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 #define CCM_MAID_LEN 48
55 #define CCM_OPCODE 1 /* CFM message opcode meaning CCM. */
56 #define CCM_RDI_MASK 0x80
58 uint8_t mdlevel_version; /* MD Level and Version */
64 uint8_t maid[CCM_MAID_LEN];
66 /* Defined by ITU-T Y.1731 should be zero */
67 ovs_be16 interval_ms_x; /* Transmission interval in ms. */
68 ovs_be64 mpid64; /* MPID in extended mode. */
70 } __attribute__((packed));
71 BUILD_ASSERT_DECL(CCM_LEN == sizeof(struct ccm));
74 char *name; /* Name of this CFM object. */
75 struct hmap_node hmap_node; /* Node in all_cfms list. */
78 bool extended; /* Extended mode. */
79 bool fault; /* Indicates connectivity fault. */
80 bool unexpected_recv; /* Received an unexpected CCM. */
82 uint32_t seq; /* The sequence number of our last CCM. */
83 uint8_t ccm_interval; /* The CCM transmission interval. */
84 int ccm_interval_ms; /* 'ccm_interval' in milliseconds. */
85 uint8_t maid[CCM_MAID_LEN]; /* The MAID of this CFM. */
87 struct timer tx_timer; /* Send CCM when expired. */
88 struct timer fault_timer; /* Check for faults when expired. */
90 struct hmap remote_mps; /* Remote MPs. */
92 /* Result of cfm_get_remote_mpids(). Updated only during fault check to
94 uint64_t *rmps_array; /* Cache of remote_mps. */
95 size_t rmps_array_len; /* Number of rmps in 'rmps_array'. */
98 /* Remote MPs represent foreign network entities that are configured to have
99 * the same MAID as this CFM instance. */
101 uint64_t mpid; /* The Maintenance Point ID of this 'remote_mp'. */
102 struct hmap_node node; /* Node in 'remote_mps' map. */
104 bool recv; /* CCM was received since last fault check. */
105 bool rdi; /* Remote Defect Indicator. Indicates remote_mp isn't
106 receiving CCMs that it's expecting to. */
109 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
110 static struct hmap all_cfms = HMAP_INITIALIZER(&all_cfms);
112 static void cfm_unixctl_show(struct unixctl_conn *, const char *args,
115 static const uint8_t *
116 cfm_ccm_addr(const struct cfm *cfm)
118 return cfm->extended ? eth_addr_ccm_x : eth_addr_ccm;
122 cfm_generate_maid(struct cfm *cfm)
124 const char *ovs_md_name = "ovs";
125 const char *ovs_ma_name = "ovs";
127 size_t md_len, ma_len;
129 memset(cfm->maid, 0, CCM_MAID_LEN);
131 md_len = strlen(ovs_md_name);
132 ma_len = strlen(ovs_ma_name);
134 assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN);
136 cfm->maid[0] = 4; /* MD name string format. */
137 cfm->maid[1] = md_len; /* MD name size. */
138 memcpy(&cfm->maid[2], ovs_md_name, md_len); /* MD name. */
140 ma_p = cfm->maid + 2 + md_len;
141 ma_p[0] = 2; /* MA name string format. */
142 ma_p[1] = ma_len; /* MA name size. */
143 memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */
147 ccm_interval_to_ms(uint8_t interval)
150 case 0: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
151 case 1: return 3; /* Not recommended due to timer resolution. */
152 case 2: return 10; /* Not recommended due to timer resolution. */
155 case 5: return 10000;
156 case 6: return 60000;
157 case 7: return 600000;
158 default: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */
165 cfm_fault_interval(struct cfm *cfm)
167 /* According to the 802.1ag specification we should assume every other MP
168 * with the same MAID has the same transmission interval that we have. If
169 * an MP has a different interval, cfm_process_heartbeat will register it
170 * as a fault (likely due to a configuration error). Thus we can check all
171 * MPs at once making this quite a bit simpler.
173 * According to the specification we should check when (ccm_interval_ms *
174 * 3.5)ms have passed. */
175 return (cfm->ccm_interval_ms * 7) / 2;
179 ms_to_ccm_interval(int interval_ms)
183 for (i = 7; i > 0; i--) {
184 if (ccm_interval_to_ms(i) <= interval_ms) {
193 hash_mpid(uint64_t mpid)
195 return hash_bytes(&mpid, sizeof mpid, 0);
199 cfm_is_valid_mpid(bool extended, uint64_t mpid)
201 /* 802.1ag specification requires MPIDs to be within the range [1, 8191].
202 * In extended mode we relax this requirement. */
203 return mpid >= 1 && (extended || mpid <= 8191);
206 static struct remote_mp *
207 lookup_remote_mp(const struct cfm *cfm, uint64_t mpid)
209 struct remote_mp *rmp;
211 HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), &cfm->remote_mps) {
212 if (rmp->mpid == mpid) {
223 unixctl_command_register("cfm/show", "interface", cfm_unixctl_show, NULL);
226 /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
227 * cfm_configure() before use. */
229 cfm_create(const char *name)
233 cfm = xzalloc(sizeof *cfm);
234 cfm->name = xstrdup(name);
235 hmap_init(&cfm->remote_mps);
236 cfm_generate_maid(cfm);
237 hmap_insert(&all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0));
242 cfm_destroy(struct cfm *cfm)
244 struct remote_mp *rmp, *rmp_next;
250 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
251 hmap_remove(&cfm->remote_mps, &rmp->node);
255 hmap_destroy(&cfm->remote_mps);
256 hmap_remove(&all_cfms, &cfm->hmap_node);
257 free(cfm->rmps_array);
262 /* Should be run periodically to update fault statistics messages. */
264 cfm_run(struct cfm *cfm)
266 if (timer_expired(&cfm->fault_timer)) {
267 long long int interval = cfm_fault_interval(cfm);
268 struct remote_mp *rmp, *rmp_next;
270 cfm->fault = cfm->unexpected_recv;
271 cfm->unexpected_recv = false;
273 cfm->rmps_array_len = 0;
274 free(cfm->rmps_array);
275 cfm->rmps_array = xmalloc(hmap_count(&cfm->remote_mps) *
276 sizeof *cfm->rmps_array);
278 HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
281 VLOG_DBG("%s: no CCM from RMP %"PRIu64" in the last %lldms",
282 cfm->name, rmp->mpid, interval);
283 hmap_remove(&cfm->remote_mps, &rmp->node);
288 if (rmp->mpid == cfm->mpid) {
289 VLOG_WARN_RL(&rl,"%s: received CCM with local MPID"
290 " %"PRIu64, cfm->name, rmp->mpid);
295 VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64, cfm->name,
300 cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid;
304 if (hmap_is_empty(&cfm->remote_mps)) {
308 timer_set_duration(&cfm->fault_timer, interval);
312 /* Should be run periodically to check if the CFM module has a CCM message it
315 cfm_should_send_ccm(struct cfm *cfm)
317 return timer_expired(&cfm->tx_timer);
320 /* Composes a CCM message into 'packet'. Messages generated with this function
321 * should be sent whenever cfm_should_send_ccm() indicates. */
323 cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet,
324 uint8_t eth_src[ETH_ADDR_LEN])
328 timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms);
329 ccm = eth_compose(packet, cfm_ccm_addr(cfm), eth_src, ETH_TYPE_CFM,
331 ccm->mdlevel_version = 0;
332 ccm->opcode = CCM_OPCODE;
333 ccm->tlv_offset = 70;
334 ccm->seq = htonl(++cfm->seq);
335 ccm->flags = cfm->ccm_interval;
336 memcpy(ccm->maid, cfm->maid, sizeof ccm->maid);
337 memset(ccm->zero, 0, sizeof ccm->zero);
340 ccm->mpid = htons(hash_mpid(cfm->mpid));
341 ccm->mpid64 = htonll(cfm->mpid);
343 ccm->mpid = htons(cfm->mpid);
344 ccm->mpid64 = htonll(0);
347 if (cfm->ccm_interval == 0) {
348 assert(cfm->extended);
349 ccm->interval_ms_x = htons(cfm->ccm_interval_ms);
352 if (hmap_is_empty(&cfm->remote_mps)) {
353 ccm->flags |= CCM_RDI_MASK;
358 cfm_wait(struct cfm *cfm)
360 timer_wait(&cfm->tx_timer);
361 timer_wait(&cfm->fault_timer);
364 /* Configures 'cfm' with settings from 's'. */
366 cfm_configure(struct cfm *cfm, const struct cfm_settings *s)
371 if (!cfm_is_valid_mpid(s->extended, s->mpid) || s->interval <= 0) {
376 cfm->extended = s->extended;
377 interval = ms_to_ccm_interval(s->interval);
378 interval_ms = ccm_interval_to_ms(interval);
380 if (cfm->extended && interval_ms != s->interval) {
382 interval_ms = MIN(s->interval, UINT16_MAX);
385 if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) {
386 cfm->ccm_interval = interval;
387 cfm->ccm_interval_ms = interval_ms;
389 timer_set_expired(&cfm->tx_timer);
390 timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm));
396 /* Returns true if 'cfm' should process packets from 'flow'. */
398 cfm_should_process_flow(const struct cfm *cfm, const struct flow *flow)
400 return (ntohs(flow->dl_type) == ETH_TYPE_CFM
401 && eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm)));
404 /* Updates internal statistics relevant to packet 'p'. Should be called on
405 * every packet whose flow returned true when passed to
406 * cfm_should_process_flow. */
408 cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
411 struct eth_header *eth;
414 ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_LEN);
417 VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.",
422 if (ccm->opcode != CCM_OPCODE) {
423 VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. "
424 "(opcode %u)", cfm->name, ccm->opcode);
428 /* According to the 802.1ag specification, reception of a CCM with an
429 * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
430 * trigger a fault. We ignore this requirement for several reasons.
432 * Faults can cause a controller or Open vSwitch to make potentially
433 * expensive changes to the network topology. It seems prudent to trigger
434 * them judiciously, especially when CFM is used to check slave status of
435 * bonds. Furthermore, faults can be maliciously triggered by crafting
437 if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) {
438 cfm->unexpected_recv = true;
439 VLOG_WARN_RL(&rl, "%s: Received unexpected remote MAID from MAC "
440 ETH_ADDR_FMT, cfm->name, ETH_ADDR_ARGS(eth->eth_src));
442 uint8_t ccm_interval = ccm->flags & 0x7;
443 bool ccm_rdi = ccm->flags & CCM_RDI_MASK;
444 uint16_t ccm_interval_ms_x = ntohs(ccm->interval_ms_x);
446 struct remote_mp *rmp;
449 ccm_mpid = cfm->extended ? ntohll(ccm->mpid64) : ntohs(ccm->mpid);
451 if (ccm_interval != cfm->ccm_interval) {
452 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid interval"
453 " (%"PRIu8") from RMP %"PRIu64, cfm->name,
454 ccm_interval, ccm_mpid);
457 if (cfm->extended && ccm_interval == 0
458 && ccm_interval_ms_x != cfm->ccm_interval_ms) {
459 VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid extended"
460 " interval (%"PRIu16"ms) from RMP %"PRIu64, cfm->name,
461 ccm_interval_ms_x, ccm_mpid);
464 rmp = lookup_remote_mp(cfm, ccm_mpid);
468 } else if (hmap_count(&cfm->remote_mps) < CFM_MAX_RMPS) {
469 rmp = xmalloc(sizeof *rmp);
470 rmp->mpid = ccm_mpid;
473 hmap_insert(&cfm->remote_mps, &rmp->node, hash_mpid(rmp->mpid));
475 cfm->unexpected_recv = true;
476 VLOG_WARN_RL(&rl, "%s: dropped CCM with MPID %"PRIu64" from MAC "
477 ETH_ADDR_FMT, cfm->name, ccm_mpid,
478 ETH_ADDR_ARGS(eth->eth_src));
481 VLOG_DBG("%s: received CCM (seq %"PRIu32") (mpid %"PRIu64")"
482 " (interval %"PRIu8") (RDI %s)", cfm->name, ntohl(ccm->seq),
483 ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false");
487 /* Gets the fault status of 'cfm'. Returns true when 'cfm' has detected
488 * connectivity problems, false otherwise. */
490 cfm_get_fault(const struct cfm *cfm)
495 /* Populates 'rmps' with an array of remote maintenance points reachable by
496 * 'cfm'. The number of remote maintenance points is written to 'n_rmps'.
497 * 'cfm' retains ownership of the array written to 'rmps' */
499 cfm_get_remote_mpids(const struct cfm *cfm, const uint64_t **rmps,
502 *rmps = cfm->rmps_array;
503 *n_rmps = cfm->rmps_array_len;
507 cfm_find(const char *name)
511 HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), &all_cfms) {
512 if (!strcmp(cfm->name, name)) {
520 cfm_unixctl_show(struct unixctl_conn *conn,
521 const char *args, void *aux OVS_UNUSED)
523 struct ds ds = DS_EMPTY_INITIALIZER;
524 const struct cfm *cfm;
525 struct remote_mp *rmp;
527 cfm = cfm_find(args);
529 unixctl_command_reply(conn, 501, "no such CFM object");
533 ds_put_format(&ds, "MPID %"PRIu64":%s%s\n", cfm->mpid,
534 cfm->fault ? " fault" : "",
535 cfm->unexpected_recv ? " unexpected_recv" : "");
537 ds_put_format(&ds, "\tinterval: %dms\n", cfm->ccm_interval_ms);
538 ds_put_format(&ds, "\tnext CCM tx: %lldms\n",
539 timer_msecs_until_expired(&cfm->tx_timer));
540 ds_put_format(&ds, "\tnext fault check: %lldms\n",
541 timer_msecs_until_expired(&cfm->fault_timer));
543 ds_put_cstr(&ds, "\n");
544 HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
545 ds_put_format(&ds, "Remote MPID %"PRIu64":%s\n",
547 rmp->rdi ? " rdi" : "");
548 ds_put_format(&ds, "\trecv since check: %s",
549 rmp->recv ? "true" : "false");
552 unixctl_command_reply(conn, 200, ds_cstr(&ds));