sys/netinet6/nd6.c

   1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
   2 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
   3
   4 /*
   5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of the project nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 #include "opt_inet.h"
  34 #include "opt_inet6.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/callout.h>
  39 #include <sys/malloc.h>
  40 #include <sys/mbuf.h>
  41 #include <sys/socket.h>
  42 #include <sys/sockio.h>
  43 #include <sys/time.h>
  44 #include <sys/kernel.h>
  45 #include <sys/protosw.h>
  46 #include <sys/errno.h>
  47 #include <sys/syslog.h>
  48 #include <sys/queue.h>
  49 #include <sys/sysctl.h>
  50 #include <sys/mutex.h>
  51
  52 #include <sys/thread2.h>
  53 #include <sys/mutex2.h>
  54
  55 #include <net/if.h>
  56 #include <net/if_dl.h>
  57 #include <net/if_types.h>
  58 #include <net/route.h>
  59 #include <net/netisr2.h>
  60 #include <net/netmsg2.h>
  61
  62 #include <netinet/in.h>
  63 #include <netinet/if_ether.h>
  64 #include <netinet6/in6_var.h>
  65 #include <netinet6/in6_ifattach.h>
  66 #include <netinet/ip6.h>
  67 #include <netinet6/ip6_var.h>
  68 #include <netinet6/nd6.h>
  69 #include <netinet/icmp6.h>
  70
  71 #include <net/net_osdep.h>
  72
  73 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
  74 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
  75
  76 #define SIN6(s) ((struct sockaddr_in6 *)s)
  77 #define SDL(s) ((struct sockaddr_dl *)s)
  78
  79 /*
  80  * Determine if the route entry is a direct neighbor on the specified
  81  * interface.  The interface test is not done if ifp is passed as NULL.
  82  * The route entry is a neighbor if all of the following are true:
  83  *
  84  *      RTF_GATEWAY is FALSE
  85  *      LLINFO is TRUE
  86  *      rt_gateway family is AF_LINK
  87  *      rt_llinfo is non-NULL
  88  *      The interfaces matches (or ifp is passed as NULL)
  89  *
  90  *      NOTE: rt_llinfo can be NULL with LLINFO set, so both must be
  91  *            tested.
  92  *
  93  *      NOTE: We can't use rt->rt_ifp to check for the interface, since
  94  *            it may be the loopback interface if the entry is for our
  95  *            own address on a non-loopback interface.  Instead, we use
  96  *            rt->rt_ifa->ifa_ifp which should specify the REAL interface.
  97  */
  98 #define ND6_IFP_MATCHES(ifp, ifa_ifp)                           \
  99         ((ifp) == NULL ||                                       \
 100          (ifa_ifp) == (ifp) ||                                  \
 101          (((ifp)->if_flags & IFF_ISBRIDGE) &&                   \
 102           (ifa_ifp)->if_bridge == (ifp)->if_softc)              \
 103         )
 104
 105 #define ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)                        \
 106         (((rt)->rt_flags & RTF_GATEWAY) == 0 &&                 \
 107          ((rt)->rt_flags & RTF_LLINFO) != 0 &&                  \
 108          (rt)->rt_gateway->sa_family == AF_LINK &&              \
 109          (rt)->rt_llinfo &&                                     \
 110          ND6_IFP_MATCHES((ifp), (rt)->rt_ifa->ifa_ifp)          \
 111         )
 112
 113 #define ND6_RTENTRY_IS_LLCLONING(rt)                            \
 114         (((rt)->rt_flags & (RTF_PRCLONING | RTF_LLINFO)) ==     \
 115          (RTF_PRCLONING | RTF_LLINFO) ||                        \
 116          ((rt)->rt_flags & RTF_CLONING))
 117
 118 /* timer values */
 119 int     nd6_prune       = 1;    /* walk list every 1 seconds */
 120 int     nd6_delay       = 5;    /* delay first probe time 5 second */
 121 int     nd6_umaxtries   = 3;    /* maximum unicast query */
 122 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
 123 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
 124 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
 125
 126 /* preventing too many loops in ND option parsing */
 127 int nd6_maxndopt = 10;  /* max # of ND options allowed */
 128
 129 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
 130
 131 #ifdef ND6_DEBUG
 132 int nd6_debug = 1;
 133 #else
 134 int nd6_debug = 0;
 135 #endif
 136
 137 /* for debugging? */
 138 static int nd6_inuse, nd6_allocated;
 139
 140 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
 141 struct nd_drhead nd_defrouter;
 142 struct nd_prhead nd_prefix = { 0 };
 143 struct mtx nd6_mtx = MTX_INITIALIZER("nd6");
 144
 145 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 146 static struct sockaddr_in6 all1_sa;
 147
 148 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
 149 static int regen_tmpaddr (struct in6_ifaddr *);
 150 static void nd6_slowtimo(void *);
 151 static void nd6_slowtimo_dispatch(netmsg_t);
 152 static void nd6_timer(void *);
 153 static void nd6_timer_dispatch(netmsg_t);
 154
 155 static struct callout nd6_slowtimo_ch;
 156 static struct netmsg_base nd6_slowtimo_netmsg;
 157
 158 static struct callout nd6_timer_ch;
 159 static struct netmsg_base nd6_timer_netmsg;
 160
 161 void
 162 nd6_init(void)
 163 {
 164         static int nd6_init_done = 0;
 165         int i;
 166
 167         if (nd6_init_done) {
 168                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
 169                 return;
 170         }
 171
 172         all1_sa.sin6_family = AF_INET6;
 173         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
 174         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
 175                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
 176
 177         /* initialization of the default router list */
 178         TAILQ_INIT(&nd_defrouter);
 179
 180         nd6_init_done = 1;
 181
 182         /* start timer */
 183         callout_init_mp(&nd6_slowtimo_ch);
 184         netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
 185             MSGF_PRIORITY, nd6_slowtimo_dispatch);
 186         callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 187             nd6_slowtimo, NULL, 0);
 188 }
 189
 190 struct nd_ifinfo *
 191 nd6_ifattach(struct ifnet *ifp)
 192 {
 193         struct nd_ifinfo *nd;
 194
 195         nd = kmalloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
 196
 197         nd->initialized = 1;
 198
 199         nd->chlim = IPV6_DEFHLIM;
 200         nd->basereachable = REACHABLE_TIME;
 201         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
 202         nd->retrans = RETRANS_TIMER;
 203
 204         nd->flags = ND6_IFF_PERFORMNUD;
 205         /* A loopback interface always has link-local address. */
 206         if (ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK))
 207                 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
 208         /* A loopback interface does not need to accept RAs. */
 209         if (ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK))
 210                 nd->flags |= ND6_IFF_ACCEPT_RTADV;
 211
 212         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
 213         nd6_setmtu0(ifp, nd);
 214
 215         return nd;
 216 }
 217
 218 void
 219 nd6_ifdetach(struct nd_ifinfo *nd)
 220 {
 221         kfree(nd, M_IP6NDP);
 222 }
 223
 224 /*
 225  * Reset ND level link MTU. This function is called when the physical MTU
 226  * changes, which means we might have to adjust the ND level MTU.
 227  */
 228 void
 229 nd6_setmtu(struct ifnet *ifp)
 230 {
 231         nd6_setmtu0(ifp, ND_IFINFO(ifp));
 232 }
 233
 234 struct netmsg_nd6setmtu {
 235         struct netmsg_base      nmsg;
 236         struct ifnet            *ifp;
 237         struct nd_ifinfo        *ndi;
 238 };
 239
 240 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
 241 static void
 242 nd6_setmtu0_dispatch(netmsg_t msg)
 243 {
 244         struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
 245         struct ifnet *ifp = nmsg->ifp;
 246         struct nd_ifinfo *ndi = nmsg->ndi;
 247         uint32_t omaxmtu;
 248
 249         omaxmtu = ndi->maxmtu;
 250
 251         switch (ifp->if_type) {
 252         case IFT_ETHER:
 253                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 254                 break;
 255         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
 256                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 257                 break;
 258 #ifdef IFT_IEEE80211
 259         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
 260                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
 261                 break;
 262 #endif
 263         default:
 264                 ndi->maxmtu = ifp->if_mtu;
 265                 break;
 266         }
 267
 268         /*
 269          * Decreasing the interface MTU under IPV6 minimum MTU may cause
 270          * undesirable situation.  We thus notify the operator of the change
 271          * explicitly.  The check for omaxmtu is necessary to restrict the
 272          * log to the case of changing the MTU, not initializing it.
 273          */
 274         if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
 275                 log(LOG_NOTICE, "nd6_setmtu0: "
 276                     "new link MTU on %s (%lu) is too small for IPv6\n",
 277                     if_name(ifp), (unsigned long)ndi->maxmtu);
 278         }
 279
 280         if (ndi->maxmtu > in6_maxmtu)
 281                 in6_setmaxmtu(); /* check all interfaces just in case */
 282
 283         lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
 284 }
 285
 286 void
 287 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
 288 {
 289         struct netmsg_nd6setmtu nmsg;
 290
 291         netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
 292             nd6_setmtu0_dispatch);
 293         nmsg.ifp = ifp;
 294         nmsg.ndi = ndi;
 295         lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
 296 }
 297
 298 void
 299 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
 300 {
 301         bzero(ndopts, sizeof(*ndopts));
 302         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
 303         ndopts->nd_opts_last
 304                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
 305
 306         if (icmp6len == 0) {
 307                 ndopts->nd_opts_done = 1;
 308                 ndopts->nd_opts_search = NULL;
 309         }
 310 }
 311
 312 /*
 313  * Take one ND option.
 314  */
 315 struct nd_opt_hdr *
 316 nd6_option(union nd_opts *ndopts)
 317 {
 318         struct nd_opt_hdr *nd_opt;
 319         int olen;
 320
 321         if (!ndopts)
 322                 panic("ndopts == NULL in nd6_option");
 323         if (!ndopts->nd_opts_last)
 324                 panic("uninitialized ndopts in nd6_option");
 325         if (!ndopts->nd_opts_search)
 326                 return NULL;
 327         if (ndopts->nd_opts_done)
 328                 return NULL;
 329
 330         nd_opt = ndopts->nd_opts_search;
 331
 332         /* make sure nd_opt_len is inside the buffer */
 333         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
 334                 bzero(ndopts, sizeof(*ndopts));
 335                 return NULL;
 336         }
 337
 338         olen = nd_opt->nd_opt_len << 3;
 339         if (olen == 0) {
 340                 /*
 341                  * Message validation requires that all included
 342                  * options have a length that is greater than zero.
 343                  */
 344                 bzero(ndopts, sizeof(*ndopts));
 345                 return NULL;
 346         }
 347
 348         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
 349         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
 350                 /* option overruns the end of buffer, invalid */
 351                 bzero(ndopts, sizeof(*ndopts));
 352                 return NULL;
 353         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
 354                 /* reached the end of options chain */
 355                 ndopts->nd_opts_done = 1;
 356                 ndopts->nd_opts_search = NULL;
 357         }
 358         return nd_opt;
 359 }
 360
 361 /*
 362  * Parse multiple ND options.
 363  * This function is much easier to use, for ND routines that do not need
 364  * multiple options of the same type.
 365  */
 366 int
 367 nd6_options(union nd_opts *ndopts)
 368 {
 369         struct nd_opt_hdr *nd_opt;
 370         int i = 0;
 371
 372         if (!ndopts)
 373                 panic("ndopts == NULL in nd6_options");
 374         if (!ndopts->nd_opts_last)
 375                 panic("uninitialized ndopts in nd6_options");
 376         if (!ndopts->nd_opts_search)
 377                 return 0;
 378
 379         while (1) {
 380                 nd_opt = nd6_option(ndopts);
 381                 if (!nd_opt && !ndopts->nd_opts_last) {
 382                         /*
 383                          * Message validation requires that all included
 384                          * options have a length that is greater than zero.
 385                          */
 386                         icmp6stat.icp6s_nd_badopt++;
 387                         bzero(ndopts, sizeof(*ndopts));
 388                         return -1;
 389                 }
 390
 391                 if (!nd_opt)
 392                         goto skip1;
 393
 394                 switch (nd_opt->nd_opt_type) {
 395                 case ND_OPT_SOURCE_LINKADDR:
 396                 case ND_OPT_TARGET_LINKADDR:
 397                 case ND_OPT_MTU:
 398                 case ND_OPT_REDIRECTED_HEADER:
 399                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 400                                 nd6log((LOG_INFO,
 401                                     "duplicated ND6 option found (type=%d)\n",
 402                                     nd_opt->nd_opt_type));
 403                                 /* XXX bark? */
 404                         } else {
 405                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 406                                         = nd_opt;
 407                         }
 408                         break;
 409                 case ND_OPT_PREFIX_INFORMATION:
 410                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
 411                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
 412                                         = nd_opt;
 413                         }
 414                         ndopts->nd_opts_pi_end =
 415                                 (struct nd_opt_prefix_info *)nd_opt;
 416                         break;
 417                 default:
 418                         /*
 419                          * Unknown options must be silently ignored,
 420                          * to accomodate future extension to the protocol.
 421                          */
 422                         nd6log((LOG_DEBUG,
 423                             "nd6_options: unsupported option %d - "
 424                             "option ignored\n", nd_opt->nd_opt_type));
 425                 }
 426
 427 skip1:
 428                 i++;
 429                 if (i > nd6_maxndopt) {
 430                         icmp6stat.icp6s_nd_toomanyopt++;
 431                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
 432                         break;
 433                 }
 434
 435                 if (ndopts->nd_opts_done)
 436                         break;
 437         }
 438
 439         return 0;
 440 }
 441
 442 /*
 443  * ND6 timer routine to expire default route list and prefix list
 444  */
 445 static void
 446 nd6_timer_dispatch(netmsg_t nmsg)
 447 {
 448         struct llinfo_nd6 *ln;
 449         struct nd_defrouter *dr;
 450         struct nd_prefix *pr;
 451         struct ifnet *ifp;
 452         struct in6_ifaddr *ia6, *nia6;
 453
 454         ASSERT_NETISR0;
 455
 456         crit_enter();
 457         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
 458         crit_exit();
 459
 460         mtx_lock(&nd6_mtx);
 461
 462         ln = llinfo_nd6.ln_next;
 463         while (ln && ln != &llinfo_nd6) {
 464                 struct rtentry *rt;
 465                 struct sockaddr_in6 *dst;
 466                 struct llinfo_nd6 *next = ln->ln_next;
 467                 /* XXX: used for the DELAY case only: */
 468                 struct nd_ifinfo *ndi = NULL;
 469
 470                 if ((rt = ln->ln_rt) == NULL) {
 471                         ln = next;
 472                         continue;
 473                 }
 474                 if ((ifp = rt->rt_ifp) == NULL) {
 475                         ln = next;
 476                         continue;
 477                 }
 478                 ndi = ND_IFINFO(ifp);
 479                 dst = (struct sockaddr_in6 *)rt_key(rt);
 480
 481                 if (ln->ln_expire > time_uptime) {
 482                         ln = next;
 483                         continue;
 484                 }
 485
 486                 /* sanity check */
 487                 if (!rt)
 488                         panic("rt=0 in nd6_timer(ln=%p)", ln);
 489                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
 490                         panic("rt_llinfo(%p) is not equal to ln(%p)",
 491                               rt->rt_llinfo, ln);
 492                 if (!dst)
 493                         panic("dst=0 in nd6_timer(ln=%p)", ln);
 494
 495                 switch (ln->ln_state) {
 496                 case ND6_LLINFO_WAITDELETE:
 497                         next = nd6_free(rt);
 498                         break;
 499                 case ND6_LLINFO_INCOMPLETE:
 500                         if (ln->ln_asked++ >= nd6_mmaxtries) {
 501                                 struct mbuf *m = ln->ln_hold;
 502                                 if (m) {
 503                                         if (rt->rt_ifp) {
 504                                                 /*
 505                                                  * Fake rcvif to make ICMP error
 506                                                  * more helpful in diagnosing
 507                                                  * for the receiver.
 508                                                  * XXX: should we consider
 509                                                  * older rcvif?
 510                                                  */
 511                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
 512                                         }
 513
 514                                         /*
 515                                          * mbuf has empty MAC header, remove
 516                                          * for icmp.  XXX layer violation.
 517                                          */
 518                                         m_adj(m, ETHER_HDR_LEN);
 519                                         icmp6_error(m, ICMP6_DST_UNREACH,
 520                                                     ICMP6_DST_UNREACH_ADDR, 0);
 521                                         ln->ln_hold = NULL;
 522                                 }
 523                                 ln->ln_state = ND6_LLINFO_WAITDELETE;
 524                                 rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
 525                         }
 526                         ln->ln_expire = time_uptime +
 527                                 ND_IFINFO(ifp)->retrans / 1000;
 528                         nd6_ns_output(ifp, NULL, &dst->sin6_addr,
 529                                 ln, 0);
 530                         break;
 531                 case ND6_LLINFO_REACHABLE:
 532                         if (ln->ln_expire) {
 533                                 ln->ln_state = ND6_LLINFO_STALE;
 534                                 ln->ln_expire = time_uptime + nd6_gctimer;
 535                         }
 536                         break;
 537
 538                 case ND6_LLINFO_STALE:
 539                         /* Garbage Collection(RFC 2461 5.3) */
 540                         if (ln->ln_expire)
 541                                 next = nd6_free(rt);
 542                         break;
 543
 544                 case ND6_LLINFO_DELAY:
 545                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
 546                                 /* We need NUD */
 547                                 ln->ln_asked = 1;
 548                                 ln->ln_state = ND6_LLINFO_PROBE;
 549                                 ln->ln_expire = time_uptime +
 550                                         ndi->retrans / 1000;
 551                                 nd6_ns_output(ifp, &dst->sin6_addr,
 552                                               &dst->sin6_addr,
 553                                               ln, 0);
 554                         } else {
 555                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
 556                                 ln->ln_expire = time_uptime + nd6_gctimer;
 557                         }
 558                         break;
 559                 case ND6_LLINFO_PROBE:
 560                         if (ln->ln_asked < nd6_umaxtries) {
 561                                 ln->ln_asked++;
 562                                 ln->ln_expire = time_uptime +
 563                                         ND_IFINFO(ifp)->retrans / 1000;
 564                                 nd6_ns_output(ifp, &dst->sin6_addr,
 565                                                &dst->sin6_addr, ln, 0);
 566                         } else {
 567                                 rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
 568                                 next = nd6_free(rt);
 569                         }
 570                         break;
 571                 }
 572                 ln = next;
 573         }
 574
 575         /* expire default router list */
 576         dr = TAILQ_FIRST(&nd_defrouter);
 577         while (dr) {
 578                 if (dr->expire && dr->expire < time_uptime) {
 579                         struct nd_defrouter *t;
 580                         t = TAILQ_NEXT(dr, dr_entry);
 581                         defrtrlist_del(dr);
 582                         dr = t;
 583                 } else {
 584                         dr = TAILQ_NEXT(dr, dr_entry);
 585                 }
 586         }
 587
 588         /*
 589          * expire interface addresses.
 590          * in the past the loop was inside prefix expiry processing.
 591          * However, from a stricter speci-confrmance standpoint, we should
 592          * rather separate address lifetimes and prefix lifetimes.
 593          */
 594 addrloop:
 595         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
 596                 nia6 = ia6->ia_next;
 597                 /* check address lifetime */
 598                 if (IFA6_IS_INVALID(ia6)) {
 599                         int regen = 0;
 600
 601                         /*
 602                          * If the expiring address is temporary, try
 603                          * regenerating a new one.  This would be useful when
 604                          * we suspended a laptop PC, then turned it on after a
 605                          * period that could invalidate all temporary
 606                          * addresses.  Although we may have to restart the
 607                          * loop (see below), it must be after purging the
 608                          * address.  Otherwise, we'd see an infinite loop of
 609                          * regeneration.
 610                          */
 611                         if (ip6_use_tempaddr &&
 612                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
 613                                 if (regen_tmpaddr(ia6) == 0)
 614                                         regen = 1;
 615                         }
 616
 617                         in6_purgeaddr(&ia6->ia_ifa);
 618
 619                         if (regen)
 620                                 goto addrloop; /* XXX: see below */
 621                         /* ia6 is no longer good, continue on to next */
 622                         continue;
 623                 }
 624                 if (IFA6_IS_DEPRECATED(ia6)) {
 625                         int oldflags = ia6->ia6_flags;
 626
 627                         if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
 628                                 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
 629                                 in6_newaddrmsg((struct ifaddr *)ia6);
 630                         }
 631
 632                         /*
 633                          * If a temporary address has just become deprecated,
 634                          * regenerate a new one if possible.
 635                          */
 636                         if (ip6_use_tempaddr &&
 637                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
 638                             !(oldflags & IN6_IFF_DEPRECATED)) {
 639
 640                                 if (regen_tmpaddr(ia6) == 0) {
 641                                         /*
 642                                          * A new temporary address is
 643                                          * generated.
 644                                          * XXX: this means the address chain
 645                                          * has changed while we are still in
 646                                          * the loop.  Although the change
 647                                          * would not cause disaster (because
 648                                          * it's not a deletion, but an
 649                                          * addition,) we'd rather restart the
 650                                          * loop just for safety.  Or does this
 651                                          * significantly reduce performance??
 652                                          */
 653                                         goto addrloop;
 654                                 }
 655                         }
 656                 } else {
 657                         /*
 658                          * A new RA might have made a deprecated address
 659                          * preferred.
 660                          */
 661                         if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
 662                                 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
 663                                 in6_newaddrmsg((struct ifaddr *)ia6);
 664                         }
 665                 }
 666         }
 667
 668         /* expire prefix list */
 669         pr = nd_prefix.lh_first;
 670         while (pr) {
 671                 /*
 672                  * check prefix lifetime.
 673                  * since pltime is just for autoconf, pltime processing for
 674                  * prefix is not necessary.
 675                  */
 676                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
 677                         struct nd_prefix *t;
 678                         t = pr->ndpr_next;
 679
 680                         /*
 681                          * address expiration and prefix expiration are
 682                          * separate.  NEVER perform in6_purgeaddr here.
 683                          */
 684
 685                         prelist_remove(pr);
 686                         pr = t;
 687                 } else
 688                         pr = pr->ndpr_next;
 689         }
 690
 691         mtx_unlock(&nd6_mtx);
 692
 693         callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
 694 }
 695
 696 static void
 697 nd6_timer(void *arg __unused)
 698 {
 699         struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
 700
 701         KASSERT(mycpuid == 0, ("not on cpu0"));
 702         crit_enter();
 703         if (lmsg->ms_flags & MSGF_DONE)
 704                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
 705         crit_exit();
 706 }
 707
 708 void
 709 nd6_timer_init(void)
 710 {
 711         callout_init_mp(&nd6_timer_ch);
 712         netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
 713             MSGF_PRIORITY, nd6_timer_dispatch);
 714         callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
 715 }
 716
 717 static int
 718 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
 719                                          address */
 720 {
 721         struct ifaddr_container *ifac;
 722         struct ifnet *ifp;
 723         struct in6_ifaddr *public_ifa6 = NULL;
 724
 725         ifp = ia6->ia_ifa.ifa_ifp;
 726         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 727                 struct ifaddr *ifa = ifac->ifa;
 728                 struct in6_ifaddr *it6;
 729
 730                 if (ifa->ifa_addr->sa_family != AF_INET6)
 731                         continue;
 732
 733                 it6 = (struct in6_ifaddr *)ifa;
 734
 735                 /* ignore no autoconf addresses. */
 736                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
 737                         continue;
 738
 739                 /* ignore autoconf addresses with different prefixes. */
 740                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
 741                         continue;
 742
 743                 /*
 744                  * Now we are looking at an autoconf address with the same
 745                  * prefix as ours.  If the address is temporary and is still
 746                  * preferred, do not create another one.  It would be rare, but
 747                  * could happen, for example, when we resume a laptop PC after
 748                  * a long period.
 749                  */
 750                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
 751                     !IFA6_IS_DEPRECATED(it6)) {
 752                         public_ifa6 = NULL;
 753                         break;
 754                 }
 755
 756                 /*
 757                  * This is a public autoconf address that has the same prefix
 758                  * as ours.  If it is preferred, keep it.  We can't break the
 759                  * loop here, because there may be a still-preferred temporary
 760                  * address with the prefix.
 761                  */
 762                 if (!IFA6_IS_DEPRECATED(it6))
 763                     public_ifa6 = it6;
 764         }
 765
 766         if (public_ifa6 != NULL) {
 767                 int e;
 768
 769                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
 770                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
 771                             " tmp addr,errno=%d\n", e);
 772                         return (-1);
 773                 }
 774                 return (0);
 775         }
 776
 777         return (-1);
 778 }
 779
 780 /*
 781  * Nuke neighbor cache/prefix/default router management table, right before
 782  * ifp goes away.
 783  */
 784 void
 785 nd6_purge(struct ifnet *ifp)
 786 {
 787         struct llinfo_nd6 *ln, *nln;
 788         struct nd_defrouter *dr, *ndr, drany;
 789         struct nd_prefix *pr, *npr;
 790
 791         /* Nuke default router list entries toward ifp */
 792         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
 793                 /*
 794                  * The first entry of the list may be stored in
 795                  * the routing table, so we'll delete it later.
 796                  */
 797                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
 798                         ndr = TAILQ_NEXT(dr, dr_entry);
 799                         if (dr->ifp == ifp)
 800                                 defrtrlist_del(dr);
 801                 }
 802                 dr = TAILQ_FIRST(&nd_defrouter);
 803                 if (dr->ifp == ifp)
 804                         defrtrlist_del(dr);
 805         }
 806
 807         /* Nuke prefix list entries toward ifp */
 808         for (pr = nd_prefix.lh_first; pr; pr = npr) {
 809                 npr = pr->ndpr_next;
 810                 if (pr->ndpr_ifp == ifp) {
 811                         /*
 812                          * Previously, pr->ndpr_addr is removed as well,
 813                          * but I strongly believe we don't have to do it.
 814                          * nd6_purge() is only called from in6_ifdetach(),
 815                          * which removes all the associated interface addresses
 816                          * by itself.
 817                          * (jinmei@kame.net 20010129)
 818                          */
 819                         prelist_remove(pr);
 820                 }
 821         }
 822
 823         /* cancel default outgoing interface setting */
 824         if (nd6_defifindex == ifp->if_index)
 825                 nd6_setdefaultiface(0);
 826
 827         if (!ip6_forwarding &&
 828             (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV)) {
 829                 /* refresh default router list */
 830                 bzero(&drany, sizeof(drany));
 831                 defrouter_delreq(&drany, 0);
 832                 defrouter_select();
 833         }
 834
 835         /*
 836          * Nuke neighbor cache entries for the ifp.
 837          * Note that rt->rt_ifp may not be the same as ifp,
 838          * due to KAME goto ours hack.  See RTM_RESOLVE case in
 839          * nd6_rtrequest(), and ip6_input().
 840          */
 841         ln = llinfo_nd6.ln_next;
 842         while (ln && ln != &llinfo_nd6) {
 843                 struct rtentry *rt;
 844                 struct sockaddr_dl *sdl;
 845
 846                 nln = ln->ln_next;
 847                 rt = ln->ln_rt;
 848                 if (rt && rt->rt_gateway &&
 849                     rt->rt_gateway->sa_family == AF_LINK) {
 850                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
 851                         if (sdl->sdl_index == ifp->if_index)
 852                                 nln = nd6_free(rt);
 853                 }
 854                 ln = nln;
 855         }
 856 }
 857
 858 struct rtentry *
 859 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
 860 {
 861         struct rtentry *rt;
 862         struct sockaddr_in6 sin6;
 863
 864         bzero(&sin6, sizeof(sin6));
 865         sin6.sin6_len = sizeof(struct sockaddr_in6);
 866         sin6.sin6_family = AF_INET6;
 867         sin6.sin6_addr = *addr6;
 868
 869         if (create)
 870                 rt = rtlookup((struct sockaddr *)&sin6);
 871         else
 872                 rt = rtpurelookup((struct sockaddr *)&sin6);
 873         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
 874                 /*
 875                  * This is the case for the default route.
 876                  * If we want to create a neighbor cache for the address, we
 877                  * should free the route for the destination and allocate an
 878                  * interface route.
 879                  */
 880                 if (create) {
 881                         --rt->rt_refcnt;
 882                         rt = NULL;
 883                 }
 884         }
 885         if (!rt) {
 886                 if (create && ifp) {
 887                         int e;
 888
 889                         /*
 890                          * If no route is available and create is set,
 891                          * we allocate a host route for the destination
 892                          * and treat it like an interface route.
 893                          * This hack is necessary for a neighbor which can't
 894                          * be covered by our own prefix.
 895                          */
 896                         struct ifaddr *ifa;
 897
 898                         ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
 899                         if (ifa == NULL)
 900                                 return (NULL);
 901
 902                         /*
 903                          * Create a new route.  RTF_LLINFO is necessary
 904                          * to create a Neighbor Cache entry for the
 905                          * destination in nd6_rtrequest which will be
 906                          * called in rtrequest via ifa->ifa_rtrequest.
 907                          */
 908                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
 909                              ifa->ifa_addr, (struct sockaddr *)&all1_sa,
 910                              (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
 911                              ~RTF_CLONING, &rt)) != 0) {
 912                                 log(LOG_ERR,
 913                                     "nd6_lookup: failed to add route for a "
 914                                     "neighbor(%s), errno=%d\n",
 915                                     ip6_sprintf(addr6), e);
 916                         }
 917                         if (rt == NULL)
 918                                 return (NULL);
 919                         if (rt->rt_llinfo) {
 920                                 struct llinfo_nd6 *ln =
 921                                     (struct llinfo_nd6 *)rt->rt_llinfo;
 922
 923                                 ln->ln_state = ND6_LLINFO_NOSTATE;
 924                         }
 925                 } else
 926                         return (NULL);
 927         }
 928         rt->rt_refcnt--;
 929
 930         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
 931                 if (create) {
 932                         log(LOG_DEBUG,
 933                             "nd6_lookup: failed to lookup %s (if = %s)\n",
 934                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
 935                         /* xxx more logs... kazu */
 936                 }
 937                 return (NULL);
 938         }
 939         return (rt);
 940 }
 941
 942 static struct rtentry *
 943 nd6_neighbor_lookup(struct in6_addr *addr6, struct ifnet *ifp)
 944 {
 945         struct rtentry *rt;
 946         struct sockaddr_in6 sin6;
 947
 948         bzero(&sin6, sizeof(sin6));
 949         sin6.sin6_len = sizeof(struct sockaddr_in6);
 950         sin6.sin6_family = AF_INET6;
 951         sin6.sin6_addr = *addr6;
 952
 953         rt = rtpurelookup((struct sockaddr *)&sin6);
 954         if (rt == NULL)
 955                 return (NULL);
 956         rt->rt_refcnt--;
 957
 958         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
 959                 if (nd6_onlink_ns_rfc4861 &&
 960                     (ND6_RTENTRY_IS_LLCLONING(rt) ||    /* not cloned yet */
 961                      (rt->rt_parent != NULL &&          /* cloning */
 962                       ND6_RTENTRY_IS_LLCLONING(rt->rt_parent)))) {
 963                         /*
 964                          * If cloning ever happened or is happening,
 965                          * rtentry for addr6 would or will become a
 966                          * neighbor cache.
 967                          */
 968                 } else {
 969                         rt = NULL;
 970                 }
 971         }
 972         return (rt);
 973 }
 974
 975 /*
 976  * Detect if a given IPv6 address identifies a neighbor on a given link.
 977  * XXX: should take care of the destination of a p2p link?
 978  */
 979 int
 980 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
 981 {
 982         struct ifaddr_container *ifac;
 983         int i;
 984
 985 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
 986 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
 987
 988         /*
 989          * A link-local address is always a neighbor.
 990          * XXX: we should use the sin6_scope_id field rather than the embedded
 991          * interface index.
 992          */
 993         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
 994             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
 995                 return (1);
 996
 997         /*
 998          * If the address matches one of our addresses,
 999          * it should be a neighbor.
1000          */
1001         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1002                 struct ifaddr *ifa = ifac->ifa;
1003
1004                 if (ifa->ifa_addr->sa_family != AF_INET6)
1005                         next: continue;
1006
1007                 for (i = 0; i < 4; i++) {
1008                         if ((IFADDR6(ifa).s6_addr32[i] ^
1009                              addr->sin6_addr.s6_addr32[i]) &
1010                             IFMASK6(ifa).s6_addr32[i])
1011                                 goto next;
1012                 }
1013                 return (1);
1014         }
1015
1016         /*
1017          * Even if the address matches none of our addresses, it might be
1018          * in the neighbor cache.
1019          */
1020         if (nd6_neighbor_lookup(&addr->sin6_addr, ifp) != NULL)
1021                 return (1);
1022
1023         return (0);
1024 #undef IFADDR6
1025 #undef IFMASK6
1026 }
1027
1028 /*
1029  * Free an nd6 llinfo entry.
1030  */
1031 struct llinfo_nd6 *
1032 nd6_free(struct rtentry *rt)
1033 {
1034         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1035         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
1036         struct nd_defrouter *dr;
1037
1038         /*
1039          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
1040          * even though it is not harmful, it was not really necessary.
1041          */
1042
1043         /* XXX: this condition too restrictive? */
1044         if (!ip6_forwarding &&
1045             (ND_IFINFO(rt->rt_ifp)->flags & ND6_IFF_ACCEPT_RTADV)) {
1046                 mtx_lock(&nd6_mtx);
1047                 dr = defrouter_lookup(
1048                     &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1049                     rt->rt_ifp);
1050
1051                 if (ln->ln_router || dr) {
1052                         /*
1053                          * rt6_flush must be called whether or not the neighbor
1054                          * is in the Default Router List.
1055                          * See a corresponding comment in nd6_na_input().
1056                          */
1057                         rt6_flush(&in6, rt->rt_ifp);
1058                 }
1059
1060                 if (dr) {
1061                         /*
1062                          * Unreachablity of a router might affect the default
1063                          * router selection and on-link detection of advertised
1064                          * prefixes.
1065                          */
1066
1067                         /*
1068                          * Temporarily fake the state to choose a new default
1069                          * router and to perform on-link determination of
1070                          * prefixes correctly.
1071                          * Below the state will be set correctly,
1072                          * or the entry itself will be deleted.
1073                          */
1074                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
1075
1076                         /*
1077                          * Since defrouter_select() does not affect the
1078                          * on-link determination and MIP6 needs the check
1079                          * before the default router selection, we perform
1080                          * the check now.
1081                          */
1082                         pfxlist_onlink_check();
1083
1084                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1085                                 /*
1086                                  * It is used as the current default router,
1087                                  * so we have to move it to the end of the
1088                                  * list and choose a new one.
1089                                  * XXX: it is not very efficient if this is
1090                                  *      the only router.
1091                                  */
1092                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1093                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1094
1095                                 defrouter_select();
1096                         }
1097                 }
1098                 mtx_unlock(&nd6_mtx);
1099         }
1100
1101         /*
1102          * Before deleting the entry, remember the next entry as the
1103          * return value.  We need this because pfxlist_onlink_check() above
1104          * might have freed other entries (particularly the old next entry) as
1105          * a side effect (XXX).
1106          */
1107         next = ln->ln_next;
1108
1109         /*
1110          * Detach the route from the routing tree and the list of neighbor
1111          * caches, and disable the route entry not to be used in already
1112          * cached routes.
1113          *
1114          * ND expiry happens under one big timer.
1115          * To avoid overflowing the route socket, don't report this.
1116          * Now that RTM_MISS is reported when an address is unresolvable
1117          * the benefit of reporting this deletion is questionable.
1118          */
1119         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1120
1121         return (next);
1122 }
1123
1124 /*
1125  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1126  *
1127  * XXX cost-effective metods?
1128  */
1129 void
1130 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1131 {
1132         struct llinfo_nd6 *ln;
1133
1134         /*
1135          * If the caller specified "rt", use that.  Otherwise, resolve the
1136          * routing table by supplied "dst6".
1137          */
1138         if (!rt) {
1139                 if (!dst6)
1140                         return;
1141                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1142                         return;
1143         }
1144
1145         if ((rt->rt_flags & RTF_GATEWAY) ||
1146             !(rt->rt_flags & RTF_LLINFO) ||
1147             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1148             rt->rt_gateway->sa_family != AF_LINK) {
1149                 /* This is not a host route. */
1150                 return;
1151         }
1152
1153         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1154         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1155                 return;
1156
1157         /*
1158          * if we get upper-layer reachability confirmation many times,
1159          * it is possible we have false information.
1160          */
1161         if (!force) {
1162                 ln->ln_byhint++;
1163                 if (ln->ln_byhint > nd6_maxnudhint)
1164                         return;
1165         }
1166
1167         ln->ln_state = ND6_LLINFO_REACHABLE;
1168         if (ln->ln_expire)
1169                 ln->ln_expire = time_uptime +
1170                         ND_IFINFO(rt->rt_ifp)->reachable;
1171 }
1172
1173 void
1174 nd6_rtrequest(int req, struct rtentry *rt)
1175 {
1176         struct sockaddr *gate = rt->rt_gateway;
1177         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1178         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1179         struct ifnet *ifp = rt->rt_ifp;
1180         struct ifaddr *ifa;
1181
1182         if ((rt->rt_flags & RTF_GATEWAY))
1183                 return;
1184
1185         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1186                 /*
1187                  * This is probably an interface direct route for a link
1188                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1189                  * We do not need special treatment below for such a route.
1190                  * Moreover, the RTF_LLINFO flag which would be set below
1191                  * would annoy the ndp(8) command.
1192                  */
1193                 return;
1194         }
1195
1196         if (req == RTM_RESOLVE &&
1197             (nd6_need_cache(ifp) == 0 || /* stf case */
1198              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1199                 /*
1200                  * FreeBSD and BSD/OS often make a cloned host route based
1201                  * on a less-specific route (e.g. the default route).
1202                  * If the less specific route does not have a "gateway"
1203                  * (this is the case when the route just goes to a p2p or an
1204                  * stf interface), we'll mistakenly make a neighbor cache for
1205                  * the host route, and will see strange neighbor solicitation
1206                  * for the corresponding destination.  In order to avoid the
1207                  * confusion, we check if the destination of the route is
1208                  * a neighbor in terms of neighbor discovery, and stop the
1209                  * process if not.  Additionally, we remove the LLINFO flag
1210                  * so that ndp(8) will not try to get the neighbor information
1211                  * of the destination.
1212                  */
1213                 rt->rt_flags &= ~RTF_LLINFO;
1214                 return;
1215         }
1216
1217         switch (req) {
1218         case RTM_ADD:
1219                 /*
1220                  * There is no backward compatibility :)
1221                  *
1222                  * if (!(rt->rt_flags & RTF_HOST) &&
1223                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1224                  *         rt->rt_flags |= RTF_CLONING;
1225                  */
1226                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1227                         /*
1228                          * Case 1: This route should come from
1229                          * a route to interface.  RTF_LLINFO flag is set
1230                          * for a host route whose destination should be
1231                          * treated as on-link.
1232                          */
1233                         rt_setgate(rt, rt_key(rt),
1234                                    (struct sockaddr *)&null_sdl);
1235                         gate = rt->rt_gateway;
1236                         SDL(gate)->sdl_type = ifp->if_type;
1237                         SDL(gate)->sdl_index = ifp->if_index;
1238                         if (ln)
1239                                 ln->ln_expire = time_uptime;
1240                         if (ln && ln->ln_expire == 0) {
1241                                 /* kludge for desktops */
1242                                 ln->ln_expire = 1;
1243                         }
1244                         if ((rt->rt_flags & RTF_CLONING))
1245                                 break;
1246                 }
1247                 /*
1248                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1249                  * We don't do that here since llinfo is not ready yet.
1250                  *
1251                  * There are also couple of other things to be discussed:
1252                  * - unsolicited NA code needs improvement beforehand
1253                  * - RFC2461 says we MAY send multicast unsolicited NA
1254                  *   (7.2.6 paragraph 4), however, it also says that we
1255                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1256                  *   we don't have anything like it right now.
1257                  *   note that the mechanism needs a mutual agreement
1258                  *   between proxies, which means that we need to implement
1259                  *   a new protocol, or a new kludge.
1260                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1261                  *   we need to check ip6forwarding before sending it.
1262                  *   (or should we allow proxy ND configuration only for
1263                  *   routers?  there's no mention about proxy ND from hosts)
1264                  */
1265 #if 0
1266                 /* XXX it does not work */
1267                 if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
1268                         nd6_na_output(ifp,
1269                               &SIN6(rt_key(rt))->sin6_addr,
1270                               &SIN6(rt_key(rt))->sin6_addr,
1271                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1272                               1, NULL);
1273                 }
1274 #endif
1275                 /* FALLTHROUGH */
1276         case RTM_RESOLVE:
1277                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1278                         /*
1279                          * Address resolution isn't necessary for a point to
1280                          * point link, so we can skip this test for a p2p link.
1281                          */
1282                         if (gate->sa_family != AF_LINK ||
1283                             gate->sa_len < sizeof(null_sdl)) {
1284                                 log(LOG_DEBUG,
1285                                     "nd6_rtrequest: bad gateway value: %s\n",
1286                                     if_name(ifp));
1287                                 break;
1288                         }
1289                         SDL(gate)->sdl_type = ifp->if_type;
1290                         SDL(gate)->sdl_index = ifp->if_index;
1291                 }
1292                 if (ln != NULL)
1293                         break;  /* This happens on a route change */
1294                 /*
1295                  * Case 2: This route may come from cloning, or a manual route
1296                  * add with a LL address.
1297                  */
1298                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1299                 rt->rt_llinfo = (caddr_t)ln;
1300                 if (!ln) {
1301                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1302                         break;
1303                 }
1304                 nd6_inuse++;
1305                 nd6_allocated++;
1306                 bzero(ln, sizeof(*ln));
1307                 ln->ln_rt = rt;
1308                 /* this is required for "ndp" command. - shin */
1309                 if (req == RTM_ADD) {
1310                         /*
1311                          * gate should have some valid AF_LINK entry,
1312                          * and ln->ln_expire should have some lifetime
1313                          * which is specified by ndp command.
1314                          */
1315                         ln->ln_state = ND6_LLINFO_REACHABLE;
1316                         ln->ln_byhint = 0;
1317                 } else {
1318                         /*
1319                          * When req == RTM_RESOLVE, rt is created and
1320                          * initialized in rtrequest(), so rt_expire is 0.
1321                          */
1322                         ln->ln_state = ND6_LLINFO_NOSTATE;
1323                         ln->ln_expire = time_uptime;
1324                 }
1325                 rt->rt_flags |= RTF_LLINFO;
1326                 ln->ln_next = llinfo_nd6.ln_next;
1327                 llinfo_nd6.ln_next = ln;
1328                 ln->ln_prev = &llinfo_nd6;
1329                 ln->ln_next->ln_prev = ln;
1330
1331                 /*
1332                  * check if rt_key(rt) is one of my address assigned
1333                  * to the interface.
1334                  */
1335                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1336                     &SIN6(rt_key(rt))->sin6_addr);
1337                 if (ifa) {
1338                         caddr_t macp = nd6_ifptomac(ifp);
1339                         ln->ln_expire = 0;
1340                         ln->ln_state = ND6_LLINFO_REACHABLE;
1341                         ln->ln_byhint = 0;
1342                         if (macp) {
1343                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1344                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1345                         }
1346                         if (nd6_useloopback) {
1347                                 rt->rt_ifp = loif;      /* XXX */
1348                                 /*
1349                                  * Make sure rt_ifa be equal to the ifaddr
1350                                  * corresponding to the address.
1351                                  * We need this because when we refer
1352                                  * rt_ifa->ia6_flags in ip6_input, we assume
1353                                  * that the rt_ifa points to the address instead
1354                                  * of the loopback address.
1355                                  */
1356                                 if (ifa != rt->rt_ifa) {
1357                                         IFAFREE(rt->rt_ifa);
1358                                         IFAREF(ifa);
1359                                         rt->rt_ifa = ifa;
1360                                 }
1361                         }
1362                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1363                         ln->ln_expire = 0;
1364                         ln->ln_state = ND6_LLINFO_REACHABLE;
1365                         ln->ln_byhint = 0;
1366
1367                         /*
1368                          * Join solicited node multicast for proxy ND, and only
1369                          * join it once on cpu0.
1370                          */
1371                         if ((ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1372                                 struct in6_addr llsol;
1373                                 int error;
1374
1375                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1376                                 llsol.s6_addr16[0] = htons(0xff02);
1377                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1378                                 llsol.s6_addr32[1] = 0;
1379                                 llsol.s6_addr32[2] = htonl(1);
1380                                 llsol.s6_addr8[12] = 0xff;
1381
1382                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1383                                         nd6log((LOG_ERR, "%s: failed to join "
1384                                             "%s (errno=%d)\n", if_name(ifp),
1385                                             ip6_sprintf(&llsol), error));
1386                                 }
1387                         }
1388                 }
1389                 break;
1390
1391         case RTM_DELETE:
1392                 if (!ln)
1393                         break;
1394                 /*
1395                  * Leave from solicited node multicast for proxy ND, and only
1396                  * leave it once on cpu0 (since we joined it once on cpu0).
1397                  */
1398                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1399                     (ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1400                         struct in6_addr llsol;
1401                         struct in6_multi *in6m;
1402
1403                         llsol = SIN6(rt_key(rt))->sin6_addr;
1404                         llsol.s6_addr16[0] = htons(0xff02);
1405                         llsol.s6_addr16[1] = htons(ifp->if_index);
1406                         llsol.s6_addr32[1] = 0;
1407                         llsol.s6_addr32[2] = htonl(1);
1408                         llsol.s6_addr8[12] = 0xff;
1409
1410                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1411                         if (in6m)
1412                                 in6_delmulti(in6m);
1413                 }
1414                 nd6_inuse--;
1415                 ln->ln_next->ln_prev = ln->ln_prev;
1416                 ln->ln_prev->ln_next = ln->ln_next;
1417                 ln->ln_prev = NULL;
1418                 rt->rt_llinfo = 0;
1419                 rt->rt_flags &= ~RTF_LLINFO;
1420                 if (ln->ln_hold)
1421                         m_freem(ln->ln_hold);
1422                 R_Free(ln);
1423         }
1424 }
1425
1426 int
1427 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1428 {
1429         struct in6_drlist *drl = (struct in6_drlist *)data;
1430         struct in6_prlist *prl = (struct in6_prlist *)data;
1431         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1432         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1433         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1434         struct nd_defrouter *dr, any;
1435         struct nd_prefix *pr;
1436         struct rtentry *rt;
1437         int i = 0, error = 0;
1438
1439         switch (cmd) {
1440         case SIOCGDRLST_IN6:
1441                 /*
1442                  * obsolete API, use sysctl under net.inet6.icmp6
1443                  */
1444                 bzero(drl, sizeof(*drl));
1445                 mtx_lock(&nd6_mtx);
1446                 dr = TAILQ_FIRST(&nd_defrouter);
1447                 while (dr && i < DRLSTSIZ) {
1448                         drl->defrouter[i].rtaddr = dr->rtaddr;
1449                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1450                                 /* XXX: need to this hack for KAME stack */
1451                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1452                         } else
1453                                 log(LOG_ERR,
1454                                     "default router list contains a "
1455                                     "non-linklocal address(%s)\n",
1456                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1457
1458                         drl->defrouter[i].flags = dr->flags;
1459                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1460                         drl->defrouter[i].expire = dr->expire;
1461                         drl->defrouter[i].if_index = dr->ifp->if_index;
1462                         i++;
1463                         dr = TAILQ_NEXT(dr, dr_entry);
1464                 }
1465                 mtx_unlock(&nd6_mtx);
1466                 break;
1467         case SIOCGPRLST_IN6:
1468                 /*
1469                  * obsolete API, use sysctl under net.inet6.icmp6
1470                  */
1471                 /*
1472                  * XXX meaning of fields, especialy "raflags", is very
1473                  * differnet between RA prefix list and RR/static prefix list.
1474                  * how about separating ioctls into two?
1475                  */
1476                 bzero(prl, sizeof(*prl));
1477                 mtx_lock(&nd6_mtx);
1478                 pr = nd_prefix.lh_first;
1479                 while (pr && i < PRLSTSIZ) {
1480                         struct nd_pfxrouter *pfr;
1481                         int j;
1482
1483                         in6_embedscope(&prl->prefix[i].prefix,
1484                             &pr->ndpr_prefix, NULL, NULL);
1485                         prl->prefix[i].raflags = pr->ndpr_raf;
1486                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1487                         prl->prefix[i].vltime = pr->ndpr_vltime;
1488                         prl->prefix[i].pltime = pr->ndpr_pltime;
1489                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1490                         prl->prefix[i].expire = pr->ndpr_expire;
1491
1492                         pfr = pr->ndpr_advrtrs.lh_first;
1493                         j = 0;
1494                         while (pfr) {
1495                                 if (j < DRLSTSIZ) {
1496 #define RTRADDR prl->prefix[i].advrtr[j]
1497                                         RTRADDR = pfr->router->rtaddr;
1498                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1499                                                 /* XXX: hack for KAME */
1500                                                 RTRADDR.s6_addr16[1] = 0;
1501                                         } else
1502                                                 log(LOG_ERR,
1503                                                     "a router(%s) advertises "
1504                                                     "a prefix with "
1505                                                     "non-link local address\n",
1506                                                     ip6_sprintf(&RTRADDR));
1507 #undef RTRADDR
1508                                 }
1509                                 j++;
1510                                 pfr = pfr->pfr_next;
1511                         }
1512                         prl->prefix[i].advrtrs = j;
1513                         prl->prefix[i].origin = PR_ORIG_RA;
1514
1515                         i++;
1516                         pr = pr->ndpr_next;
1517                 }
1518                 mtx_unlock(&nd6_mtx);
1519
1520                 break;
1521         case OSIOCGIFINFO_IN6:
1522                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1523                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1524                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1525                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1526                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1527                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1528                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1529                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1530                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1531                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1532                 break;
1533         case SIOCGIFINFO_IN6:
1534                 ndi->ndi = *ND_IFINFO(ifp);
1535                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1536                 break;
1537         case SIOCSIFINFO_IN6:
1538                 /*
1539                  * used to change host variables from userland.
1540                  * intented for a use on router to reflect RA configurations.
1541                  */
1542                 /* 0 means 'unspecified' */
1543                 if (ndi->ndi.linkmtu != 0) {
1544                         if (ndi->ndi.linkmtu < IPV6_MMTU ||
1545                             ndi->ndi.linkmtu > IN6_LINKMTU(ifp)) {
1546                                 error = EINVAL;
1547                                 break;
1548                         }
1549                         ND_IFINFO(ifp)->linkmtu = ndi->ndi.linkmtu;
1550                 }
1551
1552                 if (ndi->ndi.basereachable != 0) {
1553                         int obasereachable = ND_IFINFO(ifp)->basereachable;
1554
1555                         ND_IFINFO(ifp)->basereachable = ndi->ndi.basereachable;
1556                         if (ndi->ndi.basereachable != obasereachable)
1557                                 ND_IFINFO(ifp)->reachable =
1558                                     ND_COMPUTE_RTIME(ndi->ndi.basereachable);
1559                 }
1560                 if (ndi->ndi.retrans != 0)
1561                         ND_IFINFO(ifp)->retrans = ndi->ndi.retrans;
1562                 if (ndi->ndi.chlim != 0)
1563                         ND_IFINFO(ifp)->chlim = ndi->ndi.chlim;
1564                 /* FALLTHROUGH */
1565         case SIOCSIFINFO_FLAGS:
1566                 if ((ndi->ndi.flags & ND6_IFF_AUTO_LINKLOCAL) &&
1567                     !(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1568                         /* auto_linklocal 0->1 transision */
1569                         ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1570                         in6_ifattach(ifp, NULL);
1571                 }
1572                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1573                 break;
1574         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1575                 /* flush default router list */
1576                 /*
1577                  * xxx sumikawa: should not delete route if default
1578                  * route equals to the top of default router list
1579                  */
1580                 bzero(&any, sizeof(any));
1581                 defrouter_delreq(&any, 0);
1582                 defrouter_select();
1583                 /* xxx sumikawa: flush prefix list */
1584                 break;
1585         case SIOCSPFXFLUSH_IN6:
1586         {
1587                 /* flush all the prefix advertised by routers */
1588                 struct nd_prefix *pr, *next;
1589
1590                 mtx_lock(&nd6_mtx);
1591                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1592                         struct in6_ifaddr *ia, *ia_next;
1593
1594                         next = pr->ndpr_next;
1595
1596                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1597                                 continue; /* XXX */
1598
1599                         /* do we really have to remove addresses as well? */
1600                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1601                                 /* ia might be removed.  keep the next ptr. */
1602                                 ia_next = ia->ia_next;
1603
1604                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1605                                         continue;
1606
1607                                 if (ia->ia6_ndpr == pr)
1608                                         in6_purgeaddr(&ia->ia_ifa);
1609                         }
1610                         prelist_remove(pr);
1611                 }
1612                 mtx_unlock(&nd6_mtx);
1613                 break;
1614         }
1615         case SIOCSRTRFLUSH_IN6:
1616         {
1617                 /* flush all the default routers */
1618                 struct nd_defrouter *dr, *next;
1619
1620                 mtx_lock(&nd6_mtx);
1621                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1622                         /*
1623                          * The first entry of the list may be stored in
1624                          * the routing table, so we'll delete it later.
1625                          */
1626                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1627                                 next = TAILQ_NEXT(dr, dr_entry);
1628                                 defrtrlist_del(dr);
1629                         }
1630                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1631                 }
1632                 mtx_unlock(&nd6_mtx);
1633                 break;
1634         }
1635         case SIOCGNBRINFO_IN6:
1636         {
1637                 struct llinfo_nd6 *ln;
1638                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1639
1640                 /*
1641                  * XXX: KAME specific hack for scoped addresses
1642                  *      XXXX: for other scopes than link-local?
1643                  */
1644                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1645                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1646                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1647
1648                         if (*idp == 0)
1649                                 *idp = htons(ifp->if_index);
1650                 }
1651
1652                 mtx_lock(&nd6_mtx);
1653                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1654                         error = EINVAL;
1655                         mtx_unlock(&nd6_mtx);
1656                         break;
1657                 }
1658                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1659                 nbi->state = ln->ln_state;
1660                 nbi->asked = ln->ln_asked;
1661                 nbi->isrouter = ln->ln_router;
1662                 nbi->expire = ln->ln_expire;
1663                 mtx_unlock(&nd6_mtx);
1664
1665                 break;
1666         }
1667         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1668                 ndif->ifindex = nd6_defifindex;
1669                 break;
1670         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1671                 return (nd6_setdefaultiface(ndif->ifindex));
1672         }
1673         return (error);
1674 }
1675
1676 /*
1677  * Create neighbor cache entry and cache link-layer address,
1678  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1679  */
1680 struct rtentry *
1681 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1682                  int lladdrlen,
1683                  int type,      /* ICMP6 type */
1684                  int code       /* type dependent information */)
1685 {
1686         struct rtentry *rt = NULL;
1687         struct llinfo_nd6 *ln = NULL;
1688         int is_newentry;
1689         struct sockaddr_dl *sdl = NULL;
1690         int do_update;
1691         int olladdr;
1692         int llchange;
1693         int newstate = 0;
1694
1695         if (!ifp)
1696                 panic("ifp == NULL in nd6_cache_lladdr");
1697         if (!from)
1698                 panic("from == NULL in nd6_cache_lladdr");
1699
1700         /* nothing must be updated for unspecified address */
1701         if (IN6_IS_ADDR_UNSPECIFIED(from))
1702                 return NULL;
1703
1704         /*
1705          * Validation about ifp->if_addrlen and lladdrlen must be done in
1706          * the caller.
1707          *
1708          * XXX If the link does not have link-layer adderss, what should
1709          * we do? (ifp->if_addrlen == 0)
1710          * Spec says nothing in sections for RA, RS and NA.  There's small
1711          * description on it in NS section (RFC 2461 7.2.3).
1712          */
1713
1714         rt = nd6_lookup(from, 0, ifp);
1715         if (!rt) {
1716 #if 0
1717                 /* nothing must be done if there's no lladdr */
1718                 if (!lladdr || !lladdrlen)
1719                         return NULL;
1720 #endif
1721
1722                 rt = nd6_lookup(from, 1, ifp);
1723                 is_newentry = 1;
1724         } else {
1725                 /* do nothing if static ndp is set */
1726                 if (rt->rt_flags & RTF_STATIC)
1727                         return NULL;
1728                 is_newentry = 0;
1729         }
1730
1731         if (!rt)
1732                 return NULL;
1733         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1734 fail:
1735                 nd6_free(rt);
1736                 return NULL;
1737         }
1738         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1739         if (!ln)
1740                 goto fail;
1741         if (!rt->rt_gateway)
1742                 goto fail;
1743         if (rt->rt_gateway->sa_family != AF_LINK)
1744                 goto fail;
1745         sdl = SDL(rt->rt_gateway);
1746
1747         olladdr = (sdl->sdl_alen) ? 1 : 0;
1748         if (olladdr && lladdr) {
1749                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1750                         llchange = 1;
1751                 else
1752                         llchange = 0;
1753         } else
1754                 llchange = 0;
1755
1756         /*
1757          * newentry olladdr  lladdr  llchange   (*=record)
1758          *      0       n       n       --      (1)
1759          *      0       y       n       --      (2)
1760          *      0       n       y       --      (3) * STALE
1761          *      0       y       y       n       (4) *
1762          *      0       y       y       y       (5) * STALE
1763          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1764          *      1       --      y       --      (7) * STALE
1765          */
1766
1767         if (lladdr) {           /* (3-5) and (7) */
1768                 /*
1769                  * Record source link-layer address
1770                  * XXX is it dependent to ifp->if_type?
1771                  */
1772                 sdl->sdl_alen = ifp->if_addrlen;
1773                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1774         }
1775
1776         if (!is_newentry) {
1777                 if ((!olladdr && lladdr) ||             /* (3) */
1778                     (olladdr && lladdr && llchange)) {  /* (5) */
1779                         do_update = 1;
1780                         newstate = ND6_LLINFO_STALE;
1781                 } else {                                /* (1-2,4) */
1782                         do_update = 0;
1783                 }
1784         } else {
1785                 do_update = 1;
1786                 if (!lladdr)                            /* (6) */
1787                         newstate = ND6_LLINFO_NOSTATE;
1788                 else                                    /* (7) */
1789                         newstate = ND6_LLINFO_STALE;
1790         }
1791
1792         if (do_update) {
1793                 /*
1794                  * Update the state of the neighbor cache.
1795                  */
1796                 ln->ln_state = newstate;
1797
1798                 if (ln->ln_state == ND6_LLINFO_STALE) {
1799                         /*
1800                          * XXX: since nd6_output() below will cause
1801                          * state tansition to DELAY and reset the timer,
1802                          * we must set the timer now, although it is actually
1803                          * meaningless.
1804                          */
1805                         ln->ln_expire = time_uptime + nd6_gctimer;
1806
1807                         if (ln->ln_hold) {
1808                                 /*
1809                                  * we assume ifp is not a p2p here, so just
1810                                  * set the 2nd argument as the 1st one.
1811                                  */
1812                                 nd6_output(ifp, ifp, ln->ln_hold,
1813                                     (struct sockaddr_in6 *)rt_key(rt), rt);
1814                                 ln->ln_hold = NULL;
1815                         }
1816                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1817                         /* probe right away */
1818                         ln->ln_expire = time_uptime;
1819                 }
1820         }
1821
1822         /*
1823          * ICMP6 type dependent behavior.
1824          *
1825          * NS: clear IsRouter if new entry
1826          * RS: clear IsRouter
1827          * RA: set IsRouter if there's lladdr
1828          * redir: clear IsRouter if new entry
1829          *
1830          * RA case, (1):
1831          * The spec says that we must set IsRouter in the following cases:
1832          * - If lladdr exist, set IsRouter.  This means (1-5).
1833          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1834          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1835          * A quetion arises for (1) case.  (1) case has no lladdr in the
1836          * neighbor cache, this is similar to (6).
1837          * This case is rare but we figured that we MUST NOT set IsRouter.
1838          *
1839          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1840          *                                                      D R
1841          *      0       n       n       --      (1)     c   ?     s
1842          *      0       y       n       --      (2)     c   s     s
1843          *      0       n       y       --      (3)     c   s     s
1844          *      0       y       y       n       (4)     c   s     s
1845          *      0       y       y       y       (5)     c   s     s
1846          *      1       --      n       --      (6) c   c       c s
1847          *      1       --      y       --      (7) c   c   s   c s
1848          *
1849          *                                      (c=clear s=set)
1850          */
1851         switch (type & 0xff) {
1852         case ND_NEIGHBOR_SOLICIT:
1853                 /*
1854                  * New entry must have is_router flag cleared.
1855                  */
1856                 if (is_newentry)        /* (6-7) */
1857                         ln->ln_router = 0;
1858                 break;
1859         case ND_REDIRECT:
1860                 /*
1861                  * If the icmp is a redirect to a better router, always set the
1862                  * is_router flag.  Otherwise, if the entry is newly created,
1863                  * clear the flag.  [RFC 2461, sec 8.3]
1864                  */
1865                 if (code == ND_REDIRECT_ROUTER)
1866                         ln->ln_router = 1;
1867                 else if (is_newentry) /* (6-7) */
1868                         ln->ln_router = 0;
1869                 break;
1870         case ND_ROUTER_SOLICIT:
1871                 /*
1872                  * is_router flag must always be cleared.
1873                  */
1874                 ln->ln_router = 0;
1875                 break;
1876         case ND_ROUTER_ADVERT:
1877                 /*
1878                  * Mark an entry with lladdr as a router.
1879                  */
1880                 if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
1881                     (is_newentry && lladdr)) {                  /* (7) */
1882                         ln->ln_router = 1;
1883                 }
1884                 break;
1885         }
1886
1887         if (llchange || lladdr)
1888                 rt_rtmsg(llchange ? RTM_CHANGE : RTM_ADD, rt, rt->rt_ifp, 0);
1889
1890         /*
1891          * When the link-layer address of a router changes, select the
1892          * best router again.  In particular, when the neighbor entry is newly
1893          * created, it might affect the selection policy.
1894          * Question: can we restrict the first condition to the "is_newentry"
1895          * case?
1896          * XXX: when we hear an RA from a new router with the link-layer
1897          * address option, defrouter_select() is called twice, since
1898          * defrtrlist_update called the function as well.  However, I believe
1899          * we can compromise the overhead, since it only happens the first
1900          * time.
1901          * XXX: although defrouter_select() should not have a bad effect
1902          * for those are not autoconfigured hosts, we explicitly avoid such
1903          * cases for safety.
1904          */
1905         if (do_update && ln->ln_router && !ip6_forwarding &&
1906             (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV))
1907                 defrouter_select();
1908
1909         return rt;
1910 }
1911
1912 static void
1913 nd6_slowtimo(void *arg __unused)
1914 {
1915         struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1916
1917         KASSERT(mycpuid == 0, ("not on cpu0"));
1918         crit_enter();
1919         if (lmsg->ms_flags & MSGF_DONE)
1920                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1921         crit_exit();
1922 }
1923
1924 static void
1925 nd6_slowtimo_dispatch(netmsg_t nmsg)
1926 {
1927         const struct ifnet_array *arr;
1928         struct nd_ifinfo *nd6if;
1929         int i;
1930
1931         ASSERT_NETISR0;
1932
1933         crit_enter();
1934         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1935         crit_exit();
1936
1937         arr = ifnet_array_get();
1938
1939         mtx_lock(&nd6_mtx);
1940         for (i = 0; i < arr->ifnet_count; ++i) {
1941                 struct ifnet *ifp = arr->ifnet_arr[i];
1942
1943                 if (ifp->if_afdata[AF_INET6] == NULL)
1944                         continue;
1945                 nd6if = ND_IFINFO(ifp);
1946                 if (nd6if->basereachable && /* already initialized */
1947                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1948                         /*
1949                          * Since reachable time rarely changes by router
1950                          * advertisements, we SHOULD insure that a new random
1951                          * value gets recomputed at least once every few hours.
1952                          * (RFC 2461, 6.3.4)
1953                          */
1954                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1955                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1956                 }
1957         }
1958         mtx_unlock(&nd6_mtx);
1959
1960         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1961             nd6_slowtimo, NULL);
1962 }
1963
1964 int
1965 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1966            struct sockaddr_in6 *dst, struct rtentry *rt)
1967 {
1968         int error;
1969
1970         if (ifp->if_flags & IFF_LOOPBACK)
1971                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1972         else
1973                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1974         return error;
1975 }
1976
1977 int
1978 nd6_resolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1979     struct sockaddr *dst0, u_char *desten)
1980 {
1981         struct sockaddr_in6 *dst = SIN6(dst0);
1982         struct rtentry *rt = NULL;
1983         struct llinfo_nd6 *ln = NULL;
1984         int error;
1985
1986         if (m->m_flags & M_MCAST) {
1987                 switch (ifp->if_type) {
1988                 case IFT_ETHER:
1989 #ifdef IFT_L2VLAN
1990                 case IFT_L2VLAN:
1991 #endif
1992 #ifdef IFT_IEEE80211
1993                 case IFT_IEEE80211:
1994 #endif
1995                         ETHER_MAP_IPV6_MULTICAST(&dst->sin6_addr,
1996                                                  desten);
1997                         return 0;
1998                 case IFT_IEEE1394:
1999                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2000                         return 0;
2001                 default:
2002                         error = EAFNOSUPPORT;
2003                         goto bad;
2004                 }
2005         }
2006
2007         if (rt0 != NULL) {
2008                 error = rt_llroute(dst0, rt0, &rt);
2009                 if (error != 0)
2010                         goto bad;
2011                 ln = rt->rt_llinfo;
2012         }
2013
2014         /*
2015          * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2016          * the condition below is not very efficient.  But we believe
2017          * it is tolerable, because this should be a rare case.
2018          */
2019         if (ln == NULL && nd6_is_addr_neighbor(dst, ifp)) {
2020                 rt = nd6_lookup(&dst->sin6_addr, 1, ifp);
2021                 if (rt != NULL)
2022                         ln = rt->rt_llinfo;
2023         }
2024
2025         if (ln == NULL || rt == NULL) {
2026                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
2027                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2028                         log(LOG_DEBUG,
2029                             "nd6_output: can't allocate llinfo for %s "
2030                             "(ln=%p, rt=%p)\n",
2031                             ip6_sprintf(&dst->sin6_addr), ln, rt);
2032                         error = ENOBUFS;
2033                         goto bad;
2034                 }
2035                 return 0;
2036         }
2037
2038         /* We don't have to do link-layer address resolution on a p2p link. */
2039         if ((ifp->if_flags & IFF_POINTOPOINT) &&
2040             ln->ln_state < ND6_LLINFO_REACHABLE) {
2041                 ln->ln_state = ND6_LLINFO_STALE;
2042                 ln->ln_expire = time_uptime + nd6_gctimer;
2043         }
2044
2045         /*
2046          * The first time we send a packet to a neighbor whose entry is
2047          * STALE, we have to change the state to DELAY and a sets a timer to
2048          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2049          * neighbor unreachability detection on expiration.
2050          * (RFC 2461 7.3.3)
2051          */
2052         if (ln->ln_state == ND6_LLINFO_STALE) {
2053                 ln->ln_asked = 0;
2054                 ln->ln_state = ND6_LLINFO_DELAY;
2055                 ln->ln_expire = time_uptime + nd6_delay;
2056         }
2057
2058         /*
2059          * If the neighbor cache entry has a state other than INCOMPLETE
2060          * (i.e. its link-layer address is already resolved), return it.
2061          */
2062         if (ln->ln_state > ND6_LLINFO_INCOMPLETE) {
2063                 struct sockaddr_dl *sdl = SDL(rt->rt_gateway);
2064
2065                 KKASSERT(sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0);
2066                 bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2067                 return 0;
2068         }
2069
2070         /*
2071          * There is a neighbor cache entry, but no ethernet address
2072          * response yet.  Replace the held mbuf (if any) with this
2073          * latest one.
2074          */
2075         if (ln->ln_hold)
2076                 m_freem(ln->ln_hold);
2077         ln->ln_hold = m;
2078
2079         /*
2080          * This code conforms to the rate-limiting rule described in Section
2081          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2082          * an NS below.
2083          */
2084         if (ln->ln_state == ND6_LLINFO_NOSTATE ||
2085             ln->ln_state == ND6_LLINFO_WAITDELETE) {
2086                 /*
2087                  * This neighbor cache entry was just created; change its
2088                  * state to INCOMPLETE and start its life cycle.
2089                  *
2090                  * We force an NS output below by setting ln_expire to 1
2091                  * (nd6_rtrequest() could set it to the current time_uptime)
2092                  * and zeroing out ln_asked (XXX this may not be necessary).
2093                  */
2094                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
2095                 ln->ln_expire = 1;
2096                 ln->ln_asked = 0;
2097         }
2098         if (ln->ln_expire && ln->ln_expire < time_uptime && ln->ln_asked == 0) {
2099                 ln->ln_asked++;
2100                 ln->ln_expire = time_uptime + ND_IFINFO(ifp)->retrans / 1000;
2101                 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2102         }
2103
2104         if (ln->ln_asked >= nd6_mmaxtries)
2105                 return (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
2106                     EHOSTUNREACH : EHOSTDOWN;
2107         return EWOULDBLOCK;
2108
2109 bad:
2110         m_freem(m);
2111         return error;
2112 }
2113
2114 int
2115 nd6_need_cache(struct ifnet *ifp)
2116 {
2117         /*
2118          * XXX: we currently do not make neighbor cache on any interface
2119          * other than Ethernet and GIF.
2120          *
2121          * RFC2893 says:
2122          * - unidirectional tunnels needs no ND
2123          */
2124         switch (ifp->if_type) {
2125         case IFT_ETHER:
2126         case IFT_IEEE1394:
2127 #ifdef IFT_L2VLAN
2128         case IFT_L2VLAN:
2129 #endif
2130 #ifdef IFT_IEEE80211
2131         case IFT_IEEE80211:
2132 #endif
2133 #ifdef IFT_CARP
2134         case IFT_CARP:
2135 #endif
2136         case IFT_GIF:           /* XXX need more cases? */
2137                 return (1);
2138         default:
2139                 return (0);
2140         }
2141 }
2142
2143 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2144 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2145 #ifdef SYSCTL_DECL
2146 SYSCTL_DECL(_net_inet6_icmp6);
2147 #endif
2148 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2149         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2150 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2151         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2152
2153 static int
2154 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2155 {
2156         int error;
2157         char buf[1024];
2158         struct in6_defrouter *d, *de;
2159         struct nd_defrouter *dr;
2160
2161         if (req->newptr)
2162                 return EPERM;
2163         error = 0;
2164
2165         for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2166              dr = TAILQ_NEXT(dr, dr_entry)) {
2167                 d = (struct in6_defrouter *)buf;
2168                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2169
2170                 if (d + 1 <= de) {
2171                         bzero(d, sizeof(*d));
2172                         d->rtaddr.sin6_family = AF_INET6;
2173                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2174                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2175                             dr->ifp) != 0)
2176                                 log(LOG_ERR,
2177                                     "scope error in "
2178                                     "default router list (%s)\n",
2179                                     ip6_sprintf(&dr->rtaddr));
2180                         d->flags = dr->flags;
2181                         d->rtlifetime = dr->rtlifetime;
2182                         d->expire = dr->expire;
2183                         d->if_index = dr->ifp->if_index;
2184                 } else
2185                         panic("buffer too short");
2186
2187                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2188                 if (error)
2189                         break;
2190         }
2191         return error;
2192 }
2193
2194 static int
2195 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2196 {
2197         int error;
2198         char buf[1024];
2199         struct in6_prefix *p, *pe;
2200         struct nd_prefix *pr;
2201
2202         if (req->newptr)
2203                 return EPERM;
2204         error = 0;
2205
2206         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2207                 u_short advrtrs;
2208                 size_t advance;
2209                 struct sockaddr_in6 *sin6, *s6;
2210                 struct nd_pfxrouter *pfr;
2211
2212                 p = (struct in6_prefix *)buf;
2213                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2214
2215                 if (p + 1 <= pe) {
2216                         bzero(p, sizeof(*p));
2217                         sin6 = (struct sockaddr_in6 *)(p + 1);
2218
2219                         p->prefix = pr->ndpr_prefix;
2220                         if (in6_recoverscope(&p->prefix,
2221                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2222                                 log(LOG_ERR,
2223                                     "scope error in prefix list (%s)\n",
2224                                     ip6_sprintf(&p->prefix.sin6_addr));
2225                         p->raflags = pr->ndpr_raf;
2226                         p->prefixlen = pr->ndpr_plen;
2227                         p->vltime = pr->ndpr_vltime;
2228                         p->pltime = pr->ndpr_pltime;
2229                         p->if_index = pr->ndpr_ifp->if_index;
2230                         p->expire = pr->ndpr_expire;
2231                         p->refcnt = pr->ndpr_refcnt;
2232                         p->flags = pr->ndpr_stateflags;
2233                         p->origin = PR_ORIG_RA;
2234                         advrtrs = 0;
2235                         for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2236                              pfr = pfr->pfr_next) {
2237                                 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2238                                         advrtrs++;
2239                                         continue;
2240                                 }
2241                                 s6 = &sin6[advrtrs];
2242                                 bzero(s6, sizeof(*s6));
2243                                 s6->sin6_family = AF_INET6;
2244                                 s6->sin6_len = sizeof(*sin6);
2245                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2246                                     pfr->router->ifp) != 0)
2247                                         log(LOG_ERR,
2248                                             "scope error in "
2249                                             "prefix list (%s)\n",
2250                                             ip6_sprintf(&pfr->router->rtaddr));
2251                                 advrtrs++;
2252                         }
2253                         p->advrtrs = advrtrs;
2254                 } else {
2255                         panic("buffer too short");
2256                 }
2257
2258                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2259                 error = SYSCTL_OUT(req, buf, advance);
2260                 if (error)
2261                         break;
2262         }
2263         return error;
2264 }