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 maid[CCM_MAID_LEN]; /* The MAID of this CFM. */
102 struct timer tx_timer; /* Send CCM when expired. */
103 struct timer fault_timer; /* Check for faults when expired. */
105 struct hmap remote_mps; /* Remote MPs. */
107 /* Result of cfm_get_remote_mpids(). Updated only during fault check to
109 uint64_t *rmps_array; /* Cache of remote_mps. */
110 size_t rmps_array_len; /* Number of rmps in 'rmps_array'. */
113 /* Remote MPs represent foreign network entities that are configured to have
114 * the same MAID as this CFM instance. */
116 uint64_t mpid; /* The Maintenance Point ID of this 'remote_mp'. */
117 struct hmap_node node; /* Node in 'remote_mps' map. */
119 bool recv; /* CCM was received since last fault check. */
120 bool rdi; /* Remote Defect Indicator. Indicates remote_mp isn't
121 receiving CCMs that it's expecting to. */
122 bool opup; /* Operational State. */
125 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
126 static struct hmap all_cfms = HMAP_INITIALIZER(&all_cfms);
128 static unixctl_cb_func cfm_unixctl_show;
129 static unixctl_cb_func cfm_unixctl_set_fault;
131 static const uint8_t *
132 cfm_ccm_addr(const struct cfm *cfm)
134 return cfm->extended ? eth_addr_ccm_x : eth_addr_ccm;
138 cfm_generate_maid(struct cfm *cfm)
140 const char *ovs_md_name = "ovs";
141 const char *ovs_ma_name = "ovs";
143 size_t md_len, ma_len;
145 memset(cfm->maid, 0, CCM_MAID_LEN);
147 md_len = strlen(ovs_md_name);
148 ma_len = strlen(ovs_ma_name);
150 assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN);
152 cfm->maid[0] = 4; /* MD name string format. */
153 cfm->maid[1] = md_len; /* MD name size. */
154 memcpy(&cfm->maid[2], ovs_md_name, md_len); /* MD name. */
156 ma_p = cfm->maid + 2 + md_len;
157 ma_p[0] = 2; /* MA name string format. */
158 ma_p[1] = ma_len; /* MA name size. */
159 memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */
163 ccm_interval_to_ms(uint8_t interval)
166 case 0: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
167 case 1: return 3; /* Not recommended due to timer resolution. */
168 case 2: return 10; /* Not recommended due to timer resolution. */
171 case 5: return 10000;
172 case 6: return 60000;
173 case 7: return 600000;
174 default: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
181 cfm_fault_interval(struct cfm *cfm)
183 /* According to the 802.1ag specification we should assume every other MP
184 * with the same MAID has the same transmission interval that we have. If
185 * an MP has a different interval, cfm_process_heartbeat will register it
186 * as a fault (likely due to a configuration error). Thus we can check all
187 * MPs at once making this quite a bit simpler.
189 * According to the specification we should check when (ccm_interval_ms *
190 * 3.5)ms have passed. */
191 return (cfm->ccm_interval_ms * 7) / 2;
195 ms_to_ccm_interval(int interval_ms)
199 for (i = 7; i > 0; i--) {
200 if (ccm_interval_to_ms(i) <= interval_ms) {
209 hash_mpid(uint64_t mpid)
211 return hash_bytes(&mpid, sizeof mpid, 0);
215 cfm_is_valid_mpid(bool extended, uint64_t mpid)
217 /* 802.1ag specification requires MPIDs to be within the range [1, 8191].
218 * In extended mode we relax this requirement. */
219 return mpid >= 1 && (extended || mpid <= 8191);
222 static struct remote_mp *
223 lookup_remote_mp(const struct cfm *cfm, uint64_t mpid)
225 struct remote_mp *rmp;
227 HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), &cfm->remote_mps) {
228 if (rmp->mpid == mpid) {
239 unixctl_command_register("cfm/show", "[interface]", 0, 1, cfm_unixctl_show,
241 unixctl_command_register("cfm/set-fault", "[interface] normal|false|true",
242 1, 2, cfm_unixctl_set_fault, NULL);
245 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
246 * cfm_configure() before use. */
248 cfm_create(const char *name)
252 cfm = xzalloc(sizeof *cfm);
253 cfm->name = xstrdup(name);
254 hmap_init(&cfm->remote_mps);
255 cfm_generate_maid(cfm);
256 hmap_insert(&all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0));
257 cfm->remote_opup = true;
258 cfm->fault_override = -1;
263 cfm_destroy(struct cfm *cfm)
265 struct remote_mp *rmp, *rmp_next;
271 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
272 hmap_remove(&cfm->remote_mps, &rmp->node);
276 hmap_destroy(&cfm->remote_mps);
277 hmap_remove(&all_cfms, &cfm->hmap_node);
278 free(cfm->rmps_array);
283 /* Should be run periodically to update fault statistics messages. */
285 cfm_run(struct cfm *cfm)
287 if (timer_expired(&cfm->fault_timer)) {
288 long long int interval = cfm_fault_interval(cfm);
289 struct remote_mp *rmp, *rmp_next;
291 cfm->fault = cfm->unexpected_recv;
292 cfm->unexpected_recv = false;
294 cfm->rmps_array_len = 0;
295 free(cfm->rmps_array);
296 cfm->rmps_array = xmalloc(hmap_count(&cfm->remote_mps) *
297 sizeof *cfm->rmps_array);
299 cfm->remote_opup = true;
300 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
303 VLOG_DBG("%s: no CCM from RMP %"PRIu64" in the last %lldms",
304 cfm->name, rmp->mpid, interval);
305 hmap_remove(&cfm->remote_mps, &rmp->node);
310 if (rmp->mpid == cfm->mpid) {
311 VLOG_WARN_RL(&rl,"%s: received CCM with local MPID"
312 " %"PRIu64, cfm->name, rmp->mpid);
317 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64, cfm->name,
323 cfm->remote_opup = rmp->opup;
326 cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid;
330 if (hmap_is_empty(&cfm->remote_mps)) {
334 timer_set_duration(&cfm->fault_timer, interval);
338 /* Should be run periodically to check if the CFM module has a CCM message it
341 cfm_should_send_ccm(struct cfm *cfm)
343 return timer_expired(&cfm->tx_timer);
346 /* Composes a CCM message into 'packet'. Messages generated with this function
347 * should be sent whenever cfm_should_send_ccm() indicates. */
349 cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet,
350 uint8_t eth_src[ETH_ADDR_LEN])
354 timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms);
355 eth_compose(packet, cfm_ccm_addr(cfm), eth_src, ETH_TYPE_CFM, sizeof *ccm);
358 eth_push_vlan(packet, htons(cfm->ccm_vlan));
362 ccm->mdlevel_version = 0;
363 ccm->opcode = CCM_OPCODE;
364 ccm->tlv_offset = 70;
365 ccm->seq = htonl(++cfm->seq);
366 ccm->flags = cfm->ccm_interval;
367 memcpy(ccm->maid, cfm->maid, sizeof ccm->maid);
368 memset(ccm->zero, 0, sizeof ccm->zero);
372 ccm->mpid = htons(hash_mpid(cfm->mpid));
373 ccm->mpid64 = htonll(cfm->mpid);
374 ccm->opdown = !cfm->opup;
376 ccm->mpid = htons(cfm->mpid);
377 ccm->mpid64 = htonll(0);
381 if (cfm->ccm_interval == 0) {
382 assert(cfm->extended);
383 ccm->interval_ms_x = htons(cfm->ccm_interval_ms);
386 if (hmap_is_empty(&cfm->remote_mps)) {
387 ccm->flags |= CCM_RDI_MASK;
392 cfm_wait(struct cfm *cfm)
394 timer_wait(&cfm->tx_timer);
395 timer_wait(&cfm->fault_timer);
398 /* Configures 'cfm' with settings from 's'. */
400 cfm_configure(struct cfm *cfm, const struct cfm_settings *s)
405 if (!cfm_is_valid_mpid(s->extended, s->mpid) || s->interval <= 0) {
410 cfm->extended = s->extended;
412 interval = ms_to_ccm_interval(s->interval);
413 interval_ms = ccm_interval_to_ms(interval);
415 cfm->ccm_vlan = s->ccm_vlan & VLAN_VID_MASK;
416 if (cfm->extended && interval_ms != s->interval) {
418 interval_ms = MIN(s->interval, UINT16_MAX);
421 if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) {
422 cfm->ccm_interval = interval;
423 cfm->ccm_interval_ms = interval_ms;
425 timer_set_expired(&cfm->tx_timer);
426 timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm));
432 /* Returns true if 'cfm' should process packets from 'flow'. */
434 cfm_should_process_flow(const struct cfm *cfm, const struct flow *flow)
436 return (ntohs(flow->dl_type) == ETH_TYPE_CFM
437 && eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm)));
440 /* Updates internal statistics relevant to packet 'p'. Should be called on
441 * every packet whose flow returned true when passed to
442 * cfm_should_process_flow. */
444 cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
447 struct eth_header *eth;
450 ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_ACCEPT_LEN);
453 VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.",
458 if (ccm->opcode != CCM_OPCODE) {
459 VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. "
460 "(opcode %u)", cfm->name, ccm->opcode);
464 /* According to the 802.1ag specification, reception of a CCM with an
465 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
466 * trigger a fault. We ignore this requirement for several reasons.
468 * Faults can cause a controller or Open vSwitch to make potentially
469 * expensive changes to the network topology. It seems prudent to trigger
470 * them judiciously, especially when CFM is used to check slave status of
471 * bonds. Furthermore, faults can be maliciously triggered by crafting
473 if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) {
474 cfm->unexpected_recv = true;
475 VLOG_WARN_RL(&rl, "%s: Received unexpected remote MAID from MAC "
476 ETH_ADDR_FMT, cfm->name, ETH_ADDR_ARGS(eth->eth_src));
478 uint8_t ccm_interval = ccm->flags & 0x7;
479 bool ccm_rdi = ccm->flags & CCM_RDI_MASK;
480 uint16_t ccm_interval_ms_x = ntohs(ccm->interval_ms_x);
482 struct remote_mp *rmp;
487 ccm_mpid = ntohll(ccm->mpid64);
488 ccm_opdown = ccm->opdown;
490 ccm_mpid = ntohs(ccm->mpid);
494 if (ccm_interval != cfm->ccm_interval) {
495 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid interval"
496 " (%"PRIu8") from RMP %"PRIu64, cfm->name,
497 ccm_interval, ccm_mpid);
500 if (cfm->extended && ccm_interval == 0
501 && ccm_interval_ms_x != cfm->ccm_interval_ms) {
502 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid extended"
503 " interval (%"PRIu16"ms) from RMP %"PRIu64, cfm->name,
504 ccm_interval_ms_x, ccm_mpid);
507 rmp = lookup_remote_mp(cfm, ccm_mpid);
509 if (hmap_count(&cfm->remote_mps) < CFM_MAX_RMPS) {
510 rmp = xmalloc(sizeof *rmp);
511 hmap_insert(&cfm->remote_mps, &rmp->node, hash_mpid(ccm_mpid));
513 cfm->unexpected_recv = true;
515 "%s: dropped CCM with MPID %"PRIu64" from MAC "
516 ETH_ADDR_FMT, cfm->name, ccm_mpid,
517 ETH_ADDR_ARGS(eth->eth_src));
522 rmp->mpid = ccm_mpid;
525 rmp->opup = !ccm_opdown;
528 VLOG_DBG("%s: received CCM (seq %"PRIu32") (mpid %"PRIu64")"
529 " (interval %"PRIu8") (RDI %s)", cfm->name, ntohl(ccm->seq),
530 ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false");
534 /* Gets the fault status of 'cfm'. Returns true when 'cfm' has detected
535 * connectivity problems, false otherwise. */
537 cfm_get_fault(const struct cfm *cfm)
539 if (cfm->fault_override >= 0) {
540 return cfm->fault_override;
545 /* Gets the operational state of 'cfm'. 'cfm' is considered operationally down
546 * if it has received a CCM with the operationally down bit set from any of its
547 * remote maintenance points. Returns true if 'cfm' is operationally up. False
550 cfm_get_opup(const struct cfm *cfm)
552 return cfm->remote_opup;
555 /* Populates 'rmps' with an array of remote maintenance points reachable by
556 * 'cfm'. The number of remote maintenance points is written to 'n_rmps'.
557 * 'cfm' retains ownership of the array written to 'rmps' */
559 cfm_get_remote_mpids(const struct cfm *cfm, const uint64_t **rmps,
562 *rmps = cfm->rmps_array;
563 *n_rmps = cfm->rmps_array_len;
567 cfm_find(const char *name)
571 HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), &all_cfms) {
572 if (!strcmp(cfm->name, name)) {
580 cfm_print_details(struct ds *ds, const struct cfm *cfm)
582 struct remote_mp *rmp;
584 ds_put_format(ds, "---- %s ----\n", cfm->name);
585 ds_put_format(ds, "MPID %"PRIu64":%s%s%s%s\n", cfm->mpid,
586 cfm->extended ? " extended" : "",
587 cfm_get_fault(cfm) ? " fault" : "",
588 cfm->fault_override >= 0 ? " fault_override" : "",
589 cfm->unexpected_recv ? " unexpected_recv" : "");
591 ds_put_format(ds, "\topstate: %s\n", cfm->opup ? "up" : "down");
592 ds_put_format(ds, "\tremote_opstate: %s\n",
593 cfm->remote_opup ? "up" : "down");
594 ds_put_format(ds, "\tinterval: %dms\n", cfm->ccm_interval_ms);
595 ds_put_format(ds, "\tnext CCM tx: %lldms\n",
596 timer_msecs_until_expired(&cfm->tx_timer));
597 ds_put_format(ds, "\tnext fault check: %lldms\n",
598 timer_msecs_until_expired(&cfm->fault_timer));
600 HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
601 ds_put_format(ds, "Remote MPID %"PRIu64":%s\n",
603 rmp->rdi ? " rdi" : "");
604 ds_put_format(ds, "\trecv since check: %s\n",
605 rmp->recv ? "true" : "false");
606 ds_put_format(ds, "\topstate: %s\n", rmp->opup? "up" : "down");
611 cfm_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
612 void *aux OVS_UNUSED)
614 struct ds ds = DS_EMPTY_INITIALIZER;
615 const struct cfm *cfm;
618 cfm = cfm_find(argv[1]);
620 unixctl_command_reply(conn, 501, "no such CFM object");
623 cfm_print_details(&ds, cfm);
625 HMAP_FOR_EACH (cfm, hmap_node, &all_cfms) {
626 cfm_print_details(&ds, cfm);
630 unixctl_command_reply(conn, 200, ds_cstr(&ds));
635 cfm_unixctl_set_fault(struct unixctl_conn *conn, int argc, const char *argv[],
636 void *aux OVS_UNUSED)
638 const char *fault_str = argv[argc - 1];
642 if (!strcasecmp("true", fault_str)) {
644 } else if (!strcasecmp("false", fault_str)) {
646 } else if (!strcasecmp("normal", fault_str)) {
649 unixctl_command_reply(conn, 501, "unknown fault string");
654 cfm = cfm_find(argv[1]);
656 unixctl_command_reply(conn, 501, "no such CFM object");
659 cfm->fault_override = fault_override;
661 HMAP_FOR_EACH (cfm, hmap_node, &all_cfms) {
662 cfm->fault_override = fault_override;
666 unixctl_command_reply(conn, 200, "OK");