2 * Copyright (c) 2014 Red Hat, Inc.
4 * Based on mac-learning implementation.
6 * Licensed under the Apache License, Version 2.0 (the "License");
7 * you may not use this file except in compliance with the License.
8 * You may obtain a copy of the License at:
10 * http://www.apache.org/licenses/LICENSE-2.0
12 * Unless required by applicable law or agreed to in writing, software
13 * distributed under the License is distributed on an "AS IS" BASIS,
14 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15 * See the License for the specific language governing permissions and
16 * limitations under the License.
20 #include "mcast-snooping.h"
26 #include "byte-order.h"
30 #include "poll-loop.h"
33 #include "unaligned.h"
35 #include "vlan-bitmap.h"
36 #include "openvswitch/vlog.h"
38 COVERAGE_DEFINE(mcast_snooping_learned);
39 COVERAGE_DEFINE(mcast_snooping_expired);
41 static struct mcast_port_bundle *
42 mcast_snooping_port_lookup(struct ovs_list *list, void *port);
43 static struct mcast_mrouter_bundle *
44 mcast_snooping_mrouter_lookup(struct mcast_snooping *ms, uint16_t vlan,
46 OVS_REQ_RDLOCK(ms->rwlock);
49 mcast_snooping_enabled(const struct mcast_snooping *ms)
55 mcast_snooping_flood_unreg(const struct mcast_snooping *ms)
57 return ms->flood_unreg;
61 mcast_snooping_is_query(ovs_be16 igmp_type)
63 return igmp_type == htons(IGMP_HOST_MEMBERSHIP_QUERY);
67 mcast_snooping_is_membership(ovs_be16 igmp_type)
69 switch (ntohs(igmp_type)) {
70 case IGMP_HOST_MEMBERSHIP_REPORT:
71 case IGMPV2_HOST_MEMBERSHIP_REPORT:
72 case IGMP_HOST_LEAVE_MESSAGE:
78 /* Returns the number of seconds since multicast group 'b' was learned in a
81 mcast_bundle_age(const struct mcast_snooping *ms,
82 const struct mcast_group_bundle *b)
84 time_t remaining = b->expires - time_now();
85 return ms->idle_time - remaining;
89 mcast_table_hash(const struct mcast_snooping *ms, ovs_be32 grp_ip4,
92 return hash_3words((OVS_FORCE uint32_t) grp_ip4, vlan, ms->secret);
95 static struct mcast_group_bundle *
96 mcast_group_bundle_from_lru_node(struct ovs_list *list)
98 return CONTAINER_OF(list, struct mcast_group_bundle, bundle_node);
101 static struct mcast_group *
102 mcast_group_from_lru_node(struct ovs_list *list)
104 return CONTAINER_OF(list, struct mcast_group, group_node);
107 /* Searches 'ms' for and returns an mcast group for destination address
108 * 'dip' in 'vlan'. */
110 mcast_snooping_lookup(const struct mcast_snooping *ms, ovs_be32 dip,
112 OVS_REQ_RDLOCK(ms->rwlock)
114 struct mcast_group *grp;
117 hash = mcast_table_hash(ms, dip, vlan);
118 HMAP_FOR_EACH_WITH_HASH (grp, hmap_node, hash, &ms->table) {
119 if (grp->vlan == vlan && grp->ip4 == dip) {
126 /* If the LRU list is not empty, stores the least-recently-used entry
127 * in '*e' and returns true. Otherwise, if the LRU list is empty,
128 * stores NULL in '*e' and return false. */
130 group_get_lru(const struct mcast_snooping *ms, struct mcast_group **grp)
131 OVS_REQ_RDLOCK(ms->rwlock)
133 if (!list_is_empty(&ms->group_lru)) {
134 *grp = mcast_group_from_lru_node(ms->group_lru.next);
143 normalize_idle_time(unsigned int idle_time)
145 return (idle_time < 15 ? 15
146 : idle_time > 3600 ? 3600
150 /* Creates and returns a new mcast table with an initial mcast aging
151 * timeout of MCAST_ENTRY_DEFAULT_IDLE_TIME seconds and an initial maximum of
152 * MCAST_DEFAULT_MAX entries. */
153 struct mcast_snooping *
154 mcast_snooping_create(void)
156 struct mcast_snooping *ms;
158 ms = xmalloc(sizeof *ms);
159 hmap_init(&ms->table);
160 list_init(&ms->group_lru);
161 list_init(&ms->mrouter_lru);
162 list_init(&ms->fport_list);
163 ms->secret = random_uint32();
164 ms->idle_time = MCAST_ENTRY_DEFAULT_IDLE_TIME;
165 ms->max_entries = MCAST_DEFAULT_MAX_ENTRIES;
166 ms->need_revalidate = false;
167 ms->flood_unreg = true;
168 ovs_refcount_init(&ms->ref_cnt);
169 ovs_rwlock_init(&ms->rwlock);
173 struct mcast_snooping *
174 mcast_snooping_ref(const struct mcast_snooping *ms_)
176 struct mcast_snooping *ms = CONST_CAST(struct mcast_snooping *, ms_);
178 ovs_refcount_ref(&ms->ref_cnt);
183 /* Unreferences (and possibly destroys) mcast snooping table 'ms'. */
185 mcast_snooping_unref(struct mcast_snooping *ms)
187 if (!mcast_snooping_enabled(ms)) {
191 if (ovs_refcount_unref_relaxed(&ms->ref_cnt) == 1) {
192 mcast_snooping_flush(ms);
193 hmap_destroy(&ms->table);
194 ovs_rwlock_destroy(&ms->rwlock);
199 /* Changes the mcast aging timeout of 'ms' to 'idle_time' seconds. */
201 mcast_snooping_set_idle_time(struct mcast_snooping *ms, unsigned int idle_time)
202 OVS_REQ_WRLOCK(ms->rwlock)
204 struct mcast_group *grp;
205 struct mcast_group_bundle *b;
208 idle_time = normalize_idle_time(idle_time);
209 if (idle_time != ms->idle_time) {
210 delta = (int) idle_time - (int) ms->idle_time;
211 LIST_FOR_EACH (grp, group_node, &ms->group_lru) {
212 LIST_FOR_EACH (b, bundle_node, &grp->bundle_lru) {
216 ms->idle_time = idle_time;
220 /* Sets the maximum number of entries in 'ms' to 'max_entries', adjusting it
221 * to be within a reasonable range. */
223 mcast_snooping_set_max_entries(struct mcast_snooping *ms,
225 OVS_REQ_WRLOCK(ms->rwlock)
227 ms->max_entries = (max_entries < 10 ? 10
228 : max_entries > 1000 * 1000 ? 1000 * 1000
232 /* Sets if unregistered multicast packets should be flooded to
233 * all ports or only to ports connected to multicast routers
235 * Returns true if previous state differs from current state,
236 * false otherwise. */
238 mcast_snooping_set_flood_unreg(struct mcast_snooping *ms, bool enable)
239 OVS_REQ_WRLOCK(ms->rwlock)
241 bool prev = ms->flood_unreg;
242 ms->flood_unreg = enable;
243 return prev != enable;
246 static struct mcast_group_bundle *
247 mcast_group_bundle_lookup(struct mcast_snooping *ms OVS_UNUSED,
248 struct mcast_group *grp, void *port)
249 OVS_REQ_RDLOCK(ms->rwlock)
251 struct mcast_group_bundle *b;
253 LIST_FOR_EACH (b, bundle_node, &grp->bundle_lru) {
254 if (b->port == port) {
261 /* Insert a new bundle to the mcast group or update its
262 * position and expiration if it is already there. */
263 static struct mcast_group_bundle *
264 mcast_group_insert_bundle(struct mcast_snooping *ms OVS_UNUSED,
265 struct mcast_group *grp, void *port, int idle_time)
266 OVS_REQ_WRLOCK(ms->rwlock)
268 struct mcast_group_bundle *b;
270 b = mcast_group_bundle_lookup(ms, grp, port);
272 list_remove(&b->bundle_node);
274 b = xmalloc(sizeof *b);
275 list_init(&b->bundle_node);
279 b->expires = time_now() + idle_time;
280 list_push_back(&grp->bundle_lru, &b->bundle_node);
284 /* Return true if multicast still has bundles associated.
285 * Return false if there is no bundles. */
287 mcast_group_has_bundles(struct mcast_group *grp)
289 return !list_is_empty(&grp->bundle_lru);
292 /* Delete 'grp' from the 'ms' hash table.
293 * Caller is responsible to clean bundle lru first. */
295 mcast_snooping_flush_group__(struct mcast_snooping *ms,
296 struct mcast_group *grp)
298 ovs_assert(list_is_empty(&grp->bundle_lru));
299 hmap_remove(&ms->table, &grp->hmap_node);
300 list_remove(&grp->group_node);
304 /* Flush out mcast group and its bundles */
306 mcast_snooping_flush_group(struct mcast_snooping *ms, struct mcast_group *grp)
307 OVS_REQ_WRLOCK(ms->rwlock)
309 struct mcast_group_bundle *b, *next_b;
311 LIST_FOR_EACH_SAFE (b, next_b, bundle_node, &grp->bundle_lru) {
312 list_remove(&b->bundle_node);
315 mcast_snooping_flush_group__(ms, grp);
316 ms->need_revalidate = true;
320 /* Delete bundle returning true if it succeeds,
321 * false if it didn't find the group. */
323 mcast_group_delete_bundle(struct mcast_snooping *ms OVS_UNUSED,
324 struct mcast_group *grp, void *port)
325 OVS_REQ_WRLOCK(ms->rwlock)
327 struct mcast_group_bundle *b;
329 LIST_FOR_EACH (b, bundle_node, &grp->bundle_lru) {
330 if (b->port == port) {
331 list_remove(&b->bundle_node);
339 /* If any bundle has expired, delete it. Returns the number of deleted
342 mcast_snooping_prune_expired(struct mcast_snooping *ms,
343 struct mcast_group *grp)
344 OVS_REQ_WRLOCK(ms->rwlock)
347 struct mcast_group_bundle *b, *next_b;
348 time_t timenow = time_now();
351 LIST_FOR_EACH_SAFE (b, next_b, bundle_node, &grp->bundle_lru) {
352 /* This list is sorted on expiration time. */
353 if (b->expires > timenow) {
356 list_remove(&b->bundle_node);
361 if (!mcast_group_has_bundles(grp)) {
362 mcast_snooping_flush_group__(ms, grp);
367 ms->need_revalidate = true;
368 COVERAGE_ADD(mcast_snooping_expired, expired);
374 /* Add a multicast group to the mdb. If it exists, then
375 * move to the last position in the LRU list.
378 mcast_snooping_add_group(struct mcast_snooping *ms, ovs_be32 ip4,
379 uint16_t vlan, void *port)
380 OVS_REQ_WRLOCK(ms->rwlock)
383 struct mcast_group *grp;
385 /* Avoid duplicate packets. */
386 if (mcast_snooping_mrouter_lookup(ms, vlan, port)
387 || mcast_snooping_port_lookup(&ms->fport_list, port)) {
392 grp = mcast_snooping_lookup(ms, ip4, vlan);
394 uint32_t hash = mcast_table_hash(ms, ip4, vlan);
396 if (hmap_count(&ms->table) >= ms->max_entries) {
397 group_get_lru(ms, &grp);
398 mcast_snooping_flush_group(ms, grp);
401 grp = xmalloc(sizeof *grp);
402 hmap_insert(&ms->table, &grp->hmap_node, hash);
405 list_init(&grp->bundle_lru);
407 ms->need_revalidate = true;
408 COVERAGE_INC(mcast_snooping_learned);
410 list_remove(&grp->group_node);
412 mcast_group_insert_bundle(ms, grp, port, ms->idle_time);
414 /* Mark 'grp' as recently used. */
415 list_push_back(&ms->group_lru, &grp->group_node);
420 mcast_snooping_leave_group(struct mcast_snooping *ms, ovs_be32 ip4,
421 uint16_t vlan, void *port)
422 OVS_REQ_WRLOCK(ms->rwlock)
424 struct mcast_group *grp;
426 grp = mcast_snooping_lookup(ms, ip4, vlan);
427 if (grp && mcast_group_delete_bundle(ms, grp, port)) {
428 ms->need_revalidate = true;
437 /* Returns the number of seconds since the multicast router
438 * was learned in a port. */
440 mcast_mrouter_age(const struct mcast_snooping *ms OVS_UNUSED,
441 const struct mcast_mrouter_bundle *mrouter)
443 time_t remaining = mrouter->expires - time_now();
444 return MCAST_MROUTER_PORT_IDLE_TIME - remaining;
447 static struct mcast_mrouter_bundle *
448 mcast_mrouter_from_lru_node(struct ovs_list *list)
450 return CONTAINER_OF(list, struct mcast_mrouter_bundle, mrouter_node);
453 /* If the LRU list is not empty, stores the least-recently-used mrouter
454 * in '*m' and returns true. Otherwise, if the LRU list is empty,
455 * stores NULL in '*m' and return false. */
457 mrouter_get_lru(const struct mcast_snooping *ms,
458 struct mcast_mrouter_bundle **m)
459 OVS_REQ_RDLOCK(ms->rwlock)
461 if (!list_is_empty(&ms->mrouter_lru)) {
462 *m = mcast_mrouter_from_lru_node(ms->mrouter_lru.next);
470 static struct mcast_mrouter_bundle *
471 mcast_snooping_mrouter_lookup(struct mcast_snooping *ms, uint16_t vlan,
473 OVS_REQ_RDLOCK(ms->rwlock)
475 struct mcast_mrouter_bundle *mrouter;
477 LIST_FOR_EACH (mrouter, mrouter_node, &ms->mrouter_lru) {
478 if (mrouter->vlan == vlan && mrouter->port == port) {
486 mcast_snooping_add_mrouter(struct mcast_snooping *ms, uint16_t vlan,
488 OVS_REQ_WRLOCK(ms->rwlock)
490 struct mcast_mrouter_bundle *mrouter;
492 /* Avoid duplicate packets. */
493 if (mcast_snooping_port_lookup(&ms->fport_list, port)) {
497 mrouter = mcast_snooping_mrouter_lookup(ms, vlan, port);
499 list_remove(&mrouter->mrouter_node);
501 mrouter = xmalloc(sizeof *mrouter);
502 mrouter->vlan = vlan;
503 mrouter->port = port;
504 COVERAGE_INC(mcast_snooping_learned);
505 ms->need_revalidate = true;
508 mrouter->expires = time_now() + MCAST_MROUTER_PORT_IDLE_TIME;
509 list_push_back(&ms->mrouter_lru, &mrouter->mrouter_node);
510 return ms->need_revalidate;
514 mcast_snooping_flush_mrouter(struct mcast_mrouter_bundle *mrouter)
516 list_remove(&mrouter->mrouter_node);
522 static struct mcast_port_bundle *
523 mcast_port_from_list_node(struct ovs_list *list)
525 return CONTAINER_OF(list, struct mcast_port_bundle, node);
528 /* If the list is not empty, stores the fport in '*f' and returns true.
529 * Otherwise, if the list is empty, stores NULL in '*f' and return false. */
531 mcast_snooping_port_get(const struct ovs_list *list,
532 struct mcast_port_bundle **f)
534 if (!list_is_empty(list)) {
535 *f = mcast_port_from_list_node(list->next);
543 static struct mcast_port_bundle *
544 mcast_snooping_port_lookup(struct ovs_list *list, void *port)
546 struct mcast_port_bundle *pbundle;
548 LIST_FOR_EACH (pbundle, node, list) {
549 if (pbundle->port == port) {
557 mcast_snooping_add_port(struct ovs_list *list, void *port)
559 struct mcast_port_bundle *pbundle;
561 pbundle = xmalloc(sizeof *pbundle);
562 pbundle->port = port;
563 list_insert(list, &pbundle->node);
567 mcast_snooping_flush_port(struct mcast_port_bundle *pbundle)
569 list_remove(&pbundle->node);
576 mcast_snooping_set_port_flood(struct mcast_snooping *ms, void *port,
578 OVS_REQ_WRLOCK(ms->rwlock)
580 struct mcast_port_bundle *fbundle;
582 fbundle = mcast_snooping_port_lookup(&ms->fport_list, port);
583 if (flood && !fbundle) {
584 mcast_snooping_add_port(&ms->fport_list, port);
585 ms->need_revalidate = true;
586 } else if (!flood && fbundle) {
587 mcast_snooping_flush_port(fbundle);
588 ms->need_revalidate = true;
595 mcast_snooping_mdb_flush__(struct mcast_snooping *ms)
596 OVS_REQ_WRLOCK(ms->rwlock)
598 struct mcast_group *grp;
599 struct mcast_mrouter_bundle *mrouter;
601 while (group_get_lru(ms, &grp)) {
602 mcast_snooping_flush_group(ms, grp);
605 hmap_shrink(&ms->table);
607 while (mrouter_get_lru(ms, &mrouter)) {
608 mcast_snooping_flush_mrouter(mrouter);
613 mcast_snooping_mdb_flush(struct mcast_snooping *ms)
615 if (!mcast_snooping_enabled(ms)) {
619 ovs_rwlock_wrlock(&ms->rwlock);
620 mcast_snooping_mdb_flush__(ms);
621 ovs_rwlock_unlock(&ms->rwlock);
624 /* Flushes mdb and flood ports. */
626 mcast_snooping_flush__(struct mcast_snooping *ms)
627 OVS_REQ_WRLOCK(ms->rwlock)
629 struct mcast_group *grp;
630 struct mcast_mrouter_bundle *mrouter;
631 struct mcast_port_bundle *pbundle;
633 while (group_get_lru(ms, &grp)) {
634 mcast_snooping_flush_group(ms, grp);
637 hmap_shrink(&ms->table);
639 while (mrouter_get_lru(ms, &mrouter)) {
640 mcast_snooping_flush_mrouter(mrouter);
643 while (mcast_snooping_port_get(&ms->fport_list, &pbundle)) {
644 mcast_snooping_flush_port(pbundle);
649 mcast_snooping_flush(struct mcast_snooping *ms)
651 if (!mcast_snooping_enabled(ms)) {
655 ovs_rwlock_wrlock(&ms->rwlock);
656 mcast_snooping_flush__(ms);
657 ovs_rwlock_unlock(&ms->rwlock);
661 mcast_snooping_run__(struct mcast_snooping *ms)
662 OVS_REQ_WRLOCK(ms->rwlock)
664 bool need_revalidate;
665 struct mcast_group *grp;
666 struct mcast_mrouter_bundle *mrouter;
669 while (group_get_lru(ms, &grp)) {
670 if (hmap_count(&ms->table) > ms->max_entries) {
671 mcast_snooping_flush_group(ms, grp);
673 if (!mcast_snooping_prune_expired(ms, grp)) {
679 hmap_shrink(&ms->table);
682 while (mrouter_get_lru(ms, &mrouter)
683 && time_now() >= mrouter->expires) {
684 mcast_snooping_flush_mrouter(mrouter);
688 if (mrouter_expired) {
689 ms->need_revalidate = true;
690 COVERAGE_ADD(mcast_snooping_expired, mrouter_expired);
693 need_revalidate = ms->need_revalidate;
694 ms->need_revalidate = false;
695 return need_revalidate;
698 /* Does periodic work required by 'ms'. Returns true if something changed
699 * that may require flow revalidation. */
701 mcast_snooping_run(struct mcast_snooping *ms)
703 bool need_revalidate;
705 if (!mcast_snooping_enabled(ms)) {
709 ovs_rwlock_wrlock(&ms->rwlock);
710 need_revalidate = mcast_snooping_run__(ms);
711 ovs_rwlock_unlock(&ms->rwlock);
713 return need_revalidate;
717 mcast_snooping_wait__(struct mcast_snooping *ms)
718 OVS_REQ_RDLOCK(ms->rwlock)
720 if (hmap_count(&ms->table) > ms->max_entries
721 || ms->need_revalidate) {
722 poll_immediate_wake();
724 struct mcast_group *grp;
725 struct mcast_group_bundle *bundle;
726 struct mcast_mrouter_bundle *mrouter;
727 long long int mrouter_msec;
728 long long int msec = 0;
730 if (!list_is_empty(&ms->group_lru)) {
731 grp = mcast_group_from_lru_node(ms->group_lru.next);
732 bundle = mcast_group_bundle_from_lru_node(grp->bundle_lru.next);
733 msec = bundle->expires * 1000LL;
736 if (!list_is_empty(&ms->mrouter_lru)) {
737 mrouter = mcast_mrouter_from_lru_node(ms->mrouter_lru.next);
738 mrouter_msec = mrouter->expires * 1000LL;
739 msec = msec ? MIN(msec, mrouter_msec) : mrouter_msec;
743 poll_timer_wait_until(msec);
749 mcast_snooping_wait(struct mcast_snooping *ms)
751 if (!mcast_snooping_enabled(ms)) {
755 ovs_rwlock_rdlock(&ms->rwlock);
756 mcast_snooping_wait__(ms);
757 ovs_rwlock_unlock(&ms->rwlock);