xref: /freebsd/sys/netinet6/mld6.c (revision cba9f88105c268f73fa2a92ca4479cc15b8a3338)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote
15  *    products derived from this software without specific prior written
16  *    permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
31  */
32 
33 /*-
34  * Copyright (c) 1988 Stephen Deering.
35  * Copyright (c) 1992, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * This code is derived from software contributed to Berkeley by
39  * Stephen Deering of Stanford University.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  */
65 
66 #include "opt_inet.h"
67 #include "opt_inet6.h"
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/mbuf.h>
72 #include <sys/socket.h>
73 #include <sys/sysctl.h>
74 #include <sys/kernel.h>
75 #include <sys/callout.h>
76 #include <sys/malloc.h>
77 #include <sys/module.h>
78 #include <sys/ktr.h>
79 
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/if_private.h>
83 #include <net/route.h>
84 #include <net/vnet.h>
85 
86 #include <netinet/in.h>
87 #include <netinet/in_var.h>
88 #include <netinet6/in6_var.h>
89 #include <netinet/ip6.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #include <netinet/icmp6.h>
93 #include <netinet6/mld6.h>
94 #include <netinet6/mld6_var.h>
95 
96 #include <security/mac/mac_framework.h>
97 
98 #ifndef KTR_MLD
99 #define KTR_MLD KTR_INET6
100 #endif
101 
102 static void	mld_dispatch_packet(struct mbuf *);
103 static void	mld_dispatch_queue(struct mbufq *, int);
104 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
105 static void	mld_fasttimo_vnet(struct in6_multi_head *inmh);
106 static int	mld_handle_state_change(struct in6_multi *,
107 		    struct mld_ifsoftc *);
108 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
109 		    const int);
110 #ifdef KTR
111 static char *	mld_rec_type_to_str(const int);
112 #endif
113 static void	mld_set_version(struct mld_ifsoftc *, const int);
114 static void	mld_slowtimo_vnet(void);
115 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
116 		    /*const*/ struct mld_hdr *);
117 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
118 		    /*const*/ struct mld_hdr *);
119 static void	mld_v1_process_group_timer(struct in6_multi_head *,
120 		    struct in6_multi *);
121 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
122 static int	mld_v1_transmit_report(struct in6_multi *, const int);
123 static void	mld_v1_update_group(struct in6_multi *, const int);
124 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
125 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
126 static struct mbuf *
127 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
128 static int	mld_v2_enqueue_filter_change(struct mbufq *,
129 		    struct in6_multi *);
130 static int	mld_v2_enqueue_group_record(struct mbufq *,
131 		    struct in6_multi *, const int, const int, const int,
132 		    const int);
133 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
134 		    struct mbuf *, struct mldv2_query *, const int, const int);
135 static int	mld_v2_merge_state_changes(struct in6_multi *,
136 		    struct mbufq *);
137 static void	mld_v2_process_group_timers(struct in6_multi_head *,
138 		    struct mbufq *, struct mbufq *,
139 		    struct in6_multi *, const int);
140 static int	mld_v2_process_group_query(struct in6_multi *,
141 		    struct mld_ifsoftc *mli, int, struct mbuf *,
142 		    struct mldv2_query *, const int);
143 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
144 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
145 
146 /*
147  * Normative references: RFC 2710, RFC 3590, RFC 3810.
148  *
149  * Locking:
150  *  * The MLD subsystem lock ends up being system-wide for the moment,
151  *    but could be per-VIMAGE later on.
152  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
153  *    Any may be taken independently; if any are held at the same
154  *    time, the above lock order must be followed.
155  *  * IN6_MULTI_LOCK covers in_multi.
156  *  * MLD_LOCK covers per-link state and any global variables in this file.
157  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
158  *    per-link state iterators.
159  *
160  *  XXX LOR PREVENTION
161  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
162  *  will not accept an ifp; it wants an embedded scope ID, unlike
163  *  ip_output(), which happily takes the ifp given to it. The embedded
164  *  scope ID is only used by MLD to select the outgoing interface.
165  *
166  *  During interface attach and detach, MLD will take MLD_LOCK *after*
167  *  the LLTABLE_LOCK.
168  *  As in6_setscope() takes LLTABLE_LOCK then SCOPE_LOCK, we can't call
169  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
170  *  dispatch could work around this, but we'd rather not do that, as it
171  *  can introduce other races.
172  *
173  *  As such, we exploit the fact that the scope ID is just the interface
174  *  index, and embed it in the IPv6 destination address accordingly.
175  *  This is potentially NOT VALID for MLDv1 reports, as they
176  *  are always sent to the multicast group itself; as MLDv2
177  *  reports are always sent to ff02::16, this is not an issue
178  *  when MLDv2 is in use.
179  *
180  *  This does not however eliminate the LOR when ip6_output() itself
181  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
182  *  trigger a LOR warning in WITNESS when the ifnet is detached.
183  *
184  *  The right answer is probably to make LLTABLE_LOCK an rwlock, given
185  *  how it's used across the network stack. Here we're simply exploiting
186  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
187  *
188  * VIMAGE:
189  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
190  *    to a vnet in ifp->if_vnet.
191  */
192 static struct mtx		 mld_mtx;
193 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
194 
195 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
196 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
197 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
198 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
199 
200 /*
201  * VIMAGE-wide globals.
202  */
203 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
204 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
205 VNET_DEFINE_STATIC(int, interface_timers_running6);
206 VNET_DEFINE_STATIC(int, state_change_timers_running6);
207 VNET_DEFINE_STATIC(int, current_state_timers_running6);
208 
209 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
210 #define	V_mli_head			VNET(mli_head)
211 #define	V_interface_timers_running6	VNET(interface_timers_running6)
212 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
213 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
214 
215 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
216 
217 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
218     "IPv6 Multicast Listener Discovery");
219 
220 /*
221  * Virtualized sysctls.
222  */
223 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
224     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
225     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
226     "Rate limit for MLDv2 Group-and-Source queries in seconds");
227 
228 /*
229  * Non-virtualized sysctls.
230  */
231 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
232     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
233     "Per-interface MLDv2 state");
234 
235 VNET_DEFINE_STATIC(bool, mld_v1enable) = true;
236 #define	V_mld_v1enable	VNET(mld_v1enable)
237 SYSCTL_BOOL(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RWTUN,
238     &VNET_NAME(mld_v1enable), 0, "Enable fallback to MLDv1");
239 
240 VNET_DEFINE_STATIC(bool, mld_v2enable) = true;
241 #define	V_mld_v2enable	VNET(mld_v2enable)
242 SYSCTL_BOOL(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RWTUN,
243     &VNET_NAME(mld_v2enable), 0, "Enable MLDv2");
244 
245 VNET_DEFINE_STATIC(bool, mld_use_allow) = true;
246 #define	V_mld_use_allow	VNET(mld_use_allow)
247 SYSCTL_BOOL(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_VNET | CTLFLAG_RWTUN,
248     &VNET_NAME(mld_use_allow), 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
249 
250 /*
251  * Packed Router Alert option structure declaration.
252  */
253 struct mld_raopt {
254 	struct ip6_hbh		hbh;
255 	struct ip6_opt		pad;
256 	struct ip6_opt_router	ra;
257 } __packed;
258 
259 /*
260  * Router Alert hop-by-hop option header.
261  */
262 static struct mld_raopt mld_ra = {
263 	.hbh = { 0, 0 },
264 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
265 	.ra = {
266 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
267 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
268 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
269 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
270 	}
271 };
272 static struct ip6_pktopts mld_po;
273 
274 static __inline void
mld_save_context(struct mbuf * m,struct ifnet * ifp)275 mld_save_context(struct mbuf *m, struct ifnet *ifp)
276 {
277 
278 #ifdef VIMAGE
279 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
280 #endif /* VIMAGE */
281 	m->m_pkthdr.rcvif = ifp;
282 	m->m_pkthdr.flowid = ifp->if_index;
283 }
284 
285 static __inline void
mld_scrub_context(struct mbuf * m)286 mld_scrub_context(struct mbuf *m)
287 {
288 
289 	m->m_pkthdr.PH_loc.ptr = NULL;
290 	m->m_pkthdr.flowid = 0;
291 }
292 
293 /*
294  * Restore context from a queued output chain.
295  * Return saved ifindex.
296  *
297  * VIMAGE: The assertion is there to make sure that we
298  * actually called CURVNET_SET() with what's in the mbuf chain.
299  */
300 static __inline uint32_t
mld_restore_context(struct mbuf * m)301 mld_restore_context(struct mbuf *m)
302 {
303 
304 #if defined(VIMAGE) && defined(INVARIANTS)
305 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
306 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
307 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
308 #endif
309 	return (m->m_pkthdr.flowid);
310 }
311 
312 /*
313  * Retrieve or set threshold between group-source queries in seconds.
314  *
315  * VIMAGE: Assume curvnet set by caller.
316  * SMPng: NOTE: Serialized by MLD lock.
317  */
318 static int
sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)319 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
320 {
321 	int error;
322 	int i;
323 
324 	error = sysctl_wire_old_buffer(req, sizeof(int));
325 	if (error)
326 		return (error);
327 
328 	MLD_LOCK();
329 
330 	i = V_mld_gsrdelay.tv_sec;
331 
332 	error = sysctl_handle_int(oidp, &i, 0, req);
333 	if (error || !req->newptr)
334 		goto out_locked;
335 
336 	if (i < -1 || i >= 60) {
337 		error = EINVAL;
338 		goto out_locked;
339 	}
340 
341 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
342 	     V_mld_gsrdelay.tv_sec, i);
343 	V_mld_gsrdelay.tv_sec = i;
344 
345 out_locked:
346 	MLD_UNLOCK();
347 	return (error);
348 }
349 
350 /*
351  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
352  * For use by ifmcstat(8).
353  *
354  * VIMAGE: Assume curvnet set by caller. The node handler itself
355  * is not directly virtualized.
356  */
357 static int
sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)358 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
359 {
360 	struct epoch_tracker	 et;
361 	int			*name;
362 	int			 error;
363 	u_int			 namelen;
364 	struct ifnet		*ifp;
365 	struct mld_ifsoftc	*mli;
366 
367 	name = (int *)arg1;
368 	namelen = arg2;
369 
370 	if (req->newptr != NULL)
371 		return (EPERM);
372 
373 	if (namelen != 1)
374 		return (EINVAL);
375 
376 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
377 	if (error)
378 		return (error);
379 
380 	IN6_MULTI_LOCK();
381 	IN6_MULTI_LIST_LOCK();
382 	MLD_LOCK();
383 	NET_EPOCH_ENTER(et);
384 
385 	error = ENOENT;
386 	ifp = ifnet_byindex(name[0]);
387 	if (ifp == NULL)
388 		goto out_locked;
389 
390 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
391 		if (ifp == mli->mli_ifp) {
392 			struct mld_ifinfo info;
393 
394 			info.mli_version = mli->mli_version;
395 			info.mli_v1_timer = mli->mli_v1_timer;
396 			info.mli_v2_timer = mli->mli_v2_timer;
397 			info.mli_flags = mli->mli_flags;
398 			info.mli_rv = mli->mli_rv;
399 			info.mli_qi = mli->mli_qi;
400 			info.mli_qri = mli->mli_qri;
401 			info.mli_uri = mli->mli_uri;
402 			error = SYSCTL_OUT(req, &info, sizeof(info));
403 			break;
404 		}
405 	}
406 
407 out_locked:
408 	NET_EPOCH_EXIT(et);
409 	MLD_UNLOCK();
410 	IN6_MULTI_LIST_UNLOCK();
411 	IN6_MULTI_UNLOCK();
412 	return (error);
413 }
414 
415 /*
416  * Dispatch an entire queue of pending packet chains.
417  * VIMAGE: Assumes the vnet pointer has been set.
418  */
419 static void
mld_dispatch_queue(struct mbufq * mq,int limit)420 mld_dispatch_queue(struct mbufq *mq, int limit)
421 {
422 	struct mbuf *m;
423 
424 	while ((m = mbufq_dequeue(mq)) != NULL) {
425 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
426 		mld_dispatch_packet(m);
427 		if (--limit == 0)
428 			break;
429 	}
430 }
431 
432 /*
433  * Filter outgoing MLD report state by group.
434  *
435  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
436  * and node-local addresses. However, kernel and socket consumers
437  * always embed the KAME scope ID in the address provided, so strip it
438  * when performing comparison.
439  * Note: This is not the same as the *multicast* scope.
440  *
441  * Return zero if the given group is one for which MLD reports
442  * should be suppressed, or non-zero if reports should be issued.
443  */
444 static __inline int
mld_is_addr_reported(const struct in6_addr * addr)445 mld_is_addr_reported(const struct in6_addr *addr)
446 {
447 
448 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
449 
450 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
451 		return (0);
452 
453 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
454 		struct in6_addr tmp = *addr;
455 		in6_clearscope(&tmp);
456 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
457 			return (0);
458 	}
459 
460 	return (1);
461 }
462 
463 /*
464  * Attach MLD when PF_INET6 is attached to an interface.  Assumes that the
465  * current VNET is set by the caller.
466  */
467 void
mld_domifattach(struct ifnet * ifp)468 mld_domifattach(struct ifnet *ifp)
469 {
470 	struct mld_ifsoftc *mli = MLD_IFINFO(ifp);
471 
472 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp));
473 
474 	*mli = (struct mld_ifsoftc){
475 		.mli_ifp = ifp,
476 		.mli_version = MLD_VERSION_2,
477 		.mli_rv = MLD_RV_INIT,
478 		.mli_qi = MLD_QI_INIT,
479 		.mli_qri = MLD_QRI_INIT,
480 		.mli_uri = MLD_URI_INIT,
481 	};
482 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
483 	if ((ifp->if_flags & IFF_MULTICAST) == 0)
484 		mli->mli_flags |= MLIF_SILENT;
485 	if (V_mld_use_allow)
486 		mli->mli_flags |= MLIF_USEALLOW;
487 
488 	MLD_LOCK();
489 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
490 	MLD_UNLOCK();
491 }
492 
493 /*
494  * Hook for ifdetach.
495  *
496  * NOTE: Some finalization tasks need to run before the protocol domain
497  * is detached, but also before the link layer does its cleanup.
498  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
499  *
500  * SMPng: Caller must hold IN6_MULTI_LOCK().
501  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
502  * XXX This routine is also bitten by unlocked ifma_protospec access.
503  */
504 void
mld_ifdetach(struct ifnet * ifp,struct in6_multi_head * inmh)505 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
506 {
507 	struct epoch_tracker     et;
508 	struct mld_ifsoftc	*mli;
509 	struct ifmultiaddr	*ifma;
510 	struct in6_multi	*inm;
511 
512 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
513 	    if_name(ifp));
514 
515 	IN6_MULTI_LIST_LOCK_ASSERT();
516 	MLD_LOCK();
517 
518 	mli = MLD_IFINFO(ifp);
519 	IF_ADDR_WLOCK(ifp);
520 	/*
521 	 * Extract list of in6_multi associated with the detaching ifp
522 	 * which the PF_INET6 layer is about to release.
523 	 */
524 	NET_EPOCH_ENTER(et);
525 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
526 		inm = in6m_ifmultiaddr_get_inm(ifma);
527 		if (inm == NULL)
528 			continue;
529 		in6m_disconnect_locked(inmh, inm);
530 
531 		if (mli->mli_version == MLD_VERSION_2) {
532 			in6m_clear_recorded(inm);
533 
534 			/*
535 			 * We need to release the final reference held
536 			 * for issuing the INCLUDE {}.
537 			 */
538 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
539 				inm->in6m_state = MLD_NOT_MEMBER;
540 				in6m_rele_locked(inmh, inm);
541 			}
542 		}
543 	}
544 	NET_EPOCH_EXIT(et);
545 	IF_ADDR_WUNLOCK(ifp);
546 	MLD_UNLOCK();
547 }
548 
549 /*
550  * Hook for domifdetach.
551  * Runs after link-layer cleanup; free MLD state.
552  */
553 void
mld_domifdetach(struct ifnet * ifp)554 mld_domifdetach(struct ifnet *ifp)
555 {
556 	struct mld_ifsoftc *mli = MLD_IFINFO(ifp);
557 
558 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
559 	    __func__, ifp, if_name(ifp));
560 
561 	MLD_LOCK();
562 	LIST_REMOVE(mli, mli_link);
563 	MLD_UNLOCK();
564 	mbufq_drain(&mli->mli_gq);
565 }
566 
567 /*
568  * Process a received MLDv1 general or address-specific query.
569  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
570  *
571  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
572  * mld_addr. This is OK as we own the mbuf chain.
573  */
574 static int
mld_v1_input_query(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mld_hdr * mld)575 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
576     /*const*/ struct mld_hdr *mld)
577 {
578 	struct ifmultiaddr	*ifma;
579 	struct mld_ifsoftc	*mli;
580 	struct in6_multi	*inm;
581 	int			 is_general_query;
582 	uint16_t		 timer;
583 #ifdef KTR
584 	char			 ip6tbuf[INET6_ADDRSTRLEN];
585 #endif
586 
587 	NET_EPOCH_ASSERT();
588 
589 	is_general_query = 0;
590 
591 	if (!V_mld_v1enable) {
592 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
593 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
594 		    ifp, if_name(ifp));
595 		return (0);
596 	}
597 
598 	/*
599 	 * RFC3810 Section 6.2: MLD queries must originate from
600 	 * a router's link-local address.
601 	 */
602 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
603 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
604 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
605 		    ifp, if_name(ifp));
606 		return (0);
607 	}
608 
609 	/*
610 	 * Do address field validation upfront before we accept
611 	 * the query.
612 	 */
613 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
614 		/*
615 		 * MLDv1 General Query.
616 		 * If this was not sent to the all-nodes group, ignore it.
617 		 */
618 		struct in6_addr		 dst;
619 
620 		dst = ip6->ip6_dst;
621 		in6_clearscope(&dst);
622 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
623 			return (EINVAL);
624 		is_general_query = 1;
625 	} else {
626 		/*
627 		 * Embed scope ID of receiving interface in MLD query for
628 		 * lookup whilst we don't hold other locks.
629 		 */
630 		in6_setscope(&mld->mld_addr, ifp, NULL);
631 	}
632 
633 	IN6_MULTI_LIST_LOCK();
634 	MLD_LOCK();
635 
636 	/*
637 	 * Switch to MLDv1 host compatibility mode.
638 	 */
639 	mli = MLD_IFINFO(ifp);
640 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
641 	mld_set_version(mli, MLD_VERSION_1);
642 
643 	timer = (ntohs(mld->mld_maxdelay) * MLD_FASTHZ) / MLD_TIMER_SCALE;
644 	if (timer == 0)
645 		timer = 1;
646 
647 	if (is_general_query) {
648 		/*
649 		 * For each reporting group joined on this
650 		 * interface, kick the report timer.
651 		 */
652 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
653 			 ifp, if_name(ifp));
654 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
655 			inm = in6m_ifmultiaddr_get_inm(ifma);
656 			if (inm == NULL)
657 				continue;
658 			mld_v1_update_group(inm, timer);
659 		}
660 	} else {
661 		/*
662 		 * MLDv1 Group-Specific Query.
663 		 * If this is a group-specific MLDv1 query, we need only
664 		 * look up the single group to process it.
665 		 */
666 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
667 		if (inm != NULL) {
668 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
669 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
670 			    ifp, if_name(ifp));
671 			mld_v1_update_group(inm, timer);
672 		}
673 		/* XXX Clear embedded scope ID as userland won't expect it. */
674 		in6_clearscope(&mld->mld_addr);
675 	}
676 
677 	MLD_UNLOCK();
678 	IN6_MULTI_LIST_UNLOCK();
679 
680 	return (0);
681 }
682 
683 /*
684  * Update the report timer on a group in response to an MLDv1 query.
685  *
686  * If we are becoming the reporting member for this group, start the timer.
687  * If we already are the reporting member for this group, and timer is
688  * below the threshold, reset it.
689  *
690  * We may be updating the group for the first time since we switched
691  * to MLDv2. If we are, then we must clear any recorded source lists,
692  * and transition to REPORTING state; the group timer is overloaded
693  * for group and group-source query responses.
694  *
695  * Unlike MLDv2, the delay per group should be jittered
696  * to avoid bursts of MLDv1 reports.
697  */
698 static void
mld_v1_update_group(struct in6_multi * inm,const int timer)699 mld_v1_update_group(struct in6_multi *inm, const int timer)
700 {
701 #ifdef KTR
702 	char			 ip6tbuf[INET6_ADDRSTRLEN];
703 #endif
704 
705 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
706 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
707 	    if_name(inm->in6m_ifp), timer);
708 
709 	IN6_MULTI_LIST_LOCK_ASSERT();
710 
711 	switch (inm->in6m_state) {
712 	case MLD_NOT_MEMBER:
713 	case MLD_SILENT_MEMBER:
714 		break;
715 	case MLD_REPORTING_MEMBER:
716 		if (inm->in6m_timer != 0 &&
717 		    inm->in6m_timer <= timer) {
718 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
719 			    "skipping.", __func__);
720 			break;
721 		}
722 		/* FALLTHROUGH */
723 	case MLD_SG_QUERY_PENDING_MEMBER:
724 	case MLD_G_QUERY_PENDING_MEMBER:
725 	case MLD_IDLE_MEMBER:
726 	case MLD_LAZY_MEMBER:
727 	case MLD_AWAKENING_MEMBER:
728 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
729 		inm->in6m_state = MLD_REPORTING_MEMBER;
730 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
731 		V_current_state_timers_running6 = 1;
732 		break;
733 	case MLD_SLEEPING_MEMBER:
734 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
735 		inm->in6m_state = MLD_AWAKENING_MEMBER;
736 		break;
737 	case MLD_LEAVING_MEMBER:
738 		break;
739 	}
740 }
741 
742 /*
743  * Process a received MLDv2 general, group-specific or
744  * group-and-source-specific query.
745  *
746  * Assumes that mld points to a struct mldv2_query which is stored in
747  * contiguous memory.
748  *
749  * Return 0 if successful, otherwise an appropriate error code is returned.
750  */
751 static int
mld_v2_input_query(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mbuf * m,struct mldv2_query * mld,const int off,const int icmp6len)752 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
753     struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
754 {
755 	struct mld_ifsoftc	*mli;
756 	struct in6_multi	*inm;
757 	uint32_t		 maxdelay, nsrc, qqi;
758 	int			 is_general_query;
759 	uint16_t		 timer;
760 	uint8_t			 qrv;
761 #ifdef KTR
762 	char			 ip6tbuf[INET6_ADDRSTRLEN];
763 #endif
764 
765 	NET_EPOCH_ASSERT();
766 
767 	if (!V_mld_v2enable) {
768 		CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
769 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
770 		    ifp, if_name(ifp));
771 		return (0);
772 	}
773 
774 	/*
775 	 * RFC3810 Section 6.2: MLD queries must originate from
776 	 * a router's link-local address.
777 	 */
778 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
779 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
780 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
781 		    ifp, if_name(ifp));
782 		return (0);
783 	}
784 
785 	is_general_query = 0;
786 
787 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
788 
789 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
790 	if (maxdelay >= 32768) {
791 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
792 			   (MLD_MRC_EXP(maxdelay) + 3);
793 	}
794 	timer = (maxdelay * MLD_FASTHZ) / MLD_TIMER_SCALE;
795 	if (timer == 0)
796 		timer = 1;
797 
798 	qrv = MLD_QRV(mld->mld_misc);
799 	if (qrv < 2) {
800 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
801 		    qrv, MLD_RV_INIT);
802 		qrv = MLD_RV_INIT;
803 	}
804 
805 	qqi = mld->mld_qqi;
806 	if (qqi >= 128) {
807 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
808 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
809 	}
810 
811 	nsrc = ntohs(mld->mld_numsrc);
812 	if (nsrc > MLD_MAX_GS_SOURCES)
813 		return (EMSGSIZE);
814 	if (icmp6len < sizeof(struct mldv2_query) +
815 	    (nsrc * sizeof(struct in6_addr)))
816 		return (EMSGSIZE);
817 
818 	/*
819 	 * Do further input validation upfront to avoid resetting timers
820 	 * should we need to discard this query.
821 	 */
822 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
823 		/*
824 		 * A general query with a source list has undefined
825 		 * behaviour; discard it.
826 		 */
827 		if (nsrc > 0)
828 			return (EINVAL);
829 		is_general_query = 1;
830 	} else {
831 		/*
832 		 * Embed scope ID of receiving interface in MLD query for
833 		 * lookup whilst we don't hold other locks (due to KAME
834 		 * locking lameness). We own this mbuf chain just now.
835 		 */
836 		in6_setscope(&mld->mld_addr, ifp, NULL);
837 	}
838 
839 	IN6_MULTI_LIST_LOCK();
840 	MLD_LOCK();
841 
842 	mli = MLD_IFINFO(ifp);
843 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
844 
845 	/*
846 	 * Discard the v2 query if we're in Compatibility Mode.
847 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
848 	 * until the Old Version Querier Present timer expires.
849 	 */
850 	if (mli->mli_version != MLD_VERSION_2)
851 		goto out_locked;
852 
853 	mld_set_version(mli, MLD_VERSION_2);
854 	mli->mli_rv = qrv;
855 	mli->mli_qi = qqi;
856 	mli->mli_qri = maxdelay;
857 
858 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
859 	    maxdelay);
860 
861 	if (is_general_query) {
862 		/*
863 		 * MLDv2 General Query.
864 		 *
865 		 * Schedule a current-state report on this ifp for
866 		 * all groups, possibly containing source lists.
867 		 *
868 		 * If there is a pending General Query response
869 		 * scheduled earlier than the selected delay, do
870 		 * not schedule any other reports.
871 		 * Otherwise, reset the interface timer.
872 		 */
873 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
874 		    ifp, if_name(ifp));
875 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
876 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
877 			V_interface_timers_running6 = 1;
878 		}
879 	} else {
880 		/*
881 		 * MLDv2 Group-specific or Group-and-source-specific Query.
882 		 *
883 		 * Group-source-specific queries are throttled on
884 		 * a per-group basis to defeat denial-of-service attempts.
885 		 * Queries for groups we are not a member of on this
886 		 * link are simply ignored.
887 		 */
888 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
889 		if (inm == NULL)
890 			goto out_locked;
891 		if (nsrc > 0) {
892 			if (!ratecheck(&inm->in6m_lastgsrtv,
893 			    &V_mld_gsrdelay)) {
894 				CTR1(KTR_MLD, "%s: GS query throttled.",
895 				    __func__);
896 				goto out_locked;
897 			}
898 		}
899 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
900 		     ifp, if_name(ifp));
901 		/*
902 		 * If there is a pending General Query response
903 		 * scheduled sooner than the selected delay, no
904 		 * further report need be scheduled.
905 		 * Otherwise, prepare to respond to the
906 		 * group-specific or group-and-source query.
907 		 */
908 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
909 			mld_v2_process_group_query(inm, mli, timer, m, mld, off);
910 
911 		/* XXX Clear embedded scope ID as userland won't expect it. */
912 		in6_clearscope(&mld->mld_addr);
913 	}
914 
915 out_locked:
916 	MLD_UNLOCK();
917 	IN6_MULTI_LIST_UNLOCK();
918 
919 	return (0);
920 }
921 
922 /*
923  * Process a received MLDv2 group-specific or group-and-source-specific
924  * query.
925  * Return <0 if any error occurred. Currently this is ignored.
926  */
927 static int
mld_v2_process_group_query(struct in6_multi * inm,struct mld_ifsoftc * mli,int timer,struct mbuf * m0,struct mldv2_query * mld,const int off)928 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
929     int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
930 {
931 	int			 retval;
932 	uint16_t		 nsrc;
933 
934 	IN6_MULTI_LIST_LOCK_ASSERT();
935 	MLD_LOCK_ASSERT();
936 
937 	retval = 0;
938 
939 	switch (inm->in6m_state) {
940 	case MLD_NOT_MEMBER:
941 	case MLD_SILENT_MEMBER:
942 	case MLD_SLEEPING_MEMBER:
943 	case MLD_LAZY_MEMBER:
944 	case MLD_AWAKENING_MEMBER:
945 	case MLD_IDLE_MEMBER:
946 	case MLD_LEAVING_MEMBER:
947 		return (retval);
948 		break;
949 	case MLD_REPORTING_MEMBER:
950 	case MLD_G_QUERY_PENDING_MEMBER:
951 	case MLD_SG_QUERY_PENDING_MEMBER:
952 		break;
953 	}
954 
955 	nsrc = ntohs(mld->mld_numsrc);
956 
957 	/* Length should be checked by calling function. */
958 	KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
959 	    m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
960 	    nsrc * sizeof(struct in6_addr),
961 	    ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
962 	    m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
963 	    nsrc * sizeof(struct in6_addr), m0));
964 
965 	/*
966 	 * Deal with group-specific queries upfront.
967 	 * If any group query is already pending, purge any recorded
968 	 * source-list state if it exists, and schedule a query response
969 	 * for this group-specific query.
970 	 */
971 	if (nsrc == 0) {
972 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
973 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
974 			in6m_clear_recorded(inm);
975 			timer = min(inm->in6m_timer, timer);
976 		}
977 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
978 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
979 		V_current_state_timers_running6 = 1;
980 		return (retval);
981 	}
982 
983 	/*
984 	 * Deal with the case where a group-and-source-specific query has
985 	 * been received but a group-specific query is already pending.
986 	 */
987 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
988 		timer = min(inm->in6m_timer, timer);
989 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
990 		V_current_state_timers_running6 = 1;
991 		return (retval);
992 	}
993 
994 	/*
995 	 * Finally, deal with the case where a group-and-source-specific
996 	 * query has been received, where a response to a previous g-s-r
997 	 * query exists, or none exists.
998 	 * In this case, we need to parse the source-list which the Querier
999 	 * has provided us with and check if we have any source list filter
1000 	 * entries at T1 for these sources. If we do not, there is no need
1001 	 * schedule a report and the query may be dropped.
1002 	 * If we do, we must record them and schedule a current-state
1003 	 * report for those sources.
1004 	 */
1005 	if (inm->in6m_nsrc > 0) {
1006 		struct in6_addr		 srcaddr;
1007 		int			 i, nrecorded;
1008 		int			 soff;
1009 
1010 		soff = off + sizeof(struct mldv2_query);
1011 		nrecorded = 0;
1012 		for (i = 0; i < nsrc; i++) {
1013 			m_copydata(m0, soff, sizeof(struct in6_addr),
1014 			    (caddr_t)&srcaddr);
1015 			retval = in6m_record_source(inm, &srcaddr);
1016 			if (retval < 0)
1017 				break;
1018 			nrecorded += retval;
1019 			soff += sizeof(struct in6_addr);
1020 		}
1021 		if (nrecorded > 0) {
1022 			CTR1(KTR_MLD,
1023 			    "%s: schedule response to SG query", __func__);
1024 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1025 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1026 			V_current_state_timers_running6 = 1;
1027 		}
1028 	}
1029 
1030 	return (retval);
1031 }
1032 
1033 /*
1034  * Process a received MLDv1 host membership report.
1035  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1036  *
1037  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1038  * mld_addr. This is OK as we own the mbuf chain.
1039  */
1040 static int
mld_v1_input_report(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mld_hdr * mld)1041 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1042     /*const*/ struct mld_hdr *mld)
1043 {
1044 	struct in6_addr		 src, dst;
1045 	struct in6_ifaddr	*ia;
1046 	struct in6_multi	*inm;
1047 #ifdef KTR
1048 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1049 #endif
1050 
1051 	NET_EPOCH_ASSERT();
1052 
1053 	if (!V_mld_v1enable) {
1054 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1055 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1056 		    ifp, if_name(ifp));
1057 		return (0);
1058 	}
1059 
1060 	if (ifp->if_flags & IFF_LOOPBACK)
1061 		return (0);
1062 
1063 	/*
1064 	 * MLDv1 reports must originate from a host's link-local address,
1065 	 * or the unspecified address (when booting).
1066 	 */
1067 	src = ip6->ip6_src;
1068 	in6_clearscope(&src);
1069 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1070 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1071 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1072 		    ifp, if_name(ifp));
1073 		return (EINVAL);
1074 	}
1075 
1076 	/*
1077 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1078 	 * group, and must be directed to the group itself.
1079 	 */
1080 	dst = ip6->ip6_dst;
1081 	in6_clearscope(&dst);
1082 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1083 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1084 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1085 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1086 		    ifp, if_name(ifp));
1087 		return (EINVAL);
1088 	}
1089 
1090 	/*
1091 	 * Make sure we don't hear our own membership report, as fast
1092 	 * leave requires knowing that we are the only member of a
1093 	 * group. Assume we used the link-local address if available,
1094 	 * otherwise look for ::.
1095 	 *
1096 	 * XXX Note that scope ID comparison is needed for the address
1097 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1098 	 * performed for the on-wire address.
1099 	 */
1100 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1101 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1102 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1103 		if (ia != NULL)
1104 			ifa_free(&ia->ia_ifa);
1105 		return (0);
1106 	}
1107 	if (ia != NULL)
1108 		ifa_free(&ia->ia_ifa);
1109 
1110 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1111 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1112 
1113 	/*
1114 	 * Embed scope ID of receiving interface in MLD query for lookup
1115 	 * whilst we don't hold other locks (due to KAME locking lameness).
1116 	 */
1117 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1118 		in6_setscope(&mld->mld_addr, ifp, NULL);
1119 
1120 	IN6_MULTI_LIST_LOCK();
1121 	MLD_LOCK();
1122 
1123 	/*
1124 	 * MLDv1 report suppression.
1125 	 * If we are a member of this group, and our membership should be
1126 	 * reported, and our group timer is pending or about to be reset,
1127 	 * stop our group timer by transitioning to the 'lazy' state.
1128 	 */
1129 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1130 	if (inm != NULL) {
1131 		struct mld_ifsoftc *mli;
1132 
1133 		mli = inm->in6m_mli;
1134 		KASSERT(mli != NULL,
1135 		    ("%s: no mli for ifp %p", __func__, ifp));
1136 
1137 		/*
1138 		 * If we are in MLDv2 host mode, do not allow the
1139 		 * other host's MLDv1 report to suppress our reports.
1140 		 */
1141 		if (mli->mli_version == MLD_VERSION_2)
1142 			goto out_locked;
1143 
1144 		inm->in6m_timer = 0;
1145 
1146 		switch (inm->in6m_state) {
1147 		case MLD_NOT_MEMBER:
1148 		case MLD_SILENT_MEMBER:
1149 		case MLD_SLEEPING_MEMBER:
1150 			break;
1151 		case MLD_REPORTING_MEMBER:
1152 		case MLD_IDLE_MEMBER:
1153 		case MLD_AWAKENING_MEMBER:
1154 			CTR3(KTR_MLD,
1155 			    "report suppressed for %s on ifp %p(%s)",
1156 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1157 			    ifp, if_name(ifp));
1158 		case MLD_LAZY_MEMBER:
1159 			inm->in6m_state = MLD_LAZY_MEMBER;
1160 			break;
1161 		case MLD_G_QUERY_PENDING_MEMBER:
1162 		case MLD_SG_QUERY_PENDING_MEMBER:
1163 		case MLD_LEAVING_MEMBER:
1164 			break;
1165 		}
1166 	}
1167 
1168 out_locked:
1169 	MLD_UNLOCK();
1170 	IN6_MULTI_LIST_UNLOCK();
1171 
1172 	/* XXX Clear embedded scope ID as userland won't expect it. */
1173 	in6_clearscope(&mld->mld_addr);
1174 
1175 	return (0);
1176 }
1177 
1178 /*
1179  * MLD input path.
1180  *
1181  * Assume query messages which fit in a single ICMPv6 message header
1182  * have been pulled up.
1183  * Assume that userland will want to see the message, even if it
1184  * otherwise fails kernel input validation; do not free it.
1185  * Pullup may however free the mbuf chain m if it fails.
1186  *
1187  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1188  */
1189 int
mld_input(struct mbuf ** mp,int off,int icmp6len)1190 mld_input(struct mbuf **mp, int off, int icmp6len)
1191 {
1192 	struct ifnet	*ifp;
1193 	struct ip6_hdr	*ip6;
1194 	struct mbuf	*m;
1195 	struct mld_hdr	*mld;
1196 	int		 mldlen;
1197 
1198 	m = *mp;
1199 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1200 
1201 	ifp = m->m_pkthdr.rcvif;
1202 
1203 	/* Pullup to appropriate size. */
1204 	if (m->m_len < off + sizeof(*mld)) {
1205 		m = m_pullup(m, off + sizeof(*mld));
1206 		if (m == NULL) {
1207 			ICMP6STAT_INC(icp6s_badlen);
1208 			return (IPPROTO_DONE);
1209 		}
1210 	}
1211 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1212 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1213 	    icmp6len >= sizeof(struct mldv2_query)) {
1214 		mldlen = sizeof(struct mldv2_query);
1215 	} else {
1216 		mldlen = sizeof(struct mld_hdr);
1217 	}
1218 	if (m->m_len < off + mldlen) {
1219 		m = m_pullup(m, off + mldlen);
1220 		if (m == NULL) {
1221 			ICMP6STAT_INC(icp6s_badlen);
1222 			return (IPPROTO_DONE);
1223 		}
1224 	}
1225 	*mp = m;
1226 	ip6 = mtod(m, struct ip6_hdr *);
1227 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1228 
1229 	/*
1230 	 * Userland needs to see all of this traffic for implementing
1231 	 * the endpoint discovery portion of multicast routing.
1232 	 */
1233 	switch (mld->mld_type) {
1234 	case MLD_LISTENER_QUERY:
1235 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1236 		if (icmp6len == sizeof(struct mld_hdr)) {
1237 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1238 				return (0);
1239 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1240 			if (mld_v2_input_query(ifp, ip6, m,
1241 			    (struct mldv2_query *)mld, off, icmp6len) != 0)
1242 				return (0);
1243 		}
1244 		break;
1245 	case MLD_LISTENER_REPORT:
1246 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1247 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1248 			return (0);
1249 		break;
1250 	case MLDV2_LISTENER_REPORT:
1251 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1252 		break;
1253 	case MLD_LISTENER_DONE:
1254 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1255 		break;
1256 	default:
1257 		break;
1258 	}
1259 
1260 	return (0);
1261 }
1262 
1263 /*
1264  * Fast timeout handler (global).
1265  * VIMAGE: Timeout handlers are expected to service all vimages.
1266  */
1267 static struct callout mldfast_callout;
1268 static void
mld_fasttimo(void * arg __unused)1269 mld_fasttimo(void *arg __unused)
1270 {
1271 	struct epoch_tracker et;
1272 	struct in6_multi_head inmh;
1273 	VNET_ITERATOR_DECL(vnet_iter);
1274 
1275 	SLIST_INIT(&inmh);
1276 
1277 	NET_EPOCH_ENTER(et);
1278 	VNET_LIST_RLOCK_NOSLEEP();
1279 	VNET_FOREACH(vnet_iter) {
1280 		CURVNET_SET(vnet_iter);
1281 		mld_fasttimo_vnet(&inmh);
1282 		CURVNET_RESTORE();
1283 	}
1284 	VNET_LIST_RUNLOCK_NOSLEEP();
1285 	NET_EPOCH_EXIT(et);
1286 	in6m_release_list_deferred(&inmh);
1287 
1288 	callout_reset(&mldfast_callout, hz / MLD_FASTHZ, mld_fasttimo, NULL);
1289 }
1290 
1291 /*
1292  * Fast timeout handler (per-vnet).
1293  *
1294  * VIMAGE: Assume caller has set up our curvnet.
1295  */
1296 static void
mld_fasttimo_vnet(struct in6_multi_head * inmh)1297 mld_fasttimo_vnet(struct in6_multi_head *inmh)
1298 {
1299 	struct mbufq		 scq;	/* State-change packets */
1300 	struct mbufq		 qrq;	/* Query response packets */
1301 	struct ifnet		*ifp;
1302 	struct mld_ifsoftc	*mli;
1303 	struct ifmultiaddr	*ifma;
1304 	struct in6_multi	*inm;
1305 	int			 uri_fasthz;
1306 
1307 	uri_fasthz = 0;
1308 
1309 	/*
1310 	 * Quick check to see if any work needs to be done, in order to
1311 	 * minimize the overhead of fasttimo processing.
1312 	 * SMPng: XXX Unlocked reads.
1313 	 */
1314 	if (!V_current_state_timers_running6 &&
1315 	    !V_interface_timers_running6 &&
1316 	    !V_state_change_timers_running6)
1317 		return;
1318 
1319 	IN6_MULTI_LIST_LOCK();
1320 	MLD_LOCK();
1321 
1322 	/*
1323 	 * MLDv2 General Query response timer processing.
1324 	 */
1325 	if (V_interface_timers_running6) {
1326 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1327 
1328 		V_interface_timers_running6 = 0;
1329 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1330 			if (mli->mli_v2_timer == 0) {
1331 				/* Do nothing. */
1332 			} else if (--mli->mli_v2_timer == 0) {
1333 				mld_v2_dispatch_general_query(mli);
1334 			} else {
1335 				V_interface_timers_running6 = 1;
1336 			}
1337 		}
1338 	}
1339 
1340 	if (!V_current_state_timers_running6 &&
1341 	    !V_state_change_timers_running6)
1342 		goto out_locked;
1343 
1344 	V_current_state_timers_running6 = 0;
1345 	V_state_change_timers_running6 = 0;
1346 
1347 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1348 
1349 	/*
1350 	 * MLD host report and state-change timer processing.
1351 	 * Note: Processing a v2 group timer may remove a node.
1352 	 */
1353 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1354 		ifp = mli->mli_ifp;
1355 
1356 		if (mli->mli_version == MLD_VERSION_2) {
1357 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1358 			    MLD_FASTHZ);
1359 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1360 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1361 		}
1362 
1363 		IF_ADDR_WLOCK(ifp);
1364 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1365 			inm = in6m_ifmultiaddr_get_inm(ifma);
1366 			if (inm == NULL)
1367 				continue;
1368 			switch (mli->mli_version) {
1369 			case MLD_VERSION_1:
1370 				mld_v1_process_group_timer(inmh, inm);
1371 				break;
1372 			case MLD_VERSION_2:
1373 				mld_v2_process_group_timers(inmh, &qrq,
1374 				    &scq, inm, uri_fasthz);
1375 				break;
1376 			}
1377 		}
1378 		IF_ADDR_WUNLOCK(ifp);
1379 
1380 		switch (mli->mli_version) {
1381 		case MLD_VERSION_1:
1382 			/*
1383 			 * Transmit reports for this lifecycle.  This
1384 			 * is done while not holding IF_ADDR_LOCK
1385 			 * since this can call
1386 			 * in6ifa_ifpforlinklocal() which locks
1387 			 * IF_ADDR_LOCK internally as well as
1388 			 * ip6_output() to transmit a packet.
1389 			 */
1390 			while ((inm = SLIST_FIRST(inmh)) != NULL) {
1391 				SLIST_REMOVE_HEAD(inmh, in6m_defer);
1392 				(void)mld_v1_transmit_report(inm,
1393 				    MLD_LISTENER_REPORT);
1394 			}
1395 			break;
1396 		case MLD_VERSION_2:
1397 			mld_dispatch_queue(&qrq, 0);
1398 			mld_dispatch_queue(&scq, 0);
1399 			break;
1400 		}
1401 	}
1402 
1403 out_locked:
1404 	MLD_UNLOCK();
1405 	IN6_MULTI_LIST_UNLOCK();
1406 }
1407 
1408 /*
1409  * Update host report group timer.
1410  * Will update the global pending timer flags.
1411  */
1412 static void
mld_v1_process_group_timer(struct in6_multi_head * inmh,struct in6_multi * inm)1413 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1414 {
1415 	int report_timer_expired;
1416 
1417 	IN6_MULTI_LIST_LOCK_ASSERT();
1418 	MLD_LOCK_ASSERT();
1419 
1420 	if (inm->in6m_timer == 0) {
1421 		report_timer_expired = 0;
1422 	} else if (--inm->in6m_timer == 0) {
1423 		report_timer_expired = 1;
1424 	} else {
1425 		V_current_state_timers_running6 = 1;
1426 		return;
1427 	}
1428 
1429 	switch (inm->in6m_state) {
1430 	case MLD_NOT_MEMBER:
1431 	case MLD_SILENT_MEMBER:
1432 	case MLD_IDLE_MEMBER:
1433 	case MLD_LAZY_MEMBER:
1434 	case MLD_SLEEPING_MEMBER:
1435 	case MLD_AWAKENING_MEMBER:
1436 		break;
1437 	case MLD_REPORTING_MEMBER:
1438 		if (report_timer_expired) {
1439 			inm->in6m_state = MLD_IDLE_MEMBER;
1440 			SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
1441 		}
1442 		break;
1443 	case MLD_G_QUERY_PENDING_MEMBER:
1444 	case MLD_SG_QUERY_PENDING_MEMBER:
1445 	case MLD_LEAVING_MEMBER:
1446 		break;
1447 	}
1448 }
1449 
1450 /*
1451  * Update a group's timers for MLDv2.
1452  * Will update the global pending timer flags.
1453  * Note: Unlocked read from mli.
1454  */
1455 static void
mld_v2_process_group_timers(struct in6_multi_head * inmh,struct mbufq * qrq,struct mbufq * scq,struct in6_multi * inm,const int uri_fasthz)1456 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1457     struct mbufq *qrq, struct mbufq *scq,
1458     struct in6_multi *inm, const int uri_fasthz)
1459 {
1460 	int query_response_timer_expired;
1461 	int state_change_retransmit_timer_expired;
1462 #ifdef KTR
1463 	char ip6tbuf[INET6_ADDRSTRLEN];
1464 #endif
1465 
1466 	IN6_MULTI_LIST_LOCK_ASSERT();
1467 	MLD_LOCK_ASSERT();
1468 
1469 	query_response_timer_expired = 0;
1470 	state_change_retransmit_timer_expired = 0;
1471 
1472 	/*
1473 	 * During a transition from compatibility mode back to MLDv2,
1474 	 * a group record in REPORTING state may still have its group
1475 	 * timer active. This is a no-op in this function; it is easier
1476 	 * to deal with it here than to complicate the slow-timeout path.
1477 	 */
1478 	if (inm->in6m_timer == 0) {
1479 		query_response_timer_expired = 0;
1480 	} else if (--inm->in6m_timer == 0) {
1481 		query_response_timer_expired = 1;
1482 	} else {
1483 		V_current_state_timers_running6 = 1;
1484 	}
1485 
1486 	if (inm->in6m_sctimer == 0) {
1487 		state_change_retransmit_timer_expired = 0;
1488 	} else if (--inm->in6m_sctimer == 0) {
1489 		state_change_retransmit_timer_expired = 1;
1490 	} else {
1491 		V_state_change_timers_running6 = 1;
1492 	}
1493 
1494 	/* We are in fasttimo, so be quick about it. */
1495 	if (!state_change_retransmit_timer_expired &&
1496 	    !query_response_timer_expired)
1497 		return;
1498 
1499 	switch (inm->in6m_state) {
1500 	case MLD_NOT_MEMBER:
1501 	case MLD_SILENT_MEMBER:
1502 	case MLD_SLEEPING_MEMBER:
1503 	case MLD_LAZY_MEMBER:
1504 	case MLD_AWAKENING_MEMBER:
1505 	case MLD_IDLE_MEMBER:
1506 		break;
1507 	case MLD_G_QUERY_PENDING_MEMBER:
1508 	case MLD_SG_QUERY_PENDING_MEMBER:
1509 		/*
1510 		 * Respond to a previously pending Group-Specific
1511 		 * or Group-and-Source-Specific query by enqueueing
1512 		 * the appropriate Current-State report for
1513 		 * immediate transmission.
1514 		 */
1515 		if (query_response_timer_expired) {
1516 			int retval __unused;
1517 
1518 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1519 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1520 			    0);
1521 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1522 			    __func__, retval);
1523 			inm->in6m_state = MLD_REPORTING_MEMBER;
1524 			in6m_clear_recorded(inm);
1525 		}
1526 		/* FALLTHROUGH */
1527 	case MLD_REPORTING_MEMBER:
1528 	case MLD_LEAVING_MEMBER:
1529 		if (state_change_retransmit_timer_expired) {
1530 			/*
1531 			 * State-change retransmission timer fired.
1532 			 * If there are any further pending retransmissions,
1533 			 * set the global pending state-change flag, and
1534 			 * reset the timer.
1535 			 */
1536 			if (--inm->in6m_scrv > 0) {
1537 				inm->in6m_sctimer = uri_fasthz;
1538 				V_state_change_timers_running6 = 1;
1539 			}
1540 			/*
1541 			 * Retransmit the previously computed state-change
1542 			 * report. If there are no further pending
1543 			 * retransmissions, the mbuf queue will be consumed.
1544 			 * Update T0 state to T1 as we have now sent
1545 			 * a state-change.
1546 			 */
1547 			(void)mld_v2_merge_state_changes(inm, scq);
1548 
1549 			in6m_commit(inm);
1550 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1551 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1552 			    if_name(inm->in6m_ifp));
1553 
1554 			/*
1555 			 * If we are leaving the group for good, make sure
1556 			 * we release MLD's reference to it.
1557 			 * This release must be deferred using a SLIST,
1558 			 * as we are called from a loop which traverses
1559 			 * the in_ifmultiaddr TAILQ.
1560 			 */
1561 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1562 			    inm->in6m_scrv == 0) {
1563 				inm->in6m_state = MLD_NOT_MEMBER;
1564 				in6m_disconnect_locked(inmh, inm);
1565 				in6m_rele_locked(inmh, inm);
1566 			}
1567 		}
1568 		break;
1569 	}
1570 }
1571 
1572 /*
1573  * Switch to a different version on the given interface,
1574  * as per Section 9.12.
1575  */
1576 static void
mld_set_version(struct mld_ifsoftc * mli,const int version)1577 mld_set_version(struct mld_ifsoftc *mli, const int version)
1578 {
1579 	int old_version_timer;
1580 
1581 	MLD_LOCK_ASSERT();
1582 
1583 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1584 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1585 
1586 	if (version == MLD_VERSION_1) {
1587 		/*
1588 		 * Compute the "Older Version Querier Present" timer as per
1589 		 * Section 9.12.
1590 		 */
1591 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1592 		old_version_timer *= MLD_SLOWHZ;
1593 		mli->mli_v1_timer = old_version_timer;
1594 	}
1595 
1596 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1597 		mli->mli_version = MLD_VERSION_1;
1598 		mld_v2_cancel_link_timers(mli);
1599 	}
1600 }
1601 
1602 /*
1603  * Cancel pending MLDv2 timers for the given link and all groups
1604  * joined on it; state-change, general-query, and group-query timers.
1605  */
1606 static void
mld_v2_cancel_link_timers(struct mld_ifsoftc * mli)1607 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1608 {
1609 	struct epoch_tracker	 et;
1610 	struct in6_multi_head	 inmh;
1611 	struct ifmultiaddr	*ifma;
1612 	struct ifnet		*ifp;
1613 	struct in6_multi	*inm;
1614 
1615 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1616 	    mli->mli_ifp, if_name(mli->mli_ifp));
1617 
1618 	SLIST_INIT(&inmh);
1619 	IN6_MULTI_LIST_LOCK_ASSERT();
1620 	MLD_LOCK_ASSERT();
1621 
1622 	/*
1623 	 * Fast-track this potentially expensive operation
1624 	 * by checking all the global 'timer pending' flags.
1625 	 */
1626 	if (!V_interface_timers_running6 &&
1627 	    !V_state_change_timers_running6 &&
1628 	    !V_current_state_timers_running6)
1629 		return;
1630 
1631 	mli->mli_v2_timer = 0;
1632 
1633 	ifp = mli->mli_ifp;
1634 
1635 	IF_ADDR_WLOCK(ifp);
1636 	NET_EPOCH_ENTER(et);
1637 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1638 		inm = in6m_ifmultiaddr_get_inm(ifma);
1639 		if (inm == NULL)
1640 			continue;
1641 		switch (inm->in6m_state) {
1642 		case MLD_NOT_MEMBER:
1643 		case MLD_SILENT_MEMBER:
1644 		case MLD_IDLE_MEMBER:
1645 		case MLD_LAZY_MEMBER:
1646 		case MLD_SLEEPING_MEMBER:
1647 		case MLD_AWAKENING_MEMBER:
1648 			break;
1649 		case MLD_LEAVING_MEMBER:
1650 			/*
1651 			 * If we are leaving the group and switching
1652 			 * version, we need to release the final
1653 			 * reference held for issuing the INCLUDE {}.
1654 			 */
1655 			if (inm->in6m_refcount == 1)
1656 				in6m_disconnect_locked(&inmh, inm);
1657 			in6m_rele_locked(&inmh, inm);
1658 			/* FALLTHROUGH */
1659 		case MLD_G_QUERY_PENDING_MEMBER:
1660 		case MLD_SG_QUERY_PENDING_MEMBER:
1661 			in6m_clear_recorded(inm);
1662 			/* FALLTHROUGH */
1663 		case MLD_REPORTING_MEMBER:
1664 			inm->in6m_sctimer = 0;
1665 			inm->in6m_timer = 0;
1666 			inm->in6m_state = MLD_REPORTING_MEMBER;
1667 			/*
1668 			 * Free any pending MLDv2 state-change records.
1669 			 */
1670 			mbufq_drain(&inm->in6m_scq);
1671 			break;
1672 		}
1673 	}
1674 	NET_EPOCH_EXIT(et);
1675 	IF_ADDR_WUNLOCK(ifp);
1676 	in6m_release_list_deferred(&inmh);
1677 }
1678 
1679 /*
1680  * Global slowtimo handler.
1681  * VIMAGE: Timeout handlers are expected to service all vimages.
1682  */
1683 static struct callout mldslow_callout;
1684 static void
mld_slowtimo(void * arg __unused)1685 mld_slowtimo(void *arg __unused)
1686 {
1687 	VNET_ITERATOR_DECL(vnet_iter);
1688 
1689 	VNET_LIST_RLOCK_NOSLEEP();
1690 	VNET_FOREACH(vnet_iter) {
1691 		CURVNET_SET(vnet_iter);
1692 		mld_slowtimo_vnet();
1693 		CURVNET_RESTORE();
1694 	}
1695 	VNET_LIST_RUNLOCK_NOSLEEP();
1696 
1697 	callout_reset(&mldslow_callout, hz / MLD_SLOWHZ, mld_slowtimo, NULL);
1698 }
1699 
1700 /*
1701  * Per-vnet slowtimo handler.
1702  */
1703 static void
mld_slowtimo_vnet(void)1704 mld_slowtimo_vnet(void)
1705 {
1706 	struct mld_ifsoftc *mli;
1707 
1708 	MLD_LOCK();
1709 
1710 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1711 		mld_v1_process_querier_timers(mli);
1712 	}
1713 
1714 	MLD_UNLOCK();
1715 }
1716 
1717 /*
1718  * Update the Older Version Querier Present timers for a link.
1719  * See Section 9.12 of RFC 3810.
1720  */
1721 static void
mld_v1_process_querier_timers(struct mld_ifsoftc * mli)1722 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1723 {
1724 
1725 	MLD_LOCK_ASSERT();
1726 
1727 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1728 		/*
1729 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1730 		 */
1731 		CTR5(KTR_MLD,
1732 		    "%s: transition from v%d -> v%d on %p(%s)",
1733 		    __func__, mli->mli_version, MLD_VERSION_2,
1734 		    mli->mli_ifp, if_name(mli->mli_ifp));
1735 		mli->mli_version = MLD_VERSION_2;
1736 	}
1737 }
1738 
1739 /*
1740  * Transmit an MLDv1 report immediately.
1741  */
1742 static int
mld_v1_transmit_report(struct in6_multi * in6m,const int type)1743 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1744 {
1745 	struct ifnet		*ifp;
1746 	struct in6_ifaddr	*ia;
1747 	struct ip6_hdr		*ip6;
1748 	struct mbuf		*mh, *md;
1749 	struct mld_hdr		*mld;
1750 
1751 	NET_EPOCH_ASSERT();
1752 	IN6_MULTI_LIST_LOCK_ASSERT();
1753 	MLD_LOCK_ASSERT();
1754 
1755 	ifp = in6m->in6m_ifp;
1756 	/* in process of being freed */
1757 	if (ifp == NULL)
1758 		return (0);
1759 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1760 	/* ia may be NULL if link-local address is tentative. */
1761 
1762 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1763 	if (mh == NULL) {
1764 		if (ia != NULL)
1765 			ifa_free(&ia->ia_ifa);
1766 		return (ENOMEM);
1767 	}
1768 	md = m_get(M_NOWAIT, MT_DATA);
1769 	if (md == NULL) {
1770 		m_free(mh);
1771 		if (ia != NULL)
1772 			ifa_free(&ia->ia_ifa);
1773 		return (ENOMEM);
1774 	}
1775 	mh->m_next = md;
1776 
1777 	/*
1778 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1779 	 * that ether_output() does not need to allocate another mbuf
1780 	 * for the header in the most common case.
1781 	 */
1782 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1783 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1784 	mh->m_len = sizeof(struct ip6_hdr);
1785 
1786 	ip6 = mtod(mh, struct ip6_hdr *);
1787 	ip6->ip6_flow = 0;
1788 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1789 	ip6->ip6_vfc |= IPV6_VERSION;
1790 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1791 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1792 	ip6->ip6_dst = in6m->in6m_addr;
1793 
1794 	md->m_len = sizeof(struct mld_hdr);
1795 	mld = mtod(md, struct mld_hdr *);
1796 	mld->mld_type = type;
1797 	mld->mld_code = 0;
1798 	mld->mld_cksum = 0;
1799 	mld->mld_maxdelay = 0;
1800 	mld->mld_reserved = 0;
1801 	mld->mld_addr = in6m->in6m_addr;
1802 	in6_clearscope(&mld->mld_addr);
1803 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1804 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1805 
1806 	mld_save_context(mh, ifp);
1807 	mh->m_flags |= M_MLDV1;
1808 
1809 	mld_dispatch_packet(mh);
1810 
1811 	if (ia != NULL)
1812 		ifa_free(&ia->ia_ifa);
1813 	return (0);
1814 }
1815 
1816 /*
1817  * Process a state change from the upper layer for the given IPv6 group.
1818  *
1819  * Each socket holds a reference on the in_multi in its own ip_moptions.
1820  * The socket layer will have made the necessary updates to.the group
1821  * state, it is now up to MLD to issue a state change report if there
1822  * has been any change between T0 (when the last state-change was issued)
1823  * and T1 (now).
1824  *
1825  * We use the MLDv2 state machine at group level. The MLd module
1826  * however makes the decision as to which MLD protocol version to speak.
1827  * A state change *from* INCLUDE {} always means an initial join.
1828  * A state change *to* INCLUDE {} always means a final leave.
1829  *
1830  * If delay is non-zero, and the state change is an initial multicast
1831  * join, the state change report will be delayed by 'delay' ticks
1832  * in units of MLD_FASTHZ if MLDv1 is active on the link; otherwise
1833  * the initial MLDv2 state change report will be delayed by whichever
1834  * is sooner, a pending state-change timer or delay itself.
1835  *
1836  * VIMAGE: curvnet should have been set by caller, as this routine
1837  * is called from the socket option handlers.
1838  */
1839 int
mld_change_state(struct in6_multi * inm,const int delay)1840 mld_change_state(struct in6_multi *inm, const int delay)
1841 {
1842 	struct mld_ifsoftc *mli;
1843 	struct ifnet *ifp;
1844 	int error;
1845 
1846 	IN6_MULTI_LIST_LOCK_ASSERT();
1847 
1848 	error = 0;
1849 
1850 	/*
1851 	 * Check if the in6_multi has already been disconnected.
1852 	 */
1853 	if (inm->in6m_ifp == NULL) {
1854 		CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
1855 		return (0);
1856 	}
1857 
1858 	/*
1859 	 * Try to detect if the upper layer just asked us to change state
1860 	 * for an interface which has now gone away.
1861 	 */
1862 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1863 	ifp = inm->in6m_ifma->ifma_ifp;
1864 	if (ifp == NULL)
1865 		return (0);
1866 	/*
1867 	 * Sanity check that netinet6's notion of ifp is the
1868 	 * same as net's.
1869 	 */
1870 	KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1871 
1872 	MLD_LOCK();
1873 	mli = MLD_IFINFO(ifp);
1874 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1875 
1876 	/*
1877 	 * If we detect a state transition to or from MCAST_UNDEFINED
1878 	 * for this group, then we are starting or finishing an MLD
1879 	 * life cycle for this group.
1880 	 */
1881 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1882 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1883 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1884 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1885 			CTR1(KTR_MLD, "%s: initial join", __func__);
1886 			error = mld_initial_join(inm, mli, delay);
1887 			goto out_locked;
1888 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1889 			CTR1(KTR_MLD, "%s: final leave", __func__);
1890 			mld_final_leave(inm, mli);
1891 			goto out_locked;
1892 		}
1893 	} else {
1894 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1895 	}
1896 
1897 	error = mld_handle_state_change(inm, mli);
1898 
1899 out_locked:
1900 	MLD_UNLOCK();
1901 	return (error);
1902 }
1903 
1904 /*
1905  * Perform the initial join for an MLD group.
1906  *
1907  * When joining a group:
1908  *  If the group should have its MLD traffic suppressed, do nothing.
1909  *  MLDv1 starts sending MLDv1 host membership reports.
1910  *  MLDv2 will schedule an MLDv2 state-change report containing the
1911  *  initial state of the membership.
1912  *
1913  * If the delay argument is non-zero, then we must delay sending the
1914  * initial state change for delay ticks (in units of MLD_FASTHZ).
1915  */
1916 static int
mld_initial_join(struct in6_multi * inm,struct mld_ifsoftc * mli,const int delay)1917 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1918     const int delay)
1919 {
1920 	struct epoch_tracker     et;
1921 	struct ifnet		*ifp;
1922 	struct mbufq		*mq;
1923 	int			 error, retval, syncstates;
1924 	int			 odelay;
1925 #ifdef KTR
1926 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1927 #endif
1928 
1929 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1930 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1931 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1932 
1933 	error = 0;
1934 	syncstates = 1;
1935 
1936 	ifp = inm->in6m_ifp;
1937 
1938 	IN6_MULTI_LIST_LOCK_ASSERT();
1939 	MLD_LOCK_ASSERT();
1940 
1941 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1942 
1943 	/*
1944 	 * Groups joined on loopback or marked as 'not reported',
1945 	 * enter the MLD_SILENT_MEMBER state and
1946 	 * are never reported in any protocol exchanges.
1947 	 * All other groups enter the appropriate state machine
1948 	 * for the version in use on this link.
1949 	 * A link marked as MLIF_SILENT causes MLD to be completely
1950 	 * disabled for the link.
1951 	 */
1952 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1953 	    (mli->mli_flags & MLIF_SILENT) ||
1954 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1955 		CTR1(KTR_MLD,
1956 "%s: not kicking state machine for silent group", __func__);
1957 		inm->in6m_state = MLD_SILENT_MEMBER;
1958 		inm->in6m_timer = 0;
1959 	} else {
1960 		/*
1961 		 * Deal with overlapping in_multi lifecycle.
1962 		 * If this group was LEAVING, then make sure
1963 		 * we drop the reference we picked up to keep the
1964 		 * group around for the final INCLUDE {} enqueue.
1965 		 */
1966 		if (mli->mli_version == MLD_VERSION_2 &&
1967 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
1968 			inm->in6m_refcount--;
1969 			MPASS(inm->in6m_refcount > 0);
1970 		}
1971 		inm->in6m_state = MLD_REPORTING_MEMBER;
1972 
1973 		switch (mli->mli_version) {
1974 		case MLD_VERSION_1:
1975 			/*
1976 			 * If a delay was provided, only use it if
1977 			 * it is greater than the delay normally
1978 			 * used for an MLDv1 state change report,
1979 			 * and delay sending the initial MLDv1 report
1980 			 * by not transitioning to the IDLE state.
1981 			 */
1982 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * MLD_FASTHZ);
1983 			if (delay) {
1984 				inm->in6m_timer = max(delay, odelay);
1985 				V_current_state_timers_running6 = 1;
1986 			} else {
1987 				inm->in6m_state = MLD_IDLE_MEMBER;
1988 				NET_EPOCH_ENTER(et);
1989 				error = mld_v1_transmit_report(inm,
1990 				     MLD_LISTENER_REPORT);
1991 				NET_EPOCH_EXIT(et);
1992 				if (error == 0) {
1993 					inm->in6m_timer = odelay;
1994 					V_current_state_timers_running6 = 1;
1995 				}
1996 			}
1997 			break;
1998 
1999 		case MLD_VERSION_2:
2000 			/*
2001 			 * Defer update of T0 to T1, until the first copy
2002 			 * of the state change has been transmitted.
2003 			 */
2004 			syncstates = 0;
2005 
2006 			/*
2007 			 * Immediately enqueue a State-Change Report for
2008 			 * this interface, freeing any previous reports.
2009 			 * Don't kick the timers if there is nothing to do,
2010 			 * or if an error occurred.
2011 			 */
2012 			mq = &inm->in6m_scq;
2013 			mbufq_drain(mq);
2014 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2015 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2016 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2017 			    __func__, retval);
2018 			if (retval <= 0) {
2019 				error = retval * -1;
2020 				break;
2021 			}
2022 
2023 			/*
2024 			 * Schedule transmission of pending state-change
2025 			 * report up to RV times for this link. The timer
2026 			 * will fire at the next mld_fasttimo (~200ms),
2027 			 * giving us an opportunity to merge the reports.
2028 			 *
2029 			 * If a delay was provided to this function, only
2030 			 * use this delay if sooner than the existing one.
2031 			 */
2032 			KASSERT(mli->mli_rv > 1,
2033 			   ("%s: invalid robustness %d", __func__,
2034 			    mli->mli_rv));
2035 			inm->in6m_scrv = mli->mli_rv;
2036 			if (delay) {
2037 				if (inm->in6m_sctimer > 1) {
2038 					inm->in6m_sctimer =
2039 					    min(inm->in6m_sctimer, delay);
2040 				} else
2041 					inm->in6m_sctimer = delay;
2042 			} else
2043 				inm->in6m_sctimer = 1;
2044 			V_state_change_timers_running6 = 1;
2045 
2046 			error = 0;
2047 			break;
2048 		}
2049 	}
2050 
2051 	/*
2052 	 * Only update the T0 state if state change is atomic,
2053 	 * i.e. we don't need to wait for a timer to fire before we
2054 	 * can consider the state change to have been communicated.
2055 	 */
2056 	if (syncstates) {
2057 		in6m_commit(inm);
2058 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2059 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2060 		    if_name(inm->in6m_ifp));
2061 	}
2062 
2063 	return (error);
2064 }
2065 
2066 /*
2067  * Issue an intermediate state change during the life-cycle.
2068  */
2069 static int
mld_handle_state_change(struct in6_multi * inm,struct mld_ifsoftc * mli)2070 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2071 {
2072 	struct ifnet		*ifp;
2073 	int			 retval;
2074 #ifdef KTR
2075 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2076 #endif
2077 
2078 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2079 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2080 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2081 
2082 	ifp = inm->in6m_ifp;
2083 
2084 	IN6_MULTI_LIST_LOCK_ASSERT();
2085 	MLD_LOCK_ASSERT();
2086 
2087 	KASSERT(mli && mli->mli_ifp == ifp,
2088 	    ("%s: inconsistent ifp", __func__));
2089 
2090 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2091 	    (mli->mli_flags & MLIF_SILENT) ||
2092 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2093 	    (mli->mli_version != MLD_VERSION_2)) {
2094 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2095 			CTR1(KTR_MLD,
2096 "%s: not kicking state machine for silent group", __func__);
2097 		}
2098 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2099 		in6m_commit(inm);
2100 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2101 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2102 		    if_name(inm->in6m_ifp));
2103 		return (0);
2104 	}
2105 
2106 	mbufq_drain(&inm->in6m_scq);
2107 
2108 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2109 	    (mli->mli_flags & MLIF_USEALLOW));
2110 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2111 	if (retval <= 0)
2112 		return (-retval);
2113 
2114 	/*
2115 	 * If record(s) were enqueued, start the state-change
2116 	 * report timer for this group.
2117 	 */
2118 	inm->in6m_scrv = mli->mli_rv;
2119 	inm->in6m_sctimer = 1;
2120 	V_state_change_timers_running6 = 1;
2121 
2122 	return (0);
2123 }
2124 
2125 /*
2126  * Perform the final leave for a multicast address.
2127  *
2128  * When leaving a group:
2129  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2130  *  MLDv2 enqueues a state-change report containing a transition
2131  *  to INCLUDE {} for immediate transmission.
2132  */
2133 static void
mld_final_leave(struct in6_multi * inm,struct mld_ifsoftc * mli)2134 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2135 {
2136 	struct epoch_tracker     et;
2137 #ifdef KTR
2138 	char ip6tbuf[INET6_ADDRSTRLEN];
2139 #endif
2140 
2141 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2142 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2143 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2144 
2145 	IN6_MULTI_LIST_LOCK_ASSERT();
2146 	MLD_LOCK_ASSERT();
2147 
2148 	switch (inm->in6m_state) {
2149 	case MLD_NOT_MEMBER:
2150 	case MLD_SILENT_MEMBER:
2151 	case MLD_LEAVING_MEMBER:
2152 		/* Already leaving or left; do nothing. */
2153 		CTR1(KTR_MLD,
2154 "%s: not kicking state machine for silent group", __func__);
2155 		break;
2156 	case MLD_REPORTING_MEMBER:
2157 	case MLD_IDLE_MEMBER:
2158 	case MLD_G_QUERY_PENDING_MEMBER:
2159 	case MLD_SG_QUERY_PENDING_MEMBER:
2160 		if (mli->mli_version == MLD_VERSION_1) {
2161 #ifdef INVARIANTS
2162 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2163 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2164 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2165 			     __func__);
2166 #endif
2167 			NET_EPOCH_ENTER(et);
2168 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2169 			NET_EPOCH_EXIT(et);
2170 			inm->in6m_state = MLD_NOT_MEMBER;
2171 			V_current_state_timers_running6 = 1;
2172 		} else if (mli->mli_version == MLD_VERSION_2) {
2173 			/*
2174 			 * Stop group timer and all pending reports.
2175 			 * Immediately enqueue a state-change report
2176 			 * TO_IN {} to be sent on the next fast timeout,
2177 			 * giving us an opportunity to merge reports.
2178 			 */
2179 			mbufq_drain(&inm->in6m_scq);
2180 			inm->in6m_timer = 0;
2181 			inm->in6m_scrv = mli->mli_rv;
2182 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2183 			    "pending retransmissions.", __func__,
2184 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2185 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2186 			if (inm->in6m_scrv == 0) {
2187 				inm->in6m_state = MLD_NOT_MEMBER;
2188 				inm->in6m_sctimer = 0;
2189 			} else {
2190 				int retval __diagused;
2191 
2192 				in6m_acquire_locked(inm);
2193 
2194 				retval = mld_v2_enqueue_group_record(
2195 				    &inm->in6m_scq, inm, 1, 0, 0,
2196 				    (mli->mli_flags & MLIF_USEALLOW));
2197 				KASSERT(retval != 0,
2198 				    ("%s: enqueue record = %d", __func__,
2199 				     retval));
2200 
2201 				inm->in6m_state = MLD_LEAVING_MEMBER;
2202 				inm->in6m_sctimer = 1;
2203 				V_state_change_timers_running6 = 1;
2204 			}
2205 			break;
2206 		}
2207 		break;
2208 	case MLD_LAZY_MEMBER:
2209 	case MLD_SLEEPING_MEMBER:
2210 	case MLD_AWAKENING_MEMBER:
2211 		/* Our reports are suppressed; do nothing. */
2212 		break;
2213 	}
2214 
2215 	in6m_commit(inm);
2216 	CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2217 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2218 	    if_name(inm->in6m_ifp));
2219 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2220 	CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2221 	    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2222 }
2223 
2224 /*
2225  * Enqueue an MLDv2 group record to the given output queue.
2226  *
2227  * If is_state_change is zero, a current-state record is appended.
2228  * If is_state_change is non-zero, a state-change report is appended.
2229  *
2230  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2231  * If is_group_query is zero, and if there is a packet with free space
2232  * at the tail of the queue, it will be appended to providing there
2233  * is enough free space.
2234  * Otherwise a new mbuf packet chain is allocated.
2235  *
2236  * If is_source_query is non-zero, each source is checked to see if
2237  * it was recorded for a Group-Source query, and will be omitted if
2238  * it is not both in-mode and recorded.
2239  *
2240  * If use_block_allow is non-zero, state change reports for initial join
2241  * and final leave, on an inclusive mode group with a source list, will be
2242  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2243  *
2244  * The function will attempt to allocate leading space in the packet
2245  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2246  *
2247  * If successful the size of all data appended to the queue is returned,
2248  * otherwise an error code less than zero is returned, or zero if
2249  * no record(s) were appended.
2250  */
2251 static int
mld_v2_enqueue_group_record(struct mbufq * mq,struct in6_multi * inm,const int is_state_change,const int is_group_query,const int is_source_query,const int use_block_allow)2252 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2253     const int is_state_change, const int is_group_query,
2254     const int is_source_query, const int use_block_allow)
2255 {
2256 	struct mldv2_record	 mr;
2257 	struct mldv2_record	*pmr;
2258 	struct ifnet		*ifp;
2259 	struct ip6_msource	*ims, *nims;
2260 	struct mbuf		*m0, *m, *md;
2261 	int			 is_filter_list_change;
2262 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2263 	int			 record_has_sources;
2264 	int			 now;
2265 	int			 type;
2266 	uint8_t			 mode;
2267 #ifdef KTR
2268 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2269 #endif
2270 
2271 	IN6_MULTI_LIST_LOCK_ASSERT();
2272 
2273 	ifp = inm->in6m_ifp;
2274 	is_filter_list_change = 0;
2275 	m = NULL;
2276 	m0 = NULL;
2277 	m0srcs = 0;
2278 	msrcs = 0;
2279 	nbytes = 0;
2280 	nims = NULL;
2281 	record_has_sources = 1;
2282 	pmr = NULL;
2283 	type = MLD_DO_NOTHING;
2284 	mode = inm->in6m_st[1].iss_fmode;
2285 
2286 	/*
2287 	 * If we did not transition out of ASM mode during t0->t1,
2288 	 * and there are no source nodes to process, we can skip
2289 	 * the generation of source records.
2290 	 */
2291 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2292 	    inm->in6m_nsrc == 0)
2293 		record_has_sources = 0;
2294 
2295 	if (is_state_change) {
2296 		/*
2297 		 * Queue a state change record.
2298 		 * If the mode did not change, and there are non-ASM
2299 		 * listeners or source filters present,
2300 		 * we potentially need to issue two records for the group.
2301 		 * If there are ASM listeners, and there was no filter
2302 		 * mode transition of any kind, do nothing.
2303 		 *
2304 		 * If we are transitioning to MCAST_UNDEFINED, we need
2305 		 * not send any sources. A transition to/from this state is
2306 		 * considered inclusive with some special treatment.
2307 		 *
2308 		 * If we are rewriting initial joins/leaves to use
2309 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2310 		 * we need to send sources in all cases.
2311 		 */
2312 		if (mode != inm->in6m_st[0].iss_fmode) {
2313 			if (mode == MCAST_EXCLUDE) {
2314 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2315 				    __func__);
2316 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2317 			} else {
2318 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2319 				    __func__);
2320 				if (use_block_allow) {
2321 					/*
2322 					 * XXX
2323 					 * Here we're interested in state
2324 					 * edges either direction between
2325 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2326 					 * Perhaps we should just check
2327 					 * the group state, rather than
2328 					 * the filter mode.
2329 					 */
2330 					if (mode == MCAST_UNDEFINED) {
2331 						type = MLD_BLOCK_OLD_SOURCES;
2332 					} else {
2333 						type = MLD_ALLOW_NEW_SOURCES;
2334 					}
2335 				} else {
2336 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2337 					if (mode == MCAST_UNDEFINED)
2338 						record_has_sources = 0;
2339 				}
2340 			}
2341 		} else {
2342 			if (record_has_sources) {
2343 				is_filter_list_change = 1;
2344 			} else {
2345 				type = MLD_DO_NOTHING;
2346 			}
2347 		}
2348 	} else {
2349 		/*
2350 		 * Queue a current state record.
2351 		 */
2352 		if (mode == MCAST_EXCLUDE) {
2353 			type = MLD_MODE_IS_EXCLUDE;
2354 		} else if (mode == MCAST_INCLUDE) {
2355 			type = MLD_MODE_IS_INCLUDE;
2356 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2357 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2358 			     __func__, inm, inm->in6m_st[1].iss_asm));
2359 		}
2360 	}
2361 
2362 	/*
2363 	 * Generate the filter list changes using a separate function.
2364 	 */
2365 	if (is_filter_list_change)
2366 		return (mld_v2_enqueue_filter_change(mq, inm));
2367 
2368 	if (type == MLD_DO_NOTHING) {
2369 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2370 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2371 		    if_name(inm->in6m_ifp));
2372 		return (0);
2373 	}
2374 
2375 	/*
2376 	 * If any sources are present, we must be able to fit at least
2377 	 * one in the trailing space of the tail packet's mbuf,
2378 	 * ideally more.
2379 	 */
2380 	minrec0len = sizeof(struct mldv2_record);
2381 	if (record_has_sources)
2382 		minrec0len += sizeof(struct in6_addr);
2383 
2384 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2385 	    mld_rec_type_to_str(type),
2386 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2387 	    if_name(inm->in6m_ifp));
2388 
2389 	/*
2390 	 * Check if we have a packet in the tail of the queue for this
2391 	 * group into which the first group record for this group will fit.
2392 	 * Otherwise allocate a new packet.
2393 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2394 	 * Note: Group records for G/GSR query responses MUST be sent
2395 	 * in their own packet.
2396 	 */
2397 	m0 = mbufq_last(mq);
2398 	if (!is_group_query &&
2399 	    m0 != NULL &&
2400 	    (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2401 	    (m0->m_pkthdr.len + minrec0len) <
2402 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2403 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2404 			    sizeof(struct mldv2_record)) /
2405 			    sizeof(struct in6_addr);
2406 		m = m0;
2407 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2408 	} else {
2409 		if (mbufq_full(mq)) {
2410 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2411 			return (-ENOMEM);
2412 		}
2413 		m = NULL;
2414 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2415 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2416 		if (!is_state_change && !is_group_query)
2417 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2418 		if (m == NULL)
2419 			m = m_gethdr(M_NOWAIT, MT_DATA);
2420 		if (m == NULL)
2421 			return (-ENOMEM);
2422 
2423 		mld_save_context(m, ifp);
2424 
2425 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2426 	}
2427 
2428 	/*
2429 	 * Append group record.
2430 	 * If we have sources, we don't know how many yet.
2431 	 */
2432 	mr.mr_type = type;
2433 	mr.mr_datalen = 0;
2434 	mr.mr_numsrc = 0;
2435 	mr.mr_addr = inm->in6m_addr;
2436 	in6_clearscope(&mr.mr_addr);
2437 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2438 		if (m != m0)
2439 			m_freem(m);
2440 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2441 		return (-ENOMEM);
2442 	}
2443 	nbytes += sizeof(struct mldv2_record);
2444 
2445 	/*
2446 	 * Append as many sources as will fit in the first packet.
2447 	 * If we are appending to a new packet, the chain allocation
2448 	 * may potentially use clusters; use m_getptr() in this case.
2449 	 * If we are appending to an existing packet, we need to obtain
2450 	 * a pointer to the group record after m_append(), in case a new
2451 	 * mbuf was allocated.
2452 	 *
2453 	 * Only append sources which are in-mode at t1. If we are
2454 	 * transitioning to MCAST_UNDEFINED state on the group, and
2455 	 * use_block_allow is zero, do not include source entries.
2456 	 * Otherwise, we need to include this source in the report.
2457 	 *
2458 	 * Only report recorded sources in our filter set when responding
2459 	 * to a group-source query.
2460 	 */
2461 	if (record_has_sources) {
2462 		if (m == m0) {
2463 			md = m_last(m);
2464 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2465 			    md->m_len - nbytes);
2466 		} else {
2467 			md = m_getptr(m, 0, &off);
2468 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2469 			    off);
2470 		}
2471 		msrcs = 0;
2472 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2473 		    nims) {
2474 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2475 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2476 			now = im6s_get_mode(inm, ims, 1);
2477 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2478 			if ((now != mode) ||
2479 			    (now == mode &&
2480 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2481 				CTR1(KTR_MLD, "%s: skip node", __func__);
2482 				continue;
2483 			}
2484 			if (is_source_query && ims->im6s_stp == 0) {
2485 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2486 				    __func__);
2487 				continue;
2488 			}
2489 			CTR1(KTR_MLD, "%s: append node", __func__);
2490 			if (!m_append(m, sizeof(struct in6_addr),
2491 			    (void *)&ims->im6s_addr)) {
2492 				if (m != m0)
2493 					m_freem(m);
2494 				CTR1(KTR_MLD, "%s: m_append() failed.",
2495 				    __func__);
2496 				return (-ENOMEM);
2497 			}
2498 			nbytes += sizeof(struct in6_addr);
2499 			++msrcs;
2500 			if (msrcs == m0srcs)
2501 				break;
2502 		}
2503 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2504 		    msrcs);
2505 		pmr->mr_numsrc = htons(msrcs);
2506 		nbytes += (msrcs * sizeof(struct in6_addr));
2507 	}
2508 
2509 	if (is_source_query && msrcs == 0) {
2510 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2511 		if (m != m0)
2512 			m_freem(m);
2513 		return (0);
2514 	}
2515 
2516 	/*
2517 	 * We are good to go with first packet.
2518 	 */
2519 	if (m != m0) {
2520 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2521 		m->m_pkthdr.vt_nrecs = 1;
2522 		mbufq_enqueue(mq, m);
2523 	} else
2524 		m->m_pkthdr.vt_nrecs++;
2525 
2526 	/*
2527 	 * No further work needed if no source list in packet(s).
2528 	 */
2529 	if (!record_has_sources)
2530 		return (nbytes);
2531 
2532 	/*
2533 	 * Whilst sources remain to be announced, we need to allocate
2534 	 * a new packet and fill out as many sources as will fit.
2535 	 * Always try for a cluster first.
2536 	 */
2537 	while (nims != NULL) {
2538 		if (mbufq_full(mq)) {
2539 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2540 			return (-ENOMEM);
2541 		}
2542 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2543 		if (m == NULL)
2544 			m = m_gethdr(M_NOWAIT, MT_DATA);
2545 		if (m == NULL)
2546 			return (-ENOMEM);
2547 		mld_save_context(m, ifp);
2548 		md = m_getptr(m, 0, &off);
2549 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2550 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2551 
2552 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2553 			if (m != m0)
2554 				m_freem(m);
2555 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2556 			return (-ENOMEM);
2557 		}
2558 		m->m_pkthdr.vt_nrecs = 1;
2559 		nbytes += sizeof(struct mldv2_record);
2560 
2561 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2562 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2563 
2564 		msrcs = 0;
2565 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2566 			CTR2(KTR_MLD, "%s: visit node %s",
2567 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2568 			now = im6s_get_mode(inm, ims, 1);
2569 			if ((now != mode) ||
2570 			    (now == mode &&
2571 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2572 				CTR1(KTR_MLD, "%s: skip node", __func__);
2573 				continue;
2574 			}
2575 			if (is_source_query && ims->im6s_stp == 0) {
2576 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2577 				    __func__);
2578 				continue;
2579 			}
2580 			CTR1(KTR_MLD, "%s: append node", __func__);
2581 			if (!m_append(m, sizeof(struct in6_addr),
2582 			    (void *)&ims->im6s_addr)) {
2583 				if (m != m0)
2584 					m_freem(m);
2585 				CTR1(KTR_MLD, "%s: m_append() failed.",
2586 				    __func__);
2587 				return (-ENOMEM);
2588 			}
2589 			++msrcs;
2590 			if (msrcs == m0srcs)
2591 				break;
2592 		}
2593 		pmr->mr_numsrc = htons(msrcs);
2594 		nbytes += (msrcs * sizeof(struct in6_addr));
2595 
2596 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2597 		mbufq_enqueue(mq, m);
2598 	}
2599 
2600 	return (nbytes);
2601 }
2602 
2603 /*
2604  * Type used to mark record pass completion.
2605  * We exploit the fact we can cast to this easily from the
2606  * current filter modes on each ip_msource node.
2607  */
2608 typedef enum {
2609 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2610 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2611 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2612 	REC_FULL = REC_ALLOW | REC_BLOCK
2613 } rectype_t;
2614 
2615 /*
2616  * Enqueue an MLDv2 filter list change to the given output queue.
2617  *
2618  * Source list filter state is held in an RB-tree. When the filter list
2619  * for a group is changed without changing its mode, we need to compute
2620  * the deltas between T0 and T1 for each source in the filter set,
2621  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2622  *
2623  * As we may potentially queue two record types, and the entire R-B tree
2624  * needs to be walked at once, we break this out into its own function
2625  * so we can generate a tightly packed queue of packets.
2626  *
2627  * XXX This could be written to only use one tree walk, although that makes
2628  * serializing into the mbuf chains a bit harder. For now we do two walks
2629  * which makes things easier on us, and it may or may not be harder on
2630  * the L2 cache.
2631  *
2632  * If successful the size of all data appended to the queue is returned,
2633  * otherwise an error code less than zero is returned, or zero if
2634  * no record(s) were appended.
2635  */
2636 static int
mld_v2_enqueue_filter_change(struct mbufq * mq,struct in6_multi * inm)2637 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2638 {
2639 	static const int MINRECLEN =
2640 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2641 	struct ifnet		*ifp;
2642 	struct mldv2_record	 mr;
2643 	struct mldv2_record	*pmr;
2644 	struct ip6_msource	*ims, *nims;
2645 	struct mbuf		*m, *m0, *md;
2646 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2647 	uint8_t			 mode, now, then;
2648 	rectype_t		 crt, drt, nrt;
2649 #ifdef KTR
2650 	int			 nallow, nblock;
2651 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2652 #endif
2653 
2654 	IN6_MULTI_LIST_LOCK_ASSERT();
2655 
2656 	if (inm->in6m_nsrc == 0 ||
2657 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2658 		return (0);
2659 
2660 	ifp = inm->in6m_ifp;			/* interface */
2661 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2662 	crt = REC_NONE;	/* current group record type */
2663 	drt = REC_NONE;	/* mask of completed group record types */
2664 	nrt = REC_NONE;	/* record type for current node */
2665 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2666 	npbytes = 0;	/* # of bytes appended this packet */
2667 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2668 	rsrcs = 0;	/* # sources encoded in current record */
2669 	schanged = 0;	/* # nodes encoded in overall filter change */
2670 #ifdef KTR
2671 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2672 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2673 #endif
2674 	nims = NULL;	/* next tree node pointer */
2675 
2676 	/*
2677 	 * For each possible filter record mode.
2678 	 * The first kind of source we encounter tells us which
2679 	 * is the first kind of record we start appending.
2680 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2681 	 * as the inverse of the group's filter mode.
2682 	 */
2683 	while (drt != REC_FULL) {
2684 		do {
2685 			m0 = mbufq_last(mq);
2686 			if (m0 != NULL &&
2687 			    (m0->m_pkthdr.vt_nrecs + 1 <=
2688 			     MLD_V2_REPORT_MAXRECS) &&
2689 			    (m0->m_pkthdr.len + MINRECLEN) <
2690 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2691 				m = m0;
2692 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2693 					    sizeof(struct mldv2_record)) /
2694 					    sizeof(struct in6_addr);
2695 				CTR1(KTR_MLD,
2696 				    "%s: use previous packet", __func__);
2697 			} else {
2698 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2699 				if (m == NULL)
2700 					m = m_gethdr(M_NOWAIT, MT_DATA);
2701 				if (m == NULL) {
2702 					CTR1(KTR_MLD,
2703 					    "%s: m_get*() failed", __func__);
2704 					return (-ENOMEM);
2705 				}
2706 				m->m_pkthdr.vt_nrecs = 0;
2707 				mld_save_context(m, ifp);
2708 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2709 				    sizeof(struct mldv2_record)) /
2710 				    sizeof(struct in6_addr);
2711 				npbytes = 0;
2712 				CTR1(KTR_MLD,
2713 				    "%s: allocated new packet", __func__);
2714 			}
2715 			/*
2716 			 * Append the MLD group record header to the
2717 			 * current packet's data area.
2718 			 * Recalculate pointer to free space for next
2719 			 * group record, in case m_append() allocated
2720 			 * a new mbuf or cluster.
2721 			 */
2722 			memset(&mr, 0, sizeof(mr));
2723 			mr.mr_addr = inm->in6m_addr;
2724 			in6_clearscope(&mr.mr_addr);
2725 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2726 				if (m != m0)
2727 					m_freem(m);
2728 				CTR1(KTR_MLD,
2729 				    "%s: m_append() failed", __func__);
2730 				return (-ENOMEM);
2731 			}
2732 			npbytes += sizeof(struct mldv2_record);
2733 			if (m != m0) {
2734 				/* new packet; offset in chain */
2735 				md = m_getptr(m, npbytes -
2736 				    sizeof(struct mldv2_record), &off);
2737 				pmr = (struct mldv2_record *)(mtod(md,
2738 				    uint8_t *) + off);
2739 			} else {
2740 				/* current packet; offset from last append */
2741 				md = m_last(m);
2742 				pmr = (struct mldv2_record *)(mtod(md,
2743 				    uint8_t *) + md->m_len -
2744 				    sizeof(struct mldv2_record));
2745 			}
2746 			/*
2747 			 * Begin walking the tree for this record type
2748 			 * pass, or continue from where we left off
2749 			 * previously if we had to allocate a new packet.
2750 			 * Only report deltas in-mode at t1.
2751 			 * We need not report included sources as allowed
2752 			 * if we are in inclusive mode on the group,
2753 			 * however the converse is not true.
2754 			 */
2755 			rsrcs = 0;
2756 			if (nims == NULL) {
2757 				nims = RB_MIN(ip6_msource_tree,
2758 				    &inm->in6m_srcs);
2759 			}
2760 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2761 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2762 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2763 				now = im6s_get_mode(inm, ims, 1);
2764 				then = im6s_get_mode(inm, ims, 0);
2765 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2766 				    __func__, then, now);
2767 				if (now == then) {
2768 					CTR1(KTR_MLD,
2769 					    "%s: skip unchanged", __func__);
2770 					continue;
2771 				}
2772 				if (mode == MCAST_EXCLUDE &&
2773 				    now == MCAST_INCLUDE) {
2774 					CTR1(KTR_MLD,
2775 					    "%s: skip IN src on EX group",
2776 					    __func__);
2777 					continue;
2778 				}
2779 				nrt = (rectype_t)now;
2780 				if (nrt == REC_NONE)
2781 					nrt = (rectype_t)(~mode & REC_FULL);
2782 				if (schanged++ == 0) {
2783 					crt = nrt;
2784 				} else if (crt != nrt)
2785 					continue;
2786 				if (!m_append(m, sizeof(struct in6_addr),
2787 				    (void *)&ims->im6s_addr)) {
2788 					if (m != m0)
2789 						m_freem(m);
2790 					CTR1(KTR_MLD,
2791 					    "%s: m_append() failed", __func__);
2792 					return (-ENOMEM);
2793 				}
2794 #ifdef KTR
2795 				nallow += !!(crt == REC_ALLOW);
2796 				nblock += !!(crt == REC_BLOCK);
2797 #endif
2798 				if (++rsrcs == m0srcs)
2799 					break;
2800 			}
2801 			/*
2802 			 * If we did not append any tree nodes on this
2803 			 * pass, back out of allocations.
2804 			 */
2805 			if (rsrcs == 0) {
2806 				npbytes -= sizeof(struct mldv2_record);
2807 				if (m != m0) {
2808 					CTR1(KTR_MLD,
2809 					    "%s: m_free(m)", __func__);
2810 					m_freem(m);
2811 				} else {
2812 					CTR1(KTR_MLD,
2813 					    "%s: m_adj(m, -mr)", __func__);
2814 					m_adj(m, -((int)sizeof(
2815 					    struct mldv2_record)));
2816 				}
2817 				continue;
2818 			}
2819 			npbytes += (rsrcs * sizeof(struct in6_addr));
2820 			if (crt == REC_ALLOW)
2821 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2822 			else if (crt == REC_BLOCK)
2823 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2824 			pmr->mr_numsrc = htons(rsrcs);
2825 			/*
2826 			 * Count the new group record, and enqueue this
2827 			 * packet if it wasn't already queued.
2828 			 */
2829 			m->m_pkthdr.vt_nrecs++;
2830 			if (m != m0)
2831 				mbufq_enqueue(mq, m);
2832 			nbytes += npbytes;
2833 		} while (nims != NULL);
2834 		drt |= crt;
2835 		crt = (~crt & REC_FULL);
2836 	}
2837 
2838 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2839 	    nallow, nblock);
2840 
2841 	return (nbytes);
2842 }
2843 
2844 static int
mld_v2_merge_state_changes(struct in6_multi * inm,struct mbufq * scq)2845 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2846 {
2847 	struct mbufq	*gq;
2848 	struct mbuf	*m;		/* pending state-change */
2849 	struct mbuf	*m0;		/* copy of pending state-change */
2850 	struct mbuf	*mt;		/* last state-change in packet */
2851 	int		 docopy, domerge;
2852 	u_int		 recslen;
2853 
2854 	docopy = 0;
2855 	domerge = 0;
2856 	recslen = 0;
2857 
2858 	IN6_MULTI_LIST_LOCK_ASSERT();
2859 	MLD_LOCK_ASSERT();
2860 
2861 	/*
2862 	 * If there are further pending retransmissions, make a writable
2863 	 * copy of each queued state-change message before merging.
2864 	 */
2865 	if (inm->in6m_scrv > 0)
2866 		docopy = 1;
2867 
2868 	gq = &inm->in6m_scq;
2869 #ifdef KTR
2870 	if (mbufq_first(gq) == NULL) {
2871 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2872 		    __func__, inm);
2873 	}
2874 #endif
2875 
2876 	m = mbufq_first(gq);
2877 	while (m != NULL) {
2878 		/*
2879 		 * Only merge the report into the current packet if
2880 		 * there is sufficient space to do so; an MLDv2 report
2881 		 * packet may only contain 65,535 group records.
2882 		 * Always use a simple mbuf chain concatentation to do this,
2883 		 * as large state changes for single groups may have
2884 		 * allocated clusters.
2885 		 */
2886 		domerge = 0;
2887 		mt = mbufq_last(scq);
2888 		if (mt != NULL) {
2889 			recslen = m_length(m, NULL);
2890 
2891 			if ((mt->m_pkthdr.vt_nrecs +
2892 			    m->m_pkthdr.vt_nrecs <=
2893 			    MLD_V2_REPORT_MAXRECS) &&
2894 			    (mt->m_pkthdr.len + recslen <=
2895 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2896 				domerge = 1;
2897 		}
2898 
2899 		if (!domerge && mbufq_full(gq)) {
2900 			CTR2(KTR_MLD,
2901 			    "%s: outbound queue full, skipping whole packet %p",
2902 			    __func__, m);
2903 			mt = m->m_nextpkt;
2904 			if (!docopy)
2905 				m_freem(m);
2906 			m = mt;
2907 			continue;
2908 		}
2909 
2910 		if (!docopy) {
2911 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2912 			m0 = mbufq_dequeue(gq);
2913 			m = m0->m_nextpkt;
2914 		} else {
2915 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2916 			m0 = m_dup(m, M_NOWAIT);
2917 			if (m0 == NULL)
2918 				return (ENOMEM);
2919 			m0->m_nextpkt = NULL;
2920 			m = m->m_nextpkt;
2921 		}
2922 
2923 		if (!domerge) {
2924 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2925 			    __func__, m0, scq);
2926 			mbufq_enqueue(scq, m0);
2927 		} else {
2928 			struct mbuf *mtl;	/* last mbuf of packet mt */
2929 
2930 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2931 			    __func__, m0, mt);
2932 
2933 			mtl = m_last(mt);
2934 			m0->m_flags &= ~M_PKTHDR;
2935 			mt->m_pkthdr.len += recslen;
2936 			mt->m_pkthdr.vt_nrecs +=
2937 			    m0->m_pkthdr.vt_nrecs;
2938 
2939 			mtl->m_next = m0;
2940 		}
2941 	}
2942 
2943 	return (0);
2944 }
2945 
2946 /*
2947  * Respond to a pending MLDv2 General Query.
2948  */
2949 static void
mld_v2_dispatch_general_query(struct mld_ifsoftc * mli)2950 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2951 {
2952 	struct ifmultiaddr	*ifma;
2953 	struct ifnet		*ifp;
2954 	struct in6_multi	*inm;
2955 	int			 retval __unused;
2956 
2957 	NET_EPOCH_ASSERT();
2958 	IN6_MULTI_LIST_LOCK_ASSERT();
2959 	MLD_LOCK_ASSERT();
2960 
2961 	KASSERT(mli->mli_version == MLD_VERSION_2,
2962 	    ("%s: called when version %d", __func__, mli->mli_version));
2963 
2964 	/*
2965 	 * Check that there are some packets queued. If so, send them first.
2966 	 * For large number of groups the reply to general query can take
2967 	 * many packets, we should finish sending them before starting of
2968 	 * queuing the new reply.
2969 	 */
2970 	if (!mbufq_empty(&mli->mli_gq))
2971 		goto send;
2972 
2973 	ifp = mli->mli_ifp;
2974 
2975 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2976 		inm = in6m_ifmultiaddr_get_inm(ifma);
2977 		if (inm == NULL)
2978 			continue;
2979 		KASSERT(ifp == inm->in6m_ifp,
2980 		    ("%s: inconsistent ifp", __func__));
2981 
2982 		switch (inm->in6m_state) {
2983 		case MLD_NOT_MEMBER:
2984 		case MLD_SILENT_MEMBER:
2985 			break;
2986 		case MLD_REPORTING_MEMBER:
2987 		case MLD_IDLE_MEMBER:
2988 		case MLD_LAZY_MEMBER:
2989 		case MLD_SLEEPING_MEMBER:
2990 		case MLD_AWAKENING_MEMBER:
2991 			inm->in6m_state = MLD_REPORTING_MEMBER;
2992 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
2993 			    inm, 0, 0, 0, 0);
2994 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2995 			    __func__, retval);
2996 			break;
2997 		case MLD_G_QUERY_PENDING_MEMBER:
2998 		case MLD_SG_QUERY_PENDING_MEMBER:
2999 		case MLD_LEAVING_MEMBER:
3000 			break;
3001 		}
3002 	}
3003 
3004 send:
3005 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3006 
3007 	/*
3008 	 * Slew transmission of bursts over 500ms intervals.
3009 	 */
3010 	if (mbufq_first(&mli->mli_gq) != NULL) {
3011 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3012 		    MLD_RESPONSE_BURST_INTERVAL);
3013 		V_interface_timers_running6 = 1;
3014 	}
3015 }
3016 
3017 /*
3018  * Transmit the next pending message in the output queue.
3019  *
3020  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3021  * MRT: Nothing needs to be done, as MLD traffic is always local to
3022  * a link and uses a link-scope multicast address.
3023  */
3024 static void
mld_dispatch_packet(struct mbuf * m)3025 mld_dispatch_packet(struct mbuf *m)
3026 {
3027 	struct ip6_moptions	 im6o;
3028 	struct ifnet		*ifp;
3029 	struct ifnet		*oifp;
3030 	struct mbuf		*m0;
3031 	struct mbuf		*md;
3032 	struct ip6_hdr		*ip6;
3033 	struct mld_hdr		*mld;
3034 	int			 error;
3035 	int			 off;
3036 	int			 type;
3037 	uint32_t		 ifindex;
3038 
3039 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3040 	NET_EPOCH_ASSERT();
3041 
3042 	/*
3043 	 * Set VNET image pointer from enqueued mbuf chain
3044 	 * before doing anything else. Whilst we use interface
3045 	 * indexes to guard against interface detach, they are
3046 	 * unique to each VIMAGE and must be retrieved.
3047 	 */
3048 	ifindex = mld_restore_context(m);
3049 
3050 	/*
3051 	 * Check if the ifnet still exists. This limits the scope of
3052 	 * any race in the absence of a global ifp lock for low cost
3053 	 * (an array lookup).
3054 	 */
3055 	ifp = ifnet_byindex(ifindex);
3056 	if (ifp == NULL) {
3057 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3058 		    __func__, m, ifindex);
3059 		m_freem(m);
3060 		IP6STAT_INC(ip6s_noroute);
3061 		goto out;
3062 	}
3063 
3064 	im6o.im6o_multicast_hlim  = 1;
3065 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3066 	im6o.im6o_multicast_ifp = ifp;
3067 
3068 	if (m->m_flags & M_MLDV1) {
3069 		m0 = m;
3070 	} else {
3071 		m0 = mld_v2_encap_report(ifp, m);
3072 		if (m0 == NULL) {
3073 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3074 			IP6STAT_INC(ip6s_odropped);
3075 			goto out;
3076 		}
3077 	}
3078 
3079 	mld_scrub_context(m0);
3080 	m_clrprotoflags(m);
3081 	m0->m_pkthdr.rcvif = V_loif;
3082 
3083 	ip6 = mtod(m0, struct ip6_hdr *);
3084 #if 0
3085 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3086 #else
3087 	/*
3088 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3089 	 * occur if we called in6_setscope() at transmission.
3090 	 * See comments at top of file.
3091 	 */
3092 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3093 #endif
3094 
3095 	/*
3096 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3097 	 * so we can bump the stats.
3098 	 */
3099 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3100 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3101 	type = mld->mld_type;
3102 
3103 	oifp = NULL;
3104 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3105 	    &oifp, NULL);
3106 	if (error) {
3107 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3108 		goto out;
3109 	}
3110 	ICMP6STAT_INC2(icp6s_outhist, type);
3111 	if (oifp != NULL) {
3112 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3113 		switch (type) {
3114 		case MLD_LISTENER_REPORT:
3115 		case MLDV2_LISTENER_REPORT:
3116 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3117 			break;
3118 		case MLD_LISTENER_DONE:
3119 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3120 			break;
3121 		}
3122 	}
3123 out:
3124 	return;
3125 }
3126 
3127 /*
3128  * Encapsulate an MLDv2 report.
3129  *
3130  * KAME IPv6 requires that hop-by-hop options be passed separately,
3131  * and that the IPv6 header be prepended in a separate mbuf.
3132  *
3133  * Returns a pointer to the new mbuf chain head, or NULL if the
3134  * allocation failed.
3135  */
3136 static struct mbuf *
mld_v2_encap_report(struct ifnet * ifp,struct mbuf * m)3137 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3138 {
3139 	struct mbuf		*mh;
3140 	struct mldv2_report	*mld;
3141 	struct ip6_hdr		*ip6;
3142 	struct in6_ifaddr	*ia;
3143 	int			 mldreclen;
3144 
3145 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3146 	KASSERT((m->m_flags & M_PKTHDR),
3147 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3148 
3149 	/*
3150 	 * RFC3590: OK to send as :: or tentative during DAD.
3151 	 */
3152 	NET_EPOCH_ASSERT();
3153 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3154 	if (ia == NULL)
3155 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3156 
3157 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3158 	if (mh == NULL) {
3159 		if (ia != NULL)
3160 			ifa_free(&ia->ia_ifa);
3161 		m_freem(m);
3162 		return (NULL);
3163 	}
3164 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3165 
3166 	mldreclen = m_length(m, NULL);
3167 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3168 
3169 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3170 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3171 	    sizeof(struct mldv2_report) + mldreclen;
3172 
3173 	ip6 = mtod(mh, struct ip6_hdr *);
3174 	ip6->ip6_flow = 0;
3175 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3176 	ip6->ip6_vfc |= IPV6_VERSION;
3177 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3178 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3179 	if (ia != NULL)
3180 		ifa_free(&ia->ia_ifa);
3181 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3182 	/* scope ID will be set in netisr */
3183 
3184 	mld = (struct mldv2_report *)(ip6 + 1);
3185 	mld->mld_type = MLDV2_LISTENER_REPORT;
3186 	mld->mld_code = 0;
3187 	mld->mld_cksum = 0;
3188 	mld->mld_v2_reserved = 0;
3189 	mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3190 	m->m_pkthdr.vt_nrecs = 0;
3191 
3192 	mh->m_next = m;
3193 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3194 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3195 	return (mh);
3196 }
3197 
3198 #ifdef KTR
3199 static char *
mld_rec_type_to_str(const int type)3200 mld_rec_type_to_str(const int type)
3201 {
3202 
3203 	switch (type) {
3204 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3205 			return "TO_EX";
3206 			break;
3207 		case MLD_CHANGE_TO_INCLUDE_MODE:
3208 			return "TO_IN";
3209 			break;
3210 		case MLD_MODE_IS_EXCLUDE:
3211 			return "MODE_EX";
3212 			break;
3213 		case MLD_MODE_IS_INCLUDE:
3214 			return "MODE_IN";
3215 			break;
3216 		case MLD_ALLOW_NEW_SOURCES:
3217 			return "ALLOW_NEW";
3218 			break;
3219 		case MLD_BLOCK_OLD_SOURCES:
3220 			return "BLOCK_OLD";
3221 			break;
3222 		default:
3223 			break;
3224 	}
3225 	return "unknown";
3226 }
3227 #endif
3228 
3229 static void
mld_init(void * unused __unused)3230 mld_init(void *unused __unused)
3231 {
3232 
3233 	CTR1(KTR_MLD, "%s: initializing", __func__);
3234 	MLD_LOCK_INIT();
3235 
3236 	ip6_initpktopts(&mld_po);
3237 	mld_po.ip6po_hlim = 1;
3238 	mld_po.ip6po_hbh = &mld_ra.hbh;
3239 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3240 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3241 	mld_po.ip6po_valid = IP6PO_VALID_HLIM | IP6PO_VALID_HBH;
3242 
3243 	callout_init(&mldslow_callout, 1);
3244 	callout_reset(&mldslow_callout, hz / MLD_SLOWHZ, mld_slowtimo, NULL);
3245 	callout_init(&mldfast_callout, 1);
3246 	callout_reset(&mldfast_callout, hz / MLD_FASTHZ, mld_fasttimo, NULL);
3247 }
3248 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3249 
3250 static void
mld_uninit(void * unused __unused)3251 mld_uninit(void *unused __unused)
3252 {
3253 
3254 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3255 	callout_drain(&mldslow_callout);
3256 	callout_drain(&mldfast_callout);
3257 	MLD_LOCK_DESTROY();
3258 }
3259 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3260 
3261 static void
vnet_mld_init(const void * unused __unused)3262 vnet_mld_init(const void *unused __unused)
3263 {
3264 
3265 	CTR1(KTR_MLD, "%s: initializing", __func__);
3266 
3267 	LIST_INIT(&V_mli_head);
3268 }
3269 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3270     NULL);
3271 
3272 static void
vnet_mld_uninit(const void * unused __unused)3273 vnet_mld_uninit(const void *unused __unused)
3274 {
3275 
3276 	/* This can happen if we shutdown the network stack. */
3277 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3278 }
3279 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3280     NULL);
3281 
3282 static int
mld_modevent(module_t mod,int type,void * unused __unused)3283 mld_modevent(module_t mod, int type, void *unused __unused)
3284 {
3285 
3286     switch (type) {
3287     case MOD_LOAD:
3288     case MOD_UNLOAD:
3289 	break;
3290     default:
3291 	return (EOPNOTSUPP);
3292     }
3293     return (0);
3294 }
3295 
3296 static moduledata_t mld_mod = {
3297     "mld",
3298     mld_modevent,
3299     0
3300 };
3301 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3302