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("nd6");
 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 int regen_tmpaddr (struct in6_ifaddr *);
 118 static void nd6_slowtimo(void *);
 119 static void nd6_slowtimo_dispatch(netmsg_t);
 120 static void nd6_timer(void *);
 121 static void nd6_timer_dispatch(netmsg_t);
 122
 123 static struct callout nd6_slowtimo_ch;
 124 static struct netmsg_base nd6_slowtimo_netmsg;
 125
 126 static struct callout nd6_timer_ch;
 127 static struct netmsg_base nd6_timer_netmsg;
 128
 129 void
 130 nd6_init(void)
 131 {
 132         static int nd6_init_done = 0;
 133         int i;
 134
 135         if (nd6_init_done) {
 136                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 137                 return;
 138         }
 139
 140         all1_sa.sin6_family = AF_INET6;
 141         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 142         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 143                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 144
 145         /* initialization of the default router list */
 146         TAILQ_INIT(&nd_defrouter);
 147
 148         nd6_init_done = 1;
 149
 150         /* start timer */
 151         callout_init_mp(&nd6_slowtimo_ch);
 152         netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
 153             MSGF_PRIORITY, nd6_slowtimo_dispatch);
 154         callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 155             nd6_slowtimo, NULL, 0);
 156 }
 157
 158 struct nd_ifinfo *
 159 nd6_ifattach(struct ifnet *ifp)
 160 {
 161         struct nd_ifinfo *nd;
 162
 163         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
 164             M_WAITOK | M_ZERO);
 165
 166         nd->initialized = 1;
 167
 168         nd->linkmtu = ifp->if_mtu;
 169         nd->chlim = IPV6_DEFHLIM;
 170         nd->basereachable = REACHABLE_TIME;
 171         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 172         nd->retrans = RETRANS_TIMER;
 173         nd->receivedra = 0;
 174
 175         /*
 176          * Note that the default value of ip6_accept_rtadv is 0, which means
 177          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
 178          * here.
 179          */
 180         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
 181
 182         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 183         nd6_setmtu0(ifp, nd);
 184         return nd;
 185 }
 186
 187 void
 188 nd6_ifdetach(struct nd_ifinfo *nd)
 189 {
 190         kfree(nd, M_IP6NDP);
 191 }
 192
 193 /*
 194  * Reset ND level link MTU. This function is called when the physical MTU
 195  * changes, which means we might have to adjust the ND level MTU.
 196  */
 197 void
 198 nd6_setmtu(struct ifnet *ifp)
 199 {
 200         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 201 }
 202
 203 struct netmsg_nd6setmtu {
 204         struct netmsg_base      nmsg;
 205         struct ifnet            *ifp;
 206         struct nd_ifinfo        *ndi;
 207 };
 208
 209 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 210 static void
 211 nd6_setmtu0_dispatch(netmsg_t msg)
 212 {
 213         struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
 214         struct ifnet *ifp = nmsg->ifp;
 215         struct nd_ifinfo *ndi = nmsg->ndi;
 216         u_long oldmaxmtu;
 217         u_long oldlinkmtu;
 218
 219         oldmaxmtu = ndi->maxmtu;
 220         oldlinkmtu = ndi->linkmtu;
 221
 222         switch (ifp->if_type) {
 223         case IFT_ETHER:
 224                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 225                 break;
 226         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 227                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 228                 break;
 229 #ifdef IFT_IEEE80211
 230         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 231                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 232                 break;
 233 #endif
 234         default:
 235                 ndi->maxmtu = ifp->if_mtu;
 236                 break;
 237         }
 238
 239         if (oldmaxmtu != ndi->maxmtu) {
 240                 /*
 241                  * If the ND level MTU is not set yet, or if the maxmtu
 242                  * is reset to a smaller value than the ND level MTU,
 243                  * also reset the ND level MTU.
 244                  */
 245                 if (ndi->linkmtu == 0 ||
 246                     ndi->maxmtu < ndi->linkmtu) {
 247                         ndi->linkmtu = ndi->maxmtu;
 248                         /* also adjust in6_maxmtu if necessary. */
 249                         if (oldlinkmtu == 0) {
 250                                 /*
 251                                  * XXX: the case analysis is grotty, but
 252                                  * it is not efficient to call in6_setmaxmtu()
 253                                  * here when we are during the initialization
 254                                  * procedure.
 255                                  */
 256                                 if (in6_maxmtu < ndi->linkmtu)
 257                                         in6_maxmtu = ndi->linkmtu;
 258                         } else
 259                                 in6_setmaxmtu();
 260                 }
 261         }
 262 #undef MIN
 263
 264         lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
 265 }
 266
 267 void
 268 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 269 {
 270         struct netmsg_nd6setmtu nmsg;
 271
 272         netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
 273             nd6_setmtu0_dispatch);
 274         nmsg.ifp = ifp;
 275         nmsg.ndi = ndi;
 276         lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
 277 }
 278
 279 void
 280 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 281 {
 282         bzero(ndopts, sizeof(*ndopts));
 283         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 284         ndopts->nd_opts_last
 285                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 286
 287         if (icmp6len == 0) {
 288                 ndopts->nd_opts_done = 1;
 289                 ndopts->nd_opts_search = NULL;
 290         }
 291 }
 292
 293 /*
 294  * Take one ND option.
 295  */
 296 struct nd_opt_hdr *
 297 nd6_option(union nd_opts *ndopts)
 298 {
 299         struct nd_opt_hdr *nd_opt;
 300         int olen;
 301
 302         if (!ndopts)
 303                 panic("ndopts == NULL in nd6_option");
 304         if (!ndopts->nd_opts_last)
 305                 panic("uninitialized ndopts in nd6_option");
 306         if (!ndopts->nd_opts_search)
 307                 return NULL;
 308         if (ndopts->nd_opts_done)
 309                 return NULL;
 310
 311         nd_opt = ndopts->nd_opts_search;
 312
 313         /* make sure nd_opt_len is inside the buffer */
 314         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 315                 bzero(ndopts, sizeof(*ndopts));
 316                 return NULL;
 317         }
 318
 319         olen = nd_opt->nd_opt_len << 3;
 320         if (olen == 0) {
 321                 /*
 322                  * Message validation requires that all included
 323                  * options have a length that is greater than zero.
 324                  */
 325                 bzero(ndopts, sizeof(*ndopts));
 326                 return NULL;
 327         }
 328
 329         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 330         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 331                 /* option overruns the end of buffer, invalid */
 332                 bzero(ndopts, sizeof(*ndopts));
 333                 return NULL;
 334         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 335                 /* reached the end of options chain */
 336                 ndopts->nd_opts_done = 1;
 337                 ndopts->nd_opts_search = NULL;
 338         }
 339         return nd_opt;
 340 }
 341
 342 /*
 343  * Parse multiple ND options.
 344  * This function is much easier to use, for ND routines that do not need
 345  * multiple options of the same type.
 346  */
 347 int
 348 nd6_options(union nd_opts *ndopts)
 349 {
 350         struct nd_opt_hdr *nd_opt;
 351         int i = 0;
 352
 353         if (!ndopts)
 354                 panic("ndopts == NULL in nd6_options");
 355         if (!ndopts->nd_opts_last)
 356                 panic("uninitialized ndopts in nd6_options");
 357         if (!ndopts->nd_opts_search)
 358                 return 0;
 359
 360         while (1) {
 361                 nd_opt = nd6_option(ndopts);
 362                 if (!nd_opt && !ndopts->nd_opts_last) {
 363                         /*
 364                          * Message validation requires that all included
 365                          * options have a length that is greater than zero.
 366                          */
 367                         icmp6stat.icp6s_nd_badopt++;
 368                         bzero(ndopts, sizeof(*ndopts));
 369                         return -1;
 370                 }
 371
 372                 if (!nd_opt)
 373                         goto skip1;
 374
 375                 switch (nd_opt->nd_opt_type) {
 376                 case ND_OPT_SOURCE_LINKADDR:
 377                 case ND_OPT_TARGET_LINKADDR:
 378                 case ND_OPT_MTU:
 379                 case ND_OPT_REDIRECTED_HEADER:
 380                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 381                                 nd6log((LOG_INFO,
 382                                     "duplicated ND6 option found (type=%d)\n",
 383                                     nd_opt->nd_opt_type));
 384                                 /* XXX bark? */
 385                         } else {
 386                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 387                                         = nd_opt;
 388                         }
 389                         break;
 390                 case ND_OPT_PREFIX_INFORMATION:
 391                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 392                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 393                                         = nd_opt;
 394                         }
 395                         ndopts->nd_opts_pi_end =
 396                                 (struct nd_opt_prefix_info *)nd_opt;
 397                         break;
 398                 default:
 399                         /*
 400                          * Unknown options must be silently ignored,
 401                          * to accomodate future extension to the protocol.
 402                          */
 403                         nd6log((LOG_DEBUG,
 404                             "nd6_options: unsupported option %d - "
 405                             "option ignored\n", nd_opt->nd_opt_type));
 406                 }
 407
 408 skip1:
 409                 i++;
 410                 if (i > nd6_maxndopt) {
 411                         icmp6stat.icp6s_nd_toomanyopt++;
 412                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 413                         break;
 414                 }
 415
 416                 if (ndopts->nd_opts_done)
 417                         break;
 418         }
 419
 420         return 0;
 421 }
 422
 423 /*
 424  * ND6 timer routine to expire default route list and prefix list
 425  */
 426 static void
 427 nd6_timer_dispatch(netmsg_t nmsg)
 428 {
 429         struct llinfo_nd6 *ln;
 430         struct nd_defrouter *dr;
 431         struct nd_prefix *pr;
 432         struct ifnet *ifp;
 433         struct in6_ifaddr *ia6, *nia6;
 434
 435         KASSERT(&curthread->td_msgport == netisr_cpuport(0),
 436             ("not in netisr0"));
 437
 438         crit_enter();
 439         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
 440         crit_exit();
 441
 442         mtx_lock(&nd6_mtx);
 443
 444         ln = llinfo_nd6.ln_next;
 445         while (ln && ln != &llinfo_nd6) {
 446                 struct rtentry *rt;
 447                 struct sockaddr_in6 *dst;
 448                 struct llinfo_nd6 *next = ln->ln_next;
 449                 /* XXX: used for the DELAY case only: */
 450                 struct nd_ifinfo *ndi = NULL;
 451
 452                 if ((rt = ln->ln_rt) == NULL) {
 453                         ln = next;
 454                         continue;
 455                 }
 456                 if ((ifp = rt->rt_ifp) == NULL) {
 457                         ln = next;
 458                         continue;
 459                 }
 460                 ndi = ND_IFINFO(ifp);
 461                 dst = (struct sockaddr_in6 *)rt_key(rt);
 462
 463                 if (ln->ln_expire > time_uptime) {
 464                         ln = next;
 465                         continue;
 466                 }
 467
 468                 /* sanity check */
 469                 if (!rt)
 470                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 471                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 472                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 473                               rt->rt_llinfo, ln);
 474                 if (!dst)
 475                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 476
 477                 switch (ln->ln_state) {
 478                 case ND6_LLINFO_INCOMPLETE:
 479                         if (ln->ln_asked < nd6_mmaxtries) {
 480                                 ln->ln_asked++;
 481                                 ln->ln_expire = time_uptime +
 482                                         ND_IFINFO(ifp)->retrans / 1000;
 483                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 484                                         ln, 0);
 485                         } else {
 486                                 struct mbuf *m = ln->ln_hold;
 487                                 if (m) {
 488                                         if (rt->rt_ifp) {
 489                                                 /*
 490                                                  * Fake rcvif to make ICMP error
 491                                                  * more helpful in diagnosing
 492                                                  * for the receiver.
 493                                                  * XXX: should we consider
 494                                                  * older rcvif?
 495                                                  */
 496                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 497                                         }
 498                                         icmp6_error(m, ICMP6_DST_UNREACH,
 499                                                     ICMP6_DST_UNREACH_ADDR, 0);
 500                                         ln->ln_hold = NULL;
 501                                 }
 502                                 next = nd6_free(rt);
 503                         }
 504                         break;
 505                 case ND6_LLINFO_REACHABLE:
 506                         if (ln->ln_expire) {
 507                                 ln->ln_state = ND6_LLINFO_STALE;
 508                                 ln->ln_expire = time_uptime + nd6_gctimer;
 509                         }
 510                         break;
 511
 512                 case ND6_LLINFO_STALE:
 513                         /* Garbage Collection(RFC 2461 5.3) */
 514                         if (ln->ln_expire)
 515                                 next = nd6_free(rt);
 516                         break;
 517
 518                 case ND6_LLINFO_DELAY:
 519                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 520                                 /* We need NUD */
 521                                 ln->ln_asked = 1;
 522                                 ln->ln_state = ND6_LLINFO_PROBE;
 523                                 ln->ln_expire = time_uptime +
 524                                         ndi->retrans / 1000;
 525                                 nd6_ns_output(ifp, &dst->sin6_addr,
 526                                               &dst->sin6_addr,
 527                                               ln, 0);
 528                         } else {
 529                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 530                                 ln->ln_expire = time_uptime + nd6_gctimer;
 531                         }
 532                         break;
 533                 case ND6_LLINFO_PROBE:
 534                         if (ln->ln_asked < nd6_umaxtries) {
 535                                 ln->ln_asked++;
 536                                 ln->ln_expire = time_uptime +
 537                                         ND_IFINFO(ifp)->retrans / 1000;
 538                                 nd6_ns_output(ifp, &dst->sin6_addr,
 539                                                &dst->sin6_addr, ln, 0);
 540                         } else {
 541                                 next = nd6_free(rt);
 542                         }
 543                         break;
 544                 }
 545                 ln = next;
 546         }
 547
 548         /* expire default router list */
 549         dr = TAILQ_FIRST(&nd_defrouter);
 550         while (dr) {
 551                 if (dr->expire && dr->expire < time_uptime) {
 552                         struct nd_defrouter *t;
 553                         t = TAILQ_NEXT(dr, dr_entry);
 554                         defrtrlist_del(dr);
 555                         dr = t;
 556                 } else {
 557                         dr = TAILQ_NEXT(dr, dr_entry);
 558                 }
 559         }
 560
 561         /*
 562          * expire interface addresses.
 563          * in the past the loop was inside prefix expiry processing.
 564          * However, from a stricter speci-confrmance standpoint, we should
 565          * rather separate address lifetimes and prefix lifetimes.
 566          */
 567 addrloop:
 568         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 569                 nia6 = ia6->ia_next;
 570                 /* check address lifetime */
 571                 if (IFA6_IS_INVALID(ia6)) {
 572                         int regen = 0;
 573
 574                         /*
 575                          * If the expiring address is temporary, try
 576                          * regenerating a new one.  This would be useful when
 577                          * we suspended a laptop PC, then turned it on after a
 578                          * period that could invalidate all temporary
 579                          * addresses.  Although we may have to restart the
 580                          * loop (see below), it must be after purging the
 581                          * address.  Otherwise, we'd see an infinite loop of
 582                          * regeneration.
 583                          */
 584                         if (ip6_use_tempaddr &&
 585                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 586                                 if (regen_tmpaddr(ia6) == 0)
 587                                         regen = 1;
 588                         }
 589
 590                         in6_purgeaddr(&ia6->ia_ifa);
 591
 592                         if (regen)
 593                                 goto addrloop; /* XXX: see below */
 594                 }
 595                 if (IFA6_IS_DEPRECATED(ia6)) {
 596                         int oldflags = ia6->ia6_flags;
 597
 598                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 599
 600                         /*
 601                          * If a temporary address has just become deprecated,
 602                          * regenerate a new one if possible.
 603                          */
 604                         if (ip6_use_tempaddr &&
 605                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 606                             !(oldflags & IN6_IFF_DEPRECATED)) {
 607
 608                                 if (regen_tmpaddr(ia6) == 0) {
 609                                         /*
 610                                          * A new temporary address is
 611                                          * generated.
 612                                          * XXX: this means the address chain
 613                                          * has changed while we are still in
 614                                          * the loop.  Although the change
 615                                          * would not cause disaster (because
 616                                          * it's not a deletion, but an
 617                                          * addition,) we'd rather restart the
 618                                          * loop just for safety.  Or does this
 619                                          * significantly reduce performance??
 620                                          */
 621                                         goto addrloop;
 622                                 }
 623                         }
 624                 } else {
 625                         /*
 626                          * A new RA might have made a deprecated address
 627                          * preferred.
 628                          */
 629                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 630                 }
 631         }
 632
 633         /* expire prefix list */
 634         pr = nd_prefix.lh_first;
 635         while (pr) {
 636                 /*
 637                  * check prefix lifetime.
 638                  * since pltime is just for autoconf, pltime processing for
 639                  * prefix is not necessary.
 640                  */
 641                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
 642                         struct nd_prefix *t;
 643                         t = pr->ndpr_next;
 644
 645                         /*
 646                          * address expiration and prefix expiration are
 647                          * separate.  NEVER perform in6_purgeaddr here.
 648                          */
 649
 650                         prelist_remove(pr);
 651                         pr = t;
 652                 } else
 653                         pr = pr->ndpr_next;
 654         }
 655
 656         mtx_unlock(&nd6_mtx);
 657
 658         callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
 659 }
 660
 661 static void
 662 nd6_timer(void *arg __unused)
 663 {
 664         struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
 665
 666         KASSERT(mycpuid == 0, ("not on cpu0"));
 667         crit_enter();
 668         if (lmsg->ms_flags & MSGF_DONE)
 669                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
 670         crit_exit();
 671 }
 672
 673 void
 674 nd6_timer_init(void)
 675 {
 676         callout_init_mp(&nd6_timer_ch);
 677         netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
 678             MSGF_PRIORITY, nd6_timer_dispatch);
 679         callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
 680 }
 681
 682 static int
 683 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 684                                          address */
 685 {
 686         struct ifaddr_container *ifac;
 687         struct ifnet *ifp;
 688         struct in6_ifaddr *public_ifa6 = NULL;
 689
 690         ifp = ia6->ia_ifa.ifa_ifp;
 691         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 692                 struct ifaddr *ifa = ifac->ifa;
 693                 struct in6_ifaddr *it6;
 694
 695                 if (ifa->ifa_addr->sa_family != AF_INET6)
 696                         continue;
 697
 698                 it6 = (struct in6_ifaddr *)ifa;
 699
 700                 /* ignore no autoconf addresses. */
 701                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 702                         continue;
 703
 704                 /* ignore autoconf addresses with different prefixes. */
 705                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 706                         continue;
 707
 708                 /*
 709                  * Now we are looking at an autoconf address with the same
 710                  * prefix as ours.  If the address is temporary and is still
 711                  * preferred, do not create another one.  It would be rare, but
 712                  * could happen, for example, when we resume a laptop PC after
 713                  * a long period.
 714                  */
 715                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 716                     !IFA6_IS_DEPRECATED(it6)) {
 717                         public_ifa6 = NULL;
 718                         break;
 719                 }
 720
 721                 /*
 722                  * This is a public autoconf address that has the same prefix
 723                  * as ours.  If it is preferred, keep it.  We can't break the
 724                  * loop here, because there may be a still-preferred temporary
 725                  * address with the prefix.
 726                  */
 727                 if (!IFA6_IS_DEPRECATED(it6))
 728                     public_ifa6 = it6;
 729         }
 730
 731         if (public_ifa6 != NULL) {
 732                 int e;
 733
 734                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 735                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 736                             " tmp addr,errno=%d\n", e);
 737                         return (-1);
 738                 }
 739                 return (0);
 740         }
 741
 742         return (-1);
 743 }
 744
 745 /*
 746  * Nuke neighbor cache/prefix/default router management table, right before
 747  * ifp goes away.
 748  */
 749 void
 750 nd6_purge(struct ifnet *ifp)
 751 {
 752         struct llinfo_nd6 *ln, *nln;
 753         struct nd_defrouter *dr, *ndr, drany;
 754         struct nd_prefix *pr, *npr;
 755
 756         /* Nuke default router list entries toward ifp */
 757         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 758                 /*
 759                  * The first entry of the list may be stored in
 760                  * the routing table, so we'll delete it later.
 761                  */
 762                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 763                         ndr = TAILQ_NEXT(dr, dr_entry);
 764                         if (dr->ifp == ifp)
 765                                 defrtrlist_del(dr);
 766                 }
 767                 dr = TAILQ_FIRST(&nd_defrouter);
 768                 if (dr->ifp == ifp)
 769                         defrtrlist_del(dr);
 770         }
 771
 772         /* Nuke prefix list entries toward ifp */
 773         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 774                 npr = pr->ndpr_next;
 775                 if (pr->ndpr_ifp == ifp) {
 776                         /*
 777                          * Previously, pr->ndpr_addr is removed as well,
 778                          * but I strongly believe we don't have to do it.
 779                          * nd6_purge() is only called from in6_ifdetach(),
 780                          * which removes all the associated interface addresses
 781                          * by itself.
 782                          * (jinmei@kame.net 20010129)
 783                          */
 784                         prelist_remove(pr);
 785                 }
 786         }
 787
 788         /* cancel default outgoing interface setting */
 789         if (nd6_defifindex == ifp->if_index)
 790                 nd6_setdefaultiface(0);
 791
 792         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 793                 /* refresh default router list */
 794                 bzero(&drany, sizeof(drany));
 795                 defrouter_delreq(&drany, 0);
 796                 defrouter_select();
 797         }
 798
 799         /*
 800          * Nuke neighbor cache entries for the ifp.
 801          * Note that rt->rt_ifp may not be the same as ifp,
 802          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 803          * nd6_rtrequest(), and ip6_input().
 804          */
 805         ln = llinfo_nd6.ln_next;
 806         while (ln && ln != &llinfo_nd6) {
 807                 struct rtentry *rt;
 808                 struct sockaddr_dl *sdl;
 809
 810                 nln = ln->ln_next;
 811                 rt = ln->ln_rt;
 812                 if (rt && rt->rt_gateway &&
 813                     rt->rt_gateway->sa_family == AF_LINK) {
 814                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 815                         if (sdl->sdl_index == ifp->if_index)
 816                                 nln = nd6_free(rt);
 817                 }
 818                 ln = nln;
 819         }
 820 }
 821
 822 struct rtentry *
 823 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 824 {
 825         struct rtentry *rt;
 826         struct sockaddr_in6 sin6;
 827
 828         bzero(&sin6, sizeof(sin6));
 829         sin6.sin6_len = sizeof(struct sockaddr_in6);
 830         sin6.sin6_family = AF_INET6;
 831         sin6.sin6_addr = *addr6;
 832
 833         if (create)
 834                 rt = rtlookup((struct sockaddr *)&sin6);
 835         else
 836                 rt = rtpurelookup((struct sockaddr *)&sin6);
 837         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 838                 /*
 839                  * This is the case for the default route.
 840                  * If we want to create a neighbor cache for the address, we
 841                  * should free the route for the destination and allocate an
 842                  * interface route.
 843                  */
 844                 if (create) {
 845                         --rt->rt_refcnt;
 846                         rt = NULL;
 847                 }
 848         }
 849         if (!rt) {
 850                 if (create && ifp) {
 851                         int e;
 852
 853                         /*
 854                          * If no route is available and create is set,
 855                          * we allocate a host route for the destination
 856                          * and treat it like an interface route.
 857                          * This hack is necessary for a neighbor which can't
 858                          * be covered by our own prefix.
 859                          */
 860                         struct ifaddr *ifa =
 861                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 862                         if (ifa == NULL)
 863                                 return (NULL);
 864
 865                         /*
 866                          * Create a new route.  RTF_LLINFO is necessary
 867                          * to create a Neighbor Cache entry for the
 868                          * destination in nd6_rtrequest which will be
 869                          * called in rtrequest via ifa->ifa_rtrequest.
 870                          */
 871                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 872                                            ifa->ifa_addr,
 873                                            (struct sockaddr *)&all1_sa,
 874                                            (ifa->ifa_flags |
 875                                             RTF_HOST | RTF_LLINFO) &
 876                                            ~RTF_CLONING,
 877                                            &rt)) != 0)
 878                                 log(LOG_ERR,
 879                                     "nd6_lookup: failed to add route for a "
 880                                     "neighbor(%s), errno=%d\n",
 881                                     ip6_sprintf(addr6), e);
 882                         if (rt == NULL)
 883                                 return (NULL);
 884                         if (rt->rt_llinfo) {
 885                                 struct llinfo_nd6 *ln =
 886                                         (struct llinfo_nd6 *)rt->rt_llinfo;
 887                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 888                         }
 889                 } else
 890                         return (NULL);
 891         }
 892         rt->rt_refcnt--;
 893         /*
 894          * Validation for the entry.
 895          * Note that the check for rt_llinfo is necessary because a cloned
 896          * route from a parent route that has the L flag (e.g. the default
 897          * route to a p2p interface) may have the flag, too, while the
 898          * destination is not actually a neighbor.
 899          * XXX: we can't use rt->rt_ifp to check for the interface, since
 900          *      it might be the loopback interface if the entry is for our
 901          *      own address on a non-loopback interface. Instead, we should
 902          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
 903          *      interface.
 904          */
 905         if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
 906             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
 907             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
 908                 if (create) {
 909                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
 910                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 911                         /* xxx more logs... kazu */
 912                 }
 913                 return (NULL);
 914         }
 915         return (rt);
 916 }
 917
 918 /*
 919  * Detect if a given IPv6 address identifies a neighbor on a given link.
 920  * XXX: should take care of the destination of a p2p link?
 921  */
 922 int
 923 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 924 {
 925         struct ifaddr_container *ifac;
 926         int i;
 927
 928 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 929 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 930
 931         /*
 932          * A link-local address is always a neighbor.
 933          * XXX: we should use the sin6_scope_id field rather than the embedded
 934          * interface index.
 935          */
 936         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 937             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 938                 return (1);
 939
 940         /*
 941          * If the address matches one of our addresses,
 942          * it should be a neighbor.
 943          */
 944         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 945                 struct ifaddr *ifa = ifac->ifa;
 946
 947                 if (ifa->ifa_addr->sa_family != AF_INET6)
 948                         next: continue;
 949
 950                 for (i = 0; i < 4; i++) {
 951                         if ((IFADDR6(ifa).s6_addr32[i] ^
 952                              addr->sin6_addr.s6_addr32[i]) &
 953                             IFMASK6(ifa).s6_addr32[i])
 954                                 goto next;
 955                 }
 956                 return (1);
 957         }
 958
 959         /*
 960          * Even if the address matches none of our addresses, it might be
 961          * in the neighbor cache.
 962          */
 963         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
 964                 return (1);
 965
 966         return (0);
 967 #undef IFADDR6
 968 #undef IFMASK6
 969 }
 970
 971 /*
 972  * Free an nd6 llinfo entry.
 973  */
 974 struct llinfo_nd6 *
 975 nd6_free(struct rtentry *rt)
 976 {
 977         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
 978         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
 979         struct nd_defrouter *dr;
 980
 981         /*
 982          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
 983          * even though it is not harmful, it was not really necessary.
 984          */
 985
 986         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
 987                 mtx_lock(&nd6_mtx);
 988                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
 989                                       rt->rt_ifp);
 990
 991                 if (ln->ln_router || dr) {
 992                         /*
 993                          * rt6_flush must be called whether or not the neighbor
 994                          * is in the Default Router List.
 995                          * See a corresponding comment in nd6_na_input().
 996                          */
 997                         rt6_flush(&in6, rt->rt_ifp);
 998                 }
 999
1000                 if (dr) {
1001                         /*
1002                          * Unreachablity of a router might affect the default
1003                          * router selection and on-link detection of advertised
1004                          * prefixes.
1005                          */
1006
1007                         /*
1008                          * Temporarily fake the state to choose a new default
1009                          * router and to perform on-link determination of
1010                          * prefixes correctly.
1011                          * Below the state will be set correctly,
1012                          * or the entry itself will be deleted.
1013                          */
1014                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
1015
1016                         /*
1017                          * Since defrouter_select() does not affect the
1018                          * on-link determination and MIP6 needs the check
1019                          * before the default router selection, we perform
1020                          * the check now.
1021                          */
1022                         pfxlist_onlink_check();
1023
1024                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1025                                 /*
1026                                  * It is used as the current default router,
1027                                  * so we have to move it to the end of the
1028                                  * list and choose a new one.
1029                                  * XXX: it is not very efficient if this is
1030                                  *      the only router.
1031                                  */
1032                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1033                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1034
1035                                 defrouter_select();
1036                         }
1037                 }
1038                 mtx_unlock(&nd6_mtx);
1039         }
1040
1041         /*
1042          * Before deleting the entry, remember the next entry as the
1043          * return value.  We need this because pfxlist_onlink_check() above
1044          * might have freed other entries (particularly the old next entry) as
1045          * a side effect (XXX).
1046          */
1047         next = ln->ln_next;
1048
1049         /*
1050          * Detach the route from the routing tree and the list of neighbor
1051          * caches, and disable the route entry not to be used in already
1052          * cached routes.
1053          */
1054         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1055
1056         return (next);
1057 }
1058
1059 /*
1060  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1061  *
1062  * XXX cost-effective metods?
1063  */
1064 void
1065 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1066 {
1067         struct llinfo_nd6 *ln;
1068
1069         /*
1070          * If the caller specified "rt", use that.  Otherwise, resolve the
1071          * routing table by supplied "dst6".
1072          */
1073         if (!rt) {
1074                 if (!dst6)
1075                         return;
1076                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1077                         return;
1078         }
1079
1080         if ((rt->rt_flags & RTF_GATEWAY) ||
1081             !(rt->rt_flags & RTF_LLINFO) ||
1082             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1083             rt->rt_gateway->sa_family != AF_LINK) {
1084                 /* This is not a host route. */
1085                 return;
1086         }
1087
1088         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1089         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1090                 return;
1091
1092         /*
1093          * if we get upper-layer reachability confirmation many times,
1094          * it is possible we have false information.
1095          */
1096         if (!force) {
1097                 ln->ln_byhint++;
1098                 if (ln->ln_byhint > nd6_maxnudhint)
1099                         return;
1100         }
1101
1102         ln->ln_state = ND6_LLINFO_REACHABLE;
1103         if (ln->ln_expire)
1104                 ln->ln_expire = time_uptime +
1105                         ND_IFINFO(rt->rt_ifp)->reachable;
1106 }
1107
1108 void
1109 nd6_rtrequest(int req, struct rtentry *rt)
1110 {
1111         struct sockaddr *gate = rt->rt_gateway;
1112         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1113         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1114         struct ifnet *ifp = rt->rt_ifp;
1115         struct ifaddr *ifa;
1116
1117         if ((rt->rt_flags & RTF_GATEWAY))
1118                 return;
1119
1120         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1121                 /*
1122                  * This is probably an interface direct route for a link
1123                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1124                  * We do not need special treatment below for such a route.
1125                  * Moreover, the RTF_LLINFO flag which would be set below
1126                  * would annoy the ndp(8) command.
1127                  */
1128                 return;
1129         }
1130
1131         if (req == RTM_RESOLVE &&
1132             (nd6_need_cache(ifp) == 0 || /* stf case */
1133              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1134                 /*
1135                  * FreeBSD and BSD/OS often make a cloned host route based
1136                  * on a less-specific route (e.g. the default route).
1137                  * If the less specific route does not have a "gateway"
1138                  * (this is the case when the route just goes to a p2p or an
1139                  * stf interface), we'll mistakenly make a neighbor cache for
1140                  * the host route, and will see strange neighbor solicitation
1141                  * for the corresponding destination.  In order to avoid the
1142                  * confusion, we check if the destination of the route is
1143                  * a neighbor in terms of neighbor discovery, and stop the
1144                  * process if not.  Additionally, we remove the LLINFO flag
1145                  * so that ndp(8) will not try to get the neighbor information
1146                  * of the destination.
1147                  */
1148                 rt->rt_flags &= ~RTF_LLINFO;
1149                 return;
1150         }
1151
1152         switch (req) {
1153         case RTM_ADD:
1154                 /*
1155                  * There is no backward compatibility :)
1156                  *
1157                  * if (!(rt->rt_flags & RTF_HOST) &&
1158                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1159                  *         rt->rt_flags |= RTF_CLONING;
1160                  */
1161                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1162                         /*
1163                          * Case 1: This route should come from
1164                          * a route to interface.  RTF_LLINFO flag is set
1165                          * for a host route whose destination should be
1166                          * treated as on-link.
1167                          */
1168                         rt_setgate(rt, rt_key(rt),
1169                                    (struct sockaddr *)&null_sdl,
1170                                    RTL_DONTREPORT);
1171                         gate = rt->rt_gateway;
1172                         SDL(gate)->sdl_type = ifp->if_type;
1173                         SDL(gate)->sdl_index = ifp->if_index;
1174                         if (ln)
1175                                 ln->ln_expire = time_uptime;
1176 #if 1
1177                         if (ln && ln->ln_expire == 0) {
1178                                 /* kludge for desktops */
1179 #if 0
1180                                 kprintf("nd6_rtequest: time.tv_sec is zero; "
1181                                        "treat it as 1\n");
1182 #endif
1183                                 ln->ln_expire = 1;
1184                         }
1185 #endif
1186                         if ((rt->rt_flags & RTF_CLONING))
1187                                 break;
1188                 }
1189                 /*
1190                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1191                  * We don't do that here since llinfo is not ready yet.
1192                  *
1193                  * There are also couple of other things to be discussed:
1194                  * - unsolicited NA code needs improvement beforehand
1195                  * - RFC2461 says we MAY send multicast unsolicited NA
1196                  *   (7.2.6 paragraph 4), however, it also says that we
1197                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1198                  *   we don't have anything like it right now.
1199                  *   note that the mechanism needs a mutual agreement
1200                  *   between proxies, which means that we need to implement
1201                  *   a new protocol, or a new kludge.
1202                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1203                  *   we need to check ip6forwarding before sending it.
1204                  *   (or should we allow proxy ND configuration only for
1205                  *   routers?  there's no mention about proxy ND from hosts)
1206                  */
1207 #if 0
1208                 /* XXX it does not work */
1209                 if (rt->rt_flags & RTF_ANNOUNCE)
1210                         nd6_na_output(ifp,
1211                               &SIN6(rt_key(rt))->sin6_addr,
1212                               &SIN6(rt_key(rt))->sin6_addr,
1213                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1214                               1, NULL);
1215 #endif
1216                 /* FALLTHROUGH */
1217         case RTM_RESOLVE:
1218                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1219                         /*
1220                          * Address resolution isn't necessary for a point to
1221                          * point link, so we can skip this test for a p2p link.
1222                          */
1223                         if (gate->sa_family != AF_LINK ||
1224                             gate->sa_len < sizeof(null_sdl)) {
1225                                 log(LOG_DEBUG,
1226                                     "nd6_rtrequest: bad gateway value: %s\n",
1227                                     if_name(ifp));
1228                                 break;
1229                         }
1230                         SDL(gate)->sdl_type = ifp->if_type;
1231                         SDL(gate)->sdl_index = ifp->if_index;
1232                 }
1233                 if (ln != NULL)
1234                         break;  /* This happens on a route change */
1235                 /*
1236                  * Case 2: This route may come from cloning, or a manual route
1237                  * add with a LL address.
1238                  */
1239                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1240                 rt->rt_llinfo = (caddr_t)ln;
1241                 if (!ln) {
1242                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1243                         break;
1244                 }
1245                 nd6_inuse++;
1246                 nd6_allocated++;
1247                 bzero(ln, sizeof(*ln));
1248                 ln->ln_rt = rt;
1249                 /* this is required for "ndp" command. - shin */
1250                 if (req == RTM_ADD) {
1251                         /*
1252                          * gate should have some valid AF_LINK entry,
1253                          * and ln->ln_expire should have some lifetime
1254                          * which is specified by ndp command.
1255                          */
1256                         ln->ln_state = ND6_LLINFO_REACHABLE;
1257                         ln->ln_byhint = 0;
1258                 } else {
1259                         /*
1260                          * When req == RTM_RESOLVE, rt is created and
1261                          * initialized in rtrequest(), so rt_expire is 0.
1262                          */
1263                         ln->ln_state = ND6_LLINFO_NOSTATE;
1264                         ln->ln_expire = time_uptime;
1265                 }
1266                 rt->rt_flags |= RTF_LLINFO;
1267                 ln->ln_next = llinfo_nd6.ln_next;
1268                 llinfo_nd6.ln_next = ln;
1269                 ln->ln_prev = &llinfo_nd6;
1270                 ln->ln_next->ln_prev = ln;
1271
1272                 /*
1273                  * check if rt_key(rt) is one of my address assigned
1274                  * to the interface.
1275                  */
1276                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1277                                           &SIN6(rt_key(rt))->sin6_addr);
1278                 if (ifa) {
1279                         caddr_t macp = nd6_ifptomac(ifp);
1280                         ln->ln_expire = 0;
1281                         ln->ln_state = ND6_LLINFO_REACHABLE;
1282                         ln->ln_byhint = 0;
1283                         if (macp) {
1284                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1285                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1286                         }
1287                         if (nd6_useloopback) {
1288                                 rt->rt_ifp = &loif[0];  /* XXX */
1289                                 /*
1290                                  * Make sure rt_ifa be equal to the ifaddr
1291                                  * corresponding to the address.
1292                                  * We need this because when we refer
1293                                  * rt_ifa->ia6_flags in ip6_input, we assume
1294                                  * that the rt_ifa points to the address instead
1295                                  * of the loopback address.
1296                                  */
1297                                 if (ifa != rt->rt_ifa) {
1298                                         IFAFREE(rt->rt_ifa);
1299                                         IFAREF(ifa);
1300                                         rt->rt_ifa = ifa;
1301                                 }
1302                         }
1303                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1304                         ln->ln_expire = 0;
1305                         ln->ln_state = ND6_LLINFO_REACHABLE;
1306                         ln->ln_byhint = 0;
1307
1308                         /* join solicited node multicast for proxy ND */
1309                         if (ifp->if_flags & IFF_MULTICAST) {
1310                                 struct in6_addr llsol;
1311                                 int error;
1312
1313                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1314                                 llsol.s6_addr16[0] = htons(0xff02);
1315                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1316                                 llsol.s6_addr32[1] = 0;
1317                                 llsol.s6_addr32[2] = htonl(1);
1318                                 llsol.s6_addr8[12] = 0xff;
1319
1320                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1321                                         nd6log((LOG_ERR, "%s: failed to join "
1322                                             "%s (errno=%d)\n", if_name(ifp),
1323                                             ip6_sprintf(&llsol), error));
1324                                 }
1325                         }
1326                 }
1327                 break;
1328
1329         case RTM_DELETE:
1330                 if (!ln)
1331                         break;
1332                 /* leave from solicited node multicast for proxy ND */
1333                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1334                     (ifp->if_flags & IFF_MULTICAST)) {
1335                         struct in6_addr llsol;
1336                         struct in6_multi *in6m;
1337
1338                         llsol = SIN6(rt_key(rt))->sin6_addr;
1339                         llsol.s6_addr16[0] = htons(0xff02);
1340                         llsol.s6_addr16[1] = htons(ifp->if_index);
1341                         llsol.s6_addr32[1] = 0;
1342                         llsol.s6_addr32[2] = htonl(1);
1343                         llsol.s6_addr8[12] = 0xff;
1344
1345                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1346                         if (in6m)
1347                                 in6_delmulti(in6m);
1348                 }
1349                 nd6_inuse--;
1350                 ln->ln_next->ln_prev = ln->ln_prev;
1351                 ln->ln_prev->ln_next = ln->ln_next;
1352                 ln->ln_prev = NULL;
1353                 rt->rt_llinfo = 0;
1354                 rt->rt_flags &= ~RTF_LLINFO;
1355                 if (ln->ln_hold)
1356                         m_freem(ln->ln_hold);
1357                 Free((caddr_t)ln);
1358         }
1359 }
1360
1361 int
1362 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1363 {
1364         struct in6_drlist *drl = (struct in6_drlist *)data;
1365         struct in6_prlist *prl = (struct in6_prlist *)data;
1366         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1367         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1368         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1369         struct nd_defrouter *dr, any;
1370         struct nd_prefix *pr;
1371         struct rtentry *rt;
1372         int i = 0, error = 0;
1373
1374         switch (cmd) {
1375         case SIOCGDRLST_IN6:
1376                 /*
1377                  * obsolete API, use sysctl under net.inet6.icmp6
1378                  */
1379                 bzero(drl, sizeof(*drl));
1380                 mtx_lock(&nd6_mtx);
1381                 dr = TAILQ_FIRST(&nd_defrouter);
1382                 while (dr && i < DRLSTSIZ) {
1383                         drl->defrouter[i].rtaddr = dr->rtaddr;
1384                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1385                                 /* XXX: need to this hack for KAME stack */
1386                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1387                         } else
1388                                 log(LOG_ERR,
1389                                     "default router list contains a "
1390                                     "non-linklocal address(%s)\n",
1391                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1392
1393                         drl->defrouter[i].flags = dr->flags;
1394                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1395                         drl->defrouter[i].expire = dr->expire;
1396                         drl->defrouter[i].if_index = dr->ifp->if_index;
1397                         i++;
1398                         dr = TAILQ_NEXT(dr, dr_entry);
1399                 }
1400                 mtx_unlock(&nd6_mtx);
1401                 break;
1402         case SIOCGPRLST_IN6:
1403                 /*
1404                  * obsolete API, use sysctl under net.inet6.icmp6
1405                  */
1406                 /*
1407                  * XXX meaning of fields, especialy "raflags", is very
1408                  * differnet between RA prefix list and RR/static prefix list.
1409                  * how about separating ioctls into two?
1410                  */
1411                 bzero(prl, sizeof(*prl));
1412                 mtx_lock(&nd6_mtx);
1413                 pr = nd_prefix.lh_first;
1414                 while (pr && i < PRLSTSIZ) {
1415                         struct nd_pfxrouter *pfr;
1416                         int j;
1417
1418                         in6_embedscope(&prl->prefix[i].prefix,
1419                             &pr->ndpr_prefix, NULL, NULL);
1420                         prl->prefix[i].raflags = pr->ndpr_raf;
1421                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1422                         prl->prefix[i].vltime = pr->ndpr_vltime;
1423                         prl->prefix[i].pltime = pr->ndpr_pltime;
1424                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1425                         prl->prefix[i].expire = pr->ndpr_expire;
1426
1427                         pfr = pr->ndpr_advrtrs.lh_first;
1428                         j = 0;
1429                         while (pfr) {
1430                                 if (j < DRLSTSIZ) {
1431 #define RTRADDR prl->prefix[i].advrtr[j]
1432                                         RTRADDR = pfr->router->rtaddr;
1433                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1434                                                 /* XXX: hack for KAME */
1435                                                 RTRADDR.s6_addr16[1] = 0;
1436                                         } else
1437                                                 log(LOG_ERR,
1438                                                     "a router(%s) advertises "
1439                                                     "a prefix with "
1440                                                     "non-link local address\n",
1441                                                     ip6_sprintf(&RTRADDR));
1442 #undef RTRADDR
1443                                 }
1444                                 j++;
1445                                 pfr = pfr->pfr_next;
1446                         }
1447                         prl->prefix[i].advrtrs = j;
1448                         prl->prefix[i].origin = PR_ORIG_RA;
1449
1450                         i++;
1451                         pr = pr->ndpr_next;
1452                 }
1453                 mtx_unlock(&nd6_mtx);
1454
1455                 break;
1456         case OSIOCGIFINFO_IN6:
1457                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1458                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1459                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1460                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1461                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1462                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1463                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1464                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1465                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1466                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1467                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1468                 break;
1469         case SIOCGIFINFO_IN6:
1470                 ndi->ndi = *ND_IFINFO(ifp);
1471                 break;
1472         case SIOCSIFINFO_FLAGS:
1473                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1474                 break;
1475         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1476                 /* flush default router list */
1477                 /*
1478                  * xxx sumikawa: should not delete route if default
1479                  * route equals to the top of default router list
1480                  */
1481                 bzero(&any, sizeof(any));
1482                 defrouter_delreq(&any, 0);
1483                 defrouter_select();
1484                 /* xxx sumikawa: flush prefix list */
1485                 break;
1486         case SIOCSPFXFLUSH_IN6:
1487             {
1488                 /* flush all the prefix advertised by routers */
1489                 struct nd_prefix *pr, *next;
1490
1491                 mtx_lock(&nd6_mtx);
1492                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1493                         struct in6_ifaddr *ia, *ia_next;
1494
1495                         next = pr->ndpr_next;
1496
1497                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1498                                 continue; /* XXX */
1499
1500                         /* do we really have to remove addresses as well? */
1501                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1502                                 /* ia might be removed.  keep the next ptr. */
1503                                 ia_next = ia->ia_next;
1504
1505                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1506                                         continue;
1507
1508                                 if (ia->ia6_ndpr == pr)
1509                                         in6_purgeaddr(&ia->ia_ifa);
1510                         }
1511                         prelist_remove(pr);
1512                 }
1513                 mtx_unlock(&nd6_mtx);
1514                 break;
1515             }
1516         case SIOCSRTRFLUSH_IN6:
1517             {
1518                 /* flush all the default routers */
1519                 struct nd_defrouter *dr, *next;
1520
1521                 mtx_lock(&nd6_mtx);
1522                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1523                         /*
1524                          * The first entry of the list may be stored in
1525                          * the routing table, so we'll delete it later.
1526                          */
1527                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1528                                 next = TAILQ_NEXT(dr, dr_entry);
1529                                 defrtrlist_del(dr);
1530                         }
1531                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1532                 }
1533                 mtx_unlock(&nd6_mtx);
1534                 break;
1535             }
1536         case SIOCGNBRINFO_IN6:
1537             {
1538                 struct llinfo_nd6 *ln;
1539                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1540
1541                 /*
1542                  * XXX: KAME specific hack for scoped addresses
1543                  *      XXXX: for other scopes than link-local?
1544                  */
1545                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1546                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1547                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1548
1549                         if (*idp == 0)
1550                                 *idp = htons(ifp->if_index);
1551                 }
1552
1553                 mtx_lock(&nd6_mtx);
1554                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1555                         error = EINVAL;
1556                         mtx_unlock(&nd6_mtx);
1557                         break;
1558                 }
1559                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1560                 nbi->state = ln->ln_state;
1561                 nbi->asked = ln->ln_asked;
1562                 nbi->isrouter = ln->ln_router;
1563                 nbi->expire = ln->ln_expire;
1564                 mtx_unlock(&nd6_mtx);
1565
1566                 break;
1567             }
1568         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1569                 ndif->ifindex = nd6_defifindex;
1570                 break;
1571         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1572                 return (nd6_setdefaultiface(ndif->ifindex));
1573                 break;
1574         }
1575         return (error);
1576 }
1577
1578 /*
1579  * Create neighbor cache entry and cache link-layer address,
1580  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1581  */
1582 struct rtentry *
1583 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1584                  int lladdrlen,
1585                  int type,      /* ICMP6 type */
1586                  int code       /* type dependent information */)
1587 {
1588         struct rtentry *rt = NULL;
1589         struct llinfo_nd6 *ln = NULL;
1590         int is_newentry;
1591         struct sockaddr_dl *sdl = NULL;
1592         int do_update;
1593         int olladdr;
1594         int llchange;
1595         int newstate = 0;
1596
1597         if (!ifp)
1598                 panic("ifp == NULL in nd6_cache_lladdr");
1599         if (!from)
1600                 panic("from == NULL in nd6_cache_lladdr");
1601
1602         /* nothing must be updated for unspecified address */
1603         if (IN6_IS_ADDR_UNSPECIFIED(from))
1604                 return NULL;
1605
1606         /*
1607          * Validation about ifp->if_addrlen and lladdrlen must be done in
1608          * the caller.
1609          *
1610          * XXX If the link does not have link-layer adderss, what should
1611          * we do? (ifp->if_addrlen == 0)
1612          * Spec says nothing in sections for RA, RS and NA.  There's small
1613          * description on it in NS section (RFC 2461 7.2.3).
1614          */
1615
1616         rt = nd6_lookup(from, 0, ifp);
1617         if (!rt) {
1618 #if 0
1619                 /* nothing must be done if there's no lladdr */
1620                 if (!lladdr || !lladdrlen)
1621                         return NULL;
1622 #endif
1623
1624                 rt = nd6_lookup(from, 1, ifp);
1625                 is_newentry = 1;
1626         } else {
1627                 /* do nothing if static ndp is set */
1628                 if (rt->rt_flags & RTF_STATIC)
1629                         return NULL;
1630                 is_newentry = 0;
1631         }
1632
1633         if (!rt)
1634                 return NULL;
1635         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1636 fail:
1637                 nd6_free(rt);
1638                 return NULL;
1639         }
1640         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1641         if (!ln)
1642                 goto fail;
1643         if (!rt->rt_gateway)
1644                 goto fail;
1645         if (rt->rt_gateway->sa_family != AF_LINK)
1646                 goto fail;
1647         sdl = SDL(rt->rt_gateway);
1648
1649         olladdr = (sdl->sdl_alen) ? 1 : 0;
1650         if (olladdr && lladdr) {
1651                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1652                         llchange = 1;
1653                 else
1654                         llchange = 0;
1655         } else
1656                 llchange = 0;
1657
1658         /*
1659          * newentry olladdr  lladdr  llchange   (*=record)
1660          *      0       n       n       --      (1)
1661          *      0       y       n       --      (2)
1662          *      0       n       y       --      (3) * STALE
1663          *      0       y       y       n       (4) *
1664          *      0       y       y       y       (5) * STALE
1665          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1666          *      1       --      y       --      (7) * STALE
1667          */
1668
1669         if (lladdr) {           /* (3-5) and (7) */
1670                 /*
1671                  * Record source link-layer address
1672                  * XXX is it dependent to ifp->if_type?
1673                  */
1674                 sdl->sdl_alen = ifp->if_addrlen;
1675                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1676         }
1677
1678         if (!is_newentry) {
1679                 if ((!olladdr && lladdr)                /* (3) */
1680                  || (olladdr && lladdr && llchange)) {  /* (5) */
1681                         do_update = 1;
1682                         newstate = ND6_LLINFO_STALE;
1683                 } else                                  /* (1-2,4) */
1684                         do_update = 0;
1685         } else {
1686                 do_update = 1;
1687                 if (!lladdr)                            /* (6) */
1688                         newstate = ND6_LLINFO_NOSTATE;
1689                 else                                    /* (7) */
1690                         newstate = ND6_LLINFO_STALE;
1691         }
1692
1693         if (do_update) {
1694                 /*
1695                  * Update the state of the neighbor cache.
1696                  */
1697                 ln->ln_state = newstate;
1698
1699                 if (ln->ln_state == ND6_LLINFO_STALE) {
1700                         /*
1701                          * XXX: since nd6_output() below will cause
1702                          * state tansition to DELAY and reset the timer,
1703                          * we must set the timer now, although it is actually
1704                          * meaningless.
1705                          */
1706                         ln->ln_expire = time_uptime + nd6_gctimer;
1707
1708                         if (ln->ln_hold) {
1709                                 /*
1710                                  * we assume ifp is not a p2p here, so just
1711                                  * set the 2nd argument as the 1st one.
1712                                  */
1713                                 nd6_output(ifp, ifp, ln->ln_hold,
1714                                            (struct sockaddr_in6 *)rt_key(rt),
1715                                            rt);
1716                                 ln->ln_hold = NULL;
1717                         }
1718                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1719                         /* probe right away */
1720                         ln->ln_expire = time_uptime;
1721                 }
1722         }
1723
1724         /*
1725          * ICMP6 type dependent behavior.
1726          *
1727          * NS: clear IsRouter if new entry
1728          * RS: clear IsRouter
1729          * RA: set IsRouter if there's lladdr
1730          * redir: clear IsRouter if new entry
1731          *
1732          * RA case, (1):
1733          * The spec says that we must set IsRouter in the following cases:
1734          * - If lladdr exist, set IsRouter.  This means (1-5).
1735          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1736          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1737          * A quetion arises for (1) case.  (1) case has no lladdr in the
1738          * neighbor cache, this is similar to (6).
1739          * This case is rare but we figured that we MUST NOT set IsRouter.
1740          *
1741          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1742          *                                                      D R
1743          *      0       n       n       --      (1)     c   ?     s
1744          *      0       y       n       --      (2)     c   s     s
1745          *      0       n       y       --      (3)     c   s     s
1746          *      0       y       y       n       (4)     c   s     s
1747          *      0       y       y       y       (5)     c   s     s
1748          *      1       --      n       --      (6) c   c       c s
1749          *      1       --      y       --      (7) c   c   s   c s
1750          *
1751          *                                      (c=clear s=set)
1752          */
1753         switch (type & 0xff) {
1754         case ND_NEIGHBOR_SOLICIT:
1755                 /*
1756                  * New entry must have is_router flag cleared.
1757                  */
1758                 if (is_newentry)        /* (6-7) */
1759                         ln->ln_router = 0;
1760                 break;
1761         case ND_REDIRECT:
1762                 /*
1763                  * If the icmp is a redirect to a better router, always set the
1764                  * is_router flag. Otherwise, if the entry is newly created,
1765                  * clear the flag. [RFC 2461, sec 8.3]
1766                  */
1767                 if (code == ND_REDIRECT_ROUTER)
1768                         ln->ln_router = 1;
1769                 else if (is_newentry) /* (6-7) */
1770                         ln->ln_router = 0;
1771                 break;
1772         case ND_ROUTER_SOLICIT:
1773                 /*
1774                  * is_router flag must always be cleared.
1775                  */
1776                 ln->ln_router = 0;
1777                 break;
1778         case ND_ROUTER_ADVERT:
1779                 /*
1780                  * Mark an entry with lladdr as a router.
1781                  */
1782                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1783                  || (is_newentry && lladdr)) {                  /* (7) */
1784                         ln->ln_router = 1;
1785                 }
1786                 break;
1787         }
1788
1789         /*
1790          * When the link-layer address of a router changes, select the
1791          * best router again.  In particular, when the neighbor entry is newly
1792          * created, it might affect the selection policy.
1793          * Question: can we restrict the first condition to the "is_newentry"
1794          * case?
1795          * XXX: when we hear an RA from a new router with the link-layer
1796          * address option, defrouter_select() is called twice, since
1797          * defrtrlist_update called the function as well.  However, I believe
1798          * we can compromise the overhead, since it only happens the first
1799          * time.
1800          * XXX: although defrouter_select() should not have a bad effect
1801          * for those are not autoconfigured hosts, we explicitly avoid such
1802          * cases for safety.
1803          */
1804         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1805                 defrouter_select();
1806
1807         return rt;
1808 }
1809
1810 static void
1811 nd6_slowtimo(void *arg __unused)
1812 {
1813         struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1814
1815         KASSERT(mycpuid == 0, ("not on cpu0"));
1816         crit_enter();
1817         if (lmsg->ms_flags & MSGF_DONE)
1818                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1819         crit_exit();
1820 }
1821
1822 static void
1823 nd6_slowtimo_dispatch(netmsg_t nmsg)
1824 {
1825         const struct ifnet_array *arr;
1826         struct nd_ifinfo *nd6if;
1827         int i;
1828
1829         KASSERT(&curthread->td_msgport == netisr_cpuport(0),
1830             ("not in netisr0"));
1831
1832         crit_enter();
1833         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1834         crit_exit();
1835
1836         arr = ifnet_array_get();
1837
1838         mtx_lock(&nd6_mtx);
1839         for (i = 0; i < arr->ifnet_count; ++i) {
1840                 struct ifnet *ifp = arr->ifnet_arr[i];
1841
1842                 if (ifp->if_afdata[AF_INET6] == NULL)
1843                         continue;
1844                 nd6if = ND_IFINFO(ifp);
1845                 if (nd6if->basereachable && /* already initialized */
1846                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1847                         /*
1848                          * Since reachable time rarely changes by router
1849                          * advertisements, we SHOULD insure that a new random
1850                          * value gets recomputed at least once every few hours.
1851                          * (RFC 2461, 6.3.4)
1852                          */
1853                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1854                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1855                 }
1856         }
1857         mtx_unlock(&nd6_mtx);
1858
1859         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1860             nd6_slowtimo, NULL);
1861 }
1862
1863 #define gotoerr(e) { error = (e); goto bad;}
1864
1865 int
1866 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1867            struct sockaddr_in6 *dst, struct rtentry *rt)
1868 {
1869         struct llinfo_nd6 *ln = NULL;
1870         int error = 0;
1871
1872         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1873                 goto sendpkt;
1874
1875         if (nd6_need_cache(ifp) == 0)
1876                 goto sendpkt;
1877
1878         /*
1879          * next hop determination.  This routine is derived from ether_outpout.
1880          */
1881         if (rt != NULL) {
1882                 if (!(rt->rt_flags & RTF_UP)) {
1883                         rt = rtlookup((struct sockaddr *)dst);
1884                         if (rt == NULL)
1885                                 gotoerr(EHOSTUNREACH);
1886                         rt->rt_refcnt--;
1887                         if (rt->rt_ifp != ifp) {
1888                                 /* XXX: loop care? */
1889                                 return nd6_output(ifp, origifp, m, dst, rt);
1890                         }
1891                 }
1892                 if (rt->rt_flags & RTF_GATEWAY) {
1893                         struct sockaddr_in6 *gw6;
1894
1895                         /*
1896                          * We skip link-layer address resolution and NUD
1897                          * if the gateway is not a neighbor from ND point
1898                          * of view, regardless of the value of nd_ifinfo.flags.
1899                          * The second condition is a bit tricky; we skip
1900                          * if the gateway is our own address, which is
1901                          * sometimes used to install a route to a p2p link.
1902                          */
1903                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1904                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1905                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1906                                 /*
1907                                  * We allow this kind of tricky route only
1908                                  * when the outgoing interface is p2p.
1909                                  * XXX: we may need a more generic rule here.
1910                                  */
1911                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1912                                         gotoerr(EHOSTUNREACH);
1913
1914                                 goto sendpkt;
1915                         }
1916
1917                         if (rt->rt_gwroute == NULL) {
1918                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1919                                 if (rt->rt_gwroute == NULL)
1920                                         gotoerr(EHOSTUNREACH);
1921                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1922                                 rtfree(rt->rt_gwroute);
1923                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1924                                 if (rt->rt_gwroute == NULL)
1925                                         gotoerr(EHOSTUNREACH);
1926                         }
1927                 }
1928         }
1929
1930         /*
1931          * Address resolution or Neighbor Unreachability Detection
1932          * for the next hop.
1933          * At this point, the destination of the packet must be a unicast
1934          * or an anycast address(i.e. not a multicast).
1935          */
1936
1937         /* Look up the neighbor cache for the nexthop */
1938         if (rt && (rt->rt_flags & RTF_LLINFO))
1939                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1940         else {
1941                 /*
1942                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1943                  * the condition below is not very efficient.  But we believe
1944                  * it is tolerable, because this should be a rare case.
1945                  */
1946                 if (nd6_is_addr_neighbor(dst, ifp) &&
1947                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1948                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1949         }
1950         if (!ln || !rt) {
1951                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1952                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1953                         log(LOG_DEBUG,
1954                             "nd6_output: can't allocate llinfo for %s "
1955                             "(ln=%p, rt=%p)\n",
1956                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1957                         gotoerr(EIO);   /* XXX: good error? */
1958                 }
1959
1960                 goto sendpkt;   /* send anyway */
1961         }
1962
1963         /* We don't have to do link-layer address resolution on a p2p link. */
1964         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1965             ln->ln_state < ND6_LLINFO_REACHABLE) {
1966                 ln->ln_state = ND6_LLINFO_STALE;
1967                 ln->ln_expire = time_uptime + nd6_gctimer;
1968         }
1969
1970         /*
1971          * The first time we send a packet to a neighbor whose entry is
1972          * STALE, we have to change the state to DELAY and a sets a timer to
1973          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1974          * neighbor unreachability detection on expiration.
1975          * (RFC 2461 7.3.3)
1976          */
1977         if (ln->ln_state == ND6_LLINFO_STALE) {
1978                 ln->ln_asked = 0;
1979                 ln->ln_state = ND6_LLINFO_DELAY;
1980                 ln->ln_expire = time_uptime + nd6_delay;
1981         }
1982
1983         /*
1984          * If the neighbor cache entry has a state other than INCOMPLETE
1985          * (i.e. its link-layer address is already resolved), just
1986          * send the packet.
1987          */
1988         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1989                 goto sendpkt;
1990
1991         /*
1992          * There is a neighbor cache entry, but no ethernet address
1993          * response yet.  Replace the held mbuf (if any) with this
1994          * latest one.
1995          *
1996          * This code conforms to the rate-limiting rule described in Section
1997          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1998          * an NS below.
1999          */
2000         if (ln->ln_state == ND6_LLINFO_NOSTATE)
2001                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
2002         if (ln->ln_hold)
2003                 m_freem(ln->ln_hold);
2004         ln->ln_hold = m;
2005         if (ln->ln_expire) {
2006                 if (ln->ln_asked < nd6_mmaxtries &&
2007                     ln->ln_expire < time_uptime) {
2008                         ln->ln_asked++;
2009                         ln->ln_expire = time_uptime +
2010                                 ND_IFINFO(ifp)->retrans / 1000;
2011                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2012                 }
2013         }
2014         return (0);
2015
2016 sendpkt:
2017         if (ifp->if_flags & IFF_LOOPBACK)
2018                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
2019         else
2020                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
2021         return (error);
2022
2023 bad:
2024         m_freem(m);
2025         return (error);
2026 }
2027 #undef gotoerr
2028
2029 int
2030 nd6_need_cache(struct ifnet *ifp)
2031 {
2032         /*
2033          * XXX: we currently do not make neighbor cache on any interface
2034          * other than Ethernet and GIF.
2035          *
2036          * RFC2893 says:
2037          * - unidirectional tunnels needs no ND
2038          */
2039         switch (ifp->if_type) {
2040         case IFT_ETHER:
2041         case IFT_IEEE1394:
2042 #ifdef IFT_L2VLAN
2043         case IFT_L2VLAN:
2044 #endif
2045 #ifdef IFT_IEEE80211
2046         case IFT_IEEE80211:
2047 #endif
2048 #ifdef IFT_CARP
2049         case IFT_CARP:
2050 #endif
2051         case IFT_GIF:           /* XXX need more cases? */
2052                 return (1);
2053         default:
2054                 return (0);
2055         }
2056 }
2057
2058 int
2059 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2060                 struct sockaddr *dst, u_char *desten)
2061 {
2062         struct sockaddr_dl *sdl;
2063         struct rtentry *rt;
2064
2065
2066         if (m->m_flags & M_MCAST) {
2067                 switch (ifp->if_type) {
2068                 case IFT_ETHER:
2069 #ifdef IFT_L2VLAN
2070         case IFT_L2VLAN:
2071 #endif
2072 #ifdef IFT_IEEE80211
2073                 case IFT_IEEE80211:
2074 #endif
2075                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2076                                                  desten);
2077                         return (1);
2078                 case IFT_IEEE1394:
2079                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2080                         return (1);
2081                 default:
2082                         m_freem(m);
2083                         return (0);
2084                 }
2085         }
2086         if (rt0 == NULL) {
2087                 /* this could happen, if we could not allocate memory */
2088                 m_freem(m);
2089                 return (0);
2090         }
2091         if (rt_llroute(dst, rt0, &rt) != 0) {
2092                 m_freem(m);
2093                 return (0);
2094         }
2095         if (rt->rt_gateway->sa_family != AF_LINK) {
2096                 kprintf("nd6_storelladdr: something odd happens\n");
2097                 m_freem(m);
2098                 return (0);
2099         }
2100         sdl = SDL(rt->rt_gateway);
2101         if (sdl->sdl_alen == 0) {
2102                 /* this should be impossible, but we bark here for debugging */
2103                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2104                 m_freem(m);
2105                 return (0);
2106         }
2107
2108         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2109         return (1);
2110 }
2111
2112 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2113 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2114 #ifdef SYSCTL_DECL
2115 SYSCTL_DECL(_net_inet6_icmp6);
2116 #endif
2117 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2118         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2119 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2120         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2121
2122 static int
2123 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2124 {
2125         int error;
2126         char buf[1024];
2127         struct in6_defrouter *d, *de;
2128         struct nd_defrouter *dr;
2129
2130         if (req->newptr)
2131                 return EPERM;
2132         error = 0;
2133
2134         for (dr = TAILQ_FIRST(&nd_defrouter);
2135              dr;
2136              dr = TAILQ_NEXT(dr, dr_entry)) {
2137                 d = (struct in6_defrouter *)buf;
2138                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2139
2140                 if (d + 1 <= de) {
2141                         bzero(d, sizeof(*d));
2142                         d->rtaddr.sin6_family = AF_INET6;
2143                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2144                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2145                             dr->ifp) != 0)
2146                                 log(LOG_ERR,
2147                                     "scope error in "
2148                                     "default router list (%s)\n",
2149                                     ip6_sprintf(&dr->rtaddr));
2150                         d->flags = dr->flags;
2151                         d->rtlifetime = dr->rtlifetime;
2152                         d->expire = dr->expire;
2153                         d->if_index = dr->ifp->if_index;
2154                 } else
2155                         panic("buffer too short");
2156
2157                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2158                 if (error)
2159                         break;
2160         }
2161         return error;
2162 }
2163
2164 static int
2165 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2166 {
2167         int error;
2168         char buf[1024];
2169         struct in6_prefix *p, *pe;
2170         struct nd_prefix *pr;
2171
2172         if (req->newptr)
2173                 return EPERM;
2174         error = 0;
2175
2176         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2177                 u_short advrtrs;
2178                 size_t advance;
2179                 struct sockaddr_in6 *sin6, *s6;
2180                 struct nd_pfxrouter *pfr;
2181
2182                 p = (struct in6_prefix *)buf;
2183                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2184
2185                 if (p + 1 <= pe) {
2186                         bzero(p, sizeof(*p));
2187                         sin6 = (struct sockaddr_in6 *)(p + 1);
2188
2189                         p->prefix = pr->ndpr_prefix;
2190                         if (in6_recoverscope(&p->prefix,
2191                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2192                                 log(LOG_ERR,
2193                                     "scope error in prefix list (%s)\n",
2194                                     ip6_sprintf(&p->prefix.sin6_addr));
2195                         p->raflags = pr->ndpr_raf;
2196                         p->prefixlen = pr->ndpr_plen;
2197                         p->vltime = pr->ndpr_vltime;
2198                         p->pltime = pr->ndpr_pltime;
2199                         p->if_index = pr->ndpr_ifp->if_index;
2200                         p->expire = pr->ndpr_expire;
2201                         p->refcnt = pr->ndpr_refcnt;
2202                         p->flags = pr->ndpr_stateflags;
2203                         p->origin = PR_ORIG_RA;
2204                         advrtrs = 0;
2205                         for (pfr = pr->ndpr_advrtrs.lh_first;
2206                              pfr;
2207                              pfr = pfr->pfr_next) {
2208                                 if ((void *)&sin6[advrtrs + 1] >
2209                                     (void *)pe) {
2210                                         advrtrs++;
2211                                         continue;
2212                                 }
2213                                 s6 = &sin6[advrtrs];
2214                                 bzero(s6, sizeof(*s6));
2215                                 s6->sin6_family = AF_INET6;
2216                                 s6->sin6_len = sizeof(*sin6);
2217                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2218                                                      pfr->router->ifp) != 0)
2219                                         log(LOG_ERR,
2220                                             "scope error in "
2221                                             "prefix list (%s)\n",
2222                                             ip6_sprintf(&pfr->router->rtaddr));
2223                                 advrtrs++;
2224                         }
2225                         p->advrtrs = advrtrs;
2226                 } else
2227                         panic("buffer too short");
2228
2229                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2230                 error = SYSCTL_OUT(req, buf, advance);
2231                 if (error)
2232                         break;
2233         }
2234         return error;
2235 }