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 list_init(&ms->rport_list);
164 ms->secret = random_uint32();
165 ms->idle_time = MCAST_ENTRY_DEFAULT_IDLE_TIME;
166 ms->max_entries = MCAST_DEFAULT_MAX_ENTRIES;
167 ms->need_revalidate = false;
168 ms->flood_unreg = true;
169 ovs_refcount_init(&ms->ref_cnt);
170 ovs_rwlock_init(&ms->rwlock);
174 struct mcast_snooping *
175 mcast_snooping_ref(const struct mcast_snooping *ms_)
177 struct mcast_snooping *ms = CONST_CAST(struct mcast_snooping *, ms_);
179 ovs_refcount_ref(&ms->ref_cnt);
184 /* Unreferences (and possibly destroys) mcast snooping table 'ms'. */
186 mcast_snooping_unref(struct mcast_snooping *ms)
188 if (!mcast_snooping_enabled(ms)) {
192 if (ovs_refcount_unref_relaxed(&ms->ref_cnt) == 1) {
193 mcast_snooping_flush(ms);
194 hmap_destroy(&ms->table);
195 ovs_rwlock_destroy(&ms->rwlock);
200 /* Changes the mcast aging timeout of 'ms' to 'idle_time' seconds. */
202 mcast_snooping_set_idle_time(struct mcast_snooping *ms, unsigned int idle_time)
203 OVS_REQ_WRLOCK(ms->rwlock)
205 struct mcast_group *grp;
206 struct mcast_group_bundle *b;
209 idle_time = normalize_idle_time(idle_time);
210 if (idle_time != ms->idle_time) {
211 delta = (int) idle_time - (int) ms->idle_time;
212 LIST_FOR_EACH (grp, group_node, &ms->group_lru) {
213 LIST_FOR_EACH (b, bundle_node, &grp->bundle_lru) {
217 ms->idle_time = idle_time;
221 /* Sets the maximum number of entries in 'ms' to 'max_entries', adjusting it
222 * to be within a reasonable range. */
224 mcast_snooping_set_max_entries(struct mcast_snooping *ms,
226 OVS_REQ_WRLOCK(ms->rwlock)
228 ms->max_entries = (max_entries < 10 ? 10
229 : max_entries > 1000 * 1000 ? 1000 * 1000
233 /* Sets if unregistered multicast packets should be flooded to
234 * all ports or only to ports connected to multicast routers
236 * Returns true if previous state differs from current state,
237 * false otherwise. */
239 mcast_snooping_set_flood_unreg(struct mcast_snooping *ms, bool enable)
240 OVS_REQ_WRLOCK(ms->rwlock)
242 bool prev = ms->flood_unreg;
243 ms->flood_unreg = enable;
244 return prev != enable;
247 static struct mcast_group_bundle *
248 mcast_group_bundle_lookup(struct mcast_snooping *ms OVS_UNUSED,
249 struct mcast_group *grp, void *port)
250 OVS_REQ_RDLOCK(ms->rwlock)
252 struct mcast_group_bundle *b;
254 LIST_FOR_EACH (b, bundle_node, &grp->bundle_lru) {
255 if (b->port == port) {
262 /* Insert a new bundle to the mcast group or update its
263 * position and expiration if it is already there. */
264 static struct mcast_group_bundle *
265 mcast_group_insert_bundle(struct mcast_snooping *ms OVS_UNUSED,
266 struct mcast_group *grp, void *port, int idle_time)
267 OVS_REQ_WRLOCK(ms->rwlock)
269 struct mcast_group_bundle *b;
271 b = mcast_group_bundle_lookup(ms, grp, port);
273 list_remove(&b->bundle_node);
275 b = xmalloc(sizeof *b);
276 list_init(&b->bundle_node);
280 b->expires = time_now() + idle_time;
281 list_push_back(&grp->bundle_lru, &b->bundle_node);
285 /* Return true if multicast still has bundles associated.
286 * Return false if there is no bundles. */
288 mcast_group_has_bundles(struct mcast_group *grp)
290 return !list_is_empty(&grp->bundle_lru);
293 /* Delete 'grp' from the 'ms' hash table.
294 * Caller is responsible to clean bundle lru first. */
296 mcast_snooping_flush_group__(struct mcast_snooping *ms,
297 struct mcast_group *grp)
299 ovs_assert(list_is_empty(&grp->bundle_lru));
300 hmap_remove(&ms->table, &grp->hmap_node);
301 list_remove(&grp->group_node);
305 /* Flush out mcast group and its bundles */
307 mcast_snooping_flush_group(struct mcast_snooping *ms, struct mcast_group *grp)
308 OVS_REQ_WRLOCK(ms->rwlock)
310 struct mcast_group_bundle *b, *next_b;
312 LIST_FOR_EACH_SAFE (b, next_b, bundle_node, &grp->bundle_lru) {
313 list_remove(&b->bundle_node);
316 mcast_snooping_flush_group__(ms, grp);
317 ms->need_revalidate = true;
321 /* Delete bundle returning true if it succeeds,
322 * false if it didn't find the group. */
324 mcast_group_delete_bundle(struct mcast_snooping *ms OVS_UNUSED,
325 struct mcast_group *grp, void *port)
326 OVS_REQ_WRLOCK(ms->rwlock)
328 struct mcast_group_bundle *b;
330 LIST_FOR_EACH (b, bundle_node, &grp->bundle_lru) {
331 if (b->port == port) {
332 list_remove(&b->bundle_node);
340 /* If any bundle has expired, delete it. Returns the number of deleted
343 mcast_snooping_prune_expired(struct mcast_snooping *ms,
344 struct mcast_group *grp)
345 OVS_REQ_WRLOCK(ms->rwlock)
348 struct mcast_group_bundle *b, *next_b;
349 time_t timenow = time_now();
352 LIST_FOR_EACH_SAFE (b, next_b, bundle_node, &grp->bundle_lru) {
353 /* This list is sorted on expiration time. */
354 if (b->expires > timenow) {
357 list_remove(&b->bundle_node);
362 if (!mcast_group_has_bundles(grp)) {
363 mcast_snooping_flush_group__(ms, grp);
368 ms->need_revalidate = true;
369 COVERAGE_ADD(mcast_snooping_expired, expired);
375 /* Add a multicast group to the mdb. If it exists, then
376 * move to the last position in the LRU list.
379 mcast_snooping_add_group(struct mcast_snooping *ms, ovs_be32 ip4,
380 uint16_t vlan, void *port)
381 OVS_REQ_WRLOCK(ms->rwlock)
384 struct mcast_group *grp;
386 /* Avoid duplicate packets. */
387 if (mcast_snooping_mrouter_lookup(ms, vlan, port)
388 || mcast_snooping_port_lookup(&ms->fport_list, port)) {
393 grp = mcast_snooping_lookup(ms, ip4, vlan);
395 uint32_t hash = mcast_table_hash(ms, ip4, vlan);
397 if (hmap_count(&ms->table) >= ms->max_entries) {
398 group_get_lru(ms, &grp);
399 mcast_snooping_flush_group(ms, grp);
402 grp = xmalloc(sizeof *grp);
403 hmap_insert(&ms->table, &grp->hmap_node, hash);
406 list_init(&grp->bundle_lru);
408 ms->need_revalidate = true;
409 COVERAGE_INC(mcast_snooping_learned);
411 list_remove(&grp->group_node);
413 mcast_group_insert_bundle(ms, grp, port, ms->idle_time);
415 /* Mark 'grp' as recently used. */
416 list_push_back(&ms->group_lru, &grp->group_node);
421 mcast_snooping_leave_group(struct mcast_snooping *ms, ovs_be32 ip4,
422 uint16_t vlan, void *port)
423 OVS_REQ_WRLOCK(ms->rwlock)
425 struct mcast_group *grp;
427 /* Ports flagged to forward Reports usually have more
428 * than one host behind it, so don't leave the group
429 * on the first message and just let it expire */
430 if (mcast_snooping_port_lookup(&ms->rport_list, port)) {
434 grp = mcast_snooping_lookup(ms, ip4, vlan);
435 if (grp && mcast_group_delete_bundle(ms, grp, port)) {
436 ms->need_revalidate = true;
445 /* Returns the number of seconds since the multicast router
446 * was learned in a port. */
448 mcast_mrouter_age(const struct mcast_snooping *ms OVS_UNUSED,
449 const struct mcast_mrouter_bundle *mrouter)
451 time_t remaining = mrouter->expires - time_now();
452 return MCAST_MROUTER_PORT_IDLE_TIME - remaining;
455 static struct mcast_mrouter_bundle *
456 mcast_mrouter_from_lru_node(struct ovs_list *list)
458 return CONTAINER_OF(list, struct mcast_mrouter_bundle, mrouter_node);
461 /* If the LRU list is not empty, stores the least-recently-used mrouter
462 * in '*m' and returns true. Otherwise, if the LRU list is empty,
463 * stores NULL in '*m' and return false. */
465 mrouter_get_lru(const struct mcast_snooping *ms,
466 struct mcast_mrouter_bundle **m)
467 OVS_REQ_RDLOCK(ms->rwlock)
469 if (!list_is_empty(&ms->mrouter_lru)) {
470 *m = mcast_mrouter_from_lru_node(ms->mrouter_lru.next);
478 static struct mcast_mrouter_bundle *
479 mcast_snooping_mrouter_lookup(struct mcast_snooping *ms, uint16_t vlan,
481 OVS_REQ_RDLOCK(ms->rwlock)
483 struct mcast_mrouter_bundle *mrouter;
485 LIST_FOR_EACH (mrouter, mrouter_node, &ms->mrouter_lru) {
486 if (mrouter->vlan == vlan && mrouter->port == port) {
494 mcast_snooping_add_mrouter(struct mcast_snooping *ms, uint16_t vlan,
496 OVS_REQ_WRLOCK(ms->rwlock)
498 struct mcast_mrouter_bundle *mrouter;
500 /* Avoid duplicate packets. */
501 if (mcast_snooping_port_lookup(&ms->fport_list, port)) {
505 mrouter = mcast_snooping_mrouter_lookup(ms, vlan, port);
507 list_remove(&mrouter->mrouter_node);
509 mrouter = xmalloc(sizeof *mrouter);
510 mrouter->vlan = vlan;
511 mrouter->port = port;
512 COVERAGE_INC(mcast_snooping_learned);
513 ms->need_revalidate = true;
516 mrouter->expires = time_now() + MCAST_MROUTER_PORT_IDLE_TIME;
517 list_push_back(&ms->mrouter_lru, &mrouter->mrouter_node);
518 return ms->need_revalidate;
522 mcast_snooping_flush_mrouter(struct mcast_mrouter_bundle *mrouter)
524 list_remove(&mrouter->mrouter_node);
530 static struct mcast_port_bundle *
531 mcast_port_from_list_node(struct ovs_list *list)
533 return CONTAINER_OF(list, struct mcast_port_bundle, node);
536 /* If the list is not empty, stores the fport in '*f' and returns true.
537 * Otherwise, if the list is empty, stores NULL in '*f' and return false. */
539 mcast_snooping_port_get(const struct ovs_list *list,
540 struct mcast_port_bundle **f)
542 if (!list_is_empty(list)) {
543 *f = mcast_port_from_list_node(list->next);
551 static struct mcast_port_bundle *
552 mcast_snooping_port_lookup(struct ovs_list *list, void *port)
554 struct mcast_port_bundle *pbundle;
556 LIST_FOR_EACH (pbundle, node, list) {
557 if (pbundle->port == port) {
565 mcast_snooping_add_port(struct ovs_list *list, void *port)
567 struct mcast_port_bundle *pbundle;
569 pbundle = xmalloc(sizeof *pbundle);
570 pbundle->port = port;
571 list_insert(list, &pbundle->node);
575 mcast_snooping_flush_port(struct mcast_port_bundle *pbundle)
577 list_remove(&pbundle->node);
584 mcast_snooping_set_port_flood(struct mcast_snooping *ms, void *port,
586 OVS_REQ_WRLOCK(ms->rwlock)
588 struct mcast_port_bundle *fbundle;
590 fbundle = mcast_snooping_port_lookup(&ms->fport_list, port);
591 if (flood && !fbundle) {
592 mcast_snooping_add_port(&ms->fport_list, port);
593 ms->need_revalidate = true;
594 } else if (!flood && fbundle) {
595 mcast_snooping_flush_port(fbundle);
596 ms->need_revalidate = true;
600 /* Flood Reports ports. */
603 mcast_snooping_set_port_flood_reports(struct mcast_snooping *ms, void *port,
605 OVS_REQ_WRLOCK(ms->rwlock)
607 struct mcast_port_bundle *pbundle;
609 pbundle = mcast_snooping_port_lookup(&ms->rport_list, port);
610 if (flood && !pbundle) {
611 mcast_snooping_add_port(&ms->rport_list, port);
612 ms->need_revalidate = true;
613 } else if (!flood && pbundle) {
614 mcast_snooping_flush_port(pbundle);
615 ms->need_revalidate = true;
622 mcast_snooping_mdb_flush__(struct mcast_snooping *ms)
623 OVS_REQ_WRLOCK(ms->rwlock)
625 struct mcast_group *grp;
626 struct mcast_mrouter_bundle *mrouter;
628 while (group_get_lru(ms, &grp)) {
629 mcast_snooping_flush_group(ms, grp);
632 hmap_shrink(&ms->table);
634 while (mrouter_get_lru(ms, &mrouter)) {
635 mcast_snooping_flush_mrouter(mrouter);
640 mcast_snooping_mdb_flush(struct mcast_snooping *ms)
642 if (!mcast_snooping_enabled(ms)) {
646 ovs_rwlock_wrlock(&ms->rwlock);
647 mcast_snooping_mdb_flush__(ms);
648 ovs_rwlock_unlock(&ms->rwlock);
651 /* Flushes mdb and flood ports. */
653 mcast_snooping_flush__(struct mcast_snooping *ms)
654 OVS_REQ_WRLOCK(ms->rwlock)
656 struct mcast_group *grp;
657 struct mcast_mrouter_bundle *mrouter;
658 struct mcast_port_bundle *pbundle;
660 while (group_get_lru(ms, &grp)) {
661 mcast_snooping_flush_group(ms, grp);
664 hmap_shrink(&ms->table);
666 /* flush multicast routers */
667 while (mrouter_get_lru(ms, &mrouter)) {
668 mcast_snooping_flush_mrouter(mrouter);
671 /* flush flood ports */
672 while (mcast_snooping_port_get(&ms->fport_list, &pbundle)) {
673 mcast_snooping_flush_port(pbundle);
676 /* flush flood report ports */
677 while (mcast_snooping_port_get(&ms->rport_list, &pbundle)) {
678 mcast_snooping_flush_port(pbundle);
683 mcast_snooping_flush(struct mcast_snooping *ms)
685 if (!mcast_snooping_enabled(ms)) {
689 ovs_rwlock_wrlock(&ms->rwlock);
690 mcast_snooping_flush__(ms);
691 ovs_rwlock_unlock(&ms->rwlock);
695 mcast_snooping_run__(struct mcast_snooping *ms)
696 OVS_REQ_WRLOCK(ms->rwlock)
698 bool need_revalidate;
699 struct mcast_group *grp;
700 struct mcast_mrouter_bundle *mrouter;
703 while (group_get_lru(ms, &grp)) {
704 if (hmap_count(&ms->table) > ms->max_entries) {
705 mcast_snooping_flush_group(ms, grp);
707 if (!mcast_snooping_prune_expired(ms, grp)) {
713 hmap_shrink(&ms->table);
716 while (mrouter_get_lru(ms, &mrouter)
717 && time_now() >= mrouter->expires) {
718 mcast_snooping_flush_mrouter(mrouter);
722 if (mrouter_expired) {
723 ms->need_revalidate = true;
724 COVERAGE_ADD(mcast_snooping_expired, mrouter_expired);
727 need_revalidate = ms->need_revalidate;
728 ms->need_revalidate = false;
729 return need_revalidate;
732 /* Does periodic work required by 'ms'. Returns true if something changed
733 * that may require flow revalidation. */
735 mcast_snooping_run(struct mcast_snooping *ms)
737 bool need_revalidate;
739 if (!mcast_snooping_enabled(ms)) {
743 ovs_rwlock_wrlock(&ms->rwlock);
744 need_revalidate = mcast_snooping_run__(ms);
745 ovs_rwlock_unlock(&ms->rwlock);
747 return need_revalidate;
751 mcast_snooping_wait__(struct mcast_snooping *ms)
752 OVS_REQ_RDLOCK(ms->rwlock)
754 if (hmap_count(&ms->table) > ms->max_entries
755 || ms->need_revalidate) {
756 poll_immediate_wake();
758 struct mcast_group *grp;
759 struct mcast_group_bundle *bundle;
760 struct mcast_mrouter_bundle *mrouter;
761 long long int mrouter_msec;
762 long long int msec = 0;
764 if (!list_is_empty(&ms->group_lru)) {
765 grp = mcast_group_from_lru_node(ms->group_lru.next);
766 bundle = mcast_group_bundle_from_lru_node(grp->bundle_lru.next);
767 msec = bundle->expires * 1000LL;
770 if (!list_is_empty(&ms->mrouter_lru)) {
771 mrouter = mcast_mrouter_from_lru_node(ms->mrouter_lru.next);
772 mrouter_msec = mrouter->expires * 1000LL;
773 msec = msec ? MIN(msec, mrouter_msec) : mrouter_msec;
777 poll_timer_wait_until(msec);
783 mcast_snooping_wait(struct mcast_snooping *ms)
785 if (!mcast_snooping_enabled(ms)) {
789 ovs_rwlock_rdlock(&ms->rwlock);
790 mcast_snooping_wait__(ms);
791 ovs_rwlock_unlock(&ms->rwlock);