sys/netinet6/nd6.c

   1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
   2 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   3
   4 /*
   5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of the project nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 #include "opt_inet.h"
  34 #include "opt_inet6.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/callout.h>
  39 #include <sys/malloc.h>
  40 #include <sys/mbuf.h>
  41 #include <sys/socket.h>
  42 #include <sys/sockio.h>
  43 #include <sys/time.h>
  44 #include <sys/kernel.h>
  45 #include <sys/protosw.h>
  46 #include <sys/errno.h>
  47 #include <sys/syslog.h>
  48 #include <sys/queue.h>
  49 #include <sys/sysctl.h>
  50 #include <sys/mutex.h>
  51
  52 #include <sys/thread2.h>
  53 #include <sys/mutex2.h>
  54
  55 #include <net/if.h>
  56 #include <net/if_dl.h>
  57 #include <net/if_types.h>
  58 #include <net/route.h>
  59 #include <net/netisr2.h>
  60 #include <net/netmsg2.h>
  61
  62 #include <netinet/in.h>
  63 #include <netinet/if_ether.h>
  64 #include <netinet6/in6_var.h>
  65 #include <netinet/ip6.h>
  66 #include <netinet6/ip6_var.h>
  67 #include <netinet6/nd6.h>
  68 #include <netinet/icmp6.h>
  69
  70 #include <net/net_osdep.h>
  71
  72 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  73 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  74
  75 #define SIN6(s) ((struct sockaddr_in6 *)s)
  76 #define SDL(s) ((struct sockaddr_dl *)s)
  77
  78 /* timer values */
  79 int     nd6_prune       = 1;    /* walk list every 1 seconds */
  80 int     nd6_delay       = 5;    /* delay first probe time 5 second */
  81 int     nd6_umaxtries   = 3;    /* maximum unicast query */
  82 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
  83 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
  84 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
  85
  86 /* preventing too many loops in ND option parsing */
  87 int nd6_maxndopt = 10;  /* max # of ND options allowed */
  88
  89 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
  90
  91 #ifdef ND6_DEBUG
  92 int nd6_debug = 1;
  93 #else
  94 int nd6_debug = 0;
  95 #endif
  96
  97 /* for debugging? */
  98 static int nd6_inuse, nd6_allocated;
  99
 100 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 101 struct nd_drhead nd_defrouter;
 102 struct nd_prhead nd_prefix = { 0 };
 103 struct mtx nd6_mtx = MTX_INITIALIZER("nd6");
 104
 105 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 106 static struct sockaddr_in6 all1_sa;
 107
 108 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
 109 static int regen_tmpaddr (struct in6_ifaddr *);
 110 static void nd6_slowtimo(void *);
 111 static void nd6_slowtimo_dispatch(netmsg_t);
 112 static void nd6_timer(void *);
 113 static void nd6_timer_dispatch(netmsg_t);
 114
 115 static struct callout nd6_slowtimo_ch;
 116 static struct netmsg_base nd6_slowtimo_netmsg;
 117
 118 static struct callout nd6_timer_ch;
 119 static struct netmsg_base nd6_timer_netmsg;
 120
 121 void
 122 nd6_init(void)
 123 {
 124         static int nd6_init_done = 0;
 125         int i;
 126
 127         if (nd6_init_done) {
 128                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 129                 return;
 130         }
 131
 132         all1_sa.sin6_family = AF_INET6;
 133         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 134         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 135                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 136
 137         /* initialization of the default router list */
 138         TAILQ_INIT(&nd_defrouter);
 139
 140         nd6_init_done = 1;
 141
 142         /* start timer */
 143         callout_init_mp(&nd6_slowtimo_ch);
 144         netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
 145             MSGF_PRIORITY, nd6_slowtimo_dispatch);
 146         callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 147             nd6_slowtimo, NULL, 0);
 148 }
 149
 150 struct nd_ifinfo *
 151 nd6_ifattach(struct ifnet *ifp)
 152 {
 153         struct nd_ifinfo *nd;
 154
 155         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
 156             M_WAITOK | M_ZERO);
 157
 158         nd->initialized = 1;
 159
 160         nd->chlim = IPV6_DEFHLIM;
 161         nd->basereachable = REACHABLE_TIME;
 162         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 163         nd->retrans = RETRANS_TIMER;
 164
 165         /*
 166          * Note that the default value of ip6_accept_rtadv is 0, which means
 167          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 168          * here.
 169          */
 170         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 171
 172         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 173         nd6_setmtu0(ifp, nd);
 174         return nd;
 175 }
 176
 177 void
 178 nd6_ifdetach(struct nd_ifinfo *nd)
 179 {
 180         kfree(nd, M_IP6NDP);
 181 }
 182
 183 /*
 184  * Reset ND level link MTU. This function is called when the physical MTU
 185  * changes, which means we might have to adjust the ND level MTU.
 186  */
 187 void
 188 nd6_setmtu(struct ifnet *ifp)
 189 {
 190         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 191 }
 192
 193 struct netmsg_nd6setmtu {
 194         struct netmsg_base      nmsg;
 195         struct ifnet            *ifp;
 196         struct nd_ifinfo        *ndi;
 197 };
 198
 199 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 200 static void
 201 nd6_setmtu0_dispatch(netmsg_t msg)
 202 {
 203         struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
 204         struct ifnet *ifp = nmsg->ifp;
 205         struct nd_ifinfo *ndi = nmsg->ndi;
 206         uint32_t omaxmtu;
 207
 208         omaxmtu = ndi->maxmtu;
 209
 210         switch (ifp->if_type) {
 211         case IFT_ETHER:
 212                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 213                 break;
 214         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 215                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 216                 break;
 217 #ifdef IFT_IEEE80211
 218         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 219                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 220                 break;
 221 #endif
 222         default:
 223                 ndi->maxmtu = ifp->if_mtu;
 224                 break;
 225         }
 226
 227         /*
 228          * Decreasing the interface MTU under IPV6 minimum MTU may cause
 229          * undesirable situation.  We thus notify the operator of the change
 230          * explicitly.  The check for omaxmtu is necessary to restrict the
 231          * log to the case of changing the MTU, not initializing it.
 232          */
 233         if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
 234                 log(LOG_NOTICE, "nd6_setmtu0: "
 235                     "new link MTU on %s (%lu) is too small for IPv6\n",
 236                     if_name(ifp), (unsigned long)ndi->maxmtu);
 237         }
 238
 239         if (ndi->maxmtu > in6_maxmtu)
 240                 in6_setmaxmtu(); /* check all interfaces just in case */
 241
 242         lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
 243 }
 244
 245 void
 246 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 247 {
 248         struct netmsg_nd6setmtu nmsg;
 249
 250         netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
 251             nd6_setmtu0_dispatch);
 252         nmsg.ifp = ifp;
 253         nmsg.ndi = ndi;
 254         lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
 255 }
 256
 257 void
 258 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 259 {
 260         bzero(ndopts, sizeof(*ndopts));
 261         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 262         ndopts->nd_opts_last
 263                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 264
 265         if (icmp6len == 0) {
 266                 ndopts->nd_opts_done = 1;
 267                 ndopts->nd_opts_search = NULL;
 268         }
 269 }
 270
 271 /*
 272  * Take one ND option.
 273  */
 274 struct nd_opt_hdr *
 275 nd6_option(union nd_opts *ndopts)
 276 {
 277         struct nd_opt_hdr *nd_opt;
 278         int olen;
 279
 280         if (!ndopts)
 281                 panic("ndopts == NULL in nd6_option");
 282         if (!ndopts->nd_opts_last)
 283                 panic("uninitialized ndopts in nd6_option");
 284         if (!ndopts->nd_opts_search)
 285                 return NULL;
 286         if (ndopts->nd_opts_done)
 287                 return NULL;
 288
 289         nd_opt = ndopts->nd_opts_search;
 290
 291         /* make sure nd_opt_len is inside the buffer */
 292         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 293                 bzero(ndopts, sizeof(*ndopts));
 294                 return NULL;
 295         }
 296
 297         olen = nd_opt->nd_opt_len << 3;
 298         if (olen == 0) {
 299                 /*
 300                  * Message validation requires that all included
 301                  * options have a length that is greater than zero.
 302                  */
 303                 bzero(ndopts, sizeof(*ndopts));
 304                 return NULL;
 305         }
 306
 307         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 308         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 309                 /* option overruns the end of buffer, invalid */
 310                 bzero(ndopts, sizeof(*ndopts));
 311                 return NULL;
 312         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 313                 /* reached the end of options chain */
 314                 ndopts->nd_opts_done = 1;
 315                 ndopts->nd_opts_search = NULL;
 316         }
 317         return nd_opt;
 318 }
 319
 320 /*
 321  * Parse multiple ND options.
 322  * This function is much easier to use, for ND routines that do not need
 323  * multiple options of the same type.
 324  */
 325 int
 326 nd6_options(union nd_opts *ndopts)
 327 {
 328         struct nd_opt_hdr *nd_opt;
 329         int i = 0;
 330
 331         if (!ndopts)
 332                 panic("ndopts == NULL in nd6_options");
 333         if (!ndopts->nd_opts_last)
 334                 panic("uninitialized ndopts in nd6_options");
 335         if (!ndopts->nd_opts_search)
 336                 return 0;
 337
 338         while (1) {
 339                 nd_opt = nd6_option(ndopts);
 340                 if (!nd_opt && !ndopts->nd_opts_last) {
 341                         /*
 342                          * Message validation requires that all included
 343                          * options have a length that is greater than zero.
 344                          */
 345                         icmp6stat.icp6s_nd_badopt++;
 346                         bzero(ndopts, sizeof(*ndopts));
 347                         return -1;
 348                 }
 349
 350                 if (!nd_opt)
 351                         goto skip1;
 352
 353                 switch (nd_opt->nd_opt_type) {
 354                 case ND_OPT_SOURCE_LINKADDR:
 355                 case ND_OPT_TARGET_LINKADDR:
 356                 case ND_OPT_MTU:
 357                 case ND_OPT_REDIRECTED_HEADER:
 358                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 359                                 nd6log((LOG_INFO,
 360                                     "duplicated ND6 option found (type=%d)\n",
 361                                     nd_opt->nd_opt_type));
 362                                 /* XXX bark? */
 363                         } else {
 364                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 365                                         = nd_opt;
 366                         }
 367                         break;
 368                 case ND_OPT_PREFIX_INFORMATION:
 369                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 370                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 371                                         = nd_opt;
 372                         }
 373                         ndopts->nd_opts_pi_end =
 374                                 (struct nd_opt_prefix_info *)nd_opt;
 375                         break;
 376                 default:
 377                         /*
 378                          * Unknown options must be silently ignored,
 379                          * to accomodate future extension to the protocol.
 380                          */
 381                         nd6log((LOG_DEBUG,
 382                             "nd6_options: unsupported option %d - "
 383                             "option ignored\n", nd_opt->nd_opt_type));
 384                 }
 385
 386 skip1:
 387                 i++;
 388                 if (i > nd6_maxndopt) {
 389                         icmp6stat.icp6s_nd_toomanyopt++;
 390                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 391                         break;
 392                 }
 393
 394                 if (ndopts->nd_opts_done)
 395                         break;
 396         }
 397
 398         return 0;
 399 }
 400
 401 /*
 402  * ND6 timer routine to expire default route list and prefix list
 403  */
 404 static void
 405 nd6_timer_dispatch(netmsg_t nmsg)
 406 {
 407         struct llinfo_nd6 *ln;
 408         struct nd_defrouter *dr;
 409         struct nd_prefix *pr;
 410         struct ifnet *ifp;
 411         struct in6_ifaddr *ia6, *nia6;
 412
 413         ASSERT_NETISR0;
 414
 415         crit_enter();
 416         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
 417         crit_exit();
 418
 419         mtx_lock(&nd6_mtx);
 420
 421         ln = llinfo_nd6.ln_next;
 422         while (ln && ln != &llinfo_nd6) {
 423                 struct rtentry *rt;
 424                 struct sockaddr_in6 *dst;
 425                 struct llinfo_nd6 *next = ln->ln_next;
 426                 /* XXX: used for the DELAY case only: */
 427                 struct nd_ifinfo *ndi = NULL;
 428
 429                 if ((rt = ln->ln_rt) == NULL) {
 430                         ln = next;
 431                         continue;
 432                 }
 433                 if ((ifp = rt->rt_ifp) == NULL) {
 434                         ln = next;
 435                         continue;
 436                 }
 437                 ndi = ND_IFINFO(ifp);
 438                 dst = (struct sockaddr_in6 *)rt_key(rt);
 439
 440                 if (ln->ln_expire > time_uptime) {
 441                         ln = next;
 442                         continue;
 443                 }
 444
 445                 /* sanity check */
 446                 if (!rt)
 447                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 448                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 449                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 450                               rt->rt_llinfo, ln);
 451                 if (!dst)
 452                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 453
 454                 switch (ln->ln_state) {
 455                 case ND6_LLINFO_INCOMPLETE:
 456                         if (ln->ln_asked < nd6_mmaxtries) {
 457                                 ln->ln_asked++;
 458                                 ln->ln_expire = time_uptime +
 459                                         ND_IFINFO(ifp)->retrans / 1000;
 460                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 461                                         ln, 0);
 462                         } else {
 463                                 struct mbuf *m = ln->ln_hold;
 464                                 if (m) {
 465                                         if (rt->rt_ifp) {
 466                                                 /*
 467                                                  * Fake rcvif to make ICMP error
 468                                                  * more helpful in diagnosing
 469                                                  * for the receiver.
 470                                                  * XXX: should we consider
 471                                                  * older rcvif?
 472                                                  */
 473                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 474                                         }
 475                                         icmp6_error(m, ICMP6_DST_UNREACH,
 476                                                     ICMP6_DST_UNREACH_ADDR, 0);
 477                                         ln->ln_hold = NULL;
 478                                 }
 479                                 next = nd6_free(rt);
 480                         }
 481                         break;
 482                 case ND6_LLINFO_REACHABLE:
 483                         if (ln->ln_expire) {
 484                                 ln->ln_state = ND6_LLINFO_STALE;
 485                                 ln->ln_expire = time_uptime + nd6_gctimer;
 486                         }
 487                         break;
 488
 489                 case ND6_LLINFO_STALE:
 490                         /* Garbage Collection(RFC 2461 5.3) */
 491                         if (ln->ln_expire)
 492                                 next = nd6_free(rt);
 493                         break;
 494
 495                 case ND6_LLINFO_DELAY:
 496                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 497                                 /* We need NUD */
 498                                 ln->ln_asked = 1;
 499                                 ln->ln_state = ND6_LLINFO_PROBE;
 500                                 ln->ln_expire = time_uptime +
 501                                         ndi->retrans / 1000;
 502                                 nd6_ns_output(ifp, &dst->sin6_addr,
 503                                               &dst->sin6_addr,
 504                                               ln, 0);
 505                         } else {
 506                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 507                                 ln->ln_expire = time_uptime + nd6_gctimer;
 508                         }
 509                         break;
 510                 case ND6_LLINFO_PROBE:
 511                         if (ln->ln_asked < nd6_umaxtries) {
 512                                 ln->ln_asked++;
 513                                 ln->ln_expire = time_uptime +
 514                                         ND_IFINFO(ifp)->retrans / 1000;
 515                                 nd6_ns_output(ifp, &dst->sin6_addr,
 516                                                &dst->sin6_addr, ln, 0);
 517                         } else {
 518                                 next = nd6_free(rt);
 519                         }
 520                         break;
 521                 }
 522                 ln = next;
 523         }
 524
 525         /* expire default router list */
 526         dr = TAILQ_FIRST(&nd_defrouter);
 527         while (dr) {
 528                 if (dr->expire && dr->expire < time_uptime) {
 529                         struct nd_defrouter *t;
 530                         t = TAILQ_NEXT(dr, dr_entry);
 531                         defrtrlist_del(dr);
 532                         dr = t;
 533                 } else {
 534                         dr = TAILQ_NEXT(dr, dr_entry);
 535                 }
 536         }
 537
 538         /*
 539          * expire interface addresses.
 540          * in the past the loop was inside prefix expiry processing.
 541          * However, from a stricter speci-confrmance standpoint, we should
 542          * rather separate address lifetimes and prefix lifetimes.
 543          */
 544 addrloop:
 545         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 546                 nia6 = ia6->ia_next;
 547                 /* check address lifetime */
 548                 if (IFA6_IS_INVALID(ia6)) {
 549                         int regen = 0;
 550
 551                         /*
 552                          * If the expiring address is temporary, try
 553                          * regenerating a new one.  This would be useful when
 554                          * we suspended a laptop PC, then turned it on after a
 555                          * period that could invalidate all temporary
 556                          * addresses.  Although we may have to restart the
 557                          * loop (see below), it must be after purging the
 558                          * address.  Otherwise, we'd see an infinite loop of
 559                          * regeneration.
 560                          */
 561                         if (ip6_use_tempaddr &&
 562                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 563                                 if (regen_tmpaddr(ia6) == 0)
 564                                         regen = 1;
 565                         }
 566
 567                         in6_purgeaddr(&ia6->ia_ifa);
 568
 569                         if (regen)
 570                                 goto addrloop; /* XXX: see below */
 571                 }
 572                 if (IFA6_IS_DEPRECATED(ia6)) {
 573                         int oldflags = ia6->ia6_flags;
 574
 575                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 576
 577                         /*
 578                          * If a temporary address has just become deprecated,
 579                          * regenerate a new one if possible.
 580                          */
 581                         if (ip6_use_tempaddr &&
 582                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 583                             !(oldflags & IN6_IFF_DEPRECATED)) {
 584
 585                                 if (regen_tmpaddr(ia6) == 0) {
 586                                         /*
 587                                          * A new temporary address is
 588                                          * generated.
 589                                          * XXX: this means the address chain
 590                                          * has changed while we are still in
 591                                          * the loop.  Although the change
 592                                          * would not cause disaster (because
 593                                          * it's not a deletion, but an
 594                                          * addition,) we'd rather restart the
 595                                          * loop just for safety.  Or does this
 596                                          * significantly reduce performance??
 597                                          */
 598                                         goto addrloop;
 599                                 }
 600                         }
 601                 } else {
 602                         /*
 603                          * A new RA might have made a deprecated address
 604                          * preferred.
 605                          */
 606                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 607                 }
 608         }
 609
 610         /* expire prefix list */
 611         pr = nd_prefix.lh_first;
 612         while (pr) {
 613                 /*
 614                  * check prefix lifetime.
 615                  * since pltime is just for autoconf, pltime processing for
 616                  * prefix is not necessary.
 617                  */
 618                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
 619                         struct nd_prefix *t;
 620                         t = pr->ndpr_next;
 621
 622                         /*
 623                          * address expiration and prefix expiration are
 624                          * separate.  NEVER perform in6_purgeaddr here.
 625                          */
 626
 627                         prelist_remove(pr);
 628                         pr = t;
 629                 } else
 630                         pr = pr->ndpr_next;
 631         }
 632
 633         mtx_unlock(&nd6_mtx);
 634
 635         callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
 636 }
 637
 638 static void
 639 nd6_timer(void *arg __unused)
 640 {
 641         struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
 642
 643         KASSERT(mycpuid == 0, ("not on cpu0"));
 644         crit_enter();
 645         if (lmsg->ms_flags & MSGF_DONE)
 646                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
 647         crit_exit();
 648 }
 649
 650 void
 651 nd6_timer_init(void)
 652 {
 653         callout_init_mp(&nd6_timer_ch);
 654         netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
 655             MSGF_PRIORITY, nd6_timer_dispatch);
 656         callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
 657 }
 658
 659 static int
 660 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 661                                          address */
 662 {
 663         struct ifaddr_container *ifac;
 664         struct ifnet *ifp;
 665         struct in6_ifaddr *public_ifa6 = NULL;
 666
 667         ifp = ia6->ia_ifa.ifa_ifp;
 668         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 669                 struct ifaddr *ifa = ifac->ifa;
 670                 struct in6_ifaddr *it6;
 671
 672                 if (ifa->ifa_addr->sa_family != AF_INET6)
 673                         continue;
 674
 675                 it6 = (struct in6_ifaddr *)ifa;
 676
 677                 /* ignore no autoconf addresses. */
 678                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 679                         continue;
 680
 681                 /* ignore autoconf addresses with different prefixes. */
 682                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 683                         continue;
 684
 685                 /*
 686                  * Now we are looking at an autoconf address with the same
 687                  * prefix as ours.  If the address is temporary and is still
 688                  * preferred, do not create another one.  It would be rare, but
 689                  * could happen, for example, when we resume a laptop PC after
 690                  * a long period.
 691                  */
 692                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 693                     !IFA6_IS_DEPRECATED(it6)) {
 694                         public_ifa6 = NULL;
 695                         break;
 696                 }
 697
 698                 /*
 699                  * This is a public autoconf address that has the same prefix
 700                  * as ours.  If it is preferred, keep it.  We can't break the
 701                  * loop here, because there may be a still-preferred temporary
 702                  * address with the prefix.
 703                  */
 704                 if (!IFA6_IS_DEPRECATED(it6))
 705                     public_ifa6 = it6;
 706         }
 707
 708         if (public_ifa6 != NULL) {
 709                 int e;
 710
 711                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 712                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 713                             " tmp addr,errno=%d\n", e);
 714                         return (-1);
 715                 }
 716                 return (0);
 717         }
 718
 719         return (-1);
 720 }
 721
 722 /*
 723  * Nuke neighbor cache/prefix/default router management table, right before
 724  * ifp goes away.
 725  */
 726 void
 727 nd6_purge(struct ifnet *ifp)
 728 {
 729         struct llinfo_nd6 *ln, *nln;
 730         struct nd_defrouter *dr, *ndr, drany;
 731         struct nd_prefix *pr, *npr;
 732
 733         /* Nuke default router list entries toward ifp */
 734         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 735                 /*
 736                  * The first entry of the list may be stored in
 737                  * the routing table, so we'll delete it later.
 738                  */
 739                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 740                         ndr = TAILQ_NEXT(dr, dr_entry);
 741                         if (dr->ifp == ifp)
 742                                 defrtrlist_del(dr);
 743                 }
 744                 dr = TAILQ_FIRST(&nd_defrouter);
 745                 if (dr->ifp == ifp)
 746                         defrtrlist_del(dr);
 747         }
 748
 749         /* Nuke prefix list entries toward ifp */
 750         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 751                 npr = pr->ndpr_next;
 752                 if (pr->ndpr_ifp == ifp) {
 753                         /*
 754                          * Previously, pr->ndpr_addr is removed as well,
 755                          * but I strongly believe we don't have to do it.
 756                          * nd6_purge() is only called from in6_ifdetach(),
 757                          * which removes all the associated interface addresses
 758                          * by itself.
 759                          * (jinmei@kame.net 20010129)
 760                          */
 761                         prelist_remove(pr);
 762                 }
 763         }
 764
 765         /* cancel default outgoing interface setting */
 766         if (nd6_defifindex == ifp->if_index)
 767                 nd6_setdefaultiface(0);
 768
 769         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 770                 /* refresh default router list */
 771                 bzero(&drany, sizeof(drany));
 772                 defrouter_delreq(&drany, 0);
 773                 defrouter_select();
 774         }
 775
 776         /*
 777          * Nuke neighbor cache entries for the ifp.
 778          * Note that rt->rt_ifp may not be the same as ifp,
 779          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 780          * nd6_rtrequest(), and ip6_input().
 781          */
 782         ln = llinfo_nd6.ln_next;
 783         while (ln && ln != &llinfo_nd6) {
 784                 struct rtentry *rt;
 785                 struct sockaddr_dl *sdl;
 786
 787                 nln = ln->ln_next;
 788                 rt = ln->ln_rt;
 789                 if (rt && rt->rt_gateway &&
 790                     rt->rt_gateway->sa_family == AF_LINK) {
 791                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 792                         if (sdl->sdl_index == ifp->if_index)
 793                                 nln = nd6_free(rt);
 794                 }
 795                 ln = nln;
 796         }
 797 }
 798
 799 struct rtentry *
 800 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 801 {
 802         struct rtentry *rt;
 803         struct sockaddr_in6 sin6;
 804
 805         bzero(&sin6, sizeof(sin6));
 806         sin6.sin6_len = sizeof(struct sockaddr_in6);
 807         sin6.sin6_family = AF_INET6;
 808         sin6.sin6_addr = *addr6;
 809
 810         if (create)
 811                 rt = rtlookup((struct sockaddr *)&sin6);
 812         else
 813                 rt = rtpurelookup((struct sockaddr *)&sin6);
 814         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 815                 /*
 816                  * This is the case for the default route.
 817                  * If we want to create a neighbor cache for the address, we
 818                  * should free the route for the destination and allocate an
 819                  * interface route.
 820                  */
 821                 if (create) {
 822                         --rt->rt_refcnt;
 823                         rt = NULL;
 824                 }
 825         }
 826         if (!rt) {
 827                 if (create && ifp) {
 828                         int e;
 829
 830                         /*
 831                          * If no route is available and create is set,
 832                          * we allocate a host route for the destination
 833                          * and treat it like an interface route.
 834                          * This hack is necessary for a neighbor which can't
 835                          * be covered by our own prefix.
 836                          */
 837                         struct ifaddr *ifa;
 838
 839                         ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 840                         if (ifa == NULL)
 841                                 return (NULL);
 842
 843                         /*
 844                          * Create a new route.  RTF_LLINFO is necessary
 845                          * to create a Neighbor Cache entry for the
 846                          * destination in nd6_rtrequest which will be
 847                          * called in rtrequest via ifa->ifa_rtrequest.
 848                          */
 849                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 850                              ifa->ifa_addr, (struct sockaddr *)&all1_sa,
 851                              (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
 852                              ~RTF_CLONING, &rt)) != 0) {
 853                                 log(LOG_ERR,
 854                                     "nd6_lookup: failed to add route for a "
 855                                     "neighbor(%s), errno=%d\n",
 856                                     ip6_sprintf(addr6), e);
 857                         }
 858                         if (rt == NULL)
 859                                 return (NULL);
 860                         if (rt->rt_llinfo) {
 861                                 struct llinfo_nd6 *ln =
 862                                     (struct llinfo_nd6 *)rt->rt_llinfo;
 863
 864                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 865                         }
 866                 } else
 867                         return (NULL);
 868         }
 869         rt->rt_refcnt--;
 870         /*
 871          * Validation for the entry.
 872          * Note that the check for rt_llinfo is necessary because a cloned
 873          * route from a parent route that has the L flag (e.g. the default
 874          * route to a p2p interface) may have the flag, too, while the
 875          * destination is not actually a neighbor.
 876          * XXX: we can't use rt->rt_ifp to check for the interface, since
 877          *      it might be the loopback interface if the entry is for our
 878          *      own address on a non-loopback interface. Instead, we should
 879          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 880          *      interface.
 881          */
 882         if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
 883             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 884             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 885                 if (create) {
 886                         log(LOG_DEBUG,
 887                             "nd6_lookup: failed to lookup %s (if = %s)\n",
 888                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 889                         /* xxx more logs... kazu */
 890                 }
 891                 return (NULL);
 892         }
 893         return (rt);
 894 }
 895
 896 /*
 897  * Detect if a given IPv6 address identifies a neighbor on a given link.
 898  * XXX: should take care of the destination of a p2p link?
 899  */
 900 int
 901 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 902 {
 903         struct ifaddr_container *ifac;
 904         int i;
 905
 906 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 907 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 908
 909         /*
 910          * A link-local address is always a neighbor.
 911          * XXX: we should use the sin6_scope_id field rather than the embedded
 912          * interface index.
 913          */
 914         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 915             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 916                 return (1);
 917
 918         /*
 919          * If the address matches one of our addresses,
 920          * it should be a neighbor.
 921          */
 922         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 923                 struct ifaddr *ifa = ifac->ifa;
 924
 925                 if (ifa->ifa_addr->sa_family != AF_INET6)
 926                         next: continue;
 927
 928                 for (i = 0; i < 4; i++) {
 929                         if ((IFADDR6(ifa).s6_addr32[i] ^
 930                              addr->sin6_addr.s6_addr32[i]) &
 931                             IFMASK6(ifa).s6_addr32[i])
 932                                 goto next;
 933                 }
 934                 return (1);
 935         }
 936
 937         /*
 938          * Even if the address matches none of our addresses, it might be
 939          * in the neighbor cache.
 940          */
 941         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 942                 return (1);
 943
 944         return (0);
 945 #undef IFADDR6
 946 #undef IFMASK6
 947 }
 948
 949 /*
 950  * Free an nd6 llinfo entry.
 951  */
 952 struct llinfo_nd6 *
 953 nd6_free(struct rtentry *rt)
 954 {
 955         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 956         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 957         struct nd_defrouter *dr;
 958
 959         /*
 960          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
 961          * even though it is not harmful, it was not really necessary.
 962          */
 963
 964         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 965                 mtx_lock(&nd6_mtx);
 966                 dr = defrouter_lookup(
 967                     &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 968                     rt->rt_ifp);
 969
 970                 if (ln->ln_router || dr) {
 971                         /*
 972                          * rt6_flush must be called whether or not the neighbor
 973                          * is in the Default Router List.
 974                          * See a corresponding comment in nd6_na_input().
 975                          */
 976                         rt6_flush(&in6, rt->rt_ifp);
 977                 }
 978
 979                 if (dr) {
 980                         /*
 981                          * Unreachablity of a router might affect the default
 982                          * router selection and on-link detection of advertised
 983                          * prefixes.
 984                          */
 985
 986                         /*
 987                          * Temporarily fake the state to choose a new default
 988                          * router and to perform on-link determination of
 989                          * prefixes correctly.
 990                          * Below the state will be set correctly,
 991                          * or the entry itself will be deleted.
 992                          */
 993                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 994
 995                         /*
 996                          * Since defrouter_select() does not affect the
 997                          * on-link determination and MIP6 needs the check
 998                          * before the default router selection, we perform
 999                          * the check now.
1000                          */
1001                         pfxlist_onlink_check();
1002
1003                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1004                                 /*
1005                                  * It is used as the current default router,
1006                                  * so we have to move it to the end of the
1007                                  * list and choose a new one.
1008                                  * XXX: it is not very efficient if this is
1009                                  *      the only router.
1010                                  */
1011                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1012                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1013
1014                                 defrouter_select();
1015                         }
1016                 }
1017                 mtx_unlock(&nd6_mtx);
1018         }
1019
1020         /*
1021          * Before deleting the entry, remember the next entry as the
1022          * return value.  We need this because pfxlist_onlink_check() above
1023          * might have freed other entries (particularly the old next entry) as
1024          * a side effect (XXX).
1025          */
1026         next = ln->ln_next;
1027
1028         /*
1029          * Detach the route from the routing tree and the list of neighbor
1030          * caches, and disable the route entry not to be used in already
1031          * cached routes.
1032          */
1033         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1034
1035         return (next);
1036 }
1037
1038 /*
1039  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1040  *
1041  * XXX cost-effective metods?
1042  */
1043 void
1044 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1045 {
1046         struct llinfo_nd6 *ln;
1047
1048         /*
1049          * If the caller specified "rt", use that.  Otherwise, resolve the
1050          * routing table by supplied "dst6".
1051          */
1052         if (!rt) {
1053                 if (!dst6)
1054                         return;
1055                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1056                         return;
1057         }
1058
1059         if ((rt->rt_flags & RTF_GATEWAY) ||
1060             !(rt->rt_flags & RTF_LLINFO) ||
1061             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1062             rt->rt_gateway->sa_family != AF_LINK) {
1063                 /* This is not a host route. */
1064                 return;
1065         }
1066
1067         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1068         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1069                 return;
1070
1071         /*
1072          * if we get upper-layer reachability confirmation many times,
1073          * it is possible we have false information.
1074          */
1075         if (!force) {
1076                 ln->ln_byhint++;
1077                 if (ln->ln_byhint > nd6_maxnudhint)
1078                         return;
1079         }
1080
1081         ln->ln_state = ND6_LLINFO_REACHABLE;
1082         if (ln->ln_expire)
1083                 ln->ln_expire = time_uptime +
1084                         ND_IFINFO(rt->rt_ifp)->reachable;
1085 }
1086
1087 void
1088 nd6_rtrequest(int req, struct rtentry *rt)
1089 {
1090         struct sockaddr *gate = rt->rt_gateway;
1091         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1092         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1093         struct ifnet *ifp = rt->rt_ifp;
1094         struct ifaddr *ifa;
1095
1096         if ((rt->rt_flags & RTF_GATEWAY))
1097                 return;
1098
1099         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1100                 /*
1101                  * This is probably an interface direct route for a link
1102                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1103                  * We do not need special treatment below for such a route.
1104                  * Moreover, the RTF_LLINFO flag which would be set below
1105                  * would annoy the ndp(8) command.
1106                  */
1107                 return;
1108         }
1109
1110         if (req == RTM_RESOLVE &&
1111             (nd6_need_cache(ifp) == 0 || /* stf case */
1112              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1113                 /*
1114                  * FreeBSD and BSD/OS often make a cloned host route based
1115                  * on a less-specific route (e.g. the default route).
1116                  * If the less specific route does not have a "gateway"
1117                  * (this is the case when the route just goes to a p2p or an
1118                  * stf interface), we'll mistakenly make a neighbor cache for
1119                  * the host route, and will see strange neighbor solicitation
1120                  * for the corresponding destination.  In order to avoid the
1121                  * confusion, we check if the destination of the route is
1122                  * a neighbor in terms of neighbor discovery, and stop the
1123                  * process if not.  Additionally, we remove the LLINFO flag
1124                  * so that ndp(8) will not try to get the neighbor information
1125                  * of the destination.
1126                  */
1127                 rt->rt_flags &= ~RTF_LLINFO;
1128                 return;
1129         }
1130
1131         switch (req) {
1132         case RTM_ADD:
1133                 /*
1134                  * There is no backward compatibility :)
1135                  *
1136                  * if (!(rt->rt_flags & RTF_HOST) &&
1137                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1138                  *         rt->rt_flags |= RTF_CLONING;
1139                  */
1140                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1141                         /*
1142                          * Case 1: This route should come from
1143                          * a route to interface.  RTF_LLINFO flag is set
1144                          * for a host route whose destination should be
1145                          * treated as on-link.
1146                          */
1147                         rt_setgate(rt, rt_key(rt),
1148                                    (struct sockaddr *)&null_sdl,
1149                                    RTL_DONTREPORT);
1150                         gate = rt->rt_gateway;
1151                         SDL(gate)->sdl_type = ifp->if_type;
1152                         SDL(gate)->sdl_index = ifp->if_index;
1153                         if (ln)
1154                                 ln->ln_expire = time_uptime;
1155                         if (ln && ln->ln_expire == 0) {
1156                                 /* kludge for desktops */
1157                                 ln->ln_expire = 1;
1158                         }
1159                         if ((rt->rt_flags & RTF_CLONING))
1160                                 break;
1161                 }
1162                 /*
1163                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1164                  * We don't do that here since llinfo is not ready yet.
1165                  *
1166                  * There are also couple of other things to be discussed:
1167                  * - unsolicited NA code needs improvement beforehand
1168                  * - RFC2461 says we MAY send multicast unsolicited NA
1169                  *   (7.2.6 paragraph 4), however, it also says that we
1170                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1171                  *   we don't have anything like it right now.
1172                  *   note that the mechanism needs a mutual agreement
1173                  *   between proxies, which means that we need to implement
1174                  *   a new protocol, or a new kludge.
1175                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1176                  *   we need to check ip6forwarding before sending it.
1177                  *   (or should we allow proxy ND configuration only for
1178                  *   routers?  there's no mention about proxy ND from hosts)
1179                  */
1180 #if 0
1181                 /* XXX it does not work */
1182                 if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
1183                         nd6_na_output(ifp,
1184                               &SIN6(rt_key(rt))->sin6_addr,
1185                               &SIN6(rt_key(rt))->sin6_addr,
1186                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1187                               1, NULL);
1188                 }
1189 #endif
1190                 /* FALLTHROUGH */
1191         case RTM_RESOLVE:
1192                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1193                         /*
1194                          * Address resolution isn't necessary for a point to
1195                          * point link, so we can skip this test for a p2p link.
1196                          */
1197                         if (gate->sa_family != AF_LINK ||
1198                             gate->sa_len < sizeof(null_sdl)) {
1199                                 log(LOG_DEBUG,
1200                                     "nd6_rtrequest: bad gateway value: %s\n",
1201                                     if_name(ifp));
1202                                 break;
1203                         }
1204                         SDL(gate)->sdl_type = ifp->if_type;
1205                         SDL(gate)->sdl_index = ifp->if_index;
1206                 }
1207                 if (ln != NULL)
1208                         break;  /* This happens on a route change */
1209                 /*
1210                  * Case 2: This route may come from cloning, or a manual route
1211                  * add with a LL address.
1212                  */
1213                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1214                 rt->rt_llinfo = (caddr_t)ln;
1215                 if (!ln) {
1216                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1217                         break;
1218                 }
1219                 nd6_inuse++;
1220                 nd6_allocated++;
1221                 bzero(ln, sizeof(*ln));
1222                 ln->ln_rt = rt;
1223                 /* this is required for "ndp" command. - shin */
1224                 if (req == RTM_ADD) {
1225                         /*
1226                          * gate should have some valid AF_LINK entry,
1227                          * and ln->ln_expire should have some lifetime
1228                          * which is specified by ndp command.
1229                          */
1230                         ln->ln_state = ND6_LLINFO_REACHABLE;
1231                         ln->ln_byhint = 0;
1232                 } else {
1233                         /*
1234                          * When req == RTM_RESOLVE, rt is created and
1235                          * initialized in rtrequest(), so rt_expire is 0.
1236                          */
1237                         ln->ln_state = ND6_LLINFO_NOSTATE;
1238                         ln->ln_expire = time_uptime;
1239                 }
1240                 rt->rt_flags |= RTF_LLINFO;
1241                 ln->ln_next = llinfo_nd6.ln_next;
1242                 llinfo_nd6.ln_next = ln;
1243                 ln->ln_prev = &llinfo_nd6;
1244                 ln->ln_next->ln_prev = ln;
1245
1246                 /*
1247                  * check if rt_key(rt) is one of my address assigned
1248                  * to the interface.
1249                  */
1250                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1251                     &SIN6(rt_key(rt))->sin6_addr);
1252                 if (ifa) {
1253                         caddr_t macp = nd6_ifptomac(ifp);
1254                         ln->ln_expire = 0;
1255                         ln->ln_state = ND6_LLINFO_REACHABLE;
1256                         ln->ln_byhint = 0;
1257                         if (macp) {
1258                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1259                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1260                         }
1261                         if (nd6_useloopback) {
1262                                 rt->rt_ifp = loif;      /* XXX */
1263                                 /*
1264                                  * Make sure rt_ifa be equal to the ifaddr
1265                                  * corresponding to the address.
1266                                  * We need this because when we refer
1267                                  * rt_ifa->ia6_flags in ip6_input, we assume
1268                                  * that the rt_ifa points to the address instead
1269                                  * of the loopback address.
1270                                  */
1271                                 if (ifa != rt->rt_ifa) {
1272                                         IFAFREE(rt->rt_ifa);
1273                                         IFAREF(ifa);
1274                                         rt->rt_ifa = ifa;
1275                                 }
1276                         }
1277                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1278                         ln->ln_expire = 0;
1279                         ln->ln_state = ND6_LLINFO_REACHABLE;
1280                         ln->ln_byhint = 0;
1281
1282                         /*
1283                          * Join solicited node multicast for proxy ND, and only
1284                          * join it once on cpu0.
1285                          */
1286                         if ((ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1287                                 struct in6_addr llsol;
1288                                 int error;
1289
1290                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1291                                 llsol.s6_addr16[0] = htons(0xff02);
1292                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1293                                 llsol.s6_addr32[1] = 0;
1294                                 llsol.s6_addr32[2] = htonl(1);
1295                                 llsol.s6_addr8[12] = 0xff;
1296
1297                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1298                                         nd6log((LOG_ERR, "%s: failed to join "
1299                                             "%s (errno=%d)\n", if_name(ifp),
1300                                             ip6_sprintf(&llsol), error));
1301                                 }
1302                         }
1303                 }
1304                 break;
1305
1306         case RTM_DELETE:
1307                 if (!ln)
1308                         break;
1309                 /*
1310                  * Leave from solicited node multicast for proxy ND, and only
1311                  * leave it once on cpu0 (since we joined it once on cpu0).
1312                  */
1313                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1314                     (ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1315                         struct in6_addr llsol;
1316                         struct in6_multi *in6m;
1317
1318                         llsol = SIN6(rt_key(rt))->sin6_addr;
1319                         llsol.s6_addr16[0] = htons(0xff02);
1320                         llsol.s6_addr16[1] = htons(ifp->if_index);
1321                         llsol.s6_addr32[1] = 0;
1322                         llsol.s6_addr32[2] = htonl(1);
1323                         llsol.s6_addr8[12] = 0xff;
1324
1325                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1326                         if (in6m)
1327                                 in6_delmulti(in6m);
1328                 }
1329                 nd6_inuse--;
1330                 ln->ln_next->ln_prev = ln->ln_prev;
1331                 ln->ln_prev->ln_next = ln->ln_next;
1332                 ln->ln_prev = NULL;
1333                 rt->rt_llinfo = 0;
1334                 rt->rt_flags &= ~RTF_LLINFO;
1335                 if (ln->ln_hold)
1336                         m_freem(ln->ln_hold);
1337                 Free((caddr_t)ln);
1338         }
1339 }
1340
1341 int
1342 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1343 {
1344         struct in6_drlist *drl = (struct in6_drlist *)data;
1345         struct in6_prlist *prl = (struct in6_prlist *)data;
1346         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1347         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1348         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1349         struct nd_defrouter *dr, any;
1350         struct nd_prefix *pr;
1351         struct rtentry *rt;
1352         int i = 0, error = 0;
1353
1354         switch (cmd) {
1355         case SIOCGDRLST_IN6:
1356                 /*
1357                  * obsolete API, use sysctl under net.inet6.icmp6
1358                  */
1359                 bzero(drl, sizeof(*drl));
1360                 mtx_lock(&nd6_mtx);
1361                 dr = TAILQ_FIRST(&nd_defrouter);
1362                 while (dr && i < DRLSTSIZ) {
1363                         drl->defrouter[i].rtaddr = dr->rtaddr;
1364                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1365                                 /* XXX: need to this hack for KAME stack */
1366                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1367                         } else
1368                                 log(LOG_ERR,
1369                                     "default router list contains a "
1370                                     "non-linklocal address(%s)\n",
1371                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1372
1373                         drl->defrouter[i].flags = dr->flags;
1374                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1375                         drl->defrouter[i].expire = dr->expire;
1376                         drl->defrouter[i].if_index = dr->ifp->if_index;
1377                         i++;
1378                         dr = TAILQ_NEXT(dr, dr_entry);
1379                 }
1380                 mtx_unlock(&nd6_mtx);
1381                 break;
1382         case SIOCGPRLST_IN6:
1383                 /*
1384                  * obsolete API, use sysctl under net.inet6.icmp6
1385                  */
1386                 /*
1387                  * XXX meaning of fields, especialy "raflags", is very
1388                  * differnet between RA prefix list and RR/static prefix list.
1389                  * how about separating ioctls into two?
1390                  */
1391                 bzero(prl, sizeof(*prl));
1392                 mtx_lock(&nd6_mtx);
1393                 pr = nd_prefix.lh_first;
1394                 while (pr && i < PRLSTSIZ) {
1395                         struct nd_pfxrouter *pfr;
1396                         int j;
1397
1398                         in6_embedscope(&prl->prefix[i].prefix,
1399                             &pr->ndpr_prefix, NULL, NULL);
1400                         prl->prefix[i].raflags = pr->ndpr_raf;
1401                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1402                         prl->prefix[i].vltime = pr->ndpr_vltime;
1403                         prl->prefix[i].pltime = pr->ndpr_pltime;
1404                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1405                         prl->prefix[i].expire = pr->ndpr_expire;
1406
1407                         pfr = pr->ndpr_advrtrs.lh_first;
1408                         j = 0;
1409                         while (pfr) {
1410                                 if (j < DRLSTSIZ) {
1411 #define RTRADDR prl->prefix[i].advrtr[j]
1412                                         RTRADDR = pfr->router->rtaddr;
1413                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1414                                                 /* XXX: hack for KAME */
1415                                                 RTRADDR.s6_addr16[1] = 0;
1416                                         } else
1417                                                 log(LOG_ERR,
1418                                                     "a router(%s) advertises "
1419                                                     "a prefix with "
1420                                                     "non-link local address\n",
1421                                                     ip6_sprintf(&RTRADDR));
1422 #undef RTRADDR
1423                                 }
1424                                 j++;
1425                                 pfr = pfr->pfr_next;
1426                         }
1427                         prl->prefix[i].advrtrs = j;
1428                         prl->prefix[i].origin = PR_ORIG_RA;
1429
1430                         i++;
1431                         pr = pr->ndpr_next;
1432                 }
1433                 mtx_unlock(&nd6_mtx);
1434
1435                 break;
1436         case OSIOCGIFINFO_IN6:
1437                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1438                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1439                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1440                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1441                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1442                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1443                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1444                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1445                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1446                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1447                 break;
1448         case SIOCGIFINFO_IN6:
1449                 ndi->ndi = *ND_IFINFO(ifp);
1450                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1451                 break;
1452         case SIOCSIFINFO_FLAGS:
1453                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1454                 break;
1455         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1456                 /* flush default router list */
1457                 /*
1458                  * xxx sumikawa: should not delete route if default
1459                  * route equals to the top of default router list
1460                  */
1461                 bzero(&any, sizeof(any));
1462                 defrouter_delreq(&any, 0);
1463                 defrouter_select();
1464                 /* xxx sumikawa: flush prefix list */
1465                 break;
1466         case SIOCSPFXFLUSH_IN6:
1467         {
1468                 /* flush all the prefix advertised by routers */
1469                 struct nd_prefix *pr, *next;
1470
1471                 mtx_lock(&nd6_mtx);
1472                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1473                         struct in6_ifaddr *ia, *ia_next;
1474
1475                         next = pr->ndpr_next;
1476
1477                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1478                                 continue; /* XXX */
1479
1480                         /* do we really have to remove addresses as well? */
1481                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1482                                 /* ia might be removed.  keep the next ptr. */
1483                                 ia_next = ia->ia_next;
1484
1485                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1486                                         continue;
1487
1488                                 if (ia->ia6_ndpr == pr)
1489                                         in6_purgeaddr(&ia->ia_ifa);
1490                         }
1491                         prelist_remove(pr);
1492                 }
1493                 mtx_unlock(&nd6_mtx);
1494                 break;
1495         }
1496         case SIOCSRTRFLUSH_IN6:
1497         {
1498                 /* flush all the default routers */
1499                 struct nd_defrouter *dr, *next;
1500
1501                 mtx_lock(&nd6_mtx);
1502                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1503                         /*
1504                          * The first entry of the list may be stored in
1505                          * the routing table, so we'll delete it later.
1506                          */
1507                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1508                                 next = TAILQ_NEXT(dr, dr_entry);
1509                                 defrtrlist_del(dr);
1510                         }
1511                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1512                 }
1513                 mtx_unlock(&nd6_mtx);
1514                 break;
1515         }
1516         case SIOCGNBRINFO_IN6:
1517         {
1518                 struct llinfo_nd6 *ln;
1519                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1520
1521                 /*
1522                  * XXX: KAME specific hack for scoped addresses
1523                  *      XXXX: for other scopes than link-local?
1524                  */
1525                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1526                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1527                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1528
1529                         if (*idp == 0)
1530                                 *idp = htons(ifp->if_index);
1531                 }
1532
1533                 mtx_lock(&nd6_mtx);
1534                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1535                         error = EINVAL;
1536                         mtx_unlock(&nd6_mtx);
1537                         break;
1538                 }
1539                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1540                 nbi->state = ln->ln_state;
1541                 nbi->asked = ln->ln_asked;
1542                 nbi->isrouter = ln->ln_router;
1543                 nbi->expire = ln->ln_expire;
1544                 mtx_unlock(&nd6_mtx);
1545
1546                 break;
1547         }
1548         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1549                 ndif->ifindex = nd6_defifindex;
1550                 break;
1551         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1552                 return (nd6_setdefaultiface(ndif->ifindex));
1553         }
1554         return (error);
1555 }
1556
1557 /*
1558  * Create neighbor cache entry and cache link-layer address,
1559  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1560  */
1561 struct rtentry *
1562 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1563                  int lladdrlen,
1564                  int type,      /* ICMP6 type */
1565                  int code       /* type dependent information */)
1566 {
1567         struct rtentry *rt = NULL;
1568         struct llinfo_nd6 *ln = NULL;
1569         int is_newentry;
1570         struct sockaddr_dl *sdl = NULL;
1571         int do_update;
1572         int olladdr;
1573         int llchange;
1574         int newstate = 0;
1575
1576         if (!ifp)
1577                 panic("ifp == NULL in nd6_cache_lladdr");
1578         if (!from)
1579                 panic("from == NULL in nd6_cache_lladdr");
1580
1581         /* nothing must be updated for unspecified address */
1582         if (IN6_IS_ADDR_UNSPECIFIED(from))
1583                 return NULL;
1584
1585         /*
1586          * Validation about ifp->if_addrlen and lladdrlen must be done in
1587          * the caller.
1588          *
1589          * XXX If the link does not have link-layer adderss, what should
1590          * we do? (ifp->if_addrlen == 0)
1591          * Spec says nothing in sections for RA, RS and NA.  There's small
1592          * description on it in NS section (RFC 2461 7.2.3).
1593          */
1594
1595         rt = nd6_lookup(from, 0, ifp);
1596         if (!rt) {
1597 #if 0
1598                 /* nothing must be done if there's no lladdr */
1599                 if (!lladdr || !lladdrlen)
1600                         return NULL;
1601 #endif
1602
1603                 rt = nd6_lookup(from, 1, ifp);
1604                 is_newentry = 1;
1605         } else {
1606                 /* do nothing if static ndp is set */
1607                 if (rt->rt_flags & RTF_STATIC)
1608                         return NULL;
1609                 is_newentry = 0;
1610         }
1611
1612         if (!rt)
1613                 return NULL;
1614         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1615 fail:
1616                 nd6_free(rt);
1617                 return NULL;
1618         }
1619         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1620         if (!ln)
1621                 goto fail;
1622         if (!rt->rt_gateway)
1623                 goto fail;
1624         if (rt->rt_gateway->sa_family != AF_LINK)
1625                 goto fail;
1626         sdl = SDL(rt->rt_gateway);
1627
1628         olladdr = (sdl->sdl_alen) ? 1 : 0;
1629         if (olladdr && lladdr) {
1630                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1631                         llchange = 1;
1632                 else
1633                         llchange = 0;
1634         } else
1635                 llchange = 0;
1636
1637         /*
1638          * newentry olladdr  lladdr  llchange   (*=record)
1639          *      0       n       n       --      (1)
1640          *      0       y       n       --      (2)
1641          *      0       n       y       --      (3) * STALE
1642          *      0       y       y       n       (4) *
1643          *      0       y       y       y       (5) * STALE
1644          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1645          *      1       --      y       --      (7) * STALE
1646          */
1647
1648         if (lladdr) {           /* (3-5) and (7) */
1649                 /*
1650                  * Record source link-layer address
1651                  * XXX is it dependent to ifp->if_type?
1652                  */
1653                 sdl->sdl_alen = ifp->if_addrlen;
1654                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1655         }
1656
1657         if (!is_newentry) {
1658                 if ((!olladdr && lladdr) ||             /* (3) */
1659                     (olladdr && lladdr && llchange)) {  /* (5) */
1660                         do_update = 1;
1661                         newstate = ND6_LLINFO_STALE;
1662                 } else {                                /* (1-2,4) */
1663                         do_update = 0;
1664                 }
1665         } else {
1666                 do_update = 1;
1667                 if (!lladdr)                            /* (6) */
1668                         newstate = ND6_LLINFO_NOSTATE;
1669                 else                                    /* (7) */
1670                         newstate = ND6_LLINFO_STALE;
1671         }
1672
1673         if (do_update) {
1674                 /*
1675                  * Update the state of the neighbor cache.
1676                  */
1677                 ln->ln_state = newstate;
1678
1679                 if (ln->ln_state == ND6_LLINFO_STALE) {
1680                         /*
1681                          * XXX: since nd6_output() below will cause
1682                          * state tansition to DELAY and reset the timer,
1683                          * we must set the timer now, although it is actually
1684                          * meaningless.
1685                          */
1686                         ln->ln_expire = time_uptime + nd6_gctimer;
1687
1688                         if (ln->ln_hold) {
1689                                 /*
1690                                  * we assume ifp is not a p2p here, so just
1691                                  * set the 2nd argument as the 1st one.
1692                                  */
1693                                 nd6_output(ifp, ifp, ln->ln_hold,
1694                                     (struct sockaddr_in6 *)rt_key(rt), rt);
1695                                 ln->ln_hold = NULL;
1696                         }
1697                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1698                         /* probe right away */
1699                         ln->ln_expire = time_uptime;
1700                 }
1701         }
1702
1703         /*
1704          * ICMP6 type dependent behavior.
1705          *
1706          * NS: clear IsRouter if new entry
1707          * RS: clear IsRouter
1708          * RA: set IsRouter if there's lladdr
1709          * redir: clear IsRouter if new entry
1710          *
1711          * RA case, (1):
1712          * The spec says that we must set IsRouter in the following cases:
1713          * - If lladdr exist, set IsRouter.  This means (1-5).
1714          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1715          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1716          * A quetion arises for (1) case.  (1) case has no lladdr in the
1717          * neighbor cache, this is similar to (6).
1718          * This case is rare but we figured that we MUST NOT set IsRouter.
1719          *
1720          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1721          *                                                      D R
1722          *      0       n       n       --      (1)     c   ?     s
1723          *      0       y       n       --      (2)     c   s     s
1724          *      0       n       y       --      (3)     c   s     s
1725          *      0       y       y       n       (4)     c   s     s
1726          *      0       y       y       y       (5)     c   s     s
1727          *      1       --      n       --      (6) c   c       c s
1728          *      1       --      y       --      (7) c   c   s   c s
1729          *
1730          *                                      (c=clear s=set)
1731          */
1732         switch (type & 0xff) {
1733         case ND_NEIGHBOR_SOLICIT:
1734                 /*
1735                  * New entry must have is_router flag cleared.
1736                  */
1737                 if (is_newentry)        /* (6-7) */
1738                         ln->ln_router = 0;
1739                 break;
1740         case ND_REDIRECT:
1741                 /*
1742                  * If the icmp is a redirect to a better router, always set the
1743                  * is_router flag.  Otherwise, if the entry is newly created,
1744                  * clear the flag.  [RFC 2461, sec 8.3]
1745                  */
1746                 if (code == ND_REDIRECT_ROUTER)
1747                         ln->ln_router = 1;
1748                 else if (is_newentry) /* (6-7) */
1749                         ln->ln_router = 0;
1750                 break;
1751         case ND_ROUTER_SOLICIT:
1752                 /*
1753                  * is_router flag must always be cleared.
1754                  */
1755                 ln->ln_router = 0;
1756                 break;
1757         case ND_ROUTER_ADVERT:
1758                 /*
1759                  * Mark an entry with lladdr as a router.
1760                  */
1761                 if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
1762                     (is_newentry && lladdr)) {                  /* (7) */
1763                         ln->ln_router = 1;
1764                 }
1765                 break;
1766         }
1767
1768         /*
1769          * When the link-layer address of a router changes, select the
1770          * best router again.  In particular, when the neighbor entry is newly
1771          * created, it might affect the selection policy.
1772          * Question: can we restrict the first condition to the "is_newentry"
1773          * case?
1774          * XXX: when we hear an RA from a new router with the link-layer
1775          * address option, defrouter_select() is called twice, since
1776          * defrtrlist_update called the function as well.  However, I believe
1777          * we can compromise the overhead, since it only happens the first
1778          * time.
1779          * XXX: although defrouter_select() should not have a bad effect
1780          * for those are not autoconfigured hosts, we explicitly avoid such
1781          * cases for safety.
1782          */
1783         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1784                 defrouter_select();
1785
1786         return rt;
1787 }
1788
1789 static void
1790 nd6_slowtimo(void *arg __unused)
1791 {
1792         struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1793
1794         KASSERT(mycpuid == 0, ("not on cpu0"));
1795         crit_enter();
1796         if (lmsg->ms_flags & MSGF_DONE)
1797                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1798         crit_exit();
1799 }
1800
1801 static void
1802 nd6_slowtimo_dispatch(netmsg_t nmsg)
1803 {
1804         const struct ifnet_array *arr;
1805         struct nd_ifinfo *nd6if;
1806         int i;
1807
1808         ASSERT_NETISR0;
1809
1810         crit_enter();
1811         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1812         crit_exit();
1813
1814         arr = ifnet_array_get();
1815
1816         mtx_lock(&nd6_mtx);
1817         for (i = 0; i < arr->ifnet_count; ++i) {
1818                 struct ifnet *ifp = arr->ifnet_arr[i];
1819
1820                 if (ifp->if_afdata[AF_INET6] == NULL)
1821                         continue;
1822                 nd6if = ND_IFINFO(ifp);
1823                 if (nd6if->basereachable && /* already initialized */
1824                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1825                         /*
1826                          * Since reachable time rarely changes by router
1827                          * advertisements, we SHOULD insure that a new random
1828                          * value gets recomputed at least once every few hours.
1829                          * (RFC 2461, 6.3.4)
1830                          */
1831                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1832                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1833                 }
1834         }
1835         mtx_unlock(&nd6_mtx);
1836
1837         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1838             nd6_slowtimo, NULL);
1839 }
1840
1841 #define gotoerr(e) { error = (e); goto bad;}
1842
1843 int
1844 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1845            struct sockaddr_in6 *dst, struct rtentry *rt)
1846 {
1847         struct llinfo_nd6 *ln = NULL;
1848         int error = 0;
1849
1850         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1851                 goto sendpkt;
1852
1853         if (nd6_need_cache(ifp) == 0)
1854                 goto sendpkt;
1855
1856         /*
1857          * Next hop determination.  This routine is derived from rt_llroute.
1858          */
1859         if (rt != NULL) {
1860                 if (!(rt->rt_flags & RTF_UP)) {
1861                         rt = rtlookup((struct sockaddr *)dst);
1862                         if (rt == NULL)
1863                                 gotoerr(EHOSTUNREACH);
1864                         rt->rt_refcnt--;
1865                         if (rt->rt_ifp != ifp) {
1866                                 /* XXX: loop care? */
1867                                 return nd6_output(ifp, origifp, m, dst, rt);
1868                         }
1869                 }
1870                 if (rt->rt_flags & RTF_GATEWAY) {
1871                         struct sockaddr_in6 *gw6;
1872
1873                         /*
1874                          * We skip link-layer address resolution and NUD
1875                          * if the gateway is not a neighbor from ND point
1876                          * of view, regardless of the value of nd_ifinfo.flags.
1877                          * The second condition is a bit tricky; we skip
1878                          * if the gateway is our own address, which is
1879                          * sometimes used to install a route to a p2p link.
1880                          */
1881                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1882                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1883                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1884                                 /*
1885                                  * We allow this kind of tricky route only
1886                                  * when the outgoing interface is p2p.
1887                                  * XXX: we may need a more generic rule here.
1888                                  */
1889                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1890                                         gotoerr(EHOSTUNREACH);
1891
1892                                 goto sendpkt;
1893                         }
1894
1895                         if (rt->rt_gwroute == NULL) {
1896                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1897                                 if (rt->rt_gwroute == NULL)
1898                                         gotoerr(EHOSTUNREACH);
1899                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1900                                 rtfree(rt->rt_gwroute);
1901                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1902                                 if (rt->rt_gwroute == NULL)
1903                                         gotoerr(EHOSTUNREACH);
1904                         }
1905                         rt = rt->rt_gwroute;
1906                 }
1907         }
1908
1909         /*
1910          * Address resolution or Neighbor Unreachability Detection
1911          * for the next hop.
1912          * At this point, the destination of the packet must be a unicast
1913          * or an anycast address(i.e. not a multicast).
1914          */
1915
1916         /* Look up the neighbor cache for the nexthop */
1917         if (rt && (rt->rt_flags & RTF_LLINFO))
1918                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1919         else {
1920                 /*
1921                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1922                  * the condition below is not very efficient.  But we believe
1923                  * it is tolerable, because this should be a rare case.
1924                  */
1925                 if (nd6_is_addr_neighbor(dst, ifp) &&
1926                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1927                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1928         }
1929         if (!ln || !rt) {
1930                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1931                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1932                         log(LOG_DEBUG,
1933                             "nd6_output: can't allocate llinfo for %s "
1934                             "(ln=%p, rt=%p)\n",
1935                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1936                         gotoerr(EIO);   /* XXX: good error? */
1937                 }
1938
1939                 goto sendpkt;   /* send anyway */
1940         }
1941
1942         /* We don't have to do link-layer address resolution on a p2p link. */
1943         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1944             ln->ln_state < ND6_LLINFO_REACHABLE) {
1945                 ln->ln_state = ND6_LLINFO_STALE;
1946                 ln->ln_expire = time_uptime + nd6_gctimer;
1947         }
1948
1949         /*
1950          * The first time we send a packet to a neighbor whose entry is
1951          * STALE, we have to change the state to DELAY and a sets a timer to
1952          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1953          * neighbor unreachability detection on expiration.
1954          * (RFC 2461 7.3.3)
1955          */
1956         if (ln->ln_state == ND6_LLINFO_STALE) {
1957                 ln->ln_asked = 0;
1958                 ln->ln_state = ND6_LLINFO_DELAY;
1959                 ln->ln_expire = time_uptime + nd6_delay;
1960         }
1961
1962         /*
1963          * If the neighbor cache entry has a state other than INCOMPLETE
1964          * (i.e. its link-layer address is already resolved), just
1965          * send the packet.
1966          */
1967         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1968                 goto sendpkt;
1969
1970         /*
1971          * There is a neighbor cache entry, but no ethernet address
1972          * response yet.  Replace the held mbuf (if any) with this
1973          * latest one.
1974          *
1975          * This code conforms to the rate-limiting rule described in Section
1976          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1977          * an NS below.
1978          */
1979         if (ln->ln_state == ND6_LLINFO_NOSTATE) {
1980                 /*
1981                  * This neighbor cache entry was just created; change its
1982                  * state to INCOMPLETE and start its life cycle.
1983                  *
1984                  * We force an NS output below by setting ln_expire to 1
1985                  * (nd6_rtrequest() could set it to the current time_uptime)
1986                  * and zeroing out ln_asked (XXX this may not be necessary).
1987                  */
1988                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1989                 ln->ln_expire = 1;
1990                 ln->ln_asked = 0;
1991         }
1992         if (ln->ln_hold)
1993                 m_freem(ln->ln_hold);
1994         ln->ln_hold = m;
1995         if (ln->ln_expire) {
1996                 if (ln->ln_asked < nd6_mmaxtries &&
1997                     ln->ln_expire < time_uptime) {
1998                         ln->ln_asked++;
1999                         ln->ln_expire = time_uptime +
2000                                 ND_IFINFO(ifp)->retrans / 1000;
2001                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2002                 }
2003         }
2004         return (0);
2005
2006 sendpkt:
2007         if (ifp->if_flags & IFF_LOOPBACK)
2008                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
2009         else
2010                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
2011         return (error);
2012
2013 bad:
2014         m_freem(m);
2015         return (error);
2016 }
2017 #undef gotoerr
2018
2019 int
2020 nd6_need_cache(struct ifnet *ifp)
2021 {
2022         /*
2023          * XXX: we currently do not make neighbor cache on any interface
2024          * other than Ethernet and GIF.
2025          *
2026          * RFC2893 says:
2027          * - unidirectional tunnels needs no ND
2028          */
2029         switch (ifp->if_type) {
2030         case IFT_ETHER:
2031         case IFT_IEEE1394:
2032 #ifdef IFT_L2VLAN
2033         case IFT_L2VLAN:
2034 #endif
2035 #ifdef IFT_IEEE80211
2036         case IFT_IEEE80211:
2037 #endif
2038 #ifdef IFT_CARP
2039         case IFT_CARP:
2040 #endif
2041         case IFT_GIF:           /* XXX need more cases? */
2042                 return (1);
2043         default:
2044                 return (0);
2045         }
2046 }
2047
2048 int
2049 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2050                 struct sockaddr *dst, u_char *desten)
2051 {
2052         struct sockaddr_dl *sdl;
2053         struct rtentry *rt;
2054
2055
2056         if (m->m_flags & M_MCAST) {
2057                 switch (ifp->if_type) {
2058                 case IFT_ETHER:
2059 #ifdef IFT_L2VLAN
2060         case IFT_L2VLAN:
2061 #endif
2062 #ifdef IFT_IEEE80211
2063                 case IFT_IEEE80211:
2064 #endif
2065                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2066                                                  desten);
2067                         return (1);
2068                 case IFT_IEEE1394:
2069                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2070                         return (1);
2071                 default:
2072                         m_freem(m);
2073                         return (0);
2074                 }
2075         }
2076         if (rt0 == NULL) {
2077                 /* this could happen, if we could not allocate memory */
2078                 m_freem(m);
2079                 return (0);
2080         }
2081         if (rt_llroute(dst, rt0, &rt) != 0) {
2082                 m_freem(m);
2083                 return (0);
2084         }
2085         if (rt->rt_gateway->sa_family != AF_LINK) {
2086                 kprintf("nd6_storelladdr: something odd happens\n");
2087                 m_freem(m);
2088                 return (0);
2089         }
2090         sdl = SDL(rt->rt_gateway);
2091         if (sdl->sdl_alen == 0) {
2092                 /* this should be impossible, but we bark here for debugging */
2093                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2094                 m_freem(m);
2095                 return (0);
2096         }
2097
2098         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2099         return (1);
2100 }
2101
2102 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2103 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2104 #ifdef SYSCTL_DECL
2105 SYSCTL_DECL(_net_inet6_icmp6);
2106 #endif
2107 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2108         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2109 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2110         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2111
2112 static int
2113 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2114 {
2115         int error;
2116         char buf[1024];
2117         struct in6_defrouter *d, *de;
2118         struct nd_defrouter *dr;
2119
2120         if (req->newptr)
2121                 return EPERM;
2122         error = 0;
2123
2124         for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2125              dr = TAILQ_NEXT(dr, dr_entry)) {
2126                 d = (struct in6_defrouter *)buf;
2127                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2128
2129                 if (d + 1 <= de) {
2130                         bzero(d, sizeof(*d));
2131                         d->rtaddr.sin6_family = AF_INET6;
2132                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2133                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2134                             dr->ifp) != 0)
2135                                 log(LOG_ERR,
2136                                     "scope error in "
2137                                     "default router list (%s)\n",
2138                                     ip6_sprintf(&dr->rtaddr));
2139                         d->flags = dr->flags;
2140                         d->rtlifetime = dr->rtlifetime;
2141                         d->expire = dr->expire;
2142                         d->if_index = dr->ifp->if_index;
2143                 } else
2144                         panic("buffer too short");
2145
2146                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2147                 if (error)
2148                         break;
2149         }
2150         return error;
2151 }
2152
2153 static int
2154 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2155 {
2156         int error;
2157         char buf[1024];
2158         struct in6_prefix *p, *pe;
2159         struct nd_prefix *pr;
2160
2161         if (req->newptr)
2162                 return EPERM;
2163         error = 0;
2164
2165         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2166                 u_short advrtrs;
2167                 size_t advance;
2168                 struct sockaddr_in6 *sin6, *s6;
2169                 struct nd_pfxrouter *pfr;
2170
2171                 p = (struct in6_prefix *)buf;
2172                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2173
2174                 if (p + 1 <= pe) {
2175                         bzero(p, sizeof(*p));
2176                         sin6 = (struct sockaddr_in6 *)(p + 1);
2177
2178                         p->prefix = pr->ndpr_prefix;
2179                         if (in6_recoverscope(&p->prefix,
2180                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2181                                 log(LOG_ERR,
2182                                     "scope error in prefix list (%s)\n",
2183                                     ip6_sprintf(&p->prefix.sin6_addr));
2184                         p->raflags = pr->ndpr_raf;
2185                         p->prefixlen = pr->ndpr_plen;
2186                         p->vltime = pr->ndpr_vltime;
2187                         p->pltime = pr->ndpr_pltime;
2188                         p->if_index = pr->ndpr_ifp->if_index;
2189                         p->expire = pr->ndpr_expire;
2190                         p->refcnt = pr->ndpr_refcnt;
2191                         p->flags = pr->ndpr_stateflags;
2192                         p->origin = PR_ORIG_RA;
2193                         advrtrs = 0;
2194                         for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2195                              pfr = pfr->pfr_next) {
2196                                 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2197                                         advrtrs++;
2198                                         continue;
2199                                 }
2200                                 s6 = &sin6[advrtrs];
2201                                 bzero(s6, sizeof(*s6));
2202                                 s6->sin6_family = AF_INET6;
2203                                 s6->sin6_len = sizeof(*sin6);
2204                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2205                                     pfr->router->ifp) != 0)
2206                                         log(LOG_ERR,
2207                                             "scope error in "
2208                                             "prefix list (%s)\n",
2209                                             ip6_sprintf(&pfr->router->rtaddr));
2210                                 advrtrs++;
2211                         }
2212                         p->advrtrs = advrtrs;
2213                 } else {
2214                         panic("buffer too short");
2215                 }
2216
2217                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2218                 error = SYSCTL_OUT(req, buf, advance);
2219                 if (error)
2220                         break;
2221         }
2222         return error;
2223 }