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 #define CCM_OPCODE 1 /* CFM message opcode meaning CCM. */
43 struct list list_node; /* Node in all_cfms list. */
45 uint32_t seq; /* The sequence number of our last CCM. */
46 uint8_t ccm_interval; /* The CCM transmission interval. */
47 int ccm_interval_ms; /* 'ccm_interval' in milliseconds. */
48 uint8_t maid[CCM_MAID_LEN]; /* The MAID of this CFM. */
50 struct timer tx_timer; /* Send CCM when expired. */
51 struct timer fault_timer; /* Check for faults when expired. */
53 struct hmap remote_mps; /* Expected remote MPs. */
56 /* Remote MPs represent foreign network entities that are configured to have
57 * the same MAID as this CFM instance. */
59 uint16_t mpid; /* The Maintenance Point ID of this 'remote_mp'. */
60 struct hmap_node node; /* Node in 'remote_mps' map. */
62 bool recv; /* CCM was received since last fault check. */
63 bool fault; /* Indicates a connectivity fault. */
66 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
67 static struct list all_cfms = LIST_INITIALIZER(&all_cfms);
69 static void cfm_unixctl_show(struct unixctl_conn *, const char *args,
73 cfm_generate_maid(struct cfm_internal *cfmi)
75 const char *ovs_md_name = "ovs_md";
76 const char *ovs_ma_name = "ovs_ma";
78 size_t md_len, ma_len;
80 memset(cfmi->maid, 0, CCM_MAID_LEN);
82 md_len = strlen(ovs_md_name);
83 ma_len = strlen(ovs_ma_name);
85 assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN);
87 cfmi->maid[0] = 4; /* MD name string format. */
88 cfmi->maid[1] = md_len; /* MD name size. */
89 memcpy(&cfmi->maid[2], ovs_md_name, md_len); /* MD name. */
91 ma_p = cfmi->maid + 2 + md_len;
92 ma_p[0] = 2; /* MA name string format. */
93 ma_p[1] = ma_len; /* MA name size. */
94 memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */
98 ccm_interval_to_ms(uint8_t interval)
101 case 0: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
102 case 1: return 3; /* Not recommended due to timer resolution. */
103 case 2: return 10; /* Not recommended due to timer resolution. */
106 case 5: return 10000;
107 case 6: return 60000;
108 case 7: return 600000;
109 default: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
116 cfm_fault_interval(struct cfm_internal *cfmi)
118 /* According to the 802.1ag specification we should assume every other MP
119 * with the same MAID has the same transmission interval that we have. If
120 * an MP has a different interval, cfm_process_heartbeat will register it
121 * as a fault (likely due to a configuration error). Thus we can check all
122 * MPs at once making this quite a bit simpler.
124 * According to the specification we should check when (ccm_interval_ms *
125 * 3.5)ms have passed. */
126 return (cfmi->ccm_interval_ms * 7) / 2;
130 ms_to_ccm_interval(int interval_ms)
134 for (i = 7; i > 0; i--) {
135 if (ccm_interval_to_ms(i) <= interval_ms) {
143 static struct cfm_internal *
144 cfm_to_internal(const struct cfm *cfm)
146 return CONTAINER_OF(cfm, struct cfm_internal, cfm);
150 hash_mpid(uint8_t mpid)
152 return hash_int(mpid, 0);
156 cfm_is_valid_mpid(uint32_t mpid)
158 /* 802.1ag specification requires MPIDs to be within the range [1, 8191] */
159 return mpid >= 1 && mpid <= 8191;
162 static struct remote_mp *
163 lookup_remote_mp(const struct hmap *hmap, uint16_t mpid)
165 struct remote_mp *rmp;
167 HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), hmap) {
168 if (rmp->mpid == mpid) {
179 unixctl_command_register("cfm/show", cfm_unixctl_show, NULL);
182 /* Allocates a 'cfm' object. This object should have its 'mpid', 'maid',
183 * 'eth_src', and 'interval' filled out. cfm_configure() should be called
184 * whenever changes are made to 'cfm', and before cfm_run() is called for the
190 struct cfm_internal *cfmi;
192 cfmi = xzalloc(sizeof *cfmi);
195 hmap_init(&cfmi->remote_mps);
196 cfm_generate_maid(cfmi);
197 list_push_back(&all_cfms, &cfmi->list_node);
202 cfm_destroy(struct cfm *cfm)
204 struct cfm_internal *cfmi = cfm_to_internal(cfm);
205 struct remote_mp *rmp, *rmp_next;
211 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfmi->remote_mps) {
212 hmap_remove(&cfmi->remote_mps, &rmp->node);
216 hmap_destroy(&cfmi->remote_mps);
217 list_remove(&cfmi->list_node);
221 /* Should be run periodically to update fault statistics messages. */
223 cfm_run(struct cfm *cfm)
225 struct cfm_internal *cfmi = cfm_to_internal(cfm);
227 if (timer_expired(&cfmi->fault_timer)) {
228 long long int interval = cfm_fault_interval(cfmi);
229 struct remote_mp *rmp;
232 HMAP_FOR_EACH (rmp, node, &cfmi->remote_mps) {
233 rmp->fault = !rmp->recv;
238 VLOG_DBG("No CCM from RMP %"PRIu16" in the last %lldms",
239 rmp->mpid, interval);
244 VLOG_DBG("All RMPs received CCMs in the last %lldms", interval);
247 timer_set_duration(&cfmi->fault_timer, interval);
251 /* Should be run periodically to check if the CFM module has a CCM message it
254 cfm_should_send_ccm(struct cfm *cfm)
256 struct cfm_internal *cfmi = cfm_to_internal(cfm);
258 return timer_expired(&cfmi->tx_timer);
261 /* Composes a CCM message into 'ccm'. Messages generated with this function
262 * should be sent whenever cfm_should_send_ccm() indicates. */
264 cfm_compose_ccm(struct cfm *cfm, struct ccm *ccm)
266 struct cfm_internal *cfmi = cfm_to_internal(cfm);
268 timer_set_duration(&cfmi->tx_timer, cfmi->ccm_interval_ms);
270 ccm->mdlevel_version = 0;
271 ccm->opcode = CCM_OPCODE;
272 ccm->tlv_offset = 70;
273 ccm->seq = htonl(++cfmi->seq);
274 ccm->mpid = htons(cfmi->cfm.mpid);
275 ccm->flags = cfmi->ccm_interval;
276 memcpy(ccm->maid, cfmi->maid, sizeof ccm->maid);
280 cfm_wait(struct cfm *cfm)
282 struct cfm_internal *cfmi = cfm_to_internal(cfm);
284 timer_wait(&cfmi->tx_timer);
285 timer_wait(&cfmi->fault_timer);
288 /* Should be called whenever a client of the cfm library changes the internals
289 * of 'cfm'. Returns true if 'cfm' is valid. */
291 cfm_configure(struct cfm *cfm)
293 struct cfm_internal *cfmi = cfm_to_internal(cfm);
296 if (!cfm_is_valid_mpid(cfm->mpid) || !cfm->interval) {
300 interval = ms_to_ccm_interval(cfm->interval);
302 if (interval != cfmi->ccm_interval) {
303 cfmi->ccm_interval = interval;
304 cfmi->ccm_interval_ms = ccm_interval_to_ms(interval);
306 timer_set_expired(&cfmi->tx_timer);
307 timer_set_duration(&cfmi->fault_timer, cfm_fault_interval(cfmi));
313 /* Given an array of MPIDs, updates the 'remote_mps' map of 'cfm' to reflect
314 * it. Invalid MPIDs are skipped. */
316 cfm_update_remote_mps(struct cfm *cfm, const uint16_t *mpids, size_t n_mpids)
318 struct cfm_internal *cfmi = cfm_to_internal(cfm);
320 struct hmap new_rmps;
321 struct remote_mp *rmp, *rmp_next;
323 hmap_init(&new_rmps);
325 for (i = 0; i < n_mpids; i++) {
326 uint16_t mpid = mpids[i];
328 if (!cfm_is_valid_mpid(mpid)
329 || lookup_remote_mp(&new_rmps, mpid)) {
333 if ((rmp = lookup_remote_mp(&cfmi->remote_mps, mpid))) {
334 hmap_remove(&cfmi->remote_mps, &rmp->node);
336 rmp = xzalloc(sizeof *rmp);
340 hmap_insert(&new_rmps, &rmp->node, hash_mpid(mpid));
343 hmap_swap(&new_rmps, &cfmi->remote_mps);
345 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &new_rmps) {
346 hmap_remove(&new_rmps, &rmp->node);
350 hmap_destroy(&new_rmps);
353 /* Returns true if the CFM library should process packets from 'flow'. */
355 cfm_should_process_flow(const struct flow *flow)
357 return (ntohs(flow->dl_type) == ETH_TYPE_CFM
358 && eth_addr_equals(flow->dl_dst, eth_addr_ccm));
361 /* Updates internal statistics relevant to packet 'p'. Should be called on
362 * every packet whose flow returned true when passed to
363 * cfm_should_process_flow. */
365 cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
369 uint8_t ccm_interval;
370 struct remote_mp *rmp;
371 struct eth_header *eth;
372 struct cfm_internal *cfmi = cfm_to_internal(cfm);
375 ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_LEN);
378 VLOG_INFO_RL(&rl, "Received an un-parseable 802.1ag CCM heartbeat.");
382 if (ccm->opcode != CCM_OPCODE) {
383 VLOG_INFO_RL(&rl, "Received an unsupported 802.1ag message. "
384 "(opcode %u)", ccm->opcode);
388 /* According to the 802.1ag specification, reception of a CCM with an
389 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
390 * trigger a fault. We ignore this requirement for several reasons.
392 * Faults can cause a controller or Open vSwitch to make potentially
393 * expensive changes to the network topology. It seems prudent to trigger
394 * them judiciously, especially when CFM is used to check slave status of
395 * bonds. Furthermore, faults can be maliciously triggered by crafting
397 if (memcmp(ccm->maid, cfmi->maid, sizeof ccm->maid)) {
398 VLOG_WARN_RL(&rl, "Received unexpected remote MAID from MAC "
399 ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_src));
401 ccm_mpid = ntohs(ccm->mpid);
402 ccm_interval = ccm->flags & 0x7;
404 rmp = lookup_remote_mp(&cfmi->remote_mps, ccm_mpid);
409 if (ccm_interval != cfmi->ccm_interval) {
410 VLOG_WARN_RL(&rl, "received a CCM with an invalid interval"
411 " (%"PRIu8") from RMP %"PRIu16, ccm_interval,
415 VLOG_WARN_RL(&rl, "Received unexpected remote MPID %d from MAC "
416 ETH_ADDR_FMT, ccm_mpid, ETH_ADDR_ARGS(eth->eth_src));
419 VLOG_DBG("Received CCM (mpid %"PRIu16") (interval %"PRIu8")", ccm_mpid,
424 static struct cfm_internal *
425 cfm_find(const char *name)
427 struct cfm_internal *cfmi;
429 LIST_FOR_EACH (cfmi, list_node, &all_cfms) {
430 if (cfmi->cfm.name && !strcmp(cfmi->cfm.name, name)) {
438 cfm_unixctl_show(struct unixctl_conn *conn,
439 const char *args, void *aux OVS_UNUSED)
441 struct ds ds = DS_EMPTY_INITIALIZER;
442 const struct cfm_internal *cfmi;
443 struct remote_mp *rmp;
445 cfmi = cfm_find(args);
447 unixctl_command_reply(conn, 501, "no such CFM object");
451 ds_put_format(&ds, "MPID %"PRIu16": %s\n", cfmi->cfm.mpid,
452 cfmi->cfm.fault ? "fault" : "");
454 ds_put_format(&ds, "\tinterval: %dms\n", cfmi->ccm_interval_ms);
455 ds_put_format(&ds, "\tnext CCM tx: %lldms\n",
456 timer_msecs_until_expired(&cfmi->tx_timer));
457 ds_put_format(&ds, "\tnext fault check: %lldms\n",
458 timer_msecs_until_expired(&cfmi->fault_timer));
460 ds_put_cstr(&ds, "\n");
461 HMAP_FOR_EACH (rmp, node, &cfmi->remote_mps) {
462 ds_put_format(&ds, "Remote MPID %"PRIu16": %s\n", rmp->mpid,
463 rmp->fault ? "fault" : "");
464 ds_put_format(&ds, "\trecv since check: %s",
465 rmp->recv ? "true" : "false");
468 unixctl_command_reply(conn, 200, ds_cstr(&ds));