xref: /xnu-10002.61.3/bsd/netinet6/mld6.c (revision 0f4c859e951fba394238ab619495c4e1d54d0f34)
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 #include <os/log.h>
131 
132 /* Lock group and attribute for mld_mtx */
133 static LCK_ATTR_DECLARE(mld_mtx_attr, 0, 0);
134 static LCK_GRP_DECLARE(mld_mtx_grp, "mld_mtx");
135 
136 /*
137  * Locking and reference counting:
138  *
139  * mld_mtx mainly protects mli_head.  In cases where both mld_mtx and
140  * in6_multihead_lock must be held, the former must be acquired first in order
141  * to maintain lock ordering.  It is not a requirement that mld_mtx be
142  * acquired first before in6_multihead_lock, but in case both must be acquired
143  * in succession, the correct lock ordering must be followed.
144  *
145  * Instead of walking the if_multiaddrs list at the interface and returning
146  * the ifma_protospec value of a matching entry, we search the global list
147  * of in6_multi records and find it that way; this is done with in6_multihead
148  * lock held.  Doing so avoids the race condition issues that many other BSDs
149  * suffer from (therefore in our implementation, ifma_protospec will never be
150  * NULL for as long as the in6_multi is valid.)
151  *
152  * The above creates a requirement for the in6_multi to stay in in6_multihead
153  * list even after the final MLD leave (in MLDv2 mode) until no longer needs
154  * be retransmitted (this is not required for MLDv1.)  In order to handle
155  * this, the request and reference counts of the in6_multi are bumped up when
156  * the state changes to MLD_LEAVING_MEMBER, and later dropped in the timeout
157  * handler.  Each in6_multi holds a reference to the underlying mld_ifinfo.
158  *
159  * Thus, the permitted lock order is:
160  *
161  *	mld_mtx, in6_multihead_lock, inm6_lock, mli_lock
162  *
163  * Any may be taken independently, but if any are held at the same time,
164  * the above lock order must be followed.
165  */
166 static LCK_MTX_DECLARE_ATTR(mld_mtx, &mld_mtx_grp, &mld_mtx_attr);
167 
168 SLIST_HEAD(mld_in6m_relhead, in6_multi);
169 
170 static void     mli_initvar(struct mld_ifinfo *, struct ifnet *, int);
171 static struct mld_ifinfo *mli_alloc(zalloc_flags_t);
172 static void     mli_free(struct mld_ifinfo *);
173 static void     mli_delete(const struct ifnet *, struct mld_in6m_relhead *);
174 static void     mld_dispatch_packet(struct mbuf *);
175 static void     mld_final_leave(struct in6_multi *, struct mld_ifinfo *,
176     struct mld_tparams *);
177 static int      mld_handle_state_change(struct in6_multi *, struct mld_ifinfo *,
178     struct mld_tparams *);
179 static int      mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
180     struct mld_tparams *, const int);
181 #ifdef MLD_DEBUG
182 static const char *     mld_rec_type_to_str(const int);
183 #endif
184 static uint32_t mld_set_version(struct mld_ifinfo *, const int);
185 static void     mld_append_relq(struct mld_ifinfo *, struct in6_multi *);
186 static void     mld_flush_relq(struct mld_ifinfo *, struct mld_in6m_relhead *);
187 static void     mld_dispatch_queue_locked(struct mld_ifinfo *, struct ifqueue *, int);
188 static int      mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
189     /*const*/ struct mld_hdr *);
190 static int      mld_v1_input_report(struct ifnet *, struct mbuf *,
191     const struct ip6_hdr *, /*const*/ struct mld_hdr *);
192 static void     mld_v1_process_group_timer(struct in6_multi *, const int);
193 static void     mld_v1_process_querier_timers(struct mld_ifinfo *);
194 static int      mld_v1_transmit_report(struct in6_multi *, const uint8_t);
195 static uint32_t mld_v1_update_group(struct in6_multi *, const int);
196 static void     mld_v2_cancel_link_timers(struct mld_ifinfo *);
197 static uint32_t mld_v2_dispatch_general_query(struct mld_ifinfo *);
198 static struct mbuf *
199 mld_v2_encap_report(struct ifnet *, struct mbuf *);
200 static int      mld_v2_enqueue_filter_change(struct ifqueue *,
201     struct in6_multi *);
202 static int      mld_v2_enqueue_group_record(struct ifqueue *,
203     struct in6_multi *, const int, const int, const int,
204     const int);
205 static int      mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
206     struct mbuf *, const int, const int);
207 static int      mld_v2_merge_state_changes(struct in6_multi *,
208     struct ifqueue *);
209 static void     mld_v2_process_group_timers(struct mld_ifinfo *,
210     struct ifqueue *, struct ifqueue *,
211     struct in6_multi *, const int);
212 static int      mld_v2_process_group_query(struct in6_multi *,
213     int, struct mbuf *, const int);
214 static int      sysctl_mld_gsr SYSCTL_HANDLER_ARGS;
215 static int      sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS;
216 static int      sysctl_mld_v2enable SYSCTL_HANDLER_ARGS;
217 
218 static const uint32_t mld_timeout_delay = 1000; /* in milliseconds */
219 static const uint32_t mld_timeout_leeway = 500; /* in millseconds  */
220 static bool mld_timeout_run;             /* MLD timer is scheduled to run */
221 static bool mld_fast_timeout_run;        /* MLD fast timer is scheduled to run */
222 static void mld_timeout(thread_call_param_t, thread_call_param_t);
223 static void mld_sched_timeout(void);
224 static void mld_sched_fast_timeout(void);
225 
226 /*
227  * Normative references: RFC 2710, RFC 3590, RFC 3810.
228  */
229 static struct timeval mld_gsrdelay = {.tv_sec = 10, .tv_usec = 0};
230 static LIST_HEAD(, mld_ifinfo) mli_head;
231 
232 static int querier_present_timers_running6;
233 static int interface_timers_running6;
234 static int state_change_timers_running6;
235 static int current_state_timers_running6;
236 
237 static unsigned int mld_mli_list_genid;
238 /*
239  * Subsystem lock macros.
240  */
241 #define MLD_LOCK()                      \
242 	lck_mtx_lock(&mld_mtx)
243 #define MLD_LOCK_ASSERT_HELD()          \
244 	LCK_MTX_ASSERT(&mld_mtx, LCK_MTX_ASSERT_OWNED)
245 #define MLD_LOCK_ASSERT_NOTHELD()       \
246 	LCK_MTX_ASSERT(&mld_mtx, LCK_MTX_ASSERT_NOTOWNED)
247 #define MLD_UNLOCK()                    \
248 	lck_mtx_unlock(&mld_mtx)
249 
250 #define MLD_ADD_DETACHED_IN6M(_head, _in6m) {                           \
251 	SLIST_INSERT_HEAD(_head, _in6m, in6m_dtle);                     \
252 }
253 
254 #define MLD_REMOVE_DETACHED_IN6M(_head) {                               \
255 	struct in6_multi *_in6m, *_inm_tmp;                             \
256 	SLIST_FOREACH_SAFE(_in6m, _head, in6m_dtle, _inm_tmp) {         \
257 	        SLIST_REMOVE(_head, _in6m, in6_multi, in6m_dtle);       \
258 	        IN6M_REMREF(_in6m);                                     \
259 	}                                                               \
260 	VERIFY(SLIST_EMPTY(_head));                                     \
261 }
262 
263 static KALLOC_TYPE_DEFINE(mli_zone, struct mld_ifinfo, NET_KT_DEFAULT);
264 
265 SYSCTL_DECL(_net_inet6);        /* Note: Not in any common header. */
266 
267 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_LOCKED, 0,
268     "IPv6 Multicast Listener Discovery");
269 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
270     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
271     &mld_gsrdelay.tv_sec, 0, sysctl_mld_gsr, "I",
272     "Rate limit for MLDv2 Group-and-Source queries in seconds");
273 
274 SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_LOCKED,
275     sysctl_mld_ifinfo, "Per-interface MLDv2 state");
276 
277 static int      mld_v1enable = 1;
278 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW | CTLFLAG_LOCKED,
279     &mld_v1enable, 0, "Enable fallback to MLDv1");
280 
281 static int      mld_v2enable = 1;
282 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, v2enable,
283     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
284     &mld_v2enable, 0, sysctl_mld_v2enable, "I",
285     "Enable MLDv2 (debug purposes only)");
286 
287 static int      mld_use_allow = 1;
288 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW | CTLFLAG_LOCKED,
289     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
290 
291 #ifdef MLD_DEBUG
292 int mld_debug = 0;
293 SYSCTL_INT(_net_inet6_mld, OID_AUTO,
294     debug, CTLFLAG_RW | CTLFLAG_LOCKED, &mld_debug, 0, "");
295 #endif
296 /*
297  * Packed Router Alert option structure declaration.
298  */
299 struct mld_raopt {
300 	struct ip6_hbh          hbh;
301 	struct ip6_opt          pad;
302 	struct ip6_opt_router   ra;
303 } __packed;
304 
305 /*
306  * Router Alert hop-by-hop option header.
307  */
308 static struct mld_raopt mld_ra = {
309 	.hbh = { .ip6h_nxt = 0, .ip6h_len = 0 },
310 	.pad = { .ip6o_type = IP6OPT_PADN, .ip6o_len = 0 },
311 	.ra = {
312 		.ip6or_type = (u_int8_t)IP6OPT_ROUTER_ALERT,
313 		.ip6or_len = (u_int8_t)(IP6OPT_RTALERT_LEN - 2),
314 		.ip6or_value =  {((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
315 			         (IP6OPT_RTALERT_MLD & 0xFF) }
316 	}
317 };
318 static struct ip6_pktopts mld_po;
319 
320 /* Store MLDv2 record count in the module private scratch space */
321 #define vt_nrecs        pkt_mpriv.__mpriv_u.__mpriv32[0].__mpriv32_u.__val16[0]
322 
323 static __inline void
mld_save_context(struct mbuf * m,struct ifnet * ifp)324 mld_save_context(struct mbuf *m, struct ifnet *ifp)
325 {
326 	m->m_pkthdr.rcvif = ifp;
327 }
328 
329 static __inline void
mld_scrub_context(struct mbuf * m)330 mld_scrub_context(struct mbuf *m)
331 {
332 	m->m_pkthdr.rcvif = NULL;
333 }
334 
335 /*
336  * Restore context from a queued output chain.
337  * Return saved ifp.
338  */
339 static __inline struct ifnet *
mld_restore_context(struct mbuf * m)340 mld_restore_context(struct mbuf *m)
341 {
342 	return m->m_pkthdr.rcvif;
343 }
344 
345 /*
346  * Retrieve or set threshold between group-source queries in seconds.
347  */
348 static int
349 sysctl_mld_gsr SYSCTL_HANDLER_ARGS
350 {
351 #pragma unused(arg1, arg2)
352 	int error;
353 	int i;
354 
355 	MLD_LOCK();
356 
357 	i = (int)mld_gsrdelay.tv_sec;
358 
359 	error = sysctl_handle_int(oidp, &i, 0, req);
360 	if (error || !req->newptr) {
361 		goto out_locked;
362 	}
363 
364 	if (i < -1 || i >= 60) {
365 		error = EINVAL;
366 		goto out_locked;
367 	}
368 
369 	mld_gsrdelay.tv_sec = i;
370 
371 out_locked:
372 	MLD_UNLOCK();
373 	return error;
374 }
375 /*
376  * Expose struct mld_ifinfo to userland, keyed by ifindex.
377  * For use by ifmcstat(8).
378  *
379  */
380 static int
381 sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS
382 {
383 #pragma unused(oidp)
384 	int                     *name;
385 	int                      error;
386 	u_int                    namelen;
387 	struct ifnet            *ifp;
388 	struct mld_ifinfo       *mli;
389 	struct mld_ifinfo_u     mli_u;
390 
391 	name = (int *)arg1;
392 	namelen = arg2;
393 
394 	if (req->newptr != USER_ADDR_NULL) {
395 		return EPERM;
396 	}
397 
398 	if (namelen != 1) {
399 		return EINVAL;
400 	}
401 
402 	MLD_LOCK();
403 
404 	if (name[0] <= 0 || name[0] > (u_int)if_index) {
405 		error = ENOENT;
406 		goto out_locked;
407 	}
408 
409 	error = ENOENT;
410 
411 	ifnet_head_lock_shared();
412 	ifp = ifindex2ifnet[name[0]];
413 	ifnet_head_done();
414 	if (ifp == NULL) {
415 		goto out_locked;
416 	}
417 
418 	bzero(&mli_u, sizeof(mli_u));
419 
420 	LIST_FOREACH(mli, &mli_head, mli_link) {
421 		MLI_LOCK(mli);
422 		if (ifp != mli->mli_ifp) {
423 			MLI_UNLOCK(mli);
424 			continue;
425 		}
426 
427 		mli_u.mli_ifindex = mli->mli_ifp->if_index;
428 		mli_u.mli_version = mli->mli_version;
429 		mli_u.mli_v1_timer = mli->mli_v1_timer;
430 		mli_u.mli_v2_timer = mli->mli_v2_timer;
431 		mli_u.mli_flags = mli->mli_flags;
432 		mli_u.mli_rv = mli->mli_rv;
433 		mli_u.mli_qi = mli->mli_qi;
434 		mli_u.mli_qri = mli->mli_qri;
435 		mli_u.mli_uri = mli->mli_uri;
436 		MLI_UNLOCK(mli);
437 
438 		error = SYSCTL_OUT(req, &mli_u, sizeof(mli_u));
439 		break;
440 	}
441 
442 out_locked:
443 	MLD_UNLOCK();
444 	return error;
445 }
446 
447 static int
448 sysctl_mld_v2enable SYSCTL_HANDLER_ARGS
449 {
450 #pragma unused(arg1, arg2)
451 	int error;
452 	int i;
453 	struct mld_ifinfo *mli;
454 	struct mld_tparams mtp = { .qpt = 0, .it = 0, .cst = 0, .sct = 0 };
455 
456 	MLD_LOCK();
457 
458 	i = mld_v2enable;
459 
460 	error = sysctl_handle_int(oidp, &i, 0, req);
461 	if (error || !req->newptr) {
462 		goto out_locked;
463 	}
464 
465 	if (i < 0 || i > 1) {
466 		error = EINVAL;
467 		goto out_locked;
468 	}
469 
470 	mld_v2enable = i;
471 	/*
472 	 * If we enabled v2, the state transition will take care of upgrading
473 	 * the MLD version back to v2. Otherwise, we have to explicitly
474 	 * downgrade. Note that this functionality is to be used for debugging.
475 	 */
476 	if (mld_v2enable == 1) {
477 		goto out_locked;
478 	}
479 
480 	LIST_FOREACH(mli, &mli_head, mli_link) {
481 		MLI_LOCK(mli);
482 		if (mld_set_version(mli, MLD_VERSION_1) > 0) {
483 			mtp.qpt = 1;
484 		}
485 		MLI_UNLOCK(mli);
486 	}
487 
488 out_locked:
489 	MLD_UNLOCK();
490 
491 	mld_set_timeout(&mtp);
492 
493 	return error;
494 }
495 
496 /*
497  * Dispatch an entire queue of pending packet chains.
498  *
499  * Must not be called with in6m_lock held.
500  * XXX This routine unlocks MLD global lock and also mli locks.
501  * Make sure that the calling routine takes reference on the mli
502  * before calling this routine.
503  * Also if we are traversing mli_head, remember to check for
504  * mli list generation count and restart the loop if generation count
505  * has changed.
506  */
507 static void
mld_dispatch_queue_locked(struct mld_ifinfo * mli,struct ifqueue * ifq,int limit)508 mld_dispatch_queue_locked(struct mld_ifinfo *mli, struct ifqueue *ifq, int limit)
509 {
510 	struct mbuf *m;
511 
512 	MLD_LOCK_ASSERT_HELD();
513 
514 	if (mli != NULL) {
515 		MLI_LOCK_ASSERT_HELD(mli);
516 	}
517 
518 	for (;;) {
519 		IF_DEQUEUE(ifq, m);
520 		if (m == NULL) {
521 			break;
522 		}
523 		MLD_PRINTF(("%s: dispatch 0x%llx from 0x%llx\n", __func__,
524 		    (uint64_t)VM_KERNEL_ADDRPERM(ifq),
525 		    (uint64_t)VM_KERNEL_ADDRPERM(m)));
526 
527 		if (mli != NULL) {
528 			MLI_UNLOCK(mli);
529 		}
530 		MLD_UNLOCK();
531 
532 		mld_dispatch_packet(m);
533 
534 		MLD_LOCK();
535 		if (mli != NULL) {
536 			MLI_LOCK(mli);
537 		}
538 
539 		if (--limit == 0) {
540 			break;
541 		}
542 	}
543 
544 	if (mli != NULL) {
545 		MLI_LOCK_ASSERT_HELD(mli);
546 	}
547 }
548 
549 /*
550  * Filter outgoing MLD report state by group.
551  *
552  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
553  * and node-local addresses. However, kernel and socket consumers
554  * always embed the KAME scope ID in the address provided, so strip it
555  * when performing comparison.
556  * Note: This is not the same as the *multicast* scope.
557  *
558  * Return zero if the given group is one for which MLD reports
559  * should be suppressed, or non-zero if reports should be issued.
560  */
561 static __inline__ int
mld_is_addr_reported(const struct in6_addr * addr)562 mld_is_addr_reported(const struct in6_addr *addr)
563 {
564 	VERIFY(IN6_IS_ADDR_MULTICAST(addr));
565 
566 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) {
567 		return 0;
568 	}
569 
570 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL && !IN6_IS_ADDR_UNICAST_BASED_MULTICAST(addr)) {
571 		struct in6_addr tmp = *addr;
572 		in6_clearscope(&tmp);
573 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) {
574 			return 0;
575 		}
576 	}
577 
578 	return 1;
579 }
580 
581 /*
582  * Attach MLD when PF_INET6 is attached to an interface.
583  */
584 struct mld_ifinfo *
mld_domifattach(struct ifnet * ifp,zalloc_flags_t how)585 mld_domifattach(struct ifnet *ifp, zalloc_flags_t how)
586 {
587 	struct mld_ifinfo *mli;
588 
589 	os_log_debug(OS_LOG_DEFAULT, "%s: called for ifp %s\n", __func__,
590 	    if_name(ifp));
591 
592 	mli = mli_alloc(how);
593 	if (mli == NULL) {
594 		return NULL;
595 	}
596 
597 	MLD_LOCK();
598 
599 	MLI_LOCK(mli);
600 	mli_initvar(mli, ifp, 0);
601 	mli->mli_debug |= IFD_ATTACHED;
602 	MLI_ADDREF_LOCKED(mli); /* hold a reference for mli_head */
603 	MLI_ADDREF_LOCKED(mli); /* hold a reference for caller */
604 	MLI_UNLOCK(mli);
605 	ifnet_lock_shared(ifp);
606 	mld6_initsilent(ifp, mli);
607 	ifnet_lock_done(ifp);
608 
609 	LIST_INSERT_HEAD(&mli_head, mli, mli_link);
610 	mld_mli_list_genid++;
611 
612 	MLD_UNLOCK();
613 
614 	os_log_info(OS_LOG_DEFAULT, "%s: allocated mld_ifinfo for ifp %s\n",
615 	    __func__, if_name(ifp));
616 
617 	return mli;
618 }
619 
620 /*
621  * Attach MLD when PF_INET6 is reattached to an interface.  Caller is
622  * expected to have an outstanding reference to the mli.
623  */
624 void
mld_domifreattach(struct mld_ifinfo * mli)625 mld_domifreattach(struct mld_ifinfo *mli)
626 {
627 	struct ifnet *ifp;
628 
629 	MLD_LOCK();
630 
631 	MLI_LOCK(mli);
632 	VERIFY(!(mli->mli_debug & IFD_ATTACHED));
633 	ifp = mli->mli_ifp;
634 	VERIFY(ifp != NULL);
635 	mli_initvar(mli, ifp, 1);
636 	mli->mli_debug |= IFD_ATTACHED;
637 	MLI_ADDREF_LOCKED(mli); /* hold a reference for mli_head */
638 	MLI_UNLOCK(mli);
639 	ifnet_lock_shared(ifp);
640 	mld6_initsilent(ifp, mli);
641 	ifnet_lock_done(ifp);
642 
643 	LIST_INSERT_HEAD(&mli_head, mli, mli_link);
644 	mld_mli_list_genid++;
645 
646 	MLD_UNLOCK();
647 
648 	os_log_info(OS_LOG_DEFAULT, "%s: reattached mld_ifinfo for ifp %s\n",
649 	    __func__, if_name(ifp));
650 }
651 
652 /*
653  * Hook for domifdetach.
654  */
655 void
mld_domifdetach(struct ifnet * ifp)656 mld_domifdetach(struct ifnet *ifp)
657 {
658 	SLIST_HEAD(, in6_multi) in6m_dthead;
659 
660 	SLIST_INIT(&in6m_dthead);
661 
662 	os_log_info(OS_LOG_DEFAULT, "%s: called for ifp %s\n", __func__,
663 	    if_name(ifp));
664 
665 	MLD_LOCK();
666 	mli_delete(ifp, (struct mld_in6m_relhead *)&in6m_dthead);
667 	MLD_UNLOCK();
668 
669 	/* Now that we're dropped all locks, release detached records */
670 	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
671 }
672 
673 /*
674  * Called at interface detach time.  Note that we only flush all deferred
675  * responses and record releases; all remaining inm records and their source
676  * entries related to this interface are left intact, in order to handle
677  * the reattach case.
678  */
679 static void
mli_delete(const struct ifnet * ifp,struct mld_in6m_relhead * in6m_dthead)680 mli_delete(const struct ifnet *ifp, struct mld_in6m_relhead *in6m_dthead)
681 {
682 	struct mld_ifinfo *mli, *tmli;
683 
684 	MLD_LOCK_ASSERT_HELD();
685 
686 	LIST_FOREACH_SAFE(mli, &mli_head, mli_link, tmli) {
687 		MLI_LOCK(mli);
688 		if (mli->mli_ifp == ifp) {
689 			/*
690 			 * Free deferred General Query responses.
691 			 */
692 			IF_DRAIN(&mli->mli_gq);
693 			IF_DRAIN(&mli->mli_v1q);
694 			mld_flush_relq(mli, in6m_dthead);
695 			mli->mli_debug &= ~IFD_ATTACHED;
696 			MLI_UNLOCK(mli);
697 
698 			LIST_REMOVE(mli, mli_link);
699 			MLI_REMREF(mli); /* release mli_head reference */
700 			mld_mli_list_genid++;
701 			return;
702 		}
703 		MLI_UNLOCK(mli);
704 	}
705 	panic("%s: mld_ifinfo not found for ifp %p(%s)", __func__,
706 	    ifp, ifp->if_xname);
707 }
708 
709 __private_extern__ void
mld6_initsilent(struct ifnet * ifp,struct mld_ifinfo * mli)710 mld6_initsilent(struct ifnet *ifp, struct mld_ifinfo *mli)
711 {
712 	ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_OWNED);
713 
714 	MLI_LOCK_ASSERT_NOTHELD(mli);
715 	MLI_LOCK(mli);
716 	if (!(ifp->if_flags & IFF_MULTICAST) &&
717 	    (ifp->if_eflags & (IFEF_IPV6_ND6ALT | IFEF_LOCALNET_PRIVATE))) {
718 		mli->mli_flags |= MLIF_SILENT;
719 	} else {
720 		mli->mli_flags &= ~MLIF_SILENT;
721 	}
722 	MLI_UNLOCK(mli);
723 }
724 
725 static void
mli_initvar(struct mld_ifinfo * mli,struct ifnet * ifp,int reattach)726 mli_initvar(struct mld_ifinfo *mli, struct ifnet *ifp, int reattach)
727 {
728 	MLI_LOCK_ASSERT_HELD(mli);
729 
730 	mli->mli_ifp = ifp;
731 	if (mld_v2enable) {
732 		mli->mli_version = MLD_VERSION_2;
733 	} else {
734 		mli->mli_version = MLD_VERSION_1;
735 	}
736 	mli->mli_flags = 0;
737 	mli->mli_rv = MLD_RV_INIT;
738 	mli->mli_qi = MLD_QI_INIT;
739 	mli->mli_qri = MLD_QRI_INIT;
740 	mli->mli_uri = MLD_URI_INIT;
741 
742 	if (mld_use_allow) {
743 		mli->mli_flags |= MLIF_USEALLOW;
744 	}
745 	if (!reattach) {
746 		SLIST_INIT(&mli->mli_relinmhead);
747 	}
748 
749 	/*
750 	 * Responses to general queries are subject to bounds.
751 	 */
752 	mli->mli_gq.ifq_maxlen = MLD_MAX_RESPONSE_PACKETS;
753 	mli->mli_v1q.ifq_maxlen = MLD_MAX_RESPONSE_PACKETS;
754 }
755 
756 static struct mld_ifinfo *
mli_alloc(zalloc_flags_t how)757 mli_alloc(zalloc_flags_t how)
758 {
759 	struct mld_ifinfo *mli = zalloc_flags(mli_zone, how | Z_ZERO);
760 	if (mli != NULL) {
761 		lck_mtx_init(&mli->mli_lock, &mld_mtx_grp, &mld_mtx_attr);
762 		mli->mli_debug |= IFD_ALLOC;
763 	}
764 	return mli;
765 }
766 
767 static void
mli_free(struct mld_ifinfo * mli)768 mli_free(struct mld_ifinfo *mli)
769 {
770 	MLI_LOCK(mli);
771 	if (mli->mli_debug & IFD_ATTACHED) {
772 		panic("%s: attached mli=%p is being freed", __func__, mli);
773 		/* NOTREACHED */
774 	} else if (mli->mli_ifp != NULL) {
775 		panic("%s: ifp not NULL for mli=%p", __func__, mli);
776 		/* NOTREACHED */
777 	} else if (!(mli->mli_debug & IFD_ALLOC)) {
778 		panic("%s: mli %p cannot be freed", __func__, mli);
779 		/* NOTREACHED */
780 	} else if (mli->mli_refcnt != 0) {
781 		panic("%s: non-zero refcnt mli=%p", __func__, mli);
782 		/* NOTREACHED */
783 	}
784 	mli->mli_debug &= ~IFD_ALLOC;
785 	MLI_UNLOCK(mli);
786 
787 	lck_mtx_destroy(&mli->mli_lock, &mld_mtx_grp);
788 	zfree(mli_zone, mli);
789 }
790 
791 void
mli_addref(struct mld_ifinfo * mli,int locked)792 mli_addref(struct mld_ifinfo *mli, int locked)
793 {
794 	if (!locked) {
795 		MLI_LOCK_SPIN(mli);
796 	} else {
797 		MLI_LOCK_ASSERT_HELD(mli);
798 	}
799 
800 	if (++mli->mli_refcnt == 0) {
801 		panic("%s: mli=%p wraparound refcnt", __func__, mli);
802 		/* NOTREACHED */
803 	}
804 	if (!locked) {
805 		MLI_UNLOCK(mli);
806 	}
807 }
808 
809 void
mli_remref(struct mld_ifinfo * mli)810 mli_remref(struct mld_ifinfo *mli)
811 {
812 	SLIST_HEAD(, in6_multi) in6m_dthead;
813 	struct ifnet *ifp;
814 
815 	MLI_LOCK_SPIN(mli);
816 
817 	if (mli->mli_refcnt == 0) {
818 		panic("%s: mli=%p negative refcnt", __func__, mli);
819 		/* NOTREACHED */
820 	}
821 
822 	--mli->mli_refcnt;
823 	if (mli->mli_refcnt > 0) {
824 		MLI_UNLOCK(mli);
825 		return;
826 	}
827 
828 	ifp = mli->mli_ifp;
829 	mli->mli_ifp = NULL;
830 	IF_DRAIN(&mli->mli_gq);
831 	IF_DRAIN(&mli->mli_v1q);
832 	SLIST_INIT(&in6m_dthead);
833 	mld_flush_relq(mli, (struct mld_in6m_relhead *)&in6m_dthead);
834 	MLI_UNLOCK(mli);
835 
836 	/* Now that we're dropped all locks, release detached records */
837 	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
838 
839 	os_log(OS_LOG_DEFAULT, "%s: freeing mld_ifinfo for ifp %s\n",
840 	    __func__, if_name(ifp));
841 
842 	mli_free(mli);
843 }
844 
845 /*
846  * Process a received MLDv1 general or address-specific query.
847  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
848  *
849  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
850  * mld_addr. This is OK as we own the mbuf chain.
851  */
852 static int
mld_v1_input_query(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mld_hdr * mld)853 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
854     /*const*/ struct mld_hdr *mld)
855 {
856 	struct mld_ifinfo       *mli;
857 	struct in6_multi        *inm;
858 	int                      err = 0, is_general_query;
859 	uint16_t                 timer;
860 	struct mld_tparams       mtp = { .qpt = 0, .it = 0, .cst = 0, .sct = 0 };
861 
862 	MLD_LOCK_ASSERT_NOTHELD();
863 
864 	is_general_query = 0;
865 
866 	if (!mld_v1enable) {
867 		os_log_info(OS_LOG_DEFAULT, "%s: ignore v1 query on ifp %s\n",
868 		    __func__, if_name(ifp));
869 		goto done;
870 	}
871 
872 	/*
873 	 * RFC3810 Section 6.2: MLD queries must originate from
874 	 * a router's link-local address.
875 	 */
876 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
877 		os_log_info(OS_LOG_DEFAULT, "%s: ignore v1 query src %s on ifp %s\n",
878 		    __func__, ip6_sprintf(&ip6->ip6_src),
879 		    if_name(ifp));
880 		goto done;
881 	}
882 
883 	/*
884 	 * Do address field validation upfront before we accept
885 	 * the query.
886 	 */
887 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
888 		/*
889 		 * MLDv1 General Query.
890 		 * If this was not sent to the all-nodes group, ignore it.
891 		 */
892 		struct in6_addr          dst;
893 
894 		dst = ip6->ip6_dst;
895 		in6_clearscope(&dst);
896 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) {
897 			err = EINVAL;
898 			goto done;
899 		}
900 		is_general_query = 1;
901 	} else {
902 		/*
903 		 * Embed scope ID of receiving interface in MLD query for
904 		 * lookup whilst we don't hold other locks.
905 		 */
906 		(void)in6_setscope(&mld->mld_addr, ifp, NULL);
907 	}
908 
909 	/*
910 	 * Switch to MLDv1 host compatibility mode.
911 	 */
912 	mli = MLD_IFINFO(ifp);
913 	VERIFY(mli != NULL);
914 
915 	MLI_LOCK(mli);
916 	mtp.qpt = mld_set_version(mli, MLD_VERSION_1);
917 	MLI_UNLOCK(mli);
918 
919 	timer = ntohs(mld->mld_maxdelay) / MLD_TIMER_SCALE;
920 	if (timer == 0) {
921 		timer = 1;
922 	}
923 
924 	if (is_general_query) {
925 		struct in6_multistep step;
926 
927 		os_log_debug(OS_LOG_DEFAULT, "%s: process v1 general query on ifp %s\n",
928 		    __func__, if_name(ifp));
929 		/*
930 		 * For each reporting group joined on this
931 		 * interface, kick the report timer.
932 		 */
933 		in6_multihead_lock_shared();
934 		IN6_FIRST_MULTI(step, inm);
935 		while (inm != NULL) {
936 			IN6M_LOCK(inm);
937 			if (inm->in6m_ifp == ifp) {
938 				mtp.cst += mld_v1_update_group(inm, timer);
939 			}
940 			IN6M_UNLOCK(inm);
941 			IN6_NEXT_MULTI(step, inm);
942 		}
943 		in6_multihead_lock_done();
944 	} else {
945 		/*
946 		 * MLDv1 Group-Specific Query.
947 		 * If this is a group-specific MLDv1 query, we need only
948 		 * look up the single group to process it.
949 		 */
950 		in6_multihead_lock_shared();
951 		IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
952 		in6_multihead_lock_done();
953 
954 		if (inm != NULL) {
955 			IN6M_LOCK(inm);
956 			os_log_debug(OS_LOG_DEFAULT, "%s: process v1 query %s on "
957 			    "ifp %s\n", __func__,
958 			    ip6_sprintf(&mld->mld_addr),
959 			    if_name(ifp));
960 			mtp.cst = mld_v1_update_group(inm, timer);
961 			IN6M_UNLOCK(inm);
962 			IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
963 		}
964 		/* XXX Clear embedded scope ID as userland won't expect it. */
965 		in6_clearscope(&mld->mld_addr);
966 	}
967 done:
968 	mld_set_timeout(&mtp);
969 
970 	return err;
971 }
972 
973 /*
974  * Update the report timer on a group in response to an MLDv1 query.
975  *
976  * If we are becoming the reporting member for this group, start the timer.
977  * If we already are the reporting member for this group, and timer is
978  * below the threshold, reset it.
979  *
980  * We may be updating the group for the first time since we switched
981  * to MLDv2. If we are, then we must clear any recorded source lists,
982  * and transition to REPORTING state; the group timer is overloaded
983  * for group and group-source query responses.
984  *
985  * Unlike MLDv2, the delay per group should be jittered
986  * to avoid bursts of MLDv1 reports.
987  */
988 static uint32_t
mld_v1_update_group(struct in6_multi * inm,const int timer)989 mld_v1_update_group(struct in6_multi *inm, const int timer)
990 {
991 	IN6M_LOCK_ASSERT_HELD(inm);
992 
993 	MLD_PRINTF(("%s: %s/%s timer=%d\n", __func__,
994 	    ip6_sprintf(&inm->in6m_addr),
995 	    if_name(inm->in6m_ifp), timer));
996 
997 	switch (inm->in6m_state) {
998 	case MLD_NOT_MEMBER:
999 	case MLD_SILENT_MEMBER:
1000 		break;
1001 	case MLD_REPORTING_MEMBER:
1002 		if (inm->in6m_timer != 0 &&
1003 		    inm->in6m_timer <= timer) {
1004 			MLD_PRINTF(("%s: REPORTING and timer running, "
1005 			    "skipping.\n", __func__));
1006 			break;
1007 		}
1008 		OS_FALLTHROUGH;
1009 	case MLD_SG_QUERY_PENDING_MEMBER:
1010 	case MLD_G_QUERY_PENDING_MEMBER:
1011 	case MLD_IDLE_MEMBER:
1012 	case MLD_LAZY_MEMBER:
1013 	case MLD_AWAKENING_MEMBER:
1014 		MLD_PRINTF(("%s: ->REPORTING\n", __func__));
1015 		inm->in6m_state = MLD_REPORTING_MEMBER;
1016 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1017 		break;
1018 	case MLD_SLEEPING_MEMBER:
1019 		MLD_PRINTF(("%s: ->AWAKENING\n", __func__));
1020 		inm->in6m_state = MLD_AWAKENING_MEMBER;
1021 		break;
1022 	case MLD_LEAVING_MEMBER:
1023 		break;
1024 	}
1025 
1026 	return inm->in6m_timer;
1027 }
1028 
1029 /*
1030  * Process a received MLDv2 general, group-specific or
1031  * group-and-source-specific query.
1032  *
1033  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
1034  *
1035  * Return 0 if successful, otherwise an appropriate error code is returned.
1036  */
1037 static int
mld_v2_input_query(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mbuf * m,const int off,const int icmp6len)1038 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
1039     struct mbuf *m, const int off, const int icmp6len)
1040 {
1041 	struct mld_ifinfo       *mli;
1042 	struct mldv2_query      *mld;
1043 	struct in6_multi        *inm;
1044 	uint32_t                 maxdelay, nsrc, qqi, timer;
1045 	int                      err = 0, is_general_query;
1046 	uint8_t                  qrv;
1047 	struct mld_tparams       mtp = { .qpt = 0, .it = 0, .cst = 0, .sct = 0 };
1048 
1049 	MLD_LOCK_ASSERT_NOTHELD();
1050 
1051 	is_general_query = 0;
1052 
1053 	if (!mld_v2enable) {
1054 		os_log_info(OS_LOG_DEFAULT, "%s: ignore v2 query on ifp %s\n",
1055 		    __func__, if_name(ifp));
1056 		goto done;
1057 	}
1058 
1059 	/*
1060 	 * RFC3810 Section 6.2: MLD queries must originate from
1061 	 * a router's link-local address.
1062 	 */
1063 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
1064 		os_log_info(OS_LOG_DEFAULT,
1065 		    "%s: ignore v1 query src %s on ifp %s\n",
1066 		    __func__, ip6_sprintf(&ip6->ip6_src),
1067 		    if_name(ifp));
1068 		goto done;
1069 	}
1070 
1071 	os_log_debug(OS_LOG_DEFAULT,
1072 	    "%s: input v2 query on ifp %s\n", __func__,
1073 	    if_name(ifp));
1074 
1075 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
1076 
1077 	maxdelay = ntohs(mld->mld_maxdelay);    /* in 1/10ths of a second */
1078 	if (maxdelay > SHRT_MAX) {
1079 		maxdelay = (MLD_MRC_MANT((uint16_t)maxdelay) | 0x1000) <<
1080 		    (MLD_MRC_EXP((uint16_t)maxdelay) + 3);
1081 	}
1082 	timer = maxdelay / MLD_TIMER_SCALE;
1083 	if (timer == 0) {
1084 		timer = 1;
1085 	}
1086 
1087 	qrv = MLD_QRV(mld->mld_misc);
1088 	if (qrv < 2) {
1089 		MLD_PRINTF(("%s: clamping qrv %d to %d\n", __func__,
1090 		    qrv, MLD_RV_INIT));
1091 		qrv = MLD_RV_INIT;
1092 	}
1093 
1094 	qqi = mld->mld_qqi;
1095 	if (qqi >= 128) {
1096 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
1097 		    (MLD_QQIC_EXP(mld->mld_qqi) + 3);
1098 	}
1099 
1100 	nsrc = ntohs(mld->mld_numsrc);
1101 	if (nsrc > MLD_MAX_GS_SOURCES) {
1102 		err = EMSGSIZE;
1103 		goto done;
1104 	}
1105 	if (icmp6len < sizeof(struct mldv2_query) +
1106 	    (nsrc * sizeof(struct in6_addr))) {
1107 		err = EMSGSIZE;
1108 		goto done;
1109 	}
1110 
1111 	/*
1112 	 * Do further input validation upfront to avoid resetting timers
1113 	 * should we need to discard this query.
1114 	 */
1115 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
1116 		/*
1117 		 * A general query with a source list has undefined
1118 		 * behaviour; discard it.
1119 		 */
1120 		if (nsrc > 0) {
1121 			err = EINVAL;
1122 			goto done;
1123 		}
1124 		is_general_query = 1;
1125 	} else {
1126 		/*
1127 		 * Embed scope ID of receiving interface in MLD query for
1128 		 * lookup whilst we don't hold other locks (due to KAME
1129 		 * locking lameness). We own this mbuf chain just now.
1130 		 */
1131 		(void)in6_setscope(&mld->mld_addr, ifp, NULL);
1132 	}
1133 
1134 	mli = MLD_IFINFO(ifp);
1135 	VERIFY(mli != NULL);
1136 
1137 	MLI_LOCK(mli);
1138 	/*
1139 	 * Discard the v2 query if we're in Compatibility Mode.
1140 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
1141 	 * until the Old Version Querier Present timer expires.
1142 	 */
1143 	if (mli->mli_version != MLD_VERSION_2) {
1144 		MLI_UNLOCK(mli);
1145 		goto done;
1146 	}
1147 
1148 	mtp.qpt = mld_set_version(mli, MLD_VERSION_2);
1149 	mli->mli_rv = qrv;
1150 	mli->mli_qi = qqi;
1151 	mli->mli_qri = MAX(timer, MLD_QRI_MIN);
1152 
1153 	MLD_PRINTF(("%s: qrv %d qi %d qri %d\n", __func__, mli->mli_rv,
1154 	    mli->mli_qi, mli->mli_qri));
1155 
1156 	if (is_general_query) {
1157 		/*
1158 		 * MLDv2 General Query.
1159 		 *
1160 		 * Schedule a current-state report on this ifp for
1161 		 * all groups, possibly containing source lists.
1162 		 *
1163 		 * If there is a pending General Query response
1164 		 * scheduled earlier than the selected delay, do
1165 		 * not schedule any other reports.
1166 		 * Otherwise, reset the interface timer.
1167 		 */
1168 		os_log_debug(OS_LOG_DEFAULT, "%s: process v2 general query on ifp %s\n",
1169 		    __func__, if_name(ifp));
1170 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
1171 			mtp.it = mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
1172 		}
1173 		MLI_UNLOCK(mli);
1174 	} else {
1175 		MLI_UNLOCK(mli);
1176 		/*
1177 		 * MLDv2 Group-specific or Group-and-source-specific Query.
1178 		 *
1179 		 * Group-source-specific queries are throttled on
1180 		 * a per-group basis to defeat denial-of-service attempts.
1181 		 * Queries for groups we are not a member of on this
1182 		 * link are simply ignored.
1183 		 */
1184 		in6_multihead_lock_shared();
1185 		IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
1186 		in6_multihead_lock_done();
1187 		if (inm == NULL) {
1188 			goto done;
1189 		}
1190 
1191 		IN6M_LOCK(inm);
1192 		if (nsrc > 0) {
1193 			if (!ratecheck(&inm->in6m_lastgsrtv,
1194 			    &mld_gsrdelay)) {
1195 				os_log_info(OS_LOG_DEFAULT, "%s: GS query throttled\n",
1196 				    __func__);
1197 				IN6M_UNLOCK(inm);
1198 				IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1199 				goto done;
1200 			}
1201 		}
1202 		os_log_debug(OS_LOG_DEFAULT, "%s: process v2 group query on ifp %s\n",
1203 		    __func__, if_name(ifp));
1204 		/*
1205 		 * If there is a pending General Query response
1206 		 * scheduled sooner than the selected delay, no
1207 		 * further report need be scheduled.
1208 		 * Otherwise, prepare to respond to the
1209 		 * group-specific or group-and-source query.
1210 		 */
1211 		MLI_LOCK(mli);
1212 		mtp.it = mli->mli_v2_timer;
1213 		MLI_UNLOCK(mli);
1214 		if (mtp.it == 0 || mtp.it >= timer) {
1215 			(void) mld_v2_process_group_query(inm, timer, m, off);
1216 			mtp.cst = inm->in6m_timer;
1217 		}
1218 		IN6M_UNLOCK(inm);
1219 		IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1220 		/* XXX Clear embedded scope ID as userland won't expect it. */
1221 		in6_clearscope(&mld->mld_addr);
1222 	}
1223 done:
1224 	if (mtp.it > 0) {
1225 		os_log_debug(OS_LOG_DEFAULT, "%s: v2 general query response scheduled in "
1226 		    "T+%d seconds on ifp %s\n", __func__, mtp.it,
1227 		    if_name(ifp));
1228 	}
1229 	mld_set_timeout(&mtp);
1230 
1231 	return err;
1232 }
1233 
1234 /*
1235  * Process a recieved MLDv2 group-specific or group-and-source-specific
1236  * query.
1237  * Return <0 if any error occured. Currently this is ignored.
1238  */
1239 static int
mld_v2_process_group_query(struct in6_multi * inm,int timer,struct mbuf * m0,const int off)1240 mld_v2_process_group_query(struct in6_multi *inm, int timer, struct mbuf *m0,
1241     const int off)
1242 {
1243 	struct mldv2_query      *mld;
1244 	int                      retval;
1245 	uint16_t                 nsrc;
1246 
1247 	IN6M_LOCK_ASSERT_HELD(inm);
1248 
1249 	retval = 0;
1250 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
1251 
1252 	switch (inm->in6m_state) {
1253 	case MLD_NOT_MEMBER:
1254 	case MLD_SILENT_MEMBER:
1255 	case MLD_SLEEPING_MEMBER:
1256 	case MLD_LAZY_MEMBER:
1257 	case MLD_AWAKENING_MEMBER:
1258 	case MLD_IDLE_MEMBER:
1259 	case MLD_LEAVING_MEMBER:
1260 		return retval;
1261 	case MLD_REPORTING_MEMBER:
1262 	case MLD_G_QUERY_PENDING_MEMBER:
1263 	case MLD_SG_QUERY_PENDING_MEMBER:
1264 		break;
1265 	}
1266 
1267 	nsrc = ntohs(mld->mld_numsrc);
1268 
1269 	/*
1270 	 * Deal with group-specific queries upfront.
1271 	 * If any group query is already pending, purge any recorded
1272 	 * source-list state if it exists, and schedule a query response
1273 	 * for this group-specific query.
1274 	 */
1275 	if (nsrc == 0) {
1276 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1277 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1278 			in6m_clear_recorded(inm);
1279 			timer = min(inm->in6m_timer, timer);
1280 		}
1281 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1282 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1283 		return retval;
1284 	}
1285 
1286 	/*
1287 	 * Deal with the case where a group-and-source-specific query has
1288 	 * been received but a group-specific query is already pending.
1289 	 */
1290 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1291 		timer = min(inm->in6m_timer, timer);
1292 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1293 		return retval;
1294 	}
1295 
1296 	/*
1297 	 * Finally, deal with the case where a group-and-source-specific
1298 	 * query has been received, where a response to a previous g-s-r
1299 	 * query exists, or none exists.
1300 	 * In this case, we need to parse the source-list which the Querier
1301 	 * has provided us with and check if we have any source list filter
1302 	 * entries at T1 for these sources. If we do not, there is no need
1303 	 * schedule a report and the query may be dropped.
1304 	 * If we do, we must record them and schedule a current-state
1305 	 * report for those sources.
1306 	 */
1307 	if (inm->in6m_nsrc > 0) {
1308 		struct mbuf             *m;
1309 		struct in6_addr          addr;
1310 		int                      i, nrecorded;
1311 		int                      soff;
1312 
1313 		m = m0;
1314 		soff = off + sizeof(struct mldv2_query);
1315 		nrecorded = 0;
1316 		for (i = 0; i < nsrc; i++) {
1317 			m_copydata(m, soff, sizeof(addr), &addr);
1318 			retval = in6m_record_source(inm, &addr);
1319 			if (retval < 0) {
1320 				break;
1321 			}
1322 			nrecorded += retval;
1323 			soff += sizeof(struct in6_addr);
1324 
1325 			while (m && (soff >= m->m_len)) {
1326 				soff -= m->m_len;
1327 				m = m->m_next;
1328 			}
1329 
1330 			/* should not be possible: */
1331 			if (m == NULL) {
1332 				break;
1333 			}
1334 		}
1335 		if (nrecorded > 0) {
1336 			MLD_PRINTF(("%s: schedule response to SG query\n",
1337 			    __func__));
1338 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1339 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1340 		}
1341 	}
1342 
1343 	return retval;
1344 }
1345 
1346 /*
1347  * Process a received MLDv1 host membership report.
1348  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1349  *
1350  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1351  * mld_addr. This is OK as we own the mbuf chain.
1352  */
1353 static int
mld_v1_input_report(struct ifnet * ifp,struct mbuf * m,const struct ip6_hdr * ip6,struct mld_hdr * mld)1354 mld_v1_input_report(struct ifnet *ifp, struct mbuf *m,
1355     const struct ip6_hdr *ip6, /*const*/ struct mld_hdr *mld)
1356 {
1357 	struct in6_addr          src, dst;
1358 	struct in6_ifaddr       *ia;
1359 	struct in6_multi        *inm;
1360 
1361 	if (!mld_v1enable) {
1362 		os_log_info(OS_LOG_DEFAULT, "%s: ignore v1 report on ifp %s\n",
1363 		    __func__, if_name(ifp));
1364 		return 0;
1365 	}
1366 
1367 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1368 	    (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1369 		return 0;
1370 	}
1371 
1372 	/*
1373 	 * MLDv1 reports must originate from a host's link-local address,
1374 	 * or the unspecified address (when booting).
1375 	 */
1376 	src = ip6->ip6_src;
1377 	in6_clearscope(&src);
1378 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1379 		os_log_info(OS_LOG_DEFAULT, "%s: ignore v1 query src %s on ifp %s\n",
1380 		    __func__, ip6_sprintf(&ip6->ip6_src),
1381 		    if_name(ifp));
1382 		return EINVAL;
1383 	}
1384 
1385 	/*
1386 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1387 	 * group, and must be directed to the group itself.
1388 	 */
1389 	dst = ip6->ip6_dst;
1390 	in6_clearscope(&dst);
1391 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1392 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1393 		os_log_info(OS_LOG_DEFAULT, "%s: ignore v1 query dst %s on ifp %s\n",
1394 		    __func__, ip6_sprintf(&ip6->ip6_dst),
1395 		    if_name(ifp));
1396 		return EINVAL;
1397 	}
1398 
1399 	/*
1400 	 * Make sure we don't hear our own membership report, as fast
1401 	 * leave requires knowing that we are the only member of a
1402 	 * group. Assume we used the link-local address if available,
1403 	 * otherwise look for ::.
1404 	 *
1405 	 * XXX Note that scope ID comparison is needed for the address
1406 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1407 	 * performed for the on-wire address.
1408 	 */
1409 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
1410 	if (ia != NULL) {
1411 		IFA_LOCK(&ia->ia_ifa);
1412 		if ((IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia)))) {
1413 			IFA_UNLOCK(&ia->ia_ifa);
1414 			IFA_REMREF(&ia->ia_ifa);
1415 			return 0;
1416 		}
1417 		IFA_UNLOCK(&ia->ia_ifa);
1418 		IFA_REMREF(&ia->ia_ifa);
1419 	} else if (IN6_IS_ADDR_UNSPECIFIED(&src)) {
1420 		return 0;
1421 	}
1422 
1423 	os_log_debug(OS_LOG_DEFAULT, "%s: process v1 report %s on ifp %s\n",
1424 	    __func__, ip6_sprintf(&mld->mld_addr),
1425 	    if_name(ifp));
1426 
1427 	/*
1428 	 * Embed scope ID of receiving interface in MLD query for lookup
1429 	 * whilst we don't hold other locks (due to KAME locking lameness).
1430 	 */
1431 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
1432 		(void)in6_setscope(&mld->mld_addr, ifp, NULL);
1433 	}
1434 
1435 	/*
1436 	 * MLDv1 report suppression.
1437 	 * If we are a member of this group, and our membership should be
1438 	 * reported, and our group timer is pending or about to be reset,
1439 	 * stop our group timer by transitioning to the 'lazy' state.
1440 	 */
1441 	in6_multihead_lock_shared();
1442 	IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
1443 	in6_multihead_lock_done();
1444 
1445 	if (inm != NULL) {
1446 		struct mld_ifinfo *mli;
1447 
1448 		IN6M_LOCK(inm);
1449 		mli = inm->in6m_mli;
1450 		VERIFY(mli != NULL);
1451 
1452 		MLI_LOCK(mli);
1453 		/*
1454 		 * If we are in MLDv2 host mode, do not allow the
1455 		 * other host's MLDv1 report to suppress our reports.
1456 		 */
1457 		if (mli->mli_version == MLD_VERSION_2) {
1458 			MLI_UNLOCK(mli);
1459 			IN6M_UNLOCK(inm);
1460 			IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1461 			goto out;
1462 		}
1463 		MLI_UNLOCK(mli);
1464 
1465 		inm->in6m_timer = 0;
1466 
1467 		switch (inm->in6m_state) {
1468 		case MLD_NOT_MEMBER:
1469 		case MLD_SILENT_MEMBER:
1470 		case MLD_SLEEPING_MEMBER:
1471 			break;
1472 		case MLD_REPORTING_MEMBER:
1473 		case MLD_IDLE_MEMBER:
1474 		case MLD_AWAKENING_MEMBER:
1475 			MLD_PRINTF(("%s: report suppressed for %s on "
1476 			    "ifp 0x%llx(%s)\n", __func__,
1477 			    ip6_sprintf(&mld->mld_addr),
1478 			    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1479 			OS_FALLTHROUGH;
1480 		case MLD_LAZY_MEMBER:
1481 			inm->in6m_state = MLD_LAZY_MEMBER;
1482 			break;
1483 		case MLD_G_QUERY_PENDING_MEMBER:
1484 		case MLD_SG_QUERY_PENDING_MEMBER:
1485 		case MLD_LEAVING_MEMBER:
1486 			break;
1487 		}
1488 		IN6M_UNLOCK(inm);
1489 		IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1490 	}
1491 
1492 out:
1493 	/* XXX Clear embedded scope ID as userland won't expect it. */
1494 	in6_clearscope(&mld->mld_addr);
1495 
1496 	return 0;
1497 }
1498 
1499 /*
1500  * MLD input path.
1501  *
1502  * Assume query messages which fit in a single ICMPv6 message header
1503  * have been pulled up.
1504  * Assume that userland will want to see the message, even if it
1505  * otherwise fails kernel input validation; do not free it.
1506  * Pullup may however free the mbuf chain m if it fails.
1507  *
1508  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1509  */
1510 int
mld_input(struct mbuf * m,int off,int icmp6len)1511 mld_input(struct mbuf *m, int off, int icmp6len)
1512 {
1513 	struct ifnet    *ifp = NULL;
1514 	struct ip6_hdr  *ip6 = NULL;
1515 	struct mld_hdr  *mld = NULL;
1516 	int              mldlen = 0;
1517 
1518 	MLD_PRINTF(("%s: called w/mbuf (0x%llx,%d)\n", __func__,
1519 	    (uint64_t)VM_KERNEL_ADDRPERM(m), off));
1520 
1521 	ifp = m->m_pkthdr.rcvif;
1522 
1523 	/* Pullup to appropriate size. */
1524 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1525 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1526 	    icmp6len >= sizeof(struct mldv2_query)) {
1527 		mldlen = sizeof(struct mldv2_query);
1528 	} else {
1529 		mldlen = sizeof(struct mld_hdr);
1530 	}
1531 	// check if mldv2_query/mld_hdr fits in the first mbuf
1532 	IP6_EXTHDR_CHECK(m, off, mldlen, return IPPROTO_DONE);
1533 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1534 	if (mld == NULL) {
1535 		icmp6stat.icp6s_badlen++;
1536 		return IPPROTO_DONE;
1537 	}
1538 	ip6 = mtod(m, struct ip6_hdr *);
1539 
1540 	/*
1541 	 * Userland needs to see all of this traffic for implementing
1542 	 * the endpoint discovery portion of multicast routing.
1543 	 */
1544 	switch (mld->mld_type) {
1545 	case MLD_LISTENER_QUERY:
1546 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1547 		if (icmp6len == sizeof(struct mld_hdr)) {
1548 			if (mld_v1_input_query(ifp, ip6, mld) != 0) {
1549 				return 0;
1550 			}
1551 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1552 			if (mld_v2_input_query(ifp, ip6, m, off,
1553 			    icmp6len) != 0) {
1554 				return 0;
1555 			}
1556 		}
1557 		break;
1558 	case MLD_LISTENER_REPORT:
1559 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1560 		if (mld_v1_input_report(ifp, m, ip6, mld) != 0) {
1561 			return 0;
1562 		}
1563 		break;
1564 	case MLDV2_LISTENER_REPORT:
1565 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1566 		break;
1567 	case MLD_LISTENER_DONE:
1568 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1569 		break;
1570 	default:
1571 		break;
1572 	}
1573 
1574 	return 0;
1575 }
1576 
1577 /*
1578  * Schedule MLD timer based on various parameters; caller must ensure that
1579  * lock ordering is maintained as this routine acquires MLD global lock.
1580  */
1581 void
mld_set_timeout(struct mld_tparams * mtp)1582 mld_set_timeout(struct mld_tparams *mtp)
1583 {
1584 	MLD_LOCK_ASSERT_NOTHELD();
1585 	VERIFY(mtp != NULL);
1586 
1587 	if (mtp->qpt != 0 || mtp->it != 0 || mtp->cst != 0 || mtp->sct != 0) {
1588 		MLD_LOCK();
1589 		if (mtp->qpt != 0) {
1590 			querier_present_timers_running6 = 1;
1591 		}
1592 		if (mtp->it != 0) {
1593 			interface_timers_running6 = 1;
1594 		}
1595 		if (mtp->cst != 0) {
1596 			current_state_timers_running6 = 1;
1597 		}
1598 		if (mtp->sct != 0) {
1599 			state_change_timers_running6 = 1;
1600 		}
1601 		if (mtp->fast) {
1602 			mld_sched_fast_timeout();
1603 		} else {
1604 			mld_sched_timeout();
1605 		}
1606 		MLD_UNLOCK();
1607 	}
1608 }
1609 
1610 void
mld_set_fast_timeout(struct mld_tparams * mtp)1611 mld_set_fast_timeout(struct mld_tparams *mtp)
1612 {
1613 	VERIFY(mtp != NULL);
1614 	mtp->fast = true;
1615 	mld_set_timeout(mtp);
1616 }
1617 
1618 /*
1619  * MLD6 timer handler (per 1 second).
1620  */
1621 static void
mld_timeout(thread_call_param_t arg0,thread_call_param_t arg1 __unused)1622 mld_timeout(thread_call_param_t arg0, thread_call_param_t arg1 __unused)
1623 {
1624 	struct ifqueue           scq;   /* State-change packets */
1625 	struct ifqueue           qrq;   /* Query response packets */
1626 	struct ifnet            *ifp;
1627 	struct mld_ifinfo       *mli;
1628 	struct in6_multi        *inm;
1629 	int                      uri_sec = 0;
1630 	unsigned int genid = mld_mli_list_genid;
1631 	bool                     fast = arg0 != NULL;
1632 
1633 	SLIST_HEAD(, in6_multi) in6m_dthead;
1634 
1635 	SLIST_INIT(&in6m_dthead);
1636 
1637 	/*
1638 	 * Update coarse-grained networking timestamp (in sec.); the idea
1639 	 * is to piggy-back on the timeout callout to update the counter
1640 	 * returnable via net_uptime().
1641 	 */
1642 	net_update_uptime();
1643 
1644 	MLD_LOCK();
1645 
1646 	MLD_PRINTF(("%s: qpt %d, it %d, cst %d, sct %d, fast %d\n", __func__,
1647 	    querier_present_timers_running6, interface_timers_running6,
1648 	    current_state_timers_running6, state_change_timers_running6, fast));
1649 
1650 	if (fast) {
1651 		/*
1652 		 * When running the fast timer, skip processing
1653 		 * of "querier present" timers since they are
1654 		 * based on 1-second intervals.
1655 		 */
1656 		goto skip_query_timers;
1657 	}
1658 	/*
1659 	 * MLDv1 querier present timer processing.
1660 	 */
1661 	if (querier_present_timers_running6) {
1662 		querier_present_timers_running6 = 0;
1663 		LIST_FOREACH(mli, &mli_head, mli_link) {
1664 			MLI_LOCK(mli);
1665 			mld_v1_process_querier_timers(mli);
1666 			if (mli->mli_v1_timer > 0) {
1667 				querier_present_timers_running6 = 1;
1668 			}
1669 			MLI_UNLOCK(mli);
1670 		}
1671 	}
1672 
1673 	/*
1674 	 * MLDv2 General Query response timer processing.
1675 	 */
1676 	if (interface_timers_running6) {
1677 		MLD_PRINTF(("%s: interface timers running\n", __func__));
1678 		interface_timers_running6 = 0;
1679 		mli = LIST_FIRST(&mli_head);
1680 
1681 		while (mli != NULL) {
1682 			if (mli->mli_flags & MLIF_PROCESSED) {
1683 				mli = LIST_NEXT(mli, mli_link);
1684 				continue;
1685 			}
1686 
1687 			MLI_LOCK(mli);
1688 			if (mli->mli_version != MLD_VERSION_2) {
1689 				MLI_UNLOCK(mli);
1690 				mli = LIST_NEXT(mli, mli_link);
1691 				continue;
1692 			}
1693 			/*
1694 			 * XXX The logic below ends up calling
1695 			 * mld_dispatch_packet which can unlock mli
1696 			 * and the global MLD lock.
1697 			 * Therefore grab a reference on MLI and also
1698 			 * check for generation count to see if we should
1699 			 * iterate the list again.
1700 			 */
1701 			MLI_ADDREF_LOCKED(mli);
1702 
1703 			if (mli->mli_v2_timer == 0) {
1704 				/* Do nothing. */
1705 			} else if (--mli->mli_v2_timer == 0) {
1706 				if (mld_v2_dispatch_general_query(mli) > 0) {
1707 					interface_timers_running6 = 1;
1708 				}
1709 			} else {
1710 				interface_timers_running6 = 1;
1711 			}
1712 			mli->mli_flags |= MLIF_PROCESSED;
1713 			MLI_UNLOCK(mli);
1714 			MLI_REMREF(mli);
1715 
1716 			if (genid != mld_mli_list_genid) {
1717 				MLD_PRINTF(("%s: MLD information list changed "
1718 				    "in the middle of iteration! Restart iteration.\n",
1719 				    __func__));
1720 				mli = LIST_FIRST(&mli_head);
1721 				genid = mld_mli_list_genid;
1722 			} else {
1723 				mli = LIST_NEXT(mli, mli_link);
1724 			}
1725 		}
1726 
1727 		LIST_FOREACH(mli, &mli_head, mli_link)
1728 		mli->mli_flags &= ~MLIF_PROCESSED;
1729 	}
1730 
1731 skip_query_timers:
1732 	if (!current_state_timers_running6 &&
1733 	    !state_change_timers_running6) {
1734 		goto out_locked;
1735 	}
1736 
1737 	current_state_timers_running6 = 0;
1738 	state_change_timers_running6 = 0;
1739 
1740 	MLD_PRINTF(("%s: state change timers running\n", __func__));
1741 
1742 	memset(&qrq, 0, sizeof(struct ifqueue));
1743 	qrq.ifq_maxlen = MLD_MAX_G_GS_PACKETS;
1744 
1745 	memset(&scq, 0, sizeof(struct ifqueue));
1746 	scq.ifq_maxlen = MLD_MAX_STATE_CHANGE_PACKETS;
1747 
1748 	/*
1749 	 * MLD host report and state-change timer processing.
1750 	 * Note: Processing a v2 group timer may remove a node.
1751 	 */
1752 	mli = LIST_FIRST(&mli_head);
1753 
1754 	while (mli != NULL) {
1755 		struct in6_multistep step;
1756 
1757 		if (mli->mli_flags & MLIF_PROCESSED) {
1758 			mli = LIST_NEXT(mli, mli_link);
1759 			continue;
1760 		}
1761 
1762 		MLI_LOCK(mli);
1763 		ifp = mli->mli_ifp;
1764 		uri_sec = MLD_RANDOM_DELAY(mli->mli_uri);
1765 		MLI_UNLOCK(mli);
1766 
1767 		in6_multihead_lock_shared();
1768 		IN6_FIRST_MULTI(step, inm);
1769 		while (inm != NULL) {
1770 			IN6M_LOCK(inm);
1771 			if (inm->in6m_ifp != ifp) {
1772 				goto next;
1773 			}
1774 
1775 			MLI_LOCK(mli);
1776 			switch (mli->mli_version) {
1777 			case MLD_VERSION_1:
1778 				mld_v1_process_group_timer(inm,
1779 				    mli->mli_version);
1780 				break;
1781 			case MLD_VERSION_2:
1782 				mld_v2_process_group_timers(mli, &qrq,
1783 				    &scq, inm, uri_sec);
1784 				break;
1785 			}
1786 			MLI_UNLOCK(mli);
1787 next:
1788 			IN6M_UNLOCK(inm);
1789 			IN6_NEXT_MULTI(step, inm);
1790 		}
1791 		in6_multihead_lock_done();
1792 
1793 		/*
1794 		 * XXX The logic below ends up calling
1795 		 * mld_dispatch_packet which can unlock mli
1796 		 * and the global MLD lock.
1797 		 * Therefore grab a reference on MLI and also
1798 		 * check for generation count to see if we should
1799 		 * iterate the list again.
1800 		 */
1801 		MLI_LOCK(mli);
1802 		MLI_ADDREF_LOCKED(mli);
1803 		if (mli->mli_version == MLD_VERSION_1) {
1804 			mld_dispatch_queue_locked(mli, &mli->mli_v1q, 0);
1805 		} else if (mli->mli_version == MLD_VERSION_2) {
1806 			MLI_UNLOCK(mli);
1807 			mld_dispatch_queue_locked(NULL, &qrq, 0);
1808 			mld_dispatch_queue_locked(NULL, &scq, 0);
1809 			VERIFY(qrq.ifq_len == 0);
1810 			VERIFY(scq.ifq_len == 0);
1811 			MLI_LOCK(mli);
1812 		}
1813 		/*
1814 		 * In case there are still any pending membership reports
1815 		 * which didn't get drained at version change time.
1816 		 */
1817 		IF_DRAIN(&mli->mli_v1q);
1818 		/*
1819 		 * Release all deferred inm records, and drain any locally
1820 		 * enqueued packets; do it even if the current MLD version
1821 		 * for the link is no longer MLDv2, in order to handle the
1822 		 * version change case.
1823 		 */
1824 		mld_flush_relq(mli, (struct mld_in6m_relhead *)&in6m_dthead);
1825 		mli->mli_flags |= MLIF_PROCESSED;
1826 		MLI_UNLOCK(mli);
1827 		MLI_REMREF(mli);
1828 
1829 		IF_DRAIN(&qrq);
1830 		IF_DRAIN(&scq);
1831 
1832 		if (genid != mld_mli_list_genid) {
1833 			MLD_PRINTF(("%s: MLD information list changed "
1834 			    "in the middle of iteration! Restart iteration.\n",
1835 			    __func__));
1836 			mli = LIST_FIRST(&mli_head);
1837 			genid = mld_mli_list_genid;
1838 		} else {
1839 			mli = LIST_NEXT(mli, mli_link);
1840 		}
1841 	}
1842 
1843 	LIST_FOREACH(mli, &mli_head, mli_link)
1844 	mli->mli_flags &= ~MLIF_PROCESSED;
1845 
1846 out_locked:
1847 	/* re-arm the timer if there's work to do */
1848 	if (fast) {
1849 		mld_fast_timeout_run = false;
1850 	} else {
1851 		mld_timeout_run = false;
1852 	}
1853 	mld_sched_timeout();
1854 	MLD_UNLOCK();
1855 
1856 	/* Now that we're dropped all locks, release detached records */
1857 	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
1858 }
1859 
1860 static void
mld_sched_timeout(void)1861 mld_sched_timeout(void)
1862 {
1863 	static thread_call_t mld_timeout_tcall;
1864 	uint64_t deadline = 0, leeway = 0;
1865 
1866 	MLD_LOCK_ASSERT_HELD();
1867 	if (mld_timeout_tcall == NULL) {
1868 		mld_timeout_tcall =
1869 		    thread_call_allocate_with_options(mld_timeout,
1870 		    NULL,
1871 		    THREAD_CALL_PRIORITY_KERNEL,
1872 		    THREAD_CALL_OPTIONS_ONCE);
1873 	}
1874 
1875 	if (!mld_timeout_run &&
1876 	    (querier_present_timers_running6 || current_state_timers_running6 ||
1877 	    interface_timers_running6 || state_change_timers_running6)) {
1878 		mld_timeout_run = true;
1879 		clock_interval_to_deadline(mld_timeout_delay, NSEC_PER_MSEC,
1880 		    &deadline);
1881 		clock_interval_to_absolutetime_interval(mld_timeout_leeway,
1882 		    NSEC_PER_MSEC, &leeway);
1883 		thread_call_enter_delayed_with_leeway(mld_timeout_tcall, NULL,
1884 		    deadline, leeway,
1885 		    THREAD_CALL_DELAY_LEEWAY);
1886 	}
1887 }
1888 
1889 static void
mld_sched_fast_timeout(void)1890 mld_sched_fast_timeout(void)
1891 {
1892 	static thread_call_t mld_fast_timeout_tcall;
1893 
1894 	MLD_LOCK_ASSERT_HELD();
1895 	if (mld_fast_timeout_tcall == NULL) {
1896 		mld_fast_timeout_tcall =
1897 		    thread_call_allocate_with_options(mld_timeout,
1898 		    mld_sched_fast_timeout,
1899 		    THREAD_CALL_PRIORITY_KERNEL,
1900 		    THREAD_CALL_OPTIONS_ONCE);
1901 	}
1902 	if (!mld_fast_timeout_run &&
1903 	    (current_state_timers_running6 || state_change_timers_running6)) {
1904 		mld_fast_timeout_run = true;
1905 		thread_call_enter(mld_fast_timeout_tcall);
1906 	}
1907 }
1908 
1909 /*
1910  * Appends an in6_multi to the list to be released later.
1911  *
1912  * Caller must be holding mli_lock.
1913  */
1914 static void
mld_append_relq(struct mld_ifinfo * mli,struct in6_multi * inm)1915 mld_append_relq(struct mld_ifinfo *mli, struct in6_multi *inm)
1916 {
1917 	MLI_LOCK_ASSERT_HELD(mli);
1918 	os_log(OS_LOG_DEFAULT, "%s: adding inm %llx on relq ifp %s\n",
1919 	    __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm),
1920 	    mli->mli_ifp != NULL ? if_name(mli->mli_ifp) : "<null>");
1921 	SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, in6m_nrele);
1922 }
1923 
1924 /*
1925  * Free the in6_multi reference(s) for this MLD lifecycle.
1926  *
1927  * Caller must be holding mli_lock.
1928  */
1929 static void
mld_flush_relq(struct mld_ifinfo * mli,struct mld_in6m_relhead * in6m_dthead)1930 mld_flush_relq(struct mld_ifinfo *mli, struct mld_in6m_relhead *in6m_dthead)
1931 {
1932 	struct in6_multi *inm;
1933 	SLIST_HEAD(, in6_multi) temp_relinmhead;
1934 
1935 	/*
1936 	 * Before dropping the mli_lock, copy all the items in the
1937 	 * release list to a temporary list to prevent other threads
1938 	 * from changing mli_relinmhead while we are traversing it.
1939 	 */
1940 	MLI_LOCK_ASSERT_HELD(mli);
1941 	SLIST_INIT(&temp_relinmhead);
1942 	while ((inm = SLIST_FIRST(&mli->mli_relinmhead)) != NULL) {
1943 		SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
1944 		SLIST_INSERT_HEAD(&temp_relinmhead, inm, in6m_nrele);
1945 	}
1946 	MLI_UNLOCK(mli);
1947 	in6_multihead_lock_exclusive();
1948 	while ((inm = SLIST_FIRST(&temp_relinmhead)) != NULL) {
1949 		int lastref;
1950 
1951 		SLIST_REMOVE_HEAD(&temp_relinmhead, in6m_nrele);
1952 		IN6M_LOCK(inm);
1953 		os_log(OS_LOG_DEFAULT, "%s: flushing inm %llx on relq ifp %s\n",
1954 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm),
1955 		    inm->in6m_ifp != NULL ? if_name(inm->in6m_ifp) : "<null>");
1956 		VERIFY(inm->in6m_nrelecnt != 0);
1957 		inm->in6m_nrelecnt--;
1958 		lastref = in6_multi_detach(inm);
1959 		VERIFY(!lastref || (!(inm->in6m_debug & IFD_ATTACHED) &&
1960 		    inm->in6m_reqcnt == 0));
1961 		IN6M_UNLOCK(inm);
1962 		/* from mli_relinmhead */
1963 		IN6M_REMREF(inm);
1964 		/* from in6_multihead_list */
1965 		if (lastref) {
1966 			/*
1967 			 * Defer releasing our final reference, as we
1968 			 * are holding the MLD lock at this point, and
1969 			 * we could end up with locking issues later on
1970 			 * (while issuing SIOCDELMULTI) when this is the
1971 			 * final reference count.  Let the caller do it
1972 			 * when it is safe.
1973 			 */
1974 			MLD_ADD_DETACHED_IN6M(in6m_dthead, inm);
1975 		}
1976 	}
1977 	in6_multihead_lock_done();
1978 	MLI_LOCK(mli);
1979 }
1980 
1981 /*
1982  * Update host report group timer.
1983  * Will update the global pending timer flags.
1984  */
1985 static void
mld_v1_process_group_timer(struct in6_multi * inm,const int mld_version)1986 mld_v1_process_group_timer(struct in6_multi *inm, const int mld_version)
1987 {
1988 #pragma unused(mld_version)
1989 	int report_timer_expired;
1990 
1991 	MLD_LOCK_ASSERT_HELD();
1992 	IN6M_LOCK_ASSERT_HELD(inm);
1993 	MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
1994 
1995 	if (inm->in6m_timer == 0) {
1996 		report_timer_expired = 0;
1997 	} else if (--inm->in6m_timer == 0) {
1998 		report_timer_expired = 1;
1999 	} else {
2000 		current_state_timers_running6 = 1;
2001 		/* caller will schedule timer */
2002 		return;
2003 	}
2004 
2005 	switch (inm->in6m_state) {
2006 	case MLD_NOT_MEMBER:
2007 	case MLD_SILENT_MEMBER:
2008 	case MLD_IDLE_MEMBER:
2009 	case MLD_LAZY_MEMBER:
2010 	case MLD_SLEEPING_MEMBER:
2011 	case MLD_AWAKENING_MEMBER:
2012 		break;
2013 	case MLD_REPORTING_MEMBER:
2014 		if (report_timer_expired) {
2015 			inm->in6m_state = MLD_IDLE_MEMBER;
2016 			(void) mld_v1_transmit_report(inm,
2017 			    MLD_LISTENER_REPORT);
2018 			IN6M_LOCK_ASSERT_HELD(inm);
2019 			MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
2020 		}
2021 		break;
2022 	case MLD_G_QUERY_PENDING_MEMBER:
2023 	case MLD_SG_QUERY_PENDING_MEMBER:
2024 	case MLD_LEAVING_MEMBER:
2025 		break;
2026 	}
2027 }
2028 
2029 /*
2030  * Update a group's timers for MLDv2.
2031  * Will update the global pending timer flags.
2032  * Note: Unlocked read from mli.
2033  */
2034 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)2035 mld_v2_process_group_timers(struct mld_ifinfo *mli,
2036     struct ifqueue *qrq, struct ifqueue *scq,
2037     struct in6_multi *inm, const int uri_sec)
2038 {
2039 	int query_response_timer_expired;
2040 	int state_change_retransmit_timer_expired;
2041 
2042 	MLD_LOCK_ASSERT_HELD();
2043 	IN6M_LOCK_ASSERT_HELD(inm);
2044 	MLI_LOCK_ASSERT_HELD(mli);
2045 	VERIFY(mli == inm->in6m_mli);
2046 
2047 	query_response_timer_expired = 0;
2048 	state_change_retransmit_timer_expired = 0;
2049 
2050 	/*
2051 	 * During a transition from compatibility mode back to MLDv2,
2052 	 * a group record in REPORTING state may still have its group
2053 	 * timer active. This is a no-op in this function; it is easier
2054 	 * to deal with it here than to complicate the timeout path.
2055 	 */
2056 	if (inm->in6m_timer == 0) {
2057 		query_response_timer_expired = 0;
2058 	} else if (--inm->in6m_timer == 0) {
2059 		query_response_timer_expired = 1;
2060 	} else {
2061 		current_state_timers_running6 = 1;
2062 		/* caller will schedule timer */
2063 	}
2064 
2065 	if (inm->in6m_sctimer == 0) {
2066 		state_change_retransmit_timer_expired = 0;
2067 	} else if (--inm->in6m_sctimer == 0) {
2068 		state_change_retransmit_timer_expired = 1;
2069 	} else {
2070 		state_change_timers_running6 = 1;
2071 		/* caller will schedule timer */
2072 	}
2073 
2074 	/* We are in timer callback, so be quick about it. */
2075 	if (!state_change_retransmit_timer_expired &&
2076 	    !query_response_timer_expired) {
2077 		return;
2078 	}
2079 
2080 	switch (inm->in6m_state) {
2081 	case MLD_NOT_MEMBER:
2082 	case MLD_SILENT_MEMBER:
2083 	case MLD_SLEEPING_MEMBER:
2084 	case MLD_LAZY_MEMBER:
2085 	case MLD_AWAKENING_MEMBER:
2086 	case MLD_IDLE_MEMBER:
2087 		break;
2088 	case MLD_G_QUERY_PENDING_MEMBER:
2089 	case MLD_SG_QUERY_PENDING_MEMBER:
2090 		/*
2091 		 * Respond to a previously pending Group-Specific
2092 		 * or Group-and-Source-Specific query by enqueueing
2093 		 * the appropriate Current-State report for
2094 		 * immediate transmission.
2095 		 */
2096 		if (query_response_timer_expired) {
2097 			int retval;
2098 
2099 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
2100 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
2101 			    0);
2102 			MLD_PRINTF(("%s: enqueue record = %d\n",
2103 			    __func__, retval));
2104 			inm->in6m_state = MLD_REPORTING_MEMBER;
2105 			in6m_clear_recorded(inm);
2106 		}
2107 		OS_FALLTHROUGH;
2108 	case MLD_REPORTING_MEMBER:
2109 	case MLD_LEAVING_MEMBER:
2110 		if (state_change_retransmit_timer_expired) {
2111 			/*
2112 			 * State-change retransmission timer fired.
2113 			 * If there are any further pending retransmissions,
2114 			 * set the global pending state-change flag, and
2115 			 * reset the timer.
2116 			 */
2117 			if (--inm->in6m_scrv > 0) {
2118 				inm->in6m_sctimer = (uint16_t)uri_sec;
2119 				state_change_timers_running6 = 1;
2120 				/* caller will schedule timer */
2121 			}
2122 			/*
2123 			 * Retransmit the previously computed state-change
2124 			 * report. If there are no further pending
2125 			 * retransmissions, the mbuf queue will be consumed.
2126 			 * Update T0 state to T1 as we have now sent
2127 			 * a state-change.
2128 			 */
2129 			(void) mld_v2_merge_state_changes(inm, scq);
2130 
2131 			in6m_commit(inm);
2132 			MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2133 			    ip6_sprintf(&inm->in6m_addr),
2134 			    if_name(inm->in6m_ifp)));
2135 
2136 			/*
2137 			 * If we are leaving the group for good, make sure
2138 			 * we release MLD's reference to it.
2139 			 * This release must be deferred using a SLIST,
2140 			 * as we are called from a loop which traverses
2141 			 * the in_ifmultiaddr TAILQ.
2142 			 */
2143 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
2144 			    inm->in6m_scrv == 0) {
2145 				inm->in6m_state = MLD_NOT_MEMBER;
2146 				/*
2147 				 * A reference has already been held in
2148 				 * mld_final_leave() for this inm, so
2149 				 * no need to hold another one.  We also
2150 				 * bumped up its request count then, so
2151 				 * that it stays in in6_multihead.  Both
2152 				 * of them will be released when it is
2153 				 * dequeued later on.
2154 				 */
2155 				VERIFY(inm->in6m_nrelecnt != 0);
2156 				mld_append_relq(mli, inm);
2157 			}
2158 		}
2159 		break;
2160 	}
2161 }
2162 
2163 /*
2164  * Switch to a different version on the given interface,
2165  * as per Section 9.12.
2166  */
2167 static uint32_t
mld_set_version(struct mld_ifinfo * mli,const int mld_version)2168 mld_set_version(struct mld_ifinfo *mli, const int mld_version)
2169 {
2170 	int old_version_timer;
2171 
2172 	MLI_LOCK_ASSERT_HELD(mli);
2173 
2174 	os_log(OS_LOG_DEFAULT, "%s: switching to v%d on ifp %s\n", __func__,
2175 	    mld_version, if_name(mli->mli_ifp));
2176 
2177 	if (mld_version == MLD_VERSION_1) {
2178 		/*
2179 		 * Compute the "Older Version Querier Present" timer as per
2180 		 * Section 9.12, in seconds.
2181 		 */
2182 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
2183 		mli->mli_v1_timer = old_version_timer;
2184 	}
2185 
2186 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
2187 		mli->mli_version = MLD_VERSION_1;
2188 		mld_v2_cancel_link_timers(mli);
2189 	}
2190 
2191 	MLI_LOCK_ASSERT_HELD(mli);
2192 
2193 	return mli->mli_v1_timer;
2194 }
2195 
2196 /*
2197  * Cancel pending MLDv2 timers for the given link and all groups
2198  * joined on it; state-change, general-query, and group-query timers.
2199  *
2200  * Only ever called on a transition from v2 to Compatibility mode. Kill
2201  * the timers stone dead (this may be expensive for large N groups), they
2202  * will be restarted if Compatibility Mode deems that they must be due to
2203  * query processing.
2204  */
2205 static void
mld_v2_cancel_link_timers(struct mld_ifinfo * mli)2206 mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
2207 {
2208 	struct ifnet            *ifp;
2209 	struct in6_multi        *inm;
2210 	struct in6_multistep    step;
2211 
2212 	MLI_LOCK_ASSERT_HELD(mli);
2213 
2214 	MLD_PRINTF(("%s: cancel v2 timers on ifp 0x%llx(%s)\n", __func__,
2215 	    (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp), if_name(mli->mli_ifp)));
2216 
2217 	/*
2218 	 * Stop the v2 General Query Response on this link stone dead.
2219 	 * If timer is woken up due to interface_timers_running6,
2220 	 * the flag will be cleared if there are no pending link timers.
2221 	 */
2222 	mli->mli_v2_timer = 0;
2223 
2224 	/*
2225 	 * Now clear the current-state and state-change report timers
2226 	 * for all memberships scoped to this link.
2227 	 */
2228 	ifp = mli->mli_ifp;
2229 	MLI_UNLOCK(mli);
2230 
2231 	in6_multihead_lock_shared();
2232 	IN6_FIRST_MULTI(step, inm);
2233 	while (inm != NULL) {
2234 		IN6M_LOCK(inm);
2235 		if (inm->in6m_ifp != ifp) {
2236 			goto next;
2237 		}
2238 
2239 		switch (inm->in6m_state) {
2240 		case MLD_NOT_MEMBER:
2241 		case MLD_SILENT_MEMBER:
2242 		case MLD_IDLE_MEMBER:
2243 		case MLD_LAZY_MEMBER:
2244 		case MLD_SLEEPING_MEMBER:
2245 		case MLD_AWAKENING_MEMBER:
2246 			/*
2247 			 * These states are either not relevant in v2 mode,
2248 			 * or are unreported. Do nothing.
2249 			 */
2250 			break;
2251 		case MLD_LEAVING_MEMBER:
2252 			/*
2253 			 * If we are leaving the group and switching
2254 			 * version, we need to release the final
2255 			 * reference held for issuing the INCLUDE {}.
2256 			 * During mld_final_leave(), we bumped up both the
2257 			 * request and reference counts.  Since we cannot
2258 			 * call in6_multi_detach() here, defer this task to
2259 			 * the timer routine.
2260 			 */
2261 			VERIFY(inm->in6m_nrelecnt != 0);
2262 			MLI_LOCK(mli);
2263 			mld_append_relq(mli, inm);
2264 			MLI_UNLOCK(mli);
2265 			OS_FALLTHROUGH;
2266 		case MLD_G_QUERY_PENDING_MEMBER:
2267 		case MLD_SG_QUERY_PENDING_MEMBER:
2268 			in6m_clear_recorded(inm);
2269 			OS_FALLTHROUGH;
2270 		case MLD_REPORTING_MEMBER:
2271 			inm->in6m_state = MLD_REPORTING_MEMBER;
2272 			break;
2273 		}
2274 		/*
2275 		 * Always clear state-change and group report timers.
2276 		 * Free any pending MLDv2 state-change records.
2277 		 */
2278 		inm->in6m_sctimer = 0;
2279 		inm->in6m_timer = 0;
2280 		IF_DRAIN(&inm->in6m_scq);
2281 next:
2282 		IN6M_UNLOCK(inm);
2283 		IN6_NEXT_MULTI(step, inm);
2284 	}
2285 	in6_multihead_lock_done();
2286 
2287 	MLI_LOCK(mli);
2288 }
2289 
2290 /*
2291  * Update the Older Version Querier Present timers for a link.
2292  * See Section 9.12 of RFC 3810.
2293  */
2294 static void
mld_v1_process_querier_timers(struct mld_ifinfo * mli)2295 mld_v1_process_querier_timers(struct mld_ifinfo *mli)
2296 {
2297 	MLI_LOCK_ASSERT_HELD(mli);
2298 
2299 	if (mld_v2enable && mli->mli_version != MLD_VERSION_2 &&
2300 	    --mli->mli_v1_timer == 0) {
2301 		/*
2302 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
2303 		 */
2304 		os_log(OS_LOG_DEFAULT, "%s: transition from v%d -> v%d on %s\n",
2305 		    __func__, mli->mli_version, MLD_VERSION_2,
2306 		    if_name(mli->mli_ifp));
2307 		mli->mli_version = MLD_VERSION_2;
2308 	}
2309 }
2310 
2311 /*
2312  * Transmit an MLDv1 report immediately.
2313  */
2314 static int
mld_v1_transmit_report(struct in6_multi * in6m,const uint8_t type)2315 mld_v1_transmit_report(struct in6_multi *in6m, const uint8_t type)
2316 {
2317 	struct ifnet            *ifp;
2318 	struct in6_ifaddr       *ia;
2319 	struct ip6_hdr          *ip6;
2320 	struct mbuf             *mh, *md;
2321 	struct mld_hdr          *mld;
2322 	int                     error = 0;
2323 
2324 	IN6M_LOCK_ASSERT_HELD(in6m);
2325 	MLI_LOCK_ASSERT_HELD(in6m->in6m_mli);
2326 
2327 	ifp = in6m->in6m_ifp;
2328 	/* ia may be NULL if link-local address is tentative. */
2329 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
2330 
2331 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
2332 	if (mh == NULL) {
2333 		if (ia != NULL) {
2334 			IFA_REMREF(&ia->ia_ifa);
2335 		}
2336 		return ENOMEM;
2337 	}
2338 	MGET(md, M_DONTWAIT, MT_DATA);
2339 	if (md == NULL) {
2340 		m_free(mh);
2341 		if (ia != NULL) {
2342 			IFA_REMREF(&ia->ia_ifa);
2343 		}
2344 		return ENOMEM;
2345 	}
2346 	mh->m_next = md;
2347 
2348 	/*
2349 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
2350 	 * that ether_output() does not need to allocate another mbuf
2351 	 * for the header in the most common case.
2352 	 */
2353 	MH_ALIGN(mh, sizeof(struct ip6_hdr));
2354 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
2355 	mh->m_len = sizeof(struct ip6_hdr);
2356 
2357 	ip6 = mtod(mh, struct ip6_hdr *);
2358 	ip6->ip6_flow = 0;
2359 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
2360 	ip6->ip6_vfc |= IPV6_VERSION;
2361 	ip6->ip6_nxt = IPPROTO_ICMPV6;
2362 	if (ia != NULL) {
2363 		IFA_LOCK(&ia->ia_ifa);
2364 	}
2365 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
2366 	ip6_output_setsrcifscope(mh, IFSCOPE_NONE, ia);
2367 	if (ia != NULL) {
2368 		IFA_UNLOCK(&ia->ia_ifa);
2369 		IFA_REMREF(&ia->ia_ifa);
2370 		ia = NULL;
2371 	}
2372 	ip6->ip6_dst = in6m->in6m_addr;
2373 	ip6_output_setdstifscope(mh, in6m->ifscope, NULL);
2374 
2375 	md->m_len = sizeof(struct mld_hdr);
2376 	mld = mtod(md, struct mld_hdr *);
2377 	mld->mld_type = type;
2378 	mld->mld_code = 0;
2379 	mld->mld_cksum = 0;
2380 	mld->mld_maxdelay = 0;
2381 	mld->mld_reserved = 0;
2382 	mld->mld_addr = in6m->in6m_addr;
2383 	in6_clearscope(&mld->mld_addr);
2384 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
2385 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
2386 
2387 	mld_save_context(mh, ifp);
2388 	mh->m_flags |= M_MLDV1;
2389 
2390 	/*
2391 	 * Due to the fact that at this point we are possibly holding
2392 	 * in6_multihead_lock in shared or exclusive mode, we can't call
2393 	 * mld_dispatch_packet() here since that will eventually call
2394 	 * ip6_output(), which will try to lock in6_multihead_lock and cause
2395 	 * a deadlock.
2396 	 * Instead we defer the work to the mld_timeout() thread, thus
2397 	 * avoiding unlocking in_multihead_lock here.
2398 	 */
2399 	if (IF_QFULL(&in6m->in6m_mli->mli_v1q)) {
2400 		os_log_error(OS_LOG_DEFAULT, "%s: v1 outbound queue full\n", __func__);
2401 		error = ENOMEM;
2402 		m_freem(mh);
2403 	} else {
2404 		IF_ENQUEUE(&in6m->in6m_mli->mli_v1q, mh);
2405 		VERIFY(error == 0);
2406 	}
2407 
2408 	return error;
2409 }
2410 
2411 /*
2412  * Process a state change from the upper layer for the given IPv6 group.
2413  *
2414  * Each socket holds a reference on the in6_multi in its own ip_moptions.
2415  * The socket layer will have made the necessary updates to.the group
2416  * state, it is now up to MLD to issue a state change report if there
2417  * has been any change between T0 (when the last state-change was issued)
2418  * and T1 (now).
2419  *
2420  * We use the MLDv2 state machine at group level. The MLd module
2421  * however makes the decision as to which MLD protocol version to speak.
2422  * A state change *from* INCLUDE {} always means an initial join.
2423  * A state change *to* INCLUDE {} always means a final leave.
2424  *
2425  * If delay is non-zero, and the state change is an initial multicast
2426  * join, the state change report will be delayed by 'delay' ticks
2427  * in units of seconds if MLDv1 is active on the link; otherwise
2428  * the initial MLDv2 state change report will be delayed by whichever
2429  * is sooner, a pending state-change timer or delay itself.
2430  */
2431 int
mld_change_state(struct in6_multi * inm,struct mld_tparams * mtp,const int delay)2432 mld_change_state(struct in6_multi *inm, struct mld_tparams *mtp,
2433     const int delay)
2434 {
2435 	struct mld_ifinfo *mli;
2436 	struct ifnet *ifp;
2437 	int error = 0;
2438 
2439 	VERIFY(mtp != NULL);
2440 	bzero(mtp, sizeof(*mtp));
2441 
2442 	IN6M_LOCK_ASSERT_HELD(inm);
2443 	VERIFY(inm->in6m_mli != NULL);
2444 	MLI_LOCK_ASSERT_NOTHELD(inm->in6m_mli);
2445 
2446 	/*
2447 	 * Try to detect if the upper layer just asked us to change state
2448 	 * for an interface which has now gone away.
2449 	 */
2450 	VERIFY(inm->in6m_ifma != NULL);
2451 	ifp = inm->in6m_ifma->ifma_ifp;
2452 	/*
2453 	 * Sanity check that netinet6's notion of ifp is the same as net's.
2454 	 */
2455 	VERIFY(inm->in6m_ifp == ifp);
2456 
2457 	mli = MLD_IFINFO(ifp);
2458 	VERIFY(mli != NULL);
2459 
2460 	/*
2461 	 * If we detect a state transition to or from MCAST_UNDEFINED
2462 	 * for this group, then we are starting or finishing an MLD
2463 	 * life cycle for this group.
2464 	 */
2465 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
2466 		MLD_PRINTF(("%s: inm transition %d -> %d\n", __func__,
2467 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode));
2468 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
2469 			MLD_PRINTF(("%s: initial join\n", __func__));
2470 			error = mld_initial_join(inm, mli, mtp, delay);
2471 			goto out;
2472 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
2473 			MLD_PRINTF(("%s: final leave\n", __func__));
2474 			mld_final_leave(inm, mli, mtp);
2475 			goto out;
2476 		}
2477 	} else {
2478 		MLD_PRINTF(("%s: filter set change\n", __func__));
2479 	}
2480 
2481 	error = mld_handle_state_change(inm, mli, mtp);
2482 out:
2483 	return error;
2484 }
2485 
2486 /*
2487  * Perform the initial join for an MLD group.
2488  *
2489  * When joining a group:
2490  *  If the group should have its MLD traffic suppressed, do nothing.
2491  *  MLDv1 starts sending MLDv1 host membership reports.
2492  *  MLDv2 will schedule an MLDv2 state-change report containing the
2493  *  initial state of the membership.
2494  *
2495  * If the delay argument is non-zero, then we must delay sending the
2496  * initial state change for delay ticks (in units of seconds).
2497  */
2498 static int
mld_initial_join(struct in6_multi * inm,struct mld_ifinfo * mli,struct mld_tparams * mtp,const int delay)2499 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
2500     struct mld_tparams *mtp, const int delay)
2501 {
2502 	struct ifnet            *ifp;
2503 	struct ifqueue          *ifq;
2504 	int                      error, retval, syncstates;
2505 	int                      odelay;
2506 
2507 	IN6M_LOCK_ASSERT_HELD(inm);
2508 	MLI_LOCK_ASSERT_NOTHELD(mli);
2509 	VERIFY(mtp != NULL);
2510 
2511 	MLD_PRINTF(("%s: initial join %s on ifp 0x%llx(%s)\n",
2512 	    __func__, ip6_sprintf(&inm->in6m_addr),
2513 	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2514 	    if_name(inm->in6m_ifp)));
2515 
2516 	error = 0;
2517 	syncstates = 1;
2518 
2519 	ifp = inm->in6m_ifp;
2520 
2521 	MLI_LOCK(mli);
2522 	VERIFY(mli->mli_ifp == ifp);
2523 
2524 	/*
2525 	 * Avoid MLD if group is :
2526 	 * 1. Joined on loopback, OR
2527 	 * 2. On a link that is marked MLIF_SILENT
2528 	 * 3. rdar://problem/19227650 Is link local scoped and
2529 	 *    on cellular interface
2530 	 * 4. Is a type that should not be reported (node local
2531 	 *    or all node link local multicast.
2532 	 * All other groups enter the appropriate state machine
2533 	 * for the version in use on this link.
2534 	 */
2535 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2536 	    (mli->mli_flags & MLIF_SILENT) ||
2537 	    (IFNET_IS_CELLULAR(ifp) &&
2538 	    (IN6_IS_ADDR_MC_LINKLOCAL(&inm->in6m_addr) || IN6_IS_ADDR_MC_UNICAST_BASED_LINKLOCAL(&inm->in6m_addr))) ||
2539 	    !mld_is_addr_reported(&inm->in6m_addr)) {
2540 		MLD_PRINTF(("%s: not kicking state machine for silent group\n",
2541 		    __func__));
2542 		inm->in6m_state = MLD_SILENT_MEMBER;
2543 		inm->in6m_timer = 0;
2544 	} else {
2545 		/*
2546 		 * Deal with overlapping in6_multi lifecycle.
2547 		 * If this group was LEAVING, then make sure
2548 		 * we drop the reference we picked up to keep the
2549 		 * group around for the final INCLUDE {} enqueue.
2550 		 * Since we cannot call in6_multi_detach() here,
2551 		 * defer this task to the timer routine.
2552 		 */
2553 		if (mli->mli_version == MLD_VERSION_2 &&
2554 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
2555 			VERIFY(inm->in6m_nrelecnt != 0);
2556 			mld_append_relq(mli, inm);
2557 		}
2558 
2559 		inm->in6m_state = MLD_REPORTING_MEMBER;
2560 
2561 		switch (mli->mli_version) {
2562 		case MLD_VERSION_1:
2563 			/*
2564 			 * If a delay was provided, only use it if
2565 			 * it is greater than the delay normally
2566 			 * used for an MLDv1 state change report,
2567 			 * and delay sending the initial MLDv1 report
2568 			 * by not transitioning to the IDLE state.
2569 			 */
2570 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI);
2571 			if (delay) {
2572 				inm->in6m_timer = max(delay, odelay);
2573 				mtp->cst = 1;
2574 			} else {
2575 				inm->in6m_state = MLD_IDLE_MEMBER;
2576 				error = mld_v1_transmit_report(inm,
2577 				    MLD_LISTENER_REPORT);
2578 
2579 				IN6M_LOCK_ASSERT_HELD(inm);
2580 				MLI_LOCK_ASSERT_HELD(mli);
2581 
2582 				if (error == 0) {
2583 					inm->in6m_timer = odelay;
2584 					mtp->cst = 1;
2585 				}
2586 			}
2587 			break;
2588 
2589 		case MLD_VERSION_2:
2590 			/*
2591 			 * Defer update of T0 to T1, until the first copy
2592 			 * of the state change has been transmitted.
2593 			 */
2594 			syncstates = 0;
2595 
2596 			/*
2597 			 * Immediately enqueue a State-Change Report for
2598 			 * this interface, freeing any previous reports.
2599 			 * Don't kick the timers if there is nothing to do,
2600 			 * or if an error occurred.
2601 			 */
2602 			ifq = &inm->in6m_scq;
2603 			IF_DRAIN(ifq);
2604 			retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2605 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2606 			mtp->cst = (ifq->ifq_len > 0);
2607 			MLD_PRINTF(("%s: enqueue record = %d\n",
2608 			    __func__, retval));
2609 			if (retval <= 0) {
2610 				error = retval * -1;
2611 				break;
2612 			}
2613 
2614 			/*
2615 			 * Schedule transmission of pending state-change
2616 			 * report up to RV times for this link. The timer
2617 			 * will fire at the next mld_timeout (1 second)),
2618 			 * giving us an opportunity to merge the reports.
2619 			 *
2620 			 * If a delay was provided to this function, only
2621 			 * use this delay if sooner than the existing one.
2622 			 */
2623 			VERIFY(mli->mli_rv > 1);
2624 			inm->in6m_scrv = (uint16_t)mli->mli_rv;
2625 			if (delay) {
2626 				if (inm->in6m_sctimer > 1) {
2627 					inm->in6m_sctimer =
2628 					    MIN(inm->in6m_sctimer, (uint16_t)delay);
2629 				} else {
2630 					inm->in6m_sctimer = (uint16_t)delay;
2631 				}
2632 			} else {
2633 				inm->in6m_sctimer = 1;
2634 			}
2635 			mtp->sct = 1;
2636 			error = 0;
2637 			break;
2638 		}
2639 	}
2640 	MLI_UNLOCK(mli);
2641 
2642 	/*
2643 	 * Only update the T0 state if state change is atomic,
2644 	 * i.e. we don't need to wait for a timer to fire before we
2645 	 * can consider the state change to have been communicated.
2646 	 */
2647 	if (syncstates) {
2648 		in6m_commit(inm);
2649 		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2650 		    ip6_sprintf(&inm->in6m_addr),
2651 		    if_name(inm->in6m_ifp)));
2652 	}
2653 
2654 	return error;
2655 }
2656 
2657 /*
2658  * Issue an intermediate state change during the life-cycle.
2659  */
2660 static int
mld_handle_state_change(struct in6_multi * inm,struct mld_ifinfo * mli,struct mld_tparams * mtp)2661 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli,
2662     struct mld_tparams *mtp)
2663 {
2664 	struct ifnet            *ifp;
2665 	int                      retval = 0;
2666 
2667 	IN6M_LOCK_ASSERT_HELD(inm);
2668 	MLI_LOCK_ASSERT_NOTHELD(mli);
2669 	VERIFY(mtp != NULL);
2670 
2671 	MLD_PRINTF(("%s: state change for %s on ifp 0x%llx(%s)\n",
2672 	    __func__, ip6_sprintf(&inm->in6m_addr),
2673 	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2674 	    if_name(inm->in6m_ifp)));
2675 
2676 	ifp = inm->in6m_ifp;
2677 
2678 	MLI_LOCK(mli);
2679 	VERIFY(mli->mli_ifp == ifp);
2680 
2681 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2682 	    (mli->mli_flags & MLIF_SILENT) ||
2683 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2684 	    (mli->mli_version != MLD_VERSION_2)) {
2685 		MLI_UNLOCK(mli);
2686 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2687 			MLD_PRINTF(("%s: not kicking state machine for silent "
2688 			    "group\n", __func__));
2689 		}
2690 		MLD_PRINTF(("%s: nothing to do\n", __func__));
2691 		in6m_commit(inm);
2692 		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2693 		    ip6_sprintf(&inm->in6m_addr),
2694 		    if_name(inm->in6m_ifp)));
2695 		goto done;
2696 	}
2697 
2698 	IF_DRAIN(&inm->in6m_scq);
2699 
2700 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2701 	    (mli->mli_flags & MLIF_USEALLOW));
2702 	mtp->cst = (inm->in6m_scq.ifq_len > 0);
2703 	MLD_PRINTF(("%s: enqueue record = %d\n", __func__, retval));
2704 	if (retval <= 0) {
2705 		MLI_UNLOCK(mli);
2706 		retval *= -1;
2707 		goto done;
2708 	} else {
2709 		retval = 0;
2710 	}
2711 
2712 	/*
2713 	 * If record(s) were enqueued, start the state-change
2714 	 * report timer for this group.
2715 	 */
2716 	inm->in6m_scrv = (uint16_t)mli->mli_rv;
2717 	inm->in6m_sctimer = 1;
2718 	mtp->sct = 1;
2719 	MLI_UNLOCK(mli);
2720 
2721 done:
2722 	return retval;
2723 }
2724 
2725 /*
2726  * Perform the final leave for a multicast address.
2727  *
2728  * When leaving a group:
2729  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2730  *  MLDv2 enqueues a state-change report containing a transition
2731  *  to INCLUDE {} for immediate transmission.
2732  */
2733 static void
mld_final_leave(struct in6_multi * inm,struct mld_ifinfo * mli,struct mld_tparams * mtp)2734 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli,
2735     struct mld_tparams *mtp)
2736 {
2737 	int syncstates = 1;
2738 
2739 	IN6M_LOCK_ASSERT_HELD(inm);
2740 	MLI_LOCK_ASSERT_NOTHELD(mli);
2741 	VERIFY(mtp != NULL);
2742 
2743 	MLD_PRINTF(("%s: final leave %s on ifp 0x%llx(%s)\n",
2744 	    __func__, ip6_sprintf(&inm->in6m_addr),
2745 	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2746 	    if_name(inm->in6m_ifp)));
2747 
2748 	switch (inm->in6m_state) {
2749 	case MLD_NOT_MEMBER:
2750 	case MLD_SILENT_MEMBER:
2751 	case MLD_LEAVING_MEMBER:
2752 		/* Already leaving or left; do nothing. */
2753 		MLD_PRINTF(("%s: not kicking state machine for silent group\n",
2754 		    __func__));
2755 		break;
2756 	case MLD_REPORTING_MEMBER:
2757 	case MLD_IDLE_MEMBER:
2758 	case MLD_G_QUERY_PENDING_MEMBER:
2759 	case MLD_SG_QUERY_PENDING_MEMBER:
2760 		MLI_LOCK(mli);
2761 		if (mli->mli_version == MLD_VERSION_1) {
2762 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2763 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
2764 				panic("%s: MLDv2 state reached, not MLDv2 "
2765 				    "mode\n", __func__);
2766 				/* NOTREACHED */
2767 			}
2768 			/* scheduler timer if enqueue is successful */
2769 			mtp->cst = (mld_v1_transmit_report(inm,
2770 			    MLD_LISTENER_DONE) == 0);
2771 
2772 			IN6M_LOCK_ASSERT_HELD(inm);
2773 			MLI_LOCK_ASSERT_HELD(mli);
2774 
2775 			inm->in6m_state = MLD_NOT_MEMBER;
2776 		} else if (mli->mli_version == MLD_VERSION_2) {
2777 			/*
2778 			 * Stop group timer and all pending reports.
2779 			 * Immediately enqueue a state-change report
2780 			 * TO_IN {} to be sent on the next timeout,
2781 			 * giving us an opportunity to merge reports.
2782 			 */
2783 			IF_DRAIN(&inm->in6m_scq);
2784 			inm->in6m_timer = 0;
2785 			inm->in6m_scrv = (uint16_t)mli->mli_rv;
2786 			MLD_PRINTF(("%s: Leaving %s/%s with %d "
2787 			    "pending retransmissions.\n", __func__,
2788 			    ip6_sprintf(&inm->in6m_addr),
2789 			    if_name(inm->in6m_ifp),
2790 			    inm->in6m_scrv));
2791 			if (inm->in6m_scrv == 0) {
2792 				inm->in6m_state = MLD_NOT_MEMBER;
2793 				inm->in6m_sctimer = 0;
2794 			} else {
2795 				int retval;
2796 				/*
2797 				 * Stick around in the in6_multihead list;
2798 				 * the final detach will be issued by
2799 				 * mld_v2_process_group_timers() when
2800 				 * the retransmit timer expires.
2801 				 */
2802 				IN6M_ADDREF_LOCKED(inm);
2803 				VERIFY(inm->in6m_debug & IFD_ATTACHED);
2804 				inm->in6m_reqcnt++;
2805 				VERIFY(inm->in6m_reqcnt >= 1);
2806 				inm->in6m_nrelecnt++;
2807 				VERIFY(inm->in6m_nrelecnt != 0);
2808 
2809 				retval = mld_v2_enqueue_group_record(
2810 					&inm->in6m_scq, inm, 1, 0, 0,
2811 					(mli->mli_flags & MLIF_USEALLOW));
2812 				mtp->cst = (inm->in6m_scq.ifq_len > 0);
2813 				KASSERT(retval != 0,
2814 				    ("%s: enqueue record = %d\n", __func__,
2815 				    retval));
2816 
2817 				inm->in6m_state = MLD_LEAVING_MEMBER;
2818 				inm->in6m_sctimer = 1;
2819 				mtp->sct = 1;
2820 				syncstates = 0;
2821 			}
2822 		}
2823 		MLI_UNLOCK(mli);
2824 		break;
2825 	case MLD_LAZY_MEMBER:
2826 	case MLD_SLEEPING_MEMBER:
2827 	case MLD_AWAKENING_MEMBER:
2828 		/* Our reports are suppressed; do nothing. */
2829 		break;
2830 	}
2831 
2832 	if (syncstates) {
2833 		in6m_commit(inm);
2834 		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2835 		    ip6_sprintf(&inm->in6m_addr),
2836 		    if_name(inm->in6m_ifp)));
2837 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2838 		MLD_PRINTF(("%s: T1 now MCAST_UNDEFINED for 0x%llx/%s\n",
2839 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(&inm->in6m_addr),
2840 		    if_name(inm->in6m_ifp)));
2841 	}
2842 }
2843 
2844 /*
2845  * Enqueue an MLDv2 group record to the given output queue.
2846  *
2847  * If is_state_change is zero, a current-state record is appended.
2848  * If is_state_change is non-zero, a state-change report is appended.
2849  *
2850  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2851  * If is_group_query is zero, and if there is a packet with free space
2852  * at the tail of the queue, it will be appended to providing there
2853  * is enough free space.
2854  * Otherwise a new mbuf packet chain is allocated.
2855  *
2856  * If is_source_query is non-zero, each source is checked to see if
2857  * it was recorded for a Group-Source query, and will be omitted if
2858  * it is not both in-mode and recorded.
2859  *
2860  * If use_block_allow is non-zero, state change reports for initial join
2861  * and final leave, on an inclusive mode group with a source list, will be
2862  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2863  *
2864  * The function will attempt to allocate leading space in the packet
2865  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2866  *
2867  * If successful the size of all data appended to the queue is returned,
2868  * otherwise an error code less than zero is returned, or zero if
2869  * no record(s) were appended.
2870  */
2871 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)2872 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2873     const int is_state_change, const int is_group_query,
2874     const int is_source_query, const int use_block_allow)
2875 {
2876 	struct mldv2_record      mr;
2877 	struct mldv2_record     *pmr;
2878 	struct ifnet            *ifp;
2879 	struct ip6_msource      *ims, *nims;
2880 	struct mbuf             *m0, *m, *md;
2881 	int                      error, is_filter_list_change;
2882 	int                      minrec0len, m0srcs, msrcs, nbytes, off;
2883 	int                      record_has_sources;
2884 	int                      now;
2885 	uint8_t                  type;
2886 	uint8_t                  mode;
2887 
2888 	IN6M_LOCK_ASSERT_HELD(inm);
2889 	MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
2890 
2891 	error = 0;
2892 	ifp = inm->in6m_ifp;
2893 	is_filter_list_change = 0;
2894 	m = NULL;
2895 	m0 = NULL;
2896 	m0srcs = 0;
2897 	msrcs = 0;
2898 	nbytes = 0;
2899 	nims = NULL;
2900 	record_has_sources = 1;
2901 	pmr = NULL;
2902 	type = MLD_DO_NOTHING;
2903 	mode = (uint8_t)inm->in6m_st[1].iss_fmode;
2904 
2905 	/*
2906 	 * If we did not transition out of ASM mode during t0->t1,
2907 	 * and there are no source nodes to process, we can skip
2908 	 * the generation of source records.
2909 	 */
2910 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2911 	    inm->in6m_nsrc == 0) {
2912 		record_has_sources = 0;
2913 	}
2914 
2915 	if (is_state_change) {
2916 		/*
2917 		 * Queue a state change record.
2918 		 * If the mode did not change, and there are non-ASM
2919 		 * listeners or source filters present,
2920 		 * we potentially need to issue two records for the group.
2921 		 * If there are ASM listeners, and there was no filter
2922 		 * mode transition of any kind, do nothing.
2923 		 *
2924 		 * If we are transitioning to MCAST_UNDEFINED, we need
2925 		 * not send any sources. A transition to/from this state is
2926 		 * considered inclusive with some special treatment.
2927 		 *
2928 		 * If we are rewriting initial joins/leaves to use
2929 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2930 		 * we need to send sources in all cases.
2931 		 */
2932 		if (mode != inm->in6m_st[0].iss_fmode) {
2933 			if (mode == MCAST_EXCLUDE) {
2934 				MLD_PRINTF(("%s: change to EXCLUDE\n",
2935 				    __func__));
2936 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2937 			} else {
2938 				MLD_PRINTF(("%s: change to INCLUDE\n",
2939 				    __func__));
2940 				if (use_block_allow) {
2941 					/*
2942 					 * XXX
2943 					 * Here we're interested in state
2944 					 * edges either direction between
2945 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2946 					 * Perhaps we should just check
2947 					 * the group state, rather than
2948 					 * the filter mode.
2949 					 */
2950 					if (mode == MCAST_UNDEFINED) {
2951 						type = MLD_BLOCK_OLD_SOURCES;
2952 					} else {
2953 						type = MLD_ALLOW_NEW_SOURCES;
2954 					}
2955 				} else {
2956 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2957 					if (mode == MCAST_UNDEFINED) {
2958 						record_has_sources = 0;
2959 					}
2960 				}
2961 			}
2962 		} else {
2963 			if (record_has_sources) {
2964 				is_filter_list_change = 1;
2965 			} else {
2966 				type = MLD_DO_NOTHING;
2967 			}
2968 		}
2969 	} else {
2970 		/*
2971 		 * Queue a current state record.
2972 		 */
2973 		if (mode == MCAST_EXCLUDE) {
2974 			type = MLD_MODE_IS_EXCLUDE;
2975 		} else if (mode == MCAST_INCLUDE) {
2976 			type = MLD_MODE_IS_INCLUDE;
2977 			VERIFY(inm->in6m_st[1].iss_asm == 0);
2978 		}
2979 	}
2980 
2981 	/*
2982 	 * Generate the filter list changes using a separate function.
2983 	 */
2984 	if (is_filter_list_change) {
2985 		return mld_v2_enqueue_filter_change(ifq, inm);
2986 	}
2987 
2988 	if (type == MLD_DO_NOTHING) {
2989 		MLD_PRINTF(("%s: nothing to do for %s/%s\n",
2990 		    __func__, ip6_sprintf(&inm->in6m_addr),
2991 		    if_name(inm->in6m_ifp)));
2992 		return 0;
2993 	}
2994 
2995 	/*
2996 	 * If any sources are present, we must be able to fit at least
2997 	 * one in the trailing space of the tail packet's mbuf,
2998 	 * ideally more.
2999 	 */
3000 	minrec0len = sizeof(struct mldv2_record);
3001 	if (record_has_sources) {
3002 		minrec0len += sizeof(struct in6_addr);
3003 	}
3004 	MLD_PRINTF(("%s: queueing %s for %s/%s\n", __func__,
3005 	    mld_rec_type_to_str(type),
3006 	    ip6_sprintf(&inm->in6m_addr),
3007 	    if_name(inm->in6m_ifp)));
3008 
3009 	/*
3010 	 * Check if we have a packet in the tail of the queue for this
3011 	 * group into which the first group record for this group will fit.
3012 	 * Otherwise allocate a new packet.
3013 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
3014 	 * Note: Group records for G/GSR query responses MUST be sent
3015 	 * in their own packet.
3016 	 */
3017 	m0 = ifq->ifq_tail;
3018 	if (!is_group_query &&
3019 	    m0 != NULL &&
3020 	    (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
3021 	    (m0->m_pkthdr.len + minrec0len) <
3022 	    (ifp->if_mtu - MLD_MTUSPACE)) {
3023 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3024 		    sizeof(struct mldv2_record)) /
3025 		    sizeof(struct in6_addr);
3026 		m = m0;
3027 		MLD_PRINTF(("%s: use existing packet\n", __func__));
3028 	} else {
3029 		if (IF_QFULL(ifq)) {
3030 			os_log_error(OS_LOG_DEFAULT,
3031 			    "%s: outbound queue full\n", __func__);
3032 			return -ENOMEM;
3033 		}
3034 		m = NULL;
3035 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3036 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
3037 		if (!is_state_change && !is_group_query) {
3038 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3039 		}
3040 		if (m == NULL) {
3041 			m = m_gethdr(M_DONTWAIT, MT_DATA);
3042 		}
3043 		if (m == NULL) {
3044 			return -ENOMEM;
3045 		}
3046 
3047 		mld_save_context(m, ifp);
3048 
3049 		MLD_PRINTF(("%s: allocated first packet\n", __func__));
3050 	}
3051 
3052 	/*
3053 	 * Append group record.
3054 	 * If we have sources, we don't know how many yet.
3055 	 */
3056 	mr.mr_type = type;
3057 	mr.mr_datalen = 0;
3058 	mr.mr_numsrc = 0;
3059 	mr.mr_addr = inm->in6m_addr;
3060 	in6_clearscope(&mr.mr_addr);
3061 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
3062 		if (m != m0) {
3063 			m_freem(m);
3064 		}
3065 		os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed.\n", __func__);
3066 		return -ENOMEM;
3067 	}
3068 	nbytes += sizeof(struct mldv2_record);
3069 
3070 	/*
3071 	 * Append as many sources as will fit in the first packet.
3072 	 * If we are appending to a new packet, the chain allocation
3073 	 * may potentially use clusters; use m_getptr() in this case.
3074 	 * If we are appending to an existing packet, we need to obtain
3075 	 * a pointer to the group record after m_append(), in case a new
3076 	 * mbuf was allocated.
3077 	 *
3078 	 * Only append sources which are in-mode at t1. If we are
3079 	 * transitioning to MCAST_UNDEFINED state on the group, and
3080 	 * use_block_allow is zero, do not include source entries.
3081 	 * Otherwise, we need to include this source in the report.
3082 	 *
3083 	 * Only report recorded sources in our filter set when responding
3084 	 * to a group-source query.
3085 	 */
3086 	if (record_has_sources) {
3087 		if (m == m0) {
3088 			md = m_last(m);
3089 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
3090 			    md->m_len - nbytes);
3091 		} else {
3092 			md = m_getptr(m, 0, &off);
3093 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
3094 			    off);
3095 		}
3096 		msrcs = 0;
3097 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
3098 		    nims) {
3099 			MLD_PRINTF(("%s: visit node %s\n", __func__,
3100 			    ip6_sprintf(&ims->im6s_addr)));
3101 			now = im6s_get_mode(inm, ims, 1);
3102 			MLD_PRINTF(("%s: node is %d\n", __func__, now));
3103 			if ((now != mode) ||
3104 			    (now == mode &&
3105 			    (!use_block_allow && mode == MCAST_UNDEFINED))) {
3106 				MLD_PRINTF(("%s: skip node\n", __func__));
3107 				continue;
3108 			}
3109 			if (is_source_query && ims->im6s_stp == 0) {
3110 				MLD_PRINTF(("%s: skip unrecorded node\n",
3111 				    __func__));
3112 				continue;
3113 			}
3114 			MLD_PRINTF(("%s: append node\n", __func__));
3115 			if (!m_append(m, sizeof(struct in6_addr),
3116 			    (void *)&ims->im6s_addr)) {
3117 				if (m != m0) {
3118 					m_freem(m);
3119 				}
3120 				os_log_error(OS_LOG_DEFAULT,
3121 				    "%s: m_append() failed\n",
3122 				    __func__);
3123 				return -ENOMEM;
3124 			}
3125 			nbytes += sizeof(struct in6_addr);
3126 			++msrcs;
3127 			if (msrcs == m0srcs) {
3128 				break;
3129 			}
3130 		}
3131 		MLD_PRINTF(("%s: msrcs is %d this packet\n", __func__,
3132 		    msrcs));
3133 		pmr->mr_numsrc = htons((uint16_t)msrcs);
3134 		nbytes += (msrcs * sizeof(struct in6_addr));
3135 	}
3136 
3137 	if (is_source_query && msrcs == 0) {
3138 		MLD_PRINTF(("%s: no recorded sources to report\n", __func__));
3139 		if (m != m0) {
3140 			m_freem(m);
3141 		}
3142 		return 0;
3143 	}
3144 
3145 	/*
3146 	 * We are good to go with first packet.
3147 	 */
3148 	if (m != m0) {
3149 		MLD_PRINTF(("%s: enqueueing first packet\n", __func__));
3150 		m->m_pkthdr.vt_nrecs = 1;
3151 		IF_ENQUEUE(ifq, m);
3152 	} else {
3153 		m->m_pkthdr.vt_nrecs++;
3154 	}
3155 	/*
3156 	 * No further work needed if no source list in packet(s).
3157 	 */
3158 	if (!record_has_sources) {
3159 		return nbytes;
3160 	}
3161 
3162 	/*
3163 	 * Whilst sources remain to be announced, we need to allocate
3164 	 * a new packet and fill out as many sources as will fit.
3165 	 * Always try for a cluster first.
3166 	 */
3167 	while (nims != NULL) {
3168 		if (IF_QFULL(ifq)) {
3169 			os_log_error(OS_LOG_DEFAULT, "%s: outbound queue full\n", __func__);
3170 			return -ENOMEM;
3171 		}
3172 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3173 		if (m == NULL) {
3174 			m = m_gethdr(M_DONTWAIT, MT_DATA);
3175 		}
3176 		if (m == NULL) {
3177 			return -ENOMEM;
3178 		}
3179 		mld_save_context(m, ifp);
3180 		md = m_getptr(m, 0, &off);
3181 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
3182 		MLD_PRINTF(("%s: allocated next packet\n", __func__));
3183 
3184 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
3185 			if (m != m0) {
3186 				m_freem(m);
3187 			}
3188 			os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed.\n", __func__);
3189 			return -ENOMEM;
3190 		}
3191 		m->m_pkthdr.vt_nrecs = 1;
3192 		nbytes += sizeof(struct mldv2_record);
3193 
3194 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3195 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
3196 
3197 		msrcs = 0;
3198 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
3199 			MLD_PRINTF(("%s: visit node %s\n",
3200 			    __func__, ip6_sprintf(&ims->im6s_addr)));
3201 			now = im6s_get_mode(inm, ims, 1);
3202 			if ((now != mode) ||
3203 			    (now == mode &&
3204 			    (!use_block_allow && mode == MCAST_UNDEFINED))) {
3205 				MLD_PRINTF(("%s: skip node\n", __func__));
3206 				continue;
3207 			}
3208 			if (is_source_query && ims->im6s_stp == 0) {
3209 				MLD_PRINTF(("%s: skip unrecorded node\n",
3210 				    __func__));
3211 				continue;
3212 			}
3213 			MLD_PRINTF(("%s: append node\n", __func__));
3214 			if (!m_append(m, sizeof(struct in6_addr),
3215 			    (void *)&ims->im6s_addr)) {
3216 				if (m != m0) {
3217 					m_freem(m);
3218 				}
3219 				os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed\n",
3220 				    __func__);
3221 				return -ENOMEM;
3222 			}
3223 			++msrcs;
3224 			if (msrcs == m0srcs) {
3225 				break;
3226 			}
3227 		}
3228 		pmr->mr_numsrc = htons((uint16_t)msrcs);
3229 		nbytes += (msrcs * sizeof(struct in6_addr));
3230 
3231 		MLD_PRINTF(("%s: enqueueing next packet\n", __func__));
3232 		IF_ENQUEUE(ifq, m);
3233 	}
3234 
3235 	return nbytes;
3236 }
3237 
3238 /*
3239  * Type used to mark record pass completion.
3240  * We exploit the fact we can cast to this easily from the
3241  * current filter modes on each ip_msource node.
3242  */
3243 typedef enum {
3244 	REC_NONE = 0x00,        /* MCAST_UNDEFINED */
3245 	REC_ALLOW = 0x01,       /* MCAST_INCLUDE */
3246 	REC_BLOCK = 0x02,       /* MCAST_EXCLUDE */
3247 	REC_FULL = REC_ALLOW | REC_BLOCK
3248 } rectype_t;
3249 
3250 /*
3251  * Enqueue an MLDv2 filter list change to the given output queue.
3252  *
3253  * Source list filter state is held in an RB-tree. When the filter list
3254  * for a group is changed without changing its mode, we need to compute
3255  * the deltas between T0 and T1 for each source in the filter set,
3256  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3257  *
3258  * As we may potentially queue two record types, and the entire R-B tree
3259  * needs to be walked at once, we break this out into its own function
3260  * so we can generate a tightly packed queue of packets.
3261  *
3262  * XXX This could be written to only use one tree walk, although that makes
3263  * serializing into the mbuf chains a bit harder. For now we do two walks
3264  * which makes things easier on us, and it may or may not be harder on
3265  * the L2 cache.
3266  *
3267  * If successful the size of all data appended to the queue is returned,
3268  * otherwise an error code less than zero is returned, or zero if
3269  * no record(s) were appended.
3270  */
3271 static int
mld_v2_enqueue_filter_change(struct ifqueue * ifq,struct in6_multi * inm)3272 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
3273 {
3274 	static const int MINRECLEN =
3275 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
3276 	struct ifnet            *ifp;
3277 	struct mldv2_record      mr;
3278 	struct mldv2_record     *pmr;
3279 	struct ip6_msource      *ims, *nims;
3280 	struct mbuf             *m, *m0, *md;
3281 	int                      m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3282 	int                      nallow, nblock;
3283 	uint8_t                  mode, now, then;
3284 	rectype_t                crt, drt, nrt;
3285 
3286 	IN6M_LOCK_ASSERT_HELD(inm);
3287 
3288 	if (inm->in6m_nsrc == 0 ||
3289 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) {
3290 		return 0;
3291 	}
3292 
3293 	ifp = inm->in6m_ifp;                    /* interface */
3294 	mode = (uint8_t)inm->in6m_st[1].iss_fmode;       /* filter mode at t1 */
3295 	crt = REC_NONE; /* current group record type */
3296 	drt = REC_NONE; /* mask of completed group record types */
3297 	nrt = REC_NONE; /* record type for current node */
3298 	m0srcs = 0;     /* # source which will fit in current mbuf chain */
3299 	npbytes = 0;    /* # of bytes appended this packet */
3300 	nbytes = 0;     /* # of bytes appended to group's state-change queue */
3301 	rsrcs = 0;      /* # sources encoded in current record */
3302 	schanged = 0;   /* # nodes encoded in overall filter change */
3303 	nallow = 0;     /* # of source entries in ALLOW_NEW */
3304 	nblock = 0;     /* # of source entries in BLOCK_OLD */
3305 	nims = NULL;    /* next tree node pointer */
3306 
3307 	/*
3308 	 * For each possible filter record mode.
3309 	 * The first kind of source we encounter tells us which
3310 	 * is the first kind of record we start appending.
3311 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3312 	 * as the inverse of the group's filter mode.
3313 	 */
3314 	while (drt != REC_FULL) {
3315 		do {
3316 			m0 = ifq->ifq_tail;
3317 			if (m0 != NULL &&
3318 			    (m0->m_pkthdr.vt_nrecs + 1 <=
3319 			    MLD_V2_REPORT_MAXRECS) &&
3320 			    (m0->m_pkthdr.len + MINRECLEN) <
3321 			    (ifp->if_mtu - MLD_MTUSPACE)) {
3322 				m = m0;
3323 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3324 				    sizeof(struct mldv2_record)) /
3325 				    sizeof(struct in6_addr);
3326 				MLD_PRINTF(("%s: use previous packet\n",
3327 				    __func__));
3328 			} else {
3329 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3330 				if (m == NULL) {
3331 					m = m_gethdr(M_DONTWAIT, MT_DATA);
3332 				}
3333 				if (m == NULL) {
3334 					os_log_error(OS_LOG_DEFAULT, "%s: m_get*() failed\n",
3335 					    __func__);
3336 					return -ENOMEM;
3337 				}
3338 				m->m_pkthdr.vt_nrecs = 0;
3339 				mld_save_context(m, ifp);
3340 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3341 				    sizeof(struct mldv2_record)) /
3342 				    sizeof(struct in6_addr);
3343 				npbytes = 0;
3344 				MLD_PRINTF(("%s: allocated new packet\n",
3345 				    __func__));
3346 			}
3347 			/*
3348 			 * Append the MLD group record header to the
3349 			 * current packet's data area.
3350 			 * Recalculate pointer to free space for next
3351 			 * group record, in case m_append() allocated
3352 			 * a new mbuf or cluster.
3353 			 */
3354 			memset(&mr, 0, sizeof(mr));
3355 			mr.mr_addr = inm->in6m_addr;
3356 			in6_clearscope(&mr.mr_addr);
3357 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
3358 				if (m != m0) {
3359 					m_freem(m);
3360 				}
3361 				os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed\n",
3362 				    __func__);
3363 				return -ENOMEM;
3364 			}
3365 			npbytes += sizeof(struct mldv2_record);
3366 			if (m != m0) {
3367 				/* new packet; offset in chain */
3368 				md = m_getptr(m, npbytes -
3369 				    sizeof(struct mldv2_record), &off);
3370 				pmr = (struct mldv2_record *)(mtod(md,
3371 				    uint8_t *) + off);
3372 			} else {
3373 				/* current packet; offset from last append */
3374 				md = m_last(m);
3375 				pmr = (struct mldv2_record *)(mtod(md,
3376 				    uint8_t *) + md->m_len -
3377 				    sizeof(struct mldv2_record));
3378 			}
3379 			/*
3380 			 * Begin walking the tree for this record type
3381 			 * pass, or continue from where we left off
3382 			 * previously if we had to allocate a new packet.
3383 			 * Only report deltas in-mode at t1.
3384 			 * We need not report included sources as allowed
3385 			 * if we are in inclusive mode on the group,
3386 			 * however the converse is not true.
3387 			 */
3388 			rsrcs = 0;
3389 			if (nims == NULL) {
3390 				nims = RB_MIN(ip6_msource_tree,
3391 				    &inm->in6m_srcs);
3392 			}
3393 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
3394 				MLD_PRINTF(("%s: visit node %s\n", __func__,
3395 				    ip6_sprintf(&ims->im6s_addr)));
3396 				now = im6s_get_mode(inm, ims, 1);
3397 				then = im6s_get_mode(inm, ims, 0);
3398 				MLD_PRINTF(("%s: mode: t0 %d, t1 %d\n",
3399 				    __func__, then, now));
3400 				if (now == then) {
3401 					MLD_PRINTF(("%s: skip unchanged\n",
3402 					    __func__));
3403 					continue;
3404 				}
3405 				if (mode == MCAST_EXCLUDE &&
3406 				    now == MCAST_INCLUDE) {
3407 					MLD_PRINTF(("%s: skip IN src on EX "
3408 					    "group\n", __func__));
3409 					continue;
3410 				}
3411 				nrt = (rectype_t)now;
3412 				if (nrt == REC_NONE) {
3413 					nrt = (rectype_t)(~mode & REC_FULL);
3414 				}
3415 				if (schanged++ == 0) {
3416 					crt = nrt;
3417 				} else if (crt != nrt) {
3418 					continue;
3419 				}
3420 				if (!m_append(m, sizeof(struct in6_addr),
3421 				    (void *)&ims->im6s_addr)) {
3422 					if (m != m0) {
3423 						m_freem(m);
3424 					}
3425 					os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed\n",
3426 					    __func__);
3427 					return -ENOMEM;
3428 				}
3429 				nallow += !!(crt == REC_ALLOW);
3430 				nblock += !!(crt == REC_BLOCK);
3431 				if (++rsrcs == m0srcs) {
3432 					break;
3433 				}
3434 			}
3435 			/*
3436 			 * If we did not append any tree nodes on this
3437 			 * pass, back out of allocations.
3438 			 */
3439 			if (rsrcs == 0) {
3440 				npbytes -= sizeof(struct mldv2_record);
3441 				if (m != m0) {
3442 					MLD_PRINTF(("%s: m_free(m)\n",
3443 					    __func__));
3444 					m_freem(m);
3445 				} else {
3446 					MLD_PRINTF(("%s: m_adj(m, -mr)\n",
3447 					    __func__));
3448 					m_adj(m, -((int)sizeof(
3449 						    struct mldv2_record)));
3450 				}
3451 				continue;
3452 			}
3453 			npbytes += (rsrcs * sizeof(struct in6_addr));
3454 			if (crt == REC_ALLOW) {
3455 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
3456 			} else if (crt == REC_BLOCK) {
3457 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
3458 			}
3459 			pmr->mr_numsrc = htons((uint16_t)rsrcs);
3460 			/*
3461 			 * Count the new group record, and enqueue this
3462 			 * packet if it wasn't already queued.
3463 			 */
3464 			m->m_pkthdr.vt_nrecs++;
3465 			if (m != m0) {
3466 				IF_ENQUEUE(ifq, m);
3467 			}
3468 			nbytes += npbytes;
3469 		} while (nims != NULL);
3470 		drt |= crt;
3471 		crt = (~crt & REC_FULL);
3472 	}
3473 
3474 	MLD_PRINTF(("%s: queued %d ALLOW_NEW, %d BLOCK_OLD\n", __func__,
3475 	    nallow, nblock));
3476 
3477 	return nbytes;
3478 }
3479 
3480 static int
mld_v2_merge_state_changes(struct in6_multi * inm,struct ifqueue * ifscq)3481 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
3482 {
3483 	struct ifqueue  *gq;
3484 	struct mbuf     *m;             /* pending state-change */
3485 	struct mbuf     *m0;            /* copy of pending state-change */
3486 	struct mbuf     *mt;            /* last state-change in packet */
3487 	struct mbuf     *n;
3488 	int              docopy, domerge;
3489 	u_int            recslen;
3490 
3491 	IN6M_LOCK_ASSERT_HELD(inm);
3492 
3493 	docopy = 0;
3494 	domerge = 0;
3495 	recslen = 0;
3496 
3497 	/*
3498 	 * If there are further pending retransmissions, make a writable
3499 	 * copy of each queued state-change message before merging.
3500 	 */
3501 	if (inm->in6m_scrv > 0) {
3502 		docopy = 1;
3503 	}
3504 
3505 	gq = &inm->in6m_scq;
3506 #ifdef MLD_DEBUG
3507 	if (gq->ifq_head == NULL) {
3508 		MLD_PRINTF(("%s: WARNING: queue for inm 0x%llx is empty\n",
3509 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm)));
3510 	}
3511 #endif
3512 
3513 	/*
3514 	 * Use IF_REMQUEUE() instead of IF_DEQUEUE() below, since the
3515 	 * packet might not always be at the head of the ifqueue.
3516 	 */
3517 	m = gq->ifq_head;
3518 	while (m != NULL) {
3519 		/*
3520 		 * Only merge the report into the current packet if
3521 		 * there is sufficient space to do so; an MLDv2 report
3522 		 * packet may only contain 65,535 group records.
3523 		 * Always use a simple mbuf chain concatentation to do this,
3524 		 * as large state changes for single groups may have
3525 		 * allocated clusters.
3526 		 */
3527 		domerge = 0;
3528 		mt = ifscq->ifq_tail;
3529 		if (mt != NULL) {
3530 			recslen = m_length(m);
3531 
3532 			if ((mt->m_pkthdr.vt_nrecs +
3533 			    m->m_pkthdr.vt_nrecs <=
3534 			    MLD_V2_REPORT_MAXRECS) &&
3535 			    (mt->m_pkthdr.len + recslen <=
3536 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) {
3537 				domerge = 1;
3538 			}
3539 		}
3540 
3541 		if (!domerge && IF_QFULL(gq)) {
3542 			os_log_info(OS_LOG_DEFAULT, "%s: outbound queue full",
3543 			    __func__);
3544 			n = m->m_nextpkt;
3545 			if (!docopy) {
3546 				IF_REMQUEUE(gq, m);
3547 				m_freem(m);
3548 			}
3549 			m = n;
3550 			continue;
3551 		}
3552 
3553 		if (!docopy) {
3554 			MLD_PRINTF(("%s: dequeueing 0x%llx\n", __func__,
3555 			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3556 			n = m->m_nextpkt;
3557 			IF_REMQUEUE(gq, m);
3558 			m0 = m;
3559 			m = n;
3560 		} else {
3561 			MLD_PRINTF(("%s: copying 0x%llx\n", __func__,
3562 			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3563 			m0 = m_dup(m, M_NOWAIT);
3564 			if (m0 == NULL) {
3565 				return ENOMEM;
3566 			}
3567 			m0->m_nextpkt = NULL;
3568 			m = m->m_nextpkt;
3569 		}
3570 
3571 		if (!domerge) {
3572 			MLD_PRINTF(("%s: queueing 0x%llx to ifscq 0x%llx)\n",
3573 			    __func__, (uint64_t)VM_KERNEL_ADDRPERM(m0),
3574 			    (uint64_t)VM_KERNEL_ADDRPERM(ifscq)));
3575 			IF_ENQUEUE(ifscq, m0);
3576 		} else {
3577 			struct mbuf *mtl;       /* last mbuf of packet mt */
3578 
3579 			MLD_PRINTF(("%s: merging 0x%llx with ifscq tail "
3580 			    "0x%llx)\n", __func__,
3581 			    (uint64_t)VM_KERNEL_ADDRPERM(m0),
3582 			    (uint64_t)VM_KERNEL_ADDRPERM(mt)));
3583 
3584 			mtl = m_last(mt);
3585 			m0->m_flags &= ~M_PKTHDR;
3586 			mt->m_pkthdr.len += recslen;
3587 			mt->m_pkthdr.vt_nrecs +=
3588 			    m0->m_pkthdr.vt_nrecs;
3589 
3590 			mtl->m_next = m0;
3591 		}
3592 	}
3593 
3594 	return 0;
3595 }
3596 
3597 /*
3598  * Respond to a pending MLDv2 General Query.
3599  */
3600 static uint32_t
mld_v2_dispatch_general_query(struct mld_ifinfo * mli)3601 mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
3602 {
3603 	struct ifnet            *ifp;
3604 	struct in6_multi        *inm;
3605 	struct in6_multistep    step;
3606 	int                      retval;
3607 
3608 	MLI_LOCK_ASSERT_HELD(mli);
3609 
3610 	VERIFY(mli->mli_version == MLD_VERSION_2);
3611 
3612 	ifp = mli->mli_ifp;
3613 	MLI_UNLOCK(mli);
3614 
3615 	in6_multihead_lock_shared();
3616 	IN6_FIRST_MULTI(step, inm);
3617 	while (inm != NULL) {
3618 		IN6M_LOCK(inm);
3619 		if (inm->in6m_ifp != ifp) {
3620 			goto next;
3621 		}
3622 
3623 		switch (inm->in6m_state) {
3624 		case MLD_NOT_MEMBER:
3625 		case MLD_SILENT_MEMBER:
3626 			break;
3627 		case MLD_REPORTING_MEMBER:
3628 		case MLD_IDLE_MEMBER:
3629 		case MLD_LAZY_MEMBER:
3630 		case MLD_SLEEPING_MEMBER:
3631 		case MLD_AWAKENING_MEMBER:
3632 			inm->in6m_state = MLD_REPORTING_MEMBER;
3633 			MLI_LOCK(mli);
3634 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3635 			    inm, 0, 0, 0, 0);
3636 			MLI_UNLOCK(mli);
3637 			MLD_PRINTF(("%s: enqueue record = %d\n",
3638 			    __func__, retval));
3639 			break;
3640 		case MLD_G_QUERY_PENDING_MEMBER:
3641 		case MLD_SG_QUERY_PENDING_MEMBER:
3642 		case MLD_LEAVING_MEMBER:
3643 			break;
3644 		}
3645 next:
3646 		IN6M_UNLOCK(inm);
3647 		IN6_NEXT_MULTI(step, inm);
3648 	}
3649 	in6_multihead_lock_done();
3650 
3651 	MLI_LOCK(mli);
3652 	mld_dispatch_queue_locked(mli, &mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3653 	MLI_LOCK_ASSERT_HELD(mli);
3654 
3655 	/*
3656 	 * Slew transmission of bursts over 1 second intervals.
3657 	 */
3658 	if (mli->mli_gq.ifq_head != NULL) {
3659 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3660 			MLD_RESPONSE_BURST_INTERVAL);
3661 	}
3662 
3663 	return mli->mli_v2_timer;
3664 }
3665 
3666 /*
3667  * Transmit the next pending message in the output queue.
3668  *
3669  * Must not be called with in6m_lockm or mli_lock held.
3670  */
3671 static void
mld_dispatch_packet(struct mbuf * m)3672 mld_dispatch_packet(struct mbuf *m)
3673 {
3674 	struct ip6_moptions     *im6o;
3675 	struct ifnet            *ifp;
3676 	struct ifnet            *oifp = NULL;
3677 	struct mbuf             *m0;
3678 	struct mbuf             *md;
3679 	struct ip6_hdr          *ip6;
3680 	struct mld_hdr          *mld;
3681 	int                      error;
3682 	int                      off;
3683 	int                      type;
3684 
3685 	MLD_PRINTF(("%s: transmit 0x%llx\n", __func__,
3686 	    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3687 
3688 	/*
3689 	 * Check if the ifnet is still attached.
3690 	 */
3691 	ifp = mld_restore_context(m);
3692 	if (ifp == NULL || !ifnet_is_attached(ifp, 0)) {
3693 		os_log_error(OS_LOG_DEFAULT, "%s: dropped 0x%llx as interface went away\n",
3694 		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(m));
3695 		m_freem(m);
3696 		ip6stat.ip6s_noroute++;
3697 		return;
3698 	}
3699 
3700 	im6o = ip6_allocmoptions(Z_WAITOK);
3701 	if (im6o == NULL) {
3702 		m_freem(m);
3703 		return;
3704 	}
3705 
3706 	im6o->im6o_multicast_hlim  = 1;
3707 	im6o->im6o_multicast_loop = 0;
3708 	im6o->im6o_multicast_ifp = ifp;
3709 
3710 	if (m->m_flags & M_MLDV1) {
3711 		m0 = m;
3712 	} else {
3713 		m0 = mld_v2_encap_report(ifp, m);
3714 		if (m0 == NULL) {
3715 			os_log_error(OS_LOG_DEFAULT, "%s: dropped 0x%llx\n", __func__,
3716 			    (uint64_t)VM_KERNEL_ADDRPERM(m));
3717 			/*
3718 			 * mld_v2_encap_report() has already freed our mbuf.
3719 			 */
3720 			IM6O_REMREF(im6o);
3721 			ip6stat.ip6s_odropped++;
3722 			return;
3723 		}
3724 	}
3725 
3726 	mld_scrub_context(m0);
3727 	m->m_flags &= ~(M_PROTOFLAGS);
3728 	m0->m_pkthdr.rcvif = lo_ifp;
3729 
3730 	ip6 = mtod(m0, struct ip6_hdr *);
3731 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);
3732 	ip6_output_setdstifscope(m0, ifp->if_index, NULL);
3733 	/*
3734 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3735 	 * so we can bump the stats.
3736 	 */
3737 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3738 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3739 	type = mld->mld_type;
3740 
3741 	if (ifp->if_eflags & IFEF_TXSTART) {
3742 		/*
3743 		 * Use control service class if the outgoing
3744 		 * interface supports transmit-start model.
3745 		 */
3746 		(void) m_set_service_class(m0, MBUF_SC_CTL);
3747 	}
3748 
3749 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, im6o,
3750 	    &oifp, NULL);
3751 
3752 	IM6O_REMREF(im6o);
3753 
3754 	if (error) {
3755 		os_log_error(OS_LOG_DEFAULT, "%s: ip6_output(0x%llx) = %d\n", __func__,
3756 		    (uint64_t)VM_KERNEL_ADDRPERM(m0), error);
3757 		if (oifp != NULL) {
3758 			ifnet_release(oifp);
3759 		}
3760 		return;
3761 	}
3762 
3763 	icmp6stat.icp6s_outhist[type]++;
3764 	if (oifp != NULL) {
3765 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3766 		switch (type) {
3767 		case MLD_LISTENER_REPORT:
3768 		case MLDV2_LISTENER_REPORT:
3769 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3770 			break;
3771 		case MLD_LISTENER_DONE:
3772 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3773 			break;
3774 		}
3775 		ifnet_release(oifp);
3776 	}
3777 }
3778 
3779 /*
3780  * Encapsulate an MLDv2 report.
3781  *
3782  * KAME IPv6 requires that hop-by-hop options be passed separately,
3783  * and that the IPv6 header be prepended in a separate mbuf.
3784  *
3785  * Returns a pointer to the new mbuf chain head, or NULL if the
3786  * allocation failed.
3787  */
3788 static struct mbuf *
mld_v2_encap_report(struct ifnet * ifp,struct mbuf * m)3789 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3790 {
3791 	struct mbuf             *mh;
3792 	struct mldv2_report     *mld;
3793 	struct ip6_hdr          *ip6;
3794 	struct in6_ifaddr       *ia;
3795 	int                      mldreclen;
3796 
3797 	VERIFY(m->m_flags & M_PKTHDR);
3798 
3799 	/*
3800 	 * RFC3590: OK to send as :: or tentative during DAD.
3801 	 */
3802 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
3803 	if (ia == NULL) {
3804 		MLD_PRINTF(("%s: warning: ia is NULL\n", __func__));
3805 	}
3806 
3807 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3808 	if (mh == NULL) {
3809 		if (ia != NULL) {
3810 			IFA_REMREF(&ia->ia_ifa);
3811 		}
3812 		m_freem(m);
3813 		return NULL;
3814 	}
3815 	MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3816 
3817 	mldreclen = m_length(m);
3818 	MLD_PRINTF(("%s: mldreclen is %d\n", __func__, mldreclen));
3819 
3820 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3821 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3822 	    sizeof(struct mldv2_report) + mldreclen;
3823 
3824 	ip6 = mtod(mh, struct ip6_hdr *);
3825 	ip6->ip6_flow = 0;
3826 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3827 	ip6->ip6_vfc |= IPV6_VERSION;
3828 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3829 	if (ia != NULL) {
3830 		IFA_LOCK(&ia->ia_ifa);
3831 	}
3832 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3833 	ip6_output_setsrcifscope(mh, IFSCOPE_NONE, ia);
3834 
3835 	if (ia != NULL) {
3836 		IFA_UNLOCK(&ia->ia_ifa);
3837 		IFA_REMREF(&ia->ia_ifa);
3838 		ia = NULL;
3839 	}
3840 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3841 	ip6_output_setdstifscope(mh, ifp->if_index, NULL);
3842 	/* scope ID will be set in netisr */
3843 
3844 	mld = (struct mldv2_report *)(ip6 + 1);
3845 	mld->mld_type = MLDV2_LISTENER_REPORT;
3846 	mld->mld_code = 0;
3847 	mld->mld_cksum = 0;
3848 	mld->mld_v2_reserved = 0;
3849 	mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3850 	m->m_pkthdr.vt_nrecs = 0;
3851 	m->m_flags &= ~M_PKTHDR;
3852 
3853 	mh->m_next = m;
3854 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3855 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3856 	return mh;
3857 }
3858 
3859 #ifdef MLD_DEBUG
3860 static const char *
mld_rec_type_to_str(const int type)3861 mld_rec_type_to_str(const int type)
3862 {
3863 	switch (type) {
3864 	case MLD_CHANGE_TO_EXCLUDE_MODE:
3865 		return "TO_EX";
3866 	case MLD_CHANGE_TO_INCLUDE_MODE:
3867 		return "TO_IN";
3868 	case MLD_MODE_IS_EXCLUDE:
3869 		return "MODE_EX";
3870 	case MLD_MODE_IS_INCLUDE:
3871 		return "MODE_IN";
3872 	case MLD_ALLOW_NEW_SOURCES:
3873 		return "ALLOW_NEW";
3874 	case MLD_BLOCK_OLD_SOURCES:
3875 		return "BLOCK_OLD";
3876 	default:
3877 		break;
3878 	}
3879 	return "unknown";
3880 }
3881 #endif
3882 
3883 void
mld_init(void)3884 mld_init(void)
3885 {
3886 	os_log(OS_LOG_DEFAULT, "%s: initializing\n", __func__);
3887 
3888 	ip6_initpktopts(&mld_po);
3889 	mld_po.ip6po_hlim = 1;
3890 	mld_po.ip6po_hbh = &mld_ra.hbh;
3891 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3892 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3893 	LIST_INIT(&mli_head);
3894 }
3895