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 */
133static LCK_ATTR_DECLARE(mld_mtx_attr, 0, 0);
134static 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 */
166static LCK_MTX_DECLARE_ATTR(mld_mtx, &mld_mtx_grp, &mld_mtx_attr);
167
168SLIST_HEAD(mld_in6m_relhead, in6_multi);
169
170static void mli_initvar(struct mld_ifinfo *, struct ifnet *, int);
171static struct mld_ifinfo *mli_alloc(zalloc_flags_t);
172static void mli_free(struct mld_ifinfo *);
173static void mli_delete(const struct ifnet *, struct mld_in6m_relhead *);
174static void mld_dispatch_packet(struct mbuf *);
175static void mld_final_leave(struct in6_multi *, struct mld_ifinfo *,
176 struct mld_tparams *);
177static int mld_handle_state_change(struct in6_multi *, struct mld_ifinfo *,
178 struct mld_tparams *);
179static int mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
180 struct mld_tparams *, const int);
181#ifdef MLD_DEBUG
182static const char * mld_rec_type_to_str(const int);
183#endif
184static uint32_t mld_set_version(struct mld_ifinfo *, const int);
185static void mld_append_relq(struct mld_ifinfo *, struct in6_multi *);
186static void mld_flush_relq(struct mld_ifinfo *, struct mld_in6m_relhead *);
187static void mld_dispatch_queue_locked(struct mld_ifinfo *, struct ifqueue *, int);
188static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
189 /*const*/ struct mld_hdr *);
190static int mld_v1_input_report(struct ifnet *, struct mbuf *,
191 const struct ip6_hdr *, /*const*/ struct mld_hdr *);
192static void mld_v1_process_group_timer(struct in6_multi *, const int);
193static void mld_v1_process_querier_timers(struct mld_ifinfo *);
194static int mld_v1_transmit_report(struct in6_multi *, const uint8_t);
195static uint32_t mld_v1_update_group(struct in6_multi *, const int);
196static void mld_v2_cancel_link_timers(struct mld_ifinfo *);
197static uint32_t mld_v2_dispatch_general_query(struct mld_ifinfo *);
198static struct mbuf *
199mld_v2_encap_report(struct ifnet *, struct mbuf *);
200static int mld_v2_enqueue_filter_change(struct ifqueue *,
201 struct in6_multi *);
202static int mld_v2_enqueue_group_record(struct ifqueue *,
203 struct in6_multi *, const int, const int, const int,
204 const int);
205static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
206 struct mbuf *, const int, const int);
207static int mld_v2_merge_state_changes(struct in6_multi *,
208 struct ifqueue *);
209static void mld_v2_process_group_timers(struct mld_ifinfo *,
210 struct ifqueue *, struct ifqueue *,
211 struct in6_multi *, const int);
212static int mld_v2_process_group_query(struct in6_multi *,
213 int, struct mbuf *, const int);
214static int sysctl_mld_gsr SYSCTL_HANDLER_ARGS;
215static int sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS;
216static int sysctl_mld_v2enable SYSCTL_HANDLER_ARGS;
217
218static const uint32_t mld_timeout_delay = 1000; /* in milliseconds */
219static const uint32_t mld_timeout_leeway = 500; /* in millseconds */
220static bool mld_timeout_run; /* MLD timer is scheduled to run */
221static bool mld_fast_timeout_run; /* MLD fast timer is scheduled to run */
222static void mld_timeout(thread_call_param_t, thread_call_param_t);
223static void mld_sched_timeout(void);
224static void mld_sched_fast_timeout(void);
225
226/*
227 * Normative references: RFC 2710, RFC 3590, RFC 3810.
228 */
229static struct timeval mld_gsrdelay = {.tv_sec = 10, .tv_usec = 0};
230static LIST_HEAD(, mld_ifinfo) mli_head;
231
232static int querier_present_timers_running6;
233static int interface_timers_running6;
234static int state_change_timers_running6;
235static int current_state_timers_running6;
236
237static 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
263static KALLOC_TYPE_DEFINE(mli_zone, struct mld_ifinfo, NET_KT_DEFAULT);
264
265SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */
266
267SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_LOCKED, 0,
268 "IPv6 Multicast Listener Discovery");
269SYSCTL_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
274SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_LOCKED,
275 sysctl_mld_ifinfo, "Per-interface MLDv2 state");
276
277static int mld_v1enable = 1;
278SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW | CTLFLAG_LOCKED,
279 &mld_v1enable, 0, "Enable fallback to MLDv1");
280
281static int mld_v2enable = 1;
282SYSCTL_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
287static int mld_use_allow = 1;
288SYSCTL_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
292int mld_debug = 0;
293SYSCTL_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 */
299struct 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 */
308static 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};
318static 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
323static __inline void
324mld_save_context(struct mbuf *m, struct ifnet *ifp)
325{
326 m->m_pkthdr.rcvif = ifp;
327}
328
329static __inline void
330mld_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 */
339static __inline struct ifnet *
340mld_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 */
348static int
349sysctl_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, arg1: &i, arg2: 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
371out_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 */
380static int
381sysctl_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(s: &mli_u, n: 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
442out_locked:
443 MLD_UNLOCK();
444 return error;
445}
446
447static int
448sysctl_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, arg1: &i, arg2: 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
488out_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 */
507static void
508mld_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 */
561static __inline__ int
562mld_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 */
584struct mld_ifinfo *
585mld_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 */
624void
625mld_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 */
655void
656mld_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 */
679static void
680mli_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
710mld6_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
725static void
726mli_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
756static struct mld_ifinfo *
757mli_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(lck: &mli->mli_lock, grp: &mld_mtx_grp, attr: &mld_mtx_attr);
762 mli->mli_debug |= IFD_ALLOC;
763 }
764 return mli;
765}
766
767static void
768mli_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(lck: &mli->mli_lock, grp: &mld_mtx_grp);
788 zfree(mli_zone, mli);
789}
790
791void
792mli_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
809void
810mli_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 */
852static int
853mld_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 }
967done:
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 */
988static uint32_t
989mld_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 */
1037static int
1038mld_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(lasttime: &inm->in6m_lastgsrtv,
1194 mininterval: &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 }
1223done:
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 */
1239static int
1240mld_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(a: inm->in6m_timer, b: 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(a: inm->in6m_timer, b: 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 */
1353static int
1354mld_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(ifa: &ia->ia_ifa);
1415 return 0;
1416 }
1417 IFA_UNLOCK(&ia->ia_ifa);
1418 ifa_remref(ifa: &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
1492out:
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 */
1510int
1511mld_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 */
1581void
1582mld_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
1610void
1611mld_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 */
1621static void
1622mld_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
1731skip_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(s: &qrq, c: 0, n: sizeof(struct ifqueue));
1743 qrq.ifq_maxlen = MLD_MAX_G_GS_PACKETS;
1744
1745 memset(s: &scq, c: 0, n: 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);
1787next:
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, ifq: &mli->mli_v1q, limit: 0);
1805 } else if (mli->mli_version == MLD_VERSION_2) {
1806 MLI_UNLOCK(mli);
1807 mld_dispatch_queue_locked(NULL, ifq: &qrq, limit: 0);
1808 mld_dispatch_queue_locked(NULL, ifq: &scq, limit: 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
1846out_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
1860static void
1861mld_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(func: mld_timeout,
1870 NULL,
1871 pri: THREAD_CALL_PRIORITY_KERNEL,
1872 options: 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(interval: mld_timeout_delay, NSEC_PER_MSEC,
1880 result: &deadline);
1881 clock_interval_to_absolutetime_interval(interval: mld_timeout_leeway,
1882 NSEC_PER_MSEC, result: &leeway);
1883 thread_call_enter_delayed_with_leeway(call: mld_timeout_tcall, NULL,
1884 deadline, leeway,
1885 THREAD_CALL_DELAY_LEEWAY);
1886 }
1887}
1888
1889static void
1890mld_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(func: mld_timeout,
1898 param0: mld_sched_fast_timeout,
1899 pri: THREAD_CALL_PRIORITY_KERNEL,
1900 options: 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(call: 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 */
1914static void
1915mld_append_relq(struct mld_ifinfo *mli, struct in6_multi *inm)
1916{
1917 MLI_LOCK_ASSERT_HELD(mli);
1918 if (inm->in6m_in_nrele) {
1919 os_log_debug(OS_LOG_DEFAULT, "%s: inm %llx already on relq ifp %s\n",
1920 __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm),
1921 mli->mli_ifp != NULL ? if_name(mli->mli_ifp) : "<null>");
1922 return;
1923 }
1924 os_log_debug(OS_LOG_DEFAULT, "%s: adding inm %llx on relq ifp %s\n",
1925 __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm),
1926 mli->mli_ifp != NULL ? if_name(mli->mli_ifp) : "<null>");
1927 inm->in6m_in_nrele = true;
1928 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, in6m_nrele);
1929}
1930
1931/*
1932 * Free the in6_multi reference(s) for this MLD lifecycle.
1933 *
1934 * Caller must be holding mli_lock.
1935 */
1936static void
1937mld_flush_relq(struct mld_ifinfo *mli, struct mld_in6m_relhead *in6m_dthead)
1938{
1939 struct in6_multi *inm;
1940 SLIST_HEAD(, in6_multi) temp_relinmhead;
1941
1942 /*
1943 * Before dropping the mli_lock, copy all the items in the
1944 * release list to a temporary list to prevent other threads
1945 * from changing mli_relinmhead while we are traversing it.
1946 */
1947 MLI_LOCK_ASSERT_HELD(mli);
1948 SLIST_INIT(&temp_relinmhead);
1949 while ((inm = SLIST_FIRST(&mli->mli_relinmhead)) != NULL) {
1950 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
1951 SLIST_INSERT_HEAD(&temp_relinmhead, inm, in6m_nrele);
1952 }
1953 MLI_UNLOCK(mli);
1954 in6_multihead_lock_exclusive();
1955 while ((inm = SLIST_FIRST(&temp_relinmhead)) != NULL) {
1956 int lastref;
1957
1958 SLIST_REMOVE_HEAD(&temp_relinmhead, in6m_nrele);
1959 IN6M_LOCK(inm);
1960 os_log_debug(OS_LOG_DEFAULT, "%s: flushing inm %llx on relq ifp %s\n",
1961 __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm),
1962 inm->in6m_ifp != NULL ? if_name(inm->in6m_ifp) : "<null>");
1963 VERIFY(inm->in6m_in_nrele == true);
1964 inm->in6m_in_nrele = false;
1965 VERIFY(inm->in6m_nrelecnt != 0);
1966 inm->in6m_nrelecnt--;
1967 lastref = in6_multi_detach(inm);
1968 VERIFY(!lastref || (!(inm->in6m_debug & IFD_ATTACHED) &&
1969 inm->in6m_reqcnt == 0));
1970 IN6M_UNLOCK(inm);
1971 /* from mli_relinmhead */
1972 IN6M_REMREF(inm);
1973 /* from in6_multihead_list */
1974 if (lastref) {
1975 /*
1976 * Defer releasing our final reference, as we
1977 * are holding the MLD lock at this point, and
1978 * we could end up with locking issues later on
1979 * (while issuing SIOCDELMULTI) when this is the
1980 * final reference count. Let the caller do it
1981 * when it is safe.
1982 */
1983 MLD_ADD_DETACHED_IN6M(in6m_dthead, inm);
1984 }
1985 }
1986 in6_multihead_lock_done();
1987 MLI_LOCK(mli);
1988}
1989
1990/*
1991 * Update host report group timer.
1992 * Will update the global pending timer flags.
1993 */
1994static void
1995mld_v1_process_group_timer(struct in6_multi *inm, const int mld_version)
1996{
1997#pragma unused(mld_version)
1998 int report_timer_expired;
1999
2000 MLD_LOCK_ASSERT_HELD();
2001 IN6M_LOCK_ASSERT_HELD(inm);
2002 MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
2003
2004 if (inm->in6m_timer == 0) {
2005 report_timer_expired = 0;
2006 } else if (--inm->in6m_timer == 0) {
2007 report_timer_expired = 1;
2008 } else {
2009 current_state_timers_running6 = 1;
2010 /* caller will schedule timer */
2011 return;
2012 }
2013
2014 switch (inm->in6m_state) {
2015 case MLD_NOT_MEMBER:
2016 case MLD_SILENT_MEMBER:
2017 case MLD_IDLE_MEMBER:
2018 case MLD_LAZY_MEMBER:
2019 case MLD_SLEEPING_MEMBER:
2020 case MLD_AWAKENING_MEMBER:
2021 break;
2022 case MLD_REPORTING_MEMBER:
2023 if (report_timer_expired) {
2024 inm->in6m_state = MLD_IDLE_MEMBER;
2025 (void) mld_v1_transmit_report(inm,
2026 MLD_LISTENER_REPORT);
2027 IN6M_LOCK_ASSERT_HELD(inm);
2028 MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
2029 }
2030 break;
2031 case MLD_G_QUERY_PENDING_MEMBER:
2032 case MLD_SG_QUERY_PENDING_MEMBER:
2033 case MLD_LEAVING_MEMBER:
2034 break;
2035 }
2036}
2037
2038/*
2039 * Update a group's timers for MLDv2.
2040 * Will update the global pending timer flags.
2041 * Note: Unlocked read from mli.
2042 */
2043static void
2044mld_v2_process_group_timers(struct mld_ifinfo *mli,
2045 struct ifqueue *qrq, struct ifqueue *scq,
2046 struct in6_multi *inm, const int uri_sec)
2047{
2048 int query_response_timer_expired;
2049 int state_change_retransmit_timer_expired;
2050
2051 MLD_LOCK_ASSERT_HELD();
2052 IN6M_LOCK_ASSERT_HELD(inm);
2053 MLI_LOCK_ASSERT_HELD(mli);
2054 VERIFY(mli == inm->in6m_mli);
2055
2056 query_response_timer_expired = 0;
2057 state_change_retransmit_timer_expired = 0;
2058
2059 /*
2060 * During a transition from compatibility mode back to MLDv2,
2061 * a group record in REPORTING state may still have its group
2062 * timer active. This is a no-op in this function; it is easier
2063 * to deal with it here than to complicate the timeout path.
2064 */
2065 if (inm->in6m_timer == 0) {
2066 query_response_timer_expired = 0;
2067 } else if (--inm->in6m_timer == 0) {
2068 query_response_timer_expired = 1;
2069 } else {
2070 current_state_timers_running6 = 1;
2071 /* caller will schedule timer */
2072 }
2073
2074 if (inm->in6m_sctimer == 0) {
2075 state_change_retransmit_timer_expired = 0;
2076 } else if (--inm->in6m_sctimer == 0) {
2077 state_change_retransmit_timer_expired = 1;
2078 } else {
2079 state_change_timers_running6 = 1;
2080 /* caller will schedule timer */
2081 }
2082
2083 /* We are in timer callback, so be quick about it. */
2084 if (!state_change_retransmit_timer_expired &&
2085 !query_response_timer_expired) {
2086 return;
2087 }
2088
2089 switch (inm->in6m_state) {
2090 case MLD_NOT_MEMBER:
2091 case MLD_SILENT_MEMBER:
2092 case MLD_SLEEPING_MEMBER:
2093 case MLD_LAZY_MEMBER:
2094 case MLD_AWAKENING_MEMBER:
2095 case MLD_IDLE_MEMBER:
2096 break;
2097 case MLD_G_QUERY_PENDING_MEMBER:
2098 case MLD_SG_QUERY_PENDING_MEMBER:
2099 /*
2100 * Respond to a previously pending Group-Specific
2101 * or Group-and-Source-Specific query by enqueueing
2102 * the appropriate Current-State report for
2103 * immediate transmission.
2104 */
2105 if (query_response_timer_expired) {
2106 int retval;
2107
2108 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
2109 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
2110 0);
2111 MLD_PRINTF(("%s: enqueue record = %d\n",
2112 __func__, retval));
2113 inm->in6m_state = MLD_REPORTING_MEMBER;
2114 in6m_clear_recorded(inm);
2115 }
2116 OS_FALLTHROUGH;
2117 case MLD_REPORTING_MEMBER:
2118 case MLD_LEAVING_MEMBER:
2119 if (state_change_retransmit_timer_expired) {
2120 /*
2121 * State-change retransmission timer fired.
2122 * If there are any further pending retransmissions,
2123 * set the global pending state-change flag, and
2124 * reset the timer.
2125 */
2126 if (--inm->in6m_scrv > 0) {
2127 inm->in6m_sctimer = (uint16_t)uri_sec;
2128 state_change_timers_running6 = 1;
2129 /* caller will schedule timer */
2130 }
2131 /*
2132 * Retransmit the previously computed state-change
2133 * report. If there are no further pending
2134 * retransmissions, the mbuf queue will be consumed.
2135 * Update T0 state to T1 as we have now sent
2136 * a state-change.
2137 */
2138 (void) mld_v2_merge_state_changes(inm, scq);
2139
2140 in6m_commit(inm);
2141 MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2142 ip6_sprintf(&inm->in6m_addr),
2143 if_name(inm->in6m_ifp)));
2144
2145 /*
2146 * If we are leaving the group for good, make sure
2147 * we release MLD's reference to it.
2148 * This release must be deferred using a SLIST,
2149 * as we are called from a loop which traverses
2150 * the in_ifmultiaddr TAILQ.
2151 */
2152 if (inm->in6m_state == MLD_LEAVING_MEMBER &&
2153 inm->in6m_scrv == 0) {
2154 inm->in6m_state = MLD_NOT_MEMBER;
2155 /*
2156 * A reference has already been held in
2157 * mld_final_leave() for this inm, so
2158 * no need to hold another one. We also
2159 * bumped up its request count then, so
2160 * that it stays in in6_multihead. Both
2161 * of them will be released when it is
2162 * dequeued later on.
2163 */
2164 VERIFY(inm->in6m_nrelecnt != 0);
2165 mld_append_relq(mli, inm);
2166 }
2167 }
2168 break;
2169 }
2170}
2171
2172/*
2173 * Switch to a different version on the given interface,
2174 * as per Section 9.12.
2175 */
2176static uint32_t
2177mld_set_version(struct mld_ifinfo *mli, const int mld_version)
2178{
2179 int old_version_timer;
2180
2181 MLI_LOCK_ASSERT_HELD(mli);
2182
2183 os_log(OS_LOG_DEFAULT, "%s: switching to v%d on ifp %s\n", __func__,
2184 mld_version, if_name(mli->mli_ifp));
2185
2186 if (mld_version == MLD_VERSION_1) {
2187 /*
2188 * Compute the "Older Version Querier Present" timer as per
2189 * Section 9.12, in seconds.
2190 */
2191 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
2192 mli->mli_v1_timer = old_version_timer;
2193 }
2194
2195 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
2196 mli->mli_version = MLD_VERSION_1;
2197 mld_v2_cancel_link_timers(mli);
2198 }
2199
2200 MLI_LOCK_ASSERT_HELD(mli);
2201
2202 return mli->mli_v1_timer;
2203}
2204
2205/*
2206 * Cancel pending MLDv2 timers for the given link and all groups
2207 * joined on it; state-change, general-query, and group-query timers.
2208 *
2209 * Only ever called on a transition from v2 to Compatibility mode. Kill
2210 * the timers stone dead (this may be expensive for large N groups), they
2211 * will be restarted if Compatibility Mode deems that they must be due to
2212 * query processing.
2213 */
2214static void
2215mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
2216{
2217 struct ifnet *ifp;
2218 struct in6_multi *inm;
2219 struct in6_multistep step;
2220
2221 MLI_LOCK_ASSERT_HELD(mli);
2222
2223 MLD_PRINTF(("%s: cancel v2 timers on ifp 0x%llx(%s)\n", __func__,
2224 (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp), if_name(mli->mli_ifp)));
2225
2226 /*
2227 * Stop the v2 General Query Response on this link stone dead.
2228 * If timer is woken up due to interface_timers_running6,
2229 * the flag will be cleared if there are no pending link timers.
2230 */
2231 mli->mli_v2_timer = 0;
2232
2233 /*
2234 * Now clear the current-state and state-change report timers
2235 * for all memberships scoped to this link.
2236 */
2237 ifp = mli->mli_ifp;
2238 MLI_UNLOCK(mli);
2239
2240 in6_multihead_lock_shared();
2241 IN6_FIRST_MULTI(step, inm);
2242 while (inm != NULL) {
2243 IN6M_LOCK(inm);
2244 if (inm->in6m_ifp != ifp) {
2245 goto next;
2246 }
2247
2248 switch (inm->in6m_state) {
2249 case MLD_NOT_MEMBER:
2250 case MLD_SILENT_MEMBER:
2251 case MLD_IDLE_MEMBER:
2252 case MLD_LAZY_MEMBER:
2253 case MLD_SLEEPING_MEMBER:
2254 case MLD_AWAKENING_MEMBER:
2255 /*
2256 * These states are either not relevant in v2 mode,
2257 * or are unreported. Do nothing.
2258 */
2259 break;
2260 case MLD_LEAVING_MEMBER:
2261 /*
2262 * If we are leaving the group and switching
2263 * version, we need to release the final
2264 * reference held for issuing the INCLUDE {}.
2265 * During mld_final_leave(), we bumped up both the
2266 * request and reference counts. Since we cannot
2267 * call in6_multi_detach() here, defer this task to
2268 * the timer routine.
2269 */
2270 VERIFY(inm->in6m_nrelecnt != 0);
2271 MLI_LOCK(mli);
2272 mld_append_relq(mli, inm);
2273 MLI_UNLOCK(mli);
2274 OS_FALLTHROUGH;
2275 case MLD_G_QUERY_PENDING_MEMBER:
2276 case MLD_SG_QUERY_PENDING_MEMBER:
2277 in6m_clear_recorded(inm);
2278 OS_FALLTHROUGH;
2279 case MLD_REPORTING_MEMBER:
2280 inm->in6m_state = MLD_REPORTING_MEMBER;
2281 break;
2282 }
2283 /*
2284 * Always clear state-change and group report timers.
2285 * Free any pending MLDv2 state-change records.
2286 */
2287 inm->in6m_sctimer = 0;
2288 inm->in6m_timer = 0;
2289 IF_DRAIN(&inm->in6m_scq);
2290next:
2291 IN6M_UNLOCK(inm);
2292 IN6_NEXT_MULTI(step, inm);
2293 }
2294 in6_multihead_lock_done();
2295
2296 MLI_LOCK(mli);
2297}
2298
2299/*
2300 * Update the Older Version Querier Present timers for a link.
2301 * See Section 9.12 of RFC 3810.
2302 */
2303static void
2304mld_v1_process_querier_timers(struct mld_ifinfo *mli)
2305{
2306 MLI_LOCK_ASSERT_HELD(mli);
2307
2308 if (mld_v2enable && mli->mli_version != MLD_VERSION_2 &&
2309 --mli->mli_v1_timer == 0) {
2310 /*
2311 * MLDv1 Querier Present timer expired; revert to MLDv2.
2312 */
2313 os_log(OS_LOG_DEFAULT, "%s: transition from v%d -> v%d on %s\n",
2314 __func__, mli->mli_version, MLD_VERSION_2,
2315 if_name(mli->mli_ifp));
2316 mli->mli_version = MLD_VERSION_2;
2317 }
2318}
2319
2320/*
2321 * Transmit an MLDv1 report immediately.
2322 */
2323static int
2324mld_v1_transmit_report(struct in6_multi *in6m, const uint8_t type)
2325{
2326 struct ifnet *ifp;
2327 struct in6_ifaddr *ia;
2328 struct ip6_hdr *ip6;
2329 struct mbuf *mh, *md;
2330 struct mld_hdr *mld;
2331 int error = 0;
2332
2333 IN6M_LOCK_ASSERT_HELD(in6m);
2334 MLI_LOCK_ASSERT_HELD(in6m->in6m_mli);
2335
2336 ifp = in6m->in6m_ifp;
2337 /* ia may be NULL if link-local address is tentative. */
2338 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
2339
2340 MGETHDR(mh, M_DONTWAIT, MT_HEADER);
2341 if (mh == NULL) {
2342 if (ia != NULL) {
2343 ifa_remref(ifa: &ia->ia_ifa);
2344 }
2345 return ENOMEM;
2346 }
2347 MGET(md, M_DONTWAIT, MT_DATA);
2348 if (md == NULL) {
2349 m_free(mh);
2350 if (ia != NULL) {
2351 ifa_remref(ifa: &ia->ia_ifa);
2352 }
2353 return ENOMEM;
2354 }
2355 mh->m_next = md;
2356
2357 /*
2358 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
2359 * that ether_output() does not need to allocate another mbuf
2360 * for the header in the most common case.
2361 */
2362 MH_ALIGN(mh, sizeof(struct ip6_hdr));
2363 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
2364 mh->m_len = sizeof(struct ip6_hdr);
2365
2366 ip6 = mtod(mh, struct ip6_hdr *);
2367 ip6->ip6_flow = 0;
2368 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
2369 ip6->ip6_vfc |= IPV6_VERSION;
2370 ip6->ip6_nxt = IPPROTO_ICMPV6;
2371 if (ia != NULL) {
2372 IFA_LOCK(&ia->ia_ifa);
2373 }
2374 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
2375 ip6_output_setsrcifscope(mh, IFSCOPE_NONE, ia);
2376 if (ia != NULL) {
2377 IFA_UNLOCK(&ia->ia_ifa);
2378 ifa_remref(ifa: &ia->ia_ifa);
2379 ia = NULL;
2380 }
2381 ip6->ip6_dst = in6m->in6m_addr;
2382 ip6_output_setdstifscope(mh, in6m->ifscope, NULL);
2383
2384 md->m_len = sizeof(struct mld_hdr);
2385 mld = mtod(md, struct mld_hdr *);
2386 mld->mld_type = type;
2387 mld->mld_code = 0;
2388 mld->mld_cksum = 0;
2389 mld->mld_maxdelay = 0;
2390 mld->mld_reserved = 0;
2391 mld->mld_addr = in6m->in6m_addr;
2392 in6_clearscope(&mld->mld_addr);
2393 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
2394 sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
2395
2396 mld_save_context(m: mh, ifp);
2397 mh->m_flags |= M_MLDV1;
2398
2399 /*
2400 * Due to the fact that at this point we are possibly holding
2401 * in6_multihead_lock in shared or exclusive mode, we can't call
2402 * mld_dispatch_packet() here since that will eventually call
2403 * ip6_output(), which will try to lock in6_multihead_lock and cause
2404 * a deadlock.
2405 * Instead we defer the work to the mld_timeout() thread, thus
2406 * avoiding unlocking in_multihead_lock here.
2407 */
2408 if (IF_QFULL(&in6m->in6m_mli->mli_v1q)) {
2409 os_log_error(OS_LOG_DEFAULT, "%s: v1 outbound queue full\n", __func__);
2410 error = ENOMEM;
2411 m_freem(mh);
2412 } else {
2413 IF_ENQUEUE(&in6m->in6m_mli->mli_v1q, mh);
2414 VERIFY(error == 0);
2415 }
2416
2417 return error;
2418}
2419
2420/*
2421 * Process a state change from the upper layer for the given IPv6 group.
2422 *
2423 * Each socket holds a reference on the in6_multi in its own ip_moptions.
2424 * The socket layer will have made the necessary updates to.the group
2425 * state, it is now up to MLD to issue a state change report if there
2426 * has been any change between T0 (when the last state-change was issued)
2427 * and T1 (now).
2428 *
2429 * We use the MLDv2 state machine at group level. The MLd module
2430 * however makes the decision as to which MLD protocol version to speak.
2431 * A state change *from* INCLUDE {} always means an initial join.
2432 * A state change *to* INCLUDE {} always means a final leave.
2433 *
2434 * If delay is non-zero, and the state change is an initial multicast
2435 * join, the state change report will be delayed by 'delay' ticks
2436 * in units of seconds if MLDv1 is active on the link; otherwise
2437 * the initial MLDv2 state change report will be delayed by whichever
2438 * is sooner, a pending state-change timer or delay itself.
2439 */
2440int
2441mld_change_state(struct in6_multi *inm, struct mld_tparams *mtp,
2442 const int delay)
2443{
2444 struct mld_ifinfo *mli;
2445 struct ifnet *ifp;
2446 int error = 0;
2447
2448 VERIFY(mtp != NULL);
2449 bzero(s: mtp, n: sizeof(*mtp));
2450
2451 IN6M_LOCK_ASSERT_HELD(inm);
2452 VERIFY(inm->in6m_mli != NULL);
2453 MLI_LOCK_ASSERT_NOTHELD(inm->in6m_mli);
2454
2455 /*
2456 * Try to detect if the upper layer just asked us to change state
2457 * for an interface which has now gone away.
2458 */
2459 VERIFY(inm->in6m_ifma != NULL);
2460 ifp = inm->in6m_ifma->ifma_ifp;
2461 /*
2462 * Sanity check that netinet6's notion of ifp is the same as net's.
2463 */
2464 VERIFY(inm->in6m_ifp == ifp);
2465
2466 mli = MLD_IFINFO(ifp);
2467 VERIFY(mli != NULL);
2468
2469 /*
2470 * If we detect a state transition to or from MCAST_UNDEFINED
2471 * for this group, then we are starting or finishing an MLD
2472 * life cycle for this group.
2473 */
2474 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
2475 MLD_PRINTF(("%s: inm transition %d -> %d\n", __func__,
2476 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode));
2477 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
2478 MLD_PRINTF(("%s: initial join\n", __func__));
2479 error = mld_initial_join(inm, mli, mtp, delay);
2480 goto out;
2481 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
2482 MLD_PRINTF(("%s: final leave\n", __func__));
2483 mld_final_leave(inm, mli, mtp);
2484 goto out;
2485 }
2486 } else {
2487 MLD_PRINTF(("%s: filter set change\n", __func__));
2488 }
2489
2490 error = mld_handle_state_change(inm, mli, mtp);
2491out:
2492 return error;
2493}
2494
2495/*
2496 * Perform the initial join for an MLD group.
2497 *
2498 * When joining a group:
2499 * If the group should have its MLD traffic suppressed, do nothing.
2500 * MLDv1 starts sending MLDv1 host membership reports.
2501 * MLDv2 will schedule an MLDv2 state-change report containing the
2502 * initial state of the membership.
2503 *
2504 * If the delay argument is non-zero, then we must delay sending the
2505 * initial state change for delay ticks (in units of seconds).
2506 */
2507static int
2508mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
2509 struct mld_tparams *mtp, const int delay)
2510{
2511 struct ifnet *ifp;
2512 struct ifqueue *ifq;
2513 int error, retval, syncstates;
2514 int odelay;
2515
2516 IN6M_LOCK_ASSERT_HELD(inm);
2517 MLI_LOCK_ASSERT_NOTHELD(mli);
2518 VERIFY(mtp != NULL);
2519
2520 MLD_PRINTF(("%s: initial join %s on ifp 0x%llx(%s)\n",
2521 __func__, ip6_sprintf(&inm->in6m_addr),
2522 (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2523 if_name(inm->in6m_ifp)));
2524
2525 error = 0;
2526 syncstates = 1;
2527
2528 ifp = inm->in6m_ifp;
2529
2530 MLI_LOCK(mli);
2531 VERIFY(mli->mli_ifp == ifp);
2532
2533 /*
2534 * Avoid MLD if group is :
2535 * 1. Joined on loopback, OR
2536 * 2. On a link that is marked MLIF_SILENT
2537 * 3. rdar://problem/19227650 Is link local scoped and
2538 * on cellular interface
2539 * 4. Is a type that should not be reported (node local
2540 * or all node link local multicast.
2541 * All other groups enter the appropriate state machine
2542 * for the version in use on this link.
2543 */
2544 if ((ifp->if_flags & IFF_LOOPBACK) ||
2545 (mli->mli_flags & MLIF_SILENT) ||
2546 (IFNET_IS_CELLULAR(ifp) &&
2547 (IN6_IS_ADDR_MC_LINKLOCAL(&inm->in6m_addr) || IN6_IS_ADDR_MC_UNICAST_BASED_LINKLOCAL(&inm->in6m_addr))) ||
2548 !mld_is_addr_reported(addr: &inm->in6m_addr)) {
2549 MLD_PRINTF(("%s: not kicking state machine for silent group\n",
2550 __func__));
2551 inm->in6m_state = MLD_SILENT_MEMBER;
2552 inm->in6m_timer = 0;
2553 } else {
2554 /*
2555 * Deal with overlapping in6_multi lifecycle.
2556 * If this group was LEAVING, then make sure
2557 * we drop the reference we picked up to keep the
2558 * group around for the final INCLUDE {} enqueue.
2559 * Since we cannot call in6_multi_detach() here,
2560 * defer this task to the timer routine.
2561 */
2562 if (mli->mli_version == MLD_VERSION_2 &&
2563 inm->in6m_state == MLD_LEAVING_MEMBER) {
2564 VERIFY(inm->in6m_nrelecnt != 0);
2565 mld_append_relq(mli, inm);
2566 }
2567
2568 inm->in6m_state = MLD_REPORTING_MEMBER;
2569
2570 switch (mli->mli_version) {
2571 case MLD_VERSION_1:
2572 /*
2573 * If a delay was provided, only use it if
2574 * it is greater than the delay normally
2575 * used for an MLDv1 state change report,
2576 * and delay sending the initial MLDv1 report
2577 * by not transitioning to the IDLE state.
2578 */
2579 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI);
2580 if (delay) {
2581 inm->in6m_timer = max(a: delay, b: odelay);
2582 mtp->cst = 1;
2583 } else {
2584 inm->in6m_state = MLD_IDLE_MEMBER;
2585 error = mld_v1_transmit_report(in6m: inm,
2586 MLD_LISTENER_REPORT);
2587
2588 IN6M_LOCK_ASSERT_HELD(inm);
2589 MLI_LOCK_ASSERT_HELD(mli);
2590
2591 if (error == 0) {
2592 inm->in6m_timer = odelay;
2593 mtp->cst = 1;
2594 }
2595 }
2596 break;
2597
2598 case MLD_VERSION_2:
2599 /*
2600 * Defer update of T0 to T1, until the first copy
2601 * of the state change has been transmitted.
2602 */
2603 syncstates = 0;
2604
2605 /*
2606 * Immediately enqueue a State-Change Report for
2607 * this interface, freeing any previous reports.
2608 * Don't kick the timers if there is nothing to do,
2609 * or if an error occurred.
2610 */
2611 ifq = &inm->in6m_scq;
2612 IF_DRAIN(ifq);
2613 retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2614 0, 0, (mli->mli_flags & MLIF_USEALLOW));
2615 mtp->cst = (ifq->ifq_len > 0);
2616 MLD_PRINTF(("%s: enqueue record = %d\n",
2617 __func__, retval));
2618 if (retval <= 0) {
2619 error = retval * -1;
2620 break;
2621 }
2622
2623 /*
2624 * Schedule transmission of pending state-change
2625 * report up to RV times for this link. The timer
2626 * will fire at the next mld_timeout (1 second)),
2627 * giving us an opportunity to merge the reports.
2628 *
2629 * If a delay was provided to this function, only
2630 * use this delay if sooner than the existing one.
2631 */
2632 VERIFY(mli->mli_rv > 1);
2633 inm->in6m_scrv = (uint16_t)mli->mli_rv;
2634 if (delay) {
2635 if (inm->in6m_sctimer > 1) {
2636 inm->in6m_sctimer =
2637 MIN(inm->in6m_sctimer, (uint16_t)delay);
2638 } else {
2639 inm->in6m_sctimer = (uint16_t)delay;
2640 }
2641 } else {
2642 inm->in6m_sctimer = 1;
2643 }
2644 mtp->sct = 1;
2645 error = 0;
2646 break;
2647 }
2648 }
2649 MLI_UNLOCK(mli);
2650
2651 /*
2652 * Only update the T0 state if state change is atomic,
2653 * i.e. we don't need to wait for a timer to fire before we
2654 * can consider the state change to have been communicated.
2655 */
2656 if (syncstates) {
2657 in6m_commit(inm);
2658 MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2659 ip6_sprintf(&inm->in6m_addr),
2660 if_name(inm->in6m_ifp)));
2661 }
2662
2663 return error;
2664}
2665
2666/*
2667 * Issue an intermediate state change during the life-cycle.
2668 */
2669static int
2670mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli,
2671 struct mld_tparams *mtp)
2672{
2673 struct ifnet *ifp;
2674 int retval = 0;
2675
2676 IN6M_LOCK_ASSERT_HELD(inm);
2677 MLI_LOCK_ASSERT_NOTHELD(mli);
2678 VERIFY(mtp != NULL);
2679
2680 MLD_PRINTF(("%s: state change for %s on ifp 0x%llx(%s)\n",
2681 __func__, ip6_sprintf(&inm->in6m_addr),
2682 (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2683 if_name(inm->in6m_ifp)));
2684
2685 ifp = inm->in6m_ifp;
2686
2687 MLI_LOCK(mli);
2688 VERIFY(mli->mli_ifp == ifp);
2689
2690 if ((ifp->if_flags & IFF_LOOPBACK) ||
2691 (mli->mli_flags & MLIF_SILENT) ||
2692 !mld_is_addr_reported(addr: &inm->in6m_addr) ||
2693 (mli->mli_version != MLD_VERSION_2)) {
2694 MLI_UNLOCK(mli);
2695 if (!mld_is_addr_reported(addr: &inm->in6m_addr)) {
2696 MLD_PRINTF(("%s: not kicking state machine for silent "
2697 "group\n", __func__));
2698 }
2699 MLD_PRINTF(("%s: nothing to do\n", __func__));
2700 in6m_commit(inm);
2701 MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2702 ip6_sprintf(&inm->in6m_addr),
2703 if_name(inm->in6m_ifp)));
2704 goto done;
2705 }
2706
2707 IF_DRAIN(&inm->in6m_scq);
2708
2709 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2710 (mli->mli_flags & MLIF_USEALLOW));
2711 mtp->cst = (inm->in6m_scq.ifq_len > 0);
2712 MLD_PRINTF(("%s: enqueue record = %d\n", __func__, retval));
2713 if (retval <= 0) {
2714 MLI_UNLOCK(mli);
2715 retval *= -1;
2716 goto done;
2717 } else {
2718 retval = 0;
2719 }
2720
2721 /*
2722 * If record(s) were enqueued, start the state-change
2723 * report timer for this group.
2724 */
2725 inm->in6m_scrv = (uint16_t)mli->mli_rv;
2726 inm->in6m_sctimer = 1;
2727 mtp->sct = 1;
2728 MLI_UNLOCK(mli);
2729
2730done:
2731 return retval;
2732}
2733
2734/*
2735 * Perform the final leave for a multicast address.
2736 *
2737 * When leaving a group:
2738 * MLDv1 sends a DONE message, if and only if we are the reporter.
2739 * MLDv2 enqueues a state-change report containing a transition
2740 * to INCLUDE {} for immediate transmission.
2741 */
2742static void
2743mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli,
2744 struct mld_tparams *mtp)
2745{
2746 int syncstates = 1;
2747
2748 IN6M_LOCK_ASSERT_HELD(inm);
2749 MLI_LOCK_ASSERT_NOTHELD(mli);
2750 VERIFY(mtp != NULL);
2751
2752 MLD_PRINTF(("%s: final leave %s on ifp 0x%llx(%s)\n",
2753 __func__, ip6_sprintf(&inm->in6m_addr),
2754 (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2755 if_name(inm->in6m_ifp)));
2756
2757 switch (inm->in6m_state) {
2758 case MLD_NOT_MEMBER:
2759 case MLD_SILENT_MEMBER:
2760 case MLD_LEAVING_MEMBER:
2761 /* Already leaving or left; do nothing. */
2762 MLD_PRINTF(("%s: not kicking state machine for silent group\n",
2763 __func__));
2764 break;
2765 case MLD_REPORTING_MEMBER:
2766 case MLD_IDLE_MEMBER:
2767 case MLD_G_QUERY_PENDING_MEMBER:
2768 case MLD_SG_QUERY_PENDING_MEMBER:
2769 MLI_LOCK(mli);
2770 if (mli->mli_version == MLD_VERSION_1) {
2771 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2772 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
2773 panic("%s: MLDv2 state reached, not MLDv2 "
2774 "mode\n", __func__);
2775 /* NOTREACHED */
2776 }
2777 /* scheduler timer if enqueue is successful */
2778 mtp->cst = (mld_v1_transmit_report(in6m: inm,
2779 MLD_LISTENER_DONE) == 0);
2780
2781 IN6M_LOCK_ASSERT_HELD(inm);
2782 MLI_LOCK_ASSERT_HELD(mli);
2783
2784 inm->in6m_state = MLD_NOT_MEMBER;
2785 } else if (mli->mli_version == MLD_VERSION_2) {
2786 /*
2787 * Stop group timer and all pending reports.
2788 * Immediately enqueue a state-change report
2789 * TO_IN {} to be sent on the next timeout,
2790 * giving us an opportunity to merge reports.
2791 */
2792 IF_DRAIN(&inm->in6m_scq);
2793 inm->in6m_timer = 0;
2794 inm->in6m_scrv = (uint16_t)mli->mli_rv;
2795 MLD_PRINTF(("%s: Leaving %s/%s with %d "
2796 "pending retransmissions.\n", __func__,
2797 ip6_sprintf(&inm->in6m_addr),
2798 if_name(inm->in6m_ifp),
2799 inm->in6m_scrv));
2800 if (inm->in6m_scrv == 0) {
2801 inm->in6m_state = MLD_NOT_MEMBER;
2802 inm->in6m_sctimer = 0;
2803 } else {
2804 int retval;
2805 /*
2806 * Stick around in the in6_multihead list;
2807 * the final detach will be issued by
2808 * mld_v2_process_group_timers() when
2809 * the retransmit timer expires.
2810 */
2811 IN6M_ADDREF_LOCKED(inm);
2812 VERIFY(inm->in6m_debug & IFD_ATTACHED);
2813 inm->in6m_reqcnt++;
2814 VERIFY(inm->in6m_reqcnt >= 1);
2815 inm->in6m_nrelecnt++;
2816 VERIFY(inm->in6m_nrelecnt != 0);
2817
2818 retval = mld_v2_enqueue_group_record(
2819 &inm->in6m_scq, inm, 1, 0, 0,
2820 (mli->mli_flags & MLIF_USEALLOW));
2821 mtp->cst = (inm->in6m_scq.ifq_len > 0);
2822 KASSERT(retval != 0,
2823 ("%s: enqueue record = %d\n", __func__,
2824 retval));
2825
2826 inm->in6m_state = MLD_LEAVING_MEMBER;
2827 inm->in6m_sctimer = 1;
2828 mtp->sct = 1;
2829 syncstates = 0;
2830 }
2831 }
2832 MLI_UNLOCK(mli);
2833 break;
2834 case MLD_LAZY_MEMBER:
2835 case MLD_SLEEPING_MEMBER:
2836 case MLD_AWAKENING_MEMBER:
2837 /* Our reports are suppressed; do nothing. */
2838 break;
2839 }
2840
2841 if (syncstates) {
2842 in6m_commit(inm);
2843 MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2844 ip6_sprintf(&inm->in6m_addr),
2845 if_name(inm->in6m_ifp)));
2846 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2847 MLD_PRINTF(("%s: T1 now MCAST_UNDEFINED for 0x%llx/%s\n",
2848 __func__, (uint64_t)VM_KERNEL_ADDRPERM(&inm->in6m_addr),
2849 if_name(inm->in6m_ifp)));
2850 }
2851}
2852
2853/*
2854 * Enqueue an MLDv2 group record to the given output queue.
2855 *
2856 * If is_state_change is zero, a current-state record is appended.
2857 * If is_state_change is non-zero, a state-change report is appended.
2858 *
2859 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2860 * If is_group_query is zero, and if there is a packet with free space
2861 * at the tail of the queue, it will be appended to providing there
2862 * is enough free space.
2863 * Otherwise a new mbuf packet chain is allocated.
2864 *
2865 * If is_source_query is non-zero, each source is checked to see if
2866 * it was recorded for a Group-Source query, and will be omitted if
2867 * it is not both in-mode and recorded.
2868 *
2869 * If use_block_allow is non-zero, state change reports for initial join
2870 * and final leave, on an inclusive mode group with a source list, will be
2871 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2872 *
2873 * The function will attempt to allocate leading space in the packet
2874 * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2875 *
2876 * If successful the size of all data appended to the queue is returned,
2877 * otherwise an error code less than zero is returned, or zero if
2878 * no record(s) were appended.
2879 */
2880static int
2881mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2882 const int is_state_change, const int is_group_query,
2883 const int is_source_query, const int use_block_allow)
2884{
2885 struct mldv2_record mr;
2886 struct mldv2_record *pmr;
2887 struct ifnet *ifp;
2888 struct ip6_msource *ims, *nims;
2889 mbuf_ref_t m0, m, md;
2890 int error, is_filter_list_change;
2891 int minrec0len, m0srcs, msrcs, nbytes, off;
2892 int record_has_sources;
2893 int now;
2894 uint8_t type;
2895 uint8_t mode;
2896
2897 IN6M_LOCK_ASSERT_HELD(inm);
2898 MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
2899
2900 error = 0;
2901 ifp = inm->in6m_ifp;
2902 is_filter_list_change = 0;
2903 m = NULL;
2904 m0 = NULL;
2905 m0srcs = 0;
2906 msrcs = 0;
2907 nbytes = 0;
2908 nims = NULL;
2909 record_has_sources = 1;
2910 pmr = NULL;
2911 type = MLD_DO_NOTHING;
2912 mode = (uint8_t)inm->in6m_st[1].iss_fmode;
2913
2914 /*
2915 * If we did not transition out of ASM mode during t0->t1,
2916 * and there are no source nodes to process, we can skip
2917 * the generation of source records.
2918 */
2919 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2920 inm->in6m_nsrc == 0) {
2921 record_has_sources = 0;
2922 }
2923
2924 if (is_state_change) {
2925 /*
2926 * Queue a state change record.
2927 * If the mode did not change, and there are non-ASM
2928 * listeners or source filters present,
2929 * we potentially need to issue two records for the group.
2930 * If there are ASM listeners, and there was no filter
2931 * mode transition of any kind, do nothing.
2932 *
2933 * If we are transitioning to MCAST_UNDEFINED, we need
2934 * not send any sources. A transition to/from this state is
2935 * considered inclusive with some special treatment.
2936 *
2937 * If we are rewriting initial joins/leaves to use
2938 * ALLOW/BLOCK, and the group's membership is inclusive,
2939 * we need to send sources in all cases.
2940 */
2941 if (mode != inm->in6m_st[0].iss_fmode) {
2942 if (mode == MCAST_EXCLUDE) {
2943 MLD_PRINTF(("%s: change to EXCLUDE\n",
2944 __func__));
2945 type = MLD_CHANGE_TO_EXCLUDE_MODE;
2946 } else {
2947 MLD_PRINTF(("%s: change to INCLUDE\n",
2948 __func__));
2949 if (use_block_allow) {
2950 /*
2951 * XXX
2952 * Here we're interested in state
2953 * edges either direction between
2954 * MCAST_UNDEFINED and MCAST_INCLUDE.
2955 * Perhaps we should just check
2956 * the group state, rather than
2957 * the filter mode.
2958 */
2959 if (mode == MCAST_UNDEFINED) {
2960 type = MLD_BLOCK_OLD_SOURCES;
2961 } else {
2962 type = MLD_ALLOW_NEW_SOURCES;
2963 }
2964 } else {
2965 type = MLD_CHANGE_TO_INCLUDE_MODE;
2966 if (mode == MCAST_UNDEFINED) {
2967 record_has_sources = 0;
2968 }
2969 }
2970 }
2971 } else {
2972 if (record_has_sources) {
2973 is_filter_list_change = 1;
2974 } else {
2975 type = MLD_DO_NOTHING;
2976 }
2977 }
2978 } else {
2979 /*
2980 * Queue a current state record.
2981 */
2982 if (mode == MCAST_EXCLUDE) {
2983 type = MLD_MODE_IS_EXCLUDE;
2984 } else if (mode == MCAST_INCLUDE) {
2985 type = MLD_MODE_IS_INCLUDE;
2986 VERIFY(inm->in6m_st[1].iss_asm == 0);
2987 }
2988 }
2989
2990 /*
2991 * Generate the filter list changes using a separate function.
2992 */
2993 if (is_filter_list_change) {
2994 return mld_v2_enqueue_filter_change(ifq, inm);
2995 }
2996
2997 if (type == MLD_DO_NOTHING) {
2998 MLD_PRINTF(("%s: nothing to do for %s/%s\n",
2999 __func__, ip6_sprintf(&inm->in6m_addr),
3000 if_name(inm->in6m_ifp)));
3001 return 0;
3002 }
3003
3004 /*
3005 * If any sources are present, we must be able to fit at least
3006 * one in the trailing space of the tail packet's mbuf,
3007 * ideally more.
3008 */
3009 minrec0len = sizeof(struct mldv2_record);
3010 if (record_has_sources) {
3011 minrec0len += sizeof(struct in6_addr);
3012 }
3013 MLD_PRINTF(("%s: queueing %s for %s/%s\n", __func__,
3014 mld_rec_type_to_str(type),
3015 ip6_sprintf(&inm->in6m_addr),
3016 if_name(inm->in6m_ifp)));
3017
3018 /*
3019 * Check if we have a packet in the tail of the queue for this
3020 * group into which the first group record for this group will fit.
3021 * Otherwise allocate a new packet.
3022 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
3023 * Note: Group records for G/GSR query responses MUST be sent
3024 * in their own packet.
3025 */
3026 m0 = ifq->ifq_tail;
3027 if (!is_group_query &&
3028 m0 != NULL &&
3029 (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
3030 (m0->m_pkthdr.len + minrec0len) <
3031 (ifp->if_mtu - MLD_MTUSPACE)) {
3032 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3033 sizeof(struct mldv2_record)) /
3034 sizeof(struct in6_addr);
3035 m = m0;
3036 MLD_PRINTF(("%s: use existing packet\n", __func__));
3037 } else {
3038 if (IF_QFULL(ifq)) {
3039 os_log_error(OS_LOG_DEFAULT,
3040 "%s: outbound queue full\n", __func__);
3041 return -ENOMEM;
3042 }
3043 m = NULL;
3044 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3045 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
3046 if (!is_state_change && !is_group_query) {
3047 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3048 }
3049 if (m == NULL) {
3050 m = m_gethdr(M_DONTWAIT, MT_DATA);
3051 }
3052 if (m == NULL) {
3053 return -ENOMEM;
3054 }
3055
3056 mld_save_context(m, ifp);
3057
3058 MLD_PRINTF(("%s: allocated first packet\n", __func__));
3059 }
3060
3061 /*
3062 * Append group record.
3063 * If we have sources, we don't know how many yet.
3064 */
3065 mr.mr_type = type;
3066 mr.mr_datalen = 0;
3067 mr.mr_numsrc = 0;
3068 mr.mr_addr = inm->in6m_addr;
3069 in6_clearscope(&mr.mr_addr);
3070 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
3071 if (m != m0) {
3072 m_freem(m);
3073 }
3074 os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed.\n", __func__);
3075 return -ENOMEM;
3076 }
3077 nbytes += sizeof(struct mldv2_record);
3078
3079 /*
3080 * Append as many sources as will fit in the first packet.
3081 * If we are appending to a new packet, the chain allocation
3082 * may potentially use clusters; use m_getptr() in this case.
3083 * If we are appending to an existing packet, we need to obtain
3084 * a pointer to the group record after m_append(), in case a new
3085 * mbuf was allocated.
3086 *
3087 * Only append sources which are in-mode at t1. If we are
3088 * transitioning to MCAST_UNDEFINED state on the group, and
3089 * use_block_allow is zero, do not include source entries.
3090 * Otherwise, we need to include this source in the report.
3091 *
3092 * Only report recorded sources in our filter set when responding
3093 * to a group-source query.
3094 */
3095 if (record_has_sources) {
3096 if (m == m0) {
3097 md = m_last(m);
3098 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
3099 md->m_len - nbytes);
3100 } else {
3101 md = m_getptr(m, 0, &off);
3102 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
3103 off);
3104 }
3105 msrcs = 0;
3106 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
3107 nims) {
3108 MLD_PRINTF(("%s: visit node %s\n", __func__,
3109 ip6_sprintf(&ims->im6s_addr)));
3110 now = im6s_get_mode(inm, ims, 1);
3111 MLD_PRINTF(("%s: node is %d\n", __func__, now));
3112 if ((now != mode) ||
3113 (now == mode &&
3114 (!use_block_allow && mode == MCAST_UNDEFINED))) {
3115 MLD_PRINTF(("%s: skip node\n", __func__));
3116 continue;
3117 }
3118 if (is_source_query && ims->im6s_stp == 0) {
3119 MLD_PRINTF(("%s: skip unrecorded node\n",
3120 __func__));
3121 continue;
3122 }
3123 MLD_PRINTF(("%s: append node\n", __func__));
3124 if (!m_append(m, sizeof(struct in6_addr),
3125 (void *)&ims->im6s_addr)) {
3126 if (m != m0) {
3127 m_freem(m);
3128 }
3129 os_log_error(OS_LOG_DEFAULT,
3130 "%s: m_append() failed\n",
3131 __func__);
3132 return -ENOMEM;
3133 }
3134 nbytes += sizeof(struct in6_addr);
3135 ++msrcs;
3136 if (msrcs == m0srcs) {
3137 break;
3138 }
3139 }
3140 MLD_PRINTF(("%s: msrcs is %d this packet\n", __func__,
3141 msrcs));
3142 pmr->mr_numsrc = htons((uint16_t)msrcs);
3143 nbytes += (msrcs * sizeof(struct in6_addr));
3144 }
3145
3146 if (is_source_query && msrcs == 0) {
3147 MLD_PRINTF(("%s: no recorded sources to report\n", __func__));
3148 if (m != m0) {
3149 m_freem(m);
3150 }
3151 return 0;
3152 }
3153
3154 /*
3155 * We are good to go with first packet.
3156 */
3157 if (m != m0) {
3158 MLD_PRINTF(("%s: enqueueing first packet\n", __func__));
3159 m->m_pkthdr.vt_nrecs = 1;
3160 IF_ENQUEUE(ifq, m);
3161 } else {
3162 m->m_pkthdr.vt_nrecs++;
3163 }
3164 /*
3165 * No further work needed if no source list in packet(s).
3166 */
3167 if (!record_has_sources) {
3168 return nbytes;
3169 }
3170
3171 /*
3172 * Whilst sources remain to be announced, we need to allocate
3173 * a new packet and fill out as many sources as will fit.
3174 * Always try for a cluster first.
3175 */
3176 while (nims != NULL) {
3177 if (IF_QFULL(ifq)) {
3178 os_log_error(OS_LOG_DEFAULT, "%s: outbound queue full\n", __func__);
3179 return -ENOMEM;
3180 }
3181 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3182 if (m == NULL) {
3183 m = m_gethdr(M_DONTWAIT, MT_DATA);
3184 }
3185 if (m == NULL) {
3186 return -ENOMEM;
3187 }
3188 mld_save_context(m, ifp);
3189 md = m_getptr(m, 0, &off);
3190 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
3191 MLD_PRINTF(("%s: allocated next packet\n", __func__));
3192
3193 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
3194 if (m != m0) {
3195 m_freem(m);
3196 }
3197 os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed.\n", __func__);
3198 return -ENOMEM;
3199 }
3200 m->m_pkthdr.vt_nrecs = 1;
3201 nbytes += sizeof(struct mldv2_record);
3202
3203 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3204 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
3205
3206 msrcs = 0;
3207 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
3208 MLD_PRINTF(("%s: visit node %s\n",
3209 __func__, ip6_sprintf(&ims->im6s_addr)));
3210 now = im6s_get_mode(inm, ims, 1);
3211 if ((now != mode) ||
3212 (now == mode &&
3213 (!use_block_allow && mode == MCAST_UNDEFINED))) {
3214 MLD_PRINTF(("%s: skip node\n", __func__));
3215 continue;
3216 }
3217 if (is_source_query && ims->im6s_stp == 0) {
3218 MLD_PRINTF(("%s: skip unrecorded node\n",
3219 __func__));
3220 continue;
3221 }
3222 MLD_PRINTF(("%s: append node\n", __func__));
3223 if (!m_append(m, sizeof(struct in6_addr),
3224 (void *)&ims->im6s_addr)) {
3225 if (m != m0) {
3226 m_freem(m);
3227 }
3228 os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed\n",
3229 __func__);
3230 return -ENOMEM;
3231 }
3232 ++msrcs;
3233 if (msrcs == m0srcs) {
3234 break;
3235 }
3236 }
3237 pmr->mr_numsrc = htons((uint16_t)msrcs);
3238 nbytes += (msrcs * sizeof(struct in6_addr));
3239
3240 MLD_PRINTF(("%s: enqueueing next packet\n", __func__));
3241 IF_ENQUEUE(ifq, m);
3242 }
3243
3244 return nbytes;
3245}
3246
3247/*
3248 * Type used to mark record pass completion.
3249 * We exploit the fact we can cast to this easily from the
3250 * current filter modes on each ip_msource node.
3251 */
3252typedef enum {
3253 REC_NONE = 0x00, /* MCAST_UNDEFINED */
3254 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
3255 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
3256 REC_FULL = REC_ALLOW | REC_BLOCK
3257} rectype_t;
3258
3259/*
3260 * Enqueue an MLDv2 filter list change to the given output queue.
3261 *
3262 * Source list filter state is held in an RB-tree. When the filter list
3263 * for a group is changed without changing its mode, we need to compute
3264 * the deltas between T0 and T1 for each source in the filter set,
3265 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3266 *
3267 * As we may potentially queue two record types, and the entire R-B tree
3268 * needs to be walked at once, we break this out into its own function
3269 * so we can generate a tightly packed queue of packets.
3270 *
3271 * XXX This could be written to only use one tree walk, although that makes
3272 * serializing into the mbuf chains a bit harder. For now we do two walks
3273 * which makes things easier on us, and it may or may not be harder on
3274 * the L2 cache.
3275 *
3276 * If successful the size of all data appended to the queue is returned,
3277 * otherwise an error code less than zero is returned, or zero if
3278 * no record(s) were appended.
3279 */
3280static int
3281mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
3282{
3283 static const int MINRECLEN =
3284 sizeof(struct mldv2_record) + sizeof(struct in6_addr);
3285 struct ifnet *ifp;
3286 struct mldv2_record mr;
3287 struct mldv2_record *pmr;
3288 struct ip6_msource *ims, *nims;
3289 mbuf_ref_t m, m0, md;
3290 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3291 int nallow, nblock;
3292 uint8_t mode, now, then;
3293 rectype_t crt, drt, nrt;
3294
3295 IN6M_LOCK_ASSERT_HELD(inm);
3296
3297 if (inm->in6m_nsrc == 0 ||
3298 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) {
3299 return 0;
3300 }
3301
3302 ifp = inm->in6m_ifp; /* interface */
3303 mode = (uint8_t)inm->in6m_st[1].iss_fmode; /* filter mode at t1 */
3304 crt = REC_NONE; /* current group record type */
3305 drt = REC_NONE; /* mask of completed group record types */
3306 nrt = REC_NONE; /* record type for current node */
3307 m0srcs = 0; /* # source which will fit in current mbuf chain */
3308 npbytes = 0; /* # of bytes appended this packet */
3309 nbytes = 0; /* # of bytes appended to group's state-change queue */
3310 rsrcs = 0; /* # sources encoded in current record */
3311 schanged = 0; /* # nodes encoded in overall filter change */
3312 nallow = 0; /* # of source entries in ALLOW_NEW */
3313 nblock = 0; /* # of source entries in BLOCK_OLD */
3314 nims = NULL; /* next tree node pointer */
3315
3316 /*
3317 * For each possible filter record mode.
3318 * The first kind of source we encounter tells us which
3319 * is the first kind of record we start appending.
3320 * If a node transitioned to UNDEFINED at t1, its mode is treated
3321 * as the inverse of the group's filter mode.
3322 */
3323 while (drt != REC_FULL) {
3324 do {
3325 m0 = ifq->ifq_tail;
3326 if (m0 != NULL &&
3327 (m0->m_pkthdr.vt_nrecs + 1 <=
3328 MLD_V2_REPORT_MAXRECS) &&
3329 (m0->m_pkthdr.len + MINRECLEN) <
3330 (ifp->if_mtu - MLD_MTUSPACE)) {
3331 m = m0;
3332 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3333 sizeof(struct mldv2_record)) /
3334 sizeof(struct in6_addr);
3335 MLD_PRINTF(("%s: use previous packet\n",
3336 __func__));
3337 } else {
3338 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3339 if (m == NULL) {
3340 m = m_gethdr(M_DONTWAIT, MT_DATA);
3341 }
3342 if (m == NULL) {
3343 os_log_error(OS_LOG_DEFAULT, "%s: m_get*() failed\n",
3344 __func__);
3345 return -ENOMEM;
3346 }
3347 m->m_pkthdr.vt_nrecs = 0;
3348 mld_save_context(m, ifp);
3349 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3350 sizeof(struct mldv2_record)) /
3351 sizeof(struct in6_addr);
3352 npbytes = 0;
3353 MLD_PRINTF(("%s: allocated new packet\n",
3354 __func__));
3355 }
3356 /*
3357 * Append the MLD group record header to the
3358 * current packet's data area.
3359 * Recalculate pointer to free space for next
3360 * group record, in case m_append() allocated
3361 * a new mbuf or cluster.
3362 */
3363 memset(s: &mr, c: 0, n: sizeof(mr));
3364 mr.mr_addr = inm->in6m_addr;
3365 in6_clearscope(&mr.mr_addr);
3366 if (!m_append(m, sizeof(mr), (void *)&mr)) {
3367 if (m != m0) {
3368 m_freem(m);
3369 }
3370 os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed\n",
3371 __func__);
3372 return -ENOMEM;
3373 }
3374 npbytes += sizeof(struct mldv2_record);
3375 if (m != m0) {
3376 /* new packet; offset in chain */
3377 md = m_getptr(m, npbytes -
3378 sizeof(struct mldv2_record), &off);
3379 pmr = (struct mldv2_record *)(mtod(md,
3380 uint8_t *) + off);
3381 } else {
3382 /* current packet; offset from last append */
3383 md = m_last(m);
3384 pmr = (struct mldv2_record *)(mtod(md,
3385 uint8_t *) + md->m_len -
3386 sizeof(struct mldv2_record));
3387 }
3388 /*
3389 * Begin walking the tree for this record type
3390 * pass, or continue from where we left off
3391 * previously if we had to allocate a new packet.
3392 * Only report deltas in-mode at t1.
3393 * We need not report included sources as allowed
3394 * if we are in inclusive mode on the group,
3395 * however the converse is not true.
3396 */
3397 rsrcs = 0;
3398 if (nims == NULL) {
3399 nims = RB_MIN(ip6_msource_tree,
3400 &inm->in6m_srcs);
3401 }
3402 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
3403 MLD_PRINTF(("%s: visit node %s\n", __func__,
3404 ip6_sprintf(&ims->im6s_addr)));
3405 now = im6s_get_mode(inm, ims, 1);
3406 then = im6s_get_mode(inm, ims, 0);
3407 MLD_PRINTF(("%s: mode: t0 %d, t1 %d\n",
3408 __func__, then, now));
3409 if (now == then) {
3410 MLD_PRINTF(("%s: skip unchanged\n",
3411 __func__));
3412 continue;
3413 }
3414 if (mode == MCAST_EXCLUDE &&
3415 now == MCAST_INCLUDE) {
3416 MLD_PRINTF(("%s: skip IN src on EX "
3417 "group\n", __func__));
3418 continue;
3419 }
3420 nrt = (rectype_t)now;
3421 if (nrt == REC_NONE) {
3422 nrt = (rectype_t)(~mode & REC_FULL);
3423 }
3424 if (schanged++ == 0) {
3425 crt = nrt;
3426 } else if (crt != nrt) {
3427 continue;
3428 }
3429 if (!m_append(m, sizeof(struct in6_addr),
3430 (void *)&ims->im6s_addr)) {
3431 if (m != m0) {
3432 m_freem(m);
3433 }
3434 os_log_error(OS_LOG_DEFAULT, "%s: m_append() failed\n",
3435 __func__);
3436 return -ENOMEM;
3437 }
3438 nallow += !!(crt == REC_ALLOW);
3439 nblock += !!(crt == REC_BLOCK);
3440 if (++rsrcs == m0srcs) {
3441 break;
3442 }
3443 }
3444 /*
3445 * If we did not append any tree nodes on this
3446 * pass, back out of allocations.
3447 */
3448 if (rsrcs == 0) {
3449 npbytes -= sizeof(struct mldv2_record);
3450 if (m != m0) {
3451 MLD_PRINTF(("%s: m_free(m)\n",
3452 __func__));
3453 m_freem(m);
3454 } else {
3455 MLD_PRINTF(("%s: m_adj(m, -mr)\n",
3456 __func__));
3457 m_adj(m, -((int)sizeof(
3458 struct mldv2_record)));
3459 }
3460 continue;
3461 }
3462 npbytes += (rsrcs * sizeof(struct in6_addr));
3463 if (crt == REC_ALLOW) {
3464 pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
3465 } else if (crt == REC_BLOCK) {
3466 pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
3467 }
3468 pmr->mr_numsrc = htons((uint16_t)rsrcs);
3469 /*
3470 * Count the new group record, and enqueue this
3471 * packet if it wasn't already queued.
3472 */
3473 m->m_pkthdr.vt_nrecs++;
3474 if (m != m0) {
3475 IF_ENQUEUE(ifq, m);
3476 }
3477 nbytes += npbytes;
3478 } while (nims != NULL);
3479 drt |= crt;
3480 crt = (~crt & REC_FULL);
3481 }
3482
3483 MLD_PRINTF(("%s: queued %d ALLOW_NEW, %d BLOCK_OLD\n", __func__,
3484 nallow, nblock));
3485
3486 return nbytes;
3487}
3488
3489static int
3490mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
3491{
3492 struct ifqueue *gq;
3493 mbuf_ref_t m; /* pending state-change */
3494 mbuf_ref_t m0; /* copy of pending state-change */
3495 mbuf_ref_t mt; /* last state-change in packet */
3496 mbuf_ref_t n;
3497 int docopy, domerge;
3498 u_int recslen;
3499
3500 IN6M_LOCK_ASSERT_HELD(inm);
3501
3502 docopy = 0;
3503 domerge = 0;
3504 recslen = 0;
3505
3506 /*
3507 * If there are further pending retransmissions, make a writable
3508 * copy of each queued state-change message before merging.
3509 */
3510 if (inm->in6m_scrv > 0) {
3511 docopy = 1;
3512 }
3513
3514 gq = &inm->in6m_scq;
3515#ifdef MLD_DEBUG
3516 if (gq->ifq_head == NULL) {
3517 MLD_PRINTF(("%s: WARNING: queue for inm 0x%llx is empty\n",
3518 __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm)));
3519 }
3520#endif
3521
3522 /*
3523 * Use IF_REMQUEUE() instead of IF_DEQUEUE() below, since the
3524 * packet might not always be at the head of the ifqueue.
3525 */
3526 m = gq->ifq_head;
3527 while (m != NULL) {
3528 /*
3529 * Only merge the report into the current packet if
3530 * there is sufficient space to do so; an MLDv2 report
3531 * packet may only contain 65,535 group records.
3532 * Always use a simple mbuf chain concatentation to do this,
3533 * as large state changes for single groups may have
3534 * allocated clusters.
3535 */
3536 domerge = 0;
3537 mt = ifscq->ifq_tail;
3538 if (mt != NULL) {
3539 recslen = m_length(m);
3540
3541 if ((mt->m_pkthdr.vt_nrecs +
3542 m->m_pkthdr.vt_nrecs <=
3543 MLD_V2_REPORT_MAXRECS) &&
3544 (mt->m_pkthdr.len + recslen <=
3545 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) {
3546 domerge = 1;
3547 }
3548 }
3549
3550 if (!domerge && IF_QFULL(gq)) {
3551 os_log_info(OS_LOG_DEFAULT, "%s: outbound queue full",
3552 __func__);
3553 n = m->m_nextpkt;
3554 if (!docopy) {
3555 IF_REMQUEUE(gq, m);
3556 m_freem(m);
3557 }
3558 m = n;
3559 continue;
3560 }
3561
3562 if (!docopy) {
3563 MLD_PRINTF(("%s: dequeueing 0x%llx\n", __func__,
3564 (uint64_t)VM_KERNEL_ADDRPERM(m)));
3565 n = m->m_nextpkt;
3566 IF_REMQUEUE(gq, m);
3567 m0 = m;
3568 m = n;
3569 } else {
3570 MLD_PRINTF(("%s: copying 0x%llx\n", __func__,
3571 (uint64_t)VM_KERNEL_ADDRPERM(m)));
3572 m0 = m_dup(m, M_NOWAIT);
3573 if (m0 == NULL) {
3574 return ENOMEM;
3575 }
3576 m0->m_nextpkt = NULL;
3577 m = m->m_nextpkt;
3578 }
3579
3580 if (!domerge) {
3581 MLD_PRINTF(("%s: queueing 0x%llx to ifscq 0x%llx)\n",
3582 __func__, (uint64_t)VM_KERNEL_ADDRPERM(m0),
3583 (uint64_t)VM_KERNEL_ADDRPERM(ifscq)));
3584 IF_ENQUEUE(ifscq, m0);
3585 } else {
3586 struct mbuf *mtl; /* last mbuf of packet mt */
3587
3588 MLD_PRINTF(("%s: merging 0x%llx with ifscq tail "
3589 "0x%llx)\n", __func__,
3590 (uint64_t)VM_KERNEL_ADDRPERM(m0),
3591 (uint64_t)VM_KERNEL_ADDRPERM(mt)));
3592
3593 mtl = m_last(mt);
3594 m0->m_flags &= ~M_PKTHDR;
3595 mt->m_pkthdr.len += recslen;
3596 mt->m_pkthdr.vt_nrecs +=
3597 m0->m_pkthdr.vt_nrecs;
3598
3599 mtl->m_next = m0;
3600 }
3601 }
3602
3603 return 0;
3604}
3605
3606/*
3607 * Respond to a pending MLDv2 General Query.
3608 */
3609static uint32_t
3610mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
3611{
3612 struct ifnet *ifp;
3613 struct in6_multi *inm;
3614 struct in6_multistep step;
3615 int retval;
3616
3617 MLI_LOCK_ASSERT_HELD(mli);
3618
3619 VERIFY(mli->mli_version == MLD_VERSION_2);
3620
3621 ifp = mli->mli_ifp;
3622 MLI_UNLOCK(mli);
3623
3624 in6_multihead_lock_shared();
3625 IN6_FIRST_MULTI(step, inm);
3626 while (inm != NULL) {
3627 IN6M_LOCK(inm);
3628 if (inm->in6m_ifp != ifp) {
3629 goto next;
3630 }
3631
3632 switch (inm->in6m_state) {
3633 case MLD_NOT_MEMBER:
3634 case MLD_SILENT_MEMBER:
3635 break;
3636 case MLD_REPORTING_MEMBER:
3637 case MLD_IDLE_MEMBER:
3638 case MLD_LAZY_MEMBER:
3639 case MLD_SLEEPING_MEMBER:
3640 case MLD_AWAKENING_MEMBER:
3641 inm->in6m_state = MLD_REPORTING_MEMBER;
3642 MLI_LOCK(mli);
3643 retval = mld_v2_enqueue_group_record(ifq: &mli->mli_gq,
3644 inm, is_state_change: 0, is_group_query: 0, is_source_query: 0, use_block_allow: 0);
3645 MLI_UNLOCK(mli);
3646 MLD_PRINTF(("%s: enqueue record = %d\n",
3647 __func__, retval));
3648 break;
3649 case MLD_G_QUERY_PENDING_MEMBER:
3650 case MLD_SG_QUERY_PENDING_MEMBER:
3651 case MLD_LEAVING_MEMBER:
3652 break;
3653 }
3654next:
3655 IN6M_UNLOCK(inm);
3656 IN6_NEXT_MULTI(step, inm);
3657 }
3658 in6_multihead_lock_done();
3659
3660 MLI_LOCK(mli);
3661 mld_dispatch_queue_locked(mli, ifq: &mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3662 MLI_LOCK_ASSERT_HELD(mli);
3663
3664 /*
3665 * Slew transmission of bursts over 1 second intervals.
3666 */
3667 if (mli->mli_gq.ifq_head != NULL) {
3668 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3669 MLD_RESPONSE_BURST_INTERVAL);
3670 }
3671
3672 return mli->mli_v2_timer;
3673}
3674
3675/*
3676 * Transmit the next pending message in the output queue.
3677 *
3678 * Must not be called with in6m_lockm or mli_lock held.
3679 */
3680__attribute__((noinline))
3681static void
3682mld_dispatch_packet(struct mbuf *m)
3683{
3684 struct ip6_moptions *im6o;
3685 struct ifnet *ifp;
3686 struct ifnet *__single oifp = NULL;
3687 mbuf_ref_t m0, md;
3688 struct ip6_hdr *ip6;
3689 struct icmp6_hdr *icmp6;
3690 int error;
3691 int off;
3692 int type;
3693
3694 MLD_PRINTF(("%s: transmit 0x%llx\n", __func__,
3695 (uint64_t)VM_KERNEL_ADDRPERM(m)));
3696
3697 /*
3698 * Check if the ifnet is still attached.
3699 */
3700 ifp = mld_restore_context(m);
3701 if (ifp == NULL || !ifnet_is_attached(ifp, refio: 0)) {
3702 os_log_error(OS_LOG_DEFAULT, "%s: dropped 0x%llx as interface went away\n",
3703 __func__, (uint64_t)VM_KERNEL_ADDRPERM(m));
3704 m_freem(m);
3705 ip6stat.ip6s_noroute++;
3706 return;
3707 }
3708 im6o = ip6_allocmoptions(Z_WAITOK);
3709 if (im6o == NULL) {
3710 m_freem(m);
3711 return;
3712 }
3713
3714 im6o->im6o_multicast_hlim = 1;
3715 im6o->im6o_multicast_loop = 0;
3716 im6o->im6o_multicast_ifp = ifp;
3717 if (m->m_flags & M_MLDV1) {
3718 m0 = m;
3719 } else {
3720 m0 = mld_v2_encap_report(ifp, m);
3721 if (m0 == NULL) {
3722 os_log_error(OS_LOG_DEFAULT, "%s: dropped 0x%llx\n", __func__,
3723 (uint64_t)VM_KERNEL_ADDRPERM(m));
3724 /*
3725 * mld_v2_encap_report() has already freed our mbuf.
3726 */
3727 IM6O_REMREF(im6o);
3728 ip6stat.ip6s_odropped++;
3729 return;
3730 }
3731 }
3732 mld_scrub_context(m: m0);
3733 m->m_flags &= ~(M_PROTOFLAGS);
3734 m0->m_pkthdr.rcvif = lo_ifp;
3735
3736 ip6 = mtod(m0, struct ip6_hdr *);
3737 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL);
3738 ip6_output_setdstifscope(m0, ifp->if_index, NULL);
3739 /*
3740 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3741 * so we can bump the stats.
3742 */
3743 md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3744 icmp6 = (struct icmp6_hdr *)(mtod(md, uint8_t *) + off);
3745 type = icmp6->icmp6_type;
3746
3747 if (ifp->if_eflags & IFEF_TXSTART) {
3748 /*
3749 * Use control service class if the outgoing
3750 * interface supports transmit-start model.
3751 */
3752 (void) m_set_service_class(m0, MBUF_SC_CTL);
3753 }
3754
3755 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, im6o,
3756 &oifp, NULL);
3757
3758 IM6O_REMREF(im6o);
3759
3760 if (error) {
3761 os_log_error(OS_LOG_DEFAULT, "%s: ip6_output(0x%llx) = %d\n", __func__,
3762 (uint64_t)VM_KERNEL_ADDRPERM(m0), error);
3763 if (oifp != NULL) {
3764 ifnet_release(interface: oifp);
3765 }
3766 return;
3767 }
3768
3769 icmp6stat.icp6s_outhist[type]++;
3770 if (oifp != NULL) {
3771 icmp6_ifstat_inc(oifp, ifs6_out_msg);
3772 switch (type) {
3773 case MLD_LISTENER_REPORT:
3774 case MLDV2_LISTENER_REPORT:
3775 icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3776 break;
3777 case MLD_LISTENER_DONE:
3778 icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3779 break;
3780 }
3781 ifnet_release(interface: oifp);
3782 }
3783}
3784
3785/*
3786 * Encapsulate an MLDv2 report.
3787 *
3788 * KAME IPv6 requires that hop-by-hop options be passed separately,
3789 * and that the IPv6 header be prepended in a separate mbuf.
3790 *
3791 * Returns a pointer to the new mbuf chain head, or NULL if the
3792 * allocation failed.
3793 */
3794static struct mbuf *
3795mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3796{
3797 struct mbuf *mh;
3798 struct mldv2_report *mld;
3799 struct ip6_hdr *ip6;
3800 struct in6_ifaddr *ia;
3801 int mldreclen;
3802
3803 VERIFY(m->m_flags & M_PKTHDR);
3804
3805 /*
3806 * RFC3590: OK to send as :: or tentative during DAD.
3807 */
3808 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
3809 if (ia == NULL) {
3810 MLD_PRINTF(("%s: warning: ia is NULL\n", __func__));
3811 }
3812
3813 MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3814 if (mh == NULL) {
3815 if (ia != NULL) {
3816 ifa_remref(ifa: &ia->ia_ifa);
3817 }
3818 m_freem(m);
3819 return NULL;
3820 }
3821 MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3822
3823 mldreclen = m_length(m);
3824 MLD_PRINTF(("%s: mldreclen is %d\n", __func__, mldreclen));
3825
3826 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3827 mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3828 sizeof(struct mldv2_report) + mldreclen;
3829
3830 ip6 = mtod(mh, struct ip6_hdr *);
3831 ip6->ip6_flow = 0;
3832 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3833 ip6->ip6_vfc |= IPV6_VERSION;
3834 ip6->ip6_nxt = IPPROTO_ICMPV6;
3835 if (ia != NULL) {
3836 IFA_LOCK(&ia->ia_ifa);
3837 }
3838 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3839 ip6_output_setsrcifscope(mh, IFSCOPE_NONE, ia);
3840
3841 if (ia != NULL) {
3842 IFA_UNLOCK(&ia->ia_ifa);
3843 ifa_remref(ifa: &ia->ia_ifa);
3844 ia = NULL;
3845 }
3846 ip6->ip6_dst = in6addr_linklocal_allv2routers;
3847 ip6_output_setdstifscope(mh, ifp->if_index, NULL);
3848 /* scope ID will be set in netisr */
3849
3850 mld = (struct mldv2_report *)(ip6 + 1);
3851 mld->mld_type = MLDV2_LISTENER_REPORT;
3852 mld->mld_code = 0;
3853 mld->mld_cksum = 0;
3854 mld->mld_v2_reserved = 0;
3855 mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3856 m->m_pkthdr.vt_nrecs = 0;
3857 m->m_flags &= ~M_PKTHDR;
3858
3859 mh->m_next = m;
3860 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3861 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3862 return mh;
3863}
3864
3865#ifdef MLD_DEBUG
3866static const char *
3867mld_rec_type_to_str(const int type)
3868{
3869 switch (type) {
3870 case MLD_CHANGE_TO_EXCLUDE_MODE:
3871 return "TO_EX";
3872 case MLD_CHANGE_TO_INCLUDE_MODE:
3873 return "TO_IN";
3874 case MLD_MODE_IS_EXCLUDE:
3875 return "MODE_EX";
3876 case MLD_MODE_IS_INCLUDE:
3877 return "MODE_IN";
3878 case MLD_ALLOW_NEW_SOURCES:
3879 return "ALLOW_NEW";
3880 case MLD_BLOCK_OLD_SOURCES:
3881 return "BLOCK_OLD";
3882 default:
3883 break;
3884 }
3885 return "unknown";
3886}
3887#endif
3888
3889void
3890mld_init(void)
3891{
3892 os_log(OS_LOG_DEFAULT, "%s: initializing\n", __func__);
3893
3894 ip6_initpktopts(&mld_po);
3895 mld_po.ip6po_hlim = 1;
3896 mld_po.ip6po_hbh = &mld_ra.hbh;
3897 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3898 mld_po.ip6po_flags = IP6PO_DONTFRAG;
3899 LIST_INIT(&mli_head);
3900}
3901