sys/netinet6/nd6.c

   1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
   2 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   3
   4 /*
   5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of the project nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 /*
  34  * XXX
  35  * KAME 970409 note:
  36  * BSD/OS version heavily modifies this code, related to llinfo.
  37  * Since we don't have BSD/OS version of net/route.c in our hand,
  38  * I left the code mostly as it was in 970310.  -- itojun
  39  */
  40
  41 #include "opt_inet.h"
  42 #include "opt_inet6.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/callout.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/socket.h>
  50 #include <sys/sockio.h>
  51 #include <sys/time.h>
  52 #include <sys/kernel.h>
  53 #include <sys/protosw.h>
  54 #include <sys/errno.h>
  55 #include <sys/syslog.h>
  56 #include <sys/queue.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/mutex.h>
  59
  60 #include <sys/thread2.h>
  61 #include <sys/mutex2.h>
  62
  63 #include <net/if.h>
  64 #include <net/if_dl.h>
  65 #include <net/if_types.h>
  66 #include <net/route.h>
  67
  68 #include <netinet/in.h>
  69 #include <netinet/if_ether.h>
  70 #include <netinet6/in6_var.h>
  71 #include <netinet/ip6.h>
  72 #include <netinet6/ip6_var.h>
  73 #include <netinet6/nd6.h>
  74 #include <netinet/icmp6.h>
  75
  76 #include <net/net_osdep.h>
  77
  78 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  79 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  80
  81 #define SIN6(s) ((struct sockaddr_in6 *)s)
  82 #define SDL(s) ((struct sockaddr_dl *)s)
  83
  84 /* timer values */
  85 int     nd6_prune       = 1;    /* walk list every 1 seconds */
  86 int     nd6_delay       = 5;    /* delay first probe time 5 second */
  87 int     nd6_umaxtries   = 3;    /* maximum unicast query */
  88 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
  89 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
  90 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
  91
  92 /* preventing too many loops in ND option parsing */
  93 int nd6_maxndopt = 10;  /* max # of ND options allowed */
  94
  95 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
  96
  97 #ifdef ND6_DEBUG
  98 int nd6_debug = 1;
  99 #else
 100 int nd6_debug = 0;
 101 #endif
 102
 103 /* for debugging? */
 104 static int nd6_inuse, nd6_allocated;
 105
 106 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 107 struct nd_drhead nd_defrouter;
 108 struct nd_prhead nd_prefix = { 0 };
 109 struct mtx nd6_mtx = MTX_INITIALIZER;
 110
 111 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 112 static struct sockaddr_in6 all1_sa;
 113
 114 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
 115 static void nd6_slowtimo (void *);
 116 static int regen_tmpaddr (struct in6_ifaddr *);
 117
 118 struct callout nd6_slowtimo_ch;
 119 struct callout nd6_timer_ch;
 120 extern struct callout in6_tmpaddrtimer_ch;
 121
 122 void
 123 nd6_init(void)
 124 {
 125         static int nd6_init_done = 0;
 126         int i;
 127
 128         if (nd6_init_done) {
 129                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 130                 return;
 131         }
 132
 133         all1_sa.sin6_family = AF_INET6;
 134         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 135         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 136                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 137
 138         /* initialization of the default router list */
 139         TAILQ_INIT(&nd_defrouter);
 140
 141         nd6_init_done = 1;
 142
 143         /* start timer */
 144         callout_init(&nd6_slowtimo_ch);
 145         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 146             nd6_slowtimo, NULL);
 147 }
 148
 149 struct nd_ifinfo *
 150 nd6_ifattach(struct ifnet *ifp)
 151 {
 152         struct nd_ifinfo *nd;
 153
 154         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
 155             M_WAITOK | M_ZERO);
 156
 157         nd->initialized = 1;
 158
 159         nd->linkmtu = ifp->if_mtu;
 160         nd->chlim = IPV6_DEFHLIM;
 161         nd->basereachable = REACHABLE_TIME;
 162         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 163         nd->retrans = RETRANS_TIMER;
 164         nd->receivedra = 0;
 165
 166         /*
 167          * Note that the default value of ip6_accept_rtadv is 0, which means
 168          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 169          * here.
 170          */
 171         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 172
 173         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 174         nd6_setmtu0(ifp, nd);
 175         return nd;
 176 }
 177
 178 void
 179 nd6_ifdetach(struct nd_ifinfo *nd)
 180 {
 181         kfree(nd, M_IP6NDP);
 182 }
 183
 184 /*
 185  * Reset ND level link MTU. This function is called when the physical MTU
 186  * changes, which means we might have to adjust the ND level MTU.
 187  */
 188 void
 189 nd6_setmtu(struct ifnet *ifp)
 190 {
 191         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 192 }
 193
 194 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 195 void
 196 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 197 {
 198         u_long oldmaxmtu;
 199         u_long oldlinkmtu;
 200
 201         oldmaxmtu = ndi->maxmtu;
 202         oldlinkmtu = ndi->linkmtu;
 203
 204         switch (ifp->if_type) {
 205         case IFT_ETHER:
 206                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 207                 break;
 208         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 209                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 210                 break;
 211 #ifdef IFT_IEEE80211
 212         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 213                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 214                 break;
 215 #endif
 216         default:
 217                 ndi->maxmtu = ifp->if_mtu;
 218                 break;
 219         }
 220
 221         if (oldmaxmtu != ndi->maxmtu) {
 222                 /*
 223                  * If the ND level MTU is not set yet, or if the maxmtu
 224                  * is reset to a smaller value than the ND level MTU,
 225                  * also reset the ND level MTU.
 226                  */
 227                 if (ndi->linkmtu == 0 ||
 228                     ndi->maxmtu < ndi->linkmtu) {
 229                         ndi->linkmtu = ndi->maxmtu;
 230                         /* also adjust in6_maxmtu if necessary. */
 231                         if (oldlinkmtu == 0) {
 232                                 /*
 233                                  * XXX: the case analysis is grotty, but
 234                                  * it is not efficient to call in6_setmaxmtu()
 235                                  * here when we are during the initialization
 236                                  * procedure.
 237                                  */
 238                                 if (in6_maxmtu < ndi->linkmtu)
 239                                         in6_maxmtu = ndi->linkmtu;
 240                         } else
 241                                 in6_setmaxmtu();
 242                 }
 243         }
 244 #undef MIN
 245 }
 246
 247 void
 248 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 249 {
 250         bzero(ndopts, sizeof(*ndopts));
 251         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 252         ndopts->nd_opts_last
 253                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 254
 255         if (icmp6len == 0) {
 256                 ndopts->nd_opts_done = 1;
 257                 ndopts->nd_opts_search = NULL;
 258         }
 259 }
 260
 261 /*
 262  * Take one ND option.
 263  */
 264 struct nd_opt_hdr *
 265 nd6_option(union nd_opts *ndopts)
 266 {
 267         struct nd_opt_hdr *nd_opt;
 268         int olen;
 269
 270         if (!ndopts)
 271                 panic("ndopts == NULL in nd6_option");
 272         if (!ndopts->nd_opts_last)
 273                 panic("uninitialized ndopts in nd6_option");
 274         if (!ndopts->nd_opts_search)
 275                 return NULL;
 276         if (ndopts->nd_opts_done)
 277                 return NULL;
 278
 279         nd_opt = ndopts->nd_opts_search;
 280
 281         /* make sure nd_opt_len is inside the buffer */
 282         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 283                 bzero(ndopts, sizeof(*ndopts));
 284                 return NULL;
 285         }
 286
 287         olen = nd_opt->nd_opt_len << 3;
 288         if (olen == 0) {
 289                 /*
 290                  * Message validation requires that all included
 291                  * options have a length that is greater than zero.
 292                  */
 293                 bzero(ndopts, sizeof(*ndopts));
 294                 return NULL;
 295         }
 296
 297         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 298         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 299                 /* option overruns the end of buffer, invalid */
 300                 bzero(ndopts, sizeof(*ndopts));
 301                 return NULL;
 302         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 303                 /* reached the end of options chain */
 304                 ndopts->nd_opts_done = 1;
 305                 ndopts->nd_opts_search = NULL;
 306         }
 307         return nd_opt;
 308 }
 309
 310 /*
 311  * Parse multiple ND options.
 312  * This function is much easier to use, for ND routines that do not need
 313  * multiple options of the same type.
 314  */
 315 int
 316 nd6_options(union nd_opts *ndopts)
 317 {
 318         struct nd_opt_hdr *nd_opt;
 319         int i = 0;
 320
 321         if (!ndopts)
 322                 panic("ndopts == NULL in nd6_options");
 323         if (!ndopts->nd_opts_last)
 324                 panic("uninitialized ndopts in nd6_options");
 325         if (!ndopts->nd_opts_search)
 326                 return 0;
 327
 328         while (1) {
 329                 nd_opt = nd6_option(ndopts);
 330                 if (!nd_opt && !ndopts->nd_opts_last) {
 331                         /*
 332                          * Message validation requires that all included
 333                          * options have a length that is greater than zero.
 334                          */
 335                         icmp6stat.icp6s_nd_badopt++;
 336                         bzero(ndopts, sizeof(*ndopts));
 337                         return -1;
 338                 }
 339
 340                 if (!nd_opt)
 341                         goto skip1;
 342
 343                 switch (nd_opt->nd_opt_type) {
 344                 case ND_OPT_SOURCE_LINKADDR:
 345                 case ND_OPT_TARGET_LINKADDR:
 346                 case ND_OPT_MTU:
 347                 case ND_OPT_REDIRECTED_HEADER:
 348                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 349                                 nd6log((LOG_INFO,
 350                                     "duplicated ND6 option found (type=%d)\n",
 351                                     nd_opt->nd_opt_type));
 352                                 /* XXX bark? */
 353                         } else {
 354                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 355                                         = nd_opt;
 356                         }
 357                         break;
 358                 case ND_OPT_PREFIX_INFORMATION:
 359                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 360                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 361                                         = nd_opt;
 362                         }
 363                         ndopts->nd_opts_pi_end =
 364                                 (struct nd_opt_prefix_info *)nd_opt;
 365                         break;
 366                 default:
 367                         /*
 368                          * Unknown options must be silently ignored,
 369                          * to accomodate future extension to the protocol.
 370                          */
 371                         nd6log((LOG_DEBUG,
 372                             "nd6_options: unsupported option %d - "
 373                             "option ignored\n", nd_opt->nd_opt_type));
 374                 }
 375
 376 skip1:
 377                 i++;
 378                 if (i > nd6_maxndopt) {
 379                         icmp6stat.icp6s_nd_toomanyopt++;
 380                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 381                         break;
 382                 }
 383
 384                 if (ndopts->nd_opts_done)
 385                         break;
 386         }
 387
 388         return 0;
 389 }
 390
 391 /*
 392  * ND6 timer routine to expire default route list and prefix list
 393  */
 394 void
 395 nd6_timer(void *ignored_arg)
 396 {
 397         struct llinfo_nd6 *ln;
 398         struct nd_defrouter *dr;
 399         struct nd_prefix *pr;
 400         struct ifnet *ifp;
 401         struct in6_ifaddr *ia6, *nia6;
 402
 403         mtx_lock(&nd6_mtx);
 404         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 405                       nd6_timer, NULL);
 406
 407         ln = llinfo_nd6.ln_next;
 408         while (ln && ln != &llinfo_nd6) {
 409                 struct rtentry *rt;
 410                 struct sockaddr_in6 *dst;
 411                 struct llinfo_nd6 *next = ln->ln_next;
 412                 /* XXX: used for the DELAY case only: */
 413                 struct nd_ifinfo *ndi = NULL;
 414
 415                 if ((rt = ln->ln_rt) == NULL) {
 416                         ln = next;
 417                         continue;
 418                 }
 419                 if ((ifp = rt->rt_ifp) == NULL) {
 420                         ln = next;
 421                         continue;
 422                 }
 423                 ndi = ND_IFINFO(ifp);
 424                 dst = (struct sockaddr_in6 *)rt_key(rt);
 425
 426                 if (ln->ln_expire > time_uptime) {
 427                         ln = next;
 428                         continue;
 429                 }
 430
 431                 /* sanity check */
 432                 if (!rt)
 433                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 434                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 435                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 436                               rt->rt_llinfo, ln);
 437                 if (!dst)
 438                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 439
 440                 switch (ln->ln_state) {
 441                 case ND6_LLINFO_INCOMPLETE:
 442                         if (ln->ln_asked < nd6_mmaxtries) {
 443                                 ln->ln_asked++;
 444                                 ln->ln_expire = time_uptime +
 445                                         ND_IFINFO(ifp)->retrans / 1000;
 446                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 447                                         ln, 0);
 448                         } else {
 449                                 struct mbuf *m = ln->ln_hold;
 450                                 if (m) {
 451                                         if (rt->rt_ifp) {
 452                                                 /*
 453                                                  * Fake rcvif to make ICMP error
 454                                                  * more helpful in diagnosing
 455                                                  * for the receiver.
 456                                                  * XXX: should we consider
 457                                                  * older rcvif?
 458                                                  */
 459                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 460                                         }
 461                                         icmp6_error(m, ICMP6_DST_UNREACH,
 462                                                     ICMP6_DST_UNREACH_ADDR, 0);
 463                                         ln->ln_hold = NULL;
 464                                 }
 465                                 next = nd6_free(rt);
 466                         }
 467                         break;
 468                 case ND6_LLINFO_REACHABLE:
 469                         if (ln->ln_expire) {
 470                                 ln->ln_state = ND6_LLINFO_STALE;
 471                                 ln->ln_expire = time_uptime + nd6_gctimer;
 472                         }
 473                         break;
 474
 475                 case ND6_LLINFO_STALE:
 476                         /* Garbage Collection(RFC 2461 5.3) */
 477                         if (ln->ln_expire)
 478                                 next = nd6_free(rt);
 479                         break;
 480
 481                 case ND6_LLINFO_DELAY:
 482                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 483                                 /* We need NUD */
 484                                 ln->ln_asked = 1;
 485                                 ln->ln_state = ND6_LLINFO_PROBE;
 486                                 ln->ln_expire = time_uptime +
 487                                         ndi->retrans / 1000;
 488                                 nd6_ns_output(ifp, &dst->sin6_addr,
 489                                               &dst->sin6_addr,
 490                                               ln, 0);
 491                         } else {
 492                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 493                                 ln->ln_expire = time_uptime + nd6_gctimer;
 494                         }
 495                         break;
 496                 case ND6_LLINFO_PROBE:
 497                         if (ln->ln_asked < nd6_umaxtries) {
 498                                 ln->ln_asked++;
 499                                 ln->ln_expire = time_uptime +
 500                                         ND_IFINFO(ifp)->retrans / 1000;
 501                                 nd6_ns_output(ifp, &dst->sin6_addr,
 502                                                &dst->sin6_addr, ln, 0);
 503                         } else {
 504                                 next = nd6_free(rt);
 505                         }
 506                         break;
 507                 }
 508                 ln = next;
 509         }
 510
 511         /* expire default router list */
 512         dr = TAILQ_FIRST(&nd_defrouter);
 513         while (dr) {
 514                 if (dr->expire && dr->expire < time_uptime) {
 515                         struct nd_defrouter *t;
 516                         t = TAILQ_NEXT(dr, dr_entry);
 517                         defrtrlist_del(dr);
 518                         dr = t;
 519                 } else {
 520                         dr = TAILQ_NEXT(dr, dr_entry);
 521                 }
 522         }
 523
 524         /*
 525          * expire interface addresses.
 526          * in the past the loop was inside prefix expiry processing.
 527          * However, from a stricter speci-confrmance standpoint, we should
 528          * rather separate address lifetimes and prefix lifetimes.
 529          */
 530 addrloop:
 531         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 532                 nia6 = ia6->ia_next;
 533                 /* check address lifetime */
 534                 if (IFA6_IS_INVALID(ia6)) {
 535                         int regen = 0;
 536
 537                         /*
 538                          * If the expiring address is temporary, try
 539                          * regenerating a new one.  This would be useful when
 540                          * we suspended a laptop PC, then turned it on after a
 541                          * period that could invalidate all temporary
 542                          * addresses.  Although we may have to restart the
 543                          * loop (see below), it must be after purging the
 544                          * address.  Otherwise, we'd see an infinite loop of
 545                          * regeneration.
 546                          */
 547                         if (ip6_use_tempaddr &&
 548                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 549                                 if (regen_tmpaddr(ia6) == 0)
 550                                         regen = 1;
 551                         }
 552
 553                         in6_purgeaddr(&ia6->ia_ifa);
 554
 555                         if (regen)
 556                                 goto addrloop; /* XXX: see below */
 557                 }
 558                 if (IFA6_IS_DEPRECATED(ia6)) {
 559                         int oldflags = ia6->ia6_flags;
 560
 561                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 562
 563                         /*
 564                          * If a temporary address has just become deprecated,
 565                          * regenerate a new one if possible.
 566                          */
 567                         if (ip6_use_tempaddr &&
 568                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 569                             !(oldflags & IN6_IFF_DEPRECATED)) {
 570
 571                                 if (regen_tmpaddr(ia6) == 0) {
 572                                         /*
 573                                          * A new temporary address is
 574                                          * generated.
 575                                          * XXX: this means the address chain
 576                                          * has changed while we are still in
 577                                          * the loop.  Although the change
 578                                          * would not cause disaster (because
 579                                          * it's not a deletion, but an
 580                                          * addition,) we'd rather restart the
 581                                          * loop just for safety.  Or does this
 582                                          * significantly reduce performance??
 583                                          */
 584                                         goto addrloop;
 585                                 }
 586                         }
 587                 } else {
 588                         /*
 589                          * A new RA might have made a deprecated address
 590                          * preferred.
 591                          */
 592                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 593                 }
 594         }
 595
 596         /* expire prefix list */
 597         pr = nd_prefix.lh_first;
 598         while (pr) {
 599                 /*
 600                  * check prefix lifetime.
 601                  * since pltime is just for autoconf, pltime processing for
 602                  * prefix is not necessary.
 603                  */
 604                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
 605                         struct nd_prefix *t;
 606                         t = pr->ndpr_next;
 607
 608                         /*
 609                          * address expiration and prefix expiration are
 610                          * separate.  NEVER perform in6_purgeaddr here.
 611                          */
 612
 613                         prelist_remove(pr);
 614                         pr = t;
 615                 } else
 616                         pr = pr->ndpr_next;
 617         }
 618         mtx_unlock(&nd6_mtx);
 619 }
 620
 621 static int
 622 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 623                                          address */
 624 {
 625         struct ifaddr_container *ifac;
 626         struct ifnet *ifp;
 627         struct in6_ifaddr *public_ifa6 = NULL;
 628
 629         ifp = ia6->ia_ifa.ifa_ifp;
 630         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 631                 struct ifaddr *ifa = ifac->ifa;
 632                 struct in6_ifaddr *it6;
 633
 634                 if (ifa->ifa_addr->sa_family != AF_INET6)
 635                         continue;
 636
 637                 it6 = (struct in6_ifaddr *)ifa;
 638
 639                 /* ignore no autoconf addresses. */
 640                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 641                         continue;
 642
 643                 /* ignore autoconf addresses with different prefixes. */
 644                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 645                         continue;
 646
 647                 /*
 648                  * Now we are looking at an autoconf address with the same
 649                  * prefix as ours.  If the address is temporary and is still
 650                  * preferred, do not create another one.  It would be rare, but
 651                  * could happen, for example, when we resume a laptop PC after
 652                  * a long period.
 653                  */
 654                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 655                     !IFA6_IS_DEPRECATED(it6)) {
 656                         public_ifa6 = NULL;
 657                         break;
 658                 }
 659
 660                 /*
 661                  * This is a public autoconf address that has the same prefix
 662                  * as ours.  If it is preferred, keep it.  We can't break the
 663                  * loop here, because there may be a still-preferred temporary
 664                  * address with the prefix.
 665                  */
 666                 if (!IFA6_IS_DEPRECATED(it6))
 667                     public_ifa6 = it6;
 668         }
 669
 670         if (public_ifa6 != NULL) {
 671                 int e;
 672
 673                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 674                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 675                             " tmp addr,errno=%d\n", e);
 676                         return (-1);
 677                 }
 678                 return (0);
 679         }
 680
 681         return (-1);
 682 }
 683
 684 /*
 685  * Nuke neighbor cache/prefix/default router management table, right before
 686  * ifp goes away.
 687  */
 688 void
 689 nd6_purge(struct ifnet *ifp)
 690 {
 691         struct llinfo_nd6 *ln, *nln;
 692         struct nd_defrouter *dr, *ndr, drany;
 693         struct nd_prefix *pr, *npr;
 694
 695         /* Nuke default router list entries toward ifp */
 696         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 697                 /*
 698                  * The first entry of the list may be stored in
 699                  * the routing table, so we'll delete it later.
 700                  */
 701                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 702                         ndr = TAILQ_NEXT(dr, dr_entry);
 703                         if (dr->ifp == ifp)
 704                                 defrtrlist_del(dr);
 705                 }
 706                 dr = TAILQ_FIRST(&nd_defrouter);
 707                 if (dr->ifp == ifp)
 708                         defrtrlist_del(dr);
 709         }
 710
 711         /* Nuke prefix list entries toward ifp */
 712         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 713                 npr = pr->ndpr_next;
 714                 if (pr->ndpr_ifp == ifp) {
 715                         /*
 716                          * Previously, pr->ndpr_addr is removed as well,
 717                          * but I strongly believe we don't have to do it.
 718                          * nd6_purge() is only called from in6_ifdetach(),
 719                          * which removes all the associated interface addresses
 720                          * by itself.
 721                          * (jinmei@kame.net 20010129)
 722                          */
 723                         prelist_remove(pr);
 724                 }
 725         }
 726
 727         /* cancel default outgoing interface setting */
 728         if (nd6_defifindex == ifp->if_index)
 729                 nd6_setdefaultiface(0);
 730
 731         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 732                 /* refresh default router list */
 733                 bzero(&drany, sizeof(drany));
 734                 defrouter_delreq(&drany, 0);
 735                 defrouter_select();
 736         }
 737
 738         /*
 739          * Nuke neighbor cache entries for the ifp.
 740          * Note that rt->rt_ifp may not be the same as ifp,
 741          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 742          * nd6_rtrequest(), and ip6_input().
 743          */
 744         ln = llinfo_nd6.ln_next;
 745         while (ln && ln != &llinfo_nd6) {
 746                 struct rtentry *rt;
 747                 struct sockaddr_dl *sdl;
 748
 749                 nln = ln->ln_next;
 750                 rt = ln->ln_rt;
 751                 if (rt && rt->rt_gateway &&
 752                     rt->rt_gateway->sa_family == AF_LINK) {
 753                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 754                         if (sdl->sdl_index == ifp->if_index)
 755                                 nln = nd6_free(rt);
 756                 }
 757                 ln = nln;
 758         }
 759 }
 760
 761 struct rtentry *
 762 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 763 {
 764         struct rtentry *rt;
 765         struct sockaddr_in6 sin6;
 766
 767         bzero(&sin6, sizeof(sin6));
 768         sin6.sin6_len = sizeof(struct sockaddr_in6);
 769         sin6.sin6_family = AF_INET6;
 770         sin6.sin6_addr = *addr6;
 771
 772         if (create)
 773                 rt = rtlookup((struct sockaddr *)&sin6);
 774         else
 775                 rt = rtpurelookup((struct sockaddr *)&sin6);
 776         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 777                 /*
 778                  * This is the case for the default route.
 779                  * If we want to create a neighbor cache for the address, we
 780                  * should free the route for the destination and allocate an
 781                  * interface route.
 782                  */
 783                 if (create) {
 784                         --rt->rt_refcnt;
 785                         rt = NULL;
 786                 }
 787         }
 788         if (!rt) {
 789                 if (create && ifp) {
 790                         int e;
 791
 792                         /*
 793                          * If no route is available and create is set,
 794                          * we allocate a host route for the destination
 795                          * and treat it like an interface route.
 796                          * This hack is necessary for a neighbor which can't
 797                          * be covered by our own prefix.
 798                          */
 799                         struct ifaddr *ifa =
 800                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 801                         if (ifa == NULL)
 802                                 return (NULL);
 803
 804                         /*
 805                          * Create a new route.  RTF_LLINFO is necessary
 806                          * to create a Neighbor Cache entry for the
 807                          * destination in nd6_rtrequest which will be
 808                          * called in rtrequest via ifa->ifa_rtrequest.
 809                          */
 810                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 811                                            ifa->ifa_addr,
 812                                            (struct sockaddr *)&all1_sa,
 813                                            (ifa->ifa_flags |
 814                                             RTF_HOST | RTF_LLINFO) &
 815                                            ~RTF_CLONING,
 816                                            &rt)) != 0)
 817                                 log(LOG_ERR,
 818                                     "nd6_lookup: failed to add route for a "
 819                                     "neighbor(%s), errno=%d\n",
 820                                     ip6_sprintf(addr6), e);
 821                         if (rt == NULL)
 822                                 return (NULL);
 823                         if (rt->rt_llinfo) {
 824                                 struct llinfo_nd6 *ln =
 825                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 826                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 827                         }
 828                 } else
 829                         return (NULL);
 830         }
 831         rt->rt_refcnt--;
 832         /*
 833          * Validation for the entry.
 834          * Note that the check for rt_llinfo is necessary because a cloned
 835          * route from a parent route that has the L flag (e.g. the default
 836          * route to a p2p interface) may have the flag, too, while the
 837          * destination is not actually a neighbor.
 838          * XXX: we can't use rt->rt_ifp to check for the interface, since
 839          *      it might be the loopback interface if the entry is for our
 840          *      own address on a non-loopback interface. Instead, we should
 841          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 842          *      interface.
 843          */
 844         if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
 845             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 846             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 847                 if (create) {
 848                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 849                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 850                         /* xxx more logs... kazu */
 851                 }
 852                 return (NULL);
 853         }
 854         return (rt);
 855 }
 856
 857 /*
 858  * Detect if a given IPv6 address identifies a neighbor on a given link.
 859  * XXX: should take care of the destination of a p2p link?
 860  */
 861 int
 862 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 863 {
 864         struct ifaddr_container *ifac;
 865         int i;
 866
 867 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 868 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 869
 870         /*
 871          * A link-local address is always a neighbor.
 872          * XXX: we should use the sin6_scope_id field rather than the embedded
 873          * interface index.
 874          */
 875         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 876             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 877                 return (1);
 878
 879         /*
 880          * If the address matches one of our addresses,
 881          * it should be a neighbor.
 882          */
 883         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 884                 struct ifaddr *ifa = ifac->ifa;
 885
 886                 if (ifa->ifa_addr->sa_family != AF_INET6)
 887                         next: continue;
 888
 889                 for (i = 0; i < 4; i++) {
 890                         if ((IFADDR6(ifa).s6_addr32[i] ^
 891                              addr->sin6_addr.s6_addr32[i]) &
 892                             IFMASK6(ifa).s6_addr32[i])
 893                                 goto next;
 894                 }
 895                 return (1);
 896         }
 897
 898         /*
 899          * Even if the address matches none of our addresses, it might be
 900          * in the neighbor cache.
 901          */
 902         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 903                 return (1);
 904
 905         return (0);
 906 #undef IFADDR6
 907 #undef IFMASK6
 908 }
 909
 910 /*
 911  * Free an nd6 llinfo entry.
 912  */
 913 struct llinfo_nd6 *
 914 nd6_free(struct rtentry *rt)
 915 {
 916         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 917         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 918         struct nd_defrouter *dr;
 919
 920         /*
 921          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
 922          * even though it is not harmful, it was not really necessary.
 923          */
 924
 925         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 926                 mtx_lock(&nd6_mtx);
 927                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 928                                       rt->rt_ifp);
 929
 930                 if (ln->ln_router || dr) {
 931                         /*
 932                          * rt6_flush must be called whether or not the neighbor
 933                          * is in the Default Router List.
 934                          * See a corresponding comment in nd6_na_input().
 935                          */
 936                         rt6_flush(&in6, rt->rt_ifp);
 937                 }
 938
 939                 if (dr) {
 940                         /*
 941                          * Unreachablity of a router might affect the default
 942                          * router selection and on-link detection of advertised
 943                          * prefixes.
 944                          */
 945
 946                         /*
 947                          * Temporarily fake the state to choose a new default
 948                          * router and to perform on-link determination of
 949                          * prefixes correctly.
 950                          * Below the state will be set correctly,
 951                          * or the entry itself will be deleted.
 952                          */
 953                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 954
 955                         /*
 956                          * Since defrouter_select() does not affect the
 957                          * on-link determination and MIP6 needs the check
 958                          * before the default router selection, we perform
 959                          * the check now.
 960                          */
 961                         pfxlist_onlink_check();
 962
 963                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 964                                 /*
 965                                  * It is used as the current default router,
 966                                  * so we have to move it to the end of the
 967                                  * list and choose a new one.
 968                                  * XXX: it is not very efficient if this is
 969                                  *      the only router.
 970                                  */
 971                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
 972                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
 973
 974                                 defrouter_select();
 975                         }
 976                 }
 977                 mtx_unlock(&nd6_mtx);
 978         }
 979
 980         /*
 981          * Before deleting the entry, remember the next entry as the
 982          * return value.  We need this because pfxlist_onlink_check() above
 983          * might have freed other entries (particularly the old next entry) as
 984          * a side effect (XXX).
 985          */
 986         next = ln->ln_next;
 987
 988         /*
 989          * Detach the route from the routing tree and the list of neighbor
 990          * caches, and disable the route entry not to be used in already
 991          * cached routes.
 992          */
 993         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
 994
 995         return (next);
 996 }
 997
 998 /*
 999  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1000  *
1001  * XXX cost-effective metods?
1002  */
1003 void
1004 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1005 {
1006         struct llinfo_nd6 *ln;
1007
1008         /*
1009          * If the caller specified "rt", use that.  Otherwise, resolve the
1010          * routing table by supplied "dst6".
1011          */
1012         if (!rt) {
1013                 if (!dst6)
1014                         return;
1015                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1016                         return;
1017         }
1018
1019         if ((rt->rt_flags & RTF_GATEWAY) ||
1020             !(rt->rt_flags & RTF_LLINFO) ||
1021             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1022             rt->rt_gateway->sa_family != AF_LINK) {
1023                 /* This is not a host route. */
1024                 return;
1025         }
1026
1027         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1028         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1029                 return;
1030
1031         /*
1032          * if we get upper-layer reachability confirmation many times,
1033          * it is possible we have false information.
1034          */
1035         if (!force) {
1036                 ln->ln_byhint++;
1037                 if (ln->ln_byhint > nd6_maxnudhint)
1038                         return;
1039         }
1040
1041         ln->ln_state = ND6_LLINFO_REACHABLE;
1042         if (ln->ln_expire)
1043                 ln->ln_expire = time_uptime +
1044                         ND_IFINFO(rt->rt_ifp)->reachable;
1045 }
1046
1047 void
1048 nd6_rtrequest(int req, struct rtentry *rt)
1049 {
1050         struct sockaddr *gate = rt->rt_gateway;
1051         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1052         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1053         struct ifnet *ifp = rt->rt_ifp;
1054         struct ifaddr *ifa;
1055
1056         if ((rt->rt_flags & RTF_GATEWAY))
1057                 return;
1058
1059         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1060                 /*
1061                  * This is probably an interface direct route for a link
1062                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1063                  * We do not need special treatment below for such a route.
1064                  * Moreover, the RTF_LLINFO flag which would be set below
1065                  * would annoy the ndp(8) command.
1066                  */
1067                 return;
1068         }
1069
1070         if (req == RTM_RESOLVE &&
1071             (nd6_need_cache(ifp) == 0 || /* stf case */
1072              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1073                 /*
1074                  * FreeBSD and BSD/OS often make a cloned host route based
1075                  * on a less-specific route (e.g. the default route).
1076                  * If the less specific route does not have a "gateway"
1077                  * (this is the case when the route just goes to a p2p or an
1078                  * stf interface), we'll mistakenly make a neighbor cache for
1079                  * the host route, and will see strange neighbor solicitation
1080                  * for the corresponding destination.  In order to avoid the
1081                  * confusion, we check if the destination of the route is
1082                  * a neighbor in terms of neighbor discovery, and stop the
1083                  * process if not.  Additionally, we remove the LLINFO flag
1084                  * so that ndp(8) will not try to get the neighbor information
1085                  * of the destination.
1086                  */
1087                 rt->rt_flags &= ~RTF_LLINFO;
1088                 return;
1089         }
1090
1091         switch (req) {
1092         case RTM_ADD:
1093                 /*
1094                  * There is no backward compatibility :)
1095                  *
1096                  * if (!(rt->rt_flags & RTF_HOST) &&
1097                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1098                  *         rt->rt_flags |= RTF_CLONING;
1099                  */
1100                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1101                         /*
1102                          * Case 1: This route should come from
1103                          * a route to interface.  RTF_LLINFO flag is set
1104                          * for a host route whose destination should be
1105                          * treated as on-link.
1106                          */
1107                         rt_setgate(rt, rt_key(rt),
1108                                    (struct sockaddr *)&null_sdl,
1109                                    RTL_DONTREPORT);
1110                         gate = rt->rt_gateway;
1111                         SDL(gate)->sdl_type = ifp->if_type;
1112                         SDL(gate)->sdl_index = ifp->if_index;
1113                         if (ln)
1114                                 ln->ln_expire = time_uptime;
1115 #if 1
1116                         if (ln && ln->ln_expire == 0) {
1117                                 /* kludge for desktops */
1118 #if 0
1119                                 kprintf("nd6_rtequest: time.tv_sec is zero; "
1120                                        "treat it as 1\n");
1121 #endif
1122                                 ln->ln_expire = 1;
1123                         }
1124 #endif
1125                         if ((rt->rt_flags & RTF_CLONING))
1126                                 break;
1127                 }
1128                 /*
1129                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1130                  * We don't do that here since llinfo is not ready yet.
1131                  *
1132                  * There are also couple of other things to be discussed:
1133                  * - unsolicited NA code needs improvement beforehand
1134                  * - RFC2461 says we MAY send multicast unsolicited NA
1135                  *   (7.2.6 paragraph 4), however, it also says that we
1136                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1137                  *   we don't have anything like it right now.
1138                  *   note that the mechanism needs a mutual agreement
1139                  *   between proxies, which means that we need to implement
1140                  *   a new protocol, or a new kludge.
1141                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1142                  *   we need to check ip6forwarding before sending it.
1143                  *   (or should we allow proxy ND configuration only for
1144                  *   routers?  there's no mention about proxy ND from hosts)
1145                  */
1146 #if 0
1147                 /* XXX it does not work */
1148                 if (rt->rt_flags & RTF_ANNOUNCE)
1149                         nd6_na_output(ifp,
1150                               &SIN6(rt_key(rt))->sin6_addr,
1151                               &SIN6(rt_key(rt))->sin6_addr,
1152                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1153                               1, NULL);
1154 #endif
1155                 /* FALLTHROUGH */
1156         case RTM_RESOLVE:
1157                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1158                         /*
1159                          * Address resolution isn't necessary for a point to
1160                          * point link, so we can skip this test for a p2p link.
1161                          */
1162                         if (gate->sa_family != AF_LINK ||
1163                             gate->sa_len < sizeof(null_sdl)) {
1164                                 log(LOG_DEBUG,
1165                                     "nd6_rtrequest: bad gateway value: %s\n",
1166                                     if_name(ifp));
1167                                 break;
1168                         }
1169                         SDL(gate)->sdl_type = ifp->if_type;
1170                         SDL(gate)->sdl_index = ifp->if_index;
1171                 }
1172                 if (ln != NULL)
1173                         break;  /* This happens on a route change */
1174                 /*
1175                  * Case 2: This route may come from cloning, or a manual route
1176                  * add with a LL address.
1177                  */
1178                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1179                 rt->rt_llinfo = (caddr_t)ln;
1180                 if (!ln) {
1181                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1182                         break;
1183                 }
1184                 nd6_inuse++;
1185                 nd6_allocated++;
1186                 bzero(ln, sizeof(*ln));
1187                 ln->ln_rt = rt;
1188                 /* this is required for "ndp" command. - shin */
1189                 if (req == RTM_ADD) {
1190                         /*
1191                          * gate should have some valid AF_LINK entry,
1192                          * and ln->ln_expire should have some lifetime
1193                          * which is specified by ndp command.
1194                          */
1195                         ln->ln_state = ND6_LLINFO_REACHABLE;
1196                         ln->ln_byhint = 0;
1197                 } else {
1198                         /*
1199                          * When req == RTM_RESOLVE, rt is created and
1200                          * initialized in rtrequest(), so rt_expire is 0.
1201                          */
1202                         ln->ln_state = ND6_LLINFO_NOSTATE;
1203                         ln->ln_expire = time_uptime;
1204                 }
1205                 rt->rt_flags |= RTF_LLINFO;
1206                 ln->ln_next = llinfo_nd6.ln_next;
1207                 llinfo_nd6.ln_next = ln;
1208                 ln->ln_prev = &llinfo_nd6;
1209                 ln->ln_next->ln_prev = ln;
1210
1211                 /*
1212                  * check if rt_key(rt) is one of my address assigned
1213                  * to the interface.
1214                  */
1215                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1216                                           &SIN6(rt_key(rt))->sin6_addr);
1217                 if (ifa) {
1218                         caddr_t macp = nd6_ifptomac(ifp);
1219                         ln->ln_expire = 0;
1220                         ln->ln_state = ND6_LLINFO_REACHABLE;
1221                         ln->ln_byhint = 0;
1222                         if (macp) {
1223                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1224                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1225                         }
1226                         if (nd6_useloopback) {
1227                                 rt->rt_ifp = &loif[0];  /* XXX */
1228                                 /*
1229                                  * Make sure rt_ifa be equal to the ifaddr
1230                                  * corresponding to the address.
1231                                  * We need this because when we refer
1232                                  * rt_ifa->ia6_flags in ip6_input, we assume
1233                                  * that the rt_ifa points to the address instead
1234                                  * of the loopback address.
1235                                  */
1236                                 if (ifa != rt->rt_ifa) {
1237                                         IFAFREE(rt->rt_ifa);
1238                                         IFAREF(ifa);
1239                                         rt->rt_ifa = ifa;
1240                                 }
1241                         }
1242                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1243                         ln->ln_expire = 0;
1244                         ln->ln_state = ND6_LLINFO_REACHABLE;
1245                         ln->ln_byhint = 0;
1246
1247                         /* join solicited node multicast for proxy ND */
1248                         if (ifp->if_flags & IFF_MULTICAST) {
1249                                 struct in6_addr llsol;
1250                                 int error;
1251
1252                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1253                                 llsol.s6_addr16[0] = htons(0xff02);
1254                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1255                                 llsol.s6_addr32[1] = 0;
1256                                 llsol.s6_addr32[2] = htonl(1);
1257                                 llsol.s6_addr8[12] = 0xff;
1258
1259                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1260                                         nd6log((LOG_ERR, "%s: failed to join "
1261                                             "%s (errno=%d)\n", if_name(ifp),
1262                                             ip6_sprintf(&llsol), error));
1263                                 }
1264                         }
1265                 }
1266                 break;
1267
1268         case RTM_DELETE:
1269                 if (!ln)
1270                         break;
1271                 /* leave from solicited node multicast for proxy ND */
1272                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1273                     (ifp->if_flags & IFF_MULTICAST)) {
1274                         struct in6_addr llsol;
1275                         struct in6_multi *in6m;
1276
1277                         llsol = SIN6(rt_key(rt))->sin6_addr;
1278                         llsol.s6_addr16[0] = htons(0xff02);
1279                         llsol.s6_addr16[1] = htons(ifp->if_index);
1280                         llsol.s6_addr32[1] = 0;
1281                         llsol.s6_addr32[2] = htonl(1);
1282                         llsol.s6_addr8[12] = 0xff;
1283
1284                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1285                         if (in6m)
1286                                 in6_delmulti(in6m);
1287                 }
1288                 nd6_inuse--;
1289                 ln->ln_next->ln_prev = ln->ln_prev;
1290                 ln->ln_prev->ln_next = ln->ln_next;
1291                 ln->ln_prev = NULL;
1292                 rt->rt_llinfo = 0;
1293                 rt->rt_flags &= ~RTF_LLINFO;
1294                 if (ln->ln_hold)
1295                         m_freem(ln->ln_hold);
1296                 Free((caddr_t)ln);
1297         }
1298 }
1299
1300 int
1301 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1302 {
1303         struct in6_drlist *drl = (struct in6_drlist *)data;
1304         struct in6_prlist *prl = (struct in6_prlist *)data;
1305         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1306         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1307         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1308         struct nd_defrouter *dr, any;
1309         struct nd_prefix *pr;
1310         struct rtentry *rt;
1311         int i = 0, error = 0;
1312
1313         switch (cmd) {
1314         case SIOCGDRLST_IN6:
1315                 /*
1316                  * obsolete API, use sysctl under net.inet6.icmp6
1317                  */
1318                 bzero(drl, sizeof(*drl));
1319                 mtx_lock(&nd6_mtx);
1320                 dr = TAILQ_FIRST(&nd_defrouter);
1321                 while (dr && i < DRLSTSIZ) {
1322                         drl->defrouter[i].rtaddr = dr->rtaddr;
1323                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1324                                 /* XXX: need to this hack for KAME stack */
1325                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1326                         } else
1327                                 log(LOG_ERR,
1328                                     "default router list contains a "
1329                                     "non-linklocal address(%s)\n",
1330                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1331
1332                         drl->defrouter[i].flags = dr->flags;
1333                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1334                         drl->defrouter[i].expire = dr->expire;
1335                         drl->defrouter[i].if_index = dr->ifp->if_index;
1336                         i++;
1337                         dr = TAILQ_NEXT(dr, dr_entry);
1338                 }
1339                 mtx_unlock(&nd6_mtx);
1340                 break;
1341         case SIOCGPRLST_IN6:
1342                 /*
1343                  * obsolete API, use sysctl under net.inet6.icmp6
1344                  */
1345                 /*
1346                  * XXX meaning of fields, especialy "raflags", is very
1347                  * differnet between RA prefix list and RR/static prefix list.
1348                  * how about separating ioctls into two?
1349                  */
1350                 bzero(prl, sizeof(*prl));
1351                 mtx_lock(&nd6_mtx);
1352                 pr = nd_prefix.lh_first;
1353                 while (pr && i < PRLSTSIZ) {
1354                         struct nd_pfxrouter *pfr;
1355                         int j;
1356
1357                         in6_embedscope(&prl->prefix[i].prefix,
1358                             &pr->ndpr_prefix, NULL, NULL);
1359                         prl->prefix[i].raflags = pr->ndpr_raf;
1360                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1361                         prl->prefix[i].vltime = pr->ndpr_vltime;
1362                         prl->prefix[i].pltime = pr->ndpr_pltime;
1363                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1364                         prl->prefix[i].expire = pr->ndpr_expire;
1365
1366                         pfr = pr->ndpr_advrtrs.lh_first;
1367                         j = 0;
1368                         while (pfr) {
1369                                 if (j < DRLSTSIZ) {
1370 #define RTRADDR prl->prefix[i].advrtr[j]
1371                                         RTRADDR = pfr->router->rtaddr;
1372                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1373                                                 /* XXX: hack for KAME */
1374                                                 RTRADDR.s6_addr16[1] = 0;
1375                                         } else
1376                                                 log(LOG_ERR,
1377                                                     "a router(%s) advertises "
1378                                                     "a prefix with "
1379                                                     "non-link local address\n",
1380                                                     ip6_sprintf(&RTRADDR));
1381 #undef RTRADDR
1382                                 }
1383                                 j++;
1384                                 pfr = pfr->pfr_next;
1385                         }
1386                         prl->prefix[i].advrtrs = j;
1387                         prl->prefix[i].origin = PR_ORIG_RA;
1388
1389                         i++;
1390                         pr = pr->ndpr_next;
1391                 }
1392                 mtx_unlock(&nd6_mtx);
1393
1394                 break;
1395         case OSIOCGIFINFO_IN6:
1396                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1397                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1398                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1399                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1400                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1401                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1402                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1403                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1404                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1405                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1406                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1407                 break;
1408         case SIOCGIFINFO_IN6:
1409                 ndi->ndi = *ND_IFINFO(ifp);
1410                 break;
1411         case SIOCSIFINFO_FLAGS:
1412                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1413                 break;
1414         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1415                 /* flush default router list */
1416                 /*
1417                  * xxx sumikawa: should not delete route if default
1418                  * route equals to the top of default router list
1419                  */
1420                 bzero(&any, sizeof(any));
1421                 defrouter_delreq(&any, 0);
1422                 defrouter_select();
1423                 /* xxx sumikawa: flush prefix list */
1424                 break;
1425         case SIOCSPFXFLUSH_IN6:
1426             {
1427                 /* flush all the prefix advertised by routers */
1428                 struct nd_prefix *pr, *next;
1429
1430                 mtx_lock(&nd6_mtx);
1431                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1432                         struct in6_ifaddr *ia, *ia_next;
1433
1434                         next = pr->ndpr_next;
1435
1436                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1437                                 continue; /* XXX */
1438
1439                         /* do we really have to remove addresses as well? */
1440                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1441                                 /* ia might be removed.  keep the next ptr. */
1442                                 ia_next = ia->ia_next;
1443
1444                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1445                                         continue;
1446
1447                                 if (ia->ia6_ndpr == pr)
1448                                         in6_purgeaddr(&ia->ia_ifa);
1449                         }
1450                         prelist_remove(pr);
1451                 }
1452                 mtx_unlock(&nd6_mtx);
1453                 break;
1454             }
1455         case SIOCSRTRFLUSH_IN6:
1456             {
1457                 /* flush all the default routers */
1458                 struct nd_defrouter *dr, *next;
1459
1460                 mtx_lock(&nd6_mtx);
1461                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1462                         /*
1463                          * The first entry of the list may be stored in
1464                          * the routing table, so we'll delete it later.
1465                          */
1466                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1467                                 next = TAILQ_NEXT(dr, dr_entry);
1468                                 defrtrlist_del(dr);
1469                         }
1470                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1471                 }
1472                 mtx_unlock(&nd6_mtx);
1473                 break;
1474             }
1475         case SIOCGNBRINFO_IN6:
1476             {
1477                 struct llinfo_nd6 *ln;
1478                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1479
1480                 /*
1481                  * XXX: KAME specific hack for scoped addresses
1482                  *      XXXX: for other scopes than link-local?
1483                  */
1484                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1485                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1486                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1487
1488                         if (*idp == 0)
1489                                 *idp = htons(ifp->if_index);
1490                 }
1491
1492                 mtx_lock(&nd6_mtx);
1493                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1494                         error = EINVAL;
1495                         mtx_unlock(&nd6_mtx);
1496                         break;
1497                 }
1498                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1499                 nbi->state = ln->ln_state;
1500                 nbi->asked = ln->ln_asked;
1501                 nbi->isrouter = ln->ln_router;
1502                 nbi->expire = ln->ln_expire;
1503                 mtx_unlock(&nd6_mtx);
1504
1505                 break;
1506             }
1507         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1508                 ndif->ifindex = nd6_defifindex;
1509                 break;
1510         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1511                 return (nd6_setdefaultiface(ndif->ifindex));
1512                 break;
1513         }
1514         return (error);
1515 }
1516
1517 /*
1518  * Create neighbor cache entry and cache link-layer address,
1519  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1520  */
1521 struct rtentry *
1522 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1523                  int lladdrlen,
1524                  int type,      /* ICMP6 type */
1525                  int code       /* type dependent information */)
1526 {
1527         struct rtentry *rt = NULL;
1528         struct llinfo_nd6 *ln = NULL;
1529         int is_newentry;
1530         struct sockaddr_dl *sdl = NULL;
1531         int do_update;
1532         int olladdr;
1533         int llchange;
1534         int newstate = 0;
1535
1536         if (!ifp)
1537                 panic("ifp == NULL in nd6_cache_lladdr");
1538         if (!from)
1539                 panic("from == NULL in nd6_cache_lladdr");
1540
1541         /* nothing must be updated for unspecified address */
1542         if (IN6_IS_ADDR_UNSPECIFIED(from))
1543                 return NULL;
1544
1545         /*
1546          * Validation about ifp->if_addrlen and lladdrlen must be done in
1547          * the caller.
1548          *
1549          * XXX If the link does not have link-layer adderss, what should
1550          * we do? (ifp->if_addrlen == 0)
1551          * Spec says nothing in sections for RA, RS and NA.  There's small
1552          * description on it in NS section (RFC 2461 7.2.3).
1553          */
1554
1555         rt = nd6_lookup(from, 0, ifp);
1556         if (!rt) {
1557 #if 0
1558                 /* nothing must be done if there's no lladdr */
1559                 if (!lladdr || !lladdrlen)
1560                         return NULL;
1561 #endif
1562
1563                 rt = nd6_lookup(from, 1, ifp);
1564                 is_newentry = 1;
1565         } else {
1566                 /* do nothing if static ndp is set */
1567                 if (rt->rt_flags & RTF_STATIC)
1568                         return NULL;
1569                 is_newentry = 0;
1570         }
1571
1572         if (!rt)
1573                 return NULL;
1574         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1575 fail:
1576                 nd6_free(rt);
1577                 return NULL;
1578         }
1579         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1580         if (!ln)
1581                 goto fail;
1582         if (!rt->rt_gateway)
1583                 goto fail;
1584         if (rt->rt_gateway->sa_family != AF_LINK)
1585                 goto fail;
1586         sdl = SDL(rt->rt_gateway);
1587
1588         olladdr = (sdl->sdl_alen) ? 1 : 0;
1589         if (olladdr && lladdr) {
1590                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1591                         llchange = 1;
1592                 else
1593                         llchange = 0;
1594         } else
1595                 llchange = 0;
1596
1597         /*
1598          * newentry olladdr  lladdr  llchange   (*=record)
1599          *      0       n       n       --      (1)
1600          *      0       y       n       --      (2)
1601          *      0       n       y       --      (3) * STALE
1602          *      0       y       y       n       (4) *
1603          *      0       y       y       y       (5) * STALE
1604          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1605          *      1       --      y       --      (7) * STALE
1606          */
1607
1608         if (lladdr) {           /* (3-5) and (7) */
1609                 /*
1610                  * Record source link-layer address
1611                  * XXX is it dependent to ifp->if_type?
1612                  */
1613                 sdl->sdl_alen = ifp->if_addrlen;
1614                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1615         }
1616
1617         if (!is_newentry) {
1618                 if ((!olladdr && lladdr)                /* (3) */
1619                  || (olladdr && lladdr && llchange)) {  /* (5) */
1620                         do_update = 1;
1621                         newstate = ND6_LLINFO_STALE;
1622                 } else                                  /* (1-2,4) */
1623                         do_update = 0;
1624         } else {
1625                 do_update = 1;
1626                 if (!lladdr)                            /* (6) */
1627                         newstate = ND6_LLINFO_NOSTATE;
1628                 else                                    /* (7) */
1629                         newstate = ND6_LLINFO_STALE;
1630         }
1631
1632         if (do_update) {
1633                 /*
1634                  * Update the state of the neighbor cache.
1635                  */
1636                 ln->ln_state = newstate;
1637
1638                 if (ln->ln_state == ND6_LLINFO_STALE) {
1639                         /*
1640                          * XXX: since nd6_output() below will cause
1641                          * state tansition to DELAY and reset the timer,
1642                          * we must set the timer now, although it is actually
1643                          * meaningless.
1644                          */
1645                         ln->ln_expire = time_uptime + nd6_gctimer;
1646
1647                         if (ln->ln_hold) {
1648                                 /*
1649                                  * we assume ifp is not a p2p here, so just
1650                                  * set the 2nd argument as the 1st one.
1651                                  */
1652                                 nd6_output(ifp, ifp, ln->ln_hold,
1653                                            (struct sockaddr_in6 *)rt_key(rt),
1654                                            rt);
1655                                 ln->ln_hold = NULL;
1656                         }
1657                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1658                         /* probe right away */
1659                         ln->ln_expire = time_uptime;
1660                 }
1661         }
1662
1663         /*
1664          * ICMP6 type dependent behavior.
1665          *
1666          * NS: clear IsRouter if new entry
1667          * RS: clear IsRouter
1668          * RA: set IsRouter if there's lladdr
1669          * redir: clear IsRouter if new entry
1670          *
1671          * RA case, (1):
1672          * The spec says that we must set IsRouter in the following cases:
1673          * - If lladdr exist, set IsRouter.  This means (1-5).
1674          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1675          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1676          * A quetion arises for (1) case.  (1) case has no lladdr in the
1677          * neighbor cache, this is similar to (6).
1678          * This case is rare but we figured that we MUST NOT set IsRouter.
1679          *
1680          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1681          *                                                      D R
1682          *      0       n       n       --      (1)     c   ?     s
1683          *      0       y       n       --      (2)     c   s     s
1684          *      0       n       y       --      (3)     c   s     s
1685          *      0       y       y       n       (4)     c   s     s
1686          *      0       y       y       y       (5)     c   s     s
1687          *      1       --      n       --      (6) c   c       c s
1688          *      1       --      y       --      (7) c   c   s   c s
1689          *
1690          *                                      (c=clear s=set)
1691          */
1692         switch (type & 0xff) {
1693         case ND_NEIGHBOR_SOLICIT:
1694                 /*
1695                  * New entry must have is_router flag cleared.
1696                  */
1697                 if (is_newentry)        /* (6-7) */
1698                         ln->ln_router = 0;
1699                 break;
1700         case ND_REDIRECT:
1701                 /*
1702                  * If the icmp is a redirect to a better router, always set the
1703                  * is_router flag. Otherwise, if the entry is newly created,
1704                  * clear the flag. [RFC 2461, sec 8.3]
1705                  */
1706                 if (code == ND_REDIRECT_ROUTER)
1707                         ln->ln_router = 1;
1708                 else if (is_newentry) /* (6-7) */
1709                         ln->ln_router = 0;
1710                 break;
1711         case ND_ROUTER_SOLICIT:
1712                 /*
1713                  * is_router flag must always be cleared.
1714                  */
1715                 ln->ln_router = 0;
1716                 break;
1717         case ND_ROUTER_ADVERT:
1718                 /*
1719                  * Mark an entry with lladdr as a router.
1720                  */
1721                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1722                  || (is_newentry && lladdr)) {                  /* (7) */
1723                         ln->ln_router = 1;
1724                 }
1725                 break;
1726         }
1727
1728         /*
1729          * When the link-layer address of a router changes, select the
1730          * best router again.  In particular, when the neighbor entry is newly
1731          * created, it might affect the selection policy.
1732          * Question: can we restrict the first condition to the "is_newentry"
1733          * case?
1734          * XXX: when we hear an RA from a new router with the link-layer
1735          * address option, defrouter_select() is called twice, since
1736          * defrtrlist_update called the function as well.  However, I believe
1737          * we can compromise the overhead, since it only happens the first
1738          * time.
1739          * XXX: although defrouter_select() should not have a bad effect
1740          * for those are not autoconfigured hosts, we explicitly avoid such
1741          * cases for safety.
1742          */
1743         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1744                 defrouter_select();
1745
1746         return rt;
1747 }
1748
1749 static void
1750 nd6_slowtimo(void *ignored_arg)
1751 {
1752         struct nd_ifinfo *nd6if;
1753         struct ifnet *ifp;
1754
1755         mtx_lock(&nd6_mtx);
1756         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1757                         nd6_slowtimo, NULL);
1758         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1759                 if (ifp->if_afdata[AF_INET6] == NULL)
1760                         continue;
1761                 nd6if = ND_IFINFO(ifp);
1762                 if (nd6if->basereachable && /* already initialized */
1763                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1764                         /*
1765                          * Since reachable time rarely changes by router
1766                          * advertisements, we SHOULD insure that a new random
1767                          * value gets recomputed at least once every few hours.
1768                          * (RFC 2461, 6.3.4)
1769                          */
1770                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1771                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1772                 }
1773         }
1774         mtx_unlock(&nd6_mtx);
1775 }
1776
1777 #define gotoerr(e) { error = (e); goto bad;}
1778
1779 int
1780 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1781            struct sockaddr_in6 *dst, struct rtentry *rt)
1782 {
1783         struct llinfo_nd6 *ln = NULL;
1784         int error = 0;
1785
1786         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1787                 goto sendpkt;
1788
1789         if (nd6_need_cache(ifp) == 0)
1790                 goto sendpkt;
1791
1792         /*
1793          * next hop determination.  This routine is derived from ether_outpout.
1794          */
1795         if (rt != NULL) {
1796                 if (!(rt->rt_flags & RTF_UP)) {
1797                         rt = rtlookup((struct sockaddr *)dst);
1798                         if (rt == NULL)
1799                                 gotoerr(EHOSTUNREACH);
1800                         rt->rt_refcnt--;
1801                         if (rt->rt_ifp != ifp) {
1802                                 /* XXX: loop care? */
1803                                 return nd6_output(ifp, origifp, m, dst, rt);
1804                         }
1805                 }
1806                 if (rt->rt_flags & RTF_GATEWAY) {
1807                         struct sockaddr_in6 *gw6;
1808
1809                         /*
1810                          * We skip link-layer address resolution and NUD
1811                          * if the gateway is not a neighbor from ND point
1812                          * of view, regardless of the value of nd_ifinfo.flags.
1813                          * The second condition is a bit tricky; we skip
1814                          * if the gateway is our own address, which is
1815                          * sometimes used to install a route to a p2p link.
1816                          */
1817                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1818                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1819                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1820                                 /*
1821                                  * We allow this kind of tricky route only
1822                                  * when the outgoing interface is p2p.
1823                                  * XXX: we may need a more generic rule here.
1824                                  */
1825                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1826                                         gotoerr(EHOSTUNREACH);
1827
1828                                 goto sendpkt;
1829                         }
1830
1831                         if (rt->rt_gwroute == NULL) {
1832                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1833                                 if (rt->rt_gwroute == NULL)
1834                                         gotoerr(EHOSTUNREACH);
1835                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1836                                 rtfree(rt->rt_gwroute);
1837                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1838                                 if (rt->rt_gwroute == NULL)
1839                                         gotoerr(EHOSTUNREACH);
1840                         }
1841                 }
1842         }
1843
1844         /*
1845          * Address resolution or Neighbor Unreachability Detection
1846          * for the next hop.
1847          * At this point, the destination of the packet must be a unicast
1848          * or an anycast address(i.e. not a multicast).
1849          */
1850
1851         /* Look up the neighbor cache for the nexthop */
1852         if (rt && (rt->rt_flags & RTF_LLINFO))
1853                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1854         else {
1855                 /*
1856                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1857                  * the condition below is not very efficient.  But we believe
1858                  * it is tolerable, because this should be a rare case.
1859                  */
1860                 if (nd6_is_addr_neighbor(dst, ifp) &&
1861                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1862                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1863         }
1864         if (!ln || !rt) {
1865                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1866                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1867                         log(LOG_DEBUG,
1868                             "nd6_output: can't allocate llinfo for %s "
1869                             "(ln=%p, rt=%p)\n",
1870                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1871                         gotoerr(EIO);   /* XXX: good error? */
1872                 }
1873
1874                 goto sendpkt;   /* send anyway */
1875         }
1876
1877         /* We don't have to do link-layer address resolution on a p2p link. */
1878         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1879             ln->ln_state < ND6_LLINFO_REACHABLE) {
1880                 ln->ln_state = ND6_LLINFO_STALE;
1881                 ln->ln_expire = time_uptime + nd6_gctimer;
1882         }
1883
1884         /*
1885          * The first time we send a packet to a neighbor whose entry is
1886          * STALE, we have to change the state to DELAY and a sets a timer to
1887          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1888          * neighbor unreachability detection on expiration.
1889          * (RFC 2461 7.3.3)
1890          */
1891         if (ln->ln_state == ND6_LLINFO_STALE) {
1892                 ln->ln_asked = 0;
1893                 ln->ln_state = ND6_LLINFO_DELAY;
1894                 ln->ln_expire = time_uptime + nd6_delay;
1895         }
1896
1897         /*
1898          * If the neighbor cache entry has a state other than INCOMPLETE
1899          * (i.e. its link-layer address is already resolved), just
1900          * send the packet.
1901          */
1902         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1903                 goto sendpkt;
1904
1905         /*
1906          * There is a neighbor cache entry, but no ethernet address
1907          * response yet.  Replace the held mbuf (if any) with this
1908          * latest one.
1909          *
1910          * This code conforms to the rate-limiting rule described in Section
1911          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1912          * an NS below.
1913          */
1914         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1915                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1916         if (ln->ln_hold)
1917                 m_freem(ln->ln_hold);
1918         ln->ln_hold = m;
1919         if (ln->ln_expire) {
1920                 if (ln->ln_asked < nd6_mmaxtries &&
1921                     ln->ln_expire < time_uptime) {
1922                         ln->ln_asked++;
1923                         ln->ln_expire = time_uptime +
1924                                 ND_IFINFO(ifp)->retrans / 1000;
1925                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1926                 }
1927         }
1928         return (0);
1929
1930 sendpkt:
1931         if (ifp->if_flags & IFF_LOOPBACK)
1932                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1933         else
1934                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1935         return (error);
1936
1937 bad:
1938         m_freem(m);
1939         return (error);
1940 }
1941 #undef gotoerr
1942
1943 int
1944 nd6_need_cache(struct ifnet *ifp)
1945 {
1946         /*
1947          * XXX: we currently do not make neighbor cache on any interface
1948          * other than Ethernet and GIF.
1949          *
1950          * RFC2893 says:
1951          * - unidirectional tunnels needs no ND
1952          */
1953         switch (ifp->if_type) {
1954         case IFT_ETHER:
1955         case IFT_IEEE1394:
1956 #ifdef IFT_L2VLAN
1957         case IFT_L2VLAN:
1958 #endif
1959 #ifdef IFT_IEEE80211
1960         case IFT_IEEE80211:
1961 #endif
1962 #ifdef IFT_CARP
1963         case IFT_CARP:
1964 #endif
1965         case IFT_GIF:           /* XXX need more cases? */
1966                 return (1);
1967         default:
1968                 return (0);
1969         }
1970 }
1971
1972 int
1973 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1974                 struct sockaddr *dst, u_char *desten)
1975 {
1976         struct sockaddr_dl *sdl;
1977         struct rtentry *rt;
1978
1979
1980         if (m->m_flags & M_MCAST) {
1981                 switch (ifp->if_type) {
1982                 case IFT_ETHER:
1983 #ifdef IFT_L2VLAN
1984         case IFT_L2VLAN:
1985 #endif
1986 #ifdef IFT_IEEE80211
1987                 case IFT_IEEE80211:
1988 #endif
1989                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
1990                                                  desten);
1991                         return (1);
1992                 case IFT_IEEE1394:
1993                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1994                         return (1);
1995                 default:
1996                         m_freem(m);
1997                         return (0);
1998                 }
1999         }
2000         if (rt0 == NULL) {
2001                 /* this could happen, if we could not allocate memory */
2002                 m_freem(m);
2003                 return (0);
2004         }
2005         if (rt_llroute(dst, rt0, &rt) != 0) {
2006                 m_freem(m);
2007                 return (0);
2008         }
2009         if (rt->rt_gateway->sa_family != AF_LINK) {
2010                 kprintf("nd6_storelladdr: something odd happens\n");
2011                 m_freem(m);
2012                 return (0);
2013         }
2014         sdl = SDL(rt->rt_gateway);
2015         if (sdl->sdl_alen == 0) {
2016                 /* this should be impossible, but we bark here for debugging */
2017                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2018                 m_freem(m);
2019                 return (0);
2020         }
2021
2022         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2023         return (1);
2024 }
2025
2026 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2027 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2028 #ifdef SYSCTL_DECL
2029 SYSCTL_DECL(_net_inet6_icmp6);
2030 #endif
2031 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2032         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2033 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2034         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2035
2036 static int
2037 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2038 {
2039         int error;
2040         char buf[1024];
2041         struct in6_defrouter *d, *de;
2042         struct nd_defrouter *dr;
2043
2044         if (req->newptr)
2045                 return EPERM;
2046         error = 0;
2047
2048         for (dr = TAILQ_FIRST(&nd_defrouter);
2049              dr;
2050              dr = TAILQ_NEXT(dr, dr_entry)) {
2051                 d = (struct in6_defrouter *)buf;
2052                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2053
2054                 if (d + 1 <= de) {
2055                         bzero(d, sizeof(*d));
2056                         d->rtaddr.sin6_family = AF_INET6;
2057                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2058                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2059                             dr->ifp) != 0)
2060                                 log(LOG_ERR,
2061                                     "scope error in "
2062                                     "default router list (%s)\n",
2063                                     ip6_sprintf(&dr->rtaddr));
2064                         d->flags = dr->flags;
2065                         d->rtlifetime = dr->rtlifetime;
2066                         d->expire = dr->expire;
2067                         d->if_index = dr->ifp->if_index;
2068                 } else
2069                         panic("buffer too short");
2070
2071                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2072                 if (error)
2073                         break;
2074         }
2075         return error;
2076 }
2077
2078 static int
2079 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2080 {
2081         int error;
2082         char buf[1024];
2083         struct in6_prefix *p, *pe;
2084         struct nd_prefix *pr;
2085
2086         if (req->newptr)
2087                 return EPERM;
2088         error = 0;
2089
2090         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2091                 u_short advrtrs;
2092                 size_t advance;
2093                 struct sockaddr_in6 *sin6, *s6;
2094                 struct nd_pfxrouter *pfr;
2095
2096                 p = (struct in6_prefix *)buf;
2097                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2098
2099                 if (p + 1 <= pe) {
2100                         bzero(p, sizeof(*p));
2101                         sin6 = (struct sockaddr_in6 *)(p + 1);
2102
2103                         p->prefix = pr->ndpr_prefix;
2104                         if (in6_recoverscope(&p->prefix,
2105                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2106                                 log(LOG_ERR,
2107                                     "scope error in prefix list (%s)\n",
2108                                     ip6_sprintf(&p->prefix.sin6_addr));
2109                         p->raflags = pr->ndpr_raf;
2110                         p->prefixlen = pr->ndpr_plen;
2111                         p->vltime = pr->ndpr_vltime;
2112                         p->pltime = pr->ndpr_pltime;
2113                         p->if_index = pr->ndpr_ifp->if_index;
2114                         p->expire = pr->ndpr_expire;
2115                         p->refcnt = pr->ndpr_refcnt;
2116                         p->flags = pr->ndpr_stateflags;
2117                         p->origin = PR_ORIG_RA;
2118                         advrtrs = 0;
2119                         for (pfr = pr->ndpr_advrtrs.lh_first;
2120                              pfr;
2121                              pfr = pfr->pfr_next) {
2122                                 if ((void *)&sin6[advrtrs + 1] >
2123                                     (void *)pe) {
2124                                         advrtrs++;
2125                                         continue;
2126                                 }
2127                                 s6 = &sin6[advrtrs];
2128                                 bzero(s6, sizeof(*s6));
2129                                 s6->sin6_family = AF_INET6;
2130                                 s6->sin6_len = sizeof(*sin6);
2131                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2132                                                      pfr->router->ifp) != 0)
2133                                         log(LOG_ERR,
2134                                             "scope error in "
2135                                             "prefix list (%s)\n",
2136                                             ip6_sprintf(&pfr->router->rtaddr));
2137                                 advrtrs++;
2138                         }
2139                         p->advrtrs = advrtrs;
2140                 } else
2141                         panic("buffer too short");
2142
2143                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2144                 error = SYSCTL_OUT(req, buf, advance);
2145                 if (error)
2146                         break;
2147         }
2148         return error;
2149 }