1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 2009 Bruce Simpson.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote
15 * products derived from this software without specific prior written
16 * permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 *
30 * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
31 */
32
33 /*-
34 * Copyright (c) 1988 Stephen Deering.
35 * Copyright (c) 1992, 1993
36 * The Regents of the University of California. All rights reserved.
37 *
38 * This code is derived from software contributed to Berkeley by
39 * Stephen Deering of Stanford University.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 * notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 * notice, this list of conditions and the following disclaimer in the
48 * documentation and/or other materials provided with the distribution.
49 * 3. Neither the name of the University nor the names of its contributors
50 * may be used to endorse or promote products derived from this software
51 * without specific prior written permission.
52 *
53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63 * SUCH DAMAGE.
64 *
65 * @(#)igmp.c 8.1 (Berkeley) 7/19/93
66 */
67
68 #include <sys/cdefs.h>
69 __FBSDID("$FreeBSD$");
70
71 #include "opt_inet.h"
72 #include "opt_inet6.h"
73
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/mbuf.h>
77 #include <sys/socket.h>
78 #include <sys/protosw.h>
79 #include <sys/sysctl.h>
80 #include <sys/kernel.h>
81 #include <sys/callout.h>
82 #include <sys/malloc.h>
83 #include <sys/module.h>
84 #include <sys/ktr.h>
85
86 #include <net/if.h>
87 #include <net/if_var.h>
88 #include <net/route.h>
89 #include <net/vnet.h>
90
91 #include <netinet/in.h>
92 #include <netinet/in_var.h>
93 #include <netinet6/in6_var.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/mld6.h>
99 #include <netinet6/mld6_var.h>
100
101 #include <security/mac/mac_framework.h>
102
103 #ifndef KTR_MLD
104 #define KTR_MLD KTR_INET6
105 #endif
106
107 static struct mld_ifsoftc *
108 mli_alloc_locked(struct ifnet *);
109 static void mli_delete_locked(const struct ifnet *);
110 static void mld_dispatch_packet(struct mbuf *);
111 static void mld_dispatch_queue(struct mbufq *, int);
112 static void mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
113 static void mld_fasttimo_vnet(struct in6_multi_head *inmh);
114 static int mld_handle_state_change(struct in6_multi *,
115 struct mld_ifsoftc *);
116 static int mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
117 const int);
118 #ifdef KTR
119 static char * mld_rec_type_to_str(const int);
120 #endif
121 static void mld_set_version(struct mld_ifsoftc *, const int);
122 static void mld_slowtimo_vnet(void);
123 static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
124 /*const*/ struct mld_hdr *);
125 static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
126 /*const*/ struct mld_hdr *);
127 static void mld_v1_process_group_timer(struct in6_multi_head *,
128 struct in6_multi *);
129 static void mld_v1_process_querier_timers(struct mld_ifsoftc *);
130 static int mld_v1_transmit_report(struct in6_multi *, const int);
131 static void mld_v1_update_group(struct in6_multi *, const int);
132 static void mld_v2_cancel_link_timers(struct mld_ifsoftc *);
133 static void mld_v2_dispatch_general_query(struct mld_ifsoftc *);
134 static struct mbuf *
135 mld_v2_encap_report(struct ifnet *, struct mbuf *);
136 static int mld_v2_enqueue_filter_change(struct mbufq *,
137 struct in6_multi *);
138 static int mld_v2_enqueue_group_record(struct mbufq *,
139 struct in6_multi *, const int, const int, const int,
140 const int);
141 static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
142 struct mbuf *, struct mldv2_query *, const int, const int);
143 static int mld_v2_merge_state_changes(struct in6_multi *,
144 struct mbufq *);
145 static void mld_v2_process_group_timers(struct in6_multi_head *,
146 struct mbufq *, struct mbufq *,
147 struct in6_multi *, const int);
148 static int mld_v2_process_group_query(struct in6_multi *,
149 struct mld_ifsoftc *mli, int, struct mbuf *,
150 struct mldv2_query *, const int);
151 static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
152 static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
153
154 /*
155 * Normative references: RFC 2710, RFC 3590, RFC 3810.
156 *
157 * Locking:
158 * * The MLD subsystem lock ends up being system-wide for the moment,
159 * but could be per-VIMAGE later on.
160 * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
161 * Any may be taken independently; if any are held at the same
162 * time, the above lock order must be followed.
163 * * IN6_MULTI_LOCK covers in_multi.
164 * * MLD_LOCK covers per-link state and any global variables in this file.
165 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
166 * per-link state iterators.
167 *
168 * XXX LOR PREVENTION
169 * A special case for IPv6 is the in6_setscope() routine. ip6_output()
170 * will not accept an ifp; it wants an embedded scope ID, unlike
171 * ip_output(), which happily takes the ifp given to it. The embedded
172 * scope ID is only used by MLD to select the outgoing interface.
173 *
174 * During interface attach and detach, MLD will take MLD_LOCK *after*
175 * the IF_AFDATA_LOCK.
176 * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
177 * it with MLD_LOCK held without triggering an LOR. A netisr with indirect
178 * dispatch could work around this, but we'd rather not do that, as it
179 * can introduce other races.
180 *
181 * As such, we exploit the fact that the scope ID is just the interface
182 * index, and embed it in the IPv6 destination address accordingly.
183 * This is potentially NOT VALID for MLDv1 reports, as they
184 * are always sent to the multicast group itself; as MLDv2
185 * reports are always sent to ff02::16, this is not an issue
186 * when MLDv2 is in use.
187 *
188 * This does not however eliminate the LOR when ip6_output() itself
189 * calls in6_setscope() internally whilst MLD_LOCK is held. This will
190 * trigger a LOR warning in WITNESS when the ifnet is detached.
191 *
192 * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
193 * how it's used across the network stack. Here we're simply exploiting
194 * the fact that MLD runs at a similar layer in the stack to scope6.c.
195 *
196 * VIMAGE:
197 * * Each in6_multi corresponds to an ifp, and each ifp corresponds
198 * to a vnet in ifp->if_vnet.
199 */
200 static struct mtx mld_mtx;
201 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
202
203 #define MLD_EMBEDSCOPE(pin6, zoneid) \
204 if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \
205 IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \
206 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \
207
208 /*
209 * VIMAGE-wide globals.
210 */
211 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
212 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
213 VNET_DEFINE_STATIC(int, interface_timers_running6);
214 VNET_DEFINE_STATIC(int, state_change_timers_running6);
215 VNET_DEFINE_STATIC(int, current_state_timers_running6);
216
217 #define V_mld_gsrdelay VNET(mld_gsrdelay)
218 #define V_mli_head VNET(mli_head)
219 #define V_interface_timers_running6 VNET(interface_timers_running6)
220 #define V_state_change_timers_running6 VNET(state_change_timers_running6)
221 #define V_current_state_timers_running6 VNET(current_state_timers_running6)
222
223 SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */
224
225 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
226 "IPv6 Multicast Listener Discovery");
227
228 /*
229 * Virtualized sysctls.
230 */
231 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
232 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
233 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
234 "Rate limit for MLDv2 Group-and-Source queries in seconds");
235
236 /*
237 * Non-virtualized sysctls.
238 */
239 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
240 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
241 "Per-interface MLDv2 state");
242
243 static int mld_v1enable = 1;
244 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
245 &mld_v1enable, 0, "Enable fallback to MLDv1");
246
247 static int mld_v2enable = 1;
248 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN,
249 &mld_v2enable, 0, "Enable MLDv2");
250
251 static int mld_use_allow = 1;
252 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
253 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
254
255 /*
256 * Packed Router Alert option structure declaration.
257 */
258 struct mld_raopt {
259 struct ip6_hbh hbh;
260 struct ip6_opt pad;
261 struct ip6_opt_router ra;
262 } __packed;
263
264 /*
265 * Router Alert hop-by-hop option header.
266 */
267 static struct mld_raopt mld_ra = {
268 .hbh = { 0, 0 },
269 .pad = { .ip6o_type = IP6OPT_PADN, 0 },
270 .ra = {
271 .ip6or_type = IP6OPT_ROUTER_ALERT,
272 .ip6or_len = IP6OPT_RTALERT_LEN - 2,
273 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
274 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
275 }
276 };
277 static struct ip6_pktopts mld_po;
278
279 static __inline void
mld_save_context(struct mbuf * m,struct ifnet * ifp)280 mld_save_context(struct mbuf *m, struct ifnet *ifp)
281 {
282
283 #ifdef VIMAGE
284 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
285 #endif /* VIMAGE */
286 m->m_pkthdr.flowid = ifp->if_index;
287 }
288
289 static __inline void
mld_scrub_context(struct mbuf * m)290 mld_scrub_context(struct mbuf *m)
291 {
292
293 m->m_pkthdr.PH_loc.ptr = NULL;
294 m->m_pkthdr.flowid = 0;
295 }
296
297 /*
298 * Restore context from a queued output chain.
299 * Return saved ifindex.
300 *
301 * VIMAGE: The assertion is there to make sure that we
302 * actually called CURVNET_SET() with what's in the mbuf chain.
303 */
304 static __inline uint32_t
mld_restore_context(struct mbuf * m)305 mld_restore_context(struct mbuf *m)
306 {
307
308 #if defined(VIMAGE) && defined(INVARIANTS)
309 KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
310 ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
311 __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
312 #endif
313 return (m->m_pkthdr.flowid);
314 }
315
316 /*
317 * Retrieve or set threshold between group-source queries in seconds.
318 *
319 * VIMAGE: Assume curvnet set by caller.
320 * SMPng: NOTE: Serialized by MLD lock.
321 */
322 static int
sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)323 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
324 {
325 int error;
326 int i;
327
328 error = sysctl_wire_old_buffer(req, sizeof(int));
329 if (error)
330 return (error);
331
332 MLD_LOCK();
333
334 i = V_mld_gsrdelay.tv_sec;
335
336 error = sysctl_handle_int(oidp, &i, 0, req);
337 if (error || !req->newptr)
338 goto out_locked;
339
340 if (i < -1 || i >= 60) {
341 error = EINVAL;
342 goto out_locked;
343 }
344
345 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
346 V_mld_gsrdelay.tv_sec, i);
347 V_mld_gsrdelay.tv_sec = i;
348
349 out_locked:
350 MLD_UNLOCK();
351 return (error);
352 }
353
354 /*
355 * Expose struct mld_ifsoftc to userland, keyed by ifindex.
356 * For use by ifmcstat(8).
357 *
358 * SMPng: NOTE: Does an unlocked ifindex space read.
359 * VIMAGE: Assume curvnet set by caller. The node handler itself
360 * is not directly virtualized.
361 */
362 static int
sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)363 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
364 {
365 int *name;
366 int error;
367 u_int namelen;
368 struct ifnet *ifp;
369 struct mld_ifsoftc *mli;
370
371 name = (int *)arg1;
372 namelen = arg2;
373
374 if (req->newptr != NULL)
375 return (EPERM);
376
377 if (namelen != 1)
378 return (EINVAL);
379
380 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
381 if (error)
382 return (error);
383
384 IN6_MULTI_LOCK();
385 IN6_MULTI_LIST_LOCK();
386 MLD_LOCK();
387
388 if (name[0] <= 0 || name[0] > V_if_index) {
389 error = ENOENT;
390 goto out_locked;
391 }
392
393 error = ENOENT;
394
395 ifp = ifnet_byindex(name[0]);
396 if (ifp == NULL)
397 goto out_locked;
398
399 LIST_FOREACH(mli, &V_mli_head, mli_link) {
400 if (ifp == mli->mli_ifp) {
401 struct mld_ifinfo info;
402
403 info.mli_version = mli->mli_version;
404 info.mli_v1_timer = mli->mli_v1_timer;
405 info.mli_v2_timer = mli->mli_v2_timer;
406 info.mli_flags = mli->mli_flags;
407 info.mli_rv = mli->mli_rv;
408 info.mli_qi = mli->mli_qi;
409 info.mli_qri = mli->mli_qri;
410 info.mli_uri = mli->mli_uri;
411 error = SYSCTL_OUT(req, &info, sizeof(info));
412 break;
413 }
414 }
415
416 out_locked:
417 MLD_UNLOCK();
418 IN6_MULTI_LIST_UNLOCK();
419 IN6_MULTI_UNLOCK();
420 return (error);
421 }
422
423 /*
424 * Dispatch an entire queue of pending packet chains.
425 * VIMAGE: Assumes the vnet pointer has been set.
426 */
427 static void
mld_dispatch_queue(struct mbufq * mq,int limit)428 mld_dispatch_queue(struct mbufq *mq, int limit)
429 {
430 struct mbuf *m;
431
432 while ((m = mbufq_dequeue(mq)) != NULL) {
433 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
434 mld_dispatch_packet(m);
435 if (--limit == 0)
436 break;
437 }
438 }
439
440 /*
441 * Filter outgoing MLD report state by group.
442 *
443 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
444 * and node-local addresses. However, kernel and socket consumers
445 * always embed the KAME scope ID in the address provided, so strip it
446 * when performing comparison.
447 * Note: This is not the same as the *multicast* scope.
448 *
449 * Return zero if the given group is one for which MLD reports
450 * should be suppressed, or non-zero if reports should be issued.
451 */
452 static __inline int
mld_is_addr_reported(const struct in6_addr * addr)453 mld_is_addr_reported(const struct in6_addr *addr)
454 {
455
456 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
457
458 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
459 return (0);
460
461 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
462 struct in6_addr tmp = *addr;
463 in6_clearscope(&tmp);
464 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
465 return (0);
466 }
467
468 return (1);
469 }
470
471 /*
472 * Attach MLD when PF_INET6 is attached to an interface.
473 *
474 * SMPng: Normally called with IF_AFDATA_LOCK held.
475 */
476 struct mld_ifsoftc *
mld_domifattach(struct ifnet * ifp)477 mld_domifattach(struct ifnet *ifp)
478 {
479 struct mld_ifsoftc *mli;
480
481 CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
482 __func__, ifp, if_name(ifp));
483
484 MLD_LOCK();
485
486 mli = mli_alloc_locked(ifp);
487 if (!(ifp->if_flags & IFF_MULTICAST))
488 mli->mli_flags |= MLIF_SILENT;
489 if (mld_use_allow)
490 mli->mli_flags |= MLIF_USEALLOW;
491
492 MLD_UNLOCK();
493
494 return (mli);
495 }
496
497 /*
498 * VIMAGE: assume curvnet set by caller.
499 */
500 static struct mld_ifsoftc *
mli_alloc_locked(struct ifnet * ifp)501 mli_alloc_locked(/*const*/ struct ifnet *ifp)
502 {
503 struct mld_ifsoftc *mli;
504
505 MLD_LOCK_ASSERT();
506
507 mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
508 if (mli == NULL)
509 goto out;
510
511 mli->mli_ifp = ifp;
512 mli->mli_version = MLD_VERSION_2;
513 mli->mli_flags = 0;
514 mli->mli_rv = MLD_RV_INIT;
515 mli->mli_qi = MLD_QI_INIT;
516 mli->mli_qri = MLD_QRI_INIT;
517 mli->mli_uri = MLD_URI_INIT;
518 mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
519
520 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
521
522 CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
523 ifp, if_name(ifp));
524
525 out:
526 return (mli);
527 }
528
529 /*
530 * Hook for ifdetach.
531 *
532 * NOTE: Some finalization tasks need to run before the protocol domain
533 * is detached, but also before the link layer does its cleanup.
534 * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
535 *
536 * SMPng: Caller must hold IN6_MULTI_LOCK().
537 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
538 * XXX This routine is also bitten by unlocked ifma_protospec access.
539 */
540 void
mld_ifdetach(struct ifnet * ifp,struct in6_multi_head * inmh)541 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
542 {
543 struct epoch_tracker et;
544 struct mld_ifsoftc *mli;
545 struct ifmultiaddr *ifma;
546 struct in6_multi *inm;
547
548 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
549 if_name(ifp));
550
551 IN6_MULTI_LIST_LOCK_ASSERT();
552 MLD_LOCK();
553
554 mli = MLD_IFINFO(ifp);
555 IF_ADDR_WLOCK(ifp);
556 /*
557 * Extract list of in6_multi associated with the detaching ifp
558 * which the PF_INET6 layer is about to release.
559 */
560 NET_EPOCH_ENTER_ET(et);
561 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
562 inm = in6m_ifmultiaddr_get_inm(ifma);
563 if (inm == NULL)
564 continue;
565 in6m_disconnect_locked(inmh, inm);
566
567 if (mli->mli_version == MLD_VERSION_2) {
568 in6m_clear_recorded(inm);
569
570 /*
571 * We need to release the final reference held
572 * for issuing the INCLUDE {}.
573 */
574 if (inm->in6m_state == MLD_LEAVING_MEMBER) {
575 inm->in6m_state = MLD_NOT_MEMBER;
576 in6m_rele_locked(inmh, inm);
577 }
578 }
579 }
580 NET_EPOCH_EXIT_ET(et);
581 IF_ADDR_WUNLOCK(ifp);
582 MLD_UNLOCK();
583 }
584
585 /*
586 * Hook for domifdetach.
587 * Runs after link-layer cleanup; free MLD state.
588 *
589 * SMPng: Normally called with IF_AFDATA_LOCK held.
590 */
591 void
mld_domifdetach(struct ifnet * ifp)592 mld_domifdetach(struct ifnet *ifp)
593 {
594
595 CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
596 __func__, ifp, if_name(ifp));
597
598 MLD_LOCK();
599 mli_delete_locked(ifp);
600 MLD_UNLOCK();
601 }
602
603 static void
mli_delete_locked(const struct ifnet * ifp)604 mli_delete_locked(const struct ifnet *ifp)
605 {
606 struct mld_ifsoftc *mli, *tmli;
607
608 CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
609 __func__, ifp, if_name(ifp));
610
611 MLD_LOCK_ASSERT();
612
613 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
614 if (mli->mli_ifp == ifp) {
615 /*
616 * Free deferred General Query responses.
617 */
618 mbufq_drain(&mli->mli_gq);
619
620 LIST_REMOVE(mli, mli_link);
621
622 free(mli, M_MLD);
623 return;
624 }
625 }
626 }
627
628 /*
629 * Process a received MLDv1 general or address-specific query.
630 * Assumes that the query header has been pulled up to sizeof(mld_hdr).
631 *
632 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
633 * mld_addr. This is OK as we own the mbuf chain.
634 */
635 static int
mld_v1_input_query(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mld_hdr * mld)636 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
637 /*const*/ struct mld_hdr *mld)
638 {
639 struct ifmultiaddr *ifma;
640 struct mld_ifsoftc *mli;
641 struct in6_multi *inm;
642 int is_general_query;
643 uint16_t timer;
644 #ifdef KTR
645 char ip6tbuf[INET6_ADDRSTRLEN];
646 #endif
647
648 is_general_query = 0;
649
650 if (!mld_v1enable) {
651 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
652 ip6_sprintf(ip6tbuf, &mld->mld_addr),
653 ifp, if_name(ifp));
654 return (0);
655 }
656
657 /*
658 * RFC3810 Section 6.2: MLD queries must originate from
659 * a router's link-local address.
660 */
661 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
662 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
663 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
664 ifp, if_name(ifp));
665 return (0);
666 }
667
668 /*
669 * Do address field validation upfront before we accept
670 * the query.
671 */
672 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
673 /*
674 * MLDv1 General Query.
675 * If this was not sent to the all-nodes group, ignore it.
676 */
677 struct in6_addr dst;
678
679 dst = ip6->ip6_dst;
680 in6_clearscope(&dst);
681 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
682 return (EINVAL);
683 is_general_query = 1;
684 } else {
685 /*
686 * Embed scope ID of receiving interface in MLD query for
687 * lookup whilst we don't hold other locks.
688 */
689 in6_setscope(&mld->mld_addr, ifp, NULL);
690 }
691
692 IN6_MULTI_LIST_LOCK();
693 MLD_LOCK();
694
695 /*
696 * Switch to MLDv1 host compatibility mode.
697 */
698 mli = MLD_IFINFO(ifp);
699 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
700 mld_set_version(mli, MLD_VERSION_1);
701
702 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
703 if (timer == 0)
704 timer = 1;
705
706 IF_ADDR_RLOCK(ifp);
707 if (is_general_query) {
708 /*
709 * For each reporting group joined on this
710 * interface, kick the report timer.
711 */
712 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
713 ifp, if_name(ifp));
714 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
715 inm = in6m_ifmultiaddr_get_inm(ifma);
716 if (inm == NULL)
717 continue;
718 mld_v1_update_group(inm, timer);
719 }
720 } else {
721 /*
722 * MLDv1 Group-Specific Query.
723 * If this is a group-specific MLDv1 query, we need only
724 * look up the single group to process it.
725 */
726 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
727 if (inm != NULL) {
728 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
729 ip6_sprintf(ip6tbuf, &mld->mld_addr),
730 ifp, if_name(ifp));
731 mld_v1_update_group(inm, timer);
732 }
733 /* XXX Clear embedded scope ID as userland won't expect it. */
734 in6_clearscope(&mld->mld_addr);
735 }
736
737 IF_ADDR_RUNLOCK(ifp);
738 MLD_UNLOCK();
739 IN6_MULTI_LIST_UNLOCK();
740
741 return (0);
742 }
743
744 /*
745 * Update the report timer on a group in response to an MLDv1 query.
746 *
747 * If we are becoming the reporting member for this group, start the timer.
748 * If we already are the reporting member for this group, and timer is
749 * below the threshold, reset it.
750 *
751 * We may be updating the group for the first time since we switched
752 * to MLDv2. If we are, then we must clear any recorded source lists,
753 * and transition to REPORTING state; the group timer is overloaded
754 * for group and group-source query responses.
755 *
756 * Unlike MLDv2, the delay per group should be jittered
757 * to avoid bursts of MLDv1 reports.
758 */
759 static void
mld_v1_update_group(struct in6_multi * inm,const int timer)760 mld_v1_update_group(struct in6_multi *inm, const int timer)
761 {
762 #ifdef KTR
763 char ip6tbuf[INET6_ADDRSTRLEN];
764 #endif
765
766 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
767 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
768 if_name(inm->in6m_ifp), timer);
769
770 IN6_MULTI_LIST_LOCK_ASSERT();
771
772 switch (inm->in6m_state) {
773 case MLD_NOT_MEMBER:
774 case MLD_SILENT_MEMBER:
775 break;
776 case MLD_REPORTING_MEMBER:
777 if (inm->in6m_timer != 0 &&
778 inm->in6m_timer <= timer) {
779 CTR1(KTR_MLD, "%s: REPORTING and timer running, "
780 "skipping.", __func__);
781 break;
782 }
783 /* FALLTHROUGH */
784 case MLD_SG_QUERY_PENDING_MEMBER:
785 case MLD_G_QUERY_PENDING_MEMBER:
786 case MLD_IDLE_MEMBER:
787 case MLD_LAZY_MEMBER:
788 case MLD_AWAKENING_MEMBER:
789 CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
790 inm->in6m_state = MLD_REPORTING_MEMBER;
791 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
792 V_current_state_timers_running6 = 1;
793 break;
794 case MLD_SLEEPING_MEMBER:
795 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
796 inm->in6m_state = MLD_AWAKENING_MEMBER;
797 break;
798 case MLD_LEAVING_MEMBER:
799 break;
800 }
801 }
802
803 /*
804 * Process a received MLDv2 general, group-specific or
805 * group-and-source-specific query.
806 *
807 * Assumes that mld points to a struct mldv2_query which is stored in
808 * contiguous memory.
809 *
810 * Return 0 if successful, otherwise an appropriate error code is returned.
811 */
812 static int
mld_v2_input_query(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mbuf * m,struct mldv2_query * mld,const int off,const int icmp6len)813 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
814 struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
815 {
816 struct mld_ifsoftc *mli;
817 struct in6_multi *inm;
818 uint32_t maxdelay, nsrc, qqi;
819 int is_general_query;
820 uint16_t timer;
821 uint8_t qrv;
822 #ifdef KTR
823 char ip6tbuf[INET6_ADDRSTRLEN];
824 #endif
825
826 if (!mld_v2enable) {
827 CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
828 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
829 ifp, if_name(ifp));
830 return (0);
831 }
832
833 /*
834 * RFC3810 Section 6.2: MLD queries must originate from
835 * a router's link-local address.
836 */
837 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
838 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
839 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
840 ifp, if_name(ifp));
841 return (0);
842 }
843
844 is_general_query = 0;
845
846 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
847
848 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */
849 if (maxdelay >= 32768) {
850 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
851 (MLD_MRC_EXP(maxdelay) + 3);
852 }
853 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
854 if (timer == 0)
855 timer = 1;
856
857 qrv = MLD_QRV(mld->mld_misc);
858 if (qrv < 2) {
859 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
860 qrv, MLD_RV_INIT);
861 qrv = MLD_RV_INIT;
862 }
863
864 qqi = mld->mld_qqi;
865 if (qqi >= 128) {
866 qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
867 (MLD_QQIC_EXP(mld->mld_qqi) + 3);
868 }
869
870 nsrc = ntohs(mld->mld_numsrc);
871 if (nsrc > MLD_MAX_GS_SOURCES)
872 return (EMSGSIZE);
873 if (icmp6len < sizeof(struct mldv2_query) +
874 (nsrc * sizeof(struct in6_addr)))
875 return (EMSGSIZE);
876
877 /*
878 * Do further input validation upfront to avoid resetting timers
879 * should we need to discard this query.
880 */
881 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
882 /*
883 * A general query with a source list has undefined
884 * behaviour; discard it.
885 */
886 if (nsrc > 0)
887 return (EINVAL);
888 is_general_query = 1;
889 } else {
890 /*
891 * Embed scope ID of receiving interface in MLD query for
892 * lookup whilst we don't hold other locks (due to KAME
893 * locking lameness). We own this mbuf chain just now.
894 */
895 in6_setscope(&mld->mld_addr, ifp, NULL);
896 }
897
898 IN6_MULTI_LIST_LOCK();
899 MLD_LOCK();
900
901 mli = MLD_IFINFO(ifp);
902 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
903
904 /*
905 * Discard the v2 query if we're in Compatibility Mode.
906 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
907 * until the Old Version Querier Present timer expires.
908 */
909 if (mli->mli_version != MLD_VERSION_2)
910 goto out_locked;
911
912 mld_set_version(mli, MLD_VERSION_2);
913 mli->mli_rv = qrv;
914 mli->mli_qi = qqi;
915 mli->mli_qri = maxdelay;
916
917 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
918 maxdelay);
919
920 if (is_general_query) {
921 /*
922 * MLDv2 General Query.
923 *
924 * Schedule a current-state report on this ifp for
925 * all groups, possibly containing source lists.
926 *
927 * If there is a pending General Query response
928 * scheduled earlier than the selected delay, do
929 * not schedule any other reports.
930 * Otherwise, reset the interface timer.
931 */
932 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
933 ifp, if_name(ifp));
934 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
935 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
936 V_interface_timers_running6 = 1;
937 }
938 } else {
939 /*
940 * MLDv2 Group-specific or Group-and-source-specific Query.
941 *
942 * Group-source-specific queries are throttled on
943 * a per-group basis to defeat denial-of-service attempts.
944 * Queries for groups we are not a member of on this
945 * link are simply ignored.
946 */
947 IF_ADDR_RLOCK(ifp);
948 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
949 if (inm == NULL) {
950 IF_ADDR_RUNLOCK(ifp);
951 goto out_locked;
952 }
953 if (nsrc > 0) {
954 if (!ratecheck(&inm->in6m_lastgsrtv,
955 &V_mld_gsrdelay)) {
956 CTR1(KTR_MLD, "%s: GS query throttled.",
957 __func__);
958 IF_ADDR_RUNLOCK(ifp);
959 goto out_locked;
960 }
961 }
962 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
963 ifp, if_name(ifp));
964 /*
965 * If there is a pending General Query response
966 * scheduled sooner than the selected delay, no
967 * further report need be scheduled.
968 * Otherwise, prepare to respond to the
969 * group-specific or group-and-source query.
970 */
971 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
972 mld_v2_process_group_query(inm, mli, timer, m, mld, off);
973
974 /* XXX Clear embedded scope ID as userland won't expect it. */
975 in6_clearscope(&mld->mld_addr);
976 IF_ADDR_RUNLOCK(ifp);
977 }
978
979 out_locked:
980 MLD_UNLOCK();
981 IN6_MULTI_LIST_UNLOCK();
982
983 return (0);
984 }
985
986 /*
987 * Process a received MLDv2 group-specific or group-and-source-specific
988 * query.
989 * Return <0 if any error occurred. Currently this is ignored.
990 */
991 static int
mld_v2_process_group_query(struct in6_multi * inm,struct mld_ifsoftc * mli,int timer,struct mbuf * m0,struct mldv2_query * mld,const int off)992 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
993 int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
994 {
995 int retval;
996 uint16_t nsrc;
997
998 IN6_MULTI_LIST_LOCK_ASSERT();
999 MLD_LOCK_ASSERT();
1000
1001 retval = 0;
1002
1003 switch (inm->in6m_state) {
1004 case MLD_NOT_MEMBER:
1005 case MLD_SILENT_MEMBER:
1006 case MLD_SLEEPING_MEMBER:
1007 case MLD_LAZY_MEMBER:
1008 case MLD_AWAKENING_MEMBER:
1009 case MLD_IDLE_MEMBER:
1010 case MLD_LEAVING_MEMBER:
1011 return (retval);
1012 break;
1013 case MLD_REPORTING_MEMBER:
1014 case MLD_G_QUERY_PENDING_MEMBER:
1015 case MLD_SG_QUERY_PENDING_MEMBER:
1016 break;
1017 }
1018
1019 nsrc = ntohs(mld->mld_numsrc);
1020
1021 /* Length should be checked by calling function. */
1022 KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
1023 m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
1024 nsrc * sizeof(struct in6_addr),
1025 ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
1026 m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
1027 nsrc * sizeof(struct in6_addr), m0));
1028
1029
1030 /*
1031 * Deal with group-specific queries upfront.
1032 * If any group query is already pending, purge any recorded
1033 * source-list state if it exists, and schedule a query response
1034 * for this group-specific query.
1035 */
1036 if (nsrc == 0) {
1037 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1038 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1039 in6m_clear_recorded(inm);
1040 timer = min(inm->in6m_timer, timer);
1041 }
1042 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1043 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1044 V_current_state_timers_running6 = 1;
1045 return (retval);
1046 }
1047
1048 /*
1049 * Deal with the case where a group-and-source-specific query has
1050 * been received but a group-specific query is already pending.
1051 */
1052 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1053 timer = min(inm->in6m_timer, timer);
1054 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1055 V_current_state_timers_running6 = 1;
1056 return (retval);
1057 }
1058
1059 /*
1060 * Finally, deal with the case where a group-and-source-specific
1061 * query has been received, where a response to a previous g-s-r
1062 * query exists, or none exists.
1063 * In this case, we need to parse the source-list which the Querier
1064 * has provided us with and check if we have any source list filter
1065 * entries at T1 for these sources. If we do not, there is no need
1066 * schedule a report and the query may be dropped.
1067 * If we do, we must record them and schedule a current-state
1068 * report for those sources.
1069 */
1070 if (inm->in6m_nsrc > 0) {
1071 struct in6_addr srcaddr;
1072 int i, nrecorded;
1073 int soff;
1074
1075 soff = off + sizeof(struct mldv2_query);
1076 nrecorded = 0;
1077 for (i = 0; i < nsrc; i++) {
1078 m_copydata(m0, soff, sizeof(struct in6_addr),
1079 (caddr_t)&srcaddr);
1080 retval = in6m_record_source(inm, &srcaddr);
1081 if (retval < 0)
1082 break;
1083 nrecorded += retval;
1084 soff += sizeof(struct in6_addr);
1085 }
1086 if (nrecorded > 0) {
1087 CTR1(KTR_MLD,
1088 "%s: schedule response to SG query", __func__);
1089 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1090 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1091 V_current_state_timers_running6 = 1;
1092 }
1093 }
1094
1095 return (retval);
1096 }
1097
1098 /*
1099 * Process a received MLDv1 host membership report.
1100 * Assumes mld points to mld_hdr in pulled up mbuf chain.
1101 *
1102 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1103 * mld_addr. This is OK as we own the mbuf chain.
1104 */
1105 static int
mld_v1_input_report(struct ifnet * ifp,const struct ip6_hdr * ip6,struct mld_hdr * mld)1106 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1107 /*const*/ struct mld_hdr *mld)
1108 {
1109 struct in6_addr src, dst;
1110 struct in6_ifaddr *ia;
1111 struct in6_multi *inm;
1112 #ifdef KTR
1113 char ip6tbuf[INET6_ADDRSTRLEN];
1114 #endif
1115
1116 if (!mld_v1enable) {
1117 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1118 ip6_sprintf(ip6tbuf, &mld->mld_addr),
1119 ifp, if_name(ifp));
1120 return (0);
1121 }
1122
1123 if (ifp->if_flags & IFF_LOOPBACK)
1124 return (0);
1125
1126 /*
1127 * MLDv1 reports must originate from a host's link-local address,
1128 * or the unspecified address (when booting).
1129 */
1130 src = ip6->ip6_src;
1131 in6_clearscope(&src);
1132 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1133 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1134 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1135 ifp, if_name(ifp));
1136 return (EINVAL);
1137 }
1138
1139 /*
1140 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1141 * group, and must be directed to the group itself.
1142 */
1143 dst = ip6->ip6_dst;
1144 in6_clearscope(&dst);
1145 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1146 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1147 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1148 ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1149 ifp, if_name(ifp));
1150 return (EINVAL);
1151 }
1152
1153 /*
1154 * Make sure we don't hear our own membership report, as fast
1155 * leave requires knowing that we are the only member of a
1156 * group. Assume we used the link-local address if available,
1157 * otherwise look for ::.
1158 *
1159 * XXX Note that scope ID comparison is needed for the address
1160 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1161 * performed for the on-wire address.
1162 */
1163 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1164 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1165 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1166 if (ia != NULL)
1167 ifa_free(&ia->ia_ifa);
1168 return (0);
1169 }
1170 if (ia != NULL)
1171 ifa_free(&ia->ia_ifa);
1172
1173 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1174 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1175
1176 /*
1177 * Embed scope ID of receiving interface in MLD query for lookup
1178 * whilst we don't hold other locks (due to KAME locking lameness).
1179 */
1180 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1181 in6_setscope(&mld->mld_addr, ifp, NULL);
1182
1183 IN6_MULTI_LIST_LOCK();
1184 MLD_LOCK();
1185 IF_ADDR_RLOCK(ifp);
1186
1187 /*
1188 * MLDv1 report suppression.
1189 * If we are a member of this group, and our membership should be
1190 * reported, and our group timer is pending or about to be reset,
1191 * stop our group timer by transitioning to the 'lazy' state.
1192 */
1193 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1194 if (inm != NULL) {
1195 struct mld_ifsoftc *mli;
1196
1197 mli = inm->in6m_mli;
1198 KASSERT(mli != NULL,
1199 ("%s: no mli for ifp %p", __func__, ifp));
1200
1201 /*
1202 * If we are in MLDv2 host mode, do not allow the
1203 * other host's MLDv1 report to suppress our reports.
1204 */
1205 if (mli->mli_version == MLD_VERSION_2)
1206 goto out_locked;
1207
1208 inm->in6m_timer = 0;
1209
1210 switch (inm->in6m_state) {
1211 case MLD_NOT_MEMBER:
1212 case MLD_SILENT_MEMBER:
1213 case MLD_SLEEPING_MEMBER:
1214 break;
1215 case MLD_REPORTING_MEMBER:
1216 case MLD_IDLE_MEMBER:
1217 case MLD_AWAKENING_MEMBER:
1218 CTR3(KTR_MLD,
1219 "report suppressed for %s on ifp %p(%s)",
1220 ip6_sprintf(ip6tbuf, &mld->mld_addr),
1221 ifp, if_name(ifp));
1222 case MLD_LAZY_MEMBER:
1223 inm->in6m_state = MLD_LAZY_MEMBER;
1224 break;
1225 case MLD_G_QUERY_PENDING_MEMBER:
1226 case MLD_SG_QUERY_PENDING_MEMBER:
1227 case MLD_LEAVING_MEMBER:
1228 break;
1229 }
1230 }
1231
1232 out_locked:
1233 IF_ADDR_RUNLOCK(ifp);
1234 MLD_UNLOCK();
1235 IN6_MULTI_LIST_UNLOCK();
1236
1237 /* XXX Clear embedded scope ID as userland won't expect it. */
1238 in6_clearscope(&mld->mld_addr);
1239
1240 return (0);
1241 }
1242
1243 /*
1244 * MLD input path.
1245 *
1246 * Assume query messages which fit in a single ICMPv6 message header
1247 * have been pulled up.
1248 * Assume that userland will want to see the message, even if it
1249 * otherwise fails kernel input validation; do not free it.
1250 * Pullup may however free the mbuf chain m if it fails.
1251 *
1252 * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1253 */
1254 int
mld_input(struct mbuf * m,int off,int icmp6len)1255 mld_input(struct mbuf *m, int off, int icmp6len)
1256 {
1257 struct ifnet *ifp;
1258 struct ip6_hdr *ip6;
1259 struct mld_hdr *mld;
1260 int mldlen;
1261
1262 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1263
1264 ifp = m->m_pkthdr.rcvif;
1265
1266 ip6 = mtod(m, struct ip6_hdr *);
1267
1268 /* Pullup to appropriate size. */
1269 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1270 if (mld->mld_type == MLD_LISTENER_QUERY &&
1271 icmp6len >= sizeof(struct mldv2_query)) {
1272 mldlen = sizeof(struct mldv2_query);
1273 } else {
1274 mldlen = sizeof(struct mld_hdr);
1275 }
1276 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1277 if (mld == NULL) {
1278 ICMP6STAT_INC(icp6s_badlen);
1279 return (IPPROTO_DONE);
1280 }
1281
1282 /*
1283 * Userland needs to see all of this traffic for implementing
1284 * the endpoint discovery portion of multicast routing.
1285 */
1286 switch (mld->mld_type) {
1287 case MLD_LISTENER_QUERY:
1288 icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1289 if (icmp6len == sizeof(struct mld_hdr)) {
1290 if (mld_v1_input_query(ifp, ip6, mld) != 0)
1291 return (0);
1292 } else if (icmp6len >= sizeof(struct mldv2_query)) {
1293 if (mld_v2_input_query(ifp, ip6, m,
1294 (struct mldv2_query *)mld, off, icmp6len) != 0)
1295 return (0);
1296 }
1297 break;
1298 case MLD_LISTENER_REPORT:
1299 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1300 if (mld_v1_input_report(ifp, ip6, mld) != 0)
1301 return (0);
1302 break;
1303 case MLDV2_LISTENER_REPORT:
1304 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1305 break;
1306 case MLD_LISTENER_DONE:
1307 icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1308 break;
1309 default:
1310 break;
1311 }
1312
1313 return (0);
1314 }
1315
1316 /*
1317 * Fast timeout handler (global).
1318 * VIMAGE: Timeout handlers are expected to service all vimages.
1319 */
1320 void
mld_fasttimo(void)1321 mld_fasttimo(void)
1322 {
1323 struct in6_multi_head inmh;
1324 VNET_ITERATOR_DECL(vnet_iter);
1325
1326 SLIST_INIT(&inmh);
1327
1328 VNET_LIST_RLOCK_NOSLEEP();
1329 VNET_FOREACH(vnet_iter) {
1330 CURVNET_SET(vnet_iter);
1331 mld_fasttimo_vnet(&inmh);
1332 CURVNET_RESTORE();
1333 }
1334 VNET_LIST_RUNLOCK_NOSLEEP();
1335 in6m_release_list_deferred(&inmh);
1336 }
1337
1338 /*
1339 * Fast timeout handler (per-vnet).
1340 *
1341 * VIMAGE: Assume caller has set up our curvnet.
1342 */
1343 static void
mld_fasttimo_vnet(struct in6_multi_head * inmh)1344 mld_fasttimo_vnet(struct in6_multi_head *inmh)
1345 {
1346 struct epoch_tracker et;
1347 struct mbufq scq; /* State-change packets */
1348 struct mbufq qrq; /* Query response packets */
1349 struct ifnet *ifp;
1350 struct mld_ifsoftc *mli;
1351 struct ifmultiaddr *ifma;
1352 struct in6_multi *inm;
1353 int uri_fasthz;
1354
1355 uri_fasthz = 0;
1356
1357 /*
1358 * Quick check to see if any work needs to be done, in order to
1359 * minimize the overhead of fasttimo processing.
1360 * SMPng: XXX Unlocked reads.
1361 */
1362 if (!V_current_state_timers_running6 &&
1363 !V_interface_timers_running6 &&
1364 !V_state_change_timers_running6)
1365 return;
1366
1367 IN6_MULTI_LIST_LOCK();
1368 MLD_LOCK();
1369
1370 /*
1371 * MLDv2 General Query response timer processing.
1372 */
1373 if (V_interface_timers_running6) {
1374 CTR1(KTR_MLD, "%s: interface timers running", __func__);
1375
1376 V_interface_timers_running6 = 0;
1377 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1378 if (mli->mli_v2_timer == 0) {
1379 /* Do nothing. */
1380 } else if (--mli->mli_v2_timer == 0) {
1381 mld_v2_dispatch_general_query(mli);
1382 } else {
1383 V_interface_timers_running6 = 1;
1384 }
1385 }
1386 }
1387
1388 if (!V_current_state_timers_running6 &&
1389 !V_state_change_timers_running6)
1390 goto out_locked;
1391
1392 V_current_state_timers_running6 = 0;
1393 V_state_change_timers_running6 = 0;
1394
1395 CTR1(KTR_MLD, "%s: state change timers running", __func__);
1396
1397 /*
1398 * MLD host report and state-change timer processing.
1399 * Note: Processing a v2 group timer may remove a node.
1400 */
1401 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1402 ifp = mli->mli_ifp;
1403
1404 if (mli->mli_version == MLD_VERSION_2) {
1405 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1406 PR_FASTHZ);
1407 mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1408 mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1409 }
1410
1411 IF_ADDR_WLOCK(ifp);
1412 NET_EPOCH_ENTER_ET(et);
1413 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1414 inm = in6m_ifmultiaddr_get_inm(ifma);
1415 if (inm == NULL)
1416 continue;
1417 switch (mli->mli_version) {
1418 case MLD_VERSION_1:
1419 mld_v1_process_group_timer(inmh, inm);
1420 break;
1421 case MLD_VERSION_2:
1422 mld_v2_process_group_timers(inmh, &qrq,
1423 &scq, inm, uri_fasthz);
1424 break;
1425 }
1426 }
1427 IF_ADDR_WUNLOCK(ifp);
1428
1429 switch (mli->mli_version) {
1430 case MLD_VERSION_1:
1431 /*
1432 * Transmit reports for this lifecycle. This
1433 * is done while not holding IF_ADDR_LOCK
1434 * since this can call
1435 * in6ifa_ifpforlinklocal() which locks
1436 * IF_ADDR_LOCK internally as well as
1437 * ip6_output() to transmit a packet.
1438 */
1439 while ((inm = SLIST_FIRST(inmh)) != NULL) {
1440 SLIST_REMOVE_HEAD(inmh, in6m_defer);
1441 (void)mld_v1_transmit_report(inm,
1442 MLD_LISTENER_REPORT);
1443 }
1444 break;
1445 case MLD_VERSION_2:
1446 mld_dispatch_queue(&qrq, 0);
1447 mld_dispatch_queue(&scq, 0);
1448 break;
1449 }
1450 NET_EPOCH_EXIT_ET(et);
1451 }
1452
1453 out_locked:
1454 MLD_UNLOCK();
1455 IN6_MULTI_LIST_UNLOCK();
1456 }
1457
1458 /*
1459 * Update host report group timer.
1460 * Will update the global pending timer flags.
1461 */
1462 static void
mld_v1_process_group_timer(struct in6_multi_head * inmh,struct in6_multi * inm)1463 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1464 {
1465 int report_timer_expired;
1466
1467 IN6_MULTI_LIST_LOCK_ASSERT();
1468 MLD_LOCK_ASSERT();
1469
1470 if (inm->in6m_timer == 0) {
1471 report_timer_expired = 0;
1472 } else if (--inm->in6m_timer == 0) {
1473 report_timer_expired = 1;
1474 } else {
1475 V_current_state_timers_running6 = 1;
1476 return;
1477 }
1478
1479 switch (inm->in6m_state) {
1480 case MLD_NOT_MEMBER:
1481 case MLD_SILENT_MEMBER:
1482 case MLD_IDLE_MEMBER:
1483 case MLD_LAZY_MEMBER:
1484 case MLD_SLEEPING_MEMBER:
1485 case MLD_AWAKENING_MEMBER:
1486 break;
1487 case MLD_REPORTING_MEMBER:
1488 if (report_timer_expired) {
1489 inm->in6m_state = MLD_IDLE_MEMBER;
1490 SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
1491 }
1492 break;
1493 case MLD_G_QUERY_PENDING_MEMBER:
1494 case MLD_SG_QUERY_PENDING_MEMBER:
1495 case MLD_LEAVING_MEMBER:
1496 break;
1497 }
1498 }
1499
1500 /*
1501 * Update a group's timers for MLDv2.
1502 * Will update the global pending timer flags.
1503 * Note: Unlocked read from mli.
1504 */
1505 static void
mld_v2_process_group_timers(struct in6_multi_head * inmh,struct mbufq * qrq,struct mbufq * scq,struct in6_multi * inm,const int uri_fasthz)1506 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1507 struct mbufq *qrq, struct mbufq *scq,
1508 struct in6_multi *inm, const int uri_fasthz)
1509 {
1510 int query_response_timer_expired;
1511 int state_change_retransmit_timer_expired;
1512 #ifdef KTR
1513 char ip6tbuf[INET6_ADDRSTRLEN];
1514 #endif
1515
1516 IN6_MULTI_LIST_LOCK_ASSERT();
1517 MLD_LOCK_ASSERT();
1518
1519 query_response_timer_expired = 0;
1520 state_change_retransmit_timer_expired = 0;
1521
1522 /*
1523 * During a transition from compatibility mode back to MLDv2,
1524 * a group record in REPORTING state may still have its group
1525 * timer active. This is a no-op in this function; it is easier
1526 * to deal with it here than to complicate the slow-timeout path.
1527 */
1528 if (inm->in6m_timer == 0) {
1529 query_response_timer_expired = 0;
1530 } else if (--inm->in6m_timer == 0) {
1531 query_response_timer_expired = 1;
1532 } else {
1533 V_current_state_timers_running6 = 1;
1534 }
1535
1536 if (inm->in6m_sctimer == 0) {
1537 state_change_retransmit_timer_expired = 0;
1538 } else if (--inm->in6m_sctimer == 0) {
1539 state_change_retransmit_timer_expired = 1;
1540 } else {
1541 V_state_change_timers_running6 = 1;
1542 }
1543
1544 /* We are in fasttimo, so be quick about it. */
1545 if (!state_change_retransmit_timer_expired &&
1546 !query_response_timer_expired)
1547 return;
1548
1549 switch (inm->in6m_state) {
1550 case MLD_NOT_MEMBER:
1551 case MLD_SILENT_MEMBER:
1552 case MLD_SLEEPING_MEMBER:
1553 case MLD_LAZY_MEMBER:
1554 case MLD_AWAKENING_MEMBER:
1555 case MLD_IDLE_MEMBER:
1556 break;
1557 case MLD_G_QUERY_PENDING_MEMBER:
1558 case MLD_SG_QUERY_PENDING_MEMBER:
1559 /*
1560 * Respond to a previously pending Group-Specific
1561 * or Group-and-Source-Specific query by enqueueing
1562 * the appropriate Current-State report for
1563 * immediate transmission.
1564 */
1565 if (query_response_timer_expired) {
1566 int retval;
1567
1568 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1569 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1570 0);
1571 CTR2(KTR_MLD, "%s: enqueue record = %d",
1572 __func__, retval);
1573 inm->in6m_state = MLD_REPORTING_MEMBER;
1574 in6m_clear_recorded(inm);
1575 }
1576 /* FALLTHROUGH */
1577 case MLD_REPORTING_MEMBER:
1578 case MLD_LEAVING_MEMBER:
1579 if (state_change_retransmit_timer_expired) {
1580 /*
1581 * State-change retransmission timer fired.
1582 * If there are any further pending retransmissions,
1583 * set the global pending state-change flag, and
1584 * reset the timer.
1585 */
1586 if (--inm->in6m_scrv > 0) {
1587 inm->in6m_sctimer = uri_fasthz;
1588 V_state_change_timers_running6 = 1;
1589 }
1590 /*
1591 * Retransmit the previously computed state-change
1592 * report. If there are no further pending
1593 * retransmissions, the mbuf queue will be consumed.
1594 * Update T0 state to T1 as we have now sent
1595 * a state-change.
1596 */
1597 (void)mld_v2_merge_state_changes(inm, scq);
1598
1599 in6m_commit(inm);
1600 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1601 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1602 if_name(inm->in6m_ifp));
1603
1604 /*
1605 * If we are leaving the group for good, make sure
1606 * we release MLD's reference to it.
1607 * This release must be deferred using a SLIST,
1608 * as we are called from a loop which traverses
1609 * the in_ifmultiaddr TAILQ.
1610 */
1611 if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1612 inm->in6m_scrv == 0) {
1613 inm->in6m_state = MLD_NOT_MEMBER;
1614 in6m_disconnect_locked(inmh, inm);
1615 in6m_rele_locked(inmh, inm);
1616 }
1617 }
1618 break;
1619 }
1620 }
1621
1622 /*
1623 * Switch to a different version on the given interface,
1624 * as per Section 9.12.
1625 */
1626 static void
mld_set_version(struct mld_ifsoftc * mli,const int version)1627 mld_set_version(struct mld_ifsoftc *mli, const int version)
1628 {
1629 int old_version_timer;
1630
1631 MLD_LOCK_ASSERT();
1632
1633 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1634 version, mli->mli_ifp, if_name(mli->mli_ifp));
1635
1636 if (version == MLD_VERSION_1) {
1637 /*
1638 * Compute the "Older Version Querier Present" timer as per
1639 * Section 9.12.
1640 */
1641 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1642 old_version_timer *= PR_SLOWHZ;
1643 mli->mli_v1_timer = old_version_timer;
1644 }
1645
1646 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1647 mli->mli_version = MLD_VERSION_1;
1648 mld_v2_cancel_link_timers(mli);
1649 }
1650 }
1651
1652 /*
1653 * Cancel pending MLDv2 timers for the given link and all groups
1654 * joined on it; state-change, general-query, and group-query timers.
1655 */
1656 static void
mld_v2_cancel_link_timers(struct mld_ifsoftc * mli)1657 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1658 {
1659 struct epoch_tracker et;
1660 struct in6_multi_head inmh;
1661 struct ifmultiaddr *ifma;
1662 struct ifnet *ifp;
1663 struct in6_multi *inm;
1664
1665 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1666 mli->mli_ifp, if_name(mli->mli_ifp));
1667
1668 SLIST_INIT(&inmh);
1669 IN6_MULTI_LIST_LOCK_ASSERT();
1670 MLD_LOCK_ASSERT();
1671
1672 /*
1673 * Fast-track this potentially expensive operation
1674 * by checking all the global 'timer pending' flags.
1675 */
1676 if (!V_interface_timers_running6 &&
1677 !V_state_change_timers_running6 &&
1678 !V_current_state_timers_running6)
1679 return;
1680
1681 mli->mli_v2_timer = 0;
1682
1683 ifp = mli->mli_ifp;
1684
1685 IF_ADDR_WLOCK(ifp);
1686 NET_EPOCH_ENTER_ET(et);
1687 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1688 inm = in6m_ifmultiaddr_get_inm(ifma);
1689 if (inm == NULL)
1690 continue;
1691 switch (inm->in6m_state) {
1692 case MLD_NOT_MEMBER:
1693 case MLD_SILENT_MEMBER:
1694 case MLD_IDLE_MEMBER:
1695 case MLD_LAZY_MEMBER:
1696 case MLD_SLEEPING_MEMBER:
1697 case MLD_AWAKENING_MEMBER:
1698 break;
1699 case MLD_LEAVING_MEMBER:
1700 /*
1701 * If we are leaving the group and switching
1702 * version, we need to release the final
1703 * reference held for issuing the INCLUDE {}.
1704 */
1705 if (inm->in6m_refcount == 1)
1706 in6m_disconnect_locked(&inmh, inm);
1707 in6m_rele_locked(&inmh, inm);
1708 /* FALLTHROUGH */
1709 case MLD_G_QUERY_PENDING_MEMBER:
1710 case MLD_SG_QUERY_PENDING_MEMBER:
1711 in6m_clear_recorded(inm);
1712 /* FALLTHROUGH */
1713 case MLD_REPORTING_MEMBER:
1714 inm->in6m_sctimer = 0;
1715 inm->in6m_timer = 0;
1716 inm->in6m_state = MLD_REPORTING_MEMBER;
1717 /*
1718 * Free any pending MLDv2 state-change records.
1719 */
1720 mbufq_drain(&inm->in6m_scq);
1721 break;
1722 }
1723 }
1724 NET_EPOCH_EXIT_ET(et);
1725 IF_ADDR_WUNLOCK(ifp);
1726 in6m_release_list_deferred(&inmh);
1727 }
1728
1729 /*
1730 * Global slowtimo handler.
1731 * VIMAGE: Timeout handlers are expected to service all vimages.
1732 */
1733 void
mld_slowtimo(void)1734 mld_slowtimo(void)
1735 {
1736 VNET_ITERATOR_DECL(vnet_iter);
1737
1738 VNET_LIST_RLOCK_NOSLEEP();
1739 VNET_FOREACH(vnet_iter) {
1740 CURVNET_SET(vnet_iter);
1741 mld_slowtimo_vnet();
1742 CURVNET_RESTORE();
1743 }
1744 VNET_LIST_RUNLOCK_NOSLEEP();
1745 }
1746
1747 /*
1748 * Per-vnet slowtimo handler.
1749 */
1750 static void
mld_slowtimo_vnet(void)1751 mld_slowtimo_vnet(void)
1752 {
1753 struct mld_ifsoftc *mli;
1754
1755 MLD_LOCK();
1756
1757 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1758 mld_v1_process_querier_timers(mli);
1759 }
1760
1761 MLD_UNLOCK();
1762 }
1763
1764 /*
1765 * Update the Older Version Querier Present timers for a link.
1766 * See Section 9.12 of RFC 3810.
1767 */
1768 static void
mld_v1_process_querier_timers(struct mld_ifsoftc * mli)1769 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1770 {
1771
1772 MLD_LOCK_ASSERT();
1773
1774 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1775 /*
1776 * MLDv1 Querier Present timer expired; revert to MLDv2.
1777 */
1778 CTR5(KTR_MLD,
1779 "%s: transition from v%d -> v%d on %p(%s)",
1780 __func__, mli->mli_version, MLD_VERSION_2,
1781 mli->mli_ifp, if_name(mli->mli_ifp));
1782 mli->mli_version = MLD_VERSION_2;
1783 }
1784 }
1785
1786 /*
1787 * Transmit an MLDv1 report immediately.
1788 */
1789 static int
mld_v1_transmit_report(struct in6_multi * in6m,const int type)1790 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1791 {
1792 struct ifnet *ifp;
1793 struct in6_ifaddr *ia;
1794 struct ip6_hdr *ip6;
1795 struct mbuf *mh, *md;
1796 struct mld_hdr *mld;
1797
1798 IN6_MULTI_LIST_LOCK_ASSERT();
1799 MLD_LOCK_ASSERT();
1800
1801 ifp = in6m->in6m_ifp;
1802 /* in process of being freed */
1803 if (ifp == NULL)
1804 return (0);
1805 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1806 /* ia may be NULL if link-local address is tentative. */
1807
1808 mh = m_gethdr(M_NOWAIT, MT_DATA);
1809 if (mh == NULL) {
1810 if (ia != NULL)
1811 ifa_free(&ia->ia_ifa);
1812 return (ENOMEM);
1813 }
1814 md = m_get(M_NOWAIT, MT_DATA);
1815 if (md == NULL) {
1816 m_free(mh);
1817 if (ia != NULL)
1818 ifa_free(&ia->ia_ifa);
1819 return (ENOMEM);
1820 }
1821 mh->m_next = md;
1822
1823 /*
1824 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1825 * that ether_output() does not need to allocate another mbuf
1826 * for the header in the most common case.
1827 */
1828 M_ALIGN(mh, sizeof(struct ip6_hdr));
1829 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1830 mh->m_len = sizeof(struct ip6_hdr);
1831
1832 ip6 = mtod(mh, struct ip6_hdr *);
1833 ip6->ip6_flow = 0;
1834 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1835 ip6->ip6_vfc |= IPV6_VERSION;
1836 ip6->ip6_nxt = IPPROTO_ICMPV6;
1837 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1838 ip6->ip6_dst = in6m->in6m_addr;
1839
1840 md->m_len = sizeof(struct mld_hdr);
1841 mld = mtod(md, struct mld_hdr *);
1842 mld->mld_type = type;
1843 mld->mld_code = 0;
1844 mld->mld_cksum = 0;
1845 mld->mld_maxdelay = 0;
1846 mld->mld_reserved = 0;
1847 mld->mld_addr = in6m->in6m_addr;
1848 in6_clearscope(&mld->mld_addr);
1849 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1850 sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1851
1852 mld_save_context(mh, ifp);
1853 mh->m_flags |= M_MLDV1;
1854
1855 mld_dispatch_packet(mh);
1856
1857 if (ia != NULL)
1858 ifa_free(&ia->ia_ifa);
1859 return (0);
1860 }
1861
1862 /*
1863 * Process a state change from the upper layer for the given IPv6 group.
1864 *
1865 * Each socket holds a reference on the in_multi in its own ip_moptions.
1866 * The socket layer will have made the necessary updates to.the group
1867 * state, it is now up to MLD to issue a state change report if there
1868 * has been any change between T0 (when the last state-change was issued)
1869 * and T1 (now).
1870 *
1871 * We use the MLDv2 state machine at group level. The MLd module
1872 * however makes the decision as to which MLD protocol version to speak.
1873 * A state change *from* INCLUDE {} always means an initial join.
1874 * A state change *to* INCLUDE {} always means a final leave.
1875 *
1876 * If delay is non-zero, and the state change is an initial multicast
1877 * join, the state change report will be delayed by 'delay' ticks
1878 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1879 * the initial MLDv2 state change report will be delayed by whichever
1880 * is sooner, a pending state-change timer or delay itself.
1881 *
1882 * VIMAGE: curvnet should have been set by caller, as this routine
1883 * is called from the socket option handlers.
1884 */
1885 int
mld_change_state(struct in6_multi * inm,const int delay)1886 mld_change_state(struct in6_multi *inm, const int delay)
1887 {
1888 struct mld_ifsoftc *mli;
1889 struct ifnet *ifp;
1890 int error;
1891
1892 IN6_MULTI_LIST_LOCK_ASSERT();
1893
1894 error = 0;
1895
1896 /*
1897 * Check if the in6_multi has already been disconnected.
1898 */
1899 if (inm->in6m_ifp == NULL) {
1900 CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
1901 return (0);
1902 }
1903
1904 /*
1905 * Try to detect if the upper layer just asked us to change state
1906 * for an interface which has now gone away.
1907 */
1908 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1909 ifp = inm->in6m_ifma->ifma_ifp;
1910 if (ifp == NULL)
1911 return (0);
1912 /*
1913 * Sanity check that netinet6's notion of ifp is the
1914 * same as net's.
1915 */
1916 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1917
1918 MLD_LOCK();
1919 mli = MLD_IFINFO(ifp);
1920 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1921
1922 /*
1923 * If we detect a state transition to or from MCAST_UNDEFINED
1924 * for this group, then we are starting or finishing an MLD
1925 * life cycle for this group.
1926 */
1927 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1928 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1929 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1930 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1931 CTR1(KTR_MLD, "%s: initial join", __func__);
1932 error = mld_initial_join(inm, mli, delay);
1933 goto out_locked;
1934 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1935 CTR1(KTR_MLD, "%s: final leave", __func__);
1936 mld_final_leave(inm, mli);
1937 goto out_locked;
1938 }
1939 } else {
1940 CTR1(KTR_MLD, "%s: filter set change", __func__);
1941 }
1942
1943 error = mld_handle_state_change(inm, mli);
1944
1945 out_locked:
1946 MLD_UNLOCK();
1947 return (error);
1948 }
1949
1950 /*
1951 * Perform the initial join for an MLD group.
1952 *
1953 * When joining a group:
1954 * If the group should have its MLD traffic suppressed, do nothing.
1955 * MLDv1 starts sending MLDv1 host membership reports.
1956 * MLDv2 will schedule an MLDv2 state-change report containing the
1957 * initial state of the membership.
1958 *
1959 * If the delay argument is non-zero, then we must delay sending the
1960 * initial state change for delay ticks (in units of PR_FASTHZ).
1961 */
1962 static int
mld_initial_join(struct in6_multi * inm,struct mld_ifsoftc * mli,const int delay)1963 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1964 const int delay)
1965 {
1966 struct ifnet *ifp;
1967 struct mbufq *mq;
1968 int error, retval, syncstates;
1969 int odelay;
1970 #ifdef KTR
1971 char ip6tbuf[INET6_ADDRSTRLEN];
1972 #endif
1973
1974 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1975 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1976 inm->in6m_ifp, if_name(inm->in6m_ifp));
1977
1978 error = 0;
1979 syncstates = 1;
1980
1981 ifp = inm->in6m_ifp;
1982
1983 IN6_MULTI_LIST_LOCK_ASSERT();
1984 MLD_LOCK_ASSERT();
1985
1986 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1987
1988 /*
1989 * Groups joined on loopback or marked as 'not reported',
1990 * enter the MLD_SILENT_MEMBER state and
1991 * are never reported in any protocol exchanges.
1992 * All other groups enter the appropriate state machine
1993 * for the version in use on this link.
1994 * A link marked as MLIF_SILENT causes MLD to be completely
1995 * disabled for the link.
1996 */
1997 if ((ifp->if_flags & IFF_LOOPBACK) ||
1998 (mli->mli_flags & MLIF_SILENT) ||
1999 !mld_is_addr_reported(&inm->in6m_addr)) {
2000 CTR1(KTR_MLD,
2001 "%s: not kicking state machine for silent group", __func__);
2002 inm->in6m_state = MLD_SILENT_MEMBER;
2003 inm->in6m_timer = 0;
2004 } else {
2005 /*
2006 * Deal with overlapping in_multi lifecycle.
2007 * If this group was LEAVING, then make sure
2008 * we drop the reference we picked up to keep the
2009 * group around for the final INCLUDE {} enqueue.
2010 */
2011 if (mli->mli_version == MLD_VERSION_2 &&
2012 inm->in6m_state == MLD_LEAVING_MEMBER) {
2013 inm->in6m_refcount--;
2014 MPASS(inm->in6m_refcount > 0);
2015 }
2016 inm->in6m_state = MLD_REPORTING_MEMBER;
2017
2018 switch (mli->mli_version) {
2019 case MLD_VERSION_1:
2020 /*
2021 * If a delay was provided, only use it if
2022 * it is greater than the delay normally
2023 * used for an MLDv1 state change report,
2024 * and delay sending the initial MLDv1 report
2025 * by not transitioning to the IDLE state.
2026 */
2027 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2028 if (delay) {
2029 inm->in6m_timer = max(delay, odelay);
2030 V_current_state_timers_running6 = 1;
2031 } else {
2032 inm->in6m_state = MLD_IDLE_MEMBER;
2033 error = mld_v1_transmit_report(inm,
2034 MLD_LISTENER_REPORT);
2035 if (error == 0) {
2036 inm->in6m_timer = odelay;
2037 V_current_state_timers_running6 = 1;
2038 }
2039 }
2040 break;
2041
2042 case MLD_VERSION_2:
2043 /*
2044 * Defer update of T0 to T1, until the first copy
2045 * of the state change has been transmitted.
2046 */
2047 syncstates = 0;
2048
2049 /*
2050 * Immediately enqueue a State-Change Report for
2051 * this interface, freeing any previous reports.
2052 * Don't kick the timers if there is nothing to do,
2053 * or if an error occurred.
2054 */
2055 mq = &inm->in6m_scq;
2056 mbufq_drain(mq);
2057 retval = mld_v2_enqueue_group_record(mq, inm, 1,
2058 0, 0, (mli->mli_flags & MLIF_USEALLOW));
2059 CTR2(KTR_MLD, "%s: enqueue record = %d",
2060 __func__, retval);
2061 if (retval <= 0) {
2062 error = retval * -1;
2063 break;
2064 }
2065
2066 /*
2067 * Schedule transmission of pending state-change
2068 * report up to RV times for this link. The timer
2069 * will fire at the next mld_fasttimo (~200ms),
2070 * giving us an opportunity to merge the reports.
2071 *
2072 * If a delay was provided to this function, only
2073 * use this delay if sooner than the existing one.
2074 */
2075 KASSERT(mli->mli_rv > 1,
2076 ("%s: invalid robustness %d", __func__,
2077 mli->mli_rv));
2078 inm->in6m_scrv = mli->mli_rv;
2079 if (delay) {
2080 if (inm->in6m_sctimer > 1) {
2081 inm->in6m_sctimer =
2082 min(inm->in6m_sctimer, delay);
2083 } else
2084 inm->in6m_sctimer = delay;
2085 } else
2086 inm->in6m_sctimer = 1;
2087 V_state_change_timers_running6 = 1;
2088
2089 error = 0;
2090 break;
2091 }
2092 }
2093
2094 /*
2095 * Only update the T0 state if state change is atomic,
2096 * i.e. we don't need to wait for a timer to fire before we
2097 * can consider the state change to have been communicated.
2098 */
2099 if (syncstates) {
2100 in6m_commit(inm);
2101 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2102 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2103 if_name(inm->in6m_ifp));
2104 }
2105
2106 return (error);
2107 }
2108
2109 /*
2110 * Issue an intermediate state change during the life-cycle.
2111 */
2112 static int
mld_handle_state_change(struct in6_multi * inm,struct mld_ifsoftc * mli)2113 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2114 {
2115 struct ifnet *ifp;
2116 int retval;
2117 #ifdef KTR
2118 char ip6tbuf[INET6_ADDRSTRLEN];
2119 #endif
2120
2121 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2122 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2123 inm->in6m_ifp, if_name(inm->in6m_ifp));
2124
2125 ifp = inm->in6m_ifp;
2126
2127 IN6_MULTI_LIST_LOCK_ASSERT();
2128 MLD_LOCK_ASSERT();
2129
2130 KASSERT(mli && mli->mli_ifp == ifp,
2131 ("%s: inconsistent ifp", __func__));
2132
2133 if ((ifp->if_flags & IFF_LOOPBACK) ||
2134 (mli->mli_flags & MLIF_SILENT) ||
2135 !mld_is_addr_reported(&inm->in6m_addr) ||
2136 (mli->mli_version != MLD_VERSION_2)) {
2137 if (!mld_is_addr_reported(&inm->in6m_addr)) {
2138 CTR1(KTR_MLD,
2139 "%s: not kicking state machine for silent group", __func__);
2140 }
2141 CTR1(KTR_MLD, "%s: nothing to do", __func__);
2142 in6m_commit(inm);
2143 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2144 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2145 if_name(inm->in6m_ifp));
2146 return (0);
2147 }
2148
2149 mbufq_drain(&inm->in6m_scq);
2150
2151 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2152 (mli->mli_flags & MLIF_USEALLOW));
2153 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2154 if (retval <= 0)
2155 return (-retval);
2156
2157 /*
2158 * If record(s) were enqueued, start the state-change
2159 * report timer for this group.
2160 */
2161 inm->in6m_scrv = mli->mli_rv;
2162 inm->in6m_sctimer = 1;
2163 V_state_change_timers_running6 = 1;
2164
2165 return (0);
2166 }
2167
2168 /*
2169 * Perform the final leave for a multicast address.
2170 *
2171 * When leaving a group:
2172 * MLDv1 sends a DONE message, if and only if we are the reporter.
2173 * MLDv2 enqueues a state-change report containing a transition
2174 * to INCLUDE {} for immediate transmission.
2175 */
2176 static void
mld_final_leave(struct in6_multi * inm,struct mld_ifsoftc * mli)2177 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2178 {
2179 int syncstates;
2180 #ifdef KTR
2181 char ip6tbuf[INET6_ADDRSTRLEN];
2182 #endif
2183
2184 syncstates = 1;
2185
2186 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2187 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2188 inm->in6m_ifp, if_name(inm->in6m_ifp));
2189
2190 IN6_MULTI_LIST_LOCK_ASSERT();
2191 MLD_LOCK_ASSERT();
2192
2193 switch (inm->in6m_state) {
2194 case MLD_NOT_MEMBER:
2195 case MLD_SILENT_MEMBER:
2196 case MLD_LEAVING_MEMBER:
2197 /* Already leaving or left; do nothing. */
2198 CTR1(KTR_MLD,
2199 "%s: not kicking state machine for silent group", __func__);
2200 break;
2201 case MLD_REPORTING_MEMBER:
2202 case MLD_IDLE_MEMBER:
2203 case MLD_G_QUERY_PENDING_MEMBER:
2204 case MLD_SG_QUERY_PENDING_MEMBER:
2205 if (mli->mli_version == MLD_VERSION_1) {
2206 #ifdef INVARIANTS
2207 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2208 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2209 panic("%s: MLDv2 state reached, not MLDv2 mode",
2210 __func__);
2211 #endif
2212 mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2213 inm->in6m_state = MLD_NOT_MEMBER;
2214 V_current_state_timers_running6 = 1;
2215 } else if (mli->mli_version == MLD_VERSION_2) {
2216 /*
2217 * Stop group timer and all pending reports.
2218 * Immediately enqueue a state-change report
2219 * TO_IN {} to be sent on the next fast timeout,
2220 * giving us an opportunity to merge reports.
2221 */
2222 mbufq_drain(&inm->in6m_scq);
2223 inm->in6m_timer = 0;
2224 inm->in6m_scrv = mli->mli_rv;
2225 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2226 "pending retransmissions.", __func__,
2227 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2228 if_name(inm->in6m_ifp), inm->in6m_scrv);
2229 if (inm->in6m_scrv == 0) {
2230 inm->in6m_state = MLD_NOT_MEMBER;
2231 inm->in6m_sctimer = 0;
2232 } else {
2233 int retval;
2234
2235 in6m_acquire_locked(inm);
2236
2237 retval = mld_v2_enqueue_group_record(
2238 &inm->in6m_scq, inm, 1, 0, 0,
2239 (mli->mli_flags & MLIF_USEALLOW));
2240 KASSERT(retval != 0,
2241 ("%s: enqueue record = %d", __func__,
2242 retval));
2243
2244 inm->in6m_state = MLD_LEAVING_MEMBER;
2245 inm->in6m_sctimer = 1;
2246 V_state_change_timers_running6 = 1;
2247 syncstates = 0;
2248 }
2249 break;
2250 }
2251 break;
2252 case MLD_LAZY_MEMBER:
2253 case MLD_SLEEPING_MEMBER:
2254 case MLD_AWAKENING_MEMBER:
2255 /* Our reports are suppressed; do nothing. */
2256 break;
2257 }
2258
2259 if (syncstates) {
2260 in6m_commit(inm);
2261 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2262 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2263 if_name(inm->in6m_ifp));
2264 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2265 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2266 __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2267 }
2268 }
2269
2270 /*
2271 * Enqueue an MLDv2 group record to the given output queue.
2272 *
2273 * If is_state_change is zero, a current-state record is appended.
2274 * If is_state_change is non-zero, a state-change report is appended.
2275 *
2276 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2277 * If is_group_query is zero, and if there is a packet with free space
2278 * at the tail of the queue, it will be appended to providing there
2279 * is enough free space.
2280 * Otherwise a new mbuf packet chain is allocated.
2281 *
2282 * If is_source_query is non-zero, each source is checked to see if
2283 * it was recorded for a Group-Source query, and will be omitted if
2284 * it is not both in-mode and recorded.
2285 *
2286 * If use_block_allow is non-zero, state change reports for initial join
2287 * and final leave, on an inclusive mode group with a source list, will be
2288 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2289 *
2290 * The function will attempt to allocate leading space in the packet
2291 * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2292 *
2293 * If successful the size of all data appended to the queue is returned,
2294 * otherwise an error code less than zero is returned, or zero if
2295 * no record(s) were appended.
2296 */
2297 static int
mld_v2_enqueue_group_record(struct mbufq * mq,struct in6_multi * inm,const int is_state_change,const int is_group_query,const int is_source_query,const int use_block_allow)2298 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2299 const int is_state_change, const int is_group_query,
2300 const int is_source_query, const int use_block_allow)
2301 {
2302 struct mldv2_record mr;
2303 struct mldv2_record *pmr;
2304 struct ifnet *ifp;
2305 struct ip6_msource *ims, *nims;
2306 struct mbuf *m0, *m, *md;
2307 int is_filter_list_change;
2308 int minrec0len, m0srcs, msrcs, nbytes, off;
2309 int record_has_sources;
2310 int now;
2311 int type;
2312 uint8_t mode;
2313 #ifdef KTR
2314 char ip6tbuf[INET6_ADDRSTRLEN];
2315 #endif
2316
2317 IN6_MULTI_LIST_LOCK_ASSERT();
2318
2319 ifp = inm->in6m_ifp;
2320 is_filter_list_change = 0;
2321 m = NULL;
2322 m0 = NULL;
2323 m0srcs = 0;
2324 msrcs = 0;
2325 nbytes = 0;
2326 nims = NULL;
2327 record_has_sources = 1;
2328 pmr = NULL;
2329 type = MLD_DO_NOTHING;
2330 mode = inm->in6m_st[1].iss_fmode;
2331
2332 /*
2333 * If we did not transition out of ASM mode during t0->t1,
2334 * and there are no source nodes to process, we can skip
2335 * the generation of source records.
2336 */
2337 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2338 inm->in6m_nsrc == 0)
2339 record_has_sources = 0;
2340
2341 if (is_state_change) {
2342 /*
2343 * Queue a state change record.
2344 * If the mode did not change, and there are non-ASM
2345 * listeners or source filters present,
2346 * we potentially need to issue two records for the group.
2347 * If there are ASM listeners, and there was no filter
2348 * mode transition of any kind, do nothing.
2349 *
2350 * If we are transitioning to MCAST_UNDEFINED, we need
2351 * not send any sources. A transition to/from this state is
2352 * considered inclusive with some special treatment.
2353 *
2354 * If we are rewriting initial joins/leaves to use
2355 * ALLOW/BLOCK, and the group's membership is inclusive,
2356 * we need to send sources in all cases.
2357 */
2358 if (mode != inm->in6m_st[0].iss_fmode) {
2359 if (mode == MCAST_EXCLUDE) {
2360 CTR1(KTR_MLD, "%s: change to EXCLUDE",
2361 __func__);
2362 type = MLD_CHANGE_TO_EXCLUDE_MODE;
2363 } else {
2364 CTR1(KTR_MLD, "%s: change to INCLUDE",
2365 __func__);
2366 if (use_block_allow) {
2367 /*
2368 * XXX
2369 * Here we're interested in state
2370 * edges either direction between
2371 * MCAST_UNDEFINED and MCAST_INCLUDE.
2372 * Perhaps we should just check
2373 * the group state, rather than
2374 * the filter mode.
2375 */
2376 if (mode == MCAST_UNDEFINED) {
2377 type = MLD_BLOCK_OLD_SOURCES;
2378 } else {
2379 type = MLD_ALLOW_NEW_SOURCES;
2380 }
2381 } else {
2382 type = MLD_CHANGE_TO_INCLUDE_MODE;
2383 if (mode == MCAST_UNDEFINED)
2384 record_has_sources = 0;
2385 }
2386 }
2387 } else {
2388 if (record_has_sources) {
2389 is_filter_list_change = 1;
2390 } else {
2391 type = MLD_DO_NOTHING;
2392 }
2393 }
2394 } else {
2395 /*
2396 * Queue a current state record.
2397 */
2398 if (mode == MCAST_EXCLUDE) {
2399 type = MLD_MODE_IS_EXCLUDE;
2400 } else if (mode == MCAST_INCLUDE) {
2401 type = MLD_MODE_IS_INCLUDE;
2402 KASSERT(inm->in6m_st[1].iss_asm == 0,
2403 ("%s: inm %p is INCLUDE but ASM count is %d",
2404 __func__, inm, inm->in6m_st[1].iss_asm));
2405 }
2406 }
2407
2408 /*
2409 * Generate the filter list changes using a separate function.
2410 */
2411 if (is_filter_list_change)
2412 return (mld_v2_enqueue_filter_change(mq, inm));
2413
2414 if (type == MLD_DO_NOTHING) {
2415 CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2416 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2417 if_name(inm->in6m_ifp));
2418 return (0);
2419 }
2420
2421 /*
2422 * If any sources are present, we must be able to fit at least
2423 * one in the trailing space of the tail packet's mbuf,
2424 * ideally more.
2425 */
2426 minrec0len = sizeof(struct mldv2_record);
2427 if (record_has_sources)
2428 minrec0len += sizeof(struct in6_addr);
2429
2430 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2431 mld_rec_type_to_str(type),
2432 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2433 if_name(inm->in6m_ifp));
2434
2435 /*
2436 * Check if we have a packet in the tail of the queue for this
2437 * group into which the first group record for this group will fit.
2438 * Otherwise allocate a new packet.
2439 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2440 * Note: Group records for G/GSR query responses MUST be sent
2441 * in their own packet.
2442 */
2443 m0 = mbufq_last(mq);
2444 if (!is_group_query &&
2445 m0 != NULL &&
2446 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2447 (m0->m_pkthdr.len + minrec0len) <
2448 (ifp->if_mtu - MLD_MTUSPACE)) {
2449 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2450 sizeof(struct mldv2_record)) /
2451 sizeof(struct in6_addr);
2452 m = m0;
2453 CTR1(KTR_MLD, "%s: use existing packet", __func__);
2454 } else {
2455 if (mbufq_full(mq)) {
2456 CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2457 return (-ENOMEM);
2458 }
2459 m = NULL;
2460 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2461 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2462 if (!is_state_change && !is_group_query)
2463 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2464 if (m == NULL)
2465 m = m_gethdr(M_NOWAIT, MT_DATA);
2466 if (m == NULL)
2467 return (-ENOMEM);
2468
2469 mld_save_context(m, ifp);
2470
2471 CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2472 }
2473
2474 /*
2475 * Append group record.
2476 * If we have sources, we don't know how many yet.
2477 */
2478 mr.mr_type = type;
2479 mr.mr_datalen = 0;
2480 mr.mr_numsrc = 0;
2481 mr.mr_addr = inm->in6m_addr;
2482 in6_clearscope(&mr.mr_addr);
2483 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2484 if (m != m0)
2485 m_freem(m);
2486 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2487 return (-ENOMEM);
2488 }
2489 nbytes += sizeof(struct mldv2_record);
2490
2491 /*
2492 * Append as many sources as will fit in the first packet.
2493 * If we are appending to a new packet, the chain allocation
2494 * may potentially use clusters; use m_getptr() in this case.
2495 * If we are appending to an existing packet, we need to obtain
2496 * a pointer to the group record after m_append(), in case a new
2497 * mbuf was allocated.
2498 *
2499 * Only append sources which are in-mode at t1. If we are
2500 * transitioning to MCAST_UNDEFINED state on the group, and
2501 * use_block_allow is zero, do not include source entries.
2502 * Otherwise, we need to include this source in the report.
2503 *
2504 * Only report recorded sources in our filter set when responding
2505 * to a group-source query.
2506 */
2507 if (record_has_sources) {
2508 if (m == m0) {
2509 md = m_last(m);
2510 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2511 md->m_len - nbytes);
2512 } else {
2513 md = m_getptr(m, 0, &off);
2514 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2515 off);
2516 }
2517 msrcs = 0;
2518 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2519 nims) {
2520 CTR2(KTR_MLD, "%s: visit node %s", __func__,
2521 ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2522 now = im6s_get_mode(inm, ims, 1);
2523 CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2524 if ((now != mode) ||
2525 (now == mode &&
2526 (!use_block_allow && mode == MCAST_UNDEFINED))) {
2527 CTR1(KTR_MLD, "%s: skip node", __func__);
2528 continue;
2529 }
2530 if (is_source_query && ims->im6s_stp == 0) {
2531 CTR1(KTR_MLD, "%s: skip unrecorded node",
2532 __func__);
2533 continue;
2534 }
2535 CTR1(KTR_MLD, "%s: append node", __func__);
2536 if (!m_append(m, sizeof(struct in6_addr),
2537 (void *)&ims->im6s_addr)) {
2538 if (m != m0)
2539 m_freem(m);
2540 CTR1(KTR_MLD, "%s: m_append() failed.",
2541 __func__);
2542 return (-ENOMEM);
2543 }
2544 nbytes += sizeof(struct in6_addr);
2545 ++msrcs;
2546 if (msrcs == m0srcs)
2547 break;
2548 }
2549 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2550 msrcs);
2551 pmr->mr_numsrc = htons(msrcs);
2552 nbytes += (msrcs * sizeof(struct in6_addr));
2553 }
2554
2555 if (is_source_query && msrcs == 0) {
2556 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2557 if (m != m0)
2558 m_freem(m);
2559 return (0);
2560 }
2561
2562 /*
2563 * We are good to go with first packet.
2564 */
2565 if (m != m0) {
2566 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2567 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2568 mbufq_enqueue(mq, m);
2569 } else
2570 m->m_pkthdr.PH_vt.vt_nrecs++;
2571
2572 /*
2573 * No further work needed if no source list in packet(s).
2574 */
2575 if (!record_has_sources)
2576 return (nbytes);
2577
2578 /*
2579 * Whilst sources remain to be announced, we need to allocate
2580 * a new packet and fill out as many sources as will fit.
2581 * Always try for a cluster first.
2582 */
2583 while (nims != NULL) {
2584 if (mbufq_full(mq)) {
2585 CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2586 return (-ENOMEM);
2587 }
2588 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2589 if (m == NULL)
2590 m = m_gethdr(M_NOWAIT, MT_DATA);
2591 if (m == NULL)
2592 return (-ENOMEM);
2593 mld_save_context(m, ifp);
2594 md = m_getptr(m, 0, &off);
2595 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2596 CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2597
2598 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2599 if (m != m0)
2600 m_freem(m);
2601 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2602 return (-ENOMEM);
2603 }
2604 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2605 nbytes += sizeof(struct mldv2_record);
2606
2607 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2608 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2609
2610 msrcs = 0;
2611 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2612 CTR2(KTR_MLD, "%s: visit node %s",
2613 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2614 now = im6s_get_mode(inm, ims, 1);
2615 if ((now != mode) ||
2616 (now == mode &&
2617 (!use_block_allow && mode == MCAST_UNDEFINED))) {
2618 CTR1(KTR_MLD, "%s: skip node", __func__);
2619 continue;
2620 }
2621 if (is_source_query && ims->im6s_stp == 0) {
2622 CTR1(KTR_MLD, "%s: skip unrecorded node",
2623 __func__);
2624 continue;
2625 }
2626 CTR1(KTR_MLD, "%s: append node", __func__);
2627 if (!m_append(m, sizeof(struct in6_addr),
2628 (void *)&ims->im6s_addr)) {
2629 if (m != m0)
2630 m_freem(m);
2631 CTR1(KTR_MLD, "%s: m_append() failed.",
2632 __func__);
2633 return (-ENOMEM);
2634 }
2635 ++msrcs;
2636 if (msrcs == m0srcs)
2637 break;
2638 }
2639 pmr->mr_numsrc = htons(msrcs);
2640 nbytes += (msrcs * sizeof(struct in6_addr));
2641
2642 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2643 mbufq_enqueue(mq, m);
2644 }
2645
2646 return (nbytes);
2647 }
2648
2649 /*
2650 * Type used to mark record pass completion.
2651 * We exploit the fact we can cast to this easily from the
2652 * current filter modes on each ip_msource node.
2653 */
2654 typedef enum {
2655 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2656 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2657 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2658 REC_FULL = REC_ALLOW | REC_BLOCK
2659 } rectype_t;
2660
2661 /*
2662 * Enqueue an MLDv2 filter list change to the given output queue.
2663 *
2664 * Source list filter state is held in an RB-tree. When the filter list
2665 * for a group is changed without changing its mode, we need to compute
2666 * the deltas between T0 and T1 for each source in the filter set,
2667 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2668 *
2669 * As we may potentially queue two record types, and the entire R-B tree
2670 * needs to be walked at once, we break this out into its own function
2671 * so we can generate a tightly packed queue of packets.
2672 *
2673 * XXX This could be written to only use one tree walk, although that makes
2674 * serializing into the mbuf chains a bit harder. For now we do two walks
2675 * which makes things easier on us, and it may or may not be harder on
2676 * the L2 cache.
2677 *
2678 * If successful the size of all data appended to the queue is returned,
2679 * otherwise an error code less than zero is returned, or zero if
2680 * no record(s) were appended.
2681 */
2682 static int
mld_v2_enqueue_filter_change(struct mbufq * mq,struct in6_multi * inm)2683 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2684 {
2685 static const int MINRECLEN =
2686 sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2687 struct ifnet *ifp;
2688 struct mldv2_record mr;
2689 struct mldv2_record *pmr;
2690 struct ip6_msource *ims, *nims;
2691 struct mbuf *m, *m0, *md;
2692 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2693 int nallow, nblock;
2694 uint8_t mode, now, then;
2695 rectype_t crt, drt, nrt;
2696 #ifdef KTR
2697 char ip6tbuf[INET6_ADDRSTRLEN];
2698 #endif
2699
2700 IN6_MULTI_LIST_LOCK_ASSERT();
2701
2702 if (inm->in6m_nsrc == 0 ||
2703 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2704 return (0);
2705
2706 ifp = inm->in6m_ifp; /* interface */
2707 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */
2708 crt = REC_NONE; /* current group record type */
2709 drt = REC_NONE; /* mask of completed group record types */
2710 nrt = REC_NONE; /* record type for current node */
2711 m0srcs = 0; /* # source which will fit in current mbuf chain */
2712 npbytes = 0; /* # of bytes appended this packet */
2713 nbytes = 0; /* # of bytes appended to group's state-change queue */
2714 rsrcs = 0; /* # sources encoded in current record */
2715 schanged = 0; /* # nodes encoded in overall filter change */
2716 nallow = 0; /* # of source entries in ALLOW_NEW */
2717 nblock = 0; /* # of source entries in BLOCK_OLD */
2718 nims = NULL; /* next tree node pointer */
2719
2720 /*
2721 * For each possible filter record mode.
2722 * The first kind of source we encounter tells us which
2723 * is the first kind of record we start appending.
2724 * If a node transitioned to UNDEFINED at t1, its mode is treated
2725 * as the inverse of the group's filter mode.
2726 */
2727 while (drt != REC_FULL) {
2728 do {
2729 m0 = mbufq_last(mq);
2730 if (m0 != NULL &&
2731 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2732 MLD_V2_REPORT_MAXRECS) &&
2733 (m0->m_pkthdr.len + MINRECLEN) <
2734 (ifp->if_mtu - MLD_MTUSPACE)) {
2735 m = m0;
2736 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2737 sizeof(struct mldv2_record)) /
2738 sizeof(struct in6_addr);
2739 CTR1(KTR_MLD,
2740 "%s: use previous packet", __func__);
2741 } else {
2742 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2743 if (m == NULL)
2744 m = m_gethdr(M_NOWAIT, MT_DATA);
2745 if (m == NULL) {
2746 CTR1(KTR_MLD,
2747 "%s: m_get*() failed", __func__);
2748 return (-ENOMEM);
2749 }
2750 m->m_pkthdr.PH_vt.vt_nrecs = 0;
2751 mld_save_context(m, ifp);
2752 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2753 sizeof(struct mldv2_record)) /
2754 sizeof(struct in6_addr);
2755 npbytes = 0;
2756 CTR1(KTR_MLD,
2757 "%s: allocated new packet", __func__);
2758 }
2759 /*
2760 * Append the MLD group record header to the
2761 * current packet's data area.
2762 * Recalculate pointer to free space for next
2763 * group record, in case m_append() allocated
2764 * a new mbuf or cluster.
2765 */
2766 memset(&mr, 0, sizeof(mr));
2767 mr.mr_addr = inm->in6m_addr;
2768 in6_clearscope(&mr.mr_addr);
2769 if (!m_append(m, sizeof(mr), (void *)&mr)) {
2770 if (m != m0)
2771 m_freem(m);
2772 CTR1(KTR_MLD,
2773 "%s: m_append() failed", __func__);
2774 return (-ENOMEM);
2775 }
2776 npbytes += sizeof(struct mldv2_record);
2777 if (m != m0) {
2778 /* new packet; offset in chain */
2779 md = m_getptr(m, npbytes -
2780 sizeof(struct mldv2_record), &off);
2781 pmr = (struct mldv2_record *)(mtod(md,
2782 uint8_t *) + off);
2783 } else {
2784 /* current packet; offset from last append */
2785 md = m_last(m);
2786 pmr = (struct mldv2_record *)(mtod(md,
2787 uint8_t *) + md->m_len -
2788 sizeof(struct mldv2_record));
2789 }
2790 /*
2791 * Begin walking the tree for this record type
2792 * pass, or continue from where we left off
2793 * previously if we had to allocate a new packet.
2794 * Only report deltas in-mode at t1.
2795 * We need not report included sources as allowed
2796 * if we are in inclusive mode on the group,
2797 * however the converse is not true.
2798 */
2799 rsrcs = 0;
2800 if (nims == NULL) {
2801 nims = RB_MIN(ip6_msource_tree,
2802 &inm->in6m_srcs);
2803 }
2804 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2805 CTR2(KTR_MLD, "%s: visit node %s", __func__,
2806 ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2807 now = im6s_get_mode(inm, ims, 1);
2808 then = im6s_get_mode(inm, ims, 0);
2809 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2810 __func__, then, now);
2811 if (now == then) {
2812 CTR1(KTR_MLD,
2813 "%s: skip unchanged", __func__);
2814 continue;
2815 }
2816 if (mode == MCAST_EXCLUDE &&
2817 now == MCAST_INCLUDE) {
2818 CTR1(KTR_MLD,
2819 "%s: skip IN src on EX group",
2820 __func__);
2821 continue;
2822 }
2823 nrt = (rectype_t)now;
2824 if (nrt == REC_NONE)
2825 nrt = (rectype_t)(~mode & REC_FULL);
2826 if (schanged++ == 0) {
2827 crt = nrt;
2828 } else if (crt != nrt)
2829 continue;
2830 if (!m_append(m, sizeof(struct in6_addr),
2831 (void *)&ims->im6s_addr)) {
2832 if (m != m0)
2833 m_freem(m);
2834 CTR1(KTR_MLD,
2835 "%s: m_append() failed", __func__);
2836 return (-ENOMEM);
2837 }
2838 nallow += !!(crt == REC_ALLOW);
2839 nblock += !!(crt == REC_BLOCK);
2840 if (++rsrcs == m0srcs)
2841 break;
2842 }
2843 /*
2844 * If we did not append any tree nodes on this
2845 * pass, back out of allocations.
2846 */
2847 if (rsrcs == 0) {
2848 npbytes -= sizeof(struct mldv2_record);
2849 if (m != m0) {
2850 CTR1(KTR_MLD,
2851 "%s: m_free(m)", __func__);
2852 m_freem(m);
2853 } else {
2854 CTR1(KTR_MLD,
2855 "%s: m_adj(m, -mr)", __func__);
2856 m_adj(m, -((int)sizeof(
2857 struct mldv2_record)));
2858 }
2859 continue;
2860 }
2861 npbytes += (rsrcs * sizeof(struct in6_addr));
2862 if (crt == REC_ALLOW)
2863 pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2864 else if (crt == REC_BLOCK)
2865 pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2866 pmr->mr_numsrc = htons(rsrcs);
2867 /*
2868 * Count the new group record, and enqueue this
2869 * packet if it wasn't already queued.
2870 */
2871 m->m_pkthdr.PH_vt.vt_nrecs++;
2872 if (m != m0)
2873 mbufq_enqueue(mq, m);
2874 nbytes += npbytes;
2875 } while (nims != NULL);
2876 drt |= crt;
2877 crt = (~crt & REC_FULL);
2878 }
2879
2880 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2881 nallow, nblock);
2882
2883 return (nbytes);
2884 }
2885
2886 static int
mld_v2_merge_state_changes(struct in6_multi * inm,struct mbufq * scq)2887 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2888 {
2889 struct mbufq *gq;
2890 struct mbuf *m; /* pending state-change */
2891 struct mbuf *m0; /* copy of pending state-change */
2892 struct mbuf *mt; /* last state-change in packet */
2893 int docopy, domerge;
2894 u_int recslen;
2895
2896 docopy = 0;
2897 domerge = 0;
2898 recslen = 0;
2899
2900 IN6_MULTI_LIST_LOCK_ASSERT();
2901 MLD_LOCK_ASSERT();
2902
2903 /*
2904 * If there are further pending retransmissions, make a writable
2905 * copy of each queued state-change message before merging.
2906 */
2907 if (inm->in6m_scrv > 0)
2908 docopy = 1;
2909
2910 gq = &inm->in6m_scq;
2911 #ifdef KTR
2912 if (mbufq_first(gq) == NULL) {
2913 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2914 __func__, inm);
2915 }
2916 #endif
2917
2918 m = mbufq_first(gq);
2919 while (m != NULL) {
2920 /*
2921 * Only merge the report into the current packet if
2922 * there is sufficient space to do so; an MLDv2 report
2923 * packet may only contain 65,535 group records.
2924 * Always use a simple mbuf chain concatentation to do this,
2925 * as large state changes for single groups may have
2926 * allocated clusters.
2927 */
2928 domerge = 0;
2929 mt = mbufq_last(scq);
2930 if (mt != NULL) {
2931 recslen = m_length(m, NULL);
2932
2933 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2934 m->m_pkthdr.PH_vt.vt_nrecs <=
2935 MLD_V2_REPORT_MAXRECS) &&
2936 (mt->m_pkthdr.len + recslen <=
2937 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2938 domerge = 1;
2939 }
2940
2941 if (!domerge && mbufq_full(gq)) {
2942 CTR2(KTR_MLD,
2943 "%s: outbound queue full, skipping whole packet %p",
2944 __func__, m);
2945 mt = m->m_nextpkt;
2946 if (!docopy)
2947 m_freem(m);
2948 m = mt;
2949 continue;
2950 }
2951
2952 if (!docopy) {
2953 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2954 m0 = mbufq_dequeue(gq);
2955 m = m0->m_nextpkt;
2956 } else {
2957 CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2958 m0 = m_dup(m, M_NOWAIT);
2959 if (m0 == NULL)
2960 return (ENOMEM);
2961 m0->m_nextpkt = NULL;
2962 m = m->m_nextpkt;
2963 }
2964
2965 if (!domerge) {
2966 CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2967 __func__, m0, scq);
2968 mbufq_enqueue(scq, m0);
2969 } else {
2970 struct mbuf *mtl; /* last mbuf of packet mt */
2971
2972 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2973 __func__, m0, mt);
2974
2975 mtl = m_last(mt);
2976 m0->m_flags &= ~M_PKTHDR;
2977 mt->m_pkthdr.len += recslen;
2978 mt->m_pkthdr.PH_vt.vt_nrecs +=
2979 m0->m_pkthdr.PH_vt.vt_nrecs;
2980
2981 mtl->m_next = m0;
2982 }
2983 }
2984
2985 return (0);
2986 }
2987
2988 /*
2989 * Respond to a pending MLDv2 General Query.
2990 */
2991 static void
mld_v2_dispatch_general_query(struct mld_ifsoftc * mli)2992 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2993 {
2994 struct ifmultiaddr *ifma;
2995 struct ifnet *ifp;
2996 struct in6_multi *inm;
2997 int retval;
2998
2999 IN6_MULTI_LIST_LOCK_ASSERT();
3000 MLD_LOCK_ASSERT();
3001
3002 KASSERT(mli->mli_version == MLD_VERSION_2,
3003 ("%s: called when version %d", __func__, mli->mli_version));
3004
3005 /*
3006 * Check that there are some packets queued. If so, send them first.
3007 * For large number of groups the reply to general query can take
3008 * many packets, we should finish sending them before starting of
3009 * queuing the new reply.
3010 */
3011 if (mbufq_len(&mli->mli_gq) != 0)
3012 goto send;
3013
3014 ifp = mli->mli_ifp;
3015
3016 IF_ADDR_RLOCK(ifp);
3017 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3018 inm = in6m_ifmultiaddr_get_inm(ifma);
3019 if (inm == NULL)
3020 continue;
3021 KASSERT(ifp == inm->in6m_ifp,
3022 ("%s: inconsistent ifp", __func__));
3023
3024 switch (inm->in6m_state) {
3025 case MLD_NOT_MEMBER:
3026 case MLD_SILENT_MEMBER:
3027 break;
3028 case MLD_REPORTING_MEMBER:
3029 case MLD_IDLE_MEMBER:
3030 case MLD_LAZY_MEMBER:
3031 case MLD_SLEEPING_MEMBER:
3032 case MLD_AWAKENING_MEMBER:
3033 inm->in6m_state = MLD_REPORTING_MEMBER;
3034 retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3035 inm, 0, 0, 0, 0);
3036 CTR2(KTR_MLD, "%s: enqueue record = %d",
3037 __func__, retval);
3038 break;
3039 case MLD_G_QUERY_PENDING_MEMBER:
3040 case MLD_SG_QUERY_PENDING_MEMBER:
3041 case MLD_LEAVING_MEMBER:
3042 break;
3043 }
3044 }
3045 IF_ADDR_RUNLOCK(ifp);
3046
3047 send:
3048 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3049
3050 /*
3051 * Slew transmission of bursts over 500ms intervals.
3052 */
3053 if (mbufq_first(&mli->mli_gq) != NULL) {
3054 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3055 MLD_RESPONSE_BURST_INTERVAL);
3056 V_interface_timers_running6 = 1;
3057 }
3058 }
3059
3060 /*
3061 * Transmit the next pending message in the output queue.
3062 *
3063 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3064 * MRT: Nothing needs to be done, as MLD traffic is always local to
3065 * a link and uses a link-scope multicast address.
3066 */
3067 static void
mld_dispatch_packet(struct mbuf * m)3068 mld_dispatch_packet(struct mbuf *m)
3069 {
3070 struct ip6_moptions im6o;
3071 struct ifnet *ifp;
3072 struct ifnet *oifp;
3073 struct mbuf *m0;
3074 struct mbuf *md;
3075 struct ip6_hdr *ip6;
3076 struct mld_hdr *mld;
3077 int error;
3078 int off;
3079 int type;
3080 uint32_t ifindex;
3081
3082 CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3083
3084 /*
3085 * Set VNET image pointer from enqueued mbuf chain
3086 * before doing anything else. Whilst we use interface
3087 * indexes to guard against interface detach, they are
3088 * unique to each VIMAGE and must be retrieved.
3089 */
3090 ifindex = mld_restore_context(m);
3091
3092 /*
3093 * Check if the ifnet still exists. This limits the scope of
3094 * any race in the absence of a global ifp lock for low cost
3095 * (an array lookup).
3096 */
3097 ifp = ifnet_byindex(ifindex);
3098 if (ifp == NULL) {
3099 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3100 __func__, m, ifindex);
3101 m_freem(m);
3102 IP6STAT_INC(ip6s_noroute);
3103 goto out;
3104 }
3105
3106 im6o.im6o_multicast_hlim = 1;
3107 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3108 im6o.im6o_multicast_ifp = ifp;
3109
3110 if (m->m_flags & M_MLDV1) {
3111 m0 = m;
3112 } else {
3113 m0 = mld_v2_encap_report(ifp, m);
3114 if (m0 == NULL) {
3115 CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3116 IP6STAT_INC(ip6s_odropped);
3117 goto out;
3118 }
3119 }
3120
3121 mld_scrub_context(m0);
3122 m_clrprotoflags(m);
3123 m0->m_pkthdr.rcvif = V_loif;
3124
3125 ip6 = mtod(m0, struct ip6_hdr *);
3126 #if 0
3127 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */
3128 #else
3129 /*
3130 * XXX XXX Break some KPI rules to prevent an LOR which would
3131 * occur if we called in6_setscope() at transmission.
3132 * See comments at top of file.
3133 */
3134 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3135 #endif
3136
3137 /*
3138 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3139 * so we can bump the stats.
3140 */
3141 md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3142 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3143 type = mld->mld_type;
3144
3145 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3146 &oifp, NULL);
3147 if (error) {
3148 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3149 goto out;
3150 }
3151 ICMP6STAT_INC(icp6s_outhist[type]);
3152 if (oifp != NULL) {
3153 icmp6_ifstat_inc(oifp, ifs6_out_msg);
3154 switch (type) {
3155 case MLD_LISTENER_REPORT:
3156 case MLDV2_LISTENER_REPORT:
3157 icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3158 break;
3159 case MLD_LISTENER_DONE:
3160 icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3161 break;
3162 }
3163 }
3164 out:
3165 return;
3166 }
3167
3168 /*
3169 * Encapsulate an MLDv2 report.
3170 *
3171 * KAME IPv6 requires that hop-by-hop options be passed separately,
3172 * and that the IPv6 header be prepended in a separate mbuf.
3173 *
3174 * Returns a pointer to the new mbuf chain head, or NULL if the
3175 * allocation failed.
3176 */
3177 static struct mbuf *
mld_v2_encap_report(struct ifnet * ifp,struct mbuf * m)3178 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3179 {
3180 struct mbuf *mh;
3181 struct mldv2_report *mld;
3182 struct ip6_hdr *ip6;
3183 struct in6_ifaddr *ia;
3184 int mldreclen;
3185
3186 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3187 KASSERT((m->m_flags & M_PKTHDR),
3188 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3189
3190 /*
3191 * RFC3590: OK to send as :: or tentative during DAD.
3192 */
3193 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3194 if (ia == NULL)
3195 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3196
3197 mh = m_gethdr(M_NOWAIT, MT_DATA);
3198 if (mh == NULL) {
3199 if (ia != NULL)
3200 ifa_free(&ia->ia_ifa);
3201 m_freem(m);
3202 return (NULL);
3203 }
3204 M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3205
3206 mldreclen = m_length(m, NULL);
3207 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3208
3209 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3210 mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3211 sizeof(struct mldv2_report) + mldreclen;
3212
3213 ip6 = mtod(mh, struct ip6_hdr *);
3214 ip6->ip6_flow = 0;
3215 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3216 ip6->ip6_vfc |= IPV6_VERSION;
3217 ip6->ip6_nxt = IPPROTO_ICMPV6;
3218 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3219 if (ia != NULL)
3220 ifa_free(&ia->ia_ifa);
3221 ip6->ip6_dst = in6addr_linklocal_allv2routers;
3222 /* scope ID will be set in netisr */
3223
3224 mld = (struct mldv2_report *)(ip6 + 1);
3225 mld->mld_type = MLDV2_LISTENER_REPORT;
3226 mld->mld_code = 0;
3227 mld->mld_cksum = 0;
3228 mld->mld_v2_reserved = 0;
3229 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3230 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3231
3232 mh->m_next = m;
3233 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3234 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3235 return (mh);
3236 }
3237
3238 #ifdef KTR
3239 static char *
mld_rec_type_to_str(const int type)3240 mld_rec_type_to_str(const int type)
3241 {
3242
3243 switch (type) {
3244 case MLD_CHANGE_TO_EXCLUDE_MODE:
3245 return "TO_EX";
3246 break;
3247 case MLD_CHANGE_TO_INCLUDE_MODE:
3248 return "TO_IN";
3249 break;
3250 case MLD_MODE_IS_EXCLUDE:
3251 return "MODE_EX";
3252 break;
3253 case MLD_MODE_IS_INCLUDE:
3254 return "MODE_IN";
3255 break;
3256 case MLD_ALLOW_NEW_SOURCES:
3257 return "ALLOW_NEW";
3258 break;
3259 case MLD_BLOCK_OLD_SOURCES:
3260 return "BLOCK_OLD";
3261 break;
3262 default:
3263 break;
3264 }
3265 return "unknown";
3266 }
3267 #endif
3268
3269 static void
mld_init(void * unused __unused)3270 mld_init(void *unused __unused)
3271 {
3272
3273 CTR1(KTR_MLD, "%s: initializing", __func__);
3274 MLD_LOCK_INIT();
3275
3276 ip6_initpktopts(&mld_po);
3277 mld_po.ip6po_hlim = 1;
3278 mld_po.ip6po_hbh = &mld_ra.hbh;
3279 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3280 mld_po.ip6po_flags = IP6PO_DONTFRAG;
3281 }
3282 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3283
3284 static void
mld_uninit(void * unused __unused)3285 mld_uninit(void *unused __unused)
3286 {
3287
3288 CTR1(KTR_MLD, "%s: tearing down", __func__);
3289 MLD_LOCK_DESTROY();
3290 }
3291 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3292
3293 static void
vnet_mld_init(const void * unused __unused)3294 vnet_mld_init(const void *unused __unused)
3295 {
3296
3297 CTR1(KTR_MLD, "%s: initializing", __func__);
3298
3299 LIST_INIT(&V_mli_head);
3300 }
3301 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3302 NULL);
3303
3304 static void
vnet_mld_uninit(const void * unused __unused)3305 vnet_mld_uninit(const void *unused __unused)
3306 {
3307
3308 /* This can happen if we shutdown the network stack. */
3309 CTR1(KTR_MLD, "%s: tearing down", __func__);
3310 }
3311 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3312 NULL);
3313
3314 static int
mld_modevent(module_t mod,int type,void * unused __unused)3315 mld_modevent(module_t mod, int type, void *unused __unused)
3316 {
3317
3318 switch (type) {
3319 case MOD_LOAD:
3320 case MOD_UNLOAD:
3321 break;
3322 default:
3323 return (EOPNOTSUPP);
3324 }
3325 return (0);
3326 }
3327
3328 static moduledata_t mld_mod = {
3329 "mld",
3330 mld_modevent,
3331 0
3332 };
3333 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3334