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 "dynamic-string.h"
31 #include "poll-loop.h"
37 VLOG_DEFINE_THIS_MODULE(cfm);
39 /* Ethernet destination address of CCM packets. */
40 static const uint8_t eth_addr_ccm[6] = { 0x01, 0x80, 0xC2, 0x00, 0x00, 0x30 };
42 #define ETH_TYPE_CFM 0x8902
44 /* A 'ccm' represents a Continuity Check Message from the 802.1ag
45 * specification. Continuity Check Messages are broadcast periodically so that
46 * hosts can determine whom they have connectivity to. */
48 #define CCM_MAID_LEN 48
49 #define CCM_OPCODE 1 /* CFM message opcode meaning CCM. */
50 #define CCM_RDI_MASK 0x80
52 uint8_t mdlevel_version; /* MD Level and Version */
58 uint8_t maid[CCM_MAID_LEN];
59 uint8_t zero[16]; /* Defined by ITU-T Y.1731 should be zero */
60 } __attribute__((packed));
61 BUILD_ASSERT_DECL(CCM_LEN == sizeof(struct ccm));
64 char *name; /* Name of this CFM object. */
65 struct hmap_node hmap_node; /* Node in all_cfms list. */
68 bool fault; /* Indicates connectivity fault. */
69 bool recv_fault; /* Indicates an inability to receive CCMs. */
70 bool unexpected_recv; /* Received an unexpected CCM. */
72 uint32_t seq; /* The sequence number of our last CCM. */
73 uint8_t ccm_interval; /* The CCM transmission interval. */
74 int ccm_interval_ms; /* 'ccm_interval' in milliseconds. */
75 uint8_t maid[CCM_MAID_LEN]; /* The MAID of this CFM. */
77 struct timer tx_timer; /* Send CCM when expired. */
78 struct timer fault_timer; /* Check for faults when expired. */
80 struct hmap remote_mps; /* Expected remote MPs. */
83 /* Remote MPs represent foreign network entities that are configured to have
84 * the same MAID as this CFM instance. */
86 uint16_t mpid; /* The Maintenance Point ID of this 'remote_mp'. */
87 struct hmap_node node; /* Node in 'remote_mps' map. */
89 bool recv; /* CCM was received since last fault check. */
90 bool fault; /* Indicates a connectivity fault. */
91 bool rdi; /* Remote Defect Indicator. Indicates remote_mp isn't
92 receiving CCMs that it's expecting to. */
95 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
96 static struct hmap all_cfms = HMAP_INITIALIZER(&all_cfms);
98 static void cfm_unixctl_show(struct unixctl_conn *, const char *args,
102 cfm_generate_maid(struct cfm *cfm)
104 const char *ovs_md_name = "ovs";
105 const char *ovs_ma_name = "ovs";
107 size_t md_len, ma_len;
109 memset(cfm->maid, 0, CCM_MAID_LEN);
111 md_len = strlen(ovs_md_name);
112 ma_len = strlen(ovs_ma_name);
114 assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN);
116 cfm->maid[0] = 4; /* MD name string format. */
117 cfm->maid[1] = md_len; /* MD name size. */
118 memcpy(&cfm->maid[2], ovs_md_name, md_len); /* MD name. */
120 ma_p = cfm->maid + 2 + md_len;
121 ma_p[0] = 2; /* MA name string format. */
122 ma_p[1] = ma_len; /* MA name size. */
123 memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */
127 ccm_interval_to_ms(uint8_t interval)
130 case 0: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
131 case 1: return 3; /* Not recommended due to timer resolution. */
132 case 2: return 10; /* Not recommended due to timer resolution. */
135 case 5: return 10000;
136 case 6: return 60000;
137 case 7: return 600000;
138 default: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
145 cfm_fault_interval(struct cfm *cfm)
147 /* According to the 802.1ag specification we should assume every other MP
148 * with the same MAID has the same transmission interval that we have. If
149 * an MP has a different interval, cfm_process_heartbeat will register it
150 * as a fault (likely due to a configuration error). Thus we can check all
151 * MPs at once making this quite a bit simpler.
153 * According to the specification we should check when (ccm_interval_ms *
154 * 3.5)ms have passed. */
155 return (cfm->ccm_interval_ms * 7) / 2;
159 ms_to_ccm_interval(int interval_ms)
163 for (i = 7; i > 0; i--) {
164 if (ccm_interval_to_ms(i) <= interval_ms) {
173 hash_mpid(uint8_t mpid)
175 return hash_int(mpid, 0);
179 cfm_is_valid_mpid(uint32_t mpid)
181 /* 802.1ag specification requires MPIDs to be within the range [1, 8191] */
182 return mpid >= 1 && mpid <= 8191;
185 static struct remote_mp *
186 lookup_remote_mp(const struct hmap *hmap, uint16_t mpid)
188 struct remote_mp *rmp;
190 HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), hmap) {
191 if (rmp->mpid == mpid) {
202 unixctl_command_register("cfm/show", cfm_unixctl_show, NULL);
205 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
206 * cfm_configure() before use. */
208 cfm_create(const char *name)
212 cfm = xzalloc(sizeof *cfm);
213 cfm->name = xstrdup(name);
214 hmap_init(&cfm->remote_mps);
215 cfm_generate_maid(cfm);
216 hmap_insert(&all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0));
221 cfm_destroy(struct cfm *cfm)
223 struct remote_mp *rmp, *rmp_next;
229 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
230 hmap_remove(&cfm->remote_mps, &rmp->node);
234 hmap_destroy(&cfm->remote_mps);
235 hmap_remove(&all_cfms, &cfm->hmap_node);
240 /* Should be run periodically to update fault statistics messages. */
242 cfm_run(struct cfm *cfm)
244 if (timer_expired(&cfm->fault_timer)) {
245 long long int interval = cfm_fault_interval(cfm);
246 struct remote_mp *rmp;
248 cfm->fault = cfm->unexpected_recv;
249 cfm->recv_fault = false;
250 cfm->unexpected_recv = false;
252 HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
253 rmp->fault = !rmp->recv;
257 cfm->recv_fault = true;
258 VLOG_DBG("%s: No CCM from RMP %"PRIu16" in the last %lldms",
259 cfm->name, rmp->mpid, interval);
260 } else if (rmp->rdi) {
262 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu16, cfm->name,
267 if (cfm->recv_fault) {
270 VLOG_DBG("%s: All RMPs received CCMs in the last %lldms",
271 cfm->name, interval);
274 timer_set_duration(&cfm->fault_timer, interval);
278 /* Should be run periodically to check if the CFM module has a CCM message it
281 cfm_should_send_ccm(struct cfm *cfm)
283 return timer_expired(&cfm->tx_timer);
286 /* Composes a CCM message into 'packet'. Messages generated with this function
287 * should be sent whenever cfm_should_send_ccm() indicates. */
289 cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet,
290 uint8_t eth_src[ETH_ADDR_LEN])
294 timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms);
296 ccm = eth_compose(packet, eth_addr_ccm, eth_src, ETH_TYPE_CFM,
298 ccm->mdlevel_version = 0;
299 ccm->opcode = CCM_OPCODE;
300 ccm->tlv_offset = 70;
301 ccm->seq = htonl(++cfm->seq);
302 ccm->mpid = htons(cfm->mpid);
303 ccm->flags = cfm->ccm_interval;
304 memcpy(ccm->maid, cfm->maid, sizeof ccm->maid);
305 memset(ccm->zero, 0, sizeof ccm->zero);
307 if (cfm->recv_fault) {
308 ccm->flags |= CCM_RDI_MASK;
313 cfm_wait(struct cfm *cfm)
316 timer_wait(&cfm->tx_timer);
317 timer_wait(&cfm->fault_timer);
320 /* Configures 'cfm' with settings from 's'. */
322 cfm_configure(struct cfm *cfm, const struct cfm_settings *s)
326 struct hmap new_rmps;
327 struct remote_mp *rmp, *rmp_next;
329 if (!cfm_is_valid_mpid(s->mpid) || s->interval <= 0
330 || s->n_remote_mpids <= 0) {
335 interval = ms_to_ccm_interval(s->interval);
337 if (interval != cfm->ccm_interval) {
338 cfm->ccm_interval = interval;
339 cfm->ccm_interval_ms = ccm_interval_to_ms(interval);
341 timer_set_expired(&cfm->tx_timer);
342 timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm));
345 hmap_init(&new_rmps);
346 for (i = 0; i < s->n_remote_mpids; i++) {
347 uint16_t mpid = s->remote_mpids[i];
349 if (!cfm_is_valid_mpid(mpid)
350 || lookup_remote_mp(&new_rmps, mpid)) {
354 if ((rmp = lookup_remote_mp(&cfm->remote_mps, mpid))) {
355 hmap_remove(&cfm->remote_mps, &rmp->node);
357 rmp = xzalloc(sizeof *rmp);
361 hmap_insert(&new_rmps, &rmp->node, hash_mpid(mpid));
364 hmap_swap(&new_rmps, &cfm->remote_mps);
365 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &new_rmps) {
366 hmap_remove(&new_rmps, &rmp->node);
369 hmap_destroy(&new_rmps);
374 /* Returns true if the CFM library should process packets from 'flow'. */
376 cfm_should_process_flow(const struct flow *flow)
378 return (ntohs(flow->dl_type) == ETH_TYPE_CFM
379 && eth_addr_equals(flow->dl_dst, eth_addr_ccm));
382 /* Updates internal statistics relevant to packet 'p'. Should be called on
383 * every packet whose flow returned true when passed to
384 * cfm_should_process_flow. */
386 cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
391 uint8_t ccm_interval;
392 struct remote_mp *rmp;
393 struct eth_header *eth;
396 ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_LEN);
399 VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.",
404 if (ccm->opcode != CCM_OPCODE) {
405 VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. "
406 "(opcode %u)", cfm->name, ccm->opcode);
410 /* According to the 802.1ag specification, reception of a CCM with an
411 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
412 * trigger a fault. We ignore this requirement for several reasons.
414 * Faults can cause a controller or Open vSwitch to make potentially
415 * expensive changes to the network topology. It seems prudent to trigger
416 * them judiciously, especially when CFM is used to check slave status of
417 * bonds. Furthermore, faults can be maliciously triggered by crafting
419 if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) {
420 cfm->unexpected_recv = true;
421 VLOG_WARN_RL(&rl, "%s: Received unexpected remote MAID from MAC "
422 ETH_ADDR_FMT, cfm->name, ETH_ADDR_ARGS(eth->eth_src));
424 ccm_mpid = ntohs(ccm->mpid);
425 ccm_interval = ccm->flags & 0x7;
426 ccm_rdi = ccm->flags & CCM_RDI_MASK;
428 rmp = lookup_remote_mp(&cfm->remote_mps, ccm_mpid);
434 if (ccm_interval != cfm->ccm_interval) {
435 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid interval"
436 " (%"PRIu8") from RMP %"PRIu16, cfm->name,
437 ccm_interval, rmp->mpid);
440 cfm->unexpected_recv = true;
441 VLOG_WARN_RL(&rl, "%s: Received unexpected remote MPID %d from"
442 " MAC " ETH_ADDR_FMT, cfm->name, ccm_mpid,
443 ETH_ADDR_ARGS(eth->eth_src));
446 VLOG_DBG("%s: Received CCM (seq %"PRIu32") (mpid %"PRIu16")"
447 " (interval %"PRIu8") (RDI %s)", cfm->name, ntohl(ccm->seq),
448 ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false");
452 /* Gets the fault status of 'cfm'. Returns true when 'cfm' has detected
453 * connectivity problems, false otherwise. */
455 cfm_get_fault(const struct cfm *cfm)
461 cfm_find(const char *name)
465 HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), &all_cfms) {
466 if (!strcmp(cfm->name, name)) {
474 cfm_unixctl_show(struct unixctl_conn *conn,
475 const char *args, void *aux OVS_UNUSED)
477 struct ds ds = DS_EMPTY_INITIALIZER;
478 const struct cfm *cfm;
479 struct remote_mp *rmp;
481 cfm = cfm_find(args);
483 unixctl_command_reply(conn, 501, "no such CFM object");
487 ds_put_format(&ds, "MPID %"PRIu16":%s%s%s\n", cfm->mpid,
488 cfm->fault ? " fault" : "",
489 cfm->unexpected_recv ? " unexpected_recv" : "",
490 cfm->recv_fault ? " recv_fault" : "");
492 ds_put_format(&ds, "\tinterval: %dms\n", cfm->ccm_interval_ms);
493 ds_put_format(&ds, "\tnext CCM tx: %lldms\n",
494 timer_msecs_until_expired(&cfm->tx_timer));
495 ds_put_format(&ds, "\tnext fault check: %lldms\n",
496 timer_msecs_until_expired(&cfm->fault_timer));
498 ds_put_cstr(&ds, "\n");
499 HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
500 ds_put_format(&ds, "Remote MPID %"PRIu16": %s\n", rmp->mpid,
501 rmp->fault ? "fault" : "");
502 ds_put_format(&ds, "\trecv since check: %s",
503 rmp->recv ? "true" : "false");
506 unixctl_command_reply(conn, 200, ds_cstr(&ds));