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 1.15 2005/02/01 16:09:37 hrs Exp $ */
   3 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   4
   5 /*
   6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. Neither the name of the project nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  */
  33
  34 /*
  35  * XXX
  36  * KAME 970409 note:
  37  * BSD/OS version heavily modifies this code, related to llinfo.
  38  * Since we don't have BSD/OS version of net/route.c in our hand,
  39  * I left the code mostly as it was in 970310.  -- itojun
  40  */
  41
  42 #include "opt_inet.h"
  43 #include "opt_inet6.h"
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/callout.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mbuf.h>
  50 #include <sys/socket.h>
  51 #include <sys/sockio.h>
  52 #include <sys/time.h>
  53 #include <sys/kernel.h>
  54 #include <sys/protosw.h>
  55 #include <sys/errno.h>
  56 #include <sys/syslog.h>
  57 #include <sys/queue.h>
  58 #include <sys/sysctl.h>
  59
  60 #include <net/if.h>
  61 #include <net/if_dl.h>
  62 #include <net/if_types.h>
  63 #include <net/if_atm.h>
  64 #include <net/route.h>
  65
  66 #include <netinet/in.h>
  67 #include <netinet/if_ether.h>
  68 #include <netinet/if_fddi.h>
  69 #include <netinet6/in6_var.h>
  70 #include <netinet/ip6.h>
  71 #include <netinet6/ip6_var.h>
  72 #include <netinet6/nd6.h>
  73 #include <netinet6/in6_prefix.h>
  74 #include <netinet/icmp6.h>
  75
  76 #include "use_loop.h"
  77
  78 #include <net/net_osdep.h>
  79
  80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  82
  83 #define SIN6(s) ((struct sockaddr_in6 *)s)
  84 #define SDL(s) ((struct sockaddr_dl *)s)
  85
  86 /* timer values */
  87 int     nd6_prune       = 1;    /* walk list every 1 seconds */
  88 int     nd6_delay       = 5;    /* delay first probe time 5 second */
  89 int     nd6_umaxtries   = 3;    /* maximum unicast query */
  90 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
  91 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
  92 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
  93
  94 /* preventing too many loops in ND option parsing */
  95 int nd6_maxndopt = 10;  /* max # of ND options allowed */
  96
  97 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
  98
  99 #ifdef ND6_DEBUG
 100 int nd6_debug = 1;
 101 #else
 102 int nd6_debug = 0;
 103 #endif
 104
 105 /* for debugging? */
 106 static int nd6_inuse, nd6_allocated;
 107
 108 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 109 struct nd_drhead nd_defrouter;
 110 struct nd_prhead nd_prefix = { 0 };
 111
 112 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 113 static struct sockaddr_in6 all1_sa;
 114
 115 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
 116 static void nd6_slowtimo (void *);
 117 static int regen_tmpaddr (struct in6_ifaddr *);
 118
 119 struct callout nd6_slowtimo_ch;
 120 struct callout nd6_timer_ch;
 121 extern struct callout in6_tmpaddrtimer_ch;
 122
 123 void
 124 nd6_init(void)
 125 {
 126         static int nd6_init_done = 0;
 127         int i;
 128
 129         if (nd6_init_done) {
 130                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 131                 return;
 132         }
 133
 134         all1_sa.sin6_family = AF_INET6;
 135         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 136         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 137                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 138
 139         /* initialization of the default router list */
 140         TAILQ_INIT(&nd_defrouter);
 141
 142         nd6_init_done = 1;
 143
 144         /* start timer */
 145         callout_init(&nd6_slowtimo_ch);
 146         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 147             nd6_slowtimo, NULL);
 148 }
 149
 150 struct nd_ifinfo *
 151 nd6_ifattach(struct ifnet *ifp)
 152 {
 153         struct nd_ifinfo *nd;
 154
 155         nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
 156         bzero(nd, sizeof(*nd));
 157
 158         nd->initialized = 1;
 159
 160         nd->linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
 161         nd->chlim = IPV6_DEFHLIM;
 162         nd->basereachable = REACHABLE_TIME;
 163         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 164         nd->retrans = RETRANS_TIMER;
 165         nd->receivedra = 0;
 166
 167         /*
 168          * Note that the default value of ip6_accept_rtadv is 0, which means
 169          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 170          * here.
 171          */
 172         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 173
 174         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 175         nd6_setmtu0(ifp, nd);
 176         return nd;
 177 }
 178
 179 void
 180 nd6_ifdetach(struct nd_ifinfo *nd)
 181 {
 182         free(nd, M_IP6NDP);
 183 }
 184
 185 /*
 186  * Reset ND level link MTU. This function is called when the physical MTU
 187  * changes, which means we might have to adjust the ND level MTU.
 188  */
 189 void
 190 nd6_setmtu(struct ifnet *ifp)
 191 {
 192         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 193 }
 194
 195 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 196 void
 197 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 198 {
 199         u_long oldmaxmtu;
 200         u_long oldlinkmtu;
 201
 202         oldmaxmtu = ndi->maxmtu;
 203         oldlinkmtu = ndi->linkmtu;
 204
 205         switch (ifp->if_type) {
 206         case IFT_ARCNET:        /* XXX MTU handling needs more work */
 207                 ndi->maxmtu = MIN(60480, ifp->if_mtu);
 208                 break;
 209         case IFT_ETHER:
 210                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 211                 break;
 212         case IFT_FDDI:
 213                 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
 214                 break;
 215         case IFT_ATM:
 216                 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
 217                 break;
 218         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 219                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 220                 break;
 221 #ifdef IFT_IEEE80211
 222         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 223                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 224                 break;
 225 #endif
 226         default:
 227                 ndi->maxmtu = ifp->if_mtu;
 228                 break;
 229         }
 230
 231         if (oldmaxmtu != ndi->maxmtu) {
 232                 /*
 233                  * If the ND level MTU is not set yet, or if the maxmtu
 234                  * is reset to a smaller value than the ND level MTU,
 235                  * also reset the ND level MTU.
 236                  */
 237                 if (ndi->linkmtu == 0 ||
 238                     ndi->maxmtu < ndi->linkmtu) {
 239                         ndi->linkmtu = ndi->maxmtu;
 240                         /* also adjust in6_maxmtu if necessary. */
 241                         if (oldlinkmtu == 0) {
 242                                 /*
 243                                  * XXX: the case analysis is grotty, but
 244                                  * it is not efficient to call in6_setmaxmtu()
 245                                  * here when we are during the initialization
 246                                  * procedure.
 247                                  */
 248                                 if (in6_maxmtu < ndi->linkmtu)
 249                                         in6_maxmtu = ndi->linkmtu;
 250                         } else
 251                                 in6_setmaxmtu();
 252                 }
 253         }
 254 #undef MIN
 255 }
 256
 257 void
 258 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 259 {
 260         bzero(ndopts, sizeof(*ndopts));
 261         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 262         ndopts->nd_opts_last
 263                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 264
 265         if (icmp6len == 0) {
 266                 ndopts->nd_opts_done = 1;
 267                 ndopts->nd_opts_search = NULL;
 268         }
 269 }
 270
 271 /*
 272  * Take one ND option.
 273  */
 274 struct nd_opt_hdr *
 275 nd6_option(union nd_opts *ndopts)
 276 {
 277         struct nd_opt_hdr *nd_opt;
 278         int olen;
 279
 280         if (!ndopts)
 281                 panic("ndopts == NULL in nd6_option");
 282         if (!ndopts->nd_opts_last)
 283                 panic("uninitialized ndopts in nd6_option");
 284         if (!ndopts->nd_opts_search)
 285                 return NULL;
 286         if (ndopts->nd_opts_done)
 287                 return NULL;
 288
 289         nd_opt = ndopts->nd_opts_search;
 290
 291         /* make sure nd_opt_len is inside the buffer */
 292         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 293                 bzero(ndopts, sizeof(*ndopts));
 294                 return NULL;
 295         }
 296
 297         olen = nd_opt->nd_opt_len << 3;
 298         if (olen == 0) {
 299                 /*
 300                  * Message validation requires that all included
 301                  * options have a length that is greater than zero.
 302                  */
 303                 bzero(ndopts, sizeof(*ndopts));
 304                 return NULL;
 305         }
 306
 307         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 308         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 309                 /* option overruns the end of buffer, invalid */
 310                 bzero(ndopts, sizeof(*ndopts));
 311                 return NULL;
 312         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 313                 /* reached the end of options chain */
 314                 ndopts->nd_opts_done = 1;
 315                 ndopts->nd_opts_search = NULL;
 316         }
 317         return nd_opt;
 318 }
 319
 320 /*
 321  * Parse multiple ND options.
 322  * This function is much easier to use, for ND routines that do not need
 323  * multiple options of the same type.
 324  */
 325 int
 326 nd6_options(union nd_opts *ndopts)
 327 {
 328         struct nd_opt_hdr *nd_opt;
 329         int i = 0;
 330
 331         if (!ndopts)
 332                 panic("ndopts == NULL in nd6_options");
 333         if (!ndopts->nd_opts_last)
 334                 panic("uninitialized ndopts in nd6_options");
 335         if (!ndopts->nd_opts_search)
 336                 return 0;
 337
 338         while (1) {
 339                 nd_opt = nd6_option(ndopts);
 340                 if (!nd_opt && !ndopts->nd_opts_last) {
 341                         /*
 342                          * Message validation requires that all included
 343                          * options have a length that is greater than zero.
 344                          */
 345                         icmp6stat.icp6s_nd_badopt++;
 346                         bzero(ndopts, sizeof(*ndopts));
 347                         return -1;
 348                 }
 349
 350                 if (!nd_opt)
 351                         goto skip1;
 352
 353                 switch (nd_opt->nd_opt_type) {
 354                 case ND_OPT_SOURCE_LINKADDR:
 355                 case ND_OPT_TARGET_LINKADDR:
 356                 case ND_OPT_MTU:
 357                 case ND_OPT_REDIRECTED_HEADER:
 358                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 359                                 nd6log((LOG_INFO,
 360                                     "duplicated ND6 option found (type=%d)\n",
 361                                     nd_opt->nd_opt_type));
 362                                 /* XXX bark? */
 363                         } else {
 364                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 365                                         = nd_opt;
 366                         }
 367                         break;
 368                 case ND_OPT_PREFIX_INFORMATION:
 369                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 370                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 371                                         = nd_opt;
 372                         }
 373                         ndopts->nd_opts_pi_end =
 374                                 (struct nd_opt_prefix_info *)nd_opt;
 375                         break;
 376                 default:
 377                         /*
 378                          * Unknown options must be silently ignored,
 379                          * to accomodate future extension to the protocol.
 380                          */
 381                         nd6log((LOG_DEBUG,
 382                             "nd6_options: unsupported option %d - "
 383                             "option ignored\n", nd_opt->nd_opt_type));
 384                 }
 385
 386 skip1:
 387                 i++;
 388                 if (i > nd6_maxndopt) {
 389                         icmp6stat.icp6s_nd_toomanyopt++;
 390                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 391                         break;
 392                 }
 393
 394                 if (ndopts->nd_opts_done)
 395                         break;
 396         }
 397
 398         return 0;
 399 }
 400
 401 /*
 402  * ND6 timer routine to expire default route list and prefix list
 403  */
 404 void
 405 nd6_timer(void *ignored_arg)
 406 {
 407         int s;
 408         struct llinfo_nd6 *ln;
 409         struct nd_defrouter *dr;
 410         struct nd_prefix *pr;
 411         struct ifnet *ifp;
 412         struct in6_ifaddr *ia6, *nia6;
 413         struct in6_addrlifetime *lt6;
 414
 415         s = splnet();
 416         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 417                       nd6_timer, NULL);
 418
 419         ln = llinfo_nd6.ln_next;
 420         while (ln && ln != &llinfo_nd6) {
 421                 struct rtentry *rt;
 422                 struct sockaddr_in6 *dst;
 423                 struct llinfo_nd6 *next = ln->ln_next;
 424                 /* XXX: used for the DELAY case only: */
 425                 struct nd_ifinfo *ndi = NULL;
 426
 427                 if ((rt = ln->ln_rt) == NULL) {
 428                         ln = next;
 429                         continue;
 430                 }
 431                 if ((ifp = rt->rt_ifp) == NULL) {
 432                         ln = next;
 433                         continue;
 434                 }
 435                 ndi = ND_IFINFO(ifp);
 436                 dst = (struct sockaddr_in6 *)rt_key(rt);
 437
 438                 if (ln->ln_expire > time_second) {
 439                         ln = next;
 440                         continue;
 441                 }
 442
 443                 /* sanity check */
 444                 if (!rt)
 445                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 446                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 447                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 448                               rt->rt_llinfo, ln);
 449                 if (!dst)
 450                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 451
 452                 switch (ln->ln_state) {
 453                 case ND6_LLINFO_INCOMPLETE:
 454                         if (ln->ln_asked < nd6_mmaxtries) {
 455                                 ln->ln_asked++;
 456                                 ln->ln_expire = time_second +
 457                                         ND_IFINFO(ifp)->retrans / 1000;
 458                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 459                                         ln, 0);
 460                         } else {
 461                                 struct mbuf *m = ln->ln_hold;
 462                                 if (m) {
 463                                         if (rt->rt_ifp) {
 464                                                 /*
 465                                                  * Fake rcvif to make ICMP error
 466                                                  * more helpful in diagnosing
 467                                                  * for the receiver.
 468                                                  * XXX: should we consider
 469                                                  * older rcvif?
 470                                                  */
 471                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 472                                         }
 473                                         icmp6_error(m, ICMP6_DST_UNREACH,
 474                                                     ICMP6_DST_UNREACH_ADDR, 0);
 475                                         ln->ln_hold = NULL;
 476                                 }
 477                                 next = nd6_free(rt);
 478                         }
 479                         break;
 480                 case ND6_LLINFO_REACHABLE:
 481                         if (ln->ln_expire) {
 482                                 ln->ln_state = ND6_LLINFO_STALE;
 483                                 ln->ln_expire = time_second + nd6_gctimer;
 484                         }
 485                         break;
 486
 487                 case ND6_LLINFO_STALE:
 488                         /* Garbage Collection(RFC 2461 5.3) */
 489                         if (ln->ln_expire)
 490                                 next = nd6_free(rt);
 491                         break;
 492
 493                 case ND6_LLINFO_DELAY:
 494                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
 495                                 /* We need NUD */
 496                                 ln->ln_asked = 1;
 497                                 ln->ln_state = ND6_LLINFO_PROBE;
 498                                 ln->ln_expire = time_second +
 499                                         ndi->retrans / 1000;
 500                                 nd6_ns_output(ifp, &dst->sin6_addr,
 501                                               &dst->sin6_addr,
 502                                               ln, 0);
 503                         } else {
 504                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 505                                 ln->ln_expire = time_second + nd6_gctimer;
 506                         }
 507                         break;
 508                 case ND6_LLINFO_PROBE:
 509                         if (ln->ln_asked < nd6_umaxtries) {
 510                                 ln->ln_asked++;
 511                                 ln->ln_expire = time_second +
 512                                         ND_IFINFO(ifp)->retrans / 1000;
 513                                 nd6_ns_output(ifp, &dst->sin6_addr,
 514                                                &dst->sin6_addr, ln, 0);
 515                         } else {
 516                                 next = nd6_free(rt);
 517                         }
 518                         break;
 519                 }
 520                 ln = next;
 521         }
 522
 523         /* expire default router list */
 524         dr = TAILQ_FIRST(&nd_defrouter);
 525         while (dr) {
 526                 if (dr->expire && dr->expire < time_second) {
 527                         struct nd_defrouter *t;
 528                         t = TAILQ_NEXT(dr, dr_entry);
 529                         defrtrlist_del(dr);
 530                         dr = t;
 531                 } else {
 532                         dr = TAILQ_NEXT(dr, dr_entry);
 533                 }
 534         }
 535
 536         /*
 537          * expire interface addresses.
 538          * in the past the loop was inside prefix expiry processing.
 539          * However, from a stricter speci-confrmance standpoint, we should
 540          * rather separate address lifetimes and prefix lifetimes.
 541          */
 542   addrloop:
 543         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 544                 nia6 = ia6->ia_next;
 545                 /* check address lifetime */
 546                 lt6 = &ia6->ia6_lifetime;
 547                 if (IFA6_IS_INVALID(ia6)) {
 548                         int regen = 0;
 549
 550                         /*
 551                          * If the expiring address is temporary, try
 552                          * regenerating a new one.  This would be useful when
 553                          * we suspended a laptop PC, then turned it on after a
 554                          * period that could invalidate all temporary
 555                          * addresses.  Although we may have to restart the
 556                          * loop (see below), it must be after purging the
 557                          * address.  Otherwise, we'd see an infinite loop of
 558                          * regeneration.
 559                          */
 560                         if (ip6_use_tempaddr &&
 561                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
 562                                 if (regen_tmpaddr(ia6) == 0)
 563                                         regen = 1;
 564                         }
 565
 566                         in6_purgeaddr(&ia6->ia_ifa);
 567
 568                         if (regen)
 569                                 goto addrloop; /* XXX: see below */
 570                 }
 571                 if (IFA6_IS_DEPRECATED(ia6)) {
 572                         int oldflags = ia6->ia6_flags;
 573
 574                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 575
 576                         /*
 577                          * If a temporary address has just become deprecated,
 578                          * regenerate a new one if possible.
 579                          */
 580                         if (ip6_use_tempaddr &&
 581                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 582                             (oldflags & IN6_IFF_DEPRECATED) == 0) {
 583
 584                                 if (regen_tmpaddr(ia6) == 0) {
 585                                         /*
 586                                          * A new temporary address is
 587                                          * generated.
 588                                          * XXX: this means the address chain
 589                                          * has changed while we are still in
 590                                          * the loop.  Although the change
 591                                          * would not cause disaster (because
 592                                          * it's not a deletion, but an
 593                                          * addition,) we'd rather restart the
 594                                          * loop just for safety.  Or does this
 595                                          * significantly reduce performance??
 596                                          */
 597                                         goto addrloop;
 598                                 }
 599                         }
 600                 } else {
 601                         /*
 602                          * A new RA might have made a deprecated address
 603                          * preferred.
 604                          */
 605                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 606                 }
 607         }
 608
 609         /* expire prefix list */
 610         pr = nd_prefix.lh_first;
 611         while (pr) {
 612                 /*
 613                  * check prefix lifetime.
 614                  * since pltime is just for autoconf, pltime processing for
 615                  * prefix is not necessary.
 616                  */
 617                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
 618                         struct nd_prefix *t;
 619                         t = pr->ndpr_next;
 620
 621                         /*
 622                          * address expiration and prefix expiration are
 623                          * separate.  NEVER perform in6_purgeaddr here.
 624                          */
 625
 626                         prelist_remove(pr);
 627                         pr = t;
 628                 } else
 629                         pr = pr->ndpr_next;
 630         }
 631         splx(s);
 632 }
 633
 634 static int
 635 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 636                                          address */
 637 {
 638         struct ifaddr *ifa;
 639         struct ifnet *ifp;
 640         struct in6_ifaddr *public_ifa6 = NULL;
 641
 642         ifp = ia6->ia_ifa.ifa_ifp;
 643         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_list) {
 644                 struct in6_ifaddr *it6;
 645
 646                 if (ifa->ifa_addr->sa_family != AF_INET6)
 647                         continue;
 648
 649                 it6 = (struct in6_ifaddr *)ifa;
 650
 651                 /* ignore no autoconf addresses. */
 652                 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
 653                         continue;
 654
 655                 /* ignore autoconf addresses with different prefixes. */
 656                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 657                         continue;
 658
 659                 /*
 660                  * Now we are looking at an autoconf address with the same
 661                  * prefix as ours.  If the address is temporary and is still
 662                  * preferred, do not create another one.  It would be rare, but
 663                  * could happen, for example, when we resume a laptop PC after
 664                  * a long period.
 665                  */
 666                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
 667                     !IFA6_IS_DEPRECATED(it6)) {
 668                         public_ifa6 = NULL;
 669                         break;
 670                 }
 671
 672                 /*
 673                  * This is a public autoconf address that has the same prefix
 674                  * as ours.  If it is preferred, keep it.  We can't break the
 675                  * loop here, because there may be a still-preferred temporary
 676                  * address with the prefix.
 677                  */
 678                 if (!IFA6_IS_DEPRECATED(it6))
 679                     public_ifa6 = it6;
 680         }
 681
 682         if (public_ifa6 != NULL) {
 683                 int e;
 684
 685                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 686                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 687                             " tmp addr,errno=%d\n", e);
 688                         return(-1);
 689                 }
 690                 return(0);
 691         }
 692
 693         return(-1);
 694 }
 695
 696 /*
 697  * Nuke neighbor cache/prefix/default router management table, right before
 698  * ifp goes away.
 699  */
 700 void
 701 nd6_purge(struct ifnet *ifp)
 702 {
 703         struct llinfo_nd6 *ln, *nln;
 704         struct nd_defrouter *dr, *ndr, drany;
 705         struct nd_prefix *pr, *npr;
 706
 707         /* Nuke default router list entries toward ifp */
 708         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 709                 /*
 710                  * The first entry of the list may be stored in
 711                  * the routing table, so we'll delete it later.
 712                  */
 713                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 714                         ndr = TAILQ_NEXT(dr, dr_entry);
 715                         if (dr->ifp == ifp)
 716                                 defrtrlist_del(dr);
 717                 }
 718                 dr = TAILQ_FIRST(&nd_defrouter);
 719                 if (dr->ifp == ifp)
 720                         defrtrlist_del(dr);
 721         }
 722
 723         /* Nuke prefix list entries toward ifp */
 724         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 725                 npr = pr->ndpr_next;
 726                 if (pr->ndpr_ifp == ifp) {
 727                         /*
 728                          * Previously, pr->ndpr_addr is removed as well,
 729                          * but I strongly believe we don't have to do it.
 730                          * nd6_purge() is only called from in6_ifdetach(),
 731                          * which removes all the associated interface addresses
 732                          * by itself.
 733                          * (jinmei@kame.net 20010129)
 734                          */
 735                         prelist_remove(pr);
 736                 }
 737         }
 738
 739         /* cancel default outgoing interface setting */
 740         if (nd6_defifindex == ifp->if_index)
 741                 nd6_setdefaultiface(0);
 742
 743         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 744                 /* refresh default router list */
 745                 bzero(&drany, sizeof(drany));
 746                 defrouter_delreq(&drany, 0);
 747                 defrouter_select();
 748         }
 749
 750         /*
 751          * Nuke neighbor cache entries for the ifp.
 752          * Note that rt->rt_ifp may not be the same as ifp,
 753          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 754          * nd6_rtrequest(), and ip6_input().
 755          */
 756         ln = llinfo_nd6.ln_next;
 757         while (ln && ln != &llinfo_nd6) {
 758                 struct rtentry *rt;
 759                 struct sockaddr_dl *sdl;
 760
 761                 nln = ln->ln_next;
 762                 rt = ln->ln_rt;
 763                 if (rt && rt->rt_gateway &&
 764                     rt->rt_gateway->sa_family == AF_LINK) {
 765                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 766                         if (sdl->sdl_index == ifp->if_index)
 767                                 nln = nd6_free(rt);
 768                 }
 769                 ln = nln;
 770         }
 771 }
 772
 773 struct rtentry *
 774 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 775 {
 776         struct rtentry *rt;
 777         struct sockaddr_in6 sin6;
 778
 779         bzero(&sin6, sizeof(sin6));
 780         sin6.sin6_len = sizeof(struct sockaddr_in6);
 781         sin6.sin6_family = AF_INET6;
 782         sin6.sin6_addr = *addr6;
 783
 784         if (create)
 785                 rt = rtlookup((struct sockaddr *)&sin6);
 786         else
 787                 rt = rtpurelookup((struct sockaddr *)&sin6);
 788         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 789                 /*
 790                  * This is the case for the default route.
 791                  * If we want to create a neighbor cache for the address, we
 792                  * should free the route for the destination and allocate an
 793                  * interface route.
 794                  */
 795                 if (create) {
 796                         --rt->rt_refcnt;
 797                         rt = NULL;
 798                 }
 799         }
 800         if (!rt) {
 801                 if (create && ifp) {
 802                         int e;
 803
 804                         /*
 805                          * If no route is available and create is set,
 806                          * we allocate a host route for the destination
 807                          * and treat it like an interface route.
 808                          * This hack is necessary for a neighbor which can't
 809                          * be covered by our own prefix.
 810                          */
 811                         struct ifaddr *ifa =
 812                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 813                         if (ifa == NULL)
 814                                 return(NULL);
 815
 816                         /*
 817                          * Create a new route.  RTF_LLINFO is necessary
 818                          * to create a Neighbor Cache entry for the
 819                          * destination in nd6_rtrequest which will be
 820                          * called in rtrequest via ifa->ifa_rtrequest.
 821                          */
 822                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 823                                            ifa->ifa_addr,
 824                                            (struct sockaddr *)&all1_sa,
 825                                            (ifa->ifa_flags |
 826                                             RTF_HOST | RTF_LLINFO) &
 827                                            ~RTF_CLONING,
 828                                            &rt)) != 0)
 829                                 log(LOG_ERR,
 830                                     "nd6_lookup: failed to add route for a "
 831                                     "neighbor(%s), errno=%d\n",
 832                                     ip6_sprintf(addr6), e);
 833                         if (rt == NULL)
 834                                 return(NULL);
 835                         if (rt->rt_llinfo) {
 836                                 struct llinfo_nd6 *ln =
 837                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 838                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 839                         }
 840                 } else
 841                         return(NULL);
 842         }
 843         rt->rt_refcnt--;
 844         /*
 845          * Validation for the entry.
 846          * Note that the check for rt_llinfo is necessary because a cloned
 847          * route from a parent route that has the L flag (e.g. the default
 848          * route to a p2p interface) may have the flag, too, while the
 849          * destination is not actually a neighbor.
 850          * XXX: we can't use rt->rt_ifp to check for the interface, since
 851          *      it might be the loopback interface if the entry is for our
 852          *      own address on a non-loopback interface. Instead, we should
 853          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 854          *      interface.
 855          */
 856         if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
 857             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 858             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 859                 if (create) {
 860                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 861                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 862                         /* xxx more logs... kazu */
 863                 }
 864                 return(NULL);
 865         }
 866         return(rt);
 867 }
 868
 869 /*
 870  * Detect if a given IPv6 address identifies a neighbor on a given link.
 871  * XXX: should take care of the destination of a p2p link?
 872  */
 873 int
 874 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 875 {
 876         struct ifaddr *ifa;
 877         int i;
 878
 879 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 880 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 881
 882         /*
 883          * A link-local address is always a neighbor.
 884          * XXX: we should use the sin6_scope_id field rather than the embedded
 885          * interface index.
 886          */
 887         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 888             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 889                 return(1);
 890
 891         /*
 892          * If the address matches one of our addresses,
 893          * it should be a neighbor.
 894          */
 895         TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
 896                 if (ifa->ifa_addr->sa_family != AF_INET6)
 897                         next: continue;
 898
 899                 for (i = 0; i < 4; i++) {
 900                         if ((IFADDR6(ifa).s6_addr32[i] ^
 901                              addr->sin6_addr.s6_addr32[i]) &
 902                             IFMASK6(ifa).s6_addr32[i])
 903                                 goto next;
 904                 }
 905                 return(1);
 906         }
 907
 908         /*
 909          * Even if the address matches none of our addresses, it might be
 910          * in the neighbor cache.
 911          */
 912         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 913                 return(1);
 914
 915         return(0);
 916 #undef IFADDR6
 917 #undef IFMASK6
 918 }
 919
 920 /*
 921  * Free an nd6 llinfo entry.
 922  */
 923 struct llinfo_nd6 *
 924 nd6_free(struct rtentry *rt)
 925 {
 926         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 927         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 928         struct nd_defrouter *dr;
 929
 930         /*
 931          * we used to have pfctlinput(PRC_HOSTDEAD) here.
 932          * even though it is not harmful, it was not really necessary.
 933          */
 934
 935         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 936                 int s;
 937                 s = splnet();
 938                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 939                                       rt->rt_ifp);
 940
 941                 if (ln->ln_router || dr) {
 942                         /*
 943                          * rt6_flush must be called whether or not the neighbor
 944                          * is in the Default Router List.
 945                          * See a corresponding comment in nd6_na_input().
 946                          */
 947                         rt6_flush(&in6, rt->rt_ifp);
 948                 }
 949
 950                 if (dr) {
 951                         /*
 952                          * Unreachablity of a router might affect the default
 953                          * router selection and on-link detection of advertised
 954                          * prefixes.
 955                          */
 956
 957                         /*
 958                          * Temporarily fake the state to choose a new default
 959                          * router and to perform on-link determination of
 960                          * prefixes correctly.
 961                          * Below the state will be set correctly,
 962                          * or the entry itself will be deleted.
 963                          */
 964                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 965
 966                         /*
 967                          * Since defrouter_select() does not affect the
 968                          * on-link determination and MIP6 needs the check
 969                          * before the default router selection, we perform
 970                          * the check now.
 971                          */
 972                         pfxlist_onlink_check();
 973
 974                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 975                                 /*
 976                                  * It is used as the current default router,
 977                                  * so we have to move it to the end of the
 978                                  * list and choose a new one.
 979                                  * XXX: it is not very efficient if this is
 980                                  *      the only router.
 981                                  */
 982                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
 983                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
 984
 985                                 defrouter_select();
 986                         }
 987                 }
 988                 splx(s);
 989         }
 990
 991         /*
 992          * Before deleting the entry, remember the next entry as the
 993          * return value.  We need this because pfxlist_onlink_check() above
 994          * might have freed other entries (particularly the old next entry) as
 995          * a side effect (XXX).
 996          */
 997         next = ln->ln_next;
 998
 999         /*
1000          * Detach the route from the routing tree and the list of neighbor
1001          * caches, and disable the route entry not to be used in already
1002          * cached routes.
1003          */
1004         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1005                   rt_mask(rt), 0, (struct rtentry **)0);
1006
1007         return(next);
1008 }
1009
1010 /*
1011  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1012  *
1013  * XXX cost-effective metods?
1014  */
1015 void
1016 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1017 {
1018         struct llinfo_nd6 *ln;
1019
1020         /*
1021          * If the caller specified "rt", use that.  Otherwise, resolve the
1022          * routing table by supplied "dst6".
1023          */
1024         if (!rt) {
1025                 if (!dst6)
1026                         return;
1027                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1028                         return;
1029         }
1030
1031         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1032             (rt->rt_flags & RTF_LLINFO) == 0 ||
1033             !rt->rt_llinfo || !rt->rt_gateway ||
1034             rt->rt_gateway->sa_family != AF_LINK) {
1035                 /* This is not a host route. */
1036                 return;
1037         }
1038
1039         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1040         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1041                 return;
1042
1043         /*
1044          * if we get upper-layer reachability confirmation many times,
1045          * it is possible we have false information.
1046          */
1047         if (!force) {
1048                 ln->ln_byhint++;
1049                 if (ln->ln_byhint > nd6_maxnudhint)
1050                         return;
1051         }
1052
1053         ln->ln_state = ND6_LLINFO_REACHABLE;
1054         if (ln->ln_expire)
1055                 ln->ln_expire = time_second +
1056                         ND_IFINFO(rt->rt_ifp)->reachable;
1057 }
1058
1059 void
1060 nd6_rtrequest(int req, struct rtentry *rt,
1061               struct rt_addrinfo *info) /* xxx unused */
1062 {
1063         struct sockaddr *gate = rt->rt_gateway;
1064         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1065         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1066         struct ifnet *ifp = rt->rt_ifp;
1067         struct ifaddr *ifa;
1068
1069         if ((rt->rt_flags & RTF_GATEWAY))
1070                 return;
1071
1072         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1073                 /*
1074                  * This is probably an interface direct route for a link
1075                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1076                  * We do not need special treatment below for such a route.
1077                  * Moreover, the RTF_LLINFO flag which would be set below
1078                  * would annoy the ndp(8) command.
1079                  */
1080                 return;
1081         }
1082
1083         if (req == RTM_RESOLVE &&
1084             (nd6_need_cache(ifp) == 0 || /* stf case */
1085              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1086                 /*
1087                  * FreeBSD and BSD/OS often make a cloned host route based
1088                  * on a less-specific route (e.g. the default route).
1089                  * If the less specific route does not have a "gateway"
1090                  * (this is the case when the route just goes to a p2p or an
1091                  * stf interface), we'll mistakenly make a neighbor cache for
1092                  * the host route, and will see strange neighbor solicitation
1093                  * for the corresponding destination.  In order to avoid the
1094                  * confusion, we check if the destination of the route is
1095                  * a neighbor in terms of neighbor discovery, and stop the
1096                  * process if not.  Additionally, we remove the LLINFO flag
1097                  * so that ndp(8) will not try to get the neighbor information
1098                  * of the destination.
1099                  */
1100                 rt->rt_flags &= ~RTF_LLINFO;
1101                 return;
1102         }
1103
1104         switch (req) {
1105         case RTM_ADD:
1106                 /*
1107                  * There is no backward compatibility :)
1108                  *
1109                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1110                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1111                  *         rt->rt_flags |= RTF_CLONING;
1112                  */
1113                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1114                         /*
1115                          * Case 1: This route should come from
1116                          * a route to interface.  RTF_LLINFO flag is set
1117                          * for a host route whose destination should be
1118                          * treated as on-link.
1119                          */
1120                         rt_setgate(rt, rt_key(rt),
1121                                    (struct sockaddr *)&null_sdl);
1122                         gate = rt->rt_gateway;
1123                         SDL(gate)->sdl_type = ifp->if_type;
1124                         SDL(gate)->sdl_index = ifp->if_index;
1125                         if (ln)
1126                                 ln->ln_expire = time_second;
1127 #if 1
1128                         if (ln && ln->ln_expire == 0) {
1129                                 /* kludge for desktops */
1130 #if 0
1131                                 printf("nd6_rtequest: time.tv_sec is zero; "
1132                                        "treat it as 1\n");
1133 #endif
1134                                 ln->ln_expire = 1;
1135                         }
1136 #endif
1137                         if ((rt->rt_flags & RTF_CLONING))
1138                                 break;
1139                 }
1140                 /*
1141                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1142                  * We don't do that here since llinfo is not ready yet.
1143                  *
1144                  * There are also couple of other things to be discussed:
1145                  * - unsolicited NA code needs improvement beforehand
1146                  * - RFC2461 says we MAY send multicast unsolicited NA
1147                  *   (7.2.6 paragraph 4), however, it also says that we
1148                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1149                  *   we don't have anything like it right now.
1150                  *   note that the mechanism needs a mutual agreement
1151                  *   between proxies, which means that we need to implement
1152                  *   a new protocol, or a new kludge.
1153                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1154                  *   we need to check ip6forwarding before sending it.
1155                  *   (or should we allow proxy ND configuration only for
1156                  *   routers?  there's no mention about proxy ND from hosts)
1157                  */
1158 #if 0
1159                 /* XXX it does not work */
1160                 if (rt->rt_flags & RTF_ANNOUNCE)
1161                         nd6_na_output(ifp,
1162                               &SIN6(rt_key(rt))->sin6_addr,
1163                               &SIN6(rt_key(rt))->sin6_addr,
1164                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1165                               1, NULL);
1166 #endif
1167                 /* FALLTHROUGH */
1168         case RTM_RESOLVE:
1169                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1170                         /*
1171                          * Address resolution isn't necessary for a point to
1172                          * point link, so we can skip this test for a p2p link.
1173                          */
1174                         if (gate->sa_family != AF_LINK ||
1175                             gate->sa_len < sizeof(null_sdl)) {
1176                                 log(LOG_DEBUG,
1177                                     "nd6_rtrequest: bad gateway value: %s\n",
1178                                     if_name(ifp));
1179                                 break;
1180                         }
1181                         SDL(gate)->sdl_type = ifp->if_type;
1182                         SDL(gate)->sdl_index = ifp->if_index;
1183                 }
1184                 if (ln != NULL)
1185                         break;  /* This happens on a route change */
1186                 /*
1187                  * Case 2: This route may come from cloning, or a manual route
1188                  * add with a LL address.
1189                  */
1190                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1191                 rt->rt_llinfo = (caddr_t)ln;
1192                 if (!ln) {
1193                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1194                         break;
1195                 }
1196                 nd6_inuse++;
1197                 nd6_allocated++;
1198                 bzero(ln, sizeof(*ln));
1199                 ln->ln_rt = rt;
1200                 /* this is required for "ndp" command. - shin */
1201                 if (req == RTM_ADD) {
1202                         /*
1203                          * gate should have some valid AF_LINK entry,
1204                          * and ln->ln_expire should have some lifetime
1205                          * which is specified by ndp command.
1206                          */
1207                         ln->ln_state = ND6_LLINFO_REACHABLE;
1208                         ln->ln_byhint = 0;
1209                 } else {
1210                         /*
1211                          * When req == RTM_RESOLVE, rt is created and
1212                          * initialized in rtrequest(), so rt_expire is 0.
1213                          */
1214                         ln->ln_state = ND6_LLINFO_NOSTATE;
1215                         ln->ln_expire = time_second;
1216                 }
1217                 rt->rt_flags |= RTF_LLINFO;
1218                 ln->ln_next = llinfo_nd6.ln_next;
1219                 llinfo_nd6.ln_next = ln;
1220                 ln->ln_prev = &llinfo_nd6;
1221                 ln->ln_next->ln_prev = ln;
1222
1223                 /*
1224                  * check if rt_key(rt) is one of my address assigned
1225                  * to the interface.
1226                  */
1227                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1228                                           &SIN6(rt_key(rt))->sin6_addr);
1229                 if (ifa) {
1230                         caddr_t macp = nd6_ifptomac(ifp);
1231                         ln->ln_expire = 0;
1232                         ln->ln_state = ND6_LLINFO_REACHABLE;
1233                         ln->ln_byhint = 0;
1234                         if (macp) {
1235                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1236                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1237                         }
1238                         if (nd6_useloopback) {
1239                                 rt->rt_ifp = &loif[0];  /* XXX */
1240                                 /*
1241                                  * Make sure rt_ifa be equal to the ifaddr
1242                                  * corresponding to the address.
1243                                  * We need this because when we refer
1244                                  * rt_ifa->ia6_flags in ip6_input, we assume
1245                                  * that the rt_ifa points to the address instead
1246                                  * of the loopback address.
1247                                  */
1248                                 if (ifa != rt->rt_ifa) {
1249                                         IFAFREE(rt->rt_ifa);
1250                                         IFAREF(ifa);
1251                                         rt->rt_ifa = ifa;
1252                                 }
1253                         }
1254                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1255                         ln->ln_expire = 0;
1256                         ln->ln_state = ND6_LLINFO_REACHABLE;
1257                         ln->ln_byhint = 0;
1258
1259                         /* join solicited node multicast for proxy ND */
1260                         if (ifp->if_flags & IFF_MULTICAST) {
1261                                 struct in6_addr llsol;
1262                                 int error;
1263
1264                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1265                                 llsol.s6_addr16[0] = htons(0xff02);
1266                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1267                                 llsol.s6_addr32[1] = 0;
1268                                 llsol.s6_addr32[2] = htonl(1);
1269                                 llsol.s6_addr8[12] = 0xff;
1270
1271                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1272                                         nd6log((LOG_ERR, "%s: failed to join "
1273                                             "%s (errno=%d)\n", if_name(ifp),
1274                                             ip6_sprintf(&llsol), error));
1275                                 }
1276                         }
1277                 }
1278                 break;
1279
1280         case RTM_DELETE:
1281                 if (!ln)
1282                         break;
1283                 /* leave from solicited node multicast for proxy ND */
1284                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1285                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1286                         struct in6_addr llsol;
1287                         struct in6_multi *in6m;
1288
1289                         llsol = SIN6(rt_key(rt))->sin6_addr;
1290                         llsol.s6_addr16[0] = htons(0xff02);
1291                         llsol.s6_addr16[1] = htons(ifp->if_index);
1292                         llsol.s6_addr32[1] = 0;
1293                         llsol.s6_addr32[2] = htonl(1);
1294                         llsol.s6_addr8[12] = 0xff;
1295
1296                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1297                         if (in6m)
1298                                 in6_delmulti(in6m);
1299                 }
1300                 nd6_inuse--;
1301                 ln->ln_next->ln_prev = ln->ln_prev;
1302                 ln->ln_prev->ln_next = ln->ln_next;
1303                 ln->ln_prev = NULL;
1304                 rt->rt_llinfo = 0;
1305                 rt->rt_flags &= ~RTF_LLINFO;
1306                 if (ln->ln_hold)
1307                         m_freem(ln->ln_hold);
1308                 Free((caddr_t)ln);
1309         }
1310 }
1311
1312 int
1313 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1314 {
1315         struct in6_drlist *drl = (struct in6_drlist *)data;
1316         struct in6_prlist *prl = (struct in6_prlist *)data;
1317         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1318         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1319         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1320         struct nd_defrouter *dr, any;
1321         struct nd_prefix *pr;
1322         struct rtentry *rt;
1323         int i = 0, error = 0;
1324         int s;
1325
1326         switch (cmd) {
1327         case SIOCGDRLST_IN6:
1328                 /*
1329                  * obsolete API, use sysctl under net.inet6.icmp6
1330                  */
1331                 bzero(drl, sizeof(*drl));
1332                 s = splnet();
1333                 dr = TAILQ_FIRST(&nd_defrouter);
1334                 while (dr && i < DRLSTSIZ) {
1335                         drl->defrouter[i].rtaddr = dr->rtaddr;
1336                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1337                                 /* XXX: need to this hack for KAME stack */
1338                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1339                         } else
1340                                 log(LOG_ERR,
1341                                     "default router list contains a "
1342                                     "non-linklocal address(%s)\n",
1343                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1344
1345                         drl->defrouter[i].flags = dr->flags;
1346                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1347                         drl->defrouter[i].expire = dr->expire;
1348                         drl->defrouter[i].if_index = dr->ifp->if_index;
1349                         i++;
1350                         dr = TAILQ_NEXT(dr, dr_entry);
1351                 }
1352                 splx(s);
1353                 break;
1354         case SIOCGPRLST_IN6:
1355                 /*
1356                  * obsolete API, use sysctl under net.inet6.icmp6
1357                  */
1358                 /*
1359                  * XXX meaning of fields, especialy "raflags", is very
1360                  * differnet between RA prefix list and RR/static prefix list.
1361                  * how about separating ioctls into two?
1362                  */
1363                 bzero(prl, sizeof(*prl));
1364                 s = splnet();
1365                 pr = nd_prefix.lh_first;
1366                 while (pr && i < PRLSTSIZ) {
1367                         struct nd_pfxrouter *pfr;
1368                         int j;
1369
1370                         (void)in6_embedscope(&prl->prefix[i].prefix,
1371                             &pr->ndpr_prefix, NULL, NULL);
1372                         prl->prefix[i].raflags = pr->ndpr_raf;
1373                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1374                         prl->prefix[i].vltime = pr->ndpr_vltime;
1375                         prl->prefix[i].pltime = pr->ndpr_pltime;
1376                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1377                         prl->prefix[i].expire = pr->ndpr_expire;
1378
1379                         pfr = pr->ndpr_advrtrs.lh_first;
1380                         j = 0;
1381                         while (pfr) {
1382                                 if (j < DRLSTSIZ) {
1383 #define RTRADDR prl->prefix[i].advrtr[j]
1384                                         RTRADDR = pfr->router->rtaddr;
1385                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1386                                                 /* XXX: hack for KAME */
1387                                                 RTRADDR.s6_addr16[1] = 0;
1388                                         } else
1389                                                 log(LOG_ERR,
1390                                                     "a router(%s) advertises "
1391                                                     "a prefix with "
1392                                                     "non-link local address\n",
1393                                                     ip6_sprintf(&RTRADDR));
1394 #undef RTRADDR
1395                                 }
1396                                 j++;
1397                                 pfr = pfr->pfr_next;
1398                         }
1399                         prl->prefix[i].advrtrs = j;
1400                         prl->prefix[i].origin = PR_ORIG_RA;
1401
1402                         i++;
1403                         pr = pr->ndpr_next;
1404                 }
1405               {
1406                 struct rr_prefix *rpp;
1407
1408                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1409                      rpp = LIST_NEXT(rpp, rp_entry)) {
1410                         if (i >= PRLSTSIZ)
1411                                 break;
1412                         (void)in6_embedscope(&prl->prefix[i].prefix,
1413                             &pr->ndpr_prefix, NULL, NULL);
1414                         prl->prefix[i].raflags = rpp->rp_raf;
1415                         prl->prefix[i].prefixlen = rpp->rp_plen;
1416                         prl->prefix[i].vltime = rpp->rp_vltime;
1417                         prl->prefix[i].pltime = rpp->rp_pltime;
1418                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1419                         prl->prefix[i].expire = rpp->rp_expire;
1420                         prl->prefix[i].advrtrs = 0;
1421                         prl->prefix[i].origin = rpp->rp_origin;
1422                         i++;
1423                 }
1424               }
1425                 splx(s);
1426
1427                 break;
1428         case OSIOCGIFINFO_IN6:
1429                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1430                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1431                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1432                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1433                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1434                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1435                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1436                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1437                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1438                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1439                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1440                 break;
1441         case SIOCGIFINFO_IN6:
1442                 ndi->ndi = *ND_IFINFO(ifp);
1443                 break;
1444         case SIOCSIFINFO_FLAGS:
1445                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1446                 break;
1447         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1448                 /* flush default router list */
1449                 /*
1450                  * xxx sumikawa: should not delete route if default
1451                  * route equals to the top of default router list
1452                  */
1453                 bzero(&any, sizeof(any));
1454                 defrouter_delreq(&any, 0);
1455                 defrouter_select();
1456                 /* xxx sumikawa: flush prefix list */
1457                 break;
1458         case SIOCSPFXFLUSH_IN6:
1459             {
1460                 /* flush all the prefix advertised by routers */
1461                 struct nd_prefix *pr, *next;
1462
1463                 s = splnet();
1464                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1465                         struct in6_ifaddr *ia, *ia_next;
1466
1467                         next = pr->ndpr_next;
1468
1469                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1470                                 continue; /* XXX */
1471
1472                         /* do we really have to remove addresses as well? */
1473                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1474                                 /* ia might be removed.  keep the next ptr. */
1475                                 ia_next = ia->ia_next;
1476
1477                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1478                                         continue;
1479
1480                                 if (ia->ia6_ndpr == pr)
1481                                         in6_purgeaddr(&ia->ia_ifa);
1482                         }
1483                         prelist_remove(pr);
1484                 }
1485                 splx(s);
1486                 break;
1487             }
1488         case SIOCSRTRFLUSH_IN6:
1489             {
1490                 /* flush all the default routers */
1491                 struct nd_defrouter *dr, *next;
1492
1493                 s = splnet();
1494                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1495                         /*
1496                          * The first entry of the list may be stored in
1497                          * the routing table, so we'll delete it later.
1498                          */
1499                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1500                                 next = TAILQ_NEXT(dr, dr_entry);
1501                                 defrtrlist_del(dr);
1502                         }
1503                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1504                 }
1505                 splx(s);
1506                 break;
1507             }
1508         case SIOCGNBRINFO_IN6:
1509             {
1510                 struct llinfo_nd6 *ln;
1511                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1512
1513                 /*
1514                  * XXX: KAME specific hack for scoped addresses
1515                  *      XXXX: for other scopes than link-local?
1516                  */
1517                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1518                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1519                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1520
1521                         if (*idp == 0)
1522                                 *idp = htons(ifp->if_index);
1523                 }
1524
1525                 s = splnet();
1526                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1527                         error = EINVAL;
1528                         splx(s);
1529                         break;
1530                 }
1531                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1532                 nbi->state = ln->ln_state;
1533                 nbi->asked = ln->ln_asked;
1534                 nbi->isrouter = ln->ln_router;
1535                 nbi->expire = ln->ln_expire;
1536                 splx(s);
1537
1538                 break;
1539             }
1540         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1541                 ndif->ifindex = nd6_defifindex;
1542                 break;
1543         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1544                 return(nd6_setdefaultiface(ndif->ifindex));
1545                 break;
1546         }
1547         return(error);
1548 }
1549
1550 /*
1551  * Create neighbor cache entry and cache link-layer address,
1552  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1553  */
1554 struct rtentry *
1555 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1556                  int lladdrlen,
1557                  int type,      /* ICMP6 type */
1558                  int code       /* type dependent information */)
1559 {
1560         struct rtentry *rt = NULL;
1561         struct llinfo_nd6 *ln = NULL;
1562         int is_newentry;
1563         struct sockaddr_dl *sdl = NULL;
1564         int do_update;
1565         int olladdr;
1566         int llchange;
1567         int newstate = 0;
1568
1569         if (!ifp)
1570                 panic("ifp == NULL in nd6_cache_lladdr");
1571         if (!from)
1572                 panic("from == NULL in nd6_cache_lladdr");
1573
1574         /* nothing must be updated for unspecified address */
1575         if (IN6_IS_ADDR_UNSPECIFIED(from))
1576                 return NULL;
1577
1578         /*
1579          * Validation about ifp->if_addrlen and lladdrlen must be done in
1580          * the caller.
1581          *
1582          * XXX If the link does not have link-layer adderss, what should
1583          * we do? (ifp->if_addrlen == 0)
1584          * Spec says nothing in sections for RA, RS and NA.  There's small
1585          * description on it in NS section (RFC 2461 7.2.3).
1586          */
1587
1588         rt = nd6_lookup(from, 0, ifp);
1589         if (!rt) {
1590 #if 0
1591                 /* nothing must be done if there's no lladdr */
1592                 if (!lladdr || !lladdrlen)
1593                         return NULL;
1594 #endif
1595
1596                 rt = nd6_lookup(from, 1, ifp);
1597                 is_newentry = 1;
1598         } else {
1599                 /* do nothing if static ndp is set */
1600                 if (rt->rt_flags & RTF_STATIC)
1601                         return NULL;
1602                 is_newentry = 0;
1603         }
1604
1605         if (!rt)
1606                 return NULL;
1607         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1608 fail:
1609                 (void)nd6_free(rt);
1610                 return NULL;
1611         }
1612         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1613         if (!ln)
1614                 goto fail;
1615         if (!rt->rt_gateway)
1616                 goto fail;
1617         if (rt->rt_gateway->sa_family != AF_LINK)
1618                 goto fail;
1619         sdl = SDL(rt->rt_gateway);
1620
1621         olladdr = (sdl->sdl_alen) ? 1 : 0;
1622         if (olladdr && lladdr) {
1623                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1624                         llchange = 1;
1625                 else
1626                         llchange = 0;
1627         } else
1628                 llchange = 0;
1629
1630         /*
1631          * newentry olladdr  lladdr  llchange   (*=record)
1632          *      0       n       n       --      (1)
1633          *      0       y       n       --      (2)
1634          *      0       n       y       --      (3) * STALE
1635          *      0       y       y       n       (4) *
1636          *      0       y       y       y       (5) * STALE
1637          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1638          *      1       --      y       --      (7) * STALE
1639          */
1640
1641         if (lladdr) {           /* (3-5) and (7) */
1642                 /*
1643                  * Record source link-layer address
1644                  * XXX is it dependent to ifp->if_type?
1645                  */
1646                 sdl->sdl_alen = ifp->if_addrlen;
1647                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1648         }
1649
1650         if (!is_newentry) {
1651                 if ((!olladdr && lladdr)                /* (3) */
1652                  || (olladdr && lladdr && llchange)) {  /* (5) */
1653                         do_update = 1;
1654                         newstate = ND6_LLINFO_STALE;
1655                 } else                                  /* (1-2,4) */
1656                         do_update = 0;
1657         } else {
1658                 do_update = 1;
1659                 if (!lladdr)                            /* (6) */
1660                         newstate = ND6_LLINFO_NOSTATE;
1661                 else                                    /* (7) */
1662                         newstate = ND6_LLINFO_STALE;
1663         }
1664
1665         if (do_update) {
1666                 /*
1667                  * Update the state of the neighbor cache.
1668                  */
1669                 ln->ln_state = newstate;
1670
1671                 if (ln->ln_state == ND6_LLINFO_STALE) {
1672                         /*
1673                          * XXX: since nd6_output() below will cause
1674                          * state tansition to DELAY and reset the timer,
1675                          * we must set the timer now, although it is actually
1676                          * meaningless.
1677                          */
1678                         ln->ln_expire = time_second + nd6_gctimer;
1679
1680                         if (ln->ln_hold) {
1681                                 /*
1682                                  * we assume ifp is not a p2p here, so just
1683                                  * set the 2nd argument as the 1st one.
1684                                  */
1685                                 nd6_output(ifp, ifp, ln->ln_hold,
1686                                            (struct sockaddr_in6 *)rt_key(rt),
1687                                            rt);
1688                                 ln->ln_hold = NULL;
1689                         }
1690                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1691                         /* probe right away */
1692                         ln->ln_expire = time_second;
1693                 }
1694         }
1695
1696         /*
1697          * ICMP6 type dependent behavior.
1698          *
1699          * NS: clear IsRouter if new entry
1700          * RS: clear IsRouter
1701          * RA: set IsRouter if there's lladdr
1702          * redir: clear IsRouter if new entry
1703          *
1704          * RA case, (1):
1705          * The spec says that we must set IsRouter in the following cases:
1706          * - If lladdr exist, set IsRouter.  This means (1-5).
1707          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1708          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1709          * A quetion arises for (1) case.  (1) case has no lladdr in the
1710          * neighbor cache, this is similar to (6).
1711          * This case is rare but we figured that we MUST NOT set IsRouter.
1712          *
1713          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1714          *                                                      D R
1715          *      0       n       n       --      (1)     c   ?     s
1716          *      0       y       n       --      (2)     c   s     s
1717          *      0       n       y       --      (3)     c   s     s
1718          *      0       y       y       n       (4)     c   s     s
1719          *      0       y       y       y       (5)     c   s     s
1720          *      1       --      n       --      (6) c   c       c s
1721          *      1       --      y       --      (7) c   c   s   c s
1722          *
1723          *                                      (c=clear s=set)
1724          */
1725         switch (type & 0xff) {
1726         case ND_NEIGHBOR_SOLICIT:
1727                 /*
1728                  * New entry must have is_router flag cleared.
1729                  */
1730                 if (is_newentry)        /* (6-7) */
1731                         ln->ln_router = 0;
1732                 break;
1733         case ND_REDIRECT:
1734                 /*
1735                  * If the icmp is a redirect to a better router, always set the
1736                  * is_router flag. Otherwise, if the entry is newly created,
1737                  * clear the flag. [RFC 2461, sec 8.3]
1738                  */
1739                 if (code == ND_REDIRECT_ROUTER)
1740                         ln->ln_router = 1;
1741                 else if (is_newentry) /* (6-7) */
1742                         ln->ln_router = 0;
1743                 break;
1744         case ND_ROUTER_SOLICIT:
1745                 /*
1746                  * is_router flag must always be cleared.
1747                  */
1748                 ln->ln_router = 0;
1749                 break;
1750         case ND_ROUTER_ADVERT:
1751                 /*
1752                  * Mark an entry with lladdr as a router.
1753                  */
1754                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1755                  || (is_newentry && lladdr)) {                  /* (7) */
1756                         ln->ln_router = 1;
1757                 }
1758                 break;
1759         }
1760
1761         /*
1762          * When the link-layer address of a router changes, select the
1763          * best router again.  In particular, when the neighbor entry is newly
1764          * created, it might affect the selection policy.
1765          * Question: can we restrict the first condition to the "is_newentry"
1766          * case?
1767          * XXX: when we hear an RA from a new router with the link-layer
1768          * address option, defrouter_select() is called twice, since
1769          * defrtrlist_update called the function as well.  However, I believe
1770          * we can compromise the overhead, since it only happens the first
1771          * time.
1772          * XXX: although defrouter_select() should not have a bad effect
1773          * for those are not autoconfigured hosts, we explicitly avoid such
1774          * cases for safety.
1775          */
1776         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1777                 defrouter_select();
1778
1779         return rt;
1780 }
1781
1782 static void
1783 nd6_slowtimo(void *ignored_arg)
1784 {
1785         int s = splnet();
1786         struct nd_ifinfo *nd6if;
1787         struct ifnet *ifp;
1788
1789         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1790             nd6_slowtimo, NULL);
1791         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1792                 nd6if = ND_IFINFO(ifp);
1793                 if (nd6if->basereachable && /* already initialized */
1794                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1795                         /*
1796                          * Since reachable time rarely changes by router
1797                          * advertisements, we SHOULD insure that a new random
1798                          * value gets recomputed at least once every few hours.
1799                          * (RFC 2461, 6.3.4)
1800                          */
1801                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1802                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1803                 }
1804         }
1805         splx(s);
1806 }
1807
1808 #define senderr(e) { error = (e); goto bad;}
1809 int
1810 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1811            struct sockaddr_in6 *dst, struct rtentry *rt0)
1812 {
1813         struct mbuf *m = m0;
1814         struct rtentry *rt = rt0;
1815         struct sockaddr_in6 *gw6 = NULL;
1816         struct llinfo_nd6 *ln = NULL;
1817         int error = 0;
1818
1819         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1820                 goto sendpkt;
1821
1822         if (nd6_need_cache(ifp) == 0)
1823                 goto sendpkt;
1824
1825         /*
1826          * next hop determination.  This routine is derived from ether_outpout.
1827          */
1828         if (rt) {
1829                 if (!(rt->rt_flags & RTF_UP)) {
1830                         if ((rt0 = rt = rtlookup((struct sockaddr *)dst))) {
1831                                 rt->rt_refcnt--;
1832                                 if (rt->rt_ifp != ifp) {
1833                                         /* XXX: loop care? */
1834                                         return nd6_output(ifp, origifp, m0,
1835                                                           dst, rt);
1836                                 }
1837                         } else
1838                                 senderr(EHOSTUNREACH);
1839                 }
1840
1841                 if (rt->rt_flags & RTF_GATEWAY) {
1842                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1843
1844                         /*
1845                          * We skip link-layer address resolution and NUD
1846                          * if the gateway is not a neighbor from ND point
1847                          * of view, regardless of the value of nd_ifinfo.flags.
1848                          * The second condition is a bit tricky; we skip
1849                          * if the gateway is our own address, which is
1850                          * sometimes used to install a route to a p2p link.
1851                          */
1852                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1853                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1854                                 /*
1855                                  * We allow this kind of tricky route only
1856                                  * when the outgoing interface is p2p.
1857                                  * XXX: we may need a more generic rule here.
1858                                  */
1859                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1860                                         senderr(EHOSTUNREACH);
1861
1862                                 goto sendpkt;
1863                         }
1864
1865                         if (rt->rt_gwroute == NULL)
1866                                 goto lookup;
1867                         if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1868                                 rtfree(rt->rt_gwroute);
1869 lookup:                         rt->rt_gwroute = rtlookup(rt->rt_gateway);
1870                                 if (rt->rt_gwroute == NULL)
1871                                         senderr(EHOSTUNREACH);
1872                         }
1873                 }
1874         }
1875
1876         /*
1877          * Address resolution or Neighbor Unreachability Detection
1878          * for the next hop.
1879          * At this point, the destination of the packet must be a unicast
1880          * or an anycast address(i.e. not a multicast).
1881          */
1882
1883         /* Look up the neighbor cache for the nexthop */
1884         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1885                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1886         else {
1887                 /*
1888                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1889                  * the condition below is not very efficient.  But we believe
1890                  * it is tolerable, because this should be a rare case.
1891                  */
1892                 if (nd6_is_addr_neighbor(dst, ifp) &&
1893                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1894                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1895         }
1896         if (!ln || !rt) {
1897                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1898                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1899                         log(LOG_DEBUG,
1900                             "nd6_output: can't allocate llinfo for %s "
1901                             "(ln=%p, rt=%p)\n",
1902                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1903                         senderr(EIO);   /* XXX: good error? */
1904                 }
1905
1906                 goto sendpkt;   /* send anyway */
1907         }
1908
1909         /* We don't have to do link-layer address resolution on a p2p link. */
1910         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1911             ln->ln_state < ND6_LLINFO_REACHABLE) {
1912                 ln->ln_state = ND6_LLINFO_STALE;
1913                 ln->ln_expire = time_second + nd6_gctimer;
1914         }
1915
1916         /*
1917          * The first time we send a packet to a neighbor whose entry is
1918          * STALE, we have to change the state to DELAY and a sets a timer to
1919          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1920          * neighbor unreachability detection on expiration.
1921          * (RFC 2461 7.3.3)
1922          */
1923         if (ln->ln_state == ND6_LLINFO_STALE) {
1924                 ln->ln_asked = 0;
1925                 ln->ln_state = ND6_LLINFO_DELAY;
1926                 ln->ln_expire = time_second + nd6_delay;
1927         }
1928
1929         /*
1930          * If the neighbor cache entry has a state other than INCOMPLETE
1931          * (i.e. its link-layer address is already resolved), just
1932          * send the packet.
1933          */
1934         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1935                 goto sendpkt;
1936
1937         /*
1938          * There is a neighbor cache entry, but no ethernet address
1939          * response yet.  Replace the held mbuf (if any) with this
1940          * latest one.
1941          *
1942          * This code conforms to the rate-limiting rule described in Section
1943          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1944          * an NS below.
1945          */
1946         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1947                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1948         if (ln->ln_hold)
1949                 m_freem(ln->ln_hold);
1950         ln->ln_hold = m;
1951         if (ln->ln_expire) {
1952                 if (ln->ln_asked < nd6_mmaxtries &&
1953                     ln->ln_expire < time_second) {
1954                         ln->ln_asked++;
1955                         ln->ln_expire = time_second +
1956                                 ND_IFINFO(ifp)->retrans / 1000;
1957                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1958                 }
1959         }
1960         return(0);
1961
1962   sendpkt:
1963
1964         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1965                 return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1966                                          rt));
1967         }
1968         return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1969
1970   bad:
1971         if (m)
1972                 m_freem(m);
1973         return (error);
1974 }
1975 #undef senderr
1976
1977 int
1978 nd6_need_cache(struct ifnet *ifp)
1979 {
1980         /*
1981          * XXX: we currently do not make neighbor cache on any interface
1982          * other than ARCnet, Ethernet, FDDI and GIF.
1983          *
1984          * RFC2893 says:
1985          * - unidirectional tunnels needs no ND
1986          */
1987         switch (ifp->if_type) {
1988         case IFT_ARCNET:
1989         case IFT_ETHER:
1990         case IFT_FDDI:
1991         case IFT_IEEE1394:
1992 #ifdef IFT_L2VLAN
1993         case IFT_L2VLAN:
1994 #endif
1995 #ifdef IFT_IEEE80211
1996         case IFT_IEEE80211:
1997 #endif
1998         case IFT_GIF:           /* XXX need more cases? */
1999                 return(1);
2000         default:
2001                 return(0);
2002         }
2003 }
2004
2005 int
2006 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2007                 struct sockaddr *dst, u_char *desten)
2008 {
2009         struct sockaddr_dl *sdl;
2010         struct rtentry *rt;
2011
2012
2013         if (m->m_flags & M_MCAST) {
2014                 switch (ifp->if_type) {
2015                 case IFT_ETHER:
2016                 case IFT_FDDI:
2017 #ifdef IFT_L2VLAN
2018         case IFT_L2VLAN:
2019 #endif
2020 #ifdef IFT_IEEE80211
2021                 case IFT_IEEE80211:
2022 #endif
2023                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2024                                                  desten);
2025                         return(1);
2026                 case IFT_IEEE1394:
2027                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2028                         return(1);
2029                 case IFT_ARCNET:
2030                         *desten = 0;
2031                         return(1);
2032                 default:
2033                         m_freem(m);
2034                         return(0);
2035                 }
2036         }
2037         if (rt0 == NULL) {
2038                 /* this could happen, if we could not allocate memory */
2039                 m_freem(m);
2040                 return(0);
2041         }
2042         if (rt_llroute(dst, rt0, &rt) != 0) {
2043                 m_freem(m);
2044                 return (0);
2045         }
2046         if (rt->rt_gateway->sa_family != AF_LINK) {
2047                 printf("nd6_storelladdr: something odd happens\n");
2048                 m_freem(m);
2049                 return(0);
2050         }
2051         sdl = SDL(rt->rt_gateway);
2052         if (sdl->sdl_alen == 0) {
2053                 /* this should be impossible, but we bark here for debugging */
2054                 printf("nd6_storelladdr: sdl_alen == 0\n");
2055                 m_freem(m);
2056                 return(0);
2057         }
2058
2059         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2060         return(1);
2061 }
2062
2063 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2064 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2065 #ifdef SYSCTL_DECL
2066 SYSCTL_DECL(_net_inet6_icmp6);
2067 #endif
2068 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2069         CTLFLAG_RD, nd6_sysctl_drlist, "");
2070 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2071         CTLFLAG_RD, nd6_sysctl_prlist, "");
2072
2073 static int
2074 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2075 {
2076         int error;
2077         char buf[1024];
2078         struct in6_defrouter *d, *de;
2079         struct nd_defrouter *dr;
2080
2081         if (req->newptr)
2082                 return EPERM;
2083         error = 0;
2084
2085         for (dr = TAILQ_FIRST(&nd_defrouter);
2086              dr;
2087              dr = TAILQ_NEXT(dr, dr_entry)) {
2088                 d = (struct in6_defrouter *)buf;
2089                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2090
2091                 if (d + 1 <= de) {
2092                         bzero(d, sizeof(*d));
2093                         d->rtaddr.sin6_family = AF_INET6;
2094                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2095                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2096                             dr->ifp) != 0)
2097                                 log(LOG_ERR,
2098                                     "scope error in "
2099                                     "default router list (%s)\n",
2100                                     ip6_sprintf(&dr->rtaddr));
2101                         d->flags = dr->flags;
2102                         d->rtlifetime = dr->rtlifetime;
2103                         d->expire = dr->expire;
2104                         d->if_index = dr->ifp->if_index;
2105                 } else
2106                         panic("buffer too short");
2107
2108                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2109                 if (error)
2110                         break;
2111         }
2112         return error;
2113 }
2114
2115 static int
2116 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2117 {
2118         int error;
2119         char buf[1024];
2120         struct in6_prefix *p, *pe;
2121         struct nd_prefix *pr;
2122
2123         if (req->newptr)
2124                 return EPERM;
2125         error = 0;
2126
2127         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2128                 u_short advrtrs;
2129                 size_t advance;
2130                 struct sockaddr_in6 *sin6, *s6;
2131                 struct nd_pfxrouter *pfr;
2132
2133                 p = (struct in6_prefix *)buf;
2134                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2135
2136                 if (p + 1 <= pe) {
2137                         bzero(p, sizeof(*p));
2138                         sin6 = (struct sockaddr_in6 *)(p + 1);
2139
2140                         p->prefix = pr->ndpr_prefix;
2141                         if (in6_recoverscope(&p->prefix,
2142                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2143                                 log(LOG_ERR,
2144                                     "scope error in prefix list (%s)\n",
2145                                     ip6_sprintf(&p->prefix.sin6_addr));
2146                         p->raflags = pr->ndpr_raf;
2147                         p->prefixlen = pr->ndpr_plen;
2148                         p->vltime = pr->ndpr_vltime;
2149                         p->pltime = pr->ndpr_pltime;
2150                         p->if_index = pr->ndpr_ifp->if_index;
2151                         p->expire = pr->ndpr_expire;
2152                         p->refcnt = pr->ndpr_refcnt;
2153                         p->flags = pr->ndpr_stateflags;
2154                         p->origin = PR_ORIG_RA;
2155                         advrtrs = 0;
2156                         for (pfr = pr->ndpr_advrtrs.lh_first;
2157                              pfr;
2158                              pfr = pfr->pfr_next) {
2159                                 if ((void *)&sin6[advrtrs + 1] >
2160                                     (void *)pe) {
2161                                         advrtrs++;
2162                                         continue;
2163                                 }
2164                                 s6 = &sin6[advrtrs];
2165                                 bzero(s6, sizeof(*s6));
2166                                 s6->sin6_family = AF_INET6;
2167                                 s6->sin6_len = sizeof(*sin6);
2168                                 if (in6_recoverscope(s6,
2169                                     &pfr->router->rtaddr,
2170                                     pfr->router->ifp) != 0)
2171                                         log(LOG_ERR,
2172                                             "scope error in "
2173                                             "prefix list (%s)\n",
2174                                             ip6_sprintf(&pfr->router->rtaddr));
2175                                 advrtrs++;
2176                         }
2177                         p->advrtrs = advrtrs;
2178                 } else
2179                         panic("buffer too short");
2180
2181                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2182                 error = SYSCTL_OUT(req, buf, advance);
2183                 if (error)
2184                         break;
2185         }
2186         return error;
2187 }