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 uint32_t seq; /* The sequence number of our last CCM. */
94 uint8_t ccm_interval; /* The CCM transmission interval. */
95 int ccm_interval_ms; /* 'ccm_interval' in milliseconds. */
96 uint16_t ccm_vlan; /* Vlan tag of CCM PDUs. */
97 uint8_t maid[CCM_MAID_LEN]; /* The MAID of this CFM. */
99 struct timer tx_timer; /* Send CCM when expired. */
100 struct timer fault_timer; /* Check for faults when expired. */
102 struct hmap remote_mps; /* Remote MPs. */
104 /* Result of cfm_get_remote_mpids(). Updated only during fault check to
106 uint64_t *rmps_array; /* Cache of remote_mps. */
107 size_t rmps_array_len; /* Number of rmps in 'rmps_array'. */
110 /* Remote MPs represent foreign network entities that are configured to have
111 * the same MAID as this CFM instance. */
113 uint64_t mpid; /* The Maintenance Point ID of this 'remote_mp'. */
114 struct hmap_node node; /* Node in 'remote_mps' map. */
116 bool recv; /* CCM was received since last fault check. */
117 bool rdi; /* Remote Defect Indicator. Indicates remote_mp isn't
118 receiving CCMs that it's expecting to. */
119 bool opup; /* Operational State. */
122 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
123 static struct hmap all_cfms = HMAP_INITIALIZER(&all_cfms);
125 static unixctl_cb_func cfm_unixctl_show;
127 static const uint8_t *
128 cfm_ccm_addr(const struct cfm *cfm)
130 return cfm->extended ? eth_addr_ccm_x : eth_addr_ccm;
134 cfm_generate_maid(struct cfm *cfm)
136 const char *ovs_md_name = "ovs";
137 const char *ovs_ma_name = "ovs";
139 size_t md_len, ma_len;
141 memset(cfm->maid, 0, CCM_MAID_LEN);
143 md_len = strlen(ovs_md_name);
144 ma_len = strlen(ovs_ma_name);
146 assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN);
148 cfm->maid[0] = 4; /* MD name string format. */
149 cfm->maid[1] = md_len; /* MD name size. */
150 memcpy(&cfm->maid[2], ovs_md_name, md_len); /* MD name. */
152 ma_p = cfm->maid + 2 + md_len;
153 ma_p[0] = 2; /* MA name string format. */
154 ma_p[1] = ma_len; /* MA name size. */
155 memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */
159 ccm_interval_to_ms(uint8_t interval)
162 case 0: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
163 case 1: return 3; /* Not recommended due to timer resolution. */
164 case 2: return 10; /* Not recommended due to timer resolution. */
167 case 5: return 10000;
168 case 6: return 60000;
169 case 7: return 600000;
170 default: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
177 cfm_fault_interval(struct cfm *cfm)
179 /* According to the 802.1ag specification we should assume every other MP
180 * with the same MAID has the same transmission interval that we have. If
181 * an MP has a different interval, cfm_process_heartbeat will register it
182 * as a fault (likely due to a configuration error). Thus we can check all
183 * MPs at once making this quite a bit simpler.
185 * According to the specification we should check when (ccm_interval_ms *
186 * 3.5)ms have passed. */
187 return (cfm->ccm_interval_ms * 7) / 2;
191 ms_to_ccm_interval(int interval_ms)
195 for (i = 7; i > 0; i--) {
196 if (ccm_interval_to_ms(i) <= interval_ms) {
205 hash_mpid(uint64_t mpid)
207 return hash_bytes(&mpid, sizeof mpid, 0);
211 cfm_is_valid_mpid(bool extended, uint64_t mpid)
213 /* 802.1ag specification requires MPIDs to be within the range [1, 8191].
214 * In extended mode we relax this requirement. */
215 return mpid >= 1 && (extended || mpid <= 8191);
218 static struct remote_mp *
219 lookup_remote_mp(const struct cfm *cfm, uint64_t mpid)
221 struct remote_mp *rmp;
223 HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), &cfm->remote_mps) {
224 if (rmp->mpid == mpid) {
235 unixctl_command_register("cfm/show", "[interface]", 0, 1, cfm_unixctl_show,
239 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
240 * cfm_configure() before use. */
242 cfm_create(const char *name)
246 cfm = xzalloc(sizeof *cfm);
247 cfm->name = xstrdup(name);
248 hmap_init(&cfm->remote_mps);
249 cfm_generate_maid(cfm);
250 hmap_insert(&all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0));
251 cfm->remote_opup = true;
256 cfm_destroy(struct cfm *cfm)
258 struct remote_mp *rmp, *rmp_next;
264 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
265 hmap_remove(&cfm->remote_mps, &rmp->node);
269 hmap_destroy(&cfm->remote_mps);
270 hmap_remove(&all_cfms, &cfm->hmap_node);
271 free(cfm->rmps_array);
276 /* Should be run periodically to update fault statistics messages. */
278 cfm_run(struct cfm *cfm)
280 if (timer_expired(&cfm->fault_timer)) {
281 long long int interval = cfm_fault_interval(cfm);
282 struct remote_mp *rmp, *rmp_next;
284 cfm->fault = cfm->unexpected_recv;
285 cfm->unexpected_recv = false;
287 cfm->rmps_array_len = 0;
288 free(cfm->rmps_array);
289 cfm->rmps_array = xmalloc(hmap_count(&cfm->remote_mps) *
290 sizeof *cfm->rmps_array);
292 cfm->remote_opup = true;
293 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
296 VLOG_DBG("%s: no CCM from RMP %"PRIu64" in the last %lldms",
297 cfm->name, rmp->mpid, interval);
298 hmap_remove(&cfm->remote_mps, &rmp->node);
303 if (rmp->mpid == cfm->mpid) {
304 VLOG_WARN_RL(&rl,"%s: received CCM with local MPID"
305 " %"PRIu64, cfm->name, rmp->mpid);
310 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64, cfm->name,
316 cfm->remote_opup = rmp->opup;
319 cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid;
323 if (hmap_is_empty(&cfm->remote_mps)) {
327 timer_set_duration(&cfm->fault_timer, interval);
331 /* Should be run periodically to check if the CFM module has a CCM message it
334 cfm_should_send_ccm(struct cfm *cfm)
336 return timer_expired(&cfm->tx_timer);
339 /* Composes a CCM message into 'packet'. Messages generated with this function
340 * should be sent whenever cfm_should_send_ccm() indicates. */
342 cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet,
343 uint8_t eth_src[ETH_ADDR_LEN])
347 timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms);
348 eth_compose(packet, cfm_ccm_addr(cfm), eth_src, ETH_TYPE_CFM, sizeof *ccm);
351 eth_push_vlan(packet, htons(cfm->ccm_vlan));
355 ccm->mdlevel_version = 0;
356 ccm->opcode = CCM_OPCODE;
357 ccm->tlv_offset = 70;
358 ccm->seq = htonl(++cfm->seq);
359 ccm->flags = cfm->ccm_interval;
360 memcpy(ccm->maid, cfm->maid, sizeof ccm->maid);
361 memset(ccm->zero, 0, sizeof ccm->zero);
365 ccm->mpid = htons(hash_mpid(cfm->mpid));
366 ccm->mpid64 = htonll(cfm->mpid);
367 ccm->opdown = !cfm->opup;
369 ccm->mpid = htons(cfm->mpid);
370 ccm->mpid64 = htonll(0);
374 if (cfm->ccm_interval == 0) {
375 assert(cfm->extended);
376 ccm->interval_ms_x = htons(cfm->ccm_interval_ms);
379 if (hmap_is_empty(&cfm->remote_mps)) {
380 ccm->flags |= CCM_RDI_MASK;
385 cfm_wait(struct cfm *cfm)
387 timer_wait(&cfm->tx_timer);
388 timer_wait(&cfm->fault_timer);
391 /* Configures 'cfm' with settings from 's'. */
393 cfm_configure(struct cfm *cfm, const struct cfm_settings *s)
398 if (!cfm_is_valid_mpid(s->extended, s->mpid) || s->interval <= 0) {
403 cfm->extended = s->extended;
405 interval = ms_to_ccm_interval(s->interval);
406 interval_ms = ccm_interval_to_ms(interval);
408 cfm->ccm_vlan = s->ccm_vlan & VLAN_VID_MASK;
409 if (cfm->extended && interval_ms != s->interval) {
411 interval_ms = MIN(s->interval, UINT16_MAX);
414 if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) {
415 cfm->ccm_interval = interval;
416 cfm->ccm_interval_ms = interval_ms;
418 timer_set_expired(&cfm->tx_timer);
419 timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm));
425 /* Returns true if 'cfm' should process packets from 'flow'. */
427 cfm_should_process_flow(const struct cfm *cfm, const struct flow *flow)
429 return (ntohs(flow->dl_type) == ETH_TYPE_CFM
430 && eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm)));
433 /* Updates internal statistics relevant to packet 'p'. Should be called on
434 * every packet whose flow returned true when passed to
435 * cfm_should_process_flow. */
437 cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
440 struct eth_header *eth;
443 ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_ACCEPT_LEN);
446 VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.",
451 if (ccm->opcode != CCM_OPCODE) {
452 VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. "
453 "(opcode %u)", cfm->name, ccm->opcode);
457 /* According to the 802.1ag specification, reception of a CCM with an
458 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
459 * trigger a fault. We ignore this requirement for several reasons.
461 * Faults can cause a controller or Open vSwitch to make potentially
462 * expensive changes to the network topology. It seems prudent to trigger
463 * them judiciously, especially when CFM is used to check slave status of
464 * bonds. Furthermore, faults can be maliciously triggered by crafting
466 if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) {
467 cfm->unexpected_recv = true;
468 VLOG_WARN_RL(&rl, "%s: Received unexpected remote MAID from MAC "
469 ETH_ADDR_FMT, cfm->name, ETH_ADDR_ARGS(eth->eth_src));
471 uint8_t ccm_interval = ccm->flags & 0x7;
472 bool ccm_rdi = ccm->flags & CCM_RDI_MASK;
473 uint16_t ccm_interval_ms_x = ntohs(ccm->interval_ms_x);
475 struct remote_mp *rmp;
480 ccm_mpid = ntohll(ccm->mpid64);
481 ccm_opdown = ccm->opdown;
483 ccm_mpid = ntohs(ccm->mpid);
487 if (ccm_interval != cfm->ccm_interval) {
488 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid interval"
489 " (%"PRIu8") from RMP %"PRIu64, cfm->name,
490 ccm_interval, ccm_mpid);
493 if (cfm->extended && ccm_interval == 0
494 && ccm_interval_ms_x != cfm->ccm_interval_ms) {
495 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid extended"
496 " interval (%"PRIu16"ms) from RMP %"PRIu64, cfm->name,
497 ccm_interval_ms_x, ccm_mpid);
500 rmp = lookup_remote_mp(cfm, ccm_mpid);
502 if (hmap_count(&cfm->remote_mps) < CFM_MAX_RMPS) {
503 rmp = xmalloc(sizeof *rmp);
504 hmap_insert(&cfm->remote_mps, &rmp->node, hash_mpid(ccm_mpid));
506 cfm->unexpected_recv = true;
508 "%s: dropped CCM with MPID %"PRIu64" from MAC "
509 ETH_ADDR_FMT, cfm->name, ccm_mpid,
510 ETH_ADDR_ARGS(eth->eth_src));
515 rmp->mpid = ccm_mpid;
518 rmp->opup = !ccm_opdown;
521 VLOG_DBG("%s: received CCM (seq %"PRIu32") (mpid %"PRIu64")"
522 " (interval %"PRIu8") (RDI %s)", cfm->name, ntohl(ccm->seq),
523 ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false");
527 /* Gets the fault status of 'cfm'. Returns true when 'cfm' has detected
528 * connectivity problems, false otherwise. */
530 cfm_get_fault(const struct cfm *cfm)
535 /* Gets the operational state of 'cfm'. 'cfm' is considered operationally down
536 * if it has received a CCM with the operationally down bit set from any of its
537 * remote maintenance points. Returns true if 'cfm' is operationally up. False
540 cfm_get_opup(const struct cfm *cfm)
542 return cfm->remote_opup;
545 /* Populates 'rmps' with an array of remote maintenance points reachable by
546 * 'cfm'. The number of remote maintenance points is written to 'n_rmps'.
547 * 'cfm' retains ownership of the array written to 'rmps' */
549 cfm_get_remote_mpids(const struct cfm *cfm, const uint64_t **rmps,
552 *rmps = cfm->rmps_array;
553 *n_rmps = cfm->rmps_array_len;
557 cfm_find(const char *name)
561 HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), &all_cfms) {
562 if (!strcmp(cfm->name, name)) {
570 cfm_print_details(struct ds *ds, const struct cfm *cfm)
572 struct remote_mp *rmp;
574 ds_put_format(ds, "---- %s ----\n", cfm->name);
575 ds_put_format(ds, "MPID %"PRIu64":%s%s%s\n", cfm->mpid,
576 cfm->extended ? " extended" : "",
577 cfm->fault ? " fault" : "",
578 cfm->unexpected_recv ? " unexpected_recv" : "");
580 ds_put_format(ds, "\topstate: %s\n", cfm->opup ? "up" : "down");
581 ds_put_format(ds, "\tremote_opstate: %s\n",
582 cfm->remote_opup ? "up" : "down");
583 ds_put_format(ds, "\tinterval: %dms\n", cfm->ccm_interval_ms);
584 ds_put_format(ds, "\tnext CCM tx: %lldms\n",
585 timer_msecs_until_expired(&cfm->tx_timer));
586 ds_put_format(ds, "\tnext fault check: %lldms\n",
587 timer_msecs_until_expired(&cfm->fault_timer));
589 HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
590 ds_put_format(ds, "Remote MPID %"PRIu64":%s\n",
592 rmp->rdi ? " rdi" : "");
593 ds_put_format(ds, "\trecv since check: %s\n",
594 rmp->recv ? "true" : "false");
595 ds_put_format(ds, "\topstate: %s\n", rmp->opup? "up" : "down");
600 cfm_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
601 void *aux OVS_UNUSED)
603 struct ds ds = DS_EMPTY_INITIALIZER;
604 const struct cfm *cfm;
607 cfm = cfm_find(argv[1]);
609 unixctl_command_reply(conn, 501, "no such CFM object");
612 cfm_print_details(&ds, cfm);
614 HMAP_FOR_EACH (cfm, hmap_node, &all_cfms) {
615 cfm_print_details(&ds, cfm);
619 unixctl_command_reply(conn, 200, ds_cstr(&ds));