2 * Copyright (c) 2010, 2011, 2012, 2013, 2014, 2015 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 "dp-packet.h"
27 #include "dynamic-string.h"
32 #include "ovs-atomic.h"
34 #include "poll-loop.h"
40 #include "openvswitch/vlog.h"
42 VLOG_DEFINE_THIS_MODULE(cfm);
44 #define CFM_MAX_RMPS 256
46 /* Ethernet destination address of CCM packets. */
47 static const uint8_t eth_addr_ccm[ETH_ADDR_LEN] = {
48 0x01, 0x80, 0xC2, 0x00, 0x00, 0x30 };
49 static const uint8_t eth_addr_ccm_x[ETH_ADDR_LEN] = {
50 0x01, 0x23, 0x20, 0x00, 0x00, 0x30
53 #define ETH_TYPE_CFM 0x8902
55 /* A 'ccm' represents a Continuity Check Message from the 802.1ag
56 * specification. Continuity Check Messages are broadcast periodically so that
57 * hosts can determine whom they have connectivity to.
59 * The minimum length of a CCM as specified by IEEE 802.1ag is 75 bytes.
60 * Previous versions of Open vSwitch generated 74-byte CCM messages, so we
61 * accept such messages too. */
63 #define CCM_ACCEPT_LEN 74
64 #define CCM_MAID_LEN 48
65 #define CCM_OPCODE 1 /* CFM message opcode meaning CCM. */
66 #define CCM_RDI_MASK 0x80
67 #define CFM_HEALTH_INTERVAL 6
71 uint8_t mdlevel_version; /* MD Level and Version */
77 uint8_t maid[CCM_MAID_LEN];
79 /* Defined by ITU-T Y.1731 should be zero */
80 ovs_be16 interval_ms_x; /* Transmission interval in ms. */
81 ovs_be64 mpid64; /* MPID in extended mode. */
82 uint8_t opdown; /* Operationally down. */
88 BUILD_ASSERT_DECL(CCM_LEN == sizeof(struct ccm));
91 const char *name; /* Name of this CFM object. */
92 struct hmap_node hmap_node; /* Node in all_cfms list. */
94 struct netdev *netdev;
95 uint64_t rx_packets; /* Packets received by 'netdev'. */
98 bool demand; /* Demand mode. */
99 bool booted; /* A full fault interval has occurred. */
100 enum cfm_fault_reason fault; /* Connectivity fault status. */
101 enum cfm_fault_reason recv_fault; /* Bit mask of faults occurring on
103 bool opup; /* Operational State. */
104 bool remote_opup; /* Remote Operational State. */
106 int fault_override; /* Manual override of 'fault' status.
107 Ignored if negative. */
109 uint32_t seq; /* The sequence number of our last CCM. */
110 uint8_t ccm_interval; /* The CCM transmission interval. */
111 int ccm_interval_ms; /* 'ccm_interval' in milliseconds. */
112 uint16_t ccm_vlan; /* Vlan tag of CCM PDUs. CFM_RANDOM_VLAN if
114 uint8_t ccm_pcp; /* Priority of CCM PDUs. */
115 uint8_t maid[CCM_MAID_LEN]; /* The MAID of this CFM. */
117 struct timer tx_timer; /* Send CCM when expired. */
118 struct timer fault_timer; /* Check for faults when expired. */
120 struct hmap remote_mps; /* Remote MPs. */
122 /* Result of cfm_get_remote_mpids(). Updated only during fault check to
124 uint64_t *rmps_array; /* Cache of remote_mps. */
125 size_t rmps_array_len; /* Number of rmps in 'rmps_array'. */
127 int health; /* Percentage of the number of CCM frames
129 int health_interval; /* Number of fault_intervals since health was
131 long long int last_tx; /* Last CCM transmission time. */
133 /* These bools are atomic to allow readers to check their values
134 * without taking 'mutex'. Such readers do not assume the values they
135 * read are synchronized with any other members. */
136 atomic_bool check_tnl_key; /* Verify the tunnel key of inbound packets? */
137 atomic_bool extended; /* Extended mode. */
138 struct ovs_refcount ref_cnt;
140 uint64_t flap_count; /* Count the flaps since boot. */
142 /* True when the variables returned by cfm_get_*() are changed
143 * since last check. */
146 /* When 'cfm->demand' is set, at least one ccm is required to be received
147 * every 100 * cfm_interval. If ccm is not received within this interval,
148 * even if data packets are received, the cfm fault will be set. */
149 struct timer demand_rx_ccm_t;
152 /* Remote MPs represent foreign network entities that are configured to have
153 * the same MAID as this CFM instance. */
155 uint64_t mpid; /* The Maintenance Point ID of this 'remote_mp'. */
156 struct hmap_node node; /* Node in 'remote_mps' map. */
158 bool recv; /* CCM was received since last fault check. */
159 bool opup; /* Operational State. */
160 uint32_t seq; /* Most recently received sequence number. */
161 uint8_t num_health_ccm; /* Number of received ccm frames every
162 CFM_HEALTH_INTERVAL * 'fault_interval'. */
163 long long int last_rx; /* Last CCM reception time. */
167 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 30);
169 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
170 static struct hmap all_cfms__ = HMAP_INITIALIZER(&all_cfms__);
171 static struct hmap *const all_cfms OVS_GUARDED_BY(mutex) = &all_cfms__;
173 static unixctl_cb_func cfm_unixctl_show;
174 static unixctl_cb_func cfm_unixctl_set_fault;
177 cfm_rx_packets(const struct cfm *cfm) OVS_REQUIRES(mutex)
179 struct netdev_stats stats;
181 if (!netdev_get_stats(cfm->netdev, &stats)) {
182 return stats.rx_packets;
188 static const uint8_t *
189 cfm_ccm_addr(struct cfm *cfm)
193 atomic_read_relaxed(&cfm->extended, &extended);
195 return extended ? eth_addr_ccm_x : eth_addr_ccm;
198 /* Returns the string representation of the given cfm_fault_reason 'reason'. */
200 cfm_fault_reason_to_str(int reason)
203 #define CFM_FAULT_REASON(NAME, STR) case CFM_FAULT_##NAME: return #STR;
205 #undef CFM_FAULT_REASON
206 default: return "<unknown>";
211 ds_put_cfm_fault(struct ds *ds, int fault)
215 for (i = 0; i < CFM_FAULT_N_REASONS; i++) {
218 if (fault & reason) {
219 ds_put_format(ds, "%s ", cfm_fault_reason_to_str(reason));
227 cfm_generate_maid(struct cfm *cfm) OVS_REQUIRES(mutex)
229 const char *ovs_md_name = "ovs";
230 const char *ovs_ma_name = "ovs";
232 size_t md_len, ma_len;
234 memset(cfm->maid, 0, CCM_MAID_LEN);
236 md_len = strlen(ovs_md_name);
237 ma_len = strlen(ovs_ma_name);
239 ovs_assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN);
241 cfm->maid[0] = 4; /* MD name string format. */
242 cfm->maid[1] = md_len; /* MD name size. */
243 memcpy(&cfm->maid[2], ovs_md_name, md_len); /* MD name. */
245 ma_p = cfm->maid + 2 + md_len;
246 ma_p[0] = 2; /* MA name string format. */
247 ma_p[1] = ma_len; /* MA name size. */
248 memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */
252 ccm_interval_to_ms(uint8_t interval)
255 case 0: OVS_NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
256 case 1: return 3; /* Not recommended due to timer resolution. */
257 case 2: return 10; /* Not recommended due to timer resolution. */
260 case 5: return 10000;
261 case 6: return 60000;
262 case 7: return 600000;
263 default: OVS_NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
270 cfm_fault_interval(struct cfm *cfm) OVS_REQUIRES(mutex)
272 /* According to the 802.1ag specification we should assume every other MP
273 * with the same MAID has the same transmission interval that we have. If
274 * an MP has a different interval, cfm_process_heartbeat will register it
275 * as a fault (likely due to a configuration error). Thus we can check all
276 * MPs at once making this quite a bit simpler.
278 * When cfm is not in demand mode, we check when (ccm_interval_ms * 3.5) ms
279 * have passed. When cfm is in demand mode, we check when
280 * (MAX(ccm_interval_ms, 500) * 3.5) ms have passed. This ensures that
281 * ovs-vswitchd has enough time to pull statistics from the datapath. */
283 return (MAX(cfm->ccm_interval_ms, cfm->demand ? 500 : cfm->ccm_interval_ms)
288 ms_to_ccm_interval(int interval_ms)
292 for (i = 7; i > 0; i--) {
293 if (ccm_interval_to_ms(i) <= interval_ms) {
302 hash_mpid(uint64_t mpid)
304 return hash_uint64(mpid);
308 cfm_is_valid_mpid(bool extended, uint64_t mpid)
310 /* 802.1ag specification requires MPIDs to be within the range [1, 8191].
311 * In extended mode we relax this requirement. */
312 return mpid >= 1 && (extended || mpid <= 8191);
315 static struct remote_mp *
316 lookup_remote_mp(const struct cfm *cfm, uint64_t mpid) OVS_REQUIRES(mutex)
318 struct remote_mp *rmp;
320 HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), &cfm->remote_mps) {
321 if (rmp->mpid == mpid) {
332 unixctl_command_register("cfm/show", "[interface]", 0, 1, cfm_unixctl_show,
334 unixctl_command_register("cfm/set-fault", "[interface] normal|false|true",
335 1, 2, cfm_unixctl_set_fault, NULL);
338 /* Records the status change and changes the global connectivity seq. */
340 cfm_status_changed(struct cfm *cfm) OVS_REQUIRES(mutex)
342 seq_change(connectivity_seq_get());
343 cfm->status_changed = true;
346 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
347 * cfm_configure() before use. */
349 cfm_create(const struct netdev *netdev) OVS_EXCLUDED(mutex)
353 cfm = xzalloc(sizeof *cfm);
354 cfm->netdev = netdev_ref(netdev);
355 cfm->name = netdev_get_name(cfm->netdev);
356 hmap_init(&cfm->remote_mps);
357 cfm->remote_opup = true;
358 cfm->fault_override = -1;
362 atomic_init(&cfm->extended, false);
363 atomic_init(&cfm->check_tnl_key, false);
364 ovs_refcount_init(&cfm->ref_cnt);
366 ovs_mutex_lock(&mutex);
367 cfm_status_changed(cfm);
368 cfm_generate_maid(cfm);
369 hmap_insert(all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0));
370 ovs_mutex_unlock(&mutex);
376 cfm_unref(struct cfm *cfm) OVS_EXCLUDED(mutex)
378 struct remote_mp *rmp, *rmp_next;
384 if (ovs_refcount_unref_relaxed(&cfm->ref_cnt) != 1) {
388 ovs_mutex_lock(&mutex);
389 cfm_status_changed(cfm);
390 hmap_remove(all_cfms, &cfm->hmap_node);
391 ovs_mutex_unlock(&mutex);
393 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
394 hmap_remove(&cfm->remote_mps, &rmp->node);
398 hmap_destroy(&cfm->remote_mps);
399 netdev_close(cfm->netdev);
400 free(cfm->rmps_array);
406 cfm_ref(const struct cfm *cfm_)
408 struct cfm *cfm = CONST_CAST(struct cfm *, cfm_);
410 ovs_refcount_ref(&cfm->ref_cnt);
415 /* Should be run periodically to update fault statistics messages. */
417 cfm_run(struct cfm *cfm) OVS_EXCLUDED(mutex)
419 ovs_mutex_lock(&mutex);
420 if (timer_expired(&cfm->fault_timer)) {
421 long long int interval = cfm_fault_interval(cfm);
422 struct remote_mp *rmp, *rmp_next;
423 enum cfm_fault_reason old_cfm_fault = cfm->fault;
424 uint64_t old_flap_count = cfm->flap_count;
425 int old_health = cfm->health;
426 size_t old_rmps_array_len = cfm->rmps_array_len;
427 bool old_rmps_deleted = false;
428 bool old_rmp_opup = cfm->remote_opup;
429 bool demand_override;
430 bool rmp_set_opup = false;
431 bool rmp_set_opdown = false;
433 cfm->fault = cfm->recv_fault;
436 cfm->rmps_array_len = 0;
437 free(cfm->rmps_array);
438 cfm->rmps_array = xmalloc(hmap_count(&cfm->remote_mps) *
439 sizeof *cfm->rmps_array);
441 if (cfm->health_interval == CFM_HEALTH_INTERVAL) {
442 /* Calculate the cfm health of the interface. If the number of
443 * remote_mpids of a cfm interface is > 1, the cfm health is
444 * undefined. If the number of remote_mpids is 1, the cfm health is
445 * the percentage of the ccm frames received in the
446 * (CFM_HEALTH_INTERVAL * 3.5)ms, else it is 0. */
447 if (hmap_count(&cfm->remote_mps) > 1) {
449 } else if (hmap_is_empty(&cfm->remote_mps)) {
454 rmp = CONTAINER_OF(hmap_first(&cfm->remote_mps),
455 struct remote_mp, node);
456 exp_ccm_recvd = (CFM_HEALTH_INTERVAL * 7) / 2;
457 /* Calculate the percentage of healthy ccm frames received.
458 * Since the 'fault_interval' is (3.5 * cfm_interval), and
459 * 1 CCM packet must be received every cfm_interval,
460 * the 'remote_mpid' health reports the percentage of
461 * healthy CCM frames received every
462 * 'CFM_HEALTH_INTERVAL'th 'fault_interval'. */
463 cfm->health = (rmp->num_health_ccm * 100) / exp_ccm_recvd;
464 cfm->health = MIN(cfm->health, 100);
465 rmp->num_health_ccm = 0;
466 ovs_assert(cfm->health >= 0 && cfm->health <= 100);
468 cfm->health_interval = 0;
470 cfm->health_interval++;
472 demand_override = false;
474 uint64_t rx_packets = cfm_rx_packets(cfm);
475 demand_override = hmap_count(&cfm->remote_mps) == 1
476 && rx_packets > cfm->rx_packets
477 && !timer_expired(&cfm->demand_rx_ccm_t);
478 cfm->rx_packets = rx_packets;
481 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
483 VLOG_INFO("%s: Received no CCM from RMP %"PRIu64" in the last"
484 " %lldms", cfm->name, rmp->mpid,
485 time_msec() - rmp->last_rx);
486 if (!demand_override) {
487 old_rmps_deleted = true;
488 hmap_remove(&cfm->remote_mps, &rmp->node);
497 rmp_set_opdown = true;
500 cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid;
504 if (rmp_set_opdown) {
505 cfm->remote_opup = false;
507 else if (rmp_set_opup) {
508 cfm->remote_opup = true;
511 if (hmap_is_empty(&cfm->remote_mps)) {
512 cfm->fault |= CFM_FAULT_RECV;
515 if (old_cfm_fault != cfm->fault) {
516 if (!VLOG_DROP_INFO(&rl)) {
517 struct ds ds = DS_EMPTY_INITIALIZER;
519 ds_put_cstr(&ds, "from [");
520 ds_put_cfm_fault(&ds, old_cfm_fault);
521 ds_put_cstr(&ds, "] to [");
522 ds_put_cfm_fault(&ds, cfm->fault);
523 ds_put_char(&ds, ']');
524 VLOG_INFO("%s: CFM faults changed %s.", cfm->name, ds_cstr(&ds));
528 /* If there is a flap, increments the counter. */
529 if (old_cfm_fault == 0 || cfm->fault == 0) {
534 /* These variables represent the cfm session status, it is desirable
535 * to update them to database immediately after change. */
536 if (old_health != cfm->health
537 || old_rmp_opup != cfm->remote_opup
538 || (old_rmps_array_len != cfm->rmps_array_len || old_rmps_deleted)
539 || old_cfm_fault != cfm->fault
540 || old_flap_count != cfm->flap_count) {
541 cfm_status_changed(cfm);
545 timer_set_duration(&cfm->fault_timer, interval);
546 VLOG_DBG("%s: new fault interval", cfm->name);
548 ovs_mutex_unlock(&mutex);
551 /* Should be run periodically to check if the CFM module has a CCM message it
554 cfm_should_send_ccm(struct cfm *cfm) OVS_EXCLUDED(mutex)
558 ovs_mutex_lock(&mutex);
559 ret = timer_expired(&cfm->tx_timer);
560 ovs_mutex_unlock(&mutex);
564 /* Composes a CCM message into 'packet'. Messages generated with this function
565 * should be sent whenever cfm_should_send_ccm() indicates. */
567 cfm_compose_ccm(struct cfm *cfm, struct dp_packet *packet,
568 const uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
574 ovs_mutex_lock(&mutex);
575 timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms);
576 eth_compose(packet, cfm_ccm_addr(cfm), eth_src, ETH_TYPE_CFM, sizeof *ccm);
578 ccm_vlan = (cfm->ccm_vlan != CFM_RANDOM_VLAN
581 ccm_vlan = ccm_vlan & VLAN_VID_MASK;
583 if (ccm_vlan || cfm->ccm_pcp) {
584 uint16_t tci = ccm_vlan | (cfm->ccm_pcp << VLAN_PCP_SHIFT);
585 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), htons(tci));
588 atomic_read_relaxed(&cfm->extended, &extended);
590 ccm = dp_packet_l3(packet);
591 ccm->mdlevel_version = 0;
592 ccm->opcode = CCM_OPCODE;
593 ccm->tlv_offset = 70;
594 ccm->seq = htonl(++cfm->seq);
595 ccm->flags = cfm->ccm_interval;
596 memcpy(ccm->maid, cfm->maid, sizeof ccm->maid);
597 memset(ccm->zero, 0, sizeof ccm->zero);
601 ccm->mpid = htons(hash_mpid(cfm->mpid));
602 ccm->mpid64 = htonll(cfm->mpid);
603 ccm->opdown = !cfm->opup;
605 ccm->mpid = htons(cfm->mpid);
606 ccm->mpid64 = htonll(0);
610 if (cfm->ccm_interval == 0) {
611 ovs_assert(extended);
612 ccm->interval_ms_x = htons(cfm->ccm_interval_ms);
614 ccm->interval_ms_x = htons(0);
617 if (cfm->booted && hmap_is_empty(&cfm->remote_mps)) {
618 ccm->flags |= CCM_RDI_MASK;
622 long long int delay = time_msec() - cfm->last_tx;
623 if (delay > (cfm->ccm_interval_ms * 3 / 2)) {
624 VLOG_INFO("%s: long delay of %lldms (expected %dms) sending CCM"
625 " seq %"PRIu32, cfm->name, delay, cfm->ccm_interval_ms,
629 cfm->last_tx = time_msec();
630 ovs_mutex_unlock(&mutex);
634 cfm_wait(struct cfm *cfm) OVS_EXCLUDED(mutex)
636 long long int wake_time = cfm_wake_time(cfm);
637 poll_timer_wait_until(wake_time);
642 /* Returns the next cfm wakeup time. */
644 cfm_wake_time(struct cfm *cfm) OVS_EXCLUDED(mutex)
646 long long int retval;
652 ovs_mutex_lock(&mutex);
653 retval = MIN(cfm->tx_timer.t, cfm->fault_timer.t);
654 ovs_mutex_unlock(&mutex);
659 /* Configures 'cfm' with settings from 's'. */
661 cfm_configure(struct cfm *cfm, const struct cfm_settings *s)
667 if (!cfm_is_valid_mpid(s->extended, s->mpid) || s->interval <= 0) {
671 ovs_mutex_lock(&mutex);
674 interval = ms_to_ccm_interval(s->interval);
675 interval_ms = ccm_interval_to_ms(interval);
677 atomic_store_relaxed(&cfm->check_tnl_key, s->check_tnl_key);
678 atomic_store_relaxed(&cfm->extended, s->extended);
680 cfm->ccm_vlan = s->ccm_vlan;
681 cfm->ccm_pcp = s->ccm_pcp & (VLAN_PCP_MASK >> VLAN_PCP_SHIFT);
682 if (s->extended && interval_ms != s->interval) {
684 interval_ms = MIN(s->interval, UINT16_MAX);
687 if (s->extended && s->demand) {
690 cfm->rx_packets = cfm_rx_packets(cfm);
696 if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) {
697 cfm->ccm_interval = interval;
698 cfm->ccm_interval_ms = interval_ms;
700 timer_set_expired(&cfm->tx_timer);
701 timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm));
704 ovs_mutex_unlock(&mutex);
708 /* Must be called when the netdev owned by 'cfm' should change. */
710 cfm_set_netdev(struct cfm *cfm, const struct netdev *netdev)
713 ovs_mutex_lock(&mutex);
714 if (cfm->netdev != netdev) {
715 netdev_close(cfm->netdev);
716 cfm->netdev = netdev_ref(netdev);
718 ovs_mutex_unlock(&mutex);
721 /* Returns true if 'cfm' should process packets from 'flow'. Sets
722 * fields in 'wc' that were used to make the determination. */
724 cfm_should_process_flow(const struct cfm *cfm_, const struct flow *flow,
725 struct flow_wildcards *wc)
727 struct cfm *cfm = CONST_CAST(struct cfm *, cfm_);
730 /* Most packets are not CFM. */
731 if (OVS_LIKELY(flow->dl_type != htons(ETH_TYPE_CFM))) {
735 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
736 if (OVS_UNLIKELY(!eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm)))) {
740 atomic_read_relaxed(&cfm->check_tnl_key, &check_tnl_key);
743 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
744 return flow->tunnel.tun_id == htonll(0);
749 /* Updates internal statistics relevant to packet 'p'. Should be called on
750 * every packet whose flow returned true when passed to
751 * cfm_should_process_flow. */
753 cfm_process_heartbeat(struct cfm *cfm, const struct dp_packet *p)
757 struct eth_header *eth;
760 ovs_mutex_lock(&mutex);
762 atomic_read_relaxed(&cfm->extended, &extended);
764 eth = dp_packet_l2(p);
765 ccm = dp_packet_at(p, (uint8_t *)dp_packet_l3(p) - (uint8_t *)dp_packet_data(p),
769 VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.",
774 if (ccm->opcode != CCM_OPCODE) {
775 VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. "
776 "(opcode %u)", cfm->name, ccm->opcode);
780 /* According to the 802.1ag specification, reception of a CCM with an
781 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
782 * trigger a fault. We ignore this requirement for several reasons.
784 * Faults can cause a controller or Open vSwitch to make potentially
785 * expensive changes to the network topology. It seems prudent to trigger
786 * them judiciously, especially when CFM is used to check slave status of
787 * bonds. Furthermore, faults can be maliciously triggered by crafting
788 * unexpected CCMs. */
789 if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) {
790 cfm->recv_fault |= CFM_FAULT_MAID;
791 VLOG_WARN_RL(&rl, "%s: Received unexpected remote MAID from MAC "
792 ETH_ADDR_FMT, cfm->name, ETH_ADDR_ARGS(eth->eth_src));
794 uint8_t ccm_interval = ccm->flags & 0x7;
795 bool ccm_rdi = ccm->flags & CCM_RDI_MASK;
796 uint16_t ccm_interval_ms_x = ntohs(ccm->interval_ms_x);
798 struct remote_mp *rmp;
802 enum cfm_fault_reason cfm_fault = 0;
805 ccm_mpid = ntohll(ccm->mpid64);
806 ccm_opdown = ccm->opdown;
808 ccm_mpid = ntohs(ccm->mpid);
811 ccm_seq = ntohl(ccm->seq);
813 if (ccm_interval != cfm->ccm_interval) {
814 VLOG_WARN_RL(&rl, "%s: received a CCM with an unexpected interval"
815 " (%"PRIu8") from RMP %"PRIu64, cfm->name,
816 ccm_interval, ccm_mpid);
819 if (extended && ccm_interval == 0
820 && ccm_interval_ms_x != cfm->ccm_interval_ms) {
821 VLOG_WARN_RL(&rl, "%s: received a CCM with an unexpected extended"
822 " interval (%"PRIu16"ms) from RMP %"PRIu64, cfm->name,
823 ccm_interval_ms_x, ccm_mpid);
826 rmp = lookup_remote_mp(cfm, ccm_mpid);
828 if (hmap_count(&cfm->remote_mps) < CFM_MAX_RMPS) {
829 rmp = xzalloc(sizeof *rmp);
830 hmap_insert(&cfm->remote_mps, &rmp->node, hash_mpid(ccm_mpid));
832 cfm_fault |= CFM_FAULT_OVERFLOW;
834 "%s: dropped CCM with MPID %"PRIu64" from MAC "
835 ETH_ADDR_FMT, cfm->name, ccm_mpid,
836 ETH_ADDR_ARGS(eth->eth_src));
841 cfm_fault |= CFM_FAULT_RDI;
842 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64, cfm->name,
846 VLOG_DBG("%s: received CCM (seq %"PRIu32") (mpid %"PRIu64")"
847 " (interval %"PRIu8") (RDI %s)", cfm->name, ccm_seq,
848 ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false");
851 if (rmp->mpid == cfm->mpid) {
852 cfm_fault |= CFM_FAULT_LOOPBACK;
853 VLOG_WARN_RL(&rl,"%s: received CCM with local MPID"
854 " %"PRIu64, cfm->name, rmp->mpid);
857 if (rmp->seq && ccm_seq != (rmp->seq + 1)) {
858 VLOG_WARN_RL(&rl, "%s: (mpid %"PRIu64") detected sequence"
859 " numbers which indicate possible connectivity"
860 " problems (previous %"PRIu32") (current %"PRIu32
861 ")", cfm->name, ccm_mpid, rmp->seq, ccm_seq);
864 rmp->mpid = ccm_mpid;
866 rmp->num_health_ccm++;
868 timer_set_duration(&cfm->demand_rx_ccm_t,
869 100 * cfm->ccm_interval_ms);
873 cfm->recv_fault |= cfm_fault;
875 rmp->opup = !ccm_opdown;
876 rmp->last_rx = time_msec();
881 ovs_mutex_unlock(&mutex);
884 /* Returns and resets the 'cfm->status_changed'. */
886 cfm_check_status_change(struct cfm *cfm) OVS_EXCLUDED(mutex)
890 ovs_mutex_lock(&mutex);
891 ret = cfm->status_changed;
892 cfm->status_changed = false;
893 ovs_mutex_unlock(&mutex);
899 cfm_get_fault__(const struct cfm *cfm) OVS_REQUIRES(mutex)
901 if (cfm->fault_override >= 0) {
902 return cfm->fault_override ? CFM_FAULT_OVERRIDE : 0;
907 /* Gets the fault status of 'cfm'. Returns a bit mask of 'cfm_fault_reason's
908 * indicating the cause of the connectivity fault, or zero if there is no
911 cfm_get_fault(const struct cfm *cfm) OVS_EXCLUDED(mutex)
915 ovs_mutex_lock(&mutex);
916 fault = cfm_get_fault__(cfm);
917 ovs_mutex_unlock(&mutex);
921 /* Gets the number of cfm fault flapping since start. */
923 cfm_get_flap_count(const struct cfm *cfm) OVS_EXCLUDED(mutex)
926 ovs_mutex_lock(&mutex);
927 flap_count = cfm->flap_count;
928 ovs_mutex_unlock(&mutex);
932 /* Gets the health of 'cfm'. Returns an integer between 0 and 100 indicating
933 * the health of the link as a percentage of ccm frames received in
934 * CFM_HEALTH_INTERVAL * 'fault_interval' if there is only 1 remote_mpid,
935 * returns 0 if there are no remote_mpids, and returns -1 if there are more
936 * than 1 remote_mpids. */
938 cfm_get_health(const struct cfm *cfm) OVS_EXCLUDED(mutex)
942 ovs_mutex_lock(&mutex);
943 health = cfm->health;
944 ovs_mutex_unlock(&mutex);
949 cfm_get_opup__(const struct cfm *cfm_) OVS_REQUIRES(mutex)
951 struct cfm *cfm = CONST_CAST(struct cfm *, cfm_);
954 atomic_read_relaxed(&cfm->extended, &extended);
956 return extended ? cfm->remote_opup : -1;
959 /* Gets the operational state of 'cfm'. 'cfm' is considered operationally down
960 * if it has received a CCM with the operationally down bit set from any of its
961 * remote maintenance points. Returns 1 if 'cfm' is operationally up, 0 if
962 * 'cfm' is operationally down, or -1 if 'cfm' has no operational state
963 * (because it isn't in extended mode). */
965 cfm_get_opup(const struct cfm *cfm) OVS_EXCLUDED(mutex)
969 ovs_mutex_lock(&mutex);
970 opup = cfm_get_opup__(cfm);
971 ovs_mutex_unlock(&mutex);
977 cfm_get_remote_mpids__(const struct cfm *cfm, uint64_t **rmps, size_t *n_rmps)
980 *rmps = xmemdup(cfm->rmps_array, cfm->rmps_array_len * sizeof **rmps);
981 *n_rmps = cfm->rmps_array_len;
984 /* Populates 'rmps' with an array of remote maintenance points reachable by
985 * 'cfm'. The number of remote maintenance points is written to 'n_rmps'.
986 * 'cfm' retains ownership of the array written to 'rmps' */
988 cfm_get_remote_mpids(const struct cfm *cfm, uint64_t **rmps, size_t *n_rmps)
991 ovs_mutex_lock(&mutex);
992 cfm_get_remote_mpids__(cfm, rmps, n_rmps);
993 ovs_mutex_unlock(&mutex);
996 /* Extracts the status of 'cfm' and fills in the 's'. */
998 cfm_get_status(const struct cfm *cfm, struct cfm_status *s) OVS_EXCLUDED(mutex)
1000 ovs_mutex_lock(&mutex);
1001 s->faults = cfm_get_fault__(cfm);
1002 s->remote_opstate = cfm_get_opup__(cfm);
1003 s->flap_count = cfm->flap_count;
1004 s->health = cfm->health;
1005 cfm_get_remote_mpids__(cfm, &s->rmps, &s->n_rmps);
1006 ovs_mutex_unlock(&mutex);
1010 cfm_find(const char *name) OVS_REQUIRES(mutex)
1014 HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), all_cfms) {
1015 if (!strcmp(cfm->name, name)) {
1023 cfm_print_details(struct ds *ds, struct cfm *cfm) OVS_REQUIRES(mutex)
1025 struct remote_mp *rmp;
1029 atomic_read_relaxed(&cfm->extended, &extended);
1031 ds_put_format(ds, "---- %s ----\n", cfm->name);
1032 ds_put_format(ds, "MPID %"PRIu64":%s%s\n", cfm->mpid,
1033 extended ? " extended" : "",
1034 cfm->fault_override >= 0 ? " fault_override" : "");
1036 fault = cfm_get_fault__(cfm);
1038 ds_put_cstr(ds, "\tfault: ");
1039 ds_put_cfm_fault(ds, fault);
1040 ds_put_cstr(ds, "\n");
1043 if (cfm->health == -1) {
1044 ds_put_format(ds, "\taverage health: undefined\n");
1046 ds_put_format(ds, "\taverage health: %d\n", cfm->health);
1048 ds_put_format(ds, "\topstate: %s\n", cfm->opup ? "up" : "down");
1049 ds_put_format(ds, "\tremote_opstate: %s\n",
1050 cfm->remote_opup ? "up" : "down");
1051 ds_put_format(ds, "\tinterval: %dms\n", cfm->ccm_interval_ms);
1052 ds_put_format(ds, "\tnext CCM tx: %lldms\n",
1053 timer_msecs_until_expired(&cfm->tx_timer));
1054 ds_put_format(ds, "\tnext fault check: %lldms\n",
1055 timer_msecs_until_expired(&cfm->fault_timer));
1057 HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
1058 ds_put_format(ds, "Remote MPID %"PRIu64"\n", rmp->mpid);
1059 ds_put_format(ds, "\trecv since check: %s\n",
1060 rmp->recv ? "true" : "false");
1061 ds_put_format(ds, "\topstate: %s\n", rmp->opup? "up" : "down");
1066 cfm_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1067 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1069 struct ds ds = DS_EMPTY_INITIALIZER;
1072 ovs_mutex_lock(&mutex);
1074 cfm = cfm_find(argv[1]);
1076 unixctl_command_reply_error(conn, "no such CFM object");
1079 cfm_print_details(&ds, cfm);
1081 HMAP_FOR_EACH (cfm, hmap_node, all_cfms) {
1082 cfm_print_details(&ds, cfm);
1086 unixctl_command_reply(conn, ds_cstr(&ds));
1089 ovs_mutex_unlock(&mutex);
1093 cfm_unixctl_set_fault(struct unixctl_conn *conn, int argc, const char *argv[],
1094 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1096 const char *fault_str = argv[argc - 1];
1100 ovs_mutex_lock(&mutex);
1101 if (!strcasecmp("true", fault_str)) {
1103 } else if (!strcasecmp("false", fault_str)) {
1105 } else if (!strcasecmp("normal", fault_str)) {
1106 fault_override = -1;
1108 unixctl_command_reply_error(conn, "unknown fault string");
1113 cfm = cfm_find(argv[1]);
1115 unixctl_command_reply_error(conn, "no such CFM object");
1118 cfm->fault_override = fault_override;
1119 cfm_status_changed(cfm);
1121 HMAP_FOR_EACH (cfm, hmap_node, all_cfms) {
1122 cfm->fault_override = fault_override;
1123 cfm_status_changed(cfm);
1127 unixctl_command_reply(conn, "OK");
1130 ovs_mutex_unlock(&mutex);