211
212#define V_mld_gsrdelay VNET(mld_gsrdelay)
213#define V_mli_head VNET(mli_head)
214#define V_interface_timers_running6 VNET(interface_timers_running6)
215#define V_state_change_timers_running6 VNET(state_change_timers_running6)
216#define V_current_state_timers_running6 VNET(current_state_timers_running6)
217
218SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */
219
220SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
221 "IPv6 Multicast Listener Discovery");
222
223/*
224 * Virtualized sysctls.
225 */
226SYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
227 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
228 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
229 "Rate limit for MLDv2 Group-and-Source queries in seconds");
230
231/*
232 * Non-virtualized sysctls.
233 */
234SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
235 sysctl_mld_ifinfo, "Per-interface MLDv2 state");
236
237static int mld_v1enable = 1;
238SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW,
239 &mld_v1enable, 0, "Enable fallback to MLDv1");
240TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable);
241
242static int mld_use_allow = 1;
243SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW,
244 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
245TUNABLE_INT("net.inet6.mld.use_allow", &mld_use_allow);
246
247/*
248 * Packed Router Alert option structure declaration.
249 */
250struct mld_raopt {
251 struct ip6_hbh hbh;
252 struct ip6_opt pad;
253 struct ip6_opt_router ra;
254} __packed;
255
256/*
257 * Router Alert hop-by-hop option header.
258 */
259static struct mld_raopt mld_ra = {
260 .hbh = { 0, 0 },
261 .pad = { .ip6o_type = IP6OPT_PADN, 0 },
262 .ra = {
263 .ip6or_type = IP6OPT_ROUTER_ALERT,
264 .ip6or_len = IP6OPT_RTALERT_LEN - 2,
265 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
266 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
267 }
268};
269static struct ip6_pktopts mld_po;
270
271static __inline void
272mld_save_context(struct mbuf *m, struct ifnet *ifp)
273{
274
275#ifdef VIMAGE
276 m->m_pkthdr.header = ifp->if_vnet;
277#endif /* VIMAGE */
278 m->m_pkthdr.flowid = ifp->if_index;
279}
280
281static __inline void
282mld_scrub_context(struct mbuf *m)
283{
284
285 m->m_pkthdr.header = NULL;
286 m->m_pkthdr.flowid = 0;
287}
288
289/*
290 * Restore context from a queued output chain.
291 * Return saved ifindex.
292 *
293 * VIMAGE: The assertion is there to make sure that we
294 * actually called CURVNET_SET() with what's in the mbuf chain.
295 */
296static __inline uint32_t
297mld_restore_context(struct mbuf *m)
298{
299
300#if defined(VIMAGE) && defined(INVARIANTS)
301 KASSERT(curvnet == m->m_pkthdr.header,
302 ("%s: called when curvnet was not restored", __func__));
303#endif
304 return (m->m_pkthdr.flowid);
305}
306
307/*
308 * Retrieve or set threshold between group-source queries in seconds.
309 *
310 * VIMAGE: Assume curvnet set by caller.
311 * SMPng: NOTE: Serialized by MLD lock.
312 */
313static int
314sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
315{
316 int error;
317 int i;
318
319 error = sysctl_wire_old_buffer(req, sizeof(int));
320 if (error)
321 return (error);
322
323 MLD_LOCK();
324
325 i = V_mld_gsrdelay.tv_sec;
326
327 error = sysctl_handle_int(oidp, &i, 0, req);
328 if (error || !req->newptr)
329 goto out_locked;
330
331 if (i < -1 || i >= 60) {
332 error = EINVAL;
333 goto out_locked;
334 }
335
336 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
337 V_mld_gsrdelay.tv_sec, i);
338 V_mld_gsrdelay.tv_sec = i;
339
340out_locked:
341 MLD_UNLOCK();
342 return (error);
343}
344
345/*
346 * Expose struct mld_ifinfo to userland, keyed by ifindex.
347 * For use by ifmcstat(8).
348 *
349 * SMPng: NOTE: Does an unlocked ifindex space read.
350 * VIMAGE: Assume curvnet set by caller. The node handler itself
351 * is not directly virtualized.
352 */
353static int
354sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
355{
356 int *name;
357 int error;
358 u_int namelen;
359 struct ifnet *ifp;
360 struct mld_ifinfo *mli;
361
362 name = (int *)arg1;
363 namelen = arg2;
364
365 if (req->newptr != NULL)
366 return (EPERM);
367
368 if (namelen != 1)
369 return (EINVAL);
370
371 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
372 if (error)
373 return (error);
374
375 IN6_MULTI_LOCK();
376 MLD_LOCK();
377
378 if (name[0] <= 0 || name[0] > V_if_index) {
379 error = ENOENT;
380 goto out_locked;
381 }
382
383 error = ENOENT;
384
385 ifp = ifnet_byindex(name[0]);
386 if (ifp == NULL)
387 goto out_locked;
388
389 LIST_FOREACH(mli, &V_mli_head, mli_link) {
390 if (ifp == mli->mli_ifp) {
391 error = SYSCTL_OUT(req, mli,
392 sizeof(struct mld_ifinfo));
393 break;
394 }
395 }
396
397out_locked:
398 MLD_UNLOCK();
399 IN6_MULTI_UNLOCK();
400 return (error);
401}
402
403/*
404 * Dispatch an entire queue of pending packet chains.
405 * VIMAGE: Assumes the vnet pointer has been set.
406 */
407static void
408mld_dispatch_queue(struct ifqueue *ifq, int limit)
409{
410 struct mbuf *m;
411
412 for (;;) {
413 _IF_DEQUEUE(ifq, m);
414 if (m == NULL)
415 break;
416 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m);
417 mld_dispatch_packet(m);
418 if (--limit == 0)
419 break;
420 }
421}
422
423/*
424 * Filter outgoing MLD report state by group.
425 *
426 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
427 * and node-local addresses. However, kernel and socket consumers
428 * always embed the KAME scope ID in the address provided, so strip it
429 * when performing comparison.
430 * Note: This is not the same as the *multicast* scope.
431 *
432 * Return zero if the given group is one for which MLD reports
433 * should be suppressed, or non-zero if reports should be issued.
434 */
435static __inline int
436mld_is_addr_reported(const struct in6_addr *addr)
437{
438
439 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
440
441 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
442 return (0);
443
444 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
445 struct in6_addr tmp = *addr;
446 in6_clearscope(&tmp);
447 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
448 return (0);
449 }
450
451 return (1);
452}
453
454/*
455 * Attach MLD when PF_INET6 is attached to an interface.
456 *
457 * SMPng: Normally called with IF_AFDATA_LOCK held.
458 */
459struct mld_ifinfo *
460mld_domifattach(struct ifnet *ifp)
461{
462 struct mld_ifinfo *mli;
463
464 CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
465 __func__, ifp, ifp->if_xname);
466
467 MLD_LOCK();
468
469 mli = mli_alloc_locked(ifp);
470 if (!(ifp->if_flags & IFF_MULTICAST))
471 mli->mli_flags |= MLIF_SILENT;
472 if (mld_use_allow)
473 mli->mli_flags |= MLIF_USEALLOW;
474
475 MLD_UNLOCK();
476
477 return (mli);
478}
479
480/*
481 * VIMAGE: assume curvnet set by caller.
482 */
483static struct mld_ifinfo *
484mli_alloc_locked(/*const*/ struct ifnet *ifp)
485{
486 struct mld_ifinfo *mli;
487
488 MLD_LOCK_ASSERT();
489
490 mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO);
491 if (mli == NULL)
492 goto out;
493
494 mli->mli_ifp = ifp;
495 mli->mli_version = MLD_VERSION_2;
496 mli->mli_flags = 0;
497 mli->mli_rv = MLD_RV_INIT;
498 mli->mli_qi = MLD_QI_INIT;
499 mli->mli_qri = MLD_QRI_INIT;
500 mli->mli_uri = MLD_URI_INIT;
501
502 SLIST_INIT(&mli->mli_relinmhead);
503
504 /*
505 * Responses to general queries are subject to bounds.
506 */
507 IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
508
509 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
510
511 CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)",
512 ifp, ifp->if_xname);
513
514out:
515 return (mli);
516}
517
518/*
519 * Hook for ifdetach.
520 *
521 * NOTE: Some finalization tasks need to run before the protocol domain
522 * is detached, but also before the link layer does its cleanup.
523 * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
524 *
525 * SMPng: Caller must hold IN6_MULTI_LOCK().
526 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
527 * XXX This routine is also bitten by unlocked ifma_protospec access.
528 */
529void
530mld_ifdetach(struct ifnet *ifp)
531{
532 struct mld_ifinfo *mli;
533 struct ifmultiaddr *ifma;
534 struct in6_multi *inm, *tinm;
535
536 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
537 ifp->if_xname);
538
539 IN6_MULTI_LOCK_ASSERT();
540 MLD_LOCK();
541
542 mli = MLD_IFINFO(ifp);
543 if (mli->mli_version == MLD_VERSION_2) {
544 IF_ADDR_LOCK(ifp);
545 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
546 if (ifma->ifma_addr->sa_family != AF_INET6 ||
547 ifma->ifma_protospec == NULL)
548 continue;
549 inm = (struct in6_multi *)ifma->ifma_protospec;
550 if (inm->in6m_state == MLD_LEAVING_MEMBER) {
551 SLIST_INSERT_HEAD(&mli->mli_relinmhead,
552 inm, in6m_nrele);
553 }
554 in6m_clear_recorded(inm);
555 }
556 IF_ADDR_UNLOCK(ifp);
557 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
558 tinm) {
559 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
560 in6m_release_locked(inm);
561 }
562 }
563
564 MLD_UNLOCK();
565}
566
567/*
568 * Hook for domifdetach.
569 * Runs after link-layer cleanup; free MLD state.
570 *
571 * SMPng: Normally called with IF_AFDATA_LOCK held.
572 */
573void
574mld_domifdetach(struct ifnet *ifp)
575{
576
577 CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
578 __func__, ifp, ifp->if_xname);
579
580 MLD_LOCK();
581 mli_delete_locked(ifp);
582 MLD_UNLOCK();
583}
584
585static void
586mli_delete_locked(const struct ifnet *ifp)
587{
588 struct mld_ifinfo *mli, *tmli;
589
590 CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)",
591 __func__, ifp, ifp->if_xname);
592
593 MLD_LOCK_ASSERT();
594
595 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
596 if (mli->mli_ifp == ifp) {
597 /*
598 * Free deferred General Query responses.
599 */
600 _IF_DRAIN(&mli->mli_gq);
601
602 LIST_REMOVE(mli, mli_link);
603
604 KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
605 ("%s: there are dangling in_multi references",
606 __func__));
607
608 free(mli, M_MLD);
609 return;
610 }
611 }
612#ifdef INVARIANTS
613 panic("%s: mld_ifinfo not found for ifp %p\n", __func__, ifp);
614#endif
615}
616
617/*
618 * Process a received MLDv1 general or address-specific query.
619 * Assumes that the query header has been pulled up to sizeof(mld_hdr).
620 *
621 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
622 * mld_addr. This is OK as we own the mbuf chain.
623 */
624static int
625mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
626 /*const*/ struct mld_hdr *mld)
627{
628 struct ifmultiaddr *ifma;
629 struct mld_ifinfo *mli;
630 struct in6_multi *inm;
631 int is_general_query;
632 uint16_t timer;
633#ifdef KTR
634 char ip6tbuf[INET6_ADDRSTRLEN];
635#endif
636
637 is_general_query = 0;
638
639 if (!mld_v1enable) {
640 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
641 ip6_sprintf(ip6tbuf, &mld->mld_addr),
642 ifp, ifp->if_xname);
643 return (0);
644 }
645
646 /*
647 * RFC3810 Section 6.2: MLD queries must originate from
648 * a router's link-local address.
649 */
650 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
651 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
652 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
653 ifp, ifp->if_xname);
654 return (0);
655 }
656
657 /*
658 * Do address field validation upfront before we accept
659 * the query.
660 */
661 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
662 /*
663 * MLDv1 General Query.
664 * If this was not sent to the all-nodes group, ignore it.
665 */
666 struct in6_addr dst;
667
668 dst = ip6->ip6_dst;
669 in6_clearscope(&dst);
670 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
671 return (EINVAL);
672 is_general_query = 1;
673 } else {
674 /*
675 * Embed scope ID of receiving interface in MLD query for
676 * lookup whilst we don't hold other locks.
677 */
678 in6_setscope(&mld->mld_addr, ifp, NULL);
679 }
680
681 IN6_MULTI_LOCK();
682 MLD_LOCK();
683 IF_ADDR_LOCK(ifp);
684
685 /*
686 * Switch to MLDv1 host compatibility mode.
687 */
688 mli = MLD_IFINFO(ifp);
689 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
690 mld_set_version(mli, MLD_VERSION_1);
691
692 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
693 if (timer == 0)
694 timer = 1;
695
696 if (is_general_query) {
697 /*
698 * For each reporting group joined on this
699 * interface, kick the report timer.
700 */
701 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
702 ifp, ifp->if_xname);
703 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
704 if (ifma->ifma_addr->sa_family != AF_INET6 ||
705 ifma->ifma_protospec == NULL)
706 continue;
707 inm = (struct in6_multi *)ifma->ifma_protospec;
708 mld_v1_update_group(inm, timer);
709 }
710 } else {
711 /*
712 * MLDv1 Group-Specific Query.
713 * If this is a group-specific MLDv1 query, we need only
714 * look up the single group to process it.
715 */
716 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
717 if (inm != NULL) {
718 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
719 ip6_sprintf(ip6tbuf, &mld->mld_addr),
720 ifp, ifp->if_xname);
721 mld_v1_update_group(inm, timer);
722 }
723 /* XXX Clear embedded scope ID as userland won't expect it. */
724 in6_clearscope(&mld->mld_addr);
725 }
726
727 IF_ADDR_UNLOCK(ifp);
728 MLD_UNLOCK();
729 IN6_MULTI_UNLOCK();
730
731 return (0);
732}
733
734/*
735 * Update the report timer on a group in response to an MLDv1 query.
736 *
737 * If we are becoming the reporting member for this group, start the timer.
738 * If we already are the reporting member for this group, and timer is
739 * below the threshold, reset it.
740 *
741 * We may be updating the group for the first time since we switched
742 * to MLDv2. If we are, then we must clear any recorded source lists,
743 * and transition to REPORTING state; the group timer is overloaded
744 * for group and group-source query responses.
745 *
746 * Unlike MLDv2, the delay per group should be jittered
747 * to avoid bursts of MLDv1 reports.
748 */
749static void
750mld_v1_update_group(struct in6_multi *inm, const int timer)
751{
752#ifdef KTR
753 char ip6tbuf[INET6_ADDRSTRLEN];
754#endif
755
756 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
757 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
758 inm->in6m_ifp->if_xname, timer);
759
760 IN6_MULTI_LOCK_ASSERT();
761
762 switch (inm->in6m_state) {
763 case MLD_NOT_MEMBER:
764 case MLD_SILENT_MEMBER:
765 break;
766 case MLD_REPORTING_MEMBER:
767 if (inm->in6m_timer != 0 &&
768 inm->in6m_timer <= timer) {
769 CTR1(KTR_MLD, "%s: REPORTING and timer running, "
770 "skipping.", __func__);
771 break;
772 }
773 /* FALLTHROUGH */
774 case MLD_SG_QUERY_PENDING_MEMBER:
775 case MLD_G_QUERY_PENDING_MEMBER:
776 case MLD_IDLE_MEMBER:
777 case MLD_LAZY_MEMBER:
778 case MLD_AWAKENING_MEMBER:
779 CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
780 inm->in6m_state = MLD_REPORTING_MEMBER;
781 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
782 V_current_state_timers_running6 = 1;
783 break;
784 case MLD_SLEEPING_MEMBER:
785 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
786 inm->in6m_state = MLD_AWAKENING_MEMBER;
787 break;
788 case MLD_LEAVING_MEMBER:
789 break;
790 }
791}
792
793/*
794 * Process a received MLDv2 general, group-specific or
795 * group-and-source-specific query.
796 *
797 * Assumes that the query header has been pulled up to sizeof(mldv2_query).
798 *
799 * Return 0 if successful, otherwise an appropriate error code is returned.
800 */
801static int
802mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
803 struct mbuf *m, const int off, const int icmp6len)
804{
805 struct mld_ifinfo *mli;
806 struct mldv2_query *mld;
807 struct in6_multi *inm;
808 uint32_t maxdelay, nsrc, qqi;
809 int is_general_query;
810 uint16_t timer;
811 uint8_t qrv;
812#ifdef KTR
813 char ip6tbuf[INET6_ADDRSTRLEN];
814#endif
815
816 is_general_query = 0;
817
818 /*
819 * RFC3810 Section 6.2: MLD queries must originate from
820 * a router's link-local address.
821 */
822 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
823 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
824 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
825 ifp, ifp->if_xname);
826 return (0);
827 }
828
829 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname);
830
831 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
832
833 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */
834 if (maxdelay >= 32678) {
835 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
836 (MLD_MRC_EXP(maxdelay) + 3);
837 }
838 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
839 if (timer == 0)
840 timer = 1;
841
842 qrv = MLD_QRV(mld->mld_misc);
843 if (qrv < 2) {
844 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
845 qrv, MLD_RV_INIT);
846 qrv = MLD_RV_INIT;
847 }
848
849 qqi = mld->mld_qqi;
850 if (qqi >= 128) {
851 qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
852 (MLD_QQIC_EXP(mld->mld_qqi) + 3);
853 }
854
855 nsrc = ntohs(mld->mld_numsrc);
856 if (nsrc > MLD_MAX_GS_SOURCES)
857 return (EMSGSIZE);
858 if (icmp6len < sizeof(struct mldv2_query) +
859 (nsrc * sizeof(struct in6_addr)))
860 return (EMSGSIZE);
861
862 /*
863 * Do further input validation upfront to avoid resetting timers
864 * should we need to discard this query.
865 */
866 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
867 /*
868 * General Queries SHOULD be directed to ff02::1.
869 * A general query with a source list has undefined
870 * behaviour; discard it.
871 */
872 struct in6_addr dst;
873
874 dst = ip6->ip6_dst;
875 in6_clearscope(&dst);
876 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes) ||
877 nsrc > 0)
878 return (EINVAL);
879 is_general_query = 1;
880 } else {
881 /*
882 * Embed scope ID of receiving interface in MLD query for
883 * lookup whilst we don't hold other locks (due to KAME
884 * locking lameness). We own this mbuf chain just now.
885 */
886 in6_setscope(&mld->mld_addr, ifp, NULL);
887 }
888
889 IN6_MULTI_LOCK();
890 MLD_LOCK();
891 IF_ADDR_LOCK(ifp);
892
893 mli = MLD_IFINFO(ifp);
894 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
895
896 /*
897 * Discard the v2 query if we're in Compatibility Mode.
898 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
899 * until the Old Version Querier Present timer expires.
900 */
901 if (mli->mli_version != MLD_VERSION_2)
902 goto out_locked;
903
904 mld_set_version(mli, MLD_VERSION_2);
905 mli->mli_rv = qrv;
906 mli->mli_qi = qqi;
907 mli->mli_qri = maxdelay;
908
909 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
910 maxdelay);
911
912 if (is_general_query) {
913 /*
914 * MLDv2 General Query.
915 *
916 * Schedule a current-state report on this ifp for
917 * all groups, possibly containing source lists.
918 *
919 * If there is a pending General Query response
920 * scheduled earlier than the selected delay, do
921 * not schedule any other reports.
922 * Otherwise, reset the interface timer.
923 */
924 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
925 ifp, ifp->if_xname);
926 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
927 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
928 V_interface_timers_running6 = 1;
929 }
930 } else {
931 /*
932 * MLDv2 Group-specific or Group-and-source-specific Query.
933 *
934 * Group-source-specific queries are throttled on
935 * a per-group basis to defeat denial-of-service attempts.
936 * Queries for groups we are not a member of on this
937 * link are simply ignored.
938 */
939 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
940 if (inm == NULL)
941 goto out_locked;
942 if (nsrc > 0) {
943 if (!ratecheck(&inm->in6m_lastgsrtv,
944 &V_mld_gsrdelay)) {
945 CTR1(KTR_MLD, "%s: GS query throttled.",
946 __func__);
947 goto out_locked;
948 }
949 }
950 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
951 ifp, ifp->if_xname);
952 /*
953 * If there is a pending General Query response
954 * scheduled sooner than the selected delay, no
955 * further report need be scheduled.
956 * Otherwise, prepare to respond to the
957 * group-specific or group-and-source query.
958 */
959 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
960 mld_v2_process_group_query(inm, mli, timer, m, off);
961
962 /* XXX Clear embedded scope ID as userland won't expect it. */
963 in6_clearscope(&mld->mld_addr);
964 }
965
966out_locked:
967 IF_ADDR_UNLOCK(ifp);
968 MLD_UNLOCK();
969 IN6_MULTI_UNLOCK();
970
971 return (0);
972}
973
974/*
975 * Process a recieved MLDv2 group-specific or group-and-source-specific
976 * query.
977 * Return <0 if any error occured. Currently this is ignored.
978 */
979static int
980mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli,
981 int timer, struct mbuf *m0, const int off)
982{
983 struct mldv2_query *mld;
984 int retval;
985 uint16_t nsrc;
986
987 IN6_MULTI_LOCK_ASSERT();
988 MLD_LOCK_ASSERT();
989
990 retval = 0;
991 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
992
993 switch (inm->in6m_state) {
994 case MLD_NOT_MEMBER:
995 case MLD_SILENT_MEMBER:
996 case MLD_SLEEPING_MEMBER:
997 case MLD_LAZY_MEMBER:
998 case MLD_AWAKENING_MEMBER:
999 case MLD_IDLE_MEMBER:
1000 case MLD_LEAVING_MEMBER:
1001 return (retval);
1002 break;
1003 case MLD_REPORTING_MEMBER:
1004 case MLD_G_QUERY_PENDING_MEMBER:
1005 case MLD_SG_QUERY_PENDING_MEMBER:
1006 break;
1007 }
1008
1009 nsrc = ntohs(mld->mld_numsrc);
1010
1011 /*
1012 * Deal with group-specific queries upfront.
1013 * If any group query is already pending, purge any recorded
1014 * source-list state if it exists, and schedule a query response
1015 * for this group-specific query.
1016 */
1017 if (nsrc == 0) {
1018 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1019 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1020 in6m_clear_recorded(inm);
1021 timer = min(inm->in6m_timer, timer);
1022 }
1023 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1024 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1025 V_current_state_timers_running6 = 1;
1026 return (retval);
1027 }
1028
1029 /*
1030 * Deal with the case where a group-and-source-specific query has
1031 * been received but a group-specific query is already pending.
1032 */
1033 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1034 timer = min(inm->in6m_timer, timer);
1035 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1036 V_current_state_timers_running6 = 1;
1037 return (retval);
1038 }
1039
1040 /*
1041 * Finally, deal with the case where a group-and-source-specific
1042 * query has been received, where a response to a previous g-s-r
1043 * query exists, or none exists.
1044 * In this case, we need to parse the source-list which the Querier
1045 * has provided us with and check if we have any source list filter
1046 * entries at T1 for these sources. If we do not, there is no need
1047 * schedule a report and the query may be dropped.
1048 * If we do, we must record them and schedule a current-state
1049 * report for those sources.
1050 */
1051 if (inm->in6m_nsrc > 0) {
1052 struct mbuf *m;
1053 uint8_t *sp;
1054 int i, nrecorded;
1055 int soff;
1056
1057 m = m0;
1058 soff = off + sizeof(struct mldv2_query);
1059 nrecorded = 0;
1060 for (i = 0; i < nsrc; i++) {
1061 sp = mtod(m, uint8_t *) + soff;
1062 retval = in6m_record_source(inm,
1063 (const struct in6_addr *)sp);
1064 if (retval < 0)
1065 break;
1066 nrecorded += retval;
1067 soff += sizeof(struct in6_addr);
1068 if (soff >= m->m_len) {
1069 soff = soff - m->m_len;
1070 m = m->m_next;
1071 if (m == NULL)
1072 break;
1073 }
1074 }
1075 if (nrecorded > 0) {
1076 CTR1(KTR_MLD,
1077 "%s: schedule response to SG query", __func__);
1078 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1079 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1080 V_current_state_timers_running6 = 1;
1081 }
1082 }
1083
1084 return (retval);
1085}
1086
1087/*
1088 * Process a received MLDv1 host membership report.
1089 * Assumes mld points to mld_hdr in pulled up mbuf chain.
1090 *
1091 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1092 * mld_addr. This is OK as we own the mbuf chain.
1093 */
1094static int
1095mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1096 /*const*/ struct mld_hdr *mld)
1097{
1098 struct in6_addr src, dst;
1099 struct in6_ifaddr *ia;
1100 struct in6_multi *inm;
1101#ifdef KTR
1102 char ip6tbuf[INET6_ADDRSTRLEN];
1103#endif
1104
1105 if (!mld_v1enable) {
1106 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1107 ip6_sprintf(ip6tbuf, &mld->mld_addr),
1108 ifp, ifp->if_xname);
1109 return (0);
1110 }
1111
1112 if (ifp->if_flags & IFF_LOOPBACK)
1113 return (0);
1114
1115 /*
1116 * MLDv1 reports must originate from a host's link-local address,
1117 * or the unspecified address (when booting).
1118 */
1119 src = ip6->ip6_src;
1120 in6_clearscope(&src);
1121 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1122 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1123 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1124 ifp, ifp->if_xname);
1125 return (EINVAL);
1126 }
1127
1128 /*
1129 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1130 * group, and must be directed to the group itself.
1131 */
1132 dst = ip6->ip6_dst;
1133 in6_clearscope(&dst);
1134 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1135 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1136 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1137 ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1138 ifp, ifp->if_xname);
1139 return (EINVAL);
1140 }
1141
1142 /*
1143 * Make sure we don't hear our own membership report, as fast
1144 * leave requires knowing that we are the only member of a
1145 * group. Assume we used the link-local address if available,
1146 * otherwise look for ::.
1147 *
1148 * XXX Note that scope ID comparison is needed for the address
1149 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1150 * performed for the on-wire address.
1151 */
1152 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1153 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1154 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1155 if (ia != NULL)
1156 ifa_free(&ia->ia_ifa);
1157 return (0);
1158 }
1159 if (ia != NULL)
1160 ifa_free(&ia->ia_ifa);
1161
1162 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1163 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname);
1164
1165 /*
1166 * Embed scope ID of receiving interface in MLD query for lookup
1167 * whilst we don't hold other locks (due to KAME locking lameness).
1168 */
1169 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1170 in6_setscope(&mld->mld_addr, ifp, NULL);
1171
1172 IN6_MULTI_LOCK();
1173 MLD_LOCK();
1174 IF_ADDR_LOCK(ifp);
1175
1176 /*
1177 * MLDv1 report suppression.
1178 * If we are a member of this group, and our membership should be
1179 * reported, and our group timer is pending or about to be reset,
1180 * stop our group timer by transitioning to the 'lazy' state.
1181 */
1182 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1183 if (inm != NULL) {
1184 struct mld_ifinfo *mli;
1185
1186 mli = inm->in6m_mli;
1187 KASSERT(mli != NULL,
1188 ("%s: no mli for ifp %p", __func__, ifp));
1189
1190 /*
1191 * If we are in MLDv2 host mode, do not allow the
1192 * other host's MLDv1 report to suppress our reports.
1193 */
1194 if (mli->mli_version == MLD_VERSION_2)
1195 goto out_locked;
1196
1197 inm->in6m_timer = 0;
1198
1199 switch (inm->in6m_state) {
1200 case MLD_NOT_MEMBER:
1201 case MLD_SILENT_MEMBER:
1202 case MLD_SLEEPING_MEMBER:
1203 break;
1204 case MLD_REPORTING_MEMBER:
1205 case MLD_IDLE_MEMBER:
1206 case MLD_AWAKENING_MEMBER:
1207 CTR3(KTR_MLD,
1208 "report suppressed for %s on ifp %p(%s)",
1209 ip6_sprintf(ip6tbuf, &mld->mld_addr),
1210 ifp, ifp->if_xname);
1211 case MLD_LAZY_MEMBER:
1212 inm->in6m_state = MLD_LAZY_MEMBER;
1213 break;
1214 case MLD_G_QUERY_PENDING_MEMBER:
1215 case MLD_SG_QUERY_PENDING_MEMBER:
1216 case MLD_LEAVING_MEMBER:
1217 break;
1218 }
1219 }
1220
1221out_locked:
1222 MLD_UNLOCK();
1223 IF_ADDR_UNLOCK(ifp);
1224 IN6_MULTI_UNLOCK();
1225
1226 /* XXX Clear embedded scope ID as userland won't expect it. */
1227 in6_clearscope(&mld->mld_addr);
1228
1229 return (0);
1230}
1231
1232/*
1233 * MLD input path.
1234 *
1235 * Assume query messages which fit in a single ICMPv6 message header
1236 * have been pulled up.
1237 * Assume that userland will want to see the message, even if it
1238 * otherwise fails kernel input validation; do not free it.
1239 * Pullup may however free the mbuf chain m if it fails.
1240 *
1241 * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1242 */
1243int
1244mld_input(struct mbuf *m, int off, int icmp6len)
1245{
1246 struct ifnet *ifp;
1247 struct ip6_hdr *ip6;
1248 struct mld_hdr *mld;
1249 int mldlen;
1250
1251 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1252
1253 ifp = m->m_pkthdr.rcvif;
1254
1255 ip6 = mtod(m, struct ip6_hdr *);
1256
1257 /* Pullup to appropriate size. */
1258 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1259 if (mld->mld_type == MLD_LISTENER_QUERY &&
1260 icmp6len >= sizeof(struct mldv2_query)) {
1261 mldlen = sizeof(struct mldv2_query);
1262 } else {
1263 mldlen = sizeof(struct mld_hdr);
1264 }
1265 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1266 if (mld == NULL) {
1267 ICMP6STAT_INC(icp6s_badlen);
1268 return (IPPROTO_DONE);
1269 }
1270
1271 /*
1272 * Userland needs to see all of this traffic for implementing
1273 * the endpoint discovery portion of multicast routing.
1274 */
1275 switch (mld->mld_type) {
1276 case MLD_LISTENER_QUERY:
1277 icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1278 if (icmp6len == sizeof(struct mld_hdr)) {
1279 if (mld_v1_input_query(ifp, ip6, mld) != 0)
1280 return (0);
1281 } else if (icmp6len >= sizeof(struct mldv2_query)) {
1282 if (mld_v2_input_query(ifp, ip6, m, off,
1283 icmp6len) != 0)
1284 return (0);
1285 }
1286 break;
1287 case MLD_LISTENER_REPORT:
1288 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1289 if (mld_v1_input_report(ifp, ip6, mld) != 0)
1290 return (0);
1291 break;
1292 case MLDV2_LISTENER_REPORT:
1293 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1294 break;
1295 case MLD_LISTENER_DONE:
1296 icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1297 break;
1298 default:
1299 break;
1300 }
1301
1302 return (0);
1303}
1304
1305/*
1306 * Fast timeout handler (global).
1307 * VIMAGE: Timeout handlers are expected to service all vimages.
1308 */
1309void
1310mld_fasttimo(void)
1311{
1312 VNET_ITERATOR_DECL(vnet_iter);
1313
1314 VNET_LIST_RLOCK_NOSLEEP();
1315 VNET_FOREACH(vnet_iter) {
1316 CURVNET_SET(vnet_iter);
1317 mld_fasttimo_vnet();
1318 CURVNET_RESTORE();
1319 }
1320 VNET_LIST_RUNLOCK_NOSLEEP();
1321}
1322
1323/*
1324 * Fast timeout handler (per-vnet).
1325 *
1326 * VIMAGE: Assume caller has set up our curvnet.
1327 */
1328static void
1329mld_fasttimo_vnet(void)
1330{
1331 struct ifqueue scq; /* State-change packets */
1332 struct ifqueue qrq; /* Query response packets */
1333 struct ifnet *ifp;
1334 struct mld_ifinfo *mli;
1335 struct ifmultiaddr *ifma, *tifma;
1336 struct in6_multi *inm;
1337 int uri_fasthz;
1338
1339 uri_fasthz = 0;
1340
1341 /*
1342 * Quick check to see if any work needs to be done, in order to
1343 * minimize the overhead of fasttimo processing.
1344 * SMPng: XXX Unlocked reads.
1345 */
1346 if (!V_current_state_timers_running6 &&
1347 !V_interface_timers_running6 &&
1348 !V_state_change_timers_running6)
1349 return;
1350
1351 IN6_MULTI_LOCK();
1352 MLD_LOCK();
1353
1354 /*
1355 * MLDv2 General Query response timer processing.
1356 */
1357 if (V_interface_timers_running6) {
1358 CTR1(KTR_MLD, "%s: interface timers running", __func__);
1359
1360 V_interface_timers_running6 = 0;
1361 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1362 if (mli->mli_v2_timer == 0) {
1363 /* Do nothing. */
1364 } else if (--mli->mli_v2_timer == 0) {
1365 mld_v2_dispatch_general_query(mli);
1366 } else {
1367 V_interface_timers_running6 = 1;
1368 }
1369 }
1370 }
1371
1372 if (!V_current_state_timers_running6 &&
1373 !V_state_change_timers_running6)
1374 goto out_locked;
1375
1376 V_current_state_timers_running6 = 0;
1377 V_state_change_timers_running6 = 0;
1378
1379 CTR1(KTR_MLD, "%s: state change timers running", __func__);
1380
1381 /*
1382 * MLD host report and state-change timer processing.
1383 * Note: Processing a v2 group timer may remove a node.
1384 */
1385 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1386 ifp = mli->mli_ifp;
1387
1388 if (mli->mli_version == MLD_VERSION_2) {
1389 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1390 PR_FASTHZ);
1391
1392 memset(&qrq, 0, sizeof(struct ifqueue));
1393 IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS);
1394
1395 memset(&scq, 0, sizeof(struct ifqueue));
1396 IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1397 }
1398
1399 IF_ADDR_LOCK(ifp);
1400 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1401 tifma) {
1402 if (ifma->ifma_addr->sa_family != AF_INET6 ||
1403 ifma->ifma_protospec == NULL)
1404 continue;
1405 inm = (struct in6_multi *)ifma->ifma_protospec;
1406 switch (mli->mli_version) {
1407 case MLD_VERSION_1:
1408 /*
1409 * XXX Drop IF_ADDR lock temporarily to
1410 * avoid recursion caused by a potential
1411 * call by in6ifa_ifpforlinklocal().
1412 * rwlock candidate?
1413 */
1414 IF_ADDR_UNLOCK(ifp);
1415 mld_v1_process_group_timer(inm,
1416 mli->mli_version);
1417 IF_ADDR_LOCK(ifp);
1418 break;
1419 case MLD_VERSION_2:
1420 mld_v2_process_group_timers(mli, &qrq,
1421 &scq, inm, uri_fasthz);
1422 break;
1423 }
1424 }
1425 IF_ADDR_UNLOCK(ifp);
1426
1427 if (mli->mli_version == MLD_VERSION_2) {
1428 struct in6_multi *tinm;
1429
1430 mld_dispatch_queue(&qrq, 0);
1431 mld_dispatch_queue(&scq, 0);
1432
1433 /*
1434 * Free the in_multi reference(s) for
1435 * this lifecycle.
1436 */
1437 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1438 in6m_nrele, tinm) {
1439 SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1440 in6m_nrele);
1441 in6m_release_locked(inm);
1442 }
1443 }
1444 }
1445
1446out_locked:
1447 MLD_UNLOCK();
1448 IN6_MULTI_UNLOCK();
1449}
1450
1451/*
1452 * Update host report group timer.
1453 * Will update the global pending timer flags.
1454 */
1455static void
1456mld_v1_process_group_timer(struct in6_multi *inm, const int version)
1457{
1458 int report_timer_expired;
1459
1460 IN6_MULTI_LOCK_ASSERT();
1461 MLD_LOCK_ASSERT();
1462
1463 if (inm->in6m_timer == 0) {
1464 report_timer_expired = 0;
1465 } else if (--inm->in6m_timer == 0) {
1466 report_timer_expired = 1;
1467 } else {
1468 V_current_state_timers_running6 = 1;
1469 return;
1470 }
1471
1472 switch (inm->in6m_state) {
1473 case MLD_NOT_MEMBER:
1474 case MLD_SILENT_MEMBER:
1475 case MLD_IDLE_MEMBER:
1476 case MLD_LAZY_MEMBER:
1477 case MLD_SLEEPING_MEMBER:
1478 case MLD_AWAKENING_MEMBER:
1479 break;
1480 case MLD_REPORTING_MEMBER:
1481 if (report_timer_expired) {
1482 inm->in6m_state = MLD_IDLE_MEMBER;
1483 (void)mld_v1_transmit_report(inm,
1484 MLD_LISTENER_REPORT);
1485 }
1486 break;
1487 case MLD_G_QUERY_PENDING_MEMBER:
1488 case MLD_SG_QUERY_PENDING_MEMBER:
1489 case MLD_LEAVING_MEMBER:
1490 break;
1491 }
1492}
1493
1494/*
1495 * Update a group's timers for MLDv2.
1496 * Will update the global pending timer flags.
1497 * Note: Unlocked read from mli.
1498 */
1499static void
1500mld_v2_process_group_timers(struct mld_ifinfo *mli,
1501 struct ifqueue *qrq, struct ifqueue *scq,
1502 struct in6_multi *inm, const int uri_fasthz)
1503{
1504 int query_response_timer_expired;
1505 int state_change_retransmit_timer_expired;
1506#ifdef KTR
1507 char ip6tbuf[INET6_ADDRSTRLEN];
1508#endif
1509
1510 IN6_MULTI_LOCK_ASSERT();
1511 MLD_LOCK_ASSERT();
1512
1513 query_response_timer_expired = 0;
1514 state_change_retransmit_timer_expired = 0;
1515
1516 /*
1517 * During a transition from compatibility mode back to MLDv2,
1518 * a group record in REPORTING state may still have its group
1519 * timer active. This is a no-op in this function; it is easier
1520 * to deal with it here than to complicate the slow-timeout path.
1521 */
1522 if (inm->in6m_timer == 0) {
1523 query_response_timer_expired = 0;
1524 } else if (--inm->in6m_timer == 0) {
1525 query_response_timer_expired = 1;
1526 } else {
1527 V_current_state_timers_running6 = 1;
1528 }
1529
1530 if (inm->in6m_sctimer == 0) {
1531 state_change_retransmit_timer_expired = 0;
1532 } else if (--inm->in6m_sctimer == 0) {
1533 state_change_retransmit_timer_expired = 1;
1534 } else {
1535 V_state_change_timers_running6 = 1;
1536 }
1537
1538 /* We are in fasttimo, so be quick about it. */
1539 if (!state_change_retransmit_timer_expired &&
1540 !query_response_timer_expired)
1541 return;
1542
1543 switch (inm->in6m_state) {
1544 case MLD_NOT_MEMBER:
1545 case MLD_SILENT_MEMBER:
1546 case MLD_SLEEPING_MEMBER:
1547 case MLD_LAZY_MEMBER:
1548 case MLD_AWAKENING_MEMBER:
1549 case MLD_IDLE_MEMBER:
1550 break;
1551 case MLD_G_QUERY_PENDING_MEMBER:
1552 case MLD_SG_QUERY_PENDING_MEMBER:
1553 /*
1554 * Respond to a previously pending Group-Specific
1555 * or Group-and-Source-Specific query by enqueueing
1556 * the appropriate Current-State report for
1557 * immediate transmission.
1558 */
1559 if (query_response_timer_expired) {
1560 int retval;
1561
1562 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1563 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1564 0);
1565 CTR2(KTR_MLD, "%s: enqueue record = %d",
1566 __func__, retval);
1567 inm->in6m_state = MLD_REPORTING_MEMBER;
1568 in6m_clear_recorded(inm);
1569 }
1570 /* FALLTHROUGH */
1571 case MLD_REPORTING_MEMBER:
1572 case MLD_LEAVING_MEMBER:
1573 if (state_change_retransmit_timer_expired) {
1574 /*
1575 * State-change retransmission timer fired.
1576 * If there are any further pending retransmissions,
1577 * set the global pending state-change flag, and
1578 * reset the timer.
1579 */
1580 if (--inm->in6m_scrv > 0) {
1581 inm->in6m_sctimer = uri_fasthz;
1582 V_state_change_timers_running6 = 1;
1583 }
1584 /*
1585 * Retransmit the previously computed state-change
1586 * report. If there are no further pending
1587 * retransmissions, the mbuf queue will be consumed.
1588 * Update T0 state to T1 as we have now sent
1589 * a state-change.
1590 */
1591 (void)mld_v2_merge_state_changes(inm, scq);
1592
1593 in6m_commit(inm);
1594 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1595 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1596 inm->in6m_ifp->if_xname);
1597
1598 /*
1599 * If we are leaving the group for good, make sure
1600 * we release MLD's reference to it.
1601 * This release must be deferred using a SLIST,
1602 * as we are called from a loop which traverses
1603 * the in_ifmultiaddr TAILQ.
1604 */
1605 if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1606 inm->in6m_scrv == 0) {
1607 inm->in6m_state = MLD_NOT_MEMBER;
1608 SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1609 inm, in6m_nrele);
1610 }
1611 }
1612 break;
1613 }
1614}
1615
1616/*
1617 * Switch to a different version on the given interface,
1618 * as per Section 9.12.
1619 */
1620static void
1621mld_set_version(struct mld_ifinfo *mli, const int version)
1622{
1623 int old_version_timer;
1624
1625 MLD_LOCK_ASSERT();
1626
1627 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1628 version, mli->mli_ifp, mli->mli_ifp->if_xname);
1629
1630 if (version == MLD_VERSION_1) {
1631 /*
1632 * Compute the "Older Version Querier Present" timer as per
1633 * Section 9.12.
1634 */
1635 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1636 old_version_timer *= PR_SLOWHZ;
1637 mli->mli_v1_timer = old_version_timer;
1638 }
1639
1640 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1641 mli->mli_version = MLD_VERSION_1;
1642 mld_v2_cancel_link_timers(mli);
1643 }
1644}
1645
1646/*
1647 * Cancel pending MLDv2 timers for the given link and all groups
1648 * joined on it; state-change, general-query, and group-query timers.
1649 */
1650static void
1651mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
1652{
1653 struct ifmultiaddr *ifma;
1654 struct ifnet *ifp;
1655 struct in6_multi *inm;
1656
1657 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1658 mli->mli_ifp, mli->mli_ifp->if_xname);
1659
1660 IN6_MULTI_LOCK_ASSERT();
1661 MLD_LOCK_ASSERT();
1662
1663 /*
1664 * Fast-track this potentially expensive operation
1665 * by checking all the global 'timer pending' flags.
1666 */
1667 if (!V_interface_timers_running6 &&
1668 !V_state_change_timers_running6 &&
1669 !V_current_state_timers_running6)
1670 return;
1671
1672 mli->mli_v2_timer = 0;
1673
1674 ifp = mli->mli_ifp;
1675
1676 IF_ADDR_LOCK(ifp);
1677 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1678 if (ifma->ifma_addr->sa_family != AF_INET6)
1679 continue;
1680 inm = (struct in6_multi *)ifma->ifma_protospec;
1681 switch (inm->in6m_state) {
1682 case MLD_NOT_MEMBER:
1683 case MLD_SILENT_MEMBER:
1684 case MLD_IDLE_MEMBER:
1685 case MLD_LAZY_MEMBER:
1686 case MLD_SLEEPING_MEMBER:
1687 case MLD_AWAKENING_MEMBER:
1688 break;
1689 case MLD_LEAVING_MEMBER:
1690 /*
1691 * If we are leaving the group and switching
1692 * version, we need to release the final
1693 * reference held for issuing the INCLUDE {}.
1694 *
1695 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
1696 * around in6m_release_locked(), as it is not
1697 * a recursive mutex.
1698 */
1699 IF_ADDR_UNLOCK(ifp);
1700 in6m_release_locked(inm);
1701 IF_ADDR_LOCK(ifp);
1702 /* FALLTHROUGH */
1703 case MLD_G_QUERY_PENDING_MEMBER:
1704 case MLD_SG_QUERY_PENDING_MEMBER:
1705 in6m_clear_recorded(inm);
1706 /* FALLTHROUGH */
1707 case MLD_REPORTING_MEMBER:
1708 inm->in6m_sctimer = 0;
1709 inm->in6m_timer = 0;
1710 inm->in6m_state = MLD_REPORTING_MEMBER;
1711 /*
1712 * Free any pending MLDv2 state-change records.
1713 */
1714 _IF_DRAIN(&inm->in6m_scq);
1715 break;
1716 }
1717 }
1718 IF_ADDR_UNLOCK(ifp);
1719}
1720
1721/*
1722 * Global slowtimo handler.
1723 * VIMAGE: Timeout handlers are expected to service all vimages.
1724 */
1725void
1726mld_slowtimo(void)
1727{
1728 VNET_ITERATOR_DECL(vnet_iter);
1729
1730 VNET_LIST_RLOCK_NOSLEEP();
1731 VNET_FOREACH(vnet_iter) {
1732 CURVNET_SET(vnet_iter);
1733 mld_slowtimo_vnet();
1734 CURVNET_RESTORE();
1735 }
1736 VNET_LIST_RUNLOCK_NOSLEEP();
1737}
1738
1739/*
1740 * Per-vnet slowtimo handler.
1741 */
1742static void
1743mld_slowtimo_vnet(void)
1744{
1745 struct mld_ifinfo *mli;
1746
1747 MLD_LOCK();
1748
1749 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1750 mld_v1_process_querier_timers(mli);
1751 }
1752
1753 MLD_UNLOCK();
1754}
1755
1756/*
1757 * Update the Older Version Querier Present timers for a link.
1758 * See Section 9.12 of RFC 3810.
1759 */
1760static void
1761mld_v1_process_querier_timers(struct mld_ifinfo *mli)
1762{
1763
1764 MLD_LOCK_ASSERT();
1765
1766 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1767 /*
1768 * MLDv1 Querier Present timer expired; revert to MLDv2.
1769 */
1770 CTR5(KTR_MLD,
1771 "%s: transition from v%d -> v%d on %p(%s)",
1772 __func__, mli->mli_version, MLD_VERSION_2,
1773 mli->mli_ifp, mli->mli_ifp->if_xname);
1774 mli->mli_version = MLD_VERSION_2;
1775 }
1776}
1777
1778/*
1779 * Transmit an MLDv1 report immediately.
1780 */
1781static int
1782mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1783{
1784 struct ifnet *ifp;
1785 struct in6_ifaddr *ia;
1786 struct ip6_hdr *ip6;
1787 struct mbuf *mh, *md;
1788 struct mld_hdr *mld;
1789
1790 IN6_MULTI_LOCK_ASSERT();
1791 MLD_LOCK_ASSERT();
1792
1793 ifp = in6m->in6m_ifp;
1794 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1795 /* ia may be NULL if link-local address is tentative. */
1796
1797 MGETHDR(mh, M_DONTWAIT, MT_HEADER);
1798 if (mh == NULL) {
1799 if (ia != NULL)
1800 ifa_free(&ia->ia_ifa);
1801 return (ENOMEM);
1802 }
1803 MGET(md, M_DONTWAIT, MT_DATA);
1804 if (md == NULL) {
1805 m_free(mh);
1806 if (ia != NULL)
1807 ifa_free(&ia->ia_ifa);
1808 return (ENOMEM);
1809 }
1810 mh->m_next = md;
1811
1812 /*
1813 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1814 * that ether_output() does not need to allocate another mbuf
1815 * for the header in the most common case.
1816 */
1817 MH_ALIGN(mh, sizeof(struct ip6_hdr));
1818 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1819 mh->m_len = sizeof(struct ip6_hdr);
1820
1821 ip6 = mtod(mh, struct ip6_hdr *);
1822 ip6->ip6_flow = 0;
1823 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1824 ip6->ip6_vfc |= IPV6_VERSION;
1825 ip6->ip6_nxt = IPPROTO_ICMPV6;
1826 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1827 ip6->ip6_dst = in6m->in6m_addr;
1828
1829 md->m_len = sizeof(struct mld_hdr);
1830 mld = mtod(md, struct mld_hdr *);
1831 mld->mld_type = type;
1832 mld->mld_code = 0;
1833 mld->mld_cksum = 0;
1834 mld->mld_maxdelay = 0;
1835 mld->mld_reserved = 0;
1836 mld->mld_addr = in6m->in6m_addr;
1837 in6_clearscope(&mld->mld_addr);
1838 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1839 sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1840
1841 mld_save_context(mh, ifp);
1842 mh->m_flags |= M_MLDV1;
1843
1844 mld_dispatch_packet(mh);
1845
1846 if (ia != NULL)
1847 ifa_free(&ia->ia_ifa);
1848 return (0);
1849}
1850
1851/*
1852 * Process a state change from the upper layer for the given IPv6 group.
1853 *
1854 * Each socket holds a reference on the in_multi in its own ip_moptions.
1855 * The socket layer will have made the necessary updates to.the group
1856 * state, it is now up to MLD to issue a state change report if there
1857 * has been any change between T0 (when the last state-change was issued)
1858 * and T1 (now).
1859 *
1860 * We use the MLDv2 state machine at group level. The MLd module
1861 * however makes the decision as to which MLD protocol version to speak.
1862 * A state change *from* INCLUDE {} always means an initial join.
1863 * A state change *to* INCLUDE {} always means a final leave.
1864 *
1865 * If delay is non-zero, and the state change is an initial multicast
1866 * join, the state change report will be delayed by 'delay' ticks
1867 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1868 * the initial MLDv2 state change report will be delayed by whichever
1869 * is sooner, a pending state-change timer or delay itself.
1870 *
1871 * VIMAGE: curvnet should have been set by caller, as this routine
1872 * is called from the socket option handlers.
1873 */
1874int
1875mld_change_state(struct in6_multi *inm, const int delay)
1876{
1877 struct mld_ifinfo *mli;
1878 struct ifnet *ifp;
1879 int error;
1880
1881 IN6_MULTI_LOCK_ASSERT();
1882
1883 error = 0;
1884
1885 /*
1886 * Try to detect if the upper layer just asked us to change state
1887 * for an interface which has now gone away.
1888 */
1889 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1890 ifp = inm->in6m_ifma->ifma_ifp;
1891 if (ifp != NULL) {
1892 /*
1893 * Sanity check that netinet6's notion of ifp is the
1894 * same as net's.
1895 */
1896 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1897 }
1898
1899 MLD_LOCK();
1900
1901 mli = MLD_IFINFO(ifp);
1902 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
1903
1904 /*
1905 * If we detect a state transition to or from MCAST_UNDEFINED
1906 * for this group, then we are starting or finishing an MLD
1907 * life cycle for this group.
1908 */
1909 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1910 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1911 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1912 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1913 CTR1(KTR_MLD, "%s: initial join", __func__);
1914 error = mld_initial_join(inm, mli, delay);
1915 goto out_locked;
1916 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1917 CTR1(KTR_MLD, "%s: final leave", __func__);
1918 mld_final_leave(inm, mli);
1919 goto out_locked;
1920 }
1921 } else {
1922 CTR1(KTR_MLD, "%s: filter set change", __func__);
1923 }
1924
1925 error = mld_handle_state_change(inm, mli);
1926
1927out_locked:
1928 MLD_UNLOCK();
1929 return (error);
1930}
1931
1932/*
1933 * Perform the initial join for an MLD group.
1934 *
1935 * When joining a group:
1936 * If the group should have its MLD traffic suppressed, do nothing.
1937 * MLDv1 starts sending MLDv1 host membership reports.
1938 * MLDv2 will schedule an MLDv2 state-change report containing the
1939 * initial state of the membership.
1940 *
1941 * If the delay argument is non-zero, then we must delay sending the
1942 * initial state change for delay ticks (in units of PR_FASTHZ).
1943 */
1944static int
1945mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
1946 const int delay)
1947{
1948 struct ifnet *ifp;
1949 struct ifqueue *ifq;
1950 int error, retval, syncstates;
1951 int odelay;
1952#ifdef KTR
1953 char ip6tbuf[INET6_ADDRSTRLEN];
1954#endif
1955
1956 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1957 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1958 inm->in6m_ifp, inm->in6m_ifp->if_xname);
1959
1960 error = 0;
1961 syncstates = 1;
1962
1963 ifp = inm->in6m_ifp;
1964
1965 IN6_MULTI_LOCK_ASSERT();
1966 MLD_LOCK_ASSERT();
1967
1968 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1969
1970 /*
1971 * Groups joined on loopback or marked as 'not reported',
1972 * enter the MLD_SILENT_MEMBER state and
1973 * are never reported in any protocol exchanges.
1974 * All other groups enter the appropriate state machine
1975 * for the version in use on this link.
1976 * A link marked as MLIF_SILENT causes MLD to be completely
1977 * disabled for the link.
1978 */
1979 if ((ifp->if_flags & IFF_LOOPBACK) ||
1980 (mli->mli_flags & MLIF_SILENT) ||
1981 !mld_is_addr_reported(&inm->in6m_addr)) {
1982 CTR1(KTR_MLD,
1983"%s: not kicking state machine for silent group", __func__);
1984 inm->in6m_state = MLD_SILENT_MEMBER;
1985 inm->in6m_timer = 0;
1986 } else {
1987 /*
1988 * Deal with overlapping in_multi lifecycle.
1989 * If this group was LEAVING, then make sure
1990 * we drop the reference we picked up to keep the
1991 * group around for the final INCLUDE {} enqueue.
1992 */
1993 if (mli->mli_version == MLD_VERSION_2 &&
1994 inm->in6m_state == MLD_LEAVING_MEMBER)
1995 in6m_release_locked(inm);
1996
1997 inm->in6m_state = MLD_REPORTING_MEMBER;
1998
1999 switch (mli->mli_version) {
2000 case MLD_VERSION_1:
2001 /*
2002 * If a delay was provided, only use it if
2003 * it is greater than the delay normally
2004 * used for an MLDv1 state change report,
2005 * and delay sending the initial MLDv1 report
2006 * by not transitioning to the IDLE state.
2007 */
2008 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2009 if (delay) {
2010 inm->in6m_timer = max(delay, odelay);
2011 V_current_state_timers_running6 = 1;
2012 } else {
2013 inm->in6m_state = MLD_IDLE_MEMBER;
2014 error = mld_v1_transmit_report(inm,
2015 MLD_LISTENER_REPORT);
2016 if (error == 0) {
2017 inm->in6m_timer = odelay;
2018 V_current_state_timers_running6 = 1;
2019 }
2020 }
2021 break;
2022
2023 case MLD_VERSION_2:
2024 /*
2025 * Defer update of T0 to T1, until the first copy
2026 * of the state change has been transmitted.
2027 */
2028 syncstates = 0;
2029
2030 /*
2031 * Immediately enqueue a State-Change Report for
2032 * this interface, freeing any previous reports.
2033 * Don't kick the timers if there is nothing to do,
2034 * or if an error occurred.
2035 */
2036 ifq = &inm->in6m_scq;
2037 _IF_DRAIN(ifq);
2038 retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2039 0, 0, (mli->mli_flags & MLIF_USEALLOW));
2040 CTR2(KTR_MLD, "%s: enqueue record = %d",
2041 __func__, retval);
2042 if (retval <= 0) {
2043 error = retval * -1;
2044 break;
2045 }
2046
2047 /*
2048 * Schedule transmission of pending state-change
2049 * report up to RV times for this link. The timer
2050 * will fire at the next mld_fasttimo (~200ms),
2051 * giving us an opportunity to merge the reports.
2052 *
2053 * If a delay was provided to this function, only
2054 * use this delay if sooner than the existing one.
2055 */
2056 KASSERT(mli->mli_rv > 1,
2057 ("%s: invalid robustness %d", __func__,
2058 mli->mli_rv));
2059 inm->in6m_scrv = mli->mli_rv;
2060 if (delay) {
2061 if (inm->in6m_sctimer > 1) {
2062 inm->in6m_sctimer =
2063 min(inm->in6m_sctimer, delay);
2064 } else
2065 inm->in6m_sctimer = delay;
2066 } else
2067 inm->in6m_sctimer = 1;
2068 V_state_change_timers_running6 = 1;
2069
2070 error = 0;
2071 break;
2072 }
2073 }
2074
2075 /*
2076 * Only update the T0 state if state change is atomic,
2077 * i.e. we don't need to wait for a timer to fire before we
2078 * can consider the state change to have been communicated.
2079 */
2080 if (syncstates) {
2081 in6m_commit(inm);
2082 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2083 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2084 inm->in6m_ifp->if_xname);
2085 }
2086
2087 return (error);
2088}
2089
2090/*
2091 * Issue an intermediate state change during the life-cycle.
2092 */
2093static int
2094mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli)
2095{
2096 struct ifnet *ifp;
2097 int retval;
2098#ifdef KTR
2099 char ip6tbuf[INET6_ADDRSTRLEN];
2100#endif
2101
2102 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2103 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2104 inm->in6m_ifp, inm->in6m_ifp->if_xname);
2105
2106 ifp = inm->in6m_ifp;
2107
2108 IN6_MULTI_LOCK_ASSERT();
2109 MLD_LOCK_ASSERT();
2110
2111 KASSERT(mli && mli->mli_ifp == ifp,
2112 ("%s: inconsistent ifp", __func__));
2113
2114 if ((ifp->if_flags & IFF_LOOPBACK) ||
2115 (mli->mli_flags & MLIF_SILENT) ||
2116 !mld_is_addr_reported(&inm->in6m_addr) ||
2117 (mli->mli_version != MLD_VERSION_2)) {
2118 if (!mld_is_addr_reported(&inm->in6m_addr)) {
2119 CTR1(KTR_MLD,
2120"%s: not kicking state machine for silent group", __func__);
2121 }
2122 CTR1(KTR_MLD, "%s: nothing to do", __func__);
2123 in6m_commit(inm);
2124 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2125 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2126 inm->in6m_ifp->if_xname);
2127 return (0);
2128 }
2129
2130 _IF_DRAIN(&inm->in6m_scq);
2131
2132 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2133 (mli->mli_flags & MLIF_USEALLOW));
2134 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2135 if (retval <= 0)
2136 return (-retval);
2137
2138 /*
2139 * If record(s) were enqueued, start the state-change
2140 * report timer for this group.
2141 */
2142 inm->in6m_scrv = mli->mli_rv;
2143 inm->in6m_sctimer = 1;
2144 V_state_change_timers_running6 = 1;
2145
2146 return (0);
2147}
2148
2149/*
2150 * Perform the final leave for a multicast address.
2151 *
2152 * When leaving a group:
2153 * MLDv1 sends a DONE message, if and only if we are the reporter.
2154 * MLDv2 enqueues a state-change report containing a transition
2155 * to INCLUDE {} for immediate transmission.
2156 */
2157static void
2158mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli)
2159{
2160 int syncstates;
2161#ifdef KTR
2162 char ip6tbuf[INET6_ADDRSTRLEN];
2163#endif
2164
2165 syncstates = 1;
2166
2167 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2168 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2169 inm->in6m_ifp, inm->in6m_ifp->if_xname);
2170
2171 IN6_MULTI_LOCK_ASSERT();
2172 MLD_LOCK_ASSERT();
2173
2174 switch (inm->in6m_state) {
2175 case MLD_NOT_MEMBER:
2176 case MLD_SILENT_MEMBER:
2177 case MLD_LEAVING_MEMBER:
2178 /* Already leaving or left; do nothing. */
2179 CTR1(KTR_MLD,
2180"%s: not kicking state machine for silent group", __func__);
2181 break;
2182 case MLD_REPORTING_MEMBER:
2183 case MLD_IDLE_MEMBER:
2184 case MLD_G_QUERY_PENDING_MEMBER:
2185 case MLD_SG_QUERY_PENDING_MEMBER:
2186 if (mli->mli_version == MLD_VERSION_1) {
2187#ifdef INVARIANTS
2188 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2189 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2190 panic("%s: MLDv2 state reached, not MLDv2 mode",
2191 __func__);
2192#endif
2193 mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2194 inm->in6m_state = MLD_NOT_MEMBER;
2195 } else if (mli->mli_version == MLD_VERSION_2) {
2196 /*
2197 * Stop group timer and all pending reports.
2198 * Immediately enqueue a state-change report
2199 * TO_IN {} to be sent on the next fast timeout,
2200 * giving us an opportunity to merge reports.
2201 */
2202 _IF_DRAIN(&inm->in6m_scq);
2203 inm->in6m_timer = 0;
2204 inm->in6m_scrv = mli->mli_rv;
2205 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2206 "pending retransmissions.", __func__,
2207 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2208 inm->in6m_ifp->if_xname, inm->in6m_scrv);
2209 if (inm->in6m_scrv == 0) {
2210 inm->in6m_state = MLD_NOT_MEMBER;
2211 inm->in6m_sctimer = 0;
2212 } else {
2213 int retval;
2214
2215 in6m_acquire_locked(inm);
2216
2217 retval = mld_v2_enqueue_group_record(
2218 &inm->in6m_scq, inm, 1, 0, 0,
2219 (mli->mli_flags & MLIF_USEALLOW));
2220 KASSERT(retval != 0,
2221 ("%s: enqueue record = %d", __func__,
2222 retval));
2223
2224 inm->in6m_state = MLD_LEAVING_MEMBER;
2225 inm->in6m_sctimer = 1;
2226 V_state_change_timers_running6 = 1;
2227 syncstates = 0;
2228 }
2229 break;
2230 }
2231 break;
2232 case MLD_LAZY_MEMBER:
2233 case MLD_SLEEPING_MEMBER:
2234 case MLD_AWAKENING_MEMBER:
2235 /* Our reports are suppressed; do nothing. */
2236 break;
2237 }
2238
2239 if (syncstates) {
2240 in6m_commit(inm);
2241 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2242 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2243 inm->in6m_ifp->if_xname);
2244 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2245 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2246 __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname);
2247 }
2248}
2249
2250/*
2251 * Enqueue an MLDv2 group record to the given output queue.
2252 *
2253 * If is_state_change is zero, a current-state record is appended.
2254 * If is_state_change is non-zero, a state-change report is appended.
2255 *
2256 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2257 * If is_group_query is zero, and if there is a packet with free space
2258 * at the tail of the queue, it will be appended to providing there
2259 * is enough free space.
2260 * Otherwise a new mbuf packet chain is allocated.
2261 *
2262 * If is_source_query is non-zero, each source is checked to see if
2263 * it was recorded for a Group-Source query, and will be omitted if
2264 * it is not both in-mode and recorded.
2265 *
2266 * If use_block_allow is non-zero, state change reports for initial join
2267 * and final leave, on an inclusive mode group with a source list, will be
2268 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2269 *
2270 * The function will attempt to allocate leading space in the packet
2271 * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2272 *
2273 * If successful the size of all data appended to the queue is returned,
2274 * otherwise an error code less than zero is returned, or zero if
2275 * no record(s) were appended.
2276 */
2277static int
2278mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2279 const int is_state_change, const int is_group_query,
2280 const int is_source_query, const int use_block_allow)
2281{
2282 struct mldv2_record mr;
2283 struct mldv2_record *pmr;
2284 struct ifnet *ifp;
2285 struct ip6_msource *ims, *nims;
2286 struct mbuf *m0, *m, *md;
2287 int error, is_filter_list_change;
2288 int minrec0len, m0srcs, msrcs, nbytes, off;
2289 int record_has_sources;
2290 int now;
2291 int type;
2292 uint8_t mode;
2293#ifdef KTR
2294 char ip6tbuf[INET6_ADDRSTRLEN];
2295#endif
2296
2297 IN6_MULTI_LOCK_ASSERT();
2298
2299 error = 0;
2300 ifp = inm->in6m_ifp;
2301 is_filter_list_change = 0;
2302 m = NULL;
2303 m0 = NULL;
2304 m0srcs = 0;
2305 msrcs = 0;
2306 nbytes = 0;
2307 nims = NULL;
2308 record_has_sources = 1;
2309 pmr = NULL;
2310 type = MLD_DO_NOTHING;
2311 mode = inm->in6m_st[1].iss_fmode;
2312
2313 /*
2314 * If we did not transition out of ASM mode during t0->t1,
2315 * and there are no source nodes to process, we can skip
2316 * the generation of source records.
2317 */
2318 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2319 inm->in6m_nsrc == 0)
2320 record_has_sources = 0;
2321
2322 if (is_state_change) {
2323 /*
2324 * Queue a state change record.
2325 * If the mode did not change, and there are non-ASM
2326 * listeners or source filters present,
2327 * we potentially need to issue two records for the group.
2328 * If there are ASM listeners, and there was no filter
2329 * mode transition of any kind, do nothing.
2330 *
2331 * If we are transitioning to MCAST_UNDEFINED, we need
2332 * not send any sources. A transition to/from this state is
2333 * considered inclusive with some special treatment.
2334 *
2335 * If we are rewriting initial joins/leaves to use
2336 * ALLOW/BLOCK, and the group's membership is inclusive,
2337 * we need to send sources in all cases.
2338 */
2339 if (mode != inm->in6m_st[0].iss_fmode) {
2340 if (mode == MCAST_EXCLUDE) {
2341 CTR1(KTR_MLD, "%s: change to EXCLUDE",
2342 __func__);
2343 type = MLD_CHANGE_TO_EXCLUDE_MODE;
2344 } else {
2345 CTR1(KTR_MLD, "%s: change to INCLUDE",
2346 __func__);
2347 if (use_block_allow) {
2348 /*
2349 * XXX
2350 * Here we're interested in state
2351 * edges either direction between
2352 * MCAST_UNDEFINED and MCAST_INCLUDE.
2353 * Perhaps we should just check
2354 * the group state, rather than
2355 * the filter mode.
2356 */
2357 if (mode == MCAST_UNDEFINED) {
2358 type = MLD_BLOCK_OLD_SOURCES;
2359 } else {
2360 type = MLD_ALLOW_NEW_SOURCES;
2361 }
2362 } else {
2363 type = MLD_CHANGE_TO_INCLUDE_MODE;
2364 if (mode == MCAST_UNDEFINED)
2365 record_has_sources = 0;
2366 }
2367 }
2368 } else {
2369 if (record_has_sources) {
2370 is_filter_list_change = 1;
2371 } else {
2372 type = MLD_DO_NOTHING;
2373 }
2374 }
2375 } else {
2376 /*
2377 * Queue a current state record.
2378 */
2379 if (mode == MCAST_EXCLUDE) {
2380 type = MLD_MODE_IS_EXCLUDE;
2381 } else if (mode == MCAST_INCLUDE) {
2382 type = MLD_MODE_IS_INCLUDE;
2383 KASSERT(inm->in6m_st[1].iss_asm == 0,
2384 ("%s: inm %p is INCLUDE but ASM count is %d",
2385 __func__, inm, inm->in6m_st[1].iss_asm));
2386 }
2387 }
2388
2389 /*
2390 * Generate the filter list changes using a separate function.
2391 */
2392 if (is_filter_list_change)
2393 return (mld_v2_enqueue_filter_change(ifq, inm));
2394
2395 if (type == MLD_DO_NOTHING) {
2396 CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2397 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2398 inm->in6m_ifp->if_xname);
2399 return (0);
2400 }
2401
2402 /*
2403 * If any sources are present, we must be able to fit at least
2404 * one in the trailing space of the tail packet's mbuf,
2405 * ideally more.
2406 */
2407 minrec0len = sizeof(struct mldv2_record);
2408 if (record_has_sources)
2409 minrec0len += sizeof(struct in6_addr);
2410
2411 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2412 mld_rec_type_to_str(type),
2413 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2414 inm->in6m_ifp->if_xname);
2415
2416 /*
2417 * Check if we have a packet in the tail of the queue for this
2418 * group into which the first group record for this group will fit.
2419 * Otherwise allocate a new packet.
2420 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2421 * Note: Group records for G/GSR query responses MUST be sent
2422 * in their own packet.
2423 */
2424 m0 = ifq->ifq_tail;
2425 if (!is_group_query &&
2426 m0 != NULL &&
2427 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2428 (m0->m_pkthdr.len + minrec0len) <
2429 (ifp->if_mtu - MLD_MTUSPACE)) {
2430 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2431 sizeof(struct mldv2_record)) /
2432 sizeof(struct in6_addr);
2433 m = m0;
2434 CTR1(KTR_MLD, "%s: use existing packet", __func__);
2435 } else {
2436 if (_IF_QFULL(ifq)) {
2437 CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2438 return (-ENOMEM);
2439 }
2440 m = NULL;
2441 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2442 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2443 if (!is_state_change && !is_group_query)
2444 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2445 if (m == NULL)
2446 m = m_gethdr(M_DONTWAIT, MT_DATA);
2447 if (m == NULL)
2448 return (-ENOMEM);
2449
2450 mld_save_context(m, ifp);
2451
2452 CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2453 }
2454
2455 /*
2456 * Append group record.
2457 * If we have sources, we don't know how many yet.
2458 */
2459 mr.mr_type = type;
2460 mr.mr_datalen = 0;
2461 mr.mr_numsrc = 0;
2462 mr.mr_addr = inm->in6m_addr;
2463 in6_clearscope(&mr.mr_addr);
2464 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2465 if (m != m0)
2466 m_freem(m);
2467 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2468 return (-ENOMEM);
2469 }
2470 nbytes += sizeof(struct mldv2_record);
2471
2472 /*
2473 * Append as many sources as will fit in the first packet.
2474 * If we are appending to a new packet, the chain allocation
2475 * may potentially use clusters; use m_getptr() in this case.
2476 * If we are appending to an existing packet, we need to obtain
2477 * a pointer to the group record after m_append(), in case a new
2478 * mbuf was allocated.
2479 *
2480 * Only append sources which are in-mode at t1. If we are
2481 * transitioning to MCAST_UNDEFINED state on the group, and
2482 * use_block_allow is zero, do not include source entries.
2483 * Otherwise, we need to include this source in the report.
2484 *
2485 * Only report recorded sources in our filter set when responding
2486 * to a group-source query.
2487 */
2488 if (record_has_sources) {
2489 if (m == m0) {
2490 md = m_last(m);
2491 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2492 md->m_len - nbytes);
2493 } else {
2494 md = m_getptr(m, 0, &off);
2495 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2496 off);
2497 }
2498 msrcs = 0;
2499 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2500 nims) {
2501 CTR2(KTR_MLD, "%s: visit node %s", __func__,
2502 ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2503 now = im6s_get_mode(inm, ims, 1);
2504 CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2505 if ((now != mode) ||
2506 (now == mode &&
2507 (!use_block_allow && mode == MCAST_UNDEFINED))) {
2508 CTR1(KTR_MLD, "%s: skip node", __func__);
2509 continue;
2510 }
2511 if (is_source_query && ims->im6s_stp == 0) {
2512 CTR1(KTR_MLD, "%s: skip unrecorded node",
2513 __func__);
2514 continue;
2515 }
2516 CTR1(KTR_MLD, "%s: append node", __func__);
2517 if (!m_append(m, sizeof(struct in6_addr),
2518 (void *)&ims->im6s_addr)) {
2519 if (m != m0)
2520 m_freem(m);
2521 CTR1(KTR_MLD, "%s: m_append() failed.",
2522 __func__);
2523 return (-ENOMEM);
2524 }
2525 nbytes += sizeof(struct in6_addr);
2526 ++msrcs;
2527 if (msrcs == m0srcs)
2528 break;
2529 }
2530 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2531 msrcs);
2532 pmr->mr_numsrc = htons(msrcs);
2533 nbytes += (msrcs * sizeof(struct in6_addr));
2534 }
2535
2536 if (is_source_query && msrcs == 0) {
2537 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2538 if (m != m0)
2539 m_freem(m);
2540 return (0);
2541 }
2542
2543 /*
2544 * We are good to go with first packet.
2545 */
2546 if (m != m0) {
2547 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2548 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2549 _IF_ENQUEUE(ifq, m);
2550 } else
2551 m->m_pkthdr.PH_vt.vt_nrecs++;
2552
2553 /*
2554 * No further work needed if no source list in packet(s).
2555 */
2556 if (!record_has_sources)
2557 return (nbytes);
2558
2559 /*
2560 * Whilst sources remain to be announced, we need to allocate
2561 * a new packet and fill out as many sources as will fit.
2562 * Always try for a cluster first.
2563 */
2564 while (nims != NULL) {
2565 if (_IF_QFULL(ifq)) {
2566 CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2567 return (-ENOMEM);
2568 }
2569 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2570 if (m == NULL)
2571 m = m_gethdr(M_DONTWAIT, MT_DATA);
2572 if (m == NULL)
2573 return (-ENOMEM);
2574 mld_save_context(m, ifp);
2575 md = m_getptr(m, 0, &off);
2576 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2577 CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2578
2579 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2580 if (m != m0)
2581 m_freem(m);
2582 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2583 return (-ENOMEM);
2584 }
2585 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2586 nbytes += sizeof(struct mldv2_record);
2587
2588 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2589 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2590
2591 msrcs = 0;
2592 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2593 CTR2(KTR_MLD, "%s: visit node %s",
2594 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2595 now = im6s_get_mode(inm, ims, 1);
2596 if ((now != mode) ||
2597 (now == mode &&
2598 (!use_block_allow && mode == MCAST_UNDEFINED))) {
2599 CTR1(KTR_MLD, "%s: skip node", __func__);
2600 continue;
2601 }
2602 if (is_source_query && ims->im6s_stp == 0) {
2603 CTR1(KTR_MLD, "%s: skip unrecorded node",
2604 __func__);
2605 continue;
2606 }
2607 CTR1(KTR_MLD, "%s: append node", __func__);
2608 if (!m_append(m, sizeof(struct in6_addr),
2609 (void *)&ims->im6s_addr)) {
2610 if (m != m0)
2611 m_freem(m);
2612 CTR1(KTR_MLD, "%s: m_append() failed.",
2613 __func__);
2614 return (-ENOMEM);
2615 }
2616 ++msrcs;
2617 if (msrcs == m0srcs)
2618 break;
2619 }
2620 pmr->mr_numsrc = htons(msrcs);
2621 nbytes += (msrcs * sizeof(struct in6_addr));
2622
2623 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2624 _IF_ENQUEUE(ifq, m);
2625 }
2626
2627 return (nbytes);
2628}
2629
2630/*
2631 * Type used to mark record pass completion.
2632 * We exploit the fact we can cast to this easily from the
2633 * current filter modes on each ip_msource node.
2634 */
2635typedef enum {
2636 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2637 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2638 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2639 REC_FULL = REC_ALLOW | REC_BLOCK
2640} rectype_t;
2641
2642/*
2643 * Enqueue an MLDv2 filter list change to the given output queue.
2644 *
2645 * Source list filter state is held in an RB-tree. When the filter list
2646 * for a group is changed without changing its mode, we need to compute
2647 * the deltas between T0 and T1 for each source in the filter set,
2648 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2649 *
2650 * As we may potentially queue two record types, and the entire R-B tree
2651 * needs to be walked at once, we break this out into its own function
2652 * so we can generate a tightly packed queue of packets.
2653 *
2654 * XXX This could be written to only use one tree walk, although that makes
2655 * serializing into the mbuf chains a bit harder. For now we do two walks
2656 * which makes things easier on us, and it may or may not be harder on
2657 * the L2 cache.
2658 *
2659 * If successful the size of all data appended to the queue is returned,
2660 * otherwise an error code less than zero is returned, or zero if
2661 * no record(s) were appended.
2662 */
2663static int
2664mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
2665{
2666 static const int MINRECLEN =
2667 sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2668 struct ifnet *ifp;
2669 struct mldv2_record mr;
2670 struct mldv2_record *pmr;
2671 struct ip6_msource *ims, *nims;
2672 struct mbuf *m, *m0, *md;
2673 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2674 int nallow, nblock;
2675 uint8_t mode, now, then;
2676 rectype_t crt, drt, nrt;
2677#ifdef KTR
2678 char ip6tbuf[INET6_ADDRSTRLEN];
2679#endif
2680
2681 IN6_MULTI_LOCK_ASSERT();
2682
2683 if (inm->in6m_nsrc == 0 ||
2684 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2685 return (0);
2686
2687 ifp = inm->in6m_ifp; /* interface */
2688 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */
2689 crt = REC_NONE; /* current group record type */
2690 drt = REC_NONE; /* mask of completed group record types */
2691 nrt = REC_NONE; /* record type for current node */
2692 m0srcs = 0; /* # source which will fit in current mbuf chain */
2693 npbytes = 0; /* # of bytes appended this packet */
2694 nbytes = 0; /* # of bytes appended to group's state-change queue */
2695 rsrcs = 0; /* # sources encoded in current record */
2696 schanged = 0; /* # nodes encoded in overall filter change */
2697 nallow = 0; /* # of source entries in ALLOW_NEW */
2698 nblock = 0; /* # of source entries in BLOCK_OLD */
2699 nims = NULL; /* next tree node pointer */
2700
2701 /*
2702 * For each possible filter record mode.
2703 * The first kind of source we encounter tells us which
2704 * is the first kind of record we start appending.
2705 * If a node transitioned to UNDEFINED at t1, its mode is treated
2706 * as the inverse of the group's filter mode.
2707 */
2708 while (drt != REC_FULL) {
2709 do {
2710 m0 = ifq->ifq_tail;
2711 if (m0 != NULL &&
2712 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2713 MLD_V2_REPORT_MAXRECS) &&
2714 (m0->m_pkthdr.len + MINRECLEN) <
2715 (ifp->if_mtu - MLD_MTUSPACE)) {
2716 m = m0;
2717 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2718 sizeof(struct mldv2_record)) /
2719 sizeof(struct in6_addr);
2720 CTR1(KTR_MLD,
2721 "%s: use previous packet", __func__);
2722 } else {
2723 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2724 if (m == NULL)
2725 m = m_gethdr(M_DONTWAIT, MT_DATA);
2726 if (m == NULL) {
2727 CTR1(KTR_MLD,
2728 "%s: m_get*() failed", __func__);
2729 return (-ENOMEM);
2730 }
2731 m->m_pkthdr.PH_vt.vt_nrecs = 0;
2732 mld_save_context(m, ifp);
2733 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2734 sizeof(struct mldv2_record)) /
2735 sizeof(struct in6_addr);
2736 npbytes = 0;
2737 CTR1(KTR_MLD,
2738 "%s: allocated new packet", __func__);
2739 }
2740 /*
2741 * Append the MLD group record header to the
2742 * current packet's data area.
2743 * Recalculate pointer to free space for next
2744 * group record, in case m_append() allocated
2745 * a new mbuf or cluster.
2746 */
2747 memset(&mr, 0, sizeof(mr));
2748 mr.mr_addr = inm->in6m_addr;
2749 in6_clearscope(&mr.mr_addr);
2750 if (!m_append(m, sizeof(mr), (void *)&mr)) {
2751 if (m != m0)
2752 m_freem(m);
2753 CTR1(KTR_MLD,
2754 "%s: m_append() failed", __func__);
2755 return (-ENOMEM);
2756 }
2757 npbytes += sizeof(struct mldv2_record);
2758 if (m != m0) {
2759 /* new packet; offset in chain */
2760 md = m_getptr(m, npbytes -
2761 sizeof(struct mldv2_record), &off);
2762 pmr = (struct mldv2_record *)(mtod(md,
2763 uint8_t *) + off);
2764 } else {
2765 /* current packet; offset from last append */
2766 md = m_last(m);
2767 pmr = (struct mldv2_record *)(mtod(md,
2768 uint8_t *) + md->m_len -
2769 sizeof(struct mldv2_record));
2770 }
2771 /*
2772 * Begin walking the tree for this record type
2773 * pass, or continue from where we left off
2774 * previously if we had to allocate a new packet.
2775 * Only report deltas in-mode at t1.
2776 * We need not report included sources as allowed
2777 * if we are in inclusive mode on the group,
2778 * however the converse is not true.
2779 */
2780 rsrcs = 0;
2781 if (nims == NULL) {
2782 nims = RB_MIN(ip6_msource_tree,
2783 &inm->in6m_srcs);
2784 }
2785 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2786 CTR2(KTR_MLD, "%s: visit node %s", __func__,
2787 ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2788 now = im6s_get_mode(inm, ims, 1);
2789 then = im6s_get_mode(inm, ims, 0);
2790 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2791 __func__, then, now);
2792 if (now == then) {
2793 CTR1(KTR_MLD,
2794 "%s: skip unchanged", __func__);
2795 continue;
2796 }
2797 if (mode == MCAST_EXCLUDE &&
2798 now == MCAST_INCLUDE) {
2799 CTR1(KTR_MLD,
2800 "%s: skip IN src on EX group",
2801 __func__);
2802 continue;
2803 }
2804 nrt = (rectype_t)now;
2805 if (nrt == REC_NONE)
2806 nrt = (rectype_t)(~mode & REC_FULL);
2807 if (schanged++ == 0) {
2808 crt = nrt;
2809 } else if (crt != nrt)
2810 continue;
2811 if (!m_append(m, sizeof(struct in6_addr),
2812 (void *)&ims->im6s_addr)) {
2813 if (m != m0)
2814 m_freem(m);
2815 CTR1(KTR_MLD,
2816 "%s: m_append() failed", __func__);
2817 return (-ENOMEM);
2818 }
2819 nallow += !!(crt == REC_ALLOW);
2820 nblock += !!(crt == REC_BLOCK);
2821 if (++rsrcs == m0srcs)
2822 break;
2823 }
2824 /*
2825 * If we did not append any tree nodes on this
2826 * pass, back out of allocations.
2827 */
2828 if (rsrcs == 0) {
2829 npbytes -= sizeof(struct mldv2_record);
2830 if (m != m0) {
2831 CTR1(KTR_MLD,
2832 "%s: m_free(m)", __func__);
2833 m_freem(m);
2834 } else {
2835 CTR1(KTR_MLD,
2836 "%s: m_adj(m, -mr)", __func__);
2837 m_adj(m, -((int)sizeof(
2838 struct mldv2_record)));
2839 }
2840 continue;
2841 }
2842 npbytes += (rsrcs * sizeof(struct in6_addr));
2843 if (crt == REC_ALLOW)
2844 pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2845 else if (crt == REC_BLOCK)
2846 pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2847 pmr->mr_numsrc = htons(rsrcs);
2848 /*
2849 * Count the new group record, and enqueue this
2850 * packet if it wasn't already queued.
2851 */
2852 m->m_pkthdr.PH_vt.vt_nrecs++;
2853 if (m != m0)
2854 _IF_ENQUEUE(ifq, m);
2855 nbytes += npbytes;
2856 } while (nims != NULL);
2857 drt |= crt;
2858 crt = (~crt & REC_FULL);
2859 }
2860
2861 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2862 nallow, nblock);
2863
2864 return (nbytes);
2865}
2866
2867static int
2868mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
2869{
2870 struct ifqueue *gq;
2871 struct mbuf *m; /* pending state-change */
2872 struct mbuf *m0; /* copy of pending state-change */
2873 struct mbuf *mt; /* last state-change in packet */
2874 int docopy, domerge;
2875 u_int recslen;
2876
2877 docopy = 0;
2878 domerge = 0;
2879 recslen = 0;
2880
2881 IN6_MULTI_LOCK_ASSERT();
2882 MLD_LOCK_ASSERT();
2883
2884 /*
2885 * If there are further pending retransmissions, make a writable
2886 * copy of each queued state-change message before merging.
2887 */
2888 if (inm->in6m_scrv > 0)
2889 docopy = 1;
2890
2891 gq = &inm->in6m_scq;
2892#ifdef KTR
2893 if (gq->ifq_head == NULL) {
2894 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2895 __func__, inm);
2896 }
2897#endif
2898
2899 m = gq->ifq_head;
2900 while (m != NULL) {
2901 /*
2902 * Only merge the report into the current packet if
2903 * there is sufficient space to do so; an MLDv2 report
2904 * packet may only contain 65,535 group records.
2905 * Always use a simple mbuf chain concatentation to do this,
2906 * as large state changes for single groups may have
2907 * allocated clusters.
2908 */
2909 domerge = 0;
2910 mt = ifscq->ifq_tail;
2911 if (mt != NULL) {
2912 recslen = m_length(m, NULL);
2913
2914 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2915 m->m_pkthdr.PH_vt.vt_nrecs <=
2916 MLD_V2_REPORT_MAXRECS) &&
2917 (mt->m_pkthdr.len + recslen <=
2918 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2919 domerge = 1;
2920 }
2921
2922 if (!domerge && _IF_QFULL(gq)) {
2923 CTR2(KTR_MLD,
2924 "%s: outbound queue full, skipping whole packet %p",
2925 __func__, m);
2926 mt = m->m_nextpkt;
2927 if (!docopy)
2928 m_freem(m);
2929 m = mt;
2930 continue;
2931 }
2932
2933 if (!docopy) {
2934 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2935 _IF_DEQUEUE(gq, m0);
2936 m = m0->m_nextpkt;
2937 } else {
2938 CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2939 m0 = m_dup(m, M_NOWAIT);
2940 if (m0 == NULL)
2941 return (ENOMEM);
2942 m0->m_nextpkt = NULL;
2943 m = m->m_nextpkt;
2944 }
2945
2946 if (!domerge) {
2947 CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)",
2948 __func__, m0, ifscq);
2949 _IF_ENQUEUE(ifscq, m0);
2950 } else {
2951 struct mbuf *mtl; /* last mbuf of packet mt */
2952
2953 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2954 __func__, m0, mt);
2955
2956 mtl = m_last(mt);
2957 m0->m_flags &= ~M_PKTHDR;
2958 mt->m_pkthdr.len += recslen;
2959 mt->m_pkthdr.PH_vt.vt_nrecs +=
2960 m0->m_pkthdr.PH_vt.vt_nrecs;
2961
2962 mtl->m_next = m0;
2963 }
2964 }
2965
2966 return (0);
2967}
2968
2969/*
2970 * Respond to a pending MLDv2 General Query.
2971 */
2972static void
2973mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
2974{
2975 struct ifmultiaddr *ifma, *tifma;
2976 struct ifnet *ifp;
2977 struct in6_multi *inm;
2978 int retval;
2979
2980 IN6_MULTI_LOCK_ASSERT();
2981 MLD_LOCK_ASSERT();
2982
2983 KASSERT(mli->mli_version == MLD_VERSION_2,
2984 ("%s: called when version %d", __func__, mli->mli_version));
2985
2986 ifp = mli->mli_ifp;
2987
2988 IF_ADDR_LOCK(ifp);
2989 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
2990 if (ifma->ifma_addr->sa_family != AF_INET6 ||
2991 ifma->ifma_protospec == NULL)
2992 continue;
2993
2994 inm = (struct in6_multi *)ifma->ifma_protospec;
2995 KASSERT(ifp == inm->in6m_ifp,
2996 ("%s: inconsistent ifp", __func__));
2997
2998 switch (inm->in6m_state) {
2999 case MLD_NOT_MEMBER:
3000 case MLD_SILENT_MEMBER:
3001 break;
3002 case MLD_REPORTING_MEMBER:
3003 case MLD_IDLE_MEMBER:
3004 case MLD_LAZY_MEMBER:
3005 case MLD_SLEEPING_MEMBER:
3006 case MLD_AWAKENING_MEMBER:
3007 inm->in6m_state = MLD_REPORTING_MEMBER;
3008 retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3009 inm, 0, 0, 0, 0);
3010 CTR2(KTR_MLD, "%s: enqueue record = %d",
3011 __func__, retval);
3012 break;
3013 case MLD_G_QUERY_PENDING_MEMBER:
3014 case MLD_SG_QUERY_PENDING_MEMBER:
3015 case MLD_LEAVING_MEMBER:
3016 break;
3017 }
3018 }
3019 IF_ADDR_UNLOCK(ifp);
3020
3021 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3022
3023 /*
3024 * Slew transmission of bursts over 500ms intervals.
3025 */
3026 if (mli->mli_gq.ifq_head != NULL) {
3027 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3028 MLD_RESPONSE_BURST_INTERVAL);
3029 V_interface_timers_running6 = 1;
3030 }
3031}
3032
3033/*
3034 * Transmit the next pending message in the output queue.
3035 *
3036 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3037 * MRT: Nothing needs to be done, as MLD traffic is always local to
3038 * a link and uses a link-scope multicast address.
3039 */
3040static void
3041mld_dispatch_packet(struct mbuf *m)
3042{
3043 struct ip6_moptions im6o;
3044 struct ifnet *ifp;
3045 struct ifnet *oifp;
3046 struct mbuf *m0;
3047 struct mbuf *md;
3048 struct ip6_hdr *ip6;
3049 struct mld_hdr *mld;
3050 int error;
3051 int off;
3052 int type;
3053 uint32_t ifindex;
3054
3055 CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3056
3057 /*
3058 * Set VNET image pointer from enqueued mbuf chain
3059 * before doing anything else. Whilst we use interface
3060 * indexes to guard against interface detach, they are
3061 * unique to each VIMAGE and must be retrieved.
3062 */
3063 ifindex = mld_restore_context(m);
3064
3065 /*
3066 * Check if the ifnet still exists. This limits the scope of
3067 * any race in the absence of a global ifp lock for low cost
3068 * (an array lookup).
3069 */
3070 ifp = ifnet_byindex(ifindex);
3071 if (ifp == NULL) {
3072 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3073 __func__, m, ifindex);
3074 m_freem(m);
3075 IP6STAT_INC(ip6s_noroute);
3076 goto out;
3077 }
3078
3079 im6o.im6o_multicast_hlim = 1;
3080 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3081 im6o.im6o_multicast_ifp = ifp;
3082
3083 if (m->m_flags & M_MLDV1) {
3084 m0 = m;
3085 } else {
3086 m0 = mld_v2_encap_report(ifp, m);
3087 if (m0 == NULL) {
3088 CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3089 m_freem(m);
3090 IP6STAT_INC(ip6s_odropped);
3091 goto out;
3092 }
3093 }
3094
3095 mld_scrub_context(m0);
3096 m->m_flags &= ~(M_PROTOFLAGS);
3097 m0->m_pkthdr.rcvif = V_loif;
3098
3099 ip6 = mtod(m0, struct ip6_hdr *);
3100#if 0
3101 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */
3102#else
3103 /*
3104 * XXX XXX Break some KPI rules to prevent an LOR which would
3105 * occur if we called in6_setscope() at transmission.
3106 * See comments at top of file.
3107 */
3108 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3109#endif
3110
3111 /*
3112 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3113 * so we can bump the stats.
3114 */
3115 md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3116 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3117 type = mld->mld_type;
3118
3119 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3120 &oifp, NULL);
3121 if (error) {
3122 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3123 goto out;
3124 }
3125 ICMP6STAT_INC(icp6s_outhist[type]);
3126 if (oifp != NULL) {
3127 icmp6_ifstat_inc(oifp, ifs6_out_msg);
3128 switch (type) {
3129 case MLD_LISTENER_REPORT:
3130 case MLDV2_LISTENER_REPORT:
3131 icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3132 break;
3133 case MLD_LISTENER_DONE:
3134 icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3135 break;
3136 }
3137 }
3138out:
3139 return;
3140}
3141
3142/*
3143 * Encapsulate an MLDv2 report.
3144 *
3145 * KAME IPv6 requires that hop-by-hop options be passed separately,
3146 * and that the IPv6 header be prepended in a separate mbuf.
3147 *
3148 * Returns a pointer to the new mbuf chain head, or NULL if the
3149 * allocation failed.
3150 */
3151static struct mbuf *
3152mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3153{
3154 struct mbuf *mh;
3155 struct mldv2_report *mld;
3156 struct ip6_hdr *ip6;
3157 struct in6_ifaddr *ia;
3158 int mldreclen;
3159
3160 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3161 KASSERT((m->m_flags & M_PKTHDR),
3162 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3163
3164 /*
3165 * RFC3590: OK to send as :: or tentative during DAD.
3166 */
3167 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3168 if (ia == NULL)
3169 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3170
3171 MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3172 if (mh == NULL) {
3173 if (ia != NULL)
3174 ifa_free(&ia->ia_ifa);
3175 m_freem(m);
3176 return (NULL);
3177 }
3178 MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3179
3180 mldreclen = m_length(m, NULL);
3181 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3182
3183 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3184 mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3185 sizeof(struct mldv2_report) + mldreclen;
3186
3187 ip6 = mtod(mh, struct ip6_hdr *);
3188 ip6->ip6_flow = 0;
3189 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3190 ip6->ip6_vfc |= IPV6_VERSION;
3191 ip6->ip6_nxt = IPPROTO_ICMPV6;
3192 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3193 if (ia != NULL)
3194 ifa_free(&ia->ia_ifa);
3195 ip6->ip6_dst = in6addr_linklocal_allv2routers;
3196 /* scope ID will be set in netisr */
3197
3198 mld = (struct mldv2_report *)(ip6 + 1);
3199 mld->mld_type = MLDV2_LISTENER_REPORT;
3200 mld->mld_code = 0;
3201 mld->mld_cksum = 0;
3202 mld->mld_v2_reserved = 0;
3203 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3204 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3205
3206 mh->m_next = m;
3207 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3208 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3209 return (mh);
3210}
3211
3212#ifdef KTR
3213static char *
3214mld_rec_type_to_str(const int type)
3215{
3216
3217 switch (type) {
3218 case MLD_CHANGE_TO_EXCLUDE_MODE:
3219 return "TO_EX";
3220 break;
3221 case MLD_CHANGE_TO_INCLUDE_MODE:
3222 return "TO_IN";
3223 break;
3224 case MLD_MODE_IS_EXCLUDE:
3225 return "MODE_EX";
3226 break;
3227 case MLD_MODE_IS_INCLUDE:
3228 return "MODE_IN";
3229 break;
3230 case MLD_ALLOW_NEW_SOURCES:
3231 return "ALLOW_NEW";
3232 break;
3233 case MLD_BLOCK_OLD_SOURCES:
3234 return "BLOCK_OLD";
3235 break;
3236 default:
3237 break;
3238 }
3239 return "unknown";
3240}
3241#endif
3242
3243static void
3244mld_init(void *unused __unused)
3245{
3246
3247 CTR1(KTR_MLD, "%s: initializing", __func__);
3248 MLD_LOCK_INIT();
3249
3250 ip6_initpktopts(&mld_po);
3251 mld_po.ip6po_hlim = 1;
3252 mld_po.ip6po_hbh = &mld_ra.hbh;
3253 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3254 mld_po.ip6po_flags = IP6PO_DONTFRAG;
3255}
3256SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL);
3257
3258static void
3259mld_uninit(void *unused __unused)
3260{
3261
3262 CTR1(KTR_MLD, "%s: tearing down", __func__);
3263 MLD_LOCK_DESTROY();
3264}
3265SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL);
3266
3267static void
3268vnet_mld_init(const void *unused __unused)
3269{
3270
3271 CTR1(KTR_MLD, "%s: initializing", __func__);
3272
3273 LIST_INIT(&V_mli_head);
3274}
3275VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init,
3276 NULL);
3277
3278static void
3279vnet_mld_uninit(const void *unused __unused)
3280{
3281
3282 CTR1(KTR_MLD, "%s: tearing down", __func__);
3283
3284 KASSERT(LIST_EMPTY(&V_mli_head),
3285 ("%s: mli list not empty; ifnets not detached?", __func__));
3286}
3287VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit,
3288 NULL);
3289
3290static int
3291mld_modevent(module_t mod, int type, void *unused __unused)
3292{
3293
3294 switch (type) {
3295 case MOD_LOAD:
3296 case MOD_UNLOAD:
3297 break;
3298 default:
3299 return (EOPNOTSUPP);
3300 }
3301 return (0);
3302}
3303
3304static moduledata_t mld_mod = {
3305 "mld",
3306 mld_modevent,
3307 0
3308};
3309DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
| 211
212#define V_mld_gsrdelay VNET(mld_gsrdelay)
213#define V_mli_head VNET(mli_head)
214#define V_interface_timers_running6 VNET(interface_timers_running6)
215#define V_state_change_timers_running6 VNET(state_change_timers_running6)
216#define V_current_state_timers_running6 VNET(current_state_timers_running6)
217
218SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */
219
220SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
221 "IPv6 Multicast Listener Discovery");
222
223/*
224 * Virtualized sysctls.
225 */
226SYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
227 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
228 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
229 "Rate limit for MLDv2 Group-and-Source queries in seconds");
230
231/*
232 * Non-virtualized sysctls.
233 */
234SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
235 sysctl_mld_ifinfo, "Per-interface MLDv2 state");
236
237static int mld_v1enable = 1;
238SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW,
239 &mld_v1enable, 0, "Enable fallback to MLDv1");
240TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable);
241
242static int mld_use_allow = 1;
243SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW,
244 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
245TUNABLE_INT("net.inet6.mld.use_allow", &mld_use_allow);
246
247/*
248 * Packed Router Alert option structure declaration.
249 */
250struct mld_raopt {
251 struct ip6_hbh hbh;
252 struct ip6_opt pad;
253 struct ip6_opt_router ra;
254} __packed;
255
256/*
257 * Router Alert hop-by-hop option header.
258 */
259static struct mld_raopt mld_ra = {
260 .hbh = { 0, 0 },
261 .pad = { .ip6o_type = IP6OPT_PADN, 0 },
262 .ra = {
263 .ip6or_type = IP6OPT_ROUTER_ALERT,
264 .ip6or_len = IP6OPT_RTALERT_LEN - 2,
265 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
266 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
267 }
268};
269static struct ip6_pktopts mld_po;
270
271static __inline void
272mld_save_context(struct mbuf *m, struct ifnet *ifp)
273{
274
275#ifdef VIMAGE
276 m->m_pkthdr.header = ifp->if_vnet;
277#endif /* VIMAGE */
278 m->m_pkthdr.flowid = ifp->if_index;
279}
280
281static __inline void
282mld_scrub_context(struct mbuf *m)
283{
284
285 m->m_pkthdr.header = NULL;
286 m->m_pkthdr.flowid = 0;
287}
288
289/*
290 * Restore context from a queued output chain.
291 * Return saved ifindex.
292 *
293 * VIMAGE: The assertion is there to make sure that we
294 * actually called CURVNET_SET() with what's in the mbuf chain.
295 */
296static __inline uint32_t
297mld_restore_context(struct mbuf *m)
298{
299
300#if defined(VIMAGE) && defined(INVARIANTS)
301 KASSERT(curvnet == m->m_pkthdr.header,
302 ("%s: called when curvnet was not restored", __func__));
303#endif
304 return (m->m_pkthdr.flowid);
305}
306
307/*
308 * Retrieve or set threshold between group-source queries in seconds.
309 *
310 * VIMAGE: Assume curvnet set by caller.
311 * SMPng: NOTE: Serialized by MLD lock.
312 */
313static int
314sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
315{
316 int error;
317 int i;
318
319 error = sysctl_wire_old_buffer(req, sizeof(int));
320 if (error)
321 return (error);
322
323 MLD_LOCK();
324
325 i = V_mld_gsrdelay.tv_sec;
326
327 error = sysctl_handle_int(oidp, &i, 0, req);
328 if (error || !req->newptr)
329 goto out_locked;
330
331 if (i < -1 || i >= 60) {
332 error = EINVAL;
333 goto out_locked;
334 }
335
336 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
337 V_mld_gsrdelay.tv_sec, i);
338 V_mld_gsrdelay.tv_sec = i;
339
340out_locked:
341 MLD_UNLOCK();
342 return (error);
343}
344
345/*
346 * Expose struct mld_ifinfo to userland, keyed by ifindex.
347 * For use by ifmcstat(8).
348 *
349 * SMPng: NOTE: Does an unlocked ifindex space read.
350 * VIMAGE: Assume curvnet set by caller. The node handler itself
351 * is not directly virtualized.
352 */
353static int
354sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
355{
356 int *name;
357 int error;
358 u_int namelen;
359 struct ifnet *ifp;
360 struct mld_ifinfo *mli;
361
362 name = (int *)arg1;
363 namelen = arg2;
364
365 if (req->newptr != NULL)
366 return (EPERM);
367
368 if (namelen != 1)
369 return (EINVAL);
370
371 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
372 if (error)
373 return (error);
374
375 IN6_MULTI_LOCK();
376 MLD_LOCK();
377
378 if (name[0] <= 0 || name[0] > V_if_index) {
379 error = ENOENT;
380 goto out_locked;
381 }
382
383 error = ENOENT;
384
385 ifp = ifnet_byindex(name[0]);
386 if (ifp == NULL)
387 goto out_locked;
388
389 LIST_FOREACH(mli, &V_mli_head, mli_link) {
390 if (ifp == mli->mli_ifp) {
391 error = SYSCTL_OUT(req, mli,
392 sizeof(struct mld_ifinfo));
393 break;
394 }
395 }
396
397out_locked:
398 MLD_UNLOCK();
399 IN6_MULTI_UNLOCK();
400 return (error);
401}
402
403/*
404 * Dispatch an entire queue of pending packet chains.
405 * VIMAGE: Assumes the vnet pointer has been set.
406 */
407static void
408mld_dispatch_queue(struct ifqueue *ifq, int limit)
409{
410 struct mbuf *m;
411
412 for (;;) {
413 _IF_DEQUEUE(ifq, m);
414 if (m == NULL)
415 break;
416 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m);
417 mld_dispatch_packet(m);
418 if (--limit == 0)
419 break;
420 }
421}
422
423/*
424 * Filter outgoing MLD report state by group.
425 *
426 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
427 * and node-local addresses. However, kernel and socket consumers
428 * always embed the KAME scope ID in the address provided, so strip it
429 * when performing comparison.
430 * Note: This is not the same as the *multicast* scope.
431 *
432 * Return zero if the given group is one for which MLD reports
433 * should be suppressed, or non-zero if reports should be issued.
434 */
435static __inline int
436mld_is_addr_reported(const struct in6_addr *addr)
437{
438
439 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
440
441 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
442 return (0);
443
444 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
445 struct in6_addr tmp = *addr;
446 in6_clearscope(&tmp);
447 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
448 return (0);
449 }
450
451 return (1);
452}
453
454/*
455 * Attach MLD when PF_INET6 is attached to an interface.
456 *
457 * SMPng: Normally called with IF_AFDATA_LOCK held.
458 */
459struct mld_ifinfo *
460mld_domifattach(struct ifnet *ifp)
461{
462 struct mld_ifinfo *mli;
463
464 CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
465 __func__, ifp, ifp->if_xname);
466
467 MLD_LOCK();
468
469 mli = mli_alloc_locked(ifp);
470 if (!(ifp->if_flags & IFF_MULTICAST))
471 mli->mli_flags |= MLIF_SILENT;
472 if (mld_use_allow)
473 mli->mli_flags |= MLIF_USEALLOW;
474
475 MLD_UNLOCK();
476
477 return (mli);
478}
479
480/*
481 * VIMAGE: assume curvnet set by caller.
482 */
483static struct mld_ifinfo *
484mli_alloc_locked(/*const*/ struct ifnet *ifp)
485{
486 struct mld_ifinfo *mli;
487
488 MLD_LOCK_ASSERT();
489
490 mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO);
491 if (mli == NULL)
492 goto out;
493
494 mli->mli_ifp = ifp;
495 mli->mli_version = MLD_VERSION_2;
496 mli->mli_flags = 0;
497 mli->mli_rv = MLD_RV_INIT;
498 mli->mli_qi = MLD_QI_INIT;
499 mli->mli_qri = MLD_QRI_INIT;
500 mli->mli_uri = MLD_URI_INIT;
501
502 SLIST_INIT(&mli->mli_relinmhead);
503
504 /*
505 * Responses to general queries are subject to bounds.
506 */
507 IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
508
509 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
510
511 CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)",
512 ifp, ifp->if_xname);
513
514out:
515 return (mli);
516}
517
518/*
519 * Hook for ifdetach.
520 *
521 * NOTE: Some finalization tasks need to run before the protocol domain
522 * is detached, but also before the link layer does its cleanup.
523 * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
524 *
525 * SMPng: Caller must hold IN6_MULTI_LOCK().
526 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
527 * XXX This routine is also bitten by unlocked ifma_protospec access.
528 */
529void
530mld_ifdetach(struct ifnet *ifp)
531{
532 struct mld_ifinfo *mli;
533 struct ifmultiaddr *ifma;
534 struct in6_multi *inm, *tinm;
535
536 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
537 ifp->if_xname);
538
539 IN6_MULTI_LOCK_ASSERT();
540 MLD_LOCK();
541
542 mli = MLD_IFINFO(ifp);
543 if (mli->mli_version == MLD_VERSION_2) {
544 IF_ADDR_LOCK(ifp);
545 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
546 if (ifma->ifma_addr->sa_family != AF_INET6 ||
547 ifma->ifma_protospec == NULL)
548 continue;
549 inm = (struct in6_multi *)ifma->ifma_protospec;
550 if (inm->in6m_state == MLD_LEAVING_MEMBER) {
551 SLIST_INSERT_HEAD(&mli->mli_relinmhead,
552 inm, in6m_nrele);
553 }
554 in6m_clear_recorded(inm);
555 }
556 IF_ADDR_UNLOCK(ifp);
557 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
558 tinm) {
559 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
560 in6m_release_locked(inm);
561 }
562 }
563
564 MLD_UNLOCK();
565}
566
567/*
568 * Hook for domifdetach.
569 * Runs after link-layer cleanup; free MLD state.
570 *
571 * SMPng: Normally called with IF_AFDATA_LOCK held.
572 */
573void
574mld_domifdetach(struct ifnet *ifp)
575{
576
577 CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
578 __func__, ifp, ifp->if_xname);
579
580 MLD_LOCK();
581 mli_delete_locked(ifp);
582 MLD_UNLOCK();
583}
584
585static void
586mli_delete_locked(const struct ifnet *ifp)
587{
588 struct mld_ifinfo *mli, *tmli;
589
590 CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)",
591 __func__, ifp, ifp->if_xname);
592
593 MLD_LOCK_ASSERT();
594
595 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
596 if (mli->mli_ifp == ifp) {
597 /*
598 * Free deferred General Query responses.
599 */
600 _IF_DRAIN(&mli->mli_gq);
601
602 LIST_REMOVE(mli, mli_link);
603
604 KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
605 ("%s: there are dangling in_multi references",
606 __func__));
607
608 free(mli, M_MLD);
609 return;
610 }
611 }
612#ifdef INVARIANTS
613 panic("%s: mld_ifinfo not found for ifp %p\n", __func__, ifp);
614#endif
615}
616
617/*
618 * Process a received MLDv1 general or address-specific query.
619 * Assumes that the query header has been pulled up to sizeof(mld_hdr).
620 *
621 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
622 * mld_addr. This is OK as we own the mbuf chain.
623 */
624static int
625mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
626 /*const*/ struct mld_hdr *mld)
627{
628 struct ifmultiaddr *ifma;
629 struct mld_ifinfo *mli;
630 struct in6_multi *inm;
631 int is_general_query;
632 uint16_t timer;
633#ifdef KTR
634 char ip6tbuf[INET6_ADDRSTRLEN];
635#endif
636
637 is_general_query = 0;
638
639 if (!mld_v1enable) {
640 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
641 ip6_sprintf(ip6tbuf, &mld->mld_addr),
642 ifp, ifp->if_xname);
643 return (0);
644 }
645
646 /*
647 * RFC3810 Section 6.2: MLD queries must originate from
648 * a router's link-local address.
649 */
650 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
651 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
652 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
653 ifp, ifp->if_xname);
654 return (0);
655 }
656
657 /*
658 * Do address field validation upfront before we accept
659 * the query.
660 */
661 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
662 /*
663 * MLDv1 General Query.
664 * If this was not sent to the all-nodes group, ignore it.
665 */
666 struct in6_addr dst;
667
668 dst = ip6->ip6_dst;
669 in6_clearscope(&dst);
670 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
671 return (EINVAL);
672 is_general_query = 1;
673 } else {
674 /*
675 * Embed scope ID of receiving interface in MLD query for
676 * lookup whilst we don't hold other locks.
677 */
678 in6_setscope(&mld->mld_addr, ifp, NULL);
679 }
680
681 IN6_MULTI_LOCK();
682 MLD_LOCK();
683 IF_ADDR_LOCK(ifp);
684
685 /*
686 * Switch to MLDv1 host compatibility mode.
687 */
688 mli = MLD_IFINFO(ifp);
689 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
690 mld_set_version(mli, MLD_VERSION_1);
691
692 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
693 if (timer == 0)
694 timer = 1;
695
696 if (is_general_query) {
697 /*
698 * For each reporting group joined on this
699 * interface, kick the report timer.
700 */
701 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
702 ifp, ifp->if_xname);
703 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
704 if (ifma->ifma_addr->sa_family != AF_INET6 ||
705 ifma->ifma_protospec == NULL)
706 continue;
707 inm = (struct in6_multi *)ifma->ifma_protospec;
708 mld_v1_update_group(inm, timer);
709 }
710 } else {
711 /*
712 * MLDv1 Group-Specific Query.
713 * If this is a group-specific MLDv1 query, we need only
714 * look up the single group to process it.
715 */
716 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
717 if (inm != NULL) {
718 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
719 ip6_sprintf(ip6tbuf, &mld->mld_addr),
720 ifp, ifp->if_xname);
721 mld_v1_update_group(inm, timer);
722 }
723 /* XXX Clear embedded scope ID as userland won't expect it. */
724 in6_clearscope(&mld->mld_addr);
725 }
726
727 IF_ADDR_UNLOCK(ifp);
728 MLD_UNLOCK();
729 IN6_MULTI_UNLOCK();
730
731 return (0);
732}
733
734/*
735 * Update the report timer on a group in response to an MLDv1 query.
736 *
737 * If we are becoming the reporting member for this group, start the timer.
738 * If we already are the reporting member for this group, and timer is
739 * below the threshold, reset it.
740 *
741 * We may be updating the group for the first time since we switched
742 * to MLDv2. If we are, then we must clear any recorded source lists,
743 * and transition to REPORTING state; the group timer is overloaded
744 * for group and group-source query responses.
745 *
746 * Unlike MLDv2, the delay per group should be jittered
747 * to avoid bursts of MLDv1 reports.
748 */
749static void
750mld_v1_update_group(struct in6_multi *inm, const int timer)
751{
752#ifdef KTR
753 char ip6tbuf[INET6_ADDRSTRLEN];
754#endif
755
756 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
757 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
758 inm->in6m_ifp->if_xname, timer);
759
760 IN6_MULTI_LOCK_ASSERT();
761
762 switch (inm->in6m_state) {
763 case MLD_NOT_MEMBER:
764 case MLD_SILENT_MEMBER:
765 break;
766 case MLD_REPORTING_MEMBER:
767 if (inm->in6m_timer != 0 &&
768 inm->in6m_timer <= timer) {
769 CTR1(KTR_MLD, "%s: REPORTING and timer running, "
770 "skipping.", __func__);
771 break;
772 }
773 /* FALLTHROUGH */
774 case MLD_SG_QUERY_PENDING_MEMBER:
775 case MLD_G_QUERY_PENDING_MEMBER:
776 case MLD_IDLE_MEMBER:
777 case MLD_LAZY_MEMBER:
778 case MLD_AWAKENING_MEMBER:
779 CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
780 inm->in6m_state = MLD_REPORTING_MEMBER;
781 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
782 V_current_state_timers_running6 = 1;
783 break;
784 case MLD_SLEEPING_MEMBER:
785 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
786 inm->in6m_state = MLD_AWAKENING_MEMBER;
787 break;
788 case MLD_LEAVING_MEMBER:
789 break;
790 }
791}
792
793/*
794 * Process a received MLDv2 general, group-specific or
795 * group-and-source-specific query.
796 *
797 * Assumes that the query header has been pulled up to sizeof(mldv2_query).
798 *
799 * Return 0 if successful, otherwise an appropriate error code is returned.
800 */
801static int
802mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
803 struct mbuf *m, const int off, const int icmp6len)
804{
805 struct mld_ifinfo *mli;
806 struct mldv2_query *mld;
807 struct in6_multi *inm;
808 uint32_t maxdelay, nsrc, qqi;
809 int is_general_query;
810 uint16_t timer;
811 uint8_t qrv;
812#ifdef KTR
813 char ip6tbuf[INET6_ADDRSTRLEN];
814#endif
815
816 is_general_query = 0;
817
818 /*
819 * RFC3810 Section 6.2: MLD queries must originate from
820 * a router's link-local address.
821 */
822 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
823 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
824 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
825 ifp, ifp->if_xname);
826 return (0);
827 }
828
829 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname);
830
831 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
832
833 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */
834 if (maxdelay >= 32678) {
835 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
836 (MLD_MRC_EXP(maxdelay) + 3);
837 }
838 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
839 if (timer == 0)
840 timer = 1;
841
842 qrv = MLD_QRV(mld->mld_misc);
843 if (qrv < 2) {
844 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
845 qrv, MLD_RV_INIT);
846 qrv = MLD_RV_INIT;
847 }
848
849 qqi = mld->mld_qqi;
850 if (qqi >= 128) {
851 qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
852 (MLD_QQIC_EXP(mld->mld_qqi) + 3);
853 }
854
855 nsrc = ntohs(mld->mld_numsrc);
856 if (nsrc > MLD_MAX_GS_SOURCES)
857 return (EMSGSIZE);
858 if (icmp6len < sizeof(struct mldv2_query) +
859 (nsrc * sizeof(struct in6_addr)))
860 return (EMSGSIZE);
861
862 /*
863 * Do further input validation upfront to avoid resetting timers
864 * should we need to discard this query.
865 */
866 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
867 /*
868 * General Queries SHOULD be directed to ff02::1.
869 * A general query with a source list has undefined
870 * behaviour; discard it.
871 */
872 struct in6_addr dst;
873
874 dst = ip6->ip6_dst;
875 in6_clearscope(&dst);
876 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes) ||
877 nsrc > 0)
878 return (EINVAL);
879 is_general_query = 1;
880 } else {
881 /*
882 * Embed scope ID of receiving interface in MLD query for
883 * lookup whilst we don't hold other locks (due to KAME
884 * locking lameness). We own this mbuf chain just now.
885 */
886 in6_setscope(&mld->mld_addr, ifp, NULL);
887 }
888
889 IN6_MULTI_LOCK();
890 MLD_LOCK();
891 IF_ADDR_LOCK(ifp);
892
893 mli = MLD_IFINFO(ifp);
894 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
895
896 /*
897 * Discard the v2 query if we're in Compatibility Mode.
898 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
899 * until the Old Version Querier Present timer expires.
900 */
901 if (mli->mli_version != MLD_VERSION_2)
902 goto out_locked;
903
904 mld_set_version(mli, MLD_VERSION_2);
905 mli->mli_rv = qrv;
906 mli->mli_qi = qqi;
907 mli->mli_qri = maxdelay;
908
909 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
910 maxdelay);
911
912 if (is_general_query) {
913 /*
914 * MLDv2 General Query.
915 *
916 * Schedule a current-state report on this ifp for
917 * all groups, possibly containing source lists.
918 *
919 * If there is a pending General Query response
920 * scheduled earlier than the selected delay, do
921 * not schedule any other reports.
922 * Otherwise, reset the interface timer.
923 */
924 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
925 ifp, ifp->if_xname);
926 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
927 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
928 V_interface_timers_running6 = 1;
929 }
930 } else {
931 /*
932 * MLDv2 Group-specific or Group-and-source-specific Query.
933 *
934 * Group-source-specific queries are throttled on
935 * a per-group basis to defeat denial-of-service attempts.
936 * Queries for groups we are not a member of on this
937 * link are simply ignored.
938 */
939 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
940 if (inm == NULL)
941 goto out_locked;
942 if (nsrc > 0) {
943 if (!ratecheck(&inm->in6m_lastgsrtv,
944 &V_mld_gsrdelay)) {
945 CTR1(KTR_MLD, "%s: GS query throttled.",
946 __func__);
947 goto out_locked;
948 }
949 }
950 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
951 ifp, ifp->if_xname);
952 /*
953 * If there is a pending General Query response
954 * scheduled sooner than the selected delay, no
955 * further report need be scheduled.
956 * Otherwise, prepare to respond to the
957 * group-specific or group-and-source query.
958 */
959 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
960 mld_v2_process_group_query(inm, mli, timer, m, off);
961
962 /* XXX Clear embedded scope ID as userland won't expect it. */
963 in6_clearscope(&mld->mld_addr);
964 }
965
966out_locked:
967 IF_ADDR_UNLOCK(ifp);
968 MLD_UNLOCK();
969 IN6_MULTI_UNLOCK();
970
971 return (0);
972}
973
974/*
975 * Process a recieved MLDv2 group-specific or group-and-source-specific
976 * query.
977 * Return <0 if any error occured. Currently this is ignored.
978 */
979static int
980mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli,
981 int timer, struct mbuf *m0, const int off)
982{
983 struct mldv2_query *mld;
984 int retval;
985 uint16_t nsrc;
986
987 IN6_MULTI_LOCK_ASSERT();
988 MLD_LOCK_ASSERT();
989
990 retval = 0;
991 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
992
993 switch (inm->in6m_state) {
994 case MLD_NOT_MEMBER:
995 case MLD_SILENT_MEMBER:
996 case MLD_SLEEPING_MEMBER:
997 case MLD_LAZY_MEMBER:
998 case MLD_AWAKENING_MEMBER:
999 case MLD_IDLE_MEMBER:
1000 case MLD_LEAVING_MEMBER:
1001 return (retval);
1002 break;
1003 case MLD_REPORTING_MEMBER:
1004 case MLD_G_QUERY_PENDING_MEMBER:
1005 case MLD_SG_QUERY_PENDING_MEMBER:
1006 break;
1007 }
1008
1009 nsrc = ntohs(mld->mld_numsrc);
1010
1011 /*
1012 * Deal with group-specific queries upfront.
1013 * If any group query is already pending, purge any recorded
1014 * source-list state if it exists, and schedule a query response
1015 * for this group-specific query.
1016 */
1017 if (nsrc == 0) {
1018 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1019 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1020 in6m_clear_recorded(inm);
1021 timer = min(inm->in6m_timer, timer);
1022 }
1023 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1024 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1025 V_current_state_timers_running6 = 1;
1026 return (retval);
1027 }
1028
1029 /*
1030 * Deal with the case where a group-and-source-specific query has
1031 * been received but a group-specific query is already pending.
1032 */
1033 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1034 timer = min(inm->in6m_timer, timer);
1035 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1036 V_current_state_timers_running6 = 1;
1037 return (retval);
1038 }
1039
1040 /*
1041 * Finally, deal with the case where a group-and-source-specific
1042 * query has been received, where a response to a previous g-s-r
1043 * query exists, or none exists.
1044 * In this case, we need to parse the source-list which the Querier
1045 * has provided us with and check if we have any source list filter
1046 * entries at T1 for these sources. If we do not, there is no need
1047 * schedule a report and the query may be dropped.
1048 * If we do, we must record them and schedule a current-state
1049 * report for those sources.
1050 */
1051 if (inm->in6m_nsrc > 0) {
1052 struct mbuf *m;
1053 uint8_t *sp;
1054 int i, nrecorded;
1055 int soff;
1056
1057 m = m0;
1058 soff = off + sizeof(struct mldv2_query);
1059 nrecorded = 0;
1060 for (i = 0; i < nsrc; i++) {
1061 sp = mtod(m, uint8_t *) + soff;
1062 retval = in6m_record_source(inm,
1063 (const struct in6_addr *)sp);
1064 if (retval < 0)
1065 break;
1066 nrecorded += retval;
1067 soff += sizeof(struct in6_addr);
1068 if (soff >= m->m_len) {
1069 soff = soff - m->m_len;
1070 m = m->m_next;
1071 if (m == NULL)
1072 break;
1073 }
1074 }
1075 if (nrecorded > 0) {
1076 CTR1(KTR_MLD,
1077 "%s: schedule response to SG query", __func__);
1078 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1079 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1080 V_current_state_timers_running6 = 1;
1081 }
1082 }
1083
1084 return (retval);
1085}
1086
1087/*
1088 * Process a received MLDv1 host membership report.
1089 * Assumes mld points to mld_hdr in pulled up mbuf chain.
1090 *
1091 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1092 * mld_addr. This is OK as we own the mbuf chain.
1093 */
1094static int
1095mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1096 /*const*/ struct mld_hdr *mld)
1097{
1098 struct in6_addr src, dst;
1099 struct in6_ifaddr *ia;
1100 struct in6_multi *inm;
1101#ifdef KTR
1102 char ip6tbuf[INET6_ADDRSTRLEN];
1103#endif
1104
1105 if (!mld_v1enable) {
1106 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1107 ip6_sprintf(ip6tbuf, &mld->mld_addr),
1108 ifp, ifp->if_xname);
1109 return (0);
1110 }
1111
1112 if (ifp->if_flags & IFF_LOOPBACK)
1113 return (0);
1114
1115 /*
1116 * MLDv1 reports must originate from a host's link-local address,
1117 * or the unspecified address (when booting).
1118 */
1119 src = ip6->ip6_src;
1120 in6_clearscope(&src);
1121 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1122 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1123 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1124 ifp, ifp->if_xname);
1125 return (EINVAL);
1126 }
1127
1128 /*
1129 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1130 * group, and must be directed to the group itself.
1131 */
1132 dst = ip6->ip6_dst;
1133 in6_clearscope(&dst);
1134 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1135 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1136 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1137 ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1138 ifp, ifp->if_xname);
1139 return (EINVAL);
1140 }
1141
1142 /*
1143 * Make sure we don't hear our own membership report, as fast
1144 * leave requires knowing that we are the only member of a
1145 * group. Assume we used the link-local address if available,
1146 * otherwise look for ::.
1147 *
1148 * XXX Note that scope ID comparison is needed for the address
1149 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1150 * performed for the on-wire address.
1151 */
1152 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1153 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1154 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1155 if (ia != NULL)
1156 ifa_free(&ia->ia_ifa);
1157 return (0);
1158 }
1159 if (ia != NULL)
1160 ifa_free(&ia->ia_ifa);
1161
1162 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1163 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname);
1164
1165 /*
1166 * Embed scope ID of receiving interface in MLD query for lookup
1167 * whilst we don't hold other locks (due to KAME locking lameness).
1168 */
1169 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1170 in6_setscope(&mld->mld_addr, ifp, NULL);
1171
1172 IN6_MULTI_LOCK();
1173 MLD_LOCK();
1174 IF_ADDR_LOCK(ifp);
1175
1176 /*
1177 * MLDv1 report suppression.
1178 * If we are a member of this group, and our membership should be
1179 * reported, and our group timer is pending or about to be reset,
1180 * stop our group timer by transitioning to the 'lazy' state.
1181 */
1182 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1183 if (inm != NULL) {
1184 struct mld_ifinfo *mli;
1185
1186 mli = inm->in6m_mli;
1187 KASSERT(mli != NULL,
1188 ("%s: no mli for ifp %p", __func__, ifp));
1189
1190 /*
1191 * If we are in MLDv2 host mode, do not allow the
1192 * other host's MLDv1 report to suppress our reports.
1193 */
1194 if (mli->mli_version == MLD_VERSION_2)
1195 goto out_locked;
1196
1197 inm->in6m_timer = 0;
1198
1199 switch (inm->in6m_state) {
1200 case MLD_NOT_MEMBER:
1201 case MLD_SILENT_MEMBER:
1202 case MLD_SLEEPING_MEMBER:
1203 break;
1204 case MLD_REPORTING_MEMBER:
1205 case MLD_IDLE_MEMBER:
1206 case MLD_AWAKENING_MEMBER:
1207 CTR3(KTR_MLD,
1208 "report suppressed for %s on ifp %p(%s)",
1209 ip6_sprintf(ip6tbuf, &mld->mld_addr),
1210 ifp, ifp->if_xname);
1211 case MLD_LAZY_MEMBER:
1212 inm->in6m_state = MLD_LAZY_MEMBER;
1213 break;
1214 case MLD_G_QUERY_PENDING_MEMBER:
1215 case MLD_SG_QUERY_PENDING_MEMBER:
1216 case MLD_LEAVING_MEMBER:
1217 break;
1218 }
1219 }
1220
1221out_locked:
1222 MLD_UNLOCK();
1223 IF_ADDR_UNLOCK(ifp);
1224 IN6_MULTI_UNLOCK();
1225
1226 /* XXX Clear embedded scope ID as userland won't expect it. */
1227 in6_clearscope(&mld->mld_addr);
1228
1229 return (0);
1230}
1231
1232/*
1233 * MLD input path.
1234 *
1235 * Assume query messages which fit in a single ICMPv6 message header
1236 * have been pulled up.
1237 * Assume that userland will want to see the message, even if it
1238 * otherwise fails kernel input validation; do not free it.
1239 * Pullup may however free the mbuf chain m if it fails.
1240 *
1241 * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1242 */
1243int
1244mld_input(struct mbuf *m, int off, int icmp6len)
1245{
1246 struct ifnet *ifp;
1247 struct ip6_hdr *ip6;
1248 struct mld_hdr *mld;
1249 int mldlen;
1250
1251 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1252
1253 ifp = m->m_pkthdr.rcvif;
1254
1255 ip6 = mtod(m, struct ip6_hdr *);
1256
1257 /* Pullup to appropriate size. */
1258 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1259 if (mld->mld_type == MLD_LISTENER_QUERY &&
1260 icmp6len >= sizeof(struct mldv2_query)) {
1261 mldlen = sizeof(struct mldv2_query);
1262 } else {
1263 mldlen = sizeof(struct mld_hdr);
1264 }
1265 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1266 if (mld == NULL) {
1267 ICMP6STAT_INC(icp6s_badlen);
1268 return (IPPROTO_DONE);
1269 }
1270
1271 /*
1272 * Userland needs to see all of this traffic for implementing
1273 * the endpoint discovery portion of multicast routing.
1274 */
1275 switch (mld->mld_type) {
1276 case MLD_LISTENER_QUERY:
1277 icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1278 if (icmp6len == sizeof(struct mld_hdr)) {
1279 if (mld_v1_input_query(ifp, ip6, mld) != 0)
1280 return (0);
1281 } else if (icmp6len >= sizeof(struct mldv2_query)) {
1282 if (mld_v2_input_query(ifp, ip6, m, off,
1283 icmp6len) != 0)
1284 return (0);
1285 }
1286 break;
1287 case MLD_LISTENER_REPORT:
1288 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1289 if (mld_v1_input_report(ifp, ip6, mld) != 0)
1290 return (0);
1291 break;
1292 case MLDV2_LISTENER_REPORT:
1293 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1294 break;
1295 case MLD_LISTENER_DONE:
1296 icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1297 break;
1298 default:
1299 break;
1300 }
1301
1302 return (0);
1303}
1304
1305/*
1306 * Fast timeout handler (global).
1307 * VIMAGE: Timeout handlers are expected to service all vimages.
1308 */
1309void
1310mld_fasttimo(void)
1311{
1312 VNET_ITERATOR_DECL(vnet_iter);
1313
1314 VNET_LIST_RLOCK_NOSLEEP();
1315 VNET_FOREACH(vnet_iter) {
1316 CURVNET_SET(vnet_iter);
1317 mld_fasttimo_vnet();
1318 CURVNET_RESTORE();
1319 }
1320 VNET_LIST_RUNLOCK_NOSLEEP();
1321}
1322
1323/*
1324 * Fast timeout handler (per-vnet).
1325 *
1326 * VIMAGE: Assume caller has set up our curvnet.
1327 */
1328static void
1329mld_fasttimo_vnet(void)
1330{
1331 struct ifqueue scq; /* State-change packets */
1332 struct ifqueue qrq; /* Query response packets */
1333 struct ifnet *ifp;
1334 struct mld_ifinfo *mli;
1335 struct ifmultiaddr *ifma, *tifma;
1336 struct in6_multi *inm;
1337 int uri_fasthz;
1338
1339 uri_fasthz = 0;
1340
1341 /*
1342 * Quick check to see if any work needs to be done, in order to
1343 * minimize the overhead of fasttimo processing.
1344 * SMPng: XXX Unlocked reads.
1345 */
1346 if (!V_current_state_timers_running6 &&
1347 !V_interface_timers_running6 &&
1348 !V_state_change_timers_running6)
1349 return;
1350
1351 IN6_MULTI_LOCK();
1352 MLD_LOCK();
1353
1354 /*
1355 * MLDv2 General Query response timer processing.
1356 */
1357 if (V_interface_timers_running6) {
1358 CTR1(KTR_MLD, "%s: interface timers running", __func__);
1359
1360 V_interface_timers_running6 = 0;
1361 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1362 if (mli->mli_v2_timer == 0) {
1363 /* Do nothing. */
1364 } else if (--mli->mli_v2_timer == 0) {
1365 mld_v2_dispatch_general_query(mli);
1366 } else {
1367 V_interface_timers_running6 = 1;
1368 }
1369 }
1370 }
1371
1372 if (!V_current_state_timers_running6 &&
1373 !V_state_change_timers_running6)
1374 goto out_locked;
1375
1376 V_current_state_timers_running6 = 0;
1377 V_state_change_timers_running6 = 0;
1378
1379 CTR1(KTR_MLD, "%s: state change timers running", __func__);
1380
1381 /*
1382 * MLD host report and state-change timer processing.
1383 * Note: Processing a v2 group timer may remove a node.
1384 */
1385 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1386 ifp = mli->mli_ifp;
1387
1388 if (mli->mli_version == MLD_VERSION_2) {
1389 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1390 PR_FASTHZ);
1391
1392 memset(&qrq, 0, sizeof(struct ifqueue));
1393 IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS);
1394
1395 memset(&scq, 0, sizeof(struct ifqueue));
1396 IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1397 }
1398
1399 IF_ADDR_LOCK(ifp);
1400 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1401 tifma) {
1402 if (ifma->ifma_addr->sa_family != AF_INET6 ||
1403 ifma->ifma_protospec == NULL)
1404 continue;
1405 inm = (struct in6_multi *)ifma->ifma_protospec;
1406 switch (mli->mli_version) {
1407 case MLD_VERSION_1:
1408 /*
1409 * XXX Drop IF_ADDR lock temporarily to
1410 * avoid recursion caused by a potential
1411 * call by in6ifa_ifpforlinklocal().
1412 * rwlock candidate?
1413 */
1414 IF_ADDR_UNLOCK(ifp);
1415 mld_v1_process_group_timer(inm,
1416 mli->mli_version);
1417 IF_ADDR_LOCK(ifp);
1418 break;
1419 case MLD_VERSION_2:
1420 mld_v2_process_group_timers(mli, &qrq,
1421 &scq, inm, uri_fasthz);
1422 break;
1423 }
1424 }
1425 IF_ADDR_UNLOCK(ifp);
1426
1427 if (mli->mli_version == MLD_VERSION_2) {
1428 struct in6_multi *tinm;
1429
1430 mld_dispatch_queue(&qrq, 0);
1431 mld_dispatch_queue(&scq, 0);
1432
1433 /*
1434 * Free the in_multi reference(s) for
1435 * this lifecycle.
1436 */
1437 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1438 in6m_nrele, tinm) {
1439 SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1440 in6m_nrele);
1441 in6m_release_locked(inm);
1442 }
1443 }
1444 }
1445
1446out_locked:
1447 MLD_UNLOCK();
1448 IN6_MULTI_UNLOCK();
1449}
1450
1451/*
1452 * Update host report group timer.
1453 * Will update the global pending timer flags.
1454 */
1455static void
1456mld_v1_process_group_timer(struct in6_multi *inm, const int version)
1457{
1458 int report_timer_expired;
1459
1460 IN6_MULTI_LOCK_ASSERT();
1461 MLD_LOCK_ASSERT();
1462
1463 if (inm->in6m_timer == 0) {
1464 report_timer_expired = 0;
1465 } else if (--inm->in6m_timer == 0) {
1466 report_timer_expired = 1;
1467 } else {
1468 V_current_state_timers_running6 = 1;
1469 return;
1470 }
1471
1472 switch (inm->in6m_state) {
1473 case MLD_NOT_MEMBER:
1474 case MLD_SILENT_MEMBER:
1475 case MLD_IDLE_MEMBER:
1476 case MLD_LAZY_MEMBER:
1477 case MLD_SLEEPING_MEMBER:
1478 case MLD_AWAKENING_MEMBER:
1479 break;
1480 case MLD_REPORTING_MEMBER:
1481 if (report_timer_expired) {
1482 inm->in6m_state = MLD_IDLE_MEMBER;
1483 (void)mld_v1_transmit_report(inm,
1484 MLD_LISTENER_REPORT);
1485 }
1486 break;
1487 case MLD_G_QUERY_PENDING_MEMBER:
1488 case MLD_SG_QUERY_PENDING_MEMBER:
1489 case MLD_LEAVING_MEMBER:
1490 break;
1491 }
1492}
1493
1494/*
1495 * Update a group's timers for MLDv2.
1496 * Will update the global pending timer flags.
1497 * Note: Unlocked read from mli.
1498 */
1499static void
1500mld_v2_process_group_timers(struct mld_ifinfo *mli,
1501 struct ifqueue *qrq, struct ifqueue *scq,
1502 struct in6_multi *inm, const int uri_fasthz)
1503{
1504 int query_response_timer_expired;
1505 int state_change_retransmit_timer_expired;
1506#ifdef KTR
1507 char ip6tbuf[INET6_ADDRSTRLEN];
1508#endif
1509
1510 IN6_MULTI_LOCK_ASSERT();
1511 MLD_LOCK_ASSERT();
1512
1513 query_response_timer_expired = 0;
1514 state_change_retransmit_timer_expired = 0;
1515
1516 /*
1517 * During a transition from compatibility mode back to MLDv2,
1518 * a group record in REPORTING state may still have its group
1519 * timer active. This is a no-op in this function; it is easier
1520 * to deal with it here than to complicate the slow-timeout path.
1521 */
1522 if (inm->in6m_timer == 0) {
1523 query_response_timer_expired = 0;
1524 } else if (--inm->in6m_timer == 0) {
1525 query_response_timer_expired = 1;
1526 } else {
1527 V_current_state_timers_running6 = 1;
1528 }
1529
1530 if (inm->in6m_sctimer == 0) {
1531 state_change_retransmit_timer_expired = 0;
1532 } else if (--inm->in6m_sctimer == 0) {
1533 state_change_retransmit_timer_expired = 1;
1534 } else {
1535 V_state_change_timers_running6 = 1;
1536 }
1537
1538 /* We are in fasttimo, so be quick about it. */
1539 if (!state_change_retransmit_timer_expired &&
1540 !query_response_timer_expired)
1541 return;
1542
1543 switch (inm->in6m_state) {
1544 case MLD_NOT_MEMBER:
1545 case MLD_SILENT_MEMBER:
1546 case MLD_SLEEPING_MEMBER:
1547 case MLD_LAZY_MEMBER:
1548 case MLD_AWAKENING_MEMBER:
1549 case MLD_IDLE_MEMBER:
1550 break;
1551 case MLD_G_QUERY_PENDING_MEMBER:
1552 case MLD_SG_QUERY_PENDING_MEMBER:
1553 /*
1554 * Respond to a previously pending Group-Specific
1555 * or Group-and-Source-Specific query by enqueueing
1556 * the appropriate Current-State report for
1557 * immediate transmission.
1558 */
1559 if (query_response_timer_expired) {
1560 int retval;
1561
1562 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1563 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1564 0);
1565 CTR2(KTR_MLD, "%s: enqueue record = %d",
1566 __func__, retval);
1567 inm->in6m_state = MLD_REPORTING_MEMBER;
1568 in6m_clear_recorded(inm);
1569 }
1570 /* FALLTHROUGH */
1571 case MLD_REPORTING_MEMBER:
1572 case MLD_LEAVING_MEMBER:
1573 if (state_change_retransmit_timer_expired) {
1574 /*
1575 * State-change retransmission timer fired.
1576 * If there are any further pending retransmissions,
1577 * set the global pending state-change flag, and
1578 * reset the timer.
1579 */
1580 if (--inm->in6m_scrv > 0) {
1581 inm->in6m_sctimer = uri_fasthz;
1582 V_state_change_timers_running6 = 1;
1583 }
1584 /*
1585 * Retransmit the previously computed state-change
1586 * report. If there are no further pending
1587 * retransmissions, the mbuf queue will be consumed.
1588 * Update T0 state to T1 as we have now sent
1589 * a state-change.
1590 */
1591 (void)mld_v2_merge_state_changes(inm, scq);
1592
1593 in6m_commit(inm);
1594 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1595 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1596 inm->in6m_ifp->if_xname);
1597
1598 /*
1599 * If we are leaving the group for good, make sure
1600 * we release MLD's reference to it.
1601 * This release must be deferred using a SLIST,
1602 * as we are called from a loop which traverses
1603 * the in_ifmultiaddr TAILQ.
1604 */
1605 if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1606 inm->in6m_scrv == 0) {
1607 inm->in6m_state = MLD_NOT_MEMBER;
1608 SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1609 inm, in6m_nrele);
1610 }
1611 }
1612 break;
1613 }
1614}
1615
1616/*
1617 * Switch to a different version on the given interface,
1618 * as per Section 9.12.
1619 */
1620static void
1621mld_set_version(struct mld_ifinfo *mli, const int version)
1622{
1623 int old_version_timer;
1624
1625 MLD_LOCK_ASSERT();
1626
1627 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1628 version, mli->mli_ifp, mli->mli_ifp->if_xname);
1629
1630 if (version == MLD_VERSION_1) {
1631 /*
1632 * Compute the "Older Version Querier Present" timer as per
1633 * Section 9.12.
1634 */
1635 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1636 old_version_timer *= PR_SLOWHZ;
1637 mli->mli_v1_timer = old_version_timer;
1638 }
1639
1640 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1641 mli->mli_version = MLD_VERSION_1;
1642 mld_v2_cancel_link_timers(mli);
1643 }
1644}
1645
1646/*
1647 * Cancel pending MLDv2 timers for the given link and all groups
1648 * joined on it; state-change, general-query, and group-query timers.
1649 */
1650static void
1651mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
1652{
1653 struct ifmultiaddr *ifma;
1654 struct ifnet *ifp;
1655 struct in6_multi *inm;
1656
1657 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1658 mli->mli_ifp, mli->mli_ifp->if_xname);
1659
1660 IN6_MULTI_LOCK_ASSERT();
1661 MLD_LOCK_ASSERT();
1662
1663 /*
1664 * Fast-track this potentially expensive operation
1665 * by checking all the global 'timer pending' flags.
1666 */
1667 if (!V_interface_timers_running6 &&
1668 !V_state_change_timers_running6 &&
1669 !V_current_state_timers_running6)
1670 return;
1671
1672 mli->mli_v2_timer = 0;
1673
1674 ifp = mli->mli_ifp;
1675
1676 IF_ADDR_LOCK(ifp);
1677 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1678 if (ifma->ifma_addr->sa_family != AF_INET6)
1679 continue;
1680 inm = (struct in6_multi *)ifma->ifma_protospec;
1681 switch (inm->in6m_state) {
1682 case MLD_NOT_MEMBER:
1683 case MLD_SILENT_MEMBER:
1684 case MLD_IDLE_MEMBER:
1685 case MLD_LAZY_MEMBER:
1686 case MLD_SLEEPING_MEMBER:
1687 case MLD_AWAKENING_MEMBER:
1688 break;
1689 case MLD_LEAVING_MEMBER:
1690 /*
1691 * If we are leaving the group and switching
1692 * version, we need to release the final
1693 * reference held for issuing the INCLUDE {}.
1694 *
1695 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
1696 * around in6m_release_locked(), as it is not
1697 * a recursive mutex.
1698 */
1699 IF_ADDR_UNLOCK(ifp);
1700 in6m_release_locked(inm);
1701 IF_ADDR_LOCK(ifp);
1702 /* FALLTHROUGH */
1703 case MLD_G_QUERY_PENDING_MEMBER:
1704 case MLD_SG_QUERY_PENDING_MEMBER:
1705 in6m_clear_recorded(inm);
1706 /* FALLTHROUGH */
1707 case MLD_REPORTING_MEMBER:
1708 inm->in6m_sctimer = 0;
1709 inm->in6m_timer = 0;
1710 inm->in6m_state = MLD_REPORTING_MEMBER;
1711 /*
1712 * Free any pending MLDv2 state-change records.
1713 */
1714 _IF_DRAIN(&inm->in6m_scq);
1715 break;
1716 }
1717 }
1718 IF_ADDR_UNLOCK(ifp);
1719}
1720
1721/*
1722 * Global slowtimo handler.
1723 * VIMAGE: Timeout handlers are expected to service all vimages.
1724 */
1725void
1726mld_slowtimo(void)
1727{
1728 VNET_ITERATOR_DECL(vnet_iter);
1729
1730 VNET_LIST_RLOCK_NOSLEEP();
1731 VNET_FOREACH(vnet_iter) {
1732 CURVNET_SET(vnet_iter);
1733 mld_slowtimo_vnet();
1734 CURVNET_RESTORE();
1735 }
1736 VNET_LIST_RUNLOCK_NOSLEEP();
1737}
1738
1739/*
1740 * Per-vnet slowtimo handler.
1741 */
1742static void
1743mld_slowtimo_vnet(void)
1744{
1745 struct mld_ifinfo *mli;
1746
1747 MLD_LOCK();
1748
1749 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1750 mld_v1_process_querier_timers(mli);
1751 }
1752
1753 MLD_UNLOCK();
1754}
1755
1756/*
1757 * Update the Older Version Querier Present timers for a link.
1758 * See Section 9.12 of RFC 3810.
1759 */
1760static void
1761mld_v1_process_querier_timers(struct mld_ifinfo *mli)
1762{
1763
1764 MLD_LOCK_ASSERT();
1765
1766 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1767 /*
1768 * MLDv1 Querier Present timer expired; revert to MLDv2.
1769 */
1770 CTR5(KTR_MLD,
1771 "%s: transition from v%d -> v%d on %p(%s)",
1772 __func__, mli->mli_version, MLD_VERSION_2,
1773 mli->mli_ifp, mli->mli_ifp->if_xname);
1774 mli->mli_version = MLD_VERSION_2;
1775 }
1776}
1777
1778/*
1779 * Transmit an MLDv1 report immediately.
1780 */
1781static int
1782mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1783{
1784 struct ifnet *ifp;
1785 struct in6_ifaddr *ia;
1786 struct ip6_hdr *ip6;
1787 struct mbuf *mh, *md;
1788 struct mld_hdr *mld;
1789
1790 IN6_MULTI_LOCK_ASSERT();
1791 MLD_LOCK_ASSERT();
1792
1793 ifp = in6m->in6m_ifp;
1794 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1795 /* ia may be NULL if link-local address is tentative. */
1796
1797 MGETHDR(mh, M_DONTWAIT, MT_HEADER);
1798 if (mh == NULL) {
1799 if (ia != NULL)
1800 ifa_free(&ia->ia_ifa);
1801 return (ENOMEM);
1802 }
1803 MGET(md, M_DONTWAIT, MT_DATA);
1804 if (md == NULL) {
1805 m_free(mh);
1806 if (ia != NULL)
1807 ifa_free(&ia->ia_ifa);
1808 return (ENOMEM);
1809 }
1810 mh->m_next = md;
1811
1812 /*
1813 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1814 * that ether_output() does not need to allocate another mbuf
1815 * for the header in the most common case.
1816 */
1817 MH_ALIGN(mh, sizeof(struct ip6_hdr));
1818 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1819 mh->m_len = sizeof(struct ip6_hdr);
1820
1821 ip6 = mtod(mh, struct ip6_hdr *);
1822 ip6->ip6_flow = 0;
1823 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1824 ip6->ip6_vfc |= IPV6_VERSION;
1825 ip6->ip6_nxt = IPPROTO_ICMPV6;
1826 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1827 ip6->ip6_dst = in6m->in6m_addr;
1828
1829 md->m_len = sizeof(struct mld_hdr);
1830 mld = mtod(md, struct mld_hdr *);
1831 mld->mld_type = type;
1832 mld->mld_code = 0;
1833 mld->mld_cksum = 0;
1834 mld->mld_maxdelay = 0;
1835 mld->mld_reserved = 0;
1836 mld->mld_addr = in6m->in6m_addr;
1837 in6_clearscope(&mld->mld_addr);
1838 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1839 sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1840
1841 mld_save_context(mh, ifp);
1842 mh->m_flags |= M_MLDV1;
1843
1844 mld_dispatch_packet(mh);
1845
1846 if (ia != NULL)
1847 ifa_free(&ia->ia_ifa);
1848 return (0);
1849}
1850
1851/*
1852 * Process a state change from the upper layer for the given IPv6 group.
1853 *
1854 * Each socket holds a reference on the in_multi in its own ip_moptions.
1855 * The socket layer will have made the necessary updates to.the group
1856 * state, it is now up to MLD to issue a state change report if there
1857 * has been any change between T0 (when the last state-change was issued)
1858 * and T1 (now).
1859 *
1860 * We use the MLDv2 state machine at group level. The MLd module
1861 * however makes the decision as to which MLD protocol version to speak.
1862 * A state change *from* INCLUDE {} always means an initial join.
1863 * A state change *to* INCLUDE {} always means a final leave.
1864 *
1865 * If delay is non-zero, and the state change is an initial multicast
1866 * join, the state change report will be delayed by 'delay' ticks
1867 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1868 * the initial MLDv2 state change report will be delayed by whichever
1869 * is sooner, a pending state-change timer or delay itself.
1870 *
1871 * VIMAGE: curvnet should have been set by caller, as this routine
1872 * is called from the socket option handlers.
1873 */
1874int
1875mld_change_state(struct in6_multi *inm, const int delay)
1876{
1877 struct mld_ifinfo *mli;
1878 struct ifnet *ifp;
1879 int error;
1880
1881 IN6_MULTI_LOCK_ASSERT();
1882
1883 error = 0;
1884
1885 /*
1886 * Try to detect if the upper layer just asked us to change state
1887 * for an interface which has now gone away.
1888 */
1889 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1890 ifp = inm->in6m_ifma->ifma_ifp;
1891 if (ifp != NULL) {
1892 /*
1893 * Sanity check that netinet6's notion of ifp is the
1894 * same as net's.
1895 */
1896 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1897 }
1898
1899 MLD_LOCK();
1900
1901 mli = MLD_IFINFO(ifp);
1902 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
1903
1904 /*
1905 * If we detect a state transition to or from MCAST_UNDEFINED
1906 * for this group, then we are starting or finishing an MLD
1907 * life cycle for this group.
1908 */
1909 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1910 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1911 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1912 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1913 CTR1(KTR_MLD, "%s: initial join", __func__);
1914 error = mld_initial_join(inm, mli, delay);
1915 goto out_locked;
1916 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1917 CTR1(KTR_MLD, "%s: final leave", __func__);
1918 mld_final_leave(inm, mli);
1919 goto out_locked;
1920 }
1921 } else {
1922 CTR1(KTR_MLD, "%s: filter set change", __func__);
1923 }
1924
1925 error = mld_handle_state_change(inm, mli);
1926
1927out_locked:
1928 MLD_UNLOCK();
1929 return (error);
1930}
1931
1932/*
1933 * Perform the initial join for an MLD group.
1934 *
1935 * When joining a group:
1936 * If the group should have its MLD traffic suppressed, do nothing.
1937 * MLDv1 starts sending MLDv1 host membership reports.
1938 * MLDv2 will schedule an MLDv2 state-change report containing the
1939 * initial state of the membership.
1940 *
1941 * If the delay argument is non-zero, then we must delay sending the
1942 * initial state change for delay ticks (in units of PR_FASTHZ).
1943 */
1944static int
1945mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
1946 const int delay)
1947{
1948 struct ifnet *ifp;
1949 struct ifqueue *ifq;
1950 int error, retval, syncstates;
1951 int odelay;
1952#ifdef KTR
1953 char ip6tbuf[INET6_ADDRSTRLEN];
1954#endif
1955
1956 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1957 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1958 inm->in6m_ifp, inm->in6m_ifp->if_xname);
1959
1960 error = 0;
1961 syncstates = 1;
1962
1963 ifp = inm->in6m_ifp;
1964
1965 IN6_MULTI_LOCK_ASSERT();
1966 MLD_LOCK_ASSERT();
1967
1968 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1969
1970 /*
1971 * Groups joined on loopback or marked as 'not reported',
1972 * enter the MLD_SILENT_MEMBER state and
1973 * are never reported in any protocol exchanges.
1974 * All other groups enter the appropriate state machine
1975 * for the version in use on this link.
1976 * A link marked as MLIF_SILENT causes MLD to be completely
1977 * disabled for the link.
1978 */
1979 if ((ifp->if_flags & IFF_LOOPBACK) ||
1980 (mli->mli_flags & MLIF_SILENT) ||
1981 !mld_is_addr_reported(&inm->in6m_addr)) {
1982 CTR1(KTR_MLD,
1983"%s: not kicking state machine for silent group", __func__);
1984 inm->in6m_state = MLD_SILENT_MEMBER;
1985 inm->in6m_timer = 0;
1986 } else {
1987 /*
1988 * Deal with overlapping in_multi lifecycle.
1989 * If this group was LEAVING, then make sure
1990 * we drop the reference we picked up to keep the
1991 * group around for the final INCLUDE {} enqueue.
1992 */
1993 if (mli->mli_version == MLD_VERSION_2 &&
1994 inm->in6m_state == MLD_LEAVING_MEMBER)
1995 in6m_release_locked(inm);
1996
1997 inm->in6m_state = MLD_REPORTING_MEMBER;
1998
1999 switch (mli->mli_version) {
2000 case MLD_VERSION_1:
2001 /*
2002 * If a delay was provided, only use it if
2003 * it is greater than the delay normally
2004 * used for an MLDv1 state change report,
2005 * and delay sending the initial MLDv1 report
2006 * by not transitioning to the IDLE state.
2007 */
2008 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2009 if (delay) {
2010 inm->in6m_timer = max(delay, odelay);
2011 V_current_state_timers_running6 = 1;
2012 } else {
2013 inm->in6m_state = MLD_IDLE_MEMBER;
2014 error = mld_v1_transmit_report(inm,
2015 MLD_LISTENER_REPORT);
2016 if (error == 0) {
2017 inm->in6m_timer = odelay;
2018 V_current_state_timers_running6 = 1;
2019 }
2020 }
2021 break;
2022
2023 case MLD_VERSION_2:
2024 /*
2025 * Defer update of T0 to T1, until the first copy
2026 * of the state change has been transmitted.
2027 */
2028 syncstates = 0;
2029
2030 /*
2031 * Immediately enqueue a State-Change Report for
2032 * this interface, freeing any previous reports.
2033 * Don't kick the timers if there is nothing to do,
2034 * or if an error occurred.
2035 */
2036 ifq = &inm->in6m_scq;
2037 _IF_DRAIN(ifq);
2038 retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2039 0, 0, (mli->mli_flags & MLIF_USEALLOW));
2040 CTR2(KTR_MLD, "%s: enqueue record = %d",
2041 __func__, retval);
2042 if (retval <= 0) {
2043 error = retval * -1;
2044 break;
2045 }
2046
2047 /*
2048 * Schedule transmission of pending state-change
2049 * report up to RV times for this link. The timer
2050 * will fire at the next mld_fasttimo (~200ms),
2051 * giving us an opportunity to merge the reports.
2052 *
2053 * If a delay was provided to this function, only
2054 * use this delay if sooner than the existing one.
2055 */
2056 KASSERT(mli->mli_rv > 1,
2057 ("%s: invalid robustness %d", __func__,
2058 mli->mli_rv));
2059 inm->in6m_scrv = mli->mli_rv;
2060 if (delay) {
2061 if (inm->in6m_sctimer > 1) {
2062 inm->in6m_sctimer =
2063 min(inm->in6m_sctimer, delay);
2064 } else
2065 inm->in6m_sctimer = delay;
2066 } else
2067 inm->in6m_sctimer = 1;
2068 V_state_change_timers_running6 = 1;
2069
2070 error = 0;
2071 break;
2072 }
2073 }
2074
2075 /*
2076 * Only update the T0 state if state change is atomic,
2077 * i.e. we don't need to wait for a timer to fire before we
2078 * can consider the state change to have been communicated.
2079 */
2080 if (syncstates) {
2081 in6m_commit(inm);
2082 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2083 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2084 inm->in6m_ifp->if_xname);
2085 }
2086
2087 return (error);
2088}
2089
2090/*
2091 * Issue an intermediate state change during the life-cycle.
2092 */
2093static int
2094mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli)
2095{
2096 struct ifnet *ifp;
2097 int retval;
2098#ifdef KTR
2099 char ip6tbuf[INET6_ADDRSTRLEN];
2100#endif
2101
2102 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2103 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2104 inm->in6m_ifp, inm->in6m_ifp->if_xname);
2105
2106 ifp = inm->in6m_ifp;
2107
2108 IN6_MULTI_LOCK_ASSERT();
2109 MLD_LOCK_ASSERT();
2110
2111 KASSERT(mli && mli->mli_ifp == ifp,
2112 ("%s: inconsistent ifp", __func__));
2113
2114 if ((ifp->if_flags & IFF_LOOPBACK) ||
2115 (mli->mli_flags & MLIF_SILENT) ||
2116 !mld_is_addr_reported(&inm->in6m_addr) ||
2117 (mli->mli_version != MLD_VERSION_2)) {
2118 if (!mld_is_addr_reported(&inm->in6m_addr)) {
2119 CTR1(KTR_MLD,
2120"%s: not kicking state machine for silent group", __func__);
2121 }
2122 CTR1(KTR_MLD, "%s: nothing to do", __func__);
2123 in6m_commit(inm);
2124 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2125 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2126 inm->in6m_ifp->if_xname);
2127 return (0);
2128 }
2129
2130 _IF_DRAIN(&inm->in6m_scq);
2131
2132 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2133 (mli->mli_flags & MLIF_USEALLOW));
2134 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2135 if (retval <= 0)
2136 return (-retval);
2137
2138 /*
2139 * If record(s) were enqueued, start the state-change
2140 * report timer for this group.
2141 */
2142 inm->in6m_scrv = mli->mli_rv;
2143 inm->in6m_sctimer = 1;
2144 V_state_change_timers_running6 = 1;
2145
2146 return (0);
2147}
2148
2149/*
2150 * Perform the final leave for a multicast address.
2151 *
2152 * When leaving a group:
2153 * MLDv1 sends a DONE message, if and only if we are the reporter.
2154 * MLDv2 enqueues a state-change report containing a transition
2155 * to INCLUDE {} for immediate transmission.
2156 */
2157static void
2158mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli)
2159{
2160 int syncstates;
2161#ifdef KTR
2162 char ip6tbuf[INET6_ADDRSTRLEN];
2163#endif
2164
2165 syncstates = 1;
2166
2167 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2168 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2169 inm->in6m_ifp, inm->in6m_ifp->if_xname);
2170
2171 IN6_MULTI_LOCK_ASSERT();
2172 MLD_LOCK_ASSERT();
2173
2174 switch (inm->in6m_state) {
2175 case MLD_NOT_MEMBER:
2176 case MLD_SILENT_MEMBER:
2177 case MLD_LEAVING_MEMBER:
2178 /* Already leaving or left; do nothing. */
2179 CTR1(KTR_MLD,
2180"%s: not kicking state machine for silent group", __func__);
2181 break;
2182 case MLD_REPORTING_MEMBER:
2183 case MLD_IDLE_MEMBER:
2184 case MLD_G_QUERY_PENDING_MEMBER:
2185 case MLD_SG_QUERY_PENDING_MEMBER:
2186 if (mli->mli_version == MLD_VERSION_1) {
2187#ifdef INVARIANTS
2188 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2189 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2190 panic("%s: MLDv2 state reached, not MLDv2 mode",
2191 __func__);
2192#endif
2193 mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2194 inm->in6m_state = MLD_NOT_MEMBER;
2195 } else if (mli->mli_version == MLD_VERSION_2) {
2196 /*
2197 * Stop group timer and all pending reports.
2198 * Immediately enqueue a state-change report
2199 * TO_IN {} to be sent on the next fast timeout,
2200 * giving us an opportunity to merge reports.
2201 */
2202 _IF_DRAIN(&inm->in6m_scq);
2203 inm->in6m_timer = 0;
2204 inm->in6m_scrv = mli->mli_rv;
2205 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2206 "pending retransmissions.", __func__,
2207 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2208 inm->in6m_ifp->if_xname, inm->in6m_scrv);
2209 if (inm->in6m_scrv == 0) {
2210 inm->in6m_state = MLD_NOT_MEMBER;
2211 inm->in6m_sctimer = 0;
2212 } else {
2213 int retval;
2214
2215 in6m_acquire_locked(inm);
2216
2217 retval = mld_v2_enqueue_group_record(
2218 &inm->in6m_scq, inm, 1, 0, 0,
2219 (mli->mli_flags & MLIF_USEALLOW));
2220 KASSERT(retval != 0,
2221 ("%s: enqueue record = %d", __func__,
2222 retval));
2223
2224 inm->in6m_state = MLD_LEAVING_MEMBER;
2225 inm->in6m_sctimer = 1;
2226 V_state_change_timers_running6 = 1;
2227 syncstates = 0;
2228 }
2229 break;
2230 }
2231 break;
2232 case MLD_LAZY_MEMBER:
2233 case MLD_SLEEPING_MEMBER:
2234 case MLD_AWAKENING_MEMBER:
2235 /* Our reports are suppressed; do nothing. */
2236 break;
2237 }
2238
2239 if (syncstates) {
2240 in6m_commit(inm);
2241 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2242 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2243 inm->in6m_ifp->if_xname);
2244 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2245 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2246 __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname);
2247 }
2248}
2249
2250/*
2251 * Enqueue an MLDv2 group record to the given output queue.
2252 *
2253 * If is_state_change is zero, a current-state record is appended.
2254 * If is_state_change is non-zero, a state-change report is appended.
2255 *
2256 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2257 * If is_group_query is zero, and if there is a packet with free space
2258 * at the tail of the queue, it will be appended to providing there
2259 * is enough free space.
2260 * Otherwise a new mbuf packet chain is allocated.
2261 *
2262 * If is_source_query is non-zero, each source is checked to see if
2263 * it was recorded for a Group-Source query, and will be omitted if
2264 * it is not both in-mode and recorded.
2265 *
2266 * If use_block_allow is non-zero, state change reports for initial join
2267 * and final leave, on an inclusive mode group with a source list, will be
2268 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2269 *
2270 * The function will attempt to allocate leading space in the packet
2271 * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2272 *
2273 * If successful the size of all data appended to the queue is returned,
2274 * otherwise an error code less than zero is returned, or zero if
2275 * no record(s) were appended.
2276 */
2277static int
2278mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2279 const int is_state_change, const int is_group_query,
2280 const int is_source_query, const int use_block_allow)
2281{
2282 struct mldv2_record mr;
2283 struct mldv2_record *pmr;
2284 struct ifnet *ifp;
2285 struct ip6_msource *ims, *nims;
2286 struct mbuf *m0, *m, *md;
2287 int error, is_filter_list_change;
2288 int minrec0len, m0srcs, msrcs, nbytes, off;
2289 int record_has_sources;
2290 int now;
2291 int type;
2292 uint8_t mode;
2293#ifdef KTR
2294 char ip6tbuf[INET6_ADDRSTRLEN];
2295#endif
2296
2297 IN6_MULTI_LOCK_ASSERT();
2298
2299 error = 0;
2300 ifp = inm->in6m_ifp;
2301 is_filter_list_change = 0;
2302 m = NULL;
2303 m0 = NULL;
2304 m0srcs = 0;
2305 msrcs = 0;
2306 nbytes = 0;
2307 nims = NULL;
2308 record_has_sources = 1;
2309 pmr = NULL;
2310 type = MLD_DO_NOTHING;
2311 mode = inm->in6m_st[1].iss_fmode;
2312
2313 /*
2314 * If we did not transition out of ASM mode during t0->t1,
2315 * and there are no source nodes to process, we can skip
2316 * the generation of source records.
2317 */
2318 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2319 inm->in6m_nsrc == 0)
2320 record_has_sources = 0;
2321
2322 if (is_state_change) {
2323 /*
2324 * Queue a state change record.
2325 * If the mode did not change, and there are non-ASM
2326 * listeners or source filters present,
2327 * we potentially need to issue two records for the group.
2328 * If there are ASM listeners, and there was no filter
2329 * mode transition of any kind, do nothing.
2330 *
2331 * If we are transitioning to MCAST_UNDEFINED, we need
2332 * not send any sources. A transition to/from this state is
2333 * considered inclusive with some special treatment.
2334 *
2335 * If we are rewriting initial joins/leaves to use
2336 * ALLOW/BLOCK, and the group's membership is inclusive,
2337 * we need to send sources in all cases.
2338 */
2339 if (mode != inm->in6m_st[0].iss_fmode) {
2340 if (mode == MCAST_EXCLUDE) {
2341 CTR1(KTR_MLD, "%s: change to EXCLUDE",
2342 __func__);
2343 type = MLD_CHANGE_TO_EXCLUDE_MODE;
2344 } else {
2345 CTR1(KTR_MLD, "%s: change to INCLUDE",
2346 __func__);
2347 if (use_block_allow) {
2348 /*
2349 * XXX
2350 * Here we're interested in state
2351 * edges either direction between
2352 * MCAST_UNDEFINED and MCAST_INCLUDE.
2353 * Perhaps we should just check
2354 * the group state, rather than
2355 * the filter mode.
2356 */
2357 if (mode == MCAST_UNDEFINED) {
2358 type = MLD_BLOCK_OLD_SOURCES;
2359 } else {
2360 type = MLD_ALLOW_NEW_SOURCES;
2361 }
2362 } else {
2363 type = MLD_CHANGE_TO_INCLUDE_MODE;
2364 if (mode == MCAST_UNDEFINED)
2365 record_has_sources = 0;
2366 }
2367 }
2368 } else {
2369 if (record_has_sources) {
2370 is_filter_list_change = 1;
2371 } else {
2372 type = MLD_DO_NOTHING;
2373 }
2374 }
2375 } else {
2376 /*
2377 * Queue a current state record.
2378 */
2379 if (mode == MCAST_EXCLUDE) {
2380 type = MLD_MODE_IS_EXCLUDE;
2381 } else if (mode == MCAST_INCLUDE) {
2382 type = MLD_MODE_IS_INCLUDE;
2383 KASSERT(inm->in6m_st[1].iss_asm == 0,
2384 ("%s: inm %p is INCLUDE but ASM count is %d",
2385 __func__, inm, inm->in6m_st[1].iss_asm));
2386 }
2387 }
2388
2389 /*
2390 * Generate the filter list changes using a separate function.
2391 */
2392 if (is_filter_list_change)
2393 return (mld_v2_enqueue_filter_change(ifq, inm));
2394
2395 if (type == MLD_DO_NOTHING) {
2396 CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2397 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2398 inm->in6m_ifp->if_xname);
2399 return (0);
2400 }
2401
2402 /*
2403 * If any sources are present, we must be able to fit at least
2404 * one in the trailing space of the tail packet's mbuf,
2405 * ideally more.
2406 */
2407 minrec0len = sizeof(struct mldv2_record);
2408 if (record_has_sources)
2409 minrec0len += sizeof(struct in6_addr);
2410
2411 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2412 mld_rec_type_to_str(type),
2413 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2414 inm->in6m_ifp->if_xname);
2415
2416 /*
2417 * Check if we have a packet in the tail of the queue for this
2418 * group into which the first group record for this group will fit.
2419 * Otherwise allocate a new packet.
2420 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2421 * Note: Group records for G/GSR query responses MUST be sent
2422 * in their own packet.
2423 */
2424 m0 = ifq->ifq_tail;
2425 if (!is_group_query &&
2426 m0 != NULL &&
2427 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2428 (m0->m_pkthdr.len + minrec0len) <
2429 (ifp->if_mtu - MLD_MTUSPACE)) {
2430 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2431 sizeof(struct mldv2_record)) /
2432 sizeof(struct in6_addr);
2433 m = m0;
2434 CTR1(KTR_MLD, "%s: use existing packet", __func__);
2435 } else {
2436 if (_IF_QFULL(ifq)) {
2437 CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2438 return (-ENOMEM);
2439 }
2440 m = NULL;
2441 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2442 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2443 if (!is_state_change && !is_group_query)
2444 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2445 if (m == NULL)
2446 m = m_gethdr(M_DONTWAIT, MT_DATA);
2447 if (m == NULL)
2448 return (-ENOMEM);
2449
2450 mld_save_context(m, ifp);
2451
2452 CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2453 }
2454
2455 /*
2456 * Append group record.
2457 * If we have sources, we don't know how many yet.
2458 */
2459 mr.mr_type = type;
2460 mr.mr_datalen = 0;
2461 mr.mr_numsrc = 0;
2462 mr.mr_addr = inm->in6m_addr;
2463 in6_clearscope(&mr.mr_addr);
2464 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2465 if (m != m0)
2466 m_freem(m);
2467 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2468 return (-ENOMEM);
2469 }
2470 nbytes += sizeof(struct mldv2_record);
2471
2472 /*
2473 * Append as many sources as will fit in the first packet.
2474 * If we are appending to a new packet, the chain allocation
2475 * may potentially use clusters; use m_getptr() in this case.
2476 * If we are appending to an existing packet, we need to obtain
2477 * a pointer to the group record after m_append(), in case a new
2478 * mbuf was allocated.
2479 *
2480 * Only append sources which are in-mode at t1. If we are
2481 * transitioning to MCAST_UNDEFINED state on the group, and
2482 * use_block_allow is zero, do not include source entries.
2483 * Otherwise, we need to include this source in the report.
2484 *
2485 * Only report recorded sources in our filter set when responding
2486 * to a group-source query.
2487 */
2488 if (record_has_sources) {
2489 if (m == m0) {
2490 md = m_last(m);
2491 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2492 md->m_len - nbytes);
2493 } else {
2494 md = m_getptr(m, 0, &off);
2495 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2496 off);
2497 }
2498 msrcs = 0;
2499 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2500 nims) {
2501 CTR2(KTR_MLD, "%s: visit node %s", __func__,
2502 ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2503 now = im6s_get_mode(inm, ims, 1);
2504 CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2505 if ((now != mode) ||
2506 (now == mode &&
2507 (!use_block_allow && mode == MCAST_UNDEFINED))) {
2508 CTR1(KTR_MLD, "%s: skip node", __func__);
2509 continue;
2510 }
2511 if (is_source_query && ims->im6s_stp == 0) {
2512 CTR1(KTR_MLD, "%s: skip unrecorded node",
2513 __func__);
2514 continue;
2515 }
2516 CTR1(KTR_MLD, "%s: append node", __func__);
2517 if (!m_append(m, sizeof(struct in6_addr),
2518 (void *)&ims->im6s_addr)) {
2519 if (m != m0)
2520 m_freem(m);
2521 CTR1(KTR_MLD, "%s: m_append() failed.",
2522 __func__);
2523 return (-ENOMEM);
2524 }
2525 nbytes += sizeof(struct in6_addr);
2526 ++msrcs;
2527 if (msrcs == m0srcs)
2528 break;
2529 }
2530 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2531 msrcs);
2532 pmr->mr_numsrc = htons(msrcs);
2533 nbytes += (msrcs * sizeof(struct in6_addr));
2534 }
2535
2536 if (is_source_query && msrcs == 0) {
2537 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2538 if (m != m0)
2539 m_freem(m);
2540 return (0);
2541 }
2542
2543 /*
2544 * We are good to go with first packet.
2545 */
2546 if (m != m0) {
2547 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2548 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2549 _IF_ENQUEUE(ifq, m);
2550 } else
2551 m->m_pkthdr.PH_vt.vt_nrecs++;
2552
2553 /*
2554 * No further work needed if no source list in packet(s).
2555 */
2556 if (!record_has_sources)
2557 return (nbytes);
2558
2559 /*
2560 * Whilst sources remain to be announced, we need to allocate
2561 * a new packet and fill out as many sources as will fit.
2562 * Always try for a cluster first.
2563 */
2564 while (nims != NULL) {
2565 if (_IF_QFULL(ifq)) {
2566 CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2567 return (-ENOMEM);
2568 }
2569 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2570 if (m == NULL)
2571 m = m_gethdr(M_DONTWAIT, MT_DATA);
2572 if (m == NULL)
2573 return (-ENOMEM);
2574 mld_save_context(m, ifp);
2575 md = m_getptr(m, 0, &off);
2576 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2577 CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2578
2579 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2580 if (m != m0)
2581 m_freem(m);
2582 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2583 return (-ENOMEM);
2584 }
2585 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2586 nbytes += sizeof(struct mldv2_record);
2587
2588 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2589 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2590
2591 msrcs = 0;
2592 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2593 CTR2(KTR_MLD, "%s: visit node %s",
2594 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2595 now = im6s_get_mode(inm, ims, 1);
2596 if ((now != mode) ||
2597 (now == mode &&
2598 (!use_block_allow && mode == MCAST_UNDEFINED))) {
2599 CTR1(KTR_MLD, "%s: skip node", __func__);
2600 continue;
2601 }
2602 if (is_source_query && ims->im6s_stp == 0) {
2603 CTR1(KTR_MLD, "%s: skip unrecorded node",
2604 __func__);
2605 continue;
2606 }
2607 CTR1(KTR_MLD, "%s: append node", __func__);
2608 if (!m_append(m, sizeof(struct in6_addr),
2609 (void *)&ims->im6s_addr)) {
2610 if (m != m0)
2611 m_freem(m);
2612 CTR1(KTR_MLD, "%s: m_append() failed.",
2613 __func__);
2614 return (-ENOMEM);
2615 }
2616 ++msrcs;
2617 if (msrcs == m0srcs)
2618 break;
2619 }
2620 pmr->mr_numsrc = htons(msrcs);
2621 nbytes += (msrcs * sizeof(struct in6_addr));
2622
2623 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2624 _IF_ENQUEUE(ifq, m);
2625 }
2626
2627 return (nbytes);
2628}
2629
2630/*
2631 * Type used to mark record pass completion.
2632 * We exploit the fact we can cast to this easily from the
2633 * current filter modes on each ip_msource node.
2634 */
2635typedef enum {
2636 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2637 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2638 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2639 REC_FULL = REC_ALLOW | REC_BLOCK
2640} rectype_t;
2641
2642/*
2643 * Enqueue an MLDv2 filter list change to the given output queue.
2644 *
2645 * Source list filter state is held in an RB-tree. When the filter list
2646 * for a group is changed without changing its mode, we need to compute
2647 * the deltas between T0 and T1 for each source in the filter set,
2648 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2649 *
2650 * As we may potentially queue two record types, and the entire R-B tree
2651 * needs to be walked at once, we break this out into its own function
2652 * so we can generate a tightly packed queue of packets.
2653 *
2654 * XXX This could be written to only use one tree walk, although that makes
2655 * serializing into the mbuf chains a bit harder. For now we do two walks
2656 * which makes things easier on us, and it may or may not be harder on
2657 * the L2 cache.
2658 *
2659 * If successful the size of all data appended to the queue is returned,
2660 * otherwise an error code less than zero is returned, or zero if
2661 * no record(s) were appended.
2662 */
2663static int
2664mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
2665{
2666 static const int MINRECLEN =
2667 sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2668 struct ifnet *ifp;
2669 struct mldv2_record mr;
2670 struct mldv2_record *pmr;
2671 struct ip6_msource *ims, *nims;
2672 struct mbuf *m, *m0, *md;
2673 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2674 int nallow, nblock;
2675 uint8_t mode, now, then;
2676 rectype_t crt, drt, nrt;
2677#ifdef KTR
2678 char ip6tbuf[INET6_ADDRSTRLEN];
2679#endif
2680
2681 IN6_MULTI_LOCK_ASSERT();
2682
2683 if (inm->in6m_nsrc == 0 ||
2684 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2685 return (0);
2686
2687 ifp = inm->in6m_ifp; /* interface */
2688 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */
2689 crt = REC_NONE; /* current group record type */
2690 drt = REC_NONE; /* mask of completed group record types */
2691 nrt = REC_NONE; /* record type for current node */
2692 m0srcs = 0; /* # source which will fit in current mbuf chain */
2693 npbytes = 0; /* # of bytes appended this packet */
2694 nbytes = 0; /* # of bytes appended to group's state-change queue */
2695 rsrcs = 0; /* # sources encoded in current record */
2696 schanged = 0; /* # nodes encoded in overall filter change */
2697 nallow = 0; /* # of source entries in ALLOW_NEW */
2698 nblock = 0; /* # of source entries in BLOCK_OLD */
2699 nims = NULL; /* next tree node pointer */
2700
2701 /*
2702 * For each possible filter record mode.
2703 * The first kind of source we encounter tells us which
2704 * is the first kind of record we start appending.
2705 * If a node transitioned to UNDEFINED at t1, its mode is treated
2706 * as the inverse of the group's filter mode.
2707 */
2708 while (drt != REC_FULL) {
2709 do {
2710 m0 = ifq->ifq_tail;
2711 if (m0 != NULL &&
2712 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2713 MLD_V2_REPORT_MAXRECS) &&
2714 (m0->m_pkthdr.len + MINRECLEN) <
2715 (ifp->if_mtu - MLD_MTUSPACE)) {
2716 m = m0;
2717 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2718 sizeof(struct mldv2_record)) /
2719 sizeof(struct in6_addr);
2720 CTR1(KTR_MLD,
2721 "%s: use previous packet", __func__);
2722 } else {
2723 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2724 if (m == NULL)
2725 m = m_gethdr(M_DONTWAIT, MT_DATA);
2726 if (m == NULL) {
2727 CTR1(KTR_MLD,
2728 "%s: m_get*() failed", __func__);
2729 return (-ENOMEM);
2730 }
2731 m->m_pkthdr.PH_vt.vt_nrecs = 0;
2732 mld_save_context(m, ifp);
2733 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2734 sizeof(struct mldv2_record)) /
2735 sizeof(struct in6_addr);
2736 npbytes = 0;
2737 CTR1(KTR_MLD,
2738 "%s: allocated new packet", __func__);
2739 }
2740 /*
2741 * Append the MLD group record header to the
2742 * current packet's data area.
2743 * Recalculate pointer to free space for next
2744 * group record, in case m_append() allocated
2745 * a new mbuf or cluster.
2746 */
2747 memset(&mr, 0, sizeof(mr));
2748 mr.mr_addr = inm->in6m_addr;
2749 in6_clearscope(&mr.mr_addr);
2750 if (!m_append(m, sizeof(mr), (void *)&mr)) {
2751 if (m != m0)
2752 m_freem(m);
2753 CTR1(KTR_MLD,
2754 "%s: m_append() failed", __func__);
2755 return (-ENOMEM);
2756 }
2757 npbytes += sizeof(struct mldv2_record);
2758 if (m != m0) {
2759 /* new packet; offset in chain */
2760 md = m_getptr(m, npbytes -
2761 sizeof(struct mldv2_record), &off);
2762 pmr = (struct mldv2_record *)(mtod(md,
2763 uint8_t *) + off);
2764 } else {
2765 /* current packet; offset from last append */
2766 md = m_last(m);
2767 pmr = (struct mldv2_record *)(mtod(md,
2768 uint8_t *) + md->m_len -
2769 sizeof(struct mldv2_record));
2770 }
2771 /*
2772 * Begin walking the tree for this record type
2773 * pass, or continue from where we left off
2774 * previously if we had to allocate a new packet.
2775 * Only report deltas in-mode at t1.
2776 * We need not report included sources as allowed
2777 * if we are in inclusive mode on the group,
2778 * however the converse is not true.
2779 */
2780 rsrcs = 0;
2781 if (nims == NULL) {
2782 nims = RB_MIN(ip6_msource_tree,
2783 &inm->in6m_srcs);
2784 }
2785 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2786 CTR2(KTR_MLD, "%s: visit node %s", __func__,
2787 ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2788 now = im6s_get_mode(inm, ims, 1);
2789 then = im6s_get_mode(inm, ims, 0);
2790 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2791 __func__, then, now);
2792 if (now == then) {
2793 CTR1(KTR_MLD,
2794 "%s: skip unchanged", __func__);
2795 continue;
2796 }
2797 if (mode == MCAST_EXCLUDE &&
2798 now == MCAST_INCLUDE) {
2799 CTR1(KTR_MLD,
2800 "%s: skip IN src on EX group",
2801 __func__);
2802 continue;
2803 }
2804 nrt = (rectype_t)now;
2805 if (nrt == REC_NONE)
2806 nrt = (rectype_t)(~mode & REC_FULL);
2807 if (schanged++ == 0) {
2808 crt = nrt;
2809 } else if (crt != nrt)
2810 continue;
2811 if (!m_append(m, sizeof(struct in6_addr),
2812 (void *)&ims->im6s_addr)) {
2813 if (m != m0)
2814 m_freem(m);
2815 CTR1(KTR_MLD,
2816 "%s: m_append() failed", __func__);
2817 return (-ENOMEM);
2818 }
2819 nallow += !!(crt == REC_ALLOW);
2820 nblock += !!(crt == REC_BLOCK);
2821 if (++rsrcs == m0srcs)
2822 break;
2823 }
2824 /*
2825 * If we did not append any tree nodes on this
2826 * pass, back out of allocations.
2827 */
2828 if (rsrcs == 0) {
2829 npbytes -= sizeof(struct mldv2_record);
2830 if (m != m0) {
2831 CTR1(KTR_MLD,
2832 "%s: m_free(m)", __func__);
2833 m_freem(m);
2834 } else {
2835 CTR1(KTR_MLD,
2836 "%s: m_adj(m, -mr)", __func__);
2837 m_adj(m, -((int)sizeof(
2838 struct mldv2_record)));
2839 }
2840 continue;
2841 }
2842 npbytes += (rsrcs * sizeof(struct in6_addr));
2843 if (crt == REC_ALLOW)
2844 pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2845 else if (crt == REC_BLOCK)
2846 pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2847 pmr->mr_numsrc = htons(rsrcs);
2848 /*
2849 * Count the new group record, and enqueue this
2850 * packet if it wasn't already queued.
2851 */
2852 m->m_pkthdr.PH_vt.vt_nrecs++;
2853 if (m != m0)
2854 _IF_ENQUEUE(ifq, m);
2855 nbytes += npbytes;
2856 } while (nims != NULL);
2857 drt |= crt;
2858 crt = (~crt & REC_FULL);
2859 }
2860
2861 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2862 nallow, nblock);
2863
2864 return (nbytes);
2865}
2866
2867static int
2868mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
2869{
2870 struct ifqueue *gq;
2871 struct mbuf *m; /* pending state-change */
2872 struct mbuf *m0; /* copy of pending state-change */
2873 struct mbuf *mt; /* last state-change in packet */
2874 int docopy, domerge;
2875 u_int recslen;
2876
2877 docopy = 0;
2878 domerge = 0;
2879 recslen = 0;
2880
2881 IN6_MULTI_LOCK_ASSERT();
2882 MLD_LOCK_ASSERT();
2883
2884 /*
2885 * If there are further pending retransmissions, make a writable
2886 * copy of each queued state-change message before merging.
2887 */
2888 if (inm->in6m_scrv > 0)
2889 docopy = 1;
2890
2891 gq = &inm->in6m_scq;
2892#ifdef KTR
2893 if (gq->ifq_head == NULL) {
2894 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2895 __func__, inm);
2896 }
2897#endif
2898
2899 m = gq->ifq_head;
2900 while (m != NULL) {
2901 /*
2902 * Only merge the report into the current packet if
2903 * there is sufficient space to do so; an MLDv2 report
2904 * packet may only contain 65,535 group records.
2905 * Always use a simple mbuf chain concatentation to do this,
2906 * as large state changes for single groups may have
2907 * allocated clusters.
2908 */
2909 domerge = 0;
2910 mt = ifscq->ifq_tail;
2911 if (mt != NULL) {
2912 recslen = m_length(m, NULL);
2913
2914 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2915 m->m_pkthdr.PH_vt.vt_nrecs <=
2916 MLD_V2_REPORT_MAXRECS) &&
2917 (mt->m_pkthdr.len + recslen <=
2918 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2919 domerge = 1;
2920 }
2921
2922 if (!domerge && _IF_QFULL(gq)) {
2923 CTR2(KTR_MLD,
2924 "%s: outbound queue full, skipping whole packet %p",
2925 __func__, m);
2926 mt = m->m_nextpkt;
2927 if (!docopy)
2928 m_freem(m);
2929 m = mt;
2930 continue;
2931 }
2932
2933 if (!docopy) {
2934 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2935 _IF_DEQUEUE(gq, m0);
2936 m = m0->m_nextpkt;
2937 } else {
2938 CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2939 m0 = m_dup(m, M_NOWAIT);
2940 if (m0 == NULL)
2941 return (ENOMEM);
2942 m0->m_nextpkt = NULL;
2943 m = m->m_nextpkt;
2944 }
2945
2946 if (!domerge) {
2947 CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)",
2948 __func__, m0, ifscq);
2949 _IF_ENQUEUE(ifscq, m0);
2950 } else {
2951 struct mbuf *mtl; /* last mbuf of packet mt */
2952
2953 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2954 __func__, m0, mt);
2955
2956 mtl = m_last(mt);
2957 m0->m_flags &= ~M_PKTHDR;
2958 mt->m_pkthdr.len += recslen;
2959 mt->m_pkthdr.PH_vt.vt_nrecs +=
2960 m0->m_pkthdr.PH_vt.vt_nrecs;
2961
2962 mtl->m_next = m0;
2963 }
2964 }
2965
2966 return (0);
2967}
2968
2969/*
2970 * Respond to a pending MLDv2 General Query.
2971 */
2972static void
2973mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
2974{
2975 struct ifmultiaddr *ifma, *tifma;
2976 struct ifnet *ifp;
2977 struct in6_multi *inm;
2978 int retval;
2979
2980 IN6_MULTI_LOCK_ASSERT();
2981 MLD_LOCK_ASSERT();
2982
2983 KASSERT(mli->mli_version == MLD_VERSION_2,
2984 ("%s: called when version %d", __func__, mli->mli_version));
2985
2986 ifp = mli->mli_ifp;
2987
2988 IF_ADDR_LOCK(ifp);
2989 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
2990 if (ifma->ifma_addr->sa_family != AF_INET6 ||
2991 ifma->ifma_protospec == NULL)
2992 continue;
2993
2994 inm = (struct in6_multi *)ifma->ifma_protospec;
2995 KASSERT(ifp == inm->in6m_ifp,
2996 ("%s: inconsistent ifp", __func__));
2997
2998 switch (inm->in6m_state) {
2999 case MLD_NOT_MEMBER:
3000 case MLD_SILENT_MEMBER:
3001 break;
3002 case MLD_REPORTING_MEMBER:
3003 case MLD_IDLE_MEMBER:
3004 case MLD_LAZY_MEMBER:
3005 case MLD_SLEEPING_MEMBER:
3006 case MLD_AWAKENING_MEMBER:
3007 inm->in6m_state = MLD_REPORTING_MEMBER;
3008 retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3009 inm, 0, 0, 0, 0);
3010 CTR2(KTR_MLD, "%s: enqueue record = %d",
3011 __func__, retval);
3012 break;
3013 case MLD_G_QUERY_PENDING_MEMBER:
3014 case MLD_SG_QUERY_PENDING_MEMBER:
3015 case MLD_LEAVING_MEMBER:
3016 break;
3017 }
3018 }
3019 IF_ADDR_UNLOCK(ifp);
3020
3021 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3022
3023 /*
3024 * Slew transmission of bursts over 500ms intervals.
3025 */
3026 if (mli->mli_gq.ifq_head != NULL) {
3027 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3028 MLD_RESPONSE_BURST_INTERVAL);
3029 V_interface_timers_running6 = 1;
3030 }
3031}
3032
3033/*
3034 * Transmit the next pending message in the output queue.
3035 *
3036 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3037 * MRT: Nothing needs to be done, as MLD traffic is always local to
3038 * a link and uses a link-scope multicast address.
3039 */
3040static void
3041mld_dispatch_packet(struct mbuf *m)
3042{
3043 struct ip6_moptions im6o;
3044 struct ifnet *ifp;
3045 struct ifnet *oifp;
3046 struct mbuf *m0;
3047 struct mbuf *md;
3048 struct ip6_hdr *ip6;
3049 struct mld_hdr *mld;
3050 int error;
3051 int off;
3052 int type;
3053 uint32_t ifindex;
3054
3055 CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3056
3057 /*
3058 * Set VNET image pointer from enqueued mbuf chain
3059 * before doing anything else. Whilst we use interface
3060 * indexes to guard against interface detach, they are
3061 * unique to each VIMAGE and must be retrieved.
3062 */
3063 ifindex = mld_restore_context(m);
3064
3065 /*
3066 * Check if the ifnet still exists. This limits the scope of
3067 * any race in the absence of a global ifp lock for low cost
3068 * (an array lookup).
3069 */
3070 ifp = ifnet_byindex(ifindex);
3071 if (ifp == NULL) {
3072 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3073 __func__, m, ifindex);
3074 m_freem(m);
3075 IP6STAT_INC(ip6s_noroute);
3076 goto out;
3077 }
3078
3079 im6o.im6o_multicast_hlim = 1;
3080 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3081 im6o.im6o_multicast_ifp = ifp;
3082
3083 if (m->m_flags & M_MLDV1) {
3084 m0 = m;
3085 } else {
3086 m0 = mld_v2_encap_report(ifp, m);
3087 if (m0 == NULL) {
3088 CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3089 m_freem(m);
3090 IP6STAT_INC(ip6s_odropped);
3091 goto out;
3092 }
3093 }
3094
3095 mld_scrub_context(m0);
3096 m->m_flags &= ~(M_PROTOFLAGS);
3097 m0->m_pkthdr.rcvif = V_loif;
3098
3099 ip6 = mtod(m0, struct ip6_hdr *);
3100#if 0
3101 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */
3102#else
3103 /*
3104 * XXX XXX Break some KPI rules to prevent an LOR which would
3105 * occur if we called in6_setscope() at transmission.
3106 * See comments at top of file.
3107 */
3108 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3109#endif
3110
3111 /*
3112 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3113 * so we can bump the stats.
3114 */
3115 md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3116 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3117 type = mld->mld_type;
3118
3119 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3120 &oifp, NULL);
3121 if (error) {
3122 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3123 goto out;
3124 }
3125 ICMP6STAT_INC(icp6s_outhist[type]);
3126 if (oifp != NULL) {
3127 icmp6_ifstat_inc(oifp, ifs6_out_msg);
3128 switch (type) {
3129 case MLD_LISTENER_REPORT:
3130 case MLDV2_LISTENER_REPORT:
3131 icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3132 break;
3133 case MLD_LISTENER_DONE:
3134 icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3135 break;
3136 }
3137 }
3138out:
3139 return;
3140}
3141
3142/*
3143 * Encapsulate an MLDv2 report.
3144 *
3145 * KAME IPv6 requires that hop-by-hop options be passed separately,
3146 * and that the IPv6 header be prepended in a separate mbuf.
3147 *
3148 * Returns a pointer to the new mbuf chain head, or NULL if the
3149 * allocation failed.
3150 */
3151static struct mbuf *
3152mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3153{
3154 struct mbuf *mh;
3155 struct mldv2_report *mld;
3156 struct ip6_hdr *ip6;
3157 struct in6_ifaddr *ia;
3158 int mldreclen;
3159
3160 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3161 KASSERT((m->m_flags & M_PKTHDR),
3162 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3163
3164 /*
3165 * RFC3590: OK to send as :: or tentative during DAD.
3166 */
3167 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3168 if (ia == NULL)
3169 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3170
3171 MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3172 if (mh == NULL) {
3173 if (ia != NULL)
3174 ifa_free(&ia->ia_ifa);
3175 m_freem(m);
3176 return (NULL);
3177 }
3178 MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3179
3180 mldreclen = m_length(m, NULL);
3181 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3182
3183 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3184 mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3185 sizeof(struct mldv2_report) + mldreclen;
3186
3187 ip6 = mtod(mh, struct ip6_hdr *);
3188 ip6->ip6_flow = 0;
3189 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3190 ip6->ip6_vfc |= IPV6_VERSION;
3191 ip6->ip6_nxt = IPPROTO_ICMPV6;
3192 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3193 if (ia != NULL)
3194 ifa_free(&ia->ia_ifa);
3195 ip6->ip6_dst = in6addr_linklocal_allv2routers;
3196 /* scope ID will be set in netisr */
3197
3198 mld = (struct mldv2_report *)(ip6 + 1);
3199 mld->mld_type = MLDV2_LISTENER_REPORT;
3200 mld->mld_code = 0;
3201 mld->mld_cksum = 0;
3202 mld->mld_v2_reserved = 0;
3203 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3204 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3205
3206 mh->m_next = m;
3207 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3208 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3209 return (mh);
3210}
3211
3212#ifdef KTR
3213static char *
3214mld_rec_type_to_str(const int type)
3215{
3216
3217 switch (type) {
3218 case MLD_CHANGE_TO_EXCLUDE_MODE:
3219 return "TO_EX";
3220 break;
3221 case MLD_CHANGE_TO_INCLUDE_MODE:
3222 return "TO_IN";
3223 break;
3224 case MLD_MODE_IS_EXCLUDE:
3225 return "MODE_EX";
3226 break;
3227 case MLD_MODE_IS_INCLUDE:
3228 return "MODE_IN";
3229 break;
3230 case MLD_ALLOW_NEW_SOURCES:
3231 return "ALLOW_NEW";
3232 break;
3233 case MLD_BLOCK_OLD_SOURCES:
3234 return "BLOCK_OLD";
3235 break;
3236 default:
3237 break;
3238 }
3239 return "unknown";
3240}
3241#endif
3242
3243static void
3244mld_init(void *unused __unused)
3245{
3246
3247 CTR1(KTR_MLD, "%s: initializing", __func__);
3248 MLD_LOCK_INIT();
3249
3250 ip6_initpktopts(&mld_po);
3251 mld_po.ip6po_hlim = 1;
3252 mld_po.ip6po_hbh = &mld_ra.hbh;
3253 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3254 mld_po.ip6po_flags = IP6PO_DONTFRAG;
3255}
3256SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL);
3257
3258static void
3259mld_uninit(void *unused __unused)
3260{
3261
3262 CTR1(KTR_MLD, "%s: tearing down", __func__);
3263 MLD_LOCK_DESTROY();
3264}
3265SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL);
3266
3267static void
3268vnet_mld_init(const void *unused __unused)
3269{
3270
3271 CTR1(KTR_MLD, "%s: initializing", __func__);
3272
3273 LIST_INIT(&V_mli_head);
3274}
3275VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init,
3276 NULL);
3277
3278static void
3279vnet_mld_uninit(const void *unused __unused)
3280{
3281
3282 CTR1(KTR_MLD, "%s: tearing down", __func__);
3283
3284 KASSERT(LIST_EMPTY(&V_mli_head),
3285 ("%s: mli list not empty; ifnets not detached?", __func__));
3286}
3287VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit,
3288 NULL);
3289
3290static int
3291mld_modevent(module_t mod, int type, void *unused __unused)
3292{
3293
3294 switch (type) {
3295 case MOD_LOAD:
3296 case MOD_UNLOAD:
3297 break;
3298 default:
3299 return (EOPNOTSUPP);
3300 }
3301 return (0);
3302}
3303
3304static moduledata_t mld_mod = {
3305 "mld",
3306 mld_modevent,
3307 0
3308};
3309DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
|