *m = ln->ln_hold;
 451                                 if (m) {
 452                                         if (rt->rt_ifp) {
 453                                                 /*
 454                                                  * Fake rcvif to make ICMP error
 455                                                  * more helpful in diagnosing
 456                                                  * for the receiver.
 457                                                  * XXX: should we consider
 458                                                  * older rcvif?
 459                                                  */
 460                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 461                                         }
 462                                         icmp6_error(m, ICMP6_DST_UNREACH,
 463                                                     ICMP6_DST_UNREACH_ADDR, 0);
 464                                         ln->ln_hold = NULL;
 465                                 }
 466                                 next = nd6_free(rt);
 467                         }
 468                         break;
 469                 case ND6_LLINFO_REACHABLE:
 470                         if (ln->ln_expire) {
 471                                 ln->ln_state = ND6_LLINFO_STALE;
 472                                 ln->ln_expire = time_uptime + nd6_gctimer;
 473                         }
 474                         break;
 475
 476                 case ND6_LLINFO_STALE:
 477                         /* Garbage Collection(RFC 2461 5.3) */
 478                         if (ln->ln_expire)
 479                                 next = nd6_free(rt);
 480                         break;
 481
 482                 case ND6_LLINFO_DELAY:
 483                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 484                                 /* We need NUD */
 485                                 ln->ln_asked = 1;
 486                                 ln->ln_state = ND6_LLINFO_PROBE;
 487                                 ln->ln_expire = time_uptime +
 488                                         ndi->retrans / 1000;
 489                                 nd6_ns_output(ifp, &dst->sin6_addr,
 490                                               &dst->sin6_addr,
 491                                               ln, 0);
 492                         } else {
 493                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 494                                 ln->ln_expire = time_uptime + nd6_gctimer;
 495                         }
 496                         break;
 497                 case ND6_LLINFO_PROBE:
 498                         if (ln->ln_asked < nd6_umaxtries) {
 499                                 ln->ln_asked++;
 500                                 ln->ln_expire = time_uptime +
 501                                         ND_IFINFO(ifp)->retrans / 1000;
 502                                 nd6_ns_output(ifp, &dst->sin6_addr,
 503                                                &dst->sin6_addr, ln, 0);
 504                         } else {
 505                                 next = nd6_free(rt);
 506                         }
 507                         break;
 508                 }
 509                 ln = next;
 510         }
 511
 512         /* expire default router list */
 513         dr = TAILQ_FIRST(&nd_defrouter);
 514         while (dr) {
 515                 if (dr->expire && dr->expire < time_uptime) {
 516                         struct nd_defrouter *t;
 517                         t = TAILQ_NEXT(dr, dr_entry);
 518                         defrtrlist_del(dr);
 519                         dr = t;
 520                 } else {
 521                         dr = TAILQ_NEXT(dr, dr_entry);
 522                 }
 523         }
 524
 525         /*
 526          * expire interface addresses.
 527          * in the past the loop was inside prefix expiry processing.
 528          * However, from a stricter speci-confrmance standpoint, we should
 529          * rather separate address lifetimes and prefix lifetimes.
 530          */
 531 addrloop:
 532         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 533                 nia6 = ia6->ia_next;
 534                 /* check address lifetime */
 535                 if (IFA6_IS_INVALID(ia6)) {
 536                         int regen = 0;
 537
 538                         /*
 539                          * If the expiring address is temporary, try
 540                          * regenerating a new one.  This would be useful when
 541                          * we suspended a laptop PC, then turned it on after a
 542                          * period that could invalidate all temporary
 543                          * addresses.  Although we may have to restart the
 544                          * loop (see below), it must be after purging the
 545                          * address.  Otherwise, we'd see an infinite loop of
 546                          * regeneration.
 547                          */
 548                         if (ip6_use_tempaddr &&
 549                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 550                                 if (regen_tmpaddr(ia6) == 0)
 551                                         regen = 1;
 552                         }
 553
 554                         in6_purgeaddr(&ia6->ia_ifa);
 555
 556                         if (regen)
 557                                 goto addrloop; /* XXX: see below */
 558                 }
 559                 if (IFA6_IS_DEPRECATED(ia6)) {
 560                         int oldflags = ia6->ia6_flags;
 561
 562                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 563
 564                         /*
 565                          * If a temporary address has just become deprecated,
 566                          * regenerate a new one if possible.
 567                          */
 568                         if (ip6_use_tempaddr &&
 569                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 570                             !(oldflags & IN6_IFF_DEPRECATED)) {
 571
 572                                 if (regen_tmpaddr(ia6) == 0) {
 573                                         /*
 574                                          * A new temporary address is
 575                                          * generated.
 576                                          * XXX: this means the address chain
 577                                          * has changed while we are still in
 578                                          * the loop.  Although the change
 579                                          * would not cause disaster (because
 580                                          * it's not a deletion, but an
 581                                          * addition,) we'd rather restart the
 582                                          * loop just for safety.  Or does this
 583                                          * significantly reduce performance??
 584                                          */
 585                                         goto addrloop;
 586                                 }
 587                         }
 588                 } else {
 589                         /*
 590                          * A new RA might have made a deprecated address
 591                          * preferred.
 592                          */
 593                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 594                 }
 595         }
 596
 597         /* expire prefix list */
 598         pr = nd_prefix.lh_first;
 599         while (pr) {
 600                 /*
 601                  * check prefix lifetime.
 602                  * since pltime is just for autoconf, pltime processing for
 603                  * prefix is not necessary.
 604                  */
 605                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
 606                         struct nd_prefix *t;
 607                         t = pr->ndpr_next;
 608
 609                         /*
 610                          * address expiration and prefix expiration are
 611                          * separate.  NEVER perform in6_purgeaddr here.
 612                          */
 613
 614                         prelist_remove(pr);
 615                         pr = t;
 616                 } else
 617                         pr = pr->ndpr_next;
 618         }
 619         mtx_unlock(&nd6_mtx);
 620 }
 621
 622 static int
 623 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 624                                          address */
 625 {
 626         struct ifaddr_container *ifac;
 627         struct ifnet *ifp;
 628         struct in6_ifaddr *public_ifa6 = NULL;
 629
 630         ifp = ia6->ia_ifa.ifa_ifp;
 631         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 632                 struct ifaddr *ifa = ifac->ifa;
 633                 struct in6_ifaddr *it6;
 634
 635                 if (ifa->ifa_addr->sa_family != AF_INET6)
 636                         continue;
 637
 638                 it6 = (struct in6_ifaddr *)ifa;
 639
 640                 /* ignore no autoconf addresses. */
 641                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 642                         continue;
 643
 644                 /* ignore autoconf addresses with different prefixes. */
 645                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 646                         continue;
 647
 648                 /*
 649                  * Now we are looking at an autoconf address with the same
 650                  * prefix as ours.  If the address is temporary and is still
 651                  * preferred, do not create another one.  It would be rare, but
 652                  * could happen, for example, when we resume a laptop PC after
 653                  * a long period.
 654                  */
 655                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 656                     !IFA6_IS_DEPRECATED(it6)) {
 657                         public_ifa6 = NULL;
 658                         break;
 659                 }
 660
 661                 /*
 662                  * This is a public autoconf address that has the same prefix
 663                  * as ours.  If it is preferred, keep it.  We can't break the
 664                  * loop here, because there may be a still-preferred temporary
 665                  * address with the prefix.
 666                  */
 667                 if (!IFA6_IS_DEPRECATED(it6))
 668                     public_ifa6 = it6;
 669         }
 670
 671         if (public_ifa6 != NULL) {
 672                 int e;
 673
 674                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 675                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 676                             " tmp addr,errno=%d\n", e);
 677                         return (-1);
 678                 }
 679                 return (0);
 680         }
 681
 682         return (-1);
 683 }
 684
 685 /*
 686  * Nuke neighbor cache/prefix/default router management table, right before
 687  * ifp goes away.
 688  */
 689 void
 690 nd6_purge(struct ifnet *ifp)
 691 {
 692         struct llinfo_nd6 *ln, *nln;
 693         struct nd_defrouter *dr, *ndr, drany;
 694         struct nd_prefix *pr, *npr;
 695
 696         /* Nuke default router list entries toward ifp */
 697         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 698                 /*
 699                  * The first entry of the list may be stored in
 700                  * the routing table, so we'll delete it later.
 701                  */
 702                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 703                         ndr = TAILQ_NEXT(dr, dr_entry);
 704                         if (dr->ifp == ifp)
 705                                 defrtrlist_del(dr);
 706                 }
 707                 dr = TAILQ_FIRST(&nd_defrouter);
 708                 if (dr->ifp == ifp)
 709                         defrtrlist_del(dr);
 710         }
 711
 712         /* Nuke prefix list entries toward ifp */
 713         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 714                 npr = pr->ndpr_next;
 715                 if (pr->ndpr_ifp == ifp) {
 716                         /*
 717                          * Previously, pr->ndpr_addr is removed as well,
 718                          * but I strongly believe we don't have to do it.
 719                          * nd6_purge() is only called from in6_ifdetach(),
 720                          * which removes all the associated interface addresses
 721                          * by itself.
 722                          * (jinmei@kame.net 20010129)
 723                          */
 724                         prelist_remove(pr);
 725                 }
 726         }
 727
 728         /* cancel default outgoing interface setting */
 729         if (nd6_defifindex == ifp->if_index)
 730                 nd6_setdefaultiface(0);
 731
 732         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 733                 /* refresh default router list */
 734                 bzero(&drany, sizeof(drany));
 735                 defrouter_delreq(&drany, 0);
 736                 defrouter_select();
 737         }
 738
 739         /*
 740          * Nuke neighbor cache entries for the ifp.
 741          * Note that rt->rt_ifp may not be the same as ifp,
 742          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 743          * nd6_rtrequest(), and ip6_input().
 744          */
 745         ln = llinfo_nd6.ln_next;
 746         while (ln && ln != &llinfo_nd6) {
 747                 struct rtentry *rt;
 748                 struct sockaddr_dl *sdl;
 749
 750                 nln = ln->ln_next;
 751                 rt = ln->ln_rt;
 752                 if (rt && rt->rt_gateway &&
 753                     rt->rt_gateway->sa_family == AF_LINK) {
 754                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 755                         if (sdl->sdl_index == ifp->if_index)
 756                                 nln = nd6_free(rt);
 757                 }
 758                 ln = nln;
 759         }
 760 }
 761
 762 struct rtentry *
 763 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 764 {
 765         struct rtentry *rt;
 766         struct sockaddr_in6 sin6;
 767
 768         bzero(&sin6, sizeof(sin6));
 769         sin6.sin6_len = sizeof(struct sockaddr_in6);
 770         sin6.sin6_family = AF_INET6;
 771         sin6.sin6_addr = *addr6;
 772
 773         if (create)
 774                 rt = rtlookup((struct sockaddr *)&sin6);
 775         else
 776                 rt = rtpurelookup((struct sockaddr *)&sin6);
 777         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 778                 /*
 779                  * This is the case for the default route.
 780                  * If we want to create a neighbor cache for the address, we
 781                  * should free the route for the destination and allocate an
 782                  * interface route.
 783                  */
 784                 if (create) {
 785                         --rt->rt_refcnt;
 786                         rt = NULL;
 787                 }
 788         }
 789         if (!rt) {
 790                 if (create && ifp) {
 791                         int e;
 792
 793                         /*
 794                          * If no route is available and create is set,
 795                          * we allocate a host route for the destination
 796                          * and treat it like an interface route.
 797                          * This hack is necessary for a neighbor which can't
 798                          * be covered by our own prefix.
 799                          */
 800                         struct ifaddr *ifa =
 801                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 802                         if (ifa == NULL)
 803                                 return (NULL);
 804
 805                         /*
 806                          * Create a new route.  RTF_LLINFO is necessary
 807                          * to create a Neighbor Cache entry for the
 808                          * destination in nd6_rtrequest which will be
 809                          * called in rtrequest via ifa->ifa_rtrequest.
 810                          */
 811                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 812                                            ifa->ifa_addr,
 813                                            (struct sockaddr *)&all1_sa,
 814                                            (ifa->ifa_flags |
 815                                             RTF_HOST | RTF_LLINFO) &
 816                                            ~RTF_CLONING,
 817                                            &rt)) != 0)
 818                                 log(LOG_ERR,
 819                                     "nd6_lookup: failed to add route for a "
 820                                     "neighbor(%s), errno=%d\n",
 821                                     ip6_sprintf(addr6), e);
 822                         if (rt == NULL)
 823                                 return (NULL);
 824                         if (rt->rt_llinfo) {
 825                                 struct llinfo_nd6 *ln =
 826                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 827                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 828                         }
 829                 } else
 830                         return (NULL);
 831         }
 832         rt->rt_refcnt--;
 833         /*
 834          * Validation for the entry.
 835          * Note that the check for rt_llinfo is necessary because a cloned
 836          * route from a parent route that has the L flag (e.g. the default
 837          * route to a p2p interface) may have the flag, too, while the
 838          * destination is not actually a neighbor.
 839          * XXX: we can't use rt->rt_ifp to check for the interface, since
 840          *      it might be the loopback interface if the entry is for our
 841          *      own address on a non-loopback interface. Instead, we should
 842          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 843          *      interface.
 844          */
 845         if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
 846             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 847             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 848                 if (create) {
 849                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 850                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 851                         /* xxx more logs... kazu */
 852                 }
 853                 return (NULL);
 854         }
 855         return (rt);
 856 }
 857
 858 /*
 859  * Detect if a given IPv6 address identifies a neighbor on a given link.
 860  * XXX: should take care of the destination of a p2p link?
 861  */
 862 int
 863 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 864 {
 865         struct ifaddr_container *ifac;
 866         int i;
 867
 868 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 869 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 870
 871         /*
 872          * A link-local address is always a neighbor.
 873          * XXX: we should use the sin6_scope_id field rather than the embedded
 874          * interface index.
 875          */
 876         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 877             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 878                 return (1);
 879
 880         /*
 881          * If the address matches one of our addresses,
 882          * it should be a neighbor.
 883          */
 884         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 885                 struct ifaddr *ifa = ifac->ifa;
 886
 887                 if (ifa->ifa_addr->sa_family != AF_INET6)
 888                         next: continue;
 889
 890                 for (i = 0; i < 4; i++) {
 891                         if ((IFADDR6(ifa).s6_addr32[i] ^
 892                              addr->sin6_addr.s6_addr32[i]) &
 893                             IFMASK6(ifa).s6_addr32[i])
 894                                 goto next;
 895                 }
 896                 return (1);
 897         }
 898
 899         /*
 900          * Even if the address matches none of our addresses, it might be
 901          * in the neighbor cache.
 902          */
 903         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 904                 return (1);
 905
 906         return (0);
 907 #undef IFADDR6
 908 #undef IFMASK6
 909 }
 910
 911 /*
 912  * Free an nd6 llinfo entry.
 913  */
 914 struct llinfo_nd6 *
 915 nd6_free(struct rtentry *rt)
 916 {
 917         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 918         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 919         struct nd_defrouter *dr;
 920
 921         /*
 922          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
 923          * even though it is not harmful, it was not really necessary.
 924          */
 925
 926         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 927                 mtx_lock(&nd6_mtx);
 928                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 929                                       rt->rt_ifp);
 930
 931                 if (ln->ln_router || dr) {
 932                         /*
 933                          * rt6_flush must be called whether or not the neighbor
 934                          * is in the Default Router List.
 935                          * See a corresponding comment in nd6_na_input().
 936                          */
 937                         rt6_flush(&in6, rt->rt_ifp);
 938                 }
 939
 940                 if (dr) {
 941                         /*
 942                          * Unreachablity of a router might affect the default
 943                          * router selection and on-link detection of advertised
 944                          * prefixes.
 945                          */
 946
 947                         /*
 948                          * Temporarily fake the state to choose a new default
 949                          * router and to perform on-link determination of
 950                          * prefixes correctly.
 951                          * Below the state will be set correctly,
 952                          * or the entry itself will be deleted.
 953                          */
 954                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 955
 956                         /*
 957                          * Since defrouter_select() does not affect the
 958                          * on-link determination and MIP6 needs the check
 959                          * before the default router selection, we perform
 960                          * the check now.
 961                          */
 962                         pfxlist_onlink_check();
 963
 964                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 965                                 /*
 966                                  * It is used as the current default router,
 967                                  * so we have to move it to the end of the
 968                                  * list and choose a new one.
 969                                  * XXX: it is not very efficient if this is
 970                                  *      the only router.
 971                                  */
 972                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
 973                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
 974
 975                                 defrouter_select();
 976                         }
 977                 }
 978                 mtx_unlock(&nd6_mtx);
 979         }
 980
 981         /*
 982          * Before deleting the entry, remember the next entry as the
 983          * return value.  We need this because pfxlist_onlink_check() above
 984          * might have freed other entries (particularly the old next entry) as
 985          * a side effect (XXX).
 986          */
 987         next = ln->ln_next;
 988
 989         /*
 990          * Detach the route from the routing tree and the list of neighbor
 991          * caches, and disable the route entry not to be used in already
 992          * cached routes.
 993          */
 994         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
 995
 996         return (next);
 997 }
 998
 999 /*
1000  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1001  *
1002  * XXX cost-effective metods?
1003  */
1004 void
1005 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1006 {
1007         struct llinfo_nd6 *ln;
1008
1009         /*
1010          * If the caller specified "rt", use that.  Otherwise, resolve the
1011          * routing table by supplied "dst6".
1012          */
1013         if (!rt) {
1014                 if (!dst6)
1015                         return;
1016                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1017                         return;
1018         }
1019
1020         if ((rt->rt_flags & RTF_GATEWAY) ||
1021             !(rt->rt_flags & RTF_LLINFO) ||
1022             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1023             rt->rt_gateway->sa_family != AF_LINK) {
1024                 /* This is not a host route. */
1025                 return;
1026         }
1027
1028         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1029         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1030                 return;
1031
1032         /*
1033          * if we get upper-layer reachability confirmation many times,
1034          * it is possible we have false information.
1035          */
1036         if (!force) {
1037                 ln->ln_byhint++;
1038                 if (ln->ln_byhint > nd6_maxnudhint)
1039                         return;
1040         }
1041
1042         ln->ln_state = ND6_LLINFO_REACHABLE;
1043         if (ln->ln_expire)
1044                 ln->ln_expire = time_uptime +
1045                         ND_IFINFO(rt->rt_ifp)->reachable;
1046 }
1047
1048 void
1049 nd6_rtrequest(int req, struct rtentry *rt)
1050 {
1051         struct sockaddr *gate = rt->rt_gateway;
1052         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1053         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1054         struct ifnet *ifp = rt->rt_ifp;
1055         struct ifaddr *ifa;
1056
1057         if ((rt->rt_flags & RTF_GATEWAY))
1058                 return;
1059
1060         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1061                 /*
1062                  * This is probably an interface direct route for a link
1063                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1064                  * We do not need special treatment below for such a route.
1065                  * Moreover, the RTF_LLINFO flag which would be set below
1066                  * would annoy the ndp(8) command.
1067                  */
1068                 return;
1069         }
1070
1071         if (req == RTM_RESOLVE &&
1072             (nd6_need_cache(ifp) == 0 || /* stf case */
1073              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1074                 /*
1075                  * FreeBSD and BSD/OS often make a cloned host route based
1076                  * on a less-specific route (e.g. the default route).
1077                  * If the less specific route does not have a "gateway"
1078                  * (this is the case when the route just goes to a p2p or an
1079                  * stf interface), we'll mistakenly make a neighbor cache for
1080                  * the host route, and will see strange neighbor solicitation
1081                  * for the corresponding destination.  In order to avoid the
1082                  * confusion, we check if the destination of the route is
1083                  * a neighbor in terms of neighbor discovery, and stop the
1084                  * process if not.  Additionally, we remove the LLINFO flag
1085                  * so that ndp(8) will not try to get the neighbor information
1086                  * of the destination.
1087                  */
1088                 rt->rt_flags &= ~RTF_LLINFO;
1089                 return;
1090         }
1091
1092         switch (req) {
1093         case RTM_ADD:
1094                 /*
1095                  * There is no backward compatibility :)
1096                  *
1097                  * if (!(rt->rt_flags & RTF_HOST) &&
1098                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1099                  *         rt->rt_flags |= RTF_CLONING;
1100                  */
1101                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1102                         /*
1103                          * Case 1: This route should come from
1104                          * a route to interface.  RTF_LLINFO flag is set
1105                          * for a host route whose destination should be
1106                          * treated as on-link.
1107                          */
1108                         rt_setgate(rt, rt_key(rt),
1109                                    (struct sockaddr *)&null_sdl,
1110                                    RTL_DONTREPORT);
1111                         gate = rt->rt_gateway;
1112                         SDL(gate)->sdl_type = ifp->if_type;
1113                         SDL(gate)->sdl_index = ifp->if_index;
1114                         if (ln)
1115                                 ln->ln_expire = time_uptime;
1116 #if 1
1117                         if (ln && ln->ln_expire == 0) {
1118                                 /* kludge for desktops */
1119 #if 0
1120                                 kprintf("nd6_rtequest: time.tv_sec is zero; "
1121                                        "treat it as 1\n");
1122 #endif
1123                                 ln->ln_expire = 1;
1124                         }
1125 #endif
1126                         if ((rt->rt_flags & RTF_CLONING))
1127                                 break;
1128                 }
1129                 /*
1130                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1131                  * We don't do that here since llinfo is not ready yet.
1132                  *
1133                  * There are also couple of other things to be discussed:
1134                  * - unsolicited NA code needs improvement beforehand
1135                  * - RFC2461 says we MAY send multicast unsolicited NA
1136                  *   (7.2.6 paragraph 4), however, it also says that we
1137                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1138                  *   we don't have anything like it right now.
1139                  *   note that the mechanism needs a mutual agreement
1140                  *   between proxies, which means that we need to implement
1141                  *   a new protocol, or a new kludge.
1142                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1143                  *   we need to check ip6forwarding before sending it.
1144                  *   (or should we allow proxy ND configuration only for
1145                  *   routers?  there's no mention about proxy ND from hosts)
1146                  */
1147 #if 0
1148                 /* XXX it does not work */
1149                 if (rt->rt_flags & RTF_ANNOUNCE)
1150                         nd6_na_output(ifp,
1151                               &SIN6(rt_key(rt))->sin6_addr,
1152                               &SIN6(rt_key(rt))->sin6_addr,
1153                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1154                               1, NULL);
1155 #endif
1156                 /* FALLTHROUGH */
1157         case RTM_RESOLVE:
1158                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1159                         /*
1160                          * Address resolution isn't necessary for a point to
1161                          * point link, so we can skip this test for a p2p link.
1162                          */
1163                         if (gate->sa_family != AF_LINK ||
1164                             gate->sa_len < sizeof(null_sdl)) {
1165                                 log(LOG_DEBUG,
1166                                     "nd6_rtrequest: bad gateway value: %s\n",
1167                                     if_name(ifp));
1168                                 break;
1169                         }
1170                         SDL(gate)->sdl_type = ifp->if_type;
1171                         SDL(gate)->sdl_index = ifp->if_index;
1172                 }
1173                 if (ln != NULL)
1174                         break;  /* This happens on a route change */
1175                 /*
1176                  * Case 2: This route may come from cloning, or a manual route
1177                  * add with a LL address.
1178                  */
1179                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1180                 rt->rt_llinfo = (caddr_t)ln;
1181                 if (!ln) {
1182                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1183                         break;
1184                 }
1185                 nd6_inuse++;
1186                 nd6_allocated++;
1187                 bzero(ln, sizeof(*ln));
1188                 ln->ln_rt = rt;
1189                 /* this is required for "ndp" command. - shin */
1190                 if (req == RTM_ADD) {
1191                         /*
1192                          * gate should have some valid AF_LINK entry,
1193                          * and ln->ln_expire should have some lifetime
1194                          * which is specified by ndp command.
1195                          */
1196                         ln->ln_state = ND6_LLINFO_REACHABLE;
1197                         ln->ln_byhint = 0;
1198                 } else {
1199                         /*
1200                          * When req == RTM_RESOLVE, rt is created and
1201                          * initialized in rtrequest(), so rt_expire is 0.
1202                          */
1203                         ln->ln_state = ND6_LLINFO_NOSTATE;
1204                         ln->ln_expire = time_uptime;
1205                 }
1206                 rt->rt_flags |= RTF_LLINFO;
1207                 ln->ln_next = llinfo_nd6.ln_next;
1208                 llinfo_nd6.ln_next = ln;
1209                 ln->ln_prev = &llinfo_nd6;
1210                 ln->ln_next->ln_prev = ln;
1211
1212                 /*
1213                  * check if rt_key(rt) is one of my address assigned
1214                  * to the interface.
1215                  */
1216                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1217                                           &SIN6(rt_key(rt))->sin6_addr);
1218                 if (ifa) {
1219                         caddr_t macp = nd6_ifptomac(ifp);
1220                         ln->ln_expire = 0;
1221                         ln->ln_state = ND6_LLINFO_REACHABLE;
1222                         ln->ln_byhint = 0;
1223                         if (macp) {
1224                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1225                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1226                         }
1227                         if (nd6_useloopback) {
1228                                 rt->rt_ifp = &loif[0];  /* XXX */
1229                                 /*
1230                                  * Make sure rt_ifa be equal to the ifaddr
1231                                  * corresponding to the address.
1232                                  * We need this because when we refer
1233                                  * rt_ifa->ia6_flags in ip6_input, we assume
1234                                  * that the rt_ifa points to the address instead
1235                                  * of the loopback address.
1236                                  */
1237                                 if (ifa != rt->rt_ifa) {
1238                                         IFAFREE(rt->rt_ifa);
1239                                         IFAREF(ifa);
1240                                         rt->rt_ifa = ifa;
1241                                 }
1242                         }
1243                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1244                         ln->ln_expire = 0;
1245                         ln->ln_state = ND6_LLINFO_REACHABLE;
1246                         ln->ln_byhint = 0;
1247
1248                         /* join solicited node multicast for proxy ND */
1249                         if (ifp->if_flags & IFF_MULTICAST) {
1250                                 struct in6_addr llsol;
1251                                 int error;
1252
1253                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1254                                 llsol.s6_addr16[0] = htons(0xff02);
1255                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1256                                 llsol.s6_addr32[1] = 0;
1257                                 llsol.s6_addr32[2] = htonl(1);
1258                                 llsol.s6_addr8[12] = 0xff;
1259
1260                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1261                                         nd6log((LOG_ERR, "%s: failed to join "
1262                                             "%s (errno=%d)\n", if_name(ifp),
1263                                             ip6_sprintf(&llsol), error));
1264                                 }
1265                         }
1266                 }
1267                 break;
1268
1269         case RTM_DELETE:
1270                 if (!ln)
1271                         break;
1272                 /* leave from solicited node multicast for proxy ND */
1273                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1274                     (ifp->if_flags & IFF_MULTICAST)) {
1275                         struct in6_addr llsol;
1276                         struct in6_multi *in6m;
1277
1278                         llsol = SIN6(rt_key(rt))->sin6_addr;
1279                         llsol.s6_addr16[0] = htons(0xff02);
1280                         llsol.s6_addr16[1] = htons(ifp->if_index);
1281                         llsol.s6_addr32[1] = 0;
1282                         llsol.s6_addr32[2] = htonl(1);
1283                         llsol.s6_addr8[12] = 0xff;
1284
1285                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1286                         if (in6m)
1287                                 in6_delmulti(in6m);
1288                 }
1289                 nd6_inuse--;
1290                 ln->ln_next->ln_prev = ln->ln_prev;
1291                 ln->ln_prev->ln_next = ln->ln_next;
1292                 ln->ln_prev = NULL;
1293                 rt->rt_llinfo = 0;
1294                 rt->rt_flags &= ~RTF_LLINFO;
1295                 if (ln->ln_hold)
1296                         m_freem(ln->ln_hold);
1297                 Free((caddr_t)ln);
1298         }
1299 }
1300
1301 int
1302 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1303 {
1304         struct in6_drlist *drl = (struct in6_drlist *)data;
1305         struct in6_prlist *prl = (struct in6_prlist *)data;
1306         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1307         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1308         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1309         struct nd_defrouter *dr, any;
1310         struct nd_prefix *pr;
1311         struct rtentry *rt;
1312         int i = 0, error = 0;
1313
1314         switch (cmd) {
1315         case SIOCGDRLST_IN6:
1316                 /*
1317                  * obsolete API, use sysctl under net.inet6.icmp6
1318                  */
1319                 bzero(drl, sizeof(*drl));
1320                 mtx_lock(&nd6_mtx);
1321                 dr = TAILQ_FIRST(&nd_defrouter);
1322                 while (dr && i < DRLSTSIZ) {
1323                         drl->defrouter[i].rtaddr = dr->rtaddr;
1324                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1325                                 /* XXX: need to this hack for KAME stack */
1326                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1327                         } else
1328                                 log(LOG_ERR,
1329                                     "default router list contains a "
1330                                     "non-linklocal address(%s)\n",
1331                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1332
1333                         drl->defrouter[i].flags = dr->flags;
1334                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1335                         drl->defrouter[i].expire = dr->expire;
1336                         drl->defrouter[i].if_index = dr->ifp->if_index;
1337                         i++;
1338                         dr = TAILQ_NEXT(dr, dr_entry);
1339                 }
1340                 mtx_unlock(&nd6_mtx);
1341                 break;
1342         case SIOCGPRLST_IN6:
1343                 /*
1344                  * obsolete API, use sysctl under net.inet6.icmp6
1345                  */
1346                 /*
1347                  * XXX meaning of fields, especialy "raflags", is very
1348                  * differnet between RA prefix list and RR/static prefix list.
1349                  * how about separating ioctls into two?
1350                  */
1351                 bzero(prl, sizeof(*prl));
1352                 mtx_lock(&nd6_mtx);
1353                 pr = nd_prefix.lh_first;
1354                 while (pr && i < PRLSTSIZ) {
1355                         struct nd_pfxrouter *pfr;
1356                         int j;
1357
1358                         in6_embedscope(&prl->prefix[i].prefix,
1359                             &pr->ndpr_prefix, NULL, NULL);
1360                         prl->prefix[i].raflags = pr->ndpr_raf;
1361                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1362                         prl->prefix[i].vltime = pr->ndpr_vltime;
1363                         prl->prefix[i].pltime = pr->ndpr_pltime;
1364                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1365                         prl->prefix[i].expire = pr->ndpr_expire;
1366
1367                         pfr = pr->ndpr_advrtrs.lh_first;
1368                         j = 0;
1369                         while (pfr) {
1370                                 if (j < DRLSTSIZ) {
1371 #define RTRADDR prl->prefix[i].advrtr[j]
1372                                         RTRADDR = pfr->router->rtaddr;
1373                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1374                                                 /* XXX: hack for KAME */
1375                                                 RTRADDR.s6_addr16[1] = 0;
1376                                         } else
1377                                                 log(LOG_ERR,
1378                                                     "a router(%s) advertises "
1379                                                     "a prefix with "
1380                                                     "non-link local address\n",
1381                                                     ip6_sprintf(&RTRADDR));
1382 #undef RTRADDR
1383                                 }
1384                                 j++;
1385                                 pfr = pfr->pfr_next;
1386                         }
1387                         prl->prefix[i].advrtrs = j;
1388                         prl->prefix[i].origin = PR_ORIG_RA;
1389
1390                         i++;
1391                         pr = pr->ndpr_next;
1392                 }
1393               {
1394                 struct rr_prefix *rpp;
1395
1396                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1397                      rpp = LIST_NEXT(rpp, rp_entry)) {
1398                         if (i >= PRLSTSIZ)
1399                                 break;
1400                         in6_embedscope(&prl->prefix[i].prefix,
1401                             &pr->ndpr_prefix, NULL, NULL);
1402                         prl->prefix[i].raflags = rpp->rp_raf;
1403                         prl->prefix[i].prefixlen = rpp->rp_plen;
1404                         prl->prefix[i].vltime = rpp->rp_vltime;
1405                         prl->prefix[i].pltime = rpp->rp_pltime;
1406                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1407                         prl->prefix[i].expire = rpp->rp_expire;
1408                         prl->prefix[i].advrtrs = 0;
1409                         prl->prefix[i].origin = rpp->rp_origin;
1410                         i++;
1411                 }
1412               }
1413                 mtx_unlock(&nd6_mtx);
1414
1415                 break;
1416         case OSIOCGIFINFO_IN6:
1417                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1418                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1419                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1420                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1421                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1422                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1423                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1424                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1425                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1426                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1427                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1428                 break;
1429         case SIOCGIFINFO_IN6:
1430                 ndi->ndi = *ND_IFINFO(ifp);
1431                 break;
1432         case SIOCSIFINFO_FLAGS:
1433                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1434                 break;
1435         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1436                 /* flush default router list */
1437                 /*
1438                  * xxx sumikawa: should not delete route if default
1439                  * route equals to the top of default router list
1440                  */
1441                 bzero(&any, sizeof(any));
1442                 defrouter_delreq(&any, 0);
1443                 defrouter_select();
1444                 /* xxx sumikawa: flush prefix list */
1445                 break;
1446         case SIOCSPFXFLUSH_IN6:
1447             {
1448                 /* flush all the prefix advertised by routers */
1449                 struct nd_prefix *pr, *next;
1450
1451                 mtx_lock(&nd6_mtx);
1452                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1453                         struct in6_ifaddr *ia, *ia_next;
1454
1455                         next = pr->ndpr_next;
1456
1457                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1458                                 continue; /* XXX */
1459
1460                         /* do we really have to remove addresses as well? */
1461                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1462                                 /* ia might be removed.  keep the next ptr. */
1463                                 ia_next = ia->ia_next;
1464
1465                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1466                                         continue;
1467
1468                                 if (ia->ia6_ndpr == pr)
1469                                         in6_purgeaddr(&ia->ia_ifa);
1470                         }
1471                         prelist_remove(pr);
1472                 }
1473                 mtx_unlock(&nd6_mtx);
1474                 break;
1475             }
1476         case SIOCSRTRFLUSH_IN6:
1477             {
1478                 /* flush all the default routers */
1479                 struct nd_defrouter *dr, *next;
1480
1481                 mtx_lock(&nd6_mtx);
1482                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1483                         /*
1484                          * The first entry of the list may be stored in
1485                          * the routing table, so we'll delete it later.
1486                          */
1487                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1488                                 next = TAILQ_NEXT(dr, dr_entry);
1489                                 defrtrlist_del(dr);
1490                         }
1491                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1492                 }
1493                 mtx_unlock(&nd6_mtx);
1494                 break;
1495             }
1496         case SIOCGNBRINFO_IN6:
1497             {
1498                 struct llinfo_nd6 *ln;
1499                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1500
1501                 /*
1502                  * XXX: KAME specific hack for scoped addresses
1503                  *      XXXX: for other scopes than link-local?
1504                  */
1505                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1506                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1507                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1508
1509                         if (*idp == 0)
1510                                 *idp = htons(ifp->if_index);
1511                 }
1512
1513                 mtx_lock(&nd6_mtx);
1514                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1515                         error = EINVAL;
1516                         mtx_unlock(&nd6_mtx);
1517                         break;
1518                 }
1519                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1520                 nbi->state = ln->ln_state;
1521                 nbi->asked = ln->ln_asked;
1522                 nbi->isrouter = ln->ln_router;
1523                 nbi->expire = ln->ln_expire;
1524                 mtx_unlock(&nd6_mtx);
1525
1526                 break;
1527             }
1528         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1529                 ndif->ifindex = nd6_defifindex;
1530                 break;
1531         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1532                 return (nd6_setdefaultiface(ndif->ifindex));
1533                 break;
1534         }
1535         return (error);
1536 }
1537
1538 /*
1539  * Create neighbor cache entry and cache link-layer address,
1540  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1541  */
1542 struct rtentry *
1543 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1544                  int lladdrlen,
1545                  int type,      /* ICMP6 type */
1546                  int code       /* type dependent information */)
1547 {
1548         struct rtentry *rt = NULL;
1549         struct llinfo_nd6 *ln = NULL;
1550         int is_newentry;
1551         struct sockaddr_dl *sdl = NULL;
1552         int do_update;
1553         int olladdr;
1554         int llchange;
1555         int newstate = 0;
1556
1557         if (!ifp)
1558                 panic("ifp == NULL in nd6_cache_lladdr");
1559         if (!from)
1560                 panic("from == NULL in nd6_cache_lladdr");
1561
1562         /* nothing must be updated for unspecified address */
1563         if (IN6_IS_ADDR_UNSPECIFIED(from))
1564                 return NULL;
1565
1566         /*
1567          * Validation about ifp->if_addrlen and lladdrlen must be done in
1568          * the caller.
1569          *
1570          * XXX If the link does not have link-layer adderss, what should
1571          * we do? (ifp->if_addrlen == 0)
1572          * Spec says nothing in sections for RA, RS and NA.  There's small
1573          * description on it in NS section (RFC 2461 7.2.3).
1574          */
1575
1576         rt = nd6_lookup(from, 0, ifp);
1577         if (!rt) {
1578 #if 0
1579                 /* nothing must be done if there's no lladdr */
1580                 if (!lladdr || !lladdrlen)
1581                         return NULL;
1582 #endif
1583
1584                 rt = nd6_lookup(from, 1, ifp);
1585                 is_newentry = 1;
1586         } else {
1587                 /* do nothing if static ndp is set */
1588                 if (rt->rt_flags & RTF_STATIC)
1589                         return NULL;
1590                 is_newentry = 0;
1591         }
1592
1593         if (!rt)
1594                 return NULL;
1595         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1596 fail:
1597                 nd6_free(rt);
1598                 return NULL;
1599         }
1600         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1601         if (!ln)
1602                 goto fail;
1603         if (!rt->rt_gateway)
1604                 goto fail;
1605         if (rt->rt_gateway->sa_family != AF_LINK)
1606                 goto fail;
1607         sdl = SDL(rt->rt_gateway);
1608
1609         olladdr = (sdl->sdl_alen) ? 1 : 0;
1610         if (olladdr && lladdr) {
1611                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1612                         llchange = 1;
1613                 else
1614                         llchange = 0;
1615         } else
1616                 llchange = 0;
1617
1618         /*
1619          * newentry olladdr  lladdr  llchange   (*=record)
1620          *      0       n       n       --      (1)
1621          *      0       y       n       --      (2)
1622          *      0       n       y       --      (3) * STALE
1623          *      0       y       y       n       (4) *
1624          *      0       y       y       y       (5) * STALE
1625          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1626          *      1       --      y       --      (7) * STALE
1627          */
1628
1629         if (lladdr) {           /* (3-5) and (7) */
1630                 /*
1631                  * Record source link-layer address
1632                  * XXX is it dependent to ifp->if_type?
1633                  */
1634                 sdl->sdl_alen = ifp->if_addrlen;
1635                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1636         }
1637
1638         if (!is_newentry) {
1639                 if ((!olladdr && lladdr)                /* (3) */
1640                  || (olladdr && lladdr && llchange)) {  /* (5) */
1641                         do_update = 1;
1642                         newstate = ND6_LLINFO_STALE;
1643                 } else                                  /* (1-2,4) */
1644                         do_update = 0;
1645         } else {
1646                 do_update = 1;
1647                 if (!lladdr)                            /* (6) */
1648                         newstate = ND6_LLINFO_NOSTATE;
1649                 else                                    /* (7) */
1650                         newstate = ND6_LLINFO_STALE;
1651         }
1652
1653         if (do_update) {
1654                 /*
1655                  * Update the state of the neighbor cache.
1656                  */
1657                 ln->ln_state = newstate;
1658
1659                 if (ln->ln_state == ND6_LLINFO_STALE) {
1660                         /*
1661                          * XXX: since nd6_output() below will cause
1662                          * state tansition to DELAY and reset the timer,
1663                          * we must set the timer now, although it is actually
1664                          * meaningless.
1665                          */
1666                         ln->ln_expire = time_uptime + nd6_gctimer;
1667
1668                         if (ln->ln_hold) {
1669                                 /*
1670                                  * we assume ifp is not a p2p here, so just
1671                                  * set the 2nd argument as the 1st one.
1672                                  */
1673                                 nd6_output(ifp, ifp, ln->ln_hold,
1674                                            (struct sockaddr_in6 *)rt_key(rt),
1675                                            rt);
1676                                 ln->ln_hold = NULL;
1677                         }
1678                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1679                         /* probe right away */
1680                         ln->ln_expire = time_uptime;
1681                 }
1682         }
1683
1684         /*
1685          * ICMP6 type dependent behavior.
1686          *
1687          * NS: clear IsRouter if new entry
1688          * RS: clear IsRouter
1689          * RA: set IsRouter if there's lladdr
1690          * redir: clear IsRouter if new entry
1691          *
1692          * RA case, (1):
1693          * The spec says that we must set IsRouter in the following cases:
1694          * - If lladdr exist, set IsRouter.  This means (1-5).
1695          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1696          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1697          * A quetion arises for (1) case.  (1) case has no lladdr in the
1698          * neighbor cache, this is similar to (6).
1699          * This case is rare but we figured that we MUST NOT set IsRouter.
1700          *
1701          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1702          *                                                      D R
1703          *      0       n       n       --      (1)     c   ?     s
1704          *      0       y       n       --      (2)     c   s     s
1705          *      0       n       y       --      (3)     c   s     s
1706          *      0       y       y       n       (4)     c   s     s
1707          *      0       y       y       y       (5)     c   s     s
1708          *      1       --      n       --      (6) c   c       c s
1709          *      1       --      y       --      (7) c   c   s   c s
1710          *
1711          *                                      (c=clear s=set)
1712          */
1713         switch (type & 0xff) {
1714         case ND_NEIGHBOR_SOLICIT:
1715                 /*
1716                  * New entry must have is_router flag cleared.
1717                  */
1718                 if (is_newentry)        /* (6-7) */
1719                         ln->ln_router = 0;
1720                 break;
1721         case ND_REDIRECT:
1722                 /*
1723                  * If the icmp is a redirect to a better router, always set the
1724                  * is_router flag. Otherwise, if the entry is newly created,
1725                  * clear the flag. [RFC 2461, sec 8.3]
1726                  */
1727                 if (code == ND_REDIRECT_ROUTER)
1728                         ln->ln_router = 1;
1729                 else if (is_newentry) /* (6-7) */
1730                         ln->ln_router = 0;
1731                 break;
1732         case ND_ROUTER_SOLICIT:
1733                 /*
1734                  * is_router flag must always be cleared.
1735                  */
1736                 ln->ln_router = 0;
1737                 break;
1738         case ND_ROUTER_ADVERT:
1739                 /*
1740                  * Mark an entry with lladdr as a router.
1741                  */
1742                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1743                  || (is_newentry && lladdr)) {                  /* (7) */
1744                         ln->ln_router = 1;
1745                 }
1746                 break;
1747         }
1748
1749         /*
1750          * When the link-layer address of a router changes, select the
1751          * best router again.  In particular, when the neighbor entry is newly
1752          * created, it might affect the selection policy.
1753          * Question: can we restrict the first condition to the "is_newentry"
1754          * case?
1755          * XXX: when we hear an RA from a new router with the link-layer
1756          * address option, defrouter_select() is called twice, since
1757          * defrtrlist_update called the function as well.  However, I believe
1758          * we can compromise the overhead, since it only happens the first
1759          * time.
1760          * XXX: although defrouter_select() should not have a bad effect
1761          * for those are not autoconfigured hosts, we explicitly avoid such
1762          * cases for safety.
1763          */
1764         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1765                 defrouter_select();
1766
1767         return rt;
1768 }
1769
1770 static void
1771 nd6_slowtimo(void *ignored_arg)
1772 {
1773         struct nd_ifinfo *nd6if;
1774         struct ifnet *ifp;
1775
1776         mtx_lock(&nd6_mtx);
1777         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1778                         nd6_slowtimo, NULL);
1779         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1780                 if (ifp->if_afdata[AF_INET6] == NULL)
1781                         continue;
1782                 nd6if = ND_IFINFO(ifp);
1783                 if (nd6if->basereachable && /* already initialized */
1784                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1785                         /*
1786                          * Since reachable time rarely changes by router
1787                          * advertisements, we SHOULD insure that a new random
1788                          * value gets recomputed at least once every few hours.
1789                          * (RFC 2461, 6.3.4)
1790                          */
1791                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1792                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1793                 }
1794         }
1795         mtx_unlock(&nd6_mtx);
1796 }
1797
1798 #define gotoerr(e) { error = (e); goto bad;}
1799
1800 int
1801 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1802            struct sockaddr_in6 *dst, struct rtentry *rt)
1803 {
1804         struct llinfo_nd6 *ln = NULL;
1805         int error = 0;
1806
1807         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1808                 goto sendpkt;
1809
1810         if (nd6_need_cache(ifp) == 0)
1811                 goto sendpkt;
1812
1813         /*
1814          * next hop determination.  This routine is derived from ether_outpout.
1815          */
1816         if (rt != NULL) {
1817                 if (!(rt->rt_flags & RTF_UP)) {
1818                         rt = rtlookup((struct sockaddr *)dst);
1819                         if (rt == NULL)
1820                                 gotoerr(EHOSTUNREACH);
1821                         rt->rt_refcnt--;
1822                         if (rt->rt_ifp != ifp) {
1823                                 /* XXX: loop care? */
1824                                 return nd6_output(ifp, origifp, m, dst, rt);
1825                         }
1826                 }
1827                 if (rt->rt_flags & RTF_GATEWAY) {
1828                         struct sockaddr_in6 *gw6;
1829
1830                         /*
1831                          * We skip link-layer address resolution and NUD
1832                          * if the gateway is not a neighbor from ND point
1833                          * of view, regardless of the value of nd_ifinfo.flags.
1834                          * The second condition is a bit tricky; we skip
1835                          * if the gateway is our own address, which is
1836                          * sometimes used to install a route to a p2p link.
1837                          */
1838                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1839                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1840                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1841                                 /*
1842                                  * We allow this kind of tricky route only
1843                                  * when the outgoing interface is p2p.
1844                                  * XXX: we may need a more generic rule here.
1845                                  */
1846                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1847                                         gotoerr(EHOSTUNREACH);
1848
1849                                 goto sendpkt;
1850                         }
1851
1852                         if (rt->rt_gwroute == NULL) {
1853                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1854                                 if (rt->rt_gwroute == NULL)
1855                                         gotoerr(EHOSTUNREACH);
1856                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1857                                 rtfree(rt->rt_gwroute);
1858                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1859                                 if (rt->rt_gwroute == NULL)
1860                                         gotoerr(EHOSTUNREACH);
1861                         }
1862                 }
1863         }
1864
1865         /*
1866          * Address resolution or Neighbor Unreachability Detection
1867          * for the next hop.
1868          * At this point, the destination of the packet must be a unicast
1869          * or an anycast address(i.e. not a multicast).
1870          */
1871
1872         /* Look up the neighbor cache for the nexthop */
1873         if (rt && (rt->rt_flags & RTF_LLINFO))
1874                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1875         else {
1876                 /*
1877                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1878                  * the condition below is not very efficient.  But we believe
1879                  * it is tolerable, because this should be a rare case.
1880                  */
1881                 if (nd6_is_addr_neighbor(dst, ifp) &&
1882                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1883                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1884         }
1885         if (!ln || !rt) {
1886                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1887                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1888                         log(LOG_DEBUG,
1889                             "nd6_output: can't allocate llinfo for %s "
1890                             "(ln=%p, rt=%p)\n",
1891                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1892                         gotoerr(EIO);   /* XXX: good error? */
1893                 }
1894
1895                 goto sendpkt;   /* send anyway */
1896         }
1897
1898         /* We don't have to do link-layer address resolution on a p2p link. */
1899         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1900             ln->ln_state < ND6_LLINFO_REACHABLE) {
1901                 ln->ln_state = ND6_LLINFO_STALE;
1902                 ln->ln_expire = time_uptime + nd6_gctimer;
1903         }
1904
1905         /*
1906          * The first time we send a packet to a neighbor whose entry is
1907          * STALE, we have to change the state to DELAY and a sets a timer to
1908          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1909          * neighbor unreachability detection on expiration.
1910          * (RFC 2461 7.3.3)
1911          */
1912         if (ln->ln_state == ND6_LLINFO_STALE) {
1913                 ln->ln_asked = 0;
1914                 ln->ln_state = ND6_LLINFO_DELAY;
1915                 ln->ln_expire = time_uptime + nd6_delay;
1916         }
1917
1918         /*
1919          * If the neighbor cache entry has a state other than INCOMPLETE
1920          * (i.e. its link-layer address is already resolved), just
1921          * send the packet.
1922          */
1923         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1924                 goto sendpkt;
1925
1926         /*
1927          * There is a neighbor cache entry, but no ethernet address
1928          * response yet.  Replace the held mbuf (if any) with this
1929          * latest one.
1930          *
1931          * This code conforms to the rate-limiting rule described in Section
1932          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1933          * an NS below.
1934          */
1935         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1936                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1937         if (ln->ln_hold)
1938                 m_freem(ln->ln_hold);
1939         ln->ln_hold = m;
1940         if (ln->ln_expire) {
1941                 if (ln->ln_asked < nd6_mmaxtries &&
1942                     ln->ln_expire < time_uptime) {
1943                         ln->ln_asked++;
1944                         ln->ln_expire = time_uptime +
1945                                 ND_IFINFO(ifp)->retrans / 1000;
1946                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1947                 }
1948         }
1949         return (0);
1950
1951 sendpkt:
1952         if (ifp->if_flags & IFF_LOOPBACK)
1953                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1954         else
1955                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1956         return (error);
1957
1958 bad:
1959         m_freem(m);
1960         return (error);
1961 }
1962 #undef gotoerr
1963
1964 int
1965 nd6_need_cache(struct ifnet *ifp)
1966 {
1967         /*
1968          * XXX: we currently do not make neighbor cache on any interface
1969          * other than Ethernet and GIF.
1970          *
1971          * RFC2893 says:
1972          * - unidirectional tunnels needs no ND
1973          */
1974         switch (ifp->if_type) {
1975         case IFT_ETHER:
1976         case IFT_IEEE1394:
1977 #ifdef IFT_L2VLAN
1978         case IFT_L2VLAN:
1979 #endif
1980 #ifdef IFT_IEEE80211
1981         case IFT_IEEE80211:
1982 #endif
1983 #ifdef IFT_CARP
1984         case IFT_CARP:
1985 #endif
1986         case IFT_GIF:           /* XXX need more cases? */
1987                 return (1);
1988         default:
1989                 return (0);
1990         }
1991 }
1992
1993 int
1994 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1995                 struct sockaddr *dst, u_char *desten)
1996 {
1997         struct sockaddr_dl *sdl;
1998         struct rtentry *rt;
1999
2000
2001         if (m->m_flags & M_MCAST) {
2002                 switch (ifp->if_type) {
2003                 case IFT_ETHER:
2004 #ifdef IFT_L2VLAN
2005         case IFT_L2VLAN:
2006 #endif
2007 #ifdef IFT_IEEE80211
2008                 case IFT_IEEE80211:
2009 #endif
2010                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2011                                                  desten);
2012                         return (1);
2013                 case IFT_IEEE1394:
2014                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2015                         return (1);
2016                 default:
2017                         m_freem(m);
2018                         return (0);
2019                 }
2020         }
2021         if (rt0 == NULL) {
2022                 /* this could happen, if we could not allocate memory */
2023                 m_freem(m);
2024                 return (0);
2025         }
2026         if (rt_llroute(dst, rt0, &rt) != 0) {
2027                 m_freem(m);
2028                 return (0);
2029         }
2030         if (rt->rt_gateway->sa_family != AF_LINK) {
2031                 kprintf("nd6_storelladdr: something odd happens\n");
2032                 m_freem(m);
2033                 return (0);
2034         }
2035         sdl = SDL(rt->rt_gateway);
2036         if (sdl->sdl_alen == 0) {
2037                 /* this should be impossible, but we bark here for debugging */
2038                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2039                 m_freem(m);
2040                 return (0);
2041         }
2042
2043         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2044         return (1);
2045 }
2046
2047 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2048 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2049 #ifdef SYSCTL_DECL
2050 SYSCTL_DECL(_net_inet6_icmp6);
2051 #endif
2052 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2053         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2054 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2055         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2056
2057 static int
2058 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2059 {
2060         int error;
2061         char buf[1024];
2062         struct in6_defrouter *d, *de;
2063         struct nd_defrouter *dr;
2064
2065         if (req->newptr)
2066                 return EPERM;
2067         error = 0;
2068
2069         for (dr = TAILQ_FIRST(&nd_defrouter);
2070              dr;
2071              dr = TAILQ_NEXT(dr, dr_entry)) {
2072                 d = (struct in6_defrouter *)buf;
2073                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2074
2075                 if (d + 1 <= de) {
2076                         bzero(d, sizeof(*d));
2077                         d->rtaddr.sin6_family = AF_INET6;
2078                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2079                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2080                             dr->ifp) != 0)
2081                                 log(LOG_ERR,
2082                                     "scope error in "
2083                                     "default router list (%s)\n",
2084                                     ip6_sprintf(&dr->rtaddr));
2085                         d->flags = dr->flags;
2086                         d->rtlifetime = dr->rtlifetime;
2087                         d->expire = dr->expire;
2088                         d->if_index = dr->ifp->if_index;
2089                 } else
2090                         panic("buffer too short");
2091
2092                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2093                 if (error)
2094                         break;
2095         }
2096         return error;
2097 }
2098
2099 static int
2100 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2101 {
2102         int error;
2103         char buf[1024];
2104         struct in6_prefix *p, *pe;
2105         struct nd_prefix *pr;
2106
2107         if (req->newptr)
2108                 return EPERM;
2109         error = 0;
2110
2111         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2112                 u_short advrtrs;
2113                 size_t advance;
2114                 struct sockaddr_in6 *sin6, *s6;
2115                 struct nd_pfxrouter *pfr;
2116
2117                 p = (struct in6_prefix *)buf;
2118                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2119
2120                 if (p + 1 <= pe) {
2121                         bzero(p, sizeof(*p));
2122                         sin6 = (struct sockaddr_in6 *)(p + 1);
2123
2124                         p->prefix = pr->ndpr_prefix;
2125                         if (in6_recoverscope(&p->prefix,
2126                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2127                                 log(LOG_ERR,
2128                                     "scope error in prefix list (%s)\n",
2129                                     ip6_sprintf(&p->prefix.sin6_addr));
2130                         p->raflags = pr->ndpr_raf;
2131                         p->prefixlen = pr->ndpr_plen;
2132                         p->vltime = pr->ndpr_vltime;
2133                         p->pltime = pr->ndpr_pltime;
2134                         p->if_index = pr->ndpr_ifp->if_index;
2135                         p->expire = pr->ndpr_expire;
2136                         p->refcnt = pr->ndpr_refcnt;
2137                         p->flags = pr->ndpr_stateflags;
2138                         p->origin = PR_ORIG_RA;
2139                         advrtrs = 0;
2140                         for (pfr = pr->ndpr_advrtrs.lh_first;
2141                              pfr;
2142                              pfr = pfr->pfr_next) {
2143                                 if ((void *)&sin6[advrtrs + 1] >
2144                                     (void *)pe) {
2145                                         advrtrs++;
2146                                         continue;
2147                                 }
2148                                 s6 = &sin6[advrtrs];
2149                                 bzero(s6, sizeof(*s6));
2150                                 s6->sin6_family = AF_INET6;
2151                                 s6->sin6_len = sizeof(*sin6);
2152                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2153                                                      pfr->router->ifp) != 0)
2154                                         log(LOG_ERR,
2155                                             "scope error in "
2156                                             "prefix list (%s)\n",
2157                                             ip6_sprintf(&pfr->router->rtaddr));
2158                                 advrtrs++;
2159                         }
2160                         p->advrtrs = advrtrs;
2161                 } else
2162                         panic("buffer too short");
2163
2164                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2165                 error = SYSCTL_OUT(req, buf, advance);
2166                 if (error)
2167                         break;
2168         }
2169         return error;
2170 }