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 "openvswitch/dynamic-string.h"
30 #include "openvswitch/hmap.h"
32 #include "ovs-atomic.h"
34 #include "openvswitch/poll-loop.h"
40 #include "openvswitch/vlog.h"
43 VLOG_DEFINE_THIS_MODULE(cfm
);
45 #define CFM_MAX_RMPS 256
47 /* Ethernet destination address of CCM packets. */
48 static const struct eth_addr eth_addr_ccm
= ETH_ADDR_C(01,80,c2
,00,00,30);
49 static const struct eth_addr eth_addr_ccm_x
= ETH_ADDR_C(01,23,20,00,00,30);
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 /* These bools are atomic to allow readers to check their values
132 * without taking 'mutex'. Such readers do not assume the values they
133 * read are synchronized with any other members. */
134 atomic_bool check_tnl_key
; /* Verify the tunnel key of inbound packets? */
135 atomic_bool extended
; /* Extended mode. */
136 struct ovs_refcount ref_cnt
;
138 uint64_t flap_count
; /* Count the flaps since boot. */
140 /* True when the variables returned by cfm_get_*() are changed
141 * since last check. */
144 /* When 'cfm->demand' is set, at least one ccm is required to be received
145 * every 100 * cfm_interval. If ccm is not received within this interval,
146 * even if data packets are received, the cfm fault will be set. */
147 struct timer demand_rx_ccm_t
;
150 /* Remote MPs represent foreign network entities that are configured to have
151 * the same MAID as this CFM instance. */
153 uint64_t mpid
; /* The Maintenance Point ID of this 'remote_mp'. */
154 struct hmap_node node
; /* Node in 'remote_mps' map. */
156 bool recv
; /* CCM was received since last fault check. */
157 bool opup
; /* Operational State. */
158 uint32_t seq
; /* Most recently received sequence number. */
159 uint8_t num_health_ccm
; /* Number of received ccm frames every
160 CFM_HEALTH_INTERVAL * 'fault_interval'. */
161 long long int last_rx
; /* Last CCM reception time. */
165 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(20, 30);
167 static struct ovs_mutex mutex
= OVS_MUTEX_INITIALIZER
;
168 static struct hmap all_cfms__
= HMAP_INITIALIZER(&all_cfms__
);
169 static struct hmap
*const all_cfms
OVS_GUARDED_BY(mutex
) = &all_cfms__
;
171 static unixctl_cb_func cfm_unixctl_show
;
172 static unixctl_cb_func cfm_unixctl_set_fault
;
175 cfm_rx_packets(const struct cfm
*cfm
) OVS_REQUIRES(mutex
)
177 struct netdev_stats stats
;
179 if (!netdev_get_stats(cfm
->netdev
, &stats
)) {
180 return stats
.rx_packets
;
186 static struct eth_addr
187 cfm_ccm_addr(struct cfm
*cfm
)
191 atomic_read_relaxed(&cfm
->extended
, &extended
);
193 return extended
? eth_addr_ccm_x
: eth_addr_ccm
;
196 /* Returns the string representation of the given cfm_fault_reason 'reason'. */
198 cfm_fault_reason_to_str(int reason
)
201 #define CFM_FAULT_REASON(NAME, STR) case CFM_FAULT_##NAME: return #STR;
203 #undef CFM_FAULT_REASON
204 default: return "<unknown>";
209 ds_put_cfm_fault(struct ds
*ds
, int fault
)
213 for (i
= 0; i
< CFM_FAULT_N_REASONS
; i
++) {
216 if (fault
& reason
) {
217 ds_put_format(ds
, "%s ", cfm_fault_reason_to_str(reason
));
225 cfm_generate_maid(struct cfm
*cfm
) OVS_REQUIRES(mutex
)
227 const char *ovs_md_name
= "ovs";
228 const char *ovs_ma_name
= "ovs";
230 size_t md_len
, ma_len
;
232 memset(cfm
->maid
, 0, CCM_MAID_LEN
);
234 md_len
= strlen(ovs_md_name
);
235 ma_len
= strlen(ovs_ma_name
);
237 ovs_assert(md_len
&& ma_len
&& md_len
+ ma_len
+ 4 <= CCM_MAID_LEN
);
239 cfm
->maid
[0] = 4; /* MD name string format. */
240 cfm
->maid
[1] = md_len
; /* MD name size. */
241 memcpy(&cfm
->maid
[2], ovs_md_name
, md_len
); /* MD name. */
243 ma_p
= cfm
->maid
+ 2 + md_len
;
244 ma_p
[0] = 2; /* MA name string format. */
245 ma_p
[1] = ma_len
; /* MA name size. */
246 memcpy(&ma_p
[2], ovs_ma_name
, ma_len
); /* MA name. */
250 ccm_interval_to_ms(uint8_t interval
)
253 case 0: OVS_NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
254 case 1: return 3; /* Not recommended due to timer resolution. */
255 case 2: return 10; /* Not recommended due to timer resolution. */
258 case 5: return 10000;
259 case 6: return 60000;
260 case 7: return 600000;
261 default: OVS_NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
268 cfm_fault_interval(struct cfm
*cfm
) OVS_REQUIRES(mutex
)
270 /* According to the 802.1ag specification we should assume every other MP
271 * with the same MAID has the same transmission interval that we have. If
272 * an MP has a different interval, cfm_process_heartbeat will register it
273 * as a fault (likely due to a configuration error). Thus we can check all
274 * MPs at once making this quite a bit simpler.
276 * When cfm is not in demand mode, we check when (ccm_interval_ms * 3.5) ms
277 * have passed. When cfm is in demand mode, we check when
278 * (MAX(ccm_interval_ms, 500) * 3.5) ms have passed. This ensures that
279 * ovs-vswitchd has enough time to pull statistics from the datapath. */
281 return (MAX(cfm
->ccm_interval_ms
, cfm
->demand
? 500 : cfm
->ccm_interval_ms
)
286 ms_to_ccm_interval(int interval_ms
)
290 for (i
= 7; i
> 0; i
--) {
291 if (ccm_interval_to_ms(i
) <= interval_ms
) {
300 hash_mpid(uint64_t mpid
)
302 return hash_uint64(mpid
);
306 cfm_is_valid_mpid(bool extended
, uint64_t mpid
)
308 /* 802.1ag specification requires MPIDs to be within the range [1, 8191].
309 * In extended mode we relax this requirement. */
310 return mpid
>= 1 && (extended
|| mpid
<= 8191);
313 static struct remote_mp
*
314 lookup_remote_mp(const struct cfm
*cfm
, uint64_t mpid
) OVS_REQUIRES(mutex
)
316 struct remote_mp
*rmp
;
318 HMAP_FOR_EACH_IN_BUCKET (rmp
, node
, hash_mpid(mpid
), &cfm
->remote_mps
) {
319 if (rmp
->mpid
== mpid
) {
330 unixctl_command_register("cfm/show", "[interface]", 0, 1, cfm_unixctl_show
,
332 unixctl_command_register("cfm/set-fault", "[interface] normal|false|true",
333 1, 2, cfm_unixctl_set_fault
, NULL
);
336 /* Records the status change and changes the global connectivity seq. */
338 cfm_status_changed(struct cfm
*cfm
) OVS_REQUIRES(mutex
)
340 seq_change(connectivity_seq_get());
341 cfm
->status_changed
= true;
344 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
345 * cfm_configure() before use. */
347 cfm_create(const struct netdev
*netdev
) OVS_EXCLUDED(mutex
)
351 cfm
= xzalloc(sizeof *cfm
);
352 cfm
->netdev
= netdev_ref(netdev
);
353 cfm
->name
= netdev_get_name(cfm
->netdev
);
354 hmap_init(&cfm
->remote_mps
);
355 cfm
->remote_opup
= true;
356 cfm
->fault_override
= -1;
360 atomic_init(&cfm
->extended
, false);
361 atomic_init(&cfm
->check_tnl_key
, false);
362 ovs_refcount_init(&cfm
->ref_cnt
);
364 ovs_mutex_lock(&mutex
);
365 cfm_status_changed(cfm
);
366 cfm_generate_maid(cfm
);
367 hmap_insert(all_cfms
, &cfm
->hmap_node
, hash_string(cfm
->name
, 0));
368 ovs_mutex_unlock(&mutex
);
374 cfm_unref(struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
376 struct remote_mp
*rmp
;
382 if (ovs_refcount_unref_relaxed(&cfm
->ref_cnt
) != 1) {
386 ovs_mutex_lock(&mutex
);
387 cfm_status_changed(cfm
);
388 hmap_remove(all_cfms
, &cfm
->hmap_node
);
389 ovs_mutex_unlock(&mutex
);
391 HMAP_FOR_EACH_POP (rmp
, node
, &cfm
->remote_mps
) {
395 hmap_destroy(&cfm
->remote_mps
);
396 netdev_close(cfm
->netdev
);
397 free(cfm
->rmps_array
);
403 cfm_ref(const struct cfm
*cfm_
)
405 struct cfm
*cfm
= CONST_CAST(struct cfm
*, cfm_
);
407 ovs_refcount_ref(&cfm
->ref_cnt
);
412 /* Should be run periodically to update fault statistics messages. */
414 cfm_run(struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
416 ovs_mutex_lock(&mutex
);
417 if (timer_expired(&cfm
->fault_timer
)) {
418 long long int interval
= cfm_fault_interval(cfm
);
419 struct remote_mp
*rmp
, *rmp_next
;
420 enum cfm_fault_reason old_cfm_fault
= cfm
->fault
;
421 uint64_t old_flap_count
= cfm
->flap_count
;
422 int old_health
= cfm
->health
;
423 size_t old_rmps_array_len
= cfm
->rmps_array_len
;
424 bool old_rmps_deleted
= false;
425 bool old_rmp_opup
= cfm
->remote_opup
;
426 bool demand_override
;
427 bool rmp_set_opup
= false;
428 bool rmp_set_opdown
= false;
430 cfm
->fault
= cfm
->recv_fault
;
433 cfm
->rmps_array_len
= 0;
434 free(cfm
->rmps_array
);
435 cfm
->rmps_array
= xmalloc(hmap_count(&cfm
->remote_mps
) *
436 sizeof *cfm
->rmps_array
);
438 if (cfm
->health_interval
== CFM_HEALTH_INTERVAL
) {
439 /* Calculate the cfm health of the interface. If the number of
440 * remote_mpids of a cfm interface is > 1, the cfm health is
441 * undefined. If the number of remote_mpids is 1, the cfm health is
442 * the percentage of the ccm frames received in the
443 * (CFM_HEALTH_INTERVAL * 3.5)ms, else it is 0. */
444 if (hmap_count(&cfm
->remote_mps
) > 1) {
446 } else if (hmap_is_empty(&cfm
->remote_mps
)) {
451 rmp
= CONTAINER_OF(hmap_first(&cfm
->remote_mps
),
452 struct remote_mp
, node
);
453 exp_ccm_recvd
= (CFM_HEALTH_INTERVAL
* 7) / 2;
454 /* Calculate the percentage of healthy ccm frames received.
455 * Since the 'fault_interval' is (3.5 * cfm_interval), and
456 * 1 CCM packet must be received every cfm_interval,
457 * the 'remote_mpid' health reports the percentage of
458 * healthy CCM frames received every
459 * 'CFM_HEALTH_INTERVAL'th 'fault_interval'. */
460 cfm
->health
= (rmp
->num_health_ccm
* 100) / exp_ccm_recvd
;
461 cfm
->health
= MIN(cfm
->health
, 100);
462 rmp
->num_health_ccm
= 0;
463 ovs_assert(cfm
->health
>= 0 && cfm
->health
<= 100);
465 cfm
->health_interval
= 0;
467 cfm
->health_interval
++;
469 demand_override
= false;
471 uint64_t rx_packets
= cfm_rx_packets(cfm
);
472 demand_override
= hmap_count(&cfm
->remote_mps
) == 1
473 && rx_packets
> cfm
->rx_packets
474 && !timer_expired(&cfm
->demand_rx_ccm_t
);
475 cfm
->rx_packets
= rx_packets
;
478 HMAP_FOR_EACH_SAFE (rmp
, rmp_next
, node
, &cfm
->remote_mps
) {
480 VLOG_INFO("%s: Received no CCM from RMP %"PRIu64
" in the last"
481 " %lldms", cfm
->name
, rmp
->mpid
,
482 time_msec() - rmp
->last_rx
);
483 if (!demand_override
) {
484 old_rmps_deleted
= true;
485 hmap_remove(&cfm
->remote_mps
, &rmp
->node
);
494 rmp_set_opdown
= true;
497 cfm
->rmps_array
[cfm
->rmps_array_len
++] = rmp
->mpid
;
501 if (rmp_set_opdown
) {
502 cfm
->remote_opup
= false;
504 else if (rmp_set_opup
) {
505 cfm
->remote_opup
= true;
508 if (hmap_is_empty(&cfm
->remote_mps
)) {
509 cfm
->fault
|= CFM_FAULT_RECV
;
512 if (old_cfm_fault
!= cfm
->fault
) {
513 if (!VLOG_DROP_INFO(&rl
)) {
514 struct ds ds
= DS_EMPTY_INITIALIZER
;
516 ds_put_cstr(&ds
, "from [");
517 ds_put_cfm_fault(&ds
, old_cfm_fault
);
518 ds_put_cstr(&ds
, "] to [");
519 ds_put_cfm_fault(&ds
, cfm
->fault
);
520 ds_put_char(&ds
, ']');
521 VLOG_INFO("%s: CFM faults changed %s.", cfm
->name
, ds_cstr(&ds
));
525 /* If there is a flap, increments the counter. */
526 if (old_cfm_fault
== 0 || cfm
->fault
== 0) {
531 /* These variables represent the cfm session status, it is desirable
532 * to update them to database immediately after change. */
533 if (old_health
!= cfm
->health
534 || old_rmp_opup
!= cfm
->remote_opup
535 || (old_rmps_array_len
!= cfm
->rmps_array_len
|| old_rmps_deleted
)
536 || old_cfm_fault
!= cfm
->fault
537 || old_flap_count
!= cfm
->flap_count
) {
538 cfm_status_changed(cfm
);
542 timer_set_duration(&cfm
->fault_timer
, interval
);
543 VLOG_DBG("%s: new fault interval", cfm
->name
);
545 ovs_mutex_unlock(&mutex
);
548 /* Should be run periodically to check if the CFM module has a CCM message it
551 cfm_should_send_ccm(struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
555 ovs_mutex_lock(&mutex
);
556 ret
= timer_expired(&cfm
->tx_timer
);
557 ovs_mutex_unlock(&mutex
);
561 /* Composes a CCM message into 'packet'. Messages generated with this function
562 * should be sent whenever cfm_should_send_ccm() indicates. */
564 cfm_compose_ccm(struct cfm
*cfm
, struct dp_packet
*packet
,
565 const struct eth_addr eth_src
) OVS_EXCLUDED(mutex
)
571 ovs_mutex_lock(&mutex
);
572 timer_set_duration(&cfm
->tx_timer
, cfm
->ccm_interval_ms
);
573 eth_compose(packet
, cfm_ccm_addr(cfm
), eth_src
, ETH_TYPE_CFM
, sizeof *ccm
);
575 ccm_vlan
= (cfm
->ccm_vlan
!= CFM_RANDOM_VLAN
578 ccm_vlan
= ccm_vlan
& VLAN_VID_MASK
;
580 if (ccm_vlan
|| cfm
->ccm_pcp
) {
581 uint16_t tci
= ccm_vlan
| (cfm
->ccm_pcp
<< VLAN_PCP_SHIFT
);
582 eth_push_vlan(packet
, htons(ETH_TYPE_VLAN
), htons(tci
));
585 atomic_read_relaxed(&cfm
->extended
, &extended
);
587 ccm
= dp_packet_l3(packet
);
588 ccm
->mdlevel_version
= 0;
589 ccm
->opcode
= CCM_OPCODE
;
590 ccm
->tlv_offset
= 70;
591 ccm
->seq
= htonl(++cfm
->seq
);
592 ccm
->flags
= cfm
->ccm_interval
;
593 memcpy(ccm
->maid
, cfm
->maid
, sizeof ccm
->maid
);
594 memset(ccm
->zero
, 0, sizeof ccm
->zero
);
598 ccm
->mpid
= htons(hash_mpid(cfm
->mpid
));
599 ccm
->mpid64
= htonll(cfm
->mpid
);
600 ccm
->opdown
= !cfm
->opup
;
602 ccm
->mpid
= htons(cfm
->mpid
);
603 ccm
->mpid64
= htonll(0);
607 if (cfm
->ccm_interval
== 0) {
608 ovs_assert(extended
);
609 ccm
->interval_ms_x
= htons(cfm
->ccm_interval_ms
);
611 ccm
->interval_ms_x
= htons(0);
614 if (cfm
->booted
&& hmap_is_empty(&cfm
->remote_mps
)) {
615 ccm
->flags
|= CCM_RDI_MASK
;
619 long long int delay
= time_msec() - cfm
->last_tx
;
620 if (delay
> (cfm
->ccm_interval_ms
* 3 / 2)) {
621 VLOG_INFO("%s: long delay of %lldms (expected %dms) sending CCM"
622 " seq %"PRIu32
, cfm
->name
, delay
, cfm
->ccm_interval_ms
,
626 cfm
->last_tx
= time_msec();
627 ovs_mutex_unlock(&mutex
);
631 cfm_wait(struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
633 long long int wake_time
= cfm_wake_time(cfm
);
634 poll_timer_wait_until(wake_time
);
639 /* Returns the next cfm wakeup time. */
641 cfm_wake_time(struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
643 long long int retval
;
649 ovs_mutex_lock(&mutex
);
650 retval
= MIN(cfm
->tx_timer
.t
, cfm
->fault_timer
.t
);
651 ovs_mutex_unlock(&mutex
);
656 /* Configures 'cfm' with settings from 's'. */
658 cfm_configure(struct cfm
*cfm
, const struct cfm_settings
*s
)
664 if (!cfm_is_valid_mpid(s
->extended
, s
->mpid
) || s
->interval
<= 0) {
668 ovs_mutex_lock(&mutex
);
671 interval
= ms_to_ccm_interval(s
->interval
);
672 interval_ms
= ccm_interval_to_ms(interval
);
674 atomic_store_relaxed(&cfm
->check_tnl_key
, s
->check_tnl_key
);
675 atomic_store_relaxed(&cfm
->extended
, s
->extended
);
677 cfm
->ccm_vlan
= s
->ccm_vlan
;
678 cfm
->ccm_pcp
= s
->ccm_pcp
& (VLAN_PCP_MASK
>> VLAN_PCP_SHIFT
);
679 if (s
->extended
&& interval_ms
!= s
->interval
) {
681 interval_ms
= MIN(s
->interval
, UINT16_MAX
);
684 if (s
->extended
&& s
->demand
) {
687 cfm
->rx_packets
= cfm_rx_packets(cfm
);
693 if (interval
!= cfm
->ccm_interval
|| interval_ms
!= cfm
->ccm_interval_ms
) {
694 cfm
->ccm_interval
= interval
;
695 cfm
->ccm_interval_ms
= interval_ms
;
697 timer_set_expired(&cfm
->tx_timer
);
698 timer_set_duration(&cfm
->fault_timer
, cfm_fault_interval(cfm
));
701 ovs_mutex_unlock(&mutex
);
705 /* Must be called when the netdev owned by 'cfm' should change. */
707 cfm_set_netdev(struct cfm
*cfm
, const struct netdev
*netdev
)
710 ovs_mutex_lock(&mutex
);
711 if (cfm
->netdev
!= netdev
) {
712 netdev_close(cfm
->netdev
);
713 cfm
->netdev
= netdev_ref(netdev
);
715 ovs_mutex_unlock(&mutex
);
718 /* Returns true if 'cfm' should process packets from 'flow'. Sets
719 * fields in 'wc' that were used to make the determination. */
721 cfm_should_process_flow(const struct cfm
*cfm_
, const struct flow
*flow
,
722 struct flow_wildcards
*wc
)
724 struct cfm
*cfm
= CONST_CAST(struct cfm
*, cfm_
);
727 /* Most packets are not CFM. */
728 if (OVS_LIKELY(flow
->dl_type
!= htons(ETH_TYPE_CFM
))) {
732 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
733 if (OVS_UNLIKELY(!eth_addr_equals(flow
->dl_dst
, cfm_ccm_addr(cfm
)))) {
737 atomic_read_relaxed(&cfm
->check_tnl_key
, &check_tnl_key
);
740 memset(&wc
->masks
.tunnel
.tun_id
, 0xff, sizeof wc
->masks
.tunnel
.tun_id
);
741 return flow
->tunnel
.tun_id
== htonll(0);
746 /* Updates internal statistics relevant to packet 'p'. Should be called on
747 * every packet whose flow returned true when passed to
748 * cfm_should_process_flow. */
750 cfm_process_heartbeat(struct cfm
*cfm
, const struct dp_packet
*p
)
754 struct eth_header
*eth
;
757 ovs_mutex_lock(&mutex
);
759 atomic_read_relaxed(&cfm
->extended
, &extended
);
761 eth
= dp_packet_eth(p
);
762 ccm
= dp_packet_at(p
, (uint8_t *)dp_packet_l3(p
) - (uint8_t *)dp_packet_data(p
),
766 VLOG_INFO_RL(&rl
, "%s: Received an unparseable 802.1ag CCM heartbeat.",
771 if (ccm
->opcode
!= CCM_OPCODE
) {
772 VLOG_INFO_RL(&rl
, "%s: Received an unsupported 802.1ag message. "
773 "(opcode %u)", cfm
->name
, ccm
->opcode
);
777 /* According to the 802.1ag specification, reception of a CCM with an
778 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
779 * trigger a fault. We ignore this requirement for several reasons.
781 * Faults can cause a controller or Open vSwitch to make potentially
782 * expensive changes to the network topology. It seems prudent to trigger
783 * them judiciously, especially when CFM is used to check status of bond
784 * members. Furthermore, faults can be maliciously triggered by crafting
785 * unexpected CCMs. */
786 if (memcmp(ccm
->maid
, cfm
->maid
, sizeof ccm
->maid
)) {
787 cfm
->recv_fault
|= CFM_FAULT_MAID
;
788 VLOG_WARN_RL(&rl
, "%s: Received unexpected remote MAID from MAC "
789 ETH_ADDR_FMT
, cfm
->name
, ETH_ADDR_ARGS(eth
->eth_src
));
791 uint8_t ccm_interval
= ccm
->flags
& 0x7;
792 bool ccm_rdi
= ccm
->flags
& CCM_RDI_MASK
;
793 uint16_t ccm_interval_ms_x
= ntohs(ccm
->interval_ms_x
);
795 struct remote_mp
*rmp
;
799 enum cfm_fault_reason cfm_fault
= 0;
802 ccm_mpid
= ntohll(ccm
->mpid64
);
803 ccm_opdown
= ccm
->opdown
;
805 ccm_mpid
= ntohs(ccm
->mpid
);
808 ccm_seq
= ntohl(ccm
->seq
);
810 if (ccm_interval
!= cfm
->ccm_interval
) {
811 VLOG_WARN_RL(&rl
, "%s: received a CCM with an unexpected interval"
812 " (%"PRIu8
") from RMP %"PRIu64
, cfm
->name
,
813 ccm_interval
, ccm_mpid
);
816 if (extended
&& ccm_interval
== 0
817 && ccm_interval_ms_x
!= cfm
->ccm_interval_ms
) {
818 VLOG_WARN_RL(&rl
, "%s: received a CCM with an unexpected extended"
819 " interval (%"PRIu16
"ms) from RMP %"PRIu64
, cfm
->name
,
820 ccm_interval_ms_x
, ccm_mpid
);
823 rmp
= lookup_remote_mp(cfm
, ccm_mpid
);
825 if (hmap_count(&cfm
->remote_mps
) < CFM_MAX_RMPS
) {
826 rmp
= xzalloc(sizeof *rmp
);
827 hmap_insert(&cfm
->remote_mps
, &rmp
->node
, hash_mpid(ccm_mpid
));
829 cfm_fault
|= CFM_FAULT_OVERFLOW
;
831 "%s: dropped CCM with MPID %"PRIu64
" from MAC "
832 ETH_ADDR_FMT
, cfm
->name
, ccm_mpid
,
833 ETH_ADDR_ARGS(eth
->eth_src
));
838 cfm_fault
|= CFM_FAULT_RDI
;
839 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64
, cfm
->name
,
843 VLOG_DBG("%s: received CCM (seq %"PRIu32
") (mpid %"PRIu64
")"
844 " (interval %"PRIu8
") (RDI %s)", cfm
->name
, ccm_seq
,
845 ccm_mpid
, ccm_interval
, ccm_rdi
? "true" : "false");
848 if (rmp
->mpid
== cfm
->mpid
) {
849 cfm_fault
|= CFM_FAULT_LOOPBACK
;
850 VLOG_WARN_RL(&rl
,"%s: received CCM with local MPID"
851 " %"PRIu64
, cfm
->name
, rmp
->mpid
);
854 if (rmp
->seq
&& ccm_seq
!= (rmp
->seq
+ 1)) {
855 VLOG_WARN_RL(&rl
, "%s: (mpid %"PRIu64
") detected sequence"
856 " numbers which indicate possible connectivity"
857 " problems (previous %"PRIu32
") (current %"PRIu32
858 ")", cfm
->name
, ccm_mpid
, rmp
->seq
, ccm_seq
);
861 rmp
->mpid
= ccm_mpid
;
863 rmp
->num_health_ccm
++;
865 timer_set_duration(&cfm
->demand_rx_ccm_t
,
866 100 * cfm
->ccm_interval_ms
);
870 cfm
->recv_fault
|= cfm_fault
;
872 rmp
->opup
= !ccm_opdown
;
873 rmp
->last_rx
= time_msec();
878 ovs_mutex_unlock(&mutex
);
881 /* Returns and resets the 'cfm->status_changed'. */
883 cfm_check_status_change(struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
887 ovs_mutex_lock(&mutex
);
888 ret
= cfm
->status_changed
;
889 cfm
->status_changed
= false;
890 ovs_mutex_unlock(&mutex
);
896 cfm_get_fault__(const struct cfm
*cfm
) OVS_REQUIRES(mutex
)
898 if (cfm
->fault_override
>= 0) {
899 return cfm
->fault_override
? CFM_FAULT_OVERRIDE
: 0;
904 /* Gets the fault status of 'cfm'. Returns a bit mask of 'cfm_fault_reason's
905 * indicating the cause of the connectivity fault, or zero if there is no
908 cfm_get_fault(const struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
912 ovs_mutex_lock(&mutex
);
913 fault
= cfm_get_fault__(cfm
);
914 ovs_mutex_unlock(&mutex
);
918 /* Gets the number of cfm fault flapping since start. */
920 cfm_get_flap_count(const struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
923 ovs_mutex_lock(&mutex
);
924 flap_count
= cfm
->flap_count
;
925 ovs_mutex_unlock(&mutex
);
929 /* Gets the health of 'cfm'. Returns an integer between 0 and 100 indicating
930 * the health of the link as a percentage of ccm frames received in
931 * CFM_HEALTH_INTERVAL * 'fault_interval' if there is only 1 remote_mpid,
932 * returns 0 if there are no remote_mpids, and returns -1 if there are more
933 * than 1 remote_mpids. */
935 cfm_get_health(const struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
939 ovs_mutex_lock(&mutex
);
940 health
= cfm
->health
;
941 ovs_mutex_unlock(&mutex
);
946 cfm_get_opup__(const struct cfm
*cfm_
) OVS_REQUIRES(mutex
)
948 struct cfm
*cfm
= CONST_CAST(struct cfm
*, cfm_
);
951 atomic_read_relaxed(&cfm
->extended
, &extended
);
953 return extended
? cfm
->remote_opup
: -1;
956 /* Gets the operational state of 'cfm'. 'cfm' is considered operationally down
957 * if it has received a CCM with the operationally down bit set from any of its
958 * remote maintenance points. Returns 1 if 'cfm' is operationally up, 0 if
959 * 'cfm' is operationally down, or -1 if 'cfm' has no operational state
960 * (because it isn't in extended mode). */
962 cfm_get_opup(const struct cfm
*cfm
) OVS_EXCLUDED(mutex
)
966 ovs_mutex_lock(&mutex
);
967 opup
= cfm_get_opup__(cfm
);
968 ovs_mutex_unlock(&mutex
);
974 cfm_get_remote_mpids__(const struct cfm
*cfm
, uint64_t **rmps
, size_t *n_rmps
)
977 *rmps
= xmemdup(cfm
->rmps_array
, cfm
->rmps_array_len
* sizeof **rmps
);
978 *n_rmps
= cfm
->rmps_array_len
;
981 /* Populates 'rmps' with an array of remote maintenance points reachable by
982 * 'cfm'. The number of remote maintenance points is written to 'n_rmps'.
983 * 'cfm' retains ownership of the array written to 'rmps' */
985 cfm_get_remote_mpids(const struct cfm
*cfm
, uint64_t **rmps
, size_t *n_rmps
)
988 ovs_mutex_lock(&mutex
);
989 cfm_get_remote_mpids__(cfm
, rmps
, n_rmps
);
990 ovs_mutex_unlock(&mutex
);
993 /* Extracts the status of 'cfm' and fills in the 's'. */
995 cfm_get_status(const struct cfm
*cfm
, struct cfm_status
*s
) OVS_EXCLUDED(mutex
)
997 ovs_mutex_lock(&mutex
);
998 s
->faults
= cfm_get_fault__(cfm
);
999 s
->remote_opstate
= cfm_get_opup__(cfm
);
1000 s
->flap_count
= cfm
->flap_count
;
1001 s
->health
= cfm
->health
;
1002 cfm_get_remote_mpids__(cfm
, &s
->rmps
, &s
->n_rmps
);
1003 ovs_mutex_unlock(&mutex
);
1007 cfm_find(const char *name
) OVS_REQUIRES(mutex
)
1011 HMAP_FOR_EACH_WITH_HASH (cfm
, hmap_node
, hash_string(name
, 0), all_cfms
) {
1012 if (!strcmp(cfm
->name
, name
)) {
1020 cfm_print_details(struct ds
*ds
, struct cfm
*cfm
) OVS_REQUIRES(mutex
)
1022 struct remote_mp
*rmp
;
1026 atomic_read_relaxed(&cfm
->extended
, &extended
);
1028 ds_put_format(ds
, "---- %s ----\n", cfm
->name
);
1029 ds_put_format(ds
, "MPID %"PRIu64
":%s%s\n", cfm
->mpid
,
1030 extended
? " extended" : "",
1031 cfm
->fault_override
>= 0 ? " fault_override" : "");
1033 fault
= cfm_get_fault__(cfm
);
1035 ds_put_cstr(ds
, " fault: ");
1036 ds_put_cfm_fault(ds
, fault
);
1037 ds_put_cstr(ds
, "\n");
1040 if (cfm
->health
== -1) {
1041 ds_put_format(ds
, " average health: undefined\n");
1043 ds_put_format(ds
, " average health: %d\n", cfm
->health
);
1045 ds_put_format(ds
, " opstate: %s\n", cfm
->opup
? "up" : "down");
1046 ds_put_format(ds
, " remote_opstate: %s\n",
1047 cfm
->remote_opup
? "up" : "down");
1048 ds_put_format(ds
, " interval: %dms\n", cfm
->ccm_interval_ms
);
1049 ds_put_format(ds
, " next CCM tx: %lldms\n",
1050 timer_msecs_until_expired(&cfm
->tx_timer
));
1051 ds_put_format(ds
, " next fault check: %lldms\n",
1052 timer_msecs_until_expired(&cfm
->fault_timer
));
1054 HMAP_FOR_EACH (rmp
, node
, &cfm
->remote_mps
) {
1055 ds_put_format(ds
, "Remote MPID %"PRIu64
"\n", rmp
->mpid
);
1056 ds_put_format(ds
, " recv since check: %s\n",
1057 rmp
->recv
? "true" : "false");
1058 ds_put_format(ds
, " opstate: %s\n", rmp
->opup
? "up" : "down");
1063 cfm_unixctl_show(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
1064 void *aux OVS_UNUSED
) OVS_EXCLUDED(mutex
)
1066 struct ds ds
= DS_EMPTY_INITIALIZER
;
1069 ovs_mutex_lock(&mutex
);
1071 cfm
= cfm_find(argv
[1]);
1073 unixctl_command_reply_error(conn
, "no such CFM object");
1076 cfm_print_details(&ds
, cfm
);
1078 HMAP_FOR_EACH (cfm
, hmap_node
, all_cfms
) {
1079 cfm_print_details(&ds
, cfm
);
1083 unixctl_command_reply(conn
, ds_cstr(&ds
));
1086 ovs_mutex_unlock(&mutex
);
1090 cfm_unixctl_set_fault(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
1091 void *aux OVS_UNUSED
) OVS_EXCLUDED(mutex
)
1093 const char *fault_str
= argv
[argc
- 1];
1097 ovs_mutex_lock(&mutex
);
1098 if (!strcasecmp("true", fault_str
)) {
1100 } else if (!strcasecmp("false", fault_str
)) {
1102 } else if (!strcasecmp("normal", fault_str
)) {
1103 fault_override
= -1;
1105 unixctl_command_reply_error(conn
, "unknown fault string");
1110 cfm
= cfm_find(argv
[1]);
1112 unixctl_command_reply_error(conn
, "no such CFM object");
1115 cfm
->fault_override
= fault_override
;
1116 cfm_status_changed(cfm
);
1118 HMAP_FOR_EACH (cfm
, hmap_node
, all_cfms
) {
1119 cfm
->fault_override
= fault_override
;
1120 cfm_status_changed(cfm
);
1124 unixctl_command_reply(conn
, "OK");
1127 ovs_mutex_unlock(&mutex
);