2 * Copyright (c) 2010, 2011 Nicira Networks.
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.
25 #include "byte-order.h"
26 #include "dynamic-string.h"
32 #include "poll-loop.h"
38 VLOG_DEFINE_THIS_MODULE(cfm
);
40 #define CFM_MAX_RMPS 256
42 /* Ethernet destination address of CCM packets. */
43 static const uint8_t eth_addr_ccm
[6] = { 0x01, 0x80, 0xC2, 0x00, 0x00, 0x30 };
44 static const uint8_t eth_addr_ccm_x
[6] = {
45 0x01, 0x23, 0x20, 0x00, 0x00, 0x30
48 #define ETH_TYPE_CFM 0x8902
50 /* A 'ccm' represents a Continuity Check Message from the 802.1ag
51 * specification. Continuity Check Messages are broadcast periodically so that
52 * hosts can determine whom they have connectivity to.
54 * The minimum length of a CCM as specified by IEEE 802.1ag is 75 bytes.
55 * Previous versions of Open vSwitch generated 74-byte CCM messages, so we
56 * accept such messages too. */
58 #define CCM_ACCEPT_LEN 74
59 #define CCM_MAID_LEN 48
60 #define CCM_OPCODE 1 /* CFM message opcode meaning CCM. */
61 #define CCM_RDI_MASK 0x80
63 uint8_t mdlevel_version
; /* MD Level and Version */
69 uint8_t maid
[CCM_MAID_LEN
];
71 /* Defined by ITU-T Y.1731 should be zero */
72 ovs_be16 interval_ms_x
; /* Transmission interval in ms. */
73 ovs_be64 mpid64
; /* MPID in extended mode. */
74 uint8_t opdown
; /* Operationally down. */
79 } __attribute__((packed
));
80 BUILD_ASSERT_DECL(CCM_LEN
== sizeof(struct ccm
));
83 char *name
; /* Name of this CFM object. */
84 struct hmap_node hmap_node
; /* Node in all_cfms list. */
87 bool extended
; /* Extended mode. */
88 bool fault
; /* Indicates connectivity fault. */
89 bool unexpected_recv
; /* Received an unexpected CCM. */
90 bool opup
; /* Operational State. */
91 bool remote_opup
; /* Remote Operational State. */
93 int fault_override
; /* Manual override of 'fault' status.
94 Ignored if negative. */
96 uint32_t seq
; /* The sequence number of our last CCM. */
97 uint8_t ccm_interval
; /* The CCM transmission interval. */
98 int ccm_interval_ms
; /* 'ccm_interval' in milliseconds. */
99 uint16_t ccm_vlan
; /* Vlan tag of CCM PDUs. */
100 uint8_t ccm_pcp
; /* Priority of CCM PDUs. */
101 uint8_t maid
[CCM_MAID_LEN
]; /* The MAID of this CFM. */
103 struct timer tx_timer
; /* Send CCM when expired. */
104 struct timer fault_timer
; /* Check for faults when expired. */
106 struct hmap remote_mps
; /* Remote MPs. */
108 /* Result of cfm_get_remote_mpids(). Updated only during fault check to
110 uint64_t *rmps_array
; /* Cache of remote_mps. */
111 size_t rmps_array_len
; /* Number of rmps in 'rmps_array'. */
114 /* Remote MPs represent foreign network entities that are configured to have
115 * the same MAID as this CFM instance. */
117 uint64_t mpid
; /* The Maintenance Point ID of this 'remote_mp'. */
118 struct hmap_node node
; /* Node in 'remote_mps' map. */
120 bool recv
; /* CCM was received since last fault check. */
121 bool rdi
; /* Remote Defect Indicator. Indicates remote_mp isn't
122 receiving CCMs that it's expecting to. */
123 bool opup
; /* Operational State. */
124 uint32_t seq
; /* Most recently received sequence number. */
127 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(20, 30);
128 static struct hmap all_cfms
= HMAP_INITIALIZER(&all_cfms
);
130 static unixctl_cb_func cfm_unixctl_show
;
131 static unixctl_cb_func cfm_unixctl_set_fault
;
133 static const uint8_t *
134 cfm_ccm_addr(const struct cfm
*cfm
)
136 return cfm
->extended
? eth_addr_ccm_x
: eth_addr_ccm
;
140 cfm_generate_maid(struct cfm
*cfm
)
142 const char *ovs_md_name
= "ovs";
143 const char *ovs_ma_name
= "ovs";
145 size_t md_len
, ma_len
;
147 memset(cfm
->maid
, 0, CCM_MAID_LEN
);
149 md_len
= strlen(ovs_md_name
);
150 ma_len
= strlen(ovs_ma_name
);
152 assert(md_len
&& ma_len
&& md_len
+ ma_len
+ 4 <= CCM_MAID_LEN
);
154 cfm
->maid
[0] = 4; /* MD name string format. */
155 cfm
->maid
[1] = md_len
; /* MD name size. */
156 memcpy(&cfm
->maid
[2], ovs_md_name
, md_len
); /* MD name. */
158 ma_p
= cfm
->maid
+ 2 + md_len
;
159 ma_p
[0] = 2; /* MA name string format. */
160 ma_p
[1] = ma_len
; /* MA name size. */
161 memcpy(&ma_p
[2], ovs_ma_name
, ma_len
); /* MA name. */
165 ccm_interval_to_ms(uint8_t interval
)
168 case 0: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
169 case 1: return 3; /* Not recommended due to timer resolution. */
170 case 2: return 10; /* Not recommended due to timer resolution. */
173 case 5: return 10000;
174 case 6: return 60000;
175 case 7: return 600000;
176 default: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
183 cfm_fault_interval(struct cfm
*cfm
)
185 /* According to the 802.1ag specification we should assume every other MP
186 * with the same MAID has the same transmission interval that we have. If
187 * an MP has a different interval, cfm_process_heartbeat will register it
188 * as a fault (likely due to a configuration error). Thus we can check all
189 * MPs at once making this quite a bit simpler.
191 * According to the specification we should check when (ccm_interval_ms *
192 * 3.5)ms have passed. */
193 return (cfm
->ccm_interval_ms
* 7) / 2;
197 ms_to_ccm_interval(int interval_ms
)
201 for (i
= 7; i
> 0; i
--) {
202 if (ccm_interval_to_ms(i
) <= interval_ms
) {
211 hash_mpid(uint64_t mpid
)
213 return hash_bytes(&mpid
, sizeof mpid
, 0);
217 cfm_is_valid_mpid(bool extended
, uint64_t mpid
)
219 /* 802.1ag specification requires MPIDs to be within the range [1, 8191].
220 * In extended mode we relax this requirement. */
221 return mpid
>= 1 && (extended
|| mpid
<= 8191);
224 static struct remote_mp
*
225 lookup_remote_mp(const struct cfm
*cfm
, uint64_t mpid
)
227 struct remote_mp
*rmp
;
229 HMAP_FOR_EACH_IN_BUCKET (rmp
, node
, hash_mpid(mpid
), &cfm
->remote_mps
) {
230 if (rmp
->mpid
== mpid
) {
241 unixctl_command_register("cfm/show", "[interface]", 0, 1, cfm_unixctl_show
,
243 unixctl_command_register("cfm/set-fault", "[interface] normal|false|true",
244 1, 2, cfm_unixctl_set_fault
, NULL
);
247 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
248 * cfm_configure() before use. */
250 cfm_create(const char *name
)
254 cfm
= xzalloc(sizeof *cfm
);
255 cfm
->name
= xstrdup(name
);
256 hmap_init(&cfm
->remote_mps
);
257 cfm_generate_maid(cfm
);
258 hmap_insert(&all_cfms
, &cfm
->hmap_node
, hash_string(cfm
->name
, 0));
259 cfm
->remote_opup
= true;
260 cfm
->fault_override
= -1;
265 cfm_destroy(struct cfm
*cfm
)
267 struct remote_mp
*rmp
, *rmp_next
;
273 HMAP_FOR_EACH_SAFE (rmp
, rmp_next
, node
, &cfm
->remote_mps
) {
274 hmap_remove(&cfm
->remote_mps
, &rmp
->node
);
278 hmap_destroy(&cfm
->remote_mps
);
279 hmap_remove(&all_cfms
, &cfm
->hmap_node
);
280 free(cfm
->rmps_array
);
285 /* Should be run periodically to update fault statistics messages. */
287 cfm_run(struct cfm
*cfm
)
289 if (timer_expired(&cfm
->fault_timer
)) {
290 long long int interval
= cfm_fault_interval(cfm
);
291 struct remote_mp
*rmp
, *rmp_next
;
292 bool old_cfm_fault
= cfm
->fault
;
294 cfm
->fault
= cfm
->unexpected_recv
;
295 cfm
->unexpected_recv
= false;
297 cfm
->rmps_array_len
= 0;
298 free(cfm
->rmps_array
);
299 cfm
->rmps_array
= xmalloc(hmap_count(&cfm
->remote_mps
) *
300 sizeof *cfm
->rmps_array
);
302 cfm
->remote_opup
= true;
303 HMAP_FOR_EACH_SAFE (rmp
, rmp_next
, node
, &cfm
->remote_mps
) {
306 VLOG_DBG("%s: no CCM from RMP %"PRIu64
" in the last %lldms",
307 cfm
->name
, rmp
->mpid
, interval
);
308 hmap_remove(&cfm
->remote_mps
, &rmp
->node
);
313 if (rmp
->mpid
== cfm
->mpid
) {
314 VLOG_WARN_RL(&rl
,"%s: received CCM with local MPID"
315 " %"PRIu64
, cfm
->name
, rmp
->mpid
);
320 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64
, cfm
->name
,
326 cfm
->remote_opup
= rmp
->opup
;
329 cfm
->rmps_array
[cfm
->rmps_array_len
++] = rmp
->mpid
;
333 if (hmap_is_empty(&cfm
->remote_mps
)) {
337 if (old_cfm_fault
!= cfm
->fault
) {
338 VLOG_INFO_RL(&rl
, "%s: CFM fault status changed to %s",
339 cfm
->name
, cfm
->fault
? "true" : "false");
342 timer_set_duration(&cfm
->fault_timer
, interval
);
346 /* Should be run periodically to check if the CFM module has a CCM message it
349 cfm_should_send_ccm(struct cfm
*cfm
)
351 return timer_expired(&cfm
->tx_timer
);
354 /* Composes a CCM message into 'packet'. Messages generated with this function
355 * should be sent whenever cfm_should_send_ccm() indicates. */
357 cfm_compose_ccm(struct cfm
*cfm
, struct ofpbuf
*packet
,
358 uint8_t eth_src
[ETH_ADDR_LEN
])
362 timer_set_duration(&cfm
->tx_timer
, cfm
->ccm_interval_ms
);
363 eth_compose(packet
, cfm_ccm_addr(cfm
), eth_src
, ETH_TYPE_CFM
, sizeof *ccm
);
365 if (cfm
->ccm_vlan
|| cfm
->ccm_pcp
) {
366 uint16_t tci
= cfm
->ccm_vlan
| (cfm
->ccm_pcp
<< VLAN_PCP_SHIFT
);
367 eth_push_vlan(packet
, htons(tci
));
371 ccm
->mdlevel_version
= 0;
372 ccm
->opcode
= CCM_OPCODE
;
373 ccm
->tlv_offset
= 70;
374 ccm
->seq
= htonl(++cfm
->seq
);
375 ccm
->flags
= cfm
->ccm_interval
;
376 memcpy(ccm
->maid
, cfm
->maid
, sizeof ccm
->maid
);
377 memset(ccm
->zero
, 0, sizeof ccm
->zero
);
381 ccm
->mpid
= htons(hash_mpid(cfm
->mpid
));
382 ccm
->mpid64
= htonll(cfm
->mpid
);
383 ccm
->opdown
= !cfm
->opup
;
385 ccm
->mpid
= htons(cfm
->mpid
);
386 ccm
->mpid64
= htonll(0);
390 if (cfm
->ccm_interval
== 0) {
391 assert(cfm
->extended
);
392 ccm
->interval_ms_x
= htons(cfm
->ccm_interval_ms
);
395 if (hmap_is_empty(&cfm
->remote_mps
)) {
396 ccm
->flags
|= CCM_RDI_MASK
;
401 cfm_wait(struct cfm
*cfm
)
403 timer_wait(&cfm
->tx_timer
);
404 timer_wait(&cfm
->fault_timer
);
407 /* Configures 'cfm' with settings from 's'. */
409 cfm_configure(struct cfm
*cfm
, const struct cfm_settings
*s
)
414 if (!cfm_is_valid_mpid(s
->extended
, s
->mpid
) || s
->interval
<= 0) {
419 cfm
->extended
= s
->extended
;
421 interval
= ms_to_ccm_interval(s
->interval
);
422 interval_ms
= ccm_interval_to_ms(interval
);
424 cfm
->ccm_vlan
= s
->ccm_vlan
& VLAN_VID_MASK
;
425 cfm
->ccm_pcp
= s
->ccm_pcp
& (VLAN_PCP_MASK
>> VLAN_PCP_SHIFT
);
426 if (cfm
->extended
&& interval_ms
!= s
->interval
) {
428 interval_ms
= MIN(s
->interval
, UINT16_MAX
);
431 if (interval
!= cfm
->ccm_interval
|| interval_ms
!= cfm
->ccm_interval_ms
) {
432 cfm
->ccm_interval
= interval
;
433 cfm
->ccm_interval_ms
= interval_ms
;
435 timer_set_expired(&cfm
->tx_timer
);
436 timer_set_duration(&cfm
->fault_timer
, cfm_fault_interval(cfm
));
442 /* Returns true if 'cfm' should process packets from 'flow'. */
444 cfm_should_process_flow(const struct cfm
*cfm
, const struct flow
*flow
)
446 return (ntohs(flow
->dl_type
) == ETH_TYPE_CFM
447 && eth_addr_equals(flow
->dl_dst
, cfm_ccm_addr(cfm
)));
450 /* Updates internal statistics relevant to packet 'p'. Should be called on
451 * every packet whose flow returned true when passed to
452 * cfm_should_process_flow. */
454 cfm_process_heartbeat(struct cfm
*cfm
, const struct ofpbuf
*p
)
457 struct eth_header
*eth
;
460 ccm
= ofpbuf_at(p
, (uint8_t *)p
->l3
- (uint8_t *)p
->data
, CCM_ACCEPT_LEN
);
463 VLOG_INFO_RL(&rl
, "%s: Received an unparseable 802.1ag CCM heartbeat.",
468 if (ccm
->opcode
!= CCM_OPCODE
) {
469 VLOG_INFO_RL(&rl
, "%s: Received an unsupported 802.1ag message. "
470 "(opcode %u)", cfm
->name
, ccm
->opcode
);
474 /* According to the 802.1ag specification, reception of a CCM with an
475 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
476 * trigger a fault. We ignore this requirement for several reasons.
478 * Faults can cause a controller or Open vSwitch to make potentially
479 * expensive changes to the network topology. It seems prudent to trigger
480 * them judiciously, especially when CFM is used to check slave status of
481 * bonds. Furthermore, faults can be maliciously triggered by crafting
483 if (memcmp(ccm
->maid
, cfm
->maid
, sizeof ccm
->maid
)) {
484 cfm
->unexpected_recv
= true;
485 VLOG_WARN_RL(&rl
, "%s: Received unexpected remote MAID from MAC "
486 ETH_ADDR_FMT
, cfm
->name
, ETH_ADDR_ARGS(eth
->eth_src
));
488 uint8_t ccm_interval
= ccm
->flags
& 0x7;
489 bool ccm_rdi
= ccm
->flags
& CCM_RDI_MASK
;
490 uint16_t ccm_interval_ms_x
= ntohs(ccm
->interval_ms_x
);
492 struct remote_mp
*rmp
;
498 ccm_mpid
= ntohll(ccm
->mpid64
);
499 ccm_opdown
= ccm
->opdown
;
501 ccm_mpid
= ntohs(ccm
->mpid
);
504 ccm_seq
= ntohl(ccm
->seq
);
506 if (ccm_interval
!= cfm
->ccm_interval
) {
507 VLOG_WARN_RL(&rl
, "%s: received a CCM with an invalid interval"
508 " (%"PRIu8
") from RMP %"PRIu64
, cfm
->name
,
509 ccm_interval
, ccm_mpid
);
512 if (cfm
->extended
&& ccm_interval
== 0
513 && ccm_interval_ms_x
!= cfm
->ccm_interval_ms
) {
514 VLOG_WARN_RL(&rl
, "%s: received a CCM with an invalid extended"
515 " interval (%"PRIu16
"ms) from RMP %"PRIu64
, cfm
->name
,
516 ccm_interval_ms_x
, ccm_mpid
);
519 rmp
= lookup_remote_mp(cfm
, ccm_mpid
);
521 if (hmap_count(&cfm
->remote_mps
) < CFM_MAX_RMPS
) {
522 rmp
= xzalloc(sizeof *rmp
);
523 hmap_insert(&cfm
->remote_mps
, &rmp
->node
, hash_mpid(ccm_mpid
));
525 cfm
->unexpected_recv
= true;
527 "%s: dropped CCM with MPID %"PRIu64
" from MAC "
528 ETH_ADDR_FMT
, cfm
->name
, ccm_mpid
,
529 ETH_ADDR_ARGS(eth
->eth_src
));
533 VLOG_DBG("%s: received CCM (seq %"PRIu32
") (mpid %"PRIu64
")"
534 " (interval %"PRIu8
") (RDI %s)", cfm
->name
, ccm_seq
,
535 ccm_mpid
, ccm_interval
, ccm_rdi
? "true" : "false");
538 if (rmp
->seq
&& ccm_seq
!= (rmp
->seq
+ 1)) {
539 VLOG_WARN_RL(&rl
, "%s: (mpid %"PRIu64
") detected sequence"
540 " numbers which indicate possible connectivity"
541 " problems (previous %"PRIu32
") (current %"PRIu32
542 ")", cfm
->name
, ccm_mpid
, rmp
->seq
, ccm_seq
);
545 rmp
->mpid
= ccm_mpid
;
549 rmp
->opup
= !ccm_opdown
;
554 /* Gets the fault status of 'cfm'. Returns true when 'cfm' has detected
555 * connectivity problems, false otherwise. */
557 cfm_get_fault(const struct cfm
*cfm
)
559 if (cfm
->fault_override
>= 0) {
560 return cfm
->fault_override
;
565 /* Gets the operational state of 'cfm'. 'cfm' is considered operationally down
566 * if it has received a CCM with the operationally down bit set from any of its
567 * remote maintenance points. Returns true if 'cfm' is operationally up. False
570 cfm_get_opup(const struct cfm
*cfm
)
572 return cfm
->remote_opup
;
575 /* Populates 'rmps' with an array of remote maintenance points reachable by
576 * 'cfm'. The number of remote maintenance points is written to 'n_rmps'.
577 * 'cfm' retains ownership of the array written to 'rmps' */
579 cfm_get_remote_mpids(const struct cfm
*cfm
, const uint64_t **rmps
,
582 *rmps
= cfm
->rmps_array
;
583 *n_rmps
= cfm
->rmps_array_len
;
587 cfm_find(const char *name
)
591 HMAP_FOR_EACH_WITH_HASH (cfm
, hmap_node
, hash_string(name
, 0), &all_cfms
) {
592 if (!strcmp(cfm
->name
, name
)) {
600 cfm_print_details(struct ds
*ds
, const struct cfm
*cfm
)
602 struct remote_mp
*rmp
;
604 ds_put_format(ds
, "---- %s ----\n", cfm
->name
);
605 ds_put_format(ds
, "MPID %"PRIu64
":%s%s%s%s\n", cfm
->mpid
,
606 cfm
->extended
? " extended" : "",
607 cfm_get_fault(cfm
) ? " fault" : "",
608 cfm
->fault_override
>= 0 ? " fault_override" : "",
609 cfm
->unexpected_recv
? " unexpected_recv" : "");
611 ds_put_format(ds
, "\topstate: %s\n", cfm
->opup
? "up" : "down");
612 ds_put_format(ds
, "\tremote_opstate: %s\n",
613 cfm
->remote_opup
? "up" : "down");
614 ds_put_format(ds
, "\tinterval: %dms\n", cfm
->ccm_interval_ms
);
615 ds_put_format(ds
, "\tnext CCM tx: %lldms\n",
616 timer_msecs_until_expired(&cfm
->tx_timer
));
617 ds_put_format(ds
, "\tnext fault check: %lldms\n",
618 timer_msecs_until_expired(&cfm
->fault_timer
));
620 HMAP_FOR_EACH (rmp
, node
, &cfm
->remote_mps
) {
621 ds_put_format(ds
, "Remote MPID %"PRIu64
":%s\n",
623 rmp
->rdi
? " rdi" : "");
624 ds_put_format(ds
, "\trecv since check: %s\n",
625 rmp
->recv
? "true" : "false");
626 ds_put_format(ds
, "\topstate: %s\n", rmp
->opup
? "up" : "down");
631 cfm_unixctl_show(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
632 void *aux OVS_UNUSED
)
634 struct ds ds
= DS_EMPTY_INITIALIZER
;
635 const struct cfm
*cfm
;
638 cfm
= cfm_find(argv
[1]);
640 unixctl_command_reply(conn
, 501, "no such CFM object");
643 cfm_print_details(&ds
, cfm
);
645 HMAP_FOR_EACH (cfm
, hmap_node
, &all_cfms
) {
646 cfm_print_details(&ds
, cfm
);
650 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
655 cfm_unixctl_set_fault(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
656 void *aux OVS_UNUSED
)
658 const char *fault_str
= argv
[argc
- 1];
662 if (!strcasecmp("true", fault_str
)) {
664 } else if (!strcasecmp("false", fault_str
)) {
666 } else if (!strcasecmp("normal", fault_str
)) {
669 unixctl_command_reply(conn
, 501, "unknown fault string");
674 cfm
= cfm_find(argv
[1]);
676 unixctl_command_reply(conn
, 501, "no such CFM object");
679 cfm
->fault_override
= fault_override
;
681 HMAP_FOR_EACH (cfm
, hmap_node
, &all_cfms
) {
682 cfm
->fault_override
= fault_override
;
686 unixctl_command_reply(conn
, 200, "OK");