sys/netinet6/nd6.c

   1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
   2 /*      $DragonFly: src/sys/netinet6/nd6.c,v 1.24 2007/08/27 13:15:14 hasso Exp $       */
   3 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   4
   5 /*
   6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. Neither the name of the project nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * XXX
  36  * KAME 970409 note:
  37  * BSD/OS version heavily modifies this code, related to llinfo.
  38  * Since we don't have BSD/OS version of net/route.c in our hand,
  39  * I left the code mostly as it was in 970310.  -- itojun
  40  */
  41
  42 #include "opt_inet.h"
  43 #include "opt_inet6.h"
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/callout.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mbuf.h>
  50 #include <sys/socket.h>
  51 #include <sys/sockio.h>
  52 #include <sys/time.h>
  53 #include <sys/kernel.h>
  54 #include <sys/protosw.h>
  55 #include <sys/errno.h>
  56 #include <sys/syslog.h>
  57 #include <sys/queue.h>
  58 #include <sys/sysctl.h>
  59 #include <sys/thread2.h>
  60
  61 #include <net/if.h>
  62 #include <net/if_dl.h>
  63 #include <net/if_types.h>
  64 #include <net/if_atm.h>
  65 #include <net/route.h>
  66
  67 #include <netinet/in.h>
  68 #include <netinet/if_ether.h>
  69 #include <netinet/if_fddi.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 "use_loop.h"
  78
  79 #include <net/net_osdep.h>
  80
  81 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  82 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  83
  84 #define SIN6(s) ((struct sockaddr_in6 *)s)
  85 #define SDL(s) ((struct sockaddr_dl *)s)
  86
  87 /* timer values */
  88 int     nd6_prune       = 1;    /* walk list every 1 seconds */
  89 int     nd6_delay       = 5;    /* delay first probe time 5 second */
  90 int     nd6_umaxtries   = 3;    /* maximum unicast query */
  91 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
  92 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
  93 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
  94
  95 /* preventing too many loops in ND option parsing */
  96 int nd6_maxndopt = 10;  /* max # of ND options allowed */
  97
  98 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
  99
 100 #ifdef ND6_DEBUG
 101 int nd6_debug = 1;
 102 #else
 103 int nd6_debug = 0;
 104 #endif
 105
 106 /* for debugging? */
 107 static int nd6_inuse, nd6_allocated;
 108
 109 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 110 struct nd_drhead nd_defrouter;
 111 struct nd_prhead nd_prefix = { 0 };
 112
 113 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 114 static struct sockaddr_in6 all1_sa;
 115
 116 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
 117 static void nd6_slowtimo (void *);
 118 static int regen_tmpaddr (struct in6_ifaddr *);
 119
 120 struct callout nd6_slowtimo_ch;
 121 struct callout nd6_timer_ch;
 122 extern struct callout in6_tmpaddrtimer_ch;
 123
 124 void
 125 nd6_init(void)
 126 {
 127         static int nd6_init_done = 0;
 128         int i;
 129
 130         if (nd6_init_done) {
 131                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 132                 return;
 133         }
 134
 135         all1_sa.sin6_family = AF_INET6;
 136         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 137         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 138                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 139
 140         /* initialization of the default router list */
 141         TAILQ_INIT(&nd_defrouter);
 142
 143         nd6_init_done = 1;
 144
 145         /* start timer */
 146         callout_init(&nd6_slowtimo_ch);
 147         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 148             nd6_slowtimo, NULL);
 149 }
 150
 151 struct nd_ifinfo *
 152 nd6_ifattach(struct ifnet *ifp)
 153 {
 154         struct nd_ifinfo *nd;
 155
 156         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
 157         bzero(nd, sizeof(*nd));
 158
 159         nd->initialized = 1;
 160
 161         nd->linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
 162         nd->chlim = IPV6_DEFHLIM;
 163         nd->basereachable = REACHABLE_TIME;
 164         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 165         nd->retrans = RETRANS_TIMER;
 166         nd->receivedra = 0;
 167
 168         /*
 169          * Note that the default value of ip6_accept_rtadv is 0, which means
 170          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 171          * here.
 172          */
 173         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 174
 175         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 176         nd6_setmtu0(ifp, nd);
 177         return nd;
 178 }
 179
 180 void
 181 nd6_ifdetach(struct nd_ifinfo *nd)
 182 {
 183         kfree(nd, M_IP6NDP);
 184 }
 185
 186 /*
 187  * Reset ND level link MTU. This function is called when the physical MTU
 188  * changes, which means we might have to adjust the ND level MTU.
 189  */
 190 void
 191 nd6_setmtu(struct ifnet *ifp)
 192 {
 193         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 194 }
 195
 196 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 197 void
 198 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 199 {
 200         u_long oldmaxmtu;
 201         u_long oldlinkmtu;
 202
 203         oldmaxmtu = ndi->maxmtu;
 204         oldlinkmtu = ndi->linkmtu;
 205
 206         switch (ifp->if_type) {
 207         case IFT_ETHER:
 208                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 209                 break;
 210         case IFT_FDDI:
 211                 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
 212                 break;
 213         case IFT_ATM:
 214                 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
 215                 break;
 216         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 217                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 218                 break;
 219 #ifdef IFT_IEEE80211
 220         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 221                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 222                 break;
 223 #endif
 224         default:
 225                 ndi->maxmtu = ifp->if_mtu;
 226                 break;
 227         }
 228
 229         if (oldmaxmtu != ndi->maxmtu) {
 230                 /*
 231                  * If the ND level MTU is not set yet, or if the maxmtu
 232                  * is reset to a smaller value than the ND level MTU,
 233                  * also reset the ND level MTU.
 234                  */
 235                 if (ndi->linkmtu == 0 ||
 236                     ndi->maxmtu < ndi->linkmtu) {
 237                         ndi->linkmtu = ndi->maxmtu;
 238                         /* also adjust in6_maxmtu if necessary. */
 239                         if (oldlinkmtu == 0) {
 240                                 /*
 241                                  * XXX: the case analysis is grotty, but
 242                                  * it is not efficient to call in6_setmaxmtu()
 243                                  * here when we are during the initialization
 244                                  * procedure.
 245                                  */
 246                                 if (in6_maxmtu < ndi->linkmtu)
 247                                         in6_maxmtu = ndi->linkmtu;
 248                         } else
 249                                 in6_setmaxmtu();
 250                 }
 251         }
 252 #undef MIN
 253 }
 254
 255 void
 256 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 257 {
 258         bzero(ndopts, sizeof(*ndopts));
 259         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 260         ndopts->nd_opts_last
 261                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 262
 263         if (icmp6len == 0) {
 264                 ndopts->nd_opts_done = 1;
 265                 ndopts->nd_opts_search = NULL;
 266         }
 267 }
 268
 269 /*
 270  * Take one ND option.
 271  */
 272 struct nd_opt_hdr *
 273 nd6_option(union nd_opts *ndopts)
 274 {
 275         struct nd_opt_hdr *nd_opt;
 276         int olen;
 277
 278         if (!ndopts)
 279                 panic("ndopts == NULL in nd6_option");
 280         if (!ndopts->nd_opts_last)
 281                 panic("uninitialized ndopts in nd6_option");
 282         if (!ndopts->nd_opts_search)
 283                 return NULL;
 284         if (ndopts->nd_opts_done)
 285                 return NULL;
 286
 287         nd_opt = ndopts->nd_opts_search;
 288
 289         /* make sure nd_opt_len is inside the buffer */
 290         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 291                 bzero(ndopts, sizeof(*ndopts));
 292                 return NULL;
 293         }
 294
 295         olen = nd_opt->nd_opt_len << 3;
 296         if (olen == 0) {
 297                 /*
 298                  * Message validation requires that all included
 299                  * options have a length that is greater than zero.
 300                  */
 301                 bzero(ndopts, sizeof(*ndopts));
 302                 return NULL;
 303         }
 304
 305         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 306         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 307                 /* option overruns the end of buffer, invalid */
 308                 bzero(ndopts, sizeof(*ndopts));
 309                 return NULL;
 310         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 311                 /* reached the end of options chain */
 312                 ndopts->nd_opts_done = 1;
 313                 ndopts->nd_opts_search = NULL;
 314         }
 315         return nd_opt;
 316 }
 317
 318 /*
 319  * Parse multiple ND options.
 320  * This function is much easier to use, for ND routines that do not need
 321  * multiple options of the same type.
 322  */
 323 int
 324 nd6_options(union nd_opts *ndopts)
 325 {
 326         struct nd_opt_hdr *nd_opt;
 327         int i = 0;
 328
 329         if (!ndopts)
 330                 panic("ndopts == NULL in nd6_options");
 331         if (!ndopts->nd_opts_last)
 332                 panic("uninitialized ndopts in nd6_options");
 333         if (!ndopts->nd_opts_search)
 334                 return 0;
 335
 336         while (1) {
 337                 nd_opt = nd6_option(ndopts);
 338                 if (!nd_opt && !ndopts->nd_opts_last) {
 339                         /*
 340                          * Message validation requires that all included
 341                          * options have a length that is greater than zero.
 342                          */
 343                         icmp6stat.icp6s_nd_badopt++;
 344                         bzero(ndopts, sizeof(*ndopts));
 345                         return -1;
 346                 }
 347
 348                 if (!nd_opt)
 349                         goto skip1;
 350
 351                 switch (nd_opt->nd_opt_type) {
 352                 case ND_OPT_SOURCE_LINKADDR:
 353                 case ND_OPT_TARGET_LINKADDR:
 354                 case ND_OPT_MTU:
 355                 case ND_OPT_REDIRECTED_HEADER:
 356                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 357                                 nd6log((LOG_INFO,
 358                                     "duplicated ND6 option found (type=%d)\n",
 359                                     nd_opt->nd_opt_type));
 360                                 /* XXX bark? */
 361                         } else {
 362                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 363                                         = nd_opt;
 364                         }
 365                         break;
 366                 case ND_OPT_PREFIX_INFORMATION:
 367                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 368                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 369                                         = nd_opt;
 370                         }
 371                         ndopts->nd_opts_pi_end =
 372                                 (struct nd_opt_prefix_info *)nd_opt;
 373                         break;
 374                 default:
 375                         /*
 376                          * Unknown options must be silently ignored,
 377                          * to accomodate future extension to the protocol.
 378                          */
 379                         nd6log((LOG_DEBUG,
 380                             "nd6_options: unsupported option %d - "
 381                             "option ignored\n", nd_opt->nd_opt_type));
 382                 }
 383
 384 skip1:
 385                 i++;
 386                 if (i > nd6_maxndopt) {
 387                         icmp6stat.icp6s_nd_toomanyopt++;
 388                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 389                         break;
 390                 }
 391
 392                 if (ndopts->nd_opts_done)
 393                         break;
 394         }
 395
 396         return 0;
 397 }
 398
 399 /*
 400  * ND6 timer routine to expire default route list and prefix list
 401  */
 402 void
 403 nd6_timer(void *ignored_arg)
 404 {
 405         struct llinfo_nd6 *ln;
 406         struct nd_defrouter *dr;
 407         struct nd_prefix *pr;
 408         struct ifnet *ifp;
 409         struct in6_ifaddr *ia6, *nia6;
 410         struct in6_addrlifetime *lt6;
 411
 412         crit_enter();
 413         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 414                       nd6_timer, NULL);
 415
 416         ln = llinfo_nd6.ln_next;
 417         while (ln && ln != &llinfo_nd6) {
 418                 struct rtentry *rt;
 419                 struct sockaddr_in6 *dst;
 420                 struct llinfo_nd6 *next = ln->ln_next;
 421                 /* XXX: used for the DELAY case only: */
 422                 struct nd_ifinfo *ndi = NULL;
 423
 424                 if ((rt = ln->ln_rt) == NULL) {
 425                         ln = next;
 426                         continue;
 427                 }
 428                 if ((ifp = rt->rt_ifp) == NULL) {
 429                         ln = next;
 430                         continue;
 431                 }
 432                 ndi = ND_IFINFO(ifp);
 433                 dst = (struct sockaddr_in6 *)rt_key(rt);
 434
 435                 if (ln->ln_expire > time_second) {
 436                         ln = next;
 437                         continue;
 438                 }
 439
 440                 /* sanity check */
 441                 if (!rt)
 442                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 443                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 444                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 445                               rt->rt_llinfo, ln);
 446                 if (!dst)
 447                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 448
 449                 switch (ln->ln_state) {
 450                 case ND6_LLINFO_INCOMPLETE:
 451                         if (ln->ln_asked < nd6_mmaxtries) {
 452                                 ln->ln_asked++;
 453                                 ln->ln_expire = time_second +
 454                                         ND_IFINFO(ifp)->retrans / 1000;
 455                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 456                                         ln, 0);
 457                         } else {
 458                                 struct mbuf *m = ln->ln_hold;
 459                                 if (m) {
 460                                         if (rt->rt_ifp) {
 461                                                 /*
 462                                                  * Fake rcvif to make ICMP error
 463                                                  * more helpful in diagnosing
 464                                                  * for the receiver.
 465                                                  * XXX: should we consider
 466                                                  * older rcvif?
 467                                                  */
 468                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 469                                         }
 470                                         icmp6_error(m, ICMP6_DST_UNREACH,
 471                                                     ICMP6_DST_UNREACH_ADDR, 0);
 472                                         ln->ln_hold = NULL;
 473                                 }
 474                                 next = nd6_free(rt);
 475                         }
 476                         break;
 477                 case ND6_LLINFO_REACHABLE:
 478                         if (ln->ln_expire) {
 479                                 ln->ln_state = ND6_LLINFO_STALE;
 480                                 ln->ln_expire = time_second + nd6_gctimer;
 481                         }
 482                         break;
 483
 484                 case ND6_LLINFO_STALE:
 485                         /* Garbage Collection(RFC 2461 5.3) */
 486                         if (ln->ln_expire)
 487                                 next = nd6_free(rt);
 488                         break;
 489
 490                 case ND6_LLINFO_DELAY:
 491                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 492                                 /* We need NUD */
 493                                 ln->ln_asked = 1;
 494                                 ln->ln_state = ND6_LLINFO_PROBE;
 495                                 ln->ln_expire = time_second +
 496                                         ndi->retrans / 1000;
 497                                 nd6_ns_output(ifp, &dst->sin6_addr,
 498                                               &dst->sin6_addr,
 499                                               ln, 0);
 500                         } else {
 501                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 502                                 ln->ln_expire = time_second + nd6_gctimer;
 503                         }
 504                         break;
 505                 case ND6_LLINFO_PROBE:
 506                         if (ln->ln_asked < nd6_umaxtries) {
 507                                 ln->ln_asked++;
 508                                 ln->ln_expire = time_second +
 509                                         ND_IFINFO(ifp)->retrans / 1000;
 510                                 nd6_ns_output(ifp, &dst->sin6_addr,
 511                                                &dst->sin6_addr, ln, 0);
 512                         } else {
 513                                 next = nd6_free(rt);
 514                         }
 515                         break;
 516                 }
 517                 ln = next;
 518         }
 519
 520         /* expire default router list */
 521         dr = TAILQ_FIRST(&nd_defrouter);
 522         while (dr) {
 523                 if (dr->expire && dr->expire < time_second) {
 524                         struct nd_defrouter *t;
 525                         t = TAILQ_NEXT(dr, dr_entry);
 526                         defrtrlist_del(dr);
 527                         dr = t;
 528                 } else {
 529                         dr = TAILQ_NEXT(dr, dr_entry);
 530                 }
 531         }
 532
 533         /*
 534          * expire interface addresses.
 535          * in the past the loop was inside prefix expiry processing.
 536          * However, from a stricter speci-confrmance standpoint, we should
 537          * rather separate address lifetimes and prefix lifetimes.
 538          */
 539 addrloop:
 540         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 541                 nia6 = ia6->ia_next;
 542                 /* check address lifetime */
 543                 lt6 = &ia6->ia6_lifetime;
 544                 if (IFA6_IS_INVALID(ia6)) {
 545                         int regen = 0;
 546
 547                         /*
 548                          * If the expiring address is temporary, try
 549                          * regenerating a new one.  This would be useful when
 550                          * we suspended a laptop PC, then turned it on after a
 551                          * period that could invalidate all temporary
 552                          * addresses.  Although we may have to restart the
 553                          * loop (see below), it must be after purging the
 554                          * address.  Otherwise, we'd see an infinite loop of
 555                          * regeneration.
 556                          */
 557                         if (ip6_use_tempaddr &&
 558                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 559                                 if (regen_tmpaddr(ia6) == 0)
 560                                         regen = 1;
 561                         }
 562
 563                         in6_purgeaddr(&ia6->ia_ifa);
 564
 565                         if (regen)
 566                                 goto addrloop; /* XXX: see below */
 567                 }
 568                 if (IFA6_IS_DEPRECATED(ia6)) {
 569                         int oldflags = ia6->ia6_flags;
 570
 571                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 572
 573                         /*
 574                          * If a temporary address has just become deprecated,
 575                          * regenerate a new one if possible.
 576                          */
 577                         if (ip6_use_tempaddr &&
 578                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 579                             !(oldflags & IN6_IFF_DEPRECATED)) {
 580
 581                                 if (regen_tmpaddr(ia6) == 0) {
 582                                         /*
 583                                          * A new temporary address is
 584                                          * generated.
 585                                          * XXX: this means the address chain
 586                                          * has changed while we are still in
 587                                          * the loop.  Although the change
 588                                          * would not cause disaster (because
 589                                          * it's not a deletion, but an
 590                                          * addition,) we'd rather restart the
 591                                          * loop just for safety.  Or does this
 592                                          * significantly reduce performance??
 593                                          */
 594                                         goto addrloop;
 595                                 }
 596                         }
 597                 } else {
 598                         /*
 599                          * A new RA might have made a deprecated address
 600                          * preferred.
 601                          */
 602                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 603                 }
 604         }
 605
 606         /* expire prefix list */
 607         pr = nd_prefix.lh_first;
 608         while (pr) {
 609                 /*
 610                  * check prefix lifetime.
 611                  * since pltime is just for autoconf, pltime processing for
 612                  * prefix is not necessary.
 613                  */
 614                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
 615                         struct nd_prefix *t;
 616                         t = pr->ndpr_next;
 617
 618                         /*
 619                          * address expiration and prefix expiration are
 620                          * separate.  NEVER perform in6_purgeaddr here.
 621                          */
 622
 623                         prelist_remove(pr);
 624                         pr = t;
 625                 } else
 626                         pr = pr->ndpr_next;
 627         }
 628         crit_exit();
 629 }
 630
 631 static int
 632 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 633                                          address */
 634 {
 635         struct ifaddr *ifa;
 636         struct ifnet *ifp;
 637         struct in6_ifaddr *public_ifa6 = NULL;
 638
 639         ifp = ia6->ia_ifa.ifa_ifp;
 640         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_list) {
 641                 struct in6_ifaddr *it6;
 642
 643                 if (ifa->ifa_addr->sa_family != AF_INET6)
 644                         continue;
 645
 646                 it6 = (struct in6_ifaddr *)ifa;
 647
 648                 /* ignore no autoconf addresses. */
 649                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 650                         continue;
 651
 652                 /* ignore autoconf addresses with different prefixes. */
 653                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 654                         continue;
 655
 656                 /*
 657                  * Now we are looking at an autoconf address with the same
 658                  * prefix as ours.  If the address is temporary and is still
 659                  * preferred, do not create another one.  It would be rare, but
 660                  * could happen, for example, when we resume a laptop PC after
 661                  * a long period.
 662                  */
 663                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 664                     !IFA6_IS_DEPRECATED(it6)) {
 665                         public_ifa6 = NULL;
 666                         break;
 667                 }
 668
 669                 /*
 670                  * This is a public autoconf address that has the same prefix
 671                  * as ours.  If it is preferred, keep it.  We can't break the
 672                  * loop here, because there may be a still-preferred temporary
 673                  * address with the prefix.
 674                  */
 675                 if (!IFA6_IS_DEPRECATED(it6))
 676                     public_ifa6 = it6;
 677         }
 678
 679         if (public_ifa6 != NULL) {
 680                 int e;
 681
 682                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 683                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 684                             " tmp addr,errno=%d\n", e);
 685                         return (-1);
 686                 }
 687                 return (0);
 688         }
 689
 690         return (-1);
 691 }
 692
 693 /*
 694  * Nuke neighbor cache/prefix/default router management table, right before
 695  * ifp goes away.
 696  */
 697 void
 698 nd6_purge(struct ifnet *ifp)
 699 {
 700         struct llinfo_nd6 *ln, *nln;
 701         struct nd_defrouter *dr, *ndr, drany;
 702         struct nd_prefix *pr, *npr;
 703
 704         /* Nuke default router list entries toward ifp */
 705         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 706                 /*
 707                  * The first entry of the list may be stored in
 708                  * the routing table, so we'll delete it later.
 709                  */
 710                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 711                         ndr = TAILQ_NEXT(dr, dr_entry);
 712                         if (dr->ifp == ifp)
 713                                 defrtrlist_del(dr);
 714                 }
 715                 dr = TAILQ_FIRST(&nd_defrouter);
 716                 if (dr->ifp == ifp)
 717                         defrtrlist_del(dr);
 718         }
 719
 720         /* Nuke prefix list entries toward ifp */
 721         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 722                 npr = pr->ndpr_next;
 723                 if (pr->ndpr_ifp == ifp) {
 724                         /*
 725                          * Previously, pr->ndpr_addr is removed as well,
 726                          * but I strongly believe we don't have to do it.
 727                          * nd6_purge() is only called from in6_ifdetach(),
 728                          * which removes all the associated interface addresses
 729                          * by itself.
 730                          * (jinmei@kame.net 20010129)
 731                          */
 732                         prelist_remove(pr);
 733                 }
 734         }
 735
 736         /* cancel default outgoing interface setting */
 737         if (nd6_defifindex == ifp->if_index)
 738                 nd6_setdefaultiface(0);
 739
 740         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 741                 /* refresh default router list */
 742                 bzero(&drany, sizeof(drany));
 743                 defrouter_delreq(&drany, 0);
 744                 defrouter_select();
 745         }
 746
 747         /*
 748          * Nuke neighbor cache entries for the ifp.
 749          * Note that rt->rt_ifp may not be the same as ifp,
 750          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 751          * nd6_rtrequest(), and ip6_input().
 752          */
 753         ln = llinfo_nd6.ln_next;
 754         while (ln && ln != &llinfo_nd6) {
 755                 struct rtentry *rt;
 756                 struct sockaddr_dl *sdl;
 757
 758                 nln = ln->ln_next;
 759                 rt = ln->ln_rt;
 760                 if (rt && rt->rt_gateway &&
 761                     rt->rt_gateway->sa_family == AF_LINK) {
 762                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 763                         if (sdl->sdl_index == ifp->if_index)
 764                                 nln = nd6_free(rt);
 765                 }
 766                 ln = nln;
 767         }
 768 }
 769
 770 struct rtentry *
 771 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 772 {
 773         struct rtentry *rt;
 774         struct sockaddr_in6 sin6;
 775
 776         bzero(&sin6, sizeof(sin6));
 777         sin6.sin6_len = sizeof(struct sockaddr_in6);
 778         sin6.sin6_family = AF_INET6;
 779         sin6.sin6_addr = *addr6;
 780
 781         if (create)
 782                 rt = rtlookup((struct sockaddr *)&sin6);
 783         else
 784                 rt = rtpurelookup((struct sockaddr *)&sin6);
 785         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 786                 /*
 787                  * This is the case for the default route.
 788                  * If we want to create a neighbor cache for the address, we
 789                  * should free the route for the destination and allocate an
 790                  * interface route.
 791                  */
 792                 if (create) {
 793                         --rt->rt_refcnt;
 794                         rt = NULL;
 795                 }
 796         }
 797         if (!rt) {
 798                 if (create && ifp) {
 799                         int e;
 800
 801                         /*
 802                          * If no route is available and create is set,
 803                          * we allocate a host route for the destination
 804                          * and treat it like an interface route.
 805                          * This hack is necessary for a neighbor which can't
 806                          * be covered by our own prefix.
 807                          */
 808                         struct ifaddr *ifa =
 809                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 810                         if (ifa == NULL)
 811                                 return (NULL);
 812
 813                         /*
 814                          * Create a new route.  RTF_LLINFO is necessary
 815                          * to create a Neighbor Cache entry for the
 816                          * destination in nd6_rtrequest which will be
 817                          * called in rtrequest via ifa->ifa_rtrequest.
 818                          */
 819                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 820                                            ifa->ifa_addr,
 821                                            (struct sockaddr *)&all1_sa,
 822                                            (ifa->ifa_flags |
 823                                             RTF_HOST | RTF_LLINFO) &
 824                                            ~RTF_CLONING,
 825                                            &rt)) != 0)
 826                                 log(LOG_ERR,
 827                                     "nd6_lookup: failed to add route for a "
 828                                     "neighbor(%s), errno=%d\n",
 829                                     ip6_sprintf(addr6), e);
 830                         if (rt == NULL)
 831                                 return (NULL);
 832                         if (rt->rt_llinfo) {
 833                                 struct llinfo_nd6 *ln =
 834                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 835                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 836                         }
 837                 } else
 838                         return (NULL);
 839         }
 840         rt->rt_refcnt--;
 841         /*
 842          * Validation for the entry.
 843          * Note that the check for rt_llinfo is necessary because a cloned
 844          * route from a parent route that has the L flag (e.g. the default
 845          * route to a p2p interface) may have the flag, too, while the
 846          * destination is not actually a neighbor.
 847          * XXX: we can't use rt->rt_ifp to check for the interface, since
 848          *      it might be the loopback interface if the entry is for our
 849          *      own address on a non-loopback interface. Instead, we should
 850          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 851          *      interface.
 852          */
 853         if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
 854             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 855             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 856                 if (create) {
 857                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 858                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 859                         /* xxx more logs... kazu */
 860                 }
 861                 return (NULL);
 862         }
 863         return (rt);
 864 }
 865
 866 /*
 867  * Detect if a given IPv6 address identifies a neighbor on a given link.
 868  * XXX: should take care of the destination of a p2p link?
 869  */
 870 int
 871 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 872 {
 873         struct ifaddr *ifa;
 874         int i;
 875
 876 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 877 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 878
 879         /*
 880          * A link-local address is always a neighbor.
 881          * XXX: we should use the sin6_scope_id field rather than the embedded
 882          * interface index.
 883          */
 884         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 885             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 886                 return (1);
 887
 888         /*
 889          * If the address matches one of our addresses,
 890          * it should be a neighbor.
 891          */
 892         TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
 893                 if (ifa->ifa_addr->sa_family != AF_INET6)
 894                         next: continue;
 895
 896                 for (i = 0; i < 4; i++) {
 897                         if ((IFADDR6(ifa).s6_addr32[i] ^
 898                              addr->sin6_addr.s6_addr32[i]) &
 899                             IFMASK6(ifa).s6_addr32[i])
 900                                 goto next;
 901                 }
 902                 return (1);
 903         }
 904
 905         /*
 906          * Even if the address matches none of our addresses, it might be
 907          * in the neighbor cache.
 908          */
 909         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 910                 return (1);
 911
 912         return (0);
 913 #undef IFADDR6
 914 #undef IFMASK6
 915 }
 916
 917 /*
 918  * Free an nd6 llinfo entry.
 919  */
 920 struct llinfo_nd6 *
 921 nd6_free(struct rtentry *rt)
 922 {
 923         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 924         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 925         struct nd_defrouter *dr;
 926
 927         /*
 928          * we used to have pfctlinput(PRC_HOSTDEAD) here.
 929          * even though it is not harmful, it was not really necessary.
 930          */
 931
 932         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 933                 crit_enter();
 934                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 935                                       rt->rt_ifp);
 936
 937                 if (ln->ln_router || dr) {
 938                         /*
 939                          * rt6_flush must be called whether or not the neighbor
 940                          * is in the Default Router List.
 941                          * See a corresponding comment in nd6_na_input().
 942                          */
 943                         rt6_flush(&in6, rt->rt_ifp);
 944                 }
 945
 946                 if (dr) {
 947                         /*
 948                          * Unreachablity of a router might affect the default
 949                          * router selection and on-link detection of advertised
 950                          * prefixes.
 951                          */
 952
 953                         /*
 954                          * Temporarily fake the state to choose a new default
 955                          * router and to perform on-link determination of
 956                          * prefixes correctly.
 957                          * Below the state will be set correctly,
 958                          * or the entry itself will be deleted.
 959                          */
 960                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 961
 962                         /*
 963                          * Since defrouter_select() does not affect the
 964                          * on-link determination and MIP6 needs the check
 965                          * before the default router selection, we perform
 966                          * the check now.
 967                          */
 968                         pfxlist_onlink_check();
 969
 970                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 971                                 /*
 972                                  * It is used as the current default router,
 973                                  * so we have to move it to the end of the
 974                                  * list and choose a new one.
 975                                  * XXX: it is not very efficient if this is
 976                                  *      the only router.
 977                                  */
 978                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
 979                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
 980
 981                                 defrouter_select();
 982                         }
 983                 }
 984                 crit_exit();
 985         }
 986
 987         /*
 988          * Before deleting the entry, remember the next entry as the
 989          * return value.  We need this because pfxlist_onlink_check() above
 990          * might have freed other entries (particularly the old next entry) as
 991          * a side effect (XXX).
 992          */
 993         next = ln->ln_next;
 994
 995         /*
 996          * Detach the route from the routing tree and the list of neighbor
 997          * caches, and disable the route entry not to be used in already
 998          * cached routes.
 999          */
1000         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1001                   rt_mask(rt), 0, (struct rtentry **)0);
1002
1003         return (next);
1004 }
1005
1006 /*
1007  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1008  *
1009  * XXX cost-effective metods?
1010  */
1011 void
1012 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1013 {
1014         struct llinfo_nd6 *ln;
1015
1016         /*
1017          * If the caller specified "rt", use that.  Otherwise, resolve the
1018          * routing table by supplied "dst6".
1019          */
1020         if (!rt) {
1021                 if (!dst6)
1022                         return;
1023                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1024                         return;
1025         }
1026
1027         if ((rt->rt_flags & RTF_GATEWAY) ||
1028             !(rt->rt_flags & RTF_LLINFO) ||
1029             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1030             rt->rt_gateway->sa_family != AF_LINK) {
1031                 /* This is not a host route. */
1032                 return;
1033         }
1034
1035         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1036         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1037                 return;
1038
1039         /*
1040          * if we get upper-layer reachability confirmation many times,
1041          * it is possible we have false information.
1042          */
1043         if (!force) {
1044                 ln->ln_byhint++;
1045                 if (ln->ln_byhint > nd6_maxnudhint)
1046                         return;
1047         }
1048
1049         ln->ln_state = ND6_LLINFO_REACHABLE;
1050         if (ln->ln_expire)
1051                 ln->ln_expire = time_second +
1052                         ND_IFINFO(rt->rt_ifp)->reachable;
1053 }
1054
1055 void
1056 nd6_rtrequest(int req, struct rtentry *rt,
1057               struct rt_addrinfo *info) /* xxx unused */
1058 {
1059         struct sockaddr *gate = rt->rt_gateway;
1060         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1061         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1062         struct ifnet *ifp = rt->rt_ifp;
1063         struct ifaddr *ifa;
1064
1065         if ((rt->rt_flags & RTF_GATEWAY))
1066                 return;
1067
1068         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1069                 /*
1070                  * This is probably an interface direct route for a link
1071                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1072                  * We do not need special treatment below for such a route.
1073                  * Moreover, the RTF_LLINFO flag which would be set below
1074                  * would annoy the ndp(8) command.
1075                  */
1076                 return;
1077         }
1078
1079         if (req == RTM_RESOLVE &&
1080             (nd6_need_cache(ifp) == 0 || /* stf case */
1081              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1082                 /*
1083                  * FreeBSD and BSD/OS often make a cloned host route based
1084                  * on a less-specific route (e.g. the default route).
1085                  * If the less specific route does not have a "gateway"
1086                  * (this is the case when the route just goes to a p2p or an
1087                  * stf interface), we'll mistakenly make a neighbor cache for
1088                  * the host route, and will see strange neighbor solicitation
1089                  * for the corresponding destination.  In order to avoid the
1090                  * confusion, we check if the destination of the route is
1091                  * a neighbor in terms of neighbor discovery, and stop the
1092                  * process if not.  Additionally, we remove the LLINFO flag
1093                  * so that ndp(8) will not try to get the neighbor information
1094                  * of the destination.
1095                  */
1096                 rt->rt_flags &= ~RTF_LLINFO;
1097                 return;
1098         }
1099
1100         switch (req) {
1101         case RTM_ADD:
1102                 /*
1103                  * There is no backward compatibility :)
1104                  *
1105                  * if (!(rt->rt_flags & RTF_HOST) &&
1106                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1107                  *         rt->rt_flags |= RTF_CLONING;
1108                  */
1109                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1110                         /*
1111                          * Case 1: This route should come from
1112                          * a route to interface.  RTF_LLINFO flag is set
1113                          * for a host route whose destination should be
1114                          * treated as on-link.
1115                          */
1116                         rt_setgate(rt, rt_key(rt),
1117                                    (struct sockaddr *)&null_sdl);
1118                         gate = rt->rt_gateway;
1119                         SDL(gate)->sdl_type = ifp->if_type;
1120                         SDL(gate)->sdl_index = ifp->if_index;
1121                         if (ln)
1122                                 ln->ln_expire = time_second;
1123 #if 1
1124                         if (ln && ln->ln_expire == 0) {
1125                                 /* kludge for desktops */
1126 #if 0
1127                                 kprintf("nd6_rtequest: time.tv_sec is zero; "
1128                                        "treat it as 1\n");
1129 #endif
1130                                 ln->ln_expire = 1;
1131                         }
1132 #endif
1133                         if ((rt->rt_flags & RTF_CLONING))
1134                                 break;
1135                 }
1136                 /*
1137                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1138                  * We don't do that here since llinfo is not ready yet.
1139                  *
1140                  * There are also couple of other things to be discussed:
1141                  * - unsolicited NA code needs improvement beforehand
1142                  * - RFC2461 says we MAY send multicast unsolicited NA
1143                  *   (7.2.6 paragraph 4), however, it also says that we
1144                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1145                  *   we don't have anything like it right now.
1146                  *   note that the mechanism needs a mutual agreement
1147                  *   between proxies, which means that we need to implement
1148                  *   a new protocol, or a new kludge.
1149                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1150                  *   we need to check ip6forwarding before sending it.
1151                  *   (or should we allow proxy ND configuration only for
1152                  *   routers?  there's no mention about proxy ND from hosts)
1153                  */
1154 #if 0
1155                 /* XXX it does not work */
1156                 if (rt->rt_flags & RTF_ANNOUNCE)
1157                         nd6_na_output(ifp,
1158                               &SIN6(rt_key(rt))->sin6_addr,
1159                               &SIN6(rt_key(rt))->sin6_addr,
1160                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1161                               1, NULL);
1162 #endif
1163                 /* FALLTHROUGH */
1164         case RTM_RESOLVE:
1165                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1166                         /*
1167                          * Address resolution isn't necessary for a point to
1168                          * point link, so we can skip this test for a p2p link.
1169                          */
1170                         if (gate->sa_family != AF_LINK ||
1171                             gate->sa_len < sizeof(null_sdl)) {
1172                                 log(LOG_DEBUG,
1173                                     "nd6_rtrequest: bad gateway value: %s\n",
1174                                     if_name(ifp));
1175                                 break;
1176                         }
1177                         SDL(gate)->sdl_type = ifp->if_type;
1178                         SDL(gate)->sdl_index = ifp->if_index;
1179                 }
1180                 if (ln != NULL)
1181                         break;  /* This happens on a route change */
1182                 /*
1183                  * Case 2: This route may come from cloning, or a manual route
1184                  * add with a LL address.
1185                  */
1186                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1187                 rt->rt_llinfo = (caddr_t)ln;
1188                 if (!ln) {
1189                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1190                         break;
1191                 }
1192                 nd6_inuse++;
1193                 nd6_allocated++;
1194                 bzero(ln, sizeof(*ln));
1195                 ln->ln_rt = rt;
1196                 /* this is required for "ndp" command. - shin */
1197                 if (req == RTM_ADD) {
1198                         /*
1199                          * gate should have some valid AF_LINK entry,
1200                          * and ln->ln_expire should have some lifetime
1201                          * which is specified by ndp command.
1202                          */
1203                         ln->ln_state = ND6_LLINFO_REACHABLE;
1204                         ln->ln_byhint = 0;
1205                 } else {
1206                         /*
1207                          * When req == RTM_RESOLVE, rt is created and
1208                          * initialized in rtrequest(), so rt_expire is 0.
1209                          */
1210                         ln->ln_state = ND6_LLINFO_NOSTATE;
1211                         ln->ln_expire = time_second;
1212                 }
1213                 rt->rt_flags |= RTF_LLINFO;
1214                 ln->ln_next = llinfo_nd6.ln_next;
1215                 llinfo_nd6.ln_next = ln;
1216                 ln->ln_prev = &llinfo_nd6;
1217                 ln->ln_next->ln_prev = ln;
1218
1219                 /*
1220                  * check if rt_key(rt) is one of my address assigned
1221                  * to the interface.
1222                  */
1223                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1224                                           &SIN6(rt_key(rt))->sin6_addr);
1225                 if (ifa) {
1226                         caddr_t macp = nd6_ifptomac(ifp);
1227                         ln->ln_expire = 0;
1228                         ln->ln_state = ND6_LLINFO_REACHABLE;
1229                         ln->ln_byhint = 0;
1230                         if (macp) {
1231                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1232                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1233                         }
1234                         if (nd6_useloopback) {
1235                                 rt->rt_ifp = &loif[0];  /* XXX */
1236                                 /*
1237                                  * Make sure rt_ifa be equal to the ifaddr
1238                                  * corresponding to the address.
1239                                  * We need this because when we refer
1240                                  * rt_ifa->ia6_flags in ip6_input, we assume
1241                                  * that the rt_ifa points to the address instead
1242                                  * of the loopback address.
1243                                  */
1244                                 if (ifa != rt->rt_ifa) {
1245                                         IFAFREE(rt->rt_ifa);
1246                                         IFAREF(ifa);
1247                                         rt->rt_ifa = ifa;
1248                                 }
1249                         }
1250                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1251                         ln->ln_expire = 0;
1252                         ln->ln_state = ND6_LLINFO_REACHABLE;
1253                         ln->ln_byhint = 0;
1254
1255                         /* join solicited node multicast for proxy ND */
1256                         if (ifp->if_flags & IFF_MULTICAST) {
1257                                 struct in6_addr llsol;
1258                                 int error;
1259
1260                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1261                                 llsol.s6_addr16[0] = htons(0xff02);
1262                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1263                                 llsol.s6_addr32[1] = 0;
1264                                 llsol.s6_addr32[2] = htonl(1);
1265                                 llsol.s6_addr8[12] = 0xff;
1266
1267                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1268                                         nd6log((LOG_ERR, "%s: failed to join "
1269                                             "%s (errno=%d)\n", if_name(ifp),
1270                                             ip6_sprintf(&llsol), error));
1271                                 }
1272                         }
1273                 }
1274                 break;
1275
1276         case RTM_DELETE:
1277                 if (!ln)
1278                         break;
1279                 /* leave from solicited node multicast for proxy ND */
1280                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1281                     (ifp->if_flags & IFF_MULTICAST)) {
1282                         struct in6_addr llsol;
1283                         struct in6_multi *in6m;
1284
1285                         llsol = SIN6(rt_key(rt))->sin6_addr;
1286                         llsol.s6_addr16[0] = htons(0xff02);
1287                         llsol.s6_addr16[1] = htons(ifp->if_index);
1288                         llsol.s6_addr32[1] = 0;
1289                         llsol.s6_addr32[2] = htonl(1);
1290                         llsol.s6_addr8[12] = 0xff;
1291
1292                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1293                         if (in6m)
1294                                 in6_delmulti(in6m);
1295                 }
1296                 nd6_inuse--;
1297                 ln->ln_next->ln_prev = ln->ln_prev;
1298                 ln->ln_prev->ln_next = ln->ln_next;
1299                 ln->ln_prev = NULL;
1300                 rt->rt_llinfo = 0;
1301                 rt->rt_flags &= ~RTF_LLINFO;
1302                 if (ln->ln_hold)
1303                         m_freem(ln->ln_hold);
1304                 Free((caddr_t)ln);
1305         }
1306 }
1307
1308 int
1309 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1310 {
1311         struct in6_drlist *drl = (struct in6_drlist *)data;
1312         struct in6_prlist *prl = (struct in6_prlist *)data;
1313         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1314         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1315         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1316         struct nd_defrouter *dr, any;
1317         struct nd_prefix *pr;
1318         struct rtentry *rt;
1319         int i = 0, error = 0;
1320
1321         switch (cmd) {
1322         case SIOCGDRLST_IN6:
1323                 /*
1324                  * obsolete API, use sysctl under net.inet6.icmp6
1325                  */
1326                 bzero(drl, sizeof(*drl));
1327                 crit_enter();
1328                 dr = TAILQ_FIRST(&nd_defrouter);
1329                 while (dr && i < DRLSTSIZ) {
1330                         drl->defrouter[i].rtaddr = dr->rtaddr;
1331                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1332                                 /* XXX: need to this hack for KAME stack */
1333                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1334                         } else
1335                                 log(LOG_ERR,
1336                                     "default router list contains a "
1337                                     "non-linklocal address(%s)\n",
1338                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1339
1340                         drl->defrouter[i].flags = dr->flags;
1341                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1342                         drl->defrouter[i].expire = dr->expire;
1343                         drl->defrouter[i].if_index = dr->ifp->if_index;
1344                         i++;
1345                         dr = TAILQ_NEXT(dr, dr_entry);
1346                 }
1347                 crit_exit();
1348                 break;
1349         case SIOCGPRLST_IN6:
1350                 /*
1351                  * obsolete API, use sysctl under net.inet6.icmp6
1352                  */
1353                 /*
1354                  * XXX meaning of fields, especialy "raflags", is very
1355                  * differnet between RA prefix list and RR/static prefix list.
1356                  * how about separating ioctls into two?
1357                  */
1358                 bzero(prl, sizeof(*prl));
1359                 crit_enter();
1360                 pr = nd_prefix.lh_first;
1361                 while (pr && i < PRLSTSIZ) {
1362                         struct nd_pfxrouter *pfr;
1363                         int j;
1364
1365                         in6_embedscope(&prl->prefix[i].prefix,
1366                             &pr->ndpr_prefix, NULL, NULL);
1367                         prl->prefix[i].raflags = pr->ndpr_raf;
1368                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1369                         prl->prefix[i].vltime = pr->ndpr_vltime;
1370                         prl->prefix[i].pltime = pr->ndpr_pltime;
1371                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1372                         prl->prefix[i].expire = pr->ndpr_expire;
1373
1374                         pfr = pr->ndpr_advrtrs.lh_first;
1375                         j = 0;
1376                         while (pfr) {
1377                                 if (j < DRLSTSIZ) {
1378 #define RTRADDR prl->prefix[i].advrtr[j]
1379                                         RTRADDR = pfr->router->rtaddr;
1380                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1381                                                 /* XXX: hack for KAME */
1382                                                 RTRADDR.s6_addr16[1] = 0;
1383                                         } else
1384                                                 log(LOG_ERR,
1385                                                     "a router(%s) advertises "
1386                                                     "a prefix with "
1387                                                     "non-link local address\n",
1388                                                     ip6_sprintf(&RTRADDR));
1389 #undef RTRADDR
1390                                 }
1391                                 j++;
1392                                 pfr = pfr->pfr_next;
1393                         }
1394                         prl->prefix[i].advrtrs = j;
1395                         prl->prefix[i].origin = PR_ORIG_RA;
1396
1397                         i++;
1398                         pr = pr->ndpr_next;
1399                 }
1400               {
1401                 struct rr_prefix *rpp;
1402
1403                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1404                      rpp = LIST_NEXT(rpp, rp_entry)) {
1405                         if (i >= PRLSTSIZ)
1406                                 break;
1407                         in6_embedscope(&prl->prefix[i].prefix,
1408                             &pr->ndpr_prefix, NULL, NULL);
1409                         prl->prefix[i].raflags = rpp->rp_raf;
1410                         prl->prefix[i].prefixlen = rpp->rp_plen;
1411                         prl->prefix[i].vltime = rpp->rp_vltime;
1412                         prl->prefix[i].pltime = rpp->rp_pltime;
1413                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1414                         prl->prefix[i].expire = rpp->rp_expire;
1415                         prl->prefix[i].advrtrs = 0;
1416                         prl->prefix[i].origin = rpp->rp_origin;
1417                         i++;
1418                 }
1419               }
1420                 crit_exit();
1421
1422                 break;
1423         case OSIOCGIFINFO_IN6:
1424                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1425                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1426                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1427                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1428                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1429                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1430                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1431                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1432                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1433                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1434                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1435                 break;
1436         case SIOCGIFINFO_IN6:
1437                 ndi->ndi = *ND_IFINFO(ifp);
1438                 break;
1439         case SIOCSIFINFO_FLAGS:
1440                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1441                 break;
1442         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1443                 /* flush default router list */
1444                 /*
1445                  * xxx sumikawa: should not delete route if default
1446                  * route equals to the top of default router list
1447                  */
1448                 bzero(&any, sizeof(any));
1449                 defrouter_delreq(&any, 0);
1450                 defrouter_select();
1451                 /* xxx sumikawa: flush prefix list */
1452                 break;
1453         case SIOCSPFXFLUSH_IN6:
1454             {
1455                 /* flush all the prefix advertised by routers */
1456                 struct nd_prefix *pr, *next;
1457
1458                 crit_enter();
1459                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1460                         struct in6_ifaddr *ia, *ia_next;
1461
1462                         next = pr->ndpr_next;
1463
1464                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1465                                 continue; /* XXX */
1466
1467                         /* do we really have to remove addresses as well? */
1468                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1469                                 /* ia might be removed.  keep the next ptr. */
1470                                 ia_next = ia->ia_next;
1471
1472                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1473                                         continue;
1474
1475                                 if (ia->ia6_ndpr == pr)
1476                                         in6_purgeaddr(&ia->ia_ifa);
1477                         }
1478                         prelist_remove(pr);
1479                 }
1480                 crit_exit();
1481                 break;
1482             }
1483         case SIOCSRTRFLUSH_IN6:
1484             {
1485                 /* flush all the default routers */
1486                 struct nd_defrouter *dr, *next;
1487
1488                 crit_enter();
1489                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1490                         /*
1491                          * The first entry of the list may be stored in
1492                          * the routing table, so we'll delete it later.
1493                          */
1494                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1495                                 next = TAILQ_NEXT(dr, dr_entry);
1496                                 defrtrlist_del(dr);
1497                         }
1498                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1499                 }
1500                 crit_exit();
1501                 break;
1502             }
1503         case SIOCGNBRINFO_IN6:
1504             {
1505                 struct llinfo_nd6 *ln;
1506                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1507
1508                 /*
1509                  * XXX: KAME specific hack for scoped addresses
1510                  *      XXXX: for other scopes than link-local?
1511                  */
1512                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1513                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1514                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1515
1516                         if (*idp == 0)
1517                                 *idp = htons(ifp->if_index);
1518                 }
1519
1520                 crit_enter();
1521                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1522                         error = EINVAL;
1523                         crit_exit();
1524                         break;
1525                 }
1526                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1527                 nbi->state = ln->ln_state;
1528                 nbi->asked = ln->ln_asked;
1529                 nbi->isrouter = ln->ln_router;
1530                 nbi->expire = ln->ln_expire;
1531                 crit_exit();
1532
1533                 break;
1534             }
1535         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1536                 ndif->ifindex = nd6_defifindex;
1537                 break;
1538         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1539                 return (nd6_setdefaultiface(ndif->ifindex));
1540                 break;
1541         }
1542         return (error);
1543 }
1544
1545 /*
1546  * Create neighbor cache entry and cache link-layer address,
1547  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1548  */
1549 struct rtentry *
1550 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1551                  int lladdrlen,
1552                  int type,      /* ICMP6 type */
1553                  int code       /* type dependent information */)
1554 {
1555         struct rtentry *rt = NULL;
1556         struct llinfo_nd6 *ln = NULL;
1557         int is_newentry;
1558         struct sockaddr_dl *sdl = NULL;
1559         int do_update;
1560         int olladdr;
1561         int llchange;
1562         int newstate = 0;
1563
1564         if (!ifp)
1565                 panic("ifp == NULL in nd6_cache_lladdr");
1566         if (!from)
1567                 panic("from == NULL in nd6_cache_lladdr");
1568
1569         /* nothing must be updated for unspecified address */
1570         if (IN6_IS_ADDR_UNSPECIFIED(from))
1571                 return NULL;
1572
1573         /*
1574          * Validation about ifp->if_addrlen and lladdrlen must be done in
1575          * the caller.
1576          *
1577          * XXX If the link does not have link-layer adderss, what should
1578          * we do? (ifp->if_addrlen == 0)
1579          * Spec says nothing in sections for RA, RS and NA.  There's small
1580          * description on it in NS section (RFC 2461 7.2.3).
1581          */
1582
1583         rt = nd6_lookup(from, 0, ifp);
1584         if (!rt) {
1585 #if 0
1586                 /* nothing must be done if there's no lladdr */
1587                 if (!lladdr || !lladdrlen)
1588                         return NULL;
1589 #endif
1590
1591                 rt = nd6_lookup(from, 1, ifp);
1592                 is_newentry = 1;
1593         } else {
1594                 /* do nothing if static ndp is set */
1595                 if (rt->rt_flags & RTF_STATIC)
1596                         return NULL;
1597                 is_newentry = 0;
1598         }
1599
1600         if (!rt)
1601                 return NULL;
1602         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1603 fail:
1604                 nd6_free(rt);
1605                 return NULL;
1606         }
1607         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1608         if (!ln)
1609                 goto fail;
1610         if (!rt->rt_gateway)
1611                 goto fail;
1612         if (rt->rt_gateway->sa_family != AF_LINK)
1613                 goto fail;
1614         sdl = SDL(rt->rt_gateway);
1615
1616         olladdr = (sdl->sdl_alen) ? 1 : 0;
1617         if (olladdr && lladdr) {
1618                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1619                         llchange = 1;
1620                 else
1621                         llchange = 0;
1622         } else
1623                 llchange = 0;
1624
1625         /*
1626          * newentry olladdr  lladdr  llchange   (*=record)
1627          *      0       n       n       --      (1)
1628          *      0       y       n       --      (2)
1629          *      0       n       y       --      (3) * STALE
1630          *      0       y       y       n       (4) *
1631          *      0       y       y       y       (5) * STALE
1632          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1633          *      1       --      y       --      (7) * STALE
1634          */
1635
1636         if (lladdr) {           /* (3-5) and (7) */
1637                 /*
1638                  * Record source link-layer address
1639                  * XXX is it dependent to ifp->if_type?
1640                  */
1641                 sdl->sdl_alen = ifp->if_addrlen;
1642                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1643         }
1644
1645         if (!is_newentry) {
1646                 if ((!olladdr && lladdr)                /* (3) */
1647                  || (olladdr && lladdr && llchange)) {  /* (5) */
1648                         do_update = 1;
1649                         newstate = ND6_LLINFO_STALE;
1650                 } else                                  /* (1-2,4) */
1651                         do_update = 0;
1652         } else {
1653                 do_update = 1;
1654                 if (!lladdr)                            /* (6) */
1655                         newstate = ND6_LLINFO_NOSTATE;
1656                 else                                    /* (7) */
1657                         newstate = ND6_LLINFO_STALE;
1658         }
1659
1660         if (do_update) {
1661                 /*
1662                  * Update the state of the neighbor cache.
1663                  */
1664                 ln->ln_state = newstate;
1665
1666                 if (ln->ln_state == ND6_LLINFO_STALE) {
1667                         /*
1668                          * XXX: since nd6_output() below will cause
1669                          * state tansition to DELAY and reset the timer,
1670                          * we must set the timer now, although it is actually
1671                          * meaningless.
1672                          */
1673                         ln->ln_expire = time_second + nd6_gctimer;
1674
1675                         if (ln->ln_hold) {
1676                                 /*
1677                                  * we assume ifp is not a p2p here, so just
1678                                  * set the 2nd argument as the 1st one.
1679                                  */
1680                                 nd6_output(ifp, ifp, ln->ln_hold,
1681                                            (struct sockaddr_in6 *)rt_key(rt),
1682                                            rt);
1683                                 ln->ln_hold = NULL;
1684                         }
1685                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1686                         /* probe right away */
1687                         ln->ln_expire = time_second;
1688                 }
1689         }
1690
1691         /*
1692          * ICMP6 type dependent behavior.
1693          *
1694          * NS: clear IsRouter if new entry
1695          * RS: clear IsRouter
1696          * RA: set IsRouter if there's lladdr
1697          * redir: clear IsRouter if new entry
1698          *
1699          * RA case, (1):
1700          * The spec says that we must set IsRouter in the following cases:
1701          * - If lladdr exist, set IsRouter.  This means (1-5).
1702          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1703          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1704          * A quetion arises for (1) case.  (1) case has no lladdr in the
1705          * neighbor cache, this is similar to (6).
1706          * This case is rare but we figured that we MUST NOT set IsRouter.
1707          *
1708          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1709          *                                                      D R
1710          *      0       n       n       --      (1)     c   ?     s
1711          *      0       y       n       --      (2)     c   s     s
1712          *      0       n       y       --      (3)     c   s     s
1713          *      0       y       y       n       (4)     c   s     s
1714          *      0       y       y       y       (5)     c   s     s
1715          *      1       --      n       --      (6) c   c       c s
1716          *      1       --      y       --      (7) c   c   s   c s
1717          *
1718          *                                      (c=clear s=set)
1719          */
1720         switch (type & 0xff) {
1721         case ND_NEIGHBOR_SOLICIT:
1722                 /*
1723                  * New entry must have is_router flag cleared.
1724                  */
1725                 if (is_newentry)        /* (6-7) */
1726                         ln->ln_router = 0;
1727                 break;
1728         case ND_REDIRECT:
1729                 /*
1730                  * If the icmp is a redirect to a better router, always set the
1731                  * is_router flag. Otherwise, if the entry is newly created,
1732                  * clear the flag. [RFC 2461, sec 8.3]
1733                  */
1734                 if (code == ND_REDIRECT_ROUTER)
1735                         ln->ln_router = 1;
1736                 else if (is_newentry) /* (6-7) */
1737                         ln->ln_router = 0;
1738                 break;
1739         case ND_ROUTER_SOLICIT:
1740                 /*
1741                  * is_router flag must always be cleared.
1742                  */
1743                 ln->ln_router = 0;
1744                 break;
1745         case ND_ROUTER_ADVERT:
1746                 /*
1747                  * Mark an entry with lladdr as a router.
1748                  */
1749                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1750                  || (is_newentry && lladdr)) {                  /* (7) */
1751                         ln->ln_router = 1;
1752                 }
1753                 break;
1754         }
1755
1756         /*
1757          * When the link-layer address of a router changes, select the
1758          * best router again.  In particular, when the neighbor entry is newly
1759          * created, it might affect the selection policy.
1760          * Question: can we restrict the first condition to the "is_newentry"
1761          * case?
1762          * XXX: when we hear an RA from a new router with the link-layer
1763          * address option, defrouter_select() is called twice, since
1764          * defrtrlist_update called the function as well.  However, I believe
1765          * we can compromise the overhead, since it only happens the first
1766          * time.
1767          * XXX: although defrouter_select() should not have a bad effect
1768          * for those are not autoconfigured hosts, we explicitly avoid such
1769          * cases for safety.
1770          */
1771         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1772                 defrouter_select();
1773
1774         return rt;
1775 }
1776
1777 static void
1778 nd6_slowtimo(void *ignored_arg)
1779 {
1780         struct nd_ifinfo *nd6if;
1781         struct ifnet *ifp;
1782
1783         crit_enter();
1784         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1785             nd6_slowtimo, NULL);
1786         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1787                 nd6if = ND_IFINFO(ifp);
1788                 if (nd6if->basereachable && /* already initialized */
1789                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1790                         /*
1791                          * Since reachable time rarely changes by router
1792                          * advertisements, we SHOULD insure that a new random
1793                          * value gets recomputed at least once every few hours.
1794                          * (RFC 2461, 6.3.4)
1795                          */
1796                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1797                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1798                 }
1799         }
1800         crit_exit();
1801 }
1802
1803 #define gotoerr(e) { error = (e); goto bad;}
1804
1805 int
1806 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1807            struct sockaddr_in6 *dst, struct rtentry *rt)
1808 {
1809         struct llinfo_nd6 *ln = NULL;
1810         int error = 0;
1811
1812         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1813                 goto sendpkt;
1814
1815         if (nd6_need_cache(ifp) == 0)
1816                 goto sendpkt;
1817
1818         /*
1819          * next hop determination.  This routine is derived from ether_outpout.
1820          */
1821         if (rt != NULL) {
1822                 if (!(rt->rt_flags & RTF_UP)) {
1823                         rt = rtlookup((struct sockaddr *)dst);
1824                         if (rt == NULL)
1825                                 gotoerr(EHOSTUNREACH);
1826                         rt->rt_refcnt--;
1827                         if (rt->rt_ifp != ifp) {
1828                                 /* XXX: loop care? */
1829                                 return nd6_output(ifp, origifp, m, dst, rt);
1830                         }
1831                 }
1832                 if (rt->rt_flags & RTF_GATEWAY) {
1833                         struct sockaddr_in6 *gw6;
1834
1835                         /*
1836                          * We skip link-layer address resolution and NUD
1837                          * if the gateway is not a neighbor from ND point
1838                          * of view, regardless of the value of nd_ifinfo.flags.
1839                          * The second condition is a bit tricky; we skip
1840                          * if the gateway is our own address, which is
1841                          * sometimes used to install a route to a p2p link.
1842                          */
1843                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1844                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1845                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1846                                 /*
1847                                  * We allow this kind of tricky route only
1848                                  * when the outgoing interface is p2p.
1849                                  * XXX: we may need a more generic rule here.
1850                                  */
1851                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1852                                         gotoerr(EHOSTUNREACH);
1853
1854                                 goto sendpkt;
1855                         }
1856
1857                         if (rt->rt_gwroute == NULL) {
1858                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1859                                 if (rt->rt_gwroute == NULL)
1860                                         gotoerr(EHOSTUNREACH);
1861                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1862                                 rtfree(rt->rt_gwroute);
1863                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1864                                 if (rt->rt_gwroute == NULL)
1865                                         gotoerr(EHOSTUNREACH);
1866                         }
1867                 }
1868         }
1869
1870         /*
1871          * Address resolution or Neighbor Unreachability Detection
1872          * for the next hop.
1873          * At this point, the destination of the packet must be a unicast
1874          * or an anycast address(i.e. not a multicast).
1875          */
1876
1877         /* Look up the neighbor cache for the nexthop */
1878         if (rt && (rt->rt_flags & RTF_LLINFO))
1879                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1880         else {
1881                 /*
1882                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1883                  * the condition below is not very efficient.  But we believe
1884                  * it is tolerable, because this should be a rare case.
1885                  */
1886                 if (nd6_is_addr_neighbor(dst, ifp) &&
1887                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1888                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1889         }
1890         if (!ln || !rt) {
1891                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1892                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1893                         log(LOG_DEBUG,
1894                             "nd6_output: can't allocate llinfo for %s "
1895                             "(ln=%p, rt=%p)\n",
1896                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1897                         gotoerr(EIO);   /* XXX: good error? */
1898                 }
1899
1900                 goto sendpkt;   /* send anyway */
1901         }
1902
1903         /* We don't have to do link-layer address resolution on a p2p link. */
1904         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1905             ln->ln_state < ND6_LLINFO_REACHABLE) {
1906                 ln->ln_state = ND6_LLINFO_STALE;
1907                 ln->ln_expire = time_second + nd6_gctimer;
1908         }
1909
1910         /*
1911          * The first time we send a packet to a neighbor whose entry is
1912          * STALE, we have to change the state to DELAY and a sets a timer to
1913          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1914          * neighbor unreachability detection on expiration.
1915          * (RFC 2461 7.3.3)
1916          */
1917         if (ln->ln_state == ND6_LLINFO_STALE) {
1918                 ln->ln_asked = 0;
1919                 ln->ln_state = ND6_LLINFO_DELAY;
1920                 ln->ln_expire = time_second + nd6_delay;
1921         }
1922
1923         /*
1924          * If the neighbor cache entry has a state other than INCOMPLETE
1925          * (i.e. its link-layer address is already resolved), just
1926          * send the packet.
1927          */
1928         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1929                 goto sendpkt;
1930
1931         /*
1932          * There is a neighbor cache entry, but no ethernet address
1933          * response yet.  Replace the held mbuf (if any) with this
1934          * latest one.
1935          *
1936          * This code conforms to the rate-limiting rule described in Section
1937          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1938          * an NS below.
1939          */
1940         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1941                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1942         if (ln->ln_hold)
1943                 m_freem(ln->ln_hold);
1944         ln->ln_hold = m;
1945         if (ln->ln_expire) {
1946                 if (ln->ln_asked < nd6_mmaxtries &&
1947                     ln->ln_expire < time_second) {
1948                         ln->ln_asked++;
1949                         ln->ln_expire = time_second +
1950                                 ND_IFINFO(ifp)->retrans / 1000;
1951                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1952                 }
1953         }
1954         return (0);
1955
1956 sendpkt:
1957         lwkt_serialize_enter(ifp->if_serializer);
1958         if (ifp->if_flags & IFF_LOOPBACK) {
1959                 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1960                                           rt);
1961         } else {
1962                 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt);
1963         }
1964         lwkt_serialize_exit(ifp->if_serializer);
1965         return (error);
1966
1967 bad:
1968         m_freem(m);
1969         return (error);
1970 }
1971 #undef gotoerr
1972
1973 int
1974 nd6_need_cache(struct ifnet *ifp)
1975 {
1976         /*
1977          * XXX: we currently do not make neighbor cache on any interface
1978          * other than Ethernet, FDDI and GIF.
1979          *
1980          * RFC2893 says:
1981          * - unidirectional tunnels needs no ND
1982          */
1983         switch (ifp->if_type) {
1984         case IFT_ETHER:
1985         case IFT_FDDI:
1986         case IFT_IEEE1394:
1987 #ifdef IFT_L2VLAN
1988         case IFT_L2VLAN:
1989 #endif
1990 #ifdef IFT_IEEE80211
1991         case IFT_IEEE80211:
1992 #endif
1993 #ifdef IFT_CARP
1994         case IFT_CARP:
1995 #endif
1996         case IFT_GIF:           /* XXX need more cases? */
1997                 return (1);
1998         default:
1999                 return (0);
2000         }
2001 }
2002
2003 int
2004 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2005                 struct sockaddr *dst, u_char *desten)
2006 {
2007         struct sockaddr_dl *sdl;
2008         struct rtentry *rt;
2009
2010
2011         if (m->m_flags & M_MCAST) {
2012                 switch (ifp->if_type) {
2013                 case IFT_ETHER:
2014                 case IFT_FDDI:
2015 #ifdef IFT_L2VLAN
2016         case IFT_L2VLAN:
2017 #endif
2018 #ifdef IFT_IEEE80211
2019                 case IFT_IEEE80211:
2020 #endif
2021                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2022                                                  desten);
2023                         return (1);
2024                 case IFT_IEEE1394:
2025                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2026                         return (1);
2027                 default:
2028                         m_freem(m);
2029                         return (0);
2030                 }
2031         }
2032         if (rt0 == NULL) {
2033                 /* this could happen, if we could not allocate memory */
2034                 m_freem(m);
2035                 return (0);
2036         }
2037         if (rt_llroute(dst, rt0, &rt) != 0) {
2038                 m_freem(m);
2039                 return (0);
2040         }
2041         if (rt->rt_gateway->sa_family != AF_LINK) {
2042                 kprintf("nd6_storelladdr: something odd happens\n");
2043                 m_freem(m);
2044                 return (0);
2045         }
2046         sdl = SDL(rt->rt_gateway);
2047         if (sdl->sdl_alen == 0) {
2048                 /* this should be impossible, but we bark here for debugging */
2049                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2050                 m_freem(m);
2051                 return (0);
2052         }
2053
2054         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2055         return (1);
2056 }
2057
2058 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2059 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2060 #ifdef SYSCTL_DECL
2061 SYSCTL_DECL(_net_inet6_icmp6);
2062 #endif
2063 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2064         CTLFLAG_RD, nd6_sysctl_drlist, "");
2065 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2066         CTLFLAG_RD, nd6_sysctl_prlist, "");
2067
2068 static int
2069 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2070 {
2071         int error;
2072         char buf[1024];
2073         struct in6_defrouter *d, *de;
2074         struct nd_defrouter *dr;
2075
2076         if (req->newptr)
2077                 return EPERM;
2078         error = 0;
2079
2080         for (dr = TAILQ_FIRST(&nd_defrouter);
2081              dr;
2082              dr = TAILQ_NEXT(dr, dr_entry)) {
2083                 d = (struct in6_defrouter *)buf;
2084                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2085
2086                 if (d + 1 <= de) {
2087                         bzero(d, sizeof(*d));
2088                         d->rtaddr.sin6_family = AF_INET6;
2089                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2090                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2091                             dr->ifp) != 0)
2092                                 log(LOG_ERR,
2093                                     "scope error in "
2094                                     "default router list (%s)\n",
2095                                     ip6_sprintf(&dr->rtaddr));
2096                         d->flags = dr->flags;
2097                         d->rtlifetime = dr->rtlifetime;
2098                         d->expire = dr->expire;
2099                         d->if_index = dr->ifp->if_index;
2100                 } else
2101                         panic("buffer too short");
2102
2103                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2104                 if (error)
2105                         break;
2106         }
2107         return error;
2108 }
2109
2110 static int
2111 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2112 {
2113         int error;
2114         char buf[1024];
2115         struct in6_prefix *p, *pe;
2116         struct nd_prefix *pr;
2117
2118         if (req->newptr)
2119                 return EPERM;
2120         error = 0;
2121
2122         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2123                 u_short advrtrs;
2124                 size_t advance;
2125                 struct sockaddr_in6 *sin6, *s6;
2126                 struct nd_pfxrouter *pfr;
2127
2128                 p = (struct in6_prefix *)buf;
2129                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2130
2131                 if (p + 1 <= pe) {
2132                         bzero(p, sizeof(*p));
2133                         sin6 = (struct sockaddr_in6 *)(p + 1);
2134
2135                         p->prefix = pr->ndpr_prefix;
2136                         if (in6_recoverscope(&p->prefix,
2137                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2138                                 log(LOG_ERR,
2139                                     "scope error in prefix list (%s)\n",
2140                                     ip6_sprintf(&p->prefix.sin6_addr));
2141                         p->raflags = pr->ndpr_raf;
2142                         p->prefixlen = pr->ndpr_plen;
2143                         p->vltime = pr->ndpr_vltime;
2144                         p->pltime = pr->ndpr_pltime;
2145                         p->if_index = pr->ndpr_ifp->if_index;
2146                         p->expire = pr->ndpr_expire;
2147                         p->refcnt = pr->ndpr_refcnt;
2148                         p->flags = pr->ndpr_stateflags;
2149                         p->origin = PR_ORIG_RA;
2150                         advrtrs = 0;
2151                         for (pfr = pr->ndpr_advrtrs.lh_first;
2152                              pfr;
2153                              pfr = pfr->pfr_next) {
2154                                 if ((void *)&sin6[advrtrs + 1] >
2155                                     (void *)pe) {
2156                                         advrtrs++;
2157                                         continue;
2158                                 }
2159                                 s6 = &sin6[advrtrs];
2160                                 bzero(s6, sizeof(*s6));
2161                                 s6->sin6_family = AF_INET6;
2162                                 s6->sin6_len = sizeof(*sin6);
2163                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2164                                                      pfr->router->ifp) != 0)
2165                                         log(LOG_ERR,
2166                                             "scope error in "
2167                                             "prefix list (%s)\n",
2168                                             ip6_sprintf(&pfr->router->rtaddr));
2169                                 advrtrs++;
2170                         }
2171                         p->advrtrs = advrtrs;
2172                 } else
2173                         panic("buffer too short");
2174
2175                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2176                 error = SYSCTL_OUT(req, buf, advance);
2177                 if (error)
2178                         break;
2179         }
2180         return error;
2181 }