2 * Copyright (c) 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
24 #include "byte-order.h"
25 #include "connectivity.h"
26 #include "dynamic-string.h"
33 #include "poll-loop.h"
41 VLOG_DEFINE_THIS_MODULE(cfm);
43 #define CFM_MAX_RMPS 256
45 /* Ethernet destination address of CCM packets. */
46 static const uint8_t eth_addr_ccm[6] = { 0x01, 0x80, 0xC2, 0x00, 0x00, 0x30 };
47 static const uint8_t eth_addr_ccm_x[6] = {
48 0x01, 0x23, 0x20, 0x00, 0x00, 0x30
51 #define ETH_TYPE_CFM 0x8902
53 /* A 'ccm' represents a Continuity Check Message from the 802.1ag
54 * specification. Continuity Check Messages are broadcast periodically so that
55 * hosts can determine whom they have connectivity to.
57 * The minimum length of a CCM as specified by IEEE 802.1ag is 75 bytes.
58 * Previous versions of Open vSwitch generated 74-byte CCM messages, so we
59 * accept such messages too. */
61 #define CCM_ACCEPT_LEN 74
62 #define CCM_MAID_LEN 48
63 #define CCM_OPCODE 1 /* CFM message opcode meaning CCM. */
64 #define CCM_RDI_MASK 0x80
65 #define CFM_HEALTH_INTERVAL 6
69 uint8_t mdlevel_version; /* MD Level and Version */
75 uint8_t maid[CCM_MAID_LEN];
77 /* Defined by ITU-T Y.1731 should be zero */
78 ovs_be16 interval_ms_x; /* Transmission interval in ms. */
79 ovs_be64 mpid64; /* MPID in extended mode. */
80 uint8_t opdown; /* Operationally down. */
86 BUILD_ASSERT_DECL(CCM_LEN == sizeof(struct ccm));
89 const char *name; /* Name of this CFM object. */
90 struct hmap_node hmap_node; /* Node in all_cfms list. */
92 struct netdev *netdev;
93 uint64_t rx_packets; /* Packets received by 'netdev'. */
96 bool demand; /* Demand mode. */
97 bool booted; /* A full fault interval has occurred. */
98 enum cfm_fault_reason fault; /* Connectivity fault status. */
99 enum cfm_fault_reason recv_fault; /* Bit mask of faults occurring on
101 bool opup; /* Operational State. */
102 bool remote_opup; /* Remote Operational State. */
104 int fault_override; /* Manual override of 'fault' status.
105 Ignored if negative. */
107 uint32_t seq; /* The sequence number of our last CCM. */
108 uint8_t ccm_interval; /* The CCM transmission interval. */
109 int ccm_interval_ms; /* 'ccm_interval' in milliseconds. */
110 uint16_t ccm_vlan; /* Vlan tag of CCM PDUs. CFM_RANDOM_VLAN if
112 uint8_t ccm_pcp; /* Priority of CCM PDUs. */
113 uint8_t maid[CCM_MAID_LEN]; /* The MAID of this CFM. */
115 struct timer tx_timer; /* Send CCM when expired. */
116 struct timer fault_timer; /* Check for faults when expired. */
118 struct hmap remote_mps; /* Remote MPs. */
120 /* Result of cfm_get_remote_mpids(). Updated only during fault check to
122 uint64_t *rmps_array; /* Cache of remote_mps. */
123 size_t rmps_array_len; /* Number of rmps in 'rmps_array'. */
125 int health; /* Percentage of the number of CCM frames
127 int health_interval; /* Number of fault_intervals since health was
129 long long int last_tx; /* Last CCM transmission time. */
131 atomic_bool check_tnl_key; /* Verify the tunnel key of inbound packets? */
132 atomic_bool extended; /* Extended mode. */
133 struct ovs_refcount ref_cnt;
135 uint64_t flap_count; /* Count the flaps since boot. */
137 /* True when the variables returned by cfm_get_*() are changed
138 * since last check. */
142 /* Remote MPs represent foreign network entities that are configured to have
143 * the same MAID as this CFM instance. */
145 uint64_t mpid; /* The Maintenance Point ID of this 'remote_mp'. */
146 struct hmap_node node; /* Node in 'remote_mps' map. */
148 bool recv; /* CCM was received since last fault check. */
149 bool opup; /* Operational State. */
150 uint32_t seq; /* Most recently received sequence number. */
151 uint8_t num_health_ccm; /* Number of received ccm frames every
152 CFM_HEALTH_INTERVAL * 'fault_interval'. */
153 long long int last_rx; /* Last CCM reception time. */
157 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 30);
159 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
160 static struct hmap all_cfms__ = HMAP_INITIALIZER(&all_cfms__);
161 static struct hmap *const all_cfms OVS_GUARDED_BY(mutex) = &all_cfms__;
163 static unixctl_cb_func cfm_unixctl_show;
164 static unixctl_cb_func cfm_unixctl_set_fault;
167 cfm_rx_packets(const struct cfm *cfm) OVS_REQUIRES(mutex)
169 struct netdev_stats stats;
171 if (!netdev_get_stats(cfm->netdev, &stats)) {
172 return stats.rx_packets;
178 static const uint8_t *
179 cfm_ccm_addr(struct cfm *cfm)
182 atomic_read(&cfm->extended, &extended);
183 return extended ? eth_addr_ccm_x : eth_addr_ccm;
186 /* Returns the string representation of the given cfm_fault_reason 'reason'. */
188 cfm_fault_reason_to_str(int reason)
191 #define CFM_FAULT_REASON(NAME, STR) case CFM_FAULT_##NAME: return #STR;
193 #undef CFM_FAULT_REASON
194 default: return "<unknown>";
199 ds_put_cfm_fault(struct ds *ds, int fault)
203 for (i = 0; i < CFM_FAULT_N_REASONS; i++) {
206 if (fault & reason) {
207 ds_put_format(ds, "%s ", cfm_fault_reason_to_str(reason));
215 cfm_generate_maid(struct cfm *cfm) OVS_REQUIRES(mutex)
217 const char *ovs_md_name = "ovs";
218 const char *ovs_ma_name = "ovs";
220 size_t md_len, ma_len;
222 memset(cfm->maid, 0, CCM_MAID_LEN);
224 md_len = strlen(ovs_md_name);
225 ma_len = strlen(ovs_ma_name);
227 ovs_assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN);
229 cfm->maid[0] = 4; /* MD name string format. */
230 cfm->maid[1] = md_len; /* MD name size. */
231 memcpy(&cfm->maid[2], ovs_md_name, md_len); /* MD name. */
233 ma_p = cfm->maid + 2 + md_len;
234 ma_p[0] = 2; /* MA name string format. */
235 ma_p[1] = ma_len; /* MA name size. */
236 memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */
240 ccm_interval_to_ms(uint8_t interval)
243 case 0: OVS_NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
244 case 1: return 3; /* Not recommended due to timer resolution. */
245 case 2: return 10; /* Not recommended due to timer resolution. */
248 case 5: return 10000;
249 case 6: return 60000;
250 case 7: return 600000;
251 default: OVS_NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
258 cfm_fault_interval(struct cfm *cfm) OVS_REQUIRES(mutex)
260 /* According to the 802.1ag specification we should assume every other MP
261 * with the same MAID has the same transmission interval that we have. If
262 * an MP has a different interval, cfm_process_heartbeat will register it
263 * as a fault (likely due to a configuration error). Thus we can check all
264 * MPs at once making this quite a bit simpler.
266 * When cfm is not in demand mode, we check when (ccm_interval_ms * 3.5) ms
267 * have passed. When cfm is in demand mode, we check when
268 * (MAX(ccm_interval_ms, 500) * 3.5) ms have passed. This ensures that
269 * ovs-vswitchd has enough time to pull statistics from the datapath. */
271 return (MAX(cfm->ccm_interval_ms, cfm->demand ? 500 : cfm->ccm_interval_ms)
276 ms_to_ccm_interval(int interval_ms)
280 for (i = 7; i > 0; i--) {
281 if (ccm_interval_to_ms(i) <= interval_ms) {
290 hash_mpid(uint64_t mpid)
292 return hash_uint64(mpid);
296 cfm_is_valid_mpid(bool extended, uint64_t mpid)
298 /* 802.1ag specification requires MPIDs to be within the range [1, 8191].
299 * In extended mode we relax this requirement. */
300 return mpid >= 1 && (extended || mpid <= 8191);
303 static struct remote_mp *
304 lookup_remote_mp(const struct cfm *cfm, uint64_t mpid) OVS_REQUIRES(mutex)
306 struct remote_mp *rmp;
308 HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), &cfm->remote_mps) {
309 if (rmp->mpid == mpid) {
320 unixctl_command_register("cfm/show", "[interface]", 0, 1, cfm_unixctl_show,
322 unixctl_command_register("cfm/set-fault", "[interface] normal|false|true",
323 1, 2, cfm_unixctl_set_fault, NULL);
326 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
327 * cfm_configure() before use. */
329 cfm_create(const struct netdev *netdev) OVS_EXCLUDED(mutex)
333 cfm = xzalloc(sizeof *cfm);
334 cfm->netdev = netdev_ref(netdev);
335 cfm->name = netdev_get_name(cfm->netdev);
336 hmap_init(&cfm->remote_mps);
337 cfm->remote_opup = true;
338 cfm->fault_override = -1;
342 atomic_init(&cfm->extended, false);
343 atomic_init(&cfm->check_tnl_key, false);
344 ovs_refcount_init(&cfm->ref_cnt);
346 ovs_mutex_lock(&mutex);
347 cfm_generate_maid(cfm);
348 hmap_insert(all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0));
349 ovs_mutex_unlock(&mutex);
355 cfm_unref(struct cfm *cfm) OVS_EXCLUDED(mutex)
357 struct remote_mp *rmp, *rmp_next;
363 if (ovs_refcount_unref(&cfm->ref_cnt) != 1) {
367 ovs_mutex_lock(&mutex);
368 hmap_remove(all_cfms, &cfm->hmap_node);
369 ovs_mutex_unlock(&mutex);
371 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
372 hmap_remove(&cfm->remote_mps, &rmp->node);
376 hmap_destroy(&cfm->remote_mps);
377 netdev_close(cfm->netdev);
378 free(cfm->rmps_array);
384 cfm_ref(const struct cfm *cfm_)
386 struct cfm *cfm = CONST_CAST(struct cfm *, cfm_);
388 ovs_refcount_ref(&cfm->ref_cnt);
393 /* Records the status change and changes the global connectivity seq. */
395 cfm_status_changed(struct cfm *cfm) OVS_REQUIRES(mutex)
397 seq_change(connectivity_seq_get());
398 cfm->status_changed = true;
401 /* Should be run periodically to update fault statistics messages. */
403 cfm_run(struct cfm *cfm) OVS_EXCLUDED(mutex)
405 ovs_mutex_lock(&mutex);
406 if (timer_expired(&cfm->fault_timer)) {
407 long long int interval = cfm_fault_interval(cfm);
408 struct remote_mp *rmp, *rmp_next;
409 enum cfm_fault_reason old_cfm_fault = cfm->fault;
410 uint64_t old_flap_count = cfm->flap_count;
411 int old_health = cfm->health;
412 size_t old_rmps_array_len = cfm->rmps_array_len;
413 bool old_rmps_deleted = false;
414 bool old_rmp_opup = cfm->remote_opup;
415 bool demand_override;
416 bool rmp_set_opup = false;
417 bool rmp_set_opdown = false;
419 cfm->fault = cfm->recv_fault;
422 cfm->rmps_array_len = 0;
423 free(cfm->rmps_array);
424 cfm->rmps_array = xmalloc(hmap_count(&cfm->remote_mps) *
425 sizeof *cfm->rmps_array);
427 if (cfm->health_interval == CFM_HEALTH_INTERVAL) {
428 /* Calculate the cfm health of the interface. If the number of
429 * remote_mpids of a cfm interface is > 1, the cfm health is
430 * undefined. If the number of remote_mpids is 1, the cfm health is
431 * the percentage of the ccm frames received in the
432 * (CFM_HEALTH_INTERVAL * 3.5)ms, else it is 0. */
433 if (hmap_count(&cfm->remote_mps) > 1) {
435 } else if (hmap_is_empty(&cfm->remote_mps)) {
440 rmp = CONTAINER_OF(hmap_first(&cfm->remote_mps),
441 struct remote_mp, node);
442 exp_ccm_recvd = (CFM_HEALTH_INTERVAL * 7) / 2;
443 /* Calculate the percentage of healthy ccm frames received.
444 * Since the 'fault_interval' is (3.5 * cfm_interval), and
445 * 1 CCM packet must be received every cfm_interval,
446 * the 'remote_mpid' health reports the percentage of
447 * healthy CCM frames received every
448 * 'CFM_HEALTH_INTERVAL'th 'fault_interval'. */
449 cfm->health = (rmp->num_health_ccm * 100) / exp_ccm_recvd;
450 cfm->health = MIN(cfm->health, 100);
451 rmp->num_health_ccm = 0;
452 ovs_assert(cfm->health >= 0 && cfm->health <= 100);
454 cfm->health_interval = 0;
456 cfm->health_interval++;
458 demand_override = false;
460 uint64_t rx_packets = cfm_rx_packets(cfm);
461 demand_override = hmap_count(&cfm->remote_mps) == 1
462 && rx_packets > cfm->rx_packets;
463 cfm->rx_packets = rx_packets;
466 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
468 VLOG_INFO("%s: Received no CCM from RMP %"PRIu64" in the last"
469 " %lldms", cfm->name, rmp->mpid,
470 time_msec() - rmp->last_rx);
471 if (!demand_override) {
472 old_rmps_deleted = true;
473 hmap_remove(&cfm->remote_mps, &rmp->node);
482 rmp_set_opdown = true;
485 cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid;
489 if (rmp_set_opdown) {
490 cfm->remote_opup = false;
492 else if (rmp_set_opup) {
493 cfm->remote_opup = true;
496 if (hmap_is_empty(&cfm->remote_mps)) {
497 cfm->fault |= CFM_FAULT_RECV;
500 if (old_cfm_fault != cfm->fault) {
501 if (!VLOG_DROP_INFO(&rl)) {
502 struct ds ds = DS_EMPTY_INITIALIZER;
504 ds_put_cstr(&ds, "from [");
505 ds_put_cfm_fault(&ds, old_cfm_fault);
506 ds_put_cstr(&ds, "] to [");
507 ds_put_cfm_fault(&ds, cfm->fault);
508 ds_put_char(&ds, ']');
509 VLOG_INFO("%s: CFM faults changed %s.", cfm->name, ds_cstr(&ds));
513 /* If there is a flap, increments the counter. */
514 if (old_cfm_fault == 0 || cfm->fault == 0) {
519 /* These variables represent the cfm session status, it is desirable
520 * to update them to database immediately after change. */
521 if (old_health != cfm->health
522 || old_rmp_opup != cfm->remote_opup
523 || (old_rmps_array_len != cfm->rmps_array_len || old_rmps_deleted)
524 || old_cfm_fault != cfm->fault
525 || old_flap_count != cfm->flap_count) {
526 cfm_status_changed(cfm);
530 timer_set_duration(&cfm->fault_timer, interval);
531 VLOG_DBG("%s: new fault interval", cfm->name);
533 ovs_mutex_unlock(&mutex);
536 /* Should be run periodically to check if the CFM module has a CCM message it
539 cfm_should_send_ccm(struct cfm *cfm) OVS_EXCLUDED(mutex)
543 ovs_mutex_lock(&mutex);
544 ret = timer_expired(&cfm->tx_timer);
545 ovs_mutex_unlock(&mutex);
549 /* Composes a CCM message into 'packet'. Messages generated with this function
550 * should be sent whenever cfm_should_send_ccm() indicates. */
552 cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet,
553 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
559 ovs_mutex_lock(&mutex);
560 timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms);
561 eth_compose(packet, cfm_ccm_addr(cfm), eth_src, ETH_TYPE_CFM, sizeof *ccm);
563 ccm_vlan = (cfm->ccm_vlan != CFM_RANDOM_VLAN
566 ccm_vlan = ccm_vlan & VLAN_VID_MASK;
568 if (ccm_vlan || cfm->ccm_pcp) {
569 uint16_t tci = ccm_vlan | (cfm->ccm_pcp << VLAN_PCP_SHIFT);
570 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), htons(tci));
573 ccm = ofpbuf_l3(packet);
574 ccm->mdlevel_version = 0;
575 ccm->opcode = CCM_OPCODE;
576 ccm->tlv_offset = 70;
577 ccm->seq = htonl(++cfm->seq);
578 ccm->flags = cfm->ccm_interval;
579 memcpy(ccm->maid, cfm->maid, sizeof ccm->maid);
580 memset(ccm->zero, 0, sizeof ccm->zero);
583 atomic_read(&cfm->extended, &extended);
585 ccm->mpid = htons(hash_mpid(cfm->mpid));
586 ccm->mpid64 = htonll(cfm->mpid);
587 ccm->opdown = !cfm->opup;
589 ccm->mpid = htons(cfm->mpid);
590 ccm->mpid64 = htonll(0);
594 if (cfm->ccm_interval == 0) {
595 ovs_assert(extended);
596 ccm->interval_ms_x = htons(cfm->ccm_interval_ms);
598 ccm->interval_ms_x = htons(0);
601 if (cfm->booted && hmap_is_empty(&cfm->remote_mps)) {
602 ccm->flags |= CCM_RDI_MASK;
606 long long int delay = time_msec() - cfm->last_tx;
607 if (delay > (cfm->ccm_interval_ms * 3 / 2)) {
608 VLOG_WARN("%s: long delay of %lldms (expected %dms) sending CCM"
609 " seq %"PRIu32, cfm->name, delay, cfm->ccm_interval_ms,
613 cfm->last_tx = time_msec();
614 ovs_mutex_unlock(&mutex);
618 cfm_wait(struct cfm *cfm) OVS_EXCLUDED(mutex)
620 poll_timer_wait_until(cfm_wake_time(cfm));
624 /* Returns the next cfm wakeup time. */
626 cfm_wake_time(struct cfm *cfm) OVS_EXCLUDED(mutex)
628 long long int retval;
634 ovs_mutex_lock(&mutex);
635 retval = MIN(cfm->tx_timer.t, cfm->fault_timer.t);
636 ovs_mutex_unlock(&mutex);
641 /* Configures 'cfm' with settings from 's'. */
643 cfm_configure(struct cfm *cfm, const struct cfm_settings *s)
649 if (!cfm_is_valid_mpid(s->extended, s->mpid) || s->interval <= 0) {
653 ovs_mutex_lock(&mutex);
656 interval = ms_to_ccm_interval(s->interval);
657 interval_ms = ccm_interval_to_ms(interval);
659 atomic_store(&cfm->check_tnl_key, s->check_tnl_key);
660 atomic_store(&cfm->extended, s->extended);
662 cfm->ccm_vlan = s->ccm_vlan;
663 cfm->ccm_pcp = s->ccm_pcp & (VLAN_PCP_MASK >> VLAN_PCP_SHIFT);
664 if (s->extended && interval_ms != s->interval) {
666 interval_ms = MIN(s->interval, UINT16_MAX);
669 if (s->extended && s->demand) {
672 cfm->rx_packets = cfm_rx_packets(cfm);
678 if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) {
679 cfm->ccm_interval = interval;
680 cfm->ccm_interval_ms = interval_ms;
682 timer_set_expired(&cfm->tx_timer);
683 timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm));
686 ovs_mutex_unlock(&mutex);
690 /* Must be called when the netdev owned by 'cfm' should change. */
692 cfm_set_netdev(struct cfm *cfm, const struct netdev *netdev)
695 ovs_mutex_lock(&mutex);
696 if (cfm->netdev != netdev) {
697 netdev_close(cfm->netdev);
698 cfm->netdev = netdev_ref(netdev);
700 ovs_mutex_unlock(&mutex);
703 /* Returns true if 'cfm' should process packets from 'flow'. Sets
704 * fields in 'wc' that were used to make the determination. */
706 cfm_should_process_flow(const struct cfm *cfm_, const struct flow *flow,
707 struct flow_wildcards *wc)
709 struct cfm *cfm = CONST_CAST(struct cfm *, cfm_);
712 atomic_read(&cfm->check_tnl_key, &check_tnl_key);
713 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
715 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
717 return (ntohs(flow->dl_type) == ETH_TYPE_CFM
718 && eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm))
719 && (!check_tnl_key || flow->tunnel.tun_id == htonll(0)));
722 /* Updates internal statistics relevant to packet 'p'. Should be called on
723 * every packet whose flow returned true when passed to
724 * cfm_should_process_flow. */
726 cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
730 struct eth_header *eth;
732 ovs_mutex_lock(&mutex);
735 ccm = ofpbuf_at(p, (uint8_t *)ofpbuf_l3(p) - (uint8_t *)ofpbuf_data(p),
739 VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.",
744 if (ccm->opcode != CCM_OPCODE) {
745 VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. "
746 "(opcode %u)", cfm->name, ccm->opcode);
750 /* According to the 802.1ag specification, reception of a CCM with an
751 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
752 * trigger a fault. We ignore this requirement for several reasons.
754 * Faults can cause a controller or Open vSwitch to make potentially
755 * expensive changes to the network topology. It seems prudent to trigger
756 * them judiciously, especially when CFM is used to check slave status of
757 * bonds. Furthermore, faults can be maliciously triggered by crafting
758 * unexpected CCMs. */
759 if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) {
760 cfm->recv_fault |= CFM_FAULT_MAID;
761 VLOG_WARN_RL(&rl, "%s: Received unexpected remote MAID from MAC "
762 ETH_ADDR_FMT, cfm->name, ETH_ADDR_ARGS(eth->eth_src));
764 uint8_t ccm_interval = ccm->flags & 0x7;
765 bool ccm_rdi = ccm->flags & CCM_RDI_MASK;
766 uint16_t ccm_interval_ms_x = ntohs(ccm->interval_ms_x);
768 struct remote_mp *rmp;
773 enum cfm_fault_reason cfm_fault = 0;
775 atomic_read(&cfm->extended, &extended);
777 ccm_mpid = ntohll(ccm->mpid64);
778 ccm_opdown = ccm->opdown;
780 ccm_mpid = ntohs(ccm->mpid);
783 ccm_seq = ntohl(ccm->seq);
785 if (ccm_interval != cfm->ccm_interval) {
786 VLOG_WARN_RL(&rl, "%s: received a CCM with an unexpected interval"
787 " (%"PRIu8") from RMP %"PRIu64, cfm->name,
788 ccm_interval, ccm_mpid);
791 if (extended && ccm_interval == 0
792 && ccm_interval_ms_x != cfm->ccm_interval_ms) {
793 VLOG_WARN_RL(&rl, "%s: received a CCM with an unexpected extended"
794 " interval (%"PRIu16"ms) from RMP %"PRIu64, cfm->name,
795 ccm_interval_ms_x, ccm_mpid);
798 rmp = lookup_remote_mp(cfm, ccm_mpid);
800 if (hmap_count(&cfm->remote_mps) < CFM_MAX_RMPS) {
801 rmp = xzalloc(sizeof *rmp);
802 hmap_insert(&cfm->remote_mps, &rmp->node, hash_mpid(ccm_mpid));
804 cfm_fault |= CFM_FAULT_OVERFLOW;
806 "%s: dropped CCM with MPID %"PRIu64" from MAC "
807 ETH_ADDR_FMT, cfm->name, ccm_mpid,
808 ETH_ADDR_ARGS(eth->eth_src));
813 cfm_fault |= CFM_FAULT_RDI;
814 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64, cfm->name,
818 VLOG_DBG("%s: received CCM (seq %"PRIu32") (mpid %"PRIu64")"
819 " (interval %"PRIu8") (RDI %s)", cfm->name, ccm_seq,
820 ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false");
823 if (rmp->mpid == cfm->mpid) {
824 cfm_fault |= CFM_FAULT_LOOPBACK;
825 VLOG_WARN_RL(&rl,"%s: received CCM with local MPID"
826 " %"PRIu64, cfm->name, rmp->mpid);
829 if (rmp->seq && ccm_seq != (rmp->seq + 1)) {
830 VLOG_WARN_RL(&rl, "%s: (mpid %"PRIu64") detected sequence"
831 " numbers which indicate possible connectivity"
832 " problems (previous %"PRIu32") (current %"PRIu32
833 ")", cfm->name, ccm_mpid, rmp->seq, ccm_seq);
836 rmp->mpid = ccm_mpid;
838 rmp->num_health_ccm++;
841 cfm->recv_fault |= cfm_fault;
843 rmp->opup = !ccm_opdown;
844 rmp->last_rx = time_msec();
849 ovs_mutex_unlock(&mutex);
852 /* Returns and resets the 'cfm->status_changed'. */
854 cfm_check_status_change(struct cfm *cfm) OVS_EXCLUDED(mutex)
858 ovs_mutex_lock(&mutex);
859 ret = cfm->status_changed;
860 cfm->status_changed = false;
861 ovs_mutex_unlock(&mutex);
867 cfm_get_fault__(const struct cfm *cfm) OVS_REQUIRES(mutex)
869 if (cfm->fault_override >= 0) {
870 return cfm->fault_override ? CFM_FAULT_OVERRIDE : 0;
875 /* Gets the fault status of 'cfm'. Returns a bit mask of 'cfm_fault_reason's
876 * indicating the cause of the connectivity fault, or zero if there is no
879 cfm_get_fault(const struct cfm *cfm) OVS_EXCLUDED(mutex)
883 ovs_mutex_lock(&mutex);
884 fault = cfm_get_fault__(cfm);
885 ovs_mutex_unlock(&mutex);
889 /* Gets the number of cfm fault flapping since start. */
891 cfm_get_flap_count(const struct cfm *cfm) OVS_EXCLUDED(mutex)
894 ovs_mutex_lock(&mutex);
895 flap_count = cfm->flap_count;
896 ovs_mutex_unlock(&mutex);
900 /* Gets the health of 'cfm'. Returns an integer between 0 and 100 indicating
901 * the health of the link as a percentage of ccm frames received in
902 * CFM_HEALTH_INTERVAL * 'fault_interval' if there is only 1 remote_mpid,
903 * returns 0 if there are no remote_mpids, and returns -1 if there are more
904 * than 1 remote_mpids. */
906 cfm_get_health(const struct cfm *cfm) OVS_EXCLUDED(mutex)
910 ovs_mutex_lock(&mutex);
911 health = cfm->health;
912 ovs_mutex_unlock(&mutex);
916 /* Gets the operational state of 'cfm'. 'cfm' is considered operationally down
917 * if it has received a CCM with the operationally down bit set from any of its
918 * remote maintenance points. Returns 1 if 'cfm' is operationally up, 0 if
919 * 'cfm' is operationally down, or -1 if 'cfm' has no operational state
920 * (because it isn't in extended mode). */
922 cfm_get_opup(const struct cfm *cfm_) OVS_EXCLUDED(mutex)
924 struct cfm *cfm = CONST_CAST(struct cfm *, cfm_);
928 ovs_mutex_lock(&mutex);
929 atomic_read(&cfm->extended, &extended);
930 opup = extended ? cfm->remote_opup : -1;
931 ovs_mutex_unlock(&mutex);
936 /* Populates 'rmps' with an array of remote maintenance points reachable by
937 * 'cfm'. The number of remote maintenance points is written to 'n_rmps'.
938 * 'cfm' retains ownership of the array written to 'rmps' */
940 cfm_get_remote_mpids(const struct cfm *cfm, uint64_t **rmps, size_t *n_rmps)
943 ovs_mutex_lock(&mutex);
944 *rmps = xmemdup(cfm->rmps_array, cfm->rmps_array_len * sizeof **rmps);
945 *n_rmps = cfm->rmps_array_len;
946 ovs_mutex_unlock(&mutex);
950 cfm_find(const char *name) OVS_REQUIRES(mutex)
954 HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), all_cfms) {
955 if (!strcmp(cfm->name, name)) {
963 cfm_print_details(struct ds *ds, struct cfm *cfm) OVS_REQUIRES(mutex)
965 struct remote_mp *rmp;
969 atomic_read(&cfm->extended, &extended);
971 ds_put_format(ds, "---- %s ----\n", cfm->name);
972 ds_put_format(ds, "MPID %"PRIu64":%s%s\n", cfm->mpid,
973 extended ? " extended" : "",
974 cfm->fault_override >= 0 ? " fault_override" : "");
976 fault = cfm_get_fault__(cfm);
978 ds_put_cstr(ds, "\tfault: ");
979 ds_put_cfm_fault(ds, fault);
980 ds_put_cstr(ds, "\n");
983 if (cfm->health == -1) {
984 ds_put_format(ds, "\taverage health: undefined\n");
986 ds_put_format(ds, "\taverage health: %d\n", cfm->health);
988 ds_put_format(ds, "\topstate: %s\n", cfm->opup ? "up" : "down");
989 ds_put_format(ds, "\tremote_opstate: %s\n",
990 cfm->remote_opup ? "up" : "down");
991 ds_put_format(ds, "\tinterval: %dms\n", cfm->ccm_interval_ms);
992 ds_put_format(ds, "\tnext CCM tx: %lldms\n",
993 timer_msecs_until_expired(&cfm->tx_timer));
994 ds_put_format(ds, "\tnext fault check: %lldms\n",
995 timer_msecs_until_expired(&cfm->fault_timer));
997 HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
998 ds_put_format(ds, "Remote MPID %"PRIu64"\n", rmp->mpid);
999 ds_put_format(ds, "\trecv since check: %s\n",
1000 rmp->recv ? "true" : "false");
1001 ds_put_format(ds, "\topstate: %s\n", rmp->opup? "up" : "down");
1006 cfm_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1007 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1009 struct ds ds = DS_EMPTY_INITIALIZER;
1012 ovs_mutex_lock(&mutex);
1014 cfm = cfm_find(argv[1]);
1016 unixctl_command_reply_error(conn, "no such CFM object");
1019 cfm_print_details(&ds, cfm);
1021 HMAP_FOR_EACH (cfm, hmap_node, all_cfms) {
1022 cfm_print_details(&ds, cfm);
1026 unixctl_command_reply(conn, ds_cstr(&ds));
1029 ovs_mutex_unlock(&mutex);
1033 cfm_unixctl_set_fault(struct unixctl_conn *conn, int argc, const char *argv[],
1034 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1036 const char *fault_str = argv[argc - 1];
1040 ovs_mutex_lock(&mutex);
1041 if (!strcasecmp("true", fault_str)) {
1043 } else if (!strcasecmp("false", fault_str)) {
1045 } else if (!strcasecmp("normal", fault_str)) {
1046 fault_override = -1;
1048 unixctl_command_reply_error(conn, "unknown fault string");
1053 cfm = cfm_find(argv[1]);
1055 unixctl_command_reply_error(conn, "no such CFM object");
1058 cfm->fault_override = fault_override;
1059 cfm_status_changed(cfm);
1061 HMAP_FOR_EACH (cfm, hmap_node, all_cfms) {
1062 cfm->fault_override = fault_override;
1063 cfm_status_changed(cfm);
1067 unixctl_command_reply(conn, "OK");
1070 ovs_mutex_unlock(&mutex);