sys/netinet6/nd6.c

   1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
   2 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   3
   4 /*
   5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of the project nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 /*
  34  * XXX
  35  * KAME 970409 note:
  36  * BSD/OS version heavily modifies this code, related to llinfo.
  37  * Since we don't have BSD/OS version of net/route.c in our hand,
  38  * I left the code mostly as it was in 970310.  -- itojun
  39  */
  40
  41 #include "opt_inet.h"
  42 #include "opt_inet6.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/callout.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/socket.h>
  50 #include <sys/sockio.h>
  51 #include <sys/time.h>
  52 #include <sys/kernel.h>
  53 #include <sys/protosw.h>
  54 #include <sys/errno.h>
  55 #include <sys/syslog.h>
  56 #include <sys/queue.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/mutex.h>
  59
  60 #include <sys/thread2.h>
  61 #include <sys/mutex2.h>
  62
  63 #include <net/if.h>
  64 #include <net/if_dl.h>
  65 #include <net/if_types.h>
  66 #include <net/route.h>
  67 #include <net/netisr2.h>
  68 #include <net/netmsg2.h>
  69
  70 #include <netinet/in.h>
  71 #include <netinet/if_ether.h>
  72 #include <netinet6/in6_var.h>
  73 #include <netinet/ip6.h>
  74 #include <netinet6/ip6_var.h>
  75 #include <netinet6/nd6.h>
  76 #include <netinet/icmp6.h>
  77
  78 #include <net/net_osdep.h>
  79
  80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  82
  83 #define SIN6(s) ((struct sockaddr_in6 *)s)
  84 #define SDL(s) ((struct sockaddr_dl *)s)
  85
  86 /* timer values */
  87 int     nd6_prune       = 1;    /* walk list every 1 seconds */
  88 int     nd6_delay       = 5;    /* delay first probe time 5 second */
  89 int     nd6_umaxtries   = 3;    /* maximum unicast query */
  90 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
  91 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
  92 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
  93
  94 /* preventing too many loops in ND option parsing */
  95 int nd6_maxndopt = 10;  /* max # of ND options allowed */
  96
  97 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
  98
  99 #ifdef ND6_DEBUG
 100 int nd6_debug = 1;
 101 #else
 102 int nd6_debug = 0;
 103 #endif
 104
 105 /* for debugging? */
 106 static int nd6_inuse, nd6_allocated;
 107
 108 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 109 struct nd_drhead nd_defrouter;
 110 struct nd_prhead nd_prefix = { 0 };
 111 struct mtx nd6_mtx = MTX_INITIALIZER;
 112
 113 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 114 static struct sockaddr_in6 all1_sa;
 115
 116 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
 117 static void nd6_slowtimo (void *);
 118 static int regen_tmpaddr (struct in6_ifaddr *);
 119
 120 struct callout nd6_slowtimo_ch;
 121 struct callout nd6_timer_ch;
 122 extern struct callout in6_tmpaddrtimer_ch;
 123
 124 void
 125 nd6_init(void)
 126 {
 127         static int nd6_init_done = 0;
 128         int i;
 129
 130         if (nd6_init_done) {
 131                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 132                 return;
 133         }
 134
 135         all1_sa.sin6_family = AF_INET6;
 136         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 137         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 138                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 139
 140         /* initialization of the default router list */
 141         TAILQ_INIT(&nd_defrouter);
 142
 143         nd6_init_done = 1;
 144
 145         /* start timer */
 146         callout_init(&nd6_slowtimo_ch);
 147         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 148             nd6_slowtimo, NULL);
 149 }
 150
 151 struct nd_ifinfo *
 152 nd6_ifattach(struct ifnet *ifp)
 153 {
 154         struct nd_ifinfo *nd;
 155
 156         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
 157             M_WAITOK | M_ZERO);
 158
 159         nd->initialized = 1;
 160
 161         nd->linkmtu = ifp->if_mtu;
 162         nd->chlim = IPV6_DEFHLIM;
 163         nd->basereachable = REACHABLE_TIME;
 164         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 165         nd->retrans = RETRANS_TIMER;
 166         nd->receivedra = 0;
 167
 168         /*
 169          * Note that the default value of ip6_accept_rtadv is 0, which means
 170          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 171          * here.
 172          */
 173         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 174
 175         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 176         nd6_setmtu0(ifp, nd);
 177         return nd;
 178 }
 179
 180 void
 181 nd6_ifdetach(struct nd_ifinfo *nd)
 182 {
 183         kfree(nd, M_IP6NDP);
 184 }
 185
 186 /*
 187  * Reset ND level link MTU. This function is called when the physical MTU
 188  * changes, which means we might have to adjust the ND level MTU.
 189  */
 190 void
 191 nd6_setmtu(struct ifnet *ifp)
 192 {
 193         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 194 }
 195
 196 struct netmsg_nd6setmtu {
 197         struct netmsg_base      nmsg;
 198         struct ifnet            *ifp;
 199         struct nd_ifinfo        *ndi;
 200 };
 201
 202 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 203 static void
 204 nd6_setmtu0_dispatch(netmsg_t msg)
 205 {
 206         struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
 207         struct ifnet *ifp = nmsg->ifp;
 208         struct nd_ifinfo *ndi = nmsg->ndi;
 209         u_long oldmaxmtu;
 210         u_long oldlinkmtu;
 211
 212         oldmaxmtu = ndi->maxmtu;
 213         oldlinkmtu = ndi->linkmtu;
 214
 215         switch (ifp->if_type) {
 216         case IFT_ETHER:
 217                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 218                 break;
 219         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 220                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 221                 break;
 222 #ifdef IFT_IEEE80211
 223         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 224                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 225                 break;
 226 #endif
 227         default:
 228                 ndi->maxmtu = ifp->if_mtu;
 229                 break;
 230         }
 231
 232         if (oldmaxmtu != ndi->maxmtu) {
 233                 /*
 234                  * If the ND level MTU is not set yet, or if the maxmtu
 235                  * is reset to a smaller value than the ND level MTU,
 236                  * also reset the ND level MTU.
 237                  */
 238                 if (ndi->linkmtu == 0 ||
 239                     ndi->maxmtu < ndi->linkmtu) {
 240                         ndi->linkmtu = ndi->maxmtu;
 241                         /* also adjust in6_maxmtu if necessary. */
 242                         if (oldlinkmtu == 0) {
 243                                 /*
 244                                  * XXX: the case analysis is grotty, but
 245                                  * it is not efficient to call in6_setmaxmtu()
 246                                  * here when we are during the initialization
 247                                  * procedure.
 248                                  */
 249                                 if (in6_maxmtu < ndi->linkmtu)
 250                                         in6_maxmtu = ndi->linkmtu;
 251                         } else
 252                                 in6_setmaxmtu();
 253                 }
 254         }
 255 #undef MIN
 256
 257         lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
 258 }
 259
 260 void
 261 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 262 {
 263         struct netmsg_nd6setmtu nmsg;
 264
 265         netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
 266             nd6_setmtu0_dispatch);
 267         nmsg.ifp = ifp;
 268         nmsg.ndi = ndi;
 269         lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
 270 }
 271
 272 void
 273 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 274 {
 275         bzero(ndopts, sizeof(*ndopts));
 276         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 277         ndopts->nd_opts_last
 278                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 279
 280         if (icmp6len == 0) {
 281                 ndopts->nd_opts_done = 1;
 282                 ndopts->nd_opts_search = NULL;
 283         }
 284 }
 285
 286 /*
 287  * Take one ND option.
 288  */
 289 struct nd_opt_hdr *
 290 nd6_option(union nd_opts *ndopts)
 291 {
 292         struct nd_opt_hdr *nd_opt;
 293         int olen;
 294
 295         if (!ndopts)
 296                 panic("ndopts == NULL in nd6_option");
 297         if (!ndopts->nd_opts_last)
 298                 panic("uninitialized ndopts in nd6_option");
 299         if (!ndopts->nd_opts_search)
 300                 return NULL;
 301         if (ndopts->nd_opts_done)
 302                 return NULL;
 303
 304         nd_opt = ndopts->nd_opts_search;
 305
 306         /* make sure nd_opt_len is inside the buffer */
 307         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 308                 bzero(ndopts, sizeof(*ndopts));
 309                 return NULL;
 310         }
 311
 312         olen = nd_opt->nd_opt_len << 3;
 313         if (olen == 0) {
 314                 /*
 315                  * Message validation requires that all included
 316                  * options have a length that is greater than zero.
 317                  */
 318                 bzero(ndopts, sizeof(*ndopts));
 319                 return NULL;
 320         }
 321
 322         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 323         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 324                 /* option overruns the end of buffer, invalid */
 325                 bzero(ndopts, sizeof(*ndopts));
 326                 return NULL;
 327         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 328                 /* reached the end of options chain */
 329                 ndopts->nd_opts_done = 1;
 330                 ndopts->nd_opts_search = NULL;
 331         }
 332         return nd_opt;
 333 }
 334
 335 /*
 336  * Parse multiple ND options.
 337  * This function is much easier to use, for ND routines that do not need
 338  * multiple options of the same type.
 339  */
 340 int
 341 nd6_options(union nd_opts *ndopts)
 342 {
 343         struct nd_opt_hdr *nd_opt;
 344         int i = 0;
 345
 346         if (!ndopts)
 347                 panic("ndopts == NULL in nd6_options");
 348         if (!ndopts->nd_opts_last)
 349                 panic("uninitialized ndopts in nd6_options");
 350         if (!ndopts->nd_opts_search)
 351                 return 0;
 352
 353         while (1) {
 354                 nd_opt = nd6_option(ndopts);
 355                 if (!nd_opt && !ndopts->nd_opts_last) {
 356                         /*
 357                          * Message validation requires that all included
 358                          * options have a length that is greater than zero.
 359                          */
 360                         icmp6stat.icp6s_nd_badopt++;
 361                         bzero(ndopts, sizeof(*ndopts));
 362                         return -1;
 363                 }
 364
 365                 if (!nd_opt)
 366                         goto skip1;
 367
 368                 switch (nd_opt->nd_opt_type) {
 369                 case ND_OPT_SOURCE_LINKADDR:
 370                 case ND_OPT_TARGET_LINKADDR:
 371                 case ND_OPT_MTU:
 372                 case ND_OPT_REDIRECTED_HEADER:
 373                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 374                                 nd6log((LOG_INFO,
 375                                     "duplicated ND6 option found (type=%d)\n",
 376                                     nd_opt->nd_opt_type));
 377                                 /* XXX bark? */
 378                         } else {
 379                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 380                                         = nd_opt;
 381                         }
 382                         break;
 383                 case ND_OPT_PREFIX_INFORMATION:
 384                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 385                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 386                                         = nd_opt;
 387                         }
 388                         ndopts->nd_opts_pi_end =
 389                                 (struct nd_opt_prefix_info *)nd_opt;
 390                         break;
 391                 default:
 392                         /*
 393                          * Unknown options must be silently ignored,
 394                          * to accomodate future extension to the protocol.
 395                          */
 396                         nd6log((LOG_DEBUG,
 397                             "nd6_options: unsupported option %d - "
 398                             "option ignored\n", nd_opt->nd_opt_type));
 399                 }
 400
 401 skip1:
 402                 i++;
 403                 if (i > nd6_maxndopt) {
 404                         icmp6stat.icp6s_nd_toomanyopt++;
 405                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 406                         break;
 407                 }
 408
 409                 if (ndopts->nd_opts_done)
 410                         break;
 411         }
 412
 413         return 0;
 414 }
 415
 416 /*
 417  * ND6 timer routine to expire default route list and prefix list
 418  */
 419 void
 420 nd6_timer(void *ignored_arg)
 421 {
 422         struct llinfo_nd6 *ln;
 423         struct nd_defrouter *dr;
 424         struct nd_prefix *pr;
 425         struct ifnet *ifp;
 426         struct in6_ifaddr *ia6, *nia6;
 427
 428         mtx_lock(&nd6_mtx);
 429         callout_reset(&nd6_timer_ch, nd6_prune * hz,
 430                       nd6_timer, NULL);
 431
 432         ln = llinfo_nd6.ln_next;
 433         while (ln && ln != &llinfo_nd6) {
 434                 struct rtentry *rt;
 435                 struct sockaddr_in6 *dst;
 436                 struct llinfo_nd6 *next = ln->ln_next;
 437                 /* XXX: used for the DELAY case only: */
 438                 struct nd_ifinfo *ndi = NULL;
 439
 440                 if ((rt = ln->ln_rt) == NULL) {
 441                         ln = next;
 442                         continue;
 443                 }
 444                 if ((ifp = rt->rt_ifp) == NULL) {
 445                         ln = next;
 446                         continue;
 447                 }
 448                 ndi = ND_IFINFO(ifp);
 449                 dst = (struct sockaddr_in6 *)rt_key(rt);
 450
 451                 if (ln->ln_expire > time_uptime) {
 452                         ln = next;
 453                         continue;
 454                 }
 455
 456                 /* sanity check */
 457                 if (!rt)
 458                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 459                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 460                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 461                               rt->rt_llinfo, ln);
 462                 if (!dst)
 463                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 464
 465                 switch (ln->ln_state) {
 466                 case ND6_LLINFO_INCOMPLETE:
 467                         if (ln->ln_asked < nd6_mmaxtries) {
 468                                 ln->ln_asked++;
 469                                 ln->ln_expire = time_uptime +
 470                                         ND_IFINFO(ifp)->retrans / 1000;
 471                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 472                                         ln, 0);
 473                         } else {
 474                                 struct mbuf *m = ln->ln_hold;
 475                                 if (m) {
 476                                         if (rt->rt_ifp) {
 477                                                 /*
 478                                                  * Fake rcvif to make ICMP error
 479                                                  * more helpful in diagnosing
 480                                                  * for the receiver.
 481                                                  * XXX: should we consider
 482                                                  * older rcvif?
 483                                                  */
 484                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 485                                         }
 486                                         icmp6_error(m, ICMP6_DST_UNREACH,
 487                                                     ICMP6_DST_UNREACH_ADDR, 0);
 488                                         ln->ln_hold = NULL;
 489                                 }
 490                                 next = nd6_free(rt);
 491                         }
 492                         break;
 493                 case ND6_LLINFO_REACHABLE:
 494                         if (ln->ln_expire) {
 495                                 ln->ln_state = ND6_LLINFO_STALE;
 496                                 ln->ln_expire = time_uptime + nd6_gctimer;
 497                         }
 498                         break;
 499
 500                 case ND6_LLINFO_STALE:
 501                         /* Garbage Collection(RFC 2461 5.3) */
 502                         if (ln->ln_expire)
 503                                 next = nd6_free(rt);
 504                         break;
 505
 506                 case ND6_LLINFO_DELAY:
 507                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 508                                 /* We need NUD */
 509                                 ln->ln_asked = 1;
 510                                 ln->ln_state = ND6_LLINFO_PROBE;
 511                                 ln->ln_expire = time_uptime +
 512                                         ndi->retrans / 1000;
 513                                 nd6_ns_output(ifp, &dst->sin6_addr,
 514                                               &dst->sin6_addr,
 515                                               ln, 0);
 516                         } else {
 517                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 518                                 ln->ln_expire = time_uptime + nd6_gctimer;
 519                         }
 520                         break;
 521                 case ND6_LLINFO_PROBE:
 522                         if (ln->ln_asked < nd6_umaxtries) {
 523                                 ln->ln_asked++;
 524                                 ln->ln_expire = time_uptime +
 525                                         ND_IFINFO(ifp)->retrans / 1000;
 526                                 nd6_ns_output(ifp, &dst->sin6_addr,
 527                                                &dst->sin6_addr, ln, 0);
 528                         } else {
 529                                 next = nd6_free(rt);
 530                         }
 531                         break;
 532                 }
 533                 ln = next;
 534         }
 535
 536         /* expire default router list */
 537         dr = TAILQ_FIRST(&nd_defrouter);
 538         while (dr) {
 539                 if (dr->expire && dr->expire < time_uptime) {
 540                         struct nd_defrouter *t;
 541                         t = TAILQ_NEXT(dr, dr_entry);
 542                         defrtrlist_del(dr);
 543                         dr = t;
 544                 } else {
 545                         dr = TAILQ_NEXT(dr, dr_entry);
 546                 }
 547         }
 548
 549         /*
 550          * expire interface addresses.
 551          * in the past the loop was inside prefix expiry processing.
 552          * However, from a stricter speci-confrmance standpoint, we should
 553          * rather separate address lifetimes and prefix lifetimes.
 554          */
 555 addrloop:
 556         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 557                 nia6 = ia6->ia_next;
 558                 /* check address lifetime */
 559                 if (IFA6_IS_INVALID(ia6)) {
 560                         int regen = 0;
 561
 562                         /*
 563                          * If the expiring address is temporary, try
 564                          * regenerating a new one.  This would be useful when
 565                          * we suspended a laptop PC, then turned it on after a
 566                          * period that could invalidate all temporary
 567                          * addresses.  Although we may have to restart the
 568                          * loop (see below), it must be after purging the
 569                          * address.  Otherwise, we'd see an infinite loop of
 570                          * regeneration.
 571                          */
 572                         if (ip6_use_tempaddr &&
 573                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 574                                 if (regen_tmpaddr(ia6) == 0)
 575                                         regen = 1;
 576                         }
 577
 578                         in6_purgeaddr(&ia6->ia_ifa);
 579
 580                         if (regen)
 581                                 goto addrloop; /* XXX: see below */
 582                 }
 583                 if (IFA6_IS_DEPRECATED(ia6)) {
 584                         int oldflags = ia6->ia6_flags;
 585
 586                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 587
 588                         /*
 589                          * If a temporary address has just become deprecated,
 590                          * regenerate a new one if possible.
 591                          */
 592                         if (ip6_use_tempaddr &&
 593                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 594                             !(oldflags & IN6_IFF_DEPRECATED)) {
 595
 596                                 if (regen_tmpaddr(ia6) == 0) {
 597                                         /*
 598                                          * A new temporary address is
 599                                          * generated.
 600                                          * XXX: this means the address chain
 601                                          * has changed while we are still in
 602                                          * the loop.  Although the change
 603                                          * would not cause disaster (because
 604                                          * it's not a deletion, but an
 605                                          * addition,) we'd rather restart the
 606                                          * loop just for safety.  Or does this
 607                                          * significantly reduce performance??
 608                                          */
 609                                         goto addrloop;
 610                                 }
 611                         }
 612                 } else {
 613                         /*
 614                          * A new RA might have made a deprecated address
 615                          * preferred.
 616                          */
 617                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 618                 }
 619         }
 620
 621         /* expire prefix list */
 622         pr = nd_prefix.lh_first;
 623         while (pr) {
 624                 /*
 625                  * check prefix lifetime.
 626                  * since pltime is just for autoconf, pltime processing for
 627                  * prefix is not necessary.
 628                  */
 629                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
 630                         struct nd_prefix *t;
 631                         t = pr->ndpr_next;
 632
 633                         /*
 634                          * address expiration and prefix expiration are
 635                          * separate.  NEVER perform in6_purgeaddr here.
 636                          */
 637
 638                         prelist_remove(pr);
 639                         pr = t;
 640                 } else
 641                         pr = pr->ndpr_next;
 642         }
 643         mtx_unlock(&nd6_mtx);
 644 }
 645
 646 static int
 647 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 648                                          address */
 649 {
 650         struct ifaddr_container *ifac;
 651         struct ifnet *ifp;
 652         struct in6_ifaddr *public_ifa6 = NULL;
 653
 654         ifp = ia6->ia_ifa.ifa_ifp;
 655         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 656                 struct ifaddr *ifa = ifac->ifa;
 657                 struct in6_ifaddr *it6;
 658
 659                 if (ifa->ifa_addr->sa_family != AF_INET6)
 660                         continue;
 661
 662                 it6 = (struct in6_ifaddr *)ifa;
 663
 664                 /* ignore no autoconf addresses. */
 665                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 666                         continue;
 667
 668                 /* ignore autoconf addresses with different prefixes. */
 669                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 670                         continue;
 671
 672                 /*
 673                  * Now we are looking at an autoconf address with the same
 674                  * prefix as ours.  If the address is temporary and is still
 675                  * preferred, do not create another one.  It would be rare, but
 676                  * could happen, for example, when we resume a laptop PC after
 677                  * a long period.
 678                  */
 679                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 680                     !IFA6_IS_DEPRECATED(it6)) {
 681                         public_ifa6 = NULL;
 682                         break;
 683                 }
 684
 685                 /*
 686                  * This is a public autoconf address that has the same prefix
 687                  * as ours.  If it is preferred, keep it.  We can't break the
 688                  * loop here, because there may be a still-preferred temporary
 689                  * address with the prefix.
 690                  */
 691                 if (!IFA6_IS_DEPRECATED(it6))
 692                     public_ifa6 = it6;
 693         }
 694
 695         if (public_ifa6 != NULL) {
 696                 int e;
 697
 698                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 699                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 700                             " tmp addr,errno=%d\n", e);
 701                         return (-1);
 702                 }
 703                 return (0);
 704         }
 705
 706         return (-1);
 707 }
 708
 709 /*
 710  * Nuke neighbor cache/prefix/default router management table, right before
 711  * ifp goes away.
 712  */
 713 void
 714 nd6_purge(struct ifnet *ifp)
 715 {
 716         struct llinfo_nd6 *ln, *nln;
 717         struct nd_defrouter *dr, *ndr, drany;
 718         struct nd_prefix *pr, *npr;
 719
 720         /* Nuke default router list entries toward ifp */
 721         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 722                 /*
 723                  * The first entry of the list may be stored in
 724                  * the routing table, so we'll delete it later.
 725                  */
 726                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 727                         ndr = TAILQ_NEXT(dr, dr_entry);
 728                         if (dr->ifp == ifp)
 729                                 defrtrlist_del(dr);
 730                 }
 731                 dr = TAILQ_FIRST(&nd_defrouter);
 732                 if (dr->ifp == ifp)
 733                         defrtrlist_del(dr);
 734         }
 735
 736         /* Nuke prefix list entries toward ifp */
 737         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 738                 npr = pr->ndpr_next;
 739                 if (pr->ndpr_ifp == ifp) {
 740                         /*
 741                          * Previously, pr->ndpr_addr is removed as well,
 742                          * but I strongly believe we don't have to do it.
 743                          * nd6_purge() is only called from in6_ifdetach(),
 744                          * which removes all the associated interface addresses
 745                          * by itself.
 746                          * (jinmei@kame.net 20010129)
 747                          */
 748                         prelist_remove(pr);
 749                 }
 750         }
 751
 752         /* cancel default outgoing interface setting */
 753         if (nd6_defifindex == ifp->if_index)
 754                 nd6_setdefaultiface(0);
 755
 756         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 757                 /* refresh default router list */
 758                 bzero(&drany, sizeof(drany));
 759                 defrouter_delreq(&drany, 0);
 760                 defrouter_select();
 761         }
 762
 763         /*
 764          * Nuke neighbor cache entries for the ifp.
 765          * Note that rt->rt_ifp may not be the same as ifp,
 766          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 767          * nd6_rtrequest(), and ip6_input().
 768          */
 769         ln = llinfo_nd6.ln_next;
 770         while (ln && ln != &llinfo_nd6) {
 771                 struct rtentry *rt;
 772                 struct sockaddr_dl *sdl;
 773
 774                 nln = ln->ln_next;
 775                 rt = ln->ln_rt;
 776                 if (rt && rt->rt_gateway &&
 777                     rt->rt_gateway->sa_family == AF_LINK) {
 778                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 779                         if (sdl->sdl_index == ifp->if_index)
 780                                 nln = nd6_free(rt);
 781                 }
 782                 ln = nln;
 783         }
 784 }
 785
 786 struct rtentry *
 787 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 788 {
 789         struct rtentry *rt;
 790         struct sockaddr_in6 sin6;
 791
 792         bzero(&sin6, sizeof(sin6));
 793         sin6.sin6_len = sizeof(struct sockaddr_in6);
 794         sin6.sin6_family = AF_INET6;
 795         sin6.sin6_addr = *addr6;
 796
 797         if (create)
 798                 rt = rtlookup((struct sockaddr *)&sin6);
 799         else
 800                 rt = rtpurelookup((struct sockaddr *)&sin6);
 801         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 802                 /*
 803                  * This is the case for the default route.
 804                  * If we want to create a neighbor cache for the address, we
 805                  * should free the route for the destination and allocate an
 806                  * interface route.
 807                  */
 808                 if (create) {
 809                         --rt->rt_refcnt;
 810                         rt = NULL;
 811                 }
 812         }
 813         if (!rt) {
 814                 if (create && ifp) {
 815                         int e;
 816
 817                         /*
 818                          * If no route is available and create is set,
 819                          * we allocate a host route for the destination
 820                          * and treat it like an interface route.
 821                          * This hack is necessary for a neighbor which can't
 822                          * be covered by our own prefix.
 823                          */
 824                         struct ifaddr *ifa =
 825                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 826                         if (ifa == NULL)
 827                                 return (NULL);
 828
 829                         /*
 830                          * Create a new route.  RTF_LLINFO is necessary
 831                          * to create a Neighbor Cache entry for the
 832                          * destination in nd6_rtrequest which will be
 833                          * called in rtrequest via ifa->ifa_rtrequest.
 834                          */
 835                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 836                                            ifa->ifa_addr,
 837                                            (struct sockaddr *)&all1_sa,
 838                                            (ifa->ifa_flags |
 839                                             RTF_HOST | RTF_LLINFO) &
 840                                            ~RTF_CLONING,
 841                                            &rt)) != 0)
 842                                 log(LOG_ERR,
 843                                     "nd6_lookup: failed to add route for a "
 844                                     "neighbor(%s), errno=%d\n",
 845                                     ip6_sprintf(addr6), e);
 846                         if (rt == NULL)
 847                                 return (NULL);
 848                         if (rt->rt_llinfo) {
 849                                 struct llinfo_nd6 *ln =
 850                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 851                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 852                         }
 853                 } else
 854                         return (NULL);
 855         }
 856         rt->rt_refcnt--;
 857         /*
 858          * Validation for the entry.
 859          * Note that the check for rt_llinfo is necessary because a cloned
 860          * route from a parent route that has the L flag (e.g. the default
 861          * route to a p2p interface) may have the flag, too, while the
 862          * destination is not actually a neighbor.
 863          * XXX: we can't use rt->rt_ifp to check for the interface, since
 864          *      it might be the loopback interface if the entry is for our
 865          *      own address on a non-loopback interface. Instead, we should
 866          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 867          *      interface.
 868          */
 869         if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
 870             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 871             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 872                 if (create) {
 873                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 874                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 875                         /* xxx more logs... kazu */
 876                 }
 877                 return (NULL);
 878         }
 879         return (rt);
 880 }
 881
 882 /*
 883  * Detect if a given IPv6 address identifies a neighbor on a given link.
 884  * XXX: should take care of the destination of a p2p link?
 885  */
 886 int
 887 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 888 {
 889         struct ifaddr_container *ifac;
 890         int i;
 891
 892 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 893 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 894
 895         /*
 896          * A link-local address is always a neighbor.
 897          * XXX: we should use the sin6_scope_id field rather than the embedded
 898          * interface index.
 899          */
 900         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 901             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 902                 return (1);
 903
 904         /*
 905          * If the address matches one of our addresses,
 906          * it should be a neighbor.
 907          */
 908         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 909                 struct ifaddr *ifa = ifac->ifa;
 910
 911                 if (ifa->ifa_addr->sa_family != AF_INET6)
 912                         next: continue;
 913
 914                 for (i = 0; i < 4; i++) {
 915                         if ((IFADDR6(ifa).s6_addr32[i] ^
 916                              addr->sin6_addr.s6_addr32[i]) &
 917                             IFMASK6(ifa).s6_addr32[i])
 918                                 goto next;
 919                 }
 920                 return (1);
 921         }
 922
 923         /*
 924          * Even if the address matches none of our addresses, it might be
 925          * in the neighbor cache.
 926          */
 927         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 928                 return (1);
 929
 930         return (0);
 931 #undef IFADDR6
 932 #undef IFMASK6
 933 }
 934
 935 /*
 936  * Free an nd6 llinfo entry.
 937  */
 938 struct llinfo_nd6 *
 939 nd6_free(struct rtentry *rt)
 940 {
 941         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 942         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 943         struct nd_defrouter *dr;
 944
 945         /*
 946          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
 947          * even though it is not harmful, it was not really necessary.
 948          */
 949
 950         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 951                 mtx_lock(&nd6_mtx);
 952                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 953                                       rt->rt_ifp);
 954
 955                 if (ln->ln_router || dr) {
 956                         /*
 957                          * rt6_flush must be called whether or not the neighbor
 958                          * is in the Default Router List.
 959                          * See a corresponding comment in nd6_na_input().
 960                          */
 961                         rt6_flush(&in6, rt->rt_ifp);
 962                 }
 963
 964                 if (dr) {
 965                         /*
 966                          * Unreachablity of a router might affect the default
 967                          * router selection and on-link detection of advertised
 968                          * prefixes.
 969                          */
 970
 971                         /*
 972                          * Temporarily fake the state to choose a new default
 973                          * router and to perform on-link determination of
 974                          * prefixes correctly.
 975                          * Below the state will be set correctly,
 976                          * or the entry itself will be deleted.
 977                          */
 978                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
 979
 980                         /*
 981                          * Since defrouter_select() does not affect the
 982                          * on-link determination and MIP6 needs the check
 983                          * before the default router selection, we perform
 984                          * the check now.
 985                          */
 986                         pfxlist_onlink_check();
 987
 988                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
 989                                 /*
 990                                  * It is used as the current default router,
 991                                  * so we have to move it to the end of the
 992                                  * list and choose a new one.
 993                                  * XXX: it is not very efficient if this is
 994                                  *      the only router.
 995                                  */
 996                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
 997                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
 998
 999                                 defrouter_select();
1000                         }
1001                 }
1002                 mtx_unlock(&nd6_mtx);
1003         }
1004
1005         /*
1006          * Before deleting the entry, remember the next entry as the
1007          * return value.  We need this because pfxlist_onlink_check() above
1008          * might have freed other entries (particularly the old next entry) as
1009          * a side effect (XXX).
1010          */
1011         next = ln->ln_next;
1012
1013         /*
1014          * Detach the route from the routing tree and the list of neighbor
1015          * caches, and disable the route entry not to be used in already
1016          * cached routes.
1017          */
1018         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1019
1020         return (next);
1021 }
1022
1023 /*
1024  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1025  *
1026  * XXX cost-effective metods?
1027  */
1028 void
1029 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1030 {
1031         struct llinfo_nd6 *ln;
1032
1033         /*
1034          * If the caller specified "rt", use that.  Otherwise, resolve the
1035          * routing table by supplied "dst6".
1036          */
1037         if (!rt) {
1038                 if (!dst6)
1039                         return;
1040                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1041                         return;
1042         }
1043
1044         if ((rt->rt_flags & RTF_GATEWAY) ||
1045             !(rt->rt_flags & RTF_LLINFO) ||
1046             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1047             rt->rt_gateway->sa_family != AF_LINK) {
1048                 /* This is not a host route. */
1049                 return;
1050         }
1051
1052         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1053         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1054                 return;
1055
1056         /*
1057          * if we get upper-layer reachability confirmation many times,
1058          * it is possible we have false information.
1059          */
1060         if (!force) {
1061                 ln->ln_byhint++;
1062                 if (ln->ln_byhint > nd6_maxnudhint)
1063                         return;
1064         }
1065
1066         ln->ln_state = ND6_LLINFO_REACHABLE;
1067         if (ln->ln_expire)
1068                 ln->ln_expire = time_uptime +
1069                         ND_IFINFO(rt->rt_ifp)->reachable;
1070 }
1071
1072 void
1073 nd6_rtrequest(int req, struct rtentry *rt)
1074 {
1075         struct sockaddr *gate = rt->rt_gateway;
1076         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1077         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1078         struct ifnet *ifp = rt->rt_ifp;
1079         struct ifaddr *ifa;
1080
1081         if ((rt->rt_flags & RTF_GATEWAY))
1082                 return;
1083
1084         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1085                 /*
1086                  * This is probably an interface direct route for a link
1087                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1088                  * We do not need special treatment below for such a route.
1089                  * Moreover, the RTF_LLINFO flag which would be set below
1090                  * would annoy the ndp(8) command.
1091                  */
1092                 return;
1093         }
1094
1095         if (req == RTM_RESOLVE &&
1096             (nd6_need_cache(ifp) == 0 || /* stf case */
1097              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1098                 /*
1099                  * FreeBSD and BSD/OS often make a cloned host route based
1100                  * on a less-specific route (e.g. the default route).
1101                  * If the less specific route does not have a "gateway"
1102                  * (this is the case when the route just goes to a p2p or an
1103                  * stf interface), we'll mistakenly make a neighbor cache for
1104                  * the host route, and will see strange neighbor solicitation
1105                  * for the corresponding destination.  In order to avoid the
1106                  * confusion, we check if the destination of the route is
1107                  * a neighbor in terms of neighbor discovery, and stop the
1108                  * process if not.  Additionally, we remove the LLINFO flag
1109                  * so that ndp(8) will not try to get the neighbor information
1110                  * of the destination.
1111                  */
1112                 rt->rt_flags &= ~RTF_LLINFO;
1113                 return;
1114         }
1115
1116         switch (req) {
1117         case RTM_ADD:
1118                 /*
1119                  * There is no backward compatibility :)
1120                  *
1121                  * if (!(rt->rt_flags & RTF_HOST) &&
1122                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1123                  *         rt->rt_flags |= RTF_CLONING;
1124                  */
1125                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1126                         /*
1127                          * Case 1: This route should come from
1128                          * a route to interface.  RTF_LLINFO flag is set
1129                          * for a host route whose destination should be
1130                          * treated as on-link.
1131                          */
1132                         rt_setgate(rt, rt_key(rt),
1133                                    (struct sockaddr *)&null_sdl,
1134                                    RTL_DONTREPORT);
1135                         gate = rt->rt_gateway;
1136                         SDL(gate)->sdl_type = ifp->if_type;
1137                         SDL(gate)->sdl_index = ifp->if_index;
1138                         if (ln)
1139                                 ln->ln_expire = time_uptime;
1140 #if 1
1141                         if (ln && ln->ln_expire == 0) {
1142                                 /* kludge for desktops */
1143 #if 0
1144                                 kprintf("nd6_rtequest: time.tv_sec is zero; "
1145                                        "treat it as 1\n");
1146 #endif
1147                                 ln->ln_expire = 1;
1148                         }
1149 #endif
1150                         if ((rt->rt_flags & RTF_CLONING))
1151                                 break;
1152                 }
1153                 /*
1154                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1155                  * We don't do that here since llinfo is not ready yet.
1156                  *
1157                  * There are also couple of other things to be discussed:
1158                  * - unsolicited NA code needs improvement beforehand
1159                  * - RFC2461 says we MAY send multicast unsolicited NA
1160                  *   (7.2.6 paragraph 4), however, it also says that we
1161                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1162                  *   we don't have anything like it right now.
1163                  *   note that the mechanism needs a mutual agreement
1164                  *   between proxies, which means that we need to implement
1165                  *   a new protocol, or a new kludge.
1166                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1167                  *   we need to check ip6forwarding before sending it.
1168                  *   (or should we allow proxy ND configuration only for
1169                  *   routers?  there's no mention about proxy ND from hosts)
1170                  */
1171 #if 0
1172                 /* XXX it does not work */
1173                 if (rt->rt_flags & RTF_ANNOUNCE)
1174                         nd6_na_output(ifp,
1175                               &SIN6(rt_key(rt))->sin6_addr,
1176                               &SIN6(rt_key(rt))->sin6_addr,
1177                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1178                               1, NULL);
1179 #endif
1180                 /* FALLTHROUGH */
1181         case RTM_RESOLVE:
1182                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1183                         /*
1184                          * Address resolution isn't necessary for a point to
1185                          * point link, so we can skip this test for a p2p link.
1186                          */
1187                         if (gate->sa_family != AF_LINK ||
1188                             gate->sa_len < sizeof(null_sdl)) {
1189                                 log(LOG_DEBUG,
1190                                     "nd6_rtrequest: bad gateway value: %s\n",
1191                                     if_name(ifp));
1192                                 break;
1193                         }
1194                         SDL(gate)->sdl_type = ifp->if_type;
1195                         SDL(gate)->sdl_index = ifp->if_index;
1196                 }
1197                 if (ln != NULL)
1198                         break;  /* This happens on a route change */
1199                 /*
1200                  * Case 2: This route may come from cloning, or a manual route
1201                  * add with a LL address.
1202                  */
1203                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1204                 rt->rt_llinfo = (caddr_t)ln;
1205                 if (!ln) {
1206                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1207                         break;
1208                 }
1209                 nd6_inuse++;
1210                 nd6_allocated++;
1211                 bzero(ln, sizeof(*ln));
1212                 ln->ln_rt = rt;
1213                 /* this is required for "ndp" command. - shin */
1214                 if (req == RTM_ADD) {
1215                         /*
1216                          * gate should have some valid AF_LINK entry,
1217                          * and ln->ln_expire should have some lifetime
1218                          * which is specified by ndp command.
1219                          */
1220                         ln->ln_state = ND6_LLINFO_REACHABLE;
1221                         ln->ln_byhint = 0;
1222                 } else {
1223                         /*
1224                          * When req == RTM_RESOLVE, rt is created and
1225                          * initialized in rtrequest(), so rt_expire is 0.
1226                          */
1227                         ln->ln_state = ND6_LLINFO_NOSTATE;
1228                         ln->ln_expire = time_uptime;
1229                 }
1230                 rt->rt_flags |= RTF_LLINFO;
1231                 ln->ln_next = llinfo_nd6.ln_next;
1232                 llinfo_nd6.ln_next = ln;
1233                 ln->ln_prev = &llinfo_nd6;
1234                 ln->ln_next->ln_prev = ln;
1235
1236                 /*
1237                  * check if rt_key(rt) is one of my address assigned
1238                  * to the interface.
1239                  */
1240                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1241                                           &SIN6(rt_key(rt))->sin6_addr);
1242                 if (ifa) {
1243                         caddr_t macp = nd6_ifptomac(ifp);
1244                         ln->ln_expire = 0;
1245                         ln->ln_state = ND6_LLINFO_REACHABLE;
1246                         ln->ln_byhint = 0;
1247                         if (macp) {
1248                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1249                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1250                         }
1251                         if (nd6_useloopback) {
1252                                 rt->rt_ifp = &loif[0];  /* XXX */
1253                                 /*
1254                                  * Make sure rt_ifa be equal to the ifaddr
1255                                  * corresponding to the address.
1256                                  * We need this because when we refer
1257                                  * rt_ifa->ia6_flags in ip6_input, we assume
1258                                  * that the rt_ifa points to the address instead
1259                                  * of the loopback address.
1260                                  */
1261                                 if (ifa != rt->rt_ifa) {
1262                                         IFAFREE(rt->rt_ifa);
1263                                         IFAREF(ifa);
1264                                         rt->rt_ifa = ifa;
1265                                 }
1266                         }
1267                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1268                         ln->ln_expire = 0;
1269                         ln->ln_state = ND6_LLINFO_REACHABLE;
1270                         ln->ln_byhint = 0;
1271
1272                         /* join solicited node multicast for proxy ND */
1273                         if (ifp->if_flags & IFF_MULTICAST) {
1274                                 struct in6_addr llsol;
1275                                 int error;
1276
1277                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1278                                 llsol.s6_addr16[0] = htons(0xff02);
1279                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1280                                 llsol.s6_addr32[1] = 0;
1281                                 llsol.s6_addr32[2] = htonl(1);
1282                                 llsol.s6_addr8[12] = 0xff;
1283
1284                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1285                                         nd6log((LOG_ERR, "%s: failed to join "
1286                                             "%s (errno=%d)\n", if_name(ifp),
1287                                             ip6_sprintf(&llsol), error));
1288                                 }
1289                         }
1290                 }
1291                 break;
1292
1293         case RTM_DELETE:
1294                 if (!ln)
1295                         break;
1296                 /* leave from solicited node multicast for proxy ND */
1297                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1298                     (ifp->if_flags & IFF_MULTICAST)) {
1299                         struct in6_addr llsol;
1300                         struct in6_multi *in6m;
1301
1302                         llsol = SIN6(rt_key(rt))->sin6_addr;
1303                         llsol.s6_addr16[0] = htons(0xff02);
1304                         llsol.s6_addr16[1] = htons(ifp->if_index);
1305                         llsol.s6_addr32[1] = 0;
1306                         llsol.s6_addr32[2] = htonl(1);
1307                         llsol.s6_addr8[12] = 0xff;
1308
1309                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1310                         if (in6m)
1311                                 in6_delmulti(in6m);
1312                 }
1313                 nd6_inuse--;
1314                 ln->ln_next->ln_prev = ln->ln_prev;
1315                 ln->ln_prev->ln_next = ln->ln_next;
1316                 ln->ln_prev = NULL;
1317                 rt->rt_llinfo = 0;
1318                 rt->rt_flags &= ~RTF_LLINFO;
1319                 if (ln->ln_hold)
1320                         m_freem(ln->ln_hold);
1321                 Free((caddr_t)ln);
1322         }
1323 }
1324
1325 int
1326 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1327 {
1328         struct in6_drlist *drl = (struct in6_drlist *)data;
1329         struct in6_prlist *prl = (struct in6_prlist *)data;
1330         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1331         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1332         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1333         struct nd_defrouter *dr, any;
1334         struct nd_prefix *pr;
1335         struct rtentry *rt;
1336         int i = 0, error = 0;
1337
1338         switch (cmd) {
1339         case SIOCGDRLST_IN6:
1340                 /*
1341                  * obsolete API, use sysctl under net.inet6.icmp6
1342                  */
1343                 bzero(drl, sizeof(*drl));
1344                 mtx_lock(&nd6_mtx);
1345                 dr = TAILQ_FIRST(&nd_defrouter);
1346                 while (dr && i < DRLSTSIZ) {
1347                         drl->defrouter[i].rtaddr = dr->rtaddr;
1348                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1349                                 /* XXX: need to this hack for KAME stack */
1350                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1351                         } else
1352                                 log(LOG_ERR,
1353                                     "default router list contains a "
1354                                     "non-linklocal address(%s)\n",
1355                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1356
1357                         drl->defrouter[i].flags = dr->flags;
1358                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1359                         drl->defrouter[i].expire = dr->expire;
1360                         drl->defrouter[i].if_index = dr->ifp->if_index;
1361                         i++;
1362                         dr = TAILQ_NEXT(dr, dr_entry);
1363                 }
1364                 mtx_unlock(&nd6_mtx);
1365                 break;
1366         case SIOCGPRLST_IN6:
1367                 /*
1368                  * obsolete API, use sysctl under net.inet6.icmp6
1369                  */
1370                 /*
1371                  * XXX meaning of fields, especialy "raflags", is very
1372                  * differnet between RA prefix list and RR/static prefix list.
1373                  * how about separating ioctls into two?
1374                  */
1375                 bzero(prl, sizeof(*prl));
1376                 mtx_lock(&nd6_mtx);
1377                 pr = nd_prefix.lh_first;
1378                 while (pr && i < PRLSTSIZ) {
1379                         struct nd_pfxrouter *pfr;
1380                         int j;
1381
1382                         in6_embedscope(&prl->prefix[i].prefix,
1383                             &pr->ndpr_prefix, NULL, NULL);
1384                         prl->prefix[i].raflags = pr->ndpr_raf;
1385                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1386                         prl->prefix[i].vltime = pr->ndpr_vltime;
1387                         prl->prefix[i].pltime = pr->ndpr_pltime;
1388                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1389                         prl->prefix[i].expire = pr->ndpr_expire;
1390
1391                         pfr = pr->ndpr_advrtrs.lh_first;
1392                         j = 0;
1393                         while (pfr) {
1394                                 if (j < DRLSTSIZ) {
1395 #define RTRADDR prl->prefix[i].advrtr[j]
1396                                         RTRADDR = pfr->router->rtaddr;
1397                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1398                                                 /* XXX: hack for KAME */
1399                                                 RTRADDR.s6_addr16[1] = 0;
1400                                         } else
1401                                                 log(LOG_ERR,
1402                                                     "a router(%s) advertises "
1403                                                     "a prefix with "
1404                                                     "non-link local address\n",
1405                                                     ip6_sprintf(&RTRADDR));
1406 #undef RTRADDR
1407                                 }
1408                                 j++;
1409                                 pfr = pfr->pfr_next;
1410                         }
1411                         prl->prefix[i].advrtrs = j;
1412                         prl->prefix[i].origin = PR_ORIG_RA;
1413
1414                         i++;
1415                         pr = pr->ndpr_next;
1416                 }
1417                 mtx_unlock(&nd6_mtx);
1418
1419                 break;
1420         case OSIOCGIFINFO_IN6:
1421                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1422                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1423                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1424                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1425                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1426                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1427                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1428                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1429                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1430                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1431                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1432                 break;
1433         case SIOCGIFINFO_IN6:
1434                 ndi->ndi = *ND_IFINFO(ifp);
1435                 break;
1436         case SIOCSIFINFO_FLAGS:
1437                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1438                 break;
1439         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1440                 /* flush default router list */
1441                 /*
1442                  * xxx sumikawa: should not delete route if default
1443                  * route equals to the top of default router list
1444                  */
1445                 bzero(&any, sizeof(any));
1446                 defrouter_delreq(&any, 0);
1447                 defrouter_select();
1448                 /* xxx sumikawa: flush prefix list */
1449                 break;
1450         case SIOCSPFXFLUSH_IN6:
1451             {
1452                 /* flush all the prefix advertised by routers */
1453                 struct nd_prefix *pr, *next;
1454
1455                 mtx_lock(&nd6_mtx);
1456                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1457                         struct in6_ifaddr *ia, *ia_next;
1458
1459                         next = pr->ndpr_next;
1460
1461                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1462                                 continue; /* XXX */
1463
1464                         /* do we really have to remove addresses as well? */
1465                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1466                                 /* ia might be removed.  keep the next ptr. */
1467                                 ia_next = ia->ia_next;
1468
1469                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1470                                         continue;
1471
1472                                 if (ia->ia6_ndpr == pr)
1473                                         in6_purgeaddr(&ia->ia_ifa);
1474                         }
1475                         prelist_remove(pr);
1476                 }
1477                 mtx_unlock(&nd6_mtx);
1478                 break;
1479             }
1480         case SIOCSRTRFLUSH_IN6:
1481             {
1482                 /* flush all the default routers */
1483                 struct nd_defrouter *dr, *next;
1484
1485                 mtx_lock(&nd6_mtx);
1486                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1487                         /*
1488                          * The first entry of the list may be stored in
1489                          * the routing table, so we'll delete it later.
1490                          */
1491                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1492                                 next = TAILQ_NEXT(dr, dr_entry);
1493                                 defrtrlist_del(dr);
1494                         }
1495                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1496                 }
1497                 mtx_unlock(&nd6_mtx);
1498                 break;
1499             }
1500         case SIOCGNBRINFO_IN6:
1501             {
1502                 struct llinfo_nd6 *ln;
1503                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1504
1505                 /*
1506                  * XXX: KAME specific hack for scoped addresses
1507                  *      XXXX: for other scopes than link-local?
1508                  */
1509                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1510                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1511                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1512
1513                         if (*idp == 0)
1514                                 *idp = htons(ifp->if_index);
1515                 }
1516
1517                 mtx_lock(&nd6_mtx);
1518                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1519                         error = EINVAL;
1520                         mtx_unlock(&nd6_mtx);
1521                         break;
1522                 }
1523                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1524                 nbi->state = ln->ln_state;
1525                 nbi->asked = ln->ln_asked;
1526                 nbi->isrouter = ln->ln_router;
1527                 nbi->expire = ln->ln_expire;
1528                 mtx_unlock(&nd6_mtx);
1529
1530                 break;
1531             }
1532         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1533                 ndif->ifindex = nd6_defifindex;
1534                 break;
1535         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1536                 return (nd6_setdefaultiface(ndif->ifindex));
1537                 break;
1538         }
1539         return (error);
1540 }
1541
1542 /*
1543  * Create neighbor cache entry and cache link-layer address,
1544  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1545  */
1546 struct rtentry *
1547 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1548                  int lladdrlen,
1549                  int type,      /* ICMP6 type */
1550                  int code       /* type dependent information */)
1551 {
1552         struct rtentry *rt = NULL;
1553         struct llinfo_nd6 *ln = NULL;
1554         int is_newentry;
1555         struct sockaddr_dl *sdl = NULL;
1556         int do_update;
1557         int olladdr;
1558         int llchange;
1559         int newstate = 0;
1560
1561         if (!ifp)
1562                 panic("ifp == NULL in nd6_cache_lladdr");
1563         if (!from)
1564                 panic("from == NULL in nd6_cache_lladdr");
1565
1566         /* nothing must be updated for unspecified address */
1567         if (IN6_IS_ADDR_UNSPECIFIED(from))
1568                 return NULL;
1569
1570         /*
1571          * Validation about ifp->if_addrlen and lladdrlen must be done in
1572          * the caller.
1573          *
1574          * XXX If the link does not have link-layer adderss, what should
1575          * we do? (ifp->if_addrlen == 0)
1576          * Spec says nothing in sections for RA, RS and NA.  There's small
1577          * description on it in NS section (RFC 2461 7.2.3).
1578          */
1579
1580         rt = nd6_lookup(from, 0, ifp);
1581         if (!rt) {
1582 #if 0
1583                 /* nothing must be done if there's no lladdr */
1584                 if (!lladdr || !lladdrlen)
1585                         return NULL;
1586 #endif
1587
1588                 rt = nd6_lookup(from, 1, ifp);
1589                 is_newentry = 1;
1590         } else {
1591                 /* do nothing if static ndp is set */
1592                 if (rt->rt_flags & RTF_STATIC)
1593                         return NULL;
1594                 is_newentry = 0;
1595         }
1596
1597         if (!rt)
1598                 return NULL;
1599         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1600 fail:
1601                 nd6_free(rt);
1602                 return NULL;
1603         }
1604         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1605         if (!ln)
1606                 goto fail;
1607         if (!rt->rt_gateway)
1608                 goto fail;
1609         if (rt->rt_gateway->sa_family != AF_LINK)
1610                 goto fail;
1611         sdl = SDL(rt->rt_gateway);
1612
1613         olladdr = (sdl->sdl_alen) ? 1 : 0;
1614         if (olladdr && lladdr) {
1615                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1616                         llchange = 1;
1617                 else
1618                         llchange = 0;
1619         } else
1620                 llchange = 0;
1621
1622         /*
1623          * newentry olladdr  lladdr  llchange   (*=record)
1624          *      0       n       n       --      (1)
1625          *      0       y       n       --      (2)
1626          *      0       n       y       --      (3) * STALE
1627          *      0       y       y       n       (4) *
1628          *      0       y       y       y       (5) * STALE
1629          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1630          *      1       --      y       --      (7) * STALE
1631          */
1632
1633         if (lladdr) {           /* (3-5) and (7) */
1634                 /*
1635                  * Record source link-layer address
1636                  * XXX is it dependent to ifp->if_type?
1637                  */
1638                 sdl->sdl_alen = ifp->if_addrlen;
1639                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1640         }
1641
1642         if (!is_newentry) {
1643                 if ((!olladdr && lladdr)                /* (3) */
1644                  || (olladdr && lladdr && llchange)) {  /* (5) */
1645                         do_update = 1;
1646                         newstate = ND6_LLINFO_STALE;
1647                 } else                                  /* (1-2,4) */
1648                         do_update = 0;
1649         } else {
1650                 do_update = 1;
1651                 if (!lladdr)                            /* (6) */
1652                         newstate = ND6_LLINFO_NOSTATE;
1653                 else                                    /* (7) */
1654                         newstate = ND6_LLINFO_STALE;
1655         }
1656
1657         if (do_update) {
1658                 /*
1659                  * Update the state of the neighbor cache.
1660                  */
1661                 ln->ln_state = newstate;
1662
1663                 if (ln->ln_state == ND6_LLINFO_STALE) {
1664                         /*
1665                          * XXX: since nd6_output() below will cause
1666                          * state tansition to DELAY and reset the timer,
1667                          * we must set the timer now, although it is actually
1668                          * meaningless.
1669                          */
1670                         ln->ln_expire = time_uptime + nd6_gctimer;
1671
1672                         if (ln->ln_hold) {
1673                                 /*
1674                                  * we assume ifp is not a p2p here, so just
1675                                  * set the 2nd argument as the 1st one.
1676                                  */
1677                                 nd6_output(ifp, ifp, ln->ln_hold,
1678                                            (struct sockaddr_in6 *)rt_key(rt),
1679                                            rt);
1680                                 ln->ln_hold = NULL;
1681                         }
1682                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1683                         /* probe right away */
1684                         ln->ln_expire = time_uptime;
1685                 }
1686         }
1687
1688         /*
1689          * ICMP6 type dependent behavior.
1690          *
1691          * NS: clear IsRouter if new entry
1692          * RS: clear IsRouter
1693          * RA: set IsRouter if there's lladdr
1694          * redir: clear IsRouter if new entry
1695          *
1696          * RA case, (1):
1697          * The spec says that we must set IsRouter in the following cases:
1698          * - If lladdr exist, set IsRouter.  This means (1-5).
1699          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1700          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1701          * A quetion arises for (1) case.  (1) case has no lladdr in the
1702          * neighbor cache, this is similar to (6).
1703          * This case is rare but we figured that we MUST NOT set IsRouter.
1704          *
1705          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1706          *                                                      D R
1707          *      0       n       n       --      (1)     c   ?     s
1708          *      0       y       n       --      (2)     c   s     s
1709          *      0       n       y       --      (3)     c   s     s
1710          *      0       y       y       n       (4)     c   s     s
1711          *      0       y       y       y       (5)     c   s     s
1712          *      1       --      n       --      (6) c   c       c s
1713          *      1       --      y       --      (7) c   c   s   c s
1714          *
1715          *                                      (c=clear s=set)
1716          */
1717         switch (type & 0xff) {
1718         case ND_NEIGHBOR_SOLICIT:
1719                 /*
1720                  * New entry must have is_router flag cleared.
1721                  */
1722                 if (is_newentry)        /* (6-7) */
1723                         ln->ln_router = 0;
1724                 break;
1725         case ND_REDIRECT:
1726                 /*
1727                  * If the icmp is a redirect to a better router, always set the
1728                  * is_router flag. Otherwise, if the entry is newly created,
1729                  * clear the flag. [RFC 2461, sec 8.3]
1730                  */
1731                 if (code == ND_REDIRECT_ROUTER)
1732                         ln->ln_router = 1;
1733                 else if (is_newentry) /* (6-7) */
1734                         ln->ln_router = 0;
1735                 break;
1736         case ND_ROUTER_SOLICIT:
1737                 /*
1738                  * is_router flag must always be cleared.
1739                  */
1740                 ln->ln_router = 0;
1741                 break;
1742         case ND_ROUTER_ADVERT:
1743                 /*
1744                  * Mark an entry with lladdr as a router.
1745                  */
1746                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1747                  || (is_newentry && lladdr)) {                  /* (7) */
1748                         ln->ln_router = 1;
1749                 }
1750                 break;
1751         }
1752
1753         /*
1754          * When the link-layer address of a router changes, select the
1755          * best router again.  In particular, when the neighbor entry is newly
1756          * created, it might affect the selection policy.
1757          * Question: can we restrict the first condition to the "is_newentry"
1758          * case?
1759          * XXX: when we hear an RA from a new router with the link-layer
1760          * address option, defrouter_select() is called twice, since
1761          * defrtrlist_update called the function as well.  However, I believe
1762          * we can compromise the overhead, since it only happens the first
1763          * time.
1764          * XXX: although defrouter_select() should not have a bad effect
1765          * for those are not autoconfigured hosts, we explicitly avoid such
1766          * cases for safety.
1767          */
1768         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1769                 defrouter_select();
1770
1771         return rt;
1772 }
1773
1774 static void
1775 nd6_slowtimo(void *ignored_arg)
1776 {
1777         struct nd_ifinfo *nd6if;
1778         struct ifnet *ifp;
1779
1780         mtx_lock(&nd6_mtx);
1781         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1782                         nd6_slowtimo, NULL);
1783         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1784                 if (ifp->if_afdata[AF_INET6] == NULL)
1785                         continue;
1786                 nd6if = ND_IFINFO(ifp);
1787                 if (nd6if->basereachable && /* already initialized */
1788                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1789                         /*
1790                          * Since reachable time rarely changes by router
1791                          * advertisements, we SHOULD insure that a new random
1792                          * value gets recomputed at least once every few hours.
1793                          * (RFC 2461, 6.3.4)
1794                          */
1795                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1796                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1797                 }
1798         }
1799         mtx_unlock(&nd6_mtx);
1800 }
1801
1802 #define gotoerr(e) { error = (e); goto bad;}
1803
1804 int
1805 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1806            struct sockaddr_in6 *dst, struct rtentry *rt)
1807 {
1808         struct llinfo_nd6 *ln = NULL;
1809         int error = 0;
1810
1811         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1812                 goto sendpkt;
1813
1814         if (nd6_need_cache(ifp) == 0)
1815                 goto sendpkt;
1816
1817         /*
1818          * next hop determination.  This routine is derived from ether_outpout.
1819          */
1820         if (rt != NULL) {
1821                 if (!(rt->rt_flags & RTF_UP)) {
1822                         rt = rtlookup((struct sockaddr *)dst);
1823                         if (rt == NULL)
1824                                 gotoerr(EHOSTUNREACH);
1825                         rt->rt_refcnt--;
1826                         if (rt->rt_ifp != ifp) {
1827                                 /* XXX: loop care? */
1828                                 return nd6_output(ifp, origifp, m, dst, rt);
1829                         }
1830                 }
1831                 if (rt->rt_flags & RTF_GATEWAY) {
1832                         struct sockaddr_in6 *gw6;
1833
1834                         /*
1835                          * We skip link-layer address resolution and NUD
1836                          * if the gateway is not a neighbor from ND point
1837                          * of view, regardless of the value of nd_ifinfo.flags.
1838                          * The second condition is a bit tricky; we skip
1839                          * if the gateway is our own address, which is
1840                          * sometimes used to install a route to a p2p link.
1841                          */
1842                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1843                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1844                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1845                                 /*
1846                                  * We allow this kind of tricky route only
1847                                  * when the outgoing interface is p2p.
1848                                  * XXX: we may need a more generic rule here.
1849                                  */
1850                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1851                                         gotoerr(EHOSTUNREACH);
1852
1853                                 goto sendpkt;
1854                         }
1855
1856                         if (rt->rt_gwroute == NULL) {
1857                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1858                                 if (rt->rt_gwroute == NULL)
1859                                         gotoerr(EHOSTUNREACH);
1860                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1861                                 rtfree(rt->rt_gwroute);
1862                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1863                                 if (rt->rt_gwroute == NULL)
1864                                         gotoerr(EHOSTUNREACH);
1865                         }
1866                 }
1867         }
1868
1869         /*
1870          * Address resolution or Neighbor Unreachability Detection
1871          * for the next hop.
1872          * At this point, the destination of the packet must be a unicast
1873          * or an anycast address(i.e. not a multicast).
1874          */
1875
1876         /* Look up the neighbor cache for the nexthop */
1877         if (rt && (rt->rt_flags & RTF_LLINFO))
1878                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1879         else {
1880                 /*
1881                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1882                  * the condition below is not very efficient.  But we believe
1883                  * it is tolerable, because this should be a rare case.
1884                  */
1885                 if (nd6_is_addr_neighbor(dst, ifp) &&
1886                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1887                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1888         }
1889         if (!ln || !rt) {
1890                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1891                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1892                         log(LOG_DEBUG,
1893                             "nd6_output: can't allocate llinfo for %s "
1894                             "(ln=%p, rt=%p)\n",
1895                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1896                         gotoerr(EIO);   /* XXX: good error? */
1897                 }
1898
1899                 goto sendpkt;   /* send anyway */
1900         }
1901
1902         /* We don't have to do link-layer address resolution on a p2p link. */
1903         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1904             ln->ln_state < ND6_LLINFO_REACHABLE) {
1905                 ln->ln_state = ND6_LLINFO_STALE;
1906                 ln->ln_expire = time_uptime + nd6_gctimer;
1907         }
1908
1909         /*
1910          * The first time we send a packet to a neighbor whose entry is
1911          * STALE, we have to change the state to DELAY and a sets a timer to
1912          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1913          * neighbor unreachability detection on expiration.
1914          * (RFC 2461 7.3.3)
1915          */
1916         if (ln->ln_state == ND6_LLINFO_STALE) {
1917                 ln->ln_asked = 0;
1918                 ln->ln_state = ND6_LLINFO_DELAY;
1919                 ln->ln_expire = time_uptime + nd6_delay;
1920         }
1921
1922         /*
1923          * If the neighbor cache entry has a state other than INCOMPLETE
1924          * (i.e. its link-layer address is already resolved), just
1925          * send the packet.
1926          */
1927         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1928                 goto sendpkt;
1929
1930         /*
1931          * There is a neighbor cache entry, but no ethernet address
1932          * response yet.  Replace the held mbuf (if any) with this
1933          * latest one.
1934          *
1935          * This code conforms to the rate-limiting rule described in Section
1936          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1937          * an NS below.
1938          */
1939         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1940                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1941         if (ln->ln_hold)
1942                 m_freem(ln->ln_hold);
1943         ln->ln_hold = m;
1944         if (ln->ln_expire) {
1945                 if (ln->ln_asked < nd6_mmaxtries &&
1946                     ln->ln_expire < time_uptime) {
1947                         ln->ln_asked++;
1948                         ln->ln_expire = time_uptime +
1949                                 ND_IFINFO(ifp)->retrans / 1000;
1950                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1951                 }
1952         }
1953         return (0);
1954
1955 sendpkt:
1956         if (ifp->if_flags & IFF_LOOPBACK)
1957                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1958         else
1959                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1960         return (error);
1961
1962 bad:
1963         m_freem(m);
1964         return (error);
1965 }
1966 #undef gotoerr
1967
1968 int
1969 nd6_need_cache(struct ifnet *ifp)
1970 {
1971         /*
1972          * XXX: we currently do not make neighbor cache on any interface
1973          * other than Ethernet and GIF.
1974          *
1975          * RFC2893 says:
1976          * - unidirectional tunnels needs no ND
1977          */
1978         switch (ifp->if_type) {
1979         case IFT_ETHER:
1980         case IFT_IEEE1394:
1981 #ifdef IFT_L2VLAN
1982         case IFT_L2VLAN:
1983 #endif
1984 #ifdef IFT_IEEE80211
1985         case IFT_IEEE80211:
1986 #endif
1987 #ifdef IFT_CARP
1988         case IFT_CARP:
1989 #endif
1990         case IFT_GIF:           /* XXX need more cases? */
1991                 return (1);
1992         default:
1993                 return (0);
1994         }
1995 }
1996
1997 int
1998 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1999                 struct sockaddr *dst, u_char *desten)
2000 {
2001         struct sockaddr_dl *sdl;
2002         struct rtentry *rt;
2003
2004
2005         if (m->m_flags & M_MCAST) {
2006                 switch (ifp->if_type) {
2007                 case IFT_ETHER:
2008 #ifdef IFT_L2VLAN
2009         case IFT_L2VLAN:
2010 #endif
2011 #ifdef IFT_IEEE80211
2012                 case IFT_IEEE80211:
2013 #endif
2014                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2015                                                  desten);
2016                         return (1);
2017                 case IFT_IEEE1394:
2018                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2019                         return (1);
2020                 default:
2021                         m_freem(m);
2022                         return (0);
2023                 }
2024         }
2025         if (rt0 == NULL) {
2026                 /* this could happen, if we could not allocate memory */
2027                 m_freem(m);
2028                 return (0);
2029         }
2030         if (rt_llroute(dst, rt0, &rt) != 0) {
2031                 m_freem(m);
2032                 return (0);
2033         }
2034         if (rt->rt_gateway->sa_family != AF_LINK) {
2035                 kprintf("nd6_storelladdr: something odd happens\n");
2036                 m_freem(m);
2037                 return (0);
2038         }
2039         sdl = SDL(rt->rt_gateway);
2040         if (sdl->sdl_alen == 0) {
2041                 /* this should be impossible, but we bark here for debugging */
2042                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2043                 m_freem(m);
2044                 return (0);
2045         }
2046
2047         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2048         return (1);
2049 }
2050
2051 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2052 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2053 #ifdef SYSCTL_DECL
2054 SYSCTL_DECL(_net_inet6_icmp6);
2055 #endif
2056 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2057         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2058 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2059         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2060
2061 static int
2062 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2063 {
2064         int error;
2065         char buf[1024];
2066         struct in6_defrouter *d, *de;
2067         struct nd_defrouter *dr;
2068
2069         if (req->newptr)
2070                 return EPERM;
2071         error = 0;
2072
2073         for (dr = TAILQ_FIRST(&nd_defrouter);
2074              dr;
2075              dr = TAILQ_NEXT(dr, dr_entry)) {
2076                 d = (struct in6_defrouter *)buf;
2077                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2078
2079                 if (d + 1 <= de) {
2080                         bzero(d, sizeof(*d));
2081                         d->rtaddr.sin6_family = AF_INET6;
2082                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2083                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2084                             dr->ifp) != 0)
2085                                 log(LOG_ERR,
2086                                     "scope error in "
2087                                     "default router list (%s)\n",
2088                                     ip6_sprintf(&dr->rtaddr));
2089                         d->flags = dr->flags;
2090                         d->rtlifetime = dr->rtlifetime;
2091                         d->expire = dr->expire;
2092                         d->if_index = dr->ifp->if_index;
2093                 } else
2094                         panic("buffer too short");
2095
2096                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2097                 if (error)
2098                         break;
2099         }
2100         return error;
2101 }
2102
2103 static int
2104 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2105 {
2106         int error;
2107         char buf[1024];
2108         struct in6_prefix *p, *pe;
2109         struct nd_prefix *pr;
2110
2111         if (req->newptr)
2112                 return EPERM;
2113         error = 0;
2114
2115         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2116                 u_short advrtrs;
2117                 size_t advance;
2118                 struct sockaddr_in6 *sin6, *s6;
2119                 struct nd_pfxrouter *pfr;
2120
2121                 p = (struct in6_prefix *)buf;
2122                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2123
2124                 if (p + 1 <= pe) {
2125                         bzero(p, sizeof(*p));
2126                         sin6 = (struct sockaddr_in6 *)(p + 1);
2127
2128                         p->prefix = pr->ndpr_prefix;
2129                         if (in6_recoverscope(&p->prefix,
2130                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2131                                 log(LOG_ERR,
2132                                     "scope error in prefix list (%s)\n",
2133                                     ip6_sprintf(&p->prefix.sin6_addr));
2134                         p->raflags = pr->ndpr_raf;
2135                         p->prefixlen = pr->ndpr_plen;
2136                         p->vltime = pr->ndpr_vltime;
2137                         p->pltime = pr->ndpr_pltime;
2138                         p->if_index = pr->ndpr_ifp->if_index;
2139                         p->expire = pr->ndpr_expire;
2140                         p->refcnt = pr->ndpr_refcnt;
2141                         p->flags = pr->ndpr_stateflags;
2142                         p->origin = PR_ORIG_RA;
2143                         advrtrs = 0;
2144                         for (pfr = pr->ndpr_advrtrs.lh_first;
2145                              pfr;
2146                              pfr = pfr->pfr_next) {
2147                                 if ((void *)&sin6[advrtrs + 1] >
2148                                     (void *)pe) {
2149                                         advrtrs++;
2150                                         continue;
2151                                 }
2152                                 s6 = &sin6[advrtrs];
2153                                 bzero(s6, sizeof(*s6));
2154                                 s6->sin6_family = AF_INET6;
2155                                 s6->sin6_len = sizeof(*sin6);
2156                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2157                                                      pfr->router->ifp) != 0)
2158                                         log(LOG_ERR,
2159                                             "scope error in "
2160                                             "prefix list (%s)\n",
2161                                             ip6_sprintf(&pfr->router->rtaddr));
2162                                 advrtrs++;
2163                         }
2164                         p->advrtrs = advrtrs;
2165                 } else
2166                         panic("buffer too short");
2167
2168                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2169                 error = SYSCTL_OUT(req, buf, advance);
2170                 if (error)
2171                         break;
2172         }
2173         return error;
2174 }