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 int fault; /* Connectivity fault status. */
89 int recv_fault; /* Bit mask of faults occuring on receive. */
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;
139 /* Returns the string representation of the given cfm_fault_reason 'reason'. */
141 cfm_fault_reason_to_str(int reason) {
143 #define CFM_FAULT_REASON(NAME, STR) case CFM_FAULT_##NAME: return #STR;
145 #undef CFM_FAULT_REASON
146 default: return "<unknown>";
151 ds_put_cfm_fault(struct ds *ds, int fault)
153 size_t length = ds->length;
156 for (i = 0; i < CFM_FAULT_N_REASONS; i++) {
159 if (fault & reason) {
160 ds_put_format(ds, "%s ", cfm_fault_reason_to_str(reason));
164 if (ds->length > length) {
165 ds_truncate(ds, ds->length - 1);
170 cfm_generate_maid(struct cfm *cfm)
172 const char *ovs_md_name = "ovs";
173 const char *ovs_ma_name = "ovs";
175 size_t md_len, ma_len;
177 memset(cfm->maid, 0, CCM_MAID_LEN);
179 md_len = strlen(ovs_md_name);
180 ma_len = strlen(ovs_ma_name);
182 assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN);
184 cfm->maid[0] = 4; /* MD name string format. */
185 cfm->maid[1] = md_len; /* MD name size. */
186 memcpy(&cfm->maid[2], ovs_md_name, md_len); /* MD name. */
188 ma_p = cfm->maid + 2 + md_len;
189 ma_p[0] = 2; /* MA name string format. */
190 ma_p[1] = ma_len; /* MA name size. */
191 memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */
195 ccm_interval_to_ms(uint8_t interval)
198 case 0: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
199 case 1: return 3; /* Not recommended due to timer resolution. */
200 case 2: return 10; /* Not recommended due to timer resolution. */
203 case 5: return 10000;
204 case 6: return 60000;
205 case 7: return 600000;
206 default: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
213 cfm_fault_interval(struct cfm *cfm)
215 /* According to the 802.1ag specification we should assume every other MP
216 * with the same MAID has the same transmission interval that we have. If
217 * an MP has a different interval, cfm_process_heartbeat will register it
218 * as a fault (likely due to a configuration error). Thus we can check all
219 * MPs at once making this quite a bit simpler.
221 * According to the specification we should check when (ccm_interval_ms *
222 * 3.5)ms have passed. */
223 return (cfm->ccm_interval_ms * 7) / 2;
227 ms_to_ccm_interval(int interval_ms)
231 for (i = 7; i > 0; i--) {
232 if (ccm_interval_to_ms(i) <= interval_ms) {
241 hash_mpid(uint64_t mpid)
243 return hash_bytes(&mpid, sizeof mpid, 0);
247 cfm_is_valid_mpid(bool extended, uint64_t mpid)
249 /* 802.1ag specification requires MPIDs to be within the range [1, 8191].
250 * In extended mode we relax this requirement. */
251 return mpid >= 1 && (extended || mpid <= 8191);
254 static struct remote_mp *
255 lookup_remote_mp(const struct cfm *cfm, uint64_t mpid)
257 struct remote_mp *rmp;
259 HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), &cfm->remote_mps) {
260 if (rmp->mpid == mpid) {
271 unixctl_command_register("cfm/show", "[interface]", 0, 1, cfm_unixctl_show,
273 unixctl_command_register("cfm/set-fault", "[interface] normal|false|true",
274 1, 2, cfm_unixctl_set_fault, NULL);
277 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
278 * cfm_configure() before use. */
280 cfm_create(const char *name)
284 cfm = xzalloc(sizeof *cfm);
285 cfm->name = xstrdup(name);
286 hmap_init(&cfm->remote_mps);
287 cfm_generate_maid(cfm);
288 hmap_insert(&all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0));
289 cfm->remote_opup = true;
290 cfm->fault_override = -1;
295 cfm_destroy(struct cfm *cfm)
297 struct remote_mp *rmp, *rmp_next;
303 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
304 hmap_remove(&cfm->remote_mps, &rmp->node);
308 hmap_destroy(&cfm->remote_mps);
309 hmap_remove(&all_cfms, &cfm->hmap_node);
310 free(cfm->rmps_array);
315 /* Should be run periodically to update fault statistics messages. */
317 cfm_run(struct cfm *cfm)
319 if (timer_expired(&cfm->fault_timer)) {
320 long long int interval = cfm_fault_interval(cfm);
321 struct remote_mp *rmp, *rmp_next;
322 bool old_cfm_fault = cfm->fault;
324 cfm->fault = cfm->recv_fault;
327 cfm->rmps_array_len = 0;
328 free(cfm->rmps_array);
329 cfm->rmps_array = xmalloc(hmap_count(&cfm->remote_mps) *
330 sizeof *cfm->rmps_array);
332 cfm->remote_opup = true;
333 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
336 VLOG_DBG("%s: no CCM from RMP %"PRIu64" in the last %lldms",
337 cfm->name, rmp->mpid, interval);
338 hmap_remove(&cfm->remote_mps, &rmp->node);
343 if (rmp->mpid == cfm->mpid) {
344 VLOG_WARN_RL(&rl,"%s: received CCM with local MPID"
345 " %"PRIu64, cfm->name, rmp->mpid);
346 cfm->fault |= CFM_FAULT_LOOPBACK;
350 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64, cfm->name,
352 cfm->fault |= CFM_FAULT_RDI;
356 cfm->remote_opup = rmp->opup;
359 cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid;
363 if (hmap_is_empty(&cfm->remote_mps)) {
364 cfm->fault |= CFM_FAULT_RECV;
367 if (old_cfm_fault != cfm->fault) {
368 struct ds ds = DS_EMPTY_INITIALIZER;
370 ds_put_cfm_fault(&ds, cfm->fault);
371 VLOG_INFO_RL(&rl, "%s: CFM fault status changed: %s", cfm->name,
376 timer_set_duration(&cfm->fault_timer, interval);
380 /* Should be run periodically to check if the CFM module has a CCM message it
383 cfm_should_send_ccm(struct cfm *cfm)
385 return timer_expired(&cfm->tx_timer);
388 /* Composes a CCM message into 'packet'. Messages generated with this function
389 * should be sent whenever cfm_should_send_ccm() indicates. */
391 cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet,
392 uint8_t eth_src[ETH_ADDR_LEN])
396 timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms);
397 eth_compose(packet, cfm_ccm_addr(cfm), eth_src, ETH_TYPE_CFM, sizeof *ccm);
399 if (cfm->ccm_vlan || cfm->ccm_pcp) {
400 uint16_t tci = cfm->ccm_vlan | (cfm->ccm_pcp << VLAN_PCP_SHIFT);
401 eth_push_vlan(packet, htons(tci));
405 ccm->mdlevel_version = 0;
406 ccm->opcode = CCM_OPCODE;
407 ccm->tlv_offset = 70;
408 ccm->seq = htonl(++cfm->seq);
409 ccm->flags = cfm->ccm_interval;
410 memcpy(ccm->maid, cfm->maid, sizeof ccm->maid);
411 memset(ccm->zero, 0, sizeof ccm->zero);
415 ccm->mpid = htons(hash_mpid(cfm->mpid));
416 ccm->mpid64 = htonll(cfm->mpid);
417 ccm->opdown = !cfm->opup;
419 ccm->mpid = htons(cfm->mpid);
420 ccm->mpid64 = htonll(0);
424 if (cfm->ccm_interval == 0) {
425 assert(cfm->extended);
426 ccm->interval_ms_x = htons(cfm->ccm_interval_ms);
429 if (hmap_is_empty(&cfm->remote_mps)) {
430 ccm->flags |= CCM_RDI_MASK;
435 cfm_wait(struct cfm *cfm)
437 timer_wait(&cfm->tx_timer);
438 timer_wait(&cfm->fault_timer);
441 /* Configures 'cfm' with settings from 's'. */
443 cfm_configure(struct cfm *cfm, const struct cfm_settings *s)
448 if (!cfm_is_valid_mpid(s->extended, s->mpid) || s->interval <= 0) {
453 cfm->extended = s->extended;
455 interval = ms_to_ccm_interval(s->interval);
456 interval_ms = ccm_interval_to_ms(interval);
458 cfm->ccm_vlan = s->ccm_vlan & VLAN_VID_MASK;
459 cfm->ccm_pcp = s->ccm_pcp & (VLAN_PCP_MASK >> VLAN_PCP_SHIFT);
460 if (cfm->extended && interval_ms != s->interval) {
462 interval_ms = MIN(s->interval, UINT16_MAX);
465 if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) {
466 cfm->ccm_interval = interval;
467 cfm->ccm_interval_ms = interval_ms;
469 timer_set_expired(&cfm->tx_timer);
470 timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm));
476 /* Returns true if 'cfm' should process packets from 'flow'. */
478 cfm_should_process_flow(const struct cfm *cfm, const struct flow *flow)
480 return (ntohs(flow->dl_type) == ETH_TYPE_CFM
481 && eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm)));
484 /* Updates internal statistics relevant to packet 'p'. Should be called on
485 * every packet whose flow returned true when passed to
486 * cfm_should_process_flow. */
488 cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
491 struct eth_header *eth;
494 ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_ACCEPT_LEN);
497 VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.",
502 if (ccm->opcode != CCM_OPCODE) {
503 VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. "
504 "(opcode %u)", cfm->name, ccm->opcode);
508 /* According to the 802.1ag specification, reception of a CCM with an
509 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
510 * trigger a fault. We ignore this requirement for several reasons.
512 * Faults can cause a controller or Open vSwitch to make potentially
513 * expensive changes to the network topology. It seems prudent to trigger
514 * them judiciously, especially when CFM is used to check slave status of
515 * bonds. Furthermore, faults can be maliciously triggered by crafting
517 if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) {
518 cfm->recv_fault |= CFM_FAULT_MAID;
519 VLOG_WARN_RL(&rl, "%s: Received unexpected remote MAID from MAC "
520 ETH_ADDR_FMT, cfm->name, ETH_ADDR_ARGS(eth->eth_src));
522 uint8_t ccm_interval = ccm->flags & 0x7;
523 bool ccm_rdi = ccm->flags & CCM_RDI_MASK;
524 uint16_t ccm_interval_ms_x = ntohs(ccm->interval_ms_x);
526 struct remote_mp *rmp;
532 ccm_mpid = ntohll(ccm->mpid64);
533 ccm_opdown = ccm->opdown;
535 ccm_mpid = ntohs(ccm->mpid);
538 ccm_seq = ntohl(ccm->seq);
540 if (ccm_interval != cfm->ccm_interval) {
541 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid interval"
542 " (%"PRIu8") from RMP %"PRIu64, cfm->name,
543 ccm_interval, ccm_mpid);
546 if (cfm->extended && ccm_interval == 0
547 && ccm_interval_ms_x != cfm->ccm_interval_ms) {
548 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid extended"
549 " interval (%"PRIu16"ms) from RMP %"PRIu64, cfm->name,
550 ccm_interval_ms_x, ccm_mpid);
553 rmp = lookup_remote_mp(cfm, ccm_mpid);
555 if (hmap_count(&cfm->remote_mps) < CFM_MAX_RMPS) {
556 rmp = xzalloc(sizeof *rmp);
557 hmap_insert(&cfm->remote_mps, &rmp->node, hash_mpid(ccm_mpid));
559 cfm->recv_fault |= CFM_FAULT_OVERFLOW;
561 "%s: dropped CCM with MPID %"PRIu64" from MAC "
562 ETH_ADDR_FMT, cfm->name, ccm_mpid,
563 ETH_ADDR_ARGS(eth->eth_src));
567 VLOG_DBG("%s: received CCM (seq %"PRIu32") (mpid %"PRIu64")"
568 " (interval %"PRIu8") (RDI %s)", cfm->name, ccm_seq,
569 ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false");
572 if (rmp->seq && ccm_seq != (rmp->seq + 1)) {
573 VLOG_WARN_RL(&rl, "%s: (mpid %"PRIu64") detected sequence"
574 " numbers which indicate possible connectivity"
575 " problems (previous %"PRIu32") (current %"PRIu32
576 ")", cfm->name, ccm_mpid, rmp->seq, ccm_seq);
579 rmp->mpid = ccm_mpid;
583 rmp->opup = !ccm_opdown;
588 /* Gets the fault status of 'cfm'. Returns a bit mask of 'cfm_fault_reason's
589 * indicating the cause of the connectivity fault, or zero if there is no
592 cfm_get_fault(const struct cfm *cfm)
594 if (cfm->fault_override >= 0) {
595 return cfm->fault_override ? CFM_FAULT_OVERRIDE : 0;
600 /* Gets the operational state of 'cfm'. 'cfm' is considered operationally down
601 * if it has received a CCM with the operationally down bit set from any of its
602 * remote maintenance points. Returns true if 'cfm' is operationally up. False
605 cfm_get_opup(const struct cfm *cfm)
607 return cfm->remote_opup;
610 /* Populates 'rmps' with an array of remote maintenance points reachable by
611 * 'cfm'. The number of remote maintenance points is written to 'n_rmps'.
612 * 'cfm' retains ownership of the array written to 'rmps' */
614 cfm_get_remote_mpids(const struct cfm *cfm, const uint64_t **rmps,
617 *rmps = cfm->rmps_array;
618 *n_rmps = cfm->rmps_array_len;
622 cfm_find(const char *name)
626 HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), &all_cfms) {
627 if (!strcmp(cfm->name, name)) {
635 cfm_print_details(struct ds *ds, const struct cfm *cfm)
637 struct remote_mp *rmp;
639 ds_put_format(ds, "---- %s ----\n", cfm->name);
640 ds_put_format(ds, "MPID %"PRIu64":%s%s\n", cfm->mpid,
641 cfm->extended ? " extended" : "",
642 cfm->fault_override >= 0 ? " fault_override" : "");
645 if (cfm_get_fault(cfm)) {
646 ds_put_cstr(ds, "\tfault: ");
647 ds_put_cfm_fault(ds, cfm_get_fault(cfm));
648 ds_put_cstr(ds, "\n");
651 ds_put_format(ds, "\topstate: %s\n", cfm->opup ? "up" : "down");
652 ds_put_format(ds, "\tremote_opstate: %s\n",
653 cfm->remote_opup ? "up" : "down");
654 ds_put_format(ds, "\tinterval: %dms\n", cfm->ccm_interval_ms);
655 ds_put_format(ds, "\tnext CCM tx: %lldms\n",
656 timer_msecs_until_expired(&cfm->tx_timer));
657 ds_put_format(ds, "\tnext fault check: %lldms\n",
658 timer_msecs_until_expired(&cfm->fault_timer));
660 HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
661 ds_put_format(ds, "Remote MPID %"PRIu64":%s\n",
663 rmp->rdi ? " rdi" : "");
664 ds_put_format(ds, "\trecv since check: %s\n",
665 rmp->recv ? "true" : "false");
666 ds_put_format(ds, "\topstate: %s\n", rmp->opup? "up" : "down");
671 cfm_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
672 void *aux OVS_UNUSED)
674 struct ds ds = DS_EMPTY_INITIALIZER;
675 const struct cfm *cfm;
678 cfm = cfm_find(argv[1]);
680 unixctl_command_reply_error(conn, "no such CFM object");
683 cfm_print_details(&ds, cfm);
685 HMAP_FOR_EACH (cfm, hmap_node, &all_cfms) {
686 cfm_print_details(&ds, cfm);
690 unixctl_command_reply(conn, ds_cstr(&ds));
695 cfm_unixctl_set_fault(struct unixctl_conn *conn, int argc, const char *argv[],
696 void *aux OVS_UNUSED)
698 const char *fault_str = argv[argc - 1];
702 if (!strcasecmp("true", fault_str)) {
704 } else if (!strcasecmp("false", fault_str)) {
706 } else if (!strcasecmp("normal", fault_str)) {
709 unixctl_command_reply_error(conn, "unknown fault string");
714 cfm = cfm_find(argv[1]);
716 unixctl_command_reply_error(conn, "no such CFM object");
719 cfm->fault_override = fault_override;
721 HMAP_FOR_EACH (cfm, hmap_node, &all_cfms) {
722 cfm->fault_override = fault_override;
726 unixctl_command_reply(conn, "OK");