xref: /xnu-8792.81.2/bsd/netinet6/mld6.c (revision 19c3b8c28c31cb8130e034cfb5df6bf9ba342d90)
1 /*
2  * Copyright (c) 2000-2020 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 /*-
29  * Copyright (c) 2009 Bruce Simpson.
30  *
31  * Redistribution and use in source and binary forms, with or without
32  * modification, are permitted provided that the following conditions
33  * are met:
34  * 1. Redistributions of source code must retain the above copyright
35  *    notice, this list of conditions and the following disclaimer.
36  * 2. Redistributions in binary form must reproduce the above copyright
37  *    notice, this list of conditions and the following disclaimer in the
38  *    documentation and/or other materials provided with the distribution.
39  * 3. The name of the author may not be used to endorse or promote
40  *    products derived from this software without specific prior written
41  *    permission.
42  *
43  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
44  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
45  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
46  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
47  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
48  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
49  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
50  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
51  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
52  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53  * SUCH DAMAGE.
54  */
55 
56 /*
57  * Copyright (c) 1988 Stephen Deering.
58  * Copyright (c) 1992, 1993
59  *	The Regents of the University of California.  All rights reserved.
60  *
61  * This code is derived from software contributed to Berkeley by
62  * Stephen Deering of Stanford University.
63  *
64  * Redistribution and use in source and binary forms, with or without
65  * modification, are permitted provided that the following conditions
66  * are met:
67  * 1. Redistributions of source code must retain the above copyright
68  *    notice, this list of conditions and the following disclaimer.
69  * 2. Redistributions in binary form must reproduce the above copyright
70  *    notice, this list of conditions and the following disclaimer in the
71  *    documentation and/or other materials provided with the distribution.
72  * 3. All advertising materials mentioning features or use of this software
73  *    must display the following acknowledgement:
74  *	This product includes software developed by the University of
75  *	California, Berkeley and its contributors.
76  * 4. Neither the name of the University nor the names of its contributors
77  *    may be used to endorse or promote products derived from this software
78  *    without specific prior written permission.
79  *
80  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
81  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
82  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
83  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
84  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
85  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
86  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
87  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
88  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
89  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
90  * SUCH DAMAGE.
91  *
92  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
93  */
94 /*
95  * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
96  * support for mandatory and extensible security protections.  This notice
97  * is included in support of clause 2.2 (b) of the Apple Public License,
98  * Version 2.0.
99  */
100 
101 #include <sys/cdefs.h>
102 
103 #include <sys/param.h>
104 #include <sys/systm.h>
105 #include <sys/mbuf.h>
106 #include <sys/socket.h>
107 #include <sys/protosw.h>
108 #include <sys/sysctl.h>
109 #include <sys/kernel.h>
110 #include <sys/malloc.h>
111 #include <sys/mcache.h>
112 
113 #include <dev/random/randomdev.h>
114 
115 #include <kern/zalloc.h>
116 
117 #include <net/if.h>
118 #include <net/route.h>
119 
120 #include <netinet/in.h>
121 #include <netinet/in_var.h>
122 #include <netinet6/in6_var.h>
123 #include <netinet/ip6.h>
124 #include <netinet6/ip6_var.h>
125 #include <netinet6/scope6_var.h>
126 #include <netinet/icmp6.h>
127 #include <netinet6/mld6.h>
128 #include <netinet6/mld6_var.h>
129 
130 /* Lock group and attribute for mld_mtx */
131 static LCK_ATTR_DECLARE(mld_mtx_attr, 0, 0);
132 static LCK_GRP_DECLARE(mld_mtx_grp, "mld_mtx");
133 
134 /*
135  * Locking and reference counting:
136  *
137  * mld_mtx mainly protects mli_head.  In cases where both mld_mtx and
138  * in6_multihead_lock must be held, the former must be acquired first in order
139  * to maintain lock ordering.  It is not a requirement that mld_mtx be
140  * acquired first before in6_multihead_lock, but in case both must be acquired
141  * in succession, the correct lock ordering must be followed.
142  *
143  * Instead of walking the if_multiaddrs list at the interface and returning
144  * the ifma_protospec value of a matching entry, we search the global list
145  * of in6_multi records and find it that way; this is done with in6_multihead
146  * lock held.  Doing so avoids the race condition issues that many other BSDs
147  * suffer from (therefore in our implementation, ifma_protospec will never be
148  * NULL for as long as the in6_multi is valid.)
149  *
150  * The above creates a requirement for the in6_multi to stay in in6_multihead
151  * list even after the final MLD leave (in MLDv2 mode) until no longer needs
152  * be retransmitted (this is not required for MLDv1.)  In order to handle
153  * this, the request and reference counts of the in6_multi are bumped up when
154  * the state changes to MLD_LEAVING_MEMBER, and later dropped in the timeout
155  * handler.  Each in6_multi holds a reference to the underlying mld_ifinfo.
156  *
157  * Thus, the permitted lock order is:
158  *
159  *	mld_mtx, in6_multihead_lock, inm6_lock, mli_lock
160  *
161  * Any may be taken independently, but if any are held at the same time,
162  * the above lock order must be followed.
163  */
164 static LCK_MTX_DECLARE_ATTR(mld_mtx, &mld_mtx_grp, &mld_mtx_attr);
165 
166 SLIST_HEAD(mld_in6m_relhead, in6_multi);
167 
168 static void     mli_initvar(struct mld_ifinfo *, struct ifnet *, int);
169 static struct mld_ifinfo *mli_alloc(zalloc_flags_t);
170 static void     mli_free(struct mld_ifinfo *);
171 static void     mli_delete(const struct ifnet *, struct mld_in6m_relhead *);
172 static void     mld_dispatch_packet(struct mbuf *);
173 static void     mld_final_leave(struct in6_multi *, struct mld_ifinfo *,
174     struct mld_tparams *);
175 static int      mld_handle_state_change(struct in6_multi *, struct mld_ifinfo *,
176     struct mld_tparams *);
177 static int      mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
178     struct mld_tparams *, const int);
179 #ifdef MLD_DEBUG
180 static const char *     mld_rec_type_to_str(const int);
181 #endif
182 static uint32_t mld_set_version(struct mld_ifinfo *, const int);
183 static void     mld_flush_relq(struct mld_ifinfo *, struct mld_in6m_relhead *);
184 static void     mld_dispatch_queue_locked(struct mld_ifinfo *, struct ifqueue *, int);
185 static int      mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
186     /*const*/ struct mld_hdr *);
187 static int      mld_v1_input_report(struct ifnet *, struct mbuf *,
188     const struct ip6_hdr *, /*const*/ struct mld_hdr *);
189 static void     mld_v1_process_group_timer(struct in6_multi *, const int);
190 static void     mld_v1_process_querier_timers(struct mld_ifinfo *);
191 static int      mld_v1_transmit_report(struct in6_multi *, const uint8_t);
192 static uint32_t mld_v1_update_group(struct in6_multi *, const int);
193 static void     mld_v2_cancel_link_timers(struct mld_ifinfo *);
194 static uint32_t mld_v2_dispatch_general_query(struct mld_ifinfo *);
195 static struct mbuf *
196 mld_v2_encap_report(struct ifnet *, struct mbuf *);
197 static int      mld_v2_enqueue_filter_change(struct ifqueue *,
198     struct in6_multi *);
199 static int      mld_v2_enqueue_group_record(struct ifqueue *,
200     struct in6_multi *, const int, const int, const int,
201     const int);
202 static int      mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
203     struct mbuf *, const int, const int);
204 static int      mld_v2_merge_state_changes(struct in6_multi *,
205     struct ifqueue *);
206 static void     mld_v2_process_group_timers(struct mld_ifinfo *,
207     struct ifqueue *, struct ifqueue *,
208     struct in6_multi *, const int);
209 static int      mld_v2_process_group_query(struct in6_multi *,
210     int, struct mbuf *, const int);
211 static int      sysctl_mld_gsr SYSCTL_HANDLER_ARGS;
212 static int      sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS;
213 static int      sysctl_mld_v2enable SYSCTL_HANDLER_ARGS;
214 
215 static int mld_timeout_run;             /* MLD timer is scheduled to run */
216 static void mld_timeout(void *);
217 static void mld_sched_timeout(bool);
218 
219 /*
220  * Normative references: RFC 2710, RFC 3590, RFC 3810.
221  */
222 static struct timeval mld_gsrdelay = {.tv_sec = 10, .tv_usec = 0};
223 static LIST_HEAD(, mld_ifinfo) mli_head;
224 
225 static int querier_present_timers_running6;
226 static int interface_timers_running6;
227 static int state_change_timers_running6;
228 static int current_state_timers_running6;
229 
230 static unsigned int mld_mli_list_genid;
231 /*
232  * Subsystem lock macros.
233  */
234 #define MLD_LOCK()                      \
235 	lck_mtx_lock(&mld_mtx)
236 #define MLD_LOCK_ASSERT_HELD()          \
237 	LCK_MTX_ASSERT(&mld_mtx, LCK_MTX_ASSERT_OWNED)
238 #define MLD_LOCK_ASSERT_NOTHELD()       \
239 	LCK_MTX_ASSERT(&mld_mtx, LCK_MTX_ASSERT_NOTOWNED)
240 #define MLD_UNLOCK()                    \
241 	lck_mtx_unlock(&mld_mtx)
242 
243 #define MLD_ADD_DETACHED_IN6M(_head, _in6m) {                           \
244 	SLIST_INSERT_HEAD(_head, _in6m, in6m_dtle);                     \
245 }
246 
247 #define MLD_REMOVE_DETACHED_IN6M(_head) {                               \
248 	struct in6_multi *_in6m, *_inm_tmp;                             \
249 	SLIST_FOREACH_SAFE(_in6m, _head, in6m_dtle, _inm_tmp) {         \
250 	        SLIST_REMOVE(_head, _in6m, in6_multi, in6m_dtle);       \
251 	        IN6M_REMREF(_in6m);                                     \
252 	}                                                               \
253 	VERIFY(SLIST_EMPTY(_head));                                     \
254 }
255 
256 static ZONE_DEFINE(mli_zone, "mld_ifinfo",
257     sizeof(struct mld_ifinfo), ZC_ZFREE_CLEARMEM);
258 
259 SYSCTL_DECL(_net_inet6);        /* Note: Not in any common header. */
260 
261 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_LOCKED, 0,
262     "IPv6 Multicast Listener Discovery");
263 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
264     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
265     &mld_gsrdelay.tv_sec, 0, sysctl_mld_gsr, "I",
266     "Rate limit for MLDv2 Group-and-Source queries in seconds");
267 
268 SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_LOCKED,
269     sysctl_mld_ifinfo, "Per-interface MLDv2 state");
270 
271 static int      mld_v1enable = 1;
272 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW | CTLFLAG_LOCKED,
273     &mld_v1enable, 0, "Enable fallback to MLDv1");
274 
275 static int      mld_v2enable = 1;
276 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, v2enable,
277     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
278     &mld_v2enable, 0, sysctl_mld_v2enable, "I",
279     "Enable MLDv2 (debug purposes only)");
280 
281 static int      mld_use_allow = 1;
282 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW | CTLFLAG_LOCKED,
283     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
284 
285 #ifdef MLD_DEBUG
286 int mld_debug = 0;
287 SYSCTL_INT(_net_inet6_mld, OID_AUTO,
288     debug, CTLFLAG_RW | CTLFLAG_LOCKED, &mld_debug, 0, "");
289 #endif
290 /*
291  * Packed Router Alert option structure declaration.
292  */
293 struct mld_raopt {
294 	struct ip6_hbh          hbh;
295 	struct ip6_opt          pad;
296 	struct ip6_opt_router   ra;
297 } __packed;
298 
299 /*
300  * Router Alert hop-by-hop option header.
301  */
302 static struct mld_raopt mld_ra = {
303 	.hbh = { .ip6h_nxt = 0, .ip6h_len = 0 },
304 	.pad = { .ip6o_type = IP6OPT_PADN, .ip6o_len = 0 },
305 	.ra = {
306 		.ip6or_type = (u_int8_t)IP6OPT_ROUTER_ALERT,
307 		.ip6or_len = (u_int8_t)(IP6OPT_RTALERT_LEN - 2),
308 		.ip6or_value =  {((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
309 			         (IP6OPT_RTALERT_MLD & 0xFF) }
310 	}
311 };
312 static struct ip6_pktopts mld_po;
313 
314 /* Store MLDv2 record count in the module private scratch space */
315 #define vt_nrecs        pkt_mpriv.__mpriv_u.__mpriv32[0].__mpriv32_u.__val16[0]
316 
317 static __inline void
mld_save_context(struct mbuf * m,struct ifnet * ifp)318 mld_save_context(struct mbuf *m, struct ifnet *ifp)
319 {
320 	m->m_pkthdr.rcvif = ifp;
321 }
322 
323 static __inline void
mld_scrub_context(struct mbuf * m)324 mld_scrub_context(struct mbuf *m)
325 {
326 	m->m_pkthdr.rcvif = NULL;
327 }
328 
329 /*
330  * Restore context from a queued output chain.
331  * Return saved ifp.
332  */
333 static __inline struct ifnet *
mld_restore_context(struct mbuf * m)334 mld_restore_context(struct mbuf *m)
335 {
336 	return m->m_pkthdr.rcvif;
337 }
338 
339 /*
340  * Retrieve or set threshold between group-source queries in seconds.
341  */
342 static int
343 sysctl_mld_gsr SYSCTL_HANDLER_ARGS
344 {
345 #pragma unused(arg1, arg2)
346 	int error;
347 	int i;
348 
349 	MLD_LOCK();
350 
351 	i = (int)mld_gsrdelay.tv_sec;
352 
353 	error = sysctl_handle_int(oidp, &i, 0, req);
354 	if (error || !req->newptr) {
355 		goto out_locked;
356 	}
357 
358 	if (i < -1 || i >= 60) {
359 		error = EINVAL;
360 		goto out_locked;
361 	}
362 
363 	mld_gsrdelay.tv_sec = i;
364 
365 out_locked:
366 	MLD_UNLOCK();
367 	return error;
368 }
369 /*
370  * Expose struct mld_ifinfo to userland, keyed by ifindex.
371  * For use by ifmcstat(8).
372  *
373  */
374 static int
375 sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS
376 {
377 #pragma unused(oidp)
378 	int                     *name;
379 	int                      error;
380 	u_int                    namelen;
381 	struct ifnet            *ifp;
382 	struct mld_ifinfo       *mli;
383 	struct mld_ifinfo_u     mli_u;
384 
385 	name = (int *)arg1;
386 	namelen = arg2;
387 
388 	if (req->newptr != USER_ADDR_NULL) {
389 		return EPERM;
390 	}
391 
392 	if (namelen != 1) {
393 		return EINVAL;
394 	}
395 
396 	MLD_LOCK();
397 
398 	if (name[0] <= 0 || name[0] > (u_int)if_index) {
399 		error = ENOENT;
400 		goto out_locked;
401 	}
402 
403 	error = ENOENT;
404 
405 	ifnet_head_lock_shared();
406 	ifp = ifindex2ifnet[name[0]];
407 	ifnet_head_done();
408 	if (ifp == NULL) {
409 		goto out_locked;
410 	}
411 
412 	bzero(&mli_u, sizeof(mli_u));
413 
414 	LIST_FOREACH(mli, &mli_head, mli_link) {
415 		MLI_LOCK(mli);
416 		if (ifp != mli->mli_ifp) {
417 			MLI_UNLOCK(mli);
418 			continue;
419 		}
420 
421 		mli_u.mli_ifindex = mli->mli_ifp->if_index;
422 		mli_u.mli_version = mli->mli_version;
423 		mli_u.mli_v1_timer = mli->mli_v1_timer;
424 		mli_u.mli_v2_timer = mli->mli_v2_timer;
425 		mli_u.mli_flags = mli->mli_flags;
426 		mli_u.mli_rv = mli->mli_rv;
427 		mli_u.mli_qi = mli->mli_qi;
428 		mli_u.mli_qri = mli->mli_qri;
429 		mli_u.mli_uri = mli->mli_uri;
430 		MLI_UNLOCK(mli);
431 
432 		error = SYSCTL_OUT(req, &mli_u, sizeof(mli_u));
433 		break;
434 	}
435 
436 out_locked:
437 	MLD_UNLOCK();
438 	return error;
439 }
440 
441 static int
442 sysctl_mld_v2enable SYSCTL_HANDLER_ARGS
443 {
444 #pragma unused(arg1, arg2)
445 	int error;
446 	int i;
447 	struct mld_ifinfo *mli;
448 	struct mld_tparams mtp = { .qpt = 0, .it = 0, .cst = 0, .sct = 0 };
449 
450 	MLD_LOCK();
451 
452 	i = mld_v2enable;
453 
454 	error = sysctl_handle_int(oidp, &i, 0, req);
455 	if (error || !req->newptr) {
456 		goto out_locked;
457 	}
458 
459 	if (i < 0 || i > 1) {
460 		error = EINVAL;
461 		goto out_locked;
462 	}
463 
464 	mld_v2enable = i;
465 	/*
466 	 * If we enabled v2, the state transition will take care of upgrading
467 	 * the MLD version back to v2. Otherwise, we have to explicitly
468 	 * downgrade. Note that this functionality is to be used for debugging.
469 	 */
470 	if (mld_v2enable == 1) {
471 		goto out_locked;
472 	}
473 
474 	LIST_FOREACH(mli, &mli_head, mli_link) {
475 		MLI_LOCK(mli);
476 		if (mld_set_version(mli, MLD_VERSION_1) > 0) {
477 			mtp.qpt = 1;
478 		}
479 		MLI_UNLOCK(mli);
480 	}
481 
482 out_locked:
483 	MLD_UNLOCK();
484 
485 	mld_set_timeout(&mtp);
486 
487 	return error;
488 }
489 
490 /*
491  * Dispatch an entire queue of pending packet chains.
492  *
493  * Must not be called with in6m_lock held.
494  * XXX This routine unlocks MLD global lock and also mli locks.
495  * Make sure that the calling routine takes reference on the mli
496  * before calling this routine.
497  * Also if we are traversing mli_head, remember to check for
498  * mli list generation count and restart the loop if generation count
499  * has changed.
500  */
501 static void
mld_dispatch_queue_locked(struct mld_ifinfo * mli,struct ifqueue * ifq,int limit)502 mld_dispatch_queue_locked(struct mld_ifinfo *mli, struct ifqueue *ifq, int limit)
503 {
504 	struct mbuf *m;
505 
506 	MLD_LOCK_ASSERT_HELD();
507 
508 	if (mli != NULL) {
509 		MLI_LOCK_ASSERT_HELD(mli);
510 	}
511 
512 	for (;;) {
513 		IF_DEQUEUE(ifq, m);
514 		if (m == NULL) {
515 			break;
516 		}
517 		MLD_PRINTF(("%s: dispatch 0x%llx from 0x%llx\n", __func__,
518 		    (uint64_t)VM_KERNEL_ADDRPERM(ifq),
519 		    (uint64_t)VM_KERNEL_ADDRPERM(m)));
520 
521 		if (mli != NULL) {
522 			MLI_UNLOCK(mli);
523 		}
524 		MLD_UNLOCK();
525 
526 		mld_dispatch_packet(m);
527 
528 		MLD_LOCK();
529 		if (mli != NULL) {
530 			MLI_LOCK(mli);
531 		}
532 
533 		if (--limit == 0) {
534 			break;
535 		}
536 	}
537 
538 	if (mli != NULL) {
539 		MLI_LOCK_ASSERT_HELD(mli);
540 	}
541 }
542 
543 /*
544  * Filter outgoing MLD report state by group.
545  *
546  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
547  * and node-local addresses. However, kernel and socket consumers
548  * always embed the KAME scope ID in the address provided, so strip it
549  * when performing comparison.
550  * Note: This is not the same as the *multicast* scope.
551  *
552  * Return zero if the given group is one for which MLD reports
553  * should be suppressed, or non-zero if reports should be issued.
554  */
555 static __inline__ int
mld_is_addr_reported(const struct in6_addr * addr)556 mld_is_addr_reported(const struct in6_addr *addr)
557 {
558 	VERIFY(IN6_IS_ADDR_MULTICAST(addr));
559 
560 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) {
561 		return 0;
562 	}
563 
564 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL && !IN6_IS_ADDR_UNICAST_BASED_MULTICAST(addr)) {
565 		struct in6_addr tmp = *addr;
566 		in6_clearscope(&tmp);
567 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) {
568 			return 0;
569 		}
570 	}
571 
572 	return 1;
573 }
574 
575 /*
576  * Attach MLD when PF_INET6 is attached to an interface.
577  */
578 struct mld_ifinfo *
mld_domifattach(struct ifnet * ifp,zalloc_flags_t how)579 mld_domifattach(struct ifnet *ifp, zalloc_flags_t how)
580 {
581 	struct mld_ifinfo *mli;
582 
583 	MLD_PRINTF(("%s: called for ifp 0x%llx(%s)\n", __func__,
584 	    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
585 
586 	mli = mli_alloc(how);
587 	if (mli == NULL) {
588 		return NULL;
589 	}
590 
591 	MLD_LOCK();
592 
593 	MLI_LOCK(mli);
594 	mli_initvar(mli, ifp, 0);
595 	mli->mli_debug |= IFD_ATTACHED;
596 	MLI_ADDREF_LOCKED(mli); /* hold a reference for mli_head */
597 	MLI_ADDREF_LOCKED(mli); /* hold a reference for caller */
598 	MLI_UNLOCK(mli);
599 	ifnet_lock_shared(ifp);
600 	mld6_initsilent(ifp, mli);
601 	ifnet_lock_done(ifp);
602 
603 	LIST_INSERT_HEAD(&mli_head, mli, mli_link);
604 	mld_mli_list_genid++;
605 
606 	MLD_UNLOCK();
607 
608 	MLD_PRINTF(("%s: allocate mld_ifinfo for ifp 0x%llx(%s)\n",
609 	    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
610 
611 	return mli;
612 }
613 
614 /*
615  * Attach MLD when PF_INET6 is reattached to an interface.  Caller is
616  * expected to have an outstanding reference to the mli.
617  */
618 void
mld_domifreattach(struct mld_ifinfo * mli)619 mld_domifreattach(struct mld_ifinfo *mli)
620 {
621 	struct ifnet *ifp;
622 
623 	MLD_LOCK();
624 
625 	MLI_LOCK(mli);
626 	VERIFY(!(mli->mli_debug & IFD_ATTACHED));
627 	ifp = mli->mli_ifp;
628 	VERIFY(ifp != NULL);
629 	mli_initvar(mli, ifp, 1);
630 	mli->mli_debug |= IFD_ATTACHED;
631 	MLI_ADDREF_LOCKED(mli); /* hold a reference for mli_head */
632 	MLI_UNLOCK(mli);
633 	ifnet_lock_shared(ifp);
634 	mld6_initsilent(ifp, mli);
635 	ifnet_lock_done(ifp);
636 
637 	LIST_INSERT_HEAD(&mli_head, mli, mli_link);
638 	mld_mli_list_genid++;
639 
640 	MLD_UNLOCK();
641 
642 	MLD_PRINTF(("%s: reattached mld_ifinfo for ifp 0x%llx(%s)\n",
643 	    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
644 }
645 
646 /*
647  * Hook for domifdetach.
648  */
649 void
mld_domifdetach(struct ifnet * ifp)650 mld_domifdetach(struct ifnet *ifp)
651 {
652 	SLIST_HEAD(, in6_multi) in6m_dthead;
653 
654 	SLIST_INIT(&in6m_dthead);
655 
656 	MLD_PRINTF(("%s: called for ifp 0x%llx(%s)\n", __func__,
657 	    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
658 
659 	MLD_LOCK();
660 	mli_delete(ifp, (struct mld_in6m_relhead *)&in6m_dthead);
661 	MLD_UNLOCK();
662 
663 	/* Now that we're dropped all locks, release detached records */
664 	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
665 }
666 
667 /*
668  * Called at interface detach time.  Note that we only flush all deferred
669  * responses and record releases; all remaining inm records and their source
670  * entries related to this interface are left intact, in order to handle
671  * the reattach case.
672  */
673 static void
mli_delete(const struct ifnet * ifp,struct mld_in6m_relhead * in6m_dthead)674 mli_delete(const struct ifnet *ifp, struct mld_in6m_relhead *in6m_dthead)
675 {
676 	struct mld_ifinfo *mli, *tmli;
677 
678 	MLD_LOCK_ASSERT_HELD();
679 
680 	LIST_FOREACH_SAFE(mli, &mli_head, mli_link, tmli) {
681 		MLI_LOCK(mli);
682 		if (mli->mli_ifp == ifp) {
683 			/*
684 			 * Free deferred General Query responses.
685 			 */
686 			IF_DRAIN(&mli->mli_gq);
687 			IF_DRAIN(&mli->mli_v1q);
688 			mld_flush_relq(mli, in6m_dthead);
689 			VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
690 			mli->mli_debug &= ~IFD_ATTACHED;
691 			MLI_UNLOCK(mli);
692 
693 			LIST_REMOVE(mli, mli_link);
694 			MLI_REMREF(mli); /* release mli_head reference */
695 			mld_mli_list_genid++;
696 			return;
697 		}
698 		MLI_UNLOCK(mli);
699 	}
700 	panic("%s: mld_ifinfo not found for ifp %p(%s)", __func__,
701 	    ifp, ifp->if_xname);
702 }
703 
704 __private_extern__ void
mld6_initsilent(struct ifnet * ifp,struct mld_ifinfo * mli)705 mld6_initsilent(struct ifnet *ifp, struct mld_ifinfo *mli)
706 {
707 	ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_OWNED);
708 
709 	MLI_LOCK_ASSERT_NOTHELD(mli);
710 	MLI_LOCK(mli);
711 	if (!(ifp->if_flags & IFF_MULTICAST) &&
712 	    (ifp->if_eflags & (IFEF_IPV6_ND6ALT | IFEF_LOCALNET_PRIVATE))) {
713 		mli->mli_flags |= MLIF_SILENT;
714 	} else {
715 		mli->mli_flags &= ~MLIF_SILENT;
716 	}
717 	MLI_UNLOCK(mli);
718 }
719 
720 static void
mli_initvar(struct mld_ifinfo * mli,struct ifnet * ifp,int reattach)721 mli_initvar(struct mld_ifinfo *mli, struct ifnet *ifp, int reattach)
722 {
723 	MLI_LOCK_ASSERT_HELD(mli);
724 
725 	mli->mli_ifp = ifp;
726 	if (mld_v2enable) {
727 		mli->mli_version = MLD_VERSION_2;
728 	} else {
729 		mli->mli_version = MLD_VERSION_1;
730 	}
731 	mli->mli_flags = 0;
732 	mli->mli_rv = MLD_RV_INIT;
733 	mli->mli_qi = MLD_QI_INIT;
734 	mli->mli_qri = MLD_QRI_INIT;
735 	mli->mli_uri = MLD_URI_INIT;
736 
737 	if (mld_use_allow) {
738 		mli->mli_flags |= MLIF_USEALLOW;
739 	}
740 	if (!reattach) {
741 		SLIST_INIT(&mli->mli_relinmhead);
742 	}
743 
744 	/*
745 	 * Responses to general queries are subject to bounds.
746 	 */
747 	mli->mli_gq.ifq_maxlen = MLD_MAX_RESPONSE_PACKETS;
748 	mli->mli_v1q.ifq_maxlen = MLD_MAX_RESPONSE_PACKETS;
749 }
750 
751 static struct mld_ifinfo *
mli_alloc(zalloc_flags_t how)752 mli_alloc(zalloc_flags_t how)
753 {
754 	struct mld_ifinfo *mli = zalloc_flags(mli_zone, how | Z_ZERO);
755 	if (mli != NULL) {
756 		lck_mtx_init(&mli->mli_lock, &mld_mtx_grp, &mld_mtx_attr);
757 		mli->mli_debug |= IFD_ALLOC;
758 	}
759 	return mli;
760 }
761 
762 static void
mli_free(struct mld_ifinfo * mli)763 mli_free(struct mld_ifinfo *mli)
764 {
765 	MLI_LOCK(mli);
766 	if (mli->mli_debug & IFD_ATTACHED) {
767 		panic("%s: attached mli=%p is being freed", __func__, mli);
768 		/* NOTREACHED */
769 	} else if (mli->mli_ifp != NULL) {
770 		panic("%s: ifp not NULL for mli=%p", __func__, mli);
771 		/* NOTREACHED */
772 	} else if (!(mli->mli_debug & IFD_ALLOC)) {
773 		panic("%s: mli %p cannot be freed", __func__, mli);
774 		/* NOTREACHED */
775 	} else if (mli->mli_refcnt != 0) {
776 		panic("%s: non-zero refcnt mli=%p", __func__, mli);
777 		/* NOTREACHED */
778 	}
779 	mli->mli_debug &= ~IFD_ALLOC;
780 	MLI_UNLOCK(mli);
781 
782 	lck_mtx_destroy(&mli->mli_lock, &mld_mtx_grp);
783 	zfree(mli_zone, mli);
784 }
785 
786 void
mli_addref(struct mld_ifinfo * mli,int locked)787 mli_addref(struct mld_ifinfo *mli, int locked)
788 {
789 	if (!locked) {
790 		MLI_LOCK_SPIN(mli);
791 	} else {
792 		MLI_LOCK_ASSERT_HELD(mli);
793 	}
794 
795 	if (++mli->mli_refcnt == 0) {
796 		panic("%s: mli=%p wraparound refcnt", __func__, mli);
797 		/* NOTREACHED */
798 	}
799 	if (!locked) {
800 		MLI_UNLOCK(mli);
801 	}
802 }
803 
804 void
mli_remref(struct mld_ifinfo * mli)805 mli_remref(struct mld_ifinfo *mli)
806 {
807 	SLIST_HEAD(, in6_multi) in6m_dthead;
808 	struct ifnet *ifp;
809 
810 	MLI_LOCK_SPIN(mli);
811 
812 	if (mli->mli_refcnt == 0) {
813 		panic("%s: mli=%p negative refcnt", __func__, mli);
814 		/* NOTREACHED */
815 	}
816 
817 	--mli->mli_refcnt;
818 	if (mli->mli_refcnt > 0) {
819 		MLI_UNLOCK(mli);
820 		return;
821 	}
822 
823 	ifp = mli->mli_ifp;
824 	mli->mli_ifp = NULL;
825 	IF_DRAIN(&mli->mli_gq);
826 	IF_DRAIN(&mli->mli_v1q);
827 	SLIST_INIT(&in6m_dthead);
828 	mld_flush_relq(mli, (struct mld_in6m_relhead *)&in6m_dthead);
829 	VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
830 	MLI_UNLOCK(mli);
831 
832 	/* Now that we're dropped all locks, release detached records */
833 	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
834 
835 	MLD_PRINTF(("%s: freeing mld_ifinfo for ifp 0x%llx(%s)\n",
836 	    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
837 
838 	mli_free(mli);
839 }
840 
841 /*
842  * Process a received MLDv1 general or address-specific query.
843  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
844  *
845  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
846  * mld_addr. This is OK as we own the mbuf chain.
847  */
848 static int
mld_v1_input_query(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mld_hdr * mld)849 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
850     /*const*/ struct mld_hdr *mld)
851 {
852 	struct mld_ifinfo       *mli;
853 	struct in6_multi        *inm;
854 	int                      err = 0, is_general_query;
855 	uint16_t                 timer;
856 	struct mld_tparams       mtp = { .qpt = 0, .it = 0, .cst = 0, .sct = 0 };
857 
858 	MLD_LOCK_ASSERT_NOTHELD();
859 
860 	is_general_query = 0;
861 
862 	if (!mld_v1enable) {
863 		MLD_PRINTF(("%s: ignore v1 query %s on ifp 0x%llx(%s)\n",
864 		    __func__, ip6_sprintf(&mld->mld_addr),
865 		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
866 		goto done;
867 	}
868 
869 	/*
870 	 * RFC3810 Section 6.2: MLD queries must originate from
871 	 * a router's link-local address.
872 	 */
873 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
874 		MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
875 		    __func__, ip6_sprintf(&ip6->ip6_src),
876 		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
877 		goto done;
878 	}
879 
880 	/*
881 	 * Do address field validation upfront before we accept
882 	 * the query.
883 	 */
884 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
885 		/*
886 		 * MLDv1 General Query.
887 		 * If this was not sent to the all-nodes group, ignore it.
888 		 */
889 		struct in6_addr          dst;
890 
891 		dst = ip6->ip6_dst;
892 		in6_clearscope(&dst);
893 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) {
894 			err = EINVAL;
895 			goto done;
896 		}
897 		is_general_query = 1;
898 	} else {
899 		/*
900 		 * Embed scope ID of receiving interface in MLD query for
901 		 * lookup whilst we don't hold other locks.
902 		 */
903 		(void)in6_setscope(&mld->mld_addr, ifp, NULL);
904 	}
905 
906 	/*
907 	 * Switch to MLDv1 host compatibility mode.
908 	 */
909 	mli = MLD_IFINFO(ifp);
910 	VERIFY(mli != NULL);
911 
912 	MLI_LOCK(mli);
913 	mtp.qpt = mld_set_version(mli, MLD_VERSION_1);
914 	MLI_UNLOCK(mli);
915 
916 	timer = ntohs(mld->mld_maxdelay) / MLD_TIMER_SCALE;
917 	if (timer == 0) {
918 		timer = 1;
919 	}
920 
921 	if (is_general_query) {
922 		struct in6_multistep step;
923 
924 		MLD_PRINTF(("%s: process v1 general query on ifp 0x%llx(%s)\n",
925 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
926 		/*
927 		 * For each reporting group joined on this
928 		 * interface, kick the report timer.
929 		 */
930 		in6_multihead_lock_shared();
931 		IN6_FIRST_MULTI(step, inm);
932 		while (inm != NULL) {
933 			IN6M_LOCK(inm);
934 			if (inm->in6m_ifp == ifp) {
935 				mtp.cst += mld_v1_update_group(inm, timer);
936 			}
937 			IN6M_UNLOCK(inm);
938 			IN6_NEXT_MULTI(step, inm);
939 		}
940 		in6_multihead_lock_done();
941 	} else {
942 		/*
943 		 * MLDv1 Group-Specific Query.
944 		 * If this is a group-specific MLDv1 query, we need only
945 		 * look up the single group to process it.
946 		 */
947 		in6_multihead_lock_shared();
948 		IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
949 		in6_multihead_lock_done();
950 
951 		if (inm != NULL) {
952 			IN6M_LOCK(inm);
953 			MLD_PRINTF(("%s: process v1 query %s on "
954 			    "ifp 0x%llx(%s)\n", __func__,
955 			    ip6_sprintf(&mld->mld_addr),
956 			    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
957 			mtp.cst = mld_v1_update_group(inm, timer);
958 			IN6M_UNLOCK(inm);
959 			IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
960 		}
961 		/* XXX Clear embedded scope ID as userland won't expect it. */
962 		in6_clearscope(&mld->mld_addr);
963 	}
964 done:
965 	mld_set_timeout(&mtp);
966 
967 	return err;
968 }
969 
970 /*
971  * Update the report timer on a group in response to an MLDv1 query.
972  *
973  * If we are becoming the reporting member for this group, start the timer.
974  * If we already are the reporting member for this group, and timer is
975  * below the threshold, reset it.
976  *
977  * We may be updating the group for the first time since we switched
978  * to MLDv2. If we are, then we must clear any recorded source lists,
979  * and transition to REPORTING state; the group timer is overloaded
980  * for group and group-source query responses.
981  *
982  * Unlike MLDv2, the delay per group should be jittered
983  * to avoid bursts of MLDv1 reports.
984  */
985 static uint32_t
mld_v1_update_group(struct in6_multi * inm,const int timer)986 mld_v1_update_group(struct in6_multi *inm, const int timer)
987 {
988 	IN6M_LOCK_ASSERT_HELD(inm);
989 
990 	MLD_PRINTF(("%s: %s/%s timer=%d\n", __func__,
991 	    ip6_sprintf(&inm->in6m_addr),
992 	    if_name(inm->in6m_ifp), timer));
993 
994 	switch (inm->in6m_state) {
995 	case MLD_NOT_MEMBER:
996 	case MLD_SILENT_MEMBER:
997 		break;
998 	case MLD_REPORTING_MEMBER:
999 		if (inm->in6m_timer != 0 &&
1000 		    inm->in6m_timer <= timer) {
1001 			MLD_PRINTF(("%s: REPORTING and timer running, "
1002 			    "skipping.\n", __func__));
1003 			break;
1004 		}
1005 		OS_FALLTHROUGH;
1006 	case MLD_SG_QUERY_PENDING_MEMBER:
1007 	case MLD_G_QUERY_PENDING_MEMBER:
1008 	case MLD_IDLE_MEMBER:
1009 	case MLD_LAZY_MEMBER:
1010 	case MLD_AWAKENING_MEMBER:
1011 		MLD_PRINTF(("%s: ->REPORTING\n", __func__));
1012 		inm->in6m_state = MLD_REPORTING_MEMBER;
1013 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1014 		break;
1015 	case MLD_SLEEPING_MEMBER:
1016 		MLD_PRINTF(("%s: ->AWAKENING\n", __func__));
1017 		inm->in6m_state = MLD_AWAKENING_MEMBER;
1018 		break;
1019 	case MLD_LEAVING_MEMBER:
1020 		break;
1021 	}
1022 
1023 	return inm->in6m_timer;
1024 }
1025 
1026 /*
1027  * Process a received MLDv2 general, group-specific or
1028  * group-and-source-specific query.
1029  *
1030  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
1031  *
1032  * Return 0 if successful, otherwise an appropriate error code is returned.
1033  */
1034 static int
mld_v2_input_query(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mbuf * m,const int off,const int icmp6len)1035 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
1036     struct mbuf *m, const int off, const int icmp6len)
1037 {
1038 	struct mld_ifinfo       *mli;
1039 	struct mldv2_query      *mld;
1040 	struct in6_multi        *inm;
1041 	uint32_t                 maxdelay, nsrc, qqi, timer;
1042 	int                      err = 0, is_general_query;
1043 	uint8_t                  qrv;
1044 	struct mld_tparams       mtp = { .qpt = 0, .it = 0, .cst = 0, .sct = 0 };
1045 
1046 	MLD_LOCK_ASSERT_NOTHELD();
1047 
1048 	is_general_query = 0;
1049 
1050 	if (!mld_v2enable) {
1051 		MLD_PRINTF(("%s: ignore v2 query %s on ifp 0x%llx(%s)\n",
1052 		    __func__, ip6_sprintf(&ip6->ip6_src),
1053 		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1054 		goto done;
1055 	}
1056 
1057 	/*
1058 	 * RFC3810 Section 6.2: MLD queries must originate from
1059 	 * a router's link-local address.
1060 	 */
1061 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
1062 		MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
1063 		    __func__, ip6_sprintf(&ip6->ip6_src),
1064 		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1065 		goto done;
1066 	}
1067 
1068 	MLD_PRINTF(("%s: input v2 query on ifp 0x%llx(%s)\n", __func__,
1069 	    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1070 
1071 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
1072 
1073 	maxdelay = ntohs(mld->mld_maxdelay);    /* in 1/10ths of a second */
1074 	if (maxdelay > SHRT_MAX) {
1075 		maxdelay = (MLD_MRC_MANT((uint16_t)maxdelay) | 0x1000) <<
1076 		    (MLD_MRC_EXP((uint16_t)maxdelay) + 3);
1077 	}
1078 	timer = maxdelay / MLD_TIMER_SCALE;
1079 	if (timer == 0) {
1080 		timer = 1;
1081 	}
1082 
1083 	qrv = MLD_QRV(mld->mld_misc);
1084 	if (qrv < 2) {
1085 		MLD_PRINTF(("%s: clamping qrv %d to %d\n", __func__,
1086 		    qrv, MLD_RV_INIT));
1087 		qrv = MLD_RV_INIT;
1088 	}
1089 
1090 	qqi = mld->mld_qqi;
1091 	if (qqi >= 128) {
1092 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
1093 		    (MLD_QQIC_EXP(mld->mld_qqi) + 3);
1094 	}
1095 
1096 	nsrc = ntohs(mld->mld_numsrc);
1097 	if (nsrc > MLD_MAX_GS_SOURCES) {
1098 		err = EMSGSIZE;
1099 		goto done;
1100 	}
1101 	if (icmp6len < sizeof(struct mldv2_query) +
1102 	    (nsrc * sizeof(struct in6_addr))) {
1103 		err = EMSGSIZE;
1104 		goto done;
1105 	}
1106 
1107 	/*
1108 	 * Do further input validation upfront to avoid resetting timers
1109 	 * should we need to discard this query.
1110 	 */
1111 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
1112 		/*
1113 		 * A general query with a source list has undefined
1114 		 * behaviour; discard it.
1115 		 */
1116 		if (nsrc > 0) {
1117 			err = EINVAL;
1118 			goto done;
1119 		}
1120 		is_general_query = 1;
1121 	} else {
1122 		/*
1123 		 * Embed scope ID of receiving interface in MLD query for
1124 		 * lookup whilst we don't hold other locks (due to KAME
1125 		 * locking lameness). We own this mbuf chain just now.
1126 		 */
1127 		(void)in6_setscope(&mld->mld_addr, ifp, NULL);
1128 	}
1129 
1130 	mli = MLD_IFINFO(ifp);
1131 	VERIFY(mli != NULL);
1132 
1133 	MLI_LOCK(mli);
1134 	/*
1135 	 * Discard the v2 query if we're in Compatibility Mode.
1136 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
1137 	 * until the Old Version Querier Present timer expires.
1138 	 */
1139 	if (mli->mli_version != MLD_VERSION_2) {
1140 		MLI_UNLOCK(mli);
1141 		goto done;
1142 	}
1143 
1144 	mtp.qpt = mld_set_version(mli, MLD_VERSION_2);
1145 	mli->mli_rv = qrv;
1146 	mli->mli_qi = qqi;
1147 	mli->mli_qri = MAX(timer, MLD_QRI_MIN);
1148 
1149 	MLD_PRINTF(("%s: qrv %d qi %d qri %d\n", __func__, mli->mli_rv,
1150 	    mli->mli_qi, mli->mli_qri));
1151 
1152 	if (is_general_query) {
1153 		/*
1154 		 * MLDv2 General Query.
1155 		 *
1156 		 * Schedule a current-state report on this ifp for
1157 		 * all groups, possibly containing source lists.
1158 		 *
1159 		 * If there is a pending General Query response
1160 		 * scheduled earlier than the selected delay, do
1161 		 * not schedule any other reports.
1162 		 * Otherwise, reset the interface timer.
1163 		 */
1164 		MLD_PRINTF(("%s: process v2 general query on ifp 0x%llx(%s)\n",
1165 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1166 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
1167 			mtp.it = mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
1168 		}
1169 		MLI_UNLOCK(mli);
1170 	} else {
1171 		MLI_UNLOCK(mli);
1172 		/*
1173 		 * MLDv2 Group-specific or Group-and-source-specific Query.
1174 		 *
1175 		 * Group-source-specific queries are throttled on
1176 		 * a per-group basis to defeat denial-of-service attempts.
1177 		 * Queries for groups we are not a member of on this
1178 		 * link are simply ignored.
1179 		 */
1180 		in6_multihead_lock_shared();
1181 		IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
1182 		in6_multihead_lock_done();
1183 		if (inm == NULL) {
1184 			goto done;
1185 		}
1186 
1187 		IN6M_LOCK(inm);
1188 		if (nsrc > 0) {
1189 			if (!ratecheck(&inm->in6m_lastgsrtv,
1190 			    &mld_gsrdelay)) {
1191 				MLD_PRINTF(("%s: GS query throttled.\n",
1192 				    __func__));
1193 				IN6M_UNLOCK(inm);
1194 				IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1195 				goto done;
1196 			}
1197 		}
1198 		MLD_PRINTF(("%s: process v2 group query on ifp 0x%llx(%s)\n",
1199 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1200 		/*
1201 		 * If there is a pending General Query response
1202 		 * scheduled sooner than the selected delay, no
1203 		 * further report need be scheduled.
1204 		 * Otherwise, prepare to respond to the
1205 		 * group-specific or group-and-source query.
1206 		 */
1207 		MLI_LOCK(mli);
1208 		mtp.it = mli->mli_v2_timer;
1209 		MLI_UNLOCK(mli);
1210 		if (mtp.it == 0 || mtp.it >= timer) {
1211 			(void) mld_v2_process_group_query(inm, timer, m, off);
1212 			mtp.cst = inm->in6m_timer;
1213 		}
1214 		IN6M_UNLOCK(inm);
1215 		IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1216 		/* XXX Clear embedded scope ID as userland won't expect it. */
1217 		in6_clearscope(&mld->mld_addr);
1218 	}
1219 done:
1220 	if (mtp.it > 0) {
1221 		MLD_PRINTF(("%s: v2 general query response scheduled in "
1222 		    "T+%d seconds on ifp 0x%llx(%s)\n", __func__, mtp.it,
1223 		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1224 	}
1225 	mld_set_timeout(&mtp);
1226 
1227 	return err;
1228 }
1229 
1230 /*
1231  * Process a recieved MLDv2 group-specific or group-and-source-specific
1232  * query.
1233  * Return <0 if any error occured. Currently this is ignored.
1234  */
1235 static int
mld_v2_process_group_query(struct in6_multi * inm,int timer,struct mbuf * m0,const int off)1236 mld_v2_process_group_query(struct in6_multi *inm, int timer, struct mbuf *m0,
1237     const int off)
1238 {
1239 	struct mldv2_query      *mld;
1240 	int                      retval;
1241 	uint16_t                 nsrc;
1242 
1243 	IN6M_LOCK_ASSERT_HELD(inm);
1244 
1245 	retval = 0;
1246 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
1247 
1248 	switch (inm->in6m_state) {
1249 	case MLD_NOT_MEMBER:
1250 	case MLD_SILENT_MEMBER:
1251 	case MLD_SLEEPING_MEMBER:
1252 	case MLD_LAZY_MEMBER:
1253 	case MLD_AWAKENING_MEMBER:
1254 	case MLD_IDLE_MEMBER:
1255 	case MLD_LEAVING_MEMBER:
1256 		return retval;
1257 	case MLD_REPORTING_MEMBER:
1258 	case MLD_G_QUERY_PENDING_MEMBER:
1259 	case MLD_SG_QUERY_PENDING_MEMBER:
1260 		break;
1261 	}
1262 
1263 	nsrc = ntohs(mld->mld_numsrc);
1264 
1265 	/*
1266 	 * Deal with group-specific queries upfront.
1267 	 * If any group query is already pending, purge any recorded
1268 	 * source-list state if it exists, and schedule a query response
1269 	 * for this group-specific query.
1270 	 */
1271 	if (nsrc == 0) {
1272 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1273 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1274 			in6m_clear_recorded(inm);
1275 			timer = min(inm->in6m_timer, timer);
1276 		}
1277 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1278 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1279 		return retval;
1280 	}
1281 
1282 	/*
1283 	 * Deal with the case where a group-and-source-specific query has
1284 	 * been received but a group-specific query is already pending.
1285 	 */
1286 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1287 		timer = min(inm->in6m_timer, timer);
1288 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1289 		return retval;
1290 	}
1291 
1292 	/*
1293 	 * Finally, deal with the case where a group-and-source-specific
1294 	 * query has been received, where a response to a previous g-s-r
1295 	 * query exists, or none exists.
1296 	 * In this case, we need to parse the source-list which the Querier
1297 	 * has provided us with and check if we have any source list filter
1298 	 * entries at T1 for these sources. If we do not, there is no need
1299 	 * schedule a report and the query may be dropped.
1300 	 * If we do, we must record them and schedule a current-state
1301 	 * report for those sources.
1302 	 */
1303 	if (inm->in6m_nsrc > 0) {
1304 		struct mbuf             *m;
1305 		struct in6_addr          addr;
1306 		int                      i, nrecorded;
1307 		int                      soff;
1308 
1309 		m = m0;
1310 		soff = off + sizeof(struct mldv2_query);
1311 		nrecorded = 0;
1312 		for (i = 0; i < nsrc; i++) {
1313 			m_copydata(m, soff, sizeof(addr), &addr);
1314 			retval = in6m_record_source(inm, &addr);
1315 			if (retval < 0) {
1316 				break;
1317 			}
1318 			nrecorded += retval;
1319 			soff += sizeof(struct in6_addr);
1320 
1321 			while (m && (soff >= m->m_len)) {
1322 				soff -= m->m_len;
1323 				m = m->m_next;
1324 			}
1325 
1326 			/* should not be possible: */
1327 			if (m == NULL) {
1328 				break;
1329 			}
1330 		}
1331 		if (nrecorded > 0) {
1332 			MLD_PRINTF(("%s: schedule response to SG query\n",
1333 			    __func__));
1334 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1335 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1336 		}
1337 	}
1338 
1339 	return retval;
1340 }
1341 
1342 /*
1343  * Process a received MLDv1 host membership report.
1344  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1345  *
1346  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1347  * mld_addr. This is OK as we own the mbuf chain.
1348  */
1349 static int
mld_v1_input_report(struct ifnet * ifp,struct mbuf * m,const struct ip6_hdr * ip6,struct mld_hdr * mld)1350 mld_v1_input_report(struct ifnet *ifp, struct mbuf *m,
1351     const struct ip6_hdr *ip6, /*const*/ struct mld_hdr *mld)
1352 {
1353 	struct in6_addr          src, dst;
1354 	struct in6_ifaddr       *ia;
1355 	struct in6_multi        *inm;
1356 
1357 	if (!mld_v1enable) {
1358 		MLD_PRINTF(("%s: ignore v1 report %s on ifp 0x%llx(%s)\n",
1359 		    __func__, ip6_sprintf(&mld->mld_addr),
1360 		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1361 		return 0;
1362 	}
1363 
1364 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1365 	    (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1366 		return 0;
1367 	}
1368 
1369 	/*
1370 	 * MLDv1 reports must originate from a host's link-local address,
1371 	 * or the unspecified address (when booting).
1372 	 */
1373 	src = ip6->ip6_src;
1374 	in6_clearscope(&src);
1375 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1376 		MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
1377 		    __func__, ip6_sprintf(&ip6->ip6_src),
1378 		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1379 		return EINVAL;
1380 	}
1381 
1382 	/*
1383 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1384 	 * group, and must be directed to the group itself.
1385 	 */
1386 	dst = ip6->ip6_dst;
1387 	in6_clearscope(&dst);
1388 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1389 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1390 		MLD_PRINTF(("%s: ignore v1 query dst %s on ifp 0x%llx(%s)\n",
1391 		    __func__, ip6_sprintf(&ip6->ip6_dst),
1392 		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1393 		return EINVAL;
1394 	}
1395 
1396 	/*
1397 	 * Make sure we don't hear our own membership report, as fast
1398 	 * leave requires knowing that we are the only member of a
1399 	 * group. Assume we used the link-local address if available,
1400 	 * otherwise look for ::.
1401 	 *
1402 	 * XXX Note that scope ID comparison is needed for the address
1403 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1404 	 * performed for the on-wire address.
1405 	 */
1406 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
1407 	if (ia != NULL) {
1408 		IFA_LOCK(&ia->ia_ifa);
1409 		if ((IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia)))) {
1410 			IFA_UNLOCK(&ia->ia_ifa);
1411 			IFA_REMREF(&ia->ia_ifa);
1412 			return 0;
1413 		}
1414 		IFA_UNLOCK(&ia->ia_ifa);
1415 		IFA_REMREF(&ia->ia_ifa);
1416 	} else if (IN6_IS_ADDR_UNSPECIFIED(&src)) {
1417 		return 0;
1418 	}
1419 
1420 	MLD_PRINTF(("%s: process v1 report %s on ifp 0x%llx(%s)\n",
1421 	    __func__, ip6_sprintf(&mld->mld_addr),
1422 	    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1423 
1424 	/*
1425 	 * Embed scope ID of receiving interface in MLD query for lookup
1426 	 * whilst we don't hold other locks (due to KAME locking lameness).
1427 	 */
1428 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
1429 		(void)in6_setscope(&mld->mld_addr, ifp, NULL);
1430 	}
1431 
1432 	/*
1433 	 * MLDv1 report suppression.
1434 	 * If we are a member of this group, and our membership should be
1435 	 * reported, and our group timer is pending or about to be reset,
1436 	 * stop our group timer by transitioning to the 'lazy' state.
1437 	 */
1438 	in6_multihead_lock_shared();
1439 	IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
1440 	in6_multihead_lock_done();
1441 
1442 	if (inm != NULL) {
1443 		struct mld_ifinfo *mli;
1444 
1445 		IN6M_LOCK(inm);
1446 		mli = inm->in6m_mli;
1447 		VERIFY(mli != NULL);
1448 
1449 		MLI_LOCK(mli);
1450 		/*
1451 		 * If we are in MLDv2 host mode, do not allow the
1452 		 * other host's MLDv1 report to suppress our reports.
1453 		 */
1454 		if (mli->mli_version == MLD_VERSION_2) {
1455 			MLI_UNLOCK(mli);
1456 			IN6M_UNLOCK(inm);
1457 			IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1458 			goto out;
1459 		}
1460 		MLI_UNLOCK(mli);
1461 
1462 		inm->in6m_timer = 0;
1463 
1464 		switch (inm->in6m_state) {
1465 		case MLD_NOT_MEMBER:
1466 		case MLD_SILENT_MEMBER:
1467 		case MLD_SLEEPING_MEMBER:
1468 			break;
1469 		case MLD_REPORTING_MEMBER:
1470 		case MLD_IDLE_MEMBER:
1471 		case MLD_AWAKENING_MEMBER:
1472 			MLD_PRINTF(("%s: report suppressed for %s on "
1473 			    "ifp 0x%llx(%s)\n", __func__,
1474 			    ip6_sprintf(&mld->mld_addr),
1475 			    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1476 			OS_FALLTHROUGH;
1477 		case MLD_LAZY_MEMBER:
1478 			inm->in6m_state = MLD_LAZY_MEMBER;
1479 			break;
1480 		case MLD_G_QUERY_PENDING_MEMBER:
1481 		case MLD_SG_QUERY_PENDING_MEMBER:
1482 		case MLD_LEAVING_MEMBER:
1483 			break;
1484 		}
1485 		IN6M_UNLOCK(inm);
1486 		IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1487 	}
1488 
1489 out:
1490 	/* XXX Clear embedded scope ID as userland won't expect it. */
1491 	in6_clearscope(&mld->mld_addr);
1492 
1493 	return 0;
1494 }
1495 
1496 /*
1497  * MLD input path.
1498  *
1499  * Assume query messages which fit in a single ICMPv6 message header
1500  * have been pulled up.
1501  * Assume that userland will want to see the message, even if it
1502  * otherwise fails kernel input validation; do not free it.
1503  * Pullup may however free the mbuf chain m if it fails.
1504  *
1505  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1506  */
1507 int
mld_input(struct mbuf * m,int off,int icmp6len)1508 mld_input(struct mbuf *m, int off, int icmp6len)
1509 {
1510 	struct ifnet    *ifp = NULL;
1511 	struct ip6_hdr  *ip6 = NULL;
1512 	struct mld_hdr  *mld = NULL;
1513 	int              mldlen = 0;
1514 
1515 	MLD_PRINTF(("%s: called w/mbuf (0x%llx,%d)\n", __func__,
1516 	    (uint64_t)VM_KERNEL_ADDRPERM(m), off));
1517 
1518 	ifp = m->m_pkthdr.rcvif;
1519 
1520 	/* Pullup to appropriate size. */
1521 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1522 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1523 	    icmp6len >= sizeof(struct mldv2_query)) {
1524 		mldlen = sizeof(struct mldv2_query);
1525 	} else {
1526 		mldlen = sizeof(struct mld_hdr);
1527 	}
1528 	// check if mldv2_query/mld_hdr fits in the first mbuf
1529 	IP6_EXTHDR_CHECK(m, off, mldlen, return IPPROTO_DONE);
1530 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1531 	if (mld == NULL) {
1532 		icmp6stat.icp6s_badlen++;
1533 		return IPPROTO_DONE;
1534 	}
1535 	ip6 = mtod(m, struct ip6_hdr *);
1536 
1537 	/*
1538 	 * Userland needs to see all of this traffic for implementing
1539 	 * the endpoint discovery portion of multicast routing.
1540 	 */
1541 	switch (mld->mld_type) {
1542 	case MLD_LISTENER_QUERY:
1543 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1544 		if (icmp6len == sizeof(struct mld_hdr)) {
1545 			if (mld_v1_input_query(ifp, ip6, mld) != 0) {
1546 				return 0;
1547 			}
1548 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1549 			if (mld_v2_input_query(ifp, ip6, m, off,
1550 			    icmp6len) != 0) {
1551 				return 0;
1552 			}
1553 		}
1554 		break;
1555 	case MLD_LISTENER_REPORT:
1556 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1557 		if (mld_v1_input_report(ifp, m, ip6, mld) != 0) {
1558 			return 0;
1559 		}
1560 		break;
1561 	case MLDV2_LISTENER_REPORT:
1562 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1563 		break;
1564 	case MLD_LISTENER_DONE:
1565 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1566 		break;
1567 	default:
1568 		break;
1569 	}
1570 
1571 	return 0;
1572 }
1573 
1574 /*
1575  * Schedule MLD timer based on various parameters; caller must ensure that
1576  * lock ordering is maintained as this routine acquires MLD global lock.
1577  */
1578 void
mld_set_timeout(struct mld_tparams * mtp)1579 mld_set_timeout(struct mld_tparams *mtp)
1580 {
1581 	MLD_LOCK_ASSERT_NOTHELD();
1582 	VERIFY(mtp != NULL);
1583 
1584 	if (mtp->qpt != 0 || mtp->it != 0 || mtp->cst != 0 || mtp->sct != 0) {
1585 		MLD_LOCK();
1586 		if (mtp->qpt != 0) {
1587 			querier_present_timers_running6 = 1;
1588 		}
1589 		if (mtp->it != 0) {
1590 			interface_timers_running6 = 1;
1591 		}
1592 		if (mtp->cst != 0) {
1593 			current_state_timers_running6 = 1;
1594 		}
1595 		if (mtp->sct != 0) {
1596 			state_change_timers_running6 = 1;
1597 		}
1598 		mld_sched_timeout(mtp->fast);
1599 		MLD_UNLOCK();
1600 	}
1601 }
1602 
1603 void
mld_set_fast_timeout(struct mld_tparams * mtp)1604 mld_set_fast_timeout(struct mld_tparams *mtp)
1605 {
1606 	VERIFY(mtp != NULL);
1607 	mtp->fast = true;
1608 	mld_set_timeout(mtp);
1609 }
1610 
1611 /*
1612  * MLD6 timer handler (per 1 second).
1613  */
1614 static void
mld_timeout(void * arg)1615 mld_timeout(void *arg)
1616 {
1617 	struct ifqueue           scq;   /* State-change packets */
1618 	struct ifqueue           qrq;   /* Query response packets */
1619 	struct ifnet            *ifp;
1620 	struct mld_ifinfo       *mli;
1621 	struct in6_multi        *inm;
1622 	int                      uri_sec = 0;
1623 	unsigned int genid = mld_mli_list_genid;
1624 	bool                     fast = arg != NULL;
1625 
1626 	SLIST_HEAD(, in6_multi) in6m_dthead;
1627 
1628 	SLIST_INIT(&in6m_dthead);
1629 
1630 	/*
1631 	 * Update coarse-grained networking timestamp (in sec.); the idea
1632 	 * is to piggy-back on the timeout callout to update the counter
1633 	 * returnable via net_uptime().
1634 	 */
1635 	net_update_uptime();
1636 
1637 	MLD_LOCK();
1638 
1639 	MLD_PRINTF(("%s: qpt %d, it %d, cst %d, sct %d, fast %d\n", __func__,
1640 	    querier_present_timers_running6, interface_timers_running6,
1641 	    current_state_timers_running6, state_change_timers_running6, fast));
1642 
1643 	if (fast) {
1644 		/*
1645 		 * When running the fast timer, skip processing
1646 		 * of "querier present" timers since they are
1647 		 * based on 1-second intervals.
1648 		 */
1649 		goto skip_query_timers;
1650 	}
1651 	/*
1652 	 * MLDv1 querier present timer processing.
1653 	 */
1654 	if (querier_present_timers_running6) {
1655 		querier_present_timers_running6 = 0;
1656 		LIST_FOREACH(mli, &mli_head, mli_link) {
1657 			MLI_LOCK(mli);
1658 			mld_v1_process_querier_timers(mli);
1659 			if (mli->mli_v1_timer > 0) {
1660 				querier_present_timers_running6 = 1;
1661 			}
1662 			MLI_UNLOCK(mli);
1663 		}
1664 	}
1665 
1666 	/*
1667 	 * MLDv2 General Query response timer processing.
1668 	 */
1669 	if (interface_timers_running6) {
1670 		MLD_PRINTF(("%s: interface timers running\n", __func__));
1671 		interface_timers_running6 = 0;
1672 		mli = LIST_FIRST(&mli_head);
1673 
1674 		while (mli != NULL) {
1675 			if (mli->mli_flags & MLIF_PROCESSED) {
1676 				mli = LIST_NEXT(mli, mli_link);
1677 				continue;
1678 			}
1679 
1680 			MLI_LOCK(mli);
1681 			if (mli->mli_version != MLD_VERSION_2) {
1682 				MLI_UNLOCK(mli);
1683 				mli = LIST_NEXT(mli, mli_link);
1684 				continue;
1685 			}
1686 			/*
1687 			 * XXX The logic below ends up calling
1688 			 * mld_dispatch_packet which can unlock mli
1689 			 * and the global MLD lock.
1690 			 * Therefore grab a reference on MLI and also
1691 			 * check for generation count to see if we should
1692 			 * iterate the list again.
1693 			 */
1694 			MLI_ADDREF_LOCKED(mli);
1695 
1696 			if (mli->mli_v2_timer == 0) {
1697 				/* Do nothing. */
1698 			} else if (--mli->mli_v2_timer == 0) {
1699 				if (mld_v2_dispatch_general_query(mli) > 0) {
1700 					interface_timers_running6 = 1;
1701 				}
1702 			} else {
1703 				interface_timers_running6 = 1;
1704 			}
1705 			mli->mli_flags |= MLIF_PROCESSED;
1706 			MLI_UNLOCK(mli);
1707 			MLI_REMREF(mli);
1708 
1709 			if (genid != mld_mli_list_genid) {
1710 				MLD_PRINTF(("%s: MLD information list changed "
1711 				    "in the middle of iteration! Restart iteration.\n",
1712 				    __func__));
1713 				mli = LIST_FIRST(&mli_head);
1714 				genid = mld_mli_list_genid;
1715 			} else {
1716 				mli = LIST_NEXT(mli, mli_link);
1717 			}
1718 		}
1719 
1720 		LIST_FOREACH(mli, &mli_head, mli_link)
1721 		mli->mli_flags &= ~MLIF_PROCESSED;
1722 	}
1723 
1724 skip_query_timers:
1725 	if (!current_state_timers_running6 &&
1726 	    !state_change_timers_running6) {
1727 		goto out_locked;
1728 	}
1729 
1730 	current_state_timers_running6 = 0;
1731 	state_change_timers_running6 = 0;
1732 
1733 	MLD_PRINTF(("%s: state change timers running\n", __func__));
1734 
1735 	memset(&qrq, 0, sizeof(struct ifqueue));
1736 	qrq.ifq_maxlen = MLD_MAX_G_GS_PACKETS;
1737 
1738 	memset(&scq, 0, sizeof(struct ifqueue));
1739 	scq.ifq_maxlen = MLD_MAX_STATE_CHANGE_PACKETS;
1740 
1741 	/*
1742 	 * MLD host report and state-change timer processing.
1743 	 * Note: Processing a v2 group timer may remove a node.
1744 	 */
1745 	mli = LIST_FIRST(&mli_head);
1746 
1747 	while (mli != NULL) {
1748 		struct in6_multistep step;
1749 
1750 		if (mli->mli_flags & MLIF_PROCESSED) {
1751 			mli = LIST_NEXT(mli, mli_link);
1752 			continue;
1753 		}
1754 
1755 		MLI_LOCK(mli);
1756 		ifp = mli->mli_ifp;
1757 		uri_sec = MLD_RANDOM_DELAY(mli->mli_uri);
1758 		MLI_UNLOCK(mli);
1759 
1760 		in6_multihead_lock_shared();
1761 		IN6_FIRST_MULTI(step, inm);
1762 		while (inm != NULL) {
1763 			IN6M_LOCK(inm);
1764 			if (inm->in6m_ifp != ifp) {
1765 				goto next;
1766 			}
1767 
1768 			MLI_LOCK(mli);
1769 			switch (mli->mli_version) {
1770 			case MLD_VERSION_1:
1771 				mld_v1_process_group_timer(inm,
1772 				    mli->mli_version);
1773 				break;
1774 			case MLD_VERSION_2:
1775 				mld_v2_process_group_timers(mli, &qrq,
1776 				    &scq, inm, uri_sec);
1777 				break;
1778 			}
1779 			MLI_UNLOCK(mli);
1780 next:
1781 			IN6M_UNLOCK(inm);
1782 			IN6_NEXT_MULTI(step, inm);
1783 		}
1784 		in6_multihead_lock_done();
1785 
1786 		/*
1787 		 * XXX The logic below ends up calling
1788 		 * mld_dispatch_packet which can unlock mli
1789 		 * and the global MLD lock.
1790 		 * Therefore grab a reference on MLI and also
1791 		 * check for generation count to see if we should
1792 		 * iterate the list again.
1793 		 */
1794 		MLI_LOCK(mli);
1795 		MLI_ADDREF_LOCKED(mli);
1796 		if (mli->mli_version == MLD_VERSION_1) {
1797 			mld_dispatch_queue_locked(mli, &mli->mli_v1q, 0);
1798 		} else if (mli->mli_version == MLD_VERSION_2) {
1799 			MLI_UNLOCK(mli);
1800 			mld_dispatch_queue_locked(NULL, &qrq, 0);
1801 			mld_dispatch_queue_locked(NULL, &scq, 0);
1802 			VERIFY(qrq.ifq_len == 0);
1803 			VERIFY(scq.ifq_len == 0);
1804 			MLI_LOCK(mli);
1805 		}
1806 		/*
1807 		 * In case there are still any pending membership reports
1808 		 * which didn't get drained at version change time.
1809 		 */
1810 		IF_DRAIN(&mli->mli_v1q);
1811 		/*
1812 		 * Release all deferred inm records, and drain any locally
1813 		 * enqueued packets; do it even if the current MLD version
1814 		 * for the link is no longer MLDv2, in order to handle the
1815 		 * version change case.
1816 		 */
1817 		mld_flush_relq(mli, (struct mld_in6m_relhead *)&in6m_dthead);
1818 		VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
1819 		mli->mli_flags |= MLIF_PROCESSED;
1820 		MLI_UNLOCK(mli);
1821 		MLI_REMREF(mli);
1822 
1823 		IF_DRAIN(&qrq);
1824 		IF_DRAIN(&scq);
1825 
1826 		if (genid != mld_mli_list_genid) {
1827 			MLD_PRINTF(("%s: MLD information list changed "
1828 			    "in the middle of iteration! Restart iteration.\n",
1829 			    __func__));
1830 			mli = LIST_FIRST(&mli_head);
1831 			genid = mld_mli_list_genid;
1832 		} else {
1833 			mli = LIST_NEXT(mli, mli_link);
1834 		}
1835 	}
1836 
1837 	LIST_FOREACH(mli, &mli_head, mli_link)
1838 	mli->mli_flags &= ~MLIF_PROCESSED;
1839 
1840 out_locked:
1841 	/* re-arm the timer if there's work to do */
1842 	mld_timeout_run = 0;
1843 	mld_sched_timeout(false);
1844 	MLD_UNLOCK();
1845 
1846 	/* Now that we're dropped all locks, release detached records */
1847 	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
1848 }
1849 
1850 static void
mld_sched_timeout(bool fast)1851 mld_sched_timeout(bool fast)
1852 {
1853 	MLD_LOCK_ASSERT_HELD();
1854 
1855 	if (!mld_timeout_run &&
1856 	    (querier_present_timers_running6 || current_state_timers_running6 ||
1857 	    interface_timers_running6 || state_change_timers_running6)) {
1858 		mld_timeout_run = 1;
1859 		int sched_hz = fast ? 0 : hz;
1860 		void *arg = fast ? (void *)mld_sched_timeout : NULL;
1861 		timeout(mld_timeout, arg, sched_hz);
1862 	}
1863 }
1864 
1865 /*
1866  * Free the in6_multi reference(s) for this MLD lifecycle.
1867  *
1868  * Caller must be holding mli_lock.
1869  */
1870 static void
mld_flush_relq(struct mld_ifinfo * mli,struct mld_in6m_relhead * in6m_dthead)1871 mld_flush_relq(struct mld_ifinfo *mli, struct mld_in6m_relhead *in6m_dthead)
1872 {
1873 	struct in6_multi *inm;
1874 
1875 again:
1876 	MLI_LOCK_ASSERT_HELD(mli);
1877 	inm = SLIST_FIRST(&mli->mli_relinmhead);
1878 	if (inm != NULL) {
1879 		int lastref;
1880 
1881 		SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
1882 		MLI_UNLOCK(mli);
1883 
1884 		in6_multihead_lock_exclusive();
1885 		IN6M_LOCK(inm);
1886 		VERIFY(inm->in6m_nrelecnt != 0);
1887 		inm->in6m_nrelecnt--;
1888 		lastref = in6_multi_detach(inm);
1889 		VERIFY(!lastref || (!(inm->in6m_debug & IFD_ATTACHED) &&
1890 		    inm->in6m_reqcnt == 0));
1891 		IN6M_UNLOCK(inm);
1892 		in6_multihead_lock_done();
1893 		/* from mli_relinmhead */
1894 		IN6M_REMREF(inm);
1895 		/* from in6_multihead_list */
1896 		if (lastref) {
1897 			/*
1898 			 * Defer releasing our final reference, as we
1899 			 * are holding the MLD lock at this point, and
1900 			 * we could end up with locking issues later on
1901 			 * (while issuing SIOCDELMULTI) when this is the
1902 			 * final reference count.  Let the caller do it
1903 			 * when it is safe.
1904 			 */
1905 			MLD_ADD_DETACHED_IN6M(in6m_dthead, inm);
1906 		}
1907 		MLI_LOCK(mli);
1908 		goto again;
1909 	}
1910 }
1911 
1912 /*
1913  * Update host report group timer.
1914  * Will update the global pending timer flags.
1915  */
1916 static void
mld_v1_process_group_timer(struct in6_multi * inm,const int mld_version)1917 mld_v1_process_group_timer(struct in6_multi *inm, const int mld_version)
1918 {
1919 #pragma unused(mld_version)
1920 	int report_timer_expired;
1921 
1922 	MLD_LOCK_ASSERT_HELD();
1923 	IN6M_LOCK_ASSERT_HELD(inm);
1924 	MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
1925 
1926 	if (inm->in6m_timer == 0) {
1927 		report_timer_expired = 0;
1928 	} else if (--inm->in6m_timer == 0) {
1929 		report_timer_expired = 1;
1930 	} else {
1931 		current_state_timers_running6 = 1;
1932 		/* caller will schedule timer */
1933 		return;
1934 	}
1935 
1936 	switch (inm->in6m_state) {
1937 	case MLD_NOT_MEMBER:
1938 	case MLD_SILENT_MEMBER:
1939 	case MLD_IDLE_MEMBER:
1940 	case MLD_LAZY_MEMBER:
1941 	case MLD_SLEEPING_MEMBER:
1942 	case MLD_AWAKENING_MEMBER:
1943 		break;
1944 	case MLD_REPORTING_MEMBER:
1945 		if (report_timer_expired) {
1946 			inm->in6m_state = MLD_IDLE_MEMBER;
1947 			(void) mld_v1_transmit_report(inm,
1948 			    MLD_LISTENER_REPORT);
1949 			IN6M_LOCK_ASSERT_HELD(inm);
1950 			MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
1951 		}
1952 		break;
1953 	case MLD_G_QUERY_PENDING_MEMBER:
1954 	case MLD_SG_QUERY_PENDING_MEMBER:
1955 	case MLD_LEAVING_MEMBER:
1956 		break;
1957 	}
1958 }
1959 
1960 /*
1961  * Update a group's timers for MLDv2.
1962  * Will update the global pending timer flags.
1963  * Note: Unlocked read from mli.
1964  */
1965 static void
mld_v2_process_group_timers(struct mld_ifinfo * mli,struct ifqueue * qrq,struct ifqueue * scq,struct in6_multi * inm,const int uri_sec)1966 mld_v2_process_group_timers(struct mld_ifinfo *mli,
1967     struct ifqueue *qrq, struct ifqueue *scq,
1968     struct in6_multi *inm, const int uri_sec)
1969 {
1970 	int query_response_timer_expired;
1971 	int state_change_retransmit_timer_expired;
1972 
1973 	MLD_LOCK_ASSERT_HELD();
1974 	IN6M_LOCK_ASSERT_HELD(inm);
1975 	MLI_LOCK_ASSERT_HELD(mli);
1976 	VERIFY(mli == inm->in6m_mli);
1977 
1978 	query_response_timer_expired = 0;
1979 	state_change_retransmit_timer_expired = 0;
1980 
1981 	/*
1982 	 * During a transition from compatibility mode back to MLDv2,
1983 	 * a group record in REPORTING state may still have its group
1984 	 * timer active. This is a no-op in this function; it is easier
1985 	 * to deal with it here than to complicate the timeout path.
1986 	 */
1987 	if (inm->in6m_timer == 0) {
1988 		query_response_timer_expired = 0;
1989 	} else if (--inm->in6m_timer == 0) {
1990 		query_response_timer_expired = 1;
1991 	} else {
1992 		current_state_timers_running6 = 1;
1993 		/* caller will schedule timer */
1994 	}
1995 
1996 	if (inm->in6m_sctimer == 0) {
1997 		state_change_retransmit_timer_expired = 0;
1998 	} else if (--inm->in6m_sctimer == 0) {
1999 		state_change_retransmit_timer_expired = 1;
2000 	} else {
2001 		state_change_timers_running6 = 1;
2002 		/* caller will schedule timer */
2003 	}
2004 
2005 	/* We are in timer callback, so be quick about it. */
2006 	if (!state_change_retransmit_timer_expired &&
2007 	    !query_response_timer_expired) {
2008 		return;
2009 	}
2010 
2011 	switch (inm->in6m_state) {
2012 	case MLD_NOT_MEMBER:
2013 	case MLD_SILENT_MEMBER:
2014 	case MLD_SLEEPING_MEMBER:
2015 	case MLD_LAZY_MEMBER:
2016 	case MLD_AWAKENING_MEMBER:
2017 	case MLD_IDLE_MEMBER:
2018 		break;
2019 	case MLD_G_QUERY_PENDING_MEMBER:
2020 	case MLD_SG_QUERY_PENDING_MEMBER:
2021 		/*
2022 		 * Respond to a previously pending Group-Specific
2023 		 * or Group-and-Source-Specific query by enqueueing
2024 		 * the appropriate Current-State report for
2025 		 * immediate transmission.
2026 		 */
2027 		if (query_response_timer_expired) {
2028 			int retval;
2029 
2030 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
2031 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
2032 			    0);
2033 			MLD_PRINTF(("%s: enqueue record = %d\n",
2034 			    __func__, retval));
2035 			inm->in6m_state = MLD_REPORTING_MEMBER;
2036 			in6m_clear_recorded(inm);
2037 		}
2038 		OS_FALLTHROUGH;
2039 	case MLD_REPORTING_MEMBER:
2040 	case MLD_LEAVING_MEMBER:
2041 		if (state_change_retransmit_timer_expired) {
2042 			/*
2043 			 * State-change retransmission timer fired.
2044 			 * If there are any further pending retransmissions,
2045 			 * set the global pending state-change flag, and
2046 			 * reset the timer.
2047 			 */
2048 			if (--inm->in6m_scrv > 0) {
2049 				inm->in6m_sctimer = (uint16_t)uri_sec;
2050 				state_change_timers_running6 = 1;
2051 				/* caller will schedule timer */
2052 			}
2053 			/*
2054 			 * Retransmit the previously computed state-change
2055 			 * report. If there are no further pending
2056 			 * retransmissions, the mbuf queue will be consumed.
2057 			 * Update T0 state to T1 as we have now sent
2058 			 * a state-change.
2059 			 */
2060 			(void) mld_v2_merge_state_changes(inm, scq);
2061 
2062 			in6m_commit(inm);
2063 			MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2064 			    ip6_sprintf(&inm->in6m_addr),
2065 			    if_name(inm->in6m_ifp)));
2066 
2067 			/*
2068 			 * If we are leaving the group for good, make sure
2069 			 * we release MLD's reference to it.
2070 			 * This release must be deferred using a SLIST,
2071 			 * as we are called from a loop which traverses
2072 			 * the in_ifmultiaddr TAILQ.
2073 			 */
2074 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
2075 			    inm->in6m_scrv == 0) {
2076 				inm->in6m_state = MLD_NOT_MEMBER;
2077 				/*
2078 				 * A reference has already been held in
2079 				 * mld_final_leave() for this inm, so
2080 				 * no need to hold another one.  We also
2081 				 * bumped up its request count then, so
2082 				 * that it stays in in6_multihead.  Both
2083 				 * of them will be released when it is
2084 				 * dequeued later on.
2085 				 */
2086 				VERIFY(inm->in6m_nrelecnt != 0);
2087 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
2088 				    inm, in6m_nrele);
2089 			}
2090 		}
2091 		break;
2092 	}
2093 }
2094 
2095 /*
2096  * Switch to a different version on the given interface,
2097  * as per Section 9.12.
2098  */
2099 static uint32_t
mld_set_version(struct mld_ifinfo * mli,const int mld_version)2100 mld_set_version(struct mld_ifinfo *mli, const int mld_version)
2101 {
2102 	int old_version_timer;
2103 
2104 	MLI_LOCK_ASSERT_HELD(mli);
2105 
2106 	MLD_PRINTF(("%s: switching to v%d on ifp 0x%llx(%s)\n", __func__,
2107 	    mld_version, (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp),
2108 	    if_name(mli->mli_ifp)));
2109 
2110 	if (mld_version == MLD_VERSION_1) {
2111 		/*
2112 		 * Compute the "Older Version Querier Present" timer as per
2113 		 * Section 9.12, in seconds.
2114 		 */
2115 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
2116 		mli->mli_v1_timer = old_version_timer;
2117 	}
2118 
2119 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
2120 		mli->mli_version = MLD_VERSION_1;
2121 		mld_v2_cancel_link_timers(mli);
2122 	}
2123 
2124 	MLI_LOCK_ASSERT_HELD(mli);
2125 
2126 	return mli->mli_v1_timer;
2127 }
2128 
2129 /*
2130  * Cancel pending MLDv2 timers for the given link and all groups
2131  * joined on it; state-change, general-query, and group-query timers.
2132  *
2133  * Only ever called on a transition from v2 to Compatibility mode. Kill
2134  * the timers stone dead (this may be expensive for large N groups), they
2135  * will be restarted if Compatibility Mode deems that they must be due to
2136  * query processing.
2137  */
2138 static void
mld_v2_cancel_link_timers(struct mld_ifinfo * mli)2139 mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
2140 {
2141 	struct ifnet            *ifp;
2142 	struct in6_multi        *inm;
2143 	struct in6_multistep    step;
2144 
2145 	MLI_LOCK_ASSERT_HELD(mli);
2146 
2147 	MLD_PRINTF(("%s: cancel v2 timers on ifp 0x%llx(%s)\n", __func__,
2148 	    (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp), if_name(mli->mli_ifp)));
2149 
2150 	/*
2151 	 * Stop the v2 General Query Response on this link stone dead.
2152 	 * If timer is woken up due to interface_timers_running6,
2153 	 * the flag will be cleared if there are no pending link timers.
2154 	 */
2155 	mli->mli_v2_timer = 0;
2156 
2157 	/*
2158 	 * Now clear the current-state and state-change report timers
2159 	 * for all memberships scoped to this link.
2160 	 */
2161 	ifp = mli->mli_ifp;
2162 	MLI_UNLOCK(mli);
2163 
2164 	in6_multihead_lock_shared();
2165 	IN6_FIRST_MULTI(step, inm);
2166 	while (inm != NULL) {
2167 		IN6M_LOCK(inm);
2168 		if (inm->in6m_ifp != ifp) {
2169 			goto next;
2170 		}
2171 
2172 		switch (inm->in6m_state) {
2173 		case MLD_NOT_MEMBER:
2174 		case MLD_SILENT_MEMBER:
2175 		case MLD_IDLE_MEMBER:
2176 		case MLD_LAZY_MEMBER:
2177 		case MLD_SLEEPING_MEMBER:
2178 		case MLD_AWAKENING_MEMBER:
2179 			/*
2180 			 * These states are either not relevant in v2 mode,
2181 			 * or are unreported. Do nothing.
2182 			 */
2183 			break;
2184 		case MLD_LEAVING_MEMBER:
2185 			/*
2186 			 * If we are leaving the group and switching
2187 			 * version, we need to release the final
2188 			 * reference held for issuing the INCLUDE {}.
2189 			 * During mld_final_leave(), we bumped up both the
2190 			 * request and reference counts.  Since we cannot
2191 			 * call in6_multi_detach() here, defer this task to
2192 			 * the timer routine.
2193 			 */
2194 			VERIFY(inm->in6m_nrelecnt != 0);
2195 			MLI_LOCK(mli);
2196 			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
2197 			    in6m_nrele);
2198 			MLI_UNLOCK(mli);
2199 			OS_FALLTHROUGH;
2200 		case MLD_G_QUERY_PENDING_MEMBER:
2201 		case MLD_SG_QUERY_PENDING_MEMBER:
2202 			in6m_clear_recorded(inm);
2203 			OS_FALLTHROUGH;
2204 		case MLD_REPORTING_MEMBER:
2205 			inm->in6m_state = MLD_REPORTING_MEMBER;
2206 			break;
2207 		}
2208 		/*
2209 		 * Always clear state-change and group report timers.
2210 		 * Free any pending MLDv2 state-change records.
2211 		 */
2212 		inm->in6m_sctimer = 0;
2213 		inm->in6m_timer = 0;
2214 		IF_DRAIN(&inm->in6m_scq);
2215 next:
2216 		IN6M_UNLOCK(inm);
2217 		IN6_NEXT_MULTI(step, inm);
2218 	}
2219 	in6_multihead_lock_done();
2220 
2221 	MLI_LOCK(mli);
2222 }
2223 
2224 /*
2225  * Update the Older Version Querier Present timers for a link.
2226  * See Section 9.12 of RFC 3810.
2227  */
2228 static void
mld_v1_process_querier_timers(struct mld_ifinfo * mli)2229 mld_v1_process_querier_timers(struct mld_ifinfo *mli)
2230 {
2231 	MLI_LOCK_ASSERT_HELD(mli);
2232 
2233 	if (mld_v2enable && mli->mli_version != MLD_VERSION_2 &&
2234 	    --mli->mli_v1_timer == 0) {
2235 		/*
2236 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
2237 		 */
2238 		MLD_PRINTF(("%s: transition from v%d -> v%d on 0x%llx(%s)\n",
2239 		    __func__, mli->mli_version, MLD_VERSION_2,
2240 		    (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp),
2241 		    if_name(mli->mli_ifp)));
2242 		mli->mli_version = MLD_VERSION_2;
2243 	}
2244 }
2245 
2246 /*
2247  * Transmit an MLDv1 report immediately.
2248  */
2249 static int
mld_v1_transmit_report(struct in6_multi * in6m,const uint8_t type)2250 mld_v1_transmit_report(struct in6_multi *in6m, const uint8_t type)
2251 {
2252 	struct ifnet            *ifp;
2253 	struct in6_ifaddr       *ia;
2254 	struct ip6_hdr          *ip6;
2255 	struct mbuf             *mh, *md;
2256 	struct mld_hdr          *mld;
2257 	int                     error = 0;
2258 
2259 	IN6M_LOCK_ASSERT_HELD(in6m);
2260 	MLI_LOCK_ASSERT_HELD(in6m->in6m_mli);
2261 
2262 	ifp = in6m->in6m_ifp;
2263 	/* ia may be NULL if link-local address is tentative. */
2264 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
2265 
2266 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
2267 	if (mh == NULL) {
2268 		if (ia != NULL) {
2269 			IFA_REMREF(&ia->ia_ifa);
2270 		}
2271 		return ENOMEM;
2272 	}
2273 	MGET(md, M_DONTWAIT, MT_DATA);
2274 	if (md == NULL) {
2275 		m_free(mh);
2276 		if (ia != NULL) {
2277 			IFA_REMREF(&ia->ia_ifa);
2278 		}
2279 		return ENOMEM;
2280 	}
2281 	mh->m_next = md;
2282 
2283 	/*
2284 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
2285 	 * that ether_output() does not need to allocate another mbuf
2286 	 * for the header in the most common case.
2287 	 */
2288 	MH_ALIGN(mh, sizeof(struct ip6_hdr));
2289 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
2290 	mh->m_len = sizeof(struct ip6_hdr);
2291 
2292 	ip6 = mtod(mh, struct ip6_hdr *);
2293 	ip6->ip6_flow = 0;
2294 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
2295 	ip6->ip6_vfc |= IPV6_VERSION;
2296 	ip6->ip6_nxt = IPPROTO_ICMPV6;
2297 	if (ia != NULL) {
2298 		IFA_LOCK(&ia->ia_ifa);
2299 	}
2300 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
2301 	ip6_output_setsrcifscope(mh, IFSCOPE_NONE, ia);
2302 	if (ia != NULL) {
2303 		IFA_UNLOCK(&ia->ia_ifa);
2304 		IFA_REMREF(&ia->ia_ifa);
2305 		ia = NULL;
2306 	}
2307 	ip6->ip6_dst = in6m->in6m_addr;
2308 	ip6_output_setdstifscope(mh, in6m->ifscope, NULL);
2309 
2310 	md->m_len = sizeof(struct mld_hdr);
2311 	mld = mtod(md, struct mld_hdr *);
2312 	mld->mld_type = type;
2313 	mld->mld_code = 0;
2314 	mld->mld_cksum = 0;
2315 	mld->mld_maxdelay = 0;
2316 	mld->mld_reserved = 0;
2317 	mld->mld_addr = in6m->in6m_addr;
2318 	in6_clearscope(&mld->mld_addr);
2319 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
2320 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
2321 
2322 	mld_save_context(mh, ifp);
2323 	mh->m_flags |= M_MLDV1;
2324 
2325 	/*
2326 	 * Due to the fact that at this point we are possibly holding
2327 	 * in6_multihead_lock in shared or exclusive mode, we can't call
2328 	 * mld_dispatch_packet() here since that will eventually call
2329 	 * ip6_output(), which will try to lock in6_multihead_lock and cause
2330 	 * a deadlock.
2331 	 * Instead we defer the work to the mld_timeout() thread, thus
2332 	 * avoiding unlocking in_multihead_lock here.
2333 	 */
2334 	if (IF_QFULL(&in6m->in6m_mli->mli_v1q)) {
2335 		MLD_PRINTF(("%s: v1 outbound queue full\n", __func__));
2336 		error = ENOMEM;
2337 		m_freem(mh);
2338 	} else {
2339 		IF_ENQUEUE(&in6m->in6m_mli->mli_v1q, mh);
2340 		VERIFY(error == 0);
2341 	}
2342 
2343 	return error;
2344 }
2345 
2346 /*
2347  * Process a state change from the upper layer for the given IPv6 group.
2348  *
2349  * Each socket holds a reference on the in6_multi in its own ip_moptions.
2350  * The socket layer will have made the necessary updates to.the group
2351  * state, it is now up to MLD to issue a state change report if there
2352  * has been any change between T0 (when the last state-change was issued)
2353  * and T1 (now).
2354  *
2355  * We use the MLDv2 state machine at group level. The MLd module
2356  * however makes the decision as to which MLD protocol version to speak.
2357  * A state change *from* INCLUDE {} always means an initial join.
2358  * A state change *to* INCLUDE {} always means a final leave.
2359  *
2360  * If delay is non-zero, and the state change is an initial multicast
2361  * join, the state change report will be delayed by 'delay' ticks
2362  * in units of seconds if MLDv1 is active on the link; otherwise
2363  * the initial MLDv2 state change report will be delayed by whichever
2364  * is sooner, a pending state-change timer or delay itself.
2365  */
2366 int
mld_change_state(struct in6_multi * inm,struct mld_tparams * mtp,const int delay)2367 mld_change_state(struct in6_multi *inm, struct mld_tparams *mtp,
2368     const int delay)
2369 {
2370 	struct mld_ifinfo *mli;
2371 	struct ifnet *ifp;
2372 	int error = 0;
2373 
2374 	VERIFY(mtp != NULL);
2375 	bzero(mtp, sizeof(*mtp));
2376 
2377 	IN6M_LOCK_ASSERT_HELD(inm);
2378 	VERIFY(inm->in6m_mli != NULL);
2379 	MLI_LOCK_ASSERT_NOTHELD(inm->in6m_mli);
2380 
2381 	/*
2382 	 * Try to detect if the upper layer just asked us to change state
2383 	 * for an interface which has now gone away.
2384 	 */
2385 	VERIFY(inm->in6m_ifma != NULL);
2386 	ifp = inm->in6m_ifma->ifma_ifp;
2387 	/*
2388 	 * Sanity check that netinet6's notion of ifp is the same as net's.
2389 	 */
2390 	VERIFY(inm->in6m_ifp == ifp);
2391 
2392 	mli = MLD_IFINFO(ifp);
2393 	VERIFY(mli != NULL);
2394 
2395 	/*
2396 	 * If we detect a state transition to or from MCAST_UNDEFINED
2397 	 * for this group, then we are starting or finishing an MLD
2398 	 * life cycle for this group.
2399 	 */
2400 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
2401 		MLD_PRINTF(("%s: inm transition %d -> %d\n", __func__,
2402 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode));
2403 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
2404 			MLD_PRINTF(("%s: initial join\n", __func__));
2405 			error = mld_initial_join(inm, mli, mtp, delay);
2406 			goto out;
2407 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
2408 			MLD_PRINTF(("%s: final leave\n", __func__));
2409 			mld_final_leave(inm, mli, mtp);
2410 			goto out;
2411 		}
2412 	} else {
2413 		MLD_PRINTF(("%s: filter set change\n", __func__));
2414 	}
2415 
2416 	error = mld_handle_state_change(inm, mli, mtp);
2417 out:
2418 	return error;
2419 }
2420 
2421 /*
2422  * Perform the initial join for an MLD group.
2423  *
2424  * When joining a group:
2425  *  If the group should have its MLD traffic suppressed, do nothing.
2426  *  MLDv1 starts sending MLDv1 host membership reports.
2427  *  MLDv2 will schedule an MLDv2 state-change report containing the
2428  *  initial state of the membership.
2429  *
2430  * If the delay argument is non-zero, then we must delay sending the
2431  * initial state change for delay ticks (in units of seconds).
2432  */
2433 static int
mld_initial_join(struct in6_multi * inm,struct mld_ifinfo * mli,struct mld_tparams * mtp,const int delay)2434 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
2435     struct mld_tparams *mtp, const int delay)
2436 {
2437 	struct ifnet            *ifp;
2438 	struct ifqueue          *ifq;
2439 	int                      error, retval, syncstates;
2440 	int                      odelay;
2441 
2442 	IN6M_LOCK_ASSERT_HELD(inm);
2443 	MLI_LOCK_ASSERT_NOTHELD(mli);
2444 	VERIFY(mtp != NULL);
2445 
2446 	MLD_PRINTF(("%s: initial join %s on ifp 0x%llx(%s)\n",
2447 	    __func__, ip6_sprintf(&inm->in6m_addr),
2448 	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2449 	    if_name(inm->in6m_ifp)));
2450 
2451 	error = 0;
2452 	syncstates = 1;
2453 
2454 	ifp = inm->in6m_ifp;
2455 
2456 	MLI_LOCK(mli);
2457 	VERIFY(mli->mli_ifp == ifp);
2458 
2459 	/*
2460 	 * Avoid MLD if group is :
2461 	 * 1. Joined on loopback, OR
2462 	 * 2. On a link that is marked MLIF_SILENT
2463 	 * 3. rdar://problem/19227650 Is link local scoped and
2464 	 *    on cellular interface
2465 	 * 4. Is a type that should not be reported (node local
2466 	 *    or all node link local multicast.
2467 	 * All other groups enter the appropriate state machine
2468 	 * for the version in use on this link.
2469 	 */
2470 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2471 	    (mli->mli_flags & MLIF_SILENT) ||
2472 	    (IFNET_IS_CELLULAR(ifp) &&
2473 	    (IN6_IS_ADDR_MC_LINKLOCAL(&inm->in6m_addr) || IN6_IS_ADDR_MC_UNICAST_BASED_LINKLOCAL(&inm->in6m_addr))) ||
2474 	    !mld_is_addr_reported(&inm->in6m_addr)) {
2475 		MLD_PRINTF(("%s: not kicking state machine for silent group\n",
2476 		    __func__));
2477 		inm->in6m_state = MLD_SILENT_MEMBER;
2478 		inm->in6m_timer = 0;
2479 	} else {
2480 		/*
2481 		 * Deal with overlapping in6_multi lifecycle.
2482 		 * If this group was LEAVING, then make sure
2483 		 * we drop the reference we picked up to keep the
2484 		 * group around for the final INCLUDE {} enqueue.
2485 		 * Since we cannot call in6_multi_detach() here,
2486 		 * defer this task to the timer routine.
2487 		 */
2488 		if (mli->mli_version == MLD_VERSION_2 &&
2489 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
2490 			VERIFY(inm->in6m_nrelecnt != 0);
2491 			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
2492 			    in6m_nrele);
2493 		}
2494 
2495 		inm->in6m_state = MLD_REPORTING_MEMBER;
2496 
2497 		switch (mli->mli_version) {
2498 		case MLD_VERSION_1:
2499 			/*
2500 			 * If a delay was provided, only use it if
2501 			 * it is greater than the delay normally
2502 			 * used for an MLDv1 state change report,
2503 			 * and delay sending the initial MLDv1 report
2504 			 * by not transitioning to the IDLE state.
2505 			 */
2506 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI);
2507 			if (delay) {
2508 				inm->in6m_timer = max(delay, odelay);
2509 				mtp->cst = 1;
2510 			} else {
2511 				inm->in6m_state = MLD_IDLE_MEMBER;
2512 				error = mld_v1_transmit_report(inm,
2513 				    MLD_LISTENER_REPORT);
2514 
2515 				IN6M_LOCK_ASSERT_HELD(inm);
2516 				MLI_LOCK_ASSERT_HELD(mli);
2517 
2518 				if (error == 0) {
2519 					inm->in6m_timer = odelay;
2520 					mtp->cst = 1;
2521 				}
2522 			}
2523 			break;
2524 
2525 		case MLD_VERSION_2:
2526 			/*
2527 			 * Defer update of T0 to T1, until the first copy
2528 			 * of the state change has been transmitted.
2529 			 */
2530 			syncstates = 0;
2531 
2532 			/*
2533 			 * Immediately enqueue a State-Change Report for
2534 			 * this interface, freeing any previous reports.
2535 			 * Don't kick the timers if there is nothing to do,
2536 			 * or if an error occurred.
2537 			 */
2538 			ifq = &inm->in6m_scq;
2539 			IF_DRAIN(ifq);
2540 			retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2541 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2542 			mtp->cst = (ifq->ifq_len > 0);
2543 			MLD_PRINTF(("%s: enqueue record = %d\n",
2544 			    __func__, retval));
2545 			if (retval <= 0) {
2546 				error = retval * -1;
2547 				break;
2548 			}
2549 
2550 			/*
2551 			 * Schedule transmission of pending state-change
2552 			 * report up to RV times for this link. The timer
2553 			 * will fire at the next mld_timeout (1 second)),
2554 			 * giving us an opportunity to merge the reports.
2555 			 *
2556 			 * If a delay was provided to this function, only
2557 			 * use this delay if sooner than the existing one.
2558 			 */
2559 			VERIFY(mli->mli_rv > 1);
2560 			inm->in6m_scrv = (uint16_t)mli->mli_rv;
2561 			if (delay) {
2562 				if (inm->in6m_sctimer > 1) {
2563 					inm->in6m_sctimer =
2564 					    MIN(inm->in6m_sctimer, (uint16_t)delay);
2565 				} else {
2566 					inm->in6m_sctimer = (uint16_t)delay;
2567 				}
2568 			} else {
2569 				inm->in6m_sctimer = 1;
2570 			}
2571 			mtp->sct = 1;
2572 			error = 0;
2573 			break;
2574 		}
2575 	}
2576 	MLI_UNLOCK(mli);
2577 
2578 	/*
2579 	 * Only update the T0 state if state change is atomic,
2580 	 * i.e. we don't need to wait for a timer to fire before we
2581 	 * can consider the state change to have been communicated.
2582 	 */
2583 	if (syncstates) {
2584 		in6m_commit(inm);
2585 		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2586 		    ip6_sprintf(&inm->in6m_addr),
2587 		    if_name(inm->in6m_ifp)));
2588 	}
2589 
2590 	return error;
2591 }
2592 
2593 /*
2594  * Issue an intermediate state change during the life-cycle.
2595  */
2596 static int
mld_handle_state_change(struct in6_multi * inm,struct mld_ifinfo * mli,struct mld_tparams * mtp)2597 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli,
2598     struct mld_tparams *mtp)
2599 {
2600 	struct ifnet            *ifp;
2601 	int                      retval = 0;
2602 
2603 	IN6M_LOCK_ASSERT_HELD(inm);
2604 	MLI_LOCK_ASSERT_NOTHELD(mli);
2605 	VERIFY(mtp != NULL);
2606 
2607 	MLD_PRINTF(("%s: state change for %s on ifp 0x%llx(%s)\n",
2608 	    __func__, ip6_sprintf(&inm->in6m_addr),
2609 	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2610 	    if_name(inm->in6m_ifp)));
2611 
2612 	ifp = inm->in6m_ifp;
2613 
2614 	MLI_LOCK(mli);
2615 	VERIFY(mli->mli_ifp == ifp);
2616 
2617 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2618 	    (mli->mli_flags & MLIF_SILENT) ||
2619 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2620 	    (mli->mli_version != MLD_VERSION_2)) {
2621 		MLI_UNLOCK(mli);
2622 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2623 			MLD_PRINTF(("%s: not kicking state machine for silent "
2624 			    "group\n", __func__));
2625 		}
2626 		MLD_PRINTF(("%s: nothing to do\n", __func__));
2627 		in6m_commit(inm);
2628 		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2629 		    ip6_sprintf(&inm->in6m_addr),
2630 		    if_name(inm->in6m_ifp)));
2631 		goto done;
2632 	}
2633 
2634 	IF_DRAIN(&inm->in6m_scq);
2635 
2636 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2637 	    (mli->mli_flags & MLIF_USEALLOW));
2638 	mtp->cst = (inm->in6m_scq.ifq_len > 0);
2639 	MLD_PRINTF(("%s: enqueue record = %d\n", __func__, retval));
2640 	if (retval <= 0) {
2641 		MLI_UNLOCK(mli);
2642 		retval *= -1;
2643 		goto done;
2644 	} else {
2645 		retval = 0;
2646 	}
2647 
2648 	/*
2649 	 * If record(s) were enqueued, start the state-change
2650 	 * report timer for this group.
2651 	 */
2652 	inm->in6m_scrv = (uint16_t)mli->mli_rv;
2653 	inm->in6m_sctimer = 1;
2654 	mtp->sct = 1;
2655 	MLI_UNLOCK(mli);
2656 
2657 done:
2658 	return retval;
2659 }
2660 
2661 /*
2662  * Perform the final leave for a multicast address.
2663  *
2664  * When leaving a group:
2665  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2666  *  MLDv2 enqueues a state-change report containing a transition
2667  *  to INCLUDE {} for immediate transmission.
2668  */
2669 static void
mld_final_leave(struct in6_multi * inm,struct mld_ifinfo * mli,struct mld_tparams * mtp)2670 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli,
2671     struct mld_tparams *mtp)
2672 {
2673 	int syncstates = 1;
2674 
2675 	IN6M_LOCK_ASSERT_HELD(inm);
2676 	MLI_LOCK_ASSERT_NOTHELD(mli);
2677 	VERIFY(mtp != NULL);
2678 
2679 	MLD_PRINTF(("%s: final leave %s on ifp 0x%llx(%s)\n",
2680 	    __func__, ip6_sprintf(&inm->in6m_addr),
2681 	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2682 	    if_name(inm->in6m_ifp)));
2683 
2684 	switch (inm->in6m_state) {
2685 	case MLD_NOT_MEMBER:
2686 	case MLD_SILENT_MEMBER:
2687 	case MLD_LEAVING_MEMBER:
2688 		/* Already leaving or left; do nothing. */
2689 		MLD_PRINTF(("%s: not kicking state machine for silent group\n",
2690 		    __func__));
2691 		break;
2692 	case MLD_REPORTING_MEMBER:
2693 	case MLD_IDLE_MEMBER:
2694 	case MLD_G_QUERY_PENDING_MEMBER:
2695 	case MLD_SG_QUERY_PENDING_MEMBER:
2696 		MLI_LOCK(mli);
2697 		if (mli->mli_version == MLD_VERSION_1) {
2698 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2699 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
2700 				panic("%s: MLDv2 state reached, not MLDv2 "
2701 				    "mode\n", __func__);
2702 				/* NOTREACHED */
2703 			}
2704 			/* scheduler timer if enqueue is successful */
2705 			mtp->cst = (mld_v1_transmit_report(inm,
2706 			    MLD_LISTENER_DONE) == 0);
2707 
2708 			IN6M_LOCK_ASSERT_HELD(inm);
2709 			MLI_LOCK_ASSERT_HELD(mli);
2710 
2711 			inm->in6m_state = MLD_NOT_MEMBER;
2712 		} else if (mli->mli_version == MLD_VERSION_2) {
2713 			/*
2714 			 * Stop group timer and all pending reports.
2715 			 * Immediately enqueue a state-change report
2716 			 * TO_IN {} to be sent on the next timeout,
2717 			 * giving us an opportunity to merge reports.
2718 			 */
2719 			IF_DRAIN(&inm->in6m_scq);
2720 			inm->in6m_timer = 0;
2721 			inm->in6m_scrv = (uint16_t)mli->mli_rv;
2722 			MLD_PRINTF(("%s: Leaving %s/%s with %d "
2723 			    "pending retransmissions.\n", __func__,
2724 			    ip6_sprintf(&inm->in6m_addr),
2725 			    if_name(inm->in6m_ifp),
2726 			    inm->in6m_scrv));
2727 			if (inm->in6m_scrv == 0) {
2728 				inm->in6m_state = MLD_NOT_MEMBER;
2729 				inm->in6m_sctimer = 0;
2730 			} else {
2731 				int retval;
2732 				/*
2733 				 * Stick around in the in6_multihead list;
2734 				 * the final detach will be issued by
2735 				 * mld_v2_process_group_timers() when
2736 				 * the retransmit timer expires.
2737 				 */
2738 				IN6M_ADDREF_LOCKED(inm);
2739 				VERIFY(inm->in6m_debug & IFD_ATTACHED);
2740 				inm->in6m_reqcnt++;
2741 				VERIFY(inm->in6m_reqcnt >= 1);
2742 				inm->in6m_nrelecnt++;
2743 				VERIFY(inm->in6m_nrelecnt != 0);
2744 
2745 				retval = mld_v2_enqueue_group_record(
2746 					&inm->in6m_scq, inm, 1, 0, 0,
2747 					(mli->mli_flags & MLIF_USEALLOW));
2748 				mtp->cst = (inm->in6m_scq.ifq_len > 0);
2749 				KASSERT(retval != 0,
2750 				    ("%s: enqueue record = %d\n", __func__,
2751 				    retval));
2752 
2753 				inm->in6m_state = MLD_LEAVING_MEMBER;
2754 				inm->in6m_sctimer = 1;
2755 				mtp->sct = 1;
2756 				syncstates = 0;
2757 			}
2758 		}
2759 		MLI_UNLOCK(mli);
2760 		break;
2761 	case MLD_LAZY_MEMBER:
2762 	case MLD_SLEEPING_MEMBER:
2763 	case MLD_AWAKENING_MEMBER:
2764 		/* Our reports are suppressed; do nothing. */
2765 		break;
2766 	}
2767 
2768 	if (syncstates) {
2769 		in6m_commit(inm);
2770 		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2771 		    ip6_sprintf(&inm->in6m_addr),
2772 		    if_name(inm->in6m_ifp)));
2773 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2774 		MLD_PRINTF(("%s: T1 now MCAST_UNDEFINED for 0x%llx/%s\n",
2775 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(&inm->in6m_addr),
2776 		    if_name(inm->in6m_ifp)));
2777 	}
2778 }
2779 
2780 /*
2781  * Enqueue an MLDv2 group record to the given output queue.
2782  *
2783  * If is_state_change is zero, a current-state record is appended.
2784  * If is_state_change is non-zero, a state-change report is appended.
2785  *
2786  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2787  * If is_group_query is zero, and if there is a packet with free space
2788  * at the tail of the queue, it will be appended to providing there
2789  * is enough free space.
2790  * Otherwise a new mbuf packet chain is allocated.
2791  *
2792  * If is_source_query is non-zero, each source is checked to see if
2793  * it was recorded for a Group-Source query, and will be omitted if
2794  * it is not both in-mode and recorded.
2795  *
2796  * If use_block_allow is non-zero, state change reports for initial join
2797  * and final leave, on an inclusive mode group with a source list, will be
2798  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2799  *
2800  * The function will attempt to allocate leading space in the packet
2801  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2802  *
2803  * If successful the size of all data appended to the queue is returned,
2804  * otherwise an error code less than zero is returned, or zero if
2805  * no record(s) were appended.
2806  */
2807 static int
mld_v2_enqueue_group_record(struct ifqueue * ifq,struct in6_multi * inm,const int is_state_change,const int is_group_query,const int is_source_query,const int use_block_allow)2808 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2809     const int is_state_change, const int is_group_query,
2810     const int is_source_query, const int use_block_allow)
2811 {
2812 	struct mldv2_record      mr;
2813 	struct mldv2_record     *pmr;
2814 	struct ifnet            *ifp;
2815 	struct ip6_msource      *ims, *nims;
2816 	struct mbuf             *m0, *m, *md;
2817 	int                      error, is_filter_list_change;
2818 	int                      minrec0len, m0srcs, msrcs, nbytes, off;
2819 	int                      record_has_sources;
2820 	int                      now;
2821 	uint8_t                  type;
2822 	uint8_t                  mode;
2823 
2824 	IN6M_LOCK_ASSERT_HELD(inm);
2825 	MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
2826 
2827 	error = 0;
2828 	ifp = inm->in6m_ifp;
2829 	is_filter_list_change = 0;
2830 	m = NULL;
2831 	m0 = NULL;
2832 	m0srcs = 0;
2833 	msrcs = 0;
2834 	nbytes = 0;
2835 	nims = NULL;
2836 	record_has_sources = 1;
2837 	pmr = NULL;
2838 	type = MLD_DO_NOTHING;
2839 	mode = (uint8_t)inm->in6m_st[1].iss_fmode;
2840 
2841 	/*
2842 	 * If we did not transition out of ASM mode during t0->t1,
2843 	 * and there are no source nodes to process, we can skip
2844 	 * the generation of source records.
2845 	 */
2846 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2847 	    inm->in6m_nsrc == 0) {
2848 		record_has_sources = 0;
2849 	}
2850 
2851 	if (is_state_change) {
2852 		/*
2853 		 * Queue a state change record.
2854 		 * If the mode did not change, and there are non-ASM
2855 		 * listeners or source filters present,
2856 		 * we potentially need to issue two records for the group.
2857 		 * If there are ASM listeners, and there was no filter
2858 		 * mode transition of any kind, do nothing.
2859 		 *
2860 		 * If we are transitioning to MCAST_UNDEFINED, we need
2861 		 * not send any sources. A transition to/from this state is
2862 		 * considered inclusive with some special treatment.
2863 		 *
2864 		 * If we are rewriting initial joins/leaves to use
2865 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2866 		 * we need to send sources in all cases.
2867 		 */
2868 		if (mode != inm->in6m_st[0].iss_fmode) {
2869 			if (mode == MCAST_EXCLUDE) {
2870 				MLD_PRINTF(("%s: change to EXCLUDE\n",
2871 				    __func__));
2872 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2873 			} else {
2874 				MLD_PRINTF(("%s: change to INCLUDE\n",
2875 				    __func__));
2876 				if (use_block_allow) {
2877 					/*
2878 					 * XXX
2879 					 * Here we're interested in state
2880 					 * edges either direction between
2881 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2882 					 * Perhaps we should just check
2883 					 * the group state, rather than
2884 					 * the filter mode.
2885 					 */
2886 					if (mode == MCAST_UNDEFINED) {
2887 						type = MLD_BLOCK_OLD_SOURCES;
2888 					} else {
2889 						type = MLD_ALLOW_NEW_SOURCES;
2890 					}
2891 				} else {
2892 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2893 					if (mode == MCAST_UNDEFINED) {
2894 						record_has_sources = 0;
2895 					}
2896 				}
2897 			}
2898 		} else {
2899 			if (record_has_sources) {
2900 				is_filter_list_change = 1;
2901 			} else {
2902 				type = MLD_DO_NOTHING;
2903 			}
2904 		}
2905 	} else {
2906 		/*
2907 		 * Queue a current state record.
2908 		 */
2909 		if (mode == MCAST_EXCLUDE) {
2910 			type = MLD_MODE_IS_EXCLUDE;
2911 		} else if (mode == MCAST_INCLUDE) {
2912 			type = MLD_MODE_IS_INCLUDE;
2913 			VERIFY(inm->in6m_st[1].iss_asm == 0);
2914 		}
2915 	}
2916 
2917 	/*
2918 	 * Generate the filter list changes using a separate function.
2919 	 */
2920 	if (is_filter_list_change) {
2921 		return mld_v2_enqueue_filter_change(ifq, inm);
2922 	}
2923 
2924 	if (type == MLD_DO_NOTHING) {
2925 		MLD_PRINTF(("%s: nothing to do for %s/%s\n",
2926 		    __func__, ip6_sprintf(&inm->in6m_addr),
2927 		    if_name(inm->in6m_ifp)));
2928 		return 0;
2929 	}
2930 
2931 	/*
2932 	 * If any sources are present, we must be able to fit at least
2933 	 * one in the trailing space of the tail packet's mbuf,
2934 	 * ideally more.
2935 	 */
2936 	minrec0len = sizeof(struct mldv2_record);
2937 	if (record_has_sources) {
2938 		minrec0len += sizeof(struct in6_addr);
2939 	}
2940 	MLD_PRINTF(("%s: queueing %s for %s/%s\n", __func__,
2941 	    mld_rec_type_to_str(type),
2942 	    ip6_sprintf(&inm->in6m_addr),
2943 	    if_name(inm->in6m_ifp)));
2944 
2945 	/*
2946 	 * Check if we have a packet in the tail of the queue for this
2947 	 * group into which the first group record for this group will fit.
2948 	 * Otherwise allocate a new packet.
2949 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2950 	 * Note: Group records for G/GSR query responses MUST be sent
2951 	 * in their own packet.
2952 	 */
2953 	m0 = ifq->ifq_tail;
2954 	if (!is_group_query &&
2955 	    m0 != NULL &&
2956 	    (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2957 	    (m0->m_pkthdr.len + minrec0len) <
2958 	    (ifp->if_mtu - MLD_MTUSPACE)) {
2959 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2960 		    sizeof(struct mldv2_record)) /
2961 		    sizeof(struct in6_addr);
2962 		m = m0;
2963 		MLD_PRINTF(("%s: use existing packet\n", __func__));
2964 	} else {
2965 		if (IF_QFULL(ifq)) {
2966 			MLD_PRINTF(("%s: outbound queue full\n", __func__));
2967 			return -ENOMEM;
2968 		}
2969 		m = NULL;
2970 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2971 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2972 		if (!is_state_change && !is_group_query) {
2973 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2974 		}
2975 		if (m == NULL) {
2976 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2977 		}
2978 		if (m == NULL) {
2979 			return -ENOMEM;
2980 		}
2981 
2982 		mld_save_context(m, ifp);
2983 
2984 		MLD_PRINTF(("%s: allocated first packet\n", __func__));
2985 	}
2986 
2987 	/*
2988 	 * Append group record.
2989 	 * If we have sources, we don't know how many yet.
2990 	 */
2991 	mr.mr_type = type;
2992 	mr.mr_datalen = 0;
2993 	mr.mr_numsrc = 0;
2994 	mr.mr_addr = inm->in6m_addr;
2995 	in6_clearscope(&mr.mr_addr);
2996 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2997 		if (m != m0) {
2998 			m_freem(m);
2999 		}
3000 		MLD_PRINTF(("%s: m_append() failed.\n", __func__));
3001 		return -ENOMEM;
3002 	}
3003 	nbytes += sizeof(struct mldv2_record);
3004 
3005 	/*
3006 	 * Append as many sources as will fit in the first packet.
3007 	 * If we are appending to a new packet, the chain allocation
3008 	 * may potentially use clusters; use m_getptr() in this case.
3009 	 * If we are appending to an existing packet, we need to obtain
3010 	 * a pointer to the group record after m_append(), in case a new
3011 	 * mbuf was allocated.
3012 	 *
3013 	 * Only append sources which are in-mode at t1. If we are
3014 	 * transitioning to MCAST_UNDEFINED state on the group, and
3015 	 * use_block_allow is zero, do not include source entries.
3016 	 * Otherwise, we need to include this source in the report.
3017 	 *
3018 	 * Only report recorded sources in our filter set when responding
3019 	 * to a group-source query.
3020 	 */
3021 	if (record_has_sources) {
3022 		if (m == m0) {
3023 			md = m_last(m);
3024 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
3025 			    md->m_len - nbytes);
3026 		} else {
3027 			md = m_getptr(m, 0, &off);
3028 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
3029 			    off);
3030 		}
3031 		msrcs = 0;
3032 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
3033 		    nims) {
3034 			MLD_PRINTF(("%s: visit node %s\n", __func__,
3035 			    ip6_sprintf(&ims->im6s_addr)));
3036 			now = im6s_get_mode(inm, ims, 1);
3037 			MLD_PRINTF(("%s: node is %d\n", __func__, now));
3038 			if ((now != mode) ||
3039 			    (now == mode &&
3040 			    (!use_block_allow && mode == MCAST_UNDEFINED))) {
3041 				MLD_PRINTF(("%s: skip node\n", __func__));
3042 				continue;
3043 			}
3044 			if (is_source_query && ims->im6s_stp == 0) {
3045 				MLD_PRINTF(("%s: skip unrecorded node\n",
3046 				    __func__));
3047 				continue;
3048 			}
3049 			MLD_PRINTF(("%s: append node\n", __func__));
3050 			if (!m_append(m, sizeof(struct in6_addr),
3051 			    (void *)&ims->im6s_addr)) {
3052 				if (m != m0) {
3053 					m_freem(m);
3054 				}
3055 				MLD_PRINTF(("%s: m_append() failed.\n",
3056 				    __func__));
3057 				return -ENOMEM;
3058 			}
3059 			nbytes += sizeof(struct in6_addr);
3060 			++msrcs;
3061 			if (msrcs == m0srcs) {
3062 				break;
3063 			}
3064 		}
3065 		MLD_PRINTF(("%s: msrcs is %d this packet\n", __func__,
3066 		    msrcs));
3067 		pmr->mr_numsrc = htons((uint16_t)msrcs);
3068 		nbytes += (msrcs * sizeof(struct in6_addr));
3069 	}
3070 
3071 	if (is_source_query && msrcs == 0) {
3072 		MLD_PRINTF(("%s: no recorded sources to report\n", __func__));
3073 		if (m != m0) {
3074 			m_freem(m);
3075 		}
3076 		return 0;
3077 	}
3078 
3079 	/*
3080 	 * We are good to go with first packet.
3081 	 */
3082 	if (m != m0) {
3083 		MLD_PRINTF(("%s: enqueueing first packet\n", __func__));
3084 		m->m_pkthdr.vt_nrecs = 1;
3085 		IF_ENQUEUE(ifq, m);
3086 	} else {
3087 		m->m_pkthdr.vt_nrecs++;
3088 	}
3089 	/*
3090 	 * No further work needed if no source list in packet(s).
3091 	 */
3092 	if (!record_has_sources) {
3093 		return nbytes;
3094 	}
3095 
3096 	/*
3097 	 * Whilst sources remain to be announced, we need to allocate
3098 	 * a new packet and fill out as many sources as will fit.
3099 	 * Always try for a cluster first.
3100 	 */
3101 	while (nims != NULL) {
3102 		if (IF_QFULL(ifq)) {
3103 			MLD_PRINTF(("%s: outbound queue full\n", __func__));
3104 			return -ENOMEM;
3105 		}
3106 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3107 		if (m == NULL) {
3108 			m = m_gethdr(M_DONTWAIT, MT_DATA);
3109 		}
3110 		if (m == NULL) {
3111 			return -ENOMEM;
3112 		}
3113 		mld_save_context(m, ifp);
3114 		md = m_getptr(m, 0, &off);
3115 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
3116 		MLD_PRINTF(("%s: allocated next packet\n", __func__));
3117 
3118 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
3119 			if (m != m0) {
3120 				m_freem(m);
3121 			}
3122 			MLD_PRINTF(("%s: m_append() failed.\n", __func__));
3123 			return -ENOMEM;
3124 		}
3125 		m->m_pkthdr.vt_nrecs = 1;
3126 		nbytes += sizeof(struct mldv2_record);
3127 
3128 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3129 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
3130 
3131 		msrcs = 0;
3132 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
3133 			MLD_PRINTF(("%s: visit node %s\n",
3134 			    __func__, ip6_sprintf(&ims->im6s_addr)));
3135 			now = im6s_get_mode(inm, ims, 1);
3136 			if ((now != mode) ||
3137 			    (now == mode &&
3138 			    (!use_block_allow && mode == MCAST_UNDEFINED))) {
3139 				MLD_PRINTF(("%s: skip node\n", __func__));
3140 				continue;
3141 			}
3142 			if (is_source_query && ims->im6s_stp == 0) {
3143 				MLD_PRINTF(("%s: skip unrecorded node\n",
3144 				    __func__));
3145 				continue;
3146 			}
3147 			MLD_PRINTF(("%s: append node\n", __func__));
3148 			if (!m_append(m, sizeof(struct in6_addr),
3149 			    (void *)&ims->im6s_addr)) {
3150 				if (m != m0) {
3151 					m_freem(m);
3152 				}
3153 				MLD_PRINTF(("%s: m_append() failed.\n",
3154 				    __func__));
3155 				return -ENOMEM;
3156 			}
3157 			++msrcs;
3158 			if (msrcs == m0srcs) {
3159 				break;
3160 			}
3161 		}
3162 		pmr->mr_numsrc = htons((uint16_t)msrcs);
3163 		nbytes += (msrcs * sizeof(struct in6_addr));
3164 
3165 		MLD_PRINTF(("%s: enqueueing next packet\n", __func__));
3166 		IF_ENQUEUE(ifq, m);
3167 	}
3168 
3169 	return nbytes;
3170 }
3171 
3172 /*
3173  * Type used to mark record pass completion.
3174  * We exploit the fact we can cast to this easily from the
3175  * current filter modes on each ip_msource node.
3176  */
3177 typedef enum {
3178 	REC_NONE = 0x00,        /* MCAST_UNDEFINED */
3179 	REC_ALLOW = 0x01,       /* MCAST_INCLUDE */
3180 	REC_BLOCK = 0x02,       /* MCAST_EXCLUDE */
3181 	REC_FULL = REC_ALLOW | REC_BLOCK
3182 } rectype_t;
3183 
3184 /*
3185  * Enqueue an MLDv2 filter list change to the given output queue.
3186  *
3187  * Source list filter state is held in an RB-tree. When the filter list
3188  * for a group is changed without changing its mode, we need to compute
3189  * the deltas between T0 and T1 for each source in the filter set,
3190  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3191  *
3192  * As we may potentially queue two record types, and the entire R-B tree
3193  * needs to be walked at once, we break this out into its own function
3194  * so we can generate a tightly packed queue of packets.
3195  *
3196  * XXX This could be written to only use one tree walk, although that makes
3197  * serializing into the mbuf chains a bit harder. For now we do two walks
3198  * which makes things easier on us, and it may or may not be harder on
3199  * the L2 cache.
3200  *
3201  * If successful the size of all data appended to the queue is returned,
3202  * otherwise an error code less than zero is returned, or zero if
3203  * no record(s) were appended.
3204  */
3205 static int
mld_v2_enqueue_filter_change(struct ifqueue * ifq,struct in6_multi * inm)3206 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
3207 {
3208 	static const int MINRECLEN =
3209 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
3210 	struct ifnet            *ifp;
3211 	struct mldv2_record      mr;
3212 	struct mldv2_record     *pmr;
3213 	struct ip6_msource      *ims, *nims;
3214 	struct mbuf             *m, *m0, *md;
3215 	int                      m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3216 	int                      nallow, nblock;
3217 	uint8_t                  mode, now, then;
3218 	rectype_t                crt, drt, nrt;
3219 
3220 	IN6M_LOCK_ASSERT_HELD(inm);
3221 
3222 	if (inm->in6m_nsrc == 0 ||
3223 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) {
3224 		return 0;
3225 	}
3226 
3227 	ifp = inm->in6m_ifp;                    /* interface */
3228 	mode = (uint8_t)inm->in6m_st[1].iss_fmode;       /* filter mode at t1 */
3229 	crt = REC_NONE; /* current group record type */
3230 	drt = REC_NONE; /* mask of completed group record types */
3231 	nrt = REC_NONE; /* record type for current node */
3232 	m0srcs = 0;     /* # source which will fit in current mbuf chain */
3233 	npbytes = 0;    /* # of bytes appended this packet */
3234 	nbytes = 0;     /* # of bytes appended to group's state-change queue */
3235 	rsrcs = 0;      /* # sources encoded in current record */
3236 	schanged = 0;   /* # nodes encoded in overall filter change */
3237 	nallow = 0;     /* # of source entries in ALLOW_NEW */
3238 	nblock = 0;     /* # of source entries in BLOCK_OLD */
3239 	nims = NULL;    /* next tree node pointer */
3240 
3241 	/*
3242 	 * For each possible filter record mode.
3243 	 * The first kind of source we encounter tells us which
3244 	 * is the first kind of record we start appending.
3245 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3246 	 * as the inverse of the group's filter mode.
3247 	 */
3248 	while (drt != REC_FULL) {
3249 		do {
3250 			m0 = ifq->ifq_tail;
3251 			if (m0 != NULL &&
3252 			    (m0->m_pkthdr.vt_nrecs + 1 <=
3253 			    MLD_V2_REPORT_MAXRECS) &&
3254 			    (m0->m_pkthdr.len + MINRECLEN) <
3255 			    (ifp->if_mtu - MLD_MTUSPACE)) {
3256 				m = m0;
3257 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3258 				    sizeof(struct mldv2_record)) /
3259 				    sizeof(struct in6_addr);
3260 				MLD_PRINTF(("%s: use previous packet\n",
3261 				    __func__));
3262 			} else {
3263 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3264 				if (m == NULL) {
3265 					m = m_gethdr(M_DONTWAIT, MT_DATA);
3266 				}
3267 				if (m == NULL) {
3268 					MLD_PRINTF(("%s: m_get*() failed\n",
3269 					    __func__));
3270 					return -ENOMEM;
3271 				}
3272 				m->m_pkthdr.vt_nrecs = 0;
3273 				mld_save_context(m, ifp);
3274 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3275 				    sizeof(struct mldv2_record)) /
3276 				    sizeof(struct in6_addr);
3277 				npbytes = 0;
3278 				MLD_PRINTF(("%s: allocated new packet\n",
3279 				    __func__));
3280 			}
3281 			/*
3282 			 * Append the MLD group record header to the
3283 			 * current packet's data area.
3284 			 * Recalculate pointer to free space for next
3285 			 * group record, in case m_append() allocated
3286 			 * a new mbuf or cluster.
3287 			 */
3288 			memset(&mr, 0, sizeof(mr));
3289 			mr.mr_addr = inm->in6m_addr;
3290 			in6_clearscope(&mr.mr_addr);
3291 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
3292 				if (m != m0) {
3293 					m_freem(m);
3294 				}
3295 				MLD_PRINTF(("%s: m_append() failed\n",
3296 				    __func__));
3297 				return -ENOMEM;
3298 			}
3299 			npbytes += sizeof(struct mldv2_record);
3300 			if (m != m0) {
3301 				/* new packet; offset in chain */
3302 				md = m_getptr(m, npbytes -
3303 				    sizeof(struct mldv2_record), &off);
3304 				pmr = (struct mldv2_record *)(mtod(md,
3305 				    uint8_t *) + off);
3306 			} else {
3307 				/* current packet; offset from last append */
3308 				md = m_last(m);
3309 				pmr = (struct mldv2_record *)(mtod(md,
3310 				    uint8_t *) + md->m_len -
3311 				    sizeof(struct mldv2_record));
3312 			}
3313 			/*
3314 			 * Begin walking the tree for this record type
3315 			 * pass, or continue from where we left off
3316 			 * previously if we had to allocate a new packet.
3317 			 * Only report deltas in-mode at t1.
3318 			 * We need not report included sources as allowed
3319 			 * if we are in inclusive mode on the group,
3320 			 * however the converse is not true.
3321 			 */
3322 			rsrcs = 0;
3323 			if (nims == NULL) {
3324 				nims = RB_MIN(ip6_msource_tree,
3325 				    &inm->in6m_srcs);
3326 			}
3327 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
3328 				MLD_PRINTF(("%s: visit node %s\n", __func__,
3329 				    ip6_sprintf(&ims->im6s_addr)));
3330 				now = im6s_get_mode(inm, ims, 1);
3331 				then = im6s_get_mode(inm, ims, 0);
3332 				MLD_PRINTF(("%s: mode: t0 %d, t1 %d\n",
3333 				    __func__, then, now));
3334 				if (now == then) {
3335 					MLD_PRINTF(("%s: skip unchanged\n",
3336 					    __func__));
3337 					continue;
3338 				}
3339 				if (mode == MCAST_EXCLUDE &&
3340 				    now == MCAST_INCLUDE) {
3341 					MLD_PRINTF(("%s: skip IN src on EX "
3342 					    "group\n", __func__));
3343 					continue;
3344 				}
3345 				nrt = (rectype_t)now;
3346 				if (nrt == REC_NONE) {
3347 					nrt = (rectype_t)(~mode & REC_FULL);
3348 				}
3349 				if (schanged++ == 0) {
3350 					crt = nrt;
3351 				} else if (crt != nrt) {
3352 					continue;
3353 				}
3354 				if (!m_append(m, sizeof(struct in6_addr),
3355 				    (void *)&ims->im6s_addr)) {
3356 					if (m != m0) {
3357 						m_freem(m);
3358 					}
3359 					MLD_PRINTF(("%s: m_append() failed\n",
3360 					    __func__));
3361 					return -ENOMEM;
3362 				}
3363 				nallow += !!(crt == REC_ALLOW);
3364 				nblock += !!(crt == REC_BLOCK);
3365 				if (++rsrcs == m0srcs) {
3366 					break;
3367 				}
3368 			}
3369 			/*
3370 			 * If we did not append any tree nodes on this
3371 			 * pass, back out of allocations.
3372 			 */
3373 			if (rsrcs == 0) {
3374 				npbytes -= sizeof(struct mldv2_record);
3375 				if (m != m0) {
3376 					MLD_PRINTF(("%s: m_free(m)\n",
3377 					    __func__));
3378 					m_freem(m);
3379 				} else {
3380 					MLD_PRINTF(("%s: m_adj(m, -mr)\n",
3381 					    __func__));
3382 					m_adj(m, -((int)sizeof(
3383 						    struct mldv2_record)));
3384 				}
3385 				continue;
3386 			}
3387 			npbytes += (rsrcs * sizeof(struct in6_addr));
3388 			if (crt == REC_ALLOW) {
3389 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
3390 			} else if (crt == REC_BLOCK) {
3391 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
3392 			}
3393 			pmr->mr_numsrc = htons((uint16_t)rsrcs);
3394 			/*
3395 			 * Count the new group record, and enqueue this
3396 			 * packet if it wasn't already queued.
3397 			 */
3398 			m->m_pkthdr.vt_nrecs++;
3399 			if (m != m0) {
3400 				IF_ENQUEUE(ifq, m);
3401 			}
3402 			nbytes += npbytes;
3403 		} while (nims != NULL);
3404 		drt |= crt;
3405 		crt = (~crt & REC_FULL);
3406 	}
3407 
3408 	MLD_PRINTF(("%s: queued %d ALLOW_NEW, %d BLOCK_OLD\n", __func__,
3409 	    nallow, nblock));
3410 
3411 	return nbytes;
3412 }
3413 
3414 static int
mld_v2_merge_state_changes(struct in6_multi * inm,struct ifqueue * ifscq)3415 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
3416 {
3417 	struct ifqueue  *gq;
3418 	struct mbuf     *m;             /* pending state-change */
3419 	struct mbuf     *m0;            /* copy of pending state-change */
3420 	struct mbuf     *mt;            /* last state-change in packet */
3421 	struct mbuf     *n;
3422 	int              docopy, domerge;
3423 	u_int            recslen;
3424 
3425 	IN6M_LOCK_ASSERT_HELD(inm);
3426 
3427 	docopy = 0;
3428 	domerge = 0;
3429 	recslen = 0;
3430 
3431 	/*
3432 	 * If there are further pending retransmissions, make a writable
3433 	 * copy of each queued state-change message before merging.
3434 	 */
3435 	if (inm->in6m_scrv > 0) {
3436 		docopy = 1;
3437 	}
3438 
3439 	gq = &inm->in6m_scq;
3440 #ifdef MLD_DEBUG
3441 	if (gq->ifq_head == NULL) {
3442 		MLD_PRINTF(("%s: WARNING: queue for inm 0x%llx is empty\n",
3443 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm)));
3444 	}
3445 #endif
3446 
3447 	/*
3448 	 * Use IF_REMQUEUE() instead of IF_DEQUEUE() below, since the
3449 	 * packet might not always be at the head of the ifqueue.
3450 	 */
3451 	m = gq->ifq_head;
3452 	while (m != NULL) {
3453 		/*
3454 		 * Only merge the report into the current packet if
3455 		 * there is sufficient space to do so; an MLDv2 report
3456 		 * packet may only contain 65,535 group records.
3457 		 * Always use a simple mbuf chain concatentation to do this,
3458 		 * as large state changes for single groups may have
3459 		 * allocated clusters.
3460 		 */
3461 		domerge = 0;
3462 		mt = ifscq->ifq_tail;
3463 		if (mt != NULL) {
3464 			recslen = m_length(m);
3465 
3466 			if ((mt->m_pkthdr.vt_nrecs +
3467 			    m->m_pkthdr.vt_nrecs <=
3468 			    MLD_V2_REPORT_MAXRECS) &&
3469 			    (mt->m_pkthdr.len + recslen <=
3470 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) {
3471 				domerge = 1;
3472 			}
3473 		}
3474 
3475 		if (!domerge && IF_QFULL(gq)) {
3476 			MLD_PRINTF(("%s: outbound queue full, skipping whole "
3477 			    "packet 0x%llx\n", __func__,
3478 			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3479 			n = m->m_nextpkt;
3480 			if (!docopy) {
3481 				IF_REMQUEUE(gq, m);
3482 				m_freem(m);
3483 			}
3484 			m = n;
3485 			continue;
3486 		}
3487 
3488 		if (!docopy) {
3489 			MLD_PRINTF(("%s: dequeueing 0x%llx\n", __func__,
3490 			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3491 			n = m->m_nextpkt;
3492 			IF_REMQUEUE(gq, m);
3493 			m0 = m;
3494 			m = n;
3495 		} else {
3496 			MLD_PRINTF(("%s: copying 0x%llx\n", __func__,
3497 			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3498 			m0 = m_dup(m, M_NOWAIT);
3499 			if (m0 == NULL) {
3500 				return ENOMEM;
3501 			}
3502 			m0->m_nextpkt = NULL;
3503 			m = m->m_nextpkt;
3504 		}
3505 
3506 		if (!domerge) {
3507 			MLD_PRINTF(("%s: queueing 0x%llx to ifscq 0x%llx)\n",
3508 			    __func__, (uint64_t)VM_KERNEL_ADDRPERM(m0),
3509 			    (uint64_t)VM_KERNEL_ADDRPERM(ifscq)));
3510 			IF_ENQUEUE(ifscq, m0);
3511 		} else {
3512 			struct mbuf *mtl;       /* last mbuf of packet mt */
3513 
3514 			MLD_PRINTF(("%s: merging 0x%llx with ifscq tail "
3515 			    "0x%llx)\n", __func__,
3516 			    (uint64_t)VM_KERNEL_ADDRPERM(m0),
3517 			    (uint64_t)VM_KERNEL_ADDRPERM(mt)));
3518 
3519 			mtl = m_last(mt);
3520 			m0->m_flags &= ~M_PKTHDR;
3521 			mt->m_pkthdr.len += recslen;
3522 			mt->m_pkthdr.vt_nrecs +=
3523 			    m0->m_pkthdr.vt_nrecs;
3524 
3525 			mtl->m_next = m0;
3526 		}
3527 	}
3528 
3529 	return 0;
3530 }
3531 
3532 /*
3533  * Respond to a pending MLDv2 General Query.
3534  */
3535 static uint32_t
mld_v2_dispatch_general_query(struct mld_ifinfo * mli)3536 mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
3537 {
3538 	struct ifnet            *ifp;
3539 	struct in6_multi        *inm;
3540 	struct in6_multistep    step;
3541 	int                      retval;
3542 
3543 	MLI_LOCK_ASSERT_HELD(mli);
3544 
3545 	VERIFY(mli->mli_version == MLD_VERSION_2);
3546 
3547 	ifp = mli->mli_ifp;
3548 	MLI_UNLOCK(mli);
3549 
3550 	in6_multihead_lock_shared();
3551 	IN6_FIRST_MULTI(step, inm);
3552 	while (inm != NULL) {
3553 		IN6M_LOCK(inm);
3554 		if (inm->in6m_ifp != ifp) {
3555 			goto next;
3556 		}
3557 
3558 		switch (inm->in6m_state) {
3559 		case MLD_NOT_MEMBER:
3560 		case MLD_SILENT_MEMBER:
3561 			break;
3562 		case MLD_REPORTING_MEMBER:
3563 		case MLD_IDLE_MEMBER:
3564 		case MLD_LAZY_MEMBER:
3565 		case MLD_SLEEPING_MEMBER:
3566 		case MLD_AWAKENING_MEMBER:
3567 			inm->in6m_state = MLD_REPORTING_MEMBER;
3568 			MLI_LOCK(mli);
3569 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3570 			    inm, 0, 0, 0, 0);
3571 			MLI_UNLOCK(mli);
3572 			MLD_PRINTF(("%s: enqueue record = %d\n",
3573 			    __func__, retval));
3574 			break;
3575 		case MLD_G_QUERY_PENDING_MEMBER:
3576 		case MLD_SG_QUERY_PENDING_MEMBER:
3577 		case MLD_LEAVING_MEMBER:
3578 			break;
3579 		}
3580 next:
3581 		IN6M_UNLOCK(inm);
3582 		IN6_NEXT_MULTI(step, inm);
3583 	}
3584 	in6_multihead_lock_done();
3585 
3586 	MLI_LOCK(mli);
3587 	mld_dispatch_queue_locked(mli, &mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3588 	MLI_LOCK_ASSERT_HELD(mli);
3589 
3590 	/*
3591 	 * Slew transmission of bursts over 1 second intervals.
3592 	 */
3593 	if (mli->mli_gq.ifq_head != NULL) {
3594 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3595 			MLD_RESPONSE_BURST_INTERVAL);
3596 	}
3597 
3598 	return mli->mli_v2_timer;
3599 }
3600 
3601 /*
3602  * Transmit the next pending message in the output queue.
3603  *
3604  * Must not be called with in6m_lockm or mli_lock held.
3605  */
3606 static void
mld_dispatch_packet(struct mbuf * m)3607 mld_dispatch_packet(struct mbuf *m)
3608 {
3609 	struct ip6_moptions     *im6o;
3610 	struct ifnet            *ifp;
3611 	struct ifnet            *oifp = NULL;
3612 	struct mbuf             *m0;
3613 	struct mbuf             *md;
3614 	struct ip6_hdr          *ip6;
3615 	struct mld_hdr          *mld;
3616 	int                      error;
3617 	int                      off;
3618 	int                      type;
3619 
3620 	MLD_PRINTF(("%s: transmit 0x%llx\n", __func__,
3621 	    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3622 
3623 	/*
3624 	 * Check if the ifnet is still attached.
3625 	 */
3626 	ifp = mld_restore_context(m);
3627 	if (ifp == NULL || !ifnet_is_attached(ifp, 0)) {
3628 		MLD_PRINTF(("%s: dropped 0x%llx as ifindex %u went away.\n",
3629 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(m),
3630 		    (u_int)if_index));
3631 		m_freem(m);
3632 		ip6stat.ip6s_noroute++;
3633 		return;
3634 	}
3635 
3636 	im6o = ip6_allocmoptions(Z_WAITOK);
3637 	if (im6o == NULL) {
3638 		m_freem(m);
3639 		return;
3640 	}
3641 
3642 	im6o->im6o_multicast_hlim  = 1;
3643 	im6o->im6o_multicast_loop = 0;
3644 	im6o->im6o_multicast_ifp = ifp;
3645 
3646 	if (m->m_flags & M_MLDV1) {
3647 		m0 = m;
3648 	} else {
3649 		m0 = mld_v2_encap_report(ifp, m);
3650 		if (m0 == NULL) {
3651 			MLD_PRINTF(("%s: dropped 0x%llx\n", __func__,
3652 			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3653 			/*
3654 			 * mld_v2_encap_report() has already freed our mbuf.
3655 			 */
3656 			IM6O_REMREF(im6o);
3657 			ip6stat.ip6s_odropped++;
3658 			return;
3659 		}
3660 	}
3661 
3662 	mld_scrub_context(m0);
3663 	m->m_flags &= ~(M_PROTOFLAGS);
3664 	m0->m_pkthdr.rcvif = lo_ifp;
3665 
3666 	ip6 = mtod(m0, struct ip6_hdr *);
3667 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);
3668 	ip6_output_setdstifscope(m0, ifp->if_index, NULL);
3669 	/*
3670 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3671 	 * so we can bump the stats.
3672 	 */
3673 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3674 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3675 	type = mld->mld_type;
3676 
3677 	if (ifp->if_eflags & IFEF_TXSTART) {
3678 		/*
3679 		 * Use control service class if the outgoing
3680 		 * interface supports transmit-start model.
3681 		 */
3682 		(void) m_set_service_class(m0, MBUF_SC_CTL);
3683 	}
3684 
3685 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, im6o,
3686 	    &oifp, NULL);
3687 
3688 	IM6O_REMREF(im6o);
3689 
3690 	if (error) {
3691 		MLD_PRINTF(("%s: ip6_output(0x%llx) = %d\n", __func__,
3692 		    (uint64_t)VM_KERNEL_ADDRPERM(m0), error));
3693 		if (oifp != NULL) {
3694 			ifnet_release(oifp);
3695 		}
3696 		return;
3697 	}
3698 
3699 	icmp6stat.icp6s_outhist[type]++;
3700 	if (oifp != NULL) {
3701 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3702 		switch (type) {
3703 		case MLD_LISTENER_REPORT:
3704 		case MLDV2_LISTENER_REPORT:
3705 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3706 			break;
3707 		case MLD_LISTENER_DONE:
3708 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3709 			break;
3710 		}
3711 		ifnet_release(oifp);
3712 	}
3713 }
3714 
3715 /*
3716  * Encapsulate an MLDv2 report.
3717  *
3718  * KAME IPv6 requires that hop-by-hop options be passed separately,
3719  * and that the IPv6 header be prepended in a separate mbuf.
3720  *
3721  * Returns a pointer to the new mbuf chain head, or NULL if the
3722  * allocation failed.
3723  */
3724 static struct mbuf *
mld_v2_encap_report(struct ifnet * ifp,struct mbuf * m)3725 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3726 {
3727 	struct mbuf             *mh;
3728 	struct mldv2_report     *mld;
3729 	struct ip6_hdr          *ip6;
3730 	struct in6_ifaddr       *ia;
3731 	int                      mldreclen;
3732 
3733 	VERIFY(m->m_flags & M_PKTHDR);
3734 
3735 	/*
3736 	 * RFC3590: OK to send as :: or tentative during DAD.
3737 	 */
3738 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
3739 	if (ia == NULL) {
3740 		MLD_PRINTF(("%s: warning: ia is NULL\n", __func__));
3741 	}
3742 
3743 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3744 	if (mh == NULL) {
3745 		if (ia != NULL) {
3746 			IFA_REMREF(&ia->ia_ifa);
3747 		}
3748 		m_freem(m);
3749 		return NULL;
3750 	}
3751 	MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3752 
3753 	mldreclen = m_length(m);
3754 	MLD_PRINTF(("%s: mldreclen is %d\n", __func__, mldreclen));
3755 
3756 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3757 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3758 	    sizeof(struct mldv2_report) + mldreclen;
3759 
3760 	ip6 = mtod(mh, struct ip6_hdr *);
3761 	ip6->ip6_flow = 0;
3762 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3763 	ip6->ip6_vfc |= IPV6_VERSION;
3764 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3765 	if (ia != NULL) {
3766 		IFA_LOCK(&ia->ia_ifa);
3767 	}
3768 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3769 	ip6_output_setsrcifscope(mh, IFSCOPE_NONE, ia);
3770 
3771 	if (ia != NULL) {
3772 		IFA_UNLOCK(&ia->ia_ifa);
3773 		IFA_REMREF(&ia->ia_ifa);
3774 		ia = NULL;
3775 	}
3776 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3777 	ip6_output_setdstifscope(mh, ifp->if_index, NULL);
3778 	/* scope ID will be set in netisr */
3779 
3780 	mld = (struct mldv2_report *)(ip6 + 1);
3781 	mld->mld_type = MLDV2_LISTENER_REPORT;
3782 	mld->mld_code = 0;
3783 	mld->mld_cksum = 0;
3784 	mld->mld_v2_reserved = 0;
3785 	mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3786 	m->m_pkthdr.vt_nrecs = 0;
3787 	m->m_flags &= ~M_PKTHDR;
3788 
3789 	mh->m_next = m;
3790 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3791 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3792 	return mh;
3793 }
3794 
3795 #ifdef MLD_DEBUG
3796 static const char *
mld_rec_type_to_str(const int type)3797 mld_rec_type_to_str(const int type)
3798 {
3799 	switch (type) {
3800 	case MLD_CHANGE_TO_EXCLUDE_MODE:
3801 		return "TO_EX";
3802 	case MLD_CHANGE_TO_INCLUDE_MODE:
3803 		return "TO_IN";
3804 	case MLD_MODE_IS_EXCLUDE:
3805 		return "MODE_EX";
3806 	case MLD_MODE_IS_INCLUDE:
3807 		return "MODE_IN";
3808 	case MLD_ALLOW_NEW_SOURCES:
3809 		return "ALLOW_NEW";
3810 	case MLD_BLOCK_OLD_SOURCES:
3811 		return "BLOCK_OLD";
3812 	default:
3813 		break;
3814 	}
3815 	return "unknown";
3816 }
3817 #endif
3818 
3819 void
mld_init(void)3820 mld_init(void)
3821 {
3822 	MLD_PRINTF(("%s: initializing\n", __func__));
3823 
3824 	ip6_initpktopts(&mld_po);
3825 	mld_po.ip6po_hlim = 1;
3826 	mld_po.ip6po_hbh = &mld_ra.hbh;
3827 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3828 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3829 	LIST_INIT(&mli_head);
3830 }
3831