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  * Copyright (c) 2009 Bruce Simpson.
    3  *
    4  * Redistribution and use in source and binary forms, with or without
    5  * modification, are permitted provided that the following conditions
    6  * are met:
    7  * 1. Redistributions of source code must retain the above copyright
    8  *    notice, this list of conditions and the following disclaimer.
    9  * 2. Redistributions in binary form must reproduce the above copyright
   10  *    notice, this list of conditions and the following disclaimer in the
   11  *    documentation and/or other materials provided with the distribution.
   12  * 3. The name of the author may not be used to endorse or promote
   13  *    products derived from this software without specific prior written
   14  *    permission.
   15  *
   16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   26  * SUCH DAMAGE.
   27  *
   28  *      $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
   29  */
   30 
   31 /*-
   32  * Copyright (c) 1988 Stephen Deering.
   33  * Copyright (c) 1992, 1993
   34  *      The Regents of the University of California.  All rights reserved.
   35  *
   36  * This code is derived from software contributed to Berkeley by
   37  * Stephen Deering of Stanford University.
   38  *
   39  * Redistribution and use in source and binary forms, with or without
   40  * modification, are permitted provided that the following conditions
   41  * are met:
   42  * 1. Redistributions of source code must retain the above copyright
   43  *    notice, this list of conditions and the following disclaimer.
   44  * 2. Redistributions in binary form must reproduce the above copyright
   45  *    notice, this list of conditions and the following disclaimer in the
   46  *    documentation and/or other materials provided with the distribution.
   47  * 4. Neither the name of the University nor the names of its contributors
   48  *    may be used to endorse or promote products derived from this software
   49  *    without specific prior written permission.
   50  *
   51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   61  * SUCH DAMAGE.
   62  *
   63  *      @(#)igmp.c      8.1 (Berkeley) 7/19/93
   64  */
   65 
   66 #include <sys/cdefs.h>
   67 __FBSDID("$FreeBSD: releng/10.0/sys/netinet6/mld6.c 254804 2013-08-24 19:51:18Z andre $");
   68 
   69 #include "opt_inet.h"
   70 #include "opt_inet6.h"
   71 
   72 #include <sys/param.h>
   73 #include <sys/systm.h>
   74 #include <sys/mbuf.h>
   75 #include <sys/socket.h>
   76 #include <sys/protosw.h>
   77 #include <sys/sysctl.h>
   78 #include <sys/kernel.h>
   79 #include <sys/callout.h>
   80 #include <sys/malloc.h>
   81 #include <sys/module.h>
   82 #include <sys/ktr.h>
   83 
   84 #include <net/if.h>
   85 #include <net/route.h>
   86 #include <net/vnet.h>
   87 
   88 #include <netinet/in.h>
   89 #include <netinet/in_var.h>
   90 #include <netinet6/in6_var.h>
   91 #include <netinet/ip6.h>
   92 #include <netinet6/ip6_var.h>
   93 #include <netinet6/scope6_var.h>
   94 #include <netinet/icmp6.h>
   95 #include <netinet6/mld6.h>
   96 #include <netinet6/mld6_var.h>
   97 
   98 #include <security/mac/mac_framework.h>
   99 
  100 #ifndef KTR_MLD
  101 #define KTR_MLD KTR_INET6
  102 #endif
  103 
  104 static struct mld_ifinfo *
  105                 mli_alloc_locked(struct ifnet *);
  106 static void     mli_delete_locked(const struct ifnet *);
  107 static void     mld_dispatch_packet(struct mbuf *);
  108 static void     mld_dispatch_queue(struct ifqueue *, int);
  109 static void     mld_final_leave(struct in6_multi *, struct mld_ifinfo *);
  110 static void     mld_fasttimo_vnet(void);
  111 static int      mld_handle_state_change(struct in6_multi *,
  112                     struct mld_ifinfo *);
  113 static int      mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
  114                     const int);
  115 #ifdef KTR
  116 static char *   mld_rec_type_to_str(const int);
  117 #endif
  118 static void     mld_set_version(struct mld_ifinfo *, const int);
  119 static void     mld_slowtimo_vnet(void);
  120 static int      mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
  121                     /*const*/ struct mld_hdr *);
  122 static int      mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
  123                     /*const*/ struct mld_hdr *);
  124 static void     mld_v1_process_group_timer(struct mld_ifinfo *,
  125                     struct in6_multi *);
  126 static void     mld_v1_process_querier_timers(struct mld_ifinfo *);
  127 static int      mld_v1_transmit_report(struct in6_multi *, const int);
  128 static void     mld_v1_update_group(struct in6_multi *, const int);
  129 static void     mld_v2_cancel_link_timers(struct mld_ifinfo *);
  130 static void     mld_v2_dispatch_general_query(struct mld_ifinfo *);
  131 static struct mbuf *
  132                 mld_v2_encap_report(struct ifnet *, struct mbuf *);
  133 static int      mld_v2_enqueue_filter_change(struct ifqueue *,
  134                     struct in6_multi *);
  135 static int      mld_v2_enqueue_group_record(struct ifqueue *,
  136                     struct in6_multi *, const int, const int, const int,
  137                     const int);
  138 static int      mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
  139                     struct mbuf *, const int, const int);
  140 static int      mld_v2_merge_state_changes(struct in6_multi *,
  141                     struct ifqueue *);
  142 static void     mld_v2_process_group_timers(struct mld_ifinfo *,
  143                     struct ifqueue *, struct ifqueue *,
  144                     struct in6_multi *, const int);
  145 static int      mld_v2_process_group_query(struct in6_multi *,
  146                     struct mld_ifinfo *mli, int, struct mbuf *, const int);
  147 static int      sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
  148 static int      sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
  149 
  150 /*
  151  * Normative references: RFC 2710, RFC 3590, RFC 3810.
  152  *
  153  * Locking:
  154  *  * The MLD subsystem lock ends up being system-wide for the moment,
  155  *    but could be per-VIMAGE later on.
  156  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
  157  *    Any may be taken independently; if any are held at the same
  158  *    time, the above lock order must be followed.
  159  *  * IN6_MULTI_LOCK covers in_multi.
  160  *  * MLD_LOCK covers per-link state and any global variables in this file.
  161  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
  162  *    per-link state iterators.
  163  *
  164  *  XXX LOR PREVENTION
  165  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
  166  *  will not accept an ifp; it wants an embedded scope ID, unlike
  167  *  ip_output(), which happily takes the ifp given to it. The embedded
  168  *  scope ID is only used by MLD to select the outgoing interface.
  169  *
  170  *  During interface attach and detach, MLD will take MLD_LOCK *after*
  171  *  the IF_AFDATA_LOCK.
  172  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
  173  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
  174  *  dispatch could work around this, but we'd rather not do that, as it
  175  *  can introduce other races.
  176  *
  177  *  As such, we exploit the fact that the scope ID is just the interface
  178  *  index, and embed it in the IPv6 destination address accordingly.
  179  *  This is potentially NOT VALID for MLDv1 reports, as they
  180  *  are always sent to the multicast group itself; as MLDv2
  181  *  reports are always sent to ff02::16, this is not an issue
  182  *  when MLDv2 is in use.
  183  *
  184  *  This does not however eliminate the LOR when ip6_output() itself
  185  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
  186  *  trigger a LOR warning in WITNESS when the ifnet is detached.
  187  *
  188  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
  189  *  how it's used across the network stack. Here we're simply exploiting
  190  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
  191  *
  192  * VIMAGE:
  193  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
  194  *    to a vnet in ifp->if_vnet.
  195  */
  196 static struct mtx                mld_mtx;
  197 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
  198 
  199 #define MLD_EMBEDSCOPE(pin6, zoneid)                                    \
  200         if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||                             \
  201             IN6_IS_ADDR_MC_INTFACELOCAL(pin6))                          \
  202                 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)         \
  203 
  204 /*
  205  * VIMAGE-wide globals.
  206  */
  207 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0};
  208 static VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head);
  209 static VNET_DEFINE(int, interface_timers_running6);
  210 static VNET_DEFINE(int, state_change_timers_running6);
  211 static VNET_DEFINE(int, current_state_timers_running6);
  212 
  213 #define V_mld_gsrdelay                  VNET(mld_gsrdelay)
  214 #define V_mli_head                      VNET(mli_head)
  215 #define V_interface_timers_running6     VNET(interface_timers_running6)
  216 #define V_state_change_timers_running6  VNET(state_change_timers_running6)
  217 #define V_current_state_timers_running6 VNET(current_state_timers_running6)
  218 
  219 SYSCTL_DECL(_net_inet6);        /* Note: Not in any common header. */
  220 
  221 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
  222     "IPv6 Multicast Listener Discovery");
  223 
  224 /*
  225  * Virtualized sysctls.
  226  */
  227 SYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
  228     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
  229     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
  230     "Rate limit for MLDv2 Group-and-Source queries in seconds");
  231 
  232 /*
  233  * Non-virtualized sysctls.
  234  */
  235 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
  236     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
  237     "Per-interface MLDv2 state");
  238 
  239 static int      mld_v1enable = 1;
  240 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW,
  241     &mld_v1enable, 0, "Enable fallback to MLDv1");
  242 TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable);
  243 
  244 static int      mld_use_allow = 1;
  245 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW,
  246     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
  247 TUNABLE_INT("net.inet6.mld.use_allow", &mld_use_allow);
  248 
  249 /*
  250  * Packed Router Alert option structure declaration.
  251  */
  252 struct mld_raopt {
  253         struct ip6_hbh          hbh;
  254         struct ip6_opt          pad;
  255         struct ip6_opt_router   ra;
  256 } __packed;
  257 
  258 /*
  259  * Router Alert hop-by-hop option header.
  260  */
  261 static struct mld_raopt mld_ra = {
  262         .hbh = { 0, 0 },
  263         .pad = { .ip6o_type = IP6OPT_PADN, 0 },
  264         .ra = {
  265             .ip6or_type = IP6OPT_ROUTER_ALERT,
  266             .ip6or_len = IP6OPT_RTALERT_LEN - 2,
  267             .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
  268             .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
  269         }
  270 };
  271 static struct ip6_pktopts mld_po;
  272 
  273 static __inline void
  274 mld_save_context(struct mbuf *m, struct ifnet *ifp)
  275 {
  276 
  277 #ifdef VIMAGE
  278         m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
  279 #endif /* VIMAGE */
  280         m->m_pkthdr.flowid = ifp->if_index;
  281 }
  282 
  283 static __inline void
  284 mld_scrub_context(struct mbuf *m)
  285 {
  286 
  287         m->m_pkthdr.PH_loc.ptr = NULL;
  288         m->m_pkthdr.flowid = 0;
  289 }
  290 
  291 /*
  292  * Restore context from a queued output chain.
  293  * Return saved ifindex.
  294  *
  295  * VIMAGE: The assertion is there to make sure that we
  296  * actually called CURVNET_SET() with what's in the mbuf chain.
  297  */
  298 static __inline uint32_t
  299 mld_restore_context(struct mbuf *m)
  300 {
  301 
  302 #if defined(VIMAGE) && defined(INVARIANTS)
  303         KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
  304             ("%s: called when curvnet was not restored", __func__));
  305 #endif
  306         return (m->m_pkthdr.flowid);
  307 }
  308 
  309 /*
  310  * Retrieve or set threshold between group-source queries in seconds.
  311  *
  312  * VIMAGE: Assume curvnet set by caller.
  313  * SMPng: NOTE: Serialized by MLD lock.
  314  */
  315 static int
  316 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
  317 {
  318         int error;
  319         int i;
  320 
  321         error = sysctl_wire_old_buffer(req, sizeof(int));
  322         if (error)
  323                 return (error);
  324 
  325         MLD_LOCK();
  326 
  327         i = V_mld_gsrdelay.tv_sec;
  328 
  329         error = sysctl_handle_int(oidp, &i, 0, req);
  330         if (error || !req->newptr)
  331                 goto out_locked;
  332 
  333         if (i < -1 || i >= 60) {
  334                 error = EINVAL;
  335                 goto out_locked;
  336         }
  337 
  338         CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
  339              V_mld_gsrdelay.tv_sec, i);
  340         V_mld_gsrdelay.tv_sec = i;
  341 
  342 out_locked:
  343         MLD_UNLOCK();
  344         return (error);
  345 }
  346 
  347 /*
  348  * Expose struct mld_ifinfo to userland, keyed by ifindex.
  349  * For use by ifmcstat(8).
  350  *
  351  * SMPng: NOTE: Does an unlocked ifindex space read.
  352  * VIMAGE: Assume curvnet set by caller. The node handler itself
  353  * is not directly virtualized.
  354  */
  355 static int
  356 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
  357 {
  358         int                     *name;
  359         int                      error;
  360         u_int                    namelen;
  361         struct ifnet            *ifp;
  362         struct mld_ifinfo       *mli;
  363 
  364         name = (int *)arg1;
  365         namelen = arg2;
  366 
  367         if (req->newptr != NULL)
  368                 return (EPERM);
  369 
  370         if (namelen != 1)
  371                 return (EINVAL);
  372 
  373         error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
  374         if (error)
  375                 return (error);
  376 
  377         IN6_MULTI_LOCK();
  378         MLD_LOCK();
  379 
  380         if (name[0] <= 0 || name[0] > V_if_index) {
  381                 error = ENOENT;
  382                 goto out_locked;
  383         }
  384 
  385         error = ENOENT;
  386 
  387         ifp = ifnet_byindex(name[0]);
  388         if (ifp == NULL)
  389                 goto out_locked;
  390 
  391         LIST_FOREACH(mli, &V_mli_head, mli_link) {
  392                 if (ifp == mli->mli_ifp) {
  393                         error = SYSCTL_OUT(req, mli,
  394                             sizeof(struct mld_ifinfo));
  395                         break;
  396                 }
  397         }
  398 
  399 out_locked:
  400         MLD_UNLOCK();
  401         IN6_MULTI_UNLOCK();
  402         return (error);
  403 }
  404 
  405 /*
  406  * Dispatch an entire queue of pending packet chains.
  407  * VIMAGE: Assumes the vnet pointer has been set.
  408  */
  409 static void
  410 mld_dispatch_queue(struct ifqueue *ifq, int limit)
  411 {
  412         struct mbuf *m;
  413 
  414         for (;;) {
  415                 _IF_DEQUEUE(ifq, m);
  416                 if (m == NULL)
  417                         break;
  418                 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m);
  419                 mld_dispatch_packet(m);
  420                 if (--limit == 0)
  421                         break;
  422         }
  423 }
  424 
  425 /*
  426  * Filter outgoing MLD report state by group.
  427  *
  428  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
  429  * and node-local addresses. However, kernel and socket consumers
  430  * always embed the KAME scope ID in the address provided, so strip it
  431  * when performing comparison.
  432  * Note: This is not the same as the *multicast* scope.
  433  *
  434  * Return zero if the given group is one for which MLD reports
  435  * should be suppressed, or non-zero if reports should be issued.
  436  */
  437 static __inline int
  438 mld_is_addr_reported(const struct in6_addr *addr)
  439 {
  440 
  441         KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
  442 
  443         if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
  444                 return (0);
  445 
  446         if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
  447                 struct in6_addr tmp = *addr;
  448                 in6_clearscope(&tmp);
  449                 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
  450                         return (0);
  451         }
  452 
  453         return (1);
  454 }
  455 
  456 /*
  457  * Attach MLD when PF_INET6 is attached to an interface.
  458  *
  459  * SMPng: Normally called with IF_AFDATA_LOCK held.
  460  */
  461 struct mld_ifinfo *
  462 mld_domifattach(struct ifnet *ifp)
  463 {
  464         struct mld_ifinfo *mli;
  465 
  466         CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
  467             __func__, ifp, ifp->if_xname);
  468 
  469         MLD_LOCK();
  470 
  471         mli = mli_alloc_locked(ifp);
  472         if (!(ifp->if_flags & IFF_MULTICAST))
  473                 mli->mli_flags |= MLIF_SILENT;
  474         if (mld_use_allow)
  475                 mli->mli_flags |= MLIF_USEALLOW;
  476 
  477         MLD_UNLOCK();
  478 
  479         return (mli);
  480 }
  481 
  482 /*
  483  * VIMAGE: assume curvnet set by caller.
  484  */
  485 static struct mld_ifinfo *
  486 mli_alloc_locked(/*const*/ struct ifnet *ifp)
  487 {
  488         struct mld_ifinfo *mli;
  489 
  490         MLD_LOCK_ASSERT();
  491 
  492         mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO);
  493         if (mli == NULL)
  494                 goto out;
  495 
  496         mli->mli_ifp = ifp;
  497         mli->mli_version = MLD_VERSION_2;
  498         mli->mli_flags = 0;
  499         mli->mli_rv = MLD_RV_INIT;
  500         mli->mli_qi = MLD_QI_INIT;
  501         mli->mli_qri = MLD_QRI_INIT;
  502         mli->mli_uri = MLD_URI_INIT;
  503 
  504         SLIST_INIT(&mli->mli_relinmhead);
  505 
  506         /*
  507          * Responses to general queries are subject to bounds.
  508          */
  509         IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
  510 
  511         LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
  512 
  513         CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)",
  514              ifp, ifp->if_xname);
  515 
  516 out:
  517         return (mli);
  518 }
  519 
  520 /*
  521  * Hook for ifdetach.
  522  *
  523  * NOTE: Some finalization tasks need to run before the protocol domain
  524  * is detached, but also before the link layer does its cleanup.
  525  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
  526  *
  527  * SMPng: Caller must hold IN6_MULTI_LOCK().
  528  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
  529  * XXX This routine is also bitten by unlocked ifma_protospec access.
  530  */
  531 void
  532 mld_ifdetach(struct ifnet *ifp)
  533 {
  534         struct mld_ifinfo       *mli;
  535         struct ifmultiaddr      *ifma;
  536         struct in6_multi        *inm, *tinm;
  537 
  538         CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
  539             ifp->if_xname);
  540 
  541         IN6_MULTI_LOCK_ASSERT();
  542         MLD_LOCK();
  543 
  544         mli = MLD_IFINFO(ifp);
  545         if (mli->mli_version == MLD_VERSION_2) {
  546                 IF_ADDR_RLOCK(ifp);
  547                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
  548                         if (ifma->ifma_addr->sa_family != AF_INET6 ||
  549                             ifma->ifma_protospec == NULL)
  550                                 continue;
  551                         inm = (struct in6_multi *)ifma->ifma_protospec;
  552                         if (inm->in6m_state == MLD_LEAVING_MEMBER) {
  553                                 SLIST_INSERT_HEAD(&mli->mli_relinmhead,
  554                                     inm, in6m_nrele);
  555                         }
  556                         in6m_clear_recorded(inm);
  557                 }
  558                 IF_ADDR_RUNLOCK(ifp);
  559                 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
  560                     tinm) {
  561                         SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
  562                         in6m_release_locked(inm);
  563                 }
  564         }
  565 
  566         MLD_UNLOCK();
  567 }
  568 
  569 /*
  570  * Hook for domifdetach.
  571  * Runs after link-layer cleanup; free MLD state.
  572  *
  573  * SMPng: Normally called with IF_AFDATA_LOCK held.
  574  */
  575 void
  576 mld_domifdetach(struct ifnet *ifp)
  577 {
  578 
  579         CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
  580             __func__, ifp, ifp->if_xname);
  581 
  582         MLD_LOCK();
  583         mli_delete_locked(ifp);
  584         MLD_UNLOCK();
  585 }
  586 
  587 static void
  588 mli_delete_locked(const struct ifnet *ifp)
  589 {
  590         struct mld_ifinfo *mli, *tmli;
  591 
  592         CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)",
  593             __func__, ifp, ifp->if_xname);
  594 
  595         MLD_LOCK_ASSERT();
  596 
  597         LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
  598                 if (mli->mli_ifp == ifp) {
  599                         /*
  600                          * Free deferred General Query responses.
  601                          */
  602                         _IF_DRAIN(&mli->mli_gq);
  603 
  604                         LIST_REMOVE(mli, mli_link);
  605 
  606                         KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
  607                             ("%s: there are dangling in_multi references",
  608                             __func__));
  609 
  610                         free(mli, M_MLD);
  611                         return;
  612                 }
  613         }
  614 #ifdef INVARIANTS
  615         panic("%s: mld_ifinfo not found for ifp %p\n", __func__,  ifp);
  616 #endif
  617 }
  618 
  619 /*
  620  * Process a received MLDv1 general or address-specific query.
  621  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
  622  *
  623  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
  624  * mld_addr. This is OK as we own the mbuf chain.
  625  */
  626 static int
  627 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
  628     /*const*/ struct mld_hdr *mld)
  629 {
  630         struct ifmultiaddr      *ifma;
  631         struct mld_ifinfo       *mli;
  632         struct in6_multi        *inm;
  633         int                      is_general_query;
  634         uint16_t                 timer;
  635 #ifdef KTR
  636         char                     ip6tbuf[INET6_ADDRSTRLEN];
  637 #endif
  638 
  639         is_general_query = 0;
  640 
  641         if (!mld_v1enable) {
  642                 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
  643                     ip6_sprintf(ip6tbuf, &mld->mld_addr),
  644                     ifp, ifp->if_xname);
  645                 return (0);
  646         }
  647 
  648         /*
  649          * RFC3810 Section 6.2: MLD queries must originate from
  650          * a router's link-local address.
  651          */
  652         if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
  653                 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
  654                     ip6_sprintf(ip6tbuf, &ip6->ip6_src),
  655                     ifp, ifp->if_xname);
  656                 return (0);
  657         }
  658 
  659         /*
  660          * Do address field validation upfront before we accept
  661          * the query.
  662          */
  663         if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
  664                 /*
  665                  * MLDv1 General Query.
  666                  * If this was not sent to the all-nodes group, ignore it.
  667                  */
  668                 struct in6_addr          dst;
  669 
  670                 dst = ip6->ip6_dst;
  671                 in6_clearscope(&dst);
  672                 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
  673                         return (EINVAL);
  674                 is_general_query = 1;
  675         } else {
  676                 /*
  677                  * Embed scope ID of receiving interface in MLD query for
  678                  * lookup whilst we don't hold other locks.
  679                  */
  680                 in6_setscope(&mld->mld_addr, ifp, NULL);
  681         }
  682 
  683         IN6_MULTI_LOCK();
  684         MLD_LOCK();
  685 
  686         /*
  687          * Switch to MLDv1 host compatibility mode.
  688          */
  689         mli = MLD_IFINFO(ifp);
  690         KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
  691         mld_set_version(mli, MLD_VERSION_1);
  692 
  693         timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
  694         if (timer == 0)
  695                 timer = 1;
  696 
  697         IF_ADDR_RLOCK(ifp);
  698         if (is_general_query) {
  699                 /*
  700                  * For each reporting group joined on this
  701                  * interface, kick the report timer.
  702                  */
  703                 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
  704                     ifp, ifp->if_xname);
  705                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
  706                         if (ifma->ifma_addr->sa_family != AF_INET6 ||
  707                             ifma->ifma_protospec == NULL)
  708                                 continue;
  709                         inm = (struct in6_multi *)ifma->ifma_protospec;
  710                         mld_v1_update_group(inm, timer);
  711                 }
  712         } else {
  713                 /*
  714                  * MLDv1 Group-Specific Query.
  715                  * If this is a group-specific MLDv1 query, we need only
  716                  * look up the single group to process it.
  717                  */
  718                 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
  719                 if (inm != NULL) {
  720                         CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
  721                             ip6_sprintf(ip6tbuf, &mld->mld_addr),
  722                             ifp, ifp->if_xname);
  723                         mld_v1_update_group(inm, timer);
  724                 }
  725                 /* XXX Clear embedded scope ID as userland won't expect it. */
  726                 in6_clearscope(&mld->mld_addr);
  727         }
  728 
  729         IF_ADDR_RUNLOCK(ifp);
  730         MLD_UNLOCK();
  731         IN6_MULTI_UNLOCK();
  732 
  733         return (0);
  734 }
  735 
  736 /*
  737  * Update the report timer on a group in response to an MLDv1 query.
  738  *
  739  * If we are becoming the reporting member for this group, start the timer.
  740  * If we already are the reporting member for this group, and timer is
  741  * below the threshold, reset it.
  742  *
  743  * We may be updating the group for the first time since we switched
  744  * to MLDv2. If we are, then we must clear any recorded source lists,
  745  * and transition to REPORTING state; the group timer is overloaded
  746  * for group and group-source query responses. 
  747  *
  748  * Unlike MLDv2, the delay per group should be jittered
  749  * to avoid bursts of MLDv1 reports.
  750  */
  751 static void
  752 mld_v1_update_group(struct in6_multi *inm, const int timer)
  753 {
  754 #ifdef KTR
  755         char                     ip6tbuf[INET6_ADDRSTRLEN];
  756 #endif
  757 
  758         CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
  759             ip6_sprintf(ip6tbuf, &inm->in6m_addr),
  760             inm->in6m_ifp->if_xname, timer);
  761 
  762         IN6_MULTI_LOCK_ASSERT();
  763 
  764         switch (inm->in6m_state) {
  765         case MLD_NOT_MEMBER:
  766         case MLD_SILENT_MEMBER:
  767                 break;
  768         case MLD_REPORTING_MEMBER:
  769                 if (inm->in6m_timer != 0 &&
  770                     inm->in6m_timer <= timer) {
  771                         CTR1(KTR_MLD, "%s: REPORTING and timer running, "
  772                             "skipping.", __func__);
  773                         break;
  774                 }
  775                 /* FALLTHROUGH */
  776         case MLD_SG_QUERY_PENDING_MEMBER:
  777         case MLD_G_QUERY_PENDING_MEMBER:
  778         case MLD_IDLE_MEMBER:
  779         case MLD_LAZY_MEMBER:
  780         case MLD_AWAKENING_MEMBER:
  781                 CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
  782                 inm->in6m_state = MLD_REPORTING_MEMBER;
  783                 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
  784                 V_current_state_timers_running6 = 1;
  785                 break;
  786         case MLD_SLEEPING_MEMBER:
  787                 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
  788                 inm->in6m_state = MLD_AWAKENING_MEMBER;
  789                 break;
  790         case MLD_LEAVING_MEMBER:
  791                 break;
  792         }
  793 }
  794 
  795 /*
  796  * Process a received MLDv2 general, group-specific or
  797  * group-and-source-specific query.
  798  *
  799  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
  800  *
  801  * Return 0 if successful, otherwise an appropriate error code is returned.
  802  */
  803 static int
  804 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
  805     struct mbuf *m, const int off, const int icmp6len)
  806 {
  807         struct mld_ifinfo       *mli;
  808         struct mldv2_query      *mld;
  809         struct in6_multi        *inm;
  810         uint32_t                 maxdelay, nsrc, qqi;
  811         int                      is_general_query;
  812         uint16_t                 timer;
  813         uint8_t                  qrv;
  814 #ifdef KTR
  815         char                     ip6tbuf[INET6_ADDRSTRLEN];
  816 #endif
  817 
  818         is_general_query = 0;
  819 
  820         /*
  821          * RFC3810 Section 6.2: MLD queries must originate from
  822          * a router's link-local address.
  823          */
  824         if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
  825                 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
  826                     ip6_sprintf(ip6tbuf, &ip6->ip6_src),
  827                     ifp, ifp->if_xname);
  828                 return (0);
  829         }
  830 
  831         CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname);
  832 
  833         mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
  834 
  835         maxdelay = ntohs(mld->mld_maxdelay);    /* in 1/10ths of a second */
  836         if (maxdelay >= 32768) {
  837                 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
  838                            (MLD_MRC_EXP(maxdelay) + 3);
  839         }
  840         timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
  841         if (timer == 0)
  842                 timer = 1;
  843 
  844         qrv = MLD_QRV(mld->mld_misc);
  845         if (qrv < 2) {
  846                 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
  847                     qrv, MLD_RV_INIT);
  848                 qrv = MLD_RV_INIT;
  849         }
  850 
  851         qqi = mld->mld_qqi;
  852         if (qqi >= 128) {
  853                 qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
  854                      (MLD_QQIC_EXP(mld->mld_qqi) + 3);
  855         }
  856 
  857         nsrc = ntohs(mld->mld_numsrc);
  858         if (nsrc > MLD_MAX_GS_SOURCES)
  859                 return (EMSGSIZE);
  860         if (icmp6len < sizeof(struct mldv2_query) +
  861             (nsrc * sizeof(struct in6_addr)))
  862                 return (EMSGSIZE);
  863 
  864         /*
  865          * Do further input validation upfront to avoid resetting timers
  866          * should we need to discard this query.
  867          */
  868         if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
  869                 /*
  870                  * A general query with a source list has undefined
  871                  * behaviour; discard it.
  872                  */
  873                 if (nsrc > 0)
  874                         return (EINVAL);
  875                 is_general_query = 1;
  876         } else {
  877                 /*
  878                  * Embed scope ID of receiving interface in MLD query for
  879                  * lookup whilst we don't hold other locks (due to KAME
  880                  * locking lameness). We own this mbuf chain just now.
  881                  */
  882                 in6_setscope(&mld->mld_addr, ifp, NULL);
  883         }
  884 
  885         IN6_MULTI_LOCK();
  886         MLD_LOCK();
  887 
  888         mli = MLD_IFINFO(ifp);
  889         KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
  890 
  891         /*
  892          * Discard the v2 query if we're in Compatibility Mode.
  893          * The RFC is pretty clear that hosts need to stay in MLDv1 mode
  894          * until the Old Version Querier Present timer expires.
  895          */
  896         if (mli->mli_version != MLD_VERSION_2)
  897                 goto out_locked;
  898 
  899         mld_set_version(mli, MLD_VERSION_2);
  900         mli->mli_rv = qrv;
  901         mli->mli_qi = qqi;
  902         mli->mli_qri = maxdelay;
  903 
  904         CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
  905             maxdelay);
  906 
  907         if (is_general_query) {
  908                 /*
  909                  * MLDv2 General Query.
  910                  *
  911                  * Schedule a current-state report on this ifp for
  912                  * all groups, possibly containing source lists.
  913                  *
  914                  * If there is a pending General Query response
  915                  * scheduled earlier than the selected delay, do
  916                  * not schedule any other reports.
  917                  * Otherwise, reset the interface timer.
  918                  */
  919                 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
  920                     ifp, ifp->if_xname);
  921                 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
  922                         mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
  923                         V_interface_timers_running6 = 1;
  924                 }
  925         } else {
  926                 /*
  927                  * MLDv2 Group-specific or Group-and-source-specific Query.
  928                  *
  929                  * Group-source-specific queries are throttled on
  930                  * a per-group basis to defeat denial-of-service attempts.
  931                  * Queries for groups we are not a member of on this
  932                  * link are simply ignored.
  933                  */
  934                 IF_ADDR_RLOCK(ifp);
  935                 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
  936                 if (inm == NULL) {
  937                         IF_ADDR_RUNLOCK(ifp);
  938                         goto out_locked;
  939                 }
  940                 if (nsrc > 0) {
  941                         if (!ratecheck(&inm->in6m_lastgsrtv,
  942                             &V_mld_gsrdelay)) {
  943                                 CTR1(KTR_MLD, "%s: GS query throttled.",
  944                                     __func__);
  945                                 IF_ADDR_RUNLOCK(ifp);
  946                                 goto out_locked;
  947                         }
  948                 }
  949                 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
  950                      ifp, ifp->if_xname);
  951                 /*
  952                  * If there is a pending General Query response
  953                  * scheduled sooner than the selected delay, no
  954                  * further report need be scheduled.
  955                  * Otherwise, prepare to respond to the
  956                  * group-specific or group-and-source query.
  957                  */
  958                 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
  959                         mld_v2_process_group_query(inm, mli, timer, m, off);
  960 
  961                 /* XXX Clear embedded scope ID as userland won't expect it. */
  962                 in6_clearscope(&mld->mld_addr);
  963                 IF_ADDR_RUNLOCK(ifp);
  964         }
  965 
  966 out_locked:
  967         MLD_UNLOCK();
  968         IN6_MULTI_UNLOCK();
  969 
  970         return (0);
  971 }
  972 
  973 /*
  974  * Process a recieved MLDv2 group-specific or group-and-source-specific
  975  * query.
  976  * Return <0 if any error occured. Currently this is ignored.
  977  */
  978 static int
  979 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli,
  980     int timer, struct mbuf *m0, const int off)
  981 {
  982         struct mldv2_query      *mld;
  983         int                      retval;
  984         uint16_t                 nsrc;
  985 
  986         IN6_MULTI_LOCK_ASSERT();
  987         MLD_LOCK_ASSERT();
  988 
  989         retval = 0;
  990         mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
  991 
  992         switch (inm->in6m_state) {
  993         case MLD_NOT_MEMBER:
  994         case MLD_SILENT_MEMBER:
  995         case MLD_SLEEPING_MEMBER:
  996         case MLD_LAZY_MEMBER:
  997         case MLD_AWAKENING_MEMBER:
  998         case MLD_IDLE_MEMBER:
  999         case MLD_LEAVING_MEMBER:
 1000                 return (retval);
 1001                 break;
 1002         case MLD_REPORTING_MEMBER:
 1003         case MLD_G_QUERY_PENDING_MEMBER:
 1004         case MLD_SG_QUERY_PENDING_MEMBER:
 1005                 break;
 1006         }
 1007 
 1008         nsrc = ntohs(mld->mld_numsrc);
 1009 
 1010         /*
 1011          * Deal with group-specific queries upfront.
 1012          * If any group query is already pending, purge any recorded
 1013          * source-list state if it exists, and schedule a query response
 1014          * for this group-specific query.
 1015          */
 1016         if (nsrc == 0) {
 1017                 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
 1018                     inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
 1019                         in6m_clear_recorded(inm);
 1020                         timer = min(inm->in6m_timer, timer);
 1021                 }
 1022                 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
 1023                 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
 1024                 V_current_state_timers_running6 = 1;
 1025                 return (retval);
 1026         }
 1027 
 1028         /*
 1029          * Deal with the case where a group-and-source-specific query has
 1030          * been received but a group-specific query is already pending.
 1031          */
 1032         if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
 1033                 timer = min(inm->in6m_timer, timer);
 1034                 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
 1035                 V_current_state_timers_running6 = 1;
 1036                 return (retval);
 1037         }
 1038 
 1039         /*
 1040          * Finally, deal with the case where a group-and-source-specific
 1041          * query has been received, where a response to a previous g-s-r
 1042          * query exists, or none exists.
 1043          * In this case, we need to parse the source-list which the Querier
 1044          * has provided us with and check if we have any source list filter
 1045          * entries at T1 for these sources. If we do not, there is no need
 1046          * schedule a report and the query may be dropped.
 1047          * If we do, we must record them and schedule a current-state
 1048          * report for those sources.
 1049          */
 1050         if (inm->in6m_nsrc > 0) {
 1051                 struct mbuf             *m;
 1052                 uint8_t                 *sp;
 1053                 int                      i, nrecorded;
 1054                 int                      soff;
 1055 
 1056                 m = m0;
 1057                 soff = off + sizeof(struct mldv2_query);
 1058                 nrecorded = 0;
 1059                 for (i = 0; i < nsrc; i++) {
 1060                         sp = mtod(m, uint8_t *) + soff;
 1061                         retval = in6m_record_source(inm,
 1062                             (const struct in6_addr *)sp);
 1063                         if (retval < 0)
 1064                                 break;
 1065                         nrecorded += retval;
 1066                         soff += sizeof(struct in6_addr);
 1067                         if (soff >= m->m_len) {
 1068                                 soff = soff - m->m_len;
 1069                                 m = m->m_next;
 1070                                 if (m == NULL)
 1071                                         break;
 1072                         }
 1073                 }
 1074                 if (nrecorded > 0) {
 1075                         CTR1(KTR_MLD,
 1076                             "%s: schedule response to SG query", __func__);
 1077                         inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
 1078                         inm->in6m_timer = MLD_RANDOM_DELAY(timer);
 1079                         V_current_state_timers_running6 = 1;
 1080                 }
 1081         }
 1082 
 1083         return (retval);
 1084 }
 1085 
 1086 /*
 1087  * Process a received MLDv1 host membership report.
 1088  * Assumes mld points to mld_hdr in pulled up mbuf chain.
 1089  *
 1090  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
 1091  * mld_addr. This is OK as we own the mbuf chain.
 1092  */
 1093 static int
 1094 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
 1095     /*const*/ struct mld_hdr *mld)
 1096 {
 1097         struct in6_addr          src, dst;
 1098         struct in6_ifaddr       *ia;
 1099         struct in6_multi        *inm;
 1100 #ifdef KTR
 1101         char                     ip6tbuf[INET6_ADDRSTRLEN];
 1102 #endif
 1103 
 1104         if (!mld_v1enable) {
 1105                 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
 1106                     ip6_sprintf(ip6tbuf, &mld->mld_addr),
 1107                     ifp, ifp->if_xname);
 1108                 return (0);
 1109         }
 1110 
 1111         if (ifp->if_flags & IFF_LOOPBACK)
 1112                 return (0);
 1113 
 1114         /*
 1115          * MLDv1 reports must originate from a host's link-local address,
 1116          * or the unspecified address (when booting).
 1117          */
 1118         src = ip6->ip6_src;
 1119         in6_clearscope(&src);
 1120         if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
 1121                 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
 1122                     ip6_sprintf(ip6tbuf, &ip6->ip6_src),
 1123                     ifp, ifp->if_xname);
 1124                 return (EINVAL);
 1125         }
 1126 
 1127         /*
 1128          * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
 1129          * group, and must be directed to the group itself.
 1130          */
 1131         dst = ip6->ip6_dst;
 1132         in6_clearscope(&dst);
 1133         if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
 1134             !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
 1135                 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
 1136                     ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
 1137                     ifp, ifp->if_xname);
 1138                 return (EINVAL);
 1139         }
 1140 
 1141         /*
 1142          * Make sure we don't hear our own membership report, as fast
 1143          * leave requires knowing that we are the only member of a
 1144          * group. Assume we used the link-local address if available,
 1145          * otherwise look for ::.
 1146          *
 1147          * XXX Note that scope ID comparison is needed for the address
 1148          * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
 1149          * performed for the on-wire address.
 1150          */
 1151         ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
 1152         if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
 1153             (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
 1154                 if (ia != NULL)
 1155                         ifa_free(&ia->ia_ifa);
 1156                 return (0);
 1157         }
 1158         if (ia != NULL)
 1159                 ifa_free(&ia->ia_ifa);
 1160 
 1161         CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
 1162             ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname);
 1163 
 1164         /*
 1165          * Embed scope ID of receiving interface in MLD query for lookup
 1166          * whilst we don't hold other locks (due to KAME locking lameness).
 1167          */
 1168         if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
 1169                 in6_setscope(&mld->mld_addr, ifp, NULL);
 1170 
 1171         IN6_MULTI_LOCK();
 1172         MLD_LOCK();
 1173         IF_ADDR_RLOCK(ifp);
 1174 
 1175         /*
 1176          * MLDv1 report suppression.
 1177          * If we are a member of this group, and our membership should be
 1178          * reported, and our group timer is pending or about to be reset,
 1179          * stop our group timer by transitioning to the 'lazy' state.
 1180          */
 1181         inm = in6m_lookup_locked(ifp, &mld->mld_addr);
 1182         if (inm != NULL) {
 1183                 struct mld_ifinfo *mli;
 1184 
 1185                 mli = inm->in6m_mli;
 1186                 KASSERT(mli != NULL,
 1187                     ("%s: no mli for ifp %p", __func__, ifp));
 1188 
 1189                 /*
 1190                  * If we are in MLDv2 host mode, do not allow the
 1191                  * other host's MLDv1 report to suppress our reports.
 1192                  */
 1193                 if (mli->mli_version == MLD_VERSION_2)
 1194                         goto out_locked;
 1195 
 1196                 inm->in6m_timer = 0;
 1197 
 1198                 switch (inm->in6m_state) {
 1199                 case MLD_NOT_MEMBER:
 1200                 case MLD_SILENT_MEMBER:
 1201                 case MLD_SLEEPING_MEMBER:
 1202                         break;
 1203                 case MLD_REPORTING_MEMBER:
 1204                 case MLD_IDLE_MEMBER:
 1205                 case MLD_AWAKENING_MEMBER:
 1206                         CTR3(KTR_MLD,
 1207                             "report suppressed for %s on ifp %p(%s)",
 1208                             ip6_sprintf(ip6tbuf, &mld->mld_addr),
 1209                             ifp, ifp->if_xname);
 1210                 case MLD_LAZY_MEMBER:
 1211                         inm->in6m_state = MLD_LAZY_MEMBER;
 1212                         break;
 1213                 case MLD_G_QUERY_PENDING_MEMBER:
 1214                 case MLD_SG_QUERY_PENDING_MEMBER:
 1215                 case MLD_LEAVING_MEMBER:
 1216                         break;
 1217                 }
 1218         }
 1219 
 1220 out_locked:
 1221         IF_ADDR_RUNLOCK(ifp);
 1222         MLD_UNLOCK();
 1223         IN6_MULTI_UNLOCK();
 1224 
 1225         /* XXX Clear embedded scope ID as userland won't expect it. */
 1226         in6_clearscope(&mld->mld_addr);
 1227 
 1228         return (0);
 1229 }
 1230 
 1231 /*
 1232  * MLD input path.
 1233  *
 1234  * Assume query messages which fit in a single ICMPv6 message header
 1235  * have been pulled up.
 1236  * Assume that userland will want to see the message, even if it
 1237  * otherwise fails kernel input validation; do not free it.
 1238  * Pullup may however free the mbuf chain m if it fails.
 1239  *
 1240  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
 1241  */
 1242 int
 1243 mld_input(struct mbuf *m, int off, int icmp6len)
 1244 {
 1245         struct ifnet    *ifp;
 1246         struct ip6_hdr  *ip6;
 1247         struct mld_hdr  *mld;
 1248         int              mldlen;
 1249 
 1250         CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
 1251 
 1252         ifp = m->m_pkthdr.rcvif;
 1253 
 1254         ip6 = mtod(m, struct ip6_hdr *);
 1255 
 1256         /* Pullup to appropriate size. */
 1257         mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
 1258         if (mld->mld_type == MLD_LISTENER_QUERY &&
 1259             icmp6len >= sizeof(struct mldv2_query)) {
 1260                 mldlen = sizeof(struct mldv2_query);
 1261         } else {
 1262                 mldlen = sizeof(struct mld_hdr);
 1263         }
 1264         IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
 1265         if (mld == NULL) {
 1266                 ICMP6STAT_INC(icp6s_badlen);
 1267                 return (IPPROTO_DONE);
 1268         }
 1269 
 1270         /*
 1271          * Userland needs to see all of this traffic for implementing
 1272          * the endpoint discovery portion of multicast routing.
 1273          */
 1274         switch (mld->mld_type) {
 1275         case MLD_LISTENER_QUERY:
 1276                 icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
 1277                 if (icmp6len == sizeof(struct mld_hdr)) {
 1278                         if (mld_v1_input_query(ifp, ip6, mld) != 0)
 1279                                 return (0);
 1280                 } else if (icmp6len >= sizeof(struct mldv2_query)) {
 1281                         if (mld_v2_input_query(ifp, ip6, m, off,
 1282                             icmp6len) != 0)
 1283                                 return (0);
 1284                 }
 1285                 break;
 1286         case MLD_LISTENER_REPORT:
 1287                 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
 1288                 if (mld_v1_input_report(ifp, ip6, mld) != 0)
 1289                         return (0);
 1290                 break;
 1291         case MLDV2_LISTENER_REPORT:
 1292                 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
 1293                 break;
 1294         case MLD_LISTENER_DONE:
 1295                 icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
 1296                 break;
 1297         default:
 1298                 break;
 1299         }
 1300 
 1301         return (0);
 1302 }
 1303 
 1304 /*
 1305  * Fast timeout handler (global).
 1306  * VIMAGE: Timeout handlers are expected to service all vimages.
 1307  */
 1308 void
 1309 mld_fasttimo(void)
 1310 {
 1311         VNET_ITERATOR_DECL(vnet_iter);
 1312 
 1313         VNET_LIST_RLOCK_NOSLEEP();
 1314         VNET_FOREACH(vnet_iter) {
 1315                 CURVNET_SET(vnet_iter);
 1316                 mld_fasttimo_vnet();
 1317                 CURVNET_RESTORE();
 1318         }
 1319         VNET_LIST_RUNLOCK_NOSLEEP();
 1320 }
 1321 
 1322 /*
 1323  * Fast timeout handler (per-vnet).
 1324  *
 1325  * VIMAGE: Assume caller has set up our curvnet.
 1326  */
 1327 static void
 1328 mld_fasttimo_vnet(void)
 1329 {
 1330         struct ifqueue           scq;   /* State-change packets */
 1331         struct ifqueue           qrq;   /* Query response packets */
 1332         struct ifnet            *ifp;
 1333         struct mld_ifinfo       *mli;
 1334         struct ifmultiaddr      *ifma;
 1335         struct in6_multi        *inm, *tinm;
 1336         int                      uri_fasthz;
 1337 
 1338         uri_fasthz = 0;
 1339 
 1340         /*
 1341          * Quick check to see if any work needs to be done, in order to
 1342          * minimize the overhead of fasttimo processing.
 1343          * SMPng: XXX Unlocked reads.
 1344          */
 1345         if (!V_current_state_timers_running6 &&
 1346             !V_interface_timers_running6 &&
 1347             !V_state_change_timers_running6)
 1348                 return;
 1349 
 1350         IN6_MULTI_LOCK();
 1351         MLD_LOCK();
 1352 
 1353         /*
 1354          * MLDv2 General Query response timer processing.
 1355          */
 1356         if (V_interface_timers_running6) {
 1357                 CTR1(KTR_MLD, "%s: interface timers running", __func__);
 1358 
 1359                 V_interface_timers_running6 = 0;
 1360                 LIST_FOREACH(mli, &V_mli_head, mli_link) {
 1361                         if (mli->mli_v2_timer == 0) {
 1362                                 /* Do nothing. */
 1363                         } else if (--mli->mli_v2_timer == 0) {
 1364                                 mld_v2_dispatch_general_query(mli);
 1365                         } else {
 1366                                 V_interface_timers_running6 = 1;
 1367                         }
 1368                 }
 1369         }
 1370 
 1371         if (!V_current_state_timers_running6 &&
 1372             !V_state_change_timers_running6)
 1373                 goto out_locked;
 1374 
 1375         V_current_state_timers_running6 = 0;
 1376         V_state_change_timers_running6 = 0;
 1377 
 1378         CTR1(KTR_MLD, "%s: state change timers running", __func__);
 1379 
 1380         /*
 1381          * MLD host report and state-change timer processing.
 1382          * Note: Processing a v2 group timer may remove a node.
 1383          */
 1384         LIST_FOREACH(mli, &V_mli_head, mli_link) {
 1385                 ifp = mli->mli_ifp;
 1386 
 1387                 if (mli->mli_version == MLD_VERSION_2) {
 1388                         uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
 1389                             PR_FASTHZ);
 1390 
 1391                         memset(&qrq, 0, sizeof(struct ifqueue));
 1392                         IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS);
 1393 
 1394                         memset(&scq, 0, sizeof(struct ifqueue));
 1395                         IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
 1396                 }
 1397 
 1398                 IF_ADDR_RLOCK(ifp);
 1399                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
 1400                         if (ifma->ifma_addr->sa_family != AF_INET6 ||
 1401                             ifma->ifma_protospec == NULL)
 1402                                 continue;
 1403                         inm = (struct in6_multi *)ifma->ifma_protospec;
 1404                         switch (mli->mli_version) {
 1405                         case MLD_VERSION_1:
 1406                                 mld_v1_process_group_timer(mli, inm);
 1407                                 break;
 1408                         case MLD_VERSION_2:
 1409                                 mld_v2_process_group_timers(mli, &qrq,
 1410                                     &scq, inm, uri_fasthz);
 1411                                 break;
 1412                         }
 1413                 }
 1414                 IF_ADDR_RUNLOCK(ifp);
 1415 
 1416                 switch (mli->mli_version) {
 1417                 case MLD_VERSION_1:
 1418                         /*
 1419                          * Transmit reports for this lifecycle.  This
 1420                          * is done while not holding IF_ADDR_LOCK
 1421                          * since this can call
 1422                          * in6ifa_ifpforlinklocal() which locks
 1423                          * IF_ADDR_LOCK internally as well as
 1424                          * ip6_output() to transmit a packet.
 1425                          */
 1426                         SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
 1427                             in6m_nrele, tinm) {
 1428                                 SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
 1429                                     in6m_nrele);
 1430                                 (void)mld_v1_transmit_report(inm,
 1431                                     MLD_LISTENER_REPORT);
 1432                         }
 1433                         break;
 1434                 case MLD_VERSION_2:
 1435                         mld_dispatch_queue(&qrq, 0);
 1436                         mld_dispatch_queue(&scq, 0);
 1437 
 1438                         /*
 1439                          * Free the in_multi reference(s) for
 1440                          * this lifecycle.
 1441                          */
 1442                         SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
 1443                             in6m_nrele, tinm) {
 1444                                 SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
 1445                                     in6m_nrele);
 1446                                 in6m_release_locked(inm);
 1447                         }
 1448                         break;
 1449                 }
 1450         }
 1451 
 1452 out_locked:
 1453         MLD_UNLOCK();
 1454         IN6_MULTI_UNLOCK();
 1455 }
 1456 
 1457 /*
 1458  * Update host report group timer.
 1459  * Will update the global pending timer flags.
 1460  */
 1461 static void
 1462 mld_v1_process_group_timer(struct mld_ifinfo *mli, struct in6_multi *inm)
 1463 {
 1464         int report_timer_expired;
 1465 
 1466         IN6_MULTI_LOCK_ASSERT();
 1467         MLD_LOCK_ASSERT();
 1468 
 1469         if (inm->in6m_timer == 0) {
 1470                 report_timer_expired = 0;
 1471         } else if (--inm->in6m_timer == 0) {
 1472                 report_timer_expired = 1;
 1473         } else {
 1474                 V_current_state_timers_running6 = 1;
 1475                 return;
 1476         }
 1477 
 1478         switch (inm->in6m_state) {
 1479         case MLD_NOT_MEMBER:
 1480         case MLD_SILENT_MEMBER:
 1481         case MLD_IDLE_MEMBER:
 1482         case MLD_LAZY_MEMBER:
 1483         case MLD_SLEEPING_MEMBER:
 1484         case MLD_AWAKENING_MEMBER:
 1485                 break;
 1486         case MLD_REPORTING_MEMBER:
 1487                 if (report_timer_expired) {
 1488                         inm->in6m_state = MLD_IDLE_MEMBER;
 1489                         SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
 1490                             in6m_nrele);
 1491                 }
 1492                 break;
 1493         case MLD_G_QUERY_PENDING_MEMBER:
 1494         case MLD_SG_QUERY_PENDING_MEMBER:
 1495         case MLD_LEAVING_MEMBER:
 1496                 break;
 1497         }
 1498 }
 1499 
 1500 /*
 1501  * Update a group's timers for MLDv2.
 1502  * Will update the global pending timer flags.
 1503  * Note: Unlocked read from mli.
 1504  */
 1505 static void
 1506 mld_v2_process_group_timers(struct mld_ifinfo *mli,
 1507     struct ifqueue *qrq, struct ifqueue *scq,
 1508     struct in6_multi *inm, const int uri_fasthz)
 1509 {
 1510         int query_response_timer_expired;
 1511         int state_change_retransmit_timer_expired;
 1512 #ifdef KTR
 1513         char ip6tbuf[INET6_ADDRSTRLEN];
 1514 #endif
 1515 
 1516         IN6_MULTI_LOCK_ASSERT();
 1517         MLD_LOCK_ASSERT();
 1518 
 1519         query_response_timer_expired = 0;
 1520         state_change_retransmit_timer_expired = 0;
 1521 
 1522         /*
 1523          * During a transition from compatibility mode back to MLDv2,
 1524          * a group record in REPORTING state may still have its group
 1525          * timer active. This is a no-op in this function; it is easier
 1526          * to deal with it here than to complicate the slow-timeout path.
 1527          */
 1528         if (inm->in6m_timer == 0) {
 1529                 query_response_timer_expired = 0;
 1530         } else if (--inm->in6m_timer == 0) {
 1531                 query_response_timer_expired = 1;
 1532         } else {
 1533                 V_current_state_timers_running6 = 1;
 1534         }
 1535 
 1536         if (inm->in6m_sctimer == 0) {
 1537                 state_change_retransmit_timer_expired = 0;
 1538         } else if (--inm->in6m_sctimer == 0) {
 1539                 state_change_retransmit_timer_expired = 1;
 1540         } else {
 1541                 V_state_change_timers_running6 = 1;
 1542         }
 1543 
 1544         /* We are in fasttimo, so be quick about it. */
 1545         if (!state_change_retransmit_timer_expired &&
 1546             !query_response_timer_expired)
 1547                 return;
 1548 
 1549         switch (inm->in6m_state) {
 1550         case MLD_NOT_MEMBER:
 1551         case MLD_SILENT_MEMBER:
 1552         case MLD_SLEEPING_MEMBER:
 1553         case MLD_LAZY_MEMBER:
 1554         case MLD_AWAKENING_MEMBER:
 1555         case MLD_IDLE_MEMBER:
 1556                 break;
 1557         case MLD_G_QUERY_PENDING_MEMBER:
 1558         case MLD_SG_QUERY_PENDING_MEMBER:
 1559                 /*
 1560                  * Respond to a previously pending Group-Specific
 1561                  * or Group-and-Source-Specific query by enqueueing
 1562                  * the appropriate Current-State report for
 1563                  * immediate transmission.
 1564                  */
 1565                 if (query_response_timer_expired) {
 1566                         int retval;
 1567 
 1568                         retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
 1569                             (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
 1570                             0);
 1571                         CTR2(KTR_MLD, "%s: enqueue record = %d",
 1572                             __func__, retval);
 1573                         inm->in6m_state = MLD_REPORTING_MEMBER;
 1574                         in6m_clear_recorded(inm);
 1575                 }
 1576                 /* FALLTHROUGH */
 1577         case MLD_REPORTING_MEMBER:
 1578         case MLD_LEAVING_MEMBER:
 1579                 if (state_change_retransmit_timer_expired) {
 1580                         /*
 1581                          * State-change retransmission timer fired.
 1582                          * If there are any further pending retransmissions,
 1583                          * set the global pending state-change flag, and
 1584                          * reset the timer.
 1585                          */
 1586                         if (--inm->in6m_scrv > 0) {
 1587                                 inm->in6m_sctimer = uri_fasthz;
 1588                                 V_state_change_timers_running6 = 1;
 1589                         }
 1590                         /*
 1591                          * Retransmit the previously computed state-change
 1592                          * report. If there are no further pending
 1593                          * retransmissions, the mbuf queue will be consumed.
 1594                          * Update T0 state to T1 as we have now sent
 1595                          * a state-change.
 1596                          */
 1597                         (void)mld_v2_merge_state_changes(inm, scq);
 1598 
 1599                         in6m_commit(inm);
 1600                         CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
 1601                             ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 1602                             inm->in6m_ifp->if_xname);
 1603 
 1604                         /*
 1605                          * If we are leaving the group for good, make sure
 1606                          * we release MLD's reference to it.
 1607                          * This release must be deferred using a SLIST,
 1608                          * as we are called from a loop which traverses
 1609                          * the in_ifmultiaddr TAILQ.
 1610                          */
 1611                         if (inm->in6m_state == MLD_LEAVING_MEMBER &&
 1612                             inm->in6m_scrv == 0) {
 1613                                 inm->in6m_state = MLD_NOT_MEMBER;
 1614                                 SLIST_INSERT_HEAD(&mli->mli_relinmhead,
 1615                                     inm, in6m_nrele);
 1616                         }
 1617                 }
 1618                 break;
 1619         }
 1620 }
 1621 
 1622 /*
 1623  * Switch to a different version on the given interface,
 1624  * as per Section 9.12.
 1625  */
 1626 static void
 1627 mld_set_version(struct mld_ifinfo *mli, const int version)
 1628 {
 1629         int old_version_timer;
 1630 
 1631         MLD_LOCK_ASSERT();
 1632 
 1633         CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
 1634             version, mli->mli_ifp, mli->mli_ifp->if_xname);
 1635 
 1636         if (version == MLD_VERSION_1) {
 1637                 /*
 1638                  * Compute the "Older Version Querier Present" timer as per
 1639                  * Section 9.12.
 1640                  */
 1641                 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
 1642                 old_version_timer *= PR_SLOWHZ;
 1643                 mli->mli_v1_timer = old_version_timer;
 1644         }
 1645 
 1646         if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
 1647                 mli->mli_version = MLD_VERSION_1;
 1648                 mld_v2_cancel_link_timers(mli);
 1649         }
 1650 }
 1651 
 1652 /*
 1653  * Cancel pending MLDv2 timers for the given link and all groups
 1654  * joined on it; state-change, general-query, and group-query timers.
 1655  */
 1656 static void
 1657 mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
 1658 {
 1659         struct ifmultiaddr      *ifma;
 1660         struct ifnet            *ifp;
 1661         struct in6_multi        *inm, *tinm;
 1662 
 1663         CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
 1664             mli->mli_ifp, mli->mli_ifp->if_xname);
 1665 
 1666         IN6_MULTI_LOCK_ASSERT();
 1667         MLD_LOCK_ASSERT();
 1668 
 1669         /*
 1670          * Fast-track this potentially expensive operation
 1671          * by checking all the global 'timer pending' flags.
 1672          */
 1673         if (!V_interface_timers_running6 &&
 1674             !V_state_change_timers_running6 &&
 1675             !V_current_state_timers_running6)
 1676                 return;
 1677 
 1678         mli->mli_v2_timer = 0;
 1679 
 1680         ifp = mli->mli_ifp;
 1681 
 1682         IF_ADDR_RLOCK(ifp);
 1683         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
 1684                 if (ifma->ifma_addr->sa_family != AF_INET6)
 1685                         continue;
 1686                 inm = (struct in6_multi *)ifma->ifma_protospec;
 1687                 switch (inm->in6m_state) {
 1688                 case MLD_NOT_MEMBER:
 1689                 case MLD_SILENT_MEMBER:
 1690                 case MLD_IDLE_MEMBER:
 1691                 case MLD_LAZY_MEMBER:
 1692                 case MLD_SLEEPING_MEMBER:
 1693                 case MLD_AWAKENING_MEMBER:
 1694                         break;
 1695                 case MLD_LEAVING_MEMBER:
 1696                         /*
 1697                          * If we are leaving the group and switching
 1698                          * version, we need to release the final
 1699                          * reference held for issuing the INCLUDE {}.
 1700                          */
 1701                         SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
 1702                             in6m_nrele);
 1703                         /* FALLTHROUGH */
 1704                 case MLD_G_QUERY_PENDING_MEMBER:
 1705                 case MLD_SG_QUERY_PENDING_MEMBER:
 1706                         in6m_clear_recorded(inm);
 1707                         /* FALLTHROUGH */
 1708                 case MLD_REPORTING_MEMBER:
 1709                         inm->in6m_sctimer = 0;
 1710                         inm->in6m_timer = 0;
 1711                         inm->in6m_state = MLD_REPORTING_MEMBER;
 1712                         /*
 1713                          * Free any pending MLDv2 state-change records.
 1714                          */
 1715                         _IF_DRAIN(&inm->in6m_scq);
 1716                         break;
 1717                 }
 1718         }
 1719         IF_ADDR_RUNLOCK(ifp);
 1720         SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) {
 1721                 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
 1722                 in6m_release_locked(inm);
 1723         }
 1724 }
 1725 
 1726 /*
 1727  * Global slowtimo handler.
 1728  * VIMAGE: Timeout handlers are expected to service all vimages.
 1729  */
 1730 void
 1731 mld_slowtimo(void)
 1732 {
 1733         VNET_ITERATOR_DECL(vnet_iter);
 1734 
 1735         VNET_LIST_RLOCK_NOSLEEP();
 1736         VNET_FOREACH(vnet_iter) {
 1737                 CURVNET_SET(vnet_iter);
 1738                 mld_slowtimo_vnet();
 1739                 CURVNET_RESTORE();
 1740         }
 1741         VNET_LIST_RUNLOCK_NOSLEEP();
 1742 }
 1743 
 1744 /*
 1745  * Per-vnet slowtimo handler.
 1746  */
 1747 static void
 1748 mld_slowtimo_vnet(void)
 1749 {
 1750         struct mld_ifinfo *mli;
 1751 
 1752         MLD_LOCK();
 1753 
 1754         LIST_FOREACH(mli, &V_mli_head, mli_link) {
 1755                 mld_v1_process_querier_timers(mli);
 1756         }
 1757 
 1758         MLD_UNLOCK();
 1759 }
 1760 
 1761 /*
 1762  * Update the Older Version Querier Present timers for a link.
 1763  * See Section 9.12 of RFC 3810.
 1764  */
 1765 static void
 1766 mld_v1_process_querier_timers(struct mld_ifinfo *mli)
 1767 {
 1768 
 1769         MLD_LOCK_ASSERT();
 1770 
 1771         if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
 1772                 /*
 1773                  * MLDv1 Querier Present timer expired; revert to MLDv2.
 1774                  */
 1775                 CTR5(KTR_MLD,
 1776                     "%s: transition from v%d -> v%d on %p(%s)",
 1777                     __func__, mli->mli_version, MLD_VERSION_2,
 1778                     mli->mli_ifp, mli->mli_ifp->if_xname);
 1779                 mli->mli_version = MLD_VERSION_2;
 1780         }
 1781 }
 1782 
 1783 /*
 1784  * Transmit an MLDv1 report immediately.
 1785  */
 1786 static int
 1787 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
 1788 {
 1789         struct ifnet            *ifp;
 1790         struct in6_ifaddr       *ia;
 1791         struct ip6_hdr          *ip6;
 1792         struct mbuf             *mh, *md;
 1793         struct mld_hdr          *mld;
 1794 
 1795         IN6_MULTI_LOCK_ASSERT();
 1796         MLD_LOCK_ASSERT();
 1797 
 1798         ifp = in6m->in6m_ifp;
 1799         ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
 1800         /* ia may be NULL if link-local address is tentative. */
 1801 
 1802         mh = m_gethdr(M_NOWAIT, MT_DATA);
 1803         if (mh == NULL) {
 1804                 if (ia != NULL)
 1805                         ifa_free(&ia->ia_ifa);
 1806                 return (ENOMEM);
 1807         }
 1808         md = m_get(M_NOWAIT, MT_DATA);
 1809         if (md == NULL) {
 1810                 m_free(mh);
 1811                 if (ia != NULL)
 1812                         ifa_free(&ia->ia_ifa);
 1813                 return (ENOMEM);
 1814         }
 1815         mh->m_next = md;
 1816 
 1817         /*
 1818          * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
 1819          * that ether_output() does not need to allocate another mbuf
 1820          * for the header in the most common case.
 1821          */
 1822         MH_ALIGN(mh, sizeof(struct ip6_hdr));
 1823         mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
 1824         mh->m_len = sizeof(struct ip6_hdr);
 1825 
 1826         ip6 = mtod(mh, struct ip6_hdr *);
 1827         ip6->ip6_flow = 0;
 1828         ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
 1829         ip6->ip6_vfc |= IPV6_VERSION;
 1830         ip6->ip6_nxt = IPPROTO_ICMPV6;
 1831         ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
 1832         ip6->ip6_dst = in6m->in6m_addr;
 1833 
 1834         md->m_len = sizeof(struct mld_hdr);
 1835         mld = mtod(md, struct mld_hdr *);
 1836         mld->mld_type = type;
 1837         mld->mld_code = 0;
 1838         mld->mld_cksum = 0;
 1839         mld->mld_maxdelay = 0;
 1840         mld->mld_reserved = 0;
 1841         mld->mld_addr = in6m->in6m_addr;
 1842         in6_clearscope(&mld->mld_addr);
 1843         mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
 1844             sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
 1845 
 1846         mld_save_context(mh, ifp);
 1847         mh->m_flags |= M_MLDV1;
 1848 
 1849         mld_dispatch_packet(mh);
 1850 
 1851         if (ia != NULL)
 1852                 ifa_free(&ia->ia_ifa);
 1853         return (0);
 1854 }
 1855 
 1856 /*
 1857  * Process a state change from the upper layer for the given IPv6 group.
 1858  *
 1859  * Each socket holds a reference on the in_multi in its own ip_moptions.
 1860  * The socket layer will have made the necessary updates to.the group
 1861  * state, it is now up to MLD to issue a state change report if there
 1862  * has been any change between T0 (when the last state-change was issued)
 1863  * and T1 (now).
 1864  *
 1865  * We use the MLDv2 state machine at group level. The MLd module
 1866  * however makes the decision as to which MLD protocol version to speak.
 1867  * A state change *from* INCLUDE {} always means an initial join.
 1868  * A state change *to* INCLUDE {} always means a final leave.
 1869  *
 1870  * If delay is non-zero, and the state change is an initial multicast
 1871  * join, the state change report will be delayed by 'delay' ticks
 1872  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
 1873  * the initial MLDv2 state change report will be delayed by whichever
 1874  * is sooner, a pending state-change timer or delay itself.
 1875  *
 1876  * VIMAGE: curvnet should have been set by caller, as this routine
 1877  * is called from the socket option handlers.
 1878  */
 1879 int
 1880 mld_change_state(struct in6_multi *inm, const int delay)
 1881 {
 1882         struct mld_ifinfo *mli;
 1883         struct ifnet *ifp;
 1884         int error;
 1885 
 1886         IN6_MULTI_LOCK_ASSERT();
 1887 
 1888         error = 0;
 1889 
 1890         /*
 1891          * Try to detect if the upper layer just asked us to change state
 1892          * for an interface which has now gone away.
 1893          */
 1894         KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
 1895         ifp = inm->in6m_ifma->ifma_ifp;
 1896         if (ifp != NULL) {
 1897                 /*
 1898                  * Sanity check that netinet6's notion of ifp is the
 1899                  * same as net's.
 1900                  */
 1901                 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
 1902         }
 1903 
 1904         MLD_LOCK();
 1905 
 1906         mli = MLD_IFINFO(ifp);
 1907         KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
 1908 
 1909         /*
 1910          * If we detect a state transition to or from MCAST_UNDEFINED
 1911          * for this group, then we are starting or finishing an MLD
 1912          * life cycle for this group.
 1913          */
 1914         if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
 1915                 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
 1916                     inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
 1917                 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
 1918                         CTR1(KTR_MLD, "%s: initial join", __func__);
 1919                         error = mld_initial_join(inm, mli, delay);
 1920                         goto out_locked;
 1921                 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
 1922                         CTR1(KTR_MLD, "%s: final leave", __func__);
 1923                         mld_final_leave(inm, mli);
 1924                         goto out_locked;
 1925                 }
 1926         } else {
 1927                 CTR1(KTR_MLD, "%s: filter set change", __func__);
 1928         }
 1929 
 1930         error = mld_handle_state_change(inm, mli);
 1931 
 1932 out_locked:
 1933         MLD_UNLOCK();
 1934         return (error);
 1935 }
 1936 
 1937 /*
 1938  * Perform the initial join for an MLD group.
 1939  *
 1940  * When joining a group:
 1941  *  If the group should have its MLD traffic suppressed, do nothing.
 1942  *  MLDv1 starts sending MLDv1 host membership reports.
 1943  *  MLDv2 will schedule an MLDv2 state-change report containing the
 1944  *  initial state of the membership.
 1945  *
 1946  * If the delay argument is non-zero, then we must delay sending the
 1947  * initial state change for delay ticks (in units of PR_FASTHZ).
 1948  */
 1949 static int
 1950 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
 1951     const int delay)
 1952 {
 1953         struct ifnet            *ifp;
 1954         struct ifqueue          *ifq;
 1955         int                      error, retval, syncstates;
 1956         int                      odelay;
 1957 #ifdef KTR
 1958         char                     ip6tbuf[INET6_ADDRSTRLEN];
 1959 #endif
 1960 
 1961         CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
 1962             __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 1963             inm->in6m_ifp, inm->in6m_ifp->if_xname);
 1964 
 1965         error = 0;
 1966         syncstates = 1;
 1967 
 1968         ifp = inm->in6m_ifp;
 1969 
 1970         IN6_MULTI_LOCK_ASSERT();
 1971         MLD_LOCK_ASSERT();
 1972 
 1973         KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
 1974 
 1975         /*
 1976          * Groups joined on loopback or marked as 'not reported',
 1977          * enter the MLD_SILENT_MEMBER state and
 1978          * are never reported in any protocol exchanges.
 1979          * All other groups enter the appropriate state machine
 1980          * for the version in use on this link.
 1981          * A link marked as MLIF_SILENT causes MLD to be completely
 1982          * disabled for the link.
 1983          */
 1984         if ((ifp->if_flags & IFF_LOOPBACK) ||
 1985             (mli->mli_flags & MLIF_SILENT) ||
 1986             !mld_is_addr_reported(&inm->in6m_addr)) {
 1987                 CTR1(KTR_MLD,
 1988 "%s: not kicking state machine for silent group", __func__);
 1989                 inm->in6m_state = MLD_SILENT_MEMBER;
 1990                 inm->in6m_timer = 0;
 1991         } else {
 1992                 /*
 1993                  * Deal with overlapping in_multi lifecycle.
 1994                  * If this group was LEAVING, then make sure
 1995                  * we drop the reference we picked up to keep the
 1996                  * group around for the final INCLUDE {} enqueue.
 1997                  */
 1998                 if (mli->mli_version == MLD_VERSION_2 &&
 1999                     inm->in6m_state == MLD_LEAVING_MEMBER)
 2000                         in6m_release_locked(inm);
 2001 
 2002                 inm->in6m_state = MLD_REPORTING_MEMBER;
 2003 
 2004                 switch (mli->mli_version) {
 2005                 case MLD_VERSION_1:
 2006                         /*
 2007                          * If a delay was provided, only use it if
 2008                          * it is greater than the delay normally
 2009                          * used for an MLDv1 state change report,
 2010                          * and delay sending the initial MLDv1 report
 2011                          * by not transitioning to the IDLE state.
 2012                          */
 2013                         odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
 2014                         if (delay) {
 2015                                 inm->in6m_timer = max(delay, odelay);
 2016                                 V_current_state_timers_running6 = 1;
 2017                         } else {
 2018                                 inm->in6m_state = MLD_IDLE_MEMBER;
 2019                                 error = mld_v1_transmit_report(inm,
 2020                                      MLD_LISTENER_REPORT);
 2021                                 if (error == 0) {
 2022                                         inm->in6m_timer = odelay;
 2023                                         V_current_state_timers_running6 = 1;
 2024                                 }
 2025                         }
 2026                         break;
 2027 
 2028                 case MLD_VERSION_2:
 2029                         /*
 2030                          * Defer update of T0 to T1, until the first copy
 2031                          * of the state change has been transmitted.
 2032                          */
 2033                         syncstates = 0;
 2034 
 2035                         /*
 2036                          * Immediately enqueue a State-Change Report for
 2037                          * this interface, freeing any previous reports.
 2038                          * Don't kick the timers if there is nothing to do,
 2039                          * or if an error occurred.
 2040                          */
 2041                         ifq = &inm->in6m_scq;
 2042                         _IF_DRAIN(ifq);
 2043                         retval = mld_v2_enqueue_group_record(ifq, inm, 1,
 2044                             0, 0, (mli->mli_flags & MLIF_USEALLOW));
 2045                         CTR2(KTR_MLD, "%s: enqueue record = %d",
 2046                             __func__, retval);
 2047                         if (retval <= 0) {
 2048                                 error = retval * -1;
 2049                                 break;
 2050                         }
 2051 
 2052                         /*
 2053                          * Schedule transmission of pending state-change
 2054                          * report up to RV times for this link. The timer
 2055                          * will fire at the next mld_fasttimo (~200ms),
 2056                          * giving us an opportunity to merge the reports.
 2057                          *
 2058                          * If a delay was provided to this function, only
 2059                          * use this delay if sooner than the existing one.
 2060                          */
 2061                         KASSERT(mli->mli_rv > 1,
 2062                            ("%s: invalid robustness %d", __func__,
 2063                             mli->mli_rv));
 2064                         inm->in6m_scrv = mli->mli_rv;
 2065                         if (delay) {
 2066                                 if (inm->in6m_sctimer > 1) {
 2067                                         inm->in6m_sctimer =
 2068                                             min(inm->in6m_sctimer, delay);
 2069                                 } else
 2070                                         inm->in6m_sctimer = delay;
 2071                         } else
 2072                                 inm->in6m_sctimer = 1;
 2073                         V_state_change_timers_running6 = 1;
 2074 
 2075                         error = 0;
 2076                         break;
 2077                 }
 2078         }
 2079 
 2080         /*
 2081          * Only update the T0 state if state change is atomic,
 2082          * i.e. we don't need to wait for a timer to fire before we
 2083          * can consider the state change to have been communicated.
 2084          */
 2085         if (syncstates) {
 2086                 in6m_commit(inm);
 2087                 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
 2088                     ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2089                     inm->in6m_ifp->if_xname);
 2090         }
 2091 
 2092         return (error);
 2093 }
 2094 
 2095 /*
 2096  * Issue an intermediate state change during the life-cycle.
 2097  */
 2098 static int
 2099 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli)
 2100 {
 2101         struct ifnet            *ifp;
 2102         int                      retval;
 2103 #ifdef KTR
 2104         char                     ip6tbuf[INET6_ADDRSTRLEN];
 2105 #endif
 2106 
 2107         CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
 2108             __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2109             inm->in6m_ifp, inm->in6m_ifp->if_xname);
 2110 
 2111         ifp = inm->in6m_ifp;
 2112 
 2113         IN6_MULTI_LOCK_ASSERT();
 2114         MLD_LOCK_ASSERT();
 2115 
 2116         KASSERT(mli && mli->mli_ifp == ifp,
 2117             ("%s: inconsistent ifp", __func__));
 2118 
 2119         if ((ifp->if_flags & IFF_LOOPBACK) ||
 2120             (mli->mli_flags & MLIF_SILENT) ||
 2121             !mld_is_addr_reported(&inm->in6m_addr) ||
 2122             (mli->mli_version != MLD_VERSION_2)) {
 2123                 if (!mld_is_addr_reported(&inm->in6m_addr)) {
 2124                         CTR1(KTR_MLD,
 2125 "%s: not kicking state machine for silent group", __func__);
 2126                 }
 2127                 CTR1(KTR_MLD, "%s: nothing to do", __func__);
 2128                 in6m_commit(inm);
 2129                 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
 2130                     ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2131                     inm->in6m_ifp->if_xname);
 2132                 return (0);
 2133         }
 2134 
 2135         _IF_DRAIN(&inm->in6m_scq);
 2136 
 2137         retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
 2138             (mli->mli_flags & MLIF_USEALLOW));
 2139         CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
 2140         if (retval <= 0)
 2141                 return (-retval);
 2142 
 2143         /*
 2144          * If record(s) were enqueued, start the state-change
 2145          * report timer for this group.
 2146          */
 2147         inm->in6m_scrv = mli->mli_rv;
 2148         inm->in6m_sctimer = 1;
 2149         V_state_change_timers_running6 = 1;
 2150 
 2151         return (0);
 2152 }
 2153 
 2154 /*
 2155  * Perform the final leave for a multicast address.
 2156  *
 2157  * When leaving a group:
 2158  *  MLDv1 sends a DONE message, if and only if we are the reporter.
 2159  *  MLDv2 enqueues a state-change report containing a transition
 2160  *  to INCLUDE {} for immediate transmission.
 2161  */
 2162 static void
 2163 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli)
 2164 {
 2165         int syncstates;
 2166 #ifdef KTR
 2167         char ip6tbuf[INET6_ADDRSTRLEN];
 2168 #endif
 2169 
 2170         syncstates = 1;
 2171 
 2172         CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
 2173             __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2174             inm->in6m_ifp, inm->in6m_ifp->if_xname);
 2175 
 2176         IN6_MULTI_LOCK_ASSERT();
 2177         MLD_LOCK_ASSERT();
 2178 
 2179         switch (inm->in6m_state) {
 2180         case MLD_NOT_MEMBER:
 2181         case MLD_SILENT_MEMBER:
 2182         case MLD_LEAVING_MEMBER:
 2183                 /* Already leaving or left; do nothing. */
 2184                 CTR1(KTR_MLD,
 2185 "%s: not kicking state machine for silent group", __func__);
 2186                 break;
 2187         case MLD_REPORTING_MEMBER:
 2188         case MLD_IDLE_MEMBER:
 2189         case MLD_G_QUERY_PENDING_MEMBER:
 2190         case MLD_SG_QUERY_PENDING_MEMBER:
 2191                 if (mli->mli_version == MLD_VERSION_1) {
 2192 #ifdef INVARIANTS
 2193                         if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
 2194                             inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
 2195                         panic("%s: MLDv2 state reached, not MLDv2 mode",
 2196                              __func__);
 2197 #endif
 2198                         mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
 2199                         inm->in6m_state = MLD_NOT_MEMBER;
 2200                         V_current_state_timers_running6 = 1;
 2201                 } else if (mli->mli_version == MLD_VERSION_2) {
 2202                         /*
 2203                          * Stop group timer and all pending reports.
 2204                          * Immediately enqueue a state-change report
 2205                          * TO_IN {} to be sent on the next fast timeout,
 2206                          * giving us an opportunity to merge reports.
 2207                          */
 2208                         _IF_DRAIN(&inm->in6m_scq);
 2209                         inm->in6m_timer = 0;
 2210                         inm->in6m_scrv = mli->mli_rv;
 2211                         CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
 2212                             "pending retransmissions.", __func__,
 2213                             ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2214                             inm->in6m_ifp->if_xname, inm->in6m_scrv);
 2215                         if (inm->in6m_scrv == 0) {
 2216                                 inm->in6m_state = MLD_NOT_MEMBER;
 2217                                 inm->in6m_sctimer = 0;
 2218                         } else {
 2219                                 int retval;
 2220 
 2221                                 in6m_acquire_locked(inm);
 2222 
 2223                                 retval = mld_v2_enqueue_group_record(
 2224                                     &inm->in6m_scq, inm, 1, 0, 0,
 2225                                     (mli->mli_flags & MLIF_USEALLOW));
 2226                                 KASSERT(retval != 0,
 2227                                     ("%s: enqueue record = %d", __func__,
 2228                                      retval));
 2229 
 2230                                 inm->in6m_state = MLD_LEAVING_MEMBER;
 2231                                 inm->in6m_sctimer = 1;
 2232                                 V_state_change_timers_running6 = 1;
 2233                                 syncstates = 0;
 2234                         }
 2235                         break;
 2236                 }
 2237                 break;
 2238         case MLD_LAZY_MEMBER:
 2239         case MLD_SLEEPING_MEMBER:
 2240         case MLD_AWAKENING_MEMBER:
 2241                 /* Our reports are suppressed; do nothing. */
 2242                 break;
 2243         }
 2244 
 2245         if (syncstates) {
 2246                 in6m_commit(inm);
 2247                 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
 2248                     ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2249                     inm->in6m_ifp->if_xname);
 2250                 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
 2251                 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
 2252                     __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname);
 2253         }
 2254 }
 2255 
 2256 /*
 2257  * Enqueue an MLDv2 group record to the given output queue.
 2258  *
 2259  * If is_state_change is zero, a current-state record is appended.
 2260  * If is_state_change is non-zero, a state-change report is appended.
 2261  *
 2262  * If is_group_query is non-zero, an mbuf packet chain is allocated.
 2263  * If is_group_query is zero, and if there is a packet with free space
 2264  * at the tail of the queue, it will be appended to providing there
 2265  * is enough free space.
 2266  * Otherwise a new mbuf packet chain is allocated.
 2267  *
 2268  * If is_source_query is non-zero, each source is checked to see if
 2269  * it was recorded for a Group-Source query, and will be omitted if
 2270  * it is not both in-mode and recorded.
 2271  *
 2272  * If use_block_allow is non-zero, state change reports for initial join
 2273  * and final leave, on an inclusive mode group with a source list, will be
 2274  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
 2275  *
 2276  * The function will attempt to allocate leading space in the packet
 2277  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
 2278  *
 2279  * If successful the size of all data appended to the queue is returned,
 2280  * otherwise an error code less than zero is returned, or zero if
 2281  * no record(s) were appended.
 2282  */
 2283 static int
 2284 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
 2285     const int is_state_change, const int is_group_query,
 2286     const int is_source_query, const int use_block_allow)
 2287 {
 2288         struct mldv2_record      mr;
 2289         struct mldv2_record     *pmr;
 2290         struct ifnet            *ifp;
 2291         struct ip6_msource      *ims, *nims;
 2292         struct mbuf             *m0, *m, *md;
 2293         int                      error, is_filter_list_change;
 2294         int                      minrec0len, m0srcs, msrcs, nbytes, off;
 2295         int                      record_has_sources;
 2296         int                      now;
 2297         int                      type;
 2298         uint8_t                  mode;
 2299 #ifdef KTR
 2300         char                     ip6tbuf[INET6_ADDRSTRLEN];
 2301 #endif
 2302 
 2303         IN6_MULTI_LOCK_ASSERT();
 2304 
 2305         error = 0;
 2306         ifp = inm->in6m_ifp;
 2307         is_filter_list_change = 0;
 2308         m = NULL;
 2309         m0 = NULL;
 2310         m0srcs = 0;
 2311         msrcs = 0;
 2312         nbytes = 0;
 2313         nims = NULL;
 2314         record_has_sources = 1;
 2315         pmr = NULL;
 2316         type = MLD_DO_NOTHING;
 2317         mode = inm->in6m_st[1].iss_fmode;
 2318 
 2319         /*
 2320          * If we did not transition out of ASM mode during t0->t1,
 2321          * and there are no source nodes to process, we can skip
 2322          * the generation of source records.
 2323          */
 2324         if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
 2325             inm->in6m_nsrc == 0)
 2326                 record_has_sources = 0;
 2327 
 2328         if (is_state_change) {
 2329                 /*
 2330                  * Queue a state change record.
 2331                  * If the mode did not change, and there are non-ASM
 2332                  * listeners or source filters present,
 2333                  * we potentially need to issue two records for the group.
 2334                  * If there are ASM listeners, and there was no filter
 2335                  * mode transition of any kind, do nothing.
 2336                  *
 2337                  * If we are transitioning to MCAST_UNDEFINED, we need
 2338                  * not send any sources. A transition to/from this state is
 2339                  * considered inclusive with some special treatment.
 2340                  *
 2341                  * If we are rewriting initial joins/leaves to use
 2342                  * ALLOW/BLOCK, and the group's membership is inclusive,
 2343                  * we need to send sources in all cases.
 2344                  */
 2345                 if (mode != inm->in6m_st[0].iss_fmode) {
 2346                         if (mode == MCAST_EXCLUDE) {
 2347                                 CTR1(KTR_MLD, "%s: change to EXCLUDE",
 2348                                     __func__);
 2349                                 type = MLD_CHANGE_TO_EXCLUDE_MODE;
 2350                         } else {
 2351                                 CTR1(KTR_MLD, "%s: change to INCLUDE",
 2352                                     __func__);
 2353                                 if (use_block_allow) {
 2354                                         /*
 2355                                          * XXX
 2356                                          * Here we're interested in state
 2357                                          * edges either direction between
 2358                                          * MCAST_UNDEFINED and MCAST_INCLUDE.
 2359                                          * Perhaps we should just check
 2360                                          * the group state, rather than
 2361                                          * the filter mode.
 2362                                          */
 2363                                         if (mode == MCAST_UNDEFINED) {
 2364                                                 type = MLD_BLOCK_OLD_SOURCES;
 2365                                         } else {
 2366                                                 type = MLD_ALLOW_NEW_SOURCES;
 2367                                         }
 2368                                 } else {
 2369                                         type = MLD_CHANGE_TO_INCLUDE_MODE;
 2370                                         if (mode == MCAST_UNDEFINED)
 2371                                                 record_has_sources = 0;
 2372                                 }
 2373                         }
 2374                 } else {
 2375                         if (record_has_sources) {
 2376                                 is_filter_list_change = 1;
 2377                         } else {
 2378                                 type = MLD_DO_NOTHING;
 2379                         }
 2380                 }
 2381         } else {
 2382                 /*
 2383                  * Queue a current state record.
 2384                  */
 2385                 if (mode == MCAST_EXCLUDE) {
 2386                         type = MLD_MODE_IS_EXCLUDE;
 2387                 } else if (mode == MCAST_INCLUDE) {
 2388                         type = MLD_MODE_IS_INCLUDE;
 2389                         KASSERT(inm->in6m_st[1].iss_asm == 0,
 2390                             ("%s: inm %p is INCLUDE but ASM count is %d",
 2391                              __func__, inm, inm->in6m_st[1].iss_asm));
 2392                 }
 2393         }
 2394 
 2395         /*
 2396          * Generate the filter list changes using a separate function.
 2397          */
 2398         if (is_filter_list_change)
 2399                 return (mld_v2_enqueue_filter_change(ifq, inm));
 2400 
 2401         if (type == MLD_DO_NOTHING) {
 2402                 CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
 2403                     __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2404                     inm->in6m_ifp->if_xname);
 2405                 return (0);
 2406         }
 2407 
 2408         /*
 2409          * If any sources are present, we must be able to fit at least
 2410          * one in the trailing space of the tail packet's mbuf,
 2411          * ideally more.
 2412          */
 2413         minrec0len = sizeof(struct mldv2_record);
 2414         if (record_has_sources)
 2415                 minrec0len += sizeof(struct in6_addr);
 2416 
 2417         CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
 2418             mld_rec_type_to_str(type),
 2419             ip6_sprintf(ip6tbuf, &inm->in6m_addr),
 2420             inm->in6m_ifp->if_xname);
 2421 
 2422         /*
 2423          * Check if we have a packet in the tail of the queue for this
 2424          * group into which the first group record for this group will fit.
 2425          * Otherwise allocate a new packet.
 2426          * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
 2427          * Note: Group records for G/GSR query responses MUST be sent
 2428          * in their own packet.
 2429          */
 2430         m0 = ifq->ifq_tail;
 2431         if (!is_group_query &&
 2432             m0 != NULL &&
 2433             (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
 2434             (m0->m_pkthdr.len + minrec0len) <
 2435              (ifp->if_mtu - MLD_MTUSPACE)) {
 2436                 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
 2437                             sizeof(struct mldv2_record)) /
 2438                             sizeof(struct in6_addr);
 2439                 m = m0;
 2440                 CTR1(KTR_MLD, "%s: use existing packet", __func__);
 2441         } else {
 2442                 if (_IF_QFULL(ifq)) {
 2443                         CTR1(KTR_MLD, "%s: outbound queue full", __func__);
 2444                         return (-ENOMEM);
 2445                 }
 2446                 m = NULL;
 2447                 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
 2448                     sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
 2449                 if (!is_state_change && !is_group_query)
 2450                         m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 2451                 if (m == NULL)
 2452                         m = m_gethdr(M_NOWAIT, MT_DATA);
 2453                 if (m == NULL)
 2454                         return (-ENOMEM);
 2455 
 2456                 mld_save_context(m, ifp);
 2457 
 2458                 CTR1(KTR_MLD, "%s: allocated first packet", __func__);
 2459         }
 2460 
 2461         /*
 2462          * Append group record.
 2463          * If we have sources, we don't know how many yet.
 2464          */
 2465         mr.mr_type = type;
 2466         mr.mr_datalen = 0;
 2467         mr.mr_numsrc = 0;
 2468         mr.mr_addr = inm->in6m_addr;
 2469         in6_clearscope(&mr.mr_addr);
 2470         if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
 2471                 if (m != m0)
 2472                         m_freem(m);
 2473                 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
 2474                 return (-ENOMEM);
 2475         }
 2476         nbytes += sizeof(struct mldv2_record);
 2477 
 2478         /*
 2479          * Append as many sources as will fit in the first packet.
 2480          * If we are appending to a new packet, the chain allocation
 2481          * may potentially use clusters; use m_getptr() in this case.
 2482          * If we are appending to an existing packet, we need to obtain
 2483          * a pointer to the group record after m_append(), in case a new
 2484          * mbuf was allocated.
 2485          *
 2486          * Only append sources which are in-mode at t1. If we are
 2487          * transitioning to MCAST_UNDEFINED state on the group, and
 2488          * use_block_allow is zero, do not include source entries.
 2489          * Otherwise, we need to include this source in the report.
 2490          *
 2491          * Only report recorded sources in our filter set when responding
 2492          * to a group-source query.
 2493          */
 2494         if (record_has_sources) {
 2495                 if (m == m0) {
 2496                         md = m_last(m);
 2497                         pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
 2498                             md->m_len - nbytes);
 2499                 } else {
 2500                         md = m_getptr(m, 0, &off);
 2501                         pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
 2502                             off);
 2503                 }
 2504                 msrcs = 0;
 2505                 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
 2506                     nims) {
 2507                         CTR2(KTR_MLD, "%s: visit node %s", __func__,
 2508                             ip6_sprintf(ip6tbuf, &ims->im6s_addr));
 2509                         now = im6s_get_mode(inm, ims, 1);
 2510                         CTR2(KTR_MLD, "%s: node is %d", __func__, now);
 2511                         if ((now != mode) ||
 2512                             (now == mode &&
 2513                              (!use_block_allow && mode == MCAST_UNDEFINED))) {
 2514                                 CTR1(KTR_MLD, "%s: skip node", __func__);
 2515                                 continue;
 2516                         }
 2517                         if (is_source_query && ims->im6s_stp == 0) {
 2518                                 CTR1(KTR_MLD, "%s: skip unrecorded node",
 2519                                     __func__);
 2520                                 continue;
 2521                         }
 2522                         CTR1(KTR_MLD, "%s: append node", __func__);
 2523                         if (!m_append(m, sizeof(struct in6_addr),
 2524                             (void *)&ims->im6s_addr)) {
 2525                                 if (m != m0)
 2526                                         m_freem(m);
 2527                                 CTR1(KTR_MLD, "%s: m_append() failed.",
 2528                                     __func__);
 2529                                 return (-ENOMEM);
 2530                         }
 2531                         nbytes += sizeof(struct in6_addr);
 2532                         ++msrcs;
 2533                         if (msrcs == m0srcs)
 2534                                 break;
 2535                 }
 2536                 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
 2537                     msrcs);
 2538                 pmr->mr_numsrc = htons(msrcs);
 2539                 nbytes += (msrcs * sizeof(struct in6_addr));
 2540         }
 2541 
 2542         if (is_source_query && msrcs == 0) {
 2543                 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
 2544                 if (m != m0)
 2545                         m_freem(m);
 2546                 return (0);
 2547         }
 2548 
 2549         /*
 2550          * We are good to go with first packet.
 2551          */
 2552         if (m != m0) {
 2553                 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
 2554                 m->m_pkthdr.PH_vt.vt_nrecs = 1;
 2555                 _IF_ENQUEUE(ifq, m);
 2556         } else
 2557                 m->m_pkthdr.PH_vt.vt_nrecs++;
 2558 
 2559         /*
 2560          * No further work needed if no source list in packet(s).
 2561          */
 2562         if (!record_has_sources)
 2563                 return (nbytes);
 2564 
 2565         /*
 2566          * Whilst sources remain to be announced, we need to allocate
 2567          * a new packet and fill out as many sources as will fit.
 2568          * Always try for a cluster first.
 2569          */
 2570         while (nims != NULL) {
 2571                 if (_IF_QFULL(ifq)) {
 2572                         CTR1(KTR_MLD, "%s: outbound queue full", __func__);
 2573                         return (-ENOMEM);
 2574                 }
 2575                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 2576                 if (m == NULL)
 2577                         m = m_gethdr(M_NOWAIT, MT_DATA);
 2578                 if (m == NULL)
 2579                         return (-ENOMEM);
 2580                 mld_save_context(m, ifp);
 2581                 md = m_getptr(m, 0, &off);
 2582                 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
 2583                 CTR1(KTR_MLD, "%s: allocated next packet", __func__);
 2584 
 2585                 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
 2586                         if (m != m0)
 2587                                 m_freem(m);
 2588                         CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
 2589                         return (-ENOMEM);
 2590                 }
 2591                 m->m_pkthdr.PH_vt.vt_nrecs = 1;
 2592                 nbytes += sizeof(struct mldv2_record);
 2593 
 2594                 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
 2595                     sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
 2596 
 2597                 msrcs = 0;
 2598                 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
 2599                         CTR2(KTR_MLD, "%s: visit node %s",
 2600                             __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
 2601                         now = im6s_get_mode(inm, ims, 1);
 2602                         if ((now != mode) ||
 2603                             (now == mode &&
 2604                              (!use_block_allow && mode == MCAST_UNDEFINED))) {
 2605                                 CTR1(KTR_MLD, "%s: skip node", __func__);
 2606                                 continue;
 2607                         }
 2608                         if (is_source_query && ims->im6s_stp == 0) {
 2609                                 CTR1(KTR_MLD, "%s: skip unrecorded node",
 2610                                     __func__);
 2611                                 continue;
 2612                         }
 2613                         CTR1(KTR_MLD, "%s: append node", __func__);
 2614                         if (!m_append(m, sizeof(struct in6_addr),
 2615                             (void *)&ims->im6s_addr)) {
 2616                                 if (m != m0)
 2617                                         m_freem(m);
 2618                                 CTR1(KTR_MLD, "%s: m_append() failed.",
 2619                                     __func__);
 2620                                 return (-ENOMEM);
 2621                         }
 2622                         ++msrcs;
 2623                         if (msrcs == m0srcs)
 2624                                 break;
 2625                 }
 2626                 pmr->mr_numsrc = htons(msrcs);
 2627                 nbytes += (msrcs * sizeof(struct in6_addr));
 2628 
 2629                 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
 2630                 _IF_ENQUEUE(ifq, m);
 2631         }
 2632 
 2633         return (nbytes);
 2634 }
 2635 
 2636 /*
 2637  * Type used to mark record pass completion.
 2638  * We exploit the fact we can cast to this easily from the
 2639  * current filter modes on each ip_msource node.
 2640  */
 2641 typedef enum {
 2642         REC_NONE = 0x00,        /* MCAST_UNDEFINED */
 2643         REC_ALLOW = 0x01,       /* MCAST_INCLUDE */
 2644         REC_BLOCK = 0x02,       /* MCAST_EXCLUDE */
 2645         REC_FULL = REC_ALLOW | REC_BLOCK
 2646 } rectype_t;
 2647 
 2648 /*
 2649  * Enqueue an MLDv2 filter list change to the given output queue.
 2650  *
 2651  * Source list filter state is held in an RB-tree. When the filter list
 2652  * for a group is changed without changing its mode, we need to compute
 2653  * the deltas between T0 and T1 for each source in the filter set,
 2654  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
 2655  *
 2656  * As we may potentially queue two record types, and the entire R-B tree
 2657  * needs to be walked at once, we break this out into its own function
 2658  * so we can generate a tightly packed queue of packets.
 2659  *
 2660  * XXX This could be written to only use one tree walk, although that makes
 2661  * serializing into the mbuf chains a bit harder. For now we do two walks
 2662  * which makes things easier on us, and it may or may not be harder on
 2663  * the L2 cache.
 2664  *
 2665  * If successful the size of all data appended to the queue is returned,
 2666  * otherwise an error code less than zero is returned, or zero if
 2667  * no record(s) were appended.
 2668  */
 2669 static int
 2670 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
 2671 {
 2672         static const int MINRECLEN =
 2673             sizeof(struct mldv2_record) + sizeof(struct in6_addr);
 2674         struct ifnet            *ifp;
 2675         struct mldv2_record      mr;
 2676         struct mldv2_record     *pmr;
 2677         struct ip6_msource      *ims, *nims;
 2678         struct mbuf             *m, *m0, *md;
 2679         int                      m0srcs, nbytes, npbytes, off, rsrcs, schanged;
 2680         int                      nallow, nblock;
 2681         uint8_t                  mode, now, then;
 2682         rectype_t                crt, drt, nrt;
 2683 #ifdef KTR
 2684         char                     ip6tbuf[INET6_ADDRSTRLEN];
 2685 #endif
 2686 
 2687         IN6_MULTI_LOCK_ASSERT();
 2688 
 2689         if (inm->in6m_nsrc == 0 ||
 2690             (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
 2691                 return (0);
 2692 
 2693         ifp = inm->in6m_ifp;                    /* interface */
 2694         mode = inm->in6m_st[1].iss_fmode;       /* filter mode at t1 */
 2695         crt = REC_NONE; /* current group record type */
 2696         drt = REC_NONE; /* mask of completed group record types */
 2697         nrt = REC_NONE; /* record type for current node */
 2698         m0srcs = 0;     /* # source which will fit in current mbuf chain */
 2699         npbytes = 0;    /* # of bytes appended this packet */
 2700         nbytes = 0;     /* # of bytes appended to group's state-change queue */
 2701         rsrcs = 0;      /* # sources encoded in current record */
 2702         schanged = 0;   /* # nodes encoded in overall filter change */
 2703         nallow = 0;     /* # of source entries in ALLOW_NEW */
 2704         nblock = 0;     /* # of source entries in BLOCK_OLD */
 2705         nims = NULL;    /* next tree node pointer */
 2706 
 2707         /*
 2708          * For each possible filter record mode.
 2709          * The first kind of source we encounter tells us which
 2710          * is the first kind of record we start appending.
 2711          * If a node transitioned to UNDEFINED at t1, its mode is treated
 2712          * as the inverse of the group's filter mode.
 2713          */
 2714         while (drt != REC_FULL) {
 2715                 do {
 2716                         m0 = ifq->ifq_tail;
 2717                         if (m0 != NULL &&
 2718                             (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
 2719                              MLD_V2_REPORT_MAXRECS) &&
 2720                             (m0->m_pkthdr.len + MINRECLEN) <
 2721                              (ifp->if_mtu - MLD_MTUSPACE)) {
 2722                                 m = m0;
 2723                                 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
 2724                                             sizeof(struct mldv2_record)) /
 2725                                             sizeof(struct in6_addr);
 2726                                 CTR1(KTR_MLD,
 2727                                     "%s: use previous packet", __func__);
 2728                         } else {
 2729                                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 2730                                 if (m == NULL)
 2731                                         m = m_gethdr(M_NOWAIT, MT_DATA);
 2732                                 if (m == NULL) {
 2733                                         CTR1(KTR_MLD,
 2734                                             "%s: m_get*() failed", __func__);
 2735                                         return (-ENOMEM);
 2736                                 }
 2737                                 m->m_pkthdr.PH_vt.vt_nrecs = 0;
 2738                                 mld_save_context(m, ifp);
 2739                                 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
 2740                                     sizeof(struct mldv2_record)) /
 2741                                     sizeof(struct in6_addr);
 2742                                 npbytes = 0;
 2743                                 CTR1(KTR_MLD,
 2744                                     "%s: allocated new packet", __func__);
 2745                         }
 2746                         /*
 2747                          * Append the MLD group record header to the
 2748                          * current packet's data area.
 2749                          * Recalculate pointer to free space for next
 2750                          * group record, in case m_append() allocated
 2751                          * a new mbuf or cluster.
 2752                          */
 2753                         memset(&mr, 0, sizeof(mr));
 2754                         mr.mr_addr = inm->in6m_addr;
 2755                         in6_clearscope(&mr.mr_addr);
 2756                         if (!m_append(m, sizeof(mr), (void *)&mr)) {
 2757                                 if (m != m0)
 2758                                         m_freem(m);
 2759                                 CTR1(KTR_MLD,
 2760                                     "%s: m_append() failed", __func__);
 2761                                 return (-ENOMEM);
 2762                         }
 2763                         npbytes += sizeof(struct mldv2_record);
 2764                         if (m != m0) {
 2765                                 /* new packet; offset in chain */
 2766                                 md = m_getptr(m, npbytes -
 2767                                     sizeof(struct mldv2_record), &off);
 2768                                 pmr = (struct mldv2_record *)(mtod(md,
 2769                                     uint8_t *) + off);
 2770                         } else {
 2771                                 /* current packet; offset from last append */
 2772                                 md = m_last(m);
 2773                                 pmr = (struct mldv2_record *)(mtod(md,
 2774                                     uint8_t *) + md->m_len -
 2775                                     sizeof(struct mldv2_record));
 2776                         }
 2777                         /*
 2778                          * Begin walking the tree for this record type
 2779                          * pass, or continue from where we left off
 2780                          * previously if we had to allocate a new packet.
 2781                          * Only report deltas in-mode at t1.
 2782                          * We need not report included sources as allowed
 2783                          * if we are in inclusive mode on the group,
 2784                          * however the converse is not true.
 2785                          */
 2786                         rsrcs = 0;
 2787                         if (nims == NULL) {
 2788                                 nims = RB_MIN(ip6_msource_tree,
 2789                                     &inm->in6m_srcs);
 2790                         }
 2791                         RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
 2792                                 CTR2(KTR_MLD, "%s: visit node %s", __func__,
 2793                                     ip6_sprintf(ip6tbuf, &ims->im6s_addr));
 2794                                 now = im6s_get_mode(inm, ims, 1);
 2795                                 then = im6s_get_mode(inm, ims, 0);
 2796                                 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
 2797                                     __func__, then, now);
 2798                                 if (now == then) {
 2799                                         CTR1(KTR_MLD,
 2800                                             "%s: skip unchanged", __func__);
 2801                                         continue;
 2802                                 }
 2803                                 if (mode == MCAST_EXCLUDE &&
 2804                                     now == MCAST_INCLUDE) {
 2805                                         CTR1(KTR_MLD,
 2806                                             "%s: skip IN src on EX group",
 2807                                             __func__);
 2808                                         continue;
 2809                                 }
 2810                                 nrt = (rectype_t)now;
 2811                                 if (nrt == REC_NONE)
 2812                                         nrt = (rectype_t)(~mode & REC_FULL);
 2813                                 if (schanged++ == 0) {
 2814                                         crt = nrt;
 2815                                 } else if (crt != nrt)
 2816                                         continue;
 2817                                 if (!m_append(m, sizeof(struct in6_addr),
 2818                                     (void *)&ims->im6s_addr)) {
 2819                                         if (m != m0)
 2820                                                 m_freem(m);
 2821                                         CTR1(KTR_MLD,
 2822                                             "%s: m_append() failed", __func__);
 2823                                         return (-ENOMEM);
 2824                                 }
 2825                                 nallow += !!(crt == REC_ALLOW);
 2826                                 nblock += !!(crt == REC_BLOCK);
 2827                                 if (++rsrcs == m0srcs)
 2828                                         break;
 2829                         }
 2830                         /*
 2831                          * If we did not append any tree nodes on this
 2832                          * pass, back out of allocations.
 2833                          */
 2834                         if (rsrcs == 0) {
 2835                                 npbytes -= sizeof(struct mldv2_record);
 2836                                 if (m != m0) {
 2837                                         CTR1(KTR_MLD,
 2838                                             "%s: m_free(m)", __func__);
 2839                                         m_freem(m);
 2840                                 } else {
 2841                                         CTR1(KTR_MLD,
 2842                                             "%s: m_adj(m, -mr)", __func__);
 2843                                         m_adj(m, -((int)sizeof(
 2844                                             struct mldv2_record)));
 2845                                 }
 2846                                 continue;
 2847                         }
 2848                         npbytes += (rsrcs * sizeof(struct in6_addr));
 2849                         if (crt == REC_ALLOW)
 2850                                 pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
 2851                         else if (crt == REC_BLOCK)
 2852                                 pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
 2853                         pmr->mr_numsrc = htons(rsrcs);
 2854                         /*
 2855                          * Count the new group record, and enqueue this
 2856                          * packet if it wasn't already queued.
 2857                          */
 2858                         m->m_pkthdr.PH_vt.vt_nrecs++;
 2859                         if (m != m0)
 2860                                 _IF_ENQUEUE(ifq, m);
 2861                         nbytes += npbytes;
 2862                 } while (nims != NULL);
 2863                 drt |= crt;
 2864                 crt = (~crt & REC_FULL);
 2865         }
 2866 
 2867         CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
 2868             nallow, nblock);
 2869 
 2870         return (nbytes);
 2871 }
 2872 
 2873 static int
 2874 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
 2875 {
 2876         struct ifqueue  *gq;
 2877         struct mbuf     *m;             /* pending state-change */
 2878         struct mbuf     *m0;            /* copy of pending state-change */
 2879         struct mbuf     *mt;            /* last state-change in packet */
 2880         int              docopy, domerge;
 2881         u_int            recslen;
 2882 
 2883         docopy = 0;
 2884         domerge = 0;
 2885         recslen = 0;
 2886 
 2887         IN6_MULTI_LOCK_ASSERT();
 2888         MLD_LOCK_ASSERT();
 2889 
 2890         /*
 2891          * If there are further pending retransmissions, make a writable
 2892          * copy of each queued state-change message before merging.
 2893          */
 2894         if (inm->in6m_scrv > 0)
 2895                 docopy = 1;
 2896 
 2897         gq = &inm->in6m_scq;
 2898 #ifdef KTR
 2899         if (gq->ifq_head == NULL) {
 2900                 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
 2901                     __func__, inm);
 2902         }
 2903 #endif
 2904 
 2905         m = gq->ifq_head;
 2906         while (m != NULL) {
 2907                 /*
 2908                  * Only merge the report into the current packet if
 2909                  * there is sufficient space to do so; an MLDv2 report
 2910                  * packet may only contain 65,535 group records.
 2911                  * Always use a simple mbuf chain concatentation to do this,
 2912                  * as large state changes for single groups may have
 2913                  * allocated clusters.
 2914                  */
 2915                 domerge = 0;
 2916                 mt = ifscq->ifq_tail;
 2917                 if (mt != NULL) {
 2918                         recslen = m_length(m, NULL);
 2919 
 2920                         if ((mt->m_pkthdr.PH_vt.vt_nrecs +
 2921                             m->m_pkthdr.PH_vt.vt_nrecs <=
 2922                             MLD_V2_REPORT_MAXRECS) &&
 2923                             (mt->m_pkthdr.len + recslen <=
 2924                             (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
 2925                                 domerge = 1;
 2926                 }
 2927 
 2928                 if (!domerge && _IF_QFULL(gq)) {
 2929                         CTR2(KTR_MLD,
 2930                             "%s: outbound queue full, skipping whole packet %p",
 2931                             __func__, m);
 2932                         mt = m->m_nextpkt;
 2933                         if (!docopy)
 2934                                 m_freem(m);
 2935                         m = mt;
 2936                         continue;
 2937                 }
 2938 
 2939                 if (!docopy) {
 2940                         CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
 2941                         _IF_DEQUEUE(gq, m0);
 2942                         m = m0->m_nextpkt;
 2943                 } else {
 2944                         CTR2(KTR_MLD, "%s: copying %p", __func__, m);
 2945                         m0 = m_dup(m, M_NOWAIT);
 2946                         if (m0 == NULL)
 2947                                 return (ENOMEM);
 2948                         m0->m_nextpkt = NULL;
 2949                         m = m->m_nextpkt;
 2950                 }
 2951 
 2952                 if (!domerge) {
 2953                         CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)",
 2954                             __func__, m0, ifscq);
 2955                         _IF_ENQUEUE(ifscq, m0);
 2956                 } else {
 2957                         struct mbuf *mtl;       /* last mbuf of packet mt */
 2958 
 2959                         CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
 2960                             __func__, m0, mt);
 2961 
 2962                         mtl = m_last(mt);
 2963                         m0->m_flags &= ~M_PKTHDR;
 2964                         mt->m_pkthdr.len += recslen;
 2965                         mt->m_pkthdr.PH_vt.vt_nrecs +=
 2966                             m0->m_pkthdr.PH_vt.vt_nrecs;
 2967 
 2968                         mtl->m_next = m0;
 2969                 }
 2970         }
 2971 
 2972         return (0);
 2973 }
 2974 
 2975 /*
 2976  * Respond to a pending MLDv2 General Query.
 2977  */
 2978 static void
 2979 mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
 2980 {
 2981         struct ifmultiaddr      *ifma;
 2982         struct ifnet            *ifp;
 2983         struct in6_multi        *inm;
 2984         int                      retval;
 2985 
 2986         IN6_MULTI_LOCK_ASSERT();
 2987         MLD_LOCK_ASSERT();
 2988 
 2989         KASSERT(mli->mli_version == MLD_VERSION_2,
 2990             ("%s: called when version %d", __func__, mli->mli_version));
 2991 
 2992         ifp = mli->mli_ifp;
 2993 
 2994         IF_ADDR_RLOCK(ifp);
 2995         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
 2996                 if (ifma->ifma_addr->sa_family != AF_INET6 ||
 2997                     ifma->ifma_protospec == NULL)
 2998                         continue;
 2999 
 3000                 inm = (struct in6_multi *)ifma->ifma_protospec;
 3001                 KASSERT(ifp == inm->in6m_ifp,
 3002                     ("%s: inconsistent ifp", __func__));
 3003 
 3004                 switch (inm->in6m_state) {
 3005                 case MLD_NOT_MEMBER:
 3006                 case MLD_SILENT_MEMBER:
 3007                         break;
 3008                 case MLD_REPORTING_MEMBER:
 3009                 case MLD_IDLE_MEMBER:
 3010                 case MLD_LAZY_MEMBER:
 3011                 case MLD_SLEEPING_MEMBER:
 3012                 case MLD_AWAKENING_MEMBER:
 3013                         inm->in6m_state = MLD_REPORTING_MEMBER;
 3014                         retval = mld_v2_enqueue_group_record(&mli->mli_gq,
 3015                             inm, 0, 0, 0, 0);
 3016                         CTR2(KTR_MLD, "%s: enqueue record = %d",
 3017                             __func__, retval);
 3018                         break;
 3019                 case MLD_G_QUERY_PENDING_MEMBER:
 3020                 case MLD_SG_QUERY_PENDING_MEMBER:
 3021                 case MLD_LEAVING_MEMBER:
 3022                         break;
 3023                 }
 3024         }
 3025         IF_ADDR_RUNLOCK(ifp);
 3026 
 3027         mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
 3028 
 3029         /*
 3030          * Slew transmission of bursts over 500ms intervals.
 3031          */
 3032         if (mli->mli_gq.ifq_head != NULL) {
 3033                 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
 3034                     MLD_RESPONSE_BURST_INTERVAL);
 3035                 V_interface_timers_running6 = 1;
 3036         }
 3037 }
 3038 
 3039 /*
 3040  * Transmit the next pending message in the output queue.
 3041  *
 3042  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
 3043  * MRT: Nothing needs to be done, as MLD traffic is always local to
 3044  * a link and uses a link-scope multicast address.
 3045  */
 3046 static void
 3047 mld_dispatch_packet(struct mbuf *m)
 3048 {
 3049         struct ip6_moptions      im6o;
 3050         struct ifnet            *ifp;
 3051         struct ifnet            *oifp;
 3052         struct mbuf             *m0;
 3053         struct mbuf             *md;
 3054         struct ip6_hdr          *ip6;
 3055         struct mld_hdr          *mld;
 3056         int                      error;
 3057         int                      off;
 3058         int                      type;
 3059         uint32_t                 ifindex;
 3060 
 3061         CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
 3062 
 3063         /*
 3064          * Set VNET image pointer from enqueued mbuf chain
 3065          * before doing anything else. Whilst we use interface
 3066          * indexes to guard against interface detach, they are
 3067          * unique to each VIMAGE and must be retrieved.
 3068          */
 3069         ifindex = mld_restore_context(m);
 3070 
 3071         /*
 3072          * Check if the ifnet still exists. This limits the scope of
 3073          * any race in the absence of a global ifp lock for low cost
 3074          * (an array lookup).
 3075          */
 3076         ifp = ifnet_byindex(ifindex);
 3077         if (ifp == NULL) {
 3078                 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
 3079                     __func__, m, ifindex);
 3080                 m_freem(m);
 3081                 IP6STAT_INC(ip6s_noroute);
 3082                 goto out;
 3083         }
 3084 
 3085         im6o.im6o_multicast_hlim  = 1;
 3086         im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
 3087         im6o.im6o_multicast_ifp = ifp;
 3088 
 3089         if (m->m_flags & M_MLDV1) {
 3090                 m0 = m;
 3091         } else {
 3092                 m0 = mld_v2_encap_report(ifp, m);
 3093                 if (m0 == NULL) {
 3094                         CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
 3095                         IP6STAT_INC(ip6s_odropped);
 3096                         goto out;
 3097                 }
 3098         }
 3099 
 3100         mld_scrub_context(m0);
 3101         m_clrprotoflags(m);
 3102         m0->m_pkthdr.rcvif = V_loif;
 3103 
 3104         ip6 = mtod(m0, struct ip6_hdr *);
 3105 #if 0
 3106         (void)in6_setscope(&ip6->ip6_dst, ifp, NULL);   /* XXX LOR */
 3107 #else
 3108         /*
 3109          * XXX XXX Break some KPI rules to prevent an LOR which would
 3110          * occur if we called in6_setscope() at transmission.
 3111          * See comments at top of file.
 3112          */
 3113         MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
 3114 #endif
 3115 
 3116         /*
 3117          * Retrieve the ICMPv6 type before handoff to ip6_output(),
 3118          * so we can bump the stats.
 3119          */
 3120         md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
 3121         mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
 3122         type = mld->mld_type;
 3123 
 3124         error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
 3125             &oifp, NULL);
 3126         if (error) {
 3127                 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
 3128                 goto out;
 3129         }
 3130         ICMP6STAT_INC(icp6s_outhist[type]);
 3131         if (oifp != NULL) {
 3132                 icmp6_ifstat_inc(oifp, ifs6_out_msg);
 3133                 switch (type) {
 3134                 case MLD_LISTENER_REPORT:
 3135                 case MLDV2_LISTENER_REPORT:
 3136                         icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
 3137                         break;
 3138                 case MLD_LISTENER_DONE:
 3139                         icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
 3140                         break;
 3141                 }
 3142         }
 3143 out:
 3144         return;
 3145 }
 3146 
 3147 /*
 3148  * Encapsulate an MLDv2 report.
 3149  *
 3150  * KAME IPv6 requires that hop-by-hop options be passed separately,
 3151  * and that the IPv6 header be prepended in a separate mbuf.
 3152  *
 3153  * Returns a pointer to the new mbuf chain head, or NULL if the
 3154  * allocation failed.
 3155  */
 3156 static struct mbuf *
 3157 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
 3158 {
 3159         struct mbuf             *mh;
 3160         struct mldv2_report     *mld;
 3161         struct ip6_hdr          *ip6;
 3162         struct in6_ifaddr       *ia;
 3163         int                      mldreclen;
 3164 
 3165         KASSERT(ifp != NULL, ("%s: null ifp", __func__));
 3166         KASSERT((m->m_flags & M_PKTHDR),
 3167             ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
 3168 
 3169         /*
 3170          * RFC3590: OK to send as :: or tentative during DAD.
 3171          */
 3172         ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
 3173         if (ia == NULL)
 3174                 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
 3175 
 3176         mh = m_gethdr(M_NOWAIT, MT_DATA);
 3177         if (mh == NULL) {
 3178                 if (ia != NULL)
 3179                         ifa_free(&ia->ia_ifa);
 3180                 m_freem(m);
 3181                 return (NULL);
 3182         }
 3183         MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
 3184 
 3185         mldreclen = m_length(m, NULL);
 3186         CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
 3187 
 3188         mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
 3189         mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
 3190             sizeof(struct mldv2_report) + mldreclen;
 3191 
 3192         ip6 = mtod(mh, struct ip6_hdr *);
 3193         ip6->ip6_flow = 0;
 3194         ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
 3195         ip6->ip6_vfc |= IPV6_VERSION;
 3196         ip6->ip6_nxt = IPPROTO_ICMPV6;
 3197         ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
 3198         if (ia != NULL)
 3199                 ifa_free(&ia->ia_ifa);
 3200         ip6->ip6_dst = in6addr_linklocal_allv2routers;
 3201         /* scope ID will be set in netisr */
 3202 
 3203         mld = (struct mldv2_report *)(ip6 + 1);
 3204         mld->mld_type = MLDV2_LISTENER_REPORT;
 3205         mld->mld_code = 0;
 3206         mld->mld_cksum = 0;
 3207         mld->mld_v2_reserved = 0;
 3208         mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
 3209         m->m_pkthdr.PH_vt.vt_nrecs = 0;
 3210 
 3211         mh->m_next = m;
 3212         mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
 3213             sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
 3214         return (mh);
 3215 }
 3216 
 3217 #ifdef KTR
 3218 static char *
 3219 mld_rec_type_to_str(const int type)
 3220 {
 3221 
 3222         switch (type) {
 3223                 case MLD_CHANGE_TO_EXCLUDE_MODE:
 3224                         return "TO_EX";
 3225                         break;
 3226                 case MLD_CHANGE_TO_INCLUDE_MODE:
 3227                         return "TO_IN";
 3228                         break;
 3229                 case MLD_MODE_IS_EXCLUDE:
 3230                         return "MODE_EX";
 3231                         break;
 3232                 case MLD_MODE_IS_INCLUDE:
 3233                         return "MODE_IN";
 3234                         break;
 3235                 case MLD_ALLOW_NEW_SOURCES:
 3236                         return "ALLOW_NEW";
 3237                         break;
 3238                 case MLD_BLOCK_OLD_SOURCES:
 3239                         return "BLOCK_OLD";
 3240                         break;
 3241                 default:
 3242                         break;
 3243         }
 3244         return "unknown";
 3245 }
 3246 #endif
 3247 
 3248 static void
 3249 mld_init(void *unused __unused)
 3250 {
 3251 
 3252         CTR1(KTR_MLD, "%s: initializing", __func__);
 3253         MLD_LOCK_INIT();
 3254 
 3255         ip6_initpktopts(&mld_po);
 3256         mld_po.ip6po_hlim = 1;
 3257         mld_po.ip6po_hbh = &mld_ra.hbh;
 3258         mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
 3259         mld_po.ip6po_flags = IP6PO_DONTFRAG;
 3260 }
 3261 SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL);
 3262 
 3263 static void
 3264 mld_uninit(void *unused __unused)
 3265 {
 3266 
 3267         CTR1(KTR_MLD, "%s: tearing down", __func__);
 3268         MLD_LOCK_DESTROY();
 3269 }
 3270 SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL);
 3271 
 3272 static void
 3273 vnet_mld_init(const void *unused __unused)
 3274 {
 3275 
 3276         CTR1(KTR_MLD, "%s: initializing", __func__);
 3277 
 3278         LIST_INIT(&V_mli_head);
 3279 }
 3280 VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init,
 3281     NULL);
 3282 
 3283 static void
 3284 vnet_mld_uninit(const void *unused __unused)
 3285 {
 3286 
 3287         CTR1(KTR_MLD, "%s: tearing down", __func__);
 3288 
 3289         KASSERT(LIST_EMPTY(&V_mli_head),
 3290             ("%s: mli list not empty; ifnets not detached?", __func__));
 3291 }
 3292 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit,
 3293     NULL);
 3294 
 3295 static int
 3296 mld_modevent(module_t mod, int type, void *unused __unused)
 3297 {
 3298 
 3299     switch (type) {
 3300     case MOD_LOAD:
 3301     case MOD_UNLOAD:
 3302         break;
 3303     default:
 3304         return (EOPNOTSUPP);
 3305     }
 3306     return (0);
 3307 }
 3308 
 3309 static moduledata_t mld_mod = {
 3310     "mld",
 3311     mld_modevent,
 3312     0
 3313 };
 3314 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);

Cache object: bc235b4df997741e192c0d78acd1ed1c


[ 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.