The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/netinet6/mld6.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * SPDX-License-Identifier: BSD-3-Clause
    3  *
    4  * Copyright (c) 2009 Bruce Simpson.
    5  *
    6  * Redistribution and use in source and binary forms, with or without
    7  * modification, are permitted provided that the following conditions
    8  * are met:
    9  * 1. Redistributions of source code must retain the above copyright
   10  *    notice, this list of conditions and the following disclaimer.
   11  * 2. Redistributions in binary form must reproduce the above copyright
   12  *    notice, this list of conditions and the following disclaimer in the
   13  *    documentation and/or other materials provided with the distribution.
   14  * 3. The name of the author may not be used to endorse or promote
   15  *    products derived from this software without specific prior written
   16  *    permission.
   17  *
   18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   28  * SUCH DAMAGE.
   29  *
   30  *      $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
   31  */
   32 
   33 /*-
   34  * Copyright (c) 1988 Stephen Deering.
   35  * Copyright (c) 1992, 1993
   36  *      The Regents of the University of California.  All rights reserved.
   37  *
   38  * This code is derived from software contributed to Berkeley by
   39  * Stephen Deering of Stanford University.
   40  *
   41  * Redistribution and use in source and binary forms, with or without
   42  * modification, are permitted provided that the following conditions
   43  * are met:
   44  * 1. Redistributions of source code must retain the above copyright
   45  *    notice, this list of conditions and the following disclaimer.
   46  * 2. Redistributions in binary form must reproduce the above copyright
   47  *    notice, this list of conditions and the following disclaimer in the
   48  *    documentation and/or other materials provided with the distribution.
   49  * 3. Neither the name of the University nor the names of its contributors
   50  *    may be used to endorse or promote products derived from this software
   51  *    without specific prior written permission.
   52  *
   53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   63  * SUCH DAMAGE.
   64  *
   65  *      @(#)igmp.c      8.1 (Berkeley) 7/19/93
   66  */
   67 
   68 #include <sys/cdefs.h>
   69 __FBSDID("$FreeBSD$");
   70 
   71 #include "opt_inet.h"
   72 #include "opt_inet6.h"
   73 
   74 #include <sys/param.h>
   75 #include <sys/systm.h>
   76 #include <sys/mbuf.h>
   77 #include <sys/socket.h>
   78 #include <sys/protosw.h>
   79 #include <sys/sysctl.h>
   80 #include <sys/kernel.h>
   81 #include <sys/callout.h>
   82 #include <sys/malloc.h>
   83 #include <sys/module.h>
   84 #include <sys/ktr.h>
   85 
   86 #include <net/if.h>
   87 #include <net/if_var.h>
   88 #include <net/route.h>
   89 #include <net/vnet.h>
   90 
   91 #include <netinet/in.h>
   92 #include <netinet/in_var.h>
   93 #include <netinet6/in6_var.h>
   94 #include <netinet/ip6.h>
   95 #include <netinet6/ip6_var.h>
   96 #include <netinet6/scope6_var.h>
   97 #include <netinet/icmp6.h>
   98 #include <netinet6/mld6.h>
   99 #include <netinet6/mld6_var.h>
  100 
  101 #include <security/mac/mac_framework.h>
  102 
  103 #ifndef KTR_MLD
  104 #define KTR_MLD KTR_INET6
  105 #endif
  106 
  107 static struct mld_ifsoftc *
  108                 mli_alloc_locked(struct ifnet *);
  109 static void     mli_delete_locked(const struct ifnet *);
  110 static void     mld_dispatch_packet(struct mbuf *);
  111 static void     mld_dispatch_queue(struct mbufq *, int);
  112 static void     mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
  113 static void     mld_fasttimo_vnet(struct in6_multi_head *inmh);
  114 static int      mld_handle_state_change(struct in6_multi *,
  115                     struct mld_ifsoftc *);
  116 static int      mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
  117                     const int);
  118 #ifdef KTR
  119 static char *   mld_rec_type_to_str(const int);
  120 #endif
  121 static void     mld_set_version(struct mld_ifsoftc *, const int);
  122 static void     mld_slowtimo_vnet(void);
  123 static int      mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
  124                     /*const*/ struct mld_hdr *);
  125 static int      mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
  126                     /*const*/ struct mld_hdr *);
  127 static void     mld_v1_process_group_timer(struct in6_multi_head *,
  128                     struct in6_multi *);
  129 static void     mld_v1_process_querier_timers(struct mld_ifsoftc *);
  130 static int      mld_v1_transmit_report(struct in6_multi *, const int);
  131 static void     mld_v1_update_group(struct in6_multi *, const int);
  132 static void     mld_v2_cancel_link_timers(struct mld_ifsoftc *);
  133 static void     mld_v2_dispatch_general_query(struct mld_ifsoftc *);
  134 static struct mbuf *
  135                 mld_v2_encap_report(struct ifnet *, struct mbuf *);
  136 static int      mld_v2_enqueue_filter_change(struct mbufq *,
  137                     struct in6_multi *);
  138 static int      mld_v2_enqueue_group_record(struct mbufq *,
  139                     struct in6_multi *, const int, const int, const int,
  140                     const int);
  141 static int      mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
  142                     struct mbuf *, struct mldv2_query *, const int, const int);
  143 static int      mld_v2_merge_state_changes(struct in6_multi *,
  144                     struct mbufq *);
  145 static void     mld_v2_process_group_timers(struct in6_multi_head *,
  146                     struct mbufq *, struct mbufq *,
  147                     struct in6_multi *, const int);
  148 static int      mld_v2_process_group_query(struct in6_multi *,
  149                     struct mld_ifsoftc *mli, int, struct mbuf *,
  150                     struct mldv2_query *, const int);
  151 static int      sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
  152 static int      sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
  153 
  154 /*
  155  * Normative references: RFC 2710, RFC 3590, RFC 3810.
  156  *
  157  * Locking:
  158  *  * The MLD subsystem lock ends up being system-wide for the moment,
  159  *    but could be per-VIMAGE later on.
  160  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
  161  *    Any may be taken independently; if any are held at the same
  162  *    time, the above lock order must be followed.
  163  *  * IN6_MULTI_LOCK covers in_multi.
  164  *  * MLD_LOCK covers per-link state and any global variables in this file.
  165  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
  166  *    per-link state iterators.
  167  *
  168  *  XXX LOR PREVENTION
  169  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
  170  *  will not accept an ifp; it wants an embedded scope ID, unlike
  171  *  ip_output(), which happily takes the ifp given to it. The embedded
  172  *  scope ID is only used by MLD to select the outgoing interface.
  173  *
  174  *  During interface attach and detach, MLD will take MLD_LOCK *after*
  175  *  the IF_AFDATA_LOCK.
  176  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
  177  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
  178  *  dispatch could work around this, but we'd rather not do that, as it
  179  *  can introduce other races.
  180  *
  181  *  As such, we exploit the fact that the scope ID is just the interface
  182  *  index, and embed it in the IPv6 destination address accordingly.
  183  *  This is potentially NOT VALID for MLDv1 reports, as they
  184  *  are always sent to the multicast group itself; as MLDv2
  185  *  reports are always sent to ff02::16, this is not an issue
  186  *  when MLDv2 is in use.
  187  *
  188  *  This does not however eliminate the LOR when ip6_output() itself
  189  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
  190  *  trigger a LOR warning in WITNESS when the ifnet is detached.
  191  *
  192  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
  193  *  how it's used across the network stack. Here we're simply exploiting
  194  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
  195  *
  196  * VIMAGE:
  197  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
  198  *    to a vnet in ifp->if_vnet.
  199  */
  200 static struct mtx                mld_mtx;
  201 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
  202 
  203 #define MLD_EMBEDSCOPE(pin6, zoneid)                                    \
  204         if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||                             \
  205             IN6_IS_ADDR_MC_INTFACELOCAL(pin6))                          \
  206                 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)         \
  207 
  208 /*
  209  * VIMAGE-wide globals.
  210  */
  211 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
  212 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
  213 VNET_DEFINE_STATIC(int, interface_timers_running6);
  214 VNET_DEFINE_STATIC(int, state_change_timers_running6);
  215 VNET_DEFINE_STATIC(int, current_state_timers_running6);
  216 
  217 #define V_mld_gsrdelay                  VNET(mld_gsrdelay)
  218 #define V_mli_head                      VNET(mli_head)
  219 #define V_interface_timers_running6     VNET(interface_timers_running6)
  220 #define V_state_change_timers_running6  VNET(state_change_timers_running6)
  221 #define V_current_state_timers_running6 VNET(current_state_timers_running6)
  222 
  223 SYSCTL_DECL(_net_inet6);        /* Note: Not in any common header. */
  224 
  225 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
  226     "IPv6 Multicast Listener Discovery");
  227 
  228 /*
  229  * Virtualized sysctls.
  230  */
  231 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
  232     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
  233     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
  234     "Rate limit for MLDv2 Group-and-Source queries in seconds");
  235 
  236 /*
  237  * Non-virtualized sysctls.
  238  */
  239 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
  240     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
  241     "Per-interface MLDv2 state");
  242 
  243 static int      mld_v1enable = 1;
  244 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
  245     &mld_v1enable, 0, "Enable fallback to MLDv1");
  246 
  247 static int      mld_v2enable = 1;
  248 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN,
  249     &mld_v2enable, 0, "Enable MLDv2");
  250 
  251 static int      mld_use_allow = 1;
  252 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
  253     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
  254 
  255 /*
  256  * Packed Router Alert option structure declaration.
  257  */
  258 struct mld_raopt {
  259         struct ip6_hbh          hbh;
  260         struct ip6_opt          pad;
  261         struct ip6_opt_router   ra;
  262 } __packed;
  263 
  264 /*
  265  * Router Alert hop-by-hop option header.
  266  */
  267 static struct mld_raopt mld_ra = {
  268         .hbh = { 0, 0 },
  269         .pad = { .ip6o_type = IP6OPT_PADN, 0 },
  270         .ra = {
  271             .ip6or_type = IP6OPT_ROUTER_ALERT,
  272             .ip6or_len = IP6OPT_RTALERT_LEN - 2,
  273             .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
  274             .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
  275         }
  276 };
  277 static struct ip6_pktopts mld_po;
  278 
  279 static __inline void
  280 mld_save_context(struct mbuf *m, struct ifnet *ifp)
  281 {
  282 
  283 #ifdef VIMAGE
  284         m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
  285 #endif /* VIMAGE */
  286         m->m_pkthdr.rcvif = ifp;
  287         m->m_pkthdr.flowid = ifp->if_index;
  288 }
  289 
  290 static __inline void
  291 mld_scrub_context(struct mbuf *m)
  292 {
  293 
  294         m->m_pkthdr.PH_loc.ptr = NULL;
  295         m->m_pkthdr.flowid = 0;
  296 }
  297 
  298 /*
  299  * Restore context from a queued output chain.
  300  * Return saved ifindex.
  301  *
  302  * VIMAGE: The assertion is there to make sure that we
  303  * actually called CURVNET_SET() with what's in the mbuf chain.
  304  */
  305 static __inline uint32_t
  306 mld_restore_context(struct mbuf *m)
  307 {
  308 
  309 #if defined(VIMAGE) && defined(INVARIANTS)
  310         KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
  311             ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
  312             __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
  313 #endif
  314         return (m->m_pkthdr.flowid);
  315 }
  316 
  317 /*
  318  * Retrieve or set threshold between group-source queries in seconds.
  319  *
  320  * VIMAGE: Assume curvnet set by caller.
  321  * SMPng: NOTE: Serialized by MLD lock.
  322  */
  323 static int
  324 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
  325 {
  326         int error;
  327         int i;
  328 
  329         error = sysctl_wire_old_buffer(req, sizeof(int));
  330         if (error)
  331                 return (error);
  332 
  333         MLD_LOCK();
  334 
  335         i = V_mld_gsrdelay.tv_sec;
  336 
  337         error = sysctl_handle_int(oidp, &i, 0, req);
  338         if (error || !req->newptr)
  339                 goto out_locked;
  340 
  341         if (i < -1 || i >= 60) {
  342                 error = EINVAL;
  343                 goto out_locked;
  344         }
  345 
  346         CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
  347              V_mld_gsrdelay.tv_sec, i);
  348         V_mld_gsrdelay.tv_sec = i;
  349 
  350 out_locked:
  351         MLD_UNLOCK();
  352         return (error);
  353 }
  354 
  355 /*
  356  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
  357  * For use by ifmcstat(8).
  358  *
  359  * SMPng: NOTE: Does an unlocked ifindex space read.
  360  * VIMAGE: Assume curvnet set by caller. The node handler itself
  361  * is not directly virtualized.
  362  */
  363 static int
  364 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
  365 {
  366         int                     *name;
  367         int                      error;
  368         u_int                    namelen;
  369         struct ifnet            *ifp;
  370         struct mld_ifsoftc      *mli;
  371 
  372         name = (int *)arg1;
  373         namelen = arg2;
  374 
  375         if (req->newptr != NULL)
  376                 return (EPERM);
  377 
  378         if (namelen != 1)
  379                 return (EINVAL);
  380 
  381         error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
  382         if (error)
  383                 return (error);
  384 
  385         IN6_MULTI_LOCK();
  386         IN6_MULTI_LIST_LOCK();
  387         MLD_LOCK();
  388 
  389         if (name[0] <= 0 || name[0] > V_if_index) {
  390                 error = ENOENT;
  391                 goto out_locked;
  392         }
  393 
  394         error = ENOENT;
  395 
  396         ifp = ifnet_byindex(name[0]);
  397         if (ifp == NULL)
  398                 goto out_locked;
  399 
  400         LIST_FOREACH(mli, &V_mli_head, mli_link) {
  401                 if (ifp == mli->mli_ifp) {
  402                         struct mld_ifinfo info;
  403 
  404                         info.mli_version = mli->mli_version;
  405                         info.mli_v1_timer = mli->mli_v1_timer;
  406                         info.mli_v2_timer = mli->mli_v2_timer;
  407                         info.mli_flags = mli->mli_flags;
  408                         info.mli_rv = mli->mli_rv;
  409                         info.mli_qi = mli->mli_qi;
  410                         info.mli_qri = mli->mli_qri;
  411                         info.mli_uri = mli->mli_uri;
  412                         error = SYSCTL_OUT(req, &info, sizeof(info));
  413                         break;
  414                 }
  415         }
  416 
  417 out_locked:
  418         MLD_UNLOCK();
  419         IN6_MULTI_LIST_UNLOCK();
  420         IN6_MULTI_UNLOCK();
  421         return (error);
  422 }
  423 
  424 /*
  425  * Dispatch an entire queue of pending packet chains.
  426  * VIMAGE: Assumes the vnet pointer has been set.
  427  */
  428 static void
  429 mld_dispatch_queue(struct mbufq *mq, int limit)
  430 {
  431         struct mbuf *m;
  432 
  433         while ((m = mbufq_dequeue(mq)) != NULL) {
  434                 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
  435                 mld_dispatch_packet(m);
  436                 if (--limit == 0)
  437                         break;
  438         }
  439 }
  440 
  441 /*
  442  * Filter outgoing MLD report state by group.
  443  *
  444  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
  445  * and node-local addresses. However, kernel and socket consumers
  446  * always embed the KAME scope ID in the address provided, so strip it
  447  * when performing comparison.
  448  * Note: This is not the same as the *multicast* scope.
  449  *
  450  * Return zero if the given group is one for which MLD reports
  451  * should be suppressed, or non-zero if reports should be issued.
  452  */
  453 static __inline int
  454 mld_is_addr_reported(const struct in6_addr *addr)
  455 {
  456 
  457         KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
  458 
  459         if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
  460                 return (0);
  461 
  462         if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
  463                 struct in6_addr tmp = *addr;
  464                 in6_clearscope(&tmp);
  465                 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
  466                         return (0);
  467         }
  468 
  469         return (1);
  470 }
  471 
  472 /*
  473  * Attach MLD when PF_INET6 is attached to an interface.
  474  *
  475  * SMPng: Normally called with IF_AFDATA_LOCK held.
  476  */
  477 struct mld_ifsoftc *
  478 mld_domifattach(struct ifnet *ifp)
  479 {
  480         struct mld_ifsoftc *mli;
  481 
  482         CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
  483             __func__, ifp, if_name(ifp));
  484 
  485         MLD_LOCK();
  486 
  487         mli = mli_alloc_locked(ifp);
  488         if (!(ifp->if_flags & IFF_MULTICAST))
  489                 mli->mli_flags |= MLIF_SILENT;
  490         if (mld_use_allow)
  491                 mli->mli_flags |= MLIF_USEALLOW;
  492 
  493         MLD_UNLOCK();
  494 
  495         return (mli);
  496 }
  497 
  498 /*
  499  * VIMAGE: assume curvnet set by caller.
  500  */
  501 static struct mld_ifsoftc *
  502 mli_alloc_locked(/*const*/ struct ifnet *ifp)
  503 {
  504         struct mld_ifsoftc *mli;
  505 
  506         MLD_LOCK_ASSERT();
  507 
  508         mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
  509         if (mli == NULL)
  510                 goto out;
  511 
  512         mli->mli_ifp = ifp;
  513         mli->mli_version = MLD_VERSION_2;
  514         mli->mli_flags = 0;
  515         mli->mli_rv = MLD_RV_INIT;
  516         mli->mli_qi = MLD_QI_INIT;
  517         mli->mli_qri = MLD_QRI_INIT;
  518         mli->mli_uri = MLD_URI_INIT;
  519         mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
  520 
  521         LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
  522 
  523         CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
  524              ifp, if_name(ifp));
  525 
  526 out:
  527         return (mli);
  528 }
  529 
  530 /*
  531  * Hook for ifdetach.
  532  *
  533  * NOTE: Some finalization tasks need to run before the protocol domain
  534  * is detached, but also before the link layer does its cleanup.
  535  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
  536  *
  537  * SMPng: Caller must hold IN6_MULTI_LOCK().
  538  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
  539  * XXX This routine is also bitten by unlocked ifma_protospec access.
  540  */
  541 void
  542 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
  543 {
  544         struct epoch_tracker     et;
  545         struct mld_ifsoftc      *mli;
  546         struct ifmultiaddr      *ifma;
  547         struct in6_multi        *inm;
  548 
  549         CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
  550             if_name(ifp));
  551 
  552         IN6_MULTI_LIST_LOCK_ASSERT();
  553         MLD_LOCK();
  554 
  555         mli = MLD_IFINFO(ifp);
  556         IF_ADDR_WLOCK(ifp);
  557         /*
  558          * Extract list of in6_multi associated with the detaching ifp
  559          * which the PF_INET6 layer is about to release.
  560          */
  561         NET_EPOCH_ENTER_ET(et);
  562         CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
  563                 inm = in6m_ifmultiaddr_get_inm(ifma);
  564                 if (inm == NULL)
  565                         continue;
  566                 in6m_disconnect_locked(inmh, inm);
  567 
  568                 if (mli->mli_version == MLD_VERSION_2) {
  569                         in6m_clear_recorded(inm);
  570 
  571                         /*
  572                          * We need to release the final reference held
  573                          * for issuing the INCLUDE {}.
  574                          */
  575                         if (inm->in6m_state == MLD_LEAVING_MEMBER) {
  576                                 inm->in6m_state = MLD_NOT_MEMBER;
  577                                 in6m_rele_locked(inmh, inm);
  578                         }
  579                 }
  580         }
  581         NET_EPOCH_EXIT_ET(et);
  582         IF_ADDR_WUNLOCK(ifp);
  583         MLD_UNLOCK();
  584 }
  585 
  586 /*
  587  * Hook for domifdetach.
  588  * Runs after link-layer cleanup; free MLD state.
  589  *
  590  * SMPng: Normally called with IF_AFDATA_LOCK held.
  591  */
  592 void
  593 mld_domifdetach(struct ifnet *ifp)
  594 {
  595 
  596         CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
  597             __func__, ifp, if_name(ifp));
  598 
  599         MLD_LOCK();
  600         mli_delete_locked(ifp);
  601         MLD_UNLOCK();
  602 }
  603 
  604 static void
  605 mli_delete_locked(const struct ifnet *ifp)
  606 {
  607         struct mld_ifsoftc *mli, *tmli;
  608 
  609         CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
  610             __func__, ifp, if_name(ifp));
  611 
  612         MLD_LOCK_ASSERT();
  613 
  614         LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
  615                 if (mli->mli_ifp == ifp) {
  616                         /*
  617                          * Free deferred General Query responses.
  618                          */
  619                         mbufq_drain(&mli->mli_gq);
  620 
  621                         LIST_REMOVE(mli, mli_link);
  622 
  623                         free(mli, M_MLD);
  624                         return;
  625                 }
  626         }
  627 }
  628 
  629 /*
  630  * Process a received MLDv1 general or address-specific query.
  631  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
  632  *
  633  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
  634  * mld_addr. This is OK as we own the mbuf chain.
  635  */
  636 static int
  637 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
  638     /*const*/ struct mld_hdr *mld)
  639 {
  640         struct ifmultiaddr      *ifma;
  641         struct mld_ifsoftc      *mli;
  642         struct in6_multi        *inm;
  643         int                      is_general_query;
  644         uint16_t                 timer;
  645 #ifdef KTR
  646         char                     ip6tbuf[INET6_ADDRSTRLEN];
  647 #endif
  648 
  649         is_general_query = 0;
  650 
  651         if (!mld_v1enable) {
  652                 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
  653                     ip6_sprintf(ip6tbuf, &mld->mld_addr),
  654                     ifp, if_name(ifp));
  655                 return (0);
  656         }
  657 
  658         /*
  659          * RFC3810 Section 6.2: MLD queries must originate from
  660          * a router's link-local address.
  661          */
  662         if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
  663                 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
  664                     ip6_sprintf(ip6tbuf, &ip6->ip6_src),
  665                     ifp, if_name(ifp));
  666                 return (0);
  667         }
  668 
  669         /*
  670          * Do address field validation upfront before we accept
  671          * the query.
  672          */
  673         if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
  674                 /*
  675                  * MLDv1 General Query.
  676                  * If this was not sent to the all-nodes group, ignore it.
  677                  */
  678                 struct in6_addr          dst;
  679 
  680                 dst = ip6->ip6_dst;
  681                 in6_clearscope(&dst);
  682                 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
  683                         return (EINVAL);
  684                 is_general_query = 1;
  685         } else {
  686                 /*
  687                  * Embed scope ID of receiving interface in MLD query for
  688                  * lookup whilst we don't hold other locks.
  689                  */
  690                 in6_setscope(&mld->mld_addr, ifp, NULL);
  691         }
  692 
  693         IN6_MULTI_LIST_LOCK();
  694         MLD_LOCK();
  695 
  696         /*
  697          * Switch to MLDv1 host compatibility mode.
  698          */
  699         mli = MLD_IFINFO(ifp);
  700         KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
  701         mld_set_version(mli, MLD_VERSION_1);
  702 
  703         timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
  704         if (timer == 0)
  705                 timer = 1;
  706 
  707         IF_ADDR_RLOCK(ifp);
  708         if (is_general_query) {
  709                 /*
  710                  * For each reporting group joined on this
  711                  * interface, kick the report timer.
  712                  */
  713                 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
  714                          ifp, if_name(ifp));
  715                 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
  716                         inm = in6m_ifmultiaddr_get_inm(ifma);
  717                         if (inm == NULL)
  718                                 continue;
  719                         mld_v1_update_group(inm, timer);
  720                 }
  721         } else {
  722                 /*
  723                  * MLDv1 Group-Specific Query.
  724                  * If this is a group-specific MLDv1 query, we need only
  725                  * look up the single group to process it.
  726                  */
  727                 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
  728                 if (inm != NULL) {
  729                         CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
  730                             ip6_sprintf(ip6tbuf, &mld->mld_addr),
  731                             ifp, if_name(ifp));
  732                         mld_v1_update_group(inm, timer);
  733                 }
  734                 /* XXX Clear embedded scope ID as userland won't expect it. */
  735                 in6_clearscope(&mld->mld_addr);
  736         }
  737 
  738         IF_ADDR_RUNLOCK(ifp);
  739         MLD_UNLOCK();
  740         IN6_MULTI_LIST_UNLOCK();
  741 
  742         return (0);
  743 }
  744 
  745 /*
  746  * Update the report timer on a group in response to an MLDv1 query.
  747  *
  748  * If we are becoming the reporting member for this group, start the timer.
  749  * If we already are the reporting member for this group, and timer is
  750  * below the threshold, reset it.
  751  *
  752  * We may be updating the group for the first time since we switched
  753  * to MLDv2. If we are, then we must clear any recorded source lists,
  754  * and transition to REPORTING state; the group timer is overloaded
  755  * for group and group-source query responses. 
  756  *
  757  * Unlike MLDv2, the delay per group should be jittered
  758  * to avoid bursts of MLDv1 reports.
  759  */
  760 static void
  761 mld_v1_update_group(struct in6_multi *inm, const int timer)
  762 {
  763 #ifdef KTR
  764         char                     ip6tbuf[INET6_ADDRSTRLEN];
  765 #endif
  766 
  767         CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
  768             ip6_sprintf(ip6tbuf, &inm->in6m_addr),
  769             if_name(inm->in6m_ifp), timer);
  770 
  771         IN6_MULTI_LIST_LOCK_ASSERT();
  772 
  773         switch (inm->in6m_state) {
  774         case MLD_NOT_MEMBER:
  775         case MLD_SILENT_MEMBER:
  776                 break;
  777         case MLD_REPORTING_MEMBER:
  778                 if (inm->in6m_timer != 0 &&
  779                     inm->in6m_timer <= timer) {
  780                         CTR1(KTR_MLD, "%s: REPORTING and timer running, "
  781                             "skipping.", __func__);
  782                         break;
  783                 }
  784                 /* FALLTHROUGH */
  785         case MLD_SG_QUERY_PENDING_MEMBER:
  786         case MLD_G_QUERY_PENDING_MEMBER:
  787         case MLD_IDLE_MEMBER:
  788         case MLD_LAZY_MEMBER:
  789         case MLD_AWAKENING_MEMBER:
  790                 CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
  791                 inm->in6m_state = MLD_REPORTING_MEMBER;
  792                 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
  793                 V_current_state_timers_running6 = 1;
  794                 break;
  795         case MLD_SLEEPING_MEMBER:
  796                 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
  797                 inm->in6m_state = MLD_AWAKENING_MEMBER;
  798                 break;
  799         case MLD_LEAVING_MEMBER:
  800                 break;
  801         }
  802 }
  803 
  804 /*
  805  * Process a received MLDv2 general, group-specific or
  806  * group-and-source-specific query.
  807  *
  808  * Assumes that mld points to a struct mldv2_query which is stored in
  809  * contiguous memory.
  810  *
  811  * Return 0 if successful, otherwise an appropriate error code is returned.
  812  */
  813 static int
  814 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
  815     struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
  816 {
  817         struct mld_ifsoftc      *mli;
  818         struct in6_multi        *inm;
  819         uint32_t                 maxdelay, nsrc, qqi;
  820         int                      is_general_query;
  821         uint16_t                 timer;
  822         uint8_t                  qrv;
  823 #ifdef KTR
  824         char                     ip6tbuf[INET6_ADDRSTRLEN];
  825 #endif
  826 
  827         if (!mld_v2enable) {
  828                 CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
  829                     ip6_sprintf(ip6tbuf, &ip6->ip6_src),
  830                     ifp, if_name(ifp));
  831                 return (0);
  832         }
  833 
  834         /*
  835          * RFC3810 Section 6.2: MLD queries must originate from
  836          * a router's link-local address.
  837          */
  838         if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
  839                 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
  840                     ip6_sprintf(ip6tbuf, &ip6->ip6_src),
  841                     ifp, if_name(ifp));
  842                 return (0);
  843         }
  844 
  845         is_general_query = 0;
  846 
  847         CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
  848 
  849         maxdelay = ntohs(mld->mld_maxdelay);    /* in 1/10ths of a second */
  850         if (maxdelay >= 32768) {
  851                 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
  852                            (MLD_MRC_EXP(maxdelay) + 3);
  853         }
  854         timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
  855         if (timer == 0)
  856                 timer = 1;
  857 
  858         qrv = MLD_QRV(mld->mld_misc);
  859         if (qrv < 2) {
  860                 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
  861                     qrv, MLD_RV_INIT);
  862                 qrv = MLD_RV_INIT;
  863         }
  864 
  865         qqi = mld->mld_qqi;
  866         if (qqi >= 128) {
  867                 qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
  868                      (MLD_QQIC_EXP(mld->mld_qqi) + 3);
  869         }
  870 
  871         nsrc = ntohs(mld->mld_numsrc);
  872         if (nsrc > MLD_MAX_GS_SOURCES)
  873                 return (EMSGSIZE);
  874         if (icmp6len < sizeof(struct mldv2_query) +
  875             (nsrc * sizeof(struct in6_addr)))
  876                 return (EMSGSIZE);
  877 
  878         /*
  879          * Do further input validation upfront to avoid resetting timers
  880          * should we need to discard this query.
  881          */
  882         if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
  883                 /*
  884                  * A general query with a source list has undefined
  885                  * behaviour; discard it.
  886                  */
  887                 if (nsrc > 0)
  888                         return (EINVAL);
  889                 is_general_query = 1;
  890         } else {
  891                 /*
  892                  * Embed scope ID of receiving interface in MLD query for
  893                  * lookup whilst we don't hold other locks (due to KAME
  894                  * locking lameness). We own this mbuf chain just now.
  895                  */
  896                 in6_setscope(&mld->mld_addr, ifp, NULL);
  897         }
  898 
  899         IN6_MULTI_LIST_LOCK();
  900         MLD_LOCK();
  901 
  902         mli = MLD_IFINFO(ifp);
  903         KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
  904 
  905         /*
  906          * Discard the v2 query if we're in Compatibility Mode.
  907          * The RFC is pretty clear that hosts need to stay in MLDv1 mode
  908          * until the Old Version Querier Present timer expires.
  909          */
  910         if (mli->mli_version != MLD_VERSION_2)
  911                 goto out_locked;
  912 
  913         mld_set_version(mli, MLD_VERSION_2);
  914         mli->mli_rv = qrv;
  915         mli->mli_qi = qqi;
  916         mli->mli_qri = maxdelay;
  917 
  918         CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
  919             maxdelay);
  920 
  921         if (is_general_query) {
  922                 /*
  923                  * MLDv2 General Query.
  924                  *
  925                  * Schedule a current-state report on this ifp for
  926                  * all groups, possibly containing source lists.
  927                  *
  928                  * If there is a pending General Query response
  929                  * scheduled earlier than the selected delay, do
  930                  * not schedule any other reports.
  931                  * Otherwise, reset the interface timer.
  932                  */
  933                 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
  934                     ifp, if_name(ifp));
  935                 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
  936                         mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
  937                         V_interface_timers_running6 = 1;
  938                 }
  939         } else {
  940                 /*
  941                  * MLDv2 Group-specific or Group-and-source-specific Query.
  942                  *
  943                  * Group-source-specific queries are throttled on
  944                  * a per-group basis to defeat denial-of-service attempts.
  945                  * Queries for groups we are not a member of on this
  946                  * link are simply ignored.
  947                  */
  948                 IF_ADDR_RLOCK(ifp);
  949                 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
  950                 if (inm == NULL) {
  951                         IF_ADDR_RUNLOCK(ifp);
  952                         goto out_locked;
  953                 }
  954                 if (nsrc > 0) {
  955                         if (!ratecheck(&inm->in6m_lastgsrtv,
  956                             &V_mld_gsrdelay)) {
  957                                 CTR1(KTR_MLD, "%s: GS query throttled.",
  958                                     __func__);
  959                                 IF_ADDR_RUNLOCK(ifp);
  960                                 goto out_locked;
  961                         }
  962                 }
  963                 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
  964                      ifp, if_name(ifp));
  965                 /*
  966                  * If there is a pending General Query response
  967                  * scheduled sooner than the selected delay, no
  968                  * further report need be scheduled.
  969                  * Otherwise, prepare to respond to the
  970                  * group-specific or group-and-source query.
  971                  */
  972                 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
  973                         mld_v2_process_group_query(inm, mli, timer, m, mld, off);
  974 
  975                 /* XXX Clear embedded scope ID as userland won't expect it. */
  976                 in6_clearscope(&mld->mld_addr);
  977                 IF_ADDR_RUNLOCK(ifp);
  978         }
  979 
  980 out_locked:
  981         MLD_UNLOCK();
  982         IN6_MULTI_LIST_UNLOCK();
  983 
  984         return (0);
  985 }
  986 
  987 /*
  988  * Process a received MLDv2 group-specific or group-and-source-specific
  989  * query.
  990  * Return <0 if any error occurred. Currently this is ignored.
  991  */
  992 static int
  993 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
  994     int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
  995 {
  996         int                      retval;
  997         uint16_t                 nsrc;
  998 
  999         IN6_MULTI_LIST_LOCK_ASSERT();
 1000         MLD_LOCK_ASSERT();
 1001 
 1002         retval = 0;
 1003 
 1004         switch (inm->in6m_state) {
 1005         case MLD_NOT_MEMBER:
 1006         case MLD_SILENT_MEMBER:
 1007         case MLD_SLEEPING_MEMBER:
 1008         case MLD_LAZY_MEMBER:
 1009         case MLD_AWAKENING_MEMBER:
 1010         case MLD_IDLE_MEMBER:
 1011         case MLD_LEAVING_MEMBER:
 1012                 return (retval);
 1013                 break;
 1014         case MLD_REPORTING_MEMBER:
 1015         case MLD_G_QUERY_PENDING_MEMBER:
 1016         case MLD_SG_QUERY_PENDING_MEMBER:
 1017                 break;
 1018         }
 1019 
 1020         nsrc = ntohs(mld->mld_numsrc);
 1021 
 1022         /* Length should be checked by calling function. */
 1023         KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
 1024             m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
 1025             nsrc * sizeof(struct in6_addr),
 1026             ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
 1027             m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
 1028             nsrc * sizeof(struct in6_addr), m0));
 1029 
 1030 
 1031         /*
 1032          * Deal with group-specific queries upfront.
 1033          * If any group query is already pending, purge any recorded
 1034          * source-list state if it exists, and schedule a query response
 1035          * for this group-specific query.
 1036          */
 1037         if (nsrc == 0) {
 1038                 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
 1039                     inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
 1040                         in6m_clear_recorded(inm);
 1041                         timer = min(inm->in6m_timer, timer);
 1042                 }
 1043                 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
 1044                 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
 1045                 V_current_state_timers_running6 = 1;
 1046                 return (retval);
 1047         }
 1048 
 1049         /*
 1050          * Deal with the case where a group-and-source-specific query has
 1051          * been received but a group-specific query is already pending.
 1052          */
 1053         if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
 1054                 timer = min(inm->in6m_timer, timer);
 1055                 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
 1056                 V_current_state_timers_running6 = 1;
 1057                 return (retval);
 1058         }
 1059 
 1060         /*
 1061          * Finally, deal with the case where a group-and-source-specific
 1062          * query has been received, where a response to a previous g-s-r
 1063          * query exists, or none exists.
 1064          * In this case, we need to parse the source-list which the Querier
 1065          * has provided us with and check if we have any source list filter
 1066          * entries at T1 for these sources. If we do not, there is no need
 1067          * schedule a report and the query may be dropped.
 1068          * If we do, we must record them and schedule a current-state
 1069          * report for those sources.
 1070          */
 1071         if (inm->in6m_nsrc > 0) {
 1072                 struct in6_addr          srcaddr;
 1073                 int                      i, nrecorded;
 1074                 int                      soff;
 1075 
 1076                 soff = off + sizeof(struct mldv2_query);
 1077                 nrecorded = 0;
 1078                 for (i = 0; i < nsrc; i++) {
 1079                         m_copydata(m0, soff, sizeof(struct in6_addr),
 1080                             (caddr_t)&srcaddr);
 1081                         retval = in6m_record_source(inm, &srcaddr);
 1082                         if (retval < 0)
 1083                                 break;
 1084                         nrecorded += retval;
 1085                         soff += sizeof(struct in6_addr);
 1086                 }
 1087                 if (nrecorded > 0) {
 1088                         CTR1(KTR_MLD,
 1089                             "%s: schedule response to SG query", __func__);
 1090                         inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
 1091                         inm->in6m_timer = MLD_RANDOM_DELAY(timer);
 1092                         V_current_state_timers_running6 = 1;
 1093                 }
 1094         }
 1095 
 1096         return (retval);
 1097 }
 1098 
 1099 /*
 1100  * Process a received MLDv1 host membership report.
 1101  * Assumes mld points to mld_hdr in pulled up mbuf chain.
 1102  *
 1103  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
 1104  * mld_addr. This is OK as we own the mbuf chain.
 1105  */
 1106 static int
 1107 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
 1108     /*const*/ struct mld_hdr *mld)
 1109 {
 1110         struct in6_addr          src, dst;
 1111         struct in6_ifaddr       *ia;
 1112         struct in6_multi        *inm;
 1113 #ifdef KTR
 1114         char                     ip6tbuf[INET6_ADDRSTRLEN];
 1115 #endif
 1116 
 1117         if (!mld_v1enable) {
 1118                 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
 1119                     ip6_sprintf(ip6tbuf, &mld->mld_addr),
 1120                     ifp, if_name(ifp));
 1121                 return (0);
 1122         }
 1123 
 1124         if (ifp->if_flags & IFF_LOOPBACK)
 1125                 return (0);
 1126 
 1127         /*
 1128          * MLDv1 reports must originate from a host's link-local address,
 1129          * or the unspecified address (when booting).
 1130          */
 1131         src = ip6->ip6_src;
 1132         in6_clearscope(&src);
 1133         if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
 1134                 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
 1135                     ip6_sprintf(ip6tbuf, &ip6->ip6_src),
 1136                     ifp, if_name(ifp));
 1137                 return (EINVAL);
 1138         }
 1139 
 1140         /*
 1141          * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
 1142          * group, and must be directed to the group itself.
 1143          */
 1144         dst = ip6->ip6_dst;
 1145         in6_clearscope(&dst);
 1146         if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
 1147             !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
 1148                 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
 1149                     ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
 1150                     ifp, if_name(ifp));
 1151                 return (EINVAL);
 1152         }
 1153 
 1154         /*
 1155          * Make sure we don't hear our own membership report, as fast
 1156          * leave requires knowing that we are the only member of a
 1157          * group. Assume we used the link-local address if available,
 1158          * otherwise look for ::.
 1159          *
 1160          * XXX Note that scope ID comparison is needed for the address
 1161          * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
 1162          * performed for the on-wire address.
 1163          */
 1164         ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
 1165         if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
 1166             (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
 1167                 if (ia != NULL)
 1168                         ifa_free(&ia->ia_ifa);
 1169                 return (0);
 1170         }
 1171         if (ia != NULL)
 1172                 ifa_free(&ia->ia_ifa);
 1173 
 1174         CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
 1175             ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
 1176 
 1177         /*
 1178          * Embed scope ID of receiving interface in MLD query for lookup
 1179          * whilst we don't hold other locks (due to KAME locking lameness).
 1180          */
 1181         if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
 1182                 in6_setscope(&mld->mld_addr, ifp, NULL);
 1183 
 1184         IN6_MULTI_LIST_LOCK();
 1185         MLD_LOCK();
 1186         IF_ADDR_RLOCK(ifp);
 1187 
 1188         /*
 1189          * MLDv1 report suppression.
 1190          * If we are a member of this group, and our membership should be
 1191          * reported, and our group timer is pending or about to be reset,
 1192          * stop our group timer by transitioning to the 'lazy' state.
 1193          */
 1194         inm = in6m_lookup_locked(ifp, &mld->mld_addr);
 1195         if (inm != NULL) {
 1196                 struct mld_ifsoftc *mli;
 1197 
 1198                 mli = inm->in6m_mli;
 1199                 KASSERT(mli != NULL,
 1200                     ("%s: no mli for ifp %p", __func__, ifp));
 1201 
 1202                 /*
 1203                  * If we are in MLDv2 host mode, do not allow the
 1204                  * other host's MLDv1 report to suppress our reports.
 1205                  */
 1206                 if (mli->mli_version == MLD_VERSION_2)
 1207                         goto out_locked;
 1208 
 1209                 inm->in6m_timer = 0;
 1210 
 1211                 switch (inm->in6m_state) {
 1212                 case MLD_NOT_MEMBER:
 1213                 case MLD_SILENT_MEMBER:
 1214                 case MLD_SLEEPING_MEMBER:
 1215                         break;
 1216                 case MLD_REPORTING_MEMBER:
 1217                 case MLD_IDLE_MEMBER:
 1218                 case MLD_AWAKENING_MEMBER:
 1219                         CTR3(KTR_MLD,
 1220                             "report suppressed for %s on ifp %p(%s)",
 1221                             ip6_sprintf(ip6tbuf, &mld->mld_addr),
 1222                             ifp, if_name(ifp));
 1223                 case MLD_LAZY_MEMBER:
 1224                         inm->in6m_state = MLD_LAZY_MEMBER;
 1225                         break;
 1226                 case MLD_G_QUERY_PENDING_MEMBER:
 1227                 case MLD_SG_QUERY_PENDING_MEMBER:
 1228                 case MLD_LEAVING_MEMBER:
 1229                         break;
 1230                 }
 1231         }
 1232 
 1233 out_locked:
 1234         IF_ADDR_RUNLOCK(ifp);
 1235         MLD_UNLOCK();
 1236         IN6_MULTI_LIST_UNLOCK();
 1237 
 1238         /* XXX Clear embedded scope ID as userland won't expect it. */
 1239         in6_clearscope(&mld->mld_addr);
 1240 
 1241         return (0);
 1242 }
 1243 
 1244 /*
 1245  * MLD input path.
 1246  *
 1247  * Assume query messages which fit in a single ICMPv6 message header
 1248  * have been pulled up.
 1249  * Assume that userland will want to see the message, even if it
 1250  * otherwise fails kernel input validation; do not free it.
 1251  * Pullup may however free the mbuf chain m if it fails.
 1252  *
 1253  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
 1254  */
 1255 int
 1256 mld_input(struct mbuf **mp, int off, int icmp6len)
 1257 {
 1258         struct ifnet    *ifp;
 1259         struct ip6_hdr  *ip6;
 1260         struct mbuf     *m;
 1261         struct mld_hdr  *mld;
 1262         int              mldlen;
 1263 
 1264         m = *mp;
 1265         CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
 1266 
 1267         ifp = m->m_pkthdr.rcvif;
 1268 
 1269         /* Pullup to appropriate size. */
 1270         if (m->m_len < off + sizeof(*mld)) {
 1271                 m = m_pullup(m, off + sizeof(*mld));
 1272                 if (m == NULL) {
 1273                         ICMP6STAT_INC(icp6s_badlen);
 1274                         return (IPPROTO_DONE);
 1275                 }
 1276         }
 1277         mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
 1278         if (mld->mld_type == MLD_LISTENER_QUERY &&
 1279             icmp6len >= sizeof(struct mldv2_query)) {
 1280                 mldlen = sizeof(struct mldv2_query);
 1281         } else {
 1282                 mldlen = sizeof(struct mld_hdr);
 1283         }
 1284         if (m->m_len < off + mldlen) {
 1285                 m = m_pullup(m, off + mldlen);
 1286                 if (m == NULL) {
 1287                         ICMP6STAT_INC(icp6s_badlen);
 1288                         return (IPPROTO_DONE);
 1289                 }
 1290         }
 1291         *mp = m;
 1292         ip6 = mtod(m, struct ip6_hdr *);
 1293         mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
 1294 
 1295         /*
 1296          * Userland needs to see all of this traffic for implementing
 1297          * the endpoint discovery portion of multicast routing.
 1298          */
 1299         switch (mld->mld_type) {
 1300         case MLD_LISTENER_QUERY:
 1301                 icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
 1302                 if (icmp6len == sizeof(struct mld_hdr)) {
 1303                         if (mld_v1_input_query(ifp, ip6, mld) != 0)
 1304                                 return (0);
 1305                 } else if (icmp6len >= sizeof(struct mldv2_query)) {
 1306                         if (mld_v2_input_query(ifp, ip6, m,
 1307                             (struct mldv2_query *)mld, off, icmp6len) != 0)
 1308                                 return (0);
 1309                 }
 1310                 break;
 1311         case MLD_LISTENER_REPORT:
 1312                 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
 1313                 if (mld_v1_input_report(ifp, ip6, mld) != 0)
 1314                         return (0);
 1315                 break;
 1316         case MLDV2_LISTENER_REPORT:
 1317                 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
 1318                 break;
 1319         case MLD_LISTENER_DONE:
 1320                 icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
 1321                 break;
 1322         default:
 1323                 break;
 1324         }
 1325 
 1326         return (0);
 1327 }
 1328 
 1329 /*
 1330  * Fast timeout handler (global).
 1331  * VIMAGE: Timeout handlers are expected to service all vimages.
 1332  */
 1333 void
 1334 mld_fasttimo(void)
 1335 {
 1336         struct in6_multi_head inmh;
 1337         VNET_ITERATOR_DECL(vnet_iter);
 1338 
 1339         SLIST_INIT(&inmh);
 1340         
 1341         VNET_LIST_RLOCK_NOSLEEP();
 1342         VNET_FOREACH(vnet_iter) {
 1343                 CURVNET_SET(vnet_iter);
 1344                 mld_fasttimo_vnet(&inmh);
 1345                 CURVNET_RESTORE();
 1346         }
 1347         VNET_LIST_RUNLOCK_NOSLEEP();
 1348         in6m_release_list_deferred(&inmh);
 1349 }
 1350 
 1351 /*
 1352  * Fast timeout handler (per-vnet).
 1353  *
 1354  * VIMAGE: Assume caller has set up our curvnet.
 1355  */
 1356 static void
 1357 mld_fasttimo_vnet(struct in6_multi_head *inmh)
 1358 {
 1359         struct epoch_tracker     et;
 1360         struct mbufq             scq;   /* State-change packets */
 1361         struct mbufq             qrq;   /* Query response packets */
 1362         struct ifnet            *ifp;
 1363         struct mld_ifsoftc      *mli;
 1364         struct ifmultiaddr      *ifma;
 1365         struct in6_multi        *inm;
 1366         int                      uri_fasthz;
 1367 
 1368         uri_fasthz = 0;
 1369 
 1370         /*
 1371          * Quick check to see if any work needs to be done, in order to
 1372          * minimize the overhead of fasttimo processing.
 1373          * SMPng: XXX Unlocked reads.
 1374          */
 1375         if (!V_current_state_timers_running6 &&
 1376             !V_interface_timers_running6 &&
 1377             !V_state_change_timers_running6)
 1378                 return;
 1379 
 1380         IN6_MULTI_LIST_LOCK();
 1381         MLD_LOCK();
 1382 
 1383         /*
 1384          * MLDv2 General Query response timer processing.
 1385          */
 1386         if (V_interface_timers_running6) {
 1387                 CTR1(KTR_MLD, "%s: interface timers running", __func__);
 1388 
 1389                 V_interface_timers_running6 = 0;
 1390                 LIST_FOREACH(mli, &V_mli_head, mli_link) {
 1391                         if (mli->mli_v2_timer == 0) {
 1392                                 /* Do nothing. */
 1393                         } else if (--mli->mli_v2_timer == 0) {
 1394                                 mld_v2_dispatch_general_query(mli);
 1395                         } else {
 1396                                 V_interface_timers_running6 = 1;
 1397                         }
 1398                 }
 1399         }
 1400 
 1401         if (!V_current_state_timers_running6 &&
 1402             !V_state_change_timers_running6)
 1403                 goto out_locked;
 1404 
 1405         V_current_state_timers_running6 = 0;
 1406         V_state_change_timers_running6 = 0;
 1407 
 1408         CTR1(KTR_MLD, "%s: state change timers running", __func__);
 1409 
 1410         /*
 1411          * MLD host report and state-change timer processing.
 1412          * Note: Processing a v2 group timer may remove a node.
 1413          */
 1414         LIST_FOREACH(mli, &V_mli_head, mli_link) {
 1415                 ifp = mli->mli_ifp;
 1416 
 1417                 if (mli->mli_version == MLD_VERSION_2) {
 1418                         uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
 1419                             PR_FASTHZ);
 1420                         mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
 1421                         mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
 1422                 }
 1423 
 1424                 IF_ADDR_WLOCK(ifp);
 1425                 NET_EPOCH_ENTER_ET(et);
 1426                 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
 1427                         inm = in6m_ifmultiaddr_get_inm(ifma);
 1428                         if (inm == NULL)
 1429                                 continue;
 1430                         switch (mli->mli_version) {
 1431                         case MLD_VERSION_1:
 1432                                 mld_v1_process_group_timer(inmh, inm);
 1433                                 break;
 1434                         case MLD_VERSION_2:
 1435                                 mld_v2_process_group_timers(inmh, &qrq,
 1436                                     &scq, inm, uri_fasthz);
 1437                                 break;
 1438                         }
 1439                 }
 1440                 IF_ADDR_WUNLOCK(ifp);
 1441 
 1442                 switch (mli->mli_version) {
 1443                 case MLD_VERSION_1:
 1444                         /*
 1445                          * Transmit reports for this lifecycle.  This
 1446                          * is done while not holding IF_ADDR_LOCK
 1447                          * since this can call
 1448                          * in6ifa_ifpforlinklocal() which locks
 1449                          * IF_ADDR_LOCK internally as well as
 1450                          * ip6_output() to transmit a packet.
 1451                          */
 1452                         while ((inm = SLIST_FIRST(inmh)) != NULL) {
 1453                                 SLIST_REMOVE_HEAD(inmh, in6m_defer);
 1454                                 (void)mld_v1_transmit_report(inm,
 1455                                     MLD_LISTENER_REPORT);
 1456                         }
 1457                         break;
 1458                 case MLD_VERSION_2:
 1459                         mld_dispatch_queue(&qrq, 0);
 1460                         mld_dispatch_queue(&scq, 0);
 1461                         break;
 1462                 }
 1463                 NET_EPOCH_EXIT_ET(et);
 1464         }
 1465 
 1466 out_locked:
 1467         MLD_UNLOCK();
 1468         IN6_MULTI_LIST_UNLOCK();
 1469 }
 1470 
 1471 /*
 1472  * Update host report group timer.
 1473  * Will update the global pending timer flags.
 1474  */
 1475 static void
 1476 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
 1477 {
 1478         int report_timer_expired;
 1479 
 1480         IN6_MULTI_LIST_LOCK_ASSERT();
 1481         MLD_LOCK_ASSERT();
 1482 
 1483         if (inm->in6m_timer == 0) {
 1484                 report_timer_expired = 0;
 1485         } else if (--inm->in6m_timer == 0) {
 1486                 report_timer_expired = 1;
 1487         } else {
 1488                 V_current_state_timers_running6 = 1;
 1489                 return;
 1490         }
 1491 
 1492         switch (inm->in6m_state) {
 1493         case MLD_NOT_MEMBER:
 1494         case MLD_SILENT_MEMBER:
 1495         case MLD_IDLE_MEMBER:
 1496         case MLD_LAZY_MEMBER:
 1497         case MLD_SLEEPING_MEMBER:
 1498         case MLD_AWAKENING_MEMBER:
 1499                 break;
 1500         case MLD_REPORTING_MEMBER:
 1501                 if (report_timer_expired) {
 1502                         inm->in6m_state = MLD_IDLE_MEMBER;
 1503                         SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
 1504                 }
 1505                 break;
 1506         case MLD_G_QUERY_PENDING_MEMBER:
 1507         case MLD_SG_QUERY_PENDING_MEMBER:
 1508         case MLD_LEAVING_MEMBER:
 1509                 break;
 1510         }
 1511 }
 1512 
 1513 /*
 1514  * Update a group's timers for MLDv2.
 1515  * Will update the global pending timer flags.
 1516  * Note: Unlocked read from mli.
 1517  */
 1518 static void
 1519 mld_v2_process_group_timers(struct in6_multi_head *inmh,
 1520     struct mbufq *qrq, struct mbufq *scq,
 1521     struct in6_multi *inm, const int uri_fasthz)
 1522 {
 1523         int query_response_timer_expired;
 1524         int state_change_retransmit_timer_expired;
 1525 #ifdef KTR
 1526         char ip6tbuf[INET6_ADDRSTRLEN];
 1527 #endif
 1528 
 1529         IN6_MULTI_LIST_LOCK_ASSERT();
 1530         MLD_LOCK_ASSERT();
 1531 
 1532         query_response_timer_expired = 0;
 1533         state_change_retransmit_timer_expired = 0;
 1534 
 1535         /*
 1536          * During a transition from compatibility mode back to MLDv2,
 1537          * a group record in REPORTING state may still have its group
 1538          * timer active. This is a no-op in this function; it is easier
 1539          * to deal with it here than to complicate the slow-timeout path.
 1540          */
 1541         if (inm->in6m_timer == 0) {
 1542                 query_response_timer_expired = 0;
 1543         } else if (--inm->in6m_timer == 0) {
 1544                 query_response_timer_expired = 1;
 1545         } else {
 1546                 V_current_state_timers_running6 = 1;
 1547         }
 1548 
 1549         if (inm->in6m_sctimer == 0) {
 1550                 state_change_retransmit_timer_expired = 0;
 1551         } else if (--inm->in6m_sctimer == 0) {
 1552                 state_change_retransmit_timer_expired = 1;
 1553         } else {
 1554                 V_state_change_timers_running6 = 1;
 1555         }
 1556 
 1557         /* We are in fasttimo, so be quick about it. */
 1558         if (!state_change_retransmit_timer_expired &&
 1559             !query_response_timer_expired)
 1560                 return;
 1561 
 1562         switch (inm->in6m_state) {
 1563         case MLD_NOT_MEMBER:
 1564         case MLD_SILENT_MEMBER:
 1565         case MLD_SLEEPING_MEMBER:
 1566         case MLD_LAZY_MEMBER:
 1567         case MLD_AWAKENING_MEMBER:
 1568         case MLD_IDLE_MEMBER:
 1569                 break;
 1570         case MLD_G_QUERY_PENDING_MEMBER:
 1571         case MLD_SG_QUERY_PENDING_MEMBER:
 1572                 /*
 1573                  * Respond to a previously pending Group-Specific
 1574                  * or Group-and-Source-Specific query by enqueueing
 1575                  * the appropriate Current-State report for
 1576                  * immediate transmission.
 1577                  */
 1578                 if (query_response_timer_expired) {
 1579                         int retval;
 1580 
 1581                         retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
 1582                             (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
 1583                             0);
 1584                         CTR2(KTR_MLD, "%s: enqueue record = %d",
 1585                             __func__, retval);
 1586                         inm->in6m_state = MLD_REPORTING_MEMBER;
 1587                         in6m_clear_recorded(inm);
 1588                 }
 1589                 /* FALLTHROUGH */
 1590         case MLD_REPORTING_MEMBER:
 1591         case MLD_LEAVING_MEMBER:
 1592                 if (state_change_retransmit_timer_expired) {
 1593                         /*
 1594                          * State-change retransmission timer fired.
 1595                          * If there are any further pending retransmissions,
 1596                          * set the global pending state-change flag, and
 1597                          * reset the timer.
 1598                          */
 1599                         if (--inm->in6m_scrv > 0) {
 1600                                 inm->in6m_sctimer = uri_fasthz;
 1601                                 V_state_change_timers_running6 = 1;
 1602                         }
 1603                         /*
 1604                          * Retransmit the previously computed state-change
 1605                          * report. If there are no further pending
 1606                          * retransmissions, the mbuf queue will be consumed.
 1607                          * Update T0 state to T1 as we have now sent
 1608                          * a state-change.
 1609                          */
 1610                         (void)mld_v2_merge_state_changes(inm, scq);
 1611 
 1612                         in6m_commit(inm);
 1613                         CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
 1614                             ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 1615                             if_name(inm->in6m_ifp));
 1616 
 1617                         /*
 1618                          * If we are leaving the group for good, make sure
 1619                          * we release MLD's reference to it.
 1620                          * This release must be deferred using a SLIST,
 1621                          * as we are called from a loop which traverses
 1622                          * the in_ifmultiaddr TAILQ.
 1623                          */
 1624                         if (inm->in6m_state == MLD_LEAVING_MEMBER &&
 1625                             inm->in6m_scrv == 0) {
 1626                                 inm->in6m_state = MLD_NOT_MEMBER;
 1627                                 in6m_disconnect_locked(inmh, inm);
 1628                                 in6m_rele_locked(inmh, inm);
 1629                         }
 1630                 }
 1631                 break;
 1632         }
 1633 }
 1634 
 1635 /*
 1636  * Switch to a different version on the given interface,
 1637  * as per Section 9.12.
 1638  */
 1639 static void
 1640 mld_set_version(struct mld_ifsoftc *mli, const int version)
 1641 {
 1642         int old_version_timer;
 1643 
 1644         MLD_LOCK_ASSERT();
 1645 
 1646         CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
 1647             version, mli->mli_ifp, if_name(mli->mli_ifp));
 1648 
 1649         if (version == MLD_VERSION_1) {
 1650                 /*
 1651                  * Compute the "Older Version Querier Present" timer as per
 1652                  * Section 9.12.
 1653                  */
 1654                 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
 1655                 old_version_timer *= PR_SLOWHZ;
 1656                 mli->mli_v1_timer = old_version_timer;
 1657         }
 1658 
 1659         if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
 1660                 mli->mli_version = MLD_VERSION_1;
 1661                 mld_v2_cancel_link_timers(mli);
 1662         }
 1663 }
 1664 
 1665 /*
 1666  * Cancel pending MLDv2 timers for the given link and all groups
 1667  * joined on it; state-change, general-query, and group-query timers.
 1668  */
 1669 static void
 1670 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
 1671 {
 1672         struct epoch_tracker     et;
 1673         struct in6_multi_head    inmh;
 1674         struct ifmultiaddr      *ifma;
 1675         struct ifnet            *ifp;
 1676         struct in6_multi        *inm;
 1677 
 1678         CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
 1679             mli->mli_ifp, if_name(mli->mli_ifp));
 1680 
 1681         SLIST_INIT(&inmh);
 1682         IN6_MULTI_LIST_LOCK_ASSERT();
 1683         MLD_LOCK_ASSERT();
 1684 
 1685         /*
 1686          * Fast-track this potentially expensive operation
 1687          * by checking all the global 'timer pending' flags.
 1688          */
 1689         if (!V_interface_timers_running6 &&
 1690             !V_state_change_timers_running6 &&
 1691             !V_current_state_timers_running6)
 1692                 return;
 1693 
 1694         mli->mli_v2_timer = 0;
 1695 
 1696         ifp = mli->mli_ifp;
 1697 
 1698         IF_ADDR_WLOCK(ifp);
 1699         NET_EPOCH_ENTER_ET(et);
 1700         CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
 1701                 inm = in6m_ifmultiaddr_get_inm(ifma);
 1702                 if (inm == NULL)
 1703                         continue;
 1704                 switch (inm->in6m_state) {
 1705                 case MLD_NOT_MEMBER:
 1706                 case MLD_SILENT_MEMBER:
 1707                 case MLD_IDLE_MEMBER:
 1708                 case MLD_LAZY_MEMBER:
 1709                 case MLD_SLEEPING_MEMBER:
 1710                 case MLD_AWAKENING_MEMBER:
 1711                         break;
 1712                 case MLD_LEAVING_MEMBER:
 1713                         /*
 1714                          * If we are leaving the group and switching
 1715                          * version, we need to release the final
 1716                          * reference held for issuing the INCLUDE {}.
 1717                          */
 1718                         if (inm->in6m_refcount == 1)
 1719                                 in6m_disconnect_locked(&inmh, inm);
 1720                         in6m_rele_locked(&inmh, inm);
 1721                         /* FALLTHROUGH */
 1722                 case MLD_G_QUERY_PENDING_MEMBER:
 1723                 case MLD_SG_QUERY_PENDING_MEMBER:
 1724                         in6m_clear_recorded(inm);
 1725                         /* FALLTHROUGH */
 1726                 case MLD_REPORTING_MEMBER:
 1727                         inm->in6m_sctimer = 0;
 1728                         inm->in6m_timer = 0;
 1729                         inm->in6m_state = MLD_REPORTING_MEMBER;
 1730                         /*
 1731                          * Free any pending MLDv2 state-change records.
 1732                          */
 1733                         mbufq_drain(&inm->in6m_scq);
 1734                         break;
 1735                 }
 1736         }
 1737         NET_EPOCH_EXIT_ET(et);
 1738         IF_ADDR_WUNLOCK(ifp);
 1739         in6m_release_list_deferred(&inmh);
 1740 }
 1741 
 1742 /*
 1743  * Global slowtimo handler.
 1744  * VIMAGE: Timeout handlers are expected to service all vimages.
 1745  */
 1746 void
 1747 mld_slowtimo(void)
 1748 {
 1749         VNET_ITERATOR_DECL(vnet_iter);
 1750 
 1751         VNET_LIST_RLOCK_NOSLEEP();
 1752         VNET_FOREACH(vnet_iter) {
 1753                 CURVNET_SET(vnet_iter);
 1754                 mld_slowtimo_vnet();
 1755                 CURVNET_RESTORE();
 1756         }
 1757         VNET_LIST_RUNLOCK_NOSLEEP();
 1758 }
 1759 
 1760 /*
 1761  * Per-vnet slowtimo handler.
 1762  */
 1763 static void
 1764 mld_slowtimo_vnet(void)
 1765 {
 1766         struct mld_ifsoftc *mli;
 1767 
 1768         MLD_LOCK();
 1769 
 1770         LIST_FOREACH(mli, &V_mli_head, mli_link) {
 1771                 mld_v1_process_querier_timers(mli);
 1772         }
 1773 
 1774         MLD_UNLOCK();
 1775 }
 1776 
 1777 /*
 1778  * Update the Older Version Querier Present timers for a link.
 1779  * See Section 9.12 of RFC 3810.
 1780  */
 1781 static void
 1782 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
 1783 {
 1784 
 1785         MLD_LOCK_ASSERT();
 1786 
 1787         if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
 1788                 /*
 1789                  * MLDv1 Querier Present timer expired; revert to MLDv2.
 1790                  */
 1791                 CTR5(KTR_MLD,
 1792                     "%s: transition from v%d -> v%d on %p(%s)",
 1793                     __func__, mli->mli_version, MLD_VERSION_2,
 1794                     mli->mli_ifp, if_name(mli->mli_ifp));
 1795                 mli->mli_version = MLD_VERSION_2;
 1796         }
 1797 }
 1798 
 1799 /*
 1800  * Transmit an MLDv1 report immediately.
 1801  */
 1802 static int
 1803 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
 1804 {
 1805         struct ifnet            *ifp;
 1806         struct in6_ifaddr       *ia;
 1807         struct ip6_hdr          *ip6;
 1808         struct mbuf             *mh, *md;
 1809         struct mld_hdr          *mld;
 1810 
 1811         IN6_MULTI_LIST_LOCK_ASSERT();
 1812         MLD_LOCK_ASSERT();
 1813         
 1814         ifp = in6m->in6m_ifp;
 1815         /* in process of being freed */
 1816         if (ifp == NULL)
 1817                 return (0);
 1818         ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
 1819         /* ia may be NULL if link-local address is tentative. */
 1820 
 1821         mh = m_gethdr(M_NOWAIT, MT_DATA);
 1822         if (mh == NULL) {
 1823                 if (ia != NULL)
 1824                         ifa_free(&ia->ia_ifa);
 1825                 return (ENOMEM);
 1826         }
 1827         md = m_get(M_NOWAIT, MT_DATA);
 1828         if (md == NULL) {
 1829                 m_free(mh);
 1830                 if (ia != NULL)
 1831                         ifa_free(&ia->ia_ifa);
 1832                 return (ENOMEM);
 1833         }
 1834         mh->m_next = md;
 1835 
 1836         /*
 1837          * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
 1838          * that ether_output() does not need to allocate another mbuf
 1839          * for the header in the most common case.
 1840          */
 1841         M_ALIGN(mh, sizeof(struct ip6_hdr));
 1842         mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
 1843         mh->m_len = sizeof(struct ip6_hdr);
 1844 
 1845         ip6 = mtod(mh, struct ip6_hdr *);
 1846         ip6->ip6_flow = 0;
 1847         ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
 1848         ip6->ip6_vfc |= IPV6_VERSION;
 1849         ip6->ip6_nxt = IPPROTO_ICMPV6;
 1850         ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
 1851         ip6->ip6_dst = in6m->in6m_addr;
 1852 
 1853         md->m_len = sizeof(struct mld_hdr);
 1854         mld = mtod(md, struct mld_hdr *);
 1855         mld->mld_type = type;
 1856         mld->mld_code = 0;
 1857         mld->mld_cksum = 0;
 1858         mld->mld_maxdelay = 0;
 1859         mld->mld_reserved = 0;
 1860         mld->mld_addr = in6m->in6m_addr;
 1861         in6_clearscope(&mld->mld_addr);
 1862         mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
 1863             sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
 1864 
 1865         mld_save_context(mh, ifp);
 1866         mh->m_flags |= M_MLDV1;
 1867 
 1868         mld_dispatch_packet(mh);
 1869 
 1870         if (ia != NULL)
 1871                 ifa_free(&ia->ia_ifa);
 1872         return (0);
 1873 }
 1874 
 1875 /*
 1876  * Process a state change from the upper layer for the given IPv6 group.
 1877  *
 1878  * Each socket holds a reference on the in_multi in its own ip_moptions.
 1879  * The socket layer will have made the necessary updates to.the group
 1880  * state, it is now up to MLD to issue a state change report if there
 1881  * has been any change between T0 (when the last state-change was issued)
 1882  * and T1 (now).
 1883  *
 1884  * We use the MLDv2 state machine at group level. The MLd module
 1885  * however makes the decision as to which MLD protocol version to speak.
 1886  * A state change *from* INCLUDE {} always means an initial join.
 1887  * A state change *to* INCLUDE {} always means a final leave.
 1888  *
 1889  * If delay is non-zero, and the state change is an initial multicast
 1890  * join, the state change report will be delayed by 'delay' ticks
 1891  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
 1892  * the initial MLDv2 state change report will be delayed by whichever
 1893  * is sooner, a pending state-change timer or delay itself.
 1894  *
 1895  * VIMAGE: curvnet should have been set by caller, as this routine
 1896  * is called from the socket option handlers.
 1897  */
 1898 int
 1899 mld_change_state(struct in6_multi *inm, const int delay)
 1900 {
 1901         struct mld_ifsoftc *mli;
 1902         struct ifnet *ifp;
 1903         int error;
 1904 
 1905         IN6_MULTI_LIST_LOCK_ASSERT();
 1906 
 1907         error = 0;
 1908 
 1909         /*
 1910          * Check if the in6_multi has already been disconnected.
 1911          */
 1912         if (inm->in6m_ifp == NULL) {
 1913                 CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
 1914                 return (0);
 1915         }
 1916 
 1917         /*
 1918          * Try to detect if the upper layer just asked us to change state
 1919          * for an interface which has now gone away.
 1920          */
 1921         KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
 1922         ifp = inm->in6m_ifma->ifma_ifp;
 1923         if (ifp == NULL)
 1924                 return (0);
 1925         /*
 1926          * Sanity check that netinet6's notion of ifp is the
 1927          * same as net's.
 1928          */
 1929         KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
 1930 
 1931         MLD_LOCK();
 1932         mli = MLD_IFINFO(ifp);
 1933         KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
 1934 
 1935         /*
 1936          * If we detect a state transition to or from MCAST_UNDEFINED
 1937          * for this group, then we are starting or finishing an MLD
 1938          * life cycle for this group.
 1939          */
 1940         if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
 1941                 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
 1942                     inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
 1943                 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
 1944                         CTR1(KTR_MLD, "%s: initial join", __func__);
 1945                         error = mld_initial_join(inm, mli, delay);
 1946                         goto out_locked;
 1947                 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
 1948                         CTR1(KTR_MLD, "%s: final leave", __func__);
 1949                         mld_final_leave(inm, mli);
 1950                         goto out_locked;
 1951                 }
 1952         } else {
 1953                 CTR1(KTR_MLD, "%s: filter set change", __func__);
 1954         }
 1955 
 1956         error = mld_handle_state_change(inm, mli);
 1957 
 1958 out_locked:
 1959         MLD_UNLOCK();
 1960         return (error);
 1961 }
 1962 
 1963 /*
 1964  * Perform the initial join for an MLD group.
 1965  *
 1966  * When joining a group:
 1967  *  If the group should have its MLD traffic suppressed, do nothing.
 1968  *  MLDv1 starts sending MLDv1 host membership reports.
 1969  *  MLDv2 will schedule an MLDv2 state-change report containing the
 1970  *  initial state of the membership.
 1971  *
 1972  * If the delay argument is non-zero, then we must delay sending the
 1973  * initial state change for delay ticks (in units of PR_FASTHZ).
 1974  */
 1975 static int
 1976 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
 1977     const int delay)
 1978 {
 1979         struct ifnet            *ifp;
 1980         struct mbufq            *mq;
 1981         int                      error, retval, syncstates;
 1982         int                      odelay;
 1983 #ifdef KTR
 1984         char                     ip6tbuf[INET6_ADDRSTRLEN];
 1985 #endif
 1986 
 1987         CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
 1988             __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 1989             inm->in6m_ifp, if_name(inm->in6m_ifp));
 1990 
 1991         error = 0;
 1992         syncstates = 1;
 1993 
 1994         ifp = inm->in6m_ifp;
 1995 
 1996         IN6_MULTI_LIST_LOCK_ASSERT();
 1997         MLD_LOCK_ASSERT();
 1998 
 1999         KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
 2000 
 2001         /*
 2002          * Groups joined on loopback or marked as 'not reported',
 2003          * enter the MLD_SILENT_MEMBER state and
 2004          * are never reported in any protocol exchanges.
 2005          * All other groups enter the appropriate state machine
 2006          * for the version in use on this link.
 2007          * A link marked as MLIF_SILENT causes MLD to be completely
 2008          * disabled for the link.
 2009          */
 2010         if ((ifp->if_flags & IFF_LOOPBACK) ||
 2011             (mli->mli_flags & MLIF_SILENT) ||
 2012             !mld_is_addr_reported(&inm->in6m_addr)) {
 2013                 CTR1(KTR_MLD,
 2014 "%s: not kicking state machine for silent group", __func__);
 2015                 inm->in6m_state = MLD_SILENT_MEMBER;
 2016                 inm->in6m_timer = 0;
 2017         } else {
 2018                 /*
 2019                  * Deal with overlapping in_multi lifecycle.
 2020                  * If this group was LEAVING, then make sure
 2021                  * we drop the reference we picked up to keep the
 2022                  * group around for the final INCLUDE {} enqueue.
 2023                  */
 2024                 if (mli->mli_version == MLD_VERSION_2 &&
 2025                     inm->in6m_state == MLD_LEAVING_MEMBER) {
 2026                         inm->in6m_refcount--;
 2027                         MPASS(inm->in6m_refcount > 0);
 2028                 }
 2029                 inm->in6m_state = MLD_REPORTING_MEMBER;
 2030 
 2031                 switch (mli->mli_version) {
 2032                 case MLD_VERSION_1:
 2033                         /*
 2034                          * If a delay was provided, only use it if
 2035                          * it is greater than the delay normally
 2036                          * used for an MLDv1 state change report,
 2037                          * and delay sending the initial MLDv1 report
 2038                          * by not transitioning to the IDLE state.
 2039                          */
 2040                         odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
 2041                         if (delay) {
 2042                                 inm->in6m_timer = max(delay, odelay);
 2043                                 V_current_state_timers_running6 = 1;
 2044                         } else {
 2045                                 inm->in6m_state = MLD_IDLE_MEMBER;
 2046                                 error = mld_v1_transmit_report(inm,
 2047                                      MLD_LISTENER_REPORT);
 2048                                 if (error == 0) {
 2049                                         inm->in6m_timer = odelay;
 2050                                         V_current_state_timers_running6 = 1;
 2051                                 }
 2052                         }
 2053                         break;
 2054 
 2055                 case MLD_VERSION_2:
 2056                         /*
 2057                          * Defer update of T0 to T1, until the first copy
 2058                          * of the state change has been transmitted.
 2059                          */
 2060                         syncstates = 0;
 2061 
 2062                         /*
 2063                          * Immediately enqueue a State-Change Report for
 2064                          * this interface, freeing any previous reports.
 2065                          * Don't kick the timers if there is nothing to do,
 2066                          * or if an error occurred.
 2067                          */
 2068                         mq = &inm->in6m_scq;
 2069                         mbufq_drain(mq);
 2070                         retval = mld_v2_enqueue_group_record(mq, inm, 1,
 2071                             0, 0, (mli->mli_flags & MLIF_USEALLOW));
 2072                         CTR2(KTR_MLD, "%s: enqueue record = %d",
 2073                             __func__, retval);
 2074                         if (retval <= 0) {
 2075                                 error = retval * -1;
 2076                                 break;
 2077                         }
 2078 
 2079                         /*
 2080                          * Schedule transmission of pending state-change
 2081                          * report up to RV times for this link. The timer
 2082                          * will fire at the next mld_fasttimo (~200ms),
 2083                          * giving us an opportunity to merge the reports.
 2084                          *
 2085                          * If a delay was provided to this function, only
 2086                          * use this delay if sooner than the existing one.
 2087                          */
 2088                         KASSERT(mli->mli_rv > 1,
 2089                            ("%s: invalid robustness %d", __func__,
 2090                             mli->mli_rv));
 2091                         inm->in6m_scrv = mli->mli_rv;
 2092                         if (delay) {
 2093                                 if (inm->in6m_sctimer > 1) {
 2094                                         inm->in6m_sctimer =
 2095                                             min(inm->in6m_sctimer, delay);
 2096                                 } else
 2097                                         inm->in6m_sctimer = delay;
 2098                         } else
 2099                                 inm->in6m_sctimer = 1;
 2100                         V_state_change_timers_running6 = 1;
 2101 
 2102                         error = 0;
 2103                         break;
 2104                 }
 2105         }
 2106 
 2107         /*
 2108          * Only update the T0 state if state change is atomic,
 2109          * i.e. we don't need to wait for a timer to fire before we
 2110          * can consider the state change to have been communicated.
 2111          */
 2112         if (syncstates) {
 2113                 in6m_commit(inm);
 2114                 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
 2115                     ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2116                     if_name(inm->in6m_ifp));
 2117         }
 2118 
 2119         return (error);
 2120 }
 2121 
 2122 /*
 2123  * Issue an intermediate state change during the life-cycle.
 2124  */
 2125 static int
 2126 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
 2127 {
 2128         struct ifnet            *ifp;
 2129         int                      retval;
 2130 #ifdef KTR
 2131         char                     ip6tbuf[INET6_ADDRSTRLEN];
 2132 #endif
 2133 
 2134         CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
 2135             __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2136             inm->in6m_ifp, if_name(inm->in6m_ifp));
 2137 
 2138         ifp = inm->in6m_ifp;
 2139 
 2140         IN6_MULTI_LIST_LOCK_ASSERT();
 2141         MLD_LOCK_ASSERT();
 2142 
 2143         KASSERT(mli && mli->mli_ifp == ifp,
 2144             ("%s: inconsistent ifp", __func__));
 2145 
 2146         if ((ifp->if_flags & IFF_LOOPBACK) ||
 2147             (mli->mli_flags & MLIF_SILENT) ||
 2148             !mld_is_addr_reported(&inm->in6m_addr) ||
 2149             (mli->mli_version != MLD_VERSION_2)) {
 2150                 if (!mld_is_addr_reported(&inm->in6m_addr)) {
 2151                         CTR1(KTR_MLD,
 2152 "%s: not kicking state machine for silent group", __func__);
 2153                 }
 2154                 CTR1(KTR_MLD, "%s: nothing to do", __func__);
 2155                 in6m_commit(inm);
 2156                 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
 2157                     ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2158                     if_name(inm->in6m_ifp));
 2159                 return (0);
 2160         }
 2161 
 2162         mbufq_drain(&inm->in6m_scq);
 2163 
 2164         retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
 2165             (mli->mli_flags & MLIF_USEALLOW));
 2166         CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
 2167         if (retval <= 0)
 2168                 return (-retval);
 2169 
 2170         /*
 2171          * If record(s) were enqueued, start the state-change
 2172          * report timer for this group.
 2173          */
 2174         inm->in6m_scrv = mli->mli_rv;
 2175         inm->in6m_sctimer = 1;
 2176         V_state_change_timers_running6 = 1;
 2177 
 2178         return (0);
 2179 }
 2180 
 2181 /*
 2182  * Perform the final leave for a multicast address.
 2183  *
 2184  * When leaving a group:
 2185  *  MLDv1 sends a DONE message, if and only if we are the reporter.
 2186  *  MLDv2 enqueues a state-change report containing a transition
 2187  *  to INCLUDE {} for immediate transmission.
 2188  */
 2189 static void
 2190 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
 2191 {
 2192         int syncstates;
 2193 #ifdef KTR
 2194         char ip6tbuf[INET6_ADDRSTRLEN];
 2195 #endif
 2196 
 2197         syncstates = 1;
 2198 
 2199         CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
 2200             __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2201             inm->in6m_ifp, if_name(inm->in6m_ifp));
 2202 
 2203         IN6_MULTI_LIST_LOCK_ASSERT();
 2204         MLD_LOCK_ASSERT();
 2205 
 2206         switch (inm->in6m_state) {
 2207         case MLD_NOT_MEMBER:
 2208         case MLD_SILENT_MEMBER:
 2209         case MLD_LEAVING_MEMBER:
 2210                 /* Already leaving or left; do nothing. */
 2211                 CTR1(KTR_MLD,
 2212 "%s: not kicking state machine for silent group", __func__);
 2213                 break;
 2214         case MLD_REPORTING_MEMBER:
 2215         case MLD_IDLE_MEMBER:
 2216         case MLD_G_QUERY_PENDING_MEMBER:
 2217         case MLD_SG_QUERY_PENDING_MEMBER:
 2218                 if (mli->mli_version == MLD_VERSION_1) {
 2219 #ifdef INVARIANTS
 2220                         if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
 2221                             inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
 2222                         panic("%s: MLDv2 state reached, not MLDv2 mode",
 2223                              __func__);
 2224 #endif
 2225                         mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
 2226                         inm->in6m_state = MLD_NOT_MEMBER;
 2227                         V_current_state_timers_running6 = 1;
 2228                 } else if (mli->mli_version == MLD_VERSION_2) {
 2229                         /*
 2230                          * Stop group timer and all pending reports.
 2231                          * Immediately enqueue a state-change report
 2232                          * TO_IN {} to be sent on the next fast timeout,
 2233                          * giving us an opportunity to merge reports.
 2234                          */
 2235                         mbufq_drain(&inm->in6m_scq);
 2236                         inm->in6m_timer = 0;
 2237                         inm->in6m_scrv = mli->mli_rv;
 2238                         CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
 2239                             "pending retransmissions.", __func__,
 2240                             ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2241                             if_name(inm->in6m_ifp), inm->in6m_scrv);
 2242                         if (inm->in6m_scrv == 0) {
 2243                                 inm->in6m_state = MLD_NOT_MEMBER;
 2244                                 inm->in6m_sctimer = 0;
 2245                         } else {
 2246                                 int retval;
 2247 
 2248                                 in6m_acquire_locked(inm);
 2249 
 2250                                 retval = mld_v2_enqueue_group_record(
 2251                                     &inm->in6m_scq, inm, 1, 0, 0,
 2252                                     (mli->mli_flags & MLIF_USEALLOW));
 2253                                 KASSERT(retval != 0,
 2254                                     ("%s: enqueue record = %d", __func__,
 2255                                      retval));
 2256 
 2257                                 inm->in6m_state = MLD_LEAVING_MEMBER;
 2258                                 inm->in6m_sctimer = 1;
 2259                                 V_state_change_timers_running6 = 1;
 2260                                 syncstates = 0;
 2261                         }
 2262                         break;
 2263                 }
 2264                 break;
 2265         case MLD_LAZY_MEMBER:
 2266         case MLD_SLEEPING_MEMBER:
 2267         case MLD_AWAKENING_MEMBER:
 2268                 /* Our reports are suppressed; do nothing. */
 2269                 break;
 2270         }
 2271 
 2272         if (syncstates) {
 2273                 in6m_commit(inm);
 2274                 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
 2275                     ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2276                     if_name(inm->in6m_ifp));
 2277                 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
 2278                 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
 2279                     __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
 2280         }
 2281 }
 2282 
 2283 /*
 2284  * Enqueue an MLDv2 group record to the given output queue.
 2285  *
 2286  * If is_state_change is zero, a current-state record is appended.
 2287  * If is_state_change is non-zero, a state-change report is appended.
 2288  *
 2289  * If is_group_query is non-zero, an mbuf packet chain is allocated.
 2290  * If is_group_query is zero, and if there is a packet with free space
 2291  * at the tail of the queue, it will be appended to providing there
 2292  * is enough free space.
 2293  * Otherwise a new mbuf packet chain is allocated.
 2294  *
 2295  * If is_source_query is non-zero, each source is checked to see if
 2296  * it was recorded for a Group-Source query, and will be omitted if
 2297  * it is not both in-mode and recorded.
 2298  *
 2299  * If use_block_allow is non-zero, state change reports for initial join
 2300  * and final leave, on an inclusive mode group with a source list, will be
 2301  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
 2302  *
 2303  * The function will attempt to allocate leading space in the packet
 2304  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
 2305  *
 2306  * If successful the size of all data appended to the queue is returned,
 2307  * otherwise an error code less than zero is returned, or zero if
 2308  * no record(s) were appended.
 2309  */
 2310 static int
 2311 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
 2312     const int is_state_change, const int is_group_query,
 2313     const int is_source_query, const int use_block_allow)
 2314 {
 2315         struct mldv2_record      mr;
 2316         struct mldv2_record     *pmr;
 2317         struct ifnet            *ifp;
 2318         struct ip6_msource      *ims, *nims;
 2319         struct mbuf             *m0, *m, *md;
 2320         int                      is_filter_list_change;
 2321         int                      minrec0len, m0srcs, msrcs, nbytes, off;
 2322         int                      record_has_sources;
 2323         int                      now;
 2324         int                      type;
 2325         uint8_t                  mode;
 2326 #ifdef KTR
 2327         char                     ip6tbuf[INET6_ADDRSTRLEN];
 2328 #endif
 2329 
 2330         IN6_MULTI_LIST_LOCK_ASSERT();
 2331 
 2332         ifp = inm->in6m_ifp;
 2333         is_filter_list_change = 0;
 2334         m = NULL;
 2335         m0 = NULL;
 2336         m0srcs = 0;
 2337         msrcs = 0;
 2338         nbytes = 0;
 2339         nims = NULL;
 2340         record_has_sources = 1;
 2341         pmr = NULL;
 2342         type = MLD_DO_NOTHING;
 2343         mode = inm->in6m_st[1].iss_fmode;
 2344 
 2345         /*
 2346          * If we did not transition out of ASM mode during t0->t1,
 2347          * and there are no source nodes to process, we can skip
 2348          * the generation of source records.
 2349          */
 2350         if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
 2351             inm->in6m_nsrc == 0)
 2352                 record_has_sources = 0;
 2353 
 2354         if (is_state_change) {
 2355                 /*
 2356                  * Queue a state change record.
 2357                  * If the mode did not change, and there are non-ASM
 2358                  * listeners or source filters present,
 2359                  * we potentially need to issue two records for the group.
 2360                  * If there are ASM listeners, and there was no filter
 2361                  * mode transition of any kind, do nothing.
 2362                  *
 2363                  * If we are transitioning to MCAST_UNDEFINED, we need
 2364                  * not send any sources. A transition to/from this state is
 2365                  * considered inclusive with some special treatment.
 2366                  *
 2367                  * If we are rewriting initial joins/leaves to use
 2368                  * ALLOW/BLOCK, and the group's membership is inclusive,
 2369                  * we need to send sources in all cases.
 2370                  */
 2371                 if (mode != inm->in6m_st[0].iss_fmode) {
 2372                         if (mode == MCAST_EXCLUDE) {
 2373                                 CTR1(KTR_MLD, "%s: change to EXCLUDE",
 2374                                     __func__);
 2375                                 type = MLD_CHANGE_TO_EXCLUDE_MODE;
 2376                         } else {
 2377                                 CTR1(KTR_MLD, "%s: change to INCLUDE",
 2378                                     __func__);
 2379                                 if (use_block_allow) {
 2380                                         /*
 2381                                          * XXX
 2382                                          * Here we're interested in state
 2383                                          * edges either direction between
 2384                                          * MCAST_UNDEFINED and MCAST_INCLUDE.
 2385                                          * Perhaps we should just check
 2386                                          * the group state, rather than
 2387                                          * the filter mode.
 2388                                          */
 2389                                         if (mode == MCAST_UNDEFINED) {
 2390                                                 type = MLD_BLOCK_OLD_SOURCES;
 2391                                         } else {
 2392                                                 type = MLD_ALLOW_NEW_SOURCES;
 2393                                         }
 2394                                 } else {
 2395                                         type = MLD_CHANGE_TO_INCLUDE_MODE;
 2396                                         if (mode == MCAST_UNDEFINED)
 2397                                                 record_has_sources = 0;
 2398                                 }
 2399                         }
 2400                 } else {
 2401                         if (record_has_sources) {
 2402                                 is_filter_list_change = 1;
 2403                         } else {
 2404                                 type = MLD_DO_NOTHING;
 2405                         }
 2406                 }
 2407         } else {
 2408                 /*
 2409                  * Queue a current state record.
 2410                  */
 2411                 if (mode == MCAST_EXCLUDE) {
 2412                         type = MLD_MODE_IS_EXCLUDE;
 2413                 } else if (mode == MCAST_INCLUDE) {
 2414                         type = MLD_MODE_IS_INCLUDE;
 2415                         KASSERT(inm->in6m_st[1].iss_asm == 0,
 2416                             ("%s: inm %p is INCLUDE but ASM count is %d",
 2417                              __func__, inm, inm->in6m_st[1].iss_asm));
 2418                 }
 2419         }
 2420 
 2421         /*
 2422          * Generate the filter list changes using a separate function.
 2423          */
 2424         if (is_filter_list_change)
 2425                 return (mld_v2_enqueue_filter_change(mq, inm));
 2426 
 2427         if (type == MLD_DO_NOTHING) {
 2428                 CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
 2429                     __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2430                     if_name(inm->in6m_ifp));
 2431                 return (0);
 2432         }
 2433 
 2434         /*
 2435          * If any sources are present, we must be able to fit at least
 2436          * one in the trailing space of the tail packet's mbuf,
 2437          * ideally more.
 2438          */
 2439         minrec0len = sizeof(struct mldv2_record);
 2440         if (record_has_sources)
 2441                 minrec0len += sizeof(struct in6_addr);
 2442 
 2443         CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
 2444             mld_rec_type_to_str(type),
 2445             ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2446             if_name(inm->in6m_ifp));
 2447 
 2448         /*
 2449          * Check if we have a packet in the tail of the queue for this
 2450          * group into which the first group record for this group will fit.
 2451          * Otherwise allocate a new packet.
 2452          * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
 2453          * Note: Group records for G/GSR query responses MUST be sent
 2454          * in their own packet.
 2455          */
 2456         m0 = mbufq_last(mq);
 2457         if (!is_group_query &&
 2458             m0 != NULL &&
 2459             (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
 2460             (m0->m_pkthdr.len + minrec0len) <
 2461              (ifp->if_mtu - MLD_MTUSPACE)) {
 2462                 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
 2463                             sizeof(struct mldv2_record)) /
 2464                             sizeof(struct in6_addr);
 2465                 m = m0;
 2466                 CTR1(KTR_MLD, "%s: use existing packet", __func__);
 2467         } else {
 2468                 if (mbufq_full(mq)) {
 2469                         CTR1(KTR_MLD, "%s: outbound queue full", __func__);
 2470                         return (-ENOMEM);
 2471                 }
 2472                 m = NULL;
 2473                 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
 2474                     sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
 2475                 if (!is_state_change && !is_group_query)
 2476                         m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 2477                 if (m == NULL)
 2478                         m = m_gethdr(M_NOWAIT, MT_DATA);
 2479                 if (m == NULL)
 2480                         return (-ENOMEM);
 2481 
 2482                 mld_save_context(m, ifp);
 2483 
 2484                 CTR1(KTR_MLD, "%s: allocated first packet", __func__);
 2485         }
 2486 
 2487         /*
 2488          * Append group record.
 2489          * If we have sources, we don't know how many yet.
 2490          */
 2491         mr.mr_type = type;
 2492         mr.mr_datalen = 0;
 2493         mr.mr_numsrc = 0;
 2494         mr.mr_addr = inm->in6m_addr;
 2495         in6_clearscope(&mr.mr_addr);
 2496         if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
 2497                 if (m != m0)
 2498                         m_freem(m);
 2499                 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
 2500                 return (-ENOMEM);
 2501         }
 2502         nbytes += sizeof(struct mldv2_record);
 2503 
 2504         /*
 2505          * Append as many sources as will fit in the first packet.
 2506          * If we are appending to a new packet, the chain allocation
 2507          * may potentially use clusters; use m_getptr() in this case.
 2508          * If we are appending to an existing packet, we need to obtain
 2509          * a pointer to the group record after m_append(), in case a new
 2510          * mbuf was allocated.
 2511          *
 2512          * Only append sources which are in-mode at t1. If we are
 2513          * transitioning to MCAST_UNDEFINED state on the group, and
 2514          * use_block_allow is zero, do not include source entries.
 2515          * Otherwise, we need to include this source in the report.
 2516          *
 2517          * Only report recorded sources in our filter set when responding
 2518          * to a group-source query.
 2519          */
 2520         if (record_has_sources) {
 2521                 if (m == m0) {
 2522                         md = m_last(m);
 2523                         pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
 2524                             md->m_len - nbytes);
 2525                 } else {
 2526                         md = m_getptr(m, 0, &off);
 2527                         pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
 2528                             off);
 2529                 }
 2530                 msrcs = 0;
 2531                 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
 2532                     nims) {
 2533                         CTR2(KTR_MLD, "%s: visit node %s", __func__,
 2534                             ip6_sprintf(ip6tbuf, &ims->im6s_addr));
 2535                         now = im6s_get_mode(inm, ims, 1);
 2536                         CTR2(KTR_MLD, "%s: node is %d", __func__, now);
 2537                         if ((now != mode) ||
 2538                             (now == mode &&
 2539                              (!use_block_allow && mode == MCAST_UNDEFINED))) {
 2540                                 CTR1(KTR_MLD, "%s: skip node", __func__);
 2541                                 continue;
 2542                         }
 2543                         if (is_source_query && ims->im6s_stp == 0) {
 2544                                 CTR1(KTR_MLD, "%s: skip unrecorded node",
 2545                                     __func__);
 2546                                 continue;
 2547                         }
 2548                         CTR1(KTR_MLD, "%s: append node", __func__);
 2549                         if (!m_append(m, sizeof(struct in6_addr),
 2550                             (void *)&ims->im6s_addr)) {
 2551                                 if (m != m0)
 2552                                         m_freem(m);
 2553                                 CTR1(KTR_MLD, "%s: m_append() failed.",
 2554                                     __func__);
 2555                                 return (-ENOMEM);
 2556                         }
 2557                         nbytes += sizeof(struct in6_addr);
 2558                         ++msrcs;
 2559                         if (msrcs == m0srcs)
 2560                                 break;
 2561                 }
 2562                 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
 2563                     msrcs);
 2564                 pmr->mr_numsrc = htons(msrcs);
 2565                 nbytes += (msrcs * sizeof(struct in6_addr));
 2566         }
 2567 
 2568         if (is_source_query && msrcs == 0) {
 2569                 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
 2570                 if (m != m0)
 2571                         m_freem(m);
 2572                 return (0);
 2573         }
 2574 
 2575         /*
 2576          * We are good to go with first packet.
 2577          */
 2578         if (m != m0) {
 2579                 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
 2580                 m->m_pkthdr.PH_vt.vt_nrecs = 1;
 2581                 mbufq_enqueue(mq, m);
 2582         } else
 2583                 m->m_pkthdr.PH_vt.vt_nrecs++;
 2584 
 2585         /*
 2586          * No further work needed if no source list in packet(s).
 2587          */
 2588         if (!record_has_sources)
 2589                 return (nbytes);
 2590 
 2591         /*
 2592          * Whilst sources remain to be announced, we need to allocate
 2593          * a new packet and fill out as many sources as will fit.
 2594          * Always try for a cluster first.
 2595          */
 2596         while (nims != NULL) {
 2597                 if (mbufq_full(mq)) {
 2598                         CTR1(KTR_MLD, "%s: outbound queue full", __func__);
 2599                         return (-ENOMEM);
 2600                 }
 2601                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 2602                 if (m == NULL)
 2603                         m = m_gethdr(M_NOWAIT, MT_DATA);
 2604                 if (m == NULL)
 2605                         return (-ENOMEM);
 2606                 mld_save_context(m, ifp);
 2607                 md = m_getptr(m, 0, &off);
 2608                 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
 2609                 CTR1(KTR_MLD, "%s: allocated next packet", __func__);
 2610 
 2611                 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
 2612                         if (m != m0)
 2613                                 m_freem(m);
 2614                         CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
 2615                         return (-ENOMEM);
 2616                 }
 2617                 m->m_pkthdr.PH_vt.vt_nrecs = 1;
 2618                 nbytes += sizeof(struct mldv2_record);
 2619 
 2620                 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
 2621                     sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
 2622 
 2623                 msrcs = 0;
 2624                 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
 2625                         CTR2(KTR_MLD, "%s: visit node %s",
 2626                             __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
 2627                         now = im6s_get_mode(inm, ims, 1);
 2628                         if ((now != mode) ||
 2629                             (now == mode &&
 2630                              (!use_block_allow && mode == MCAST_UNDEFINED))) {
 2631                                 CTR1(KTR_MLD, "%s: skip node", __func__);
 2632                                 continue;
 2633                         }
 2634                         if (is_source_query && ims->im6s_stp == 0) {
 2635                                 CTR1(KTR_MLD, "%s: skip unrecorded node",
 2636                                     __func__);
 2637                                 continue;
 2638                         }
 2639                         CTR1(KTR_MLD, "%s: append node", __func__);
 2640                         if (!m_append(m, sizeof(struct in6_addr),
 2641                             (void *)&ims->im6s_addr)) {
 2642                                 if (m != m0)
 2643                                         m_freem(m);
 2644                                 CTR1(KTR_MLD, "%s: m_append() failed.",
 2645                                     __func__);
 2646                                 return (-ENOMEM);
 2647                         }
 2648                         ++msrcs;
 2649                         if (msrcs == m0srcs)
 2650                                 break;
 2651                 }
 2652                 pmr->mr_numsrc = htons(msrcs);
 2653                 nbytes += (msrcs * sizeof(struct in6_addr));
 2654 
 2655                 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
 2656                 mbufq_enqueue(mq, m);
 2657         }
 2658 
 2659         return (nbytes);
 2660 }
 2661 
 2662 /*
 2663  * Type used to mark record pass completion.
 2664  * We exploit the fact we can cast to this easily from the
 2665  * current filter modes on each ip_msource node.
 2666  */
 2667 typedef enum {
 2668         REC_NONE = 0x00,        /* MCAST_UNDEFINED */
 2669         REC_ALLOW = 0x01,       /* MCAST_INCLUDE */
 2670         REC_BLOCK = 0x02,       /* MCAST_EXCLUDE */
 2671         REC_FULL = REC_ALLOW | REC_BLOCK
 2672 } rectype_t;
 2673 
 2674 /*
 2675  * Enqueue an MLDv2 filter list change to the given output queue.
 2676  *
 2677  * Source list filter state is held in an RB-tree. When the filter list
 2678  * for a group is changed without changing its mode, we need to compute
 2679  * the deltas between T0 and T1 for each source in the filter set,
 2680  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
 2681  *
 2682  * As we may potentially queue two record types, and the entire R-B tree
 2683  * needs to be walked at once, we break this out into its own function
 2684  * so we can generate a tightly packed queue of packets.
 2685  *
 2686  * XXX This could be written to only use one tree walk, although that makes
 2687  * serializing into the mbuf chains a bit harder. For now we do two walks
 2688  * which makes things easier on us, and it may or may not be harder on
 2689  * the L2 cache.
 2690  *
 2691  * If successful the size of all data appended to the queue is returned,
 2692  * otherwise an error code less than zero is returned, or zero if
 2693  * no record(s) were appended.
 2694  */
 2695 static int
 2696 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
 2697 {
 2698         static const int MINRECLEN =
 2699             sizeof(struct mldv2_record) + sizeof(struct in6_addr);
 2700         struct ifnet            *ifp;
 2701         struct mldv2_record      mr;
 2702         struct mldv2_record     *pmr;
 2703         struct ip6_msource      *ims, *nims;
 2704         struct mbuf             *m, *m0, *md;
 2705         int                      m0srcs, nbytes, npbytes, off, rsrcs, schanged;
 2706         int                      nallow, nblock;
 2707         uint8_t                  mode, now, then;
 2708         rectype_t                crt, drt, nrt;
 2709 #ifdef KTR
 2710         char                     ip6tbuf[INET6_ADDRSTRLEN];
 2711 #endif
 2712 
 2713         IN6_MULTI_LIST_LOCK_ASSERT();
 2714 
 2715         if (inm->in6m_nsrc == 0 ||
 2716             (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
 2717                 return (0);
 2718 
 2719         ifp = inm->in6m_ifp;                    /* interface */
 2720         mode = inm->in6m_st[1].iss_fmode;       /* filter mode at t1 */
 2721         crt = REC_NONE; /* current group record type */
 2722         drt = REC_NONE; /* mask of completed group record types */
 2723         nrt = REC_NONE; /* record type for current node */
 2724         m0srcs = 0;     /* # source which will fit in current mbuf chain */
 2725         npbytes = 0;    /* # of bytes appended this packet */
 2726         nbytes = 0;     /* # of bytes appended to group's state-change queue */
 2727         rsrcs = 0;      /* # sources encoded in current record */
 2728         schanged = 0;   /* # nodes encoded in overall filter change */
 2729         nallow = 0;     /* # of source entries in ALLOW_NEW */
 2730         nblock = 0;     /* # of source entries in BLOCK_OLD */
 2731         nims = NULL;    /* next tree node pointer */
 2732 
 2733         /*
 2734          * For each possible filter record mode.
 2735          * The first kind of source we encounter tells us which
 2736          * is the first kind of record we start appending.
 2737          * If a node transitioned to UNDEFINED at t1, its mode is treated
 2738          * as the inverse of the group's filter mode.
 2739          */
 2740         while (drt != REC_FULL) {
 2741                 do {
 2742                         m0 = mbufq_last(mq);
 2743                         if (m0 != NULL &&
 2744                             (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
 2745                              MLD_V2_REPORT_MAXRECS) &&
 2746                             (m0->m_pkthdr.len + MINRECLEN) <
 2747                              (ifp->if_mtu - MLD_MTUSPACE)) {
 2748                                 m = m0;
 2749                                 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
 2750                                             sizeof(struct mldv2_record)) /
 2751                                             sizeof(struct in6_addr);
 2752                                 CTR1(KTR_MLD,
 2753                                     "%s: use previous packet", __func__);
 2754                         } else {
 2755                                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 2756                                 if (m == NULL)
 2757                                         m = m_gethdr(M_NOWAIT, MT_DATA);
 2758                                 if (m == NULL) {
 2759                                         CTR1(KTR_MLD,
 2760                                             "%s: m_get*() failed", __func__);
 2761                                         return (-ENOMEM);
 2762                                 }
 2763                                 m->m_pkthdr.PH_vt.vt_nrecs = 0;
 2764                                 mld_save_context(m, ifp);
 2765                                 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
 2766                                     sizeof(struct mldv2_record)) /
 2767                                     sizeof(struct in6_addr);
 2768                                 npbytes = 0;
 2769                                 CTR1(KTR_MLD,
 2770                                     "%s: allocated new packet", __func__);
 2771                         }
 2772                         /*
 2773                          * Append the MLD group record header to the
 2774                          * current packet's data area.
 2775                          * Recalculate pointer to free space for next
 2776                          * group record, in case m_append() allocated
 2777                          * a new mbuf or cluster.
 2778                          */
 2779                         memset(&mr, 0, sizeof(mr));
 2780                         mr.mr_addr = inm->in6m_addr;
 2781                         in6_clearscope(&mr.mr_addr);
 2782                         if (!m_append(m, sizeof(mr), (void *)&mr)) {
 2783                                 if (m != m0)
 2784                                         m_freem(m);
 2785                                 CTR1(KTR_MLD,
 2786                                     "%s: m_append() failed", __func__);
 2787                                 return (-ENOMEM);
 2788                         }
 2789                         npbytes += sizeof(struct mldv2_record);
 2790                         if (m != m0) {
 2791                                 /* new packet; offset in chain */
 2792                                 md = m_getptr(m, npbytes -
 2793                                     sizeof(struct mldv2_record), &off);
 2794                                 pmr = (struct mldv2_record *)(mtod(md,
 2795                                     uint8_t *) + off);
 2796                         } else {
 2797                                 /* current packet; offset from last append */
 2798                                 md = m_last(m);
 2799                                 pmr = (struct mldv2_record *)(mtod(md,
 2800                                     uint8_t *) + md->m_len -
 2801                                     sizeof(struct mldv2_record));
 2802                         }
 2803                         /*
 2804                          * Begin walking the tree for this record type
 2805                          * pass, or continue from where we left off
 2806                          * previously if we had to allocate a new packet.
 2807                          * Only report deltas in-mode at t1.
 2808                          * We need not report included sources as allowed
 2809                          * if we are in inclusive mode on the group,
 2810                          * however the converse is not true.
 2811                          */
 2812                         rsrcs = 0;
 2813                         if (nims == NULL) {
 2814                                 nims = RB_MIN(ip6_msource_tree,
 2815                                     &inm->in6m_srcs);
 2816                         }
 2817                         RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
 2818                                 CTR2(KTR_MLD, "%s: visit node %s", __func__,
 2819                                     ip6_sprintf(ip6tbuf, &ims->im6s_addr));
 2820                                 now = im6s_get_mode(inm, ims, 1);
 2821                                 then = im6s_get_mode(inm, ims, 0);
 2822                                 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
 2823                                     __func__, then, now);
 2824                                 if (now == then) {
 2825                                         CTR1(KTR_MLD,
 2826                                             "%s: skip unchanged", __func__);
 2827                                         continue;
 2828                                 }
 2829                                 if (mode == MCAST_EXCLUDE &&
 2830                                     now == MCAST_INCLUDE) {
 2831                                         CTR1(KTR_MLD,
 2832                                             "%s: skip IN src on EX group",
 2833                                             __func__);
 2834                                         continue;
 2835                                 }
 2836                                 nrt = (rectype_t)now;
 2837                                 if (nrt == REC_NONE)
 2838                                         nrt = (rectype_t)(~mode & REC_FULL);
 2839                                 if (schanged++ == 0) {
 2840                                         crt = nrt;
 2841                                 } else if (crt != nrt)
 2842                                         continue;
 2843                                 if (!m_append(m, sizeof(struct in6_addr),
 2844                                     (void *)&ims->im6s_addr)) {
 2845                                         if (m != m0)
 2846                                                 m_freem(m);
 2847                                         CTR1(KTR_MLD,
 2848                                             "%s: m_append() failed", __func__);
 2849                                         return (-ENOMEM);
 2850                                 }
 2851                                 nallow += !!(crt == REC_ALLOW);
 2852                                 nblock += !!(crt == REC_BLOCK);
 2853                                 if (++rsrcs == m0srcs)
 2854                                         break;
 2855                         }
 2856                         /*
 2857                          * If we did not append any tree nodes on this
 2858                          * pass, back out of allocations.
 2859                          */
 2860                         if (rsrcs == 0) {
 2861                                 npbytes -= sizeof(struct mldv2_record);
 2862                                 if (m != m0) {
 2863                                         CTR1(KTR_MLD,
 2864                                             "%s: m_free(m)", __func__);
 2865                                         m_freem(m);
 2866                                 } else {
 2867                                         CTR1(KTR_MLD,
 2868                                             "%s: m_adj(m, -mr)", __func__);
 2869                                         m_adj(m, -((int)sizeof(
 2870                                             struct mldv2_record)));
 2871                                 }
 2872                                 continue;
 2873                         }
 2874                         npbytes += (rsrcs * sizeof(struct in6_addr));
 2875                         if (crt == REC_ALLOW)
 2876                                 pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
 2877                         else if (crt == REC_BLOCK)
 2878                                 pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
 2879                         pmr->mr_numsrc = htons(rsrcs);
 2880                         /*
 2881                          * Count the new group record, and enqueue this
 2882                          * packet if it wasn't already queued.
 2883                          */
 2884                         m->m_pkthdr.PH_vt.vt_nrecs++;
 2885                         if (m != m0)
 2886                                 mbufq_enqueue(mq, m);
 2887                         nbytes += npbytes;
 2888                 } while (nims != NULL);
 2889                 drt |= crt;
 2890                 crt = (~crt & REC_FULL);
 2891         }
 2892 
 2893         CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
 2894             nallow, nblock);
 2895 
 2896         return (nbytes);
 2897 }
 2898 
 2899 static int
 2900 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
 2901 {
 2902         struct mbufq    *gq;
 2903         struct mbuf     *m;             /* pending state-change */
 2904         struct mbuf     *m0;            /* copy of pending state-change */
 2905         struct mbuf     *mt;            /* last state-change in packet */
 2906         int              docopy, domerge;
 2907         u_int            recslen;
 2908 
 2909         docopy = 0;
 2910         domerge = 0;
 2911         recslen = 0;
 2912 
 2913         IN6_MULTI_LIST_LOCK_ASSERT();
 2914         MLD_LOCK_ASSERT();
 2915 
 2916         /*
 2917          * If there are further pending retransmissions, make a writable
 2918          * copy of each queued state-change message before merging.
 2919          */
 2920         if (inm->in6m_scrv > 0)
 2921                 docopy = 1;
 2922 
 2923         gq = &inm->in6m_scq;
 2924 #ifdef KTR
 2925         if (mbufq_first(gq) == NULL) {
 2926                 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
 2927                     __func__, inm);
 2928         }
 2929 #endif
 2930 
 2931         m = mbufq_first(gq);
 2932         while (m != NULL) {
 2933                 /*
 2934                  * Only merge the report into the current packet if
 2935                  * there is sufficient space to do so; an MLDv2 report
 2936                  * packet may only contain 65,535 group records.
 2937                  * Always use a simple mbuf chain concatentation to do this,
 2938                  * as large state changes for single groups may have
 2939                  * allocated clusters.
 2940                  */
 2941                 domerge = 0;
 2942                 mt = mbufq_last(scq);
 2943                 if (mt != NULL) {
 2944                         recslen = m_length(m, NULL);
 2945 
 2946                         if ((mt->m_pkthdr.PH_vt.vt_nrecs +
 2947                             m->m_pkthdr.PH_vt.vt_nrecs <=
 2948                             MLD_V2_REPORT_MAXRECS) &&
 2949                             (mt->m_pkthdr.len + recslen <=
 2950                             (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
 2951                                 domerge = 1;
 2952                 }
 2953 
 2954                 if (!domerge && mbufq_full(gq)) {
 2955                         CTR2(KTR_MLD,
 2956                             "%s: outbound queue full, skipping whole packet %p",
 2957                             __func__, m);
 2958                         mt = m->m_nextpkt;
 2959                         if (!docopy)
 2960                                 m_freem(m);
 2961                         m = mt;
 2962                         continue;
 2963                 }
 2964 
 2965                 if (!docopy) {
 2966                         CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
 2967                         m0 = mbufq_dequeue(gq);
 2968                         m = m0->m_nextpkt;
 2969                 } else {
 2970                         CTR2(KTR_MLD, "%s: copying %p", __func__, m);
 2971                         m0 = m_dup(m, M_NOWAIT);
 2972                         if (m0 == NULL)
 2973                                 return (ENOMEM);
 2974                         m0->m_nextpkt = NULL;
 2975                         m = m->m_nextpkt;
 2976                 }
 2977 
 2978                 if (!domerge) {
 2979                         CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
 2980                             __func__, m0, scq);
 2981                         mbufq_enqueue(scq, m0);
 2982                 } else {
 2983                         struct mbuf *mtl;       /* last mbuf of packet mt */
 2984 
 2985                         CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
 2986                             __func__, m0, mt);
 2987 
 2988                         mtl = m_last(mt);
 2989                         m0->m_flags &= ~M_PKTHDR;
 2990                         mt->m_pkthdr.len += recslen;
 2991                         mt->m_pkthdr.PH_vt.vt_nrecs +=
 2992                             m0->m_pkthdr.PH_vt.vt_nrecs;
 2993 
 2994                         mtl->m_next = m0;
 2995                 }
 2996         }
 2997 
 2998         return (0);
 2999 }
 3000 
 3001 /*
 3002  * Respond to a pending MLDv2 General Query.
 3003  */
 3004 static void
 3005 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
 3006 {
 3007         struct ifmultiaddr      *ifma;
 3008         struct ifnet            *ifp;
 3009         struct in6_multi        *inm;
 3010         int                      retval;
 3011 
 3012         IN6_MULTI_LIST_LOCK_ASSERT();
 3013         MLD_LOCK_ASSERT();
 3014 
 3015         KASSERT(mli->mli_version == MLD_VERSION_2,
 3016             ("%s: called when version %d", __func__, mli->mli_version));
 3017 
 3018         /*
 3019          * Check that there are some packets queued. If so, send them first.
 3020          * For large number of groups the reply to general query can take
 3021          * many packets, we should finish sending them before starting of
 3022          * queuing the new reply.
 3023          */
 3024         if (mbufq_len(&mli->mli_gq) != 0)
 3025                 goto send;
 3026 
 3027         ifp = mli->mli_ifp;
 3028 
 3029         IF_ADDR_RLOCK(ifp);
 3030         CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
 3031                 inm = in6m_ifmultiaddr_get_inm(ifma);
 3032                 if (inm == NULL)
 3033                         continue;
 3034                 KASSERT(ifp == inm->in6m_ifp,
 3035                     ("%s: inconsistent ifp", __func__));
 3036 
 3037                 switch (inm->in6m_state) {
 3038                 case MLD_NOT_MEMBER:
 3039                 case MLD_SILENT_MEMBER:
 3040                         break;
 3041                 case MLD_REPORTING_MEMBER:
 3042                 case MLD_IDLE_MEMBER:
 3043                 case MLD_LAZY_MEMBER:
 3044                 case MLD_SLEEPING_MEMBER:
 3045                 case MLD_AWAKENING_MEMBER:
 3046                         inm->in6m_state = MLD_REPORTING_MEMBER;
 3047                         retval = mld_v2_enqueue_group_record(&mli->mli_gq,
 3048                             inm, 0, 0, 0, 0);
 3049                         CTR2(KTR_MLD, "%s: enqueue record = %d",
 3050                             __func__, retval);
 3051                         break;
 3052                 case MLD_G_QUERY_PENDING_MEMBER:
 3053                 case MLD_SG_QUERY_PENDING_MEMBER:
 3054                 case MLD_LEAVING_MEMBER:
 3055                         break;
 3056                 }
 3057         }
 3058         IF_ADDR_RUNLOCK(ifp);
 3059 
 3060 send:
 3061         mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
 3062 
 3063         /*
 3064          * Slew transmission of bursts over 500ms intervals.
 3065          */
 3066         if (mbufq_first(&mli->mli_gq) != NULL) {
 3067                 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
 3068                     MLD_RESPONSE_BURST_INTERVAL);
 3069                 V_interface_timers_running6 = 1;
 3070         }
 3071 }
 3072 
 3073 /*
 3074  * Transmit the next pending message in the output queue.
 3075  *
 3076  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
 3077  * MRT: Nothing needs to be done, as MLD traffic is always local to
 3078  * a link and uses a link-scope multicast address.
 3079  */
 3080 static void
 3081 mld_dispatch_packet(struct mbuf *m)
 3082 {
 3083         struct ip6_moptions      im6o;
 3084         struct ifnet            *ifp;
 3085         struct ifnet            *oifp;
 3086         struct mbuf             *m0;
 3087         struct mbuf             *md;
 3088         struct ip6_hdr          *ip6;
 3089         struct mld_hdr          *mld;
 3090         int                      error;
 3091         int                      off;
 3092         int                      type;
 3093         uint32_t                 ifindex;
 3094 
 3095         CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
 3096 
 3097         /*
 3098          * Set VNET image pointer from enqueued mbuf chain
 3099          * before doing anything else. Whilst we use interface
 3100          * indexes to guard against interface detach, they are
 3101          * unique to each VIMAGE and must be retrieved.
 3102          */
 3103         ifindex = mld_restore_context(m);
 3104 
 3105         /*
 3106          * Check if the ifnet still exists. This limits the scope of
 3107          * any race in the absence of a global ifp lock for low cost
 3108          * (an array lookup).
 3109          */
 3110         ifp = ifnet_byindex(ifindex);
 3111         if (ifp == NULL) {
 3112                 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
 3113                     __func__, m, ifindex);
 3114                 m_freem(m);
 3115                 IP6STAT_INC(ip6s_noroute);
 3116                 goto out;
 3117         }
 3118 
 3119         im6o.im6o_multicast_hlim  = 1;
 3120         im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
 3121         im6o.im6o_multicast_ifp = ifp;
 3122 
 3123         if (m->m_flags & M_MLDV1) {
 3124                 m0 = m;
 3125         } else {
 3126                 m0 = mld_v2_encap_report(ifp, m);
 3127                 if (m0 == NULL) {
 3128                         CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
 3129                         IP6STAT_INC(ip6s_odropped);
 3130                         goto out;
 3131                 }
 3132         }
 3133 
 3134         mld_scrub_context(m0);
 3135         m_clrprotoflags(m);
 3136         m0->m_pkthdr.rcvif = V_loif;
 3137 
 3138         ip6 = mtod(m0, struct ip6_hdr *);
 3139 #if 0
 3140         (void)in6_setscope(&ip6->ip6_dst, ifp, NULL);   /* XXX LOR */
 3141 #else
 3142         /*
 3143          * XXX XXX Break some KPI rules to prevent an LOR which would
 3144          * occur if we called in6_setscope() at transmission.
 3145          * See comments at top of file.
 3146          */
 3147         MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
 3148 #endif
 3149 
 3150         /*
 3151          * Retrieve the ICMPv6 type before handoff to ip6_output(),
 3152          * so we can bump the stats.
 3153          */
 3154         md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
 3155         mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
 3156         type = mld->mld_type;
 3157 
 3158         oifp = NULL;
 3159         error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
 3160             &oifp, NULL);
 3161         if (error) {
 3162                 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
 3163                 goto out;
 3164         }
 3165         ICMP6STAT_INC(icp6s_outhist[type]);
 3166         if (oifp != NULL) {
 3167                 icmp6_ifstat_inc(oifp, ifs6_out_msg);
 3168                 switch (type) {
 3169                 case MLD_LISTENER_REPORT:
 3170                 case MLDV2_LISTENER_REPORT:
 3171                         icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
 3172                         break;
 3173                 case MLD_LISTENER_DONE:
 3174                         icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
 3175                         break;
 3176                 }
 3177         }
 3178 out:
 3179         return;
 3180 }
 3181 
 3182 /*
 3183  * Encapsulate an MLDv2 report.
 3184  *
 3185  * KAME IPv6 requires that hop-by-hop options be passed separately,
 3186  * and that the IPv6 header be prepended in a separate mbuf.
 3187  *
 3188  * Returns a pointer to the new mbuf chain head, or NULL if the
 3189  * allocation failed.
 3190  */
 3191 static struct mbuf *
 3192 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
 3193 {
 3194         struct mbuf             *mh;
 3195         struct mldv2_report     *mld;
 3196         struct ip6_hdr          *ip6;
 3197         struct in6_ifaddr       *ia;
 3198         int                      mldreclen;
 3199 
 3200         KASSERT(ifp != NULL, ("%s: null ifp", __func__));
 3201         KASSERT((m->m_flags & M_PKTHDR),
 3202             ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
 3203 
 3204         /*
 3205          * RFC3590: OK to send as :: or tentative during DAD.
 3206          */
 3207         ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
 3208         if (ia == NULL)
 3209                 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
 3210 
 3211         mh = m_gethdr(M_NOWAIT, MT_DATA);
 3212         if (mh == NULL) {
 3213                 if (ia != NULL)
 3214                         ifa_free(&ia->ia_ifa);
 3215                 m_freem(m);
 3216                 return (NULL);
 3217         }
 3218         M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
 3219 
 3220         mldreclen = m_length(m, NULL);
 3221         CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
 3222 
 3223         mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
 3224         mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
 3225             sizeof(struct mldv2_report) + mldreclen;
 3226 
 3227         ip6 = mtod(mh, struct ip6_hdr *);
 3228         ip6->ip6_flow = 0;
 3229         ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
 3230         ip6->ip6_vfc |= IPV6_VERSION;
 3231         ip6->ip6_nxt = IPPROTO_ICMPV6;
 3232         ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
 3233         if (ia != NULL)
 3234                 ifa_free(&ia->ia_ifa);
 3235         ip6->ip6_dst = in6addr_linklocal_allv2routers;
 3236         /* scope ID will be set in netisr */
 3237 
 3238         mld = (struct mldv2_report *)(ip6 + 1);
 3239         mld->mld_type = MLDV2_LISTENER_REPORT;
 3240         mld->mld_code = 0;
 3241         mld->mld_cksum = 0;
 3242         mld->mld_v2_reserved = 0;
 3243         mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
 3244         m->m_pkthdr.PH_vt.vt_nrecs = 0;
 3245 
 3246         mh->m_next = m;
 3247         mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
 3248             sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
 3249         return (mh);
 3250 }
 3251 
 3252 #ifdef KTR
 3253 static char *
 3254 mld_rec_type_to_str(const int type)
 3255 {
 3256 
 3257         switch (type) {
 3258                 case MLD_CHANGE_TO_EXCLUDE_MODE:
 3259                         return "TO_EX";
 3260                         break;
 3261                 case MLD_CHANGE_TO_INCLUDE_MODE:
 3262                         return "TO_IN";
 3263                         break;
 3264                 case MLD_MODE_IS_EXCLUDE:
 3265                         return "MODE_EX";
 3266                         break;
 3267                 case MLD_MODE_IS_INCLUDE:
 3268                         return "MODE_IN";
 3269                         break;
 3270                 case MLD_ALLOW_NEW_SOURCES:
 3271                         return "ALLOW_NEW";
 3272                         break;
 3273                 case MLD_BLOCK_OLD_SOURCES:
 3274                         return "BLOCK_OLD";
 3275                         break;
 3276                 default:
 3277                         break;
 3278         }
 3279         return "unknown";
 3280 }
 3281 #endif
 3282 
 3283 static void
 3284 mld_init(void *unused __unused)
 3285 {
 3286 
 3287         CTR1(KTR_MLD, "%s: initializing", __func__);
 3288         MLD_LOCK_INIT();
 3289 
 3290         ip6_initpktopts(&mld_po);
 3291         mld_po.ip6po_hlim = 1;
 3292         mld_po.ip6po_hbh = &mld_ra.hbh;
 3293         mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
 3294         mld_po.ip6po_flags = IP6PO_DONTFRAG;
 3295 }
 3296 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
 3297 
 3298 static void
 3299 mld_uninit(void *unused __unused)
 3300 {
 3301 
 3302         CTR1(KTR_MLD, "%s: tearing down", __func__);
 3303         MLD_LOCK_DESTROY();
 3304 }
 3305 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
 3306 
 3307 static void
 3308 vnet_mld_init(const void *unused __unused)
 3309 {
 3310 
 3311         CTR1(KTR_MLD, "%s: initializing", __func__);
 3312 
 3313         LIST_INIT(&V_mli_head);
 3314 }
 3315 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
 3316     NULL);
 3317 
 3318 static void
 3319 vnet_mld_uninit(const void *unused __unused)
 3320 {
 3321 
 3322         /* This can happen if we shutdown the network stack. */
 3323         CTR1(KTR_MLD, "%s: tearing down", __func__);
 3324 }
 3325 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
 3326     NULL);
 3327 
 3328 static int
 3329 mld_modevent(module_t mod, int type, void *unused __unused)
 3330 {
 3331 
 3332     switch (type) {
 3333     case MOD_LOAD:
 3334     case MOD_UNLOAD:
 3335         break;
 3336     default:
 3337         return (EOPNOTSUPP);
 3338     }
 3339     return (0);
 3340 }
 3341 
 3342 static moduledata_t mld_mod = {
 3343     "mld",
 3344     mld_modevent,
 3345     0
 3346 };
 3347 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);

Cache object: 624dadd864cbe23a1af962051b9da0c7


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.